* [PATCH 1/2] f2fs: use vmalloc instead of kvmalloc in .init_{,de}compress_ctx
@ 2025-05-13 5:57 Chao Yu
2025-05-13 5:57 ` [PATCH 2/2] f2fs: introduce FAULT_VMALLOC Chao Yu
` (2 more replies)
0 siblings, 3 replies; 5+ messages in thread
From: Chao Yu @ 2025-05-13 5:57 UTC (permalink / raw)
To: jaegeuk; +Cc: linux-f2fs-devel, linux-kernel, Chao Yu
.init_{,de}compress_ctx uses kvmalloc() to alloc memory, it will try
to allocate physically continuous page first, it may cause more memory
allocation pressure, let's use vmalloc instead to mitigate it.
[Test]
cd /data/local/tmp
touch file
f2fs_io setflags compression file
f2fs_io getflags file
for i in $(seq 1 10); do sync; echo 3 > /proc/sys/vm/drop_caches;\
time f2fs_io write 512 0 4096 zero osync file; truncate -s 0 file;\
done
[Result]
Before After Delta
21.243 21.694 -2.12%
For compression, we recommend to use ioctl to compress file data in
background for workaround.
For decompression, only zstd will be affected.
Signed-off-by: Chao Yu <chao@kernel.org>
---
fs/f2fs/compress.c | 23 ++++++++++-------------
fs/f2fs/f2fs.h | 5 +++++
2 files changed, 15 insertions(+), 13 deletions(-)
diff --git a/fs/f2fs/compress.c b/fs/f2fs/compress.c
index 1e62fdffda07..37d18e2a3327 100644
--- a/fs/f2fs/compress.c
+++ b/fs/f2fs/compress.c
@@ -180,8 +180,7 @@ void f2fs_compress_ctx_add_page(struct compress_ctx *cc, struct folio *folio)
#ifdef CONFIG_F2FS_FS_LZO
static int lzo_init_compress_ctx(struct compress_ctx *cc)
{
- cc->private = f2fs_kvmalloc(F2FS_I_SB(cc->inode),
- LZO1X_MEM_COMPRESS, GFP_NOFS);
+ cc->private = f2fs_vmalloc(LZO1X_MEM_COMPRESS);
if (!cc->private)
return -ENOMEM;
@@ -191,7 +190,7 @@ static int lzo_init_compress_ctx(struct compress_ctx *cc)
static void lzo_destroy_compress_ctx(struct compress_ctx *cc)
{
- kvfree(cc->private);
+ vfree(cc->private);
cc->private = NULL;
}
@@ -248,7 +247,7 @@ static int lz4_init_compress_ctx(struct compress_ctx *cc)
size = LZ4HC_MEM_COMPRESS;
#endif
- cc->private = f2fs_kvmalloc(F2FS_I_SB(cc->inode), size, GFP_NOFS);
+ cc->private = f2fs_vmalloc(size);
if (!cc->private)
return -ENOMEM;
@@ -263,7 +262,7 @@ static int lz4_init_compress_ctx(struct compress_ctx *cc)
static void lz4_destroy_compress_ctx(struct compress_ctx *cc)
{
- kvfree(cc->private);
+ vfree(cc->private);
cc->private = NULL;
}
@@ -344,8 +343,7 @@ static int zstd_init_compress_ctx(struct compress_ctx *cc)
params = zstd_get_params(level, cc->rlen);
workspace_size = zstd_cstream_workspace_bound(¶ms.cParams);
- workspace = f2fs_kvmalloc(F2FS_I_SB(cc->inode),
- workspace_size, GFP_NOFS);
+ workspace = f2fs_vmalloc(workspace_size);
if (!workspace)
return -ENOMEM;
@@ -353,7 +351,7 @@ static int zstd_init_compress_ctx(struct compress_ctx *cc)
if (!stream) {
f2fs_err_ratelimited(F2FS_I_SB(cc->inode),
"%s zstd_init_cstream failed", __func__);
- kvfree(workspace);
+ vfree(workspace);
return -EIO;
}
@@ -366,7 +364,7 @@ static int zstd_init_compress_ctx(struct compress_ctx *cc)
static void zstd_destroy_compress_ctx(struct compress_ctx *cc)
{
- kvfree(cc->private);
+ vfree(cc->private);
cc->private = NULL;
cc->private2 = NULL;
}
@@ -425,8 +423,7 @@ static int zstd_init_decompress_ctx(struct decompress_io_ctx *dic)
workspace_size = zstd_dstream_workspace_bound(max_window_size);
- workspace = f2fs_kvmalloc(F2FS_I_SB(dic->inode),
- workspace_size, GFP_NOFS);
+ workspace = f2fs_vmalloc(workspace_size);
if (!workspace)
return -ENOMEM;
@@ -434,7 +431,7 @@ static int zstd_init_decompress_ctx(struct decompress_io_ctx *dic)
if (!stream) {
f2fs_err_ratelimited(F2FS_I_SB(dic->inode),
"%s zstd_init_dstream failed", __func__);
- kvfree(workspace);
+ vfree(workspace);
return -EIO;
}
@@ -446,7 +443,7 @@ static int zstd_init_decompress_ctx(struct decompress_io_ctx *dic)
static void zstd_destroy_decompress_ctx(struct decompress_io_ctx *dic)
{
- kvfree(dic->private);
+ vfree(dic->private);
dic->private = NULL;
dic->private2 = NULL;
}
diff --git a/fs/f2fs/f2fs.h b/fs/f2fs/f2fs.h
index 9432fd15766a..6e9615b5cdc3 100644
--- a/fs/f2fs/f2fs.h
+++ b/fs/f2fs/f2fs.h
@@ -3540,6 +3540,11 @@ static inline void *f2fs_kvzalloc(struct f2fs_sb_info *sbi,
return f2fs_kvmalloc(sbi, size, flags | __GFP_ZERO);
}
+static inline void *f2fs_vmalloc(size_t size)
+{
+ return vmalloc(size);
+}
+
static inline int get_extra_isize(struct inode *inode)
{
return F2FS_I(inode)->i_extra_isize / sizeof(__le32);
--
2.49.0
^ permalink raw reply related [flat|nested] 5+ messages in thread
* [PATCH 2/2] f2fs: introduce FAULT_VMALLOC
2025-05-13 5:57 [PATCH 1/2] f2fs: use vmalloc instead of kvmalloc in .init_{,de}compress_ctx Chao Yu
@ 2025-05-13 5:57 ` Chao Yu
2025-05-13 6:01 ` [PATCH 1/2] f2fs: use vmalloc instead of kvmalloc in .init_{,de}compress_ctx Christoph Hellwig
2025-05-14 15:40 ` [f2fs-dev] [PATCH 1/2] f2fs: use vmalloc instead of kvmalloc in .init_{, de}compress_ctx patchwork-bot+f2fs
2 siblings, 0 replies; 5+ messages in thread
From: Chao Yu @ 2025-05-13 5:57 UTC (permalink / raw)
To: jaegeuk; +Cc: linux-f2fs-devel, linux-kernel, Chao Yu
Introduce a new fault type FAULT_VMALLOC to simulate no memory error in
f2fs_vmalloc().
Signed-off-by: Chao Yu <chao@kernel.org>
---
Documentation/ABI/testing/sysfs-fs-f2fs | 1 +
Documentation/filesystems/f2fs.rst | 1 +
fs/f2fs/compress.c | 9 +++++----
fs/f2fs/f2fs.h | 6 +++++-
fs/f2fs/super.c | 1 +
5 files changed, 13 insertions(+), 5 deletions(-)
diff --git a/Documentation/ABI/testing/sysfs-fs-f2fs b/Documentation/ABI/testing/sysfs-fs-f2fs
index e060798f9fc1..bf03263b9f46 100644
--- a/Documentation/ABI/testing/sysfs-fs-f2fs
+++ b/Documentation/ABI/testing/sysfs-fs-f2fs
@@ -736,6 +736,7 @@ Description: Support configuring fault injection type, should be
FAULT_NO_SEGMENT 0x00100000
FAULT_INCONSISTENT_FOOTER 0x00200000
FAULT_TIMEOUT 0x00400000 (1000ms)
+ FAULT_VMALLOC 0x00800000
=========================== ==========
What: /sys/fs/f2fs/<disk>/discard_io_aware_gran
diff --git a/Documentation/filesystems/f2fs.rst b/Documentation/filesystems/f2fs.rst
index 724fc5e2889a..440e4ae74e44 100644
--- a/Documentation/filesystems/f2fs.rst
+++ b/Documentation/filesystems/f2fs.rst
@@ -208,6 +208,7 @@ fault_type=%d Support configuring fault injection type, should be
FAULT_NO_SEGMENT 0x00100000
FAULT_INCONSISTENT_FOOTER 0x00200000
FAULT_TIMEOUT 0x00400000 (1000ms)
+ FAULT_VMALLOC 0x00800000
=========================== ==========
mode=%s Control block allocation mode which supports "adaptive"
and "lfs". In "lfs" mode, there should be no random
diff --git a/fs/f2fs/compress.c b/fs/f2fs/compress.c
index 37d18e2a3327..2a9f7b68a6c6 100644
--- a/fs/f2fs/compress.c
+++ b/fs/f2fs/compress.c
@@ -180,7 +180,8 @@ void f2fs_compress_ctx_add_page(struct compress_ctx *cc, struct folio *folio)
#ifdef CONFIG_F2FS_FS_LZO
static int lzo_init_compress_ctx(struct compress_ctx *cc)
{
- cc->private = f2fs_vmalloc(LZO1X_MEM_COMPRESS);
+ cc->private = f2fs_vmalloc(F2FS_I_SB(cc->inode),
+ LZO1X_MEM_COMPRESS);
if (!cc->private)
return -ENOMEM;
@@ -247,7 +248,7 @@ static int lz4_init_compress_ctx(struct compress_ctx *cc)
size = LZ4HC_MEM_COMPRESS;
#endif
- cc->private = f2fs_vmalloc(size);
+ cc->private = f2fs_vmalloc(F2FS_I_SB(cc->inode), size);
if (!cc->private)
return -ENOMEM;
@@ -343,7 +344,7 @@ static int zstd_init_compress_ctx(struct compress_ctx *cc)
params = zstd_get_params(level, cc->rlen);
workspace_size = zstd_cstream_workspace_bound(¶ms.cParams);
- workspace = f2fs_vmalloc(workspace_size);
+ workspace = f2fs_vmalloc(F2FS_I_SB(cc->inode), workspace_size);
if (!workspace)
return -ENOMEM;
@@ -423,7 +424,7 @@ static int zstd_init_decompress_ctx(struct decompress_io_ctx *dic)
workspace_size = zstd_dstream_workspace_bound(max_window_size);
- workspace = f2fs_vmalloc(workspace_size);
+ workspace = f2fs_vmalloc(F2FS_I_SB(dic->inode), workspace_size);
if (!workspace)
return -ENOMEM;
diff --git a/fs/f2fs/f2fs.h b/fs/f2fs/f2fs.h
index 6e9615b5cdc3..bf6056aa09d8 100644
--- a/fs/f2fs/f2fs.h
+++ b/fs/f2fs/f2fs.h
@@ -64,6 +64,7 @@ enum {
FAULT_NO_SEGMENT,
FAULT_INCONSISTENT_FOOTER,
FAULT_TIMEOUT,
+ FAULT_VMALLOC,
FAULT_MAX,
};
@@ -3540,8 +3541,11 @@ static inline void *f2fs_kvzalloc(struct f2fs_sb_info *sbi,
return f2fs_kvmalloc(sbi, size, flags | __GFP_ZERO);
}
-static inline void *f2fs_vmalloc(size_t size)
+static inline void *f2fs_vmalloc(struct f2fs_sb_info *sbi, size_t size)
{
+ if (time_to_inject(sbi, FAULT_VMALLOC))
+ return NULL;
+
return vmalloc(size);
}
diff --git a/fs/f2fs/super.c b/fs/f2fs/super.c
index 1860edc2a8fb..32f2abac19cf 100644
--- a/fs/f2fs/super.c
+++ b/fs/f2fs/super.c
@@ -68,6 +68,7 @@ const char *f2fs_fault_name[FAULT_MAX] = {
[FAULT_NO_SEGMENT] = "no free segment",
[FAULT_INCONSISTENT_FOOTER] = "inconsistent footer",
[FAULT_TIMEOUT] = "timeout",
+ [FAULT_VMALLOC] = "vmalloc",
};
int f2fs_build_fault_attr(struct f2fs_sb_info *sbi, unsigned long rate,
--
2.49.0
^ permalink raw reply related [flat|nested] 5+ messages in thread
* Re: [PATCH 1/2] f2fs: use vmalloc instead of kvmalloc in .init_{,de}compress_ctx
2025-05-13 5:57 [PATCH 1/2] f2fs: use vmalloc instead of kvmalloc in .init_{,de}compress_ctx Chao Yu
2025-05-13 5:57 ` [PATCH 2/2] f2fs: introduce FAULT_VMALLOC Chao Yu
@ 2025-05-13 6:01 ` Christoph Hellwig
2025-05-13 6:34 ` Chao Yu
2025-05-14 15:40 ` [f2fs-dev] [PATCH 1/2] f2fs: use vmalloc instead of kvmalloc in .init_{, de}compress_ctx patchwork-bot+f2fs
2 siblings, 1 reply; 5+ messages in thread
From: Christoph Hellwig @ 2025-05-13 6:01 UTC (permalink / raw)
To: Chao Yu; +Cc: jaegeuk, linux-f2fs-devel, linux-kernel, linux-mm
On Tue, May 13, 2025 at 01:57:20PM +0800, Chao Yu wrote:
> .init_{,de}compress_ctx uses kvmalloc() to alloc memory, it will try
> to allocate physically continuous page first, it may cause more memory
> allocation pressure, let's use vmalloc instead to mitigate it.
Shouldn't this be handled in kvmalloc instead of working around it in
callers?
^ permalink raw reply [flat|nested] 5+ messages in thread
* Re: [PATCH 1/2] f2fs: use vmalloc instead of kvmalloc in .init_{,de}compress_ctx
2025-05-13 6:01 ` [PATCH 1/2] f2fs: use vmalloc instead of kvmalloc in .init_{,de}compress_ctx Christoph Hellwig
@ 2025-05-13 6:34 ` Chao Yu
0 siblings, 0 replies; 5+ messages in thread
From: Chao Yu @ 2025-05-13 6:34 UTC (permalink / raw)
To: Christoph Hellwig; +Cc: chao, jaegeuk, linux-f2fs-devel, linux-kernel, linux-mm
On 5/13/25 14:01, Christoph Hellwig wrote:
> On Tue, May 13, 2025 at 01:57:20PM +0800, Chao Yu wrote:
>> .init_{,de}compress_ctx uses kvmalloc() to alloc memory, it will try
>> to allocate physically continuous page first, it may cause more memory
>> allocation pressure, let's use vmalloc instead to mitigate it.
>
> Shouldn't this be handled in kvmalloc instead of working around it in
> callers?
kvmalloc() will allocate physically continuous page first, so it may race
w/ other physically continuous page allocator who uses kmalloc(), in
scenario of there are few physically continuous pages.
Thanks,
^ permalink raw reply [flat|nested] 5+ messages in thread
* Re: [f2fs-dev] [PATCH 1/2] f2fs: use vmalloc instead of kvmalloc in .init_{, de}compress_ctx
2025-05-13 5:57 [PATCH 1/2] f2fs: use vmalloc instead of kvmalloc in .init_{,de}compress_ctx Chao Yu
2025-05-13 5:57 ` [PATCH 2/2] f2fs: introduce FAULT_VMALLOC Chao Yu
2025-05-13 6:01 ` [PATCH 1/2] f2fs: use vmalloc instead of kvmalloc in .init_{,de}compress_ctx Christoph Hellwig
@ 2025-05-14 15:40 ` patchwork-bot+f2fs
2 siblings, 0 replies; 5+ messages in thread
From: patchwork-bot+f2fs @ 2025-05-14 15:40 UTC (permalink / raw)
To: Chao Yu; +Cc: jaegeuk, linux-kernel, linux-f2fs-devel
Hello:
This series was applied to jaegeuk/f2fs.git (dev)
by Jaegeuk Kim <jaegeuk@kernel.org>:
On Tue, 13 May 2025 13:57:20 +0800 you wrote:
> .init_{,de}compress_ctx uses kvmalloc() to alloc memory, it will try
> to allocate physically continuous page first, it may cause more memory
> allocation pressure, let's use vmalloc instead to mitigate it.
>
> [Test]
> cd /data/local/tmp
> touch file
> f2fs_io setflags compression file
> f2fs_io getflags file
> for i in $(seq 1 10); do sync; echo 3 > /proc/sys/vm/drop_caches;\
> time f2fs_io write 512 0 4096 zero osync file; truncate -s 0 file;\
> done
>
> [...]
Here is the summary with links:
- [f2fs-dev,1/2] f2fs: use vmalloc instead of kvmalloc in .init_{, de}compress_ctx
https://git.kernel.org/jaegeuk/f2fs/c/10b26e772b10
- [f2fs-dev,2/2] f2fs: introduce FAULT_VMALLOC
https://git.kernel.org/jaegeuk/f2fs/c/2c19d65bab04
You are awesome, thank you!
--
Deet-doot-dot, I am a bot.
https://korg.docs.kernel.org/patchwork/pwbot.html
^ permalink raw reply [flat|nested] 5+ messages in thread
end of thread, other threads:[~2025-05-14 15:39 UTC | newest]
Thread overview: 5+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2025-05-13 5:57 [PATCH 1/2] f2fs: use vmalloc instead of kvmalloc in .init_{,de}compress_ctx Chao Yu
2025-05-13 5:57 ` [PATCH 2/2] f2fs: introduce FAULT_VMALLOC Chao Yu
2025-05-13 6:01 ` [PATCH 1/2] f2fs: use vmalloc instead of kvmalloc in .init_{,de}compress_ctx Christoph Hellwig
2025-05-13 6:34 ` Chao Yu
2025-05-14 15:40 ` [f2fs-dev] [PATCH 1/2] f2fs: use vmalloc instead of kvmalloc in .init_{, de}compress_ctx patchwork-bot+f2fs
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).