From: | Peter Avalos <pavalos@xxxxxxxxxxxx> |
Date: | Fri, 9 Sep 2005 00:09:54 -0700 |
Mail-followup-to: | bugs@crater.dragonflybsd.org |
On Wed, Sep 07, 2005 at 05:46:24PM -0700, Matthew Dillon wrote: > > ::After a week of uptime: > :: > ::panic: assertion: tok->t_cpu == gd in lwkt_reltoken > ::panic messages: > > I found another case that could cause the panic. It's a fairly rare > case but 1-week-to-crash is believable if this turns out to be the > cqause. Peter, I would appreciate it if you would run the enclosed > patch in addition to the KTR options I previously suggested. > > Other people with SMP boxes who would like to help test, please run > this patch as well! I will commit it in a week if no issues crop up. > > On a more general note, I am clearly going to have to rethink the whole > token business. I like the API, but my implementation seems a little > more bug-prone then I would like. > Ok, so I applied the patch. After about 20 hours, the machine starting acting weird. It would still ping, but I could not connect via ssh. So, I went to the console. I was able to enter username and password, and after that, nothing. After about 5 minutes, I ctrl-alt-esc'd, and panic'd the machine. The core and kernel is uploading to leaf: ~/public/*.2 --Peter
Attachment:
pgp00004.pgp
Description: PGP signature