Kernel KVM virtualization development
 help / color / mirror / Atom feed
From: sashiko-bot@kernel.org
To: "Hemanth Selam" <hemanth.selam@gmail.com>
Cc: kvm@vger.kernel.org
Subject: Re: [PATCH] KVM: selftests: Add a test for nested virtual NMI support
Date: Fri, 11 Sep 2026 10:30:40 +0000	[thread overview]
Message-ID: <20260911103041.4E7A71F00899@smtp.kernel.org> (raw)
In-Reply-To: <20260911102105.1927773-1-hemanth.selam@gmail.com>

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
- [Medium] Missing memory barrier before VM-exit allows compiler to reorder global variable store, causing test flakiness.
--

commit 369b941892ff4fd633ee8a8bca1eef89e730d9a0
Author: Hemanth Selam <hemanth.selam@gmail.com>

KVM: selftests: Add a test for nested virtual NMI support

KVM's virtual NMI support for nested guests has no test coverage.
This commit adds a test covering the vNMI controls L1 provides in vmcb12,
including delivery of virtual NMIs to L2 and verifying the state of
V_NMI_PENDING and V_NMI_BLOCKING flags.

> diff --git a/tools/testing/selftests/kvm/x86/svm_nested_vnmi_test.c b/tools/testing/selftests/kvm/x86/svm_nested_vnmi_test.c
> new file mode 100644
> index 0000000000000..4e9916a837ddd
> --- /dev/null
> +++ b/tools/testing/selftests/kvm/x86/svm_nested_vnmi_test.c

[ ... ]

> +static unsigned int nmi_fired;
> +
> +static void guest_nmi_handler(struct ex_regs *regs)
> +{
> +	nmi_fired++;
> +
> +	/*
> +	 * Exit to L1 from NMI context, i.e. before this handler's IRET, so
> +	 * that L1 can observe V_NMI_BLOCKING while the NMI is in service.
> +	 */
> +	vmmcall();
> +}

[Severity: Medium]
Could an optimizing compiler reorder the store to nmi_fired after the
vmmcall()?

Since nmi_fired is a plain global variable and vmmcall() is implemented as
an inline assembly block without a "memory" clobber, the C standard allows
the compiler to reorder the memory store across the vmmcall() instruction.

If instruction scheduling places the memory store after the vmmcall(), L2
will execute the VM-exit before the store is committed, causing the host (L1)
to read a stale value and making the GUEST_ASSERT_EQ(nmi_fired, i) assertion
fail sporadically.

Would adding a compiler barrier or using WRITE_ONCE() before the vmmcall()
help ensure the store is committed before exiting to the host?

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20260911102105.1927773-1-hemanth.selam@gmail.com?part=1

      reply	other threads:[~2026-09-11 10:30 UTC|newest]

Thread overview: 2+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2026-09-11 10:21 [PATCH] KVM: selftests: Add a test for nested virtual NMI support Hemanth Selam
2026-09-11 10:30 ` sashiko-bot [this message]

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20260911103041.4E7A71F00899@smtp.kernel.org \
    --to=sashiko-bot@kernel.org \
    --cc=hemanth.selam@gmail.com \
    --cc=kvm@vger.kernel.org \
    --cc=sashiko-reviews@lists.linux.dev \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox