linux-kernel - Re: [PATCH v2 00/10] Independent per-CPU data section for nVHE

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: <20200916115404.rhv4dkyjz35e4x25@google.com>
Date:   Wed, 16 Sep 2020 12:54:04 +0100
From:   David Brazdil <dbrazdil@...gle.com>
To:     Will Deacon <will@...nel.org>
Cc:     Marc Zyngier <maz@...nel.org>,
        Catalin Marinas <catalin.marinas@....com>,
        Dennis Zhou <dennis@...nel.org>, Tejun Heo <tj@...nel.org>,
        Christoph Lameter <cl@...ux.com>,
        Arnd Bergmann <arnd@...db.de>,
        James Morse <james.morse@....com>,
        Julien Thierry <julien.thierry.kdev@...il.com>,
        Suzuki K Poulose <suzuki.poulose@....com>,
        linux-arm-kernel@...ts.infradead.org, linux-kernel@...r.kernel.org,
        kvmarm@...ts.cs.columbia.edu, linux-arch@...r.kernel.org,
        kernel-team@...roid.com
Subject: Re: [PATCH v2 00/10] Independent per-CPU data section for nVHE

Hi Will,

On Mon, Sep 14, 2020 at 06:40:09PM +0100, Will Deacon wrote:
> Hi David,
> 
> On Thu, Sep 03, 2020 at 11:17:02AM +0200, David Brazdil wrote:
> > Introduce '.hyp.data..percpu' as part of ongoing effort to make nVHE
> > hyp code self-contained and independent of the rest of the kernel.
> > 
> > The series builds on top of the "Split off nVHE hyp code" series which
> > used objcopy to rename '.text' to '.hyp.text' and prefix all ELF
> > symbols with '__kvm_nvhe' for all object files under kvm/hyp/nvhe.
> 
> I've been playing around with this series this afternoon, trying to see
> if we can reduce the coupling between the nVHE code and the core code. I've
> ended up with the diff below on top of your series, but I think it actually
> removes the need to change the core code at all. The idea is to collapse
> the percpu sections during prelink, and then we can just deal with the
> resulting data section a bit like we do for .hyp.text already.
> 
> Have I missed something critical?

I was wondering whether this approach would be sufficient as well because of
the simplicity. We'd just need to be careful about correctly preserving the
semantics of the different .data..percpu..* sections.

For instance, I've noticed you make .hyp..data..percpu page-aligned rather than
cacheline-aligned. We need that for stage-2 unmapping but it also happens to
correctly align DEFINE_PER_CPU_PAGE_ALIGNED variables when collapsed into the
single hyp section. The reason why I ended up reusing the global macro was to
avoid introducing subtleties like that into the arm64 linker script. Do you
think it's a worthwhile trade off?

One place where this approach doesn't work is DEFINE_PER_CPU_FIRST. But I'm
guessing that's something we can live without.

I was also wondering about another approach - using the PERCPU_SECTION macro
unchanged in the hyp linker script. It would lay out a single .data..percpu and
we would then prefix it with .hyp and the symbols with __kvm_nvhe_ as with
everything else. WDYT? Haven't tried that yet, could be a naive idea. 

Thanks for reviewing,
David