Panic with recent -CURRENT kernel in EC2

[Colin Percival]

On 01/12/18 08:16, Pieper, Jeffrey E wrote:
> Hrm…
> 
>> taskqgroup_attach_cpu: setaffinity failed: 22
>> taskqgroup_attach_cpu: setaffinity failed: 22
> 
> Is this a fresh snapshot? That should have been fixed in
> https://svnweb.freebsd.org/base?view=revision&revision=327013 . Can you
> please file a bug on this? This looks like an iflib issue.

I can confirm that this happens with the 12.0-CURRENT r327524 snasphot (on a
c4.8xlarge EC2 instance, in case it matters).

Colin Percival

> On 1/12/18, 2:43 AM, "owner-freebsd-net at freebsd.org on behalf of Colin Percival" <owner-freebsd-net at freebsd.org on behalf of cperciva at tarsnap.com> wrote:
> 
>     This sounds like it may be related to r327031, which make ixgbe (and thus ixv)
>     use iflib... maybe erj@ will have some ideas here?
>     
>     In case it helps, dhclient sets the MTU, which in the past has caused
>     problems due to drivers resetting -- the fact that sys_ioctl is ending
>     up in iflib_init_locked makes me wonder if this is a similar issue.  (If
>     it is, there's a locking problem somewhere...)
>     
>     Colin Percival
>     
>     On 01/12/18 01:23, Bradley T. Hughes wrote:
>     > (Cross-posting to -net, but I'm not subscribed there... yet)
>     > 
>     > I was wondering if anyone else has experienced this panic when dhclient
>     tries to get an IP address after booting a recent 12.0-CURRENT kernel? Using
>     the 12.0-CURRENT AMI from early November works, but with the latest snapshot
>     or when using a kernel built from source I get this:
>     > 
>     > ixv0: <Intel(R) PRO/10GbE Virtual Function Network Driver> mem 0xf3000000-0xf3003fff,0xf3004000-0xf3007fff at device 3.0 on pci0
>     > ixv0: using 2048 tx descriptors and 2048 rx descriptors
>     > ixv0: msix_init qsets capped at 1
>     > ixv0: pxm cpus: 2 queue msgs: 2 admincnt: 1
>     > ixv0: using 1 rx queues 1 tx queues 
>     > ixv0: Using MSIX interrupts with 2 vectors
>     > ixv0: allocated for 1 queues
>     > ixv0: allocated for 1 rx queues
>     > taskqgroup_attach_cpu: setaffinity failed: 22
>     > taskqgroup_attach_cpu: setaffinity failed: 22
>     > ixv0: Ethernet address: 0a:ad:68:70:ea:78
>     > ixv0: netmap queues/slots: TX 1/2048, RX 1/2048
>     > ...
>     > panic: Assertion if_getdrvflags(ifp) == i failed at /usr/src/sys/net/iflib.c:2199
>     > cpuid = 3
>     > time = 1515673176
>     > KDB: stack backtrace:
>     > db_trace_self_wrapper() at db_trace_self_wrapper+0x2b/frame 0xfffffe008a8ee600
>     > vpanic() at vpanic+0x19c/frame 0xfffffe008a8ee680
>     > kassert_panic() at kassert_panic+0x126/frame 0xfffffe008a8ee6f0
>     > iflib_init_locked() at iflib_init_locked+0x75f/frame 0xfffffe008a8ee750
>     > iflib_if_ioctl() at iflib_if_ioctl+0x744/frame 0xfffffe008a8ee7b0
>     > ifioctl() at ifioctl+0x1903/frame 0xfffffe008a8ee850
>     > kern_ioctl() at kern_ioctl+0x2c4/frame 0xfffffe008a8ee8b0
>     > sys_ioctl() at sys_ioctl+0x15c/frame 0xfffffe008a8ee980
>     > amd64_syscall() at amd64_syscall+0x79b/frame 0xfffffe008a8eeab0
>     > Xfast_syscall() at Xfast_syscall+0xfb/frame 0xfffffe008a8eeab0
>     > --- syscall (54, FreeBSD ELF64, sys_ioctl), rip = 0x80100a4fa, rsp = 0x7fffffffe408, rbp = 0x7fffffffe460 ---
>     > 
>     > Thanks in advance :)

-- 
Colin Percival
Security Officer Emeritus, FreeBSD | The power to serve
Founder, Tarsnap | www.tarsnap.com | Online backups for the truly paranoid