Re: [PATCH net-next 00/13] selftests: ncdevmem: Add ncdevmem to ksft
From: Stanislav Fomichev <hidden>
Date: 2024-12-13 20:33:55
On 12/13, Mina Almasry wrote:
On Thu, Dec 12, 2024 at 5:19 PM Stanislav Fomichev [off-list ref] wrote:quoted
On 09/12, Stanislav Fomichev wrote:quoted
On 09/12, Mina Almasry wrote:quoted
On Thu, Sep 12, 2024 at 12:48 PM Stanislav Fomichev [off-list ref] wrote:quoted
On 09/12, Stanislav Fomichev wrote:quoted
The goal of the series is to simplify and make it possible to use ncdevmem in an automated way from the ksft python wrapper. ncdevmem is slowly mutated into a state where it uses stdout to print the payload and the python wrapper is added to make sure the arrived payload matches the expected one.Mina, what's your plan/progress on the upstreamable TX side? I hope you're still gonna finish it up?I'm very open to someone pushing the TX side, but there is a bit of a need here to get the TX side done sooner than later. In reality I don't think anyone cares as much as me to push this ASAP so I plan/hope to look into it. I have made some progress but a bit to be worked through at the moment. I hope to have something ready as the merge window reopens; very likely doable.Perfect!Hey Mina, Any updates on this? Any chance getting something out this merge window? I was hoping you'd post something in the previous merge window (late Sep), but if you're still busy with other things, should I post a v2 RFC? I have moved to the mode which you suggested where tx dmabuf is associated with the socket; this lets me drop all tx ref counts (socket holds dmabuf, skb holds socket until tx completion). I also moved to a more performant offset->dma_addr resolution logic in tcp_sendmsg and fixed a bunch of things on the test side...My apologies for the delay. I have the TX path ready, but having trouble running the performance tests unrelated to the patch itself. I'm going to send the TX path after a round of internal reviews, likely by the end of next week.
Awesome, looking forward to it! I put whatever I have on my side on gh as well: https://github.com/fomichev/linux/commits/upstream/net-next/tx/2/ Take a look in case something makes sense to update on your side. I reworked the test to a more conventional 'cfg.remote.deploy' model and I also simplified lookup of dma_addr on tx (my previous tx code was super slow and buggy).