[PATCH 0/2] docs: Remove the unsupported preempt= modes

[PATCH 0/2] docs: Remove the unsupported preempt= modes

[Jinjie Ruan]

The none and voluntary preemption models can no longer be selected with
CONFIG_PREEMPT_DYNAMIC:

- commit 7dadeaa6e851 ("sched: Further restrict the preemption modes")
  stopped PREEMPT_NONE and PREEMPT_VOLUNTARY from being offered on
  architectures that select ARCH_HAS_PREEMPT_LAZY;

- commit d3d16750693b ("sched: dynamic: Make PREEMPT_DYNAMIC depend on
  ARCH_HAS_PREEMPT_LAZY") made that dependency official and removed the
  code that parsed "preempt=none" and "preempt=voluntary".

Two documentation files still describe both as runtime options. The first
patch drops them from the preempt= entry in kernel-parameters.txt, the
second fixes the model table in sched-preemption.rst, which also misstates
why an architecture may or may not offer them.

Nothing in the code changes.

Jinjie Ruan (2):
  docs: kernel-parameters: Drop unsupported preempt= modes
  docs: sched-preemption: Fix the list of runtime-selectable models

 Documentation/admin-guide/kernel-parameters.txt |  2 --
 Documentation/scheduler/sched-preemption.rst    | 13 ++++++-------
 2 files changed, 6 insertions(+), 9 deletions(-)

-- 
2.34.1

[PATCH 1/2] docs: kernel-parameters: Drop unsupported preempt= modes

[Jinjie Ruan]

Since commit d3d16750693b ("sched: dynamic: Make PREEMPT_DYNAMIC depend
on ARCH_HAS_PREEMPT_LAZY") only "full" and "lazy" are accepted with
CONFIG_PREEMPT_DYNAMIC: every architecture that supports it selects
ARCH_HAS_PREEMPT_LAZY, which excludes the none and voluntary models.
Passing either is rejected with "Dynamic Preempt: unsupported mode".

Remove the two lines so the parameter reference matches the code.

Signed-off-by: Jinjie Ruan <ruanjinjie@huawei.com>
---
 Documentation/admin-guide/kernel-parameters.txt | 2 --
 1 file changed, 2 deletions(-)

diff --git a/Documentation/admin-guide/kernel-parameters.txt b/Documentation/admin-guide/kernel-parameters.txt
index fce0d52b9cfe..11f0b700d56a 100644
--- a/Documentation/admin-guide/kernel-parameters.txt
+++ b/Documentation/admin-guide/kernel-parameters.txt
@@ -5428,8 +5428,6 @@ Kernel parameters
 
 	preempt=	[KNL]
 			Select preemption mode if you have CONFIG_PREEMPT_DYNAMIC
-			none - Limited to cond_resched() calls
-			voluntary - Limited to cond_resched() and might_sleep() calls
 			full - Any section that isn't explicitly preempt disabled
 			       can be preempted anytime.  Tasks will also yield
 			       contended spinlocks (if the critical section isn't
-- 
2.34.1

[PATCH 2/2] docs: sched-preemption: Fix the list of runtime-selectable models

[Jinjie Ruan]

CONFIG_PREEMPT_DYNAMIC now depends on CONFIG_ARCH_HAS_PREEMPT_LAZY, so the
none and voluntary models can no longer be selected at runtime; passing
either to preempt= is rejected with "Dynamic Preempt: unsupported mode".

Drop them from the table of runtime-selectable models, and describe the
build-time situation instead of implying the choice simply depends on the
architecture offering lazy preemption.  Keep the per-model descriptions
below, which still explain how a scheduling request is honoured under
models used by other kernels.

Signed-off-by: Jinjie Ruan <ruanjinjie@huawei.com>
---
 Documentation/scheduler/sched-preemption.rst | 13 ++++++-------
 1 file changed, 6 insertions(+), 7 deletions(-)

diff --git a/Documentation/scheduler/sched-preemption.rst b/Documentation/scheduler/sched-preemption.rst
index b48e53587a7a..0b5ddf7d26f2 100644
--- a/Documentation/scheduler/sched-preemption.rst
+++ b/Documentation/scheduler/sched-preemption.rst
@@ -20,9 +20,6 @@ change.
 The models selectable at runtime are:
 
 	=========	====================================================
-	none		No preemption of kernel code other than at explicit
-			``cond_resched()`` / blocking points.
-	voluntary	As ``none``, plus ``might_sleep()`` sites.
 	full		Any section that is not explicitly preempt disabled
 			may be preempted at any time.
 	lazy		As ``full``, except that the scheduling request is
@@ -31,10 +28,12 @@ The models selectable at runtime are:
 			apply to real-time tasks.
 	=========	====================================================
 
-Not every model is available on every kernel.  An architecture that provides
-lazy-preempt support offers only ``full`` and ``lazy``, and ``none`` and
-``voluntary`` are rejected there.  A PREEMPT_RT kernel always offers ``full``,
-and ``lazy`` as well if the architecture provides it.
+CONFIG_PREEMPT_DYNAMIC depends on CONFIG_ARCH_HAS_PREEMPT_LAZY, so ``none``
+and ``voluntary`` are never selectable at runtime; passing either is
+rejected with "Dynamic Preempt: unsupported mode".  The two models are only
+left as build-time choices on architectures without lazy-preempt support.
+A PREEMPT_RT kernel always offers ``full``, and ``lazy`` as well if the
+architecture provides it.
 
 The model that is actually active is reported in the boot log::
 
-- 
2.34.1

Re: [PATCH 1/2] docs: kernel-parameters: Drop unsupported preempt= modes

[Daniel Pereira]

Em ter., 27 de out. de 2026 às 23:45, Jinjie Ruan
<ruanjinjie@huawei.com> escreveu:
>
> Since commit d3d16750693b ("sched: dynamic: Make PREEMPT_DYNAMIC depend
> on ARCH_HAS_PREEMPT_LAZY") only "full" and "lazy" are accepted with
> CONFIG_PREEMPT_DYNAMIC: every architecture that supports it selects
> ARCH_HAS_PREEMPT_LAZY, which excludes the none and voluntary models.
> Passing either is rejected with "Dynamic Preempt: unsupported mode".
>
Hi Jinjie,

Thanks for the patch. I checked it against current mainline:
PREEMPT_DYNAMIC now depends on ARCH_HAS_PREEMPT_LAZY (d3d16750693b),
sched_dynamic_mode() only accepts "full" and "lazy", and all six
architectures that select HAVE_PREEMPT_DYNAMIC also select
ARCH_HAS_PREEMPT_LAZY. So dropping "none" and "voluntary" from the
preempt= entry matches the code.

checkpatch is clean, and the admin-guide build (SPHINXDIRS=admin-guide)
renders the preempt= entry correctly with only "full" and "lazy".

Reviewed-by: Daniel Pereira <danielmaraboo@gmail.com>

Best regards,
Daniel

Re: [PATCH 2/2] docs: sched-preemption: Fix the list of runtime-selectable models

[Daniel Pereira]

On Tue, Oct 27, 2026 11:44 PM, Jinjie Ruan <ruanjinjie@huawei.com> wrote:
>
> CONFIG_PREEMPT_DYNAMIC now depends on CONFIG_ARCH_HAS_PREEMPT_LAZY, so the
> none and voluntary models can no longer be selected at runtime; passing
> either to preempt= is rejected with "Dynamic Preempt: unsupported mode".
>
> Drop them from the table of runtime-selectable models, and describe the
> build-time situation instead of implying the choice simply depends on the
> architecture offering lazy preemption.  Keep the per-model descriptions
> below, which still explain how a scheduling request is honoured under
> models used by other kernels.
>
> Signed-off-by: Jinjie Ruan <ruanjinjie@huawei.com>
> ---
>  Documentation/scheduler/sched-preemption.rst | 13 ++++++-------
>  1 file changed, 6 insertions(+), 7 deletions(-)

Hi Jinjie,

Thanks for the patch. I validated it: checkpatch and the scheduler docs
build (SPHINXDIRS=scheduler) report no errors or new warnings, the table
renders as expected, and the changes match current mainline
(PREEMPT_DYNAMIC depends on ARCH_HAS_PREEMPT_LAZY, d3d16750693b).

Reviewed-by: Daniel Pereira <danielmaraboo@gmail.com>

Best regards,
Daniel

Re: [PATCH 2/2] docs: sched-preemption: Fix the list of runtime-selectable models

[Sebastian Andrzej Siewior]

On 2026-10-28 10:43:33 [+0800], Jinjie Ruan wrote:
> --- a/Documentation/scheduler/sched-preemption.rst
> +++ b/Documentation/scheduler/sched-preemption.rst
> @@ -20,9 +20,6 @@ change.
>  The models selectable at runtime are:
>  
>  	=========	====================================================
> -	none		No preemption of kernel code other than at explicit
> -			``cond_resched()`` / blocking points.
> -	voluntary	As ``none``, plus ``might_sleep()`` sites.
>  	full		Any section that is not explicitly preempt disabled
>  			may be preempted at any time.
>  	lazy		As ``full``, except that the scheduling request is

This is fine.

> @@ -31,10 +28,12 @@ The models selectable at runtime are:
>  			apply to real-time tasks.
>  	=========	====================================================
>  
> -Not every model is available on every kernel.  An architecture that provides
> -lazy-preempt support offers only ``full`` and ``lazy``, and ``none`` and
> -``voluntary`` are rejected there.  A PREEMPT_RT kernel always offers ``full``,
> -and ``lazy`` as well if the architecture provides it.
> +CONFIG_PREEMPT_DYNAMIC depends on CONFIG_ARCH_HAS_PREEMPT_LAZY, so ``none``
> +and ``voluntary`` are never selectable at runtime; passing either is
> +rejected with "Dynamic Preempt: unsupported mode".  The two models are only
> +left as build-time choices on architectures without lazy-preempt support.
> +A PREEMPT_RT kernel always offers ``full``, and ``lazy`` as well if the
> +architecture provides it.

What about:
 Not every model is available on every kernel.  An architecture that provides
 preemption support supports ``full``. If additionally lazy-preempt is
 supported then the choice ``lazy`` becomes available. An architecture
 without preemption support can not enable CONFIG_PREEMPT_DYNAMIC. On
 such an architecture the preemption model can only be selected at
 compile time. 

Sebastian

Re: [PATCH 1/2] docs: kernel-parameters: Drop unsupported preempt= modes

[Sebastian Andrzej Siewior]

On 2026-10-28 10:43:32 [+0800], Jinjie Ruan wrote:
> Since commit d3d16750693b ("sched: dynamic: Make PREEMPT_DYNAMIC depend
> on ARCH_HAS_PREEMPT_LAZY") only "full" and "lazy" are accepted with
> CONFIG_PREEMPT_DYNAMIC: every architecture that supports it selects
> ARCH_HAS_PREEMPT_LAZY, which excludes the none and voluntary models.
> Passing either is rejected with "Dynamic Preempt: unsupported mode".
> 
> Remove the two lines so the parameter reference matches the code.
> 
> Signed-off-by: Jinjie Ruan <ruanjinjie@huawei.com>

Reviewed-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>

Sebastian