[src/trunk]: src/doc note the sparc / GCC 4.5.3 / NULL savefpstate IPI hack.

[mrg <mrg%NetBSD.org@localhost>]

details:   https://anonhg.NetBSD.org/src/rev/bd1e52112ffa
branches:  trunk
changeset: 768352:bd1e52112ffa
user:      mrg <mrg%NetBSD.org@localhost>
date:      Mon Aug 15 02:29:38 2011 +0000

description:
note the sparc / GCC 4.5.3 / NULL savefpstate IPI hack.

diffstat:

 doc/HACKS |  22 +++++++++++++++++++++-
 1 files changed, 21 insertions(+), 1 deletions(-)

diffs (43 lines):

diff -r a9aa0f0512ed -r bd1e52112ffa doc/HACKS
--- a/doc/HACKS Mon Aug 15 02:22:46 2011 +0000
+++ b/doc/HACKS Mon Aug 15 02:29:38 2011 +0000
@@ -1,4 +1,4 @@
-# $NetBSD: HACKS,v 1.115 2011/03/12 08:03:06 matt Exp $
+# $NetBSD: HACKS,v 1.116 2011/08/15 02:29:38 mrg Exp $
 #
 # This file is intended to document workarounds for currently unsolved
 # (mostly) compiler bugs.
@@ -562,6 +562,25 @@
        kcah
 
 
+port   sparc
+
+       hack    avoid NULL pointer derefs in savefpstate IPIs and GCC 4.5.3
+       cdate   Sun Aug 14 19:26:48 PDT 2011
+       who     mrg
+       file    sys/arch/sparc64/sparc/cpu.c : 1.234
+       file    sys/arch/sparc64/sparc/cpuvar.h : 1.90
+       file    sys/arch/sparc64/sparc/genassym.cf : 1.67
+       file    sys/arch/sparc64/sparc/locore.s : 1.265
+       descr
+               Something is wrong with GCC 4.5.3 and the savefpstate IPI.
+               Post newlock2 there was a bug where a lock was reduced from
+               IPL_SCHED to IPL_VM (?) and occasionally savefpstate IPI
+               would crash due to NULL IPI.  This was fixed by re-using the
+               right IPL value.  However, GCC 4.5.3 build kernels have the
+               same problems.  For now, the hack is re-instated.
+       kcah
+
+
 port   sparc64
 
        hack    disable optimzations for uvm_bio.c on 32 bit kernels
@@ -574,6 +593,7 @@
                any optimization. This results in ubc_release panics.
        kcah
 
+
 port   mips
 
        hack    mips-shared-linker-load-address