On Wed, Sep 07, 2022 at 07:51:38PM +0000, Parav Pandit wrote:

quoted

From: Michael S. Tsirkin <mst@redhat.com>
Sent: Wednesday, September 7, 2022 3:36 PM

On Wed, Sep 07, 2022 at 07:27:16PM +0000, Parav Pandit wrote:

quoted

quoted

From: Michael S. Tsirkin <mst@redhat.com>
Sent: Wednesday, September 7, 2022 3:24 PM

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

quoted

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/PH0PR12MB54812EC7F4711C1EA4CAA119DC

quoted

quoted

419@

quoted

PH0PR12MB5481.namprd12.prod.outlook.com/

quoted

does the device not support indirect?

Yes, indirect feature bit is disabled on the device.

OK that explains it.

So can we proceed with v6 to contain
(a) updated commit message and
(b) function name change you suggested to drop _fields suffix?

(c) replace mtu = 0 with sensibly not calling the function when mtu is
unknown.

quoted

And I'd like commit log to include results of perf testing
- with indirect feature on

Which device do you suggest using for this test?

AFAIK most devices support INDIRECT, e.g. don't nvidia cards do this?

quoted

- with mtu feature off

Why is this needed when it is not touching the area of mtu being not offered?

I don't really like it that instead of checking the MTU feature bit
everywhere the patch sets mtu variable to 0. Because of this
it wasn't all that obvious that the patch did not affect !MTU
performance (the code does change).

Rereading it afresh I think it's ok. But explicit check for !MTU
would be better imho making it obvious we do not need to test !MTU.

-- 
MST