1=====================
2GPIO Driver Interface
3=====================
4
5This document serves as a guide for writers of GPIO chip drivers.
6
7Each GPIO controller driver needs to include the following header, which defines
8the structures used to define a GPIO driver::
9
10	#include <linux/gpio/driver.h>
11
12
13Internal Representation of GPIOs
14================================
15
16A GPIO chip handles one or more GPIO lines. To be considered a GPIO chip, the
17lines must conform to the definition: General Purpose Input/Output. If the
18line is not general purpose, it is not GPIO and should not be handled by a
19GPIO chip. The use case is the indicative: certain lines in a system may be
20called GPIO but serve a very particular purpose thus not meeting the criteria
21of a general purpose I/O. On the other hand a LED driver line may be used as a
22GPIO and should therefore still be handled by a GPIO chip driver.
23
24Inside a GPIO driver, individual GPIO lines are identified by their hardware
25number, sometime also referred to as ``offset``, which is a unique number
26between 0 and n-1, n being the number of GPIOs managed by the chip.
27
28The hardware GPIO number should be something intuitive to the hardware, for
29example if a system uses a memory-mapped set of I/O-registers where 32 GPIO
30lines are handled by one bit per line in a 32-bit register, it makes sense to
31use hardware offsets 0..31 for these, corresponding to bits 0..31 in the
32register.
33
34This number is purely internal: the hardware number of a particular GPIO
35line is never made visible outside of the driver.
36
37On top of this internal number, each GPIO line also needs to have a global
38number in the integer GPIO namespace so that it can be used with the legacy GPIO
39interface. Each chip must thus have a "base" number (which can be automatically
40assigned), and for each GPIO line the global number will be (base + hardware
41number). Although the integer representation is considered deprecated, it still
42has many users and thus needs to be maintained.
43
44So for example one platform could use global numbers 32-159 for GPIOs, with a
45controller defining 128 GPIOs at a "base" of 32 ; while another platform uses
46global numbers 0..63 with one set of GPIO controllers, 64-79 with another type
47of GPIO controller, and on one particular board 80-95 with an FPGA. The legacy
48numbers need not be contiguous; either of those platforms could also use numbers
492000-2063 to identify GPIO lines in a bank of I2C GPIO expanders.
50
51
52Controller Drivers: gpio_chip
53=============================
54
55In the gpiolib framework each GPIO controller is packaged as a "struct
56gpio_chip" (see <linux/gpio/driver.h> for its complete definition) with members
57common to each controller of that type, these should be assigned by the
58driver code:
59
60 - methods to establish GPIO line direction
61 - methods used to access GPIO line values
62 - method to set electrical configuration for a given GPIO line
63 - method to return the IRQ number associated to a given GPIO line
64 - flag saying whether calls to its methods may sleep
65 - optional line names array to identify lines
66 - optional debugfs dump method (showing extra state information)
67 - optional base number (will be automatically assigned if omitted)
68 - optional label for diagnostics and GPIO chip mapping using platform data
69
70The code implementing a gpio_chip should support multiple instances of the
71controller, preferably using the driver model. That code will configure each
72gpio_chip and issue gpiochip_add(), gpiochip_add_data(), or
73devm_gpiochip_add_data().  Removing a GPIO controller should be rare; use
74gpiochip_remove() when it is unavoidable.
75
76Often a gpio_chip is part of an instance-specific structure with states not
77exposed by the GPIO interfaces, such as addressing, power management, and more.
78Chips such as audio codecs will have complex non-GPIO states.
79
80Any debugfs dump method should normally ignore lines which haven't been
81requested. They can use gpiochip_is_requested(), which returns either
82NULL or the label associated with that GPIO line when it was requested.
83
84Realtime considerations: the GPIO driver should not use spinlock_t or any
85sleepable APIs (like PM runtime) in its gpio_chip implementation (.get/.set
86and direction control callbacks) if it is expected to call GPIO APIs from
87atomic context on realtime kernels (inside hard IRQ handlers and similar
88contexts). Normally this should not be required.
89
90
91GPIO electrical configuration
92-----------------------------
93
94GPIO lines can be configured for several electrical modes of operation by using
95the .set_config() callback. Currently this API supports setting:
96
97- Debouncing
98- Single-ended modes (open drain/open source)
99- Pull up and pull down resistor enablement
100
101These settings are described below.
102
103The .set_config() callback uses the same enumerators and configuration
104semantics as the generic pin control drivers. This is not a coincidence: it is
105possible to assign the .set_config() to the function gpiochip_generic_config()
106which will result in pinctrl_gpio_set_config() being called and eventually
107ending up in the pin control back-end "behind" the GPIO controller, usually
108closer to the actual pins. This way the pin controller can manage the below
109listed GPIO configurations.
110
111If a pin controller back-end is used, the GPIO controller or hardware
112description needs to provide "GPIO ranges" mapping the GPIO line offsets to pin
113numbers on the pin controller so they can properly cross-reference each other.
114
115
116GPIO lines with debounce support
117--------------------------------
118
119Debouncing is a configuration set to a pin indicating that it is connected to
120a mechanical switch or button, or similar that may bounce. Bouncing means the
121line is pulled high/low quickly at very short intervals for mechanical
122reasons. This can result in the value being unstable or irqs firing repeatedly
123unless the line is debounced.
124
125Debouncing in practice involves setting up a timer when something happens on
126the line, wait a little while and then sample the line again, so see if it
127still has the same value (low or high). This could also be repeated by a clever
128state machine, waiting for a line to become stable. In either case, it sets
129a certain number of milliseconds for debouncing, or just "on/off" if that time
130is not configurable.
131
132
133GPIO lines with open drain/source support
134-----------------------------------------
135
136Open drain (CMOS) or open collector (TTL) means the line is not actively driven
137high: instead you provide the drain/collector as output, so when the transistor
138is not open, it will present a high-impedance (tristate) to the external rail::
139
140
141   CMOS CONFIGURATION      TTL CONFIGURATION
142
143            ||--- out              +--- out
144     in ----||                   |/
145            ||--+         in ----|
146                |                |\
147               GND	           GND
148
149This configuration is normally used as a way to achieve one of two things:
150
151- Level-shifting: to reach a logical level higher than that of the silicon
152  where the output resides.
153
154- Inverse wire-OR on an I/O line, for example a GPIO line, making it possible
155  for any driving stage on the line to drive it low even if any other output
156  to the same line is simultaneously driving it high. A special case of this
157  is driving the SCL and SDA lines of an I2C bus, which is by definition a
158  wire-OR bus.
159
160Both use cases require that the line be equipped with a pull-up resistor. This
161resistor will make the line tend to high level unless one of the transistors on
162the rail actively pulls it down.
163
164The level on the line will go as high as the VDD on the pull-up resistor, which
165may be higher than the level supported by the transistor, achieving a
166level-shift to the higher VDD.
167
168Integrated electronics often have an output driver stage in the form of a CMOS
169"totem-pole" with one N-MOS and one P-MOS transistor where one of them drives
170the line high and one of them drives the line low. This is called a push-pull
171output. The "totem-pole" looks like so::
172
173                     VDD
174                      |
175            OD    ||--+
176         +--/ ---o||     P-MOS-FET
177         |        ||--+
178    IN --+            +----- out
179         |        ||--+
180         +--/ ----||     N-MOS-FET
181            OS    ||--+
182                      |
183                     GND
184
185The desired output signal (e.g. coming directly from some GPIO output register)
186arrives at IN. The switches named "OD" and "OS" are normally closed, creating
187a push-pull circuit.
188
189Consider the little "switches" named "OD" and "OS" that enable/disable the
190P-MOS or N-MOS transistor right after the split of the input. As you can see,
191either transistor will go totally numb if this switch is open. The totem-pole
192is then halved and give high impedance instead of actively driving the line
193high or low respectively. That is usually how software-controlled open
194drain/source works.
195
196Some GPIO hardware come in open drain / open source configuration. Some are
197hard-wired lines that will only support open drain or open source no matter
198what: there is only one transistor there. Some are software-configurable:
199by flipping a bit in a register the output can be configured as open drain
200or open source, in practice by flicking open the switches labeled "OD" and "OS"
201in the drawing above.
202
203By disabling the P-MOS transistor, the output can be driven between GND and
204high impedance (open drain), and by disabling the N-MOS transistor, the output
205can be driven between VDD and high impedance (open source). In the first case,
206a pull-up resistor is needed on the outgoing rail to complete the circuit, and
207in the second case, a pull-down resistor is needed on the rail.
208
209Hardware that supports open drain or open source or both, can implement a
210special callback in the gpio_chip: .set_config() that takes a generic
211pinconf packed value telling whether to configure the line as open drain,
212open source or push-pull. This will happen in response to the
213GPIO_OPEN_DRAIN or GPIO_OPEN_SOURCE flag set in the machine file, or coming
214from other hardware descriptions.
215
216If this state can not be configured in hardware, i.e. if the GPIO hardware does
217not support open drain/open source in hardware, the GPIO library will instead
218use a trick: when a line is set as output, if the line is flagged as open
219drain, and the IN output value is low, it will be driven low as usual. But
220if the IN output value is set to high, it will instead *NOT* be driven high,
221instead it will be switched to input, as input mode is high impedance, thus
222achieving an "open drain emulation" of sorts: electrically the behaviour will
223be identical, with the exception of possible hardware glitches when switching
224the mode of the line.
225
226For open source configuration the same principle is used, just that instead
227of actively driving the line low, it is set to input.
228
229
230GPIO lines with pull up/down resistor support
231---------------------------------------------
232
233A GPIO line can support pull-up/down using the .set_config() callback. This
234means that a pull up or pull-down resistor is available on the output of the
235GPIO line, and this resistor is software controlled.
236
237In discrete designs, a pull-up or pull-down resistor is simply soldered on
238the circuit board. This is not something we deal with or model in software. The
239most you will think about these lines is that they will very likely be
240configured as open drain or open source (see the section above).
241
242The .set_config() callback can only turn pull up or down on and off, and will
243no have any semantic knowledge about the resistance used. It will only say
244switch a bit in a register enabling or disabling pull-up or pull-down.
245
246If the GPIO line supports shunting in different resistance values for the
247pull-up or pull-down resistor, the GPIO chip callback .set_config() will not
248suffice. For these complex use cases, a combined GPIO chip and pin controller
249need to be implemented, as the pin config interface of a pin controller
250supports more versatile control over electrical properties and can handle
251different pull-up or pull-down resistance values.
252
253
254GPIO drivers providing IRQs
255===========================
256
257It is custom that GPIO drivers (GPIO chips) are also providing interrupts,
258most often cascaded off a parent interrupt controller, and in some special
259cases the GPIO logic is melded with a SoC's primary interrupt controller.
260
261The IRQ portions of the GPIO block are implemented using an irq_chip, using
262the header <linux/irq.h>. So this combined driver is utilizing two sub-
263systems simultaneously: gpio and irq.
264
265It is legal for any IRQ consumer to request an IRQ from any irqchip even if it
266is a combined GPIO+IRQ driver. The basic premise is that gpio_chip and
267irq_chip are orthogonal, and offering their services independent of each
268other.
269
270gpiod_to_irq() is just a convenience function to figure out the IRQ for a
271certain GPIO line and should not be relied upon to have been called before
272the IRQ is used.
273
274Always prepare the hardware and make it ready for action in respective
275callbacks from the GPIO and irq_chip APIs. Do not rely on gpiod_to_irq() having
276been called first.
277
278We can divide GPIO irqchips in two broad categories:
279
280- CASCADED INTERRUPT CHIPS: this means that the GPIO chip has one common
281  interrupt output line, which is triggered by any enabled GPIO line on that
282  chip. The interrupt output line will then be routed to an parent interrupt
283  controller one level up, in the most simple case the systems primary
284  interrupt controller. This is modeled by an irqchip that will inspect bits
285  inside the GPIO controller to figure out which line fired it. The irqchip
286  part of the driver needs to inspect registers to figure this out and it
287  will likely also need to acknowledge that it is handling the interrupt
288  by clearing some bit (sometime implicitly, by just reading a status
289  register) and it will often need to set up the configuration such as
290  edge sensitivity (rising or falling edge, or high/low level interrupt for
291  example).
292
293- HIERARCHICAL INTERRUPT CHIPS: this means that each GPIO line has a dedicated
294  irq line to a parent interrupt controller one level up. There is no need
295  to inquire the GPIO hardware to figure out which line has fired, but it
296  may still be necessary to acknowledge the interrupt and set up configuration
297  such as edge sensitivity.
298
299Realtime considerations: a realtime compliant GPIO driver should not use
300spinlock_t or any sleepable APIs (like PM runtime) as part of its irqchip
301implementation.
302
303- spinlock_t should be replaced with raw_spinlock_t.[1]
304- If sleepable APIs have to be used, these can be done from the .irq_bus_lock()
305  and .irq_bus_unlock() callbacks, as these are the only slowpath callbacks
306  on an irqchip. Create the callbacks if needed.[2]
307
308
309Cascaded GPIO irqchips
310----------------------
311
312Cascaded GPIO irqchips usually fall in one of three categories:
313
314- CHAINED CASCADED GPIO IRQCHIPS: these are usually the type that is embedded on
315  an SoC. This means that there is a fast IRQ flow handler for the GPIOs that
316  gets called in a chain from the parent IRQ handler, most typically the
317  system interrupt controller. This means that the GPIO irqchip handler will
318  be called immediately from the parent irqchip, while holding the IRQs
319  disabled. The GPIO irqchip will then end up calling something like this
320  sequence in its interrupt handler::
321
322    static irqreturn_t foo_gpio_irq(int irq, void *data)
323        chained_irq_enter(...);
324        generic_handle_irq(...);
325        chained_irq_exit(...);
326
327  Chained GPIO irqchips typically can NOT set the .can_sleep flag on
328  struct gpio_chip, as everything happens directly in the callbacks: no
329  slow bus traffic like I2C can be used.
330
331  Realtime considerations: Note that chained IRQ handlers will not be forced
332  threaded on -RT. As a result, spinlock_t or any sleepable APIs (like PM
333  runtime) can't be used in a chained IRQ handler.
334
335  If required (and if it can't be converted to the nested threaded GPIO irqchip,
336  see below) a chained IRQ handler can be converted to generic irq handler and
337  this way it will become a threaded IRQ handler on -RT and a hard IRQ handler
338  on non-RT (for example, see [3]).
339
340  The generic_handle_irq() is expected to be called with IRQ disabled,
341  so the IRQ core will complain if it is called from an IRQ handler which is
342  forced to a thread. The "fake?" raw lock can be used to work around this
343  problem::
344
345    raw_spinlock_t wa_lock;
346    static irqreturn_t omap_gpio_irq_handler(int irq, void *gpiobank)
347        unsigned long wa_lock_flags;
348        raw_spin_lock_irqsave(&bank->wa_lock, wa_lock_flags);
349        generic_handle_irq(irq_find_mapping(bank->chip.irq.domain, bit));
350        raw_spin_unlock_irqrestore(&bank->wa_lock, wa_lock_flags);
351
352- GENERIC CHAINED GPIO IRQCHIPS: these are the same as "CHAINED GPIO irqchips",
353  but chained IRQ handlers are not used. Instead GPIO IRQs dispatching is
354  performed by generic IRQ handler which is configured using request_irq().
355  The GPIO irqchip will then end up calling something like this sequence in
356  its interrupt handler::
357
358    static irqreturn_t gpio_rcar_irq_handler(int irq, void *dev_id)
359        for each detected GPIO IRQ
360            generic_handle_irq(...);
361
362  Realtime considerations: this kind of handlers will be forced threaded on -RT,
363  and as result the IRQ core will complain that generic_handle_irq() is called
364  with IRQ enabled and the same work-around as for "CHAINED GPIO irqchips" can
365  be applied.
366
367- NESTED THREADED GPIO IRQCHIPS: these are off-chip GPIO expanders and any
368  other GPIO irqchip residing on the other side of a sleeping bus such as I2C
369  or SPI.
370
371  Of course such drivers that need slow bus traffic to read out IRQ status and
372  similar, traffic which may in turn incur other IRQs to happen, cannot be
373  handled in a quick IRQ handler with IRQs disabled. Instead they need to spawn
374  a thread and then mask the parent IRQ line until the interrupt is handled
375  by the driver. The hallmark of this driver is to call something like
376  this in its interrupt handler::
377
378    static irqreturn_t foo_gpio_irq(int irq, void *data)
379        ...
380        handle_nested_irq(irq);
381
382  The hallmark of threaded GPIO irqchips is that they set the .can_sleep
383  flag on struct gpio_chip to true, indicating that this chip may sleep
384  when accessing the GPIOs.
385
386  These kinds of irqchips are inherently realtime tolerant as they are
387  already set up to handle sleeping contexts.
388
389
390Infrastructure helpers for GPIO irqchips
391----------------------------------------
392
393To help out in handling the set-up and management of GPIO irqchips and the
394associated irqdomain and resource allocation callbacks. These are activated
395by selecting the Kconfig symbol GPIOLIB_IRQCHIP. If the symbol
396IRQ_DOMAIN_HIERARCHY is also selected, hierarchical helpers will also be
397provided. A big portion of overhead code will be managed by gpiolib,
398under the assumption that your interrupts are 1-to-1-mapped to the
399GPIO line index:
400
401.. csv-table::
402    :header: GPIO line offset, Hardware IRQ
403
404    0,0
405    1,1
406    2,2
407    ...,...
408    ngpio-1, ngpio-1
409
410
411If some GPIO lines do not have corresponding IRQs, the bitmask valid_mask
412and the flag need_valid_mask in gpio_irq_chip can be used to mask off some
413lines as invalid for associating with IRQs.
414
415The preferred way to set up the helpers is to fill in the
416struct gpio_irq_chip inside struct gpio_chip before adding the gpio_chip.
417If you do this, the additional irq_chip will be set up by gpiolib at the
418same time as setting up the rest of the GPIO functionality. The following
419is a typical example of a chained cascaded interrupt handler using
420the gpio_irq_chip. Note how the mask/unmask (or disable/enable) functions
421call into the core gpiolib code:
422
423.. code-block:: c
424
425  /* Typical state container */
426  struct my_gpio {
427      struct gpio_chip gc;
428  };
429
430  static void my_gpio_mask_irq(struct irq_data *d)
431  {
432      struct gpio_chip *gc = irq_data_get_irq_chip_data(d);
433      irq_hw_number_t hwirq = irqd_to_hwirq(d);
434
435      /*
436       * Perform any necessary action to mask the interrupt,
437       * and then call into the core code to synchronise the
438       * state.
439       */
440
441      gpiochip_disable_irq(gc, hwirq);
442  }
443
444  static void my_gpio_unmask_irq(struct irq_data *d)
445  {
446      struct gpio_chip *gc = irq_data_get_irq_chip_data(d);
447      irq_hw_number_t hwirq = irqd_to_hwirq(d);
448
449      gpiochip_enable_irq(gc, hwirq);
450
451      /*
452       * Perform any necessary action to unmask the interrupt,
453       * after having called into the core code to synchronise
454       * the state.
455       */
456  }
457
458  /*
459   * Statically populate the irqchip. Note that it is made const
460   * (further indicated by the IRQCHIP_IMMUTABLE flag), and that
461   * the GPIOCHIP_IRQ_RESOURCE_HELPER macro adds some extra
462   * callbacks to the structure.
463   */
464  static const struct irq_chip my_gpio_irq_chip = {
465      .name		= "my_gpio_irq",
466      .irq_ack		= my_gpio_ack_irq,
467      .irq_mask		= my_gpio_mask_irq,
468      .irq_unmask	= my_gpio_unmask_irq,
469      .irq_set_type	= my_gpio_set_irq_type,
470      .flags		= IRQCHIP_IMMUTABLE,
471      /* Provide the gpio resource callbacks */
472      GPIOCHIP_IRQ_RESOURCE_HELPERS,
473  };
474
475  int irq; /* from platform etc */
476  struct my_gpio *g;
477  struct gpio_irq_chip *girq;
478
479  /* Get a pointer to the gpio_irq_chip */
480  girq = &g->gc.irq;
481  gpio_irq_chip_set_chip(girq, &my_gpio_irq_chip);
482  girq->parent_handler = ftgpio_gpio_irq_handler;
483  girq->num_parents = 1;
484  girq->parents = devm_kcalloc(dev, 1, sizeof(*girq->parents),
485                               GFP_KERNEL);
486  if (!girq->parents)
487      return -ENOMEM;
488  girq->default_type = IRQ_TYPE_NONE;
489  girq->handler = handle_bad_irq;
490  girq->parents[0] = irq;
491
492  return devm_gpiochip_add_data(dev, &g->gc, g);
493
494The helper supports using threaded interrupts as well. Then you just request
495the interrupt separately and go with it:
496
497.. code-block:: c
498
499  /* Typical state container */
500  struct my_gpio {
501      struct gpio_chip gc;
502  };
503
504  static void my_gpio_mask_irq(struct irq_data *d)
505  {
506      struct gpio_chip *gc = irq_data_get_irq_chip_data(d);
507      irq_hw_number_t hwirq = irqd_to_hwirq(d);
508
509      /*
510       * Perform any necessary action to mask the interrupt,
511       * and then call into the core code to synchronise the
512       * state.
513       */
514
515      gpiochip_disable_irq(gc, hwirq);
516  }
517
518  static void my_gpio_unmask_irq(struct irq_data *d)
519  {
520      struct gpio_chip *gc = irq_data_get_irq_chip_data(d);
521      irq_hw_number_t hwirq = irqd_to_hwirq(d);
522
523      gpiochip_enable_irq(gc, hwirq);
524
525      /*
526       * Perform any necessary action to unmask the interrupt,
527       * after having called into the core code to synchronise
528       * the state.
529       */
530  }
531
532  /*
533   * Statically populate the irqchip. Note that it is made const
534   * (further indicated by the IRQCHIP_IMMUTABLE flag), and that
535   * the GPIOCHIP_IRQ_RESOURCE_HELPER macro adds some extra
536   * callbacks to the structure.
537   */
538  static const struct irq_chip my_gpio_irq_chip = {
539      .name		= "my_gpio_irq",
540      .irq_ack		= my_gpio_ack_irq,
541      .irq_mask		= my_gpio_mask_irq,
542      .irq_unmask	= my_gpio_unmask_irq,
543      .irq_set_type	= my_gpio_set_irq_type,
544      .flags		= IRQCHIP_IMMUTABLE,
545      /* Provide the gpio resource callbacks */
546      GPIOCHIP_IRQ_RESOURCE_HELPERS,
547  };
548
549  int irq; /* from platform etc */
550  struct my_gpio *g;
551  struct gpio_irq_chip *girq;
552
553  ret = devm_request_threaded_irq(dev, irq, NULL,
554		irq_thread_fn, IRQF_ONESHOT, "my-chip", g);
555  if (ret < 0)
556	return ret;
557
558  /* Get a pointer to the gpio_irq_chip */
559  girq = &g->gc.irq;
560  gpio_irq_chip_set_chip(girq, &my_gpio_irq_chip);
561  /* This will let us handle the parent IRQ in the driver */
562  girq->parent_handler = NULL;
563  girq->num_parents = 0;
564  girq->parents = NULL;
565  girq->default_type = IRQ_TYPE_NONE;
566  girq->handler = handle_bad_irq;
567
568  return devm_gpiochip_add_data(dev, &g->gc, g);
569
570The helper supports using hierarchical interrupt controllers as well.
571In this case the typical set-up will look like this:
572
573.. code-block:: c
574
575  /* Typical state container with dynamic irqchip */
576  struct my_gpio {
577      struct gpio_chip gc;
578      struct fwnode_handle *fwnode;
579  };
580
581  static void my_gpio_mask_irq(struct irq_data *d)
582  {
583      struct gpio_chip *gc = irq_data_get_irq_chip_data(d);
584      irq_hw_number_t hwirq = irqd_to_hwirq(d);
585
586      /*
587       * Perform any necessary action to mask the interrupt,
588       * and then call into the core code to synchronise the
589       * state.
590       */
591
592      gpiochip_disable_irq(gc, hwirq);
593      irq_mask_mask_parent(d);
594  }
595
596  static void my_gpio_unmask_irq(struct irq_data *d)
597  {
598      struct gpio_chip *gc = irq_data_get_irq_chip_data(d);
599      irq_hw_number_t hwirq = irqd_to_hwirq(d);
600
601      gpiochip_enable_irq(gc, hwirq);
602
603      /*
604       * Perform any necessary action to unmask the interrupt,
605       * after having called into the core code to synchronise
606       * the state.
607       */
608
609      irq_mask_unmask_parent(d);
610  }
611
612  /*
613   * Statically populate the irqchip. Note that it is made const
614   * (further indicated by the IRQCHIP_IMMUTABLE flag), and that
615   * the GPIOCHIP_IRQ_RESOURCE_HELPER macro adds some extra
616   * callbacks to the structure.
617   */
618  static const struct irq_chip my_gpio_irq_chip = {
619      .name		= "my_gpio_irq",
620      .irq_ack		= my_gpio_ack_irq,
621      .irq_mask		= my_gpio_mask_irq,
622      .irq_unmask	= my_gpio_unmask_irq,
623      .irq_set_type	= my_gpio_set_irq_type,
624      .flags		= IRQCHIP_IMMUTABLE,
625      /* Provide the gpio resource callbacks */
626      GPIOCHIP_IRQ_RESOURCE_HELPERS,
627  };
628
629  struct my_gpio *g;
630  struct gpio_irq_chip *girq;
631
632  /* Get a pointer to the gpio_irq_chip */
633  girq = &g->gc.irq;
634  gpio_irq_chip_set_chip(girq, &my_gpio_irq_chip);
635  girq->default_type = IRQ_TYPE_NONE;
636  girq->handler = handle_bad_irq;
637  girq->fwnode = g->fwnode;
638  girq->parent_domain = parent;
639  girq->child_to_parent_hwirq = my_gpio_child_to_parent_hwirq;
640
641  return devm_gpiochip_add_data(dev, &g->gc, g);
642
643As you can see pretty similar, but you do not supply a parent handler for
644the IRQ, instead a parent irqdomain, an fwnode for the hardware and
645a function .child_to_parent_hwirq() that has the purpose of looking up
646the parent hardware irq from a child (i.e. this gpio chip) hardware irq.
647As always it is good to look at examples in the kernel tree for advice
648on how to find the required pieces.
649
650If there is a need to exclude certain GPIO lines from the IRQ domain handled by
651these helpers, we can set .irq.need_valid_mask of the gpiochip before
652devm_gpiochip_add_data() or gpiochip_add_data() is called. This allocates an
653.irq.valid_mask with as many bits set as there are GPIO lines in the chip, each
654bit representing line 0..n-1. Drivers can exclude GPIO lines by clearing bits
655from this mask. The mask can be filled in the init_valid_mask() callback
656that is part of the struct gpio_irq_chip.
657
658To use the helpers please keep the following in mind:
659
660- Make sure to assign all relevant members of the struct gpio_chip so that
661  the irqchip can initialize. E.g. .dev and .can_sleep shall be set up
662  properly.
663
664- Nominally set gpio_irq_chip.handler to handle_bad_irq. Then, if your irqchip
665  is cascaded, set the handler to handle_level_irq() and/or handle_edge_irq()
666  in the irqchip .set_type() callback depending on what your controller
667  supports and what is requested by the consumer.
668
669
670Locking IRQ usage
671-----------------
672
673Since GPIO and irq_chip are orthogonal, we can get conflicts between different
674use cases. For example a GPIO line used for IRQs should be an input line,
675it does not make sense to fire interrupts on an output GPIO.
676
677If there is competition inside the subsystem which side is using the
678resource (a certain GPIO line and register for example) it needs to deny
679certain operations and keep track of usage inside of the gpiolib subsystem.
680
681Input GPIOs can be used as IRQ signals. When this happens, a driver is requested
682to mark the GPIO as being used as an IRQ::
683
684	int gpiochip_lock_as_irq(struct gpio_chip *chip, unsigned int offset)
685
686This will prevent the use of non-irq related GPIO APIs until the GPIO IRQ lock
687is released::
688
689	void gpiochip_unlock_as_irq(struct gpio_chip *chip, unsigned int offset)
690
691When implementing an irqchip inside a GPIO driver, these two functions should
692typically be called in the .startup() and .shutdown() callbacks from the
693irqchip.
694
695When using the gpiolib irqchip helpers, these callbacks are automatically
696assigned.
697
698
699Disabling and enabling IRQs
700---------------------------
701
702In some (fringe) use cases, a driver may be using a GPIO line as input for IRQs,
703but occasionally switch that line over to drive output and then back to being
704an input with interrupts again. This happens on things like CEC (Consumer
705Electronics Control).
706
707When a GPIO is used as an IRQ signal, then gpiolib also needs to know if
708the IRQ is enabled or disabled. In order to inform gpiolib about this,
709the irqchip driver should call::
710
711	void gpiochip_disable_irq(struct gpio_chip *chip, unsigned int offset)
712
713This allows drivers to drive the GPIO as an output while the IRQ is
714disabled. When the IRQ is enabled again, a driver should call::
715
716	void gpiochip_enable_irq(struct gpio_chip *chip, unsigned int offset)
717
718When implementing an irqchip inside a GPIO driver, these two functions should
719typically be called in the .irq_disable() and .irq_enable() callbacks from the
720irqchip.
721
722When IRQCHIP_IMMUTABLE is not advertised by the irqchip, these callbacks
723are automatically assigned. This behaviour is deprecated and on its way
724to be removed from the kernel.
725
726
727Real-Time compliance for GPIO IRQ chips
728---------------------------------------
729
730Any provider of irqchips needs to be carefully tailored to support Real-Time
731preemption. It is desirable that all irqchips in the GPIO subsystem keep this
732in mind and do the proper testing to assure they are real time-enabled.
733
734So, pay attention on above realtime considerations in the documentation.
735
736The following is a checklist to follow when preparing a driver for real-time
737compliance:
738
739- ensure spinlock_t is not used as part irq_chip implementation
740- ensure that sleepable APIs are not used as part irq_chip implementation
741  If sleepable APIs have to be used, these can be done from the .irq_bus_lock()
742  and .irq_bus_unlock() callbacks
743- Chained GPIO irqchips: ensure spinlock_t or any sleepable APIs are not used
744  from the chained IRQ handler
745- Generic chained GPIO irqchips: take care about generic_handle_irq() calls and
746  apply corresponding work-around
747- Chained GPIO irqchips: get rid of the chained IRQ handler and use generic irq
748  handler if possible
749- regmap_mmio: it is possible to disable internal locking in regmap by setting
750  .disable_locking and handling the locking in the GPIO driver
751- Test your driver with the appropriate in-kernel real-time test cases for both
752  level and edge IRQs
753
754* [1] http://www.spinics.net/lists/linux-omap/msg120425.html
755* [2] https://lore.kernel.org/r/1443209283-20781-2-git-send-email-grygorii.strashko@ti.com
756* [3] https://lore.kernel.org/r/1443209283-20781-3-git-send-email-grygorii.strashko@ti.com
757
758
759Requesting self-owned GPIO pins
760===============================
761
762Sometimes it is useful to allow a GPIO chip driver to request its own GPIO
763descriptors through the gpiolib API. A GPIO driver can use the following
764functions to request and free descriptors::
765
766	struct gpio_desc *gpiochip_request_own_desc(struct gpio_desc *desc,
767						    u16 hwnum,
768						    const char *label,
769						    enum gpiod_flags flags)
770
771	void gpiochip_free_own_desc(struct gpio_desc *desc)
772
773Descriptors requested with gpiochip_request_own_desc() must be released with
774gpiochip_free_own_desc().
775
776These functions must be used with care since they do not affect module use
777count. Do not use the functions to request gpio descriptors not owned by the
778calling driver.
779