Comment 238 for bug 1667750

you could try to flush the endpoint ringing PCI write, and see if it helps
starting the endpint, but I don't have high hopes for this, a PCI write should
be flushed anyway, especially in 25 seconds.

maybe also add trace to re-read the endpoint state after flushing pci write:
(untested)

diff --git a/drivers/usb/host/xhci-ring.c b/drivers/usb/host/xhci-ring.c
index e74518e7de6a..20e209b64551 100644
--- a/drivers/usb/host/xhci-ring.c
+++ b/drivers/usb/host/xhci-ring.c
@@ -408,6 +408,7 @@ void xhci_ring_ep_doorbell(struct xhci_hcd *xhci,
        trace_xhci_ring_ep_doorbell(slot_id, DB_VALUE(ep_index, stream_id));

        writel(DB_VALUE(ep_index, stream_id), db_addr);
+ readl(db_addr);
        /* The CPU has better things to do at this point than wait for a
         * write-posting flush. It'll get there soon enough.
         */
@@ -1176,6 +1177,8 @@ static void xhci_handle_cmd_reset_ep(struct xhci_hcd *xhci, int slot_id,
        /* if this was a soft reset, then restart */
        if ((le32_to_cpu(trb->generic.field[3])) & TRB_TSP)
                ring_doorbell_for_active_rings(xhci, slot_id, ep_index);
+
+ trace_xhci_handle_cmd_reset_ep(ep_ctx);
 }