Re: [RESEND][PATCH] xen/arm: sched: Ensure the vCPU context is seen before vcpu_pause() returns

[Bertrand Marquis <Bertrand.Marquis@xxxxxxx>]

Hi Julien,

> On 22 Sep 2020, at 20:31, Julien Grall <julien@xxxxxxx> wrote:
> 
> From: Julien Grall <jgrall@xxxxxxxxxx>
> 
> Some callers of vcpu_pause() will expect to access the latest vcpu
> context when the function returns (see XENDOMCTL_{set,get}vcpucontext}.
> 
> However, the latest vCPU context can only be observed after
> v->is_running has been observed to be false.
> 
> As there is no memory barrier instruction generated, a processor could
> try to speculatively access the vCPU context before it was observed.

The function vcpu_context_saved does contain a memory barrier already.

Shouldn’t we make sure instead that any time is_running is modified to
false there is a barrier before (which is the case in vcpu_context_saved) ?

I understand the goal here but the barrier seem very far from the modification
of is_running.

Cheers,
Bertrand

> 
> To prevent the corruption of the vCPU context, we need to insert a
> memory barrier instruction after v->is_running is observed and before
> the context is accessed. This barrier is added in sync_vcpu_execstate()
> as it seems to be the place where we expect the synchronization to
> happen.
> 
> Signed-off-by: Julien Grall <jgrall@xxxxxxxxxx>
> 
> ---
> 
> Cc: Andrew Cooper <andrew.cooper3@xxxxxxxxxx>
> Cc: Jan Beulich <jbeulich@xxxxxxxx>
> Cc: George Dunlap <george.dunlap@xxxxxxxxxx>
> Cc: Dario Faggioli <dfaggioli@xxxxxxxx>
> Cc: Bertrand Marquis <Bertrand.Marquis@xxxxxxx>
> 
> I am also adding the x86 and scheduler maintainers because I am not sure
> whether this barrier should be part of the common code instead.
> ---
> xen/arch/arm/domain.c | 15 ++++++++++++++-
> 1 file changed, 14 insertions(+), 1 deletion(-)
> 
> diff --git a/xen/arch/arm/domain.c b/xen/arch/arm/domain.c
> index 9258f6d3faa2..3b37f899b9da 100644
> --- a/xen/arch/arm/domain.c
> +++ b/xen/arch/arm/domain.c
> @@ -371,7 +371,20 @@ void sync_local_execstate(void)
> 
> void sync_vcpu_execstate(struct vcpu *v)
> {
> -    /* Nothing to do -- no lazy switching */
> +    /*
> +     * We don't support lazy switching.
> +     *
> +     * However the context may have been saved from a remote pCPU so we
> +     * need a barrier to ensure it is observed before continuing.
> +     *
> +     * Per vcpu_context_saved(), the context can be observed when
> +     * v->is_running is false (the caller should check it before calling
> +     * this function).
> +     *
> +     * Note this is a full barrier to also prevent update of the context
> +     * to happen before it was observed.
> +     */
> +    smp_mb();
> }
> 
> #define NEXT_ARG(fmt, args)                                                 \
> -- 
> 2.17.1
>