Aurelien Jarno wrote:
On Sun, Jul 15, 2007 at 09:23:31AM +0300, Avi Kivity wrote:
Can you bisect kvm to find the offending commit? Basically, you do a
git clone git://git.kernel.org/pub/scm/linux/kernel/git/avi/kvm.git
cd kvm
git bisect start drivers/kvm/
bit bisect bad kvm-29
git bisect good kvm-28
git will check out a test candidate; go to your kvm userspace directory
and do
make -C kernel sync LINUX=/path/to/the/kvm/git/directory
make -C kernel
sudo make -C kernel install
reload the module, test, and issue 'git bisect good' or 'git bisect bad'
according to the result. As there are only 25 commits you should be
done in 5 cycles.
And the result is:
commit ba9c20c048726037664d303362b688759fdf6e9d
Author: Luca Tettamanti <[EMAIL PROTECTED]>
Date: Tue Jun 19 22:41:20 2007 +0200
KVM: Fix x86 emulator writeback
When the old value and new one are the same the emulator skips the
write; this is undesirable when the destination is a MMIO area and the
write shall be performed regardless of the previous value. This
optimization breaks e.g. a Linux guest APIC compiled without
X86_GOOD_APIC.
Remove the check and perform the writeback stage in the emulation unless
it's explicitly disabled (currently push and some 2 bytes instructions
may disable the writeback).
Signed-Off-By: Luca Tettamanti <[EMAIL PROTECTED]>
Signed-off-by: Avi Kivity <[EMAIL PROTECTED]>
I have tried to revert this patch directly into kvm-29, and it also
fixes the problem.
From a cursory inspection, looks like the cmov instructions were broken
by the patch. Can you try the attached patch on top of kvm-29?
--
error compiling committee.c: too many arguments to function
diff --git a/drivers/kvm/x86_emulate.c b/drivers/kvm/x86_emulate.c
index f60012d..7974012 100644
--- a/drivers/kvm/x86_emulate.c
+++ b/drivers/kvm/x86_emulate.c
@@ -1225,40 +1225,40 @@ twobyte_insn:
break;
case 0x40 ... 0x4f: /* cmov */
dst.val = dst.orig_val = src.val;
- d &= ~Mov; /* default to no move */
+ no_wb = 1; /* default to no move */
/*
* First, assume we're decoding an even cmov opcode
* (lsb == 0).
*/
switch ((b & 15) >> 1) {
case 0: /* cmovo */
- d |= (_eflags & EFLG_OF) ? Mov : 0;
+ no_wb &= (_eflags & EFLG_OF) ? 0 : 1;
break;
case 1: /* cmovb/cmovc/cmovnae */
- d |= (_eflags & EFLG_CF) ? Mov : 0;
+ no_wb &= (_eflags & EFLG_CF) ? 0 : 1;
break;
case 2: /* cmovz/cmove */
- d |= (_eflags & EFLG_ZF) ? Mov : 0;
+ no_wb &= (_eflags & EFLG_ZF) ? 0 : 1;
break;
case 3: /* cmovbe/cmovna */
- d |= (_eflags & (EFLG_CF | EFLG_ZF)) ? Mov : 0;
+ no_wb &= (_eflags & (EFLG_CF | EFLG_ZF)) ? 0 : 1;
break;
case 4: /* cmovs */
- d |= (_eflags & EFLG_SF) ? Mov : 0;
+ no_wb &= (_eflags & EFLG_SF) ? 0 : 1;
break;
case 5: /* cmovp/cmovpe */
- d |= (_eflags & EFLG_PF) ? Mov : 0;
+ no_wb &= (_eflags & EFLG_PF) ? 0 : 1;
break;
case 7: /* cmovle/cmovng */
- d |= (_eflags & EFLG_ZF) ? Mov : 0;
+ no_wb &= (_eflags & EFLG_ZF) ? 0 : 1;
/* fall through */
case 6: /* cmovl/cmovnge */
- d |= (!(_eflags & EFLG_SF) !=
- !(_eflags & EFLG_OF)) ? Mov : 0;
+ no_wb &= (!(_eflags & EFLG_SF) !=
+ !(_eflags & EFLG_OF)) ? 0 : 1;
break;
}
/* Odd cmov opcodes (lsb == 1) have inverted sense. */
- d ^= (b & 1) ? Mov : 0;
+ no_wb ^= (b & 1) ? 1 : 0;
break;
case 0xb0 ... 0xb1: /* cmpxchg */
/*
-------------------------------------------------------------------------
This SF.net email is sponsored by DB2 Express
Download DB2 Express C - the FREE version of DB2 express and take
control of your XML. No limits. Just data. Click to get it now.
http://sourceforge.net/powerbar/db2/
_______________________________________________
kvm-devel mailing list
kvm-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/kvm-devel
