Re: [PATCH 1/6] firmware: arm_scmi: Add atomic mode support to virtio transport

[Cristian Marussi]

On Thu, Jan 27, 2022 at 09:07:59AM +0000, Cristian Marussi wrote:
> On Wed, Jan 26, 2022 at 03:28:52PM +0100, Peter Hilber wrote:
> > On 24.01.22 11:03, Cristian Marussi wrote:
> > > Add support for .mark_txdone and .poll_done transport operations to SCMI
> > > VirtIO transport as pre-requisites to enable atomic operations.
> > > 
> > > Add a Kernel configuration option to enable SCMI VirtIO transport polling
> > > and atomic mode for selected SCMI transactions while leaving it default
> > > disabled.
> > > 
> > 

Hi Peter,

> > Hi Cristian,
> > 
> 
> Hi Peter,
> 
> > please see one remark below.
> > 
> > Best regards,
> > 
> > Peter
> > 
> > > Cc: "Michael S. Tsirkin" <mst@xxxxxxxxxx>
> > > Cc: Igor Skalkin <igor.skalkin@xxxxxxxxxxxxxxx>
> > > Cc: Peter Hilber <peter.hilber@xxxxxxxxxxxxxxx>
> > > Cc: virtualization@xxxxxxxxxxxxxxxxxxxxxxxxxx
> > > Signed-off-by: Cristian Marussi <cristian.marussi@xxxxxxx>
> > > ---
> 
> [snip]
> 
> > > +/**
> > > + * virtio_poll_done  - Provide polling support for VirtIO transport
> > > + *
> > > + * @cinfo: SCMI channel info
> > > + * @xfer: Reference to the transfer being poll for.
> > > + *
> > > + * VirtIO core provides a polling mechanism based only on last used indexes:
> > > + * this means that it is possible to poll the virtqueues waiting for something
> > > + * new to arrive from the host side but the only way to check if the freshly
> > > + * arrived buffer was what we were waiting for is to compare the newly arrived
> > > + * message descriptors with the one we are polling on.
> > > + *
> > > + * As a consequence it can happen to dequeue something different from the buffer
> > > + * we were poll-waiting for: if that is the case such early fetched buffers are
> > > + * then added to a the @pending_cmds_list list for later processing by a
> > > + * dedicated deferred worker.
> > > + *
> > > + * So, basically, once something new is spotted we proceed to de-queue all the
> > > + * freshly received used buffers until we found the one we were polling on, or,
> > > + * we have 'seemingly' emptied the virtqueue; if some buffers are still pending
> > > + * in the vqueue at the end of the polling loop (possible due to inherent races
> > > + * in virtqueues handling mechanisms), we similarly kick the deferred worker
> > > + * and let it process those, to avoid indefinitely looping in the .poll_done
> > > + * helper.
> > > + *
> > > + * Note that, since we do NOT have per-message suppress notification mechanism,
> > > + * the message we are polling for could be delivered via usual IRQs callbacks
> > > + * on another core which happened to have IRQs enabled: in such case it will be
> > > + * handled as such by scmi_rx_callback() and the polling loop in the
> > > + * SCMI Core TX path will be transparently terminated anyway.
> > > + *
> > > + * Return: True once polling has successfully completed.
> > > + */
> > > +static bool virtio_poll_done(struct scmi_chan_info *cinfo,
> > > +			     struct scmi_xfer *xfer)
> > > +{
> > > +	bool pending, ret = false;
> > > +	unsigned int length, any_prefetched = 0;
> > > +	unsigned long flags;
> > > +	struct scmi_vio_msg *next_msg, *msg = xfer->priv;
> > > +	struct scmi_vio_channel *vioch = cinfo->transport_info;
> > > +
> > > +	if (!msg)
> > > +		return true;
> > > +
> > > +	spin_lock_irqsave(&vioch->lock, flags);
> > 
> > If now acquiring vioch->lock here, I see no need to virtqueue_poll() any more.
> > After checking msg->poll_status, we could just directly try virtqueue_get_buf().
> > 
> > On the other hand, always taking the vioch->lock in a busy loop might better be
> > avoided (I assumed before that taking it was omitted on purpose), since it
> > might hamper tx channel progress in other cores (but I'm not sure about the
> > actual impact).
> > 
> > Also, I don't yet understand why the vioch->lock would need to be taken here.
> 
> There was a race I could reproduce between the below check against
> VIO_MSG_POLL_DONE and the poll_idx later update near the end of the poll loop
> where another thread could have set VIO_MSG_POLL_DONE after this thread
> had checked it and then this same thread would have cleared it rewriting
> the new poll_idx; so at first I needlessly enlanrged the spinlocked section
> (even though I knew was subtopimal given virtqueue_poll does not need
> serialization) and then forget to properly review this thing.
> 
> BUT now that, following your suggestion, I introduced a dedicated
> poll_status that race is gone, so I shrinked back the spinlocked section
> as before and works fine (even poll_idx itself does not need to be
> protected really given it can be accessed only here)
> 
> I'll post the fix in -rc2 together with the core change in the
> virtio-core I proposed last week to Michael (if not too costly as perfs)
> 

Looking again at the polling support, beside the above fixes, I realized that
also the handling of the virtqueue ready status was not addressed at all
in polling mode, so I got rid of it and introduced a dedicated refcount
mechanism that should serve the same purpose; moreover I shrinked a bit the
main spinlocked areas introducing dedicated spinlocks for free_list and
pending_lists. (and removed some redundant locks on msg...hopefully :P)

..and everuything still works apparently in my test scenario ...
( but ....  you know... :P)

So just a heads up that the next series will include a couple of
preparatory patches, and the core virtio_ring fix for the ABA problem,
before the virtio polling one.

Thanks,
Cristian