fs: btrfs: Fix cache alignment bugs
PATCH 1/1 fs: btrfs: Fix cache alignment bugs
The btrfs implementation passes cache-unaligned buffers into the block layer, which triggers cache alignment problems down in the block device drivers. Align the buffers to prevent this.
--- fs/btrfs/ctree.c | 24 +++++++++++++----------- fs/btrfs/extent-io.c | 3 ++- fs/btrfs/super.c | 3 ++- 3 files changed, 17 insertions(+), 13 deletions(-)
fs/btrfs/ctree.c +13 -11
@@ -7,6 +7,7 @@#include "btrfs.h" #include <malloc.h> +#include <memalign.h> int btrfs_comp_keys (struct btrfs_key *a,struct btrfs_key *b) {
@@ -105,23 +106,24 @@ void btrfs_free_path(struct btrfs_path *p)static int read_tree_node (u64 physical,union btrfs_tree_node **buf) { -struct btrfs_header hdr; -unsigned long size, offset =sizeof (hdr); + ALLOC_CACHE_ALIGN_BUFFER(struct btrfs_header, hdr, + sizeof(struct btrfs_header)); +unsigned long size, offset =sizeof (*hdr);union btrfs_tree_node *res;u32 i; -if (!btrfs_devread (physical,sizeof (hdr), &hdr)) +if (!btrfs_devread (physical,sizeof (*hdr), hdr))return -1 ; -btrfs_header_to_cpu (&hdr); +btrfs_header_to_cpu (hdr); -if (hdr.level) +if (hdr->level) size =sizeof (struct btrfs_node ) - + hdr.nritems *sizeof (struct btrfs_key_ptr ); + + hdr->nritems *sizeof (struct btrfs_key_ptr );else size = btrfs_info.sb.nodesize; - res =malloc (size); + res =malloc_cache_aligned (size);if (!res) {debug ("%s: malloc failed\n" , __func__);return -1 ;
@@ -133,12 +135,12 @@ static int read_tree_node(u64 physical, union btrfs_tree_node **buf)return -1 ; } - res->header = hdr; -if (hdr.level) -for (i =0 ; i < hdr.nritems; ++i) +memcpy (&res->header, hdr,sizeof (*hdr)); +if (hdr->level) +for (i =0 ; i < hdr->nritems; ++i)btrfs_key_ptr_to_cpu (&res->node.ptrs[i]);else -for (i =0 ; i < hdr.nritems; ++i) +for (i =0 ; i < hdr->nritems; ++i)btrfs_item_to_cpu (&res->leaf.items[i]); *buf = res;
fs/btrfs/extent-io.c +2 -1
@@ -7,6 +7,7 @@#include "btrfs.h" #include <malloc.h> +#include <memalign.h> u64 btrfs_read_extent_inline(struct btrfs_path *path, struct btrfs_file_extent_item *extent, u64 offset,
@@ -89,7 +90,7 @@ u64 btrfs_read_extent_reg(struct btrfs_path *path,return size; } - cbuf =malloc (dlen > size ? clen + dlen : clen); + cbuf =malloc_cache_aligned (dlen > size ? clen + dlen : clen);if (!cbuf)return -1ULL ;
fs/btrfs/super.c +2 -1
@@ -6,6 +6,7 @@*/ #include "btrfs.h" +#include <memalign.h> #define BTRFS_SUPER_FLAG_SUPP (BTRFS_HEADER_FLAG_WRITTEN \ | BTRFS_HEADER_FLAG_RELOC \
@@ -179,7 +180,7 @@ int btrfs_read_superblock(void)0x4000000000ull , 0x4000000000000ull }; -char raw_sb[BTRFS_SUPER_INFO_SIZE ]; +ALLOC_CACHE_ALIGN_BUFFER (char, raw_sb,BTRFS_SUPER_INFO_SIZE );struct btrfs_super_block *sb = (struct btrfs_super_block *) raw_sb;u64 dev_total_bytes;int i;
Signature · 1 line
2.18.0
Quoted 6 lines · Marek Vasut
The btrfs implementation passes cache-unaligned buffers into the block layer, which triggers cache alignment problems down in the block device drivers. Align the buffers to prevent this.
Signed-off-by Marek Vasut <[email protected]>Cc Marek Behun <[email protected]>
Applied to u-boot/master, thanks!
Signature · 8 lines
Tom -------------- next part -------------- A non-text attachment was scrubbed... Name: signature.asc Type: application/pgp-signature Size: 819 bytes Desc: not available URL: <http://lists.denx.de/pipermail/u-boot/attachments/20180930/14e8f327/attachment.sig>
PATCH 1/1 fs: btrfs: Fix cache alignment bugs
In reply to Marek VasutQuoted 115 lines · On Sat, 22 Sep 2018 04:13:35 +0200 Marek Vasut <[email protected]> wrote:
The btrfs implementation passes cache-unaligned buffers into the block layer, which triggers cache alignment problems down in the block device drivers. Align the buffers to prevent this.
Signed-off-by Marek Vasut <[email protected]>Cc Marek Behun <[email protected]>--- fs/btrfs/ctree.c | 24 +++++++++++++----------- fs/btrfs/extent-io.c | 3 ++- fs/btrfs/super.c | 3 ++- 3 files changed, 17 insertions(+), 13 deletions(-)fs/btrfs/ctree.c +13
-11@@ -7,6 +7,7 @@#include "btrfs.h" #include <malloc.h> +#include <memalign.h> int btrfs_comp_keys (struct btrfs_key *a,struct btrfs_key *b) {@@ -105,23 +106,24 @@ void btrfs_free_path(struct btrfs_path *p)static int read_tree_node (u64 physical,union btrfs_tree_node **buf) { -struct btrfs_header hdr; -unsigned long size, offset =sizeof (hdr); + ALLOC_CACHE_ALIGN_BUFFER(struct btrfs_header, hdr, + sizeof(struct btrfs_header)); +unsigned long size, offset =sizeof (*hdr);union btrfs_tree_node *res;u32 i; -if (!btrfs_devread (physical,sizeof (hdr), &hdr)) +if (!btrfs_devread (physical,sizeof (*hdr), hdr))return -1 ; -btrfs_header_to_cpu (&hdr); +btrfs_header_to_cpu (hdr); -if (hdr.level) +if (hdr->level) size =sizeof (struct btrfs_node ) - + hdr.nritems *sizeof (struct btrfs_key_ptr ); + + hdr->nritems *sizeof (struct btrfs_key_ptr );else size = btrfs_info.sb.nodesize; - res =malloc (size); + res =malloc_cache_aligned (size);if (!res) {debug ("%s: malloc failed\n" , __func__);return -1 ;@@ -133,12 +135,12 @@ static int read_tree_node(u64 physical, union btrfs_tree_node **buf) return -1; } - res->header = hdr; -if (hdr.level) -for (i =0 ; i < hdr.nritems; ++i) +memcpy (&res->header, hdr,sizeof (*hdr)); +if (hdr->level) +for (i =0 ; i < hdr->nritems; ++i)btrfs_key_ptr_to_cpu (&res->node.ptrs[i]);else -for (i =0 ; i < hdr.nritems; ++i) +for (i =0 ; i < hdr->nritems; ++i)btrfs_item_to_cpu (&res->leaf.items[i]); *buf = res;fs/btrfs/extent-io.c +1
-0@@ -7,6 +7,7 @@#include "btrfs.h" #include <malloc.h> +#include <memalign.h> u64 btrfs_read_extent_inline(struct btrfs_path *path, struct btrfs_file_extent_item *extent,u64 offset, @@ -89,7 +90,7 @@ u64 btrfs_read_extent_reg(struct
btrfs_path*path, return size;fs/btrfs/extent-io.c +1
-1} - cbuf =malloc (dlen > size ? clen + dlen : clen); + cbuf = malloc_cache_aligned(dlen > size ? clen + dlen :clen); if (!cbuf)
fs/btrfs/extent-io.c +0
-0return -1ULL ;fs/btrfs/super.c +2
-1@@ -6,6 +6,7 @@*/ #include "btrfs.h" +#include <memalign.h> #define BTRFS_SUPER_FLAG_SUPP (BTRFS_HEADER_FLAG_WRITTEN \ | BTRFS_HEADER_FLAG_RELOC \@@ -179,7 +180,7 @@ int btrfs_read_superblock(void)0x4000000000ull ,0x4000000000000ull }; -char raw_sb[BTRFS_SUPER_INFO_SIZE ]; + ALLOC_CACHE_ALIGN_BUFFER(char, raw_sb,
BTRFS_SUPER_INFO_SIZE); structbtrfs_super_block*sb = (structbtrfs_super_block*)raw_sb; u64dev_total_bytes;
int i;
PATCH 1/1 fs: btrfs: Fix cache alignment bugs
Quoted 1 line · Marek Behún
Tested-by Marek Behún <[email protected]>
btw don't you see those warnings/problems on the Armada platform ? I presume this is mostly used on the Omnia. Maybe the cache alignment checking there is not as verbose as on other platforms.
Quoted 119 lines
Quoted 115 lines · On Sat, 22 Sep 2018 04:13:35 +0200 Marek Vasut <[email protected]> wrote:
The btrfs implementation passes cache-unaligned buffers into the block layer, which triggers cache alignment problems down in the block device drivers. Align the buffers to prevent this.
Signed-off-by Marek Vasut <[email protected]>Cc Marek Behun <[email protected]>--- fs/btrfs/ctree.c | 24 +++++++++++++----------- fs/btrfs/extent-io.c | 3 ++- fs/btrfs/super.c | 3 ++- 3 files changed, 17 insertions(+), 13 deletions(-)fs/btrfs/ctree.c +13
-11@@ -7,6 +7,7 @@#include "btrfs.h" #include <malloc.h> +#include <memalign.h> int btrfs_comp_keys (struct btrfs_key *a,struct btrfs_key *b) {@@ -105,23 +106,24 @@ void btrfs_free_path(struct btrfs_path *p)static int read_tree_node (u64 physical,union btrfs_tree_node **buf) { -struct btrfs_header hdr; -unsigned long size, offset =sizeof (hdr); + ALLOC_CACHE_ALIGN_BUFFER(struct btrfs_header, hdr, + sizeof(struct btrfs_header)); +unsigned long size, offset =sizeof (*hdr);union btrfs_tree_node *res;u32 i; -if (!btrfs_devread (physical,sizeof (hdr), &hdr)) +if (!btrfs_devread (physical,sizeof (*hdr), hdr))return -1 ; -btrfs_header_to_cpu (&hdr); +btrfs_header_to_cpu (hdr); -if (hdr.level) +if (hdr->level) size =sizeof (struct btrfs_node ) - + hdr.nritems *sizeof (struct btrfs_key_ptr ); + + hdr->nritems *sizeof (struct btrfs_key_ptr );else size = btrfs_info.sb.nodesize; - res =malloc (size); + res =malloc_cache_aligned (size);if (!res) {debug ("%s: malloc failed\n" , __func__);return -1 ;@@ -133,12 +135,12 @@ static int read_tree_node(u64 physical, union btrfs_tree_node **buf) return -1; } - res->header = hdr; -if (hdr.level) -for (i =0 ; i < hdr.nritems; ++i) +memcpy (&res->header, hdr,sizeof (*hdr)); +if (hdr->level) +for (i =0 ; i < hdr->nritems; ++i)btrfs_key_ptr_to_cpu (&res->node.ptrs[i]);else -for (i =0 ; i < hdr.nritems; ++i) +for (i =0 ; i < hdr->nritems; ++i)btrfs_item_to_cpu (&res->leaf.items[i]); *buf = res;fs/btrfs/extent-io.c +1
-0@@ -7,6 +7,7 @@#include "btrfs.h" #include <malloc.h> +#include <memalign.h> u64 btrfs_read_extent_inline(struct btrfs_path *path, struct btrfs_file_extent_item *extent,u64 offset, @@ -89,7 +90,7 @@ u64 btrfs_read_extent_reg(struct
btrfs_path*path, return size;fs/btrfs/extent-io.c +1
-1} - cbuf =malloc (dlen > size ? clen + dlen : clen); + cbuf = malloc_cache_aligned(dlen > size ? clen + dlen :clen); if (!cbuf)
fs/btrfs/extent-io.c +0
-0return -1ULL ;fs/btrfs/super.c +2
-1@@ -6,6 +6,7 @@*/ #include "btrfs.h" +#include <memalign.h> #define BTRFS_SUPER_FLAG_SUPP (BTRFS_HEADER_FLAG_WRITTEN \ | BTRFS_HEADER_FLAG_RELOC \@@ -179,7 +180,7 @@ int btrfs_read_superblock(void)0x4000000000ull ,0x4000000000000ull }; -char raw_sb[BTRFS_SUPER_INFO_SIZE ]; + ALLOC_CACHE_ALIGN_BUFFER(char, raw_sb,
BTRFS_SUPER_INFO_SIZE); structbtrfs_super_block*sb = (structbtrfs_super_block*)raw_sb; u64dev_total_bytes;
int i;
Signature · 2 lines
Best regards, Marek Vasut
PATCH 1/1 fs: btrfs: Fix cache alignment bugs
Quoted 6 lines · On Tue, 2 Oct 2018 13:43:55 +0200 Marek Vasut <[email protected]> wrote:
Quoted 1 line · Marek Behún
Tested-by Marek Behún <[email protected]>btw don't you see those warnings/problems on the Armada platform ? I presume this is mostly used on the Omnia. Maybe the cache alignment checking there is not as verbose as on other platforms.
Hi, no, I cannot say that I had problem with that (but maybe I don't remember), but the last 12 months I have been working on 64bit Armada, so maybe the problem does not present on AArch64.
Marek
PATCH 1/1 fs: btrfs: Fix cache alignment bugs
Quoted 6 lines · Marek Behún
Quoted 3 lines · On Tue, 2 Oct 2018 13:43:55 +0200 Marek Vasut <[email protected]> wrote:
Quoted 1 line · Marek Behún
Tested-by Marek Behún <[email protected]>
Quoted 7 lines
Quoted 3 lines
btw don't you see those warnings/problems on the Armada platform ? I presume this is mostly used on the Omnia. Maybe the cache alignment checking there is not as verbose as on other platforms.
Hi, no, I cannot say that I had problem with that (but maybe I don't remember), but the last 12 months I have been working on 64bit Armada, so maybe the problem does not present on AArch64.
It is present on all systems with data cache enabled and with block drivers that do DMA, since those structures are not aligned.
Signature · 2 lines
Best regards, Marek Vasut