ARM: Implement PatchStackCheckCodeAt and RevertStackCheckCode.

ARM: Implement PatchStackCheckCodeAt and RevertStackCheckCode. 

Remove a failing test expectation from mjsunit.status.

[Søren Thygesen Gjesse, 2011-02-04 13:54:27]

LGTM with comments.

http://codereview.chromium.org/6410029/diff/2003/src/arm/deoptimizer-arm.cc
File src/arm/deoptimizer-arm.cc (right):

http://codereview.chromium.org/6410029/diff/2003/src/arm/deoptimizer-arm.cc#n...
src/arm/deoptimizer-arm.cc:129: //  2a ?? ?? ??       bcs ok
?? ?? ?? -> 00 00 01, the restoring puts in 1 here.

http://codereview.chromium.org/6410029/diff/2003/src/arm/deoptimizer-arm.cc#n...
src/arm/deoptimizer-arm.cc:132: ASSERT(Memory::uint32_at(pc_after - 4) ==
0xe12fff3c);
0xe12fff3c -> al | B24 | B21 | 15*B16 | 15*B12 | 15*B8 | BLX | ip.code()

http://codereview.chromium.org/6410029/diff/2003/src/arm/deoptimizer-arm.cc#n...
src/arm/deoptimizer-arm.cc:132: ASSERT(Memory::uint32_at(pc_after - 4) ==
0xe12fff3c);
4 -> kInstrSize

http://codereview.chromium.org/6410029/diff/2003/src/arm/deoptimizer-arm.cc#n...
src/arm/deoptimizer-arm.cc:133: ASSERT(Memory::uint8_at(pc_after - 5) == 0xe5);
Use Assembler::IsLdrPcImmediateOffset() to check this. Maybe add
Assembler::IsLdrPcImmediateOffset() which also takes a register and checks that
it is ldr ip, [...

http://codereview.chromium.org/6410029/diff/2003/src/arm/deoptimizer-arm.cc#n...
src/arm/deoptimizer-arm.cc:145: Memory::uint32_at(pc_after - 12) = 0xe1a00000;
I think you should be able to use the code patcher here:

CodePatcher patcher(pc_after - 3 * kInstrSize, 1)
patcher.masm()->nop();

if not

0xe1a00000 -> al | MOV

http://codereview.chromium.org/6410029/diff/2003/src/arm/deoptimizer-arm.cc#n...
src/arm/deoptimizer-arm.cc:167: Memory::uint32_at(pc_after - 12) = 0x2a000001;
Use CodePatcher if possible, or at least:

0x2a000001 -> cs | B27 | B25 | 1

http://codereview.chromium.org/6410029/diff/2003/src/deoptimizer.cc
File src/deoptimizer.cc (right):

http://codereview.chromium.org/6410029/diff/2003/src/deoptimizer.cc#newcode837
src/deoptimizer.cc:837: uint32_t length = last_site - first_site + patch_size();
If you use the CodePatcher the i-cache flushing is automatic.

http://codereview.chromium.org/6410029/diff/2003/src/deoptimizer.cc#newcode867
src/deoptimizer.cc:867: uint32_t length = last_site - first_site + patch_size();
Ditto.

[Karl Klose, 2011-02-04 18:06:49]

http://codereview.chromium.org/6410029/diff/2003/src/arm/deoptimizer-arm.cc
File src/arm/deoptimizer-arm.cc (right):

http://codereview.chromium.org/6410029/diff/2003/src/arm/deoptimizer-arm.cc#n...
src/arm/deoptimizer-arm.cc:129: //  2a ?? ?? ??       bcs ok
On 2011/02/04 13:54:27, Søren Gjesse wrote:
> ?? ?? ?? -> 00 00 01, the restoring puts in 1 here.

Done.

http://codereview.chromium.org/6410029/diff/2003/src/arm/deoptimizer-arm.cc#n...
src/arm/deoptimizer-arm.cc:132: ASSERT(Memory::uint32_at(pc_after - 4) ==
0xe12fff3c);
On 2011/02/04 13:54:27, Søren Gjesse wrote:
> 0xe12fff3c -> al | B24 | B21 | 15*B16 | 15*B12 | 15*B8 | BLX | ip.code()

Done.

http://codereview.chromium.org/6410029/diff/2003/src/arm/deoptimizer-arm.cc#n...
src/arm/deoptimizer-arm.cc:132: ASSERT(Memory::uint32_at(pc_after - 4) ==
0xe12fff3c);
On 2011/02/04 13:54:27, Søren Gjesse wrote:
> 4 -> kInstrSize

Done.

http://codereview.chromium.org/6410029/diff/2003/src/arm/deoptimizer-arm.cc#n...
src/arm/deoptimizer-arm.cc:133: ASSERT(Memory::uint8_at(pc_after - 5) == 0xe5);
On 2011/02/04 13:54:27, Søren Gjesse wrote:
> Use Assembler::IsLdrPcImmediateOffset() to check this. Maybe add
> Assembler::IsLdrPcImmediateOffset() which also takes a register and checks
that
> it is ldr ip, [...

Done.

http://codereview.chromium.org/6410029/diff/2003/src/arm/deoptimizer-arm.cc#n...
src/arm/deoptimizer-arm.cc:145: Memory::uint32_at(pc_after - 12) = 0xe1a00000;
On 2011/02/04 13:54:27, Søren Gjesse wrote:
> I think you should be able to use the code patcher here:
> 
> CodePatcher patcher(pc_after - 3 * kInstrSize, 1)
> patcher.masm()->nop();
> 
> if not
> 
> 0xe1a00000 -> al | MOV

Done.

http://codereview.chromium.org/6410029/diff/2003/src/arm/deoptimizer-arm.cc#n...
src/arm/deoptimizer-arm.cc:167: Memory::uint32_at(pc_after - 12) = 0x2a000001;
I used the CodePatcher.

http://codereview.chromium.org/6410029/diff/2003/src/deoptimizer.cc
File src/deoptimizer.cc (right):

http://codereview.chromium.org/6410029/diff/2003/src/deoptimizer.cc#newcode837
src/deoptimizer.cc:837: uint32_t length = last_site - first_site + patch_size();
Done. I used the code patcher and removed the flushing here.

http://codereview.chromium.org/6410029/diff/2003/src/deoptimizer.cc#newcode867
src/deoptimizer.cc:867: uint32_t length = last_site - first_site + patch_size();
Done. I used the code patcher and removed the flushing here.

[Søren Thygesen Gjesse, 2011-02-09 12:56:53]

LGTM

[Karl Klose, 2011-02-14 10:19:26]

Soeren, could you have another look? I have removed a test expectation for
simple-osr.js, which is now succeeding on ARM.

[Søren Thygesen Gjesse, 2011-02-14 11:09:13]

LGTM