Skip to contentSkip to conversation

Add cpuidle support for at91

PATCH Superseded Nicolas Ferre 3 participants 6 messages Sep 29, 2009 – Oct 22, 2009 6 messages not in the archive

lore
Share & tools
lore

A message that is not in the archive (sent off-list or not yet received): 1247680190-19674-1-git-send-email-albin.tonnerre@free-electrons.com

A message that is not in the archive (sent off-list or not yet received): [email protected]

A message that is not in the archive (sent off-list or not yet received): [email protected]

A message that is not in the archive (sent off-list or not yet received): [email protected]

A message that is not in the archive (sent off-list or not yet received): [email protected]

A message that is not in the archive (sent off-list or not yet received): [email protected]

#6 lore raw

PATCH v2 1/1 Add cpuidle support for at91

Salut Albin !

Albin Tonnerre :

Quoted 1 line

Nicolas, Andrew: ping?

Indeed it's been a long time...

Quoted 8 lines · Albin Tonnerre
Quoted 7 lines

This patch adds the support for cpuidle on AT91 SoCs, taken from the cpuidle support in mach-kirkwood.
cpuidle needs sdram_selfrefresh_enable and _disable, so move their definition to a separate header file instead of duplicating the code already used in pm.c.

[nicolas.ferre at atmel.com: tested on at91sam9263ek]

Works very well and allow to save some power. After quick measures, we run at 172mA at 12V instead of 190mA at 12V so ~2.06W and a saving of ~10% :
nice job !
This is of course in my test conditions.

Just one precision: what is the difference, entering state0 with only the current cpu_do_idle() that is activated by default ?

Quoted 294 lines
Quoted 294 lines

---
Changelog since V1:
Updated thanks to feedback from Marc Pignat:
- On AT91RM200, it is necessary to make sure that the Low-power mode is left before entering the self-refresh mode
- Document the fact that restoring the low-power mode when we're not sure we're out of self-refresh mode is not recommended on RM200

 arch/arm/mach-at91/Makefile  |    1 +
 arch/arm/mach-at91/cpuidle.c |  101 ++++++++++++++++++++++++++++++++++++++++++
 arch/arm/mach-at91/pm.c      |   62 ++------------------------
 arch/arm/mach-at91/pm.h      |   62 ++++++++++++++++++++++++++
 4 files changed, 168 insertions(+), 58 deletions(-)
 mode change 100644 => 100755 arch/arm/mach-at91/Makefile
 create mode 100644 arch/arm/mach-at91/cpuidle.c
 create mode 100644 arch/arm/mach-at91/pm.h
arch/arm/mach-at91/Makefile +1 -0
old mode 100644
new mode 100755
index c69ff23..06d189d
@@ -67,6 +67,7 @@ obj-y				+= leds.o
 # Power Management
 obj-$(CONFIG_PM)		+= pm.o
 obj-$(CONFIG_AT91_SLOW_CLOCK)	+= pm_slowclock.o
+obj-$(CONFIG_CPU_IDLE)	+= cpuidle.o
 
 ifeq ($(CONFIG_PM_DEBUG),y)
 CFLAGS_pm.o += -DDEBUG
arch/arm/mach-at91/cpuidle.c new file +101 -0
new file mode 100644
index 0000000..9146135
@@ -0,0 +1,101 @@
+/*
+ * based on arch/arm/mach-kirkwood/cpuidle.c
+ *
+ * CPU idle support for AT91 SoC
+ *
+ * This file is licensed under the terms of the GNU General Public
+ * License version 2.  This program is licensed "as is" without any
+ * warranty of any kind, whether express or implied.
+ *
+ * The cpu idle uses wait-for-interrupt and RAM self refresh in order
+ * to implement two idle states -
+ * #1 wait-for-interrupt
+ * #2 wait-for-interrupt and RAM self refresh
+ */
+
+#include <linux/kernel.h>
+#include <linux/init.h>
+#include <linux/platform_device.h>
+#include <linux/cpuidle.h>
+#include <asm/proc-fns.h>
+#include <linux/io.h>
+
+#include "pm.h"
+
+#define AT91_MAX_STATES	2
+
+static DEFINE_PER_CPU(struct cpuidle_device, at91_cpuidle_device);
+
+static struct cpuidle_driver at91_idle_driver = {
+	.name =         "at91_idle",
+	.owner =        THIS_MODULE,
+};
+
+/* Actual code that puts the SoC in different idle states */
+static int at91_enter_idle(struct cpuidle_device *dev,
+			       struct cpuidle_state *state)
+{
+	struct timeval before, after;
+	int idle_time;
+	u32 saved_lpr;
+
+	local_irq_disable();
+	do_gettimeofday(&before);
+	if (state == &dev->states[0])
+		/* Wait for interrupt state */
+		cpu_do_idle();
+	else if (state == &dev->states[1]) {
+		asm("b 1f; .align 5; 1:");
+		asm("mcr p15, 0, r0, c7, c10, 4");	/* drain write buffer */
+		saved_lpr = sdram_selfrefresh_enable();
+		cpu_do_idle();
+		/*
+		 * On AT91RM200, self-refresh mode is exited as soon as a memory access
+		 * is made, but we don't know for sure when that happens. However, we
+		 * need to restore the low-power mode if it was enabled before going
+		 * idle. Restoring low-power mode while still in self-refresh is "not
+		 * recommended", but seems to work.
+		 */
+		sdram_selfrefresh_disable(saved_lpr);
+	}
+	do_gettimeofday(&after);
+	local_irq_enable();
+	idle_time = (after.tv_sec - before.tv_sec) * USEC_PER_SEC +
+			(after.tv_usec - before.tv_usec);
+	return idle_time;
+}
+
+/* Initialize CPU idle by registering the idle states */
+static int at91_init_cpuidle(void)
+{
+	struct cpuidle_device *device;
+
+	cpuidle_register_driver(&at91_idle_driver);
+
+	device = &per_cpu(at91_cpuidle_device, smp_processor_id());
+	device->state_count = AT91_MAX_STATES;
+
+	/* Wait for interrupt state */
+	device->states[0].enter = at91_enter_idle;
+	device->states[0].exit_latency = 1;
+	device->states[0].target_residency = 10000;
+	device->states[0].flags = CPUIDLE_FLAG_TIME_VALID;
+	strcpy(device->states[0].name, "WFI");
+	strcpy(device->states[0].desc, "Wait for interrupt");
+
+	/* Wait for interrupt and RAM self refresh state */
+	device->states[1].enter = at91_enter_idle;
+	device->states[1].exit_latency = 10;
+	device->states[1].target_residency = 10000;
+	device->states[1].flags = CPUIDLE_FLAG_TIME_VALID;
+	strcpy(device->states[1].name, "RAM_SR");
+	strcpy(device->states[1].desc, "WFI and RAM Self Refresh");
+
+	if (cpuidle_register_device(device)) {
+		printk(KERN_ERR "at91_init_cpuidle: Failed registering\n");
+		return -EIO;
+	}
+	return 0;
+}
+
+device_initcall(at91_init_cpuidle);
arch/arm/mach-at91/pm.c +4 -58
index e26c4fe..0a68c46 100644
@@ -29,62 +29,7 @@
 #include <mach/cpu.h>
 
 #include "generic.h"
-
-#ifdef CONFIG_ARCH_AT91RM9200
-#include <mach/at91rm9200_mc.h>
-
-/*
- * The AT91RM9200 goes into self-refresh mode with this command, and will
- * terminate self-refresh automatically on the next SDRAM access.
- */
-#define sdram_selfrefresh_enable()	at91_sys_write(AT91_SDRAMC_SRR, 1)
-#define sdram_selfrefresh_disable()	do {} while (0)
-
-#elif defined(CONFIG_ARCH_AT91CAP9)
-#include <mach/at91cap9_ddrsdr.h>
-
-static u32 saved_lpr;
-
-static inline void sdram_selfrefresh_enable(void)
-{
-	u32 lpr;
-
-	saved_lpr = at91_sys_read(AT91_DDRSDRC_LPR);
-
-	lpr = saved_lpr & ~AT91_DDRSDRC_LPCB;
-	at91_sys_write(AT91_DDRSDRC_LPR, lpr | AT91_DDRSDRC_LPCB_SELF_REFRESH);
-}
-
-#define sdram_selfrefresh_disable()	at91_sys_write(AT91_DDRSDRC_LPR, saved_lpr)
-
-#else
-#include <mach/at91sam9_sdramc.h>
-
-#ifdef CONFIG_ARCH_AT91SAM9263
-/*
- * FIXME either or both the SDRAM controllers (EB0, EB1) might be in use;
- * handle those cases both here and in the Suspend-To-RAM support.
- */
-#define	AT91_SDRAMC	AT91_SDRAMC0
-#warning Assuming EB1 SDRAM controller is *NOT* used
-#endif
-
-static u32 saved_lpr;
-
-static inline void sdram_selfrefresh_enable(void)
-{
-	u32 lpr;
-
-	saved_lpr = at91_sys_read(AT91_SDRAMC_LPR);
-
-	lpr = saved_lpr & ~AT91_SDRAMC_LPCB;
-	at91_sys_write(AT91_SDRAMC_LPR, lpr | AT91_SDRAMC_LPCB_SELF_REFRESH);
-}
-
-#define sdram_selfrefresh_disable()	at91_sys_write(AT91_SDRAMC_LPR, saved_lpr)
-
-#endif
-
+#include "pm.h"
 
 /*
  * Show the reason for the previous system reset.
@@ -259,6 +204,7 @@ extern u32 at91_slow_clock_sz;
 
 static int at91_pm_enter(suspend_state_t state)
 {
+	u32 saved_lpr;
 	at91_gpio_suspend();
 	at91_irq_suspend();
 
@@ -314,9 +260,9 @@ static int at91_pm_enter(suspend_state_t state)
 			 */
 			asm("b 1f; .align 5; 1:");
 			asm("mcr p15, 0, r0, c7, c10, 4");	/* drain write buffer */
-			sdram_selfrefresh_enable();
+			saved_lpr = sdram_selfrefresh_enable();
 			asm("mcr p15, 0, r0, c7, c0, 4");	/* wait for interrupt */
-			sdram_selfrefresh_disable();
+			sdram_selfrefresh_disable(saved_lpr);
 			break;
 
 		case PM_SUSPEND_ON:
arch/arm/mach-at91/pm.h new file +62 -0
new file mode 100644
index 0000000..0b8b717
@@ -0,0 +1,62 @@
+#ifdef CONFIG_ARCH_AT91RM9200
+#include <mach/at91rm9200_mc.h>
+
+/*
+ * The AT91RM9200 goes into self-refresh mode with this command, and will
+ * terminate self-refresh automatically on the next SDRAM access.
+ */
+
+static inline u32 sdram_selfrefresh_enable(void)
+{
+	u32 saved_lpr = at91_sys_read(AT91_SDRAMC_LPR);
+
+	at91_sys_write(AT91_SDRAMC_LPR, 0);
+	at91_sys_write(AT91_SDRAMC_SRR, 1);
+	return saved_lpr;
+}
+
+#define sdram_selfrefresh_disable(saved_lpr)	do { at91_sys_write(AT91_SDRAMC_LPR, saved_lpr); } while (0)
+
+#elif defined(CONFIG_ARCH_AT91CAP9)
+#include <mach/at91cap9_ddrsdr.h>
+
+
+static inline u32 sdram_selfrefresh_enable(void)
+{
+	u32 saved_lpr, lpr;
+
+	saved_lpr = at91_sys_read(AT91_DDRSDRC_LPR);
+
+	lpr = saved_lpr & ~AT91_DDRSDRC_LPCB;
+	at91_sys_write(AT91_DDRSDRC_LPR, lpr | AT91_DDRSDRC_LPCB_SELF_REFRESH);
+	return saved_lpr;
+}
+
+#define sdram_selfrefresh_disable(saved_lpr)	at91_sys_write(AT91_DDRSDRC_LPR, saved_lpr)
+
+#else
+#include <mach/at91sam9_sdramc.h>
+
+#ifdef CONFIG_ARCH_AT91SAM9263
+/*
+ * FIXME either or both the SDRAM controllers (EB0, EB1) might be in use;
+ * handle those cases both here and in the Suspend-To-RAM support.
+ */
+#define	AT91_SDRAMC	AT91_SDRAMC0
+#warning Assuming EB1 SDRAM controller is *NOT* used
+#endif
+
+static inline u32 sdram_selfrefresh_enable(void)
+{
+	u32 saved_lpr, lpr;
+
+	saved_lpr = at91_sys_read(AT91_SDRAMC_LPR);
+
+	lpr = saved_lpr & ~AT91_SDRAMC_LPCB;
+	at91_sys_write(AT91_SDRAMC_LPR, lpr | AT91_SDRAMC_LPCB_SELF_REFRESH);
+	return saved_lpr;
+}
+
+#define sdram_selfrefresh_disable(saved_lpr)	at91_sys_write(AT91_SDRAMC_LPR, saved_lpr)
+
+#endif
Signature · 1 line
Nicolas Ferre
#7 lore raw

PATCH v2 1/1 Add cpuidle support for at91

[Moving the discussion to l-a-k at lists.infradead.org]

Salut Nicolas !

Thanks a lot for your answer.

Quoted 2 lines · Nicolas Ferre

Works very well and allow to save some power. After quick measures, we run at 172mA at 12V instead of 190mA at 12V so ~2.06W and a saving of ~10% :

... and you haven't tried cpufreq yet :)

Quoted 5 lines

nice job !
This is of course in my test conditions.

Just one precision: what is the difference, entering state0 with only the current cpu_do_idle() that is activated by default ?

Sorry, I don't get what you mean. Would you mind elaborating a bit?

Regards,

Signature · 11 lines
Albin Tonnerre, Free Electrons
Kernel, drivers and embedded Linux development,
consulting, training and support.
http://free-electrons.com
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 836 bytes
Desc: Digital signature
URL: <http://lists.infradead.org/pipermail/linux-arm-kernel/attachments/20090930/35313a82/attachment.sig>
#8 lore raw

PATCH v2 1/1 Add cpuidle support for at91

Albin Tonnerre :

Quoted 4 lines
Quoted 2 lines

Just one precision: what is the difference, entering state0 with only the current cpu_do_idle() that is activated by default ?

Sorry, I don't get what you mean. Would you mind elaborating a bit?

I guess that during idle time, even without the cpuidle infrastructure, the SOC enters arch_idle() that calls cpu_do_idle().

So, now that I have your patch applied, I wonder what is the difference between the old situation and the first state of the cpuidle table.

In other words, only state "wait-for-interrupt *and* RAM self refresh" brings some more power saving. Indeed, cpu_do_idle() that correspond to the "WFI" state was already the way of dealing with idle cpu, even without your patch. Am I correct ?

Best regards,

Signature · 1 line
Nicolas Ferre
#9 lore raw

PATCH v2 1/1 Add cpuidle support for at91

Quoted 5 lines

In other words, only state "wait-for-interrupt *and* RAM self refresh" brings some more power saving. Indeed, cpu_do_idle() that correspond to the "WFI" state was already the way of dealing with idle cpu, even without your patch. Am I correct ?

Yes, but it has a higher exit latency. Those data is provided to the cpuidle framework, so it can decide which state to use.

I wonder if other people can confirm a similar amount of power savings? I am trying to implement the RAM self refresh for i.MX and don't see much improvment. I guess I still do something wrong...

Regards,

Wolfram

Signature · 9 lines
Pengutronix e.K.                           | Wolfram Sang                |
Industrial Linux Solutions                 | http://www.pengutronix.de/  |
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 197 bytes
Desc: Digital signature
URL: <http://lists.infradead.org/pipermail/linux-arm-kernel/attachments/20091002/4f2c5efa/attachment.sig>
#10 lore raw

PATCH v2 1/1 Add cpuidle support for at91

In reply to Nicolas Ferre
Quoted 9 lines · Nicolas Ferre

Albin Tonnerre :

Quoted 4 lines
Quoted 2 lines

Just one precision: what is the difference, entering state0 with only the current cpu_do_idle() that is activated by default ?

Sorry, I don't get what you mean. Would you mind elaborating a bit?

I guess that during idle time, even without the cpuidle infrastructure, the SOC enters arch_idle() that calls cpu_do_idle().

Quoted 7 lines

So, now that I have your patch applied, I wonder what is the difference between the old situation and the first state of the cpuidle table.

In other words, only state "wait-for-interrupt *and* RAM self refresh" brings some more power saving. Indeed, cpu_do_idle() that correspond to the "WFI" state was already the way of dealing with idle cpu, even without your patch. Am I correct ?

Indeed, only the second mode brings us some gain.

Regards,

Signature · 11 lines
Albin Tonnerre, Free Electrons
Kernel, drivers and embedded Linux development,
consulting, training and support.
http://free-electrons.com
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 835 bytes
Desc: Digital signature
URL: <http://lists.infradead.org/pipermail/linux-arm-kernel/attachments/20091009/20844aa2/attachment.sig>
#11 lore raw

PATCH v2 1/1 Add cpuidle support for at91

In reply to Nicolas Ferre

[Sending it to the right list this time, sorry for the noise]

Quoted 11 lines · Nicolas Ferre
Quoted 8 lines · Albin Tonnerre
Quoted 7 lines

This patch adds the support for cpuidle on AT91 SoCs, taken from the cpuidle support in mach-kirkwood.
cpuidle needs sdram_selfrefresh_enable and _disable, so move their definition to a separate header file instead of duplicating the code already used in pm.c.

[nicolas.ferre at atmel.com: tested on at91sam9263ek]

Andrew: what's your opinion on this? If it's OK, I'd really like to see it pushed forward.

Regards,

Signature · 4 lines
Albin Tonnerre, Free Electrons
Kernel, drivers and embedded Linux development,
consulting, training and support.
http://free-electrons.com