Make sure we always have room for patching the reloc info during lazy deoptim...

Make sure we always have room for patching the reloc info during lazy deoptimization (fixes issue 1156).

Before we could have calls to builtins that would not be in the
relocation info since this used a register as target. Whenever we have
this case (from lithium codegen) we now emit a comment in the reloc
info.


Committed: http://code.google.com/p/v8/source/detail?r=6795

[Kevin Millikin (Chromium), 2011-02-15 12:19:14]

I think the approach will work.

We also need to do the same for MacroAssembler::InvokeFunction, and any other
place that calls through a register.

We might want a comment in RelocWriter::Write (or wherever we emit the code
comments) to say that we rely on using the long encoding for comments (so we can
patch it with anything).

It would be nice if we could somehow assert for each safepoint that the reloc
info had changed by "delta" between before the call and after the call.

http://codereview.chromium.org/6499015/diff/2002/src/ia32/assembler-ia32.h
File src/ia32/assembler-ia32.h (right):

http://codereview.chromium.org/6499015/diff/2002/src/ia32/assembler-ia32.h#ne...
src/ia32/assembler-ia32.h:960: // Use --code-comments to enable.
We should probably update this comment to mention that force overrides
--nocode-comments.

http://codereview.chromium.org/6499015/diff/2002/src/ia32/lithium-codegen-ia3...
File src/ia32/lithium-codegen-ia32.cc (right):

http://codereview.chromium.org/6499015/diff/2002/src/ia32/lithium-codegen-ia3...
src/ia32/lithium-codegen-ia32.cc:58: codegen_->masm()->RecordComment(NULL,
true);
I guess printing comments works with NULL strings?

We might want to put some distinct static constant string so we can recognize
these when dumping code and reloc info.  (I suggest "Kick me!").

[Rico, 2011-02-15 13:33:00]

>> It would be nice if we could somehow assert for each safepoint that the reloc
>> info had changed by "delta" between before the call and after the call.

We are not sure how much space is actually needed for a call, this depends on
usage of pc-jump instructions, so a call can be 1, 3, 4, 5 or 6 bytes long. We
are, however, guaranteed that for normal calls that have reloc info we do
actually have room for this since the original call will also be preceded by a
pc jump (since the pc delta between old and new calls will be the same)

http://codereview.chromium.org/6499015/diff/2002/src/ia32/assembler-ia32.h
File src/ia32/assembler-ia32.h (right):

http://codereview.chromium.org/6499015/diff/2002/src/ia32/assembler-ia32.h#ne...
src/ia32/assembler-ia32.h:960: // Use --code-comments to enable.
On 2011/02/15 12:19:15, Kevin Millikin wrote:
> We should probably update this comment to mention that force overrides
> --nocode-comments.

Done.

http://codereview.chromium.org/6499015/diff/2002/src/ia32/lithium-codegen-ia3...
File src/ia32/lithium-codegen-ia32.cc (right):

http://codereview.chromium.org/6499015/diff/2002/src/ia32/lithium-codegen-ia3...
src/ia32/lithium-codegen-ia32.cc:58: codegen_->masm()->RecordComment(NULL,
true);
On 2011/02/15 12:19:15, Kevin Millikin wrote:
> I guess printing comments works with NULL strings?
> 
> We might want to put some distinct static constant string so we can recognize
> these when dumping code and reloc info.  (I suggest "Kick me!").
Printing with null works, but I changed this to a static string in the assembler
class

[Rico, 2011-02-15 13:34:53]

Comments addressed, please have another look. Added Soeren and Mads as reviewers
as well, we should be very sure this is right since this needs to be landed on
the 3.0 branch as well.

[Søren Thygesen Gjesse, 2011-02-15 13:42:44]

LGTM

[Rico, 2011-02-15 14:16:04]

Added src/assembler.cc to the cl (initialization of the padding string);