On 17 June 2011 18:27, Alan Stern <stern@xxxxxxxxxxxxxxxxxxx> wrote:
> On Wed, 15 Jun 2011, Matej Kenda wrote:
>
>> Patched Linux kernel is now being used on four systems (COMe type) for
>> almost two weeks. USB disk where the complete OS is installed did not
>> unexpectedly disconnect on any of these embedded devices.
>>
>> It seems that the your patch solves the problem.
>
> I've got a new version of this patch for you to test. In some respects
> it simplifies things compared to the old patch. (It looks worse than
> it really is because as part of the simplification, a whole chunk of
> code got moved out one indentation level.)
>
> Let me know what happens with it. I think this will apply to 3.0-rc3,
> and it should apply to earlier kernels with only minor adjustments.
> (Fair warning: I haven't tried this on my own system yet.)
The patch was adapted for 3.6.38 (Ubuntu Natty).
I will install patched kernel and run it for some time.
Please find attached modified patch that applies to 2.6.38.
Regards,
Matej
diff --git a/debian.master/changelog b/debian.master/changelog
index 6a9f46e..65549e0 100644
--- a/debian.master/changelog
+++ b/debian.master/changelog
@@ -1,3 +1,12 @@
+linux (2.6.38-10.299) natty; urgency=low
+
+ [ Alan Stern ]
+
+ * Patch for unwanted USB disconnect
+ - LP: #701011
+
+ -- Matej Kenda <matejken@xxxxxxxxx> Mon, 20 Jun 2011 09:50:22 +0200
+
linux (2.6.38-10.44) natty-proposed; urgency=low
[ Steve Conklin ]
diff --git a/drivers/usb/host/ehci-hcd.c b/drivers/usb/host/ehci-hcd.c
index 74dcf49..65a40e9 100644
--- a/drivers/usb/host/ehci-hcd.c
+++ b/drivers/usb/host/ehci-hcd.c
@@ -90,7 +90,8 @@ static const char hcd_name [] = "ehci_hcd";
#define EHCI_IAA_MSECS 10 /* arbitrary */
#define EHCI_IO_JIFFIES (HZ/10) /* io watchdog > irq_thresh */
#define EHCI_ASYNC_JIFFIES (HZ/20) /* async idle timeout */
-#define EHCI_SHRINK_FRAMES 5 /* async qh unlink delay */
+#define EHCI_SHRINK_JIFFIES (DIV_ROUND_UP(HZ, 200) + 1)
+ /* async qh unlink delay */
/* Initial IRQ latency: faster than hw default */
static int log2_irq_thresh = 0; // 0 to 6
@@ -150,10 +151,7 @@ timer_action(struct ehci_hcd *ehci, enum ehci_timer_action action)
break;
/* case TIMER_ASYNC_SHRINK: */
default:
- /* add a jiffie since we synch against the
- * 8 KHz uframe counter.
- */
- t = DIV_ROUND_UP(EHCI_SHRINK_FRAMES * HZ, 1000) + 1;
+ t = EHCI_SHRINK_JIFFIES;
break;
}
mod_timer(&ehci->watchdog, t + jiffies);
diff --git a/drivers/usb/host/ehci-q.c b/drivers/usb/host/ehci-q.c
index baf7362..8be6351 100644
--- a/drivers/usb/host/ehci-q.c
+++ b/drivers/usb/host/ehci-q.c
@@ -988,6 +988,8 @@ static void qh_link_async (struct ehci_hcd *ehci, struct ehci_qh *qh)
/* splice right after start */
qh->qh_next = head->qh_next;
qh->hw->hw_next = head->hw->hw_next;
+ if (ehci->qh_scan_next == qh)
+ ehci->qh_scan_next = qh->qh_next.qh;
wmb ();
head->qh_next.qh = qh;
@@ -1249,53 +1251,48 @@ static void scan_async (struct ehci_hcd *ehci)
struct ehci_qh *qh;
enum ehci_timer_action action = TIMER_IO_WATCHDOG;
- ehci->stamp = ehci_readl(ehci, &ehci->regs->frame_index);
timer_action_done (ehci, TIMER_ASYNC_SHRINK);
-rescan:
stopped = !HC_IS_RUNNING(ehci_to_hcd(ehci)->state);
- qh = ehci->async->qh_next.qh;
- if (likely (qh != NULL)) {
- do {
- /* clean any finished work for this qh */
- if (!list_empty(&qh->qtd_list) && (stopped ||
- qh->stamp != ehci->stamp)) {
- int temp;
-
- /* unlinks could happen here; completion
- * reporting drops the lock. rescan using
- * the latest schedule, but don't rescan
- * qhs we already finished (no looping)
- * unless the controller is stopped.
- */
- qh = qh_get (qh);
- qh->stamp = ehci->stamp;
- temp = qh_completions (ehci, qh);
- if (qh->needs_rescan)
- unlink_async(ehci, qh);
- qh_put (qh);
- if (temp != 0) {
- goto rescan;
- }
- }
-
- /* unlink idle entries, reducing DMA usage as well
- * as HCD schedule-scanning costs. delay for any qh
- * we just scanned, there's a not-unusual case that it
- * doesn't stay idle for long.
- * (plus, avoids some kind of re-activation race.)
- */
- if (list_empty(&qh->qtd_list)
- && qh->qh_state == QH_STATE_LINKED) {
- if (!ehci->reclaim && (stopped ||
- ((ehci->stamp - qh->stamp) & 0x1fff)
- >= EHCI_SHRINK_FRAMES * 8))
- start_unlink_async(ehci, qh);
- else
- action = TIMER_ASYNC_SHRINK;
- }
+ ehci->qh_scan_next = ehci->async->qh_next.qh;
+ while (ehci->qh_scan_next) {
+ qh = ehci->qh_scan_next;
+ ehci->qh_scan_next = qh->qh_next.qh;
+rescan:
+ /* clean any finished work for this qh */
+ if (!list_empty(&qh->qtd_list)) {
+ int temp;
+
+ /*
+ * Unlinks could happen here; completion reporting
+ * drops the lock. That's why ehci->qh_scan_next
+ * always holds the next qh to scan; if the next qh
+ * gets unlinked then ehci->qh_scan_next is adjusted
+ * in start_unlink_async().
+ */
+ qh = qh_get(qh);
+ temp = qh_completions(ehci, qh);
+ if (qh->needs_rescan)
+ unlink_async(ehci, qh);
+ qh->unlink_time = jiffies + EHCI_SHRINK_JIFFIES;
+ qh_put(qh);
+ if (temp != 0)
+ goto rescan;
+ }
- qh = qh->qh_next.qh;
- } while (qh);
+ /* unlink idle entries, reducing DMA usage as well
+ * as HCD schedule-scanning costs. delay for any qh
+ * we just scanned, there's a not-unusual case that it
+ * doesn't stay idle for long.
+ * (plus, avoids some kind of re-activation race.)
+ */
+ if (list_empty(&qh->qtd_list)
+ && qh->qh_state == QH_STATE_LINKED) {
+ if (!ehci->reclaim && (stopped ||
+ time_after_eq(jiffies, qh->unlink_time)))
+ start_unlink_async(ehci, qh);
+ else
+ action = TIMER_ASYNC_SHRINK;
+ }
}
if (action == TIMER_ASYNC_SHRINK)
timer_action (ehci, TIMER_ASYNC_SHRINK);
diff --git a/drivers/usb/host/ehci.h b/drivers/usb/host/ehci.h
index 799ac16..e0e5472 100644
--- a/drivers/usb/host/ehci.h
+++ b/drivers/usb/host/ehci.h
@@ -75,6 +75,7 @@ struct ehci_hcd { /* one per controller */
struct ehci_qh *async;
struct ehci_qh *dummy; /* For AMD quirk use */
struct ehci_qh *reclaim;
+ struct ehci_qh *qh_scan_next;
unsigned scanning : 1;
/* periodic schedule support */
@@ -336,6 +337,7 @@ struct ehci_qh {
struct ehci_qh *reclaim; /* next to reclaim */
struct ehci_hcd *ehci;
+ unsigned long unlink_time;
/*
* Do NOT use atomic operations for QH refcounting. On some CPUs
