[ help ] vcpu handle MWAIT different

["Longpeng (Mike)" <longpeng2@huawei.com>]

Hi guys,

I wrote a testcase which repeat execute MWAIT insn in kvm-unit-test and I tested
it on different hardware, but got different results.


1) cpu: E5620 (hardware don't support MONITOR/MWAIT)

This works well because I saw the following info in trace:

       CPU 0/KVM-7301  [015] .... 1534719.099573: kvm_exit: reason
MWAIT_INSTRUCTION rip 0x400387 info 0 0
       CPU 0/KVM-7301  [015] d... 1534719.099573: kvm_entry: vcpu 0
       CPU 0/KVM-7301  [015] .... 1534719.099574: kvm_exit: reason
MWAIT_INSTRUCTION rip 0x400387 info 0 0
       CPU 0/KVM-7301  [015] d... 1534719.099574: kvm_entry: vcpu 0


2) cpu: E5-2620 v3 and E5-2690 v4 (both hardwares support MONITOR/MWAIT)

This...

       CPU 0/KVM-1683  [017] .... 4844125.856164: kvm_exit: reason EXCEPTION_NMI
rip 0x400382 info 0 80000306
       CPU 0/KVM-1683  [017] .... 4844125.856166: kvm_emulate_insn: 0:400382:0f
01 c9 (prot64)
       CPU 0/KVM-1683  [017] .... 4844125.856166: kvm_inj_exception: #UD (0x0)
       CPU 0/KVM-1683  [017] d... 4844125.856167: kvm_entry: vcpu 0

and the kvm-unit-test guest reported:

Unhandled exception 6 #UD at ip 0000000000400387
error_code=0000      rflags=00010002      cs=00000008
rax=0000000000000000 rcx=0000000000000000 rdx=0000000200000000 rbx=0000000000000000
rbp=000000000044db60 rsi=0000000000000001 rdi=0000000000004020
 r8=000000000000000a  r9=00000000000003f8 r10=000000000044d1e2 r11=000000000044d1e1
r12=0000000000000000 r13=0000000000000000 r14=0000000000000000 r15=0000000000000000
cr0=0000000080010011 cr2=0000000000000000 cr3=0000000007fff000 cr4=0000000000000020
cr8=0000000000000000
	STACK: @400387 40028f
0x0000000000400387: mwait at x86/perf_exit.c:17
        	/* "mwait %eax, %ecx;" */
      > 	asm volatile(".byte 0x0f, 0x01, 0xc9;"
        		:: "a" (eax), "c" (ecx));
 (inlined by) main at x86/perf_exit.c:93
        		mb();
      > 		mwait(0, 0);
        		mb();
0x000000000040028e: start64 at x86/cstart64.S:206
        	lea __argv(%rip), %rsi
      > 	call main
        	mov %eax, %edi

In result 2, the trace shows that vcpu is in 64bit protect-mode (the testcase
didn't jump to ring3 manually, so it's in ring0), as intel vol2 said the MWAIT
will #UD in this situation when:

#UD 	If the current privilege level is not 0.
	If CPUID.01H:ECX.MONITOR[bit 3] = 0.

I think CPUID.01H:ECX.MONITOR does not matter, so I'm so curious about the
result 2 .

Do you have any idea ? Thanks.

PS: Wanpeng tested this on his laptop (maybe the cpu is Core-i7) and it worked
well as result 1.

kvm version : 4.4.11
host kernel : 3.10


The testcase is:

#include "libcflat.h"
#include "smp.h"
#include "vm.h"
#include "processor.h"
#include "desc.h"
#include "apic.h"
#include "apic-defs.h"
#include "asm/barrier.h"

[...]

static inline void mwait(unsigned long eax, unsigned long ecx)
{
        /* "mwait %eax, %ecx;" */
        asm volatile(".byte 0x0f, 0x01, 0xc9;"
                :: "a" (eax), "c" (ecx));
}

[...]

int main(int argc, char **argv)
{
        setup_vm();
        smp_init();

	while (test_cnt--) {
                [...]
                mwait(0, 0);
                [...]
        }
	[...]
}


-- 
Regards,
Longpeng(Mike)