Re: [patch 1/3] Text Edit Lock - i386

[Mathieu Desnoyers <mathieu.desnoyers@polymtl.ca>]

* Chuck Ebbert (cebbert@redhat.com) wrote:
> On 06/18/2007 05:58 PM, Mathieu Desnoyers wrote:
> > Interface to use for code patching : uses a mutex to insure mutual edit
> > exclusion and makes sure the page is writable.
> > 
> ... 
> > +/* Mutex protecting text section modification (dynamic code patching) */
> > +static DEFINE_MUTEX(text_mutex);
> > +
> 
> Probably should be a spinlock.
> 
> And it just occurred to me, how does smp_alternatives deal with this?
> Is it broken now when the text section is read-only?

(note that the implementation I just posted is a proof of concept: I
just noticed that I need to keep track of wether or not I am called
before or after the mark_rodata is done, so the apply alternatives does
not crash at early boot because of the global_flush_tlb().)

A spinlock it will be then :)

SMP alternatives deals with this by simply disabling the whole
protection:

mark_rodata_ro():

#ifdef CONFIG_HOTPLUG_CPU
        /* It must still be possible to apply SMP alternatives. */
        if (num_possible_cpus() <= 1)
#endif
        {
                change_page_attr(virt_to_page(start),
                                 size >> PAGE_SHIFT, PAGE_KERNEL_RX);
                printk("Write protecting the kernel text: %luk\n", size >> 10);
        }

So it's ok if no CPU can be hotplugged, since the CPUs are brought up
before the mark_rodata_ro is done, but not if HOTPLUG is selected.

Mathieu

-- 
Mathieu Desnoyers
Computer Engineering Ph.D. Student, Ecole Polytechnique de Montreal
OpenPGP key fingerprint: 8CD5 52C3 8E3C 4140 715F  BA06 3F25 A8FE 3BAE 9A68