linux-kernel - Re: [PATCH] KVM: arm64: Check cpu_has_spe() before initializing PMSCR

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] [day] [month] [year] [list]

Message-ID: <0b610239-6e90-464a-bf01-5332de62719f@arm.com>
Date: Thu, 9 Oct 2025 10:11:20 +0100
From: Suzuki K Poulose <suzuki.poulose@....com>
To: Oliver Upton <oliver.upton@...ux.dev>, Marc Zyngier <maz@...nel.org>
Cc: Mukesh Ojha <mukesh.ojha@....qualcomm.com>, joey.gouly@....com,
 yuzenghui@...wei.com, catalin.marinas@....com, will@...nel.org,
 alexandru.elisei@....com, linux-arm-kernel@...ts.infradead.org,
 kvmarm@...ts.linux.dev, linux-kernel@...r.kernel.org
Subject: Re: [PATCH] KVM: arm64: Check cpu_has_spe() before initializing
 PMSCR_EL1 in VHE

On 08/10/2025 19:26, Oliver Upton wrote:
> On Wed, Oct 08, 2025 at 11:46:55AM +0100, Marc Zyngier wrote:
>> On Tue, 07 Oct 2025 19:31:45 +0100,
>> Oliver Upton <oliver.upton@...ux.dev> wrote:
>>>
>>> Hi Mukesh,
>>>
>>> I find it a bit odd to refer to cpu_has_spe() in the shortlog, which
>>> doesn't exist prior to this patch.
>>>
>>> On Tue, Oct 07, 2025 at 11:53:56PM +0530, Mukesh Ojha wrote:
>>>> commit efad60e46057 ("KVM: arm64: Initialize PMSCR_EL1 when in VHE")
>>>> initializes PMSCR_EL1 to 0 which is making the boot up stuck when KVM
>>>> runs in VHE mode and reverting the change is fixing the issue.
>>>>
>>>> [    2.967447] RPC: Registered tcp NFSv4.1 backchannel transport module.
>>>> [    2.974061] PCI: CLS 0 bytes, default 64
>>>> [    2.978171] Unpacking initramfs...
>>>> [    2.982889] kvm [1]: nv: 568 coarse grained trap handlers
>>>> [    2.988573] kvm [1]: IPA Size Limit: 40 bits
>>>>
>>>> Lets guard the change with cpu_has_spe() check so that it only affects
>>>> the cpu which has SPE feature supported.
>>>
>>> This could benefit from being spelled out a bit more. In both cases we
>>> check for the presence of FEAT_SPE, however I believe the issue you
>>> observe is EL3 hasn't delegated ownership of the Profiling Buffer to
>>> Non-secure nor does it reinject an UNDEF in response to the sysreg trap.
>>>
>>> I agree that the change is correct but the rationale needs to be clear.
>>
>> To me, this smells a lot more like some sort of papering over a
>> firmware bug. Why isn't SPE available the first place?
> 
> While I agree this points the finger at a half-assed EL3, the
> architecture explicitly allows this sort of crap and we cope with the
> accessibility of SPE in almost every other case.

One of the reasons behind this control is to work around errata, where
the higher EL could prevent the OS from using a broken implementation.
Also, for KVM could use the EL2 controls to disable SPE for Guest.

Suzuki

> 
> We should at least be consistent in how we handle an inaccessible SPE.
> 
> Thanks,
> Oliver