1This is a place for planning the ongoing long-term work in the GPIO 2subsystem. 3 4=============================================================================== 5 6GPIO descriptors 7 8Starting with commit 79a9becda894 the GPIO subsystem embarked on a journey 9to move away from the global GPIO numberspace and toward a descriptor-based 10approach. This means that GPIO consumers, drivers and machine descriptions 11ideally have no use or idea of the global GPIO numberspace that has/was 12used in the inception of the GPIO subsystem. 13 14The numberspace issue is the same as to why irq is moving away from irq 15numbers to IRQ descriptors. 16 17The underlying motivation for this is that the GPIO numberspace has become 18unmanageable: machine board files tend to become full of macros trying to 19establish the numberspace at compile-time, making it hard to add any numbers 20in the middle (such as if you missed a pin on a chip) without the numberspace 21breaking. 22 23Machine descriptions such as device tree or ACPI does not have a concept of the 24Linux GPIO number as those descriptions are external to the Linux kernel 25and treat GPIO lines as abstract entities. 26 27The runtime-assigned GPIO numberspace (what you get if you assign the GPIO 28base as -1 in struct gpio_chip) has also became unpredictable due to factors 29such as probe ordering and the introduction of -EPROBE_DEFER making probe 30ordering of independent GPIO chips essentially unpredictable, as their base 31number will be assigned on a first come first serve basis. 32 33The best way to get out of the problem is to make the global GPIO numbers 34unimportant by simply not using them. GPIO descriptors deal with this. 35 36Work items: 37 38- Convert all GPIO device drivers to only #include <linux/gpio/driver.h> 39 40- Convert all consumer drivers to only #include <linux/gpio/consumer.h> 41 42- Convert all machine descriptors in "boardfiles" to only 43 #include <linux/gpio/machine.h>, the other option being to convert it 44 to a machine description such as device tree, ACPI or fwnode that 45 implicitly does not use global GPIO numbers. 46 47- Fix drivers to not read back struct gpio_chip::base. Some drivers do 48 that and would be broken by attempts to poison it or make it dynamic. 49 Example in AT91 pinctrl driver: 50 https://lore.kernel.org/all/1d00c056-3d61-4c22-bedd-3bae0bf1ddc4@pengutronix.de/ 51 This particular driver is also DT-only, so with the above fixed, the 52 base can be made dynamic (set to -1) if CONFIG_GPIO_SYSFS is disabled. 53 54- When this work is complete (will require some of the items in the 55 following ongoing work as well) we can delete the old global 56 numberspace accessors from <linux/gpio.h> and eventually delete 57 <linux/gpio.h> altogether. 58 59------------------------------------------------------------------------------- 60 61Get rid of <linux/of_gpio.h> 62 63This header and helpers appeared at one point when there was no proper 64driver infrastructure for doing simpler MMIO GPIO devices and there was 65no core support for parsing device tree GPIOs from the core library with 66the [devm_]gpiod_get() calls we have today that will implicitly go into 67the device tree back-end. It is legacy and should not be used in new code. 68 69Work items: 70 71- Change all consumer drivers that #include <linux/of_gpio.h> to 72 #include <linux/gpio/consumer.h> and stop doing custom parsing of the 73 GPIO lines from the device tree. This can be tricky and often involves 74 changing board files, etc. 75 76- Pull semantics for legacy device tree (OF) GPIO lookups into 77 gpiolib-of.c: in some cases subsystems are doing custom flags and 78 lookups for polarity inversion, open drain and what not. As we now 79 handle this with generic OF bindings, pull all legacy handling into 80 gpiolib so the library API becomes narrow and deep and handle all 81 legacy bindings internally. (See e.g. commits 6953c57ab172, 82 6a537d48461d etc) 83 84- Delete <linux/of_gpio.h> when all the above is complete and everything 85 uses <linux/gpio/consumer.h> or <linux/gpio/driver.h> instead. 86 87------------------------------------------------------------------------------- 88 89Get rid of <linux/gpio/legacy-of-mm-gpiochip.h> 90 91Work items: 92 93- Get rid of struct of_mm_gpio_chip altogether: use the generic MMIO 94 GPIO for all current users (see below). Delete struct of_mm_gpio_chip, 95 to_of_mm_gpio_chip(), of_mm_gpiochip_add_data(), of_mm_gpiochip_remove(), 96 CONFIG_OF_GPIO_MM_GPIOCHIP from the kernel. 97 98------------------------------------------------------------------------------- 99 100Collect drivers 101 102Collect GPIO drivers from arch/* and other places that should be placed 103in drivers/gpio/gpio-*. Augment platforms to create platform devices or 104similar and probe a proper driver in the gpiolib subsystem. 105 106In some cases it makes sense to create a GPIO chip from the local driver 107for a few GPIOs. Those should stay where they are. 108 109At the same time it makes sense to get rid of code duplication in existing or 110new coming drivers. For example, gpio-ml-ioh should be incorporated into 111gpio-pch. 112 113------------------------------------------------------------------------------- 114 115Generic MMIO GPIO 116 117The GPIO drivers can utilize the generic MMIO helper library in many 118cases, and the helper library should be as helpful as possible for MMIO 119drivers. (drivers/gpio/gpio-mmio.c) 120 121Work items: 122 123- Look over and identify any remaining easily converted drivers and 124 dry-code conversions to MMIO GPIO for maintainers to test 125 126- Expand the MMIO GPIO or write a new library for regmap-based I/O 127 helpers for GPIO drivers on regmap that simply use offsets 128 0..n in some register to drive GPIO lines 129 130- Expand the MMIO GPIO or write a new library for port-mapped I/O 131 helpers (x86 inb()/outb()) and convert port-mapped I/O drivers to use 132 this with dry-coding and sending to maintainers to test 133 134------------------------------------------------------------------------------- 135 136Generic regmap GPIO 137 138In the very similar way to Generic MMIO GPIO convert the users which can 139take advantage of using regmap over direct IO accessors. Note, even in 140MMIO case the regmap MMIO with gpio-regmap.c is preferable over gpio-mmio.c. 141 142------------------------------------------------------------------------------- 143 144GPIOLIB irqchip 145 146The GPIOLIB irqchip is a helper irqchip for "simple cases" that should 147try to cover any generic kind of irqchip cascaded from a GPIO. 148 149- Look over and identify any remaining easily converted drivers and 150 dry-code conversions to gpiolib irqchip for maintainers to test 151 152------------------------------------------------------------------------------- 153 154Moving over to immutable irq_chip structures 155 156Most of the gpio chips implementing interrupt support rely on gpiolib 157intercepting some of the irq_chip callbacks, preventing the structures 158from being made read-only and forcing duplication of structures that 159should otherwise be unique. 160 161The solution is to call into the gpiolib code when needed (resource 162management, enable/disable or unmask/mask callbacks), and to let the 163core code know about that by exposing a flag (IRQCHIP_IMMUTABLE) in 164the irq_chip structure. The irq_chip structure can then be made unique 165and const. 166 167A small number of drivers have been converted (pl061, tegra186, msm, 168amd, apple), and can be used as examples of how to proceed with this 169conversion. Note that drivers using the generic irqchip framework 170cannot be converted yet, but watch this space! 171 172------------------------------------------------------------------------------- 173 174Convert all GPIO chips to using the new, value returning line setters 175 176struct gpio_chip's set() and set_multiple() callbacks are now deprecated. They 177return void and thus do not allow drivers to indicate failure to set the line 178value back to the caller. 179 180We've now added new variants - set_rv() and set_multiple_rv() that return an 181integer. Let's convert all GPIO drivers treewide to use the new callbacks, 182remove the old ones and finally rename the new ones back to the old names. 183 184------------------------------------------------------------------------------- 185 186Extend the sysfs ABI to allow exporting lines by their HW offsets 187 188The need to support the sysfs GPIO class is one of the main obstacles to 189removing the global GPIO numberspace from the kernel. In order to wean users 190off using global numbers from user-space, extend the existing interface with 191new per-gpiochip export/unexport attributes that allow to refer to GPIOs using 192their hardware offsets within the chip. 193 194Encourage users to switch to using them and eventually remove the existing 195global export/unexport attribues. 196 197------------------------------------------------------------------------------- 198 199Remove GPIOD_FLAGS_BIT_NONEXCLUSIVE 200 201GPIOs in the linux kernel are meant to be an exclusive resource. This means 202that the GPIO descriptors (the software representation of the hardware concept) 203are not reference counted and - in general - only one user at a time can 204request a GPIO line and control its settings. The consumer API is designed 205around full control of the line's state as evidenced by the fact that, for 206instance, gpiod_set_value() does indeed drive the line as requested, instead 207of bumping an enable counter of some sort. 208 209A problematic use-case for GPIOs is when two consumers want to use the same 210descriptor independently. An example of such a user is the regulator subsystem 211which may instantiate several struct regulator_dev instances containing 212a struct device but using the same enable GPIO line. 213 214A workaround was introduced in the form of the GPIOD_FLAGS_BIT_NONEXCLUSIVE 215flag but its implementation is problematic: it does not provide any 216synchronization of usage nor did it introduce any enable count meaning the 217non-exclusive users of the same descriptor will in fact "fight" for the 218control over it. This flag should be removed and replaced with a better 219solution, possibly based on the new power sequencing subsystem. 220 221------------------------------------------------------------------------------- 222 223Remove devm_gpiod_unhinge() 224 225devm_gpiod_unhinge() is provided as a way to transfer the ownership of managed 226enable GPIOs to the regulator core. Rather than doing that however, we should 227make it possible for the regulator subsystem to deal with GPIO resources the 228lifetime of which it doesn't control as logically, a GPIO obtained by a caller 229should also be freed by it. 230