Add commands Jump, Call and Return...

Add commands Jump, Call and Return

This is the minimal change version.

There were 3 possible solution:

#1) Remove these commands

#2) Change the command buffers so the client side
can not see the "get" pointer and therefore force
the client side to use tokens to say in sync.

It's the cleaner solution. It lets you put commands
in any shared memory thereby getting rid of the ring
buffer commands. 

But, it required major changes. After going down
that path for an entire day I punted for...

#3) Implement the commands assuming there is only
one buffer the commands can be in (what is currently
called the ring_buffer). With the jump commands
available it no longer has to be used as a ring
buffer. A Pepper or NaCl app can use it however
they want but, unlike solution #2 they only get
the one buffer.

TEST=none
BUG=none


Committed: http://src.chromium.org/viewvc/chrome?view=rev&revision=37446

[greggman, 2010-01-27 08:46:53]

http://codereview.chromium.org/555129/diff/1/3
File gpu/command_buffer/service/cmd_buffer_engine.h (right):

http://codereview.chromium.org/555129/diff/1/3#newcode16
gpu/command_buffer/service/cmd_buffer_engine.h:16: class CommandBufferEngine {
I really don't quite understand the point of this class. There's hardly anything
in it and the name nor the comment above make it clear what this is for.

[apatrick_chromium, 2010-01-27 19:20:56]

http://codereview.chromium.org/555129/diff/19/1023
File gpu/command_buffer/service/common_decoder_unittest.cc (right):

http://codereview.chromium.org/555129/diff/19/1023#newcode237
gpu/command_buffer/service/common_decoder_unittest.cc:237:
cmd.Init(MockCommandBufferEngine::kValidOffset);
Should we enforce a max stack depth? I think a malicious client could
potentially make the GPU process crash by recursing until the stack overflows.

http://codereview.chromium.org/555129/diff/19/1031
File gpu/command_buffer/service/gpu_processor.cc (right):

http://codereview.chromium.org/555129/diff/19/1031#newcode84
gpu/command_buffer/service/gpu_processor.cc:84: return parser_->set_get(offset);
I need to check when I get to my desk but I can't remember whether this should
be applied to the command buffer or the parser.

[apatrick_chromium, 2010-01-27 19:38:45]

http://codereview.chromium.org/555129/diff/19/1031
File gpu/command_buffer/service/gpu_processor.cc (right):

http://codereview.chromium.org/555129/diff/19/1031#newcode84
gpu/command_buffer/service/gpu_processor.cc:84: return parser_->set_get(offset);
This is kind of gross but I think it needs to set the "get" offset on both the
command buffer and the parser.

[greggman, 2010-01-28 10:15:33]

http://codereview.chromium.org/555129/diff/19/1023
File gpu/command_buffer/service/common_decoder_unittest.cc (right):

http://codereview.chromium.org/555129/diff/19/1023#newcode237
gpu/command_buffer/service/common_decoder_unittest.cc:237:
cmd.Init(MockCommandBufferEngine::kValidOffset);
On 2010/01/27 19:20:57, apatrick_chromium wrote:
> Should we enforce a max stack depth? I think a malicious client could
> potentially make the GPU process crash by recursing until the stack overflows.

Done.

http://codereview.chromium.org/555129/diff/19/1031
File gpu/command_buffer/service/gpu_processor.cc (right):

http://codereview.chromium.org/555129/diff/19/1031#newcode84
gpu/command_buffer/service/gpu_processor.cc:84: return parser_->set_get(offset);
On 2010/01/27 19:38:46, apatrick_chromium wrote:
> This is kind of gross but I think it needs to set the "get" offset on both the
> command buffer and the parser.

Done.

[apatrick, 2010-01-28 19:14:05]

lgtm