Once we start supporting uas hardware, and as more and more uas devices
become available, we will likely start seeing broken devices. This patch
prepares for the inevitable need for blacklisting those devices from
using the uas driver (they will use usb-storage instead).
Signed-off-by: NHans de Goede <hdegoede@redhat.com>
Signed-off-by: NSarah Sharp <sarah.a.sharp@linux.intel.com>

On disconnect USB3 protocol ports transit from U0 to SS.Inactive to Rx.Detect,
on a recoverable error, the port stays in SS.Inactive and we recover from it by
doing a warm-reset (through usb_device_reset if we have a udev for the port).

If this really is a disconnect we may end up trying the warm-reset anyways,
since khubd may run before the SS.Inactive to Rx.Detect transition, or it
may get skipped if the transition to Rx.Detect happens before khubd gets run.

With a loose connector, or in the case which actually led me to debugging this
bad ACPI firmware toggling Vbus off and on in quick succession, the port
may transition from Rx.Detect to U0 again before khubd gets run. In this case
the device state is unknown really, but khubd happily goes into the resuscitate
an existing device path, and the device driver never gets notified about the
device state being messed up.

If the above scenario happens with a streams using device, as soon as an urb
is submitted to an endpoint with streams, the following appears in dmesg:

ERROR Transfer event for disabled endpoint or incorrect stream ring
@0000000036807420 00000000 00000000 04000000 04078000

Notice how the TRB address is all zeros. I've seen this both on Intel
Pantherpoint and Nec xhci hosts.

Luckily we can detect the U0 to SS.Inactive to Rx.Detect to U0 all having
happened before khubd runs case since the C_LINK_STATE bit gets set in the
portchange bits on the U0 -> SS.Inactive change. This bit will also be set on
suspend / resume, but then it gets cleared by port_hub_init before khubd runs.

So if the C_LINK_STATE bit is set and a warm-reset is not needed, iow the port
is not still in SS.Inactive, and the port still has a connection, then the
device needs to be reset to put it back in a known state.

I've verified that doing the device reset also fixes the transfer event with
all zeros address issue.
Signed-off-by: NHans de Goede <hdegoede@redhat.com>
Signed-off-by: NSarah Sharp <sarah.a.sharp@linux.intel.com>

If streams are still allocated on device-reset or set-interface then the hcd
code implictly frees the streams. Clear host_endpoint->streams in this case
so that if a driver later tries to re-allocate them it won't run afoul of the
device already having streams check in usb_alloc_streams().

Note normally streams still being allocated at reset / set-intf  would be a
driver bug, but this can happen without it being a driver bug on reset-resume.
Signed-off-by: NHans de Goede <hdegoede@redhat.com>
Signed-off-by: NSarah Sharp <sarah.a.sharp@linux.intel.com>

Signed-off-by: NHans de Goede <hdegoede@redhat.com>
Signed-off-by: NSarah Sharp <sarah.a.sharp@linux.intel.com>

If we align segment dma pool memory to 64 bytes, then a segment can be located
at 0x10000040 - 0x1000043f, and a segment from another ring at 0x10000440 -
0x1000083f. The last trb in the first segment at 0x10000430 will then translate
to the same radix tree key as the first trb of the second segment, while they
are in different rings!

This patches fixes this by changing the alignment of the dma pool to be 1KB
rather then 64 bytes. An alternative fix would be to reduce the shift used
to calculate the radix tree keys, but that would (slighlty) grow the radix
trees so I believe this is the better fix.

Note this patch is mostly theoretical since in practice I've not seen
the dma_pool actually return not 1KB aligned memory.
Signed-off-by: NHans de Goede <hdegoede@redhat.com>
Signed-off-by: NSarah Sharp <sarah.a.sharp@linux.intel.com>

cmd_ring_reserved_trbs gets decremented by xhci_free_stream_info(), so set it
to 0 after freeing all rings, otherwise it wraps around to a very large value
when rings with streams are free-ed.

Before this patch the wrap-around could be triggered when xhci_resume
calls xhci_mem_cleanup if the controller resume fails.
Signed-off-by: NHans de Goede <hdegoede@redhat.com>
Signed-off-by: NSarah Sharp <sarah.a.sharp@linux.intel.com>

This is a preparation patch for teaching usb-storage to not bind to
uas devices.
Signed-off-by: NHans de Goede <hdegoede@redhat.com>
Signed-off-by: NSarah Sharp <sarah.a.sharp@linux.intel.com>

This is a preparation patch for teaching usb-storage to not bind to
uas devices.
Signed-off-by: NHans de Goede <hdegoede@redhat.com>
Signed-off-by: NSarah Sharp <sarah.a.sharp@linux.intel.com>

If we get ie 16 streams we can use stream-id 1-16, not 1-15.
Signed-off-by: NHans de Goede <hdegoede@redhat.com>
Signed-off-by: NSarah Sharp <sarah.a.sharp@linux.intel.com>

Signed-off-by: NHans de Goede <hdegoede@redhat.com>
Signed-off-by: NSarah Sharp <sarah.a.sharp@linux.intel.com>

Handle usb-device resets not triggered from uas_eh_bus_reset_handler(), when
this happens, disable cmd queuing during the reset, and wait for existing
requests to finish in pre_reset.
Signed-off-by: NHans de Goede <hdegoede@redhat.com>
Signed-off-by: NSarah Sharp <sarah.a.sharp@linux.intel.com>

Fix the uas_eh_bus_reset_handler not properly taking the usbdev lock
before calling usb_device_reset, the usb-core expects this lock to be
taken when usb_device_reset is called.
Signed-off-by: NHans de Goede <hdegoede@redhat.com>
Signed-off-by: NSarah Sharp <sarah.a.sharp@linux.intel.com>

I thought it would be a good idea to also test uas with usb-2, and it turns out
it was, as it did not work. The problem is that the uas driver was passing the
bEndpointAddress' direction bit to usb_rcvbulkpipe, the xhci code seems to not
care about this, but with the ehci code this causes usb_submit_urb failure.

With this fixed the uas code works nicely with an uas device plugged into
an ehci port.
Signed-off-by: NHans de Goede <hdegoede@redhat.com>
Signed-off-by: NSarah Sharp <sarah.a.sharp@linux.intel.com>

The cmd endpoint never has streams, so the stream_id parameter is unused.
Signed-off-by: NHans de Goede <hdegoede@redhat.com>
Signed-off-by: NSarah Sharp <sarah.a.sharp@linux.intel.com>

All callers of unlink_data_urbs drop devinfo->lock before calling it, and
then immediately take it again after the call. And the first thing
unlink_data_urbs does is take the lock again, and the last thing it does
is drop it. This commit removes all the unnecessary lock dropping and taking.
Signed-off-by: NHans de Goede <hdegoede@redhat.com>
Signed-off-by: NSarah Sharp <sarah.a.sharp@linux.intel.com>

- Rename labels to properly reflect this
- Don't skip free-ing the streams when scsi_init_shared_tag_map fails
Signed-off-by: NHans de Goede <hdegoede@redhat.com>
Signed-off-by: NSarah Sharp <sarah.a.sharp@linux.intel.com>

Otherwise they may complete before they get anchored and thus never get
unanchored (as the unanchoring is done by the usb core on completion).

This commit also remove the usb_get_urb / usb_put_urb around cmd submission +
anchoring, since if done in the proper order this is not necessary.
Signed-off-by: NHans de Goede <hdegoede@redhat.com>
Signed-off-by: NSarah Sharp <sarah.a.sharp@linux.intel.com>

Signed-off-by: NGerd Hoffmann <kraxel@redhat.com>
Signed-off-by: NHans de Goede <hdegoede@redhat.com>
Signed-off-by: NSarah Sharp <sarah.a.sharp@linux.intel.com>

This patch adds a new list where all requests which are canceled are
added to, so we don't loose them.  Then, after killing all inflight
urbs on bus reset (and disconnect) we'll walk over the list and clean
them up.

Without this we can end up with aborted requests lingering around in
case of status pipe transfer errors.
Signed-off-by: NGerd Hoffmann <kraxel@redhat.com>
Signed-off-by: NHans de Goede <hdegoede@redhat.com>
Signed-off-by: NSarah Sharp <sarah.a.sharp@linux.intel.com>

Simplifies locking, we'll protect the list with the device spin lock.
Also plugs races which can happen when two devices operate on the
global list.

While being at it rename the list head from "list" to "work", preparing
for the addition of a second list.
Signed-off-by: NGerd Hoffmann <kraxel@redhat.com>
Signed-off-by: NHans de Goede <hdegoede@redhat.com>
Signed-off-by: NSarah Sharp <sarah.a.sharp@linux.intel.com>

Signed-off-by: NGerd Hoffmann <kraxel@redhat.com>
Signed-off-by: NHans de Goede <hdegoede@redhat.com>
Signed-off-by: NSarah Sharp <sarah.a.sharp@linux.intel.com>

This allows userspace to use bulk-streams, just like in kernel drivers, see
Documentation/usb/bulk-streams.txt for details on the in kernel API. This
is exported pretty much one on one to userspace.

To use streams an app must first make a USBDEVFS_ALLOC_STREAMS ioctl,
on success this will return the number of streams available (which may be
less then requested). If there are n streams the app can then submit
usbdevfs_urb-s with their stream_id member set to 1-n to use a specific
stream. IE if USBDEVFS_ALLOC_STREAMS returns 4 then stream_id 1-4 can be
used.

When the app is done using streams it should call USBDEVFS_FREE_STREAMS

Note applications are advised to use libusb rather then using the
usbdevfs api directly. The latest version of libusb has support for streams.
Signed-off-by: NHans de Goede <hdegoede@redhat.com>
Signed-off-by: NSarah Sharp <sarah.a.sharp@linux.intel.com>

Signed-off-by: NHans de Goede <hdegoede@redhat.com>
Signed-off-by: NSarah Sharp <sarah.a.sharp@linux.intel.com>

This patch makes it possible to specify a bulk stream id when submitting
an urb using the async usbfs API. It overloads the number_of_packets
usbdevfs_urb field for this. This is not pretty, but given other
constraints it is the best we can do. The reasoning leading to this goes
as follows:

1) We want to support bulk streams in the usbfs API
2) We do not want to extend the usbdevfs_urb struct with a new member, as
   that would mean defining new ioctl numbers for all async API ioctls +
   adding compat versions for the old ones (times 2 for 32 bit support)
3) 1 + 2 means we need to re-use an existing field
4) number_of_packets is only used for isoc urbs, and streams are bulk only
   so it is the best (and only) candidate for re-using

Note that:
1) This patch only uses number_of_packets as stream_id if the app has
   actually allocated streams on the ep, so that old apps which may have
   garbage in there (as it was unused until now in the bulk case), will not
   break
2) This patch does not add support for allocating / freeing bulk-streams, that
   is done in a follow up patch
Signed-off-by: NHans de Goede <hdegoede@redhat.com>
Signed-off-by: NSarah Sharp <sarah.a.sharp@linux.intel.com>

This is a preparation patch for adding support for bulk streams.
Signed-off-by: NHans de Goede <hdegoede@redhat.com>
Signed-off-by: NSarah Sharp <sarah.a.sharp@linux.intel.com>

The usb_set_interface documentation says:

 * Also, drivers must not change altsettings while urbs are scheduled for
 * endpoints in that interface; all such urbs must first be completed
 * (perhaps forced by unlinking).

For in kernel drivers we trust the drivers to get this right, but we
cannot trust userspace to get this right, so enforce it by killing any
urbs still pending on the interface.
Signed-off-by: NHans de Goede <hdegoede@redhat.com>
Signed-off-by: NSarah Sharp <sarah.a.sharp@linux.intel.com>

Documentation/usb/bulk-streams.txt says:

All stream IDs will be deallocated when the driver releases the interface, to
ensure that drivers that don't support streams will be able to use the endpoint

This commit actually implements this.
Signed-off-by: NHans de Goede <hdegoede@redhat.com>
Signed-off-by: NSarah Sharp <sarah.a.sharp@linux.intel.com>

This is a preparation patch for adding support for bulk streams to usbfs.
Signed-off-by: NHans de Goede <hdegoede@redhat.com>
Signed-off-by: NSarah Sharp <sarah.a.sharp@linux.intel.com>

So that it can be used in other places too.
Signed-off-by: NHans de Goede <hdegoede@redhat.com>
Signed-off-by: NSarah Sharp <sarah.a.sharp@linux.intel.com>

Signed-off-by: NHans de Goede <hdegoede@redhat.com>
Signed-off-by: NSarah Sharp <sarah.a.sharp@linux.intel.com>

If we're expanding a stream ring, we want to make sure we can add those
ring segments to the radix tree that maps segments to ring pointers.
Try the radix tree insert after the new ring segments have been allocated
(the last segment in the new ring chunk will point to the first newly
allocated segment), but before the new ring segments are linked into the
old ring.

If insert fails on any one segment, remove each segment from the radix
tree, deallocate the new segments, and return.  Otherwise, link the new
segments into the tree.

HdG: Add a check to only update stream mappings in xhci_ring_expansion when
the ring is a stream ring.
Signed-off-by: NSarah Sharp <sarah.a.sharp@linux.intel.com>
Signed-off-by: NHans de Goede <hdegoede@redhat.com>
Signed-off-by: NSarah Sharp <sarah.a.sharp@linux.intel.com>

The ss_ep_comp bmAttributes filed can contain more info then just the
streams, use usb_ss_max_streams to properly get max streams.
Signed-off-by: NHans de Goede <hdegoede@redhat.com>
Signed-off-by: NSarah Sharp <sarah.a.sharp@linux.intel.com>

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: NHans de Goede <hdegoede@redhat.com>
Signed-off-by: NSarah 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: NHans de Goede <hdegoede@redhat.com>
Signed-off-by: NSarah Sharp <sarah.a.sharp@linux.intel.com>

Signed-off-by: NHans de Goede <hdegoede@redhat.com>
Signed-off-by: NSarah Sharp <sarah.a.sharp@linux.intel.com>

Before this a device needing ie 32 stream ctxs would end up with an entry from
the small_streams_pool which has 256 bytes entries, where as 32 stream ctxs
need 512 bytes. Things actually keep running for a surprisingly long time
before crashing because of this.
Signed-off-by: NHans de Goede <hdegoede@redhat.com>
Signed-off-by: NSarah Sharp <sarah.a.sharp@linux.intel.com>

And warn about this, as that would be a driver bug.

Like wise drivers should ensure that streams are properly free-ed before a
device is reset. So lets warn about that too. This already causes warnings
in the form of:

[   96.982398] xhci_hcd 0000:01:00.0: WARN Can't disable streams for endpoint 0x81
, streams are already disabled!
[   96.982400] xhci_hcd 0000:01:00.0: WARN xhci_free_streams() called with non-streams endpoint

But it is better to also warn about the actual cause of this later warnings.
Signed-off-by: NHans de Goede <hdegoede@redhat.com>
Signed-off-by: NSarah 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: NGerd Hoffmann <kraxel@redhat.com>
Signed-off-by: NHans de Goede <hdegoede@redhat.com>
Signed-off-by: NSarah Sharp <sarah.a.sharp@linux.intel.com>

This changes debug messages and warnings in xhci-ring.c
to be on a single line so grep can find them. grep must
have precedence over the 80 column limit.

[Sarah fixed two checkpatch.pl issues with split lines
introduced by this commit.]
Signed-off-by: NOliver Neukum <oneukum@suse.de>
Signed-off-by: NSarah Sharp <sarah.a.sharp@linux.intel.com>

The xHCI driver currently uses a USB core internal field,
udev->lpm_capable, to indicate the xHCI driver knows how to calculate
the LPM timeout values.  If this value is set for the host controller
udev, it means Link PM can be enabled for child devices under that host.

Change the code so the xHCI driver isn't mucking with USB core internal
fields.  Instead, indicate the xHCI driver doesn't support Link PM on
this host by clearing the U1 and U2 exit latencies in the roothub
SuperSpeed Extended Capabilities BOS descriptor.

The code to check for the roothub setting U1 and U2 exit latencies to
zero will also disable LPM for external devices that do that same.  This
was already effectively done with commit
ae8963ad "usb: Don't enable LPM if the
exit latency is zero."  Leave that code in place, so that if a device
sets one exit latency value to zero, but the other is set to a valid
value, LPM is only enabled for the U1 or U2 state that had the valid
value.  This is the same behavior the code had before.

Also, change messages about missing Link PM information from warning
level to info level.  Only print a warning about the first device that
doesn't support LPM, to avoid log spam.  Further, cleanup some
unnecessary line breaks to help people to grep for the error messages.
Signed-off-by: NSarah Sharp <sarah.a.sharp@linux.intel.com>
Cc: Alan Stern <stern@rowland.harvard.edu>