summaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
authorLen Brown <len.brown@intel.com>2014-04-04 01:21:07 -0400
committerLen Brown <len.brown@intel.com>2014-04-04 12:55:29 -0400
commit0138d8f0755b5b28d0acdb0a758bcfcaf441fc58 (patch)
treee51f4df361e41e656d9d6df0ea708f81cfcfce5c
parentacead1b0fac5b10d0ae3f1cc5f7820b9f9f924f5 (diff)
downloadlinux-0138d8f0755b5b28d0acdb0a758bcfcaf441fc58.tar.bz2
intel_idle: fine-tune IVT residency targets
Ivy Town processors have slightly different properties than Ivy Bridge processors, particuarly as socket count grows. Here we add dedicated tables covering 1-2 socket, 3-4 socket, and > 4 socket IVT configurations. This reduces the frequency of deep transitions on those systems, which can impact throughput. Signed-off-by: Len Brown <len.brown@intel.com>
-rw-r--r--drivers/idle/intel_idle.c141
1 files changed, 140 insertions, 1 deletions
diff --git a/drivers/idle/intel_idle.c b/drivers/idle/intel_idle.c
index 710ab54ab04b..9ddbf550bafa 100644
--- a/drivers/idle/intel_idle.c
+++ b/drivers/idle/intel_idle.c
@@ -283,6 +283,105 @@ static struct cpuidle_state ivb_cstates[] = {
.enter = NULL }
};
+static struct cpuidle_state ivt_cstates[] = {
+ {
+ .name = "C1-IVT",
+ .desc = "MWAIT 0x00",
+ .flags = MWAIT2flg(0x00) | CPUIDLE_FLAG_TIME_VALID,
+ .exit_latency = 1,
+ .target_residency = 1,
+ .enter = &intel_idle },
+ {
+ .name = "C1E-IVT",
+ .desc = "MWAIT 0x01",
+ .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_TIME_VALID,
+ .exit_latency = 10,
+ .target_residency = 80,
+ .enter = &intel_idle },
+ {
+ .name = "C3-IVT",
+ .desc = "MWAIT 0x10",
+ .flags = MWAIT2flg(0x10) | CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
+ .exit_latency = 59,
+ .target_residency = 156,
+ .enter = &intel_idle },
+ {
+ .name = "C6-IVT",
+ .desc = "MWAIT 0x20",
+ .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
+ .exit_latency = 82,
+ .target_residency = 300,
+ .enter = &intel_idle },
+ {
+ .enter = NULL }
+};
+
+static struct cpuidle_state ivt_cstates_4s[] = {
+ {
+ .name = "C1-IVT-4S",
+ .desc = "MWAIT 0x00",
+ .flags = MWAIT2flg(0x00) | CPUIDLE_FLAG_TIME_VALID,
+ .exit_latency = 1,
+ .target_residency = 1,
+ .enter = &intel_idle },
+ {
+ .name = "C1E-IVT-4S",
+ .desc = "MWAIT 0x01",
+ .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_TIME_VALID,
+ .exit_latency = 10,
+ .target_residency = 250,
+ .enter = &intel_idle },
+ {
+ .name = "C3-IVT-4S",
+ .desc = "MWAIT 0x10",
+ .flags = MWAIT2flg(0x10) | CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
+ .exit_latency = 59,
+ .target_residency = 300,
+ .enter = &intel_idle },
+ {
+ .name = "C6-IVT-4S",
+ .desc = "MWAIT 0x20",
+ .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
+ .exit_latency = 84,
+ .target_residency = 400,
+ .enter = &intel_idle },
+ {
+ .enter = NULL }
+};
+
+static struct cpuidle_state ivt_cstates_8s[] = {
+ {
+ .name = "C1-IVT-8S",
+ .desc = "MWAIT 0x00",
+ .flags = MWAIT2flg(0x00) | CPUIDLE_FLAG_TIME_VALID,
+ .exit_latency = 1,
+ .target_residency = 1,
+ .enter = &intel_idle },
+ {
+ .name = "C1E-IVT-8S",
+ .desc = "MWAIT 0x01",
+ .flags = MWAIT2flg(0x01) | CPUIDLE_FLAG_TIME_VALID,
+ .exit_latency = 10,
+ .target_residency = 500,
+ .enter = &intel_idle },
+ {
+ .name = "C3-IVT-8S",
+ .desc = "MWAIT 0x10",
+ .flags = MWAIT2flg(0x10) | CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
+ .exit_latency = 59,
+ .target_residency = 600,
+ .enter = &intel_idle },
+ {
+ .name = "C6-IVT-8S",
+ .desc = "MWAIT 0x20",
+ .flags = MWAIT2flg(0x20) | CPUIDLE_FLAG_TIME_VALID | CPUIDLE_FLAG_TLB_FLUSHED,
+ .exit_latency = 88,
+ .target_residency = 700,
+ .enter = &intel_idle },
+ {
+ .enter = NULL }
+};
+
static struct cpuidle_state hsw_cstates[] = {
{
.name = "C1-HSW",
@@ -521,6 +620,11 @@ static const struct idle_cpu idle_cpu_ivb = {
.disable_promotion_to_c1e = true,
};
+static const struct idle_cpu idle_cpu_ivt = {
+ .state_table = ivt_cstates,
+ .disable_promotion_to_c1e = true,
+};
+
static const struct idle_cpu idle_cpu_hsw = {
.state_table = hsw_cstates,
.disable_promotion_to_c1e = true,
@@ -549,7 +653,7 @@ static const struct x86_cpu_id intel_idle_ids[] = {
ICPU(0x36, idle_cpu_atom),
ICPU(0x37, idle_cpu_byt),
ICPU(0x3a, idle_cpu_ivb),
- ICPU(0x3e, idle_cpu_ivb),
+ ICPU(0x3e, idle_cpu_ivt),
ICPU(0x3c, idle_cpu_hsw),
ICPU(0x3f, idle_cpu_hsw),
ICPU(0x45, idle_cpu_hsw),
@@ -626,6 +730,39 @@ static void intel_idle_cpuidle_devices_uninit(void)
free_percpu(intel_idle_cpuidle_devices);
return;
}
+
+/*
+ * intel_idle_state_table_update()
+ *
+ * Update the default state_table for this CPU-id
+ *
+ * Currently used to access tuned IVT multi-socket targets
+ * Assumption: num_sockets == (max_package_num + 1)
+ */
+void intel_idle_state_table_update(void)
+{
+ /* IVT uses a different table for 1-2, 3-4, and > 4 sockets */
+ if (boot_cpu_data.x86_model == 0x3e) { /* IVT */
+ int cpu, package_num, num_sockets = 1;
+
+ for_each_online_cpu(cpu) {
+ package_num = topology_physical_package_id(cpu);
+ if (package_num + 1 > num_sockets) {
+ num_sockets = package_num + 1;
+
+ if (num_sockets > 4)
+ cpuidle_state_table = ivt_cstates_8s;
+ return;
+ }
+ }
+
+ if (num_sockets > 2)
+ cpuidle_state_table = ivt_cstates_4s;
+ /* else, 1 and 2 socket systems use default ivt_cstates */
+ }
+ return;
+}
+
/*
* intel_idle_cpuidle_driver_init()
* allocate, initialize cpuidle_states
@@ -635,6 +772,8 @@ static int __init intel_idle_cpuidle_driver_init(void)
int cstate;
struct cpuidle_driver *drv = &intel_idle_driver;
+ intel_idle_state_table_update();
+
drv->state_count = 1;
for (cstate = 0; cstate < CPUIDLE_STATE_MAX; ++cstate) {