Re: BUG? i386 TLB Flushing of Global Pages

From: giampapa@us.ibm.com
Date: Mon Jan 17 2000 - 14:06:38 EST


Hi Manfred,
> I think we should map _all_ kernel memory with PAGE_GLOBAL: most
> distributions heavily use modules, and they would win from mapping
> everything with _PAGE_GLOBAL.

  One thing that is not specified by Intel, as far as I can tell, is
whether or not global and non-global TLB entries compete fairly/equally
for normal TLB replacement. Although my assumption is that they compete
fairly, given how little memory the TLB's actually map, I have always
taken the approach of using _PAGE_GLOBAL sparingly. Clearly interrupt
and trap handlers, the scheduler, etc. should be global, but infrequently
executed kernel code need not be global.
  It would be worthwhile to benchmark all global versus judicious global
to see if any gain could be had. If so, perhaps another kernel segment
could be created (i.e., "__global", like "__init" now). Just a thought...
Thanks,
Mark

Mark Giampapa
Parallel Systems
IBM TJ Watson Research
1-914-945-3133
giampapa@us.ibm.com

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/



This archive was generated by hypermail 2b29 : Sun Jan 23 2000 - 21:00:16 EST