more bogomips :)

more bogomips :)

[Christian Kujau]

I know bogomips are kinda "bogus by definition", but I'll ask anyway, 
out of curiosity:

--- ppc-cpuinfo.2.6.20-rc6      2007-01-27 12:33:00.000000000 +0000
+++ ppc-cpuinfo.2.6.21-rc4-git6 2007-03-24 16:25:26.000000000 +0000
@@ -1,8 +1,8 @@
  processor      : 0
  cpu            : 7447A, altivec supported
-clock          : 1199.999997MHz
+clock          : 1199.999000MHz
  revision       : 0.1 (pvr 8003 0101)
-bogomips       : 36.73
+bogomips       : 73.47
  timebase       : 18432000
  platform       : PowerMac
  machine                : PowerBook6,5

It seems that with a more recent kernel, bogomips have doubled (both 
with CONFIG_HZ=1000, if this matters). How comes?

Thanks,
Christian
-- 
BOFH excuse #108:

The air conditioning water supply pipe ruptured over the machine room

Re: more bogomips :)

[Benjamin Herrenschmidt]

On Sat, 2007-03-24 at 16:48 +0000, Christian Kujau wrote:
> I know bogomips are kinda "bogus by definition", but I'll ask anyway, 
> out of curiosity:
> 
> --- ppc-cpuinfo.2.6.20-rc6      2007-01-27 12:33:00.000000000 +0000
> +++ ppc-cpuinfo.2.6.21-rc4-git6 2007-03-24 16:25:26.000000000 +0000
> @@ -1,8 +1,8 @@
>   processor      : 0
>   cpu            : 7447A, altivec supported
> -clock          : 1199.999997MHz
> +clock          : 1199.999000MHz
>   revision       : 0.1 (pvr 8003 0101)
> -bogomips       : 36.73
> +bogomips       : 73.47
>   timebase       : 18432000
>   platform       : PowerMac
>   machine                : PowerBook6,5
> 
> It seems that with a more recent kernel, bogomips have doubled (both 
> with CONFIG_HZ=1000, if this matters). How comes?

Not sure what the bogomips are about nowadays but they are irrelevant on
most macs as we use the timebase for short delays.

Ben.
 

Re: more bogomips :)

[Johannes Berg]

[-- Attachment #1: Type: text/plain, Size: 845 bytes --]

On Sat, 2007-03-24 at 16:48 +0000, Christian Kujau wrote:

> -bogomips       : 36.73

This value was retrieved when your CPU was set to the lower speed by cpu
freq.

> +bogomips       : 73.47

And that when it was at the higher speed.

> It seems that with a more recent kernel, bogomips have doubled (both 
> with CONFIG_HZ=1000, if this matters). How comes?

They've always switched between those values afaik.

Interesting. This does raise the question of whether it's the right
thing to do, and it looks like a bug to me. The cpufreq core code scales
the loops per jiffy simply by the clock frequency difference which
(since we boot with slow frequency at least on my powerbook) means that
all our delays get to be twice as long during clocked-up operation
because the timebase doesn't change with cpufreq.

johannes

[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 190 bytes --]

Re: more bogomips :)

[Johannes Berg]

[-- Attachment #1: Type: text/plain, Size: 296 bytes --]

On Sun, 2007-03-25 at 10:26 +0200, Johannes Berg wrote:

> Interesting. This does raise the question of whether it's the right
> thing to do, and it looks like a bug to me.
[...]
> because the timebase doesn't change with cpufreq.

Thinko. We don't use lpj for udelay anyway.

johannes

[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 190 bytes --]