On Wed, Feb 19, 2025 at 03:03:36PM +0100, Kurt Kanzenbach wrote: > On Tue Feb 18 2025, Joe Damato wrote: > > On Tue, Feb 18, 2025 at 04:18:19PM -0500, Joe Damato wrote: > >> On Mon, Feb 17, 2025 at 12:31:20PM +0100, Kurt Kanzenbach wrote: > >> > This is a follow up for the igb XDP/ZC implementation. The first two > >> > patches link the IRQs and queues to NAPI instances. This is required to > >> > bring back the XDP/ZC busy polling support. The last patch removes > >> > undesired IRQs (injected via igb watchdog) while busy polling with > >> > napi_defer_hard_irqs and gro_flush_timeout set. > >> > > >> > Signed-off-by: Kurt Kanzenbach <[email protected]> > >> > --- > >> > Changes in v2: > >> > - Take RTNL lock in PCI error handlers (Joe) > >> > - Fix typo in commit message (Gerhard) > >> > - Use netif_napi_add_config() (Joe) > >> > - Link to v1: > >> > https://lore.kernel.org/r/[email protected] > >> > >> Thanks for sending a v2. > >> > >> My comment from the previous series still stands, which simply that > >> I have no idea if the maintainers will accept changes using this API > >> or prefer to wait until Stanislav's work [1] is completed to remove > >> the RTNL requirement from this API altogether. > > > > Also, may be worth running the newly added XSK test with the NETIF > > env var set to the igb device? Assuming eth0 is your igb device: > > > > NETIF=eth0 ./tools/testing/selftests/drivers/net/queues.py > > > > should output: > > > > KTAP version 1 > > 1..4 > > ok 1 queues.get_queues > > ok 2 queues.addremove_queues > > ok 3 queues.check_down > > ok 4 queues.check_xdp > > # Totals: pass:4 fail:0 xfail:0 xpass:0 skip:0 error:0 > > > > Note the check_xdp line above. > > > > Sure, why not. Seems to work.
Thanks for testing it. > |root@apl1:~/linux# uname -a > |Linux apl1 6.14.0-rc2+ #2 SMP PREEMPT_RT Wed Feb 19 14:41:23 CET 2025 x86_64 > GNU/Linux > |root@apl1:~/linux# NETIF=enp2s0 > ./tools/testing/selftests/drivers/net/queues.py > |KTAP version 1 > |1..4 > |ok 1 queues.get_queues > |ok 2 queues.addremove_queues > |ok 3 queues.check_down > |ok 4 queues.check_xdp > |# Totals: pass:4 fail:0 xfail:0 xpass:0 skip:0 error:0 > > Has this xsk netlink attribute been added fairly recently? The test > failed on my kernel from a few days ago (kernel from today works). Yes, it was just merged, see the commit date here: https://git.kernel.org/pub/scm/linux/kernel/git/netdev/net-next.git/commit/?id=788e52e2b66844301fe09f3372d46d8c62f6ebe4 > I think there's room for improvement though: > > |root@apl1:~/linux# NETIF=enp2s0 > ./tools/testing/selftests/drivers/net/queues.py > |KTAP version 1 > |1..4 > |ok 1 queues.get_queues > |ok 2 queues.addremove_queues > |ok 3 queues.check_down > |# Exception| Traceback (most recent call last): > |# Exception| File > "/root/linux/tools/testing/selftests/net/lib/py/ksft.py", line 218, in > ksft_run > |# Exception| case(*args) > |# Exception| File > "/root/linux/./tools/testing/selftests/drivers/net/queues.py", line 53, in > check_xdp > |# Exception| ksft_eq(q['xsk'], {}) > |# Exception| ~^^^^^^^ > |# Exception| KeyError: 'xsk' > |not ok 4 queues.check_xdp > |# Totals: pass:3 fail:1 xfail:0 xpass:0 skip:0 error:0 > > I'd assume this shouldn't be a Python exception, but rather say > something like "Expected xsk attribute, but none found. Fix the driver!" :) > > While at it would you mind to add a newline to the xdp_helper usage > line (and fix the one typo)? Jakub currently has a series out to change the test a bit and improve it overall, see: https://lore.kernel.org/netdev/[email protected]/ It looks like your concerns (the typo, newline, and better error) may still remain. If so, I can submit a follow-up once his work has been merged to address your concerns - unless you'd like to do that?
