too low MAX_MP_BUSSES

too low MAX_MP_BUSSES

[Vitezslav Samel]

	Hi!

  I tried upgrading our server (i386 arch) from 2.6.16 to 2.6.17 but there
were some odd messages in dmesg:

	MP table busid value (32) for bustype ISA is too large, max. supported is 31

and (repeated 315 times):

	unknown bus type 32

I found out that 50% of the processor time was spent in softirq and the timers
ran too fast. I didn't look for what else was wrong.

  Tracked down to this change in 2.6.17-rc2:

diff -urN linux-2.6.17-rc1/arch/i386/kernel/mpparse.c linux-2.6.17-rc2/arch/i386/kernel/mpparse.c
+       if (m->mpc_busid >= MAX_MP_BUSSES) {
+               printk(KERN_WARNING "MP table busid value (%d) for bustype %s "
+                       " is too large, max. supported is %d\n",
+                       m->mpc_busid, str, MAX_MP_BUSSES - 1);
+               return;
+       }

 Uping the MAX_MP_BUSSES value in include/asm-i386/mach-default/mach_mpspec.h
to 64 makes the machine work O.K.
The system is HP DL380 g4 with 1 Xeon CPU, kernel compiled non-SMP.

  Here is excerpt from mptable output:
---
Bus:            Bus ID  Type
                 0       PCI
                 1       PCI
                 2       PCI
                 3       PCI
                 4       PCI
                 5       PCI
                 6       PCI
                10       PCI
                32       ISA
---
The last item is the offending one.

Please, can you consider up the default value of MAX_MP_BUSSES?

P.S.: also tested 2.6.18-rc3, the same - bad - result

	Cheers,
		Vita

Re: too low MAX_MP_BUSSES

[Andrew Morton]

On Mon, 31 Jul 2006 13:55:45 +0200
Vitezslav Samel <samel@mail.cz> wrote:

> 	Hi!
> 
>   I tried upgrading our server (i386 arch) from 2.6.16 to 2.6.17 but there
> were some odd messages in dmesg:
> 
> 	MP table busid value (32) for bustype ISA is too large, max. supported is 31
> 
> and (repeated 315 times):
> 
> 	unknown bus type 32
> 
> I found out that 50% of the processor time was spent in softirq and the timers
> ran too fast. I didn't look for what else was wrong.
> 
>   Tracked down to this change in 2.6.17-rc2:
> 
> diff -urN linux-2.6.17-rc1/arch/i386/kernel/mpparse.c linux-2.6.17-rc2/arch/i386/kernel/mpparse.c
> +       if (m->mpc_busid >= MAX_MP_BUSSES) {
> +               printk(KERN_WARNING "MP table busid value (%d) for bustype %s "
> +                       " is too large, max. supported is %d\n",
> +                       m->mpc_busid, str, MAX_MP_BUSSES - 1);
> +               return;
> +       }
> 
>  Uping the MAX_MP_BUSSES value in include/asm-i386/mach-default/mach_mpspec.h
> to 64 makes the machine work O.K.
> The system is HP DL380 g4 with 1 Xeon CPU, kernel compiled non-SMP.
> 
>   Here is excerpt from mptable output:
> ---
> Bus:            Bus ID  Type
>                  0       PCI
>                  1       PCI
>                  2       PCI
>                  3       PCI
>                  4       PCI
>                  5       PCI
>                  6       PCI
>                 10       PCI
>                 32       ISA
> ---
> The last item is the offending one.
> 
> Please, can you consider up the default value of MAX_MP_BUSSES?
> 
> P.S.: also tested 2.6.18-rc3, the same - bad - result
> 

mach-default uses 32 and mach-generic uses 260, so I doubt if there's a big
downside to increasing mach-default.  I expect distros ship with
mach-generic, so you're a rare case.

<tries to remember who works on this and fails>

Andi?  Can you see any problems with increasing the mach-default setting?

Thanks.

Re: too low MAX_MP_BUSSES

[Andi Kleen]

> mach-default uses 32 and mach-generic uses 260, so I doubt if there's a big
> downside to increasing mach-default.  I expect distros ship with
> mach-generic, so you're a rare case.
> 
> <tries to remember who works on this and fails>
> 
> Andi?  Can you see any problems with increasing the mach-default setting?

No, except for wasting a bit of memory (maybe make it dependent on CONFIG_TINY?)

-Andi

Re: too low MAX_MP_BUSSES

[Andrew Morton]

On 1 Aug 2006 10:46:37 +0200
Andi Kleen <ak@muc.de> wrote:

> > mach-default uses 32 and mach-generic uses 260, so I doubt if there's a big
> > downside to increasing mach-default.  I expect distros ship with
> > mach-generic, so you're a rare case.
> > 
> > <tries to remember who works on this and fails>
> > 
> > Andi?  Can you see any problems with increasing the mach-default setting?
> 
> No, except for wasting a bit of memory (maybe make it dependent on CONFIG_TINY?)

Sounds sane, thanks.  I'll do a patch.