[parisc-linux] init_idle and local_irq_save

[parisc-linux] init_idle and local_irq_save

[Grant Grundler]

Hi all,
I'm trying to get 2.6 SMP working on a UP box and was walking
through the init sequence. I have two questions related to init_idle()
in kernel/sched.c:

o local_irq_save()/local_irq_restore are called without an
  obvious local_irq_disable() call in between. Is this really
  needed for some other reason or is the local_irq_disable() missing?

o double_rq_lock() acquires the locks with lowest address first.
  double_rq_unlock() does not release those in the inverse order.
  Is that a problem?
  I'm just nervous since I didn't see the local_irq_disable()
  in init_idle() which would guarantee both unlock's happen atomically.

thanks,
grant

Re: [parisc-linux] init_idle and local_irq_save

[James Bottomley]

On Mon, 2004-03-29 at 11:03, Grant Grundler wrote:
> o local_irq_save()/local_irq_restore are called without an
>   obvious local_irq_disable() call in between. Is this really
>   needed for some other reason or is the local_irq_disable() missing?

local_irq_save() does do an irq disable.  You're thinking of
local_save_flags() which only saves the flags but doesn't disable
interrupts.

> o double_rq_lock() acquires the locks with lowest address first.
>   double_rq_unlock() does not release those in the inverse order.
>   Is that a problem?
>   I'm just nervous since I didn't see the local_irq_disable()
>   in init_idle() which would guarantee both unlock's happen atomically.

No.  The deadlock potential comes only from lock acquisition order, not
lock release order:  nothing can block a lock release from proceeding.

James

Re: [parisc-linux] init_idle and local_irq_save

[Matthew Wilcox]

On Mon, Mar 29, 2004 at 09:03:57AM -0700, Grant Grundler wrote:
> o double_rq_lock() acquires the locks with lowest address first.
>   double_rq_unlock() does not release those in the inverse order.
>   Is that a problem?

No.  You have two possibilities (locks A, B; acquire and release):

1. Aa Ba Ar Br
2. Aa Ba Br Ar

It doesn't matter how you combine them; either you are waiting for A or
you have A and are waiting for B.  You never have B and are waiting for A.
Case 2 is the more common idiom since B is then entirely nested within A,
but lock A ranks lock B in both cases, so you're safe.

-- 
"Next the statesmen will invent cheap lies, putting the blame upon 
the nation that is attacked, and every man will be glad of those
conscience-soothing falsities, and will diligently study them, and refuse
to examine any refutations of them; and thus he will by and by convince 
himself that the war is just, and will thank God for the better sleep 
he enjoys after this process of grotesque self-deception." -- Mark Twain