linux/Documentation/IRQ-domain.txt
<<
>>
Prefs
   1irq_domain interrupt number mapping library
   2
   3The current design of the Linux kernel uses a single large number
   4space where each separate IRQ source is assigned a different number.
   5This is simple when there is only one interrupt controller, but in
   6systems with multiple interrupt controllers the kernel must ensure
   7that each one gets assigned non-overlapping allocations of Linux
   8IRQ numbers.
   9
  10The number of interrupt controllers registered as unique irqchips
  11show a rising tendency: for example subdrivers of different kinds
  12such as GPIO controllers avoid reimplementing identical callback
  13mechanisms as the IRQ core system by modelling their interrupt
  14handlers as irqchips, i.e. in effect cascading interrupt controllers.
  15
  16Here the interrupt number loose all kind of correspondence to
  17hardware interrupt numbers: whereas in the past, IRQ numbers could
  18be chosen so they matched the hardware IRQ line into the root
  19interrupt controller (i.e. the component actually fireing the
  20interrupt line to the CPU) nowadays this number is just a number.
  21
  22For this reason we need a mechanism to separate controller-local
  23interrupt numbers, called hardware irq's, from Linux IRQ numbers.
  24
  25The irq_alloc_desc*() and irq_free_desc*() APIs provide allocation of
  26irq numbers, but they don't provide any support for reverse mapping of
  27the controller-local IRQ (hwirq) number into the Linux IRQ number
  28space.
  29
  30The irq_domain library adds mapping between hwirq and IRQ numbers on
  31top of the irq_alloc_desc*() API.  An irq_domain to manage mapping is
  32preferred over interrupt controller drivers open coding their own
  33reverse mapping scheme.
  34
  35irq_domain also implements translation from an abstract irq_fwspec
  36structure to hwirq numbers (Device Tree and ACPI GSI so far), and can
  37be easily extended to support other IRQ topology data sources.
  38
  39=== irq_domain usage ===
  40An interrupt controller driver creates and registers an irq_domain by
  41calling one of the irq_domain_add_*() functions (each mapping method
  42has a different allocator function, more on that later).  The function
  43will return a pointer to the irq_domain on success.  The caller must
  44provide the allocator function with an irq_domain_ops structure.
  45
  46In most cases, the irq_domain will begin empty without any mappings
  47between hwirq and IRQ numbers.  Mappings are added to the irq_domain
  48by calling irq_create_mapping() which accepts the irq_domain and a
  49hwirq number as arguments.  If a mapping for the hwirq doesn't already
  50exist then it will allocate a new Linux irq_desc, associate it with
  51the hwirq, and call the .map() callback so the driver can perform any
  52required hardware setup.
  53
  54When an interrupt is received, irq_find_mapping() function should
  55be used to find the Linux IRQ number from the hwirq number.
  56
  57The irq_create_mapping() function must be called *atleast once*
  58before any call to irq_find_mapping(), lest the descriptor will not
  59be allocated.
  60
  61If the driver has the Linux IRQ number or the irq_data pointer, and
  62needs to know the associated hwirq number (such as in the irq_chip
  63callbacks) then it can be directly obtained from irq_data->hwirq.
  64
  65=== Types of irq_domain mappings ===
  66There are several mechanisms available for reverse mapping from hwirq
  67to Linux irq, and each mechanism uses a different allocation function.
  68Which reverse map type should be used depends on the use case.  Each
  69of the reverse map types are described below:
  70
  71==== Linear ====
  72irq_domain_add_linear()
  73irq_domain_create_linear()
  74
  75The linear reverse map maintains a fixed size table indexed by the
  76hwirq number.  When a hwirq is mapped, an irq_desc is allocated for
  77the hwirq, and the IRQ number is stored in the table.
  78
  79The Linear map is a good choice when the maximum number of hwirqs is
  80fixed and a relatively small number (~ < 256).  The advantages of this
  81map are fixed time lookup for IRQ numbers, and irq_descs are only
  82allocated for in-use IRQs.  The disadvantage is that the table must be
  83as large as the largest possible hwirq number.
  84
  85irq_domain_add_linear() and irq_domain_create_linear() are functionally
  86equivalent, except for the first argument is different - the former
  87accepts an Open Firmware specific 'struct device_node', while the latter
  88accepts a more general abstraction 'struct fwnode_handle'.
  89
  90The majority of drivers should use the linear map.
  91
  92==== Tree ====
  93irq_domain_add_tree()
  94irq_domain_create_tree()
  95
  96The irq_domain maintains a radix tree map from hwirq numbers to Linux
  97IRQs.  When an hwirq is mapped, an irq_desc is allocated and the
  98hwirq is used as the lookup key for the radix tree.
  99
 100The tree map is a good choice if the hwirq number can be very large
 101since it doesn't need to allocate a table as large as the largest
 102hwirq number.  The disadvantage is that hwirq to IRQ number lookup is
 103dependent on how many entries are in the table.
 104
 105irq_domain_add_tree() and irq_domain_create_tree() are functionally
 106equivalent, except for the first argument is different - the former
 107accepts an Open Firmware specific 'struct device_node', while the latter
 108accepts a more general abstraction 'struct fwnode_handle'.
 109
 110Very few drivers should need this mapping.
 111
 112==== No Map ===-
 113irq_domain_add_nomap()
 114
 115The No Map mapping is to be used when the hwirq number is
 116programmable in the hardware.  In this case it is best to program the
 117Linux IRQ number into the hardware itself so that no mapping is
 118required.  Calling irq_create_direct_mapping() will allocate a Linux
 119IRQ number and call the .map() callback so that driver can program the
 120Linux IRQ number into the hardware.
 121
 122Most drivers cannot use this mapping.
 123
 124==== Legacy ====
 125irq_domain_add_simple()
 126irq_domain_add_legacy()
 127irq_domain_add_legacy_isa()
 128
 129The Legacy mapping is a special case for drivers that already have a
 130range of irq_descs allocated for the hwirqs.  It is used when the
 131driver cannot be immediately converted to use the linear mapping.  For
 132example, many embedded system board support files use a set of #defines
 133for IRQ numbers that are passed to struct device registrations.  In that
 134case the Linux IRQ numbers cannot be dynamically assigned and the legacy
 135mapping should be used.
 136
 137The legacy map assumes a contiguous range of IRQ numbers has already
 138been allocated for the controller and that the IRQ number can be
 139calculated by adding a fixed offset to the hwirq number, and
 140visa-versa.  The disadvantage is that it requires the interrupt
 141controller to manage IRQ allocations and it requires an irq_desc to be
 142allocated for every hwirq, even if it is unused.
 143
 144The legacy map should only be used if fixed IRQ mappings must be
 145supported.  For example, ISA controllers would use the legacy map for
 146mapping Linux IRQs 0-15 so that existing ISA drivers get the correct IRQ
 147numbers.
 148
 149Most users of legacy mappings should use irq_domain_add_simple() which
 150will use a legacy domain only if an IRQ range is supplied by the
 151system and will otherwise use a linear domain mapping. The semantics
 152of this call are such that if an IRQ range is specified then
 153descriptors will be allocated on-the-fly for it, and if no range is
 154specified it will fall through to irq_domain_add_linear() which means
 155*no* irq descriptors will be allocated.
 156
 157A typical use case for simple domains is where an irqchip provider
 158is supporting both dynamic and static IRQ assignments.
 159
 160In order to avoid ending up in a situation where a linear domain is
 161used and no descriptor gets allocated it is very important to make sure
 162that the driver using the simple domain call irq_create_mapping()
 163before any irq_find_mapping() since the latter will actually work
 164for the static IRQ assignment case.
 165
 166==== Hierarchy IRQ domain ====
 167On some architectures, there may be multiple interrupt controllers
 168involved in delivering an interrupt from the device to the target CPU.
 169Let's look at a typical interrupt delivering path on x86 platforms:
 170
 171Device --> IOAPIC -> Interrupt remapping Controller -> Local APIC -> CPU
 172
 173There are three interrupt controllers involved:
 1741) IOAPIC controller
 1752) Interrupt remapping controller
 1763) Local APIC controller
 177
 178To support such a hardware topology and make software architecture match
 179hardware architecture, an irq_domain data structure is built for each
 180interrupt controller and those irq_domains are organized into hierarchy.
 181When building irq_domain hierarchy, the irq_domain near to the device is
 182child and the irq_domain near to CPU is parent. So a hierarchy structure
 183as below will be built for the example above.
 184        CPU Vector irq_domain (root irq_domain to manage CPU vectors)
 185                ^
 186                |
 187        Interrupt Remapping irq_domain (manage irq_remapping entries)
 188                ^
 189                |
 190        IOAPIC irq_domain (manage IOAPIC delivery entries/pins)
 191
 192There are four major interfaces to use hierarchy irq_domain:
 1931) irq_domain_alloc_irqs(): allocate IRQ descriptors and interrupt
 194   controller related resources to deliver these interrupts.
 1952) irq_domain_free_irqs(): free IRQ descriptors and interrupt controller
 196   related resources associated with these interrupts.
 1973) irq_domain_activate_irq(): activate interrupt controller hardware to
 198   deliver the interrupt.
 1994) irq_domain_deactivate_irq(): deactivate interrupt controller hardware
 200   to stop delivering the interrupt.
 201
 202Following changes are needed to support hierarchy irq_domain.
 2031) a new field 'parent' is added to struct irq_domain; it's used to
 204   maintain irq_domain hierarchy information.
 2052) a new field 'parent_data' is added to struct irq_data; it's used to
 206   build hierarchy irq_data to match hierarchy irq_domains. The irq_data
 207   is used to store irq_domain pointer and hardware irq number.
 2083) new callbacks are added to struct irq_domain_ops to support hierarchy
 209   irq_domain operations.
 210
 211With support of hierarchy irq_domain and hierarchy irq_data ready, an
 212irq_domain structure is built for each interrupt controller, and an
 213irq_data structure is allocated for each irq_domain associated with an
 214IRQ. Now we could go one step further to support stacked(hierarchy)
 215irq_chip. That is, an irq_chip is associated with each irq_data along
 216the hierarchy. A child irq_chip may implement a required action by
 217itself or by cooperating with its parent irq_chip.
 218
 219With stacked irq_chip, interrupt controller driver only needs to deal
 220with the hardware managed by itself and may ask for services from its
 221parent irq_chip when needed. So we could achieve a much cleaner
 222software architecture.
 223
 224For an interrupt controller driver to support hierarchy irq_domain, it
 225needs to:
 2261) Implement irq_domain_ops.alloc and irq_domain_ops.free
 2272) Optionally implement irq_domain_ops.activate and
 228   irq_domain_ops.deactivate.
 2293) Optionally implement an irq_chip to manage the interrupt controller
 230   hardware.
 2314) No need to implement irq_domain_ops.map and irq_domain_ops.unmap,
 232   they are unused with hierarchy irq_domain.
 233
 234Hierarchy irq_domain may also be used to support other architectures,
 235such as ARM, ARM64 etc.
 236