On Wed, May 25, 2022 at 04:46:03PM +0200, David Jander wrote:
> David Jander <david@xxxxxxxxxxx> wrote:
> > +static void __spi_transfer_message_noqueue(struct spi_controller *ctlr, struct spi_message *msg)
> > +{
> > + bool was_busy;
> > + int ret;
> > +
> > + mutex_lock(&ctlr->io_mutex);
> > +
> > + /* If another context is idling the device then wait */
> > + while (ctlr->idling) {
> > + printk(KERN_INFO "spi sync message processing: controller is idling!\n");
> > + usleep_range(10000, 11000);
> > + }
> This is dead ugly of course, and it needs to be removed. Not yet sure how,
> hence the RFC. Maybe the idle -> not busy transition can be included inside
> the io_mutex? That way this while will never be hit and can be removed...
I'm not sure it's even quite right from a safety point of view - idling
is protected by queue_lock but this now only takes io_mutex. Moving
idling (and all the was_busy stuff) within the io_mutex would definitely
resolve the issue, the async submission context is the only one that
really needs the spinlock and it doesn't care about idling. I can't
think what you could do with the io_mutex when idling so it seems to
fit.
Attachment:
signature.asc
Description: PGP signature
