take 2: request for review: backport of sx and rwlocks from 7.0 to 6-stable

Updated patch attached:

sizeof(struct sx) should be the same.
garbage removed from kern_sx.c.
alignment of "struct thread" set to constant
rather than magic number.
uma_zalloc of "struct thread" fixed to use
constant.



-- 
- Alfred Perlstein

_______________________________________________
freebsd-smpfreebsd.org mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-smp
To unsubscribe, send any mail to
"freebsd-smp-unsubscribefreebsd.org"

On Monday 03 September 2007 10:31:52 pm Alfred Perlstein
wrote:
> Updated patch attached:
> 
> sizeof(struct sx) should be the same.
> garbage removed from kern_sx.c.
> alignment of "struct thread" set to constant
rather than magic number.
> uma_zalloc of "struct thread" fixed to use
constant.

I would not add gratuitous diffs to HEAD (the thread
alignment thing).  If you 
want to change that, you should change it in HEAD first.  I
would keep the 
diffs with HEAD minimal in the MFC though as it makes
backporting future 
fixes easier.

I would add new fields to 'struct thread' at the bottom of
the structure in 
6.x to minimize ABI trauma.  I think if you put td_tsqueue
before td_locks 
you will actually not adjust the ABI since there should be a
"hole" for a 
char there due to alignment already.

The locking notes for the new fields in 'struct thread'
should be (j) not (t).  
(t) doesn't exist in 6.x.

I'm not sure what the debug.turnstile_nullowners thing is
(maybe added with 
threadlock?) but it probably isn't relevant to 6.x.

Why did you remove the 'show sleepq' DDB command completely?
 It's very 
useful. :(

6.x already has a LOCK_CLASS() macro in sys/lock.h.  No need
to duplicate it 
in kern_sx.c.

-- 
John Baldwin
_______________________________________________
freebsd-smpfreebsd.org mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-smp
To unsubscribe, send any mail to
"freebsd-smp-unsubscribefreebsd.org"

2007/9/19, John Baldwin <jhbfreebsd.org>:
> Why did you remove the 'show sleepq' DDB command
completely?  It's very
> useful. :(

Honestly, if there is something I really dislike in this
MFC, except
the removal of 'show sleepq' command is the backporting of
lock_profile architecture.
It shouldn't have been happened.
it is still not mature for 7.0 I think and kmacy is going to
replace
with a better mechanism for 8.0 certainly.
So Alfred, if it is feasible for you, can you remove the
lock_profile
backporting from 6.0?
I'm going to help you with this if you find any trouble.

Respect,
Attilio


-- 
Peace can only be achieved by understanding - A. Einstein
_______________________________________________
freebsd-smpfreebsd.org mailing list
http://lists.freebsd.org/mailman/listinfo/freebsd-smp
To unsubscribe, send any mail to
"freebsd-smp-unsubscribefreebsd.org"