Messages in this thread | | | Date | Fri, 8 May 2015 15:27:00 +0200 | From | Ingo Molnar <> | Subject | Re: [PATCH 3/3] context_tracking,x86: remove extraneous irq disable & enable from context tracking on syscall entry |
| |
* Andy Lutomirski <luto@amacapital.net> wrote:
> > on return from idle we'd do: > > > > rcu_qs_ctr += 4-2+1; /* Register QS, clear bit 1, set bit 0 */ > > > > etc. On all boundary transitions we can use a constant ADD on a > > suitable percpu variable. > > Sounds good to me, except that we need to be careful to distinguish > between non-syscall entries from quiescent states > and non-syscall entries from quiescent states.
It might be hard to make that distinction! ;-)
I suspect you wanted to raise the issue of various contexts nesting on each other, such as syscall triggering a page fault, which gets an irq nested, etc. - versus non-nested contexts such as user-space triggering a page fault or user-space getting an irq?
> [...] We could save the old state (as the current exception_enter > code does) or we could allocate enough low bits for the state that > the problem goes away.
So I think, if it all works in practice just as well as it does in email, we might be better off with more state bits: that would tell any remote statistics/sampling code more as well.
It might also be easier to patch in/out, because this kind of state tracking will affect non-RT CPUs as well. (Later on we could do a separate IDT for RT CPUs as well, with a patched version of the entry code.)
> I don't think the TIF_RCU_QS variant is worthwhile -- merging the > counter and state is probably both easier and faster.
Yeah.
Thanks,
Ingo
| |