BUG: 2.6.6-rc3 on SMP/SPARC64 (Sun E3000)? & MD

BUG: 2.6.6-rc3 on SMP/SPARC64 (Sun E3000)? & MD

[Patrick Finnegan]

I've been trying to get 2.6.6-rc3 to work in SMP mode on my E3000 without
much success yet.  It boots fine with a uniprocessor kernel, but trying
to enable SMP gives me this as the last few lines of the kernel messages
(booted with -p early printk option):

----------------------------------------------------------------
CENTRAL: Detected 4 slot Enterprise system. cfreg[aa] cver[fc]
FHC(board 1): Version[1] PartID[fa0] Manuf[3e] (CENTRAL)
FHC(board 5): Version[1] PartID[fa0] Manuf[3e] (JTAG Master)
FHC(board 7): Version[1] PartID[fa0] Manuf[3e]
FHC(board 1): Version[1] PartID[fa0] Manuf[3e]
FHC(board 3): Version[1] PartID[fa0] Manuf[3e]
Built 1 zonelists
Kernel command line: root=/dev/sda1 ro -p
PID hash table entries: 2048 (order 11: 32768 bytes)
Unable to handle kernel NULL pointer dereference
tsk->{mm,active_mm}->context = 0000000000000000
tsk->{mm,active_mm}->pgd = fffff8000040f000
              \|/ ____ \|/
              "@'/ .. \`@"
              /_| \__/ |_\
                 \__U_/
swapper(0): Oops [#1]
TSTATE: 0000009980f01603 TPC: 000000000044b704 TNPC: 000000000044b708 Y: 000000d
TPC: <wake_up_forked_process+0x1e4/0x280>
g0: 00000000006c5c00 g1: 00000000000007f0 g2: 0000000000000000 g3: 0000000000628
g4: 0000000000622000 g5: 0000000000000000 g6: 000000000061e000 g7: 0000000000000
o0: 000000000000007d o1: 0000000000008000 o2: fffff8000005c440 o3: 00000000006a0
o4: 0000000000621d98 o5: 0000000000000008 sp: 00000000006214e1 ret_pc: 00000000c
RPC: <wake_up_forked_process+0x15c/0x280>
l0: 00000000006a11b8 l1: 00000000006c7400 l2: 00000000006c5c00 l3: 0000000000620
l4: 000000000000fe83 l5: 0000000000000003 l6: 0000000000000000 l7: 0000000000620
i0: 0000000000622000 i1: 0000000000000002 i2: 00000000006a11b8 i3: fffff80000030
i4: 0000000000000001 i5: fffff8000003de98 i6: 00000000006215b1 i7: 000000000068c
I7: <sched_init+0x12c/0x1a0>
Instruction DUMP: 9a01e008  82006020  c2762038 <c4586008> c6706008  82102001  c
Kernel panic: Attempted to kill the idle task!
In idle task - not syncing
 <0>Press L1-A to return to the boot prom
---------------------------

I've posted my .config, System.map, (uniproc) dmesg (from klogd), and 
(2.4.26) /proc/cpuinfo files to:
http://x-ray.rcs.purdue.edu/linux-2.6.6-rc3-e3000.config
http://x-ray.rcs.purdue.edu/linux-2.6.6-rc3-e3000.System.map
http://x-ray.rcs.purdue.edu/linux-2.6.6-rc3-e3000.dmesg
http://x-ray.rcs.purdue.edu/linux-2.4.26-e3000.cpuinfo

I have not yet tested any other versions on it.

Also, I've noticed the md driver seems to have problems with data corruption 
(it overwrites other partitions on the same disk, or the disklabel even).  I've
noticed this problem under 2.6.5 on my Sun Ultra 60, as well.  Is it an
UltrSparc issue, or is it more general?

Pat
-- 
Purdue University ITAP/RCS        --- http://www.itap.purdue.edu/rcs/
The Computer Refuge               --- http://computer-refuge.org

Re: BUG: 2.6.6-rc3 on SMP/SPARC64 (Sun E3000) & MD / ksymoops output

[Patrick Finnegan]

Output from ksymoops is attached below.

On Friday 07 May 2004 10:49, you wrote:
> I've been trying to get 2.6.6-rc3 to work in SMP mode on my E3000
> without much success yet.  It boots fine with a uniprocessor kernel,
> but trying to enable SMP gives me this as the last few lines of the
> kernel messages (booted with -p early printk option):
>

ksymoops 2.4.9 on sparc64 2.4.26-sparc64-smp.  Options used
     -v /usr/src/linux-2.6.6-rc3/vmlinux (specified)
     -K (specified)
     -L (specified)
     -o /lib/modules/2.4.26-sparc64-smp/ (default)
     -m /usr/src/linux-2.6.6-rc3/System.map (specified)

No modules in ksyms, skipping objects
Unable to handle kernel NULL pointer dereference
tsk->{mm,active_mm}->context = 0000000000000000
tsk->{mm,active_mm}->pgd = fffff8000040f000
              \|/ ____ \|/
              "@'/ .. \`@"
              /_| \__/ |_\
                 \__U_/
swapper(0): Oops [#1]
TSTATE: 0000009980f01603 TPC: 000000000044b704 TNPC: 000000000044b708 Y: 
000000d
Using defaults from ksymoops -t elf32-sparc -a sparc
g0: 00000000006c5c00 g1: 00000000000007f0 g2: 0000000000000000 g3: 
0000000000628
g4: 0000000000622000 g5: 0000000000000000 g6: 000000000061e000 g7: 
0000000000000
o0: 000000000000007d o1: 0000000000008000 o2: fffff8000005c440 o3: 
00000000006a0
o4: 0000000000621d98 o5: 0000000000000008 sp: 00000000006214e1 ret_pc: 
00000000c
l0: 00000000006a11b8 l1: 00000000006c7400 l2: 00000000006c5c00 l3: 
0000000000620
l4: 000000000000fe83 l5: 0000000000000003 l6: 0000000000000000 l7: 
0000000000620
i0: 0000000000622000 i1: 0000000000000002 i2: 00000000006a11b8 i3: 
fffff80000030
i4: 0000000000000001 i5: fffff8000003de98 i6: 00000000006215b1 i7: 
000000000068c


>>PC;  0044b704 <wake_up_forked_process+1e4/280>   <=====

>>g0; 006c5c00 <__log_buf+7d50/8000>
>>g4; 00622000 <init_task+0/580>
>>g6; 0061e000 <init_thread_union+0/4000>
>>o4; 00621d98 <init_thread_union+3d98/4000>
>>sp; 006214e1 <init_thread_union+34e1/4000>
>>l0; 006a11b8 <per_cpu__runqueues+0/1260>
>>l1; 006c7400 <uidhash_table+ef0/1000>
>>l2; 006c5c00 <__log_buf+7d50/8000>
>>i0; 00622000 <init_task+0/580>
>>i2; 006a11b8 <per_cpu__runqueues+0/1260>
>>i6; 006215b1 <init_thread_union+35b1/4000>

Instruction DUMP: 9a01e008  82006020  c2762038 <c4586008> c6706008  
82102001  c
Error (Oops_code_values): invalid value 0xc in Code line, must be 2, 4, 
8 or 16 digits, value ignored


Code;  0044b6f8 <wake_up_forked_process+1d8/280>
00000000 <_PC>:
Code;  0044b6f8 <wake_up_forked_process+1d8/280>
   0:   9a 01 e0 08       add  %g7, 8, %o5
Code;  0044b6fc <wake_up_forked_process+1dc/280>
   4:   82 00 60 20       add  %g1, 0x20, %g1
Code;  0044b700 <wake_up_forked_process+1e0/280>
   8:   c2 76 20 38       unknown
Code;  0044b704 <wake_up_forked_process+1e4/280>
   c:   c4 58 60 08       unknown
Code;  0044b708 <wake_up_forked_process+1e8/280>
  10:   c6 70 60 08       unknown
Code;  0044b70c <wake_up_forked_process+1ec/280>
  14:   82 10 20 01       mov  1, %g1

Kernel panic: Attempted to kill the idle task!

> ----------------------------------------------------------------
> CENTRAL: Detected 4 slot Enterprise system. cfreg[aa] cver[fc]
> FHC(board 1): Version[1] PartID[fa0] Manuf[3e] (CENTRAL)
> FHC(board 5): Version[1] PartID[fa0] Manuf[3e] (JTAG Master)
> FHC(board 7): Version[1] PartID[fa0] Manuf[3e]
> FHC(board 1): Version[1] PartID[fa0] Manuf[3e]
> FHC(board 3): Version[1] PartID[fa0] Manuf[3e]
> Built 1 zonelists
> Kernel command line: root=/dev/sda1 ro -p
> PID hash table entries: 2048 (order 11: 32768 bytes)
> Unable to handle kernel NULL pointer dereference
> tsk->{mm,active_mm}->context = 0000000000000000
> tsk->{mm,active_mm}->pgd = fffff8000040f000
>               \|/ ____ \|/
>               "@'/ .. \`@"
>               /_| \__/ |_\
>                  \__U_/
> swapper(0): Oops [#1]
> TSTATE: 0000009980f01603 TPC: 000000000044b704 TNPC: 000000000044b708
> Y: 000000d TPC: <wake_up_forked_process+0x1e4/0x280>
> g0: 00000000006c5c00 g1: 00000000000007f0 g2: 0000000000000000 g3:
> 0000000000628 g4: 0000000000622000 g5: 0000000000000000 g6:
> 000000000061e000 g7: 0000000000000 o0: 000000000000007d o1:
> 0000000000008000 o2: fffff8000005c440 o3: 00000000006a0 o4:
> 0000000000621d98 o5: 0000000000000008 sp: 00000000006214e1 ret_pc:
> 00000000c RPC: <wake_up_forked_process+0x15c/0x280>
> l0: 00000000006a11b8 l1: 00000000006c7400 l2: 00000000006c5c00 l3:
> 0000000000620 l4: 000000000000fe83 l5: 0000000000000003 l6:
> 0000000000000000 l7: 0000000000620 i0: 0000000000622000 i1:
> 0000000000000002 i2: 00000000006a11b8 i3: fffff80000030 i4:
> 0000000000000001 i5: fffff8000003de98 i6: 00000000006215b1 i7:
> 000000000068c I7: <sched_init+0x12c/0x1a0>
> Instruction DUMP: 9a01e008  82006020  c2762038 <c4586008> c6706008 
> 82102001  c Kernel panic: Attempted to kill the idle task!
> In idle task - not syncing
>  <0>Press L1-A to return to the boot prom
> ---------------------------
>
> I've posted my .config, System.map, (uniproc) dmesg (from klogd), and
> (2.4.26) /proc/cpuinfo files to:
> http://x-ray.rcs.purdue.edu/linux-2.6.6-rc3-e3000.config
> http://x-ray.rcs.purdue.edu/linux-2.6.6-rc3-e3000.System.map
> http://x-ray.rcs.purdue.edu/linux-2.6.6-rc3-e3000.dmesg
> http://x-ray.rcs.purdue.edu/linux-2.4.26-e3000.cpuinfo
>
> I have not yet tested any other versions on it.
>
> Also, I've noticed the md driver seems to have problems with data
> corruption (it overwrites other partitions on the same disk, or the
> disklabel even).  I've noticed this problem under 2.6.5 on my Sun
> Ultra 60, as well.  Is it an UltrSparc issue, or is it more general?
>
> Pat

-- 
Purdue University ITAP/RCS        ---  http://www.itap.purdue.edu/rcs/
The Computer Refuge               ---  http://computer-refuge.org

[PATCH] Re: BUG: 2.6.6-rc3 on SMP/SPARC64 (Sun E3000)?

[Patrick Finnegan]

On Friday 07 May 2004 10:49, you wrote:
> I've been trying to get 2.6.6-rc3 to work in SMP mode on my E3000
> without much success yet.  It boots fine with a uniprocessor kernel,
> but trying to enable SMP gives me this as the last few lines of the
> kernel messages (booted with -p early printk option):

It appears that the problem is that on Sparc64, smp_processor_id() gives
the hardware ID, not a logical number (ie 0..n for the first n+1 
processors).  I had NR_CPUS set to 8, and the first (boot) CPU was 
numbered by the hardware to be CPU 10.

The patch catches if the boot CPU is greater than NR_CPUS, since that is
possible with Sparc64.  I see that the condition is explicitly checked 
for in smp_tick_init(), but that must not get called soon enough to 
catch the problem (which manifested in sched_init()s call to
wake_up_forked_process() ), so I moved it to smp_prepare_boot_cpu().

Comments?

Pat
-- 
Purdue University ITAP/RCS        ---  http://www.itap.purdue.edu/rcs/
The Computer Refuge               ---  http://computer-refuge.org

--- linux-2.6.6.orig/arch/sparc64/kernel/smp.c  2004-05-09 21:31:55.000000000 -0500
+++ linux-2.6.6/arch/sparc64/kernel/smp.c       2004-05-11 19:46:15.692007000 -0500
@@ -1108,11 +1108,6 @@
        boot_cpu_id = hard_smp_processor_id();
        current_tick_offset = timer_tick_offset;

-       if (boot_cpu_id >= NR_CPUS) {
-               prom_printf("Serious problem, boot cpu id >= NR_CPUS\n");
-               prom_halt();
-       }
-
        cpu_set(boot_cpu_id, cpu_online_map);
        prof_counter(boot_cpu_id) = prof_multiplier(boot_cpu_id) = 1;
 }
@@ -1254,6 +1249,11 @@

 void __devinit smp_prepare_boot_cpu(void)
 {
+       if (hard_smp_processor_id() >= NR_CPUS) {
+               prom_printf("Serious problem, boot cpu id >= NR_CPUS\n");
+               prom_halt();
+       }
+
        current_thread_info()->cpu = hard_smp_processor_id();
        cpu_set(smp_processor_id(), cpu_online_map);
        cpu_set(smp_processor_id(), phys_cpu_present_map);

Re: [PATCH] Re: BUG: 2.6.6-rc3 on SMP/SPARC64 (Sun E3000)?

[David S. Miller]

Good patch, but I had to apply it by hand since your email client
changed all the tabs into spaces.  Please use an attachment or
similar for the patch in the future.

Thanks.

Re: [PATCH] Re: BUG: 2.6.6-rc3 on SMP/SPARC64 (Sun E3000)?

[Paul Jackson]

> since your email client changed all the tabs into spaces.

Might not have been his email client.  Another easy way to botch the
tabs is to cut+paste the patch into the email's message composition
window, for those using gui-style mailers, such as Kmail (Patrick) or
Sylpheed (myself).  One can still inline patches with such tools - one
just has to be careful to read them into (File:Insert, perhaps) the
outgoing message from a file, rather than via cut+paste.

-- 
                          I won't rest till it's the best ...
                          Programmer, Linux Scalability
                          Paul Jackson <pj@sgi.com> 1.650.933.1373

Re: [PATCH] Re: BUG: 2.6.6-rc3 on SMP/SPARC64 (Sun E3000)?

[Pete Zaitcev]

On Thu, 20 May 2004 02:30:04 -0700
Paul Jackson <pj@sgi.com> wrote:

> Might not have been his email client.  Another easy way to botch the
> tabs is to cut+paste the patch into the email's message composition
> window, for those using gui-style mailers, such as Kmail (Patrick) or
> Sylpheed (myself). [...]

sylpheed does no such thing. It's your terminal emulatior, most likely.
The truth is very easy to establish by running xcb and xclipboard.

-- Pete

Re: [PATCH] Re: BUG: 2.6.6-rc3 on SMP/SPARC64 (Sun E3000)?

[Paul Jackson]

> sylpheed does no such thing.

Yes.   Exactly as I was saying - the cut+paste, not the the
email client, is dropping tabs in this scenario.

We are in vigourous agreement.

-- 
                          I won't rest till it's the best ...
                          Programmer, Linux Scalability
                          Paul Jackson <pj@sgi.com> 1.650.933.1373