[Cbe-oss-dev] [patch 01/18] cell: add per BE structure with info about its SPUs

[André Detsch]

Jeremy Kerr wrote:
> Andre,
> 
>> * number of reserved (non scheduleable) spus.
> 
> Do we actually need this information?
> 
> As far as I can tell, the only thing we use the reserved_spus variable 
> for is in spufs_assert_affinity(), and I'm not convinced that the check 
> there is really necessary (however, it's not doing any harm...). The 
> NOSCHED contexts may be as short-lived as the normal ones, and the SPEs 
> may be be available for use when needed.

It really depends on what we expect to be the typical case for NOSCHED 
contexts. By doing the check, we can avoid some starvation situations, 
but, in fact, this is only effective if NOSCHED ctxs are typically long 
lived.

Do we already have a better view of what NOSCHED scenarios we expect to 
have? If this includes short-lived NOSCHED, we can drop the use of the 
'number of non scheduleable' field. However, as you mentioned, having it 
for now does not hurt.

 > I'm not vetoing this, just asking for more info :)

Ok, np :)

--
Andre Detsch