Skip to contentSkip to conversation

musb/dsps: suspend related patches for 3.14 (rebased)

PATCH Superseded Daniel Mack 6 participants 31 messages Nov 25, 2013 – Nov 26, 2013 11 messages not in the archive

Subsystems (2)

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore
#0 lore raw

cover musb/dsps: suspend related patches for 3.14 (rebased)

Hi Felipe,

These five patches for musb/dsps didn't make it into 3.13, which is no big deal, at least as far as I'm concerned :)

I rebased the missing bits on top of 3.13-rc1. There are no functional changes in comparison to what was posted earlier on this list.

Successfully tested on a AM33xx board.

Thanks,
Daniel

Daniel Mack (5):
usb: musb: conditionally save and restore the context on suspend
usb: musb: call musb_port_suspend from musb_bus_suspend
usb: musb: dsps: add {tx,rx}_mode to wrapper
usb: musb: dsps: add support for suspend and resume
usb: musb: dsps: indentation and whitespace fixes

 drivers/usb/musb/musb_core.c    | 17 +++++++-
 drivers/usb/musb/musb_core.h    |  1 +
 drivers/usb/musb/musb_dsps.c    | 86 +++++++++++++++++++++++++++++++++++------
 drivers/usb/musb/musb_host.c    |  2 +
 drivers/usb/musb/musb_host.h    |  4 ++
 drivers/usb/musb/musb_virthub.c |  4 +-
 include/linux/usb/musb.h        |  3 ++
 7 files changed, 102 insertions(+), 15 deletions(-)
Signature · 6 lines
1.8.4.2

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

PATCH 3/5 usb: musb: dsps: add {tx,rx}_mode to wrapper

In reply to a message not in the archive

rx_mode and tx_mode need to be read at suspend time and restored on resume for dsps platforms. So add it to the wrapper struct first, and initialize the values.

---
 drivers/usb/musb/musb_dsps.c | 4 ++++
 1 file changed, 4 insertions(+)
drivers/usb/musb/musb_dsps.c +4 -0
index 1901f6f..e57d712 100644
@@ -83,6 +83,8 @@ struct dsps_musb_wrapper {
 	u16	coreintr_status;
 	u16	phy_utmi;
 	u16	mode;
+	u16	tx_mode;
+	u16	rx_mode;
 
 	/* bit positions for control */
 	unsigned	reset:5;
@@ -605,6 +607,8 @@ static const struct dsps_musb_wrapper am33xx_driver_data = {
 	.coreintr_status	= 0x34,
 	.phy_utmi		= 0xe0,
 	.mode			= 0xe8,
+	.tx_mode		= 0x70,
+	.rx_mode		= 0x74,
 	.reset			= 0,
 	.otg_disable		= 21,
 	.iddig			= 8,
Signature · 6 lines
1.8.4.2

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

PATCH 1/5 usb: musb: conditionally save and restore the context on suspend

In reply to Daniel Mack

It appears not all platforms featuring a musb core need to save the musb core registers at suspend time and restore them on resume.

The dsps platform does, however. So add a bit in struct musb_hdrc_platform_data to let platforms specify their need of such action being taken.

---
 drivers/usb/musb/musb_core.c | 17 ++++++++++++++++-
 include/linux/usb/musb.h     |  3 +++
 2 files changed, 19 insertions(+), 1 deletion(-)
drivers/usb/musb/musb_core.c +16 -1
index 0a43329..a8ded57 100644
@@ -2202,6 +2202,7 @@ static int musb_suspend(struct device *dev)
 {
 	struct musb	*musb = dev_to_musb(dev);
 	unsigned long	flags;
+	struct musb_hdrc_platform_data *plat = dev_get_platdata(dev);
 
 	spin_lock_irqsave(&musb->lock, flags);
 
@@ -2215,16 +2216,30 @@ static int musb_suspend(struct device *dev)
 		 */
 	}
 
+	if (plat->restore_after_suspend)
+		musb_save_context(musb);
+
 	spin_unlock_irqrestore(&musb->lock, flags);
 	return 0;
 }
 
 static int musb_resume_noirq(struct device *dev)
 {
-	/* for static cmos like DaVinci, register values were preserved
+	struct musb	*musb = dev_to_musb(dev);
+	struct musb_hdrc_platform_data *plat = dev_get_platdata(dev);
+
+	/*
+	 * For static cmos like DaVinci, register values were preserved
 	 * unless for some reason the whole soc powered down or the USB
 	 * module got reset through the PSC (vs just being disabled).
+	 *
+	 * The plaform data tells us about the necessity of saving and
+	 * restoring the context across a suspend cycle.
 	 */
+
+	if (plat->restore_after_suspend)
+		musb_restore_context(musb);
+
 	return 0;
 }
 
include/linux/usb/musb.h +3 -0
index eb50525..e5a581c 100644
@@ -99,6 +99,9 @@ struct musb_hdrc_platform_data {
 	/* MUSB_HOST, MUSB_PERIPHERAL, or MUSB_OTG */
 	u8		mode;
 
+	/* should the musb core restore registers after suspend? */
+	u8		restore_after_suspend:1;
+
 	/* for clk_get() */
 	const char	*clock;
 
Signature · 1 line
1.8.4.2
#11 lore raw

PATCH 2/5 usb: musb: call musb_port_suspend from musb_bus_suspend

In reply to Daniel Mack

Make musb_port_suspend() externally available, and call it when to host goes into suspend. This allows the core to go into suspend while a device is connected.

---
 drivers/usb/musb/musb_host.c    | 2 ++
 drivers/usb/musb/musb_host.h    | 2 ++
 drivers/usb/musb/musb_virthub.c | 2 +-
 3 files changed, 5 insertions(+), 1 deletion(-)
drivers/usb/musb/musb_host.c +2 -0
index 6582a20..81caf9f 100644
@@ -2433,6 +2433,8 @@ static int musb_bus_suspend(struct usb_hcd *hcd)
 	struct musb	*musb = hcd_to_musb(hcd);
 	u8		devctl;
 
+	musb_port_suspend(musb, true);
+
 	if (!is_host_active(musb))
 		return 0;
 
drivers/usb/musb/musb_host.h +2 -0
index 960d735..e660af9 100644
@@ -92,6 +92,7 @@ extern void musb_host_rx(struct musb *, u8);
 extern void musb_root_disconnect(struct musb *musb);
 extern void musb_host_resume_root_hub(struct musb *musb);
 extern void musb_host_poke_root_hub(struct musb *musb);
+extern void musb_port_suspend(struct musb *musb, bool do_suspend);
 #else
 static inline struct musb *hcd_to_musb(struct usb_hcd *hcd)
 {
@@ -121,6 +122,7 @@ static inline void musb_root_disconnect(struct musb *musb)	{}
 static inline void musb_host_resume_root_hub(struct musb *musb)	{}
 static inline void musb_host_poll_rh_status(struct musb *musb)	{}
 static inline void musb_host_poke_root_hub(struct musb *musb)	{}
+static inline void musb_port_suspend(struct musb *musb, bool do_suspend) {}
 #endif
 
 struct usb_hcd;
drivers/usb/musb/musb_virthub.c +1 -1
index 9af6bba..e977441 100644
@@ -44,7 +44,7 @@
 
 #include "musb_core.h"
 
-static void musb_port_suspend(struct musb *musb, bool do_suspend)
+void musb_port_suspend(struct musb *musb, bool do_suspend)
 {
 	struct usb_otg	*otg = musb->xceiv->otg;
 	u8		power;
Signature · 1 line
1.8.4.2
#28 lore raw

PATCH 4/5 usb: musb: dsps: add support for suspend and resume

In reply to Daniel Mack

The dsps platform needs to save save some registers at suspend time and restore them after resume. This patch adds a struct for these registers, and also lets the musb core know that the core registers need to be saved as well.

We also have to call musb_port_reset() for this platform upon resume, so this function has to be made non-static.

---
 drivers/usb/musb/musb_core.h    |  1 +
 drivers/usb/musb/musb_dsps.c    | 59 +++++++++++++++++++++++++++++++++++++++++
 drivers/usb/musb/musb_host.h    |  2 ++
 drivers/usb/musb/musb_virthub.c |  2 +-
 4 files changed, 63 insertions(+), 1 deletion(-)
drivers/usb/musb/musb_core.h +1 -0
index 29f7cd7..a423037 100644
@@ -295,6 +295,7 @@ struct musb {
 
 	irqreturn_t		(*isr)(int, void *);
 	struct work_struct	irq_work;
+
 	u16			hwvers;
 
 	u16			intrrxe;
drivers/usb/musb/musb_dsps.c +59 -0

88 lines not on this page · raw message

drivers/usb/musb/musb_host.h +2 -0
index e660af9..7436c24 100644
@@ -93,6 +93,7 @@ extern void musb_root_disconnect(struct musb *musb);
 extern void musb_host_resume_root_hub(struct musb *musb);
 extern void musb_host_poke_root_hub(struct musb *musb);
 extern void musb_port_suspend(struct musb *musb, bool do_suspend);
+extern void musb_port_reset(struct musb *musb, bool do_reset);
 #else
 static inline struct musb *hcd_to_musb(struct usb_hcd *hcd)
 {
@@ -123,6 +124,7 @@ static inline void musb_host_resume_root_hub(struct musb *musb)	{}
 static inline void musb_host_poll_rh_status(struct musb *musb)	{}
 static inline void musb_host_poke_root_hub(struct musb *musb)	{}
 static inline void musb_port_suspend(struct musb *musb, bool do_suspend) {}
+static inline void musb_port_reset(struct musb *musb)		{}
 #endif
 
 struct usb_hcd;
drivers/usb/musb/musb_virthub.c +1 -1
index e977441..24e46c0 100644
@@ -109,7 +109,7 @@ void musb_port_suspend(struct musb *musb, bool do_suspend)
 	}
 }
 
-static void musb_port_reset(struct musb *musb, bool do_reset)
+void musb_port_reset(struct musb *musb, bool do_reset)
 {
 	u8		power;
 	void __iomem	*mbase = musb->mregs;
Signature · 1 line
1.8.4.2
#38 lore raw

PATCH 5/5 usb: musb: dsps: indentation and whitespace fixes

In reply to Daniel Mack

Just a cosmetic thing, no functional change.

---
 drivers/usb/musb/musb_dsps.c | 23 +++++++++++------------
 1 file changed, 11 insertions(+), 12 deletions(-)
drivers/usb/musb/musb_dsps.c +11 -12
index 361ddf8..b27c072 100644
@@ -131,9 +131,9 @@ struct dsps_context {
 struct dsps_glue {
 	struct device *dev;
 	struct platform_device *musb;	/* child musb pdev */
-	const struct dsps_musb_wrapper *wrp; /* wrapper register offsets */
+	const struct dsps_musb_wrapper *wrp;	/* wrapper register offsets */
 	struct timer_list timer;	/* otg_workaround timer */
-	unsigned long last_timer;    /* last timer data for each instance */
+	unsigned long last_timer;	/* last timer data for each instance */
 
 	struct dsps_context context;
 };
@@ -213,7 +213,7 @@ static void dsps_musb_disable(struct musb *musb)
 
 	dsps_writel(reg_base, wrp->coreintr_clear, wrp->usb_bitmap);
 	dsps_writel(reg_base, wrp->epintr_clear,
-			 wrp->txep_bitmap | wrp->rxep_bitmap);
+		    wrp->txep_bitmap | wrp->rxep_bitmap);
 	dsps_writeb(musb->mregs, MUSB_DEVCTL, 0);
 }
 
@@ -234,7 +234,7 @@ static void otg_timer(unsigned long _musb)
 	 */
 	devctl = dsps_readb(mregs, MUSB_DEVCTL);
 	dev_dbg(musb->controller, "Poll devctl %02x (%s)\n", devctl,
-				usb_otg_state_string(musb->xceiv->state));
+		usb_otg_state_string(musb->xceiv->state));
 
 	spin_lock_irqsave(&musb->lock, flags);
 	switch (musb->xceiv->state) {
@@ -298,7 +298,7 @@ static irqreturn_t dsps_interrupt(int irq, void *hci)
 		dsps_writel(reg_base, wrp->coreintr_status, usbintr);
 
 	dev_dbg(musb->controller, "usbintr (%x) epintr(%x)\n",
-			usbintr, epintr);
+		usbintr, epintr);
 	/*
 	 * DRVVBUS IRQs are the only proxy we have (a very poor one!) for
 	 * DSPS IP's missing ID change IRQ.  We need an ID change IRQ to
@@ -348,10 +348,9 @@ static irqreturn_t dsps_interrupt(int irq, void *hci)
 
 		/* NOTE: this must complete power-on within 100 ms. */
 		dev_dbg(musb->controller, "VBUS %s (%s)%s, devctl %02x\n",
-				drvvbus ? "on" : "off",
-				usb_otg_state_string(musb->xceiv->state),
-				err ? " ERROR" : "",
-				devctl);
+			drvvbus ? "on" : "off",
+			usb_otg_state_string(musb->xceiv->state),
+			err ? " ERROR" : "", devctl);
 		ret = IRQ_HANDLED;
 	}
 
@@ -692,9 +691,9 @@ static SIMPLE_DEV_PM_OPS(dsps_pm_ops, dsps_suspend, dsps_resume);
 
 static struct platform_driver dsps_usbss_driver = {
 	.probe		= dsps_probe,
-	.remove         = dsps_remove,
-	.driver         = {
-		.name   = "musb-dsps",
+	.remove		= dsps_remove,
+	.driver		= {
+		.name	= "musb-dsps",
 		.pm	= &dsps_pm_ops,
 		.of_match_table	= musb_dsps_of_match,
 	},
Signature · 1 line
1.8.4.2
#4 lore raw

Re: [PATCH 1/5] usb: musb: conditionally save and restore the context on suspend

In reply to Daniel Mack

Hi,

Quoted 18 lines · Daniel Mack

It appears not all platforms featuring a musb core need to save the musb core registers at suspend time and restore them on resume.

The dsps platform does, however. So add a bit in struct musb_hdrc_platform_data to let platforms specify their need of such action being taken.

---
 drivers/usb/musb/musb_core.c | 17 ++++++++++++++++-
 include/linux/usb/musb.h     |  3 +++
 2 files changed, 19 insertions(+), 1 deletion(-)
drivers/usb/musb/musb_core.c +0 -0
index 0a43329..a8ded57 100644

drivers/usb/musb/musb_core.c static int musb_suspend(struct device *dev) · line 2205

Quoted 4 lines
drivers/usb/musb/musb_core.c +1 -0
@@ -2202,6 +2202,7 @@ static int musb_suspend(struct device *dev)
 {
 	struct musb	*musb = dev_to_musb(dev);
 	unsigned long	flags;
+	struct musb_hdrc_platform_data *plat = dev_get_platdata(dev);

we don't want to have platform_data on DT-based boot. It's best to just save those registers unconditionally as it doesn't hurt.

Signature · 1 line
balbi
#13 lore raw

Re: [PATCH 2/5] usb: musb: call musb_port_suspend from musb_bus_suspend

In reply to a message not in the archive
Quoted 16 lines · Daniel Mack

Make musb_port_suspend() externally available, and call it when to host goes into suspend. This allows the core to go into suspend while a device is connected.

---
 drivers/usb/musb/musb_host.c    | 2 ++
 drivers/usb/musb/musb_host.h    | 2 ++
 drivers/usb/musb/musb_virthub.c | 2 +-
 3 files changed, 5 insertions(+), 1 deletion(-)
drivers/usb/musb/musb_host.c +0 -0
index 6582a20..81caf9f 100644

drivers/usb/musb/musb_host.c static int musb_bus_suspend(struct usb_hcd *hcd) · line 2436

Quoted 4 lines
drivers/usb/musb/musb_host.c +1 -0
@@ -2433,6 +2433,8 @@ static int musb_bus_suspend(struct usb_hcd *hcd)
 	struct musb	*musb = hcd_to_musb(hcd);
 	u8		devctl;
 
+	musb_port_suspend(musb, true);

have you considered the fact that when musb looses context it'll cause a disconnect on the bus because soft_connect bit is lost ?

What if you have a mounted file system on a pendrive ? Should we allow suspend in that case ?

Alan, Greg, any hints ?

Signature · 1 line
balbi
#30 lore raw

Re: [PATCH 4/5] usb: musb: dsps: add support for suspend and resume

In reply to a message not in the archive
Quoted 149 lines · Daniel Mack

The dsps platform needs to save save some registers at suspend time and restore them after resume. This patch adds a struct for these registers, and also lets the musb core know that the core registers need to be saved as well.

We also have to call musb_port_reset() for this platform upon resume, so this function has to be made non-static.

---
 drivers/usb/musb/musb_core.h    |  1 +
 drivers/usb/musb/musb_dsps.c    | 59 +++++++++++++++++++++++++++++++++++++++++
 drivers/usb/musb/musb_host.h    |  2 ++
 drivers/usb/musb/musb_virthub.c |  2 +-
 4 files changed, 63 insertions(+), 1 deletion(-)
drivers/usb/musb/musb_core.h +1 -0
index 29f7cd7..a423037 100644
@@ -295,6 +295,7 @@ struct musb {
 
 	irqreturn_t		(*isr)(int, void *);
 	struct work_struct	irq_work;
+
 	u16			hwvers;
 
 	u16			intrrxe;
drivers/usb/musb/musb_dsps.c +59 -0
index e57d712..361ddf8 100644
@@ -112,6 +112,19 @@ struct dsps_musb_wrapper {
 	u8		poll_seconds;
 };
 
+/*
+ * register shadow for suspend
+ */
+struct dsps_context {
+	u32 control;
+	u32 epintr;
+	u32 coreintr;
+	u32 phy_utmi;
+	u32 mode;
+	u32 tx_mode;
+	u32 rx_mode;
+};
+
 /**
  * DSPS glue structure.
  */
@@ -121,6 +134,8 @@ struct dsps_glue {
 	const struct dsps_musb_wrapper *wrp; /* wrapper register offsets */
 	struct timer_list timer;	/* otg_workaround timer */
 	unsigned long last_timer;    /* last timer data for each instance */
+
+	struct dsps_context context;
 };
 
 static void dsps_musb_try_idle(struct musb *musb, unsigned long timeout)
@@ -506,6 +521,7 @@ static int dsps_create_musb_pdev(struct dsps_glue *glue,
 	}
 	pdata.config = config;
 	pdata.platform_ops = &dsps_ops;
+	pdata.restore_after_suspend = 1;
 
 	config->num_eps = get_int_prop(dn, "mentor,num-eps");
 	config->ram_bits = get_int_prop(dn, "mentor,ram-bits");
@@ -632,11 +648,54 @@ static const struct of_device_id musb_dsps_of_match[] = {
 };
 MODULE_DEVICE_TABLE(of, musb_dsps_of_match);
 
+#ifdef CONFIG_PM
+static int dsps_suspend(struct device *dev)
+{
+	struct dsps_glue *glue = dev_get_drvdata(dev);
+	const struct dsps_musb_wrapper *wrp = glue->wrp;
+	struct musb *musb = platform_get_drvdata(glue->musb);
+	void __iomem *mbase = musb->ctrl_base;
+
+	glue->context.control = dsps_readl(mbase, wrp->control);
+	glue->context.epintr = dsps_readl(mbase, wrp->epintr_set);
+	glue->context.coreintr = dsps_readl(mbase, wrp->coreintr_set);
+	glue->context.phy_utmi = dsps_readl(mbase, wrp->phy_utmi);
+	glue->context.mode = dsps_readl(mbase, wrp->mode);
+	glue->context.tx_mode = dsps_readl(mbase, wrp->tx_mode);
+	glue->context.rx_mode = dsps_readl(mbase, wrp->rx_mode);
+
+	return 0;
+}
+
+static int dsps_resume(struct device *dev)
+{
+	struct dsps_glue *glue = dev_get_drvdata(dev);
+	const struct dsps_musb_wrapper *wrp = glue->wrp;
+	struct musb *musb = platform_get_drvdata(glue->musb);
+	void __iomem *mbase = musb->ctrl_base;
+
+	dsps_writel(mbase, wrp->control, glue->context.control);
+	dsps_writel(mbase, wrp->epintr_set, glue->context.epintr);
+	dsps_writel(mbase, wrp->coreintr_set, glue->context.coreintr);
+	dsps_writel(mbase, wrp->phy_utmi, glue->context.phy_utmi);
+	dsps_writel(mbase, wrp->mode, glue->context.mode);
+	dsps_writel(mbase, wrp->tx_mode, glue->context.tx_mode);
+	dsps_writel(mbase, wrp->rx_mode, glue->context.rx_mode);
+
+	musb_port_reset(musb, false);
+
+	return 0;
+}
+#endif
+
+static SIMPLE_DEV_PM_OPS(dsps_pm_ops, dsps_suspend, dsps_resume);
+
 static struct platform_driver dsps_usbss_driver = {
 	.probe		= dsps_probe,
 	.remove         = dsps_remove,
 	.driver         = {
 		.name   = "musb-dsps",
+		.pm	= &dsps_pm_ops,
 		.of_match_table	= musb_dsps_of_match,
 	},
 };
drivers/usb/musb/musb_host.h +2 -0
index e660af9..7436c24 100644
@@ -93,6 +93,7 @@ extern void musb_root_disconnect(struct musb *musb);
 extern void musb_host_resume_root_hub(struct musb *musb);
 extern void musb_host_poke_root_hub(struct musb *musb);
 extern void musb_port_suspend(struct musb *musb, bool do_suspend);
+extern void musb_port_reset(struct musb *musb, bool do_reset);
 #else
 static inline struct musb *hcd_to_musb(struct usb_hcd *hcd)
 {
@@ -123,6 +124,7 @@ static inline void musb_host_resume_root_hub(struct musb *musb)	{}
 static inline void musb_host_poll_rh_status(struct musb *musb)	{}
 static inline void musb_host_poke_root_hub(struct musb *musb)	{}
 static inline void musb_port_suspend(struct musb *musb, bool do_suspend) {}
+static inline void musb_port_reset(struct musb *musb)		{}
 #endif
 
 struct usb_hcd;
drivers/usb/musb/musb_virthub.c +0 -0
index e977441..24e46c0 100644

drivers/usb/musb/musb_virthub.c void musb_port_suspend(struct musb *musb, bool do_suspend) · line 112

Quoted 5 lines
drivers/usb/musb/musb_virthub.c +1 -1
@@ -109,7 +109,7 @@ void musb_port_suspend(struct musb *musb, bool do_suspend)
 	}
 }
 
-static void musb_port_reset(struct musb *musb, bool do_reset)
+void musb_port_reset(struct musb *musb, bool do_reset)

NAK, this should not be called from the glue layers at all. If anything call from musb_core's resume callback. That will only be called after parent's resume has been called anyway.

Signature · 1 line
balbi
#40 lore raw

Re: [PATCH 5/5] usb: musb: dsps: indentation and whitespace fixes

In reply to a message not in the archive
Quoted 24 lines · Daniel Mack

Just a cosmetic thing, no functional change.

---
 drivers/usb/musb/musb_dsps.c | 23 +++++++++++------------
 1 file changed, 11 insertions(+), 12 deletions(-)
drivers/usb/musb/musb_dsps.c +2 -2
index 361ddf8..b27c072 100644
@@ -131,9 +131,9 @@ struct dsps_context {
 struct dsps_glue {
 	struct device *dev;
 	struct platform_device *musb;	/* child musb pdev */
-	const struct dsps_musb_wrapper *wrp; /* wrapper register offsets */
+	const struct dsps_musb_wrapper *wrp;	/* wrapper register offsets */
 	struct timer_list timer;	/* otg_workaround timer */
-	unsigned long last_timer;    /* last timer data for each instance */
+	unsigned long last_timer;	/* last timer data for each instance */
 
 	struct dsps_context context;
 };

drivers/usb/musb/musb_dsps.c static void dsps_musb_disable(struct musb *musb) · line 216

Quoted 5 lines
drivers/usb/musb/musb_dsps.c +1 -1
@@ -213,7 +213,7 @@ static void dsps_musb_disable(struct musb *musb)
 
 	dsps_writel(reg_base, wrp->coreintr_clear, wrp->usb_bitmap);
 	dsps_writel(reg_base, wrp->epintr_clear,
-			 wrp->txep_bitmap | wrp->rxep_bitmap);
+		    wrp->txep_bitmap | wrp->rxep_bitmap);

quite frankly, I don't what checkpatch.pl is trying to force here. I much prefer a tab-only indentation, matching with ( characters forces us to add some spaces. I would rather not take this patch.

Signature · 1 line
balbi
#15 lore raw

Re: [PATCH 2/5] usb: musb: call musb_port_suspend from musb_bus_suspend

In reply to a message not in the archive
Quoted 20 lines · Felipe Balbi
Quoted 19 lines · Daniel Mack

Make musb_port_suspend() externally available, and call it when to host goes into suspend. This allows the core to go into suspend while a device is connected.

---
 drivers/usb/musb/musb_host.c    | 2 ++
 drivers/usb/musb/musb_host.h    | 2 ++
 drivers/usb/musb/musb_virthub.c | 2 +-
 3 files changed, 5 insertions(+), 1 deletion(-)
drivers/usb/musb/musb_host.c +0 -0
index 6582a20..81caf9f 100644
@@ -2433,6 +2433,8 @@ static int musb_bus_suspend(struct usb_hcd *hcd)
 	struct musb	*musb = hcd_to_musb(hcd);
 	u8		devctl;
 
Quoted 7 lines

drivers/usb/musb/musb_host.c static int musb_bus_suspend(struct usb_hcd *hcd) · line 2436

Quoted 1 line
drivers/usb/musb/musb_host.c +1 -0
+	musb_port_suspend(musb, true);

have you considered the fact that when musb looses context it'll cause a disconnect on the bus because soft_connect bit is lost ?

What if you have a mounted file system on a pendrive ? Should we allow suspend in that case ?

Well, I would have expected that, but in fact, the opposite is true. With 3.12, mounting a filesystem on a USB media and accessing it after resume was exactly my test case, and it worked just fine with that patch.

In 3.13, something about the parition table reading seems to be broken currently, I'll have a closer look.

Thanks,
Daniel

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

#41 lore raw

Re: [PATCH 5/5] usb: musb: dsps: indentation and whitespace fixes

In reply to Felipe Balbi
Quoted 26 lines · Felipe Balbi
Quoted 25 lines · Daniel Mack

Just a cosmetic thing, no functional change.

---
 drivers/usb/musb/musb_dsps.c | 23 +++++++++++------------
 1 file changed, 11 insertions(+), 12 deletions(-)
drivers/usb/musb/musb_dsps.c +2 -2
index 361ddf8..b27c072 100644
@@ -131,9 +131,9 @@ struct dsps_context {
 struct dsps_glue {
 	struct device *dev;
 	struct platform_device *musb;	/* child musb pdev */
-	const struct dsps_musb_wrapper *wrp; /* wrapper register offsets */
+	const struct dsps_musb_wrapper *wrp;	/* wrapper register offsets */
 	struct timer_list timer;	/* otg_workaround timer */
-	unsigned long last_timer;    /* last timer data for each instance */
+	unsigned long last_timer;	/* last timer data for each instance */
 
 	struct dsps_context context;
 };
@@ -213,7 +213,7 @@ static void dsps_musb_disable(struct musb *musb)
 
Quoted 9 lines

drivers/usb/musb/musb_dsps.c static void dsps_musb_disable(struct musb *musb) · line 216

Quoted 4 lines
drivers/usb/musb/musb_dsps.c +1 -1
 	dsps_writel(reg_base, wrp->coreintr_clear, wrp->usb_bitmap);
 	dsps_writel(reg_base, wrp->epintr_clear,
-			 wrp->txep_bitmap | wrp->rxep_bitmap);
+		    wrp->txep_bitmap | wrp->rxep_bitmap);

quite frankly, I don't what checkpatch.pl is trying to force here. I much prefer a tab-only indentation, matching with ( characters forces us to add some spaces. I would rather not take this patch.

Alright, I really don't care much. I'll drop the patch when resending.

#16 lore raw

Re: [PATCH 2/5] usb: musb: call musb_port_suspend from musb_bus_suspend

In reply to Daniel Mack

Hi,

Quoted 26 lines · Daniel Mack
Quoted 25 lines · Felipe Balbi
Quoted 16 lines · Daniel Mack

Make musb_port_suspend() externally available, and call it when to host goes into suspend. This allows the core to go into suspend while a device is connected.

---
 drivers/usb/musb/musb_host.c    | 2 ++
 drivers/usb/musb/musb_host.h    | 2 ++
 drivers/usb/musb/musb_virthub.c | 2 +-
 3 files changed, 5 insertions(+), 1 deletion(-)
drivers/usb/musb/musb_host.c +0 -0
index 6582a20..81caf9f 100644

drivers/usb/musb/musb_host.c static int musb_bus_suspend(struct usb_hcd *hcd) · line 2436

Quoted 4 lines
drivers/usb/musb/musb_host.c +1 -0
@@ -2433,6 +2433,8 @@ static int musb_bus_suspend(struct usb_hcd *hcd)
 	struct musb	*musb = hcd_to_musb(hcd);
 	u8		devctl;
 
+	musb_port_suspend(musb, true);

have you considered the fact that when musb looses context it'll cause a disconnect on the bus because soft_connect bit is lost ?

Quoted 6 lines
Quoted 2 lines

What if you have a mounted file system on a pendrive ? Should we allow suspend in that case ?

Well, I would have expected that, but in fact, the opposite is true. With 3.12, mounting a filesystem on a USB media and accessing it after resume was exactly my test case, and it worked just fine with that patch.

so you had a mounted file system, suspend, resume, it was still mounted? or did it reenumerate and you remounted it ?

Try to do the same with transfers in flight, it's likely to corrupt your file system.

Quoted 2 lines

In 3.13, something about the parition table reading seems to be broken currently, I'll have a closer look.

cool

Signature · 1 line
balbi
#31 lore raw

Re: [PATCH 4/5] usb: musb: dsps: add support for suspend and resume

In reply to Felipe Balbi
Quoted 153 lines · Felipe Balbi
Quoted 152 lines · Daniel Mack

The dsps platform needs to save save some registers at suspend time and restore them after resume. This patch adds a struct for these registers, and also lets the musb core know that the core registers need to be saved as well.

We also have to call musb_port_reset() for this platform upon resume, so this function has to be made non-static.

---
 drivers/usb/musb/musb_core.h    |  1 +
 drivers/usb/musb/musb_dsps.c    | 59 +++++++++++++++++++++++++++++++++++++++++
 drivers/usb/musb/musb_host.h    |  2 ++
 drivers/usb/musb/musb_virthub.c |  2 +-
 4 files changed, 63 insertions(+), 1 deletion(-)
drivers/usb/musb/musb_core.h +1 -0
index 29f7cd7..a423037 100644
@@ -295,6 +295,7 @@ struct musb {
 
 	irqreturn_t		(*isr)(int, void *);
 	struct work_struct	irq_work;
+
 	u16			hwvers;
 
 	u16			intrrxe;
drivers/usb/musb/musb_dsps.c +59 -0
index e57d712..361ddf8 100644
@@ -112,6 +112,19 @@ struct dsps_musb_wrapper {
 	u8		poll_seconds;
 };
 
+/*
+ * register shadow for suspend
+ */
+struct dsps_context {
+	u32 control;
+	u32 epintr;
+	u32 coreintr;
+	u32 phy_utmi;
+	u32 mode;
+	u32 tx_mode;
+	u32 rx_mode;
+};
+
 /**
  * DSPS glue structure.
  */
@@ -121,6 +134,8 @@ struct dsps_glue {
 	const struct dsps_musb_wrapper *wrp; /* wrapper register offsets */
 	struct timer_list timer;	/* otg_workaround timer */
 	unsigned long last_timer;    /* last timer data for each instance */
+
+	struct dsps_context context;
 };
 
 static void dsps_musb_try_idle(struct musb *musb, unsigned long timeout)
@@ -506,6 +521,7 @@ static int dsps_create_musb_pdev(struct dsps_glue *glue,
 	}
 	pdata.config = config;
 	pdata.platform_ops = &dsps_ops;
+	pdata.restore_after_suspend = 1;
 
 	config->num_eps = get_int_prop(dn, "mentor,num-eps");
 	config->ram_bits = get_int_prop(dn, "mentor,ram-bits");
@@ -632,11 +648,54 @@ static const struct of_device_id musb_dsps_of_match[] = {
 };
 MODULE_DEVICE_TABLE(of, musb_dsps_of_match);
 
+#ifdef CONFIG_PM
+static int dsps_suspend(struct device *dev)
+{
+	struct dsps_glue *glue = dev_get_drvdata(dev);
+	const struct dsps_musb_wrapper *wrp = glue->wrp;
+	struct musb *musb = platform_get_drvdata(glue->musb);
+	void __iomem *mbase = musb->ctrl_base;
+
+	glue->context.control = dsps_readl(mbase, wrp->control);
+	glue->context.epintr = dsps_readl(mbase, wrp->epintr_set);
+	glue->context.coreintr = dsps_readl(mbase, wrp->coreintr_set);
+	glue->context.phy_utmi = dsps_readl(mbase, wrp->phy_utmi);
+	glue->context.mode = dsps_readl(mbase, wrp->mode);
+	glue->context.tx_mode = dsps_readl(mbase, wrp->tx_mode);
+	glue->context.rx_mode = dsps_readl(mbase, wrp->rx_mode);
+
+	return 0;
+}
+
+static int dsps_resume(struct device *dev)
+{
+	struct dsps_glue *glue = dev_get_drvdata(dev);
+	const struct dsps_musb_wrapper *wrp = glue->wrp;
+	struct musb *musb = platform_get_drvdata(glue->musb);
+	void __iomem *mbase = musb->ctrl_base;
+
+	dsps_writel(mbase, wrp->control, glue->context.control);
+	dsps_writel(mbase, wrp->epintr_set, glue->context.epintr);
+	dsps_writel(mbase, wrp->coreintr_set, glue->context.coreintr);
+	dsps_writel(mbase, wrp->phy_utmi, glue->context.phy_utmi);
+	dsps_writel(mbase, wrp->mode, glue->context.mode);
+	dsps_writel(mbase, wrp->tx_mode, glue->context.tx_mode);
+	dsps_writel(mbase, wrp->rx_mode, glue->context.rx_mode);
+
+	musb_port_reset(musb, false);
+
+	return 0;
+}
+#endif
+
+static SIMPLE_DEV_PM_OPS(dsps_pm_ops, dsps_suspend, dsps_resume);
+
 static struct platform_driver dsps_usbss_driver = {
 	.probe		= dsps_probe,
 	.remove         = dsps_remove,
 	.driver         = {
 		.name   = "musb-dsps",
+		.pm	= &dsps_pm_ops,
 		.of_match_table	= musb_dsps_of_match,
 	},
 };
drivers/usb/musb/musb_host.h +2 -0
index e660af9..7436c24 100644
@@ -93,6 +93,7 @@ extern void musb_root_disconnect(struct musb *musb);
 extern void musb_host_resume_root_hub(struct musb *musb);
 extern void musb_host_poke_root_hub(struct musb *musb);
 extern void musb_port_suspend(struct musb *musb, bool do_suspend);
+extern void musb_port_reset(struct musb *musb, bool do_reset);
 #else
 static inline struct musb *hcd_to_musb(struct usb_hcd *hcd)
 {
@@ -123,6 +124,7 @@ static inline void musb_host_resume_root_hub(struct musb *musb)	{}
 static inline void musb_host_poll_rh_status(struct musb *musb)	{}
 static inline void musb_host_poke_root_hub(struct musb *musb)	{}
 static inline void musb_port_suspend(struct musb *musb, bool do_suspend) {}
+static inline void musb_port_reset(struct musb *musb)		{}
 #endif
 
 struct usb_hcd;
drivers/usb/musb/musb_virthub.c +0 -0
index e977441..24e46c0 100644
@@ -109,7 +109,7 @@ void musb_port_suspend(struct musb *musb, bool do_suspend)
 	}
 }
 
Quoted 6 lines

drivers/usb/musb/musb_virthub.c void musb_port_suspend(struct musb *musb, bool do_suspend) · line 112

Quoted 2 lines
drivers/usb/musb/musb_virthub.c +1 -1
-static void musb_port_reset(struct musb *musb, bool do_reset)
+void musb_port_reset(struct musb *musb, bool do_reset)

NAK, this should not be called from the glue layers at all. If anything call from musb_core's resume callback. That will only be called after parent's resume has been called anyway.

Given that this driver is successfully used with suspend/resume on other platforms without that reset, but it's clearly needed for dsps, I'm fairly sure this should be a glue-layer specific thing. Hence the change.

I think it'll break things on other platforms if we unconditionally reset the port after resume, but I can't test it. Anyone?

Thanks,
Daniel

#33 lore raw

Re: [PATCH 4/5] usb: musb: dsps: add support for suspend and resume

In reply to a message not in the archive

Hi,

Quoted 15 lines · Daniel Mack
Quoted 8 lines
Quoted 8 lines
drivers/usb/musb/musb_virthub.c +0 -0
index e977441..24e46c0 100644
@@ -109,7 +109,7 @@ void musb_port_suspend(struct musb *musb, bool do_suspend)
 	}
 }
 
Quoted 6 lines

drivers/usb/musb/musb_virthub.c void musb_port_suspend(struct musb *musb, bool do_suspend) · line 112

Quoted 2 lines
drivers/usb/musb/musb_virthub.c +1 -1
-static void musb_port_reset(struct musb *musb, bool do_reset)
+void musb_port_reset(struct musb *musb, bool do_reset)

NAK, this should not be called from the glue layers at all. If anything call from musb_core's resume callback. That will only be called after parent's resume has been called anyway.

Quoted 6 lines

Given that this driver is successfully used with suspend/resume on other platforms without that reset, but it's clearly needed for dsps, I'm fairly sure this should be a glue-layer specific thing. Hence the change.

I think it'll break things on other platforms if we unconditionally reset the port after resume, but I can't test it. Anyone?

your original commit log only says "we need to issue port reset" but it never explains why. It could very well be you just found a device which needs to be reset when coming out of suspend and it misses a quirk flag ?

In any case, those functions should never be called by the glue, if reset needs to be called, it must be called by musb-hdrc.ko, if you need a flag, pass a flag but I need a really good explanation in your commit log of why that's necessary.

Signature · 1 line
balbi
#5 lore raw

Re: [PATCH 1/5] usb: musb: conditionally save and restore the context on suspend

In reply to Felipe Balbi
Quoted 16 lines · Felipe Balbi

Hi,

Quoted 13 lines · Daniel Mack

It appears not all platforms featuring a musb core need to save the musb core registers at suspend time and restore them on resume.

The dsps platform does, however. So add a bit in struct musb_hdrc_platform_data to let platforms specify their need of such action being taken.

---
 drivers/usb/musb/musb_core.c | 17 ++++++++++++++++-
 include/linux/usb/musb.h     |  3 +++
 2 files changed, 19 insertions(+), 1 deletion(-)
Quoted 13 lines

drivers/usb/musb/musb_core.c static int musb_suspend(struct device *dev) · line 2205

Quoted 9 lines
drivers/usb/musb/musb_core.c +1 -0
index 0a43329..a8ded57 100644
@@ -2202,6 +2202,7 @@ static int musb_suspend(struct device *dev)
 {
 	struct musb	*musb = dev_to_musb(dev);
 	unsigned long	flags;
+	struct musb_hdrc_platform_data *plat = dev_get_platdata(dev);

we don't want to have platform_data on DT-based boot. It's best to just save those registers unconditionally as it doesn't hurt.

My concern about doing it unconditionally from the core is simply that I fear regressions for other platforms. I can of course drop it if you're certain that that's correct.

I can only test this on a dsps glue layer, and I have no documentation for the musb core. All I'm left with here is fishing in muddy waters :/

Daniel

#6 lore raw

Re: [PATCH 1/5] usb: musb: conditionally save and restore the context on suspend

Hi,

Quoted 31 lines · Daniel Mack
Quoted 16 lines · Felipe Balbi

Hi,

Quoted 13 lines · Daniel Mack

It appears not all platforms featuring a musb core need to save the musb core registers at suspend time and restore them on resume.

The dsps platform does, however. So add a bit in struct musb_hdrc_platform_data to let platforms specify their need of such action being taken.

---
 drivers/usb/musb/musb_core.c | 17 ++++++++++++++++-
 include/linux/usb/musb.h     |  3 +++
 2 files changed, 19 insertions(+), 1 deletion(-)
Quoted 13 lines

drivers/usb/musb/musb_core.c static int musb_suspend(struct device *dev) · line 2205

Quoted 9 lines
drivers/usb/musb/musb_core.c +1 -0
index 0a43329..a8ded57 100644
@@ -2202,6 +2202,7 @@ static int musb_suspend(struct device *dev)
 {
 	struct musb	*musb = dev_to_musb(dev);
 	unsigned long	flags;
+	struct musb_hdrc_platform_data *plat = dev_get_platdata(dev);

we don't want to have platform_data on DT-based boot. It's best to just save those registers unconditionally as it doesn't hurt.

Quoted 6 lines

My concern about doing it unconditionally from the core is simply that I fear regressions for other platforms. I can of course drop it if you're certain that that's correct.

I can only test this on a dsps glue layer, and I have no documentation for the musb core. All I'm left with here is fishing in muddy waters :/

I really don't think it should cause any issues with any other platform.

We can make to leave this soaking in linux-next for a long time and hope people test it.

Signature · 1 line
balbi
#17 lore raw

Re: [PATCH 2/5] usb: musb: call musb_port_suspend from musb_bus_suspend

In reply to Felipe Balbi
Quoted 29 lines · Felipe Balbi

Hi,

Quoted 26 lines · Daniel Mack
Quoted 25 lines · Felipe Balbi
Quoted 16 lines · Daniel Mack

Make musb_port_suspend() externally available, and call it when to host goes into suspend. This allows the core to go into suspend while a device is connected.

---
 drivers/usb/musb/musb_host.c    | 2 ++
 drivers/usb/musb/musb_host.h    | 2 ++
 drivers/usb/musb/musb_virthub.c | 2 +-
 3 files changed, 5 insertions(+), 1 deletion(-)
drivers/usb/musb/musb_host.c +0 -0
index 6582a20..81caf9f 100644

drivers/usb/musb/musb_host.c static int musb_bus_suspend(struct usb_hcd *hcd) · line 2436

Quoted 4 lines
drivers/usb/musb/musb_host.c +1 -0
@@ -2433,6 +2433,8 @@ static int musb_bus_suspend(struct usb_hcd *hcd)
 	struct musb	*musb = hcd_to_musb(hcd);
 	u8		devctl;
 
+	musb_port_suspend(musb, true);

have you considered the fact that when musb looses context it'll cause a disconnect on the bus because soft_connect bit is lost ?

Quoted 8 lines
Quoted 6 lines
Quoted 2 lines

What if you have a mounted file system on a pendrive ? Should we allow suspend in that case ?

Well, I would have expected that, but in fact, the opposite is true. With 3.12, mounting a filesystem on a USB media and accessing it after resume was exactly my test case, and it worked just fine with that patch.

so you had a mounted file system, suspend, resume, it was still mounted?

Yes, exactly.

Quoted 1 line

or did it reenumerate and you remounted it ?

No, it did not reenumerate, but the core did a reset on it. See the following log.

# mount /dev/sda /mnt
# md5sum /mnt/sine_left.wav
bcd90548c1ebf293289072f1a6161f0f  /mnt/sine_left.wav
#
#
# echo mem > /sys/power/state
[   31.553425] PM: Syncing filesystems ... done.
[   31.575462] Freezing user space processes ... (elapsed 0.002 seconds)
done.
[   31.585275] Freezing remaining freezable tasks ... (elapsed 0.002
seconds) done.
[   31.595953] PM: Sending message for entering DeepSleep mode
[   31.628887] PM: suspend of devices complete after 21.097 msecs
[   31.641082] PM: late suspend of devices complete after 5.906 msecs
[   31.653834] PM: noirq suspend of devices complete after 6.135 msecs
[   31.660558] PM: GFX domain entered low power state
[   31.660558] PM: Successfully put all powerdomains to target state
[   31.660558] PM: Wakeup source GPIO0
[   31.679130] PM: noirq resume of devices complete after 17.941 msecs
[   31.693371] PM: early resume of devices complete after 4.388 msecs
[   31.702750] net eth0: initializing cpsw version 1.12 (0)
[   31.712084] net eth0: phy found : id is : 0x4dd076
[   31.717317] libphy: PHY  not found
[   31.722448] net eth0: phy  not found on slave 1
[   32.030792] usb 1-1: reset high-speed USB device number 2 using
musb-hdrc <<<< !
[   32.395192] PM: resume of devices complete after 695.279 msecs
[   32.407816] PM: Sending message for resetting M3 state machine
[   32.414683] Restarting tasks ... done.
[   36.701568] libphy: 4a101000.mdio:04 - Link is Up - 100/Full
#
#
# md5sum /mnt/sine_left.wav
bcd90548c1ebf293289072f1a6161f0f  /mnt/sine_left.wav
Quoted 2 lines

Try to do the same with transfers in flight, it's likely to corrupt your file system.

I did that as well, and it appeared stable to me. After all, the USB device was properly put into suspend before VBUS went away, and the usb storage driver caught up after the core's reset.

In my tests, this was the only approach that would work after resume.

Daniel

#34 lore raw

Re: [PATCH 4/5] usb: musb: dsps: add support for suspend and resume

In reply to Felipe Balbi
Quoted 20 lines · Felipe Balbi
Quoted 19 lines · Daniel Mack
Quoted 8 lines
Quoted 8 lines
drivers/usb/musb/musb_virthub.c +0 -0
index e977441..24e46c0 100644
@@ -109,7 +109,7 @@ void musb_port_suspend(struct musb *musb, bool do_suspend)
 	}
 }
 
Quoted 6 lines

drivers/usb/musb/musb_virthub.c void musb_port_suspend(struct musb *musb, bool do_suspend) · line 112

Quoted 2 lines
drivers/usb/musb/musb_virthub.c +1 -1
-static void musb_port_reset(struct musb *musb, bool do_reset)
+void musb_port_reset(struct musb *musb, bool do_reset)

NAK, this should not be called from the glue layers at all. If anything call from musb_core's resume callback. That will only be called after parent's resume has been called anyway.

Given that this driver is successfully used with suspend/resume on other platforms without that reset, but it's clearly needed for dsps, I'm fairly sure this should be a glue-layer specific thing. Hence the change.

Quoted 6 lines
Quoted 2 lines

I think it'll break things on other platforms if we unconditionally reset the port after resume, but I can't test it. Anyone?

your original commit log only says "we need to issue port reset" but it never explains why. It could very well be you just found a device which needs to be reset when coming out of suspend and it misses a quirk flag ?

I think I can really rule that out, as I tested with multiple USB sticks and also tested the same sticks on other embedded platforms.

Quoted 4 lines

In any case, those functions should never be called by the glue, if reset needs to be called, it must be called by musb-hdrc.ko, if you need a flag, pass a flag but I need a really good explanation in your commit log of why that's necessary.

Well, if I'd only knew exactly why. All these patches are really the result of many days of trial and error, with multiple drivers (musb, cppi41, ...) involved. And as I said - I have no documentation of the musb core itself.

So yes, I can do it the other way around and pass a flag, but what I can't provide is a good explanantion why the dsps glue behaves differently here than others. I'm curious myself, and all I know is that with this reset in place, things work as expected on AM33xx.

Thanks for your feedback,
Daniel

#8 lore raw

Re: [PATCH 1/5] usb: musb: conditionally save and restore the context on suspend

In reply to a message not in the archive
Quoted 38 lines · Felipe Balbi

Hi,

Quoted 35 lines · Daniel Mack
Quoted 16 lines · Felipe Balbi

Hi,

Quoted 13 lines · Daniel Mack

It appears not all platforms featuring a musb core need to save the musb core registers at suspend time and restore them on resume.

The dsps platform does, however. So add a bit in struct musb_hdrc_platform_data to let platforms specify their need of such action being taken.

---
 drivers/usb/musb/musb_core.c | 17 ++++++++++++++++-
 include/linux/usb/musb.h     |  3 +++
 2 files changed, 19 insertions(+), 1 deletion(-)
Quoted 13 lines

drivers/usb/musb/musb_core.c static int musb_suspend(struct device *dev) · line 2205

Quoted 9 lines
drivers/usb/musb/musb_core.c +1 -0
index 0a43329..a8ded57 100644
@@ -2202,6 +2202,7 @@ static int musb_suspend(struct device *dev)
 {
 	struct musb	*musb = dev_to_musb(dev);
 	unsigned long	flags;
+	struct musb_hdrc_platform_data *plat = dev_get_platdata(dev);

we don't want to have platform_data on DT-based boot. It's best to just save those registers unconditionally as it doesn't hurt.

My concern about doing it unconditionally from the core is simply that I fear regressions for other platforms. I can of course drop it if you're certain that that's correct.

Quoted 8 lines
Quoted 2 lines

I can only test this on a dsps glue layer, and I have no documentation for the musb core. All I'm left with here is fishing in muddy waters :/

I really don't think it should cause any issues with any other platform.

We can make to leave this soaking in linux-next for a long time and hope people test it.

Ok, sounds good. I'll do it and repost, along with the other changes.

Many thanks,
Daniel

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

#27 lore raw

Re: [PATCH 2/5] usb: musb: call musb_port_suspend from musb_bus_suspend

In reply to a message not in the archive
Quoted 5 lines · Felipe Balbi
Quoted 3 lines · Daniel Mack

Make musb_port_suspend() externally available, and call it when to host goes into suspend. This allows the core to go into suspend while a device is connected.

Quoted 7 lines

have you considered the fact that when musb looses context it'll cause a disconnect on the bus because soft_connect bit is lost ?

What if you have a mounted file system on a pendrive ? Should we allow suspend in that case ?

Alan, Greg, any hints ?

An HCD should strive to avoid losing power sessions over a suspend/resume. But if this is unavoidable, the USB Persist mechanism should be able to cope by issuing a reset-resume. (Unless userspace has turned Persist off, of course -- I believe some distributions do this. Chrome?)

Alan Stern

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

#19 lore raw

Re: [PATCH 2/5] usb: musb: call musb_port_suspend from musb_bus_suspend

In reply to a message not in the archive
Quoted 15 lines · Daniel Mack
Quoted 8 lines
Quoted 6 lines
Quoted 2 lines

What if you have a mounted file system on a pendrive ? Should we allow suspend in that case ?

Well, I would have expected that, but in fact, the opposite is true. With 3.12, mounting a filesystem on a USB media and accessing it after resume was exactly my test case, and it worked just fine with that patch.

so you had a mounted file system, suspend, resume, it was still mounted?

Yes, exactly.

Quoted 1 line

or did it reenumerate and you remounted it ?

No, it did not reenumerate, but the core did a reset on it. See the following log.

...

Quoted 2 lines
[   32.030792] usb 1-1: reset high-speed USB device number 2 using

musb-hdrc <<<< !

Like I said, you get a reset-resume.

Quoted 2 lines
Quoted 2 lines

Try to do the same with transfers in flight, it's likely to corrupt your file system.

It's not so easy to suspend a system with USB mass-storage transfers in flight. For one thing, userspace gets frozen before the suspend starts, so there's nothing to initiate new I/O requests (except perhaps for dirty-block writebacks).

Also, the child SCSI device gets suspended before the USB device, so its I/O queue stops running. In addition, the usb-storage suspend routine and the main I/O thread are mutually exclusive (they both grab the private mutex), so a suspend can't occur while a SCSI command is underway.

Alan Stern

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

#21 lore raw

Re: [PATCH 2/5] usb: musb: call musb_port_suspend from musb_bus_suspend

In reply to a message not in the archive
Quoted 15 lines · Alan Stern
Quoted 13 lines · Daniel Mack
Quoted 8 lines
Quoted 6 lines
Quoted 2 lines

What if you have a mounted file system on a pendrive ? Should we allow suspend in that case ?

Well, I would have expected that, but in fact, the opposite is true. With 3.12, mounting a filesystem on a USB media and accessing it after resume was exactly my test case, and it worked just fine with that patch.

so you had a mounted file system, suspend, resume, it was still mounted?

Yes, exactly.

Quoted 1 line

or did it reenumerate and you remounted it ?

Quoted 8 lines
Quoted 2 lines

No, it did not reenumerate, but the core did a reset on it. See the following log.

...

Quoted 2 lines
[   32.030792] usb 1-1: reset high-speed USB device number 2 using

musb-hdrc <<<< !

Like I said, you get a reset-resume.

Yup. We can't avoid losing VBUS on this platform, and even if we could, we wouldn't want that, in order to save as much power as possible.

Quoted 8 lines
Quoted 2 lines
Quoted 2 lines

Try to do the same with transfers in flight, it's likely to corrupt your file system.

It's not so easy to suspend a system with USB mass-storage transfers in flight. For one thing, userspace gets frozen before the suspend starts, so there's nothing to initiate new I/O requests (except perhaps for dirty-block writebacks).

Quoted 5 lines

Also, the child SCSI device gets suspended before the USB device, so its I/O queue stops running. In addition, the usb-storage suspend routine and the main I/O thread are mutually exclusive (they both grab the private mutex), so a suspend can't occur while a SCSI command is underway.

Thanks Alan for the explanation. That was my understanding as well.

Best regards,
Daniel

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

#22 lore raw

Re: [PATCH 2/5] usb: musb: call musb_port_suspend from musb_bus_suspend

In reply to a message not in the archive

Hi,

Quoted 32 lines · Alan Stern
Quoted 15 lines · Daniel Mack
Quoted 8 lines
Quoted 6 lines
Quoted 2 lines

What if you have a mounted file system on a pendrive ? Should we allow suspend in that case ?

Well, I would have expected that, but in fact, the opposite is true. With 3.12, mounting a filesystem on a USB media and accessing it after resume was exactly my test case, and it worked just fine with that patch.

so you had a mounted file system, suspend, resume, it was still mounted?

Yes, exactly.

Quoted 1 line

or did it reenumerate and you remounted it ?

No, it did not reenumerate, but the core did a reset on it. See the following log.

...

Quoted 2 lines
[   32.030792] usb 1-1: reset high-speed USB device number 2 using

musb-hdrc <<<< !

Like I said, you get a reset-resume.

Quoted 2 lines
Quoted 2 lines

Try to do the same with transfers in flight, it's likely to corrupt your file system.

It's not so easy to suspend a system with USB mass-storage transfers in flight. For one thing, userspace gets frozen before the suspend starts, so there's nothing to initiate new I/O requests (except perhaps for dirty-block writebacks).

Quoted 5 lines

Also, the child SCSI device gets suspended before the USB device, so its I/O queue stops running. In addition, the usb-storage suspend routine and the main I/O thread are mutually exclusive (they both grab the private mutex), so a suspend can't occur while a SCSI command is underway.

awesome, thanks for the explanation Alan :-)

Signature · 1 line
balbi
#24 lore raw

Re: [PATCH 2/5] usb: musb: call musb_port_suspend from musb_bus_suspend

In reply to a message not in the archive
Quoted 6 lines · Felipe Balbi
Quoted 2 lines · Daniel Mack

In 3.13, something about the parition table reading seems to be broken currently, I'll have a closer look.

cool

Ah, and forget about this one. I didn't look close enough. There's no such issue in 3.13. Sorry for the noise.

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

#35 lore raw

Re: [PATCH 4/5] usb: musb: dsps: add support for suspend and resume

In reply to Daniel Mack

Hi,

Quoted 41 lines · Daniel Mack
Quoted 20 lines · Felipe Balbi
Quoted 19 lines · Daniel Mack
Quoted 8 lines
Quoted 8 lines
drivers/usb/musb/musb_virthub.c +0 -0
index e977441..24e46c0 100644
@@ -109,7 +109,7 @@ void musb_port_suspend(struct musb *musb, bool do_suspend)
 	}
 }
 
Quoted 6 lines

drivers/usb/musb/musb_virthub.c void musb_port_suspend(struct musb *musb, bool do_suspend) · line 112

Quoted 2 lines
drivers/usb/musb/musb_virthub.c +1 -1
-static void musb_port_reset(struct musb *musb, bool do_reset)
+void musb_port_reset(struct musb *musb, bool do_reset)

NAK, this should not be called from the glue layers at all. If anything call from musb_core's resume callback. That will only be called after parent's resume has been called anyway.

Given that this driver is successfully used with suspend/resume on other platforms without that reset, but it's clearly needed for dsps, I'm fairly sure this should be a glue-layer specific thing. Hence the change.

Quoted 6 lines
Quoted 2 lines

I think it'll break things on other platforms if we unconditionally reset the port after resume, but I can't test it. Anyone?

your original commit log only says "we need to issue port reset" but it never explains why. It could very well be you just found a device which needs to be reset when coming out of suspend and it misses a quirk flag ?

I think I can really rule that out, as I tested with multiple USB sticks and also tested the same sticks on other embedded platforms.

Quoted 4 lines

In any case, those functions should never be called by the glue, if reset needs to be called, it must be called by musb-hdrc.ko, if you need a flag, pass a flag but I need a really good explanation in your commit log of why that's necessary.

Well, if I'd only knew exactly why. All these patches are really the result of many days of trial and error, with multiple drivers (musb, cppi41, ...) involved. And as I said - I have no documentation of the musb core itself.

Quoted 4 lines

So yes, I can do it the other way around and pass a flag, but what I can't provide is a good explanantion why the dsps glue behaves differently here than others. I'm curious myself, and all I know is that with this reset in place, things work as expected on AM33xx.

ok, then let's pass a flag. Meanwhile I'll try to figure out internally why we need that reset. As Alan said, usb-persist should already for a bus-reset when resuming, right ?

Signature · 1 line
balbi
#26 lore raw

Re: [PATCH 2/5] usb: musb: call musb_port_suspend from musb_bus_suspend

In reply to a message not in the archive
Quoted 10 lines · Daniel Mack
Quoted 6 lines · Felipe Balbi
Quoted 2 lines · Daniel Mack

In 3.13, something about the parition table reading seems to be broken currently, I'll have a closer look.

cool

Ah, and forget about this one. I didn't look close enough. There's no such issue in 3.13. Sorry for the noise.

no problem :-)

Signature · 1 line
balbi
#36 lore raw

Re: [PATCH 4/5] usb: musb: dsps: add support for suspend and resume

In reply to Felipe Balbi
Quoted 44 lines · Felipe Balbi

Hi,

Quoted 41 lines · Daniel Mack
Quoted 20 lines · Felipe Balbi
Quoted 19 lines · Daniel Mack
Quoted 8 lines
Quoted 8 lines
drivers/usb/musb/musb_virthub.c +0 -0
index e977441..24e46c0 100644
@@ -109,7 +109,7 @@ void musb_port_suspend(struct musb *musb, bool do_suspend)
 	}
 }
 
Quoted 6 lines

drivers/usb/musb/musb_virthub.c void musb_port_suspend(struct musb *musb, bool do_suspend) · line 112

Quoted 2 lines
drivers/usb/musb/musb_virthub.c +1 -1
-static void musb_port_reset(struct musb *musb, bool do_reset)
+void musb_port_reset(struct musb *musb, bool do_reset)

NAK, this should not be called from the glue layers at all. If anything call from musb_core's resume callback. That will only be called after parent's resume has been called anyway.

Given that this driver is successfully used with suspend/resume on other platforms without that reset, but it's clearly needed for dsps, I'm fairly sure this should be a glue-layer specific thing. Hence the change.

Quoted 6 lines
Quoted 2 lines

I think it'll break things on other platforms if we unconditionally reset the port after resume, but I can't test it. Anyone?

your original commit log only says "we need to issue port reset" but it never explains why. It could very well be you just found a device which needs to be reset when coming out of suspend and it misses a quirk flag ?

I think I can really rule that out, as I tested with multiple USB sticks and also tested the same sticks on other embedded platforms.

Quoted 4 lines

In any case, those functions should never be called by the glue, if reset needs to be called, it must be called by musb-hdrc.ko, if you need a flag, pass a flag but I need a really good explanation in your commit log of why that's necessary.

Well, if I'd only knew exactly why. All these patches are really the result of many days of trial and error, with multiple drivers (musb, cppi41, ...) involved. And as I said - I have no documentation of the musb core itself.

Quoted 7 lines
Quoted 4 lines

So yes, I can do it the other way around and pass a flag, but what I can't provide is a good explanantion why the dsps glue behaves differently here than others. I'm curious myself, and all I know is that with this reset in place, things work as expected on AM33xx.

ok, then let's pass a flag. Meanwhile I'll try to figure out internally why we need that reset.

Well, we don't need a reset, I wasn't quite precise here. What we need is to manually *de*assert the reset when we resume. Note the 'false' flag that is passed to musb_port_reset().

Quoted 2 lines

As Alan said, usb-persist should already for a
bus-reset when resuming, right ?

The explicit un-reset is really needed, otherwise the port will reenumerate the device, and a previously mounted filesystem will become invalid of course. The log looks like this in that case:

[   17.045641] PM: resume of devices complete after 166.084 msecs
[   17.056461] PM: Sending message for resetting M3 state machine
[   17.063451] Restarting tasks ... [   17.071767] usb 1-1: USB
disconnect, device number 2
done.
[   17.403402] usb 1-1: new high-speed USB device number 3 using musb-hdrc
[   17.766959] usb 1-1: New USB device found, idVendor=058f, idProduct=6387
[   17.774849] usb 1-1: New USB device strings: Mfr=1, Product=2,

SerialNumber=3

Thanks,
Daniel

#37 lore raw

Re: [PATCH 4/5] usb: musb: dsps: add support for suspend and resume

Hi,

Quoted 15 lines · Daniel Mack
Quoted 2 lines

As Alan said, usb-persist should already for a
bus-reset when resuming, right ?

The explicit un-reset is really needed, otherwise the port will reenumerate the device, and a previously mounted filesystem will become invalid of course. The log looks like this in that case:

[   17.045641] PM: resume of devices complete after 166.084 msecs
[   17.056461] PM: Sending message for resetting M3 state machine
[   17.063451] Restarting tasks ... [   17.071767] usb 1-1: USB
disconnect, device number 2
done.
[   17.403402] usb 1-1: new high-speed USB device number 3 using musb-hdrc
[   17.766959] usb 1-1: New USB device found, idVendor=058f, idProduct=6387
[   17.774849] usb 1-1: New USB device strings: Mfr=1, Product=2,
Quoted 1 line

SerialNumber=3

got it now ;-) Thanks

Signature · 1 line
balbi
#9 lore raw

Re: [PATCH 1/5] usb: musb: conditionally save and restore the context on suspend

In reply to Daniel Mack

Hello.

Quoted 8 lines · Daniel Mack

It appears not all platforms featuring a musb core need to save the musb core registers at suspend time and restore them on resume.

The dsps platform does, however. So add a bit in struct musb_hdrc_platform_data to let platforms specify their need of such action being taken.

[...]

include/linux/usb/musb.h struct musb_hdrc_platform_data { · line 104

Quoted 11 lines
include/linux/usb/musb.h +3 -0
index eb50525..e5a581c 100644
@@ -99,6 +99,9 @@ struct musb_hdrc_platform_data {
  	/* MUSB_HOST, MUSB_PERIPHERAL, or MUSB_OTG */
  	u8		mode;
 
+	/* should the musb core restore registers after suspend? */
+	u8		restore_after_suspend:1;
+

Better placement seems to be with 'extvbus' field which is also 1-bit --
that would save space in the structure.

WBR, Sergei

#10 lore raw

Re: [PATCH 1/5] usb: musb: conditionally save and restore the context on suspend

Quoted 20 lines · Sergei Shtylyov
Quoted 8 lines · Daniel Mack

It appears not all platforms featuring a musb core need to save the musb core registers at suspend time and restore them on resume.

The dsps platform does, however. So add a bit in struct musb_hdrc_platform_data to let platforms specify their need of such action being taken.

[...]

Quoted 8 lines
include/linux/usb/musb.h +0 -0
index eb50525..e5a581c 100644
@@ -99,6 +99,9 @@ struct musb_hdrc_platform_data {
  	/* MUSB_HOST, MUSB_PERIPHERAL, or MUSB_OTG */
  	u8		mode;
 
Quoted 6 lines

include/linux/usb/musb.h struct musb_hdrc_platform_data { · line 104

Quoted 3 lines
include/linux/usb/musb.h +3 -0
+	/* should the musb core restore registers after suspend? */
+	u8		restore_after_suspend:1;
+

Better placement seems to be with 'extvbus' field which is also 1-bit --
that would save space in the structure.

That patch is deprecated, as things are now done unconditionally, without consulting this flag.

Daniel