Re: mdelay() implementation patch

Alexander Kjeldaas (astor@guardian.no)
Fri, 22 May 1998 13:56:51 +0200

On Fri, May 22, 1998 at 05:27:44AM +1000, Paul Gortmaker wrote:
> >
> > > My question is, do we:
> > >
> > > (a) ignore it
> > > (b) when >1000, replace each occurence with:
> > > {unsigned int msec = XXXX; while (--msec) udelay(1000);}
> > > (c) indroduce a "mdelay()" that does (b)
> > > (d) order a pizza
> >
> > My answer currently seems to be "all of the above".
> >
> > (c) would certainly make sense.
> >
> > Linus
>
> Here is a diff for (c) as (d) doesn't uuencode well. The semi
> interesting bits are in <linux/delay.h> as the rest was mostly a job
> for sed. With this patch, you can easily tune (on a per-architecture
> basis if you want) the max. number of milliseconds udelay will be
> used with (currently set to 5). Patch is against 2.1.103.
>

What is the semantic difference between mdelay and udelay? I mean, why
don't you just change the implementation of udelay to handle longer
delays using __builtin_constant_p et al? Having both mdelay and udelay
is confusing.

if you redefine the current udelay to __native_udelay for example, then
instead of doing

+#define mdelay(n) (\
+ (__builtin_constant_p(n) && (n)<=MAX_UDELAY_MS) ? udelay((n)*1000) : \
+ ({unsigned long msec=(n); while (msec--) udelay(1000);}))

you could do roughly

#define udelay(n) do { \
if (__builtin_constant_p(n) && (n)<=MAX_UDELAY_MS*1000) \
__native_udelay(n); \
else { \
unsigned long msec=(n)/1000; \
while (msec--) \
__native_udelay(1000); \
__native_udelay((n)%1000); \
} \
} while (0)

and we wouldn't have two separate functions.

astor

-- 
 Alexander Kjeldaas, Guardian Networks AS, Trondheim, Norway
 http://www.guardian.no/


-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu