Skip to contentSkip to conversation

powernow-k8: add core performance boost support

PATCH In review Mark Langsdorf 3 participants 3 messages Mar 3, 2010 – Mar 4, 2010

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 2/3 powernow-k8: add core performance boost support

Quoted 1 line

From 7b1c9670d8f04d67cde9f810ef462ec8a8d3adbf Mon Sep 17 00:00:00 2001

From: Mark Langsdorf <[email protected]>
Date: Wed, 3 Mar 2010 14:33:43 -0600
Subject: [PATCH 1/2] powernow-k8: add core performance boost support

Add CPUID check for hardware CPB support

Also, update copyright while at it.

---
 arch/x86/kernel/cpu/cpufreq/powernow-k8.c |   41 ++++++++++++++++++++++++++---
 arch/x86/kernel/cpu/cpufreq/powernow-k8.h |    3 +-
 include/linux/cpufreq.h                   |    4 +++
 3 files changed, 42 insertions(+), 6 deletions(-)
arch/x86/kernel/cpu/cpufreq/powernow-k8.c +37 -4
index 07f18a4..0a6f1f8 100644
@@ -1,6 +1,5 @@
-
 /*
- *   (c) 2003-2006 Advanced Micro Devices, Inc.
+ *   (c) 2003-2009 Advanced Micro Devices, Inc.
  *  Your use of this code is subject to the terms and conditions of the
  *  GNU general public license version 2. See "COPYING" or
  *  http://www.gnu.org/licenses/gpl.html
@@ -57,6 +56,9 @@ static int cpu_family = CPU_OPTERON;
 
 static struct cpufreq_driver cpufreq_amd64_driver;
 
+/* core performance boost */
+static bool cpb_capable;
+
 #ifndef CONFIG_SMP
 static inline const struct cpumask *cpu_core_mask(int cpu)
 {
@@ -511,6 +513,24 @@ static int core_voltage_post_transition(struct powernow_k8_data *data,
 	return 0;
 }
 
+/* CPB 0=disable, 1=enable. */
+static void cpb_toggle(u32 t)
+{
+	if (cpb_capable) {
+		u32 lo, hi;
+		rdmsr(MSR_K7_HWCR, lo, hi);
+
+		if (t)
+			lo &= ~(1 << 25);
+		else
+			lo |= (1 << 25);
+
+		wrmsr(MSR_K7_HWCR, lo, hi);
+
+		printk(KERN_ERR "CPB: %s.\n", (t ? "on" : "off"));
+	}
+}
+
 static void check_supported_cpu(void *_rc)
 {
 	u32 eax, ebx, ecx, edx;
@@ -1173,8 +1193,8 @@ static int powernowk8_target(struct cpufreq_policy *pol,
 		}
 	}
 
-	if (cpufreq_frequency_table_target(pol, data->powernow_table,
-				targfreq, relation, &newstate))
+	if (cpufreq_frequency_table_target(pol, data->powernow_table, targfreq,
+					   relation, &newstate))
 		goto err_out;
 
 	mutex_lock(&fidvid_mutex);
@@ -1191,6 +1211,9 @@ static int powernowk8_target(struct cpufreq_policy *pol,
 		mutex_unlock(&fidvid_mutex);
 		goto err_out;
 	}
+
+	cpb_toggle(pol->flags.cpb);
+
 	mutex_unlock(&fidvid_mutex);
 
 	if (cpu_family == CPU_HW_PSTATE)
@@ -1330,6 +1353,7 @@ static int __cpuinit powernowk8_cpu_init(struct cpufreq_policy *pol)
 	/* Check for APERF/MPERF support in hardware */
 	if (cpu_has(c, X86_FEATURE_APERFMPERF))
 		cpufreq_amd64_driver.getavg = get_measured_perf;
+	pol->flags.cpb = cpb_capable;
 
 	cpufreq_frequency_table_get_attr(data->powernow_table, pol->cpu);
 
@@ -1419,11 +1443,20 @@ static struct cpufreq_driver cpufreq_amd64_driver = {
 	.attr		= powernow_k8_attr,
 };
 
+static bool __cpuinit check_cpb_capable(void)
+{
+	u32 tmp = cpuid_edx(CPUID_FREQ_VOLT_CAPABILITIES);
+
+	return (tmp & CPB_CAPABLE);
+}
+
 /* driver entry point for init */
 static int __cpuinit powernowk8_init(void)
 {
 	unsigned int i, supported_cpus = 0;
 
+	cpb_capable = check_cpb_capable();
+
 	for_each_online_cpu(i) {
 		int rc;
 		smp_call_function_single(i, check_supported_cpu, &rc, 1);
arch/x86/kernel/cpu/cpufreq/powernow-k8.h +1 -2
index 02ce824..3be308e 100644
@@ -5,7 +5,6 @@
  *  http://www.gnu.org/licenses/gpl.html
  */
 
-
 enum pstate {
 	HW_PSTATE_INVALID = 0xff,
 	HW_PSTATE_0 = 0,
@@ -55,7 +54,6 @@ struct powernow_k8_data {
 	struct cpumask *available_cores;
 };
 
-
 /* processor's cpuid instruction support */
 #define CPUID_PROCESSOR_SIGNATURE	1	/* function 1 */
 #define CPUID_XFAM			0x0ff00000	/* extended family */
@@ -67,6 +65,7 @@ struct powernow_k8_data {
 #define CPUID_GET_MAX_CAPABILITIES	0x80000000
 #define CPUID_FREQ_VOLT_CAPABILITIES	0x80000007
 #define P_STATE_TRANSITION_CAPABLE	6
+#define CPB_CAPABLE			0x00000200
 
 /* Model Specific Registers for p-state transitions. MSRs are 64-bit. For     */
 /* writes (wrmsr - opcode 0f 30), the register number is placed in ecx, and   */
include/linux/cpufreq.h +4 -0
index 4de02b1..e738576 100644
@@ -104,6 +104,10 @@ struct cpufreq_policy {
 
 	struct kobject		kobj;
 	struct completion	kobj_unregister;
+
+	struct flags {
+		unsigned long cpb:1;	/* toggle CPB on this cpu */
+	} flags;
 };
 
 #define CPUFREQ_ADJUST		(0)
Signature · 1 line
1.6.0.2
#1 lore raw
Quoted 40 lines · On Wed, 3 Mar 2010 14:58:27 -0600 Mark Langsdorf <[email protected]> wrote:
Quoted 1 line

From 7b1c9670d8f04d67cde9f810ef462ec8a8d3adbf Mon Sep 17 00:00:00 2001

From: Mark Langsdorf <[email protected]>
Date: Wed, 3 Mar 2010 14:33:43 -0600
Subject: [PATCH 1/2] powernow-k8: add core performance boost support

Add CPUID check for hardware CPB support

Also, update copyright while at it.

---
 arch/x86/kernel/cpu/cpufreq/powernow-k8.c |   41 ++++++++++++++++++++++++++---
 arch/x86/kernel/cpu/cpufreq/powernow-k8.h |    3 +-
 include/linux/cpufreq.h                   |    4 +++
 3 files changed, 42 insertions(+), 6 deletions(-)
arch/x86/kernel/cpu/cpufreq/powernow-k8.c +4 -2
index 07f18a4..0a6f1f8 100644
@@ -1,6 +1,5 @@
-
 /*
- *   (c) 2003-2006 Advanced Micro Devices, Inc.
+ *   (c) 2003-2009 Advanced Micro Devices, Inc.
  *  Your use of this code is subject to the terms and conditions of the
  *  GNU general public license version 2. See "COPYING" or
  *  http://www.gnu.org/licenses/gpl.html
@@ -57,6 +56,9 @@ static int cpu_family = CPU_OPTERON;
 
 static struct cpufreq_driver cpufreq_amd64_driver;
 
+/* core performance boost */
+static bool cpb_capable;
+
 #ifndef CONFIG_SMP
 static inline const struct cpumask *cpu_core_mask(int cpu)
 {

arch/x86/kernel/cpu/cpufreq/powernow-k8.c static int core_voltage_post_transition(struct powernow_k8_data *data, · line 522

Quoted 10 lines
arch/x86/kernel/cpu/cpufreq/powernow-k8.c +7 -0
@@ -511,6 +513,24 @@ static int core_voltage_post_transition(struct powernow_k8_data *data,
 	return 0;
 }
 
+/* CPB 0=disable, 1=enable. */
+static void cpb_toggle(u32 t)
+{
+	if (cpb_capable) {
+		u32 lo, hi;
+		rdmsr(MSR_K7_HWCR, lo, hi);
+

The newline usually goes after end-of-locals, before start-of-code.

arch/x86/kernel/cpu/cpufreq/powernow-k8.c static int core_voltage_post_transition(struct powernow_k8_data *data, · line 530

Quoted 8 lines
arch/x86/kernel/cpu/cpufreq/powernow-k8.c +8 -0
+		if (t)
+			lo &= ~(1 << 25);
+		else
+			lo |= (1 << 25);
+
+		wrmsr(MSR_K7_HWCR, lo, hi);
+
+		printk(KERN_ERR "CPB: %s.\n", (t ? "on" : "off"));

Why KERN_ERR? It's not an error?

Do we want a printk here at all? Under which circumstances will it come out? Does it have sufficient context for people to be able to understand what it means, and which subsystem it's referring to? If you phone your Aunt Tillie and tell her "CPB: on", will she understand what you mean?

include/linux/cpufreq.h struct cpufreq_policy { · line 111

Quoted 11 lines
include/linux/cpufreq.h +4 -0
@@ -104,6 +104,10 @@ struct cpufreq_policy {
 
 	struct kobject		kobj;
 	struct completion	kobj_unregister;
+
+	struct flags {
+		unsigned long cpb:1;	/* toggle CPB on this cpu */
+	} flags;
 };

Bear in mind that the compiler provides no atomicity support for bitfields. So if someone later comes along and adds a new field to flags', they will need to provide external synchronisation (ie: a lock) to protect that field during modifications to cpb'.

IOW, this is a bit of a hand-grenade.

#2 lore raw

From: Andrew Morton <[email protected]>
Date: Thu, Mar 04, 2010 at 01:23:28PM -0800

Hi Andrew,

[..]

Quoted 9 lines

arch/x86/kernel/cpu/cpufreq/powernow-k8.c static int core_voltage_post_transition(struct powernow_k8_data *data, · line 522

Quoted 7 lines
arch/x86/kernel/cpu/cpufreq/powernow-k8.c +7 -0
+/* CPB 0=disable, 1=enable. */
+static void cpb_toggle(u32 t)
+{
+	if (cpb_capable) {
+		u32 lo, hi;
+		rdmsr(MSR_K7_HWCR, lo, hi);
+

The newline usually goes after end-of-locals, before start-of-code.

Done.

Quoted 11 lines

arch/x86/kernel/cpu/cpufreq/powernow-k8.c static int core_voltage_post_transition(struct powernow_k8_data *data, · line 530

Quoted 8 lines
arch/x86/kernel/cpu/cpufreq/powernow-k8.c +8 -0
+		if (t)
+			lo &= ~(1 << 25);
+		else
+			lo |= (1 << 25);
+
+		wrmsr(MSR_K7_HWCR, lo, hi);
+
+		printk(KERN_ERR "CPB: %s.\n", (t ? "on" : "off"));

Why KERN_ERR? It's not an error?

Quoted 5 lines

Do we want a printk here at all? Under which circumstances will it come out? Does it have sufficient context for people to be able to understand what it means, and which subsystem it's referring to? If you phone your Aunt Tillie and tell her "CPB: on", will she understand what you mean?

This isn't actually supposed to be there - it was there only for debugging.

These patches weren't supposed to go out just yet so please drop them from your tree for now - I'll have corrected versions with proper changelogs soon.

Quoted 17 lines

include/linux/cpufreq.h struct cpufreq_policy { · line 111

Quoted 11 lines
include/linux/cpufreq.h +4 -0
@@ -104,6 +104,10 @@ struct cpufreq_policy {
 
 	struct kobject		kobj;
 	struct completion	kobj_unregister;
+
+	struct flags {
+		unsigned long cpb:1;	/* toggle CPB on this cpu */
+	} flags;
 };

Bear in mind that the compiler provides no atomicity support for bitfields. So if someone later comes along and adds a new field to flags', they will need to provide external synchronisation (ie: a lock) to protect that field during modifications to cpb'.

Quoted 1 line

IOW, this is a bit of a hand-grenade.

Ok, I'll switch to a unsigned long for the flags.

Thanks.

Signature · 6 lines
Regards/Gruss,
Boris.

-
Advanced Micro Devices, Inc.
Operating Systems Research Center