Re: [PATCH 1/2] xen/x86: Move irq allocation from Xen smp_op.cpu_up()

[Boris Ostrovsky <boris.ostrovsky@oracle.com>]

On 03/25/2016 11:10 AM, Konrad Rzeszutek Wilk wrote:
> On Thu, Mar 17, 2016 at 09:33:32AM -0400, Boris Ostrovsky wrote:
>> Commit ce0d3c0a6fb1 ("genirq: Revert sparse irq locking around
>> __cpu_up() and move it to x86 for now") reverted irq locking
>> introduced by commit a89941816726 ("hotplug: Prevent alloc/free
>> of irq descriptors during cpu up/down") because of Xen allocating
>> irqs in both of its cpu_up ops.
>>
>> We can move those allocations into CPU notifiers so that original
>> patch can be reinstated.
> Original being "hotplug: Prevent alloc/free..." ?

Yes.

>
> -static int xen_hvm_cpu_notify(struct notifier_block *self, unsigned long action,
> -			      void *hcpu)
> +static int xen_cpu_notify(struct notifier_block *self, unsigned long action,
> +			  void *hcpu)
>   {
>   	int cpu = (long)hcpu;
> +	int rc;
> +
>   	switch (action) {
>   	case CPU_UP_PREPARE:
> -		xen_vcpu_setup(cpu);
> -		if (xen_have_vector_callback) {
> -			if (xen_feature(XENFEAT_hvm_safe_pvclock))
> -				xen_setup_timer(cpu);
> +		if (xen_hvm_domain()) {
> +			/*
> +			 * This can happen if CPU was offlined earlier and
> +			 * offlining timed out in common_cpu_die().
> +			 */
> +			if (cpu_report_state(cpu) == CPU_DEAD_FROZEN) {
> +				xen_smp_intr_free(cpu);
> +				xen_uninit_lock_cpu(cpu);
> +			}
> +
> +			xen_vcpu_setup(cpu);
>   		}
> +
> +		if (xen_pv_domain() ||
> +		    (xen_have_vector_callback &&
> +		     xen_feature(XENFEAT_hvm_safe_pvclock)))
> +			xen_setup_timer(cpu);
> +
> +		rc = xen_smp_intr_init(cpu);
> +		if (rc) {
> +			WARN(1, "xen_smp_intr_init() for CPU %d failed: %d\n",
> +			     cpu, rc);
> +			return NOTIFY_BAD;
> +		}
> +
> +		break;
> +	case CPU_ONLINE:
> +		xen_init_lock_cpu(cpu);
> +		break;
> +	case CPU_UP_CANCELED:
> +		xen_smp_intr_free(cpu);
> xen_uninit_lock_cpu ?

I don't think this is needed: we initialize lock in CPU_ONLINE notifier 
which can only be called after CPU_UP_CANCELED would have run (in which 
case we'll never do CPU_ONLINE)

-boris

>
>
>> +		if (xen_pv_domain() ||
>> +		    (xen_have_vector_callback &&
>> +		     xen_feature(XENFEAT_hvm_safe_pvclock)))
>> +			xen_teardown_timer(cpu);
>>   		break;
>>   	default:
>>   		break;