linux-kernel - Re: [PATCH 2/7] KVM: x86/MMU: Move rmap

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: <Y5kf2KI5oharI0xZ@google.com>
Date:   Wed, 14 Dec 2022 00:59:04 +0000
From:   Sean Christopherson <seanjc@...gle.com>
To:     Ben Gardon <bgardon@...gle.com>
Cc:     David Matlack <dmatlack@...gle.com>, linux-kernel@...r.kernel.org,
        kvm@...r.kernel.org, Paolo Bonzini <pbonzini@...hat.com>,
        Peter Xu <peterx@...hat.com>, Vipin Sharma <vipinsh@...gle.com>
Subject: Re: [PATCH 2/7] KVM: x86/MMU: Move rmap_iterator to rmap.h

On Tue, Dec 13, 2022, Ben Gardon wrote:
> On Fri, Dec 9, 2022 at 3:04 PM David Matlack <dmatlack@...gle.com> wrote:
> >
> > > +/*
> > > + * Used by the following functions to iterate through the sptes linked by a
> > > + * rmap.  All fields are private and not assumed to be used outside.
> > > + */
> > > +struct rmap_iterator {
> > > +     /* private fields */
> > > +     struct pte_list_desc *desc;     /* holds the sptep if not NULL */
> > > +     int pos;                        /* index of the sptep */
> > > +};
> > > +
> > > +u64 *rmap_get_first(struct kvm_rmap_head *rmap_head,
> > > +                 struct rmap_iterator *iter);
> > > +u64 *rmap_get_next(struct rmap_iterator *iter);
> > > +
> > > +#define for_each_rmap_spte(_rmap_head_, _iter_, _spte_)                      \
> > > +     for (_spte_ = rmap_get_first(_rmap_head_, _iter_);              \
> > > +          _spte_; _spte_ = rmap_get_next(_iter_))
> > > +
> >
> > I always found these function names and kvm_rmap_head confusing since

Heh, you definitely aren't the only one.

> > they are about iterating through the pte_list_desc data structure. The
> > rmap (gfn -> list of sptes) is a specific application of the
> > pte_list_desc structure, but not the only application. There's also
> > parent_ptes in struct kvm_mmu_page, which is not an rmap, just a plain
> > old list of ptes.
>
> > While you are refactoring this code, what do you think about doing the
> > following renames?
> >
> >   struct kvm_rmap_head  -> struct pte_list_head
> >   struct rmap_iterator  -> struct pte_list_iterator
> >   rmap_get_first()      -> pte_list_get_first()
> >   rmap_get_next()       -> pte_list_get_next()
> >   for_each_rmap_spte()  -> for_each_pte_list_entry()

I would strongly prefer to keep "spte" in this one regardless of what other naming
changes we do (see below).  Maybe just for_each_spte()?  IMO, "pte_list_entry"
unnecessarily obfuscates that it's a list of SPTEs.

> > Then we can reserve the term "rmap" just for the actual rmap
> > (slot->arch.rmap), and code that deals with sp->parent_ptes will become
> > a lot more clear IMO (because it will not longer mention rmap).
> >
> > e.g. We go from this:
> >
> >   struct rmap_iterator iter;
> >   u64 *sptep;
> >
> >   for_each_rmap_spte(&sp->parent_ptes, &iter, sptep) {
> >      ...
> >   }
> >
> > To this:
> >
> >   struct pte_list_iterator iter;
> >   u64 *sptep;
> >
> >   for_each_pte_list_entry(&sp->parent_ptes, &iter, sptep) {
> >      ...
> >   }
> 
> I like this suggestion, and I do think it'll make things more
> readable. It's going to be a huge patch to rename all the instances of
> kvm_rmap_head, but it's probably worth it.

I generally like this idea too, but tying into my above comment, before jumping
in I think we should figure out what end state we want, i.e. get the bikeshedding
out of the way now to hopefully avoid dragging out a series while various things
get nitpicked.

E.g. if we if we just rename the structs and their macros, then we'll end up with
things like

	static bool slot_rmap_write_protect(struct kvm *kvm,
					    struct pte_list_head *rmap_head,
					    const struct kvm_memory_slot *slot)
	{
		return rmap_write_protect(rmap_head, false);
	}

which isn't terrible, but there's still opportunity for cleanup, e.g.
rmap_write_protect() could easily be sptes_write_protect() or write_protect_sptes().

That will generate a naming conflict of sorts with pte_list_head if we don't also
rename that to spte_list_head.  And I think capturing that it's a list of SPTEs and
not guest PTEs will be helpful in general.

And if we rename pte_list_head, then we might as well commit 100% and use consisnent
nomenclature across the board, e.g. end up with

	static bool sptes_clear_dirty(struct kvm *kvm, struct sptes_list_head *head,
				      const struct kvm_memory_slot *slot)
	{
		u64 *sptep;
		struct spte_list_iterator iter;
		bool flush = false;

		for_each_spte(head, &iter, sptep) {
			if (spte_ad_need_write_protect(*sptep))
				flush |= spte_wrprot_for_clear_dirty(sptep);
			else
				flush |= spte_clear_dirty(sptep);
		}

		return flush;
	}

versus the current

	static bool __rmap_clear_dirty(struct kvm *kvm, struct kvm_rmap_head *rmap_head,
				       const struct kvm_memory_slot *slot)
	{
		u64 *sptep;
		struct rmap_iterator iter;
		bool flush = false;

		for_each_rmap_spte(rmap_head, &iter, sptep)
			if (spte_ad_need_write_protect(*sptep))
				flush |= spte_wrprot_for_clear_dirty(sptep);
			else
				flush |= spte_clear_dirty(sptep);

		return flush;
	}