Re: [PATCH] sched: Tweak default dynamic preempt mode selection

From: Valentin Schneider
Date: Tue Nov 09 2021 - 04:53:17 EST


On 09/11/21 06:30, Mike Galbraith wrote:
> Not seeing your v2 land yet, I grabbed my mallet and had a go at goal
> reconciliation over morning java. Non-lovely result seems to work.
>

Yeah so I went down a debatable path, gave up on that and started something
different, and gave up on that because it was late :-)

Now interestingly my second attempt is pretty close to what you have
below.

> sched, Kconfig: Fix preemption model selection
>
> Switch PREEMPT_DYNAMIC/PREEMPT_RT dependency around so PREEMPT_RT
> can be selected during the initial preemption model selection.
> Further, since PREEMPT_DYNAMIC requires PREEMPT, make it depend
> upon it instead of selecting it, and add a menu to allow selection
> of the boot time behavior, this to allow arches that do not support
> PREEMPT_DYNAMIC to retain their various configs untouched.
>

Have some nits below, but otherwise where I stand right now I think it's
the least ugly way of tackling this :)

Reviewed-by: Valentin Schneider <valentin.schneider@xxxxxxx>

> Signed-off-by: Mike Galbraith <efault@xxxxxx>
> ---
> kernel/Kconfig.preempt | 46 +++++++++++++++++++++++++---------------
> ------
> 1 file changed, 25 insertions(+), 21 deletions(-)
>
> --- a/kernel/Kconfig.preempt
> +++ b/kernel/Kconfig.preempt
> @@ -2,11 +2,10 @@
>
> choice
> prompt "Preemption Model"
> - default PREEMPT_NONE_BEHAVIOUR
> + default PREEMPT_NONE
>
> -config PREEMPT_NONE_BEHAVIOUR
> +config PREEMPT_NONE
> bool "No Forced Preemption (Server)"
> - select PREEMPT_NONE if !PREEMPT_DYNAMIC
> help
> This is the traditional Linux preemption model, geared
> towards
> throughput. It will still provide good latencies most of the
> @@ -18,10 +17,9 @@ config PREEMPT_NONE_BEHAVIOUR
> raw processing power of the kernel, irrespective of
> scheduling
> latencies.
>
> -config PREEMPT_VOLUNTARY_BEHAVIOUR
> +config PREEMPT_VOLUNTARY
> bool "Voluntary Kernel Preemption (Desktop)"
> depends on !ARCH_NO_PREEMPT
> - select PREEMPT_VOLUNTARY if !PREEMPT_DYNAMIC
> help
> This option reduces the latency of the kernel by adding more
> "explicit preemption points" to the kernel code. These new
> @@ -37,10 +35,11 @@ config PREEMPT_VOLUNTARY_BEHAVIOUR
>
> Select this if you are building a kernel for a desktop
> system.
>
> -config PREEMPT_BEHAVIOUR
> +config PREEMPT
> bool "Preemptible Kernel (Low-Latency Desktop)"
> depends on !ARCH_NO_PREEMPT
> - select PREEMPT
> + select PREEMPTION
> + select UNINLINE_SPIN_UNLOCK if !ARCH_INLINE_SPIN_UNLOCK
> help
> This option reduces the latency of the kernel by making
> all kernel code (that is not executing in a critical
> section)
> @@ -58,7 +57,7 @@ config PREEMPT_BEHAVIOUR
>
> config PREEMPT_RT
> bool "Fully Preemptible Kernel (Real-Time)"
> - depends on EXPERT && ARCH_SUPPORTS_RT && !PREEMPT_DYNAMIC
> + depends on EXPERT && ARCH_SUPPORTS_RT
> select PREEMPTION
> help
> This option turns the kernel into a real-time kernel by
> replacing

This we can't get away from IMO - it's pretty much the same issue as

b8d3349803ba ("sched/rt, Kconfig: Unbreak def/oldconfig with CONFIG_PREEMPT=y")

IOW we can't really touch those preempt configs, but rather add stuff
around them.

> @@ -75,17 +74,6 @@ config PREEMPT_RT
>
> endchoice
>
> -config PREEMPT_NONE
> - bool
> -
> -config PREEMPT_VOLUNTARY
> - bool
> -
> -config PREEMPT
> - bool
> - select PREEMPTION
> - select UNINLINE_SPIN_UNLOCK if !ARCH_INLINE_SPIN_UNLOCK
> -
> config PREEMPT_COUNT
> bool
>
> @@ -95,8 +83,7 @@ config PREEMPTION
>
> config PREEMPT_DYNAMIC
> bool "Preemption behaviour defined on boot"
> - depends on HAVE_PREEMPT_DYNAMIC
> - select PREEMPT
> + depends on HAVE_PREEMPT_DYNAMIC && PREEMPT

I got there too. If we look at x86 defconfig, it selects
PREEMPT_VOLUNTARY. I initially tried making it so PREEMPT_VOLUNTARY means
both:

- VOLUNTARY model if !PREEMPT_DYNAMIC
- VOLUNTARY model as default boot-time mode if PREEMPT_DYNAMIC
(& similar for other preemption modes)

which, now that I've played with it, is a sad wreck. Having PREEMPT_DYNAMIC
depend on PREEMPT at least makes things clear, and defconfig choices remain
respected (they just won't get PREEMPT_DYNAMIC if they didn't pick
PREEMPT).

> default y
> help
> This option allows to define the preemption model on the
> kernel
> @@ -114,6 +101,23 @@ config PREEMPT_DYNAMIC
> Interesting if you want the same pre-built kernel should be
> used for
> both Server and Desktop workloads.
>
> +if PREEMPT_DYNAMIC

AFAICT that should be

depends on PREEMPT_DYNAMIC

within the choice itself. See e.g. "Kernel compression mode" choice in
init/Kconfig.

> +choice
> + prompt "Boot Time Preemption Model"

Default boot-time [...]? This is only used if the cmdline parameter is
missing. Also a help-text wouldn't hurt. I had:

This option defines the default preemption model of the kernel
if it hasn't been specified by the command line parameter.

> + default PREEMPT_NONE_BEHAVIOR
> +
> +config PREEMPT_NONE_BEHAVIOR
> + bool "No Forced Preemption (Server)"
> +
> +config PREEMPT_VOLUNTARY_BEHAVIOR
> + bool "Voluntary Kernel Preemption (Desktop)"
> +
> +config PREEMPT_BEHAVIOR
> + bool "Preemptible Kernel (Low-Latency Desktop)"
> +
> +endchoice
> +endif # PREEMPT_DYNAMIC
> +
> config SCHED_CORE
> bool "Core Scheduling for SMT"
> depends on SCHED_SMT