Re: [XEN PATCH v1 5/5] xen/arm: ffa: support notification

[Jens Wiklander <jens.wiklander@xxxxxxxxxx>]

Hi Bertrand,

On Wed, Apr 10, 2024 at 6:30 PM Bertrand Marquis
<Bertrand.Marquis@xxxxxxx> wrote:
>
> Hi Jens,
>
> > On 10 Apr 2024, at 17:45, Jens Wiklander <jens.wiklander@xxxxxxxxxx> wrote:
> >
> > On Tue, Apr 9, 2024 at 5:36 PM Jens Wiklander <jens.wiklander@xxxxxxxxxx> 
> > wrote:
> >>
> >> Add support for FF-A notifications, currently limited to an SP (Secure
> >> Partition) sending an asynchronous notification to a guest.
> >>
> >> Guests and Xen itself are made aware of pending notifications with an
> >> interrupt. The interrupt handler retrieves the notifications using the
> >> FF-A ABI and deliver them to their destinations.
> >>
> >> Signed-off-by: Jens Wiklander <jens.wiklander@xxxxxxxxxx>
> >> ---
[snip]
> >> +    case FFA_FEATURE_NOTIF_PEND_INTR:
> >> +        if ( ctx->notif.enabled )
> >> +            ffa_set_regs_success(regs, FFA_NOTIF_PEND_INTR_ID, 0);
> >> +        else
> >> +            ffa_set_regs_error(regs, FFA_RET_NOT_SUPPORTED);
> >> +        break;
> >> +    case FFA_FEATURE_SCHEDULE_RECV_INTR:
> >> +        if ( ctx->notif.enabled )
> >> +            ffa_set_regs_success(regs, FFA_NOTIF_PEND_INTR_ID, 0);
> >> +        else
> >> +            ffa_set_regs_error(regs, FFA_RET_NOT_SUPPORTED);
> >> +        break;
> >
> > With the recently posted kernel patch
> > https://lore.kernel.org/all/20240410-ffa_npi_support-v1-3-1a5223391bd1@xxxxxxx/
> > we need to decide which feature interrupt to return since the kernel
> > will only install a handle for the first it finds. Right now I propose
> > to to not report FFA_FEATURE_SCHEDULE_RECV_INTR. When the time comes
> > to use a secondary scheduler we'll need to report the SRI instead.
>
>
> We just had a meeting with Sudeep to discuss that matter and he agreed that
> he would register the interrupt handler for all the interrupts available so 
> that
> we can keep a model where we will generate SRIs only to a secondary scheduler
> and NPI for notification interrupts (so that the VM does not do a INFO_GET 
> when
> not required).
>
> We will have to report both as any VM could act as secondary scheduler for SPs
> in theory (we might need at some point a parameter for that) but for now those
> should only be generated to Dom0 if there are pending notifications for SPs.

OK, thanks. I'll keep both then.

Cheers,
Jens