device driver wake up problem on Ubuntu 7.04 (kernel 2.6.20)
- From: brian.keen@xxxxxxxxx
- Date: Sat, 08 Sep 2007 05:03:31 -0700
I have a problem with a PCI card character device driver that uses
dma.
We are using wait_event_interruptible_timeout() to block inside the
driver read() until a dma transfer completes. We call
wake_up_interruptible() in the interrupt handler to wake up the read
process on dma completion.
The read process actually uses a number of separate threads but a
semaphore in the driver ensures that there can only ever be one thread
on the wait queue at any time.
Everything works fine when using Ubuntu 6.06 (kernel 2.6.15) runing on
an Intel core duo processor.
We are now attempting to install the software onto a system that uses
an Intel quad core processor.
We found that the Ubuntu 6.06 installer would not run on this
processor, so rebuilt our code for Ubuntu 7.04 (kernel 2.6.20).
We now find that the call to wait_event_interruptible_timeout()
frequently fails to return after wake_up_interruptible() has been
called from the interrupt handler.
Eventually, after a hundred milliseconds or so, the hardware reports a
buffer overrun.
It does not fail every time - usually we get several successful dma
transfers before it hangs.
This suggests a timing/race problem.
I have added a few printk()'s in the driver to confirm that the dma
completion interrupt is being received and that
wake_up_interruptible() has been called from the ISR. A printk()
immediately before and after the call to
wait_event_interruptible_timeout() confirms that the read process
remains stuck on the wait queue.
I have tried adding a call to waitqueue_active() just before the call
to wake_up_interruptible() in the interrupt handler just to check that
the read process is ready and waiting on the queue.
Thinking that it may be an SMP issue, I tried printing the cpu number
in the ISR and the read process. This turned out to be 0 in all cases.
At the moment we are unsure whether the problem is related to the
kernel change from 2.6.15 to 2.6.20 or to the change from a duo core
to a quad core cpu.
It is possible that we have a timing problem in the driver that only
manifests itself with this combination.
We are hoping to try the Ubuntu 7.04 build on a dual-core processor
next week to try and pin the problem down further.
In the mean time, is anyone aware of any timing problems or SMP
pitfalls in this area that we should be aware of?
Regards,
Brian.
.
- Prev by Date: Re: writing drivers using C++
- Next by Date: Re: writing drivers using C++
- Previous by thread: kernel threads and signals
- Next by thread: Stupid question about c structures
- Index(es):
Relevant Pages
|
