EHCI: don't try to clear the IAAD bit
authorAlan Stern <stern@rowland.harvard.edu>
Tue, 17 Apr 2012 19:23:25 +0000 (15:23 -0400)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Wed, 18 Apr 2012 20:55:40 +0000 (13:55 -0700)
This patch (as1541) corrects a small mistake in ehci-hcd.  The IAAD
(Interrupt on Async Advance Doorbell) bit in the USBCMD register is
designed, as its name says, to act as a "doorbell".  That is, the
driver activates the bit by setting it to 1, and the hardware
deactivates it later by setting it back to 0.  The driver cannot clear
the bit by writing a 0 to it; such writes are simply ignored.

Therefore there is no reason for ehci-hcd to try to clear the bit.
The patch removes the two instances where such attempts occur.

Signed-off-by: Alan Stern <stern@rowland.harvard.edu>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
drivers/usb/host/ehci-hcd.c

index 057cdda7a48978999ebf5ac4cbb8ecee7acc60e0..343f40c23b5ff824420940e0901d203613d92a37 100644 (file)
@@ -415,9 +415,6 @@ static void ehci_iaa_watchdog(unsigned long param)
                 * CMD_IAAD when it sets STS_IAA.)
                 */
                cmd = ehci_readl(ehci, &ehci->regs->command);
-               if (cmd & CMD_IAAD)
-                       ehci_writel(ehci, cmd & ~CMD_IAAD,
-                                       &ehci->regs->command);
 
                /* If IAA is set here it either legitimately triggered
                 * before we cleared IAAD above (but _way_ late, so we'll
@@ -887,11 +884,8 @@ static irqreturn_t ehci_irq (struct usb_hcd *hcd)
        /* complete the unlinking of some qh [4.15.2.3] */
        if (status & STS_IAA) {
                /* guard against (alleged) silicon errata */
-               if (cmd & CMD_IAAD) {
-                       ehci_writel(ehci, cmd & ~CMD_IAAD,
-                                       &ehci->regs->command);
+               if (cmd & CMD_IAAD)
                        ehci_dbg(ehci, "IAA with IAAD still set?\n");
-               }
                if (ehci->reclaim) {
                        COUNT(ehci->stats.reclaim);
                        end_unlink_async(ehci);