Re: [PATCH] Avoid calling down_read and down_write during startup

[Alan Stern <stern@rowland.harvard.edu>]

On Fri, 24 Feb 2006, Benjamin LaHaise wrote:

> On Thu, Feb 23, 2006 at 04:16:31PM -0800, Andrew Morton wrote:
> > down_write() unconditionally (and reasonably) does local_irq_enable() in
> > the uncontended case.  And enabling local interrupts early in boot can
> > cause crashes.
> 
> Why not do a down_write_trylock() instead first?  Then the code doesn't 
> have the dependancy on system_state, which seems horribly fragile.

I suggested this to Andrew.  His reply was as follows:

> > which means we can't avoid calling down_write.  The 
> >  only solution I can think of is to use down_write_trylock in the 
> >  blocking_notifier_chain_register and unregister routines, even though 
> >  doing that is a crock.
> > 
> >  Or else change __down_read and __down_write to use spin_lock_irqsave 
> >  instead of spin_lock_irq.  What do you think would be best?
> 
> Nothing's pretty.  Perhaps look at system_state and not do any locking at all
> in early boot?

I admit that the whole things is fragile.  IMO the safest approach would
be for __down_read and __down_write not to assume that interrupts are
currently enabled.  But that would introduce more overhead as well; at
least this way the overhead is confined to the notifier chain registration
routines.

Alan Stern

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/