1 <?xml version="1.0" encoding="UTF-8"?>
2 <!DOCTYPE book PUBLIC "-//OASIS//DTD DocBook XML V4.1.2//EN"
3 "http://www.oasis-open.org/docbook/xml/4.1.2/docbookx.dtd" []>
5 <book id="Generic-IRQ-Guide">
7 <title>Linux generic IRQ handling</title>
11 <firstname>Thomas</firstname>
12 <surname>Gleixner</surname>
15 <email>tglx@linutronix.de</email>
20 <firstname>Ingo</firstname>
21 <surname>Molnar</surname>
24 <email>mingo@elte.hu</email>
31 <year>2005-2006</year>
32 <holder>Thomas Gleixner</holder>
35 <year>2005-2006</year>
36 <holder>Ingo Molnar</holder>
41 This documentation is free software; you can redistribute
42 it and/or modify it under the terms of the GNU General Public
43 License version 2 as published by the Free Software Foundation.
47 This program is distributed in the hope that it will be
48 useful, but WITHOUT ANY WARRANTY; without even the implied
49 warranty of MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
50 See the GNU General Public License for more details.
54 You should have received a copy of the GNU General Public
55 License along with this program; if not, write to the Free
56 Software Foundation, Inc., 59 Temple Place, Suite 330, Boston,
61 For more details see the file COPYING in the source
62 distribution of Linux.
70 <title>Introduction</title>
72 The generic interrupt handling layer is designed to provide a
73 complete abstraction of interrupt handling for device drivers.
74 It is able to handle all the different types of interrupt controller
75 hardware. Device drivers use generic API functions to request, enable,
76 disable and free interrupts. The drivers do not have to know anything
77 about interrupt hardware details, so they can be used on different
78 platforms without code changes.
81 This documentation is provided to developers who want to implement
82 an interrupt subsystem based for their architecture, with the help
83 of the generic IRQ handling layer.
87 <chapter id="rationale">
88 <title>Rationale</title>
90 The original implementation of interrupt handling in Linux is using
91 the __do_IRQ() super-handler, which is able to deal with every
92 type of interrupt logic.
95 Originally, Russell King identified different types of handlers to
96 build a quite universal set for the ARM interrupt handler
97 implementation in Linux 2.5/2.6. He distinguished between:
99 <listitem><para>Level type</para></listitem>
100 <listitem><para>Edge type</para></listitem>
101 <listitem><para>Simple type</para></listitem>
103 In the SMP world of the __do_IRQ() super-handler another type
106 <listitem><para>Per CPU type</para></listitem>
110 This split implementation of highlevel IRQ handlers allows us to
111 optimize the flow of the interrupt handling for each specific
112 interrupt type. This reduces complexity in that particular codepath
113 and allows the optimized handling of a given type.
116 The original general IRQ implementation used hw_interrupt_type
117 structures and their ->ack(), ->end() [etc.] callbacks to
118 differentiate the flow control in the super-handler. This leads to
119 a mix of flow logic and lowlevel hardware logic, and it also leads
120 to unnecessary code duplication: for example in i386, there is a
121 ioapic_level_irq and a ioapic_edge_irq irq-type which share many
122 of the lowlevel details but have different flow handling.
125 A more natural abstraction is the clean separation of the
126 'irq flow' and the 'chip details'.
129 Analysing a couple of architecture's IRQ subsystem implementations
130 reveals that most of them can use a generic set of 'irq flow'
131 methods and only need to add the chip level specific code.
132 The separation is also valuable for (sub)architectures
133 which need specific quirks in the irq flow itself but not in the
134 chip-details - and thus provides a more transparent IRQ subsystem
138 Each interrupt descriptor is assigned its own highlevel flow
139 handler, which is normally one of the generic
140 implementations. (This highlevel flow handler implementation also
141 makes it simple to provide demultiplexing handlers which can be
142 found in embedded platforms on various architectures.)
145 The separation makes the generic interrupt handling layer more
146 flexible and extensible. For example, an (sub)architecture can
147 use a generic irq-flow implementation for 'level type' interrupts
148 and add a (sub)architecture specific 'edge type' implementation.
151 To make the transition to the new model easier and prevent the
152 breakage of existing implementations, the __do_IRQ() super-handler
153 is still available. This leads to a kind of duality for the time
154 being. Over time the new model should be used in more and more
155 architectures, as it enables smaller and cleaner IRQ subsystems.
159 <title>Known Bugs And Assumptions</title>
161 None (knock on wood).
165 <chapter id="Abstraction">
166 <title>Abstraction layers</title>
168 There are three main levels of abstraction in the interrupt code:
170 <listitem><para>Highlevel driver API</para></listitem>
171 <listitem><para>Highlevel IRQ flow handlers</para></listitem>
172 <listitem><para>Chiplevel hardware encapsulation</para></listitem>
176 <title>Interrupt control flow</title>
178 Each interrupt is described by an interrupt descriptor structure
179 irq_desc. The interrupt is referenced by an 'unsigned int' numeric
180 value which selects the corresponding interrupt decription structure
181 in the descriptor structures array.
182 The descriptor structure contains status information and pointers
183 to the interrupt flow method and the interrupt chip structure
184 which are assigned to this interrupt.
187 Whenever an interrupt triggers, the lowlevel arch code calls into
188 the generic interrupt code by calling desc->handle_irq().
189 This highlevel IRQ handling function only uses desc->chip primitives
190 referenced by the assigned chip descriptor structure.
194 <title>Highlevel Driver API</title>
196 The highlevel Driver API consists of following functions:
198 <listitem><para>request_irq()</para></listitem>
199 <listitem><para>free_irq()</para></listitem>
200 <listitem><para>disable_irq()</para></listitem>
201 <listitem><para>enable_irq()</para></listitem>
202 <listitem><para>disable_irq_nosync() (SMP only)</para></listitem>
203 <listitem><para>synchronize_irq() (SMP only)</para></listitem>
204 <listitem><para>set_irq_type()</para></listitem>
205 <listitem><para>set_irq_wake()</para></listitem>
206 <listitem><para>set_irq_data()</para></listitem>
207 <listitem><para>set_irq_chip()</para></listitem>
208 <listitem><para>set_irq_chip_data()</para></listitem>
210 See the autogenerated function documentation for details.
214 <title>Highlevel IRQ flow handlers</title>
216 The generic layer provides a set of pre-defined irq-flow methods:
218 <listitem><para>handle_level_irq</para></listitem>
219 <listitem><para>handle_edge_irq</para></listitem>
220 <listitem><para>handle_simple_irq</para></listitem>
221 <listitem><para>handle_percpu_irq</para></listitem>
223 The interrupt flow handlers (either predefined or architecture
224 specific) are assigned to specific interrupts by the architecture
225 either during bootup or during device initialization.
228 <title>Default flow implementations</title>
230 <title>Helper functions</title>
232 The helper functions call the chip primitives and
233 are used by the default flow implementations.
234 The following helper functions are implemented (simplified excerpt):
238 desc->chip->unmask(irq);
243 if (!delay_disable(irq))
244 desc->chip->mask(irq);
252 default_mask_ack(irq)
254 if (chip->mask_ack) {
271 <title>Default flow handler implementations</title>
273 <title>Default Level IRQ flow handler</title>
275 handle_level_irq provides a generic implementation
276 for level-triggered interrupts.
279 The following control flow is implemented (simplified excerpt):
282 handle_IRQ_event(desc->action);
288 <title>Default Edge IRQ flow handler</title>
290 handle_edge_irq provides a generic implementation
291 for edge-triggered interrupts.
294 The following control flow is implemented (simplified excerpt):
296 if (desc->status & running) {
298 desc->status |= pending | masked;
302 desc->status |= running;
304 if (desc->status & masked)
305 desc->chip->enable();
306 desc-status &= ~pending;
307 handle_IRQ_event(desc->action);
308 } while (status & pending);
309 desc-status &= ~running;
315 <title>Default simple IRQ flow handler</title>
317 handle_simple_irq provides a generic implementation
318 for simple interrupts.
321 Note: The simple flow handler does not call any
322 handler/chip primitives.
325 The following control flow is implemented (simplified excerpt):
327 handle_IRQ_event(desc->action);
332 <title>Default per CPU flow handler</title>
334 handle_percpu_irq provides a generic implementation
335 for per CPU interrupts.
338 Per CPU interrupts are only available on SMP and
339 the handler provides a simplified version without
343 The following control flow is implemented (simplified excerpt):
346 handle_IRQ_event(desc->action);
353 <title>Quirks and optimizations</title>
355 The generic functions are intended for 'clean' architectures and chips,
356 which have no platform-specific IRQ handling quirks. If an architecture
357 needs to implement quirks on the 'flow' level then it can do so by
358 overriding the highlevel irq-flow handler.
362 <title>Delayed interrupt disable</title>
364 This per interrupt selectable feature, which was introduced by Russell
365 King in the ARM interrupt implementation, does not mask an interrupt
366 at the hardware level when disable_irq() is called. The interrupt is
367 kept enabled and is masked in the flow handler when an interrupt event
368 happens. This prevents losing edge interrupts on hardware which does
369 not store an edge interrupt event while the interrupt is disabled at
370 the hardware level. When an interrupt arrives while the IRQ_DISABLED
371 flag is set, then the interrupt is masked at the hardware level and
372 the IRQ_PENDING bit is set. When the interrupt is re-enabled by
373 enable_irq() the pending bit is checked and if it is set, the
374 interrupt is resent either via hardware or by a software resend
375 mechanism. (It's necessary to enable CONFIG_HARDIRQS_SW_RESEND when
376 you want to use the delayed interrupt disable feature and your
377 hardware is not capable of retriggering an interrupt.)
378 The delayed interrupt disable can be runtime enabled, per interrupt,
379 by setting the IRQ_DELAYED_DISABLE flag in the irq_desc status field.
384 <title>Chiplevel hardware encapsulation</title>
386 The chip level hardware descriptor structure irq_chip
387 contains all the direct chip relevant functions, which
388 can be utilized by the irq flow implementations.
390 <listitem><para>ack()</para></listitem>
391 <listitem><para>mask_ack() - Optional, recommended for performance</para></listitem>
392 <listitem><para>mask()</para></listitem>
393 <listitem><para>unmask()</para></listitem>
394 <listitem><para>retrigger() - Optional</para></listitem>
395 <listitem><para>set_type() - Optional</para></listitem>
396 <listitem><para>set_wake() - Optional</para></listitem>
398 These primitives are strictly intended to mean what they say: ack means
399 ACK, masking means masking of an IRQ line, etc. It is up to the flow
400 handler(s) to use these basic units of lowlevel functionality.
406 <title>__do_IRQ entry point</title>
408 The original implementation __do_IRQ() is an alternative entry
409 point for all types of interrupts.
412 This handler turned out to be not suitable for all
413 interrupt hardware and was therefore reimplemented with split
414 functionality for egde/level/simple/percpu interrupts. This is not
415 only a functional optimization. It also shortens code paths for
419 To make use of the split implementation, replace the call to
420 __do_IRQ by a call to desc->chip->handle_irq() and associate
421 the appropriate handler function to desc->chip->handle_irq().
422 In most cases the generic handler implementations should
427 <chapter id="locking">
428 <title>Locking on SMP</title>
430 The locking of chip registers is up to the architecture that
431 defines the chip primitives. There is a chip->lock field that can be used
432 for serialization, but the generic layer does not touch it. The per-irq
433 structure is protected via desc->lock, by the generic layer.
436 <chapter id="structs">
437 <title>Structures</title>
439 This chapter contains the autogenerated documentation of the structures which are
440 used in the generic IRQ layer.
442 !Iinclude/linux/irq.h
445 <chapter id="pubfunctions">
446 <title>Public Functions Provided</title>
448 This chapter contains the autogenerated documentation of the kernel API functions
451 !Ekernel/irq/manage.c
455 <chapter id="intfunctions">
456 <title>Internal Functions Provided</title>
458 This chapter contains the autogenerated documentation of the internal functions.
460 !Ikernel/irq/handle.c
464 <chapter id="credits">
465 <title>Credits</title>
467 The following people have contributed to this document:
469 <listitem><para>Thomas Gleixner<email>tglx@linutronix.de</email></para></listitem>
470 <listitem><para>Ingo Molnar<email>mingo@elte.hu</email></para></listitem>