Skip to content

Commit

Permalink
KVM: x86: Dirty the dest op page on cmpxchg emulation
Browse files Browse the repository at this point in the history
Intel SDM says for CMPXCHG: "To simplify the interface to the processor’s bus,
the destination operand receives a write cycle without regard to the result of
the comparison.". This means the destination page should be dirtied.

Fix it to by writing back the original value if cmpxchg failed.

Signed-off-by: Nadav Amit <namit@cs.technion.ac.il>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
  • Loading branch information
anadav authored and bonzini committed Jan 26, 2015
1 parent 8fff5e3 commit 2fcf5c8
Showing 1 changed file with 7 additions and 4 deletions.
11 changes: 7 additions & 4 deletions arch/x86/kvm/emulate.c
Original file line number Diff line number Diff line change
Expand Up @@ -2205,12 +2205,15 @@ static int em_cmpxchg(struct x86_emulate_ctxt *ctxt)
fastop(ctxt, em_cmp);

if (ctxt->eflags & EFLG_ZF) {
/* Success: write back to memory. */
/* Success: write back to memory; no update of EAX */
ctxt->src.type = OP_NONE;
ctxt->dst.val = ctxt->src.orig_val;
} else {
/* Failure: write the value we saw to EAX. */
ctxt->dst.type = OP_REG;
ctxt->dst.addr.reg = reg_rmw(ctxt, VCPU_REGS_RAX);
ctxt->src.type = OP_REG;
ctxt->src.addr.reg = reg_rmw(ctxt, VCPU_REGS_RAX);
ctxt->src.val = ctxt->dst.orig_val;
/* Create write-cycle to dest by writing the same value */
ctxt->dst.val = ctxt->dst.orig_val;
}
return X86EMUL_CONTINUE;
Expand Down Expand Up @@ -4157,7 +4160,7 @@ static const struct opcode twobyte_table[256] = {
F(DstMem | SrcReg | Src2CL | ModRM, em_shrd),
GD(0, &group15), F(DstReg | SrcMem | ModRM, em_imul),
/* 0xB0 - 0xB7 */
I2bv(DstMem | SrcReg | ModRM | Lock | PageTable, em_cmpxchg),
I2bv(DstMem | SrcReg | ModRM | Lock | PageTable | SrcWrite, em_cmpxchg),
I(DstReg | SrcMemFAddr | ModRM | Src2SS, em_lseg),
F(DstMem | SrcReg | ModRM | BitOp | Lock, em_btr),
I(DstReg | SrcMemFAddr | ModRM | Src2FS, em_lseg),
Expand Down

0 comments on commit 2fcf5c8

Please sign in to comment.