Re: Ideas for v2.1

Werner Almesberger (
Tue, 11 Jun 1996 20:54:08 +0200 (MET DST)

Alan Cox wrote:
> The TLB one is likely to be even nastier as people go to stuff like DMA
> to/from user space and do things like creating windows for interrupts to
> occur between the set_pte() and the mmu reload or tlb flush.
> Werner - I don't think you handle this yet for the ATM code do you ?

No, at quite a few places, an interrupt could trip over some stale
TLB information. I'm currently trying to figure out where it actually
matters (e.g. it should be okay if the page is just in the process of
getting locked in physical memory), and in particular how to use all
those (comparably) new flush_tlb* functions efficiently. The code also
isn't SMP-aware at all.

I hope to have something reasonable by next week or so.

- Werner

 / Werner Almesberger, DI-LRC,EPFL,CH /