Skip to contentSkip to conversation

too low MAX_MP_BUSSES

DISCUSSION Vitezslav Samel 3 participants 4 messages Jul 31, 2006 – Aug 1, 2006

lore
Share & tools
lore
#0 lore raw
	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

#1 lore raw
Quoted 40 lines · On Mon, 31 Jul 2006 13:55:45 +0200 Vitezslav Samel <[email protected]> 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

(quoted diff) +6 -0
+       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
Quoted 7 lines

---
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.

#2 lore raw
Quoted 7 lines

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

#3 lore raw
Quoted 9 lines · On 1 Aug 2006 10:46:37 +0200 Andi Kleen <[email protected]> wrote:
Quoted 7 lines

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.