[RFC PATCH 0/2] sched/fair: Remove tunable scaling

[RFC PATCH 0/2] sched/fair: Remove tunable scaling

[Tor Vic]

This series removes the scaling of tunables, which was introduced around
Linux 2.6.33.

I split this into two separate patches:

The first removes the "none" and "linear" scaling which are likely not used much
and have limited usefulness, as one can always set sched_base_slice through
debugfs.
It keeps the logarithmic scaling which is already the default.

The second removes the entire scaling mecanism and sets a fixed sched_base_slice
of 2 milliseconds, a value which has been chosen somewhat arbitrarily.

Currently, the base_slice multiplied by the log2 factor can take the values of
0.7 (1 CPU), 1.4 (2-3 CPUs), 2.1 (4-7 CPUs) and 2.8 msecs (8+ CPUs), so it seems
that the new value of 2 msecs is a sane default, which I also use on my consumer
x86_64 machines.
However I'll leave that open for discussion. Note that the base_slice value has
changed quite often over the years, lastly in 2025 [1].

The series applies to 7.3-rc5, and is build- and boot tested on a measly 2C/4T
Skylake machine. I marked it as RFC because I don't know whether these changes
are actually wanted by the sched people - and because I'm not a developer.
It does however remove quite a lot of code.

[1] commit 2ae891b826958b60919ea21c727f77bcd6ffcc2c

Tor Vic (2):
  sched/fair: Remove the 'none' and 'linear' scaling of tunables
  sched/fair: Remove sched_base_slice scaling altogether

 include/linux/sched/sysctl.h |  7 ----
 kernel/sched/core.c          |  1 -
 kernel/sched/debug.c         | 52 ------------------------
 kernel/sched/fair.c          | 79 ++----------------------------------
 kernel/sched/sched.h         |  5 ---
 5 files changed, 3 insertions(+), 141 deletions(-)

-- 
2.55.0

[RFC PATCH 1/2] sched/fair: Remove the 'none' and 'linear' scaling of tunables

[Tor Vic]

Get rid of the tunable scalings 'none' and 'linear' as these do not
appear to be very useful.

Only keep the 'log' scaling, which is already the default.

Signed-off-by: Tor Vic <torvic9@mailbox.org>
---
 include/linux/sched/sysctl.h |  7 -----
 kernel/sched/debug.c         | 52 ------------------------------------
 kernel/sched/fair.c          | 28 +------------------
 kernel/sched/sched.h         |  2 --
 4 files changed, 1 insertion(+), 88 deletions(-)

diff --git a/include/linux/sched/sysctl.h b/include/linux/sched/sysctl.h
index 5a64582b086b..0b923cc771e0 100644
--- a/include/linux/sched/sysctl.h
+++ b/include/linux/sched/sysctl.h
@@ -12,13 +12,6 @@ extern unsigned long sysctl_hung_task_timeout_secs;
 enum { sysctl_hung_task_timeout_secs = 0 };
 #endif
 
-enum sched_tunable_scaling {
-	SCHED_TUNABLESCALING_NONE,
-	SCHED_TUNABLESCALING_LOG,
-	SCHED_TUNABLESCALING_LINEAR,
-	SCHED_TUNABLESCALING_END,
-};
-
 #define NUMA_BALANCING_DISABLED		0x0
 #define NUMA_BALANCING_NORMAL		0x1
 #define NUMA_BALANCING_MEMORY_TIERING	0x2
diff --git a/kernel/sched/debug.c b/kernel/sched/debug.c
index 72236db67983..b159d7a1cfa2 100644
--- a/kernel/sched/debug.c
+++ b/kernel/sched/debug.c
@@ -170,46 +170,6 @@ static const struct file_operations sched_feat_fops = {
 	.release	= single_release,
 };
 
-static ssize_t sched_scaling_write(struct file *filp, const char __user *ubuf,
-				   size_t cnt, loff_t *ppos)
-{
-	unsigned int scaling;
-	int ret;
-
-	ret = kstrtouint_from_user(ubuf, cnt, 10, &scaling);
-	if (ret)
-		return ret;
-
-	if (scaling >= SCHED_TUNABLESCALING_END)
-		return -EINVAL;
-
-	sysctl_sched_tunable_scaling = scaling;
-	if (sched_update_scaling())
-		return -EINVAL;
-
-	*ppos += cnt;
-	return cnt;
-}
-
-static int sched_scaling_show(struct seq_file *m, void *v)
-{
-	seq_printf(m, "%d\n", sysctl_sched_tunable_scaling);
-	return 0;
-}
-
-static int sched_scaling_open(struct inode *inode, struct file *filp)
-{
-	return single_open(filp, sched_scaling_show, NULL);
-}
-
-static const struct file_operations sched_scaling_fops = {
-	.open		= sched_scaling_open,
-	.write		= sched_scaling_write,
-	.read		= seq_read,
-	.llseek		= seq_lseek,
-	.release	= single_release,
-};
-
 #ifdef CONFIG_SCHED_CACHE
 static ssize_t
 sched_cache_enable_write(struct file *filp, const char __user *ubuf,
@@ -726,7 +686,6 @@ static __init int sched_init_debug(void)
 	debugfs_create_u32("latency_warn_ms", 0644, debugfs_sched, &sysctl_resched_latency_warn_ms);
 	debugfs_create_u32("latency_warn_once", 0644, debugfs_sched, &sysctl_resched_latency_warn_once);
 
-	debugfs_create_file("tunable_scaling", 0644, debugfs_sched, NULL, &sched_scaling_fops);
 	debugfs_create_u32("migration_cost_ns", 0644, debugfs_sched, &sysctl_sched_migration_cost);
 	debugfs_create_u32("nr_migrate", 0644, debugfs_sched, &sysctl_sched_nr_migrate);
 
@@ -1238,12 +1197,6 @@ do {									\
 	SEQ_printf(m, "\n");
 }
 
-static const char *sched_tunable_scaling_names[] = {
-	"none",
-	"logarithmic",
-	"linear"
-};
-
 static void sched_debug_header(struct seq_file *m)
 {
 	u64 ktime, sched_clk, cpu_clk;
@@ -1286,11 +1239,6 @@ static void sched_debug_header(struct seq_file *m)
 #undef PN
 #undef P
 
-	SEQ_printf(m, "  .%-40s: %d (%s)\n",
-		"sysctl_sched_tunable_scaling",
-		sysctl_sched_tunable_scaling,
-		sched_tunable_scaling_names[sysctl_sched_tunable_scaling]);
-	SEQ_printf(m, "\n");
 }
 
 static int sched_debug_show(struct seq_file *m, void *v)
diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
index 57360f5cdde4..2f9b7d2fe813 100644
--- a/kernel/sched/fair.c
+++ b/kernel/sched/fair.c
@@ -61,19 +61,6 @@
 #include "stats.h"
 #include "autogroup.h"
 
-/*
- * The initial- and re-scaling of tunables is configurable
- *
- * Options are:
- *
- *   SCHED_TUNABLESCALING_NONE - unscaled, always *1
- *   SCHED_TUNABLESCALING_LOG - scaled logarithmically, *1+ilog(ncpus)
- *   SCHED_TUNABLESCALING_LINEAR - scaled linear, *ncpus
- *
- * (default SCHED_TUNABLESCALING_LOG = *(1+ilog(ncpus))
- */
-unsigned int sysctl_sched_tunable_scaling = SCHED_TUNABLESCALING_LOG;
-
 /*
  * Default base time slice (request size r_i) for SCHED_NORMAL/SCHED_BATCH:
  *
@@ -198,20 +185,7 @@ static inline void update_load_set(struct load_weight *lw, unsigned long w)
 static unsigned int get_update_sysctl_factor(void)
 {
 	unsigned int cpus = min_t(unsigned int, num_online_cpus(), 8);
-	unsigned int factor;
-
-	switch (sysctl_sched_tunable_scaling) {
-	case SCHED_TUNABLESCALING_NONE:
-		factor = 1;
-		break;
-	case SCHED_TUNABLESCALING_LINEAR:
-		factor = cpus;
-		break;
-	case SCHED_TUNABLESCALING_LOG:
-	default:
-		factor = 1 + ilog2(cpus);
-		break;
-	}
+	unsigned int factor = 1 + ilog2(cpus);
 
 	return factor;
 }
diff --git a/kernel/sched/sched.h b/kernel/sched/sched.h
index e656c7059bf8..7a36b3210757 100644
--- a/kernel/sched/sched.h
+++ b/kernel/sched/sched.h
@@ -3153,8 +3153,6 @@ extern unsigned int sysctl_sched_base_slice;
 extern int sysctl_resched_latency_warn_ms;
 extern int sysctl_resched_latency_warn_once;
 
-extern unsigned int sysctl_sched_tunable_scaling;
-
 extern unsigned int sysctl_numa_balancing_scan_delay;
 extern unsigned int sysctl_numa_balancing_scan_period_min;
 extern unsigned int sysctl_numa_balancing_scan_period_max;
-- 
2.55.0

[RFC PATCH 2/2] sched/fair: Remove sched_base_slice scaling altogether

[Tor Vic]

After having removed the 'none' and 'linear' scalings, remove the
entire tunable scaling mechanism altogether, and just set a fixed
default sched_base_slice of 2 msec.

This can still be changed through debugfs.

Signed-off-by: Tor Vic <torvic9@mailbox.org>
---
 kernel/sched/core.c  |  1 -
 kernel/sched/fair.c  | 53 +++-----------------------------------------
 kernel/sched/sched.h |  3 ---
 3 files changed, 3 insertions(+), 54 deletions(-)

diff --git a/kernel/sched/core.c b/kernel/sched/core.c
index 1fe40de6ebe3..2a8a242e20e5 100644
--- a/kernel/sched/core.c
+++ b/kernel/sched/core.c
@@ -8942,7 +8942,6 @@ void __init sched_init_smp(void)
 	if (set_cpus_allowed_ptr(current, housekeeping_cpumask(HK_TYPE_DOMAIN)) < 0)
 		BUG();
 	current->flags &= ~PF_NO_SETAFFINITY;
-	sched_init_granularity();
 
 	init_sched_rt_class();
 	init_sched_dl_class();
diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
index 2f9b7d2fe813..3d59af44abdf 100644
--- a/kernel/sched/fair.c
+++ b/kernel/sched/fair.c
@@ -67,10 +67,10 @@
  * Under EEVDF this is the request size used to compute the virtual
  * deadline; see update_deadline().
  *
- * (default: 0.70 msec * (1 + ilog(ncpus)), units: nanoseconds)
+ * (default: 2 msec)
  */
-unsigned int sysctl_sched_base_slice			= 700000ULL;
-static unsigned int normalized_sysctl_sched_base_slice	= 700000ULL;
+unsigned int sysctl_sched_base_slice			= 2000000ULL;
+static unsigned int normalized_sysctl_sched_base_slice	= 2000000ULL;
 
 __read_mostly unsigned int sysctl_sched_migration_cost	= 500000UL;
 
@@ -173,38 +173,6 @@ static inline void update_load_set(struct load_weight *lw, unsigned long w)
 	lw->inv_weight = 0;
 }
 
-/*
- * Increase the granularity value when there are more CPUs,
- * because with more CPUs the 'effective latency' as visible
- * to users decreases. But the relationship is not linear,
- * so pick a second-best guess by going with the log2 of the
- * number of CPUs.
- *
- * This idea comes from the SD scheduler of Con Kolivas:
- */
-static unsigned int get_update_sysctl_factor(void)
-{
-	unsigned int cpus = min_t(unsigned int, num_online_cpus(), 8);
-	unsigned int factor = 1 + ilog2(cpus);
-
-	return factor;
-}
-
-static void update_sysctl(void)
-{
-	unsigned int factor = get_update_sysctl_factor();
-
-#define SET_SYSCTL(name) \
-	(sysctl_##name = (factor) * normalized_sysctl_##name)
-	SET_SYSCTL(sched_base_slice);
-#undef SET_SYSCTL
-}
-
-void __init sched_init_granularity(void)
-{
-	update_sysctl();
-}
-
 #ifndef CONFIG_64BIT
 #define WMULT_CONST	(~0U)
 #define WMULT_SHIFT	32
@@ -1242,17 +1210,6 @@ struct sched_entity *__pick_last_entity(struct cfs_rq *cfs_rq)
 /**************************************************************
  * Scheduling class statistics methods:
  */
-int sched_update_scaling(void)
-{
-	unsigned int factor = get_update_sysctl_factor();
-
-#define WRT_SYSCTL(name) \
-	(normalized_sysctl_##name = sysctl_##name / (factor))
-	WRT_SYSCTL(sched_base_slice);
-#undef WRT_SYSCTL
-
-	return 0;
-}
 
 static void clear_buddies(struct cfs_rq *cfs_rq, struct sched_entity *se);
 
@@ -15011,15 +14968,11 @@ void sched_balance_trigger(struct rq *rq)
 
 static void rq_online_fair(struct rq *rq)
 {
-	update_sysctl();
-
 	update_runtime_enabled(rq);
 }
 
 static void rq_offline_fair(struct rq *rq)
 {
-	update_sysctl();
-
 	/* Ensure any throttled groups are reachable by pick_next_task */
 	unthrottle_offline_cfs_rqs(rq);
 
diff --git a/kernel/sched/sched.h b/kernel/sched/sched.h
index 7a36b3210757..a323fdeb7125 100644
--- a/kernel/sched/sched.h
+++ b/kernel/sched/sched.h
@@ -2302,8 +2302,6 @@ extern int group_balance_cpu(struct sched_group *sg);
 extern void update_sched_domain_debugfs(void);
 extern void dirty_sched_domain_sysctl(int cpu);
 
-extern int sched_update_scaling(void);
-
 static inline const struct cpumask *task_user_cpus(struct task_struct *p)
 {
 	if (!p->user_cpus_ptr)
@@ -2987,7 +2985,6 @@ extern void schedule_idle(void);
 asmlinkage void schedule_user(void);
 
 extern void sysrq_sched_debug_show(void);
-extern void sched_init_granularity(void);
 extern void update_max_interval(void);
 
 extern void init_sched_dl_class(void);
-- 
2.55.0