| Age | Commit message (Collapse) | Author |
|---|
|
Contiguous PF GGTT VMAs can be scarce after creating VFs.
Increase the GuC buffer cache size to 8M for PF so that we can fit GuC
migration data (which currently maxes out at just over 4M) and use the
cache instead of allocating fresh BOs.
Reviewed-by: Michal Wajdeczko <michal.wajdeczko@intel.com>
Link: https://patch.msgid.link/20251112132220.516975-13-michal.winiarski@intel.com
Signed-off-by: Michał Winiarski <michal.winiarski@intel.com>
|
|
An upcoming change will use GuC buffer cache as a place where GuC
migration data will be stored, and the memory requirement for that is
larger than indirect data.
Allow the caller to pass the size based on the intended usecase.
Reviewed-by: Michal Wajdeczko <michal.wajdeczko@intel.com>
Link: https://patch.msgid.link/20251112132220.516975-12-michal.winiarski@intel.com
Signed-off-by: Michał Winiarski <michal.winiarski@intel.com>
|
|
In upcoming changes the cached buffers are going to be used to read data
produced by the GuC. Add a counterpart to flush, which synchronizes the
CPU-side of suballocation with the GPU data and propagate the interface
to GuC Buffer Cache.
Reviewed-by: Michal Wajdeczko <michal.wajdeczko@intel.com>
Link: https://patch.msgid.link/20251112132220.516975-11-michal.winiarski@intel.com
Signed-off-by: Michał Winiarski <michal.winiarski@intel.com>
|
|
The size is normally used to make a decision on when to stop the device
(mainly when it's in a pre_copy state).
Reviewed-by: Michal Wajdeczko <michal.wajdeczko@intel.com>
Link: https://patch.msgid.link/20251112132220.516975-10-michal.winiarski@intel.com
Signed-off-by: Michał Winiarski <michal.winiarski@intel.com>
|
|
The descriptor reuses the KLV format used by GuC and contains metadata
that can be used to quickly fail migration when source is incompatible
with destination.
Reviewed-by: Michal Wajdeczko <michal.wajdeczko@intel.com>
Link: https://patch.msgid.link/20251112132220.516975-9-michal.winiarski@intel.com
Signed-off-by: Michał Winiarski <michal.winiarski@intel.com>
|
|
Add debugfs handlers for migration state and handle bitstream
.read()/.write() to convert from bitstream to/from migration data
packets.
As descriptor/trailer are handled at this layer - add handling for both
save and restore side.
Reviewed-by: Michal Wajdeczko <michal.wajdeczko@intel.com>
Link: https://patch.msgid.link/20251112132220.516975-8-michal.winiarski@intel.com
Signed-off-by: Michał Winiarski <michal.winiarski@intel.com>
|
|
Now that it's possible to free the packets - connect the restore
handling logic with the ring.
The helpers will also be used in upcoming changes that will start
producing migration data packets.
Reviewed-by: Michal Wajdeczko <michal.wajdeczko@intel.com>
Link: https://patch.msgid.link/20251112132220.516975-7-michal.winiarski@intel.com
Signed-off-by: Michał Winiarski <michal.winiarski@intel.com>
|
|
Migration data is queued in a per-GT ptr_ring to decouple the worker
responsible for handling the data transfer from the .read() and .write()
syscalls.
Add the data structures and handlers that will be used in future
commits.
Reviewed-by: Michal Wajdeczko <michal.wajdeczko@intel.com>
Link: https://patch.msgid.link/20251112132220.516975-6-michal.winiarski@intel.com
Signed-off-by: Michał Winiarski <michal.winiarski@intel.com>
|
|
The states will be used by upcoming changes to produce (in case of save)
or consume (in case of resume) the VF migration data.
Reviewed-by: Michal Wajdeczko <michal.wajdeczko@intel.com>
Link: https://patch.msgid.link/20251112132220.516975-5-michal.winiarski@intel.com
Signed-off-by: Michał Winiarski <michal.winiarski@intel.com>
|
|
In upcoming changes, the number of states will increase as a result of
introducing SAVE and RESTORE states.
This means that using unsigned long as underlying storage won't work on
32-bit architectures, as we'll run out of bits.
Use bitmap instead.
Reported-by: kernel test robot <lkp@intel.com>
Closes: https://lore.kernel.org/oe-kbuild-all/202510231918.XlOqymLC-lkp@intel.com/
Reviewed-by: Michal Wajdeczko <michal.wajdeczko@intel.com>
Link: https://patch.msgid.link/20251112132220.516975-4-michal.winiarski@intel.com
Signed-off-by: Michał Winiarski <michal.winiarski@intel.com>
|
|
Upcoming changes will allow users to control VF state and obtain its
migration data with a device-level granularity (not tile/gt).
Change the data structures to reflect that and move the GT-level
migration init to happen after device-level init.
Reviewed-by: Michal Wajdeczko <michal.wajdeczko@intel.com>
Link: https://patch.msgid.link/20251112132220.516975-3-michal.winiarski@intel.com
Signed-off-by: Michał Winiarski <michal.winiarski@intel.com>
|
|
Since commit 4eb0aab6e4434 ("drm/xe/guc: Bump minimum required GuC
version to v70.29.2"), the minimum GuC version required by the driver
is v70.29.2, which should already include everything that we need for
migration.
Remove the version check.
Suggested-by: Michal Wajdeczko <michal.wajdeczko@intel.com>
Reviewed-by: Michal Wajdeczko <michal.wajdeczko@intel.com>
Link: https://patch.msgid.link/20251112132220.516975-2-michal.winiarski@intel.com
Signed-off-by: Michał Winiarski <michal.winiarski@intel.com>
|
|
Add a v4l2 sub-device driver for the Sony IMX111 image sensor. This is a
camera sensor using the i2c bus for control and the csi-2 bus for data.
The following features are supported:
- manual exposure, digital and analog gain control support
- pixel rate/link freq control support
- supported resolution up to 3280x2464 for single shot capture
- supported resolution up to 1920x1080 @ 30fps for video
- supported bayer order output SGBRG10 and SGBRG8
Camera module seems to be partially compatible with Nokia SMIA but it
lacks a few registers required for clock calculations and has different
vendor-specific per-mode configurations which makes it incompatible with
existing CCS driver.
Signed-off-by: Svyatoslav Ryhel <clamor95@gmail.com>
[Sakari Ailus: Fix set_fmt() callback a bit and idle after autosuspend.]
Signed-off-by: Sakari Ailus <sakari.ailus@linux.intel.com>
Signed-off-by: Hans Verkuil <hverkuil+cisco@kernel.org>
|
|
Add a default case to the switch case statement in mipi_exec_send_packet
to prevent attempts to read an uninitialized ret value. It's unlikely
the default case will ever occur during regular exeuction, but if more
MIPI DSI Processor-to-Peripheral transaction types are ever added, then
having this in place will be a useful safety guard.
Signed-off-by: Jonathan Cavitt <jonathan.cavitt@intel.com>
Fixes: 23db1577ce2d ("drm/i915/dsi: log send packet sequence errors")
Reviewed-by: Jani Nikula <jani.nikula@intel.com>
Link: https://patch.msgid.link/20251104164150.16795-2-jonathan.cavitt@intel.com
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
|
|
Introduce IMX219_EXPOSURE_OFFSET (4) and use it instead of the literal
'4' when computing the maximum coarse exposure. The IMX219 datasheet
specifies the maximum storage time as frame_length_lines - 4.
(Ref: Datasheet section 5-7-1)
Also fix one indentation issue for consistency.
Signed-off-by: Tarang Raval <tarang.raval@siliconsignals.io>
Reviewed-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Signed-off-by: Sakari Ailus <sakari.ailus@linux.intel.com>
Signed-off-by: Hans Verkuil <hverkuil+cisco@kernel.org>
|
|
Propagate return values from __v4l2_ctrl_modify_range() and
__v4l2_ctrl_s_ctrl() in imx219_set_ctrl() and imx219_set_pad_format().
This ensures proper error handling instead of ignoring possible
failures. Also return the result of imx219_set_pad_format() from
imx219_init_state().
Signed-off-by: Tarang Raval <tarang.raval@siliconsignals.io>
Reviewed-by: Isaac Scott <isaac.scott@ideasonboard.com>
Reviewed-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
Signed-off-by: Sakari Ailus <sakari.ailus@linux.intel.com>
Signed-off-by: Hans Verkuil <hverkuil+cisco@kernel.org>
|
|
Add support for monochrome media bus formats, for use with monochrome/IR
camera sensors.
This has been tested on a Dell XPS 13 9320 with OV1A1B IR sensor.
Signed-off-by: Hans de Goede <hansg@kernel.org>
Reviewed-by: Bingbu Cao <bingbu.cao@intel.com>
Signed-off-by: Sakari Ailus <sakari.ailus@linux.intel.com>
Signed-off-by: Hans Verkuil <hverkuil+cisco@kernel.org>
|
|
Smatch is confused by the control initialization logic. It assumes that
there can be a case where imx214->link_freq can be NULL and ctrls_init
returns 0.
Re-arrange the function to make smatch happy.
This patch fixes this smatch error:
drivers/media/i2c/imx214.c:1109 imx214_ctrls_init() error: we previously assumed 'imx214->link_freq' could be null (see line 1017)
Signed-off-by: Ricardo Ribalda <ribalda@chromium.org>
Signed-off-by: Sakari Ailus <sakari.ailus@linux.intel.com>
Signed-off-by: Hans Verkuil <hverkuil+cisco@kernel.org>
|
|
Call on v4l2_get_active_data_lanes() to check if the driver reports that
the number of lanes actively used by the MIPI CSI transmitter differs to
the maximum defined in device tree.
If the number of active data lanes reported by the driver is invalid,
catch and return the error. If the operation is not supported, fall back
to the number of allowed data lanes.
Signed-off-by: Isaac Scott <isaac.scott@ideasonboard.com>
Signed-off-by: Sakari Ailus <sakari.ailus@linux.intel.com>
Signed-off-by: Hans Verkuil <hverkuil+cisco@kernel.org>
|
|
Add the num_data_lanes field to the mipi_csis_device struct, and set it
equal to csis->bus.num_data_lanes. This is in preparation to support
cases when the data lanes actively used differs from the maximum
supported data lanes.
No functional changes intended by this commit.
Reviewed-by: Frank Li <Frank.Li@nxp.com>
Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Signed-off-by: Isaac Scott <isaac.scott@ideasonboard.com>
Signed-off-by: Sakari Ailus <sakari.ailus@linux.intel.com>
Signed-off-by: Hans Verkuil <hverkuil+cisco@kernel.org>
|
|
The number of data lanes is already printed as part of
mipi_csis_async_register(), making the first part of this print
redundant. Remove the redundant print, and move the debug print for
clock frequency to mipi_csis_parse_dt().
Reviewed-by: Frank Li <Frank.Li@nxp.com>
Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Signed-off-by: Isaac Scott <isaac.scott@ideasonboard.com>
Signed-off-by: Sakari Ailus <sakari.ailus@linux.intel.com>
Signed-off-by: Hans Verkuil <hverkuil+cisco@kernel.org>
|
|
Sometimes, users will not use all of the MIPI CSI 2 lanes available when
connecting to the MIPI CSI receiver of their device. Add a helper
function that checks the mbus_config for the device driver to allow
users to define the number of active data lanes through the
get_mbus_config op.
If the driver does not implement this op, fall back to using the maximum
number of lanes available.
Reviewed-by: Frank Li <Frank.Li@nxp.com>
Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Signed-off-by: Isaac Scott <isaac.scott@ideasonboard.com>
Signed-off-by: Sakari Ailus <sakari.ailus@linux.intel.com>
Signed-off-by: Hans Verkuil <hverkuil+cisco@kernel.org>
|
|
Switch from s_stream to enable_streams and disable_streams callbacks.
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Signed-off-by: Jai Luthra <jai.luthra@ideasonboard.com>
Signed-off-by: Sakari Ailus <sakari.ailus@linux.intel.com>
Signed-off-by: Hans Verkuil <hverkuil+cisco@kernel.org>
|
|
Introduce 2x2 binning mode (1312x972@60fps). Since there are multiple
modes now, use v4l2_find_nearest_size() to select the appropriate mode
in .set_fmt().
For 2x2 binning the minimum shutter value supported is 17 instead of 9.
This effects the maximum allowed exposure time, and if not enforced then
the sensor produces very dark frames when the minimum shutter limit is
violated.
Lastly, update the crop size reported to the userspace. When trying 2x2
binning with the datasheet suggested pixel array size (i.e. 2592 / 2 =>
1296), on some platforms (Raspberry Pi 5) artefacts are introduced on
the right edge of the output image. Instead, using a higher width of
1312 works fine on all platforms.
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Signed-off-by: Jai Luthra <jai.luthra@ideasonboard.com>
Signed-off-by: Sakari Ailus <sakari.ailus@linux.intel.com>
Signed-off-by: Hans Verkuil <hverkuil+cisco@kernel.org>
|
|
Port the driver to use the subdev active state. This simplifies locking,
and makes it easier to support different crop sizes for binned modes, by
storing the crop rectangle inside the subdev state.
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Signed-off-by: Jai Luthra <jai.luthra@ideasonboard.com>
Signed-off-by: Sakari Ailus <sakari.ailus@linux.intel.com>
Signed-off-by: Hans Verkuil <hverkuil+cisco@kernel.org>
|
|
Simplify .s_stream callback implementation by moving the runtime PM
calls to the leaf functions. This patch should not affect any
functionality.
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Signed-off-by: Jai Luthra <jai.luthra@ideasonboard.com>
Signed-off-by: Sakari Ailus <sakari.ailus@linux.intel.com>
Signed-off-by: Hans Verkuil <hverkuil+cisco@kernel.org>
|
|
While switching modes, updating to a different value of HBLANK isn't
sufficient, as this is a read-only control with a single allowed value,
and thus hblank_min == hblank_max == hblank of the default mode.
So to correctly update the user-facing value of the HBLANK parameter,
which is necessary for correct framerate calculation, update the whole
range when switching modes.
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Signed-off-by: Jai Luthra <jai.luthra@ideasonboard.com>
Signed-off-by: Sakari Ailus <sakari.ailus@linux.intel.com>
Signed-off-by: Hans Verkuil <hverkuil+cisco@kernel.org>
|
|
The sensor datasheet reports actual total number of pixels as 2696x2044.
This becomes important for supporting 2x2 binning modes that can go
beyond the current maximum pixel array width set here.
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Signed-off-by: Jai Luthra <jai.luthra@ideasonboard.com>
Signed-off-by: Sakari Ailus <sakari.ailus@linux.intel.com>
Signed-off-by: Hans Verkuil <hverkuil+cisco@kernel.org>
|
|
Support vertical flip by setting REG_VREVERSE.
Additional registers also needs to be set per mode, according
to the readout direction (normal/inverted) as mentioned in the
data sheet.
Since the register IMX335_REG_AREA3_ST_ADR_1 is based on the
flip (and is set via vflip related registers), it has been
moved out of the 2592x1944 mode regs.
Signed-off-by: Umang Jain <umang.jain@ideasonboard.com>
Reviewed-by: Tommaso Merciai <tomm.merciai@gmail.com>
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Signed-off-by: Jai Luthra <jai.luthra@ideasonboard.com>
Signed-off-by: Sakari Ailus <sakari.ailus@linux.intel.com>
Signed-off-by: Hans Verkuil <hverkuil+cisco@kernel.org>
|
|
In commit 81495a59baeb ("media: imx335: Fix active area height
discrepency") the height for the mode struct was rectified to '1944'.
However, the name of mode struct is still reflecting to '1940'. Update
it.
Signed-off-by: Umang Jain <umang.jain@ideasonboard.com>
Reviewed-by: Tommaso Merciai <tomm.merciai@gmail.com>
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Signed-off-by: Jai Luthra <jai.luthra@ideasonboard.com>
Signed-off-by: Sakari Ailus <sakari.ailus@linux.intel.com>
Signed-off-by: Hans Verkuil <hverkuil+cisco@kernel.org>
|
|
Add the OV05C10 sensor to the list of supported sensors.
Signed-off-by: Hao Yao <hao.yao@intel.com>
Signed-off-by: Hans de Goede <hansg@kernel.org>
Signed-off-by: Sakari Ailus <sakari.ailus@linux.intel.com>
Signed-off-by: Hans Verkuil <hverkuil+cisco@kernel.org>
|
|
Add the IMX471 sensor, which is used with a SONY471A ACPI HID on
the Lenovo ThinkPad X9 15 laptop to the list of supported sensors.
Link: https://bugzilla.redhat.com/show_bug.cgi?id=2362064
Signed-off-by: Jimmy Su <jimmy.su@intel.com>
Signed-off-by: Hans de Goede <hansg@kernel.org>
Signed-off-by: Sakari Ailus <sakari.ailus@linux.intel.com>
Signed-off-by: Hans Verkuil <hverkuil+cisco@kernel.org>
|
|
The `pwm_th1520` Rust driver calls C functions from the `PWM` namespace,
triggering `modpost` warnings due to missing namespace import
declarations in its `.modinfo` section.
Fix these warnings and simplify the module declaration by switching from
the generic `kernel::module_platform_driver!` macro to the newly
introduced PWM-specific `kernel::module_pwm_platform_driver!` macro.
The new macro automatically handles the required `imports_ns: ["PWM"]`
declaration.
Signed-off-by: Michal Wilczynski <m.wilczynski@samsung.com>
Reviewed-by: Troy Mitchell <troy.mitchell@linux.dev>
Reviewed-by: Elle Rhumsaa <elle@weathered-steel.dev>
Link: https://patch.msgid.link/20251028-pwm_fixes-v1-3-25a532d31998@samsung.com
Signed-off-by: Uwe Kleine-König <ukleinek@kernel.org>
|
|
Clippy warns about redundant struct field initialization when the field
name and the variable name are the same (e.g., `status: status`).
No functional change.
Signed-off-by: Michal Wilczynski <m.wilczynski@samsung.com>
Reviewed-by: Elle Rhumsaa <elle@weathered-steel.dev>
Link: https://patch.msgid.link/20251028-pwm_fixes-v1-4-25a532d31998@samsung.com
Signed-off-by: Uwe Kleine-König <ukleinek@kernel.org>
|
|
We do our best to keep the repository `rustfmt`-clean [1], thus run the
tool to fix the formatting issue.
A trailing empty comment [2] is added in order to preserve the wanted
style for imports (otherwise the tool will compact the first two items).
Link: https://rust-for-linux.com/contributing#submit-checklist-addendum [1]
Link: https://docs.kernel.org/rust/coding-guidelines.html#style-formatting [2]
Fixes: d8046cd50879 ("rust: pwm: Add complete abstraction layer")
Fixes: 7b3dce814a15 ("rust: pwm: Add Kconfig and basic data structures")
Fixes: e03724aac758 ("pwm: Add Rust driver for T-HEAD TH1520 SoC")
Signed-off-by: Miguel Ojeda <ojeda@kernel.org>
Link: https://patch.msgid.link/20251029182502.783392-1-ojeda@kernel.org
Signed-off-by: Uwe Kleine-König <ukleinek@kernel.org>
|
|
Introduce a PWM driver for the T-HEAD TH1520 SoC, written in Rust and
utilizing the safe PWM abstractions from the preceding commit.
The driver implements the pwm::PwmOps trait using the modern waveform
API (round_waveform_tohw, write_waveform, etc.) to support configuration
of period, duty cycle, and polarity for the TH1520's PWM channels.
Resource management is handled using idiomatic Rust patterns. The PWM
chip object is allocated via pwm::Chip::new and its registration with
the PWM core is managed by the pwm::Registration RAII guard. This
ensures pwmchip_remove is always called when the driver unbinds,
preventing resource leaks. Device managed resources are used for the
MMIO region, and the clock lifecycle is correctly managed in the
driver's private data Drop implementation.
The driver's core logic is written entirely in safe Rust, with no unsafe
blocks, except for the Send and Sync implementations for the driver
data, which are explained in the comments.
Reviewed-by: Elle Rhumsaa <elle@weathered-steel.dev>
Signed-off-by: Michal Wilczynski <m.wilczynski@samsung.com>
Link: https://patch.msgid.link/20251016-rust-next-pwm-working-fan-for-sending-v16-4-a5df2405d2bd@samsung.com
Signed-off-by: Uwe Kleine-König <ukleinek@kernel.org>
|
|
This has become unused and unnecessary. Remove.
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Link: https://patch.msgid.link/20251112185547.172113-1-jani.nikula@intel.com
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
|
|
Display code stopped using i915_utils.h in favour of
intel_display_utils.h. Fix recent additions.
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Link: https://patch.msgid.link/20251112181342.107911-1-jani.nikula@intel.com
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
|
|
with AutoFDO
When compiling the kernel with -ffunction-sections enabled, the split()
function gets compiled into the .text.split section. In some cases it
can even be cloned into .text.split.constprop.0 or .text.split.isra.0.
However, .text.split.* is already reserved for use by the Clang
-fsplit-machine-functions flag, which is used by AutoFDO. That may
place part of a function's code in a .text.split.<func> section.
This naming conflict causes the vmlinux linker script to wrongly place
split() with other .text.split.* code, rather than where it belongs with
regular text.
Fix it by renaming split() to split_strings().
Fixes: 6568f14cb5ae ("vmlinux.lds: Exclude .text.startup and .text.exit from TEXT_MAIN")
Signed-off-by: Josh Poimboeuf <jpoimboe@kernel.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: live-patching@vger.kernel.org
Cc: Juergen Gross <jgross@suse.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Link: https://patch.msgid.link/92a194234a0f757765e275b288bb1a7236c2c35c.1762991150.git.jpoimboe@kernel.org
|
|
-ffunction-sections
When compiling the kernel with -ffunction-sections (e.g., for LTO,
livepatch, dead code elimination, AutoFDO, or Propeller), the startup()
function gets compiled into the .text.startup section. In some cases it
can even be cloned into .text.startup.constprop.0 or
.text.startup.isra.0.
However, the .text.startup and .text.startup.* section names are already
reserved for use by the compiler for __attribute__((constructor)) code.
This naming conflict causes the vmlinux linker script to wrongly place
startup() function code in .init.text, which gets freed during boot.
Fix that by renaming startup() to ov2722_startup().
Fixes: 6568f14cb5ae ("vmlinux.lds: Exclude .text.startup and .text.exit from TEXT_MAIN")
Signed-off-by: Josh Poimboeuf <jpoimboe@kernel.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: live-patching@vger.kernel.org
Cc: Hans de Goede <hansg@kernel.org>
Cc: Mauro Carvalho Chehab <mchehab@kernel.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Link: https://patch.msgid.link/bf8cd823a3f11f64cc82167913be5013c72afa57.1762991150.git.jpoimboe@kernel.org
|
|
Signed-off-by: Ingo Molnar <mingo@kernel.org>
|
|
Marco Crivellari <marco.crivellari@suse.com> says:
Hi,
=== Current situation: problems ===
Let's consider a nohz_full system with isolated CPUs: wq_unbound_cpumask is
set to the housekeeping CPUs, for !WQ_UNBOUND the local CPU is selected.
This leads to different scenarios if a work item is scheduled on an
isolated CPU where "delay" value is 0 or greater then 0:
schedule_delayed_work(, 0);
This will be handled by __queue_work() that will queue the work item on the
current local (isolated) CPU, while:
schedule_delayed_work(, 1);
Will move the timer on an housekeeping CPU, and schedule the work there.
Currently if a user enqueue a work item using schedule_delayed_work() the
used wq is "system_wq" (per-cpu wq) while queue_delayed_work() use
WORK_CPU_UNBOUND (used when a cpu is not specified). The same applies to
schedule_work() that is using system_wq and queue_work(), that makes use
again of WORK_CPU_UNBOUND.
This lack of consistency cannot be addressed without refactoring the API.
=== Recent changes to the WQ API ===
The following, address the recent changes in the Workqueue API:
- commit 128ea9f6ccfb ("workqueue: Add system_percpu_wq and system_dfl_wq")
- commit 930c2ea566af ("workqueue: Add new WQ_PERCPU flag")
The old workqueues will be removed in a future release cycle.
=== Introduced Changes by this series ===
1) [P 1] Replace uses of system_wq and system_unbound_wq
system_unbound_wq is to be used when locality is not required.
Because of that, system_unbound_wq has been replaced with
system_dfl_wq, to make clear it should be used when locality
is not required.
2) [P 2-3-4] WQ_PERCPU added to alloc_workqueue()
This change adds a new WQ_PERCPU flag to explicitly request
alloc_workqueue() to be per-cpu when WQ_UNBOUND has not been specified.
Thanks!
Link: https://patch.msgid.link/20251031095643.74246-1-marco.crivellari@suse.com
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
|
|
Currently if a user enqueues a work item using schedule_delayed_work()
the used wq is "system_wq" (per-cpu wq) while queue_delayed_work() use
WORK_CPU_UNBOUND (used when a cpu is not specified). The same applies to
schedule_work() that is using system_wq and queue_work(), that makes use
again of WORK_CPU_UNBOUND. This lack of consistency cannot be addressed
without refactoring the API.
alloc_workqueue() treats all queues as per-CPU by default, while unbound
workqueues must opt-in via WQ_UNBOUND.
This default is suboptimal: most workloads benefit from unbound queues,
allowing the scheduler to place worker threads where they’re needed and
reducing noise when CPUs are isolated.
This continues the effort to refactor workqueue APIs, which began with
the introduction of new workqueues and a new alloc_workqueue flag in:
commit 128ea9f6ccfb ("workqueue: Add system_percpu_wq and system_dfl_wq")
commit 930c2ea566af ("workqueue: Add new WQ_PERCPU flag")
This adds a new WQ_PERCPU flag to explicitly request to alloc_workqueue()
to be per-cpu when WQ_UNBOUND has not been specified.
With the introduction of the WQ_PERCPU flag (equivalent to !WQ_UNBOUND),
any alloc_workqueue() caller that doesn’t explicitly specify WQ_UNBOUND
must now use WQ_PERCPU.
Once migration is complete, WQ_UNBOUND can be removed and unbound will
become the implicit default.
Suggested-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Marco Crivellari <marco.crivellari@suse.com>
Link: https://patch.msgid.link/20251107155257.316728-1-marco.crivellari@suse.com
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
|
|
Currently if a user enqueues a work item using schedule_delayed_work()
the used wq is "system_wq" (per-cpu wq) while queue_delayed_work() use
WORK_CPU_UNBOUND (used when a cpu is not specified). The same applies to
schedule_work() that is using system_wq and queue_work(), that makes use
again of WORK_CPU_UNBOUND. This lack of consistency cannot be addressed
without refactoring the API.
alloc_workqueue() treats all queues as per-CPU by default, while unbound
workqueues must opt-in via WQ_UNBOUND.
This default is suboptimal: most workloads benefit from unbound queues,
allowing the scheduler to place worker threads where they’re needed and
reducing noise when CPUs are isolated.
This continues the effort to refactor workqueue APIs, which began with
the introduction of new workqueues and a new alloc_workqueue flag in:
commit 128ea9f6ccfb ("workqueue: Add system_percpu_wq and system_dfl_wq")
commit 930c2ea566af ("workqueue: Add new WQ_PERCPU flag")
This change adds a new WQ_PERCPU flag to explicitly request
alloc_workqueue() to be per-cpu when WQ_UNBOUND has not been specified.
With the introduction of the WQ_PERCPU flag (equivalent to !WQ_UNBOUND),
any alloc_workqueue() caller that doesn’t explicitly specify WQ_UNBOUND
must now use WQ_PERCPU.
Once migration is complete, WQ_UNBOUND can be removed and unbound will
become the implicit default.
Suggested-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Marco Crivellari <marco.crivellari@suse.com>
Link: https://patch.msgid.link/20251107154008.304127-1-marco.crivellari@suse.com
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
|
|
Currently if a user enqueues a work item using schedule_delayed_work()
the used wq is "system_wq" (per-cpu wq) while queue_delayed_work() use
WORK_CPU_UNBOUND (used when a cpu is not specified). The same applies to
schedule_work() that is using system_wq and queue_work(), that makes use
again of WORK_CPU_UNBOUND. This lack of consistency cannot be addressed
without refactoring the API.
alloc_workqueue() treats all queues as per-CPU by default, while unbound
workqueues must opt-in via WQ_UNBOUND.
This default is suboptimal: most workloads benefit from unbound queues,
allowing the scheduler to place worker threads where they’re needed and
reducing noise when CPUs are isolated.
This continues the effort to refactor workqueue APIs, which began with
the introduction of new workqueues and a new alloc_workqueue flag in:
commit 128ea9f6ccfb ("workqueue: Add system_percpu_wq and system_dfl_wq")
commit 930c2ea566af ("workqueue: Add new WQ_PERCPU flag")
This change adds a new WQ_PERCPU flag to explicitly request
alloc_workqueue() to be per-cpu when WQ_UNBOUND has not been specified.
With the introduction of the WQ_PERCPU flag (equivalent to !WQ_UNBOUND),
any alloc_workqueue() caller that doesn’t explicitly specify WQ_UNBOUND
must now use WQ_PERCPU.
Once migration is complete, WQ_UNBOUND can be removed and unbound will
become the implicit default.
Suggested-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Marco Crivellari <marco.crivellari@suse.com>
Link: https://patch.msgid.link/20251107151618.281250-1-marco.crivellari@suse.com
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
|
|
Currently if a user enqueues a work item using schedule_delayed_work()
the used wq is "system_wq" (per-cpu wq) while queue_delayed_work() use
WORK_CPU_UNBOUND (used when a cpu is not specified). The same applies to
schedule_work() that is using system_wq and queue_work(), that makes use
again of WORK_CPU_UNBOUND. This lack of consistency cannot be addressed
without refactoring the API.
alloc_workqueue() treats all queues as per-CPU by default, while unbound
workqueues must opt-in via WQ_UNBOUND.
This default is suboptimal: most workloads benefit from unbound queues,
allowing the scheduler to place worker threads where they’re needed and
reducing noise when CPUs are isolated.
This continues the effort to refactor workqueue APIs, which began with
the introduction of new workqueues and a new alloc_workqueue flag in:
commit 128ea9f6ccfb ("workqueue: Add system_percpu_wq and system_dfl_wq")
commit 930c2ea566af ("workqueue: Add new WQ_PERCPU flag")
This change adds a new WQ_PERCPU flag to explicitly request
alloc_workqueue() to be per-cpu when WQ_UNBOUND has not been specified.
With the introduction of the WQ_PERCPU flag (equivalent to !WQ_UNBOUND),
any alloc_workqueue() caller that doesn’t explicitly specify WQ_UNBOUND
must now use WQ_PERCPU.
Once migration is complete, WQ_UNBOUND can be removed and unbound will
become the implicit default.
Suggested-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Marco Crivellari <marco.crivellari@suse.com>
Link: https://patch.msgid.link/20251107150542.271229-1-marco.crivellari@suse.com
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
|
|
Currently if a user enqueues a work item using schedule_delayed_work()
the used wq is "system_wq" (per-cpu wq) while queue_delayed_work() use
WORK_CPU_UNBOUND (used when a cpu is not specified). The same applies to
schedule_work() that is using system_wq and queue_work(), that makes use
again of WORK_CPU_UNBOUND. This lack of consistency cannot be addressed
without refactoring the API.
alloc_workqueue() treats all queues as per-CPU by default, while unbound
workqueues must opt-in via WQ_UNBOUND.
This default is suboptimal: most workloads benefit from unbound queues,
allowing the scheduler to place worker threads where they’re needed and
reducing noise when CPUs are isolated.
This continues the effort to refactor workqueue APIs, which began with
the introduction of new workqueues and a new alloc_workqueue flag in:
commit 128ea9f6ccfb ("workqueue: Add system_percpu_wq and system_dfl_wq")
commit 930c2ea566af ("workqueue: Add new WQ_PERCPU flag")
This change adds a new WQ_PERCPU flag to explicitly request
alloc_workqueue() to be per-cpu when WQ_UNBOUND has not been specified.
With the introduction of the WQ_PERCPU flag (equivalent to !WQ_UNBOUND),
any alloc_workqueue() caller that doesn’t explicitly specify WQ_UNBOUND
must now use WQ_PERCPU.
Once migration is complete, WQ_UNBOUND can be removed and unbound will
become the implicit default.
Suggested-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Marco Crivellari <marco.crivellari@suse.com>
Link: https://patch.msgid.link/20251107150155.267651-3-marco.crivellari@suse.com
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
|
|
Currently if a user enqueues a work item using schedule_delayed_work()
the used wq is "system_wq" (per-cpu wq) while queue_delayed_work() use
WORK_CPU_UNBOUND (used when a cpu is not specified). The same applies to
schedule_work() that is using system_wq and queue_work(), that makes use
again of WORK_CPU_UNBOUND. This lack of consistency cannot be addressed
without refactoring the API.
alloc_workqueue() treats all queues as per-CPU by default, while unbound
workqueues must opt-in via WQ_UNBOUND.
This default is suboptimal: most workloads benefit from unbound queues,
allowing the scheduler to place worker threads where they’re needed and
reducing noise when CPUs are isolated.
This continues the effort to refactor workqueue APIs, which began with
the introduction of new workqueues and a new alloc_workqueue flag in:
commit 128ea9f6ccfb ("workqueue: Add system_percpu_wq and system_dfl_wq")
commit 930c2ea566af ("workqueue: Add new WQ_PERCPU flag")
This change adds a new WQ_PERCPU flag to explicitly request
alloc_workqueue() to be per-cpu when WQ_UNBOUND has not been specified.
With the introduction of the WQ_PERCPU flag (equivalent to !WQ_UNBOUND),
any alloc_workqueue() caller that doesn’t explicitly specify WQ_UNBOUND
must now use WQ_PERCPU.
Once migration is complete, WQ_UNBOUND can be removed and unbound will
become the implicit default.
Suggested-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Marco Crivellari <marco.crivellari@suse.com>
Link: https://patch.msgid.link/20251107150155.267651-2-marco.crivellari@suse.com
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
|
|
Currently if a user enqueues a work item using schedule_delayed_work()
the used wq is "system_wq" (per-cpu wq) while queue_delayed_work() use
WORK_CPU_UNBOUND (used when a cpu is not specified). The same applies to
schedule_work() that is using system_wq and queue_work(), that makes use
again of WORK_CPU_UNBOUND. This lack of consistency cannot be addressed
without refactoring the API.
alloc_workqueue() treats all queues as per-CPU by default, while unbound
workqueues must opt-in via WQ_UNBOUND.
This default is suboptimal: most workloads benefit from unbound queues,
allowing the scheduler to place worker threads where they’re needed and
reducing noise when CPUs are isolated.
This continues the effort to refactor workqueue APIs, which began with
the introduction of new workqueues and a new alloc_workqueue flag in:
commit 128ea9f6ccfb ("workqueue: Add system_percpu_wq and system_dfl_wq")
commit 930c2ea566af ("workqueue: Add new WQ_PERCPU flag")
This change adds a new WQ_PERCPU flag to explicitly request
alloc_workqueue() to be per-cpu when WQ_UNBOUND has not been specified.
With the introduction of the WQ_PERCPU flag (equivalent to !WQ_UNBOUND),
any alloc_workqueue() caller that doesn’t explicitly specify WQ_UNBOUND
must now use WQ_PERCPU.
Once migration is complete, WQ_UNBOUND can be removed and unbound will
become the implicit default.
Suggested-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Marco Crivellari <marco.crivellari@suse.com>
Link: https://patch.msgid.link/20251107144949.256894-1-marco.crivellari@suse.com
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
|
|
Currently if a user enqueues a work item using schedule_delayed_work()
the used wq is "system_wq" (per-cpu wq) while queue_delayed_work() use
WORK_CPU_UNBOUND (used when a cpu is not specified). The same applies to
schedule_work() that is using system_wq and queue_work(), that makes use
again of WORK_CPU_UNBOUND. This lack of consistency cannot be addressed
without refactoring the API.
alloc_workqueue() treats all queues as per-CPU by default, while unbound
workqueues must opt-in via WQ_UNBOUND.
This default is suboptimal: most workloads benefit from unbound queues,
allowing the scheduler to place worker threads where they’re needed and
reducing noise when CPUs are isolated.
This continues the effort to refactor workqueue APIs, which began with
the introduction of new workqueues and a new alloc_workqueue flag in:
commit 128ea9f6ccfb ("workqueue: Add system_percpu_wq and system_dfl_wq")
commit 930c2ea566af ("workqueue: Add new WQ_PERCPU flag")
This change adds a new WQ_PERCPU flag to explicitly request
alloc_workqueue() to be per-cpu when WQ_UNBOUND has not been specified.
With the introduction of the WQ_PERCPU flag (equivalent to !WQ_UNBOUND),
any alloc_workqueue() caller that doesn’t explicitly specify WQ_UNBOUND
must now use WQ_PERCPU.
Once migration is complete, WQ_UNBOUND can be removed and unbound will
become the implicit default.
Suggested-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Marco Crivellari <marco.crivellari@suse.com>
Link: https://patch.msgid.link/20251107141458.225119-1-marco.crivellari@suse.com
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
|
