Re: sky2 panic in 2.6.32.1 under load (new oops)

[Michael Breuer <mbreuer@majjas.com>]

More results... Seems that this is related to the tty (or maybe vt or 
maybe console) driver. I had disabled mingetty on tty1 as a workaround 
to a Fedora/KDE issue. When I reenabled it, I was then able to recreate 
this problem (sky2) in runlevel 3.

Retesting, it seems that the following conditions must be met in order 
for the interrupt errors to happen:

1) mingetty running on tty1 before the start of the test run
2) High sustained incoming load - 25MB/sec or more (as reported by nethogs)
3) I've only been able to recreate this using samba & a Windows 7 backup 
- but there could be other triggers

As it happens, much of my other data all came down to mingetty starting 
or not starting on tty1 at specific points of test preparation. At this 
point I can recreate 100% of the time if I ensure that there is a 
mingetty running on tty1, regardless of runlevel; and I can not recreate 
100% of the time when there is no mingetty running on tty1 at the start 
of the test. The results do not change if I start or stop mingetty on 
tty1 during the test. It's possible that X running on tty1 may also 
trigger the issue, but I haven't tested that. Note that running mingetty 
on the other tty's does not affect the outcome of the test. The test is 
also unaffected if I chvt to or away from tty1 before the test. I do not 
have to be logged in at the console (on any vt) to trigger the issue.

In a nutshell, to recreate: make sure there is a mingetty running on 
tty1, blast traffic into the system on eth0 (sky2).




On 12/31/2009 1:09 PM, Michael Breuer wrote:
> Did some more digging today... Haven't nailed it, but there's 
> something going on with X and tty...
>
> Among other things, when I telinit 3 && telinit 5 the tty keeps 
> switching between 7 and 8 (nothing else running on either tty). It 
> would appear that somehow the tty deallocation isn't complete when X 
> restarts. Also, X grabbing a tty seems to be a requisite step in 
> recreating the sky2 issue.
>
> On 12/30/2009 1:10 PM, Stephen Hemminger wrote:
>> On Wed, 30 Dec 2009 10:40:56 -0500
>> Michael Breuer<mbreuer@majjas.com>  wrote:
>>
>>> Dec 30 10:37:14 mail kernel: sky2 0000:06:00.0: error interrupt
>>> status=0x40000008
>>> Dec 30 10:37:14 mail kernel: sky2 Tx ring pending=31...40 report=35 
>>> done=35
>>> Dec 30 10:37:14 mail kernel: 30: 0x0: 0xeb50c0ca(9014)
>>> Dec 30 10:37:14 mail kernel: 32: 0x0: 0xeb5080ca(9014)
>>> Dec 30 10:37:14 mail kernel: 34: 0x0: 0xeb5040ca(9014)
>>> Dec 30 10:37:14 mail kernel: 36: 0x0: 0xeb5000ca(9014)
>>> Dec 30 10:37:14 mail kernel: 38: 0x0: 0xeb4fc0ca(9014)
>> You must be using 9K MTU.  Do you see the problem with smaller
>> MTU?
>>
>>> Dec 30 10:37:14 mail kernel: sky2 0000:06:00.0: error interrupt
>>> status=0x40000008
>>> Dec 30 10:37:14 mail kernel: sky2 Tx ring pending=87...92 report=89 
>>> done=91
>>> Dec 30 10:37:14 mail kernel: 86: 0x0: 0xeafd80ca(9014)
>>> Dec 30 10:37:14 mail kernel: 88: 0x0: 0xeafd40ca(9014)
>>> Dec 30 10:37:14 mail kernel: 90: 0x0: 0xeafd00ca(9014)
>>> Dec 30 10:37:15 mail kernel: sky2 0000:06:00.0: error interrupt 
>>> status=0x8
>>> Dec 30 10:37:15 mail kernel: sky2 Tx ring pending=39...42 report=39 
>>> done=41
>>> Dec 30 10:37:15 mail kernel: 38: 0x0: 0xe5fec0ca(9014)
>>> Dec 30 10:37:15 mail kernel: 40: 0x0: 0xe5fe80ca(9014)
>>> Dec 30 10:37:15 mail kernel: sky2 0000:06:00.0: error interrupt
>>> status=0x40000008
>>> Dec 30 10:37:15 mail kernel: sky2 Tx ring pending=117...124 report=119
>>> done=119
>>> Dec 30 10:37:15 mail kernel: 116: 0x0: 0xe5d6c0ca(9014)
>>> Dec 30 10:37:15 mail kernel: 118: 0x0: 0xe5d680ca(9014)
>>> Dec 30 10:37:15 mail kernel: 120: 0x0: 0xe5d660ca(5938)
>>> Dec 30 10:37:15 mail kernel: 122: 0x0: 0xe5d640ca(4213)
>