Problem - ohci1394.c:ohci_devctl ends up calling dma_pool_destroy from 
invalid context. Below is the dmesg output when I exit Kino after video 
capture -

Debug: sleeping function called from invalid context at 
include/asm/semaphore.h:107
in_atomic():1, irqs_disabled():1
 [<c0104c2e>] dump_stack+0x1e/0x20
 [<c011f8a2>] __might_sleep+0xa2/0xc0
 [<c028c660>] dma_pool_destroy+0x20/0x140
 [<f0b7affe>] free_dma_rcv_ctx+0x8e/0x150 [ohci1394]
 [<f0b774a4>] ohci_devctl+0x214/0x9b0 [ohci1394]
 [<f0e7aa99>] handle_iso_listen+0x2d9/0x310 [raw1394]
 [<f0e7f22b>] state_connected+0x29b/0x2b0 [raw1394]
 [<f0e7f2de>] raw1394_write+0x9e/0xd0 [raw1394]
 [<c0183ef2>] vfs_write+0xc2/0x170
 [<c018406b>] sys_write+0x4b/0x80
 [<c0103cb1>] sysenter_past_esp+0x52/0x75

Attached patch against 2.6.11-rc2 (tested to work normally with a  
camcorder device) enables ohci_devctl to defer the work of destroying 
the dma pool by using a work queue, so the dma pool is destroyed in a 
valid context and we no longer get an error in dmesg (duh :)

Note : There still exists  one more similar problem with ohci1394 - that 
is with dma_pool_create being called from invalid context - this happens 
in response to ISO_LISTEN_CHANNEL. I get dmesg debug for this case which 
is similar to the above. My analysis is that this one is non-trivial to 
fix. More on this in a separate mail.

Patch is attached since T'Bird messes up with the inlined one. (Pls. 
suggest a good email client for kernel patch stuff :)

Signed-off-by: Parag Warudkar (kernel-stuff@comcast.net)


Parag