| .. | .. |
|---|
| 3 | 3 | * Copyright (c) 2000-2005 Silicon Graphics, Inc. |
|---|
| 4 | 4 | * All Rights Reserved. |
|---|
| 5 | 5 | */ |
|---|
| 6 | | -#include <linux/mm.h> |
|---|
| 7 | | -#include <linux/sched/mm.h> |
|---|
| 8 | | -#include <linux/highmem.h> |
|---|
| 9 | | -#include <linux/slab.h> |
|---|
| 10 | | -#include <linux/swap.h> |
|---|
| 11 | | -#include <linux/blkdev.h> |
|---|
| 6 | +#include "xfs.h" |
|---|
| 12 | 7 | #include <linux/backing-dev.h> |
|---|
| 13 | | -#include "kmem.h" |
|---|
| 14 | 8 | #include "xfs_message.h" |
|---|
| 9 | +#include "xfs_trace.h" |
|---|
| 15 | 10 | |
|---|
| 16 | 11 | void * |
|---|
| 17 | 12 | kmem_alloc(size_t size, xfs_km_flags_t flags) |
|---|
| .. | .. |
|---|
| 20 | 15 | gfp_t lflags = kmem_flags_convert(flags); |
|---|
| 21 | 16 | void *ptr; |
|---|
| 22 | 17 | |
|---|
| 18 | + trace_kmem_alloc(size, flags, _RET_IP_); |
|---|
| 19 | + |
|---|
| 23 | 20 | do { |
|---|
| 24 | 21 | ptr = kmalloc(size, lflags); |
|---|
| 25 | | - if (ptr || (flags & (KM_MAYFAIL|KM_NOSLEEP))) |
|---|
| 22 | + if (ptr || (flags & KM_MAYFAIL)) |
|---|
| 26 | 23 | return ptr; |
|---|
| 27 | 24 | if (!(++retries % 100)) |
|---|
| 28 | 25 | xfs_err(NULL, |
|---|
| .. | .. |
|---|
| 33 | 30 | } while (1); |
|---|
| 34 | 31 | } |
|---|
| 35 | 32 | |
|---|
| 36 | | -void * |
|---|
| 37 | | -kmem_alloc_large(size_t size, xfs_km_flags_t flags) |
|---|
| 33 | + |
|---|
| 34 | +/* |
|---|
| 35 | + * __vmalloc() will allocate data pages and auxiliary structures (e.g. |
|---|
| 36 | + * pagetables) with GFP_KERNEL, yet we may be under GFP_NOFS context here. Hence |
|---|
| 37 | + * we need to tell memory reclaim that we are in such a context via |
|---|
| 38 | + * PF_MEMALLOC_NOFS to prevent memory reclaim re-entering the filesystem here |
|---|
| 39 | + * and potentially deadlocking. |
|---|
| 40 | + */ |
|---|
| 41 | +static void * |
|---|
| 42 | +__kmem_vmalloc(size_t size, xfs_km_flags_t flags) |
|---|
| 38 | 43 | { |
|---|
| 39 | 44 | unsigned nofs_flag = 0; |
|---|
| 40 | 45 | void *ptr; |
|---|
| 41 | | - gfp_t lflags; |
|---|
| 46 | + gfp_t lflags = kmem_flags_convert(flags); |
|---|
| 42 | 47 | |
|---|
| 43 | | - ptr = kmem_alloc(size, flags | KM_MAYFAIL); |
|---|
| 44 | | - if (ptr) |
|---|
| 45 | | - return ptr; |
|---|
| 46 | | - |
|---|
| 47 | | - /* |
|---|
| 48 | | - * __vmalloc() will allocate data pages and auxillary structures (e.g. |
|---|
| 49 | | - * pagetables) with GFP_KERNEL, yet we may be under GFP_NOFS context |
|---|
| 50 | | - * here. Hence we need to tell memory reclaim that we are in such a |
|---|
| 51 | | - * context via PF_MEMALLOC_NOFS to prevent memory reclaim re-entering |
|---|
| 52 | | - * the filesystem here and potentially deadlocking. |
|---|
| 53 | | - */ |
|---|
| 54 | 48 | if (flags & KM_NOFS) |
|---|
| 55 | 49 | nofs_flag = memalloc_nofs_save(); |
|---|
| 56 | 50 | |
|---|
| 57 | | - lflags = kmem_flags_convert(flags); |
|---|
| 58 | | - ptr = __vmalloc(size, lflags, PAGE_KERNEL); |
|---|
| 51 | + ptr = __vmalloc(size, lflags); |
|---|
| 59 | 52 | |
|---|
| 60 | 53 | if (flags & KM_NOFS) |
|---|
| 61 | 54 | memalloc_nofs_restore(nofs_flag); |
|---|
| .. | .. |
|---|
| 63 | 56 | return ptr; |
|---|
| 64 | 57 | } |
|---|
| 65 | 58 | |
|---|
| 59 | +/* |
|---|
| 60 | + * Same as kmem_alloc_large, except we guarantee the buffer returned is aligned |
|---|
| 61 | + * to the @align_mask. We only guarantee alignment up to page size, we'll clamp |
|---|
| 62 | + * alignment at page size if it is larger. vmalloc always returns a PAGE_SIZE |
|---|
| 63 | + * aligned region. |
|---|
| 64 | + */ |
|---|
| 66 | 65 | void * |
|---|
| 67 | | -kmem_realloc(const void *old, size_t newsize, xfs_km_flags_t flags) |
|---|
| 66 | +kmem_alloc_io(size_t size, int align_mask, xfs_km_flags_t flags) |
|---|
| 68 | 67 | { |
|---|
| 69 | | - int retries = 0; |
|---|
| 70 | | - gfp_t lflags = kmem_flags_convert(flags); |
|---|
| 71 | 68 | void *ptr; |
|---|
| 72 | 69 | |
|---|
| 73 | | - do { |
|---|
| 74 | | - ptr = krealloc(old, newsize, lflags); |
|---|
| 75 | | - if (ptr || (flags & (KM_MAYFAIL|KM_NOSLEEP))) |
|---|
| 70 | + trace_kmem_alloc_io(size, flags, _RET_IP_); |
|---|
| 71 | + |
|---|
| 72 | + if (WARN_ON_ONCE(align_mask >= PAGE_SIZE)) |
|---|
| 73 | + align_mask = PAGE_SIZE - 1; |
|---|
| 74 | + |
|---|
| 75 | + ptr = kmem_alloc(size, flags | KM_MAYFAIL); |
|---|
| 76 | + if (ptr) { |
|---|
| 77 | + if (!((uintptr_t)ptr & align_mask)) |
|---|
| 76 | 78 | return ptr; |
|---|
| 77 | | - if (!(++retries % 100)) |
|---|
| 78 | | - xfs_err(NULL, |
|---|
| 79 | | - "%s(%u) possible memory allocation deadlock size %zu in %s (mode:0x%x)", |
|---|
| 80 | | - current->comm, current->pid, |
|---|
| 81 | | - newsize, __func__, lflags); |
|---|
| 82 | | - congestion_wait(BLK_RW_ASYNC, HZ/50); |
|---|
| 83 | | - } while (1); |
|---|
| 79 | + kfree(ptr); |
|---|
| 80 | + } |
|---|
| 81 | + return __kmem_vmalloc(size, flags); |
|---|
| 84 | 82 | } |
|---|
| 85 | 83 | |
|---|
| 86 | 84 | void * |
|---|
| 87 | | -kmem_zone_alloc(kmem_zone_t *zone, xfs_km_flags_t flags) |
|---|
| 85 | +kmem_alloc_large(size_t size, xfs_km_flags_t flags) |
|---|
| 88 | 86 | { |
|---|
| 89 | | - int retries = 0; |
|---|
| 90 | | - gfp_t lflags = kmem_flags_convert(flags); |
|---|
| 91 | 87 | void *ptr; |
|---|
| 92 | 88 | |
|---|
| 93 | | - do { |
|---|
| 94 | | - ptr = kmem_cache_alloc(zone, lflags); |
|---|
| 95 | | - if (ptr || (flags & (KM_MAYFAIL|KM_NOSLEEP))) |
|---|
| 96 | | - return ptr; |
|---|
| 97 | | - if (!(++retries % 100)) |
|---|
| 98 | | - xfs_err(NULL, |
|---|
| 99 | | - "%s(%u) possible memory allocation deadlock in %s (mode:0x%x)", |
|---|
| 100 | | - current->comm, current->pid, |
|---|
| 101 | | - __func__, lflags); |
|---|
| 102 | | - congestion_wait(BLK_RW_ASYNC, HZ/50); |
|---|
| 103 | | - } while (1); |
|---|
| 89 | + trace_kmem_alloc_large(size, flags, _RET_IP_); |
|---|
| 90 | + |
|---|
| 91 | + ptr = kmem_alloc(size, flags | KM_MAYFAIL); |
|---|
| 92 | + if (ptr) |
|---|
| 93 | + return ptr; |
|---|
| 94 | + return __kmem_vmalloc(size, flags); |
|---|
| 104 | 95 | } |
|---|