Fix lazy deoptimization in the presence of exceptions

Fix lazy deoptimization in the presence of exceptions

Catch block entries need to be considered when doing lazy deoptimization.

In addition to the return, also patch all catch entry blocks in a function scheduled for lazy deoptimization.

Also, move restoring the pool pointer to the jump-to-handler stub to simplify patching the catch entry.

BUG=#27419
R=rmacnak@google.com

Committed: https://github.com/dart-lang/sdk/commit/7ca3da1a3aabcc52cef226380870a67fd0a5f98d

[Florian Schneider, 2016-09-22 23:54:16]

Description was changed from

==========
Fix lazy deoptimization in the presence of exceptions

BUG=
==========

to

==========
Fix lazy deoptimization in the presence of exceptions

Catch block entries need to be considered when doing lazy deoptimization.

In addition to the return, also patch all catch entry blocks in a function
scheduled for lazy deoptimization.

Also, move restoring the pool pointer to the jump-to-handler stub to simplify
patching the catch entry.

BUG=
==========

[Florian Schneider, 2016-09-23 18:18:25]

Description was changed from

==========
Fix lazy deoptimization in the presence of exceptions

Catch block entries need to be considered when doing lazy deoptimization.

In addition to the return, also patch all catch entry blocks in a function
scheduled for lazy deoptimization.

Also, move restoring the pool pointer to the jump-to-handler stub to simplify
patching the catch entry.

BUG=
==========

to

==========
Fix lazy deoptimization in the presence of exceptions

Catch block entries need to be considered when doing lazy deoptimization.

In addition to the return, also patch all catch entry blocks in a function
scheduled for lazy deoptimization.

Also, move restoring the pool pointer to the jump-to-handler stub to simplify
patching the catch entry.

BUG=#27419
==========

[Florian Schneider, 2016-09-23 19:53:33]

fschneider@google.com changed reviewers:
+ rmacnak@google.com

[Florian Schneider, 2016-09-23 19:53:34]

https://codereview.chromium.org/2357343003/diff/40001/runtime/vm/flow_graph_c...
File runtime/vm/flow_graph_compiler.h (left):

https://codereview.chromium.org/2357343003/diff/40001/runtime/vm/flow_graph_c...
runtime/vm/flow_graph_compiler.h:539: void AddDeoptIndexAtCall(intptr_t
deopt_id, TokenPosition token_pos);
token_pos was unused here.

https://codereview.chromium.org/2357343003/diff/40001/runtime/vm/intermediate...
File runtime/vm/intermediate_language_ia32.cc (right):

https://codereview.chromium.org/2357343003/diff/40001/runtime/vm/intermediate...
runtime/vm/intermediate_language_ia32.cc:2552: __
nop(CallPattern::pattern_length_in_bytes());
Needed to ensure, patching a return-address does not overlap with a catch entry
block (also on x64). On ARM/MIPS/ARM64, there is always space after the call
(one instructions dropping the arguments)

https://codereview.chromium.org/2357343003/diff/40001/runtime/vm/intermediate...
File runtime/vm/intermediate_language_x64.cc (left):

https://codereview.chromium.org/2357343003/diff/40001/runtime/vm/intermediate...
runtime/vm/intermediate_language_x64.cc:2575: __ LoadPoolPointer(PP);
This now happens in the JumpToExceptionHandler stub.

[rmacnak, 2016-09-26 20:21:48]

LGTM

https://chromiumcodereview.appspot.com/2357343003/diff/40001/runtime/vm/inter...
File runtime/vm/intermediate_language_ia32.cc (right):

https://chromiumcodereview.appspot.com/2357343003/diff/40001/runtime/vm/inter...
runtime/vm/intermediate_language_ia32.cc:2552: __
nop(CallPattern::pattern_length_in_bytes());
On 2016/09/23 19:53:34, Florian Schneider wrote:
> Needed to ensure, patching a return-address does not overlap with a catch
entry
> block (also on x64). On ARM/MIPS/ARM64, there is always space after the call
> (one instructions dropping the arguments)

Just noting that space might disappear on the RISC architectures with a register
calling convention.

[Florian Schneider, 2016-09-26 23:05:28]

Ptal

https://codereview.chromium.org/2357343003/diff/40001/runtime/vm/intermediate...
File runtime/vm/intermediate_language_ia32.cc (right):

https://codereview.chromium.org/2357343003/diff/40001/runtime/vm/intermediate...
runtime/vm/intermediate_language_ia32.cc:2552: __
nop(CallPattern::pattern_length_in_bytes());
On 2016/09/26 20:21:48, rmacnak wrote:
> On 2016/09/23 19:53:34, Florian Schneider wrote:
> > Needed to ensure, patching a return-address does not overlap with a catch
> entry
> > block (also on x64). On ARM/MIPS/ARM64, there is always space after the call
> > (one instructions dropping the arguments)
> 
> Just noting that space might disappear on the RISC architectures with a
register
> calling convention.

Actually, I'll add space on ARM/MIPS archs as well. The call pattern can be up
to 5 instructions which may overlap with the entry.

Also, it only is needed in JIT mode, and in optimized code.

[rmacnak, 2016-09-26 23:21:13]

lgtm

[Florian Schneider, 2016-09-26 23:24:28]

Description was changed from

==========
Fix lazy deoptimization in the presence of exceptions

Catch block entries need to be considered when doing lazy deoptimization.

In addition to the return, also patch all catch entry blocks in a function
scheduled for lazy deoptimization.

Also, move restoring the pool pointer to the jump-to-handler stub to simplify
patching the catch entry.

BUG=#27419
==========

to

==========
Fix lazy deoptimization in the presence of exceptions

Catch block entries need to be considered when doing lazy deoptimization.

In addition to the return, also patch all catch entry blocks in a function
scheduled for lazy deoptimization.

Also, move restoring the pool pointer to the jump-to-handler stub to simplify
patching the catch entry.

BUG=#27419
R=rmacnak@google.com

Committed:
https://github.com/dart-lang/sdk/commit/7ca3da1a3aabcc52cef226380870a67fd0a5f98d
==========

[Florian Schneider, 2016-09-26 23:24:29]

Committed patchset #4 (id:60001) manually as
7ca3da1a3aabcc52cef226380870a67fd0a5f98d (presubmit successful).