Re: [PATCH 3/5] of/irq: introduce of_irq_init

[Rob Herring <robherring2-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org>]

Grant,

On 09/17/2011 06:53 PM, Grant Likely wrote:
> On Wed, Sep 14, 2011 at 11:31:38AM -0500, Rob Herring wrote:
>> From: Rob Herring <rob.herring-bsGFqQB8/DxBDgjK7y7TUQ@public.gmane.org>
>>
>> of_irq_init will scan the devicetree for matching interrupt controller
>> nodes. Then it calls an initialization function for each found controller
>> in the proper order with parent nodes initialized before child nodes.
>>
>> Based on initial pseudo code from Grant Likely.
>>
>> Signed-off-by: Rob Herring <rob.herring-bsGFqQB8/DxBDgjK7y7TUQ@public.gmane.org>
>> Cc: Grant Likely <grant.likely-s3s/WqlpOiPyB63q8FvJNQ@public.gmane.org>
>> ---
>>  drivers/of/irq.c       |   96 ++++++++++++++++++++++++++++++++++++++++++++++++
>>  include/linux/of_irq.h |    1 +
>>  2 files changed, 97 insertions(+), 0 deletions(-)
>>
>> diff --git a/drivers/of/irq.c b/drivers/of/irq.c
>> index 9f689f1..a0cd7e8 100644
>> --- a/drivers/of/irq.c
>> +++ b/drivers/of/irq.c
>> @@ -19,10 +19,13 @@
>>   */
>>  
>>  #include <linux/errno.h>
>> +#include <linux/list.h>
>> +#include <linux/list_sort.h>
>>  #include <linux/module.h>
>>  #include <linux/of.h>
>>  #include <linux/of_irq.h>
>>  #include <linux/string.h>
>> +#include <linux/slab.h>
>>  
>>  /* For archs that don't support NO_IRQ (such as x86), provide a dummy value */
>>  #ifndef NO_IRQ
>> @@ -386,3 +389,96 @@ int of_irq_to_resource_table(struct device_node *dev, struct resource *res,
>>  
>>  	return i;
>>  }
>> +
>> +struct intc_desc {
>> +	struct list_head	list;
>> +	struct device_node	*dev;
>> +	struct device_node	*parent;
>> +};
>> +
>> +typedef void (*irq_init_cb_t)(struct device_node *, struct device_node *);
>> +
>> +static int __init irq_cmp_intc_desc(void *unused, struct list_head *a,
>> +				    struct list_head *b)
>> +{
>> +	const struct intc_desc *da = list_entry(a, typeof(*da), list);
>> +	const struct intc_desc *db = list_entry(b, typeof(*db), list);
>> +
>> +	/* same parent, so order doesn't matter */
>> +	if (da->parent == db->parent)
>> +		return 0;
>> +
>> +	/* NULL parent comes first */
>> +	if (!da->parent && db->parent)
>> +		return -1;
>> +	if (!db->parent && da->parent)
>> +		return 1;
>> +
>> +	/* parent node must be before child node */
>> +	if (da->dev == db->parent)
>> +		return -1;
>> +	if (db->dev == da->parent)
>> +		return 1;
> 
> Does sort_list work for relationships 4 or more levels deep?  ie. if
> there was a relationship of A <- B <- C <- D, then B compared with D
> would return 0 from this function which could potentially result in an
> incorrectly ordered list.
> 

Doh! Um, 3 levels is enough for everyone!? ;)

> The other option for implementing this would be to take the probe
> deferral approach and not try to sort the list, but instead allow
> probe functions to fail & request retry if the parent hasn't yet been
> probed.  I haven't thought enough about it though to say which would
> be the best approach.
> 

Considering the list will typically be only a few entries, it is
probably not so important how efficiently we sort or walk the list.

The only way I see controller code knowing if it needs to defer init is
if of_irq_create_mapping fails. The core code could simply do this
itself. However, I would imagine sorting it would be faster than that path.

How about something like this (untested):

int find_order(struct intc_desc *node)
{
	struct intc_desc *d;

	list_for_each_entry(d, &intc_desc_list, list) {
		if (node->parent != d->dev)
			continue;

		if (d->order < 0)
			find_order(d);

		node->order = d->order + 1;
		break;
	}
}


Then rather than sorting, do this:


	list_for_each_entry(desc, &intc_desc_list, list)
		find_order(desc);

	for (order = 0; !list_empty(&intc_desc_list); order++) {
		list_for_each_entry_safe(desc, temp_desc, &intc_desc_list, list) {
			if (desc->order != order)
				continue;
			
			match = of_match_node(matches, desc->dev);
			if (match && match->data) {
				irq_init_cb_t irq_init_cb = match->data;
				pr_debug("of_irq_init: init %s @ %p, parent %p\n",
					 match->compatible, desc->dev, desc->parent);
				irq_init_cb(desc->dev, desc->parent);
			}
			list_del(&desc->list);
			kfree(desc);
		}
	}


>> +
>> +	return 0;
>> +}
>> +
>> +/**
>> + * of_irq_init - Scan the device tree for matching interrupt controllers and
>> + * call their initialization functions in order with parents first.
>> + * @matches: 0 terminated array of nodes to match and initialization function
>> + * to call on match
>> + */
>> +void __init of_irq_init(const struct of_device_id *matches)
>> +{
>> +	struct device_node *np;
>> +	const struct of_device_id *match;
>> +	struct intc_desc *desc;
>> +	struct intc_desc *temp_desc;
>> +	struct list_head intc_desc_list;
>> +
>> +	INIT_LIST_HEAD(&intc_desc_list);
>> +
>> +	for_each_matching_node(np, matches) {
>> +		if (!of_find_property(np, "interrupt-controller", NULL))
>> +			continue;
>> +		/* Here, we allocate and populate an intc_desc with the node
>> +		* pointer, interrupt-parent device_node etc. */
>> +		desc = kzalloc(sizeof(*desc), GFP_KERNEL);
>> +		if (!desc) {
>> +			WARN_ON(1);
>> +			goto err;
>> +		}
>> +		desc->dev = np;
>> +		desc->parent = of_irq_find_parent(np);
>> +		list_add(&desc->list, &intc_desc_list);
>> +	}
>> +	if (list_empty(&intc_desc_list))
>> +		return;
>> +
>> +	/*
>> +	 * The root irq controller is the one without an interrupt-parent.
>> +	 * That one goes first, followed by the controllers that reference it,
>> +	 * followed by the ones that reference the 2nd level controllers, etc
>> +	 */
> 
> I don't believe that this actually turns out to be true (and yes I
> know it is how I originally described it).  :-)  When the
> interrupt-parent property is at the root of the tree, then the root
> interrupt controller may very well inherit itself as it's interrupt
> parent, and of_irq_find_parent() will still return a value.  This
> should probably be considered a bug in of_irq_find_parent(), and it
> should return NULL if the parent is itself.

I did hit this exact issue. There is an easy, but not obvious fix to the
device tree. Simply adding "interupt-parent;" to the root interrupt
controller node will do the trick and override the value in the tree root.

> 
> of_irq_find_parent should probably be implemented thusly (completely
> untested); although the only functional change is the line:
> 	return (p == child) ? NULL : p;
> 
> /**
>  * of_irq_find_parent - Given a device node, find its interrupt parent node
>  * @child: pointer to device node
>  *
>  * Returns a pointer to the interrupt parent node, or NULL if the
>  * interrupt parent could not be determined.
>  */
> struct device_node *of_irq_find_parent(struct device_node *child)
> {
> 	struct device_node *p, *c = child;
> 	const __be32 *parp;
> 
> 	if (!of_node_get(c))
> 		return NULL;
> 
> 	do {
> 		p = of_parse_phandle(c, "interrupt-parent", 0);
> 
> 		if (!p && (of_irq_workarounds & OF_IMAP_NO_PHANDLE) &&
> 		    of_find_property(c, "interrupt-parent", NULL))
> 			p = of_node_get(of_irq_dflt_pic);
> 
> 		if (!p)
> 			p = of_get_parent(c);
> 
> 		of_node_put(c);
> 		c = p;
> 	} while (p && !of_find_property(p, "#interrupt-cells", NULL));
> 
> 	return (p == child) ? NULL : p;
> }
> 

This change should probably be implemented as well as this is likely a
common occurrence that will be stumbled over or existing device trees
won't have this. I'll test and add to the next series.

Rob