[RESEND PATCH v3] usb: xhci: fix lack of short packet event trb handling

classic Classic list List threaded Threaded
2 messages Options
Reply | Threaded
Open this post in threaded view
|

[RESEND PATCH v3] usb: xhci: fix lack of short packet event trb handling

Ran Wang
For bulk IN transfer, the codes will set ISP flag to request event TRB
being generated by xHC for case of short packet. So for the case of
buffer-cross-64K-boundary (which we will divide payload and enqueuqe
more than 1 transfer TRB), if the first TRB ends up with a short packet
condition it will trigger an short packet code transfer event per that
flag and cause more than 1 event TRB generated for this transfer.

However, current codes will only handle the first transfer event TRB
then mark current transfer completed, causing next transfer
failure due to event TRB mis-match.

Such issue has been observed on some Layerscape platforms (LS1028A,
LS1088A, etc) with USB ethernet device.

This patch adds a loop to make sure the event TRB for last transfer TRB
has been handled in time.

Signed-off-by: Ran Wang <[hidden email]>
---
Change in v3:
 - Update commit message according to v2 feedback .
 - Replace (void *) with (uintptr_t) to fix below armv7 compile warning:
 ...
 drivers/usb/host/xhci-ring.c: In function 'xhci_bulk_tx':
 drivers/usb/host/xhci-ring.c:726:6: warning: cast to pointer from integer of different size [-Wint-to-pointer-cast]
   726 |  if ((void *)le64_to_cpu(event->trans_event.buffer) != last_transfer_trb_addr) {
       |       ^
 ...


Change in v2:
 - Re-write commit message to describe context more clearly.
 - Add prefix 'le64_to_cpu' for 'event->trans_event.buffer'.

 drivers/usb/host/xhci-ring.c | 18 +++++++++++++++---
 1 file changed, 15 insertions(+), 3 deletions(-)

diff --git a/drivers/usb/host/xhci-ring.c b/drivers/usb/host/xhci-ring.c
index 13065d7..d708fc9 100644
--- a/drivers/usb/host/xhci-ring.c
+++ b/drivers/usb/host/xhci-ring.c
@@ -580,10 +580,13 @@ int xhci_bulk_tx(struct usb_device *udev, unsigned long pipe,
  int ret;
  u32 trb_fields[4];
  u64 val_64 = virt_to_phys(buffer);
+ void *last_transfer_trb_addr;
+ int available_length;
 
  debug("dev=%p, pipe=%lx, buffer=%p, length=%d\n",
  udev, pipe, buffer, length);
 
+ available_length = length;
  ep_index = usb_pipe_ep_index(pipe);
  virt_dev = ctrl->devs[slot_id];
 
@@ -697,7 +700,7 @@ int xhci_bulk_tx(struct usb_device *udev, unsigned long pipe,
  trb_fields[2] = length_field;
  trb_fields[3] = field | TRB_TYPE(TRB_NORMAL);
 
- queue_trb(ctrl, ring, (num_trbs > 1), trb_fields);
+ last_transfer_trb_addr = queue_trb(ctrl, ring, (num_trbs > 1), trb_fields);
 
  --num_trbs;
 
@@ -710,6 +713,7 @@ int xhci_bulk_tx(struct usb_device *udev, unsigned long pipe,
 
  giveback_first_trb(udev, ep_index, start_cycle, start_trb);
 
+again:
  event = xhci_wait_for_event(ctrl, TRB_TRANSFER);
  if (!event) {
  debug("XHCI bulk transfer timed out, aborting...\n");
@@ -718,12 +722,20 @@ int xhci_bulk_tx(struct usb_device *udev, unsigned long pipe,
  udev->act_len = 0;
  return -ETIMEDOUT;
  }
- field = le32_to_cpu(event->trans_event.flags);
 
+ if ((uintptr_t)(le64_to_cpu(event->trans_event.buffer))
+ != (uintptr_t)last_transfer_trb_addr) {
+ available_length -=
+ (int)EVENT_TRB_LEN(le32_to_cpu(event->trans_event.transfer_len));
+ xhci_acknowledge_event(ctrl);
+ goto again;
+ }
+
+ field = le32_to_cpu(event->trans_event.flags);
  BUG_ON(TRB_TO_SLOT_ID(field) != slot_id);
  BUG_ON(TRB_TO_EP_INDEX(field) != ep_index);
 
- record_transfer_result(udev, event, length);
+ record_transfer_result(udev, event, available_length);
  xhci_acknowledge_event(ctrl);
  xhci_inval_cache((uintptr_t)buffer, length);
 
--
2.7.4

Reply | Threaded
Open this post in threaded view
|

Re: [RESEND PATCH v3] usb: xhci: fix lack of short packet event trb handling

Bin Meng-2
On Wed, Nov 18, 2020 at 11:46 AM Ran Wang <[hidden email]> wrote:
>
> For bulk IN transfer, the codes will set ISP flag to request event TRB
> being generated by xHC for case of short packet. So for the case of

Looks you missed this one: for the case of

> buffer-cross-64K-boundary (which we will divide payload and enqueuqe
> more than 1 transfer TRB), if the first TRB ends up with a short packet
> condition it will trigger an short packet code transfer event per that
> flag and cause more than 1 event TRB generated for this transfer.
>
> However, current codes will only handle the first transfer event TRB
> then mark current transfer completed, causing next transfer
> failure due to event TRB mis-match.
>
> Such issue has been observed on some Layerscape platforms (LS1028A,
> LS1088A, etc) with USB ethernet device.
>
> This patch adds a loop to make sure the event TRB for last transfer TRB
> has been handled in time.
>
> Signed-off-by: Ran Wang <[hidden email]>
> ---
> Change in v3:
>  - Update commit message according to v2 feedback .
>  - Replace (void *) with (uintptr_t) to fix below armv7 compile warning:
>  ...
>  drivers/usb/host/xhci-ring.c: In function 'xhci_bulk_tx':
>  drivers/usb/host/xhci-ring.c:726:6: warning: cast to pointer from integer of different size [-Wint-to-pointer-cast]
>    726 |  if ((void *)le64_to_cpu(event->trans_event.buffer) != last_transfer_trb_addr) {
>        |       ^
>  ...
>
>
> Change in v2:
>  - Re-write commit message to describe context more clearly.
>  - Add prefix 'le64_to_cpu' for 'event->trans_event.buffer'.
>
>  drivers/usb/host/xhci-ring.c | 18 +++++++++++++++---
>  1 file changed, 15 insertions(+), 3 deletions(-)
>

Looks good to me otherwise:
Reviewed-by: Bin Meng <[hidden email]>