PartitionAlloc: Make in-place realloc() less likely to happen

PartitionAlloc: Make in-place realloc() less likely to happen

On direct mapped allocations, we support downsizing realloc() in-place.
As part of trying to figure out what causes apparent OOM crashes, do this
only in much more restricted circumstances:

 - Only down to 80 % of the size of the original mapping.
 - Only down to sizes larger than the maximum bucketed allocation size.

BUG=368406
Committed: https://src.chromium.org/viewvc/blink?view=rev&revision=172932

[Tom Sepez, 2014-04-29 20:27:01]

LGTM.

https://codereview.chromium.org/254103003/diff/1/Source/wtf/PartitionAllocTes...
File Source/wtf/PartitionAllocTest.cpp (right):

https://codereview.chromium.org/254103003/diff/1/Source/wtf/PartitionAllocTes...
Source/wtf/PartitionAllocTest.cpp:701: EXPECT_EQ(actualSize - 8 *
WTF::kSystemPageSize, partitionAllocGetSize(ptr2));
It would be nice to have a test case that shows, say, that going from say 2MB to
16B is not in-place.  Do we have this?

[Jens Widell, 2014-04-29 20:29:48]

On 2014/04/29 20:27:01, Tom Sepez wrote:
> LGTM.
> 
>
https://codereview.chromium.org/254103003/diff/1/Source/wtf/PartitionAllocTes...
> File Source/wtf/PartitionAllocTest.cpp (right):
> 
>
https://codereview.chromium.org/254103003/diff/1/Source/wtf/PartitionAllocTes...
> Source/wtf/PartitionAllocTest.cpp:701: EXPECT_EQ(actualSize - 8 *
> WTF::kSystemPageSize, partitionAllocGetSize(ptr2));
> It would be nice to have a test case that shows, say, that going from say 2MB
to
> 16B is not in-place.  Do we have this?

We don't, but the test I modified could easily be extended to do that. I'll
extend it.

[Jens Widell, 2014-04-29 20:35:52]

On 2014/04/29 20:29:48, Jens Lindström wrote:
> On 2014/04/29 20:27:01, Tom Sepez wrote:
> > LGTM.
> > 
> >
>
https://codereview.chromium.org/254103003/diff/1/Source/wtf/PartitionAllocTes...
> > File Source/wtf/PartitionAllocTest.cpp (right):
> > 
> >
>
https://codereview.chromium.org/254103003/diff/1/Source/wtf/PartitionAllocTes...
> > Source/wtf/PartitionAllocTest.cpp:701: EXPECT_EQ(actualSize - 8 *
> > WTF::kSystemPageSize, partitionAllocGetSize(ptr2));
> > It would be nice to have a test case that shows, say, that going from say
2MB
> to
> > 16B is not in-place.  Do we have this?
> 
> We don't, but the test I modified could easily be extended to do that. I'll
> extend it.

Patch updated.

[Tom Sepez, 2014-04-29 20:38:29]

> Patch updated.
Excellent.  Unfortunately, I'm not in wtf/OWNERS so we'll need eric's approval
before landing. (cevans is out for some time).

[Jens Widell, 2014-04-29 20:38:38]

The CQ bit was checked by jl@opera.com

[commit-bot: I haz the power, 2014-04-29 20:39:49]

CQ is trying da patch. Follow status at
https://chromium-status.appspot.com/cq/jl@opera.com/254103003/20001

[Jens Widell, 2014-04-29 20:43:21]

On 2014/04/29 20:38:29, Tom Sepez wrote:
> > Patch updated.
> Excellent.  Unfortunately, I'm not in wtf/OWNERS so we'll need eric's approval
> before landing. (cevans is out for some time).

Right. It's getting late over here, so I'm signing out. If an owner approves,
please feel free to land it for me. Or I'll do it myself in the morning.

[eseidel, 2014-04-29 20:45:35]

lgtm

Hopefully this helps.

[commit-bot: I haz the power, 2014-04-30 04:56:26]

Change committed as 172932