On 03/08/2012 09:57 PM, Grant Likely wrote:
On Sat, 03 Mar 2012 21:09:32 -0800, David Daney<firstname.lastname@example.org> wrote:
On 03/03/2012 11:35 AM, Rob Herring wrote:
On 03/02/2012 01:29 PM, David Daney wrote:
On 03/02/2012 11:07 AM, Grant Likely wrote:
+static void __init octeon_irq_set_ciu_mapping(unsigned int irq,
+ unsigned int line,
+ unsigned int bit,
+ struct irq_domain *domain,
struct irq_chip *chip,
+ struct irq_data *irqd;
union octeon_ciu_chip_data cd;
irq_set_chip_and_handler(irq, chip, handler);
cd.l = 0;
cd.s.line = line;
cd.s.bit = bit;
octeon_irq_ciu_to_irq[line][bit] = irq;
+ irqd = irq_get_irq_data(irq);
+ irqd->hwirq = line<< 6 | bit;
+ irqd->domain = domain;
I think the domain code will set these.
It is my understanding that the domain code only does this for:
We use none of those. So I do it here.
If there is a better way, I am open to suggestions.
irq_create_mapping is called by irq_create_of_mapping() which is
in turn called by irq_of_parse_and-map(). irq_domain always
manages the hwirq and domain values. Driver code cannot manipulate
I really must be missing something.
1) I must have a mapping between hwirq and irq that I control so that
non-OF code using the OCTEON_IRQ_* constants continues to work.
Those defines are what you need to work to get rid of.
We are not starting from a blank slate here. There is a lot of in-tree
code using these symbols. We cannot make them disappear with wishful
The first step is a switch to irq_domains using the existing mappings.
After we do that, I have patches to transition some drivers to use the
OF mapping via irq_domains. After those are merged, we can work toward
getting rid of OCTEON_IRQ_*. But I think it must be the last step in
the process, not the first.
2) irq_create_mapping() will allocate a random irq value if none is
already assigned to the hwirq.
Therefore: To avoid having random irq values assigned, I must manually
So you should be using legacy domain if you need to maintain fixed hwirq
to linux irq numbers. "linear" is a bit confusing as it doesn't mean
linear 1:1 irq number assignment, but linear search.
My reading of Grant's code in linux-next directly contradicts this
statement. There is no code in irqdomain.c, that I can see, that allows
me to have an arbitrary mapping of irq<--> hwirq values.
There are 4 kinds of mappings available; legacy, linear, radix and nomap.
Yes, I had discovered that.
Ignore nomap and radix; you don't want them.
legacy maps a contiguous range of hwirq numbers to a contiguous range of
linux irq numbers. To preserve the exising #define mappings but still add
DT support, this is the one that you want.
This is precisely the point that you and Rob seem to have missed in the
last three or four back-and-forths about this.
Probably I have not explained well enough why legacy will not work.
We have three different interrupt controllers (although only one is
currently in-tree). hwirq to irq mapping for them is more or less as
irq hwirqCIU hwirqCIU2 hwirqCIU3
OCTEON_IRQ_USB0 56 81 934562
OCTEON_IRQ_TWSI 45 224 100543
OCTEON_IRQ_UART0 34 228 4572
Now what we notice here is that there is no possible 1:1 linearly
increasing mapping possible for the irq and *all* three hwirq sets. We
want a single binary that contains support for all three interrupt
controllers, so the OCTEON_IRQ_* values have to be the same for all
three interrupt controllers. Because of this, legacy mapping is
Since the possible ranges of the hwirq values is very large and quite
sparse, probably the radix mapping will be required.
Also to support non-OF drivers and architecture specific code for the
near future, I really think the existing IRQ values *must* be preserved.
Therefore, as I said above, we need a way for my SOC/board code to
specify the mapping.
Perhaps we need to add an optional function to struct irq_domain_ops
that would allow the default mapping to be overridden on a per
Otherwise, I think I will have to keep poking into the internal
irq_domain data structures to get the mappings I want.
What do you think?