Andrei Dorofeev wrote:
Hi Ethan,
I haven't reviewed everything yet, but I have these
comments/questions so far.
usr/src/uts/i86pc/os/intr.c
intr_thread_prolog():294
How come intrtime gets set to "t->t_intr_start - tsc_read()"
and not "tsc_read() - t->t_intr_start" like everywhere else?
Is it intentional?
Ick. That's the way the code has been. I touched it, but didn't
notice that it was already backwards. That definitely needs to be reversed.
cpu_intr_swtch_enter():717
Why are you using atomic_add_64() here? It seems that
we're updating local CPU's counters everywhere else
without using atomic ops.
This function isn't called with interrupts disabled, unlikely all
the other instances in intr.c. atomic ops aren't needed to protect
intrstat[pil] because no other interrupt can arrive at this pil level,
but a higher level interrupt could arrive and update the same
intracct[mstate].
-- Ethan
_______________________________________________
perf-discuss mailing list
perf-discuss@opensolaris.org
