Re: kernel 2.6 : cdc_acm problem

Re: kernel 2.6 : cdc_acm problem

[David Brownell]

The problem is that cdc_acm calls a "softirq-only" routine
in a hardirq context.  See this patch:

http://marc.theaimsgroup.com/?l=linux-usb-devel&m=106764585001038&w=2

It's not clear that'll make it into 2.6.0-final.

- Dave

Re: kernel 2.6 : cdc_acm problem

[Greg KH]

On Sun, Nov 09, 2003 at 09:21:59AM -0800, David Brownell wrote:
> The problem is that cdc_acm calls a "softirq-only" routine
> in a hardirq context.  See this patch:
> 
> http://marc.theaimsgroup.com/?l=linux-usb-devel&m=106764585001038&w=2
> 
> It's not clear that'll make it into 2.6.0-final.

I've not planned to submit it for 2.6.0 as it's a relativly big change,
and I don't have the hardware to test it out.  Anyone have any other
thoughts about this?

thanks,

greg k-h

Re: kernel 2.6 : cdc_acm problem

[David Brownell]

>>The problem is that cdc_acm calls a "softirq-only" routine
>>in a hardirq context.  See this patch:
>>
>>http://marc.theaimsgroup.com/?l=linux-usb-devel&m=106764585001038&w=2
>>
>>It's not clear that'll make it into 2.6.0-final.
> 
> 
> I've not planned to submit it for 2.6.0 as it's a relativly big change,
> and I don't have the hardware to test it out.  Anyone have any other
> thoughts about this?

Doesn't seem big to me.  It could be shrunk a smidgeon, but
that's the version that's gotten the positive test results.

The folk who have this kind of hardware have reported this
happening for quite a few months now, and it does seem to
fill up log buffers with catastrophic-seeming stack traces.

Colin, does it fix your problem?  Can you eke more than
twenty minutes from your laptop battery now?  :)

- Dave

Re: kernel 2.6 : cdc_acm problem

[Colin Leroy]

On 10 Nov 2003 at 04h21, David Brownell wrote:

Hi,

>>> http://marc.theaimsgroup.com/?l=linux-usb-devel&m=106764585001038&w=2

> Colin, does it fix your problem?  Can you eke more than
> twenty minutes from your laptop battery now?  :)

That works great :-)
Many thanks for the pointer. I think it'd be nice if this was in 2.6.0 :)

-- 
Colin
http://www.colino.net/

Re: kernel 2.6 : cdc_acm problem

[Peter Matthias]

Greg KH schrieb:

> On Sun, Nov 09, 2003 at 09:21:59AM -0800, David Brownell wrote:
>> The problem is that cdc_acm calls a "softirq-only" routine
>> in a hardirq context.  See this patch:
>> 
>> http://marc.theaimsgroup.com/?l=linux-usb-devel&m=106764585001038&w=2
>> 
>> It's not clear that'll make it into 2.6.0-final.
> 
> I've not planned to submit it for 2.6.0 as it's a relativly big change,
> and I don't have the hardware to test it out.  Anyone have any other
> thoughts about this?

IMO it's a must be fixed issue for 2.6.0 and it works on my system.

Peter

Re: kernel 2.6 : cdc_acm problem

[Benjamin Herrenschmidt]

On Sun, 2003-11-09 at 22:20, Colin Leroy wrote:
> Hi,
> 
> just to report a problem with kernel 2.6.0-tes9-benh.. Using my GPRS phone
> to access internet spits out lots (real lots) of these to syslog:
> 
> Nov  8 23:36:21 chloe kernel: Call trace:
> Nov  8 23:36:21 chloe kernel:  [c000b604] dump_stack+0x18/0x28
> Nov  8 23:36:21 chloe kernel:  [c000885c] check_bug_trap+0x84/0x98
> Nov  8 23:36:21 chloe kernel:  [c0008940] ProgramCheckException+0xd0/0x170
> Nov  8 23:36:21 chloe kernel:  [c0007ee0] ret_from_except_full+0x0/0x4c
> Nov  8 23:36:21 chloe kernel:  [c0021250] local_bh_enable+0x28/0x60
> Nov  8 23:36:21 chloe kernel:  [c0109c68] ppp_input+0x204/0x21c
> Nov  8 23:36:21 chloe kernel:  [c010e99c] ppp_async_input+0x50c/0x618
> Nov  8 23:36:21 chloe kernel:  [c010d890] ppp_asynctty_receive+0x50/0xac
> Nov  8 23:36:21 chloe kernel:  [c00d5c7c] flush_to_ldisc+0xa0/0xb0
> Nov  8 23:36:21 chloe kernel:  [c00d5d34] tty_flip_buffer_push+0x1c/0x3c
> Nov  8 23:36:21 chloe kernel:  [d98fb250] acm_read_bulk+0xdc/0x138
> [cdc_acm]
> Nov  8 23:36:21 chloe kernel:  [c017e298] usb_hcd_giveback_urb+0x90/0xac
> Nov  8 23:36:21 chloe kernel:  [c0187ac8] finish_urb+0xd8/0xec
> Nov  8 23:36:21 chloe kernel:  [c0188f28] dl_done_list+0x7c/0x128
> Nov  8 23:36:21 chloe kernel:  [c0189a48] ohci_irq+0xec/0x1b4
> Nov  8 23:36:21 chloe kernel: Badness in local_bh_enable at
> kernel/softirq.c:121
> 
> This makes the computer very irresponsive and the log-storm emptied the
> battery in about 20 minutes :-)

Looks like a bug calling local_bh_enable with irq off, though it isn't
clear who is disabling them at this point. I don't have time to track
it down more precisely right now, it could be either OHCI finishing the
urbs within a spinlock_irqsave or some TTY or PPP issue, the ACM driver
doesn't seem to play with irq masking by itself.

Ben.