Re: panic: assertion failed in iflib_txq_drain()
- In reply to: Drew Gallatin: "Re: panic: assertion failed in iflib_txq_drain()"
- Go to: [ bottom of page ] [ top of archives ] [ this month ]
Date: Sat, 18 Apr 2026 20:06:17 UTC
> On 18. Apr 2026, at 21:51, Drew Gallatin <gallatin@freebsd.org> wrote: > > > That is indeed the proper fix. I just reviewed & tested it. Thank you so much! Thanks for the quick review. The fix is committed in: https://cgit.FreeBSD.org/src/commit/?id=cca22c36c306dfabe13b1d1de10e8d27ef3c3bce <https://cgit.freebsd.org/src/commit/?id=cca22c36c306dfabe13b1d1de10e8d27ef3c3bce> Best regards Michael > > Until it lands, another workaround is to use the new, cheaper, faster, simpler iflib transmit routine by setting the tunable > > dev.$NIC.$UNIT.iflib.simple_tx=1 > > Eg, dev.bnxt.0.iflib.simple_tx=1 > > In loader.conf > > I plan to deprecate the current complex transmit routine (mp_ring) in favor of the simple mechanism, and I confess that I sometimes don't test "straightforward" changes like this on mp_ring. I'll try to get better about that. > > Drew > > On Sat, Apr 18, 2026, at 3:12 PM, Michael Tuexen wrote: >> >> >> > On 18. Apr 2026, at 21:09, Drew Gallatin <gallatin@freebsd.org> wrote: >> > >> > Sorry, looking into this now. I did not test this change with the mp_ring path.. i don't see how fixing the counters could cause this panic. Looking at it now.. >> Hi Drew, >> >> I think https://reviews.freebsd.org/D56509 fixes it. >> >> Best regards >> Michael >> > >> > Drew >> > >> > On Sat, Apr 18, 2026, at 1:38 PM, Michael Tuexen wrote: >> >> >> >> >> >> > On 18. Apr 2026, at 19:14, David Wolfskill <david@catwhisker.org> wrote: >> >> > >> >> > On Sat, Apr 18, 2026 at 03:53:33PM +0200, Michael Tuexen wrote: >> >> >> ... >> >> >>> So... one of the machines on which I track head got this, as well -- my >> >> >>> (mostly-)headless build machine. (2 laptops, updated in sync with the >> >> >>> build machine, did not panic. One of those also uses a wired NIC.) >> >> >> Do the network drivers of the machines not being affected use iflib? >> >> > >> >> > Apparently not -- they are em(4), iwm(4), & iwn(4), while the panicking >> >> > machine uses igb(4). >> >> > >> >> >> I do see the problem also on one of my machines and local testing shows that >> >> >> https://cgit.FreeBSD.org/src/commit/?id=3fade68cfdf95ee0b517b5d69b270bd8da633404 <https://cgit.freebsd.org/src/commit/?id=3fade68cfdf95ee0b517b5d69b270bd8da633404> >> >> >> is the problem. At least reverting it locally resolves the problem. >> >> >> Can you confirm this? >> >> > >> >> > I can, and do: after reverting main-n285140-3fade68cfdf9 & rebuilding, >> >> > I am able to ssh in without issue; the machine reports: >> >> > >> >> > freebeast(16.0-C)[2] uname -aUK >> >> > FreeBSD freebeast.catwhisker.org 16.0-CURRENT FreeBSD 16.0-CURRENT #549 main-n285180-23a84efeb191: Sat Apr 18 17:01:25 UTC 2026 root@freebeast.catwhisker.org:/common/S3/obj/usr/src/amd64.amd64/sys/GENERIC amd64 1600015 1600015 >> >> > >> >> > Thanks! >> >> You could also disable TSO. I will put up a review for a fix soon. >> >> >> >> Best regards >> >> Michael >> >> > >> >> > Peace, >> >> > david >> >> > -- >> >> > David H. Wolfskill david@catwhisker.org >> >> > >> >> > See https://www.catwhisker.org/~david/publickey.gpg for my public key. >> >> >> >> >> >> >> > >> >> >