Re: [PATCH v2] spi: bcm2835: Enable shared interrupt support

[Robin Murphy <robin.murphy@arm.com>]

On 2020-06-15 18:04, Florian Fainelli wrote:
> 
> 
> On 6/15/2020 10:00 AM, Mark Brown wrote:
>> On Mon, Jun 15, 2020 at 09:34:58AM -0700, Florian Fainelli wrote:
>>
>>> OK, so this has been dropped for spi/for-next right? How do we move from
>>> there?
>>
>> Well, I actually have it queued up for applying so unless I pull it
>> before my scripts get that far through the stuff I queued over the merge
>> window it'll go in (I dropped it due to it not being a bugfix).  If it
>> were me I'd go with the two instruction hit from checking the flag TBH
>> but otherwise I guess __always_inline should work for compilers that
>> misoptimize.  None of this is getting in the way of the framework so if
>> everyone involved in the driver is happy to spend time optimising it
>> and dealing with the fragility then it's fine by me.
> 
> OK, how about I send you an increment patch (would a fixup be okay?)
> that adds __always_inline since we know from this thread that some
> compilers may mis-optimize the function inlining?

Now that I've been inclined to go and look up the documentation, are we 
sure this so-very-contentious check is even correct? From my reading of 
things we're checking whether the RXR interrupt function is *enabled*, 
which still says nothing about whether either condition for the 
interrupt being *asserted* is true (RXR = 1 or DONE = 1). Thus if more 
than one SPI instance is active at once we could still end up trying to 
service an IRQ on a controller that didn't raise it.

Robin.