Re: [RFC 00/15] x86_64: Optimize percpu accesses

[Christoph Lameter]

Jeremy Fitzhardinge wrote:
> Christoph Lameter wrote:
>> Jeremy Fitzhardinge wrote:
>>
>>  
>>> You want to virtually map the percpu area?  How and when would it get
>>> extended?
>>>     
>>
>> It would get extended when cpu_alloc() is called and the allocator
>> finds that there is no per cpu memory available.
>>   
> 
> Which, I take it, allocates percpu memory.  It would have the same
> caveats as vmalloc memory, with respect to accessing it during fault
> handlers and nmi handlers, I take it.

Right. One would not want to allocate per cpu memory in those contexts. The current allocpercpu() functions already have those restrictions.

> How would cpu_alloc() actually get used?  It doesn't make much sense for
> general code, since we don't have the notion of a percpu pointer to
> memory (vs a pointer to percpu memory).  Is the intended use for
> allocating percpu memory in modules?  What other uses?

Argh. Do I have to reexplain all of this all over again? Please look at the latest cpu_alloc patchset and the related discussions.

The cpu_alloc patchset introduces the concept of a pointer to percpu memory. Or you could call it an offset into the percpu segment that can be treated (in a restricted way) like a pointer....



--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/