On 5/2/2025 4:53 AM, mlevitsk@xxxxxxxxxx wrote:
> On Thu, 2025-05-01 at 16:41 -0400, mlevitsk@xxxxxxxxxx wrote:
>> On Tue, 2025-04-22 at 16:41 -0700, Sean Christopherson wrote:
>>> On Tue, Apr 15, 2025, Maxim Levitsky wrote:
>>>> Pass through the host's DEBUGCTL.DEBUGCTLMSR_FREEZE_IN_SMM to the guest
>>>> GUEST_IA32_DEBUGCTL without the guest seeing this value.
>>>>
>>>> Note that in the future we might allow the guest to set this bit as well,
>>>> when we implement PMU freezing on VM own, virtual SMM entry.
>>>>
>>>> Since the value of the host DEBUGCTL can in theory change between VM runs,
>>>> check if has changed, and if yes, then reload the GUEST_IA32_DEBUGCTL with
>>>> the new value of the host portion of it (currently only the
>>>> DEBUGCTLMSR_FREEZE_IN_SMM bit)
>>> No, it can't. DEBUGCTLMSR_FREEZE_IN_SMM can be toggled via IPI callback, but
>>> IRQs are disabled for the entirety of the inner run loop. And if I'm somehow
>>> wrong, this change movement absolutely belongs in a separate patch.
>
> Hi,
>
> You are right here - reading MSR_IA32_DEBUGCTLMSR in the inner loop is a performance
> regression.
>
>
> Any ideas on how to solve this then? Since currently its the common code that
> reads the current value of the MSR_IA32_DEBUGCTLMSR and it doesn't leave any indication
> about if it changed I can do either
>
> 1. store old value as well, something like 'vcpu->arch.host_debugctl_old' Ugly IMHO.
>
> 2. add DEBUG_CTL to the set of the 'dirty' registers, e.g add new bit for kvm_register_mark_dirty
> It looks a bit overkill to me
>
> 3. Add new x86 callback for something like .sync_debugctl(). I vote for this option.
>
> What do you think/prefer?
Hmm, not sure if I missed something, but why to move the reading host
debug_ctrl MSR from the original place into inner loop? The interrupt has
been disabled before reading host debug_ctrl for original code, suppose
host debug_ctrl won't changed after reading it?
>
> Best regards,
> Maxim Levitsky
>
>>>> Signed-off-by: Maxim Levitsky <mlevitsk@xxxxxxxxxx>
>>>> ---
>>>> arch/x86/kvm/svm/svm.c | 2 ++
>>>> arch/x86/kvm/vmx/vmx.c | 28 +++++++++++++++++++++++++++-
>>>> arch/x86/kvm/x86.c | 2 --
>>>> 3 files changed, 29 insertions(+), 3 deletions(-)
>>>>
>>>> diff --git a/arch/x86/kvm/svm/svm.c b/arch/x86/kvm/svm/svm.c
>>>> index cc1c721ba067..fda0660236d8 100644
>>>> --- a/arch/x86/kvm/svm/svm.c
>>>> +++ b/arch/x86/kvm/svm/svm.c
>>>> @@ -4271,6 +4271,8 @@ static __no_kcsan fastpath_t svm_vcpu_run(struct kvm_vcpu *vcpu,
>>>> svm->vmcb->save.rsp = vcpu->arch.regs[VCPU_REGS_RSP];
>>>> svm->vmcb->save.rip = vcpu->arch.regs[VCPU_REGS_RIP];
>>>>
>>>> + vcpu->arch.host_debugctl = get_debugctlmsr();
>>>> +
>>>> /*
>>>> * Disable singlestep if we're injecting an interrupt/exception.
>>>> * We don't want our modified rflags to be pushed on the stack where
>>>> diff --git a/arch/x86/kvm/vmx/vmx.c b/arch/x86/kvm/vmx/vmx.c
>>>> index c9208a4acda4..e0bc31598d60 100644
>>>> --- a/arch/x86/kvm/vmx/vmx.c
>>>> +++ b/arch/x86/kvm/vmx/vmx.c
>>>> @@ -2194,6 +2194,17 @@ static u64 vmx_get_supported_debugctl(struct kvm_vcpu *vcpu, bool host_initiated
>>>> return debugctl;
>>>> }
>>>>
>>>> +static u64 vmx_get_host_preserved_debugctl(struct kvm_vcpu *vcpu)
>>> No, just open code handling DEBUGCTLMSR_FREEZE_IN_SMM, or make it a #define.
>>> I'm not remotely convinced that we'll ever want to emulate DEBUGCTLMSR_FREEZE_IN_SMM,
>>> and trying to plan for that possibility and adds complexity for no immediate value.
>> Hi,
>>
>> The problem here is a bit different: we indeed are very unlikely to emulate the
>> DEBUGCTLMSR_FREEZE_IN_SMM but however, when I wrote this patch I was sure that this bit is
>> mandatory with PMU version of 2 or more, but looks like it is optional after all:
>>
>> "
>> Note that system software must check if the processor supports the IA32_DEBUGCTL.FREEZE_WHILE_SMM
>> control bit. IA32_DEBUGCTL.FREEZE_WHILE_SMM is supported if IA32_PERF_CAPABIL-
>> ITIES.FREEZE_WHILE_SMM[Bit 12] is reporting 1. See Section 20.8 for details of detecting the presence of
>> IA32_PERF_CAPABILITIES MSR."
>>
>> KVM indeed doesn't set the bit 12 of IA32_PERF_CAPABILITIES.
>>
>> However, note that the Linux kernel silently sets this bit without checking the aforementioned capability
>> bit and ends up with a #GP exception, which it silently ignores.... (I checked this with a trace...)
>>
>> This led me to believe that this bit should be unconditionally supported,
>> meaning that KVM should at least fake setting it without triggering a #GP.
>>
>> Since that is not the case, I can revert to the simpler model of exclusively using GUEST_IA32_DEBUGCTL
>> while hiding the bit from the guest, however I do vote to keep the guest/host separation.
>>
>>>> +{
>>>> + /*
>>>> + * Bits of host's DEBUGCTL that we should preserve while the guest is
>>>> + * running.
>>>> + *
>>>> + * Some of those bits might still be emulated for the guest own use.
>>>> + */
>>>> + return DEBUGCTLMSR_FREEZE_IN_SMM;
>>>>
>>>> u64 vmx_get_guest_debugctl(struct kvm_vcpu *vcpu)
>>>> {
>>>> return to_vmx(vcpu)->msr_ia32_debugctl;
>>>> @@ -2202,9 +2213,11 @@ u64 vmx_get_guest_debugctl(struct kvm_vcpu *vcpu)
>>>> static void __vmx_set_guest_debugctl(struct kvm_vcpu *vcpu, u64 data)
>>>> {
>>>> struct vcpu_vmx *vmx = to_vmx(vcpu);
>>>> + u64 host_mask = vmx_get_host_preserved_debugctl(vcpu);
>>>>
>>>> vmx->msr_ia32_debugctl = data;
>>>> - vmcs_write64(GUEST_IA32_DEBUGCTL, data);
>>>> + vmcs_write64(GUEST_IA32_DEBUGCTL,
>>>> + (vcpu->arch.host_debugctl & host_mask) | (data & ~host_mask));
>>>> }
>>>>
>>>> bool vmx_set_guest_debugctl(struct kvm_vcpu *vcpu, u64 data, bool host_initiated)
>>>> @@ -2232,6 +2245,7 @@ bool vmx_set_guest_debugctl(struct kvm_vcpu *vcpu, u64 data, bool host_initiated
>>>> return true;
>>>> }
>>>>
>>>> +
>>> Spurious newline.
>>>
>>>> /*
>>>> * Writes msr value into the appropriate "register".
>>>> * Returns 0 on success, non-0 otherwise.
>>>> @@ -7349,6 +7363,7 @@ fastpath_t vmx_vcpu_run(struct kvm_vcpu *vcpu, bool force_immediate_exit)
>>>> {
>>>> struct vcpu_vmx *vmx = to_vmx(vcpu);
>>>> unsigned long cr3, cr4;
>>>> + u64 old_debugctl;
>>>>
>>>> /* Record the guest's net vcpu time for enforced NMI injections. */
>>>> if (unlikely(!enable_vnmi &&
>>>> @@ -7379,6 +7394,17 @@ fastpath_t vmx_vcpu_run(struct kvm_vcpu *vcpu, bool force_immediate_exit)
>>>> vmcs_write32(PLE_WINDOW, vmx->ple_window);
>>>> }
>>>>
>>>> + old_debugctl = vcpu->arch.host_debugctl;
>>>> + vcpu->arch.host_debugctl = get_debugctlmsr();
>>>> +
>>>> + /*
>>>> + * In case the host DEBUGCTL had changed since the last time we
>>>> + * read it, update the guest's GUEST_IA32_DEBUGCTL with
>>>> + * the host's bits.
>>>> + */
>>>> + if (old_debugctl != vcpu->arch.host_debugctl)
>>> This can and should be optimized to only do an update if a host-preserved bit
>>> is toggled.
>> True, I will do this in the next version.
>>
>>>> + __vmx_set_guest_debugctl(vcpu, vmx->msr_ia32_debugctl);
>>> I would rather have a helper that explicitly writes the VMCS field, not one that
>>> sets the guest value *and* writes the VMCS field.
>>> The usage in init_vmcs() doesn't need to write vmx->msr_ia32_debugctl because the
>>> vCPU is zero allocated, and this usage doesn't change vmx->msr_ia32_debugctl.
>>> So the only path that actually needs to modify vmx->msr_ia32_debugctl is
>>> vmx_set_guest_debugctl().
>>
>> But what about nested entry? nested entry pretty much sets the MSR to a value given by the guest.
>>
>> Also technically the intel_pmu_legacy_freezing_lbrs_on_pmi also changes the guest value by emulating what the real hardware does.
>>
>> Best regards,
>> Maxim Levitsky
>>
>>
>>>> +
>>>> /*
>>>> * We did this in prepare_switch_to_guest, because it needs to
>>>> * be within srcu_read_lock.
>>>> diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
>>>> index 844e81ee1d96..05e866ed345d 100644
>>>> --- a/arch/x86/kvm/x86.c
>>>> +++ b/arch/x86/kvm/x86.c
>>>> @@ -11020,8 +11020,6 @@ static int vcpu_enter_guest(struct kvm_vcpu *vcpu)
>>>> set_debugreg(0, 7);
>>>> }
>>>>
>>>> - vcpu->arch.host_debugctl = get_debugctlmsr();
>>>> -
>>>> guest_timing_enter_irqoff();
>>>>
>>>> for (;;) {
>>>> --
>>>> 2.26.3
>>>>
>
