Skip to contentSkip to conversation

Revert "ARM: OMAP3: PM: call pre/post transition per powerdomain"

PATCH Superseded Kevin Hilman 2 participants 3 messages Aug 7, 2012 – Aug 8, 2012

Subsystems (4)
lore
Share & tools
lore
#0 lore raw

PATCH v2 1/1 Revert "ARM: OMAP3: PM: call pre/post transition per powerdomain"

This reverts commit 58f0829b7186150318c79515f0e0850c5e7a9c89.

Converstion to per-pwrdm per/post transition calls was a bit premature. Only tracking MPU, PER & CORE in the idle path means we lose the accounting for all the other powerdomains which may also transition in idle. On OMAP3, due to autodeps, several powerdomains transition along with MPU (e.g. DSS, USBHOST), and the accounting for these was lost with this patch. Since the accounting includes the context loss counters, drivers for devices in those power domains would never notice context lost, so would likely hang after any off-mode transitions.

This patch should be revisited when the upcoming clkdm/pwrmdm/voltdm use-counting seires is merged since then we can properly do accounting without relying on a call in the idle path.

In addition, the original patch had another bug because the PER powerdomain accounting was not updated until after the GPIO resume hook is called. Since gpio_resume_after_idle() checks the context loss count (which is not yet updated) it would not properly restore context, leaving the GPIO banks in an undefined state.

---
 arch/arm/mach-omap2/pm34xx.c |   19 ++++---------------
 1 file changed, 4 insertions(+), 15 deletions(-)
arch/arm/mach-omap2/pm34xx.c +4 -15
index e4fc88c..05bd8f0 100644
@@ -272,21 +272,16 @@ void omap_sram_idle(void)
 	per_next_state = pwrdm_read_next_pwrst(per_pwrdm);
 	core_next_state = pwrdm_read_next_pwrst(core_pwrdm);
 
-	if (mpu_next_state < PWRDM_POWER_ON) {
-		pwrdm_pre_transition(mpu_pwrdm);
-		pwrdm_pre_transition(neon_pwrdm);
-	}
+	pwrdm_pre_transition(NULL);
 
 	/* PER */
 	if (per_next_state < PWRDM_POWER_ON) {
-		pwrdm_pre_transition(per_pwrdm);
 		per_going_off = (per_next_state == PWRDM_POWER_OFF) ? 1 : 0;
 		omap2_gpio_prepare_for_idle(per_going_off);
 	}
 
 	/* CORE */
 	if (core_next_state < PWRDM_POWER_ON) {
-		pwrdm_pre_transition(core_pwrdm);
 		if (core_next_state == PWRDM_POWER_OFF) {
 			omap3_core_save_context();
 			omap3_cm_save_context();
@@ -339,20 +334,14 @@ void omap_sram_idle(void)
 			omap2_prm_clear_mod_reg_bits(OMAP3430_AUTO_OFF_MASK,
 					       OMAP3430_GR_MOD,
 					       OMAP3_PRM_VOLTCTRL_OFFSET);
-		pwrdm_post_transition(core_pwrdm);
 	}
 	omap3_intc_resume_idle();
 
+	pwrdm_post_transition(NULL);
+
 	/* PER */
-	if (per_next_state < PWRDM_POWER_ON) {
+	if (per_next_state < PWRDM_POWER_ON)
 		omap2_gpio_resume_after_idle();
-		pwrdm_post_transition(per_pwrdm);
-	}
-
-	if (mpu_next_state < PWRDM_POWER_ON) {
-		pwrdm_post_transition(mpu_pwrdm);
-		pwrdm_post_transition(neon_pwrdm);
-	}
 }
 
 static void omap3_pm_idle(void)
Signature · 1 line
1.7.9.2
#1 lore raw

PATCH v2 1/1 Revert "ARM: OMAP3: PM: call pre/post transition per powerdomain"

Quoted 11 lines · Kevin Hilman

This reverts commit 58f0829b7186150318c79515f0e0850c5e7a9c89.

Converstion to per-pwrdm per/post transition calls was a bit premature. Only tracking MPU, PER & CORE in the idle path means we lose the accounting for all the other powerdomains which may also transition in idle. On OMAP3, due to autodeps, several powerdomains transition along with MPU (e.g. DSS, USBHOST), and the accounting for these was lost with this patch. Since the accounting includes the context loss counters, drivers for devices in those power domains would never notice context lost, so would likely hang after any off-mode transitions.

That's a shame, pwrdm_pre_transition/pwrdm_post_transition are the main contributors to idle latency and cause large performance loss on small and frequent loads, like short DMAs. Could we perhaps only do it when PM_DEBUG is on or when off transitions happen instead?

Quoted 3 lines

This patch should be revisited when the upcoming clkdm/pwrmdm/voltdm use-counting seires is merged since then we can properly do accounting without relying on a call in the idle path.

So all hope of getting rid of those pre/post transition calls goes here then? Small typo with 'seires'..

Quoted 73 lines

In addition, the original patch had another bug because the PER powerdomain accounting was not updated until after the GPIO resume hook is called. Since gpio_resume_after_idle() checks the context loss count (which is not yet updated) it would not properly restore context, leaving the GPIO banks in an undefined state.

---
 arch/arm/mach-omap2/pm34xx.c |   19 ++++---------------
 1 file changed, 4 insertions(+), 15 deletions(-)
arch/arm/mach-omap2/pm34xx.c +4 -16
index e4fc88c..05bd8f0 100644
@@ -272,21 +272,16 @@ void omap_sram_idle(void)
        per_next_state = pwrdm_read_next_pwrst(per_pwrdm);
        core_next_state = pwrdm_read_next_pwrst(core_pwrdm);
 
-       if (mpu_next_state < PWRDM_POWER_ON) {
-               pwrdm_pre_transition(mpu_pwrdm);
-               pwrdm_pre_transition(neon_pwrdm);
-       }
+       pwrdm_pre_transition(NULL);
 
        /* PER */
        if (per_next_state < PWRDM_POWER_ON) {
-               pwrdm_pre_transition(per_pwrdm);
                per_going_off = (per_next_state == PWRDM_POWER_OFF) ? 1 : 0;
                omap2_gpio_prepare_for_idle(per_going_off);
        }
 
        /* CORE */
        if (core_next_state < PWRDM_POWER_ON) {
-               pwrdm_pre_transition(core_pwrdm);
                if (core_next_state == PWRDM_POWER_OFF) {
                        omap3_core_save_context();
                        omap3_cm_save_context();
@@ -339,20 +334,14 @@ void omap_sram_idle(void)
                        omap2_prm_clear_mod_reg_bits(OMAP3430_AUTO_OFF_MASK,
                                               OMAP3430_GR_MOD,
                                               OMAP3_PRM_VOLTCTRL_OFFSET);
-               pwrdm_post_transition(core_pwrdm);
        }
        omap3_intc_resume_idle();
 
+       pwrdm_post_transition(NULL);
+
        /* PER */
-       if (per_next_state < PWRDM_POWER_ON) {
+       if (per_next_state < PWRDM_POWER_ON)
                omap2_gpio_resume_after_idle();
-               pwrdm_post_transition(per_pwrdm);
-       }
-
-       if (mpu_next_state < PWRDM_POWER_ON) {
-               pwrdm_post_transition(mpu_pwrdm);
-               pwrdm_post_transition(neon_pwrdm);
-       }
 }
 
 static void omap3_pm_idle(void)
--

1.7.9.2

--
To unsubscribe from this list: send the line "unsubscribe linux-omap" in the body of a message to majordomo at vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html

Signature · 1 line
Gra?vydas
#2 lore raw

PATCH v2 1/1 Revert "ARM: OMAP3: PM: call pre/post transition per powerdomain"

Quoted 13 lines · Grazvydas Ignotas
Quoted 11 lines · Kevin Hilman

This reverts commit 58f0829b7186150318c79515f0e0850c5e7a9c89.

Converstion to per-pwrdm per/post transition calls was a bit premature. Only tracking MPU, PER & CORE in the idle path means we lose the accounting for all the other powerdomains which may also transition in idle. On OMAP3, due to autodeps, several powerdomains transition along with MPU (e.g. DSS, USBHOST), and the accounting for these was lost with this patch. Since the accounting includes the context loss counters, drivers for devices in those power domains would never notice context lost, so would likely hang after any off-mode transitions.

Quoted 4 lines

That's a shame, pwrdm_pre_transition/pwrdm_post_transition are the main contributors to idle latency and cause large performance loss on small and frequent loads, like short DMAs. Could we perhaps only do it when PM_DEBUG is on or when off transitions happen instead?

Unfortunately, that's also where we're currently doing the accounting for when power domains lose context. Drivers rely on this to determine whether or not to restore context, so it's not a debug only feature.

Quoted 6 lines
Quoted 3 lines

This patch should be revisited when the upcoming clkdm/pwrmdm/voltdm use-counting seires is merged since then we can properly do accounting without relying on a call in the idle path.

So all hope of getting rid of those pre/post transition calls goes here then? Small typo with 'seires'..

Yes. The goal of the use-counting series is so that proper accounting can be done when the usecount changes so that not all of them have to be done in the idle path.

Kevin

Quoted 73 lines
Quoted 73 lines

In addition, the original patch had another bug because the PER powerdomain accounting was not updated until after the GPIO resume hook is called. Since gpio_resume_after_idle() checks the context loss count (which is not yet updated) it would not properly restore context, leaving the GPIO banks in an undefined state.

---
 arch/arm/mach-omap2/pm34xx.c |   19 ++++---------------
 1 file changed, 4 insertions(+), 15 deletions(-)
arch/arm/mach-omap2/pm34xx.c +4 -16
index e4fc88c..05bd8f0 100644
@@ -272,21 +272,16 @@ void omap_sram_idle(void)
        per_next_state = pwrdm_read_next_pwrst(per_pwrdm);
        core_next_state = pwrdm_read_next_pwrst(core_pwrdm);
 
-       if (mpu_next_state < PWRDM_POWER_ON) {
-               pwrdm_pre_transition(mpu_pwrdm);
-               pwrdm_pre_transition(neon_pwrdm);
-       }
+       pwrdm_pre_transition(NULL);
 
        /* PER */
        if (per_next_state < PWRDM_POWER_ON) {
-               pwrdm_pre_transition(per_pwrdm);
                per_going_off = (per_next_state == PWRDM_POWER_OFF) ? 1 : 0;
                omap2_gpio_prepare_for_idle(per_going_off);
        }
 
        /* CORE */
        if (core_next_state < PWRDM_POWER_ON) {
-               pwrdm_pre_transition(core_pwrdm);
                if (core_next_state == PWRDM_POWER_OFF) {
                        omap3_core_save_context();
                        omap3_cm_save_context();
@@ -339,20 +334,14 @@ void omap_sram_idle(void)
                        omap2_prm_clear_mod_reg_bits(OMAP3430_AUTO_OFF_MASK,
                                               OMAP3430_GR_MOD,
                                               OMAP3_PRM_VOLTCTRL_OFFSET);
-               pwrdm_post_transition(core_pwrdm);
        }
        omap3_intc_resume_idle();
 
+       pwrdm_post_transition(NULL);
+
        /* PER */
-       if (per_next_state < PWRDM_POWER_ON) {
+       if (per_next_state < PWRDM_POWER_ON)
                omap2_gpio_resume_after_idle();
-               pwrdm_post_transition(per_pwrdm);
-       }
-
-       if (mpu_next_state < PWRDM_POWER_ON) {
-               pwrdm_post_transition(mpu_pwrdm);
-               pwrdm_post_transition(neon_pwrdm);
-       }
 }
 
 static void omap3_pm_idle(void)
--

1.7.9.2

--
To unsubscribe from this list: send the line "unsubscribe linux-omap" in the body of a message to majordomo at vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html