Adding support for reporting addresses of JIT compiled code to OProfile

Adding support for reporting addresses of JIT compiled code to OProfile.

Please be warned that current gHardy versions have OProfile 0.9.3
which doesn't have JIT API. You need to install OProfile 0.9.4
with a 32-bit version of opagent library. Instructions are on the
internal Wiki page: http://wiki/Main/V8UsingOProfile

Committed: http://code.google.com/p/v8/source/detail?r=1426

[Kasper Lund, 2009-03-05 10:20:53]

LGTM. Biggest concern is the atexit call from within V8.

http://codereview.chromium.org/39179/diff/1/8
File src/objects.h (right):

http://codereview.chromium.org/39179/diff/1/8#newcode2263
Line 2263: int ObjectSize() {
This name is probably too generic since it doesn't include the entire object
(reloc info, scope info, and padding does not count). It's not really the entire
size of the object.

http://codereview.chromium.org/39179/diff/1/9
File src/oprofile-agent.cc (right):

http://codereview.chromium.org/39179/diff/1/9#newcode44
Line 44: // Disable code moving by GC
Terminate comment with .

http://codereview.chromium.org/39179/diff/1/10
File src/oprofile-agent.h (right):

http://codereview.chromium.org/39179/diff/1/10#newcode60
Line 60: // Size of the buffer that is used for composing code areas names
Terminate comment with .

http://codereview.chromium.org/39179/diff/1/11
File src/v8.cc (right):

http://codereview.chromium.org/39179/diff/1/11#newcode93
Line 93: atexit(V8::TearDown);
I'm not sure this is a good idea in the context of Chromium. Maybe you should
just make the shell sample and d8 do explicit tear down or use the atexit from
there? It really should be up to the embedder to invoke V8::TearDown.

[Mikhail Naganov, 2009-03-05 10:51:03]

http://codereview.chromium.org/39179/diff/1/8
File src/objects.h (right):

http://codereview.chromium.org/39179/diff/1/8#newcode2263
Line 2263: int ObjectSize() {
On 2009/03/05 10:20:53, Kasper Lund wrote:
> This name is probably too generic since it doesn't include the entire object
> (reloc info, scope info, and padding does not count). It's not really the
entire
> size of the object.

Renamed to "ExecutableSize", since it is the size of the executable area of the
object.

http://codereview.chromium.org/39179/diff/1/9
File src/oprofile-agent.cc (right):

http://codereview.chromium.org/39179/diff/1/9#newcode44
Line 44: // Disable code moving by GC
On 2009/03/05 10:20:53, Kasper Lund wrote:
> Terminate comment with .

Done.

http://codereview.chromium.org/39179/diff/1/10
File src/oprofile-agent.h (right):

http://codereview.chromium.org/39179/diff/1/10#newcode60
Line 60: // Size of the buffer that is used for composing code areas names
On 2009/03/05 10:20:53, Kasper Lund wrote:
> Terminate comment with .

Done.

http://codereview.chromium.org/39179/diff/1/11
File src/v8.cc (right):

http://codereview.chromium.org/39179/diff/1/11#newcode93
Line 93: atexit(V8::TearDown);
On 2009/03/05 10:20:53, Kasper Lund wrote:
> I'm not sure this is a good idea in the context of Chromium. Maybe you should
> just make the shell sample and d8 do explicit tear down or use the atexit from
> there? It really should be up to the embedder to invoke V8::TearDown.

OK. The problem is that I don't see any shutdown-related methods in the public
header file (include/v8.h).

I think it's not a big problem for OProfile to have an unclosed handle.
Application can crash, for example.

But it's an interesting issue though that shutting down V8 isn't put explicitly
anywhere. Do you think we need to track this?