[lkml]   [2003]   [May]   [7]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
Messages in this thread
SubjectRe: [PATCH]: DELAYTIMER_MAX is defined
I would like to give some thought to setting this value rather low, 
say 1000 or so. The issue is that the update to the expire time is
done by adding the timer increment repeatedly until a time that has
not yet passed is found. THIS TAKES TIME and it is done under an irq
spin_lock! I do think we should limit the time in this loop and,
also, possibly finding a way to do it with out holding the spin_lock,
so that other tasks could at least preempt the caller. (As it is, it
already done in the callers task and not in an interrupt context.)

This, of course, also says that we should not only limit the value of
overrun, but also do something different when it happens.


Eric Piel wrote:
> Ulrich Drepper wrote:
>>This is not correct. The constant does not have to be defined. Like
>>all the various *_MAX constants they only have to be defined if there is
>>a fixed limit the implementation has. If there is none or it can only
>>be defined dynamically at runtime the the macro must not be defined.
>>Instead sysconf() can provide the value. But not even this is
>>necessary. sysconf() can return -1.
> Sorry, I wasn't aware about this POSIX rule, thank you for pointing out
> this. Knowing that it would obviously be better providing support of the
> constant _SC_DELAYTIMER_MAX for sysconf() and return a value.
> This would imply there is now some job for glibc. However it's still
> unclear for me how the glibc would know about wich value it should
> return while this question would be so easy to answer for the kernel!
> Also AFAIK, right now, only the NPTL supports the new syscalls for the
> timers (and none supports the clocks syscalls). Does it mean a special
> __sysconf() is necessary in the NPTL? Well, probably that's why you
> suggested -1 as a return value I guess :-)
>>Anyway, in this specific case the implementation should be protected
>>against the ever so improbable overflow of the counter, yes. If you
>>want a fixed value, fine. If you want to use ULONG_MAX (or whatever),
>>good too. Whether we advertise this limit is another thing.
>>Advertising it in the macro means it never can be changed.
> You are right also here, of course with this point of view it's
> better not putting any constant.
> So the patch could become something like that: ?
> diff -ur linux-2.5.67-ia64-hrtcore/include/linux/posix-timers.h linux-2.5.67-ia64-hrtimers/include/linux/posix-timers.h
> --- linux-2.5.67-ia64-hrtcore/include/linux/posix-timers.h 2003-04-22 11:10:44.000000000 +0200
> +++ linux-2.5.67-ia64-hrtimers/include/linux/posix-timers.h 2003-05-06 16:07:56.000000000 +0200
> @@ -25,6 +59,7 @@
> #define posix_bump_timer(timr) do { \
> (timr)->it_timer.expires += (timr)->it_incr; \
> - (timr)->it_overrun++; \
> + if ((timr)->it_overrun < INT_MAX)\
> + (timr)->it_overrun++; \
> }while (0)
> #endif
> Eric

George Anzinger
Preemption patch:

To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to
More majordomo info at
Please read the FAQ at

 \ /
  Last update: 2005-03-22 13:35    [W:0.096 / U:0.196 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site