commit 0090114d336a9604aa2d90bc83f20f7cd121b76c upstream.

The CPPI 4.1 driver polls register to workaround the premature TX
interrupt issue, but it causes audio playback underrun when triggered in
Isoch transfers.

Isoch doesn't do back-to-back transfers, the TX should be done by the
time the next transfer is scheduled. So skip this polling workaround for
Isoch transfer.

Fixes: a655f481d83d6 ("usb: musb: musb_cppi41: handle pre-mature TX complete interrupt")
Reported-by: Alexandre Bailon <abailon@baylibre.com>
Acked-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Tested-by: Alexandre Bailon <abailon@baylibre.com>
Signed-off-by: Bin Liu <b-liu@ti.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>

After a few iterations of start/stop UVC camera streaming, the streaming
stops.

This patch adds 250us delay in the cppi channel abort path to let cppi
drain properly.

Using 50us delay seems to be too aggressive, some webcams are still
broken. 250us is the original value used in TI 3.2 kernel.

Signed-off-by: Bin Liu <b-liu@ti.com>
Signed-off-by: Felipe Balbi <balbi@ti.com>

since commit 33c300cb90a6 ("usb: musb: dsps:
don't fake of_node to musb core") we have been
preventing CPPI 4.1 from probing due to NULL
of_node. We can't revert said commit otherwise
a different regression would show up, so the fix
is to look for the parent device's (glue layer's)
of_node instead, since that's the thing which
is actually described in DTS.

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>

connect AR9271(USB wifi) to AM335x, and send a flood ping from Mac OSX,
AR9271 is stopped.

on USB bus, the following occurs.

 - OUT transaction is ACKed (NYET).
 - IN transaction is ACKed (512bytes).
 - PING-NAK transaction is continued for about 2 seconds (AR9271 timeout?).

In current imprementation, IN-transaction is not completed because it
checks the empty of TX-FIFO in cppi41_dma_callback().
As a result, communication to AR9271 stops.

This patch modified to check the empty of TX-FIFO only when OUT-transaction.

Signed-off-by: Takeyoshi Kikuchi <kikuchi@centurysys.co.jp>
Signed-off-by: Felipe Balbi <balbi@ti.com>

according to comment in code, HS completion
will happen pretty fast, instead of using
udelay(), let's just busy loop and drop a
cpu_relax() where udelay() was.

Signed-off-by: Felipe Balbi <balbi@ti.com>

as soon as we find out tx fifo is empty, there's
no need to break out of the loop just to have another
branch to complete the transfer. We can just complete
transfer and exit early.

Signed-off-by: Felipe Balbi <balbi@ti.com>

no functional changes, clean up only.

Signed-off-by: Felipe Balbi <balbi@ti.com>

1. set AUTOREQ to NONE at the beginning of teardown;

2. add delay for dma pipeline to drain;

3. Do not set USB_TDOWN bit for RX teardown.

  The CPPI hw has an issue that when tearing down a RX channel, if
  another RX channel is receiving data, the CPPI will lockup.

  To workaround the issue, do not set the CPPI TD bit. The steps before
  this point ensures the CPPI channel will be torn down properly.

Signed-off-by: Bin Liu <b-liu@ti.com>
Signed-off-by: Felipe Balbi <balbi@ti.com>

The macro EP_MODE_AUTOREG_* should be called EP_MODE_AUTOREQ_*, as they
are used for register AUTOREQ.

Signed-off-by: Bin Liu <b-liu@ti.com>
Signed-off-by: Felipe Balbi <balbi@ti.com>