pvrusb2 Linux Driver Change History
There are 464 changes across 106 snapshots shown here.
Minor internal adjustments that don't affect functionality.
Fix 7MHz and 8MHz DVB-T tuner support for HVR-1900 rev D1F5 (Mike Krufky).
Enable use of ioctl2 internal V4L interface. This is a large change but
it was a very long time in coming. Thanks to Hans Verkuil for blazing the
trail on this. This change completely alters how the pvrusb2 driver
interacts with the V4L API as seen by applications. Previously the
pvrusb2 driver implemented the actual ioctl() functions directly; now this
is done in the V4L core via a function dispatch table out to the driver
for each specific API function. This strategy allows for more interesting
flexibility in that the V4L core can directly implement things that would
be a nuisance for the driver (like video standards enumeration). All
ongoing V4L development uses ioctl2 so we really must participate in this
in order to keep up. Note: Since the stand-alone driver keeps backwards
compatibility with older kernels, this functionality is only compiled-in
when compiling against kernel versions 3.0.x or newer. The earliest this
change will find its way to the kernel tree will probably not be until
3.4.x or 3.5.x.
Tie-in querystd operation (new with previous snapshot) with use of ioctl2.
Make output of querystd available via sysfs interface, by implementing as
an internal control ID.
Fix truncated video standard name strings that sometimes happened in sysfs
Adjust V4L's enumeration of video standards on behalf of this driver
(which happens now with use of ioctl2) to be limited to just those
standards that we know the hardware supports.
Fix querystd implementation to be limited to just those standards that we
know the hardware supports.
Remove pvrusb2 driver's own implementation for enumerating video
standards. With the use of ioctl2, this now happens on our behalf inside
the V4L core. I have NOT checked that the standards list generated is the
same, but empirical testing so far looks OK. Also note that in order to
remain backwards compatible, the driver internal implementation is still
compiled in if ioctl2 is not used. Said another way, this change only
affects those who compile the standalone driver against kernel versions
3.0.x or newer. This was another large alteration to the driver.
IR interface updates to remain compatible with surrounding environment.
Code added to support videodev_ioctl2(), however it is deactivated for the
moment until further testing can be done.
g/s_ tuner support fix needed to remain compatible with surrounding
Minor allocation failure check added by a kernel janitor.
Remove old, dead, useless module/driver version info which dated back to
the earliest days of this driver.
Return ENOTTY instead of EINVAL if an unsupported ioctl() is requested in
the V4L2 API.
Various other miscellaneous tweaks to be compatible with recent kernel
Clean up s-video input for Terratec Grabster. Thanks go to Sven Barth for
figuring this out.
Various cosmetic cleanups.
Various updates to keep up with the moving target that is the Linux
kernel... (up to 2.6.38)
Fix potential memory leak involving device tear-down.
Provide more information about IR units to IR subsystem, other related
changes to improve IR support.
Remove V4L1 compatibility when built against 2.6.38 or later.
Resync with some very minor changes from upstream
Make compilable under 2.6.37
Fix missing mutex release in video buffer handling path. This was a
longstanding bug in an error leg (apparently never hit), detected via
source code analysis by Julia Lawall.
Very small tweak in an attempt to make possible sliced VBI capture. No,
VBI still does not work...
Long anticipated cropping tweaks from Servaas Vandenberghe.
Various internal changes to support correct operation for kernels 2.6.35
Various internal changes to support compatible operation with recent
changes to v4l-dvb.
Fix race condition during tear-down of the sysfs interface. This bug
apparently has been present forever but only became malignant with the
2.6.34 kernel. Nodes under /sys/class/pvrusb2 are represented by kernel
device structures. The /sys/class/pvrusb2 head however is mapped by a
kernel "class" structure. Each underlying device structure
points back to the class but there's no protection in the kernel against
cases where the class might get deleted while the device instances are
still around. Thus the device instances can end up pointing to deleted
memory if this is not guarded by the driver. From looking at the kernel
code there appears to be an assumption that a class definition is
logically static compared to the device instances. So, change the pvrusb2
driver to not delete the /sys/class/pvrusb2 parent until after everything
else in the driver has been shut down first (and thus the device instances
Increase sysfs attribute array size by one. This is an array of pointers;
its size is learned by the kernel by noticing a trailing null pointer. We
weren't allowing enough space for the worse case. Again, this hadn't been
causing any outward symptoms, yet.
Perform correct reference counting of the usb device pointer as referenced
through the sysfs interface. This hadn't been causing any outward
Call sysfs_attr_init() on all sysfs attributes - this is something new for
kernel 2.6.34 or later. Failing to do this generates a lot of noise in
the kernel log.
Add missing #include <linux/slab.h> to pvrusb2-v4l2.c that
apparently wasn't a real problem until compiling with the 2.6.34 kernel.
Fix kernel oops caused by touching deleted memory after unregistration.
This bug was introduced when we had started using video_device_node_name()
to generate the device name. Rearranged the code a bit to let
video_device_node_name() to do its thing before we actually unregister the
Implement ability to mark support for specific devices as being
experimental. The Terratec Grabster AV400 is one such device (note that
it isn't yet listed in any of the pvrusb2 documentation either).
Avoid use of stack-allocated memory for USB buffer I/O. The pvrusb2
driver has been doing this without issue for years; only now has this
apparently become a problem in the Linux kernel. With the 2.6.34 kernel,
stack-based USB buffers generate nasty messages to the kernel log.
Fix operation of RF tuner with Gotview hardware. This bug was introduced
back when the driver implemented the sub-device model of module control.
The root cause was that the tda9887 module needs to be explicitly loaded
with the new model (with the old model it was automatic), and I had missed
with this the Gotview device definition.
Selectively enforce a 300msec quiet period between the time when the video
digitizer is started and when the encoder is started. For devices which
use an saa7115 video digitizer, this has been observed to solve sporadic
black video problems on streaming startup. The delay is only implemented
when a saa7115 is being used. Thanks go to Martin Dauskardt
<firstname.lastname@example.org> for solving this. With this change, we
also shorten the encoder configuration wait (now determined to have not
Pull in change from v4l-dvb which implements stream on/off control for
audio sub-devices, when the functionality is available. This may reduce
incidences of tinny audio on startup.
Misc other changes to maintain interoperability with the v4l-dvb subsystem
(all implemented here to be backwards compatible).
Further fix to support use of larger FX2 16KB firmware. Missed one spot
in the firmware loader and forgot to include HVR-1900 alongside HVR-1950
in the set of devices which could load the larger firmware.
Bind I2C address 0x71 for Zilog IR devices (needed specifically for
Shorten hardware descriptive text to fit in too-small V4L API buffer.
(This was not a showstopper problem - the driver never overran the buffer
but just correctly truncated the contents.)
Support use of larger FX2 16KB firmware.
Log a description of the hardware when the driver is associated with said
Include hardware description in sysfs debuginfo output.
Implement some additional logic during initialization to differentiate
between failures due to jammed hardware vs failures due to missing /
messed up sub-device kernel modules. This should result in slightly more
useful kernel messages about the problem.
Include patch from Mike Krufky to "disable the tda18271 slave tuner
output / loop thru options when in low power mode" (this is specific
to HVR-1950 and is only relevant for kernel 2.6.32 or later).
Minor internal changes that track ongoing changes within v4l-dvb.
Associate all pvrusb2-driven V4L device nodes in sysfs with the underlying
USB device. This allows udev to see the hardware device properties, which
in turn makes possible customization of the V4L device node setup via
rules added to udev. Note that this implementation only works for kernels
back to 2.6.24 - before that point the feature is disabled because I
cannot find a way to cleanly break the V4L->USB device association when
the USB device goes away before the V4L device (a fact of life in a
Fix compilation issue in pvrusb2-devattr.c for kernels older than 2.6.31.
An earlier tweak there relies on a struct change that only appears in
kernel 2.6.31 (or v4l-dvb). Conditionalize the change appropriate to the
Enable DVB usage in the standalone driver when compiling under kernels
2.6.26 or later. Previously this was only enabled when building against
v4l-dvb. Without this change, support for the digital side of the
HVR-1950/HVR-1900 would have been missing. Several people have gotten hit
by this but I had failed to realize the root cause - that the DVB portions
of the pvrusb2 driver were simply disabled - until now. D'Oh!!!
Fix problem involving correct determination of undefined audio / video
routing. Routing tables for various chips are sparse arrays but their
"sparseness" was incorrectly set up. With this change we can
now handle detection of undefined routing setup.
Change default television frequency to US broadcast channel 3 (previously
was 7). This has zero impact on apps; it only helps with my own testing.
With the phase out of OTA analog television in the US, the previous value
became useless to me. I have to set some kind of default here, so I might
as well choose something useful (which in this case corresponds to my
analog RF modulator).
Initialize VBI for cx25840 any time the video standard is changed. This
is because the VBI initialization code in the cx25840 appears to use
aspects of the video standard as input so we want to make sure that this
code runs any time the standard is changed. No, this does not enable use
of VBI; it is a further refinement on the video scaling bug fix.
Fix horizontal video scaling bug involving analog capture where a cx25840
is used. This is an old bug - previously fixed back in Nov 2006 (video
was being distorted / corrupted for horizontal resolutions other than 720
pixels). The problem was that the fix had been rendered ineffective by
other changes. VBI is now initialized in the cx25840 driver only after the
chip has been initialized (fixes 24xxx) and again after any time the chip
is reset (fixes HVR-1950).
Additional tweaks to HVR-1900 device configuration having to do with I/F
frequencies in the tuner (from Steve Toth).
Instantiate ir_video I2C device by default - basically change the default
value of the "disable_autoload_ir_video" option to zero.
Configuration tweaks from Steve Toth for Hauppauge HVR-1900.
Implement experimental support for Terratec Grabster AV400. This is very
experimental; I am including this so that others can test and help improve
things. I strongly suspect that since this does not include any help for
the CS5340 that I'm told is on the board that audio might still not quite
be right... Credit goes to Andrea Venturis, Tobias Seiler, and Dave Craig
for providing the critical details to make this possible.
Remove use of usb_settoggle(); this is no longer available with later
Attempt to autoload ir-kbd (v4l-dvb's module for some IR receivers) where
appropriate. Note! This is new behavior and may interfere if lirc was
being used. A new module option is defined to suppress this:
disable_autoload_ir_kbd. It is entirely possible that this option may
disappear in the future, if/when lirc is updated to operate with the new
i2c binding mechanism. For now, we default this option to "1",
i.e. we won't try to autoload anything to handle IR.
Include configuration data to select and track actual IR scheme in use.
This change includes reporting of the IR scheme in use via the driver's
debug info output.
Correctly initialize v4l2_routing structures everywhere. This wasn't
causing a problem, but this is the sort of thing that should be cleaned
Account for various API changes related to v4l2-subdev changes since
Fix unitialized struct element during tuner setup. Stack allocated
foreign struct instances should always be memset() to zero in order to
guard against fields being added later (and thus being left
uninitialized). This bug was causing bogus agc warnings from the tuner
module when using an HVR-1950. This bug has also been around for a very
In sysfs, show default values symbolically, same as current values, where
Fix longstanding bug (since Aug 2008) that caused corrupted values to be
returned for control default values when the control is not an integer.
Actually that bug propagated from another bug dating back to Apr 2006
where default values for non-integer valued controls was always returning
Various miscellaneous tweaks to preserve reasonably correct compilation
for older kernels (test compiled back to 22.214.171.124).
Implement and enable composite and s-video inputs for OnAir devices.
Remove ancient IVTV_IOC_G_CODEC and IVTV_IOC_S_CODEC ioctl functions.
These are very very old, were non-standard, and were only present to keep
MythTV happy (their implementation did nothing except to report success).
That was long ago; no recent versions of MythTV should require this
MAJOR: Implement a new client modules layer, based upon the v4l2-subdev
framework. Note that this is only active in the standalone driver when
building against v4l-dvb or in a kernel that is 2.6.29 or later. The
v4l-dvb core has a new mechanism known as "v4l2-subdev". This
is a whole new method for finding, attaching to, and communicating with
all the various chip-level drivers. This is the "official" way
of controlling all the "sub-drivers" now. This is also expected
to be required starting with kernel 2.6.30 (i.e. support for the old i2c
specific mechanism will be removed). The pvrusb2 standalone driver for
now still has the old method implemented alongside the new one - compile
time options select which method to build depending on the environment
(i.e. kernel version). So backwards compatibility for now is not affected
and the driver should _in theory_ continue to work all the way back to
2.6.12 (conditioned of course on the other various things mentioned in the
setup documentation). But at some point in the future, I expect it will
become too cumbersome to keep both methods working, at which point I will
remove the old method and the standalone driver will then no longer
directly support kernels older than approximately 2.6.29 (the point when
the v4l2-subdev mechanism first became workable and resident in the
Fix build so that V4LDIR trick works again for building against v4l-dvb.
Use usb_make_path (where possible) to report device bus location. This
produces a more stable result than what had been done before.
Stop advertising VBI capability through the V4L interface. It's never
been supported by this driver.
Fix build breakage for 2.6.28
Changes to hopefully support 2.6.29
Issue VIDIOC_INT_INIT when modules first bind to the driver. This is
needed because cx25840 in particular now depends on being told to
Fix deadlock problem that started with kernel 2.6.27.x. New behavior in
this kernel causes a USB disconnect if a USB device reset is requested by
the driver. Previously there was no disconnect and the driver retained
control of the device. The pvrusb2 driver, as controlled by the
initusbreset option, can trigger such a reset as part of device
initialization - and I've had this enabled by default. Unfortunately now
with a disconnect taking place, then the driver ends up cleaning up and
then starting over with the next connect - and then triggering another
reset. Worse still, the USB core is issuing the disconnect callback in
the same thread context as the thread which requested the reset, which
results in a recursive lock entry inside the pvrusb2 driver, thus setting
up the deadlock problem. Once in this state, the only way out is a
reboot. Really, the function of "initusbreset" amounted to
"chicken soup" anyway; it didn't hurt and might help. So this
fix simply removes the function. It is no longer possible to
automatically reset the device on initialization. The change is
conditioned on the kernel version, so if you build against a kernel older
than 2.6.27 then the option (for what it's worth) will still be present.
Implement a minor tweak to the mpeg encoder programming to hopefully keep
PTSes from "drifting away". This fix was discovered empirically
by Boris Dores.
Disable "virtual" IR receiver on for 24xxx devices that have an
internal IR blaster. In that case there's another another IR receiver
present and to leave the virtual receiver available just causes confusion.
This means that 24xxx users will no longer see a phantom IR chip.
Plumbing added to support cropping. Note that this will only work if
corresponding support exists and works in cx25840.ko or saa7115.ko
(depending on the device you have). Credit to Servaas Vandenberghe for
laying the groundwork for this. Access is implemented for both the normal
V4L ioctl interface published by the driver and sysfs.
Implement "def_val" in sysfs interface. Controls already
support default values, this makes those values visible in sysfs.
Detect and trap attempts to use an "alien" USB ID. The driver
expects a descriptor associated with the USB ID; if that descriptor is
missing, print a message and fail. This can happen if the user tries to
insert a new USB ID via sysfs - without this change the driver can
otherwise cause an oops.
Various minor tweaks to keep up with the surrounding kernel and v4l-dvb
Add USB ID 2040:2950 - this is a really old 29xxx device. It lacks and IR
receiver but is otherwise functionally identical to the 2040:2900 device.
Since the IR detection for this device is a separate action anyway, we can
safely alias the 2950 id to 2900.
Include v4l2-ioctl.h when kernel is 2.6.27 or later. This is a critical
change to support kernel 2.6.27.
Rework lots of odds and ends to maintain the ability to build old kernels
in spite of what has changed in the v4l-dvb repository. At this point
v4l-dvb apparently no longer supports kernels older than 2.6.16. This
driver however *should* still compile and function in at least a limited
fashion for kernels going back to 2.6.12, provided the documentation's
instructions for setup are followed for such kernels.
For HVR-1950, the IR chip is held in reset at powerup. Implement a small
change to ensure that the chip has been released from reset.
Select an appropriate GOP size value for the video standard in effect and
set it when the video standard is changed. This is not a big stretch from
what the driver had already been doing with the vertical capture
Track changes to cx2341x API (fixes compilation issue for kernel versions
2.6.27 or later).
Port hardware name and description nodes over to legacy sysfs
implementation from the more recent sysfs implementation.
Increase duration of enforced encoder "quiet period" from 50msec
to 100msec. This seems to improve the odds that the encoder will
correctly start streaming.
Implement tv_freq and radio_freq module options. These set the initial
default TV and radio frequencies upon initialization. Under normal use one
does not care about this (since the app is always going to set these
anyway); this is however useful for debugging.
Rework pvrusb2-sysfs.c to eliminate a giant pile of hacked up crap that
dates from the early days of the driver. Back then the kernel's class
device mechanism did not have a means for passing per-attribute
information so an ugly thunking mechanism had been implemented as a
workaround. These days we use the kernel's device mechanism which had
been enhanced to handle sysfs classes - and with that we can pass per
attribute info. So remove all the gunk and clean this up.
Enable FM radio support for "Gotview USB2.0 DVD 2".
Implement new mechanism to enforce input choices depending on interface
being used. An interface can now declare what inputs it will handle, and
the driver core will enforce it - possibly doing a mode switch as a
result. Thus the DVB interface will only handle the dtv input, the v4l
radio interface will only handle the radio input, and the v4l video
interface will handle all except the dtv input. Hopefully further in the
future it will become possible to stream digital over v4l or mpeg-encoded
analog streams over dvb; then these constraints will be relaxed. But for
now the outward effect is that digital streaming will be automatic when
DVB is used and analog streaming (via mpeg data) will be automatic when
V4L is used. This is a significant change, as this finally enables full
use of supported hybrid devices in digital mode. (Though at the moment,
the set of such devices is rather small.)
Apparently the kernel developers no longer consider it proper etiquette to
use __FUNCTION__; everyone must instead use __func__ (even though it
breaks with older compilers). And worse still, actual effort is being
expended to sweep this change throughout the kernel source tree. Don't
these people have better things to do? So... completely clean out all use
of __FUNCTION__ from the pvrusb2 driver (it was just in the sysfs
interface). I'm not going to use __func__ either. So there.
Fix build so that DVB interface build can succeed when using the V4LDIR
trick to perform a standalone driver build against a v4l-dvb repository.
Fix connect / disconnect race. The driver wasn't properly sequencing
disconnects against connects. Said another way, if a disconnect happens
while the driver is still initializing, then Bad Things can happen. The
fix uses a lock to sequence this properly, and does a check at the top of
initialization (after entering the lock) to ensure that the USB core is
Fix race on module unload. Absolutely amazing that this bug apparently
went undetected since the early days of the driver. ALL references into
the pvrusb2 module must be gone before pvr_exit() can be allowed to
return. That includes any running kernel threads, however previously the
driver wasn't tracking kernel thread lifetime at all. This could lead to
an oops if a thread was still cleaning up after the module exit function
returns. Fixing this required rewriting a large fraction of
pvrusb2-context.c. The good news however is that now the driver only
needs one control thread instead of one control thread per instance.
(However there is still a worker thread per instance.)
Implement streaming quirk for OnAir: Ensure that encoder has run for at
least 1/4 second before digital streaming is attempted. This was
determined through experimentation. And while this seemingly simple
requirement should be easy to do, it wasn't because we also had to account
for cases where the encoder's firmware might get reloaded or on the
off-chance that any previous streaming in analog mode didn't meet the 1/4
second minimum. Plus we still have to recover if that quick attempt to
run the encoder happens to fail. So a number of state machine changes
were needed. This is all keyed by setting the OnAir digital streaming
scheme in the device attribute table.
New device attribute defined to indicate that the encoder is required to
be healthy in order to do digital streaming. Normally the encoder should
only be needed for analog streaming, but at least one case (OnAir) has
been identified where the encoder still has to be able to do stuff before
digital streaming can start. If this attribute is set, then the device
will require encoder firmware before it will stream anything.
Fix stuck feed thread in new (still experimental) DVB interface. If the
device fails to stream, the feed thread will block forever waiting for
buffers. But while in this state it was not looking for an exit condition
from the driver DVB interface. This caused the thread to jam.
Add digital streaming definitions for Hauppauge 75xxx models (new hardware
coming up). Also define additional USB IDs for this device.
Add digital streaming definitions for OnAir hardware.
Rework code which starts / stops USB streaming. Previous mechanism
assumed too much about the device type. This helps streaming with OnAir.
Implement some new infrastructure for sending simple FX2 commands. This
allows more intelligent handling / logging of such commands and simplifies
everyone else that can use it now.
Implement logic to collect USB buffering statistics. Output can be seen
by cat'ing debugcmd or debuginfo (which also triggers entries into the
kernel log) via sysfs.
Tuner type tweaks in pvrusb2-devattr.c for OnAir, needed for new v4l-dvb
snapshots (and kernels in the future).
Merge in a new DVB interface. No, this is not ready for prime-time yet.
Compilation of this interface is controlled by the variable
CONFIG_PVRUSB2_VIDEO_DVB, which is kept off for the standalone build. For
now. Coordination between the DVB interface and the rest of the driver is
not quite entirely there yet.
Add USB IDs for new Hauppauge devices (73xxx and 75xxx).
Set default video standard for OnAir devices to NTSC.
Fix various compilation warnings caused by a shifting kernel environment.
Clean up a spelling error or two as well. Fix some C99-questionable
Completely rework context handling and initialization. A new kernel
thread is created per-instance to manage overall device state. Hardware
initialization is moved into this thread. Initialization of all interfaces
now happens in the context of this thread. Previously these two steps
took place in the context of a work queue item, but that could cause a
deadlock if the initialization code itself needed another work queue item
to run (e.g. core state change). The deadlock risk for this is actually
zero since no interfaces try this, but there's a new interface coming
which might need to operate some driver controls during initialization.
Interface tear-down and notification updates also happen through this new
thread. Also with the use of this new thread, the master context mutex is
no longer required during interface initialization, which solves a very
very longstanding mutex deadlock race (task 1 grabs mutex A then B while
task 2 grabs mutex B then A) during driver initialization. Nobody I know
of has ever actually been directly burned by this, but the kernel's own
mutex deadlock detector can find it. Short summary: New kernel thread has
been introduced to manage overall driver instance handling and this solves
a number of nagging (though currently benign but later to be malignant)
Close a potential (but improbable) timer race during driver instance
tear-down. This could have caused a kernel oops, though I've never gotten
an actual report of this happening.
Close a potential (but extremely unlikely) thread race during
Increase staging buffer size for code which prints video standard strings.
There's enough possible combinations now that the previous buffer wasn't
large enough. (The buffer printing code tracks available size so this
wasn't an overrun problem, just truncated text.)
Enable support for "GOTVIEW USB2.0 DVD Deluxe" hardware.
Device attribute structure is entirely read-only and is intended to be
compact. It does not exist in any context involving user application
access nor can it be touched in any external manner which might be
sensitive to packing order. Therefore, shrink the space used by switching
enumerations to char type and using single bit fields for flag items.
Abstract LED control out so that it can be handled as a device-specific
Allow digital streaming to work even if the encoder firmware could not be
loaded (because the encoder is not in the digital streaming path so it
does not matter if the encoder is working or not).
Lots of new generic control logic implemented to handle streaming from a
digital source. This is a major change. None of this is actually in use
yet - these changes just lay the foundation for digital streaming control.
Fix oops in stream allocation logic (not possible in V4L interface, was
found while debugging a new interface).
Improve input selection logic so that only selections which make sense are
made available. For example, if the device doesn't have an FM radio, then
there's no radio choice. Similarly, if the device doesn't have a digital
tuner, then no dtv choice either. This determination is made for all
input choices where known. Device attributes are used to control this.
Foundation laid for new input choice: dtv. This eventually will play a
role in supporting devices that also have a digital tuner.
Implement analog support for Hauppauge HVR-1950.
Implement experimental support for the OnAir Creator and older OnAir USB2
tuners. These are hybrid devices. Support here however is currently
limited to just the analog side. Only television input has been
successfully tested, so far.
Fix Gotview misconfiguation - mark that a cx25840 is in use there. This
has implications in various parts of the driver that must adjust to the
presence of that chip.
Recognize and print out video standard mask bit values for ATSC standards.
(This is currently of no real use, but will likely play a part if/when we
start handling hybrid devices.)
Remove use of volatile for various bits of state that might be shared
across threads. I have become convinced that such use of volatile should
not be required here.
Fix critical 1-character typo in logic which attempts to make available
PAL-D1 for PAL-D/K devices. Patch from Roel Kluin
Implement the ability to set up a default video standard based on detected
Enable support for "GOTVIEW USB2.0 DVD2" hardware.
Add device_hardware_type and device_hardware_description nodes to the
sysfs interface. Reading these files will return a single token (for
device_hardware_type) or a one line description (for
device_hardware_description), based on the actual device type being
Collect all information that describes specific device types and move it
all into a single module (pvrusb2-devattr.c). No more scattered logic
that tries to adjust behavior on perceived device type. Instead, a
central descriptor contains information about device attributes which is
then used through the driver to tailor the correct behavior.
Implement concept of device-specific audio / video routing scheme. It is
possible now to specify the signal routing scheme as part of the device
description - making possible support of other devices which use different
Note: This driver version implements a massive pile of changes mostly
having to do with control of the streaming pipeline. This is the largest
amount of churn since the 20051231 snapshot back in Dec 2005 when the I2C
module control mechanism was reworked. I have however beat on the driver
quite a bit and it looks very solid. The most obvious outward effect of
these changes is that video corruption now should be far far more
difficult to cause. Maybe it's completely gone now (I can no longer cause
it to corrupt at all.) Thanks go to Hans Verkuil, who provided me with
the key clues here that resulted in these changes. But with all that
said, this is a big block of changes and I definitely want to hear if
people see any new problems crop up. (I'd also like to hear if people
actually see any improvements too.)
Rip out subsystem style pipeline control logic and replace it with
separate independent state machines, one per bit of state being managed.
Each bit of state tracks one aspect of the pipeline and each state machine
manages that bit of state autonomously, according to surrounding driver
state and input from the controls and pipeline streaming state.
Implement additional state control to correctly prevent streaming from
initiating if there's no associated digitizer chip-level driver. And
automatically recover if/when the driver in fact does later on appear.
Move kernel thread operation from surrounding pvrusb2 context structure
and place it directly in the driver core. Along the way, discover and fix
a race condition that could cause a kernel oops on driver removal (this
has been there all along).
Move all pipeline control into the kernel thread. Operation of streaming
start / stop now happens in the kernel thread context and not in any
process context from an app. To start / stop streaming, an app now sets a
request bit and waits for the kernel thread to act on the request.
Split apart control update such that the actual update / commit now takes
place in the kernel thread - synchronized with the pipeline control. This
enables far better control of when/how control changes are reflected into
Implement a 50 msec timer to determine when the video digitizer has
quiesced after being stopped. The kernel thread now uses this (via the
afore-mentioned state machines) to ensure that the encoder won't be
configured until the digitizer is known to be "quiet". Hans
Verkuil has discovered that it is not safe to initialize the encoder until
(a) the digitizer is not streaming and (b) the digitizer has been idle for
at least 1 frame time. Since we can't easily measure a frame time, the
driver just assumes the worst case and measures 50msec.
Implement another 50 msec timer to enforce a short idle interval just
before the encoder is configured. I was still able to (very rarely) see
corrupted video even with the previously mentioned "wait for the
digitizer to be quiescent" fix. This change enforces a much more
strict delay - the kernel thread won't even try to touch the encoder until
the hardware has been basically not messed with for 50 msec. With this
change I am no longer able to trigger video corruption at all. I suspect
that with this change, the other 50msec timer can be eliminated, but for
the moment I am leaving both in place.
The cx2341x module can notice certain classes of control changes that are
not safe while the encoder is running. The driver now obeys this case -
when it happens, the kernel thread will handle it by temporarily shutting
down the pipeline, then it will execute the control changes and start the
stream up again. Under most cases the entire sequence should happen in a
fraction of a second. NOTE: This change is only of benefit if the driver
has been compiled to use the cx2341x module, and that only will happen if
the driver is built under kernel 2.6.18 or later (or as part of the
v4l-dvb distribution under any 2.6.x kernel supported there).
Permit the driver to still be partially functional even if the cx23416
firmware could not be located. Obviously streaming won't work in this
case, but now it is possible to reach the driver through sysfs and use it,
for example, to capture the fx2 firmware.
Fix kernel oops involving sysfs support when unloading the driver. This
problem was triggered by change over to the new /sys/class mechanism
Fix minor bug which affected handling of token lists with certain
controls. This issue has been around for quite a while; nobody noticed it
until a kernel janitor audited the code.
Additional logic to attempt a better job at tracking the health of the
mpeg2 encoder chip.
Stop handling I2C_M_NOSTART flag; this is deprecated and going away.
Get rid of use of USB_DEVICE(0,0), seems to break gcc in some cases. Yes,
this appears to be a compiler bug!
Internal list handling cleanup from Trent Piepho (slight performance
improvement, code maintenance improvement).
The kernel-approved method for handling /sys/class changed a while back
(around 2.6.19, usage of struct class_device --> changes to struct
device). While the "old" way still works as of 2.6.22, it is on
its way out and we here have to move forward. Unfortunately the needed
changes break compatibility with older kernels. To deal with that here,
I've forked the pvrusb2-sysfs module such that the older version will be
built if the kernel target is older than 2.6.22. For anything newer, the
newer pvrusb2-sysfs implementation will be compiled. Note: An implication
of this is that when using the in-V4L driver (not the standalone or
in-kernel versions), the sysfs interface will no longer be supported at
all for kernels older than roughly 2.6.19.
Do a better job at guessing the initial video standard to set, based upon
which standards the device reports it can be handle. Also log this
information with a separate debug mask control bit.
Fix oops that might happen if a devices fails to correctly initialize
(missing a null pointer check).
Improve code which discriminates device types, so that it will be easier
to add other types. This is just a minor cleanup, not a refactoring.
Implement programmatic mechanism to allow direct retrieval of device ROM
Update use of cx2341x module to match changed API for kernel 2.6.24 or
later or when built against v4l-dvb.
Fix compilation breakage for kernels older than 2.6.16 (bug was introduced
in previous snapshot).
Implement module option to prevent IR reception on a per instance basis.
This is useful in situations where there are multiple tuners and we only
want to designate one tuner as the one with the IR receiver.
Implement a few encoder tweaks to better support PAL-60. Changes are from
Stop using sysfs class attribute owner fields for kernels newer than
2.6.21. This is apparently no longer needed for newer kernsl and kernel
developers are trying to get rid of this entanglement.
Fix build issue in 2.6.21; a feature had been patched in that I had not
expected until 2.6.22. Thanks to Mark Brand for spotting and fixing this.
Reorganize pvrusb2-options.h a bit for better maintainability.
Convert usage of a semaphore into a mutex (a mutex is more efficient).
This comes from a kernel janitor.
Added bus_info_str sysfs item which will show a string representation of
the USB path / address of the connected device. Modified VIDIOC_QUERYCAP
to also report this string in the bus_info structure element returned.
Some minor benign fixes for issues detected by the sparse tool (change
instances of 0 to NULL).
Pulled up a pile of minor changes from V4L, including a compat.h reorg in
the V4L repo and a changed register access API. The register access
changes are merged here in a manner that should maintain backwards
compatibility with older kernels.
Structure change in source code: Options moved from compat.h to
pvrusb2-options.h. This became necessary because the use of compat.h
within v4l has changed in a way that makes continued use of that header
Get rid of I2C configuration warning in system log. This issue appears to
have been around all along, but the warning message began appearing in
Be able to handle encoder firmware whose size is not a multiple of 8KB.
Encoder setup tweaks to get rid of video corruption during rapid channel
changes. This seems to fix the problem for both 24xxx and 29xxx devices.
Minor rework in the encoder communications protocol to exploit updated
knowledge for how the encoder works. This should hopefully put an end to
the "It is likely that this is a bad idea..." errors.
Eliminate some recently introduced compilation warnings for AMD64 (from
Mauro Carvalho Chehab <email@example.com>).
Change all FX2 command byte values to be symbolic constants, and collect
all the definitions into a header file. Clean up done by Michael Krufky
Only respond to VIDIOC_G_TUNER for tuner #0. For all other tuner indices
return -EINVAL. This is how an app figures out how many tuners there are.
Without this filter in place, an app (e.g. v4l-info) thinks there will be
an infinite number of tuners.
Implement some hackery in pvrusb2-v4l2.c in order to get mplayer working
in pvr:// mode. Credit to Pantelis Koukousoulas
<firstname.lastname@example.org> for figuring this out.
Enable reception of FM radio in stereo mode. Root cause was due to
VIDIOC_S_TUNER not being properly broadcast to all I2C chip level drivers.
Code cleanups: Use kzalloc() in place of kmalloc()+memset() wherever we
can. Also replace sizeof(x)/sizeof(x) idiom with standard ARRAY_SIZE
macro (most of these changes came from a kernel janitor).
Fix misplaced #endif which had the effect of leaving an empty control
definition in the driver. This empty definition propagated out to the
sysfs interface, which resulted in various null pointers and other kernel
badness taking place. The #endif which caused all this havoc was added as
part of the change to disable auto_mode_switch, so this bug was only
introduced in the previous snapshot.
If a VIDIOC_[S|G|TRY]_EXT_CTRLS command is issued by an app where the
number of controls passed in is zero, let the command succeed. Previously
it returned a failure.
Allow mpeg streaming from /dev/radioX, same as /dev/videoX. This change
was made on advice from the V4L maintainer. V4L radio apps generally
expect the audio to come from the sound card input via a loop-around cable
so there's no problem spitting mpeg out of the radio device.
Clean up handling of VIDIOC_G_TUNER in pvrusb2-v4l2.c. Proper frequency
ranges should be returned now for all modes. With this change, kradio
Rework internal architecture with respect to how the driver executes
status gathering from I2C chip-level drivers. This was a long-standing
piece of cruft, now gone. The driver now can report signal strength (as
opposed to just a signal present flag). Also, proper audio mode (stereo /
mono / lang1 / lang2) is reported.
Numerous minor clode cleanups.
Remove auto_mode_switch feature. Currently this is just ifdef'ed out.
This concept was not really safe (in some area FM radio band can overlap
TV band) and it got in the way of the VIDIOC_G_TUNER cleanup.
Fix missing break statement pvrusb2-v4l2.c; this bug caused bogus
VIDIOC_S_TUNER errors reported by xawtv.
When /dev/radioX is closed, set the input selection back to what it had
been previously set to, but only if it was still set to the radio.
When /dev/radioX is opened and the input control is switched to the radio,
we also have to commit this in order for the control change to take
Clean up various hacks in pvrusb2-v4l2.c that are conditionalized on the
v4l type being the radio. These hacks are needed to prevent streaming of
the radio device node. Better solution: Make the radio's device node
stream type be pcm, then only allow streaming on ANY device node which has
a valid stream pointer inside of it. Since pcm is not implemented at the
moment, then the stream pointer will be NULL and thus everything else will
work itself out without needing any special hacks.
Clean up how pvr2_config is dealt with; this enum really describes the
type of stream not the type of v4l device node. So we use normal
VFL_TYPE_xxxx symbols in pvrusb2-v4l2.c to track information about the
device node and then let _that_ type drive the actual stream type. Along
the way, change the pvr2_config enumeration so it more closely maps the
expected stream types we think we can support.
Fix cut/paste bug involving auto_mode_switch control.
Implement various bits of V4L1 functionality so that radio mode hopefully
even works with older kernels. This is _untested_.
Clean up V4L2 frequency set / get operations in pvrusb2-v4l2.c to account
for radio mode.
Implement ability to execute a television <-> radio mode switch
based on the frequency being set. This behavior is controlled by a flag
and is disabled by default. See the "auto_mode_switch" control
and the "auto_mode_switch" module parameter.
Implement intelligent handling of the current frequency, by storing two
values - one for television mode and one for radio mode. When switching
modes, the frequency is transparently switched at the same time. Thus we
should always have a valid frequency set.
Implement radio support for 24xxx devices. (Pantelis was only able to
test on a 29xxx device.)
Implement radio mode. Many thanks to Pantelis Koukousoulas
<email@example.com> for implementing the key pieces of this
puzzle. To use the radio, just change the input selection to
Change default volume setting from 65536 to 62000. I can't re-scale the
volume unfortunately because at least one app (MythTV) implies this
scaling already. So if I scale it down (i.e. make 65536 correspond to a
reasonable maximum) then MythTV's audio will be out of whack.
Some minor tweaks / cleanups committed by kernel janitors.
Minor fixes to permit compilation in kernel 2.6.20.
Eliminate 3:2 pulldown API usage; Hans reports this function is a nop in
Fix long-standing issue which prevented use of 24xxx devices at any
horizontal resolution other an 720. The problem is in the cx25840 module,
if no attempt is made to initialize VBI there. Such a thing should not be
needed, but the pvrusb2 driver is updated now to deal with this.
Minor changes to permit clean compilation under 2.6.19
Refactored a bunch of stuff in decode_log.cpp to permit more flexible
tracking of different types of I2C chips.
Implement IR reception for 24xxx series devices. This is implemented by
emulating the presence of the IR receiver that 29xxx devices contain. The
emulation is very low overhead and it is only in the execution path when a
24xxx device is being controlled. The net effect of this is that IR for
24xxx devices externally now works *exactly* the same way as it has all
along for 29xxx devices.
Update fwextract.pl to work with latest Hauppauge driver package. (This
does not affect the driver itself.)
Fix some nasty bugs in decode_log utility that caused SEGV's and bad
statistics. (This does not affect the driver itself.)
Fixes for VIDIOC_INT_S_REGISTER and VIDIOC_INT_G_REGISTER commands - it
actually works properly now.
Rework how driver context pointers are handled in pvrusb2-v4l2.c. This
fixes a circular locking problem introduced when I have to get rid of any
use of video_set_drvdata() and video_get_drvdata().
Implement VIDIOC_INT_S_REGISTER and VIDIOC_INT_G_REGISTER commands. This
is a debugging aid.
Change resolution range limits. Minimum vertical and horizontal
resolutions now are set to empirically tested values. Maximum horizontal
resolution is 720. Maximum vertical resolution is set based on video
standard in use (480 or 576). Note that 24XXX specific horizontal hack is
still in place.
Fix flipped handling of vertical / horizontal resolution limits in the
implementation of VIDIOC_S_FMT. (This gets filed into the category of
"I can't believe this bug survived this long".)
Eliminate all internal uses of CONFIG_VIDEO_PVRUSB2_24XXX. We always turn
that on for standalone builds and now the V4L build no longer requires it.
So it goes away...
Correctly handle failures to create various parts of the sysfs interface.
Hack the horizontal resolution setting control to not allow anything
except 720 if we're dealing with a 24xxx device. This is a workaround for
an unsolved bug involving capture resolution on the 24xxx device type. It
should reduce the pain a bit for apps (e.g. MythTV) which try to use a
different default value (e.g. 480).
Support PAL-M in places where NTSC-M is supported by the hardware.
Prioritize NTSC if NTSC is available along with PAL and/or SECAM. This is
needed so that the driver doesn't default-choose the less common PAL-M
standard over NTSC-M.
Always printk device registration / unregistration messages to the log
regardless of debug mask setting.
Stop using tda9887 module when under kernel 2.6.18 or later (module goes
Sanitize code against Linus' sparse tool. Implement tiny hack in Makefile
to enable automatic sparse checking of the driver: If you pass SPARSE=y on
the make command line then sparse will be used. Note: You must have
sparse installed of course for this to work.
Fix bug where changes to encoding parameters were not having any effect
until the stream was restarted. This problem had been introduced when
cx2341x was integrated.
Fix mixed up audio sampling rate setting values; problem was introduced
when cx2341x was integrated and only manifests itself when using cx2341x.
Harmonize the names and functions of all the various encoding controls
when not using cx2341x to match behavior when cx2341x is in use. This
allows for uniformity of sysfs controls and behavior for older kernels
where cx2341x would not otherwise be available.
Turn off lots of debug output (it can still be turned on later).
Fix build problem involving cx2341x.h
Cut back on the amount of noise that goes into the system log.
Don't stop / start the encoder when changing encoder attributes. Leaving
the stream going doesn't seem to have any ill effects and the stop / start
action seems to cause other issues.
Fix a misbehavior during encoder error recovery - under certain scenarios,
the firmware wasn't being reloaded when the recovery procedure was
signaling for it.
Support more efficient method for V4L2 apps that are trying to discover
what controls are implemented. The driver can be told to find the
"next" control id given a previous id. This is better than a
brute force scan.
Correctly handle V4L flags attributes for all V4L controls.
Defined new internal boolean control type. Implemented new type sysfs
member for unambiguously discovering the type of a control. Converted a
number of controls to this type.
Implement new V4L extended controls (should appear in 2.6.18).
Implement new logic to take advantage of new cx2341x module in V4L. This
is enabled at compile time through the use of the PVR2_ENABLE_CX2341XMOD
Remove longstanding cruft with behavior of VIDIOC_[S|G|TRY]_FORMAT
commands. Just set the resolution requested and cleanly return what's
been set. The business with interlace has been a broken piece of junk for
far too long.
Make possible the compilation of the standalone driver against a V4L
source tree snapshot, by passing V4LDIR=path_to_v4l_tree on the make
command line. This is something you don't normally need to do (especially
since there's a pvrusb2 driver already in the v4l tree), but I implemented
this to help with my own debugging of the stanalone driver under certain
Conditionally change how v4l2 context tracking is done, in preparation for
upcoming changes in V4L API (likely to be present starting with 2.6.18
Arrange things so that pvrusb2-audio-old is optionally built and then
disable it for any kernel 2.6.17 or later. This module implemented an old
obsolete interface for communicating with msp3400. For recent kernels it
is useless anyway, and starting with 2.6.17, it stopped building correctly
due to a missing structure (which had disappeared from the kernel as part
of the removal of the obsolete interface).
Fix compilation error involving how cx2341x.h is used when building
Fix composite and s-video input for 24xxx devices when running under the
2.6.15 kernel. Previously audio wasn't working at all and only composite
video worked. Note that this only affected 24xxx hardware.
Use cx23416 firmware file name from cx2341x.h (this doesn't change
Include config.h where potentially needed (this has no immediate impact
but becomes important for future in-kernel builds).
ifdef-out the definition of msp_matrix in the ivtv-resident version of
msp3400.h when the kernel is 2.6.16 or newer. This header gets included
by the pvrusb2 driver when we aren't doing "new routing" (which
is the case for kernels 2.6.16 or older). However specifically in 2.6.16,
msp_matrix is already being defined in v4l2-common.h, so we don't need or
want it here. In my own debugging I never saw this problem because I have
been using a kernel source tree which has an updated msp3400.h within it
which was overriding the use of this header. Distro-generated kernel
source trees tend not to include this header because it is in an unusual
spot - and that is in fact the reason why I dropped that msp3400.h into
the ivtv directory in the first place...
Update wm8775 support in the driver to use new routing API that is
currently in the V4L repository and will be in the next kernel release
Work around cx25840 bad behavior where cx25840 will stomp on our internal
state. The problem is that cx25840 is trying to reach through an I2C
adapter-private context pointer in an attempt to manipulate the driver.
The problem however is that it is assuming a specific type of I2C adapter
driver which is incorrect in the case of pvrusb2. The cx25840 module
really has no business trying to touch the "private parts" of
another driver. This is being fixed in the main V4L repo, but for older
kernels (2.6.15 and probably also 2.6.16) we still need to do a
workaround. What this will do is pad the front of the pvr2_hdw structure
so that what cx25840 messes with won't actually harm anything.
Add V4L2_STD_NTSC_M_KR to pvrusb2-std.c.
B/G and D/K video standard combinations are "special". We need
to allow for selection of B+G or D+K together because they can each be
used at the same time. When that happens "B" (or "D")
are for VHF reception while "G" (or "K") are used for
UHF reception. Modify video standard enumeration generation so that these
combinations are also generated as enumerated values if their constituent
components are present.
Fix off-by-1 problem with V4L video standards enumeration API.
Fix problems involving the return value not being initialized in the V4L2
ioctl() implementation. This bug resulted in none of the controls (e.g.
volume) being visible to a V4L app.
Adjust build configuration such that setting a CONFIG option to 'n'
produces the desired result.
Make logic that maps additional supported video standards be table-driven
(previously it was just open-coded). Add a table entry to enable PAL-N if
any of PAL-B/G/D/K/I are reported by tveeprom.
Note: Model 24xxx series hardware should work considerably better with
this driver version.
Move change history out of the main pvrusb2.html file and into
change_history.txt, transformed into a simpler format. A perl script is
used now to generate an html version of the same data, to which
pvrusb2.html is linked now. The main page was getting too long, mainly
due to the change history.
Make it possible to not have to compile in sysfs support or the debug
interface. These choices will become kernel CONFIG variables.
Make support for new 24xxx hardware a configuration option (normally
Eliminate translation of control ids in the V4L interface.
Internal "controls" logic reworked. Ability to handle bit masks
implemented in sysfs interface. This touches lots of code in the driver -
nearly everything. Controls handling should be cleaner now, though
somewhat more verbose than before.
Completely new logic to deal with handling various video standards. This
should make possible support of any V4L defined video standard provided
that the hardware can handle it. Also, the driver now will only advertise
video standards that it knows the hardware can support. (Previously the
driver was working with a hardcoded subset regardless of hardware.)
New video_std module option, which is a per-device instance array. If
this is non-zero, it is interpreted to be a v4l2_std_id bit position and
will be the default video standard to use. Note: If the chosen standard
falls outside the list of eeprom-reported supported standards, then the
supported list is expanded to include this chosen value.
Modify output of eeprom video standard query a bit so that if NTSC-M is
reported, also make NTSC-Mj available. The white sticker says that these
devices are also NTSC-Mj capable, yet tveeprom appears not to be reporting
Change cx23416 SET_OUTPUT_PORT command to send 0x00 as the second argument
instead of 0x02. This undocumented behavior seems to still let things
work, and now xawtv is happy again. A lengthy comment about this is
written up in pvrusb2-encoder.c.
Include linux/mutex.h not asm/mutex.h everywhere. Fixes compilation
problems on amd64. Mutexes were introduced in the 2.6.16 kernel; so this
change is only relevant when compiling for kernel 2.6.16 or later.
Implement new I2C transfer filter framework, which allows the driver to
take special actions on behalf of an I2C chip to which a transfer is being
Implement I2C filter for wm8775. On new (24xxx) hardware this filter is
activated. The only special thing it does is return success on probe
attempts; all other operations pass through. This should nicely work
around the fact that we can't autodetect the wm8775 chip.
Implement I2C filter for cx25840. On new (24xxx) hardware this filter is
activated. Things are a little complex here. Basically this filter
watches for a specific attempt to read either revision byte out of the
chip. Until that successfully happens, all other I2C accesses are blocked
(and probe attempts will silently return success). Once the revision
register is read, the returned byte is snooped for one of two bad values.
If either bad value appears, the chip is considered wedged and the driver
is rendered useless (along with some warning messages to the log
explaining what has happened). Otherwise everything is good to go and
then the filter will unlink itself, and all further transfers to the part
will go straight through.
Modify request_module() behavior - only request modules to be loaded that
are relevant for the model of device that we are dealing with. The
previous approach technically worked, but it was an elephant gun strategy.
Now we take aim and only load what we really need.
Enhance I2C filter for cx25840 to detect attempts to send firmware data to
the chip where the chunk size is too large. When this situation is
detected, break the data into smaller bite-sized chunks and perform the
load operation. This effectively eliminates the need for the FWSEND patch
in the in-kernel version of cx25840 for kernels older than 2.6.17.
Stop calling pvr2_reset_ctl_endpoints() during initialization. This is an
unsafe operation because it can stomp on a command that might be underway
with the hardware (going on in a different thread). While we can make
this thread safe, there's really no reason to even keep this around. This
logic is very old code, and it's been lurking undetected in the driver for
over a year. It's time to just get rid of this step in the
initialization. This fixes a recently discovered kernel oops problem (and
is the reason for this snapshot).
Note: While this driver version includes even more changes to support
operation of the newer 24xxx model series, we're still not quite there
Don't start the I2C adapter until after the powerup command has been sent.
It seems that some I2C chips won't respond until after this command has
been issued to the FX2.
Implement mechanism to issue a reset to the decoder module.
New logic to control cx25840 module.
New logic to control wm8775 module.
Adjust encoder module to properly operate new 24xxx model series hardware.
Integrate new input routing API from V4L; this is needed when operating
modules from current V4L repository sources or when in any kernel 2.6.17
Return control's default value in addition to min & max when a V4L app
queries about a control.
Fix compilation error involving pvrusb2-audio-old with older kernels.
Note: While this driver version includes a number of changes to support
operation of the newer 24xxx model series, we're still not quite there
Don't hardcode eeprom I2C address anymore in the driver. Query the
controller for the right address. This is needed because newer PVR USB2
devices appear to be using a different eeprom now.
Implemented VIDIOC_LOG_STATUS out to all V4L2 modules. This may be
directly triggered via a V4L ioctl, or by cat'ing debuginfo in sysfs.
Modify I2C adapter implementation slightly so it will look appetizing to
cx25840.ko - which is needed for newer tuner devices.
Rename encoder firmware to v4l-cx2341x-enc.fw and controller firmware to
v4l-fx2-pvrusb2.fw. Driver will attempt to load the new names first but
fall back to the old names if a given new name could not be located.
System log messages have been improved significantly as well so that the
actual firmware file loaded is logged and if there is a failure loading
firmware, the name(s) tried will also be logged.
Enforce a configurable pause (currently zero) after forcing a hardware
reset during initialization. The delay is measured in msec and configured
through a new module option init_pause_msec. This can potentially be
needed to deal with a race where we might try to perform I2C operations to
the device before it is ready (but at the moment it appears not to be a
New decode_eeprom program in utils.
New detection method implemented to discover if the FX2 firmware needs to
be loaded. This method should work with the newer PVR USB2 hardware.
Renamed FX2 firmware to v4l-pvrusb2-29xxx-01.fw, since we know that newer
hardware devices require different firmware. Damnit.
Adjust legal upper bound on tuned frequency to 850MHz. Some areas need to
tune that high.
New logic to keep track of specific hardware type of device. This is
largely a hack right now, but it is needed immediately. With that logic in
place, we now adapt FX2 firmware name to track with hardware device type.
Minor textual cosmetic fixes to conform with V4L standards.
Copy pvrusb2-audio to pvrusb2-audio-old, update pvrusb2-audio to V4L2
standard. This should improve compatibility with various msp3400 versions
floating around too.
Rework pvrusb2-ioread.c to implement a synchronization filter that will
discard all data at the front of the stream until a specifically given key
sequence is seen. The key sequence is programmed in pvrusb2-context.c,
and is set to be the beginning of a legitimate mpeg2 frame. This allows
the driver to discard garbage at the front of the streaming buffer that
the device appears to be emitting. With model 29xxx this problem was
benign but with model 24xxx device this problem became malignant due to
the garbage data having a legitimnate packet header (followed by more
garbage). This fix is for both models of device, but it is critical for
model 24xxx devices.
Fix minor bug involving incorrect audio mode switching (but the supporting
module is still busted so this doesn't really fix anything yet).
Use mutexes for 2.6.16 or later. Apply same hackish trick as was done in
the V4L repository (by defining the new API functions to expand to the old
semaphore code if compiling for earlier than 2.6.16).
Hack up ivtv/tveeprom.c a bit such that its function prototype will
hopefully now match the corresponding prototype in the kernel. This
should in theory put an end to the symbol version mismatch errors that
people encounter when using the snapshot's tveeprom.ko in a kernel with
CONFIG_MODVERSIONS enabled. Note however that this change is really only
useful for older kernels; for kernels 2.6.15 or later just use what is in
the kernel and ignore the modules here.
Update decode_log.cpp in utils to address a SEGV bug, and also account for
more recent knowledge about the known device commands.
New pvrui.sh utility; implements a dialog-driven front end that uses
sysfs. Known to work for Debian, unsure about other distributions (needs
the dialog program which exists in Debian).
Changed sense of various configuration options in compat.h so that things
are not as confusing as before.
Define more configuration settings to improve compilability within V4L CVS
when using old kernel versions.
Use cx2341x.h from V4L where possible for definition of encoder commands.
Use v4l_print_ioctl() wherever possible.
Change global "debug" variable to namespace-cleaner
Various changes / fixes to permit successful compilation in kernel
Detect tda9887 module also using its driver ID if available.
Detect tda9887 module by searching for substring "tda9887" in
its name if available.
Improve information reported about detected / driven support modules.
Implemented the ability to optionally tolerate a certain level of
streaming errors. To activate, you must use the new "tolerance"
module parameter (see the module parameters description on the main
pvrusb2 page for more information on how this parameter works). A debug
mask bit has been defined to track ignored errors if they happen. The
implementation of this was inspired by some work done by Ingo
Bit mask definitions for debug parameter and subsystem definitions changed
to be shift expressions instead of raw values (purely cosmetic).
Driver now detects if a firmware image load fails due to file(s) being
missing and will warn the user of this fact in the log. Also solved of
problem of the misleading log message about firmware having been reloaded
when in fact the load had actually failed.
Removed some module parameters that weren't actually being used (piece of
bit rot from the distant past).
Fix compilation problem when building under 2.6.15 (bug introduced in
Fixed compilation problem for kernel versions older than 2.6.13 (thanks
for Michael Roitzsch for finding and fixing this).
New debug code added to log I2C transactions (off by default).
Fixed problem where side effects (specifically regarding the vertical
capture resolution) were not happening correctly when the video standard
is first initialized. (This problem had been around all along and only
came to the surface with the recent PAL B/G vs SECAM change.)
Minor configuration change to maintain better compatibility with V4L CVS
Fixed initialization code for saa711x which didn't always initialize all
the properties of the driver. (This was the same issue that was already
fixed for saa7115.) Note: You are only affected by this if you are
running under 2.6.15 and using saa711x instead of saa7115.
Minor tweak to tveeprom.c to remove the inclusion of asm/segment.h, a
header that we don't seem to need and which causes problems with other
architectures. (Yes this violates my policy of exactly tracking ivtv
however ivtv is being merged into V4L now and it already isn't supplying
tveeprom.c anymore so there's nothing left to track.)
Fixed initialization code for saa7115 and audio which didn't always
initialize all the properties of the driver. This was a simple stupid
thing involving incorrect prioritization of elements in a bit shifting
expression. Credit to Per Christian Henden for hitting it first and
helping with the bug chase and solution.
Adjust use of V4L header files in several modules in order to stay
compatible with upcoming V4L changes for 2.6.15 and onward.
Fixed compilation breakage in saa711x support, introduced by previous
large change. This only affects builds under 2.6.15 (where saa711x might
be present). Also fixed a few cosmetic issues in the code.
Note: This snapshot implements a considerable amount of new logic
surrounding operation of the support modules. There might be some new
instabilities introduced here. If you have new problems with this
snapshot, please post on the pvrusb2 list. (So far however it looks
Prefer PAL-B/G to SECAM-L. NOTE! If you have a multistandard tuner and
you are in SECAM territory, you will now have to reset the video standard
(either through sysfs or the V4L app you are using). This is a change in
behavior from earlier snapshots; the reasoning is that PAL as a standard
is far more common and thus it is the more likely correct choice. For
folks in SECAM-land, sorry about this.
Updated decode_log to recognize tda9887 transactions.
Completely new I2C module control system in place. This version is
capable of broadcasting to multiple I2C driver modules, while still
selectively sending specific commands where it makes sense. In addition,
logic is rearranged to contain the actual "configuration"
choices for the I2C modules into one small area which can be easily edited
/ replaced as module APIs change. With this change, the pvrusb2 driver
should do a much better job of keeping the support modules in sync with
the overall state of the driver. (In particular, you should now be able
to actually rmmod / insmod modules on the fly while the driver is running
and the driver will re-update each module when it (re)appears.)
Automatically command tda9887 to set port2=0 and port1=0 when the tuner
type is PHILIPS_FM1216ME_MK3 or PHILIPS_FM1236_MK3. This hopefully should
remove any need for manually loading tda9887 with port2=0 as a module
Fix busted i2c scan logic - this has been broken for quite some time and
had gone unnoticed until recently.
Modify debuginfo sysfs variable such that a list of attached I2C modules
and their configuration state (as viewed by pvrusb2) is dumped as part of
Fix error in tracking V4L_minor_number - the driver was erasing this when
the V4L device was closed (something that happens routinely) rather than
when it was removed (the actual point when the minor number became
Remove hackish I2C code in pvrusb2-tuner.c which had been issuing
"magic" commands to the device that we didn't understand. It's
clear now that this is what tda9887 is supposed to be doing; now we must
just stay out of its way.
Fix fatal typo in fwextract.pl (stupid quoting error when I added the new
Use NTSC_M instead of NTSC when communicating with support modules, needed
for more recent v4l tuner.ko module (and should be harmless for other
New firmware signature added (credit to Fedor Bezrukov).
Support PAL-M (credit to Mauro Carvalho Chehab)
Fix uninitialized data problem when setting tuner frequency. This problem
surfaced when the v4l2_frequency struct in v4l added a new field that we
weren't setting. Now we memset() the whole structure to zero first, so
that fields we don't care about will take on a default value.
Support use of tda9887 module where appropriate. This is needed for newer
versions of the PVR USB2 device which contains a different (type 103)
tuner. Changes are harmless for older devices. Also it appears that
tda9887 may be merged into v4l's tuner core; if that happens this change
will be obsolete, but it will also be harmless (but you'll need it for
running at least up to 2.6.15 since in those cases tda9887 still exists).
New global configuration switches in compat.h. This is a special header.
It is set up to flip switches various ways appropriate to the build
environment. In particular, it is set up so that when the header is not
present, the settings will be appropriate for compilation as part of v4l
itself (in the v4l environment this header is replaced with something
else). The idea here is to ease the transition into v4l.
Reworked pvrusb2-video so that actual choice can be made at run-time
depending on which type of driver is found. Arranged things so that newer
driver (which is now pvrusb2-video-v4l) is only compiled with 2.6.15 or
later. Older driver renamed from pvrusb2-video-old to pvrusb2-video-ivtv.
Restructured pvrusb2-eeprom so that it can play nicer within v4l (by using
Note that this snapshot does not implement ANY new features. The changes
here are all code reorganization and adaptations in preparation for
possible inclusion into v4l.
Remove download links from pvrusb2.html and place into a separate file,
Move driver version info into a new file pvrusb2-version.h.
Split pvrusb2-video.c into two versions, pvrusb2-video.c and
pvrusb2-video-old.c. The file pvrusb2-video-old.c has the original
ivtv-compatible saa7115 glue logic, while pvrusb2-video.c is a rewritten
version designed to operate with the new saa7115 module that has just been
placed into v4l (and should show up in the 2.6.15 kernel).
New script in the utilities area, release_snapshot.pl. I plan on use this
to automate the snapshot process better. With the appearance of v4l,
things are too complex to continue doing manually. This script automates
tagging and tarring of the source tree, creating pvrusb2-version.h, and it
can create pvrusb2-download.html as well.
Minor tweak made to Makefile so that KDIR or KREL set in the environment
will be properly noticed.
Modify the build slightly so that msp3400.h can be pulled from the ivtv
subdirectory of the driver snapshot if needed by pvrusb2-audio.c.
Normally that header can be found in the kernel tree under
drivers/media/video but that subtree apparently is not always fully there
in some distributions' kernel development package.
Note that this snapshot does not implement ANY new features. The changes
here are all code reorganization in preparation for possible inclusion
Source tree reorganized to physically separate the ivtv-provided parts
from the rest of the driver. Build files have been adjusted accordingly
and the documentation has been updated to account for the two sibling
build areas that now exist. This is a precursor towards subsequently
jettisoning the ivtv parts and finally merging the driver itself into the
v4l source tree.
pvrusb2-main.[ch] renamed to pvrusb2-context.[ch], to better reflect what
it is really for. The pvr2_main struct exported by this module has also
been renamed everywhere to pvr2_context.
pvrusb2-msp34x5.[ch] renamed to pvrusb2-audio.[ch], to avoid confusion
from others examining the source tree.
pvrusb2-saa7115.[ch] renamed to pvrusb2-video.[ch], to avoid confusion
from others examining the source tree.
pvrusb2-hdw-low and pvrusb2-firmware merged into pvrusb2-hdw. The entire
semi-public hardware interface (i.e. that which is used by the various
high level interfaces) is now in pvrusb2-hdw.h.
driver/README expanded considerably to document the overall function(s) of
each source file.
Tested each control in sysfs and documented the resulting effect or
breakage (see doc/controls.txt). Updated sysfs description in main web
page to point out this new information.
Fixed the setting negative values for bass, treble, balance, hue in sysfs.
This was due to the fact that the number parser in pvrusb2-sysfs.c didn't
bother noticing or dealing with the '-' sign prefix.
Noticed broken bass, treble, balance and audio mode settings. Problem
can't be fixed in pvrusb2 itself. FAQ items added to explain this.
Updated to modules from ivtv 0.4.0. Implemented a hack in
pvrusb2-saa7115.c to permit detection and operation with both older
saa7115.ko and newer saa7115.ko modules (a critical command code had
A lot of random changes to support compilation and operation in kernel
2.6.14. (Successfully tested for continued correct operation with older
kernels back to 126.96.36.199.)
Adopt tveeprom as an attached i2c client if we can. This is needed if
pvrusb2-eeprom.c is configured to use the indirect method of tveeprom.c to
read PROM contents (normally this is not the case).
Rework pvrusb2-eeprom.c so that it can continue to operate correctly with
either tveeprom provided by either ivtv (copied into this driver) or v4l.
Large comment added into pvrusb2-eeprom.c explaining the issues and
choices here. New HTML file (pvrusb2-eeprom.html) created that explains
all of this, and references to it added in the FAQ and in the main web
Modify sysfs code to use a unit id in the class path if there is no serial
number available (which can happen if there is trouble reading the device
Fixed kernel oops caused by linked list foul-up in pvrusb2-io.c. This was
a rare hard to reproduce race condition, which has apparently been present
all along but only recently noticed. When shutting down a stream, pending
URBs are flushed, but if every once in a long while one completes during
the flushing operation and the flush loop happened to hold a pointer to
that buffer at the same instant, then the flush loop will leak over to the
wrong list, resulting a wild pointer dereference upon exhausting that
wrong list. (Long story.) The flush function has been rewritten to no
longer be sensitive to buffers completing at the same time as when
flushing takes place.
Fix utils/decode_log.cpp so that gcc 4.x can compile it - a silly stupid
problem involving a forward class declaration implied by being named in a
friend declaration that no longer appears to work in gcc 4.x - probably
because the previous behavior wasn't very clean. Duh.
Included contributed channel changing script (utils/change-channel.sh)
from Per Christian Henden.
Fixed kernel oops when cable was yanked while video was streaming. The
driver internally doesn't tear everything down inside the USB core's
disconnect hook due to the need to wait for other references (e.g. open
file handles) to go away first. When it finally does tear things down, it
was also disconnecting at that time its association(s) with the USB core.
But that's too late, and doing that can result in dangling pointers being
dereferenced from within the core. This is not a USB problem; the root
cause is that the driver must remove itself from the USB core before the
disconnect hook returns. Now the driver disassociates from USB
immediately on device disconnect, while still delaying its own internal
tear-down until other references go away, as usual.
Fixed memory leak in pvrusb2-io.c - buffer control structures were never
being freed (happens only when streaming is stopped). Also cleaned up a
potential dangling pointer (which actually should never have been touched
anyway, but I'm paranoid).
Updated misc/lircd.conf to match the "Hauppauge grey" remote
more typically included with the PVR USB2 hardware. (Some corresponding
tweaks to misc/lircrc.) This is a conf file I've been using with a
PVR-250 (it's the same remote).
Added sanity / debug code to buffer management, in an attempt to trap any
possibly corrupted buffer control structures. This is turned off by
default right now. To enable, turn on the SANITY_CHECK_BUFFERS symbol
pvrusb2-io.c. When on, there will be some additional noise during stream
start / stop but otherwise no additional log messages will appear unless a
bad buffer has been spotted.
Got rid of dead pvrusb2-data.h. Cleaned up pvrusb2.h (removed remaining
old junk which had long since stopped being used).
Implemented unit number concept which is used to uniquely name driver /
device instances. Other parts of the driver will use the unit number to
select per-unit module options (like tuner override, minor device number,
Implemented tuner=x module option to allow overriding of tuner type. This
is an array, so the tuner type can be individually overridden for each
pvrusb2 unit you might have (yeah, as if that will ever actually be
Changed video_nr module option from a scalar to an array. Similar to the
tuner option, this can be specified on a per-instance basis.
The v4l minor device number is published now in sysfs as v4l_minor_number.
The unit number is published now in sysfs as unit_number.
Updated ivtv-sourced modules to those from ivtv 0.3.7i.
Fixed pvrusb2-tuner.c so that driver works under 2.6.13 (verified with
Fixed longstanding and nasty bug which prevented tuning any frequency
below 76.00MHz (roughly below US broadcast channel 5). It's amazing that
something like this went undetected for so long. Cause: brain-fart while
reading the FM1236 datasheet back in February (read off the floor
frequency for FM instead of VHF). Thanks go to a pvrusb2 user for
tracking this one down.
Minor clarifications / updates to web page info based on feedback from
New firmware signature added to fwextract.pl (corresponding to driver
2.5.22329 found on a CDROM distribution).
Modified fwextract.pl to exit with a non-zero status if extraction is not
Contributed script fwfind.sh added to utilities and described. Firmware
extraction procedure reworked to include fwfind.sh, and extra information
added about where to place firmware files after extraction.
Adjusted log messages slightly which are related to firmware upload. We
don't want to mislead the user if the upload had not succeeded.
Various tweaks / improvements to firmware extraction script.
Teach extraction script how to pull firmware from 3 additional Hauppauge
New firmware extraction procedure, along with new tools and a driver
feature to support it.
Moved all files into Subversion. Rearranged distribution layout; there
are now separate utils, doc, and driver subdirectories. Utilities now
included as part of driver snapshot.
Added pvrusb2-utils.html web page, which describes detailed firmware
extraction info, along with how to use the decompiler.
Detect when encoder stops responding to commands and return an error code
to higher level code in the driver. We don't know why this can happen,
but at least now we can detect when it happens...
Rewrote low level request / response logic (pvr2_send_request()). New
version can issue simultaneous write / read pairs, and eliminates some of
the hacks that previously existed (and from what I can tell over 5 months
of working on this driver that these hacks were completely unneeded).
Defined encoder firmware reloading as a callable configuration step.
Reworked device configuration function (pvr2_hdw_subsys_bit_chg_no_lock())
to react to failures and retry appropriate pieces as needed.
Specifically, we can recover from a hung encoder now by causing an encoder
firmware reload (and reconfiguration of all affected pieces).
Reworked all code involved in encoder register read/write to use
little-endian ordering instead of big-endian. (This wasn't a bug, but now
we match what ivtv is doing which makes it easier to analyze other
Fixed error in pvrusb2-encoder.c which caused too many words to be read
back from encoder (was 32, should have been 16).
Always send 16 words to the encoder for every request instead of 4 plus
the number of arguments.
Fixed bug in pvrusb2-encoder.c which caused it to misbehave if more than 8
words were being sent to the encoder.
Implemented "reset firmware" command in debug interface, which
can be used to force an encoder firmware reload.
Added lots of new stuff to protocol-info.txt, all regarding operation of
the encoder chip. (See a running theme here yet for this snapshot?...)
Implemented functions in pvrusb2-hdw-low.c for direct manipulation of GPIO
Discovered that bit position d7 of the GPIO direction(?) register appears
to control the LED. Confirmed that flipping this bit does not seem to
impact video streaming, so now we ensure that the LED is on when the
encoder is running and off when not the encoder is stopped. (We really
need to understand these GPIO bits a lot better.)
After auditing the encoder configuration command sequence against the same
sequence snooped from the Windows driver, one unexplained difference was
found. Changing this driver to match the Windows driver seems to have
cleared up video corruption problems. Because of this, streaming control
now toggles the USB link again and also now reconfigures the encoder when
the encoder is started / stopped.
Additional debug abilities added to help debug driver lockups.
Updated ivtv-sourced modules to those from ivtv 0.3.6n.
All pointers are formatted with %p now instead of my usual pattern of
hacking, I mean uh casting the value into an unsigned int and printing
with %x. Using %p is more portable across machine architectures, and the
previous method caused a flood of warnings when compiled on AMD64. A few
related AMD64 compilation warning were also fixed.
Internal state handling code has been reworked so that we can track when
we need to configure various parts ("subsystems") of the device
and know what parts of the stream pipeline are running. This is all in an
effort to better implement stream start / stop, which so far has been very
unstable. It's now possible to easily configure what gets manipulated
when streaming is started and stopped.
A new command parsing debug interface has been implemented, currently
accessible through the sysfs interface under the name debugifc. cat'ing
out this file will dump a human-readable report of the driver's internal
state (which subsystems are running or configured, whether streaming is in
progress, speed of USB link). Commands can be echoed into it to adjust
driver operation and / or (attempt to) recover from strange errors. This
is not meant for routine access; it is purely a debugging aid.
Identified the true purpose of a number of previously mysterious commands
sent over USB and abstracted them to separate functions. Numerous ways to
reset internal device state have been uncovered, and a means for querying
the device's view of the link speed is available now (see ctl_usb_speed in
the sysfs interface). The speed query function is also a useful way to
ping the device - it's a single command & response requiring virtually
nothing else in order to work. So if you can read the USB speed value,
then you can be assured that communication is working between the device
and the driver.
Reverse-engineered the actual underlying encoder communication protocol
and reworked pvrusb2-encoder.c to expose and exploit that. This makes our
operation of the encoder somewhat more honest than before (unfortunately
this work didn't actually fix any known problems). Interestingly enough,
our communication with the encoder bears a spooky resemblence to the
mailbox communication used within the ivtv driver. Coincidence? I think
Adjusted driver's streaming start / stop function to no longer shut down
the USB link portion of the video streaming pipeline when streaming is
stopped. Numerous experiments seem to show that leaving the USB link
running at all times seems to avoid the rampant accumulating video
corruption problems that I and many others have observed (at least in the
tests I've done with mplayer).
Added a text file to the driver sources protocol-info.txt which documents
the control protocol that the device expects. This is of course far from
complete, but I have spent far too long trying to fill in this picture
already and it doesn't do any good if the knowledge is kept only in my
Ensured that "make install" does the right thing (this bug had
been introduced in the 20050527 snapshot).
Fixed a severe bug in the frequency table handling that essentially made
the frequency table useless. This problem had been introduced a few
Combed through pvrusb2-encoder.c and symbolically name all of the commands
being sent to the encoder chip. The information source for this was the
ivtv driver; this is a minor steps towards trying to understand how the
encoder works. No functional change here, not yet at least.
Cleaned up and regularized copyright attributions and CVS Id keywords
Tweaked sysfs implementation a bit - if a control file is written with
multiple value now, each value will be set in rapid sequence. This is a
mostly useless feature, but it is good for debugging scenarios where one
wants to rapidly toggle something on / off (like chasing the streaming
Eliminated race condition that was possible during a disconnect or
permanent failure where somebody might be attempting a control transaction
at the same time. The logic which does the disconnect or marks the
permanent failure now takes the same semaphore as the logic which performs
transactions. This is needed because I2C chip drivers might try to
operate autonomously with respect to this driver.
Cleaned up build process; we no longer have to specify KREL.
Kbuild-specific stuff is now moved to its own file. The resulting
Makefile is significantly shorter. Added lots of comments.
Turned on a debug log message ("ZERO Request? Returning zero.")
that will report when somebody attempts a read of length zero. The driver
responds to such a read by returning zero, but unfortunately xawtv then
treats that as an EOF and closes the channel. This is an xawtv problem;
the log statement is there so it's possible to detect when it happens.
Fixed a bunch of brain-damage involving the file permissions mask for all
the module parameters. Now all the permissions are reasonable.
Seriously improved error handling in the streaming logic. If the cable is
yanked while streaming video, the device will now cause the stream to fail
gracefully. This should cause the app to close the channel, after which
point the driver should deactivate itself.
Found and fixed a nasty race condition that can happen if the cable is
yanked while it is being initialized.
Implemented a set of informational summary log messages that print at the
end of device initialization which explain what state the device is
probably in. If it detects that the device is jammed, a message will
print suggesting that the device be power-cycled. It will also print a
message if it is allowed to reload the firmware and decide to do so. If
initialization is successful, a message to that affect will also be
printed. If initialization is incomplete due to a firmware reload, an
advisory message will print that a reconnect is to be expected.
Implemented experimental code to force a new firmware download if the
device fails to initialize properly. Unfortunately this doesn't seem to
help. This is controlled by the "procreload" module parameter.
For now I've left it off by default.
Perform a device reset during set up (after primary firmware download).
Doing this allows recovery from some inconsistent device states. This
step is controlled by the "initusbreset" module parameter. It
is on by default.
Fix call to I2C adapter tear-down function so that it is called every
time, not just when the device has been successfully initialized. Why?
Because we might have initialized the adapter anyway; the overall
initialization could have failed after adapter initialization, thus
without this fix we could leave the kernel's I2C core with dangling
pointers - which would eventually end badly with a kernel oops.
Fix the driver version info to match the snapshot. D'oh!!
The old streaming logic has been completely ripped out and replaced. This
is by far the biggest change for this snapshot. Previously logic was in
pvrusb2-buffer.c and pvrusb2-stream.c. New logic is pvrusb2-io.[ch] and
pvrusb2-ioread.[ch]. The new logic currently only implements the
"read" method, however even though the old logic did all three
v4l methods, it appears that the other two methods were broken there
anyway. The new streaming logic has two well-defined internal shearing
layers. The lowest level handles all USB buffering, and the middle level
translates that into stream data for the read() system call. Other middle
levels can later be implemented alongside for other forms of I/O as
The entire v4l API implementation has now been isolated and stuffed into a
single place, pvrusb2-v4l2.c. The old streaming logic had been entangled
with v4l, so this step could not happen until that the new streaming logic
had been implemented first.
Debug logging has been further cleaned up. High-bandwidth log messages
are isolated behind specific debug mask bits and their formatting has been
regularized so that a post-processing script may scrape useful data out of
the kernel log. There's enough information available now to theoretically
reconstruct blow-by-blow events in the streaming path. (Obviously that
level of logging should normally remain off.)
Pulled in the following modules from ivtv: tuner.ko, msp3400.ko, and
tveeprom.ko. Requiring people to hunt these down from other sources
seemed to be too much trouble. The final straw was the type 58 tuner
problem in v4l. However that doesn't mean I've forked the code. It
should still be possible to continue using other versions. The copies in
this driver snapshot are exact copies from ivtv and I intend to keep it
that way. I hope.
Isolated usage struct tveeprom; this is an internal structure from
tveeprom.ko which may be subject to thrashing and changes. It is now only
visible inside of pvrusb2-eeprom.c.
Defined more controls that can be accessed through sysfs. Signal presence
(ctl_signal_present) can be queried (but obviously not set), the streaming
state (ctl_streaming_enabled) can be queried (but not set), and there's a
control present (ctl_streaming_force_suspend) that can be used to
explicitly, temporarily suspend streaming (for debugging).
We now tell tuner.ko about which video standard we are using. Most of the
time this probably isn't going to matter much since it seems that the
tuner type also implies the standard. But the module has the ability to
be told and there probably are some cases where it matters, so we're
telling it now.
Do a better job of logging which video standards have been detected from
the eeprom. This should make things easier to understand for
multi-standard tuners where we happen to "pick" the wrong
choice. (Picking the wrong standard is no big deal since it is easily
corrected through sysfs and a v4l app should probably be setting the
choice of standard anyway.)
The video standard is now initialized to a value that is determined by
what has been read from the Hauppauge eeprom. Previously it had been
default-initialized to NTSC.
A frequency table has been implemented in the driver and made available
Fixed bug in video standard enumeration. This problem was introduced in
the previous snapshot due to the v4l driver abstraction layer cleanup.
New sysfs control interface. This is the big change here for this
snapshot. It's possible now to operate all driver controls just by
examining / changing files under /sys/class/pvrusb2/sn-XXXX (XXXX is the
device's serial number).
More junk gutted out of pvrusb2-main.c.
New driver abstraction layer implemented, the definition for which can be
found in pvrusb2-base.h. This is an attempt to isolate code which manages
the driver as a whole, and also provides a shearing layer where we can
plug in additional control interfaces. The existing v4l code is hacked up
to work through this abstraction layer as just "another
interface". The previously-mentioned "shell" remnant
struct in pvrusb2.h (see further back in the history) is now completely
removed, in favor of this new layer.
First attempt made at trying to understand what is going on in
pvrusb2-encoder.c (the part which handles the mpeg2 encoder hardware). As
part of this, a large amount of debug code is added. Bugs fixed here;
these problems were collateral damage from all the other rework.
Implemented a kernel thread for the driver. Each driver instance gets a
kernel thread now, to handle whatever background tasks are needed which
otherwise would block the caller.
All major device initialization moved to the kernel thread. All firmware
loading happens here now instead of being inline with the USB probe
function. This seems to have noticably sped up driver initialization.
And hotplug auto-loading of the driver works now. Yay!
Fixed problem with video standard handling and audio processing. We need
to tell msp3400.ko what video standard is in use, because that driver uses
it in determining how to configure the audio processor. Previously this
was working "by accident", but a recent change to msp3400.ko
exposed this bug in the pvrusb2 driver
Brought in new saa7115.c from ivtv. The only real change is a largely
cosmetic fix required due to a change in the kernel's I2C subsystem.
Unfortunately lots of lines changed anyway because it appears that the
ivtv driver author decided to fix all the tabbing / spacing in the driver
at this time.
Driver author info updated.
Most of the debug code has been scrubbed and cleaned up. It's now
possible to selectively enable different kinds of debugging printk()'s via
the debug module parameter which is now a bit mask. Possible bit mask
choices are defined in pvrusb2-debug.h
The first really big change: New hardware abstraction layer implemented,
core interface is defined pvrusb2-hdw.h. Implementation is spread across
other new modules: pvrusb2-hdw-internal.h, pvrusb2-hdw.c,
pvrusb2-hdw-low.c. All other modules which touch the hardware except the
streaming code are restructured to now hide behind this interface. All
driver controls are abstracted as part of this, with a unified interface
laid out in pvrusb2-hdw.h. The driver's ioctl() (where much of the v4l
part of this puzzle exists) is reworked to operate through this interface.
The master control structure in pvrusb2.h is gutted down to a shell, and
fitted with an instance of the new layer. The driver's global state
variable is history now, no longer needed.
Moved all code which handles firmware loading into its own module:
New streaming code written, which can be found in pvrusb2-io.c,
pvrusb2-ioread.c, and pvrusb2-iommap.c. All of it compiles, but none of
it is being put to use it. This big shakeup must wait until later...
Beginnings of a new driver abstraction layer put together but not put in
Removed all code for handling the video capture chip, and replaced it with
far simpler code that merely calls out to saa7115.ko, a module available
from the ivtv driver (not currently in v4l).
Imported CVS snapshots of ivtv driver, and copied saa7115.c from there
into this driver. What is in this driver is an exact copy of what came
from ivtv - no local changes. I intend to keep it exact. Several header
files were also brought over and one was hacked up in order to make it
possible for saa7115.c to compile cleanly outside of ivtv.
More I2C low level fixes.
Minor cleanup in driver's ioctl() function.
Additional eeprom handling cleanup.
More I2C adapter driver cleanup / fixes.
Removed all code for handling audio, and replaced it with far simpler code
that merely calls out to msp3400.ko which is already in v4l.
Reworked existing pvrusb2-saa7115.c code (video digitizer) to use new I2C
Additional tuner handling cleanup.
Control handling cleanup in the driver's ioctl() implementation.
Added code to module initialization such that other needed I2C based
modules (e.g. tuner.ko) are explicitly loaded at the same time, if not
New module: pvrusb2-i2c.c, containing low level logic for appropriate
interfacing with kernel's I2C layer.
Removed IR handling logic, killing an instability problem when the driver
Removed all code for handling the Hauppauge eeprom, and replaced it with
far simpler code that merely calls out to tveeprom.ko which is already in
Removed all tuner handling code (scattered around in various places), and
replaced it with far simpler code that merely passes off the actual work
to tuner.ko which is already in v4l.
Fixed USB run-time warning: changed call to usb_unlink_urb() to
Defined two new tuner types.
Minor tuner handling tweak.
Renamed video standard string names and video input string names to better
match expectations of xawtv.
(Aurelien's 07112004 snapshot starting point)