[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-Id: <CTHZKBJCQRHR.11CFMPICJ8XQP@wheely>
Date: Wed, 21 Jun 2023 12:51:31 +1000
From: "Nicholas Piggin" <npiggin@...il.com>
To: "Yu Zhao" <yuzhao@...gle.com>
Cc: "Andrew Morton" <akpm@...ux-foundation.org>,
"Paolo Bonzini" <pbonzini@...hat.com>,
"Alistair Popple" <apopple@...dia.com>,
"Anup Patel" <anup@...infault.org>,
"Ben Gardon" <bgardon@...gle.com>,
"Borislav Petkov" <bp@...en8.de>,
"Catalin Marinas" <catalin.marinas@....com>,
"Chao Peng" <chao.p.peng@...ux.intel.com>,
"Christophe Leroy" <christophe.leroy@...roup.eu>,
"Dave Hansen" <dave.hansen@...ux.intel.com>,
"Fabiano Rosas" <farosas@...ux.ibm.com>,
"Gaosheng Cui" <cuigaosheng1@...wei.com>,
"Gavin Shan" <gshan@...hat.com>, "H. Peter Anvin" <hpa@...or.com>,
"Ingo Molnar" <mingo@...hat.com>,
"James Morse" <james.morse@....com>,
"Jason A. Donenfeld" <Jason@...c4.com>,
"Jason Gunthorpe" <jgg@...pe.ca>,
"Jonathan Corbet" <corbet@....net>,
"Marc Zyngier" <maz@...nel.org>,
"Masami Hiramatsu" <mhiramat@...nel.org>,
"Michael Ellerman" <mpe@...erman.id.au>,
"Michael Larabel" <michael@...haellarabel.com>,
"Mike Rapoport" <rppt@...nel.org>,
"Oliver Upton" <oliver.upton@...ux.dev>,
"Paul Mackerras" <paulus@...abs.org>,
"Peter Xu" <peterx@...hat.com>,
"Sean Christopherson" <seanjc@...gle.com>,
"Steven Rostedt" <rostedt@...dmis.org>,
"Suzuki K Poulose" <suzuki.poulose@....com>,
"Thomas Gleixner" <tglx@...utronix.de>,
"Thomas Huth" <thuth@...hat.com>, "Will Deacon" <will@...nel.org>,
"Zenghui Yu" <yuzenghui@...wei.com>, <kvmarm@...ts.linux.dev>,
<kvm@...r.kernel.org>, <linux-arm-kernel@...ts.infradead.org>,
<linux-doc@...r.kernel.org>, <linux-kernel@...r.kernel.org>,
<linux-mm@...ck.org>, <linuxppc-dev@...ts.ozlabs.org>,
<linux-trace-kernel@...r.kernel.org>, <x86@...nel.org>,
<linux-mm@...gle.com>
Subject: Re: [PATCH mm-unstable v2 07/10] kvm/powerpc: add
kvm_arch_test_clear_young()
On Wed Jun 21, 2023 at 10:38 AM AEST, Yu Zhao wrote:
> On Tue, Jun 20, 2023 at 1:48 AM Nicholas Piggin <npiggin@...il.com> wrote:
> >
> > On Sat May 27, 2023 at 9:44 AM AEST, Yu Zhao wrote:
> > > Implement kvm_arch_test_clear_young() to support the fast path in
> > > mmu_notifier_ops->test_clear_young().
> > >
> > > It focuses on a simple case, i.e., radix MMU sets the accessed bit in
> > > KVM PTEs and VMs are not nested, where it can rely on RCU and
> > > pte_xchg() to safely clear the accessed bit without taking
> > > kvm->mmu_lock. Complex cases fall back to the existing slow path
> > > where kvm->mmu_lock is then taken.
> > >
> > > Signed-off-by: Yu Zhao <yuzhao@...gle.com>
> > > ---
> > > arch/powerpc/include/asm/kvm_host.h | 8 ++++
> > > arch/powerpc/include/asm/kvm_ppc.h | 1 +
> > > arch/powerpc/kvm/book3s.c | 6 +++
> > > arch/powerpc/kvm/book3s.h | 1 +
> > > arch/powerpc/kvm/book3s_64_mmu_radix.c | 59 ++++++++++++++++++++++++++
> > > arch/powerpc/kvm/book3s_hv.c | 5 +++
> > > 6 files changed, 80 insertions(+)
> > >
> > > diff --git a/arch/powerpc/include/asm/kvm_host.h b/arch/powerpc/include/asm/kvm_host.h
> > > index 14ee0dece853..75c260ea8a9e 100644
> > > --- a/arch/powerpc/include/asm/kvm_host.h
> > > +++ b/arch/powerpc/include/asm/kvm_host.h
> > > @@ -883,4 +883,12 @@ static inline void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu) {}
> > > static inline void kvm_arch_vcpu_blocking(struct kvm_vcpu *vcpu) {}
> > > static inline void kvm_arch_vcpu_unblocking(struct kvm_vcpu *vcpu) {}
> > >
> > > +#define kvm_arch_has_test_clear_young kvm_arch_has_test_clear_young
> > > +static inline bool kvm_arch_has_test_clear_young(void)
> > > +{
> > > + return IS_ENABLED(CONFIG_KVM_BOOK3S_HV_POSSIBLE) &&
> > > + cpu_has_feature(CPU_FTR_HVMODE) && cpu_has_feature(CPU_FTR_ARCH_300) &&
> > > + radix_enabled();
> >
> > This could probably be radix_enabled() && !kvmhv_on_pseries().
>
> Will do. (I used !kvmhv_on_pseries() in v1 but had second thoughts on
> moving kvmhv_on_pseries() into this file.)
That should be okay. kvmhv_on_pseries is a property of the host so it
seems reasonable to move it here if needed.
> > Although unclear why not nested hypervisor... I'd have to think about it a bit
> > more. Do you have any idea what might go wrong, or just didn't have the
> > time to consider it? (Not just powerpc nested but any others).
>
> Yes, this series excludes nested KVM support on all architures. The
> common reason for such a decision on powerpc and x86 (aarch64 doesn't
> support nested yet) is that it's quite challenging to make the rmap, a
> complex data structure that maps one PFN to multiple GFNs, lockless.
> (See kvmhv_insert_nest_rmap().) It might be possible, however, the
> potential ROI would be in question.
Okay just wondering. rmap (at least the powerpc one) is just a list
I think, with a few details. If that is all it is, it might not be
so hard to make that lock-free or a fine-grained lock on the rmap
chains maybe. But fine to ignore it to start with.
> > > +}
> > > +
> > > #endif /* __POWERPC_KVM_HOST_H__ */
> > > diff --git a/arch/powerpc/include/asm/kvm_ppc.h b/arch/powerpc/include/asm/kvm_ppc.h
> > > index 79a9c0bb8bba..ff1af6a7b44f 100644
> > > --- a/arch/powerpc/include/asm/kvm_ppc.h
> > > +++ b/arch/powerpc/include/asm/kvm_ppc.h
> > > @@ -287,6 +287,7 @@ struct kvmppc_ops {
> > > bool (*unmap_gfn_range)(struct kvm *kvm, struct kvm_gfn_range *range);
> > > bool (*age_gfn)(struct kvm *kvm, struct kvm_gfn_range *range);
> > > bool (*test_age_gfn)(struct kvm *kvm, struct kvm_gfn_range *range);
> > > + bool (*test_clear_young)(struct kvm *kvm, struct kvm_gfn_range *range);
> > > bool (*set_spte_gfn)(struct kvm *kvm, struct kvm_gfn_range *range);
> > > void (*free_memslot)(struct kvm_memory_slot *slot);
> > > int (*init_vm)(struct kvm *kvm);
> > > diff --git a/arch/powerpc/kvm/book3s.c b/arch/powerpc/kvm/book3s.c
> > > index 686d8d9eda3e..37bf40b0c4ff 100644
> > > --- a/arch/powerpc/kvm/book3s.c
> > > +++ b/arch/powerpc/kvm/book3s.c
> > > @@ -899,6 +899,12 @@ bool kvm_test_age_gfn(struct kvm *kvm, struct kvm_gfn_range *range)
> > > return kvm->arch.kvm_ops->test_age_gfn(kvm, range);
> > > }
> > >
> > > +bool kvm_arch_test_clear_young(struct kvm *kvm, struct kvm_gfn_range *range)
> > > +{
> > > + return !kvm->arch.kvm_ops->test_clear_young ||
> > > + kvm->arch.kvm_ops->test_clear_young(kvm, range);
> > > +}
> > > +
> > > bool kvm_set_spte_gfn(struct kvm *kvm, struct kvm_gfn_range *range)
> > > {
> > > return kvm->arch.kvm_ops->set_spte_gfn(kvm, range);
> > > diff --git a/arch/powerpc/kvm/book3s.h b/arch/powerpc/kvm/book3s.h
> > > index 58391b4b32ed..fa2659e21ccc 100644
> > > --- a/arch/powerpc/kvm/book3s.h
> > > +++ b/arch/powerpc/kvm/book3s.h
> > > @@ -12,6 +12,7 @@ extern void kvmppc_core_flush_memslot_hv(struct kvm *kvm,
> > > extern bool kvm_unmap_gfn_range_hv(struct kvm *kvm, struct kvm_gfn_range *range);
> > > extern bool kvm_age_gfn_hv(struct kvm *kvm, struct kvm_gfn_range *range);
> > > extern bool kvm_test_age_gfn_hv(struct kvm *kvm, struct kvm_gfn_range *range);
> > > +extern bool kvm_test_clear_young_hv(struct kvm *kvm, struct kvm_gfn_range *range);
> > > extern bool kvm_set_spte_gfn_hv(struct kvm *kvm, struct kvm_gfn_range *range);
> > >
> > > extern int kvmppc_mmu_init_pr(struct kvm_vcpu *vcpu);
> > > diff --git a/arch/powerpc/kvm/book3s_64_mmu_radix.c b/arch/powerpc/kvm/book3s_64_mmu_radix.c
> > > index 3b65b3b11041..0a392e9a100a 100644
> > > --- a/arch/powerpc/kvm/book3s_64_mmu_radix.c
> > > +++ b/arch/powerpc/kvm/book3s_64_mmu_radix.c
> > > @@ -1088,6 +1088,65 @@ bool kvm_test_age_radix(struct kvm *kvm, struct kvm_memory_slot *memslot,
> > > return ref;
> > > }
> > >
> > > +bool kvm_test_clear_young_hv(struct kvm *kvm, struct kvm_gfn_range *range)
> > > +{
> > > + bool err;
> > > + gfn_t gfn = range->start;
> > > +
> > > + rcu_read_lock();
> > > +
> > > + err = !kvm_is_radix(kvm);
> > > + if (err)
> > > + goto unlock;
> > > +
> > > + /*
> > > + * Case 1: This function kvmppc_switch_mmu_to_hpt()
> > > + *
> > > + * rcu_read_lock()
> > > + * Test kvm_is_radix() kvm->arch.radix = 0
> > > + * Use kvm->arch.pgtable synchronize_rcu()
> > > + * rcu_read_unlock()
> > > + * kvmppc_free_radix()
> > > + *
> > > + *
> > > + * Case 2: This function kvmppc_switch_mmu_to_radix()
> > > + *
> > > + * kvmppc_init_vm_radix()
> > > + * smp_wmb()
> > > + * Test kvm_is_radix() kvm->arch.radix = 1
> > > + * smp_rmb()
> > > + * Use kvm->arch.pgtable
> > > + */
> > > + smp_rmb();
> >
> > Comment could stand to expand slightly on what you are solving, in
> > words.
>
> Will do.
>
> > If you use synchronize_rcu() on both sides, you wouldn't need the
> > barrier, right?
>
> Case 2 is about memory ordering, which is orthogonal to case 1 (RCU
> freeing). So we need the r/w barrier regardless.
RCU can take care of memory ordering too though. If you had
synchronize_rcu() where smp_wmb() is, then no smp_rmb() neeed here.
>
> > > + while (gfn < range->end) {
> > > + pte_t *ptep;
> > > + pte_t old, new;
> > > + unsigned int shift;
> > > +
> > > + ptep = find_kvm_secondary_pte_unlocked(kvm, gfn * PAGE_SIZE, &shift);
> > > + if (!ptep)
> > > + goto next;
> > > +
> > > + VM_WARN_ON_ONCE(!page_count(virt_to_page(ptep)));
> >
> > Not really appropriate at the KVM level. mm enforces this kind of
> > thing (with notifiers).
>
> Will remove this.
>
> > > +
> > > + old = READ_ONCE(*ptep);
> > > + if (!pte_present(old) || !pte_young(old))
> > > + goto next;
> > > +
> > > + new = pte_mkold(old);
> > > +
> > > + if (kvm_should_clear_young(range, gfn))
> > > + pte_xchg(ptep, old, new);
> >
> > *Probably* will work. I can't think of a reason why not at the
> > moment anyway :)
>
> My reasoning:
> * It should work if we only change the dedicated A bit, i.e., not
> shared for other purposes, because races are safe (the case here).
> * It may not work for x86 EPT without the A bit (excluded in this
> series) where accesses are trapped by the R/X bits, because races in
> changing the R/X bits can be unsafe.
(For the benefit of others reading, it works because powerpc's pte_xchg
is actually a cmpxchg, for some reason which we really should fix).
Although it can fail to clear the bit if the cmpxchg fails.
I think pte_xchg is only used on with hash MMU in Linux before this
change. I think we may want to keep it that way and use something
like kvmppc_radix_update_pte() here to clear out the bit. But don't
worry too much about fine details so much before sorting out the
core changes I will have a better look after that.
Thanks,
Nick
Powered by blists - more mailing lists