Message-ID: <DDWPZY3AA7BX.1Y05FOYIHAI07@google.com>
Date: Fri, 31 Oct 2025 18:31:12 +0000
From: Brendan Jackman <jackmanb@...gle.com>
To: Brendan Jackman <jackmanb@...gle.com>, Dave Hansen <dave.hansen@...el.com>, 
	"Roy, Patrick" <roypat@...zon.co.uk>
Cc: "pbonzini@...hat.com" <pbonzini@...hat.com>, "corbet@....net" <corbet@....net>, 
	"maz@...nel.org" <maz@...nel.org>, "oliver.upton@...ux.dev" <oliver.upton@...ux.dev>, 
	"joey.gouly@....com" <joey.gouly@....com>, "suzuki.poulose@....com" <suzuki.poulose@....com>, 
	"yuzenghui@...wei.com" <yuzenghui@...wei.com>, "catalin.marinas@....com" <catalin.marinas@....com>, 
	"will@...nel.org" <will@...nel.org>, "tglx@...utronix.de" <tglx@...utronix.de>, 
	"mingo@...hat.com" <mingo@...hat.com>, "bp@...en8.de" <bp@...en8.de>, 
	"dave.hansen@...ux.intel.com" <dave.hansen@...ux.intel.com>, "x86@...nel.org" <x86@...nel.org>, 
	"hpa@...or.com" <hpa@...or.com>, "luto@...nel.org" <luto@...nel.org>, 
	"peterz@...radead.org" <peterz@...radead.org>, "willy@...radead.org" <willy@...radead.org>, 
	"akpm@...ux-foundation.org" <akpm@...ux-foundation.org>, "david@...hat.com" <david@...hat.com>, 
	"lorenzo.stoakes@...cle.com" <lorenzo.stoakes@...cle.com>, 
	"Liam.Howlett@...cle.com" <Liam.Howlett@...cle.com>, "vbabka@...e.cz" <vbabka@...e.cz>, 
	"rppt@...nel.org" <rppt@...nel.org>, "surenb@...gle.com" <surenb@...gle.com>, "mhocko@...e.com" <mhocko@...e.com>, 
	"song@...nel.org" <song@...nel.org>, "jolsa@...nel.org" <jolsa@...nel.org>, "ast@...nel.org" <ast@...nel.org>, 
	"daniel@...earbox.net" <daniel@...earbox.net>, "andrii@...nel.org" <andrii@...nel.org>, 
	"martin.lau@...ux.dev" <martin.lau@...ux.dev>, "eddyz87@...il.com" <eddyz87@...il.com>, 
	"yonghong.song@...ux.dev" <yonghong.song@...ux.dev>, 
	"john.fastabend@...il.com" <john.fastabend@...il.com>, "kpsingh@...nel.org" <kpsingh@...nel.org>, 
	"sdf@...ichev.me" <sdf@...ichev.me>, "haoluo@...gle.com" <haoluo@...gle.com>, "jgg@...pe.ca" <jgg@...pe.ca>, 
	"jhubbard@...dia.com" <jhubbard@...dia.com>, "peterx@...hat.com" <peterx@...hat.com>, 
	"jannh@...gle.com" <jannh@...gle.com>, "pfalcato@...e.de" <pfalcato@...e.de>, 
	"shuah@...nel.org" <shuah@...nel.org>, "seanjc@...gle.com" <seanjc@...gle.com>, 
	"kvm@...r.kernel.org" <kvm@...r.kernel.org>, 
	"linux-doc@...r.kernel.org" <linux-doc@...r.kernel.org>, 
	"linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>, 
	"linux-arm-kernel@...ts.infradead.org" <linux-arm-kernel@...ts.infradead.org>, 
	"kvmarm@...ts.linux.dev" <kvmarm@...ts.linux.dev>, 
	"linux-fsdevel@...r.kernel.org" <linux-fsdevel@...r.kernel.org>, "linux-mm@...ck.org" <linux-mm@...ck.org>, 
	"bpf@...r.kernel.org" <bpf@...r.kernel.org>, 
	"linux-kselftest@...r.kernel.org" <linux-kselftest@...r.kernel.org>, "Cali, Marco" <xmarcalx@...zon.co.uk>, 
	"Kalyazin, Nikita" <kalyazin@...zon.co.uk>, "Thomson, Jack" <jackabt@...zon.co.uk>, 
	"derekmn@...zon.co.uk" <derekmn@...zon.co.uk>, "tabba@...gle.com" <tabba@...gle.com>, 
	"ackerleytng@...gle.com" <ackerleytng@...gle.com>
Subject: Re: [PATCH v7 06/12] KVM: guest_memfd: add module param for disabling
 TLB flushing

On Thu Oct 30, 2025 at 4:05 PM UTC, Brendan Jackman wrote:
> On Thu Sep 25, 2025 at 6:27 PM UTC, Dave Hansen wrote:
>> On 9/24/25 08:22, Roy, Patrick wrote:
>>> Add an option to not perform TLB flushes after direct map manipulations.
>>
>> I'd really prefer this be left out for now. It's a massive can of worms.
>> Let's agree on something that works and has well-defined behavior before
>> we go breaking it on purpose.
>
> As David pointed out in the MM Alignment Session yesterday, I might be
> able to help here. In [0] I've proposed a way to break up the direct map
> by ASI's "sensitivity" concept, which is weaker than the "totally absent
> from the direct map" being proposed here, but it has kinda similar
> implementation challenges.
>
> Basically it introduces a thing called a "freetype" that extends the
> idea of migratetype. Like the existing idea of migratetype, it's used to
> physically group pages when allocating, and you can index free pages by
> it, i.e. each freetype gets its own freelist. But it can also encode
> other information than mobility (and the other stuff that's encoded in
> migratetype...).
>
> Could it make sense to use that logic to just have entire pageblocks
> that are absent from the direct map? Then when allocating memory for the
> guest_memfd we get it from one of those pageblocks. Then we only have to
> flush the TLB if there's no memory left in pageblocks of this freetype
> (so the allocator has to flip another pageblock over to the "no direct
> map" freetype, after removing it from the direct map).
>
> I haven't yet investigated this properly, I'll start doing that now.
> But I thought I'd immediately drop this note in case anyone can
> immediately see a reason why this doesn't work.

I spent some time poking around and I think there's only one issue here:
in this design the mapping/unmapping of the direct map happens while
allocating. But, it might need to allocate a pagetable to break down a
page.

In my ASI-specific presentation of that feature, I dodged this issue by
just requiring the whole ASI direct map to be set up at pageblock
granularity. This totally dodges the recursion issue since we just never
have to break down pages. (Actually, Dave Hansen suggested for the
initial implementation I simplify it by just doing all the ASI stuff at
4k, which achieves the same thing).

I guess we'd like to avoid globally fragmenting the whole direct map
just in case someone wants to use guest_memfd at some point? And, I
guess we could just instantaneously fragment it all at the instant that
someone wants to do that, but that's still a bit yucky.

If we just ignore this issue and try to allocate pagetables, it's
possible for a pathological physmap state to emerge where we get into
the allocator path that [un]maps a pageblock, but then need to allocate
a page to [un]map it, and that allocation in turn gets into the
[un]mapping path, and suddenly, turtles.

I think the simplest answer to that is to just fail the [un]map path if
we detect we're recursive, with something like a PF_MEMALLOC_* flag.
But this feels a bit yucky.

Other ideas might include: don't actually fragment the whole physmap,
but at least pre-allocate the pagetables down to pageblock granularity.
Or alternatively, this could point to an issue in the way I injected
[un]mapping into the allocator, and fixing that design flaw would solve
the problem.

I'll have to think about this some more on Monday but sharing my
thoughts now in case anyone has an idea already...

I've dumped the (untested) branch where I've adapted [0] for the
NO_DIRECT_MAP usecase here:

https://github.com/bjackman/linux/tree/demo-guest_memfd-physmap

> [0] https://lore.kernel.org/all/20250924-b4-asi-page-alloc-v1-0-2d861768041f@google.com/T/#t

Powered by blists - more mailing lists