On Wed, Feb 10, 2021 at 1:14 AM Michael S. Tsirkin [off-list ref] wrote:

On Tue, Feb 09, 2021 at 10:00:22AM -0800, Wei Wang wrote:

quoted

On Tue, Feb 9, 2021 at 6:58 AM Willem de Bruijn
[off-list ref] wrote:

quoted

quoted

quoted

quoted

quoted

I have no preference. Just curious, especially if it complicates the patch.

My understanding is that. It's probably ok for net. But we probably need
to document the assumptions to make sure it was not abused in other drivers.

Introduce new parameters for find_vqs() can help to eliminate the subtle
stuffs but I agree it looks like a overkill.

(Btw, I forget the numbers but wonder how much difference if we simple
remove the free_old_xmits() from the rx NAPI path?)

The committed patchset did not record those numbers, but I found them
in an earlier iteration:

   [PATCH net-next 0/3] virtio-net tx napi
   https://lists.openwall.net/netdev/2017/04/02/55

It did seem to significantly reduce compute cycles ("Gcyc") at the
time. For instance:

     TCP_RR Latency (us):
     1x:
       p50              24       24       21
       p99              27       27       27
       Gcycles         299      432      308

I'm concerned that removing it now may cause a regression report in a
few months. That is higher risk than the spurious interrupt warning
that was only reported after years of use.

Right.

So if Michael is fine with this approach, I'm ok with it. But we
probably need to a TODO to invent the interrupt handlers that can be
used for more than one virtqueues. When MSI-X is enabled, the interrupt
handler (vring_interrup()) assumes the interrupt is used by a single
virtqueue.

Thanks.

The approach to schedule tx-napi from virtnet_poll_cleantx instead of
cleaning directly in this rx-napi function was not effective at
suppressing the warning, I understand.

Correct. I tried the approach to schedule tx napi instead of directly
do free_old_xmit_skbs() in virtnet_poll_cleantx(). But the warning
still happens.

Two questions here: is the device using packed or split vqs?
And is event index enabled?

The device is indeed using split vqs with event index enabled.

I think one issue is that at the moment with split and event index we
don't actually disable events at all.

You mean we don't disable 'interrupts' right?
What is the reason for that?

static void virtqueue_disable_cb_split(struct virtqueue *_vq)
{
        struct vring_virtqueue *vq = to_vvq(_vq);

        if (!(vq->split.avail_flags_shadow & VRING_AVAIL_F_NO_INTERRUPT)) {
                vq->split.avail_flags_shadow |= VRING_AVAIL_F_NO_INTERRUPT;
                if (!vq->event)
                        vq->split.vring.avail->flags =
                                cpu_to_virtio16(_vq->vdev,
                                                vq->split.avail_flags_shadow);
        }
}

Can you try your napi patch + disable event index?

Thanks for the suggestion.
I've run the reproducer with  napi patch + disable event index, and so
far, I did not see the warning getting triggered. Will keep it running
for a bit longer.

--
MST