code review 44820047: runtime: fix and improve CPU profiling

runtime: fix and improve CPU profiling

- do not lose profiling signals when we have no mcache (possible for syscalls/cgo)
- do not lose any profiling signals on windows
- fix profiling of cgo programs on windows (they had no m->thread setup)
- properly setup tls in cgo programs on windows
- check _beginthread return value

Fixes issue 6417.
Fixes issue 6986.

[dvyukov, 2013-12-23 18:11:21]

Hello alex.brainman@gmail.com, rsc@golang.org (cc:
golang-codereviews@googlegroups.com),

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

[dvyukov, 2013-12-23 18:13:38]

https://codereview.appspot.com/44820047/diff/80001/src/pkg/runtime/os_windows.c
File src/pkg/runtime/os_windows.c (right):

https://codereview.appspot.com/44820047/diff/80001/src/pkg/runtime/os_windows...
src/pkg/runtime/os_windows.c:249: runtime·atomicstorep(&m->thread, thandle);
this location works for main thread, normal additional threads and cgo
additional threads
previously m->thread was not set for cgo threads, so profiling was not working
at all

https://codereview.appspot.com/44820047/diff/80001/src/pkg/runtime/proc.c
File src/pkg/runtime/proc.c (left):

https://codereview.appspot.com/44820047/diff/80001/src/pkg/runtime/proc.c#old...
src/pkg/runtime/proc.c:2051: if(!Windows && (m == nil || m->mcache == nil))
if windows can do it w/o mcache, then other OSes must be able too

[brainman, 2013-12-24 03:15:45]

LGTM

but what do I know about this. So please wait for rsc.

Thank you. I was fighting with this for a long time.

Alex

[dvyukov, 2013-12-26 09:38:22]

https://codereview.appspot.com/44820047/diff/80001/src/pkg/runtime/proc.c
File src/pkg/runtime/proc.c (left):

https://codereview.appspot.com/44820047/diff/80001/src/pkg/runtime/proc.c#old...
src/pkg/runtime/proc.c:2051: if(!Windows && (m == nil || m->mcache == nil))
On 2013/12/23 18:13:38, dvyukov wrote:
> if windows can do it w/o mcache, then other OSes must be able too

we can do even better, and enforce that sigprof does not allocate

[dvyukov, 2014-01-21 06:58:58]

Russ, please take a look at this as well.

[rsc, 2014-01-22 02:02:01]

LGTM

Thanks.

[dvyukov, 2014-01-22 06:30:20]

*** Submitted as https://code.google.com/p/go/source/detail?r=72c0dfd50949 ***

runtime: fix and improve CPU profiling

- do not lose profiling signals when we have no mcache (possible for
syscalls/cgo)
- do not lose any profiling signals on windows
- fix profiling of cgo programs on windows (they had no m->thread setup)
- properly setup tls in cgo programs on windows
- check _beginthread return value

Fixes issue 6417.
Fixes issue 6986.

R=alex.brainman, rsc
CC=golang-codereviews
https://codereview.appspot.com/44820047

[brainman, 2014-01-23 00:38:56]

Dmitriy,

windows-race builder is still broken:

...
BenchmarkReaderCopyOptimal-4	*** Test killed: ran too long (10m0s).
FAIL	bufio	600.010s
...

Somehow it is getting wedged. I cannot reproduce it here on a small pc:

c:\go\root\src\pkg\bufio>go test -race -run=nothingplease -bench=.*
-benchtime=.1s -cpu=4
PASS
BenchmarkReaderCopyOptimal-4       10000             23347 ns/op
BenchmarkReaderCopyUnoptimal-4      5000             45132 ns/op
BenchmarkReaderCopyNoWriteTo-4      2000             87432 ns/op
BenchmarkWriterCopyOptimal-4        2000             69359 ns/op
BenchmarkWriterCopyUnoptimal-4     10000             43472 ns/op
BenchmarkWriterCopyNoReadFrom-4     2000             69848 ns/op
BenchmarkReaderEmpty-4     20000              8401 ns/op            4225 B/op   
   3 allocs/op
BenchmarkWriterEmpty-4     10000             12015 ns/op            4160 B/op   
   2 allocs/op
BenchmarkWriterFlush-4    200000               825 ns/op               0 B/op   
   0 allocs/op
ok      bufio   3.943s

Alex

[dvyukov, 2014-01-23 10:10:16]

mailed
https://codereview.appspot.com/51360046/

for me it hangs even w/o -race, but with -cpu=4


On Thu, Jan 23, 2014 at 4:38 AM,  <alex.brainman@gmail.com> wrote:
> Dmitriy,
>
> windows-race builder is still broken:
>
> ...
> BenchmarkReaderCopyOptimal-4    *** Test killed: ran too long (10m0s).
> FAIL    bufio   600.010s
> ...
>
> Somehow it is getting wedged. I cannot reproduce it here on a small pc:
>
> c:\go\root\src\pkg\bufio>go test -race -run=nothingplease -bench=.*
> -benchtime=.1s -cpu=4
> PASS
> BenchmarkReaderCopyOptimal-4       10000             23347 ns/op
> BenchmarkReaderCopyUnoptimal-4      5000             45132 ns/op
> BenchmarkReaderCopyNoWriteTo-4      2000             87432 ns/op
> BenchmarkWriterCopyOptimal-4        2000             69359 ns/op
> BenchmarkWriterCopyUnoptimal-4     10000             43472 ns/op
> BenchmarkWriterCopyNoReadFrom-4     2000             69848 ns/op
> BenchmarkReaderEmpty-4     20000              8401 ns/op            4225
> B/op       3 allocs/op
> BenchmarkWriterEmpty-4     10000             12015 ns/op            4160
> B/op       2 allocs/op
> BenchmarkWriterFlush-4    200000               825 ns/op               0
> B/op       0 allocs/op
> ok      bufio   3.943s
>
> Alex
>
> https://codereview.appspot.com/44820047/