![]() | |||||||||||||
Messages in this thread |
On 22/11/06, Linus Torvalds <torvalds@osdl.org> wrote: > > > On Wed, 22 Nov 2006, Jesper Juhl wrote: > > > > So it *seems* to be somehow related to running low on RAM and swap > > starting to be used. > > Does it happen if you just do some simple "use all memory" script, eg run > a few copies of > > #define SIZE (100<<20) > > char *buf = malloc(SIZE); > memset(buf, SIZE, 0); > sleep(100); > > on your box? > I'll try, when I get home from work. I'll let you know later. > > The box has 2GB of RAM and 768MB swap. > > I wonder.. It _used_ to be true that we were pretty good at making swap be > "extra" memory. But maybe we've lost some of that, and we have trouble > with having more physical memory. We could end up in a situation where we > allocate it all very quickly (because we don't actually page it out, we > just allocate backing store for the pages), and we screw something up. > > But stupid bugs there should still leave us trivially able to do the SysRQ > things, so.. > Well, it's a fact that sysrq works just fine before the lockup but does not work at all after a lockup, so... > Is it highmem-related? Some bounce-buffering problem while having to swap? I can try building a kernel without highmem support and see if I can still cause it to lockup. Would be an interresting datapoint. I'll also try reproducing the lockup without any swap active to see if that makes a difference. > What block device driver do you use for the swap device? > It's a swap partition on a IBM Ultra160 10K RPM SCSI disk. The controller is an Adaptec 29160N. Using the SCSI_AIC7XXX driver. > I don't think we use any irq-disable locking in the VM itself, but I could > imagine some nasty situation with the block device layer getting into a > deadlock with interrupts disabled when it runs out of queue entries and > cannot allocate more memory.. > Just let me know what you would like me to try/test to prove/disprove that. -- Jesper Juhl <jesper.juhl@gmail.com> Don't top-post http://www.catb.org/~esr/jargon/html/T/top-post.html Plain text mails only, please http://www.expita.com/nomime.html - To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/ | ||||||||||||
| Last update: 2006-11-22 11:05 [W:0.147 / U:0.020 seconds] ©2003-2008 Jasper Spaans | |||||||||||||