[PATCH 15/17] media: st_rc: Don't stay on an IRQ handler forever

Mauro Carvalho Chehab mchehab at s-opensource.com
Fri Apr 13 03:00:50 PDT 2018


Em Fri, 13 Apr 2018 10:40:05 +0100
Sean Young <sean at mess.org> escreveu:

> On Fri, Apr 13, 2018 at 06:06:46AM -0300, Mauro Carvalho Chehab wrote:
> > Hi Sean,
> > 
> > Em Thu, 12 Apr 2018 23:21:32 +0100
> > Sean Young <sean at mess.org> escreveu:
> >   
> > > On Thu, Apr 12, 2018 at 11:24:07AM -0400, Mauro Carvalho Chehab wrote:  
> > > > As warned by smatch:
> > > > 	drivers/media/rc/st_rc.c:110 st_rc_rx_interrupt() warn: this loop depends on readl() succeeding
> > > > 
> > > > If something goes wrong at readl(), the logic will stay there
> > > > inside an IRQ code forever. This is not the nicest thing to
> > > > do :-)
> > > > 
> > > > So, add a timeout there, preventing staying inside the IRQ
> > > > for more than 10ms.    
> > > 
> > > If we knew how large the fifo was, then we could limit the loop to that many
> > > iterations (maybe a few extra in case IR arrives while we a reading, but
> > > IR is much slower than a cpu executing this loop of course).  
> > 
> > IR is slower, but this code is called at IRQ time, e. g. when the
> > controller already received the IR data. Also, it reads directly
> > via a memory mapped register, with should be fast.  
> 
> There is a chance that a new IR edge occurs whilst reading the fifo. All of
> this is academic since we don't know the size of the fifo.
> 
> > I suspect that
> > 10ms is a lot more time than what would be required to go though
> > all the FIFO data.  
> 
> 10ms seems far too much. The serial_ir prints a warning if it loops more
> than 255 times (and breaks out of the loop). Since we don't know the size
> of the fifo, that will be more than enough. We could also limit it to 
> 512 times, since the raw IR kfifo is 512 and any more than that would
> not fit in the kfifo anyway. This would exit much quicker than 10ms.
> 
> At least winbond-cir has a similar loop, there might be other drivers.

Yeah, we could limit it to run only 512 times (or some other reasonable
quantity), but in order to do that, we need to be sure that, on each read(),
the FIFO will shift - e. g. no risk of needing to do more than one read
to get the next element. That would work if the FIFO is implemented via
flip-flops. But if it is implemented via some slow memory, or if the
shift logic is implemented via some software on a micro-controller, it
may need a few interactions to get the next value.

Without knowing about the hardware implementation, I'd say that setting
a max time for the whole FIFO interaction is safer.

Thanks,
Mauro



More information about the linux-arm-kernel mailing list