Re: [PATCH] hangcheck-timer is broken on x86

[Yury Polyanskiy <ypolyans@princeton.edu>]

[-- Attachment #1: Type: text/plain, Size: 1544 bytes --]

On Fri, 26 Mar 2010 17:57:11 -0700
Joel Becker <Joel.Becker@oracle.com> wrote:

> On Fri, Mar 26, 2010 at 06:00:25PM -0400, Yury Polyanskiy wrote:
> > On Fri, 26 Mar 2010 14:46:49 -0700
> > Joel Becker <Joel.Becker@oracle.com> wrote:
> > 
> > > On Tue, Mar 23, 2010 at 11:36:11PM -0400, Yury Polyanskiy wrote:
> > 
> > > 1) Does getrawmonotonic() satisfy hangcheck-timer?  What I mean is, will
> > > it always return the wallclock nanoseconds even in the face of CPU speed
> > > changes, suspend, udelay, or any other suspension of kernel operation?
> > > Yes, I know this is a tougher standard than rdtsc(), but this is what
> > > hangcheck-timer wants.  rdtsc() at least satisfied udelay and PCI hangs.
> > 
> > Yes, as far as I can tell. Note that rdtsc is hosed on suspend-resume.
> 
> 	Yeah, I know.  rdtsc hangcheck-timer really required no suspend
> or cpufreq.  Since it is only really used by servers, this wasn't a
> terrible restriction.  Then virtualization came along...

Joel, just realized there is a slight mistake in what I said before.
getrawmonotonic() is a refined jiffies (and actually resolves to
get_cycles() on my system in the end). Thus it doesn't count while in
suspend. However, jiffies-based timers (aka timer-wheel) are also
stopped while in suspend. So getrawmonotonic() is the right call to
check the precision of the jiffies-based timer (i.e. you dont need to
make a correction by calling monotonic_to_bootbased()). 

So my comment about rdtsc and suspend-resume is wrong.

Yury

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 198 bytes --]