Re: Lockup 2.1.6* => kmalloc/slab

Alan Cox (alan@lxorguk.ukuu.org.uk)
Wed, 12 Nov 1997 15:19:27 +0000 (GMT)

> Of course, the lock-up may not be in the networking code. The allocations
> the code performs triggers a page-reaping (race) bug somewhere else. This
> does sounds v. plausable - I'll walk some code paths this evening. Your
> not SMP are you?

With 2.1.62+CVS SMP I got what looked very like lockups until I caught it
with the right tools handy. The machine was trying to do a pile of NFS
writes to swap out pages of a read only binary, and while having some
memory had nothing over 4K. That froze X11 trying to get the NFS pages
it wanted and it sure looked liek a crash. Running gdb then quitting it
sorted the total memory fragmentation.

Unless I run with the page colouring disabled this is fairly repeatable