i#1114 JIT optimization: cache consistency interface

Commit log for first patchset:
---------------
i#1114 JIT optimization: cache consistency interface

Implements the annotations to un/manage a code area,
and integrates the optimization with bb construction
and vmarea management operations.
---------------

[Byron, 2015-12-14 10:51:08]

Commit log for latest patchset:
---------------
i#1114 JIT optimization: cache consistency interface

Implements the annotations to un/manage a code area,
and integrates the optimization with bb construction
and vmarea management operations.

Review-URL: https://codereview.appspot.com/278490043
---------------

[bruening, 2015-12-14 20:38:35]

LGTM if fixed for ARM

https://codereview.appspot.com/278490043/diff/20001/core/jit_opt.c
File core/jit_opt.c (right):

https://codereview.appspot.com/278490043/diff/20001/core/jit_opt.c#newcode16
core/jit_opt.c:16: LOG(GLOBAL, LOG_ANNOTATIONS, 1,
Is 1 too low?

https://codereview.appspot.com/278490043/diff/20001/core/jit_opt.c#newcode31
core/jit_opt.c:31: LOG(GLOBAL, LOG_ANNOTATIONS, 1,
Is 1 too low?

https://codereview.appspot.com/278490043/diff/20001/core/jit_opt.c#newcode35
core/jit_opt.c:35: mutex_lock(&thread_initexit_lock);
Are annotations invoked straight from the cache?  If so, they are another new
source of whereami==fcache running DR code, which will cause problems for
prof_pcs even without a client.

https://codereview.appspot.com/278490043/diff/20001/core/vmareas.c
File core/vmareas.c (right):

https://codereview.appspot.com/278490043/diff/20001/core/vmareas.c#newcode468
core/vmareas.c:468: #define DEBUG_SIGILL_OPCODE 0x0b0f
This is x86-specific, and should be with the other raw sequences: add to
RAW_OPCODE_ enum instead.

https://codereview.appspot.com/278490043/diff/20001/core/vmareas.c#newcode6277
core/vmareas.c:6277: vm_area_t *region;
Should this assert that -opt_jit was specified?

https://codereview.appspot.com/278490043/diff/20001/core/vmareas.c#newcode9005
core/vmareas.c:9005: *(ushort *) f->start_pc = DEBUG_SIGILL_OPCODE;
incorrect code for ARM

https://codereview.appspot.com/278490043/diff/20001/core/vmareas.c#newcode9159
core/vmareas.c:9159: *(ushort *) f->start_pc = DEBUG_SIGILL_OPCODE;
Ditto

[Byron, 2015-12-15 00:08:20]

Committed as
https://github.com/DynamoRIO/dynamorio/commit/c11a8afb05da6750345c8e1f3728a46...

Final commit log: 
---------------
i#1114 JIT optimization: cache consistency interface

Implements the annotations to un/manage a code area,
and integrates the optimization with bb construction
and vmarea management operations.

Review-URL: https://codereview.appspot.com/278490043
---------------

[Byron, 2015-12-15 00:08:54]

https://codereview.appspot.com/278490043/diff/20001/core/jit_opt.c
File core/jit_opt.c (right):

https://codereview.appspot.com/278490043/diff/20001/core/jit_opt.c#newcode16
core/jit_opt.c:16: LOG(GLOBAL, LOG_ANNOTATIONS, 1,
On 2015/12/14 20:38:35, bruening wrote:
> Is 1 too low?

Done.

https://codereview.appspot.com/278490043/diff/20001/core/jit_opt.c#newcode31
core/jit_opt.c:31: LOG(GLOBAL, LOG_ANNOTATIONS, 1,
On 2015/12/14 20:38:34, bruening wrote:
> Is 1 too low?

Done.

https://codereview.appspot.com/278490043/diff/20001/core/jit_opt.c#newcode35
core/jit_opt.c:35: mutex_lock(&thread_initexit_lock);
On 2015/12/14 20:38:35, bruening wrote:
> Are annotations invoked straight from the cache?  If so, they are another new
> source of whereami==fcache running DR code, which will cause problems for
> prof_pcs even without a client.

Will fix in a separate CL by adding WHERE_CLEAN_CALLEE and setting it in the
clean call instrumentation (for -prof_pcs only).

https://codereview.appspot.com/278490043/diff/20001/core/vmareas.c
File core/vmareas.c (right):

https://codereview.appspot.com/278490043/diff/20001/core/vmareas.c#newcode468
core/vmareas.c:468: #define DEBUG_SIGILL_OPCODE 0x0b0f
On 2015/12/14 20:38:35, bruening wrote:
> This is x86-specific, and should be with the other raw sequences: add to
> RAW_OPCODE_ enum instead.

Done.

https://codereview.appspot.com/278490043/diff/20001/core/vmareas.c#newcode6277
core/vmareas.c:6277: vm_area_t *region;
On 2015/12/14 20:38:35, bruening wrote:
> Should this assert that -opt_jit was specified?

Done.

https://codereview.appspot.com/278490043/diff/20001/core/vmareas.c#newcode9005
core/vmareas.c:9005: *(ushort *) f->start_pc = DEBUG_SIGILL_OPCODE;
On 2015/12/14 20:38:35, bruening wrote:
> incorrect code for ARM

Done.

https://codereview.appspot.com/278490043/diff/20001/core/vmareas.c#newcode9159
core/vmareas.c:9159: *(ushort *) f->start_pc = DEBUG_SIGILL_OPCODE;
On 2015/12/14 20:38:35, bruening wrote:
> Ditto

Done.