code review 11551044: runtime: reclaim another word in cgocallback_gofunc

runtime: more cgocallback_gofunc work

Debugging the Windows breakage I noticed that SEH
only exists on 386, so we can balance the two stacks
a little more on amd64 and reclaim another word.

Now we're down to just one word consumed by
cgocallback_gofunc, having reclaimed 25% of the
overall budget (4 words out of 16).

Separately, fix windows/386 - the SEH must be on the
m0 stack, as must the saved SP, so we are forced to have
a three-word frame for 386. It matters much less for
386, because there 128 bytes gives 32 words to use.

[rsc, 2013-07-24 03:10:55]

Hello dvyukov (cc: golang-dev@googlegroups.com),

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

[brainman, 2013-07-24 03:28:39]

Test passes on windows/amd64, but windows/386 is broken in
misc/cgo/test/TestCthread. I can have a look at it, but I am slow.

Alex

[rsc, 2013-07-24 04:32:35]

Sorry about that. I cannot read, apparently.

I have updated this CL to include a windows/386 fix, but it's untested. Can
you try it?

Thanks.
Russ

[brainman, 2013-07-24 04:47:10]

LGTM. Both windows/386 and windows/amd64 builds are fine. It is really
complicated. :-(

Alex

https://codereview.appspot.com/11551044/diff/13001/src/pkg/runtime/asm_386.s
File src/pkg/runtime/asm_386.s (right):

https://codereview.appspot.com/11551044/diff/13001/src/pkg/runtime/asm_386.s#...
src/pkg/runtime/asm_386.s:557: // On Windows, the SEH is at 4(SP) and 8(SP).
Maybe you want this line after:

// 4(SP) and 8(SP) are unused.

down below.

[rsc, 2013-07-24 04:50:47]

https://codereview.appspot.com/11551044/diff/13001/src/pkg/runtime/asm_386.s
File src/pkg/runtime/asm_386.s (right):

https://codereview.appspot.com/11551044/diff/13001/src/pkg/runtime/asm_386.s#...
src/pkg/runtime/asm_386.s:557: // On Windows, the SEH is at 4(SP) and 8(SP).
On 2013/07/24 04:47:11, brainman wrote:
> Maybe you want this line after:
> // 4(SP) and 8(SP) are unused.
> down below.

No, that was the bug. The comment here is describing one stack; the comment
below is describing a different stack ("in the new goroutine"). The SEH was on
the wrong stack.

[brainman, 2013-07-24 05:18:34]

Thank you.

Alex

https://codereview.appspot.com/11551044/diff/13001/src/pkg/runtime/asm_386.s
File src/pkg/runtime/asm_386.s (right):

https://codereview.appspot.com/11551044/diff/13001/src/pkg/runtime/asm_386.s#...
src/pkg/runtime/asm_386.s:557: // On Windows, the SEH is at 4(SP) and 8(SP).
On 2013/07/24 04:50:48, rsc wrote:
> ... The comment here is describing one stack; the comment
> below is describing a different stack ("in the new goroutine"). The SEH was on
> the wrong stack.

Yes. You are right about that.

[dvyukov, 2013-07-24 05:31:28]

LGTM
I think I even understand what happens here

https://codereview.appspot.com/11551044/diff/13001/src/pkg/runtime/proc.c
File src/pkg/runtime/proc.c (right):

https://codereview.appspot.com/11551044/diff/13001/src/pkg/runtime/proc.c#new...
src/pkg/runtime/proc.c:659: m->seh = (SEH*)((uintptr*)&x + 1);
please set m->seh to nil otherwise, we are reusing an old M so it's not all
zeros

[rsc, 2013-07-24 13:01:11]

https://codereview.appspot.com/11551044/diff/13001/src/pkg/runtime/proc.c
File src/pkg/runtime/proc.c (right):

https://codereview.appspot.com/11551044/diff/13001/src/pkg/runtime/proc.c#new...
src/pkg/runtime/proc.c:659: m->seh = (SEH*)((uintptr*)&x + 1);
On 2013/07/24 05:31:29, dvyukov wrote:
> please set m->seh to nil otherwise, we are reusing an old M so it's not all
> zeros

I changed the other assignment to m->seh to only happen on windows/386. It will
now be nil always, because Ms are zeroed to start with.

[rsc, 2013-07-24 13:02:04]

*** Submitted as https://code.google.com/p/go/source/detail?r=e7b0d65f9e5b ***

runtime: more cgocallback_gofunc work

Debugging the Windows breakage I noticed that SEH
only exists on 386, so we can balance the two stacks
a little more on amd64 and reclaim another word.

Now we're down to just one word consumed by
cgocallback_gofunc, having reclaimed 25% of the
overall budget (4 words out of 16).

Separately, fix windows/386 - the SEH must be on the
m0 stack, as must the saved SP, so we are forced to have
a three-word frame for 386. It matters much less for
386, because there 128 bytes gives 32 words to use.

R=dvyukov, alex.brainman
CC=golang-dev
https://codereview.appspot.com/11551044