mmotm 2010-09-15 - BUG in mmc driver calling led_trigger_event()

[Valdis . Kletnieks]

On Wed, 15 Sep 2010 16:21:43 PDT, akpm@xxxxxxxxxxxxxxxxxxxx said:
> The mm-of-the-moment snapshot 2010-09-15-16-21 has been uploaded to
> 
>    http://userweb.kernel.org/~akpm/mmotm/

Dell Latitude E6500. lspci -v says:

03:01.2 SD Host controller: Ricoh Co Ltd R5C822 SD/SDIO/MMC/MS/MSPro Host Adapter (rev 21) (prog-if 01)
        Subsystem: Dell Device 024f
        Flags: bus master, medium devsel, latency 64, IRQ 18
        Memory at f1bff600 (32-bit, non-prefetchable) [size=256]
        Capabilities: [80] Power Management version 2
        Kernel driver in use: sdhci-pci

Not consistently repeatable - I had 3 clean boots of this kernel, hit this the 4th time.

[    2.928661] Freeing unused kernel memory: 892k freed
[    2.934051] BUG: sleeping function called from invalid context at kernel/mutex.c:278
[    2.934440] in_atomic(): 1, irqs_disabled(): 0, pid: 11, name: kworker/0:1
[    2.934777] 3 locks held by kworker/0:1/11:
[    2.934780]  #0:  (((dev_name(&(mmc)->class_dev)))){+.+...}, at: [<ffffffff81052388>] process_one_work+0x1b5/0x49d
[    2.934796]  #1:  ((&host->finish_work)){+.+...}, at: [<ffffffff81052388>] process_one_work+0x1b5/0x49d
[    2.934806]  #2:  (&trigger->leddev_list_lock){.+.+..}, at: [<ffffffff813d464e>] led_trigger_event+0x22/0x75
[    2.934821] Pid: 11, comm: kworker/0:1 Not tainted 2.6.36-rc4-mmotm0915 #2
[    2.934825] Call Trace:
[    2.934833]  [<ffffffff8102e456>] __might_sleep+0x124/0x129
[    2.934840]  [<ffffffff8156559e>] mutex_lock_nested+0x20/0x39
[    2.934846]  [<ffffffff813d1412>] sdhci_led_control+0x24/0x52
[    2.934852]  [<ffffffff813d464e>] ? led_trigger_event+0x22/0x75
[    2.934858]  [<ffffffff813d4682>] led_trigger_event+0x56/0x75
[    2.934865]  [<ffffffff813c74c8>] mmc_request_done+0x5c/0x7a
[    2.934871]  [<ffffffff813d1aba>] sdhci_finish_work+0xe6/0xef
[    2.934877]  [<ffffffff81052472>] process_one_work+0x29f/0x49d
[    2.934882]  [<ffffffff81052388>] ? process_one_work+0x1b5/0x49d
[    2.934888]  [<ffffffff813d19d4>] ? sdhci_finish_work+0x0/0xef
[    2.934895]  [<ffffffff81052d10>] worker_thread+0x17e/0x251
[    2.934901]  [<ffffffff81052b92>] ? worker_thread+0x0/0x251
[    2.934908]  [<ffffffff81056882>] kthread+0x7d/0x85
[    2.934915]  [<ffffffff81003554>] kernel_thread_helper+0x4/0x10
[    2.934922]  [<ffffffff815676c0>] ? restore_args+0x0/0x30
[    2.934928]  [<ffffffff81056805>] ? kthread+0x0/0x85
[    2.934934]  [<ffffffff81003550>] ? kernel_thread_helper+0x0/0x10
[    2.941048] mmc0: mmc_rescan: trying to init card at 200000 Hz

Only config diff between the first 3 boots and this one:

 #
 CONFIG_BINFMT_ELF=y
 CONFIG_COMPAT_BINFMT_ELF=y
-CONFIG_CORE_DUMP_DEFAULT_ELF_HEADERS=y
+# CONFIG_CORE_DUMP_DEFAULT_ELF_HEADERS is not set
 # CONFIG_HAVE_AOUT is not set
 CONFIG_BINFMT_MISC=y
 CONFIG_IA32_EMULATION=y

which shouldn't matter.

Anything obvious to you guys? Race condition of some sort, or somebody else
leaving a dangling in_atomic() status for us to get ambushed by? Maybe Hein's
patches to retry at multiple freqs are causing one of the retries to happen at a
bad time? I'm mystified.

Attachment: pgp00000.pgp
Description: PGP signature