user/sven/linux.git/include/linux/irqdomain.h, branch v3.16.3

irqdomain: make irq_linear_revmap() a fast path again

2013-06-24T13:02:41Z

Over the years, irq_linear_revmap() gained tests and checks to make sure callers were using it safely, which while important, also make it less of a fast path. After the irqdomain refactoring done recently, it is now possible to make irq_linear_revmap() a fast path again. This patch moves irq_linear_revmap() to the header file and makes it a static inline so that interrupt controller drivers using a linear mapping can decode the virq from a hwirq in just a couple of instructions. Signed-off-by: Grant Likely

irqdomain: remove irq_domain_generate_simple()

2013-06-24T13:02:41Z

Nobody calls it; remove the function Signed-off-by: Grant Likely

irqdomain: Refactor irq_domain_associate_many()

2013-06-24T13:01:42Z

Originally, irq_domain_associate_many() was designed to unwind the mapped irqs on a failure of any individual association. However, that proved to be a problem with certain IRQ controllers. Some of them only support a subset of irqs, and will fail when attempting to map a reserved IRQ. In those cases we want to map as many IRQs as possible, so instead it is better for irq_domain_associate_many() to make a best-effort attempt to map irqs, but not fail if any or all of them don't succeed. If a caller really cares about how many irqs got associated, then it should instead go back and check that all of the irqs is cares about were mapped. The original design open-coded the individual association code into the body of irq_domain_associate_many(), but with no longer needing to unwind associations, the code becomes simpler to split out irq_domain_associate() to contain the bulk of the logic, and irq_domain_associate_many() to be a simple loop wrapper. This patch also adds a new error check to the associate path to make sure it isn't called for an irq larger than the controller can handle, and adds locking so that the irq_domain_mutex is held while setting up a new association. v3: Fixup missing change to irq_domain_add_tree() v2: Fixup x86 warning. irq_domain_associate_many() no longer returns an error code, but reports errors to the printk log directly. In the majority of cases we don't actually want to fail if there is a problem, but rather log it and still try to boot the system. Signed-off-by: Grant Likely irqdomain: Fix flubbed irq_domain_associate_many refactoring commit d39046ec72, "irqdomain: Refactor irq_domain_associate_many()" was missing the following hunk which causes a boot failure on anything using irq_domain_add_tree() to allocate an irq domain. Signed-off-by: Grant Likely Cc: Michael Neuling Cc: Benjamin Herrenschmidt , Cc: Thomas Gleixner , Cc: Stephen Rothwell

irqdomain: Clean up aftermath of irq_domain refactoring

2013-06-10T10:52:09Z

After refactoring the irqdomain code, there are a number of API functions that are merely empty wrappers around core code. Drop those wrappers out of the C file and replace them with static inlines in the header. Signed-off-by: Grant Likely

irqdomain: Eliminate revmap type

2013-06-10T10:52:09Z

The NOMAP irq_domain type is only used by a handful of interrupt controllers and it unnecessarily complicates the code by adding special cases on how to look up mappings and different revmap functions are used for each type which need to validate the correct type is passed to it before performing the reverse map. Eliminating the revmap_type and making a single reverse mapping function simplifies the code. It also shouldn't be any slower than having separate revmap functions because the type of the revmap needed to be checked anyway. The linear and tree revmap types were already merged in a previous patch. This patch rolls the NOMAP or direct mapping behaviour into the same domain code making is possible for an irq domain to do any mapping type; linear, tree or direct; and that the mapping will be transparent to the interrupt controller driver. With this change, direct mappings will get stored in the linear or tree mapping for consistency. Reverse mapping from the hwirq to virq will go through the normal lookup process. However, any controller using a direct mapping can take advantage of knowing that hwirq==virq for any mapped interrupts skip doing a revmap lookup when handling IRQs. Signed-off-by: Grant Likely

irqdomain: merge linear and tree reverse mappings.

2013-06-10T10:52:09Z

Keeping them separate makes irq_domain more complex and adds a lot of code (as proven by the diffstat). Merging them simplifies the whole scheme. This change makes it so both the tree and linear methods can be used by the same irq_domain instance. If the hwirq is less than the ->linear_size, then the linear map is used to reverse map the hwirq. Otherwise the radix tree is used. The test for which map to use is no more expensive that the existing code, so the performance of fast path is preserved. It also means that complex interrupt controllers can use both the linear map and a tree in the same domain. This may be useful for an interrupt controller with a base set of core irqs and a large number of GPIOs which might be used as irqs. The linear map could cover the core irqs, and the tree used for thas irqs. The linear map could cover the core irqs, and the tree used for the gpios. v2: Drop reorganization of revmap data Signed-off-by: Grant Likely Cc: Paul Mundt Cc: Benjamin Herrenschmidt Cc: Thomas Gleixner Cc: Rob Herring

irqdomain: Add a name field

2013-06-10T10:52:09Z

This patch adds a name field to the irq_domain structure to help mere mortals understand the mappings between irq domains and virqs. It also converts a number of places that have open-coded some kind of fudging an irqdomain name to use the new field. This means a more consistent display of names in irq domain log messages and debugfs output. Signed-off-by: Grant Likely

irqdomain: Replace LEGACY mapping with LINEAR

2013-06-10T10:52:09Z

The LEGACY mapping unnecessarily complicates the irqdomain code and can easily be implemented with a linear mapping. By ripping it out and replacing it with the LINEAR mapping the object size of irqdomain.c shrinks by about 330 bytes (ARMv7) which offsets the additional allocation required by the linear map. It also makes it possible for current LEGACY map users to pre-allocate irq_descs for a subset of the hwirqs and dynamically allocate the rest as needed. Signed-off-by: Grant Likely Cc: Paul Mundt Cc: Benjamin Herrenschmidt Cc: Thomas Gleixner Cc: Rob Herring

genirq: Generic chip: Add linear irq domain support

2013-05-29T08:57:11Z

Provide infrastructure for irq chip implementations which work on linear irq domains. - Interface to allocate multiple generic chips which are associated to the irq domain. - Interface to get the generic chip pointer for a particular hardware interrupt in the domain. - irq domain mapping function to install the chip for a particular interrupt. Note: This lacks a removal function for now. [ Sebastian Hesselbarth: Mask cache and pointer math fixups ] Signed-off-by: Thomas Gleixner Cc: Thomas Petazzoni Cc: Andrew Lunn Cc: Russell King - ARM Linux Cc: Jason Cooper Cc: Arnd Bergmann Cc: Jean-Francois Moine Cc: devicetree-discuss@lists.ozlabs.org Cc: Rob Herring Cc: Jason Gunthorpe Cc: Gregory Clement Cc: Gerlando Falauto Cc: Rob Landley Acked-by: Grant Likely Cc: Maxime Ripard Cc: Ezequiel Garcia Cc: linux-arm-kernel@lists.infradead.org Cc: Sebastian Hesselbarth Link: http://lkml.kernel.org/r/20130506142539.450634298@linutronix.de Signed-off-by: Thomas Gleixner

irqdomain: Eliminate dedicated radix lookup functions

2012-07-11T15:16:00Z

In preparation to remove the slow revmap path, eliminate the public radix revmap lookup functions. This simplifies the code and makes the slowpath removal patch a lot simpler. Signed-off-by: Grant Likely Cc: Paul Mundt Cc: Benjamin Herrenschmidt Cc: Thomas Gleixner Cc: Rob Herring