korg/linux
0
0
Fork 0
mirror of https://mirrors.bfsu.edu.cn/git/linux.git synced 2024-12-19 17:14:40 +08:00
Code Issues Packages Projects Releases Wiki Activity
bb677d527e
linux/drivers/misc/habanalabs/gaudi2/gaudi2.c

10254 lines
335 KiB
C
Raw Normal View History

habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
// SPDX-License-Identifier: GPL-2.0
/*
* Copyright 2020-2022 HabanaLabs, Ltd.
* All Rights Reserved.
*/
#include "gaudi2P.h"
#include "gaudi2_masks.h"
#include "../include/hw_ip/mmu/mmu_general.h"
#include "../include/hw_ip/mmu/mmu_v2_0.h"
#include "../include/gaudi2/gaudi2_packets.h"
#include "../include/gaudi2/gaudi2_reg_map.h"
#include "../include/gaudi2/gaudi2_async_ids_map_extended.h"
#include "../include/gaudi2/arc/gaudi2_arc_common_packets.h"
#include <linux/module.h>
#include <linux/pci.h>
#include <linux/hwmon.h>
#include <linux/iommu.h>
#define GAUDI2_DMA_POOL_BLK_SIZE SZ_256 /* 256 bytes */
habanalabs/gaudi2: increase hard-reset sleep time to 2 sec The access to the device registers is blocked during hard reset, until preboot runs and allows the access to specific registers, including the PSOC BTM_FSM register which is used to know when the reset is done. Between the reset request and until this register is polled there is a small delay of 500 msec which is not enough for F/W to process the reset and for preboot to run, so the register might be accessed while it is blocked. To avoid it, increase the delay to 2 sec. Signed-off-by: Tomer Tayar <ttayar@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-09-06 21:37:14 +08:00
#define GAUDI2_RESET_TIMEOUT_MSEC 2000 /* 2000ms */
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
#define GAUDI2_RESET_POLL_TIMEOUT_USEC 50000 /* 50ms */
#define GAUDI2_PLDM_HRESET_TIMEOUT_MSEC 25000 /* 25s */
#define GAUDI2_PLDM_SRESET_TIMEOUT_MSEC 25000 /* 25s */
#define GAUDI2_PLDM_RESET_POLL_TIMEOUT_USEC 3000000 /* 3s */
#define GAUDI2_RESET_POLL_CNT 3
#define GAUDI2_RESET_WAIT_MSEC 1 /* 1ms */
#define GAUDI2_CPU_RESET_WAIT_MSEC 100 /* 100ms */
#define GAUDI2_PLDM_RESET_WAIT_MSEC 1000 /* 1s */
#define GAUDI2_CB_POOL_CB_CNT 512
#define GAUDI2_CB_POOL_CB_SIZE SZ_128K /* 128KB */
#define GAUDI2_MSG_TO_CPU_TIMEOUT_USEC 4000000 /* 4s */
#define GAUDI2_WAIT_FOR_BL_TIMEOUT_USEC 25000000 /* 25s */
#define GAUDI2_TEST_QUEUE_WAIT_USEC 100000 /* 100ms */
#define GAUDI2_PLDM_TEST_QUEUE_WAIT_USEC 1000000 /* 1s */
#define GAUDI2_ALLOC_CPU_MEM_RETRY_CNT 3
/*
* since the code already has built-in support for binning of up to MAX_FAULTY_TPCS TPCs
* and the code relies on that value (for array size etc..) we define another value
* for MAX faulty TPCs which reflects the cluster binning requirements
*/
#define MAX_CLUSTER_BINNING_FAULTY_TPCS 1
#define MAX_FAULTY_XBARS 1
#define MAX_FAULTY_EDMAS 1
#define MAX_FAULTY_DECODERS 1
#define GAUDI2_TPC_FULL_MASK 0x1FFFFFF
#define GAUDI2_HIF_HMMU_FULL_MASK 0xFFFF
#define GAUDI2_DECODER_FULL_MASK 0x3FF
#define GAUDI2_NUM_OF_QM_ERR_CAUSE 18
#define GAUDI2_NUM_OF_QM_LCP_ERR_CAUSE 25
#define GAUDI2_NUM_OF_QM_ARB_ERR_CAUSE 3
#define GAUDI2_NUM_OF_ARC_SEI_ERR_CAUSE 14
#define GAUDI2_NUM_OF_CPU_SEI_ERR_CAUSE 3
#define GAUDI2_NUM_OF_QM_SEI_ERR_CAUSE 2
#define GAUDI2_NUM_OF_ROT_ERR_CAUSE 22
#define GAUDI2_NUM_OF_TPC_INTR_CAUSE 30
#define GAUDI2_NUM_OF_DEC_ERR_CAUSE 25
#define GAUDI2_NUM_OF_MME_ERR_CAUSE 16
#define GAUDI2_NUM_OF_MME_SBTE_ERR_CAUSE 5
#define GAUDI2_NUM_OF_MME_WAP_ERR_CAUSE 7
#define GAUDI2_NUM_OF_DMA_CORE_INTR_CAUSE 8
#define GAUDI2_NUM_OF_MMU_SPI_SEI_CAUSE 19
#define GAUDI2_NUM_OF_HBM_SEI_CAUSE 9
#define GAUDI2_NUM_OF_SM_SEI_ERR_CAUSE 3
#define GAUDI2_NUM_OF_PCIE_ADDR_DEC_ERR_CAUSE 3
#define GAUDI2_NUM_OF_PMMU_FATAL_ERR_CAUSE 2
#define GAUDI2_NUM_OF_HIF_FATAL_ERR_CAUSE 2
#define GAUDI2_NUM_OF_AXI_DRAIN_ERR_CAUSE 2
#define GAUDI2_NUM_OF_HBM_MC_SPI_CAUSE 5
#define GAUDI2_MMU_CACHE_INV_TIMEOUT_USEC (MMU_CONFIG_TIMEOUT_USEC * 10)
#define GAUDI2_PLDM_MMU_TIMEOUT_USEC (MMU_CONFIG_TIMEOUT_USEC * 200)
#define GAUDI2_ARB_WDT_TIMEOUT (0x1000000)
#define GAUDI2_VDEC_TIMEOUT_USEC 10000 /* 10ms */
#define GAUDI2_PLDM_VDEC_TIMEOUT_USEC (GAUDI2_VDEC_TIMEOUT_USEC * 100)
#define KDMA_TIMEOUT_USEC USEC_PER_SEC
#define IS_DMA_IDLE(dma_core_idle_ind_mask) \
(!((dma_core_idle_ind_mask) & \
((DCORE0_EDMA0_CORE_IDLE_IND_MASK_DESC_CNT_STS_MASK) | \
(DCORE0_EDMA0_CORE_IDLE_IND_MASK_COMP_MASK))))
#define IS_MME_IDLE(mme_arch_sts) (((mme_arch_sts) & MME_ARCH_IDLE_MASK) == MME_ARCH_IDLE_MASK)
#define IS_TPC_IDLE(tpc_cfg_sts) (((tpc_cfg_sts) & (TPC_IDLE_MASK)) == (TPC_IDLE_MASK))
#define IS_QM_IDLE(qm_glbl_sts0, qm_glbl_sts1, qm_cgm_sts) \
((((qm_glbl_sts0) & (QM_IDLE_MASK)) == (QM_IDLE_MASK)) && \
(((qm_glbl_sts1) & (QM_ARC_IDLE_MASK)) == (QM_ARC_IDLE_MASK)) && \
(((qm_cgm_sts) & (CGM_IDLE_MASK)) == (CGM_IDLE_MASK)))
#define PCIE_DEC_EN_MASK 0x300
#define DEC_WORK_STATE_IDLE 0
#define DEC_WORK_STATE_PEND 3
#define IS_DEC_IDLE(dec_swreg15) \
(((dec_swreg15) & DCORE0_DEC0_CMD_SWREG15_SW_WORK_STATE_MASK) == DEC_WORK_STATE_IDLE || \
((dec_swreg15) & DCORE0_DEC0_CMD_SWREG15_SW_WORK_STATE_MASK) == DEC_WORK_STATE_PEND)
/* HBM MMU address scrambling parameters */
#define GAUDI2_HBM_MMU_SCRM_MEM_SIZE SZ_8M
#define GAUDI2_HBM_MMU_SCRM_DIV_SHIFT 26
#define GAUDI2_HBM_MMU_SCRM_MOD_SHIFT 0
#define GAUDI2_HBM_MMU_SCRM_ADDRESS_MASK DRAM_VA_HINT_MASK
#define GAUDI2_COMPENSATE_TLB_PAGE_SIZE_FACTOR 16
#define MMU_RANGE_INV_VA_LSB_SHIFT 12
#define MMU_RANGE_INV_VA_MSB_SHIFT 44
#define MMU_RANGE_INV_EN_SHIFT 0
#define MMU_RANGE_INV_ASID_EN_SHIFT 1
#define MMU_RANGE_INV_ASID_SHIFT 2
habanalabs/gaudi2: enable all MMU SPI/SEI interrupts Currently only part of the MMU SPI/SEI interrupts are enabled, although there is no real reason to not enable all. The only exception is "burst_fifo_full" which is expected for PMMU because it has a 2 entries FIFO, and thus is it not enabled for it. Signed-off-by: Tomer Tayar <ttayar@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-07-12 23:19:11 +08:00
/* The last SPI_SEI cause bit, "burst_fifo_full", is expected to be triggered in PMMU because it has
* a 2 entries FIFO, and hence it is not enabled for it.
*/
#define GAUDI2_PMMU_SPI_SEI_ENABLE_MASK GENMASK(GAUDI2_NUM_OF_MMU_SPI_SEI_CAUSE - 2, 0)
#define GAUDI2_HMMU_SPI_SEI_ENABLE_MASK GENMASK(GAUDI2_NUM_OF_MMU_SPI_SEI_CAUSE - 1, 0)
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
#define GAUDI2_MAX_STRING_LEN 64
#define GAUDI2_VDEC_MSIX_ENTRIES (GAUDI2_IRQ_NUM_SHARED_DEC1_ABNRM - \
GAUDI2_IRQ_NUM_DCORE0_DEC0_NRM + 1)
enum hl_pmmu_fatal_cause {
LATENCY_RD_OUT_FIFO_OVERRUN,
LATENCY_WR_OUT_FIFO_OVERRUN,
};
enum hl_pcie_drain_ind_cause {
LBW_AXI_DRAIN_IND,
HBW_AXI_DRAIN_IND
};
static const u32 cluster_hmmu_hif_enabled_mask[GAUDI2_HBM_NUM] = {
[HBM_ID0] = 0xFFFC,
[HBM_ID1] = 0xFFCF,
[HBM_ID2] = 0xF7F7,
[HBM_ID3] = 0x7F7F,
[HBM_ID4] = 0xFCFF,
[HBM_ID5] = 0xCFFF,
};
static const u8 xbar_edge_to_hbm_cluster[EDMA_ID_SIZE] = {
[0] = HBM_ID0,
[1] = HBM_ID1,
[2] = HBM_ID4,
[3] = HBM_ID5,
};
static const u8 edma_to_hbm_cluster[EDMA_ID_SIZE] = {
[EDMA_ID_DCORE0_INSTANCE0] = HBM_ID0,
[EDMA_ID_DCORE0_INSTANCE1] = HBM_ID2,
[EDMA_ID_DCORE1_INSTANCE0] = HBM_ID1,
[EDMA_ID_DCORE1_INSTANCE1] = HBM_ID3,
[EDMA_ID_DCORE2_INSTANCE0] = HBM_ID2,
[EDMA_ID_DCORE2_INSTANCE1] = HBM_ID4,
[EDMA_ID_DCORE3_INSTANCE0] = HBM_ID3,
[EDMA_ID_DCORE3_INSTANCE1] = HBM_ID5,
};
static const int gaudi2_qman_async_event_id[] = {
[GAUDI2_QUEUE_ID_PDMA_0_0] = GAUDI2_EVENT_PDMA0_QM,
[GAUDI2_QUEUE_ID_PDMA_0_1] = GAUDI2_EVENT_PDMA0_QM,
[GAUDI2_QUEUE_ID_PDMA_0_2] = GAUDI2_EVENT_PDMA0_QM,
[GAUDI2_QUEUE_ID_PDMA_0_3] = GAUDI2_EVENT_PDMA0_QM,
[GAUDI2_QUEUE_ID_PDMA_1_0] = GAUDI2_EVENT_PDMA1_QM,
[GAUDI2_QUEUE_ID_PDMA_1_1] = GAUDI2_EVENT_PDMA1_QM,
[GAUDI2_QUEUE_ID_PDMA_1_2] = GAUDI2_EVENT_PDMA1_QM,
[GAUDI2_QUEUE_ID_PDMA_1_3] = GAUDI2_EVENT_PDMA1_QM,
[GAUDI2_QUEUE_ID_DCORE0_EDMA_0_0] = GAUDI2_EVENT_HDMA0_QM,
[GAUDI2_QUEUE_ID_DCORE0_EDMA_0_1] = GAUDI2_EVENT_HDMA0_QM,
[GAUDI2_QUEUE_ID_DCORE0_EDMA_0_2] = GAUDI2_EVENT_HDMA0_QM,
[GAUDI2_QUEUE_ID_DCORE0_EDMA_0_3] = GAUDI2_EVENT_HDMA0_QM,
[GAUDI2_QUEUE_ID_DCORE0_EDMA_1_0] = GAUDI2_EVENT_HDMA1_QM,
[GAUDI2_QUEUE_ID_DCORE0_EDMA_1_1] = GAUDI2_EVENT_HDMA1_QM,
[GAUDI2_QUEUE_ID_DCORE0_EDMA_1_2] = GAUDI2_EVENT_HDMA1_QM,
[GAUDI2_QUEUE_ID_DCORE0_EDMA_1_3] = GAUDI2_EVENT_HDMA1_QM,
[GAUDI2_QUEUE_ID_DCORE0_MME_0_0] = GAUDI2_EVENT_MME0_QM,
[GAUDI2_QUEUE_ID_DCORE0_MME_0_1] = GAUDI2_EVENT_MME0_QM,
[GAUDI2_QUEUE_ID_DCORE0_MME_0_2] = GAUDI2_EVENT_MME0_QM,
[GAUDI2_QUEUE_ID_DCORE0_MME_0_3] = GAUDI2_EVENT_MME0_QM,
[GAUDI2_QUEUE_ID_DCORE0_TPC_0_0] = GAUDI2_EVENT_TPC0_QM,
[GAUDI2_QUEUE_ID_DCORE0_TPC_0_1] = GAUDI2_EVENT_TPC0_QM,
[GAUDI2_QUEUE_ID_DCORE0_TPC_0_2] = GAUDI2_EVENT_TPC0_QM,
[GAUDI2_QUEUE_ID_DCORE0_TPC_0_3] = GAUDI2_EVENT_TPC0_QM,
[GAUDI2_QUEUE_ID_DCORE0_TPC_1_0] = GAUDI2_EVENT_TPC1_QM,
[GAUDI2_QUEUE_ID_DCORE0_TPC_1_1] = GAUDI2_EVENT_TPC1_QM,
[GAUDI2_QUEUE_ID_DCORE0_TPC_1_2] = GAUDI2_EVENT_TPC1_QM,
[GAUDI2_QUEUE_ID_DCORE0_TPC_1_3] = GAUDI2_EVENT_TPC1_QM,
[GAUDI2_QUEUE_ID_DCORE0_TPC_2_0] = GAUDI2_EVENT_TPC2_QM,
[GAUDI2_QUEUE_ID_DCORE0_TPC_2_1] = GAUDI2_EVENT_TPC2_QM,
[GAUDI2_QUEUE_ID_DCORE0_TPC_2_2] = GAUDI2_EVENT_TPC2_QM,
[GAUDI2_QUEUE_ID_DCORE0_TPC_2_3] = GAUDI2_EVENT_TPC2_QM,
[GAUDI2_QUEUE_ID_DCORE0_TPC_3_0] = GAUDI2_EVENT_TPC3_QM,
[GAUDI2_QUEUE_ID_DCORE0_TPC_3_1] = GAUDI2_EVENT_TPC3_QM,
[GAUDI2_QUEUE_ID_DCORE0_TPC_3_2] = GAUDI2_EVENT_TPC3_QM,
[GAUDI2_QUEUE_ID_DCORE0_TPC_3_3] = GAUDI2_EVENT_TPC3_QM,
[GAUDI2_QUEUE_ID_DCORE0_TPC_4_0] = GAUDI2_EVENT_TPC4_QM,
[GAUDI2_QUEUE_ID_DCORE0_TPC_4_1] = GAUDI2_EVENT_TPC4_QM,
[GAUDI2_QUEUE_ID_DCORE0_TPC_4_2] = GAUDI2_EVENT_TPC4_QM,
[GAUDI2_QUEUE_ID_DCORE0_TPC_4_3] = GAUDI2_EVENT_TPC4_QM,
[GAUDI2_QUEUE_ID_DCORE0_TPC_5_0] = GAUDI2_EVENT_TPC5_QM,
[GAUDI2_QUEUE_ID_DCORE0_TPC_5_1] = GAUDI2_EVENT_TPC5_QM,
[GAUDI2_QUEUE_ID_DCORE0_TPC_5_2] = GAUDI2_EVENT_TPC5_QM,
[GAUDI2_QUEUE_ID_DCORE0_TPC_5_3] = GAUDI2_EVENT_TPC5_QM,
[GAUDI2_QUEUE_ID_DCORE0_TPC_6_0] = GAUDI2_EVENT_TPC24_QM,
[GAUDI2_QUEUE_ID_DCORE0_TPC_6_1] = GAUDI2_EVENT_TPC24_QM,
[GAUDI2_QUEUE_ID_DCORE0_TPC_6_2] = GAUDI2_EVENT_TPC24_QM,
[GAUDI2_QUEUE_ID_DCORE0_TPC_6_3] = GAUDI2_EVENT_TPC24_QM,
[GAUDI2_QUEUE_ID_DCORE1_EDMA_0_0] = GAUDI2_EVENT_HDMA2_QM,
[GAUDI2_QUEUE_ID_DCORE1_EDMA_0_1] = GAUDI2_EVENT_HDMA2_QM,
[GAUDI2_QUEUE_ID_DCORE1_EDMA_0_2] = GAUDI2_EVENT_HDMA2_QM,
[GAUDI2_QUEUE_ID_DCORE1_EDMA_0_3] = GAUDI2_EVENT_HDMA2_QM,
[GAUDI2_QUEUE_ID_DCORE1_EDMA_1_0] = GAUDI2_EVENT_HDMA3_QM,
[GAUDI2_QUEUE_ID_DCORE1_EDMA_1_1] = GAUDI2_EVENT_HDMA3_QM,
[GAUDI2_QUEUE_ID_DCORE1_EDMA_1_2] = GAUDI2_EVENT_HDMA3_QM,
[GAUDI2_QUEUE_ID_DCORE1_EDMA_1_3] = GAUDI2_EVENT_HDMA3_QM,
[GAUDI2_QUEUE_ID_DCORE1_MME_0_0] = GAUDI2_EVENT_MME1_QM,
[GAUDI2_QUEUE_ID_DCORE1_MME_0_1] = GAUDI2_EVENT_MME1_QM,
[GAUDI2_QUEUE_ID_DCORE1_MME_0_2] = GAUDI2_EVENT_MME1_QM,
[GAUDI2_QUEUE_ID_DCORE1_MME_0_3] = GAUDI2_EVENT_MME1_QM,
[GAUDI2_QUEUE_ID_DCORE1_TPC_0_0] = GAUDI2_EVENT_TPC6_QM,
[GAUDI2_QUEUE_ID_DCORE1_TPC_0_1] = GAUDI2_EVENT_TPC6_QM,
[GAUDI2_QUEUE_ID_DCORE1_TPC_0_2] = GAUDI2_EVENT_TPC6_QM,
[GAUDI2_QUEUE_ID_DCORE1_TPC_0_3] = GAUDI2_EVENT_TPC6_QM,
[GAUDI2_QUEUE_ID_DCORE1_TPC_1_0] = GAUDI2_EVENT_TPC7_QM,
[GAUDI2_QUEUE_ID_DCORE1_TPC_1_1] = GAUDI2_EVENT_TPC7_QM,
[GAUDI2_QUEUE_ID_DCORE1_TPC_1_2] = GAUDI2_EVENT_TPC7_QM,
[GAUDI2_QUEUE_ID_DCORE1_TPC_1_3] = GAUDI2_EVENT_TPC7_QM,
[GAUDI2_QUEUE_ID_DCORE1_TPC_2_0] = GAUDI2_EVENT_TPC8_QM,
[GAUDI2_QUEUE_ID_DCORE1_TPC_2_1] = GAUDI2_EVENT_TPC8_QM,
[GAUDI2_QUEUE_ID_DCORE1_TPC_2_2] = GAUDI2_EVENT_TPC8_QM,
[GAUDI2_QUEUE_ID_DCORE1_TPC_2_3] = GAUDI2_EVENT_TPC8_QM,
[GAUDI2_QUEUE_ID_DCORE1_TPC_3_0] = GAUDI2_EVENT_TPC9_QM,
[GAUDI2_QUEUE_ID_DCORE1_TPC_3_1] = GAUDI2_EVENT_TPC9_QM,
[GAUDI2_QUEUE_ID_DCORE1_TPC_3_2] = GAUDI2_EVENT_TPC9_QM,
[GAUDI2_QUEUE_ID_DCORE1_TPC_3_3] = GAUDI2_EVENT_TPC9_QM,
[GAUDI2_QUEUE_ID_DCORE1_TPC_4_0] = GAUDI2_EVENT_TPC10_QM,
[GAUDI2_QUEUE_ID_DCORE1_TPC_4_1] = GAUDI2_EVENT_TPC10_QM,
[GAUDI2_QUEUE_ID_DCORE1_TPC_4_2] = GAUDI2_EVENT_TPC10_QM,
[GAUDI2_QUEUE_ID_DCORE1_TPC_4_3] = GAUDI2_EVENT_TPC10_QM,
[GAUDI2_QUEUE_ID_DCORE1_TPC_5_0] = GAUDI2_EVENT_TPC11_QM,
[GAUDI2_QUEUE_ID_DCORE1_TPC_5_1] = GAUDI2_EVENT_TPC11_QM,
[GAUDI2_QUEUE_ID_DCORE1_TPC_5_2] = GAUDI2_EVENT_TPC11_QM,
[GAUDI2_QUEUE_ID_DCORE1_TPC_5_3] = GAUDI2_EVENT_TPC11_QM,
[GAUDI2_QUEUE_ID_DCORE2_EDMA_0_0] = GAUDI2_EVENT_HDMA4_QM,
[GAUDI2_QUEUE_ID_DCORE2_EDMA_0_1] = GAUDI2_EVENT_HDMA4_QM,
[GAUDI2_QUEUE_ID_DCORE2_EDMA_0_2] = GAUDI2_EVENT_HDMA4_QM,
[GAUDI2_QUEUE_ID_DCORE2_EDMA_0_3] = GAUDI2_EVENT_HDMA4_QM,
[GAUDI2_QUEUE_ID_DCORE2_EDMA_1_0] = GAUDI2_EVENT_HDMA5_QM,
[GAUDI2_QUEUE_ID_DCORE2_EDMA_1_1] = GAUDI2_EVENT_HDMA5_QM,
[GAUDI2_QUEUE_ID_DCORE2_EDMA_1_2] = GAUDI2_EVENT_HDMA5_QM,
[GAUDI2_QUEUE_ID_DCORE2_EDMA_1_3] = GAUDI2_EVENT_HDMA5_QM,
[GAUDI2_QUEUE_ID_DCORE2_MME_0_0] = GAUDI2_EVENT_MME2_QM,
[GAUDI2_QUEUE_ID_DCORE2_MME_0_1] = GAUDI2_EVENT_MME2_QM,
[GAUDI2_QUEUE_ID_DCORE2_MME_0_2] = GAUDI2_EVENT_MME2_QM,
[GAUDI2_QUEUE_ID_DCORE2_MME_0_3] = GAUDI2_EVENT_MME2_QM,
[GAUDI2_QUEUE_ID_DCORE2_TPC_0_0] = GAUDI2_EVENT_TPC12_QM,
[GAUDI2_QUEUE_ID_DCORE2_TPC_0_1] = GAUDI2_EVENT_TPC12_QM,
[GAUDI2_QUEUE_ID_DCORE2_TPC_0_2] = GAUDI2_EVENT_TPC12_QM,
[GAUDI2_QUEUE_ID_DCORE2_TPC_0_3] = GAUDI2_EVENT_TPC12_QM,
[GAUDI2_QUEUE_ID_DCORE2_TPC_1_0] = GAUDI2_EVENT_TPC13_QM,
[GAUDI2_QUEUE_ID_DCORE2_TPC_1_1] = GAUDI2_EVENT_TPC13_QM,
[GAUDI2_QUEUE_ID_DCORE2_TPC_1_2] = GAUDI2_EVENT_TPC13_QM,
[GAUDI2_QUEUE_ID_DCORE2_TPC_1_3] = GAUDI2_EVENT_TPC13_QM,
[GAUDI2_QUEUE_ID_DCORE2_TPC_2_0] = GAUDI2_EVENT_TPC14_QM,
[GAUDI2_QUEUE_ID_DCORE2_TPC_2_1] = GAUDI2_EVENT_TPC14_QM,
[GAUDI2_QUEUE_ID_DCORE2_TPC_2_2] = GAUDI2_EVENT_TPC14_QM,
[GAUDI2_QUEUE_ID_DCORE2_TPC_2_3] = GAUDI2_EVENT_TPC14_QM,
[GAUDI2_QUEUE_ID_DCORE2_TPC_3_0] = GAUDI2_EVENT_TPC15_QM,
[GAUDI2_QUEUE_ID_DCORE2_TPC_3_1] = GAUDI2_EVENT_TPC15_QM,
[GAUDI2_QUEUE_ID_DCORE2_TPC_3_2] = GAUDI2_EVENT_TPC15_QM,
[GAUDI2_QUEUE_ID_DCORE2_TPC_3_3] = GAUDI2_EVENT_TPC15_QM,
[GAUDI2_QUEUE_ID_DCORE2_TPC_4_0] = GAUDI2_EVENT_TPC16_QM,
[GAUDI2_QUEUE_ID_DCORE2_TPC_4_1] = GAUDI2_EVENT_TPC16_QM,
[GAUDI2_QUEUE_ID_DCORE2_TPC_4_2] = GAUDI2_EVENT_TPC16_QM,
[GAUDI2_QUEUE_ID_DCORE2_TPC_4_3] = GAUDI2_EVENT_TPC16_QM,
[GAUDI2_QUEUE_ID_DCORE2_TPC_5_0] = GAUDI2_EVENT_TPC17_QM,
[GAUDI2_QUEUE_ID_DCORE2_TPC_5_1] = GAUDI2_EVENT_TPC17_QM,
[GAUDI2_QUEUE_ID_DCORE2_TPC_5_2] = GAUDI2_EVENT_TPC17_QM,
[GAUDI2_QUEUE_ID_DCORE2_TPC_5_3] = GAUDI2_EVENT_TPC17_QM,
[GAUDI2_QUEUE_ID_DCORE3_EDMA_0_0] = GAUDI2_EVENT_HDMA6_QM,
[GAUDI2_QUEUE_ID_DCORE3_EDMA_0_1] = GAUDI2_EVENT_HDMA6_QM,
[GAUDI2_QUEUE_ID_DCORE3_EDMA_0_2] = GAUDI2_EVENT_HDMA6_QM,
[GAUDI2_QUEUE_ID_DCORE3_EDMA_0_3] = GAUDI2_EVENT_HDMA6_QM,
[GAUDI2_QUEUE_ID_DCORE3_EDMA_1_0] = GAUDI2_EVENT_HDMA7_QM,
[GAUDI2_QUEUE_ID_DCORE3_EDMA_1_1] = GAUDI2_EVENT_HDMA7_QM,
[GAUDI2_QUEUE_ID_DCORE3_EDMA_1_2] = GAUDI2_EVENT_HDMA7_QM,
[GAUDI2_QUEUE_ID_DCORE3_EDMA_1_3] = GAUDI2_EVENT_HDMA7_QM,
[GAUDI2_QUEUE_ID_DCORE3_MME_0_0] = GAUDI2_EVENT_MME3_QM,
[GAUDI2_QUEUE_ID_DCORE3_MME_0_1] = GAUDI2_EVENT_MME3_QM,
[GAUDI2_QUEUE_ID_DCORE3_MME_0_2] = GAUDI2_EVENT_MME3_QM,
[GAUDI2_QUEUE_ID_DCORE3_MME_0_3] = GAUDI2_EVENT_MME3_QM,
[GAUDI2_QUEUE_ID_DCORE3_TPC_0_0] = GAUDI2_EVENT_TPC18_QM,
[GAUDI2_QUEUE_ID_DCORE3_TPC_0_1] = GAUDI2_EVENT_TPC18_QM,
[GAUDI2_QUEUE_ID_DCORE3_TPC_0_2] = GAUDI2_EVENT_TPC18_QM,
[GAUDI2_QUEUE_ID_DCORE3_TPC_0_3] = GAUDI2_EVENT_TPC18_QM,
[GAUDI2_QUEUE_ID_DCORE3_TPC_1_0] = GAUDI2_EVENT_TPC19_QM,
[GAUDI2_QUEUE_ID_DCORE3_TPC_1_1] = GAUDI2_EVENT_TPC19_QM,
[GAUDI2_QUEUE_ID_DCORE3_TPC_1_2] = GAUDI2_EVENT_TPC19_QM,
[GAUDI2_QUEUE_ID_DCORE3_TPC_1_3] = GAUDI2_EVENT_TPC19_QM,
[GAUDI2_QUEUE_ID_DCORE3_TPC_2_0] = GAUDI2_EVENT_TPC20_QM,
[GAUDI2_QUEUE_ID_DCORE3_TPC_2_1] = GAUDI2_EVENT_TPC20_QM,
[GAUDI2_QUEUE_ID_DCORE3_TPC_2_2] = GAUDI2_EVENT_TPC20_QM,
[GAUDI2_QUEUE_ID_DCORE3_TPC_2_3] = GAUDI2_EVENT_TPC20_QM,
[GAUDI2_QUEUE_ID_DCORE3_TPC_3_0] = GAUDI2_EVENT_TPC21_QM,
[GAUDI2_QUEUE_ID_DCORE3_TPC_3_1] = GAUDI2_EVENT_TPC21_QM,
[GAUDI2_QUEUE_ID_DCORE3_TPC_3_2] = GAUDI2_EVENT_TPC21_QM,
[GAUDI2_QUEUE_ID_DCORE3_TPC_3_3] = GAUDI2_EVENT_TPC21_QM,
[GAUDI2_QUEUE_ID_DCORE3_TPC_4_0] = GAUDI2_EVENT_TPC22_QM,
[GAUDI2_QUEUE_ID_DCORE3_TPC_4_1] = GAUDI2_EVENT_TPC22_QM,
[GAUDI2_QUEUE_ID_DCORE3_TPC_4_2] = GAUDI2_EVENT_TPC22_QM,
[GAUDI2_QUEUE_ID_DCORE3_TPC_4_3] = GAUDI2_EVENT_TPC22_QM,
[GAUDI2_QUEUE_ID_DCORE3_TPC_5_0] = GAUDI2_EVENT_TPC23_QM,
[GAUDI2_QUEUE_ID_DCORE3_TPC_5_1] = GAUDI2_EVENT_TPC23_QM,
[GAUDI2_QUEUE_ID_DCORE3_TPC_5_2] = GAUDI2_EVENT_TPC23_QM,
[GAUDI2_QUEUE_ID_DCORE3_TPC_5_3] = GAUDI2_EVENT_TPC23_QM,
[GAUDI2_QUEUE_ID_NIC_0_0] = GAUDI2_EVENT_NIC0_QM0,
[GAUDI2_QUEUE_ID_NIC_0_1] = GAUDI2_EVENT_NIC0_QM0,
[GAUDI2_QUEUE_ID_NIC_0_2] = GAUDI2_EVENT_NIC0_QM0,
[GAUDI2_QUEUE_ID_NIC_0_3] = GAUDI2_EVENT_NIC0_QM0,
[GAUDI2_QUEUE_ID_NIC_1_0] = GAUDI2_EVENT_NIC0_QM1,
[GAUDI2_QUEUE_ID_NIC_1_1] = GAUDI2_EVENT_NIC0_QM1,
[GAUDI2_QUEUE_ID_NIC_1_2] = GAUDI2_EVENT_NIC0_QM1,
[GAUDI2_QUEUE_ID_NIC_1_3] = GAUDI2_EVENT_NIC0_QM1,
[GAUDI2_QUEUE_ID_NIC_2_0] = GAUDI2_EVENT_NIC1_QM0,
[GAUDI2_QUEUE_ID_NIC_2_1] = GAUDI2_EVENT_NIC1_QM0,
[GAUDI2_QUEUE_ID_NIC_2_2] = GAUDI2_EVENT_NIC1_QM0,
[GAUDI2_QUEUE_ID_NIC_2_3] = GAUDI2_EVENT_NIC1_QM0,
[GAUDI2_QUEUE_ID_NIC_3_0] = GAUDI2_EVENT_NIC1_QM1,
[GAUDI2_QUEUE_ID_NIC_3_1] = GAUDI2_EVENT_NIC1_QM1,
[GAUDI2_QUEUE_ID_NIC_3_2] = GAUDI2_EVENT_NIC1_QM1,
[GAUDI2_QUEUE_ID_NIC_3_3] = GAUDI2_EVENT_NIC1_QM1,
[GAUDI2_QUEUE_ID_NIC_4_0] = GAUDI2_EVENT_NIC2_QM0,
[GAUDI2_QUEUE_ID_NIC_4_1] = GAUDI2_EVENT_NIC2_QM0,
[GAUDI2_QUEUE_ID_NIC_4_2] = GAUDI2_EVENT_NIC2_QM0,
[GAUDI2_QUEUE_ID_NIC_4_3] = GAUDI2_EVENT_NIC2_QM0,
[GAUDI2_QUEUE_ID_NIC_5_0] = GAUDI2_EVENT_NIC2_QM1,
[GAUDI2_QUEUE_ID_NIC_5_1] = GAUDI2_EVENT_NIC2_QM1,
[GAUDI2_QUEUE_ID_NIC_5_2] = GAUDI2_EVENT_NIC2_QM1,
[GAUDI2_QUEUE_ID_NIC_5_3] = GAUDI2_EVENT_NIC2_QM1,
[GAUDI2_QUEUE_ID_NIC_6_0] = GAUDI2_EVENT_NIC3_QM0,
[GAUDI2_QUEUE_ID_NIC_6_1] = GAUDI2_EVENT_NIC3_QM0,
[GAUDI2_QUEUE_ID_NIC_6_2] = GAUDI2_EVENT_NIC3_QM0,
[GAUDI2_QUEUE_ID_NIC_6_3] = GAUDI2_EVENT_NIC3_QM0,
[GAUDI2_QUEUE_ID_NIC_7_0] = GAUDI2_EVENT_NIC3_QM1,
[GAUDI2_QUEUE_ID_NIC_7_1] = GAUDI2_EVENT_NIC3_QM1,
[GAUDI2_QUEUE_ID_NIC_7_2] = GAUDI2_EVENT_NIC3_QM1,
[GAUDI2_QUEUE_ID_NIC_7_3] = GAUDI2_EVENT_NIC3_QM1,
[GAUDI2_QUEUE_ID_NIC_8_0] = GAUDI2_EVENT_NIC4_QM0,
[GAUDI2_QUEUE_ID_NIC_8_1] = GAUDI2_EVENT_NIC4_QM0,
[GAUDI2_QUEUE_ID_NIC_8_2] = GAUDI2_EVENT_NIC4_QM0,
[GAUDI2_QUEUE_ID_NIC_8_3] = GAUDI2_EVENT_NIC4_QM0,
[GAUDI2_QUEUE_ID_NIC_9_0] = GAUDI2_EVENT_NIC4_QM1,
[GAUDI2_QUEUE_ID_NIC_9_1] = GAUDI2_EVENT_NIC4_QM1,
[GAUDI2_QUEUE_ID_NIC_9_2] = GAUDI2_EVENT_NIC4_QM1,
[GAUDI2_QUEUE_ID_NIC_9_3] = GAUDI2_EVENT_NIC4_QM1,
[GAUDI2_QUEUE_ID_NIC_10_0] = GAUDI2_EVENT_NIC5_QM0,
[GAUDI2_QUEUE_ID_NIC_10_1] = GAUDI2_EVENT_NIC5_QM0,
[GAUDI2_QUEUE_ID_NIC_10_2] = GAUDI2_EVENT_NIC5_QM0,
[GAUDI2_QUEUE_ID_NIC_10_3] = GAUDI2_EVENT_NIC5_QM0,
[GAUDI2_QUEUE_ID_NIC_11_0] = GAUDI2_EVENT_NIC5_QM1,
[GAUDI2_QUEUE_ID_NIC_11_1] = GAUDI2_EVENT_NIC5_QM1,
[GAUDI2_QUEUE_ID_NIC_11_2] = GAUDI2_EVENT_NIC5_QM1,
[GAUDI2_QUEUE_ID_NIC_11_3] = GAUDI2_EVENT_NIC5_QM1,
[GAUDI2_QUEUE_ID_NIC_12_0] = GAUDI2_EVENT_NIC6_QM0,
[GAUDI2_QUEUE_ID_NIC_12_1] = GAUDI2_EVENT_NIC6_QM0,
[GAUDI2_QUEUE_ID_NIC_12_2] = GAUDI2_EVENT_NIC6_QM0,
[GAUDI2_QUEUE_ID_NIC_12_3] = GAUDI2_EVENT_NIC6_QM0,
[GAUDI2_QUEUE_ID_NIC_13_0] = GAUDI2_EVENT_NIC6_QM1,
[GAUDI2_QUEUE_ID_NIC_13_1] = GAUDI2_EVENT_NIC6_QM1,
[GAUDI2_QUEUE_ID_NIC_13_2] = GAUDI2_EVENT_NIC6_QM1,
[GAUDI2_QUEUE_ID_NIC_13_3] = GAUDI2_EVENT_NIC6_QM1,
[GAUDI2_QUEUE_ID_NIC_14_0] = GAUDI2_EVENT_NIC7_QM0,
[GAUDI2_QUEUE_ID_NIC_14_1] = GAUDI2_EVENT_NIC7_QM0,
[GAUDI2_QUEUE_ID_NIC_14_2] = GAUDI2_EVENT_NIC7_QM0,
[GAUDI2_QUEUE_ID_NIC_14_3] = GAUDI2_EVENT_NIC7_QM0,
[GAUDI2_QUEUE_ID_NIC_15_0] = GAUDI2_EVENT_NIC7_QM1,
[GAUDI2_QUEUE_ID_NIC_15_1] = GAUDI2_EVENT_NIC7_QM1,
[GAUDI2_QUEUE_ID_NIC_15_2] = GAUDI2_EVENT_NIC7_QM1,
[GAUDI2_QUEUE_ID_NIC_15_3] = GAUDI2_EVENT_NIC7_QM1,
[GAUDI2_QUEUE_ID_NIC_16_0] = GAUDI2_EVENT_NIC8_QM0,
[GAUDI2_QUEUE_ID_NIC_16_1] = GAUDI2_EVENT_NIC8_QM0,
[GAUDI2_QUEUE_ID_NIC_16_2] = GAUDI2_EVENT_NIC8_QM0,
[GAUDI2_QUEUE_ID_NIC_16_3] = GAUDI2_EVENT_NIC8_QM0,
[GAUDI2_QUEUE_ID_NIC_17_0] = GAUDI2_EVENT_NIC8_QM1,
[GAUDI2_QUEUE_ID_NIC_17_1] = GAUDI2_EVENT_NIC8_QM1,
[GAUDI2_QUEUE_ID_NIC_17_2] = GAUDI2_EVENT_NIC8_QM1,
[GAUDI2_QUEUE_ID_NIC_17_3] = GAUDI2_EVENT_NIC8_QM1,
[GAUDI2_QUEUE_ID_NIC_18_0] = GAUDI2_EVENT_NIC9_QM0,
[GAUDI2_QUEUE_ID_NIC_18_1] = GAUDI2_EVENT_NIC9_QM0,
[GAUDI2_QUEUE_ID_NIC_18_2] = GAUDI2_EVENT_NIC9_QM0,
[GAUDI2_QUEUE_ID_NIC_18_3] = GAUDI2_EVENT_NIC9_QM0,
[GAUDI2_QUEUE_ID_NIC_19_0] = GAUDI2_EVENT_NIC9_QM1,
[GAUDI2_QUEUE_ID_NIC_19_1] = GAUDI2_EVENT_NIC9_QM1,
[GAUDI2_QUEUE_ID_NIC_19_2] = GAUDI2_EVENT_NIC9_QM1,
[GAUDI2_QUEUE_ID_NIC_19_3] = GAUDI2_EVENT_NIC9_QM1,
[GAUDI2_QUEUE_ID_NIC_20_0] = GAUDI2_EVENT_NIC10_QM0,
[GAUDI2_QUEUE_ID_NIC_20_1] = GAUDI2_EVENT_NIC10_QM0,
[GAUDI2_QUEUE_ID_NIC_20_2] = GAUDI2_EVENT_NIC10_QM0,
[GAUDI2_QUEUE_ID_NIC_20_3] = GAUDI2_EVENT_NIC10_QM0,
[GAUDI2_QUEUE_ID_NIC_21_0] = GAUDI2_EVENT_NIC10_QM1,
[GAUDI2_QUEUE_ID_NIC_21_1] = GAUDI2_EVENT_NIC10_QM1,
[GAUDI2_QUEUE_ID_NIC_21_2] = GAUDI2_EVENT_NIC10_QM1,
[GAUDI2_QUEUE_ID_NIC_21_3] = GAUDI2_EVENT_NIC10_QM1,
[GAUDI2_QUEUE_ID_NIC_22_0] = GAUDI2_EVENT_NIC11_QM0,
[GAUDI2_QUEUE_ID_NIC_22_1] = GAUDI2_EVENT_NIC11_QM0,
[GAUDI2_QUEUE_ID_NIC_22_2] = GAUDI2_EVENT_NIC11_QM0,
[GAUDI2_QUEUE_ID_NIC_22_3] = GAUDI2_EVENT_NIC11_QM0,
[GAUDI2_QUEUE_ID_NIC_23_0] = GAUDI2_EVENT_NIC11_QM1,
[GAUDI2_QUEUE_ID_NIC_23_1] = GAUDI2_EVENT_NIC11_QM1,
[GAUDI2_QUEUE_ID_NIC_23_2] = GAUDI2_EVENT_NIC11_QM1,
[GAUDI2_QUEUE_ID_NIC_23_3] = GAUDI2_EVENT_NIC11_QM1,
[GAUDI2_QUEUE_ID_ROT_0_0] = GAUDI2_EVENT_ROTATOR0_ROT0_QM,
[GAUDI2_QUEUE_ID_ROT_0_1] = GAUDI2_EVENT_ROTATOR0_ROT0_QM,
[GAUDI2_QUEUE_ID_ROT_0_2] = GAUDI2_EVENT_ROTATOR0_ROT0_QM,
[GAUDI2_QUEUE_ID_ROT_0_3] = GAUDI2_EVENT_ROTATOR0_ROT0_QM,
[GAUDI2_QUEUE_ID_ROT_1_0] = GAUDI2_EVENT_ROTATOR1_ROT1_QM,
[GAUDI2_QUEUE_ID_ROT_1_1] = GAUDI2_EVENT_ROTATOR1_ROT1_QM,
[GAUDI2_QUEUE_ID_ROT_1_2] = GAUDI2_EVENT_ROTATOR1_ROT1_QM,
[GAUDI2_QUEUE_ID_ROT_1_3] = GAUDI2_EVENT_ROTATOR1_ROT1_QM
};
static const int gaudi2_dma_core_async_event_id[] = {
[DMA_CORE_ID_EDMA0] = GAUDI2_EVENT_HDMA0_CORE,
[DMA_CORE_ID_EDMA1] = GAUDI2_EVENT_HDMA1_CORE,
[DMA_CORE_ID_EDMA2] = GAUDI2_EVENT_HDMA2_CORE,
[DMA_CORE_ID_EDMA3] = GAUDI2_EVENT_HDMA3_CORE,
[DMA_CORE_ID_EDMA4] = GAUDI2_EVENT_HDMA4_CORE,
[DMA_CORE_ID_EDMA5] = GAUDI2_EVENT_HDMA5_CORE,
[DMA_CORE_ID_EDMA6] = GAUDI2_EVENT_HDMA6_CORE,
[DMA_CORE_ID_EDMA7] = GAUDI2_EVENT_HDMA7_CORE,
[DMA_CORE_ID_PDMA0] = GAUDI2_EVENT_PDMA0_CORE,
[DMA_CORE_ID_PDMA1] = GAUDI2_EVENT_PDMA1_CORE,
[DMA_CORE_ID_KDMA] = GAUDI2_EVENT_KDMA0_CORE,
};
static const char * const gaudi2_qm_sei_error_cause[GAUDI2_NUM_OF_QM_SEI_ERR_CAUSE] = {
"qman sei intr",
"arc sei intr"
};
static const char * const gaudi2_cpu_sei_error_cause[GAUDI2_NUM_OF_CPU_SEI_ERR_CAUSE] = {
"AXI_TERMINATOR WR",
"AXI_TERMINATOR RD",
"AXI SPLIT SEI Status"
};
static const char * const gaudi2_arc_sei_error_cause[GAUDI2_NUM_OF_ARC_SEI_ERR_CAUSE] = {
"cbu_bresp_sei_intr_cause",
"cbu_rresp_sei_intr_cause",
"lbu_bresp_sei_intr_cause",
"lbu_rresp_sei_intr_cause",
"cbu_axi_split_intr_cause",
"lbu_axi_split_intr_cause",
"arc_ip_excptn_sei_intr_cause",
"dmi_bresp_sei_intr_cause",
"aux2apb_err_sei_intr_cause",
"cfg_lbw_wr_terminated_intr_cause",
"cfg_lbw_rd_terminated_intr_cause",
"cfg_dccm_wr_terminated_intr_cause",
"cfg_dccm_rd_terminated_intr_cause",
"cfg_hbw_rd_terminated_intr_cause"
};
static const char * const gaudi2_dec_error_cause[GAUDI2_NUM_OF_DEC_ERR_CAUSE] = {
"msix_vcd_hbw_sei",
"msix_l2c_hbw_sei",
"msix_nrm_hbw_sei",
"msix_abnrm_hbw_sei",
"msix_vcd_lbw_sei",
"msix_l2c_lbw_sei",
"msix_nrm_lbw_sei",
"msix_abnrm_lbw_sei",
"apb_vcd_lbw_sei",
"apb_l2c_lbw_sei",
"apb_nrm_lbw_sei",
"apb_abnrm_lbw_sei",
"dec_sei",
"dec_apb_sei",
"trc_apb_sei",
"lbw_mstr_if_sei",
"axi_split_bresp_err_sei",
"hbw_axi_wr_viol_sei",
"hbw_axi_rd_viol_sei",
"lbw_axi_wr_viol_sei",
"lbw_axi_rd_viol_sei",
"vcd_spi",
"l2c_spi",
"nrm_spi",
"abnrm_spi",
};
static const char * const gaudi2_qman_error_cause[GAUDI2_NUM_OF_QM_ERR_CAUSE] = {
"PQ AXI HBW error",
"CQ AXI HBW error",
"CP AXI HBW error",
"CP error due to undefined OPCODE",
"CP encountered STOP OPCODE",
"CP AXI LBW error",
"CP WRREG32 or WRBULK returned error",
"N/A",
"FENCE 0 inc over max value and clipped",
"FENCE 1 inc over max value and clipped",
"FENCE 2 inc over max value and clipped",
"FENCE 3 inc over max value and clipped",
"FENCE 0 dec under min value and clipped",
"FENCE 1 dec under min value and clipped",
"FENCE 2 dec under min value and clipped",
"FENCE 3 dec under min value and clipped",
"CPDMA Up overflow",
"PQC L2H error"
};
static const char * const gaudi2_qman_lower_cp_error_cause[GAUDI2_NUM_OF_QM_LCP_ERR_CAUSE] = {
"RSVD0",
"CQ AXI HBW error",
"CP AXI HBW error",
"CP error due to undefined OPCODE",
"CP encountered STOP OPCODE",
"CP AXI LBW error",
"CP WRREG32 or WRBULK returned error",
"N/A",
"FENCE 0 inc over max value and clipped",
"FENCE 1 inc over max value and clipped",
"FENCE 2 inc over max value and clipped",
"FENCE 3 inc over max value and clipped",
"FENCE 0 dec under min value and clipped",
"FENCE 1 dec under min value and clipped",
"FENCE 2 dec under min value and clipped",
"FENCE 3 dec under min value and clipped",
"CPDMA Up overflow",
"RSVD17",
"CQ_WR_IFIFO_CI_ERR",
"CQ_WR_CTL_CI_ERR",
"ARC_CQF_RD_ERR",
"ARC_CQ_WR_IFIFO_CI_ERR",
"ARC_CQ_WR_CTL_CI_ERR",
"ARC_AXI_ERR",
"CP_SWITCH_WDT_ERR"
};
static const char * const gaudi2_qman_arb_error_cause[GAUDI2_NUM_OF_QM_ARB_ERR_CAUSE] = {
"Choice push while full error",
"Choice Q watchdog error",
"MSG AXI LBW returned with error"
};
static const char * const guadi2_rot_error_cause[GAUDI2_NUM_OF_ROT_ERR_CAUSE] = {
"qm_axi_err",
"qm_trace_fence_events",
"qm_sw_err",
"qm_cp_sw_stop",
"lbw_mstr_rresp_err",
"lbw_mstr_bresp_err",
"lbw_msg_slverr",
"hbw_msg_slverr",
"wbc_slverr",
"hbw_mstr_rresp_err",
"hbw_mstr_bresp_err",
"sb_resp_intr",
"mrsb_resp_intr",
"core_dw_status_0",
"core_dw_status_1",
"core_dw_status_2",
"core_dw_status_3",
"core_dw_status_4",
"core_dw_status_5",
"core_dw_status_6",
"core_dw_status_7",
"async_arc2cpu_sei_intr",
};
static const char * const gaudi2_tpc_interrupts_cause[GAUDI2_NUM_OF_TPC_INTR_CAUSE] = {
"tpc_address_exceed_slm",
"tpc_div_by_0",
"tpc_spu_mac_overflow",
"tpc_spu_addsub_overflow",
"tpc_spu_abs_overflow",
"tpc_spu_fma_fp_dst_nan",
"tpc_spu_fma_fp_dst_inf",
"tpc_spu_convert_fp_dst_nan",
"tpc_spu_convert_fp_dst_inf",
"tpc_spu_fp_dst_denorm",
"tpc_vpu_mac_overflow",
"tpc_vpu_addsub_overflow",
"tpc_vpu_abs_overflow",
"tpc_vpu_convert_fp_dst_nan",
"tpc_vpu_convert_fp_dst_inf",
"tpc_vpu_fma_fp_dst_nan",
"tpc_vpu_fma_fp_dst_inf",
"tpc_vpu_fp_dst_denorm",
"tpc_assertions",
"tpc_illegal_instruction",
"tpc_pc_wrap_around",
"tpc_qm_sw_err",
"tpc_hbw_rresp_err",
"tpc_hbw_bresp_err",
"tpc_lbw_rresp_err",
"tpc_lbw_bresp_err",
"st_unlock_already_locked",
"invalid_lock_access",
"LD_L protection violation",
"ST_L protection violation",
};
static const char * const guadi2_mme_error_cause[GAUDI2_NUM_OF_MME_ERR_CAUSE] = {
"agu_resp_intr",
"qman_axi_err",
"wap sei (wbc axi err)",
"arc sei",
habanalabs/gaudi: rename mme cfg error response print Current description is misleading hence we rename it to a more suitable error description. Signed-off-by: Ofir Bitton <obitton@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-08-30 20:07:51 +08:00
"cfg access error",
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
"qm_sw_err",
"sbte_dbg_intr_0",
"sbte_dbg_intr_1",
"sbte_dbg_intr_2",
"sbte_dbg_intr_3",
"sbte_dbg_intr_4",
"sbte_prtn_intr_0",
"sbte_prtn_intr_1",
"sbte_prtn_intr_2",
"sbte_prtn_intr_3",
"sbte_prtn_intr_4",
};
static const char * const guadi2_mme_sbte_error_cause[GAUDI2_NUM_OF_MME_SBTE_ERR_CAUSE] = {
"i0",
"i1",
"i2",
"i3",
"i4",
};
static const char * const guadi2_mme_wap_error_cause[GAUDI2_NUM_OF_MME_WAP_ERR_CAUSE] = {
"WBC ERR RESP_0",
"WBC ERR RESP_1",
"AP SOURCE POS INF",
"AP SOURCE NEG INF",
"AP SOURCE NAN",
"AP RESULT POS INF",
"AP RESULT NEG INF",
};
static const char * const gaudi2_dma_core_interrupts_cause[GAUDI2_NUM_OF_DMA_CORE_INTR_CAUSE] = {
"HBW Read returned with error RRESP",
"HBW write returned with error BRESP",
"LBW write returned with error BRESP",
"descriptor_fifo_overflow",
"KDMA SB LBW Read returned with error",
"KDMA WBC LBW Write returned with error",
"TRANSPOSE ENGINE DESC FIFO OVERFLOW",
"WRONG CFG FOR COMMIT IN LIN DMA"
};
static const char * const gaudi2_kdma_core_interrupts_cause[GAUDI2_NUM_OF_DMA_CORE_INTR_CAUSE] = {
"HBW/LBW Read returned with error RRESP",
"HBW/LBW write returned with error BRESP",
"LBW write returned with error BRESP",
"descriptor_fifo_overflow",
"KDMA SB LBW Read returned with error",
"KDMA WBC LBW Write returned with error",
"TRANSPOSE ENGINE DESC FIFO OVERFLOW",
"WRONG CFG FOR COMMIT IN LIN DMA"
};
struct gaudi2_sm_sei_cause_data {
const char *cause_name;
const char *log_name;
u32 log_mask;
};
static const struct gaudi2_sm_sei_cause_data
gaudi2_sm_sei_cause[GAUDI2_NUM_OF_SM_SEI_ERR_CAUSE] = {
{"calculated SO value overflow/underflow", "SOB group ID", 0x7FF},
{"payload address of monitor is not aligned to 4B", "monitor addr", 0xFFFF},
{"armed monitor write got BRESP (SLVERR or DECERR)", "AXI id", 0xFFFF},
};
static const char * const
gaudi2_pmmu_fatal_interrupts_cause[GAUDI2_NUM_OF_PMMU_FATAL_ERR_CAUSE] = {
"LATENCY_RD_OUT_FIFO_OVERRUN",
"LATENCY_WR_OUT_FIFO_OVERRUN",
};
static const char * const
gaudi2_hif_fatal_interrupts_cause[GAUDI2_NUM_OF_HIF_FATAL_ERR_CAUSE] = {
"LATENCY_RD_OUT_FIFO_OVERRUN",
"LATENCY_WR_OUT_FIFO_OVERRUN",
};
static const char * const
gaudi2_psoc_axi_drain_interrupts_cause[GAUDI2_NUM_OF_AXI_DRAIN_ERR_CAUSE] = {
"AXI drain HBW",
"AXI drain LBW",
};
static const char * const
gaudi2_pcie_addr_dec_error_cause[GAUDI2_NUM_OF_PCIE_ADDR_DEC_ERR_CAUSE] = {
"HBW error response",
"LBW error response",
"TLP is blocked by RR"
};
const u32 gaudi2_qm_blocks_bases[GAUDI2_QUEUE_ID_SIZE] = {
[GAUDI2_QUEUE_ID_PDMA_0_0] = mmPDMA0_QM_BASE,
[GAUDI2_QUEUE_ID_PDMA_0_1] = mmPDMA0_QM_BASE,
[GAUDI2_QUEUE_ID_PDMA_0_2] = mmPDMA0_QM_BASE,
[GAUDI2_QUEUE_ID_PDMA_0_3] = mmPDMA0_QM_BASE,
[GAUDI2_QUEUE_ID_PDMA_1_0] = mmPDMA1_QM_BASE,
[GAUDI2_QUEUE_ID_PDMA_1_1] = mmPDMA1_QM_BASE,
[GAUDI2_QUEUE_ID_PDMA_1_2] = mmPDMA1_QM_BASE,
[GAUDI2_QUEUE_ID_PDMA_1_3] = mmPDMA1_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE0_EDMA_0_0] = mmDCORE0_EDMA0_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE0_EDMA_0_1] = mmDCORE0_EDMA0_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE0_EDMA_0_2] = mmDCORE0_EDMA0_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE0_EDMA_0_3] = mmDCORE0_EDMA0_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE0_EDMA_1_0] = mmDCORE0_EDMA1_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE0_EDMA_1_1] = mmDCORE0_EDMA1_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE0_EDMA_1_2] = mmDCORE0_EDMA1_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE0_EDMA_1_3] = mmDCORE0_EDMA1_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE0_MME_0_0] = mmDCORE0_MME_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE0_MME_0_1] = mmDCORE0_MME_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE0_MME_0_2] = mmDCORE0_MME_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE0_MME_0_3] = mmDCORE0_MME_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE0_TPC_0_0] = mmDCORE0_TPC0_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE0_TPC_0_1] = mmDCORE0_TPC0_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE0_TPC_0_2] = mmDCORE0_TPC0_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE0_TPC_0_3] = mmDCORE0_TPC0_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE0_TPC_1_0] = mmDCORE0_TPC1_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE0_TPC_1_1] = mmDCORE0_TPC1_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE0_TPC_1_2] = mmDCORE0_TPC1_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE0_TPC_1_3] = mmDCORE0_TPC1_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE0_TPC_2_0] = mmDCORE0_TPC2_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE0_TPC_2_1] = mmDCORE0_TPC2_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE0_TPC_2_2] = mmDCORE0_TPC2_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE0_TPC_2_3] = mmDCORE0_TPC2_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE0_TPC_3_0] = mmDCORE0_TPC3_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE0_TPC_3_1] = mmDCORE0_TPC3_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE0_TPC_3_2] = mmDCORE0_TPC3_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE0_TPC_3_3] = mmDCORE0_TPC3_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE0_TPC_4_0] = mmDCORE0_TPC4_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE0_TPC_4_1] = mmDCORE0_TPC4_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE0_TPC_4_2] = mmDCORE0_TPC4_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE0_TPC_4_3] = mmDCORE0_TPC4_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE0_TPC_5_0] = mmDCORE0_TPC5_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE0_TPC_5_1] = mmDCORE0_TPC5_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE0_TPC_5_2] = mmDCORE0_TPC5_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE0_TPC_5_3] = mmDCORE0_TPC5_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE0_TPC_6_0] = mmDCORE0_TPC6_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE0_TPC_6_1] = mmDCORE0_TPC6_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE0_TPC_6_2] = mmDCORE0_TPC6_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE0_TPC_6_3] = mmDCORE0_TPC6_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE1_EDMA_0_0] = mmDCORE1_EDMA0_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE1_EDMA_0_1] = mmDCORE1_EDMA0_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE1_EDMA_0_2] = mmDCORE1_EDMA0_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE1_EDMA_0_3] = mmDCORE1_EDMA0_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE1_EDMA_1_0] = mmDCORE1_EDMA1_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE1_EDMA_1_1] = mmDCORE1_EDMA1_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE1_EDMA_1_2] = mmDCORE1_EDMA1_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE1_EDMA_1_3] = mmDCORE1_EDMA1_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE1_MME_0_0] = mmDCORE1_MME_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE1_MME_0_1] = mmDCORE1_MME_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE1_MME_0_2] = mmDCORE1_MME_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE1_MME_0_3] = mmDCORE1_MME_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE1_TPC_0_0] = mmDCORE1_TPC0_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE1_TPC_0_1] = mmDCORE1_TPC0_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE1_TPC_0_2] = mmDCORE1_TPC0_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE1_TPC_0_3] = mmDCORE1_TPC0_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE1_TPC_1_0] = mmDCORE1_TPC1_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE1_TPC_1_1] = mmDCORE1_TPC1_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE1_TPC_1_2] = mmDCORE1_TPC1_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE1_TPC_1_3] = mmDCORE1_TPC1_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE1_TPC_2_0] = mmDCORE1_TPC2_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE1_TPC_2_1] = mmDCORE1_TPC2_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE1_TPC_2_2] = mmDCORE1_TPC2_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE1_TPC_2_3] = mmDCORE1_TPC2_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE1_TPC_3_0] = mmDCORE1_TPC3_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE1_TPC_3_1] = mmDCORE1_TPC3_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE1_TPC_3_2] = mmDCORE1_TPC3_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE1_TPC_3_3] = mmDCORE1_TPC3_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE1_TPC_4_0] = mmDCORE1_TPC4_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE1_TPC_4_1] = mmDCORE1_TPC4_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE1_TPC_4_2] = mmDCORE1_TPC4_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE1_TPC_4_3] = mmDCORE1_TPC4_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE1_TPC_5_0] = mmDCORE1_TPC5_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE1_TPC_5_1] = mmDCORE1_TPC5_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE1_TPC_5_2] = mmDCORE1_TPC5_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE1_TPC_5_3] = mmDCORE1_TPC5_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE2_EDMA_0_0] = mmDCORE2_EDMA0_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE2_EDMA_0_1] = mmDCORE2_EDMA0_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE2_EDMA_0_2] = mmDCORE2_EDMA0_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE2_EDMA_0_3] = mmDCORE2_EDMA0_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE2_EDMA_1_0] = mmDCORE2_EDMA1_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE2_EDMA_1_1] = mmDCORE2_EDMA1_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE2_EDMA_1_2] = mmDCORE2_EDMA1_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE2_EDMA_1_3] = mmDCORE2_EDMA1_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE2_MME_0_0] = mmDCORE2_MME_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE2_MME_0_1] = mmDCORE2_MME_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE2_MME_0_2] = mmDCORE2_MME_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE2_MME_0_3] = mmDCORE2_MME_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE2_TPC_0_0] = mmDCORE2_TPC0_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE2_TPC_0_1] = mmDCORE2_TPC0_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE2_TPC_0_2] = mmDCORE2_TPC0_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE2_TPC_0_3] = mmDCORE2_TPC0_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE2_TPC_1_0] = mmDCORE2_TPC1_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE2_TPC_1_1] = mmDCORE2_TPC1_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE2_TPC_1_2] = mmDCORE2_TPC1_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE2_TPC_1_3] = mmDCORE2_TPC1_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE2_TPC_2_0] = mmDCORE2_TPC2_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE2_TPC_2_1] = mmDCORE2_TPC2_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE2_TPC_2_2] = mmDCORE2_TPC2_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE2_TPC_2_3] = mmDCORE2_TPC2_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE2_TPC_3_0] = mmDCORE2_TPC3_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE2_TPC_3_1] = mmDCORE2_TPC3_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE2_TPC_3_2] = mmDCORE2_TPC3_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE2_TPC_3_3] = mmDCORE2_TPC3_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE2_TPC_4_0] = mmDCORE2_TPC4_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE2_TPC_4_1] = mmDCORE2_TPC4_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE2_TPC_4_2] = mmDCORE2_TPC4_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE2_TPC_4_3] = mmDCORE2_TPC4_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE2_TPC_5_0] = mmDCORE2_TPC5_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE2_TPC_5_1] = mmDCORE2_TPC5_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE2_TPC_5_2] = mmDCORE2_TPC5_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE2_TPC_5_3] = mmDCORE2_TPC5_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE3_EDMA_0_0] = mmDCORE3_EDMA0_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE3_EDMA_0_1] = mmDCORE3_EDMA0_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE3_EDMA_0_2] = mmDCORE3_EDMA0_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE3_EDMA_0_3] = mmDCORE3_EDMA0_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE3_EDMA_1_0] = mmDCORE3_EDMA1_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE3_EDMA_1_1] = mmDCORE3_EDMA1_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE3_EDMA_1_2] = mmDCORE3_EDMA1_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE3_EDMA_1_3] = mmDCORE3_EDMA1_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE3_MME_0_0] = mmDCORE3_MME_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE3_MME_0_1] = mmDCORE3_MME_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE3_MME_0_2] = mmDCORE3_MME_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE3_MME_0_3] = mmDCORE3_MME_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE3_TPC_0_0] = mmDCORE3_TPC0_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE3_TPC_0_1] = mmDCORE3_TPC0_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE3_TPC_0_2] = mmDCORE3_TPC0_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE3_TPC_0_3] = mmDCORE3_TPC0_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE3_TPC_1_0] = mmDCORE3_TPC1_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE3_TPC_1_1] = mmDCORE3_TPC1_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE3_TPC_1_2] = mmDCORE3_TPC1_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE3_TPC_1_3] = mmDCORE3_TPC1_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE3_TPC_2_0] = mmDCORE3_TPC2_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE3_TPC_2_1] = mmDCORE3_TPC2_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE3_TPC_2_2] = mmDCORE3_TPC2_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE3_TPC_2_3] = mmDCORE3_TPC2_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE3_TPC_3_0] = mmDCORE3_TPC3_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE3_TPC_3_1] = mmDCORE3_TPC3_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE3_TPC_3_2] = mmDCORE3_TPC3_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE3_TPC_3_3] = mmDCORE3_TPC3_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE3_TPC_4_0] = mmDCORE3_TPC4_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE3_TPC_4_1] = mmDCORE3_TPC4_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE3_TPC_4_2] = mmDCORE3_TPC4_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE3_TPC_4_3] = mmDCORE3_TPC4_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE3_TPC_5_0] = mmDCORE3_TPC5_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE3_TPC_5_1] = mmDCORE3_TPC5_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE3_TPC_5_2] = mmDCORE3_TPC5_QM_BASE,
[GAUDI2_QUEUE_ID_DCORE3_TPC_5_3] = mmDCORE3_TPC5_QM_BASE,
[GAUDI2_QUEUE_ID_NIC_0_0] = mmNIC0_QM0_BASE,
[GAUDI2_QUEUE_ID_NIC_0_1] = mmNIC0_QM0_BASE,
[GAUDI2_QUEUE_ID_NIC_0_2] = mmNIC0_QM0_BASE,
[GAUDI2_QUEUE_ID_NIC_0_3] = mmNIC0_QM0_BASE,
[GAUDI2_QUEUE_ID_NIC_1_0] = mmNIC0_QM1_BASE,
[GAUDI2_QUEUE_ID_NIC_1_1] = mmNIC0_QM1_BASE,
[GAUDI2_QUEUE_ID_NIC_1_2] = mmNIC0_QM1_BASE,
[GAUDI2_QUEUE_ID_NIC_1_3] = mmNIC0_QM1_BASE,
[GAUDI2_QUEUE_ID_NIC_2_0] = mmNIC1_QM0_BASE,
[GAUDI2_QUEUE_ID_NIC_2_1] = mmNIC1_QM0_BASE,
[GAUDI2_QUEUE_ID_NIC_2_2] = mmNIC1_QM0_BASE,
[GAUDI2_QUEUE_ID_NIC_2_3] = mmNIC1_QM0_BASE,
[GAUDI2_QUEUE_ID_NIC_3_0] = mmNIC1_QM1_BASE,
[GAUDI2_QUEUE_ID_NIC_3_1] = mmNIC1_QM1_BASE,
[GAUDI2_QUEUE_ID_NIC_3_2] = mmNIC1_QM1_BASE,
[GAUDI2_QUEUE_ID_NIC_3_3] = mmNIC1_QM1_BASE,
[GAUDI2_QUEUE_ID_NIC_4_0] = mmNIC2_QM0_BASE,
[GAUDI2_QUEUE_ID_NIC_4_1] = mmNIC2_QM0_BASE,
[GAUDI2_QUEUE_ID_NIC_4_2] = mmNIC2_QM0_BASE,
[GAUDI2_QUEUE_ID_NIC_4_3] = mmNIC2_QM0_BASE,
[GAUDI2_QUEUE_ID_NIC_5_0] = mmNIC2_QM1_BASE,
[GAUDI2_QUEUE_ID_NIC_5_1] = mmNIC2_QM1_BASE,
[GAUDI2_QUEUE_ID_NIC_5_2] = mmNIC2_QM1_BASE,
[GAUDI2_QUEUE_ID_NIC_5_3] = mmNIC2_QM1_BASE,
[GAUDI2_QUEUE_ID_NIC_6_0] = mmNIC3_QM0_BASE,
[GAUDI2_QUEUE_ID_NIC_6_1] = mmNIC3_QM0_BASE,
[GAUDI2_QUEUE_ID_NIC_6_2] = mmNIC3_QM0_BASE,
[GAUDI2_QUEUE_ID_NIC_6_3] = mmNIC3_QM0_BASE,
[GAUDI2_QUEUE_ID_NIC_7_0] = mmNIC3_QM1_BASE,
[GAUDI2_QUEUE_ID_NIC_7_1] = mmNIC3_QM1_BASE,
[GAUDI2_QUEUE_ID_NIC_7_2] = mmNIC3_QM1_BASE,
[GAUDI2_QUEUE_ID_NIC_7_3] = mmNIC3_QM1_BASE,
[GAUDI2_QUEUE_ID_NIC_8_0] = mmNIC4_QM0_BASE,
[GAUDI2_QUEUE_ID_NIC_8_1] = mmNIC4_QM0_BASE,
[GAUDI2_QUEUE_ID_NIC_8_2] = mmNIC4_QM0_BASE,
[GAUDI2_QUEUE_ID_NIC_8_3] = mmNIC4_QM0_BASE,
[GAUDI2_QUEUE_ID_NIC_9_0] = mmNIC4_QM1_BASE,
[GAUDI2_QUEUE_ID_NIC_9_1] = mmNIC4_QM1_BASE,
[GAUDI2_QUEUE_ID_NIC_9_2] = mmNIC4_QM1_BASE,
[GAUDI2_QUEUE_ID_NIC_9_3] = mmNIC4_QM1_BASE,
[GAUDI2_QUEUE_ID_NIC_10_0] = mmNIC5_QM0_BASE,
[GAUDI2_QUEUE_ID_NIC_10_1] = mmNIC5_QM0_BASE,
[GAUDI2_QUEUE_ID_NIC_10_2] = mmNIC5_QM0_BASE,
[GAUDI2_QUEUE_ID_NIC_10_3] = mmNIC5_QM0_BASE,
[GAUDI2_QUEUE_ID_NIC_11_0] = mmNIC5_QM1_BASE,
[GAUDI2_QUEUE_ID_NIC_11_1] = mmNIC5_QM1_BASE,
[GAUDI2_QUEUE_ID_NIC_11_2] = mmNIC5_QM1_BASE,
[GAUDI2_QUEUE_ID_NIC_11_3] = mmNIC5_QM1_BASE,
[GAUDI2_QUEUE_ID_NIC_12_0] = mmNIC6_QM0_BASE,
[GAUDI2_QUEUE_ID_NIC_12_1] = mmNIC6_QM0_BASE,
[GAUDI2_QUEUE_ID_NIC_12_2] = mmNIC6_QM0_BASE,
[GAUDI2_QUEUE_ID_NIC_12_3] = mmNIC6_QM0_BASE,
[GAUDI2_QUEUE_ID_NIC_13_0] = mmNIC6_QM1_BASE,
[GAUDI2_QUEUE_ID_NIC_13_1] = mmNIC6_QM1_BASE,
[GAUDI2_QUEUE_ID_NIC_13_2] = mmNIC6_QM1_BASE,
[GAUDI2_QUEUE_ID_NIC_13_3] = mmNIC6_QM1_BASE,
[GAUDI2_QUEUE_ID_NIC_14_0] = mmNIC7_QM0_BASE,
[GAUDI2_QUEUE_ID_NIC_14_1] = mmNIC7_QM0_BASE,
[GAUDI2_QUEUE_ID_NIC_14_2] = mmNIC7_QM0_BASE,
[GAUDI2_QUEUE_ID_NIC_14_3] = mmNIC7_QM0_BASE,
[GAUDI2_QUEUE_ID_NIC_15_0] = mmNIC7_QM1_BASE,
[GAUDI2_QUEUE_ID_NIC_15_1] = mmNIC7_QM1_BASE,
[GAUDI2_QUEUE_ID_NIC_15_2] = mmNIC7_QM1_BASE,
[GAUDI2_QUEUE_ID_NIC_15_3] = mmNIC7_QM1_BASE,
[GAUDI2_QUEUE_ID_NIC_16_0] = mmNIC8_QM0_BASE,
[GAUDI2_QUEUE_ID_NIC_16_1] = mmNIC8_QM0_BASE,
[GAUDI2_QUEUE_ID_NIC_16_2] = mmNIC8_QM0_BASE,
[GAUDI2_QUEUE_ID_NIC_16_3] = mmNIC8_QM0_BASE,
[GAUDI2_QUEUE_ID_NIC_17_0] = mmNIC8_QM1_BASE,
[GAUDI2_QUEUE_ID_NIC_17_1] = mmNIC8_QM1_BASE,
[GAUDI2_QUEUE_ID_NIC_17_2] = mmNIC8_QM1_BASE,
[GAUDI2_QUEUE_ID_NIC_17_3] = mmNIC8_QM1_BASE,
[GAUDI2_QUEUE_ID_NIC_18_0] = mmNIC9_QM0_BASE,
[GAUDI2_QUEUE_ID_NIC_18_1] = mmNIC9_QM0_BASE,
[GAUDI2_QUEUE_ID_NIC_18_2] = mmNIC9_QM0_BASE,
[GAUDI2_QUEUE_ID_NIC_18_3] = mmNIC9_QM0_BASE,
[GAUDI2_QUEUE_ID_NIC_19_0] = mmNIC9_QM1_BASE,
[GAUDI2_QUEUE_ID_NIC_19_1] = mmNIC9_QM1_BASE,
[GAUDI2_QUEUE_ID_NIC_19_2] = mmNIC9_QM1_BASE,
[GAUDI2_QUEUE_ID_NIC_19_3] = mmNIC9_QM1_BASE,
[GAUDI2_QUEUE_ID_NIC_20_0] = mmNIC10_QM0_BASE,
[GAUDI2_QUEUE_ID_NIC_20_1] = mmNIC10_QM0_BASE,
[GAUDI2_QUEUE_ID_NIC_20_2] = mmNIC10_QM0_BASE,
[GAUDI2_QUEUE_ID_NIC_20_3] = mmNIC10_QM0_BASE,
[GAUDI2_QUEUE_ID_NIC_21_0] = mmNIC10_QM1_BASE,
[GAUDI2_QUEUE_ID_NIC_21_1] = mmNIC10_QM1_BASE,
[GAUDI2_QUEUE_ID_NIC_21_2] = mmNIC10_QM1_BASE,
[GAUDI2_QUEUE_ID_NIC_21_3] = mmNIC10_QM1_BASE,
[GAUDI2_QUEUE_ID_NIC_22_0] = mmNIC11_QM0_BASE,
[GAUDI2_QUEUE_ID_NIC_22_1] = mmNIC11_QM0_BASE,
[GAUDI2_QUEUE_ID_NIC_22_2] = mmNIC11_QM0_BASE,
[GAUDI2_QUEUE_ID_NIC_22_3] = mmNIC11_QM0_BASE,
[GAUDI2_QUEUE_ID_NIC_23_0] = mmNIC11_QM1_BASE,
[GAUDI2_QUEUE_ID_NIC_23_1] = mmNIC11_QM1_BASE,
[GAUDI2_QUEUE_ID_NIC_23_2] = mmNIC11_QM1_BASE,
[GAUDI2_QUEUE_ID_NIC_23_3] = mmNIC11_QM1_BASE,
[GAUDI2_QUEUE_ID_ROT_0_0] = mmROT0_QM_BASE,
[GAUDI2_QUEUE_ID_ROT_0_1] = mmROT0_QM_BASE,
[GAUDI2_QUEUE_ID_ROT_0_2] = mmROT0_QM_BASE,
[GAUDI2_QUEUE_ID_ROT_0_3] = mmROT0_QM_BASE,
[GAUDI2_QUEUE_ID_ROT_1_0] = mmROT1_QM_BASE,
[GAUDI2_QUEUE_ID_ROT_1_1] = mmROT1_QM_BASE,
[GAUDI2_QUEUE_ID_ROT_1_2] = mmROT1_QM_BASE,
[GAUDI2_QUEUE_ID_ROT_1_3] = mmROT1_QM_BASE
};
static const u32 gaudi2_arc_blocks_bases[NUM_ARC_CPUS] = {
[CPU_ID_SCHED_ARC0] = mmARC_FARM_ARC0_AUX_BASE,
[CPU_ID_SCHED_ARC1] = mmARC_FARM_ARC1_AUX_BASE,
[CPU_ID_SCHED_ARC2] = mmARC_FARM_ARC2_AUX_BASE,
[CPU_ID_SCHED_ARC3] = mmARC_FARM_ARC3_AUX_BASE,
[CPU_ID_SCHED_ARC4] = mmDCORE1_MME_QM_ARC_AUX_BASE,
[CPU_ID_SCHED_ARC5] = mmDCORE3_MME_QM_ARC_AUX_BASE,
[CPU_ID_TPC_QMAN_ARC0] = mmDCORE0_TPC0_QM_ARC_AUX_BASE,
[CPU_ID_TPC_QMAN_ARC1] = mmDCORE0_TPC1_QM_ARC_AUX_BASE,
[CPU_ID_TPC_QMAN_ARC2] = mmDCORE0_TPC2_QM_ARC_AUX_BASE,
[CPU_ID_TPC_QMAN_ARC3] = mmDCORE0_TPC3_QM_ARC_AUX_BASE,
[CPU_ID_TPC_QMAN_ARC4] = mmDCORE0_TPC4_QM_ARC_AUX_BASE,
[CPU_ID_TPC_QMAN_ARC5] = mmDCORE0_TPC5_QM_ARC_AUX_BASE,
[CPU_ID_TPC_QMAN_ARC6] = mmDCORE1_TPC0_QM_ARC_AUX_BASE,
[CPU_ID_TPC_QMAN_ARC7] = mmDCORE1_TPC1_QM_ARC_AUX_BASE,
[CPU_ID_TPC_QMAN_ARC8] = mmDCORE1_TPC2_QM_ARC_AUX_BASE,
[CPU_ID_TPC_QMAN_ARC9] = mmDCORE1_TPC3_QM_ARC_AUX_BASE,
[CPU_ID_TPC_QMAN_ARC10] = mmDCORE1_TPC4_QM_ARC_AUX_BASE,
[CPU_ID_TPC_QMAN_ARC11] = mmDCORE1_TPC5_QM_ARC_AUX_BASE,
[CPU_ID_TPC_QMAN_ARC12] = mmDCORE2_TPC0_QM_ARC_AUX_BASE,
[CPU_ID_TPC_QMAN_ARC13] = mmDCORE2_TPC1_QM_ARC_AUX_BASE,
[CPU_ID_TPC_QMAN_ARC14] = mmDCORE2_TPC2_QM_ARC_AUX_BASE,
[CPU_ID_TPC_QMAN_ARC15] = mmDCORE2_TPC3_QM_ARC_AUX_BASE,
[CPU_ID_TPC_QMAN_ARC16] = mmDCORE2_TPC4_QM_ARC_AUX_BASE,
[CPU_ID_TPC_QMAN_ARC17] = mmDCORE2_TPC5_QM_ARC_AUX_BASE,
[CPU_ID_TPC_QMAN_ARC18] = mmDCORE3_TPC0_QM_ARC_AUX_BASE,
[CPU_ID_TPC_QMAN_ARC19] = mmDCORE3_TPC1_QM_ARC_AUX_BASE,
[CPU_ID_TPC_QMAN_ARC20] = mmDCORE3_TPC2_QM_ARC_AUX_BASE,
[CPU_ID_TPC_QMAN_ARC21] = mmDCORE3_TPC3_QM_ARC_AUX_BASE,
[CPU_ID_TPC_QMAN_ARC22] = mmDCORE3_TPC4_QM_ARC_AUX_BASE,
[CPU_ID_TPC_QMAN_ARC23] = mmDCORE3_TPC5_QM_ARC_AUX_BASE,
[CPU_ID_TPC_QMAN_ARC24] = mmDCORE0_TPC6_QM_ARC_AUX_BASE,
[CPU_ID_MME_QMAN_ARC0] = mmDCORE0_MME_QM_ARC_AUX_BASE,
[CPU_ID_MME_QMAN_ARC1] = mmDCORE2_MME_QM_ARC_AUX_BASE,
[CPU_ID_EDMA_QMAN_ARC0] = mmDCORE0_EDMA0_QM_ARC_AUX_BASE,
[CPU_ID_EDMA_QMAN_ARC1] = mmDCORE0_EDMA1_QM_ARC_AUX_BASE,
[CPU_ID_EDMA_QMAN_ARC2] = mmDCORE1_EDMA0_QM_ARC_AUX_BASE,
[CPU_ID_EDMA_QMAN_ARC3] = mmDCORE1_EDMA1_QM_ARC_AUX_BASE,
[CPU_ID_EDMA_QMAN_ARC4] = mmDCORE2_EDMA0_QM_ARC_AUX_BASE,
[CPU_ID_EDMA_QMAN_ARC5] = mmDCORE2_EDMA1_QM_ARC_AUX_BASE,
[CPU_ID_EDMA_QMAN_ARC6] = mmDCORE3_EDMA0_QM_ARC_AUX_BASE,
[CPU_ID_EDMA_QMAN_ARC7] = mmDCORE3_EDMA1_QM_ARC_AUX_BASE,
[CPU_ID_PDMA_QMAN_ARC0] = mmPDMA0_QM_ARC_AUX_BASE,
[CPU_ID_PDMA_QMAN_ARC1] = mmPDMA1_QM_ARC_AUX_BASE,
[CPU_ID_ROT_QMAN_ARC0] = mmROT0_QM_ARC_AUX_BASE,
[CPU_ID_ROT_QMAN_ARC1] = mmROT1_QM_ARC_AUX_BASE,
[CPU_ID_NIC_QMAN_ARC0] = mmNIC0_QM_ARC_AUX0_BASE,
[CPU_ID_NIC_QMAN_ARC1] = mmNIC0_QM_ARC_AUX1_BASE,
[CPU_ID_NIC_QMAN_ARC2] = mmNIC1_QM_ARC_AUX0_BASE,
[CPU_ID_NIC_QMAN_ARC3] = mmNIC1_QM_ARC_AUX1_BASE,
[CPU_ID_NIC_QMAN_ARC4] = mmNIC2_QM_ARC_AUX0_BASE,
[CPU_ID_NIC_QMAN_ARC5] = mmNIC2_QM_ARC_AUX1_BASE,
[CPU_ID_NIC_QMAN_ARC6] = mmNIC3_QM_ARC_AUX0_BASE,
[CPU_ID_NIC_QMAN_ARC7] = mmNIC3_QM_ARC_AUX1_BASE,
[CPU_ID_NIC_QMAN_ARC8] = mmNIC4_QM_ARC_AUX0_BASE,
[CPU_ID_NIC_QMAN_ARC9] = mmNIC4_QM_ARC_AUX1_BASE,
[CPU_ID_NIC_QMAN_ARC10] = mmNIC5_QM_ARC_AUX0_BASE,
[CPU_ID_NIC_QMAN_ARC11] = mmNIC5_QM_ARC_AUX1_BASE,
[CPU_ID_NIC_QMAN_ARC12] = mmNIC6_QM_ARC_AUX0_BASE,
[CPU_ID_NIC_QMAN_ARC13] = mmNIC6_QM_ARC_AUX1_BASE,
[CPU_ID_NIC_QMAN_ARC14] = mmNIC7_QM_ARC_AUX0_BASE,
[CPU_ID_NIC_QMAN_ARC15] = mmNIC7_QM_ARC_AUX1_BASE,
[CPU_ID_NIC_QMAN_ARC16] = mmNIC8_QM_ARC_AUX0_BASE,
[CPU_ID_NIC_QMAN_ARC17] = mmNIC8_QM_ARC_AUX1_BASE,
[CPU_ID_NIC_QMAN_ARC18] = mmNIC9_QM_ARC_AUX0_BASE,
[CPU_ID_NIC_QMAN_ARC19] = mmNIC9_QM_ARC_AUX1_BASE,
[CPU_ID_NIC_QMAN_ARC20] = mmNIC10_QM_ARC_AUX0_BASE,
[CPU_ID_NIC_QMAN_ARC21] = mmNIC10_QM_ARC_AUX1_BASE,
[CPU_ID_NIC_QMAN_ARC22] = mmNIC11_QM_ARC_AUX0_BASE,
[CPU_ID_NIC_QMAN_ARC23] = mmNIC11_QM_ARC_AUX1_BASE,
};
static const u32 gaudi2_arc_dccm_bases[NUM_ARC_CPUS] = {
[CPU_ID_SCHED_ARC0] = mmARC_FARM_ARC0_DCCM0_BASE,
[CPU_ID_SCHED_ARC1] = mmARC_FARM_ARC1_DCCM0_BASE,
[CPU_ID_SCHED_ARC2] = mmARC_FARM_ARC2_DCCM0_BASE,
[CPU_ID_SCHED_ARC3] = mmARC_FARM_ARC3_DCCM0_BASE,
[CPU_ID_SCHED_ARC4] = mmDCORE1_MME_QM_ARC_DCCM_BASE,
[CPU_ID_SCHED_ARC5] = mmDCORE3_MME_QM_ARC_DCCM_BASE,
[CPU_ID_TPC_QMAN_ARC0] = mmDCORE0_TPC0_QM_DCCM_BASE,
[CPU_ID_TPC_QMAN_ARC1] = mmDCORE0_TPC1_QM_DCCM_BASE,
[CPU_ID_TPC_QMAN_ARC2] = mmDCORE0_TPC2_QM_DCCM_BASE,
[CPU_ID_TPC_QMAN_ARC3] = mmDCORE0_TPC3_QM_DCCM_BASE,
[CPU_ID_TPC_QMAN_ARC4] = mmDCORE0_TPC4_QM_DCCM_BASE,
[CPU_ID_TPC_QMAN_ARC5] = mmDCORE0_TPC5_QM_DCCM_BASE,
[CPU_ID_TPC_QMAN_ARC6] = mmDCORE1_TPC0_QM_DCCM_BASE,
[CPU_ID_TPC_QMAN_ARC7] = mmDCORE1_TPC1_QM_DCCM_BASE,
[CPU_ID_TPC_QMAN_ARC8] = mmDCORE1_TPC2_QM_DCCM_BASE,
[CPU_ID_TPC_QMAN_ARC9] = mmDCORE1_TPC3_QM_DCCM_BASE,
[CPU_ID_TPC_QMAN_ARC10] = mmDCORE1_TPC4_QM_DCCM_BASE,
[CPU_ID_TPC_QMAN_ARC11] = mmDCORE1_TPC5_QM_DCCM_BASE,
[CPU_ID_TPC_QMAN_ARC12] = mmDCORE2_TPC0_QM_DCCM_BASE,
[CPU_ID_TPC_QMAN_ARC13] = mmDCORE2_TPC1_QM_DCCM_BASE,
[CPU_ID_TPC_QMAN_ARC14] = mmDCORE2_TPC2_QM_DCCM_BASE,
[CPU_ID_TPC_QMAN_ARC15] = mmDCORE2_TPC3_QM_DCCM_BASE,
[CPU_ID_TPC_QMAN_ARC16] = mmDCORE2_TPC4_QM_DCCM_BASE,
[CPU_ID_TPC_QMAN_ARC17] = mmDCORE2_TPC5_QM_DCCM_BASE,
[CPU_ID_TPC_QMAN_ARC18] = mmDCORE3_TPC0_QM_DCCM_BASE,
[CPU_ID_TPC_QMAN_ARC19] = mmDCORE3_TPC1_QM_DCCM_BASE,
[CPU_ID_TPC_QMAN_ARC20] = mmDCORE3_TPC2_QM_DCCM_BASE,
[CPU_ID_TPC_QMAN_ARC21] = mmDCORE3_TPC3_QM_DCCM_BASE,
[CPU_ID_TPC_QMAN_ARC22] = mmDCORE3_TPC4_QM_DCCM_BASE,
[CPU_ID_TPC_QMAN_ARC23] = mmDCORE3_TPC5_QM_DCCM_BASE,
[CPU_ID_TPC_QMAN_ARC24] = mmDCORE0_TPC6_QM_DCCM_BASE,
[CPU_ID_MME_QMAN_ARC0] = mmDCORE0_MME_QM_ARC_DCCM_BASE,
[CPU_ID_MME_QMAN_ARC1] = mmDCORE2_MME_QM_ARC_DCCM_BASE,
[CPU_ID_EDMA_QMAN_ARC0] = mmDCORE0_EDMA0_QM_DCCM_BASE,
[CPU_ID_EDMA_QMAN_ARC1] = mmDCORE0_EDMA1_QM_DCCM_BASE,
[CPU_ID_EDMA_QMAN_ARC2] = mmDCORE1_EDMA0_QM_DCCM_BASE,
[CPU_ID_EDMA_QMAN_ARC3] = mmDCORE1_EDMA1_QM_DCCM_BASE,
[CPU_ID_EDMA_QMAN_ARC4] = mmDCORE2_EDMA0_QM_DCCM_BASE,
[CPU_ID_EDMA_QMAN_ARC5] = mmDCORE2_EDMA1_QM_DCCM_BASE,
[CPU_ID_EDMA_QMAN_ARC6] = mmDCORE3_EDMA0_QM_DCCM_BASE,
[CPU_ID_EDMA_QMAN_ARC7] = mmDCORE3_EDMA1_QM_DCCM_BASE,
[CPU_ID_PDMA_QMAN_ARC0] = mmPDMA0_QM_ARC_DCCM_BASE,
[CPU_ID_PDMA_QMAN_ARC1] = mmPDMA1_QM_ARC_DCCM_BASE,
[CPU_ID_ROT_QMAN_ARC0] = mmROT0_QM_ARC_DCCM_BASE,
[CPU_ID_ROT_QMAN_ARC1] = mmROT1_QM_ARC_DCCM_BASE,
[CPU_ID_NIC_QMAN_ARC0] = mmNIC0_QM_DCCM0_BASE,
[CPU_ID_NIC_QMAN_ARC1] = mmNIC0_QM_DCCM1_BASE,
[CPU_ID_NIC_QMAN_ARC2] = mmNIC1_QM_DCCM0_BASE,
[CPU_ID_NIC_QMAN_ARC3] = mmNIC1_QM_DCCM1_BASE,
[CPU_ID_NIC_QMAN_ARC4] = mmNIC2_QM_DCCM0_BASE,
[CPU_ID_NIC_QMAN_ARC5] = mmNIC2_QM_DCCM1_BASE,
[CPU_ID_NIC_QMAN_ARC6] = mmNIC3_QM_DCCM0_BASE,
[CPU_ID_NIC_QMAN_ARC7] = mmNIC3_QM_DCCM1_BASE,
[CPU_ID_NIC_QMAN_ARC8] = mmNIC4_QM_DCCM0_BASE,
[CPU_ID_NIC_QMAN_ARC9] = mmNIC4_QM_DCCM1_BASE,
[CPU_ID_NIC_QMAN_ARC10] = mmNIC5_QM_DCCM0_BASE,
[CPU_ID_NIC_QMAN_ARC11] = mmNIC5_QM_DCCM1_BASE,
[CPU_ID_NIC_QMAN_ARC12] = mmNIC6_QM_DCCM0_BASE,
[CPU_ID_NIC_QMAN_ARC13] = mmNIC6_QM_DCCM1_BASE,
[CPU_ID_NIC_QMAN_ARC14] = mmNIC7_QM_DCCM0_BASE,
[CPU_ID_NIC_QMAN_ARC15] = mmNIC7_QM_DCCM1_BASE,
[CPU_ID_NIC_QMAN_ARC16] = mmNIC8_QM_DCCM0_BASE,
[CPU_ID_NIC_QMAN_ARC17] = mmNIC8_QM_DCCM1_BASE,
[CPU_ID_NIC_QMAN_ARC18] = mmNIC9_QM_DCCM0_BASE,
[CPU_ID_NIC_QMAN_ARC19] = mmNIC9_QM_DCCM1_BASE,
[CPU_ID_NIC_QMAN_ARC20] = mmNIC10_QM_DCCM0_BASE,
[CPU_ID_NIC_QMAN_ARC21] = mmNIC10_QM_DCCM1_BASE,
[CPU_ID_NIC_QMAN_ARC22] = mmNIC11_QM_DCCM0_BASE,
[CPU_ID_NIC_QMAN_ARC23] = mmNIC11_QM_DCCM1_BASE,
};
const u32 gaudi2_mme_ctrl_lo_blocks_bases[MME_ID_SIZE] = {
[MME_ID_DCORE0] = mmDCORE0_MME_CTRL_LO_BASE,
[MME_ID_DCORE1] = mmDCORE1_MME_CTRL_LO_BASE,
[MME_ID_DCORE2] = mmDCORE2_MME_CTRL_LO_BASE,
[MME_ID_DCORE3] = mmDCORE3_MME_CTRL_LO_BASE,
};
static const u32 gaudi2_queue_id_to_arc_id[GAUDI2_QUEUE_ID_SIZE] = {
[GAUDI2_QUEUE_ID_PDMA_0_0] = CPU_ID_PDMA_QMAN_ARC0,
[GAUDI2_QUEUE_ID_PDMA_0_1] = CPU_ID_PDMA_QMAN_ARC0,
[GAUDI2_QUEUE_ID_PDMA_0_2] = CPU_ID_PDMA_QMAN_ARC0,
[GAUDI2_QUEUE_ID_PDMA_0_3] = CPU_ID_PDMA_QMAN_ARC0,
[GAUDI2_QUEUE_ID_PDMA_1_0] = CPU_ID_PDMA_QMAN_ARC1,
[GAUDI2_QUEUE_ID_PDMA_1_1] = CPU_ID_PDMA_QMAN_ARC1,
[GAUDI2_QUEUE_ID_PDMA_1_2] = CPU_ID_PDMA_QMAN_ARC1,
[GAUDI2_QUEUE_ID_PDMA_1_3] = CPU_ID_PDMA_QMAN_ARC1,
[GAUDI2_QUEUE_ID_DCORE0_EDMA_0_0] = CPU_ID_EDMA_QMAN_ARC0,
[GAUDI2_QUEUE_ID_DCORE0_EDMA_0_1] = CPU_ID_EDMA_QMAN_ARC0,
[GAUDI2_QUEUE_ID_DCORE0_EDMA_0_2] = CPU_ID_EDMA_QMAN_ARC0,
[GAUDI2_QUEUE_ID_DCORE0_EDMA_0_3] = CPU_ID_EDMA_QMAN_ARC0,
[GAUDI2_QUEUE_ID_DCORE0_EDMA_1_0] = CPU_ID_EDMA_QMAN_ARC1,
[GAUDI2_QUEUE_ID_DCORE0_EDMA_1_1] = CPU_ID_EDMA_QMAN_ARC1,
[GAUDI2_QUEUE_ID_DCORE0_EDMA_1_2] = CPU_ID_EDMA_QMAN_ARC1,
[GAUDI2_QUEUE_ID_DCORE0_EDMA_1_3] = CPU_ID_EDMA_QMAN_ARC1,
[GAUDI2_QUEUE_ID_DCORE0_MME_0_0] = CPU_ID_MME_QMAN_ARC0,
[GAUDI2_QUEUE_ID_DCORE0_MME_0_1] = CPU_ID_MME_QMAN_ARC0,
[GAUDI2_QUEUE_ID_DCORE0_MME_0_2] = CPU_ID_MME_QMAN_ARC0,
[GAUDI2_QUEUE_ID_DCORE0_MME_0_3] = CPU_ID_MME_QMAN_ARC0,
[GAUDI2_QUEUE_ID_DCORE0_TPC_0_0] = CPU_ID_TPC_QMAN_ARC0,
[GAUDI2_QUEUE_ID_DCORE0_TPC_0_1] = CPU_ID_TPC_QMAN_ARC0,
[GAUDI2_QUEUE_ID_DCORE0_TPC_0_2] = CPU_ID_TPC_QMAN_ARC0,
[GAUDI2_QUEUE_ID_DCORE0_TPC_0_3] = CPU_ID_TPC_QMAN_ARC0,
[GAUDI2_QUEUE_ID_DCORE0_TPC_1_0] = CPU_ID_TPC_QMAN_ARC1,
[GAUDI2_QUEUE_ID_DCORE0_TPC_1_1] = CPU_ID_TPC_QMAN_ARC1,
[GAUDI2_QUEUE_ID_DCORE0_TPC_1_2] = CPU_ID_TPC_QMAN_ARC1,
[GAUDI2_QUEUE_ID_DCORE0_TPC_1_3] = CPU_ID_TPC_QMAN_ARC1,
[GAUDI2_QUEUE_ID_DCORE0_TPC_2_0] = CPU_ID_TPC_QMAN_ARC2,
[GAUDI2_QUEUE_ID_DCORE0_TPC_2_1] = CPU_ID_TPC_QMAN_ARC2,
[GAUDI2_QUEUE_ID_DCORE0_TPC_2_2] = CPU_ID_TPC_QMAN_ARC2,
[GAUDI2_QUEUE_ID_DCORE0_TPC_2_3] = CPU_ID_TPC_QMAN_ARC2,
[GAUDI2_QUEUE_ID_DCORE0_TPC_3_0] = CPU_ID_TPC_QMAN_ARC3,
[GAUDI2_QUEUE_ID_DCORE0_TPC_3_1] = CPU_ID_TPC_QMAN_ARC3,
[GAUDI2_QUEUE_ID_DCORE0_TPC_3_2] = CPU_ID_TPC_QMAN_ARC3,
[GAUDI2_QUEUE_ID_DCORE0_TPC_3_3] = CPU_ID_TPC_QMAN_ARC3,
[GAUDI2_QUEUE_ID_DCORE0_TPC_4_0] = CPU_ID_TPC_QMAN_ARC4,
[GAUDI2_QUEUE_ID_DCORE0_TPC_4_1] = CPU_ID_TPC_QMAN_ARC4,
[GAUDI2_QUEUE_ID_DCORE0_TPC_4_2] = CPU_ID_TPC_QMAN_ARC4,
[GAUDI2_QUEUE_ID_DCORE0_TPC_4_3] = CPU_ID_TPC_QMAN_ARC4,
[GAUDI2_QUEUE_ID_DCORE0_TPC_5_0] = CPU_ID_TPC_QMAN_ARC5,
[GAUDI2_QUEUE_ID_DCORE0_TPC_5_1] = CPU_ID_TPC_QMAN_ARC5,
[GAUDI2_QUEUE_ID_DCORE0_TPC_5_2] = CPU_ID_TPC_QMAN_ARC5,
[GAUDI2_QUEUE_ID_DCORE0_TPC_5_3] = CPU_ID_TPC_QMAN_ARC5,
[GAUDI2_QUEUE_ID_DCORE0_TPC_6_0] = CPU_ID_TPC_QMAN_ARC24,
[GAUDI2_QUEUE_ID_DCORE0_TPC_6_1] = CPU_ID_TPC_QMAN_ARC24,
[GAUDI2_QUEUE_ID_DCORE0_TPC_6_2] = CPU_ID_TPC_QMAN_ARC24,
[GAUDI2_QUEUE_ID_DCORE0_TPC_6_3] = CPU_ID_TPC_QMAN_ARC24,
[GAUDI2_QUEUE_ID_DCORE1_EDMA_0_0] = CPU_ID_EDMA_QMAN_ARC2,
[GAUDI2_QUEUE_ID_DCORE1_EDMA_0_1] = CPU_ID_EDMA_QMAN_ARC2,
[GAUDI2_QUEUE_ID_DCORE1_EDMA_0_2] = CPU_ID_EDMA_QMAN_ARC2,
[GAUDI2_QUEUE_ID_DCORE1_EDMA_0_3] = CPU_ID_EDMA_QMAN_ARC2,
[GAUDI2_QUEUE_ID_DCORE1_EDMA_1_0] = CPU_ID_EDMA_QMAN_ARC3,
[GAUDI2_QUEUE_ID_DCORE1_EDMA_1_1] = CPU_ID_EDMA_QMAN_ARC3,
[GAUDI2_QUEUE_ID_DCORE1_EDMA_1_2] = CPU_ID_EDMA_QMAN_ARC3,
[GAUDI2_QUEUE_ID_DCORE1_EDMA_1_3] = CPU_ID_EDMA_QMAN_ARC3,
[GAUDI2_QUEUE_ID_DCORE1_MME_0_0] = CPU_ID_SCHED_ARC4,
[GAUDI2_QUEUE_ID_DCORE1_MME_0_1] = CPU_ID_SCHED_ARC4,
[GAUDI2_QUEUE_ID_DCORE1_MME_0_2] = CPU_ID_SCHED_ARC4,
[GAUDI2_QUEUE_ID_DCORE1_MME_0_3] = CPU_ID_SCHED_ARC4,
[GAUDI2_QUEUE_ID_DCORE1_TPC_0_0] = CPU_ID_TPC_QMAN_ARC6,
[GAUDI2_QUEUE_ID_DCORE1_TPC_0_1] = CPU_ID_TPC_QMAN_ARC6,
[GAUDI2_QUEUE_ID_DCORE1_TPC_0_2] = CPU_ID_TPC_QMAN_ARC6,
[GAUDI2_QUEUE_ID_DCORE1_TPC_0_3] = CPU_ID_TPC_QMAN_ARC6,
[GAUDI2_QUEUE_ID_DCORE1_TPC_1_0] = CPU_ID_TPC_QMAN_ARC7,
[GAUDI2_QUEUE_ID_DCORE1_TPC_1_1] = CPU_ID_TPC_QMAN_ARC7,
[GAUDI2_QUEUE_ID_DCORE1_TPC_1_2] = CPU_ID_TPC_QMAN_ARC7,
[GAUDI2_QUEUE_ID_DCORE1_TPC_1_3] = CPU_ID_TPC_QMAN_ARC7,
[GAUDI2_QUEUE_ID_DCORE1_TPC_2_0] = CPU_ID_TPC_QMAN_ARC8,
[GAUDI2_QUEUE_ID_DCORE1_TPC_2_1] = CPU_ID_TPC_QMAN_ARC8,
[GAUDI2_QUEUE_ID_DCORE1_TPC_2_2] = CPU_ID_TPC_QMAN_ARC8,
[GAUDI2_QUEUE_ID_DCORE1_TPC_2_3] = CPU_ID_TPC_QMAN_ARC8,
[GAUDI2_QUEUE_ID_DCORE1_TPC_3_0] = CPU_ID_TPC_QMAN_ARC9,
[GAUDI2_QUEUE_ID_DCORE1_TPC_3_1] = CPU_ID_TPC_QMAN_ARC9,
[GAUDI2_QUEUE_ID_DCORE1_TPC_3_2] = CPU_ID_TPC_QMAN_ARC9,
[GAUDI2_QUEUE_ID_DCORE1_TPC_3_3] = CPU_ID_TPC_QMAN_ARC9,
[GAUDI2_QUEUE_ID_DCORE1_TPC_4_0] = CPU_ID_TPC_QMAN_ARC10,
[GAUDI2_QUEUE_ID_DCORE1_TPC_4_1] = CPU_ID_TPC_QMAN_ARC10,
[GAUDI2_QUEUE_ID_DCORE1_TPC_4_2] = CPU_ID_TPC_QMAN_ARC10,
[GAUDI2_QUEUE_ID_DCORE1_TPC_4_3] = CPU_ID_TPC_QMAN_ARC10,
[GAUDI2_QUEUE_ID_DCORE1_TPC_5_0] = CPU_ID_TPC_QMAN_ARC11,
[GAUDI2_QUEUE_ID_DCORE1_TPC_5_1] = CPU_ID_TPC_QMAN_ARC11,
[GAUDI2_QUEUE_ID_DCORE1_TPC_5_2] = CPU_ID_TPC_QMAN_ARC11,
[GAUDI2_QUEUE_ID_DCORE1_TPC_5_3] = CPU_ID_TPC_QMAN_ARC11,
[GAUDI2_QUEUE_ID_DCORE2_EDMA_0_0] = CPU_ID_EDMA_QMAN_ARC4,
[GAUDI2_QUEUE_ID_DCORE2_EDMA_0_1] = CPU_ID_EDMA_QMAN_ARC4,
[GAUDI2_QUEUE_ID_DCORE2_EDMA_0_2] = CPU_ID_EDMA_QMAN_ARC4,
[GAUDI2_QUEUE_ID_DCORE2_EDMA_0_3] = CPU_ID_EDMA_QMAN_ARC4,
[GAUDI2_QUEUE_ID_DCORE2_EDMA_1_0] = CPU_ID_EDMA_QMAN_ARC5,
[GAUDI2_QUEUE_ID_DCORE2_EDMA_1_1] = CPU_ID_EDMA_QMAN_ARC5,
[GAUDI2_QUEUE_ID_DCORE2_EDMA_1_2] = CPU_ID_EDMA_QMAN_ARC5,
[GAUDI2_QUEUE_ID_DCORE2_EDMA_1_3] = CPU_ID_EDMA_QMAN_ARC5,
[GAUDI2_QUEUE_ID_DCORE2_MME_0_0] = CPU_ID_MME_QMAN_ARC1,
[GAUDI2_QUEUE_ID_DCORE2_MME_0_1] = CPU_ID_MME_QMAN_ARC1,
[GAUDI2_QUEUE_ID_DCORE2_MME_0_2] = CPU_ID_MME_QMAN_ARC1,
[GAUDI2_QUEUE_ID_DCORE2_MME_0_3] = CPU_ID_MME_QMAN_ARC1,
[GAUDI2_QUEUE_ID_DCORE2_TPC_0_0] = CPU_ID_TPC_QMAN_ARC12,
[GAUDI2_QUEUE_ID_DCORE2_TPC_0_1] = CPU_ID_TPC_QMAN_ARC12,
[GAUDI2_QUEUE_ID_DCORE2_TPC_0_2] = CPU_ID_TPC_QMAN_ARC12,
[GAUDI2_QUEUE_ID_DCORE2_TPC_0_3] = CPU_ID_TPC_QMAN_ARC12,
[GAUDI2_QUEUE_ID_DCORE2_TPC_1_0] = CPU_ID_TPC_QMAN_ARC13,
[GAUDI2_QUEUE_ID_DCORE2_TPC_1_1] = CPU_ID_TPC_QMAN_ARC13,
[GAUDI2_QUEUE_ID_DCORE2_TPC_1_2] = CPU_ID_TPC_QMAN_ARC13,
[GAUDI2_QUEUE_ID_DCORE2_TPC_1_3] = CPU_ID_TPC_QMAN_ARC13,
[GAUDI2_QUEUE_ID_DCORE2_TPC_2_0] = CPU_ID_TPC_QMAN_ARC14,
[GAUDI2_QUEUE_ID_DCORE2_TPC_2_1] = CPU_ID_TPC_QMAN_ARC14,
[GAUDI2_QUEUE_ID_DCORE2_TPC_2_2] = CPU_ID_TPC_QMAN_ARC14,
[GAUDI2_QUEUE_ID_DCORE2_TPC_2_3] = CPU_ID_TPC_QMAN_ARC14,
[GAUDI2_QUEUE_ID_DCORE2_TPC_3_0] = CPU_ID_TPC_QMAN_ARC15,
[GAUDI2_QUEUE_ID_DCORE2_TPC_3_1] = CPU_ID_TPC_QMAN_ARC15,
[GAUDI2_QUEUE_ID_DCORE2_TPC_3_2] = CPU_ID_TPC_QMAN_ARC15,
[GAUDI2_QUEUE_ID_DCORE2_TPC_3_3] = CPU_ID_TPC_QMAN_ARC15,
[GAUDI2_QUEUE_ID_DCORE2_TPC_4_0] = CPU_ID_TPC_QMAN_ARC16,
[GAUDI2_QUEUE_ID_DCORE2_TPC_4_1] = CPU_ID_TPC_QMAN_ARC16,
[GAUDI2_QUEUE_ID_DCORE2_TPC_4_2] = CPU_ID_TPC_QMAN_ARC16,
[GAUDI2_QUEUE_ID_DCORE2_TPC_4_3] = CPU_ID_TPC_QMAN_ARC16,
[GAUDI2_QUEUE_ID_DCORE2_TPC_5_0] = CPU_ID_TPC_QMAN_ARC17,
[GAUDI2_QUEUE_ID_DCORE2_TPC_5_1] = CPU_ID_TPC_QMAN_ARC17,
[GAUDI2_QUEUE_ID_DCORE2_TPC_5_2] = CPU_ID_TPC_QMAN_ARC17,
[GAUDI2_QUEUE_ID_DCORE2_TPC_5_3] = CPU_ID_TPC_QMAN_ARC17,
[GAUDI2_QUEUE_ID_DCORE3_EDMA_0_0] = CPU_ID_EDMA_QMAN_ARC6,
[GAUDI2_QUEUE_ID_DCORE3_EDMA_0_1] = CPU_ID_EDMA_QMAN_ARC6,
[GAUDI2_QUEUE_ID_DCORE3_EDMA_0_2] = CPU_ID_EDMA_QMAN_ARC6,
[GAUDI2_QUEUE_ID_DCORE3_EDMA_0_3] = CPU_ID_EDMA_QMAN_ARC6,
[GAUDI2_QUEUE_ID_DCORE3_EDMA_1_0] = CPU_ID_EDMA_QMAN_ARC7,
[GAUDI2_QUEUE_ID_DCORE3_EDMA_1_1] = CPU_ID_EDMA_QMAN_ARC7,
[GAUDI2_QUEUE_ID_DCORE3_EDMA_1_2] = CPU_ID_EDMA_QMAN_ARC7,
[GAUDI2_QUEUE_ID_DCORE3_EDMA_1_3] = CPU_ID_EDMA_QMAN_ARC7,
[GAUDI2_QUEUE_ID_DCORE3_MME_0_0] = CPU_ID_SCHED_ARC5,
[GAUDI2_QUEUE_ID_DCORE3_MME_0_1] = CPU_ID_SCHED_ARC5,
[GAUDI2_QUEUE_ID_DCORE3_MME_0_2] = CPU_ID_SCHED_ARC5,
[GAUDI2_QUEUE_ID_DCORE3_MME_0_3] = CPU_ID_SCHED_ARC5,
[GAUDI2_QUEUE_ID_DCORE3_TPC_0_0] = CPU_ID_TPC_QMAN_ARC18,
[GAUDI2_QUEUE_ID_DCORE3_TPC_0_1] = CPU_ID_TPC_QMAN_ARC18,
[GAUDI2_QUEUE_ID_DCORE3_TPC_0_2] = CPU_ID_TPC_QMAN_ARC18,
[GAUDI2_QUEUE_ID_DCORE3_TPC_0_3] = CPU_ID_TPC_QMAN_ARC18,
[GAUDI2_QUEUE_ID_DCORE3_TPC_1_0] = CPU_ID_TPC_QMAN_ARC19,
[GAUDI2_QUEUE_ID_DCORE3_TPC_1_1] = CPU_ID_TPC_QMAN_ARC19,
[GAUDI2_QUEUE_ID_DCORE3_TPC_1_2] = CPU_ID_TPC_QMAN_ARC19,
[GAUDI2_QUEUE_ID_DCORE3_TPC_1_3] = CPU_ID_TPC_QMAN_ARC19,
[GAUDI2_QUEUE_ID_DCORE3_TPC_2_0] = CPU_ID_TPC_QMAN_ARC20,
[GAUDI2_QUEUE_ID_DCORE3_TPC_2_1] = CPU_ID_TPC_QMAN_ARC20,
[GAUDI2_QUEUE_ID_DCORE3_TPC_2_2] = CPU_ID_TPC_QMAN_ARC20,
[GAUDI2_QUEUE_ID_DCORE3_TPC_2_3] = CPU_ID_TPC_QMAN_ARC20,
[GAUDI2_QUEUE_ID_DCORE3_TPC_3_0] = CPU_ID_TPC_QMAN_ARC21,
[GAUDI2_QUEUE_ID_DCORE3_TPC_3_1] = CPU_ID_TPC_QMAN_ARC21,
[GAUDI2_QUEUE_ID_DCORE3_TPC_3_2] = CPU_ID_TPC_QMAN_ARC21,
[GAUDI2_QUEUE_ID_DCORE3_TPC_3_3] = CPU_ID_TPC_QMAN_ARC21,
[GAUDI2_QUEUE_ID_DCORE3_TPC_4_0] = CPU_ID_TPC_QMAN_ARC22,
[GAUDI2_QUEUE_ID_DCORE3_TPC_4_1] = CPU_ID_TPC_QMAN_ARC22,
[GAUDI2_QUEUE_ID_DCORE3_TPC_4_2] = CPU_ID_TPC_QMAN_ARC22,
[GAUDI2_QUEUE_ID_DCORE3_TPC_4_3] = CPU_ID_TPC_QMAN_ARC22,
[GAUDI2_QUEUE_ID_DCORE3_TPC_5_0] = CPU_ID_TPC_QMAN_ARC23,
[GAUDI2_QUEUE_ID_DCORE3_TPC_5_1] = CPU_ID_TPC_QMAN_ARC23,
[GAUDI2_QUEUE_ID_DCORE3_TPC_5_2] = CPU_ID_TPC_QMAN_ARC23,
[GAUDI2_QUEUE_ID_DCORE3_TPC_5_3] = CPU_ID_TPC_QMAN_ARC23,
[GAUDI2_QUEUE_ID_NIC_0_0] = CPU_ID_NIC_QMAN_ARC0,
[GAUDI2_QUEUE_ID_NIC_0_1] = CPU_ID_NIC_QMAN_ARC0,
[GAUDI2_QUEUE_ID_NIC_0_2] = CPU_ID_NIC_QMAN_ARC0,
[GAUDI2_QUEUE_ID_NIC_0_3] = CPU_ID_NIC_QMAN_ARC0,
[GAUDI2_QUEUE_ID_NIC_1_0] = CPU_ID_NIC_QMAN_ARC1,
[GAUDI2_QUEUE_ID_NIC_1_1] = CPU_ID_NIC_QMAN_ARC1,
[GAUDI2_QUEUE_ID_NIC_1_2] = CPU_ID_NIC_QMAN_ARC1,
[GAUDI2_QUEUE_ID_NIC_1_3] = CPU_ID_NIC_QMAN_ARC1,
[GAUDI2_QUEUE_ID_NIC_2_0] = CPU_ID_NIC_QMAN_ARC2,
[GAUDI2_QUEUE_ID_NIC_2_1] = CPU_ID_NIC_QMAN_ARC2,
[GAUDI2_QUEUE_ID_NIC_2_2] = CPU_ID_NIC_QMAN_ARC2,
[GAUDI2_QUEUE_ID_NIC_2_3] = CPU_ID_NIC_QMAN_ARC2,
[GAUDI2_QUEUE_ID_NIC_3_0] = CPU_ID_NIC_QMAN_ARC3,
[GAUDI2_QUEUE_ID_NIC_3_1] = CPU_ID_NIC_QMAN_ARC3,
[GAUDI2_QUEUE_ID_NIC_3_2] = CPU_ID_NIC_QMAN_ARC3,
[GAUDI2_QUEUE_ID_NIC_3_3] = CPU_ID_NIC_QMAN_ARC3,
[GAUDI2_QUEUE_ID_NIC_4_0] = CPU_ID_NIC_QMAN_ARC4,
[GAUDI2_QUEUE_ID_NIC_4_1] = CPU_ID_NIC_QMAN_ARC4,
[GAUDI2_QUEUE_ID_NIC_4_2] = CPU_ID_NIC_QMAN_ARC4,
[GAUDI2_QUEUE_ID_NIC_4_3] = CPU_ID_NIC_QMAN_ARC4,
[GAUDI2_QUEUE_ID_NIC_5_0] = CPU_ID_NIC_QMAN_ARC5,
[GAUDI2_QUEUE_ID_NIC_5_1] = CPU_ID_NIC_QMAN_ARC5,
[GAUDI2_QUEUE_ID_NIC_5_2] = CPU_ID_NIC_QMAN_ARC5,
[GAUDI2_QUEUE_ID_NIC_5_3] = CPU_ID_NIC_QMAN_ARC5,
[GAUDI2_QUEUE_ID_NIC_6_0] = CPU_ID_NIC_QMAN_ARC6,
[GAUDI2_QUEUE_ID_NIC_6_1] = CPU_ID_NIC_QMAN_ARC6,
[GAUDI2_QUEUE_ID_NIC_6_2] = CPU_ID_NIC_QMAN_ARC6,
[GAUDI2_QUEUE_ID_NIC_6_3] = CPU_ID_NIC_QMAN_ARC6,
[GAUDI2_QUEUE_ID_NIC_7_0] = CPU_ID_NIC_QMAN_ARC7,
[GAUDI2_QUEUE_ID_NIC_7_1] = CPU_ID_NIC_QMAN_ARC7,
[GAUDI2_QUEUE_ID_NIC_7_2] = CPU_ID_NIC_QMAN_ARC7,
[GAUDI2_QUEUE_ID_NIC_7_3] = CPU_ID_NIC_QMAN_ARC7,
[GAUDI2_QUEUE_ID_NIC_8_0] = CPU_ID_NIC_QMAN_ARC8,
[GAUDI2_QUEUE_ID_NIC_8_1] = CPU_ID_NIC_QMAN_ARC8,
[GAUDI2_QUEUE_ID_NIC_8_2] = CPU_ID_NIC_QMAN_ARC8,
[GAUDI2_QUEUE_ID_NIC_8_3] = CPU_ID_NIC_QMAN_ARC8,
[GAUDI2_QUEUE_ID_NIC_9_0] = CPU_ID_NIC_QMAN_ARC9,
[GAUDI2_QUEUE_ID_NIC_9_1] = CPU_ID_NIC_QMAN_ARC9,
[GAUDI2_QUEUE_ID_NIC_9_2] = CPU_ID_NIC_QMAN_ARC9,
[GAUDI2_QUEUE_ID_NIC_9_3] = CPU_ID_NIC_QMAN_ARC9,
[GAUDI2_QUEUE_ID_NIC_10_0] = CPU_ID_NIC_QMAN_ARC10,
[GAUDI2_QUEUE_ID_NIC_10_1] = CPU_ID_NIC_QMAN_ARC10,
[GAUDI2_QUEUE_ID_NIC_10_2] = CPU_ID_NIC_QMAN_ARC10,
[GAUDI2_QUEUE_ID_NIC_10_3] = CPU_ID_NIC_QMAN_ARC10,
[GAUDI2_QUEUE_ID_NIC_11_0] = CPU_ID_NIC_QMAN_ARC11,
[GAUDI2_QUEUE_ID_NIC_11_1] = CPU_ID_NIC_QMAN_ARC11,
[GAUDI2_QUEUE_ID_NIC_11_2] = CPU_ID_NIC_QMAN_ARC11,
[GAUDI2_QUEUE_ID_NIC_11_3] = CPU_ID_NIC_QMAN_ARC11,
[GAUDI2_QUEUE_ID_NIC_12_0] = CPU_ID_NIC_QMAN_ARC12,
[GAUDI2_QUEUE_ID_NIC_12_1] = CPU_ID_NIC_QMAN_ARC12,
[GAUDI2_QUEUE_ID_NIC_12_2] = CPU_ID_NIC_QMAN_ARC12,
[GAUDI2_QUEUE_ID_NIC_12_3] = CPU_ID_NIC_QMAN_ARC12,
[GAUDI2_QUEUE_ID_NIC_13_0] = CPU_ID_NIC_QMAN_ARC13,
[GAUDI2_QUEUE_ID_NIC_13_1] = CPU_ID_NIC_QMAN_ARC13,
[GAUDI2_QUEUE_ID_NIC_13_2] = CPU_ID_NIC_QMAN_ARC13,
[GAUDI2_QUEUE_ID_NIC_13_3] = CPU_ID_NIC_QMAN_ARC13,
[GAUDI2_QUEUE_ID_NIC_14_0] = CPU_ID_NIC_QMAN_ARC14,
[GAUDI2_QUEUE_ID_NIC_14_1] = CPU_ID_NIC_QMAN_ARC14,
[GAUDI2_QUEUE_ID_NIC_14_2] = CPU_ID_NIC_QMAN_ARC14,
[GAUDI2_QUEUE_ID_NIC_14_3] = CPU_ID_NIC_QMAN_ARC14,
[GAUDI2_QUEUE_ID_NIC_15_0] = CPU_ID_NIC_QMAN_ARC15,
[GAUDI2_QUEUE_ID_NIC_15_1] = CPU_ID_NIC_QMAN_ARC15,
[GAUDI2_QUEUE_ID_NIC_15_2] = CPU_ID_NIC_QMAN_ARC15,
[GAUDI2_QUEUE_ID_NIC_15_3] = CPU_ID_NIC_QMAN_ARC15,
[GAUDI2_QUEUE_ID_NIC_16_0] = CPU_ID_NIC_QMAN_ARC16,
[GAUDI2_QUEUE_ID_NIC_16_1] = CPU_ID_NIC_QMAN_ARC16,
[GAUDI2_QUEUE_ID_NIC_16_2] = CPU_ID_NIC_QMAN_ARC16,
[GAUDI2_QUEUE_ID_NIC_16_3] = CPU_ID_NIC_QMAN_ARC16,
[GAUDI2_QUEUE_ID_NIC_17_0] = CPU_ID_NIC_QMAN_ARC17,
[GAUDI2_QUEUE_ID_NIC_17_1] = CPU_ID_NIC_QMAN_ARC17,
[GAUDI2_QUEUE_ID_NIC_17_2] = CPU_ID_NIC_QMAN_ARC17,
[GAUDI2_QUEUE_ID_NIC_17_3] = CPU_ID_NIC_QMAN_ARC17,
[GAUDI2_QUEUE_ID_NIC_18_0] = CPU_ID_NIC_QMAN_ARC18,
[GAUDI2_QUEUE_ID_NIC_18_1] = CPU_ID_NIC_QMAN_ARC18,
[GAUDI2_QUEUE_ID_NIC_18_2] = CPU_ID_NIC_QMAN_ARC18,
[GAUDI2_QUEUE_ID_NIC_18_3] = CPU_ID_NIC_QMAN_ARC18,
[GAUDI2_QUEUE_ID_NIC_19_0] = CPU_ID_NIC_QMAN_ARC19,
[GAUDI2_QUEUE_ID_NIC_19_1] = CPU_ID_NIC_QMAN_ARC19,
[GAUDI2_QUEUE_ID_NIC_19_2] = CPU_ID_NIC_QMAN_ARC19,
[GAUDI2_QUEUE_ID_NIC_19_3] = CPU_ID_NIC_QMAN_ARC19,
[GAUDI2_QUEUE_ID_NIC_20_0] = CPU_ID_NIC_QMAN_ARC20,
[GAUDI2_QUEUE_ID_NIC_20_1] = CPU_ID_NIC_QMAN_ARC20,
[GAUDI2_QUEUE_ID_NIC_20_2] = CPU_ID_NIC_QMAN_ARC20,
[GAUDI2_QUEUE_ID_NIC_20_3] = CPU_ID_NIC_QMAN_ARC20,
[GAUDI2_QUEUE_ID_NIC_21_0] = CPU_ID_NIC_QMAN_ARC21,
[GAUDI2_QUEUE_ID_NIC_21_1] = CPU_ID_NIC_QMAN_ARC21,
[GAUDI2_QUEUE_ID_NIC_21_2] = CPU_ID_NIC_QMAN_ARC21,
[GAUDI2_QUEUE_ID_NIC_21_3] = CPU_ID_NIC_QMAN_ARC21,
[GAUDI2_QUEUE_ID_NIC_22_0] = CPU_ID_NIC_QMAN_ARC22,
[GAUDI2_QUEUE_ID_NIC_22_1] = CPU_ID_NIC_QMAN_ARC22,
[GAUDI2_QUEUE_ID_NIC_22_2] = CPU_ID_NIC_QMAN_ARC22,
[GAUDI2_QUEUE_ID_NIC_22_3] = CPU_ID_NIC_QMAN_ARC22,
[GAUDI2_QUEUE_ID_NIC_23_0] = CPU_ID_NIC_QMAN_ARC23,
[GAUDI2_QUEUE_ID_NIC_23_1] = CPU_ID_NIC_QMAN_ARC23,
[GAUDI2_QUEUE_ID_NIC_23_2] = CPU_ID_NIC_QMAN_ARC23,
[GAUDI2_QUEUE_ID_NIC_23_3] = CPU_ID_NIC_QMAN_ARC23,
[GAUDI2_QUEUE_ID_ROT_0_0] = CPU_ID_ROT_QMAN_ARC0,
[GAUDI2_QUEUE_ID_ROT_0_1] = CPU_ID_ROT_QMAN_ARC0,
[GAUDI2_QUEUE_ID_ROT_0_2] = CPU_ID_ROT_QMAN_ARC0,
[GAUDI2_QUEUE_ID_ROT_0_3] = CPU_ID_ROT_QMAN_ARC0,
[GAUDI2_QUEUE_ID_ROT_1_0] = CPU_ID_ROT_QMAN_ARC1,
[GAUDI2_QUEUE_ID_ROT_1_1] = CPU_ID_ROT_QMAN_ARC1,
[GAUDI2_QUEUE_ID_ROT_1_2] = CPU_ID_ROT_QMAN_ARC1,
[GAUDI2_QUEUE_ID_ROT_1_3] = CPU_ID_ROT_QMAN_ARC1
};
const u32 gaudi2_dma_core_blocks_bases[DMA_CORE_ID_SIZE] = {
[DMA_CORE_ID_PDMA0] = mmPDMA0_CORE_BASE,
[DMA_CORE_ID_PDMA1] = mmPDMA1_CORE_BASE,
[DMA_CORE_ID_EDMA0] = mmDCORE0_EDMA0_CORE_BASE,
[DMA_CORE_ID_EDMA1] = mmDCORE0_EDMA1_CORE_BASE,
[DMA_CORE_ID_EDMA2] = mmDCORE1_EDMA0_CORE_BASE,
[DMA_CORE_ID_EDMA3] = mmDCORE1_EDMA1_CORE_BASE,
[DMA_CORE_ID_EDMA4] = mmDCORE2_EDMA0_CORE_BASE,
[DMA_CORE_ID_EDMA5] = mmDCORE2_EDMA1_CORE_BASE,
[DMA_CORE_ID_EDMA6] = mmDCORE3_EDMA0_CORE_BASE,
[DMA_CORE_ID_EDMA7] = mmDCORE3_EDMA1_CORE_BASE,
[DMA_CORE_ID_KDMA] = mmARC_FARM_KDMA_BASE
};
const u32 gaudi2_mme_acc_blocks_bases[MME_ID_SIZE] = {
[MME_ID_DCORE0] = mmDCORE0_MME_ACC_BASE,
[MME_ID_DCORE1] = mmDCORE1_MME_ACC_BASE,
[MME_ID_DCORE2] = mmDCORE2_MME_ACC_BASE,
[MME_ID_DCORE3] = mmDCORE3_MME_ACC_BASE
};
static const u32 gaudi2_tpc_cfg_blocks_bases[TPC_ID_SIZE] = {
[TPC_ID_DCORE0_TPC0] = mmDCORE0_TPC0_CFG_BASE,
[TPC_ID_DCORE0_TPC1] = mmDCORE0_TPC1_CFG_BASE,
[TPC_ID_DCORE0_TPC2] = mmDCORE0_TPC2_CFG_BASE,
[TPC_ID_DCORE0_TPC3] = mmDCORE0_TPC3_CFG_BASE,
[TPC_ID_DCORE0_TPC4] = mmDCORE0_TPC4_CFG_BASE,
[TPC_ID_DCORE0_TPC5] = mmDCORE0_TPC5_CFG_BASE,
[TPC_ID_DCORE1_TPC0] = mmDCORE1_TPC0_CFG_BASE,
[TPC_ID_DCORE1_TPC1] = mmDCORE1_TPC1_CFG_BASE,
[TPC_ID_DCORE1_TPC2] = mmDCORE1_TPC2_CFG_BASE,
[TPC_ID_DCORE1_TPC3] = mmDCORE1_TPC3_CFG_BASE,
[TPC_ID_DCORE1_TPC4] = mmDCORE1_TPC4_CFG_BASE,
[TPC_ID_DCORE1_TPC5] = mmDCORE1_TPC5_CFG_BASE,
[TPC_ID_DCORE2_TPC0] = mmDCORE2_TPC0_CFG_BASE,
[TPC_ID_DCORE2_TPC1] = mmDCORE2_TPC1_CFG_BASE,
[TPC_ID_DCORE2_TPC2] = mmDCORE2_TPC2_CFG_BASE,
[TPC_ID_DCORE2_TPC3] = mmDCORE2_TPC3_CFG_BASE,
[TPC_ID_DCORE2_TPC4] = mmDCORE2_TPC4_CFG_BASE,
[TPC_ID_DCORE2_TPC5] = mmDCORE2_TPC5_CFG_BASE,
[TPC_ID_DCORE3_TPC0] = mmDCORE3_TPC0_CFG_BASE,
[TPC_ID_DCORE3_TPC1] = mmDCORE3_TPC1_CFG_BASE,
[TPC_ID_DCORE3_TPC2] = mmDCORE3_TPC2_CFG_BASE,
[TPC_ID_DCORE3_TPC3] = mmDCORE3_TPC3_CFG_BASE,
[TPC_ID_DCORE3_TPC4] = mmDCORE3_TPC4_CFG_BASE,
[TPC_ID_DCORE3_TPC5] = mmDCORE3_TPC5_CFG_BASE,
[TPC_ID_DCORE0_TPC6] = mmDCORE0_TPC6_CFG_BASE,
};
const u32 gaudi2_rot_blocks_bases[ROTATOR_ID_SIZE] = {
[ROTATOR_ID_0] = mmROT0_BASE,
[ROTATOR_ID_1] = mmROT1_BASE
};
static const u32 gaudi2_tpc_id_to_queue_id[TPC_ID_SIZE] = {
[TPC_ID_DCORE0_TPC0] = GAUDI2_QUEUE_ID_DCORE0_TPC_0_0,
[TPC_ID_DCORE0_TPC1] = GAUDI2_QUEUE_ID_DCORE0_TPC_1_0,
[TPC_ID_DCORE0_TPC2] = GAUDI2_QUEUE_ID_DCORE0_TPC_2_0,
[TPC_ID_DCORE0_TPC3] = GAUDI2_QUEUE_ID_DCORE0_TPC_3_0,
[TPC_ID_DCORE0_TPC4] = GAUDI2_QUEUE_ID_DCORE0_TPC_4_0,
[TPC_ID_DCORE0_TPC5] = GAUDI2_QUEUE_ID_DCORE0_TPC_5_0,
[TPC_ID_DCORE1_TPC0] = GAUDI2_QUEUE_ID_DCORE1_TPC_0_0,
[TPC_ID_DCORE1_TPC1] = GAUDI2_QUEUE_ID_DCORE1_TPC_1_0,
[TPC_ID_DCORE1_TPC2] = GAUDI2_QUEUE_ID_DCORE1_TPC_2_0,
[TPC_ID_DCORE1_TPC3] = GAUDI2_QUEUE_ID_DCORE1_TPC_3_0,
[TPC_ID_DCORE1_TPC4] = GAUDI2_QUEUE_ID_DCORE1_TPC_4_0,
[TPC_ID_DCORE1_TPC5] = GAUDI2_QUEUE_ID_DCORE1_TPC_5_0,
[TPC_ID_DCORE2_TPC0] = GAUDI2_QUEUE_ID_DCORE2_TPC_0_0,
[TPC_ID_DCORE2_TPC1] = GAUDI2_QUEUE_ID_DCORE2_TPC_1_0,
[TPC_ID_DCORE2_TPC2] = GAUDI2_QUEUE_ID_DCORE2_TPC_2_0,
[TPC_ID_DCORE2_TPC3] = GAUDI2_QUEUE_ID_DCORE2_TPC_3_0,
[TPC_ID_DCORE2_TPC4] = GAUDI2_QUEUE_ID_DCORE2_TPC_4_0,
[TPC_ID_DCORE2_TPC5] = GAUDI2_QUEUE_ID_DCORE2_TPC_5_0,
[TPC_ID_DCORE3_TPC0] = GAUDI2_QUEUE_ID_DCORE3_TPC_0_0,
[TPC_ID_DCORE3_TPC1] = GAUDI2_QUEUE_ID_DCORE3_TPC_1_0,
[TPC_ID_DCORE3_TPC2] = GAUDI2_QUEUE_ID_DCORE3_TPC_2_0,
[TPC_ID_DCORE3_TPC3] = GAUDI2_QUEUE_ID_DCORE3_TPC_3_0,
[TPC_ID_DCORE3_TPC4] = GAUDI2_QUEUE_ID_DCORE3_TPC_4_0,
[TPC_ID_DCORE3_TPC5] = GAUDI2_QUEUE_ID_DCORE3_TPC_5_0,
[TPC_ID_DCORE0_TPC6] = GAUDI2_QUEUE_ID_DCORE0_TPC_6_0,
};
static const u32 gaudi2_rot_id_to_queue_id[ROTATOR_ID_SIZE] = {
[ROTATOR_ID_0] = GAUDI2_QUEUE_ID_ROT_0_0,
[ROTATOR_ID_1] = GAUDI2_QUEUE_ID_ROT_1_0,
};
const u32 edma_stream_base[NUM_OF_EDMA_PER_DCORE * NUM_OF_DCORES] = {
GAUDI2_QUEUE_ID_DCORE0_EDMA_0_0,
GAUDI2_QUEUE_ID_DCORE0_EDMA_1_0,
GAUDI2_QUEUE_ID_DCORE1_EDMA_0_0,
GAUDI2_QUEUE_ID_DCORE1_EDMA_1_0,
GAUDI2_QUEUE_ID_DCORE2_EDMA_0_0,
GAUDI2_QUEUE_ID_DCORE2_EDMA_1_0,
GAUDI2_QUEUE_ID_DCORE3_EDMA_0_0,
GAUDI2_QUEUE_ID_DCORE3_EDMA_1_0,
};
static const char gaudi2_vdec_irq_name[GAUDI2_VDEC_MSIX_ENTRIES][GAUDI2_MAX_STRING_LEN] = {
"gaudi2 vdec 0_0", "gaudi2 vdec 0_0 abnormal",
"gaudi2 vdec 0_1", "gaudi2 vdec 0_1 abnormal",
"gaudi2 vdec 1_0", "gaudi2 vdec 1_0 abnormal",
"gaudi2 vdec 1_1", "gaudi2 vdec 1_1 abnormal",
"gaudi2 vdec 2_0", "gaudi2 vdec 2_0 abnormal",
"gaudi2 vdec 2_1", "gaudi2 vdec 2_1 abnormal",
"gaudi2 vdec 3_0", "gaudi2 vdec 3_0 abnormal",
"gaudi2 vdec 3_1", "gaudi2 vdec 3_1 abnormal",
"gaudi2 vdec s_0", "gaudi2 vdec s_0 abnormal",
"gaudi2 vdec s_1", "gaudi2 vdec s_1 abnormal"
};
static const u32 rtr_coordinates_to_rtr_id[NUM_OF_RTR_PER_DCORE * NUM_OF_DCORES] = {
RTR_ID_X_Y(2, 4),
RTR_ID_X_Y(3, 4),
RTR_ID_X_Y(4, 4),
RTR_ID_X_Y(5, 4),
RTR_ID_X_Y(6, 4),
RTR_ID_X_Y(7, 4),
RTR_ID_X_Y(8, 4),
RTR_ID_X_Y(9, 4),
RTR_ID_X_Y(10, 4),
RTR_ID_X_Y(11, 4),
RTR_ID_X_Y(12, 4),
RTR_ID_X_Y(13, 4),
RTR_ID_X_Y(14, 4),
RTR_ID_X_Y(15, 4),
RTR_ID_X_Y(16, 4),
RTR_ID_X_Y(17, 4),
RTR_ID_X_Y(2, 11),
RTR_ID_X_Y(3, 11),
RTR_ID_X_Y(4, 11),
RTR_ID_X_Y(5, 11),
RTR_ID_X_Y(6, 11),
RTR_ID_X_Y(7, 11),
RTR_ID_X_Y(8, 11),
RTR_ID_X_Y(9, 11),
RTR_ID_X_Y(0, 0),/* 24 no id */
RTR_ID_X_Y(0, 0),/* 25 no id */
RTR_ID_X_Y(0, 0),/* 26 no id */
RTR_ID_X_Y(0, 0),/* 27 no id */
RTR_ID_X_Y(14, 11),
RTR_ID_X_Y(15, 11),
RTR_ID_X_Y(16, 11),
RTR_ID_X_Y(17, 11)
};
habanalabs/gaudi2: dump detailed information upon RAZWI In order to improve debuggability, we add all available information when a RAZWI event occur. Signed-off-by: Ofir Bitton <obitton@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-08-21 18:50:51 +08:00
enum rtr_id {
DCORE0_RTR0,
DCORE0_RTR1,
DCORE0_RTR2,
DCORE0_RTR3,
DCORE0_RTR4,
DCORE0_RTR5,
DCORE0_RTR6,
DCORE0_RTR7,
DCORE1_RTR0,
DCORE1_RTR1,
DCORE1_RTR2,
DCORE1_RTR3,
DCORE1_RTR4,
DCORE1_RTR5,
DCORE1_RTR6,
DCORE1_RTR7,
DCORE2_RTR0,
DCORE2_RTR1,
DCORE2_RTR2,
DCORE2_RTR3,
DCORE2_RTR4,
DCORE2_RTR5,
DCORE2_RTR6,
DCORE2_RTR7,
DCORE3_RTR0,
DCORE3_RTR1,
DCORE3_RTR2,
DCORE3_RTR3,
DCORE3_RTR4,
DCORE3_RTR5,
DCORE3_RTR6,
DCORE3_RTR7,
};
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
static const u32 gaudi2_tpc_initiator_rtr_id[NUM_OF_TPC_PER_DCORE * NUM_OF_DCORES + 1] = {
habanalabs/gaudi2: dump detailed information upon RAZWI In order to improve debuggability, we add all available information when a RAZWI event occur. Signed-off-by: Ofir Bitton <obitton@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-08-21 18:50:51 +08:00
DCORE0_RTR1, DCORE0_RTR1, DCORE0_RTR2, DCORE0_RTR2, DCORE0_RTR3, DCORE0_RTR3,
DCORE1_RTR6, DCORE1_RTR6, DCORE1_RTR5, DCORE1_RTR5, DCORE1_RTR4, DCORE1_RTR4,
DCORE2_RTR3, DCORE2_RTR3, DCORE2_RTR2, DCORE2_RTR2, DCORE2_RTR1, DCORE2_RTR1,
DCORE3_RTR4, DCORE3_RTR4, DCORE3_RTR5, DCORE3_RTR5, DCORE3_RTR6, DCORE3_RTR6,
DCORE0_RTR0
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
};
static const u32 gaudi2_dec_initiator_rtr_id[NUMBER_OF_DEC] = {
habanalabs/gaudi2: dump detailed information upon RAZWI In order to improve debuggability, we add all available information when a RAZWI event occur. Signed-off-by: Ofir Bitton <obitton@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-08-21 18:50:51 +08:00
DCORE0_RTR0, DCORE0_RTR0, DCORE1_RTR7, DCORE1_RTR7, DCORE2_RTR0, DCORE2_RTR0,
DCORE3_RTR7, DCORE3_RTR7, DCORE0_RTR0, DCORE0_RTR0
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
};
static const u32 gaudi2_nic_initiator_rtr_id[NIC_NUMBER_OF_MACROS] = {
habanalabs/gaudi2: dump detailed information upon RAZWI In order to improve debuggability, we add all available information when a RAZWI event occur. Signed-off-by: Ofir Bitton <obitton@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-08-21 18:50:51 +08:00
DCORE1_RTR7, DCORE1_RTR7, DCORE1_RTR7, DCORE1_RTR7, DCORE1_RTR7, DCORE2_RTR0,
DCORE2_RTR0, DCORE2_RTR0, DCORE2_RTR0, DCORE3_RTR7, DCORE3_RTR7, DCORE3_RTR7
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
};
struct sft_info {
u8 interface_id;
u8 dcore_id;
};
static const struct sft_info gaudi2_edma_initiator_sft_id[NUM_OF_EDMA_PER_DCORE * NUM_OF_DCORES] = {
{0, 0}, {1, 0}, {0, 1}, {1, 1}, {1, 2}, {1, 3}, {0, 2}, {0, 3},
};
static const u32 gaudi2_pdma_initiator_rtr_id[NUM_OF_PDMA] = {
habanalabs/gaudi2: dump detailed information upon RAZWI In order to improve debuggability, we add all available information when a RAZWI event occur. Signed-off-by: Ofir Bitton <obitton@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-08-21 18:50:51 +08:00
DCORE0_RTR0, DCORE0_RTR0
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
};
static const u32 gaudi2_rot_initiator_rtr_id[NUM_OF_ROT] = {
habanalabs/gaudi2: dump detailed information upon RAZWI In order to improve debuggability, we add all available information when a RAZWI event occur. Signed-off-by: Ofir Bitton <obitton@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-08-21 18:50:51 +08:00
DCORE2_RTR0, DCORE3_RTR7
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
};
struct mme_initiators_rtr_id {
u32 wap0;
u32 wap1;
u32 write;
u32 read;
u32 sbte0;
u32 sbte1;
u32 sbte2;
u32 sbte3;
u32 sbte4;
};
enum mme_initiators {
MME_WAP0 = 0,
MME_WAP1,
MME_WRITE,
MME_READ,
MME_SBTE0,
MME_SBTE1,
MME_SBTE2,
MME_SBTE3,
MME_SBTE4,
MME_INITIATORS_MAX
};
static const struct mme_initiators_rtr_id
gaudi2_mme_initiator_rtr_id[NUM_OF_MME_PER_DCORE * NUM_OF_DCORES] = {
{ .wap0 = 5, .wap1 = 7, .write = 6, .read = 7,
.sbte0 = 7, .sbte1 = 4, .sbte2 = 4, .sbte3 = 5, .sbte4 = 6},
{ .wap0 = 10, .wap1 = 8, .write = 9, .read = 8,
.sbte0 = 11, .sbte1 = 11, .sbte2 = 10, .sbte3 = 9, .sbte4 = 8},
{ .wap0 = 21, .wap1 = 23, .write = 22, .read = 23,
.sbte0 = 20, .sbte1 = 20, .sbte2 = 21, .sbte3 = 22, .sbte4 = 23},
{ .wap0 = 30, .wap1 = 28, .write = 29, .read = 30,
.sbte0 = 31, .sbte1 = 31, .sbte2 = 30, .sbte3 = 29, .sbte4 = 28},
};
enum razwi_event_sources {
RAZWI_TPC,
RAZWI_MME,
RAZWI_EDMA,
RAZWI_PDMA,
RAZWI_NIC,
RAZWI_DEC,
RAZWI_ROT
};
struct hbm_mc_error_causes {
u32 mask;
char cause[50];
};
static struct hbm_mc_error_causes hbm_mc_spi[GAUDI2_NUM_OF_HBM_MC_SPI_CAUSE] = {
{HBM_MC_SPI_TEMP_PIN_CHG_MASK, "temperature pins changed"},
{HBM_MC_SPI_THR_ENG_MASK, "temperature-based throttling engaged"},
{HBM_MC_SPI_THR_DIS_ENG_MASK, "temperature-based throttling disengaged"},
{HBM_MC_SPI_IEEE1500_COMP_MASK, "IEEE1500 op comp"},
{HBM_MC_SPI_IEEE1500_PAUSED_MASK, "IEEE1500 op paused"},
};
static const char * const hbm_mc_sei_cause[GAUDI2_NUM_OF_HBM_SEI_CAUSE] = {
[HBM_SEI_CMD_PARITY_EVEN] = "SEI C/A parity even",
[HBM_SEI_CMD_PARITY_ODD] = "SEI C/A parity odd",
[HBM_SEI_READ_ERR] = "SEI read data error",
[HBM_SEI_WRITE_DATA_PARITY_ERR] = "SEI write data parity error",
[HBM_SEI_CATTRIP] = "SEI CATTRIP asserted",
[HBM_SEI_MEM_BIST_FAIL] = "SEI memory BIST fail",
[HBM_SEI_DFI] = "SEI DFI error",
[HBM_SEI_INV_TEMP_READ_OUT] = "SEI invalid temp read",
[HBM_SEI_BIST_FAIL] = "SEI BIST fail"
};
struct mmu_spi_sei_cause {
char cause[50];
int clear_bit;
};
static const struct mmu_spi_sei_cause gaudi2_mmu_spi_sei[GAUDI2_NUM_OF_MMU_SPI_SEI_CAUSE] = {
{"page fault", 1}, /* INTERRUPT_CLR[1] */
{"page access", 1}, /* INTERRUPT_CLR[1] */
{"bypass ddr", 2}, /* INTERRUPT_CLR[2] */
{"multi hit", 2}, /* INTERRUPT_CLR[2] */
{"mmu rei0", -1}, /* no clear register bit */
{"mmu rei1", -1}, /* no clear register bit */
{"stlb rei0", -1}, /* no clear register bit */
{"stlb rei1", -1}, /* no clear register bit */
{"rr privileged write hit", 2}, /* INTERRUPT_CLR[2] */
{"rr privileged read hit", 2}, /* INTERRUPT_CLR[2] */
{"rr secure write hit", 2}, /* INTERRUPT_CLR[2] */
{"rr secure read hit", 2}, /* INTERRUPT_CLR[2] */
{"bist_fail no use", 2}, /* INTERRUPT_CLR[2] */
{"bist_fail no use", 2}, /* INTERRUPT_CLR[2] */
{"bist_fail no use", 2}, /* INTERRUPT_CLR[2] */
{"bist_fail no use", 2}, /* INTERRUPT_CLR[2] */
{"slave error", 16}, /* INTERRUPT_CLR[16] */
{"dec error", 17}, /* INTERRUPT_CLR[17] */
{"burst fifo full", 2} /* INTERRUPT_CLR[2] */
};
struct gaudi2_cache_invld_params {
u64 start_va;
u64 end_va;
u32 inv_start_val;
u32 flags;
bool range_invalidation;
};
struct gaudi2_tpc_idle_data {
habanalabs: removed seq_file parameter from is_idle asic functions Change is_idle functions so it would be more usable outside debugfs. Do this by replacing seq_file parameter with regular string. Signed-off-by: Dani Liberman <dliberman@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-07-03 22:40:57 +08:00
struct engines_data *e;
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
unsigned long *mask;
bool *is_idle;
const char *tpc_fmt;
};
struct gaudi2_tpc_mmu_data {
u32 rw_asid;
};
static s64 gaudi2_state_dump_specs_props[SP_MAX] = {0};
static int gaudi2_memset_device_memory(struct hl_device *hdev, u64 addr, u64 size, u64 val);
static bool gaudi2_is_queue_enabled(struct hl_device *hdev, u32 hw_queue_id);
static bool gaudi2_is_arc_enabled(struct hl_device *hdev, u64 arc_id);
static void gaudi2_clr_arc_id_cap(struct hl_device *hdev, u64 arc_id);
static void gaudi2_set_arc_id_cap(struct hl_device *hdev, u64 arc_id);
static void gaudi2_memset_device_lbw(struct hl_device *hdev, u32 addr, u32 size, u32 val);
static int gaudi2_send_job_to_kdma(struct hl_device *hdev, u64 src_addr, u64 dst_addr, u32 size,
bool is_memset);
static u64 gaudi2_mmu_scramble_addr(struct hl_device *hdev, u64 raw_addr);
static void gaudi2_init_scrambler_hbm(struct hl_device *hdev)
{
}
static u32 gaudi2_get_signal_cb_size(struct hl_device *hdev)
{
return sizeof(struct packet_msg_short);
}
static u32 gaudi2_get_wait_cb_size(struct hl_device *hdev)
{
return sizeof(struct packet_msg_short) * 4 + sizeof(struct packet_fence);
}
void gaudi2_iterate_tpcs(struct hl_device *hdev, struct iterate_module_ctx *ctx)
{
struct asic_fixed_properties *prop = &hdev->asic_prop;
int dcore, inst, tpc_seq;
u32 offset;
habanalabs: add return code field to module iterator Up until now the module iterator called void callback functions and so caller activating callback that may fail suffered from 2 issues: 1. The need to "plant" return called in the private data. This is a drawback since the iterator itself should not be aware of the private data of the caller. 2. Due to 1 even in a failure the iterator would keep iterating instead of break upon error. To overcome this an optional rc field added to the iterator context. Signed-off-by: Ohad Sharabi <osharabi@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-07-13 18:47:23 +08:00
/* init the return code */
ctx->rc = 0;
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
for (dcore = 0; dcore < NUM_OF_DCORES; dcore++) {
for (inst = 0; inst < NUM_OF_TPC_PER_DCORE; inst++) {
tpc_seq = dcore * NUM_OF_TPC_PER_DCORE + inst;
if (!(prop->tpc_enabled_mask & BIT(tpc_seq)))
continue;
offset = (DCORE_OFFSET * dcore) + (DCORE_TPC_OFFSET * inst);
habanalabs: add return code field to module iterator Up until now the module iterator called void callback functions and so caller activating callback that may fail suffered from 2 issues: 1. The need to "plant" return called in the private data. This is a drawback since the iterator itself should not be aware of the private data of the caller. 2. Due to 1 even in a failure the iterator would keep iterating instead of break upon error. To overcome this an optional rc field added to the iterator context. Signed-off-by: Ohad Sharabi <osharabi@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-07-13 18:47:23 +08:00
ctx->fn(hdev, dcore, inst, offset, ctx);
if (ctx->rc) {
dev_err(hdev->dev, "TPC iterator failed for DCORE%d TPC%d\n",
dcore, inst);
return;
}
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
}
}
if (!(prop->tpc_enabled_mask & BIT(TPC_ID_DCORE0_TPC6)))
return;
/* special check for PCI TPC (DCORE0_TPC6) */
offset = DCORE_TPC_OFFSET * (NUM_DCORE0_TPC - 1);
habanalabs: add return code field to module iterator Up until now the module iterator called void callback functions and so caller activating callback that may fail suffered from 2 issues: 1. The need to "plant" return called in the private data. This is a drawback since the iterator itself should not be aware of the private data of the caller. 2. Due to 1 even in a failure the iterator would keep iterating instead of break upon error. To overcome this an optional rc field added to the iterator context. Signed-off-by: Ohad Sharabi <osharabi@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-07-13 18:47:23 +08:00
ctx->fn(hdev, 0, NUM_DCORE0_TPC - 1, offset, ctx);
if (ctx->rc)
dev_err(hdev->dev, "TPC iterator failed for DCORE0 TPC6\n");
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
}
static bool gaudi2_host_phys_addr_valid(u64 addr)
{
if ((addr < HOST_PHYS_BASE_0 + HOST_PHYS_SIZE_0) || (addr >= HOST_PHYS_BASE_1))
return true;
return false;
}
static int set_number_of_functional_hbms(struct hl_device *hdev)
{
struct asic_fixed_properties *prop = &hdev->asic_prop;
u8 faulty_hbms = hweight64(hdev->dram_binning);
/* check if all HBMs should be used */
if (!faulty_hbms) {
dev_dbg(hdev->dev, "All HBM are in use (no binning)\n");
prop->num_functional_hbms = GAUDI2_HBM_NUM;
return 0;
}
/*
* check for error condition in which number of binning
* candidates is higher than the maximum supported by the
* driver (in which case binning mask shall be ignored and driver will
* set the default)
*/
if (faulty_hbms > MAX_FAULTY_HBMS) {
dev_err(hdev->dev,
"HBM binning supports max of %d faulty HBMs, supplied mask 0x%llx.\n",
MAX_FAULTY_HBMS, hdev->dram_binning);
return -EINVAL;
}
/*
* by default, number of functional HBMs in Gaudi2 is always
* GAUDI2_HBM_NUM - 1.
*/
prop->num_functional_hbms = GAUDI2_HBM_NUM - faulty_hbms;
return 0;
}
static int gaudi2_set_dram_properties(struct hl_device *hdev)
{
struct asic_fixed_properties *prop = &hdev->asic_prop;
u32 basic_hbm_page_size;
int rc;
rc = set_number_of_functional_hbms(hdev);
if (rc)
return -EINVAL;
/*
* Due to HW bug in which TLB size is x16 smaller than expected we use a workaround
* in which we are using x16 bigger page size to be able to populate the entire
* HBM mappings in the TLB
*/
basic_hbm_page_size = prop->num_functional_hbms * SZ_8M;
prop->dram_page_size = GAUDI2_COMPENSATE_TLB_PAGE_SIZE_FACTOR * basic_hbm_page_size;
prop->device_mem_alloc_default_page_size = prop->dram_page_size;
prop->dram_size = prop->num_functional_hbms * SZ_16G;
prop->dram_base_address = DRAM_PHYS_BASE;
prop->dram_end_address = prop->dram_base_address + prop->dram_size;
prop->dram_supports_virtual_memory = true;
prop->dram_user_base_address = DRAM_PHYS_BASE + prop->dram_page_size;
prop->dram_hints_align_mask = ~GAUDI2_HBM_MMU_SCRM_ADDRESS_MASK;
prop->hints_dram_reserved_va_range.start_addr = RESERVED_VA_RANGE_FOR_ARC_ON_HBM_START;
prop->hints_dram_reserved_va_range.end_addr = RESERVED_VA_RANGE_FOR_ARC_ON_HBM_END;
habanalabs/gaudi2: use DIV_ROUND_UP_SECTOR_T instead of roundup roundup will create an error in 32-bit architectures as we use 64-bit variables. Reported-by: kernel test robot <lkp@intel.com> Signed-off-by: Ohad Sharabi <osharabi@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-07-07 21:03:14 +08:00
/* since DRAM page size differs from DMMU page size we need to allocate
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
* DRAM memory in units of dram_page size and mapping this memory in
* units of DMMU page size. we overcome this size mismatch using a
habanalabs/gaudi2: use DIV_ROUND_UP_SECTOR_T instead of roundup roundup will create an error in 32-bit architectures as we use 64-bit variables. Reported-by: kernel test robot <lkp@intel.com> Signed-off-by: Ohad Sharabi <osharabi@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-07-07 21:03:14 +08:00
* scrambling routine which takes a DRAM page and converts it to a DMMU
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
* page.
* We therefore:
* 1. partition the virtual address space to DRAM-page (whole) pages.
* (suppose we get n such pages)
* 2. limit the amount of virtual address space we got from 1 above to
* a multiple of 64M as we don't want the scrambled address to cross
* the DRAM virtual address space.
* ( m = (n * DRAM_page_size) / DMMU_page_size).
* 3. determine the and address accordingly
* end_addr = start_addr + m * 48M
*
* the DRAM address MSBs (63:48) are not part of the roundup calculation
*/
prop->dmmu.start_addr = prop->dram_base_address +
habanalabs/gaudi2: use DIV_ROUND_UP_SECTOR_T instead of roundup roundup will create an error in 32-bit architectures as we use 64-bit variables. Reported-by: kernel test robot <lkp@intel.com> Signed-off-by: Ohad Sharabi <osharabi@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-07-07 21:03:14 +08:00
(prop->dram_page_size *
DIV_ROUND_UP_SECTOR_T(prop->dram_size, prop->dram_page_size));
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
prop->dmmu.end_addr = prop->dmmu.start_addr + prop->dram_page_size *
div_u64((VA_HBM_SPACE_END - prop->dmmu.start_addr), prop->dmmu.page_size);
return 0;
}
static int gaudi2_set_fixed_properties(struct hl_device *hdev)
{
struct asic_fixed_properties *prop = &hdev->asic_prop;
struct hw_queue_properties *q_props;
u32 num_sync_stream_queues = 0;
int i;
prop->max_queues = GAUDI2_QUEUE_ID_SIZE;
prop->hw_queues_props = kcalloc(prop->max_queues, sizeof(struct hw_queue_properties),
GFP_KERNEL);
if (!prop->hw_queues_props)
return -ENOMEM;
q_props = prop->hw_queues_props;
for (i = 0 ; i < GAUDI2_QUEUE_ID_CPU_PQ ; i++) {
q_props[i].type = QUEUE_TYPE_HW;
q_props[i].driver_only = 0;
if (i >= GAUDI2_QUEUE_ID_NIC_0_0 && i <= GAUDI2_QUEUE_ID_NIC_23_3) {
q_props[i].supports_sync_stream = 0;
} else {
q_props[i].supports_sync_stream = 1;
num_sync_stream_queues++;
}
q_props[i].cb_alloc_flags = CB_ALLOC_USER;
}
q_props[GAUDI2_QUEUE_ID_CPU_PQ].type = QUEUE_TYPE_CPU;
q_props[GAUDI2_QUEUE_ID_CPU_PQ].driver_only = 1;
q_props[GAUDI2_QUEUE_ID_CPU_PQ].cb_alloc_flags = CB_ALLOC_KERNEL;
prop->cache_line_size = DEVICE_CACHE_LINE_SIZE;
prop->cfg_base_address = CFG_BASE;
prop->device_dma_offset_for_host_access = HOST_PHYS_BASE_0;
prop->host_base_address = HOST_PHYS_BASE_0;
prop->host_end_address = prop->host_base_address + HOST_PHYS_SIZE_0;
prop->max_pending_cs = GAUDI2_MAX_PENDING_CS;
prop->completion_queues_count = GAUDI2_RESERVED_CQ_NUMBER;
prop->user_dec_intr_count = NUMBER_OF_DEC;
prop->user_interrupt_count = GAUDI2_IRQ_NUM_USER_LAST - GAUDI2_IRQ_NUM_USER_FIRST + 1;
prop->completion_mode = HL_COMPLETION_MODE_CS;
habanalabs/gaudi2: replace defines for reserved sob/mob with enums Following patches are going to add more reserved sync objects and monitors. To make the counting of these reserved resources simpler, replace the existing RESERVED_* defines with enumerations. Signed-off-by: Tomer Tayar <ttayar@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-30 00:20:38 +08:00
prop->sync_stream_first_sob = GAUDI2_RESERVED_SOB_NUMBER;
prop->sync_stream_first_mon = GAUDI2_RESERVED_MON_NUMBER;
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
prop->sram_base_address = SRAM_BASE_ADDR;
prop->sram_size = SRAM_SIZE;
prop->sram_end_address = prop->sram_base_address + prop->sram_size;
prop->sram_user_base_address = prop->sram_base_address + SRAM_USER_BASE_OFFSET;
prop->hints_range_reservation = true;
if (hdev->pldm)
prop->mmu_pgt_size = 0x800000; /* 8MB */
else
prop->mmu_pgt_size = MMU_PAGE_TABLES_INITIAL_SIZE;
prop->mmu_pte_size = HL_PTE_SIZE;
prop->mmu_hop_table_size = HOP_TABLE_SIZE_512_PTE;
prop->mmu_hop0_tables_total_size = HOP0_512_PTE_TABLES_TOTAL_SIZE;
prop->dmmu.hop_shifts[MMU_HOP0] = DHOP0_SHIFT;
prop->dmmu.hop_shifts[MMU_HOP1] = DHOP1_SHIFT;
prop->dmmu.hop_shifts[MMU_HOP2] = DHOP2_SHIFT;
prop->dmmu.hop_shifts[MMU_HOP3] = DHOP3_SHIFT;
prop->dmmu.hop_shifts[MMU_HOP4] = DHOP4_SHIFT;
prop->dmmu.hop_masks[MMU_HOP0] = DHOP0_MASK;
prop->dmmu.hop_masks[MMU_HOP1] = DHOP1_MASK;
prop->dmmu.hop_masks[MMU_HOP2] = DHOP2_MASK;
prop->dmmu.hop_masks[MMU_HOP3] = DHOP3_MASK;
prop->dmmu.hop_masks[MMU_HOP4] = DHOP4_MASK;
prop->dmmu.page_size = PAGE_SIZE_1GB;
prop->dmmu.num_hops = MMU_ARCH_6_HOPS;
prop->dmmu.last_mask = LAST_MASK;
prop->dmmu.host_resident = 1;
/* TODO: will be duplicated until implementing per-MMU props */
prop->dmmu.hop_table_size = prop->mmu_hop_table_size;
prop->dmmu.hop0_tables_total_size = prop->mmu_hop0_tables_total_size;
/*
* this is done in order to be able to validate FW descriptor (i.e. validating that
* the addresses and allocated space for FW image does not cross memory bounds).
* for this reason we set the DRAM size to the minimum possible and later it will
* be modified according to what reported in the cpucp info packet
*/
prop->dram_size = (GAUDI2_HBM_NUM - 1) * SZ_16G;
hdev->pmmu_huge_range = true;
prop->pmmu.host_resident = 1;
prop->pmmu.num_hops = MMU_ARCH_6_HOPS;
prop->pmmu.last_mask = LAST_MASK;
/* TODO: will be duplicated until implementing per-MMU props */
prop->pmmu.hop_table_size = prop->mmu_hop_table_size;
prop->pmmu.hop0_tables_total_size = prop->mmu_hop0_tables_total_size;
habanalabs/gaudi2: map virtual MSI-X doorbell memory for user Upon the initialization of a user context, map the host memory page of the virtual MSI-X doorbell in the device MMU. A reserved VA is used for this purpose, so user can use it directly without any allocation/map operation. Signed-off-by: Tomer Tayar <ttayar@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-07-01 03:05:51 +08:00
prop->hints_host_reserved_va_range.start_addr = RESERVED_VA_FOR_VIRTUAL_MSIX_DOORBELL_START;
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
prop->hints_host_reserved_va_range.end_addr = RESERVED_VA_RANGE_FOR_ARC_ON_HOST_END;
prop->hints_host_hpage_reserved_va_range.start_addr =
RESERVED_VA_RANGE_FOR_ARC_ON_HOST_HPAGE_START;
prop->hints_host_hpage_reserved_va_range.end_addr =
RESERVED_VA_RANGE_FOR_ARC_ON_HOST_HPAGE_END;
if (PAGE_SIZE == SZ_64K) {
prop->pmmu.hop_shifts[MMU_HOP0] = HOP0_SHIFT_64K;
prop->pmmu.hop_shifts[MMU_HOP1] = HOP1_SHIFT_64K;
prop->pmmu.hop_shifts[MMU_HOP2] = HOP2_SHIFT_64K;
prop->pmmu.hop_shifts[MMU_HOP3] = HOP3_SHIFT_64K;
prop->pmmu.hop_shifts[MMU_HOP4] = HOP4_SHIFT_64K;
prop->pmmu.hop_shifts[MMU_HOP5] = HOP5_SHIFT_64K;
prop->pmmu.hop_masks[MMU_HOP0] = HOP0_MASK_64K;
prop->pmmu.hop_masks[MMU_HOP1] = HOP1_MASK_64K;
prop->pmmu.hop_masks[MMU_HOP2] = HOP2_MASK_64K;
prop->pmmu.hop_masks[MMU_HOP3] = HOP3_MASK_64K;
prop->pmmu.hop_masks[MMU_HOP4] = HOP4_MASK_64K;
prop->pmmu.hop_masks[MMU_HOP5] = HOP5_MASK_64K;
prop->pmmu.start_addr = VA_HOST_SPACE_PAGE_START;
prop->pmmu.end_addr = VA_HOST_SPACE_PAGE_END;
prop->pmmu.page_size = PAGE_SIZE_64KB;
/* shifts and masks are the same in PMMU and HPMMU */
memcpy(&prop->pmmu_huge, &prop->pmmu, sizeof(prop->pmmu));
prop->pmmu_huge.page_size = PAGE_SIZE_16MB;
prop->pmmu_huge.start_addr = VA_HOST_SPACE_HPAGE_START;
prop->pmmu_huge.end_addr = VA_HOST_SPACE_HPAGE_END;
} else {
prop->pmmu.hop_shifts[MMU_HOP0] = HOP0_SHIFT_4K;
prop->pmmu.hop_shifts[MMU_HOP1] = HOP1_SHIFT_4K;
prop->pmmu.hop_shifts[MMU_HOP2] = HOP2_SHIFT_4K;
prop->pmmu.hop_shifts[MMU_HOP3] = HOP3_SHIFT_4K;
prop->pmmu.hop_shifts[MMU_HOP4] = HOP4_SHIFT_4K;
prop->pmmu.hop_shifts[MMU_HOP5] = HOP5_SHIFT_4K;
prop->pmmu.hop_masks[MMU_HOP0] = HOP0_MASK_4K;
prop->pmmu.hop_masks[MMU_HOP1] = HOP1_MASK_4K;
prop->pmmu.hop_masks[MMU_HOP2] = HOP2_MASK_4K;
prop->pmmu.hop_masks[MMU_HOP3] = HOP3_MASK_4K;
prop->pmmu.hop_masks[MMU_HOP4] = HOP4_MASK_4K;
prop->pmmu.hop_masks[MMU_HOP5] = HOP5_MASK_4K;
prop->pmmu.start_addr = VA_HOST_SPACE_PAGE_START;
prop->pmmu.end_addr = VA_HOST_SPACE_PAGE_END;
prop->pmmu.page_size = PAGE_SIZE_4KB;
/* shifts and masks are the same in PMMU and HPMMU */
memcpy(&prop->pmmu_huge, &prop->pmmu, sizeof(prop->pmmu));
prop->pmmu_huge.page_size = PAGE_SIZE_2MB;
prop->pmmu_huge.start_addr = VA_HOST_SPACE_HPAGE_START;
prop->pmmu_huge.end_addr = VA_HOST_SPACE_HPAGE_END;
}
habanalabs/gaudi2: new API to control engine cores running mode The current flow of halting the engine cores is implemented by command buffers built by the user space and sent towards the Driver. This current flow is broken since the user space does not know when the cores actually halt as sending a workload is async op. Therefore the application can not free the memory that is mapped to the engine cores. This new API allows the user space to control the running mode. The API call is sync (returns after the cores are set to the requested mode). Signed-off-by: Tal Cohen <talcohen@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-07-07 23:42:47 +08:00
prop->num_engine_cores = CPU_ID_MAX;
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
prop->cfg_size = CFG_SIZE;
prop->max_asid = MAX_ASID;
prop->num_of_events = GAUDI2_EVENT_SIZE;
prop->dc_power_default = DC_POWER_DEFAULT;
prop->cb_pool_cb_cnt = GAUDI2_CB_POOL_CB_CNT;
prop->cb_pool_cb_size = GAUDI2_CB_POOL_CB_SIZE;
prop->pcie_dbi_base_address = CFG_BASE + mmPCIE_DBI_BASE;
prop->pcie_aux_dbi_reg_addr = CFG_BASE + mmPCIE_AUX_DBI;
strncpy(prop->cpucp_info.card_name, GAUDI2_DEFAULT_CARD_NAME, CARD_NAME_MAX_LEN);
prop->mme_master_slave_mode = 1;
habanalabs/gaudi2: replace defines for reserved sob/mob with enums Following patches are going to add more reserved sync objects and monitors. To make the counting of these reserved resources simpler, replace the existing RESERVED_* defines with enumerations. Signed-off-by: Tomer Tayar <ttayar@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-30 00:20:38 +08:00
prop->first_available_user_sob[0] = GAUDI2_RESERVED_SOB_NUMBER +
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
(num_sync_stream_queues * HL_RSVD_SOBS);
habanalabs/gaudi2: replace defines for reserved sob/mob with enums Following patches are going to add more reserved sync objects and monitors. To make the counting of these reserved resources simpler, replace the existing RESERVED_* defines with enumerations. Signed-off-by: Tomer Tayar <ttayar@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-30 00:20:38 +08:00
prop->first_available_user_mon[0] = GAUDI2_RESERVED_MON_NUMBER +
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
(num_sync_stream_queues * HL_RSVD_MONS);
prop->first_available_user_interrupt = GAUDI2_IRQ_NUM_USER_FIRST;
prop->first_available_cq[0] = GAUDI2_RESERVED_CQ_NUMBER;
prop->fw_cpu_boot_dev_sts0_valid = false;
prop->fw_cpu_boot_dev_sts1_valid = false;
prop->hard_reset_done_by_fw = false;
prop->gic_interrupts_enable = true;
prop->server_type = HL_SERVER_TYPE_UNKNOWN;
prop->max_dec = NUMBER_OF_DEC;
prop->clk_pll_index = HL_GAUDI2_MME_PLL;
prop->dma_mask = 64;
return 0;
}
static int gaudi2_pci_bars_map(struct hl_device *hdev)
{
static const char * const name[] = {"CFG_SRAM", "MSIX", "DRAM"};
bool is_wc[3] = {false, false, true};
int rc;
rc = hl_pci_bars_map(hdev, name, is_wc);
if (rc)
return rc;
hdev->rmmio = hdev->pcie_bar[SRAM_CFG_BAR_ID] + (CFG_BASE - STM_FLASH_BASE_ADDR);
return 0;
}
static u64 gaudi2_set_hbm_bar_base(struct hl_device *hdev, u64 addr)
{
struct gaudi2_device *gaudi2 = hdev->asic_specific;
struct hl_inbound_pci_region pci_region;
u64 old_addr = addr;
int rc;
if ((gaudi2) && (gaudi2->dram_bar_cur_addr == addr))
return old_addr;
if (hdev->asic_prop.iatu_done_by_fw)
return U64_MAX;
/* Inbound Region 2 - Bar 4 - Point to DRAM */
pci_region.mode = PCI_BAR_MATCH_MODE;
pci_region.bar = DRAM_BAR_ID;
pci_region.addr = addr;
rc = hl_pci_set_inbound_region(hdev, 2, &pci_region);
if (rc)
return U64_MAX;
if (gaudi2) {
old_addr = gaudi2->dram_bar_cur_addr;
gaudi2->dram_bar_cur_addr = addr;
}
return old_addr;
}
static int gaudi2_init_iatu(struct hl_device *hdev)
{
struct hl_inbound_pci_region inbound_region;
struct hl_outbound_pci_region outbound_region;
u32 bar_addr_low, bar_addr_high;
int rc;
if (hdev->asic_prop.iatu_done_by_fw)
return 0;
/* Temporary inbound Region 0 - Bar 0 - Point to CFG
* We must map this region in BAR match mode in order to
* fetch BAR physical base address
*/
inbound_region.mode = PCI_BAR_MATCH_MODE;
inbound_region.bar = SRAM_CFG_BAR_ID;
/* Base address must be aligned to Bar size which is 256 MB */
inbound_region.addr = STM_FLASH_BASE_ADDR - STM_FLASH_ALIGNED_OFF;
rc = hl_pci_set_inbound_region(hdev, 0, &inbound_region);
if (rc)
return rc;
/* Fetch physical BAR address */
bar_addr_high = RREG32(mmPCIE_DBI_BAR1_REG + STM_FLASH_ALIGNED_OFF);
bar_addr_low = RREG32(mmPCIE_DBI_BAR0_REG + STM_FLASH_ALIGNED_OFF) & ~0xF;
hdev->pcie_bar_phys[SRAM_CFG_BAR_ID] = (u64)bar_addr_high << 32 | bar_addr_low;
/* Inbound Region 0 - Bar 0 - Point to CFG */
inbound_region.mode = PCI_ADDRESS_MATCH_MODE;
inbound_region.bar = SRAM_CFG_BAR_ID;
inbound_region.offset_in_bar = 0;
inbound_region.addr = STM_FLASH_BASE_ADDR;
inbound_region.size = CFG_REGION_SIZE;
rc = hl_pci_set_inbound_region(hdev, 0, &inbound_region);
if (rc)
return rc;
/* Inbound Region 1 - Bar 0 - Point to BAR0_RESERVED + SRAM */
inbound_region.mode = PCI_ADDRESS_MATCH_MODE;
inbound_region.bar = SRAM_CFG_BAR_ID;
inbound_region.offset_in_bar = CFG_REGION_SIZE;
inbound_region.addr = BAR0_RSRVD_BASE_ADDR;
inbound_region.size = BAR0_RSRVD_SIZE + SRAM_SIZE;
rc = hl_pci_set_inbound_region(hdev, 1, &inbound_region);
if (rc)
return rc;
/* Inbound Region 2 - Bar 4 - Point to DRAM */
inbound_region.mode = PCI_BAR_MATCH_MODE;
inbound_region.bar = DRAM_BAR_ID;
inbound_region.addr = DRAM_PHYS_BASE;
rc = hl_pci_set_inbound_region(hdev, 2, &inbound_region);
if (rc)
return rc;
/* Outbound Region 0 - Point to Host */
outbound_region.addr = HOST_PHYS_BASE_0;
outbound_region.size = HOST_PHYS_SIZE_0;
rc = hl_pci_set_outbound_region(hdev, &outbound_region);
return rc;
}
static enum hl_device_hw_state gaudi2_get_hw_state(struct hl_device *hdev)
{
return RREG32(mmHW_STATE);
}
static int gaudi2_tpc_binning_init_prop(struct hl_device *hdev)
{
struct asic_fixed_properties *prop = &hdev->asic_prop;
/*
* check for error condition in which number of binning candidates
* is higher than the maximum supported by the driver
*/
if (hweight64(hdev->tpc_binning) > MAX_CLUSTER_BINNING_FAULTY_TPCS) {
dev_err(hdev->dev, "TPC binning is supported for max of %d faulty TPCs, provided mask 0x%llx\n",
MAX_CLUSTER_BINNING_FAULTY_TPCS,
hdev->tpc_binning);
return -EINVAL;
}
prop->tpc_binning_mask = hdev->tpc_binning;
prop->tpc_enabled_mask = GAUDI2_TPC_FULL_MASK;
return 0;
}
static int gaudi2_set_tpc_binning_masks(struct hl_device *hdev)
{
struct asic_fixed_properties *prop = &hdev->asic_prop;
struct hw_queue_properties *q_props = prop->hw_queues_props;
u64 tpc_binning_mask;
u8 subst_idx = 0;
int i, rc;
rc = gaudi2_tpc_binning_init_prop(hdev);
if (rc)
return rc;
tpc_binning_mask = prop->tpc_binning_mask;
for (i = 0 ; i < MAX_FAULTY_TPCS ; i++) {
u8 subst_seq, binned, qid_base;
if (tpc_binning_mask == 0)
break;
if (subst_idx == 0) {
subst_seq = TPC_ID_DCORE0_TPC6;
qid_base = GAUDI2_QUEUE_ID_DCORE0_TPC_6_0;
} else {
subst_seq = TPC_ID_DCORE3_TPC5;
qid_base = GAUDI2_QUEUE_ID_DCORE3_TPC_5_0;
}
/* clear bit from mask */
binned = __ffs(tpc_binning_mask);
/*
* Coverity complains about possible out-of-bound access in
* clear_bit
*/
if (binned >= TPC_ID_SIZE) {
dev_err(hdev->dev,
"Invalid binned TPC (binning mask: %llx)\n",
tpc_binning_mask);
return -EINVAL;
}
clear_bit(binned, (unsigned long *)&tpc_binning_mask);
/* also clear replacing TPC bit from enabled mask */
clear_bit(subst_seq, (unsigned long *)&prop->tpc_enabled_mask);
/* bin substite TPC's Qs */
q_props[qid_base].binned = 1;
q_props[qid_base + 1].binned = 1;
q_props[qid_base + 2].binned = 1;
q_props[qid_base + 3].binned = 1;
subst_idx++;
}
return 0;
}
static int gaudi2_set_dec_binning_masks(struct hl_device *hdev)
{
struct asic_fixed_properties *prop = &hdev->asic_prop;
u8 num_faulty;
num_faulty = hweight32(hdev->decoder_binning);
/*
* check for error condition in which number of binning candidates
* is higher than the maximum supported by the driver
*/
if (num_faulty > MAX_FAULTY_DECODERS) {
dev_err(hdev->dev, "decoder binning is supported for max of single faulty decoder, provided mask 0x%x\n",
hdev->decoder_binning);
return -EINVAL;
}
prop->decoder_binning_mask = (hdev->decoder_binning & GAUDI2_DECODER_FULL_MASK);
if (prop->decoder_binning_mask)
prop->decoder_enabled_mask = (GAUDI2_DECODER_FULL_MASK & ~BIT(DEC_ID_PCIE_VDEC1));
else
prop->decoder_enabled_mask = GAUDI2_DECODER_FULL_MASK;
return 0;
}
static void gaudi2_set_dram_binning_masks(struct hl_device *hdev)
{
struct asic_fixed_properties *prop = &hdev->asic_prop;
/* check if we should override default binning */
if (!hdev->dram_binning) {
prop->dram_binning_mask = 0;
prop->dram_enabled_mask = GAUDI2_DRAM_FULL_MASK;
return;
}
/* set DRAM binning constraints */
prop->faulty_dram_cluster_map |= hdev->dram_binning;
prop->dram_binning_mask = hdev->dram_binning;
prop->dram_enabled_mask = GAUDI2_DRAM_FULL_MASK & ~BIT(HBM_ID5);
}
static int gaudi2_set_edma_binning_masks(struct hl_device *hdev)
{
struct asic_fixed_properties *prop = &hdev->asic_prop;
struct hw_queue_properties *q_props;
u8 seq, num_faulty;
num_faulty = hweight32(hdev->edma_binning);
/*
* check for error condition in which number of binning candidates
* is higher than the maximum supported by the driver
*/
if (num_faulty > MAX_FAULTY_EDMAS) {
dev_err(hdev->dev,
"EDMA binning is supported for max of single faulty EDMA, provided mask 0x%x\n",
hdev->edma_binning);
return -EINVAL;
}
if (!hdev->edma_binning) {
prop->edma_binning_mask = 0;
prop->edma_enabled_mask = GAUDI2_EDMA_FULL_MASK;
return 0;
}
seq = __ffs((unsigned long)hdev->edma_binning);
/* set binning constraints */
prop->faulty_dram_cluster_map |= BIT(edma_to_hbm_cluster[seq]);
prop->edma_binning_mask = hdev->edma_binning;
prop->edma_enabled_mask = GAUDI2_EDMA_FULL_MASK & ~BIT(EDMA_ID_DCORE3_INSTANCE1);
/* bin substitute EDMA's queue */
q_props = prop->hw_queues_props;
q_props[GAUDI2_QUEUE_ID_DCORE3_EDMA_1_0].binned = 1;
q_props[GAUDI2_QUEUE_ID_DCORE3_EDMA_1_1].binned = 1;
q_props[GAUDI2_QUEUE_ID_DCORE3_EDMA_1_2].binned = 1;
q_props[GAUDI2_QUEUE_ID_DCORE3_EDMA_1_3].binned = 1;
return 0;
}
static int gaudi2_set_xbar_edge_enable_mask(struct hl_device *hdev, u32 xbar_edge_iso_mask)
{
struct asic_fixed_properties *prop = &hdev->asic_prop;
u8 num_faulty, seq;
/* check if we should override default binning */
if (!xbar_edge_iso_mask) {
prop->xbar_edge_enabled_mask = GAUDI2_XBAR_EDGE_FULL_MASK;
return 0;
}
/*
* note that it can be set to value other than 0 only after cpucp packet (i.e.
* only the FW can set a redundancy value). for user it'll always be 0.
*/
num_faulty = hweight32(xbar_edge_iso_mask);
/*
* check for error condition in which number of binning candidates
* is higher than the maximum supported by the driver
*/
if (num_faulty > MAX_FAULTY_XBARS) {
dev_err(hdev->dev, "we cannot have more than %d faulty XBAR EDGE\n",
MAX_FAULTY_XBARS);
return -EINVAL;
}
seq = __ffs((unsigned long)xbar_edge_iso_mask);
/* set binning constraints */
prop->faulty_dram_cluster_map |= BIT(xbar_edge_to_hbm_cluster[seq]);
prop->xbar_edge_enabled_mask = (~xbar_edge_iso_mask) & GAUDI2_XBAR_EDGE_FULL_MASK;
return 0;
}
static int gaudi2_set_cluster_binning_masks_common(struct hl_device *hdev, u8 xbar_edge_iso_mask)
{
int rc;
/*
* mark all clusters as good, each component will "fail" cluster
* based on eFuse/user values.
* If more than single cluster is faulty- the chip is unusable
*/
hdev->asic_prop.faulty_dram_cluster_map = 0;
gaudi2_set_dram_binning_masks(hdev);
rc = gaudi2_set_edma_binning_masks(hdev);
if (rc)
return rc;
rc = gaudi2_set_xbar_edge_enable_mask(hdev, xbar_edge_iso_mask);
if (rc)
return rc;
/* always initially set to full mask */
hdev->asic_prop.hmmu_hif_enabled_mask = GAUDI2_HIF_HMMU_FULL_MASK;
return 0;
}
static int gaudi2_set_cluster_binning_masks(struct hl_device *hdev)
{
struct asic_fixed_properties *prop = &hdev->asic_prop;
int rc;
rc = gaudi2_set_cluster_binning_masks_common(hdev, prop->cpucp_info.xbar_binning_mask);
if (rc)
return rc;
/* if we have DRAM binning reported by FW we should perform cluster config */
if (prop->faulty_dram_cluster_map) {
u8 cluster_seq = __ffs((unsigned long)prop->faulty_dram_cluster_map);
prop->hmmu_hif_enabled_mask = cluster_hmmu_hif_enabled_mask[cluster_seq];
}
return 0;
}
static int gaudi2_cpucp_info_get(struct hl_device *hdev)
{
struct gaudi2_device *gaudi2 = hdev->asic_specific;
struct asic_fixed_properties *prop = &hdev->asic_prop;
long max_power;
u64 dram_size;
int rc;
if (!(gaudi2->hw_cap_initialized & HW_CAP_CPU_Q))
return 0;
/* No point of asking this information again when not doing hard reset, as the device
* CPU hasn't been reset
*/
habanalabs: rename soft reset to compute reset Doing compute reset can be the traditional inference soft reset that is supported only in Goya. Or it can be the new reset upon device release, which is supported in Gaudi2 and above. Therefore, wherever suitable, use the terminology of compute reset instead of soft reset. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-07-07 17:00:24 +08:00
if (hdev->reset_info.in_compute_reset)
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
return 0;
rc = hl_fw_cpucp_handshake(hdev, mmCPU_BOOT_DEV_STS0, mmCPU_BOOT_DEV_STS1, mmCPU_BOOT_ERR0,
mmCPU_BOOT_ERR1);
if (rc)
return rc;
dram_size = le64_to_cpu(prop->cpucp_info.dram_size);
if (dram_size) {
/* we can have wither 5 or 6 HBMs. other values are invalid */
if ((dram_size != ((GAUDI2_HBM_NUM - 1) * SZ_16G)) &&
habanalabs/gaudi2: use DIV_ROUND_UP_SECTOR_T instead of roundup roundup will create an error in 32-bit architectures as we use 64-bit variables. Reported-by: kernel test robot <lkp@intel.com> Signed-off-by: Ohad Sharabi <osharabi@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-07-07 21:03:14 +08:00
(dram_size != (GAUDI2_HBM_NUM * SZ_16G))) {
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
dev_err(hdev->dev,
"F/W reported invalid DRAM size %llu. Trying to use default size %llu\n",
dram_size, prop->dram_size);
dram_size = prop->dram_size;
}
prop->dram_size = dram_size;
prop->dram_end_address = prop->dram_base_address + dram_size;
}
if (!strlen(prop->cpucp_info.card_name))
strncpy(prop->cpucp_info.card_name, GAUDI2_DEFAULT_CARD_NAME, CARD_NAME_MAX_LEN);
/* Overwrite binning masks with the actual binning values from F/W */
hdev->dram_binning = prop->cpucp_info.dram_binning_mask;
hdev->edma_binning = prop->cpucp_info.edma_binning_mask;
hdev->tpc_binning = le64_to_cpu(prop->cpucp_info.tpc_binning_mask);
hdev->decoder_binning = lower_32_bits(le64_to_cpu(prop->cpucp_info.decoder_binning_mask));
/*
* at this point the DRAM parameters need to be updated according to data obtained
* from the FW
*/
rc = gaudi2_set_dram_properties(hdev);
if (rc)
return rc;
rc = gaudi2_set_cluster_binning_masks(hdev);
if (rc)
return rc;
rc = gaudi2_set_tpc_binning_masks(hdev);
if (rc)
return rc;
rc = gaudi2_set_dec_binning_masks(hdev);
if (rc)
return rc;
max_power = hl_fw_get_max_power(hdev);
if (max_power < 0)
return max_power;
prop->max_power_default = (u64) max_power;
return 0;
}
static int gaudi2_fetch_psoc_frequency(struct hl_device *hdev)
{
struct gaudi2_device *gaudi2 = hdev->asic_specific;
u16 pll_freq_arr[HL_PLL_NUM_OUTPUTS];
int rc;
if (!(gaudi2->hw_cap_initialized & HW_CAP_CPU_Q))
return 0;
rc = hl_fw_cpucp_pll_info_get(hdev, HL_GAUDI2_CPU_PLL, pll_freq_arr);
if (rc)
return rc;
hdev->asic_prop.psoc_timestamp_frequency = pll_freq_arr[3];
return 0;
}
static int gaudi2_early_init(struct hl_device *hdev)
{
struct asic_fixed_properties *prop = &hdev->asic_prop;
struct pci_dev *pdev = hdev->pdev;
resource_size_t pci_bar_size;
int rc;
rc = gaudi2_set_fixed_properties(hdev);
if (rc)
return rc;
/* Check BAR sizes */
pci_bar_size = pci_resource_len(pdev, SRAM_CFG_BAR_ID);
if (pci_bar_size != CFG_BAR_SIZE) {
dev_err(hdev->dev, "Not " HL_NAME "? BAR %d size %pa, expecting %llu\n",
SRAM_CFG_BAR_ID, &pci_bar_size, CFG_BAR_SIZE);
rc = -ENODEV;
goto free_queue_props;
}
pci_bar_size = pci_resource_len(pdev, MSIX_BAR_ID);
if (pci_bar_size != MSIX_BAR_SIZE) {
dev_err(hdev->dev, "Not " HL_NAME "? BAR %d size %pa, expecting %llu\n",
MSIX_BAR_ID, &pci_bar_size, MSIX_BAR_SIZE);
rc = -ENODEV;
goto free_queue_props;
}
prop->dram_pci_bar_size = pci_resource_len(pdev, DRAM_BAR_ID);
hdev->dram_pci_bar_start = pci_resource_start(pdev, DRAM_BAR_ID);
habanalabs/gaudi2: change device f/w security check On Gaudi2 the f/w always configures the PCIe iATU and allows access to scratchpad registers. Therefore, we can know if the f/w is secured by reading a status bit from the f/w registers. Signed-off-by: farah kassabri <fkassabri@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-07-20 18:53:37 +08:00
/*
* Only in pldm driver config iATU
*/
if (hdev->pldm)
hdev->asic_prop.iatu_done_by_fw = false;
else
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
hdev->asic_prop.iatu_done_by_fw = true;
rc = hl_pci_init(hdev);
if (rc)
goto free_queue_props;
/* Before continuing in the initialization, we need to read the preboot
* version to determine whether we run with a security-enabled firmware
*/
habanalabs: wait for preboot ready after hard reset Currently we are not waiting for preboot ready after hard reset. This leads to a race in which COMMs protocol begins but will get no response from the f/w. Signed-off-by: Ohad Sharabi <osharabi@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-28 17:09:21 +08:00
rc = hl_fw_read_preboot_status(hdev);
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
if (rc) {
if (hdev->reset_on_preboot_fail)
hdev->asic_funcs->hw_fini(hdev, true, false);
goto pci_fini;
}
if (gaudi2_get_hw_state(hdev) == HL_DEVICE_HW_STATE_DIRTY) {
habanalabs: move h/w dirty message to debug H/W being dirty during initialization is completely expected in case f/w tools are used before loading the driver. As it is not an error, and as it doesn't give any meaningful information to the user, no point of printing it. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-07-10 19:14:04 +08:00
dev_dbg(hdev->dev, "H/W state is dirty, must reset before initializing\n");
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
hdev->asic_funcs->hw_fini(hdev, true, false);
}
return 0;
pci_fini:
hl_pci_fini(hdev);
free_queue_props:
kfree(hdev->asic_prop.hw_queues_props);
return rc;
}
static int gaudi2_early_fini(struct hl_device *hdev)
{
kfree(hdev->asic_prop.hw_queues_props);
hl_pci_fini(hdev);
return 0;
}
static bool gaudi2_is_arc_nic_owned(u64 arc_id)
{
switch (arc_id) {
case CPU_ID_NIC_QMAN_ARC0...CPU_ID_NIC_QMAN_ARC23:
return true;
default:
return false;
}
}
static bool gaudi2_is_arc_tpc_owned(u64 arc_id)
{
switch (arc_id) {
case CPU_ID_TPC_QMAN_ARC0...CPU_ID_TPC_QMAN_ARC24:
return true;
default:
return false;
}
}
static void gaudi2_init_arcs(struct hl_device *hdev)
{
struct gaudi2_device *gaudi2 = hdev->asic_specific;
u64 arc_id;
u32 i;
for (i = CPU_ID_SCHED_ARC0 ; i <= CPU_ID_SCHED_ARC3 ; i++) {
if (gaudi2_is_arc_enabled(hdev, i))
continue;
gaudi2_set_arc_id_cap(hdev, i);
}
for (i = GAUDI2_QUEUE_ID_PDMA_0_0 ; i < GAUDI2_QUEUE_ID_CPU_PQ ; i += 4) {
if (!gaudi2_is_queue_enabled(hdev, i))
continue;
arc_id = gaudi2_queue_id_to_arc_id[i];
if (gaudi2_is_arc_enabled(hdev, arc_id))
continue;
if (gaudi2_is_arc_nic_owned(arc_id) &&
!(hdev->nic_ports_mask & BIT_ULL(arc_id - CPU_ID_NIC_QMAN_ARC0)))
continue;
if (gaudi2_is_arc_tpc_owned(arc_id) && !(gaudi2->tpc_hw_cap_initialized &
BIT_ULL(arc_id - CPU_ID_TPC_QMAN_ARC0)))
continue;
gaudi2_set_arc_id_cap(hdev, arc_id);
}
}
static int gaudi2_scrub_arc_dccm(struct hl_device *hdev, u32 cpu_id)
{
u32 reg_base, reg_val;
int rc;
switch (cpu_id) {
case CPU_ID_SCHED_ARC0 ... CPU_ID_SCHED_ARC3:
/* Each ARC scheduler has 2 consecutive DCCM blocks */
rc = gaudi2_send_job_to_kdma(hdev, 0, CFG_BASE + gaudi2_arc_dccm_bases[cpu_id],
ARC_DCCM_BLOCK_SIZE * 2, true);
if (rc)
return rc;
break;
case CPU_ID_SCHED_ARC4:
case CPU_ID_SCHED_ARC5:
case CPU_ID_MME_QMAN_ARC0:
case CPU_ID_MME_QMAN_ARC1:
reg_base = gaudi2_arc_blocks_bases[cpu_id];
/* Scrub lower DCCM block */
rc = gaudi2_send_job_to_kdma(hdev, 0, CFG_BASE + gaudi2_arc_dccm_bases[cpu_id],
ARC_DCCM_BLOCK_SIZE, true);
if (rc)
return rc;
/* Switch to upper DCCM block */
reg_val = FIELD_PREP(ARC_FARM_ARC0_AUX_MME_ARC_UPPER_DCCM_EN_VAL_MASK, 1);
WREG32(reg_base + ARC_DCCM_UPPER_EN_OFFSET, reg_val);
/* Scrub upper DCCM block */
rc = gaudi2_send_job_to_kdma(hdev, 0, CFG_BASE + gaudi2_arc_dccm_bases[cpu_id],
ARC_DCCM_BLOCK_SIZE, true);
if (rc)
return rc;
/* Switch to lower DCCM block */
reg_val = FIELD_PREP(ARC_FARM_ARC0_AUX_MME_ARC_UPPER_DCCM_EN_VAL_MASK, 0);
WREG32(reg_base + ARC_DCCM_UPPER_EN_OFFSET, reg_val);
break;
default:
rc = gaudi2_send_job_to_kdma(hdev, 0, CFG_BASE + gaudi2_arc_dccm_bases[cpu_id],
ARC_DCCM_BLOCK_SIZE, true);
if (rc)
return rc;
}
return 0;
}
static void gaudi2_scrub_arcs_dccm(struct hl_device *hdev)
{
u16 arc_id;
for (arc_id = CPU_ID_SCHED_ARC0 ; arc_id < CPU_ID_MAX ; arc_id++) {
if (!gaudi2_is_arc_enabled(hdev, arc_id))
continue;
gaudi2_scrub_arc_dccm(hdev, arc_id);
}
}
static int gaudi2_late_init(struct hl_device *hdev)
{
habanalabs/gaudi2: configure virtual MSI-X doorbell interface Due to a watchdog timer in the LBW path, writes to the MSI-X doorbell can return sporadic error responses. To work-around this issue, a virtual MSI-X doorbell on the HBW path is configured, using the MSI-X AXI slave interface in the PCIe controller. Upon an access to a configured HBW host address, the controller will generate MSI-X interrupt instead of treating the access as regular host memory access. This patch allocates the dedicate host memory page, and communicate the address to F/W, so it will configure the relevant address match registers in the controller, and will use this address to generate MSI-X interrupts for F/W events. Following patches will handle other initiators in the device, to move them to use the virtual MSI-X doorbell. Signed-off-by: Tomer Tayar <ttayar@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-29 21:37:55 +08:00
struct gaudi2_device *gaudi2 = hdev->asic_specific;
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
int rc;
habanalabs: add support for new cpucp return codes Firmware now responds with a more detailed cpucp return codes. Driver can now distinguish between error and debug return codes. Signed-off-by: Ofir Bitton <obitton@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-08-23 21:58:38 +08:00
hdev->asic_prop.supports_advanced_cpucp_rc = true;
habanalabs/gaudi2: configure virtual MSI-X doorbell interface Due to a watchdog timer in the LBW path, writes to the MSI-X doorbell can return sporadic error responses. To work-around this issue, a virtual MSI-X doorbell on the HBW path is configured, using the MSI-X AXI slave interface in the PCIe controller. Upon an access to a configured HBW host address, the controller will generate MSI-X interrupt instead of treating the access as regular host memory access. This patch allocates the dedicate host memory page, and communicate the address to F/W, so it will configure the relevant address match registers in the controller, and will use this address to generate MSI-X interrupts for F/W events. Following patches will handle other initiators in the device, to move them to use the virtual MSI-X doorbell. Signed-off-by: Tomer Tayar <ttayar@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-29 21:37:55 +08:00
rc = hl_fw_send_pci_access_msg(hdev, CPUCP_PACKET_ENABLE_PCI_ACCESS,
gaudi2->virt_msix_db_dma_addr);
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
if (rc) {
dev_err(hdev->dev, "Failed to enable PCI access from CPU\n");
return rc;
}
rc = gaudi2_fetch_psoc_frequency(hdev);
if (rc) {
dev_err(hdev->dev, "Failed to fetch psoc frequency\n");
goto disable_pci_access;
}
gaudi2_init_arcs(hdev);
gaudi2_scrub_arcs_dccm(hdev);
habanalabs/gaudi2: add gaudi2 security module Use the generic security module to block all registers in the ASIC and then open only those that are needed to be accessed by the user. Signed-off-by: Ofir Bitton <obitton@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-27 02:30:25 +08:00
gaudi2_init_security(hdev);
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
return 0;
disable_pci_access:
habanalabs: add a value field to hl_fw_send_pci_access_msg() For gaudi2 we need to send a value to F/W as part of the PCI_ACCESS packet. As a preparation, modify hl_fw_send_pci_access_msg() to have a 'value' field. Signed-off-by: Tomer Tayar <ttayar@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-07-06 15:20:54 +08:00
hl_fw_send_pci_access_msg(hdev, CPUCP_PACKET_DISABLE_PCI_ACCESS, 0x0);
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
return rc;
}
static void gaudi2_late_fini(struct hl_device *hdev)
{
habanalabs: unify hwmon resources clean up Since hwmon fini code is common for all asics, unified it to common function. Signed-off-by: Dani Liberman <dliberman@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-28 03:06:51 +08:00
hl_hwmon_release_resources(hdev);
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
}
static void gaudi2_user_mapped_dec_init(struct gaudi2_device *gaudi2, u32 start_idx)
{
struct user_mapped_block *blocks = gaudi2->mapped_blocks;
HL_USR_MAPPED_BLK_INIT(&blocks[start_idx++], mmDCORE0_DEC0_CMD_BASE, HL_BLOCK_SIZE);
HL_USR_MAPPED_BLK_INIT(&blocks[start_idx++], mmDCORE0_DEC1_CMD_BASE, HL_BLOCK_SIZE);
HL_USR_MAPPED_BLK_INIT(&blocks[start_idx++], mmDCORE1_DEC0_CMD_BASE, HL_BLOCK_SIZE);
HL_USR_MAPPED_BLK_INIT(&blocks[start_idx++], mmDCORE1_DEC1_CMD_BASE, HL_BLOCK_SIZE);
HL_USR_MAPPED_BLK_INIT(&blocks[start_idx++], mmDCORE2_DEC0_CMD_BASE, HL_BLOCK_SIZE);
HL_USR_MAPPED_BLK_INIT(&blocks[start_idx++], mmDCORE2_DEC1_CMD_BASE, HL_BLOCK_SIZE);
HL_USR_MAPPED_BLK_INIT(&blocks[start_idx++], mmDCORE3_DEC0_CMD_BASE, HL_BLOCK_SIZE);
HL_USR_MAPPED_BLK_INIT(&blocks[start_idx++], mmDCORE3_DEC1_CMD_BASE, HL_BLOCK_SIZE);
HL_USR_MAPPED_BLK_INIT(&blocks[start_idx++], mmPCIE_DEC0_CMD_BASE, HL_BLOCK_SIZE);
HL_USR_MAPPED_BLK_INIT(&blocks[start_idx], mmPCIE_DEC1_CMD_BASE, HL_BLOCK_SIZE);
}
static void gaudi2_user_mapped_blocks_init(struct hl_device *hdev)
{
struct gaudi2_device *gaudi2 = hdev->asic_specific;
struct user_mapped_block *blocks = gaudi2->mapped_blocks;
u32 block_size, umr_start_idx, num_umr_blocks;
int i;
for (i = 0 ; i < NUM_ARC_CPUS ; i++) {
if (i >= CPU_ID_SCHED_ARC0 && i <= CPU_ID_SCHED_ARC3)
block_size = ARC_DCCM_BLOCK_SIZE * 2;
else
block_size = ARC_DCCM_BLOCK_SIZE;
blocks[i].address = gaudi2_arc_dccm_bases[i];
blocks[i].size = block_size;
}
blocks[NUM_ARC_CPUS].address = mmARC_FARM_ARC0_ACP_ENG_BASE;
blocks[NUM_ARC_CPUS].size = HL_BLOCK_SIZE;
blocks[NUM_ARC_CPUS + 1].address = mmARC_FARM_ARC1_ACP_ENG_BASE;
blocks[NUM_ARC_CPUS + 1].size = HL_BLOCK_SIZE;
blocks[NUM_ARC_CPUS + 2].address = mmARC_FARM_ARC2_ACP_ENG_BASE;
blocks[NUM_ARC_CPUS + 2].size = HL_BLOCK_SIZE;
blocks[NUM_ARC_CPUS + 3].address = mmARC_FARM_ARC3_ACP_ENG_BASE;
blocks[NUM_ARC_CPUS + 3].size = HL_BLOCK_SIZE;
blocks[NUM_ARC_CPUS + 4].address = mmDCORE0_MME_QM_ARC_ACP_ENG_BASE;
blocks[NUM_ARC_CPUS + 4].size = HL_BLOCK_SIZE;
blocks[NUM_ARC_CPUS + 5].address = mmDCORE1_MME_QM_ARC_ACP_ENG_BASE;
blocks[NUM_ARC_CPUS + 5].size = HL_BLOCK_SIZE;
blocks[NUM_ARC_CPUS + 6].address = mmDCORE2_MME_QM_ARC_ACP_ENG_BASE;
blocks[NUM_ARC_CPUS + 6].size = HL_BLOCK_SIZE;
blocks[NUM_ARC_CPUS + 7].address = mmDCORE3_MME_QM_ARC_ACP_ENG_BASE;
blocks[NUM_ARC_CPUS + 7].size = HL_BLOCK_SIZE;
umr_start_idx = NUM_ARC_CPUS + NUM_OF_USER_ACP_BLOCKS;
num_umr_blocks = NIC_NUMBER_OF_ENGINES * NUM_OF_USER_NIC_UMR_BLOCKS;
for (i = 0 ; i < num_umr_blocks ; i++) {
u8 nic_id, umr_block_id;
nic_id = i / NUM_OF_USER_NIC_UMR_BLOCKS;
umr_block_id = i % NUM_OF_USER_NIC_UMR_BLOCKS;
blocks[umr_start_idx + i].address =
mmNIC0_UMR0_0_UNSECURE_DOORBELL0_BASE +
(nic_id / NIC_NUMBER_OF_QM_PER_MACRO) * NIC_OFFSET +
(nic_id % NIC_NUMBER_OF_QM_PER_MACRO) * NIC_QM_OFFSET +
umr_block_id * NIC_UMR_OFFSET;
blocks[umr_start_idx + i].size = HL_BLOCK_SIZE;
}
/* Expose decoder HW configuration block to user */
gaudi2_user_mapped_dec_init(gaudi2, USR_MAPPED_BLK_DEC_START_IDX);
for (i = 1; i < NUM_OF_DCORES; ++i) {
blocks[USR_MAPPED_BLK_SM_START_IDX + 2 * (i - 1)].size = SM_OBJS_BLOCK_SIZE;
blocks[USR_MAPPED_BLK_SM_START_IDX + 2 * (i - 1) + 1].size = HL_BLOCK_SIZE;
blocks[USR_MAPPED_BLK_SM_START_IDX + 2 * (i - 1)].address =
mmDCORE0_SYNC_MNGR_OBJS_BASE + i * DCORE_OFFSET;
blocks[USR_MAPPED_BLK_SM_START_IDX + 2 * (i - 1) + 1].address =
mmDCORE0_SYNC_MNGR_GLBL_BASE + i * DCORE_OFFSET;
}
}
static int gaudi2_alloc_cpu_accessible_dma_mem(struct hl_device *hdev)
{
dma_addr_t dma_addr_arr[GAUDI2_ALLOC_CPU_MEM_RETRY_CNT] = {}, end_addr;
void *virt_addr_arr[GAUDI2_ALLOC_CPU_MEM_RETRY_CNT] = {};
int i, j, rc = 0;
/* The device ARC works with 32-bits addresses, and because there is a single HW register
* that holds the extension bits (49..28), these bits must be identical in all the allocated
* range.
*/
for (i = 0 ; i < GAUDI2_ALLOC_CPU_MEM_RETRY_CNT ; i++) {
virt_addr_arr[i] = hl_asic_dma_alloc_coherent(hdev, HL_CPU_ACCESSIBLE_MEM_SIZE,
&dma_addr_arr[i], GFP_KERNEL | __GFP_ZERO);
if (!virt_addr_arr[i]) {
rc = -ENOMEM;
goto free_dma_mem_arr;
}
end_addr = dma_addr_arr[i] + HL_CPU_ACCESSIBLE_MEM_SIZE - 1;
if (GAUDI2_ARC_PCI_MSB_ADDR(dma_addr_arr[i]) == GAUDI2_ARC_PCI_MSB_ADDR(end_addr))
break;
}
if (i == GAUDI2_ALLOC_CPU_MEM_RETRY_CNT) {
dev_err(hdev->dev,
"MSB of ARC accessible DMA memory are not identical in all range\n");
rc = -EFAULT;
goto free_dma_mem_arr;
}
hdev->cpu_accessible_dma_mem = virt_addr_arr[i];
hdev->cpu_accessible_dma_address = dma_addr_arr[i];
free_dma_mem_arr:
for (j = 0 ; j < i ; j++)
hl_asic_dma_free_coherent(hdev, HL_CPU_ACCESSIBLE_MEM_SIZE, virt_addr_arr[j],
dma_addr_arr[j]);
return rc;
}
static void gaudi2_set_pci_memory_regions(struct hl_device *hdev)
{
struct asic_fixed_properties *prop = &hdev->asic_prop;
struct pci_mem_region *region;
/* CFG */
region = &hdev->pci_mem_region[PCI_REGION_CFG];
region->region_base = CFG_BASE;
region->region_size = CFG_SIZE;
region->offset_in_bar = CFG_BASE - STM_FLASH_BASE_ADDR;
region->bar_size = CFG_BAR_SIZE;
region->bar_id = SRAM_CFG_BAR_ID;
region->used = 1;
/* SRAM */
region = &hdev->pci_mem_region[PCI_REGION_SRAM];
region->region_base = SRAM_BASE_ADDR;
region->region_size = SRAM_SIZE;
region->offset_in_bar = CFG_REGION_SIZE + BAR0_RSRVD_SIZE;
region->bar_size = CFG_BAR_SIZE;
region->bar_id = SRAM_CFG_BAR_ID;
region->used = 1;
/* DRAM */
region = &hdev->pci_mem_region[PCI_REGION_DRAM];
region->region_base = DRAM_PHYS_BASE;
region->region_size = hdev->asic_prop.dram_size;
region->offset_in_bar = 0;
region->bar_size = prop->dram_pci_bar_size;
region->bar_id = DRAM_BAR_ID;
region->used = 1;
}
static void gaudi2_user_interrupt_setup(struct hl_device *hdev)
{
struct asic_fixed_properties *prop = &hdev->asic_prop;
int i, j, k;
habanalabs: naming refactor of user interrupt flow Current naming convention can be misleading. Hence renaming some variables and defines in order to be more explicit. Signed-off-by: Ofir Bitton <obitton@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-29 02:05:28 +08:00
/* Initialize common user CQ interrupt */
HL_USR_INTR_STRUCT_INIT(hdev->common_user_cq_interrupt, hdev,
HL_COMMON_USER_CQ_INTERRUPT_ID, false);
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
habanalabs: add support for common decoder interrupts User application should be able to get notification for any decoder completion. Hence, we introduce a new interface in which a user can wait for all current decoder pending interrupts. Signed-off-by: Ofir Bitton <obitton@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-28 23:34:58 +08:00
/* Initialize common decoder interrupt */
HL_USR_INTR_STRUCT_INIT(hdev->common_decoder_interrupt, hdev,
HL_COMMON_DEC_INTERRUPT_ID, true);
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
/* User interrupts structure holds both decoder and user interrupts from various engines.
* We first initialize the decoder interrupts and then we add the user interrupts.
* The only limitation is that the last decoder interrupt id must be smaller
* then GAUDI2_IRQ_NUM_USER_FIRST. This is checked at compilation time.
*/
/* Initialize decoder interrupts, expose only normal interrupts,
* error interrupts to be handled by driver
*/
for (i = GAUDI2_IRQ_NUM_DCORE0_DEC0_NRM, j = 0 ; i <= GAUDI2_IRQ_NUM_SHARED_DEC1_NRM;
i += 2, j++)
HL_USR_INTR_STRUCT_INIT(hdev->user_interrupt[j], hdev, i, true);
for (i = GAUDI2_IRQ_NUM_USER_FIRST, k = 0 ; k < prop->user_interrupt_count; i++, j++, k++)
HL_USR_INTR_STRUCT_INIT(hdev->user_interrupt[j], hdev, i, false);
}
static inline int gaudi2_get_non_zero_random_int(void)
{
int rand = get_random_int();
return rand ? rand : 1;
}
static int gaudi2_sw_init(struct hl_device *hdev)
{
habanalabs/gaudi2: configure virtual MSI-X doorbell interface Due to a watchdog timer in the LBW path, writes to the MSI-X doorbell can return sporadic error responses. To work-around this issue, a virtual MSI-X doorbell on the HBW path is configured, using the MSI-X AXI slave interface in the PCIe controller. Upon an access to a configured HBW host address, the controller will generate MSI-X interrupt instead of treating the access as regular host memory access. This patch allocates the dedicate host memory page, and communicate the address to F/W, so it will configure the relevant address match registers in the controller, and will use this address to generate MSI-X interrupts for F/W events. Following patches will handle other initiators in the device, to move them to use the virtual MSI-X doorbell. Signed-off-by: Tomer Tayar <ttayar@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-29 21:37:55 +08:00
struct asic_fixed_properties *prop = &hdev->asic_prop;
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
struct gaudi2_device *gaudi2;
int i, rc;
/* Allocate device structure */
gaudi2 = kzalloc(sizeof(*gaudi2), GFP_KERNEL);
if (!gaudi2)
return -ENOMEM;
for (i = 0 ; i < ARRAY_SIZE(gaudi2_irq_map_table) ; i++) {
if (gaudi2_irq_map_table[i].msg || !gaudi2_irq_map_table[i].valid)
continue;
if (gaudi2->num_of_valid_hw_events == GAUDI2_EVENT_SIZE) {
dev_err(hdev->dev, "H/W events array exceeds the limit of %u events\n",
GAUDI2_EVENT_SIZE);
rc = -EINVAL;
goto free_gaudi2_device;
}
gaudi2->hw_events[gaudi2->num_of_valid_hw_events++] = gaudi2_irq_map_table[i].fc_id;
}
for (i = 0 ; i < MME_NUM_OF_LFSR_SEEDS ; i++)
gaudi2->lfsr_rand_seeds[i] = gaudi2_get_non_zero_random_int();
gaudi2->cpucp_info_get = gaudi2_cpucp_info_get;
hdev->asic_specific = gaudi2;
/* Create DMA pool for small allocations.
* Use DEVICE_CACHE_LINE_SIZE for alignment since the NIC memory-mapped
* PI/CI registers allocated from this pool have this restriction
*/
hdev->dma_pool = dma_pool_create(dev_name(hdev->dev), &hdev->pdev->dev,
GAUDI2_DMA_POOL_BLK_SIZE, DEVICE_CACHE_LINE_SIZE, 0);
if (!hdev->dma_pool) {
dev_err(hdev->dev, "failed to create DMA pool\n");
rc = -ENOMEM;
goto free_gaudi2_device;
}
rc = gaudi2_alloc_cpu_accessible_dma_mem(hdev);
if (rc)
goto free_dma_pool;
hdev->cpu_accessible_dma_pool = gen_pool_create(ilog2(32), -1);
if (!hdev->cpu_accessible_dma_pool) {
dev_err(hdev->dev, "Failed to create CPU accessible DMA pool\n");
rc = -ENOMEM;
goto free_cpu_dma_mem;
}
rc = gen_pool_add(hdev->cpu_accessible_dma_pool, (uintptr_t) hdev->cpu_accessible_dma_mem,
HL_CPU_ACCESSIBLE_MEM_SIZE, -1);
if (rc) {
dev_err(hdev->dev, "Failed to add memory to CPU accessible DMA pool\n");
rc = -EFAULT;
goto free_cpu_accessible_dma_pool;
}
habanalabs/gaudi2: configure virtual MSI-X doorbell interface Due to a watchdog timer in the LBW path, writes to the MSI-X doorbell can return sporadic error responses. To work-around this issue, a virtual MSI-X doorbell on the HBW path is configured, using the MSI-X AXI slave interface in the PCIe controller. Upon an access to a configured HBW host address, the controller will generate MSI-X interrupt instead of treating the access as regular host memory access. This patch allocates the dedicate host memory page, and communicate the address to F/W, so it will configure the relevant address match registers in the controller, and will use this address to generate MSI-X interrupts for F/W events. Following patches will handle other initiators in the device, to move them to use the virtual MSI-X doorbell. Signed-off-by: Tomer Tayar <ttayar@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-29 21:37:55 +08:00
gaudi2->virt_msix_db_cpu_addr = hl_cpu_accessible_dma_pool_alloc(hdev, prop->pmmu.page_size,
&gaudi2->virt_msix_db_dma_addr);
if (!gaudi2->virt_msix_db_cpu_addr) {
dev_err(hdev->dev, "Failed to allocate DMA memory for virtual MSI-X doorbell\n");
rc = -ENOMEM;
goto free_cpu_accessible_dma_pool;
}
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
spin_lock_init(&gaudi2->hw_queues_lock);
gaudi2->scratchpad_kernel_address = hl_asic_dma_alloc_coherent(hdev, PAGE_SIZE,
&gaudi2->scratchpad_bus_address,
GFP_KERNEL | __GFP_ZERO);
if (!gaudi2->scratchpad_kernel_address) {
rc = -ENOMEM;
habanalabs/gaudi2: configure virtual MSI-X doorbell interface Due to a watchdog timer in the LBW path, writes to the MSI-X doorbell can return sporadic error responses. To work-around this issue, a virtual MSI-X doorbell on the HBW path is configured, using the MSI-X AXI slave interface in the PCIe controller. Upon an access to a configured HBW host address, the controller will generate MSI-X interrupt instead of treating the access as regular host memory access. This patch allocates the dedicate host memory page, and communicate the address to F/W, so it will configure the relevant address match registers in the controller, and will use this address to generate MSI-X interrupts for F/W events. Following patches will handle other initiators in the device, to move them to use the virtual MSI-X doorbell. Signed-off-by: Tomer Tayar <ttayar@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-29 21:37:55 +08:00
goto free_virt_msix_db_mem;
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
}
gaudi2_user_mapped_blocks_init(hdev);
/* Initialize user interrupts */
gaudi2_user_interrupt_setup(hdev);
hdev->supports_coresight = true;
hdev->supports_sync_stream = true;
hdev->supports_cb_mapping = true;
hdev->supports_wait_for_multi_cs = false;
habanalabs: rename soft reset to compute reset Doing compute reset can be the traditional inference soft reset that is supported only in Goya. Or it can be the new reset upon device release, which is supported in Gaudi2 and above. Therefore, wherever suitable, use the terminology of compute reset instead of soft reset. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-07-07 17:00:24 +08:00
prop->supports_compute_reset = true;
habanalabs/gaudi2: configure virtual MSI-X doorbell interface Due to a watchdog timer in the LBW path, writes to the MSI-X doorbell can return sporadic error responses. To work-around this issue, a virtual MSI-X doorbell on the HBW path is configured, using the MSI-X AXI slave interface in the PCIe controller. Upon an access to a configured HBW host address, the controller will generate MSI-X interrupt instead of treating the access as regular host memory access. This patch allocates the dedicate host memory page, and communicate the address to F/W, so it will configure the relevant address match registers in the controller, and will use this address to generate MSI-X interrupts for F/W events. Following patches will handle other initiators in the device, to move them to use the virtual MSI-X doorbell. Signed-off-by: Tomer Tayar <ttayar@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-29 21:37:55 +08:00
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
hdev->asic_funcs->set_pci_memory_regions(hdev);
return 0;
habanalabs/gaudi2: configure virtual MSI-X doorbell interface Due to a watchdog timer in the LBW path, writes to the MSI-X doorbell can return sporadic error responses. To work-around this issue, a virtual MSI-X doorbell on the HBW path is configured, using the MSI-X AXI slave interface in the PCIe controller. Upon an access to a configured HBW host address, the controller will generate MSI-X interrupt instead of treating the access as regular host memory access. This patch allocates the dedicate host memory page, and communicate the address to F/W, so it will configure the relevant address match registers in the controller, and will use this address to generate MSI-X interrupts for F/W events. Following patches will handle other initiators in the device, to move them to use the virtual MSI-X doorbell. Signed-off-by: Tomer Tayar <ttayar@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-29 21:37:55 +08:00
free_virt_msix_db_mem:
hl_cpu_accessible_dma_pool_free(hdev, prop->pmmu.page_size, gaudi2->virt_msix_db_cpu_addr);
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
free_cpu_accessible_dma_pool:
gen_pool_destroy(hdev->cpu_accessible_dma_pool);
free_cpu_dma_mem:
hl_asic_dma_free_coherent(hdev, HL_CPU_ACCESSIBLE_MEM_SIZE, hdev->cpu_accessible_dma_mem,
hdev->cpu_accessible_dma_address);
free_dma_pool:
dma_pool_destroy(hdev->dma_pool);
free_gaudi2_device:
kfree(gaudi2);
return rc;
}
static int gaudi2_sw_fini(struct hl_device *hdev)
{
habanalabs/gaudi2: configure virtual MSI-X doorbell interface Due to a watchdog timer in the LBW path, writes to the MSI-X doorbell can return sporadic error responses. To work-around this issue, a virtual MSI-X doorbell on the HBW path is configured, using the MSI-X AXI slave interface in the PCIe controller. Upon an access to a configured HBW host address, the controller will generate MSI-X interrupt instead of treating the access as regular host memory access. This patch allocates the dedicate host memory page, and communicate the address to F/W, so it will configure the relevant address match registers in the controller, and will use this address to generate MSI-X interrupts for F/W events. Following patches will handle other initiators in the device, to move them to use the virtual MSI-X doorbell. Signed-off-by: Tomer Tayar <ttayar@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-29 21:37:55 +08:00
struct asic_fixed_properties *prop = &hdev->asic_prop;
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
struct gaudi2_device *gaudi2 = hdev->asic_specific;
habanalabs/gaudi2: configure virtual MSI-X doorbell interface Due to a watchdog timer in the LBW path, writes to the MSI-X doorbell can return sporadic error responses. To work-around this issue, a virtual MSI-X doorbell on the HBW path is configured, using the MSI-X AXI slave interface in the PCIe controller. Upon an access to a configured HBW host address, the controller will generate MSI-X interrupt instead of treating the access as regular host memory access. This patch allocates the dedicate host memory page, and communicate the address to F/W, so it will configure the relevant address match registers in the controller, and will use this address to generate MSI-X interrupts for F/W events. Following patches will handle other initiators in the device, to move them to use the virtual MSI-X doorbell. Signed-off-by: Tomer Tayar <ttayar@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-29 21:37:55 +08:00
hl_cpu_accessible_dma_pool_free(hdev, prop->pmmu.page_size, gaudi2->virt_msix_db_cpu_addr);
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
gen_pool_destroy(hdev->cpu_accessible_dma_pool);
hl_asic_dma_free_coherent(hdev, HL_CPU_ACCESSIBLE_MEM_SIZE, hdev->cpu_accessible_dma_mem,
hdev->cpu_accessible_dma_address);
hl_asic_dma_free_coherent(hdev, PAGE_SIZE, gaudi2->scratchpad_kernel_address,
gaudi2->scratchpad_bus_address);
dma_pool_destroy(hdev->dma_pool);
kfree(gaudi2);
return 0;
}
static void gaudi2_stop_qman_common(struct hl_device *hdev, u32 reg_base)
{
WREG32(reg_base + QM_GLBL_CFG1_OFFSET, QM_GLBL_CFG1_PQF_STOP |
QM_GLBL_CFG1_CQF_STOP |
QM_GLBL_CFG1_CP_STOP);
/* stop also the ARC */
WREG32(reg_base + QM_GLBL_CFG2_OFFSET, QM_GLBL_CFG2_ARC_CQF_STOP);
}
static void gaudi2_flush_qman_common(struct hl_device *hdev, u32 reg_base)
{
WREG32(reg_base + QM_GLBL_CFG1_OFFSET, QM_GLBL_CFG1_PQF_FLUSH |
QM_GLBL_CFG1_CQF_FLUSH |
QM_GLBL_CFG1_CP_FLUSH);
}
static void gaudi2_flush_qman_arc_common(struct hl_device *hdev, u32 reg_base)
{
WREG32(reg_base + QM_GLBL_CFG2_OFFSET, QM_GLBL_CFG2_ARC_CQF_FLUSH);
}
/**
* gaudi2_clear_qm_fence_counters_common - clear QM's fence counters
*
* @hdev: pointer to the habanalabs device structure
* @queue_id: queue to clear fence counters to
* @skip_fence: if true set maximum fence value to all fence counters to avoid
* getting stuck on any fence value. otherwise set all fence
* counters to 0 (standard clear of fence counters)
*/
static void gaudi2_clear_qm_fence_counters_common(struct hl_device *hdev, u32 queue_id,
bool skip_fence)
{
u32 size, reg_base;
u32 addr, val;
reg_base = gaudi2_qm_blocks_bases[queue_id];
addr = reg_base + QM_CP_FENCE0_CNT_0_OFFSET;
size = mmPDMA0_QM_CP_BARRIER_CFG - mmPDMA0_QM_CP_FENCE0_CNT_0;
/*
* in case we want to make sure that QM that is stuck on a fence will
* be released we should set the fence counter to a higher value that
* the value the QM waiting for. to comply with any fence counter of
* any value we set maximum fence value to all counters
*/
val = skip_fence ? U32_MAX : 0;
gaudi2_memset_device_lbw(hdev, addr, size, val);
}
static void gaudi2_qman_manual_flush_common(struct hl_device *hdev, u32 queue_id)
{
u32 reg_base = gaudi2_qm_blocks_bases[queue_id];
gaudi2_clear_qm_fence_counters_common(hdev, queue_id, true);
gaudi2_flush_qman_common(hdev, reg_base);
gaudi2_flush_qman_arc_common(hdev, reg_base);
}
static void gaudi2_stop_dma_qmans(struct hl_device *hdev)
{
struct gaudi2_device *gaudi2 = hdev->asic_specific;
int dcore, inst;
if (!(gaudi2->hw_cap_initialized & HW_CAP_PDMA_MASK))
goto stop_edma_qmans;
/* Stop CPs of PDMA QMANs */
gaudi2_stop_qman_common(hdev, mmPDMA0_QM_BASE);
gaudi2_stop_qman_common(hdev, mmPDMA1_QM_BASE);
stop_edma_qmans:
if (!(gaudi2->hw_cap_initialized & HW_CAP_EDMA_MASK))
return;
for (dcore = 0 ; dcore < NUM_OF_DCORES ; dcore++) {
for (inst = 0 ; inst < NUM_OF_EDMA_PER_DCORE ; inst++) {
u8 seq = dcore * NUM_OF_EDMA_PER_DCORE + inst;
u32 qm_base;
if (!(gaudi2->hw_cap_initialized & BIT_ULL(HW_CAP_EDMA_SHIFT + seq)))
continue;
qm_base = mmDCORE0_EDMA0_QM_BASE + dcore * DCORE_OFFSET +
inst * DCORE_EDMA_OFFSET;
/* Stop CPs of EDMA QMANs */
gaudi2_stop_qman_common(hdev, qm_base);
}
}
}
static void gaudi2_stop_mme_qmans(struct hl_device *hdev)
{
struct gaudi2_device *gaudi2 = hdev->asic_specific;
u32 offset, i;
offset = mmDCORE1_MME_QM_BASE - mmDCORE0_MME_QM_BASE;
for (i = 0 ; i < NUM_OF_DCORES ; i++) {
if (!(gaudi2->hw_cap_initialized & BIT_ULL(HW_CAP_MME_SHIFT + i)))
continue;
gaudi2_stop_qman_common(hdev, mmDCORE0_MME_QM_BASE + (i * offset));
}
}
static void gaudi2_stop_tpc_qmans(struct hl_device *hdev)
{
struct gaudi2_device *gaudi2 = hdev->asic_specific;
u32 reg_base;
int i;
if (!(gaudi2->tpc_hw_cap_initialized & HW_CAP_TPC_MASK))
return;
for (i = 0 ; i < TPC_ID_SIZE ; i++) {
if (!(gaudi2->tpc_hw_cap_initialized & BIT_ULL(HW_CAP_TPC_SHIFT + i)))
continue;
reg_base = gaudi2_qm_blocks_bases[gaudi2_tpc_id_to_queue_id[i]];
gaudi2_stop_qman_common(hdev, reg_base);
}
}
static void gaudi2_stop_rot_qmans(struct hl_device *hdev)
{
struct gaudi2_device *gaudi2 = hdev->asic_specific;
u32 reg_base;
int i;
if (!(gaudi2->hw_cap_initialized & HW_CAP_ROT_MASK))
return;
for (i = 0 ; i < ROTATOR_ID_SIZE ; i++) {
if (!(gaudi2->hw_cap_initialized & BIT_ULL(HW_CAP_ROT_SHIFT + i)))
continue;
reg_base = gaudi2_qm_blocks_bases[gaudi2_rot_id_to_queue_id[i]];
gaudi2_stop_qman_common(hdev, reg_base);
}
}
static void gaudi2_stop_nic_qmans(struct hl_device *hdev)
{
struct gaudi2_device *gaudi2 = hdev->asic_specific;
u32 reg_base, queue_id;
int i;
if (!(gaudi2->nic_hw_cap_initialized & HW_CAP_NIC_MASK))
return;
queue_id = GAUDI2_QUEUE_ID_NIC_0_0;
for (i = 0 ; i < NIC_NUMBER_OF_ENGINES ; i++, queue_id += NUM_OF_PQ_PER_QMAN) {
if (!(hdev->nic_ports_mask & BIT(i)))
continue;
reg_base = gaudi2_qm_blocks_bases[queue_id];
gaudi2_stop_qman_common(hdev, reg_base);
}
}
static void gaudi2_stall_dma_common(struct hl_device *hdev, u32 reg_base)
{
u32 reg_val;
reg_val = FIELD_PREP(PDMA0_CORE_CFG_1_HALT_MASK, 0x1);
WREG32(reg_base + DMA_CORE_CFG_1_OFFSET, reg_val);
}
static void gaudi2_dma_stall(struct hl_device *hdev)
{
struct gaudi2_device *gaudi2 = hdev->asic_specific;
int dcore, inst;
if (!(gaudi2->hw_cap_initialized & HW_CAP_PDMA_MASK))
goto stall_edma;
gaudi2_stall_dma_common(hdev, mmPDMA0_CORE_BASE);
gaudi2_stall_dma_common(hdev, mmPDMA1_CORE_BASE);
stall_edma:
if (!(gaudi2->hw_cap_initialized & HW_CAP_EDMA_MASK))
return;
for (dcore = 0 ; dcore < NUM_OF_DCORES ; dcore++) {
for (inst = 0 ; inst < NUM_OF_EDMA_PER_DCORE ; inst++) {
u8 seq = dcore * NUM_OF_EDMA_PER_DCORE + inst;
u32 core_base;
if (!(gaudi2->hw_cap_initialized & BIT_ULL(HW_CAP_EDMA_SHIFT + seq)))
continue;
core_base = mmDCORE0_EDMA0_CORE_BASE + dcore * DCORE_OFFSET +
inst * DCORE_EDMA_OFFSET;
/* Stall CPs of EDMA QMANs */
gaudi2_stall_dma_common(hdev, core_base);
}
}
}
static void gaudi2_mme_stall(struct hl_device *hdev)
{
struct gaudi2_device *gaudi2 = hdev->asic_specific;
u32 offset, i;
offset = mmDCORE1_MME_CTRL_LO_QM_STALL - mmDCORE0_MME_CTRL_LO_QM_STALL;
for (i = 0 ; i < NUM_OF_DCORES ; i++)
if (gaudi2->hw_cap_initialized & BIT_ULL(HW_CAP_MME_SHIFT + i))
WREG32(mmDCORE0_MME_CTRL_LO_QM_STALL + (i * offset), 1);
}
static void gaudi2_tpc_stall(struct hl_device *hdev)
{
struct gaudi2_device *gaudi2 = hdev->asic_specific;
u32 reg_base;
int i;
if (!(gaudi2->tpc_hw_cap_initialized & HW_CAP_TPC_MASK))
return;
for (i = 0 ; i < TPC_ID_SIZE ; i++) {
if (!(gaudi2->tpc_hw_cap_initialized & BIT_ULL(HW_CAP_TPC_SHIFT + i)))
continue;
reg_base = gaudi2_tpc_cfg_blocks_bases[i];
WREG32(reg_base + TPC_CFG_STALL_OFFSET, 1);
}
}
static void gaudi2_rotator_stall(struct hl_device *hdev)
{
struct gaudi2_device *gaudi2 = hdev->asic_specific;
u32 reg_val;
int i;
if (!(gaudi2->hw_cap_initialized & HW_CAP_ROT_MASK))
return;
reg_val = FIELD_PREP(ROT_MSS_HALT_WBC_MASK, 0x1) |
FIELD_PREP(ROT_MSS_HALT_RSB_MASK, 0x1) |
FIELD_PREP(ROT_MSS_HALT_MRSB_MASK, 0x1);
for (i = 0 ; i < ROTATOR_ID_SIZE ; i++) {
if (!(gaudi2->hw_cap_initialized & BIT_ULL(HW_CAP_ROT_SHIFT + i)))
continue;
WREG32(mmROT0_MSS_HALT + i * ROT_OFFSET, reg_val);
}
}
static void gaudi2_disable_qman_common(struct hl_device *hdev, u32 reg_base)
{
WREG32(reg_base + QM_GLBL_CFG0_OFFSET, 0);
}
static void gaudi2_disable_dma_qmans(struct hl_device *hdev)
{
struct gaudi2_device *gaudi2 = hdev->asic_specific;
int dcore, inst;
if (!(gaudi2->hw_cap_initialized & HW_CAP_PDMA_MASK))
goto stop_edma_qmans;
gaudi2_disable_qman_common(hdev, mmPDMA0_QM_BASE);
gaudi2_disable_qman_common(hdev, mmPDMA1_QM_BASE);
stop_edma_qmans:
if (!(gaudi2->hw_cap_initialized & HW_CAP_EDMA_MASK))
return;
for (dcore = 0 ; dcore < NUM_OF_DCORES ; dcore++) {
for (inst = 0 ; inst < NUM_OF_EDMA_PER_DCORE ; inst++) {
u8 seq = dcore * NUM_OF_EDMA_PER_DCORE + inst;
u32 qm_base;
if (!(gaudi2->hw_cap_initialized & BIT_ULL(HW_CAP_EDMA_SHIFT + seq)))
continue;
qm_base = mmDCORE0_EDMA0_QM_BASE + dcore * DCORE_OFFSET +
inst * DCORE_EDMA_OFFSET;
/* Disable CPs of EDMA QMANs */
gaudi2_disable_qman_common(hdev, qm_base);
}
}
}
static void gaudi2_disable_mme_qmans(struct hl_device *hdev)
{
struct gaudi2_device *gaudi2 = hdev->asic_specific;
u32 offset, i;
offset = mmDCORE1_MME_QM_BASE - mmDCORE0_MME_QM_BASE;
for (i = 0 ; i < NUM_OF_DCORES ; i++)
if (gaudi2->hw_cap_initialized & BIT_ULL(HW_CAP_MME_SHIFT + i))
gaudi2_disable_qman_common(hdev, mmDCORE0_MME_QM_BASE + (i * offset));
}
static void gaudi2_disable_tpc_qmans(struct hl_device *hdev)
{
struct gaudi2_device *gaudi2 = hdev->asic_specific;
u32 reg_base;
int i;
if (!(gaudi2->tpc_hw_cap_initialized & HW_CAP_TPC_MASK))
return;
for (i = 0 ; i < TPC_ID_SIZE ; i++) {
if (!(gaudi2->tpc_hw_cap_initialized & BIT_ULL(HW_CAP_TPC_SHIFT + i)))
continue;
reg_base = gaudi2_qm_blocks_bases[gaudi2_tpc_id_to_queue_id[i]];
gaudi2_disable_qman_common(hdev, reg_base);
}
}
static void gaudi2_disable_rot_qmans(struct hl_device *hdev)
{
struct gaudi2_device *gaudi2 = hdev->asic_specific;
u32 reg_base;
int i;
if (!(gaudi2->hw_cap_initialized & HW_CAP_ROT_MASK))
return;
for (i = 0 ; i < ROTATOR_ID_SIZE ; i++) {
if (!(gaudi2->hw_cap_initialized & BIT_ULL(HW_CAP_ROT_SHIFT + i)))
continue;
reg_base = gaudi2_qm_blocks_bases[gaudi2_rot_id_to_queue_id[i]];
gaudi2_disable_qman_common(hdev, reg_base);
}
}
static void gaudi2_disable_nic_qmans(struct hl_device *hdev)
{
struct gaudi2_device *gaudi2 = hdev->asic_specific;
u32 reg_base, queue_id;
int i;
if (!(gaudi2->nic_hw_cap_initialized & HW_CAP_NIC_MASK))
return;
queue_id = GAUDI2_QUEUE_ID_NIC_0_0;
for (i = 0 ; i < NIC_NUMBER_OF_ENGINES ; i++, queue_id += NUM_OF_PQ_PER_QMAN) {
if (!(hdev->nic_ports_mask & BIT(i)))
continue;
reg_base = gaudi2_qm_blocks_bases[queue_id];
gaudi2_disable_qman_common(hdev, reg_base);
}
}
static void gaudi2_enable_timestamp(struct hl_device *hdev)
{
/* Disable the timestamp counter */
WREG32(mmPSOC_TIMESTAMP_BASE, 0);
/* Zero the lower/upper parts of the 64-bit counter */
WREG32(mmPSOC_TIMESTAMP_BASE + 0xC, 0);
WREG32(mmPSOC_TIMESTAMP_BASE + 0x8, 0);
/* Enable the counter */
WREG32(mmPSOC_TIMESTAMP_BASE, 1);
}
static void gaudi2_disable_timestamp(struct hl_device *hdev)
{
/* Disable the timestamp counter */
WREG32(mmPSOC_TIMESTAMP_BASE, 0);
}
static const char *gaudi2_irq_name(u16 irq_number)
{
switch (irq_number) {
case GAUDI2_IRQ_NUM_EVENT_QUEUE:
return "gaudi2 cpu eq";
case GAUDI2_IRQ_NUM_COMPLETION:
return "gaudi2 completion";
case GAUDI2_IRQ_NUM_DCORE0_DEC0_NRM ... GAUDI2_IRQ_NUM_SHARED_DEC1_ABNRM:
return gaudi2_vdec_irq_name[irq_number - GAUDI2_IRQ_NUM_DCORE0_DEC0_NRM];
case GAUDI2_IRQ_NUM_USER_FIRST ... GAUDI2_IRQ_NUM_USER_LAST:
return "gaudi2 user completion";
default:
return "invalid";
}
}
static void gaudi2_dec_disable_msix(struct hl_device *hdev, u32 max_irq_num)
{
int i, irq, relative_idx;
struct hl_dec *dec;
for (i = GAUDI2_IRQ_NUM_DCORE0_DEC0_NRM ; i < max_irq_num ; i++) {
irq = pci_irq_vector(hdev->pdev, i);
relative_idx = i - GAUDI2_IRQ_NUM_DCORE0_DEC0_NRM;
dec = hdev->dec + relative_idx / 2;
/* We pass different structures depending on the irq handler. For the abnormal
* interrupt we pass hl_dec and for the regular interrupt we pass the relevant
* user_interrupt entry
*/
free_irq(irq, ((relative_idx % 2) ?
(void *) dec :
(void *) &hdev->user_interrupt[dec->core_id]));
}
}
static int gaudi2_dec_enable_msix(struct hl_device *hdev)
{
int rc, i, irq_init_cnt, irq, relative_idx;
irq_handler_t irq_handler;
struct hl_dec *dec;
for (i = GAUDI2_IRQ_NUM_DCORE0_DEC0_NRM, irq_init_cnt = 0;
i <= GAUDI2_IRQ_NUM_SHARED_DEC1_ABNRM;
i++, irq_init_cnt++) {
irq = pci_irq_vector(hdev->pdev, i);
relative_idx = i - GAUDI2_IRQ_NUM_DCORE0_DEC0_NRM;
irq_handler = (relative_idx % 2) ?
hl_irq_handler_dec_abnrm :
hl_irq_handler_user_interrupt;
dec = hdev->dec + relative_idx / 2;
/* We pass different structures depending on the irq handler. For the abnormal
* interrupt we pass hl_dec and for the regular interrupt we pass the relevant
* user_interrupt entry
*/
rc = request_irq(irq, irq_handler, 0, gaudi2_irq_name(i),
((relative_idx % 2) ?
(void *) dec :
(void *) &hdev->user_interrupt[dec->core_id]));
if (rc) {
dev_err(hdev->dev, "Failed to request IRQ %d", irq);
goto free_dec_irqs;
}
}
return 0;
free_dec_irqs:
gaudi2_dec_disable_msix(hdev, (GAUDI2_IRQ_NUM_DCORE0_DEC0_NRM + irq_init_cnt));
return rc;
}
static int gaudi2_enable_msix(struct hl_device *hdev)
{
struct asic_fixed_properties *prop = &hdev->asic_prop;
struct gaudi2_device *gaudi2 = hdev->asic_specific;
int rc, irq, i, j, user_irq_init_cnt;
irq_handler_t irq_handler;
struct hl_cq *cq;
if (gaudi2->hw_cap_initialized & HW_CAP_MSIX)
return 0;
rc = pci_alloc_irq_vectors(hdev->pdev, GAUDI2_MSIX_ENTRIES, GAUDI2_MSIX_ENTRIES,
PCI_IRQ_MSIX);
if (rc < 0) {
dev_err(hdev->dev, "MSI-X: Failed to enable support -- %d/%d\n",
GAUDI2_MSIX_ENTRIES, rc);
return rc;
}
irq = pci_irq_vector(hdev->pdev, GAUDI2_IRQ_NUM_COMPLETION);
habanalabs/gaudi2: replace defines for reserved sob/mob with enums Following patches are going to add more reserved sync objects and monitors. To make the counting of these reserved resources simpler, replace the existing RESERVED_* defines with enumerations. Signed-off-by: Tomer Tayar <ttayar@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-30 00:20:38 +08:00
cq = &hdev->completion_queue[GAUDI2_RESERVED_CQ_CS_COMPLETION];
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
rc = request_irq(irq, hl_irq_handler_cq, 0, gaudi2_irq_name(GAUDI2_IRQ_NUM_COMPLETION), cq);
if (rc) {
dev_err(hdev->dev, "Failed to request IRQ %d", irq);
goto free_irq_vectors;
}
irq = pci_irq_vector(hdev->pdev, GAUDI2_IRQ_NUM_EVENT_QUEUE);
rc = request_irq(irq, hl_irq_handler_eq, 0, gaudi2_irq_name(GAUDI2_IRQ_NUM_EVENT_QUEUE),
&hdev->event_queue);
if (rc) {
dev_err(hdev->dev, "Failed to request IRQ %d", irq);
goto free_completion_irq;
}
rc = gaudi2_dec_enable_msix(hdev);
if (rc) {
dev_err(hdev->dev, "Failed to enable decoder IRQ");
habanalabs/gaudi2: free event irq if init fails In case initialization fails after event irq was requested, we need to release that irq. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-09-04 15:39:27 +08:00
goto free_event_irq;
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
}
for (i = GAUDI2_IRQ_NUM_USER_FIRST, j = prop->user_dec_intr_count, user_irq_init_cnt = 0;
user_irq_init_cnt < prop->user_interrupt_count;
i++, j++, user_irq_init_cnt++) {
irq = pci_irq_vector(hdev->pdev, i);
irq_handler = hl_irq_handler_user_interrupt;
rc = request_irq(irq, irq_handler, 0, gaudi2_irq_name(i), &hdev->user_interrupt[j]);
if (rc) {
dev_err(hdev->dev, "Failed to request IRQ %d", irq);
goto free_user_irq;
}
}
gaudi2->hw_cap_initialized |= HW_CAP_MSIX;
return 0;
free_user_irq:
for (i = GAUDI2_IRQ_NUM_USER_FIRST, j = prop->user_dec_intr_count;
i < GAUDI2_IRQ_NUM_USER_FIRST + user_irq_init_cnt ; i++, j++) {
irq = pci_irq_vector(hdev->pdev, i);
free_irq(irq, &hdev->user_interrupt[j]);
}
gaudi2_dec_disable_msix(hdev, GAUDI2_IRQ_NUM_SHARED_DEC1_ABNRM + 1);
habanalabs/gaudi2: free event irq if init fails In case initialization fails after event irq was requested, we need to release that irq. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-09-04 15:39:27 +08:00
free_event_irq:
irq = pci_irq_vector(hdev->pdev, GAUDI2_IRQ_NUM_EVENT_QUEUE);
free_irq(irq, cq);
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
free_completion_irq:
irq = pci_irq_vector(hdev->pdev, GAUDI2_IRQ_NUM_COMPLETION);
free_irq(irq, cq);
free_irq_vectors:
pci_free_irq_vectors(hdev->pdev);
return rc;
}
static void gaudi2_sync_irqs(struct hl_device *hdev)
{
struct gaudi2_device *gaudi2 = hdev->asic_specific;
int i, j;
int irq;
if (!(gaudi2->hw_cap_initialized & HW_CAP_MSIX))
return;
/* Wait for all pending IRQs to be finished */
synchronize_irq(pci_irq_vector(hdev->pdev, GAUDI2_IRQ_NUM_COMPLETION));
for (i = GAUDI2_IRQ_NUM_DCORE0_DEC0_NRM ; i <= GAUDI2_IRQ_NUM_SHARED_DEC1_ABNRM ; i++) {
irq = pci_irq_vector(hdev->pdev, i);
synchronize_irq(irq);
}
for (i = GAUDI2_IRQ_NUM_USER_FIRST, j = 0 ; j < hdev->asic_prop.user_interrupt_count;
i++, j++) {
irq = pci_irq_vector(hdev->pdev, i);
synchronize_irq(irq);
}
synchronize_irq(pci_irq_vector(hdev->pdev, GAUDI2_IRQ_NUM_EVENT_QUEUE));
}
static void gaudi2_disable_msix(struct hl_device *hdev)
{
struct asic_fixed_properties *prop = &hdev->asic_prop;
struct gaudi2_device *gaudi2 = hdev->asic_specific;
struct hl_cq *cq;
int irq, i, j, k;
if (!(gaudi2->hw_cap_initialized & HW_CAP_MSIX))
return;
gaudi2_sync_irqs(hdev);
irq = pci_irq_vector(hdev->pdev, GAUDI2_IRQ_NUM_EVENT_QUEUE);
free_irq(irq, &hdev->event_queue);
gaudi2_dec_disable_msix(hdev, GAUDI2_IRQ_NUM_SHARED_DEC1_ABNRM + 1);
for (i = GAUDI2_IRQ_NUM_USER_FIRST, j = prop->user_dec_intr_count, k = 0;
k < hdev->asic_prop.user_interrupt_count ; i++, j++, k++) {
irq = pci_irq_vector(hdev->pdev, i);
free_irq(irq, &hdev->user_interrupt[j]);
}
irq = pci_irq_vector(hdev->pdev, GAUDI2_IRQ_NUM_COMPLETION);
habanalabs/gaudi2: replace defines for reserved sob/mob with enums Following patches are going to add more reserved sync objects and monitors. To make the counting of these reserved resources simpler, replace the existing RESERVED_* defines with enumerations. Signed-off-by: Tomer Tayar <ttayar@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-30 00:20:38 +08:00
cq = &hdev->completion_queue[GAUDI2_RESERVED_CQ_CS_COMPLETION];
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
free_irq(irq, cq);
pci_free_irq_vectors(hdev->pdev);
gaudi2->hw_cap_initialized &= ~HW_CAP_MSIX;
}
static void gaudi2_stop_dcore_dec(struct hl_device *hdev, int dcore_id)
{
u32 reg_val = FIELD_PREP(DCORE0_VDEC0_BRDG_CTRL_GRACEFUL_STOP_MASK, 0x1);
u32 graceful_pend_mask = DCORE0_VDEC0_BRDG_CTRL_GRACEFUL_PEND_MASK;
u32 timeout_usec, dec_id, dec_bit, offset, graceful;
int rc;
if (hdev->pldm)
timeout_usec = GAUDI2_PLDM_VDEC_TIMEOUT_USEC;
else
timeout_usec = GAUDI2_VDEC_TIMEOUT_USEC;
for (dec_id = 0 ; dec_id < NUM_OF_DEC_PER_DCORE ; dec_id++) {
dec_bit = dcore_id * NUM_OF_DEC_PER_DCORE + dec_id;
if (!(hdev->asic_prop.decoder_enabled_mask & BIT(dec_bit)))
continue;
offset = dcore_id * DCORE_OFFSET + dec_id * DCORE_VDEC_OFFSET;
WREG32(mmDCORE0_DEC0_CMD_SWREG16 + offset, 0);
WREG32(mmDCORE0_VDEC0_BRDG_CTRL_GRACEFUL + offset, reg_val);
/* Wait till all traffic from decoder stops
* before apply core reset.
*/
rc = hl_poll_timeout(
hdev,
mmDCORE0_VDEC0_BRDG_CTRL_GRACEFUL + offset,
graceful,
(graceful & graceful_pend_mask),
100,
timeout_usec);
if (rc)
dev_err(hdev->dev,
"Failed to stop traffic from DCORE%d Decoder %d\n",
dcore_id, dec_id);
}
}
static void gaudi2_stop_pcie_dec(struct hl_device *hdev)
{
u32 reg_val = FIELD_PREP(DCORE0_VDEC0_BRDG_CTRL_GRACEFUL_STOP_MASK, 0x1);
u32 graceful_pend_mask = PCIE_VDEC0_BRDG_CTRL_GRACEFUL_PEND_MASK;
u32 timeout_usec, dec_id, dec_bit, offset, graceful;
int rc;
if (hdev->pldm)
timeout_usec = GAUDI2_PLDM_VDEC_TIMEOUT_USEC;
else
timeout_usec = GAUDI2_VDEC_TIMEOUT_USEC;
for (dec_id = 0 ; dec_id < NUM_OF_DEC_PER_DCORE ; dec_id++) {
dec_bit = PCIE_DEC_SHIFT + dec_id;
if (!(hdev->asic_prop.decoder_enabled_mask & BIT(dec_bit)))
continue;
offset = dec_id * PCIE_VDEC_OFFSET;
WREG32(mmPCIE_DEC0_CMD_SWREG16 + offset, 0);
WREG32(mmPCIE_VDEC0_BRDG_CTRL_GRACEFUL + offset, reg_val);
/* Wait till all traffic from decoder stops
* before apply core reset.
*/
rc = hl_poll_timeout(
hdev,
mmPCIE_VDEC0_BRDG_CTRL_GRACEFUL + offset,
graceful,
(graceful & graceful_pend_mask),
100,
timeout_usec);
if (rc)
dev_err(hdev->dev,
"Failed to stop traffic from PCIe Decoder %d\n",
dec_id);
}
}
static void gaudi2_stop_dec(struct hl_device *hdev)
{
struct gaudi2_device *gaudi2 = hdev->asic_specific;
int dcore_id;
if ((gaudi2->dec_hw_cap_initialized & HW_CAP_DEC_MASK) == 0)
return;
for (dcore_id = 0 ; dcore_id < NUM_OF_DCORES ; dcore_id++)
gaudi2_stop_dcore_dec(hdev, dcore_id);
gaudi2_stop_pcie_dec(hdev);
}
habanalabs/gaudi2: new API to control engine cores running mode The current flow of halting the engine cores is implemented by command buffers built by the user space and sent towards the Driver. This current flow is broken since the user space does not know when the cores actually halt as sending a workload is async op. Therefore the application can not free the memory that is mapped to the engine cores. This new API allows the user space to control the running mode. The API call is sync (returns after the cores are set to the requested mode). Signed-off-by: Tal Cohen <talcohen@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-07-07 23:42:47 +08:00
static void gaudi2_set_arc_running_mode(struct hl_device *hdev, u32 cpu_id, u32 run_mode)
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
{
u32 reg_base, reg_val;
reg_base = gaudi2_arc_blocks_bases[cpu_id];
habanalabs/gaudi2: new API to control engine cores running mode The current flow of halting the engine cores is implemented by command buffers built by the user space and sent towards the Driver. This current flow is broken since the user space does not know when the cores actually halt as sending a workload is async op. Therefore the application can not free the memory that is mapped to the engine cores. This new API allows the user space to control the running mode. The API call is sync (returns after the cores are set to the requested mode). Signed-off-by: Tal Cohen <talcohen@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-07-07 23:42:47 +08:00
if (run_mode == HL_ENGINE_CORE_RUN)
reg_val = FIELD_PREP(ARC_FARM_ARC0_AUX_RUN_HALT_REQ_RUN_REQ_MASK, 1);
else
reg_val = FIELD_PREP(ARC_FARM_ARC0_AUX_RUN_HALT_REQ_HALT_REQ_MASK, 1);
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
WREG32(reg_base + ARC_HALT_REQ_OFFSET, reg_val);
}
static void gaudi2_halt_arcs(struct hl_device *hdev)
{
u16 arc_id;
for (arc_id = CPU_ID_SCHED_ARC0; arc_id < CPU_ID_MAX; arc_id++) {
if (gaudi2_is_arc_enabled(hdev, arc_id))
habanalabs/gaudi2: new API to control engine cores running mode The current flow of halting the engine cores is implemented by command buffers built by the user space and sent towards the Driver. This current flow is broken since the user space does not know when the cores actually halt as sending a workload is async op. Therefore the application can not free the memory that is mapped to the engine cores. This new API allows the user space to control the running mode. The API call is sync (returns after the cores are set to the requested mode). Signed-off-by: Tal Cohen <talcohen@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-07-07 23:42:47 +08:00
gaudi2_set_arc_running_mode(hdev, arc_id, HL_ENGINE_CORE_HALT);
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
}
}
habanalabs/gaudi2: new API to control engine cores running mode The current flow of halting the engine cores is implemented by command buffers built by the user space and sent towards the Driver. This current flow is broken since the user space does not know when the cores actually halt as sending a workload is async op. Therefore the application can not free the memory that is mapped to the engine cores. This new API allows the user space to control the running mode. The API call is sync (returns after the cores are set to the requested mode). Signed-off-by: Tal Cohen <talcohen@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-07-07 23:42:47 +08:00
static int gaudi2_verify_arc_running_mode(struct hl_device *hdev, u32 cpu_id, u32 run_mode)
{
int rc;
u32 reg_base, val, ack_mask, timeout_usec = 100000;
if (hdev->pldm)
timeout_usec *= 100;
reg_base = gaudi2_arc_blocks_bases[cpu_id];
if (run_mode == HL_ENGINE_CORE_RUN)
ack_mask = ARC_FARM_ARC0_AUX_RUN_HALT_ACK_RUN_ACK_MASK;
else
ack_mask = ARC_FARM_ARC0_AUX_RUN_HALT_ACK_HALT_ACK_MASK;
rc = hl_poll_timeout(hdev, reg_base + ARC_HALT_ACK_OFFSET,
val, ((val & ack_mask) == ack_mask),
1000, timeout_usec);
if (!rc) {
/* Clear */
val = FIELD_PREP(ARC_FARM_ARC0_AUX_RUN_HALT_REQ_RUN_REQ_MASK, 0);
WREG32(reg_base + ARC_HALT_REQ_OFFSET, val);
}
return rc;
}
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
static void gaudi2_reset_arcs(struct hl_device *hdev)
{
struct gaudi2_device *gaudi2 = hdev->asic_specific;
u16 arc_id;
if (!gaudi2)
return;
for (arc_id = CPU_ID_SCHED_ARC0; arc_id < CPU_ID_MAX; arc_id++)
if (gaudi2_is_arc_enabled(hdev, arc_id))
gaudi2_clr_arc_id_cap(hdev, arc_id);
}
static void gaudi2_nic_qmans_manual_flush(struct hl_device *hdev)
{
struct gaudi2_device *gaudi2 = hdev->asic_specific;
u32 queue_id;
int i;
if (!(gaudi2->nic_hw_cap_initialized & HW_CAP_NIC_MASK))
return;
queue_id = GAUDI2_QUEUE_ID_NIC_0_0;
habanalabs/gaudi2: new API to control engine cores running mode The current flow of halting the engine cores is implemented by command buffers built by the user space and sent towards the Driver. This current flow is broken since the user space does not know when the cores actually halt as sending a workload is async op. Therefore the application can not free the memory that is mapped to the engine cores. This new API allows the user space to control the running mode. The API call is sync (returns after the cores are set to the requested mode). Signed-off-by: Tal Cohen <talcohen@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-07-07 23:42:47 +08:00
for (i = 0 ; i < NIC_NUMBER_OF_ENGINES ; i++, queue_id += NUM_OF_PQ_PER_QMAN) {
if (!(hdev->nic_ports_mask & BIT(i)))
continue;
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
gaudi2_qman_manual_flush_common(hdev, queue_id);
habanalabs/gaudi2: new API to control engine cores running mode The current flow of halting the engine cores is implemented by command buffers built by the user space and sent towards the Driver. This current flow is broken since the user space does not know when the cores actually halt as sending a workload is async op. Therefore the application can not free the memory that is mapped to the engine cores. This new API allows the user space to control the running mode. The API call is sync (returns after the cores are set to the requested mode). Signed-off-by: Tal Cohen <talcohen@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-07-07 23:42:47 +08:00
}
}
static int gaudi2_set_engine_cores(struct hl_device *hdev, u32 *core_ids,
u32 num_cores, u32 core_command)
{
int i, rc;
for (i = 0 ; i < num_cores ; i++) {
if (gaudi2_is_arc_enabled(hdev, core_ids[i]))
gaudi2_set_arc_running_mode(hdev, core_ids[i], core_command);
}
for (i = 0 ; i < num_cores ; i++) {
if (gaudi2_is_arc_enabled(hdev, core_ids[i])) {
rc = gaudi2_verify_arc_running_mode(hdev, core_ids[i], core_command);
if (rc) {
dev_err(hdev->dev, "failed to %s arc: %d\n",
(core_command == HL_ENGINE_CORE_HALT) ?
"HALT" : "RUN", core_ids[i]);
return -1;
}
}
}
return 0;
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
}
static void gaudi2_halt_engines(struct hl_device *hdev, bool hard_reset, bool fw_reset)
{
u32 wait_timeout_ms;
if (hdev->pldm)
wait_timeout_ms = GAUDI2_PLDM_RESET_WAIT_MSEC;
else
wait_timeout_ms = GAUDI2_RESET_WAIT_MSEC;
if (fw_reset)
goto skip_engines;
gaudi2_stop_dma_qmans(hdev);
gaudi2_stop_mme_qmans(hdev);
gaudi2_stop_tpc_qmans(hdev);
gaudi2_stop_rot_qmans(hdev);
gaudi2_stop_nic_qmans(hdev);
msleep(wait_timeout_ms);
gaudi2_halt_arcs(hdev);
gaudi2_dma_stall(hdev);
gaudi2_mme_stall(hdev);
gaudi2_tpc_stall(hdev);
gaudi2_rotator_stall(hdev);
msleep(wait_timeout_ms);
gaudi2_stop_dec(hdev);
/*
* in case of soft reset do a manual flush for QMANs (currently called
* only for NIC QMANs
*/
if (!hard_reset)
gaudi2_nic_qmans_manual_flush(hdev);
gaudi2_disable_dma_qmans(hdev);
gaudi2_disable_mme_qmans(hdev);
gaudi2_disable_tpc_qmans(hdev);
gaudi2_disable_rot_qmans(hdev);
gaudi2_disable_nic_qmans(hdev);
gaudi2_disable_timestamp(hdev);
skip_engines:
if (hard_reset) {
gaudi2_disable_msix(hdev);
return;
}
gaudi2_sync_irqs(hdev);
}
habanalabs: wait for preboot ready after hard reset Currently we are not waiting for preboot ready after hard reset. This leads to a race in which COMMs protocol begins but will get no response from the f/w. Signed-off-by: Ohad Sharabi <osharabi@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-28 17:09:21 +08:00
static void gaudi2_init_firmware_preload_params(struct hl_device *hdev)
{
struct pre_fw_load_props *pre_fw_load = &hdev->fw_loader.pre_fw_load;
pre_fw_load->cpu_boot_status_reg = mmPSOC_GLOBAL_CONF_CPU_BOOT_STATUS;
pre_fw_load->sts_boot_dev_sts0_reg = mmCPU_BOOT_DEV_STS0;
pre_fw_load->sts_boot_dev_sts1_reg = mmCPU_BOOT_DEV_STS1;
pre_fw_load->boot_err0_reg = mmCPU_BOOT_ERR0;
pre_fw_load->boot_err1_reg = mmCPU_BOOT_ERR1;
pre_fw_load->wait_for_preboot_timeout = GAUDI2_PREBOOT_REQ_TIMEOUT_USEC;
}
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
static void gaudi2_init_firmware_loader(struct hl_device *hdev)
{
struct fw_load_mgr *fw_loader = &hdev->fw_loader;
struct dynamic_fw_load_mgr *dynamic_loader;
struct cpu_dyn_regs *dyn_regs;
/* fill common fields */
fw_loader->fw_comp_loaded = FW_TYPE_NONE;
fw_loader->boot_fit_img.image_name = GAUDI2_BOOT_FIT_FILE;
fw_loader->linux_img.image_name = GAUDI2_LINUX_FW_FILE;
fw_loader->boot_fit_timeout = GAUDI2_BOOT_FIT_REQ_TIMEOUT_USEC;
fw_loader->skip_bmc = false;
fw_loader->sram_bar_id = SRAM_CFG_BAR_ID;
fw_loader->dram_bar_id = DRAM_BAR_ID;
if (hdev->asic_type == ASIC_GAUDI2 || hdev->asic_type == ASIC_GAUDI2_SEC)
fw_loader->cpu_timeout = GAUDI2_CPU_TIMEOUT_USEC;
else /* ASIC_GAUDI2_FPGA */
fw_loader->cpu_timeout = GAUDI2_FPGA_CPU_TIMEOUT;
/* here we update initial values for few specific dynamic regs (as
* before reading the first descriptor from FW those value has to be
* hard-coded). in later stages of the protocol those values will be
* updated automatically by reading the FW descriptor so data there
* will always be up-to-date
*/
dynamic_loader = &hdev->fw_loader.dynamic_loader;
dyn_regs = &dynamic_loader->comm_desc.cpu_dyn_regs;
dyn_regs->kmd_msg_to_cpu = cpu_to_le32(mmPSOC_GLOBAL_CONF_KMD_MSG_TO_CPU);
dyn_regs->cpu_cmd_status_to_host = cpu_to_le32(mmCPU_CMD_STATUS_TO_HOST);
dynamic_loader->wait_for_bl_timeout = GAUDI2_WAIT_FOR_BL_TIMEOUT_USEC;
}
static int gaudi2_init_cpu(struct hl_device *hdev)
{
struct gaudi2_device *gaudi2 = hdev->asic_specific;
int rc;
if (!(hdev->fw_components & FW_TYPE_PREBOOT_CPU))
return 0;
if (gaudi2->hw_cap_initialized & HW_CAP_CPU)
return 0;
rc = hl_fw_init_cpu(hdev);
if (rc)
return rc;
gaudi2->hw_cap_initialized |= HW_CAP_CPU;
return 0;
}
static int gaudi2_init_cpu_queues(struct hl_device *hdev, u32 cpu_timeout)
{
struct hl_hw_queue *cpu_pq = &hdev->kernel_queues[GAUDI2_QUEUE_ID_CPU_PQ];
struct asic_fixed_properties *prop = &hdev->asic_prop;
struct gaudi2_device *gaudi2 = hdev->asic_specific;
struct cpu_dyn_regs *dyn_regs;
struct hl_eq *eq;
u32 status;
int err;
if (!hdev->cpu_queues_enable)
return 0;
if (gaudi2->hw_cap_initialized & HW_CAP_CPU_Q)
return 0;
eq = &hdev->event_queue;
dyn_regs = &hdev->fw_loader.dynamic_loader.comm_desc.cpu_dyn_regs;
WREG32(mmCPU_IF_PQ_BASE_ADDR_LOW, lower_32_bits(cpu_pq->bus_address));
WREG32(mmCPU_IF_PQ_BASE_ADDR_HIGH, upper_32_bits(cpu_pq->bus_address));
WREG32(mmCPU_IF_EQ_BASE_ADDR_LOW, lower_32_bits(eq->bus_address));
WREG32(mmCPU_IF_EQ_BASE_ADDR_HIGH, upper_32_bits(eq->bus_address));
WREG32(mmCPU_IF_CQ_BASE_ADDR_LOW, lower_32_bits(hdev->cpu_accessible_dma_address));
WREG32(mmCPU_IF_CQ_BASE_ADDR_HIGH, upper_32_bits(hdev->cpu_accessible_dma_address));
WREG32(mmCPU_IF_PQ_LENGTH, HL_QUEUE_SIZE_IN_BYTES);
WREG32(mmCPU_IF_EQ_LENGTH, HL_EQ_SIZE_IN_BYTES);
WREG32(mmCPU_IF_CQ_LENGTH, HL_CPU_ACCESSIBLE_MEM_SIZE);
/* Used for EQ CI */
WREG32(mmCPU_IF_EQ_RD_OFFS, 0);
WREG32(mmCPU_IF_PF_PQ_PI, 0);
WREG32(mmCPU_IF_QUEUE_INIT, PQ_INIT_STATUS_READY_FOR_CP);
/* Let the ARC know we are ready as it is now handling those queues */
WREG32(le32_to_cpu(dyn_regs->gic_host_pi_upd_irq),
gaudi2_irq_map_table[GAUDI2_EVENT_CPU_PI_UPDATE].cpu_id);
err = hl_poll_timeout(
hdev,
mmCPU_IF_QUEUE_INIT,
status,
(status == PQ_INIT_STATUS_READY_FOR_HOST),
1000,
cpu_timeout);
if (err) {
dev_err(hdev->dev, "Failed to communicate with device CPU (timeout)\n");
return -EIO;
}
/* update FW application security bits */
if (prop->fw_cpu_boot_dev_sts0_valid)
prop->fw_app_cpu_boot_dev_sts0 = RREG32(mmCPU_BOOT_DEV_STS0);
if (prop->fw_cpu_boot_dev_sts1_valid)
prop->fw_app_cpu_boot_dev_sts1 = RREG32(mmCPU_BOOT_DEV_STS1);
gaudi2->hw_cap_initialized |= HW_CAP_CPU_Q;
return 0;
}
static void gaudi2_init_qman_pq(struct hl_device *hdev, u32 reg_base,
u32 queue_id_base)
{
struct hl_hw_queue *q;
u32 pq_id, pq_offset;
for (pq_id = 0 ; pq_id < NUM_OF_PQ_PER_QMAN ; pq_id++) {
q = &hdev->kernel_queues[queue_id_base + pq_id];
pq_offset = pq_id * 4;
WREG32(reg_base + QM_PQ_BASE_LO_0_OFFSET + pq_offset,
lower_32_bits(q->bus_address));
WREG32(reg_base + QM_PQ_BASE_HI_0_OFFSET + pq_offset,
upper_32_bits(q->bus_address));
WREG32(reg_base + QM_PQ_SIZE_0_OFFSET + pq_offset, ilog2(HL_QUEUE_LENGTH));
WREG32(reg_base + QM_PQ_PI_0_OFFSET + pq_offset, 0);
WREG32(reg_base + QM_PQ_CI_0_OFFSET + pq_offset, 0);
}
}
static void gaudi2_init_qman_cp(struct hl_device *hdev, u32 reg_base)
{
u32 cp_id, cp_offset, mtr_base_lo, mtr_base_hi, so_base_lo, so_base_hi;
mtr_base_lo = lower_32_bits(CFG_BASE + mmDCORE0_SYNC_MNGR_OBJS_MON_PAY_ADDRL_0);
mtr_base_hi = upper_32_bits(CFG_BASE + mmDCORE0_SYNC_MNGR_OBJS_MON_PAY_ADDRL_0);
so_base_lo = lower_32_bits(CFG_BASE + mmDCORE0_SYNC_MNGR_OBJS_SOB_OBJ_0);
so_base_hi = upper_32_bits(CFG_BASE + mmDCORE0_SYNC_MNGR_OBJS_SOB_OBJ_0);
for (cp_id = 0 ; cp_id < NUM_OF_CP_PER_QMAN; cp_id++) {
cp_offset = cp_id * 4;
WREG32(reg_base + QM_CP_MSG_BASE0_ADDR_LO_0_OFFSET + cp_offset, mtr_base_lo);
WREG32(reg_base + QM_CP_MSG_BASE0_ADDR_HI_0_OFFSET + cp_offset, mtr_base_hi);
WREG32(reg_base + QM_CP_MSG_BASE1_ADDR_LO_0_OFFSET + cp_offset, so_base_lo);
WREG32(reg_base + QM_CP_MSG_BASE1_ADDR_HI_0_OFFSET + cp_offset, so_base_hi);
}
/* allow QMANs to accept work from ARC CQF */
WREG32(reg_base + QM_CP_CFG_OFFSET, FIELD_PREP(PDMA0_QM_CP_CFG_SWITCH_EN_MASK, 0x1));
}
static void gaudi2_init_qman_pqc(struct hl_device *hdev, u32 reg_base,
u32 queue_id_base)
{
struct gaudi2_device *gaudi2 = hdev->asic_specific;
u32 pq_id, pq_offset, so_base_lo, so_base_hi;
so_base_lo = lower_32_bits(CFG_BASE + mmDCORE0_SYNC_MNGR_OBJS_SOB_OBJ_0);
so_base_hi = upper_32_bits(CFG_BASE + mmDCORE0_SYNC_MNGR_OBJS_SOB_OBJ_0);
for (pq_id = 0 ; pq_id < NUM_OF_PQ_PER_QMAN ; pq_id++) {
pq_offset = pq_id * 4;
/* Configure QMAN HBW to scratchpad as it is not needed */
WREG32(reg_base + QM_PQC_HBW_BASE_LO_0_OFFSET + pq_offset,
lower_32_bits(gaudi2->scratchpad_bus_address));
WREG32(reg_base + QM_PQC_HBW_BASE_HI_0_OFFSET + pq_offset,
upper_32_bits(gaudi2->scratchpad_bus_address));
WREG32(reg_base + QM_PQC_SIZE_0_OFFSET + pq_offset,
ilog2(PAGE_SIZE / sizeof(struct hl_cq_entry)));
WREG32(reg_base + QM_PQC_PI_0_OFFSET + pq_offset, 0);
WREG32(reg_base + QM_PQC_LBW_WDATA_0_OFFSET + pq_offset, QM_PQC_LBW_WDATA);
WREG32(reg_base + QM_PQC_LBW_BASE_LO_0_OFFSET + pq_offset, so_base_lo);
WREG32(reg_base + QM_PQC_LBW_BASE_HI_0_OFFSET + pq_offset, so_base_hi);
}
/* Enable QMAN H/W completion */
WREG32(reg_base + QM_PQC_CFG_OFFSET, 1 << PDMA0_QM_PQC_CFG_EN_SHIFT);
}
static u32 gaudi2_get_dyn_sp_reg(struct hl_device *hdev, u32 queue_id_base)
{
struct cpu_dyn_regs *dyn_regs = &hdev->fw_loader.dynamic_loader.comm_desc.cpu_dyn_regs;
u32 sp_reg_addr;
switch (queue_id_base) {
case GAUDI2_QUEUE_ID_PDMA_0_0...GAUDI2_QUEUE_ID_PDMA_1_3:
fallthrough;
case GAUDI2_QUEUE_ID_DCORE0_EDMA_0_0...GAUDI2_QUEUE_ID_DCORE0_EDMA_1_3:
fallthrough;
case GAUDI2_QUEUE_ID_DCORE1_EDMA_0_0...GAUDI2_QUEUE_ID_DCORE1_EDMA_1_3:
fallthrough;
case GAUDI2_QUEUE_ID_DCORE2_EDMA_0_0...GAUDI2_QUEUE_ID_DCORE2_EDMA_1_3:
fallthrough;
case GAUDI2_QUEUE_ID_DCORE3_EDMA_0_0...GAUDI2_QUEUE_ID_DCORE3_EDMA_1_3:
sp_reg_addr = le32_to_cpu(dyn_regs->gic_dma_qm_irq_ctrl);
break;
case GAUDI2_QUEUE_ID_DCORE0_MME_0_0...GAUDI2_QUEUE_ID_DCORE0_MME_0_3:
fallthrough;
case GAUDI2_QUEUE_ID_DCORE1_MME_0_0...GAUDI2_QUEUE_ID_DCORE1_MME_0_3:
fallthrough;
case GAUDI2_QUEUE_ID_DCORE2_MME_0_0...GAUDI2_QUEUE_ID_DCORE2_MME_0_3:
fallthrough;
case GAUDI2_QUEUE_ID_DCORE3_MME_0_0...GAUDI2_QUEUE_ID_DCORE3_MME_0_3:
sp_reg_addr = le32_to_cpu(dyn_regs->gic_mme_qm_irq_ctrl);
break;
case GAUDI2_QUEUE_ID_DCORE0_TPC_0_0 ... GAUDI2_QUEUE_ID_DCORE0_TPC_6_3:
fallthrough;
case GAUDI2_QUEUE_ID_DCORE1_TPC_0_0 ... GAUDI2_QUEUE_ID_DCORE1_TPC_5_3:
fallthrough;
case GAUDI2_QUEUE_ID_DCORE2_TPC_0_0 ... GAUDI2_QUEUE_ID_DCORE2_TPC_5_3:
fallthrough;
case GAUDI2_QUEUE_ID_DCORE3_TPC_0_0 ... GAUDI2_QUEUE_ID_DCORE3_TPC_5_3:
sp_reg_addr = le32_to_cpu(dyn_regs->gic_tpc_qm_irq_ctrl);
break;
case GAUDI2_QUEUE_ID_ROT_0_0...GAUDI2_QUEUE_ID_ROT_1_3:
sp_reg_addr = le32_to_cpu(dyn_regs->gic_rot_qm_irq_ctrl);
break;
case GAUDI2_QUEUE_ID_NIC_0_0...GAUDI2_QUEUE_ID_NIC_23_3:
sp_reg_addr = le32_to_cpu(dyn_regs->gic_nic_qm_irq_ctrl);
break;
default:
dev_err(hdev->dev, "Unexpected h/w queue %d\n", queue_id_base);
return 0;
}
return sp_reg_addr;
}
static void gaudi2_init_qman_common(struct hl_device *hdev, u32 reg_base,
u32 queue_id_base)
{
u32 glbl_prot = QMAN_MAKE_TRUSTED, irq_handler_offset;
int map_table_entry;
WREG32(reg_base + QM_GLBL_PROT_OFFSET, glbl_prot);
irq_handler_offset = gaudi2_get_dyn_sp_reg(hdev, queue_id_base);
WREG32(reg_base + QM_GLBL_ERR_ADDR_LO_OFFSET, lower_32_bits(CFG_BASE + irq_handler_offset));
WREG32(reg_base + QM_GLBL_ERR_ADDR_HI_OFFSET, upper_32_bits(CFG_BASE + irq_handler_offset));
map_table_entry = gaudi2_qman_async_event_id[queue_id_base];
WREG32(reg_base + QM_GLBL_ERR_WDATA_OFFSET,
gaudi2_irq_map_table[map_table_entry].cpu_id);
WREG32(reg_base + QM_ARB_ERR_MSG_EN_OFFSET, QM_ARB_ERR_MSG_EN_MASK);
WREG32(reg_base + QM_ARB_SLV_CHOISE_WDT_OFFSET, GAUDI2_ARB_WDT_TIMEOUT);
WREG32(reg_base + QM_GLBL_CFG1_OFFSET, 0);
WREG32(reg_base + QM_GLBL_CFG2_OFFSET, 0);
/* Enable the QMAN channel.
habanalabs/gaudi2: assigning PQFs for ARC f/w in PDMA Assigning 3 PQFs in PDMA1 and 2 PQFs in PDMA0 for ARC firmware usage. Signed-off-by: Rajarama Manjukody Bhat <rmbhat@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-08-12 14:28:20 +08:00
* PDMA QMAN configuration is different, as we do not allow user to
* access some of the CPs.
* PDMA0: CP2/3 are reserved for the ARC usage.
* PDMA1: CP1/2/3 are reserved for the ARC usage.
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
*/
if (reg_base == gaudi2_qm_blocks_bases[GAUDI2_QUEUE_ID_PDMA_1_0])
WREG32(reg_base + QM_GLBL_CFG0_OFFSET, PDMA1_QMAN_ENABLE);
habanalabs/gaudi2: assigning PQFs for ARC f/w in PDMA Assigning 3 PQFs in PDMA1 and 2 PQFs in PDMA0 for ARC firmware usage. Signed-off-by: Rajarama Manjukody Bhat <rmbhat@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-08-12 14:28:20 +08:00
else if (reg_base == gaudi2_qm_blocks_bases[GAUDI2_QUEUE_ID_PDMA_0_0])
WREG32(reg_base + QM_GLBL_CFG0_OFFSET, PDMA0_QMAN_ENABLE);
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
else
WREG32(reg_base + QM_GLBL_CFG0_OFFSET, QMAN_ENABLE);
}
static void gaudi2_init_qman(struct hl_device *hdev, u32 reg_base,
u32 queue_id_base)
{
u32 pq_id;
for (pq_id = 0 ; pq_id < NUM_OF_PQ_PER_QMAN ; pq_id++)
habanalabs/gaudi2: replace defines for reserved sob/mob with enums Following patches are going to add more reserved sync objects and monitors. To make the counting of these reserved resources simpler, replace the existing RESERVED_* defines with enumerations. Signed-off-by: Tomer Tayar <ttayar@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-30 00:20:38 +08:00
hdev->kernel_queues[queue_id_base + pq_id].cq_id = GAUDI2_RESERVED_CQ_CS_COMPLETION;
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
gaudi2_init_qman_pq(hdev, reg_base, queue_id_base);
gaudi2_init_qman_cp(hdev, reg_base);
gaudi2_init_qman_pqc(hdev, reg_base, queue_id_base);
gaudi2_init_qman_common(hdev, reg_base, queue_id_base);
}
static void gaudi2_init_dma_core(struct hl_device *hdev, u32 reg_base,
u32 dma_core_id, bool is_secure)
{
u32 prot, irq_handler_offset;
struct cpu_dyn_regs *dyn_regs;
int map_table_entry;
prot = 1 << ARC_FARM_KDMA_PROT_ERR_VAL_SHIFT;
if (is_secure)
prot |= 1 << ARC_FARM_KDMA_PROT_VAL_SHIFT;
WREG32(reg_base + DMA_CORE_PROT_OFFSET, prot);
dyn_regs = &hdev->fw_loader.dynamic_loader.comm_desc.cpu_dyn_regs;
irq_handler_offset = le32_to_cpu(dyn_regs->gic_dma_core_irq_ctrl);
WREG32(reg_base + DMA_CORE_ERRMSG_ADDR_LO_OFFSET,
lower_32_bits(CFG_BASE + irq_handler_offset));
WREG32(reg_base + DMA_CORE_ERRMSG_ADDR_HI_OFFSET,
upper_32_bits(CFG_BASE + irq_handler_offset));
map_table_entry = gaudi2_dma_core_async_event_id[dma_core_id];
WREG32(reg_base + DMA_CORE_ERRMSG_WDATA_OFFSET,
gaudi2_irq_map_table[map_table_entry].cpu_id);
/* Enable the DMA channel */
WREG32(reg_base + DMA_CORE_CFG_0_OFFSET, 1 << ARC_FARM_KDMA_CFG_0_EN_SHIFT);
}
static void gaudi2_init_kdma(struct hl_device *hdev)
{
struct gaudi2_device *gaudi2 = hdev->asic_specific;
u32 reg_base;
if ((gaudi2->hw_cap_initialized & HW_CAP_KDMA) == HW_CAP_KDMA)
return;
reg_base = gaudi2_dma_core_blocks_bases[DMA_CORE_ID_KDMA];
gaudi2_init_dma_core(hdev, reg_base, DMA_CORE_ID_KDMA, true);
gaudi2->hw_cap_initialized |= HW_CAP_KDMA;
}
static void gaudi2_init_pdma(struct hl_device *hdev)
{
struct gaudi2_device *gaudi2 = hdev->asic_specific;
u32 reg_base;
if ((gaudi2->hw_cap_initialized & HW_CAP_PDMA_MASK) == HW_CAP_PDMA_MASK)
return;
reg_base = gaudi2_dma_core_blocks_bases[DMA_CORE_ID_PDMA0];
gaudi2_init_dma_core(hdev, reg_base, DMA_CORE_ID_PDMA0, false);
reg_base = gaudi2_qm_blocks_bases[GAUDI2_QUEUE_ID_PDMA_0_0];
gaudi2_init_qman(hdev, reg_base, GAUDI2_QUEUE_ID_PDMA_0_0);
reg_base = gaudi2_dma_core_blocks_bases[DMA_CORE_ID_PDMA1];
gaudi2_init_dma_core(hdev, reg_base, DMA_CORE_ID_PDMA1, false);
reg_base = gaudi2_qm_blocks_bases[GAUDI2_QUEUE_ID_PDMA_1_0];
gaudi2_init_qman(hdev, reg_base, GAUDI2_QUEUE_ID_PDMA_1_0);
gaudi2->hw_cap_initialized |= HW_CAP_PDMA_MASK;
}
static void gaudi2_init_edma_instance(struct hl_device *hdev, u8 seq)
{
u32 reg_base, base_edma_core_id, base_edma_qman_id;
base_edma_core_id = DMA_CORE_ID_EDMA0 + seq;
base_edma_qman_id = edma_stream_base[seq];
reg_base = gaudi2_dma_core_blocks_bases[base_edma_core_id];
gaudi2_init_dma_core(hdev, reg_base, base_edma_core_id, false);
reg_base = gaudi2_qm_blocks_bases[base_edma_qman_id];
gaudi2_init_qman(hdev, reg_base, base_edma_qman_id);
}
static void gaudi2_init_edma(struct hl_device *hdev)
{
struct asic_fixed_properties *prop = &hdev->asic_prop;
struct gaudi2_device *gaudi2 = hdev->asic_specific;
int dcore, inst;
if ((gaudi2->hw_cap_initialized & HW_CAP_EDMA_MASK) == HW_CAP_EDMA_MASK)
return;
for (dcore = 0 ; dcore < NUM_OF_DCORES ; dcore++) {
for (inst = 0 ; inst < NUM_OF_EDMA_PER_DCORE ; inst++) {
u8 seq = dcore * NUM_OF_EDMA_PER_DCORE + inst;
if (!(prop->edma_enabled_mask & BIT(seq)))
continue;
gaudi2_init_edma_instance(hdev, seq);
gaudi2->hw_cap_initialized |= BIT_ULL(HW_CAP_EDMA_SHIFT + seq);
}
}
}
habanalabs/gaudi2: modify decoder to use virtual MSI-X doorbell Modify the decoder wrapper blocks to generate interrupts using the virtual MSI-X doorbell. As a decoder wrapper block cannot write directly to HBW upon completion, it writes instead to SOB which is monitored by a master monitor. When resolved, this monitor will be the one to actually write to the virtual MSI-X doorbell. Signed-off-by: Tomer Tayar <ttayar@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-30 16:22:54 +08:00
/*
* gaudi2_arm_monitors_for_virt_msix_db() - Arm monitors for writing to the virtual MSI-X doorbell.
* @hdev: pointer to habanalabs device structure.
* @sob_id: sync object ID.
* @first_mon_id: ID of first monitor out of 3 consecutive monitors.
* @interrupt_id: interrupt ID.
*
* Some initiators cannot have HBW address in their completion address registers, and thus cannot
* write directly to the HBW host memory of the virtual MSI-X doorbell.
* Instead, they are configured to LBW write to a sync object, and a monitor will do the HBW write.
*
* The mechanism in the sync manager block is composed of a master monitor with 3 messages.
* In addition to the HBW write, the other 2 messages are for preparing the monitor to next
* completion, by decrementing the sync object value and re-arming the monitor.
*/
static void gaudi2_arm_monitors_for_virt_msix_db(struct hl_device *hdev, u32 sob_id,
u32 first_mon_id, u32 interrupt_id)
{
u32 sob_offset, first_mon_offset, mon_offset, payload, sob_group, mode, arm, config;
struct gaudi2_device *gaudi2 = hdev->asic_specific;
u64 addr;
u8 mask;
/* Reset the SOB value */
sob_offset = sob_id * sizeof(u32);
WREG32(mmDCORE0_SYNC_MNGR_OBJS_SOB_OBJ_0 + sob_offset, 0);
/* Configure 3 monitors:
* 1. Write interrupt ID to the virtual MSI-X doorbell (master monitor)
* 2. Decrement SOB value by 1.
* 3. Re-arm the master monitor.
*/
first_mon_offset = first_mon_id * sizeof(u32);
/* 2nd monitor: Decrement SOB value by 1 */
mon_offset = first_mon_offset + sizeof(u32);
addr = CFG_BASE + mmDCORE0_SYNC_MNGR_OBJS_SOB_OBJ_0 + sob_offset;
WREG32(mmDCORE0_SYNC_MNGR_OBJS_MON_PAY_ADDRL_0 + mon_offset, lower_32_bits(addr));
WREG32(mmDCORE0_SYNC_MNGR_OBJS_MON_PAY_ADDRH_0 + mon_offset, upper_32_bits(addr));
payload = FIELD_PREP(DCORE0_SYNC_MNGR_OBJS_SOB_OBJ_VAL_MASK, 0x7FFF) | /* "-1" */
FIELD_PREP(DCORE0_SYNC_MNGR_OBJS_SOB_OBJ_SIGN_MASK, 1) |
FIELD_PREP(DCORE0_SYNC_MNGR_OBJS_SOB_OBJ_INC_MASK, 1);
WREG32(mmDCORE0_SYNC_MNGR_OBJS_MON_PAY_DATA_0 + mon_offset, payload);
/* 3rd monitor: Re-arm the master monitor */
mon_offset = first_mon_offset + 2 * sizeof(u32);
addr = CFG_BASE + mmDCORE0_SYNC_MNGR_OBJS_MON_ARM_0 + first_mon_offset;
WREG32(mmDCORE0_SYNC_MNGR_OBJS_MON_PAY_ADDRL_0 + mon_offset, lower_32_bits(addr));
WREG32(mmDCORE0_SYNC_MNGR_OBJS_MON_PAY_ADDRH_0 + mon_offset, upper_32_bits(addr));
sob_group = sob_id / 8;
mask = ~BIT(sob_id & 0x7);
mode = 0; /* comparison mode is "greater than or equal to" */
arm = FIELD_PREP(DCORE0_SYNC_MNGR_OBJS_MON_ARM_SID_MASK, sob_group) |
FIELD_PREP(DCORE0_SYNC_MNGR_OBJS_MON_ARM_MASK_MASK, mask) |
FIELD_PREP(DCORE0_SYNC_MNGR_OBJS_MON_ARM_SOP_MASK, mode) |
FIELD_PREP(DCORE0_SYNC_MNGR_OBJS_MON_ARM_SOD_MASK, 1);
payload = arm;
WREG32(mmDCORE0_SYNC_MNGR_OBJS_MON_PAY_DATA_0 + mon_offset, payload);
/* 1st monitor (master): Write interrupt ID to the virtual MSI-X doorbell */
mon_offset = first_mon_offset;
config = FIELD_PREP(DCORE0_SYNC_MNGR_OBJS_MON_CONFIG_WR_NUM_MASK, 2); /* "2": 3 writes */
WREG32(mmDCORE0_SYNC_MNGR_OBJS_MON_CONFIG_0 + mon_offset, config);
addr = gaudi2->virt_msix_db_dma_addr;
WREG32(mmDCORE0_SYNC_MNGR_OBJS_MON_PAY_ADDRL_0 + mon_offset, lower_32_bits(addr));
WREG32(mmDCORE0_SYNC_MNGR_OBJS_MON_PAY_ADDRH_0 + mon_offset, upper_32_bits(addr));
payload = interrupt_id;
WREG32(mmDCORE0_SYNC_MNGR_OBJS_MON_PAY_DATA_0 + mon_offset, payload);
WREG32(mmDCORE0_SYNC_MNGR_OBJS_MON_ARM_0 + mon_offset, arm);
}
static void gaudi2_prepare_sm_for_virt_msix_db(struct hl_device *hdev)
{
u32 decoder_id, sob_id, first_mon_id, interrupt_id;
struct asic_fixed_properties *prop = &hdev->asic_prop;
/* Decoder normal/abnormal interrupts */
for (decoder_id = 0 ; decoder_id < NUMBER_OF_DEC ; ++decoder_id) {
if (!(prop->decoder_enabled_mask & BIT(decoder_id)))
continue;
sob_id = GAUDI2_RESERVED_SOB_DEC_NRM_FIRST + decoder_id;
first_mon_id = GAUDI2_RESERVED_MON_DEC_NRM_FIRST + 3 * decoder_id;
interrupt_id = GAUDI2_IRQ_NUM_DCORE0_DEC0_NRM + 2 * decoder_id;
gaudi2_arm_monitors_for_virt_msix_db(hdev, sob_id, first_mon_id, interrupt_id);
sob_id = GAUDI2_RESERVED_SOB_DEC_ABNRM_FIRST + decoder_id;
first_mon_id = GAUDI2_RESERVED_MON_DEC_ABNRM_FIRST + 3 * decoder_id;
interrupt_id += 1;
gaudi2_arm_monitors_for_virt_msix_db(hdev, sob_id, first_mon_id, interrupt_id);
}
}
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
static void gaudi2_init_sm(struct hl_device *hdev)
{
habanalabs/gaudi2: modify CS completion CQ to use virtual MSI-X doorbell Modify the CQ which is used for CS completion, to use the virtual MSI-X doorbell. Signed-off-by: Tomer Tayar <ttayar@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-30 23:40:59 +08:00
struct gaudi2_device *gaudi2 = hdev->asic_specific;
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
u64 cq_address;
u32 reg_val;
int i;
/* Enable HBW/LBW CQ for completion monitors */
reg_val = FIELD_PREP(DCORE0_SYNC_MNGR_OBJS_MON_CONFIG_CQ_EN_MASK, 1);
reg_val |= FIELD_PREP(DCORE0_SYNC_MNGR_OBJS_MON_CONFIG_LBW_EN_MASK, 1);
for (i = 0 ; i < GAUDI2_MAX_PENDING_CS ; i++)
WREG32(mmDCORE0_SYNC_MNGR_OBJS_MON_CONFIG_0 + (4 * i), reg_val);
/* Enable only HBW CQ for KDMA completion monitor */
reg_val = FIELD_PREP(DCORE0_SYNC_MNGR_OBJS_MON_CONFIG_CQ_EN_MASK, 1);
WREG32(mmDCORE0_SYNC_MNGR_OBJS_MON_CONFIG_0 + (4 * i), reg_val);
/* Init CQ0 DB */
/* Configure the monitor to trigger MSI-X interrupt */
habanalabs/gaudi2: modify CS completion CQ to use virtual MSI-X doorbell Modify the CQ which is used for CS completion, to use the virtual MSI-X doorbell. Signed-off-by: Tomer Tayar <ttayar@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-30 23:40:59 +08:00
/* TODO:
* Remove the if statement when virtual MSI-X doorbell is supported in simulator (SW-93022)
* and in F/W (SW-93024).
*/
if (!hdev->pdev || hdev->asic_prop.fw_security_enabled) {
u64 msix_db_reg = CFG_BASE + mmPCIE_DBI_MSIX_DOORBELL_OFF;
WREG32(mmDCORE0_SYNC_MNGR_GLBL_LBW_ADDR_L_0, lower_32_bits(msix_db_reg));
WREG32(mmDCORE0_SYNC_MNGR_GLBL_LBW_ADDR_H_0, upper_32_bits(msix_db_reg));
} else {
WREG32(mmDCORE0_SYNC_MNGR_GLBL_LBW_ADDR_L_0,
lower_32_bits(gaudi2->virt_msix_db_dma_addr));
WREG32(mmDCORE0_SYNC_MNGR_GLBL_LBW_ADDR_H_0,
upper_32_bits(gaudi2->virt_msix_db_dma_addr));
}
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
WREG32(mmDCORE0_SYNC_MNGR_GLBL_LBW_DATA_0, GAUDI2_IRQ_NUM_COMPLETION);
for (i = 0 ; i < GAUDI2_RESERVED_CQ_NUMBER ; i++) {
cq_address =
hdev->completion_queue[i].bus_address;
WREG32(mmDCORE0_SYNC_MNGR_GLBL_CQ_BASE_ADDR_L_0 + (4 * i),
lower_32_bits(cq_address));
WREG32(mmDCORE0_SYNC_MNGR_GLBL_CQ_BASE_ADDR_H_0 + (4 * i),
upper_32_bits(cq_address));
WREG32(mmDCORE0_SYNC_MNGR_GLBL_CQ_SIZE_LOG2_0 + (4 * i),
ilog2(HL_CQ_SIZE_IN_BYTES));
}
/* Configure kernel ASID and MMU BP*/
WREG32(mmDCORE0_SYNC_MNGR_GLBL_ASID_SEC, 0x10000);
WREG32(mmDCORE0_SYNC_MNGR_GLBL_ASID_NONE_SEC_PRIV, 0);
habanalabs/gaudi2: modify decoder to use virtual MSI-X doorbell Modify the decoder wrapper blocks to generate interrupts using the virtual MSI-X doorbell. As a decoder wrapper block cannot write directly to HBW upon completion, it writes instead to SOB which is monitored by a master monitor. When resolved, this monitor will be the one to actually write to the virtual MSI-X doorbell. Signed-off-by: Tomer Tayar <ttayar@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-30 16:22:54 +08:00
/* Initialize sync objects and monitors which are used for the virtual MSI-X doorbell */
gaudi2_prepare_sm_for_virt_msix_db(hdev);
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
}
static void gaudi2_init_mme_acc(struct hl_device *hdev, u32 reg_base)
{
struct gaudi2_device *gaudi2 = hdev->asic_specific;
u32 reg_val;
int i;
reg_val = FIELD_PREP(MME_ACC_INTR_MASK_WBC_ERR_RESP_MASK, 0);
reg_val |= FIELD_PREP(MME_ACC_INTR_MASK_AP_SRC_POS_INF_MASK, 1);
reg_val |= FIELD_PREP(MME_ACC_INTR_MASK_AP_SRC_NEG_INF_MASK, 1);
reg_val |= FIELD_PREP(MME_ACC_INTR_MASK_AP_SRC_NAN_MASK, 1);
reg_val |= FIELD_PREP(MME_ACC_INTR_MASK_AP_RESULT_POS_INF_MASK, 1);
reg_val |= FIELD_PREP(MME_ACC_INTR_MASK_AP_RESULT_NEG_INF_MASK, 1);
WREG32(reg_base + MME_ACC_INTR_MASK_OFFSET, reg_val);
WREG32(reg_base + MME_ACC_AP_LFSR_POLY_OFFSET, 0x80DEADAF);
for (i = 0 ; i < MME_NUM_OF_LFSR_SEEDS ; i++) {
WREG32(reg_base + MME_ACC_AP_LFSR_SEED_SEL_OFFSET, i);
WREG32(reg_base + MME_ACC_AP_LFSR_SEED_WDATA_OFFSET, gaudi2->lfsr_rand_seeds[i]);
}
}
static void gaudi2_init_dcore_mme(struct hl_device *hdev, int dcore_id,
bool config_qman_only)
{
u32 queue_id_base, reg_base, clk_en_addr = 0;
switch (dcore_id) {
case 0:
queue_id_base = GAUDI2_QUEUE_ID_DCORE0_MME_0_0;
break;
case 1:
queue_id_base = GAUDI2_QUEUE_ID_DCORE1_MME_0_0;
clk_en_addr = mmDCORE1_MME_CTRL_LO_QM_SLV_CLK_EN;
break;
case 2:
queue_id_base = GAUDI2_QUEUE_ID_DCORE2_MME_0_0;
break;
case 3:
queue_id_base = GAUDI2_QUEUE_ID_DCORE3_MME_0_0;
clk_en_addr = mmDCORE3_MME_CTRL_LO_QM_SLV_CLK_EN;
break;
default:
dev_err(hdev->dev, "Invalid dcore id %u\n", dcore_id);
return;
}
if (clk_en_addr && !(hdev->fw_components & FW_TYPE_BOOT_CPU))
WREG32(clk_en_addr, 0x1);
if (!config_qman_only) {
reg_base = gaudi2_mme_acc_blocks_bases[dcore_id];
gaudi2_init_mme_acc(hdev, reg_base);
}
reg_base = gaudi2_qm_blocks_bases[queue_id_base];
gaudi2_init_qman(hdev, reg_base, queue_id_base);
}
static void gaudi2_init_mme(struct hl_device *hdev)
{
struct gaudi2_device *gaudi2 = hdev->asic_specific;
int i;
if ((gaudi2->hw_cap_initialized & HW_CAP_MME_MASK) == HW_CAP_MME_MASK)
return;
for (i = 0 ; i < NUM_OF_DCORES ; i++) {
gaudi2_init_dcore_mme(hdev, i, false);
gaudi2->hw_cap_initialized |= BIT_ULL(HW_CAP_MME_SHIFT + i);
}
}
static void gaudi2_init_tpc_cfg(struct hl_device *hdev, u32 reg_base)
{
/* Mask arithmetic and QM interrupts in TPC */
WREG32(reg_base + TPC_CFG_TPC_INTR_MASK_OFFSET, 0x23FFFE);
/* Set 16 cache lines */
WREG32(reg_base + TPC_CFG_MSS_CONFIG_OFFSET,
2 << DCORE0_TPC0_CFG_MSS_CONFIG_ICACHE_FETCH_LINE_NUM_SHIFT);
}
struct gaudi2_tpc_init_cfg_data {
enum gaudi2_queue_id dcore_tpc_qid_base[NUM_OF_DCORES];
};
static void gaudi2_init_tpc_config(struct hl_device *hdev, int dcore, int inst,
habanalabs: add return code field to module iterator Up until now the module iterator called void callback functions and so caller activating callback that may fail suffered from 2 issues: 1. The need to "plant" return called in the private data. This is a drawback since the iterator itself should not be aware of the private data of the caller. 2. Due to 1 even in a failure the iterator would keep iterating instead of break upon error. To overcome this an optional rc field added to the iterator context. Signed-off-by: Ohad Sharabi <osharabi@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-07-13 18:47:23 +08:00
u32 offset, struct iterate_module_ctx *ctx)
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
{
struct gaudi2_device *gaudi2 = hdev->asic_specific;
habanalabs: add return code field to module iterator Up until now the module iterator called void callback functions and so caller activating callback that may fail suffered from 2 issues: 1. The need to "plant" return called in the private data. This is a drawback since the iterator itself should not be aware of the private data of the caller. 2. Due to 1 even in a failure the iterator would keep iterating instead of break upon error. To overcome this an optional rc field added to the iterator context. Signed-off-by: Ohad Sharabi <osharabi@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-07-13 18:47:23 +08:00
struct gaudi2_tpc_init_cfg_data *cfg_data = ctx->data;
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
u32 queue_id_base;
u8 seq;
queue_id_base = cfg_data->dcore_tpc_qid_base[dcore] + (inst * NUM_OF_PQ_PER_QMAN);
if (dcore == 0 && inst == (NUM_DCORE0_TPC - 1))
/* gets last sequence number */
seq = NUM_OF_DCORES * NUM_OF_TPC_PER_DCORE;
else
seq = dcore * NUM_OF_TPC_PER_DCORE + inst;
gaudi2_init_tpc_cfg(hdev, mmDCORE0_TPC0_CFG_BASE + offset);
gaudi2_init_qman(hdev, mmDCORE0_TPC0_QM_BASE + offset, queue_id_base);
gaudi2->tpc_hw_cap_initialized |= BIT_ULL(HW_CAP_TPC_SHIFT + seq);
}
static void gaudi2_init_tpc(struct hl_device *hdev)
{
struct gaudi2_device *gaudi2 = hdev->asic_specific;
struct gaudi2_tpc_init_cfg_data init_cfg_data;
struct iterate_module_ctx tpc_iter;
if (!hdev->asic_prop.tpc_enabled_mask)
return;
if ((gaudi2->tpc_hw_cap_initialized & HW_CAP_TPC_MASK) == HW_CAP_TPC_MASK)
return;
init_cfg_data.dcore_tpc_qid_base[0] = GAUDI2_QUEUE_ID_DCORE0_TPC_0_0;
init_cfg_data.dcore_tpc_qid_base[1] = GAUDI2_QUEUE_ID_DCORE1_TPC_0_0;
init_cfg_data.dcore_tpc_qid_base[2] = GAUDI2_QUEUE_ID_DCORE2_TPC_0_0;
init_cfg_data.dcore_tpc_qid_base[3] = GAUDI2_QUEUE_ID_DCORE3_TPC_0_0;
tpc_iter.fn = &gaudi2_init_tpc_config;
tpc_iter.data = &init_cfg_data;
gaudi2_iterate_tpcs(hdev, &tpc_iter);
}
static void gaudi2_init_rotator(struct hl_device *hdev)
{
struct gaudi2_device *gaudi2 = hdev->asic_specific;
u32 i, reg_base, queue_id;
queue_id = GAUDI2_QUEUE_ID_ROT_0_0;
for (i = 0 ; i < NUM_OF_ROT ; i++, queue_id += NUM_OF_PQ_PER_QMAN) {
reg_base = gaudi2_qm_blocks_bases[queue_id];
gaudi2_init_qman(hdev, reg_base, queue_id);
gaudi2->hw_cap_initialized |= BIT_ULL(HW_CAP_ROT_SHIFT + i);
}
}
habanalabs/gaudi2: modify decoder to use virtual MSI-X doorbell Modify the decoder wrapper blocks to generate interrupts using the virtual MSI-X doorbell. As a decoder wrapper block cannot write directly to HBW upon completion, it writes instead to SOB which is monitored by a master monitor. When resolved, this monitor will be the one to actually write to the virtual MSI-X doorbell. Signed-off-by: Tomer Tayar <ttayar@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-30 16:22:54 +08:00
static void gaudi2_init_vdec_brdg_ctrl(struct hl_device *hdev, u64 base_addr, u32 decoder_id)
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
{
habanalabs/gaudi2: modify decoder to use virtual MSI-X doorbell Modify the decoder wrapper blocks to generate interrupts using the virtual MSI-X doorbell. As a decoder wrapper block cannot write directly to HBW upon completion, it writes instead to SOB which is monitored by a master monitor. When resolved, this monitor will be the one to actually write to the virtual MSI-X doorbell. Signed-off-by: Tomer Tayar <ttayar@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-30 16:22:54 +08:00
u32 sob_id;
/* TODO:
* Remove when virtual MSI-X doorbell is supported in simulator (SW-93022) and in F/W
* (SW-93024).
*/
if (!hdev->pdev || hdev->asic_prop.fw_security_enabled) {
u32 interrupt_id = GAUDI2_IRQ_NUM_DCORE0_DEC0_NRM + 2 * decoder_id;
WREG32(base_addr + BRDG_CTRL_NRM_MSIX_LBW_AWADDR, mmPCIE_DBI_MSIX_DOORBELL_OFF);
WREG32(base_addr + BRDG_CTRL_NRM_MSIX_LBW_WDATA, interrupt_id);
WREG32(base_addr + BRDG_CTRL_ABNRM_MSIX_LBW_AWADDR, mmPCIE_DBI_MSIX_DOORBELL_OFF);
WREG32(base_addr + BRDG_CTRL_ABNRM_MSIX_LBW_WDATA, interrupt_id + 1);
return;
}
/* VCMD normal interrupt */
sob_id = GAUDI2_RESERVED_SOB_DEC_NRM_FIRST + decoder_id;
WREG32(base_addr + BRDG_CTRL_NRM_MSIX_LBW_AWADDR,
mmDCORE0_SYNC_MNGR_OBJS_SOB_OBJ_0 + sob_id * sizeof(u32));
WREG32(base_addr + BRDG_CTRL_NRM_MSIX_LBW_WDATA, GAUDI2_SOB_INCREMENT_BY_ONE);
/* VCMD abnormal interrupt */
sob_id = GAUDI2_RESERVED_SOB_DEC_ABNRM_FIRST + decoder_id;
WREG32(base_addr + BRDG_CTRL_ABNRM_MSIX_LBW_AWADDR,
mmDCORE0_SYNC_MNGR_OBJS_SOB_OBJ_0 + sob_id * sizeof(u32));
WREG32(base_addr + BRDG_CTRL_ABNRM_MSIX_LBW_WDATA, GAUDI2_SOB_INCREMENT_BY_ONE);
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
}
static void gaudi2_init_dec(struct hl_device *hdev)
{
struct gaudi2_device *gaudi2 = hdev->asic_specific;
habanalabs/gaudi2: modify decoder to use virtual MSI-X doorbell Modify the decoder wrapper blocks to generate interrupts using the virtual MSI-X doorbell. As a decoder wrapper block cannot write directly to HBW upon completion, it writes instead to SOB which is monitored by a master monitor. When resolved, this monitor will be the one to actually write to the virtual MSI-X doorbell. Signed-off-by: Tomer Tayar <ttayar@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-30 16:22:54 +08:00
u32 dcore_id, dec_id, dec_bit;
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
u64 base_addr;
if (!hdev->asic_prop.decoder_enabled_mask)
return;
if ((gaudi2->dec_hw_cap_initialized & HW_CAP_DEC_MASK) == HW_CAP_DEC_MASK)
return;
for (dcore_id = 0 ; dcore_id < NUM_OF_DCORES ; dcore_id++)
for (dec_id = 0 ; dec_id < NUM_OF_DEC_PER_DCORE ; dec_id++) {
dec_bit = dcore_id * NUM_OF_DEC_PER_DCORE + dec_id;
if (!(hdev->asic_prop.decoder_enabled_mask & BIT(dec_bit)))
continue;
base_addr = mmDCORE0_DEC0_CMD_BASE +
BRDG_CTRL_BLOCK_OFFSET +
dcore_id * DCORE_OFFSET +
dec_id * DCORE_VDEC_OFFSET;
habanalabs/gaudi2: modify decoder to use virtual MSI-X doorbell Modify the decoder wrapper blocks to generate interrupts using the virtual MSI-X doorbell. As a decoder wrapper block cannot write directly to HBW upon completion, it writes instead to SOB which is monitored by a master monitor. When resolved, this monitor will be the one to actually write to the virtual MSI-X doorbell. Signed-off-by: Tomer Tayar <ttayar@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-30 16:22:54 +08:00
gaudi2_init_vdec_brdg_ctrl(hdev, base_addr, dec_bit);
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
gaudi2->dec_hw_cap_initialized |= BIT_ULL(HW_CAP_DEC_SHIFT + dec_bit);
}
for (dec_id = 0 ; dec_id < NUM_OF_PCIE_VDEC ; dec_id++) {
dec_bit = PCIE_DEC_SHIFT + dec_id;
if (!(hdev->asic_prop.decoder_enabled_mask & BIT(dec_bit)))
continue;
base_addr = mmPCIE_DEC0_CMD_BASE + BRDG_CTRL_BLOCK_OFFSET +
dec_id * DCORE_VDEC_OFFSET;
habanalabs/gaudi2: modify decoder to use virtual MSI-X doorbell Modify the decoder wrapper blocks to generate interrupts using the virtual MSI-X doorbell. As a decoder wrapper block cannot write directly to HBW upon completion, it writes instead to SOB which is monitored by a master monitor. When resolved, this monitor will be the one to actually write to the virtual MSI-X doorbell. Signed-off-by: Tomer Tayar <ttayar@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-30 16:22:54 +08:00
gaudi2_init_vdec_brdg_ctrl(hdev, base_addr, dec_bit);
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
gaudi2->dec_hw_cap_initialized |= BIT_ULL(HW_CAP_DEC_SHIFT + dec_bit);
}
}
static void gaudi2_init_msix_gw_table(struct hl_device *hdev)
{
u32 first_reg_offset, last_reg_offset, msix_gw_table_base;
u8 first_bit, last_bit;
int i;
msix_gw_table_base = mmPCIE_WRAP_MSIX_GW_TABLE_0;
first_reg_offset = (GAUDI2_IRQ_NUM_USER_FIRST >> 5) << 2;
first_bit = GAUDI2_IRQ_NUM_USER_FIRST % 32;
last_reg_offset = (GAUDI2_IRQ_NUM_USER_LAST >> 5) << 2;
last_bit = GAUDI2_IRQ_NUM_USER_LAST % 32;
if (first_reg_offset == last_reg_offset) {
WREG32(msix_gw_table_base + first_reg_offset, GENMASK(last_bit, first_bit));
return;
}
WREG32(msix_gw_table_base + first_reg_offset, GENMASK(31, first_bit));
WREG32(msix_gw_table_base + last_reg_offset, GENMASK(last_bit, 0));
for (i = first_reg_offset + 4; i < last_reg_offset ; i += 4)
WREG32(msix_gw_table_base + i, 0xFFFFFFFF);
}
static int gaudi2_mmu_update_asid_hop0_addr(struct hl_device *hdev,
u32 stlb_base, u32 asid, u64 phys_addr)
{
u32 status, timeout_usec;
int rc;
if (hdev->pldm || !hdev->pdev)
timeout_usec = GAUDI2_PLDM_MMU_TIMEOUT_USEC;
else
timeout_usec = MMU_CONFIG_TIMEOUT_USEC;
WREG32(stlb_base + STLB_ASID_OFFSET, asid);
WREG32(stlb_base + STLB_HOP0_PA43_12_OFFSET, phys_addr >> MMU_HOP0_PA43_12_SHIFT);
WREG32(stlb_base + STLB_HOP0_PA63_44_OFFSET, phys_addr >> MMU_HOP0_PA63_44_SHIFT);
WREG32(stlb_base + STLB_BUSY_OFFSET, 0x80000000);
rc = hl_poll_timeout(
hdev,
stlb_base + STLB_BUSY_OFFSET,
status,
!(status & 0x80000000),
1000,
timeout_usec);
if (rc) {
dev_err(hdev->dev, "Timeout during MMU hop0 config of asid %d\n", asid);
return rc;
}
return 0;
}
static void gaudi2_mmu_send_invalidate_cache_cmd(struct hl_device *hdev, u32 stlb_base,
u32 start_offset, u32 inv_start_val,
u32 flags)
{
/* clear PMMU mem line cache (only needed in mmu range invalidation) */
if (flags & MMU_OP_CLEAR_MEMCACHE)
WREG32(mmPMMU_HBW_STLB_MEM_CACHE_INVALIDATION, 0x1);
if (flags & MMU_OP_SKIP_LOW_CACHE_INV)
return;
WREG32(stlb_base + start_offset, inv_start_val);
}
static int gaudi2_mmu_invalidate_cache_status_poll(struct hl_device *hdev, u32 stlb_base,
struct gaudi2_cache_invld_params *inv_params)
{
u32 status, timeout_usec, start_offset;
int rc;
timeout_usec = (hdev->pldm) ? GAUDI2_PLDM_MMU_TIMEOUT_USEC :
GAUDI2_MMU_CACHE_INV_TIMEOUT_USEC;
/* poll PMMU mem line cache (only needed in mmu range invalidation) */
if (inv_params->flags & MMU_OP_CLEAR_MEMCACHE) {
rc = hl_poll_timeout(
hdev,
mmPMMU_HBW_STLB_MEM_CACHE_INV_STATUS,
status,
status & 0x1,
1000,
timeout_usec);
if (rc)
return rc;
/* Need to manually reset the status to 0 */
WREG32(mmPMMU_HBW_STLB_MEM_CACHE_INV_STATUS, 0x0);
}
/* Lower cache does not work with cache lines, hence we can skip its
* invalidation upon map and invalidate only upon unmap
*/
if (inv_params->flags & MMU_OP_SKIP_LOW_CACHE_INV)
return 0;
start_offset = inv_params->range_invalidation ?
STLB_RANGE_CACHE_INVALIDATION_OFFSET : STLB_INV_ALL_START_OFFSET;
rc = hl_poll_timeout(
hdev,
stlb_base + start_offset,
status,
!(status & 0x1),
1000,
timeout_usec);
return rc;
}
bool gaudi2_is_hmmu_enabled(struct hl_device *hdev, int dcore_id, int hmmu_id)
{
struct gaudi2_device *gaudi2 = hdev->asic_specific;
u32 hw_cap;
hw_cap = HW_CAP_DCORE0_DMMU0 << (NUM_OF_HMMU_PER_DCORE * dcore_id + hmmu_id);
if (gaudi2->hw_cap_initialized & hw_cap)
return true;
return false;
}
/* this function shall be called only for HMMUs for which capability bit is set */
static inline u32 get_hmmu_stlb_base(int dcore_id, int hmmu_id)
{
u32 offset;
offset = (u32) (dcore_id * DCORE_OFFSET + hmmu_id * DCORE_HMMU_OFFSET);
return (u32)(mmDCORE0_HMMU0_STLB_BASE + offset);
}
static void gaudi2_mmu_invalidate_cache_trigger(struct hl_device *hdev, u32 stlb_base,
struct gaudi2_cache_invld_params *inv_params)
{
u32 start_offset;
if (inv_params->range_invalidation) {
/* Set the addresses range
* Note: that the start address we set in register, is not included in
* the range of the invalidation, by design.
* that's why we need to set lower address than the one we actually
* want to be included in the range invalidation.
*/
u64 start = inv_params->start_va - 1;
start_offset = STLB_RANGE_CACHE_INVALIDATION_OFFSET;
WREG32(stlb_base + STLB_RANGE_INV_START_LSB_OFFSET,
start >> MMU_RANGE_INV_VA_LSB_SHIFT);
WREG32(stlb_base + STLB_RANGE_INV_START_MSB_OFFSET,
start >> MMU_RANGE_INV_VA_MSB_SHIFT);
WREG32(stlb_base + STLB_RANGE_INV_END_LSB_OFFSET,
inv_params->end_va >> MMU_RANGE_INV_VA_LSB_SHIFT);
WREG32(stlb_base + STLB_RANGE_INV_END_MSB_OFFSET,
inv_params->end_va >> MMU_RANGE_INV_VA_MSB_SHIFT);
} else {
start_offset = STLB_INV_ALL_START_OFFSET;
}
gaudi2_mmu_send_invalidate_cache_cmd(hdev, stlb_base, start_offset,
inv_params->inv_start_val, inv_params->flags);
}
static inline void gaudi2_hmmu_invalidate_cache_trigger(struct hl_device *hdev,
int dcore_id, int hmmu_id,
struct gaudi2_cache_invld_params *inv_params)
{
u32 stlb_base = get_hmmu_stlb_base(dcore_id, hmmu_id);
gaudi2_mmu_invalidate_cache_trigger(hdev, stlb_base, inv_params);
}
static inline int gaudi2_hmmu_invalidate_cache_status_poll(struct hl_device *hdev,
int dcore_id, int hmmu_id,
struct gaudi2_cache_invld_params *inv_params)
{
u32 stlb_base = get_hmmu_stlb_base(dcore_id, hmmu_id);
return gaudi2_mmu_invalidate_cache_status_poll(hdev, stlb_base, inv_params);
}
static int gaudi2_hmmus_invalidate_cache(struct hl_device *hdev,
struct gaudi2_cache_invld_params *inv_params)
{
int dcore_id, hmmu_id;
/* first send all invalidation commands */
for (dcore_id = 0 ; dcore_id < NUM_OF_DCORES ; dcore_id++) {
for (hmmu_id = 0 ; hmmu_id < NUM_OF_HMMU_PER_DCORE ; hmmu_id++) {
if (!gaudi2_is_hmmu_enabled(hdev, dcore_id, hmmu_id))
continue;
gaudi2_hmmu_invalidate_cache_trigger(hdev, dcore_id, hmmu_id, inv_params);
}
}
/* next, poll all invalidations status */
for (dcore_id = 0 ; dcore_id < NUM_OF_DCORES ; dcore_id++) {
for (hmmu_id = 0 ; hmmu_id < NUM_OF_HMMU_PER_DCORE ; hmmu_id++) {
int rc;
if (!gaudi2_is_hmmu_enabled(hdev, dcore_id, hmmu_id))
continue;
rc = gaudi2_hmmu_invalidate_cache_status_poll(hdev, dcore_id, hmmu_id,
inv_params);
if (rc)
return rc;
}
}
return 0;
}
static int gaudi2_mmu_invalidate_cache(struct hl_device *hdev, bool is_hard, u32 flags)
{
struct gaudi2_device *gaudi2 = hdev->asic_specific;
struct gaudi2_cache_invld_params invld_params;
int rc = 0;
if (hdev->reset_info.hard_reset_pending)
return rc;
invld_params.range_invalidation = false;
invld_params.inv_start_val = 1;
if ((flags & MMU_OP_USERPTR) && (gaudi2->hw_cap_initialized & HW_CAP_PMMU)) {
invld_params.flags = flags;
gaudi2_mmu_invalidate_cache_trigger(hdev, mmPMMU_HBW_STLB_BASE, &invld_params);
rc = gaudi2_mmu_invalidate_cache_status_poll(hdev, mmPMMU_HBW_STLB_BASE,
&invld_params);
} else if (flags & MMU_OP_PHYS_PACK) {
invld_params.flags = 0;
rc = gaudi2_hmmus_invalidate_cache(hdev, &invld_params);
}
return rc;
}
static int gaudi2_mmu_invalidate_cache_range(struct hl_device *hdev, bool is_hard,
u32 flags, u32 asid, u64 va, u64 size)
{
struct gaudi2_cache_invld_params invld_params = {0};
struct gaudi2_device *gaudi2 = hdev->asic_specific;
u64 start_va, end_va;
u32 inv_start_val;
int rc = 0;
if (hdev->reset_info.hard_reset_pending)
return 0;
inv_start_val = (1 << MMU_RANGE_INV_EN_SHIFT |
1 << MMU_RANGE_INV_ASID_EN_SHIFT |
asid << MMU_RANGE_INV_ASID_SHIFT);
start_va = va;
end_va = start_va + size;
if ((flags & MMU_OP_USERPTR) && (gaudi2->hw_cap_initialized & HW_CAP_PMMU)) {
/* As range invalidation does not support zero address we will
* do full invalidation in this case
*/
if (start_va) {
invld_params.range_invalidation = true;
invld_params.start_va = start_va;
invld_params.end_va = end_va;
invld_params.inv_start_val = inv_start_val;
invld_params.flags = flags | MMU_OP_CLEAR_MEMCACHE;
} else {
invld_params.range_invalidation = false;
invld_params.inv_start_val = 1;
invld_params.flags = flags;
}
gaudi2_mmu_invalidate_cache_trigger(hdev, mmPMMU_HBW_STLB_BASE, &invld_params);
rc = gaudi2_mmu_invalidate_cache_status_poll(hdev, mmPMMU_HBW_STLB_BASE,
&invld_params);
if (rc)
return rc;
} else if (flags & MMU_OP_PHYS_PACK) {
invld_params.start_va = gaudi2_mmu_scramble_addr(hdev, start_va);
invld_params.end_va = gaudi2_mmu_scramble_addr(hdev, end_va);
invld_params.inv_start_val = inv_start_val;
invld_params.flags = flags;
rc = gaudi2_hmmus_invalidate_cache(hdev, &invld_params);
}
return rc;
}
static int gaudi2_mmu_update_hop0_addr(struct hl_device *hdev, u32 stlb_base)
{
struct asic_fixed_properties *prop = &hdev->asic_prop;
u64 hop0_addr;
u32 asid, max_asid = prop->max_asid;
int rc;
/* it takes too much time to init all of the ASIDs on palladium */
if (hdev->pldm)
max_asid = min((u32) 8, max_asid);
for (asid = 0 ; asid < max_asid ; asid++) {
hop0_addr = hdev->mmu_priv.hr.mmu_asid_hop0[asid].phys_addr;
rc = gaudi2_mmu_update_asid_hop0_addr(hdev, stlb_base, asid, hop0_addr);
if (rc) {
dev_err(hdev->dev, "failed to set hop0 addr for asid %d\n", asid);
return rc;
}
}
return 0;
}
habanalabs/gaudi2: enable all MMU SPI/SEI interrupts Currently only part of the MMU SPI/SEI interrupts are enabled, although there is no real reason to not enable all. The only exception is "burst_fifo_full" which is expected for PMMU because it has a 2 entries FIFO, and thus is it not enabled for it. Signed-off-by: Tomer Tayar <ttayar@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-07-12 23:19:11 +08:00
static int gaudi2_mmu_init_common(struct hl_device *hdev, u32 mmu_base, u32 stlb_base)
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
{
u32 status, timeout_usec;
int rc;
if (hdev->pldm || !hdev->pdev)
timeout_usec = GAUDI2_PLDM_MMU_TIMEOUT_USEC;
else
timeout_usec = GAUDI2_MMU_CACHE_INV_TIMEOUT_USEC;
WREG32(stlb_base + STLB_INV_ALL_START_OFFSET, 1);
rc = hl_poll_timeout(
hdev,
stlb_base + STLB_SRAM_INIT_OFFSET,
status,
!status,
1000,
timeout_usec);
if (rc)
dev_notice_ratelimited(hdev->dev, "Timeout when waiting for MMU SRAM init\n");
rc = gaudi2_mmu_update_hop0_addr(hdev, stlb_base);
if (rc)
return rc;
WREG32(mmu_base + MMU_BYPASS_OFFSET, 0);
rc = hl_poll_timeout(
hdev,
stlb_base + STLB_INV_ALL_START_OFFSET,
status,
!status,
1000,
timeout_usec);
if (rc)
dev_notice_ratelimited(hdev->dev, "Timeout when waiting for MMU invalidate all\n");
WREG32(mmu_base + MMU_ENABLE_OFFSET, 1);
return rc;
}
static int gaudi2_pci_mmu_init(struct hl_device *hdev)
{
struct gaudi2_device *gaudi2 = hdev->asic_specific;
u32 mmu_base, stlb_base;
int rc;
if (gaudi2->hw_cap_initialized & HW_CAP_PMMU)
return 0;
mmu_base = mmPMMU_HBW_MMU_BASE;
stlb_base = mmPMMU_HBW_STLB_BASE;
RMWREG32(stlb_base + STLB_HOP_CONFIGURATION_OFFSET,
(0 << PMMU_HBW_STLB_HOP_CONFIGURATION_FIRST_HOP_SHIFT) |
(5 << PMMU_HBW_STLB_HOP_CONFIGURATION_FIRST_LOOKUP_HOP_SMALL_P_SHIFT) |
(4 << PMMU_HBW_STLB_HOP_CONFIGURATION_FIRST_LOOKUP_HOP_LARGE_P_SHIFT) |
(5 << PMMU_HBW_STLB_HOP_CONFIGURATION_LAST_HOP_SHIFT) |
(5 << PMMU_HBW_STLB_HOP_CONFIGURATION_FOLLOWER_HOP_SHIFT),
PMMU_HBW_STLB_HOP_CONFIGURATION_FIRST_HOP_MASK |
PMMU_HBW_STLB_HOP_CONFIGURATION_FIRST_LOOKUP_HOP_SMALL_P_MASK |
PMMU_HBW_STLB_HOP_CONFIGURATION_FIRST_LOOKUP_HOP_LARGE_P_MASK |
PMMU_HBW_STLB_HOP_CONFIGURATION_LAST_HOP_MASK |
PMMU_HBW_STLB_HOP_CONFIGURATION_FOLLOWER_HOP_MASK);
WREG32(stlb_base + STLB_LL_LOOKUP_MASK_63_32_OFFSET, 0);
if (PAGE_SIZE == SZ_64K) {
/* Set page sizes to 64K on hop5 and 16M on hop4 + enable 8 bit hops */
RMWREG32(mmu_base + MMU_STATIC_MULTI_PAGE_SIZE_OFFSET,
FIELD_PREP(DCORE0_HMMU0_MMU_STATIC_MULTI_PAGE_SIZE_HOP5_PAGE_SIZE_MASK, 4) |
FIELD_PREP(DCORE0_HMMU0_MMU_STATIC_MULTI_PAGE_SIZE_HOP4_PAGE_SIZE_MASK, 3) |
FIELD_PREP(
DCORE0_HMMU0_MMU_STATIC_MULTI_PAGE_SIZE_CFG_8_BITS_HOP_MODE_EN_MASK,
1),
DCORE0_HMMU0_MMU_STATIC_MULTI_PAGE_SIZE_HOP5_PAGE_SIZE_MASK |
DCORE0_HMMU0_MMU_STATIC_MULTI_PAGE_SIZE_HOP4_PAGE_SIZE_MASK |
DCORE0_HMMU0_MMU_STATIC_MULTI_PAGE_SIZE_CFG_8_BITS_HOP_MODE_EN_MASK);
}
habanalabs/gaudi2: enable all MMU SPI/SEI interrupts Currently only part of the MMU SPI/SEI interrupts are enabled, although there is no real reason to not enable all. The only exception is "burst_fifo_full" which is expected for PMMU because it has a 2 entries FIFO, and thus is it not enabled for it. Signed-off-by: Tomer Tayar <ttayar@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-07-12 23:19:11 +08:00
WREG32(mmu_base + MMU_SPI_SEI_MASK_OFFSET, GAUDI2_PMMU_SPI_SEI_ENABLE_MASK);
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
rc = gaudi2_mmu_init_common(hdev, mmu_base, stlb_base);
if (rc)
return rc;
gaudi2->hw_cap_initialized |= HW_CAP_PMMU;
return 0;
}
static int gaudi2_dcore_hmmu_init(struct hl_device *hdev, int dcore_id,
int hmmu_id)
{
struct asic_fixed_properties *prop = &hdev->asic_prop;
struct gaudi2_device *gaudi2 = hdev->asic_specific;
u32 offset, mmu_base, stlb_base, hw_cap;
u8 dmmu_seq;
int rc;
dmmu_seq = NUM_OF_HMMU_PER_DCORE * dcore_id + hmmu_id;
hw_cap = HW_CAP_DCORE0_DMMU0 << dmmu_seq;
/*
* return if DMMU is already initialized or if it's not out of
* isolation (due to cluster binning)
*/
if ((gaudi2->hw_cap_initialized & hw_cap) || !(prop->hmmu_hif_enabled_mask & BIT(dmmu_seq)))
return 0;
offset = (u32) (dcore_id * DCORE_OFFSET + hmmu_id * DCORE_HMMU_OFFSET);
mmu_base = mmDCORE0_HMMU0_MMU_BASE + offset;
stlb_base = mmDCORE0_HMMU0_STLB_BASE + offset;
RMWREG32(mmu_base + MMU_STATIC_MULTI_PAGE_SIZE_OFFSET, 5 /* 64MB */,
MMU_STATIC_MULTI_PAGE_SIZE_HOP4_PAGE_SIZE_MASK);
RMWREG32(stlb_base + STLB_HOP_CONFIGURATION_OFFSET,
FIELD_PREP(DCORE0_HMMU0_STLB_HOP_CONFIGURATION_FIRST_HOP_MASK, 0) |
FIELD_PREP(DCORE0_HMMU0_STLB_HOP_CONFIGURATION_FIRST_LOOKUP_HOP_SMALL_P_MASK, 3) |
FIELD_PREP(DCORE0_HMMU0_STLB_HOP_CONFIGURATION_FIRST_LOOKUP_HOP_LARGE_P_MASK, 3) |
FIELD_PREP(DCORE0_HMMU0_STLB_HOP_CONFIGURATION_LAST_HOP_MASK, 3) |
FIELD_PREP(DCORE0_HMMU0_STLB_HOP_CONFIGURATION_FOLLOWER_HOP_MASK, 3),
DCORE0_HMMU0_STLB_HOP_CONFIGURATION_FIRST_HOP_MASK |
DCORE0_HMMU0_STLB_HOP_CONFIGURATION_FIRST_LOOKUP_HOP_SMALL_P_MASK |
DCORE0_HMMU0_STLB_HOP_CONFIGURATION_FIRST_LOOKUP_HOP_LARGE_P_MASK |
DCORE0_HMMU0_STLB_HOP_CONFIGURATION_LAST_HOP_MASK |
DCORE0_HMMU0_STLB_HOP_CONFIGURATION_FOLLOWER_HOP_MASK);
RMWREG32(stlb_base + STLB_HOP_CONFIGURATION_OFFSET, 1,
STLB_HOP_CONFIGURATION_ONLY_LARGE_PAGE_MASK);
habanalabs/gaudi2: enable all MMU SPI/SEI interrupts Currently only part of the MMU SPI/SEI interrupts are enabled, although there is no real reason to not enable all. The only exception is "burst_fifo_full" which is expected for PMMU because it has a 2 entries FIFO, and thus is it not enabled for it. Signed-off-by: Tomer Tayar <ttayar@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-07-12 23:19:11 +08:00
WREG32(mmu_base + MMU_SPI_SEI_MASK_OFFSET, GAUDI2_HMMU_SPI_SEI_ENABLE_MASK);
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
rc = gaudi2_mmu_init_common(hdev, mmu_base, stlb_base);
if (rc)
return rc;
gaudi2->hw_cap_initialized |= hw_cap;
return 0;
}
static int gaudi2_hbm_mmu_init(struct hl_device *hdev)
{
int rc, dcore_id, hmmu_id;
for (dcore_id = 0 ; dcore_id < NUM_OF_DCORES ; dcore_id++)
for (hmmu_id = 0 ; hmmu_id < NUM_OF_HMMU_PER_DCORE; hmmu_id++) {
rc = gaudi2_dcore_hmmu_init(hdev, dcore_id, hmmu_id);
if (rc)
return rc;
}
return 0;
}
static int gaudi2_mmu_init(struct hl_device *hdev)
{
int rc;
rc = gaudi2_pci_mmu_init(hdev);
if (rc)
return rc;
rc = gaudi2_hbm_mmu_init(hdev);
if (rc)
return rc;
return 0;
}
static int gaudi2_hw_init(struct hl_device *hdev)
{
struct gaudi2_device *gaudi2 = hdev->asic_specific;
int rc;
/* Let's mark in the H/W that we have reached this point. We check
* this value in the reset_before_init function to understand whether
* we need to reset the chip before doing H/W init. This register is
* cleared by the H/W upon H/W reset
*/
WREG32(mmHW_STATE, HL_DEVICE_HW_STATE_DIRTY);
/* Perform read from the device to make sure device is up */
RREG32(mmHW_STATE);
/* If iATU is done by FW, the HBM bar ALWAYS points to DRAM_PHYS_BASE.
* So we set it here and if anyone tries to move it later to
* a different address, there will be an error
*/
if (hdev->asic_prop.iatu_done_by_fw)
gaudi2->dram_bar_cur_addr = DRAM_PHYS_BASE;
/*
* Before pushing u-boot/linux to device, need to set the hbm bar to
* base address of dram
*/
if (gaudi2_set_hbm_bar_base(hdev, DRAM_PHYS_BASE) == U64_MAX) {
dev_err(hdev->dev, "failed to map HBM bar to DRAM base address\n");
return -EIO;
}
rc = gaudi2_init_cpu(hdev);
if (rc) {
dev_err(hdev->dev, "failed to initialize CPU\n");
return rc;
}
gaudi2_init_msix_gw_table(hdev);
gaudi2_init_scrambler_hbm(hdev);
gaudi2_init_kdma(hdev);
rc = gaudi2_init_cpu_queues(hdev, GAUDI2_CPU_TIMEOUT_USEC);
if (rc) {
dev_err(hdev->dev, "failed to initialize CPU H/W queues %d\n", rc);
return rc;
}
rc = gaudi2->cpucp_info_get(hdev);
if (rc) {
dev_err(hdev->dev, "Failed to get cpucp info\n");
return rc;
}
rc = gaudi2_mmu_init(hdev);
if (rc)
return rc;
gaudi2_init_pdma(hdev);
gaudi2_init_edma(hdev);
gaudi2_init_sm(hdev);
gaudi2_init_tpc(hdev);
gaudi2_init_mme(hdev);
gaudi2_init_rotator(hdev);
gaudi2_init_dec(hdev);
gaudi2_enable_timestamp(hdev);
habanalabs/gaudi2: add gaudi2 profiler module Add the Gaudi2 code to initialize the ASIC's profiler. The profile receives its initialization values from the user, same as in Gaudi2, but the code to initialize is in the driver because the configuration space of the device is not directly exposed to the user. Signed-off-by: Benjamin Dotan <bdotan@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-27 02:35:07 +08:00
rc = gaudi2_coresight_init(hdev);
if (rc)
goto disable_queues;
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
rc = gaudi2_enable_msix(hdev);
if (rc)
goto disable_queues;
/* Perform read from the device to flush all configuration */
RREG32(mmHW_STATE);
return 0;
disable_queues:
gaudi2_disable_dma_qmans(hdev);
gaudi2_disable_mme_qmans(hdev);
gaudi2_disable_tpc_qmans(hdev);
gaudi2_disable_rot_qmans(hdev);
gaudi2_disable_nic_qmans(hdev);
gaudi2_disable_timestamp(hdev);
return rc;
}
/**
* gaudi2_send_hard_reset_cmd - common function to handle reset
*
* @hdev: pointer to the habanalabs device structure
*
* This function handles the various possible scenarios for reset.
* It considers if reset is handled by driver\FW and what FW components are loaded
*/
static void gaudi2_send_hard_reset_cmd(struct hl_device *hdev)
{
struct cpu_dyn_regs *dyn_regs = &hdev->fw_loader.dynamic_loader.comm_desc.cpu_dyn_regs;
bool heartbeat_reset, preboot_only, cpu_initialized = false;
struct gaudi2_device *gaudi2 = hdev->asic_specific;
u32 cpu_boot_status;
preboot_only = (hdev->fw_loader.fw_comp_loaded == FW_TYPE_PREBOOT_CPU);
heartbeat_reset = (hdev->reset_info.curr_reset_cause == HL_RESET_CAUSE_HEARTBEAT);
/*
* Handle corner case where failure was at cpu management app load,
* and driver didn't detect any failure while loading the FW,
* then at such scenario driver will send only HALT_MACHINE
* and no one will respond to this request since FW already back to preboot
* and it cannot handle such cmd.
* In this case next time the management app loads it'll check on events register
* which will still have the halt indication, and will reboot the device.
* The solution is to let preboot clear all relevant registers before next boot
* once driver send COMMS_RST_DEV.
*/
cpu_boot_status = RREG32(mmPSOC_GLOBAL_CONF_CPU_BOOT_STATUS);
if (gaudi2 && (gaudi2->hw_cap_initialized & HW_CAP_CPU) &&
(cpu_boot_status == CPU_BOOT_STATUS_SRAM_AVAIL))
cpu_initialized = true;
/*
* when Linux/Bootfit exist this write to the SP can be interpreted in 2 ways:
* 1. FW reset: FW initiate the reset sequence
* 2. driver reset: FW will start HALT sequence (the preparations for the
* reset but not the reset itself as it is not implemented
* on their part) and LKD will wait to let FW complete the
* sequence before issuing the reset
*/
if (!preboot_only && cpu_initialized) {
WREG32(le32_to_cpu(dyn_regs->gic_host_halt_irq),
gaudi2_irq_map_table[GAUDI2_EVENT_CPU_HALT_MACHINE].cpu_id);
msleep(GAUDI2_CPU_RESET_WAIT_MSEC);
}
/*
* When working with preboot (without Linux/Boot fit) we can
* communicate only using the COMMS commands to issue halt/reset.
*
* For the case in which we are working with Linux/Bootfit this is a hail-mary
* attempt to revive the card in the small chance that the f/w has
* experienced a watchdog event, which caused it to return back to preboot.
* In that case, triggering reset through GIC won't help. We need to
* trigger the reset as if Linux wasn't loaded.
*
* We do it only if the reset cause was HB, because that would be the
* indication of such an event.
*
* In case watchdog hasn't expired but we still got HB, then this won't
* do any damage.
*/
if (heartbeat_reset || preboot_only || !cpu_initialized) {
if (hdev->asic_prop.hard_reset_done_by_fw)
hl_fw_ask_hard_reset_without_linux(hdev);
else
hl_fw_ask_halt_machine_without_linux(hdev);
}
}
/**
* gaudi2_execute_hard_reset - execute hard reset by driver/FW
*
* @hdev: pointer to the habanalabs device structure
* @reset_sleep_ms: sleep time in msec after reset
*
* This function executes hard reset based on if driver/FW should do the reset
*/
static void gaudi2_execute_hard_reset(struct hl_device *hdev, u32 reset_sleep_ms)
{
if (hdev->asic_prop.hard_reset_done_by_fw) {
gaudi2_send_hard_reset_cmd(hdev);
return;
}
/* Set device to handle FLR by H/W as we will put the device
* CPU to halt mode
*/
WREG32(mmPCIE_AUX_FLR_CTRL,
(PCIE_AUX_FLR_CTRL_HW_CTRL_MASK | PCIE_AUX_FLR_CTRL_INT_MASK_MASK));
gaudi2_send_hard_reset_cmd(hdev);
WREG32(mmPSOC_RESET_CONF_SW_ALL_RST, 1);
}
/**
* gaudi2_execute_soft_reset - execute soft reset by driver/FW
*
* @hdev: pointer to the habanalabs device structure
* @reset_sleep_ms: sleep time in msec after reset
* @driver_performs_reset: true if driver should perform reset instead of f/w.
*
* This function executes soft reset based on if driver/FW should do the reset
*/
static void gaudi2_execute_soft_reset(struct hl_device *hdev, u32 reset_sleep_ms,
bool driver_performs_reset)
{
struct cpu_dyn_regs *dyn_regs = &hdev->fw_loader.dynamic_loader.comm_desc.cpu_dyn_regs;
if (!driver_performs_reset) {
/* set SP to indicate reset request sent to FW */
habanalabs/gaudi2: get f/w reset status register dynamically Get the firmware reset status address from the dynamic registers we read from the firmware instead of using a define. Signed-off-by: farah kassabri <fkassabri@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-08-29 21:56:28 +08:00
if (dyn_regs->cpu_rst_status)
WREG32(le32_to_cpu(dyn_regs->cpu_rst_status), CPU_RST_STATUS_NA);
else
WREG32(mmCPU_RST_STATUS_TO_HOST, CPU_RST_STATUS_NA);
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
WREG32(le32_to_cpu(dyn_regs->gic_host_soft_rst_irq),
gaudi2_irq_map_table[GAUDI2_EVENT_CPU_SOFT_RESET].cpu_id);
return;
}
habanalabs/gaudi2: add gaudi2 security module Use the generic security module to block all registers in the ASIC and then open only those that are needed to be accessed by the user. Signed-off-by: Ofir Bitton <obitton@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-27 02:30:25 +08:00
/* Block access to engines, QMANs and SM during reset, these
* RRs will be reconfigured after soft reset.
* PCIE_MSIX is left unsecured to allow NIC packets processing during the reset.
*/
gaudi2_write_rr_to_all_lbw_rtrs(hdev, RR_TYPE_LONG, NUM_LONG_LBW_RR - 1,
mmDCORE0_TPC0_QM_DCCM_BASE, mmPCIE_MSIX_BASE);
gaudi2_write_rr_to_all_lbw_rtrs(hdev, RR_TYPE_LONG, NUM_LONG_LBW_RR - 2,
mmPCIE_MSIX_BASE + HL_BLOCK_SIZE,
mmPCIE_VDEC1_MSTR_IF_RR_SHRD_HBW_BASE + HL_BLOCK_SIZE);
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
WREG32(mmPSOC_RESET_CONF_SOFT_RST, 1);
}
static void gaudi2_poll_btm_indication(struct hl_device *hdev, u32 reset_sleep_ms,
u32 poll_timeout_us)
{
int i, rc = 0;
u32 reg_val;
/* without this sleep reset will not work */
msleep(reset_sleep_ms);
/* We poll the BTM done indication multiple times after reset due to
* a HW errata 'GAUDI2_0300'
*/
for (i = 0 ; i < GAUDI2_RESET_POLL_CNT ; i++)
rc = hl_poll_timeout(
hdev,
mmPSOC_GLOBAL_CONF_BTM_FSM,
reg_val,
reg_val == 0,
1000,
poll_timeout_us);
if (rc)
dev_err(hdev->dev, "Timeout while waiting for device to reset 0x%x\n", reg_val);
}
static void gaudi2_get_soft_rst_done_indication(struct hl_device *hdev, u32 poll_timeout_us)
{
int i, rc = 0;
u32 reg_val;
for (i = 0 ; i < GAUDI2_RESET_POLL_CNT ; i++)
rc = hl_poll_timeout(
hdev,
mmCPU_RST_STATUS_TO_HOST,
reg_val,
reg_val == CPU_RST_STATUS_SOFT_RST_DONE,
1000,
poll_timeout_us);
if (rc)
dev_err(hdev->dev, "Timeout while waiting for FW to complete soft reset (0x%x)\n",
reg_val);
}
static void gaudi2_hw_fini(struct hl_device *hdev, bool hard_reset, bool fw_reset)
{
struct gaudi2_device *gaudi2 = hdev->asic_specific;
u32 poll_timeout_us, reset_sleep_ms;
bool driver_performs_reset = false;
if (hdev->pldm) {
reset_sleep_ms = hard_reset ? GAUDI2_PLDM_HRESET_TIMEOUT_MSEC :
GAUDI2_PLDM_SRESET_TIMEOUT_MSEC;
poll_timeout_us = GAUDI2_PLDM_RESET_POLL_TIMEOUT_USEC;
} else {
reset_sleep_ms = GAUDI2_RESET_TIMEOUT_MSEC;
poll_timeout_us = GAUDI2_RESET_POLL_TIMEOUT_USEC;
}
if (fw_reset)
goto skip_reset;
gaudi2_reset_arcs(hdev);
if (hard_reset) {
driver_performs_reset = !hdev->asic_prop.hard_reset_done_by_fw;
gaudi2_execute_hard_reset(hdev, reset_sleep_ms);
} else {
/*
* As we have to support also work with preboot only (which does not supports
* soft reset) we have to make sure that security is disabled before letting driver
* do the reset. user shall control the BFE flags to avoid asking soft reset in
* secured device with preboot only.
*/
driver_performs_reset = (hdev->fw_components == FW_TYPE_PREBOOT_CPU &&
!hdev->asic_prop.fw_security_enabled);
gaudi2_execute_soft_reset(hdev, reset_sleep_ms, driver_performs_reset);
}
skip_reset:
if (driver_performs_reset || hard_reset)
gaudi2_poll_btm_indication(hdev, reset_sleep_ms, poll_timeout_us);
else
gaudi2_get_soft_rst_done_indication(hdev, poll_timeout_us);
if (!gaudi2)
return;
gaudi2->dec_hw_cap_initialized &= ~(HW_CAP_DEC_MASK);
gaudi2->tpc_hw_cap_initialized &= ~(HW_CAP_TPC_MASK);
/*
* Clear NIC capability mask in order for driver to re-configure
* NIC QMANs. NIC ports will not be re-configured during soft
* reset as we call gaudi2_nic_init only during hard reset
*/
gaudi2->nic_hw_cap_initialized &= ~(HW_CAP_NIC_MASK);
if (hard_reset) {
gaudi2->hw_cap_initialized &=
~(HW_CAP_DRAM | HW_CAP_CLK_GATE | HW_CAP_HBM_SCRAMBLER_MASK |
HW_CAP_PMMU | HW_CAP_CPU | HW_CAP_CPU_Q |
HW_CAP_SRAM_SCRAMBLER | HW_CAP_DMMU_MASK |
HW_CAP_PDMA_MASK | HW_CAP_EDMA_MASK | HW_CAP_KDMA |
HW_CAP_MME_MASK | HW_CAP_ROT_MASK);
memset(gaudi2->events_stat, 0, sizeof(gaudi2->events_stat));
} else {
gaudi2->hw_cap_initialized &=
~(HW_CAP_CLK_GATE | HW_CAP_HBM_SCRAMBLER_SW_RESET |
HW_CAP_PDMA_MASK | HW_CAP_EDMA_MASK | HW_CAP_MME_MASK |
HW_CAP_ROT_MASK);
}
}
static int gaudi2_suspend(struct hl_device *hdev)
{
int rc;
habanalabs: add a value field to hl_fw_send_pci_access_msg() For gaudi2 we need to send a value to F/W as part of the PCI_ACCESS packet. As a preparation, modify hl_fw_send_pci_access_msg() to have a 'value' field. Signed-off-by: Tomer Tayar <ttayar@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-07-06 15:20:54 +08:00
rc = hl_fw_send_pci_access_msg(hdev, CPUCP_PACKET_DISABLE_PCI_ACCESS, 0x0);
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
if (rc)
dev_err(hdev->dev, "Failed to disable PCI access from CPU\n");
return rc;
}
static int gaudi2_resume(struct hl_device *hdev)
{
return gaudi2_init_iatu(hdev);
}
static int gaudi2_mmap(struct hl_device *hdev, struct vm_area_struct *vma,
void *cpu_addr, dma_addr_t dma_addr, size_t size)
{
int rc;
vma->vm_flags |= VM_IO | VM_PFNMAP | VM_DONTEXPAND | VM_DONTDUMP |
VM_DONTCOPY | VM_NORESERVE;
#ifdef _HAS_DMA_MMAP_COHERENT
rc = dma_mmap_coherent(hdev->dev, vma, cpu_addr, dma_addr, size);
if (rc)
dev_err(hdev->dev, "dma_mmap_coherent error %d", rc);
#else
rc = remap_pfn_range(vma, vma->vm_start,
virt_to_phys(cpu_addr) >> PAGE_SHIFT,
size, vma->vm_page_prot);
if (rc)
dev_err(hdev->dev, "remap_pfn_range error %d", rc);
#endif
return rc;
}
static bool gaudi2_is_queue_enabled(struct hl_device *hdev, u32 hw_queue_id)
{
struct gaudi2_device *gaudi2 = hdev->asic_specific;
u64 hw_cap_mask = 0;
u64 hw_tpc_cap_bit = 0;
u64 hw_nic_cap_bit = 0;
u64 hw_test_cap_bit = 0;
switch (hw_queue_id) {
habanalabs/gaudi2: assigning PQFs for ARC f/w in PDMA Assigning 3 PQFs in PDMA1 and 2 PQFs in PDMA0 for ARC firmware usage. Signed-off-by: Rajarama Manjukody Bhat <rmbhat@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-08-12 14:28:20 +08:00
case GAUDI2_QUEUE_ID_PDMA_0_0:
case GAUDI2_QUEUE_ID_PDMA_0_1:
case GAUDI2_QUEUE_ID_PDMA_1_0:
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
hw_cap_mask = HW_CAP_PDMA_MASK;
break;
case GAUDI2_QUEUE_ID_DCORE0_EDMA_0_0...GAUDI2_QUEUE_ID_DCORE0_EDMA_1_3:
hw_test_cap_bit = HW_CAP_EDMA_SHIFT +
((hw_queue_id - GAUDI2_QUEUE_ID_DCORE0_EDMA_0_0) >> 2);
break;
case GAUDI2_QUEUE_ID_DCORE1_EDMA_0_0...GAUDI2_QUEUE_ID_DCORE1_EDMA_1_3:
hw_test_cap_bit = HW_CAP_EDMA_SHIFT + NUM_OF_EDMA_PER_DCORE +
((hw_queue_id - GAUDI2_QUEUE_ID_DCORE1_EDMA_0_0) >> 2);
break;
case GAUDI2_QUEUE_ID_DCORE2_EDMA_0_0...GAUDI2_QUEUE_ID_DCORE2_EDMA_1_3:
hw_test_cap_bit = HW_CAP_EDMA_SHIFT + 2 * NUM_OF_EDMA_PER_DCORE +
((hw_queue_id - GAUDI2_QUEUE_ID_DCORE2_EDMA_0_0) >> 2);
break;
case GAUDI2_QUEUE_ID_DCORE3_EDMA_0_0...GAUDI2_QUEUE_ID_DCORE3_EDMA_1_3:
hw_test_cap_bit = HW_CAP_EDMA_SHIFT + 3 * NUM_OF_EDMA_PER_DCORE +
((hw_queue_id - GAUDI2_QUEUE_ID_DCORE3_EDMA_0_0) >> 2);
break;
case GAUDI2_QUEUE_ID_DCORE0_MME_0_0 ... GAUDI2_QUEUE_ID_DCORE0_MME_0_3:
hw_test_cap_bit = HW_CAP_MME_SHIFT;
break;
case GAUDI2_QUEUE_ID_DCORE1_MME_0_0 ... GAUDI2_QUEUE_ID_DCORE1_MME_0_3:
hw_test_cap_bit = HW_CAP_MME_SHIFT + 1;
break;
case GAUDI2_QUEUE_ID_DCORE2_MME_0_0 ... GAUDI2_QUEUE_ID_DCORE2_MME_0_3:
hw_test_cap_bit = HW_CAP_MME_SHIFT + 2;
break;
case GAUDI2_QUEUE_ID_DCORE3_MME_0_0 ... GAUDI2_QUEUE_ID_DCORE3_MME_0_3:
hw_test_cap_bit = HW_CAP_MME_SHIFT + 3;
break;
case GAUDI2_QUEUE_ID_DCORE0_TPC_0_0 ... GAUDI2_QUEUE_ID_DCORE0_TPC_5_3:
hw_tpc_cap_bit = HW_CAP_TPC_SHIFT +
((hw_queue_id - GAUDI2_QUEUE_ID_DCORE0_TPC_0_0) >> 2);
/* special case where cap bit refers to the first queue id */
if (!hw_tpc_cap_bit)
return !!(gaudi2->tpc_hw_cap_initialized & BIT_ULL(0));
break;
case GAUDI2_QUEUE_ID_DCORE1_TPC_0_0 ... GAUDI2_QUEUE_ID_DCORE1_TPC_5_3:
hw_tpc_cap_bit = HW_CAP_TPC_SHIFT + NUM_OF_TPC_PER_DCORE +
((hw_queue_id - GAUDI2_QUEUE_ID_DCORE1_TPC_0_0) >> 2);
break;
case GAUDI2_QUEUE_ID_DCORE2_TPC_0_0 ... GAUDI2_QUEUE_ID_DCORE2_TPC_5_3:
hw_tpc_cap_bit = HW_CAP_TPC_SHIFT + (2 * NUM_OF_TPC_PER_DCORE) +
((hw_queue_id - GAUDI2_QUEUE_ID_DCORE2_TPC_0_0) >> 2);
break;
case GAUDI2_QUEUE_ID_DCORE3_TPC_0_0 ... GAUDI2_QUEUE_ID_DCORE3_TPC_5_3:
hw_tpc_cap_bit = HW_CAP_TPC_SHIFT + (3 * NUM_OF_TPC_PER_DCORE) +
((hw_queue_id - GAUDI2_QUEUE_ID_DCORE3_TPC_0_0) >> 2);
break;
case GAUDI2_QUEUE_ID_DCORE0_TPC_6_0 ... GAUDI2_QUEUE_ID_DCORE0_TPC_6_3:
hw_tpc_cap_bit = HW_CAP_TPC_SHIFT + (4 * NUM_OF_TPC_PER_DCORE);
break;
case GAUDI2_QUEUE_ID_ROT_0_0 ... GAUDI2_QUEUE_ID_ROT_1_3:
hw_test_cap_bit = HW_CAP_ROT_SHIFT + ((hw_queue_id - GAUDI2_QUEUE_ID_ROT_0_0) >> 2);
break;
case GAUDI2_QUEUE_ID_NIC_0_0 ... GAUDI2_QUEUE_ID_NIC_23_3:
hw_nic_cap_bit = HW_CAP_NIC_SHIFT + ((hw_queue_id - GAUDI2_QUEUE_ID_NIC_0_0) >> 2);
/* special case where cap bit refers to the first queue id */
if (!hw_nic_cap_bit)
return !!(gaudi2->nic_hw_cap_initialized & BIT_ULL(0));
break;
case GAUDI2_QUEUE_ID_CPU_PQ:
return !!(gaudi2->hw_cap_initialized & HW_CAP_CPU_Q);
default:
return false;
}
if (hw_tpc_cap_bit)
return !!(gaudi2->tpc_hw_cap_initialized & BIT_ULL(hw_tpc_cap_bit));
if (hw_nic_cap_bit)
return !!(gaudi2->nic_hw_cap_initialized & BIT_ULL(hw_nic_cap_bit));
if (hw_test_cap_bit)
hw_cap_mask = BIT_ULL(hw_test_cap_bit);
return !!(gaudi2->hw_cap_initialized & hw_cap_mask);
}
static bool gaudi2_is_arc_enabled(struct hl_device *hdev, u64 arc_id)
{
struct gaudi2_device *gaudi2 = hdev->asic_specific;
switch (arc_id) {
case CPU_ID_SCHED_ARC0 ... CPU_ID_SCHED_ARC5:
case CPU_ID_MME_QMAN_ARC0...CPU_ID_ROT_QMAN_ARC1:
return !!(gaudi2->active_hw_arc & BIT_ULL(arc_id));
case CPU_ID_TPC_QMAN_ARC0...CPU_ID_TPC_QMAN_ARC24:
return !!(gaudi2->active_tpc_arc & BIT_ULL(arc_id - CPU_ID_TPC_QMAN_ARC0));
case CPU_ID_NIC_QMAN_ARC0...CPU_ID_NIC_QMAN_ARC23:
return !!(gaudi2->active_nic_arc & BIT_ULL(arc_id - CPU_ID_NIC_QMAN_ARC0));
default:
return false;
}
}
static void gaudi2_clr_arc_id_cap(struct hl_device *hdev, u64 arc_id)
{
struct gaudi2_device *gaudi2 = hdev->asic_specific;
switch (arc_id) {
case CPU_ID_SCHED_ARC0 ... CPU_ID_SCHED_ARC5:
case CPU_ID_MME_QMAN_ARC0...CPU_ID_ROT_QMAN_ARC1:
gaudi2->active_hw_arc &= ~(BIT_ULL(arc_id));
break;
case CPU_ID_TPC_QMAN_ARC0...CPU_ID_TPC_QMAN_ARC24:
gaudi2->active_tpc_arc &= ~(BIT_ULL(arc_id - CPU_ID_TPC_QMAN_ARC0));
break;
case CPU_ID_NIC_QMAN_ARC0...CPU_ID_NIC_QMAN_ARC23:
gaudi2->active_nic_arc &= ~(BIT_ULL(arc_id - CPU_ID_NIC_QMAN_ARC0));
break;
default:
return;
}
}
static void gaudi2_set_arc_id_cap(struct hl_device *hdev, u64 arc_id)
{
struct gaudi2_device *gaudi2 = hdev->asic_specific;
switch (arc_id) {
case CPU_ID_SCHED_ARC0 ... CPU_ID_SCHED_ARC5:
case CPU_ID_MME_QMAN_ARC0...CPU_ID_ROT_QMAN_ARC1:
gaudi2->active_hw_arc |= BIT_ULL(arc_id);
break;
case CPU_ID_TPC_QMAN_ARC0...CPU_ID_TPC_QMAN_ARC24:
gaudi2->active_tpc_arc |= BIT_ULL(arc_id - CPU_ID_TPC_QMAN_ARC0);
break;
case CPU_ID_NIC_QMAN_ARC0...CPU_ID_NIC_QMAN_ARC23:
gaudi2->active_nic_arc |= BIT_ULL(arc_id - CPU_ID_NIC_QMAN_ARC0);
break;
default:
return;
}
}
static void gaudi2_ring_doorbell(struct hl_device *hdev, u32 hw_queue_id, u32 pi)
{
struct cpu_dyn_regs *dyn_regs = &hdev->fw_loader.dynamic_loader.comm_desc.cpu_dyn_regs;
u32 pq_offset, reg_base, db_reg_offset, db_value;
if (hw_queue_id != GAUDI2_QUEUE_ID_CPU_PQ) {
/*
* QMAN has 4 successive PQ_PI registers, 1 for each of the QMAN PQs.
* Masking the H/W queue ID with 0x3 extracts the QMAN internal PQ
* number.
*/
pq_offset = (hw_queue_id & 0x3) * 4;
reg_base = gaudi2_qm_blocks_bases[hw_queue_id];
db_reg_offset = reg_base + QM_PQ_PI_0_OFFSET + pq_offset;
} else {
db_reg_offset = mmCPU_IF_PF_PQ_PI;
}
db_value = pi;
/* ring the doorbell */
WREG32(db_reg_offset, db_value);
if (hw_queue_id == GAUDI2_QUEUE_ID_CPU_PQ) {
/* make sure device CPU will read latest data from host */
mb();
WREG32(le32_to_cpu(dyn_regs->gic_host_pi_upd_irq),
gaudi2_irq_map_table[GAUDI2_EVENT_CPU_PI_UPDATE].cpu_id);
}
}
static void gaudi2_pqe_write(struct hl_device *hdev, __le64 *pqe, struct hl_bd *bd)
{
__le64 *pbd = (__le64 *) bd;
/* The QMANs are on the host memory so a simple copy suffice */
pqe[0] = pbd[0];
pqe[1] = pbd[1];
}
static void *gaudi2_dma_alloc_coherent(struct hl_device *hdev, size_t size,
dma_addr_t *dma_handle, gfp_t flags)
{
return dma_alloc_coherent(&hdev->pdev->dev, size, dma_handle, flags);
}
static void gaudi2_dma_free_coherent(struct hl_device *hdev, size_t size,
void *cpu_addr, dma_addr_t dma_handle)
{
dma_free_coherent(&hdev->pdev->dev, size, cpu_addr, dma_handle);
}
static int gaudi2_send_cpu_message(struct hl_device *hdev, u32 *msg, u16 len,
u32 timeout, u64 *result)
{
struct gaudi2_device *gaudi2 = hdev->asic_specific;
if (!(gaudi2->hw_cap_initialized & HW_CAP_CPU_Q)) {
if (result)
*result = 0;
return 0;
}
if (!timeout)
timeout = GAUDI2_MSG_TO_CPU_TIMEOUT_USEC;
return hl_fw_send_cpu_message(hdev, GAUDI2_QUEUE_ID_CPU_PQ, msg, len, timeout, result);
}
static void *gaudi2_dma_pool_zalloc(struct hl_device *hdev, size_t size,
gfp_t mem_flags, dma_addr_t *dma_handle)
{
if (size > GAUDI2_DMA_POOL_BLK_SIZE)
return NULL;
return dma_pool_zalloc(hdev->dma_pool, mem_flags, dma_handle);
}
static void gaudi2_dma_pool_free(struct hl_device *hdev, void *vaddr, dma_addr_t dma_addr)
{
dma_pool_free(hdev->dma_pool, vaddr, dma_addr);
}
static void *gaudi2_cpu_accessible_dma_pool_alloc(struct hl_device *hdev, size_t size,
dma_addr_t *dma_handle)
{
return hl_fw_cpu_accessible_dma_pool_alloc(hdev, size, dma_handle);
}
static void gaudi2_cpu_accessible_dma_pool_free(struct hl_device *hdev, size_t size, void *vaddr)
{
hl_fw_cpu_accessible_dma_pool_free(hdev, size, vaddr);
}
static dma_addr_t gaudi2_dma_map_single(struct hl_device *hdev, void *addr, int len,
enum dma_data_direction dir)
{
dma_addr_t dma_addr;
dma_addr = dma_map_single(&hdev->pdev->dev, addr, len, dir);
if (unlikely(dma_mapping_error(&hdev->pdev->dev, dma_addr)))
return 0;
return dma_addr;
}
static void gaudi2_dma_unmap_single(struct hl_device *hdev, dma_addr_t addr, int len,
enum dma_data_direction dir)
{
dma_unmap_single(&hdev->pdev->dev, addr, len, dir);
}
static int gaudi2_validate_cb_address(struct hl_device *hdev, struct hl_cs_parser *parser)
{
struct asic_fixed_properties *asic_prop = &hdev->asic_prop;
struct gaudi2_device *gaudi2 = hdev->asic_specific;
if (!gaudi2_is_queue_enabled(hdev, parser->hw_queue_id)) {
dev_err(hdev->dev, "h/w queue %d is disabled\n", parser->hw_queue_id);
return -EINVAL;
}
/* Just check if CB address is valid */
if (hl_mem_area_inside_range((u64) (uintptr_t) parser->user_cb,
parser->user_cb_size,
asic_prop->sram_user_base_address,
asic_prop->sram_end_address))
return 0;
if (hl_mem_area_inside_range((u64) (uintptr_t) parser->user_cb,
parser->user_cb_size,
asic_prop->dram_user_base_address,
asic_prop->dram_end_address))
return 0;
if ((gaudi2->hw_cap_initialized & HW_CAP_DMMU_MASK) &&
hl_mem_area_inside_range((u64) (uintptr_t) parser->user_cb,
parser->user_cb_size,
asic_prop->dmmu.start_addr,
asic_prop->dmmu.end_addr))
return 0;
if (gaudi2->hw_cap_initialized & HW_CAP_PMMU) {
if (hl_mem_area_inside_range((u64) (uintptr_t) parser->user_cb,
parser->user_cb_size,
asic_prop->pmmu.start_addr,
asic_prop->pmmu.end_addr) ||
hl_mem_area_inside_range(
(u64) (uintptr_t) parser->user_cb,
parser->user_cb_size,
asic_prop->pmmu_huge.start_addr,
asic_prop->pmmu_huge.end_addr))
return 0;
} else if (gaudi2_host_phys_addr_valid((u64) (uintptr_t) parser->user_cb)) {
if (!hdev->pdev)
return 0;
if (!device_iommu_mapped(&hdev->pdev->dev))
return 0;
}
dev_err(hdev->dev, "CB address %p + 0x%x for internal QMAN is not valid\n",
parser->user_cb, parser->user_cb_size);
return -EFAULT;
}
static int gaudi2_cs_parser(struct hl_device *hdev, struct hl_cs_parser *parser)
{
struct gaudi2_device *gaudi2 = hdev->asic_specific;
if (!parser->is_kernel_allocated_cb)
return gaudi2_validate_cb_address(hdev, parser);
if (!(gaudi2->hw_cap_initialized & HW_CAP_PMMU)) {
dev_err(hdev->dev, "PMMU not initialized - Unsupported mode in Gaudi2\n");
return -EINVAL;
}
return 0;
}
static int gaudi2_send_heartbeat(struct hl_device *hdev)
{
struct gaudi2_device *gaudi2 = hdev->asic_specific;
if (!(gaudi2->hw_cap_initialized & HW_CAP_CPU_Q))
return 0;
return hl_fw_send_heartbeat(hdev);
}
/* This is an internal helper function, used to update the KDMA mmu props.
* Should be called with a proper kdma lock.
*/
static void gaudi2_kdma_set_mmbp_asid(struct hl_device *hdev,
bool mmu_bypass, u32 asid)
{
u32 rw_asid, rw_mmu_bp;
rw_asid = (asid << ARC_FARM_KDMA_CTX_AXUSER_HB_ASID_RD_SHIFT) |
(asid << ARC_FARM_KDMA_CTX_AXUSER_HB_ASID_WR_SHIFT);
rw_mmu_bp = (!!mmu_bypass << ARC_FARM_KDMA_CTX_AXUSER_HB_MMU_BP_RD_SHIFT) |
(!!mmu_bypass << ARC_FARM_KDMA_CTX_AXUSER_HB_MMU_BP_WR_SHIFT);
WREG32(mmARC_FARM_KDMA_CTX_AXUSER_HB_ASID, rw_asid);
WREG32(mmARC_FARM_KDMA_CTX_AXUSER_HB_MMU_BP, rw_mmu_bp);
}
habanalabs/gaudi2: replace defines for reserved sob/mob with enums Following patches are going to add more reserved sync objects and monitors. To make the counting of these reserved resources simpler, replace the existing RESERVED_* defines with enumerations. Signed-off-by: Tomer Tayar <ttayar@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-30 00:20:38 +08:00
static void gaudi2_arm_cq_monitor(struct hl_device *hdev, u32 sob_id, u32 mon_id, u32 cq_id,
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
u32 mon_payload, u32 sync_value)
{
habanalabs/gaudi2: replace defines for reserved sob/mob with enums Following patches are going to add more reserved sync objects and monitors. To make the counting of these reserved resources simpler, replace the existing RESERVED_* defines with enumerations. Signed-off-by: Tomer Tayar <ttayar@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-30 00:20:38 +08:00
u32 sob_offset, mon_offset, sync_group_id, mode, mon_arm;
habanalabs/gaudi2: SM mask can only be 8-bit Otherwise, due to how we calculate it, we might fail in FIELD_PREP checks. Reported-by: kernel test robot <lkp@intel.com> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-30 16:53:33 +08:00
u8 mask;
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
habanalabs/gaudi2: replace defines for reserved sob/mob with enums Following patches are going to add more reserved sync objects and monitors. To make the counting of these reserved resources simpler, replace the existing RESERVED_* defines with enumerations. Signed-off-by: Tomer Tayar <ttayar@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-30 00:20:38 +08:00
sob_offset = sob_id * 4;
mon_offset = mon_id * 4;
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
/* Reset the SOB value */
habanalabs/gaudi2: replace defines for reserved sob/mob with enums Following patches are going to add more reserved sync objects and monitors. To make the counting of these reserved resources simpler, replace the existing RESERVED_* defines with enumerations. Signed-off-by: Tomer Tayar <ttayar@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-30 00:20:38 +08:00
WREG32(mmDCORE0_SYNC_MNGR_OBJS_SOB_OBJ_0 + sob_offset, 0);
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
/* Configure this address with CQ_ID 0 because CQ_EN is set */
habanalabs/gaudi2: replace defines for reserved sob/mob with enums Following patches are going to add more reserved sync objects and monitors. To make the counting of these reserved resources simpler, replace the existing RESERVED_* defines with enumerations. Signed-off-by: Tomer Tayar <ttayar@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-30 00:20:38 +08:00
WREG32(mmDCORE0_SYNC_MNGR_OBJS_MON_PAY_ADDRL_0 + mon_offset, cq_id);
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
/* Configure this address with CS index because CQ_EN is set */
habanalabs/gaudi2: replace defines for reserved sob/mob with enums Following patches are going to add more reserved sync objects and monitors. To make the counting of these reserved resources simpler, replace the existing RESERVED_* defines with enumerations. Signed-off-by: Tomer Tayar <ttayar@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-30 00:20:38 +08:00
WREG32(mmDCORE0_SYNC_MNGR_OBJS_MON_PAY_DATA_0 + mon_offset, mon_payload);
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
habanalabs/gaudi2: replace defines for reserved sob/mob with enums Following patches are going to add more reserved sync objects and monitors. To make the counting of these reserved resources simpler, replace the existing RESERVED_* defines with enumerations. Signed-off-by: Tomer Tayar <ttayar@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-30 00:20:38 +08:00
sync_group_id = sob_id / 8;
mask = ~(1 << (sob_id & 0x7));
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
mode = 1; /* comparison mode is "equal to" */
mon_arm = FIELD_PREP(DCORE0_SYNC_MNGR_OBJS_MON_ARM_SOD_MASK, sync_value);
mon_arm |= FIELD_PREP(DCORE0_SYNC_MNGR_OBJS_MON_ARM_SOP_MASK, mode);
mon_arm |= FIELD_PREP(DCORE0_SYNC_MNGR_OBJS_MON_ARM_MASK_MASK, mask);
mon_arm |= FIELD_PREP(DCORE0_SYNC_MNGR_OBJS_MON_ARM_SID_MASK, sync_group_id);
habanalabs/gaudi2: replace defines for reserved sob/mob with enums Following patches are going to add more reserved sync objects and monitors. To make the counting of these reserved resources simpler, replace the existing RESERVED_* defines with enumerations. Signed-off-by: Tomer Tayar <ttayar@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-30 00:20:38 +08:00
WREG32(mmDCORE0_SYNC_MNGR_OBJS_MON_ARM_0 + mon_offset, mon_arm);
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
}
/* This is an internal helper function used by gaudi2_send_job_to_kdma only */
static int gaudi2_send_job_to_kdma(struct hl_device *hdev,
u64 src_addr, u64 dst_addr,
u32 size, bool is_memset)
{
u32 comp_val, commit_mask, *polling_addr, timeout, status = 0;
struct hl_cq_entry *cq_base;
struct hl_cq *cq;
u64 comp_addr;
int rc;
habanalabs/gaudi2: replace defines for reserved sob/mob with enums Following patches are going to add more reserved sync objects and monitors. To make the counting of these reserved resources simpler, replace the existing RESERVED_* defines with enumerations. Signed-off-by: Tomer Tayar <ttayar@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-30 00:20:38 +08:00
gaudi2_arm_cq_monitor(hdev, GAUDI2_RESERVED_SOB_KDMA_COMPLETION,
GAUDI2_RESERVED_MON_KDMA_COMPLETION,
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
GAUDI2_RESERVED_CQ_KDMA_COMPLETION, 1, 1);
comp_addr = CFG_BASE + mmDCORE0_SYNC_MNGR_OBJS_SOB_OBJ_0 +
habanalabs/gaudi2: replace defines for reserved sob/mob with enums Following patches are going to add more reserved sync objects and monitors. To make the counting of these reserved resources simpler, replace the existing RESERVED_* defines with enumerations. Signed-off-by: Tomer Tayar <ttayar@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-30 00:20:38 +08:00
(GAUDI2_RESERVED_SOB_KDMA_COMPLETION * sizeof(u32));
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
comp_val = FIELD_PREP(DCORE0_SYNC_MNGR_OBJS_SOB_OBJ_INC_MASK, 1) |
FIELD_PREP(DCORE0_SYNC_MNGR_OBJS_SOB_OBJ_VAL_MASK, 1);
WREG32(mmARC_FARM_KDMA_CTX_SRC_BASE_LO, lower_32_bits(src_addr));
WREG32(mmARC_FARM_KDMA_CTX_SRC_BASE_HI, upper_32_bits(src_addr));
WREG32(mmARC_FARM_KDMA_CTX_DST_BASE_LO, lower_32_bits(dst_addr));
WREG32(mmARC_FARM_KDMA_CTX_DST_BASE_HI, upper_32_bits(dst_addr));
WREG32(mmARC_FARM_KDMA_CTX_WR_COMP_ADDR_LO, lower_32_bits(comp_addr));
WREG32(mmARC_FARM_KDMA_CTX_WR_COMP_ADDR_HI, upper_32_bits(comp_addr));
WREG32(mmARC_FARM_KDMA_CTX_WR_COMP_WDATA, comp_val);
WREG32(mmARC_FARM_KDMA_CTX_DST_TSIZE_0, size);
commit_mask = FIELD_PREP(ARC_FARM_KDMA_CTX_COMMIT_LIN_MASK, 1) |
FIELD_PREP(ARC_FARM_KDMA_CTX_COMMIT_WR_COMP_EN_MASK, 1);
if (is_memset)
commit_mask |= FIELD_PREP(ARC_FARM_KDMA_CTX_COMMIT_MEM_SET_MASK, 1);
WREG32(mmARC_FARM_KDMA_CTX_COMMIT, commit_mask);
/* Wait for completion */
cq = &hdev->completion_queue[GAUDI2_RESERVED_CQ_KDMA_COMPLETION];
cq_base = cq->kernel_address;
polling_addr = (u32 *)&cq_base[cq->ci];
if (hdev->pldm)
/* for each 1MB 20 second of timeout */
timeout = ((size / SZ_1M) + 1) * USEC_PER_SEC * 20;
else
timeout = KDMA_TIMEOUT_USEC;
/* Polling */
rc = hl_poll_timeout_memory(
hdev,
polling_addr,
status,
(status == 1),
1000,
timeout,
true);
*polling_addr = 0;
if (rc) {
dev_err(hdev->dev, "Timeout while waiting for KDMA to be idle\n");
WREG32(mmARC_FARM_KDMA_CFG_1, 1 << ARC_FARM_KDMA_CFG_1_HALT_SHIFT);
return rc;
}
cq->ci = hl_cq_inc_ptr(cq->ci);
return 0;
}
static void gaudi2_memset_device_lbw(struct hl_device *hdev, u32 addr, u32 size, u32 val)
{
u32 i;
for (i = 0 ; i < size ; i += sizeof(u32))
WREG32(addr + i, val);
}
static void gaudi2_qman_set_test_mode(struct hl_device *hdev, u32 hw_queue_id, bool enable)
{
u32 reg_base = gaudi2_qm_blocks_bases[hw_queue_id];
if (enable) {
WREG32(reg_base + QM_GLBL_PROT_OFFSET, QMAN_MAKE_TRUSTED_TEST_MODE);
WREG32(reg_base + QM_PQC_CFG_OFFSET, 0);
} else {
WREG32(reg_base + QM_GLBL_PROT_OFFSET, QMAN_MAKE_TRUSTED);
WREG32(reg_base + QM_PQC_CFG_OFFSET, 1 << PDMA0_QM_PQC_CFG_EN_SHIFT);
}
}
static int gaudi2_test_queue(struct hl_device *hdev, u32 hw_queue_id)
{
u32 sob_offset = hdev->asic_prop.first_available_user_sob[0] * 4;
u32 sob_addr = mmDCORE0_SYNC_MNGR_OBJS_SOB_OBJ_0 + sob_offset;
u32 timeout_usec, tmp, sob_base = 1, sob_val = 0x5a5a;
struct packet_msg_short *msg_short_pkt;
dma_addr_t pkt_dma_addr;
size_t pkt_size;
int rc;
if (hdev->pldm)
timeout_usec = GAUDI2_PLDM_TEST_QUEUE_WAIT_USEC;
else
timeout_usec = GAUDI2_TEST_QUEUE_WAIT_USEC;
pkt_size = sizeof(*msg_short_pkt);
msg_short_pkt = hl_asic_dma_pool_zalloc(hdev, pkt_size, GFP_KERNEL, &pkt_dma_addr);
if (!msg_short_pkt) {
dev_err(hdev->dev, "Failed to allocate packet for H/W queue %d testing\n",
hw_queue_id);
return -ENOMEM;
}
tmp = (PACKET_MSG_SHORT << GAUDI2_PKT_CTL_OPCODE_SHIFT) |
(1 << GAUDI2_PKT_CTL_EB_SHIFT) |
(1 << GAUDI2_PKT_CTL_MB_SHIFT) |
(sob_base << GAUDI2_PKT_SHORT_CTL_BASE_SHIFT) |
(sob_offset << GAUDI2_PKT_SHORT_CTL_ADDR_SHIFT);
msg_short_pkt->value = cpu_to_le32(sob_val);
msg_short_pkt->ctl = cpu_to_le32(tmp);
/* Reset the SOB value */
WREG32(sob_addr, 0);
rc = hl_hw_queue_send_cb_no_cmpl(hdev, hw_queue_id, pkt_size, pkt_dma_addr);
if (rc) {
dev_err(hdev->dev, "Failed to send msg_short packet to H/W queue %d\n",
hw_queue_id);
goto free_pkt;
}
rc = hl_poll_timeout(
hdev,
sob_addr,
tmp,
(tmp == sob_val),
1000,
timeout_usec);
if (rc == -ETIMEDOUT) {
dev_err(hdev->dev, "H/W queue %d test failed (SOB_OBJ_0 == 0x%x)\n",
hw_queue_id, tmp);
rc = -EIO;
}
/* Reset the SOB value */
WREG32(sob_addr, 0);
free_pkt:
hl_asic_dma_pool_free(hdev, (void *) msg_short_pkt, pkt_dma_addr);
return rc;
}
static int gaudi2_test_cpu_queue(struct hl_device *hdev)
{
struct gaudi2_device *gaudi2 = hdev->asic_specific;
/*
* check capability here as send_cpu_message() won't update the result
* value if no capability
*/
if (!(gaudi2->hw_cap_initialized & HW_CAP_CPU_Q))
return 0;
return hl_fw_test_cpu_queue(hdev);
}
static int gaudi2_test_queues(struct hl_device *hdev)
{
int i, rc, ret_val = 0;
for (i = GAUDI2_QUEUE_ID_PDMA_0_0 ; i < GAUDI2_QUEUE_ID_CPU_PQ; i++) {
if (!gaudi2_is_queue_enabled(hdev, i))
continue;
gaudi2_qman_set_test_mode(hdev, i, true);
rc = gaudi2_test_queue(hdev, i);
gaudi2_qman_set_test_mode(hdev, i, false);
if (rc) {
ret_val = -EINVAL;
goto done;
}
}
rc = gaudi2_test_cpu_queue(hdev);
if (rc) {
ret_val = -EINVAL;
goto done;
}
done:
return ret_val;
}
habanalabs: rename non_hard_reset to compute_reset In order to be more explicit we should use the term compute_reset for describing the reset in which only the compute engines gets reset. Signed-off-by: Ofir Bitton <obitton@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-07-13 20:08:09 +08:00
static int gaudi2_compute_reset_late_init(struct hl_device *hdev)
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
{
struct gaudi2_device *gaudi2 = hdev->asic_specific;
size_t irq_arr_size;
/* TODO: missing gaudi2_nic_resume.
* Until implemented nic_hw_cap_initialized will remain zeroed
*/
gaudi2_init_arcs(hdev);
gaudi2_scrub_arcs_dccm(hdev);
habanalabs/gaudi2: add gaudi2 security module Use the generic security module to block all registers in the ASIC and then open only those that are needed to be accessed by the user. Signed-off-by: Ofir Bitton <obitton@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-27 02:30:25 +08:00
gaudi2_init_security(hdev);
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
/* Unmask all IRQs since some could have been received during the soft reset */
irq_arr_size = gaudi2->num_of_valid_hw_events * sizeof(gaudi2->hw_events[0]);
return hl_fw_unmask_irq_arr(hdev, gaudi2->hw_events, irq_arr_size);
}
static void gaudi2_is_tpc_engine_idle(struct hl_device *hdev, int dcore, int inst, u32 offset,
habanalabs: add return code field to module iterator Up until now the module iterator called void callback functions and so caller activating callback that may fail suffered from 2 issues: 1. The need to "plant" return called in the private data. This is a drawback since the iterator itself should not be aware of the private data of the caller. 2. Due to 1 even in a failure the iterator would keep iterating instead of break upon error. To overcome this an optional rc field added to the iterator context. Signed-off-by: Ohad Sharabi <osharabi@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-07-13 18:47:23 +08:00
struct iterate_module_ctx *ctx)
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
{
habanalabs: add return code field to module iterator Up until now the module iterator called void callback functions and so caller activating callback that may fail suffered from 2 issues: 1. The need to "plant" return called in the private data. This is a drawback since the iterator itself should not be aware of the private data of the caller. 2. Due to 1 even in a failure the iterator would keep iterating instead of break upon error. To overcome this an optional rc field added to the iterator context. Signed-off-by: Ohad Sharabi <osharabi@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-07-13 18:47:23 +08:00
struct gaudi2_tpc_idle_data *idle_data = ctx->data;
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
u32 tpc_cfg_sts, qm_glbl_sts0, qm_glbl_sts1, qm_cgm_sts;
bool is_eng_idle;
int engine_idx;
if ((dcore == 0) && (inst == (NUM_DCORE0_TPC - 1)))
engine_idx = GAUDI2_DCORE0_ENGINE_ID_TPC_6;
else
engine_idx = GAUDI2_DCORE0_ENGINE_ID_TPC_0 +
dcore * GAUDI2_ENGINE_ID_DCORE_OFFSET + inst;
tpc_cfg_sts = RREG32(mmDCORE0_TPC0_CFG_STATUS + offset);
qm_glbl_sts0 = RREG32(mmDCORE0_TPC0_QM_GLBL_STS0 + offset);
qm_glbl_sts1 = RREG32(mmDCORE0_TPC0_QM_GLBL_STS1 + offset);
qm_cgm_sts = RREG32(mmDCORE0_TPC0_QM_CGM_STS + offset);
is_eng_idle = IS_QM_IDLE(qm_glbl_sts0, qm_glbl_sts1, qm_cgm_sts) &&
IS_TPC_IDLE(tpc_cfg_sts);
*(idle_data->is_idle) &= is_eng_idle;
if (idle_data->mask && !is_eng_idle)
set_bit(engine_idx, idle_data->mask);
habanalabs: removed seq_file parameter from is_idle asic functions Change is_idle functions so it would be more usable outside debugfs. Do this by replacing seq_file parameter with regular string. Signed-off-by: Dani Liberman <dliberman@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-07-03 22:40:57 +08:00
if (idle_data->e)
hl_engine_data_sprintf(idle_data->e,
idle_data->tpc_fmt, dcore, inst,
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
is_eng_idle ? "Y" : "N",
qm_glbl_sts0, qm_cgm_sts, tpc_cfg_sts);
}
habanalabs: removed seq_file parameter from is_idle asic functions Change is_idle functions so it would be more usable outside debugfs. Do this by replacing seq_file parameter with regular string. Signed-off-by: Dani Liberman <dliberman@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-07-03 22:40:57 +08:00
static bool gaudi2_is_device_idle(struct hl_device *hdev, u64 *mask_arr, u8 mask_len,
struct engines_data *e)
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
{
u32 qm_glbl_sts0, qm_glbl_sts1, qm_cgm_sts, dma_core_idle_ind_mask,
mme_arch_sts, dec_swreg15, dec_enabled_bit;
struct asic_fixed_properties *prop = &hdev->asic_prop;
const char *rot_fmt = "%-6d%-5d%-9s%#-14x%#-12x%s\n";
unsigned long *mask = (unsigned long *) mask_arr;
const char *edma_fmt = "%-6d%-6d%-9s%#-14x%#x\n";
const char *mme_fmt = "%-5d%-6s%-9s%#-14x%#x\n";
const char *nic_fmt = "%-5d%-9s%#-14x%#-12x\n";
const char *pdma_fmt = "%-6d%-9s%#-14x%#x\n";
const char *pcie_dec_fmt = "%-10d%-9s%#x\n";
const char *dec_fmt = "%-6d%-5d%-9s%#x\n";
bool is_idle = true, is_eng_idle;
u64 offset;
struct gaudi2_tpc_idle_data tpc_idle_data = {
.tpc_fmt = "%-6d%-5d%-9s%#-14x%#-12x%#x\n",
habanalabs: removed seq_file parameter from is_idle asic functions Change is_idle functions so it would be more usable outside debugfs. Do this by replacing seq_file parameter with regular string. Signed-off-by: Dani Liberman <dliberman@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-07-03 22:40:57 +08:00
.e = e,
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
.mask = mask,
.is_idle = &is_idle,
};
struct iterate_module_ctx tpc_iter = {
.fn = &gaudi2_is_tpc_engine_idle,
.data = &tpc_idle_data,
};
int engine_idx, i, j;
/* EDMA, Two engines per Dcore */
habanalabs: removed seq_file parameter from is_idle asic functions Change is_idle functions so it would be more usable outside debugfs. Do this by replacing seq_file parameter with regular string. Signed-off-by: Dani Liberman <dliberman@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-07-03 22:40:57 +08:00
if (e)
hl_engine_data_sprintf(e,
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
"\nCORE EDMA is_idle QM_GLBL_STS0 DMA_CORE_IDLE_IND_MASK\n"
"---- ---- ------- ------------ ----------------------\n");
for (i = 0; i < NUM_OF_DCORES; i++) {
for (j = 0 ; j < NUM_OF_EDMA_PER_DCORE ; j++) {
int seq = i * NUM_OF_EDMA_PER_DCORE + j;
if (!(prop->edma_enabled_mask & BIT(seq)))
continue;
engine_idx = GAUDI2_DCORE0_ENGINE_ID_EDMA_0 +
i * GAUDI2_ENGINE_ID_DCORE_OFFSET + j;
offset = i * DCORE_OFFSET + j * DCORE_EDMA_OFFSET;
dma_core_idle_ind_mask =
RREG32(mmDCORE0_EDMA0_CORE_IDLE_IND_MASK + offset);
qm_glbl_sts0 = RREG32(mmDCORE0_EDMA0_QM_GLBL_STS0 + offset);
qm_glbl_sts1 = RREG32(mmDCORE0_EDMA0_QM_GLBL_STS1 + offset);
qm_cgm_sts = RREG32(mmDCORE0_EDMA0_QM_CGM_STS + offset);
is_eng_idle = IS_QM_IDLE(qm_glbl_sts0, qm_glbl_sts1, qm_cgm_sts) &&
IS_DMA_IDLE(dma_core_idle_ind_mask);
is_idle &= is_eng_idle;
if (mask && !is_eng_idle)
set_bit(engine_idx, mask);
habanalabs: removed seq_file parameter from is_idle asic functions Change is_idle functions so it would be more usable outside debugfs. Do this by replacing seq_file parameter with regular string. Signed-off-by: Dani Liberman <dliberman@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-07-03 22:40:57 +08:00
if (e)
hl_engine_data_sprintf(e, edma_fmt, i, j,
is_eng_idle ? "Y" : "N",
qm_glbl_sts0,
dma_core_idle_ind_mask);
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
}
}
/* PDMA, Two engines in Full chip */
habanalabs: removed seq_file parameter from is_idle asic functions Change is_idle functions so it would be more usable outside debugfs. Do this by replacing seq_file parameter with regular string. Signed-off-by: Dani Liberman <dliberman@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-07-03 22:40:57 +08:00
if (e)
hl_engine_data_sprintf(e,
"\nPDMA is_idle QM_GLBL_STS0 DMA_CORE_IDLE_IND_MASK\n"
"---- ------- ------------ ----------------------\n");
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
for (i = 0 ; i < NUM_OF_PDMA ; i++) {
engine_idx = GAUDI2_ENGINE_ID_PDMA_0 + i;
offset = i * PDMA_OFFSET;
dma_core_idle_ind_mask = RREG32(mmPDMA0_CORE_IDLE_IND_MASK + offset);
qm_glbl_sts0 = RREG32(mmPDMA0_QM_GLBL_STS0 + offset);
qm_glbl_sts1 = RREG32(mmPDMA0_QM_GLBL_STS1 + offset);
qm_cgm_sts = RREG32(mmPDMA0_QM_CGM_STS + offset);
is_eng_idle = IS_QM_IDLE(qm_glbl_sts0, qm_glbl_sts1, qm_cgm_sts) &&
IS_DMA_IDLE(dma_core_idle_ind_mask);
is_idle &= is_eng_idle;
if (mask && !is_eng_idle)
set_bit(engine_idx, mask);
habanalabs: removed seq_file parameter from is_idle asic functions Change is_idle functions so it would be more usable outside debugfs. Do this by replacing seq_file parameter with regular string. Signed-off-by: Dani Liberman <dliberman@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-07-03 22:40:57 +08:00
if (e)
hl_engine_data_sprintf(e, pdma_fmt, i, is_eng_idle ? "Y" : "N",
qm_glbl_sts0, dma_core_idle_ind_mask);
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
}
/* NIC, twelve macros in Full chip */
habanalabs: removed seq_file parameter from is_idle asic functions Change is_idle functions so it would be more usable outside debugfs. Do this by replacing seq_file parameter with regular string. Signed-off-by: Dani Liberman <dliberman@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-07-03 22:40:57 +08:00
if (e && hdev->nic_ports_mask)
hl_engine_data_sprintf(e,
"\nNIC is_idle QM_GLBL_STS0 QM_CGM_STS\n"
"--- ------- ------------ ----------\n");
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
for (i = 0 ; i < NIC_NUMBER_OF_ENGINES ; i++) {
if (!(i & 1))
offset = i / 2 * NIC_OFFSET;
else
offset += NIC_QM_OFFSET;
if (!(hdev->nic_ports_mask & BIT(i)))
continue;
engine_idx = GAUDI2_ENGINE_ID_NIC0_0 + i;
qm_glbl_sts0 = RREG32(mmNIC0_QM0_GLBL_STS0 + offset);
qm_glbl_sts1 = RREG32(mmNIC0_QM0_GLBL_STS1 + offset);
qm_cgm_sts = RREG32(mmNIC0_QM0_CGM_STS + offset);
is_eng_idle = IS_QM_IDLE(qm_glbl_sts0, qm_glbl_sts1, qm_cgm_sts);
is_idle &= is_eng_idle;
if (mask && !is_eng_idle)
set_bit(engine_idx, mask);
habanalabs: removed seq_file parameter from is_idle asic functions Change is_idle functions so it would be more usable outside debugfs. Do this by replacing seq_file parameter with regular string. Signed-off-by: Dani Liberman <dliberman@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-07-03 22:40:57 +08:00
if (e)
hl_engine_data_sprintf(e, nic_fmt, i, is_eng_idle ? "Y" : "N",
qm_glbl_sts0, qm_cgm_sts);
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
}
habanalabs: removed seq_file parameter from is_idle asic functions Change is_idle functions so it would be more usable outside debugfs. Do this by replacing seq_file parameter with regular string. Signed-off-by: Dani Liberman <dliberman@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-07-03 22:40:57 +08:00
if (e)
hl_engine_data_sprintf(e,
"\nMME Stub is_idle QM_GLBL_STS0 MME_ARCH_STATUS\n"
"--- ---- ------- ------------ ---------------\n");
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
/* MME, one per Dcore */
for (i = 0 ; i < NUM_OF_DCORES ; i++) {
engine_idx = GAUDI2_DCORE0_ENGINE_ID_MME + i * GAUDI2_ENGINE_ID_DCORE_OFFSET;
offset = i * DCORE_OFFSET;
qm_glbl_sts0 = RREG32(mmDCORE0_MME_QM_GLBL_STS0 + offset);
qm_glbl_sts1 = RREG32(mmDCORE0_MME_QM_GLBL_STS1 + offset);
qm_cgm_sts = RREG32(mmDCORE0_MME_QM_CGM_STS + offset);
is_eng_idle = IS_QM_IDLE(qm_glbl_sts0, qm_glbl_sts1, qm_cgm_sts);
is_idle &= is_eng_idle;
mme_arch_sts = RREG32(mmDCORE0_MME_CTRL_LO_ARCH_STATUS + offset);
is_eng_idle &= IS_MME_IDLE(mme_arch_sts);
is_idle &= is_eng_idle;
habanalabs: removed seq_file parameter from is_idle asic functions Change is_idle functions so it would be more usable outside debugfs. Do this by replacing seq_file parameter with regular string. Signed-off-by: Dani Liberman <dliberman@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-07-03 22:40:57 +08:00
if (e)
hl_engine_data_sprintf(e, mme_fmt, i, "N",
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
is_eng_idle ? "Y" : "N",
qm_glbl_sts0,
mme_arch_sts);
if (mask && !is_eng_idle)
set_bit(engine_idx, mask);
}
/*
* TPC
*/
habanalabs: removed seq_file parameter from is_idle asic functions Change is_idle functions so it would be more usable outside debugfs. Do this by replacing seq_file parameter with regular string. Signed-off-by: Dani Liberman <dliberman@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-07-03 22:40:57 +08:00
if (e && prop->tpc_enabled_mask)
hl_engine_data_sprintf(e,
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
"\nCORE TPC is_idle QM_GLBL_STS0 QM_CGM_STS DMA_CORE_IDLE_IND_MASK\n"
"---- --- -------- ------------ ---------- ----------------------\n");
gaudi2_iterate_tpcs(hdev, &tpc_iter);
/* Decoders, two each Dcore and two shared PCIe decoders */
habanalabs: removed seq_file parameter from is_idle asic functions Change is_idle functions so it would be more usable outside debugfs. Do this by replacing seq_file parameter with regular string. Signed-off-by: Dani Liberman <dliberman@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-07-03 22:40:57 +08:00
if (e && (prop->decoder_enabled_mask & (~PCIE_DEC_EN_MASK)))
hl_engine_data_sprintf(e,
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
"\nCORE DEC is_idle VSI_CMD_SWREG15\n"
"---- --- ------- ---------------\n");
for (i = 0 ; i < NUM_OF_DCORES ; i++) {
for (j = 0 ; j < NUM_OF_DEC_PER_DCORE ; j++) {
dec_enabled_bit = 1 << (i * NUM_OF_DEC_PER_DCORE + j);
if (!(prop->decoder_enabled_mask & dec_enabled_bit))
continue;
engine_idx = GAUDI2_DCORE0_ENGINE_ID_DEC_0 +
i * GAUDI2_ENGINE_ID_DCORE_OFFSET + j;
offset = i * DCORE_OFFSET + j * DCORE_DEC_OFFSET;
dec_swreg15 = RREG32(mmDCORE0_DEC0_CMD_SWREG15 + offset);
is_eng_idle = IS_DEC_IDLE(dec_swreg15);
is_idle &= is_eng_idle;
if (mask && !is_eng_idle)
set_bit(engine_idx, mask);
habanalabs: removed seq_file parameter from is_idle asic functions Change is_idle functions so it would be more usable outside debugfs. Do this by replacing seq_file parameter with regular string. Signed-off-by: Dani Liberman <dliberman@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-07-03 22:40:57 +08:00
if (e)
hl_engine_data_sprintf(e, dec_fmt, i, j,
is_eng_idle ? "Y" : "N", dec_swreg15);
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
}
}
habanalabs: removed seq_file parameter from is_idle asic functions Change is_idle functions so it would be more usable outside debugfs. Do this by replacing seq_file parameter with regular string. Signed-off-by: Dani Liberman <dliberman@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-07-03 22:40:57 +08:00
if (e && (prop->decoder_enabled_mask & PCIE_DEC_EN_MASK))
hl_engine_data_sprintf(e,
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
"\nPCIe DEC is_idle VSI_CMD_SWREG15\n"
"-------- ------- ---------------\n");
/* Check shared(PCIe) decoders */
for (i = 0 ; i < NUM_OF_DEC_PER_DCORE ; i++) {
dec_enabled_bit = PCIE_DEC_SHIFT + i;
if (!(prop->decoder_enabled_mask & BIT(dec_enabled_bit)))
continue;
engine_idx = GAUDI2_PCIE_ENGINE_ID_DEC_0 + i;
offset = i * DCORE_DEC_OFFSET;
dec_swreg15 = RREG32(mmPCIE_DEC0_CMD_SWREG15 + offset);
is_eng_idle = IS_DEC_IDLE(dec_swreg15);
is_idle &= is_eng_idle;
if (mask && !is_eng_idle)
set_bit(engine_idx, mask);
habanalabs: removed seq_file parameter from is_idle asic functions Change is_idle functions so it would be more usable outside debugfs. Do this by replacing seq_file parameter with regular string. Signed-off-by: Dani Liberman <dliberman@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-07-03 22:40:57 +08:00
if (e)
hl_engine_data_sprintf(e, pcie_dec_fmt, i,
is_eng_idle ? "Y" : "N", dec_swreg15);
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
}
habanalabs: removed seq_file parameter from is_idle asic functions Change is_idle functions so it would be more usable outside debugfs. Do this by replacing seq_file parameter with regular string. Signed-off-by: Dani Liberman <dliberman@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-07-03 22:40:57 +08:00
if (e)
hl_engine_data_sprintf(e,
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
"\nCORE ROT is_idle QM_GLBL_STS0 QM_CGM_STS DMA_CORE_STS0\n"
"---- ---- ------- ------------ ---------- -------------\n");
for (i = 0 ; i < NUM_OF_ROT ; i++) {
engine_idx = GAUDI2_ENGINE_ID_ROT_0 + i;
offset = i * ROT_OFFSET;
qm_glbl_sts0 = RREG32(mmROT0_QM_GLBL_STS0 + offset);
qm_glbl_sts1 = RREG32(mmROT0_QM_GLBL_STS1 + offset);
qm_cgm_sts = RREG32(mmROT0_QM_CGM_STS + offset);
is_eng_idle = IS_QM_IDLE(qm_glbl_sts0, qm_glbl_sts1, qm_cgm_sts);
is_idle &= is_eng_idle;
if (mask && !is_eng_idle)
set_bit(engine_idx, mask);
habanalabs: removed seq_file parameter from is_idle asic functions Change is_idle functions so it would be more usable outside debugfs. Do this by replacing seq_file parameter with regular string. Signed-off-by: Dani Liberman <dliberman@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-07-03 22:40:57 +08:00
if (e)
hl_engine_data_sprintf(e, rot_fmt, i, 0, is_eng_idle ? "Y" : "N",
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
qm_glbl_sts0, qm_cgm_sts, "-");
}
return is_idle;
}
static void gaudi2_hw_queues_lock(struct hl_device *hdev)
__acquires(&gaudi2->hw_queues_lock)
{
struct gaudi2_device *gaudi2 = hdev->asic_specific;
spin_lock(&gaudi2->hw_queues_lock);
}
static void gaudi2_hw_queues_unlock(struct hl_device *hdev)
__releases(&gaudi2->hw_queues_lock)
{
struct gaudi2_device *gaudi2 = hdev->asic_specific;
spin_unlock(&gaudi2->hw_queues_lock);
}
static u32 gaudi2_get_pci_id(struct hl_device *hdev)
{
return hdev->pdev->device;
}
static int gaudi2_get_eeprom_data(struct hl_device *hdev, void *data, size_t max_size)
{
struct gaudi2_device *gaudi2 = hdev->asic_specific;
if (!(gaudi2->hw_cap_initialized & HW_CAP_CPU_Q))
return 0;
return hl_fw_get_eeprom_data(hdev, data, max_size);
}
static void gaudi2_update_eq_ci(struct hl_device *hdev, u32 val)
{
WREG32(mmCPU_IF_EQ_RD_OFFS, val);
}
static void *gaudi2_get_events_stat(struct hl_device *hdev, bool aggregate, u32 *size)
{
struct gaudi2_device *gaudi2 = hdev->asic_specific;
if (aggregate) {
*size = (u32) sizeof(gaudi2->events_stat_aggregate);
return gaudi2->events_stat_aggregate;
}
*size = (u32) sizeof(gaudi2->events_stat);
return gaudi2->events_stat;
}
static void gaudi2_mmu_vdec_dcore_prepare(struct hl_device *hdev, int dcore_id,
int dcore_vdec_id, u32 rw_asid, u32 rw_mmu_bp)
{
u32 offset = (mmDCORE0_VDEC1_BRDG_CTRL_BASE - mmDCORE0_VDEC0_BRDG_CTRL_BASE) *
dcore_vdec_id + DCORE_OFFSET * dcore_id;
WREG32(mmDCORE0_VDEC0_BRDG_CTRL_AXUSER_DEC_HB_MMU_BP + offset, rw_mmu_bp);
WREG32(mmDCORE0_VDEC0_BRDG_CTRL_AXUSER_DEC_HB_ASID + offset, rw_asid);
WREG32(mmDCORE0_VDEC0_BRDG_CTRL_AXUSER_MSIX_ABNRM_HB_MMU_BP + offset, rw_mmu_bp);
WREG32(mmDCORE0_VDEC0_BRDG_CTRL_AXUSER_MSIX_ABNRM_HB_ASID + offset, rw_asid);
WREG32(mmDCORE0_VDEC0_BRDG_CTRL_AXUSER_MSIX_L2C_HB_MMU_BP + offset, rw_mmu_bp);
WREG32(mmDCORE0_VDEC0_BRDG_CTRL_AXUSER_MSIX_L2C_HB_ASID + offset, rw_asid);
WREG32(mmDCORE0_VDEC0_BRDG_CTRL_AXUSER_MSIX_NRM_HB_MMU_BP + offset, rw_mmu_bp);
WREG32(mmDCORE0_VDEC0_BRDG_CTRL_AXUSER_MSIX_NRM_HB_ASID + offset, rw_asid);
WREG32(mmDCORE0_VDEC0_BRDG_CTRL_AXUSER_MSIX_VCD_HB_MMU_BP + offset, rw_mmu_bp);
WREG32(mmDCORE0_VDEC0_BRDG_CTRL_AXUSER_MSIX_VCD_HB_ASID + offset, rw_asid);
}
static void gaudi2_mmu_dcore_prepare(struct hl_device *hdev, int dcore_id, u32 asid)
{
u32 rw_asid = (asid << ARC_FARM_KDMA_CTX_AXUSER_HB_ASID_RD_SHIFT) |
(asid << ARC_FARM_KDMA_CTX_AXUSER_HB_ASID_WR_SHIFT);
struct asic_fixed_properties *prop = &hdev->asic_prop;
u32 dcore_offset = dcore_id * DCORE_OFFSET;
u32 vdec_id, i, ports_offset, reg_val;
u8 edma_seq_base;
/* EDMA */
edma_seq_base = dcore_id * NUM_OF_EDMA_PER_DCORE;
if (prop->edma_enabled_mask & BIT(edma_seq_base)) {
WREG32(mmDCORE0_EDMA0_QM_AXUSER_NONSECURED_HB_MMU_BP + dcore_offset, 0);
WREG32(mmDCORE0_EDMA0_QM_AXUSER_NONSECURED_HB_ASID + dcore_offset, rw_asid);
WREG32(mmDCORE0_EDMA0_CORE_CTX_AXUSER_HB_MMU_BP + dcore_offset, 0);
WREG32(mmDCORE0_EDMA0_CORE_CTX_AXUSER_HB_ASID + dcore_offset, rw_asid);
}
if (prop->edma_enabled_mask & BIT(edma_seq_base + 1)) {
WREG32(mmDCORE0_EDMA1_QM_AXUSER_NONSECURED_HB_MMU_BP + dcore_offset, 0);
WREG32(mmDCORE0_EDMA1_QM_AXUSER_NONSECURED_HB_ASID + dcore_offset, rw_asid);
WREG32(mmDCORE0_EDMA1_CORE_CTX_AXUSER_HB_ASID + dcore_offset, rw_asid);
WREG32(mmDCORE0_EDMA1_CORE_CTX_AXUSER_HB_MMU_BP + dcore_offset, 0);
}
/* Sync Mngr */
WREG32(mmDCORE0_SYNC_MNGR_GLBL_ASID_NONE_SEC_PRIV + dcore_offset, asid);
/*
* Sync Mngrs on dcores 1 - 3 are exposed to user, so must use user ASID
* for any access type
*/
if (dcore_id > 0) {
reg_val = (asid << DCORE0_SYNC_MNGR_MSTR_IF_AXUSER_HB_ASID_RD_SHIFT) |
(asid << DCORE0_SYNC_MNGR_MSTR_IF_AXUSER_HB_ASID_WR_SHIFT);
WREG32(mmDCORE0_SYNC_MNGR_MSTR_IF_AXUSER_HB_ASID + dcore_offset, reg_val);
WREG32(mmDCORE0_SYNC_MNGR_MSTR_IF_AXUSER_HB_MMU_BP + dcore_offset, 0);
}
WREG32(mmDCORE0_MME_CTRL_LO_MME_AXUSER_HB_MMU_BP + dcore_offset, 0);
WREG32(mmDCORE0_MME_CTRL_LO_MME_AXUSER_HB_ASID + dcore_offset, rw_asid);
for (i = 0 ; i < NUM_OF_MME_SBTE_PORTS ; i++) {
ports_offset = i * DCORE_MME_SBTE_OFFSET;
WREG32(mmDCORE0_MME_SBTE0_MSTR_IF_AXUSER_HB_MMU_BP +
dcore_offset + ports_offset, 0);
WREG32(mmDCORE0_MME_SBTE0_MSTR_IF_AXUSER_HB_ASID +
dcore_offset + ports_offset, rw_asid);
}
for (i = 0 ; i < NUM_OF_MME_WB_PORTS ; i++) {
ports_offset = i * DCORE_MME_WB_OFFSET;
WREG32(mmDCORE0_MME_WB0_MSTR_IF_AXUSER_HB_MMU_BP +
dcore_offset + ports_offset, 0);
WREG32(mmDCORE0_MME_WB0_MSTR_IF_AXUSER_HB_ASID +
dcore_offset + ports_offset, rw_asid);
}
WREG32(mmDCORE0_MME_QM_AXUSER_NONSECURED_HB_MMU_BP + dcore_offset, 0);
WREG32(mmDCORE0_MME_QM_AXUSER_NONSECURED_HB_ASID + dcore_offset, rw_asid);
/*
* Decoders
*/
for (vdec_id = 0 ; vdec_id < NUM_OF_DEC_PER_DCORE ; vdec_id++) {
if (prop->decoder_enabled_mask & BIT(dcore_id * NUM_OF_DEC_PER_DCORE + vdec_id))
gaudi2_mmu_vdec_dcore_prepare(hdev, dcore_id, vdec_id, rw_asid, 0);
}
}
static void gudi2_mmu_vdec_shared_prepare(struct hl_device *hdev,
int shared_vdec_id, u32 rw_asid, u32 rw_mmu_bp)
{
u32 offset = (mmPCIE_VDEC1_BRDG_CTRL_BASE - mmPCIE_VDEC0_BRDG_CTRL_BASE) * shared_vdec_id;
WREG32(mmPCIE_VDEC0_BRDG_CTRL_AXUSER_DEC_HB_MMU_BP + offset, rw_mmu_bp);
WREG32(mmPCIE_VDEC0_BRDG_CTRL_AXUSER_DEC_HB_ASID + offset, rw_asid);
WREG32(mmPCIE_VDEC0_BRDG_CTRL_AXUSER_MSIX_ABNRM_HB_MMU_BP + offset, rw_mmu_bp);
WREG32(mmPCIE_VDEC0_BRDG_CTRL_AXUSER_MSIX_ABNRM_HB_ASID + offset, rw_asid);
WREG32(mmPCIE_VDEC0_BRDG_CTRL_AXUSER_MSIX_L2C_HB_MMU_BP + offset, rw_mmu_bp);
WREG32(mmPCIE_VDEC0_BRDG_CTRL_AXUSER_MSIX_L2C_HB_ASID + offset, rw_asid);
WREG32(mmPCIE_VDEC0_BRDG_CTRL_AXUSER_MSIX_NRM_HB_MMU_BP + offset, rw_mmu_bp);
WREG32(mmPCIE_VDEC0_BRDG_CTRL_AXUSER_MSIX_NRM_HB_ASID + offset, rw_asid);
WREG32(mmPCIE_VDEC0_BRDG_CTRL_AXUSER_MSIX_VCD_HB_MMU_BP + offset, rw_mmu_bp);
WREG32(mmPCIE_VDEC0_BRDG_CTRL_AXUSER_MSIX_VCD_HB_ASID + offset, rw_asid);
}
static void gudi2_mmu_arc_farm_arc_dup_eng_prepare(struct hl_device *hdev, int arc_farm_id,
u32 rw_asid, u32 rw_mmu_bp)
{
u32 offset = (mmARC_FARM_ARC1_DUP_ENG_BASE - mmARC_FARM_ARC0_DUP_ENG_BASE) * arc_farm_id;
WREG32(mmARC_FARM_ARC0_DUP_ENG_AXUSER_HB_MMU_BP + offset, rw_mmu_bp);
WREG32(mmARC_FARM_ARC0_DUP_ENG_AXUSER_HB_ASID + offset, rw_asid);
}
static void gaudi2_arc_mmu_prepare(struct hl_device *hdev, u32 cpu_id, u32 asid)
{
u32 reg_base, reg_offset, reg_val = 0;
reg_base = gaudi2_arc_blocks_bases[cpu_id];
/* Enable MMU and configure asid for all relevant ARC regions */
reg_val = FIELD_PREP(ARC_FARM_ARC0_AUX_ARC_REGION_CFG_MMU_BP_MASK, 0);
reg_val |= FIELD_PREP(ARC_FARM_ARC0_AUX_ARC_REGION_CFG_0_ASID_MASK, asid);
reg_offset = ARC_REGION_CFG_OFFSET(ARC_REGION3_GENERAL);
WREG32(reg_base + reg_offset, reg_val);
reg_offset = ARC_REGION_CFG_OFFSET(ARC_REGION4_HBM0_FW);
WREG32(reg_base + reg_offset, reg_val);
reg_offset = ARC_REGION_CFG_OFFSET(ARC_REGION5_HBM1_GC_DATA);
WREG32(reg_base + reg_offset, reg_val);
reg_offset = ARC_REGION_CFG_OFFSET(ARC_REGION6_HBM2_GC_DATA);
WREG32(reg_base + reg_offset, reg_val);
reg_offset = ARC_REGION_CFG_OFFSET(ARC_REGION7_HBM3_GC_DATA);
WREG32(reg_base + reg_offset, reg_val);
reg_offset = ARC_REGION_CFG_OFFSET(ARC_REGION9_PCIE);
WREG32(reg_base + reg_offset, reg_val);
reg_offset = ARC_REGION_CFG_OFFSET(ARC_REGION10_GENERAL);
WREG32(reg_base + reg_offset, reg_val);
reg_offset = ARC_REGION_CFG_OFFSET(ARC_REGION11_GENERAL);
WREG32(reg_base + reg_offset, reg_val);
reg_offset = ARC_REGION_CFG_OFFSET(ARC_REGION12_GENERAL);
WREG32(reg_base + reg_offset, reg_val);
reg_offset = ARC_REGION_CFG_OFFSET(ARC_REGION13_GENERAL);
WREG32(reg_base + reg_offset, reg_val);
reg_offset = ARC_REGION_CFG_OFFSET(ARC_REGION14_GENERAL);
WREG32(reg_base + reg_offset, reg_val);
}
static int gaudi2_arc_mmu_prepare_all(struct hl_device *hdev, u32 asid)
{
int i;
if (hdev->fw_components & FW_TYPE_BOOT_CPU)
return hl_fw_cpucp_engine_core_asid_set(hdev, asid);
for (i = CPU_ID_SCHED_ARC0 ; i < NUM_OF_ARC_FARMS_ARC ; i++)
gaudi2_arc_mmu_prepare(hdev, i, asid);
for (i = GAUDI2_QUEUE_ID_PDMA_0_0 ; i < GAUDI2_QUEUE_ID_CPU_PQ ; i += 4) {
if (!gaudi2_is_queue_enabled(hdev, i))
continue;
gaudi2_arc_mmu_prepare(hdev, gaudi2_queue_id_to_arc_id[i], asid);
}
return 0;
}
static int gaudi2_mmu_shared_prepare(struct hl_device *hdev, u32 asid)
{
struct asic_fixed_properties *prop = &hdev->asic_prop;
u32 rw_asid, offset;
int rc, i;
rw_asid = FIELD_PREP(ARC_FARM_KDMA_CTX_AXUSER_HB_ASID_RD_MASK, asid) |
FIELD_PREP(ARC_FARM_KDMA_CTX_AXUSER_HB_ASID_WR_MASK, asid);
WREG32(mmPDMA0_QM_AXUSER_NONSECURED_HB_ASID, rw_asid);
WREG32(mmPDMA0_QM_AXUSER_NONSECURED_HB_MMU_BP, 0);
WREG32(mmPDMA0_CORE_CTX_AXUSER_HB_ASID, rw_asid);
WREG32(mmPDMA0_CORE_CTX_AXUSER_HB_MMU_BP, 0);
WREG32(mmPDMA1_QM_AXUSER_NONSECURED_HB_ASID, rw_asid);
WREG32(mmPDMA1_QM_AXUSER_NONSECURED_HB_MMU_BP, 0);
WREG32(mmPDMA1_CORE_CTX_AXUSER_HB_ASID, rw_asid);
WREG32(mmPDMA1_CORE_CTX_AXUSER_HB_MMU_BP, 0);
/* ROT */
for (i = 0 ; i < NUM_OF_ROT ; i++) {
offset = i * ROT_OFFSET;
WREG32(mmROT0_QM_AXUSER_NONSECURED_HB_ASID + offset, rw_asid);
WREG32(mmROT0_QM_AXUSER_NONSECURED_HB_MMU_BP + offset, 0);
RMWREG32(mmROT0_CPL_QUEUE_AWUSER + offset, asid, MMUBP_ASID_MASK);
RMWREG32(mmROT0_DESC_HBW_ARUSER_LO + offset, asid, MMUBP_ASID_MASK);
RMWREG32(mmROT0_DESC_HBW_AWUSER_LO + offset, asid, MMUBP_ASID_MASK);
}
/* Shared Decoders are the last bits in the decoders mask */
if (prop->decoder_enabled_mask & BIT(NUM_OF_DCORES * NUM_OF_DEC_PER_DCORE + 0))
gudi2_mmu_vdec_shared_prepare(hdev, 0, rw_asid, 0);
if (prop->decoder_enabled_mask & BIT(NUM_OF_DCORES * NUM_OF_DEC_PER_DCORE + 1))
gudi2_mmu_vdec_shared_prepare(hdev, 1, rw_asid, 0);
/* arc farm arc dup eng */
for (i = 0 ; i < NUM_OF_ARC_FARMS_ARC ; i++)
gudi2_mmu_arc_farm_arc_dup_eng_prepare(hdev, i, rw_asid, 0);
rc = gaudi2_arc_mmu_prepare_all(hdev, asid);
if (rc)
return rc;
return 0;
}
static void gaudi2_tpc_mmu_prepare(struct hl_device *hdev, int dcore, int inst, u32 offset,
habanalabs: add return code field to module iterator Up until now the module iterator called void callback functions and so caller activating callback that may fail suffered from 2 issues: 1. The need to "plant" return called in the private data. This is a drawback since the iterator itself should not be aware of the private data of the caller. 2. Due to 1 even in a failure the iterator would keep iterating instead of break upon error. To overcome this an optional rc field added to the iterator context. Signed-off-by: Ohad Sharabi <osharabi@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-07-13 18:47:23 +08:00
struct iterate_module_ctx *ctx)
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
{
habanalabs: add return code field to module iterator Up until now the module iterator called void callback functions and so caller activating callback that may fail suffered from 2 issues: 1. The need to "plant" return called in the private data. This is a drawback since the iterator itself should not be aware of the private data of the caller. 2. Due to 1 even in a failure the iterator would keep iterating instead of break upon error. To overcome this an optional rc field added to the iterator context. Signed-off-by: Ohad Sharabi <osharabi@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-07-13 18:47:23 +08:00
struct gaudi2_tpc_mmu_data *mmu_data = ctx->data;
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
WREG32(mmDCORE0_TPC0_CFG_AXUSER_HB_MMU_BP + offset, 0);
WREG32(mmDCORE0_TPC0_CFG_AXUSER_HB_ASID + offset, mmu_data->rw_asid);
WREG32(mmDCORE0_TPC0_QM_AXUSER_NONSECURED_HB_MMU_BP + offset, 0);
WREG32(mmDCORE0_TPC0_QM_AXUSER_NONSECURED_HB_ASID + offset, mmu_data->rw_asid);
}
/* zero the MMUBP and set the ASID */
static int gaudi2_mmu_prepare(struct hl_device *hdev, u32 asid)
{
struct gaudi2_device *gaudi2 = hdev->asic_specific;
struct gaudi2_tpc_mmu_data tpc_mmu_data;
struct iterate_module_ctx tpc_iter = {
.fn = &gaudi2_tpc_mmu_prepare,
.data = &tpc_mmu_data,
};
int rc, i;
if (asid & ~DCORE0_HMMU0_STLB_ASID_ASID_MASK) {
dev_crit(hdev->dev, "asid %u is too big\n", asid);
return -EINVAL;
}
if (!(gaudi2->hw_cap_initialized & HW_CAP_MMU_MASK))
return 0;
rc = gaudi2_mmu_shared_prepare(hdev, asid);
if (rc)
return rc;
/* configure DCORE MMUs */
tpc_mmu_data.rw_asid = (asid << ARC_FARM_KDMA_CTX_AXUSER_HB_ASID_RD_SHIFT) |
(asid << ARC_FARM_KDMA_CTX_AXUSER_HB_ASID_WR_SHIFT);
gaudi2_iterate_tpcs(hdev, &tpc_iter);
for (i = 0 ; i < NUM_OF_DCORES ; i++)
gaudi2_mmu_dcore_prepare(hdev, i, asid);
return 0;
}
static inline bool is_info_event(u32 event)
{
switch (event) {
case GAUDI2_EVENT_CPU_CPLD_SHUTDOWN_CAUSE:
return true;
default:
return false;
}
}
static void gaudi2_print_irq_info(struct hl_device *hdev, u16 event_type)
{
char desc[64] = "";
bool event_valid = false;
/* return in case of NIC status event - these events are received periodically and not as
* an indication to an error, thus not printed.
*/
if (event_type >= GAUDI2_EVENT_CPU0_STATUS_NIC0_ENG0 &&
event_type <= GAUDI2_EVENT_CPU11_STATUS_NIC11_ENG1)
return;
if (gaudi2_irq_map_table[event_type].valid) {
snprintf(desc, sizeof(desc), gaudi2_irq_map_table[event_type].name);
event_valid = true;
}
if (!event_valid)
snprintf(desc, sizeof(desc), "N/A");
if (is_info_event(event_type))
dev_info_ratelimited(hdev->dev, "Received H/W interrupt %d [\"%s\"]\n",
event_type, desc);
else
dev_err_ratelimited(hdev->dev, "Received H/W interrupt %d [\"%s\"]\n",
event_type, desc);
}
habanalabs/gaudi2: reset device upon critical ECC event Correctable ECC events are not fatal, but as they accumulate, the f/w can decide that a hard-rest is required. This indication is propagated to the host using the existing ECC event interface. Signed-off-by: Ofir Bitton <obitton@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-28 13:34:28 +08:00
static bool gaudi2_handle_ecc_event(struct hl_device *hdev, u16 event_type,
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
struct hl_eq_ecc_data *ecc_data)
{
u64 ecc_address = 0, ecc_syndrom = 0;
u8 memory_wrapper_idx = 0;
ecc_address = le64_to_cpu(ecc_data->ecc_address);
ecc_syndrom = le64_to_cpu(ecc_data->ecc_syndrom);
memory_wrapper_idx = ecc_data->memory_wrapper_idx;
habanalabs/gaudi2: reset device upon critical ECC event Correctable ECC events are not fatal, but as they accumulate, the f/w can decide that a hard-rest is required. This indication is propagated to the host using the existing ECC event interface. Signed-off-by: Ofir Bitton <obitton@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-28 13:34:28 +08:00
dev_err(hdev->dev,
"ECC error detected. address: %#llx. Syndrom: %#llx. block id %u. critical %u.\n",
ecc_address, ecc_syndrom, memory_wrapper_idx, ecc_data->is_critical);
return !!ecc_data->is_critical;
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
}
/*
* gaudi2_queue_idx_dec - decrement queue index (pi/ci) and handle wrap
*
* @idx: the current pi/ci value
* @q_len: the queue length (power of 2)
*
* @return the cyclically decremented index
*/
static inline u32 gaudi2_queue_idx_dec(u32 idx, u32 q_len)
{
u32 mask = q_len - 1;
/*
* modular decrement is equivalent to adding (queue_size -1)
* later we take LSBs to make sure the value is in the
* range [0, queue_len - 1]
*/
return (idx + q_len - 1) & mask;
}
/**
* gaudi2_print_sw_config_stream_data - print SW config stream data
*
* @hdev: pointer to the habanalabs device structure
* @stream: the QMAN's stream
* @qman_base: base address of QMAN registers block
*/
static void gaudi2_print_sw_config_stream_data(struct hl_device *hdev,
u32 stream, u64 qman_base)
{
u64 cq_ptr_lo, cq_ptr_hi, cq_tsize, cq_ptr;
u32 cq_ptr_lo_off, size;
cq_ptr_lo_off = mmDCORE0_TPC0_QM_CQ_PTR_LO_1 - mmDCORE0_TPC0_QM_CQ_PTR_LO_0;
cq_ptr_lo = qman_base + (mmDCORE0_TPC0_QM_CQ_PTR_LO_0 - mmDCORE0_TPC0_QM_BASE) +
stream * cq_ptr_lo_off;
cq_ptr_hi = cq_ptr_lo + (mmDCORE0_TPC0_QM_CQ_PTR_HI_0 - mmDCORE0_TPC0_QM_CQ_PTR_LO_0);
cq_tsize = cq_ptr_lo + (mmDCORE0_TPC0_QM_CQ_TSIZE_0 - mmDCORE0_TPC0_QM_CQ_PTR_LO_0);
cq_ptr = (((u64) RREG32(cq_ptr_hi)) << 32) | RREG32(cq_ptr_lo);
size = RREG32(cq_tsize);
dev_info(hdev->dev, "stop on err: stream: %u, addr: %#llx, size: %x\n",
stream, cq_ptr, size);
}
/**
* gaudi2_print_last_pqes_on_err - print last PQEs on error
*
* @hdev: pointer to the habanalabs device structure
* @qid_base: first QID of the QMAN (out of 4 streams)
* @stream: the QMAN's stream
* @qman_base: base address of QMAN registers block
* @pr_sw_conf: if true print the SW config stream data (CQ PTR and SIZE)
*/
static void gaudi2_print_last_pqes_on_err(struct hl_device *hdev, u32 qid_base, u32 stream,
u64 qman_base, bool pr_sw_conf)
{
u32 ci, qm_ci_stream_off;
struct hl_hw_queue *q;
u64 pq_ci;
int i;
q = &hdev->kernel_queues[qid_base + stream];
qm_ci_stream_off = mmDCORE0_TPC0_QM_PQ_CI_1 - mmDCORE0_TPC0_QM_PQ_CI_0;
pq_ci = qman_base + (mmDCORE0_TPC0_QM_PQ_CI_0 - mmDCORE0_TPC0_QM_BASE) +
stream * qm_ci_stream_off;
hdev->asic_funcs->hw_queues_lock(hdev);
if (pr_sw_conf)
gaudi2_print_sw_config_stream_data(hdev, stream, qman_base);
ci = RREG32(pq_ci);
/* we should start printing form ci -1 */
ci = gaudi2_queue_idx_dec(ci, HL_QUEUE_LENGTH);
for (i = 0; i < PQ_FETCHER_CACHE_SIZE; i++) {
struct hl_bd *bd;
u64 addr;
u32 len;
bd = q->kernel_address;
bd += ci;
len = le32_to_cpu(bd->len);
/* len 0 means uninitialized entry- break */
if (!len)
break;
addr = le64_to_cpu(bd->ptr);
dev_info(hdev->dev, "stop on err PQE(stream %u): ci: %u, addr: %#llx, size: %x\n",
stream, ci, addr, len);
/* get previous ci, wrap if needed */
ci = gaudi2_queue_idx_dec(ci, HL_QUEUE_LENGTH);
}
hdev->asic_funcs->hw_queues_unlock(hdev);
}
/**
* print_qman_data_on_err - extract QMAN data on error
*
* @hdev: pointer to the habanalabs device structure
* @qid_base: first QID of the QMAN (out of 4 streams)
* @stream: the QMAN's stream
* @qman_base: base address of QMAN registers block
*
* This function attempt to extract as much data as possible on QMAN error.
* On upper CP print the SW config stream data and last 8 PQEs.
* On lower CP print SW config data and last PQEs of ALL 4 upper CPs
*/
static void print_qman_data_on_err(struct hl_device *hdev, u32 qid_base, u32 stream, u64 qman_base)
{
u32 i;
if (stream != QMAN_STREAMS) {
gaudi2_print_last_pqes_on_err(hdev, qid_base, stream, qman_base, true);
return;
}
gaudi2_print_sw_config_stream_data(hdev, stream, qman_base);
for (i = 0 ; i < QMAN_STREAMS ; i++)
gaudi2_print_last_pqes_on_err(hdev, qid_base, i, qman_base, false);
}
static void gaudi2_handle_qman_err_generic(struct hl_device *hdev, const char *qm_name,
u64 qman_base, u32 qid_base)
{
habanalabs/gaudi2: remove unused variable glbl_sts_clr_val was set but never used Reported-by: kernel test robot <lkp@intel.com> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-30 16:07:59 +08:00
u32 i, j, glbl_sts_val, arb_err_val, num_error_causes;
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
u64 glbl_sts_addr, arb_err_addr;
char reg_desc[32];
glbl_sts_addr = qman_base + (mmDCORE0_TPC0_QM_GLBL_ERR_STS_0 - mmDCORE0_TPC0_QM_BASE);
arb_err_addr = qman_base + (mmDCORE0_TPC0_QM_ARB_ERR_CAUSE - mmDCORE0_TPC0_QM_BASE);
/* Iterate through all stream GLBL_ERR_STS registers + Lower CP */
for (i = 0 ; i < QMAN_STREAMS + 1 ; i++) {
glbl_sts_val = RREG32(glbl_sts_addr + 4 * i);
if (!glbl_sts_val)
continue;
if (i == QMAN_STREAMS) {
snprintf(reg_desc, ARRAY_SIZE(reg_desc), "LowerCP");
num_error_causes = GAUDI2_NUM_OF_QM_LCP_ERR_CAUSE;
} else {
snprintf(reg_desc, ARRAY_SIZE(reg_desc), "stream%u", i);
num_error_causes = GAUDI2_NUM_OF_QM_ERR_CAUSE;
}
habanalabs/gaudi2: remove unused variable glbl_sts_clr_val was set but never used Reported-by: kernel test robot <lkp@intel.com> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-30 16:07:59 +08:00
for (j = 0 ; j < num_error_causes ; j++)
if (glbl_sts_val & BIT(j))
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
dev_err_ratelimited(hdev->dev, "%s %s. err cause: %s\n",
qm_name, reg_desc,
i == QMAN_STREAMS ?
gaudi2_qman_lower_cp_error_cause[j] :
gaudi2_qman_error_cause[j]);
print_qman_data_on_err(hdev, qid_base, i, qman_base);
}
arb_err_val = RREG32(arb_err_addr);
if (!arb_err_val)
return;
for (j = 0 ; j < GAUDI2_NUM_OF_QM_ARB_ERR_CAUSE ; j++) {
if (arb_err_val & BIT(j))
dev_err_ratelimited(hdev->dev, "%s ARB_ERR. err cause: %s\n",
qm_name, gaudi2_qman_arb_error_cause[j]);
}
}
static void gaudi2_razwi_rr_hbw_shared_printf_info(struct hl_device *hdev,
u64 rtr_mstr_if_base_addr, bool is_write, char *name,
bool read_razwi_regs, struct hl_eq_razwi_info *razwi_info)
{
u32 razwi_hi, razwi_lo, razwi_xy;
if (is_write) {
if (read_razwi_regs) {
razwi_hi = RREG32(rtr_mstr_if_base_addr + RR_SHRD_HBW_AW_RAZWI_HI);
razwi_lo = RREG32(rtr_mstr_if_base_addr + RR_SHRD_HBW_AW_RAZWI_LO);
razwi_xy = RREG32(rtr_mstr_if_base_addr + RR_SHRD_HBW_AW_RAZWI_XY);
} else {
razwi_hi = le32_to_cpu(razwi_info->hbw.rr_aw_razwi_hi_reg);
razwi_lo = le32_to_cpu(razwi_info->hbw.rr_aw_razwi_lo_reg);
razwi_xy = le32_to_cpu(razwi_info->hbw.rr_aw_razwi_id_reg);
}
} else {
if (read_razwi_regs) {
razwi_hi = RREG32(rtr_mstr_if_base_addr + RR_SHRD_HBW_AR_RAZWI_HI);
razwi_lo = RREG32(rtr_mstr_if_base_addr + RR_SHRD_HBW_AR_RAZWI_LO);
razwi_xy = RREG32(rtr_mstr_if_base_addr + RR_SHRD_HBW_AR_RAZWI_XY);
} else {
razwi_hi = le32_to_cpu(razwi_info->hbw.rr_ar_razwi_hi_reg);
razwi_lo = le32_to_cpu(razwi_info->hbw.rr_ar_razwi_lo_reg);
razwi_xy = le32_to_cpu(razwi_info->hbw.rr_ar_razwi_id_reg);
}
}
habanalabs/gaudi2: dump detailed information upon RAZWI In order to improve debuggability, we add all available information when a RAZWI event occur. Signed-off-by: Ofir Bitton <obitton@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-08-21 18:50:51 +08:00
dev_err_ratelimited(hdev->dev,
"%s-RAZWI SHARED RR HBW %s error, address %#llx, Initiator coordinates 0x%x\n",
name, is_write ? "WR" : "RD", (u64)razwi_hi << 32 | razwi_lo, razwi_xy);
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
}
static void gaudi2_razwi_rr_lbw_shared_printf_info(struct hl_device *hdev,
u64 rtr_mstr_if_base_addr, bool is_write, char *name,
bool read_razwi_regs, struct hl_eq_razwi_info *razwi_info)
{
u32 razwi_addr, razwi_xy;
if (is_write) {
if (read_razwi_regs) {
razwi_addr = RREG32(rtr_mstr_if_base_addr + RR_SHRD_LBW_AW_RAZWI);
razwi_xy = RREG32(rtr_mstr_if_base_addr + RR_SHRD_LBW_AW_RAZWI_XY);
} else {
razwi_addr = le32_to_cpu(razwi_info->lbw.rr_aw_razwi_reg);
razwi_xy = le32_to_cpu(razwi_info->lbw.rr_aw_razwi_id_reg);
}
dev_err_ratelimited(hdev->dev,
"%s-RAZWI SHARED RR LBW WR error, mstr_if 0x%llx, captured address 0x%x, Initiator coordinates 0x%x\n",
name, rtr_mstr_if_base_addr, razwi_addr, razwi_xy);
} else {
if (read_razwi_regs) {
razwi_addr = RREG32(rtr_mstr_if_base_addr + RR_SHRD_LBW_AR_RAZWI);
razwi_xy = RREG32(rtr_mstr_if_base_addr + RR_SHRD_LBW_AR_RAZWI_XY);
} else {
razwi_addr = le32_to_cpu(razwi_info->lbw.rr_ar_razwi_reg);
razwi_xy = le32_to_cpu(razwi_info->lbw.rr_ar_razwi_id_reg);
}
dev_err_ratelimited(hdev->dev,
"%s-RAZWI SHARED RR LBW AR error, mstr_if 0x%llx, captured address 0x%x Initiator coordinates 0x%x\n",
name, rtr_mstr_if_base_addr, razwi_addr, razwi_xy);
}
}
/*
* This function handles RR(Range register) hit events.
* raised be initiators not PSOC RAZWI.
*/
static void gaudi2_ack_module_razwi_event_handler(struct hl_device *hdev,
enum razwi_event_sources module, u8 module_idx,
u8 module_sub_idx, struct hl_eq_razwi_info *razwi_info)
{
bool via_sft = false, read_razwi_regs = false;
u32 rtr_id, dcore_id, dcore_rtr_id, sft_id;
u64 rtr_mstr_if_base_addr;
u32 hbw_shrd_aw = 0, hbw_shrd_ar = 0;
u32 lbw_shrd_aw = 0, lbw_shrd_ar = 0;
char initiator_name[64];
if (hdev->pldm || !(hdev->fw_components & FW_TYPE_LINUX) || !razwi_info)
read_razwi_regs = true;
switch (module) {
case RAZWI_TPC:
rtr_id = gaudi2_tpc_initiator_rtr_id[module_idx];
sprintf(initiator_name, "TPC_%u", module_idx);
break;
case RAZWI_MME:
sprintf(initiator_name, "MME_%u", module_idx);
switch (module_sub_idx) {
case MME_WAP0:
rtr_id = gaudi2_mme_initiator_rtr_id[module_idx].wap0;
break;
case MME_WAP1:
rtr_id = gaudi2_mme_initiator_rtr_id[module_idx].wap1;
break;
case MME_WRITE:
rtr_id = gaudi2_mme_initiator_rtr_id[module_idx].write;
break;
case MME_READ:
rtr_id = gaudi2_mme_initiator_rtr_id[module_idx].read;
break;
case MME_SBTE0:
rtr_id = gaudi2_mme_initiator_rtr_id[module_idx].sbte0;
break;
case MME_SBTE1:
rtr_id = gaudi2_mme_initiator_rtr_id[module_idx].sbte1;
break;
case MME_SBTE2:
rtr_id = gaudi2_mme_initiator_rtr_id[module_idx].sbte2;
break;
case MME_SBTE3:
rtr_id = gaudi2_mme_initiator_rtr_id[module_idx].sbte3;
break;
case MME_SBTE4:
rtr_id = gaudi2_mme_initiator_rtr_id[module_idx].sbte4;
break;
default:
return;
}
break;
case RAZWI_EDMA:
sft_id = gaudi2_edma_initiator_sft_id[module_idx].interface_id;
dcore_id = gaudi2_edma_initiator_sft_id[module_idx].dcore_id;
via_sft = true;
sprintf(initiator_name, "EDMA_%u", module_idx);
break;
case RAZWI_PDMA:
rtr_id = gaudi2_pdma_initiator_rtr_id[module_idx];
sprintf(initiator_name, "PDMA_%u", module_idx);
break;
case RAZWI_NIC:
rtr_id = gaudi2_nic_initiator_rtr_id[module_idx];
sprintf(initiator_name, "NIC_%u", module_idx);
break;
case RAZWI_DEC:
rtr_id = gaudi2_dec_initiator_rtr_id[module_idx];
sprintf(initiator_name, "DEC_%u", module_idx);
break;
case RAZWI_ROT:
rtr_id = gaudi2_rot_initiator_rtr_id[module_idx];
sprintf(initiator_name, "ROT_%u", module_idx);
break;
default:
return;
}
if (!read_razwi_regs) {
if (le32_to_cpu(razwi_info->razwi_happened_mask) & RAZWI_HAPPENED_HBW) {
hbw_shrd_aw = le32_to_cpu(razwi_info->razwi_happened_mask) &
RAZWI_HAPPENED_AW;
hbw_shrd_ar = le32_to_cpu(razwi_info->razwi_happened_mask) &
RAZWI_HAPPENED_AR;
} else if (le32_to_cpu(razwi_info->razwi_happened_mask) & RAZWI_HAPPENED_LBW) {
lbw_shrd_aw = le32_to_cpu(razwi_info->razwi_happened_mask) &
RAZWI_HAPPENED_AW;
lbw_shrd_ar = le32_to_cpu(razwi_info->razwi_happened_mask) &
RAZWI_HAPPENED_AR;
}
rtr_mstr_if_base_addr = 0;
goto dump_info;
}
/* Find router mstr_if register base */
if (via_sft) {
rtr_mstr_if_base_addr = mmSFT0_HBW_RTR_IF0_RTR_CTRL_BASE +
dcore_id * SFT_DCORE_OFFSET +
sft_id * SFT_IF_OFFSET +
RTR_MSTR_IF_OFFSET;
} else {
dcore_id = rtr_id / NUM_OF_RTR_PER_DCORE;
dcore_rtr_id = rtr_id % NUM_OF_RTR_PER_DCORE;
rtr_mstr_if_base_addr = mmDCORE0_RTR0_CTRL_BASE +
dcore_id * DCORE_OFFSET +
dcore_rtr_id * DCORE_RTR_OFFSET +
RTR_MSTR_IF_OFFSET;
}
/* Find out event cause by reading "RAZWI_HAPPENED" registers */
hbw_shrd_aw = RREG32(rtr_mstr_if_base_addr + RR_SHRD_HBW_AW_RAZWI_HAPPENED);
hbw_shrd_ar = RREG32(rtr_mstr_if_base_addr + RR_SHRD_HBW_AR_RAZWI_HAPPENED);
if (via_sft) {
/* SFT has separate MSTR_IF for LBW, only there we can
* read the LBW razwi related registers
*/
u64 base;
base = mmSFT0_HBW_RTR_IF0_RTR_CTRL_BASE + dcore_id * SFT_DCORE_OFFSET +
RTR_LBW_MSTR_IF_OFFSET;
lbw_shrd_aw = RREG32(base + RR_SHRD_LBW_AW_RAZWI_HAPPENED);
lbw_shrd_ar = RREG32(base + RR_SHRD_LBW_AR_RAZWI_HAPPENED);
} else {
lbw_shrd_aw = RREG32(rtr_mstr_if_base_addr + RR_SHRD_LBW_AW_RAZWI_HAPPENED);
lbw_shrd_ar = RREG32(rtr_mstr_if_base_addr + RR_SHRD_LBW_AR_RAZWI_HAPPENED);
}
dump_info:
/* check if there is no RR razwi indication at all */
if (!hbw_shrd_aw && !hbw_shrd_ar && !lbw_shrd_aw && !lbw_shrd_ar)
return;
if (hbw_shrd_aw) {
gaudi2_razwi_rr_hbw_shared_printf_info(hdev, rtr_mstr_if_base_addr, true,
initiator_name, read_razwi_regs, razwi_info);
/* Clear event indication */
if (read_razwi_regs)
WREG32(rtr_mstr_if_base_addr + RR_SHRD_HBW_AW_RAZWI_HAPPENED, hbw_shrd_aw);
}
if (hbw_shrd_ar) {
gaudi2_razwi_rr_hbw_shared_printf_info(hdev, rtr_mstr_if_base_addr, false,
initiator_name, read_razwi_regs, razwi_info);
/* Clear event indication */
if (read_razwi_regs)
WREG32(rtr_mstr_if_base_addr + RR_SHRD_HBW_AR_RAZWI_HAPPENED, hbw_shrd_ar);
}
if (lbw_shrd_aw) {
gaudi2_razwi_rr_lbw_shared_printf_info(hdev, rtr_mstr_if_base_addr, true,
initiator_name, read_razwi_regs, razwi_info);
/* Clear event indication */
if (read_razwi_regs)
WREG32(rtr_mstr_if_base_addr + RR_SHRD_LBW_AW_RAZWI_HAPPENED, lbw_shrd_aw);
}
if (lbw_shrd_ar) {
gaudi2_razwi_rr_lbw_shared_printf_info(hdev, rtr_mstr_if_base_addr, false,
initiator_name, read_razwi_regs, razwi_info);
/* Clear event indication */
if (read_razwi_regs)
WREG32(rtr_mstr_if_base_addr + RR_SHRD_LBW_AR_RAZWI_HAPPENED, lbw_shrd_ar);
}
}
static void gaudi2_check_if_razwi_happened(struct hl_device *hdev)
{
struct asic_fixed_properties *prop = &hdev->asic_prop;
u8 mod_idx, sub_mod;
/* check all TPCs */
for (mod_idx = 0 ; mod_idx < (NUM_OF_TPC_PER_DCORE * NUM_OF_DCORES + 1) ; mod_idx++) {
if (prop->tpc_enabled_mask & BIT(mod_idx))
gaudi2_ack_module_razwi_event_handler(hdev, RAZWI_TPC, mod_idx, 0, NULL);
}
/* check all MMEs */
for (mod_idx = 0 ; mod_idx < (NUM_OF_MME_PER_DCORE * NUM_OF_DCORES) ; mod_idx++)
for (sub_mod = MME_WAP0 ; sub_mod < MME_INITIATORS_MAX ; sub_mod++)
gaudi2_ack_module_razwi_event_handler(hdev, RAZWI_MME, mod_idx,
sub_mod, NULL);
/* check all EDMAs */
for (mod_idx = 0 ; mod_idx < (NUM_OF_EDMA_PER_DCORE * NUM_OF_DCORES) ; mod_idx++)
if (prop->edma_enabled_mask & BIT(mod_idx))
gaudi2_ack_module_razwi_event_handler(hdev, RAZWI_EDMA, mod_idx, 0, NULL);
/* check all PDMAs */
for (mod_idx = 0 ; mod_idx < NUM_OF_PDMA ; mod_idx++)
gaudi2_ack_module_razwi_event_handler(hdev, RAZWI_PDMA, mod_idx, 0, NULL);
/* check all NICs */
for (mod_idx = 0 ; mod_idx < NIC_NUMBER_OF_PORTS ; mod_idx++)
if (hdev->nic_ports_mask & BIT(mod_idx))
gaudi2_ack_module_razwi_event_handler(hdev, RAZWI_NIC, mod_idx >> 1, 0,
NULL);
/* check all DECs */
for (mod_idx = 0 ; mod_idx < NUMBER_OF_DEC ; mod_idx++)
if (prop->decoder_enabled_mask & BIT(mod_idx))
gaudi2_ack_module_razwi_event_handler(hdev, RAZWI_DEC, mod_idx, 0, NULL);
/* check all ROTs */
for (mod_idx = 0 ; mod_idx < NUM_OF_ROT ; mod_idx++)
gaudi2_ack_module_razwi_event_handler(hdev, RAZWI_ROT, mod_idx, 0, NULL);
}
habanalabs/gaudi2: dump detailed information upon RAZWI In order to improve debuggability, we add all available information when a RAZWI event occur. Signed-off-by: Ofir Bitton <obitton@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-08-21 18:50:51 +08:00
static const char *gaudi2_get_initiators_name(u32 rtr_id)
{
switch (rtr_id) {
case DCORE0_RTR0:
return "DEC0/1/8/9, TPC24, PDMA0/1, PMMU, PCIE_IF, EDMA0/2, HMMU0/2/4/6, CPU";
case DCORE0_RTR1:
return "TPC0/1";
case DCORE0_RTR2:
return "TPC2/3";
case DCORE0_RTR3:
return "TPC4/5";
case DCORE0_RTR4:
return "MME0_SBTE0/1";
case DCORE0_RTR5:
return "MME0_WAP0/SBTE2";
case DCORE0_RTR6:
return "MME0_CTRL_WR/SBTE3";
case DCORE0_RTR7:
return "MME0_WAP1/CTRL_RD/SBTE4";
case DCORE1_RTR0:
return "MME1_WAP1/CTRL_RD/SBTE4";
case DCORE1_RTR1:
return "MME1_CTRL_WR/SBTE3";
case DCORE1_RTR2:
return "MME1_WAP0/SBTE2";
case DCORE1_RTR3:
return "MME1_SBTE0/1";
case DCORE1_RTR4:
return "TPC10/11";
case DCORE1_RTR5:
return "TPC8/9";
case DCORE1_RTR6:
return "TPC6/7";
case DCORE1_RTR7:
return "DEC2/3, NIC0/1/2/3/4, ARC_FARM, KDMA, EDMA1/3, HMMU1/3/5/7";
case DCORE2_RTR0:
return "DEC4/5, NIC5/6/7/8, EDMA4/6, HMMU8/10/12/14, ROT0";
case DCORE2_RTR1:
return "TPC16/17";
case DCORE2_RTR2:
return "TPC14/15";
case DCORE2_RTR3:
return "TPC12/13";
case DCORE2_RTR4:
return "MME2_SBTE0/1";
case DCORE2_RTR5:
return "MME2_WAP0/SBTE2";
case DCORE2_RTR6:
return "MME2_CTRL_WR/SBTE3";
case DCORE2_RTR7:
return "MME2_WAP1/CTRL_RD/SBTE4";
case DCORE3_RTR0:
return "MME3_WAP1/CTRL_RD/SBTE4";
case DCORE3_RTR1:
return "MME3_CTRL_WR/SBTE3";
case DCORE3_RTR2:
return "MME3_WAP0/SBTE2";
case DCORE3_RTR3:
return "MME3_SBTE0/1";
case DCORE3_RTR4:
return "TPC18/19";
case DCORE3_RTR5:
return "TPC20/21";
case DCORE3_RTR6:
return "TPC22/23";
case DCORE3_RTR7:
return "DEC6/7, NIC9/10/11, EDMA5/7, HMMU9/11/13/15, ROT1, PSOC";
default:
return "N/A";
}
}
static void gaudi2_razwi_unmapped_addr_hbw_printf_info(struct hl_device *hdev, u32 rtr_id,
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
u64 rtr_ctrl_base_addr, bool is_write)
{
u32 razwi_hi, razwi_lo;
if (is_write) {
razwi_hi = RREG32(rtr_ctrl_base_addr + DEC_RAZWI_HBW_AW_ADDR_HI);
razwi_lo = RREG32(rtr_ctrl_base_addr + DEC_RAZWI_HBW_AW_ADDR_LO);
/* Clear set indication */
WREG32(rtr_ctrl_base_addr + DEC_RAZWI_HBW_AW_SET, 0x1);
} else {
razwi_hi = RREG32(rtr_ctrl_base_addr + DEC_RAZWI_HBW_AR_ADDR_HI);
razwi_lo = RREG32(rtr_ctrl_base_addr + DEC_RAZWI_HBW_AR_ADDR_LO);
/* Clear set indication */
WREG32(rtr_ctrl_base_addr + DEC_RAZWI_HBW_AR_SET, 0x1);
}
habanalabs/gaudi2: dump detailed information upon RAZWI In order to improve debuggability, we add all available information when a RAZWI event occur. Signed-off-by: Ofir Bitton <obitton@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-08-21 18:50:51 +08:00
dev_err_ratelimited(hdev->dev,
"RAZWI PSOC unmapped HBW %s error, rtr id %u, address %#llx\n",
is_write ? "WR" : "RD", rtr_id, (u64)razwi_hi << 32 | razwi_lo);
dev_err_ratelimited(hdev->dev,
"Initiators: %s\n", gaudi2_get_initiators_name(rtr_id));
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
}
habanalabs/gaudi2: dump detailed information upon RAZWI In order to improve debuggability, we add all available information when a RAZWI event occur. Signed-off-by: Ofir Bitton <obitton@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-08-21 18:50:51 +08:00
static void gaudi2_razwi_unmapped_addr_lbw_printf_info(struct hl_device *hdev, u32 rtr_id,
u64 rtr_ctrl_base_addr, bool is_write)
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
{
u32 razwi_addr;
if (is_write) {
razwi_addr = RREG32(rtr_ctrl_base_addr + DEC_RAZWI_LBW_AW_ADDR);
/* Clear set indication */
WREG32(rtr_ctrl_base_addr + DEC_RAZWI_LBW_AW_SET, 0x1);
} else {
razwi_addr = RREG32(rtr_ctrl_base_addr + DEC_RAZWI_LBW_AR_ADDR);
/* Clear set indication */
WREG32(rtr_ctrl_base_addr + DEC_RAZWI_LBW_AR_SET, 0x1);
}
habanalabs/gaudi2: dump detailed information upon RAZWI In order to improve debuggability, we add all available information when a RAZWI event occur. Signed-off-by: Ofir Bitton <obitton@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-08-21 18:50:51 +08:00
dev_err_ratelimited(hdev->dev,
"RAZWI PSOC unmapped LBW %s error, rtr id %u, address %#x\n",
is_write ? "WR" : "RD", rtr_id, razwi_addr);
dev_err_ratelimited(hdev->dev,
"Initiators: %s\n", gaudi2_get_initiators_name(rtr_id));
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
}
/* PSOC RAZWI interrupt occurs only when trying to access a bad address */
static void gaudi2_ack_psoc_razwi_event_handler(struct hl_device *hdev)
{
u32 hbw_aw_set, hbw_ar_set, lbw_aw_set, lbw_ar_set, rtr_id, dcore_id, dcore_rtr_id, xy,
razwi_mask_info, razwi_intr = 0;
int rtr_map_arr_len = NUM_OF_RTR_PER_DCORE * NUM_OF_DCORES;
u64 rtr_ctrl_base_addr;
if (hdev->pldm || !(hdev->fw_components & FW_TYPE_LINUX)) {
razwi_intr = RREG32(mmPSOC_GLOBAL_CONF_RAZWI_INTERRUPT);
if (!razwi_intr)
return;
}
razwi_mask_info = RREG32(mmPSOC_GLOBAL_CONF_RAZWI_MASK_INFO);
habanalabs/gaudi2: dump detailed information upon RAZWI In order to improve debuggability, we add all available information when a RAZWI event occur. Signed-off-by: Ofir Bitton <obitton@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-08-21 18:50:51 +08:00
xy = FIELD_GET(PSOC_GLOBAL_CONF_RAZWI_MASK_INFO_AXUSER_L_MASK, razwi_mask_info);
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
dev_err_ratelimited(hdev->dev,
habanalabs/gaudi2: dump detailed information upon RAZWI In order to improve debuggability, we add all available information when a RAZWI event occur. Signed-off-by: Ofir Bitton <obitton@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-08-21 18:50:51 +08:00
"PSOC RAZWI interrupt: Mask %d, AR %d, AW %d, AXUSER_L 0x%x AXUSER_H 0x%x\n",
FIELD_GET(PSOC_GLOBAL_CONF_RAZWI_MASK_INFO_MASK_MASK, razwi_mask_info),
FIELD_GET(PSOC_GLOBAL_CONF_RAZWI_MASK_INFO_WAS_AR_MASK, razwi_mask_info),
FIELD_GET(PSOC_GLOBAL_CONF_RAZWI_MASK_INFO_WAS_AW_MASK, razwi_mask_info),
xy,
FIELD_GET(PSOC_GLOBAL_CONF_RAZWI_MASK_INFO_AXUSER_H_MASK, razwi_mask_info));
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
if (xy == 0) {
dev_err_ratelimited(hdev->dev,
"PSOC RAZWI interrupt: received event from 0 rtr coordinates\n");
goto clear;
}
/* Find router id by router coordinates */
for (rtr_id = 0 ; rtr_id < rtr_map_arr_len ; rtr_id++)
if (rtr_coordinates_to_rtr_id[rtr_id] == xy)
break;
if (rtr_id == rtr_map_arr_len) {
dev_err_ratelimited(hdev->dev,
"PSOC RAZWI interrupt: invalid rtr coordinates (0x%x)\n", xy);
goto clear;
}
/* Find router mstr_if register base */
dcore_id = rtr_id / NUM_OF_RTR_PER_DCORE;
dcore_rtr_id = rtr_id % NUM_OF_RTR_PER_DCORE;
rtr_ctrl_base_addr = mmDCORE0_RTR0_CTRL_BASE + dcore_id * DCORE_OFFSET +
dcore_rtr_id * DCORE_RTR_OFFSET;
hbw_aw_set = RREG32(rtr_ctrl_base_addr + DEC_RAZWI_HBW_AW_SET);
hbw_ar_set = RREG32(rtr_ctrl_base_addr + DEC_RAZWI_HBW_AR_SET);
lbw_aw_set = RREG32(rtr_ctrl_base_addr + DEC_RAZWI_LBW_AW_SET);
lbw_ar_set = RREG32(rtr_ctrl_base_addr + DEC_RAZWI_LBW_AR_SET);
if (hbw_aw_set)
habanalabs/gaudi2: dump detailed information upon RAZWI In order to improve debuggability, we add all available information when a RAZWI event occur. Signed-off-by: Ofir Bitton <obitton@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-08-21 18:50:51 +08:00
gaudi2_razwi_unmapped_addr_hbw_printf_info(hdev, rtr_id,
rtr_ctrl_base_addr, true);
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
if (hbw_ar_set)
habanalabs/gaudi2: dump detailed information upon RAZWI In order to improve debuggability, we add all available information when a RAZWI event occur. Signed-off-by: Ofir Bitton <obitton@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-08-21 18:50:51 +08:00
gaudi2_razwi_unmapped_addr_hbw_printf_info(hdev, rtr_id,
rtr_ctrl_base_addr, false);
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
if (lbw_aw_set)
habanalabs/gaudi2: dump detailed information upon RAZWI In order to improve debuggability, we add all available information when a RAZWI event occur. Signed-off-by: Ofir Bitton <obitton@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-08-21 18:50:51 +08:00
gaudi2_razwi_unmapped_addr_lbw_printf_info(hdev, rtr_id,
rtr_ctrl_base_addr, true);
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
if (lbw_ar_set)
habanalabs/gaudi2: dump detailed information upon RAZWI In order to improve debuggability, we add all available information when a RAZWI event occur. Signed-off-by: Ofir Bitton <obitton@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-08-21 18:50:51 +08:00
gaudi2_razwi_unmapped_addr_lbw_printf_info(hdev, rtr_id,
rtr_ctrl_base_addr, false);
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
clear:
/* Clear Interrupts only on pldm or if f/w doesn't handle interrupts */
if (hdev->pldm || !(hdev->fw_components & FW_TYPE_LINUX))
WREG32(mmPSOC_GLOBAL_CONF_RAZWI_INTERRUPT, razwi_intr);
}
static void _gaudi2_handle_qm_sei_err(struct hl_device *hdev, u64 qman_base)
{
u32 i, sts_val, sts_clr_val = 0;
sts_val = RREG32(qman_base + QM_SEI_STATUS_OFFSET);
for (i = 0 ; i < GAUDI2_NUM_OF_QM_SEI_ERR_CAUSE ; i++) {
if (sts_val & BIT(i)) {
dev_err_ratelimited(hdev->dev, "QM SEI. err cause: %s\n",
gaudi2_qm_sei_error_cause[i]);
sts_clr_val |= BIT(i);
}
}
WREG32(qman_base + QM_SEI_STATUS_OFFSET, sts_clr_val);
}
static void gaudi2_handle_qm_sei_err(struct hl_device *hdev, u16 event_type,
struct hl_eq_razwi_info *razwi_info)
{
u64 qman_base;
u8 index;
switch (event_type) {
case GAUDI2_EVENT_TPC0_AXI_ERR_RSP ... GAUDI2_EVENT_TPC23_AXI_ERR_RSP:
index = event_type - GAUDI2_EVENT_TPC0_AXI_ERR_RSP;
qman_base = mmDCORE0_TPC0_QM_BASE +
(index / NUM_OF_TPC_PER_DCORE) * DCORE_OFFSET +
(index % NUM_OF_TPC_PER_DCORE) * DCORE_TPC_OFFSET;
break;
case GAUDI2_EVENT_TPC24_AXI_ERR_RSP:
qman_base = mmDCORE0_TPC6_QM_BASE;
break;
case GAUDI2_EVENT_MME0_CTRL_AXI_ERROR_RESPONSE:
case GAUDI2_EVENT_MME1_CTRL_AXI_ERROR_RESPONSE:
case GAUDI2_EVENT_MME2_CTRL_AXI_ERROR_RESPONSE:
case GAUDI2_EVENT_MME3_CTRL_AXI_ERROR_RESPONSE:
index = (event_type - GAUDI2_EVENT_MME0_CTRL_AXI_ERROR_RESPONSE) /
(GAUDI2_EVENT_MME1_CTRL_AXI_ERROR_RESPONSE -
GAUDI2_EVENT_MME0_CTRL_AXI_ERROR_RESPONSE);
qman_base = mmDCORE0_MME_QM_BASE + index * DCORE_OFFSET;
break;
case GAUDI2_EVENT_PDMA_CH0_AXI_ERR_RSP:
case GAUDI2_EVENT_PDMA_CH1_AXI_ERR_RSP:
index = event_type - GAUDI2_EVENT_PDMA_CH0_AXI_ERR_RSP;
qman_base = mmPDMA0_QM_BASE + index * PDMA_OFFSET;
break;
case GAUDI2_EVENT_ROTATOR0_AXI_ERROR_RESPONSE:
case GAUDI2_EVENT_ROTATOR1_AXI_ERROR_RESPONSE:
index = event_type - GAUDI2_EVENT_ROTATOR0_AXI_ERROR_RESPONSE;
qman_base = mmROT0_QM_BASE + index * ROT_OFFSET;
break;
default:
return;
}
_gaudi2_handle_qm_sei_err(hdev, qman_base);
/* There is a single event per NIC macro, so should check its both QMAN blocks */
if (event_type >= GAUDI2_EVENT_NIC0_AXI_ERROR_RESPONSE &&
event_type <= GAUDI2_EVENT_NIC11_AXI_ERROR_RESPONSE)
_gaudi2_handle_qm_sei_err(hdev, qman_base + NIC_QM_OFFSET);
/* check if RAZWI happened */
if (razwi_info)
gaudi2_ack_module_razwi_event_handler(hdev, RAZWI_PDMA, 0, 0, razwi_info);
}
static void gaudi2_handle_qman_err(struct hl_device *hdev, u16 event_type)
{
u32 qid_base;
u64 qman_base;
char desc[32];
u8 index;
switch (event_type) {
case GAUDI2_EVENT_TPC0_QM ... GAUDI2_EVENT_TPC5_QM:
index = event_type - GAUDI2_EVENT_TPC0_QM;
qid_base = GAUDI2_QUEUE_ID_DCORE0_TPC_0_0 + index * QMAN_STREAMS;
qman_base = mmDCORE0_TPC0_QM_BASE + index * DCORE_TPC_OFFSET;
snprintf(desc, ARRAY_SIZE(desc), "DCORE0_TPC%d_QM", index);
break;
case GAUDI2_EVENT_TPC6_QM ... GAUDI2_EVENT_TPC11_QM:
index = event_type - GAUDI2_EVENT_TPC6_QM;
qid_base = GAUDI2_QUEUE_ID_DCORE1_TPC_0_0 + index * QMAN_STREAMS;
qman_base = mmDCORE1_TPC0_QM_BASE + index * DCORE_TPC_OFFSET;
snprintf(desc, ARRAY_SIZE(desc), "DCORE1_TPC%d_QM", index);
break;
case GAUDI2_EVENT_TPC12_QM ... GAUDI2_EVENT_TPC17_QM:
index = event_type - GAUDI2_EVENT_TPC12_QM;
qid_base = GAUDI2_QUEUE_ID_DCORE2_TPC_0_0 + index * QMAN_STREAMS;
qman_base = mmDCORE2_TPC0_QM_BASE + index * DCORE_TPC_OFFSET;
snprintf(desc, ARRAY_SIZE(desc), "DCORE2_TPC%d_QM", index);
break;
case GAUDI2_EVENT_TPC18_QM ... GAUDI2_EVENT_TPC23_QM:
index = event_type - GAUDI2_EVENT_TPC18_QM;
qid_base = GAUDI2_QUEUE_ID_DCORE3_TPC_0_0 + index * QMAN_STREAMS;
qman_base = mmDCORE3_TPC0_QM_BASE + index * DCORE_TPC_OFFSET;
snprintf(desc, ARRAY_SIZE(desc), "DCORE3_TPC%d_QM", index);
break;
case GAUDI2_EVENT_TPC24_QM:
qid_base = GAUDI2_QUEUE_ID_DCORE0_TPC_6_0;
qman_base = mmDCORE0_TPC6_QM_BASE;
snprintf(desc, ARRAY_SIZE(desc), "DCORE0_TPC6_QM");
break;
case GAUDI2_EVENT_MME0_QM:
qid_base = GAUDI2_QUEUE_ID_DCORE0_MME_0_0;
qman_base = mmDCORE0_MME_QM_BASE;
snprintf(desc, ARRAY_SIZE(desc), "DCORE0_MME_QM");
break;
case GAUDI2_EVENT_MME1_QM:
qid_base = GAUDI2_QUEUE_ID_DCORE1_MME_0_0;
qman_base = mmDCORE1_MME_QM_BASE;
snprintf(desc, ARRAY_SIZE(desc), "DCORE1_MME_QM");
break;
case GAUDI2_EVENT_MME2_QM:
qid_base = GAUDI2_QUEUE_ID_DCORE2_MME_0_0;
qman_base = mmDCORE2_MME_QM_BASE;
snprintf(desc, ARRAY_SIZE(desc), "DCORE2_MME_QM");
break;
case GAUDI2_EVENT_MME3_QM:
qid_base = GAUDI2_QUEUE_ID_DCORE3_MME_0_0;
qman_base = mmDCORE3_MME_QM_BASE;
snprintf(desc, ARRAY_SIZE(desc), "DCORE3_MME_QM");
break;
case GAUDI2_EVENT_HDMA0_QM:
qid_base = GAUDI2_QUEUE_ID_DCORE0_EDMA_0_0;
qman_base = mmDCORE0_EDMA0_QM_BASE;
snprintf(desc, ARRAY_SIZE(desc), "DCORE0_EDMA0_QM");
break;
case GAUDI2_EVENT_HDMA1_QM:
qid_base = GAUDI2_QUEUE_ID_DCORE0_EDMA_1_0;
qman_base = mmDCORE0_EDMA1_QM_BASE;
snprintf(desc, ARRAY_SIZE(desc), "DCORE0_EDMA1_QM");
break;
case GAUDI2_EVENT_HDMA2_QM:
qid_base = GAUDI2_QUEUE_ID_DCORE1_EDMA_0_0;
qman_base = mmDCORE1_EDMA0_QM_BASE;
snprintf(desc, ARRAY_SIZE(desc), "DCORE1_EDMA0_QM");
break;
case GAUDI2_EVENT_HDMA3_QM:
qid_base = GAUDI2_QUEUE_ID_DCORE1_EDMA_1_0;
qman_base = mmDCORE1_EDMA1_QM_BASE;
snprintf(desc, ARRAY_SIZE(desc), "DCORE1_EDMA1_QM");
break;
case GAUDI2_EVENT_HDMA4_QM:
qid_base = GAUDI2_QUEUE_ID_DCORE2_EDMA_0_0;
qman_base = mmDCORE2_EDMA0_QM_BASE;
snprintf(desc, ARRAY_SIZE(desc), "DCORE2_EDMA0_QM");
break;
case GAUDI2_EVENT_HDMA5_QM:
qid_base = GAUDI2_QUEUE_ID_DCORE2_EDMA_1_0;
qman_base = mmDCORE2_EDMA1_QM_BASE;
snprintf(desc, ARRAY_SIZE(desc), "DCORE2_EDMA1_QM");
break;
case GAUDI2_EVENT_HDMA6_QM:
qid_base = GAUDI2_QUEUE_ID_DCORE3_EDMA_0_0;
qman_base = mmDCORE3_EDMA0_QM_BASE;
snprintf(desc, ARRAY_SIZE(desc), "DCORE3_EDMA0_QM");
break;
case GAUDI2_EVENT_HDMA7_QM:
qid_base = GAUDI2_QUEUE_ID_DCORE3_EDMA_1_0;
qman_base = mmDCORE3_EDMA1_QM_BASE;
snprintf(desc, ARRAY_SIZE(desc), "DCORE3_EDMA1_QM");
break;
case GAUDI2_EVENT_PDMA0_QM:
qid_base = GAUDI2_QUEUE_ID_PDMA_0_0;
qman_base = mmPDMA0_QM_BASE;
snprintf(desc, ARRAY_SIZE(desc), "PDMA0_QM");
break;
case GAUDI2_EVENT_PDMA1_QM:
qid_base = GAUDI2_QUEUE_ID_PDMA_1_0;
qman_base = mmPDMA1_QM_BASE;
snprintf(desc, ARRAY_SIZE(desc), "PDMA1_QM");
break;
case GAUDI2_EVENT_ROTATOR0_ROT0_QM:
qid_base = GAUDI2_QUEUE_ID_ROT_0_0;
qman_base = mmROT0_QM_BASE;
snprintf(desc, ARRAY_SIZE(desc), "ROTATOR0_QM");
break;
case GAUDI2_EVENT_ROTATOR1_ROT1_QM:
qid_base = GAUDI2_QUEUE_ID_ROT_1_0;
qman_base = mmROT1_QM_BASE;
snprintf(desc, ARRAY_SIZE(desc), "ROTATOR1_QM");
break;
default:
return;
}
gaudi2_handle_qman_err_generic(hdev, desc, qman_base, qid_base);
/* Handle EDMA QM SEI here because there is no AXI error response event for EDMA */
if (event_type >= GAUDI2_EVENT_HDMA2_QM && event_type <= GAUDI2_EVENT_HDMA5_QM)
_gaudi2_handle_qm_sei_err(hdev, qman_base);
}
static void gaudi2_handle_arc_farm_sei_err(struct hl_device *hdev)
{
u32 i, sts_val, sts_clr_val = 0;
sts_val = RREG32(mmARC_FARM_ARC0_AUX_ARC_SEI_INTR_STS);
for (i = 0 ; i < GAUDI2_NUM_OF_ARC_SEI_ERR_CAUSE ; i++) {
if (sts_val & BIT(i)) {
dev_err_ratelimited(hdev->dev, "ARC SEI. err cause: %s\n",
gaudi2_arc_sei_error_cause[i]);
sts_clr_val |= BIT(i);
}
}
WREG32(mmARC_FARM_ARC0_AUX_ARC_SEI_INTR_CLR, sts_clr_val);
}
static void gaudi2_handle_cpu_sei_err(struct hl_device *hdev)
{
u32 i, sts_val, sts_clr_val = 0;
sts_val = RREG32(mmCPU_IF_CPU_SEI_INTR_STS);
for (i = 0 ; i < GAUDI2_NUM_OF_CPU_SEI_ERR_CAUSE ; i++) {
if (sts_val & BIT(i)) {
dev_err_ratelimited(hdev->dev, "CPU SEI. err cause: %s\n",
gaudi2_cpu_sei_error_cause[i]);
sts_clr_val |= BIT(i);
}
}
WREG32(mmCPU_IF_CPU_SEI_INTR_CLR, sts_clr_val);
}
static void gaudi2_handle_rot_err(struct hl_device *hdev, u8 rot_index,
struct hl_eq_razwi_with_intr_cause *razwi_with_intr_cause)
{
u64 intr_cause_data = le64_to_cpu(razwi_with_intr_cause->intr_cause.intr_cause_data);
int i;
for (i = 0 ; i < GAUDI2_NUM_OF_ROT_ERR_CAUSE ; i++)
if (intr_cause_data & BIT(i))
dev_err_ratelimited(hdev->dev, "ROT%u. err cause: %s\n",
rot_index, guadi2_rot_error_cause[i]);
/* check if RAZWI happened */
gaudi2_ack_module_razwi_event_handler(hdev, RAZWI_ROT, rot_index, 0,
&razwi_with_intr_cause->razwi_info);
}
static void gaudi2_tpc_ack_interrupts(struct hl_device *hdev, u8 tpc_index, char *interrupt_name,
struct hl_eq_razwi_with_intr_cause *razwi_with_intr_cause)
{
u64 intr_cause_data = le64_to_cpu(razwi_with_intr_cause->intr_cause.intr_cause_data);
int i;
for (i = 0 ; i < GAUDI2_NUM_OF_TPC_INTR_CAUSE ; i++)
if (intr_cause_data & BIT(i))
dev_err_ratelimited(hdev->dev, "TPC%d_%s interrupt cause: %s\n",
tpc_index, interrupt_name, gaudi2_tpc_interrupts_cause[i]);
/* check if RAZWI happened */
gaudi2_ack_module_razwi_event_handler(hdev, RAZWI_TPC, tpc_index, 0,
&razwi_with_intr_cause->razwi_info);
}
static void gaudi2_handle_dec_err(struct hl_device *hdev, u8 dec_index, const char *interrupt_name,
struct hl_eq_razwi_info *razwi_info)
{
u32 sts_addr, sts_val, sts_clr_val = 0;
int i;
if (dec_index < NUM_OF_VDEC_PER_DCORE * NUM_OF_DCORES)
/* DCORE DEC */
sts_addr = mmDCORE0_VDEC0_BRDG_CTRL_CAUSE_INTR +
DCORE_OFFSET * (dec_index / NUM_OF_DEC_PER_DCORE) +
DCORE_VDEC_OFFSET * (dec_index % NUM_OF_DEC_PER_DCORE);
else
/* PCIE DEC */
sts_addr = mmPCIE_VDEC0_BRDG_CTRL_CAUSE_INTR + PCIE_VDEC_OFFSET *
(dec_index - NUM_OF_VDEC_PER_DCORE * NUM_OF_DCORES);
sts_val = RREG32(sts_addr);
for (i = 0 ; i < GAUDI2_NUM_OF_DEC_ERR_CAUSE ; i++) {
if (sts_val & BIT(i)) {
dev_err_ratelimited(hdev->dev, "DEC%u_%s err cause: %s\n",
dec_index, interrupt_name, gaudi2_dec_error_cause[i]);
sts_clr_val |= BIT(i);
}
}
/* check if RAZWI happened */
gaudi2_ack_module_razwi_event_handler(hdev, RAZWI_DEC, dec_index, 0, razwi_info);
/* Write 1 clear errors */
WREG32(sts_addr, sts_clr_val);
}
static void gaudi2_handle_mme_err(struct hl_device *hdev, u8 mme_index, const char *interrupt_name,
struct hl_eq_razwi_info *razwi_info)
{
u32 sts_addr, sts_val, sts_clr_addr, sts_clr_val = 0;
int i;
sts_addr = mmDCORE0_MME_CTRL_LO_INTR_CAUSE + DCORE_OFFSET * mme_index;
sts_clr_addr = mmDCORE0_MME_CTRL_LO_INTR_CLEAR + DCORE_OFFSET * mme_index;
sts_val = RREG32(sts_addr);
for (i = 0 ; i < GAUDI2_NUM_OF_MME_ERR_CAUSE ; i++) {
if (sts_val & BIT(i)) {
dev_err_ratelimited(hdev->dev, "MME%u_%s err cause: %s\n",
mme_index, interrupt_name, guadi2_mme_error_cause[i]);
sts_clr_val |= BIT(i);
}
}
/* check if RAZWI happened */
for (i = MME_WRITE ; i < MME_INITIATORS_MAX ; i++)
gaudi2_ack_module_razwi_event_handler(hdev, RAZWI_MME, mme_index, i, razwi_info);
WREG32(sts_clr_addr, sts_clr_val);
}
static void gaudi2_handle_mme_sbte_err(struct hl_device *hdev, u8 mme_index, u8 sbte_index,
u64 intr_cause_data)
{
int i;
for (i = 0 ; i < GAUDI2_NUM_OF_MME_SBTE_ERR_CAUSE ; i++)
if (intr_cause_data & BIT(i))
dev_err_ratelimited(hdev->dev, "MME%uSBTE%u_AXI_ERR_RSP err cause: %s\n",
mme_index, sbte_index, guadi2_mme_sbte_error_cause[i]);
}
static void gaudi2_handle_mme_wap_err(struct hl_device *hdev, u8 mme_index,
struct hl_eq_razwi_info *razwi_info)
{
u32 sts_addr, sts_val, sts_clr_addr, sts_clr_val = 0;
int i;
sts_addr = mmDCORE0_MME_ACC_INTR_CAUSE + DCORE_OFFSET * mme_index;
sts_clr_addr = mmDCORE0_MME_ACC_INTR_CLEAR + DCORE_OFFSET * mme_index;
sts_val = RREG32(sts_addr);
for (i = 0 ; i < GAUDI2_NUM_OF_MME_WAP_ERR_CAUSE ; i++) {
if (sts_val & BIT(i)) {
dev_err_ratelimited(hdev->dev,
"MME%u_WAP_SOURCE_RESULT_INVALID err cause: %s\n",
mme_index, guadi2_mme_wap_error_cause[i]);
sts_clr_val |= BIT(i);
}
}
/* check if RAZWI happened on WAP0/1 */
gaudi2_ack_module_razwi_event_handler(hdev, RAZWI_MME, mme_index, MME_WAP0, razwi_info);
gaudi2_ack_module_razwi_event_handler(hdev, RAZWI_MME, mme_index, MME_WAP1, razwi_info);
WREG32(sts_clr_addr, sts_clr_val);
}
static void gaudi2_handle_kdma_core_event(struct hl_device *hdev, u64 intr_cause_data)
{
int i;
/* If an AXI read or write error is received, an error is reported and
* interrupt message is sent. Due to an HW errata, when reading the cause
* register of the KDMA engine, the reported error is always HBW even if
* the actual error caused by a LBW KDMA transaction.
*/
for (i = 0 ; i < GAUDI2_NUM_OF_DMA_CORE_INTR_CAUSE ; i++)
if (intr_cause_data & BIT(i))
dev_err_ratelimited(hdev->dev, "kdma core err cause: %s\n",
gaudi2_kdma_core_interrupts_cause[i]);
}
static void gaudi2_handle_dma_core_event(struct hl_device *hdev, u64 intr_cause_data)
{
int i;
for (i = 0 ; i < GAUDI2_NUM_OF_DMA_CORE_INTR_CAUSE ; i++)
if (intr_cause_data & BIT(i))
dev_err_ratelimited(hdev->dev, "dma core err cause: %s\n",
gaudi2_dma_core_interrupts_cause[i]);
}
habanalabs/gaudi2: print RAZWI info upon PCIe access error Add the dump of the RAZWI information when a PCIe access is blocked by RR. Signed-off-by: Tomer Tayar <ttayar@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-09-05 21:24:21 +08:00
static void gaudi2_print_pcie_mstr_rr_mstr_if_razwi_info(struct hl_device *hdev)
{
u32 mstr_if_base_addr = mmPCIE_MSTR_RR_MSTR_IF_RR_SHRD_HBW_BASE, razwi_happened_addr;
razwi_happened_addr = mstr_if_base_addr + RR_SHRD_HBW_AW_RAZWI_HAPPENED;
if (RREG32(razwi_happened_addr)) {
gaudi2_razwi_rr_hbw_shared_printf_info(hdev, mstr_if_base_addr, true, "PCIE", true,
NULL);
WREG32(razwi_happened_addr, 0x1);
}
razwi_happened_addr = mstr_if_base_addr + RR_SHRD_HBW_AR_RAZWI_HAPPENED;
if (RREG32(razwi_happened_addr)) {
gaudi2_razwi_rr_hbw_shared_printf_info(hdev, mstr_if_base_addr, false, "PCIE", true,
NULL);
WREG32(razwi_happened_addr, 0x1);
}
razwi_happened_addr = mstr_if_base_addr + RR_SHRD_LBW_AW_RAZWI_HAPPENED;
if (RREG32(razwi_happened_addr)) {
gaudi2_razwi_rr_lbw_shared_printf_info(hdev, mstr_if_base_addr, true, "PCIE", true,
NULL);
WREG32(razwi_happened_addr, 0x1);
}
razwi_happened_addr = mstr_if_base_addr + RR_SHRD_LBW_AR_RAZWI_HAPPENED;
if (RREG32(razwi_happened_addr)) {
gaudi2_razwi_rr_lbw_shared_printf_info(hdev, mstr_if_base_addr, false, "PCIE", true,
NULL);
WREG32(razwi_happened_addr, 0x1);
}
}
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
static void gaudi2_print_pcie_addr_dec_info(struct hl_device *hdev, u64 intr_cause_data)
{
int i;
habanalabs/gaudi2: print RAZWI info upon PCIe access error Add the dump of the RAZWI information when a PCIe access is blocked by RR. Signed-off-by: Tomer Tayar <ttayar@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-09-05 21:24:21 +08:00
for (i = 0 ; i < GAUDI2_NUM_OF_PCIE_ADDR_DEC_ERR_CAUSE ; i++) {
if (!(intr_cause_data & BIT_ULL(i)))
continue;
dev_err_ratelimited(hdev->dev, "PCIE ADDR DEC Error: %s\n",
gaudi2_pcie_addr_dec_error_cause[i]);
switch (intr_cause_data & BIT_ULL(i)) {
case PCIE_WRAP_PCIE_IC_SEI_INTR_IND_AXI_LBW_ERR_INTR_MASK:
break;
case PCIE_WRAP_PCIE_IC_SEI_INTR_IND_BAD_ACCESS_INTR_MASK:
gaudi2_print_pcie_mstr_rr_mstr_if_razwi_info(hdev);
break;
}
}
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
}
static void gaudi2_handle_pif_fatal(struct hl_device *hdev, u64 intr_cause_data)
{
int i;
for (i = 0 ; i < GAUDI2_NUM_OF_PMMU_FATAL_ERR_CAUSE ; i++) {
if (intr_cause_data & BIT_ULL(i))
dev_err_ratelimited(hdev->dev, "PMMU PIF err cause: %s\n",
gaudi2_pmmu_fatal_interrupts_cause[i]);
}
}
static void gaudi2_handle_hif_fatal(struct hl_device *hdev, u16 event_type, u64 intr_cause_data)
{
u32 dcore_id, hif_id;
int i;
dcore_id = (event_type - GAUDI2_EVENT_HIF0_FATAL) / 4;
hif_id = (event_type - GAUDI2_EVENT_HIF0_FATAL) % 4;
for (i = 0 ; i < GAUDI2_NUM_OF_HIF_FATAL_ERR_CAUSE ; i++) {
if (intr_cause_data & BIT_ULL(i))
dev_err_ratelimited(hdev->dev, "DCORE%u_HIF%u: %s\n", dcore_id, hif_id,
gaudi2_hif_fatal_interrupts_cause[i]);
}
}
static void gaudi2_handle_page_error(struct hl_device *hdev, u64 mmu_base, bool is_pmmu)
{
u32 valid, val;
u64 addr;
valid = RREG32(mmu_base + MMU_OFFSET(mmDCORE0_HMMU0_MMU_ACCESS_PAGE_ERROR_VALID));
if (!(valid & DCORE0_HMMU0_MMU_ACCESS_PAGE_ERROR_VALID_PAGE_ERR_VALID_ENTRY_MASK))
return;
val = RREG32(mmu_base + MMU_OFFSET(mmDCORE0_HMMU0_MMU_PAGE_ERROR_CAPTURE));
addr = val & DCORE0_HMMU0_MMU_PAGE_ERROR_CAPTURE_VA_63_32_MASK;
addr <<= 32;
addr |= RREG32(mmu_base + MMU_OFFSET(mmDCORE0_HMMU0_MMU_PAGE_ERROR_CAPTURE_VA));
dev_err_ratelimited(hdev->dev, "%s page fault on va 0x%llx\n",
is_pmmu ? "PMMU" : "HMMU", addr);
WREG32(mmu_base + MMU_OFFSET(mmDCORE0_HMMU0_MMU_PAGE_ERROR_CAPTURE), 0);
}
static void gaudi2_handle_access_error(struct hl_device *hdev, u64 mmu_base, bool is_pmmu)
{
u32 valid, val;
u64 addr;
valid = RREG32(mmu_base + MMU_OFFSET(mmDCORE0_HMMU0_MMU_ACCESS_PAGE_ERROR_VALID));
if (!(valid & DCORE0_HMMU0_MMU_ACCESS_PAGE_ERROR_VALID_ACCESS_ERR_VALID_ENTRY_MASK))
return;
val = RREG32(mmu_base + MMU_OFFSET(mmDCORE0_HMMU0_MMU_ACCESS_ERROR_CAPTURE));
addr = val & DCORE0_HMMU0_MMU_ACCESS_ERROR_CAPTURE_VA_63_32_MASK;
addr <<= 32;
addr |= RREG32(mmu_base + MMU_OFFSET(mmDCORE0_HMMU0_MMU_ACCESS_ERROR_CAPTURE_VA));
dev_err_ratelimited(hdev->dev, "%s access error on va 0x%llx\n",
is_pmmu ? "PMMU" : "HMMU", addr);
WREG32(mmu_base + MMU_OFFSET(mmDCORE0_HMMU0_MMU_ACCESS_ERROR_CAPTURE), 0);
}
static void gaudi2_handle_mmu_spi_sei_generic(struct hl_device *hdev, const char *mmu_name,
u64 mmu_base, bool is_pmmu)
{
u32 spi_sei_cause, interrupt_clr = 0x0;
int i;
spi_sei_cause = RREG32(mmu_base + MMU_SPI_SEI_CAUSE_OFFSET);
for (i = 0 ; i < GAUDI2_NUM_OF_MMU_SPI_SEI_CAUSE ; i++) {
if (spi_sei_cause & BIT(i)) {
dev_err_ratelimited(hdev->dev, "%s SPI_SEI ERR. err cause: %s\n",
mmu_name, gaudi2_mmu_spi_sei[i].cause);
if (i == 0)
gaudi2_handle_page_error(hdev, mmu_base, is_pmmu);
else if (i == 1)
gaudi2_handle_access_error(hdev, mmu_base, is_pmmu);
if (gaudi2_mmu_spi_sei[i].clear_bit >= 0)
interrupt_clr |= BIT(gaudi2_mmu_spi_sei[i].clear_bit);
}
}
/* Clear cause */
WREG32_AND(mmu_base + MMU_SPI_SEI_CAUSE_OFFSET, ~spi_sei_cause);
/* Clear interrupt */
WREG32(mmu_base + MMU_INTERRUPT_CLR_OFFSET, interrupt_clr);
}
static bool gaudi2_handle_sm_err(struct hl_device *hdev, u8 sm_index)
{
u32 sei_cause_addr, sei_cause_val, sei_cause_cause, sei_cause_log;
u32 cq_intr_addr, cq_intr_val, cq_intr_queue_index;
bool reset = true;
int i;
sei_cause_addr = mmDCORE0_SYNC_MNGR_GLBL_SM_SEI_CAUSE + DCORE_OFFSET * sm_index;
cq_intr_addr = mmDCORE0_SYNC_MNGR_GLBL_CQ_INTR + DCORE_OFFSET * sm_index;
sei_cause_val = RREG32(sei_cause_addr);
sei_cause_cause = FIELD_GET(DCORE0_SYNC_MNGR_GLBL_SM_SEI_CAUSE_CAUSE_MASK, sei_cause_val);
cq_intr_val = RREG32(cq_intr_addr);
/* SEI interrupt */
if (sei_cause_cause) {
/* There are corresponding SEI_CAUSE_log bits for every SEI_CAUSE_cause bit */
sei_cause_log = FIELD_GET(DCORE0_SYNC_MNGR_GLBL_SM_SEI_CAUSE_LOG_MASK,
sei_cause_val);
for (i = 0 ; i < GAUDI2_NUM_OF_SM_SEI_ERR_CAUSE ; i++) {
if (!(sei_cause_cause & BIT(i)))
continue;
dev_err_ratelimited(hdev->dev, "SM%u SEI ERR. err cause: %s. %s: 0x%X\n",
sm_index,
gaudi2_sm_sei_cause[i].cause_name,
gaudi2_sm_sei_cause[i].log_name,
sei_cause_log & gaudi2_sm_sei_cause[i].log_mask);
/* Due to a potential H/W issue, do not reset upon BRESP errors */
if (i == 2)
reset = false;
break;
}
/* Clear SM_SEI_CAUSE */
WREG32(sei_cause_addr, 0);
}
/* CQ interrupt */
if (cq_intr_val & DCORE0_SYNC_MNGR_GLBL_CQ_INTR_CQ_SEC_INTR_MASK) {
cq_intr_queue_index =
FIELD_GET(DCORE0_SYNC_MNGR_GLBL_CQ_INTR_CQ_INTR_QUEUE_INDEX_MASK,
cq_intr_val);
dev_err_ratelimited(hdev->dev, "SM%u err. err cause: CQ_INTR. queue index: %u\n",
sm_index, cq_intr_queue_index);
/* Clear CQ_INTR */
WREG32(cq_intr_addr, 0);
}
return reset;
}
static void gaudi2_handle_mmu_spi_sei_err(struct hl_device *hdev, u16 event_type)
{
bool is_pmmu = false;
char desc[32];
u64 mmu_base;
u8 index;
switch (event_type) {
case GAUDI2_EVENT_HMMU0_PAGE_FAULT_OR_WR_PERM ... GAUDI2_EVENT_HMMU3_SECURITY_ERROR:
index = (event_type - GAUDI2_EVENT_HMMU0_PAGE_FAULT_OR_WR_PERM) / 3;
mmu_base = mmDCORE0_HMMU0_MMU_BASE + index * DCORE_HMMU_OFFSET;
snprintf(desc, ARRAY_SIZE(desc), "DCORE0_HMMU%d", index);
break;
case GAUDI2_EVENT_HMMU_0_AXI_ERR_RSP ... GAUDI2_EVENT_HMMU_3_AXI_ERR_RSP:
index = (event_type - GAUDI2_EVENT_HMMU_0_AXI_ERR_RSP);
mmu_base = mmDCORE0_HMMU0_MMU_BASE + index * DCORE_HMMU_OFFSET;
snprintf(desc, ARRAY_SIZE(desc), "DCORE0_HMMU%d", index);
break;
case GAUDI2_EVENT_HMMU8_PAGE_FAULT_WR_PERM ... GAUDI2_EVENT_HMMU11_SECURITY_ERROR:
index = (event_type - GAUDI2_EVENT_HMMU8_PAGE_FAULT_WR_PERM) / 3;
mmu_base = mmDCORE1_HMMU0_MMU_BASE + index * DCORE_HMMU_OFFSET;
snprintf(desc, ARRAY_SIZE(desc), "DCORE1_HMMU%d", index);
break;
case GAUDI2_EVENT_HMMU_8_AXI_ERR_RSP ... GAUDI2_EVENT_HMMU_11_AXI_ERR_RSP:
index = (event_type - GAUDI2_EVENT_HMMU_8_AXI_ERR_RSP);
mmu_base = mmDCORE1_HMMU0_MMU_BASE + index * DCORE_HMMU_OFFSET;
snprintf(desc, ARRAY_SIZE(desc), "DCORE1_HMMU%d", index);
break;
case GAUDI2_EVENT_HMMU7_PAGE_FAULT_WR_PERM ... GAUDI2_EVENT_HMMU4_SECURITY_ERROR:
index = (event_type - GAUDI2_EVENT_HMMU7_PAGE_FAULT_WR_PERM) / 3;
mmu_base = mmDCORE2_HMMU0_MMU_BASE + index * DCORE_HMMU_OFFSET;
snprintf(desc, ARRAY_SIZE(desc), "DCORE2_HMMU%d", index);
break;
case GAUDI2_EVENT_HMMU_7_AXI_ERR_RSP ... GAUDI2_EVENT_HMMU_4_AXI_ERR_RSP:
index = (event_type - GAUDI2_EVENT_HMMU_7_AXI_ERR_RSP);
mmu_base = mmDCORE2_HMMU0_MMU_BASE + index * DCORE_HMMU_OFFSET;
snprintf(desc, ARRAY_SIZE(desc), "DCORE2_HMMU%d", index);
break;
case GAUDI2_EVENT_HMMU15_PAGE_FAULT_WR_PERM ... GAUDI2_EVENT_HMMU12_SECURITY_ERROR:
index = (event_type - GAUDI2_EVENT_HMMU15_PAGE_FAULT_WR_PERM) / 3;
mmu_base = mmDCORE3_HMMU0_MMU_BASE + index * DCORE_HMMU_OFFSET;
snprintf(desc, ARRAY_SIZE(desc), "DCORE3_HMMU%d", index);
break;
case GAUDI2_EVENT_HMMU_15_AXI_ERR_RSP ... GAUDI2_EVENT_HMMU_12_AXI_ERR_RSP:
index = (event_type - GAUDI2_EVENT_HMMU_15_AXI_ERR_RSP);
mmu_base = mmDCORE3_HMMU0_MMU_BASE + index * DCORE_HMMU_OFFSET;
snprintf(desc, ARRAY_SIZE(desc), "DCORE3_HMMU%d", index);
break;
case GAUDI2_EVENT_PMMU0_PAGE_FAULT_WR_PERM ... GAUDI2_EVENT_PMMU0_SECURITY_ERROR:
case GAUDI2_EVENT_PMMU_AXI_ERR_RSP_0:
is_pmmu = true;
mmu_base = mmPMMU_HBW_MMU_BASE;
snprintf(desc, ARRAY_SIZE(desc), "PMMU");
break;
default:
return;
}
gaudi2_handle_mmu_spi_sei_generic(hdev, desc, mmu_base, is_pmmu);
}
/* returns true if hard reset is required (ECC DERR or Read parity), false otherwise (ECC SERR) */
static bool gaudi2_hbm_sei_handle_read_err(struct hl_device *hdev,
struct hl_eq_hbm_sei_read_err_intr_info *rd_err_data, u32 err_cnt)
{
u32 addr, beat, beat_shift;
bool rc = false;
dev_err_ratelimited(hdev->dev,
"READ ERROR count: ECC SERR: %d, ECC DERR: %d, RD_PARITY: %d\n",
FIELD_GET(HBM_ECC_SERR_CNTR_MASK, err_cnt),
FIELD_GET(HBM_ECC_DERR_CNTR_MASK, err_cnt),
FIELD_GET(HBM_RD_PARITY_CNTR_MASK, err_cnt));
addr = le32_to_cpu(rd_err_data->dbg_rd_err_addr.rd_addr_val);
dev_err_ratelimited(hdev->dev,
"READ ERROR address: sid(%u), bg(%u), ba(%u), col(%u), row(%u)\n",
FIELD_GET(HBM_RD_ADDR_SID_MASK, addr),
FIELD_GET(HBM_RD_ADDR_BG_MASK, addr),
FIELD_GET(HBM_RD_ADDR_BA_MASK, addr),
FIELD_GET(HBM_RD_ADDR_COL_MASK, addr),
FIELD_GET(HBM_RD_ADDR_ROW_MASK, addr));
/* For each beat (RDQS edge), look for possible errors and print relevant info */
for (beat = 0 ; beat < 4 ; beat++) {
if (le32_to_cpu(rd_err_data->dbg_rd_err_misc) &
(HBM_RD_ERR_SERR_BEAT0_MASK << beat))
dev_err_ratelimited(hdev->dev, "Beat%d ECC SERR: DM: %#x, Syndrome: %#x\n",
beat,
le32_to_cpu(rd_err_data->dbg_rd_err_dm),
le32_to_cpu(rd_err_data->dbg_rd_err_syndrome));
if (le32_to_cpu(rd_err_data->dbg_rd_err_misc) &
(HBM_RD_ERR_DERR_BEAT0_MASK << beat)) {
dev_err_ratelimited(hdev->dev, "Beat%d ECC DERR: DM: %#x, Syndrome: %#x\n",
beat,
le32_to_cpu(rd_err_data->dbg_rd_err_dm),
le32_to_cpu(rd_err_data->dbg_rd_err_syndrome));
rc |= true;
}
beat_shift = beat * HBM_RD_ERR_BEAT_SHIFT;
if (le32_to_cpu(rd_err_data->dbg_rd_err_misc) &
(HBM_RD_ERR_PAR_ERR_BEAT0_MASK << beat_shift)) {
dev_err_ratelimited(hdev->dev,
"Beat%d read PARITY: DM: %#x, PAR data: %#x\n",
beat,
le32_to_cpu(rd_err_data->dbg_rd_err_dm),
(le32_to_cpu(rd_err_data->dbg_rd_err_misc) &
(HBM_RD_ERR_PAR_DATA_BEAT0_MASK << beat_shift)) >>
(HBM_RD_ERR_PAR_DATA_BEAT0_SHIFT + beat_shift));
rc |= true;
}
dev_err_ratelimited(hdev->dev, "Beat%d DQ data:\n", beat);
dev_err_ratelimited(hdev->dev, "\t0x%08x\n",
le32_to_cpu(rd_err_data->dbg_rd_err_data[beat * 2]));
dev_err_ratelimited(hdev->dev, "\t0x%08x\n",
le32_to_cpu(rd_err_data->dbg_rd_err_data[beat * 2 + 1]));
}
return rc;
}
static void gaudi2_hbm_sei_print_wr_par_info(struct hl_device *hdev,
struct hl_eq_hbm_sei_wr_par_intr_info *wr_par_err_data, u32 err_cnt)
{
struct hbm_sei_wr_cmd_address *wr_cmd_addr = wr_par_err_data->dbg_last_wr_cmds;
u32 i, curr_addr, derr = wr_par_err_data->dbg_derr;
dev_err_ratelimited(hdev->dev, "WRITE PARITY ERROR count: %d\n", err_cnt);
dev_err_ratelimited(hdev->dev, "CK-0 DERR: 0x%02x, CK-1 DERR: 0x%02x\n",
derr & 0x3, derr & 0xc);
/* JIRA H6-3286 - the following prints may not be valid */
dev_err_ratelimited(hdev->dev, "Last latched write commands addresses:\n");
for (i = 0 ; i < HBM_WR_PAR_CMD_LIFO_LEN ; i++) {
curr_addr = le32_to_cpu(wr_cmd_addr[i].dbg_wr_cmd_addr);
dev_err_ratelimited(hdev->dev,
"\twrite cmd[%u]: Address: SID(%u) BG(%u) BA(%u) COL(%u).\n",
i,
FIELD_GET(WR_PAR_LAST_CMD_SID_MASK, curr_addr),
FIELD_GET(WR_PAR_LAST_CMD_BG_MASK, curr_addr),
FIELD_GET(WR_PAR_LAST_CMD_BA_MASK, curr_addr),
FIELD_GET(WR_PAR_LAST_CMD_COL_MASK, curr_addr));
}
}
static void gaudi2_hbm_sei_print_ca_par_info(struct hl_device *hdev,
struct hl_eq_hbm_sei_ca_par_intr_info *ca_par_err_data, u32 err_cnt)
{
__le32 *col_cmd = ca_par_err_data->dbg_col;
__le16 *row_cmd = ca_par_err_data->dbg_row;
u32 i;
dev_err_ratelimited(hdev->dev, "CA ERROR count: %d\n", err_cnt);
dev_err_ratelimited(hdev->dev, "Last latched C&R bus commands:\n");
for (i = 0 ; i < HBM_CA_ERR_CMD_LIFO_LEN ; i++)
dev_err_ratelimited(hdev->dev, "cmd%u: ROW(0x%04x) COL(0x%05x)\n", i,
le16_to_cpu(row_cmd[i]) & (u16)GENMASK(13, 0),
le32_to_cpu(col_cmd[i]) & (u32)GENMASK(17, 0));
}
/* Returns true if hard reset is needed or false otherwise */
static bool gaudi2_handle_hbm_mc_sei_err(struct hl_device *hdev, u16 event_type,
struct hl_eq_hbm_sei_data *sei_data)
{
bool require_hard_reset = false;
u32 hbm_id, mc_id, cause_idx;
hbm_id = (event_type - GAUDI2_EVENT_HBM0_MC0_SEI_SEVERE) / 4;
mc_id = ((event_type - GAUDI2_EVENT_HBM0_MC0_SEI_SEVERE) / 2) % 2;
cause_idx = sei_data->hdr.sei_cause;
if (cause_idx > GAUDI2_NUM_OF_HBM_SEI_CAUSE - 1) {
dev_err_ratelimited(hdev->dev, "Invalid HBM SEI event cause (%d) provided by FW\n",
cause_idx);
return true;
}
habanalabs/gaudi2: log critical events with no rate limit When we have a storm of errors of HBM ECC SERR we can reach a situation where driver start hard reset flow without logging the error cause that caused the hard reset due to logs rate limiting. Signed-off-by: farah kassabri <fkassabri@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-08-23 22:41:52 +08:00
if (sei_data->hdr.is_critical)
dev_err(hdev->dev,
"System Critical Error Interrupt - HBM(%u) MC(%u) MC_CH(%u) MC_PC(%u). Error cause: %s\n",
hbm_id, mc_id, sei_data->hdr.mc_channel, sei_data->hdr.mc_pseudo_channel,
hbm_mc_sei_cause[cause_idx]);
else
dev_err_ratelimited(hdev->dev,
"System Non-Critical Error Interrupt - HBM(%u) MC(%u) MC_CH(%u) MC_PC(%u). Error cause: %s\n",
hbm_id, mc_id, sei_data->hdr.mc_channel, sei_data->hdr.mc_pseudo_channel,
hbm_mc_sei_cause[cause_idx]);
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
/* Print error-specific info */
switch (cause_idx) {
case HBM_SEI_CATTRIP:
require_hard_reset = true;
break;
case HBM_SEI_CMD_PARITY_EVEN:
gaudi2_hbm_sei_print_ca_par_info(hdev, &sei_data->ca_parity_even_info,
le32_to_cpu(sei_data->hdr.cnt));
require_hard_reset = true;
break;
case HBM_SEI_CMD_PARITY_ODD:
gaudi2_hbm_sei_print_ca_par_info(hdev, &sei_data->ca_parity_odd_info,
le32_to_cpu(sei_data->hdr.cnt));
require_hard_reset = true;
break;
case HBM_SEI_WRITE_DATA_PARITY_ERR:
gaudi2_hbm_sei_print_wr_par_info(hdev, &sei_data->wr_parity_info,
le32_to_cpu(sei_data->hdr.cnt));
require_hard_reset = true;
break;
case HBM_SEI_READ_ERR:
/* Unlike other SEI events, read error requires further processing of the
* raw data in order to determine the root cause.
*/
require_hard_reset = gaudi2_hbm_sei_handle_read_err(hdev,
&sei_data->read_err_info,
le32_to_cpu(sei_data->hdr.cnt));
break;
default:
break;
}
habanalabs/gaudi2: reset device upon critical ECC event Correctable ECC events are not fatal, but as they accumulate, the f/w can decide that a hard-rest is required. This indication is propagated to the host using the existing ECC event interface. Signed-off-by: Ofir Bitton <obitton@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-28 13:34:28 +08:00
require_hard_reset |= !!sei_data->hdr.is_critical;
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
return require_hard_reset;
}
static void gaudi2_handle_hbm_cattrip(struct hl_device *hdev, u64 intr_cause_data)
{
dev_err(hdev->dev,
"HBM catastrophic temperature error (CATTRIP) cause %#llx\n",
intr_cause_data);
}
static void gaudi2_handle_hbm_mc_spi(struct hl_device *hdev, u64 intr_cause_data)
{
u32 i;
for (i = 0 ; i < GAUDI2_NUM_OF_HBM_MC_SPI_CAUSE ; i++)
if (intr_cause_data & hbm_mc_spi[i].mask)
dev_dbg(hdev->dev, "HBM spi event: notification cause(%s)\n",
hbm_mc_spi[i].cause);
}
static void gaudi2_print_clk_change_info(struct hl_device *hdev, u16 event_type)
{
ktime_t zero_time = ktime_set(0, 0);
mutex_lock(&hdev->clk_throttling.lock);
switch (event_type) {
case GAUDI2_EVENT_CPU_FIX_POWER_ENV_S:
hdev->clk_throttling.current_reason |= HL_CLK_THROTTLE_POWER;
hdev->clk_throttling.aggregated_reason |= HL_CLK_THROTTLE_POWER;
hdev->clk_throttling.timestamp[HL_CLK_THROTTLE_TYPE_POWER].start = ktime_get();
hdev->clk_throttling.timestamp[HL_CLK_THROTTLE_TYPE_POWER].end = zero_time;
dev_info_ratelimited(hdev->dev, "Clock throttling due to power consumption\n");
break;
case GAUDI2_EVENT_CPU_FIX_POWER_ENV_E:
hdev->clk_throttling.current_reason &= ~HL_CLK_THROTTLE_POWER;
hdev->clk_throttling.timestamp[HL_CLK_THROTTLE_TYPE_POWER].end = ktime_get();
dev_info_ratelimited(hdev->dev, "Power envelop is safe, back to optimal clock\n");
break;
case GAUDI2_EVENT_CPU_FIX_THERMAL_ENV_S:
hdev->clk_throttling.current_reason |= HL_CLK_THROTTLE_THERMAL;
hdev->clk_throttling.aggregated_reason |= HL_CLK_THROTTLE_THERMAL;
hdev->clk_throttling.timestamp[HL_CLK_THROTTLE_TYPE_THERMAL].start = ktime_get();
hdev->clk_throttling.timestamp[HL_CLK_THROTTLE_TYPE_THERMAL].end = zero_time;
dev_info_ratelimited(hdev->dev, "Clock throttling due to overheating\n");
break;
case GAUDI2_EVENT_CPU_FIX_THERMAL_ENV_E:
hdev->clk_throttling.current_reason &= ~HL_CLK_THROTTLE_THERMAL;
hdev->clk_throttling.timestamp[HL_CLK_THROTTLE_TYPE_THERMAL].end = ktime_get();
dev_info_ratelimited(hdev->dev, "Thermal envelop is safe, back to optimal clock\n");
break;
default:
dev_err(hdev->dev, "Received invalid clock change event %d\n", event_type);
break;
}
mutex_unlock(&hdev->clk_throttling.lock);
}
static void gaudi2_print_out_of_sync_info(struct hl_device *hdev,
struct cpucp_pkt_sync_err *sync_err)
{
struct hl_hw_queue *q = &hdev->kernel_queues[GAUDI2_QUEUE_ID_CPU_PQ];
dev_err(hdev->dev, "Out of sync with FW, FW: pi=%u, ci=%u, LKD: pi=%u, ci=%u\n",
sync_err->pi, sync_err->ci, q->pi, atomic_read(&q->ci));
}
static void gaudi2_handle_pcie_p2p_msix(struct hl_device *hdev)
{
u32 p2p_intr, msix_gw_intr;
p2p_intr = RREG32(mmPCIE_WRAP_P2P_INTR);
msix_gw_intr = RREG32(mmPCIE_WRAP_MSIX_GW_INTR);
if (p2p_intr) {
dev_err_ratelimited(hdev->dev,
"pcie p2p transaction terminated due to security, req_id(0x%x)\n",
RREG32(mmPCIE_WRAP_P2P_REQ_ID));
WREG32(mmPCIE_WRAP_P2P_INTR, 0x1);
}
if (msix_gw_intr) {
dev_err_ratelimited(hdev->dev,
"pcie msi-x gen denied due to vector num check failure, vec(0x%X)\n",
RREG32(mmPCIE_WRAP_MSIX_GW_VEC));
WREG32(mmPCIE_WRAP_MSIX_GW_INTR, 0x1);
}
}
static void gaudi2_handle_pcie_drain(struct hl_device *hdev,
struct hl_eq_pcie_drain_ind_data *drain_data)
{
u64 lbw_rd, lbw_wr, hbw_rd, hbw_wr, cause;
cause = le64_to_cpu(drain_data->intr_cause.intr_cause_data);
lbw_rd = le64_to_cpu(drain_data->drain_rd_addr_lbw);
lbw_wr = le64_to_cpu(drain_data->drain_wr_addr_lbw);
hbw_rd = le64_to_cpu(drain_data->drain_rd_addr_hbw);
hbw_wr = le64_to_cpu(drain_data->drain_wr_addr_hbw);
if (cause & BIT_ULL(0))
dev_err_ratelimited(hdev->dev,
"PCIE AXI drain LBW completed, read_err %u, write_err %u\n",
!!lbw_rd, !!lbw_wr);
if (cause & BIT_ULL(1))
dev_err_ratelimited(hdev->dev,
"PCIE AXI drain HBW completed, raddr %#llx, waddr %#llx\n",
hbw_rd, hbw_wr);
}
static void gaudi2_handle_psoc_drain(struct hl_device *hdev, u64 intr_cause_data)
{
int i;
for (i = 0 ; i < GAUDI2_NUM_OF_AXI_DRAIN_ERR_CAUSE ; i++) {
if (intr_cause_data & BIT_ULL(i))
dev_err_ratelimited(hdev->dev, "PSOC %s completed\n",
gaudi2_psoc_axi_drain_interrupts_cause[i]);
}
}
static void gaudi2_print_cpu_pkt_failure_info(struct hl_device *hdev,
struct cpucp_pkt_sync_err *sync_err)
{
struct hl_hw_queue *q = &hdev->kernel_queues[GAUDI2_QUEUE_ID_CPU_PQ];
dev_warn(hdev->dev,
"FW reported sanity check failure, FW: pi=%u, ci=%u, LKD: pi=%u, ci=%u\n",
sync_err->pi, sync_err->ci, q->pi, atomic_read(&q->ci));
}
static void hl_arc_event_handle(struct hl_device *hdev,
struct hl_eq_engine_arc_intr_data *data)
{
struct hl_engine_arc_dccm_queue_full_irq *q;
u32 intr_type, engine_id;
u64 payload;
intr_type = le32_to_cpu(data->intr_type);
engine_id = le32_to_cpu(data->engine_id);
payload = le64_to_cpu(data->payload);
switch (intr_type) {
case ENGINE_ARC_DCCM_QUEUE_FULL_IRQ:
q = (struct hl_engine_arc_dccm_queue_full_irq *) &payload;
dev_err_ratelimited(hdev->dev,
"ARC DCCM Full event: EngId: %u, Intr_type: %u, Qidx: %u\n",
engine_id, intr_type, q->queue_index);
break;
default:
dev_err_ratelimited(hdev->dev, "Unknown ARC event type\n");
}
}
static void gaudi2_handle_eqe(struct hl_device *hdev, struct hl_eq_entry *eq_entry)
{
u32 ctl, reset_flags = HL_DRV_RESET_HARD | HL_DRV_RESET_DELAY;
struct gaudi2_device *gaudi2 = hdev->asic_specific;
habanalabs/gaudi2: reset device upon critical ECC event Correctable ECC events are not fatal, but as they accumulate, the f/w can decide that a hard-rest is required. This indication is propagated to the host using the existing ECC event interface. Signed-off-by: Ofir Bitton <obitton@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-28 13:34:28 +08:00
bool reset_required = false, skip_reset = false;
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
int index, sbte_index;
habanalabs: new notifier events for device state Add new notifier events that inform several device states. General H/W error raised on device general H/W error occurs. User engine error is raised when a device engine informs of an error. Signed-off-by: Tal Cohen <talcohen@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-08-18 17:54:23 +08:00
u64 event_mask = 0;
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
u16 event_type;
ctl = le32_to_cpu(eq_entry->hdr.ctl);
event_type = ((ctl & EQ_CTL_EVENT_TYPE_MASK) >> EQ_CTL_EVENT_TYPE_SHIFT);
if (event_type >= GAUDI2_EVENT_SIZE) {
dev_err(hdev->dev, "Event type %u exceeds maximum of %u",
event_type, GAUDI2_EVENT_SIZE - 1);
return;
}
gaudi2->events_stat[event_type]++;
gaudi2->events_stat_aggregate[event_type]++;
gaudi2_print_irq_info(hdev, event_type);
switch (event_type) {
case GAUDI2_EVENT_PCIE_CORE_SERR ... GAUDI2_EVENT_ARC0_ECC_DERR:
fallthrough;
case GAUDI2_EVENT_ROTATOR0_SERR ... GAUDI2_EVENT_ROTATOR1_DERR:
reset_flags |= HL_DRV_RESET_FW_FATAL_ERR;
habanalabs: new notifier events for device state Add new notifier events that inform several device states. General H/W error raised on device general H/W error occurs. User engine error is raised when a device engine informs of an error. Signed-off-by: Tal Cohen <talcohen@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-08-18 17:54:23 +08:00
event_mask |= HL_NOTIFIER_EVENT_GENERAL_HW_ERR;
habanalabs/gaudi2: reset device upon critical ECC event Correctable ECC events are not fatal, but as they accumulate, the f/w can decide that a hard-rest is required. This indication is propagated to the host using the existing ECC event interface. Signed-off-by: Ofir Bitton <obitton@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-28 13:34:28 +08:00
reset_required = gaudi2_handle_ecc_event(hdev, event_type, &eq_entry->ecc_data);
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
break;
case GAUDI2_EVENT_TPC0_QM ... GAUDI2_EVENT_PDMA1_QM:
fallthrough;
case GAUDI2_EVENT_ROTATOR0_ROT0_QM ... GAUDI2_EVENT_ROTATOR1_ROT1_QM:
fallthrough;
case GAUDI2_EVENT_NIC0_QM0 ... GAUDI2_EVENT_NIC11_QM1:
gaudi2_handle_qman_err(hdev, event_type);
habanalabs: new notifier events for device state Add new notifier events that inform several device states. General H/W error raised on device general H/W error occurs. User engine error is raised when a device engine informs of an error. Signed-off-by: Tal Cohen <talcohen@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-08-18 17:54:23 +08:00
event_mask |= HL_NOTIFIER_EVENT_USER_ENGINE_ERR;
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
break;
case GAUDI2_EVENT_ARC_AXI_ERROR_RESPONSE_0:
reset_flags |= HL_DRV_RESET_FW_FATAL_ERR;
gaudi2_handle_arc_farm_sei_err(hdev);
habanalabs: new notifier events for device state Add new notifier events that inform several device states. General H/W error raised on device general H/W error occurs. User engine error is raised when a device engine informs of an error. Signed-off-by: Tal Cohen <talcohen@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-08-18 17:54:23 +08:00
event_mask |= HL_NOTIFIER_EVENT_GENERAL_HW_ERR;
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
break;
case GAUDI2_EVENT_CPU_AXI_ERR_RSP:
gaudi2_handle_cpu_sei_err(hdev);
habanalabs: new notifier events for device state Add new notifier events that inform several device states. General H/W error raised on device general H/W error occurs. User engine error is raised when a device engine informs of an error. Signed-off-by: Tal Cohen <talcohen@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-08-18 17:54:23 +08:00
event_mask |= HL_NOTIFIER_EVENT_GENERAL_HW_ERR;
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
break;
case GAUDI2_EVENT_PDMA_CH0_AXI_ERR_RSP:
case GAUDI2_EVENT_PDMA_CH1_AXI_ERR_RSP:
reset_flags |= HL_DRV_RESET_FW_FATAL_ERR;
gaudi2_handle_qm_sei_err(hdev, event_type, &eq_entry->razwi_info);
habanalabs: new notifier events for device state Add new notifier events that inform several device states. General H/W error raised on device general H/W error occurs. User engine error is raised when a device engine informs of an error. Signed-off-by: Tal Cohen <talcohen@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-08-18 17:54:23 +08:00
event_mask |= HL_NOTIFIER_EVENT_USER_ENGINE_ERR;
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
break;
case GAUDI2_EVENT_ROTATOR0_AXI_ERROR_RESPONSE:
case GAUDI2_EVENT_ROTATOR1_AXI_ERROR_RESPONSE:
index = event_type - GAUDI2_EVENT_ROTATOR0_AXI_ERROR_RESPONSE;
gaudi2_handle_rot_err(hdev, index, &eq_entry->razwi_with_intr_cause);
gaudi2_handle_qm_sei_err(hdev, event_type, NULL);
habanalabs: new notifier events for device state Add new notifier events that inform several device states. General H/W error raised on device general H/W error occurs. User engine error is raised when a device engine informs of an error. Signed-off-by: Tal Cohen <talcohen@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-08-18 17:54:23 +08:00
event_mask |= HL_NOTIFIER_EVENT_USER_ENGINE_ERR;
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
break;
case GAUDI2_EVENT_TPC0_AXI_ERR_RSP ... GAUDI2_EVENT_TPC24_AXI_ERR_RSP:
index = event_type - GAUDI2_EVENT_TPC0_AXI_ERR_RSP;
gaudi2_tpc_ack_interrupts(hdev, index, "AXI_ERR_RSP",
&eq_entry->razwi_with_intr_cause);
gaudi2_handle_qm_sei_err(hdev, event_type, NULL);
habanalabs: new notifier events for device state Add new notifier events that inform several device states. General H/W error raised on device general H/W error occurs. User engine error is raised when a device engine informs of an error. Signed-off-by: Tal Cohen <talcohen@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-08-18 17:54:23 +08:00
event_mask |= HL_NOTIFIER_EVENT_USER_ENGINE_ERR;
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
break;
case GAUDI2_EVENT_DEC0_AXI_ERR_RSPONSE ... GAUDI2_EVENT_DEC9_AXI_ERR_RSPONSE:
index = event_type - GAUDI2_EVENT_DEC0_AXI_ERR_RSPONSE;
gaudi2_handle_dec_err(hdev, index, "AXI_ERR_RESPONSE", &eq_entry->razwi_info);
habanalabs: new notifier events for device state Add new notifier events that inform several device states. General H/W error raised on device general H/W error occurs. User engine error is raised when a device engine informs of an error. Signed-off-by: Tal Cohen <talcohen@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-08-18 17:54:23 +08:00
event_mask |= HL_NOTIFIER_EVENT_USER_ENGINE_ERR;
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
break;
case GAUDI2_EVENT_TPC0_KERNEL_ERR:
case GAUDI2_EVENT_TPC1_KERNEL_ERR:
case GAUDI2_EVENT_TPC2_KERNEL_ERR:
case GAUDI2_EVENT_TPC3_KERNEL_ERR:
case GAUDI2_EVENT_TPC4_KERNEL_ERR:
case GAUDI2_EVENT_TPC5_KERNEL_ERR:
case GAUDI2_EVENT_TPC6_KERNEL_ERR:
case GAUDI2_EVENT_TPC7_KERNEL_ERR:
case GAUDI2_EVENT_TPC8_KERNEL_ERR:
case GAUDI2_EVENT_TPC9_KERNEL_ERR:
case GAUDI2_EVENT_TPC10_KERNEL_ERR:
case GAUDI2_EVENT_TPC11_KERNEL_ERR:
case GAUDI2_EVENT_TPC12_KERNEL_ERR:
case GAUDI2_EVENT_TPC13_KERNEL_ERR:
case GAUDI2_EVENT_TPC14_KERNEL_ERR:
case GAUDI2_EVENT_TPC15_KERNEL_ERR:
case GAUDI2_EVENT_TPC16_KERNEL_ERR:
case GAUDI2_EVENT_TPC17_KERNEL_ERR:
case GAUDI2_EVENT_TPC18_KERNEL_ERR:
case GAUDI2_EVENT_TPC19_KERNEL_ERR:
case GAUDI2_EVENT_TPC20_KERNEL_ERR:
case GAUDI2_EVENT_TPC21_KERNEL_ERR:
case GAUDI2_EVENT_TPC22_KERNEL_ERR:
case GAUDI2_EVENT_TPC23_KERNEL_ERR:
case GAUDI2_EVENT_TPC24_KERNEL_ERR:
index = (event_type - GAUDI2_EVENT_TPC0_KERNEL_ERR) /
(GAUDI2_EVENT_TPC1_KERNEL_ERR - GAUDI2_EVENT_TPC0_KERNEL_ERR);
gaudi2_tpc_ack_interrupts(hdev, index, "KRN_ERR", &eq_entry->razwi_with_intr_cause);
habanalabs: new notifier events for device state Add new notifier events that inform several device states. General H/W error raised on device general H/W error occurs. User engine error is raised when a device engine informs of an error. Signed-off-by: Tal Cohen <talcohen@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-08-18 17:54:23 +08:00
event_mask |= HL_NOTIFIER_EVENT_USER_ENGINE_ERR;
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
break;
case GAUDI2_EVENT_DEC0_SPI:
case GAUDI2_EVENT_DEC1_SPI:
case GAUDI2_EVENT_DEC2_SPI:
case GAUDI2_EVENT_DEC3_SPI:
case GAUDI2_EVENT_DEC4_SPI:
case GAUDI2_EVENT_DEC5_SPI:
case GAUDI2_EVENT_DEC6_SPI:
case GAUDI2_EVENT_DEC7_SPI:
case GAUDI2_EVENT_DEC8_SPI:
case GAUDI2_EVENT_DEC9_SPI:
index = (event_type - GAUDI2_EVENT_DEC0_SPI) /
(GAUDI2_EVENT_DEC1_SPI - GAUDI2_EVENT_DEC0_SPI);
gaudi2_handle_dec_err(hdev, index, "SPI", &eq_entry->razwi_info);
habanalabs: new notifier events for device state Add new notifier events that inform several device states. General H/W error raised on device general H/W error occurs. User engine error is raised when a device engine informs of an error. Signed-off-by: Tal Cohen <talcohen@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-08-18 17:54:23 +08:00
event_mask |= HL_NOTIFIER_EVENT_USER_ENGINE_ERR;
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
break;
case GAUDI2_EVENT_MME0_CTRL_AXI_ERROR_RESPONSE:
case GAUDI2_EVENT_MME1_CTRL_AXI_ERROR_RESPONSE:
case GAUDI2_EVENT_MME2_CTRL_AXI_ERROR_RESPONSE:
case GAUDI2_EVENT_MME3_CTRL_AXI_ERROR_RESPONSE:
index = (event_type - GAUDI2_EVENT_MME0_CTRL_AXI_ERROR_RESPONSE) /
(GAUDI2_EVENT_MME1_CTRL_AXI_ERROR_RESPONSE -
GAUDI2_EVENT_MME0_CTRL_AXI_ERROR_RESPONSE);
gaudi2_handle_mme_err(hdev, index,
"CTRL_AXI_ERROR_RESPONSE", &eq_entry->razwi_info);
gaudi2_handle_qm_sei_err(hdev, event_type, NULL);
habanalabs: new notifier events for device state Add new notifier events that inform several device states. General H/W error raised on device general H/W error occurs. User engine error is raised when a device engine informs of an error. Signed-off-by: Tal Cohen <talcohen@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-08-18 17:54:23 +08:00
event_mask |= HL_NOTIFIER_EVENT_USER_ENGINE_ERR;
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
break;
case GAUDI2_EVENT_MME0_QMAN_SW_ERROR:
case GAUDI2_EVENT_MME1_QMAN_SW_ERROR:
case GAUDI2_EVENT_MME2_QMAN_SW_ERROR:
case GAUDI2_EVENT_MME3_QMAN_SW_ERROR:
index = (event_type - GAUDI2_EVENT_MME0_QMAN_SW_ERROR) /
(GAUDI2_EVENT_MME1_QMAN_SW_ERROR -
GAUDI2_EVENT_MME0_QMAN_SW_ERROR);
gaudi2_handle_mme_err(hdev, index, "QMAN_SW_ERROR", &eq_entry->razwi_info);
habanalabs: new notifier events for device state Add new notifier events that inform several device states. General H/W error raised on device general H/W error occurs. User engine error is raised when a device engine informs of an error. Signed-off-by: Tal Cohen <talcohen@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-08-18 17:54:23 +08:00
event_mask |= HL_NOTIFIER_EVENT_USER_ENGINE_ERR;
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
break;
case GAUDI2_EVENT_MME0_WAP_SOURCE_RESULT_INVALID:
case GAUDI2_EVENT_MME1_WAP_SOURCE_RESULT_INVALID:
case GAUDI2_EVENT_MME2_WAP_SOURCE_RESULT_INVALID:
case GAUDI2_EVENT_MME3_WAP_SOURCE_RESULT_INVALID:
index = (event_type - GAUDI2_EVENT_MME0_WAP_SOURCE_RESULT_INVALID) /
(GAUDI2_EVENT_MME1_WAP_SOURCE_RESULT_INVALID -
GAUDI2_EVENT_MME0_WAP_SOURCE_RESULT_INVALID);
gaudi2_handle_mme_wap_err(hdev, index, &eq_entry->razwi_info);
habanalabs: new notifier events for device state Add new notifier events that inform several device states. General H/W error raised on device general H/W error occurs. User engine error is raised when a device engine informs of an error. Signed-off-by: Tal Cohen <talcohen@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-08-18 17:54:23 +08:00
event_mask |= HL_NOTIFIER_EVENT_USER_ENGINE_ERR;
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
break;
case GAUDI2_EVENT_KDMA_CH0_AXI_ERR_RSP:
case GAUDI2_EVENT_KDMA0_CORE:
gaudi2_handle_kdma_core_event(hdev,
le64_to_cpu(eq_entry->intr_cause.intr_cause_data));
habanalabs: new notifier events for device state Add new notifier events that inform several device states. General H/W error raised on device general H/W error occurs. User engine error is raised when a device engine informs of an error. Signed-off-by: Tal Cohen <talcohen@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-08-18 17:54:23 +08:00
event_mask |= HL_NOTIFIER_EVENT_GENERAL_HW_ERR;
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
break;
case GAUDI2_EVENT_HDMA2_CORE ... GAUDI2_EVENT_PDMA1_CORE:
gaudi2_handle_dma_core_event(hdev,
le64_to_cpu(eq_entry->intr_cause.intr_cause_data));
habanalabs: new notifier events for device state Add new notifier events that inform several device states. General H/W error raised on device general H/W error occurs. User engine error is raised when a device engine informs of an error. Signed-off-by: Tal Cohen <talcohen@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-08-18 17:54:23 +08:00
event_mask |= HL_NOTIFIER_EVENT_USER_ENGINE_ERR;
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
break;
case GAUDI2_EVENT_PCIE_ADDR_DEC_ERR:
gaudi2_print_pcie_addr_dec_info(hdev,
le64_to_cpu(eq_entry->intr_cause.intr_cause_data));
habanalabs/gaudi2: mark PCIE access error as fatal F/W events are enabled in a late phase of the device init, so an event for a PCIE access error during the init, can be received after the init is already done and considered as successful. A resulting device reset, which does the same H/W init, can end similarly with this event right after the reset is done and considered as successful, and a loop of this sequence can continue. To avoid it mark the PCIE access error as a fatal event, so after 2 consecutive events no more resets will be done. Signed-off-by: Tomer Tayar <ttayar@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-07-21 01:02:20 +08:00
reset_flags |= HL_DRV_RESET_FW_FATAL_ERR;
habanalabs: new notifier events for device state Add new notifier events that inform several device states. General H/W error raised on device general H/W error occurs. User engine error is raised when a device engine informs of an error. Signed-off-by: Tal Cohen <talcohen@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-08-18 17:54:23 +08:00
event_mask |= HL_NOTIFIER_EVENT_GENERAL_HW_ERR;
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
break;
case GAUDI2_EVENT_HMMU0_PAGE_FAULT_OR_WR_PERM ... GAUDI2_EVENT_HMMU12_SECURITY_ERROR:
case GAUDI2_EVENT_HMMU_0_AXI_ERR_RSP ... GAUDI2_EVENT_HMMU_12_AXI_ERR_RSP:
habanalabs/gaudi2: add handling to pmmu events in eqe handler In order to get the error cause and the captured address in case of page fault, added pmmu events to eqe handler. Signed-off-by: Dani Liberman <dliberman@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-09-08 23:24:41 +08:00
case GAUDI2_EVENT_PMMU0_PAGE_FAULT_WR_PERM ... GAUDI2_EVENT_PMMU0_SECURITY_ERROR:
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
case GAUDI2_EVENT_PMMU_AXI_ERR_RSP_0:
gaudi2_handle_mmu_spi_sei_err(hdev, event_type);
reset_flags |= HL_DRV_RESET_FW_FATAL_ERR;
habanalabs: new notifier events for device state Add new notifier events that inform several device states. General H/W error raised on device general H/W error occurs. User engine error is raised when a device engine informs of an error. Signed-off-by: Tal Cohen <talcohen@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-08-18 17:54:23 +08:00
event_mask |= HL_NOTIFIER_EVENT_USER_ENGINE_ERR;
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
break;
case GAUDI2_EVENT_HIF0_FATAL ... GAUDI2_EVENT_HIF12_FATAL:
gaudi2_handle_hif_fatal(hdev, event_type,
le64_to_cpu(eq_entry->intr_cause.intr_cause_data));
reset_flags |= HL_DRV_RESET_FW_FATAL_ERR;
habanalabs: new notifier events for device state Add new notifier events that inform several device states. General H/W error raised on device general H/W error occurs. User engine error is raised when a device engine informs of an error. Signed-off-by: Tal Cohen <talcohen@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-08-18 17:54:23 +08:00
event_mask |= HL_NOTIFIER_EVENT_GENERAL_HW_ERR;
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
break;
case GAUDI2_EVENT_PMMU_FATAL_0:
gaudi2_handle_pif_fatal(hdev,
le64_to_cpu(eq_entry->intr_cause.intr_cause_data));
reset_flags |= HL_DRV_RESET_FW_FATAL_ERR;
habanalabs: new notifier events for device state Add new notifier events that inform several device states. General H/W error raised on device general H/W error occurs. User engine error is raised when a device engine informs of an error. Signed-off-by: Tal Cohen <talcohen@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-08-18 17:54:23 +08:00
event_mask |= HL_NOTIFIER_EVENT_GENERAL_HW_ERR;
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
break;
case GAUDI2_EVENT_PSOC63_RAZWI_OR_PID_MIN_MAX_INTERRUPT:
gaudi2_ack_psoc_razwi_event_handler(hdev);
habanalabs: new notifier events for device state Add new notifier events that inform several device states. General H/W error raised on device general H/W error occurs. User engine error is raised when a device engine informs of an error. Signed-off-by: Tal Cohen <talcohen@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-08-18 17:54:23 +08:00
event_mask |= HL_NOTIFIER_EVENT_USER_ENGINE_ERR;
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
break;
case GAUDI2_EVENT_HBM0_MC0_SEI_SEVERE ... GAUDI2_EVENT_HBM5_MC1_SEI_NON_SEVERE:
habanalabs: new notifier events for device state Add new notifier events that inform several device states. General H/W error raised on device general H/W error occurs. User engine error is raised when a device engine informs of an error. Signed-off-by: Tal Cohen <talcohen@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-08-18 17:54:23 +08:00
event_mask |= HL_NOTIFIER_EVENT_GENERAL_HW_ERR;
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
if (gaudi2_handle_hbm_mc_sei_err(hdev, event_type, &eq_entry->sei_data)) {
reset_flags |= HL_DRV_RESET_FW_FATAL_ERR;
habanalabs/gaudi2: reset device upon critical ECC event Correctable ECC events are not fatal, but as they accumulate, the f/w can decide that a hard-rest is required. This indication is propagated to the host using the existing ECC event interface. Signed-off-by: Ofir Bitton <obitton@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-28 13:34:28 +08:00
reset_required = true;
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
}
break;
case GAUDI2_EVENT_HBM_CATTRIP_0 ... GAUDI2_EVENT_HBM_CATTRIP_5:
gaudi2_handle_hbm_cattrip(hdev, le64_to_cpu(eq_entry->intr_cause.intr_cause_data));
habanalabs: new notifier events for device state Add new notifier events that inform several device states. General H/W error raised on device general H/W error occurs. User engine error is raised when a device engine informs of an error. Signed-off-by: Tal Cohen <talcohen@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-08-18 17:54:23 +08:00
event_mask |= HL_NOTIFIER_EVENT_GENERAL_HW_ERR;
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
break;
case GAUDI2_EVENT_HBM0_MC0_SPI ... GAUDI2_EVENT_HBM5_MC1_SPI:
gaudi2_handle_hbm_mc_spi(hdev, le64_to_cpu(eq_entry->intr_cause.intr_cause_data));
habanalabs: new notifier events for device state Add new notifier events that inform several device states. General H/W error raised on device general H/W error occurs. User engine error is raised when a device engine informs of an error. Signed-off-by: Tal Cohen <talcohen@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-08-18 17:54:23 +08:00
event_mask |= HL_NOTIFIER_EVENT_GENERAL_HW_ERR;
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
break;
case GAUDI2_EVENT_PCIE_DRAIN_COMPLETE:
gaudi2_handle_pcie_drain(hdev, &eq_entry->pcie_drain_ind_data);
habanalabs: new notifier events for device state Add new notifier events that inform several device states. General H/W error raised on device general H/W error occurs. User engine error is raised when a device engine informs of an error. Signed-off-by: Tal Cohen <talcohen@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-08-18 17:54:23 +08:00
event_mask |= HL_NOTIFIER_EVENT_GENERAL_HW_ERR;
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
break;
case GAUDI2_EVENT_PSOC59_RPM_ERROR_OR_DRAIN:
gaudi2_handle_psoc_drain(hdev, le64_to_cpu(eq_entry->intr_cause.intr_cause_data));
habanalabs: new notifier events for device state Add new notifier events that inform several device states. General H/W error raised on device general H/W error occurs. User engine error is raised when a device engine informs of an error. Signed-off-by: Tal Cohen <talcohen@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-08-18 17:54:23 +08:00
event_mask |= HL_NOTIFIER_EVENT_GENERAL_HW_ERR;
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
break;
case GAUDI2_EVENT_CPU_AXI_ECC:
reset_flags |= HL_DRV_RESET_FW_FATAL_ERR;
habanalabs: new notifier events for device state Add new notifier events that inform several device states. General H/W error raised on device general H/W error occurs. User engine error is raised when a device engine informs of an error. Signed-off-by: Tal Cohen <talcohen@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-08-18 17:54:23 +08:00
event_mask |= HL_NOTIFIER_EVENT_GENERAL_HW_ERR;
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
break;
case GAUDI2_EVENT_CPU_L2_RAM_ECC:
reset_flags |= HL_DRV_RESET_FW_FATAL_ERR;
habanalabs: new notifier events for device state Add new notifier events that inform several device states. General H/W error raised on device general H/W error occurs. User engine error is raised when a device engine informs of an error. Signed-off-by: Tal Cohen <talcohen@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-08-18 17:54:23 +08:00
event_mask |= HL_NOTIFIER_EVENT_GENERAL_HW_ERR;
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
break;
case GAUDI2_EVENT_MME0_SBTE0_AXI_ERR_RSP ... GAUDI2_EVENT_MME0_SBTE4_AXI_ERR_RSP:
case GAUDI2_EVENT_MME1_SBTE0_AXI_ERR_RSP ... GAUDI2_EVENT_MME1_SBTE4_AXI_ERR_RSP:
case GAUDI2_EVENT_MME2_SBTE0_AXI_ERR_RSP ... GAUDI2_EVENT_MME2_SBTE4_AXI_ERR_RSP:
case GAUDI2_EVENT_MME3_SBTE0_AXI_ERR_RSP ... GAUDI2_EVENT_MME3_SBTE4_AXI_ERR_RSP:
index = (event_type - GAUDI2_EVENT_MME0_SBTE0_AXI_ERR_RSP) /
(GAUDI2_EVENT_MME1_SBTE0_AXI_ERR_RSP -
GAUDI2_EVENT_MME0_SBTE0_AXI_ERR_RSP);
sbte_index = (event_type - GAUDI2_EVENT_MME0_SBTE0_AXI_ERR_RSP) %
(GAUDI2_EVENT_MME1_SBTE0_AXI_ERR_RSP -
GAUDI2_EVENT_MME0_SBTE0_AXI_ERR_RSP);
gaudi2_handle_mme_sbte_err(hdev, index, sbte_index,
le64_to_cpu(eq_entry->intr_cause.intr_cause_data));
habanalabs: new notifier events for device state Add new notifier events that inform several device states. General H/W error raised on device general H/W error occurs. User engine error is raised when a device engine informs of an error. Signed-off-by: Tal Cohen <talcohen@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-08-18 17:54:23 +08:00
event_mask |= HL_NOTIFIER_EVENT_USER_ENGINE_ERR;
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
break;
case GAUDI2_EVENT_VM0_ALARM_A ... GAUDI2_EVENT_VM3_ALARM_B:
reset_flags |= HL_DRV_RESET_FW_FATAL_ERR;
habanalabs: new notifier events for device state Add new notifier events that inform several device states. General H/W error raised on device general H/W error occurs. User engine error is raised when a device engine informs of an error. Signed-off-by: Tal Cohen <talcohen@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-08-18 17:54:23 +08:00
event_mask |= HL_NOTIFIER_EVENT_GENERAL_HW_ERR;
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
break;
case GAUDI2_EVENT_PSOC_AXI_ERR_RSP:
habanalabs: new notifier events for device state Add new notifier events that inform several device states. General H/W error raised on device general H/W error occurs. User engine error is raised when a device engine informs of an error. Signed-off-by: Tal Cohen <talcohen@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-08-18 17:54:23 +08:00
event_mask |= HL_NOTIFIER_EVENT_GENERAL_HW_ERR;
break;
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
case GAUDI2_EVENT_PSOC_PRSTN_FALL:
habanalabs: new notifier events for device state Add new notifier events that inform several device states. General H/W error raised on device general H/W error occurs. User engine error is raised when a device engine informs of an error. Signed-off-by: Tal Cohen <talcohen@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-08-18 17:54:23 +08:00
event_mask |= HL_NOTIFIER_EVENT_GENERAL_HW_ERR;
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
break;
case GAUDI2_EVENT_PCIE_APB_TIMEOUT:
reset_flags |= HL_DRV_RESET_FW_FATAL_ERR;
habanalabs: new notifier events for device state Add new notifier events that inform several device states. General H/W error raised on device general H/W error occurs. User engine error is raised when a device engine informs of an error. Signed-off-by: Tal Cohen <talcohen@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-08-18 17:54:23 +08:00
event_mask |= HL_NOTIFIER_EVENT_GENERAL_HW_ERR;
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
break;
case GAUDI2_EVENT_PCIE_FATAL_ERR:
habanalabs: new notifier events for device state Add new notifier events that inform several device states. General H/W error raised on device general H/W error occurs. User engine error is raised when a device engine informs of an error. Signed-off-by: Tal Cohen <talcohen@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-08-18 17:54:23 +08:00
event_mask |= HL_NOTIFIER_EVENT_GENERAL_HW_ERR;
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
break;
case GAUDI2_EVENT_TPC0_BMON_SPMU:
case GAUDI2_EVENT_TPC1_BMON_SPMU:
case GAUDI2_EVENT_TPC2_BMON_SPMU:
case GAUDI2_EVENT_TPC3_BMON_SPMU:
case GAUDI2_EVENT_TPC4_BMON_SPMU:
case GAUDI2_EVENT_TPC5_BMON_SPMU:
case GAUDI2_EVENT_TPC6_BMON_SPMU:
case GAUDI2_EVENT_TPC7_BMON_SPMU:
case GAUDI2_EVENT_TPC8_BMON_SPMU:
case GAUDI2_EVENT_TPC9_BMON_SPMU:
case GAUDI2_EVENT_TPC10_BMON_SPMU:
case GAUDI2_EVENT_TPC11_BMON_SPMU:
case GAUDI2_EVENT_TPC12_BMON_SPMU:
case GAUDI2_EVENT_TPC13_BMON_SPMU:
case GAUDI2_EVENT_TPC14_BMON_SPMU:
case GAUDI2_EVENT_TPC15_BMON_SPMU:
case GAUDI2_EVENT_TPC16_BMON_SPMU:
case GAUDI2_EVENT_TPC17_BMON_SPMU:
case GAUDI2_EVENT_TPC18_BMON_SPMU:
case GAUDI2_EVENT_TPC19_BMON_SPMU:
case GAUDI2_EVENT_TPC20_BMON_SPMU:
case GAUDI2_EVENT_TPC21_BMON_SPMU:
case GAUDI2_EVENT_TPC22_BMON_SPMU:
case GAUDI2_EVENT_TPC23_BMON_SPMU:
case GAUDI2_EVENT_TPC24_BMON_SPMU:
case GAUDI2_EVENT_MME0_CTRL_BMON_SPMU:
case GAUDI2_EVENT_MME0_SBTE_BMON_SPMU:
case GAUDI2_EVENT_MME0_WAP_BMON_SPMU:
case GAUDI2_EVENT_MME1_CTRL_BMON_SPMU:
case GAUDI2_EVENT_MME1_SBTE_BMON_SPMU:
case GAUDI2_EVENT_MME1_WAP_BMON_SPMU:
case GAUDI2_EVENT_MME2_CTRL_BMON_SPMU:
case GAUDI2_EVENT_MME2_SBTE_BMON_SPMU:
case GAUDI2_EVENT_MME2_WAP_BMON_SPMU:
case GAUDI2_EVENT_MME3_CTRL_BMON_SPMU:
case GAUDI2_EVENT_MME3_SBTE_BMON_SPMU:
case GAUDI2_EVENT_MME3_WAP_BMON_SPMU:
case GAUDI2_EVENT_HDMA2_BM_SPMU ... GAUDI2_EVENT_PDMA1_BM_SPMU:
fallthrough;
case GAUDI2_EVENT_DEC0_BMON_SPMU:
case GAUDI2_EVENT_DEC1_BMON_SPMU:
case GAUDI2_EVENT_DEC2_BMON_SPMU:
case GAUDI2_EVENT_DEC3_BMON_SPMU:
case GAUDI2_EVENT_DEC4_BMON_SPMU:
case GAUDI2_EVENT_DEC5_BMON_SPMU:
case GAUDI2_EVENT_DEC6_BMON_SPMU:
case GAUDI2_EVENT_DEC7_BMON_SPMU:
case GAUDI2_EVENT_DEC8_BMON_SPMU:
case GAUDI2_EVENT_DEC9_BMON_SPMU:
case GAUDI2_EVENT_ROTATOR0_BMON_SPMU ... GAUDI2_EVENT_SM3_BMON_SPMU:
habanalabs: new notifier events for device state Add new notifier events that inform several device states. General H/W error raised on device general H/W error occurs. User engine error is raised when a device engine informs of an error. Signed-off-by: Tal Cohen <talcohen@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-08-18 17:54:23 +08:00
event_mask |= HL_NOTIFIER_EVENT_USER_ENGINE_ERR;
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
break;
case GAUDI2_EVENT_CPU_FIX_POWER_ENV_S:
case GAUDI2_EVENT_CPU_FIX_POWER_ENV_E:
case GAUDI2_EVENT_CPU_FIX_THERMAL_ENV_S:
case GAUDI2_EVENT_CPU_FIX_THERMAL_ENV_E:
gaudi2_print_clk_change_info(hdev, event_type);
habanalabs: new notifier events for device state Add new notifier events that inform several device states. General H/W error raised on device general H/W error occurs. User engine error is raised when a device engine informs of an error. Signed-off-by: Tal Cohen <talcohen@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-08-18 17:54:23 +08:00
event_mask |= HL_NOTIFIER_EVENT_USER_ENGINE_ERR;
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
break;
case GAUDI2_EVENT_CPU_PKT_QUEUE_OUT_SYNC:
gaudi2_print_out_of_sync_info(hdev, &eq_entry->pkt_sync_err);
habanalabs: new notifier events for device state Add new notifier events that inform several device states. General H/W error raised on device general H/W error occurs. User engine error is raised when a device engine informs of an error. Signed-off-by: Tal Cohen <talcohen@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-08-18 17:54:23 +08:00
event_mask |= HL_NOTIFIER_EVENT_GENERAL_HW_ERR;
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
break;
case GAUDI2_EVENT_PCIE_FLR_REQUESTED:
habanalabs: new notifier events for device state Add new notifier events that inform several device states. General H/W error raised on device general H/W error occurs. User engine error is raised when a device engine informs of an error. Signed-off-by: Tal Cohen <talcohen@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-08-18 17:54:23 +08:00
event_mask |= HL_NOTIFIER_EVENT_GENERAL_HW_ERR;
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
/* Do nothing- FW will handle it */
break;
case GAUDI2_EVENT_PCIE_P2P_MSIX:
gaudi2_handle_pcie_p2p_msix(hdev);
habanalabs: new notifier events for device state Add new notifier events that inform several device states. General H/W error raised on device general H/W error occurs. User engine error is raised when a device engine informs of an error. Signed-off-by: Tal Cohen <talcohen@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-08-18 17:54:23 +08:00
event_mask |= HL_NOTIFIER_EVENT_USER_ENGINE_ERR;
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
break;
case GAUDI2_EVENT_SM0_AXI_ERROR_RESPONSE ... GAUDI2_EVENT_SM3_AXI_ERROR_RESPONSE:
index = event_type - GAUDI2_EVENT_SM0_AXI_ERROR_RESPONSE;
skip_reset = !gaudi2_handle_sm_err(hdev, index);
habanalabs: new notifier events for device state Add new notifier events that inform several device states. General H/W error raised on device general H/W error occurs. User engine error is raised when a device engine informs of an error. Signed-off-by: Tal Cohen <talcohen@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-08-18 17:54:23 +08:00
event_mask |= HL_NOTIFIER_EVENT_USER_ENGINE_ERR;
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
break;
case GAUDI2_EVENT_PSOC_MME_PLL_LOCK_ERR ... GAUDI2_EVENT_DCORE2_HBM_PLL_LOCK_ERR:
habanalabs: new notifier events for device state Add new notifier events that inform several device states. General H/W error raised on device general H/W error occurs. User engine error is raised when a device engine informs of an error. Signed-off-by: Tal Cohen <talcohen@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-08-18 17:54:23 +08:00
event_mask |= HL_NOTIFIER_EVENT_GENERAL_HW_ERR;
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
break;
case GAUDI2_EVENT_CPU_CPLD_SHUTDOWN_CAUSE:
dev_info(hdev->dev, "CPLD shutdown cause, reset reason: 0x%llx\n",
le64_to_cpu(eq_entry->data[0]));
habanalabs: new notifier events for device state Add new notifier events that inform several device states. General H/W error raised on device general H/W error occurs. User engine error is raised when a device engine informs of an error. Signed-off-by: Tal Cohen <talcohen@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-08-18 17:54:23 +08:00
event_mask |= HL_NOTIFIER_EVENT_GENERAL_HW_ERR;
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
break;
case GAUDI2_EVENT_CPU_CPLD_SHUTDOWN_EVENT:
dev_err(hdev->dev, "CPLD shutdown event, reset reason: 0x%llx\n",
le64_to_cpu(eq_entry->data[0]));
habanalabs: new notifier events for device state Add new notifier events that inform several device states. General H/W error raised on device general H/W error occurs. User engine error is raised when a device engine informs of an error. Signed-off-by: Tal Cohen <talcohen@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-08-18 17:54:23 +08:00
event_mask |= HL_NOTIFIER_EVENT_GENERAL_HW_ERR;
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
break;
case GAUDI2_EVENT_CPU_PKT_SANITY_FAILED:
gaudi2_print_cpu_pkt_failure_info(hdev, &eq_entry->pkt_sync_err);
habanalabs: new notifier events for device state Add new notifier events that inform several device states. General H/W error raised on device general H/W error occurs. User engine error is raised when a device engine informs of an error. Signed-off-by: Tal Cohen <talcohen@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-08-18 17:54:23 +08:00
event_mask |= HL_NOTIFIER_EVENT_GENERAL_HW_ERR;
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
break;
case GAUDI2_EVENT_ARC_DCCM_FULL:
hl_arc_event_handle(hdev, &eq_entry->arc_data);
habanalabs: new notifier events for device state Add new notifier events that inform several device states. General H/W error raised on device general H/W error occurs. User engine error is raised when a device engine informs of an error. Signed-off-by: Tal Cohen <talcohen@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-08-18 17:54:23 +08:00
event_mask |= HL_NOTIFIER_EVENT_USER_ENGINE_ERR;
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
break;
default:
if (gaudi2_irq_map_table[event_type].valid)
dev_err_ratelimited(hdev->dev, "Cannot find handler for event %d\n",
event_type);
}
habanalabs/gaudi2: reset device upon critical ECC event Correctable ECC events are not fatal, but as they accumulate, the f/w can decide that a hard-rest is required. This indication is propagated to the host using the existing ECC event interface. Signed-off-by: Ofir Bitton <obitton@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-28 13:34:28 +08:00
if ((gaudi2_irq_map_table[event_type].reset || reset_required) && !skip_reset)
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
goto reset_device;
/* Send unmask irq only for interrupts not classified as MSG */
if (!gaudi2_irq_map_table[event_type].msg)
hl_fw_unmask_irq(hdev, event_type);
habanalabs: new notifier events for device state Add new notifier events that inform several device states. General H/W error raised on device general H/W error occurs. User engine error is raised when a device engine informs of an error. Signed-off-by: Tal Cohen <talcohen@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-08-18 17:54:23 +08:00
if (event_mask)
hl_notifier_event_send_all(hdev, event_mask);
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
return;
reset_device:
if (hdev->hard_reset_on_fw_events) {
hl_device_reset(hdev, reset_flags);
habanalabs: new notifier events for device state Add new notifier events that inform several device states. General H/W error raised on device general H/W error occurs. User engine error is raised when a device engine informs of an error. Signed-off-by: Tal Cohen <talcohen@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-08-18 17:54:23 +08:00
event_mask |= HL_NOTIFIER_EVENT_DEVICE_RESET;
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
} else {
if (!gaudi2_irq_map_table[event_type].msg)
hl_fw_unmask_irq(hdev, event_type);
}
habanalabs: new notifier events for device state Add new notifier events that inform several device states. General H/W error raised on device general H/W error occurs. User engine error is raised when a device engine informs of an error. Signed-off-by: Tal Cohen <talcohen@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-08-18 17:54:23 +08:00
if (event_mask)
hl_notifier_event_send_all(hdev, event_mask);
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
}
static int gaudi2_memset_device_memory(struct hl_device *hdev, u64 addr, u64 size, u64 val)
{
struct asic_fixed_properties *prop = &hdev->asic_prop;
u64 comp_addr, cur_addr = addr, end_addr = addr + size;
u32 chunk_size, busy, dcore, edma_idx, sob_offset, sob_addr, comp_val, edma_commit;
u32 old_mmubp, mmubp;
int rc = 0;
sob_offset = hdev->asic_prop.first_available_user_sob[0] * 4;
sob_addr = mmDCORE0_SYNC_MNGR_OBJS_SOB_OBJ_0 + sob_offset;
comp_addr = CFG_BASE + sob_addr;
comp_val = FIELD_PREP(DCORE0_SYNC_MNGR_OBJS_SOB_OBJ_INC_MASK, 1) |
FIELD_PREP(DCORE0_SYNC_MNGR_OBJS_SOB_OBJ_VAL_MASK, 1);
edma_commit = FIELD_PREP(ARC_FARM_KDMA_CTX_COMMIT_LIN_MASK, 1) |
FIELD_PREP(ARC_FARM_KDMA_CTX_COMMIT_MEM_SET_MASK, 1) |
FIELD_PREP(ARC_FARM_KDMA_CTX_COMMIT_WR_COMP_EN_MASK, 1);
mmubp = FIELD_PREP(ARC_FARM_KDMA_CTX_AXUSER_HB_MMU_BP_WR_MASK, 1) |
FIELD_PREP(ARC_FARM_KDMA_CTX_AXUSER_HB_MMU_BP_RD_MASK, 1);
if (prop->edma_enabled_mask == 0) {
dev_info(hdev->dev, "non of the EDMA engines is enabled - skip dram scrubbing\n");
return -EIO;
}
/*
* set mmu bypass for the scrubbing - all ddmas are configured the same so save
* only the first one to restore later
*/
old_mmubp = RREG32(mmDCORE0_EDMA0_CORE_CTX_AXUSER_HB_MMU_BP);
for (dcore = 0 ; dcore < NUM_OF_DCORES ; dcore++) {
for (edma_idx = 0 ; edma_idx < NUM_OF_EDMA_PER_DCORE ; edma_idx++) {
u32 edma_offset = dcore * DCORE_OFFSET + edma_idx * DCORE_EDMA_OFFSET;
u32 edma_bit = dcore * NUM_OF_EDMA_PER_DCORE + edma_idx;
if (!(prop->edma_enabled_mask & BIT(edma_bit)))
continue;
WREG32(mmDCORE0_EDMA0_CORE_CTX_AXUSER_HB_MMU_BP +
edma_offset, mmubp);
}
}
while (cur_addr < end_addr) {
int dma_num = 0;
WREG32(sob_addr, 0);
for (dcore = 0 ; dcore < NUM_OF_DCORES ; dcore++) {
for (edma_idx = 0 ; edma_idx < NUM_OF_EDMA_PER_DCORE ; edma_idx++) {
u32 edma_offset = dcore * DCORE_OFFSET +
edma_idx * DCORE_EDMA_OFFSET;
u32 edma_bit = dcore * NUM_OF_EDMA_PER_DCORE + edma_idx;
if (!(prop->edma_enabled_mask & BIT(edma_bit)))
continue;
chunk_size = min_t(u64, SZ_2G, end_addr - cur_addr);
WREG32(mmDCORE0_EDMA0_CORE_CTX_SRC_BASE_LO + edma_offset,
lower_32_bits(val));
WREG32(mmDCORE0_EDMA0_CORE_CTX_SRC_BASE_HI + edma_offset,
upper_32_bits(val));
WREG32(mmDCORE0_EDMA0_CORE_CTX_DST_BASE_LO + edma_offset,
lower_32_bits(cur_addr));
WREG32(mmDCORE0_EDMA0_CORE_CTX_DST_BASE_HI + edma_offset,
upper_32_bits(cur_addr));
WREG32(mmDCORE0_EDMA0_CORE_CTX_WR_COMP_ADDR_LO + edma_offset,
lower_32_bits(comp_addr));
WREG32(mmDCORE0_EDMA0_CORE_CTX_WR_COMP_ADDR_HI + edma_offset,
upper_32_bits(comp_addr));
WREG32(mmDCORE0_EDMA0_CORE_CTX_WR_COMP_WDATA + edma_offset,
comp_val);
WREG32(mmDCORE0_EDMA0_CORE_CTX_DST_TSIZE_0 + edma_offset,
chunk_size);
WREG32(mmDCORE0_EDMA0_CORE_CTX_COMMIT + edma_offset, edma_commit);
dma_num++;
cur_addr += chunk_size;
if (cur_addr == end_addr)
goto poll;
}
}
poll:
rc = hl_poll_timeout(hdev, sob_addr, busy, (busy == dma_num), 1000, 1000000);
if (rc) {
dev_err(hdev->dev, "DMA Timeout during HBM scrubbing\n");
goto end;
}
}
end:
for (dcore = 0 ; dcore < NUM_OF_DCORES ; dcore++) {
for (edma_idx = 0 ; edma_idx < NUM_OF_EDMA_PER_DCORE ; edma_idx++) {
u32 edma_offset = dcore * DCORE_OFFSET + edma_idx * DCORE_EDMA_OFFSET;
u32 edma_bit = dcore * NUM_OF_EDMA_PER_DCORE + edma_idx;
if (!(prop->edma_enabled_mask & BIT(edma_bit)))
continue;
WREG32(mmDCORE0_EDMA0_CORE_CTX_AXUSER_HB_MMU_BP + edma_offset, old_mmubp);
}
}
WREG32(sob_addr, 0);
return rc;
}
static int gaudi2_scrub_device_dram(struct hl_device *hdev, u64 val)
{
int rc;
struct asic_fixed_properties *prop = &hdev->asic_prop;
u64 size = prop->dram_end_address - prop->dram_user_base_address;
rc = gaudi2_memset_device_memory(hdev, prop->dram_user_base_address, size, val);
if (rc)
dev_err(hdev->dev, "Failed to scrub dram, address: 0x%llx size: %llu\n",
prop->dram_user_base_address, size);
return rc;
}
static int gaudi2_scrub_device_mem(struct hl_device *hdev)
{
int rc;
struct asic_fixed_properties *prop = &hdev->asic_prop;
u64 val = hdev->memory_scrub_val;
u64 addr, size;
if (!hdev->memory_scrub)
return 0;
/* scrub SRAM */
addr = prop->sram_user_base_address;
size = hdev->pldm ? 0x10000 : (prop->sram_size - SRAM_USER_BASE_OFFSET);
dev_dbg(hdev->dev, "Scrubbing SRAM: 0x%09llx - 0x%09llx, val: 0x%llx\n",
addr, addr + size, val);
rc = gaudi2_memset_device_memory(hdev, addr, size, val);
if (rc) {
dev_err(hdev->dev, "scrubbing SRAM failed (%d)\n", rc);
return rc;
}
/* scrub DRAM */
rc = gaudi2_scrub_device_dram(hdev, val);
if (rc) {
dev_err(hdev->dev, "scrubbing DRAM failed (%d)\n", rc);
return rc;
}
return 0;
}
static void gaudi2_restore_user_sm_registers(struct hl_device *hdev)
{
u64 addr, mon_sts_addr, mon_cfg_addr, cq_lbw_l_addr, cq_lbw_h_addr,
cq_lbw_data_addr, cq_base_l_addr, cq_base_h_addr, cq_size_addr;
u32 val, size, offset;
int dcore_id;
offset = hdev->asic_prop.first_available_cq[0] * 4;
cq_lbw_l_addr = mmDCORE0_SYNC_MNGR_GLBL_LBW_ADDR_L_0 + offset;
cq_lbw_h_addr = mmDCORE0_SYNC_MNGR_GLBL_LBW_ADDR_H_0 + offset;
cq_lbw_data_addr = mmDCORE0_SYNC_MNGR_GLBL_LBW_DATA_0 + offset;
cq_base_l_addr = mmDCORE0_SYNC_MNGR_GLBL_CQ_BASE_ADDR_L_0 + offset;
cq_base_h_addr = mmDCORE0_SYNC_MNGR_GLBL_CQ_BASE_ADDR_H_0 + offset;
cq_size_addr = mmDCORE0_SYNC_MNGR_GLBL_CQ_SIZE_LOG2_0 + offset;
size = mmDCORE0_SYNC_MNGR_GLBL_LBW_ADDR_H_0 -
(mmDCORE0_SYNC_MNGR_GLBL_LBW_ADDR_L_0 + offset);
/* memset dcore0 CQ registers */
gaudi2_memset_device_lbw(hdev, cq_lbw_l_addr, size, 0);
gaudi2_memset_device_lbw(hdev, cq_lbw_h_addr, size, 0);
gaudi2_memset_device_lbw(hdev, cq_lbw_data_addr, size, 0);
gaudi2_memset_device_lbw(hdev, cq_base_l_addr, size, 0);
gaudi2_memset_device_lbw(hdev, cq_base_h_addr, size, 0);
cq_lbw_l_addr = mmDCORE0_SYNC_MNGR_GLBL_LBW_ADDR_L_0 + DCORE_OFFSET;
cq_lbw_h_addr = mmDCORE0_SYNC_MNGR_GLBL_LBW_ADDR_H_0 + DCORE_OFFSET;
cq_lbw_data_addr = mmDCORE0_SYNC_MNGR_GLBL_LBW_DATA_0 + DCORE_OFFSET;
cq_base_l_addr = mmDCORE0_SYNC_MNGR_GLBL_CQ_BASE_ADDR_L_0 + DCORE_OFFSET;
cq_base_h_addr = mmDCORE0_SYNC_MNGR_GLBL_CQ_BASE_ADDR_H_0 + DCORE_OFFSET;
cq_size_addr = mmDCORE0_SYNC_MNGR_GLBL_CQ_SIZE_LOG2_0 + DCORE_OFFSET;
size = mmDCORE0_SYNC_MNGR_GLBL_LBW_ADDR_H_0 - mmDCORE0_SYNC_MNGR_GLBL_LBW_ADDR_L_0;
for (dcore_id = 1 ; dcore_id < NUM_OF_DCORES ; dcore_id++) {
gaudi2_memset_device_lbw(hdev, cq_lbw_l_addr, size, 0);
gaudi2_memset_device_lbw(hdev, cq_lbw_h_addr, size, 0);
gaudi2_memset_device_lbw(hdev, cq_lbw_data_addr, size, 0);
gaudi2_memset_device_lbw(hdev, cq_base_l_addr, size, 0);
gaudi2_memset_device_lbw(hdev, cq_base_h_addr, size, 0);
gaudi2_memset_device_lbw(hdev, cq_size_addr, size, 0);
cq_lbw_l_addr += DCORE_OFFSET;
cq_lbw_h_addr += DCORE_OFFSET;
cq_lbw_data_addr += DCORE_OFFSET;
cq_base_l_addr += DCORE_OFFSET;
cq_base_h_addr += DCORE_OFFSET;
cq_size_addr += DCORE_OFFSET;
}
offset = hdev->asic_prop.first_available_user_mon[0] * 4;
addr = mmDCORE0_SYNC_MNGR_OBJS_MON_STATUS_0 + offset;
val = 1 << DCORE0_SYNC_MNGR_OBJS_MON_STATUS_PROT_SHIFT;
size = mmDCORE0_SYNC_MNGR_OBJS_SM_SEC_0 - (mmDCORE0_SYNC_MNGR_OBJS_MON_STATUS_0 + offset);
/* memset dcore0 monitors */
gaudi2_memset_device_lbw(hdev, addr, size, val);
addr = mmDCORE0_SYNC_MNGR_OBJS_MON_CONFIG_0 + offset;
gaudi2_memset_device_lbw(hdev, addr, size, 0);
mon_sts_addr = mmDCORE0_SYNC_MNGR_OBJS_MON_STATUS_0 + DCORE_OFFSET;
mon_cfg_addr = mmDCORE0_SYNC_MNGR_OBJS_MON_CONFIG_0 + DCORE_OFFSET;
size = mmDCORE0_SYNC_MNGR_OBJS_SM_SEC_0 - mmDCORE0_SYNC_MNGR_OBJS_MON_STATUS_0;
for (dcore_id = 1 ; dcore_id < NUM_OF_DCORES ; dcore_id++) {
gaudi2_memset_device_lbw(hdev, mon_sts_addr, size, val);
gaudi2_memset_device_lbw(hdev, mon_cfg_addr, size, 0);
mon_sts_addr += DCORE_OFFSET;
mon_cfg_addr += DCORE_OFFSET;
}
offset = hdev->asic_prop.first_available_user_sob[0] * 4;
addr = mmDCORE0_SYNC_MNGR_OBJS_SOB_OBJ_0 + offset;
val = 0;
size = mmDCORE0_SYNC_MNGR_OBJS_MON_PAY_ADDRL_0 -
(mmDCORE0_SYNC_MNGR_OBJS_SOB_OBJ_0 + offset);
/* memset dcore0 sobs */
gaudi2_memset_device_lbw(hdev, addr, size, val);
addr = mmDCORE0_SYNC_MNGR_OBJS_SOB_OBJ_0 + DCORE_OFFSET;
size = mmDCORE0_SYNC_MNGR_OBJS_MON_PAY_ADDRL_0 - mmDCORE0_SYNC_MNGR_OBJS_SOB_OBJ_0;
for (dcore_id = 1 ; dcore_id < NUM_OF_DCORES ; dcore_id++) {
gaudi2_memset_device_lbw(hdev, addr, size, val);
addr += DCORE_OFFSET;
}
/* Flush all WREG to prevent race */
val = RREG32(mmDCORE0_SYNC_MNGR_OBJS_SOB_OBJ_0 + offset);
}
static void gaudi2_restore_user_qm_registers(struct hl_device *hdev)
{
u32 reg_base, hw_queue_id;
for (hw_queue_id = GAUDI2_QUEUE_ID_PDMA_0_0 ; hw_queue_id <= GAUDI2_QUEUE_ID_ROT_1_0;
hw_queue_id += NUM_OF_PQ_PER_QMAN) {
if (!gaudi2_is_queue_enabled(hdev, hw_queue_id))
continue;
gaudi2_clear_qm_fence_counters_common(hdev, hw_queue_id, false);
reg_base = gaudi2_qm_blocks_bases[hw_queue_id];
WREG32(reg_base + QM_ARB_CFG_0_OFFSET, 0);
}
/* Flush all WREG to prevent race */
RREG32(mmPDMA0_QM_ARB_CFG_0);
}
static void gaudi2_restore_nic_qm_registers(struct hl_device *hdev)
{
u32 reg_base, hw_queue_id;
for (hw_queue_id = GAUDI2_QUEUE_ID_NIC_0_0 ; hw_queue_id <= GAUDI2_QUEUE_ID_NIC_23_3;
hw_queue_id += NUM_OF_PQ_PER_QMAN) {
if (!gaudi2_is_queue_enabled(hdev, hw_queue_id))
continue;
gaudi2_clear_qm_fence_counters_common(hdev, hw_queue_id, false);
reg_base = gaudi2_qm_blocks_bases[hw_queue_id];
WREG32(reg_base + QM_ARB_CFG_0_OFFSET, 0);
}
/* Flush all WREG to prevent race */
RREG32(mmPDMA0_QM_ARB_CFG_0);
}
static int gaudi2_context_switch(struct hl_device *hdev, u32 asid)
{
return 0;
}
static void gaudi2_restore_phase_topology(struct hl_device *hdev)
{
}
static void gaudi2_init_block_instances(struct hl_device *hdev, u32 block_idx,
struct dup_block_ctx *cfg_ctx)
{
u64 block_base = cfg_ctx->base + block_idx * cfg_ctx->block_off;
u8 seq;
int i;
for (i = 0 ; i < cfg_ctx->instances ; i++) {
seq = block_idx * cfg_ctx->instances + i;
/* skip disabled instance */
if (!(cfg_ctx->enabled_mask & BIT_ULL(seq)))
continue;
cfg_ctx->instance_cfg_fn(hdev, block_base + i * cfg_ctx->instance_off,
cfg_ctx->data);
}
}
static void gaudi2_init_blocks_with_mask(struct hl_device *hdev, struct dup_block_ctx *cfg_ctx,
u64 mask)
{
int i;
cfg_ctx->enabled_mask = mask;
for (i = 0 ; i < cfg_ctx->blocks ; i++)
gaudi2_init_block_instances(hdev, i, cfg_ctx);
}
void gaudi2_init_blocks(struct hl_device *hdev, struct dup_block_ctx *cfg_ctx)
{
gaudi2_init_blocks_with_mask(hdev, cfg_ctx, U64_MAX);
}
static int gaudi2_debugfs_read_dma(struct hl_device *hdev, u64 addr, u32 size, void *blob_addr)
{
void *host_mem_virtual_addr;
dma_addr_t host_mem_dma_addr;
u64 reserved_va_base;
u32 pos, size_left, size_to_dma;
struct hl_ctx *ctx;
int rc = 0;
/* Fetch the ctx */
ctx = hl_get_compute_ctx(hdev);
if (!ctx) {
dev_err(hdev->dev, "No ctx available\n");
return -EINVAL;
}
/* Allocate buffers for read and for poll */
host_mem_virtual_addr = hl_asic_dma_alloc_coherent(hdev, SZ_2M, &host_mem_dma_addr,
GFP_KERNEL | __GFP_ZERO);
if (host_mem_virtual_addr == NULL) {
dev_err(hdev->dev, "Failed to allocate memory for KDMA read\n");
rc = -ENOMEM;
goto put_ctx;
}
/* Reserve VM region on asic side */
reserved_va_base = hl_reserve_va_block(hdev, ctx, HL_VA_RANGE_TYPE_HOST, SZ_2M,
HL_MMU_VA_ALIGNMENT_NOT_NEEDED);
if (!reserved_va_base) {
dev_err(hdev->dev, "Failed to reserve vmem on asic\n");
rc = -ENOMEM;
goto free_data_buffer;
}
/* Create mapping on asic side */
habanalabs: MMU invalidation h/w is per device The code used the mmu mutex to protect access to the context's page tables and invalidation of the MMU cache. Because pgt are per context, the mmu mutex was a member of the context object. The problem is that the device has a single MMU invalidation h/w (per MMU). Therefore, the mmu mutex should not be a property of the context but a property of the device. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-09-05 22:14:45 +08:00
mutex_lock(&hdev->mmu_lock);
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
rc = hl_mmu_map_contiguous(ctx, reserved_va_base, host_mem_dma_addr, SZ_2M);
hl_mmu_invalidate_cache_range(hdev, false,
MMU_OP_USERPTR | MMU_OP_SKIP_LOW_CACHE_INV,
ctx->asid, reserved_va_base, SZ_2M);
habanalabs: MMU invalidation h/w is per device The code used the mmu mutex to protect access to the context's page tables and invalidation of the MMU cache. Because pgt are per context, the mmu mutex was a member of the context object. The problem is that the device has a single MMU invalidation h/w (per MMU). Therefore, the mmu mutex should not be a property of the context but a property of the device. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-09-05 22:14:45 +08:00
mutex_unlock(&hdev->mmu_lock);
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
if (rc) {
dev_err(hdev->dev, "Failed to create mapping on asic mmu\n");
goto unreserve_va;
}
/* Enable MMU on KDMA */
gaudi2_kdma_set_mmbp_asid(hdev, false, ctx->asid);
pos = 0;
size_left = size;
size_to_dma = SZ_2M;
while (size_left > 0) {
if (size_left < SZ_2M)
size_to_dma = size_left;
rc = gaudi2_send_job_to_kdma(hdev, addr, reserved_va_base, size_to_dma, false);
if (rc)
break;
memcpy(blob_addr + pos, host_mem_virtual_addr, size_to_dma);
if (size_left <= SZ_2M)
break;
pos += SZ_2M;
addr += SZ_2M;
size_left -= SZ_2M;
}
gaudi2_kdma_set_mmbp_asid(hdev, true, HL_KERNEL_ASID_ID);
habanalabs: MMU invalidation h/w is per device The code used the mmu mutex to protect access to the context's page tables and invalidation of the MMU cache. Because pgt are per context, the mmu mutex was a member of the context object. The problem is that the device has a single MMU invalidation h/w (per MMU). Therefore, the mmu mutex should not be a property of the context but a property of the device. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-09-05 22:14:45 +08:00
mutex_lock(&hdev->mmu_lock);
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
hl_mmu_unmap_contiguous(ctx, reserved_va_base, SZ_2M);
hl_mmu_invalidate_cache_range(hdev, false, MMU_OP_USERPTR,
ctx->asid, reserved_va_base, SZ_2M);
habanalabs: MMU invalidation h/w is per device The code used the mmu mutex to protect access to the context's page tables and invalidation of the MMU cache. Because pgt are per context, the mmu mutex was a member of the context object. The problem is that the device has a single MMU invalidation h/w (per MMU). Therefore, the mmu mutex should not be a property of the context but a property of the device. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-09-05 22:14:45 +08:00
mutex_unlock(&hdev->mmu_lock);
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
unreserve_va:
hl_unreserve_va_block(hdev, ctx, reserved_va_base, SZ_2M);
free_data_buffer:
hl_asic_dma_free_coherent(hdev, SZ_2M, host_mem_virtual_addr, host_mem_dma_addr);
put_ctx:
hl_ctx_put(ctx);
return rc;
}
static int gaudi2_internal_cb_pool_init(struct hl_device *hdev, struct hl_ctx *ctx)
{
struct gaudi2_device *gaudi2 = hdev->asic_specific;
int min_alloc_order, rc;
if (!(gaudi2->hw_cap_initialized & HW_CAP_PMMU))
return 0;
hdev->internal_cb_pool_virt_addr = hl_asic_dma_alloc_coherent(hdev,
HOST_SPACE_INTERNAL_CB_SZ,
&hdev->internal_cb_pool_dma_addr,
GFP_KERNEL | __GFP_ZERO);
if (!hdev->internal_cb_pool_virt_addr)
return -ENOMEM;
min_alloc_order = ilog2(min(gaudi2_get_signal_cb_size(hdev),
gaudi2_get_wait_cb_size(hdev)));
hdev->internal_cb_pool = gen_pool_create(min_alloc_order, -1);
if (!hdev->internal_cb_pool) {
dev_err(hdev->dev, "Failed to create internal CB pool\n");
rc = -ENOMEM;
goto free_internal_cb_pool;
}
rc = gen_pool_add(hdev->internal_cb_pool, (uintptr_t) hdev->internal_cb_pool_virt_addr,
HOST_SPACE_INTERNAL_CB_SZ, -1);
if (rc) {
dev_err(hdev->dev, "Failed to add memory to internal CB pool\n");
rc = -EFAULT;
goto destroy_internal_cb_pool;
}
hdev->internal_cb_va_base = hl_reserve_va_block(hdev, ctx, HL_VA_RANGE_TYPE_HOST,
HOST_SPACE_INTERNAL_CB_SZ, HL_MMU_VA_ALIGNMENT_NOT_NEEDED);
if (!hdev->internal_cb_va_base) {
rc = -ENOMEM;
goto destroy_internal_cb_pool;
}
habanalabs: MMU invalidation h/w is per device The code used the mmu mutex to protect access to the context's page tables and invalidation of the MMU cache. Because pgt are per context, the mmu mutex was a member of the context object. The problem is that the device has a single MMU invalidation h/w (per MMU). Therefore, the mmu mutex should not be a property of the context but a property of the device. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-09-05 22:14:45 +08:00
mutex_lock(&hdev->mmu_lock);
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
rc = hl_mmu_map_contiguous(ctx, hdev->internal_cb_va_base, hdev->internal_cb_pool_dma_addr,
HOST_SPACE_INTERNAL_CB_SZ);
hl_mmu_invalidate_cache(hdev, false, MMU_OP_USERPTR);
habanalabs: MMU invalidation h/w is per device The code used the mmu mutex to protect access to the context's page tables and invalidation of the MMU cache. Because pgt are per context, the mmu mutex was a member of the context object. The problem is that the device has a single MMU invalidation h/w (per MMU). Therefore, the mmu mutex should not be a property of the context but a property of the device. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-09-05 22:14:45 +08:00
mutex_unlock(&hdev->mmu_lock);
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
if (rc)
goto unreserve_internal_cb_pool;
return 0;
unreserve_internal_cb_pool:
hl_unreserve_va_block(hdev, ctx, hdev->internal_cb_va_base, HOST_SPACE_INTERNAL_CB_SZ);
destroy_internal_cb_pool:
gen_pool_destroy(hdev->internal_cb_pool);
free_internal_cb_pool:
hl_asic_dma_free_coherent(hdev, HOST_SPACE_INTERNAL_CB_SZ, hdev->internal_cb_pool_virt_addr,
hdev->internal_cb_pool_dma_addr);
return rc;
}
static void gaudi2_internal_cb_pool_fini(struct hl_device *hdev, struct hl_ctx *ctx)
{
struct gaudi2_device *gaudi2 = hdev->asic_specific;
if (!(gaudi2->hw_cap_initialized & HW_CAP_PMMU))
return;
habanalabs: MMU invalidation h/w is per device The code used the mmu mutex to protect access to the context's page tables and invalidation of the MMU cache. Because pgt are per context, the mmu mutex was a member of the context object. The problem is that the device has a single MMU invalidation h/w (per MMU). Therefore, the mmu mutex should not be a property of the context but a property of the device. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-09-05 22:14:45 +08:00
mutex_lock(&hdev->mmu_lock);
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
hl_mmu_unmap_contiguous(ctx, hdev->internal_cb_va_base, HOST_SPACE_INTERNAL_CB_SZ);
hl_unreserve_va_block(hdev, ctx, hdev->internal_cb_va_base, HOST_SPACE_INTERNAL_CB_SZ);
hl_mmu_invalidate_cache(hdev, true, MMU_OP_USERPTR);
habanalabs: MMU invalidation h/w is per device The code used the mmu mutex to protect access to the context's page tables and invalidation of the MMU cache. Because pgt are per context, the mmu mutex was a member of the context object. The problem is that the device has a single MMU invalidation h/w (per MMU). Therefore, the mmu mutex should not be a property of the context but a property of the device. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-09-05 22:14:45 +08:00
mutex_unlock(&hdev->mmu_lock);
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
gen_pool_destroy(hdev->internal_cb_pool);
hl_asic_dma_free_coherent(hdev, HOST_SPACE_INTERNAL_CB_SZ, hdev->internal_cb_pool_virt_addr,
hdev->internal_cb_pool_dma_addr);
}
static void gaudi2_restore_user_registers(struct hl_device *hdev)
{
gaudi2_restore_user_sm_registers(hdev);
gaudi2_restore_user_qm_registers(hdev);
}
habanalabs/gaudi2: map virtual MSI-X doorbell memory for user Upon the initialization of a user context, map the host memory page of the virtual MSI-X doorbell in the device MMU. A reserved VA is used for this purpose, so user can use it directly without any allocation/map operation. Signed-off-by: Tomer Tayar <ttayar@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-07-01 03:05:51 +08:00
static int gaudi2_map_virtual_msix_doorbell_memory(struct hl_ctx *ctx)
{
struct hl_device *hdev = ctx->hdev;
struct asic_fixed_properties *prop = &hdev->asic_prop;
struct gaudi2_device *gaudi2 = hdev->asic_specific;
int rc;
rc = hl_mmu_map_page(ctx, RESERVED_VA_FOR_VIRTUAL_MSIX_DOORBELL_START,
gaudi2->virt_msix_db_dma_addr, prop->pmmu.page_size, true);
if (rc)
dev_err(hdev->dev, "Failed to map VA %#llx for virtual MSI-X doorbell memory\n",
RESERVED_VA_FOR_VIRTUAL_MSIX_DOORBELL_START);
return rc;
}
static void gaudi2_unmap_virtual_msix_doorbell_memory(struct hl_ctx *ctx)
{
struct hl_device *hdev = ctx->hdev;
struct asic_fixed_properties *prop = &hdev->asic_prop;
int rc;
rc = hl_mmu_unmap_page(ctx, RESERVED_VA_FOR_VIRTUAL_MSIX_DOORBELL_START,
prop->pmmu.page_size, true);
if (rc)
dev_err(hdev->dev, "Failed to unmap VA %#llx of virtual MSI-X doorbell memory\n",
RESERVED_VA_FOR_VIRTUAL_MSIX_DOORBELL_START);
}
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
static int gaudi2_ctx_init(struct hl_ctx *ctx)
{
int rc;
rc = gaudi2_mmu_prepare(ctx->hdev, ctx->asid);
if (rc)
return rc;
/* No need to clear user registers if the device has just
* performed reset, we restore only nic qm registers
*/
if (ctx->hdev->reset_upon_device_release)
gaudi2_restore_nic_qm_registers(ctx->hdev);
else
gaudi2_restore_user_registers(ctx->hdev);
habanalabs/gaudi2: map virtual MSI-X doorbell memory for user Upon the initialization of a user context, map the host memory page of the virtual MSI-X doorbell in the device MMU. A reserved VA is used for this purpose, so user can use it directly without any allocation/map operation. Signed-off-by: Tomer Tayar <ttayar@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-07-01 03:05:51 +08:00
rc = gaudi2_internal_cb_pool_init(ctx->hdev, ctx);
if (rc)
return rc;
rc = gaudi2_map_virtual_msix_doorbell_memory(ctx);
if (rc)
gaudi2_internal_cb_pool_fini(ctx->hdev, ctx);
return rc;
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
}
static void gaudi2_ctx_fini(struct hl_ctx *ctx)
{
if (ctx->asid == HL_KERNEL_ASID_ID)
return;
gaudi2_internal_cb_pool_fini(ctx->hdev, ctx);
habanalabs/gaudi2: map virtual MSI-X doorbell memory for user Upon the initialization of a user context, map the host memory page of the virtual MSI-X doorbell in the device MMU. A reserved VA is used for this purpose, so user can use it directly without any allocation/map operation. Signed-off-by: Tomer Tayar <ttayar@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-07-01 03:05:51 +08:00
gaudi2_unmap_virtual_msix_doorbell_memory(ctx);
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
}
static int gaudi2_pre_schedule_cs(struct hl_cs *cs)
{
struct hl_device *hdev = cs->ctx->hdev;
int index = cs->sequence & (hdev->asic_prop.max_pending_cs - 1);
habanalabs/gaudi2: replace defines for reserved sob/mob with enums Following patches are going to add more reserved sync objects and monitors. To make the counting of these reserved resources simpler, replace the existing RESERVED_* defines with enumerations. Signed-off-by: Tomer Tayar <ttayar@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-30 00:20:38 +08:00
u32 mon_payload, sob_id, mon_id;
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
if (!cs_needs_completion(cs))
return 0;
/*
habanalabs/gaudi2: replace defines for reserved sob/mob with enums Following patches are going to add more reserved sync objects and monitors. To make the counting of these reserved resources simpler, replace the existing RESERVED_* defines with enumerations. Signed-off-by: Tomer Tayar <ttayar@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-30 00:20:38 +08:00
* First 64 SOB/MON are reserved for driver for QMAN auto completion
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
* mechanism. Each SOB/MON pair are used for a pending CS with the same
* cyclic index. The SOB value is increased when each of the CS jobs is
* completed. When the SOB reaches the number of CS jobs, the monitor
* generates MSI-X interrupt.
*/
habanalabs/gaudi2: replace defines for reserved sob/mob with enums Following patches are going to add more reserved sync objects and monitors. To make the counting of these reserved resources simpler, replace the existing RESERVED_* defines with enumerations. Signed-off-by: Tomer Tayar <ttayar@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-30 00:20:38 +08:00
sob_id = mon_id = index;
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
mon_payload = (1 << CQ_ENTRY_SHADOW_INDEX_VALID_SHIFT) |
(1 << CQ_ENTRY_READY_SHIFT) | index;
habanalabs/gaudi2: replace defines for reserved sob/mob with enums Following patches are going to add more reserved sync objects and monitors. To make the counting of these reserved resources simpler, replace the existing RESERVED_* defines with enumerations. Signed-off-by: Tomer Tayar <ttayar@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-30 00:20:38 +08:00
gaudi2_arm_cq_monitor(hdev, sob_id, mon_id, GAUDI2_RESERVED_CQ_CS_COMPLETION, mon_payload,
cs->jobs_cnt);
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
return 0;
}
static u32 gaudi2_get_queue_id_for_cq(struct hl_device *hdev, u32 cq_idx)
{
return HL_INVALID_QUEUE;
}
static u32 gaudi2_gen_signal_cb(struct hl_device *hdev, void *data, u16 sob_id, u32 size, bool eb)
{
habanalabs/gaudi2: Remove unnecessary (void*) conversions The void pointer object can be directly assigned to different structure objects, it does not need to be cast. Signed-off-by: Li zeming <zeming@nfschina.com> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-09-19 10:20:54 +08:00
struct hl_cb *cb = data;
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
struct packet_msg_short *pkt;
u32 value, ctl, pkt_size = sizeof(*pkt);
pkt = (struct packet_msg_short *) (uintptr_t) (cb->kernel_address + size);
memset(pkt, 0, pkt_size);
/* Inc by 1, Mode ADD */
value = FIELD_PREP(GAUDI2_PKT_SHORT_VAL_SOB_SYNC_VAL_MASK, 1);
value |= FIELD_PREP(GAUDI2_PKT_SHORT_VAL_SOB_MOD_MASK, 1);
ctl = FIELD_PREP(GAUDI2_PKT_SHORT_CTL_ADDR_MASK, sob_id * 4);
ctl |= FIELD_PREP(GAUDI2_PKT_SHORT_CTL_BASE_MASK, 1); /* SOB base */
ctl |= FIELD_PREP(GAUDI2_PKT_CTL_OPCODE_MASK, PACKET_MSG_SHORT);
ctl |= FIELD_PREP(GAUDI2_PKT_CTL_EB_MASK, eb);
ctl |= FIELD_PREP(GAUDI2_PKT_CTL_MB_MASK, 1);
pkt->value = cpu_to_le32(value);
pkt->ctl = cpu_to_le32(ctl);
return size + pkt_size;
}
static u32 gaudi2_add_mon_msg_short(struct packet_msg_short *pkt, u32 value, u16 addr)
{
u32 ctl, pkt_size = sizeof(*pkt);
memset(pkt, 0, pkt_size);
ctl = FIELD_PREP(GAUDI2_PKT_SHORT_CTL_ADDR_MASK, addr);
ctl |= FIELD_PREP(GAUDI2_PKT_SHORT_CTL_BASE_MASK, 0); /* MON base */
ctl |= FIELD_PREP(GAUDI2_PKT_CTL_OPCODE_MASK, PACKET_MSG_SHORT);
ctl |= FIELD_PREP(GAUDI2_PKT_CTL_EB_MASK, 0);
ctl |= FIELD_PREP(GAUDI2_PKT_CTL_MB_MASK, 0);
pkt->value = cpu_to_le32(value);
pkt->ctl = cpu_to_le32(ctl);
return pkt_size;
}
static u32 gaudi2_add_arm_monitor_pkt(struct hl_device *hdev, struct packet_msg_short *pkt,
u16 sob_base, u8 sob_mask, u16 sob_val, u16 addr)
{
u32 ctl, value, pkt_size = sizeof(*pkt);
u8 mask;
if (hl_gen_sob_mask(sob_base, sob_mask, &mask)) {
dev_err(hdev->dev, "sob_base %u (mask %#x) is not valid\n", sob_base, sob_mask);
return 0;
}
memset(pkt, 0, pkt_size);
value = FIELD_PREP(GAUDI2_PKT_SHORT_VAL_MON_SYNC_GID_MASK, sob_base / 8);
value |= FIELD_PREP(GAUDI2_PKT_SHORT_VAL_MON_SYNC_VAL_MASK, sob_val);
value |= FIELD_PREP(GAUDI2_PKT_SHORT_VAL_MON_MODE_MASK, 0); /* GREATER OR EQUAL*/
value |= FIELD_PREP(GAUDI2_PKT_SHORT_VAL_MON_MASK_MASK, mask);
ctl = FIELD_PREP(GAUDI2_PKT_SHORT_CTL_ADDR_MASK, addr);
ctl |= FIELD_PREP(GAUDI2_PKT_SHORT_CTL_BASE_MASK, 0); /* MON base */
ctl |= FIELD_PREP(GAUDI2_PKT_CTL_OPCODE_MASK, PACKET_MSG_SHORT);
ctl |= FIELD_PREP(GAUDI2_PKT_CTL_EB_MASK, 0);
ctl |= FIELD_PREP(GAUDI2_PKT_CTL_MB_MASK, 1);
pkt->value = cpu_to_le32(value);
pkt->ctl = cpu_to_le32(ctl);
return pkt_size;
}
static u32 gaudi2_add_fence_pkt(struct packet_fence *pkt)
{
u32 ctl, cfg, pkt_size = sizeof(*pkt);
memset(pkt, 0, pkt_size);
cfg = FIELD_PREP(GAUDI2_PKT_FENCE_CFG_DEC_VAL_MASK, 1);
cfg |= FIELD_PREP(GAUDI2_PKT_FENCE_CFG_TARGET_VAL_MASK, 1);
cfg |= FIELD_PREP(GAUDI2_PKT_FENCE_CFG_ID_MASK, 2);
ctl = FIELD_PREP(GAUDI2_PKT_CTL_OPCODE_MASK, PACKET_FENCE);
ctl |= FIELD_PREP(GAUDI2_PKT_CTL_EB_MASK, 0);
ctl |= FIELD_PREP(GAUDI2_PKT_CTL_MB_MASK, 1);
pkt->cfg = cpu_to_le32(cfg);
pkt->ctl = cpu_to_le32(ctl);
return pkt_size;
}
static u32 gaudi2_gen_wait_cb(struct hl_device *hdev, struct hl_gen_wait_properties *prop)
{
habanalabs/gaudi2: Remove unnecessary (void*) conversions The void pointer object can be directly assigned to different structure objects, it does not need to be cast. Signed-off-by: Li zeming <zeming@nfschina.com> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-09-19 10:20:54 +08:00
struct hl_cb *cb = prop->data;
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
void *buf = (void *) (uintptr_t) (cb->kernel_address);
u64 monitor_base, fence_addr = 0;
u32 stream_index, size = prop->size;
u16 msg_addr_offset;
stream_index = prop->q_idx % 4;
fence_addr = CFG_BASE + gaudi2_qm_blocks_bases[prop->q_idx] +
QM_FENCE2_OFFSET + stream_index * 4;
/*
* monitor_base should be the content of the base0 address registers,
* so it will be added to the msg short offsets
*/
monitor_base = mmDCORE0_SYNC_MNGR_OBJS_MON_PAY_ADDRL_0;
/* First monitor config packet: low address of the sync */
msg_addr_offset = (mmDCORE0_SYNC_MNGR_OBJS_MON_PAY_ADDRL_0 + prop->mon_id * 4) -
monitor_base;
size += gaudi2_add_mon_msg_short(buf + size, (u32) fence_addr, msg_addr_offset);
/* Second monitor config packet: high address of the sync */
msg_addr_offset = (mmDCORE0_SYNC_MNGR_OBJS_MON_PAY_ADDRH_0 + prop->mon_id * 4) -
monitor_base;
size += gaudi2_add_mon_msg_short(buf + size, (u32) (fence_addr >> 32), msg_addr_offset);
/*
* Third monitor config packet: the payload, i.e. what to write when the
* sync triggers
*/
msg_addr_offset = (mmDCORE0_SYNC_MNGR_OBJS_MON_PAY_DATA_0 + prop->mon_id * 4) -
monitor_base;
size += gaudi2_add_mon_msg_short(buf + size, 1, msg_addr_offset);
/* Fourth monitor config packet: bind the monitor to a sync object */
msg_addr_offset = (mmDCORE0_SYNC_MNGR_OBJS_MON_ARM_0 + prop->mon_id * 4) - monitor_base;
size += gaudi2_add_arm_monitor_pkt(hdev, buf + size, prop->sob_base, prop->sob_mask,
prop->sob_val, msg_addr_offset);
/* Fence packet */
size += gaudi2_add_fence_pkt(buf + size);
return size;
}
static void gaudi2_reset_sob(struct hl_device *hdev, void *data)
{
habanalabs/gaudi2: Remove unnecessary (void*) conversions The void pointer object can be directly assigned to different structure objects, it does not need to be cast. Signed-off-by: Li zeming <zeming@nfschina.com> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-09-19 10:20:54 +08:00
struct hl_hw_sob *hw_sob = data;
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
dev_dbg(hdev->dev, "reset SOB, q_idx: %d, sob_id: %d\n", hw_sob->q_idx, hw_sob->sob_id);
WREG32(mmDCORE0_SYNC_MNGR_OBJS_SOB_OBJ_0 + hw_sob->sob_id * 4, 0);
kref_init(&hw_sob->kref);
}
static void gaudi2_reset_sob_group(struct hl_device *hdev, u16 sob_group)
{
}
static u64 gaudi2_get_device_time(struct hl_device *hdev)
{
u64 device_time = ((u64) RREG32(mmPSOC_TIMESTAMP_CNTCVU)) << 32;
return device_time | RREG32(mmPSOC_TIMESTAMP_CNTCVL);
}
static int gaudi2_collective_wait_init_cs(struct hl_cs *cs)
{
return 0;
}
static int gaudi2_collective_wait_create_jobs(struct hl_device *hdev, struct hl_ctx *ctx,
struct hl_cs *cs, u32 wait_queue_id,
u32 collective_engine_id, u32 encaps_signal_offset)
{
return -EINVAL;
}
/*
* hl_mmu_scramble - converts a dram (non power of 2) page-size aligned address
* to DMMU page-size address (64MB) before mapping it in
* the MMU.
* The operation is performed on both the virtual and physical addresses.
* for device with 6 HBMs the scramble is:
* (addr[47:0] / 48M) * 64M + addr % 48M + addr[63:48]
*
* Example:
* =============================================================================
* Allocated DRAM Reserved VA scrambled VA for MMU mapping Scrambled PA
* Phys address in MMU last
* HOP
* =============================================================================
* PA1 0x3000000 VA1 0x9C000000 SVA1= (VA1/48M)*64M 0xD0000000 <- PA1/48M 0x1
* PA2 0x9000000 VA2 0x9F000000 SVA2= (VA2/48M)*64M 0xD4000000 <- PA2/48M 0x3
* =============================================================================
*/
static u64 gaudi2_mmu_scramble_addr(struct hl_device *hdev, u64 raw_addr)
{
struct asic_fixed_properties *prop = &hdev->asic_prop;
u32 divisor, mod_va;
u64 div_va;
/* accept any address in the DRAM address space */
if (hl_mem_area_inside_range(raw_addr, sizeof(raw_addr), DRAM_PHYS_BASE,
VA_HBM_SPACE_END)) {
divisor = prop->num_functional_hbms * GAUDI2_HBM_MMU_SCRM_MEM_SIZE;
div_va = div_u64_rem(raw_addr & GAUDI2_HBM_MMU_SCRM_ADDRESS_MASK, divisor, &mod_va);
return (raw_addr & ~GAUDI2_HBM_MMU_SCRM_ADDRESS_MASK) |
(div_va << GAUDI2_HBM_MMU_SCRM_DIV_SHIFT) |
(mod_va << GAUDI2_HBM_MMU_SCRM_MOD_SHIFT);
}
return raw_addr;
}
static u64 gaudi2_mmu_descramble_addr(struct hl_device *hdev, u64 scrambled_addr)
{
struct asic_fixed_properties *prop = &hdev->asic_prop;
u32 divisor, mod_va;
u64 div_va;
/* accept any address in the DRAM address space */
if (hl_mem_area_inside_range(scrambled_addr, sizeof(scrambled_addr), DRAM_PHYS_BASE,
VA_HBM_SPACE_END)) {
divisor = prop->num_functional_hbms * GAUDI2_HBM_MMU_SCRM_MEM_SIZE;
div_va = div_u64_rem(scrambled_addr & GAUDI2_HBM_MMU_SCRM_ADDRESS_MASK,
PAGE_SIZE_64MB, &mod_va);
return ((scrambled_addr & ~GAUDI2_HBM_MMU_SCRM_ADDRESS_MASK) +
(div_va * divisor + mod_va));
}
return scrambled_addr;
}
static u32 gaudi2_get_dec_base_addr(struct hl_device *hdev, u32 core_id)
{
u32 base = 0, dcore_id, dec_id;
if (core_id >= NUMBER_OF_DEC) {
dev_err(hdev->dev, "Unexpected core number %d for DEC\n", core_id);
goto out;
}
if (core_id < 8) {
dcore_id = core_id / NUM_OF_DEC_PER_DCORE;
dec_id = core_id % NUM_OF_DEC_PER_DCORE;
base = mmDCORE0_DEC0_CMD_BASE + dcore_id * DCORE_OFFSET +
dec_id * DCORE_VDEC_OFFSET;
} else {
/* PCIe Shared Decoder */
base = mmPCIE_DEC0_CMD_BASE + ((core_id % 8) * PCIE_VDEC_OFFSET);
}
out:
return base;
}
static int gaudi2_get_hw_block_id(struct hl_device *hdev, u64 block_addr,
u32 *block_size, u32 *block_id)
{
struct gaudi2_device *gaudi2 = hdev->asic_specific;
int i;
for (i = 0 ; i < NUM_USER_MAPPED_BLOCKS ; i++) {
if (block_addr == CFG_BASE + gaudi2->mapped_blocks[i].address) {
*block_id = i;
if (block_size)
*block_size = gaudi2->mapped_blocks[i].size;
return 0;
}
}
dev_err(hdev->dev, "Invalid block address %#llx", block_addr);
return -EINVAL;
}
static int gaudi2_block_mmap(struct hl_device *hdev, struct vm_area_struct *vma,
u32 block_id, u32 block_size)
{
struct gaudi2_device *gaudi2 = hdev->asic_specific;
u64 offset_in_bar;
u64 address;
int rc;
if (block_id >= NUM_USER_MAPPED_BLOCKS) {
dev_err(hdev->dev, "Invalid block id %u", block_id);
return -EINVAL;
}
/* we allow mapping only an entire block */
if (block_size != gaudi2->mapped_blocks[block_id].size) {
dev_err(hdev->dev, "Invalid block size %u", block_size);
return -EINVAL;
}
offset_in_bar = CFG_BASE + gaudi2->mapped_blocks[block_id].address - STM_FLASH_BASE_ADDR;
address = pci_resource_start(hdev->pdev, SRAM_CFG_BAR_ID) + offset_in_bar;
vma->vm_flags |= VM_IO | VM_PFNMAP | VM_DONTEXPAND | VM_DONTDUMP |
VM_DONTCOPY | VM_NORESERVE;
rc = remap_pfn_range(vma, vma->vm_start, address >> PAGE_SHIFT,
block_size, vma->vm_page_prot);
if (rc)
dev_err(hdev->dev, "remap_pfn_range error %d", rc);
return rc;
}
static void gaudi2_enable_events_from_fw(struct hl_device *hdev)
{
struct gaudi2_device *gaudi2 = hdev->asic_specific;
struct cpu_dyn_regs *dyn_regs = &hdev->fw_loader.dynamic_loader.comm_desc.cpu_dyn_regs;
u32 irq_handler_offset = le32_to_cpu(dyn_regs->gic_host_ints_irq);
if (gaudi2->hw_cap_initialized & HW_CAP_CPU_Q)
WREG32(irq_handler_offset,
gaudi2_irq_map_table[GAUDI2_EVENT_CPU_INTS_REGISTER].cpu_id);
}
static int gaudi2_get_mmu_base(struct hl_device *hdev, u64 mmu_id, u32 *mmu_base)
{
switch (mmu_id) {
case HW_CAP_DCORE0_DMMU0:
*mmu_base = mmDCORE0_HMMU0_MMU_BASE;
break;
case HW_CAP_DCORE0_DMMU1:
*mmu_base = mmDCORE0_HMMU1_MMU_BASE;
break;
case HW_CAP_DCORE0_DMMU2:
*mmu_base = mmDCORE0_HMMU2_MMU_BASE;
break;
case HW_CAP_DCORE0_DMMU3:
*mmu_base = mmDCORE0_HMMU3_MMU_BASE;
break;
case HW_CAP_DCORE1_DMMU0:
*mmu_base = mmDCORE1_HMMU0_MMU_BASE;
break;
case HW_CAP_DCORE1_DMMU1:
*mmu_base = mmDCORE1_HMMU1_MMU_BASE;
break;
case HW_CAP_DCORE1_DMMU2:
*mmu_base = mmDCORE1_HMMU2_MMU_BASE;
break;
case HW_CAP_DCORE1_DMMU3:
*mmu_base = mmDCORE1_HMMU3_MMU_BASE;
break;
case HW_CAP_DCORE2_DMMU0:
*mmu_base = mmDCORE2_HMMU0_MMU_BASE;
break;
case HW_CAP_DCORE2_DMMU1:
*mmu_base = mmDCORE2_HMMU1_MMU_BASE;
break;
case HW_CAP_DCORE2_DMMU2:
*mmu_base = mmDCORE2_HMMU2_MMU_BASE;
break;
case HW_CAP_DCORE2_DMMU3:
*mmu_base = mmDCORE2_HMMU3_MMU_BASE;
break;
case HW_CAP_DCORE3_DMMU0:
*mmu_base = mmDCORE3_HMMU0_MMU_BASE;
break;
case HW_CAP_DCORE3_DMMU1:
*mmu_base = mmDCORE3_HMMU1_MMU_BASE;
break;
case HW_CAP_DCORE3_DMMU2:
*mmu_base = mmDCORE3_HMMU2_MMU_BASE;
break;
case HW_CAP_DCORE3_DMMU3:
*mmu_base = mmDCORE3_HMMU3_MMU_BASE;
break;
case HW_CAP_PMMU:
*mmu_base = mmPMMU_HBW_MMU_BASE;
break;
default:
return -EINVAL;
}
return 0;
}
static void gaudi2_ack_mmu_error(struct hl_device *hdev, u64 mmu_id)
{
habanalabs: Simplify bool conversion Fix the following coccicheck warning: ./drivers/misc/habanalabs/gaudi2/gaudi2.c:9727:48-53: WARNING: conversion to bool not needed here Signed-off-by: Yang Li <yang.lee@linux.alibaba.com> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-07-14 09:29:08 +08:00
bool is_pmmu = (mmu_id == HW_CAP_PMMU);
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
struct gaudi2_device *gaudi2 = hdev->asic_specific;
u32 mmu_base;
if (!(gaudi2->hw_cap_initialized & mmu_id))
return;
if (gaudi2_get_mmu_base(hdev, mmu_id, &mmu_base))
return;
gaudi2_handle_page_error(hdev, mmu_base, is_pmmu);
gaudi2_handle_access_error(hdev, mmu_base, is_pmmu);
}
static int gaudi2_ack_mmu_page_fault_or_access_error(struct hl_device *hdev, u64 mmu_cap_mask)
{
u32 i, mmu_id, num_of_hmmus = NUM_OF_HMMU_PER_DCORE * NUM_OF_DCORES;
/* check all HMMUs */
for (i = 0 ; i < num_of_hmmus ; i++) {
mmu_id = HW_CAP_DCORE0_DMMU0 << i;
if (mmu_cap_mask & mmu_id)
gaudi2_ack_mmu_error(hdev, mmu_id);
}
/* check PMMU */
if (mmu_cap_mask & HW_CAP_PMMU)
gaudi2_ack_mmu_error(hdev, HW_CAP_PMMU);
return 0;
}
static void gaudi2_get_msi_info(__le32 *table)
{
table[CPUCP_EVENT_QUEUE_MSI_TYPE] = cpu_to_le32(GAUDI2_EVENT_QUEUE_MSIX_IDX);
}
static int gaudi2_map_pll_idx_to_fw_idx(u32 pll_idx)
{
switch (pll_idx) {
case HL_GAUDI2_CPU_PLL: return CPU_PLL;
case HL_GAUDI2_PCI_PLL: return PCI_PLL;
case HL_GAUDI2_NIC_PLL: return NIC_PLL;
case HL_GAUDI2_DMA_PLL: return DMA_PLL;
case HL_GAUDI2_MESH_PLL: return MESH_PLL;
case HL_GAUDI2_MME_PLL: return MME_PLL;
case HL_GAUDI2_TPC_PLL: return TPC_PLL;
case HL_GAUDI2_IF_PLL: return IF_PLL;
case HL_GAUDI2_SRAM_PLL: return SRAM_PLL;
case HL_GAUDI2_HBM_PLL: return HBM_PLL;
case HL_GAUDI2_VID_PLL: return VID_PLL;
case HL_GAUDI2_MSS_PLL: return MSS_PLL;
default: return -EINVAL;
}
}
static int gaudi2_gen_sync_to_engine_map(struct hl_device *hdev, struct hl_sync_to_engine_map *map)
{
/* Not implemented */
return 0;
}
static int gaudi2_monitor_valid(struct hl_mon_state_dump *mon)
{
/* Not implemented */
return 0;
}
static int gaudi2_print_single_monitor(char **buf, size_t *size, size_t *offset,
struct hl_device *hdev, struct hl_mon_state_dump *mon)
{
/* Not implemented */
return 0;
}
static int gaudi2_print_fences_single_engine(struct hl_device *hdev, u64 base_offset,
u64 status_base_offset, enum hl_sync_engine_type engine_type,
u32 engine_id, char **buf, size_t *size, size_t *offset)
{
/* Not implemented */
return 0;
}
static struct hl_state_dump_specs_funcs gaudi2_state_dump_funcs = {
.monitor_valid = gaudi2_monitor_valid,
.print_single_monitor = gaudi2_print_single_monitor,
.gen_sync_to_engine_map = gaudi2_gen_sync_to_engine_map,
.print_fences_single_engine = gaudi2_print_fences_single_engine,
};
static void gaudi2_state_dump_init(struct hl_device *hdev)
{
/* Not implemented */
hdev->state_dump_specs.props = gaudi2_state_dump_specs_props;
hdev->state_dump_specs.funcs = gaudi2_state_dump_funcs;
}
static u32 gaudi2_get_sob_addr(struct hl_device *hdev, u32 sob_id)
{
return 0;
}
static u32 *gaudi2_get_stream_master_qid_arr(void)
{
return NULL;
}
static void gaudi2_add_device_attr(struct hl_device *hdev, struct attribute_group *dev_clk_attr_grp,
struct attribute_group *dev_vrm_attr_grp)
{
hl_sysfs_add_dev_clk_attr(hdev, dev_clk_attr_grp);
hl_sysfs_add_dev_vrm_attr(hdev, dev_vrm_attr_grp);
}
static int gaudi2_mmu_get_real_page_size(struct hl_device *hdev, struct hl_mmu_properties *mmu_prop,
u32 page_size, u32 *real_page_size, bool is_dram_addr)
{
struct asic_fixed_properties *prop = &hdev->asic_prop;
/* for host pages the page size must be */
if (!is_dram_addr) {
if (page_size % mmu_prop->page_size)
goto page_size_err;
*real_page_size = mmu_prop->page_size;
return 0;
}
if ((page_size % prop->dram_page_size) || (prop->dram_page_size > mmu_prop->page_size))
goto page_size_err;
/*
* MMU page size is different from DRAM page size (more precisely, DMMU page is greater
* than DRAM page size).
* for this reason work with the DRAM page size and let the MMU scrambling routine handle
* this mismatch when calculating the address to place in the MMU page table.
* (in that case also make sure that the dram_page_size is not greater than the
* mmu page size)
*/
*real_page_size = prop->dram_page_size;
return 0;
page_size_err:
dev_err(hdev->dev, "page size of %u is not %uKB aligned, can't map\n",
page_size, mmu_prop->page_size >> 10);
return -EFAULT;
}
static int gaudi2_get_monitor_dump(struct hl_device *hdev, void *data)
{
return -EOPNOTSUPP;
}
habanalabs: send device active message to f/w As part of the RAS that is done by the f/w, we should send a message to the f/w when a user either acquires or releases the device. Signed-off-by: farah kassabri <fkassabri@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-08-17 22:43:43 +08:00
int gaudi2_send_device_activity(struct hl_device *hdev, bool open)
{
struct gaudi2_device *gaudi2 = hdev->asic_specific;
if (!(gaudi2->hw_cap_initialized & HW_CAP_CPU_Q) || hdev->fw_major_version < 37)
return 0;
/* TODO: add check for FW version using minor ver once it's known */
return hl_fw_send_device_activity(hdev, open);
}
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
static const struct hl_asic_funcs gaudi2_funcs = {
.early_init = gaudi2_early_init,
.early_fini = gaudi2_early_fini,
.late_init = gaudi2_late_init,
.late_fini = gaudi2_late_fini,
.sw_init = gaudi2_sw_init,
.sw_fini = gaudi2_sw_fini,
.hw_init = gaudi2_hw_init,
.hw_fini = gaudi2_hw_fini,
.halt_engines = gaudi2_halt_engines,
.suspend = gaudi2_suspend,
.resume = gaudi2_resume,
.mmap = gaudi2_mmap,
.ring_doorbell = gaudi2_ring_doorbell,
.pqe_write = gaudi2_pqe_write,
.asic_dma_alloc_coherent = gaudi2_dma_alloc_coherent,
.asic_dma_free_coherent = gaudi2_dma_free_coherent,
.scrub_device_mem = gaudi2_scrub_device_mem,
.scrub_device_dram = gaudi2_scrub_device_dram,
.get_int_queue_base = NULL,
.test_queues = gaudi2_test_queues,
.asic_dma_pool_zalloc = gaudi2_dma_pool_zalloc,
.asic_dma_pool_free = gaudi2_dma_pool_free,
.cpu_accessible_dma_pool_alloc = gaudi2_cpu_accessible_dma_pool_alloc,
.cpu_accessible_dma_pool_free = gaudi2_cpu_accessible_dma_pool_free,
.asic_dma_unmap_single = gaudi2_dma_unmap_single,
.asic_dma_map_single = gaudi2_dma_map_single,
.hl_dma_unmap_sgtable = hl_dma_unmap_sgtable,
.cs_parser = gaudi2_cs_parser,
.asic_dma_map_sgtable = hl_dma_map_sgtable,
.add_end_of_cb_packets = NULL,
.update_eq_ci = gaudi2_update_eq_ci,
.context_switch = gaudi2_context_switch,
.restore_phase_topology = gaudi2_restore_phase_topology,
.debugfs_read_dma = gaudi2_debugfs_read_dma,
.add_device_attr = gaudi2_add_device_attr,
.handle_eqe = gaudi2_handle_eqe,
.get_events_stat = gaudi2_get_events_stat,
.read_pte = NULL,
.write_pte = NULL,
.mmu_invalidate_cache = gaudi2_mmu_invalidate_cache,
.mmu_invalidate_cache_range = gaudi2_mmu_invalidate_cache_range,
.mmu_prefetch_cache_range = NULL,
.send_heartbeat = gaudi2_send_heartbeat,
habanalabs/gaudi2: add gaudi2 profiler module Add the Gaudi2 code to initialize the ASIC's profiler. The profile receives its initialization values from the user, same as in Gaudi2, but the code to initialize is in the driver because the configuration space of the device is not directly exposed to the user. Signed-off-by: Benjamin Dotan <bdotan@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-27 02:35:07 +08:00
.debug_coresight = gaudi2_debug_coresight,
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
.is_device_idle = gaudi2_is_device_idle,
habanalabs: rename non_hard_reset to compute_reset In order to be more explicit we should use the term compute_reset for describing the reset in which only the compute engines gets reset. Signed-off-by: Ofir Bitton <obitton@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-07-13 20:08:09 +08:00
.compute_reset_late_init = gaudi2_compute_reset_late_init,
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
.hw_queues_lock = gaudi2_hw_queues_lock,
.hw_queues_unlock = gaudi2_hw_queues_unlock,
.get_pci_id = gaudi2_get_pci_id,
.get_eeprom_data = gaudi2_get_eeprom_data,
.get_monitor_dump = gaudi2_get_monitor_dump,
.send_cpu_message = gaudi2_send_cpu_message,
.pci_bars_map = gaudi2_pci_bars_map,
.init_iatu = gaudi2_init_iatu,
.rreg = hl_rreg,
.wreg = hl_wreg,
habanalabs/gaudi2: add gaudi2 profiler module Add the Gaudi2 code to initialize the ASIC's profiler. The profile receives its initialization values from the user, same as in Gaudi2, but the code to initialize is in the driver because the configuration space of the device is not directly exposed to the user. Signed-off-by: Benjamin Dotan <bdotan@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-27 02:35:07 +08:00
.halt_coresight = gaudi2_halt_coresight,
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
.ctx_init = gaudi2_ctx_init,
.ctx_fini = gaudi2_ctx_fini,
.pre_schedule_cs = gaudi2_pre_schedule_cs,
.get_queue_id_for_cq = gaudi2_get_queue_id_for_cq,
.load_firmware_to_device = NULL,
.load_boot_fit_to_device = NULL,
.get_signal_cb_size = gaudi2_get_signal_cb_size,
.get_wait_cb_size = gaudi2_get_wait_cb_size,
.gen_signal_cb = gaudi2_gen_signal_cb,
.gen_wait_cb = gaudi2_gen_wait_cb,
.reset_sob = gaudi2_reset_sob,
.reset_sob_group = gaudi2_reset_sob_group,
.get_device_time = gaudi2_get_device_time,
habanalabs/gaudi2: add gaudi2 security module Use the generic security module to block all registers in the ASIC and then open only those that are needed to be accessed by the user. Signed-off-by: Ofir Bitton <obitton@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-27 02:30:25 +08:00
.pb_print_security_errors = gaudi2_pb_print_security_errors,
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
.collective_wait_init_cs = gaudi2_collective_wait_init_cs,
.collective_wait_create_jobs = gaudi2_collective_wait_create_jobs,
.get_dec_base_addr = gaudi2_get_dec_base_addr,
.scramble_addr = gaudi2_mmu_scramble_addr,
.descramble_addr = gaudi2_mmu_descramble_addr,
habanalabs/gaudi2: add gaudi2 security module Use the generic security module to block all registers in the ASIC and then open only those that are needed to be accessed by the user. Signed-off-by: Ofir Bitton <obitton@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-27 02:30:25 +08:00
.ack_protection_bits_errors = gaudi2_ack_protection_bits_errors,
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
.get_hw_block_id = gaudi2_get_hw_block_id,
.hw_block_mmap = gaudi2_block_mmap,
.enable_events_from_fw = gaudi2_enable_events_from_fw,
.ack_mmu_errors = gaudi2_ack_mmu_page_fault_or_access_error,
.get_msi_info = gaudi2_get_msi_info,
.map_pll_idx_to_fw_idx = gaudi2_map_pll_idx_to_fw_idx,
habanalabs: wait for preboot ready after hard reset Currently we are not waiting for preboot ready after hard reset. This leads to a race in which COMMs protocol begins but will get no response from the f/w. Signed-off-by: Ohad Sharabi <osharabi@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-28 17:09:21 +08:00
.init_firmware_preload_params = gaudi2_init_firmware_preload_params,
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
.init_firmware_loader = gaudi2_init_firmware_loader,
.init_cpu_scrambler_dram = gaudi2_init_scrambler_hbm,
.state_dump_init = gaudi2_state_dump_init,
.get_sob_addr = &gaudi2_get_sob_addr,
.set_pci_memory_regions = gaudi2_set_pci_memory_regions,
.get_stream_master_qid_arr = gaudi2_get_stream_master_qid_arr,
.check_if_razwi_happened = gaudi2_check_if_razwi_happened,
.mmu_get_real_page_size = gaudi2_mmu_get_real_page_size,
.access_dev_mem = hl_access_dev_mem,
.set_dram_bar_base = gaudi2_set_hbm_bar_base,
habanalabs/gaudi2: new API to control engine cores running mode The current flow of halting the engine cores is implemented by command buffers built by the user space and sent towards the Driver. This current flow is broken since the user space does not know when the cores actually halt as sending a workload is async op. Therefore the application can not free the memory that is mapped to the engine cores. This new API allows the user space to control the running mode. The API call is sync (returns after the cores are set to the requested mode). Signed-off-by: Tal Cohen <talcohen@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-07-07 23:42:47 +08:00
.set_engine_cores = gaudi2_set_engine_cores,
habanalabs: send device active message to f/w As part of the RAS that is done by the f/w, we should send a message to the f/w when a user either acquires or releases the device. Signed-off-by: farah kassabri <fkassabri@habana.ai> Reviewed-by: Oded Gabbay <ogabbay@kernel.org> Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-08-17 22:43:43 +08:00
.send_device_activity = gaudi2_send_device_activity,
habanalabs: add gaudi2 asic-specific code Add the ASIC-specific code for Gaudi2. Supply (almost) all of the function callbacks that the driver's common code need to initialize, finalize and submit workloads to the Gaudi2 ASIC. It also contains the code to initialize the F/W of the Gaudi2 ASIC and to receive events from the F/W. It contains new debugfs entry to dump razwi events. razwi is a case where the device's engines create a transaction that reaches an invalid destination. Signed-off-by: Oded Gabbay <ogabbay@kernel.org>
2022-06-26 23:20:03 +08:00
};
void gaudi2_set_asic_funcs(struct hl_device *hdev)
{
hdev->asic_funcs = &gaudi2_funcs;
}
Reference in New Issue Copy Permalink