Skip to contentSkip to conversation

OMAP clock and clockdomain fixes for 2.6.29-rc5

PATCH In review Paul Walmsley 2 participants 5 messages Oct 14, 2009 – Oct 15, 2009

Subsystems (4)
lore
Share & tools
lore

Paul Walmsley

author maintainer
#0 lore raw

cover OMAP clock and clockdomain fixes for 2.6.29-rc5

(resending due to old l-a-k host address)

Hello,

This series contains:

- one OMAP2 clock fix that allows N800 to boot further (there are unrelated problems with both the current linux-omap head and omap-fixes branch that cause booting to fail due to other reasons), and

- one OMAP2/3 clockdomain fix that can cause clockdomain transitions to be missed by the clockdomain code -- this mostly affects OMAP3 OFF mode.

These patches are also available as a git branch (based on Linus' v2.6.32-rc4 tag) from

git://git.pwsan.com/linux-2.6 2_6_32rc4_fixes

- Paul

---

Kalle Jokiniemi (1):
OMAP: Fix race condition with autodeps

Paul Walmsley (1):
OMAP2xxx clock: set up clockdomain pointer in struct clk

 arch/arm/mach-omap2/clock24xx.c   |    1 +
 arch/arm/mach-omap2/clockdomain.c |   74 ++++++++++++++++++++++---------------
 2 files changed, 45 insertions(+), 30 deletions(-)

size:
text data bss dec hex filename
3588474 198624 105152 3892250 3b641a vmlinux.omap3beagle.orig
3588562 198624 105152 3892338 3b6472 vmlinux.omap3beagle.patched

author maintainer
#1 lore raw

PATCH 1/2 OMAP: Fix race condition with autodeps

From: Kalle Jokiniemi <[email protected]>

There is a possible race condition in clockdomain code handling hw supported idle transitions.

When multiple autodeps dependencies are being added or removed, a transition of still remaining dependent powerdomain can result in false readings of the state counter. This is especially fatal for off mode state counter, as it could result in a driver not noticing a context loss.

Fixed by disabling hw supported state transitions when autodeps are being changed.

---
 arch/arm/mach-omap2/clockdomain.c |   74 ++++++++++++++++++++++---------------
 1 files changed, 44 insertions(+), 30 deletions(-)
arch/arm/mach-omap2/clockdomain.c +44 -30
index 4ef7b4f..58aff84 100644
@@ -137,6 +137,36 @@ static void _clkdm_del_autodeps(struct clockdomain *clkdm)
 	}
 }
 
+/*
+ * _omap2_clkdm_set_hwsup - set the hwsup idle transition bit
+ * @clkdm: struct clockdomain *
+ * @enable: int 0 to disable, 1 to enable
+ *
+ * Internal helper for actually switching the bit that controls hwsup
+ * idle transitions for clkdm.
+ */
+static void _omap2_clkdm_set_hwsup(struct clockdomain *clkdm, int enable)
+{
+	u32 v;
+
+	if (cpu_is_omap24xx()) {
+		if (enable)
+			v = OMAP24XX_CLKSTCTRL_ENABLE_AUTO;
+		else
+			v = OMAP24XX_CLKSTCTRL_DISABLE_AUTO;
+	} else if (cpu_is_omap34xx()) {
+		if (enable)
+			v = OMAP34XX_CLKSTCTRL_ENABLE_AUTO;
+		else
+			v = OMAP34XX_CLKSTCTRL_DISABLE_AUTO;
+	} else {
+		BUG();
+	}
+
+	cm_rmw_mod_reg_bits(clkdm->clktrctrl_mask,
+			    v << __ffs(clkdm->clktrctrl_mask),
+			    clkdm->pwrdm.ptr->prcm_offs, CM_CLKSTCTRL);
+}
 
 static struct clockdomain *_clkdm_lookup(const char *name)
 {
@@ -456,8 +486,6 @@ int omap2_clkdm_wakeup(struct clockdomain *clkdm)
  */
 void omap2_clkdm_allow_idle(struct clockdomain *clkdm)
 {
-	u32 v;
-
 	if (!clkdm)
 		return;
 
@@ -473,18 +501,7 @@ void omap2_clkdm_allow_idle(struct clockdomain *clkdm)
 	if (atomic_read(&clkdm->usecount) > 0)
 		_clkdm_add_autodeps(clkdm);
 
-	if (cpu_is_omap24xx())
-		v = OMAP24XX_CLKSTCTRL_ENABLE_AUTO;
-	else if (cpu_is_omap34xx())
-		v = OMAP34XX_CLKSTCTRL_ENABLE_AUTO;
-	else
-		BUG();
-
-
-	cm_rmw_mod_reg_bits(clkdm->clktrctrl_mask,
-			    v << __ffs(clkdm->clktrctrl_mask),
-			    clkdm->pwrdm.ptr->prcm_offs,
-			    CM_CLKSTCTRL);
+	_omap2_clkdm_set_hwsup(clkdm, 1);
 
 	pwrdm_clkdm_state_switch(clkdm);
 }
@@ -500,8 +517,6 @@ void omap2_clkdm_allow_idle(struct clockdomain *clkdm)
  */
 void omap2_clkdm_deny_idle(struct clockdomain *clkdm)
 {
-	u32 v;
-
 	if (!clkdm)
 		return;
 
@@ -514,16 +529,7 @@ void omap2_clkdm_deny_idle(struct clockdomain *clkdm)
 	pr_debug("clockdomain: disabling automatic idle transitions for %s\n",
 		 clkdm->name);
 
-	if (cpu_is_omap24xx())
-		v = OMAP24XX_CLKSTCTRL_DISABLE_AUTO;
-	else if (cpu_is_omap34xx())
-		v = OMAP34XX_CLKSTCTRL_DISABLE_AUTO;
-	else
-		BUG();
-
-	cm_rmw_mod_reg_bits(clkdm->clktrctrl_mask,
-			    v << __ffs(clkdm->clktrctrl_mask),
-			    clkdm->pwrdm.ptr->prcm_offs, CM_CLKSTCTRL);
+	_omap2_clkdm_set_hwsup(clkdm, 0);
 
 	if (atomic_read(&clkdm->usecount) > 0)
 		_clkdm_del_autodeps(clkdm);
@@ -569,10 +575,14 @@ int omap2_clkdm_clk_enable(struct clockdomain *clkdm, struct clk *clk)
 	v = omap2_clkdm_clktrctrl_read(clkdm);
 
 	if ((cpu_is_omap34xx() && v == OMAP34XX_CLKSTCTRL_ENABLE_AUTO) ||
-	    (cpu_is_omap24xx() && v == OMAP24XX_CLKSTCTRL_ENABLE_AUTO))
+	    (cpu_is_omap24xx() && v == OMAP24XX_CLKSTCTRL_ENABLE_AUTO)) {
+		/* Disable HW transitions when we are changing deps */
+		_omap2_clkdm_set_hwsup(clkdm, 0);
 		_clkdm_add_autodeps(clkdm);
-	else
+		_omap2_clkdm_set_hwsup(clkdm, 1);
+	} else {
 		omap2_clkdm_wakeup(clkdm);
+	}
 
 	pwrdm_wait_transition(clkdm->pwrdm.ptr);
 	pwrdm_clkdm_state_switch(clkdm);
@@ -623,10 +633,14 @@ int omap2_clkdm_clk_disable(struct clockdomain *clkdm, struct clk *clk)
 	v = omap2_clkdm_clktrctrl_read(clkdm);
 
 	if ((cpu_is_omap34xx() && v == OMAP34XX_CLKSTCTRL_ENABLE_AUTO) ||
-	    (cpu_is_omap24xx() && v == OMAP24XX_CLKSTCTRL_ENABLE_AUTO))
+	    (cpu_is_omap24xx() && v == OMAP24XX_CLKSTCTRL_ENABLE_AUTO)) {
+		/* Disable HW transitions when we are changing deps */
+		_omap2_clkdm_set_hwsup(clkdm, 0);
 		_clkdm_del_autodeps(clkdm);
-	else
+		_omap2_clkdm_set_hwsup(clkdm, 1);
+	} else {
 		omap2_clkdm_sleep(clkdm);
+	}
 
 	pwrdm_clkdm_state_switch(clkdm);
 
author maintainer
#2 lore raw

PATCH 2/2 OMAP2xxx clock: set up clockdomain pointer in struct clk

In reply to Paul Walmsley

clock24xx.c is missing a omap2_init_clk_clkdm() in its omap2_clk_init() function. Among other bad effects, this causes the OMAP hwmod layer to oops on boot.

Thanks to Carlos Aguiar <[email protected]> and Stefano Panella <[email protected]> for reporting this bug. Thanks to Tony Lindgren <[email protected]> for N800 booting advice.

Cc Carlos Aguiar <[email protected]>
Cc Stefano Panella <[email protected]>
Cc Tony Lindgren <[email protected]>
---
 arch/arm/mach-omap2/clock24xx.c |    1 +
 1 files changed, 1 insertions(+), 0 deletions(-)
arch/arm/mach-omap2/clock24xx.c +1 -0
index bc5d3ac..e2dbedd 100644
@@ -769,6 +769,7 @@ int __init omap2_clk_init(void)
 		if (c->cpu & cpu_mask) {
 			clkdev_add(&c->lk);
 			clk_register(c->lk.clk);
+			omap2_init_clk_clkdm(c->lk.clk);
 		}
 
 	/* Check the MPU rate set by bootloader */
author maintainer
#3 lore raw

cover OMAP clock and clockdomain fixes for 2.6.29-rc5

In reply to Paul Walmsley

Time for a subject line correction of my own: Kevin kindly pointed out that the subject should read "2.6.32-rc5"

- Paul

maintainer
#4 lore raw
In reply to Paul Walmsley
Quoted 8 lines · Paul Walmsley

Hello,

This series contains:

- one OMAP2 clock fix that allows N800 to boot further (there are unrelated problems with both the current linux-omap head and omap-fixes branch that cause booting to fail due to other reasons), and

Quoted 7 lines

- one OMAP2/3 clockdomain fix that can cause clockdomain transitions to be missed by the clockdomain code -- this mostly affects OMAP3 OFF mode.

These patches are also available as a git branch (based on Tony's omap-fixes branch) from

git://git.pwsan.com/linux-2.6 2_6_32rc4_fixes

I've pulled these on top of the fixes I posted yesterday.

Regards,

Tony

Quoted 25 lines

- Paul

---

Kalle Jokiniemi (1):
OMAP: Fix race condition with autodeps

Paul Walmsley (1):
OMAP2xxx clock: set up clockdomain pointer in struct clk

 arch/arm/mach-omap2/clock24xx.c   |    1 +
 arch/arm/mach-omap2/clockdomain.c |   74 ++++++++++++++++++++++---------------
 2 files changed, 45 insertions(+), 30 deletions(-)

size:
text data bss dec hex filename
3588474 198624 105152 3892250 3b641a vmlinux.omap3beagle.orig
3588562 198624 105152 3892338 3b6472 vmlinux.omap3beagle.patched

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