Re: lib/38482: C compiler can generate non-restartable code within a RAS

[Andrew Doran <ad%netbsd.org@localhost>]

The following reply was made to PR lib/38482; it has been noted by GNATS.

From: Andrew Doran <ad%netbsd.org@localhost>
To: gnats-bugs%NetBSD.org@localhost
Cc: 
Subject: Re: lib/38482: C compiler can generate non-restartable code within a 
RAS
Date: Tue, 22 Apr 2008 12:08:15 +0100

 On Mon, Apr 21, 2008 at 09:00:01PM +0000, scw%netbsd.org@localhost wrote:
 
 > Rather than fighting the compiler, I propose we implement _atomic_cas_up()
 > in asm code for each affected arch. I can do ARM and m68000.
 
 Yes please. I can do mips and perhaps alpha. IIRC the assembler should
 recognise the .hidden keyword so none of the symbols in the atomic CAS
 file need to be visible outside libc.
  
 > In addition, the rasctl(2) manual page should be updated to strongly
 > recommend that RASs are implemented in asm code only. Perhaps we should go
 > so far as removing the RAS_START/RAS_END macros from the API, though that
 > may preclude using in-line asm...
 
 I also agree with this. C is too hit-and-miss..
 
 > An audit of the tree for similar code sequences would also be a good idea.
 > For example: pthread_lock.c.
 
 I think we may already have assembly sequences in libpthread for some
 architectures.
 
 Thanks,
 Andrew