panic: SACK scoreboard must not be empty

[Peter Holm]

On Thu, Apr 23, 2015 at 08:33:00AM +1000, Lawrence Stewart wrote:
> On 04/22/15 19:19, Peter Holm wrote:
> > On Wed, Apr 22, 2015 at 11:45:21AM +1000, Lawrence Stewart wrote:
> >> Hi Peter,
> >>
> >> On 04/15/15 21:04, Peter Holm wrote:
> >>> Seen during i386 stress tests.
> >>>
> >>> db_trace_self_wrapper(c11e20fc,0,c11b79d4,1eb,e47578e0,...) at db_trace_self_wrapper+0x2a/frame 0xe47578b0
> >>> kdb_backtrace(c139e637,2,c120851d,e4757984,e4757940,...) at kdb_backtrace+0x2d/frame 0xe4757918
> >>> vpanic(c155b86a,100,c120851d,e4757984,e4757984,...) at vpanic+0x11d/frame 0xe4757954
> >>> kassert_panic(c120851d,0,8,164,16,...) at kassert_panic+0xe6/frame 0xe4757978
> >>> tcp_sack_doack(e047d900,e4757a48,e9eeab30,5f8,c15a9298,...) at tcp_sack_doack+0x405/frame 0xe47579f0
> >>> tcp_do_segment(df801720,e047d900,40,0,0,...) at tcp_do_segment+0x21c2/frame 0xe4757aa0
> >>> tcp_input(e4757bec,e4757be8,6,0,e4757be8,...) at tcp_input+0xfbc/frame 0xe4757ba0
> >>> ip_input(e1cb5800,0,c11f6228,2f0,45,...) at ip_input+0x17c/frame 0xe4757c0c
> >>> swi_net(c1c25880,c11d6fed,54f,8f6da840,c8a209c8,...) at swi_net+0x1a3/frame 0xe4757c4c
> >>> intr_event_execute_handlers(c156a310,c8a20980,c11d6fed,54f,0,...) at intr_event_execute_handlers+0xdb/frame 0xe4757c74
> >>> ithread_loop(c88d4f50,e4757ce8,c11d6d7c,3dc,0,...) at ithread_loop+0x87/frame 0xe4757cac
> >>> fork_exit(c0b2c9d0,c88d4f50,e4757ce8) at fork_exit+0x7e/frame 0xe4757cd4
> >>> fork_trampoline() at fork_trampoline+0x8/frame 0xe4757cd4
> >>>
> >>> Details:
> >>> https://people.freebsd.org/~pho/stress/log/kostik794.txt
> >>>
> >>
> >> >From the looks of the log, you don't have a core?
> >>
> > 
> > Sure, I have.
> > https://people.freebsd.org/~pho/kernel+vmcore.792.txz
> > 
> 
> Oh, great. For some reason I thought I saw a filesystem full message
> related to capturing the dump in the log, but it was just my brain
> playing tricks on me :)
> 
> I'll take a look at this panic. Thanks for the report and crash dump.
> 
> Cheers,
> Lawrence

Lawrence,

I can reproduce this on a i386 kernel, without too much trouble.
It is probably worth noting that the test box is out of VM:
vm_cnt.v_free_count: 2.

Latest panic: https://people.freebsd.org/~pho/stress/log/nfs9-2.txt
http://people.freebsd.org/~pho/kernel+vmcore.959.txz

-- 
Peter