[PATCH 3/5] comedi: ni_660x: don't poll without a DMA channel

[Greg Kroah-Hartman <gregkh@linuxfoundation.org>]

From: Josef Bacik <josef@toxicpanda.com>

A counter command owns a MITE DMA channel from ni_660x_cmd() to
ni_660x_cancel(). A command does not only end by COMEDI_CANCEL: when the
reader is too slow mite_sync_input_dma() sets COMEDI_CB_OVERFLOW, and
counter errors set COMEDI_CB_ERROR, and comedi_handle_events() then calls
->cancel() from the interrupt handler. That releases the channel and
clears counter->mite_chan, but s->busy stays set until the owner has read
the rest of the data and seen the end of the acquisition.

do_poll_ioctl() only looks at s->busy, so a COMEDI_POLL in that state
ends in mite_sync_dma(NULL, s), which dereferences the channel.

Test the channel like ni_pcidio_poll() and ni_sync_ai_dma() do. The
pointer only changes under dev->mutex, which the ioctl holds, or in the
interrupt handler under interrupt_lock, which is taken here.

Fixes: 894db1197347 ("Staging: Comedi: ni_600x: Added support for comedi_poll.")
Cc: stable <stable@kernel.org>
Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
---
 drivers/comedi/drivers/ni_660x.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/drivers/comedi/drivers/ni_660x.c b/drivers/comedi/drivers/ni_660x.c
index 5508cdb1b0ba..9238475dda07 100644
--- a/drivers/comedi/drivers/ni_660x.c
+++ b/drivers/comedi/drivers/ni_660x.c
@@ -459,7 +459,8 @@ static int ni_660x_input_poll(struct comedi_device *dev,
 
 	/* lock to avoid race with comedi_poll */
 	spin_lock_irqsave(&devpriv->interrupt_lock, flags);
-	mite_sync_dma(counter->mite_chan, s);
+	if (counter->mite_chan)
+		mite_sync_dma(counter->mite_chan, s);
 	spin_unlock_irqrestore(&devpriv->interrupt_lock, flags);
 	return comedi_buf_read_n_available(s);
 }

-- 
2.56.0