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>
musb-core was setting the ops to the default indexed or flat handlers after
checking for platform overrides. Reverse the order of this so that platform
overrides actually work.
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Felipe Balbi <balbi@ti.com>
The generic fifo functions already use non wrapped accesses in various
cases through the iowrite#_rep functions, and all platforms which override
the default musb_read[b|w] / _write[b|w] functions also provide their own
fifo access functions, so we can safely drop the unnecessary indirection
from the fifo access functions.
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Felipe Balbi <balbi@ti.com>
The Allwinner (sunxi) implementation of the musb has its busctl registers
indexed by the MUSB_INDEX register rather then in a flat address space.
This commit turns MUSB_BUSCTL_OFFSET from a macro into an io-op which can
be overridden from the platform ops.
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Felipe Balbi <balbi@ti.com>
For some reason the musb_write_rxfun* and musb_write_rxhub* functions had
a different function prototype and some extra magic needed on the caller side
compared to their tx counterparts, this commit makes them work the same as
their tx counterparts.
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Felipe Balbi <balbi@ti.com>
Currently we always assign one of the two common implementations of
ep_offset and ep_select operations, overwriting any platform-specific
implementations.
Fixes: d026e9c76a ("usb: musb: Change end point selection to use ...")
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
Signed-off-by: Felipe Balbi <balbi@ti.com>
For musb_core.c we can now just drop the DMA related
ifdef and use the already existing runtime test for
!is_cppi_enabled(musb) instead.
Signed-off-by: Tony Lindgren <tony@atomide.com>
Signed-off-by: Felipe Balbi <balbi@ti.com>
Set up function pointers for DMA so get closer to
being able to build in all the DMA engines.
Signed-off-by: Tony Lindgren <tony@atomide.com>
Signed-off-by: Felipe Balbi <balbi@ti.com>
Pass struct musb to tusb_dma_omap() and is_cppi_enabled(),
and add macros for the other DMA controllers. Populate the
platform specific quirks with the DMA type and use it during
runtime.
Note that platform glue layers with no custom DMA code are
tagged with MUSB_DMA_INVENTRA which may have a chance of
working. Looks like the defconfigs for these use PIO_ONLY,
so this should not break existing configs.
Signed-off-by: Tony Lindgren <tony@atomide.com>
Signed-off-by: Felipe Balbi <balbi@ti.com>
Make sure we're using the new macro, so our
resume signaling will always pass certification.
Based on original work by Bin Liu <Bin Liu <b-liu@ti.com>>
Cc: Bin Liu <b-liu@ti.com>
Cc: <stable@vger.kernel.org> # v3.10+
Signed-off-by: Felipe Balbi <balbi@ti.com>
we can also have babble conditions with LS/FS
and we also want to recover in that case.
Because of that we will drop the check of HSMODE
and always try to run babble recovery.
Suggested-by: Bin Liu <b-liu@ti.com>
Signed-off-by: Felipe Balbi <balbi@ti.com>
There's no point is splitting those anymore.
We're now also able to drop another forward
declaration.
Tested-by: Bin Liu <b-liu@ti.com>
Signed-off-by: Felipe Balbi <balbi@ti.com>
that's not needed anymore. Everything that we
call is irq-safe, so we might as well not
have a delayed work for babble recovery.
Tested-by: Bin Liu <b-liu@ti.com>
Signed-off-by: Felipe Balbi <balbi@ti.com>
recover is a much better name than reset, considering
we don't really reset the IP, just run platform-specific
babble recovery algorithm.
while at that, also fix a typo in comment and add kdoc
for recover memeber of platform_ops.
Tested-by: Bin Liu <b-liu@ti.com>
Signed-off-by: Felipe Balbi <balbi@ti.com>
we're not resetting musb at all, just restarting
the session. This means we don't need to touch PHYs
or VBUS or anything like that. Just make sure session
bit is reenabled after MUSB dropped it.
while at that, make sure to tell usbcore that we're
dropping the session and, thus, disconnecting the
device.
Tested-by: Bin Liu <b-liu@ti.com>
Signed-off-by: Felipe Balbi <balbi@ti.com>
When babble IRQ happens, we need to wait only
5.3us (320 cycles of 60MHz clock), we will give
it some slack and schedule our work a 10 usecs into
the future.
Tested-by: Bin Liu <b-liu@ti.com>
Signed-off-by: Felipe Balbi <balbi@ti.com>
We do *not* want to touch devctl at all when
trying to recover from babble. All we want to
do is mask IRQs until we're done without our
babble recovery, at which point we will unmask
IRQs.
Tested-by: Bin Liu <b-liu@ti.com>
Signed-off-by: Felipe Balbi <balbi@ti.com>
sometimes we want to just mask/unmask interrupts
without touching devctl register. For those
cases, let's introduce musb_enable_interrupts and
musb_disable_interrupts()
Tested-by: Bin Liu <b-liu@ti.com>
Signed-off-by: Felipe Balbi <balbi@ti.com>
Whenever babble happens, MUSB controller will
drop session automatically.
The only case where it won't drop the session,
is when we're running on AM335x and SW_SESSION_CTRL
bit has been set. In that case, controller will
not touch session bit so SW has a chance to recover
from babble condition.
Tested-by: Bin Liu <b-liu@ti.com>
Signed-off-by: Felipe Balbi <balbi@ti.com>
musb->int_usb already contains the correct
information for musb-core to handle babble.
In fact, this very check was just causing a
nonsensical babble interrupt storm.
With this I can get test.sh to run and, even though
all tests fail with timeout, that's still better
than locking up the system due to IRQ storm.
Also, if I remove g_zero and load g_mass_storage,
then everything works fine again.
Tested-by: Bin Liu <b-liu@ti.com>
Signed-off-by: Felipe Balbi <balbi@ti.com>
when musb is operating as host and a remote wakeup
fires up, a resume interrupt will be raised. At that
point SUSPENDM bit is automatically cleared and
RESUME bit is automatically set.
Remove those two from IRQ handler.
Tested-by: Bin Liu <b-liu@ti.com>
Signed-off-by: Felipe Balbi <balbi@ti.com>
There was already a proper place where we were
checking for babble interrupts, move babble
recovery there.
Tested-by: Bin Liu <b-liu@ti.com>
Signed-off-by: Felipe Balbi <balbi@ti.com>
FSDEV is set for both HIGH and FULL speeds,
the correct HIGHSPEED check is done through
power register's HSMODE bit.
Tested-by: Bin Liu <b-liu@ti.com>
Signed-off-by: Felipe Balbi <balbi@ti.com>
devctl & MUSB_DEVCTL_HM represents a single bit,
just check for the bit, there's really no need
to compare the result against 0.
Tested-by: Bin Liu <b-liu@ti.com>
Signed-off-by: Felipe Balbi <balbi@ti.com>
instead of using manually spelled out bit-shits
and iterate over each of the 16-bits (one for
each endpoint) on each direction, we can make use
of for_each_set_bit() which internally uses
find_first_bit().
This makes the code slightly more readable while
also making we only iterate over bits which are
actually set.
Signed-off-by: Felipe Balbi <balbi@ti.com>
As per Mentor Graphics' documentation, we should
always handle TX endpoints before RX endpoints.
This patch fixes that error while also updating
some hard-to-read comments which were scattered
around musb_interrupt().
This patch should be backported as far back as
possible since this error has been in the driver
since it's conception.
Cc: <stable@vger.kernel.org>
Signed-off-by: Felipe Balbi <balbi@ti.com>
We need a pm_runtime_get_sync() call from
within musb_gadget_pullup() to make sure
registers are accessible at that time.
The problem is that musb_gadget_pullup() is
called with IRQs disabled and, because of that,
we need to tell pm_runtime that this pm_runtime_get_sync()
is IRQ safe.
We can simply add pm_runtime_irq_safe(), however, because
we need to make our read/write accessor function pointers
have been initialized before trying to use them. This means
that all pm_runtime initialization for musb_core needs to
be moved down so that when we call pm_runtime_irq_safe(),
the pm_runtime_get_sync() that it calls on the parent, won't
cause a crash due to NULL musb_read/write accessors.
Reported-by: Pali Rohár <pali.rohar@gmail.com>
Signed-off-by: Felipe Balbi <balbi@ti.com>
The commit 889ad3b "usb: musb: try a race-free wakeup" breaks device
hotplug enumeraitonn when the device is connected behind a hub while usb
autosuspend is enabled.
Adding finish_resume_work into runtime resume callback fixes the issue.
Also resume root hub is required to resume the bus from runtime suspend,
so move musb_host_resume_root_hub() back to its original location, where
handles RESUME interrupt.
Signed-off-by: Bin Liu <b-liu@ti.com>
Tested-by: Tony Lindgren <tony@atomide.com>
Signed-off-by: Felipe Balbi <balbi@ti.com>
Here's the set of driver core patches for 3.19-rc1.
They are dominated by the removal of the .owner field in platform
drivers. They touch a lot of files, but they are "simple" changes, just
removing a line in a structure.
Other than that, a few minor driver core and debugfs changes. There are
some ath9k patches coming in through this tree that have been acked by
the wireless maintainers as they relied on the debugfs changes.
Everything has been in linux-next for a while.
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v2
iEYEABECAAYFAlSOD20ACgkQMUfUDdst+ylLPACg2QrW1oHhdTMT9WI8jihlHVRM
53kAoLeteByQ3iVwWurwwseRPiWa8+MI
=OVRS
-----END PGP SIGNATURE-----
Merge tag 'driver-core-3.19-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/driver-core
Pull driver core update from Greg KH:
"Here's the set of driver core patches for 3.19-rc1.
They are dominated by the removal of the .owner field in platform
drivers. They touch a lot of files, but they are "simple" changes,
just removing a line in a structure.
Other than that, a few minor driver core and debugfs changes. There
are some ath9k patches coming in through this tree that have been
acked by the wireless maintainers as they relied on the debugfs
changes.
Everything has been in linux-next for a while"
* tag 'driver-core-3.19-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/driver-core: (324 commits)
Revert "ath: ath9k: use debugfs_create_devm_seqfile() helper for seq_file entries"
fs: debugfs: add forward declaration for struct device type
firmware class: Deletion of an unnecessary check before the function call "vunmap"
firmware loader: fix hung task warning dump
devcoredump: provide a one-way disable function
device: Add dev_<level>_once variants
ath: ath9k: use debugfs_create_devm_seqfile() helper for seq_file entries
ath: use seq_file api for ath9k debugfs files
debugfs: add helper function to create device related seq_file
drivers/base: cacheinfo: remove noisy error boot message
Revert "core: platform: add warning if driver has no owner"
drivers: base: support cpu cache information interface to userspace via sysfs
drivers: base: add cpu_device_create to support per-cpu devices
topology: replace custom attribute macros with standard DEVICE_ATTR*
cpumask: factor out show_cpumap into separate helper function
driver core: Fix unbalanced device reference in drivers_probe
driver core: fix race with userland in device_add()
sysfs/kernfs: make read requests on pre-alloc files use the buffer.
sysfs/kernfs: allow attributes to request write buffer be pre-allocated.
fs: sysfs: return EGBIG on write if offset is larger than file size
...
This removes the ifdef clutter a bit and saves few lines.
It also makes it easier to detect the remaining places
where we have conditional building of code done based
on if defined for things like DMA.
Signed-off-by: Tony Lindgren <tony@atomide.com>
Signed-off-by: Felipe Balbi <balbi@ti.com>
This allows the endpoints to work when multiple MUSB glue
layers are built in.
Cc: Fabio Baltieri <fabio.baltieri@linaro.org>
Cc: Lee Jones <lee.jones@linaro.org>
Cc: Lars-Peter Clausen <lars@metafoo.de>
Acked-by: Linus Walleij <linus.walleij@linaro.org>
Acked-by: Apelete Seketeli <apelete@seketeli.net>
Signed-off-by: Tony Lindgren <tony@atomide.com>
Signed-off-by: Felipe Balbi <balbi@ti.com>
Change to use new IO access. This allows us to build in multiple
MUSB glue layers.
[ balbi@ti.com : switch to EXPORT_SYMBOL_GPL()
fix long lines ]
Cc: Fabio Baltieri <fabio.baltieri@linaro.org>
Cc: Lee Jones <lee.jones@linaro.org>
Cc: Linus Walleij <linus.walleij@linaro.org>
Signed-off-by: Tony Lindgren <tony@atomide.com>
Signed-off-by: Felipe Balbi <balbi@ti.com>
Disable the MUSB interrupts till MUSB is recovered fully from BABBLE
condition. There are chances that we could get multiple interrupts
till the time the babble recover work gets scheduled. Sometimes
this could even end up in an endless loop making MUSB itself unusable.
Reported-by: Felipe Balbi <balbi@ti.com>
Signed-off-by: George Cherian <george.cherian@ti.com>
Signed-off-by: Felipe Balbi <balbi@ti.com>
On am335x-evm with musb in host mode and using it as a wakeup source the
following happens once the CPU comes out of suspend to ram:
|PM: Wakeup source MPU_WAKE
|PM: noirq resume of devices complete after 15.453 msecs
|PM: early resume of devices complete after 2.222 msecs
|PM: resume of devices complete after 507.351 msecs
|Restarting tasks ...
|------------[ cut here ]------------
|WARNING: CPU: 0 PID: 322 at drivers/usb/core/urb.c:339 usb_submit_urb+0x494/0x4c8()
|URB cc0db380 submitted while active
|[<c0348e64>] (usb_submit_urb) from [<c0340f94>] (hub_activate+0x2b8/0x49c)
|[<c0340f94>] (hub_activate) from [<c03411dc>] (hub_resume+0x14/0x1c)
|[<c03411dc>] (hub_resume) from [<c034be10>] (usb_resume_interface.isra.4+0xdc/0x110)
|[<c034be10>] (usb_resume_interface.isra.4) from [<c034beb0>] (usb_resume_both+0x6c/0x13c)
|[<c034beb0>] (usb_resume_both) from [<c034cca4>] (usb_runtime_resume+0x10/0x14)
|[<c034cca4>] (usb_runtime_resume) from [<c02bbd80>] (__rpm_callback+0x2c/0x60)
|[<c02bbd80>] (__rpm_callback) from [<c02bbdd4>] (rpm_callback+0x20/0x74)
|[<c02bbdd4>] (rpm_callback) from [<c02bcc48>] (rpm_resume+0x380/0x548)
|[<c02bcc48>] (rpm_resume) from [<c02bcb00>] (rpm_resume+0x238/0x548)
|[<c02bcb00>] (rpm_resume) from [<c02bd08c>] (__pm_runtime_resume+0x64/0x94)
|[<c02bd08c>] (__pm_runtime_resume) from [<c034b5a4>] (usb_autopm_get_interface+0x18/0x5c)
|[<c034b5a4>] (usb_autopm_get_interface) from [<c03438b8>] (hub_thread+0x10c/0x115c)
|[<c03438b8>] (hub_thread) from [<c005a70c>] (kthread+0xbc/0xd8)
|---[ end trace 036aa5fe78203142 ]---
|hub 1-0:1.0: activate --> -16
|hub 2-0:1.0: activate --> -16
The reason for this backtrace is the attempt of the USB code to resume
the HUB twice and thus enqueue the status URB twice.
Alan Stern was a great help by explaining how the USB code supposed to
work and what is most likely the problem. The root problem is that after
resume the musb runtime-suspend state remains RPM_SUSPENDED.
According to git log it RPM was added for the omap2430 platform. If I
understand it correct the omap2430 invokes a get on musb once a cable is
connected and a put once the cable is gone. In between the device could
go auto-idle/off. Not sure what happens when the device goes into suspend
but then I guess it was gadget only.
On DSPS I see only a get in probe and put in remove function. This would
forbid RPM from working but then the devices enterns suspended state
anyway :)
To get rid of this warning, I set the device state to RPM_ACTIVE which
the expected state.
Cc: Alan Stern <stern@rowland.harvard.edu>
Cc: <stable@vger.kernel.org>
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Felipe Balbi <balbi@ti.com>
Attaching a keyboard, using it as a wakeup via
|for f in $(find /sys/devices/ocp.3/47400000.usb -name wakeup)
|do
| echo enabled > $f
|done
going into standby
| echo standby > /sys/power/state
and now a wake up by a pressing a key.
What happens is that the system wakes up but the USB device is dead. The
USB stack tries to send a few control URBs but nothing comes back.
Eventually it gaves up and the device remains dead:
|[ 632.559678] PM: Wakeup source USB1_PHY
|[ 632.581074] PM: noirq resume of devices complete after 21.261 msecs
|[ 632.607521] PM: early resume of devices complete after 10.360 msecs
|[ 632.616854] net eth2: initializing cpsw version 1.12 (0)
|[ 632.704126] net eth2: phy found : id is : 0x4dd074
|[ 636.704048] libphy: 4a101000.mdio:00 - Link is Up - 1000/Full
|[ 638.444620] usb 1-1: reset low-speed USB device number 2 using musb-hdrc
|[ 653.713435] usb 1-1: device descriptor read/64, error -110
|[ 669.093435] usb 1-1: device descriptor read/64, error -110
|[ 669.473424] usb 1-1: reset low-speed USB device number 2 using musb-hdrc
|[ 684.743436] usb 1-1: device descriptor read/64, error -110
|[ 690.065097] PM: resume of devices complete after 57450.744 msecs
|[ 690.076601] PM: Finishing wakeup.
|[ 690.076627] Restarting tasks ...
It seems that since we got woken up via MUSB_INTR_RESUME the
musb_host_finish_resume() callback is executed before the
resume-callbacks of the PHY and glue layer are invoked. If I delay it
until the glue layer resumed then I don't see this problem.
I also move musb_host_resume_root_hub() into that callback since I don't
see any reason in doing anything resume-link if there are still pieces
not restored.
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Felipe Balbi <balbi@ti.com>
The am335x-evmsk support two kinds of suspend:
- standby
the USB device remains powered while the system goes into suspend
- mem
the USB device becomes powerless while the system goes into suspend.
In the "standby" case the device resumes quickly. In the "mem" case the
system hangs for a few seconds. It seems to me that the USB-device has
no address (it was disconnected) and the USB stack thinks that it is
fully operational and GetPortStatus returns the status from before the
suspend so it is not a big help here.
This adds a check in the resume path to see if the device mode (A or B)
and the speed is the same. If the device went missing between
suspend/resume (VBUS went down) then MUSB seems to go into B mode and
HS/FS bits are cleared. In that case we clear the port1_status bits and
assume a disconnect. Once the stack learns this it does a "logical
disconnect" and removes the USB-device quickly. Should the device remain
connected during the suspend then MUSB will receives a "CONNECT" interrupt.
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Felipe Balbi <balbi@ti.com>
Before using the PHY framework instead of the USB PHY one, we need to
move the OTG state into another place, since it won't be available when
USB PHY isn't used. This patch moves the OTG state into the OTG
structure, and makes all the needed modifications in the drivers
using the OTG state.
[ balbi@ti.com : fix build regressions with phy-tahvo.c, musb_dsps.c,
phy-isp1301-omap, and chipidea's debug.c ]
Acked-by: Kishon Vijay Abraham I <kishon@ti.com>
Acked-by: Peter Chen <peter.chen@freescale.com>
Signed-off-by: Antoine Tenart <antoine.tenart@free-electrons.com>
Signed-off-by: Felipe Balbi <balbi@ti.com>
In commit 0ec8fd70fb ("USB: musb: fix possible panic while resuming")
musb_resume() became musb_resume_early() in order to enable the clocks
early on resume. This piece of the resume code was removed later in
commit 034917612 ("usb: musb: move clock handling to glue layer").
In between the function was renamed from musb_resume_early() to
musb_resume_noirq() by commit commit 48fea9659e ("USB: Rework musb
suspend()/resume_early()").
Now I see that first musb_core is resumed followed by phy and glue
layer and I ask myself is this really what we intend to do? This kind
of revoked the purpose of the first commit (0ec8fd70fb).
Because of this and because it looks wrong to resume (core) before the
glue layer I push it to the ->resume callback.
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Felipe Balbi <balbi@ti.com>
Sanity check on resource happening with devm_ioremap_resource().
Signed-off-by: Varka Bhadram <varkab@cdac.in>
Signed-off-by: Felipe Balbi <balbi@ti.com>
Currently musb_platform_reset() is only used by dsps.
In case of BABBLE interrupt for other platforms the musb_platform_reset()
is a NOP. In such situations no need to re-initialize the endpoints.
Also in the latest silicon revision of AM335x, we do have a babble recovery
mechanism without resetting the IP block. In preperation to add that support
its better to have a rest_done return for musb_platform_reset().
Signed-off-by: George Cherian <george.cherian@ti.com>
Tested-by: Bin Liu <b-liu@ti.com>
Signed-off-by: Felipe Balbi <balbi@ti.com>
During babble condition both first disconnect of devices are
initiated. Make sure MUSB controller is reset and re-initialized
after all disconnects.
To acheive this schedule a delayed work for babble recovery.
While at that convert udelay to usleep_range.
Refer Documentation/timers/timers-howto.txt
Signed-off-by: George Cherian <george.cherian@ti.com>
Tested-by: Bin Liu <b-liu@ti.com>
Signed-off-by: Felipe Balbi <balbi@ti.com>
BABBLE and RESET share the same interrupt. The interrupt
is considered to be RESET if MUSB is in peripheral mode and
as a BABBLE if MUSB is in HOST mode.
Handle babble condition iff MUSB is in HOST mode.
Signed-off-by: George Cherian <george.cherian@ti.com>
Tested-by: Bin Liu <b-liu@ti.com>
Signed-off-by: Felipe Balbi <balbi@ti.com>
On AM33xx platforms, unplugging a device in the middle of an active
transfer leads to a drop of MUSB_DEVCTL_HM in MUSB_DEVCTL before the
system is informed about a disconnect. This consequently makes the musb
core call the gadget code to handle the interrupt request, which then
crashes the kernel because the relevant pointers haven't been set up
for gadget mode.
To fix this, use is_host_active() rather than (devctl & MUSB_DEVCTL_HM)
in musb_interrupt() and musb_dma_completion() to detect whether the
controller is in host or peripheral mode. This information is provided
by the driver logic and does not rely on register contents.
Signed-off-by: Daniel Mack <zonque@gmail.com>
Signed-off-by: Felipe Balbi <balbi@ti.com>
Handle BABBLE interrupt error conditions from a work struct handler.
This indirection is necessary as we can't be certain that the phy
functions don't sleep.
Platform layer implementation may pass a babble error down to the core
in order to handle it.
Signed-off-by: Daniel Mack <zonque@gmail.com>
Signed-off-by: Felipe Balbi <balbi@ti.com>
