| Message ID | 20240722150738.61486-1-ukaszb@chromium.org (mailing list archive) |
|---|---|
| State | Superseded |
| Headers | show |
| Series | [v1] xhci: dbc: fix handling ClearFeature Halt requests | expand |
On Mon, Jul 22, 2024 at 03:07:38PM +0000, Łukasz Bartosik wrote: > DbC driver does not handle ClearFeature Halt requests correctly > which in turn may lead to dropping packets on the receive path. > > Below is a trace capture where due to incorrect handling of > ClearFeature Halt packet gets dropped on the receive path. > > /sys/kernel/debug/tracing # cat trace > 1) kworker/10:3-514 [010] d..1. 2925.601843: xhci_dbc_handle_event: > EVENT: TRB 000000019588c0e0 status 'Stall Error' len 0 slot 1 ep 2 > type 'Transfer Event' flags e:C > > 2) kworker/10:3-514 [010] d..1. 2925.613285: xhci_dbc_handle_event: > EVENT: TRB 000000019588bc80 status 'Stall Error' len 1024 slot 1 > ep 3 type 'Transfer Event' flags e:C > > 3) kworker/10:3-514 [010] d..1. 2925.619024: xhci_dbc_handle_transfer: > BULK: Buffer 0000000244b49800 length 1024 TD size 0 intr 0 type > 'Normal' flags b:i:I:c:s:i:e:C > > 4) kworker/10:3-514 [010] d..1. 2925.619025: xhci_dbc_giveback_request: > bulk-in: req 00000000a70b5ad2 length 0/1024 ==> -6 > > 5) kworker/10:3-514 [010] dNs2. 2925.623820: xhci_dbc_gadget_ep_queue: > BULK: Buffer 0000000244b49800 length 1024 TD size 0 intr 0 type > 'Normal' flags b:i:I:c:s:i:e:c > > 6) kworker/10:3-514 [010] dNs1. 2925.623823: xhci_dbc_queue_request: > bulk-in: req 00000000a70b5ad2 length 0/1024 ==> -115 > > 7) kworker/10:3-514 [010] d..1. 2925.629865: xhci_dbc_handle_event: > EVENT: TRB 000000019588bc80 status 'Short Packet' len 1000 slot 1 > ep 3 type 'Transfer Event' flags e:C > > 8) kworker/10:3-514 [010] d..1. 2925.635540: xhci_dbc_handle_event: > EVENT: TRB 000000019588bc90 status 'Short Packet' len 763 slot 1 > ep 3 type 'Transfer Event' flags e:C > > 9) kworker/10:3-514 [010] d..1. 2925.635540: xhci_dbc_handle_transfer: > BULK: Buffer 0000000244b49c00 length 1024 TD size 0 intr 0 type > 'Normal' flags b:i:I:c:s:i:e:C > > 10) kworker/10:3-514 [010] d..1. 2925.635541: xhci_dbc_giveback_request: > bulk-in: req 00000000b4ec77d7 length 261/1024 ==> 0 > > 11) kworker/10:3-514 [010] dNs2. 2925.635561: xhci_dbc_gadget_ep_queue: > BULK: Buffer 0000000244b49c00 length 1024 TD size 0 intr 0 type > 'Normal' flags b:i:I:c:s:i:e:c > > 12) kworker/10:3-514 [010] dNs1. 2925.635562: xhci_dbc_queue_request: > bulk-in: req 00000000b4ec77d7 length 0/1024 ==> -115 > > Lines 1 and 2 are Endpoints OUT and IN responses to receiving ClearFeature > Halt requests. > > Line 7 notifies of reception of 24 bytes packet. > > Line 8 notifies of reception of 261 bytes packet > > In Lines [9, 12] 261 bytes packet is being processed. > > However 24 bytes packet gets dropped. The kernel log includes entry which > is an indication of a packet drop: > [ 127.651845] xhci_hcd 0000:00:0d.0: no matched request > > This fix adds correct handling of ClearFeature Halt requests > by restarting an endpoint which received the request. > > Fixes: dfba2174dc42 ("usb: xhci: Add DbC support in xHCI driver") > Signed-off-by: Łukasz Bartosik <ukaszb@chromium.org> > --- > drivers/usb/host/xhci-dbgcap.c | 29 +++++++++++++++++++---------- > drivers/usb/host/xhci-dbgtty.c | 5 +++++ > 2 files changed, 24 insertions(+), 10 deletions(-) > > diff --git a/drivers/usb/host/xhci-dbgcap.c b/drivers/usb/host/xhci-dbgcap.c > index 872d9cddbcef..b7af3a5ac98a 100644 > --- a/drivers/usb/host/xhci-dbgcap.c > +++ b/drivers/usb/host/xhci-dbgcap.c > @@ -173,7 +173,7 @@ static void xhci_dbc_giveback(struct dbc_request *req, int status) > spin_lock(&dbc->lock); > } > > -static void xhci_dbc_flush_single_request(struct dbc_request *req) > +static void xhci_dbc_flush_single_request(struct dbc_request *req, int status) So now we need to look up what "status" is here for every place this is called? That's going to be a pain, as you didn't even document it :( thanks, greg k-h
On Mon, Jul 22, 2024 at 6:31 PM Greg Kroah-Hartman <gregkh@linuxfoundation.org> wrote: > > On Mon, Jul 22, 2024 at 03:07:38PM +0000, Łukasz Bartosik wrote: > > DbC driver does not handle ClearFeature Halt requests correctly > > which in turn may lead to dropping packets on the receive path. > > > > Below is a trace capture where due to incorrect handling of > > ClearFeature Halt packet gets dropped on the receive path. > > > > /sys/kernel/debug/tracing # cat trace > > 1) kworker/10:3-514 [010] d..1. 2925.601843: xhci_dbc_handle_event: > > EVENT: TRB 000000019588c0e0 status 'Stall Error' len 0 slot 1 ep 2 > > type 'Transfer Event' flags e:C > > > > 2) kworker/10:3-514 [010] d..1. 2925.613285: xhci_dbc_handle_event: > > EVENT: TRB 000000019588bc80 status 'Stall Error' len 1024 slot 1 > > ep 3 type 'Transfer Event' flags e:C > > > > 3) kworker/10:3-514 [010] d..1. 2925.619024: xhci_dbc_handle_transfer: > > BULK: Buffer 0000000244b49800 length 1024 TD size 0 intr 0 type > > 'Normal' flags b:i:I:c:s:i:e:C > > > > 4) kworker/10:3-514 [010] d..1. 2925.619025: xhci_dbc_giveback_request: > > bulk-in: req 00000000a70b5ad2 length 0/1024 ==> -6 > > > > 5) kworker/10:3-514 [010] dNs2. 2925.623820: xhci_dbc_gadget_ep_queue: > > BULK: Buffer 0000000244b49800 length 1024 TD size 0 intr 0 type > > 'Normal' flags b:i:I:c:s:i:e:c > > > > 6) kworker/10:3-514 [010] dNs1. 2925.623823: xhci_dbc_queue_request: > > bulk-in: req 00000000a70b5ad2 length 0/1024 ==> -115 > > > > 7) kworker/10:3-514 [010] d..1. 2925.629865: xhci_dbc_handle_event: > > EVENT: TRB 000000019588bc80 status 'Short Packet' len 1000 slot 1 > > ep 3 type 'Transfer Event' flags e:C > > > > 8) kworker/10:3-514 [010] d..1. 2925.635540: xhci_dbc_handle_event: > > EVENT: TRB 000000019588bc90 status 'Short Packet' len 763 slot 1 > > ep 3 type 'Transfer Event' flags e:C > > > > 9) kworker/10:3-514 [010] d..1. 2925.635540: xhci_dbc_handle_transfer: > > BULK: Buffer 0000000244b49c00 length 1024 TD size 0 intr 0 type > > 'Normal' flags b:i:I:c:s:i:e:C > > > > 10) kworker/10:3-514 [010] d..1. 2925.635541: xhci_dbc_giveback_request: > > bulk-in: req 00000000b4ec77d7 length 261/1024 ==> 0 > > > > 11) kworker/10:3-514 [010] dNs2. 2925.635561: xhci_dbc_gadget_ep_queue: > > BULK: Buffer 0000000244b49c00 length 1024 TD size 0 intr 0 type > > 'Normal' flags b:i:I:c:s:i:e:c > > > > 12) kworker/10:3-514 [010] dNs1. 2925.635562: xhci_dbc_queue_request: > > bulk-in: req 00000000b4ec77d7 length 0/1024 ==> -115 > > > > Lines 1 and 2 are Endpoints OUT and IN responses to receiving ClearFeature > > Halt requests. > > > > Line 7 notifies of reception of 24 bytes packet. > > > > Line 8 notifies of reception of 261 bytes packet > > > > In Lines [9, 12] 261 bytes packet is being processed. > > > > However 24 bytes packet gets dropped. The kernel log includes entry which > > is an indication of a packet drop: > > [ 127.651845] xhci_hcd 0000:00:0d.0: no matched request > > > > This fix adds correct handling of ClearFeature Halt requests > > by restarting an endpoint which received the request. > > > > Fixes: dfba2174dc42 ("usb: xhci: Add DbC support in xHCI driver") > > Signed-off-by: Łukasz Bartosik <ukaszb@chromium.org> > > --- > > drivers/usb/host/xhci-dbgcap.c | 29 +++++++++++++++++++---------- > > drivers/usb/host/xhci-dbgtty.c | 5 +++++ > > 2 files changed, 24 insertions(+), 10 deletions(-) > > > > diff --git a/drivers/usb/host/xhci-dbgcap.c b/drivers/usb/host/xhci-dbgcap.c > > index 872d9cddbcef..b7af3a5ac98a 100644 > > --- a/drivers/usb/host/xhci-dbgcap.c > > +++ b/drivers/usb/host/xhci-dbgcap.c > > @@ -173,7 +173,7 @@ static void xhci_dbc_giveback(struct dbc_request *req, int status) > > spin_lock(&dbc->lock); > > } > > > > -static void xhci_dbc_flush_single_request(struct dbc_request *req) > > +static void xhci_dbc_flush_single_request(struct dbc_request *req, int status) > > > So now we need to look up what "status" is here for every place this is > called? That's going to be a pain, as you didn't even document it :( > The xhci_dbc_flush_single_request is called only in the function xhci_dbc_flush_endpoint_requests which follows xhci_dbc_flush_single_request but I will document it. Thanks, Lukasz > thanks, > > greg k-h
diff --git a/drivers/usb/host/xhci-dbgcap.c b/drivers/usb/host/xhci-dbgcap.c index 872d9cddbcef..b7af3a5ac98a 100644 --- a/drivers/usb/host/xhci-dbgcap.c +++ b/drivers/usb/host/xhci-dbgcap.c @@ -173,7 +173,7 @@ static void xhci_dbc_giveback(struct dbc_request *req, int status) spin_lock(&dbc->lock); } -static void xhci_dbc_flush_single_request(struct dbc_request *req) +static void xhci_dbc_flush_single_request(struct dbc_request *req, int status) { union xhci_trb *trb = req->trb; @@ -183,21 +183,27 @@ static void xhci_dbc_flush_single_request(struct dbc_request *req) trb->generic.field[3] &= cpu_to_le32(TRB_CYCLE); trb->generic.field[3] |= cpu_to_le32(TRB_TYPE(TRB_TR_NOOP)); - xhci_dbc_giveback(req, -ESHUTDOWN); + xhci_dbc_giveback(req, status); } -static void xhci_dbc_flush_endpoint_requests(struct dbc_ep *dep) +static void xhci_dbc_flush_endpoint_requests(struct dbc_ep *dep, bool restart) { + struct list_head *list = &dep->list_pending; struct dbc_request *req, *tmp; + int status = -ESHUTDOWN; + + list_for_each_entry_safe(req, tmp, list, list_pending) { + if (restart && list_is_last(&req->list_pending, list)) + status = -ERESTART; - list_for_each_entry_safe(req, tmp, &dep->list_pending, list_pending) - xhci_dbc_flush_single_request(req); + xhci_dbc_flush_single_request(req, status); + } } static void xhci_dbc_flush_requests(struct xhci_dbc *dbc) { - xhci_dbc_flush_endpoint_requests(&dbc->eps[BULK_OUT]); - xhci_dbc_flush_endpoint_requests(&dbc->eps[BULK_IN]); + xhci_dbc_flush_endpoint_requests(&dbc->eps[BULK_OUT], false); + xhci_dbc_flush_endpoint_requests(&dbc->eps[BULK_IN], false); } struct dbc_request * @@ -718,10 +724,13 @@ static void dbc_handle_xfer_event(struct xhci_dbc *dbc, union xhci_trb *event) case COMP_TRB_ERROR: case COMP_BABBLE_DETECTED_ERROR: case COMP_USB_TRANSACTION_ERROR: - case COMP_STALL_ERROR: dev_warn(dbc->dev, "tx error %d detected\n", comp_code); status = -comp_code; break; + case COMP_STALL_ERROR: + /* Restart endpoint */ + xhci_dbc_flush_endpoint_requests(dep, true); + return; default: dev_err(dbc->dev, "unknown tx error %d\n", comp_code); status = -comp_code; @@ -823,12 +832,12 @@ static enum evtreturn xhci_dbc_do_handle_events(struct xhci_dbc *dbc) if (ctrl & DBC_CTRL_HALT_IN_TR) { dep = get_in_ep(dbc); - xhci_dbc_flush_endpoint_requests(dep); + xhci_dbc_flush_endpoint_requests(dep, false); } if (ctrl & DBC_CTRL_HALT_OUT_TR) { dep = get_out_ep(dbc); - xhci_dbc_flush_endpoint_requests(dep); + xhci_dbc_flush_endpoint_requests(dep, false); } return EVT_DONE; diff --git a/drivers/usb/host/xhci-dbgtty.c b/drivers/usb/host/xhci-dbgtty.c index b74e98e94393..bdf80aa2b28c 100644 --- a/drivers/usb/host/xhci-dbgtty.c +++ b/drivers/usb/host/xhci-dbgtty.c @@ -121,6 +121,7 @@ static void dbc_write_complete(struct xhci_dbc *dbc, struct dbc_request *req) list_add(&req->list_pool, &port->write_pool); switch (req->status) { case 0: + case -ERESTART: dbc_start_tx(port); break; case -ESHUTDOWN: @@ -318,6 +319,10 @@ static void dbc_rx_push(struct tasklet_struct *t) case -ESHUTDOWN: disconnect = true; break; + case -ERESTART: + disconnect = false; + req->actual = 0; + break; default: pr_warn("ttyDBC0: unexpected RX status %d\n", req->status);
DbC driver does not handle ClearFeature Halt requests correctly which in turn may lead to dropping packets on the receive path. Below is a trace capture where due to incorrect handling of ClearFeature Halt packet gets dropped on the receive path. /sys/kernel/debug/tracing # cat trace 1) kworker/10:3-514 [010] d..1. 2925.601843: xhci_dbc_handle_event: EVENT: TRB 000000019588c0e0 status 'Stall Error' len 0 slot 1 ep 2 type 'Transfer Event' flags e:C 2) kworker/10:3-514 [010] d..1. 2925.613285: xhci_dbc_handle_event: EVENT: TRB 000000019588bc80 status 'Stall Error' len 1024 slot 1 ep 3 type 'Transfer Event' flags e:C 3) kworker/10:3-514 [010] d..1. 2925.619024: xhci_dbc_handle_transfer: BULK: Buffer 0000000244b49800 length 1024 TD size 0 intr 0 type 'Normal' flags b:i:I:c:s:i:e:C 4) kworker/10:3-514 [010] d..1. 2925.619025: xhci_dbc_giveback_request: bulk-in: req 00000000a70b5ad2 length 0/1024 ==> -6 5) kworker/10:3-514 [010] dNs2. 2925.623820: xhci_dbc_gadget_ep_queue: BULK: Buffer 0000000244b49800 length 1024 TD size 0 intr 0 type 'Normal' flags b:i:I:c:s:i:e:c 6) kworker/10:3-514 [010] dNs1. 2925.623823: xhci_dbc_queue_request: bulk-in: req 00000000a70b5ad2 length 0/1024 ==> -115 7) kworker/10:3-514 [010] d..1. 2925.629865: xhci_dbc_handle_event: EVENT: TRB 000000019588bc80 status 'Short Packet' len 1000 slot 1 ep 3 type 'Transfer Event' flags e:C 8) kworker/10:3-514 [010] d..1. 2925.635540: xhci_dbc_handle_event: EVENT: TRB 000000019588bc90 status 'Short Packet' len 763 slot 1 ep 3 type 'Transfer Event' flags e:C 9) kworker/10:3-514 [010] d..1. 2925.635540: xhci_dbc_handle_transfer: BULK: Buffer 0000000244b49c00 length 1024 TD size 0 intr 0 type 'Normal' flags b:i:I:c:s:i:e:C 10) kworker/10:3-514 [010] d..1. 2925.635541: xhci_dbc_giveback_request: bulk-in: req 00000000b4ec77d7 length 261/1024 ==> 0 11) kworker/10:3-514 [010] dNs2. 2925.635561: xhci_dbc_gadget_ep_queue: BULK: Buffer 0000000244b49c00 length 1024 TD size 0 intr 0 type 'Normal' flags b:i:I:c:s:i:e:c 12) kworker/10:3-514 [010] dNs1. 2925.635562: xhci_dbc_queue_request: bulk-in: req 00000000b4ec77d7 length 0/1024 ==> -115 Lines 1 and 2 are Endpoints OUT and IN responses to receiving ClearFeature Halt requests. Line 7 notifies of reception of 24 bytes packet. Line 8 notifies of reception of 261 bytes packet In Lines [9, 12] 261 bytes packet is being processed. However 24 bytes packet gets dropped. The kernel log includes entry which is an indication of a packet drop: [ 127.651845] xhci_hcd 0000:00:0d.0: no matched request This fix adds correct handling of ClearFeature Halt requests by restarting an endpoint which received the request. Fixes: dfba2174dc42 ("usb: xhci: Add DbC support in xHCI driver") Signed-off-by: Łukasz Bartosik <ukaszb@chromium.org> --- drivers/usb/host/xhci-dbgcap.c | 29 +++++++++++++++++++---------- drivers/usb/host/xhci-dbgtty.c | 5 +++++ 2 files changed, 24 insertions(+), 10 deletions(-)