linux/mm/dmapool.c

525 lines
13 KiB
C
Raw Normal View History

// SPDX-License-Identifier: GPL-2.0-only
/*
* DMA Pool allocator
*
* Copyright 2001 David Brownell
* Copyright 2007 Intel Corporation
* Author: Matthew Wilcox <willy@linux.intel.com>
*
* This allocator returns small blocks of a given size which are DMA-able by
* the given device. It uses the dma_alloc_coherent page allocator to get
* new pages, then splits them up into blocks of the required size.
* Many older drivers still have their own code to do this.
*
* The current design of this allocator is fairly simple. The pool is
* represented by the 'struct dma_pool' which keeps a doubly-linked list of
* allocated pages. Each page in the page_list is split into blocks of at
* least 'size' bytes. Free blocks are tracked in an unsorted singly-linked
dmapool: link blocks across pages The allocated dmapool pages are never freed for the lifetime of the pool. There is no need for the two level list+stack lookup for finding a free block since nothing is ever removed from the list. Just use a simple stack, reducing time complexity to constant. The implementation inserts the stack linking elements and the dma handle of the block within itself when freed. This means the smallest possible dmapool block is increased to at most 16 bytes to accommodate these fields, but there are no exisiting users requesting a dma pool smaller than that anyway. Removing the list has a significant change in performance. Using the kernel's micro-benchmarking self test: Before: # modprobe dmapool_test dmapool test: size:16 blocks:8192 time:57282 dmapool test: size:64 blocks:8192 time:172562 dmapool test: size:256 blocks:8192 time:789247 dmapool test: size:1024 blocks:2048 time:371823 dmapool test: size:4096 blocks:1024 time:362237 After: # modprobe dmapool_test dmapool test: size:16 blocks:8192 time:24997 dmapool test: size:64 blocks:8192 time:26584 dmapool test: size:256 blocks:8192 time:33542 dmapool test: size:1024 blocks:2048 time:9022 dmapool test: size:4096 blocks:1024 time:6045 The module test allocates quite a few blocks that may not accurately represent how these pools are used in real life. For a more marco level benchmark, running fio high-depth + high-batched on nvme, this patch shows submission and completion latency reduced by ~100usec each, 1% IOPs improvement, and perf record's time spent in dma_pool_alloc/free were reduced by half. [kbusch@kernel.org: push new blocks in ascending order] Link: https://lkml.kernel.org/r/20230221165400.1595247-1-kbusch@meta.com Link: https://lkml.kernel.org/r/20230126215125.4069751-12-kbusch@meta.com Fixes: 2d55c16c0c54 ("dmapool: create/destroy cleanup") Signed-off-by: Keith Busch <kbusch@kernel.org> Reviewed-by: Christoph Hellwig <hch@lst.de> Tested-by: Bryan O'Donoghue <bryan.odonoghue@linaro.org> Cc: Matthew Wilcox <willy@infradead.org> Cc: Tony Battersby <tonyb@cybernetics.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2023-01-27 05:51:24 +08:00
* list of free blocks across all pages. Used blocks aren't tracked, but we
* keep a count of how many are currently allocated from each page.
*/
#include <linux/device.h>
#include <linux/dma-mapping.h>
#include <linux/dmapool.h>
#include <linux/kernel.h>
#include <linux/list.h>
#include <linux/export.h>
#include <linux/mutex.h>
#include <linux/poison.h>
#include <linux/sched.h>
#include <linux/sched/mm.h>
#include <linux/slab.h>
#include <linux/stat.h>
#include <linux/spinlock.h>
#include <linux/string.h>
#include <linux/types.h>
#include <linux/wait.h>
#ifdef CONFIG_SLUB_DEBUG_ON
#define DMAPOOL_DEBUG 1
#endif
dmapool: link blocks across pages The allocated dmapool pages are never freed for the lifetime of the pool. There is no need for the two level list+stack lookup for finding a free block since nothing is ever removed from the list. Just use a simple stack, reducing time complexity to constant. The implementation inserts the stack linking elements and the dma handle of the block within itself when freed. This means the smallest possible dmapool block is increased to at most 16 bytes to accommodate these fields, but there are no exisiting users requesting a dma pool smaller than that anyway. Removing the list has a significant change in performance. Using the kernel's micro-benchmarking self test: Before: # modprobe dmapool_test dmapool test: size:16 blocks:8192 time:57282 dmapool test: size:64 blocks:8192 time:172562 dmapool test: size:256 blocks:8192 time:789247 dmapool test: size:1024 blocks:2048 time:371823 dmapool test: size:4096 blocks:1024 time:362237 After: # modprobe dmapool_test dmapool test: size:16 blocks:8192 time:24997 dmapool test: size:64 blocks:8192 time:26584 dmapool test: size:256 blocks:8192 time:33542 dmapool test: size:1024 blocks:2048 time:9022 dmapool test: size:4096 blocks:1024 time:6045 The module test allocates quite a few blocks that may not accurately represent how these pools are used in real life. For a more marco level benchmark, running fio high-depth + high-batched on nvme, this patch shows submission and completion latency reduced by ~100usec each, 1% IOPs improvement, and perf record's time spent in dma_pool_alloc/free were reduced by half. [kbusch@kernel.org: push new blocks in ascending order] Link: https://lkml.kernel.org/r/20230221165400.1595247-1-kbusch@meta.com Link: https://lkml.kernel.org/r/20230126215125.4069751-12-kbusch@meta.com Fixes: 2d55c16c0c54 ("dmapool: create/destroy cleanup") Signed-off-by: Keith Busch <kbusch@kernel.org> Reviewed-by: Christoph Hellwig <hch@lst.de> Tested-by: Bryan O'Donoghue <bryan.odonoghue@linaro.org> Cc: Matthew Wilcox <willy@infradead.org> Cc: Tony Battersby <tonyb@cybernetics.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2023-01-27 05:51:24 +08:00
struct dma_block {
struct dma_block *next_block;
dma_addr_t dma;
};
struct dma_pool { /* the pool */
struct list_head page_list;
spinlock_t lock;
dmapool: link blocks across pages The allocated dmapool pages are never freed for the lifetime of the pool. There is no need for the two level list+stack lookup for finding a free block since nothing is ever removed from the list. Just use a simple stack, reducing time complexity to constant. The implementation inserts the stack linking elements and the dma handle of the block within itself when freed. This means the smallest possible dmapool block is increased to at most 16 bytes to accommodate these fields, but there are no exisiting users requesting a dma pool smaller than that anyway. Removing the list has a significant change in performance. Using the kernel's micro-benchmarking self test: Before: # modprobe dmapool_test dmapool test: size:16 blocks:8192 time:57282 dmapool test: size:64 blocks:8192 time:172562 dmapool test: size:256 blocks:8192 time:789247 dmapool test: size:1024 blocks:2048 time:371823 dmapool test: size:4096 blocks:1024 time:362237 After: # modprobe dmapool_test dmapool test: size:16 blocks:8192 time:24997 dmapool test: size:64 blocks:8192 time:26584 dmapool test: size:256 blocks:8192 time:33542 dmapool test: size:1024 blocks:2048 time:9022 dmapool test: size:4096 blocks:1024 time:6045 The module test allocates quite a few blocks that may not accurately represent how these pools are used in real life. For a more marco level benchmark, running fio high-depth + high-batched on nvme, this patch shows submission and completion latency reduced by ~100usec each, 1% IOPs improvement, and perf record's time spent in dma_pool_alloc/free were reduced by half. [kbusch@kernel.org: push new blocks in ascending order] Link: https://lkml.kernel.org/r/20230221165400.1595247-1-kbusch@meta.com Link: https://lkml.kernel.org/r/20230126215125.4069751-12-kbusch@meta.com Fixes: 2d55c16c0c54 ("dmapool: create/destroy cleanup") Signed-off-by: Keith Busch <kbusch@kernel.org> Reviewed-by: Christoph Hellwig <hch@lst.de> Tested-by: Bryan O'Donoghue <bryan.odonoghue@linaro.org> Cc: Matthew Wilcox <willy@infradead.org> Cc: Tony Battersby <tonyb@cybernetics.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2023-01-27 05:51:24 +08:00
struct dma_block *next_block;
size_t nr_blocks;
size_t nr_active;
size_t nr_pages;
struct device *dev;
unsigned int size;
unsigned int allocation;
unsigned int boundary;
char name[32];
struct list_head pools;
};
struct dma_page { /* cacheable header for 'allocation' bytes */
struct list_head page_list;
void *vaddr;
dma_addr_t dma;
};
static DEFINE_MUTEX(pools_lock);
mm: dmapool: add/remove sysfs file outside of the pool lock lock cat /sys/.../pools followed by removal the device leads to: |====================================================== |[ INFO: possible circular locking dependency detected ] |3.17.0-rc4+ #1498 Not tainted |------------------------------------------------------- |rmmod/2505 is trying to acquire lock: | (s_active#28){++++.+}, at: [<c017f754>] kernfs_remove_by_name_ns+0x3c/0x88 | |but task is already holding lock: | (pools_lock){+.+.+.}, at: [<c011494c>] dma_pool_destroy+0x18/0x17c | |which lock already depends on the new lock. |the existing dependency chain (in reverse order) is: | |-> #1 (pools_lock){+.+.+.}: | [<c0114ae8>] show_pools+0x30/0xf8 | [<c0313210>] dev_attr_show+0x1c/0x48 | [<c0180e84>] sysfs_kf_seq_show+0x88/0x10c | [<c017f960>] kernfs_seq_show+0x24/0x28 | [<c013efc4>] seq_read+0x1b8/0x480 | [<c011e820>] vfs_read+0x8c/0x148 | [<c011ea10>] SyS_read+0x40/0x8c | [<c000e960>] ret_fast_syscall+0x0/0x48 | |-> #0 (s_active#28){++++.+}: | [<c017e9ac>] __kernfs_remove+0x258/0x2ec | [<c017f754>] kernfs_remove_by_name_ns+0x3c/0x88 | [<c0114a7c>] dma_pool_destroy+0x148/0x17c | [<c03ad288>] hcd_buffer_destroy+0x20/0x34 | [<c03a4780>] usb_remove_hcd+0x110/0x1a4 The problem is the lock order of pools_lock and kernfs_mutex in dma_pool_destroy() vs show_pools() call path. This patch breaks out the creation of the sysfs file outside of the pools_lock mutex. The newly added pools_reg_lock ensures that there is no race of create vs destroy code path in terms whether or not the sysfs file has to be deleted (and was it deleted before we try to create a new one) and what to do if device_create_file() failed. Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2014-10-10 06:28:50 +08:00
static DEFINE_MUTEX(pools_reg_lock);
static ssize_t pools_show(struct device *dev, struct device_attribute *attr, char *buf)
{
struct dma_pool *pool;
dmapool: link blocks across pages The allocated dmapool pages are never freed for the lifetime of the pool. There is no need for the two level list+stack lookup for finding a free block since nothing is ever removed from the list. Just use a simple stack, reducing time complexity to constant. The implementation inserts the stack linking elements and the dma handle of the block within itself when freed. This means the smallest possible dmapool block is increased to at most 16 bytes to accommodate these fields, but there are no exisiting users requesting a dma pool smaller than that anyway. Removing the list has a significant change in performance. Using the kernel's micro-benchmarking self test: Before: # modprobe dmapool_test dmapool test: size:16 blocks:8192 time:57282 dmapool test: size:64 blocks:8192 time:172562 dmapool test: size:256 blocks:8192 time:789247 dmapool test: size:1024 blocks:2048 time:371823 dmapool test: size:4096 blocks:1024 time:362237 After: # modprobe dmapool_test dmapool test: size:16 blocks:8192 time:24997 dmapool test: size:64 blocks:8192 time:26584 dmapool test: size:256 blocks:8192 time:33542 dmapool test: size:1024 blocks:2048 time:9022 dmapool test: size:4096 blocks:1024 time:6045 The module test allocates quite a few blocks that may not accurately represent how these pools are used in real life. For a more marco level benchmark, running fio high-depth + high-batched on nvme, this patch shows submission and completion latency reduced by ~100usec each, 1% IOPs improvement, and perf record's time spent in dma_pool_alloc/free were reduced by half. [kbusch@kernel.org: push new blocks in ascending order] Link: https://lkml.kernel.org/r/20230221165400.1595247-1-kbusch@meta.com Link: https://lkml.kernel.org/r/20230126215125.4069751-12-kbusch@meta.com Fixes: 2d55c16c0c54 ("dmapool: create/destroy cleanup") Signed-off-by: Keith Busch <kbusch@kernel.org> Reviewed-by: Christoph Hellwig <hch@lst.de> Tested-by: Bryan O'Donoghue <bryan.odonoghue@linaro.org> Cc: Matthew Wilcox <willy@infradead.org> Cc: Tony Battersby <tonyb@cybernetics.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2023-01-27 05:51:24 +08:00
unsigned size;
size = sysfs_emit(buf, "poolinfo - 0.1\n");
mutex_lock(&pools_lock);
list_for_each_entry(pool, &dev->dma_pools, pools) {
/* per-pool info, no real statistics yet */
dmapool: link blocks across pages The allocated dmapool pages are never freed for the lifetime of the pool. There is no need for the two level list+stack lookup for finding a free block since nothing is ever removed from the list. Just use a simple stack, reducing time complexity to constant. The implementation inserts the stack linking elements and the dma handle of the block within itself when freed. This means the smallest possible dmapool block is increased to at most 16 bytes to accommodate these fields, but there are no exisiting users requesting a dma pool smaller than that anyway. Removing the list has a significant change in performance. Using the kernel's micro-benchmarking self test: Before: # modprobe dmapool_test dmapool test: size:16 blocks:8192 time:57282 dmapool test: size:64 blocks:8192 time:172562 dmapool test: size:256 blocks:8192 time:789247 dmapool test: size:1024 blocks:2048 time:371823 dmapool test: size:4096 blocks:1024 time:362237 After: # modprobe dmapool_test dmapool test: size:16 blocks:8192 time:24997 dmapool test: size:64 blocks:8192 time:26584 dmapool test: size:256 blocks:8192 time:33542 dmapool test: size:1024 blocks:2048 time:9022 dmapool test: size:4096 blocks:1024 time:6045 The module test allocates quite a few blocks that may not accurately represent how these pools are used in real life. For a more marco level benchmark, running fio high-depth + high-batched on nvme, this patch shows submission and completion latency reduced by ~100usec each, 1% IOPs improvement, and perf record's time spent in dma_pool_alloc/free were reduced by half. [kbusch@kernel.org: push new blocks in ascending order] Link: https://lkml.kernel.org/r/20230221165400.1595247-1-kbusch@meta.com Link: https://lkml.kernel.org/r/20230126215125.4069751-12-kbusch@meta.com Fixes: 2d55c16c0c54 ("dmapool: create/destroy cleanup") Signed-off-by: Keith Busch <kbusch@kernel.org> Reviewed-by: Christoph Hellwig <hch@lst.de> Tested-by: Bryan O'Donoghue <bryan.odonoghue@linaro.org> Cc: Matthew Wilcox <willy@infradead.org> Cc: Tony Battersby <tonyb@cybernetics.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2023-01-27 05:51:24 +08:00
size += sysfs_emit_at(buf, size, "%-16s %4zu %4zu %4u %2zu\n",
pool->name, pool->nr_active,
pool->nr_blocks, pool->size,
pool->nr_pages);
}
mutex_unlock(&pools_lock);
return size;
}
static DEVICE_ATTR_RO(pools);
#ifdef DMAPOOL_DEBUG
dmapool: link blocks across pages The allocated dmapool pages are never freed for the lifetime of the pool. There is no need for the two level list+stack lookup for finding a free block since nothing is ever removed from the list. Just use a simple stack, reducing time complexity to constant. The implementation inserts the stack linking elements and the dma handle of the block within itself when freed. This means the smallest possible dmapool block is increased to at most 16 bytes to accommodate these fields, but there are no exisiting users requesting a dma pool smaller than that anyway. Removing the list has a significant change in performance. Using the kernel's micro-benchmarking self test: Before: # modprobe dmapool_test dmapool test: size:16 blocks:8192 time:57282 dmapool test: size:64 blocks:8192 time:172562 dmapool test: size:256 blocks:8192 time:789247 dmapool test: size:1024 blocks:2048 time:371823 dmapool test: size:4096 blocks:1024 time:362237 After: # modprobe dmapool_test dmapool test: size:16 blocks:8192 time:24997 dmapool test: size:64 blocks:8192 time:26584 dmapool test: size:256 blocks:8192 time:33542 dmapool test: size:1024 blocks:2048 time:9022 dmapool test: size:4096 blocks:1024 time:6045 The module test allocates quite a few blocks that may not accurately represent how these pools are used in real life. For a more marco level benchmark, running fio high-depth + high-batched on nvme, this patch shows submission and completion latency reduced by ~100usec each, 1% IOPs improvement, and perf record's time spent in dma_pool_alloc/free were reduced by half. [kbusch@kernel.org: push new blocks in ascending order] Link: https://lkml.kernel.org/r/20230221165400.1595247-1-kbusch@meta.com Link: https://lkml.kernel.org/r/20230126215125.4069751-12-kbusch@meta.com Fixes: 2d55c16c0c54 ("dmapool: create/destroy cleanup") Signed-off-by: Keith Busch <kbusch@kernel.org> Reviewed-by: Christoph Hellwig <hch@lst.de> Tested-by: Bryan O'Donoghue <bryan.odonoghue@linaro.org> Cc: Matthew Wilcox <willy@infradead.org> Cc: Tony Battersby <tonyb@cybernetics.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2023-01-27 05:51:24 +08:00
static void pool_check_block(struct dma_pool *pool, struct dma_block *block,
gfp_t mem_flags)
{
dmapool: link blocks across pages The allocated dmapool pages are never freed for the lifetime of the pool. There is no need for the two level list+stack lookup for finding a free block since nothing is ever removed from the list. Just use a simple stack, reducing time complexity to constant. The implementation inserts the stack linking elements and the dma handle of the block within itself when freed. This means the smallest possible dmapool block is increased to at most 16 bytes to accommodate these fields, but there are no exisiting users requesting a dma pool smaller than that anyway. Removing the list has a significant change in performance. Using the kernel's micro-benchmarking self test: Before: # modprobe dmapool_test dmapool test: size:16 blocks:8192 time:57282 dmapool test: size:64 blocks:8192 time:172562 dmapool test: size:256 blocks:8192 time:789247 dmapool test: size:1024 blocks:2048 time:371823 dmapool test: size:4096 blocks:1024 time:362237 After: # modprobe dmapool_test dmapool test: size:16 blocks:8192 time:24997 dmapool test: size:64 blocks:8192 time:26584 dmapool test: size:256 blocks:8192 time:33542 dmapool test: size:1024 blocks:2048 time:9022 dmapool test: size:4096 blocks:1024 time:6045 The module test allocates quite a few blocks that may not accurately represent how these pools are used in real life. For a more marco level benchmark, running fio high-depth + high-batched on nvme, this patch shows submission and completion latency reduced by ~100usec each, 1% IOPs improvement, and perf record's time spent in dma_pool_alloc/free were reduced by half. [kbusch@kernel.org: push new blocks in ascending order] Link: https://lkml.kernel.org/r/20230221165400.1595247-1-kbusch@meta.com Link: https://lkml.kernel.org/r/20230126215125.4069751-12-kbusch@meta.com Fixes: 2d55c16c0c54 ("dmapool: create/destroy cleanup") Signed-off-by: Keith Busch <kbusch@kernel.org> Reviewed-by: Christoph Hellwig <hch@lst.de> Tested-by: Bryan O'Donoghue <bryan.odonoghue@linaro.org> Cc: Matthew Wilcox <willy@infradead.org> Cc: Tony Battersby <tonyb@cybernetics.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2023-01-27 05:51:24 +08:00
u8 *data = (void *)block;
int i;
dmapool: link blocks across pages The allocated dmapool pages are never freed for the lifetime of the pool. There is no need for the two level list+stack lookup for finding a free block since nothing is ever removed from the list. Just use a simple stack, reducing time complexity to constant. The implementation inserts the stack linking elements and the dma handle of the block within itself when freed. This means the smallest possible dmapool block is increased to at most 16 bytes to accommodate these fields, but there are no exisiting users requesting a dma pool smaller than that anyway. Removing the list has a significant change in performance. Using the kernel's micro-benchmarking self test: Before: # modprobe dmapool_test dmapool test: size:16 blocks:8192 time:57282 dmapool test: size:64 blocks:8192 time:172562 dmapool test: size:256 blocks:8192 time:789247 dmapool test: size:1024 blocks:2048 time:371823 dmapool test: size:4096 blocks:1024 time:362237 After: # modprobe dmapool_test dmapool test: size:16 blocks:8192 time:24997 dmapool test: size:64 blocks:8192 time:26584 dmapool test: size:256 blocks:8192 time:33542 dmapool test: size:1024 blocks:2048 time:9022 dmapool test: size:4096 blocks:1024 time:6045 The module test allocates quite a few blocks that may not accurately represent how these pools are used in real life. For a more marco level benchmark, running fio high-depth + high-batched on nvme, this patch shows submission and completion latency reduced by ~100usec each, 1% IOPs improvement, and perf record's time spent in dma_pool_alloc/free were reduced by half. [kbusch@kernel.org: push new blocks in ascending order] Link: https://lkml.kernel.org/r/20230221165400.1595247-1-kbusch@meta.com Link: https://lkml.kernel.org/r/20230126215125.4069751-12-kbusch@meta.com Fixes: 2d55c16c0c54 ("dmapool: create/destroy cleanup") Signed-off-by: Keith Busch <kbusch@kernel.org> Reviewed-by: Christoph Hellwig <hch@lst.de> Tested-by: Bryan O'Donoghue <bryan.odonoghue@linaro.org> Cc: Matthew Wilcox <willy@infradead.org> Cc: Tony Battersby <tonyb@cybernetics.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2023-01-27 05:51:24 +08:00
for (i = sizeof(struct dma_block); i < pool->size; i++) {
if (data[i] == POOL_POISON_FREED)
continue;
dmapool: link blocks across pages The allocated dmapool pages are never freed for the lifetime of the pool. There is no need for the two level list+stack lookup for finding a free block since nothing is ever removed from the list. Just use a simple stack, reducing time complexity to constant. The implementation inserts the stack linking elements and the dma handle of the block within itself when freed. This means the smallest possible dmapool block is increased to at most 16 bytes to accommodate these fields, but there are no exisiting users requesting a dma pool smaller than that anyway. Removing the list has a significant change in performance. Using the kernel's micro-benchmarking self test: Before: # modprobe dmapool_test dmapool test: size:16 blocks:8192 time:57282 dmapool test: size:64 blocks:8192 time:172562 dmapool test: size:256 blocks:8192 time:789247 dmapool test: size:1024 blocks:2048 time:371823 dmapool test: size:4096 blocks:1024 time:362237 After: # modprobe dmapool_test dmapool test: size:16 blocks:8192 time:24997 dmapool test: size:64 blocks:8192 time:26584 dmapool test: size:256 blocks:8192 time:33542 dmapool test: size:1024 blocks:2048 time:9022 dmapool test: size:4096 blocks:1024 time:6045 The module test allocates quite a few blocks that may not accurately represent how these pools are used in real life. For a more marco level benchmark, running fio high-depth + high-batched on nvme, this patch shows submission and completion latency reduced by ~100usec each, 1% IOPs improvement, and perf record's time spent in dma_pool_alloc/free were reduced by half. [kbusch@kernel.org: push new blocks in ascending order] Link: https://lkml.kernel.org/r/20230221165400.1595247-1-kbusch@meta.com Link: https://lkml.kernel.org/r/20230126215125.4069751-12-kbusch@meta.com Fixes: 2d55c16c0c54 ("dmapool: create/destroy cleanup") Signed-off-by: Keith Busch <kbusch@kernel.org> Reviewed-by: Christoph Hellwig <hch@lst.de> Tested-by: Bryan O'Donoghue <bryan.odonoghue@linaro.org> Cc: Matthew Wilcox <willy@infradead.org> Cc: Tony Battersby <tonyb@cybernetics.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2023-01-27 05:51:24 +08:00
dev_err(pool->dev, "%s %s, %p (corrupted)\n", __func__,
pool->name, block);
/*
* Dump the first 4 bytes even if they are not
* POOL_POISON_FREED
*/
print_hex_dump(KERN_ERR, "", DUMP_PREFIX_OFFSET, 16, 1,
data, pool->size, 1);
break;
}
dmapool: link blocks across pages The allocated dmapool pages are never freed for the lifetime of the pool. There is no need for the two level list+stack lookup for finding a free block since nothing is ever removed from the list. Just use a simple stack, reducing time complexity to constant. The implementation inserts the stack linking elements and the dma handle of the block within itself when freed. This means the smallest possible dmapool block is increased to at most 16 bytes to accommodate these fields, but there are no exisiting users requesting a dma pool smaller than that anyway. Removing the list has a significant change in performance. Using the kernel's micro-benchmarking self test: Before: # modprobe dmapool_test dmapool test: size:16 blocks:8192 time:57282 dmapool test: size:64 blocks:8192 time:172562 dmapool test: size:256 blocks:8192 time:789247 dmapool test: size:1024 blocks:2048 time:371823 dmapool test: size:4096 blocks:1024 time:362237 After: # modprobe dmapool_test dmapool test: size:16 blocks:8192 time:24997 dmapool test: size:64 blocks:8192 time:26584 dmapool test: size:256 blocks:8192 time:33542 dmapool test: size:1024 blocks:2048 time:9022 dmapool test: size:4096 blocks:1024 time:6045 The module test allocates quite a few blocks that may not accurately represent how these pools are used in real life. For a more marco level benchmark, running fio high-depth + high-batched on nvme, this patch shows submission and completion latency reduced by ~100usec each, 1% IOPs improvement, and perf record's time spent in dma_pool_alloc/free were reduced by half. [kbusch@kernel.org: push new blocks in ascending order] Link: https://lkml.kernel.org/r/20230221165400.1595247-1-kbusch@meta.com Link: https://lkml.kernel.org/r/20230126215125.4069751-12-kbusch@meta.com Fixes: 2d55c16c0c54 ("dmapool: create/destroy cleanup") Signed-off-by: Keith Busch <kbusch@kernel.org> Reviewed-by: Christoph Hellwig <hch@lst.de> Tested-by: Bryan O'Donoghue <bryan.odonoghue@linaro.org> Cc: Matthew Wilcox <willy@infradead.org> Cc: Tony Battersby <tonyb@cybernetics.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2023-01-27 05:51:24 +08:00
if (!want_init_on_alloc(mem_flags))
dmapool: link blocks across pages The allocated dmapool pages are never freed for the lifetime of the pool. There is no need for the two level list+stack lookup for finding a free block since nothing is ever removed from the list. Just use a simple stack, reducing time complexity to constant. The implementation inserts the stack linking elements and the dma handle of the block within itself when freed. This means the smallest possible dmapool block is increased to at most 16 bytes to accommodate these fields, but there are no exisiting users requesting a dma pool smaller than that anyway. Removing the list has a significant change in performance. Using the kernel's micro-benchmarking self test: Before: # modprobe dmapool_test dmapool test: size:16 blocks:8192 time:57282 dmapool test: size:64 blocks:8192 time:172562 dmapool test: size:256 blocks:8192 time:789247 dmapool test: size:1024 blocks:2048 time:371823 dmapool test: size:4096 blocks:1024 time:362237 After: # modprobe dmapool_test dmapool test: size:16 blocks:8192 time:24997 dmapool test: size:64 blocks:8192 time:26584 dmapool test: size:256 blocks:8192 time:33542 dmapool test: size:1024 blocks:2048 time:9022 dmapool test: size:4096 blocks:1024 time:6045 The module test allocates quite a few blocks that may not accurately represent how these pools are used in real life. For a more marco level benchmark, running fio high-depth + high-batched on nvme, this patch shows submission and completion latency reduced by ~100usec each, 1% IOPs improvement, and perf record's time spent in dma_pool_alloc/free were reduced by half. [kbusch@kernel.org: push new blocks in ascending order] Link: https://lkml.kernel.org/r/20230221165400.1595247-1-kbusch@meta.com Link: https://lkml.kernel.org/r/20230126215125.4069751-12-kbusch@meta.com Fixes: 2d55c16c0c54 ("dmapool: create/destroy cleanup") Signed-off-by: Keith Busch <kbusch@kernel.org> Reviewed-by: Christoph Hellwig <hch@lst.de> Tested-by: Bryan O'Donoghue <bryan.odonoghue@linaro.org> Cc: Matthew Wilcox <willy@infradead.org> Cc: Tony Battersby <tonyb@cybernetics.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2023-01-27 05:51:24 +08:00
memset(block, POOL_POISON_ALLOCATED, pool->size);
}
dmapool: link blocks across pages The allocated dmapool pages are never freed for the lifetime of the pool. There is no need for the two level list+stack lookup for finding a free block since nothing is ever removed from the list. Just use a simple stack, reducing time complexity to constant. The implementation inserts the stack linking elements and the dma handle of the block within itself when freed. This means the smallest possible dmapool block is increased to at most 16 bytes to accommodate these fields, but there are no exisiting users requesting a dma pool smaller than that anyway. Removing the list has a significant change in performance. Using the kernel's micro-benchmarking self test: Before: # modprobe dmapool_test dmapool test: size:16 blocks:8192 time:57282 dmapool test: size:64 blocks:8192 time:172562 dmapool test: size:256 blocks:8192 time:789247 dmapool test: size:1024 blocks:2048 time:371823 dmapool test: size:4096 blocks:1024 time:362237 After: # modprobe dmapool_test dmapool test: size:16 blocks:8192 time:24997 dmapool test: size:64 blocks:8192 time:26584 dmapool test: size:256 blocks:8192 time:33542 dmapool test: size:1024 blocks:2048 time:9022 dmapool test: size:4096 blocks:1024 time:6045 The module test allocates quite a few blocks that may not accurately represent how these pools are used in real life. For a more marco level benchmark, running fio high-depth + high-batched on nvme, this patch shows submission and completion latency reduced by ~100usec each, 1% IOPs improvement, and perf record's time spent in dma_pool_alloc/free were reduced by half. [kbusch@kernel.org: push new blocks in ascending order] Link: https://lkml.kernel.org/r/20230221165400.1595247-1-kbusch@meta.com Link: https://lkml.kernel.org/r/20230126215125.4069751-12-kbusch@meta.com Fixes: 2d55c16c0c54 ("dmapool: create/destroy cleanup") Signed-off-by: Keith Busch <kbusch@kernel.org> Reviewed-by: Christoph Hellwig <hch@lst.de> Tested-by: Bryan O'Donoghue <bryan.odonoghue@linaro.org> Cc: Matthew Wilcox <willy@infradead.org> Cc: Tony Battersby <tonyb@cybernetics.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2023-01-27 05:51:24 +08:00
static struct dma_page *pool_find_page(struct dma_pool *pool, dma_addr_t dma)
{
dmapool: link blocks across pages The allocated dmapool pages are never freed for the lifetime of the pool. There is no need for the two level list+stack lookup for finding a free block since nothing is ever removed from the list. Just use a simple stack, reducing time complexity to constant. The implementation inserts the stack linking elements and the dma handle of the block within itself when freed. This means the smallest possible dmapool block is increased to at most 16 bytes to accommodate these fields, but there are no exisiting users requesting a dma pool smaller than that anyway. Removing the list has a significant change in performance. Using the kernel's micro-benchmarking self test: Before: # modprobe dmapool_test dmapool test: size:16 blocks:8192 time:57282 dmapool test: size:64 blocks:8192 time:172562 dmapool test: size:256 blocks:8192 time:789247 dmapool test: size:1024 blocks:2048 time:371823 dmapool test: size:4096 blocks:1024 time:362237 After: # modprobe dmapool_test dmapool test: size:16 blocks:8192 time:24997 dmapool test: size:64 blocks:8192 time:26584 dmapool test: size:256 blocks:8192 time:33542 dmapool test: size:1024 blocks:2048 time:9022 dmapool test: size:4096 blocks:1024 time:6045 The module test allocates quite a few blocks that may not accurately represent how these pools are used in real life. For a more marco level benchmark, running fio high-depth + high-batched on nvme, this patch shows submission and completion latency reduced by ~100usec each, 1% IOPs improvement, and perf record's time spent in dma_pool_alloc/free were reduced by half. [kbusch@kernel.org: push new blocks in ascending order] Link: https://lkml.kernel.org/r/20230221165400.1595247-1-kbusch@meta.com Link: https://lkml.kernel.org/r/20230126215125.4069751-12-kbusch@meta.com Fixes: 2d55c16c0c54 ("dmapool: create/destroy cleanup") Signed-off-by: Keith Busch <kbusch@kernel.org> Reviewed-by: Christoph Hellwig <hch@lst.de> Tested-by: Bryan O'Donoghue <bryan.odonoghue@linaro.org> Cc: Matthew Wilcox <willy@infradead.org> Cc: Tony Battersby <tonyb@cybernetics.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2023-01-27 05:51:24 +08:00
struct dma_page *page;
dmapool: link blocks across pages The allocated dmapool pages are never freed for the lifetime of the pool. There is no need for the two level list+stack lookup for finding a free block since nothing is ever removed from the list. Just use a simple stack, reducing time complexity to constant. The implementation inserts the stack linking elements and the dma handle of the block within itself when freed. This means the smallest possible dmapool block is increased to at most 16 bytes to accommodate these fields, but there are no exisiting users requesting a dma pool smaller than that anyway. Removing the list has a significant change in performance. Using the kernel's micro-benchmarking self test: Before: # modprobe dmapool_test dmapool test: size:16 blocks:8192 time:57282 dmapool test: size:64 blocks:8192 time:172562 dmapool test: size:256 blocks:8192 time:789247 dmapool test: size:1024 blocks:2048 time:371823 dmapool test: size:4096 blocks:1024 time:362237 After: # modprobe dmapool_test dmapool test: size:16 blocks:8192 time:24997 dmapool test: size:64 blocks:8192 time:26584 dmapool test: size:256 blocks:8192 time:33542 dmapool test: size:1024 blocks:2048 time:9022 dmapool test: size:4096 blocks:1024 time:6045 The module test allocates quite a few blocks that may not accurately represent how these pools are used in real life. For a more marco level benchmark, running fio high-depth + high-batched on nvme, this patch shows submission and completion latency reduced by ~100usec each, 1% IOPs improvement, and perf record's time spent in dma_pool_alloc/free were reduced by half. [kbusch@kernel.org: push new blocks in ascending order] Link: https://lkml.kernel.org/r/20230221165400.1595247-1-kbusch@meta.com Link: https://lkml.kernel.org/r/20230126215125.4069751-12-kbusch@meta.com Fixes: 2d55c16c0c54 ("dmapool: create/destroy cleanup") Signed-off-by: Keith Busch <kbusch@kernel.org> Reviewed-by: Christoph Hellwig <hch@lst.de> Tested-by: Bryan O'Donoghue <bryan.odonoghue@linaro.org> Cc: Matthew Wilcox <willy@infradead.org> Cc: Tony Battersby <tonyb@cybernetics.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2023-01-27 05:51:24 +08:00
list_for_each_entry(page, &pool->page_list, page_list) {
if (dma < page->dma)
continue;
if ((dma - page->dma) < pool->allocation)
return page;
}
return NULL;
}
static bool pool_block_err(struct dma_pool *pool, void *vaddr, dma_addr_t dma)
{
struct dma_block *block = pool->next_block;
struct dma_page *page;
page = pool_find_page(pool, dma);
if (!page) {
dev_err(pool->dev, "%s %s, %p/%pad (bad dma)\n",
__func__, pool->name, vaddr, &dma);
return true;
}
dmapool: link blocks across pages The allocated dmapool pages are never freed for the lifetime of the pool. There is no need for the two level list+stack lookup for finding a free block since nothing is ever removed from the list. Just use a simple stack, reducing time complexity to constant. The implementation inserts the stack linking elements and the dma handle of the block within itself when freed. This means the smallest possible dmapool block is increased to at most 16 bytes to accommodate these fields, but there are no exisiting users requesting a dma pool smaller than that anyway. Removing the list has a significant change in performance. Using the kernel's micro-benchmarking self test: Before: # modprobe dmapool_test dmapool test: size:16 blocks:8192 time:57282 dmapool test: size:64 blocks:8192 time:172562 dmapool test: size:256 blocks:8192 time:789247 dmapool test: size:1024 blocks:2048 time:371823 dmapool test: size:4096 blocks:1024 time:362237 After: # modprobe dmapool_test dmapool test: size:16 blocks:8192 time:24997 dmapool test: size:64 blocks:8192 time:26584 dmapool test: size:256 blocks:8192 time:33542 dmapool test: size:1024 blocks:2048 time:9022 dmapool test: size:4096 blocks:1024 time:6045 The module test allocates quite a few blocks that may not accurately represent how these pools are used in real life. For a more marco level benchmark, running fio high-depth + high-batched on nvme, this patch shows submission and completion latency reduced by ~100usec each, 1% IOPs improvement, and perf record's time spent in dma_pool_alloc/free were reduced by half. [kbusch@kernel.org: push new blocks in ascending order] Link: https://lkml.kernel.org/r/20230221165400.1595247-1-kbusch@meta.com Link: https://lkml.kernel.org/r/20230126215125.4069751-12-kbusch@meta.com Fixes: 2d55c16c0c54 ("dmapool: create/destroy cleanup") Signed-off-by: Keith Busch <kbusch@kernel.org> Reviewed-by: Christoph Hellwig <hch@lst.de> Tested-by: Bryan O'Donoghue <bryan.odonoghue@linaro.org> Cc: Matthew Wilcox <willy@infradead.org> Cc: Tony Battersby <tonyb@cybernetics.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2023-01-27 05:51:24 +08:00
while (block) {
if (block != vaddr) {
block = block->next_block;
continue;
}
dev_err(pool->dev, "%s %s, dma %pad already free\n",
__func__, pool->name, &dma);
return true;
}
dmapool: link blocks across pages The allocated dmapool pages are never freed for the lifetime of the pool. There is no need for the two level list+stack lookup for finding a free block since nothing is ever removed from the list. Just use a simple stack, reducing time complexity to constant. The implementation inserts the stack linking elements and the dma handle of the block within itself when freed. This means the smallest possible dmapool block is increased to at most 16 bytes to accommodate these fields, but there are no exisiting users requesting a dma pool smaller than that anyway. Removing the list has a significant change in performance. Using the kernel's micro-benchmarking self test: Before: # modprobe dmapool_test dmapool test: size:16 blocks:8192 time:57282 dmapool test: size:64 blocks:8192 time:172562 dmapool test: size:256 blocks:8192 time:789247 dmapool test: size:1024 blocks:2048 time:371823 dmapool test: size:4096 blocks:1024 time:362237 After: # modprobe dmapool_test dmapool test: size:16 blocks:8192 time:24997 dmapool test: size:64 blocks:8192 time:26584 dmapool test: size:256 blocks:8192 time:33542 dmapool test: size:1024 blocks:2048 time:9022 dmapool test: size:4096 blocks:1024 time:6045 The module test allocates quite a few blocks that may not accurately represent how these pools are used in real life. For a more marco level benchmark, running fio high-depth + high-batched on nvme, this patch shows submission and completion latency reduced by ~100usec each, 1% IOPs improvement, and perf record's time spent in dma_pool_alloc/free were reduced by half. [kbusch@kernel.org: push new blocks in ascending order] Link: https://lkml.kernel.org/r/20230221165400.1595247-1-kbusch@meta.com Link: https://lkml.kernel.org/r/20230126215125.4069751-12-kbusch@meta.com Fixes: 2d55c16c0c54 ("dmapool: create/destroy cleanup") Signed-off-by: Keith Busch <kbusch@kernel.org> Reviewed-by: Christoph Hellwig <hch@lst.de> Tested-by: Bryan O'Donoghue <bryan.odonoghue@linaro.org> Cc: Matthew Wilcox <willy@infradead.org> Cc: Tony Battersby <tonyb@cybernetics.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2023-01-27 05:51:24 +08:00
memset(vaddr, POOL_POISON_FREED, pool->size);
return false;
}
static void pool_init_page(struct dma_pool *pool, struct dma_page *page)
{
memset(page->vaddr, POOL_POISON_FREED, pool->allocation);
}
#else
dmapool: link blocks across pages The allocated dmapool pages are never freed for the lifetime of the pool. There is no need for the two level list+stack lookup for finding a free block since nothing is ever removed from the list. Just use a simple stack, reducing time complexity to constant. The implementation inserts the stack linking elements and the dma handle of the block within itself when freed. This means the smallest possible dmapool block is increased to at most 16 bytes to accommodate these fields, but there are no exisiting users requesting a dma pool smaller than that anyway. Removing the list has a significant change in performance. Using the kernel's micro-benchmarking self test: Before: # modprobe dmapool_test dmapool test: size:16 blocks:8192 time:57282 dmapool test: size:64 blocks:8192 time:172562 dmapool test: size:256 blocks:8192 time:789247 dmapool test: size:1024 blocks:2048 time:371823 dmapool test: size:4096 blocks:1024 time:362237 After: # modprobe dmapool_test dmapool test: size:16 blocks:8192 time:24997 dmapool test: size:64 blocks:8192 time:26584 dmapool test: size:256 blocks:8192 time:33542 dmapool test: size:1024 blocks:2048 time:9022 dmapool test: size:4096 blocks:1024 time:6045 The module test allocates quite a few blocks that may not accurately represent how these pools are used in real life. For a more marco level benchmark, running fio high-depth + high-batched on nvme, this patch shows submission and completion latency reduced by ~100usec each, 1% IOPs improvement, and perf record's time spent in dma_pool_alloc/free were reduced by half. [kbusch@kernel.org: push new blocks in ascending order] Link: https://lkml.kernel.org/r/20230221165400.1595247-1-kbusch@meta.com Link: https://lkml.kernel.org/r/20230126215125.4069751-12-kbusch@meta.com Fixes: 2d55c16c0c54 ("dmapool: create/destroy cleanup") Signed-off-by: Keith Busch <kbusch@kernel.org> Reviewed-by: Christoph Hellwig <hch@lst.de> Tested-by: Bryan O'Donoghue <bryan.odonoghue@linaro.org> Cc: Matthew Wilcox <willy@infradead.org> Cc: Tony Battersby <tonyb@cybernetics.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2023-01-27 05:51:24 +08:00
static void pool_check_block(struct dma_pool *pool, struct dma_block *block,
gfp_t mem_flags)
{
}
dmapool: link blocks across pages The allocated dmapool pages are never freed for the lifetime of the pool. There is no need for the two level list+stack lookup for finding a free block since nothing is ever removed from the list. Just use a simple stack, reducing time complexity to constant. The implementation inserts the stack linking elements and the dma handle of the block within itself when freed. This means the smallest possible dmapool block is increased to at most 16 bytes to accommodate these fields, but there are no exisiting users requesting a dma pool smaller than that anyway. Removing the list has a significant change in performance. Using the kernel's micro-benchmarking self test: Before: # modprobe dmapool_test dmapool test: size:16 blocks:8192 time:57282 dmapool test: size:64 blocks:8192 time:172562 dmapool test: size:256 blocks:8192 time:789247 dmapool test: size:1024 blocks:2048 time:371823 dmapool test: size:4096 blocks:1024 time:362237 After: # modprobe dmapool_test dmapool test: size:16 blocks:8192 time:24997 dmapool test: size:64 blocks:8192 time:26584 dmapool test: size:256 blocks:8192 time:33542 dmapool test: size:1024 blocks:2048 time:9022 dmapool test: size:4096 blocks:1024 time:6045 The module test allocates quite a few blocks that may not accurately represent how these pools are used in real life. For a more marco level benchmark, running fio high-depth + high-batched on nvme, this patch shows submission and completion latency reduced by ~100usec each, 1% IOPs improvement, and perf record's time spent in dma_pool_alloc/free were reduced by half. [kbusch@kernel.org: push new blocks in ascending order] Link: https://lkml.kernel.org/r/20230221165400.1595247-1-kbusch@meta.com Link: https://lkml.kernel.org/r/20230126215125.4069751-12-kbusch@meta.com Fixes: 2d55c16c0c54 ("dmapool: create/destroy cleanup") Signed-off-by: Keith Busch <kbusch@kernel.org> Reviewed-by: Christoph Hellwig <hch@lst.de> Tested-by: Bryan O'Donoghue <bryan.odonoghue@linaro.org> Cc: Matthew Wilcox <willy@infradead.org> Cc: Tony Battersby <tonyb@cybernetics.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2023-01-27 05:51:24 +08:00
static bool pool_block_err(struct dma_pool *pool, void *vaddr, dma_addr_t dma)
{
if (want_init_on_free())
memset(vaddr, 0, pool->size);
return false;
}
static void pool_init_page(struct dma_pool *pool, struct dma_page *page)
{
}
#endif
dmapool: link blocks across pages The allocated dmapool pages are never freed for the lifetime of the pool. There is no need for the two level list+stack lookup for finding a free block since nothing is ever removed from the list. Just use a simple stack, reducing time complexity to constant. The implementation inserts the stack linking elements and the dma handle of the block within itself when freed. This means the smallest possible dmapool block is increased to at most 16 bytes to accommodate these fields, but there are no exisiting users requesting a dma pool smaller than that anyway. Removing the list has a significant change in performance. Using the kernel's micro-benchmarking self test: Before: # modprobe dmapool_test dmapool test: size:16 blocks:8192 time:57282 dmapool test: size:64 blocks:8192 time:172562 dmapool test: size:256 blocks:8192 time:789247 dmapool test: size:1024 blocks:2048 time:371823 dmapool test: size:4096 blocks:1024 time:362237 After: # modprobe dmapool_test dmapool test: size:16 blocks:8192 time:24997 dmapool test: size:64 blocks:8192 time:26584 dmapool test: size:256 blocks:8192 time:33542 dmapool test: size:1024 blocks:2048 time:9022 dmapool test: size:4096 blocks:1024 time:6045 The module test allocates quite a few blocks that may not accurately represent how these pools are used in real life. For a more marco level benchmark, running fio high-depth + high-batched on nvme, this patch shows submission and completion latency reduced by ~100usec each, 1% IOPs improvement, and perf record's time spent in dma_pool_alloc/free were reduced by half. [kbusch@kernel.org: push new blocks in ascending order] Link: https://lkml.kernel.org/r/20230221165400.1595247-1-kbusch@meta.com Link: https://lkml.kernel.org/r/20230126215125.4069751-12-kbusch@meta.com Fixes: 2d55c16c0c54 ("dmapool: create/destroy cleanup") Signed-off-by: Keith Busch <kbusch@kernel.org> Reviewed-by: Christoph Hellwig <hch@lst.de> Tested-by: Bryan O'Donoghue <bryan.odonoghue@linaro.org> Cc: Matthew Wilcox <willy@infradead.org> Cc: Tony Battersby <tonyb@cybernetics.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2023-01-27 05:51:24 +08:00
static struct dma_block *pool_block_pop(struct dma_pool *pool)
{
struct dma_block *block = pool->next_block;
if (block) {
pool->next_block = block->next_block;
pool->nr_active++;
}
return block;
}
static void pool_block_push(struct dma_pool *pool, struct dma_block *block,
dma_addr_t dma)
{
block->dma = dma;
block->next_block = pool->next_block;
pool->next_block = block;
}
/**
* dma_pool_create - Creates a pool of consistent memory blocks, for dma.
* @name: name of pool, for diagnostics
* @dev: device that will be doing the DMA
* @size: size of the blocks in this pool.
* @align: alignment requirement for blocks; must be a power of two
* @boundary: returned blocks won't cross this power of two boundary
* Context: not in_interrupt()
*
* Given one of these pools, dma_pool_alloc()
* may be used to allocate memory. Such memory will all have "consistent"
* DMA mappings, accessible by the device and its driver without using
* cache flushing primitives. The actual size of blocks allocated may be
* larger than requested because of alignment.
*
* If @boundary is nonzero, objects returned from dma_pool_alloc() won't
* cross that size boundary. This is useful for devices which have
* addressing restrictions on individual DMA transfers, such as not crossing
* boundaries of 4KBytes.
*
* Return: a dma allocation pool with the requested characteristics, or
* %NULL if one can't be created.
*/
struct dma_pool *dma_pool_create(const char *name, struct device *dev,
size_t size, size_t align, size_t boundary)
{
struct dma_pool *retval;
size_t allocation;
bool empty;
if (!dev)
return NULL;
if (align == 0)
align = 1;
else if (align & (align - 1))
return NULL;
if (size == 0 || size > INT_MAX)
return NULL;
dmapool: link blocks across pages The allocated dmapool pages are never freed for the lifetime of the pool. There is no need for the two level list+stack lookup for finding a free block since nothing is ever removed from the list. Just use a simple stack, reducing time complexity to constant. The implementation inserts the stack linking elements and the dma handle of the block within itself when freed. This means the smallest possible dmapool block is increased to at most 16 bytes to accommodate these fields, but there are no exisiting users requesting a dma pool smaller than that anyway. Removing the list has a significant change in performance. Using the kernel's micro-benchmarking self test: Before: # modprobe dmapool_test dmapool test: size:16 blocks:8192 time:57282 dmapool test: size:64 blocks:8192 time:172562 dmapool test: size:256 blocks:8192 time:789247 dmapool test: size:1024 blocks:2048 time:371823 dmapool test: size:4096 blocks:1024 time:362237 After: # modprobe dmapool_test dmapool test: size:16 blocks:8192 time:24997 dmapool test: size:64 blocks:8192 time:26584 dmapool test: size:256 blocks:8192 time:33542 dmapool test: size:1024 blocks:2048 time:9022 dmapool test: size:4096 blocks:1024 time:6045 The module test allocates quite a few blocks that may not accurately represent how these pools are used in real life. For a more marco level benchmark, running fio high-depth + high-batched on nvme, this patch shows submission and completion latency reduced by ~100usec each, 1% IOPs improvement, and perf record's time spent in dma_pool_alloc/free were reduced by half. [kbusch@kernel.org: push new blocks in ascending order] Link: https://lkml.kernel.org/r/20230221165400.1595247-1-kbusch@meta.com Link: https://lkml.kernel.org/r/20230126215125.4069751-12-kbusch@meta.com Fixes: 2d55c16c0c54 ("dmapool: create/destroy cleanup") Signed-off-by: Keith Busch <kbusch@kernel.org> Reviewed-by: Christoph Hellwig <hch@lst.de> Tested-by: Bryan O'Donoghue <bryan.odonoghue@linaro.org> Cc: Matthew Wilcox <willy@infradead.org> Cc: Tony Battersby <tonyb@cybernetics.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2023-01-27 05:51:24 +08:00
if (size < sizeof(struct dma_block))
size = sizeof(struct dma_block);
size = ALIGN(size, align);
allocation = max_t(size_t, size, PAGE_SIZE);
if (!boundary)
boundary = allocation;
else if ((boundary < size) || (boundary & (boundary - 1)))
return NULL;
boundary = min(boundary, allocation);
dmapool: link blocks across pages The allocated dmapool pages are never freed for the lifetime of the pool. There is no need for the two level list+stack lookup for finding a free block since nothing is ever removed from the list. Just use a simple stack, reducing time complexity to constant. The implementation inserts the stack linking elements and the dma handle of the block within itself when freed. This means the smallest possible dmapool block is increased to at most 16 bytes to accommodate these fields, but there are no exisiting users requesting a dma pool smaller than that anyway. Removing the list has a significant change in performance. Using the kernel's micro-benchmarking self test: Before: # modprobe dmapool_test dmapool test: size:16 blocks:8192 time:57282 dmapool test: size:64 blocks:8192 time:172562 dmapool test: size:256 blocks:8192 time:789247 dmapool test: size:1024 blocks:2048 time:371823 dmapool test: size:4096 blocks:1024 time:362237 After: # modprobe dmapool_test dmapool test: size:16 blocks:8192 time:24997 dmapool test: size:64 blocks:8192 time:26584 dmapool test: size:256 blocks:8192 time:33542 dmapool test: size:1024 blocks:2048 time:9022 dmapool test: size:4096 blocks:1024 time:6045 The module test allocates quite a few blocks that may not accurately represent how these pools are used in real life. For a more marco level benchmark, running fio high-depth + high-batched on nvme, this patch shows submission and completion latency reduced by ~100usec each, 1% IOPs improvement, and perf record's time spent in dma_pool_alloc/free were reduced by half. [kbusch@kernel.org: push new blocks in ascending order] Link: https://lkml.kernel.org/r/20230221165400.1595247-1-kbusch@meta.com Link: https://lkml.kernel.org/r/20230126215125.4069751-12-kbusch@meta.com Fixes: 2d55c16c0c54 ("dmapool: create/destroy cleanup") Signed-off-by: Keith Busch <kbusch@kernel.org> Reviewed-by: Christoph Hellwig <hch@lst.de> Tested-by: Bryan O'Donoghue <bryan.odonoghue@linaro.org> Cc: Matthew Wilcox <willy@infradead.org> Cc: Tony Battersby <tonyb@cybernetics.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2023-01-27 05:51:24 +08:00
retval = kzalloc(sizeof(*retval), GFP_KERNEL);
if (!retval)
return retval;
strscpy(retval->name, name, sizeof(retval->name));
retval->dev = dev;
INIT_LIST_HEAD(&retval->page_list);
spin_lock_init(&retval->lock);
retval->size = size;
retval->boundary = boundary;
retval->allocation = allocation;
INIT_LIST_HEAD(&retval->pools);
mm: dmapool: add/remove sysfs file outside of the pool lock lock cat /sys/.../pools followed by removal the device leads to: |====================================================== |[ INFO: possible circular locking dependency detected ] |3.17.0-rc4+ #1498 Not tainted |------------------------------------------------------- |rmmod/2505 is trying to acquire lock: | (s_active#28){++++.+}, at: [<c017f754>] kernfs_remove_by_name_ns+0x3c/0x88 | |but task is already holding lock: | (pools_lock){+.+.+.}, at: [<c011494c>] dma_pool_destroy+0x18/0x17c | |which lock already depends on the new lock. |the existing dependency chain (in reverse order) is: | |-> #1 (pools_lock){+.+.+.}: | [<c0114ae8>] show_pools+0x30/0xf8 | [<c0313210>] dev_attr_show+0x1c/0x48 | [<c0180e84>] sysfs_kf_seq_show+0x88/0x10c | [<c017f960>] kernfs_seq_show+0x24/0x28 | [<c013efc4>] seq_read+0x1b8/0x480 | [<c011e820>] vfs_read+0x8c/0x148 | [<c011ea10>] SyS_read+0x40/0x8c | [<c000e960>] ret_fast_syscall+0x0/0x48 | |-> #0 (s_active#28){++++.+}: | [<c017e9ac>] __kernfs_remove+0x258/0x2ec | [<c017f754>] kernfs_remove_by_name_ns+0x3c/0x88 | [<c0114a7c>] dma_pool_destroy+0x148/0x17c | [<c03ad288>] hcd_buffer_destroy+0x20/0x34 | [<c03a4780>] usb_remove_hcd+0x110/0x1a4 The problem is the lock order of pools_lock and kernfs_mutex in dma_pool_destroy() vs show_pools() call path. This patch breaks out the creation of the sysfs file outside of the pools_lock mutex. The newly added pools_reg_lock ensures that there is no race of create vs destroy code path in terms whether or not the sysfs file has to be deleted (and was it deleted before we try to create a new one) and what to do if device_create_file() failed. Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2014-10-10 06:28:50 +08:00
/*
* pools_lock ensures that the ->dma_pools list does not get corrupted.
* pools_reg_lock ensures that there is not a race between
* dma_pool_create() and dma_pool_destroy() or within dma_pool_create()
* when the first invocation of dma_pool_create() failed on
* device_create_file() and the second assumes that it has been done (I
* know it is a short window).
*/
mutex_lock(&pools_reg_lock);
mutex_lock(&pools_lock);
empty = list_empty(&dev->dma_pools);
mm: dmapool: add/remove sysfs file outside of the pool lock lock cat /sys/.../pools followed by removal the device leads to: |====================================================== |[ INFO: possible circular locking dependency detected ] |3.17.0-rc4+ #1498 Not tainted |------------------------------------------------------- |rmmod/2505 is trying to acquire lock: | (s_active#28){++++.+}, at: [<c017f754>] kernfs_remove_by_name_ns+0x3c/0x88 | |but task is already holding lock: | (pools_lock){+.+.+.}, at: [<c011494c>] dma_pool_destroy+0x18/0x17c | |which lock already depends on the new lock. |the existing dependency chain (in reverse order) is: | |-> #1 (pools_lock){+.+.+.}: | [<c0114ae8>] show_pools+0x30/0xf8 | [<c0313210>] dev_attr_show+0x1c/0x48 | [<c0180e84>] sysfs_kf_seq_show+0x88/0x10c | [<c017f960>] kernfs_seq_show+0x24/0x28 | [<c013efc4>] seq_read+0x1b8/0x480 | [<c011e820>] vfs_read+0x8c/0x148 | [<c011ea10>] SyS_read+0x40/0x8c | [<c000e960>] ret_fast_syscall+0x0/0x48 | |-> #0 (s_active#28){++++.+}: | [<c017e9ac>] __kernfs_remove+0x258/0x2ec | [<c017f754>] kernfs_remove_by_name_ns+0x3c/0x88 | [<c0114a7c>] dma_pool_destroy+0x148/0x17c | [<c03ad288>] hcd_buffer_destroy+0x20/0x34 | [<c03a4780>] usb_remove_hcd+0x110/0x1a4 The problem is the lock order of pools_lock and kernfs_mutex in dma_pool_destroy() vs show_pools() call path. This patch breaks out the creation of the sysfs file outside of the pools_lock mutex. The newly added pools_reg_lock ensures that there is no race of create vs destroy code path in terms whether or not the sysfs file has to be deleted (and was it deleted before we try to create a new one) and what to do if device_create_file() failed. Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2014-10-10 06:28:50 +08:00
list_add(&retval->pools, &dev->dma_pools);
mutex_unlock(&pools_lock);
mm: dmapool: add/remove sysfs file outside of the pool lock lock cat /sys/.../pools followed by removal the device leads to: |====================================================== |[ INFO: possible circular locking dependency detected ] |3.17.0-rc4+ #1498 Not tainted |------------------------------------------------------- |rmmod/2505 is trying to acquire lock: | (s_active#28){++++.+}, at: [<c017f754>] kernfs_remove_by_name_ns+0x3c/0x88 | |but task is already holding lock: | (pools_lock){+.+.+.}, at: [<c011494c>] dma_pool_destroy+0x18/0x17c | |which lock already depends on the new lock. |the existing dependency chain (in reverse order) is: | |-> #1 (pools_lock){+.+.+.}: | [<c0114ae8>] show_pools+0x30/0xf8 | [<c0313210>] dev_attr_show+0x1c/0x48 | [<c0180e84>] sysfs_kf_seq_show+0x88/0x10c | [<c017f960>] kernfs_seq_show+0x24/0x28 | [<c013efc4>] seq_read+0x1b8/0x480 | [<c011e820>] vfs_read+0x8c/0x148 | [<c011ea10>] SyS_read+0x40/0x8c | [<c000e960>] ret_fast_syscall+0x0/0x48 | |-> #0 (s_active#28){++++.+}: | [<c017e9ac>] __kernfs_remove+0x258/0x2ec | [<c017f754>] kernfs_remove_by_name_ns+0x3c/0x88 | [<c0114a7c>] dma_pool_destroy+0x148/0x17c | [<c03ad288>] hcd_buffer_destroy+0x20/0x34 | [<c03a4780>] usb_remove_hcd+0x110/0x1a4 The problem is the lock order of pools_lock and kernfs_mutex in dma_pool_destroy() vs show_pools() call path. This patch breaks out the creation of the sysfs file outside of the pools_lock mutex. The newly added pools_reg_lock ensures that there is no race of create vs destroy code path in terms whether or not the sysfs file has to be deleted (and was it deleted before we try to create a new one) and what to do if device_create_file() failed. Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2014-10-10 06:28:50 +08:00
if (empty) {
int err;
err = device_create_file(dev, &dev_attr_pools);
if (err) {
mutex_lock(&pools_lock);
list_del(&retval->pools);
mutex_unlock(&pools_lock);
mutex_unlock(&pools_reg_lock);
kfree(retval);
return NULL;
}
}
mutex_unlock(&pools_reg_lock);
return retval;
}
EXPORT_SYMBOL(dma_pool_create);
static void pool_initialise_page(struct dma_pool *pool, struct dma_page *page)
{
dmapool: link blocks across pages The allocated dmapool pages are never freed for the lifetime of the pool. There is no need for the two level list+stack lookup for finding a free block since nothing is ever removed from the list. Just use a simple stack, reducing time complexity to constant. The implementation inserts the stack linking elements and the dma handle of the block within itself when freed. This means the smallest possible dmapool block is increased to at most 16 bytes to accommodate these fields, but there are no exisiting users requesting a dma pool smaller than that anyway. Removing the list has a significant change in performance. Using the kernel's micro-benchmarking self test: Before: # modprobe dmapool_test dmapool test: size:16 blocks:8192 time:57282 dmapool test: size:64 blocks:8192 time:172562 dmapool test: size:256 blocks:8192 time:789247 dmapool test: size:1024 blocks:2048 time:371823 dmapool test: size:4096 blocks:1024 time:362237 After: # modprobe dmapool_test dmapool test: size:16 blocks:8192 time:24997 dmapool test: size:64 blocks:8192 time:26584 dmapool test: size:256 blocks:8192 time:33542 dmapool test: size:1024 blocks:2048 time:9022 dmapool test: size:4096 blocks:1024 time:6045 The module test allocates quite a few blocks that may not accurately represent how these pools are used in real life. For a more marco level benchmark, running fio high-depth + high-batched on nvme, this patch shows submission and completion latency reduced by ~100usec each, 1% IOPs improvement, and perf record's time spent in dma_pool_alloc/free were reduced by half. [kbusch@kernel.org: push new blocks in ascending order] Link: https://lkml.kernel.org/r/20230221165400.1595247-1-kbusch@meta.com Link: https://lkml.kernel.org/r/20230126215125.4069751-12-kbusch@meta.com Fixes: 2d55c16c0c54 ("dmapool: create/destroy cleanup") Signed-off-by: Keith Busch <kbusch@kernel.org> Reviewed-by: Christoph Hellwig <hch@lst.de> Tested-by: Bryan O'Donoghue <bryan.odonoghue@linaro.org> Cc: Matthew Wilcox <willy@infradead.org> Cc: Tony Battersby <tonyb@cybernetics.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2023-01-27 05:51:24 +08:00
unsigned int next_boundary = pool->boundary, offset = 0;
struct dma_block *block, *first = NULL, *last = NULL;
pool_init_page(pool, page);
dmapool: link blocks across pages The allocated dmapool pages are never freed for the lifetime of the pool. There is no need for the two level list+stack lookup for finding a free block since nothing is ever removed from the list. Just use a simple stack, reducing time complexity to constant. The implementation inserts the stack linking elements and the dma handle of the block within itself when freed. This means the smallest possible dmapool block is increased to at most 16 bytes to accommodate these fields, but there are no exisiting users requesting a dma pool smaller than that anyway. Removing the list has a significant change in performance. Using the kernel's micro-benchmarking self test: Before: # modprobe dmapool_test dmapool test: size:16 blocks:8192 time:57282 dmapool test: size:64 blocks:8192 time:172562 dmapool test: size:256 blocks:8192 time:789247 dmapool test: size:1024 blocks:2048 time:371823 dmapool test: size:4096 blocks:1024 time:362237 After: # modprobe dmapool_test dmapool test: size:16 blocks:8192 time:24997 dmapool test: size:64 blocks:8192 time:26584 dmapool test: size:256 blocks:8192 time:33542 dmapool test: size:1024 blocks:2048 time:9022 dmapool test: size:4096 blocks:1024 time:6045 The module test allocates quite a few blocks that may not accurately represent how these pools are used in real life. For a more marco level benchmark, running fio high-depth + high-batched on nvme, this patch shows submission and completion latency reduced by ~100usec each, 1% IOPs improvement, and perf record's time spent in dma_pool_alloc/free were reduced by half. [kbusch@kernel.org: push new blocks in ascending order] Link: https://lkml.kernel.org/r/20230221165400.1595247-1-kbusch@meta.com Link: https://lkml.kernel.org/r/20230126215125.4069751-12-kbusch@meta.com Fixes: 2d55c16c0c54 ("dmapool: create/destroy cleanup") Signed-off-by: Keith Busch <kbusch@kernel.org> Reviewed-by: Christoph Hellwig <hch@lst.de> Tested-by: Bryan O'Donoghue <bryan.odonoghue@linaro.org> Cc: Matthew Wilcox <willy@infradead.org> Cc: Tony Battersby <tonyb@cybernetics.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2023-01-27 05:51:24 +08:00
while (offset + pool->size <= pool->allocation) {
if (offset + pool->size > next_boundary) {
offset = next_boundary;
next_boundary += pool->boundary;
dmapool: link blocks across pages The allocated dmapool pages are never freed for the lifetime of the pool. There is no need for the two level list+stack lookup for finding a free block since nothing is ever removed from the list. Just use a simple stack, reducing time complexity to constant. The implementation inserts the stack linking elements and the dma handle of the block within itself when freed. This means the smallest possible dmapool block is increased to at most 16 bytes to accommodate these fields, but there are no exisiting users requesting a dma pool smaller than that anyway. Removing the list has a significant change in performance. Using the kernel's micro-benchmarking self test: Before: # modprobe dmapool_test dmapool test: size:16 blocks:8192 time:57282 dmapool test: size:64 blocks:8192 time:172562 dmapool test: size:256 blocks:8192 time:789247 dmapool test: size:1024 blocks:2048 time:371823 dmapool test: size:4096 blocks:1024 time:362237 After: # modprobe dmapool_test dmapool test: size:16 blocks:8192 time:24997 dmapool test: size:64 blocks:8192 time:26584 dmapool test: size:256 blocks:8192 time:33542 dmapool test: size:1024 blocks:2048 time:9022 dmapool test: size:4096 blocks:1024 time:6045 The module test allocates quite a few blocks that may not accurately represent how these pools are used in real life. For a more marco level benchmark, running fio high-depth + high-batched on nvme, this patch shows submission and completion latency reduced by ~100usec each, 1% IOPs improvement, and perf record's time spent in dma_pool_alloc/free were reduced by half. [kbusch@kernel.org: push new blocks in ascending order] Link: https://lkml.kernel.org/r/20230221165400.1595247-1-kbusch@meta.com Link: https://lkml.kernel.org/r/20230126215125.4069751-12-kbusch@meta.com Fixes: 2d55c16c0c54 ("dmapool: create/destroy cleanup") Signed-off-by: Keith Busch <kbusch@kernel.org> Reviewed-by: Christoph Hellwig <hch@lst.de> Tested-by: Bryan O'Donoghue <bryan.odonoghue@linaro.org> Cc: Matthew Wilcox <willy@infradead.org> Cc: Tony Battersby <tonyb@cybernetics.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2023-01-27 05:51:24 +08:00
continue;
}
dmapool: link blocks across pages The allocated dmapool pages are never freed for the lifetime of the pool. There is no need for the two level list+stack lookup for finding a free block since nothing is ever removed from the list. Just use a simple stack, reducing time complexity to constant. The implementation inserts the stack linking elements and the dma handle of the block within itself when freed. This means the smallest possible dmapool block is increased to at most 16 bytes to accommodate these fields, but there are no exisiting users requesting a dma pool smaller than that anyway. Removing the list has a significant change in performance. Using the kernel's micro-benchmarking self test: Before: # modprobe dmapool_test dmapool test: size:16 blocks:8192 time:57282 dmapool test: size:64 blocks:8192 time:172562 dmapool test: size:256 blocks:8192 time:789247 dmapool test: size:1024 blocks:2048 time:371823 dmapool test: size:4096 blocks:1024 time:362237 After: # modprobe dmapool_test dmapool test: size:16 blocks:8192 time:24997 dmapool test: size:64 blocks:8192 time:26584 dmapool test: size:256 blocks:8192 time:33542 dmapool test: size:1024 blocks:2048 time:9022 dmapool test: size:4096 blocks:1024 time:6045 The module test allocates quite a few blocks that may not accurately represent how these pools are used in real life. For a more marco level benchmark, running fio high-depth + high-batched on nvme, this patch shows submission and completion latency reduced by ~100usec each, 1% IOPs improvement, and perf record's time spent in dma_pool_alloc/free were reduced by half. [kbusch@kernel.org: push new blocks in ascending order] Link: https://lkml.kernel.org/r/20230221165400.1595247-1-kbusch@meta.com Link: https://lkml.kernel.org/r/20230126215125.4069751-12-kbusch@meta.com Fixes: 2d55c16c0c54 ("dmapool: create/destroy cleanup") Signed-off-by: Keith Busch <kbusch@kernel.org> Reviewed-by: Christoph Hellwig <hch@lst.de> Tested-by: Bryan O'Donoghue <bryan.odonoghue@linaro.org> Cc: Matthew Wilcox <willy@infradead.org> Cc: Tony Battersby <tonyb@cybernetics.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2023-01-27 05:51:24 +08:00
block = page->vaddr + offset;
block->dma = page->dma + offset;
block->next_block = NULL;
if (last)
last->next_block = block;
else
first = block;
last = block;
offset += pool->size;
pool->nr_blocks++;
}
last->next_block = pool->next_block;
pool->next_block = first;
list_add(&page->page_list, &pool->page_list);
pool->nr_pages++;
}
static struct dma_page *pool_alloc_page(struct dma_pool *pool, gfp_t mem_flags)
{
struct dma_page *page;
page = kmalloc(sizeof(*page), mem_flags);
if (!page)
return NULL;
page->vaddr = dma_alloc_coherent(pool->dev, pool->allocation,
&page->dma, mem_flags);
if (!page->vaddr) {
kfree(page);
return NULL;
}
return page;
}
/**
* dma_pool_destroy - destroys a pool of dma memory blocks.
* @pool: dma pool that will be destroyed
* Context: !in_interrupt()
*
* Caller guarantees that no more memory from the pool is in use,
* and that nothing will try to use the pool after this call.
*/
void dma_pool_destroy(struct dma_pool *pool)
{
struct dma_page *page, *tmp;
bool empty, busy = false;
mm: dmapool: add/remove sysfs file outside of the pool lock lock cat /sys/.../pools followed by removal the device leads to: |====================================================== |[ INFO: possible circular locking dependency detected ] |3.17.0-rc4+ #1498 Not tainted |------------------------------------------------------- |rmmod/2505 is trying to acquire lock: | (s_active#28){++++.+}, at: [<c017f754>] kernfs_remove_by_name_ns+0x3c/0x88 | |but task is already holding lock: | (pools_lock){+.+.+.}, at: [<c011494c>] dma_pool_destroy+0x18/0x17c | |which lock already depends on the new lock. |the existing dependency chain (in reverse order) is: | |-> #1 (pools_lock){+.+.+.}: | [<c0114ae8>] show_pools+0x30/0xf8 | [<c0313210>] dev_attr_show+0x1c/0x48 | [<c0180e84>] sysfs_kf_seq_show+0x88/0x10c | [<c017f960>] kernfs_seq_show+0x24/0x28 | [<c013efc4>] seq_read+0x1b8/0x480 | [<c011e820>] vfs_read+0x8c/0x148 | [<c011ea10>] SyS_read+0x40/0x8c | [<c000e960>] ret_fast_syscall+0x0/0x48 | |-> #0 (s_active#28){++++.+}: | [<c017e9ac>] __kernfs_remove+0x258/0x2ec | [<c017f754>] kernfs_remove_by_name_ns+0x3c/0x88 | [<c0114a7c>] dma_pool_destroy+0x148/0x17c | [<c03ad288>] hcd_buffer_destroy+0x20/0x34 | [<c03a4780>] usb_remove_hcd+0x110/0x1a4 The problem is the lock order of pools_lock and kernfs_mutex in dma_pool_destroy() vs show_pools() call path. This patch breaks out the creation of the sysfs file outside of the pools_lock mutex. The newly added pools_reg_lock ensures that there is no race of create vs destroy code path in terms whether or not the sysfs file has to be deleted (and was it deleted before we try to create a new one) and what to do if device_create_file() failed. Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2014-10-10 06:28:50 +08:00
if (unlikely(!pool))
return;
mm: dmapool: add/remove sysfs file outside of the pool lock lock cat /sys/.../pools followed by removal the device leads to: |====================================================== |[ INFO: possible circular locking dependency detected ] |3.17.0-rc4+ #1498 Not tainted |------------------------------------------------------- |rmmod/2505 is trying to acquire lock: | (s_active#28){++++.+}, at: [<c017f754>] kernfs_remove_by_name_ns+0x3c/0x88 | |but task is already holding lock: | (pools_lock){+.+.+.}, at: [<c011494c>] dma_pool_destroy+0x18/0x17c | |which lock already depends on the new lock. |the existing dependency chain (in reverse order) is: | |-> #1 (pools_lock){+.+.+.}: | [<c0114ae8>] show_pools+0x30/0xf8 | [<c0313210>] dev_attr_show+0x1c/0x48 | [<c0180e84>] sysfs_kf_seq_show+0x88/0x10c | [<c017f960>] kernfs_seq_show+0x24/0x28 | [<c013efc4>] seq_read+0x1b8/0x480 | [<c011e820>] vfs_read+0x8c/0x148 | [<c011ea10>] SyS_read+0x40/0x8c | [<c000e960>] ret_fast_syscall+0x0/0x48 | |-> #0 (s_active#28){++++.+}: | [<c017e9ac>] __kernfs_remove+0x258/0x2ec | [<c017f754>] kernfs_remove_by_name_ns+0x3c/0x88 | [<c0114a7c>] dma_pool_destroy+0x148/0x17c | [<c03ad288>] hcd_buffer_destroy+0x20/0x34 | [<c03a4780>] usb_remove_hcd+0x110/0x1a4 The problem is the lock order of pools_lock and kernfs_mutex in dma_pool_destroy() vs show_pools() call path. This patch breaks out the creation of the sysfs file outside of the pools_lock mutex. The newly added pools_reg_lock ensures that there is no race of create vs destroy code path in terms whether or not the sysfs file has to be deleted (and was it deleted before we try to create a new one) and what to do if device_create_file() failed. Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2014-10-10 06:28:50 +08:00
mutex_lock(&pools_reg_lock);
mutex_lock(&pools_lock);
list_del(&pool->pools);
empty = list_empty(&pool->dev->dma_pools);
mutex_unlock(&pools_lock);
mm: dmapool: add/remove sysfs file outside of the pool lock lock cat /sys/.../pools followed by removal the device leads to: |====================================================== |[ INFO: possible circular locking dependency detected ] |3.17.0-rc4+ #1498 Not tainted |------------------------------------------------------- |rmmod/2505 is trying to acquire lock: | (s_active#28){++++.+}, at: [<c017f754>] kernfs_remove_by_name_ns+0x3c/0x88 | |but task is already holding lock: | (pools_lock){+.+.+.}, at: [<c011494c>] dma_pool_destroy+0x18/0x17c | |which lock already depends on the new lock. |the existing dependency chain (in reverse order) is: | |-> #1 (pools_lock){+.+.+.}: | [<c0114ae8>] show_pools+0x30/0xf8 | [<c0313210>] dev_attr_show+0x1c/0x48 | [<c0180e84>] sysfs_kf_seq_show+0x88/0x10c | [<c017f960>] kernfs_seq_show+0x24/0x28 | [<c013efc4>] seq_read+0x1b8/0x480 | [<c011e820>] vfs_read+0x8c/0x148 | [<c011ea10>] SyS_read+0x40/0x8c | [<c000e960>] ret_fast_syscall+0x0/0x48 | |-> #0 (s_active#28){++++.+}: | [<c017e9ac>] __kernfs_remove+0x258/0x2ec | [<c017f754>] kernfs_remove_by_name_ns+0x3c/0x88 | [<c0114a7c>] dma_pool_destroy+0x148/0x17c | [<c03ad288>] hcd_buffer_destroy+0x20/0x34 | [<c03a4780>] usb_remove_hcd+0x110/0x1a4 The problem is the lock order of pools_lock and kernfs_mutex in dma_pool_destroy() vs show_pools() call path. This patch breaks out the creation of the sysfs file outside of the pools_lock mutex. The newly added pools_reg_lock ensures that there is no race of create vs destroy code path in terms whether or not the sysfs file has to be deleted (and was it deleted before we try to create a new one) and what to do if device_create_file() failed. Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2014-10-10 06:28:50 +08:00
if (empty)
device_remove_file(pool->dev, &dev_attr_pools);
mutex_unlock(&pools_reg_lock);
dmapool: link blocks across pages The allocated dmapool pages are never freed for the lifetime of the pool. There is no need for the two level list+stack lookup for finding a free block since nothing is ever removed from the list. Just use a simple stack, reducing time complexity to constant. The implementation inserts the stack linking elements and the dma handle of the block within itself when freed. This means the smallest possible dmapool block is increased to at most 16 bytes to accommodate these fields, but there are no exisiting users requesting a dma pool smaller than that anyway. Removing the list has a significant change in performance. Using the kernel's micro-benchmarking self test: Before: # modprobe dmapool_test dmapool test: size:16 blocks:8192 time:57282 dmapool test: size:64 blocks:8192 time:172562 dmapool test: size:256 blocks:8192 time:789247 dmapool test: size:1024 blocks:2048 time:371823 dmapool test: size:4096 blocks:1024 time:362237 After: # modprobe dmapool_test dmapool test: size:16 blocks:8192 time:24997 dmapool test: size:64 blocks:8192 time:26584 dmapool test: size:256 blocks:8192 time:33542 dmapool test: size:1024 blocks:2048 time:9022 dmapool test: size:4096 blocks:1024 time:6045 The module test allocates quite a few blocks that may not accurately represent how these pools are used in real life. For a more marco level benchmark, running fio high-depth + high-batched on nvme, this patch shows submission and completion latency reduced by ~100usec each, 1% IOPs improvement, and perf record's time spent in dma_pool_alloc/free were reduced by half. [kbusch@kernel.org: push new blocks in ascending order] Link: https://lkml.kernel.org/r/20230221165400.1595247-1-kbusch@meta.com Link: https://lkml.kernel.org/r/20230126215125.4069751-12-kbusch@meta.com Fixes: 2d55c16c0c54 ("dmapool: create/destroy cleanup") Signed-off-by: Keith Busch <kbusch@kernel.org> Reviewed-by: Christoph Hellwig <hch@lst.de> Tested-by: Bryan O'Donoghue <bryan.odonoghue@linaro.org> Cc: Matthew Wilcox <willy@infradead.org> Cc: Tony Battersby <tonyb@cybernetics.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2023-01-27 05:51:24 +08:00
if (pool->nr_active) {
dev_err(pool->dev, "%s %s busy\n", __func__, pool->name);
busy = true;
}
list_for_each_entry_safe(page, tmp, &pool->page_list, page_list) {
dmapool: link blocks across pages The allocated dmapool pages are never freed for the lifetime of the pool. There is no need for the two level list+stack lookup for finding a free block since nothing is ever removed from the list. Just use a simple stack, reducing time complexity to constant. The implementation inserts the stack linking elements and the dma handle of the block within itself when freed. This means the smallest possible dmapool block is increased to at most 16 bytes to accommodate these fields, but there are no exisiting users requesting a dma pool smaller than that anyway. Removing the list has a significant change in performance. Using the kernel's micro-benchmarking self test: Before: # modprobe dmapool_test dmapool test: size:16 blocks:8192 time:57282 dmapool test: size:64 blocks:8192 time:172562 dmapool test: size:256 blocks:8192 time:789247 dmapool test: size:1024 blocks:2048 time:371823 dmapool test: size:4096 blocks:1024 time:362237 After: # modprobe dmapool_test dmapool test: size:16 blocks:8192 time:24997 dmapool test: size:64 blocks:8192 time:26584 dmapool test: size:256 blocks:8192 time:33542 dmapool test: size:1024 blocks:2048 time:9022 dmapool test: size:4096 blocks:1024 time:6045 The module test allocates quite a few blocks that may not accurately represent how these pools are used in real life. For a more marco level benchmark, running fio high-depth + high-batched on nvme, this patch shows submission and completion latency reduced by ~100usec each, 1% IOPs improvement, and perf record's time spent in dma_pool_alloc/free were reduced by half. [kbusch@kernel.org: push new blocks in ascending order] Link: https://lkml.kernel.org/r/20230221165400.1595247-1-kbusch@meta.com Link: https://lkml.kernel.org/r/20230126215125.4069751-12-kbusch@meta.com Fixes: 2d55c16c0c54 ("dmapool: create/destroy cleanup") Signed-off-by: Keith Busch <kbusch@kernel.org> Reviewed-by: Christoph Hellwig <hch@lst.de> Tested-by: Bryan O'Donoghue <bryan.odonoghue@linaro.org> Cc: Matthew Wilcox <willy@infradead.org> Cc: Tony Battersby <tonyb@cybernetics.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2023-01-27 05:51:24 +08:00
if (!busy)
dma_free_coherent(pool->dev, pool->allocation,
page->vaddr, page->dma);
list_del(&page->page_list);
kfree(page);
}
kfree(pool);
}
EXPORT_SYMBOL(dma_pool_destroy);
/**
* dma_pool_alloc - get a block of consistent memory
* @pool: dma pool that will produce the block
* @mem_flags: GFP_* bitmask
* @handle: pointer to dma address of block
*
* Return: the kernel virtual address of a currently unused block,
* and reports its dma address through the handle.
* If such a memory block can't be allocated, %NULL is returned.
*/
void *dma_pool_alloc(struct dma_pool *pool, gfp_t mem_flags,
dma_addr_t *handle)
{
dmapool: link blocks across pages The allocated dmapool pages are never freed for the lifetime of the pool. There is no need for the two level list+stack lookup for finding a free block since nothing is ever removed from the list. Just use a simple stack, reducing time complexity to constant. The implementation inserts the stack linking elements and the dma handle of the block within itself when freed. This means the smallest possible dmapool block is increased to at most 16 bytes to accommodate these fields, but there are no exisiting users requesting a dma pool smaller than that anyway. Removing the list has a significant change in performance. Using the kernel's micro-benchmarking self test: Before: # modprobe dmapool_test dmapool test: size:16 blocks:8192 time:57282 dmapool test: size:64 blocks:8192 time:172562 dmapool test: size:256 blocks:8192 time:789247 dmapool test: size:1024 blocks:2048 time:371823 dmapool test: size:4096 blocks:1024 time:362237 After: # modprobe dmapool_test dmapool test: size:16 blocks:8192 time:24997 dmapool test: size:64 blocks:8192 time:26584 dmapool test: size:256 blocks:8192 time:33542 dmapool test: size:1024 blocks:2048 time:9022 dmapool test: size:4096 blocks:1024 time:6045 The module test allocates quite a few blocks that may not accurately represent how these pools are used in real life. For a more marco level benchmark, running fio high-depth + high-batched on nvme, this patch shows submission and completion latency reduced by ~100usec each, 1% IOPs improvement, and perf record's time spent in dma_pool_alloc/free were reduced by half. [kbusch@kernel.org: push new blocks in ascending order] Link: https://lkml.kernel.org/r/20230221165400.1595247-1-kbusch@meta.com Link: https://lkml.kernel.org/r/20230126215125.4069751-12-kbusch@meta.com Fixes: 2d55c16c0c54 ("dmapool: create/destroy cleanup") Signed-off-by: Keith Busch <kbusch@kernel.org> Reviewed-by: Christoph Hellwig <hch@lst.de> Tested-by: Bryan O'Donoghue <bryan.odonoghue@linaro.org> Cc: Matthew Wilcox <willy@infradead.org> Cc: Tony Battersby <tonyb@cybernetics.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2023-01-27 05:51:24 +08:00
struct dma_block *block;
struct dma_page *page;
dmapool: link blocks across pages The allocated dmapool pages are never freed for the lifetime of the pool. There is no need for the two level list+stack lookup for finding a free block since nothing is ever removed from the list. Just use a simple stack, reducing time complexity to constant. The implementation inserts the stack linking elements and the dma handle of the block within itself when freed. This means the smallest possible dmapool block is increased to at most 16 bytes to accommodate these fields, but there are no exisiting users requesting a dma pool smaller than that anyway. Removing the list has a significant change in performance. Using the kernel's micro-benchmarking self test: Before: # modprobe dmapool_test dmapool test: size:16 blocks:8192 time:57282 dmapool test: size:64 blocks:8192 time:172562 dmapool test: size:256 blocks:8192 time:789247 dmapool test: size:1024 blocks:2048 time:371823 dmapool test: size:4096 blocks:1024 time:362237 After: # modprobe dmapool_test dmapool test: size:16 blocks:8192 time:24997 dmapool test: size:64 blocks:8192 time:26584 dmapool test: size:256 blocks:8192 time:33542 dmapool test: size:1024 blocks:2048 time:9022 dmapool test: size:4096 blocks:1024 time:6045 The module test allocates quite a few blocks that may not accurately represent how these pools are used in real life. For a more marco level benchmark, running fio high-depth + high-batched on nvme, this patch shows submission and completion latency reduced by ~100usec each, 1% IOPs improvement, and perf record's time spent in dma_pool_alloc/free were reduced by half. [kbusch@kernel.org: push new blocks in ascending order] Link: https://lkml.kernel.org/r/20230221165400.1595247-1-kbusch@meta.com Link: https://lkml.kernel.org/r/20230126215125.4069751-12-kbusch@meta.com Fixes: 2d55c16c0c54 ("dmapool: create/destroy cleanup") Signed-off-by: Keith Busch <kbusch@kernel.org> Reviewed-by: Christoph Hellwig <hch@lst.de> Tested-by: Bryan O'Donoghue <bryan.odonoghue@linaro.org> Cc: Matthew Wilcox <willy@infradead.org> Cc: Tony Battersby <tonyb@cybernetics.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2023-01-27 05:51:24 +08:00
unsigned long flags;
might_alloc(mem_flags);
spin_lock_irqsave(&pool->lock, flags);
dmapool: link blocks across pages The allocated dmapool pages are never freed for the lifetime of the pool. There is no need for the two level list+stack lookup for finding a free block since nothing is ever removed from the list. Just use a simple stack, reducing time complexity to constant. The implementation inserts the stack linking elements and the dma handle of the block within itself when freed. This means the smallest possible dmapool block is increased to at most 16 bytes to accommodate these fields, but there are no exisiting users requesting a dma pool smaller than that anyway. Removing the list has a significant change in performance. Using the kernel's micro-benchmarking self test: Before: # modprobe dmapool_test dmapool test: size:16 blocks:8192 time:57282 dmapool test: size:64 blocks:8192 time:172562 dmapool test: size:256 blocks:8192 time:789247 dmapool test: size:1024 blocks:2048 time:371823 dmapool test: size:4096 blocks:1024 time:362237 After: # modprobe dmapool_test dmapool test: size:16 blocks:8192 time:24997 dmapool test: size:64 blocks:8192 time:26584 dmapool test: size:256 blocks:8192 time:33542 dmapool test: size:1024 blocks:2048 time:9022 dmapool test: size:4096 blocks:1024 time:6045 The module test allocates quite a few blocks that may not accurately represent how these pools are used in real life. For a more marco level benchmark, running fio high-depth + high-batched on nvme, this patch shows submission and completion latency reduced by ~100usec each, 1% IOPs improvement, and perf record's time spent in dma_pool_alloc/free were reduced by half. [kbusch@kernel.org: push new blocks in ascending order] Link: https://lkml.kernel.org/r/20230221165400.1595247-1-kbusch@meta.com Link: https://lkml.kernel.org/r/20230126215125.4069751-12-kbusch@meta.com Fixes: 2d55c16c0c54 ("dmapool: create/destroy cleanup") Signed-off-by: Keith Busch <kbusch@kernel.org> Reviewed-by: Christoph Hellwig <hch@lst.de> Tested-by: Bryan O'Donoghue <bryan.odonoghue@linaro.org> Cc: Matthew Wilcox <willy@infradead.org> Cc: Tony Battersby <tonyb@cybernetics.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2023-01-27 05:51:24 +08:00
block = pool_block_pop(pool);
if (!block) {
/*
* pool_alloc_page() might sleep, so temporarily drop
* &pool->lock
*/
spin_unlock_irqrestore(&pool->lock, flags);
dmapool: link blocks across pages The allocated dmapool pages are never freed for the lifetime of the pool. There is no need for the two level list+stack lookup for finding a free block since nothing is ever removed from the list. Just use a simple stack, reducing time complexity to constant. The implementation inserts the stack linking elements and the dma handle of the block within itself when freed. This means the smallest possible dmapool block is increased to at most 16 bytes to accommodate these fields, but there are no exisiting users requesting a dma pool smaller than that anyway. Removing the list has a significant change in performance. Using the kernel's micro-benchmarking self test: Before: # modprobe dmapool_test dmapool test: size:16 blocks:8192 time:57282 dmapool test: size:64 blocks:8192 time:172562 dmapool test: size:256 blocks:8192 time:789247 dmapool test: size:1024 blocks:2048 time:371823 dmapool test: size:4096 blocks:1024 time:362237 After: # modprobe dmapool_test dmapool test: size:16 blocks:8192 time:24997 dmapool test: size:64 blocks:8192 time:26584 dmapool test: size:256 blocks:8192 time:33542 dmapool test: size:1024 blocks:2048 time:9022 dmapool test: size:4096 blocks:1024 time:6045 The module test allocates quite a few blocks that may not accurately represent how these pools are used in real life. For a more marco level benchmark, running fio high-depth + high-batched on nvme, this patch shows submission and completion latency reduced by ~100usec each, 1% IOPs improvement, and perf record's time spent in dma_pool_alloc/free were reduced by half. [kbusch@kernel.org: push new blocks in ascending order] Link: https://lkml.kernel.org/r/20230221165400.1595247-1-kbusch@meta.com Link: https://lkml.kernel.org/r/20230126215125.4069751-12-kbusch@meta.com Fixes: 2d55c16c0c54 ("dmapool: create/destroy cleanup") Signed-off-by: Keith Busch <kbusch@kernel.org> Reviewed-by: Christoph Hellwig <hch@lst.de> Tested-by: Bryan O'Donoghue <bryan.odonoghue@linaro.org> Cc: Matthew Wilcox <willy@infradead.org> Cc: Tony Battersby <tonyb@cybernetics.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2023-01-27 05:51:24 +08:00
page = pool_alloc_page(pool, mem_flags & (~__GFP_ZERO));
if (!page)
return NULL;
dmapool: link blocks across pages The allocated dmapool pages are never freed for the lifetime of the pool. There is no need for the two level list+stack lookup for finding a free block since nothing is ever removed from the list. Just use a simple stack, reducing time complexity to constant. The implementation inserts the stack linking elements and the dma handle of the block within itself when freed. This means the smallest possible dmapool block is increased to at most 16 bytes to accommodate these fields, but there are no exisiting users requesting a dma pool smaller than that anyway. Removing the list has a significant change in performance. Using the kernel's micro-benchmarking self test: Before: # modprobe dmapool_test dmapool test: size:16 blocks:8192 time:57282 dmapool test: size:64 blocks:8192 time:172562 dmapool test: size:256 blocks:8192 time:789247 dmapool test: size:1024 blocks:2048 time:371823 dmapool test: size:4096 blocks:1024 time:362237 After: # modprobe dmapool_test dmapool test: size:16 blocks:8192 time:24997 dmapool test: size:64 blocks:8192 time:26584 dmapool test: size:256 blocks:8192 time:33542 dmapool test: size:1024 blocks:2048 time:9022 dmapool test: size:4096 blocks:1024 time:6045 The module test allocates quite a few blocks that may not accurately represent how these pools are used in real life. For a more marco level benchmark, running fio high-depth + high-batched on nvme, this patch shows submission and completion latency reduced by ~100usec each, 1% IOPs improvement, and perf record's time spent in dma_pool_alloc/free were reduced by half. [kbusch@kernel.org: push new blocks in ascending order] Link: https://lkml.kernel.org/r/20230221165400.1595247-1-kbusch@meta.com Link: https://lkml.kernel.org/r/20230126215125.4069751-12-kbusch@meta.com Fixes: 2d55c16c0c54 ("dmapool: create/destroy cleanup") Signed-off-by: Keith Busch <kbusch@kernel.org> Reviewed-by: Christoph Hellwig <hch@lst.de> Tested-by: Bryan O'Donoghue <bryan.odonoghue@linaro.org> Cc: Matthew Wilcox <willy@infradead.org> Cc: Tony Battersby <tonyb@cybernetics.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2023-01-27 05:51:24 +08:00
spin_lock_irqsave(&pool->lock, flags);
pool_initialise_page(pool, page);
block = pool_block_pop(pool);
}
spin_unlock_irqrestore(&pool->lock, flags);
mm: add support for __GFP_ZERO flag to dma_pool_alloc() Currently a call to dma_pool_alloc() with a ___GFP_ZERO flag returns a non-zeroed memory region. This patchset adds support for the __GFP_ZERO flag to dma_pool_alloc(), adds 2 wrapper functions for allocing zeroed memory from a pool, and provides a coccinelle script for finding & replacing instances of dma_pool_alloc() followed by memset(0) with a single dma_pool_zalloc() call. There was some concern that this always calls memset() to zero, instead of passing __GFP_ZERO into the page allocator. [https://lkml.org/lkml/2015/7/15/881] I ran a test on my system to get an idea of how often dma_pool_alloc() calls into pool_alloc_page(). After Boot: [ 30.119863] alloc_calls:541, page_allocs:7 After an hour: [ 3600.951031] alloc_calls:9566, page_allocs:12 After copying 1GB file onto a USB drive: [ 4260.657148] alloc_calls:17225, page_allocs:12 It doesn't look like dma_pool_alloc() calls down to the page allocator very often (at least on my system). This patch (of 4): Currently the __GFP_ZERO flag is ignored by dma_pool_alloc(). Make dma_pool_alloc() zero the memory if this flag is set. Signed-off-by: Sean O. Stalley <sean.stalley@intel.com> Acked-by: David Rientjes <rientjes@google.com> Cc: Vinod Koul <vinod.koul@intel.com> Cc: Bjorn Helgaas <bhelgaas@google.com> Cc: Gilles Muller <Gilles.Muller@lip6.fr> Cc: Nicolas Palix <nicolas.palix@imag.fr> Cc: Michal Marek <mmarek@suse.cz> Cc: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Cc: Jonathan Corbet <corbet@lwn.net> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2015-09-09 06:02:24 +08:00
dmapool: link blocks across pages The allocated dmapool pages are never freed for the lifetime of the pool. There is no need for the two level list+stack lookup for finding a free block since nothing is ever removed from the list. Just use a simple stack, reducing time complexity to constant. The implementation inserts the stack linking elements and the dma handle of the block within itself when freed. This means the smallest possible dmapool block is increased to at most 16 bytes to accommodate these fields, but there are no exisiting users requesting a dma pool smaller than that anyway. Removing the list has a significant change in performance. Using the kernel's micro-benchmarking self test: Before: # modprobe dmapool_test dmapool test: size:16 blocks:8192 time:57282 dmapool test: size:64 blocks:8192 time:172562 dmapool test: size:256 blocks:8192 time:789247 dmapool test: size:1024 blocks:2048 time:371823 dmapool test: size:4096 blocks:1024 time:362237 After: # modprobe dmapool_test dmapool test: size:16 blocks:8192 time:24997 dmapool test: size:64 blocks:8192 time:26584 dmapool test: size:256 blocks:8192 time:33542 dmapool test: size:1024 blocks:2048 time:9022 dmapool test: size:4096 blocks:1024 time:6045 The module test allocates quite a few blocks that may not accurately represent how these pools are used in real life. For a more marco level benchmark, running fio high-depth + high-batched on nvme, this patch shows submission and completion latency reduced by ~100usec each, 1% IOPs improvement, and perf record's time spent in dma_pool_alloc/free were reduced by half. [kbusch@kernel.org: push new blocks in ascending order] Link: https://lkml.kernel.org/r/20230221165400.1595247-1-kbusch@meta.com Link: https://lkml.kernel.org/r/20230126215125.4069751-12-kbusch@meta.com Fixes: 2d55c16c0c54 ("dmapool: create/destroy cleanup") Signed-off-by: Keith Busch <kbusch@kernel.org> Reviewed-by: Christoph Hellwig <hch@lst.de> Tested-by: Bryan O'Donoghue <bryan.odonoghue@linaro.org> Cc: Matthew Wilcox <willy@infradead.org> Cc: Tony Battersby <tonyb@cybernetics.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2023-01-27 05:51:24 +08:00
*handle = block->dma;
pool_check_block(pool, block, mem_flags);
mm: security: introduce init_on_alloc=1 and init_on_free=1 boot options Patch series "add init_on_alloc/init_on_free boot options", v10. Provide init_on_alloc and init_on_free boot options. These are aimed at preventing possible information leaks and making the control-flow bugs that depend on uninitialized values more deterministic. Enabling either of the options guarantees that the memory returned by the page allocator and SL[AU]B is initialized with zeroes. SLOB allocator isn't supported at the moment, as its emulation of kmem caches complicates handling of SLAB_TYPESAFE_BY_RCU caches correctly. Enabling init_on_free also guarantees that pages and heap objects are initialized right after they're freed, so it won't be possible to access stale data by using a dangling pointer. As suggested by Michal Hocko, right now we don't let the heap users to disable initialization for certain allocations. There's not enough evidence that doing so can speed up real-life cases, and introducing ways to opt-out may result in things going out of control. This patch (of 2): The new options are needed to prevent possible information leaks and make control-flow bugs that depend on uninitialized values more deterministic. This is expected to be on-by-default on Android and Chrome OS. And it gives the opportunity for anyone else to use it under distros too via the boot args. (The init_on_free feature is regularly requested by folks where memory forensics is included in their threat models.) init_on_alloc=1 makes the kernel initialize newly allocated pages and heap objects with zeroes. Initialization is done at allocation time at the places where checks for __GFP_ZERO are performed. init_on_free=1 makes the kernel initialize freed pages and heap objects with zeroes upon their deletion. This helps to ensure sensitive data doesn't leak via use-after-free accesses. Both init_on_alloc=1 and init_on_free=1 guarantee that the allocator returns zeroed memory. The two exceptions are slab caches with constructors and SLAB_TYPESAFE_BY_RCU flag. Those are never zero-initialized to preserve their semantics. Both init_on_alloc and init_on_free default to zero, but those defaults can be overridden with CONFIG_INIT_ON_ALLOC_DEFAULT_ON and CONFIG_INIT_ON_FREE_DEFAULT_ON. If either SLUB poisoning or page poisoning is enabled, those options take precedence over init_on_alloc and init_on_free: initialization is only applied to unpoisoned allocations. Slowdown for the new features compared to init_on_free=0, init_on_alloc=0: hackbench, init_on_free=1: +7.62% sys time (st.err 0.74%) hackbench, init_on_alloc=1: +7.75% sys time (st.err 2.14%) Linux build with -j12, init_on_free=1: +8.38% wall time (st.err 0.39%) Linux build with -j12, init_on_free=1: +24.42% sys time (st.err 0.52%) Linux build with -j12, init_on_alloc=1: -0.13% wall time (st.err 0.42%) Linux build with -j12, init_on_alloc=1: +0.57% sys time (st.err 0.40%) The slowdown for init_on_free=0, init_on_alloc=0 compared to the baseline is within the standard error. The new features are also going to pave the way for hardware memory tagging (e.g. arm64's MTE), which will require both on_alloc and on_free hooks to set the tags for heap objects. With MTE, tagging will have the same cost as memory initialization. Although init_on_free is rather costly, there are paranoid use-cases where in-memory data lifetime is desired to be minimized. There are various arguments for/against the realism of the associated threat models, but given that we'll need the infrastructure for MTE anyway, and there are people who want wipe-on-free behavior no matter what the performance cost, it seems reasonable to include it in this series. [glider@google.com: v8] Link: http://lkml.kernel.org/r/20190626121943.131390-2-glider@google.com [glider@google.com: v9] Link: http://lkml.kernel.org/r/20190627130316.254309-2-glider@google.com [glider@google.com: v10] Link: http://lkml.kernel.org/r/20190628093131.199499-2-glider@google.com Link: http://lkml.kernel.org/r/20190617151050.92663-2-glider@google.com Signed-off-by: Alexander Potapenko <glider@google.com> Acked-by: Kees Cook <keescook@chromium.org> Acked-by: Michal Hocko <mhocko@suse.cz> [page and dmapool parts Acked-by: James Morris <jamorris@linux.microsoft.com>] Cc: Christoph Lameter <cl@linux.com> Cc: Masahiro Yamada <yamada.masahiro@socionext.com> Cc: "Serge E. Hallyn" <serge@hallyn.com> Cc: Nick Desaulniers <ndesaulniers@google.com> Cc: Kostya Serebryany <kcc@google.com> Cc: Dmitry Vyukov <dvyukov@google.com> Cc: Sandeep Patil <sspatil@android.com> Cc: Laura Abbott <labbott@redhat.com> Cc: Randy Dunlap <rdunlap@infradead.org> Cc: Jann Horn <jannh@google.com> Cc: Mark Rutland <mark.rutland@arm.com> Cc: Marco Elver <elver@google.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2019-07-12 11:59:19 +08:00
if (want_init_on_alloc(mem_flags))
dmapool: link blocks across pages The allocated dmapool pages are never freed for the lifetime of the pool. There is no need for the two level list+stack lookup for finding a free block since nothing is ever removed from the list. Just use a simple stack, reducing time complexity to constant. The implementation inserts the stack linking elements and the dma handle of the block within itself when freed. This means the smallest possible dmapool block is increased to at most 16 bytes to accommodate these fields, but there are no exisiting users requesting a dma pool smaller than that anyway. Removing the list has a significant change in performance. Using the kernel's micro-benchmarking self test: Before: # modprobe dmapool_test dmapool test: size:16 blocks:8192 time:57282 dmapool test: size:64 blocks:8192 time:172562 dmapool test: size:256 blocks:8192 time:789247 dmapool test: size:1024 blocks:2048 time:371823 dmapool test: size:4096 blocks:1024 time:362237 After: # modprobe dmapool_test dmapool test: size:16 blocks:8192 time:24997 dmapool test: size:64 blocks:8192 time:26584 dmapool test: size:256 blocks:8192 time:33542 dmapool test: size:1024 blocks:2048 time:9022 dmapool test: size:4096 blocks:1024 time:6045 The module test allocates quite a few blocks that may not accurately represent how these pools are used in real life. For a more marco level benchmark, running fio high-depth + high-batched on nvme, this patch shows submission and completion latency reduced by ~100usec each, 1% IOPs improvement, and perf record's time spent in dma_pool_alloc/free were reduced by half. [kbusch@kernel.org: push new blocks in ascending order] Link: https://lkml.kernel.org/r/20230221165400.1595247-1-kbusch@meta.com Link: https://lkml.kernel.org/r/20230126215125.4069751-12-kbusch@meta.com Fixes: 2d55c16c0c54 ("dmapool: create/destroy cleanup") Signed-off-by: Keith Busch <kbusch@kernel.org> Reviewed-by: Christoph Hellwig <hch@lst.de> Tested-by: Bryan O'Donoghue <bryan.odonoghue@linaro.org> Cc: Matthew Wilcox <willy@infradead.org> Cc: Tony Battersby <tonyb@cybernetics.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2023-01-27 05:51:24 +08:00
memset(block, 0, pool->size);
mm: add support for __GFP_ZERO flag to dma_pool_alloc() Currently a call to dma_pool_alloc() with a ___GFP_ZERO flag returns a non-zeroed memory region. This patchset adds support for the __GFP_ZERO flag to dma_pool_alloc(), adds 2 wrapper functions for allocing zeroed memory from a pool, and provides a coccinelle script for finding & replacing instances of dma_pool_alloc() followed by memset(0) with a single dma_pool_zalloc() call. There was some concern that this always calls memset() to zero, instead of passing __GFP_ZERO into the page allocator. [https://lkml.org/lkml/2015/7/15/881] I ran a test on my system to get an idea of how often dma_pool_alloc() calls into pool_alloc_page(). After Boot: [ 30.119863] alloc_calls:541, page_allocs:7 After an hour: [ 3600.951031] alloc_calls:9566, page_allocs:12 After copying 1GB file onto a USB drive: [ 4260.657148] alloc_calls:17225, page_allocs:12 It doesn't look like dma_pool_alloc() calls down to the page allocator very often (at least on my system). This patch (of 4): Currently the __GFP_ZERO flag is ignored by dma_pool_alloc(). Make dma_pool_alloc() zero the memory if this flag is set. Signed-off-by: Sean O. Stalley <sean.stalley@intel.com> Acked-by: David Rientjes <rientjes@google.com> Cc: Vinod Koul <vinod.koul@intel.com> Cc: Bjorn Helgaas <bhelgaas@google.com> Cc: Gilles Muller <Gilles.Muller@lip6.fr> Cc: Nicolas Palix <nicolas.palix@imag.fr> Cc: Michal Marek <mmarek@suse.cz> Cc: Sebastian Andrzej Siewior <bigeasy@linutronix.de> Cc: Jonathan Corbet <corbet@lwn.net> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2015-09-09 06:02:24 +08:00
dmapool: link blocks across pages The allocated dmapool pages are never freed for the lifetime of the pool. There is no need for the two level list+stack lookup for finding a free block since nothing is ever removed from the list. Just use a simple stack, reducing time complexity to constant. The implementation inserts the stack linking elements and the dma handle of the block within itself when freed. This means the smallest possible dmapool block is increased to at most 16 bytes to accommodate these fields, but there are no exisiting users requesting a dma pool smaller than that anyway. Removing the list has a significant change in performance. Using the kernel's micro-benchmarking self test: Before: # modprobe dmapool_test dmapool test: size:16 blocks:8192 time:57282 dmapool test: size:64 blocks:8192 time:172562 dmapool test: size:256 blocks:8192 time:789247 dmapool test: size:1024 blocks:2048 time:371823 dmapool test: size:4096 blocks:1024 time:362237 After: # modprobe dmapool_test dmapool test: size:16 blocks:8192 time:24997 dmapool test: size:64 blocks:8192 time:26584 dmapool test: size:256 blocks:8192 time:33542 dmapool test: size:1024 blocks:2048 time:9022 dmapool test: size:4096 blocks:1024 time:6045 The module test allocates quite a few blocks that may not accurately represent how these pools are used in real life. For a more marco level benchmark, running fio high-depth + high-batched on nvme, this patch shows submission and completion latency reduced by ~100usec each, 1% IOPs improvement, and perf record's time spent in dma_pool_alloc/free were reduced by half. [kbusch@kernel.org: push new blocks in ascending order] Link: https://lkml.kernel.org/r/20230221165400.1595247-1-kbusch@meta.com Link: https://lkml.kernel.org/r/20230126215125.4069751-12-kbusch@meta.com Fixes: 2d55c16c0c54 ("dmapool: create/destroy cleanup") Signed-off-by: Keith Busch <kbusch@kernel.org> Reviewed-by: Christoph Hellwig <hch@lst.de> Tested-by: Bryan O'Donoghue <bryan.odonoghue@linaro.org> Cc: Matthew Wilcox <willy@infradead.org> Cc: Tony Battersby <tonyb@cybernetics.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2023-01-27 05:51:24 +08:00
return block;
}
EXPORT_SYMBOL(dma_pool_alloc);
/**
* dma_pool_free - put block back into dma pool
* @pool: the dma pool holding the block
* @vaddr: virtual address of block
* @dma: dma address of block
*
* Caller promises neither device nor driver will again touch this block
* unless it is first re-allocated.
*/
void dma_pool_free(struct dma_pool *pool, void *vaddr, dma_addr_t dma)
{
dmapool: link blocks across pages The allocated dmapool pages are never freed for the lifetime of the pool. There is no need for the two level list+stack lookup for finding a free block since nothing is ever removed from the list. Just use a simple stack, reducing time complexity to constant. The implementation inserts the stack linking elements and the dma handle of the block within itself when freed. This means the smallest possible dmapool block is increased to at most 16 bytes to accommodate these fields, but there are no exisiting users requesting a dma pool smaller than that anyway. Removing the list has a significant change in performance. Using the kernel's micro-benchmarking self test: Before: # modprobe dmapool_test dmapool test: size:16 blocks:8192 time:57282 dmapool test: size:64 blocks:8192 time:172562 dmapool test: size:256 blocks:8192 time:789247 dmapool test: size:1024 blocks:2048 time:371823 dmapool test: size:4096 blocks:1024 time:362237 After: # modprobe dmapool_test dmapool test: size:16 blocks:8192 time:24997 dmapool test: size:64 blocks:8192 time:26584 dmapool test: size:256 blocks:8192 time:33542 dmapool test: size:1024 blocks:2048 time:9022 dmapool test: size:4096 blocks:1024 time:6045 The module test allocates quite a few blocks that may not accurately represent how these pools are used in real life. For a more marco level benchmark, running fio high-depth + high-batched on nvme, this patch shows submission and completion latency reduced by ~100usec each, 1% IOPs improvement, and perf record's time spent in dma_pool_alloc/free were reduced by half. [kbusch@kernel.org: push new blocks in ascending order] Link: https://lkml.kernel.org/r/20230221165400.1595247-1-kbusch@meta.com Link: https://lkml.kernel.org/r/20230126215125.4069751-12-kbusch@meta.com Fixes: 2d55c16c0c54 ("dmapool: create/destroy cleanup") Signed-off-by: Keith Busch <kbusch@kernel.org> Reviewed-by: Christoph Hellwig <hch@lst.de> Tested-by: Bryan O'Donoghue <bryan.odonoghue@linaro.org> Cc: Matthew Wilcox <willy@infradead.org> Cc: Tony Battersby <tonyb@cybernetics.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2023-01-27 05:51:24 +08:00
struct dma_block *block = vaddr;
unsigned long flags;
spin_lock_irqsave(&pool->lock, flags);
dmapool: link blocks across pages The allocated dmapool pages are never freed for the lifetime of the pool. There is no need for the two level list+stack lookup for finding a free block since nothing is ever removed from the list. Just use a simple stack, reducing time complexity to constant. The implementation inserts the stack linking elements and the dma handle of the block within itself when freed. This means the smallest possible dmapool block is increased to at most 16 bytes to accommodate these fields, but there are no exisiting users requesting a dma pool smaller than that anyway. Removing the list has a significant change in performance. Using the kernel's micro-benchmarking self test: Before: # modprobe dmapool_test dmapool test: size:16 blocks:8192 time:57282 dmapool test: size:64 blocks:8192 time:172562 dmapool test: size:256 blocks:8192 time:789247 dmapool test: size:1024 blocks:2048 time:371823 dmapool test: size:4096 blocks:1024 time:362237 After: # modprobe dmapool_test dmapool test: size:16 blocks:8192 time:24997 dmapool test: size:64 blocks:8192 time:26584 dmapool test: size:256 blocks:8192 time:33542 dmapool test: size:1024 blocks:2048 time:9022 dmapool test: size:4096 blocks:1024 time:6045 The module test allocates quite a few blocks that may not accurately represent how these pools are used in real life. For a more marco level benchmark, running fio high-depth + high-batched on nvme, this patch shows submission and completion latency reduced by ~100usec each, 1% IOPs improvement, and perf record's time spent in dma_pool_alloc/free were reduced by half. [kbusch@kernel.org: push new blocks in ascending order] Link: https://lkml.kernel.org/r/20230221165400.1595247-1-kbusch@meta.com Link: https://lkml.kernel.org/r/20230126215125.4069751-12-kbusch@meta.com Fixes: 2d55c16c0c54 ("dmapool: create/destroy cleanup") Signed-off-by: Keith Busch <kbusch@kernel.org> Reviewed-by: Christoph Hellwig <hch@lst.de> Tested-by: Bryan O'Donoghue <bryan.odonoghue@linaro.org> Cc: Matthew Wilcox <willy@infradead.org> Cc: Tony Battersby <tonyb@cybernetics.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2023-01-27 05:51:24 +08:00
if (!pool_block_err(pool, vaddr, dma)) {
pool_block_push(pool, block, dma);
pool->nr_active--;
}
spin_unlock_irqrestore(&pool->lock, flags);
}
EXPORT_SYMBOL(dma_pool_free);
/*
* Managed DMA pool
*/
static void dmam_pool_release(struct device *dev, void *res)
{
struct dma_pool *pool = *(struct dma_pool **)res;
dma_pool_destroy(pool);
}
static int dmam_pool_match(struct device *dev, void *res, void *match_data)
{
return *(struct dma_pool **)res == match_data;
}
/**
* dmam_pool_create - Managed dma_pool_create()
* @name: name of pool, for diagnostics
* @dev: device that will be doing the DMA
* @size: size of the blocks in this pool.
* @align: alignment requirement for blocks; must be a power of two
* @allocation: returned blocks won't cross this boundary (or zero)
*
* Managed dma_pool_create(). DMA pool created with this function is
* automatically destroyed on driver detach.
*
* Return: a managed dma allocation pool with the requested
* characteristics, or %NULL if one can't be created.
*/
struct dma_pool *dmam_pool_create(const char *name, struct device *dev,
size_t size, size_t align, size_t allocation)
{
struct dma_pool **ptr, *pool;
ptr = devres_alloc(dmam_pool_release, sizeof(*ptr), GFP_KERNEL);
if (!ptr)
return NULL;
pool = *ptr = dma_pool_create(name, dev, size, align, allocation);
if (pool)
devres_add(dev, ptr);
else
devres_free(ptr);
return pool;
}
EXPORT_SYMBOL(dmam_pool_create);
/**
* dmam_pool_destroy - Managed dma_pool_destroy()
* @pool: dma pool that will be destroyed
*
* Managed dma_pool_destroy().
*/
void dmam_pool_destroy(struct dma_pool *pool)
{
struct device *dev = pool->dev;
WARN_ON(devres_release(dev, dmam_pool_release, dmam_pool_match, pool));
}
EXPORT_SYMBOL(dmam_pool_destroy);