Skip to contentSkip to conversation

lockdep support caused an MCA

DISCUSSION Alex Chiang 3 participants 5 messages Jan 11, 2010 – Jan 14, 2010

lore
Share & tools
lore
#0 lore raw

This patch 2f02b4a12b24c9f077c6d739ac23c9b90840ccca:

Author: Luming Yu <[email protected]> 2009-12-04 09:14:30

[IA64] lockdep support

Basic functionality for lockdep on ia64.

Known issues:
0. dynamic allocate per CPU area in lockdep.
1. lock held in leave kernel.
2. enabling CONFIG_DEBUG_LOCKDEP triggers the lockdep warning.
3. Need to implement save_stack functions
4. Replace some cpp macros with inline functions
5. Add nmi_enter/nmi_exit

causes a very early MCA on an HP rx8640.

Unfortunately, even with early_printk turned on, the machine encounters the MCA and reboots before any output occurs, so I don't have any further information.

I haven't done any other triage on this as the patch has already been dropped from Tony's tree, but I'd be happy to test the next go-around.

Thanks,
/ac

#1 lore raw

Let me check if I can get access to an HP rx8640 to re-spin the patch.

-----Original Message-----
From: Alex Chiang [mailto:[email protected]]
Sent: Tuesday, January 12, 2010 3:58 AM
To: Yu, Luming; Luck, Tony
Cc: [email protected]
Subject: lockdep support caused an MCA

This patch 2f02b4a12b24c9f077c6d739ac23c9b90840ccca:

Author: Luming Yu <[email protected]> 2009-12-04 09:14:30

[IA64] lockdep support

Basic functionality for lockdep on ia64.

Known issues:
0. dynamic allocate per CPU area in lockdep.
1. lock held in leave kernel.
2. enabling CONFIG_DEBUG_LOCKDEP triggers the lockdep warning.
3. Need to implement save_stack functions
4. Replace some cpp macros with inline functions
5. Add nmi_enter/nmi_exit

causes a very early MCA on an HP rx8640.

Unfortunately, even with early_printk turned on, the machine encounters the MCA and reboots before any output occurs, so I don't have any further information.

I haven't done any other triage on this as the patch has already been dropped from Tony's tree, but I'd be happy to test the next go-around.

Thanks,
/ac

#2 lore raw
In reply to Alex Chiang
Quoted 32 lines · Luming

Let me check if I can get access to an HP rx8640 to re-spin the patch.

-----Original Message-----
From: Alex Chiang [mailto:[email protected]]
Sent: Tuesday, January 12, 2010 3:58 AM
To: Yu, Luming; Luck, Tony
Cc: [email protected]
Subject: lockdep support caused an MCA

This patch 2f02b4a12b24c9f077c6d739ac23c9b90840ccca:

Author: Luming Yu <[email protected]>  2009-12-04 09:14:30

   [IA64] lockdep support

   Basic functionality for lockdep on ia64.

   Known issues:
   0. dynamic allocate per CPU area in lockdep.
   1. lock held in leave kernel.
   2. enabling CONFIG_DEBUG_LOCKDEP triggers the lockdep warning.    3. Need to implement save_stack functions
   4. Replace some cpp macros with inline functions
   5. Add nmi_enter/nmi_exit

   Signed-off-by: Bob Picco <[email protected]>    Signed-off-by: Yu Luming <[email protected]>    Signed-off-by: Tony Luck <[email protected]>

causes a very early MCA on an HP rx8640.

Quoted 7 lines

Unfortunately, even with early_printk turned on, the machine encounters the MCA and reboots before any output occurs, so I don't have any further information.

I haven't done any other triage on this as the patch has already been dropped from Tony's tree, but I'd be happy to test the next go-around.

I have managed to get the patch run on a HP rx8640 system with latest next tree + ia64-lockdep patch. The following is the part of boot log with lockdep footprints..The interesting thing is lockdep code seems to have detected possible recursive locking with the kernel on the platform.

Kernel command line:
BOOT_IMAGE=scsi0:EFI\redhat\vmlinuz-2.6.33-rc3-next-20100113 root=/dev/VolGroup00/LogVol00 ro
PID hash table entries: 4096 (order: 1, 32768 bytes)
Memory: 33235200k/33370096k available (8346k code, 171728k reserved, 10446k data, 2400k init)
Hierarchical RCU implementation.
RCU-based detection of stalled CPUs is enabled. NR_IRQS:1024

======================[ INFO: possible recursive locking detected ] 2.6.33-rc3-next-20100113 #1
---------------------------------------------
swapper/0 is trying to acquire lock:
(&(&parent->list_lock)->rlock){......}, at: [<a0000001001e9450>]
kmem_cache_free+0x2f0/0x560

but task is already holding lock:
(&(&parent->list_lock)->rlock){......}, at: [<a0000001001ea490>]
kfree+0x470/0x620

other info that might help us debug this:
2 locks held by swapper/0:
#0: (cache_chain_mutex){+.+...}, at: [<a000000100acf350>]

kmem_cache_init_late+0x30/0x240
 #1:  (&(&parent->list_lock)->rlock){......}, at: [<a0000001001ea490>]
kfree+0x470/0x620

stack backtrace:

Call Trace:
 [<a000000100015fb0>] show_stack+0x50/0xa0
                                sp 00000100f07bc0 bsp 00000100f01ec0
 [<a000000100016030>] dump_stack+0x30/0x60
                                sp 00000100f07d90 bsp 00000100f01ea0
 [<a0000001000f3d10>] validate_chain+0x1010/0x2660
                                sp 00000100f07d90 bsp 00000100f01e28
 [<a0000001000f6880>] __lock_acquire+0x1520/0x1660
                                sp 00000100f07de0 bsp 00000100f01d90
 [<a0000001000f6b50>] lock_acquire+0x190/0x200
                                sp 00000100f07de0 bsp 00000100f01d30
 [<a00000010081ba50>] _raw_spin_lock+0x50/0xe0
                                sp 00000100f07df0 bsp 00000100f01cf8
 [<a0000001001e9450>] kmem_cache_free+0x2f0/0x560
                                sp 00000100f07df0 bsp 00000100f01ca8
 [<a0000001001e97a0>] slab_destroy+0xe0/0x100
                                sp 00000100f07e00 bsp 00000100f01c70
 [<a0000001001e9a20>] free_block+0x260/0x2e0
                                sp 00000100f07e10 bsp 00000100f01c10
 [<a0000001001ea4b0>] kfree+0x490/0x620
                                sp 00000100f07e10 bsp 00000100f01bb8
 [<a0000001001eafb0>] do_tune_cpucache+0x450/0xba0
                                sp 00000100f07e20 bsp 00000100f01b40
 [<a0000001001ebbb0>] enable_cpucache+0x110/0x1e0
                                sp 00000100f07e20 bsp 00000100f01b08
 [<a000000100acf380>] kmem_cache_init_late+0x60/0x240
                                sp 00000100f07e20 bsp 00000100f01ae0
 [<a000000100aa92b0>] start_kernel+0x570/0x980
                                sp 00000100f07e20 bsp 00000100f01a60
 [<a0000001007eaa00>] start_ap+0x760/0x780
                                sp 00000100f07e30 bsp 00000100f019c0
Console: colour dummy device 80x25
Lock dependency validator: Copyright (c) 2006 Red Hat, Inc., Ingo Molnar
... MAX_LOCKDEP_SUBCLASSES:  8
... MAX_LOCK_DEPTH:          48
... MAX_LOCKDEP_KEYS:        1023
... CLASSHASH_SIZE:          512
... MAX_LOCKDEP_ENTRIES:     16384
... MAX_LOCKDEP_CHAINS:      32768
... CHAINHASH_SIZE:          16384

memory used by lock dependency info: 2839 kB
per task-struct memory footprint: 2688 bytes
----------------------------------
| Locking API testsuite disabled |
----------------------------------

caught one possbile recurive locking

#3 lore raw
In reply to Alex Chiang

* Luming Yu <[email protected]>:

Quoted 41 lines · Luming
Quoted 32 lines

Let me check if I can get access to an HP rx8640 to re-spin the patch.

-----Original Message-----
From: Alex Chiang [mailto:[email protected]]
Sent: Tuesday, January 12, 2010 3:58 AM
To: Yu, Luming; Luck, Tony
Cc: [email protected]
Subject: lockdep support caused an MCA

This patch 2f02b4a12b24c9f077c6d739ac23c9b90840ccca:

Author: Luming Yu <[email protected]>  2009-12-04 09:14:30

   [IA64] lockdep support

   Basic functionality for lockdep on ia64.

   Known issues:
   0. dynamic allocate per CPU area in lockdep.
   1. lock held in leave kernel.
   2. enabling CONFIG_DEBUG_LOCKDEP triggers the lockdep warning.
   3. Need to implement save_stack functions
   4. Replace some cpp macros with inline functions
   5. Add nmi_enter/nmi_exit

causes a very early MCA on an HP rx8640.

Quoted 7 lines

Unfortunately, even with early_printk turned on, the machine encounters the MCA and reboots before any output occurs, so I don't have any further information.

I haven't done any other triage on this as the patch has already been dropped from Tony's tree, but I'd be happy to test the next go-around.

Quoted 5 lines

I have managed to get the patch run on a HP rx8640 system with latest next tree + ia64-lockdep patch. The following is the part of boot log with lockdep footprints..The interesting thing is lockdep code seems to have detected possible recursive locking with the kernel on the platform.

How many cells was your system? Mine was 2 cells (for a total of 16 CPUs). I just mention this because booting on a 1 cell config is quite different from a multi-cell config.

Thanks for getting it going though.

/ac

#4 lore raw
In reply to Alex Chiang
Quoted 4 lines

How many cells was your system? Mine was 2 cells (for a total of 16 CPUs). I just mention this because booting on a 1 cell config is quite different from a multi-cell config.

Looks like mine is a 2 cells system too, judging by the 16 CPUS brought up from another part of boot log, which comes from same boot as my previous mail in this thread.

...
Brought up 16 CPUs
Total of 16 processors activated (50987.00 BogoMIPS).
...

Thanks,
Luming