[PATCH 4/7] soc: fsl: qe: replace spin_event_timeout by readx_poll_timeout_atomic

[Michael Ellerman]

Rasmus Villemoes <linux at rasmusvillemoes.dk> writes:
> On 18/10/2019 18.08, Christoph Hellwig wrote:
>> On Fri, Oct 18, 2019 at 02:52:31PM +0200, Rasmus Villemoes wrote:
>>>  	/* wait for the QE_CR_FLG to clear */
>>> -	ret = spin_event_timeout((ioread32be(&qe_immr->cp.cecr) & QE_CR_FLG) == 0,
>>> -				 100, 0);
>>> -	/* On timeout (e.g. failure), the expression will be false (ret == 0),
>>> -	   otherwise it will be true (ret == 1). */
>>> +	ret = readx_poll_timeout_atomic(ioread32be, &qe_immr->cp.cecr, val, (val & QE_CR_FLG) == 0,
>> 
>> This creates an overly long line.
>
> Yeah, readx_poll_timeout_atomic is a mouthful, and then one also has to
> put in the name of the accessor... I'll wrap it when I respin the
> series, thanks.
>
>> Btw, given how few users of spin_event_timeout we have it might be good
>> idea to just kill it entirely.
>
> Maybe. That's for the ppc folks to comment on; the iopoll.h helpers are
> not completely equivalent (because obviously they don't read tbl
> directly).

AFAICS it was added by and only ever used by Freescale folks, most of
whom have now moved on to other things.

I'd be happy to see it replaced with something generic.

I created an issue in our github to remind us to do that cleanup. Though
in practice that probably just means we'll never get around to it:

  https://github.com/linuxppc/issues/issues/280

> Maybe the generic versions should be taught
> spin_begin/spin_end/spin_cpu_relax so at least that part would be
> drop-in replacement.

That would be nice. Though TBH the intersection of platforms that use
spin_event_timeout() and also define a non-empty spin_begin() etc. is
close to if not zero.

cheers