On processors starting from Panther Lake, additional workload type hints
are provided.
The hardware analyzes workload residencies over an extended period to
determine whether the workload classification tends toward idle/battery
life states or sustained/performance states. Based on this long-term
analysis, it classifies:
Power Classification: If the workload exhibits more idle or battery
life residencies, it is classified as "power". This is indicated by
setting bit 4 of the current workload type.
Performance Classification: If the workload exhibits more sustained
or performance residencies, it is classified as "performance". This
is indicated by clearing bit 4 of the current workload type.
This approach enables applications to ignore short-term workload
fluctuations and instead respond to longer-term power vs. performance
trends. Hints of this type are called slow workload hints.
To get notifications for slow workload hints, bit 22 in the thermal
mailbox can be used for configuring workload interrupts. It is possible
to exclusively enable slow workload hints or enable them in addition to
the current workload hints.
To enable slow workload hints, a new sysfs attribute is added to the
existing workload hint attributes:
workload_slow_hint_enable (RW): Write 1 to enable, 0 to disable.
Reading this attribute shows the current state.
This attribute is not present on any previous generation of processors.
Signed-off-by: Srinivas Pandruvada <srinivas.pandruvada@linux.intel.com>
[ rjw: Dropped redundant local variables, changelog edits ]
Link: https://patch.msgid.link/20251218222559.4110027-2-srinivas.pandruvada@linux.intel.com
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Corrected a few spelling mistakes
functionalty ==> functionality
in Documentation/driver-api/cxl/devices/device-types.rst
adjascent ==> adjacent
in Documentation/driver-api/cxl/platform/example-configurations/one-dev-per-hb.rst
succeessful ==> successful
in Documentation/driver-api/thermal/exynos_thermal_emulation.rst
Signed-off-by: Ranganath V N <vnranganath.20@gmail.com>
Reviewed-by: Randy Dunlap <rdunlap@infradead.org>
Signed-off-by: Jonathan Corbet <corbet@lwn.net>
Link: https://lore.kernel.org/r/20250814184304.20448-1-vnranganath.20@gmail.com
Firmware-based thermal temperature control loops may aggressively
throttle performance to prevent temperature overshoots relative to the
defined target temperature. This can negatively impact performance. User
space may prefer to prioritize performance, even if it results in
temperature overshoots with in acceptable range.
For example, user space might tolerate temperature overshoots when the
device is placed on a desk, as opposed to when it's on a lap. To
accommodate such scenarios, an optional attribute is provided to specify
a tolerance level for temperature overshoots while maintaining acceptable
performance.
Attribute:
thermal_tolerance: This attribute ranges from 0 to 7, where 0 represents
the most aggressive control to avoid any temperature overshoots, and 7
represents a more graceful approach, favoring performance even at the
expense of temperature overshoots.
Note: This level may not scale linearly. For example, a value of 3 does not
necessarily imply a 50% improvement in performance compared to a value of 0.
Signed-off-by: Srinivas Pandruvada <srinivas.pandruvada@linux.intel.com>
Reviewed-by: Zhang Rui <rui.zhang@intel.com>
Link: https://patch.msgid.link/20250613214923.2910397-1-srinivas.pandruvada@linux.intel.com
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
There are no more callers of thermal_zone_bind_cooling_device() and
thermal_zone_unbind_cooling_device(), so drop them along with all of
the corresponding headers, code and documentation.
Moreover, because the .bind() and .unbind() thermal zone callbacks would
only be used when the above functions, respectively, were called, drop
them as well along with all of the code related to them.
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Reviewed-by: Zhang Rui <rui.zhang@intel.com>
Acked-by: Huisong Li <lihuisong@huawei.com>
Reviewed-by: Daniel Lezcano <daniel.lezcano@linaro.org>
Link: https://patch.msgid.link/4251116.1IzOArtZ34@rjwysocki.net
There are no more users of get_thermal_instance(), so drop it.
While at it, replace get_instance() returning a pointer to struct
thermal_instance with thermal_instance_present() returning a bool
which is more straightforward.
No functional impact.
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Reviewed-by: Lukasz Luba <lukasz.luba@arm.com>
Link: https://patch.msgid.link/2014591.usQuhbGJ8B@rjwysocki.net
[ rjw: Dropped get_thermal_instance() documentation ]
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
The thermal sysfs API document is outdated. One of the problems with
it is that is still documents thermal_zone_device_register() which
does not exit any more and it does not reflect the current thermal
zone operations definition.
Replace the thermal_zone_device_register() description in it with
a thermal_zone_device_register_with_trips() description, including
an update of the thermal zone operations list.
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Reviewed-by: Lukasz Luba <lukasz.luba@arm.com>
Link: https://patch.msgid.link/2767845.mvXUDI8C0e@rjwysocki.net
When the hardware reduces the power to the minimum possible, the power
floor is notified via an interrupt.
This can happen when user space requests a power limit via powercap RAPL
interface, which forces the system to enter to the lowest power. This
power floor indication can be used as a hint to resort to other methods
of reducing power than via RAPL power limit.
Before power floor status can be read or the firmware can trigger
notifications regarding it, it needs to be configured via a mailbox
command. The actual power floor status is read via bit 39 of MMIO
offset 0x5B18 of the processor thermal PCI device.
To show the current power floor status and get notification
on a sysfs attribute, add 2 new attributes to
/sys/bus/pci/devices/0000\:00\:04.0/power_limits/
power_floor_enable : This attribute is present when power floor
notifications are supported. This attribute allows to enable/disable
power floor notifications.
power_floor_status : This attribute is present when power floor
notifications are supported. When enabled via power_floor_enable, this
attribute shows the current power floor status.
The power floor implementation provides interfaces which are called
from the sysfs callbacks to enable/disable and read power floor
status. It also provides two additional interfaces to check if the
current processor thermal device interrupt is for power floor status
and to send notifications to user space.
Signed-off-by: Srinivas Pandruvada <srinivas.pandruvada@linux.intel.com>
[ rjw: Changelog and documentation changes edits ]
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Prior to Meteor Lake processor generation, user space can pass workload
type request to the firmware. Then firmware can optimize power based on
the indicated workload type. User space also uses workload type requests
to implement its own heuristics.
The firmware in Meteor Lake processor generation is capable of predicting
workload type without software help.
To avoid duplicate processing, add a sysfs interface allowing user space
to obtain the workload hint from the firmware instead of trying to
predict the workload type by itself.
This workload hint is passed from the firmware via MMIO offset 0x5B18 of
the processor thermal PCI device. Before workload hints can be produced by
the firmware, it needs to be configured via a mailbox command. This
mailbox command turns ON the workload hint and it allows to program a
notification delay to control the rate of notifications.
The notification delay can be changed from user space vis sysfs.
Attribute group 'workload_hint' in sysfs is used for implementing the
workload hints interface between user space and the kernel.
It contains the following attributes:
workload_type_enable:
Enables/disables workload type hints from the firmware.
notification_delay_ms:
Notification delay in milliseconds.
workload_type_index:
The current workload type index predicted by the firmware (see
the documentation changes below for supported index values and
their meaning).
Signed-off-by: Srinivas Pandruvada <srinivas.pandruvada@linux.intel.com>
[ rjw: Changelog edits, documentation edits, whitespace adjustments ]
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Add support for DLVR (Digital Linear Voltage Regulator) attributes,
which can be used to control RFIM.
Here instead of "fivr" another directory "dlvr" is created with DLVR
attributes:
/sys/bus/pci/devices/0000:00:04.0/dlvr
├── dlvr_freq_mhz
├── dlvr_freq_select
├── dlvr_hardware_rev
├── dlvr_pll_busy
├── dlvr_rfim_enable
└── dlvr_spread_spectrum_pct
└── dlvr_control_mode
└── dlvr_control_lock
Attributes
dlvr_freq_mhz (RO):
Current DLVR PLL frequency in MHz.
dlvr_freq_select (RW):
Sets DLVR PLL clock frequency.
dlvr_hardware_rev (RO):
DLVR hardware revision.
dlvr_pll_busy (RO):
PLL can't accept frequency change when set.
dlvr_rfim_enable (RW):
0: Disable RF frequency hopping, 1: Enable RF frequency hopping.
dlvr_control_mode (RW):
Specifies how frequencies are spread. 0: Down spread, 1: Spread in Center.
dlvr_control_lock (RW):
1: future writes are ignored.
dlvr_spread_spectrum_pct (RW)
A write to this register updates the DLVR spread spectrum percent value.
Signed-off-by: Srinivas Pandruvada <srinivas.pandruvada@linux.intel.com>
[ rjw: Subject edits ]
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Create a folder "thermal" under Documentation/admin-guide and move
intel_powerclamp documentation to this folder.
Signed-off-by: Srinivas Pandruvada <srinivas.pandruvada@linux.intel.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
It is possible that the system manufacturer locks down thermal tuning
beyond what is usually done on the given platform. In that case user
space calibration tools should not try to adjust the thermal
configuration of the system.
To allow user space to check if that is the case, add a new sysfs
attribute "production_mode" that will be present when the ACPI DCFG
method is present under the INT3400 device object in the ACPI Namespace.
Signed-off-by: Srinivas Pandruvada <srinivas.pandruvada@linux.intel.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
The thermal ABI and the internal development details are described in:
Documentation/driver-api/thermal/sysfs-api.rst
Move the sysfs ABI description to Documentation/ABI, ensuring that
scripts/get_abi.pl will properly parse it.
While here, also update the thermal record in MAINTAINERS to point to
the documentation.
Signed-off-by: Mauro Carvalho Chehab <mchehab+huawei@kernel.org>
Acked-by: Daniel Lezcano <daniel.lezcano@linaro.org>
[ rjw: Changelog end subject edits ]
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Required bump from v5.13-rc3 to v5.14-rc3, and to pick up sysfb compilation fixes.
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Pull documentation updates from Jonathan Corbet:
"It has been a relatively quiet cycle in docsland.
- As promised, the minimum Sphinx version to build the docs is now
1.7, and we have dropped support for Python 2 entirely. That
allowed the removal of a bunch of compatibility code.
- A set of treewide warning fixups from Mauro that I applied after it
became clear nobody else was going to deal with them.
- The automarkup mechanism can now create cross-references from
relative paths to RST files.
- More translations, typo fixes, and warning fixes"
* tag 'docs-5.12' of git://git.lwn.net/linux: (75 commits)
docs: kernel-hacking: be more civil
docs: Remove the Microsoft rhetoric
Documentation/admin-guide: kernel-parameters: Update nohlt section
doc/admin-guide: fix spelling mistake: "perfomance" -> "performance"
docs: Document cross-referencing using relative path
docs: Enable usage of relative paths to docs on automarkup
docs: thermal: fix spelling mistakes
Documentation: admin-guide: Update kvm/xen config option
docs: Make syscalls' helpers naming consistent
coding-style.rst: Avoid comma statements
Documentation: /proc/loadavg: add 3 more field descriptions
Documentation/submitting-patches: Add blurb about backtraces in commit messages
Docs: drop Python 2 support
Move our minimum Sphinx version to 1.7
Documentation: input: define ABS_PRESSURE/ABS_MT_PRESSURE resolution as grams
scripts/kernel-doc: add internal hyperlink to DOC: sections
Update Documentation/admin-guide/sysctl/fs.rst
docs: Update DTB format references
docs: zh_CN: add iio index.rst translation
docs/zh_CN: add iio ep93xx_adc.rst translation
...
The code was reorganized in 2012 with the commit 0c01ebbfd3.
The main change is a loop on the trip points array and a unconditional
call to the throttle() ops of the governors for each of them even if
the trip temperature is not reached yet.
With this change, the 'forced_passive' is no longer checked in the
thermal_zone_device_update() function but in the step wise governor's
throttle() callback.
As the force_passive does no belong to the trip point array, the
thermal_zone_device_update() can not compare with the specified
passive temperature, thus does not detect the passive limit has been
crossed. Consequently, throttle() is never called and the
'forced_passive' branch is unreached.
In addition, the default processor cooling device is not automatically
bound to the thermal zone if there is not passive trip point, thus the
'forced_passive' can not operate.
If there is an active trip point, then the throttle function will be
called to mitigate at this temperature and the 'forced_passive' will
override the mitigation of the active trip point in this case but with
the default cooling device bound to the thermal zone, so usually a
fan, and that is not a passive cooling effect.
Given the regression exists since more than 8 years, nobody complained
and at the best of my knowledge there is no bug open in
https://bugzilla.kernel.org, it is reasonable to say it is unused.
Remove the 'forced_passive' related code.
Signed-off-by: Daniel Lezcano <daniel.lezcano@linaro.org>
Reviewed-by: Thara Gopinath <thara.gopinath@linaro.org>
Link: https://lore.kernel.org/r/20201214233811.485669-1-daniel.lezcano@linaro.org
