Re: [PULL] x86 cpumask work

[Rusty Russell <rusty@rustcorp.com.au>]

On Sunday 15 March 2009 16:30:44 Ingo Molnar wrote:
> 
> >  	} else if (check_c1e_idle(c)) {
> >  		printk(KERN_INFO "using C1E aware idle routine\n");
> > -		alloc_cpumask_var(&c1e_mask, GFP_KERNEL);
> > -		cpumask_clear(c1e_mask);
> > +		/* c1e_mask can only be NULL during boot of first cpu. */
> > +		if (c1e_mask == NULL) {
> > +			alloc_cpumask_var(&c1e_mask, GFP_KERNEL);
> 
> Sigh, there are two bugs here:
> 
>    1) what if the GFP_KERNEL allocation fails?

As the comments says, it can only be NULL during boot of the first CPU.

start_kernel -> check_bugs -> identify_boot_cpu -> identify_cpu
  -> select_idle_routine.

Did you want me to panic if it fails?

>    2) this code is called with interrupts disabled, so a 
>       GFP_KERNEL allocation can be lethal.

I don't see that in my analysis of the code.

> c1e_mask should stay a static cpumask...
> 
> Why do we convert static, standalone masks to cpumask_var?

Because we can't undefine struct cpumask while there are any left.  If we
really want to we can make it a bitmap, but that's not a good thing to
encourage.

Hope that clarifies,
Rusty.