Platform Profile support
Subsystems (2)
cover Platform Profile support
This patch series is for the implementation of the platform-profile feature - the ability to determine which mode the platform is in and to change the mode using a sysfs entry.
The first patch is an update of the document I've been working on with review and help from the kernel community. Thank you to everybody for their input.
The second patch implements the platform-profile sysfs and API's needed.
The third patch has Lenovo specific changes in thinkpad_acpi.c to use the new platform-profile implementation and be able to switch between low, medium and high power modes.
Thanks
Mark
Mark Pearson (3):
Documentation: Add documentation for new platform_profile sysfs
attribute
ACPI: platform-profile: Add platform profile support platform/x86: thinkpad_acpi: Add platform profile support
.../ABI/testing/sysfs-platform_profile | 66 +++++ MAINTAINERS | 8 + drivers/acpi/Kconfig | 19 ++ drivers/acpi/Makefile | 1 + drivers/acpi/platform_profile.c | 171 ++++++++++++ drivers/platform/x86/thinkpad_acpi.c | 261 +++++++++++++++++- include/linux/platform_profile.h | 36 +++ 7 files changed, 550 insertions(+), 12 deletions(-) create mode 100644 Documentation/ABI/testing/sysfs-platform_profile create mode 100644 drivers/acpi/platform_profile.c create mode 100644 include/linux/platform_profile.h
Signature · 1 line
2.28.0
PATCH 1/3 Documentation: Add documentation for new platform_profile sysfs attribute
On modern systems the platform performance, temperature, fan and other hardware related characteristics are often dynamically configurable. The profile is often automatically adjusted to the load by somei automatic-mechanism (which may very well live outside the kernel).
These auto platform-adjustment mechanisms often can be configured with one of several 'platform-profiles', with either a bias towards low-power consumption or towards performance (and higher power consumption and thermals).
Introduce a new platform_profile sysfs API which offers a generic API for selecting the performance-profile of these automatic-mechanisms.
--- .../ABI/testing/sysfs-platform_profile | 66 +++++++++++++++++++ 1 file changed, 66 insertions(+) create mode 100644 Documentation/ABI/testing/sysfs-platform_profile
Documentation/ABI/testing/sysfs-platform_profile new file +66 -0
@@ -0,0 +1,66 @@ +Platform profile selection (e.g. /sys/firmware/acpi/platform_profile) + +On modern systems the platform performance, temperature, fan and other +hardware related characteristics are often dynamically configurable. The +profile is often automatically adjusted to the load by some +automatic mechanism (which may very well live outside the kernel). + +These auto platform adjustment mechanisms often can be configured with +one of several platform profiles, with either a bias towards low power +operation or towards performance. + +The purpose of the platform_profile attribute is to offer a generic sysfs +API for selecting the platform profile of these automatic mechanisms. + +Note that this API is only for selecting the platform profile, it is +NOT a goal of this API to allow monitoring the resulting performance +characteristics. Monitoring performance is best done with device/vendor +specific tools such as e.g. turbostat. + +Specifically when selecting a high performance profile the actual achieved +performance may be limited by various factors such as: the heat generated +by other components, room temperature, free air flow at the bottom of a +laptop, etc. It is explicitly NOT a goal of this API to let userspace know +about any sub-optimal conditions which are impeding reaching the requested +performance level. + +Since numbers on their own cannot represent the multiple variables that a +profile will adjust (power consumption, heat generation, etc) this API +uses strings to describe the various profiles. To make sure that userspace +gets a consistent experience this API document defines a fixed set of +profile names. Drivers *must* map their internal profile representation +onto this fixed set. + + +If there is no good match when mapping then a new profile name may be +added. Drivers which wish to introduce new profile names must: +1. Explain why the existing profile names canot be used. +2. Add the new profile name, along with a clear description of the + expected behaviour, to the documentation. + +What: /sys/firmware/acpi/platform_profile_choices +Date: October 2020 +Contact: Hans de Goede <[email protected]> +Description: + This file contains a space-separated list of profiles + supported for this device. + + Drivers must use the following standard profile-names: + + low-power: Low power consumption + cool: Cooler operation + quiet: Quieter operation + balanced: Balance between low power consumption + and performance + performance: High performance operation + + Userspace may expect drivers to offer more than one of these + standard profile names. + +What: /sys/firmware/acpi/platform_profile +Date: October 2020 +Contact: Hans de Goede <[email protected]> +Description: + Reading this file gives the current selected profile for this + device. Writing this file with one of the strings from + available_profiles changes the profile to the new value.
Signature · 1 line
2.28.0
Re: [PATCH 1/3] Documentation: Add documentation for new platform_profile sysfs attribute
Quoted 11 lines · Mark Pearson
On modern systems the platform performance, temperature, fan and other hardware related characteristics are often dynamically configurable. The profile is often automatically adjusted to the load by somei automatic-mechanism (which may very well live outside the kernel).
These auto platform-adjustment mechanisms often can be configured with one of several 'platform-profiles', with either a bias towards low-power consumption or towards performance (and higher power consumption and thermals).
Quoted 4 lines
Introduce a new
platform_profilesysfs API which offers a generic API for selecting the performance-profile of these automatic-mechanisms..../ABI/testing/sysfs-platform_profile | 66 +++++++++++++++++++
Is this in reST format?
Mauro have been converting ABI to be reST compatible (at least).
Signature · 2 lines
With Best Regards, Andy Shevchenko
Re: [External] Re: [PATCH 1/3] Documentation: Add documentation for new platform_profile sysfs attribute
Thanks Andy
Quoted 12 lines · Andy Shevchenko
Quoted 11 lines · Mark Pearson
On modern systems the platform performance, temperature, fan and other hardware related characteristics are often dynamically configurable. The profile is often automatically adjusted to the load by somei automatic-mechanism (which may very well live outside the kernel).
These auto platform-adjustment mechanisms often can be configured with one of several 'platform-profiles', with either a bias towards low-power consumption or towards performance (and higher power consumption and thermals).
Quoted 8 lines
Quoted 4 lines
Introduce a new
platform_profilesysfs API which offers a generic API for selecting the performance-profile of these automatic-mechanisms..../ABI/testing/sysfs-platform_profile | 66 +++++++++++++++++++Is this in reST format?
Mauro have been converting ABI to be reST compatible (at least).
I'm assuming reST is "reStructured Text"? (it's a new one for me)
I'll have a look and check it's compatible and see if I can fix up anything missing. If you have any pointers to a document that has already been updated and is a good reference point that would be appreciated
Thanks
Mark
PATCH 2/3 ACPI: platform-profile: Add platform profile support
In reply to Mark PearsonThis is the initial implementation of the platform-profile feature. It provides the details discussed and outlined in the sysfs-platform_profile document.
Many modern systems have the ability to modify the operating profile to control aspects like fan speed, temperature and power levels. This module provides a common sysfs interface that platform modules can register against to control their individual profile options.
--- MAINTAINERS | 8 ++ drivers/acpi/Kconfig | 19 ++++ drivers/acpi/Makefile | 1 + drivers/acpi/platform_profile.c | 172 +++++++++++++++++++++++++++++++ include/linux/platform_profile.h | 36 +++++++ 5 files changed, 236 insertions(+) create mode 100644 drivers/acpi/platform_profile.c create mode 100644 include/linux/platform_profile.h
MAINTAINERS +8 -0
@@ -436,6 +436,14 @@ S: Orphan F: drivers/platform/x86/wmi.c F: include/uapi/linux/wmi.h +ACPI PLATFORM PROFILE DRIVER +M: Mark Pearson <[email protected]> +L: [email protected] +S: Supported +W: https://01.org/linux-acpi +B: https://bugzilla.kernel.org +F: drivers/acpi/platform_profile.c + AD1889 ALSA SOUND DRIVER L: [email protected] S: Maintained
drivers/acpi/Kconfig +19 -0
@@ -601,3 +601,22 @@ config X86_PM_TIMER You should nearly always say Y here because many modern systems require this timer. + +config ACPI_PLATFORM_PROFILE + tristate "ACPI Platform Profile Driver" + default y + help + This driver adds support for platform-profiles on platforms that + support it. + + Platform-profiles can be used to control the platform behaviour. For + example whether to operate in a lower power mode, in a higher + power performance mode or between the two. + + This driver provides the sysfs interface and is used as the registration + point for platform specific drivers. + + Which profiles are supported is determined on a per-platform basis and + should be obtained from the platform specific driver. + +
drivers/acpi/Makefile +1 -0
@@ -93,6 +93,7 @@ obj-$(CONFIG_ACPI_CPPC_LIB) += cppc_acpi.o obj-$(CONFIG_ACPI_SPCR_TABLE) += spcr.o obj-$(CONFIG_ACPI_DEBUGGER_USER) += acpi_dbg.o obj-$(CONFIG_ACPI_PPTT) += pptt.o +obj-$(CONFIG_ACPI_PLATFORM_PROFILE) += platform_profile.o # processor has its own "processor." module_param namespace processor-y := processor_driver.o
drivers/acpi/platform_profile.c new file +172 -0
172 lines not on this page · raw message
include/linux/platform_profile.h new file +36 -0
@@ -0,0 +1,36 @@ +/* SPDX-License-Identifier: GPL-2.0-or-later */ +/* +* platform_profile.h - platform profile sysfs interface +* +* See Documentation/ABI/testing/sysfs-platform_profile for more information. +*/ + +#ifndef _PLATFORM_PROFILE_H_ +#define _PLATFORM_PROFILE_H_ + +/* +* If more options are added please update profile_str +* array in platform-profile.c +*/ + +enum profile_option { + profile_low, + profile_cool, + profile_quiet, + profile_balance, + profile_perform, + profile_unknown/* Must always be last */ +}; + +struct platform_profile { +unsigned int choices;/* bitmap of available choices */ +int cur_profile;/* Current active profile */ +int (*profile_get)(void ); +int (*profile_set)(int profile); +}; + +extern int platform_profile_register (struct platform_profile *pprof); +extern int platform_profile_unregister (void ); +extern int platform_profile_notify (void ); + +#endif /*_PLATFORM_PROFILE_H_*/
Signature · 1 line
2.28.0
Re: [PATCH 2/3] ACPI: platform-profile: Add platform profile support
Hi
I've added some questions and comments inline.
2020. november 10., kedd 4:31 keltezéssel, Mark Pearson írta:
Quoted 15 lines
[...]
drivers/acpi/platform_profile.c new file +8
-0@@ -0,0 +1,172 @@ +// SPDX-License-Identifier: GPL-2.0-or-later + +/* +* platform_profile.c - Platform profile sysfs interface +*/ + +#include <linux/module.h> +#include <linux/printk.h>
drivers/acpi/platform_profile.c line 18
Quoted 10 lines
drivers/acpi/platform_profile.c new file +10
-0+#include <linux/kobject.h> +#include <linux/sysfs.h> +#include <linux/init.h> +#include <linux/fs.h> +#include <linux/string.h> +#include <linux/device.h> +#include <linux/acpi.h> +#include <linux/mutex.h> +#include <acpi/acpi_bus.h> +#include <linux/platform_profile.h>
This should preferably be alphabetically sorted.
drivers/acpi/platform_profile.c line 20
Quoted 2 lines
drivers/acpi/platform_profile.c new file +2
-0+ +struct platform_profile *cur_profile;
This should be static.
drivers/acpi/platform_profile.c line 23
Quoted 3 lines
drivers/acpi/platform_profile.c new file +3
-0+DEFINE_MUTEX (profile_lock); + +/* Ensure the first char of each profile is unique */
I wholeheartedly disagree that only the first character should be considered. It is not future-proof, potentially subverts user expectation, and even worse, someone could rely on this (undocumented) behaviour.
drivers/acpi/platform_profile.c line 24
Quoted 1 line
drivers/acpi/platform_profile.c new file +1
-0+static char *profile_str[] = {
Why is it not const?
drivers/acpi/platform_profile.c line 30
Quoted 6 lines
drivers/acpi/platform_profile.c new file +6
-0+"Low-power" , +"Cool" , +"Quiet" , +"Balance" , +"Performance" , +"Unknown"
"unknown" is not documented, yet it may be returned to userspace.
drivers/acpi/platform_profile.c line 31
Quoted 1 line
drivers/acpi/platform_profile.c new file +1
-0+};
The documentation has the names in all-lowercase, and in my opinion it'd be better to use lowercase names here as well.
Quoted 7 lines
drivers/acpi/platform_profile.c new file +7
-0+ +static ssize_t platform_profile_choices_show (struct device *dev, +struct device_attribute *attr, +char *buf) +{ +int i; +int ret, count =0 ;
drivers/acpi/platform_profile.c line 48
Quoted 10 lines
drivers/acpi/platform_profile.c new file +10
-0+ +mutex_lock (&profile_lock); +if (!cur_profile) { +mutex_unlock (&profile_lock); +return -ENODEV ; + } + +if (!cur_profile->choices) { +mutex_unlock (&profile_lock); +return snprintf (buf,PAGE_SIZE ,"None" );
"None" is not documented anywhere as far as I can see, maybe an empty line would be better in this case?
drivers/acpi/platform_profile.c line 52
Quoted 4 lines
drivers/acpi/platform_profile.c new file +4
-0+ } + +for (i = profile_low; i < profile_unknown; i++) { +if (cur_profile->choices & (1 << i)) {
BIT(i)?
drivers/acpi/platform_profile.c line 53
Quoted 1 line
drivers/acpi/platform_profile.c new file +1
-0+ ret =snprintf (buf+count,PAGE_SIZE ,"%s " , profile_str[i]);
You could use sysfs_emit_at(). ret is only used in this block, so it could be defined here.
drivers/acpi/platform_profile.c line 55
Quoted 2 lines
drivers/acpi/platform_profile.c new file +2
-0+if (ret <0 ) +break ;
However unlikely this case is, I'm not sure if providing partial values is better than not providing any data at all.
drivers/acpi/platform_profile.c line 59
Quoted 4 lines
drivers/acpi/platform_profile.c new file +4
-0+ count += ret; + } + } +mutex_unlock (&profile_lock);
I think a newline character should be written at the end (possibly overwriting the last space).
Quoted 6 lines
drivers/acpi/platform_profile.c new file +6
-0+return count; +} + +static ssize_t platform_profile_show (struct device *dev, +struct device_attribute *attr, +char *buf)
drivers/acpi/platform_profile.c line 75
Quoted 10 lines
drivers/acpi/platform_profile.c new file +10
-0+{ +enum profile_option profile = profile_unknown; + +mutex_lock (&profile_lock); +if (!cur_profile) { +mutex_unlock (&profile_lock); +return -ENODEV ; + } +if (cur_profile->profile_get) + profile = cur_profile->profile_get ();
I'd assume that profile_get() can return any arbitrary errno, which is then propagated to the "reader", but it seems that's not the case?
I think returning -EOPNOTSUPP would be better if profile_get is NULL.
drivers/acpi/platform_profile.c line 78
Quoted 3 lines
drivers/acpi/platform_profile.c new file +3
-0+mutex_unlock (&profile_lock); + +return snprintf (buf,PAGE_SIZE ,"%s" , profile_str[profile]);
There is sysfs_emit(), as far as I know it is supposed to replace this exact snprintf(...) idiom. Directly indexing the profile_str with an unchecked value here is rather unsafe in my opinion.
Quoted 16 lines
drivers/acpi/platform_profile.c new file +16
-0+} + +static ssize_t platform_profile_store (struct device *dev, +struct device_attribute *attr, +const char *buf,size_t count) +{ +enum profile_option profile; + +mutex_lock (&profile_lock); +if (!cur_profile) { +mutex_unlock (&profile_lock); +return -ENODEV ; + } + +/* Scan for a matching profile */ +for (profile = profile_low; profile < profile_unknown; profile++) {
drivers/acpi/platform_profile.c line 104
Quoted 10 lines
drivers/acpi/platform_profile.c new file +10
-0+if (toupper (buf[0 ]) == profile_str[profile][0 ]) +break ; + } +if (profile == profile_unknown) { +mutex_unlock (&profile_lock); +return -EINVAL ; + } + +if (cur_profile->profile_set) + cur_profile->profile_set (profile);
The return value is entirely discarded? I'd assume it's returned to the "writer". I'm also not sure if ignoring if profile_set is NULL is the best course of action. Maybe returning -EOPNOTSUPP would be better?
Quoted 7 lines
drivers/acpi/platform_profile.c new file +7
-0+ +mutex_unlock (&profile_lock); +return count; +} + +static DEVICE_ATTR_RO (platform_profile_choices); +static DEVICE_ATTR_RW (platform_profile);
drivers/acpi/platform_profile.c line 121
Quoted 10 lines
drivers/acpi/platform_profile.c new file +10
-0+ +static struct attribute *platform_profile_attributes[] = { + &dev_attr_platform_profile_choices.attr, + &dev_attr_platform_profile.attr, +NULL , +}; + +static const struct attribute_group platform_profile_attr_group = { + .attrs = platform_profile_attributes, +};
It's a minor thing, but there is an ATTRIBUTE_GROUPS() macro which could possibly simplify the above part.
Quoted 27 lines
drivers/acpi/platform_profile.c new file +27
-0+ +int platform_profile_notify (void ) +{ +if (!cur_profile) +return -ENODEV ; +sysfs_notify (acpi_kobj,NULL ,"platform_profile" ); +return 0 ; +} +EXPORT_SYMBOL_GPL (platform_profile_notify); + +int platform_profile_register (struct platform_profile *pprof) +{ +mutex_lock (&profile_lock); +/* We can only have one active profile */ +if (cur_profile) { +mutex_unlock (&profile_lock); +return -EEXIST ; + } + cur_profile = pprof; +mutex_unlock (&profile_lock); +return sysfs_create_group (acpi_kobj, &platform_profile_attr_group); +} +EXPORT_SYMBOL_GPL (platform_profile_register); + +int platform_profile_unregister (void ) +{ +mutex_lock (&profile_lock);
drivers/acpi/platform_profile.c line 158
Quoted 10 lines
drivers/acpi/platform_profile.c new file +10
-0+sysfs_remove_group (acpi_kobj, &platform_profile_attr_group); + cur_profile =NULL ; +mutex_unlock (&profile_lock); +return 0 ; +} +EXPORT_SYMBOL_GPL (platform_profile_unregister); + +static int __init platform_profile_init (void ) +{ + cur_profile =NULL ;
If I'm not missing anything, cur_profile will be initialized to NULL, thus this is not needed.
drivers/acpi/platform_profile.c line 162
Quoted 4 lines
drivers/acpi/platform_profile.c new file +4
-0+return 0 ; +} + +static void platform_profile_exit (void )
This should be marked __exit.
Quoted 29 lines
drivers/acpi/platform_profile.c new file +10
-0+{ +sysfs_remove_group (acpi_kobj, &platform_profile_attr_group); + cur_profile =NULL ; +} + +MODULE_AUTHOR ("Mark Pearson <[email protected]>" ); +MODULE_LICENSE ("GPL" ); + +module_init (platform_profile_init); +module_exit (platform_profile_exit);include/linux/platform_profile.h new file +13
-0@@ -0,0 +1,36 @@ +/* SPDX-License-Identifier: GPL-2.0-or-later */ +/* +* platform_profile.h - platform profile sysfs interface +* +* See Documentation/ABI/testing/sysfs-platform_profile for more information. +*/ + +#ifndef _PLATFORM_PROFILE_H_ +#define _PLATFORM_PROFILE_H_ + +/* +* If more options are added please update profile_str +* array in platform-profile.c
include/linux/platform_profile.h line 23
Quoted 10 lines
include/linux/platform_profile.h new file +10
-0+*/ + +enum profile_option { + profile_low, + profile_cool, + profile_quiet, + profile_balance, + profile_perform, + profile_unknown/* Must always be last */ +};
Shouldn't these be prefixed by platform_? And I think it'd be better to have profile_unknown as the first value in the enumeration.
include/linux/platform_profile.h line 25
Quoted 2 lines
include/linux/platform_profile.h new file +2
-0+ +struct platform_profile {
Personally, I think a name like platform_profile_(handler|provider) would be a better fit.
include/linux/platform_profile.h line 26
Quoted 1 line
include/linux/platform_profile.h new file +1
-0+unsigned int choices;/* bitmap of available choices */
Most comments are capitalized.
include/linux/platform_profile.h line 27
Quoted 1 line
include/linux/platform_profile.h new file +1
-0+int cur_profile;/* Current active profile */
cur_profile field doesn't seem to be used here. I see that it's utilized in the thinkpad_acpi driver, but I feel like this does not "belong" here.
include/linux/platform_profile.h line 29
Quoted 2 lines
include/linux/platform_profile.h new file +2
-0+int (*profile_get)(void ); +int (*profile_set)(int profile);
Why does it take an int instead of enum profile_option?
include/linux/platform_profile.h line 35
Quoted 6 lines
include/linux/platform_profile.h new file +6
-0+}; + +extern int platform_profile_register (struct platform_profile *pprof); +extern int platform_profile_unregister (void ); +extern int platform_profile_notify (void ); +
extern could be omitted from here. Although it seems rather "unregulated" whether extern is to be present in header files or not.
Quoted 3 lines
include/linux/platform_profile.h new file +1
-1+#endif /*_PLATFORM_PROFILE_H_*/ --2.28.0
Regards,
Barnabás Pőcze
Re: [External] Re: [PATCH 2/3] ACPI: platform-profile: Add platform profile support
Thanks Barnabas
Quoted 27 lines · Barnabás Pőcze
Hi
I've added some questions and comments inline.
2020. november 10., kedd 4:31 keltezéssel, Mark Pearson írta:
Quoted 19 lines
[...]
drivers/acpi/platform_profile.c new file +12
-0@@ -0,0 +1,172 @@ +// SPDX-License-Identifier: GPL-2.0-or-later + +/* +* platform_profile.c - Platform profile sysfs interface +*/ + +#include <linux/module.h> +#include <linux/printk.h> +#include <linux/kobject.h> +#include <linux/sysfs.h> +#include <linux/init.h> +#include <linux/fs.h>
Quoted 8 lines
drivers/acpi/platform_profile.c line 18
Quoted 6 lines
drivers/acpi/platform_profile.c new file +6
-0+#include <linux/string.h> +#include <linux/device.h> +#include <linux/acpi.h> +#include <linux/mutex.h> +#include <acpi/acpi_bus.h> +#include <linux/platform_profile.h> This should preferably be alphabetically sorted.
Ack - I hadn't realised that was a requirement. I'll update
Quoted 6 lines
drivers/acpi/platform_profile.c line 20
Quoted 4 lines
drivers/acpi/platform_profile.c new file +2
-0+ +struct platform_profile *cur_profile;This should be
static.
Agreed.
Quoted 9 lines
drivers/acpi/platform_profile.c line 23
Quoted 5 lines
drivers/acpi/platform_profile.c new file +3
-0+DEFINE_MUTEX (profile_lock); + +/* Ensure the first char of each profile is unique */ I wholeheartedly disagree that only the first character should be considered. It is not future-proof, potentially subverts user expectation, and even worse, someone could rely on this (undocumented) behaviour.
OK, fair enough. I debated about it and it's nice just to do
echo 'L' > platform_profile
instead of typing in a whole string, but I appreciate it's lazy.
I'm OK with fixing based on your points.
Quoted 5 lines
drivers/acpi/platform_profile.c line 24
Quoted 3 lines
drivers/acpi/platform_profile.c new file +1
-0+static char *profile_str[] = {Why is it not
const?
My mistake. I will fix
Quoted 10 lines
drivers/acpi/platform_profile.c line 30
Quoted 8 lines
drivers/acpi/platform_profile.c new file +6
-0+"Low-power" , +"Cool" , +"Quiet" , +"Balance" , +"Performance" , +"Unknown" "unknown" is not documented, yet it may be returned to userspace.
Ack - I'll look into if it's really needed, but it seemed sensible to have it whilst doing the implementation.
Quoted 6 lines
drivers/acpi/platform_profile.c line 31
Quoted 3 lines
drivers/acpi/platform_profile.c new file +1
-0+};The documentation has the names in all-lowercase, and in my opinion it'd be better to use lowercase names here as well.
Agreed - my bad. I'll fix.
Quoted 14 lines
Quoted 14 lines
drivers/acpi/platform_profile.c new file +12
-0+ +static ssize_t platform_profile_choices_show (struct device *dev, +struct device_attribute *attr, +char *buf) +{ +int i; +int ret, count =0 ; + +mutex_lock (&profile_lock); +if (!cur_profile) { +mutex_unlock (&profile_lock); +return -ENODEV ;
Quoted 8 lines
drivers/acpi/platform_profile.c line 48
Quoted 5 lines
drivers/acpi/platform_profile.c new file +5
-0+ } + +if (!cur_profile->choices) { +mutex_unlock (&profile_lock); +return snprintf (buf,PAGE_SIZE ,"None" );"None" is not documented anywhere as far as I can see, maybe an empty line would be better in this case?
Agreed, I'll fix.
Quoted 8 lines
drivers/acpi/platform_profile.c line 52
Quoted 6 lines
drivers/acpi/platform_profile.c new file +4
-0+ } + +for (i = profile_low; i < profile_unknown; i++) { +if (cur_profile->choices & (1 << i)) {
BIT(i)?
Yes, that would be better.
Quoted 6 lines
drivers/acpi/platform_profile.c line 53
Quoted 3 lines
drivers/acpi/platform_profile.c new file +1
-0+ ret =snprintf (buf+count,PAGE_SIZE ,"%s " , profile_str[i]);You could use
sysfs_emit_at().retis only used in this block, so it could be defined here.
OK - I hadn't come across that function. I'll update to use it.
Quoted 7 lines
drivers/acpi/platform_profile.c line 55
Quoted 4 lines
drivers/acpi/platform_profile.c new file +2
-0+if (ret <0 ) +break ;However unlikely this case is, I'm not sure if providing partial values is better than not providing any data at all.
Interesting point, I think I agree.
I'll look at returning an empty string if an error occurs.
Quoted 9 lines
drivers/acpi/platform_profile.c line 59
Quoted 6 lines
drivers/acpi/platform_profile.c new file +4
-0+ count += ret; + } + } +mutex_unlock (&profile_lock);I think a newline character should be written at the end (possibly overwriting the last space).
OK.
Quoted 14 lines
Quoted 14 lines
drivers/acpi/platform_profile.c new file +12
-0+return count; +} + +static ssize_t platform_profile_show (struct device *dev, +struct device_attribute *attr, +char *buf) +{ +enum profile_option profile = profile_unknown; + +mutex_lock (&profile_lock); +if (!cur_profile) { +mutex_unlock (&profile_lock);
Quoted 8 lines
drivers/acpi/platform_profile.c line 75
Quoted 4 lines
drivers/acpi/platform_profile.c new file +4
-0+return -ENODEV ; + } +if (cur_profile->profile_get) + profile = cur_profile->profile_get ();I'd assume that
profile_get()can return any arbitrary errno, which is then propagated to the "reader", but it seems that's not the case?
I think returning-EOPNOTSUPPwould be better ifprofile_getis NULL.
OK - I went through a few iterations here and it's a bit weak. I'll look at it again.
Quoted 9 lines
drivers/acpi/platform_profile.c line 78
Quoted 5 lines
drivers/acpi/platform_profile.c new file +3
-0+mutex_unlock (&profile_lock); + +return snprintf (buf,PAGE_SIZE ,"%s" , profile_str[profile]);There is
sysfs_emit(), as far as I know it is supposed to replace this exact snprintf(...) idiom. Directly indexing theprofile_strwith an unchecked value here is rather unsafe in my opinion.
Agreed - I'll fix this.
Quoted 24 lines
Quoted 24 lines
drivers/acpi/platform_profile.c new file +22
-0+} + +static ssize_t platform_profile_store (struct device *dev, +struct device_attribute *attr, +const char *buf,size_t count) +{ +enum profile_option profile; + +mutex_lock (&profile_lock); +if (!cur_profile) { +mutex_unlock (&profile_lock); +return -ENODEV ; + } + +/* Scan for a matching profile */ +for (profile = profile_low; profile < profile_unknown; profile++) { +if (toupper (buf[0 ]) == profile_str[profile][0 ]) +break ; + } +if (profile == profile_unknown) { +mutex_unlock (&profile_lock); +return -EINVAL ;
Quoted 8 lines
drivers/acpi/platform_profile.c line 104
Quoted 4 lines
drivers/acpi/platform_profile.c new file +4
-0+ } + +if (cur_profile->profile_set) + cur_profile->profile_set (profile);The return value is entirely discarded? I'd assume it's returned to the "writer". I'm also not sure if ignoring if
profile_setis NULL is the best course of action. Maybe returning-EOPNOTSUPPwould be better?
OK.
Quoted 14 lines
Quoted 14 lines
drivers/acpi/platform_profile.c new file +12
-0+ +mutex_unlock (&profile_lock); +return count; +} + +static DEVICE_ATTR_RO (platform_profile_choices); +static DEVICE_ATTR_RW (platform_profile); + +static struct attribute *platform_profile_attributes[] = { + &dev_attr_platform_profile_choices.attr, + &dev_attr_platform_profile.attr, +NULL ,
Quoted 8 lines
drivers/acpi/platform_profile.c line 121
Quoted 5 lines
drivers/acpi/platform_profile.c new file +5
-0+}; + +static const struct attribute_group platform_profile_attr_group = { + .attrs = platform_profile_attributes, +};It's a minor thing, but there is an
ATTRIBUTE_GROUPS()macro which could possibly simplify the above part.
OK - I'd not come across that and will look at using it.
Quoted 34 lines
Quoted 34 lines
drivers/acpi/platform_profile.c new file +32
-0+ +int platform_profile_notify (void ) +{ +if (!cur_profile) +return -ENODEV ; +sysfs_notify (acpi_kobj,NULL ,"platform_profile" ); +return 0 ; +} +EXPORT_SYMBOL_GPL (platform_profile_notify); + +int platform_profile_register (struct platform_profile *pprof) +{ +mutex_lock (&profile_lock); +/* We can only have one active profile */ +if (cur_profile) { +mutex_unlock (&profile_lock); +return -EEXIST ; + } + cur_profile = pprof; +mutex_unlock (&profile_lock); +return sysfs_create_group (acpi_kobj, &platform_profile_attr_group); +} +EXPORT_SYMBOL_GPL (platform_profile_register); + +int platform_profile_unregister (void ) +{ +mutex_lock (&profile_lock); +sysfs_remove_group (acpi_kobj, &platform_profile_attr_group); + cur_profile =NULL ; +mutex_unlock (&profile_lock); +return 0 ; +}
Quoted 8 lines
drivers/acpi/platform_profile.c line 158
Quoted 5 lines
drivers/acpi/platform_profile.c new file +5
-0+EXPORT_SYMBOL_GPL (platform_profile_unregister); + +static int __init platform_profile_init (void ) +{ + cur_profile =NULL ;If I'm not missing anything,
cur_profilewill be initialized to NULL, thus this is not needed.
I guess I was just playing safe here, I think you're right. I'll remove.
Quoted 8 lines
drivers/acpi/platform_profile.c line 162
Quoted 6 lines
drivers/acpi/platform_profile.c new file +4
-0+return 0 ; +} + +static void platform_profile_exit (void )This should be marked
__exit.
Not sure how I missed that one.... Thanks.
Quoted 36 lines
Quoted 36 lines
drivers/acpi/platform_profile.c new file +10
-0+{ +sysfs_remove_group (acpi_kobj, &platform_profile_attr_group); + cur_profile =NULL ; +} + +MODULE_AUTHOR ("Mark Pearson <[email protected]>" ); +MODULE_LICENSE ("GPL" ); + +module_init (platform_profile_init); +module_exit (platform_profile_exit);include/linux/platform_profile.h new file +18
-0@@ -0,0 +1,36 @@ +/* SPDX-License-Identifier: GPL-2.0-or-later */ +/* +* platform_profile.h - platform profile sysfs interface +* +* See Documentation/ABI/testing/sysfs-platform_profile for more information. +*/ + +#ifndef _PLATFORM_PROFILE_H_ +#define _PLATFORM_PROFILE_H_ + +/* +* If more options are added please update profile_str +* array in platform-profile.c +*/ + +enum profile_option { + profile_low, + profile_cool,
Quoted 8 lines
include/linux/platform_profile.h line 23
Quoted 5 lines
include/linux/platform_profile.h new file +5
-0+ profile_quiet, + profile_balance, + profile_perform, + profile_unknown /* Must always be last */ +};Shouldn't these be prefixed by
platform_? And I think it'd be better to haveprofile_unknownas the first value in the enumeration.
I can add 'platform_'
I liked having profile_unknown as the last value as it makes scanning from 'low' to 'unknown' more future proof if other profiles get added (e.g in platform_profile_choices_show). Is this something you feel strongly about?
Quoted 7 lines
include/linux/platform_profile.h line 25
Quoted 4 lines
include/linux/platform_profile.h new file +2
-0+ +struct platform_profile {Personally, I think a name like
platform_profile_(handler|provider)would be a better fit.
OK - happy to update that.
Quoted 5 lines
include/linux/platform_profile.h line 26
Quoted 3 lines
include/linux/platform_profile.h new file +1
-0+unsigned int choices;/* bitmap of available choices */ Most comments are capitalized.
Ack.
Quoted 6 lines
include/linux/platform_profile.h line 27
Quoted 3 lines
include/linux/platform_profile.h new file +1
-0+int cur_profile;/* Current active profile */
cur_profilefield doesn't seem to be used here. I see that it's utilized in thethinkpad_acpidriver, but I feel like this does not "belong" here.
It seemed a logical place to keep it to me. I understand where you're coming from, and I can change it so it's just internal to thinkpad_acpi.c but it just seemed a nice place to keep it and I'm guessing other implementations would like to have it too.
I'd be interested to see what others have to say on this one.
Quoted 6 lines
include/linux/platform_profile.h line 29
Quoted 4 lines
include/linux/platform_profile.h new file +2
-0+int (*profile_get)(void ); +int (*profile_set)(int profile);Why does it take an
intinstead ofenum profile_option?
Mostly for error conditions to be propagated if needs be, though I have some improvements to do based on the comments above :)
Quoted 11 lines
include/linux/platform_profile.h line 35
Quoted 8 lines
include/linux/platform_profile.h new file +6
-0+}; + +extern int platform_profile_register (struct platform_profile *pprof); +extern int platform_profile_unregister (void ); +extern int platform_profile_notify (void ); +
externcould be omitted from here. Although it seems rather "unregulated" whetherexternis to be present in header files or not.
OK.
Quoted 10 lines
Quoted 5 lines
include/linux/platform_profile.h new file +1
-1+#endif /*_PLATFORM_PROFILE_H_*/ --2.28.0
Regards,
Barnabás Pőcze
Thank you for the detailed review - really appreciate the time you spent going through this.
Mark
Re: [External] Re: [PATCH 2/3] ACPI: platform-profile: Add platform profile support
Hi
Quoted 15 lines
[...]
Quoted 3 lines
Quoted 1 line
+static char *profile_str[] = {
Why is it not
const?My mistake. I will fix
Quoted 10 lines
drivers/acpi/platform_profile.c line 30
Quoted 8 lines
drivers/acpi/platform_profile.c new file +6
-0+"Low-power" , +"Cool" , +"Quiet" , +"Balance" , +"Performance" , +"Unknown" "unknown" is not documented, yet it may be returned to userspace.
Quoted 2 lines
Ack - I'll look into if it's really needed, but it seemed sensible to have it whilst doing the implementation.
I don't advocate for its removal, just that it be documented if it may be returned to userspace.
Quoted 23 lines
Quoted 3 lines
Quoted 3 lines
+};
[...]
Quoted 11 lines
Quoted 11 lines
include/linux/platform_profile.h new file +11
-0+#ifndef _PLATFORM_PROFILE_H_ +#define _PLATFORM_PROFILE_H_ + +/* +* If more options are added please update profile_str +* array in platform-profile.c +*/ + +enum profile_option { + profile_low, + profile_cool,Quoted 8 lines
include/linux/platform_profile.h line 23
Quoted 5 lines
include/linux/platform_profile.h new file +5
-0+ profile_quiet, + profile_balance, + profile_perform, + profile_unknown /* Must always be last */ +};Shouldn't these be prefixed by
platform_? And I think it'd be better to haveprofile_unknownas the first value in the enumeration.
Quoted 5 lines
I can add '
platform_'
I liked havingprofile_unknownas the last value as it makes scanning from 'low' to 'unknown' more future proof if other profiles get added (e.g inplatform_profile_choices_show).
Is this something you feel strongly about?
I don't feel strongly about it, just that right now, for all practical purposes profile_unknown feels like a first-class profile option in the current form of the patch, and it didn't seem right that it can just change. I'd do something like
```
enum performance_profile_option {
performance_profile_unknown,
performance_profile_low,
...
performance_profile_max, /* must be last */
};
```But I don't have a strong preference for either one of them. Maybe someone could chime in and tell us which one is more prevalent/preferred.
And as a side note, I think you could put something like static_assert(ARRAY_SIZE(profile_str) == performance_profile_max); in the code somewhere to make sure there are as many strings in the array as profile options.
You might actually do the following as well:
```
static const char *profile_str[] = {
[performance_profile_unknown] = "unknown",
[performance_profile_low] = "low",
...
};
```I realize I might be a bit too paranoid here. :-) But if you do these three things, or something similar, then the chances of the enum and the array being out of sync (by accident) will be very slim.
Quoted 1 line
[...]
Regards,
Barnabás Pőcze
Re: [External] Re: [PATCH 2/3] ACPI: platform-profile: Add platform profile support
In reply to Mark PearsonHi,
Quoted 10 lines · Mark Pearson
Thanks Barnabas
Quoted 7 lines · Barnabás Pőcze
Hi
I've added some questions and comments inline.
2020. november 10., kedd 4:31 keltezéssel, Mark Pearson írta:
<snip>
Quoted 9 lines
Quoted 8 lines
Quoted 6 lines
+ "Low-power",
+ "Cool",
+ "Quiet",
+ "Balance",
+ "Performance",
+ "Unknown""unknown" is not documented, yet it may be returned to userspace.
Ack - I'll look into if it's really needed, but it seemed sensible to have it whilst doing the implementation.
AFAIK with the currently agreed upon version of the API we do not need this, please remove it. We can always add it later if necessary, but for now I would like to avoid drivers being tempted to use this.
Regards,
Hans
Re: [PATCH 2/3] ACPI: platform-profile: Add platform profile support
In reply to Mark PearsonQuoted 9 lines · Mark Pearson
This is the initial implementation of the platform-profile feature. It provides the details discussed and outlined in the sysfs-platform_profile document.
Many modern systems have the ability to modify the operating profile to control aspects like fan speed, temperature and power levels. This module provides a common sysfs interface that platform modules can register against to control their individual profile options.
...
Quoted 9 lines
drivers/acpi/Kconfig +9
-0+config ACPI_PLATFORM_PROFILE + tristate "ACPI Platform Profile Driver" + default y + help + This driver adds support for platform-profiles on platforms that + support it. + + Platform-profiles can be used to control the platform behaviour. For + example whether to operate in a lower power mode, in a higher
drivers/acpi/Kconfig config X86_PM_TIMER · line 622
Quoted 10 lines
drivers/acpi/Kconfig +9
-0+ power performance mode or between the two. + + This driver provides the sysfs interface and is used as the registration + point for platform specific drivers. + + Which profiles are supported is determined on a per-platform basis and + should be obtained from the platform specific driver. + +
None of the blank lines is enough. But can you consider to find perhaps better place (I imply some logical group of options in the file).
...
Quoted 4 lines
obj-$(CONFIG_ACPI_SPCR_TABLE) += spcr.o
obj-$(CONFIG_ACPI_DEBUGGER_USER) += acpi_dbg.o
obj-$(CONFIG_ACPI_PPTT) += pptt.o
+obj-$(CONFIG_ACPI_PLATFORM_PROFILE) += platform_profile.o
...yes, and this becomes consistent with the above.
...
Quoted 3 lines
+/*
+ *platform_profile.c- Platform profile sysfs interface
+ */
One line. PLease, don't put the file name into the file. If we want to rename it, it will give additional churn and as shown in practice people often forget this change to follow.
...
drivers/acpi/platform_profile.c line 18
Quoted 12 lines
drivers/acpi/platform_profile.c new file +12
-0+#include <linux/module.h> +#include <linux/printk.h> +#include <linux/kobject.h> +#include <linux/sysfs.h> +#include <linux/init.h> +#include <linux/fs.h> +#include <linux/string.h> +#include <linux/device.h> +#include <linux/acpi.h> +#include <linux/mutex.h> +#include <acpi/acpi_bus.h> +#include <linux/platform_profile.h>
Perhaps sorted?
Why do you need a specific acpi_bus.h? I thought acpi.h includes it already, no?
...
Quoted 1 line
+struct
platform_profile*cur_profile;
Better naming since it's a global variable.
Is it supposed to be exported to modules?
...
Quoted 1 line
+DEFINE_MUTEX(profile_lock);
No static?
...
drivers/acpi/platform_profile.c line 24
Quoted 2 lines
drivers/acpi/platform_profile.c new file +2
-0+/* Ensure the first char of each profile is unique */ +static char *profile_str[] = {
static const char * const profile_names[]
Also naming (perhaps like I proposed above?).
drivers/acpi/platform_profile.c line 30
Quoted 7 lines
drivers/acpi/platform_profile.c new file +6
-0+"Low-power" , +"Cool" , +"Quiet" , +"Balance" , +"Performance" , +"Unknown"
Leave the comma here.
Quoted 1 line
+};
...
drivers/acpi/platform_profile.c line 38
Quoted 2 lines
drivers/acpi/platform_profile.c new file +2
-0+int i; +int ret, count =0 ;
count AFAICS should be size_t (or ssize_t).
Can you make them in reversed xmas tree order?
...
Quoted 1 line
+ return
snprintf(buf, PAGE_SIZE, "%s", profile_str[profile]);
Nowadays we have sysfs_emit(), use it.
...
drivers/acpi/platform_profile.c line 97
Quoted 5 lines
drivers/acpi/platform_profile.c new file +5
-0+/* Scan for a matching profile */ +for (profile = profile_low; profile < profile_unknown; profile++) { +if (toupper (buf[0 ]) == profile_str[profile][0 ]) +break ; + }
match_string() / sysfs_match_string() ?
...
drivers/acpi/platform_profile.c line 116
Quoted 5 lines
drivers/acpi/platform_profile.c new file +4
-0+static struct attribute *platform_profile_attributes[] = { + &dev_attr_platform_profile_choices.attr, + &dev_attr_platform_profile.attr, +NULL ,
Drop comma in terminator line.
Quoted 1 line
+};
...
drivers/acpi/platform_profile.c line 172
Quoted 2 lines
drivers/acpi/platform_profile.c new file +2
-0+module_init (platform_profile_init); +module_exit (platform_profile_exit);
Attach them to respective functions.
...
Quoted 2 lines
+/*
+ *platform_profile.h- platform profile sysfs interface
No file name.
Quoted 3 lines
+ *
+ * SeeDocumentation/ABI/testing/sysfs-platform_profilefor more information.
+ */
...
Quoted 4 lines
+/*
+ * If more options are added please updateprofile_str
+ * array inplatform-profile.c
+ */
Kernel doc?
include/linux/platform_profile.h line 22
Quoted 8 lines
include/linux/platform_profile.h new file +7
-0+enum profile_option { + profile_low, + profile_cool, + profile_quiet, + profile_balance, + profile_perform, + profile_unknown/* Must always be last */
Comment is semi-useless. Comma at the end (or its absence) is usually enough to give a clue, but okay, comment makes this explicit.
...
include/linux/platform_profile.h line 27
Quoted 3 lines
include/linux/platform_profile.h new file +3
-0+struct platform_profile { +unsigned int choices;/* bitmap of available choices */ +int cur_profile;/* Current active profile */
Kernel doc?
Quoted 3 lines
include/linux/platform_profile.h new file +3
-0+int (*profile_get)(void ); +int (*profile_set)(int profile); +};
...
include/linux/platform_profile.h line 34
Quoted 3 lines
include/linux/platform_profile.h new file +3
-0+extern int platform_profile_register (struct platform_profile *pprof); +extern int platform_profile_unregister (void ); +extern int platform_profile_notify (void );
extern is not needed.
Signature · 2 lines
With Best Regards, Andy Shevchenko
Re: [External] Re: [PATCH 2/3] ACPI: platform-profile: Add platform profile support
Hi Andy,
Quoted 30 lines · Andy Shevchenko
Quoted 9 lines · Mark Pearson
This is the initial implementation of the platform-profile feature. It provides the details discussed and outlined in the sysfs-platform_profile document.
Many modern systems have the ability to modify the operating profile to control aspects like fan speed, temperature and power levels. This module provides a common sysfs interface that platform modules can register against to control their individual profile options.
...
Quoted 16 lines
drivers/acpi/Kconfig +16
-0+config ACPI_PLATFORM_PROFILE + tristate "ACPI Platform Profile Driver" + default y + help + This driver adds support for platform-profiles on platforms that + support it. + + Platform-profiles can be used to control the platform behaviour. For + example whether to operate in a lower power mode, in a higher + power performance mode or between the two. + + This driver provides the sysfs interface and is used as the registration + point for platform specific drivers. + + Which profiles are supported is determined on a per-platform basis and + should be obtained from the platform specific driver.
Quoted 6 lines
Review
Quoted 2 lines
(quoted diff) +2
-0+ +None of the blank lines is enough. But can you consider to find perhaps better place (I imply some logical group of options in the file).
Will do.
Quoted 12 lines
...
Quoted 4 lines
obj-$(CONFIG_ACPI_SPCR_TABLE) += spcr.o
obj-$(CONFIG_ACPI_DEBUGGER_USER) += acpi_dbg.o
obj-$(CONFIG_ACPI_PPTT) += pptt.o
+obj-$(CONFIG_ACPI_PLATFORM_PROFILE) += platform_profile.o...yes, and this becomes consistent with the above.
...
Quoted 7 lines
Quoted 3 lines
+/*
+ *platform_profile.c- Platform profile sysfs interface
+ */One line. PLease, don't put the file name into the file. If we want to rename it, it will give additional churn and as shown in practice people often forget this change to follow.
Ack.
Quoted 18 lines
...
drivers/acpi/platform_profile.c line 18
Quoted 12 lines
drivers/acpi/platform_profile.c new file +12
-0+#include <linux/module.h> +#include <linux/printk.h> +#include <linux/kobject.h> +#include <linux/sysfs.h> +#include <linux/init.h> +#include <linux/fs.h> +#include <linux/string.h> +#include <linux/device.h> +#include <linux/acpi.h> +#include <linux/mutex.h> +#include <acpi/acpi_bus.h> +#include <linux/platform_profile.h> Perhaps sorted?
Why do you need a specificacpi_bus.h? I thoughtacpi.hincludes it already, no?
Sure - I hadn't realised sorted was a requirement :)
I'll check on acpi_bus.h - I think I initially added acpi_bus.h thinking it was all I needed and then added acpi.h later. I'll clean up if I can.
Quoted 7 lines
...
Quoted 1 line
+struct
platform_profile*cur_profile;Better naming since it's a global variable.
Is it supposed to be exported to modules?
It's internal only, but agreed on improving the naming. (Barabas pointed out that it should be a static too)
Quoted 6 lines
...
Quoted 1 line
+DEFINE_MUTEX(profile_lock);
No static?
Yes, it should be.
Quoted 9 lines
...
drivers/acpi/platform_profile.c line 24
Quoted 2 lines
drivers/acpi/platform_profile.c new file +2
-0+/* Ensure the first char of each profile is unique */ +static char *profile_str[] = {static const char * const profile_names[]
Also naming (perhaps like I proposed above?).
Ack.
Quoted 10 lines
drivers/acpi/platform_profile.c line 30
Quoted 8 lines
drivers/acpi/platform_profile.c new file +6
-0+"Low-power" , +"Cool" , +"Quiet" , +"Balance" , +"Performance" , +"Unknown" Leave the comma here.
Ack.
Quoted 10 lines
Quoted 2 lines
+};
...
drivers/acpi/platform_profile.c line 38
Quoted 2 lines
drivers/acpi/platform_profile.c new file +2
-0+int i; +int ret, count =0 ;count AFAICS should be
size_t(orssize_t).
Can you make them in reversed xmas tree order?
Sure - will fix.
Quoted 6 lines
...
Quoted 1 line
+ return
snprintf(buf, PAGE_SIZE, "%s", profile_str[profile]);Nowadays we have
sysfs_emit(), use it.
OK - that was a new one to me. Will update.
Quoted 10 lines
...
drivers/acpi/platform_profile.c line 97
Quoted 5 lines
drivers/acpi/platform_profile.c new file +5
-0+/* Scan for a matching profile */ +for (profile = profile_low; profile < profile_unknown; profile++) { +if (toupper (buf[0 ]) == profile_str[profile][0 ]) +break ; + }
match_string()/sysfs_match_string()?
Yes, Barnabas picked up on this too. I'll correct.
Quoted 10 lines
...
drivers/acpi/platform_profile.c line 116
Quoted 5 lines
drivers/acpi/platform_profile.c new file +4
-0+static struct attribute *platform_profile_attributes[] = { + &dev_attr_platform_profile_choices.attr, + &dev_attr_platform_profile.attr, +NULL ,Drop comma in terminator line.
Ack.
Quoted 9 lines
Quoted 2 lines
+};
...
drivers/acpi/platform_profile.c line 172
Quoted 2 lines
drivers/acpi/platform_profile.c new file +2
-0+module_init (platform_profile_init); +module_exit (platform_profile_exit);Attach them to respective functions.
OK.
Quoted 7 lines
...
Quoted 2 lines
+/*
+ *platform_profile.h- platform profile sysfs interfaceNo file name.
Ack.
Quoted 5 lines
Quoted 4 lines
+ *
+ * SeeDocumentation/ABI/testing/sysfs-platform_profilefor more information.
+ */
Quoted 8 lines
...
Quoted 4 lines
+/*
+ * If more options are added please updateprofile_str
+ * array inplatform-profile.c
+ */Kernel doc?
OK.
Just in case - I assume this should go int the sysfs-platform-profile doc (patch 1 of the series). Let me know if that's not the expectation.
Quoted 8 lines
Quoted 7 lines
include/linux/platform_profile.h new file +6
-0+enum profile_option { + profile_low, + profile_cool, + profile_quiet, + profile_balance, + profile_perform,
Quoted 4 lines
Quoted 1 line
+
profile_unknown/* Must always be last */Comment is semi-useless. Comma at the end (or its absence) is usually enough to give a clue, but okay, comment makes this explicit.
I prefer explicit but let me know if you feel strongly about this one.
Quoted 8 lines
...
include/linux/platform_profile.h line 27
Quoted 3 lines
include/linux/platform_profile.h new file +3
-0+struct platform_profile { +unsigned int choices;/* bitmap of available choices */ +int cur_profile;/* Current active profile */ Kernel doc?
Sure
Quoted 5 lines
include/linux/platform_profile.h line 30
Quoted 4 lines
include/linux/platform_profile.h new file +3
-0+int (*profile_get)(void ); +int (*profile_set)(int profile); +};
Quoted 8 lines
...
include/linux/platform_profile.h line 34
Quoted 3 lines
include/linux/platform_profile.h new file +3
-0+extern int platform_profile_register (struct platform_profile *pprof); +extern int platform_profile_unregister (void ); +extern int platform_profile_notify (void );extern is not needed.
I will remove.
Thanks for the detailed review - very much appreciated. Mark
Rafael J. Wysocki
maintainerRe: [PATCH 2/3] ACPI: platform-profile: Add platform profile support
In reply to Mark PearsonQuoted 26 lines · Mark Pearson
This is the initial implementation of the platform-profile feature. It provides the details discussed and outlined in the sysfs-platform_profile document.
Many modern systems have the ability to modify the operating profile to control aspects like fan speed, temperature and power levels. This module provides a common sysfs interface that platform modules can register against to control their individual profile options.
Signed-off-by Mark Pearson <[email protected]>--- MAINTAINERS | 8 ++ drivers/acpi/Kconfig | 19 ++++ drivers/acpi/Makefile | 1 + drivers/acpi/platform_profile.c | 172 +++++++++++++++++++++++++++++++ include/linux/platform_profile.h | 36 +++++++ 5 files changed, 236 insertions(+) create mode 100644 drivers/acpi/platform_profile.c create mode 100644 include/linux/platform_profile.hMAINTAINERS +0
-0
MAINTAINERS S: Orphan · line 445
Quoted 10 lines
MAINTAINERS +7
-0@@ -436,6 +436,14 @@ S: Orphan F: drivers/platform/x86/wmi.c F: include/uapi/linux/wmi.h +ACPI PLATFORM PROFILE DRIVER +M: Mark Pearson <[email protected]> +L: [email protected] +S: Supported +W: https://01.org/linux-acpi +B: https://bugzilla.kernel.org +F: drivers/acpi/platform_profile.c
IMO it is premature to add a MAINTAINERS entry for this until it turns out to be really necessary.
Re: [External] Re: [PATCH 2/3] ACPI: platform-profile: Add platform profile support
Hi Rafael
Quoted 30 lines · Rafael J. Wysocki
Quoted 29 lines · Mark Pearson
This is the initial implementation of the platform-profile feature. It provides the details discussed and outlined in the sysfs-platform_profile document.
Many modern systems have the ability to modify the operating profile to control aspects like fan speed, temperature and power levels. This module provides a common sysfs interface that platform modules can register against to control their individual profile options.
Signed-off-by Mark Pearson <[email protected]>--- MAINTAINERS | 8 ++ drivers/acpi/Kconfig | 19 ++++ drivers/acpi/Makefile | 1 + drivers/acpi/platform_profile.c | 172 +++++++++++++++++++++++++++++++ include/linux/platform_profile.h | 36 +++++++ 5 files changed, 236 insertions(+)create mode 100644
drivers/acpi/platform_profile.ccreate mode 100644include/linux/platform_profile.hMAINTAINERS +0
-0@@ -436,6 +436,14 @@ S: Orphan F: drivers/platform/x86/wmi.c F: include/uapi/linux/wmi.h
Quoted 11 lines
MAINTAINERS S: Orphan · line 445
Quoted 7 lines
MAINTAINERS +7
-0+ACPI PLATFORM PROFILE DRIVER +M: Mark Pearson <[email protected]> +L: [email protected] +S: Supported +W: https://01.org/linux-acpi +B: https://bugzilla.kernel.org +F: drivers/acpi/platform_profile.cIMO it is premature to add a
MAINTAINERSentry for this until it turns out to be really necessary.
Not a problem - I'll remove this. I wasn't particularly sure about the etiquette here and I ran checkpatch and it suggested it.
I'm obviously happy to help with fixes etc. My plan is not to dump and run :)
Mark
PATCH 3/3 platform/x86: thinkpad_acpi: Add platform profile support
In reply to Mark PearsonAdd support to thinkpad_acpi for Lenovo platforms that have DYTC version 5 support or newer to use the platform profile feature.
This will allow users to determine and control the platform modes between low-power, balanced operation and performance modes.
--- drivers/platform/x86/thinkpad_acpi.c | 263 +++++++++++++++++++++++++-- 1 file changed, 246 insertions(+), 17 deletions(-)
drivers/platform/x86/thinkpad_acpi.c +246 -17
@@ -72,6 +72,7 @@#include <linux/uaccess.h> #include <acpi/battery.h> #include <acpi/video.h> +#include <linux/platform_profile.h> /* ThinkPad CMOS commands */ #define TP_CMOS_VOLUME_DOWN 0
@@ -9824,10 +9825,40 @@ static struct ibm_struct lcdshadow_driver_data = {* DYTC subdriver, for the Lenovo lapmode feature */ +#define DYTC_CMD_QUERY 0/* To get DYTC status - enable/revision */ +#define DYTC_CMD_SET 1/* To enable/disable IC function mode */ #define DYTC_CMD_GET 2/* To get current IC function and mode */ #define DYTC_GET_LAPMODE_BIT 17/* Set when in lapmode */ +#define DYTC_CMD_RESET 0x1ff/* To reset back to default */ + +#define DYTC_QUERY_ENABLE_BIT 8/* Bit 8 - 0 = disabled, 1 = enabled */ +#define DYTC_QUERY_SUBREV_BIT 16/* Bits 16 - 27 - sub revisision */ +#define DYTC_QUERY_REV_BIT 28/* Bits 28 - 31 - revision */ + +#define DYTC_GET_FUNCTION_BIT 8/* Bits 8-11 - function setting */ +#define DYTC_GET_MODE_BIT 12/* Bits 12-15 - mode setting */ +#define DYTC_GET_LAPMODE_BIT 17/* Bit 17 - lapmode. Set when on lap */ + +#define DYTC_SET_FUNCTION_BIT 12/* Bits 12-15 - funct setting */ +#define DYTC_SET_MODE_BIT 16/* Bits 16-19 - mode setting */ +#define DYTC_SET_VALID_BIT 20/* Bit 20 - 1 = on, 0 = off */ + +#define DYTC_FUNCTION_STD 0/* Function = 0, standard mode */ +#define DYTC_FUNCTION_CQL 1/* Function = 1, lap mode */ +#define DYTC_FUNCTION_MMC 11/* Function = 11, desk mode */ + +#define DYTC_MODE_PERFORM 2/* High power mode aka performance */ +#define DYTC_MODE_QUIET 3/* low power mode aka quiet */ +#define DYTC_MODE_BALANCE 0xF/* default mode aka balance */ + +#define DYTC_DISABLE_CQL ((DYTC_MODE_BALANCE << DYTC_SET_MODE_BIT) | \ + (DYTC_FUNCTION_CQL << DYTC_SET_FUNCTION_BIT) | \ + DYTC_CMD_SET) +#define DYTC_ENABLE_CQL (DYTC_DISABLE_CQL | (1 << DYTC_SET_VALID_BIT))static bool dytc_lapmode; +static bool dytc_available; +static bool dytc_ignore_next_event;static void dytc_lapmode_notify_change (void ) {
@@ -9881,22 +9912,172 @@ static ssize_t dytc_lapmode_show(struct device *dev,static DEVICE_ATTR_RO (dytc_lapmode); -static struct attribute *dytc_attributes[] = { - &dev_attr_dytc_lapmode.attr, -NULL , -}; +#if defined(CONFIG_ACPI_PLATFORM_PROFILE ) +static struct platform_profile dytc_profile; -static const struct attribute_group dytc_attr_group = { - .attrs = dytc_attributes, -}; +static int convert_dytc_to_profile (int dytcmode) +{ +switch (dytcmode) { +case DYTC_MODE_QUIET : +return profile_low; +case DYTC_MODE_BALANCE : +return profile_balance; +case DYTC_MODE_PERFORM : +return profile_perform; +default :/* Unknown mode */ +return profile_unknown; + } +} -static int tpacpi_dytc_init (struct ibm_init_struct *iibm) +static int convert_profile_to_dytc (int profile) +{ +switch (profile) { +case profile_low: +return DYTC_MODE_QUIET ; +case profile_balance: +return DYTC_MODE_BALANCE ; +case profile_perform: +return DYTC_MODE_PERFORM ; +default :/* Unknown profile */ +return -EINVAL ; + } +} + +static int dytc_perfmode_get (int *perfmode,int *funcmode) +{ +int output, err; + +if (!dytc_available) +return -ENODEV ; + + err =dytc_command (DYTC_CMD_GET , &output); +if (err) +return err; + + *funcmode = (output >>DYTC_GET_FUNCTION_BIT ) &0xF ; +if (*funcmode ==DYTC_FUNCTION_CQL ) { +int dummy; +/* +* We can't get the mode when in CQL mode - so we disable CQL +* mode retrieve the mode and then enable it again. +* As disabling/enabling CQL triggers an event we set a flag to +* ignore these events. This will be cleared by the event handler +*/ + dytc_ignore_next_event = true; + err =dytc_command (DYTC_DISABLE_CQL , &dummy); +if (err) +return err; + err =dytc_command (DYTC_CMD_GET , &output); +if (err) +return err; +/* Again ignore this event */ + dytc_ignore_next_event = true; + err =dytc_command (DYTC_ENABLE_CQL , &dummy); +if (err) +return err; + } + *perfmode = (output >>DYTC_GET_MODE_BIT ) &0xF ; +return 0 ; +} + +/* +* dytc_profile_get: Function to register with platform_profile +* handler. Returns current platform profile. +*/ +int dytc_profile_get (void ) {int err; +int funcmode, perfmode; + + err =dytc_perfmode_get (&perfmode, &funcmode); +if (err) +return err; - err =dytc_lapmode_get (&dytc_lapmode); -/* If support isn't available (ENODEV) then don't return an error -* but just don't create the sysfs group +/* Convert Lenovo DYTC profile to platform_profile */ + dytc_profile.cur_profile =convert_dytc_to_profile (perfmode); +return dytc_profile.cur_profile; +} + +/* +* dytc_profile_set: Function to register with platform_profile +* handler. Sets current platform profile. +*/ +int dytc_profile_set (int profile) +{ +int err, dytc_set; +int output; +int cur_perfmode, cur_funcmode; + +if (!dytc_available) +return -ENODEV ; + +if (profile == profile_balance) { +/* To get back to balance mode we just issue a reset command */ + err =dytc_command (DYTC_CMD_RESET , &output); +if (err) +return err; + }else { +int perfmode =convert_profile_to_dytc (profile); + +if (perfmode <0 ) +return perfmode; + +/* Determine if we are in CQL mode. This alters the commands we do */ + err =dytc_perfmode_get (&cur_perfmode, &cur_funcmode); +if (err) +return err; + +if (cur_funcmode ==DYTC_FUNCTION_CQL ) { +/* To set the mode we need to disable CQL first*/ + dytc_ignore_next_event = true;/*ignore event*/ + err =dytc_command (DYTC_DISABLE_CQL , &output); +if (err) +return err; + } + dytc_set = (1 <<DYTC_SET_VALID_BIT ) | + (DYTC_FUNCTION_MMC <<DYTC_SET_FUNCTION_BIT ) | + (perfmode <<DYTC_SET_MODE_BIT ) | +DYTC_CMD_SET ; + err =dytc_command (dytc_set, &output); +if (err) +return err; +if (cur_funcmode ==DYTC_FUNCTION_CQL ) { + dytc_ignore_next_event = true;/*ignore event*/ + err =dytc_command (DYTC_ENABLE_CQL , &output); +if (err) +return err; + } + } +/* Success - update current profile */ + dytc_profile.cur_profile = profile; +return 0 ; +} + +static void dytc_profile_refresh (void ) +{ +int perfmode, funcmode, profile; +int err; + + err =dytc_perfmode_get (&perfmode, &funcmode); +if (err) +return ; + + profile =convert_dytc_to_profile (perfmode); +if (profile != dytc_profile.cur_profile) { + dytc_profile.cur_profile = profile; +platform_profile_notify (); + } +} +#endif + +static int tpacpi_dytc_init (struct ibm_init_struct *iibm) +{ +int err, output; + + err =dytc_command (DYTC_CMD_QUERY , &output); +/* +* If support isn't available (ENODEV) then don't return an error +* and don't create the sysfs group */ if (err == -ENODEV )return 0 ;
@@ -9904,14 +10085,55 @@ static int tpacpi_dytc_init(struct ibm_init_struct *iibm)if (err)return err; -/* Platform supports this feature - create the group */ - err =sysfs_create_group (&tpacpi_pdev->dev.kobj, &dytc_attr_group); -return err; +/* Check DYTC is enabled and supports mode setting */ + dytc_available = false; + dytc_ignore_next_event = false; + +if (output &BIT (DYTC_QUERY_ENABLE_BIT )) { +/* Only DYTC v5.0 and later has this feature. */ +int dytc_version; + + dytc_version = (output >>DYTC_QUERY_REV_BIT ) &0xF ; +if (dytc_version >=5 ) { +dbg_printk (TPACPI_DBG_INIT , +"DYTC version %d: thermal mode available\n" , dytc_version); + dytc_available = true; +#if defined(CONFIG_ACPI_PLATFORM_PROFILE ) +/* Create platform_profile structure and register */ + dytc_profile.choices = (1 << profile_low) | + (1 << profile_balance) | + (1 << profile_perform); + dytc_profile.profile_get = dytc_profile_get; + dytc_profile.profile_set = dytc_profile_set; + dytc_profile.cur_profile = profile_unknown; + err =platform_profile_register (&dytc_profile); +/* +* If for some reason platform_profiles aren't enabled +* don't quit terminally. +*/ +if (err) +return 0 ; +#endif +/* +* Note - this has been deprecated by the input sensor implementation, +* but can't be removed until we confirm user space is no longer using +*/ +dytc_lapmode_get (&dytc_lapmode); +return device_create_file (&tpacpi_pdev->dev, &dev_attr_dytc_lapmode); + } + } +return 0 ; }static void dytc_exit (void ) { -sysfs_remove_group (&tpacpi_pdev->dev.kobj, &dytc_attr_group); +if (dytc_available) { +device_remove_file (&tpacpi_pdev->dev, &dev_attr_dytc_lapmode); +#if defined(CONFIG_ACPI_PLATFORM_PROFILE ) +platform_profile_unregister (); +#endif + dytc_available = false; + } }static struct ibm_struct dytc_driver_data = {
@@ -10095,8 +10317,15 @@ static void tpacpi_driver_event(const unsigned int hkey_event) }if (hkey_event ==TP_HKEY_EV_THM_CSM_COMPLETED ) { -dytc_lapmode_refresh (); -lapsensor_refresh (); +if (dytc_ignore_next_event) + dytc_ignore_next_event = false;/*clear setting*/ +else { +dytc_lapmode_refresh (); +#if defined(CONFIG_ACPI_PLATFORM_PROFILE ) +dytc_profile_refresh (); +#endif +lapsensor_refresh (); + } } }
Signature · 1 line
2.28.0
Re: [PATCH 3/3] platform/x86: thinkpad_acpi: Add platform profile support
Quoted 6 lines · Mark Pearson
Add support to
thinkpad_acpifor Lenovo platforms that have DYTC version 5 support or newer to use the platform profile feature.This will allow users to determine and control the platform modes between low-power, balanced operation and performance modes.
...
drivers/platform/x86/thinkpad_acpi.c static int tpacpi_dytc_init(struct ibm_init_struct *iibm)
Quoted 8 lines
drivers/platform/x86/thinkpad_acpi.c +7
-0+#if defined(CONFIG_ACPI_PLATFORM_PROFILE ) +platform_profile_unregister (); +#endif + dytc_available = false; +#if defined(CONFIG_ACPI_PLATFORM_PROFILE ) +dytc_profile_refresh (); +#endif
Better to use (e.g. test coverage) if (IS_BUILTIN()) / if (IS_ENABLE()) / etc.
Signature · 2 lines
With Best Regards, Andy Shevchenko
Re: [External] Re: [PATCH 3/3] platform/x86: thinkpad_acpi: Add platform profile support
Hi Andy,
Quoted 15 lines · Andy Shevchenko
Quoted 6 lines · Mark Pearson
Add support to
thinkpad_acpifor Lenovo platforms that have DYTC version 5 support or newer to use the platform profile feature.This will allow users to determine and control the platform modes between low-power, balanced operation and performance modes.
...
Quoted 4 lines
drivers/platform/x86/thinkpad_acpi.c +4
-0+#if defined(CONFIG_ACPI_PLATFORM_PROFILE ) +platform_profile_unregister (); +#endif + dytc_available = false;
Quoted 6 lines
drivers/platform/x86/thinkpad_acpi.c static void tpacpi_driver_event(const unsigned int hkey_event) · line 10326
Quoted 3 lines
drivers/platform/x86/thinkpad_acpi.c +3
-0+#if defined(CONFIG_ACPI_PLATFORM_PROFILE ) +dytc_profile_refresh (); +#endif Better to use (e.g. test coverage) if (
IS_BUILTIN()) / if (IS_ENABLE()) / etc.
Thanks - I wasn't aware of those. I'll update
Mark
Quoted 9 lines · Mark Pearson
This patch series is for the implementation of the platform-profile feature - the ability to determine which mode the platform is in and to change the mode using a sysfs entry.
The first patch is an update of the document I've been working on with
review and help from the kernel community. Thank you to everybody for their input.
Quoted 6 lines
The second patch implements the platform-profile sysfs and API's needed.
The third patch has Lenovo specific changes in
thinkpad_acpi.cto use the new platform-profile implementation and be able to switch between low, medium and high power modes.
I've implemented this in power-profiles-daemon:
https://gitlab.freedesktop.org/hadess/power-profiles-daemon/-/merge_requests/41
Was pretty straight forward to implement, so good job :)