Re: x86 rwsem: up_read() does not check active count in fast path

[David Howells]

Michel Lespinasse <walken@xxxxxxxxxx> wrote:

> Looking at the x86 rwsem code, I have been wondering about the up_read()
> path.  __rwsem_do_wake() comment mentions that one should check the active
> count on the way there; however I could not find that check when coming from
> up_read().

Look in call_rwsem_wake(), which is implemented in assembly in two places in
the x86 arch:

	arch/x86/lib/rwsem_64.S
	arch/x86/lib/semaphore_32.S

Note that the x86_64 version contains a bug, a patch for which is attached.

David
---
From: David Howells <dhowells@xxxxxxxxxx>
Subject: [PATCH] Fix the x86_64 implementation of call_rwsem_wait()

The x86_64 call_rwsem_wait() treats the active state counter part of the R/W
semaphore state as being 16-bit when it's actually 32-bit (it's half of the
64-bit state).  It should do "decl %edx" not "decw %dx".

Signed-off-by: David Howells <dhowells@xxxxxxxxxx>
---

 arch/x86/lib/rwsem_64.S |    2 +-
 1 files changed, 1 insertions(+), 1 deletions(-)


diff --git a/arch/x86/lib/rwsem_64.S b/arch/x86/lib/rwsem_64.S
index 15acecf..41fcf00 100644
--- a/arch/x86/lib/rwsem_64.S
+++ b/arch/x86/lib/rwsem_64.S
@@ -60,7 +60,7 @@ ENTRY(call_rwsem_down_write_failed)
 	ENDPROC(call_rwsem_down_write_failed)
 
 ENTRY(call_rwsem_wake)
-	decw %dx    /* do nothing if still outstanding active readers */
+	decl %edx	/* do nothing if still outstanding active readers */
 	jnz 1f
 	save_common_regs
 	movq %rax,%rdi
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/