On 2018-09-10 16:56, Mark Brown wrote:

On Mon, Sep 10, 2018 at 09:27:09AM +0530, dkota@codeaurora.org wrote:

quoted

quoted

The thing is, we want it to be 100% reliable, not 99.9% reliable.  Is
it somehow wrong to add the spinlock?  ...or are you noticing
performance problems with the spinlock there?  It's just nice not to
have to think about it.

quoted

As I said, timeout will be handled after the calculated time as per 
data
size and speed. Enough time is given for interrupt, there is no chance 
of
interrupt occurrence during the handle_fifo_timeout(). So there is no 
need
of spinlock.

Assuming nothing goes wrong - the system isn't under unusually heavy
load for example, there's some oversight in the code, there's no impact
from power management causing things to run more slowly than you were
expecting, someone uses the driver on a new bit of hardware where there
are extra considerations or whatever else might go wrong.  Like Doug
says unless we're in some performance critical situation where it's
worth thinking *really* hard about how things really are actually safe
even though they might not look it it's both easier and more
maintainable to just write software that's obviously safe to 
inspection.

Agree with this perspective. There wont be any performance impact with 
spinlock. I will include the spinlock in the code.