1How to Get Your Patch Accepted Into the Hwmon Subsystem 2======================================================= 3 4This text is a collection of suggestions for people writing patches or 5drivers for the hwmon subsystem. Following these suggestions will greatly 6increase the chances of your change being accepted. 7 8 91. General 10---------- 11 12* It should be unnecessary to mention, but please read and follow: 13 14 - Documentation/process/submit-checklist.rst 15 - Documentation/process/submitting-patches.rst 16 - Documentation/process/coding-style.rst 17 18* Please run your patch through 'checkpatch --strict'. There should be no 19 errors, no warnings, and few if any check messages. If there are any 20 messages, please be prepared to explain. 21 22* Please use the standard multi-line comment style. Do not mix C and C++ 23 style comments in a single driver (with the exception of the SPDX license 24 identifier). 25 26* If your patch generates checkpatch errors, warnings, or check messages, 27 please refrain from explanations such as "I prefer that coding style". 28 Keep in mind that each unnecessary message helps hiding a real problem, 29 and a consistent coding style makes it easier for others to understand 30 and review the code. 31 32* Please test your patch thoroughly. We are not your test group. 33 Sometimes a patch can not or not completely be tested because of missing 34 hardware. In such cases, you should test-build the code on at least one 35 architecture. If run-time testing was not achieved, it should be written 36 explicitly below the patch header. 37 38* If your patch (or the driver) is affected by configuration options such as 39 CONFIG_SMP, make sure it compiles for all configuration variants. 40 41 422. Adding functionality to existing drivers 43------------------------------------------- 44 45* Make sure the documentation in Documentation/hwmon/<driver_name>.rst is up to 46 date. 47 48* Make sure the information in Kconfig is up to date. 49 50* If the added functionality requires some cleanup or structural changes, split 51 your patch into a cleanup part and the actual addition. This makes it easier 52 to review your changes, and to bisect any resulting problems. 53 54* Never mix bug fixes, cleanup, and functional enhancements in a single patch. 55 56 573. New drivers 58-------------- 59 60* Running your patch or driver file(s) through checkpatch does not mean its 61 formatting is clean. If unsure about formatting in your new driver, run it 62 through Lindent. Lindent is not perfect, and you may have to do some minor 63 cleanup, but it is a good start. 64 65* Consider adding yourself to MAINTAINERS. 66 67* Document the driver in Documentation/hwmon/<driver_name>.rst. 68 69* Add the driver to Kconfig and Makefile in alphabetical order. 70 71* Make sure that all dependencies are listed in Kconfig. 72 73* Please list include files in alphabetic order. 74 75* Please align continuation lines with '(' on the previous line. 76 77* Avoid forward declarations if you can. Rearrange the code if necessary. 78 79* Avoid macros to generate groups of sensor attributes. It not only confuses 80 checkpatch, but also makes it more difficult to review the code. 81 82* Avoid calculations in macros and macro-generated functions. While such macros 83 may save a line or so in the source, it obfuscates the code and makes code 84 review more difficult. It may also result in code which is more complicated 85 than necessary. Such macros may also evaluate their arguments multiple times. 86 This leads to Time-of-Check to Time-of-Use (TOCTOU) race conditions when 87 accessing shared data without locking, for example when calculating values in 88 sysfs show functions. Use inline functions or just regular functions instead. 89 90* Limit the number of kernel log messages. In general, your driver should not 91 generate an error message just because a runtime operation failed. Report 92 errors to user space instead, using an appropriate error code. Keep in mind 93 that kernel error log messages not only fill up the kernel log, but also are 94 printed synchronously, most likely with interrupt disabled, often to a serial 95 console. Excessive logging can seriously affect system performance. 96 97* Use devres functions whenever possible to allocate resources. For rationale 98 and supported functions, please see Documentation/driver-api/driver-model/devres.rst. 99 If a function is not supported by devres, consider using devm_add_action(). 100 101* If the driver has a detect function, make sure it is silent. Debug messages 102 and messages printed after a successful detection are acceptable, but it 103 must not print messages such as "Chip XXX not found/supported". 104 105 Keep in mind that the detect function will run for all drivers supporting an 106 address if a chip is detected on that address. Unnecessary messages will just 107 pollute the kernel log and not provide any value. 108 109* Provide a detect function if and only if a chip can be detected reliably. 110 111* Only the following I2C addresses shall be probed: 0x18-0x1f, 0x28-0x2f, 112 0x48-0x4f, 0x58, 0x5c, 0x73 and 0x77. Probing other addresses is strongly 113 discouraged as it is known to cause trouble with other (non-hwmon) I2C 114 chips. If your chip lives at an address which can't be probed then the 115 device will have to be instantiated explicitly (which is always better 116 anyway.) 117 118* Avoid writing to chip registers in the detect function. If you have to write, 119 only do it after you have already gathered enough data to be certain that the 120 detection is going to be successful. 121 122 Keep in mind that the chip might not be what your driver believes it is, and 123 writing to it might cause a bad misconfiguration. 124 125* Make sure there are no race conditions in the probe function. Specifically, 126 completely initialize your chip and your driver first, then register with 127 the hwmon subsystem. 128 129* Use devm_hwmon_device_register_with_info() or, if your driver needs a remove 130 function, hwmon_device_register_with_info() to register your driver with the 131 hwmon subsystem. Try using devm_add_action() instead of a remove function if 132 possible. Do not use any of the deprecated registration functions. 133 134* Your driver should be buildable as module. If not, please be prepared to 135 explain why it has to be built into the kernel. 136 137* Do not provide support for deprecated sysfs attributes. 138 139* Do not create non-standard attributes unless really needed. If you have to use 140 non-standard attributes, or you believe you do, discuss it on the mailing list 141 first. Either case, provide a detailed explanation why you need the 142 non-standard attribute(s). 143 Standard attributes are specified in Documentation/hwmon/sysfs-interface.rst. 144 145* When deciding which sysfs attributes to support, look at the chip's 146 capabilities. While we do not expect your driver to support everything the 147 chip may offer, it should at least support all limits and alarms. 148 149* Last but not least, please check if a driver for your chip already exists 150 before starting to write a new driver. Especially for temperature sensors, 151 new chips are often variants of previously released chips. In some cases, 152 a presumably new chip may simply have been relabeled. 153