1=============================================== 2The irq_domain interrupt number mapping library 3=============================================== 4 5The current design of the Linux kernel uses a single large number 6space where each separate IRQ source is assigned a different number. 7This is simple when there is only one interrupt controller, but in 8systems with multiple interrupt controllers the kernel must ensure 9that each one gets assigned non-overlapping allocations of Linux 10IRQ numbers. 11 12The number of interrupt controllers registered as unique irqchips 13show a rising tendency: for example subdrivers of different kinds 14such as GPIO controllers avoid reimplementing identical callback 15mechanisms as the IRQ core system by modelling their interrupt 16handlers as irqchips, i.e. in effect cascading interrupt controllers. 17 18Here the interrupt number loose all kind of correspondence to 19hardware interrupt numbers: whereas in the past, IRQ numbers could 20be chosen so they matched the hardware IRQ line into the root 21interrupt controller (i.e. the component actually fireing the 22interrupt line to the CPU) nowadays this number is just a number. 23 24For this reason we need a mechanism to separate controller-local 25interrupt numbers, called hardware irq's, from Linux IRQ numbers. 26 27The irq_alloc_desc*() and irq_free_desc*() APIs provide allocation of 28irq numbers, but they don't provide any support for reverse mapping of 29the controller-local IRQ (hwirq) number into the Linux IRQ number 30space. 31 32The irq_domain library adds mapping between hwirq and IRQ numbers on 33top of the irq_alloc_desc*() API. An irq_domain to manage mapping is 34preferred over interrupt controller drivers open coding their own 35reverse mapping scheme. 36 37irq_domain also implements translation from an abstract irq_fwspec 38structure to hwirq numbers (Device Tree and ACPI GSI so far), and can 39be easily extended to support other IRQ topology data sources. 40 41irq_domain usage 42================ 43 44An interrupt controller driver creates and registers an irq_domain by 45calling one of the irq_domain_add_*() or irq_domain_create_*() functions 46(each mapping method has a different allocator function, more on that later). 47The function will return a pointer to the irq_domain on success. The caller 48must provide the allocator function with an irq_domain_ops structure. 49 50In most cases, the irq_domain will begin empty without any mappings 51between hwirq and IRQ numbers. Mappings are added to the irq_domain 52by calling irq_create_mapping() which accepts the irq_domain and a 53hwirq number as arguments. If a mapping for the hwirq doesn't already 54exist then it will allocate a new Linux irq_desc, associate it with 55the hwirq, and call the .map() callback so the driver can perform any 56required hardware setup. 57 58Once a mapping has been established, it can be retrieved or used via a 59variety of methods: 60 61- irq_resolve_mapping() returns a pointer to the irq_desc structure 62 for a given domain and hwirq number, and NULL if there was no 63 mapping. 64- irq_find_mapping() returns a Linux IRQ number for a given domain and 65 hwirq number, and 0 if there was no mapping 66- irq_linear_revmap() is now identical to irq_find_mapping(), and is 67 deprecated 68- generic_handle_domain_irq() handles an interrupt described by a 69 domain and a hwirq number 70- handle_domain_irq() does the same thing for root interrupt 71 controllers and deals with the set_irq_reg()/irq_enter() sequences 72 that most architecture requires 73 74Note that irq domain lookups must happen in contexts that are 75compatible with a RCU read-side critical section. 76 77The irq_create_mapping() function must be called *atleast once* 78before any call to irq_find_mapping(), lest the descriptor will not 79be allocated. 80 81If the driver has the Linux IRQ number or the irq_data pointer, and 82needs to know the associated hwirq number (such as in the irq_chip 83callbacks) then it can be directly obtained from irq_data->hwirq. 84 85Types of irq_domain mappings 86============================ 87 88There are several mechanisms available for reverse mapping from hwirq 89to Linux irq, and each mechanism uses a different allocation function. 90Which reverse map type should be used depends on the use case. Each 91of the reverse map types are described below: 92 93Linear 94------ 95 96:: 97 98 irq_domain_add_linear() 99 irq_domain_create_linear() 100 101The linear reverse map maintains a fixed size table indexed by the 102hwirq number. When a hwirq is mapped, an irq_desc is allocated for 103the hwirq, and the IRQ number is stored in the table. 104 105The Linear map is a good choice when the maximum number of hwirqs is 106fixed and a relatively small number (~ < 256). The advantages of this 107map are fixed time lookup for IRQ numbers, and irq_descs are only 108allocated for in-use IRQs. The disadvantage is that the table must be 109as large as the largest possible hwirq number. 110 111irq_domain_add_linear() and irq_domain_create_linear() are functionally 112equivalent, except for the first argument is different - the former 113accepts an Open Firmware specific 'struct device_node', while the latter 114accepts a more general abstraction 'struct fwnode_handle'. 115 116The majority of drivers should use the linear map. 117 118Tree 119---- 120 121:: 122 123 irq_domain_add_tree() 124 irq_domain_create_tree() 125 126The irq_domain maintains a radix tree map from hwirq numbers to Linux 127IRQs. When an hwirq is mapped, an irq_desc is allocated and the 128hwirq is used as the lookup key for the radix tree. 129 130The tree map is a good choice if the hwirq number can be very large 131since it doesn't need to allocate a table as large as the largest 132hwirq number. The disadvantage is that hwirq to IRQ number lookup is 133dependent on how many entries are in the table. 134 135irq_domain_add_tree() and irq_domain_create_tree() are functionally 136equivalent, except for the first argument is different - the former 137accepts an Open Firmware specific 'struct device_node', while the latter 138accepts a more general abstraction 'struct fwnode_handle'. 139 140Very few drivers should need this mapping. 141 142No Map 143------ 144 145:: 146 147 irq_domain_add_nomap() 148 149The No Map mapping is to be used when the hwirq number is 150programmable in the hardware. In this case it is best to program the 151Linux IRQ number into the hardware itself so that no mapping is 152required. Calling irq_create_direct_mapping() will allocate a Linux 153IRQ number and call the .map() callback so that driver can program the 154Linux IRQ number into the hardware. 155 156Most drivers cannot use this mapping, and it is now gated on the 157CONFIG_IRQ_DOMAIN_NOMAP option. Please refrain from introducing new 158users of this API. 159 160Legacy 161------ 162 163:: 164 165 irq_domain_add_simple() 166 irq_domain_add_legacy() 167 irq_domain_create_simple() 168 irq_domain_create_legacy() 169 170The Legacy mapping is a special case for drivers that already have a 171range of irq_descs allocated for the hwirqs. It is used when the 172driver cannot be immediately converted to use the linear mapping. For 173example, many embedded system board support files use a set of #defines 174for IRQ numbers that are passed to struct device registrations. In that 175case the Linux IRQ numbers cannot be dynamically assigned and the legacy 176mapping should be used. 177 178As the name implies, the \*_legacy() functions are deprecated and only 179exist to ease the support of ancient platforms. No new users should be 180added. Same goes for the \*_simple() functions when their use results 181in the legacy behaviour. 182 183The legacy map assumes a contiguous range of IRQ numbers has already 184been allocated for the controller and that the IRQ number can be 185calculated by adding a fixed offset to the hwirq number, and 186visa-versa. The disadvantage is that it requires the interrupt 187controller to manage IRQ allocations and it requires an irq_desc to be 188allocated for every hwirq, even if it is unused. 189 190The legacy map should only be used if fixed IRQ mappings must be 191supported. For example, ISA controllers would use the legacy map for 192mapping Linux IRQs 0-15 so that existing ISA drivers get the correct IRQ 193numbers. 194 195Most users of legacy mappings should use irq_domain_add_simple() or 196irq_domain_create_simple() which will use a legacy domain only if an IRQ range 197is supplied by the system and will otherwise use a linear domain mapping. 198The semantics of this call are such that if an IRQ range is specified then 199descriptors will be allocated on-the-fly for it, and if no range is 200specified it will fall through to irq_domain_add_linear() or 201irq_domain_create_linear() which means *no* irq descriptors will be allocated. 202 203A typical use case for simple domains is where an irqchip provider 204is supporting both dynamic and static IRQ assignments. 205 206In order to avoid ending up in a situation where a linear domain is 207used and no descriptor gets allocated it is very important to make sure 208that the driver using the simple domain call irq_create_mapping() 209before any irq_find_mapping() since the latter will actually work 210for the static IRQ assignment case. 211 212irq_domain_add_simple() and irq_domain_create_simple() as well as 213irq_domain_add_legacy() and irq_domain_create_legacy() are functionally 214equivalent, except for the first argument is different - the former 215accepts an Open Firmware specific 'struct device_node', while the latter 216accepts a more general abstraction 'struct fwnode_handle'. 217 218Hierarchy IRQ domain 219-------------------- 220 221On some architectures, there may be multiple interrupt controllers 222involved in delivering an interrupt from the device to the target CPU. 223Let's look at a typical interrupt delivering path on x86 platforms:: 224 225 Device --> IOAPIC -> Interrupt remapping Controller -> Local APIC -> CPU 226 227There are three interrupt controllers involved: 228 2291) IOAPIC controller 2302) Interrupt remapping controller 2313) Local APIC controller 232 233To support such a hardware topology and make software architecture match 234hardware architecture, an irq_domain data structure is built for each 235interrupt controller and those irq_domains are organized into hierarchy. 236When building irq_domain hierarchy, the irq_domain near to the device is 237child and the irq_domain near to CPU is parent. So a hierarchy structure 238as below will be built for the example above:: 239 240 CPU Vector irq_domain (root irq_domain to manage CPU vectors) 241 ^ 242 | 243 Interrupt Remapping irq_domain (manage irq_remapping entries) 244 ^ 245 | 246 IOAPIC irq_domain (manage IOAPIC delivery entries/pins) 247 248There are four major interfaces to use hierarchy irq_domain: 249 2501) irq_domain_alloc_irqs(): allocate IRQ descriptors and interrupt 251 controller related resources to deliver these interrupts. 2522) irq_domain_free_irqs(): free IRQ descriptors and interrupt controller 253 related resources associated with these interrupts. 2543) irq_domain_activate_irq(): activate interrupt controller hardware to 255 deliver the interrupt. 2564) irq_domain_deactivate_irq(): deactivate interrupt controller hardware 257 to stop delivering the interrupt. 258 259Following changes are needed to support hierarchy irq_domain: 260 2611) a new field 'parent' is added to struct irq_domain; it's used to 262 maintain irq_domain hierarchy information. 2632) a new field 'parent_data' is added to struct irq_data; it's used to 264 build hierarchy irq_data to match hierarchy irq_domains. The irq_data 265 is used to store irq_domain pointer and hardware irq number. 2663) new callbacks are added to struct irq_domain_ops to support hierarchy 267 irq_domain operations. 268 269With support of hierarchy irq_domain and hierarchy irq_data ready, an 270irq_domain structure is built for each interrupt controller, and an 271irq_data structure is allocated for each irq_domain associated with an 272IRQ. Now we could go one step further to support stacked(hierarchy) 273irq_chip. That is, an irq_chip is associated with each irq_data along 274the hierarchy. A child irq_chip may implement a required action by 275itself or by cooperating with its parent irq_chip. 276 277With stacked irq_chip, interrupt controller driver only needs to deal 278with the hardware managed by itself and may ask for services from its 279parent irq_chip when needed. So we could achieve a much cleaner 280software architecture. 281 282For an interrupt controller driver to support hierarchy irq_domain, it 283needs to: 284 2851) Implement irq_domain_ops.alloc and irq_domain_ops.free 2862) Optionally implement irq_domain_ops.activate and 287 irq_domain_ops.deactivate. 2883) Optionally implement an irq_chip to manage the interrupt controller 289 hardware. 2904) No need to implement irq_domain_ops.map and irq_domain_ops.unmap, 291 they are unused with hierarchy irq_domain. 292 293Hierarchy irq_domain is in no way x86 specific, and is heavily used to 294support other architectures, such as ARM, ARM64 etc. 295 296Debugging 297========= 298 299Most of the internals of the IRQ subsystem are exposed in debugfs by 300turning CONFIG_GENERIC_IRQ_DEBUGFS on. 301