Enable TCMalloc on Linux by default.

Enable TCMalloc on Linux by default.
This change also reworks the tcmalloc dependency to be added only to chrome and test_shell, instead of base.  This is necessary since otherwise tcmalloc will be double initialized (by both the main executable and dlopen'd shared objects like the npapitestplugin.so).
Add valgrind suppressions.  This are invalid reads on static initialization in the VDSOSupport module.  I haven't investigated it yet, but I suspect they're benign.
BUG=http://crbug.com/28149, http://crbug.com/28385

Committed: http://src.chromium.org/viewvc/chrome?view=rev&revision=33010

[willchan no longer on Chromium, 2009-11-18 21:00:01]

In my experiments, I saw memory for the 12 tab test go up 20% or so (mbelshe
said that this was the case for Windows too) while the perf bots improved by 10%
in general.  Mike said that the DOM benchmark on Windows improved by 40% or so,
but I don't recall seeing that for the Linux perf bot.  I forget though, so I'll
look again this time for any such changes.

[jar (doing other things), 2009-11-20 19:45:36]

LGTM+

Please file a bug to optimize de-commit strategy of TCMalloc on linux as you
land.  I'm pretty sure that some of the Window's tricks need to be applied to
our fork.  You also suggested that you adjust the command line build flag to get
more regular scavenging, so that would be good to mention is such a bug.

Thanks!

[willchan no longer on Chromium, 2009-11-20 20:34:49]

Bug filed.  Just waiting for the trybots to verify the latest round of changes.

On 2009/11/20 19:45:36, jar wrote:
> LGTM+
> 
> Please file a bug to optimize de-commit strategy of TCMalloc on linux as you
> land.  I'm pretty sure that some of the Window's tricks need to be applied to
> our fork.  You also suggested that you adjust the command line build flag to
get
> more regular scavenging, so that would be good to mention is such a bug.
> 
> Thanks!

[willchan no longer on Chromium, 2009-11-23 20:47:44]

There are two changes since you last looked.
(1) I moved the tcmalloc dependency out to the chrome executable instead of the
base library's process_util.  This prevents re-running TCMalloc static
initializers when dynamically loading a shared library that has also been built
with the base library, since TCMalloc's static initializers would then exist in
both.
(2) I had to fork tcmalloc.cc in Linux to get tcmalloc to respect tc_new_mode
for realloc/calloc/memalign/etc since it was only used for malloc/new.  I've
sent a patch to fix this upstream too.

[willchan no longer on Chromium, 2009-11-23 20:47:51]

There are two changes since you last looked.
(1) I moved the tcmalloc dependency out to the chrome executable instead of the
base library's process_util.  This prevents re-running TCMalloc static
initializers when dynamically loading a shared library that has also been built
with the base library, since TCMalloc's static initializers would then exist in
both.
(2) I had to fork tcmalloc.cc in Linux to get tcmalloc to respect tc_new_mode
for realloc/calloc/memalign/etc since it was only used for malloc/new.  I've
sent a patch to fix this upstream too.

[jar (doing other things), 2009-11-23 22:28:41]

http://codereview.chromium.org/399081/diff/8001/8008
File chrome/browser/renderer_host/resource_dispatcher_host.cc (right):

http://codereview.chromium.org/399081/diff/8001/8008#newcode1556
chrome/browser/renderer_host/resource_dispatcher_host.cc:1556: //        
(request->status().status() == URLRequestStatus::CANCELED) ||
Can you justify this change?
Is it an accidental inclusion in this CL?
If the DCHECK is bogus, it should go.  If it is a temporary comment-out, there
should be a TODO()...bug=...

http://codereview.chromium.org/399081/diff/8001/8011
File third_party/tcmalloc/tcmalloc_linux.cc (right):

http://codereview.chromium.org/399081/diff/8001/8011#newcode1
third_party/tcmalloc/tcmalloc_linux.cc:1: // Copyright (c) 2005, Google Inc.
Could you add an inline comment to identify your changes in this file?

[willchan no longer on Chromium, 2009-11-23 22:57:50]

Sorry for adding an unnecessary file.  Thanks for catching that.

With regard to the tcmalloc_linux.cc changes, I've already emailed csilvers and
the changes for calloc/realloc/etc will get accepted shortly.  We're unclear
about memalign() and friends though, but I think they'll get in too.  So
hopefully this fork will be unforked fairly soon.

http://codereview.chromium.org/399081/diff/8001/8011
File third_party/tcmalloc/tcmalloc_linux.cc (right):

http://codereview.chromium.org/399081/diff/8001/8011#newcode362
third_party/tcmalloc/tcmalloc_linux.cc:362: static int tc_new_mode = 0;  // See
tc_set_new_mode().
Moved here from previous location to be early enough to be used.

http://codereview.chromium.org/399081/diff/8001/8011#newcode784
third_party/tcmalloc/tcmalloc_linux.cc:784: inline void* cpp_alloc(size_t size,
bool nothrow);
784-797 are all new.  Declarations and simple wrappers.

http://codereview.chromium.org/399081/diff/8001/8011#newcode849
third_party/tcmalloc/tcmalloc_linux.cc:849: void* result = cpp_or_malloc(size,
false);
s/do_malloc/cpp_or_malloc/g

http://codereview.chromium.org/399081/diff/8001/8011#newcode958
third_party/tcmalloc/tcmalloc_linux.cc:958: new_ptr =
cpp_or_malloc(lower_bound_to_grow, false);
s/do_malloc/cpp_or_malloc/g

http://codereview.chromium.org/399081/diff/8001/8011#newcode962
third_party/tcmalloc/tcmalloc_linux.cc:962: new_ptr = cpp_or_malloc(new_size,
false);
s/do_malloc/cpp_or_malloc/g

http://codereview.chromium.org/399081/diff/8001/8011#newcode1148
third_party/tcmalloc/tcmalloc_linux.cc:1148: inline void* cpp_memalign(size_t
align, size_t size, bool nothrow) {
This function is new.

http://codereview.chromium.org/399081/diff/8001/8011#newcode1227
third_party/tcmalloc/tcmalloc_linux.cc:1227: void* result = cpp_or_malloc(size,
false);
Replaced ternary operation with inline function.

http://codereview.chromium.org/399081/diff/8001/8011#newcode1252
third_party/tcmalloc/tcmalloc_linux.cc:1252: void* result =
cpp_or_malloc(new_size, false);
s/do_malloc/cpp_or_malloc/g

http://codereview.chromium.org/399081/diff/8001/8011#newcode1326
third_party/tcmalloc/tcmalloc_linux.cc:1326: void* result =
cpp_or_memalign(align, size, false);
s/do_memalign/cpp_or_memalign/g

http://codereview.chromium.org/399081/diff/8001/8011#newcode1339
third_party/tcmalloc/tcmalloc_linux.cc:1339: void* result =
cpp_or_memalign(align, size, false);
s/do_memalign/cpp_or_memalign/g

http://codereview.chromium.org/399081/diff/8001/8011#newcode1354
third_party/tcmalloc/tcmalloc_linux.cc:1354: void* result =
cpp_or_memalign(pagesize, size, false);
s/do_memalign/cpp_or_memalign/g

http://codereview.chromium.org/399081/diff/8001/8011#newcode1366
third_party/tcmalloc/tcmalloc_linux.cc:1366: void* result =
cpp_or_memalign(pagesize, size, false);
s/do_memalign/cpp_or_memalign/g

[willchan no longer on Chromium, 2009-11-24 21:14:06]

ping

[jar (doing other things), 2009-11-25 00:09:27]

LGTM