The readq() and writeq() helpers are available in the
asm-generic/io-64-nonatomic-hi-lo.h and asm-generic/io-64-nonatomic-lo-hi.h
headers. Replace custom implementation by the generic helpers.
Signed-off-by: Andy Shevchenko <andriy.shevchenko@linux.intel.com>
Signed-off-by: Mathias Nyman <mathias.nyman@linux.intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
xhci versions 1.0 and later report the untransferred data remaining in a
TD a bit differently than older hosts.
We used to have separate functions for these, and needed to check host
version before calling the right function.
Now Mediatek host has an additional quirk on how it uses the TD Size
field for remaining data. To prevent yet another function for calculating
remainder we instead want to make one quirk friendly unified function.
Tested-by: Chunfeng Yun <chunfeng.yun@mediatek.com>
Signed-off-by: Mathias Nyman <mathias.nyman@linux.intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
USB 3.1 adds different types of Get Port Status request.
The Get Extended Port Status request returns 4 additional bytes
after the normal portstatus and portchange words containing
link speed and lane information about a connected enhanced super
speed device
Signed-off-by: Mathias Nyman <mathias.nyman@linux.intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
USB 3.1 capable xhci controllers use a new default speed ID "5" in the
PORTSC register to represent a 10Gbps connection speed of a SuperSpeedPlus
device
Make sure the xhci driver can handle the returned SuperSpeedPlus speed ID
properly
Signed-off-by: Mathias Nyman <mathias.nyman@linux.intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
xhci 1.1 controllers that support USB 3.1 must provide a protocol speed ID
(PSI) list to inform the driver of the supported speeds.
The PSI list can be read from the xhci supported protocol extended
capabilities.
The PSI values will be used to create a USB 3.1 SuperSpeedPlus capability
descriptor for the xhci USB 3.1 roothub.
Signed-off-by: Mathias Nyman <mathias.nyman@linux.intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
xhci 1.1 capable controllers have a new HCCPARAMS2 registers
with bits indicating support for new xhci 1.1 capabilities.
Also add support for the new xhci 1.1 bits in the config operational
opertational register that used to be reserved
Signed-off-by: Lu Baolu <baolu.lu@linux.intel.com>
[modified and left out parts not related to HCCPARAMS2 -Mathias]
Signed-off-by: Mathias Nyman <mathias.nyman@linux.intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
This patch enables xhci driver to support SPC by handling
Stopped - Short Packet event in transfer event path.
If SPC = '1' and the stop endpoint command is executed, after a Short
Packet condition has been detected, but before the end of the TD has been
reached, (i.e. the TD is in progress for pipe), then a Transfer Event TRB
with its Completion Code set to Stopped - Short Packet and its TRB
Transfer Length set to value of the EDTLA shall be forced for the
interrupted TRB, irrespective of whether its IOC or ISP flags are set.
This Transfer Event TRB will precede the Command Completion Event TRB for
the command, and is referred to as a Stopped Transfer Event.
Signed-off-by: Lu Baolu <baolu.lu@linux.intel.com>
Signed-off-by: Mathias Nyman <mathias.nyman@linux.intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
If the Contiguous Frame ID Capability is supported (CFC = 1),
then the xHC shall match the Frame ID in every Isoch TD with
SIA = 0 against the Frame Index of the MFINDEX register. This
rule ensures resynchronization of Isoch TDs even if some are
dropped due to Missed Service Errors or Stopping the endpoint.
This patch enables xHCI driver to support CFC by calculating
and setting the Frame ID field of an Isoch TRB.
[made some dbg messages checkpatch friendly -Mathias]
Signed-off-by: Lu Baolu <baolu.lu@linux.intel.com>
Signed-off-by: Mathias Nyman <mathias.nyman@linux.intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
When the link is just waken, it's in Resume state, and driver sets PLS to
U0. This refers to Phase 1. Phase 2 refers to when the link has completed
the transition from Resume state to U0.
With the fix of xhci: report U3 when link is in resume state, it also
exposes an issue that usb3 roothub and controller can suspend right
after phase 1, and this causes a hard hang in controller.
To fix the issue, we need to prevent usb3 bus suspend if any port is
resuming in phase 1.
[merge separate USB2 and USB3 port resume checking to one -Mathias]
Cc: <stable@vger.kernel.org>
Signed-off-by: Zhuang Jin Can <jin.can.zhuang@intel.com>
Signed-off-by: Mathias Nyman <mathias.nyman@linux.intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
This resolves a merge issue in musb_core.c and we want the fixes that
were in Linus's tree in this branch as well for testing.
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
HCD core allocates memory for HCD private data in
usb_create_[shared_]hcd() so make use of that
mechanism to allocate the struct xhci_hcd.
Introduce struct xhci_driver_overrides to provide
the size of HCD private data and hc_driver operation
overrides. As of now we only need to override the
reset and start methods.
Signed-off-by: Roger Quadros <rogerq@ti.com>
Signed-off-by: Mathias Nyman <mathias.nyman@linux.intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Regression in commit 638139eb95 ("usb: hub: allow to process more usb
hub events in parallel")
The regression resulted in intermittent failure to initialise a 10-port
hub (with three internal VL812 4-port hub controllers) on boot, with a
failure rate of around 8%, due to multiple race conditions when
accessing addr_dev and slot_id in struct xhci_hcd.
This regression also exposed a problem with xhci_setup_device, which
"should be protected by the usb_address0_mutex" but no longer is due to
commit 6fecd4f2a5 ("USB: separate usb_address0 mutexes for each bus")
With separate buses (and locks) it is no longer the case that a single
lock will protect xhci_setup_device from accesses by two parallel
threads processing events on the two buses.
Fix this by adding a mutex to protect addr_dev and slot_id in struct
xhci_hcd, and by making the assignment of slot_id atomic.
Fixes multiple boot errors:
[ 0.583008] xhci_hcd 0000:00:14.0: Bad Slot ID 2
[ 0.583009] xhci_hcd 0000:00:14.0: Could not allocate xHCI USB device data structures
[ 0.583012] usb usb1-port3: couldn't allocate usb_device
And:
[ 0.637409] xhci_hcd 0000:00:14.0: Error while assigning device slot ID
[ 0.637417] xhci_hcd 0000:00:14.0: Max number of devices this xHCI host supports is 32.
[ 0.637421] usb usb1-port1: couldn't allocate usb_device
And:
[ 0.753372] xhci_hcd 0000:00:14.0: ERROR: unexpected setup context command completion code 0x0.
[ 0.753373] usb 1-3: hub failed to enable device, error -22
[ 0.753400] xhci_hcd 0000:00:14.0: Error while assigning device slot ID
[ 0.753402] xhci_hcd 0000:00:14.0: Max number of devices this xHCI host supports is 32.
[ 0.753403] usb usb1-port3: couldn't allocate usb_device
And:
[ 11.018386] usb 1-3: device descriptor read/all, error -110
And:
[ 5.753838] xhci_hcd 0000:00:14.0: Timeout while waiting for setup device command
Tested with 200 reboots, resulting in no USB hub init related errors.
Fixes: 638139eb95 ("usb: hub: allow to process more usb hub events in parallel")
Link: https://lkml.kernel.org/g/CAP-bSRb=A0iEYobdGCLpwynS7pkxpt_9ZnwyZTPVAoy0Y=Zo3Q@mail.gmail.com
Signed-off-by: Chris Bainbridge <chris.bainbridge@gmail.com>
Cc: <stable@vger.kernel.org> # 3.18+
[changed git commit description style for checkpatch -Mathias]
Signed-off-by: Mathias Nyman <mathias.nyman@linux.intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Our event ring consists of only one segment, and we risk filling
the event ring in case we get isoc transfers with short intervals
such as webcams that fill a TD every microframe (125us)
With 64 TRB segment size one usb camera could fill the event ring in 8ms.
A setup with several cameras and other devices can fill up the
event ring as it is shared between all devices.
This has occurred when uvcvideo queues 5 * 32TD URBs which then
get cancelled when the video mode changes. The cancelled URBs are returned
in the xhci interrupt context and blocks the interrupt handler from
handling the new events.
A full event ring will block xhci from scheduling traffic and affect all
devices conneted to the xhci, will see errors such as Missed Service
Intervals for isoc devices, and and Split transaction errors for LS/FS
interrupt devices.
Increasing the TRB_PER_SEGMENT will also increase the default endpoint ring
size, which is welcome as for most isoc transfer we had to dynamically
expand the endpoint ring anyway to be able to queue the 5 * 32TDs uvcvideo
queues.
The default size used to be 64 TRBs per segment
Cc: <stable@vger.kernel.org>
Signed-off-by: Mathias Nyman <mathias.nyman@linux.intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
This reverts commit 27082e2654 ("xhci: Clear the host side toggle manually")
Turns out this fix to enable soft resetting endpoints wasn't mature enough.
It caused regression with some usb DVB-T devices and needs some more tuning
to get the endpiont ring pointers set correctly.
The original commit was tagged for stable 3.18, and should be reverted
from there as well.
Cc: stable <stable@vger.kernel.org> # v3.18
Signed-off-by: Mathias Nyman <mathias.nyman@linux.intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The xhci in Intel Sunrisepoint and Cherryview platforms need a driver
workaround for a Stuck PME that might either block PME events in suspend,
or create spurious PME events preventing runtime suspend.
Workaround is to clear a internal PME flag, BIT(28) in a vendor specific
PMCTRL register at offset 0x80a4, in both suspend resume callbacks
Without this, xhci connected usb devices might never be able to wake up the
system from suspend, or prevent device from going to suspend (xhci d3)
Cc: <stable@vger.kernel.org>
Signed-off-by: Mathias Nyman <mathias.nyman@linux.intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
When a control transfer has a short data stage, the xHCI controller generates
two transfer events: a COMP_SHORT_TX event that specifies the untransferred
amount, and a COMP_SUCCESS event. But when the data stage is not short, only the
COMP_SUCCESS event occurs. Therefore, xhci-hcd must set urb->actual_length to
urb->transfer_buffer_length while processing the COMP_SUCCESS event, unless
urb->actual_length was set already by a previous COMP_SHORT_TX event.
The driver checks this by seeing whether urb->actual_length == 0, but this alone
is the wrong test, as it is entirely possible for a short transfer to have an
urb->actual_length = 0.
This patch changes the xhci driver to rely on a new td->urb_length_set flag,
which is set to true when a COMP_SHORT_TX event is received and the URB length
updated at that stage.
This fixes a bug which affected the HSO plugin, which relies on URBs with
urb->actual_length == 0 to halt re-submitting the RX URB in the control
endpoint.
Cc: <stable@vger.kernel.org>
Signed-off-by: Aleksander Morgado <aleksander@aleksander.es>
Signed-off-by: Mathias Nyman <mathias.nyman@linux.intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Main benefit of this is to get xhci connected USB scanners to work.
Some devices use a clear endpoint halt request as a 'soft reset' even if
the endpoint is not halted. This will clear the toggle and sequence on the
device side. xHCI however refuses to reset a non-halted endpoint, so instead
we need to issue a configure endpoint command on xHCI to clear its host side
toggle and sequence, and get it in sync with the device side.
Tested-by: Mike Mammarella <mikem@crystalorb.net>
Cc: <stable@vger.kernel.org> # v3.18
Signed-off-by: Mathias Nyman <mathias.nyman@linux.intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Include the high order bit fields for Max scratchpad buffers when
calculating how many scratchpad buffers are needed.
I'm suprised this hasn't caused more issues, we never allocated more than
32 buffers even if xhci needed more. Either we got lucky and xhci never
really used past that area, or then we got enough zeroed dma memory anyway.
Should be backported as far back as possible
Reported-by: Tim Chen <tim.c.chen@linux.intel.com>
Tested-by: Tim Chen <tim.c.chen@linux.intel.com>
Signed-off-by: Mathias Nyman <mathias.nyman@linux.intel.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
To help debugging xhci problems.
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Mathias Nyman <mathias.nyman@linux.intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Some parameters are not used by functions in xhci-mem.c, just
remove it.
Changes compared to v1:
- Rebase to the latest usb-next branch
Signed-off-by: Lin Wang <lin.x.wang@intel.com>
Signed-off-by: Mathias Nyman <mathias.nyman@linux.intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Parameter 'xhci' is no longer be used in function xhci_handshake(),
just remove it.
Signed-off-by: Lin Wang <lin.x.wang@intel.com>
Signed-off-by: Mathias Nyman <mathias.nyman@linux.intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
According to xHCI specification, PORT_DEV_REMOVE(bit 30) in PORTSC
true means "Device is non-removable".
Reported-by: Juro Bystricky <jurobystricky@hotmail.com>
Signed-off-by: Lu Baolu <baolu.lu@linux.intel.com>
Signed-off-by: Mathias Nyman <mathias.nyman@linux.intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Endpoints halted on errors, and endpoints stopped manually both used
the same ep->stopped_td to store the halted or stopped td. this causes
confusion and possible races.
There is no longer a need to use the ep->stopped_td variable to store
the halted TD. A halted endpoint is handled immediately and we can pass
it to the handling function directly.
Signed-off-by: Mathias Nyman <mathias.nyman@linux.intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
When system is being suspended, if host device is not allowed to do wakeup,
xhci_suspend() needs to clear all root port wake on bits. Otherwise, some
platforms may generate spurious wakeup, even if PCI PME# is disabled.
The initial commit ff8cbf250b ("xhci: clear root port wake on bits"),
which also got into stable, turned out to not work correctly and had to
be reverted, and is now rewritten.
Cc: stable <stable@vger.kernel.org> # v3.2+
Signed-off-by: Lu Baolu <baolu.lu@linux.intel.com>
Suggested-by: Alan Stern <stern@rowland.harvard.edu>
Acked-by: Alan Stern <stern@rowland.harvard.edu>
[Mathias Nyman: reword commit message]
Signed-off-by: Mathias Nyman <mathias.nyman@linux.intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Instead of building all of the xHCI code into a single module, separate
it out into the core (xhci-hcd), PCI (xhci-pci, now selected by the new
config option CONFIG_USB_XHCI_PCI), and platform (xhci-plat) drivers.
Also update the PCI/platform drivers with module descriptions/licenses
and have them register their respective drivers in their initcalls.
Signed-off-by: Andrew Bresticker <abrestic@chromium.org>
Signed-off-by: Mathias Nyman <mathias.nyman@linux.intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Instead of calling xhci_compliance_mode_recovery_timer_quirk_check() again
in the PCI suspend path, just check for XHCI_COMP_MODE_QUIRK which will
have been set based on xhci_compliance_mode_recovery_timer_quirk_check()
in xhci_init().
Signed-off-by: Andrew Bresticker <abrestic@chromium.org>
Signed-off-by: Mathias Nyman <mathias.nyman@linux.intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Since the struct hc_driver is mostly the same across the xhci-pci,
xhci-plat, and the upcoming xhci-tegra driver, introduce the function
xhci_init_driver() which will populate the hc_driver with the default
xHCI operations. The caller must supply a setup function which will
be used as the hc_driver's reset callback.
Note that xhci-plat also overrides the default ->start() callback so
that it can do rcar-specific initialization.
Signed-off-by: Andrew Bresticker <abrestic@chromium.org>
Signed-off-by: Mathias Nyman <mathias.nyman@linux.intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Lately (with the use of uas / bulk-streams) we have been seeing several
cases where this error triggers (which should never happen).
Add some extra logging to make debugging these errors easier.
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Mathias Nyman <mathias.nyman@linux.intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
There are multiple reasons for this:
1) This fixes a missing check for xhci_alloc_command failing in
xhci_handle_cmd_stop_ep()
2) This adds a warning when we cannot set the new dequeue state because of
xhci_alloc_command failing
3) It puts the allocation of the command after the sanity checks in
queue_set_tr_deq(), avoiding leaking the command if those fail
4) Since queue_set_tr_deq now owns the command it can free it if queue_command
fails
5) It reduces code duplication
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Mathias Nyman <mathias.nyman@linux.intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Streams on the EJ168 do not work as they should. I've spend 2 days trying
to get them to work, but without success.
The first problem is that when ever you ring the stream-ring doorbell, the
controller starts executing trbs at the beginning of the first ring segment,
event if it ended somewhere else previously. This can be worked around by
allowing enqueing only one td (not a problem with how streams are typically
used) and then resetting our copies of the enqueueing en dequeueing pointers
on a td completion to match what the controller seems to be doing.
This way things seem to start working with uas and instead of being able
to complete only the very first scsi command, the scsi core can probe the disk.
But then things break later on when td-s get enqueued with more then one
trb. The controller does seem to increase its dequeue pointer while executing
a stream-ring (data transfer events I inserted for debugging do trigger).
However execution seems to stop at the final normal trb of a multi trb td,
even if there is a data transfer event inserted after the final trb.
The first problem alone is a serious deviation from the spec, and esp.
dealing with cancellation would have been very tricky if not outright
impossible, but the second problem simply is a deal breaker altogether,
so this patch simply disables streams.
Note this will cause the usb-storage + uas driver pair to automatically switch
to using usb-storage instead of uas on these devices, essentially reverting
to the 3.14 and earlier behavior when uas was marked CONFIG_BROKEN.
https://bugzilla.redhat.com/show_bug.cgi?id=1121288https://bugzilla.kernel.org/show_bug.cgi?id=80101
Cc: stable@vger.kernel.org # 3.15
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Some platforms (such as the Armada 38x ones) can gate the clock of
their USB controller. This patch adds the support for one clock in
xhci-plat, by enabling it during probe and disabling it on remove.
To achieve this, it adds a 'struct clk *' member in xhci_hcd. While
only used for now in xhci-plat, it might be used by other drivers in
the future. Moreover, the xhci_hcd structure already holds other
members such as msix_count and msix_entries, which are MSI-X specific,
and therefore only used by xhci-pci.
Signed-off-by: Gregory CLEMENT <gregory.clement@free-electrons.com>
Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
Acked-by: Felipe Balbi <balbi@ti.com>
Acked-by: Mathias Nyman <mathias.nyman@linux.intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Use one timer to control command timeout.
start/kick the timer every time a command is completed and a
new command is waiting, or a new command is added to a empty list.
If the timer runs out, then tag the current command as "aborted", and
start the xhci command abortion process.
Previously each function that submitted a command had its own timer.
If that command timed out, a new command structure for the
command was created and it was put on a cancel_cmd_list list,
then a pci write to abort the command ring was issued.
when the ring was aborted, it checked if the current command
was the one to be canceled, later when the ring was stopped the
driver got ownership of the TRBs in the command ring,
compared then to the TRBs in the cancel_cmd_list,
and turned them into No-ops.
Now, instead, at timeout we tag the status of the command in the
command queue to be aborted, and start the ring abortion.
Ring abortion stops the command ring and gives control of the
commands to us.
All the aborted commands are now turned into No-ops.
If the ring is already stopped when the command times outs its not possible
to start the ring abortion, in this case the command is turnd to No-op
right away.
All these changes allows us to remove the entire cancel_cmd_list code.
The functions waiting for a command to finish no longer have their own timeouts.
They will wait either until the command completes normally,
or until the whole command abortion is done.
Signed-off-by: Mathias Nyman <mathias.nyman@linux.intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Remove the per-device command list and handle_cmd_in_cmd_wait_list()
and use the completion and status variables found in the
command structure in the global command list.
Signed-off-by: Mathias Nyman <mathias.nyman@linux.intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Create a list to store command structures, add a structure to it every time
a command is submitted, and remove it from the list once we get a
command completion event matching the command.
Callers that wait for completion will free their command structures themselves.
The other command structures are freed in the command completion event handler.
Also add a check that prevents queuing commands if host is dying
Signed-off-by: Mathias Nyman <mathias.nyman@linux.intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
To create a global command queue we require that each command put on the
command ring is submitted with a command structure.
Functions that queue commands and wait for completion need to allocate a command
before submitting it, and free it once completed. The following command queuing
functions need to be modified.
xhci_configure_endpoint()
xhci_address_device()
xhci_queue_slot_control()
xhci_queue_stop_endpoint()
xhci_queue_new_dequeue_state()
xhci_queue_reset_ep()
xhci_configure_endpoint()
xhci_configure_endpoint() could already be called with a command structure,
and only xhci_check_maxpacket and xhci_check_bandwidth did not do so. These
are changed and a command structure is now required. This change also simplifies
the configure endpoint command completion handling and the "goto bandwidth_change"
handling code can be removed.
In some cases the command queuing function is called in interrupt context.
These commands needs to be allocated atomically, and they can't wait for
completion. These commands will in this patch be freed directly after queuing,
but freeing will be moved to the command completion event handler in a later
patch once we get the global command queue up.(Just so that we won't leak
memory in the middle of the patch set)
Signed-off-by: Mathias Nyman <mathias.nyman@linux.intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Using the IS_ENABLED() macro can make the code shorter and easier to read.
Signed-off-by: Fabio Estevam <fabio.estevam@freescale.com>
Signed-off-by: Mathias Nyman <mathias.nyman@linux.intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
We have observed a rare cycle state desync bug after Set TR Dequeue
Pointer commands on Intel LynxPoint xHCs (resulting in an endpoint that
doesn't fetch new TRBs and thus an unresponsive USB device). It always
triggers when a previous Set TR Dequeue Pointer command has set the
pointer to the final Link TRB of a segment, and then another URB gets
enqueued and cancelled again before it can be completed. Further
investigation showed that the xHC had returned the Link TRB in the TRB
Pointer field of the Transfer Event (CC == Stopped -- Length Invalid),
but when xhci_find_new_dequeue_state() later accesses the Endpoint
Context's TR Dequeue Pointer field it is set to the first TRB of the
next segment.
The driver expects those two values to be the same in this situation,
and uses the cycle state of the latter together with the address of the
former. This should be fine according to the XHCI specification, since
the endpoint ring should be stopped when returning the Transfer Event
and thus should not advance over the Link TRB before it gets restarted.
However, real-world XHCI implementations apparently don't really care
that much about these details, so the driver should follow a more
defensive approach to try to work around HC spec violations.
This patch removes the stopped_trb variable that had been used to store
the TRB Pointer from the last Transfer Event of a stopped TRB. Instead,
xhci_find_new_dequeue_state() now relies only on the Endpoint Context,
requiring a small amount of additional processing to find the virtual
address corresponding to the TR Dequeue Pointer. Some other parts of the
function were slightly rearranged to better fit into this model.
This patch should be backported to kernels as old as 2.6.31 that contain
the commit ae63674714 "USB: xhci: URB
cancellation support."
Signed-off-by: Julius Werner <jwerner@chromium.org>
Cc: stable@vger.kernel.org
Signed-off-by: Mathias Nyman <mathias.nyman@linux.intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
This fixes TR dequeue validation failing on Intel XHCI controllers with the
following warning:
Mismatch between completed Set TR Deq Ptr command & xHCI internal state.
Interestingly enough reading the deq ptr from the ep ctx after a
TR Deq Ptr command does work on a Nec XHCI controller, it seems the Nec
writes the ptr to both the ep and stream contexts when streams are used.
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Sarah Sharp <sarah.a.sharp@linux.intel.com>
Nec XHCI controllers don't seem to care, but without this Intel XHCI
controllers reject Set TR dequeue commands with a COMP_TRB_ERR, leading
to the following warning:
WARN Set TR Deq Ptr cmd invalid because of stream ID configuration
And very shortly after this the system completely freezes.
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Sarah Sharp <sarah.a.sharp@linux.intel.com>
xhci maintains a radix tree for each stream endpoint because it must
be able to map a trb address to the stream ring. Each ring segment
must be added to the ring for this to work. Currently xhci sticks
only the first segment of each stream ring into the radix tree.
Result is that things work initially, but as soon as the first segment
is full xhci can't map the trb address from the completion event to the
stream ring any more -> BOOM. You'll find this message in the logs:
ERROR Transfer event for disabled endpoint or incorrect stream ring
This patch adds a helper function to update the radix tree, and a
function to remove ring segments from the tree. Both functions loop
over the segment list and handles all segments instead of just the
first.
[Note: Sarah changed this patch to add radix_tree_maybe_preload() and
radix_tree_preload_end() calls around the radix tree insert, since we
can now insert entries in interrupt context. There are now two helper
functions to make the code cleaner, and those functions are moved to
make them static.]
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Sarah Sharp <sarah.a.sharp@linux.intel.com>
This reverts commit f2d9b991c5.
We are ripping out commit 35773dac5f "usb:
xhci: Link TRB must not occur within a USB payload burst" because it's a
hack that caused regressions in the usb-storage and userspace USB
drivers that use usbfs and libusb. This commit attempted to fix the
issues with that patch.
Signed-off-by: Sarah Sharp <sarah.a.sharp@linux.intel.com>
Cc: stable@vger.kernel.org #3.12
This reverts commit e8b373326d. Many xHCI
host controllers can only handle 32-bit addresses, and writing 64-bits
at a time causes them to fail. Reading 64-bits at a time may also cause
them to return 0xffffffff, so revert this commit as well.
Signed-off-by: Sarah Sharp <sarah.a.sharp@linux.intel.com>
This reverts commit 7dd09a1af2.
Many xHCI host controllers can only handle 32-bit addresses, and writing
64-bits at a time causes them to fail. Rafał reports that USB devices
simply do not enumerate, and reverting this patch helps. Branimir
reports that his host controller doesn't respond to an Enable Slot
command and dies:
[ 75.576160] xhci_hcd 0000:03:00.0: Timeout while waiting for a slot
[ 88.991634] xhci_hcd 0000:03:00.0: Stopped the command ring failed, maybe the host is dead
[ 88.991748] xhci_hcd 0000:03:00.0: Abort command ring failed
[ 88.991845] xhci_hcd 0000:03:00.0: HC died; cleaning up
[ 93.985489] xhci_hcd 0000:03:00.0: Timeout while waiting for a slot
[ 93.985494] xhci_hcd 0000:03:00.0: Abort the command ring, but the xHCI is dead.
[ 98.982586] xhci_hcd 0000:03:00.0: Timeout while waiting for a slot
[ 98.982591] xhci_hcd 0000:03:00.0: Abort the command ring, but the xHCI is dead.
[ 103.979696] xhci_hcd 0000:03:00.0: Timeout while waiting for a slot
[ 103.979702] xhci_hcd 0000:03:00.0: Abort the command ring, but the xHCI is dead
Signed-off-by: Sarah Sharp <sarah.a.sharp@intel.com>
Reported-by: Rafał Miłecki <zajec5@gmail.com>
Reported-by: Branimir Maksimovic <branimir.maksimovic@gmail.com>
Cc: Xenia Ragiadakou <burzalodowa@gmail.com>
Commit 35773dac5f "usb: xhci: Link TRB
must not occur within a USB payload burst" attempted to fix an issue
found with USB ethernet adapters, and inadvertently broke USB storage
devices. The patch attempts to ensure that transfers never span a
segment, and rejects transfers that have more than 63 entries (or
possibly less, if some entries cross 64KB boundaries).
usb-storage limits the maximum transfer size to 120K, and we had assumed
the block layer would pass a scatter-gather list of 4K entries,
resulting in no more than 31 sglist entries:
http://marc.info/?l=linux-usb&m=138498190419312&w=2
That assumption was wrong, since we've seen the driver reject a write
that was 218 sectors long (of probably 512 bytes each):
Jan 1 07:04:49 jidanni5 kernel: [ 559.624704] xhci_hcd 0000:00:14.0: Too many fragments 79, max 63
...
Jan 1 07:04:58 jidanni5 kernel: [ 568.622583] Write(10): 2a 00 00 06 85 0e 00 00 da 00
Limit the number of scatter-gather entries to half a ring segment. That
should be margin enough in case some entries cross 64KB boundaries.
Increase the number of TRBs per segment from 64 to 256, which should
result in ring segments fitting on a 4K page.
Signed-off-by: Sarah Sharp <sarah.a.sharp@linux.intel.com>
Reported-by: jidanni@jidanni.org
References: http://bugs.debian.org/733907
Fixes: 35773dac5f ('usb: xhci: Link TRB must not occur within a USB payload burst')
Cc: stable <stable@vger.kernel.org> # 3.12
Change the default enumeration scheme for xhci attached non-SuperSpeed
devices from:
Reset
SetAddress [xhci address-device BSR = 0]
GetDescriptor(8)
GetDescriptor(18)
...to:
Reset
[xhci address-device BSR = 1]
GetDescriptor(64)
Reset
SetAddress [xhci address-device BSR = 0]
GetDescriptor(18)
...as some devices misbehave when encountering a SetAddress command
prior to GetDescriptor. There are known legacy devices that require
this scheme, but testing has found at least one USB3 device that fails
enumeration when presented with this ordering. For now, follow the ehci
case and enable 'new scheme' by default for non-SuperSpeed devices.
To support this enumeration scheme on xhci the AddressDevice operation
needs to be performed twice. The first instance of the command enables
the HC's device and slot context info for the device, but omits sending
the device a SetAddress command (BSR == block set address request).
Then, after GetDescriptor completes, follow up with the full
AddressDevice+SetAddress operation.
As mentioned before, this ordering of events with USB3 devices causes an
extra state transition to be exposed to xhci. Previously USB3 devices
would transition directly from 'enabled' to 'addressed' and never need
to underrun responses to 'get descriptor'. We do see the 64-byte
descriptor fetch the correct data, but the following 18-byte descriptor
read after the reset gets:
bLength = 0
bDescriptorType = 0
bcdUSB = 0
bDeviceClass = 0
bDeviceSubClass = 0
bDeviceProtocol = 0
bMaxPacketSize0 = 9
instead of:
bLength = 12
bDescriptorType = 1
bcdUSB = 300
bDeviceClass = 0
bDeviceSubClass = 0
bDeviceProtocol = 0
bMaxPacketSize0 = 9
which results in the discovery process looping until falling back to
'old scheme' enumeration.
Acked-by: Alan Stern <stern@rowland.harvard.edu>
Reported-by: David Moore <david.moore@gmail.com>
Suggested-by: Sarah Sharp <sarah.a.sharp@linux.intel.com>
Reported-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
Signed-off-by: Sarah Sharp <sarah.a.sharp@linux.intel.com>
Function xhci_write_64() is used to write 64bit xHC registers residing in MMIO.
On 32bit systems, xHC registers need to be written with 32bit accesses by
writing first the lower 32bits and then the higher 32bits. The header file
asm-generic/io-64-nonatomic-lo-hi.h ensures that on 32bit systems writeq() will
will write 64bit registers in 32bit chunks with low-high order.
Replace all calls to xhci_write_64() with calls to writeq().
This is done to reduce code duplication since 64bit low-high write logic
is already implemented and to take advantage of inherent "atomic" 64bit
write operations on 64bit systems.
Signed-off-by: Xenia Ragiadakou <burzalodowa@gmail.com>
Signed-off-by: Sarah Sharp <sarah.a.sharp@linux.intel.com>
Function xhci_read_64() is used to read 64bit xHC registers residing in MMIO.
On 32bit systems, xHC registers need to be read with 32bit accesses by
reading first the lower 32bits and then the higher 32bits.
Replace all calls to xhci_read_64() with calls to readq() and include
asm-generic/io-64-nonatomic-lo-hi.h header file, so that if the system
is not 64bit, readq() will read registers in 32bit chunks with low-high order.
This is done to reduce code duplication since 64bit low-high read logic
is already implemented and to take advantage of inherent "atomic" 64bit
read operations on 64bit systems.
Signed-off-by: Xenia Ragiadakou <burzalodowa@gmail.com>
Signed-off-by: Sarah Sharp <sarah.a.sharp@linux.intel.com>
Function xhci_writel() is used to write a 32bit value in xHC registers residing
in MMIO address space. It takes as first argument a pointer to the xhci_hcd
although it does not use it. xhci_writel() internally simply calls writel().
This creates an illusion that xhci_writel() is an xhci specific function that
has to be called in a context where a pointer to xhci_hcd is available.
Remove xhci_writel() wrapper function and replace its calls with calls to
writel() to make the code more straight-forward.
Signed-off-by: Xenia Ragiadakou <burzalodowa@gmail.com>
Signed-off-by: Sarah Sharp <sarah.a.sharp@linux.intel.com>
Function xhci_readl() is used to read 32bit xHC registers residing in MMIO
address space. It takes as first argument a pointer to the xhci_hcd although
it does not use it. xhci_readl() internally simply calls readl(). This creates
an illusion that xhci_readl() is an xhci specific function that has to be
called in a context where a pointer to xhci_hcd is available.
Remove the unnecessary xhci_readl() wrapper function and replace its calls to
with calls to readl() to make the code more straightforward.
Signed-off-by: Xenia Ragiadakou <burzalodowa@gmail.com>
Signed-off-by: Sarah Sharp <sarah.a.sharp@linux.intel.com>
SCT_FOR_CTX(p) is defined as (((p) << 1) & 0x7) in which case if we want
to set the stream context type to SCT_SSA_256 i.e 0x7 (although secondary
stream arrays are not yet supported) using this macro definition we will
get actually 0x6 which is not what we want.
This patch fixes the above issue by defining the SCT_FOR_CTX(p) macro as
(((p) & 0x7) << 1)
Signed-off-by: Xenia Ragiadakou <burzalodowa@gmail.com>
Signed-off-by: Sarah Sharp <sarah.a.sharp@linux.intel.com>
Andy Shevchenko