ARM: OMAP AM3517/05: hwmod data: block WFI when EMAC active

According to Mark Greer, on OMAP AM3517/3505 chips, the EMAC is unable to wake the ARM up from WFI:

    http://www.spinics.net/lists/arm-kernel/msg174734.html

Further troubleshooting was unable to narrow the problem down. So we don't have much choice other than to block WFI when the EMAC is active with the HWMOD_BLOCK_WFI flag. However, for some reason, we don't have an EMAC hwmod -- probably some bug in the data -- so set the flag on the MDIO hwmod data instead.

Based on Mark's original patch. We're removing the omap_device-based pm_lats code, so a different approach was needed.

---

Applies after the WFI cleanup series -- intended for a post-cleanup fixes pull request or a new feature pull request.

 arch/arm/mach-omap2/omap_hwmod_3xxx_data.c |    8 +++++++-
 1 file changed, 7 insertions(+), 1 deletion(-)

index 8bb2628..4a68735 100644
@@ -3439,7 +3439,13 @@ static struct omap_hwmod_class am35xx_mdio_class = {
 static struct omap_hwmod am35xx_mdio_hwmod = {
 	.name		= "davinci_mdio",
 	.class		= &am35xx_mdio_class,
-	.flags		= HWMOD_NO_IDLEST,
+	/*
+	 * According to Mark Greer, the MPU will not return from WFI
+	 * when the EMAC signals an interrupt.  We're missing an EMAC
+	 * hwmod for some reason, so add the flag to the MDIO instead.
+	 * http://www.spinics.net/lists/arm-kernel/msg174734.html
+	 */
+	.flags		= (HWMOD_NO_IDLEST | HWMOD_BLOCK_WFI),
 };
 
 /*

1.7.10.4

Mark and I discussed this in private E-mail. Looks like I managed to confuse AM33xx and AM3517 :-( Here's the updated patch.

- Paul

From: Paul Walmsley <[email protected]>
Date: Sun, 30 Dec 2012 10:36:36 -0700
Subject: [PATCH] ARM: OMAP AM3517/05: hwmod data: block WFI when EMAC active

According to Mark Greer, on OMAP AM3517/3505 chips, the EMAC is unable to wake the ARM up from WFI:

    http://www.spinics.net/lists/arm-kernel/msg174734.html

Further troubleshooting was unable to narrow the problem down. So we don't have much choice other than to block WFI when the EMAC is active with the HWMOD_BLOCK_WFI flag.

Based on Mark's original patch. We're removing the omap_device-based pm_lats code, so a different approach was needed.

This second version contains some corrections thanks to Mark's review.

---
 arch/arm/mach-omap2/omap_hwmod_3xxx_data.c |    8 +++++++-
 1 file changed, 7 insertions(+), 1 deletion(-)

index 8bb2628..7af28b7 100644
@@ -3493,7 +3493,13 @@ static struct omap_hwmod am35xx_emac_hwmod = {
 	.name		= "davinci_emac",
 	.mpu_irqs	= am35xx_emac_mpu_irqs,
 	.class		= &am35xx_emac_class,
-	.flags		= HWMOD_NO_IDLEST,
+	/*
+	 * According to Mark Greer, the MPU will not return from WFI
+	 * when the EMAC signals an interrupt.  We're missing an EMAC
+	 * hwmod for some reason, so add the flag to the MDIO instead.
+	 * http://www.spinics.net/lists/arm-kernel/msg174734.html
+	 */
+	.flags		= (HWMOD_NO_IDLEST | HWMOD_BLOCK_WFI),
 };
 
 /* l3_core -> davinci emac interface */

1.7.10.4

Hi Paul.

Hi Mark

You are absolutely right, and the correction is very much appreciated. Updated patch follows.

- Paul

From: Paul Walmsley <[email protected]>
Date: Sun, 30 Dec 2012 10:36:36 -0700
Subject: [PATCH] ARM: OMAP AM3517/05: hwmod data: block WFI when EMAC active

According to Mark Greer, on OMAP AM3517/3505 chips, the EMAC is unable to wake the ARM up from WFI:

    http://www.spinics.net/lists/arm-kernel/msg174734.html

Further troubleshooting was unable to narrow the problem down. So we don't have much choice other than to block WFI when the EMAC is active with the HWMOD_BLOCK_WFI flag.

Based on Mark's original patch. We're removing the omap_device-based pm_lats code, so a different approach was needed.

This third version contains some corrections thanks to Mark's review.

---
 arch/arm/mach-omap2/omap_hwmod_3xxx_data.c |    7 ++++++-
 1 file changed, 6 insertions(+), 1 deletion(-)

index 8bb2628..ac7e03e 100644
@@ -3493,7 +3493,12 @@ static struct omap_hwmod am35xx_emac_hwmod = {
 	.name		= "davinci_emac",
 	.mpu_irqs	= am35xx_emac_mpu_irqs,
 	.class		= &am35xx_emac_class,
-	.flags		= HWMOD_NO_IDLEST,
+	/*
+	 * According to Mark Greer, the MPU will not return from WFI
+	 * when the EMAC signals an interrupt.
+	 * http://www.spinics.net/lists/arm-kernel/msg174734.html
+	 */
+	.flags		= (HWMOD_NO_IDLEST | HWMOD_BLOCK_WFI),
 };
 
 /* l3_core -> davinci emac interface */

1.7.10.4

FWIW,