Re: [PATCH v6 4/6] x86/sev: Add interface to re-enable RMP optimizations.

[sashiko-bot@kernel.org]

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:

Pre-existing issues:
- [High] The leader execution branches in `rmpopt_work_handler` lack `cond_resched()`, potentially causing softlockups on non-preemptible kernels during the expensive RMP scan.
--

commit 083ce9f1c90cf08302bc96625ddfd8a7dbfd5a45
Author: Ashish Kalra <ashish.kalra@amd.com>

x86/sev: Add interface to re-enable RMP optimizations.

This commit adds an interface to explicitly re-enable RMP optimizations.
It allows the processor to recover optimizations that may have been cleared
by events like RMPUPDATE by re-queuing the RMPOPT delayed work.

> diff --git a/arch/x86/virt/svm/sev.c b/arch/x86/virt/svm/sev.c
> index d7e40a5fe5ca6..4442ecae3d18f 100644
> --- a/arch/x86/virt/svm/sev.c
> +++ b/arch/x86/virt/svm/sev.c
> @@ -741,6 +741,21 @@ static void rmpopt_work_handler(struct work_struct *work)
>  	free_cpumask_var(follower_mask);
>  }

[Severity: High]
This is a pre-existing issue, but since the new code provides an interface
to explicitly re-queue this work, is there a risk of CPU stalls in the 
work handler?

In rmpopt_work_handler() in arch/x86/virt/svm/sev.c, the leader execution
paths loop over up to 2TB of memory in 1GB chunks (up to 2048 iterations)
without yielding:

	for (pa = rmpopt_pa_start; pa < rmpopt_pa_end; pa += SZ_1G)
		rmpopt(pa);

It also happens in the IPI loop:

	for (pa = rmpopt_pa_start; pa < rmpopt_pa_end; pa += SZ_1G)
		smp_call_function_single(leader_cpu, rmpopt_smp,
					 (void *)pa, true);

Since RMPOPT is an expensive hardware instruction, could executing up to 2048
iterations sequentially without cond_resched() cause workqueue lockups and
softlockup warnings on non-preemptible kernels?

>  
> +void snp_rmpopt_all_physmem(void)
> +{
> +	if (!cpu_feature_enabled(X86_FEATURE_RMPOPT))
> +		return;
> +
> +	guard(mutex)(&rmpopt_wq_mutex);
> +
> +	if (!rmpopt_wq)
> +		return;
> +
> +	queue_delayed_work(rmpopt_wq, &rmpopt_delayed_work,
> +			   msecs_to_jiffies(RMPOPT_WORK_TIMEOUT));
> +}

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/cover.1780427587.git.ashish.kalra@amd.com?part=4