On Wed, Apr 15, 2015 at 04:00:32PM +0800, Huiquan Zhong wrote:
> master->pump_messages workqueue can be queued by spi_start_queue(),
> __spi_queued_transfer().
>
> there is one case that if one resume thread call spi_start_queue(),
> and at the same time another spi_device thread call spi_queued_transfer()
> to do spi transfer, then the first workqueue will start the spi transfer,
> but the next workqueue queued before SPI transfer complete, will unusual
> terminate spi transfer.
I've applied this, it took a while mostly because it is very hard for me
to understand your commit message - I'm not clear what mechanism will
cause a problem for a running transfer ("unusual terminate spi
transfer"). I can see we might end up with the work being added to the
queue more often than is needed but not how that ends up causing
anything other than a little wasted work.
Your commit message suggests it's to do with multiple devices adding to
the queue simultaneously but the queue access in __spi_queued_transfer()
and __spi_pump_messages() all appears to be done with the queue lock
held...
I'm still not 100% clear that this is doing anything other than making
the locking a bit more consistent.
Attachment:
signature.asc
Description: Digital signature
