LED heartbeat trigger - in_atomic() while allocs

I was playing around with the LED heartbeat trigger and it caught the
BUG() code when trying to activate.  Does anyone know why this call
stack would be in_atomic()?  Or any ideas for a patch that would allow
the heartbeat trigger to allocate when it is not in atomic context?

BUG: sleeping function called from invalid context at mm/slab.c:3024
in_atomic():1, irqs_disabled():0
Call Trace:
[def3dd90] [c0008e60] show_stack+0x48/0x194 (unreliable)
[def3ddc0] [c0014160] __might_sleep+0xd0/0xdc
[def3dde0] [c0063e60] kmem_cache_zalloc+0xdc/0x12c
[def3de00] [c01975c0] heartbeat_trig_activate+0x24/0x80
[def3de20] [c0196844] led_trigger_set+0x128/0x160
[def3de40] [c0196988] led_trigger_store+0x10c/0x1a8
[def3deb0] [c0162f40] class_device_attr_store+0x44/0x5c
[def3dec0] [c00ae614] sysfs_write_file+0x114/0x1dc
[def3def0] [c0068ea4] vfs_write+0x9c/0x110
[def3df10] [c0068ff4] sys_write+0x4c/0x90
[def3df40] [c000fc00] ret_from_syscall+0x0/0x38

--- Exception: c01 at 0xfe83818
    LR =3D 0xfe35448

Kernel 2.6.22-4, powerpc, targeting an MPC 8347.  I don't think this
issue is PowerPC specific but I was hoping other embedded people might
have experience with the LED class driver.

Thank you,
	Michael

Benedict, Michael wrote:

From a brief glance at the code, looks like led_trigger_store is

holding a rwlock while calling led_trigger_set.

Thanks, yeah I see that now.  I just changed the alloc to be GPF_ATOMIC
for now, but I don't really know how this was intended to work.
	-Michael=20

-----Original Message-----
From:=20
linuxppc-embedded-bounces+mbenedict=3Dtwacs.com@ozlabs.org=20
[mailto:linuxppc-embedded-bounces+mbenedict=3Dtwacs.com@ozlabs.o

rg] On Behalf Of Nathan Lynch

Sent: Monday, November 26, 2007 2:37 PM
To: Benedict, Michael
Cc: linuxppc-embedded@ozlabs.org
Subject: Re: LED heartbeat trigger - in_atomic() while allocs
=20
Benedict, Michael wrote:

quoted

I was playing around with the LED heartbeat trigger and it=20

caught the

quoted

BUG() code when trying to activate.  Does anyone know why this call
stack would be in_atomic()?  Or any ideas for a patch that=20

would allow

quoted

the heartbeat trigger to allocate when it is not in atomic context?
=20
BUG: sleeping function called from invalid context at mm/slab.c:3024
in_atomic():1, irqs_disabled():0
Call Trace:
[def3dd90] [c0008e60] show_stack+0x48/0x194 (unreliable)
[def3ddc0] [c0014160] __might_sleep+0xd0/0xdc
[def3dde0] [c0063e60] kmem_cache_zalloc+0xdc/0x12c
[def3de00] [c01975c0] heartbeat_trig_activate+0x24/0x80
[def3de20] [c0196844] led_trigger_set+0x128/0x160
[def3de40] [c0196988] led_trigger_store+0x10c/0x1a8
[def3deb0] [c0162f40] class_device_attr_store+0x44/0x5c
[def3dec0] [c00ae614] sysfs_write_file+0x114/0x1dc
[def3def0] [c0068ea4] vfs_write+0x9c/0x110
[def3df10] [c0068ff4] sys_write+0x4c/0x90
[def3df40] [c000fc00] ret_from_syscall+0x0/0x38

--- Exception: c01 at 0xfe83818
    LR =3D 0xfe35448

=20
Kernel 2.6.22-4, powerpc, targeting an MPC 8347.

=20
From a brief glance at the code, looks like led_trigger_store is
holding a rwlock while calling led_trigger_set.
_______________________________________________
Linuxppc-embedded mailing list
Linuxppc-embedded@ozlabs.org
https://ozlabs.org/mailman/listinfo/linuxppc-embedded
=20
=20