[Cbe-oss-dev] Should rndcntl_stop disappear?

[Julio M. Merino Vidal]

On 17/03/2007, at 23:53, Arnd Bergmann wrote:

> On Saturday 17 March 2007 23:42:22 Julio M. Merino Vidal wrote:
>> This function is defined in spufs.h, struct spu_context_ops.  After
>> some inspection, I found that this has been in the code since the
>> beginning, and was latter removed in commit
>> b9e3bd774bb1a90fee9b90f461a51e4ba295fe6d, available in both
>> branches.  However, it was reintroduced in commit
>> f071f0230b4e5c062ed3899adacf2f00fc1a67e2, only found in the ps3-linux
>> branch.
>>
>> So I'm wondering... should rndcntl_stop be really available in struct
>> spu_context_ops?  Or should it die again in favour of rndcntl_write
>> (which is what all callers do)?  I'll happily provide a trivial patch
>> for this if it is desired.
>
> We have a yet unresolved problem in the ps3 linux tree because the
> master control, which is supposed to replace the run bit in the
> runcntl register, is not accessible there.
>
> In short, runcntl_stop should go away, but just removing the function
> is not the real solution, we must actually make sure that the kernel
> accessing the runcntl variable does not race against users accessing
> it, in particular while a context is loaded into an spu but not  
> currently
> executing.

Ah, OK; thanks for the details.  I just thought its readdition was  
something that slipped in when committing some bigger changes and  
thus could be "fixed for cleanliness".

-- 
Julio M. Merino Vidal <jmerino at ac.upc.edu>