code review 713041: runtime: add memory profiling, disabled.

runtime: add memory profiling, disabled.
no way to get the data out yet.

add prototype for runtime.Callers,
missing from last CL.

[rsc, 2010-03-24 01:01:57]

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

I'd like you to review this change.

[r, 2010-03-24 02:39:06]

LGTM

http://codereview.appspot.com/713041/diff/4001/5002
File src/pkg/runtime/extern.go (right):

http://codereview.appspot.com/713041/diff/4001/5002#newcode32
src/pkg/runtime/extern.go:32: // to skip before recording in pc.  It returns the
number of entries written to pc.
this function should appear in the CL description

http://codereview.appspot.com/713041/diff/4001/5002#newcode32
src/pkg/runtime/extern.go:32: // to skip before recording in pc.  It returns the
number of entries written to pc.
define what zero means for skip

http://codereview.appspot.com/713041/diff/4001/5002#newcode177
src/pkg/runtime/extern.go:177: // A MemProfileKind specifies how frequently to
record
s/A //

http://codereview.appspot.com/713041/diff/4001/5004
File src/pkg/runtime/malloc.cgo (right):

http://codereview.appspot.com/713041/diff/4001/5004#newcode19
src/pkg/runtime/malloc.cgo:19: fastrand1(void)
doesn't this need to be interlocked? i guess it's protected by the gc locks but
you should comment to that effect. also say where the algorithm comes from and
why you can't use the one in chan.c

http://codereview.appspot.com/713041/diff/4001/5008
File src/pkg/runtime/mprof.cgo (right):

http://codereview.appspot.com/713041/diff/4001/5008#newcode16
src/pkg/runtime/mprof.cgo:16: // NOTE(rsc): Everything here use cas if
contention became an issue.
grammar?

http://codereview.appspot.com/713041/diff/4001/5008#newcode35
src/pkg/runtime/mprof.cgo:35: static Bucket *buckhash[179999];
that's huge. can't you allocate this only if you need it? or is there no point
because this is always on?

[rsc, 2010-03-24 03:23:02]

> http://codereview.appspot.com/713041/diff/4001/5008#newcode35
> src/pkg/runtime/mprof.cgo:35: static Bucket *buckhash[179999];
> that's huge. can't you allocate this only if you need it? or is there no
> point because this is always on?

if the profiling is off, it's just address space.
i could allocate it the first time, but i don't
think it matters much - bss is free.
i do hope to be able to have it always on
once we've done appropriate tuning.

russ

[rsc, 2010-03-24 03:31:31]

All done.

I changed buckhash to allocate anyway.

Regarding fastrand1, it probably should be interlocked
but I don't care too much if two different goroutines stumble
upon the same random number.  Comment added.

Russ

[rsc, 2010-03-24 03:48:33]

*** Submitted as http://code.google.com/p/go/source/detail?r=4c2146eece2e ***

runtime: add memory profiling, disabled.
no way to get the data out yet.

add prototype for runtime.Callers,
missing from last CL.

R=r
CC=golang-dev
http://codereview.appspot.com/713041