From: David Matlack <dmatlack@google.com>
To: Sean Christopherson <seanjc@google.com>
Cc: kvm@vger.kernel.org, Ben Gardon <bgardon@google.com>,
Joerg Roedel <joro@8bytes.org>, Jim Mattson <jmattson@google.com>,
Wanpeng Li <wanpengli@tencent.com>,
Vitaly Kuznetsov <vkuznets@redhat.com>,
Paolo Bonzini <pbonzini@redhat.com>,
Junaid Shahid <junaids@google.com>,
Andrew Jones <drjones@redhat.com>
Subject: Re: [PATCH 5/8] KVM: x86/mmu: Also record spteps in shadow_page_walk
Date: Mon, 14 Jun 2021 23:39:17 +0000 [thread overview]
Message-ID: <YMfopaDSRKvlsH0Y@google.com> (raw)
In-Reply-To: <YMffQriMoxWw2V1f@google.com>
On Mon, Jun 14, 2021 at 10:59:14PM +0000, Sean Christopherson wrote:
> On Fri, Jun 11, 2021, David Matlack wrote:
> > In order to use walk_shadow_page_lockless() in fast_page_fault() we need
> > to also record the spteps.
> >
> > No functional change intended.
> >
> > Signed-off-by: David Matlack <dmatlack@google.com>
> > ---
> > arch/x86/kvm/mmu/mmu.c | 1 +
> > arch/x86/kvm/mmu/mmu_internal.h | 3 +++
> > arch/x86/kvm/mmu/tdp_mmu.c | 1 +
> > 3 files changed, 5 insertions(+)
> >
> > diff --git a/arch/x86/kvm/mmu/mmu.c b/arch/x86/kvm/mmu/mmu.c
> > index 8140c262f4d3..765f5b01768d 100644
> > --- a/arch/x86/kvm/mmu/mmu.c
> > +++ b/arch/x86/kvm/mmu/mmu.c
> > @@ -3538,6 +3538,7 @@ static bool walk_shadow_page_lockless(struct kvm_vcpu *vcpu, u64 addr,
> > spte = mmu_spte_get_lockless(it.sptep);
> > walk->last_level = it.level;
> > walk->sptes[it.level] = spte;
> > + walk->spteps[it.level] = it.sptep;
> >
> > if (!is_shadow_present_pte(spte))
> > break;
> > diff --git a/arch/x86/kvm/mmu/mmu_internal.h b/arch/x86/kvm/mmu/mmu_internal.h
> > index 26da6ca30fbf..0fefbd5d6c95 100644
> > --- a/arch/x86/kvm/mmu/mmu_internal.h
> > +++ b/arch/x86/kvm/mmu/mmu_internal.h
> > @@ -178,6 +178,9 @@ struct shadow_page_walk {
> >
> > /* The spte value at each level. */
> > u64 sptes[PT64_ROOT_MAX_LEVEL + 1];
> > +
> > + /* The spte pointers at each level. */
> > + u64 *spteps[PT64_ROOT_MAX_LEVEL + 1];
>
> Hrm. I'm not sure how I feel about this patch, or about shadow_page_walk in
> general. On the one hand, I like having a common API. On the other hand, I
> really don't like mixing two different protection schemes, e.g. this really
> should be
>
> tdp_ptep_t spteps;
>
> in order to gain the RCU annotations for TDP, but those RCU annotations are
> problematic because the legacy MMU doesn't use RCU to protect its page tables.
>
> I also don't like forcing the caller to hold the "lock" for longer than is
> necessary, e.g. get_mmio_spte() doesn't require access to the page tables after
> the initial walk, but the common spteps[] kinda sorta forces its hand.
Yeah this felt gross implementing. I like your idea to create separate
APIs instead.
>
> The two use cases (and the only common use cases I can see) have fairly different
> requirements. The MMIO check wants the SPTEs at _all_ levels, whereas the fast
> page fault handler wants the SPTE _and_ its pointer at a single level. So I
> wonder if by providing a super generic API we'd actually increase complexity.
>
> I.e. rather than provide a completely generic API, maybe it would be better to
> have two distinct API. That wouldn't fix the tdp_ptep_t issue, but it would at
> least bound it to some degree and make the code more obvious.
Does the tdp_ptep_t issue go away if kvm_tdp_mmu_get_spte_lockless
returns an rcu_dereference'd version of the pointer? See below.
> I suspect it would
> also reduce the code churn, though that's not necessarily a goal in and of itself.
Makes sense. So keep walk_shadow_page_lockless_{begin,end}() as a
generic API but provide separate helper functions for get_mmio_spte and
fast_page_fault to get each exactly what each needs.
>
> E.g. for fast_page_fault():
>
> walk_shadow_page_lockless_begin(vcpu);
>
> do {
> sptep = get_spte_lockless(..., &spte);
> if (!is_shadow_present_pte(spte))
> break;
>
> sp = sptep_to_sp(sptep);
> if (!is_last_spte(spte, sp->role.level))
> break;
>
> ...
> } while(true);
>
> walk_shadow_page_lockless_end(vcpu);
>
>
> and for get_mmio_spte():
> walk_shadow_page_lockless_begin(vcpu);
> leaf = get_sptes_lockless(vcpu, addr, sptes, &root);
> if (unlikely(leaf < 0)) {
> *sptep = 0ull;
> return reserved;
> }
>
> walk_shadow_page_lockless_end(vcpu);
>
>
> And if we look at the TDP MMU implementations, they aren't sharing _that_ much
> code, and the code that is shared isn't all that interesting, e.g. if we really
> wanted to we could macro-magic away the boilerplate, but I think even I would
> balk at the result :-)
Agreed.
>
> int kvm_tdp_mmu_get_sptes_lockless(struct kvm_vcpu *vcpu, u64 addr, u64 *sptes,
> int *root_level)
> {
> struct tdp_iter iter;
> struct kvm_mmu *mmu = vcpu->arch.mmu;
> gfn_t gfn = addr >> PAGE_SHIFT;
> int leaf = -1;
>
> *root_level = vcpu->arch.mmu->shadow_root_level;
>
> tdp_mmu_for_each_pte(iter, mmu, gfn, gfn + 1) {
> leaf = iter.level;
> sptes[leaf] = iter.old_spte;
> }
>
> return leaf;
> }
>
> u64 *kvm_tdp_mmu_get_spte_lockless(struct kvm_vcpu *vcpu, u64 addr, u64 *spte)
> {
> struct kvm_mmu *mmu = vcpu->arch.mmu;
> gfn_t gfn = addr >> PAGE_SHIFT;
> struct tdp_iter iter;
> u64 *sptep = NULL;
>
> *spte = 0ull;
>
> tdp_mmu_for_each_pte(iter, mmu, gfn, gfn + 1) {
> /*
> * Here be a comment about the unfortunate differences between
> * the TDP MMU and the legacy MMU.
> */
> sptep = (u64 * __force)iter.sptep;
Instead, should this be:
sptep = rcu_dereference(iter.sptep);
?
> *spte = iter.old_spte;
> }
> return sptep;
> }
>
next prev parent reply other threads:[~2021-06-14 23:40 UTC|newest]
Thread overview: 32+ messages / expand[flat|nested] mbox.gz Atom feed top
2021-06-11 23:56 [PATCH 0/8] KVM: x86/mmu: Fast page fault support for the TDP MMU David Matlack
2021-06-11 23:56 ` [PATCH 1/8] KVM: x86/mmu: Refactor is_tdp_mmu_root() David Matlack
2021-06-14 17:56 ` Ben Gardon
2021-06-14 19:07 ` Sean Christopherson
2021-06-14 21:23 ` David Matlack
2021-06-14 21:39 ` Sean Christopherson
2021-06-14 22:01 ` David Matlack
2021-06-11 23:56 ` [PATCH 2/8] KVM: x86/mmu: Rename cr2_or_gpa to gpa in fast_page_fault David Matlack
2021-06-14 17:56 ` Ben Gardon
2021-06-11 23:56 ` [PATCH 3/8] KVM: x86/mmu: Fix use of enums in trace_fast_page_fault David Matlack
2021-06-11 23:56 ` [PATCH 4/8] KVM: x86/mmu: Common API for lockless shadow page walks David Matlack
2021-06-14 17:56 ` Ben Gardon
2021-06-11 23:56 ` [PATCH 5/8] KVM: x86/mmu: Also record spteps in shadow_page_walk David Matlack
2021-06-14 17:56 ` Ben Gardon
2021-06-14 22:27 ` David Matlack
2021-06-14 22:59 ` Sean Christopherson
2021-06-14 23:39 ` David Matlack [this message]
2021-06-15 0:22 ` Sean Christopherson
2021-06-11 23:56 ` [PATCH 6/8] KVM: x86/mmu: fast_page_fault support for the TDP MMU David Matlack
2021-06-11 23:59 ` David Matlack
2021-06-14 17:56 ` Ben Gardon
2021-06-14 22:34 ` David Matlack
2021-06-11 23:57 ` [PATCH 7/8] KVM: selftests: Fix missing break in dirty_log_perf_test arg parsing David Matlack
2021-06-14 17:56 ` Ben Gardon
2021-06-11 23:57 ` [PATCH 8/8] KVM: selftests: Introduce access_tracking_perf_test David Matlack
2021-06-14 17:56 ` Ben Gardon
2021-06-14 21:47 ` David Matlack
2021-06-14 9:54 ` [PATCH 0/8] KVM: x86/mmu: Fast page fault support for the TDP MMU Paolo Bonzini
2021-06-14 21:08 ` David Matlack
2021-06-15 7:16 ` Paolo Bonzini
2021-06-16 19:27 ` David Matlack
2021-06-16 19:31 ` 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=YMfopaDSRKvlsH0Y@google.com \
--to=dmatlack@google.com \
--cc=bgardon@google.com \
--cc=drjones@redhat.com \
--cc=jmattson@google.com \
--cc=joro@8bytes.org \
--cc=junaids@google.com \
--cc=kvm@vger.kernel.org \
--cc=pbonzini@redhat.com \
--cc=seanjc@google.com \
--cc=vkuznets@redhat.com \
--cc=wanpengli@tencent.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).