RE: [PATCH iwl-net v1] igc: fix race condition in TX timestamp read for register 0

["Bouska, Zdenek" <zdenek.bouska@siemens.com>]

On 18/09/2025 19:38, Chwee-Lin Choong wrote:
> This fix also prevents TX unit hangs observed under heavy
> timestamping load.
> 

I have tested this patch with heavy TX timestamps load.

I have not reproduced any hangs when I apply this patch to 6.15.0.

But when I apply it to 6.17.0, then I still have some hangs, but not as much as without the patch on 6.17.0.

Used git bisect with this patch and I found that this is the first commit where this patch does not avoid all hangs:

commit 0d58cdc902dace2298406b3972def04f55e3d775
Author: Faizal Rahim <faizal.abdul.rahim@linux.intel.com>
Date:   Mon Mar 17 23:07:35 2025 -0400

    igc: optimize TX packet buffer utilization for TSN mode

    In preparation for upcoming frame preemption patches, optimize the TX
    packet buffer size. The total packet buffer size (RX + TX) is 64KB, with
    a maximum of 34KB for either RX or TX. Split the buffer evenly,
    allocating 32KB to each.

    For TX, assign 7KB to each of the four TX packet buffers (total 28KB)
    and reserve 4KB for BMC.

When I revert this commit and apply this patch on 6.17.0 I have no hangs.

Best regards,
Zdenek Bouska

--
Siemens, s.r.o
Foundational Technologies