public inbox for libc-ports@sourceware.org
 help / color / mirror / Atom feed
* [PATCH roland/arm-atomic-warn] Fiddle ARM atomic.h to avoid -Wvolatile-register-var warnings.
@ 2012-08-08 20:44 Roland McGrath
  2012-08-08 20:52 ` Joseph S. Myers
  2012-08-08 22:02 ` Richard Henderson
  0 siblings, 2 replies; 4+ messages in thread
From: Roland McGrath @ 2012-08-08 20:44 UTC (permalink / raw)
  To: libc-ports

Trunk GCC has -Wvolatile-register-var on by default and this hits in
some atomic.h macros because __typeof (*ptr) is leaking volatile-ness
into the non-pointer declarations and being treated as 'register volatile'.

I only did a basic compile test (with trunk gcc for arm-linux-gnueabi).
I looked at one file where I'd been getting the warning (nptl-init.c)
and the generated code differed only in reordering a couple of (non-atomic)
loads whose order shouldn't matter.

Ok?


Thanks,
Roland


ports/ChangeLog.arm
	* sysdeps/unix/sysv/linux/arm/nptl/bits/atomic.h
	[!__GCC_HAVE_SYNC_COMPARE_AND_SWAP_4]
	(__arch_compare_and_exchange_val_32_acq): Use uint32_t rather than
	__typeof (...) for non-pointer variables derived from the arguments.

diff --git a/ports/sysdeps/unix/sysv/linux/arm/nptl/bits/atomic.h b/ports/sysdeps/unix/sysv/linux/arm/nptl/bits/atomic.h
index 4e810a2..c9ad50d 100644
--- a/ports/sysdeps/unix/sysv/linux/arm/nptl/bits/atomic.h
+++ b/ports/sysdeps/unix/sysv/linux/arm/nptl/bits/atomic.h
@@ -1,4 +1,4 @@
-/* Copyright (C) 2002, 2003, 2004, 2005 Free Software Foundation, Inc.
+/* Copyright (C) 2002-2012 Free Software Foundation, Inc.
    This file is part of the GNU C Library.
 
    The GNU C Library is free software; you can redistribute it and/or
@@ -74,17 +74,24 @@ void __arm_link_error (void);
 
 /* It doesn't matter what register is used for a_oldval2, but we must
    specify one to work around GCC PR rtl-optimization/21223.  Otherwise
-   it may cause a_oldval or a_tmp to be moved to a different register.  */
+   it may cause a_oldval or a_tmp to be moved to a different register.
 
+   We use the union trick rather than simply using __typeof (...) in the
+   declarations of A_OLDVAL et al because when NEWVAL or OLDVAL is of the
+   form *PTR and PTR has a 'volatile ... *' type, then __typeof (*PTR) has
+   a 'volatile ...' type and this triggers -Wvolatile-register-var to
+   complain about 'register volatile ... asm ("reg")'.  */
 #elif defined __thumb2__
 /* Thumb-2 has ldrex/strex.  However it does not have barrier instructions,
    so we still need to use the kernel helper.  */
 #define __arch_compare_and_exchange_val_32_acq(mem, newval, oldval) \
-  ({ register __typeof (oldval) a_oldval asm ("r0");			      \
-     register __typeof (oldval) a_newval asm ("r1") = (newval);		      \
+  ({ union { __typeof (oldval) a; uint32_t v; } oldval_arg = { .a = (oldval) };\
+     union { __typeof (newval) a; uint32_t v; } newval_arg = { .a = (newval) };\
+     register uint32_t a_oldval asm ("r0");				      \
+     register uint32_t a_newval asm ("r1") = newval_arg.v;		      \
      register __typeof (mem) a_ptr asm ("r2") = (mem);			      \
-     register __typeof (oldval) a_tmp asm ("r3");			      \
-     register __typeof (oldval) a_oldval2 asm ("r4") = (oldval);	      \
+     register uint32_t a_tmp asm ("r3");				      \
+     register uint32_t a_oldval2 asm ("r4") = oldval_arg.v;		      \
      __asm__ __volatile__						      \
 	     ("0:\tldr\t%[tmp],[%[ptr]]\n\t"				      \
 	      "cmp\t%[tmp], %[old2]\n\t"				      \
@@ -100,14 +107,16 @@ void __arm_link_error (void);
 	      : [new] "r" (a_newval), [ptr] "r" (a_ptr),		      \
 		[old2] "r" (a_oldval2)					      \
 	      : "ip", "lr", "cc", "memory");				      \
-     a_tmp; })
+     (__typeof (oldval)) a_tmp; })
 #else
-#define __arch_compare_and_exchange_val_32_acq(mem, newval, oldval) \
-  ({ register __typeof (oldval) a_oldval asm ("r0");			      \
-     register __typeof (oldval) a_newval asm ("r1") = (newval);		      \
+#define __arch_compare_and_exchange_val_32_acq(mem, newval, oldval)	      \
+  ({ union { __typeof (oldval) a; uint32_t v; } oldval_arg = { .a = (oldval) };\
+     union { __typeof (newval) a; uint32_t v; } newval_arg = { .a = (newval) };\
+     register uint32_t a_oldval asm ("r0");				      \
+     register uint32_t a_newval asm ("r1") = newval_arg.v;		      \
      register __typeof (mem) a_ptr asm ("r2") = (mem);			      \
-     register __typeof (oldval) a_tmp asm ("r3");			      \
-     register __typeof (oldval) a_oldval2 asm ("r4") = (oldval);	      \
+     register uint32_t a_tmp asm ("r3");				      \
+     register uint32_t a_oldval2 asm ("r4") = oldval_arg.v;		      \
      __asm__ __volatile__						      \
 	     ("0:\tldr\t%[tmp],[%[ptr]]\n\t"				      \
 	      "cmp\t%[tmp], %[old2]\n\t"				      \
@@ -123,7 +132,7 @@ void __arm_link_error (void);
 	      : [new] "r" (a_newval), [ptr] "r" (a_ptr),		      \
 		[old2] "r" (a_oldval2)					      \
 	      : "ip", "lr", "cc", "memory");				      \
-     a_tmp; })
+     (__typeof (oldval)) a_tmp; })
 #endif
 
 #define __arch_compare_and_exchange_val_64_acq(mem, newval, oldval) \

^ permalink raw reply	[flat|nested] 4+ messages in thread

* Re: [PATCH roland/arm-atomic-warn] Fiddle ARM atomic.h to avoid -Wvolatile-register-var warnings.
  2012-08-08 20:44 [PATCH roland/arm-atomic-warn] Fiddle ARM atomic.h to avoid -Wvolatile-register-var warnings Roland McGrath
@ 2012-08-08 20:52 ` Joseph S. Myers
  2012-08-08 22:02 ` Richard Henderson
  1 sibling, 0 replies; 4+ messages in thread
From: Joseph S. Myers @ 2012-08-08 20:52 UTC (permalink / raw)
  To: Roland McGrath; +Cc: libc-ports

On Wed, 8 Aug 2012, Roland McGrath wrote:

> Trunk GCC has -Wvolatile-register-var on by default and this hits in
> some atomic.h macros because __typeof (*ptr) is leaking volatile-ness
> into the non-pointer declarations and being treated as 'register volatile'.
> 
> I only did a basic compile test (with trunk gcc for arm-linux-gnueabi).
> I looked at one file where I'd been getting the warning (nptl-init.c)
> and the generated code differed only in reordering a couple of (non-atomic)
> loads whose order shouldn't matter.
> 
> Ok?

OK.

-- 
Joseph S. Myers
joseph@codesourcery.com

^ permalink raw reply	[flat|nested] 4+ messages in thread

* Re: [PATCH roland/arm-atomic-warn] Fiddle ARM atomic.h to avoid -Wvolatile-register-var warnings.
  2012-08-08 20:44 [PATCH roland/arm-atomic-warn] Fiddle ARM atomic.h to avoid -Wvolatile-register-var warnings Roland McGrath
  2012-08-08 20:52 ` Joseph S. Myers
@ 2012-08-08 22:02 ` Richard Henderson
  2012-08-08 22:13   ` Roland McGrath
  1 sibling, 1 reply; 4+ messages in thread
From: Richard Henderson @ 2012-08-08 22:02 UTC (permalink / raw)
  To: Roland McGrath; +Cc: libc-ports

On 08/08/2012 01:44 PM, Roland McGrath wrote:
> +   We use the union trick rather than simply using __typeof (...) in the
> +   declarations of A_OLDVAL et al because when NEWVAL or OLDVAL is of the
> +   form *PTR and PTR has a 'volatile ... *' type, then __typeof (*PTR) has
> +   a 'volatile ...' type and this triggers -Wvolatile-register-var to
> +   complain about 'register volatile ... asm ("reg")'.  */
>  #elif defined __thumb2__
>  /* Thumb-2 has ldrex/strex.  However it does not have barrier instructions,
>     so we still need to use the kernel helper.  */
>  #define __arch_compare_and_exchange_val_32_acq(mem, newval, oldval) \
> -  ({ register __typeof (oldval) a_oldval asm ("r0");			      \
> -     register __typeof (oldval) a_newval asm ("r1") = (newval);		      \
> +  ({ union { __typeof (oldval) a; uint32_t v; } oldval_arg = { .a = (oldval) };\
> +     union { __typeof (newval) a; uint32_t v; } newval_arg = { .a = (newval) };\
> +     register uint32_t a_oldval asm ("r0");				      \
> +     register uint32_t a_newval asm ("r1") = newval_arg.v;		      \
>       register __typeof (mem) a_ptr asm ("r2") = (mem);			      \
> -     register __typeof (oldval) a_tmp asm ("r3");			      \
> -     register __typeof (oldval) a_oldval2 asm ("r4") = (oldval);	      \
> +     register uint32_t a_tmp asm ("r3");				      \
> +     register uint32_t a_oldval2 asm ("r4") = oldval_arg.v;		      \
>       __asm__ __volatile__						      \
>  	     ("0:\tldr\t%[tmp],[%[ptr]]\n\t"				      \

Is it really worth working so hard to avoid the gcc builtins?
Yes, gcc before 4.7 doesn't have __atomic to avoid the extra
barriers, but at least 4.7 and later will.

May I ask what gcc version you're targeting, anyway?


r~

^ permalink raw reply	[flat|nested] 4+ messages in thread

* Re: [PATCH roland/arm-atomic-warn] Fiddle ARM atomic.h to avoid -Wvolatile-register-var warnings.
  2012-08-08 22:02 ` Richard Henderson
@ 2012-08-08 22:13   ` Roland McGrath
  0 siblings, 0 replies; 4+ messages in thread
From: Roland McGrath @ 2012-08-08 22:13 UTC (permalink / raw)
  To: Richard Henderson; +Cc: libc-ports

> Is it really worth working so hard to avoid the gcc builtins?
> Yes, gcc before 4.7 doesn't have __atomic to avoid the extra
> barriers, but at least 4.7 and later will.

That file uses them #ifdef __GCC_HAVE_SYNC_COMPARE_AND_SWAP_4.
I was just killing a warning without worrying about why the code
path was being used.

> May I ask what gcc version you're targeting, anyway?

I'm using today's trunk, cross --target=arm-linux-gnueabi and no other
special options.  I don't know why __GCC_HAVE_SYNC_COMPARE_AND_SWAP_4 is
not defined.  Probably it would be with appropriate -march= settings.


Thanks,
Roland

^ permalink raw reply	[flat|nested] 4+ messages in thread

end of thread, other threads:[~2012-08-08 22:13 UTC | newest]

Thread overview: 4+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2012-08-08 20:44 [PATCH roland/arm-atomic-warn] Fiddle ARM atomic.h to avoid -Wvolatile-register-var warnings Roland McGrath
2012-08-08 20:52 ` Joseph S. Myers
2012-08-08 22:02 ` Richard Henderson
2012-08-08 22:13   ` Roland McGrath

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 read-only IMAP folder(s) and NNTP newsgroup(s).