Re: [Industrypack-devel] [PATCH] ipoctal: protect only the real critical section

["Samuel Iglesias Gonsálvez" <siglesias@igalia.com>]

[-- Attachment #1: Type: text/plain, Size: 2482 bytes --]

On Thu, 2014-06-26 at 09:52 +0200, Samuel Iglesias Gonsálvez wrote:
> On Thu, 2014-06-26 at 09:46 +0200, Federico Vaga wrote:
> > In some conditions (echo or particular sequence of special
> > characters), on buffer push, the tty layer calls the write operation
> > while we are holding the spinlock. This means deadlock within the same
> > process on kernels version < 3.12. It seems not a problem on recent
> > kernel, but the patch still valid as locking optimization.
> > 
> > The protected variables by the spinlock are: xmit_buf, nb_bytes,
> > pointer_read and pointer_write. So, this patch reduces the locked area
> > in the IRQ handler only to these variables. Most of the code inside the
> > locked area in the IRQ handler is not protected elsewhere; it means
> > that it is not protected at all.
> > 
> > Signed-off-by: Federico Vaga <federico.vaga@cern.ch>
> > ---
> >  drivers/ipack/devices/ipoctal.c |    5 ++---
> >  1 file changed, 2 insertions(+), 3 deletions(-)
> > 
> > diff --git a/drivers/ipack/devices/ipoctal.c b/drivers/ipack/devices/ipoctal.c
> > index 141094e..69687f1 100644
> > --- a/drivers/ipack/devices/ipoctal.c
> > +++ b/drivers/ipack/devices/ipoctal.c
> > @@ -177,19 +177,20 @@ static void ipoctal_irq_tx(struct ipoctal_channel *channel)
> >  	if (channel->nb_bytes == 0)
> >  		return;
> >  
> > +	spin_lock(&channel->lock);
> >  	value = channel->tty_port.xmit_buf[*pointer_write];
> >  	iowrite8(value, &channel->regs->w.thr);
> >  	channel->stats.tx++;
> >  	(*pointer_write)++;
> >  	*pointer_write = *pointer_write % PAGE_SIZE;
> >  	channel->nb_bytes--;
> > +	spin_unlock(&channel->lock);
> >  }
> >  
> >  static void ipoctal_irq_channel(struct ipoctal_channel *channel)
> >  {
> >  	u8 isr, sr;
> >  
> > -	spin_lock(&channel->lock);
> >  	/* The HW is organized in pair of channels.  See which register we need
> >  	 * to read from */
> >  	isr = ioread8(&channel->block_regs->r.isr);
> > @@ -213,8 +214,6 @@ static void ipoctal_irq_channel(struct ipoctal_channel *channel)
> >  	/* TX of each character */
> >  	if ((isr & channel->isr_tx_rdy_mask) && (sr & SR_TX_READY))
> >  		ipoctal_irq_tx(channel);
> > -
> > -	spin_unlock(&channel->lock);
> >  }
> >  
> >  static irqreturn_t ipoctal_irq_handler(void *arg)
> 
> Acked-by: Samuel Iglesias Gonsalvez <siglesias@igalia.com>
> 
> Sam

Greg, Would you mind picking this patch through your driver-core tree?

Thanks,

Sam

[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 819 bytes --]