Without this condition, all controllers will do this delay,
and increase the resume time.
Only enabled and unsuspended port needs this delay, but
Some buggy hardware(like Synopsys usb controller) will
clear suspend bit once they receive/send resume signal,
so it takes resume bit as consideration.
Tested it at Freescale i.mx6q Sabrelite board.
Signed-off-by: Peter Chen <peter.chen@xxxxxxxxxxxxx>
---
drivers/usb/host/ehci-hub.c | 22 +++++++++++++++++-----
1 files changed, 17 insertions(+), 5 deletions(-)
diff --git a/drivers/usb/host/ehci-hub.c b/drivers/usb/host/ehci-hub.c
index c788022..6505063 100644
--- a/drivers/usb/host/ehci-hub.c
+++ b/drivers/usb/host/ehci-hub.c
@@ -384,11 +384,23 @@ static int ehci_bus_resume (struct usb_hcd *hcd)
ehci_writel(ehci, ehci->command, &ehci->regs->command);
ehci->rh_state = EHCI_RH_RUNNING;
- /* Some controller/firmware combinations need a delay during which
- * they set up the port statuses. See Bugzilla #8190. */
- spin_unlock_irq(&ehci->lock);
- msleep(8);
- spin_lock_irq(&ehci->lock);
+ /*
+ * According to Bugzilla #8190, the port status for some controllers
+ * will be wrong without a delay. At their wrong status, the port
+ * is enabled, but not suspended neither resumed.
+ */
+ i = HCS_N_PORTS (ehci->hcs_params);
+ while (i--) {
+ temp = ehci_readl(ehci, &ehci->regs->port_status [i]);
+ if ((temp & PORT_PE) &&
+ !(temp & (PORT_SUSPEND | PORT_RESUME))) {
+ ehci_warn(ehci, "Port status(0x%x) is error\n", temp);
+ spin_unlock_irq(&ehci->lock);
+ msleep(8);
+ spin_lock_irq(&ehci->lock);
+ }
+ }
+
if (ehci->shutdown)
goto shutdown;
--
1.7.0.4
--
To unsubscribe from this list: send the line "unsubscribe linux-usb" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
