linux-kernel - Re: [PATCH 19/22] KVM: x86/mmu: Add infrastructure to allow walking rmaps outside of mmu

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: <Zrt5eNArfQA7x1qj@google.com>
Date: Tue, 13 Aug 2024 08:19:20 -0700
From: Sean Christopherson <seanjc@...gle.com>
To: Lai Jiangshan <jiangshanlai@...il.com>
Cc: Paolo Bonzini <pbonzini@...hat.com>, kvm@...r.kernel.org, linux-kernel@...r.kernel.org, 
	Oliver Upton <oliver.upton@...ux.dev>, Marc Zyngier <maz@...nel.org>, Peter Xu <peterx@...hat.com>, 
	James Houghton <jthoughton@...gle.com>
Subject: Re: [PATCH 19/22] KVM: x86/mmu: Add infrastructure to allow walking
 rmaps outside of mmu_lock

On Tue, Aug 13, 2024, Lai Jiangshan wrote:
> On Mon, Aug 12, 2024 at 11:22 PM Sean Christopherson <seanjc@...gle.com> wrote:
> 
> >
> > Oh yeah, duh, re-read after PAUSE, not before.
> >
> > Definitely holler if you have any alternative ideas for walking rmaps
> > without taking mmu_lock, I guarantee you've spent more time than me
> > thinking about the shadow MMU :-)
> 
> We use the same bit and the same way for the rmap lock.
> 
> We just use bit_spin_lock() and the optimization for empty rmap_head is
> handled out of kvm_rmap_lock().

Hmm, I'm leaning towards keeping the custom locking.  There are a handful of
benefits, none of which are all that meaningful on their own, but do add up.

 - Callers don't need to manually check for an empty rmap_head.
 - Can avoid the redundant preempt_{disable,enable}() entirely in the common case
   of being called while mmu_lock is held.
 - Handles the (likely super rare) edge case where a read-only walker encounters
   an rmap that was just emptied (rmap_head->val goes to zero after the initial
   check to elide the lock).
 - Avoids an atomic when releasing the lock, and any extra instructions entirely
   for writers since they always write the full rmap_head->val when releasing.

> bit_spin_lock() has the most-needed preempt_disable(). I'm not sure if the
> new kvm_rmap_age_gfn_range_lockless() is called in a preempt disabled region.

Oof, it doesn't.  Disabling IRQs crossed my mind, but I completely forgot about
preemption.

Thanks much!