On Wed, Sep 07, 2022 at 07:18:06PM +0000, Parav Pandit wrote:

quoted

From: Michael S. Tsirkin <mst@redhat.com>
Sent: Wednesday, September 7, 2022 3:12 PM

quoted

quoted

Because of shallow queue of 16 entries deep.

but why is the queue just 16 entries?

I explained the calculation in [1] about 16 entries.

[1] https://lore.kernel.org/netdev/PH0PR12MB54812EC7F4711C1EA4CAA119DC419@PH0PR12MB5481.namprd12.prod.outlook.com/ (local)

quoted

does the device not support indirect?

Yes, indirect feature bit is disabled on the device.

OK that explains it.

quoted

because with indirect you get 256 entries, with 16 s/g each.

Sure. I explained below that indirect comes with 7x memory cost that is not desired.
(Ignored the table memory allocation cost and extra latency).

Oh sure, it's a waste. I wonder what effect does the patch have
on bandwidth with indirect enabled though.

Hence don't want to enable indirect in this scenario.
This optimization also works with indirect with smaller indirect table.

quoted

quoted

With driver turn around time to repost buffers, device is idle without any

RQ buffers.

quoted

With this improvement, device has 85 buffers instead of 16 to receive

packets.

quoted

Enabling indirect in device can help at cost of 7x higher memory per VQ in

the guest VM.