[PATCH v5] introduce macro spin_event_timeout()

[Josh Boyer]

On Tue, Mar 10, 2009 at 05:58:58PM -0500, Scott Wood wrote:
> Josh Boyer wrote:
>> On Tue, Mar 10, 2009 at 05:33:08PM -0500, Scott Wood wrote:
>>> Timur Tabi wrote:
>>>> The macro spin_event_timeout() takes a condition and timeout value
>>>> (in microseconds) as parameters.  It spins until either the condition is true
>>>> or the timeout expires.  It returns zero if the timeout expires first, non-zero
>>>> otherwise.
>>>>
>>>> This primary purpose of this macro is to poll on a hardware register until a
>>>> status bit changes.  The timeout ensures that the loop still terminates if the
>>>> bit doesn't change as expected.  This macro makes it easier for driver
>>>> developers to perform this kind of operation properly.
>>>>
>>>> Signed-off-by: Timur Tabi <timur at freescale.com>
>>>> ---
>>>>
>>>> v5: ported to arch/powerpc, made it powerpc-specific, eliminated udelay
>>> Why make it powerpc-specific?  This would be nice to have in   
>>> arch-independent code.
>>
>> That's just mean.  He already posted it to lkml and was told to make it
>> powerpc specific by Alan.
>
> Well, that's what happens when a discussion hops mailing lists with no  
> backreference. :-P
>
> I don't see anywhere where he says it should be architecture dependent,  
> but rather a general "I don't like this, get off my lawn!" response.
>
> I cannot agree with the "we shouldn't be encouraging this" sentiment;  
> people don't generally do spin loops because they're lazy[1], but rather  
> because the hardware demands it -- and it's hardly only on powerpc (much  
> less just "some Freescale drivers") that I've encountered hardware that  
> demands it, typiclally during reset/initialization or similarly non-hot  
> paths.  Why not provide something less likely to have bugs (the timeout  
> case is unlikely to be well tested), more easily seen when reviewing a  
> patch, and more likely to result in spin loops *with* a timeout rather  
> than without?

Excellent questions.  Did you send them to lkml and Alan?

josh