Skip to contentSkip to conversation

xhci-hub.c: handle command_trb that may be link TRB

PATCH In review Xiao Jin 2 participants 3 messages Oct 9, 2013 – Oct 10, 2013

Subsystems (2)
lore
Share & tools
lore

Xiao Jin

author
#0 lore raw

PATCH 1/1 xhci-hub.c: handle command_trb that may be link TRB

From: xiao jin <[email protected]>
Date: Wed, 9 Oct 2013 09:38:45 +0800
Subject: [PATCH] xhci-hub.c: handle command_trb that may be link TRB

When xhci stop device, it's possible cmd_ring enqueue point to link TRB after queue the last but one stop endpoint. We must handle the command_trb point to the next segment trb. Otherwise xhci stop devie will timeout because command_trb can't match with cmd_ring dequeue.

The patch is to let command_trb point to the next segment trb if cmd_ring enqueue point to link TRB.

---
 drivers/usb/host/xhci-hub.c |    7 +++++++
 1 files changed, 7 insertions(+), 0 deletions(-)
drivers/usb/host/xhci-hub.c +7 -0
index 1d35459..4872640 100644
@@ -287,6 +287,13 @@ static int xhci_stop_device(struct xhci_hcd *xhci, int slot_id, int suspend)
 			xhci_queue_stop_endpoint(xhci, slot_id, i, suspend);
 	}
 	cmd->command_trb = xhci->cmd_ring->enqueue;
+	/* Enqueue pointer can be left pointing to the link TRB,
+	 * we must handle that
+	 */
+	if (TRB_TYPE_LINK_LE32(cmd->command_trb->link.control))
+		cmd->command_trb =
+			xhci->cmd_ring->enq_seg->next->trbs;
+
 	list_add_tail(&cmd->cmd_list, &virt_dev->cmd_list);
 	xhci_queue_stop_endpoint(xhci, slot_id, 0, suspend);
 	xhci_ring_cmd_db(xhci);
Signature · 1 line
1.7.1
#1 lore raw

Hi Xiao,

Thanks for taking the time to submit this patch. Comments below.

Quoted 3 lines · Xiao Jin

From: xiao jin <[email protected]>
Date: Wed, 9 Oct 2013 09:38:45 +0800
Subject: [PATCH] xhci-hub.c: handle command_trb that may be link TRB

I won't be able to apply your patch because it has these extra lines in the body ^^^. I suspect you used git format-patch to produce this, and then tried to copy-paste it into your mail client? You can't do that, because your mail client will probably word-wrap your patch, and possibly turn tabs into spaces. You need to use git send-email instead to send your patches.

Quoted 22 lines

When xhci stop device, it's possible cmd_ring enqueue point to link TRB after queue the last but one stop endpoint. We must handle the command_trb point to the next segment trb. Otherwise xhci stop devie will timeout because command_trb can't match with cmd_ring dequeue.

The patch is to let command_trb point to the next segment trb if cmd_ring enqueue point to link TRB.

---
 drivers/usb/host/xhci-hub.c |    7 +++++++
 1 files changed, 7 insertions(+), 0 deletions(-)
drivers/usb/host/xhci-hub.c +0 -0
index 1d35459..4872640 100644
@@ -287,6 +287,13 @@ static int xhci_stop_device(struct xhci_hcd *xhci, int slot_id, int suspend)
 			xhci_queue_stop_endpoint(xhci, slot_id, i, suspend);
 	}
 	cmd->command_trb = xhci->cmd_ring->enqueue;

drivers/usb/host/xhci-hub.c static int xhci_stop_device(struct xhci_hcd *xhci, int slot_id, int suspend) · line 299

Quoted 10 lines
drivers/usb/host/xhci-hub.c +7 -0
+	/* Enqueue pointer can be left pointing to the link TRB,
+	 * we must handle that
+	 */
+	if (TRB_TYPE_LINK_LE32(cmd->command_trb->link.control))
+		cmd->command_trb =
+			xhci->cmd_ring->enq_seg->next->trbs;
+
 	list_add_tail(&cmd->cmd_list, &virt_dev->cmd_list);
 	xhci_queue_stop_endpoint(xhci, slot_id, 0, suspend);
 	xhci_ring_cmd_db(xhci);

What kernel version or tree is it against? I ask because there's already a fix in the 3.12-rc4 kernel for this:

static int xhci_stop_device(struct xhci_hcd *xhci, int slot_id, int suspend) {
...

        spin_lock_irqsave(&xhci->lock, flags);
        for (i = LAST_EP_INDEX; i > 0; i--) {
                if (virt_dev->eps[i].ring && virt_dev->eps[i].ring->dequeue)
                        xhci_queue_stop_endpoint(xhci, slot_id, i, suspend);
        }
        cmd->command_trb = xhci_find_next_enqueue(xhci->cmd_ring);

...

Where xhci_find_next_enqueue is defined as:

union xhci_trb *xhci_find_next_enqueue(struct xhci_ring *ring)

{
        /* Enqueue pointer can be left pointing to the link TRB,
         * we must handle that
         */
        if (TRB_TYPE_LINK_LE32(ring->enqueue->link.control))
                return ring->enq_seg->next->trbs;
        return ring->enqueue;
}

That was added in commit ec7e43e2d98173483866fe2e4e690143626b659c "xhci:
Ensure a command structure points to the correct trb on the command ring" It's in (or should be in soon) the stable kernels as well.

Sarah Sharp

author
#2 lore raw

Sarah,

Thanks for the reminding. The kernel base of the patch is k3.10, I didn't notice k3.12 commit ec7e43e2d98173483866fe2e4e690143626b659c at that time. I will backport the patch from k3.12 for use.

As you mentioned in another email, we meet with dpm timeout when xhci suspend, the root cause are what the patch aimed to solve. FYI.

Quoted 80 lines · Sarah Sharp

Hi Xiao,

Thanks for taking the time to submit this patch. Comments below.

Quoted 3 lines · Xiao Jin

From: xiao jin <[email protected]>
Date: Wed, 9 Oct 2013 09:38:45 +0800
Subject: [PATCH] xhci-hub.c: handle command_trb that may be link TRB

I won't be able to apply your patch because it has these extra lines in the body ^^^. I suspect you used git format-patch to produce this, and then tried to copy-paste it into your mail client? You can't do that, because your mail client will probably word-wrap your patch, and possibly turn tabs into spaces. You need to use git send-email instead to send your patches.

Quoted 22 lines

When xhci stop device, it's possible cmd_ring enqueue point to link TRB after queue the last but one stop endpoint. We must handle the command_trb point to the next segment trb. Otherwise xhci stop devie will timeout because command_trb can't match with cmd_ring dequeue.

The patch is to let command_trb point to the next segment trb if cmd_ring enqueue point to link TRB.

---
 drivers/usb/host/xhci-hub.c |    7 +++++++
 1 files changed, 7 insertions(+), 0 deletions(-)
drivers/usb/host/xhci-hub.c +0 -0
index 1d35459..4872640 100644
@@ -287,6 +287,13 @@ static int xhci_stop_device(struct xhci_hcd *xhci, int slot_id, int suspend)
 			xhci_queue_stop_endpoint(xhci, slot_id, i, suspend);
 	}
 	cmd->command_trb = xhci->cmd_ring->enqueue;

drivers/usb/host/xhci-hub.c static int xhci_stop_device(struct xhci_hcd *xhci, int slot_id, int suspend) · line 299

Quoted 10 lines
drivers/usb/host/xhci-hub.c +7 -0
+	/* Enqueue pointer can be left pointing to the link TRB,
+	 * we must handle that
+	 */
+	if (TRB_TYPE_LINK_LE32(cmd->command_trb->link.control))
+		cmd->command_trb =
+			xhci->cmd_ring->enq_seg->next->trbs;
+
 	list_add_tail(&cmd->cmd_list, &virt_dev->cmd_list);
 	xhci_queue_stop_endpoint(xhci, slot_id, 0, suspend);
 	xhci_ring_cmd_db(xhci);

What kernel version or tree is it against? I ask because there's already a fix in the 3.12-rc4 kernel for this:

static int xhci_stop_device(struct xhci_hcd *xhci, int slot_id, int suspend) {
...

        spin_lock_irqsave(&xhci->lock, flags);
        for (i = LAST_EP_INDEX; i > 0; i--) {
                if (virt_dev->eps[i].ring && virt_dev->eps[i].ring->dequeue)
                        xhci_queue_stop_endpoint(xhci, slot_id, i, suspend);
        }
        cmd->command_trb = xhci_find_next_enqueue(xhci->cmd_ring);

...

Where xhci_find_next_enqueue is defined as:

union xhci_trb *xhci_find_next_enqueue(struct xhci_ring *ring)

{
        /* Enqueue pointer can be left pointing to the link TRB,
         * we must handle that
         */
        if (TRB_TYPE_LINK_LE32(ring->enqueue->link.control))
                return ring->enq_seg->next->trbs;
        return ring->enqueue;
}

That was added in commit ec7e43e2d98173483866fe2e4e690143626b659c "xhci:
Ensure a command structure points to the correct trb on the command ring" It's in (or should be in soon) the stable kernels as well.

Sarah Sharp