On Thu, 9 Aug 2001, Andrea Arcangeli wrote:
> > I'm trying to track down a problem which seems to be a race condition
> > vmalloc page table delayed PTE copying", or "you must never
> > call free_kiovec in an interrupt context" etc..)
>
> The latter is certainly the case. The former could be the case, however
Had some nasty bugs in that case before..
> if you know you're running any free_kiovec from irqs remove it and try
> again first.
Yes, I found it. The irq which signals end-of-I/O also frees the kiovec,
which is bad. But it seemed that what would happen then would be deadlocks
(due to the spinlocks) and not races - but I'll try to make it free in a
task or something instead (there is no kernel context for the "calling"
process because the driver is asynchronous so the context which did the
alloc_kiovec etc. has exited when the irq comes later) and see if it works
better.
> BTW, you should also avoid all the kiobuf allocation in all the fast
> paths, try to pre-allocate it in a slow path to deliver higher
> performance. (shortly we'll split the bh/blocks array out of the kiobuf
Ok, that's probably not a problem since the allocation is done in a normal
system-call context.
/Bjorn
-
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/
This archive was generated by hypermail 2b29 : Wed Aug 15 2001 - 21:00:26 EST