Hi
TJ wrote:
Thanks for the responses. I think ddi_periodic_add is what I was looking for.
I have been using trapstat, but I also need information on the virtual
addresses on which the TLB is missing, so trapstat falls a bit short.
In that case have a look at the TRAPTRACE code that is active in DEBUG
kernels - it'll give you some inspiration. Depending on your needs
it could even be enough - you generally need to enlarge the per-cpu
buffer it uses to get any meaningful time period covered, and it
can help to eliminate traptrace records on window traps as they
take 90% of the records (both will need recompile). You
can dump with ::ttrace [-x] - for tlb and tsb misses one
of the F1-4 parameters is the VA that missed. There is also
a trap freeze mechanism that can be triggered via a userland
fasttrap instruction - a handy way of freezing tracing once
you have reproduced the issue you are investigating.
Not as useful for long-term performance profiling.
Cheers
Gavin
_______________________________________________
opensolaris-code mailing list
opensolaris-code@opensolaris.org
http://mail.opensolaris.org/mailman/listinfo/opensolaris-code
