Initialize blink only when needed in utility processes.

Initialize blink only when needed in utility processes.

Currently, in a utility process containing a v8 proxy resolver, both
blink and the proxy resolver attempt to use v8 in incompatible ways,
causing a crash on shutdown. This cl changes the utility process to
only initialize blink when needed.

On Windows, initializing blink with v8 before enabling the sandbox
calls RtlGenRandom(), which leaves it available from within the sandbox.
To avoid breaking base::RandBytes(), this cl also adds a call to
base::RandBytes before enabling the sandbox.

BUG=506439
Committed: https://crrev.com/6a2e8d4d42cd4c15755d338313b94cdf308219fe
Cr-Commit-Position: refs/heads/master@{#340835}

[Sam McNally, 2015-07-24 07:31:03]

Patchset #2 (id:20001) has been deleted

[Sam McNally, 2015-07-27 04:01:20]

Patchset #1 (id:1) has been deleted

[Sam McNally, 2015-07-27 04:05:15]

sammc@chromium.org changed reviewers:
+ jochen@chromium.org

[jam, 2015-07-27 05:08:42]

jam@chromium.org changed reviewers:
+ jam@chromium.org

[jam, 2015-07-27 05:08:43]

this probably has legacy reasons (i.e. first users of utility process all used
blink). but perhaps we shouldn't initialize it anymore,a nd only users that need
it can initialize it (in whatever mode they desire)

[Sam McNally, 2015-07-27 05:52:31]

Patchset #2 (id:60001) has been deleted

[Sam McNally, 2015-07-27 05:56:32]

Patchset #2 (id:80001) has been deleted

[Sam McNally, 2015-07-27 06:31:57]

Patchset #2 (id:100001) has been deleted

[Sam McNally, 2015-07-27 07:22:22]

Patchset #3 (id:140001) has been deleted

[Sam McNally, 2015-07-27 08:03:50]

On 2015/07/27 05:08:43, jam wrote:
> this probably has legacy reasons (i.e. first users of utility process all used
> blink). but perhaps we shouldn't initialize it anymore,a nd only users that
need
> it can initialize it (in whatever mode they desire)

Done.

[jochen (gone - plz use gerrit), 2015-07-27 08:58:51]

On 2015/07/27 at 08:03:50, sammc wrote:
> On 2015/07/27 05:08:43, jam wrote:
> > this probably has legacy reasons (i.e. first users of utility process all
used
> > blink). but perhaps we shouldn't initialize it anymore,a nd only users that
need
> > it can initialize it (in whatever mode they desire)
> 
> Done.

are there actually any users that need this? maybe instead of introducing
EnsureBlinkInitialized() without any user that actually needs v8, just always
initialize blink without v8

[Sam McNally, 2015-07-27 10:39:04]

On 2015/07/27 08:58:51, jochen wrote:
> On 2015/07/27 at 08:03:50, sammc wrote:
> > On 2015/07/27 05:08:43, jam wrote:
> > > this probably has legacy reasons (i.e. first users of utility process all
> used
> > > blink). but perhaps we shouldn't initialize it anymore,a nd only users
that
> need
> > > it can initialize it (in whatever mode they desire)
> > 
> > Done.
> 
> are there actually any users that need this? maybe instead of introducing
> EnsureBlinkInitialized() without any user that actually needs v8, just always
> initialize blink without v8

Changed EnsureBlinkInitialized() to always initialize without v8. It turns out
that not initializing blink reduces proxy resolver utility process memory usage
by 1mb on Linux. I don't think anything needs fine-grained control over this,
but most users don't need blink at all.

[jochen (gone - plz use gerrit), 2015-07-27 11:35:21]

https://codereview.chromium.org/1251823002/diff/180001/content/public/utility...
File content/public/utility/utility_thread.h (right):

https://codereview.chromium.org/1251823002/diff/180001/content/public/utility...
content/public/utility/utility_thread.h:26: virtual void
EnsureBlinkInitialized() = 0;
why is this API needed? why not just call initializeWithoutV8() where we used to
call initialize()?

[jam, 2015-07-27 15:14:49]

https://codereview.chromium.org/1251823002/diff/180001/content/public/utility...
File content/public/utility/utility_thread.h (right):

https://codereview.chromium.org/1251823002/diff/180001/content/public/utility...
content/public/utility/utility_thread.h:26: virtual void
EnsureBlinkInitialized() = 0;
On 2015/07/27 11:35:21, jochen wrote:
> why is this API needed? why not just call initializeWithoutV8() where we used
to
> call initialize()?

(see my first message)

to expand on it also, utility process is now for hosting many things that don't
use blink (i.e. mojo apps). so we shouldn't do the work of initializing blink if
we don't need to.

[jochen (gone - plz use gerrit), 2015-07-27 15:19:00]

ok, makes sense, lgtm

[Sam McNally, 2015-07-28 00:32:12]

sammc@chromium.org changed reviewers:
+ asargent@chromium.org

[Sam McNally, 2015-07-28 00:32:13]

+asargent for extensions/utility/utility_handler.cc

[asargent_no_longer_on_chrome, 2015-07-28 17:59:55]

lgtm

[Sam McNally, 2015-07-28 23:54:24]

The CQ bit was checked by sammc@chromium.org

[commit-bot: I haz the power, 2015-07-28 23:54:58]

CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/1251823002/180001
View timeline at
 https://chromium-cq-status.appspot.com/patch-timeline/1251823002/180001

[commit-bot: I haz the power, 2015-07-29 01:52:30]

Committed patchset #4 (id:180001)

[commit-bot: I haz the power, 2015-07-29 01:53:08]

Patchset 4 (id:??) landed as
https://crrev.com/6a2e8d4d42cd4c15755d338313b94cdf308219fe
Cr-Commit-Position: refs/heads/master@{#340835}