linux-kernel - Re: [PATCH kvmtool v4 01/15] Allow pausing the VM from vcpu thread

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: <861pk0mbkv.wl-maz@kernel.org>
Date: Thu, 08 Jan 2026 14:19:12 +0000
From: Marc Zyngier <maz@...nel.org>
To: Suzuki K Poulose <suzuki.poulose@....com>
Cc: kvmarm@...ts.linux.dev,
	kvm@...r.kernel.org,
	linux-kernel@...r.kernel.org,
	will@...nel.org,
	oliver.upton@...ux.dev,
	alexandru.elisei@....com,
	aneesh.kumar@...nel.org,
	steven.price@....com,
	tabba@...gle.com
Subject: Re: [PATCH kvmtool v4 01/15] Allow pausing the VM from vcpu thread

On Tue, 30 Sep 2025 11:31:15 +0100,
Suzuki K Poulose <suzuki.poulose@....com> wrote:
> 
> Pausing the VM from a vCPU thread doesn't work today, as it waits indefinitely
> for a signal that never comes. By using the "current_kvm_cpu", enlighten the
> kvm__pause() to skip the current CPU and do it inline. This also brings in a
> restriction that a following kvm__continue() must be called from the same vCPU
> thread.
> 
> Cc: Will Deacon <will@...nel.org>
> Cc: Oliver Upton <oliver.upton@...ux.dev>
> Link: https://lore.kernel.org/all/20230918104028.GA17744@willie-the-truck/
> Signed-off-by: Suzuki K Poulose <suzuki.poulose@....com>
> ---
>  kvm.c | 35 +++++++++++++++++++++++++++++++----
>  1 file changed, 31 insertions(+), 4 deletions(-)
> 
> diff --git a/kvm.c b/kvm.c
> index 07089cf1..cc25ecdb 100644
> --- a/kvm.c
> +++ b/kvm.c
> @@ -59,6 +59,8 @@ const char *kvm_exit_reasons[] = {
>  
>  static int pause_event;
>  static DEFINE_MUTEX(pause_lock);
> +static struct kvm_cpu *pause_req_cpu;
> +
>  extern struct kvm_ext kvm_req_ext[];
>  
>  static char kvm_dir[PATH_MAX];
> @@ -573,9 +575,25 @@ void kvm__reboot(struct kvm *kvm)
>  
>  void kvm__continue(struct kvm *kvm)
>  {
> +	/*
> +	 * We must ensure that the resume request comes from the same context
> +	 * as the one requested the pause, especially if it was issued from a
> +	 * vCPU thread.
> +	 */
> +	if (current_kvm_cpu) {
> +		if (pause_req_cpu != current_kvm_cpu ||
> +		    !current_kvm_cpu->paused)
> +			die("Trying to resume VM from invalid context");
> +		current_kvm_cpu->paused = 0;
> +	}
>  	mutex_unlock(&pause_lock);
>  }
>  
> +/*
> + * Mark all active CPUs as paused, until kvm__continue() is issued.
> + * NOTE: If this is called from a cpu thread, kvm__continue() must
> + * be called from the same thread.
> + */
>  void kvm__pause(struct kvm *kvm)
>  {
>  	int i, paused_vcpus = 0;
> @@ -590,10 +608,17 @@ void kvm__pause(struct kvm *kvm)
>  	if (pause_event < 0)
>  		die("Failed creating pause notification event");
>  	for (i = 0; i < kvm->nrcpus; i++) {
> -		if (kvm->cpus[i]->is_running && kvm->cpus[i]->paused == 0)
> -			pthread_kill(kvm->cpus[i]->thread, SIGKVMPAUSE);
> -		else
> -			paused_vcpus++;
> +		if (kvm->cpus[i]->is_running && kvm->cpus[i]->paused == 0) {
> +			if (current_kvm_cpu != kvm->cpus[i]) {
> +				pthread_kill(kvm->cpus[i]->thread, SIGKVMPAUSE);
> +				continue;
> +			} else if (current_kvm_cpu) {
> +				current_kvm_cpu->paused = 1;
> +				pause_req_cpu = current_kvm_cpu;

This is also set as we leave the function. Why do we need to set it
twice?

> +				/* fall through to update our count */
> +			}
> +		}
> +		paused_vcpus++;
>  	}
>  
>  	while (paused_vcpus < kvm->nrcpus) {
> @@ -604,6 +629,8 @@ void kvm__pause(struct kvm *kvm)
>  		paused_vcpus += cur_read;
>  	}
>  	close(pause_event);
> +	/* Remember the context requesting pause */
> +	pause_req_cpu = current_kvm_cpu;
>  }
>  
>  void kvm__notify_paused(void)

Thanks,

	M.

-- 
Without deviation from the norm, progress is not possible.