USB: musb: fix late external abort on suspend

The musb delayed irq work was never flushed on suspend, something which since 4.9 can lead to an external abort if the work is scheduled after the grandparent's clock has been disabled:

PM: Suspending system (mem)
PM: suspend of devices complete after 125.224 msecs
PM: suspend devices took 0.132 seconds
PM: late suspend of devices complete after 7.423 msecs
PM: noirq suspend of devices complete after 7.083 msecs
suspend debug: Waiting for 5 second(s).
Unhandled fault: external abort on non-linefetch (0x1008) at 0xd0262c60
...
[<c054880c>] (musb_default_readb) from [<c0547b5c>] (musb_irq_work+0x48/0x220) [<c0547b5c>] (musb_irq_work) from [<c014f8a4>] (process_one_work+0x1f4/0x758) [<c014f8a4>] (process_one_work) from [<c014fe5c>] (worker_thread+0x54/0x514) [<c014fe5c>] (worker_thread) from [<c015704c>] (kthread+0x128/0x158) [<c015704c>] (kthread) from [<c0109330>] (ret_from_fork+0x14/0x24)

Commit 2bff3916fda9 ("usb: musb: Fix PM for hub disconnect") started scheduling musb_irq_work with a delay of up to a second and with retries thereby making this easy to trigger, for example, by suspending shortly after a disconnect.

Note that the session flag is cleared and runtime-PM usage count balanced after having cancelled the irq work instead of always rescheduling on resume as the latter would have broken PM for host mode, which always has the session bit set.

---
 drivers/usb/musb/musb_core.c | 12 +++++++++++-
 1 file changed, 11 insertions(+), 1 deletion(-)

index b67692857daf..8645432337cb 100644
@@ -2681,8 +2681,18 @@ static int musb_suspend(struct device *dev)
 
 	musb_platform_disable(musb);
 	musb_disable_interrupts(musb);
-	if (!(musb->io.quirks & MUSB_PRESERVE_SESSION))
+
+	cancel_delayed_work_sync(&musb->irq_work);
+
+	if (!(musb->io.quirks & MUSB_PRESERVE_SESSION)) {
 		musb_writeb(musb->mregs, MUSB_DEVCTL, 0);
+
+		if (musb->session) {
+			musb->session = false;
+			pm_runtime_put_noidle(musb->controller);
+		}
+	}
+
 	WARN_ON(!list_empty(&musb->pending_list));
 
 	spin_lock_irqsave(&musb->lock, flags);

2.14.1

The current session-bit quirk implementation does not prevent the retry counter from underflowing, something which could break runtime PM and keep the device active for a very long time (about 2^32 seconds) after a disconnect.

This notably breaks the B-device timeout case, but could potentially cause problems also when the controller is operating as an A-device.

---
 drivers/usb/musb/musb_core.c | 8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)

index 8645432337cb..8c40fc555eb0 100644
@@ -1861,22 +1861,22 @@ static void musb_pm_runtime_check_session(struct musb *musb)
 		MUSB_DEVCTL_HR;
 	switch (devctl & ~s) {
 	case MUSB_QUIRK_B_INVALID_VBUS_91:
-		if (musb->quirk_retries--) {
+		if (musb->quirk_retries) {
 			musb_dbg(musb,
 				 "Poll devctl on invalid vbus, assume no session");
 			schedule_delayed_work(&musb->irq_work,
 					      msecs_to_jiffies(1000));
-
+			musb->quirk_retries--;
 			return;
 		}
 		/* fall through */
 	case MUSB_QUIRK_A_DISCONNECT_19:
-		if (musb->quirk_retries--) {
+		if (musb->quirk_retries) {
 			musb_dbg(musb,
 				 "Poll devctl on possible host mode disconnect");
 			schedule_delayed_work(&musb->irq_work,
 					      msecs_to_jiffies(1000));
-
+			musb->quirk_retries--;
 			return;
 		}
 		if (!musb->session)

2.14.1