code review 7388043: runtime: prepare for M's running w/o mcache

runtime: prepare for M's running w/o mcache
Can not happen ATM. In preparation for the new scheduler.

[dvyukov, 2013-02-20 16:21:12]

Hello golang-dev@googlegroups.com,

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

[rsc, 2013-02-20 16:22:04]

How will this be possible in the new scheduler? Doesn't every m allocate
mcache very early? Is it that there will be fewer mcache's than m's?

[dvyukov, 2013-02-20 16:49:25]

In the new scheduler mcache sits in a separate P struct. P structs along
with mcache's are allocated during bootstrap or in GOMAXPROCS(). M's
acquire P (and mcache) before running Go code, and release P (and mcache)
before parking or before entering syscall.
So, yes, there are fewer mcaches' than M's.
Some other resources will be moved to P as well, e.g. cache of dead G's;
potentially cache of stack segments can be moved as well, but it may be
tricky; and this will be a foundation for user proc-local data (caches,
stats, etc).






On Wed, Feb 20, 2013 at 8:22 PM, Russ Cox <rsc@golang.org> wrote:

> How will this be possible in the new scheduler? Doesn't every m allocate
> mcache very early? Is it that there will be fewer mcache's than m's?
>
>
>

[rsc, 2013-02-20 16:58:04]

LGTM

[dvyukov, 2013-02-20 17:18:04]

*** Submitted as https://code.google.com/p/go/source/detail?r=592fa6bfc380 ***

runtime: prepare for M's running w/o mcache
Can not happen ATM. In preparation for the new scheduler.

R=golang-dev, rsc
CC=golang-dev
https://codereview.appspot.com/7388043