Re: [Xen-devel] [RFC PATCH v1] Replace tasklets with per-cpu implementation.

["Jan Beulich" <JBeulich@xxxxxxxx>]

>>> On 09.09.14 at 16:37, <konrad.wilk@xxxxxxxxxx> wrote:
> On Tue, Sep 09, 2014 at 10:01:09AM +0100, Jan Beulich wrote:
>> >>> On 08.09.14 at 21:01, <konrad.wilk@xxxxxxxxxx> wrote:
>> > +static int cpu_callback(
>> > +    struct notifier_block *nfb, unsigned long action, void *hcpu)
>> > +{
>> > +    unsigned int cpu = (unsigned long)hcpu;
>> > +
>> > +    switch ( action )
>> > +    {
>> > +    case CPU_UP_PREPARE:
>> > +        INIT_LIST_HEAD(&per_cpu(dpci_list, cpu));
>> > +        break;
>> > +    case CPU_UP_CANCELED:
>> > +    case CPU_DEAD:
>> > +        migrate_tasklets_from_cpu(cpu, (&per_cpu(dpci_list, cpu)));
>> 
>> Can CPUs go down while softirqs are pending on them?
> 
> No. By the time we get here, the CPU is no longer "hearing" them.

So what's that code (also still present in the newer patch you
had attached here) for then?

> +void dpci_kill(struct domain *d)
> +{
> +    while ( test_and_set_bit(STATE_SCHED, &d->state) )
> +    {
> +        do {
> +            process_pending_softirqs();
> +        } while ( test_bit(STATE_SCHED, &d->state) );
> +    }
> +
> +    while ( test_bit(STATE_RUN, &d->state) )
> +    {
> +        cpu_relax();
> +    }
> +    clear_bit(STATE_SCHED, &d->state);

Does all this perhaps need preemption handling? The caller
(pci_release_devices()) is direct descendant from
domain_relinquish_resources(), so even bubbling -EAGAIN or
-ERESTART back up instead of spinning would seem like an
option.

Jan


_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
http://lists.xen.org/xen-devel