LOCKDEP warning if CONFIG_DETECT_SOFTLOCKUP=y

[Zilvinas Valinskas]

Hello lkml, 

Booting ixp4xx/ARM/BE  with lockdep enabled to test my code, I got a
lockdep warning (see below). Does it mean that soft-lockup detector is
started too early ?  Any advice how to approach and fix the problem are
appreciated very much. If CONFIG_DETECT_SOFTLOCKUP=n, lockdep does not
complain.

Kernel is vanilla 2.6.21-rc4 + lockdep-stacktrace.diff from
http://svn.wantstofly.org/kernel/ repository.


[    0.000000] Linux version 2.6.21-rc4 (zilvinas@zv) (gcc version 4.2.0 20070207 (prerelease) (http://www.wilibox.com)) #12 Fri Mar 16 16:05:39 EET 2007
[    0.000000] CPU: XScale-IXP42x Family [690541f1] revision 1 (ARMv5TE), cr=000039ff
[    0.000000] Machine: Intel IXDP425 Development Platform
[    0.000000] Memory policy: ECC disabled, Data cache writeback
...

[    0.000000] Built 1 zonelists.  Total pages: 16256
[    0.000000] Kernel command line: initcall_debug loglevel=8 i2c_debug=9 root=/dev/mtdblock2 console=ttyS0,115200 ro init=/linuxrc
[    0.000000] PID hash table entries: 256 (order: 8, 1024 bytes)
[    0.000000] Lock dependency validator: Copyright (c) 2006 Red Hat, Inc., Ingo Molnar
[    0.000000] ... MAX_LOCKDEP_SUBCLASSES:    8
[    0.000000] ... MAX_LOCK_DEPTH:          30
[    0.000000] ... MAX_LOCKDEP_KEYS:        2048
[    0.000000] ... CLASSHASH_SIZE:           1024
[    0.000000] ... MAX_LOCKDEP_ENTRIES:     8192
[    0.000000] ... MAX_LOCKDEP_CHAINS:      16384
[    0.000000] ... CHAINHASH_SIZE:          8192
[    0.000000]  memory used by lock dependency info: 1096 kB
[    0.000000]  per task-struct memory footprint: 1200 bytes
[    0.000000] ------------------------
[    0.000000] | Locking API testsuite:

... <skiped> ...

[    0.590000] -------------------------------------------------------
[    0.590000] Good, all 218 testcases passed! |
[    0.590000] ---------------------------------
[    0.590000] Dentry cache hash table entries: 8192 (order: 3, 32768 bytes)
[    0.600000] Calibrating delay loop... 265.42 BogoMIPS (lpj=1327104)
[    0.840000] Mount-cache hash table entries: 512
[    0.840000] CPU: Testing write buffer coherency: ok
[    0.840000] INFO: trying to register non-static key.
[    0.840000] the code is fine but needs lockdep annotation.
[    0.840000] turning off the locking correctness validator.

Don't know how to fix that.

[    0.840000] [<c0021018>] (dump_stack+0x0/0x14) from [<c0052ccc>] (__lock_acquire+0x640/0x1110)
[    0.840000] [<c005268c>] (__lock_acquire+0x0/0x1110) from [<c0053820>] (lock_acquire+0x84/0x98)
[    0.840000] [<c005379c>] (lock_acquire+0x0/0x98) from [<c01858b4>] (_spin_lock_irqsave+0x48/0x5c)
[    0.840000] [<c018586c>] (_spin_lock_irqsave+0x0/0x5c) from [<c002f358>] (sched_setscheduler+0xd4/0x2a8)
[    0.840000]  r6 = C04D6000  r5 = C04D1820  r4 = 00000001 
[    0.840000] [<c002f284>] (sched_setscheduler+0x0/0x2a8) from [<c0058848>] (watchdog+0x30/0x6c)
[    0.840000] [<c0058818>] (watchdog+0x0/0x6c) from [<c00494ec>] (kthread+0x108/0x118)
[    0.840000]  r4 = 00000000 
[    0.840000] [<c00493e4>] (kthread+0x0/0x118) from [<c0036098>] (do_exit+0x0/0x894)
[    0.840000]  r7 = 00000000  r6 = 00000000  r5 = 00000000  r4 = 00000000
[    0.840000] Calling initcall 0xc0009980: ptrace_break_init+0x0/0x2c()

With gdb it points to *__task_rq_lock(struct task_struct *p)

(gdb) l *sched_setscheduler+0xd4
0x1644 is in sched_setscheduler (kernel/sched.c:395).
390     {
391             struct rq *rq;
392
393     repeat_lock_task:
394             rq = task_rq(p);
395             spin_lock(&rq->lock);
396             if (unlikely(rq != task_rq(p))) {
397                     spin_unlock(&rq->lock);
398                     goto repeat_lock_task;
399             }
(gdb) 

I think it is really about ->pi_lock, on line 4134, that lockdep doesn't know 
anything about.

4130         /*
4131          * make sure no PI-waiters arrive (or leave) while we are
4132          * changing the priority of the task:
4133          */
4134         spin_lock_irqsave(&p->pi_lock, flags);
4135         /*
4136          * To be able to change p->policy safely, the apropriate
4137          * runqueue lock must be held.
4138          */
4139         rq = __task_rq_lock(p);


Best regards,
Åilvinas Valinskas

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/