btrfs: make ASSERT no-op in release builds
PATCH v2 1/1 btrfs: make ASSERT no-op in release builds
The current definition of ASSERT(cond) as (void)(cond) is redundant, since these checks have no side effects and don't affect code logic.
However, some checks contain READ_ONCE or other compiler-unfriendly constructs. For example, ASSERT(list_empty) in btrfs_add_dealloc_inode was compiled to a redundant mov instruction due to this issue.
This patch defines ASSERT as BUILD_BUG_ON_INVALID for !CONFIG_BTRFS_ASSERT builds. It also marks full_page_sectors_uptodate as __maybe_unused to suppress "unneeded declaration" warning (it's needed in compile time)
---
Changes from v1:
- Annotate full_page_sectors_uptodate as __maybe_unused to avoid compiler warning
Link to v1: https://lore.kernel.org/linux-btrfs/[email protected]/
--- fs/btrfs/messages.h | 2 +- fs/btrfs/raid56.c | 4 ++-- 2 files changed, 3 insertions(+), 3 deletions(-)
fs/btrfs/messages.h +1 -1
@@ -168,7 +168,7 @@ do { \#endif #else -#define ASSERT(cond, args...) (void)(cond) +#define ASSERT(cond, args...) BUILD_BUG_ON_INVALID(cond)#endif #ifdef CONFIG_BTRFS_DEBUG
fs/btrfs/raid56.c +2 -2
@@ -299,8 +299,8 @@ static int rbio_bucket(struct btrfs_raid_bio *rbio)return hash_64 (num >>16 ,BTRFS_STRIPE_HASH_TABLE_BITS ); } -static bool full_page_sectors_uptodate (struct btrfs_raid_bio *rbio, -unsigned int page_nr) +static __maybe_unused bool full_page_sectors_uptodate (struct btrfs_raid_bio *rbio, +unsigned int page_nr) {const u32 sectorsize = rbio->bioc->fs_info->sectorsize;const u32 sectors_per_page =PAGE_SIZE / sectorsize;
base-commit: e53642b87a4f4b03a8d7e5f8507fc3cd0c595ea6
Signature · 1 line
2.51.1.dirty
Quoted 2 lines · Gladyshev Ilya
The current definition of
ASSERT(cond)as(void)(cond)is redundant, since these checks have no side effects and don't affect code logic.
Have you checked that none of the assert expressions really don't have side effects other than touching the memory?
Quoted 25 lines
However, some checks contain
READ_ONCEor other compiler-unfriendly constructs. For example,ASSERT(list_empty)inbtrfs_add_dealloc_inodewas compiled to a redundant mov instruction due to this issue.This patch defines ASSERT as
BUILD_BUG_ON_INVALIDfor !CONFIG_BTRFS_ASSERT builds. It also marksfull_page_sectors_uptodateas__maybe_unusedto suppress "unneeded declaration" warning (it's needed in compile time)Signed-off-by Gladyshev Ilya <[email protected]>---
Changes from v1:
- Annotatefull_page_sectors_uptodateas__maybe_unusedto avoid compiler warningLink to v1: https://lore.kernel.org/linux-btrfs/[email protected]/
--- fs/btrfs/messages.h | 2 +- fs/btrfs/raid56.c | 4 ++-- 2 files changed, 3 insertions(+), 3 deletions(-)fs/btrfs/messages.h +0
-0
fs/btrfs/messages.h do { \ · line 171
Quoted 5 lines
fs/btrfs/messages.h +1
-1@@ -168,7 +168,7 @@ do { \#endif #else -#define ASSERT(cond, args...) (void)(cond) +#define ASSERT(cond, args...) BUILD_BUG_ON_INVALID(cond)
I'd rather have the expression open coded rather than using BUILD_BUG_ON_INVALID, the name is confusing as it's not checking build time condtitons.
Quoted 6 lines · David Sterba
Quoted 2 lines · Gladyshev Ilya
The current definition of
ASSERT(cond)as(void)(cond)is redundant, since these checks have no side effects and don't affect code logic.Have you checked that none of the assert expressions really don't have side effects other than touching the memory?
Yes, but visually only. Most checks are plain C comparisons, and some call folio/btrfs/refcount _check/test_ functions where I didn't find side effects.
However, fs/btrfs/ has ~880 asserts, so if you know more robust verification methods, I'd be glad to try them.
Quoted 27 lines
Quoted 27 lines
However, some checks contain
READ_ONCEor other compiler-unfriendly constructs. For example,ASSERT(list_empty)inbtrfs_add_dealloc_inodewas compiled to a redundant mov instruction due to this issue.This patch defines ASSERT as
BUILD_BUG_ON_INVALIDfor !CONFIG_BTRFS_ASSERT builds. It also marksfull_page_sectors_uptodateas__maybe_unusedto suppress "unneeded declaration" warning (it's needed in compile time)Signed-off-by Gladyshev Ilya <[email protected]>---
Changes from v1:
- Annotatefull_page_sectors_uptodateas__maybe_unusedto avoid compiler warningLink to v1: https://lore.kernel.org/linux-btrfs/[email protected]/
--- fs/btrfs/messages.h | 2 +- fs/btrfs/raid56.c | 4 ++-- 2 files changed, 3 insertions(+), 3 deletions(-)fs/btrfs/messages.h +0
-0@@ -168,7 +168,7 @@ do { \#endif
Quoted 7 lines
fs/btrfs/messages.h do { \ · line 171
Quoted 3 lines
fs/btrfs/messages.h +1
-1#else -#define ASSERT(cond, args...) (void)(cond) +#define ASSERT(cond, args...) BUILD_BUG_ON_INVALID(cond)I'd rather have the expression open coded rather than using
BUILD_BUG_ON_INVALID, the name is confusing as it's not checking build time condtitons.
The name kinda indicates that it triggers on invalid conditions, not false ones, but I understand that it can be confusing. While we could use direct sizeof() magic here, I prefer reusing the same infrastructure as VM_BUG_ON(), VFS_*_ON() and others.
Maybe adding a comment about its semantics above ASSERT definition will help clarify the usage? But if you prefer the sizeof() approach, I can change it - it's not a big deal.
Quoted 8 lines · Gladyshev Ilya
Quoted 6 lines · David Sterba
Quoted 2 lines · Gladyshev Ilya
The current definition of
ASSERT(cond)as(void)(cond)is redundant, since these checks have no side effects and don't affect code logic.Have you checked that none of the assert expressions really don't have side effects other than touching the memory?
Quoted 6 lines
Yes, but visually only. Most checks are plain C comparisons, and some call folio/btrfs/refcount _check/test_ functions where I didn't find side effects.
However,
fs/btrfs/has ~880 asserts, so if you know more robust verification methods, I'd be glad to try them.
Good, thanks. I tried the same, with some random grep filters for possible function calls but nothing out of scope found so I guess this is sufficient.
Quoted 35 lines
Quoted 27 lines
Quoted 27 lines
However, some checks contain
READ_ONCEor other compiler-unfriendly constructs. For example,ASSERT(list_empty)inbtrfs_add_dealloc_inodewas compiled to a redundant mov instruction due to this issue.This patch defines ASSERT as
BUILD_BUG_ON_INVALIDfor !CONFIG_BTRFS_ASSERT builds. It also marksfull_page_sectors_uptodateas__maybe_unusedto suppress "unneeded declaration" warning (it's needed in compile time)Signed-off-by Gladyshev Ilya <[email protected]>---
Changes from v1:
- Annotatefull_page_sectors_uptodateas__maybe_unusedto avoid compiler warningLink to v1: https://lore.kernel.org/linux-btrfs/[email protected]/
--- fs/btrfs/messages.h | 2 +- fs/btrfs/raid56.c | 4 ++-- 2 files changed, 3 insertions(+), 3 deletions(-)fs/btrfs/messages.h +0
-0@@ -168,7 +168,7 @@ do { \#endif Quoted 7 lines
fs/btrfs/messages.h do { \ · line 171
Quoted 3 lines
fs/btrfs/messages.h +1
-1#else -#define ASSERT(cond, args...) (void)(cond) +#define ASSERT(cond, args...) BUILD_BUG_ON_INVALID(cond)I'd rather have the expression open coded rather than using
BUILD_BUG_ON_INVALID, the name is confusing as it's not checking build time condtitons.
Quoted 8 lines
The name kinda indicates that it triggers on invalid conditions, not false ones, but I understand that it can be confusing. While we could use direct
sizeof()magic here, I prefer reusing the same infrastructure asVM_BUG_ON(), VFS_*_ON() and others.Maybe adding a comment about its semantics above ASSERT definition will help clarify the usage? But if you prefer the
sizeof()approach, I can change it - it's not a big deal.
A comment for ASSERT works too. The BUILD_BUG_ON_INVALID is indeed widely used so I don't expect any sudden change in semantics. As adding the comment is simple I'll do that, no need to resend the patch. Thanks.