VBI on PVR-500 stopped working between kernels 3.6 and 3.13

[Christopher Neufeld <media-alias@cneufeld.ca>]

I've been using a PVR-500 to record shows in MythTV, and to capture the VBI
part of the stream from the standard-definition output of my STB when it
records high definition.  This has worked for about 7 years now.

I recently updated my LinHES MythTV distribution, and part of the update
involved moving to a new kernel.  The old kernel went by the code
3.6.7-1-ARCH, while the new one is 3.13.7-2-ARCH.

With the updated kernel, my VBI captures no longer function.
Standard-definition recordings made from MythTV using the PVR-500 before
the update have caption data in the stream, those made after do not.

The retrieval of caption data for high-definition shows involves some
manual scripting to set the modes for the PVR-500, after which I run
ccextractor on the V4L device node and just pull out the captions data (the
audio and video being output separately on high-definition outputs of the
STB, and captured by a HD-PVR).

The script that I use to set up captions invokes this command:
v4l2-ctl -d <DEV> --set-fmt-sliced-vbi=cc --set-ctrl=stream_vbi_format=1

This now errors out.  Part of that is a parsing bug in v4l2-ctl, it wants
to see more text after the 'cc'.  I can change it to 
v4l2-ctl -d <DEV> --set-fmt-sliced-vbi=cc=1 --set-ctrl=stream_vbi_format=1

with this change, it no longer complains about the command line, but it
errors out in the ioctls.  This behaviour is seen with three versions of
v4l2-ctl: the old one packaged with the old kernel, the new one packaged
with the newer kernel, and the git-head, compiled against the headers of
the new kernel.

I strace-d the v4l2-ctl command.  The relevant output is:
open("/dev/pvr_500_1", O_RDWR)          = 3
ioctl(3, VIDIOC_QUERYCAP or VT_OPENQRY, 0x7fff836aac10) = 0
ioctl(3, VIDIOC_QUERYCTRL, 0x7fff836aaa70) = 0
ioctl(3, VIDIOC_QUERYCTRL, 0x7fff836aaa70) = 0
brk(0)                                  = 0x12ca000
brk(0x12eb000)                          = 0x12eb000
ioctl(3, VIDIOC_QUERYCTRL, 0x7fff836aaa70) = 0
			<<<PREVIOUS LINE REPEATS 41 TIMES>>>
ioctl(3, VIDIOC_QUERYCTRL, 0x7fff836aaa70) = 0
ioctl(3, VIDIOC_QUERYCTRL, 0x7fff836aaa70) = -1 EINVAL (Invalid argument)
ioctl(3, VIDIOC_S_FMT or VT_RELDISP, 0x62eae0) = -1 EINVAL (Invalid argument)
fstat(1, {st_mode=S_IFCHR|0600, st_rdev=makedev(136, 0), ...}) = 0
mmap(NULL, 4096, PROT_READ|PROT_WRITE, MAP_PRIVATE|MAP_ANONYMOUS, -1, 0) = 0x7fe8233bf000
write(1, "VIDIOC_S_FMT: failed: Invalid ar"..., 39VIDIOC_S_FMT: failed: Invalid argument
) = 39
close(3)                                = 0
exit_group(-1)                          = ?

I did once see VBI data arriving from one of the paired devices in the
PVR-500, and not from the other.  I would guess that to be because it
started up in that state.  When this happened, I ran v4l2-ctl --all on both
devices, captured the output, and stored it to files.  I did not see any
relevent differences in these outputs, but I present the diff here:

--- file1       2014-10-25 13:40:36.698703171 -0400
+++ file2       2014-10-25 13:40:41.248614481 -0400
@@ -1,15 +1,14 @@
 Driver Info (not using libv4l2):
        Driver name   : ivtv
-       Card type     : WinTV PVR 500 (unit #1)
-       Bus info      : PCI:0000:06:08.0
+       Card type     : WinTV PVR 500 (unit #2)
+       Bus info      : PCI:0000:06:09.0
        Driver version: 3.13.7
-       Capabilities  : 0x81070051
+       Capabilities  : 0x81030051
                Video Capture
                VBI Capture
                Sliced VBI Capture
                Tuner
                Audio
-               Radio
                Read/Write
                Device Capabilities
        Device Caps   : 0x01030001
@@ -18,7 +17,7 @@
                Audio
                Read/Write
 Priority: 2
-Frequency for tuner 0: 4148 (259.250000 MHz)
+Frequency for tuner 0: 884 (55.250000 MHz)
 Tuner 0:
        Name                 : ivtv TV Tuner
        Capabilities         : 62.5 kHz multi-standard stereo lang1 lang2 freq-bands 


The fact that I once saw valid VBI data suggests that the driver is still
capable of the feature, but something about the ioctl invocations in
v4l2-ctl and in MythTV 0.27.4 is wrong for getting the driver reliably into
the state where VBI data is present in the stream coming from the device.


-- 
 Christopher Neufeld
 Home page:  http://www.cneufeld.ca/neufeld
 "Don't edit reality for the sake of simplicity"