xenpm: Add a small scheduler knob "sched_smt_power_savings"

Current scheduler only care performance, thus always picks pCPU from
the most idle package. This knob provides another option to pick pCPU from
least idle package, for user who want performance power balance.

Signed-off-by: Yu Ke <ke.yu@intel.com>
Signed-off-by: Tian Kevin <kevin.tian@intel.com>

diff --git a/tools/libxc/xc_pm.c b/tools/libxc/xc_pm.c
index 918c60401086110545709feba7025659b9177cf4..13342a31213f32ead12f19196e4eb5e1187ffc78 100644 (file)
--- a/tools/libxc/xc_pm.c
+++ b/tools/libxc/xc_pm.c
@@ -345,3 +345,20 @@ int xc_get_cputopo(int xc_handle, struct xc_get_cputopo *info)
     return rc;
 }
 
+/* value:   0 - disable sched_smt_power_savings 
+            1 - enable sched_smt_power_savings
+ */
+int xc_set_sched_opt_smt(int xc_handle, uint32_t value)
+{
+   int rc;
+   DECLARE_SYSCTL;
+
+   sysctl.cmd = XEN_SYSCTL_pm_op;
+   sysctl.u.pm_op.cmd = XEN_SYSCTL_pm_op_set_sched_opt_smt;
+   sysctl.u.pm_op.cpuid = 0;
+   sysctl.u.pm_op.set_sched_opt_smt = value;
+   rc = do_sysctl(xc_handle, &sysctl);
+
+   return rc;
+}
+

diff --git a/tools/libxc/xenctrl.h b/tools/libxc/xenctrl.h
index 4a30c68fe47699b94d3f865f72c36723c32caaf1..9ce228620a870f33c4c6661f6ad3a9c08f02e15e 100644 (file)
--- a/tools/libxc/xenctrl.h
+++ b/tools/libxc/xenctrl.h
@@ -1260,4 +1260,6 @@ struct xc_get_cputopo {
 
 int xc_get_cputopo(int xc_handle, struct xc_get_cputopo *info);
 
+int xc_set_sched_opt_smt(int xc_handle, uint32_t value);
+
 #endif /* XENCTRL_H */

diff --git a/tools/misc/xenpm.c b/tools/misc/xenpm.c
index 3540ab2de8a89fdc7fea53302bd2ba69d4e57b5a..39eef6517f4bc3483f244a23375a825e434353a4 100644 (file)
--- a/tools/misc/xenpm.c
+++ b/tools/misc/xenpm.c
@@ -56,6 +56,7 @@ void show_help(void)
             " set-up-threshold      [cpuid] <num> set up threshold on CPU <cpuid> or all\n"
             "                                     it is used in ondemand governor.\n"
             " get-cpu-topology                    get thread/core/socket topology info\n"
+            " set-sched-smt           enable|disable enable/disable scheduler smt power saving\n"
             " start [seconds]                     start collect Cx/Px statistics,\n"
             "                                     output after CTRL-C or SIGINT or several seconds.\n"
             );
@@ -838,6 +839,36 @@ void cpu_topology_func(int argc, char *argv[])
     return ;
 }
 
+void set_sched_smt_func(int argc, char *argv[])
+{
+    int value, rc;
+
+    if (argc != 1){
+        show_help();
+        exit(-1);
+    }
+
+    if ( !strncmp(argv[0], "disable", sizeof("disable")) )
+    {
+        value = 0;
+    }
+    else if ( !strncmp(argv[0], "enable", sizeof("enable")) )
+    {
+        value = 1;
+    }
+    else
+    {
+        show_help();
+        exit(-1);
+    }
+
+    rc = xc_set_sched_opt_smt(xc_fd, value);
+    printf("%s sched_smt_power_savings %s\n", argv[0],
+                    rc? "failed":"successeed" );
+
+    return;
+}
+
 struct {
     const char *name;
     void (*function)(int argc, char *argv[]);
@@ -854,6 +885,7 @@ struct {
     { "set-sampling-rate", scaling_sampling_rate_func },
     { "set-up-threshold", scaling_up_threshold_func },
     { "get-cpu-topology", cpu_topology_func},
+    { "set-sched-smt", set_sched_smt_func},
 };
 
 int main(int argc, char *argv[])

diff --git a/xen/common/sched_credit.c b/xen/common/sched_credit.c
index f2c3cde0fdd481ecf6ebe6e110d30f7839c021ef..d724293bc164496c59289bc58763f5ac4d6c4e8f 100644 (file)
--- a/xen/common/sched_credit.c
+++ b/xen/common/sched_credit.c
@@ -387,7 +387,7 @@ csched_cpu_pick(struct vcpu *vc)
     {
         cpumask_t cpu_idlers;
         cpumask_t nxt_idlers;
-        int nxt;
+        int nxt, weight_cpu, weight_nxt;
 
         nxt = cycle_cpu(cpu, cpus);
 
@@ -404,7 +404,10 @@ csched_cpu_pick(struct vcpu *vc)
             cpus_and(nxt_idlers, idlers, cpu_core_map[nxt]);
         }
 
-        if ( cpus_weight(cpu_idlers) < cpus_weight(nxt_idlers) )
+        weight_cpu = cpus_weight(cpu_idlers);
+        weight_nxt = cpus_weight(nxt_idlers);
+        if ( ( (weight_cpu < weight_nxt) ^ sched_smt_power_savings )
+                && (weight_cpu != weight_nxt) )
         {
             cpu = nxt;
             cpu_clear(cpu, cpus);

diff --git a/xen/common/schedule.c b/xen/common/schedule.c
index 7ce18b14e9b3209962b9dcb54d8ffd7e60917c2c..5e91f6c85d818744f981f29010925241bb471b8b 100644 (file)
--- a/xen/common/schedule.c
+++ b/xen/common/schedule.c
@@ -38,6 +38,13 @@
 static char opt_sched[10] = "credit";
 string_param("sched", opt_sched);
 
+/* if sched_smt_power_savings is set,
+ * scheduler will give preferrence to partially idle package compared to
+ * the full idle package, when picking pCPU to schedule vCPU.
+ */
+int sched_smt_power_savings = 0;
+boolean_param("sched_smt_power_savings", sched_smt_power_savings);
+
 #define TIME_SLOP      (s32)MICROSECS(50)     /* allow time to slip a bit */
 
 /* Various timer handlers. */
@@ -942,6 +949,8 @@ void dump_runq(unsigned char key)
 
     printk("Scheduler: %s (%s)\n", ops.name, ops.opt_name);
     SCHED_OP(dump_settings);
+    printk("sched_smt_power_savings: %s\n",
+            sched_smt_power_savings? "enabled":"disabled");
     printk("NOW=0x%08X%08X\n",  (u32)(now>>32), (u32)now);
 
     for_each_online_cpu ( i )

diff --git a/xen/drivers/acpi/pmstat.c b/xen/drivers/acpi/pmstat.c
index 76e91a4288128f4502a6c918c6d9b4b9a83c009c..1ac35c8237082dc691d96e964d43efa6282e0bc0 100644 (file)
--- a/xen/drivers/acpi/pmstat.c
+++ b/xen/drivers/acpi/pmstat.c
@@ -516,6 +516,17 @@ int do_pm_op(struct xen_sysctl_pm_op *op)
         break;
     }
 
+    case XEN_SYSCTL_pm_op_set_sched_opt_smt:
+    {
+        uint32_t saved_value;
+
+        saved_value = sched_smt_power_savings;
+        sched_smt_power_savings = !!op->set_sched_opt_smt;
+        op->set_sched_opt_smt = saved_value;
+
+        break;
+    }
+
     default:
         printk("not defined sub-hypercall @ do_pm_op\n");
         ret = -ENOSYS;

diff --git a/xen/include/public/sysctl.h b/xen/include/public/sysctl.h
index 2c0c5dcc33a18aae755c8952f5a1931fa6ae61cc..48d327c06775eaea875bbc78e972be3cbf6c107c 100644 (file)
--- a/xen/include/public/sysctl.h
+++ b/xen/include/public/sysctl.h
@@ -348,11 +348,9 @@ struct xen_set_cpufreq_para {
 
     uint32_t ctrl_type;
     uint32_t ctrl_value;
-}
-;
-/* Get physical CPU topology information. */
-
+};
 
+/* Get physical CPU topology information. */
 #define INVALID_TOPOLOGY_ID  (~0U)
 struct xen_get_cputopo {
      /* IN: maximum addressable entry in
@@ -381,6 +379,9 @@ struct xen_sysctl_pm_op {
     /* get CPU topology */
     #define XEN_SYSCTL_pm_op_get_cputopo  0x20
 
+    /* set/reset scheduler power saving option */
+    #define XEN_SYSCTL_pm_op_set_sched_opt_smt    0x21
+
     uint32_t cmd;
     uint32_t cpuid;
     union {
@@ -389,6 +390,7 @@ struct xen_sysctl_pm_op {
         struct xen_set_cpufreq_para set_para;
         uint64_t get_avgfreq;
         struct xen_get_cputopo      get_topo;
+        uint32_t                    set_sched_opt_smt;
     };
 };
 

diff --git a/xen/include/xen/sched.h b/xen/include/xen/sched.h
index 5978985c42fa679b13bed8c10ba4cc5f99280f61..efaec7e9b7fa66902991dd53e707dc2e2c407653 100644 (file)
--- a/xen/include/xen/sched.h
+++ b/xen/include/xen/sched.h
@@ -550,6 +550,8 @@ uint64_t get_cpu_idle_time(unsigned int cpu);
 #define is_hvm_vcpu(v)   (is_hvm_domain(v->domain))
 #define need_iommu(d)    ((d)->need_iommu && !(d)->is_hvm)
 
+extern int sched_smt_power_savings;
+
 extern enum cpufreq_controller {
     FREQCTL_none, FREQCTL_dom0_kernel, FREQCTL_xen
 } cpufreq_controller;