Hi Pete,

> At OLS I talked to Anton Blanchard (he works on ppc6 with Paul
> Mackerras), and he indicated an interest in making a kernel
> running with an 8K software page size on top of 4K hardware
> page size. I understand that the order-2 patch from 2.4.17
> level update reduced the urgency of this for s390x; please
> tell me if this is correct.

Reduce the urgency yes, but it still isn't a complete fix.

> Do we want something like this?

The problem we were seeing with this approach is that it is
user-space visible: mmap on 4K (but not 8K) boundary would now
fail.  This is in particular fatal for applications running
in 31-bit emu mode (e.g. ELF sections are aligned to 4K) ...

> If yes, what workload triggers allocation failures (for reproducible
> testing)?

I'm not aware of any small test case.  Running very long batch
jobs on a SAP R/3 app server under heavy load used to trigger
the problem (before the order-2 patch).

Bye,
Ulrich

--
  Dr. Ulrich Weigand
  [EMAIL PROTECTED]

Reply via email to