Re: [PATCH RESEND] genirq: Machine-parsable version of /proc/interrupts

[Thomas Gleixner]

On Tue, 26 Jul 2016, Craig Gallek wrote:
>  /*
>   * Core internal functions to deal with irq descriptors
> @@ -92,6 +93,7 @@ struct irq_desc {
>  	int			parent_irq;
>  	struct module		*owner;
>  	const char		*name;
> +	struct kobject		kobj;

Lacks a docbook comment.

> @@ -121,6 +122,145 @@ EXPORT_SYMBOL_GPL(nr_irqs);
>  static DEFINE_MUTEX(sparse_irq_lock);
>  static DECLARE_BITMAP(allocated_irqs, IRQ_BITMAP_BITS);
>  
> +struct kobject *irq_kobj;

That wants to be static. Aside of that the variable name
sucks. irq_kobj_base or such would be self explaining.

> +#define IRQ_ATTR_RO(_name) \
> +static struct kobj_attribute _name##_attr = __ATTR_RO(_name)

All these show() functions want to be conditional on CONFIG_SYSFS

> +static ssize_t per_cpu_count_show(struct kobject *kobj,
> +				  struct kobj_attribute *attr, char *buf)
> +{
> +	struct irq_desc *desc = container_of(kobj, struct irq_desc, kobj);
> +	int i;
> +	unsigned long flags;
> +	ssize_t ret = 0;

Can you please order the variables as we do everywhere else?

+	struct irq_desc *desc = container_of(kobj, struct irq_desc, kobj);
+	unsigned long flags;
+	ssize_t ret = 0;
+	int i;

Hmm?

> +
> +	raw_spin_lock_irqsave(&desc->lock, flags);

Why would we need locking for this? The count is unsigned int, so you can
access it lockless.

> +	for_each_online_cpu(i) {
> +		unsigned int c = kstat_irqs_cpu(desc->irq_data.irq, i);

Missing new line between variable declaration and code.

> +		if (!ret)
> +			ret = sprintf(buf, "%u", c);
> +		else
> +			ret += scnprintf(buf+ret, PAGE_SIZE-ret, ",%u", c);

  			'buf + ret', PAGE_SIZE - ret,
Please.

> +	}
> +	raw_spin_unlock_irqrestore(&desc->lock, flags);
> +
> +	if (ret)

This conditional is pointless. There is always at least one cpu online
otherwise you would not be executing this function ....

> +		ret += scnprintf(buf+ret, PAGE_SIZE-ret, "\n");

And the whole thing can be simplified to:

	int cpu, irq = desc->irq_data.irq;
	ssize_t ret = 0;
	char *p = "";

	for_each_online_cpu(cpu) {
		unsigned int c = kstat_irqs_cpu(irq, cpu);

		ret += scnprintf(buf + ret , PAGE_SIZE - ret, "%s%u", p, c);
		p = ",";
	}

	ret += scnprintf(buf + ret , PAGE_SIZE - ret, "\n");
	return ret;

Hmm?

> +IRQ_ATTR_RO(per_cpu_count);
> +
> +static ssize_t chip_name_show(struct kobject *kobj,
> +			      struct kobj_attribute *attr, char *buf)
> +{
> +	struct irq_desc *desc = container_of(kobj, struct irq_desc, kobj);
> +	unsigned long flags;
> +	ssize_t ret = 0;
> +
> +	raw_spin_lock_irqsave(&desc->lock, flags);

raw_spin_lock_irq() because this is always called from non interrupt
disabled code. No need for flags.

> +	if (desc->irq_data.chip->name)

chip can be NULL .....

> +		ret = scnprintf(buf, PAGE_SIZE, "%s\n",
> +				desc->irq_data.chip->name);

If you break the line of the conditional statement, please add 
a pair of curly braces.

> +static ssize_t actions_show(struct kobject *kobj,
> +			    struct kobj_attribute *attr, char *buf)
> +{
> +	struct irq_desc *desc = container_of(kobj, struct irq_desc, kobj);
> +	struct irqaction *action;
> +	unsigned long flags;
> +	ssize_t ret = 0;
> +
> +	raw_spin_lock_irqsave(&desc->lock, flags);
> +	for (action = desc->action; action != NULL; action = action->next) {
> +		if (!ret)
> +			ret = scnprintf(buf, PAGE_SIZE, "%s", action->name);
> +		else
> +			ret += scnprintf(buf+ret, PAGE_SIZE-ret, ",%s",
> +					 action->name);
> +	}
> +	raw_spin_unlock_irqrestore(&desc->lock, flags);
> +
> +	if (ret)
> +		ret += scnprintf(buf+ret, PAGE_SIZE-ret, "\n");

Can be deuglified as well.

> @@ -261,6 +410,12 @@ static int alloc_descs(unsigned int start, unsigned int cnt, int node,
>  			goto err;
>  		mutex_lock(&sparse_irq_lock);
>  		irq_insert_desc(start + i, desc);
> +		if (irq_kobj) {
> +			if (kobject_add(&desc->kobj, irq_kobj, "%d", start + i))
> +				printk(KERN_WARNING
> +				       "Fail to add kobject for irq %d\n",
> +				       start + i);

pr_warn please

> +		}

Please split that out into a a helper function. irq_sysfs_add() or such so
you avoid that mess in the code and you can reuse that for the init stuff.

> +static void irq_kobj_release(struct kobject *kobj)
> +{
> +	struct irq_desc *desc = container_of(kobj, struct irq_desc, kobj);
> +
> +	/*
> +	 * This irq_desc is statically allocated.  Simply zero its kobject
> +	 * rather than kfree it.
> +	 */
> +	memset(&desc->kobj, 0, sizeof(desc->kobj));

This will nicely explode when the irq descriptor is handed out again before
this function is called.

There is no point to make this work for !SPARSE_IRQ. Just ignore it.

> +static int __init irq_sysfs_init(void)

Again: CONFIG_SYSFS ....

> +{
> +	int i;
> +
> +	irq_kobj = kobject_create_and_add("irq", kernel_kobj);
> +	if (!irq_kobj)
> +		return -ENOMEM;
> +
> +	irq_lock_sparse();
> +	for (i = 0; i < nr_irqs; ++i) {

  for_each_irq_desc() ....

> +		struct irq_desc *desc = irq_to_desc(i);
> +
> +		if (!desc)
> +			continue;
> +
> +		if (kobject_add(&desc->kobj, irq_kobj, "%d", i))
> +			printk(KERN_WARNING
> +			       "Fail to add kobject for irq %d\n", i);

So you managed to copy the same thing 3 times ....

Thanks,

	tglx