code review 11883045: runtime: fix arm preemption

runtime: fix arm preemption

Preemption during the software floating point code
could cause m (R9) to change, so that when the
original registers were restored at the end of the
floating point handler, the changed and correct m
would be replaced by the old and incorrect m.

[rsc, 2013-08-01 04:16:29]

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

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

[rsc, 2013-08-01 04:16:35]

*** Submitted as https://code.google.com/p/go/source/detail?r=0e50ba9a0494 ***

runtime: fix arm preemption

Preemption during the software floating point code
could cause m (R9) to change, so that when the
original registers were restored at the end of the
floating point handler, the changed and correct m
would be replaced by the old and incorrect m.

TBR=dvyukov
CC=golang-dev
https://codereview.appspot.com/11883045

[dave_cheney.net, 2013-08-01 04:21:17]

https://codereview.appspot.com/11883045/diff/4001/src/pkg/runtime/vlop_arm.s
File src/pkg/runtime/vlop_arm.s (right):

https://codereview.appspot.com/11883045/diff/4001/src/pkg/runtime/vlop_arm.s#...
src/pkg/runtime/vlop_arm.s:70: MOVM.IA.W	(R0), [R11-R12]
Thanks rsc. Could this also be written as

MOVM.IA.W (R0), [R1,R2,R3,R4,R5,R6,R7,R8,R11,R12] ?

It probably doesn't matter that much

[dvyukov, 2013-08-01 10:40:42]

rubber stamp LGTM

[remyoudompheng, 2013-08-01 10:52:02]

The linux arm builders seem to still complain about numeric
corruptions, not sure what is the reason.

Rémy.

2013/8/1, dvyukov@google.com <dvyukov@google.com>:
> rubber stamp LGTM
>
> https://codereview.appspot.com/11883045/
>
> --
>
> ---
> You received this message because you are subscribed to the Google Groups
> "golang-dev" group.
> To unsubscribe from this group and stop receiving emails from it, send an
> email to golang-dev+unsubscribe@googlegroups.com.
> For more options, visit https://groups.google.com/groups/opt_out.
>
>
>

[rsc, 2013-08-01 14:50:57]

https://codereview.appspot.com/11883045/diff/4001/src/pkg/runtime/vlop_arm.s
File src/pkg/runtime/vlop_arm.s (right):

https://codereview.appspot.com/11883045/diff/4001/src/pkg/runtime/vlop_arm.s#...
src/pkg/runtime/vlop_arm.s:70: MOVM.IA.W	(R0), [R11-R12]
On 2013/08/01 04:21:17, dfc wrote:
> Thanks rsc. Could this also be written as
> 
> MOVM.IA.W (R0), [R1,R2,R3,R4,R5,R6,R7,R8,R11,R12] ?
> 
> It probably doesn't matter that much 

No, it can't, because I need to skip over the R9 and R10 values that are on the
stack. I didn't change the form of the struct being passed to _sfloat, because
it assumes it can index into the list using the register number.

It's possible that if R0 is being updated that I could have done ADD $8, R0
instead of MOVW $52(R13), R0, but both are a single instruction and I was more
sure of the latter.