sched: Make tunable scaling style configurable

As scaling now takes place on all kind of cpu add/remove events a user that configures values via proc should be able to configure if his set values are still rescaled or kept whatever happens. As the comments state that log2 was just a second guess that worked the interface is not just designed for on/off, but to choose a scaling type. Currently this allows none, log and linear, but more important it allwos us to keep the interface even if someone has an even better idea how to scale the values. Signed-off-by: Christian Ehrhardt <ehrhardt@linux.vnet.ibm.com> Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl> LKML-Reference: <1259579808-11357-3-git-send-email-ehrhardt@linux.vnet.ibm.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>

sched: Make tunable scaling style configurable
As scaling now takes place on all kind of cpu add/remove events a user that configures values via proc should be able to configure if his set values are still rescaled or kept whatever happens. As the comments state that log2 was just a second guess that worked the interface is not just designed for on/off, but to choose a scaling type. Currently this allows none, log and linear, but more important it allwos us to keep the interface even if someone has an even better idea how to scale the values. Signed-off-by: Christian Ehrhardt <ehrhardt@linux.vnet.ibm.com> Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl> LKML-Reference: <1259579808-11357-3-git-send-email-ehrhardt@linux.vnet.ibm.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
Christian Ehrhardt · Ingo Molnar
1 parent 0bcdcf28c9
Showing 5 changed files with 61 additions and 2 deletions Side-by-side Diff
include/linux/sched.h
kernel/sched.c
kernel/sched_debug.c
kernel/sched_fair.c
kernel/sysctl.c
@@ -1902,13 +1902,22 @@
 extern unsigned int sysctl_sched_shares_ratelimit;
 extern unsigned int sysctl_sched_shares_thresh;
 extern unsigned int sysctl_sched_child_runs_first;
+
+enum sched_tunable_scaling {
+	SCHED_TUNABLESCALING_NONE,
+	SCHED_TUNABLESCALING_LOG,
+	SCHED_TUNABLESCALING_LINEAR,
+	SCHED_TUNABLESCALING_END,
+};
+extern enum sched_tunable_scaling sysctl_sched_tunable_scaling;
+
 #ifdef CONFIG_SCHED_DEBUG
 extern unsigned int sysctl_sched_migration_cost;
 extern unsigned int sysctl_sched_nr_migrate;
 extern unsigned int sysctl_sched_time_avg;
 extern unsigned int sysctl_timer_migration;
  
-int sched_nr_latency_handler(struct ctl_table *table, int write,
+int sched_proc_update_handler(struct ctl_table *table, int write,
 		void __user *buffer, size_t *length,
 		loff_t *ppos);
 #endif
@@ -7033,7 +7033,20 @@
 static void update_sysctl(void)
 {
 	unsigned int cpus = min(num_online_cpus(), 8U);
-	unsigned int factor = 1 + ilog2(cpus);
+	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;
+	}
  
 #define SET_SYSCTL(name) \
 	(sysctl_##name = (factor) * normalized_sysctl_##name)
@@ -309,6 +309,12 @@
 	print_rq(m, rq, cpu);
 }
  
+static const char *sched_tunable_scaling_names[] = {
+	"none",
+	"logaritmic",
+	"linear"
+};
+
 static int sched_debug_show(struct seq_file *m, void *v)
 {
 	u64 now = ktime_to_ns(ktime_get());
@@ -333,6 +339,10 @@
 	P(sysctl_sched_features);
 #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]);
  
 	for_each_online_cpu(cpu)
 		print_cpu(m, cpu);
@@ -21,6 +21,7 @@
  */
  
 #include <linux/latencytop.h>
+#include <linux/sched.h>
  
 /*
  * Targeted preemption latency for CPU-bound tasks:
@@ -36,6 +37,18 @@
  */
 unsigned int sysctl_sched_latency = 5000000ULL;
 unsigned int normalized_sysctl_sched_latency = 5000000ULL;
+
+/*
+ * The initial- and re-scaling of tunables is configurable
+ * (default SCHED_TUNABLESCALING_LOG = *(1+ilog(ncpus))
+ *
+ * Options are:
+ * SCHED_TUNABLESCALING_NONE - unscaled, always *1
+ * SCHED_TUNABLESCALING_LOG - scaled logarithmical, *1+ilog(ncpus)
+ * SCHED_TUNABLESCALING_LINEAR - scaled linear, *ncpus
+ */
+enum sched_tunable_scaling sysctl_sched_tunable_scaling
+	= SCHED_TUNABLESCALING_LOG;
  
 /*
  * Minimal preemption granularity for CPU-bound tasks:
@@ -251,6 +251,8 @@
 static int max_sched_granularity_ns = NSEC_PER_SEC;	/* 1 second */
 static int min_wakeup_granularity_ns;			/* 0 usecs */
 static int max_wakeup_granularity_ns = NSEC_PER_SEC;	/* 1 second */
+static int min_sched_tunable_scaling = SCHED_TUNABLESCALING_NONE;
+static int max_sched_tunable_scaling = SCHED_TUNABLESCALING_END-1;
 #endif
  
 static struct ctl_table kern_table[] = {
@@ -304,6 +306,18 @@
 		.mode		= 0644,
 		.proc_handler	= &proc_dointvec,
 	},
+	{
+		.ctl_name	= CTL_UNNUMBERED,
+		.procname	= "sched_tunable_scaling",
+		.data		= &sysctl_sched_tunable_scaling,
+		.maxlen		= sizeof(enum sched_tunable_scaling),
+		.mode		= 0644,
+		.proc_handler	= &proc_dointvec_minmax,
+		.strategy	= &sysctl_intvec,
+		.extra1		= &min_sched_tunable_scaling,
+		.extra2		= &max_sched_tunable_scaling,
+	},
+
 	{
 		.ctl_name	= CTL_UNNUMBERED,
 		.procname	= "sched_shares_thresh",
...	...	@@ -1902,13 +1902,22 @@
1902	1902	extern unsigned int sysctl_sched_shares_ratelimit;
1903	1903	extern unsigned int sysctl_sched_shares_thresh;
1904	1904	extern unsigned int sysctl_sched_child_runs_first;
	1905	+
	1906	+enum sched_tunable_scaling {
	1907	+ SCHED_TUNABLESCALING_NONE,
	1908	+ SCHED_TUNABLESCALING_LOG,
	1909	+ SCHED_TUNABLESCALING_LINEAR,
	1910	+ SCHED_TUNABLESCALING_END,
	1911	+};
	1912	+extern enum sched_tunable_scaling sysctl_sched_tunable_scaling;
	1913	+
1905	1914	#ifdef CONFIG_SCHED_DEBUG
1906	1915	extern unsigned int sysctl_sched_migration_cost;
1907	1916	extern unsigned int sysctl_sched_nr_migrate;
1908	1917	extern unsigned int sysctl_sched_time_avg;
1909	1918	extern unsigned int sysctl_timer_migration;
1910	1919
1911		-int sched_nr_latency_handler(struct ctl_table *table, int write,
	1920	+int sched_proc_update_handler(struct ctl_table *table, int write,
1912	1921	void __user buffer, size_t length,
1913	1922	loff_t *ppos);
1914	1923	#endif
...	...	@@ -7033,7 +7033,20 @@
7033	7033	static void update_sysctl(void)
7034	7034	{
7035	7035	unsigned int cpus = min(num_online_cpus(), 8U);
7036		- unsigned int factor = 1 + ilog2(cpus);
	7036	+ unsigned int factor;
	7037	+
	7038	+ switch (sysctl_sched_tunable_scaling) {
	7039	+ case SCHED_TUNABLESCALING_NONE:
	7040	+ factor = 1;
	7041	+ break;
	7042	+ case SCHED_TUNABLESCALING_LINEAR:
	7043	+ factor = cpus;
	7044	+ break;
	7045	+ case SCHED_TUNABLESCALING_LOG:
	7046	+ default:
	7047	+ factor = 1 + ilog2(cpus);
	7048	+ break;
	7049	+ }
7037	7050
7038	7051	#define SET_SYSCTL(name) \
7039	7052	(sysctl_##name = (factor) * normalized_sysctl_##name)
...	...	@@ -309,6 +309,12 @@
309	309	print_rq(m, rq, cpu);
310	310	}
311	311
	312	+static const char *sched_tunable_scaling_names[] = {
	313	+ "none",
	314	+ "logaritmic",
	315	+ "linear"
	316	+};
	317	+
312	318	static int sched_debug_show(struct seq_file m, void v)
313	319	{
314	320	u64 now = ktime_to_ns(ktime_get());
...	...	@@ -333,6 +339,10 @@
333	339	P(sysctl_sched_features);
334	340	#undef PN
335	341	#undef P
	342	+
	343	+ SEQ_printf(m, " .%-40s: %d (%s)\n", "sysctl_sched_tunable_scaling",
	344	+ sysctl_sched_tunable_scaling,
	345	+ sched_tunable_scaling_names[sysctl_sched_tunable_scaling]);
336	346
337	347	for_each_online_cpu(cpu)
338	348	print_cpu(m, cpu);
...	...	@@ -21,6 +21,7 @@
21	21	*/
22	22
23	23	#include <linux/latencytop.h>
	24	+#include <linux/sched.h>
24	25
25	26	/*
26	27	* Targeted preemption latency for CPU-bound tasks:
...	...	@@ -36,6 +37,18 @@
36	37	*/
37	38	unsigned int sysctl_sched_latency = 5000000ULL;
38	39	unsigned int normalized_sysctl_sched_latency = 5000000ULL;
	40	+
	41	+/*
	42	+ * The initial- and re-scaling of tunables is configurable
	43	+ * (default SCHED_TUNABLESCALING_LOG = *(1+ilog(ncpus))
	44	+ *
	45	+ * Options are:
	46	+ * SCHED_TUNABLESCALING_NONE - unscaled, always *1
	47	+ * SCHED_TUNABLESCALING_LOG - scaled logarithmical, *1+ilog(ncpus)
	48	+ * SCHED_TUNABLESCALING_LINEAR - scaled linear, *ncpus
	49	+ */
	50	+enum sched_tunable_scaling sysctl_sched_tunable_scaling
	51	+ = SCHED_TUNABLESCALING_LOG;
39	52
40	53	/*
41	54	* Minimal preemption granularity for CPU-bound tasks:
...	...	@@ -251,6 +251,8 @@
251	251	static int max_sched_granularity_ns = NSEC_PER_SEC; /* 1 second */
252	252	static int min_wakeup_granularity_ns; /* 0 usecs */
253	253	static int max_wakeup_granularity_ns = NSEC_PER_SEC; /* 1 second */
	254	+static int min_sched_tunable_scaling = SCHED_TUNABLESCALING_NONE;
	255	+static int max_sched_tunable_scaling = SCHED_TUNABLESCALING_END-1;
254	256	#endif
255	257
256	258	static struct ctl_table kern_table[] = {
...	...	@@ -304,6 +306,18 @@
304	306	.mode = 0644,
305	307	.proc_handler = &proc_dointvec,
306	308	},
	309	+ {
	310	+ .ctl_name = CTL_UNNUMBERED,
	311	+ .procname = "sched_tunable_scaling",
	312	+ .data = &sysctl_sched_tunable_scaling,
	313	+ .maxlen = sizeof(enum sched_tunable_scaling),
	314	+ .mode = 0644,
	315	+ .proc_handler = &proc_dointvec_minmax,
	316	+ .strategy = &sysctl_intvec,
	317	+ .extra1 = &min_sched_tunable_scaling,
	318	+ .extra2 = &max_sched_tunable_scaling,
	319	+ },
	320	+
307	321	{
308	322	.ctl_name = CTL_UNNUMBERED,
309	323	.procname = "sched_shares_thresh",