code review 6810095: runtime/race: more precise handling of finalizers

runtime/race: more precise handling of finalizers
Currently race detector runtime just disables race detection in the finalizer goroutine.
It has false positives when a finalizer writes to shared memory -- the race with finalizer is reported in a normal goroutine that accesses the same memory.
After this change I am going to synchronize the finalizer goroutine with the rest of the world in racefingo(). This is closer to what happens in reality and so
does not have false positives.
And also add README file with instructions how to build the runtime.

[dvyukov, 2012-11-08 14:27:59]

Hello golang-dev@googlegroups.com,

I'd like you to review this change to
https://dvyukov%40google.com@code.google.com/p/go/

[dvyukov, 2012-11-09 11:24:00]

On 2012/11/08 14:27:59, dvyukov wrote:
> Hello mailto:golang-dev@googlegroups.com,
> 
> I'd like you to review this change to
> https://dvyukov%2540google.com%40code.google.com/p/go/

I've added here necessary runtime changes and the README file.
The runtime is built on revision 167609.

[minux1, 2012-11-09 11:29:28]

please update the description.

https://codereview.appspot.com/6810095/diff/5002/src/pkg/runtime/race/README
File src/pkg/runtime/race/README (right):

https://codereview.appspot.com/6810095/diff/5002/src/pkg/runtime/race/README#...
src/pkg/runtime/race/README:10: You may need gcc >= 4.6.1. gcc 4.3 produces bad
syso files with weak symbols
is this still the case? CL 6783050 just modified cmd/ld to support weak symbols.
sorry i don't have gcc 4.3 to test.
if it's still true, could you please mail me a copy of bad linux/amd64 syso off
list so that i can fix the issue in cmd/6l?

[dvyukov, 2012-11-09 11:31:55]

On 2012/11/09 11:29:28, minux wrote:
> please update the description.

Done.

> https://codereview.appspot.com/6810095/diff/5002/src/pkg/runtime/race/README
> File src/pkg/runtime/race/README (right):
> 
>
https://codereview.appspot.com/6810095/diff/5002/src/pkg/runtime/race/README#...
> src/pkg/runtime/race/README:10: You may need gcc >= 4.6.1. gcc 4.3 produces
bad
> syso files with weak symbols
> is this still the case? CL 6783050 just modified cmd/ld to support weak
symbols.
> sorry i don't have gcc 4.3 to test.
> if it's still true, could you please mail me a copy of bad linux/amd64 syso
off
> list so that i can fix the issue in cmd/6l?

I will try to reproduce it.

[dvyukov, 2012-11-09 11:43:38]

On 2012/11/09 11:31:55, dvyukov wrote:
> On 2012/11/09 11:29:28, minux wrote:
> > please update the description.
> 
> Done.
> 
> > https://codereview.appspot.com/6810095/diff/5002/src/pkg/runtime/race/README
> > File src/pkg/runtime/race/README (right):
> > 
> >
>
https://codereview.appspot.com/6810095/diff/5002/src/pkg/runtime/race/README#...
> > src/pkg/runtime/race/README:10: You may need gcc >= 4.6.1. gcc 4.3 produces
> bad
> > syso files with weak symbols
> > is this still the case? CL 6783050 just modified cmd/ld to support weak
> symbols.
> > sorry i don't have gcc 4.3 to test.
> > if it's still true, could you please mail me a copy of bad linux/amd64 syso
> off
> > list so that i can fix the issue in cmd/6l?
> 
> I will try to reproduce it.

Seem to work now.

However gcc 4.4 still fails when I build in debug mode:

gotsan.cc:1485: sorry, unimplemented: inlining failed in call to ‘void
__tsan::MemoryAccessImpl(__tsan::ThreadState*, __sanitizer::uptr, int, bool,
__tsan::FastState, __sanitizer::u64*, __tsan::Shadow)’: function not inlinable

I rephrased it as "Tested with gcc 4.6.1 and 4.7.0."

[minux1, 2012-11-11 17:48:41]

On Fri, Nov 9, 2012 at 7:43 PM, <dvyukov@google.com> wrote:
>
> Seem to work now.
>
good to hear.

>
> However gcc 4.4 still fails when I build in debug mode:
>
> gotsan.cc:1485: sorry, unimplemented: inlining failed in call to ‘void
> __tsan::MemoryAccessImpl(__**tsan::ThreadState*, __sanitizer::uptr, int,
> bool, __tsan::FastState, __sanitizer::u64*, __tsan::Shadow)’: function
> not inlinable
>
> I rephrased it as "Tested with gcc 4.6.1 and 4.7.0.“
>
the README looks fine to me now.

[rsc, 2012-11-12 20:41:03]

LGTM

[dvyukov, 2012-11-14 13:23:50]

*** Submitted as http://code.google.com/p/go/source/detail?r=d1854dd425b2 ***

runtime/race: more precise handling of finalizers
Currently race detector runtime just disables race detection in the finalizer
goroutine.
It has false positives when a finalizer writes to shared memory -- the race with
finalizer is reported in a normal goroutine that accesses the same memory.
After this change I am going to synchronize the finalizer goroutine with the
rest of the world in racefingo(). This is closer to what happens in reality and
so
does not have false positives.
And also add README file with instructions how to build the runtime.

R=golang-dev, minux.ma, rsc
CC=golang-dev
http://codereview.appspot.com/6810095