From: "Stamatis, Ilias" <ilstam@amazon.com>
To: "kvm@vger.kernel.org" <kvm@vger.kernel.org>,
"pbonzini@redhat.com" <pbonzini@redhat.com>,
"mlevitsk@redhat.com" <mlevitsk@redhat.com>,
"ilstam@mailbox.org" <ilstam@mailbox.org>
Cc: "jmattson@google.com" <jmattson@google.com>,
"Woodhouse, David" <dwmw@amazon.co.uk>,
"vkuznets@redhat.com" <vkuznets@redhat.com>,
"joro@8bytes.org" <joro@8bytes.org>,
"mtosatti@redhat.com" <mtosatti@redhat.com>,
"zamsden@gmail.com" <zamsden@gmail.com>,
"seanjc@google.com" <seanjc@google.com>,
"wanpengli@tencent.com" <wanpengli@tencent.com>
Subject: Re: [PATCH 3/8] KVM: X86: Pass an additional 'L1' argument to kvm_scale_tsc()
Date: Mon, 10 May 2021 15:44:31 +0000 [thread overview]
Message-ID: <041e087ab930f33cff5563204c79438368c9d694.camel@amazon.com> (raw)
In-Reply-To: <b87ca34b3251f06c807e5d46bbf821756e57ff5b.camel@redhat.com>
On Mon, 2021-05-10 at 16:52 +0300, Maxim Levitsky wrote:
> On Thu, 2021-05-06 at 10:32 +0000, ilstam@mailbox.org wrote:
> > From: Ilias Stamatis <ilstam@amazon.com>
> >
> > Sometimes kvm_scale_tsc() needs to use the current scaling ratio and
> > other times (like when reading the TSC from user space) it needs to use
> > L1's scaling ratio. Have the caller specify this by passing an
> > additional boolean argument.
> >
> > Signed-off-by: Ilias Stamatis <ilstam@amazon.com>
> > ---
> > arch/x86/include/asm/kvm_host.h | 2 +-
> > arch/x86/kvm/x86.c | 21 +++++++++++++--------
> > 2 files changed, 14 insertions(+), 9 deletions(-)
> >
> > diff --git a/arch/x86/include/asm/kvm_host.h b/arch/x86/include/asm/kvm_host.h
> > index 132e820525fb..cdddbf0b1177 100644
> > --- a/arch/x86/include/asm/kvm_host.h
> > +++ b/arch/x86/include/asm/kvm_host.h
> > @@ -1779,7 +1779,7 @@ int kvm_pv_send_ipi(struct kvm *kvm, unsigned long ipi_bitmap_low,
> > void kvm_define_user_return_msr(unsigned index, u32 msr);
> > int kvm_set_user_return_msr(unsigned index, u64 val, u64 mask);
> >
> > -u64 kvm_scale_tsc(struct kvm_vcpu *vcpu, u64 tsc);
> > +u64 kvm_scale_tsc(struct kvm_vcpu *vcpu, u64 tsc, bool l1);
> > u64 kvm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc);
> >
> > unsigned long kvm_get_linear_rip(struct kvm_vcpu *vcpu);
> > diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
> > index 7bc5155ac6fd..26a4c0f46f15 100644
> > --- a/arch/x86/kvm/x86.c
> > +++ b/arch/x86/kvm/x86.c
> > @@ -2241,10 +2241,14 @@ static inline u64 __scale_tsc(u64 ratio, u64 tsc)
> > return mul_u64_u64_shr(tsc, ratio, kvm_tsc_scaling_ratio_frac_bits);
> > }
> >
> > -u64 kvm_scale_tsc(struct kvm_vcpu *vcpu, u64 tsc)
> > +/*
> > + * If l1 is true the TSC is scaled using L1's scaling ratio, otherwise
> > + * the current scaling ratio is used.
> > + */
> > +u64 kvm_scale_tsc(struct kvm_vcpu *vcpu, u64 tsc, bool l1)
> > {
> > u64 _tsc = tsc;
> > - u64 ratio = vcpu->arch.tsc_scaling_ratio;
> > + u64 ratio = l1 ? vcpu->arch.l1_tsc_scaling_ratio : vcpu->arch.tsc_scaling_ratio;
> >
> > if (ratio != kvm_default_tsc_scaling_ratio)
> > _tsc = __scale_tsc(ratio, tsc);
>
> I do wonder if it is better to have kvm_scale_tsc_l1 and kvm_scale_tsc instead for better
> readablility?
>
That makes sense. Will do.
>
> > @@ -2257,14 +2261,14 @@ static u64 kvm_compute_tsc_offset(struct kvm_vcpu *vcpu, u64 target_tsc)
> > {
> > u64 tsc;
> >
> > - tsc = kvm_scale_tsc(vcpu, rdtsc());
> > + tsc = kvm_scale_tsc(vcpu, rdtsc(), true);
>
> Here we have a somewhat dangerous assumption that this function
> will always be used with L1 tsc values.
>
> The kvm_compute_tsc_offset should at least be renamed to
> kvm_compute_tsc_offset_l1 or something like that.
>
> Currently the assumption holds though:
>
> We call the kvm_compute_tsc_offset in:
>
> -> kvm_synchronize_tsc which is nowadays thankfully only called
> on TSC writes from qemu, which are assumed to be L1 values.
>
> (this is pending a rework of the whole thing which I started
> some time ago but haven't had a chance to finish it yet)
>
> -> Guest write of MSR_IA32_TSC. The value written is in L1 units,
> since TSC offset/scaling only covers RDTSC and RDMSR of the IA32_TSC,
> while WRMSR should be intercepted by L1 and emulated.
> If it is not emulated, then L2 would just write L1 value.
>
> -> in kvm_arch_vcpu_load, when TSC is unstable, we always try to resume
> the guest from the same TSC value as it had seen last time,
> and then catchup.
Yes. I wasn't sure about kvm_compute_tsc_offset but my understanding was
that all of its callers wanted the L1 value scaled.
Renaming it to kvm_scale_tsc_l1 sounds like a great idea.
> Also host TSC values are used, and after reading this function,
> I recommend to rename the vcpu->arch.last_guest_tsc
> to vcpu->arch.last_guest_tsc_l1 to document this.
OK
> >
> > return target_tsc - tsc;
> > }
> >
> > u64 kvm_read_l1_tsc(struct kvm_vcpu *vcpu, u64 host_tsc)
> > {
> > - return vcpu->arch.l1_tsc_offset + kvm_scale_tsc(vcpu, host_tsc);
> > + return vcpu->arch.l1_tsc_offset + kvm_scale_tsc(vcpu, host_tsc, true);
>
> OK
> > }
> > EXPORT_SYMBOL_GPL(kvm_read_l1_tsc);
> >
> > @@ -2395,9 +2399,9 @@ static inline void adjust_tsc_offset_guest(struct kvm_vcpu *vcpu,
> >
> > static inline void adjust_tsc_offset_host(struct kvm_vcpu *vcpu, s64 adjustment)
> > {
> > - if (vcpu->arch.tsc_scaling_ratio != kvm_default_tsc_scaling_ratio)
> > + if (vcpu->arch.l1_tsc_scaling_ratio != kvm_default_tsc_scaling_ratio)
> > WARN_ON(adjustment < 0);
>
> This should belong to patch 2 IMHO.
>
Right, I will move it.
> > - adjustment = kvm_scale_tsc(vcpu, (u64) adjustment);
> > + adjustment = kvm_scale_tsc(vcpu, (u64) adjustment, true);
>
> OK
> > adjust_tsc_offset_guest(vcpu, adjustment);
> > }
> >
> > @@ -2780,7 +2784,7 @@ static int kvm_guest_time_update(struct kvm_vcpu *v)
> > /* With all the info we got, fill in the values */
> >
> > if (kvm_has_tsc_control)
> > - tgt_tsc_khz = kvm_scale_tsc(v, tgt_tsc_khz);
> > + tgt_tsc_khz = kvm_scale_tsc(v, tgt_tsc_khz, true);
>
> OK (kvmclock is for L1 only, L1 hypervisor is free to expose its own kvmclock to L2)
> >
> > if (unlikely(vcpu->hw_tsc_khz != tgt_tsc_khz)) {
> > kvm_get_time_scale(NSEC_PER_SEC, tgt_tsc_khz * 1000LL,
> > @@ -3474,7 +3478,8 @@ int kvm_get_msr_common(struct kvm_vcpu *vcpu, struct msr_data *msr_info)
> > u64 tsc_offset = msr_info->host_initiated ? vcpu->arch.l1_tsc_offset :
> > vcpu->arch.tsc_offset;
> >
> > - msr_info->data = kvm_scale_tsc(vcpu, rdtsc()) + tsc_offset;
> > + msr_info->data = kvm_scale_tsc(vcpu, rdtsc(),
> > + msr_info->host_initiated) + tsc_offset;
>
> Since we now do two things that depend on msr_info->host_initiated, I
> think I would prefer to convert this back to regular 'if'.
> I don't have a strong opinion on this though.
>
Agreed.
Thanks!
Ilias
>
> > break;
> > }
> > case MSR_MTRRcap:
>
>
> Best regards,
> Maxim Levitsky
>
>
next prev parent reply other threads:[~2021-05-10 15:44 UTC|newest]
Thread overview: 36+ messages / expand[flat|nested] mbox.gz Atom feed top
2021-05-06 10:32 [PATCH 0/8] KVM: VMX: Implement nested TSC scaling ilstam
2021-05-06 10:32 ` [PATCH 1/8] KVM: VMX: Add a TSC multiplier field in VMCS12 ilstam
2021-05-06 14:50 ` kernel test robot
2021-05-06 17:36 ` Jim Mattson
2021-05-10 13:42 ` Maxim Levitsky
2021-05-06 10:32 ` [PATCH 2/8] KVM: X86: Store L1's TSC scaling ratio in 'struct kvm_vcpu_arch' ilstam
2021-05-10 13:43 ` Maxim Levitsky
2021-05-06 10:32 ` [PATCH 3/8] KVM: X86: Pass an additional 'L1' argument to kvm_scale_tsc() ilstam
2021-05-10 13:52 ` Maxim Levitsky
2021-05-10 15:44 ` Stamatis, Ilias [this message]
2021-05-06 10:32 ` [PATCH 4/8] KVM: VMX: Adjust the TSC-related VMCS fields on L2 entry and exit ilstam
2021-05-06 11:32 ` Paolo Bonzini
2021-05-06 17:35 ` Stamatis, Ilias
2021-05-10 14:11 ` Paolo Bonzini
2021-05-10 13:53 ` Maxim Levitsky
2021-05-10 14:44 ` Stamatis, Ilias
2021-05-11 12:38 ` Maxim Levitsky
2021-05-11 15:11 ` Stamatis, Ilias
2021-05-06 10:32 ` [PATCH 5/8] KVM: X86: Move tracing outside write_l1_tsc_offset() ilstam
2021-05-10 13:54 ` Maxim Levitsky
2021-05-06 10:32 ` [PATCH 6/8] KVM: VMX: Make vmx_write_l1_tsc_offset() work with nested TSC scaling ilstam
2021-05-10 13:54 ` Maxim Levitsky
2021-05-10 16:08 ` Stamatis, Ilias
2021-05-11 12:44 ` Maxim Levitsky
2021-05-11 17:44 ` Stamatis, Ilias
2021-05-06 10:32 ` [PATCH 7/8] KVM: VMX: Expose TSC scaling to L2 ilstam
2021-05-10 13:56 ` Maxim Levitsky
2021-05-06 10:32 ` [PATCH 8/8] KVM: selftests: x86: Add vmx_nested_tsc_scaling_test ilstam
2021-05-10 13:59 ` Maxim Levitsky
2021-05-11 11:16 ` Stamatis, Ilias
2021-05-11 12:47 ` Maxim Levitsky
2021-05-11 14:02 ` Stamatis, Ilias
2021-05-06 17:16 ` [PATCH 0/8] KVM: VMX: Implement nested TSC scaling Jim Mattson
2021-05-06 17:48 ` Stamatis, Ilias
2021-05-10 13:43 ` Maxim Levitsky
2021-05-10 14:29 ` Paolo Bonzini
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=041e087ab930f33cff5563204c79438368c9d694.camel@amazon.com \
--to=ilstam@amazon.com \
--cc=dwmw@amazon.co.uk \
--cc=ilstam@mailbox.org \
--cc=jmattson@google.com \
--cc=joro@8bytes.org \
--cc=kvm@vger.kernel.org \
--cc=mlevitsk@redhat.com \
--cc=mtosatti@redhat.com \
--cc=pbonzini@redhat.com \
--cc=seanjc@google.com \
--cc=vkuznets@redhat.com \
--cc=wanpengli@tencent.com \
--cc=zamsden@gmail.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).