lkml.org 
[lkml]   [2020]   [Dec]   [15]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [PATCH v2] KVM: SVM: use vmsave/vmload for saving/restoring additional host state
On Mon, Dec 14, 2020 at 02:29:46PM -0800, Andy Lutomirski wrote:
>
>
> > On Dec 14, 2020, at 2:02 PM, Michael Roth <michael.roth@amd.com> wrote:
> >
> > On Mon, Dec 14, 2020 at 11:38:23AM -0800, Sean Christopherson wrote:
> >> +Andy, who provided a lot of feedback on v1.
> >> On Mon, Dec 14, 2020, Michael Roth wrote:
> >> Cc: Andy Lutomirski <luto@kernel.org>
> >>> Suggested-by: Tom Lendacky <thomas.lendacky@amd.com>
> >>> Signed-off-by: Michael Roth <michael.roth@amd.com>
> >>> ---
> >>> v2:
> >>> * rebase on latest kvm/next
> >>> * move VMLOAD to just after vmexit so we can use it to handle all FS/GS
> >>> host state restoration and rather than relying on loadsegment() and
> >>> explicit write to MSR_GS_BASE (Andy)
> >>> * drop 'host' field from struct vcpu_svm since it is no longer needed
> >>> for storing FS/GS/LDT state (Andy)
> >>> ---
> >>> arch/x86/kvm/svm/svm.c | 44 ++++++++++++++++--------------------------
> >>> arch/x86/kvm/svm/svm.h | 14 +++-----------
> >>> 2 files changed, 20 insertions(+), 38 deletions(-)
> >>> diff --git a/arch/x86/kvm/svm/svm.c b/arch/x86/kvm/svm/svm.c
> >>> index 0e52fac4f5ae..fb15b7bd461f 100644
> >>> --- a/arch/x86/kvm/svm/svm.c
> >>> +++ b/arch/x86/kvm/svm/svm.c
> >>> @@ -1367,15 +1367,19 @@ static void svm_vcpu_load(struct kvm_vcpu *vcpu, int cpu)
> >>> vmcb_mark_all_dirty(svm->vmcb);
> >>> }
> >>> -#ifdef CONFIG_X86_64
> >>> - rdmsrl(MSR_GS_BASE, to_svm(vcpu)->host.gs_base);
> >>> -#endif
> >>> - savesegment(fs, svm->host.fs);
> >>> - savesegment(gs, svm->host.gs);
> >>> - svm->host.ldt = kvm_read_ldt();
> >>> -
> >>> - for (i = 0; i < NR_HOST_SAVE_USER_MSRS; i++)
> >>> + for (i = 0; i < NR_HOST_SAVE_USER_MSRS; i++) {
> >>> rdmsrl(host_save_user_msrs[i], svm->host_user_msrs[i]);
> >>> + }
> >
> > Hi Sean,
> >
> > Hopefully I've got my email situation sorted out now...
> >
> >> Unnecessary change that violates preferred coding style. Checkpatch explicitly
> >> complains about this.
> >> WARNING: braces {} are not necessary for single statement blocks
> >> #132: FILE: arch/x86/kvm/svm/svm.c:1370:
> >> + for (i = 0; i < NR_HOST_SAVE_USER_MSRS; i++) {
> >> rdmsrl(host_save_user_msrs[i], svm->host_user_msrs[i]);
> >> +
> >
> > Sorry, that was an artifact from an earlier version of the patch that I
> > failed to notice. I'll make sure to run everything through checkpatch
> > going forward.
> >
> >>> +
> >>> + asm volatile(__ex("vmsave")
> >>> + : : "a" (page_to_pfn(sd->save_area) << PAGE_SHIFT)
> >> I'm pretty sure this can be page_to_phys().
> >>> + : "memory");
> >> I think we can defer this until we're actually planning on running the guest,
> >> i.e. put this in svm_prepare_guest_switch().
> >
> > One downside to that is that we'd need to do the VMSAVE on every
> > iteration of vcpu_run(), as opposed to just once when we enter from
> > userspace via KVM_RUN. It ends up being a similar situation to Andy's
> > earlier suggestion of moving VMLOAD just after vmexit, but in that case
> > we were able to remove an MSR write to MSR_GS_BASE, which cancelled out
> > the overhead, but in this case I think it could only cost us extra.
>
> If you want to micro-optimize, there is a trick you could play: use WRGSBASE if available. If X86_FEATURE_GSBASE is available, you could use WRGSBASE to restore GSBASE and defer VMLOAD to vcpu_put(). This would need benchmarking on Zen 3 to see if it’s worthwhile.

I'll give this a try. The vmsave only seems to be 100-200 in and of itself so
I'm not sure there's much to be gained, but would be good to know either
way.

>
>

\
 
 \ /
  Last update: 2020-12-15 19:58    [W:0.112 / U:0.880 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site