linux-kernel - Re: [RFC PATCH v3 14/21] KVM: x86: Kill cur_tsc

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <Zr4tIK5I17NcIxRz@google.com>
Date: Thu, 15 Aug 2024 09:30:24 -0700
From: Sean Christopherson <seanjc@...gle.com>
To: David Woodhouse <dwmw2@...radead.org>
Cc: kvm@...r.kernel.org, Paolo Bonzini <pbonzini@...hat.com>, 
	Jonathan Corbet <corbet@....net>, Thomas Gleixner <tglx@...utronix.de>, Ingo Molnar <mingo@...hat.com>, 
	Borislav Petkov <bp@...en8.de>, Dave Hansen <dave.hansen@...ux.intel.com>, x86@...nel.org, 
	"H. Peter Anvin" <hpa@...or.com>, Paul Durrant <paul@....org>, Peter Zijlstra <peterz@...radead.org>, 
	Juri Lelli <juri.lelli@...hat.com>, Vincent Guittot <vincent.guittot@...aro.org>, 
	Dietmar Eggemann <dietmar.eggemann@....com>, Steven Rostedt <rostedt@...dmis.org>, 
	Ben Segall <bsegall@...gle.com>, Mel Gorman <mgorman@...e.de>, 
	Daniel Bristot de Oliveira <bristot@...hat.com>, Valentin Schneider <vschneid@...hat.com>, Shuah Khan <shuah@...nel.org>, 
	linux-doc@...r.kernel.org, linux-kernel@...r.kernel.org, 
	jalliste@...zon.co.uk, sveith@...zon.de, zide.chen@...el.com, 
	Dongli Zhang <dongli.zhang@...cle.com>, Chenyi Qiang <chenyi.qiang@...el.com>
Subject: Re: [RFC PATCH v3 14/21] KVM: x86: Kill cur_tsc_{nsec,offset,write} fields

On Wed, May 22, 2024, David Woodhouse wrote:
> From: David Woodhouse <dwmw@...zon.co.uk>
> 
> These pointlessly duplicate the last_tsc_{nsec,offset,write} values.
> 
> The only place they were used was where the TSC is stable and a new vCPU
> is being synchronized to the previous setting, in which case the 'last_'
> value is definitely identical.
> 
> Signed-off-by: David Woodhouse <dwmw@...zon.co.uk>
> ---
>  arch/x86/include/asm/kvm_host.h |  3 ---
>  arch/x86/kvm/x86.c              | 19 ++++++++-----------
>  2 files changed, 8 insertions(+), 14 deletions(-)
> 
> diff --git a/arch/x86/include/asm/kvm_host.h b/arch/x86/include/asm/kvm_host.h
> index b01c1d000fff..7d06f389a607 100644
> --- a/arch/x86/include/asm/kvm_host.h
> +++ b/arch/x86/include/asm/kvm_host.h
> @@ -1354,9 +1354,6 @@ struct kvm_arch {
>  	u32 last_tsc_khz;
>  	u64 last_tsc_offset;
>  	u64 last_tsc_scaling_ratio;
> -	u64 cur_tsc_nsec;
> -	u64 cur_tsc_write;
> -	u64 cur_tsc_offset;
>  	u64 cur_tsc_generation;
>  	int nr_vcpus_matched_tsc;
>  
> diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
> index 6ec43f39bdb0..ab5d55071253 100644
> --- a/arch/x86/kvm/x86.c
> +++ b/arch/x86/kvm/x86.c
> @@ -2713,11 +2713,9 @@ static void __kvm_synchronize_tsc(struct kvm_vcpu *vcpu, u64 offset, u64 tsc,
>  	lockdep_assert_held(&kvm->arch.tsc_write_lock);
>  
>  	/*
> -	 * We also track th most recent recorded KHZ, write and time to
> -	 * allow the matching interval to be extended at each write.
> +	 * Track the last recorded kHz (and associated scaling ratio for
> +	 * calculating the guest TSC), and offset.
>  	 */
> -	kvm->arch.last_tsc_nsec = ns;
> -	kvm->arch.last_tsc_write = tsc;
>  	kvm->arch.last_tsc_khz = vcpu->arch.virtual_tsc_khz;
>  	kvm->arch.last_tsc_scaling_ratio = vcpu->arch.l1_tsc_scaling_ratio;
>  	kvm->arch.last_tsc_offset = offset;
> @@ -2736,10 +2734,9 @@ static void __kvm_synchronize_tsc(struct kvm_vcpu *vcpu, u64 offset, u64 tsc,
>  		 *
>  		 * These values are tracked in kvm->arch.cur_xxx variables.

This comment is now stale, as most of the fields are now .last_xxx, not cur_xxx.

However...

>  		 */
> +		kvm->arch.last_tsc_nsec = ns;


There is a functional change here, and it's either incorrect or misleading (I
think the latter).  If the TSC is unstable, "ns" in kvm_synchronize_tsc() will
come from get_kvmclock_base_ns(), and only the TSC frequency is checked for a
match when synchronizing.

That results in .last_tsc_nsec not being updated, and so subsequent syncs will
compute a larger elapsed time (relative to the current generation's timestamp,
not the "last" timestamp).

Functionally, I think that's ok?  So long as all vCPUs sync against the same
baseline, it should work?  I think.

But if that's the case, then I would prefer to delete last_tsc_{nsec,write,offset},
not the cur_xxx versions.  For nsec and write it shows that they are valid/used
only in the context of the current generation.

And for the offset, updating it _outside_ of the loop makes it more obvious that
the offset can change (by design) within a generation if the TSC is unstable.

Ooh, and if I'm reading the code correctly, last_tsc_khz can be renamed to
cur_tsc_khz and moved in the !matched statement too, as it's guaranteed to be
vcpu->arch.virtual_tsc_khz if matched==true.

Ah, right, and last_tsc_scaling_ratio is just an deriviation of virtual_tsc_khz,
so it too can be cur_xxx and put under !matched.

Am I missing something?  That seems too easy...