Wherever possible, replace constructs that match either
generic_handle_irq(irq_find_mapping()) or
generic_handle_irq(irq_linear_revmap()) to a single call to
generic_handle_domain_irq().

Reviewed-by: Geert Uytterhoeven <geert+renesas@glider.be>
Acked-by: Krzysztof Kozlowski <krzysztof.kozlowski@canonical.com>
Signed-off-by: Marc Zyngier <maz@kernel.org>

Despite the name, handle_domain_irq() deals with non-irqdomain
handling for the sake of a handful of legacy ARM platforms.

Move such handling into ARM's handle_IRQ(), allowing for better
code generation for everyone else. This allows us get rid of
some complexity, and to rearrange the guards on the various helpers
in a more logical way.

Signed-off-by: Marc Zyngier <maz@kernel.org>

Provide generic_handle_domain_irq() as a pendent to handle_domain_irq()
for non-root interrupt controllers

Signed-off-by: Marc Zyngier <maz@kernel.org>

Given that the nvic driver is fully irqdomain aware, there is no
reason for it to use the arch-specific handle_IRQ(), and it can
be moved over to handle_domain_irq().

Signed-off-by: Marc Zyngier <maz@kernel.org>

It appears that the comment about a NULL domain meaning anything
has always been wrong. Fix it.

Signed-off-by: Marc Zyngier <maz@kernel.org>

In order to start reaping the benefits of irq_resolve_mapping(),
start using it in __handle_domain_irq() and handle_domain_nmi().

This involves splitting generic_handle_irq() to be able to directly
provide the irq_desc.

Signed-off-by: Marc Zyngier <maz@kernel.org>

Rework irq_find_mapping() to return an both an irq_desc pointer,
optionally the virtual irq number, and rename the result to
__irq_resolve_mapping(). a new helper called irq_resolve_mapping()
is provided for code that doesn't need the virtual irq number.

irq_find_mapping() is also rewritten in terms of __irq_resolve_mapping().

Signed-off-by: Marc Zyngier <maz@kernel.org>

It is pretty odd that the radix tree uses RCU while the linear
portion doesn't, leading to potential surprises for the users,
depending on how the irqdomain has been created.

Fix this by moving the update of the linear revmap under
the mutex, and the lookup under the RCU read-side lock.

The mutex name is updated to reflect that it doesn't only
cover the radix-tree anymore.

Signed-off-by: Marc Zyngier <maz@kernel.org>

Caching a virq number in the revmap is pretty inefficient, as
it means we will need to convert it back to either an irq_data
or irq_desc to do anything with it.

It is also a bit odd, as the radix tree does cache irq_data
pointers.

Change the revmap type to be an irq_data pointer instead of
an unsigned int, and preserve the current API for now.

Signed-off-by: Marc Zyngier <maz@kernel.org>

Instead of open-coding the size computation of struct irqdomain,
use the struct_size() helper instead.

This is going to be handy as we change the type of the revmap
array.

Signed-off-by: Marc Zyngier <maz@kernel.org>

Direct mappings are completely exclusive of normal mappings, meaning
that we can refactor the code slightly so that we can get rid of
the revmap_direct_max_irq field and use the revmap_size field
instead, reducing the size of the irqdomain structure.

Signed-off-by: Marc Zyngier <maz@kernel.org>

Only a handful of old PPC systems are still using the old 'nomap'
variant of the irqdomain library. Move the associated definitions
behind a configuration option, which will allow us to make some
more radical changes.

Signed-off-by: Marc Zyngier <maz@kernel.org>

irq_linear_revmap() is supposed to be a fast path for domain
lookups, but it only exposes low-level details of the irqdomain
implementation, details which are better kept private.

The *overhead* between the two is only a function call and
a couple of tests, so it is likely that noone can show any
meaningful difference compared to the cost of taking an
interrupt.

Reimplement irq_linear_revmap() with irq_find_mapping()
in order to preserve source code compatibility, and
rename the internal field for a measure.

Signed-off-by: Marc Zyngier <maz@kernel.org>