From: Ard Biesheuvel <ardb+git@google.com>
To: linux-kernel@vger.kernel.org
Cc: Ard Biesheuvel <ardb@kernel.org>,
Sebastian Andrzej Siewior <bigeasy@linutronix.de>,
Peter Zijlstra <peterz@infradead.org>,
Ingo Molnar <mingo@redhat.com>, Will Deacon <will@kernel.org>,
Boqun Feng <boqun@kernel.org>, Waiman Long <longman@redhat.com>,
Nathan Chancellor <nathan@kernel.org>,
Nick Desaulniers <ndesaulniers@google.com>,
Bill Wendling <morbo@google.com>,
Justin Stitt <justinstitt@google.com>,
llvm@lists.linux.dev
Subject: [PATCH] locking/local_lock: Don't evaluate the per-CPU lock pointer if unused
Date: Sun, 22 Nov 2026 11:56:47 +0100 [thread overview]
Message-ID: <20261122105646.2223131-2-ardb+git@google.com> (raw)
From: Ard Biesheuvel <ardb@kernel.org>
When lock debugging is disabled, local_lock_acquire() and friends are
empty, and so the per-CPU lock pointer that is passed to them, which is
usually computed using this_cpu_ptr(), is not used at all.
However, on x86, when the compiler does not support named address spaces
(i.e., Clang), this_cpu_ptr() is based on inline asm with a memory
operand, and Clang does not discard such inline asm even if its result
is unused. So the computation of the per-CPU address of the lock is
emitted regardless, e.g., on every local_lock_nested_bh() in
process_backlog() and on every local_lock() in the folio batching code.
This was addressed before by commit
9983a9d577db ("locking/local_lock: Make the empty local_lock_*() function a macro.")
but that change was reverted by commit
2d2f8f083ef2 ("Revert "locking/local_lock: Make the empty local_lock_*() function a macro."")
once commit
1c1e7e3c23dd ("x86/percpu: Remove volatile from arch_raw_cpu_ptr().")
had dropped the volatile qualifier from the inline asm, on the basis that
the compiler no longer emits the unused per-CPU reference in that case.
This is true for GCC, but not for Clang.
So implement the !CONFIG_DEBUG_LOCK_ALLOC versions of these helpers as
macros that only reference their argument in dead code, so that it is
type checked but never evaluated. Referencing it only in an unevaluated
context instead, as the typecheck() in the original macros did, makes
Clang complain about per-CPU locks that are only used via these helpers
(-Wunneeded-internal-declaration), which is what commit
a4812d47deff ("mm/page_alloc: mark pagesets as __maybe_unused")
had to work around at the time.
Also, only evaluate the lock pointer in __local_lock_acquire() and
__local_lock_release() when it is actually needed, i.e., for
local_trylock_t.
Cc: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Will Deacon <will@kernel.org>
Cc: Boqun Feng <boqun@kernel.org>
Cc: Waiman Long <longman@redhat.com>
Cc: Nathan Chancellor <nathan@kernel.org>
Cc: Nick Desaulniers <ndesaulniers@google.com>
Cc: Bill Wendling <morbo@google.com>
Cc: Justin Stitt <justinstitt@google.com>
Cc: llvm@lists.linux.dev
Fixes: 2d2f8f083ef2 ("Revert "locking/local_lock: Make the empty local_lock_*() function a macro."")
Signed-off-by: Ard Biesheuvel <ardb@kernel.org>
---
include/linux/local_lock_internal.h | 47 +++++++++++++-------
1 file changed, 30 insertions(+), 17 deletions(-)
diff --git a/include/linux/local_lock_internal.h b/include/linux/local_lock_internal.h
index 234be7f12c15..7c3191438c90 100644
--- a/include/linux/local_lock_internal.h
+++ b/include/linux/local_lock_internal.h
@@ -69,10 +69,21 @@ static inline void local_lock_debug_init(local_lock_t *l)
#else /* CONFIG_DEBUG_LOCK_ALLOC */
# define LOCAL_LOCK_DEBUG_INIT(lockname)
# define LOCAL_TRYLOCK_DEBUG_INIT(lockname)
-static inline void local_lock_acquire(local_lock_t *l) { }
-static inline void local_trylock_acquire(local_lock_t *l) { }
-static inline void local_lock_release(local_lock_t *l) { }
static inline void local_lock_debug_init(local_lock_t *l) { }
+
+/*
+ * The argument passed to these is usually a this_cpu_ptr() expression, which
+ * is implemented using inline asm when the compiler lacks support for named
+ * address spaces (e.g., Clang). Clang may fail to discard such inline asm
+ * even if its result is never used, so only reference the argument in dead
+ * code, which type checks it without evaluating it. (Referencing it only in an
+ * unevaluated context such as sizeof() would make Clang complain about per-CPU
+ * locks that are only used via these helpers being unneeded.)
+ */
+static inline void __local_lock_typecheck(local_lock_t *l) { }
+# define local_lock_acquire(l) ({ if (0) __local_lock_typecheck(l); })
+# define local_trylock_acquire(l) ({ if (0) __local_lock_typecheck(l); })
+# define local_lock_release(l) ({ if (0) __local_lock_typecheck(l); })
#endif /* !CONFIG_DEBUG_LOCK_ALLOC */
#define INIT_LOCAL_LOCK(lockname) { LOCAL_LOCK_DEBUG_INIT(lockname) }
@@ -105,20 +116,23 @@ do { \
local_lock_debug_init(lock); \
} while (0)
+/*
+ * Only evaluate @lock if it is actually used, i.e., for local_trylock_t, or
+ * when lock debugging is enabled (see local_lock_acquire() above).
+ */
#define __local_lock_acquire(lock) \
do { \
- local_trylock_t *__tl; \
- local_lock_t *__l; \
- \
- __l = (local_lock_t *)(lock); \
- __tl = (local_trylock_t *)__l; \
_Generic((lock), \
local_trylock_t *: ({ \
+ local_trylock_t *__tl = (local_trylock_t *)(lock); \
+ \
lockdep_assert(__tl->acquired == 0); \
WRITE_ONCE(__tl->acquired, 1); \
+ local_lock_acquire((local_lock_t *)__tl); \
}), \
- local_lock_t *: (void)0); \
- local_lock_acquire(__l); \
+ local_lock_t *: ({ \
+ local_lock_acquire((local_lock_t *)(lock)); \
+ })); \
} while (0)
#define __local_lock(lock) \
@@ -181,18 +195,17 @@ do { \
#define __local_lock_release(lock) \
do { \
- local_trylock_t *__tl; \
- local_lock_t *__l; \
- \
- __l = (local_lock_t *)(lock); \
- __tl = (local_trylock_t *)__l; \
- local_lock_release(__l); \
_Generic((lock), \
local_trylock_t *: ({ \
+ local_trylock_t *__tl = (local_trylock_t *)(lock); \
+ \
+ local_lock_release((local_lock_t *)__tl); \
lockdep_assert(__tl->acquired == 1); \
WRITE_ONCE(__tl->acquired, 0); \
}), \
- local_lock_t *: (void)0); \
+ local_lock_t *: ({ \
+ local_lock_release((local_lock_t *)(lock)); \
+ })); \
} while (0)
#define __local_unlock(lock) \
--
2.56.1.750.gf70dc682cf-goog
reply other threads:[~2026-11-22 10:56 UTC|newest]
Thread overview: [no followups] expand[flat|nested] mbox.gz Atom feed
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20261122105646.2223131-2-ardb+git@google.com \
--to=ardb+git@google.com \
--cc=ardb@kernel.org \
--cc=bigeasy@linutronix.de \
--cc=boqun@kernel.org \
--cc=justinstitt@google.com \
--cc=linux-kernel@vger.kernel.org \
--cc=llvm@lists.linux.dev \
--cc=longman@redhat.com \
--cc=mingo@redhat.com \
--cc=morbo@google.com \
--cc=nathan@kernel.org \
--cc=ndesaulniers@google.com \
--cc=peterz@infradead.org \
--cc=will@kernel.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox