Re: [PATCH V2 0/3] improve -overcommit cpu-pm=on|off

["Chen, Zide" <zide.chen@intel.com>]

On 5/29/2024 5:46 AM, Igor Mammedov wrote:
> On Tue, 28 May 2024 11:16:59 -0700
> "Chen, Zide" <zide.chen@intel.com> wrote:
> 
>> On 5/28/2024 2:23 AM, Igor Mammedov wrote:
>>> On Fri, 24 May 2024 13:00:14 -0700
>>> Zide Chen <zide.chen@intel.com> wrote:
>>>   
>>>> Currently, if running "-overcommit cpu-pm=on" on hosts that don't
>>>> have MWAIT support, the MWAIT/MONITOR feature is advertised to the
>>>> guest and executing MWAIT/MONITOR on the guest triggers #UD.  
>>>
>>> this is missing proper description how do you trigger issue
>>> with reproducer and detailed description why guest sees MWAIT
>>> when it's not supported by host.  
>>
>> If "overcommit cpu-pm=on" and "-cpu host" are present, as shown in the
> it's bette to provide full QEMU CLI and host/guest kernels used and what
> hardware was used if it's relevant so others can reproduce problem.

I ever reproduced this on an older Intel Icelake machine, a
Sapphire Rapids and a Sierra Forest, but I believe this is a x86 generic
issue, not specific to particular models.

For the CLI, I think the only command line options that matter are
 -overcommit cpu-pm=on: to set enable_cpu_pm
 -cpu host: so that cpu->max_features is set

For QEMU version, as long as it's after this commit: 662175b91ff2
("i386: reorder call to cpu_exec_realizefn")

The guest fails to boot:

[ 24.825568] smpboot: x86: Booting SMP configuration:
[ 24.826377] .... node #0, CPUs: #1 #2 #3 #4 #5 #6 #7 #8 #9 #10 #11 #12
#13 #14 #15 #17
[ 24.985799] .... node #1, CPUs: #128 #129 #130 #131 #132 #133 #134 #135
#136 #137 #138 #139 #140 #141 #142 #143 #145
[ 25.136955] invalid opcode: 0000 1 PREEMPT SMP NOPTI
[ 25.137790] CPU: 0 PID: 0 Comm: swapper/0 Not tainted 6.8.0 #2
[ 25.137790] Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS
rel-1.16.3-0-ga6ed6b701f0a-prebuilt.qemu.org 04/04
[ 25.137790] RIP: 0010:mwait_idle+0x35/0x80
[ 25.137790] Code: 6f f0 80 48 02 20 48 8b 10 83 e2 08 75 3e 65 48 8b 15
47 d6 56 6f 48 0f ba e2 27 72 41 31 d2 48 89 d8
[ 25.137790] RSP: 0000:ffffffff91403e70 EFLAGS: 00010046
[ 25.137790] RAX: ffffffff9140a980 RBX: ffffffff9140a980 RCX:
0000000000000000
[ 25.137790] RDX: 0000000000000000 RSI: ffff97f1ade21b20 RDI:
0000000000000004
[ 25.137790] RBP: 0000000000000000 R08: 00000005da4709cb R09:
0000000000000001
[ 25.137790] R10: 0000000000005da4 R11: 0000000000000009 R12:
0000000000000000
[ 25.137790] R13: ffff98573ff90fc0 R14: ffffffff9140a038 R15:
0000000000093ff0
[ 25.137790] FS: 0000000000000000(0000) GS:ffff97f1ade00000(0000)
knlGS:0000000000000000
[ 25.137790] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[ 25.137790] CR2: ffff97d8aa801000 CR3: 00000049e9430001 CR4:
0000000000770ef0
[ 25.137790] DR0: 0000000000000000 DR1: 0000000000000000 DR2:
0000000000000000
[ 25.137790] DR3: 0000000000000000 DR6: 00000000ffff07f0 DR7:
0000000000000400
[ 25.137790] PKRU: 55555554
[ 25.137790] Call Trace:
[ 25.137790] <TASK>
[ 25.137790] ? die+0x37/0x90
[ 25.137790] ? do_trap+0xe3/0x110
[ 25.137790] ? mwait_idle+0x35/0x80
[ 25.137790] ? do_error_trap+0x6a/0x90
[ 25.137790] ? mwait_idle+0x35/0x80
[ 25.137790] ? exc_invalid_op+0x52/0x70
[ 25.137790] ? mwait_idle+0x35/0x80
[ 25.137790] ? asm_exc_invalid_op+0x1a/0x20
[ 25.137790] ? mwait_idle+0x35/0x80
[ 25.137790] default_idle_call+0x30/0x100
[ 25.137790] cpuidle_idle_call+0x12c/0x170
[ 25.137790] ? tsc_verify_tsc_adjust+0x73/0xd0
[ 25.137790] do_idle+0x7f/0xd0
[ 25.137790] cpu_startup_entry+0x29/0x30
[ 25.137790] rest_init+0xcc/0xd0
[ 25.137790] start_kernel+0x396/0x5d0
[ 25.137790] x86_64_start_reservations+0x18/0x30
[ 25.137790] x86_64_start_kernel+0xe7/0xf0
[ 25.137790] common_startup_64+0x13e/0x148
[ 25.137790] </TASK>
[ 25.137790] Modules linked in:
[ 25.137790] --[ end trace 0000000000000000 ]--
[ 25.137790] invalid opcode: 0000 2 PREEMPT SMP NOPTI
[ 25.137790] RIP: 0010:mwait_idle+0x35/0x80
[ 25.137790] Code: 6f f0 80 48 02 20 48 8b 10 83 e2 08 75 3e 65 48 8b 15
47 d6 56 6f 48 0f ba e2 27 72 41 31 d2 48 89 d8

> 
>> following, CPUID_EXT_MONITOR is set after x86_cpu_filter_features(), so
>> that it doesn't have a chance to check MWAIT against host features and
>> will be advertised to the guest regardless of whether it's supported by
>> the host or not.
>>
>> x86_cpu_realizefn()
>>   x86_cpu_filter_features()
>>   cpu_exec_realizefn()
>>     kvm_cpu_realizefn
>>       host_cpu_realizefn
>>         host_cpu_enable_cpu_pm
>>           env->features[FEAT_1_ECX] |= CPUID_EXT_MONITOR;
>>
>>
>> If it's not supported by the host, executing MONITOR or MWAIT
>> instructions from the guest triggers #UD, no matter MWAIT_EXITING
>> control is set or not.
> 
> If I recall right, kvm was able to emulate mwait/monitor.
> So question is why it leads to exception instead?

KVM can come to play only iff it can trigger MWAIT/MONITOR VM exits. I
didn't find explicit proof from Intel SDM that #UD exceptions take
precedence over MWAIT/MONITOR VM exits, but this is my speculation. For
example, in ancient machines which don't support MWAIT yet, the only way
it can do is #UD, not MWAIT VM exit?