Re: [PATCH] Revert "serial: 8250_dma: don't bother DMA with small transfers"

[Andy Shevchenko <andriy.shevchenko-VuQAYsv1563Yd54FQh9/CA@public.gmane.org>]

On Wed, 2015-09-30 at 09:52 -0400, Peter Hurley wrote:
> [ + Andy Shevchenko ]
> 
> On 09/30/2015 08:46 AM, Frederic Danis wrote:
> > This reverts commit 9119fba0cfeda6d415c9f068df66838a104b87cb.
> 
> Hi Frederic,
> 
> It's customary to mail the commit author in patches which revert
> their work.
> 
> > This commit prevents from sending "big" file using Bluetooth.
> > When sending a lot of data quickly through the Bluetooth interface, 
> > and
> > after a variable amount of data sent, transfer fails with error:
> >     kernel: [  415.247453] Bluetooth: hci0 hardware error 0x00
> 
> Hmmm, was there any other information regarding why this happened?

Frederic informed us privately about this issue. I tried to hunt the
root cause but mostly (there was one observation but apparently seems
it is not related to this issue) nothing looks wrong.

> Eg., did the driver mistakenly re-order i/o (DMA occurred after PIO)?

I have few tracing dumps which Frederic provides to me, but I couldn't
see such a problem. Maybe I missed it. My commit 9119fba0cfed
introduces interleaved IO (DMA/PIO) which makes it like DMA -> PIO ->
DMA -> ...


> Or, might there be some race condition in the dmaengine driver on 
> this
> platform?

There is [1] might be a clue, though I don't think it's directly
related.

[1] http://www.spinics.net/lists/linux-spi/msg01151.html
The change prevents interleaved DMA/PIO IO.

> 
> Regards,
> Peter Hurley
> 
> > Found on T100TA.
> > 
> > After reverting this commit, send works fine for any file size.
> > 
> > Signed-off-by: Frederic Danis <frederic.danis-VuQAYsv1563Yd54FQh9/CA@public.gmane.org>
> > ---
> >  drivers/tty/serial/8250/8250_dma.c | 4 ----
> >  1 file changed, 4 deletions(-)
> > 
> > diff --git a/drivers/tty/serial/8250/8250_dma.c 
> > b/drivers/tty/serial/8250/8250_dma.c
> > index 21d01a4..e508939 100644
> > --- a/drivers/tty/serial/8250/8250_dma.c
> > +++ b/drivers/tty/serial/8250/8250_dma.c
> > @@ -80,10 +80,6 @@ int serial8250_tx_dma(struct uart_8250_port *p)
> >  		return 0;
> >  
> >  	dma->tx_size = CIRC_CNT_TO_END(xmit->head, xmit->tail, 
> > UART_XMIT_SIZE);
> > -	if (dma->tx_size < p->port.fifosize) {
> > -		ret = -EINVAL;
> > -		goto err;
> > -	}
> >  
> >  	desc = dmaengine_prep_slave_single(dma->txchan,
> >  					   dma->tx_addr + xmit
> > ->tail,
> > 
> 

-- 
Andy Shevchenko <andriy.shevchenko-VuQAYsv1563Yd54FQh9/CA@public.gmane.org>
Intel Finland Oy