This refactor results in a cleaner state machine code and promotes
consistency, readability, and maintanability of this driver.

This refactor state machine was well tested and it is currently running in
production code and devices.
Signed-off-by: Felipe F. Tonello <eu@felipetonello.com>
Signed-off-by: Felipe Balbi <felipe.balbi@linux.intel.com>

512 is the value used by wMaxPacketSize, as specified by the USB Spec. This
makes sure this driver uses, by default, the most optimal value for IN and OUT
endpoint requests.
Signed-off-by: Felipe F. Tonello <eu@felipetonello.com>
Signed-off-by: Felipe Balbi <felipe.balbi@linux.intel.com>

The new version of alloc_ep_req() already aligns the buffer size to
wMaxPacketSize on OUT endpoints.
Signed-off-by: Felipe F. Tonello <eu@felipetonello.com>
Signed-off-by: Felipe Balbi <felipe.balbi@linux.intel.com>

Using usb_ep_align() makes sure that the buffer size for OUT endpoints is
always aligned with wMaxPacketSize (512 usually). This makes sure
that no buffer has the wrong size, which can cause nasty bugs.
Signed-off-by: Felipe F. Tonello <eu@felipetonello.com>
Signed-off-by: Felipe Balbi <felipe.balbi@linux.intel.com>

Length of buffers should be of type size_t whenever possible. Altough
recommended, this change has no real practical change, unless a driver has a
uses a huge or negative buffer size - it might help find these bugs.
Signed-off-by: Felipe F. Tonello <eu@felipetonello.com>
Signed-off-by: Felipe Balbi <felipe.balbi@linux.intel.com>

USB spec specifies wMaxPacketSize to be little endian (as other properties),
so when using this variable in the driver we should convert to the current
CPU endianness if necessary.

This patch also introduces usb_ep_align() which does always returns the
aligned buffer size for an endpoint. This is useful to be used by USB requests
allocator functions.
Signed-off-by: Felipe F. Tonello <eu@felipetonello.com>
Signed-off-by: Felipe Balbi <felipe.balbi@linux.intel.com>

Introduce an attribute "inquiry_string" to the lun.

In some environments, e. g. BIOS boot menus, the inquiry string
is the only information about devices presented to the user. The
default string depends on the "cdrom" bit of the first lun as
well as the kernel version and allows no further customization.
So without access to the client it is not obvious which gadget is
active at a given point and what any of the available luns might
contain.

If "inquiry_string" is ignored or set to the empty string, the
old behavior is preserved.
Signed-off-by: Philipp Gesang <philipp.gesang@intra2net.com>
Signed-off-by: Felipe Balbi <felipe.balbi@linux.intel.com>

Rockchip platform merely enable usb3 clocks and
populate its children. So we can use this generic
glue layer to support Rockchip dwc3.
Signed-off-by: William Wu <william.wu@rock-chips.com>
Signed-off-by: Felipe Balbi <felipe.balbi@linux.intel.com>

This patch adds the devicetree documentation required for Rockchip
USB3.0 core wrapper consisting of USB3.0 IP from Synopsys.

It supports DRD mode, and could operate in device mode (SS, HS, FS)
and host mode (SS, HS, FS, LS).
Signed-off-by: William Wu <william.wu@rock-chips.com>
Acked-by: Rob Herring <robh@kernel.org>
Signed-off-by: Felipe Balbi <felipe.balbi@linux.intel.com>

Add a quirk to clear the GUSB3PIPECTL.DELAYP1TRANS bit,
which specifies whether disable delay PHY power change
from P0 to P1/P2/P3 when link state changing from U0
to U1/U2/U3 respectively.
Signed-off-by: William Wu <william.wu@rock-chips.com>
Acked-by: Rob Herring <robh@kernel.org>
Signed-off-by: Felipe Balbi <felipe.balbi@linux.intel.com>

Support to configure the UTMI+ PHY with an 8- or 16-bit
interface via DT. The UTMI+ PHY interface is a hardware
capability, and it's platform dependent. Normally, the
PHYIF can be configured during coreconsultant.

But for some specific USB cores(e.g. rk3399 SoC DWC3),
the default PHYIF configuration value is false, so we
need to reconfigure it by software.
Signed-off-by: William Wu <william.wu@rock-chips.com>
Acked-by: Rob Herring <robh@kernel.org>
Signed-off-by: Felipe Balbi <felipe.balbi@linux.intel.com>

Add a quirk to clear the GUSB2PHYCFG.U2_FREECLK_EXISTS bit,
which specifies whether the USB2.0 PHY provides a free-running
PHY clock, which is active when the clock control input is active.
Signed-off-by: William Wu <william.wu@rock-chips.com>
Acked-by: Rob Herring <robh@kernel.org>
Signed-off-by: Felipe Balbi <felipe.balbi@linux.intel.com>

With composite gadget (ACM + NCM), USB3380 to host TCP transfer
speed dropped to 150 Mbit/s compared to 900 Mbit/s with NCM
gadget. Problem seems to be that net2280/USB3380 has only four
DMA channels and those DMA channels are allocated to first HW
endpoints. Endpoint match function was mapping endpoint names
directly, so NCM did not get DMA for bulk endpoints.

This patch changed match_ep to prefer DMA enabled hw endpoints
for bulk usb endpoints and PIO for interrupt usb endpoints.
Signed-off-by: Jussi Kivilinna <jussi.kivilinna@haltian.com>
Signed-off-by: Felipe Balbi <felipe.balbi@linux.intel.com>

With SuperSpeed CDC NCM gadget, net2280 would get stuck in
'handle_ep_small' function. Triggering issue requires large
TCP transfer from host to USB3380.

Patch adds check for stuck condition and prevents hard lockup.
Signed-off-by: Jussi Kivilinna <jussi.kivilinna@haltian.com>
Signed-off-by: Felipe Balbi <felipe.balbi@linux.intel.com>

Patch enables SuperSpeed for NCM gadget.

Tested with USB3380 and measured TCP throughput with two Intel PCs:
  udc to host: 920 Mbit/s
  host to udc: 550 Mbit/s
Signed-off-by: Jussi Kivilinna <jussi.kivilinna@haltian.com>
Signed-off-by: Felipe Balbi <felipe.balbi@linux.intel.com>

alloc_ordered_workqueue replaces the deprecated
create_singlethread_workqueue.

There are multiple work items on the work queue, which require
ordering. Hence, an ordered workqueue has been used.

The workqueue "wq_otg" is not being used on a memory reclaim path.
Hence, WQ_MEM_RECLAIM has not been set.
Acked-by: John Youn <johnyoun@synopsys.com>
Signed-off-by: Bhaktipriya Shridhar <bhaktipriya96@gmail.com>
Signed-off-by: Felipe Balbi <felipe.balbi@linux.intel.com>

The assignment ret = ret is redundant and can be removed.
Reviewed-by: Peter Chen <peter.chen@nxp.com>
Signed-off-by: Colin Ian King <colin.king@canonical.com>
Signed-off-by: Felipe Balbi <felipe.balbi@linux.intel.com>

Disabling USB gadget functions configured through configfs is something
that can happen in normal use cases. Keep the existing log for this type
of event, but only as information, not as an error.
Signed-off-by: Romain Izard <romain.izard.pro@gmail.com>
Signed-off-by: Felipe Balbi <felipe.balbi@linux.intel.com>

As the last known user, ie. pxa27x_udc relying on calls to
usb_gadget_xxx() was amended to use the phy notifier, remove a bit the
USB stack adherence.

Actually the driver still uses the gadget API for structures definition,
but the implementation of USB gadget specific function usb_gadget_*() is
not necessary anymore.
Signed-off-by: Robert Jarzmik <robert.jarzmik@free.fr>
Signed-off-by: Felipe Balbi <felipe.balbi@linux.intel.com>

In the gpio based case, the status of the phy is known at start by
reading the VBus gpio.

Actually, this is a fix, as this initial state, when not set up,
prevents a gadget to answer to the enumeration phase, as there is no
notification in this case (the VBus is already high when kernel boots)
so no interrupt is triggered, and the flow is :
 - gadget initializes
 - gadget gets its phy-generic with a xxx_get_phy_xxx() call type
 - gadget does a "set_peripheral()" call type
   => here if the otg->state is correctly filled, the proper vbus
   handling will be called, and the gadget will be aware it should
   answer enumeration and go forth

Without this fix, the USB cable must be removed and replugged for any
gadget relying on phy-generic and its gpio vbus handling to work.

The problem was seen on a pxa27x architecture based board on a
devicetree build.
Signed-off-by: Robert Jarzmik <robert.jarzmik@free.fr>
Signed-off-by: Felipe Balbi <felipe.balbi@linux.intel.com>

In the legacy behavior, and USB phy, upon detection a VBus signal, was
calling usb_gadget_vbus_(dis)connect().

This model doesn't work if the phy is generic and doesn't have an
adherence to the gadget API.

Instead of relying on the phy to call the gadget API, hook up the phy
notifier to report the VBus event, and upon it call the usb gadget API
ourselves.

This brings a new ordering problem, as before even if the usb_get_phy()
was failing because the UDC was probed before the phy, the phy would
call the gadget anyway, making the VBus connection event forwarded to
the gadget. Now we rely on the notifier, we have to ensure the
xxx_get_phy() does indeed work.

In order to cope with this, it is assumed that :
 - for legacy platform_data machine, as the ordering cannot be ensured,
   the phy must call usb_gadget_vbus_(dis)connect, such as
   phy-gpio-vbus-usb.c
 - for new devicetree platforms, we'll rely on the probe deferral, and
   the phy can be gadget API agnostic.
Signed-off-by: Robert Jarzmik <robert.jarzmik@free.fr>
Signed-off-by: Felipe Balbi <felipe.balbi@linux.intel.com>

No functional changes, just a slight cosmetic
change.
Signed-off-by: Felipe Balbi <felipe.balbi@linux.intel.com>

We don't use LST bit anymore, so this condition will
never trigger.
Signed-off-by: Felipe Balbi <felipe.balbi@linux.intel.com>

Upon transfer completion after a full ring, let's
add more TRBs to our ring in order to complete our
request successfully.
Signed-off-by: Felipe Balbi <felipe.balbi@linux.intel.com>

If the ring is full and we are processing a big
sglist, then let's interrupt so we can, later, add
more TRBs to the ring.
Signed-off-by: Felipe Balbi <felipe.balbi@linux.intel.com>

These two fields will be used in a follow-up patch
to track how many entries of request's sglist we
have already processed. The reason is that if a
gadget driver sends an sglist with more entries then
we can fit in the ring, we will have to continue
processing remaining afterwards.
Signed-off-by: Felipe Balbi <felipe.balbi@linux.intel.com>

We know that we have to iterate over the list of
started requests. Instead of looping forever, we can
rely on list_for_each_entry(). Likewise, instead of
a do {} while loop over all, maybe available,
scatterlist entries, we can detect if $this request
uses scatterlist and rely on for_each_sg().

This makes the code easier to follow while making
sure that we will *always* break out of the loop.
Signed-off-by: Felipe Balbi <felipe.balbi@linux.intel.com>

Many of the comments in that function are really
outdated and don't match what the driver is
doing. Moreover, recent patches combined programming
model for all non-control endpoints, this gives us
an opportunity to get rid of our special cases in
__dwc3_gadget_ep_queue().
Signed-off-by: Felipe Balbi <felipe.balbi@linux.intel.com>

We always need to decrement our index by at least
one. Simplify the implementation by using a
temporary local variable and making sure that we
will always decrement one extra if tmp == 0.
Signed-off-by: Felipe Balbi <felipe.balbi@linux.intel.com>

Instead of waiting until giveback before
incrementing the dequeue pointer, we can increment
it from dwc3_cleanup_done_reqs(), that way we avoid
an extra loop over all TRBs during giveback.

While at that, also avoid using req->first_trb_index
as that's completely unnecessary. A follow-up patch
will clean up further uses of that and remove the
field altogether.
Signed-off-by: Felipe Balbi <felipe.balbi@linux.intel.com>

The only endpoint which actually requires LST bit
and XferComplete is ep0/1. Let's save some time by
completely removing LST bit support and
XferComplete.

This simplifies and consolidates endpoint handling
for all other 3 transfer types while also avoiding
extra interrupts.
Signed-off-by: Felipe Balbi <felipe.balbi@linux.intel.com>

Pull two parisc fixes from Helge Deller:
 "The first patch ensures that the high-res cr16 clocksource (which was
  added in kernel 4.7) gets choosen as default clocksource for parisc.

  The second patch moves the #define of EREFUSED down inside errno.h and
  thus unbreaks building the gccgo compiler"

* 'parisc-4.8-2' of git://git.kernel.org/pub/scm/linux/kernel/git/deller/parisc-linux:
  parisc: Fix order of EREFUSED define in errno.h
  parisc: Fix automatic selection of cr16 clocksource

This is an entirely new driver instead of yet another set of patches
to sb_edac.c because:

1) Mapping from PCI devices to socket/memory controller is significantly
   different. Skylake scatters devices on a socket across a number of
   PCI buses.
2) There is an extra level of interleaving via the "mcroute" register
   that would be a little messy to squeeze into the old driver.
3) Validation is getting too expensive. Changes to sb_edac need to
   be checked against Sandy Bridge, Ivy Bridge, Haswell, Broadwell and
   Knights Landing.
Acked-by: Aristeu Rozanski <aris@redhat.com>
Acked-by: Borislav Petkov <bp@suse.de>
Signed-off-by: Tony Luck <tony.luck@intel.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

When building gccgo in userspace, errno.h gets parsed and the go include file
sysinfo.go is generated.

Since EREFUSED is defined to the same value as ECONNREFUSED, and ECONNREFUSED
is defined later on in errno.h, this leads to go complaining that EREFUSED
isn't defined yet.

Fix this trivial problem by moving the define of EREFUSED down after
ECONNREFUSED in errno.h (and clean up the indenting while touching this line).
Signed-off-by: Helge Deller <deller@gmx.de>
Cc: stable@vger.kernel.org

Commit 54b66800 (parisc: Add native high-resolution sched_clock()
implementation) added support to use the CPU-internal cr16 counters as reliable
clocksource with the help of HAVE_UNSTABLE_SCHED_CLOCK.

Sadly the commit missed to remove the hack which prevented cr16 to become the
default clocksource even on SMP systems.
Signed-off-by: Helge Deller <deller@gmx.de>
Cc: stable@vger.kernel.org # 4.7+

The kernel test robot reported a usercopy failure in the new hardened
sanity checks, due to a page-crossing copy of the FPU state into the
task structure.

This happened because the kernel test robot was testing with SLOB, which
doesn't actually do the required book-keeping for slab allocations, and
as a result the hardening code didn't realize that the task struct
allocation was one single allocation - and the sanity checks fail.

Since SLOB doesn't even claim to support hardening (and you really
shouldn't use it), the straightforward solution is to just make the
usercopy hardening code depend on the allocator supporting it.
Reported-by: kernel test robot <xiaolong.ye@intel.com>
Cc: Kees Cook <keescook@chromium.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

Pull i2c fixes from Wolfram Sang:
 "I2C has some pretty standard driver bugfixes and one minor cleanup"

* 'i2c/for-current' of git://git.kernel.org/pub/scm/linux/kernel/git/wsa/linux:
  i2c: meson: Use complete() instead of complete_all()
  i2c: brcmstb: Use complete() instead of complete_all()
  i2c: bcm-kona: Use complete() instead of complete_all()
  i2c: bcm-iproc: Use complete() instead of complete_all()
  i2c: at91: fix support of the "alternative command" feature
  i2c: ocores: add missed clk_disable_unprepare() on failure paths
  i2c: cros-ec-tunnel: Fix usage of cros_ec_cmd_xfer()
  i2c: mux: demux-pinctrl: properly roll back when adding adapter fails

Pull device mapper fixes from Mike Snitzer:

 - a stable fix for DM round robin multipath path selector to disable
   preemption before using this_cpu_ptr()

 - a slight increase in DM crypt's mempool reserves to make swap ontop
   of DM crypt more performant

 - a few DM raid fixes to issues found while testing changes that were
   merged in v4.8-rc1

* tag 'dm-4.8-fixes-2' of git://git.kernel.org/pub/scm/linux/kernel/git/device-mapper/linux-dm:
  dm raid: support raid0 with missing metadata devices
  dm raid: enhance attempt_restore_of_faulty_devices() to support more devices
  dm raid: fix restoring of failed devices regression
  dm raid: fix frozen recovery regression
  dm crypt: increase mempool reserve to better support swapping
  dm round robin: do not use this_cpu_ptr() without having preemption disabled

Pull SCSI fixes from James Bottomley:
 "Six fairly small fixes.  The ipr, mpt3sas and ses ones all trigger
  oopses.  The megaraid one fixes an attach failure on io mapped only
  cards, the fcoe one is an obvious problem in the error path and the
  aacraid one is a theoretical security issue (ability to trick the
  kernel into a buffer overrun)"

* tag 'scsi-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/jejb/scsi:
  ses: Fix racy cleanup of /sys in remove_dev()
  mpt3sas: Fix resume on WarpDrive flash cards
  ipr: Fix sync scsi scan
  megaraid_sas: Fix probing cards without io port
  aacraid: Check size values after double-fetch from user
  fcoe: Use kfree_skb() instead of kfree()