> I assume it's one or both of these loops in arch/x86/xen/time.c 
quoted text
> do_stolen_accounting() that are being optimized into a divide which 
> generates a libgcc call:
>
> 	while (stolen >= NS_PER_TICK) {
> 		ticks++;
> 		stolen -= NS_PER_TICK;
> 	}
>
> or
>
> 	while (blocked >= NS_PER_TICK) {
> 		ticks++;
> 		blocked -= NS_PER_TICK;
> 	}

That looks plausible.

quoted text
> I seem to recall in one previous case we added some dummy assembly 
> code to stop gcc from doing this.

I think you refer to 38332cb9, "time: prevent the loop in
timespec_add_ns() from being optimised away".


         while(unlikely(ns >= NSEC_PER_SEC)) {
+               /* The following asm() prevents the compiler from
+                * optimising this loop into a modulo operation.  */
+               asm("" : "+r"(ns));
+
                 ns -= NSEC_PER_SEC;
                 a->tv_sec++;
         }


quoted text
> Not sure if that is a sustainable fix, though..

It should be.  The asm() arg tells GCC that the asm() could modify
"ns" in some way, so GCC cannot optimise away the loop, since it
doesn't have the required info about the induction variable to do
that.


Segher

--
unsubscribe notice
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/