[lkml]   [2013]   [Dec]   [4]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
Messages in this thread
SubjectRe: [PATCH v2 08/23] mm/memblock: Add memblock memory allocation apis
On Wednesday 04 December 2013 11:07 AM, Tejun Heo wrote:
> Hello,
> On Wed, Dec 04, 2013 at 10:54:47AM -0500, Santosh Shilimkar wrote:
>> Well as you know there are architectures still using bootmem even after
>> this series. Changing MAX_NUMNODES to NUMA_NO_NODE is too invasive and
>> actually should be done in a separate series. As commented, the best
>> time to do that would be when all remaining architectures moves to
>> memblock.
>> Just to give you perspective, look at the patch end of the email which
>> Grygorrii cooked up. It doesn't cover all the users of MAX_NUMNODES
>> and we are bot even sure whether the change is correct and its
>> impact on the code which we can't even tests. I would really want to
>> avoid touching all the architectures and keep the scope of the series
>> to core code as we aligned initially.
>> May be you have better idea to handle this change so do
>> let us know how to proceed with it. With such a invasive change the
>> $subject series can easily get into circles again :-(
> But we don't have to use MAX_NUMNODES for the new interface, no? Or
> do you think that it'd be more confusing because it ends up mixing the
> two?
The issue is memblock code already using MAX_NUMNODES. Please
look at __next_free_mem_range() and __next_free_mem_range_rev().
The new API use the above apis and hence use MAX_NUMNODES. If the
usage of these constant was consistent across bootmem and memblock
then we wouldn't have had the whole confusion.

It kinda really bothers me this patchset is expanding the usage
> of the wrong constant with only very far-out plan to fix that. All
> archs converting to nobootmem will take a *long* time, that is, if
> that happens at all. I don't really care about the order of things
> happening but "this is gonna be fixed when everyone moves off
> MAX_NUMNODES" really isn't good enough.
Fair enough though the patchset continue to use the constant
which is already used by few memblock APIs ;-)

If we can fix the __next_free_mem_range() and __next_free_mem_range_rev()
to not use MAX_NUMNODES then we can potentially avoid the wrong
usage of constant.


 \ /
  Last update: 2013-12-04 18:41    [W:0.083 / U:0.420 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site