[Gold] Fix race condition on parallel link

[Gold] Fix race condition on parallel link

Avoid running Write_after_input_sections_task while Write_sections_task is running.

BUG=161942
Committed: https://crrev.com/5fa22ca7e604e8d7d941099ff56247f05f0288b5
Cr-Commit-Position: refs/heads/master@{#296397}

[tzik, 2014-09-23 17:29:46]

tzik@chromium.org changed reviewers:
+ ccoutant@google.com, thakis@chromium.org

[tzik, 2014-09-23 17:43:53]

PTL.

More readable diff for the patch is available at:
https://user.git.corp.google.com/tzik/binutils-gdb/+/race_fix%5E%21

I recently hit "internal error in do_sized_write, at ehframe.cc:170" on my local
build of chrome.
It happens once per 100 link for official release build with --thread-count=4.
--debug=all output of the error case looks like below. 
IIUC,running Write_after_input_sections_task while Write_sections_task is
running causes the issue.

This CL sets up the task scheduler to ensure the task ordering.

~/toolchain/src/binutils-gdb/gold/ld-new:   0 awake                             
                                       
~/toolchain/src/binutils-gdb/gold/ld-new:   0 running   task Write_sections_task
                                       
~/toolchain/src/binutils-gdb/gold/ld-new:   1 completed task Write_symbols_task
(user: 0.000000 sys: 0.000000 wall: 0.0\
00000)
~/toolchain/src/binutils-gdb/gold/ld-new:   1 running   task Write_data_task    
                                       
~/toolchain/src/binutils-gdb/gold/ld-new:   1 completed task Write_data_task
(user: 0.000000 sys: 0.000000 wall: 0.0000\
00)
~/toolchain/src/binutils-gdb/gold/ld-new:   1 running   task Relocate_task
lib/mojo_views_support_internal.views_init_i\
nternal.o
~/toolchain/src/binutils-gdb/gold/ld-new:   1 completed task Relocate_task
lib/mojo_views_support_internal.views_init_i\
nternal.o (user: 0.000000 sys: 0.000000 wall: 0.000000)
~/toolchain/src/binutils-gdb/gold/ld-new:   1 running   task
Write_after_input_sections_task
~/toolchain/src/binutils-gdb/gold/ld-new: internal error in do_sized_write, at
ehframe.cc:170
~/toolchain/src/binutils-gdb/gold/ld-new:   0 completed task Write_sections_task
(user: 0.010000 sys: 0.010000 wall: 0.\
000000)
~/toolchain/src/binutils-gdb/gold/ld-new:   0 sleeping

[Nico, 2014-09-23 17:50:50]

thakis@chromium.org changed reviewers:
+ thestig@chromium.org

[Nico, 2014-09-23 17:50:52]

Awesome!

Lei, do you know where the eh patch came from?

[Lei Zhang, 2014-09-23 18:20:33]

On 2014/09/23 17:50:52, Nico (hiding) wrote:
> Awesome!
> 
> Lei, do you know where the eh patch came from?

IIRC, Cary Coutant asked me to try applying that and see if it fixed bug 161942.

Have you tried submitting race_fix.patch to binutils?

[tzik, 2014-09-24 08:54:24]

On 2014/09/23 18:20:33, Lei Zhang wrote:
> On 2014/09/23 17:50:52, Nico (hiding) wrote:
> > Awesome!
> > 
> > Lei, do you know where the eh patch came from?
> 
> IIRC, Cary Coutant asked me to try applying that and see if it fixed bug
161942.
> 
> Have you tried submitting race_fix.patch to binutils?

Updated the CL based on the upstream fix.
Now, Cary landed a fix to upstream.

[Lei Zhang, 2014-09-24 09:00:36]

lgtm

[Lei Zhang, 2014-09-24 09:01:05]

Maybe Friday I can do a new binutils build and enable parallel linking again?

[tzik, 2014-09-24 12:03:06]

The CQ bit was checked by tzik@chromium.org

[commit-bot: I haz the power, 2014-09-24 12:03:49]

CQ is trying da patch. Follow status at
 https://chromium-cq-status.appspot.com/patch-status/596853002/20001

[commit-bot: I haz the power, 2014-09-24 12:43:28]

Committed patchset #2 (id:20001) as dd6e63f8c95e9f92df380777b40de15f3230338d

[commit-bot: I haz the power, 2014-09-24 12:44:26]

Patchset 2 (id:??) landed as
https://crrev.com/5fa22ca7e604e8d7d941099ff56247f05f0288b5
Cr-Commit-Position: refs/heads/master@{#296397}