Signed-off-by: NLaurent Pinchart <laurent.pinchart@ideasonboard.com>
Signed-off-by: NMauro Carvalho Chehab <mchehab@redhat.com>

The bpp field currently stores the sample width and is aligned to the
next multiple of 8 bits when computing data size in memory. This won't
work anymore for YUYV8_2X8 formats. Split the bpp field into a sample
width and a bytes per pixel value.
Signed-off-by: NLaurent Pinchart <laurent.pinchart@ideasonboard.com>
Acked-by: NSakari Ailus <sakari.ailus@iki.fi>
Signed-off-by: NMauro Carvalho Chehab <mchehab@redhat.com>

Rearrange the CFA interpolation coefficients table based on the Bayer
pattern. Support for non-Bayer CFA patterns is dropped as they were not
correctly supported, and have never been tested.
Signed-off-by: NLaurent Pinchart <laurent.pinchart@ideasonboard.com>
Acked-by: NSakari Ailus <sakari.ailus@iki.fi>
Signed-off-by: NMauro Carvalho Chehab <mchehab@redhat.com>

Enabling gamma bypass disables gamma correction and vice versa. Merge
the two parameters.
Signed-off-by: NLaurent Pinchart <laurent.pinchart@ideasonboard.com>
Acked-by: NSakari Ailus <sakari.ailus@iki.fi>
Signed-off-by: NMauro Carvalho Chehab <mchehab@redhat.com>

Reorder the configuration and enable functions to match the parameters
order.
Signed-off-by: NLaurent Pinchart <laurent.pinchart@ideasonboard.com>
Acked-by: NSakari Ailus <sakari.ailus@iki.fi>
Signed-off-by: NMauro Carvalho Chehab <mchehab@redhat.com>

Instead of using void pointers and offset arithmetics to compute a
pointer to configuration parameters in a generic way, pass the complete
parameters structure to configuration functions and let them access the
parameters they need.

Also modify the enable functions to use a bool enable parameter instead
of a u8.
Signed-off-by: NLaurent Pinchart <laurent.pinchart@ideasonboard.com>
Acked-by: NSakari Ailus <sakari.ailus@iki.fi>
Signed-off-by: NMauro Carvalho Chehab <mchehab@redhat.com>

The feature isn't fully implemented and doesn't work, remove it.
Signed-off-by: NLaurent Pinchart <laurent.pinchart@ideasonboard.com>
Acked-by: NSakari Ailus <sakari.ailus@iki.fi>
Signed-off-by: NMauro Carvalho Chehab <mchehab@redhat.com>

Signed-off-by: NMichael Jones <michael.jones@matrix-vision.de>
Signed-off-by: NLaurent Pinchart <laurent.pinchart@ideasonboard.com>
Signed-off-by: NMauro Carvalho Chehab <mchehab@redhat.com>

Signed-off-by: NPeter Meerwald <pmeerw@pmeerw.net>
Signed-off-by: NLaurent Pinchart <laurent.pinchart@ideasonboard.com>
Signed-off-by: NMauro Carvalho Chehab <mchehab@redhat.com>

This needs to be performed before enabling interrupts as the sensor
might be free-running and the ISP default setting (HS edge) would put an
unnecessary burden on the CPU.
Signed-off-by: NLaurent Pinchart <laurent.pinchart@ideasonboard.com>
Acked-by: NSakari Ailus <sakari.ailus@iki.fi>
Signed-off-by: NMauro Carvalho Chehab <mchehab@redhat.com>

Use the existing omap3isp_subclk_enable() and omap3isp_subclk_disable()
functions instead.
Signed-off-by: NLaurent Pinchart <laurent.pinchart@ideasonboard.com>
Acked-by: NSakari Ailus <sakari.ailus@iki.fi>
Signed-off-by: NMauro Carvalho Chehab <mchehab@redhat.com>

Signed-off-by: NHans Verkuil <hans.verkuil@cisco.com>
Signed-off-by: NMauro Carvalho Chehab <mchehab@redhat.com>

Signed-off-by: NHans Verkuil <hans.verkuil@cisco.com>
Signed-off-by: NMauro Carvalho Chehab <mchehab@redhat.com>

Two of the four possible YUV formats got the wrong colorspace.
Signed-off-by: NHans Verkuil <hans.verkuil@cisco.com>
Signed-off-by: NMauro Carvalho Chehab <mchehab@redhat.com>

The register read function is referencing the dev->ctrlmsg structure outside
of the dev->mutex lock, which can cause corruption of the value if multiple
callers are invoking au0828_readreg() simultaneously.

Use a stack variable to hold the result, and copy the buffer returned by
usb_control_msg() to that variable.

In reality, the whole recv_control_msg() function can probably be collapsed
into au0288_readreg() since it is the only caller.

Also get rid of cmd_msg_dump() since the only case in which the function is
ever called only is ever passed a single byte for the response (and it is
already logged).
Signed-off-by: NDevin Heitmueller <dheitmueller@kernellabs.com>
Signed-off-by: NMauro Carvalho Chehab <mchehab@redhat.com>

Signed-off-by: NEzequiel Garcia <elezegarcia@gmail.com>
Signed-off-by: NMauro Carvalho Chehab <mchehab@redhat.com>

This patch seems for rectifying a typo. But actually the difference between
mantis_vp2040.c and mantis_vp2033.c code is a card name only.
Signed-off-by: NIgor M. Liplianin <liplianin@me.by>
Signed-off-by: NMauro Carvalho Chehab <mchehab@redhat.com>

The original xc5000c driver support was based on a beta version of the
firmware, and there were no redistribution rights.  Change over to using
the release version, for which freely redistributable firmware can be
found here:

http://kernellabs.com/firmware/xc5000/README.xc5000c
http://kernellabs.com/firmware/xc5000/dvb-fe-xc5000c-4.1.30.7.fw

Thanks to Ramon Cazares from Cresta Technology for making the firmware
available as well as working out the licensing issues.

[mchehab@redhat.com: Fix a merge conflict with the patch that added support
 for MODULE_FIRMWARE() macro]
Signed-off-by: NDevin Heitmueller <dheitmueller@kernellabs.com>
Cc: Michael Krufky <mkrufky@kernellabs.com>
Signed-off-by: NMauro Carvalho Chehab <mchehab@redhat.com>

Make the firmware speedup work for the 5000c as well as the original
xc5000.  This cuts firmware load time in half.

Thanks to John Casey at Hauppauge for loaning me a board for testing.
Signed-off-by: NDevin Heitmueller <dheitmueller@kernellabs.com>
Signed-off-by: NMauro Carvalho Chehab <mchehab@redhat.com>

This patch addresses a couple of cases where I forgot to pop open the
gate when in analog mode (a correlary to fix the change made in patch
1c58d5b4a5fca42dce5428bd79b9405878017735).
Signed-off-by: NDevin Heitmueller <dheitmueller@kernellabs.com>
Signed-off-by: NMauro Carvalho Chehab <mchehab@redhat.com>

The driver prints out a dotted version number but it's in hex.  As a
result, the version doesn't visibly match the filename for the firmware,
and it caused a bunch of confusion while discussing different versions
with the chip manufacturer.

Change the firmware printout to be in decimal.
Signed-off-by: NDevin Heitmueller <dheitmueller@kernellabs.com>
Signed-off-by: NMauro Carvalho Chehab <mchehab@redhat.com>

The hack I put in a couple of years ago to avoid clock stretching issues
when talking to the xc5000 worked fine for writes, but intermittently
fails for register reads, because the xc5000 may stretch the clock for
longer between bytes (I was seeing cases of 21 us on the analyzer).

The problem manifested itself as the xc5000 firmware version and PLL
lock register intermittently showing garbage values.

Slow down the i2c bus from 30 KHz to 20 KHz to accommodate.
Signed-off-by: NDevin Heitmueller <dheitmueller@kernellabs.com>
Signed-off-by: NMauro Carvalho Chehab <mchehab@redhat.com>

The xc5000c and newer versions of the xc5000a firmware need minor revisions
to their initialization process.  Add support for validating the firmware
was properly loaded, as well as checking the init status after initialization.

Based on advice from CrestaTech support as well as xc5000 datasheet v2.3.
Signed-off-by: NDevin Heitmueller <dheitmueller@kernellabs.com>
Signed-off-by: NMauro Carvalho Chehab <mchehab@redhat.com>

It's possible for the xc5000 to enter an unknown state such that all
subsequent tuning requests fail.  The only way to recover is to reset the
tuner and reload the firmware.  This problem was detected after several days
straight of issuing tuning requests every five seconds.

Reset the firmware in the event that the PLL is in an unlocked state.  This
solution was provided by the engineer at CrestaTech (the company that acquired
Xceive).
Signed-off-by: NDevin Heitmueller <dheitmueller@kernellabs.com>
Signed-off-by: NMauro Carvalho Chehab <mchehab@redhat.com>

Adjust the retry timeout and number of retries to speed up xc5000
firmware download.  With this change it goes from 4.2 seconds to 2.9.
The net time waited is pretty much the same, but we just poll more
often.

Tested at 250 KHz as well as 30 KHz.
Signed-off-by: NDevin Heitmueller <dheitmueller@kernellabs.com>
Signed-off-by: NMauro Carvalho Chehab <mchehab@redhat.com>

There are no cases where a control message is ever sent to the au0828
with an actual buffer defined.  Remove the reference to dev->ctrlmsg,
which currently requires us to hold a mutex since it is shared with the
read function.
Signed-off-by: NDevin Heitmueller <dheitmueller@kernellabs.com>
Signed-off-by: NMauro Carvalho Chehab <mchehab@redhat.com>

Put a hack in place to speed up the firmware load in the case that the
xc5000 has just been reset.  The chip can safely do 400 KHz in this mode,
while in normal operation it can only do 100 KHz.

This reduces the firmware load time from 6.9 seconds to 4.2.
Signed-off-by: NDevin Heitmueller <dheitmueller@kernellabs.com>
Signed-off-by: NMauro Carvalho Chehab <mchehab@redhat.com>

We weren't checking whether the resource was in use before calling
res_free(), so applications which called STREAMOFF on a v4l2 device that
wasn't already streaming would cause a BUG() to be hit (MythTV).
Reported-by: NLarry Finger <larry.finger@lwfinger.net>
Reported-by: NJay Harbeston <jharbestonus@gmail.com>
Signed-off-by: NDevin Heitmueller <dheitmueller@kernellabs.com>
Signed-off-by: NMauro Carvalho Chehab <mchehab@redhat.com>

The original implementation of the analog support would use an
i2c_gate_ctrl function when using the digital side of the au8522, but on
the analog side we would always just force the gate open and leave it
open all the time.

This can have adverse effects on the xc5000 given the tuner is receiving
all the spurious i2c traffic (a problem which can be exaggerated due to
bugs in the au0828 i2c hardware implementation).

Rework the existing hack to only open/close the gate when actually
talking to the tuner.

This logic might need to be reworked a bit if anybody ever tries to add
support for a board that has the au0828/au8522 but doesn't have digital
support implemented (because the i2c_gate_ctrl callback is being set in
the DVB attach).  However given how few different models are in
circulation, this can be deferred until such a situation arises (if
ever).
Signed-off-by: NDevin Heitmueller <dheitmueller@kernellabs.com>
Signed-off-by: NMauro Carvalho Chehab <mchehab@redhat.com>

The current code invokes the auto calibration of the tuner whenever the
init routine is called (whenever the DVB frontend opens the device).
However we should really only be invoking the calibration if we actually
did reset the device and reload the firmware.

Rework the routine to only do calibration if reset and firmware load was
performed.  Also because the called function is now a no-op if the
firmware is already loaded, the caller no longer needs to invoke
is_firmware_loaded().
Signed-off-by: NDevin Heitmueller <dheitmueller@kernellabs.com>
Signed-off-by: NMauro Carvalho Chehab <mchehab@redhat.com>

The au8522 driver was broken into three modules (dig, decoder, common),
and as a result the debug modprobe option doesn't work for any of the
common functions.

Copy the module macros over to the common module so that the debug
option works again.
Signed-off-by: NDevin Heitmueller <dheitmueller@kernellabs.com>
Signed-off-by: NMauro Carvalho Chehab <mchehab@redhat.com>

If the user performs a tuning attempt without explicitly calling the
s_std ioctl(), a value of zero is sent from tuner-core to xc5000.  This
causes the xc5000 driver to leave the standard unchanged.  The problem
was masked by the fact that the xc5000 driver defaulted to NTSC, but if
you happened to perform an ATSC/ClearQAM tuning attempt and then do an
analog tune, the net effect is an analog tune with the standard still
set to DTV6.

Keep track of whether the standard has ever been sent to tuner-core.  We
don't make an s_std subdev call explicitly during probe because that
will cause a firmware load (which is very time consuming on the 950q).
With the logic in this patch, the s_std call will occur automatically on
the s_freq call if it hasn't already been set.
Signed-off-by: NDevin Heitmueller <dheitmueller@kernellabs.com>
Signed-off-by: NMauro Carvalho Chehab <mchehab@redhat.com>

In some cases users would see the xc5000_attach() call failing for the
digital side of the tuner on initialization.  This is because of udev
running v4l-id while the digital side of the board is still coming up.

This is the exact same race condition which was present in em28xx (not
surprising since I copied all the locking logic from that driver when I
added analog support).  Reproduce Mauro's fix from the em28xx driver in
au0828.
Reported-by: NRick Harding <rharding@mitechie.com>
Signed-off-by: NDevin Heitmueller <dheitmueller@kernellabs.com>
Signed-off-by: NMauro Carvalho Chehab <mchehab@redhat.com>

The logic as written would *never* actually return an error condition,
since the loop would run until the counter hit zero but the check was
for a value less than zero.
Signed-off-by: NDevin Heitmueller <dheitmueller@kernellabs.com>
Signed-off-by: NMauro Carvalho Chehab <mchehab@redhat.com>

When debugging is enabled, also show the analog SNR and the total gain
status values.
Signed-off-by: NDevin Heitmueller <dheitmueller@kernellabs.com>
Signed-off-by: NMauro Carvalho Chehab <mchehab@redhat.com>

The quality register only has relevant data in bits 2-0, so discard the
other bits (which results in a value being printed that is consistent
with the expected 0-7 range).
Signed-off-by: NDevin Heitmueller <dheitmueller@kernellabs.com>
Signed-off-by: NMauro Carvalho Chehab <mchehab@redhat.com>

The g_reg and s_reg calls worked properly if acting on subdev registers
(such as the au8522), but didn't work against the au0828 itself.  Copy
the logic over from em28xx.
Signed-off-by: NDevin Heitmueller <dheitmueller@kernellabs.com>
Signed-off-by: NMauro Carvalho Chehab <mchehab@redhat.com>

There is an apparent bug in the au8522 TS clocking which can result in
it delivering a TS payload to the au0828 that is shifted by some number
of bits. For example, the device will announce a packet containing "FA
38 FF F8" which if you shift left one bit is "1F 47 1F FF F0..."

This presents itself as no TS stream being delivered from the kernel to
userland, since the kernel demux will drop every packet.

In the event that this condition occurs, restart the DVB stream.

Also, this patch includes a couple of lines of cleanup to not change the
FIFO configuration while the FIFO is running (which can screw up the
state machine), and dequeue the buffers before turning off the FIFO.
This puts the logic in sync with the Windows driver.
Signed-off-by: NDevin Heitmueller <dheitmueller@kernellabs.com>
Signed-off-by: NMauro Carvalho Chehab <mchehab@redhat.com>

The table of valid SNR values for QAM 256 is off by one, and as a result
if the SNR is oscillating between 40.0 and 39.9 dB, tools like azap show
it going back and forth between 40.0 and 0 (misleading some people,
including myself, to think signal lock is being lost or there is a
problem with register reads).

Fix the table so that 40.0 dB is properly represented.

Cc: Steven Toth <stoth@kernellabs.com>
Signed-off-by: NDevin Heitmueller <dheitmueller@kernellabs.com>
Signed-off-by: NMauro Carvalho Chehab <mchehab@redhat.com>

It turns up the autodetection for the video standard in the au8522 is
prone to hanging the chip until a reset is performed.  This condition is
trivial to reproduce simply by tuning to a station and then rapidly
unplugging/ replugging the coax feed.

Because we've never claimed to support anything other than NTSC-M, just
disable the video-standard autodetection logic and force it to always be
NTSC-M.
Signed-off-by: NDevin Heitmueller <dheitmueller@kernellabs.com>
Signed-off-by: NMauro Carvalho Chehab <mchehab@redhat.com>