1 .. SPDX-License-Identifier: GPL-2.0
3 ======================================
4 _DSD Device Properties Related to GPIO
5 ======================================
7 With the release of ACPI 5.1, the _DSD configuration object finally
8 allows names to be given to GPIOs (and other things as well) returned
9 by _CRS. Previously, we were only able to use an integer index to find
10 the corresponding GPIO, which is pretty error prone (it depends on
11 the _CRS output ordering, for example).
13 With _DSD we can now query GPIOs using a name instead of an integer
14 index, like the ASL example below shows::
16 // Bluetooth device with reset and shutdown GPIOs
21 Name (_CRS, ResourceTemplate ()
23 GpioIo (Exclusive, PullUp, 0, 0, IoRestrictionOutputOnly,
24 "\\_SB.GPO0", 0, ResourceConsumer) { 15 }
25 GpioIo (Exclusive, PullUp, 0, 0, IoRestrictionOutputOnly,
26 "\\_SB.GPO0", 0, ResourceConsumer) { 27, 31 }
29 Name (_DSD, Package ()
31 ToUUID("daffd814-6eba-4d8c-8a91-bc9bbf4aa301"),
34 Package () { "reset-gpios", Package () { ^BTH, 1, 1, 0 } },
35 Package () { "shutdown-gpios", Package () { ^BTH, 0, 0, 0 } },
40 The format of the supported GPIO property is::
42 Package () { "name", Package () { ref, index, pin, active_low }}
45 The device that has _CRS containing GpioIo()/GpioInt() resources,
46 typically this is the device itself (BTH in our case).
48 Index of the GpioIo()/GpioInt() resource in _CRS starting from zero.
50 Pin in the GpioIo()/GpioInt() resource. Typically this is zero.
52 If 1, the GPIO is marked as active_low.
54 Since ACPI GpioIo() resource does not have a field saying whether it is
55 active low or high, the "active_low" argument can be used here. Setting
56 it to 1 marks the GPIO as active low.
58 Note, active_low in _DSD does not make sense for GpioInt() resource and
59 must be 0. GpioInt() resource has its own means of defining it.
61 In our Bluetooth example the "reset-gpios" refers to the second GpioIo()
62 resource, second pin in that resource with the GPIO number of 31.
64 The GpioIo() resource unfortunately doesn't explicitly provide an initial
65 state of the output pin which driver should use during its initialization.
67 Linux tries to use common sense here and derives the state from the bias
68 and polarity settings. The table below shows the expectations:
70 +-------------+-------------+-----------------------------------------------+
71 | Pull Bias | Polarity | Requested... |
72 +=============+=============+===============================================+
74 +-------------+-------------+-----------------------------------------------+
75 | **Default** | x | AS IS (assumed firmware configured it for us) |
76 +-------------+-------------+-----------------------------------------------+
78 +-------------+-------------+-----------------------------------------------+
79 | **None** | x | AS IS (assumed firmware configured it for us) |
80 | | | with no Pull Bias |
81 +-------------+-------------+-----------------------------------------------+
82 | **Up** | x (no _DSD) | |
83 | +-------------+ as high, assuming non-active |
85 | +-------------+-----------------------------------------------+
86 | | High | as high, assuming active |
87 +-------------+-------------+-----------------------------------------------+
88 | **Down** | x (no _DSD) | |
89 | +-------------+ as low, assuming non-active |
91 | +-------------+-----------------------------------------------+
92 | | Low | as low, assuming active |
93 +-------------+-------------+-----------------------------------------------+
95 That said, for our above example the both GPIOs, since the bias setting
96 is explicit and _DSD is present, will be treated as active with a high
97 polarity and Linux will configure the pins in this state until a driver
98 reprograms them differently.
100 It is possible to leave holes in the array of GPIOs. This is useful in
101 cases like with SPI host controllers where some chip selects may be
102 implemented as GPIOs and some as native signals. For example a SPI host
103 controller can have chip selects 0 and 2 implemented as GPIOs and 1 as
109 ^GPIO, 19, 0, 0, // chip select 0: GPIO
110 0, // chip select 1: native signal
111 ^GPIO, 20, 0, 0, // chip select 2: GPIO
115 Note, that historically ACPI has no means of the GPIO polarity and thus
116 the SPISerialBus() resource defines it on the per-chip basis. In order
117 to avoid a chain of negations, the GPIO polarity is considered being
118 Active High. Even for the cases when _DSD() is involved (see the example
119 above) the GPIO CS polarity must be defined Active High to avoid ambiguity.
121 Other supported properties
122 ==========================
124 Following Device Tree compatible device properties are also supported by
125 _DSD device properties for GPIO controllers:
135 Name (_DSD, Package () {
136 // _DSD Hierarchical Properties Extension UUID
137 ToUUID("dbb8e3e6-5886-4ba6-8795-1319f52a966b"),
139 Package () { "hog-gpio8", "G8PU" }
143 Name (G8PU, Package () {
144 ToUUID("daffd814-6eba-4d8c-8a91-bc9bbf4aa301"),
146 Package () { "gpio-hog", 1 },
147 Package () { "gpios", Package () { 8, 0 } },
148 Package () { "output-high", 1 },
149 Package () { "line-name", "gpio8-pullup" },
155 The ``gpio-line-names`` declaration is a list of strings ("names"), which
156 describes each line/pin of a GPIO controller/expander. This list, contained in
157 a package, must be inserted inside the GPIO controller declaration of an ACPI
158 table (typically inside the DSDT). The ``gpio-line-names`` list must respect the
159 following rules (see also the examples):
161 - the first name in the list corresponds with the first line/pin of the GPIO
163 - the names inside the list must be consecutive (no "holes" are permitted)
164 - the list can be incomplete and can end before the last GPIO line: in
165 other words, it is not mandatory to fill all the GPIO lines
166 - empty names are allowed (two quotation marks ``""`` correspond to an empty
168 - names inside one GPIO controller/expander must be unique
170 Example of a GPIO controller of 16 lines, with an incomplete list with two
185 At runtime, the above declaration produces the following result (using the
188 root@debian:~# gpioinfo gpiochip4
189 gpiochip4 - 16 lines:
190 line 0: "pin_0" unused input active-high
191 line 1: "pin_1" unused input active-high
192 line 2: unnamed unused input active-high
193 line 3: unnamed unused input active-high
194 line 4: "pin_3" unused input active-high
195 line 5: "pin_4_push_button" unused input active-high
196 line 6: unnamed unused input active-high
197 line 7 unnamed unused input active-high
198 line 8: unnamed unused input active-high
199 line 9: unnamed unused input active-high
200 line 10: unnamed unused input active-high
201 line 11: unnamed unused input active-high
202 line 12: unnamed unused input active-high
203 line 13: unnamed unused input active-high
204 line 14: unnamed unused input active-high
205 line 15: unnamed unused input active-high
206 root@debian:~# gpiofind pin_4_push_button
215 "SPI0_CS_N", "EXP2_INT", "MUX6_IO", "UART0_RXD",
216 "MUX7_IO", "LVL_C_A1", "MUX0_IO", "SPI1_MISO",
220 See Documentation/devicetree/bindings/gpio/gpio.txt for more information
221 about these properties.
223 ACPI GPIO Mappings Provided by Drivers
224 ======================================
226 There are systems in which the ACPI tables do not contain _DSD but provide _CRS
227 with GpioIo()/GpioInt() resources and device drivers still need to work with
230 In those cases ACPI device identification objects, _HID, _CID, _CLS, _SUB, _HRV,
231 available to the driver can be used to identify the device and that is supposed
232 to be sufficient to determine the meaning and purpose of all of the GPIO lines
233 listed by the GpioIo()/GpioInt() resources returned by _CRS. In other words,
234 the driver is supposed to know what to use the GpioIo()/GpioInt() resources for
235 once it has identified the device. Having done that, it can simply assign names
236 to the GPIO lines it is going to use and provide the GPIO subsystem with a
237 mapping between those names and the ACPI GPIO resources corresponding to them.
239 To do that, the driver needs to define a mapping table as a NULL-terminated
240 array of struct acpi_gpio_mapping objects that each contains a name, a pointer
241 to an array of line data (struct acpi_gpio_params) objects and the size of that
242 array. Each struct acpi_gpio_params object consists of three fields,
243 crs_entry_index, line_index, active_low, representing the index of the target
244 GpioIo()/GpioInt() resource in _CRS starting from zero, the index of the target
245 line in that resource starting from zero, and the active-low flag for that line,
246 respectively, in analogy with the _DSD GPIO property format specified above.
248 For the example Bluetooth device discussed previously the data structures in
249 question would look like this::
251 static const struct acpi_gpio_params reset_gpio = { 1, 1, false };
252 static const struct acpi_gpio_params shutdown_gpio = { 0, 0, false };
254 static const struct acpi_gpio_mapping bluetooth_acpi_gpios[] = {
255 { "reset-gpios", &reset_gpio, 1 },
256 { "shutdown-gpios", &shutdown_gpio, 1 },
260 Next, the mapping table needs to be passed as the second argument to
261 acpi_dev_add_driver_gpios() or its managed analogue that will
262 register it with the ACPI device object pointed to by its first
263 argument. That should be done in the driver's .probe() routine.
264 On removal, the driver should unregister its GPIO mapping table by
265 calling acpi_dev_remove_driver_gpios() on the ACPI device object where that
266 table was previously registered.
268 Using the _CRS fallback
269 =======================
271 If a device does not have _DSD or the driver does not create ACPI GPIO
272 mapping, the Linux GPIO framework refuses to return any GPIOs. This is
273 because the driver does not know what it actually gets. For example if we
274 have a device like below::
280 Name (_CRS, ResourceTemplate () {
281 GpioIo (Exclusive, PullNone, 0, 0, IoRestrictionNone,
282 "\\_SB.GPO0", 0, ResourceConsumer) { 15 }
283 GpioIo (Exclusive, PullNone, 0, 0, IoRestrictionNone,
284 "\\_SB.GPO0", 0, ResourceConsumer) { 27 }
288 The driver might expect to get the right GPIO when it does::
290 desc = gpiod_get(dev, "reset", GPIOD_OUT_LOW);
294 but since there is no way to know the mapping between "reset" and
295 the GpioIo() in _CRS desc will hold ERR_PTR(-ENOENT).
297 The driver author can solve this by passing the mapping explicitly
298 (this is the recommended way and it's documented in the above chapter).
300 The ACPI GPIO mapping tables should not contaminate drivers that are not
301 knowing about which exact device they are servicing on. It implies that
302 the ACPI GPIO mapping tables are hardly linked to an ACPI ID and certain
303 objects, as listed in the above chapter, of the device in question.
305 Getting GPIO descriptor
306 =======================
308 There are two main approaches to get GPIO resource from ACPI::
310 desc = gpiod_get(dev, connection_id, flags);
311 desc = gpiod_get_index(dev, connection_id, index, flags);
313 We may consider two different cases here, i.e. when connection ID is
314 provided and otherwise.
318 desc = gpiod_get(dev, "non-null-connection-id", flags);
319 desc = gpiod_get_index(dev, "non-null-connection-id", index, flags);
323 desc = gpiod_get(dev, NULL, flags);
324 desc = gpiod_get_index(dev, NULL, index, flags);
326 Case 1 assumes that corresponding ACPI device description must have
327 defined device properties and will prevent to getting any GPIO resources
330 Case 2 explicitly tells GPIO core to look for resources in _CRS.
332 Be aware that gpiod_get_index() in cases 1 and 2, assuming that there
333 are two versions of ACPI device description provided and no mapping is
334 present in the driver, will return different resources. That's why a
335 certain driver has to handle them carefully as explained in the previous