[Fuchsia] Implements the thread interrupter for CPU profiling.

[Fuchsia] Implements the thread interrupter for CPU profiling.

R=asiva@google.com

Committed: https://github.com/dart-lang/sdk/commit/904aefe6e3b371cf82d41c92772f5995f3c3c40c

[zra, 2017-05-12 22:47:01]

zra@google.com changed reviewers:
+ asiva@google.com, rmacnak@google.com

[zra, 2017-05-12 22:47:01]

zra@google.com changed reviewers:
+ asiva@google.com, rmacnak@google.com

[zra, 2017-05-12 22:47:02]

This enables the the profiler's thread interrupter for Fuchsia, but does not
attempt to read a suspended thread's state, as that is not yet supported.

[zra, 2017-05-12 22:47:02]

This enables the the profiler's thread interrupter for Fuchsia, but does not
attempt to read a suspended thread's state, as that is not yet supported.

[siva, 2017-05-18 00:11:30]

lgtm

https://codereview.chromium.org/2796143002/diff/20001/runtime/vm/thread_inter...
File runtime/vm/thread_interrupter_fuchsia.cc (right):

https://codereview.chromium.org/2796143002/diff/20001/runtime/vm/thread_inter...
runtime/vm/thread_interrupter_fuchsia.cc:40: mx_status_get_string(status));
OS::PrintErr here?

https://codereview.chromium.org/2796143002/diff/20001/runtime/vm/thread_inter...
runtime/vm/thread_interrupter_fuchsia.cc:54: state->fp =
static_cast<uintptr_t>(regs->r[29]);
can we use a name here instead of 29

[rmacnak, 2017-05-18 01:31:27]

https://codereview.chromium.org/2796143002/diff/20001/runtime/vm/thread_inter...
File runtime/vm/thread_interrupter_fuchsia.cc (right):

https://codereview.chromium.org/2796143002/diff/20001/runtime/vm/thread_inter...
runtime/vm/thread_interrupter_fuchsia.cc:77: status =
mx_object_get_child(mx_process_self(),
Why use the koid as the ThreadId and keep creating new handles instead of using
the handle as the ThreadId?

[zra, 2017-05-23 16:09:43]

https://codereview.chromium.org/2796143002/diff/20001/runtime/vm/thread_inter...
File runtime/vm/thread_interrupter_fuchsia.cc (right):

https://codereview.chromium.org/2796143002/diff/20001/runtime/vm/thread_inter...
runtime/vm/thread_interrupter_fuchsia.cc:40: mx_status_get_string(status));
On 2017/05/18 00:11:30, siva wrote:
> OS::PrintErr here?

Done.

https://codereview.chromium.org/2796143002/diff/20001/runtime/vm/thread_inter...
runtime/vm/thread_interrupter_fuchsia.cc:54: state->fp =
static_cast<uintptr_t>(regs->r[29]);
On 2017/05/18 00:11:30, siva wrote:
> can we use a name here instead of 29

Done.

https://codereview.chromium.org/2796143002/diff/20001/runtime/vm/thread_inter...
runtime/vm/thread_interrupter_fuchsia.cc:77: status =
mx_object_get_child(mx_process_self(),
On 2017/05/18 01:31:27, rmacnak wrote:
> Why use the koid as the ThreadId and keep creating new handles instead of
using
> the handle as the ThreadId?

Mostly because it's not entirely clear whether an open handle can be shared
across threads without passing it over a channel, but also because I don't want
to have long-lived open handles around in case they leak or become dangling. We
do that on Windows only because there is no other option.

It is probably inefficient to open/close these handles for every sample, but
this whole approach is temporary until MG-430 is resolved. I'll add a comment to
that effect.

[zra, 2017-05-23 17:15:44]

I've had to disable the profiler by default because suspending a thread in a
call to pthread_cond_wait() causes pthread_cond_wait() to return ETIMEDOUT. I'll
file a bug, and then land a work-around depending on the response.

[zra, 2017-05-23 17:22:11]

Description was changed from

==========
[Fuchsia] Implements the thread interrupter for CPU profiling.
==========

to

==========
[Fuchsia] Implements the thread interrupter for CPU profiling.

R=asiva@google.com

Committed:
https://github.com/dart-lang/sdk/commit/904aefe6e3b371cf82d41c92772f5995f3c3c40c
==========

[zra, 2017-05-23 17:22:12]

Committed patchset #6 (id:100001) manually as
904aefe6e3b371cf82d41c92772f5995f3c3c40c (presubmit successful).