lkml.org 
[lkml]   [2014]   [Dec]   [8]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [RFC PATCH V2 0/4] Reducing parameters of alloc_pages* family of functions
On 12/06/2014 02:07 AM, Linus Torvalds wrote:
> On Fri, Dec 5, 2014 at 11:59 AM, Vlastimil Babka <vbabka@suse.cz> wrote:
>> Hey all,
>>
>> this is a V2 of attempting something that has been discussed when Minchan
>> proposed to expand the x86 kernel stack [1], namely the reduction of huge
>> number of parameters that the alloc_pages* family and get_page_from_freelist()
>> functions have.
>
> So I generally like this, but looking at that "struct alloc_context",
> one member kind of stands out: the "order" parameter doesn't fit in
> with all the other members.
>
> Most everything else is describing where or what kind of pages to work
> with. The "order" in contrast, really is separate.
>
> So conceptually, my reaction is that it looks like a good cleanup even
> aside from the code/stack size reduction, but that the alloc_context
> definition is a bit odd.
>
> Quite frankly, I think the :"order" really fits much more closely with
> "alloc_flags", not with the alloc_context. Because like alloc_flags,.
> it really describes how we need to allocate things within the context,
> I'd argue.
>
> In fact, I think that the order could actually be packed with the
> alloc_flags in a single register, even on 32-bit (using a single-word
> structure, perhaps). If we really care about number of parameters.
>
> I'd rather go for "makes conceptual sense" over "packs order in
> because it kind of works" and we don't modify it".
>
> Hmm?

Thanks for the suggestions, order indeed stands out. I'll check if it
makes more sense to have it separately, or pack as you suggest. Packing
could perhaps bring more complexity than the benefit of less parameters.
But the suggestion made me realize that migratetype could be also packed
into alloc_flags and it would be more straightforward for that than order.

With order and migratetype out, everything left in alloc_context would
be about nodes and zones, which is also good I guess. Maybe a different
name for the structure then?

Vlastimil

>
> Linus
>



\
 
 \ /
  Last update: 2014-12-08 10:01    [W:0.061 / U:1.344 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site