Hi, John Youn <John.Youn@xxxxxxxxxxxx> writes: > On 2/12/2016 2:05 AM, Felipe Balbi wrote: >> >> Hi, >> >> John Youn <John.Youn@xxxxxxxxxxxx> writes: >>> On 2/10/2016 1:07 PM, Felipe Balbi wrote: >>>> John Youn <John.Youn@xxxxxxxxxxxx> writes: >>>>>>> Basically assign all the resources in advance. >>>>>> >>>>>> I thought about that, but wouldn't this, essentially, enable all >>>>>> endpoints unconditionally ? This could, potentially, increase power >>>>>> consumption on some systems, right ? This could also cause "spurious" >>>>>> interrupts if a bogus host tries to move data on an endpoint which >>>>>> hasn't been enabled yet. >>>>> >>>>> No, I mean to just assign resources withouth configuring or enabling >>>>> the endpoint. I have tested this approach and it works. But I still >>>> >>>> oh ok. >>>> >>>>> need to verify that it won't conflict with anything, such as streams. >>>> >>>> yeah, we would probably have an issue with streams. IIRC, we allocate >>>> one transfer resource per stream, right ? >>> >>> Ends up that is not a concern. Streams always use a single resource >>> per endpoint, not stream. >> >> hey, that's great. So what's the idea ? static resource assignment on >> endpoint initialization ? >> > > Yes that's it. I will go ahead and submit this fix. See the commit > message for details. I verified with engineers and did a round of > testing and so far no problems. > > If you prefer to only assign resources as needed, I have a separate > fix that I can submit if you want. > > Also, I think we need to handle backporting separately as neither > patch applies cleanly to 4.3. when backporting, we will receive emails that Greg couldn't apply. All we have to do is reply with the proper backport ;-) -- balbi
Attachment:
signature.asc
Description: PGP signature
