Patch: improve huge page support

[Andrew Haley]

On 05/09/2011 02:20 PM, Dr Andrew John Hughes wrote:
> On 13:00 Thu 05 May     , Andrew Haley wrote:
>> On 05/04/2011 04:53 PM, Dr Andrew John Hughes wrote:
>>> On 15:49 Wed 04 May     , Andrew Haley wrote:
>>>> My big patch to improve huge page support in now in HotSpot upstream.
>>>> This is a backport.
>>>>
>>>> The question is: to what branches should we apply this?  It makes the
>>>> use of huge pages much, much easier.  You no longer need root access
>>>> for anything except setting the size of the huge page pool, and the
>>>> huge pages aren't locked in memory, so it's a big improvement for
>>>> usability.  If the new huge page support doesn't work in the kernel,
>>>> it falls back to the old SysV-style huge page allocator.
>>>>
>>>> Thoughts?
>>>>
>>>
>>> I notice this doesn't actually enable the patch in the Makefile.am.  Have
>>> you built this on any version of IcedTea yet?
>>
>> Yes, but it's missing from the email, that's all.
> 
> So which version?

Trunk; I always begin with trunk.

>>> I would prefer that this was two separate patches, corresponding to each
>>> upstream changeset.  It makes it easier to just delete the patch file
>>> when it gets backported upstream in OpenJDK6 (and, prior to that, makes
>>> it easier to see what we need to request for backport).
>>
>> The patches aren't separate, and shouldn't be applied separately.  There
>> are no circumstances in which it's appropriate to apply just one, IMO.
> 
> They are separate changesets so should be separate patches.  It's very hard
> to tell what's in the current patch as it doesn't correspond directly to
> an OpenJDK7 changeset.  If one was to appear in OpenJDK6 and not the other,
> it would be difficult to work out which parts were from which changeset.

'Mkay, it's for changeset tracking, then.  Fair enough.

>>> As to branches, things could get tricky, as each of the currently supported
>>> branches has two HotSpot variants as follows:
>>>
>>> 1.8: hs14 & hs16
>>> 1.9: hs17 (default) & hs19
>>> 1.10: hs19 (default) & hs20
>>> 1.11 (HEAD): hs19 & hs20 (default)
>>>
>>> So that's potentially five versions of HotSpot to patch (hs14, hs16, hs17, hs19, hs20)
>>> but it may be that this code has changed little and the same patch can be applied to
>>> several versions.
>>
>> I suspect so, but I had to do a fair bit of work on the backport,
>> as this file has some churn.
> 
> Well, merging the two changesets makes this harder as it's now difficult to tell which
> bits of the patch came from which changeset, and which bits came from neither but are
> a result of backporting.

Sure, but that's going to be difficult no matter what.  There isn't
any patch that will apply that corresponds exactly to the upstream
changesets.

Andrew.