We've the same info doubled in both the inflight list and the cmnd array,
drop the list.
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The data urbs are all killed before calling zap_pending, and their completion
handler should have cleared their inflight flag.
Do not 0 the data inflight flags, and add a check for try_complete succeeding,
as it should always succeed when called from zap_pending.
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Drop the whole dance with first moving cmnds to a dead-list. The resetting
flag ensures that no new cmds / urbs will be submitted, and that any urb
completions are short-circuited without trying to complete the scsi cmnd.
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Now that we no longer drop our lock to unlink the data urbs, we can simply
free them on completion, making their handling consistent with the other urbs.
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
There is no need for all the trickery with dropping the lock, we can
simply reference the urbs while we hold the lock to ensure the urbs don't
disappear beneath us, and do the actual unlink (+ unreference) after we've
dropped the lock.
This also fixes a race where we may loose of cmnd ownership to the scsi
midlayer without holding the lock due to the midlayer re-claiming ownership
through an abort (which will be handled by a future patch in this series).
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The status urb should not complete before the command has been submitted, nor
should we get a second status urb for the same tag after a IU_ID_STATUS.
Data urbs should not complete before the command has been submitted, but may
complete after the IU_ID_STATUS.
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Using scsi_host_find_tag with tags returned by the device is unsafe for
multiple reasons:
1) It returns tags->rqs[tag], which may be non NULL even when the cmnd is
not owned by us
2) It returns tags->rqs[tag], without holding any locks protecting it
3) It returns tags->rqs[tag], without doing any boundary checking
Instead keep our own list which maps tags -> inflight cmnds.
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Factor out the mapping of scsi-tags -> uas-tags/stream-ids to a helper function
so that there is a single place where this "magic" happens.
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
- Make sure we always hold the lock when setting / checking resetting
- Check resetting before checking urb->status
- Add missing check for resetting to uas_data_cmplt
- Add missing check for resetting to uas_do_work
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
There are various bug reports about oopses / hangs with the uas driver,
which all point to the abort-command and logical-unit-reset (task-management)
error handling paths.
Getting these right is very hard, there are quite a few corner cases, and
testing is almost impossible since under normal operation these code paths
are not used at all.
Another problem is that there are also some cases where it simply is not clear
what to do at all. E.g. over usb-2 multiple outstanding commands share the same
endpoint. What if a command gets aborted while its sense urb is half way
through completing (so some data has been transfered but not all). Since the
urb is not yet complete we don't know if the sense urb is actually for this
command, or for one of the other oustanding commands. If it is for one of the
other commands and we cancel it, then we end up in an undefined state. But if
it is actually for the command we're aborting, and the abort succeeds, then it
may never complete...
This exact same problem applies to logical unit resets too, if there are
multiple luns, then commands outstanding on both luns share the sense
endpoint. If there is only a single lun, then doing a logical unit reset is
little better then doing a full usb device reset.
So summarizing because:
1) abort / lun-reset is very tricky to get right
2) Not being able to test the tricky code, which means it will have bugs
3) This being a code path which under normal operation will never happen,
so being slow / sub-optimal here is not really an issue
4) Under error conditions we will still be able to recover through usb
device resets.
5) This may be a bit slower in some cases, but this is actually faster in
cases where the bridge ship has locked up, which seems to be the most
common error case sofar.
This commit removes the abort / lun-reset error handling paths, and also the
taks-mgmt code since those are the only 2 task-mgmt users. Leaving only the
(tested and testable) usb-device-reset error handling path in place.
Note I realize that this is somewhat of a big hammer, but currently people
are seeing very hard to debug oopses with uas. First let focus on making uas
work reliable, then we can later look into adding more fine grained error
handling.
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Besides the ASM1051 (*) needing sdev->no_report_opcodes = 1, it turns out that
the JMicron JMS567 also needs it to work properly with uas (usb-storage always
sets it). Since some of the scsi devs were not to keen on the idea to
outrightly set sdev->no_report_opcodes = 1 for all uas devices, so add a quirk
for this, and set it for the JMS567.
*) Which has become a non-issue since we've completely blacklisted uas on
the ASM1051 for other reasons
Cc: stable@vger.kernel.org
Reported-and-tested-by: Claudio Bizzarri <claudio.bizzarri@gmail.com>
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
And set this quirk for the Seagate Expansion Desk (0bc2:2312), as that one
seems to hang upon receiving an ATA_12 or ATA_16 command.
https://bugzilla.kernel.org/show_bug.cgi?id=79511https://bbs.archlinux.org/viewtopic.php?id=183190
While at it also add missing documentation for the u value for usb-storage
quirks.
Cc: stable@vger.kernel.org # 3.16, 3.17
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
--
Changes in v2: Add documentation for new t and u usb-storage.quirks flags
Changes in v3: Fix typo in documentation
Changes in v4: Also apply the quirk to (0bc2:3312)
Changes in v5: Rebased on 3.17-rc5, drop u documentation, already upstream
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
on some architecture spin_is_locked() always return false in
uniprocessor configuration and therefore it would be advise
to replace with lockdep_assert_held().
Signed-off-by: Sanjeev Sharma <Sanjeev_Sharma@mentor.com>
Acked-by: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Some jmicron uas chipsets act up (they disconnect from the bus) when sending
more then 32 commands to them at once.
Rather then building an ever growing list with usb-id based quirks for
devices using this chipset, simply reduce the qdepth to 32 when connected
over usb-2. 32 should be plenty to keep things close to maximum
possible throughput on usb-2.
Cc: stable@vger.kernel.org
Tested-and-reported-by: Laszlo T. <tlacix@gmail.com>
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
There are also two allocations with GFP_KERNEL in the pre-/post_reset
code paths. That is no good because that is a part of the SCSI error handler.
Signed-off-by: Oliver Neukum <oliver@neukum.org>
Reviewed-by: Hans de Goede <hdegoede@redhat.com>
Acked-by: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
intfdata is set only after scsi_scan(). uas_pre_reset() however
needs intfdata to be valid and will follow the NULL pointer
killing khubd. intfdata must be preemptively set before the
host is registered and undone in the error case.
Signed-off-by: Oliver Neukum <oliver@neukum.org>
Reviewed-by: Hans de Goede <hdegoede@redhat.com>
Acked-by: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Quote Dan:
The patch e36e64930cff: "uas: Use GFP_NOIO rather then GFP_ATOMIC
where possible" from Nov 7, 2013, leads to the following static
checker warning:
drivers/usb/storage/uas.c:806 uas_eh_task_mgmt()
error: scheduling with locks held: 'spin_lock:lock'
Some other allocations under spinlock are not caught.
The fix essentially reverts e36e64930c
Signed-off-by: Oliver Neukum <oliver@neukum.org>
Reported-by: Dan Carpenter <dan.carpenter@oracle.com>
Reviewed-by: Hans de Goede <hdegoede@redhat.com>
Acked-by: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Although an interesting concept, I don't think that this is a good idea:
-This will result in lots of "virtual" scsi controllers confusing users
-If we get a scsi-bus-reset we will now need to do a usb-device-reset of all
uas devices on the same usb bus, which is something to avoid if possible
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Sarah Sharp <sarah.a.sharp@linux.intel.com>
At the kernel-summit Sarah Sharp asked me if I was willing to become the
uas maintainer. I said yes, and here is a patch to make this official.
Also remove Matthew Wilcox and Sarah Sharp as maintainers at their request.
I've also added myself to the module's author tag, so that if people look there
rather then in maintainers they will know they should bug me about uas too.
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Sarah Sharp <sarah.a.sharp@linux.intel.com>
Copy the sg alignment trick from the usb-storage driver, without this I'm
seeing intermittent errors when using uas devices with an ehci controller.
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Sarah Sharp <sarah.a.sharp@linux.intel.com>
The scsi error handling path re-uses previously queued up (and errored-out)
cmds. If such a re-used cmd had a data-phase then cmdinfo will have
data_in_urb / data_out_urb still set to the free-ed urbs from the errored-out
cmd, and they will get free-ed a second time when the error handling cmd
completes, corrupting the kernel heap.
Clearing cmdinfo on command queue-ing fixes this, and seems like a good idea
in general.
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Sarah Sharp <sarah.a.sharp@linux.intel.com>
The scsi-host structure is refcounted, scsi_remove_host tears down the
scsi-host but does not decrement the refcount, so we need to call
scsi_put_host on disconnect to get the underlying memory to be freed.
After calling scsi_remove_host, the scsi-core may still hold a reference to
the scsi-host, iow we may still get called after uas_disconnect, but we
do our own life cycle management of uas_devinfo, freeing it on disconnect,
and thus may end up using devinfo after it has been freed. Switch to letting
scsi_host_alloc allocate and manage the memory for us.
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Sarah Sharp <sarah.a.sharp@linux.intel.com>
cmds are either on the inflight list or on the dead list, never both, so
we only need one list head.
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Sarah Sharp <sarah.a.sharp@linux.intel.com>
Before this commit the uas driver would keep track of scsi commands which still
need to have some urbs submitted to the device, and complete this with an
ABORT result code on bus-reset or disconnect, but in flight scsi commands
which have all their urbs submitted, and thus are not part of the work list,
would never get their done callback called.
The problem is killed sense urbs don't have any tag info, so it is impossible
to tell which scsi cmd they belong to, so merely making sure all the urbs
have completed one way or the other is not enough.
This commit fixes this by changing the work list to an inflight list, which
keeps tracks of all inflight scsi cmnds, using the IS_IN_WORK_LIST flag to
determine if actual work needs to be done in uas_do_work(), and by moving
marking all inflight scsi commands as aborted and moving them to the dead list
on bus-reset or disconnect.
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Sarah Sharp <sarah.a.sharp@linux.intel.com>
uas_alloc_data_urb always gets called with a stream_id value of 0 when not
using streams. Removing the check makes it consistent with uas_alloc_sense_urb.
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Sarah Sharp <sarah.a.sharp@linux.intel.com>
For USB-2 connections the stream-id must always be 0.
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Sarah Sharp <sarah.a.sharp@linux.intel.com>
Some BIOS-es will hang on reboot when an uas device is attached and left in
uas mode on reboot.
This commit adds a shutdown handler which on reboot puts the device back into
usb-storage mode, fixing the hang on reboot on these systems.
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Sarah Sharp <sarah.a.sharp@linux.intel.com>
We can sleep in our own workqueue (which is the whole reason for having
it), and scsi error handlers are also always called from a context which
may sleep.
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Sarah Sharp <sarah.a.sharp@linux.intel.com>
Since we use a fixed tag / stream for tasks we cannot allow more then one
to run at the same time. This could happen before this time if a task timed
out.
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Sarah Sharp <sarah.a.sharp@linux.intel.com>
The fixed endpoint config code was only necessary to deal with an early
uas prototype which has never been released, so lets drop it and enforce
proper uas endpoint descriptors.
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Sarah Sharp <sarah.a.sharp@linux.intel.com>
The loop uses up to 3 bytes of the endpoint extra data.
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Sarah Sharp <sarah.a.sharp@linux.intel.com>
This is a preparation patch for adding better descriptor validation.
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Sarah Sharp <sarah.a.sharp@linux.intel.com>
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: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Sarah 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: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Sarah 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: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Sarah 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: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Sarah 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: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Sarah 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: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Sarah 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: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Sarah Sharp <sarah.a.sharp@linux.intel.com>
The cmd endpoint never has streams, so the stream_id parameter is unused.
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Sarah 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: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Sarah 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: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Sarah 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: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Sarah Sharp <sarah.a.sharp@linux.intel.com>
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 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: 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>
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: 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>
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>
256 luns is what the sam-4 address method 0 can handle and what
the qemu uas emulation supports. So pick that for now.
[ v2: unlike the other two max_* fields max_channel isn't max+1 ]
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Add new function to unlink and abort requests from the work
list, call it on bus reset and disconnect where we kill all
in-flight urbs. Also reorder calls in disconnect to first
cancel transfers, then remove the scsi hba.
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Two changes. First we check whenever the request is linked in the work
list and if so take it out. Second check whenever the command is
actually in flight before asking the device to cancel it via task
management, and in case it isn't just zap the data urbs and finish it.
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Keep track whenever the request is linked into the work list or not.
Needed for request abort.
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
uas_unlink_data_urbs uses this to make sure the the scsi command is
not released while looking at it. This will be needed when we start
calling uas_unlink_data_urbs in the request cancel code paths.
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Add uas_unlink_data_urbs function to cancel in-flight data urbs.
Moves existing code into a separate function.
[ v2: also drop the locking, just call usb_unlink_urb no matter what,
which is safe because the usb core guarantees the completion
callback is called only once ]
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Streamline control flow so it is easier for gcc to follow which paths
can be taken and which can't.
Fixes "warning: 'cmdinfo' may be used uninitialized in this function"
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
Reported-by: Fengguang Wu <fengguang.wu@intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Properly report aborted commands.
Also don't access cmdinfo after kicking task management,
it may not be valid any more once it returns.
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Allocate one tag for task management functions and
use it properly.
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Add task management support, wind up in abort and device reset error
handlers. Cancel all in-flight urbs in bus reset handler.
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Use separate anchors for data and sense urbs, which
I think will be useful when implementing error handling.
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
(1) Handle data pipe errors: When the data urb failed we
didn't transfer anything, update scsi_cmnd accordingly.
(2) Cancel data transfers when we got back an error on the
status pipe.
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Set state bits after submitting data urbs & command urbs, so we know
what is in flight. Clear data bits when the data urb is finished, clear
command bit when we see the status urb for the command. Finish the scsi
command after running both status and data completion handlers for the
command.
Add a cmd status logging function for debugging purposes. Hook it into
the error handler, so we see in the log what status a command is in
which the scsi layer wants cancel.
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Stop reusing sense urbs, just allocate a fresh one each time and free it
when done.
Stop storing a sense urb pointer in the scsi request, all you can do
with it is misusing. For example requeuing the sense urb, then f*ck it
up by picking the wrong one in case tagged requests don't finish in the
same order you've submitted them. Also note that (not-yet supported)
task management ops don't have a scsi request in the first place.
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
This reverts commit ceb3f91fd5.
IMO the real bug is assigning status urbs to scsi requests. First there
is no such link in the non-stream case. Also there isn't nessesarely a
scsi request in the first place, for example when submitting task
management requests.
This patch just papers over the real bug and introduces different status
urb handling in the stream/non-stream case for no good reason.
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
This reverts commit e4d8318a85.
This patch makes uas.c call usb_unlink_urb on data urbs. The data urbs
get freed in the completion callback. This is illegal according to the
usb_unlink_urb documentation.
This patch also makes the code expect the data completion callback
being called before the status completion callback. This isn't
guaranteed to be the case, even though the actual data transfer should
be finished by the time the status is received.
Background: The ehci irq handler for example only know that there are
finished transfers, it then has go check the QHs & TDs to see which
transfers did actually finish. It has no way to figure in which order
the transfers did complete. The xhci driver can call the callbacks in
completion order thanks to the event queue. This does nicely explain
why the driver is solid on a (usb2) xhci port whereas it goes crazy on
ehci in my testing.
Signed-off-by: Gerd Hoffmann <kraxel@redhat.com>
Cc: stable <stable@vger.kernel.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Just run into the following:
- new disk arrived in the system
- udev couldn't wait to get its hands on to to run ata_id /dev/sda
- this sent the cdb 0xa1 to the device.
- my UAS-gadget recevied the cdb and had no idea what to do with it. It
decided to send a status URB back with sense set to invalid opcode.
- the host side received it status and completed the scsi command.
- the host sent another scsi with 4kib data buffer
- Now I was confused why the data transfer is only 512 bytes instead of
4kib since the host is always allocating the complete transfer in one
go.
- Finally the system crashed while walking through the sg list.
This patch adds three new flags in order to distinguish between DATA
URB completed and outstanding. If we receive status before data, we
cancel data and let data complete the command.
This solves the problem for IN and OUT transfers but does not work for
BIDI.
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Sarah Sharp <sarah.a.sharp@linux.intel.com>
The protocol specific structures and defines which are used by UAS are
moved into a header files by this patch so it can be accessed by the UAS
gadget as well.
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Sarah Sharp <sarah.a.sharp@linux.intel.com>
The UAS driver requires SG support by the HCD operating the device. This
patch stops UAS from operating on a HCD without sg support and prints a
message to let him know.
The spec says:
|For [USB2] backward compatibility, the device shall present [BOT] as
|alternate interface zero (primary) and [UAS] as alternate interface one
|(secondary). A device which does not need backward compatibility with
|[BOT] shall present [UAS] as alternate interface zero. In [USB2]
|systems, the [BOT] driver or an associated filter driver may need to
|issue a SET INTERFACE request for alternate interface one and then allow
|the [UAS] driver to load.
If the user used usb_modeswitch to switch to UAS then he can go back to
BOT or use a different HCD. In case UAS is the only interface then there
is currently no way out.
In future usb_sg_wait() should be extended to provide a non-blocking
interface so it can work with the UAS driver.
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Sarah Sharp <sarah.a.sharp@linux.intel.com>
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.11 (GNU/Linux)
iQIcBAABAgAGBQJO8x85AAoJEHuW6BYqjPXRXWgQAI/+xBuTNKwfe/zAcIYxH1I5
Lq3jQaaqZ/TYWMZH2BwHYPXLLayLpHyvYSmub4hJm9okPp/iWJeE6nnYg8Ige8CD
jmpGa8p25pvdP+Rn8GW0EVIA+X0ZVdUV1llz88+21piY4ydobBQOtUtXAIl5jsBv
EVjqCZ6kmn0X7utdutpTHQ421/5k/9x5bh7mHOQTLiMcouXDyA/KVYjet4DVt5PU
Umqrk+uOMUjnygShRZ3C+FLBi/cfJ2jKUQo34dJbhxtD4hZyDe3CXM0UKnvZ5dso
EAykudlN32L29vT000F5KCocEUjQfnqvlBKK+acI7gvdmHCerDn0X/luHx/skSQV
li+pqCqfcfnX6Xg68WWLod9xrt5TtUlD1vAwhvi+ObDIFcVkCqLIdD5lBl+pZz3s
L2j/NgxVkAEVr4VotCNCHCAw17nWYSyle7Px5rLzU1+gfRFTGXiepBxnp52ZPXFf
l+rM+U+OZzrEtuXQq2AEIqh07G4EeovEigWB2v0Eyl77ZmE+SprMHoFVLTtB5Xqc
iau4YMG3nuJdIEyxIHMTrIX91iEgKFfo2Zc030GIe1EH7Gllr8VSYnoQLZf6/tF7
D4NzlqAVQUd6RIDlwQn6XpOGm4FLMF0rJL+Di3QqZ1vk/CRfSmGPSP7mFLbaCNOg
YhMYuzWknxi80kYRMZgy
=GTUe
-----END PGP SIGNATURE-----
Merge tag 'uas_for_sarah' of git://linutronix.de/users/bigeasy/linux into for-uas-next
Merge UAS bug fixes from Sebastian Andrzej Siewior, including some patches of
mine that he signed.
UAS fixes for Sarah
The status/sense URB is allocated on per-command basis. A read/write
looks the following way on a stream-less connection:
- send cmd tag X, queue status
- receive status, oh it is a read for tag X. queue status & read
- receive read
- receive status, oh I'm done for tag X. Cool call complete and free
status urb.
This block repeats itself 1:1 for further commands and looks great so
far. Lets take a look now what happens if we do allow multiple commands:
- send cmd tag X, queue statusX (belongs to the command with the X tag)
- send cmd tag Y, queue statusY (belongs to the command with the Y tag)
- receive statusX, oh it is a read for tag X. queue statusX & a read
- receive read
- receive statusY, oh I'm done for tag X. Cool call complete and free statusY.
- receive statusX, oh it is a read for tag Y. queue statusY & before we
queue the read the the following message can be observed:
|sd 0:0:0:0: [sda] sense urb submission failure
followed by a second attempt with the same result.
In order to address this problem we will use only one status URB for
each scsi host in case we don't have stream support (as suggested by
Matthew). This URB is requeued until the device removed. Nothing changes
on stream based endpoints.
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
In "usb/uas: use unique tags for all LUNs" we make sure to create unique
tags across all LUNs. This patch uses scsi_host_find_tag() to obtain the
correct command which is associated with the tag.
The following changes are required:
- don't use sdev->current_cmnd anymore
Since we can have devices which don't support command queueing we must
ensure that we can tell the two commands apart. Even if a device
supports comand queuing we send the INQUIRY command "untagged" for
LUN1 while we can send a tagged command to LUN0 at the same time.
devinfo->cmnd is used for stashing the one "untagged" command.
- tag number is altered. If stream support is used then the tag number
must match the stream number. Therefore we can't use tag 0 and must
start at tag 1.
In case we have untagged commands (at least the first command) we must
be able to distinguish between command tag 0 (which becomes 1) and
untagged command (which becomes curently also 1).
The following tag numbers are used:
0: never
1: for untagged commands (devinfo->cmnd)
2+: tagged commands.
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
I observed that on a device with multiple LUNs UAS was re-using the same
tag number for requests which were issued at the same time to both LUNs.
This patch uses scsi_init_shared_tag_map() to use unique tags for all
LUNs. With this patch I haven't seen the same tag number during the init
sequence anymore. Tag 1 is used for devices which do not adverise
command queueing.
This patch initilizes the queue before adding the scsi host like the
other two user in tree.
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
In the UAS status URB completion handler, we need to free the URB, no
matter what happens. Fix a bug where we would leak the URB (and its
buffer) if we couldn't find a SCSI command that is associated with this
status phase.
Signed-off-by: Sarah Sharp <sarah.a.sharp@linux.intel.com>
Cc: Matthew Wilcox <willy@linux.intel.com>
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
UAS can work with either USB 3.0 devices that support bulk streams, or
USB 2.0 devices that do not support bulk streams. When we're working
with a non-streams device, we need to be able to uniquely identify a
SCSI command with a tag in the IU. Devices will barf and abort all
queued commands if they find a duplicate tag.
uas_queuecommand_lck() sets cmdinfo->stream to zero if the device
doesn't support streams, which is later passed into uas_alloc_cmd_urb()
as the variable stream. This means the UAS driver was setting the tag
in all commands to zero for non-stream devices. So the UAS driver won't
currently work with USB 2.0 devices.
Use the SCSI command tag instead of the stream ID for the command IU
tag. We have to add one to the SCSI command tag because SCSI tags are
zero-based, but stream IDs are one-based, and the command tag must match
the stream ID that we're queueing the data IUs for. Untagged SCSI
commands use stream ID 1.
Signed-off-by: Sarah Sharp <sarah.a.sharp@linux.intel.com>
Cc: Matthew Wilcox <willy@linux.intel.com>
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
If the original submission (or allocation) of the URBs for a SCSI
command fails, the UAS driver sticks the command structure in a
workqueue and schedules uas_do_work() to run. That function removes the
entire queue before walking across it and attempting to resubmit.
Unfortunately, if the second submission fails, we will leak memory
(because an allocated URB was not submitted) and possibly leave the SCSI
command partially enqueued on some of the stream rings. Fix this by
checking whether the second submission failed and re-queueing the
command to the UAS workqueue and scheduling it.
Signed-off-by: Sarah Sharp <sarah.a.sharp@linux.intel.com>
Cc: Matthew Wilcox <willy@linux.intel.com>
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
This converts the drivers in drivers/usb/* to use the
module_usb_driver() macro which makes the code smaller and a bit
simpler.
Added bonus is that it removes some unneeded kernel log messages about
drivers loading and/or unloading.
Cc: Simon Arlott <cxacru@fire.lp0.eu>
Cc: Duncan Sands <duncan.sands@free.fr>
Cc: Matthieu CASTET <castet.matthieu@free.fr>
Cc: Stanislaw Gruszka <stf_xl@wp.pl>
Cc: Pete Zaitcev <zaitcev@redhat.com>
Cc: Oliver Neukum <oliver@neukum.name>
Cc: Juergen Stuber <starblue@users.sourceforge.net>
Cc: Cesar Miquel <miquel@df.uba.ar>
Cc: Matthew Dharm <mdharm-usb@one-eyed-alien.net>
Cc: Matthew Wilcox <willy@linux.intel.com>
Cc: Sarah Sharp <sarah.a.sharp@linux.intel.com>
Cc: Kuninori Morimoto <kuninori.morimoto.gx@renesas.com>
Cc: Felipe Balbi <balbi@ti.com>
Cc: Lucas De Marchi <lucas.demarchi@profusion.mobi>
Cc: Michael Hund <mhund@ld-didactic.de>
Cc: Zack Parsons <k3bacon@gmail.com>
Cc: Melchior FRANZ <mfranz@aon.at>
Cc: Tomoki Sekiyama <tomoki.sekiyama@gmail.com>
Cc: Dan Carpenter <error27@gmail.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
The situation up to this point meant that module.h was pretty
much everywhere, regardless of whether you asked for it or not.
We are fixing that, so give the USB folks who want it an actual
include of it.
Signed-off-by: Paul Gortmaker <paul.gortmaker@windriver.com>
* usb-next: (132 commits)
USB: uas: Use GFP_NOIO instead of GFP_KERNEL in I/O submission path
USB: uas: Ensure we only bind to a UAS interface
USB: uas: Rename sense pipe and sense urb to status pipe and status urb
USB: uas: Use kzalloc instead of kmalloc
USB: uas: Fix up the Sense IU
usb: musb: core: kill unneeded #include's
DA8xx: assign name to MUSB IRQ resource
usb: gadget: g_ncm added
usb: gadget: f_ncm.c added
usb: gadget: u_ether: prepare for NCM
usb: pch_udc: Fix setup transfers with data out
usb: pch_udc: Fix compile error, warnings and checkpatch warnings
usb: add ab8500 usb transceiver driver
USB: gadget: Implement runtime PM for MSM bus glue driver
USB: gadget: Implement runtime PM for ci13xxx gadget
USB: gadget: Add USB controller driver for MSM SoC
USB: gadget: Introduce ci13xxx_udc_driver struct
USB: gadget: Initialize ci13xxx gadget device's coherent DMA mask
USB: gadget: Fix "scheduling while atomic" bugs in ci13xxx_udc
USB: gadget: Separate out PCI bus code from ci13xxx_udc
...
If swap is on a UAS device, we could recurse into the driver by using
GFP_KERNEL. Using GFP_NOIO ensures we won't.
Reported-by: James Bottomley <James.Bottomley@suse.de>
Signed-off-by: Matthew Wilcox <willy@linux.intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
While all existing UAS devices use alternate interface 1, this is not
guaranteed, and it has caused confusion with people trying to bind the
uas driver to non-uas devices.
Signed-off-by: Matthew Wilcox <willy@linux.intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
The spec calls this the status pipe. While it is used to receive sense IUs,
it is also used to receive other IUs, so this can be confusing.
Reported-by: Luben Tuikov <ltuikov@yahoo.com>
Signed-off-by: Matthew Wilcox <willy@linux.intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
The IUs are not being fully initialised by the driver (due to the reserved
space). Since we should be zeroing reserved fields, use kzalloc to do
it for us.
Reported-by: Luben Tuikov <ltuikov@yahoo.com>
Signed-off-by: Matthew Wilcox <willy@linux.intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
Add a comment to the Sense IU data structure that it's also used for Read
Ready and Write Ready. Remove the 'service response' element since it's
gone from the current draft (04).
Signed-off-by: Matthew Wilcox <willy@linux.intel.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
Move the mid-layer's ->queuecommand() invocation from being locked
with the host lock to being unlocked to facilitate speeding up the
critical path for drivers who don't need this lock taken anyway.
The patch below presents a simple SCSI host lock push-down as an
equivalent transformation. No locking or other behavior should change
with this patch. All existing bugs and locking orders are preserved.
Additionally, add one parameter to queuecommand,
struct Scsi_Host *
and remove one parameter from queuecommand,
void (*done)(struct scsi_cmnd *)
Scsi_Host* is a convenient pointer that most host drivers need anyway,
and 'done' is redundant to struct scsi_cmnd->scsi_done.
Minimal code disturbance was attempted with this change. Most drivers
needed only two one-line modifications for their host lock push-down.
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>
Acked-by: James Bottomley <James.Bottomley@suse.de>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
REQ_HARDBARRIER is dead now, so remove the leftovers. What's left
at this point is:
- various checks inside the block layer.
- sanity checks in bio based drivers.
- now unused bio_empty_barrier helper.
- Xen blockfront use of BLKIF_OP_WRITE_BARRIER - it's dead for a while,
but Xen really needs to sort out it's barrier situaton.
- setting of ordered tags in uas - dead code copied from old scsi
drivers.
- scsi different retry for barriers - it's dead and should have been
removed when flushes were converted to FS requests.
- blktrace handling of barriers - removed. Someone who knows blktrace
better should add support for REQ_FLUSH and REQ_FUA, though.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <jaxboe@fusionio.com>
USB Attached SCSI is a new protocol specified jointly by the SCSI T10
committee and the USB Implementors Forum.
Signed-off-by: Matthew Wilcox <willy@linux.intel.com>
Cc: Matthew Dharm <mdharm-usb@one-eyed-alien.net>
[mina86@mina86.com: updated to use new USB_ prefix]
Signed-off-by: Michal Nazarewicz <mina86@mina86.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>