CVS log for DVB/driver/dvb_frontend.c

Up to [DVB] / DVB / driver

Request diff between arbitrary revisions

Keyword substitution: kv
Default branch: MAIN

Initialize '.readers' in dvb_frontend.c, required due to 2003-08-20 changes

applied patch by Juergen Peitz to introduce FE_CAN_RECOVER

use aquire_signal flag to call FE_RESET only after tuning
until FE_HAS_LOCK has been signalled, and not when
FE_HAS_LOCK drops out for short periods of time later

- remove udelay()s from alps_bsrv2.c;
- instead schedule_timeout(1) in dvb_frontend.c after setting
  frontend and before waking up frontend thread
- do FE_RESET in each iteration of frontend thread if !FE_HAS_LOCK
- printk() -> dprintk() for AFC in ves1820.c

don't bend for frontends on different adapters

Deadlock. Bei einem blockierenden FE_GET_EVENT wird der fe-thread am
weiterarbeiten gehindert. Folglich auch keine weiteren Events...
found by wjoost

patch from Jaakko Hyvatti http://hyvatti.iki.fi/~jaakko/dvb/dvb6.diff:
- Makes zigzag to not change frequency for DVB-c at all. It only resets
  the parameters over and over again.
- The IF of 35937500 might be better to be 36000000. This might be what the
  card designer meant. My card shows the AFC fine tuning offset to be closest
  to zero with 35937500 (one step below 36MHz). Some other cards might be
  different. I have not heard of any complaints about my patch not working
  for someone in this respect.
- Has anyone ever tested this auto inversion thing? DVB-c is far too slow to
  tune and lock for this auto inversion code to work. Therefore you need to
  manually set it. Maybe zigzag changes above allow the auto inversion to work.
- QAM registers have been tweaked to non-specification values. I do not know
  if someone has counterexamples, but this works for a lot of people.

fixed remaining frontend thread issues (I hope ;-)
(workaround needed because kernel_thread() fails when the process
opening the frontend device is ptraced)

trying to fix frontend thread start/stop problems

ensure that poll_wait() is always called in poll()

use writers count, not users count to decide if we want to shut down the frontend.

Now the frontend will get closed when no more users with write access have the frontend opened.

apply fix regarding first tune after device open reported by Andrew de Quincey <adq_dvb@lidskialf.net>

Do not return values greater than zero

allow multiple read only users

apply patch suggested by Uwe Scheffler

stop fe thread without FE_SLEEP if shutdown timeout is 0

 - rename kdvb-fe to kdvb-fe-%i:%i (adapter/id)
 - rename arm_mon thread to kdvb-av7110
 - don't start high priority frontend thread before first tuning

remove last ~0 voodoo - my fault

don't stop ~0 thread

applied patch suggested by Oliver Endriss <o.endriss@gmx.de> in order to avoid
reace conditions in starting threads

call reparent_to_init() after daemonize ()

replaced down_interruptible() by non-interruptible down() and changed
kdvb-fe thread exit code - hopefully this resolves the problems reported
with occasional oopses at deinitialisation time...

fixed a stupid bug regarding FE_CAN_CLEAN_SETUP

add FE_CAN_CLEAN_SETUP bit to avoid unnecessairy zigzag delay when using
demodulators which can set new parameters without artifacts
VS: ----------------------------------------------------------------------

wait a little more again before starting the zigzag scan
Klaus Schmidinger reported some problems with the new value on BSRV2 based cards

frontend notifier fixes by Florian Schirmer

make the required number of lost syncs that cause a recovery call a little
smaller

make "DVB: initialising frontend..." a debug message ind introduce
"DVB: registering frontend..."

user counter direction changed a long time ago, this screwed up the power
management code. The bug never showed up so late since we disabled this
feature by default...

enable power management by default if driver is compiled into the kernel

don't tune to any default transponder anymore, don't set any default PIDs

backport of 2.5 driver changes

the Schmidinger Patch

zigzag tuning

C99 patches by <aschultz@apophis.eastlink.de>

rewrote dvb_frontend_bend_frequency()

remove a debug message

remove a debug message

zigzag tuning...

replaced event spinlock by semaphore

the new HEAD

make room for the new HEAD trunk

ves1820: initialize stepsize
bend_frequency(): bail out after 10 iterations

some commenting

mute TS only for frontends which can't mute TS when they lose lock

return value of dvb_register_frontend is int, not a pointer

compatibility crap for kernels from the stone age

count users positive again

deactivate recovery for now, we need much more testing, this gets tricky...

dvb_call_frontend_notifiers(fe,0) before reinitializing

set parameters when reinitializing the frontend

recovery experiment

recovery experiment

recovery experiment

recovery experiment

recovery experiment

- use list_for_each_safe() again when modifying the lists -- this caused
   segfaults on 2.4.10 kernels

- add MODULE_XXX() macros to dvb_demux to suppress warnings at module load time

- handle av7110_irq in tasklet. The spinlock there might be a bit paranoid,
   we can check and beautify this later.

- wait_for_debi_done_irq() is obsolete now, use wait_for_debi_done() everywhere

- suppress av7110_ir debug messages by default

fixed a recursive locking problem, tune to n-tv again on module loading

don't initialize frontend when registering before/after ioctls.
For some reason this causes lookups with many cards, I'll have to
investigate this further...

- works on SMP systems again

- added lotsa debug messages while hunting a bug

use down_interruptibe everywhere

clean exiting from threads in the case down_interruptible failed

use down_interruptible where possible to fix driver unloading problems

more restrictive mutex locking

use -EOVERFLOW instead of self-defined error value

 - link related object files together in order to get less kernel modules
 - link related object files together in order to get less kernel modules
 - configurable IR keymaps using /proc/av7110_ir
 - make frontend code a little less verbose
 - removed some old/unused stuff

- fixed dvb_bend_frequency() for continous frequency sweeps
- added DMX_STREAM_X enums

clear timeout bit once more...

some license string cleanups

tuning support for NOVA-CI (not tested)

modified js-compatibility-flicker-bit behaviour

"§!$%&/

empty event queue at device open time

less verbosity

FE_TIMEDOUT event implementation

modified init message

renamed frontend thread to "kdvb-fe"

Power management is implemented now, LNB power and frontend is shut down
<shutdown_timeout> seconds after close().
<shutdown_timeout> is a module parameter.

cvs sucks. sometimes. always when it forgets files to ci

adapt to change from dvb_device_t ** to * in dvbdev.c

delay frontend notifier on sync to ensure a stable TS. delay is configurable in the dvb_frontend_info struct

keep last status in dvb_frontend_data struct and set it to zero
if diseqc or 22KHz changes

also notify everybody of lost sync if diseqc or tone ioctls are called

 - struct dvb_frontend_info has now a FrontendCapabilities field which
    specifies which modes the frontend supports, if it can mute the TS
    when running out of sync etc.

use the DVB device registrar

allow an additional data argument in frontend notifiers

- removed the static device table from dvbdev.c, replaced by a list, we don't
   have any limit for the number of devices now.

   Since the Nokia API specifies that each open() call for a demux device
   allocates a new filter we can't use the file->private_data pointer to
   save the dvbdev struct, but have to search our device list linearly for
   each open/ioctl/release call. Should not be a real problem since we will
   rarely have more than a few DVB devices, but anyway - the code is ugly.

   Did I ever told you that the Nokia API sucks?

- put all EXPORT_SYMBOL() macros in dvb_ksyms.c and av7110_ksyms.c,
   now we don't have to build the whole driver from scratch when we touch
   a file exporting symbols...

the not yet working cable frontend are not a bug but a missing feature.
That rev2.1 support is broken is just not true, I have a rev2.1 card with
Alps BSRU6 tuner here, it's just working fine. What card exactly is not
working? (what frontend?)

For the disable-demux-before-tuning/enable-again-whe-TS-is-in-sync we have
a clean framework using the frontend notifiers now, the AV7110 just has to
call the register functions...

I'll care about these issues next week.

made frontend_ioctl_list and frontend_list static

remove some verbosity

 - av7110: introduced something like a av7110/saa7146 namespace

 - ported saa7146 i2c driver to the new internal driver API, fixed
    problems with concatenated messages

 - tested and fixed Alps BSRU6, BSRV2, Grundig 29504-401 frontend code

 - initialize dvb device directory when initializing the DVB bridge

 - ported DiSEqC, 13/18V and 22kHz stuff to the new frontend API

 - ported szap and tzap

 - some new tiny test utilities (setpid, set22k, setvoltage, diseqc)

 - DMX_SATELLITE_FE, DMX_CABLE_FE and DMX_TERRESTRIAL_FE in demux.h are gone,
    defined DMX_FRONTEND_0, DMX_FRONTEND_1, DMX_FRONTEND_2 and DMX_FRONTEND_3

there are still a lot of FIXME's and debug messages, but the stuff is basically
working...

Peter Hettkamp's driver for Pinnacle's PCTV Sat cards
To be merged into the NEWSTRUCT branch

auto inversion for stv0299 and ves1820

DiSEqC 2.0 related frontend API changes + some files that got lost in my last commit

- replace broken include files
- minor changes in the frontend API
- no typedef'd structs anymore
- renamed grundig frontend because the old name was ambigous

ppc fixes

stricter locking of filters

some changes in kernel demux API

started adapting margi code to new structure

new fronten stuff by Holger Waechtler
user interface not tested, probably does not
work as it is right now

use dvb_frontend_old.* for old interface

kernel config files

fix frontend polling and adapt szap to new include paths

new dvbdev registering (still in flux and only av7110 works)

more restructuring (much more soon)

change include files

use do_gettimeofday

new directory structure

moved some more stuff from dvb.c in dvb_frontend.c

semaphores for A/V playback control/writing
cleanups in frontend inits

CI link layer device (experimental)

fixed fail status (by Falck Kenneth <kennu@swelcom.fi>)

check if MODULE_LICENSE exists (for older 2.4.x kernels)

add separate module for frontend control