diff --git a/arch/tile/Kconfig b/arch/tile/Kconfig index 0ad771f7a7e1..557e3a381ca0 100644 --- a/arch/tile/Kconfig +++ b/arch/tile/Kconfig @@ -3,6 +3,8 @@ config TILE def_bool y + select HAVE_DMA_ATTRS + select HAVE_DMA_API_DEBUG select HAVE_KVM if !TILEGX select GENERIC_FIND_FIRST_BIT select USE_GENERIC_SMP_HELPERS @@ -79,6 +81,9 @@ config ARCH_DMA_ADDR_T_64BIT config NEED_DMA_MAP_STATE def_bool y +config ARCH_HAS_DMA_SET_COHERENT_MASK + bool + config LOCKDEP_SUPPORT def_bool y @@ -215,6 +220,19 @@ config HIGHMEM config ZONE_DMA def_bool y +config IOMMU_HELPER + bool + +config NEED_SG_DMA_LENGTH + bool + +config SWIOTLB + bool + default TILEGX + select IOMMU_HELPER + select NEED_SG_DMA_LENGTH + select ARCH_HAS_DMA_SET_COHERENT_MASK + # We do not currently support disabling NUMA. config NUMA bool # "NUMA Memory Allocation and Scheduler Support" diff --git a/arch/tile/include/asm/Kbuild b/arch/tile/include/asm/Kbuild index 143473e3a0bb..fb7c65ae8de0 100644 --- a/arch/tile/include/asm/Kbuild +++ b/arch/tile/include/asm/Kbuild @@ -9,7 +9,6 @@ header-y += hardwall.h generic-y += bug.h generic-y += bugs.h generic-y += cputime.h -generic-y += device.h generic-y += div64.h generic-y += emergency-restart.h generic-y += errno.h diff --git a/arch/tile/include/asm/device.h b/arch/tile/include/asm/device.h new file mode 100644 index 000000000000..5182705bd056 --- /dev/null +++ b/arch/tile/include/asm/device.h @@ -0,0 +1,33 @@ +/* + * Copyright 2010 Tilera Corporation. All Rights Reserved. + * + * This program is free software; you can redistribute it and/or + * modify it under the terms of the GNU General Public License + * as published by the Free Software Foundation, version 2. + * + * This program is distributed in the hope that it will be useful, but + * WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or + * NON INFRINGEMENT. See the GNU General Public License for + * more details. + * Arch specific extensions to struct device + */ + +#ifndef _ASM_TILE_DEVICE_H +#define _ASM_TILE_DEVICE_H + +struct dev_archdata { + /* DMA operations on that device */ + struct dma_map_ops *dma_ops; + + /* Offset of the DMA address from the PA. */ + dma_addr_t dma_offset; + + /* Highest DMA address that can be generated by this device. */ + dma_addr_t max_direct_dma_addr; +}; + +struct pdev_archdata { +}; + +#endif /* _ASM_TILE_DEVICE_H */ diff --git a/arch/tile/include/asm/dma-mapping.h b/arch/tile/include/asm/dma-mapping.h index eaa06d175b39..4b6247d1a315 100644 --- a/arch/tile/include/asm/dma-mapping.h +++ b/arch/tile/include/asm/dma-mapping.h @@ -20,69 +20,80 @@ #include #include -/* - * Note that on x86 and powerpc, there is a "struct dma_mapping_ops" - * that is used for all the DMA operations. For now, we don't have an - * equivalent on tile, because we only have a single way of doing DMA. - * (Tilera bug 7994 to use dma_mapping_ops.) - */ +extern struct dma_map_ops *tile_dma_map_ops; +extern struct dma_map_ops *gx_pci_dma_map_ops; +extern struct dma_map_ops *gx_legacy_pci_dma_map_ops; -#define dma_alloc_noncoherent(d, s, h, f) dma_alloc_coherent(d, s, h, f) -#define dma_free_noncoherent(d, s, v, h) dma_free_coherent(d, s, v, h) +static inline struct dma_map_ops *get_dma_ops(struct device *dev) +{ + if (dev && dev->archdata.dma_ops) + return dev->archdata.dma_ops; + else + return tile_dma_map_ops; +} -extern dma_addr_t dma_map_single(struct device *dev, void *ptr, size_t size, - enum dma_data_direction); -extern void dma_unmap_single(struct device *dev, dma_addr_t dma_addr, - size_t size, enum dma_data_direction); -extern int dma_map_sg(struct device *dev, struct scatterlist *sg, int nents, - enum dma_data_direction); -extern void dma_unmap_sg(struct device *dev, struct scatterlist *sg, - int nhwentries, enum dma_data_direction); -extern dma_addr_t dma_map_page(struct device *dev, struct page *page, - unsigned long offset, size_t size, - enum dma_data_direction); -extern void dma_unmap_page(struct device *dev, dma_addr_t dma_address, - size_t size, enum dma_data_direction); -extern void dma_sync_sg_for_cpu(struct device *dev, struct scatterlist *sg, - int nelems, enum dma_data_direction); -extern void dma_sync_sg_for_device(struct device *dev, struct scatterlist *sg, - int nelems, enum dma_data_direction); +static inline dma_addr_t get_dma_offset(struct device *dev) +{ + return dev->archdata.dma_offset; +} +static inline void set_dma_offset(struct device *dev, dma_addr_t off) +{ + dev->archdata.dma_offset = off; +} -void *dma_alloc_coherent(struct device *dev, size_t size, - dma_addr_t *dma_handle, gfp_t flag); +static inline dma_addr_t phys_to_dma(struct device *dev, phys_addr_t paddr) +{ + return paddr + get_dma_offset(dev); +} -void dma_free_coherent(struct device *dev, size_t size, - void *vaddr, dma_addr_t dma_handle); +static inline phys_addr_t dma_to_phys(struct device *dev, dma_addr_t daddr) +{ + return daddr - get_dma_offset(dev); +} -extern void dma_sync_single_for_cpu(struct device *, dma_addr_t, size_t, - enum dma_data_direction); -extern void dma_sync_single_for_device(struct device *, dma_addr_t, - size_t, enum dma_data_direction); -extern void dma_sync_single_range_for_cpu(struct device *, dma_addr_t, - unsigned long offset, size_t, - enum dma_data_direction); -extern void dma_sync_single_range_for_device(struct device *, dma_addr_t, - unsigned long offset, size_t, - enum dma_data_direction); -extern void dma_cache_sync(struct device *dev, void *vaddr, size_t, - enum dma_data_direction); +static inline void dma_mark_clean(void *addr, size_t size) {} + +#include + +static inline void set_dma_ops(struct device *dev, struct dma_map_ops *ops) +{ + dev->archdata.dma_ops = ops; +} + +static inline bool dma_capable(struct device *dev, dma_addr_t addr, size_t size) +{ + if (!dev->dma_mask) + return 0; + + return addr + size - 1 <= *dev->dma_mask; +} static inline int dma_mapping_error(struct device *dev, dma_addr_t dma_addr) { - return 0; + return get_dma_ops(dev)->mapping_error(dev, dma_addr); } static inline int dma_supported(struct device *dev, u64 mask) { - return 1; + return get_dma_ops(dev)->dma_supported(dev, mask); } static inline int dma_set_mask(struct device *dev, u64 mask) { + struct dma_map_ops *dma_ops = get_dma_ops(dev); + + /* Handle legacy PCI devices with limited memory addressability. */ + if ((dma_ops == gx_pci_dma_map_ops) && (mask <= DMA_BIT_MASK(32))) { + set_dma_ops(dev, gx_legacy_pci_dma_map_ops); + set_dma_offset(dev, 0); + if (mask > dev->archdata.max_direct_dma_addr) + mask = dev->archdata.max_direct_dma_addr; + } + if (!dev->dma_mask || !dma_supported(dev, mask)) return -EIO; @@ -91,4 +102,43 @@ dma_set_mask(struct device *dev, u64 mask) return 0; } +static inline void *dma_alloc_attrs(struct device *dev, size_t size, + dma_addr_t *dma_handle, gfp_t flag, + struct dma_attrs *attrs) +{ + struct dma_map_ops *dma_ops = get_dma_ops(dev); + void *cpu_addr; + + cpu_addr = dma_ops->alloc(dev, size, dma_handle, flag, attrs); + + debug_dma_alloc_coherent(dev, size, *dma_handle, cpu_addr); + + return cpu_addr; +} + +static inline void dma_free_attrs(struct device *dev, size_t size, + void *cpu_addr, dma_addr_t dma_handle, + struct dma_attrs *attrs) +{ + struct dma_map_ops *dma_ops = get_dma_ops(dev); + + debug_dma_free_coherent(dev, size, cpu_addr, dma_handle); + + dma_ops->free(dev, size, cpu_addr, dma_handle, attrs); +} + +#define dma_alloc_coherent(d, s, h, f) dma_alloc_attrs(d, s, h, f, NULL) +#define dma_alloc_noncoherent(d, s, h, f) dma_alloc_attrs(d, s, h, f, NULL) +#define dma_free_coherent(d, s, v, h) dma_free_attrs(d, s, v, h, NULL) +#define dma_free_noncoherent(d, s, v, h) dma_free_attrs(d, s, v, h, NULL) + +/* + * dma_alloc_noncoherent() is #defined to return coherent memory, + * so there's no need to do any flushing here. + */ +static inline void dma_cache_sync(struct device *dev, void *vaddr, size_t size, + enum dma_data_direction direction) +{ +} + #endif /* _ASM_TILE_DMA_MAPPING_H */ diff --git a/arch/tile/include/asm/pci.h b/arch/tile/include/asm/pci.h index 2c224c47d8ab..553b7ff018c4 100644 --- a/arch/tile/include/asm/pci.h +++ b/arch/tile/include/asm/pci.h @@ -15,6 +15,7 @@ #ifndef _ASM_TILE_PCI_H #define _ASM_TILE_PCI_H +#include #include #include #include @@ -53,6 +54,16 @@ static inline void pci_iounmap(struct pci_dev *dev, void __iomem *addr) {} #define TILE_NUM_PCIE 2 +/* + * The hypervisor maps the entirety of CPA-space as bus addresses, so + * bus addresses are physical addresses. The networking and block + * device layers use this boolean for bounce buffer decisions. + */ +#define PCI_DMA_BUS_IS_PHYS 1 + +/* generic pci stuff */ +#include + #else #include @@ -85,7 +96,47 @@ static inline void pci_iounmap(struct pci_dev *dev, void __iomem *addr) {} /* * Each Mem-Map interrupt region occupies 4KB. */ -#define MEM_MAP_INTR_REGION_SIZE (1<< TRIO_MAP_MEM_LIM__ADDR_SHIFT) +#define MEM_MAP_INTR_REGION_SIZE (1 << TRIO_MAP_MEM_LIM__ADDR_SHIFT) + +/* + * Allocate the PCI BAR window right below 4GB. + */ +#define TILE_PCI_BAR_WINDOW_TOP (1ULL << 32) + +/* + * Allocate 1GB for the PCI BAR window. + */ +#define TILE_PCI_BAR_WINDOW_SIZE (1 << 30) + +/* + * This is the highest bus address targeting the host memory that + * can be generated by legacy PCI devices with 32-bit or less + * DMA capability, dictated by the BAR window size and location. + */ +#define TILE_PCI_MAX_DIRECT_DMA_ADDRESS \ + (TILE_PCI_BAR_WINDOW_TOP - TILE_PCI_BAR_WINDOW_SIZE - 1) + +/* + * We shift the PCI bus range for all the physical memory up by the whole PA + * range. The corresponding CPA of an incoming PCI request will be the PCI + * address minus TILE_PCI_MEM_MAP_BASE_OFFSET. This also implies + * that the 64-bit capable devices will be given DMA addresses as + * the CPA plus TILE_PCI_MEM_MAP_BASE_OFFSET. To support 32-bit + * devices, we create a separate map region that handles the low + * 4GB. + */ +#define TILE_PCI_MEM_MAP_BASE_OFFSET (1ULL << CHIP_PA_WIDTH()) + +/* + * End of the PCI memory resource. + */ +#define TILE_PCI_MEM_END \ + ((1ULL << CHIP_PA_WIDTH()) + TILE_PCI_BAR_WINDOW_TOP) + +/* + * Start of the PCI memory resource. + */ +#define TILE_PCI_MEM_START (TILE_PCI_MEM_END - TILE_PCI_BAR_WINDOW_SIZE) /* * Structure of a PCI controller (host bridge) on Gx. @@ -108,6 +159,8 @@ struct pci_controller { int index; /* PCI domain number */ struct pci_bus *root_bus; + uint64_t mem_offset; /* cpu->bus memory mapping offset. */ + int last_busno; struct pci_ops *ops; @@ -126,14 +179,22 @@ extern gxio_trio_context_t trio_contexts[TILEGX_NUM_TRIO]; extern void pci_iounmap(struct pci_dev *dev, void __iomem *); -#endif /* __tilegx__ */ +extern void +pcibios_resource_to_bus(struct pci_dev *dev, struct pci_bus_region *region, + struct resource *res); + +extern void +pcibios_bus_to_resource(struct pci_dev *dev, struct resource *res, + struct pci_bus_region *region); /* - * The hypervisor maps the entirety of CPA-space as bus addresses, so - * bus addresses are physical addresses. The networking and block - * device layers use this boolean for bounce buffer decisions. + * The PCI address space does not equal the physical memory address + * space (we have an IOMMU). The IDE and SCSI device layers use this + * boolean for bounce buffer decisions. */ -#define PCI_DMA_BUS_IS_PHYS 1 +#define PCI_DMA_BUS_IS_PHYS 0 + +#endif /* __tilegx__ */ int __init tile_pci_init(void); int __init pcibios_init(void); @@ -169,7 +230,4 @@ static inline int pcibios_assign_all_busses(void) /* implement the pci_ DMA API in terms of the generic device dma_ one */ #include -/* generic pci stuff */ -#include - #endif /* _ASM_TILE_PCI_H */ diff --git a/arch/tile/kernel/pci-dma.c b/arch/tile/kernel/pci-dma.c index edd856a000c5..b9fe80ec1089 100644 --- a/arch/tile/kernel/pci-dma.c +++ b/arch/tile/kernel/pci-dma.c @@ -14,6 +14,7 @@ #include #include +#include #include #include #include @@ -31,10 +32,9 @@ #define PAGE_HOME_DMA PAGE_HOME_HASH #endif -void *dma_alloc_coherent(struct device *dev, - size_t size, - dma_addr_t *dma_handle, - gfp_t gfp) +static void *tile_dma_alloc_coherent(struct device *dev, size_t size, + dma_addr_t *dma_handle, gfp_t gfp, + struct dma_attrs *attrs) { u64 dma_mask = dev->coherent_dma_mask ?: DMA_BIT_MASK(32); int node = dev_to_node(dev); @@ -68,19 +68,19 @@ void *dma_alloc_coherent(struct device *dev, } *dma_handle = addr; + return page_address(pg); } -EXPORT_SYMBOL(dma_alloc_coherent); /* - * Free memory that was allocated with dma_alloc_coherent. + * Free memory that was allocated with tile_dma_alloc_coherent. */ -void dma_free_coherent(struct device *dev, size_t size, - void *vaddr, dma_addr_t dma_handle) +static void tile_dma_free_coherent(struct device *dev, size_t size, + void *vaddr, dma_addr_t dma_handle, + struct dma_attrs *attrs) { homecache_free_pages((unsigned long)vaddr, get_order(size)); } -EXPORT_SYMBOL(dma_free_coherent); /* * The map routines "map" the specified address range for DMA @@ -199,38 +199,9 @@ static void __dma_complete_pa_range(dma_addr_t dma_addr, size_t size, } } - -/* - * dma_map_single can be passed any memory address, and there appear - * to be no alignment constraints. - * - * There is a chance that the start of the buffer will share a cache - * line with some other data that has been touched in the meantime. - */ -dma_addr_t dma_map_single(struct device *dev, void *ptr, size_t size, - enum dma_data_direction direction) -{ - dma_addr_t dma_addr = __pa(ptr); - - BUG_ON(!valid_dma_direction(direction)); - WARN_ON(size == 0); - - __dma_prep_pa_range(dma_addr, size, direction); - - return dma_addr; -} -EXPORT_SYMBOL(dma_map_single); - -void dma_unmap_single(struct device *dev, dma_addr_t dma_addr, size_t size, - enum dma_data_direction direction) -{ - BUG_ON(!valid_dma_direction(direction)); - __dma_complete_pa_range(dma_addr, size, direction); -} -EXPORT_SYMBOL(dma_unmap_single); - -int dma_map_sg(struct device *dev, struct scatterlist *sglist, int nents, - enum dma_data_direction direction) +static int tile_dma_map_sg(struct device *dev, struct scatterlist *sglist, + int nents, enum dma_data_direction direction, + struct dma_attrs *attrs) { struct scatterlist *sg; int i; @@ -242,14 +213,17 @@ int dma_map_sg(struct device *dev, struct scatterlist *sglist, int nents, for_each_sg(sglist, sg, nents, i) { sg->dma_address = sg_phys(sg); __dma_prep_pa_range(sg->dma_address, sg->length, direction); +#ifdef CONFIG_NEED_SG_DMA_LENGTH + sg->dma_length = sg->length; +#endif } return nents; } -EXPORT_SYMBOL(dma_map_sg); -void dma_unmap_sg(struct device *dev, struct scatterlist *sglist, int nents, - enum dma_data_direction direction) +static void tile_dma_unmap_sg(struct device *dev, struct scatterlist *sglist, + int nents, enum dma_data_direction direction, + struct dma_attrs *attrs) { struct scatterlist *sg; int i; @@ -261,46 +235,50 @@ void dma_unmap_sg(struct device *dev, struct scatterlist *sglist, int nents, direction); } } -EXPORT_SYMBOL(dma_unmap_sg); -dma_addr_t dma_map_page(struct device *dev, struct page *page, - unsigned long offset, size_t size, - enum dma_data_direction direction) +static dma_addr_t tile_dma_map_page(struct device *dev, struct page *page, + unsigned long offset, size_t size, + enum dma_data_direction direction, + struct dma_attrs *attrs) { BUG_ON(!valid_dma_direction(direction)); BUG_ON(offset + size > PAGE_SIZE); __dma_prep_page(page, offset, size, direction); + return page_to_pa(page) + offset; } -EXPORT_SYMBOL(dma_map_page); -void dma_unmap_page(struct device *dev, dma_addr_t dma_address, size_t size, - enum dma_data_direction direction) +static void tile_dma_unmap_page(struct device *dev, dma_addr_t dma_address, + size_t size, enum dma_data_direction direction, + struct dma_attrs *attrs) { BUG_ON(!valid_dma_direction(direction)); + __dma_complete_page(pfn_to_page(PFN_DOWN(dma_address)), dma_address & PAGE_OFFSET, size, direction); } -EXPORT_SYMBOL(dma_unmap_page); -void dma_sync_single_for_cpu(struct device *dev, dma_addr_t dma_handle, - size_t size, enum dma_data_direction direction) +static void tile_dma_sync_single_for_cpu(struct device *dev, + dma_addr_t dma_handle, + size_t size, + enum dma_data_direction direction) { BUG_ON(!valid_dma_direction(direction)); + __dma_complete_pa_range(dma_handle, size, direction); } -EXPORT_SYMBOL(dma_sync_single_for_cpu); -void dma_sync_single_for_device(struct device *dev, dma_addr_t dma_handle, - size_t size, enum dma_data_direction direction) +static void tile_dma_sync_single_for_device(struct device *dev, + dma_addr_t dma_handle, size_t size, + enum dma_data_direction direction) { __dma_prep_pa_range(dma_handle, size, direction); } -EXPORT_SYMBOL(dma_sync_single_for_device); -void dma_sync_sg_for_cpu(struct device *dev, struct scatterlist *sglist, - int nelems, enum dma_data_direction direction) +static void tile_dma_sync_sg_for_cpu(struct device *dev, + struct scatterlist *sglist, int nelems, + enum dma_data_direction direction) { struct scatterlist *sg; int i; @@ -313,10 +291,10 @@ void dma_sync_sg_for_cpu(struct device *dev, struct scatterlist *sglist, sg_dma_len(sg), direction); } } -EXPORT_SYMBOL(dma_sync_sg_for_cpu); -void dma_sync_sg_for_device(struct device *dev, struct scatterlist *sglist, - int nelems, enum dma_data_direction direction) +static void tile_dma_sync_sg_for_device(struct device *dev, + struct scatterlist *sglist, int nelems, + enum dma_data_direction direction) { struct scatterlist *sg; int i; @@ -329,31 +307,280 @@ void dma_sync_sg_for_device(struct device *dev, struct scatterlist *sglist, sg_dma_len(sg), direction); } } -EXPORT_SYMBOL(dma_sync_sg_for_device); -void dma_sync_single_range_for_cpu(struct device *dev, dma_addr_t dma_handle, - unsigned long offset, size_t size, - enum dma_data_direction direction) +static inline int +tile_dma_mapping_error(struct device *dev, dma_addr_t dma_addr) { - dma_sync_single_for_cpu(dev, dma_handle + offset, size, direction); + return 0; } -EXPORT_SYMBOL(dma_sync_single_range_for_cpu); -void dma_sync_single_range_for_device(struct device *dev, - dma_addr_t dma_handle, - unsigned long offset, size_t size, - enum dma_data_direction direction) +static inline int +tile_dma_supported(struct device *dev, u64 mask) { - dma_sync_single_for_device(dev, dma_handle + offset, size, direction); + return 1; +} + +static struct dma_map_ops tile_default_dma_map_ops = { + .alloc = tile_dma_alloc_coherent, + .free = tile_dma_free_coherent, + .map_page = tile_dma_map_page, + .unmap_page = tile_dma_unmap_page, + .map_sg = tile_dma_map_sg, + .unmap_sg = tile_dma_unmap_sg, + .sync_single_for_cpu = tile_dma_sync_single_for_cpu, + .sync_single_for_device = tile_dma_sync_single_for_device, + .sync_sg_for_cpu = tile_dma_sync_sg_for_cpu, + .sync_sg_for_device = tile_dma_sync_sg_for_device, + .mapping_error = tile_dma_mapping_error, + .dma_supported = tile_dma_supported +}; + +struct dma_map_ops *tile_dma_map_ops = &tile_default_dma_map_ops; +EXPORT_SYMBOL(tile_dma_map_ops); + +/* Generic PCI DMA mapping functions */ + +static void *tile_pci_dma_alloc_coherent(struct device *dev, size_t size, + dma_addr_t *dma_handle, gfp_t gfp, + struct dma_attrs *attrs) +{ + int node = dev_to_node(dev); + int order = get_order(size); + struct page *pg; + dma_addr_t addr; + + gfp |= __GFP_ZERO; + + pg = homecache_alloc_pages_node(node, gfp, order, PAGE_HOME_DMA); + if (pg == NULL) + return NULL; + + addr = page_to_phys(pg); + + *dma_handle = phys_to_dma(dev, addr); + + return page_address(pg); } -EXPORT_SYMBOL(dma_sync_single_range_for_device); /* - * dma_alloc_noncoherent() is #defined to return coherent memory, - * so there's no need to do any flushing here. + * Free memory that was allocated with tile_pci_dma_alloc_coherent. */ -void dma_cache_sync(struct device *dev, void *vaddr, size_t size, - enum dma_data_direction direction) +static void tile_pci_dma_free_coherent(struct device *dev, size_t size, + void *vaddr, dma_addr_t dma_handle, + struct dma_attrs *attrs) { + homecache_free_pages((unsigned long)vaddr, get_order(size)); } -EXPORT_SYMBOL(dma_cache_sync); + +static int tile_pci_dma_map_sg(struct device *dev, struct scatterlist *sglist, + int nents, enum dma_data_direction direction, + struct dma_attrs *attrs) +{ + struct scatterlist *sg; + int i; + + BUG_ON(!valid_dma_direction(direction)); + + WARN_ON(nents == 0 || sglist->length == 0); + + for_each_sg(sglist, sg, nents, i) { + sg->dma_address = sg_phys(sg); + __dma_prep_pa_range(sg->dma_address, sg->length, direction); + + sg->dma_address = phys_to_dma(dev, sg->dma_address); +#ifdef CONFIG_NEED_SG_DMA_LENGTH + sg->dma_length = sg->length; +#endif + } + + return nents; +} + +static void tile_pci_dma_unmap_sg(struct device *dev, + struct scatterlist *sglist, int nents, + enum dma_data_direction direction, + struct dma_attrs *attrs) +{ + struct scatterlist *sg; + int i; + + BUG_ON(!valid_dma_direction(direction)); + for_each_sg(sglist, sg, nents, i) { + sg->dma_address = sg_phys(sg); + __dma_complete_pa_range(sg->dma_address, sg->length, + direction); + } +} + +static dma_addr_t tile_pci_dma_map_page(struct device *dev, struct page *page, + unsigned long offset, size_t size, + enum dma_data_direction direction, + struct dma_attrs *attrs) +{ + BUG_ON(!valid_dma_direction(direction)); + + BUG_ON(offset + size > PAGE_SIZE); + __dma_prep_page(page, offset, size, direction); + + return phys_to_dma(dev, page_to_pa(page) + offset); +} + +static void tile_pci_dma_unmap_page(struct device *dev, dma_addr_t dma_address, + size_t size, + enum dma_data_direction direction, + struct dma_attrs *attrs) +{ + BUG_ON(!valid_dma_direction(direction)); + + dma_address = dma_to_phys(dev, dma_address); + + __dma_complete_page(pfn_to_page(PFN_DOWN(dma_address)), + dma_address & PAGE_OFFSET, size, direction); +} + +static void tile_pci_dma_sync_single_for_cpu(struct device *dev, + dma_addr_t dma_handle, + size_t size, + enum dma_data_direction direction) +{ + BUG_ON(!valid_dma_direction(direction)); + + dma_handle = dma_to_phys(dev, dma_handle); + + __dma_complete_pa_range(dma_handle, size, direction); +} + +static void tile_pci_dma_sync_single_for_device(struct device *dev, + dma_addr_t dma_handle, + size_t size, + enum dma_data_direction + direction) +{ + dma_handle = dma_to_phys(dev, dma_handle); + + __dma_prep_pa_range(dma_handle, size, direction); +} + +static void tile_pci_dma_sync_sg_for_cpu(struct device *dev, + struct scatterlist *sglist, + int nelems, + enum dma_data_direction direction) +{ + struct scatterlist *sg; + int i; + + BUG_ON(!valid_dma_direction(direction)); + WARN_ON(nelems == 0 || sglist->length == 0); + + for_each_sg(sglist, sg, nelems, i) { + dma_sync_single_for_cpu(dev, sg->dma_address, + sg_dma_len(sg), direction); + } +} + +static void tile_pci_dma_sync_sg_for_device(struct device *dev, + struct scatterlist *sglist, + int nelems, + enum dma_data_direction direction) +{ + struct scatterlist *sg; + int i; + + BUG_ON(!valid_dma_direction(direction)); + WARN_ON(nelems == 0 || sglist->length == 0); + + for_each_sg(sglist, sg, nelems, i) { + dma_sync_single_for_device(dev, sg->dma_address, + sg_dma_len(sg), direction); + } +} + +static inline int +tile_pci_dma_mapping_error(struct device *dev, dma_addr_t dma_addr) +{ + return 0; +} + +static inline int +tile_pci_dma_supported(struct device *dev, u64 mask) +{ + return 1; +} + +static struct dma_map_ops tile_pci_default_dma_map_ops = { + .alloc = tile_pci_dma_alloc_coherent, + .free = tile_pci_dma_free_coherent, + .map_page = tile_pci_dma_map_page, + .unmap_page = tile_pci_dma_unmap_page, + .map_sg = tile_pci_dma_map_sg, + .unmap_sg = tile_pci_dma_unmap_sg, + .sync_single_for_cpu = tile_pci_dma_sync_single_for_cpu, + .sync_single_for_device = tile_pci_dma_sync_single_for_device, + .sync_sg_for_cpu = tile_pci_dma_sync_sg_for_cpu, + .sync_sg_for_device = tile_pci_dma_sync_sg_for_device, + .mapping_error = tile_pci_dma_mapping_error, + .dma_supported = tile_pci_dma_supported +}; + +struct dma_map_ops *gx_pci_dma_map_ops = &tile_pci_default_dma_map_ops; +EXPORT_SYMBOL(gx_pci_dma_map_ops); + +/* PCI DMA mapping functions for legacy PCI devices */ + +#ifdef CONFIG_SWIOTLB +static void *tile_swiotlb_alloc_coherent(struct device *dev, size_t size, + dma_addr_t *dma_handle, gfp_t gfp, + struct dma_attrs *attrs) +{ + gfp |= GFP_DMA; + return swiotlb_alloc_coherent(dev, size, dma_handle, gfp); +} + +static void tile_swiotlb_free_coherent(struct device *dev, size_t size, + void *vaddr, dma_addr_t dma_addr, + struct dma_attrs *attrs) +{ + swiotlb_free_coherent(dev, size, vaddr, dma_addr); +} + +static struct dma_map_ops pci_swiotlb_dma_ops = { + .alloc = tile_swiotlb_alloc_coherent, + .free = tile_swiotlb_free_coherent, + .map_page = swiotlb_map_page, + .unmap_page = swiotlb_unmap_page, + .map_sg = swiotlb_map_sg_attrs, + .unmap_sg = swiotlb_unmap_sg_attrs, + .sync_single_for_cpu = swiotlb_sync_single_for_cpu, + .sync_single_for_device = swiotlb_sync_single_for_device, + .sync_sg_for_cpu = swiotlb_sync_sg_for_cpu, + .sync_sg_for_device = swiotlb_sync_sg_for_device, + .dma_supported = swiotlb_dma_supported, + .mapping_error = swiotlb_dma_mapping_error, +}; + +struct dma_map_ops *gx_legacy_pci_dma_map_ops = &pci_swiotlb_dma_ops; +#else +struct dma_map_ops *gx_legacy_pci_dma_map_ops; +#endif +EXPORT_SYMBOL(gx_legacy_pci_dma_map_ops); + +#ifdef CONFIG_ARCH_HAS_DMA_SET_COHERENT_MASK +int dma_set_coherent_mask(struct device *dev, u64 mask) +{ + struct dma_map_ops *dma_ops = get_dma_ops(dev); + + /* Handle legacy PCI devices with limited memory addressability. */ + if (((dma_ops == gx_pci_dma_map_ops) || + (dma_ops == gx_legacy_pci_dma_map_ops)) && + (mask <= DMA_BIT_MASK(32))) { + if (mask > dev->archdata.max_direct_dma_addr) + mask = dev->archdata.max_direct_dma_addr; + } + + if (!dma_supported(dev, mask)) + return -EIO; + dev->coherent_dma_mask = mask; + return 0; +} +EXPORT_SYMBOL(dma_set_coherent_mask); +#endif diff --git a/arch/tile/kernel/pci_gx.c b/arch/tile/kernel/pci_gx.c index 1b996bb628f1..27f7ab021137 100644 --- a/arch/tile/kernel/pci_gx.c +++ b/arch/tile/kernel/pci_gx.c @@ -40,22 +40,8 @@ #include /* - * Initialization flow and process - * ------------------------------- - * - * This files containes the routines to search for PCI buses, + * This file containes the routines to search for PCI buses, * enumerate the buses, and configure any attached devices. - * - * There are two entry points here: - * 1) tile_pci_init - * This sets up the pci_controller structs, and opens the - * FDs to the hypervisor. This is called from setup_arch() early - * in the boot process. - * 2) pcibios_init - * This probes the PCI bus(es) for any attached hardware. It's - * called by subsys_initcall. All of the real work is done by the - * generic Linux PCI layer. - * */ #define DEBUG_PCI_CFG 0 @@ -110,6 +96,21 @@ static struct pci_ops tile_cfg_ops; /* Mask of CPUs that should receive PCIe interrupts. */ static struct cpumask intr_cpus_map; +/* PCI I/O space support is not implemented. */ +static struct resource pci_ioport_resource = { + .name = "PCI IO", + .start = 0, + .end = 0, + .flags = IORESOURCE_IO, +}; + +static struct resource pci_iomem_resource = { + .name = "PCI mem", + .start = TILE_PCI_MEM_START, + .end = TILE_PCI_MEM_END, + .flags = IORESOURCE_MEM, +}; + /* * We don't need to worry about the alignment of resources. */ @@ -334,8 +335,6 @@ free_irqs: } /* - * First initialization entry point, called from setup_arch(). - * * Find valid controllers and fill in pci_controller structs for each * of them. * @@ -583,10 +582,7 @@ static int __devinit setup_pcie_rc_delay(char *str) early_param("pcie_rc_delay", setup_pcie_rc_delay); /* - * Second PCI initialization entry point, called by subsys_initcall. - * - * The controllers have been set up by the time we get here, by a call to - * tile_pci_init. + * PCI initialization entry point, called by subsys_initcall. */ int __init pcibios_init(void) { @@ -594,15 +590,13 @@ int __init pcibios_init(void) LIST_HEAD(resources); int i; + tile_pci_init(); + if (num_rc_controllers == 0 && num_ep_controllers == 0) return 0; - pr_info("PCI: Probing PCI hardware\n"); - /* * We loop over all the TRIO shims and set up the MMIO mappings. - * This step can't be done in tile_pci_init because the MM subsystem - * hasn't been initialized then. */ for (i = 0; i < TILEGX_NUM_TRIO; i++) { gxio_trio_context_t *context = &trio_contexts[i]; @@ -645,9 +639,7 @@ int __init pcibios_init(void) unsigned int class_code_revision; int trio_index; int mac; -#ifndef USE_SHARED_PCIE_CONFIG_REGION int ret; -#endif if (trio_context->fd < 0) continue; @@ -802,8 +794,6 @@ int __init pcibios_init(void) pr_err("PCI: PCI CFG PIO alloc failure for mac %d " "on TRIO %d, give up\n", mac, trio_index); - /* TBD: cleanup ... */ - continue; } @@ -819,8 +809,6 @@ int __init pcibios_init(void) pr_err("PCI: PCI CFG PIO init failure for mac %d " "on TRIO %d, give up\n", mac, trio_index); - /* TBD: cleanup ... */ - continue; } @@ -837,8 +825,6 @@ int __init pcibios_init(void) pr_err("PCI: PIO map failure for mac %d on TRIO %d\n", mac, trio_index); - /* TBD: cleanup ... */ - continue; } @@ -852,7 +838,14 @@ int __init pcibios_init(void) continue; } - pci_add_resource(&resources, &iomem_resource); + /* + * The PCI memory resource is located above the PA space. + * The memory range for the PCI root bus should not overlap + * with the physical RAM + */ + pci_add_resource_offset(&resources, &iomem_resource, + 1ULL << CHIP_PA_WIDTH()); + bus = pci_scan_root_bus(NULL, 0, controller->ops, controller, &resources); controller->root_bus = bus; @@ -922,11 +915,6 @@ int __init pcibios_init(void) continue; } - /* - * We always assign 32-bit PCI bus BAR ranges. - */ - BUG_ON(bus_address_hi != 0); - /* * Alloc a PIO region for PCI memory access for each RC port. */ @@ -936,8 +924,6 @@ int __init pcibios_init(void) "give up\n", controller->trio_index, controller->mac); - /* TBD: cleanup ... */ - continue; } @@ -950,15 +936,13 @@ int __init pcibios_init(void) ret = gxio_trio_init_pio_region_aux(trio_context, controller->pio_mem_index, controller->mac, - bus_address_hi, + 0, 0); if (ret < 0) { pr_err("PCI: MEM PIO init failure on TRIO %d mac %d, " "give up\n", controller->trio_index, controller->mac); - /* TBD: cleanup ... */ - continue; } @@ -980,8 +964,6 @@ int __init pcibios_init(void) controller->trio_index, controller->mac, j); - /* TBD: cleanup ... */ - goto alloc_mem_map_failed; } @@ -991,9 +973,13 @@ int __init pcibios_init(void) * Initialize the Mem-Map and the I/O MMU so that all * the physical memory can be accessed by the endpoint * devices. The base bus address is set to the base CPA - * of this memory controller, so is the base VA. The + * of this memory controller plus an offset (see pci.h). + * The region's base VA is set to the base CPA. The * I/O MMU table essentially translates the CPA to - * the real PA. + * the real PA. Implicitly, for node 0, we create + * a separate Mem-Map region that serves as the inbound + * window for legacy 32-bit devices. This is a direct + * map of the low 4GB CPA space. */ ret = gxio_trio_init_memory_map_mmu_aux(trio_context, controller->mem_maps[j], @@ -1001,7 +987,8 @@ int __init pcibios_init(void) nr_pages << PAGE_SHIFT, trio_context->asid, controller->mac, - start_pfn << PAGE_SHIFT, + (start_pfn << PAGE_SHIFT) + + TILE_PCI_MEM_MAP_BASE_OFFSET, j, GXIO_TRIO_ORDER_MODE_UNORDERED); if (ret < 0) { @@ -1010,11 +997,8 @@ int __init pcibios_init(void) controller->trio_index, controller->mac, j); - /* TBD: cleanup ... */ - goto alloc_mem_map_failed; } - continue; alloc_mem_map_failed: @@ -1028,11 +1012,19 @@ alloc_mem_map_failed: subsys_initcall(pcibios_init); /* - * No bus fixups needed. + * PCI scan code calls the arch specific pcibios_fixup_bus() each time it scans + * a new bridge. Called after each bus is probed, but before its children are + * examined. */ void __devinit pcibios_fixup_bus(struct pci_bus *bus) { - /* Nothing needs to be done. */ + struct pci_dev *dev = bus->self; + + if (!dev) { + /* This is the root bus. */ + bus->resource[0] = &pci_ioport_resource; + bus->resource[1] = &pci_iomem_resource; + } } /* @@ -1069,6 +1061,17 @@ int pcibios_enable_device(struct pci_dev *dev, int mask) return pci_enable_resources(dev, mask); } +/* Called for each device after PCI setup is done. */ +static void __init +pcibios_fixup_final(struct pci_dev *pdev) +{ + set_dma_ops(&pdev->dev, gx_pci_dma_map_ops); + set_dma_offset(&pdev->dev, TILE_PCI_MEM_MAP_BASE_OFFSET); + pdev->dev.archdata.max_direct_dma_addr = + TILE_PCI_MAX_DIRECT_DMA_ADDRESS; +} +DECLARE_PCI_FIXUP_FINAL(PCI_ANY_ID, PCI_ANY_ID, pcibios_fixup_final); + /* Map a PCI MMIO bus address into VA space. */ void __iomem *ioremap(resource_size_t phys_addr, unsigned long size) { @@ -1127,7 +1130,7 @@ got_it: * We need to keep the PCI bus address's in-page offset in the VA. */ return iorpc_ioremap(trio_fd, offset, size) + - (phys_addr & (PAGE_SIZE - 1)); + (phys_addr & (PAGE_SIZE - 1)); } EXPORT_SYMBOL(ioremap); diff --git a/arch/tile/kernel/setup.c b/arch/tile/kernel/setup.c index fdde3b6986e5..2b8b689e596d 100644 --- a/arch/tile/kernel/setup.c +++ b/arch/tile/kernel/setup.c @@ -23,6 +23,7 @@ #include #include #include +#include #include #include #include @@ -109,7 +110,7 @@ static unsigned int __initdata maxnodemem_pfn[MAX_NUMNODES] = { }; static nodemask_t __initdata isolnodes; -#ifdef CONFIG_PCI +#if defined(CONFIG_PCI) && !defined(__tilegx__) enum { DEFAULT_PCI_RESERVE_MB = 64 }; static unsigned int __initdata pci_reserve_mb = DEFAULT_PCI_RESERVE_MB; unsigned long __initdata pci_reserve_start_pfn = -1U; @@ -160,7 +161,7 @@ static int __init setup_isolnodes(char *str) } early_param("isolnodes", setup_isolnodes); -#ifdef CONFIG_PCI +#if defined(CONFIG_PCI) && !defined(__tilegx__) static int __init setup_pci_reserve(char* str) { unsigned long mb; @@ -171,7 +172,7 @@ static int __init setup_pci_reserve(char* str) pci_reserve_mb = mb; pr_info("Reserving %dMB for PCIE root complex mappings\n", - pci_reserve_mb); + pci_reserve_mb); return 0; } early_param("pci_reserve", setup_pci_reserve); @@ -411,7 +412,7 @@ static void __init setup_memory(void) continue; } #endif -#ifdef CONFIG_PCI +#if defined(CONFIG_PCI) && !defined(__tilegx__) /* * Blocks that overlap the pci reserved region must * have enough space to hold the maximum percpu data @@ -604,11 +605,9 @@ static void __init setup_bootmem_allocator_node(int i) /* Free all the space back into the allocator. */ free_bootmem(PFN_PHYS(start), PFN_PHYS(end - start)); -#if defined(CONFIG_PCI) +#if defined(CONFIG_PCI) && !defined(__tilegx__) /* - * Throw away any memory aliased by the PCI region. FIXME: this - * is a temporary hack to work around bug 10502, and needs to be - * fixed properly. + * Throw away any memory aliased by the PCI region. */ if (pci_reserve_start_pfn < end && pci_reserve_end_pfn > start) reserve_bootmem(PFN_PHYS(pci_reserve_start_pfn), @@ -1353,8 +1352,7 @@ void __init setup_arch(char **cmdline_p) setup_cpu_maps(); -#ifdef CONFIG_PCI -#if !defined (__tilegx__) +#if defined(CONFIG_PCI) && !defined(__tilegx__) /* * Initialize the PCI structures. This is done before memory * setup so that we know whether or not a pci_reserve region @@ -1362,7 +1360,6 @@ void __init setup_arch(char **cmdline_p) */ if (tile_pci_init() == 0) pci_reserve_mb = 0; -#endif /* PCI systems reserve a region just below 4GB for mapping iomem. */ pci_reserve_end_pfn = (1 << (32 - PAGE_SHIFT)); @@ -1384,6 +1381,10 @@ void __init setup_arch(char **cmdline_p) * any memory using the bootmem allocator. */ +#ifdef CONFIG_SWIOTLB + swiotlb_init(0); +#endif + paging_init(); setup_numa_mapping(); zone_sizes_init(); @@ -1391,10 +1392,6 @@ void __init setup_arch(char **cmdline_p) setup_cpu(1); setup_clock(); load_hv_initrd(); - -#if defined(CONFIG_PCI) && defined (__tilegx__) - tile_pci_init(); -#endif } @@ -1538,11 +1535,11 @@ static struct resource code_resource = { }; /* - * We reserve all resources above 4GB so that PCI won't try to put + * On Pro, we reserve all resources above 4GB so that PCI won't try to put * mappings above 4GB; the standard allows that for some devices but * the probing code trunates values to 32 bits. */ -#ifdef CONFIG_PCI +#if defined(CONFIG_PCI) && !defined(__tilegx__) static struct resource* __init insert_non_bus_resource(void) { @@ -1588,7 +1585,7 @@ static int __init request_standard_resources(void) enum { CODE_DELTA = MEM_SV_INTRPT - PAGE_OFFSET }; iomem_resource.end = -1LL; -#ifdef CONFIG_PCI +#if defined(CONFIG_PCI) && !defined(__tilegx__) insert_non_bus_resource(); #endif @@ -1596,7 +1593,7 @@ static int __init request_standard_resources(void) u64 start_pfn = node_start_pfn[i]; u64 end_pfn = node_end_pfn[i]; -#ifdef CONFIG_PCI +#if defined(CONFIG_PCI) && !defined(__tilegx__) if (start_pfn <= pci_reserve_start_pfn && end_pfn > pci_reserve_start_pfn) { if (end_pfn > pci_reserve_end_pfn)