usb: musb: Fix LapDock enumeration on omap for boot and slow cable insertion

Looks like we can get VBUS interrupt before the ID interrupt at least with a Motorola LapDock if the USB mini-A cable is inserted slowly at the omap end. This issue also prevents enumerating the LapDock during the boot.

It seems that the issue is caused by musb getting confused and trying to do things on it's own as the LapDock provides VBUS. Fix the issue by disconnecting the transceiver before setting the host mode if the previous state was VBUS, and check that musb is in a sane state with the session bit cleared before we attempt to start the session.

With this patch the LapDock is properly enumerated during the boot and on slow insertion after a cold boot. This means that the LapDock can be used for both power and keyboard with at least omap panda es.

However, there seems to be a separate issue with musb not resetting properly after a warm reset which also leads to devices not enumerating.

---

Can you guys please test that this does not cause regressions on other devices when connecting USB devices?

@@ -251,6 +251,20 @@ void omap_musb_mailbox(enum omap_musb_vbus_id_status status)
 }
 EXPORT_SYMBOL_GPL(omap_musb_mailbox);
 
+static void omap_musb_disconnect(struct musb *musb, struct omap2430_glue *glue)
+{
+	int loops = 100;
+
+	omap_control_usb_set_mode(glue->control_otghs,
+				  USB_MODE_DISCONNECT);
+	while (loops--) {
+		u8 devctl = musb_readb(musb->mregs, MUSB_DEVCTL);
+		if (!(devctl & MUSB_DEVCTL_SESSION))
+			break;
+		dev_dbg(dev, "disconnecting: %02x\n", devctl);
+	}
+}
+
 static void omap_musb_set_mailbox(struct omap2430_glue *glue)
 {
 	struct musb *musb = glue_to_musb(glue);

@@ -258,6 +272,7 @@ static void omap_musb_set_mailbox(struct omap2430_glue *glue)
 	struct musb_hdrc_platform_data *pdata = dev->platform_data;
 	struct omap_musb_board_data *data = pdata->board_data;
 	struct usb_otg *otg = musb->xceiv->otg;
+	int disconnect = 0;
 
 	switch (glue->status) {
 	case OMAP_MUSB_ID_GROUND:

@@ -265,9 +280,13 @@ static void omap_musb_set_mailbox(struct omap2430_glue *glue)
 
 		otg->default_a = true;
 		musb->xceiv->state = OTG_STATE_A_IDLE;
+		if (musb->xceiv->last_event == USB_EVENT_VBUS)
+			disconnect = 1;
 		musb->xceiv->last_event = USB_EVENT_ID;
 		if (musb->gadget_driver) {
 			pm_runtime_get_sync(dev);
+			if (disconnect)
+				omap_musb_disconnect(musb, glue);
 			omap_control_usb_set_mode(glue->control_otghs,
 				USB_MODE_HOST);
 			omap2430_musb_set_vbus(musb, 1);

Sorry forgot to stg refresh, one line was missing. Updated patch below.

Regards,

Tony

From: Tony Lindgren <[email protected]>
Date: Wed, 1 May 2013 14:28:46 -0700
Subject: [PATCH] usb: musb: Fix LapDock enumeration on omap for boot and slow cable insertion

Looks like we can get VBUS interrupt before the ID interrupt at least with a Motorola LapDock if the USB mini-A cable is inserted slowly at the omap end. This issue also prevents enumerating the LapDock during the boot.

It seems that the issue is caused by musb getting confused and trying to do things on it's own as the LapDock provides VBUS. Fix the issue by disconnecting the transceiver before setting the host mode if the previous state was VBUS, and check that musb is in a sane state with the session bit cleared before we attempt to start the session.

With this patch the LapDock is properly enumerated during the boot and on slow insertion after a cold boot. This means that the LapDock can be used for both power and keyboard with at least omap panda es.

However, there seems to be a separate issue with musb not resetting properly after a warm reset which also leads to devices not enumerating.

@@ -251,6 +251,21 @@ void omap_musb_mailbox(enum omap_musb_vbus_id_status status)
 }
 EXPORT_SYMBOL_GPL(omap_musb_mailbox);
 
+static void omap_musb_disconnect(struct musb *musb, struct omap2430_glue *glue)
+{
+	struct device *dev = musb->controller;
+	int loops = 100;
+
+	omap_control_usb_set_mode(glue->control_otghs,
+				  USB_MODE_DISCONNECT);
+	while (loops--) {
+		u8 devctl = musb_readb(musb->mregs, MUSB_DEVCTL);
+		if (!(devctl & MUSB_DEVCTL_SESSION))
+			break;
+		dev_dbg(dev, "disconnecting: %02x\n", devctl);
+	}
+}
+
 static void omap_musb_set_mailbox(struct omap2430_glue *glue)
 {
 	struct musb *musb = glue_to_musb(glue);

@@ -258,6 +273,7 @@ static void omap_musb_set_mailbox(struct omap2430_glue *glue)
 	struct musb_hdrc_platform_data *pdata = dev->platform_data;
 	struct omap_musb_board_data *data = pdata->board_data;
 	struct usb_otg *otg = musb->xceiv->otg;
+	int disconnect = 0;
 
 	switch (glue->status) {
 	case OMAP_MUSB_ID_GROUND:

@@ -265,9 +281,13 @@ static void omap_musb_set_mailbox(struct omap2430_glue *glue)
 
 		otg->default_a = true;
 		musb->xceiv->state = OTG_STATE_A_IDLE;
+		if (musb->xceiv->last_event == USB_EVENT_VBUS)
+			disconnect = 1;
 		musb->xceiv->last_event = USB_EVENT_ID;
 		if (musb->gadget_driver) {
 			pm_runtime_get_sync(dev);
+			if (disconnect)
+				omap_musb_disconnect(musb, glue);
 			omap_control_usb_set_mode(glue->control_otghs,
 				USB_MODE_HOST);
 			omap2430_musb_set_vbus(musb, 1);

how can this happen ? VBUS interrupt happens when you connect to a port which is sourcing VBUS to you, while ID interrupt happens when ID is grounded, meaning that you should be sourcing VBUS.

Have you hacked a Hub to backfeed 5V to OMAP by any chance ?

balbi

Yes, in this case we get both interrupts and the order depends on how fast/slow the cable is inserted.

..as that's how the LapDock seems to behave backfeeding 5V.

It would be interesting to take a look at the signaling on it, but I think my old beagle sniffer is fried.

Looking at the "Figure 6-1: Common State Diagram" on page 32 in "USB_OTG_and_EH_3-0_release_1_1_10May2012.pdf" the logic is the following depending on the order of interrupt:

start -> id ground -> a_idle -> a_wait_vrise -> a_wait_bcon...

or

start -> vbus -> b_idle -> id ground -> a_idle -> a_wait_vrise -> a_wait_bcon...

I don't think having the VBUS there actually violates that if we just should follow the ID state and accept that a_wait_vrise is already satisfied and not even try to turn the VBUS at MUSB end in that case.

Regards,

Tony