Re: [PATCH 3/3] xhci: rework cycle bit checking for new dequeue pointers

From: Joseph Salisbury
Date: Wed Aug 20 2014 - 18:06:22 EST


On 08/19/2014 08:17 AM, Mathias Nyman wrote:
> When we manually need to move the TR dequeue pointer we need to set the
> correct cycle bit as well. Previously we used the trb pointer from the
> last event received as a base, but this was changed in
> commit 1f81b6d22a59 ("usb: xhci: Prefer endpoint context dequeue pointer")
> to use the dequeue pointer from the endpoint context instead
>
> It turns out some Asmedia controllers advance the dequeue pointer
> stored in the endpoint context past the event triggering TRB, and
> this messed up the way the cycle bit was calculated.
>
> Instead of adding a quirk or complicating the already hard to follow cycle bit
> code, the whole cycle bit calculation is now simplified and adapted to handle
> event and endpoint context dequeue pointer differences.
>
> Fixes: 1f81b6d22a59 ("usb: xhci: Prefer endpoint context dequeue pointer")
> Reported-by: Maciej Puzio <mx34567@xxxxxxxxx>
> Reported-by: Evan Langlois <uudruid74@xxxxxxxxx>
> Reviewed-by: Julius Werner <jwerner@xxxxxxxxxxxx>
> Tested-by: Maciej Puzio <mx34567@xxxxxxxxx>
> Tested-by: Evan Langlois <uudruid74@xxxxxxxxx>
> Signed-off-by: Mathias Nyman <mathias.nyman@xxxxxxxxxxxxxxx>
> Cc: stable@xxxxxxxxxxxxxxx
> ---
> drivers/usb/host/xhci-ring.c | 101 +++++++++++++++++--------------------------
> drivers/usb/host/xhci.c | 3 ++
> 2 files changed, 42 insertions(+), 62 deletions(-)
>
> diff --git a/drivers/usb/host/xhci-ring.c b/drivers/usb/host/xhci-ring.c
> index ac8cf23..abed30b 100644
> --- a/drivers/usb/host/xhci-ring.c
> +++ b/drivers/usb/host/xhci-ring.c
> @@ -364,32 +364,6 @@ static void ring_doorbell_for_active_rings(struct xhci_hcd *xhci,
> }
> }
>
> -/*
> - * Find the segment that trb is in. Start searching in start_seg.
> - * If we must move past a segment that has a link TRB with a toggle cycle state
> - * bit set, then we will toggle the value pointed at by cycle_state.
> - */
> -static struct xhci_segment *find_trb_seg(
> - struct xhci_segment *start_seg,
> - union xhci_trb *trb, int *cycle_state)
> -{
> - struct xhci_segment *cur_seg = start_seg;
> - struct xhci_generic_trb *generic_trb;
> -
> - while (cur_seg->trbs > trb ||
> - &cur_seg->trbs[TRBS_PER_SEGMENT - 1] < trb) {
> - generic_trb = &cur_seg->trbs[TRBS_PER_SEGMENT - 1].generic;
> - if (generic_trb->field[3] & cpu_to_le32(LINK_TOGGLE))
> - *cycle_state ^= 0x1;
> - cur_seg = cur_seg->next;
> - if (cur_seg == start_seg)
> - /* Looped over the entire list. Oops! */
> - return NULL;
> - }
> - return cur_seg;
> -}
> -
> -
> static struct xhci_ring *xhci_triad_to_transfer_ring(struct xhci_hcd *xhci,
> unsigned int slot_id, unsigned int ep_index,
> unsigned int stream_id)
> @@ -459,9 +433,12 @@ void xhci_find_new_dequeue_state(struct xhci_hcd *xhci,
> struct xhci_virt_device *dev = xhci->devs[slot_id];
> struct xhci_virt_ep *ep = &dev->eps[ep_index];
> struct xhci_ring *ep_ring;
> - struct xhci_generic_trb *trb;
> + struct xhci_segment *new_seg;
> + union xhci_trb *new_deq;
> dma_addr_t addr;
> u64 hw_dequeue;
> + bool cycle_found = false;
> + bool td_last_trb_found = false;
>
> ep_ring = xhci_triad_to_transfer_ring(xhci, slot_id,
> ep_index, stream_id);
> @@ -486,45 +463,45 @@ void xhci_find_new_dequeue_state(struct xhci_hcd *xhci,
> hw_dequeue = le64_to_cpu(ep_ctx->deq);
> }
>
> - /* Find virtual address and segment of hardware dequeue pointer */
> - state->new_deq_seg = ep_ring->deq_seg;
> - state->new_deq_ptr = ep_ring->dequeue;
> - while (xhci_trb_virt_to_dma(state->new_deq_seg, state->new_deq_ptr)
> - != (dma_addr_t)(hw_dequeue & ~0xf)) {
> - next_trb(xhci, ep_ring, &state->new_deq_seg,
> - &state->new_deq_ptr);
> - if (state->new_deq_ptr == ep_ring->dequeue) {
> - WARN_ON(1);
> - return;
> - }
> - }
> + new_seg = ep_ring->deq_seg;
> + new_deq = ep_ring->dequeue;
> + state->new_cycle_state = hw_dequeue & 0x1;
> +
> /*
> - * Find cycle state for last_trb, starting at old cycle state of
> - * hw_dequeue. If there is only one segment ring, find_trb_seg() will
> - * return immediately and cannot toggle the cycle state if this search
> - * wraps around, so add one more toggle manually in that case.
> + * We want to find the pointer, segment and cycle state of the new trb
> + * (the one after current TD's last_trb). We know the cycle state at
> + * hw_dequeue, so walk the ring until both hw_dequeue and last_trb are
> + * found.
> */
> - state->new_cycle_state = hw_dequeue & 0x1;
> - if (ep_ring->first_seg == ep_ring->first_seg->next &&
> - cur_td->last_trb < state->new_deq_ptr)
> - state->new_cycle_state ^= 0x1;
> + do {
> + if (!cycle_found && xhci_trb_virt_to_dma(new_seg, new_deq)
> + == (dma_addr_t)(hw_dequeue & ~0xf)) {
> + cycle_found = true;
> + if (td_last_trb_found)
> + break;
> + }
> + if (new_deq == cur_td->last_trb)
> + td_last_trb_found = true;
>
> - state->new_deq_ptr = cur_td->last_trb;
> - xhci_dbg_trace(xhci, trace_xhci_dbg_cancel_urb,
> - "Finding segment containing last TRB in TD.");
> - state->new_deq_seg = find_trb_seg(state->new_deq_seg,
> - state->new_deq_ptr, &state->new_cycle_state);
> - if (!state->new_deq_seg) {
> - WARN_ON(1);
> - return;
> - }
> + if (cycle_found &&
> + TRB_TYPE_LINK_LE32(new_deq->generic.field[3]) &&
> + new_deq->generic.field[3] & cpu_to_le32(LINK_TOGGLE))
> + state->new_cycle_state ^= 0x1;
> +
> + next_trb(xhci, ep_ring, &new_seg, &new_deq);
> +
> + /* Search wrapped around, bail out */
> + if (new_deq == ep->ring->dequeue) {
> + xhci_err(xhci, "Error: Failed finding new dequeue state\n");
> + state->new_deq_seg = NULL;
> + state->new_deq_ptr = NULL;
> + return;
> + }
> +
> + } while (!cycle_found || !td_last_trb_found);
>
> - /* Increment to find next TRB after last_trb. Cycle if appropriate. */
> - trb = &state->new_deq_ptr->generic;
> - if (TRB_TYPE_LINK_LE32(trb->field[3]) &&
> - (trb->field[3] & cpu_to_le32(LINK_TOGGLE)))
> - state->new_cycle_state ^= 0x1;
> - next_trb(xhci, ep_ring, &state->new_deq_seg, &state->new_deq_ptr);
> + state->new_deq_seg = new_seg;
> + state->new_deq_ptr = new_deq;
>
> /* Don't update the ring cycle state for the producer (us). */
> xhci_dbg_trace(xhci, trace_xhci_dbg_cancel_urb,
> diff --git a/drivers/usb/host/xhci.c b/drivers/usb/host/xhci.c
> index b6f2117..c020b09 100644
> --- a/drivers/usb/host/xhci.c
> +++ b/drivers/usb/host/xhci.c
> @@ -2880,6 +2880,9 @@ void xhci_cleanup_stalled_ring(struct xhci_hcd *xhci,
> ep_index, ep->stopped_stream, ep->stopped_td,
> &deq_state);
>
> + if (!deq_state.new_deq_ptr || !deq_state.new_deq_seg)
> + return;
> +
> /* HW with the reset endpoint quirk will use the saved dequeue state to
> * issue a configure endpoint command later.
> */

Hi Mathias,

Some of the stable kernel versions fail to build with this patch, 3.2.y
and 3.13.y for example. This is because the function 'find_trb_seg' is
still called by xhci_cmd_to_noop, which is removed from mainline but
still exists in the stable kernels. The patch removes the definition of
find_trb_seg, which is what causes the build to fail.

Should we leave the definition of find_trb_seg in your patch for the
stable kernels, or do you have other ideas?

Thanks,

Joe


--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/