testing/packages/linux-5.14.x/kernel-source-5.14.14_smp-noarch-2.txz
These kernels enable CONFIG_PREEMPT=y and CONFIG_PREEMPT_DYNAMIC=y allowing
the kernel preemption model to be specified on the kernel command line
with one of these options: preempt=none, preempt=voluntary, and preempt=full.
Since there is no .config option to set a default, and the default in the
kernel sources is “full” (which is probably not a good default), the
kernel-source.SlackBuild has been modified to add support for an environment
variable CONFIG_PREEMPT_DEFAULT_MODE which can be set to none, voluntary, or
full to set the default kernel preemption model when a command line option
is not provided. These kernels have been built with a preemption model of
“none” (presumably the safest choice which will behave like the kernels we
have shipped before.) The runtime overhead on 64-bit should be negligible.
On 32-bit we lack support for HAVE_STATIC_CALL_INLINE, so spinlocks and
mutexes will have to be approached through a trampoline, adding a very small
amount of overhead. I feel this is probably worth it in order to have the
option to run a kernel with voluntary or full preemption, especially for
gaming or desktop purposes. The reduction in input lag with these modes is
actually quite noticable.
To check the current preemption model, you may use debugfs:
mount -t debugfs none /sys/kernel/debug
cat /sys/kernel/debug/sched/preempt
(none) voluntary full
You may change to a different preemption model on the fly once debugfs is
mounted:
echo voluntary > /sys/kernel/debug/sched/preempt
cat /sys/kernel/debug/sched/preempt
none (voluntary) full
Thanks to Daedra.
-DRM_I810 n
-INLINE_READ_UNLOCK y
-INLINE_READ_UNLOCK_IRQ y
-INLINE_SPIN_UNLOCK_IRQ y
-INLINE_WRITE_UNLOCK y
-INLINE_WRITE_UNLOCK_IRQ y
PREEMPT n → y
PREEMPT_VOLUNTARY y → n
+CEC_GPIO n
+DEBUG_PREEMPT y
+PREEMPTION y
+PREEMPT_COUNT y
+PREEMPT_DYNAMIC y
+PREEMPT_RCU y
+PREEMPT_TRACER n
+RCU_BOOST n
+TASKS_RCU y
+UNINLINE_SPIN_UNLOCK y
Giuseppe Di Terlizzi