Setup R10/EDX to be valid Oops before stub calls

Setup R10/EDX to be valid Oops before stub calls so when the debug stub saves them to the stack the GC doesn't try to deference out of the heap.

BUG=http://dartbug.com/19331
R=asiva@google.com, iposva@google.com

Committed: https://code.google.com/p/dart/source/detail?r=37297

[Ivan Posva, 2014-06-12 22:55:09]

LGTM with Siva's comment.

-Ivan

https://codereview.chromium.org/335443002/diff/160001/runtime/vm/flow_graph_c...
File runtime/vm/flow_graph_compiler_mips.cc (right):

https://codereview.chromium.org/335443002/diff/160001/runtime/vm/flow_graph_c...
runtime/vm/flow_graph_compiler_mips.cc:1352: __ LoadImmediate(S4, 0);  //
Garbage that looks like an object.
1

Maybe you should define a constant for this?

[siva, 2014-06-12 23:11:34]

lgtm

https://codereview.chromium.org/335443002/diff/160001/runtime/vm/debugger_api...
File runtime/vm/debugger_api_impl_test.cc (right):

https://codereview.chromium.org/335443002/diff/160001/runtime/vm/debugger_api...
runtime/vm/debugger_api_impl_test.cc:989: OS::Print("Hit line %" Pd "\n",
line_number);
Is the print necessary?

You could just do
intptr_t line_number = -1;
res = Dart_Activation......
EXPECT_NE(line_number, -1);

https://codereview.chromium.org/335443002/diff/160001/runtime/vm/debugger_api...
runtime/vm/debugger_api_impl_test.cc:1011:
TEST_CASE(Debug_BreakpointStubPatching) {
Maybe you should add a comment here that if somebody changes this
script here they have to also account for the line numbers in
breakpoint_lines array below.

[rmacnak, 2014-06-12 23:50:15]

https://codereview.chromium.org/335443002/diff/160001/runtime/vm/debugger_api...
File runtime/vm/debugger_api_impl_test.cc (right):

https://codereview.chromium.org/335443002/diff/160001/runtime/vm/debugger_api...
runtime/vm/debugger_api_impl_test.cc:989: OS::Print("Hit line %" Pd "\n",
line_number);
On 2014/06/12 23:11:34, siva wrote:
> Is the print necessary?
> 
> You could just do
> intptr_t line_number = -1;
> res = Dart_Activation......
> EXPECT_NE(line_number, -1);

This was for debugging why I wasn't hitting the expected number of breakpoints
developing the test, to see which were missing. Adding expectation that we get a
value and hiding the print under verbose.

https://codereview.chromium.org/335443002/diff/160001/runtime/vm/debugger_api...
runtime/vm/debugger_api_impl_test.cc:1011:
TEST_CASE(Debug_BreakpointStubPatching) {
On 2014/06/12 23:11:34, siva wrote:
> Maybe you should add a comment here that if somebody changes this
> script here they have to also account for the line numbers in
> breakpoint_lines array below.

Added.

https://codereview.chromium.org/335443002/diff/160001/runtime/vm/flow_graph_c...
File runtime/vm/flow_graph_compiler_mips.cc (right):

https://codereview.chromium.org/335443002/diff/160001/runtime/vm/flow_graph_c...
runtime/vm/flow_graph_compiler_mips.cc:1352: __ LoadImmediate(S4, 0);  //
Garbage that looks like an object.
On 2014/06/12 22:55:09, Ivan Posva wrote:
> 1
> 
> Maybe you should define a constant for this?

Changed to kInvalidObjectPointer.

[rmacnak, 2014-06-12 23:51:03]

Committed patchset #11 manually as r37297 (presubmit successful).

[Florian Schneider, 2014-06-20 12:48:54]

dbc:

https://codereview.chromium.org/335443002/diff/200001/runtime/vm/flow_graph_c...
File runtime/vm/flow_graph_compiler_ia32.cc (right):

https://codereview.chromium.org/335443002/diff/200001/runtime/vm/flow_graph_c...
runtime/vm/flow_graph_compiler_ia32.cc:1185: __ movl(EDX, Immediate(0));
This is bad for unoptimized code size.

[Ivan Posva, 2014-06-20 14:03:31]

https://codereview.chromium.org/335443002/diff/200001/runtime/vm/flow_graph_c...
File runtime/vm/flow_graph_compiler_ia32.cc (right):

https://codereview.chromium.org/335443002/diff/200001/runtime/vm/flow_graph_c...
runtime/vm/flow_graph_compiler_ia32.cc:1185: __ movl(EDX, Immediate(0));
On 2014/06/20 12:48:54, Florian Schneider wrote:
> This is bad for unoptimized code size.

We were fully aware of the code size issue when we implemented this fix. The
real fix will be to have different stubs, based on the call patterns and the
corresponding live registers.