* Re: [PATCH net 0/6][pull request] Intel Wired LAN Driver Updates 2026-11-13 (ice) [not found] <20261113231225.1864810-1-anthony.l.nguyen@intel.com> @ 2026-11-18 11:59 ` Petr Oros 2026-11-18 13:37 ` Magnus Karlsson 0 siblings, 1 reply; 5+ messages in thread From: Petr Oros @ 2026-11-18 11:59 UTC (permalink / raw) To: Tony Nguyen Cc: Petr Oros, netdev, Jakub Kicinski, Paolo Abeni, edumazet, davem, andrew+netdev, horms, Jacob Keller, Przemek Kitszel, Maciej Fijalkowski, Marcin Szycik, Aleksandr Loktionov, intel-wired-lan I think this pull is a good moment to ask how fixes for Intel drivers should flow, because it is not the first time a fix spent months on intel-wired-lan and then hit a wall on net. My patches 2/6 and 3/6 have been on intel-wired-lan since June, v4 on July 8 [1]. They were reviewed and tested by Intel validation there, and four months later Sashiko found real problems in them on this pull. I have a v5 and will send it once it is tested. Jake's E82x timestamp fixes went through three pulls before a part of them was applied [2], and Maciej described the same loop on the i40e pull in September [3]. I know Intel patches go through iwl and are not applied to net directly [4]. But for fixes it means that the review which decides happens on netdev months after the iwl review and validation, and every respin goes through both again. And if iwl stops cross-posting to netdev, as Jakub suggested [5], Sashiko will see Intel fixes only in the pull. So does the iwl round still make sense for fixes? Would it be better to send fixes straight to net with iwl in Cc, and have Tony's ack and the Tested-by from Intel validation on netdev? [1] https://lore.kernel.org/all/20260708125755.706263-1-poros@redhat.com/ [2] https://lore.kernel.org/all/20261021-jk-e825c-timestamp-processing-logic-fixes-srcu-v5-0-72289a4ac8eb@intel.com/ [3] https://lore.kernel.org/all/are4rBUJO3mzrkO8@boxer/ [4] https://lore.kernel.org/all/20260721174642.7749143e@kernel.org/ [5] https://lore.kernel.org/all/20260922112216.41e6f8cf@kernel.org/ Regards, Petr ^ permalink raw reply [flat|nested] 5+ messages in thread
* Re: [PATCH net 0/6][pull request] Intel Wired LAN Driver Updates 2026-11-13 (ice) 2026-11-18 11:59 ` [PATCH net 0/6][pull request] Intel Wired LAN Driver Updates 2026-11-13 (ice) Petr Oros @ 2026-11-18 13:37 ` Magnus Karlsson 2026-11-18 13:48 ` Maciej Fijalkowski 0 siblings, 1 reply; 5+ messages in thread From: Magnus Karlsson @ 2026-11-18 13:37 UTC (permalink / raw) To: Petr Oros Cc: Tony Nguyen, netdev, Jakub Kicinski, Paolo Abeni, edumazet, davem, andrew+netdev, horms, Jacob Keller, Przemek Kitszel, Maciej Fijalkowski, Marcin Szycik, Aleksandr Loktionov, intel-wired-lan On Wed, 18 Nov 2026 at 13:00, Petr Oros <poros@redhat.com> wrote: > > I think this pull is a good moment to ask how fixes for Intel drivers > should flow, because it is not the first time a fix spent months on > intel-wired-lan and then hit a wall on net. > > My patches 2/6 and 3/6 have been on intel-wired-lan since June, v4 on > July 8 [1]. They were reviewed and tested by Intel validation there, > and four months later Sashiko found real problems in them on this > pull. I have a v5 and will send it once it is tested. > > Jake's E82x timestamp fixes went through three pulls before a part of > them was applied [2], and Maciej described the same loop on the i40e > pull in September [3]. > > I know Intel patches go through iwl and are not applied to net > directly [4]. But for fixes it means that the review which decides > happens on netdev months after the iwl review and validation, and every > respin goes through both again. And if iwl stops cross-posting to > netdev, as Jakub suggested [5], Sashiko will see Intel fixes only in > the pull. > > So does the iwl round still make sense for fixes? Would it be better to > send fixes straight to net with iwl in Cc, and have Tony's ack and the > Tested-by from Intel validation on netdev? Glad you brought this up. IMO, it makes no sense for fixes to go through iwl and it makes contributing fixes (or new features/optimizations) a lot more painful. Sometimes even to the point that you just stop contributing something. I am not in the NIC team at Intel, but sometimes I have to do network core work that also touches our drivers. In this case, that patch set goes straight to net-next (as iwl cannot and should not take core patches), but any driver fix that I find on the way needs to go to iwl-net so that it can go to net and be backported. So the fixes that my patch set are dependent on and that should go in first are instead sent to a slow-path (i.e. iwl-net) and after X weeks/months appear in net and then net-next. But my patch set in net-next is dependent on these fixes so I have to either wait or just cross my fingers that my new feature is stable enough even without those fixes. Not optimal in my opinion. So anything that could be done to speed up this process would be very welcome. Sending fixes straight to net would be great. The validation team could pick those patches up from the list and run the regression tests against them, or would that take too long? Anyway, good initiative. I am sure there is a solution to this that we will all like. /Magnus > [1] https://lore.kernel.org/all/20260708125755.706263-1-poros@redhat.com/ > [2] https://lore.kernel.org/all/20261021-jk-e825c-timestamp-processing-logic-fixes-srcu-v5-0-72289a4ac8eb@intel.com/ > [3] https://lore.kernel.org/all/are4rBUJO3mzrkO8@boxer/ > [4] https://lore.kernel.org/all/20260721174642.7749143e@kernel.org/ > [5] https://lore.kernel.org/all/20260922112216.41e6f8cf@kernel.org/ > > Regards, > Petr > > ^ permalink raw reply [flat|nested] 5+ messages in thread
* Re: [PATCH net 0/6][pull request] Intel Wired LAN Driver Updates 2026-11-13 (ice) 2026-11-18 13:37 ` Magnus Karlsson @ 2026-11-18 13:48 ` Maciej Fijalkowski 2026-11-18 13:51 ` Petr Oros 0 siblings, 1 reply; 5+ messages in thread From: Maciej Fijalkowski @ 2026-11-18 13:48 UTC (permalink / raw) To: Magnus Karlsson Cc: Petr Oros, Tony Nguyen, netdev, Jakub Kicinski, Paolo Abeni, edumazet, davem, andrew+netdev, horms, Jacob Keller, Przemek Kitszel, Marcin Szycik, Aleksandr Loktionov, intel-wired-lan On Wed, Nov 18, 2026 at 02:37:06PM +0100, Magnus Karlsson wrote: > On Wed, 18 Nov 2026 at 13:00, Petr Oros <poros@redhat.com> wrote: > > > > I think this pull is a good moment to ask how fixes for Intel drivers > > should flow, because it is not the first time a fix spent months on > > intel-wired-lan and then hit a wall on net. > > > > My patches 2/6 and 3/6 have been on intel-wired-lan since June, v4 on > > July 8 [1]. They were reviewed and tested by Intel validation there, > > and four months later Sashiko found real problems in them on this > > pull. I have a v5 and will send it once it is tested. > > > > Jake's E82x timestamp fixes went through three pulls before a part of > > them was applied [2], and Maciej described the same loop on the i40e > > pull in September [3]. > > > > I know Intel patches go through iwl and are not applied to net > > directly [4]. But for fixes it means that the review which decides > > happens on netdev months after the iwl review and validation, and every > > respin goes through both again. And if iwl stops cross-posting to > > netdev, as Jakub suggested [5], Sashiko will see Intel fixes only in > > the pull. > > > > So does the iwl round still make sense for fixes? Would it be better to > > send fixes straight to net with iwl in Cc, and have Tony's ack and the > > Tested-by from Intel validation on netdev? > > Glad you brought this up. IMO, it makes no sense for fixes to go > through iwl and it makes contributing fixes (or new > features/optimizations) a lot more painful. Sometimes even to the > point that you just stop contributing something. I am not in the NIC > team at Intel, but sometimes I have to do network core work that also > touches our drivers. In this case, that patch set goes straight to > net-next (as iwl cannot and should not take core patches), but any > driver fix that I find on the way needs to go to iwl-net so that it > can go to net and be backported. So the fixes that my patch set are > dependent on and that should go in first are instead sent to a > slow-path (i.e. iwl-net) and after X weeks/months appear in net and > then net-next. But my patch set in net-next is dependent on these > fixes so I have to either wait or just cross my fingers that my new > feature is stable enough even without those fixes. Not optimal in my > opinion. > > So anything that could be done to speed up this process would be very > welcome. Sending fixes straight to net would be great. The validation > team could pick those patches up from the list and run the regression > tests against them, or would that take too long? Regarding providing Tested-by tags, could we lift the policy where only Intel validation is allowed to bless the patch? Nowadays with AI regular dev can spin up a nice pile of tests by themselves. So if you send a fix and you can find a peer that will test and review it for you, I don't think we have to wait for validation folks taking care of it if they are swamped. I also don't think it would backfire on us and we would introduce some bloody regressions, given we have {s,cl}ashikos in the pipeline. > > Anyway, good initiative. I am sure there is a solution to this that we > will all like. > > /Magnus > > > [1] https://lore.kernel.org/all/20260708125755.706263-1-poros@redhat.com/ > > [2] https://lore.kernel.org/all/20261021-jk-e825c-timestamp-processing-logic-fixes-srcu-v5-0-72289a4ac8eb@intel.com/ > > [3] https://lore.kernel.org/all/are4rBUJO3mzrkO8@boxer/ > > [4] https://lore.kernel.org/all/20260721174642.7749143e@kernel.org/ > > [5] https://lore.kernel.org/all/20260922112216.41e6f8cf@kernel.org/ > > > > Regards, > > Petr > > > > ^ permalink raw reply [flat|nested] 5+ messages in thread
* Re: [PATCH net 0/6][pull request] Intel Wired LAN Driver Updates 2026-11-13 (ice) 2026-11-18 13:48 ` Maciej Fijalkowski @ 2026-11-18 13:51 ` Petr Oros 2026-11-18 23:00 ` Jacob Keller 0 siblings, 1 reply; 5+ messages in thread From: Petr Oros @ 2026-11-18 13:51 UTC (permalink / raw) To: Maciej Fijalkowski, Magnus Karlsson Cc: Tony Nguyen, netdev, Jakub Kicinski, Paolo Abeni, edumazet, davem, andrew+netdev, horms, Jacob Keller, Przemek Kitszel, Marcin Szycik, Aleksandr Loktionov, intel-wired-lan On 11/18/26 2:48 PM, Maciej Fijalkowski wrote: > On Wed, Nov 18, 2026 at 02:37:06PM +0100, Magnus Karlsson wrote: >> On Wed, 18 Nov 2026 at 13:00, Petr Oros <poros@redhat.com> wrote: >>> >>> I think this pull is a good moment to ask how fixes for Intel drivers >>> should flow, because it is not the first time a fix spent months on >>> intel-wired-lan and then hit a wall on net. >>> >>> My patches 2/6 and 3/6 have been on intel-wired-lan since June, v4 on >>> July 8 [1]. They were reviewed and tested by Intel validation there, >>> and four months later Sashiko found real problems in them on this >>> pull. I have a v5 and will send it once it is tested. >>> >>> Jake's E82x timestamp fixes went through three pulls before a part of >>> them was applied [2], and Maciej described the same loop on the i40e >>> pull in September [3]. >>> >>> I know Intel patches go through iwl and are not applied to net >>> directly [4]. But for fixes it means that the review which decides >>> happens on netdev months after the iwl review and validation, and every >>> respin goes through both again. And if iwl stops cross-posting to >>> netdev, as Jakub suggested [5], Sashiko will see Intel fixes only in >>> the pull. >>> >>> So does the iwl round still make sense for fixes? Would it be better to >>> send fixes straight to net with iwl in Cc, and have Tony's ack and the >>> Tested-by from Intel validation on netdev? >> >> Glad you brought this up. IMO, it makes no sense for fixes to go >> through iwl and it makes contributing fixes (or new >> features/optimizations) a lot more painful. Sometimes even to the >> point that you just stop contributing something. I am not in the NIC >> team at Intel, but sometimes I have to do network core work that also >> touches our drivers. In this case, that patch set goes straight to >> net-next (as iwl cannot and should not take core patches), but any >> driver fix that I find on the way needs to go to iwl-net so that it >> can go to net and be backported. So the fixes that my patch set are >> dependent on and that should go in first are instead sent to a >> slow-path (i.e. iwl-net) and after X weeks/months appear in net and >> then net-next. But my patch set in net-next is dependent on these >> fixes so I have to either wait or just cross my fingers that my new >> feature is stable enough even without those fixes. Not optimal in my >> opinion. >> >> So anything that could be done to speed up this process would be very >> welcome. Sending fixes straight to net would be great. The validation >> team could pick those patches up from the list and run the regression >> tests against them, or would that take too long? > > Regarding providing Tested-by tags, could we lift the policy where only > Intel validation is allowed to bless the patch? Nowadays with AI regular > dev can spin up a nice pile of tests by themselves. > > So if you send a fix and you can find a peer that will test and review it > for you, I don't think we have to wait for validation folks taking care of > it if they are swamped. > > I also don't think it would backfire on us and we would introduce some > bloody regressions, given we have {s,cl}ashikos in the pipeline. BIG +1 > >> >> Anyway, good initiative. I am sure there is a solution to this that we >> will all like. >> >> /Magnus >> >>> [1] https://lore.kernel.org/all/20260708125755.706263-1-poros@redhat.com/ >>> [2] https://lore.kernel.org/all/20261021-jk-e825c-timestamp-processing-logic-fixes-srcu-v5-0-72289a4ac8eb@intel.com/ >>> [3] https://lore.kernel.org/all/are4rBUJO3mzrkO8@boxer/ >>> [4] https://lore.kernel.org/all/20260721174642.7749143e@kernel.org/ >>> [5] https://lore.kernel.org/all/20260922112216.41e6f8cf@kernel.org/ >>> >>> Regards, >>> Petr >>> >>> > ^ permalink raw reply [flat|nested] 5+ messages in thread
* Re: [PATCH net 0/6][pull request] Intel Wired LAN Driver Updates 2026-11-13 (ice) 2026-11-18 13:51 ` Petr Oros @ 2026-11-18 23:00 ` Jacob Keller 0 siblings, 0 replies; 5+ messages in thread From: Jacob Keller @ 2026-11-18 23:00 UTC (permalink / raw) To: Petr Oros, Maciej Fijalkowski, Magnus Karlsson Cc: Tony Nguyen, netdev, Jakub Kicinski, Paolo Abeni, edumazet, davem, andrew+netdev, horms, Przemek Kitszel, Marcin Szycik, Aleksandr Loktionov, intel-wired-lan On 11/18/2026 5:51 AM, Petr Oros wrote: > > > On 11/18/26 2:48 PM, Maciej Fijalkowski wrote: >> On Wed, Nov 18, 2026 at 02:37:06PM +0100, Magnus Karlsson wrote: >>> On Wed, 18 Nov 2026 at 13:00, Petr Oros <poros@redhat.com> wrote: >>>> >>>> I think this pull is a good moment to ask how fixes for Intel drivers >>>> should flow, because it is not the first time a fix spent months on >>>> intel-wired-lan and then hit a wall on net. >>>> >>>> My patches 2/6 and 3/6 have been on intel-wired-lan since June, v4 on >>>> July 8 [1]. They were reviewed and tested by Intel validation there, >>>> and four months later Sashiko found real problems in them on this >>>> pull. I have a v5 and will send it once it is tested. >>>> >>>> Jake's E82x timestamp fixes went through three pulls before a part of >>>> them was applied [2], and Maciej described the same loop on the i40e >>>> pull in September [3]. >>>> Part of this is that local Sashiko runs do not seem to mimic the same results as netdev Sashiko does. Partly this is due to lack of access to models, and possibly differences from the version netdev runs. I am still trying to figure out how to get our local setup working within the netdev fork, but haven't had good success yet. My series ultimately blocked Tony (and myself while I was covering for him) from submitting anything else for *months*, due to its size and scope. This is the result of the 15 patch limit of outstanding submissions. Because only Tony "officially" submits work to netdev, all of the Intel drivers are limited to effectively 15 patches per week at best (assuming 48 hrs to resubmit a new revision after feedback and needing at least one revision before being accepted, which seems to be the case these days, if not worse). This is across multiple supported drivers including ixgbe, i40e, ice, idpf, now ixd, and the client drivers like igc and e1000e. >>>> I know Intel patches go through iwl and are not applied to net >>>> directly [4]. But for fixes it means that the review which decides >>>> happens on netdev months after the iwl review and validation, and every >>>> respin goes through both again. And if iwl stops cross-posting to >>>> netdev, as Jakub suggested [5], Sashiko will see Intel fixes only in >>>> the pull. >>>> Yep. Right now we get reports from Sashiko.dev for Intel Wired LAN but these only seem to report a *fraction* of what netdev does, and my own attempts at internal reproduction also fail to reproduce the same problems while consuming a large part of my allotted token budget. Using the larger models sometimes helps but rapidly burns my available budget. >>>> So does the iwl round still make sense for fixes? Would it be better to >>>> send fixes straight to net with iwl in Cc, and have Tony's ack and the >>>> Tested-by from Intel validation on netdev? >>> >>> Glad you brought this up. IMO, it makes no sense for fixes to go >>> through iwl and it makes contributing fixes (or new >>> features/optimizations) a lot more painful. Sometimes even to the >>> point that you just stop contributing something. I am not in the NIC >>> team at Intel, but sometimes I have to do network core work that also >>> touches our drivers. In this case, that patch set goes straight to >>> net-next (as iwl cannot and should not take core patches), but any >>> driver fix that I find on the way needs to go to iwl-net so that it >>> can go to net and be backported. So the fixes that my patch set are >>> dependent on and that should go in first are instead sent to a >>> slow-path (i.e. iwl-net) and after X weeks/months appear in net and >>> then net-next. But my patch set in net-next is dependent on these >>> fixes so I have to either wait or just cross my fingers that my new >>> feature is stable enough even without those fixes. Not optimal in my >>> opinion. >>> >>> So anything that could be done to speed up this process would be very >>> welcome. Sending fixes straight to net would be great. The validation >>> team could pick those patches up from the list and run the regression >>> tests against them, or would that take too long? >> >> Regarding providing Tested-by tags, could we lift the policy where only >> Intel validation is allowed to bless the patch? Nowadays with AI regular >> dev can spin up a nice pile of tests by themselves. >> >> So if you send a fix and you can find a peer that will test and review it >> for you, I don't think we have to wait for validation folks taking >> care of >> it if they are swamped. >> >> I also don't think it would backfire on us and we would introduce some >> bloody regressions, given we have {s,cl}ashikos in the pipeline. > > BIG +1 > This is really a question for the maintainers. I suspect that they may not want to increase the number of submissions. Checking now, the netdevbpf patchwork is already at nearly 1200 patches waiting, with the oldest being only from the 4th. By contrast, Intel Wired LAN has ~300ish patches in the queue with the oldest being from March 27th. Personally, I'd be happy with anything that increases our velocity of getting fixes out. Frankly, I'm very frustrated by the run around with Sashiko AI reviews. I do find that the end result has been for a higher quality, but the delays and difficulties are straining. Its completely understandable that many other contributors have pulled back as the burden increase. Part of the problem is our own mistakes and legacy of poor driver code. As a result, we're accumulating more fixes than we can submit with the 15 patch limit, and this doesn't seem to be slowing. The rate of acceptance is certainly not going up. Plus pretty much every bug fix I've worked on in the last year has come with 2 to 3 other "side quests" reported by Sashiko that need investigation and development. I'm not sure what compromise the netdev maintainers would accept, given they have already repeatedly stated that they're flooded with work. However, I think Intel Wired LAN is nearly at a breaking point given how long fixes sit in the queue, blocked while waiting on other priorities. >> >>> >>> Anyway, good initiative. I am sure there is a solution to this that we >>> will all like. >>> >>> /Magnus >>> >>>> [1] https://lore.kernel.org/all/20260708125755.706263-1- >>>> poros@redhat.com/ >>>> [2] https://lore.kernel.org/all/20261021-jk-e825c-timestamp- >>>> processing-logic-fixes-srcu-v5-0-72289a4ac8eb@intel.com/ >>>> [3] https://lore.kernel.org/all/are4rBUJO3mzrkO8@boxer/ >>>> [4] https://lore.kernel.org/all/20260721174642.7749143e@kernel.org/ >>>> [5] https://lore.kernel.org/all/20260922112216.41e6f8cf@kernel.org/ >>>> >>>> Regards, >>>> Petr >>>> >>>> >> > ^ permalink raw reply [flat|nested] 5+ messages in thread
end of thread, other threads:[~2026-11-18 23:01 UTC | newest]
Thread overview: 5+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
[not found] <20261113231225.1864810-1-anthony.l.nguyen@intel.com>
2026-11-18 11:59 ` [PATCH net 0/6][pull request] Intel Wired LAN Driver Updates 2026-11-13 (ice) Petr Oros
2026-11-18 13:37 ` Magnus Karlsson
2026-11-18 13:48 ` Maciej Fijalkowski
2026-11-18 13:51 ` Petr Oros
2026-11-18 23:00 ` Jacob Keller
This is a public inbox, see mirroring instructions for how to clone and mirror all data and code used for this inbox