Messages in this thread | | | Date | Tue, 28 Apr 2015 08:53:18 -0400 | From | Rik van Riel <> | Subject | Re: [PATCH v2] context_tracking: remove local_irq_save from __acct_update_integrals |
| |
On 04/27/2015 07:18 AM, Heiko Carstens wrote: > On Sat, Apr 25, 2015 at 08:50:49AM -0400, Rik van Riel wrote: >> On 04/25/2015 05:43 AM, Heiko Carstens wrote: >>> ...the READ_ONCE() doesn't give you any guarantees about reading >>> tsk->acct_timexpd in an atomic way. >>> Well, actually you don't need atomic semantics, but only to make sure that >>> the read access happens with a single instruction, since you want to protect >>> against interrupts. >>> But still: if the size of acct_timexpd is 64 bit READ_ONCE() may still result >>> in two instructions on 32 bit architectures. >>> (or isn't there currently no 32 bit architecture with 64 bit cputime_t left?) >> >> Even if there is (maybe some ARM system?), can we even guarantee >> that a single instruction to read 64 bits exists on such a system? > > I wouldn't bet on it. I can only talk for s390 and there is an instruction > available which would do that. But since s390 is now a 64 bit only architecture > it doesn't matter anyway. > For other architectures I'd say: no, you can't rely on that.
So what can I do to move forward with this patch?
It speeds up syscall entry / exit by 7% when nohz_full is enabled on a CPU...
Should I have the irq block compiled in only when sizeof(cputime_t) > sizeof(long) ?
-- All rights reversed.
| |