| Message ID | 20090404142837.3e12824c@hyperion.delvare (mailing list archive) |
|---|---|
| State | RFC |
| Headers | show |
On Sat, 2009-04-04 at 14:28 +0200, Jean Delvare wrote: > Let card drivers probe for IR receiver devices and instantiate them if > found. Ultimately it would be better if we could stop probing > completely, but I suspect this won't be possible for all card types. > > There's certainly room for cleanups. For example, some drivers are > sharing I2C adapter IDs, so they also had to share the list of I2C > addresses being probed for an IR receiver. Now that each driver > explicitly says which addresses should be probed, maybe some addresses > can be dropped from some drivers. > > Also, the special cases in saa7134-i2c should probably be handled on a > per-board basis. This would be more efficient and less risky than always > probing extra addresses on all boards. I'll give it a try later. > > Signed-off-by: Jean Delvare <khali@linux-fr.org> > Cc: Mauro Carvalho Chehab <mchehab@infradead.org> > Cc: Hans Verkuil <hverkuil@xs4all.nl> > Cc: Andy Walls <awalls@radix.net> > Cc: Mike Isely <isely@pobox.com> > --- > linux/drivers/media/video/cx18/cx18-i2c.c | 30 ++ > linux/drivers/media/video/ivtv/ivtv-i2c.c | 31 ++ > linux/include/media/ir-kbd-i2c.h | 2 > 17 files changed, 284 insertions(+), 196 deletions(-) > > --- v4l-dvb.orig/linux/drivers/media/video/cx18/cx18-i2c.c 2009-04-04 10:53:15.000000000 +0200 > +++ v4l-dvb/linux/drivers/media/video/cx18/cx18-i2c.c 2009-04-04 10:58:36.000000000 +0200 > @@ -211,7 +211,32 @@ static struct i2c_algo_bit_data cx18_i2c > .timeout = CX18_ALGO_BIT_TIMEOUT*HZ /* jiffies */ > }; > > -/* init + register i2c algo-bit adapter */ > +static void init_cx18_i2c_ir(struct cx18 *cx) > +{ > + struct i2c_board_info info; > + /* The external IR receiver is at i2c address 0x34 (0x35 for > + reads). Future Hauppauge cards will have an internal > + receiver at 0x30 (0x31 for reads). In theory, both can be > + fitted, and Hauppauge suggest an external overrides an > + internal. > + > + That's why we probe 0x1a (~0x34) first. CB > + */ > + const unsigned short addr_list[] = { > + 0x1a, 0x18, 0x64, 0x30, > + I2C_CLIENT_END > + }; I think this is way out of date for cx18 based boards. The only IR chip I know of so far is the Zilog Z8F0811 sitting at 7 bit addresses 0x70-0x74. I guess 0x71 is the proper address for Rx. I'll let you know when I test. > + memset(&info, 0, sizeof(struct i2c_board_info)); > + strlcpy(info.type, "ir-kbd", I2C_NAME_SIZE); > + > + /* The IR receiver device can be on either I2C bus */ > + if (i2c_new_probed_device(&cx->i2c_adap[0], &info, addr_list)) > + return; > + i2c_new_probed_device(&cx->i2c_adap[1], &info, addr_list); > +} > + > +/* init + register i2c adapters + instantiate IR receiver */ > int init_cx18_i2c(struct cx18 *cx) > { > int i, err; > @@ -279,6 +304,9 @@ int init_cx18_i2c(struct cx18 *cx) > err = i2c_bit_add_bus(&cx->i2c_adap[1]); > if (err) > goto err_del_bus_0; > + > + /* Instantiate the IR receiver device, if present */ > + init_cx18_i2c_ir(cx); > return 0; I have an I2C related question. If the cx18 or ivtv driver autoloads "ir-kbd-i2c" and registers an I2C client on the bus, does that preclude lirc_i2c, lirc_pvr150 or lirc_zilog from using the device? LIRC users may notice, if it does. If that is the case, then we probably shouldn't autoload the ir-kbd module after the CX23418 i2c adapters are initialized. I'm not sure what's the best solution: 1. A module option to the cx18 driver to tell it to call init_cx18_i2c_ir() from cx18_probe() or not? (Easiest solution) 2. Some involved programmatic way for IR device modules to query bridge drivers about what IR devices they may have, and on which I2C bus, and at what addresses to probe, and whether a driver/module has already claimed that device? (Gold plated solution) Regards, Andy -- To unsubscribe from this list: send the line "unsubscribe linux-media" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Nacked-by: Mike Isely <isely@pobox.com> This will interfere with the alternative use of LIRC drivers (which work in more cases that ir-kbd). It will thus break some peoples' use of the driver. Also we have better information on what i2c addresses needed to be probed based on the model of the device - and some devices supported by this device are not from Hauppauge so you are making a too-strong assumption that IR should be probed this way in all cases. Also, unless ir-kbd has suddenly improved, this will not work at all for HVR-1950 class devices nor MCE type PVR-24xxx devices (different incompatible IR receiver). This is why the pvrusb2 driver has never directly attempted to load ir-kbd. -Mike On Sat, 4 Apr 2009, Jean Delvare wrote: > --- v4l-dvb.orig/linux/drivers/media/video/pvrusb2/pvrusb2-i2c-core.c 2009-04-04 10:53:08.000000000 +0200 > +++ v4l-dvb/linux/drivers/media/video/pvrusb2/pvrusb2-i2c-core.c 2009-04-04 10:58:36.000000000 +0200 > @@ -649,6 +649,27 @@ static void do_i2c_scan(struct pvr2_hdw > printk(KERN_INFO "%s: i2c scan done.\n", hdw->name); > } > > +static void pvr2_i2c_register_ir(struct i2c_adapter *i2c_adap) > +{ > + struct i2c_board_info info; > + /* The external IR receiver is at i2c address 0x34 (0x35 for > + reads). Future Hauppauge cards will have an internal > + receiver at 0x30 (0x31 for reads). In theory, both can be > + fitted, and Hauppauge suggest an external overrides an > + internal. > + > + That's why we probe 0x1a (~0x34) first. CB > + */ > + const unsigned short addr_list[] = { > + 0x1a, 0x18, 0x4b, 0x64, 0x30, > + I2C_CLIENT_END > + }; > + > + memset(&info, 0, sizeof(struct i2c_board_info)); > + strlcpy(info.type, "ir-kbd", I2C_NAME_SIZE); > + i2c_new_probed_device(i2c_adap, &info, addr_list); > +} > + > void pvr2_i2c_core_init(struct pvr2_hdw *hdw) > { > unsigned int idx; > @@ -696,6 +717,9 @@ void pvr2_i2c_core_init(struct pvr2_hdw > } > } > if (i2c_scan) do_i2c_scan(hdw); > + > + /* Instantiate the IR receiver device, if present */ > + pvr2_i2c_register_ir(&hdw->i2c_adap); > } > > void pvr2_i2c_core_done(struct pvr2_hdw *hdw)
On Sat, 4 Apr 2009, Andy Walls wrote: [...] > > I have an I2C related question. If the cx18 or ivtv driver autoloads > "ir-kbd-i2c" and registers an I2C client on the bus, does that preclude > lirc_i2c, lirc_pvr150 or lirc_zilog from using the device? LIRC users > may notice, if it does. > > If that is the case, then we probably shouldn't autoload the ir-kbd > module after the CX23418 i2c adapters are initialized. > > I'm not sure what's the best solution: > > 1. A module option to the cx18 driver to tell it to call > init_cx18_i2c_ir() from cx18_probe() or not? (Easiest solution) > > 2. Some involved programmatic way for IR device modules to query bridge > drivers about what IR devices they may have, and on which I2C bus, and > at what addresses to probe, and whether a driver/module has already > claimed that device? (Gold plated solution) > > Regards, > Andy Ah, glad to see I'm not the only one concerned about this. I suppose I could instead add a module option to the pvrusb2 driver to control autoloading of ir-kbd (option 1). It also should probably be a per-device attribute, since AFAIK, ir-kbd only even works when using older Hauppauge IR receivers (i.e. lirc_i2c - cases that would otherwise use lirc_pvr150 or lirc_zilog I believe do not work with ir-kbd). Some devices handled by the pvrusb2 driver are not from Hauppauge. Too bad if this is the case, it was easier to let the user decide just by choosing which actual module to load. -Mike
On Sat, 2009-04-04 at 11:05 -0500, Mike Isely wrote: > On Sat, 4 Apr 2009, Andy Walls wrote: > > [...] > > > > > I have an I2C related question. If the cx18 or ivtv driver autoloads > > "ir-kbd-i2c" and registers an I2C client on the bus, does that preclude > > lirc_i2c, lirc_pvr150 or lirc_zilog from using the device? LIRC users > > may notice, if it does. > > > > If that is the case, then we probably shouldn't autoload the ir-kbd > > module after the CX23418 i2c adapters are initialized. > > > > I'm not sure what's the best solution: > > > > 1. A module option to the cx18 driver to tell it to call > > init_cx18_i2c_ir() from cx18_probe() or not? (Easiest solution) > > > > 2. Some involved programmatic way for IR device modules to query bridge > > drivers about what IR devices they may have, and on which I2C bus, and > > at what addresses to probe, and whether a driver/module has already > > claimed that device? (Gold plated solution) I was thinking about this while mowing the lawn today.... The objectives seem to be: 1. Avoid auto probing 2. Leverage the bridge driver's knowledge of what should be available 3. Allow the user to dictate which kernel IR driver module be used with the subordinate device. So my rough outline of an idea (which probably runs slightly afoul of Hans' media_controller device, but we don't have it yet): 1. Add a function to the v4l2 framework to iterate over all the v4l2_device's that are registered (if there isn't one already). 2. Add a method to the v4l2_device class to return the IR subdevice for a given v4l2_device: v4l2_subdev *v4l2_device_get_ir_subdev(v4l2_device *dev); and if it returns NULL, that device has no IR chip. 3. To the v4l2_subdev framework add: struct v4l2_subdev_ir_ops { (*enumerate) (v4l2_subdev *sd, /* bus_type, bus #, addr for Rx, addr for Tx */); (*claim) (v4l2_subdev *sd, /* claiming driver name string, going-away callback function pointer */); (*release) (v4l2_subdev *sd, /* handle */); bool (*is_claimed) (v4l2_subdev *sd, /* output string of the "owner" */); /* Or maybe just */ (*send) (v4l2_subdev *sd, /* data buffer */); (*receive) (v4l2_subdev *sd, /* data buffer */); } and have the bridge driver support these. (I also had some in mind for the IR micro-controller debug/programming port, but the above will fit the task at hand I think.) OK so that's all a bit rough around the edges. The idea is a uniform call in for ir-kdb-i2c or lirc_foo or ir_foo to get at an IR chip behind a bridge device, that the bridge device driver itself cares about very little. *Except* ir driver modules would be coordinated by the bridge driver in what they can and cannot do to get at the IR device. This coordination prevents bad things on the bridge chip's I2C bus(es) or from having modules racing to get the IR device. That way whatever module the user loads will get first shot at claiming the IR chip. This also provides a discovery mechanism four use by ir driver modules that is informed by the bridge chip driver. I think lirc_foo can also still use it's current way of doing business too. It really just looks like a small subset of what Hans intended for the media controller, so maybe this would be a good chance to get some "lessons learned." Regards, Andy > > Regards, > > Andy > > Ah, glad to see I'm not the only one concerned about this. > > I suppose I could instead add a module option to the pvrusb2 driver to > control autoloading of ir-kbd (option 1). It also should probably be a > per-device attribute, since AFAIK, ir-kbd only even works when using > older Hauppauge IR receivers (i.e. lirc_i2c - cases that would otherwise > use lirc_pvr150 or lirc_zilog I believe do not work with ir-kbd). Some > devices handled by the pvrusb2 driver are not from Hauppauge. Too bad > if this is the case, it was easier to let the user decide just by > choosing which actual module to load. I think we can get there and still have the random probing reduced. Regards, Andy > -Mike -- To unsubscribe from this list: send the line "unsubscribe linux-media" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On Sat, 2009-04-04 at 18:25 -0400, Andy Walls wrote: > On Sat, 2009-04-04 at 11:05 -0500, Mike Isely wrote: > So my rough outline of an idea (which probably runs slightly afoul of > Hans' media_controller device, but we don't have it yet): > > 1. Add a function to the v4l2 framework to iterate over all the > v4l2_device's that are registered (if there isn't one already). > > 2. Add a method to the v4l2_device class to return the IR subdevice for > a given v4l2_device: > > v4l2_subdev *v4l2_device_get_ir_subdev(v4l2_device *dev); > > and if it returns NULL, that device has no IR chip. > > > 3. To the v4l2_subdev framework add: > > struct v4l2_subdev_ir_ops { > (*enumerate) (v4l2_subdev *sd, /* bus_type, bus #, addr for Rx, addr for Tx */); > (*claim) (v4l2_subdev *sd, /* claiming driver name string, going-away callback function pointer */); > (*release) (v4l2_subdev *sd, /* handle */); > bool (*is_claimed) (v4l2_subdev *sd, /* output string of the "owner" */); > /* Or maybe just */ > (*send) (v4l2_subdev *sd, /* data buffer */); > (*receive) (v4l2_subdev *sd, /* data buffer */); > } > > and have the bridge driver support these. (I also had some in mind for > the IR micro-controller debug/programming port, but the above will fit > the task at hand I think.) > > > OK so that's all a bit rough around the edges. The idea is a uniform > call in for ir-kdb-i2c or lirc_foo or ir_foo to get at an IR chip behind > a bridge device, that the bridge device driver itself cares about very > little. *Except* ir driver modules would be coordinated by the bridge > driver in what they can and cannot do to get at the IR device. This > coordination prevents bad things on the bridge chip's I2C bus(es) or > from having modules racing to get the IR device. That way whatever > module the user loads will get first shot at claiming the IR chip. This > also provides a discovery mechanism four use by ir driver modules that > is informed by the bridge chip driver. I think lirc_foo can also still > use it's current way of doing business too. > > It really just looks like a small subset of what Hans intended for the > media controller, so maybe this would be a good chance to get some > "lessons learned." Oops. That leaves DTV only devices with IR out in the cold, unless they start to implement a v4l2_device and IR v4l2_subdev as well, or unless they were never used with ir-kbd-i2c in the first place. Regards, Andy -- To unsubscribe from this list: send the line "unsubscribe linux-media" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Hi Andy, On Sat, 04 Apr 2009 09:42:09 -0400, Andy Walls wrote: > On Sat, 2009-04-04 at 14:28 +0200, Jean Delvare wrote: > > Let card drivers probe for IR receiver devices and instantiate them if > > found. Ultimately it would be better if we could stop probing > > completely, but I suspect this won't be possible for all card types. > > > > There's certainly room for cleanups. For example, some drivers are > > sharing I2C adapter IDs, so they also had to share the list of I2C > > addresses being probed for an IR receiver. Now that each driver > > explicitly says which addresses should be probed, maybe some addresses > > can be dropped from some drivers. > > > > Also, the special cases in saa7134-i2c should probably be handled on a > > per-board basis. This would be more efficient and less risky than always > > probing extra addresses on all boards. I'll give it a try later. > > > > Signed-off-by: Jean Delvare <khali@linux-fr.org> > > Cc: Mauro Carvalho Chehab <mchehab@infradead.org> > > Cc: Hans Verkuil <hverkuil@xs4all.nl> > > Cc: Andy Walls <awalls@radix.net> > > Cc: Mike Isely <isely@pobox.com> > > --- > > linux/drivers/media/video/cx18/cx18-i2c.c | 30 ++ > > linux/drivers/media/video/ivtv/ivtv-i2c.c | 31 ++ > > linux/include/media/ir-kbd-i2c.h | 2 > > 17 files changed, 284 insertions(+), 196 deletions(-) > > > > > --- v4l-dvb.orig/linux/drivers/media/video/cx18/cx18-i2c.c 2009-04-04 10:53:15.000000000 +0200 > > +++ v4l-dvb/linux/drivers/media/video/cx18/cx18-i2c.c 2009-04-04 10:58:36.000000000 +0200 > > @@ -211,7 +211,32 @@ static struct i2c_algo_bit_data cx18_i2c > > .timeout = CX18_ALGO_BIT_TIMEOUT*HZ /* jiffies */ > > }; > > > > -/* init + register i2c algo-bit adapter */ > > +static void init_cx18_i2c_ir(struct cx18 *cx) > > +{ > > + struct i2c_board_info info; > > + /* The external IR receiver is at i2c address 0x34 (0x35 for > > + reads). Future Hauppauge cards will have an internal > > + receiver at 0x30 (0x31 for reads). In theory, both can be > > + fitted, and Hauppauge suggest an external overrides an > > + internal. > > + > > + That's why we probe 0x1a (~0x34) first. CB > > + */ > > + const unsigned short addr_list[] = { > > + 0x1a, 0x18, 0x64, 0x30, > > + I2C_CLIENT_END > > + }; > > > I think this is way out of date for cx18 based boards. The only IR chip > I know of so far is the Zilog Z8F0811 sitting at 7 bit addresses > 0x70-0x74. I guess 0x71 is the proper address for Rx. I'll let you > know when I test. This address list comes from the ir-kbd-i2c driver. The cx18 driver happens to use the same I2C adapter ID as the ivtv driver (I2C_HW_B_CX2341X) and this is what the ir-kbd-i2c driver used to decide which addresses to probe. As I don't know anything about the hardware, I had to keep the new code compatible with the old one and keep probing the same addresses. Now, if you tell me that this list doesn't make sense for cx18 boards, we can change it. As addresses 0x70-0x74 were not probed so far on cx18 boards, I guess that IR support never worked for cx18 (at least not with ir-kbd-i2c)? Does ir-kbd-i2c support the Zilog Z8F0811 at all? If IR on the cx18 is not supported (by the ir-kbd-i2c driver) then I can simplify my patch set and omit the cx18 entirely. > > + memset(&info, 0, sizeof(struct i2c_board_info)); > > + strlcpy(info.type, "ir-kbd", I2C_NAME_SIZE); > > + > > + /* The IR receiver device can be on either I2C bus */ > > + if (i2c_new_probed_device(&cx->i2c_adap[0], &info, addr_list)) > > + return; > > + i2c_new_probed_device(&cx->i2c_adap[1], &info, addr_list); > > +} > > + > > +/* init + register i2c adapters + instantiate IR receiver */ > > int init_cx18_i2c(struct cx18 *cx) > > { > > int i, err; > > @@ -279,6 +304,9 @@ int init_cx18_i2c(struct cx18 *cx) > > err = i2c_bit_add_bus(&cx->i2c_adap[1]); > > if (err) > > goto err_del_bus_0; > > + > > + /* Instantiate the IR receiver device, if present */ > > + init_cx18_i2c_ir(cx); > > return 0; > > I have an I2C related question. If the cx18 or ivtv driver autoloads > "ir-kbd-i2c" and registers an I2C client on the bus, does that preclude > lirc_i2c, lirc_pvr150 or lirc_zilog from using the device? LIRC users > may notice, if it does. I don't know anything about lirc_i2c, lirc_pvr150 or lirc_zilog. I tend to ignore all the code that is neither in the Linux kernel tree nor in the v4l-dvb tree. If you want me to answer this question, you'll have to tell me what exactly these drivers are doing as far as I2C is concerned. Do they instantiate I2C clients? Or do they do raw I2C transfers? Do they check for address business before they do? On what basis do they attach to I2C devices? Please note that my conversion doesn't actually change anything to the autoloading of the ir-kbd-i2c driver. The bridge drivers which were loading ir-kbd-i2c (saa7134, cx23885, em28xx and cx88) still are. Those which were not, still aren't. The ir-kbd-i2c driver doesn't include a MODULE_ALIAS call as most I2C drivers do, to prevent udev from loading this driver automatically. What my conversion changes is that an "ir-kbd" I2C device may be instantiated if a probe is successful. This will make the address in question show as busy (except to i2c-dev which only considers an address as busy when a driver is bound to the device.) But that's about it. > If that is the case, then we probably shouldn't autoload the ir-kbd > module after the CX23418 i2c adapters are initialized. > > I'm not sure what's the best solution: > > 1. A module option to the cx18 driver to tell it to call > init_cx18_i2c_ir() from cx18_probe() or not? (Easiest solution) Sounds perfectly sensible. I seem to remember that Hans Verkuil told me he wanted something like this for ivtv. As a matter of fact, the saa7134, em28xx and cx231xx already have such a module parameter (disable_ir). Implementing the same for bttv, cx88, cx18, ivtv or any other driver should be fairly trivial. > 2. Some involved programmatic way for IR device modules to query bridge > drivers about what IR devices they may have, and on which I2C bus, and > at what addresses to probe, and whether a driver/module has already > claimed that device? (Gold plated solution) I'd rather name this the over-engineered solution. It's really looking at the situation by the wrong end (that is, with the legacy i2c binding model still in mind.) Bridge drivers know which IR receivers can be present and at which address, it is up to them to instantiate the appropriate I2C devices on the bus, possibly with platform data to help the I2C drivers (be they ir-kbd-i2c, lirc or whatever.) This is exactly what my code does. The fact that the same IR chip can be handled by 2 or more I2C drivers is a bad idea to start with. Why the hell did we do that in the first place? If you want a clean solution to the problem, it clearly starts with getting rid of this mess and having each IR receiver chip on I2C supported by exactly one I2C driver and make sure the driver in question is in the Linux kernel tree. Spending time on any other "clean solution" is wasting time IMHO. Still, note that it is totally possible to have several I2C drivers support the same device. The new model supports this, just like the old model did. I2C devices are instantiated by bridge drivers, which give them a _name_. Several I2C drivers are allowed to support that chip name, and the first one loaded will get to bind to the device. The "ir-i2c" devices created by cx18, ivtv etc. can be requested by other drivers than ir-kbd-i2c if you want to do that. This will require some changes to lirc_i2c and friends, but at this point changes to these are very needed anyway. I hope I managed to clarify the situation a bit.
Hi Mike, On Sat, 4 Apr 2009 10:51:01 -0500 (CDT), Mike Isely wrote: > > Nacked-by: Mike Isely <isely@pobox.com> > > This will interfere with the alternative use of LIRC drivers (which work > in more cases that ir-kbd). Why then is ir-kbd in the kernel tree and not LIRC drivers? > It will thus break some peoples' use of the driver. Do you think it will, or did you test and it actually does? If it indeed breaks, please explain why, so that a solution can be found. > Also we have better information on what i2c addresses needed to > be probed based on the model of the device This is excellent news. As I said in the header comment of the patch, avoiding probing when we know what the IR receiver is and at which address it sits is the way to go. Please send me all the information you have and I'll be happy to add a patch to the series, that skips probing whenever possible. Or write that patch yourself if you prefer. > - and some devices supported > by this device are not from Hauppauge so you are making a too-strong > assumption that IR should be probed this way in all cases. I didn't make any assumption, sorry. I simply copied the code from ir-kbd-i2c. If my code does the wrong thing for some devices, that was already the case before. And this will certainly be easier to fix after my changes than before. On top of that, the "Hauppauge trick" is really only the order in which the addresses are probed. Just because a specific order is better for Hauppauge boards, doesn't mean it won't work for non-Hauppauge boards. > Also, unless > ir-kbd has suddenly improved, this will not work at all for HVR-1950 > class devices nor MCE type PVR-24xxx devices (different incompatible IR > receiver). I'm sorry but you can't blame me for ir-kbd-i2c not supporting some devices. I updated the driver to make use of the new binding model, but that's about all I did. > This is why the pvrusb2 driver has never directly attempted to load > ir-kbd. The pvrusb2 driver however abuses the bttv driver's I2C adapter ID (I2C_HW_B_BT848) and was thus affected when ir-kbd-i2c is loaded. This is the only reason why my patch touches the pvrusb2 driver. If you tell me you want the ir-kbd-i2c driver to leave pvrusb2 alone, I can drop all the related changes from my patch, that's very easy.
On Sun, 5 Apr 2009, Jean Delvare wrote: > Hi Mike, > > On Sat, 4 Apr 2009 10:51:01 -0500 (CDT), Mike Isely wrote: > > > > Nacked-by: Mike Isely <isely@pobox.com> > > > > This will interfere with the alternative use of LIRC drivers (which work > > in more cases that ir-kbd). > > Why then is ir-kbd in the kernel tree and not LIRC drivers? > How should I know? I don't maintain either. But I know they are both used and it's not my place to force usage of one over the other. > > It will thus break some peoples' use of the driver. > > Do you think it will, or did you test and it actually does? If it > indeed breaks, please explain why, so that a solution can be found. As I interpret this, your patch will cause the pvrusb2 to probe for the IR receiver's I2C address and bind ir-kbd-i2c to what it finds. That will break anyone's usage of the driver where another IR driver (e.g. something from LIRC) is used instead. > > > Also we have better information on what i2c addresses needed to > > be probed based on the model of the device > > This is excellent news. As I said in the header comment of the patch, > avoiding probing when we know what the IR receiver is and at which > address it sits is the way to go. Please send me all the information > you have and I'll be happy to add a patch to the series, that skips > probing whenever possible. Or write that patch yourself if you prefer. I have samples of most of the devices in question and can code proper I2C addresses for each of those for each resident I2C client. The pvrusb2 driver's device attribute structure already has allowance for specification of the correct I2C addresses to use for chips specific to each device (part of the v4l2-subdev rework I recently did). Right now the driver has built in defaults, and if a particular model needs something else, it can be overridden as part of the device's profile in pvrusb2-devattr.c. > > > - and some devices supported > > by this device are not from Hauppauge so you are making a too-strong > > assumption that IR should be probed this way in all cases. > > I didn't make any assumption, sorry. I simply copied the code from > ir-kbd-i2c. If my code does the wrong thing for some devices, that was > already the case before. And this will certainly be easier to fix after > my changes than before. No, I think the point you are missing is that by moving that logic from ir-kbd-i2c into each driver (e.g. pvrusb2) you are moving logic that *might* be executed into a spot where it *will* be executed. Remember that the pvrusb2 driver did not autoload ir-kbd-i2c before this patch. Before this change, a user could elect not to use ir-i2c-kbd simply by not loading it. The pvrusb2 driver didn't request it, because the intent all along there is that the user makes the choice by loading the desired driver. Now with this change, the pvrusb2 driver is going to attempt to load ir-kbd-i2c where asked for or not. And if not, the resulting binding will prevent lirc_i2c from later on loading. If the user had been loading lirc_i2c previously, this will cause his/her usage of IR to break. No, it's not perfect, but it worked. I'm all for improving things, but not by removing an ability that people are using. > > On top of that, the "Hauppauge trick" is really only the order in which > the addresses are probed. Just because a specific order is better for > Hauppauge boards, doesn't mean it won't work for non-Hauppauge boards. "Hauppauge trick"? > > > Also, unless > > ir-kbd has suddenly improved, this will not work at all for HVR-1950 > > class devices nor MCE type PVR-24xxx devices (different incompatible IR > > receiver). > > I'm sorry but you can't blame me for ir-kbd-i2c not supporting some > devices. I updated the driver to make use of the new binding model, but > that's about all I did. Yes, but I can point out that this change now will cause ir-kbd-i2c to be loaded in cases where the user didn't want it and will risk interference with the driver that the user *did* want. > > > This is why the pvrusb2 driver has never directly attempted to load > > ir-kbd. > > The pvrusb2 driver however abuses the bttv driver's I2C adapter ID > (I2C_HW_B_BT848) and was thus affected when ir-kbd-i2c is loaded. This > is the only reason why my patch touches the pvrusb2 driver. If you tell > me you want the ir-kbd-i2c driver to leave pvrusb2 alone, I can drop > all the related changes from my patch, that's very easy. That "abuse" is a separate issue. The pvrusb2 driver started a long time ago out-of-tree and at that time it was the only way to get IR to work at all. Personally I forgot all about it back in 2005 and was only recently reminded that this situation still exists. It should be fixed and I'd welcome the correct patch to fix this. I haven't fixed it myself, because, well I've had much bigger fish to fry here. I really don't want to throw rocks here; it's always better to work out the solution than simply block any changes at all. I realize that something has to be done here in order to keep ir-kbd-i2c viable as a choice for users of the pvrusb2 driver. To that end, how about this strategy: 1. Just drop the part of the patch for the pvrusb2 driver and get the rest merged. Yes, I realize that this will break use of ir-kbd-i2c in cooperation with the pvrusb2 driver. 2. Once ir-kbd-i2c has been updated, I will push another set of changes into the pvrusb2 driver that will make it usable there. Basically what I have in mind is similar to what you tried but I'm going to integrate it with the device profiles so that it can be appropriately loaded based on device model, with the correct I2C address in each case. And most importantly, I will add a module option to enable/disable loading or ir-kbd-i2c. This will fix my main objection, since then it will still allow lirc to be usable, for now... 3. I'd like to fix the "abuse" you mention regarding I2C_HW_B_BT848. I'm unclear on what the cleanest solution is there, but if you like to (a) point at some documentation, (b) describe what I should do, or (c) suggest a patch, I will work to deal with the problem. -Mike
On Sun, 2009-04-05 at 00:51 +0200, Jean Delvare wrote: > Hi Andy, > > On Sat, 04 Apr 2009 09:42:09 -0400, Andy Walls wrote: > > On Sat, 2009-04-04 at 14:28 +0200, Jean Delvare wrote: > > > Let card drivers probe for IR receiver devices and instantiate them if > > > found. Ultimately it would be better if we could stop probing > > > completely, but I suspect this won't be possible for all card types. > > > > > > There's certainly room for cleanups. For example, some drivers are > > > sharing I2C adapter IDs, so they also had to share the list of I2C > > > addresses being probed for an IR receiver. Now that each driver > > > explicitly says which addresses should be probed, maybe some addresses > > > can be dropped from some drivers. > > > > > > Also, the special cases in saa7134-i2c should probably be handled on a > > > per-board basis. This would be more efficient and less risky than always > > > probing extra addresses on all boards. I'll give it a try later. > > > > > > Signed-off-by: Jean Delvare <khali@linux-fr.org> > > > Cc: Mauro Carvalho Chehab <mchehab@infradead.org> > > > Cc: Hans Verkuil <hverkuil@xs4all.nl> > > > Cc: Andy Walls <awalls@radix.net> > > > Cc: Mike Isely <isely@pobox.com> > > > --- > > > linux/drivers/media/video/cx18/cx18-i2c.c | 30 ++ > > > linux/drivers/media/video/ivtv/ivtv-i2c.c | 31 ++ > > > linux/include/media/ir-kbd-i2c.h | 2 > > > 17 files changed, 284 insertions(+), 196 deletions(-) > > > > > > > > --- v4l-dvb.orig/linux/drivers/media/video/cx18/cx18-i2c.c 2009-04-04 10:53:15.000000000 +0200 > > > +++ v4l-dvb/linux/drivers/media/video/cx18/cx18-i2c.c 2009-04-04 10:58:36.000000000 +0200 > > > @@ -211,7 +211,32 @@ static struct i2c_algo_bit_data cx18_i2c > > > .timeout = CX18_ALGO_BIT_TIMEOUT*HZ /* jiffies */ > > > }; > > > > > > -/* init + register i2c algo-bit adapter */ > > > +static void init_cx18_i2c_ir(struct cx18 *cx) > > > +{ > > > + struct i2c_board_info info; > > > + /* The external IR receiver is at i2c address 0x34 (0x35 for > > > + reads). Future Hauppauge cards will have an internal > > > + receiver at 0x30 (0x31 for reads). In theory, both can be > > > + fitted, and Hauppauge suggest an external overrides an > > > + internal. > > > + > > > + That's why we probe 0x1a (~0x34) first. CB > > > + */ > > > + const unsigned short addr_list[] = { > > > + 0x1a, 0x18, 0x64, 0x30, > > > + I2C_CLIENT_END > > > + }; > > > > > > I think this is way out of date for cx18 based boards. The only IR chip > > I know of so far is the Zilog Z8F0811 sitting at 7 bit addresses > > 0x70-0x74. I guess 0x71 is the proper address for Rx. I'll let you > > know when I test. > > This address list comes from the ir-kbd-i2c driver. The cx18 driver > happens to use the same I2C adapter ID as the ivtv driver > (I2C_HW_B_CX2341X) and this is what the ir-kbd-i2c driver used to > decide which addresses to probe. As I don't know anything about the > hardware, I had to keep the new code compatible with the old one and > keep probing the same addresses. This is the i2cdetect output from my HVR-1600 - the only cx18 based card known or reported to have an IR chip: [root@morgan ~]# i2cdetect -l i2c-0 smbus SMBus PIIX4 adapter at 0b00 SMBus adapter i2c-1 i2c ivtv i2c driver #0 I2C adapter i2c-2 i2c cx18 i2c driver #0-0 I2C adapter i2c-3 i2c cx18 i2c driver #0-1 I2C adapter [root@morgan ~]# i2cdetect 2 WARNING! This program can confuse your I2C bus, cause data loss and worse! I will probe file /dev/i2c-2. I will probe address range 0x03-0x77. Continue? [Y/n] y 0 1 2 3 4 5 6 7 8 9 a b c d e f 00: -- -- -- -- -- -- -- -- -- -- -- -- -- 10: -- -- -- -- -- -- -- -- -- 19 -- -- -- -- -- -- 20: -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- 30: -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- 40: -- -- -- -- -- -- -- -- -- -- -- -- UU -- -- -- 50: 50 -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- 60: -- -- -- 63 -- -- -- -- -- -- -- -- -- -- -- -- 70: 70 71 72 73 -- -- -- -- [root@morgan ~]# i2cdetect 3 WARNING! This program can confuse your I2C bus, cause data loss and worse! I will probe file /dev/i2c-3. I will probe address range 0x03-0x77. Continue? [Y/n] y 0 1 2 3 4 5 6 7 8 9 a b c d e f 00: -- -- -- -- -- -- -- -- -- -- -- -- -- 10: -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- 20: -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- 30: -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- 40: -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- 50: -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- 60: -- UU -- -- -- -- -- -- -- -- -- -- -- -- -- -- 70: -- -- -- -- -- -- -- -- The Zilog is at 0x70-0x73. The standard IR Tx and RX addresses are 0x70 and 0x71 > Now, if you tell me that this list doesn't make sense for cx18 boards, > we can change it. I owe you the right address to probe. I think it is 0x71, but I want to double check. > As addresses 0x70-0x74 were not probed so far on cx18 > boards, I guess that IR support never worked for cx18 (at least not with > ir-kbd-i2c)? No, the lirc_i2c, lirc_pvr150, and lirc_zilog come in via the i2c subsystem. > Does ir-kbd-i2c support the Zilog Z8F0811 at all? > > If IR on the cx18 is not supported (by the ir-kbd-i2c driver) then I > can simplify my patch set and omit the cx18 entirely. The HVR-1600 could have been supported by ir-kbd-i2c. It's submission was redirected slightly here: http://lkml.org/lkml/2009/2/3/118 And deferred here: http://www.spinics.net/lists/linux-media/msg03883.html until your changes were done. > > > + memset(&info, 0, sizeof(struct i2c_board_info)); > > > + strlcpy(info.type, "ir-kbd", I2C_NAME_SIZE); > > > + > > > + /* The IR receiver device can be on either I2C bus */ > > > + if (i2c_new_probed_device(&cx->i2c_adap[0], &info, addr_list)) > > > + return; > > > + i2c_new_probed_device(&cx->i2c_adap[1], &info, addr_list); > > > +} > > > + > > > +/* init + register i2c adapters + instantiate IR receiver */ > > > int init_cx18_i2c(struct cx18 *cx) > > > { > > > int i, err; > > > @@ -279,6 +304,9 @@ int init_cx18_i2c(struct cx18 *cx) > > > err = i2c_bit_add_bus(&cx->i2c_adap[1]); > > > if (err) > > > goto err_del_bus_0; > > > + > > > + /* Instantiate the IR receiver device, if present */ > > > + init_cx18_i2c_ir(cx); > > > return 0; > > > > I have an I2C related question. If the cx18 or ivtv driver autoloads > > "ir-kbd-i2c" and registers an I2C client on the bus, does that preclude > > lirc_i2c, lirc_pvr150 or lirc_zilog from using the device? LIRC users > > may notice, if it does. > > I don't know anything about lirc_i2c, lirc_pvr150 or lirc_zilog. I tend > to ignore all the code that is neither in the Linux kernel tree nor in > the v4l-dvb tree. lirc_pvr150 has always been out of kernel and likely always will be. lirc_i2c and lirc_zilog, the stripped down version of lirc_pvr150, was submitted by Janne and Jarrod: http://lkml.org/lkml/2008/9/9/19 I do not know if it any of the lirc drivers made it in. There were lots of comments. > If you want me to answer this question, you'll have > to tell me what exactly these drivers are doing as far as I2C is > concerned. Do they instantiate I2C clients? Or do they do raw I2C > transfers? Do they check for address business before they do? On what > basis do they attach to I2C devices? Let me point you to the lirc_i2c.c file and I think you'll understand it faster than I could explain it. Here it is in Jarrod's patch submission: http://lkml.org/lkml/2008/9/9/3 Essentially, it 1. loads a bunch of bridge chip modules 2. creates and adds I2C driver 3. for every adapter that tries to "attach" the driver a. checks the i2c hw id b. probes a list of possible addresses based on the id > Please note that my conversion doesn't actually change anything to the > autoloading of the ir-kbd-i2c driver. The bridge drivers which were > loading ir-kbd-i2c (saa7134, cx23885, em28xx and cx88) still are. Those > which were not, still aren't. The ir-kbd-i2c driver doesn't include a > MODULE_ALIAS call as most I2C drivers do, to prevent udev from loading > this driver automatically. > > What my conversion changes is that an "ir-kbd" I2C device may be > instantiated if a probe is successful. This will make the address in > question show as busy (except to i2c-dev which only considers an > address as busy when a driver is bound to the device.) But that's about > it. OK. I didn't quite grok if the "ir-kbd" in i2c_new_probed_device() call would load the driver module or not. Tying up the address and making it unavailable for lirc modules is my concern. > > If that is the case, then we probably shouldn't autoload the ir-kbd > > module after the CX23418 i2c adapters are initialized. > > > > I'm not sure what's the best solution: > > > > 1. A module option to the cx18 driver to tell it to call > > init_cx18_i2c_ir() from cx18_probe() or not? (Easiest solution) > > Sounds perfectly sensible. I seem to remember that Hans Verkuil told me > he wanted something like this for ivtv. As a matter of fact, the > saa7134, em28xx and cx231xx already have such a module parameter > (disable_ir). Implementing the same for bttv, cx88, cx18, ivtv or any > other driver should be fairly trivial. Yes it's the most expedient thing to do. > > 2. Some involved programmatic way for IR device modules to query bridge > > drivers about what IR devices they may have, and on which I2C bus, and > > at what addresses to probe, and whether a driver/module has already > > claimed that device? (Gold plated solution) > > I'd rather name this the over-engineered solution. It's really looking > at the situation by the wrong end (that is, with the legacy i2c binding > model still in mind.) Bridge drivers know which IR receivers can be > present and at which address, it is up to them to instantiate the > appropriate I2C devices on the bus, possibly with platform data to help > the I2C drivers (be they ir-kbd-i2c, lirc or whatever.) This is exactly > what my code does. > > The fact that the same IR chip can be handled by 2 or more I2C drivers > is a bad idea to start with. Why the hell did we do that in the first > place? Accident of history? IR receive vs. IR blast/transmit? Why do we have ir-kbd-i2c.c trying to handle a laundry list of devices (somewhat like tvaudio)? User space apps such as MythTV and mplayer have specific support for LIRC. I guess LIRC's user space components abstract away a lot of the differences of various IR transmitters, receivers and remote controls to make things easier for application writers. Someone with an infradead.org email address can probably speak to LIRC's strengths and weaknesses better than I. I was wondering why we had ir-kbd-i2c. :) Mark Lord did say, in one of his posts to get the HVR-1600 support in ir-kbd-i2c, he didn't want the LIRC bloat for what he needed. > If you want a clean solution to the problem, it clearly starts > with getting rid of this mess and having each IR receiver chip on I2C > supported by exactly one I2C driver and make sure the driver in > question is in the Linux kernel tree. Spending time on any other "clean > solution" is wasting time IMHO. Makes sense to me. > > Still, note that it is totally possible to have several I2C drivers > support the same device. The new model supports this, just like the old > model did. I2C devices are instantiated by bridge drivers, which give > them a _name_. Several I2C drivers are allowed to support that chip > name, and the first one loaded will get to bind to the device. The > "ir-i2c" devices created by cx18, ivtv etc. can be requested by other > drivers than ir-kbd-i2c if you want to do that. Yes. OK. That's the part I didn't understand. So a hypothetical kernel ir-haup-zilog-i2c.ko module would look for devices with a name of "haup-zilog-ir", right? And the i2c adapter # and address can be used to differentiate different instances of the chip with the same name, so names don't have to be unique? Am I correct in my understanding? > This will require some > changes to lirc_i2c and friends, but at this point changes to these are > very needed anyway. Yes, it looks like LIRC's kernel space components that use I2C may get broken with upcoming I2C subsystem changes. > I hope I managed to clarify the situation a bit. A bit. I'm not totally clear, but once I play with it a little, I'll probably get it. Regards, Andy -- To unsubscribe from this list: send the line "unsubscribe linux-media" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On Sunday 05 April 2009 01:05:39 Jean Delvare wrote: > Hi Mike, > > On Sat, 4 Apr 2009 10:51:01 -0500 (CDT), Mike Isely wrote: > > Nacked-by: Mike Isely <isely@pobox.com> > > > > This will interfere with the alternative use of LIRC drivers (which > > work in more cases that ir-kbd). > > Why then is ir-kbd in the kernel tree and not LIRC drivers? > > > It will thus break some peoples' use of the driver. > > Do you think it will, or did you test and it actually does? If it > indeed breaks, please explain why, so that a solution can be found. > > > Also we have better information on what i2c addresses needed to > > be probed based on the model of the device > > This is excellent news. As I said in the header comment of the patch, > avoiding probing when we know what the IR receiver is and at which > address it sits is the way to go. Please send me all the information > you have and I'll be happy to add a patch to the series, that skips > probing whenever possible. Or write that patch yourself if you prefer. > > > - and some devices supported > > by this device are not from Hauppauge so you are making a too-strong > > assumption that IR should be probed this way in all cases. > > I didn't make any assumption, sorry. I simply copied the code from > ir-kbd-i2c. If my code does the wrong thing for some devices, that was > already the case before. And this will certainly be easier to fix after > my changes than before. > > On top of that, the "Hauppauge trick" is really only the order in which > the addresses are probed. Just because a specific order is better for > Hauppauge boards, doesn't mean it won't work for non-Hauppauge boards. > > > Also, unless > > ir-kbd has suddenly improved, this will not work at all for HVR-1950 > > class devices nor MCE type PVR-24xxx devices (different incompatible IR > > receiver). > > I'm sorry but you can't blame me for ir-kbd-i2c not supporting some > devices. I updated the driver to make use of the new binding model, but > that's about all I did. > > > This is why the pvrusb2 driver has never directly attempted to load > > ir-kbd. > > The pvrusb2 driver however abuses the bttv driver's I2C adapter ID > (I2C_HW_B_BT848) and was thus affected when ir-kbd-i2c is loaded. This > is the only reason why my patch touches the pvrusb2 driver. If you tell > me you want the ir-kbd-i2c driver to leave pvrusb2 alone, I can drop > all the related changes from my patch, that's very easy. Let's keep it simple: add a 'load_ir_kbd_i2c' module option for those drivers that did not autoload this module. The driver author can refine things later (I'll definitely will do that for ivtv). It will be interesting if someone can find out whether lirc will work at all once autoprobing is removed from i2c. If it isn't, then perhaps that will wake them up to the realization that they really need to move to the kernel. The new mechanism is the right way to do it: the adapter driver has all the information if, where and what IR is used and so should be the one to tell the kernel what to do. Attempting to autodetect and magically figure out what IR might be there is awkward and probably impossible to get right 100%. Hell, it's wrong already: if you have another board that already loads ir-kbd-i2c then if you load ivtv or pvrusb2 afterwards you get ir-kbd-i2c whether you like it or not, because ir-kbd-i2c will connect to your i2c adapter like a leech. So with the addition of a module option you at least give back control of this to the user. When this initial conversion is done I'm pretty sure we can improve ir-kbd-i2c to make it easier to let the adapter driver tell it what to do. So we don't need those horrible adapter ID tests and other magic that's going on in that driver. But that's phase two. Regards, Hans
On Sun, 5 Apr 2009 07:46:47 +0200 Hans Verkuil <hverkuil@xs4all.nl> wrote: > On Sunday 05 April 2009 01:05:39 Jean Delvare wrote: > > Hi Mike, > > > > On Sat, 4 Apr 2009 10:51:01 -0500 (CDT), Mike Isely wrote: > > > Nacked-by: Mike Isely <isely@pobox.com> > > > > > > This will interfere with the alternative use of LIRC drivers (which > > > work in more cases that ir-kbd). > > > > Why then is ir-kbd in the kernel tree and not LIRC drivers? > > > > > It will thus break some peoples' use of the driver. > > > > Do you think it will, or did you test and it actually does? If it > > indeed breaks, please explain why, so that a solution can be found. > > > > > Also we have better information on what i2c addresses needed to > > > be probed based on the model of the device > > > > This is excellent news. As I said in the header comment of the patch, > > avoiding probing when we know what the IR receiver is and at which > > address it sits is the way to go. Please send me all the information > > you have and I'll be happy to add a patch to the series, that skips > > probing whenever possible. Or write that patch yourself if you prefer. > > > > > - and some devices supported > > > by this device are not from Hauppauge so you are making a too-strong > > > assumption that IR should be probed this way in all cases. > > > > I didn't make any assumption, sorry. I simply copied the code from > > ir-kbd-i2c. If my code does the wrong thing for some devices, that was > > already the case before. And this will certainly be easier to fix after > > my changes than before. > > > > On top of that, the "Hauppauge trick" is really only the order in which > > the addresses are probed. Just because a specific order is better for > > Hauppauge boards, doesn't mean it won't work for non-Hauppauge boards. > > > > > Also, unless > > > ir-kbd has suddenly improved, this will not work at all for HVR-1950 > > > class devices nor MCE type PVR-24xxx devices (different incompatible IR > > > receiver). > > > > I'm sorry but you can't blame me for ir-kbd-i2c not supporting some > > devices. I updated the driver to make use of the new binding model, but > > that's about all I did. > > > > > This is why the pvrusb2 driver has never directly attempted to load > > > ir-kbd. > > > > The pvrusb2 driver however abuses the bttv driver's I2C adapter ID > > (I2C_HW_B_BT848) and was thus affected when ir-kbd-i2c is loaded. This > > is the only reason why my patch touches the pvrusb2 driver. If you tell > > me you want the ir-kbd-i2c driver to leave pvrusb2 alone, I can drop > > all the related changes from my patch, that's very easy. > > Let's keep it simple: add a 'load_ir_kbd_i2c' module option for those > drivers that did not autoload this module. The driver author can refine > things later (I'll definitely will do that for ivtv). > > It will be interesting if someone can find out whether lirc will work at all > once autoprobing is removed from i2c. If it isn't, then perhaps that will > wake them up to the realization that they really need to move to the > kernel. > > The new mechanism is the right way to do it: the adapter driver has all the > information if, where and what IR is used and so should be the one to tell > the kernel what to do. Attempting to autodetect and magically figure out > what IR might be there is awkward and probably impossible to get right > 100%. > > Hell, it's wrong already: if you have another board that already loads > ir-kbd-i2c then if you load ivtv or pvrusb2 afterwards you get ir-kbd-i2c > whether you like it or not, because ir-kbd-i2c will connect to your i2c > adapter like a leech. So with the addition of a module option you at least > give back control of this to the user. > > When this initial conversion is done I'm pretty sure we can improve > ir-kbd-i2c to make it easier to let the adapter driver tell it what to do. > So we don't need those horrible adapter ID tests and other magic that's > going on in that driver. But that's phase two. IMO, doing all those tricks to support an out-of-tree driver is the wrong approach. This is just postponing a more serious discussion about what should be done in kernel, in order to better support IR's. In the case of lirc, the userspace part has already an event interface. If the drivers are doing the right thing with their IR part, lirc can just use the event interface for all drivers. This seems to be the proper approach. From what I got from Andy and Mike's comments is that the real issue is that the IR kernel code is incomplete, broken or bad designed. So, several users and userspace apps don't rely on the kernel code but, instead, use lirc as an alternative. That's said, I propose a different approach: 1) Add some entry at feature-removal-schedule.txt posting a date to end support for out-of-tree I2C IR modules; 2) Start discussing with lirc people (and input/event maintainers if needed) about what is needed to properly support the required functionalities for a better lirc usage; 3) Propose a few API additions in order to support those functionalities; 4) apply IR patches on kernel to support the missing functionalities; 5) remove the support for out-of-tree i2c IR modules. Cheers, Mauro -- To unsubscribe from this list: send the line "unsubscribe linux-media" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On Sun, 2009-04-05 at 06:14 -0300, Mauro Carvalho Chehab wrote: > > IMO, doing all those tricks to support an out-of-tree driver is the wrong > approach. This is just postponing a more serious discussion about what should > be done in kernel, in order to better support IR's. The "tricks" were to not break the current user experience by saddling them with a kernel module that they may not want. (If the tricks are needed at all - I'm will test later today.) I agree that better in kernel infrastructure needs to be in place for IR. LIRC's kernel space infrastructure module, lirc_dev, probably isn't a bad model for support of IR devices. The individual LIRC modules for supporting specific sets of devices are the ones that have problems to varying degrees. > In the case of lirc, the userspace part has already an event interface. If the > drivers are doing the right thing with their IR part, lirc can just use the > event interface for all drivers. This seems to be the proper approach. Input event interfaces alone neglect IR blasters. > >From what I got from Andy and Mike's comments is that the real issue is that > the IR kernel code is incomplete, broken or bad designed. So, several users and > userspace apps don't rely on the kernel code but, instead, use lirc as an > alternative. There is at least one other motivation: LIRC also handles a number of other hardware interfaces that are not I2C: serial ports (/dev/ttySX), parallel port, USB, etc. I happen to use the lirc_mceusb2 module for my Phillips Home IR receiver/blaster (I'm not sure if the blaster works under linux.) > That's said, I propose a different approach: > > 1) Add some entry at feature-removal-schedule.txt posting a date to end support > for out-of-tree I2C IR modules; > > 2) Start discussing with lirc people (and input/event maintainers if needed) > about what is needed to properly support the required functionalities for a > better lirc usage; > > 3) Propose a few API additions in order to support those functionalities; > 4) apply IR patches on kernel to support the missing functionalities; The scope of a complete kernel IR infrastructure goes a bit beyond I2C bus devices that are only input devices. What's the scope of what you want to tackle here? I certainly don't want to reinvent something that's going to look just like the LIRC driver model: http://www.lirc.org/html/technical.html Which already has an infrastructure for IR driver module writers: http://www.lirc.org/html/technical.html#lirc_dev Do we just convert lirc_dev, lirc_i2c, and lirc_zilog to a cleaned up set of in kernel modules? lirc_i2c can certainly be broken up into several modules: 1 per supported device. Should these create an input device as well to maintain compatability with apps expecting an ir-kbd-i2c like function? Or do we split up ir-kbd-i2c into per device modules and in addition to the input event interface, have it register with the lirc_dev module? Do we leverage LIRC's lirc_dev infrastructure module at all? (I think it would be a waste of time not to do so.) Regards, Andy > 5) remove the support for out-of-tree i2c IR modules. > > Cheers, > Mauro > -- To unsubscribe from this list: send the line "unsubscribe linux-media" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Hi Andy, On Sat, 04 Apr 2009 21:50:08 -0400, Andy Walls wrote: > On Sun, 2009-04-05 at 00:51 +0200, Jean Delvare wrote: > > On Sat, 04 Apr 2009 09:42:09 -0400, Andy Walls wrote: > > > I think this is way out of date for cx18 based boards. The only IR chip > > > I know of so far is the Zilog Z8F0811 sitting at 7 bit addresses > > > 0x70-0x74. I guess 0x71 is the proper address for Rx. I'll let you > > > know when I test. > > > > This address list comes from the ir-kbd-i2c driver. The cx18 driver > > happens to use the same I2C adapter ID as the ivtv driver > > (I2C_HW_B_CX2341X) and this is what the ir-kbd-i2c driver used to > > decide which addresses to probe. As I don't know anything about the > > hardware, I had to keep the new code compatible with the old one and > > keep probing the same addresses. > > This is the i2cdetect output from my HVR-1600 - the only cx18 based card > known or reported to have an IR chip: > > [root@morgan ~]# i2cdetect -l > i2c-0 smbus SMBus PIIX4 adapter at 0b00 SMBus adapter > i2c-1 i2c ivtv i2c driver #0 I2C adapter > i2c-2 i2c cx18 i2c driver #0-0 I2C adapter > i2c-3 i2c cx18 i2c driver #0-1 I2C adapter > [root@morgan ~]# i2cdetect 2 > WARNING! This program can confuse your I2C bus, cause data loss and worse! > I will probe file /dev/i2c-2. > I will probe address range 0x03-0x77. > Continue? [Y/n] y > 0 1 2 3 4 5 6 7 8 9 a b c d e f > 00: -- -- -- -- -- -- -- -- -- -- -- -- -- > 10: -- -- -- -- -- -- -- -- -- 19 -- -- -- -- -- -- > 20: -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- > 30: -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- > 40: -- -- -- -- -- -- -- -- -- -- -- -- UU -- -- -- > 50: 50 -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- > 60: -- -- -- 63 -- -- -- -- -- -- -- -- -- -- -- -- > 70: 70 71 72 73 -- -- -- -- > [root@morgan ~]# i2cdetect 3 > WARNING! This program can confuse your I2C bus, cause data loss and worse! > I will probe file /dev/i2c-3. > I will probe address range 0x03-0x77. > Continue? [Y/n] y > 0 1 2 3 4 5 6 7 8 9 a b c d e f > 00: -- -- -- -- -- -- -- -- -- -- -- -- -- > 10: -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- > 20: -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- > 30: -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- > 40: -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- > 50: -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- > 60: -- UU -- -- -- -- -- -- -- -- -- -- -- -- -- -- > 70: -- -- -- -- -- -- -- -- > > The Zilog is at 0x70-0x73. The standard IR Tx and RX addresses are 0x70 > and 0x71 > > > > Now, if you tell me that this list doesn't make sense for cx18 boards, > > we can change it. > > I owe you the right address to probe. I think it is 0x71, but I want to > double check. Well, it doesn't really matter to me at this point. All I care about is that ir-kbd-i2c didn't support cx18 adapters before, so my patch can just ignore them. Support can be added later. More interestingly, if only one board needs to be supported for now and you have all the information about the IR receiver, then we simply don't need auto-detection of IR devices on cx18 at all. We can directly take the clean route of device declaration. As far as I know the cx18 does that very well for all other chips (tuner, decoder etc.) much like the ivtv driver does, so adding support for IR should be easy. > > As addresses 0x70-0x74 were not probed so far on cx18 > > boards, I guess that IR support never worked for cx18 (at least not with > > ir-kbd-i2c)? > > No, the lirc_i2c, lirc_pvr150, and lirc_zilog come in via the i2c > subsystem. > > > Does ir-kbd-i2c support the Zilog Z8F0811 at all? > > > > If IR on the cx18 is not supported (by the ir-kbd-i2c driver) then I > > can simplify my patch set and omit the cx18 entirely. Which I just did... > The HVR-1600 could have been supported by ir-kbd-i2c. > > It's submission was redirected slightly here: > > http://lkml.org/lkml/2009/2/3/118 > > And deferred here: > > http://www.spinics.net/lists/linux-media/msg03883.html > > until your changes were done. OK. Then let's indeed get my changes merged first, and then we can see the best way to add support for the HVR-1600 IR. > > > I have an I2C related question. If the cx18 or ivtv driver autoloads > > > "ir-kbd-i2c" and registers an I2C client on the bus, does that preclude > > > lirc_i2c, lirc_pvr150 or lirc_zilog from using the device? LIRC users > > > may notice, if it does. > > > > I don't know anything about lirc_i2c, lirc_pvr150 or lirc_zilog. I tend > > to ignore all the code that is neither in the Linux kernel tree nor in > > the v4l-dvb tree. > > lirc_pvr150 has always been out of kernel and likely always will be. Any valid reason? Out-of-free drivers are a pain for users :( > lirc_i2c and lirc_zilog, the stripped down version of lirc_pvr150, was > submitted by Janne and Jarrod: > > http://lkml.org/lkml/2008/9/9/19 > > I do not know if it any of the lirc drivers made it in. There were lots > of comments. I can't see any piece of lirc in the v4l-dvb repository so I'd guess nothing made it in. But at least if they are trying to get their code merged, that's a good thing. > > If you want me to answer this question, you'll have > > to tell me what exactly these drivers are doing as far as I2C is > > concerned. Do they instantiate I2C clients? Or do they do raw I2C > > transfers? Do they check for address business before they do? On what > > basis do they attach to I2C devices? > > Let me point you to the lirc_i2c.c file and I think you'll understand it > faster than I could explain it. Here it is in Jarrod's patch submission: > > http://lkml.org/lkml/2008/9/9/3 > > Essentially, it > > 1. loads a bunch of bridge chip modules Yeah, I've seen it. How ugly :( > 2. creates and adds I2C driver A legacy driver... which will break as soon as the legacy model is removed from the kernel - really soon now. > 3. for every adapter that tries to "attach" the driver > a. checks the i2c hw id A deprecated approach... > b. probes a list of possible addresses based on the id Something the new probing model doesn't easily support. All in all, the lirc code looks a lot like the old ir-kbd-i2c driver (before my changes.) So the exact same changes I just made, will have to be done there as well. It will be more difficult for an out-of-tree driver, but OTOH they can benefit from the work I just did. Nevertheless, I don't see ir-kbd-i2c and lirc_i2c (and possibly others) being developed in parallel as a viable solution for the future. It's about time to come up with a unified solution which supports all the hardware out there properly. I don't know much about input drivers and IR, but my initial feeling is that handling the remote control as a regular keyboard-like input is the way to go. After all a remote control is just a set of keys, nothing really original. Now, if the lirc interface offers something more that can't be implemented via the input interface, fine with me, let's keep it. But the user shouldn't have to manually chose between 2 drivers. > > Please note that my conversion doesn't actually change anything to the > > autoloading of the ir-kbd-i2c driver. The bridge drivers which were > > loading ir-kbd-i2c (saa7134, cx23885, em28xx and cx88) still are. Those > > which were not, still aren't. The ir-kbd-i2c driver doesn't include a > > MODULE_ALIAS call as most I2C drivers do, to prevent udev from loading > > this driver automatically. > > > > What my conversion changes is that an "ir-kbd" I2C device may be > > instantiated if a probe is successful. This will make the address in > > question show as busy (except to i2c-dev which only considers an > > address as busy when a driver is bound to the device.) But that's about > > it. > > OK. I didn't quite grok if the "ir-kbd" in i2c_new_probed_device() call > would load the driver module or not. Tying up the address and making it > unavailable for lirc modules is my concern. No, the ir-kbd-i2c driver module doesn't get loaded just because a device is created. That _would_ be the case if we added a MODULE_ALIAS() to the driver, and udev is running. But for now there is no MODULE_ALIAS() exactly because of existing alternatives to ir-kbd-i2c. That being said, looking at the lirc code, it _will_ be broken by my changes, because the i2c core won't let a legacy i2c device be created at the same address as a new-style i2c device. So users will have to load the bridge driver with disable_ir=1 if they want to use lirc_i2c. This sounds like a good reason to add a disable_ir module parameter to bridge drivers which don't have it yet. Then again, lirc_i2c will break in the next months with the removal of the legacy model anyway. It will have to be converted to the new binding model, at which point we can discuss a new temporary strategy to keep all ir modules usable. More on this below. > > > If that is the case, then we probably shouldn't autoload the ir-kbd > > > module after the CX23418 i2c adapters are initialized. > > > > > > I'm not sure what's the best solution: > > > > > > 1. A module option to the cx18 driver to tell it to call > > > init_cx18_i2c_ir() from cx18_probe() or not? (Easiest solution) > > > > Sounds perfectly sensible. I seem to remember that Hans Verkuil told me > > he wanted something like this for ivtv. As a matter of fact, the > > saa7134, em28xx and cx231xx already have such a module parameter > > (disable_ir). Implementing the same for bttv, cx88, cx18, ivtv or any > > other driver should be fairly trivial. > > Yes it's the most expedient thing to do. Done, I'll post the patch later today. > > > 2. Some involved programmatic way for IR device modules to query bridge > > > drivers about what IR devices they may have, and on which I2C bus, and > > > at what addresses to probe, and whether a driver/module has already > > > claimed that device? (Gold plated solution) > > > > I'd rather name this the over-engineered solution. It's really looking > > at the situation by the wrong end (that is, with the legacy i2c binding > > model still in mind.) Bridge drivers know which IR receivers can be > > present and at which address, it is up to them to instantiate the > > appropriate I2C devices on the bus, possibly with platform data to help > > the I2C drivers (be they ir-kbd-i2c, lirc or whatever.) This is exactly > > what my code does. > > > > The fact that the same IR chip can be handled by 2 or more I2C drivers > > is a bad idea to start with. Why the hell did we do that in the first > > place? > > Accident of history? IR receive vs. IR blast/transmit? Why do we have > ir-kbd-i2c.c trying to handle a laundry list of devices (somewhat like > tvaudio)? Yeah, that's another good question. But the ir-kbd-i2c driver is pretty small, so that's not necessarily a problem in practice. A single driver (cleanly) supporting several IR receiver devices shouldn't be a problem. > User space apps such as MythTV and mplayer have specific support for > LIRC. I guess LIRC's user space components abstract away a lot of the > differences of various IR transmitters, receivers and remote controls to > make things easier for application writers. Someone with an > infradead.org email address can probably speak to LIRC's strengths and > weaknesses better than I. Well, ir-kbd-i2c exposes the IR as a standard keyboard-like input, this seems like a reasonable abstraction to me too. But, just like you, I really don't know enough to compare. > I was wondering why we had ir-kbd-i2c. :) Mark Lord did say, in one of > his posts to get the HVR-1600 support in ir-kbd-i2c, he didn't want the > LIRC bloat for what he needed. > > > If you want a clean solution to the problem, it clearly starts > > with getting rid of this mess and having each IR receiver chip on I2C > > supported by exactly one I2C driver and make sure the driver in > > question is in the Linux kernel tree. Spending time on any other "clean > > solution" is wasting time IMHO. > > Makes sense to me. > > > Still, note that it is totally possible to have several I2C drivers > > support the same device. The new model supports this, just like the old > > model did. I2C devices are instantiated by bridge drivers, which give > > them a _name_. Several I2C drivers are allowed to support that chip > > name, and the first one loaded will get to bind to the device. The > > "ir-i2c" devices created by cx18, ivtv etc. can be requested by other > > drivers than ir-kbd-i2c if you want to do that. > > Yes. OK. That's the part I didn't understand. > > So a hypothetical kernel ir-haup-zilog-i2c.ko module would look for > devices with a name of "haup-zilog-ir", right? Depends. If your bridge driver creates an I2C named "haup-zilog-ir" and specifically expects ir-haup-zilog-i2c.ko and only ir-haup-zilog-i2c.ko to bind to it, then yes. But if ir-haup-zilog-i2c.ko if only an alternative to ir-kbd-i2c for this piece of hardware, then the bridge driver would create "ir-kbd" (or we might even make the name driver-neutral such as "v4l-ir") and both ir-kbd-i2c.ko and ir-haup-zilog-i2c.ko would use that device name (and the one loaded gets to grab it.) This is the easiest way to support ir-kbd-i2c and lirc_i2c in parallel in the new binding model. This might do for short term. > And the i2c adapter # > and address can be used to differentiate different instances of the chip > with the same name, so names don't have to be unique? Am I correct in > my understanding? Yes, the device name (e.g. "ir-kbd") says what device type this is, it isn't a unique identifier (you can have more than one IR receiver in a given system). The unique ID is made of the i2c adapter # and the main i2c device address. > > This will require some > > changes to lirc_i2c and friends, but at this point changes to these are > > very needed anyway. > > Yes, it looks like LIRC's kernel space components that use I2C may get > broken with upcoming I2C subsystem changes. Yes, it will be broken.
Hi Hans, On Sun, 5 Apr 2009 07:46:47 +0200, Hans Verkuil wrote: > Let's keep it simple: add a 'load_ir_kbd_i2c' module option for those > drivers that did not autoload this module. The driver author can refine > things later (I'll definitely will do that for ivtv). I'd rather name the parameter "disable_ir", to make it consistent with what other bridge drivers already use, and also because what the parameter really does is preventing I2C devices from being instantiated, _not_ preventing the ir-kbd-i2c module from loading. I have a patch doing that already, it's pretty simple, I'll post it in a minute. > It will be interesting if someone can find out whether lirc will work at all > once autoprobing is removed from i2c. If it isn't, then perhaps that will > wake them up to the realization that they really need to move to the > kernel. lirc_i2c will break, it still uses the legacy binding model. That's what you get for living outside the kernel tree... Upgrading it shouldn't be too hard, given that the difficult part is to update the bridge drivers and I've already taken care of this part (although lirc_i2c might need to probe more addresses than ir-kbd-i2c does). > The new mechanism is the right way to do it: the adapter driver has all the > information if, where and what IR is used and so should be the one to tell > the kernel what to do. Attempting to autodetect and magically figure out > what IR might be there is awkward and probably impossible to get right > 100%. I wholeheartedly agree. > Hell, it's wrong already: if you have another board that already loads > ir-kbd-i2c then if you load ivtv or pvrusb2 afterwards you get ir-kbd-i2c > whether you like it or not, because ir-kbd-i2c will connect to your i2c > adapter like a leech. So with the addition of a module option you at least > give back control of this to the user. Totally correct. The current mess^Wmodel vaguely works when a single TV card is present, but mixing different TV cards in the same system would certainly break. Admittedly this is probably not a very common case, I guess the vast majority of users have a single TV card. > When this initial conversion is done I'm pretty sure we can improve > ir-kbd-i2c to make it easier to let the adapter driver tell it what to do. > So we don't need those horrible adapter ID tests and other magic that's > going on in that driver. But that's phase two. Please note that my conversion _already_ no longer makes any use of adapter IDs. What it still does is probing, except for a few selected cards (AVerMedia Cardbus and MSI TV@nywhere Plus). The idea is clearly to turn probing into a fallback and use per-card data for IR device instantiation the first choice for as many cards as possible in the future. Updated patch set is available at: http://jdelvare.pck.nerim.net/linux/ir-kbd-i2c/ Changes since previous version: * Dropped cx18 changes on request by Andy Walls. * Added disable_ir module parameter to all bridge drivers which didn't have it.
Hi Mike, Selected answers, as most points have already been discussed elsewhere meanwhile... On Sat, 4 Apr 2009 18:29:35 -0500 (CDT), Mike Isely wrote: > On Sun, 5 Apr 2009, Jean Delvare wrote: > > This is excellent news. As I said in the header comment of the patch, > > avoiding probing when we know what the IR receiver is and at which > > address it sits is the way to go. Please send me all the information > > you have and I'll be happy to add a patch to the series, that skips > > probing whenever possible. Or write that patch yourself if you prefer. > > I have samples of most of the devices in question and can code proper > I2C addresses for each of those for each resident I2C client. The > pvrusb2 driver's device attribute structure already has allowance for > specification of the correct I2C addresses to use for chips specific to > each device (part of the v4l2-subdev rework I recently did). Right now > the driver has built in defaults, and if a particular model needs > something else, it can be overridden as part of the device's profile in > pvrusb2-devattr.c. Good. Declaring the right IR receiver device based on board information is clearly the way to go. > > I didn't make any assumption, sorry. I simply copied the code from > > ir-kbd-i2c. If my code does the wrong thing for some devices, that was > > already the case before. And this will certainly be easier to fix after > > my changes than before. > > No, I think the point you are missing is that by moving that logic from > ir-kbd-i2c into each driver (e.g. pvrusb2) you are moving logic that > *might* be executed into a spot where it *will* be executed. Yes, you are right. > Remember > that the pvrusb2 driver did not autoload ir-kbd-i2c before this patch. > Before this change, a user could elect not to use ir-i2c-kbd simply by > not loading it. Which was pretty fragile because another bridge driver could still have loaded it. > The pvrusb2 driver didn't request it, because the > intent all along there is that the user makes the choice by loading the > desired driver. Which simply underlines how weird the current situation is... Forcing the choice on the user is pretty bad from a usability perspective. > Now with this change, the pvrusb2 driver is going to > attempt to load ir-kbd-i2c where asked for or not. Not exactly, only the device will be the instantiated, the drivers won't be loaded, but the result is indeed the same: it's getting in lirc_i2c's way if that's what the user wants to use. > And if not, the > resulting binding will prevent lirc_i2c from later on loading. If the > user had been loading lirc_i2c previously, this will cause his/her usage > of IR to break. No, it's not perfect, but it worked. I'm all for > improving things, but not by removing an ability that people are using. I have just added a disable_ir module parameter to work around this issue. Set it to 1 and no "ir-kbd" device will be instantiated, letting lirc_i2c do whatever it wants with the IR receiver device. You might argue that this is still a regression because the user now has to pass an extra parameter to get the same result as before, but OTOH lirc_i2c will need changes pretty soon anyway, its behavior will have to be changed and the users will notice. There's no way we can go from the current weird situation to a sane situation without changing the (unfortunate) user habits. > (...) > I really don't want to throw rocks here; it's always better to work out > the solution than simply block any changes at all. I realize that > something has to be done here in order to keep ir-kbd-i2c viable as a > choice for users of the pvrusb2 driver. To that end, how about this > strategy: > > 1. Just drop the part of the patch for the pvrusb2 driver and get the > rest merged. Yes, I realize that this will break use of ir-kbd-i2c in > cooperation with the pvrusb2 driver. As Mauro already said: not acceptable. Breaking an in-tree driver (ir-kbd-i2c) is worse than breaking an out-of-tree driver (lirc_i2c) regardless of the respective number of users or usefulness of these drivers. > 2. Once ir-kbd-i2c has been updated, I will push another set of changes > into the pvrusb2 driver that will make it usable there. Basically what > I have in mind is similar to what you tried but I'm going to integrate > it with the device profiles so that it can be appropriately loaded based > on device model, with the correct I2C address in each case. And most > importantly, I will add a module option to enable/disable loading or > ir-kbd-i2c. This will fix my main objection, since then it will still > allow lirc to be usable, for now... This sounds like a good idea. > 3. I'd like to fix the "abuse" you mention regarding I2C_HW_B_BT848. > I'm unclear on what the cleanest solution is there, but if you like to > (a) point at some documentation, (b) describe what I should do, or (c) > suggest a patch, I will work to deal with the problem. Ideally these adapter IDs will no longer be needed within a couple weeks, so it's probably better to leave this alone and just let it die in silence.
On Sun, Apr 05, 2009 at 07:46:47AM +0200, Hans Verkuil wrote: > > Let's keep it simple: add a 'load_ir_kbd_i2c' module option for those > drivers that did not autoload this module. The driver author can refine > things later (I'll definitely will do that for ivtv). > > It will be interesting if someone can find out whether lirc will work at all > once autoprobing is removed from i2c. If it isn't, then perhaps that will > wake them up to the realization that they really need to move to the > kernel. I would guess that it won't work. There is an effort to merge lirc. It's currently stalled though. A git tree is available at git://git.wilsonet.com/linux-2.6-lirc.git Jared Wilson and I were working on it (mainly last september). Since the IR on the HD PVR is also driven by the same zilog chip as on other hauppauge devices I'll take of lirc_zilog. Help converting the i2c drivers to the new i2c model is welcome. General cleanup of lirc to make it ready for mainline is of course wellcome too. Janne -- To unsubscribe from this list: send the line "unsubscribe linux-media" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Hi Janne, On Sun, 5 Apr 2009 16:37:49 +0200, Janne Grunau wrote: > On Sun, Apr 05, 2009 at 07:46:47AM +0200, Hans Verkuil wrote: > > > > Let's keep it simple: add a 'load_ir_kbd_i2c' module option for those > > drivers that did not autoload this module. The driver author can refine > > things later (I'll definitely will do that for ivtv). > > > > It will be interesting if someone can find out whether lirc will work at all > > once autoprobing is removed from i2c. If it isn't, then perhaps that will > > wake them up to the realization that they really need to move to the > > kernel. > > I would guess that it won't work. There is an effort to merge lirc. It's > currently stalled though. A git tree is available at > > git://git.wilsonet.com/linux-2.6-lirc.git I tried to clone this but it failed: git.wilsonet.com[0: 72.93.233.4]: errno=Connection timed out fatal: unable to connect a socket (Connection timed out) > Jared Wilson and I were working on it (mainly last september). Since the > IR on the HD PVR is also driven by the same zilog chip as on other > hauppauge devices I'll take of lirc_zilog. Help converting the i2c > drivers to the new i2c model is welcome. General cleanup of lirc to make > it ready for mainline is of course wellcome too. I will happily help you with the i2c side of things. Without an access to your git tree however, I don't know the latest state of your code. And I can't see any lirc_zilog module in the CVS repository of lirc either. But if you show me the i2c code you're worried about, I'll let you know what I think about it.
On Sun, Apr 05, 2009 at 06:37:43PM +0200, Jean Delvare wrote: > Hi Janne, > > On Sun, 5 Apr 2009 16:37:49 +0200, Janne Grunau wrote: > > On Sun, Apr 05, 2009 at 07:46:47AM +0200, Hans Verkuil wrote: > > > > > > Let's keep it simple: add a 'load_ir_kbd_i2c' module option for those > > > drivers that did not autoload this module. The driver author can refine > > > things later (I'll definitely will do that for ivtv). > > > > > > It will be interesting if someone can find out whether lirc will work at all > > > once autoprobing is removed from i2c. If it isn't, then perhaps that will > > > wake them up to the realization that they really need to move to the > > > kernel. > > > > I would guess that it won't work. There is an effort to merge lirc. It's > > currently stalled though. A git tree is available at > > > > git://git.wilsonet.com/linux-2.6-lirc.git > > I tried to clone this but it failed: > > git.wilsonet.com[0: 72.93.233.4]: errno=Connection timed out > fatal: unable to connect a socket (Connection timed out) hmm, getting that here too. I'll send Jarod a mail. You can use my clone instead: git://git.jannau.net/linux-2.6-lirc.git > I will happily help you with the i2c side of things. Without an access > to your git tree however, I don't know the latest state of your code. > And I can't see any lirc_zilog module in the CVS repository of lirc > either. But if you show me the i2c code you're worried about, I'll let > you know what I think about it. lirc_zilog or lirc_pvr is not in lirc's cvs since it requires a 'firmware' for mapping keys to sequences for the ir blaster. Janne -- To unsubscribe from this list: send the line "unsubscribe linux-media" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Hi Janne, On Sun, 2009-04-05 at 16:37 +0200, Janne Grunau wrote: > On Sun, Apr 05, 2009 at 07:46:47AM +0200, Hans Verkuil wrote: > > > > Let's keep it simple: add a 'load_ir_kbd_i2c' module option for those > > drivers that did not autoload this module. The driver author can refine > > things later (I'll definitely will do that for ivtv). > > > > It will be interesting if someone can find out whether lirc will work at all > > once autoprobing is removed from i2c. If it isn't, then perhaps that will > > wake them up to the realization that they really need to move to the > > kernel. > > I would guess that it won't work. There is an effort to merge lirc. It's > currently stalled though. Perhaps you and Jarrod and Christopher have already discussed this, but... Instead of trying to push all of the LIRC kernel components through in one big patch set, perhaps it would be easier to just get the lirc_dev and any other needed infrastructure components in first. If one focuses on satisfying the LKML comments to lirc_dev and the Makefile to get that kernel module in the kernel, then, at least for video card hosted IR devices, there is an infrastructure to which to hook new or rewritten i2c IR driver modules. > A git tree is available at > > git://git.wilsonet.com/linux-2.6-lirc.git > > Jared Wilson and I were working on it (mainly last september). Since the > IR on the HD PVR is also driven by the same zilog chip as on other > hauppauge devices I'll take of lirc_zilog. Help converting the i2c > drivers to the new i2c model is welcome. General cleanup of lirc to make > it ready for mainline is of course wellcome too. I can help with this. I'm mainly concerned with lirc_dev, lirc_i2c (for Rx only use of the zilog at 0x71), lirc_zilog, and lirc_mceusb2. That's because, of course, I have devices that use those modules. :) lirc_dev and the API header would be my first priority, if you need help. Did anyone consolidate all the comments from the LKML on Jarrod's patch submission? Regards, Andy > Janne -- To unsubscribe from this list: send the line "unsubscribe linux-media" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On Sun, 2009-04-05 at 15:08 +0200, Jean Delvare wrote: > Hi Andy, > > > > If IR on the cx18 is not supported (by the ir-kbd-i2c driver) then I > > > can simplify my patch set and omit the cx18 entirely. > > Which I just did... > > > The HVR-1600 could have been supported by ir-kbd-i2c. > > > > It's submission was redirected slightly here: > > > > http://lkml.org/lkml/2009/2/3/118 > > > > And deferred here: > > > > http://www.spinics.net/lists/linux-media/msg03883.html > > > > until your changes were done. > > OK. Then let's indeed get my changes merged first, and then we can see > the best way to add support for the HVR-1600 IR. OK. I'll test your change anyway if I can. > > lirc_pvr150 has always been out of kernel and likely always will be. > > Any valid reason? Out-of-free drivers are a pain for users :( Well, like many of the lirc modules, it's a little kludged. The main problem is this: 1. lirc_pvr150, in the past, needed to make a direct call into the ivtv module to reset the IR chip, if it detected that the chip was hung up. That's why it tries to load the ivtv module, to make sure that symbol is in the kernel. This could cause problems, if it was a Z8 chip that was supported by some other bridge driver. I wrote a patch for lirc_pvr150 for cx18 devices for users who needed it. lirc_zilog is the cut down version of lirc_pvr150 module that was submitted in the patchset to the LKML, and no longer has the reset logic. The reset logic is not needed anymore as far as I can tell, and thus the cx18 specific patch is probably irrelevant for lirc_zilog. Other weird things include: 2. In lirc_pvr150 and lirc_zilog, both the IR Rx and IR Tx support are in one module, which is a break from the normal LIRC driver modules that keep those functions separate. This was done for the sake of detecting if the chip had hung up and to call the reset logic, AFAICT. 3. lirc_pvr150 and lirc_zilog have an IR blaster "firmware" image that is really an encoding of a bunch of captured sequences between the Windows driver and the Z8F0811 chip. It allows the lirc_zilog or lirc_pvr150 driver to do IR blasting by essentially performing a replay attack on the Z8F0811. Since the Zilog EULA that comes with the Hauppauge Windows IR driver for the Z8F0811 is pretty draconian, replaying captured snoops is probably the best that can be done legally to stimulate the microcontroller IR Tx code in the Z8 as delivered. Regards, Andy -- To unsubscribe from this list: send the line "unsubscribe linux-media" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On Sun, Apr 05, 2009 at 01:39:33PM -0400, Andy Walls wrote: > On Sun, 2009-04-05 at 16:37 +0200, Janne Grunau wrote: > > > > I would guess that it won't work. There is an effort to merge lirc. It's > > currently stalled though. > > Perhaps you and Jarrod and Christopher have already discussed this, > but... > > Instead of trying to push all of the LIRC kernel components through in > one big patch set, perhaps it would be easier to just get the lirc_dev > and any other needed infrastructure components in first. > > If one focuses on satisfying the LKML comments to lirc_dev and the > Makefile to get that kernel module in the kernel, then, at least for > video card hosted IR devices, there is an infrastructure to which to > hook new or rewritten i2c IR driver modules. I guess lkml would NAK patches adding infrastructure only bits but we will probably for the next patchset concentrate on a few lirc drivers. Christopher doesn't participate in the merge attempt. > > A git tree is available at > > > > git://git.wilsonet.com/linux-2.6-lirc.git > > > > Jared Wilson and I were working on it (mainly last september). Since the > > IR on the HD PVR is also driven by the same zilog chip as on other > > hauppauge devices I'll take of lirc_zilog. Help converting the i2c > > drivers to the new i2c model is welcome. General cleanup of lirc to make > > it ready for mainline is of course wellcome too. > > I can help with this. I'm mainly concerned with lirc_dev, lirc_i2c (for > Rx only use of the zilog at 0x71), lirc_zilog, and lirc_mceusb2. That's > because, of course, I have devices that use those modules. :) I have devices for lirc_zilog (which should probably be merged with lirc_i2c) and lirc serial. Jarod has at least mce usb and imon devices. That are probably the devices we'll concentrate on the next submission. > lirc_dev and the API header would be my first priority, if you need > help. Did anyone consolidate all the comments from the LKML on Jarrod's > patch submission? no and I lost track which comments were already handled. Janne -- To unsubscribe from this list: send the line "unsubscribe linux-media" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Jean: More comments interspersed below... On Sun, 5 Apr 2009, Jean Delvare wrote: [...] > > > Remember > > that the pvrusb2 driver did not autoload ir-kbd-i2c before this patch. > > Before this change, a user could elect not to use ir-i2c-kbd simply by > > not loading it. > > Which was pretty fragile because another bridge driver could still have > loaded it. Well I didn't say it was perfect. But this has been a valid use-case and I know people have been using this it this way. > > > The pvrusb2 driver didn't request it, because the > > intent all along there is that the user makes the choice by loading the > > desired driver. > > Which simply underlines how weird the current situation is... Forcing > the choice on the user is pretty bad from a usability perspective. What's worse is taking away a choice when the alternatives are not interchangeable. In this case we have two different drivers with two different interfaces which do not cover the same range of hardware. The lirc case happens to work quite well in MythTV. In addition, lirc covers hardware situations involving the pvrusb2 driver that ir-kbd-i2c does not. Also, lirc makes it possible to userspace map the underlying IR codes to keybindings and associate multiple different remotes - all of that is apparently hardcoded in ir-kbd-i2c. Wierd or not, your change makes it hard(er) on those who legitimately wish to use lirc. Here's an interesting summary: If fact, the only pvrusb2-driven hardware from Hauppauge that is known to work with ir-kbd-i2c are the old 29xxx and 24xxx model series (not the "MCE" series). Those devices are out of production, AFAIK. The current devices being sold by Hauppauge don't work at all with ir-kbd-i2c and probably never will. Perhaps one can conclude that there hasn't been a lot of pressure (that I know about) to deal with updating / enhancing / replacing ir-kbd-i2c because lirc happens to be filling the niche better in many cases. > > > Now with this change, the pvrusb2 driver is going to > > attempt to load ir-kbd-i2c where asked for or not. > > Not exactly, only the device will be the instantiated, the drivers > won't be loaded, but the result is indeed the same: it's getting in > lirc_i2c's way if that's what the user wants to use. Well yes. The binding of the address is enough to mess things up. > > > And if not, the > > resulting binding will prevent lirc_i2c from later on loading. If the > > user had been loading lirc_i2c previously, this will cause his/her usage > > of IR to break. No, it's not perfect, but it worked. I'm all for > > improving things, but not by removing an ability that people are using. > > I have just added a disable_ir module parameter to work around this > issue. Set it to 1 and no "ir-kbd" device will be instantiated, letting > lirc_i2c do whatever it wants with the IR receiver device. > > You might argue that this is still a regression because the user now > has to pass an extra parameter to get the same result as before, but > OTOH lirc_i2c will need changes pretty soon anyway, its behavior will > have to be changed and the users will notice. There's no way we can go > from the current weird situation to a sane situation without changing > the (unfortunate) user habits. Yes, I would argue that this is still a regression. I really think this should be an *enable* switch in order to match previous behavior. In the long term I agree that this is less than optimal / undesirable, however in the long term none of it is going to matter anyway, since it looks like lirc without further changes is going to have problems here. Once (if) lirc is changed to use the new binding model then this whole argument becomes moot. Given that, then anything done here is a short term strategy, designed to avoid a regression, so I would much rather this be an enable not a disable switch. > > > (...) > > I really don't want to throw rocks here; it's always better to work out > > the solution than simply block any changes at all. I realize that > > something has to be done here in order to keep ir-kbd-i2c viable as a > > choice for users of the pvrusb2 driver. To that end, how about this > > strategy: > > > > 1. Just drop the part of the patch for the pvrusb2 driver and get the > > rest merged. Yes, I realize that this will break use of ir-kbd-i2c in > > cooperation with the pvrusb2 driver. > > As Mauro already said: not acceptable. Breaking an in-tree driver > (ir-kbd-i2c) is worse than breaking an out-of-tree driver (lirc_i2c) > regardless of the respective number of users or usefulness of these > drivers. *I* didn't break ir-kbd-i2c. And now because of what has effectively "broken" (viewed in just the wrong light) you are asking me to accept changes in the pvrusb2 driver to make up the difference? Normally I wouldn't object, except those proposed changes in turn will break another driver (lirc). That is why I nacked the change. I don't care if the driver is in-tree or out of tree, I am not going to accept a change that breaks without recourse a known use-case that people are using. I consider Mauro's reasoning flawed here. What I am suggesting for (1) does not "break" ir-kbd-i2c. Yes, it impacts its use but only within the pvrusb2 driver, and only until I get (2) implemented, which unless something comes up I plan on dealing with today. So (1) should be acceptable. > > > 2. Once ir-kbd-i2c has been updated, I will push another set of changes > > into the pvrusb2 driver that will make it usable there. Basically what > > I have in mind is similar to what you tried but I'm going to integrate > > it with the device profiles so that it can be appropriately loaded based > > on device model, with the correct I2C address in each case. And most > > importantly, I will add a module option to enable/disable loading or > > ir-kbd-i2c. This will fix my main objection, since then it will still > > allow lirc to be usable, for now... > > This sounds like a good idea. > > > 3. I'd like to fix the "abuse" you mention regarding I2C_HW_B_BT848. > > I'm unclear on what the cleanest solution is there, but if you like to > > (a) point at some documentation, (b) describe what I should do, or (c) > > suggest a patch, I will work to deal with the problem. > > Ideally these adapter IDs will no longer be needed within a couple > weeks, so it's probably better to leave this alone and just let it die > in silence. > OK, no problem. I'm happy with letting it die :-) -Mike
On Sun, 5 Apr 2009, Hans Verkuil wrote: [...] > > Let's keep it simple: add a 'load_ir_kbd_i2c' module option for those > drivers that did not autoload this module. The driver author can refine > things later (I'll definitely will do that for ivtv). Yes, and I will do the same for pvrusb2. Simple is good. > > It will be interesting if someone can find out whether lirc will work at all > once autoprobing is removed from i2c. If it isn't, then perhaps that will > wake them up to the realization that they really need to move to the > kernel. It's probably going to break, and that will wake them up. There's no choice otherwise. > > The new mechanism is the right way to do it: the adapter driver has all the > information if, where and what IR is used and so should be the one to tell > the kernel what to do. Attempting to autodetect and magically figure out > what IR might be there is awkward and probably impossible to get right > 100%. > > Hell, it's wrong already: if you have another board that already loads > ir-kbd-i2c then if you load ivtv or pvrusb2 afterwards you get ir-kbd-i2c > whether you like it or not, because ir-kbd-i2c will connect to your i2c > adapter like a leech. So with the addition of a module option you at least > give back control of this to the user. Yes, I know this is a possibility. It sucks and long term the new mechanism is the solution. I don't want anyone to think I am against the new mechanism. I'm for it. But I'd like to minimize the damage potential on the way there. > > When this initial conversion is done I'm pretty sure we can improve > ir-kbd-i2c to make it easier to let the adapter driver tell it what to do. > So we don't need those horrible adapter ID tests and other magic that's > going on in that driver. But that's phase two. My impression (at least for pvrusb2-driven devices) is that the later IR receivers require a completely different driver to work properly; one can't just bolt additional features into ir-kbd-i2c for this. In lirc's case, a different driver is in fact used. But you know this already. I haven't looked at ir-kbd-i2c in a while, but one other thing I seriously did not like about it was that the mapping of IR codes to key events was effectively hardcoded in the driver. That makes it difficult to make the same driver work with different kinds of remotes. Even if the bridge driver (e.g. pvrusb2) were to supply such a map, that's still wrong because it's within the realm of possibility that the user might actually want to use a different remote transmitter (provided it is compatible enough to work with the receiver hardware). The lirc architecture solves this easily via a conf file in userspace. In fact, lirc can map multiple mappings to a single receiver, permitting it to work concurrently with more than one remote. But is such a thing even possible with ir-kbd-i2c? I know this is one reason people tend to choose lirc. -Mike
On Sun, 2009-04-05 at 20:31 +0200, Janne Grunau wrote: > On Sun, Apr 05, 2009 at 01:39:33PM -0400, Andy Walls wrote: > > If one focuses on satisfying the LKML comments to lirc_dev and the > > Makefile to get that kernel module in the kernel, then, at least for > > video card hosted IR devices, there is an infrastructure to which to > > hook new or rewritten i2c IR driver modules. > > I guess lkml would NAK patches adding infrastructure only bits but we > will probably for the next patchset concentrate on a few lirc drivers. > Christopher doesn't participate in the merge attempt. Oh, OK. > > > A git tree is available at > > > > > > git://git.wilsonet.com/linux-2.6-lirc.git > > > > > > Jared Wilson and I were working on it (mainly last september). Since the > > > IR on the HD PVR is also driven by the same zilog chip as on other > > > hauppauge devices I'll take of lirc_zilog. Help converting the i2c > > > drivers to the new i2c model is welcome. General cleanup of lirc to make > > > it ready for mainline is of course wellcome too. > > > > I can help with this. I'm mainly concerned with lirc_dev, lirc_i2c (for > > Rx only use of the zilog at 0x71), lirc_zilog, and lirc_mceusb2. That's > > because, of course, I have devices that use those modules. :) > > I have devices for lirc_zilog (which should probably be merged with > lirc_i2c) Hmmm. Following Jean's reasoning, that may be the wrong way to go, if you want to avoid probing. A module to handle each specific type of I2C IR chip, splitting up lirc_i2c and leaving lirc_zilog as is, may be better in the long run. I'd personally leave lirc_zilog separate since it handles Tx and RX for one specific chip, and lirc_i2c is Rx only for a number of chips. Perhaps dropping Rx support for the Zilog Z8 in lirc_i2c and then modifying lirc_zilog to still do Rx, even if the "firmware" wasn't available for Tx, is a better way to go. > and lirc serial. Jarod has at least mce usb and imon devices. > That are probably the devices we'll concentrate on the next submission. OK. > > lirc_dev and the API header would be my first priority, if you need > > help. Did anyone consolidate all the comments from the LKML on Jarrod's > > patch submission? > > no and I lost track which comments were already handled. Hmm. Well good luck. Let me know if I can help. I have 2 cards with the Zilog and a USB unit that is supported by lirc_mceusb2. Regards, Andy > Janne -- To unsubscribe from this list: send the line "unsubscribe linux-media" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On Sun, 2009-04-05 at 16:05 +0200, Jean Delvare wrote: > Hi Hans, Hi Jean, > Updated patch set is available at: > http://jdelvare.pck.nerim.net/linux/ir-kbd-i2c/ > --- v4l-dvb.orig/linux/drivers/media/video/ivtv/ivtv-i2c.c 2009-04-04 10:53:08.000000000 +0200 +++ v4l-dvb/linux/drivers/media/video/ivtv/ivtv-i2c.c 2009-04-04 10:58:36.000000000 +0200 [snip] - + const unsigned short addr_list[] = { + 0x1a, 0x18, 0x64, 0x30, + I2C_CLIENT_END + }; [snip] I just noticed you're missing address 0x71 for ivtv. At least some PVR-150 boards have a Zilog chip at that address. Regards, Andy -- To unsubscribe from this list: send the line "unsubscribe linux-media" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On Sun, 2009-04-05 at 13:33 -0500, Mike Isely wrote: > Also, lirc makes it possible to userspace map the underlying > IR codes to keybindings and associate multiple different remotes - all > of that is apparently hardcoded in ir-kbd-i2c. Yes. My first remote for my PC was an HP OEM'ed and customized unit. LIRC's text configuration files made support for the remote's non-standard keys a task for 'vi' and not 'make'. Key mappings belong in configuration files - not hard coded in the kernel. > Wierd or not, your > change makes it hard(er) on those who legitimately wish to use lirc. > Here's an interesting summary: > > If fact, the only pvrusb2-driven hardware from Hauppauge that is known > to work with ir-kbd-i2c are the old 29xxx and 24xxx model series (not > the "MCE" series). Those devices are out of production, AFAIK. The > current devices being sold by Hauppauge don't work at all with > ir-kbd-i2c and probably never will. > > Perhaps one can conclude that there hasn't been a lot of pressure (that > I know about) to deal with updating / enhancing / replacing ir-kbd-i2c > because lirc happens to be filling the niche better in many cases. Here is where LIRC may be its own worst enemy. LIRC has filled some shortcomings in the kernel for support of IR device functions for so long (LWN says LIRC is 10 years old), that large numbers of users have come to depend on its operation, while at the same time apparently removing impetus for doing much to update the in kernel IR device support. Some of the discussion on the LKML about why an input layer device is sufficient for handling most, but not all cases is interesting: http://lkml.org/lkml/2008/9/11/63 Gerd also enlightens us on why we have ir-kdb-i2c in kernel in that post. It appears that ir-kbd-i2c was written to avoid using LIRC for TV card I2C IR devices for the most common use cases. As such, it's a 90% (80%, 70% ?) solution: no blasting, no raw IR parsing for unknown protocols, only the most "common" remotes supported, and, of course, no support for non-I2C devices. My needs don't fit that unfortunately. I need IR blasting, an uncommon remote supported, and both USB and I2C IR device support. Regards, Andy -- To unsubscribe from this list: send the line "unsubscribe linux-media" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On Sun, 05 Apr 2009 14:58:17 -0400, Andy Walls wrote: > On Sun, 2009-04-05 at 20:31 +0200, Janne Grunau wrote: > > I have devices for lirc_zilog (which should probably be merged with > > lirc_i2c) > > Hmmm. Following Jean's reasoning, that may be the wrong way to go, if > you want to avoid probing. A module to handle each specific type of I2C > IR chip, splitting up lirc_i2c and leaving lirc_zilog as is, may be > better in the long run. This really doesn't matter. With the new binding model, probing is under control. You can do probing on some cards and not others, and you can probe some addresses and not others. And one i2c drivers can cleanly support more than one device type. What should be considered to decide whether two devices should be supported by the same driver or not, is how much their supporting code has in common.
Am Sonntag, den 05.04.2009, 22:22 +0200 schrieb Jean Delvare: > On Sun, 05 Apr 2009 14:58:17 -0400, Andy Walls wrote: > > On Sun, 2009-04-05 at 20:31 +0200, Janne Grunau wrote: > > > I have devices for lirc_zilog (which should probably be merged with > > > lirc_i2c) > > > > Hmmm. Following Jean's reasoning, that may be the wrong way to go, if > > you want to avoid probing. A module to handle each specific type of I2C > > IR chip, splitting up lirc_i2c and leaving lirc_zilog as is, may be > > better in the long run. > > This really doesn't matter. With the new binding model, probing is > under control. You can do probing on some cards and not others, and you > can probe some addresses and not others. And one i2c drivers can > cleanly support more than one device type. Hmm, I'm still "happy" with the broken DVB-T for saa7134 on 2.6.29, tasting some Chianti vine now and need to sleep soon, but I'm also not that confident that your saa7134 MSI TV@nywhere Plus i2c remote does work addressing it directly, since previous reports always said it becomes only visible at all after other devices are probed previously. Unfortunately I can't test it, but will try to reach some with such hardware and ask for testing, likely not on the list currently. > What should be considered to decide whether two devices should be > supported by the same driver or not, is how much their supporting code > has in common. What can not be translated to the input system I would like to know. Andy seems to have closer looked into that. To have it was a need after 2.5.x turned into 2.6.x. It was not even in sight if and when lirc would be ever usable on it again. You have it from Gerd. I also think we currently have lots of users with all sorts of "TV" cards. Triple cards are still quite expensive and can have only one DVB and one analog stream at once, Quad cards depend on special PCI slots and PCIe multi capable cards are not yet supported. Because of that, there are lots of mythtv and similar machines stuffed with all sort of cards in every free PCI slot I think. Cheers, Hermann -- To unsubscribe from this list: send the line "unsubscribe linux-media" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On Sun, 2009-04-05 at 23:22 +0200, hermann pitton wrote: > Am Sonntag, den 05.04.2009, 22:22 +0200 schrieb Jean Delvare: > > On Sun, 05 Apr 2009 14:58:17 -0400, Andy Walls wrote: > What can not be translated to the input system I would like to know. > Andy seems to have closer looked into that. 1. IR blasting: sending IR codes to transmit out to a cable convertor box, DTV to analog convertor box, or similar devices to change channels before recording starts. An input interface doesn't work well for output. 2. Sending raw IR samples to user space: user space applications can then decode or match an unknown or non-standard IR remote protocol in user space software. Timing information to go along with the sample data probably needs to be preserved. I'm assuming the input interface currently doesn't support that. That's all the Gerd mentioned. One more nice feature to have, that I'm not sure how easily the input system could support: 3. specifying remote control code to key/button translations with a configuration file instead of recompiling a module. In effect there are actually two devices the ir-kbd-i2c input driver is supporting in various combinations: an IR receiver and an IR remote. Regards, Andy -- To unsubscribe from this list: send the line "unsubscribe linux-media" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Am Sonntag, den 05.04.2009, 18:00 -0400 schrieb Andy Walls: > On Sun, 2009-04-05 at 23:22 +0200, hermann pitton wrote: > > Am Sonntag, den 05.04.2009, 22:22 +0200 schrieb Jean Delvare: > > > On Sun, 05 Apr 2009 14:58:17 -0400, Andy Walls wrote: > > > > What can not be translated to the input system I would like to know. > > Andy seems to have closer looked into that. > > 1. IR blasting: sending IR codes to transmit out to a cable convertor > box, DTV to analog convertor box, or similar devices to change channels > before recording starts. An input interface doesn't work well for > output. > > 2. Sending raw IR samples to user space: user space applications can > then decode or match an unknown or non-standard IR remote protocol in > user space software. Timing information to go along with the sample > data probably needs to be preserved. I'm assuming the input interface > currently doesn't support that. > > That's all the Gerd mentioned. Hmmm .... > One more nice feature to have, that I'm not sure how easily the input > system could support: > > 3. specifying remote control code to key/button translations with a > configuration file instead of recompiling a module. Maybe try that from 5 years back. Or what Mauro has. Who ever had to recompile a module for changing the key tables? > In effect there are actually two devices the ir-kbd-i2c input driver is > supporting in various combinations: an IR receiver and an IR remote. > > > Regards, > Andy > I doubt you are down to it. Cheers, Hermann -- To unsubscribe from this list: send the line "unsubscribe linux-media" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Hi Andy, Am Sonntag, den 05.04.2009, 18:00 -0400 schrieb Andy Walls: > On Sun, 2009-04-05 at 23:22 +0200, hermann pitton wrote: > > Am Sonntag, den 05.04.2009, 22:22 +0200 schrieb Jean Delvare: > > > On Sun, 05 Apr 2009 14:58:17 -0400, Andy Walls wrote: > > > > What can not be translated to the input system I would like to know. > > Andy seems to have closer looked into that. > > 1. IR blasting: sending IR codes to transmit out to a cable convertor > box, DTV to analog convertor box, or similar devices to change channels > before recording starts. An input interface doesn't work well for > output. > > 2. Sending raw IR samples to user space: user space applications can > then decode or match an unknown or non-standard IR remote protocol in > user space software. Timing information to go along with the sample > data probably needs to be preserved. I'm assuming the input interface > currently doesn't support that. > > That's all the Gerd mentioned. > > > One more nice feature to have, that I'm not sure how easily the input > system could support: > > 3. specifying remote control code to key/button translations with a > configuration file instead of recompiling a module. > > In effect there are actually two devices the ir-kbd-i2c input driver is > supporting in various combinations: an IR receiver and an IR remote. > > > Regards, > Andy > you always end up with something transmitting series of 0s and 1s. It does not matter if the medium is infrared, infradead ;), wireless or whats or ever. If it spares a chip for the remote and you have to get it from IRQs triggered, you have to do that. It could also be some voltage change, a ultrasound beeper or what ever. The first place for such is the input layer and nothing out of the kernel. Cheers, Hermann -- To unsubscribe from this list: send the line "unsubscribe linux-media" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On Sun, 05 Apr 2009 18:00:04 -0400 Andy Walls <awalls@radix.net> wrote: > On Sun, 2009-04-05 at 23:22 +0200, hermann pitton wrote: > > Am Sonntag, den 05.04.2009, 22:22 +0200 schrieb Jean Delvare: > > > On Sun, 05 Apr 2009 14:58:17 -0400, Andy Walls wrote: > > > > What can not be translated to the input system I would like to know. > > Andy seems to have closer looked into that. > > 1. IR blasting: sending IR codes to transmit out to a cable convertor > box, DTV to analog convertor box, or similar devices to change channels > before recording starts. An input interface doesn't work well for > output. On my understanding, IR output is a separate issue. AFAIK, only a very few ivtv devices support IR output. I'm not sure how this is currently implemented. > 2. Sending raw IR samples to user space: user space applications can > then decode or match an unknown or non-standard IR remote protocol in > user space software. Timing information to go along with the sample > data probably needs to be preserved. I'm assuming the input interface > currently doesn't support that. If the driver processes correctly the IR samples, I don't see why you would need to pass the raw protocols to userspace. Maybe we need to add some ioctls at the API to allow certain controls, like, for example, ask kernel to decode IR using RC4 instead or RC5, on devices that supports more than one IR protocol. > That's all the Gerd mentioned. > > > One more nice feature to have, that I'm not sure how easily the input > system could support: > > 3. specifying remote control code to key/button translations with a > configuration file instead of recompiling a module. The input and the current drivers that use input already supports this feature. You just need to load a new code table to replace the existing one. See v4l2-apps/util/keytable.c to see how easy is to change a key code. It contains a complete code to fully replace a key code table. Also, the Makefile there will extract the current keytables for the in-kernel drivers. Btw, with only 12 lines, you can create a keycode replace "hello world!": #include <fcntl.h> /* due to O_RDONLY */ #include <stdio.h> /* open() */ #include <linux/input.h> /* input ioctls and keycode macros */ #include <sys/ioctl.h> /* ioctl() */ void main(void) { int codes[2]; int fd = open("/dev/video0", O_RDONLY); /* Hmm.. in real apps, we should check for errors */ codes[0] = 10; /* Scan code */ codes[1] = KEY_UP; /* Key code */ ioctl(fd, EVIOCSKEYCODE, codes); /* hello world! */ } Cheers, Mauro -- To unsubscribe from this list: send the line "unsubscribe linux-media" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On Sun, 5 Apr 2009, Mauro Carvalho Chehab wrote: > On Sun, 05 Apr 2009 18:00:04 -0400 > Andy Walls <awalls@radix.net> wrote: > > > On Sun, 2009-04-05 at 23:22 +0200, hermann pitton wrote: > > > Am Sonntag, den 05.04.2009, 22:22 +0200 schrieb Jean Delvare: > > > > On Sun, 05 Apr 2009 14:58:17 -0400, Andy Walls wrote: > > > > > > > What can not be translated to the input system I would like to know. > > > Andy seems to have closer looked into that. > > > > 1. IR blasting: sending IR codes to transmit out to a cable convertor > > box, DTV to analog convertor box, or similar devices to change channels > > before recording starts. An input interface doesn't work well for > > output. > > On my understanding, IR output is a separate issue. AFAIK, only a very few ivtv > devices support IR output. I'm not sure how this is currently implemented. For the pvrusb2 driver, MCE style 24xxx devices (2nd generation 24xxx) and HVR-1950 devices have IR blasting capabilities. At the moment, people have gotten this to work on the 24xxx model with the appropriate lirc driver. In theory it should be doable for HVR-1950 as well (and the pvrusb2 does what is needed to make it possible) but I don't think anyone has succeeded there yet. Sure IR output as a concept and interface is a separate issue. But it can be implemented in the same chip (which is the case in the two examples I list above). So the issue is not separate; it must be dealt with as a whole. Two drivers implementing different features but trying to share one chip is just not fun. > > > > 2. Sending raw IR samples to user space: user space applications can > > then decode or match an unknown or non-standard IR remote protocol in > > user space software. Timing information to go along with the sample > > data probably needs to be preserved. I'm assuming the input interface > > currently doesn't support that. > > If the driver processes correctly the IR samples, I don't see why you would > need to pass the raw protocols to userspace. Maybe we need to add some ioctls > at the API to allow certain controls, like, for example, ask kernel to decode > IR using RC4 instead or RC5, on devices that supports more than one IR protocol. Ugh. Why should v4l-dvb get into this business when it's already solved somewhere else? In userspace even. I see in so many other places people arguing for V4L functionality that needs to be kicked out of the kernel and put into userspace. For example, there's all that silliness over pixel formats that I'm soon going to have to deal with... Yet in this case with IR, there already exists a subsystem that does *more* than ir-kbd-i2c.c, AND it does all the crazy configuration / control in userspace - and yet you argue that ir-kbd-i2c.c should be preferred? Purely because lirc is not in-tree? Well heck, lirc should be in-tree. Let's help them get there and forget ever having to deal with IR again ourselves. Let them do it. > > > That's all the Gerd mentioned. > > > > > > One more nice feature to have, that I'm not sure how easily the input > > system could support: > > > > 3. specifying remote control code to key/button translations with a > > configuration file instead of recompiling a module. > > The input and the current drivers that use input already supports this feature. > You just need to load a new code table to replace the existing one. > > See v4l2-apps/util/keytable.c to see how easy is to change a key code. It > contains a complete code to fully replace a key code table. Also, the Makefile > there will extract the current keytables for the in-kernel drivers. > > Btw, with only 12 lines, you can create a keycode replace "hello world!": > > #include <fcntl.h> /* due to O_RDONLY */ > #include <stdio.h> /* open() */ > #include <linux/input.h> /* input ioctls and keycode macros */ > #include <sys/ioctl.h> /* ioctl() */ > void main(void) > { > int codes[2]; > int fd = open("/dev/video0", O_RDONLY); /* Hmm.. in real apps, we should check for errors */ > codes[0] = 10; /* Scan code */ > codes[1] = KEY_UP; /* Key code */ > ioctl(fd, EVIOCSKEYCODE, codes); /* hello world! */ > } I just looked at this. I freely admit I haven't noticed this before, but having looked at it now, and having examined ir-kbd-i2c.c, I still don't see the whole picture here: 1. The switch statement in ir-kbd-i2c.c:ir_attach() is apparently implicitly trying to assume a particular type of remote based on the I2C address of the IR receiver it's talking to. Yuck. That's really not right at all. The IR receiver used does not automatically mean which remote is used. What if the vendor switches remotes? That's happened with the PVR-USB2 hardware in the past (based on photos I've seen). Who's to say the next remote to be supplied is compatible? 2. Your example above is opening the video device endpoint and issuing ioctl()s that are not part of V4L. That is supposed to work?!? 3. A given IR remote may be described by much more than what 'scan codes' it produces. I don't know a lot about IR, but looking at the typical lirc definition for a remote, there's obvious timing and protocol parameters as well. Just being able to swap scan codes around is not always going to be enough. 4. I imagine that the input event framework in the kernel has a means for programmatic mapping of scan codes to key codes, but looking at ir-kbd-i2c.c, it appears to only be selecting from among a very small set of kernel-compiled translation tables. I must be missing something here. In an earlier post (from Andy?) some history was dug up about ir-kbd-i2c.c. From what I understand, the only reason ir-kbd-i2c.c came into existence was because lirc was late in supporting 2.6.x series kernels and Gerd needed *something* to allow IR to work. So he created this module, knowing full well that it didn't cover all possible cases. Rather it covered the common cases he cared about. That was a while ago. And we need to do all the cases - or at least not mess up what already exists elsewhere that does handle the "uncommon" cases. The lirc drivers do work in 2.6. And apparently they were on the scene before ir-kbd-i2c.c, just unfortunately not in-tree. The lirc drivers really need to get into the kernel. From where I'm sitting the long term goal should be to get lirc into the kernel. -Mike
Am Sonntag, den 05.04.2009, 21:52 -0500 schrieb Mike Isely: > On Sun, 5 Apr 2009, Mauro Carvalho Chehab wrote: > > > On Sun, 05 Apr 2009 18:00:04 -0400 > > Andy Walls <awalls@radix.net> wrote: > > > > > On Sun, 2009-04-05 at 23:22 +0200, hermann pitton wrote: > > > > Am Sonntag, den 05.04.2009, 22:22 +0200 schrieb Jean Delvare: > > > > > On Sun, 05 Apr 2009 14:58:17 -0400, Andy Walls wrote: > > > > > > > > > > What can not be translated to the input system I would like to know. > > > > Andy seems to have closer looked into that. > > > > > > 1. IR blasting: sending IR codes to transmit out to a cable convertor > > > box, DTV to analog convertor box, or similar devices to change channels > > > before recording starts. An input interface doesn't work well for > > > output. > > > > On my understanding, IR output is a separate issue. AFAIK, only a very few ivtv > > devices support IR output. I'm not sure how this is currently implemented. > > For the pvrusb2 driver, MCE style 24xxx devices (2nd generation 24xxx) > and HVR-1950 devices have IR blasting capabilities. At the moment, > people have gotten this to work on the 24xxx model with the appropriate > lirc driver. In theory it should be doable for HVR-1950 as well (and > the pvrusb2 does what is needed to make it possible) but I don't think > anyone has succeeded there yet. > > Sure IR output as a concept and interface is a separate issue. But it > can be implemented in the same chip (which is the case in the two > examples I list above). So the issue is not separate; it must be dealt > with as a whole. Two drivers implementing different features but trying > to share one chip is just not fun. > > > > > > > > > 2. Sending raw IR samples to user space: user space applications can > > > then decode or match an unknown or non-standard IR remote protocol in > > > user space software. Timing information to go along with the sample > > > data probably needs to be preserved. I'm assuming the input interface > > > currently doesn't support that. > > > > If the driver processes correctly the IR samples, I don't see why you would > > need to pass the raw protocols to userspace. Maybe we need to add some ioctls > > at the API to allow certain controls, like, for example, ask kernel to decode > > IR using RC4 instead or RC5, on devices that supports more than one IR protocol. > > Ugh. Why should v4l-dvb get into this business when it's already solved > somewhere else? In userspace even. > > I see in so many other places people arguing for V4L functionality that > needs to be kicked out of the kernel and put into userspace. For > example, there's all that silliness over pixel formats that I'm soon > going to have to deal with... > > Yet in this case with IR, there already exists a subsystem that does > *more* than ir-kbd-i2c.c, AND it does all the crazy configuration / > control in userspace - and yet you argue that ir-kbd-i2c.c should be > preferred? Purely because lirc is not in-tree? Well heck, lirc should > be in-tree. Let's help them get there and forget ever having to deal > with IR again ourselves. Let them do it. > > > > > > > That's all the Gerd mentioned. > > > > > > > > > One more nice feature to have, that I'm not sure how easily the input > > > system could support: > > > > > > 3. specifying remote control code to key/button translations with a > > > configuration file instead of recompiling a module. > > > > The input and the current drivers that use input already supports this feature. > > You just need to load a new code table to replace the existing one. > > > > See v4l2-apps/util/keytable.c to see how easy is to change a key code. It > > contains a complete code to fully replace a key code table. Also, the Makefile > > there will extract the current keytables for the in-kernel drivers. > > > > Btw, with only 12 lines, you can create a keycode replace "hello world!": > > > > #include <fcntl.h> /* due to O_RDONLY */ > > #include <stdio.h> /* open() */ > > #include <linux/input.h> /* input ioctls and keycode macros */ > > #include <sys/ioctl.h> /* ioctl() */ > > void main(void) > > { > > int codes[2]; > > int fd = open("/dev/video0", O_RDONLY); /* Hmm.. in real apps, we should check for errors */ > > codes[0] = 10; /* Scan code */ > > codes[1] = KEY_UP; /* Key code */ > > ioctl(fd, EVIOCSKEYCODE, codes); /* hello world! */ > > } > > I just looked at this. I freely admit I haven't noticed this before, > but having looked at it now, and having examined ir-kbd-i2c.c, I still > don't see the whole picture here: > > 1. The switch statement in ir-kbd-i2c.c:ir_attach() is apparently > implicitly trying to assume a particular type of remote based on the I2C > address of the IR receiver it's talking to. Yuck. That's really not > right at all. The IR receiver used does not automatically mean which > remote is used. What if the vendor switches remotes? That's happened > with the PVR-USB2 hardware in the past (based on photos I've seen). > Who's to say the next remote to be supplied is compatible? > > 2. Your example above is opening the video device endpoint and issuing > ioctl()s that are not part of V4L. That is supposed to work?!? > > 3. A given IR remote may be described by much more than what 'scan > codes' it produces. I don't know a lot about IR, but looking at the > typical lirc definition for a remote, there's obvious timing and > protocol parameters as well. Just being able to swap scan codes around > is not always going to be enough. > > 4. I imagine that the input event framework in the kernel has a means > for programmatic mapping of scan codes to key codes, but looking at > ir-kbd-i2c.c, it appears to only be selecting from among a very small > set of kernel-compiled translation tables. I must be missing something > here. > > In an earlier post (from Andy?) some history was dug up about > ir-kbd-i2c.c. From what I understand, the only reason ir-kbd-i2c.c came > into existence was because lirc was late in supporting 2.6.x series > kernels and Gerd needed *something* to allow IR to work. So he created > this module, knowing full well that it didn't cover all possible cases. > Rather it covered the common cases he cared about. That was a while > ago. And we need to do all the cases - or at least not mess up what > already exists elsewhere that does handle the "uncommon" cases. The > lirc drivers do work in 2.6. And apparently they were on the scene > before ir-kbd-i2c.c, just unfortunately not in-tree. The lirc drivers > really need to get into the kernel. From where I'm sitting the long > term goal should be to get lirc into the kernel. > > -Mike > ir_kbd_i2c is much older than Gerd. It comes from bttv. lirc was unusable on 2.6.x. I do confirm that. If you like to have some more on the ways, the just merged dvb did state the whole in kernel i2c is crap, likely with some good reasons. Mike, I'm too old now to start on it again, but feel free to give us more insight. Cheers, Hermann -- To unsubscribe from this list: send the line "unsubscribe linux-media" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On Sun, 5 Apr 2009, Andy Walls wrote: > > Here is where LIRC may be its own worst enemy. LIRC has filled some > shortcomings in the kernel for support of IR device functions for so > long (LWN says LIRC is 10 years old), that large numbers of users have > come to depend on its operation, while at the same time apparently > removing impetus for doing much to update the in kernel IR device > support. More than that. In 1997 I bought a serial port remote off ebay and tried to get it to work with linux. I found an abandoned project from the Metzler brothers called LIRC, though it didn't work. I wrote a new protocol for the serial port driver, which was the only one at the time, rewrote the remote pulse decoding code and came up a new socket based the client/server protocol and wrote the x-event client. At that point remotes were defined in header files so make was still needed to add a new one. -- To unsubscribe from this list: send the line "unsubscribe linux-media" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Jean: Here's a description of what I've got on the front burner right now: 1. The pvrusb2 driver now can unambiguously know if it is dealing with a device that is known to have ir-kbd-i2c compatible IR capabilities. 2. There is a new module option, "disable_autoload_ir_kbd", which if present and set to 1 will indicate that ir-kbd should not be loaded. 2. Based upon (1) and (2), the driver will optionally attempt to load ir-kbd using the code from your patch. 3. In the pvrusb2 case, the only i2c address that currently matters is 0x18 (though I have some suspicions about another case but that can be dealt with later), so I trimmed the probe list in the register function you had added. Since calling i2c_new_probed_device() for a non-existent target driver doesn't cause any harm, then merging the above now should not result in any kind of regression. So it can go in even before the rest of your changes. That I believe also removes the objection Mauro had - this way there's no issues / dependencies. I've tested this enough to know that it at least doesn't do any further harm. I will put this up in a changeset shortly. With all that said, we should not ignore lirc and instead do whatever is reasonable to help ensure it continues to work. Though admittedly there's been plenty of opportunity to update and this whole transition has been going on for a long time. The stuff I describe above should at least keep the pvrusb2 driver out of the fray for now. -Mike
On Sun, 5 Apr 2009, Mike Isely wrote: > 1. The switch statement in ir-kbd-i2c.c:ir_attach() is apparently > implicitly trying to assume a particular type of remote based on the I2C > address of the IR receiver it's talking to. Yuck. That's really not > right at all. The IR receiver used does not automatically mean which > remote is used. What if the vendor switches remotes? That's happened > with the PVR-USB2 hardware in the past (based on photos I've seen). > Who's to say the next remote to be supplied is compatible? IMHO, the way the remote supported is compiled into the kernel is absurd. The system I setup 12 years ago was better than this. At least the remote was compiled into a userspace daemon and multiple remotes were supported at the same time. The keycode system I used had a remote id/key id split, so you could have volume up key on any remote control the mixer but make the power buttons on different remotes turn on different apps. > 3. A given IR remote may be described by much more than what 'scan > codes' it produces. I don't know a lot about IR, but looking at the > typical lirc definition for a remote, there's obvious timing and > protocol parameters as well. Just being able to swap scan codes around > is not always going to be enough. A remote typically sends a header sequence of a long pulse and space before the code. The length of the pulse on the space varies greatly by remote and makes a good way to identify the remote when multiple ones are supported. Then a pulse coded remote sends a sequence bits, usually 8 to 32. The length of the pulse identifies 1s or 0s. Different remotes have different pulse lengths and different spaces between them. RC5 remotes use Manchester encoding for this part. When you hold a key down some remotes just repeat the same sequence over and over again. Some repeat the scan code but omit the header part. Some send out a special pulse sequence to indicate the last key is being held down. With the latter two methods you can tell the difference between a key being held down and a key being pressed repeatedly. With the first you have guess based on how fast the repeats are coming in. This is very different than a keyboard, which sends a code when you press a key and another when you release it. The rate at which remotes repeat varies greatly. You might find that one remote makes your volume change annoying slowly while another is much too fast to be usable. Remote keys usually start repeating without delay, so if you let a toggle like 'mute' repeat it becomes almost impossible to hit it just once. Entering numbers becomes impossible as well. -- To unsubscribe from this list: send the line "unsubscribe linux-media" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On Sun, 05 Apr 2009 23:22:03 +0200, drunk and tired hermann pitton wrote: > Hmm, I'm still "happy" with the broken DVB-T for saa7134 on 2.6.29, > tasting some Chianti vine now and need to sleep soon, but I'm also not > that confident that your saa7134 MSI TV@nywhere Plus i2c remote does > work addressing it directly, since previous reports always said it > becomes only visible at all after other devices are probed previously. > > Unfortunately I can't test it, but will try to reach some with such > hardware and ask for testing, likely not on the list currently. Thanks for the heads up. I was curious about this as well. The original comment said that the MSI TV@nywhere Plus IR receiver would not respond to _probes_ before another device on the I2C bus was accessed. I didn't know for sure if this only applied to the probe sequence or to any attempt to access the IR receiver. As we no longer need to probe for the device, I thought it may be OK to remove the extra code. But probably the removal of the extra code should be delayed until we find one tester to confirm the exact behavior. Here, done. Anyone out there with a MSI TV@nywhere Plus that could help with testing?
Hi Andy, On Sun, 05 Apr 2009 15:35:52 -0400, Andy Walls wrote: > --- v4l-dvb.orig/linux/drivers/media/video/ivtv/ivtv-i2c.c 2009-04-04 10:53:08.000000000 +0200 > +++ v4l-dvb/linux/drivers/media/video/ivtv/ivtv-i2c.c 2009-04-04 10:58:36.000000000 +0200 > [snip] > - > + const unsigned short addr_list[] = { > + 0x1a, 0x18, 0x64, 0x30, > + I2C_CLIENT_END > + }; > [snip] > > > I just noticed you're missing address 0x71 for ivtv. At least some > PVR-150 boards have a Zilog chip at that address. Thanks for reporting. The list above is taken directly from the original ir-kbd-i2c driver (minus address 0x4b which Hans Verkuil told me was useless for the ivtv and cx18 adapters). I'm all for adding support for more boards, however I'd rather do this _after_ the i2c model conversion is done, so that we have a proper changelog entry saying that we added support for the PVR-150, and that it gets proper testing. Hiding support addition in a larger patch would probably do as much harm as good.
On Sun, 5 Apr 2009 13:48:02 -0500 (CDT) Mike Isely <isely@isely.net> wrote: > My impression (at least for pvrusb2-driven devices) is that the later IR > receivers require a completely different driver to work properly; one > can't just bolt additional features into ir-kbd-i2c for this. This is the old approach: adding more stuff into ir-kbd-i2c. The new approach is to let ir-kbd-i2c to prepare for IR, but filling the protocol decoding routines and IR names after having it bound on i2c bus. So, the IR routines will be properly filled by the bridge driver (pvrusb2, in this case). > In lirc's case, a different driver is in fact used. But you know this already. > > I haven't looked at ir-kbd-i2c in a while, but one other thing I > seriously did not like about it was that the mapping of IR codes to key > events was effectively hardcoded in the driver. That makes it difficult > to make the same driver work with different kinds of remotes. Even if > the bridge driver (e.g. pvrusb2) were to supply such a map, that's still > wrong because it's within the realm of possibility that the user might > actually want to use a different remote transmitter (provided it is > compatible enough to work with the receiver hardware). The hardcoded keytables are just the default ones for that keyboard. As I've shown already in this thread, it can be easily replaced on userspace, and we have already an userspace tool on v4l2-apps that replaces those tables. > The lirc architecture solves this easily via a conf file in userspace. In fact, > lirc can map multiple mappings to a single receiver, permitting it to > work concurrently with more than one remote. But is such a thing even > possible with ir-kbd-i2c? I know this is one reason people tend to > choose lirc. Multiple mappings based on what userspace app you're using can't be done internally. However, you can keep using lirc with event driver for those cases where you want to have multiple mappings, and this works fine. The only drawback i saw when I used lirc the last time (a long time ago) is that, when you remove an usb device, it used to flood the logs with errors (several errors by second, warning that event interface disappeared). Probably they already solved this issue. Cheers, Mauro -- To unsubscribe from this list: send the line "unsubscribe linux-media" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On Mon, 2009-04-06 at 11:04 +0200, Jean Delvare wrote: > Hi Andy, > I'm all for adding > support for more boards, however I'd rather do this _after_ the i2c > model conversion is done, so that we have a proper changelog entry > saying that we added support for the PVR-150, and that it gets proper > testing. Hiding support addition in a larger patch would probably do > as much harm as good. Makes sens to me. Especially when I just simply, blindly added 0x71 in my initial testing, I got a kernel Oops. Regards, Andy -- To unsubscribe from this list: send the line "unsubscribe linux-media" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On Sun, 5 Apr 2009 21:52:31 -0500 (CDT) Mike Isely <isely@isely.net> wrote: > On Sun, 5 Apr 2009, Mauro Carvalho Chehab wrote: > > > On Sun, 05 Apr 2009 18:00:04 -0400 > > Andy Walls <awalls@radix.net> wrote: > > > > > On Sun, 2009-04-05 at 23:22 +0200, hermann pitton wrote: > > > > Am Sonntag, den 05.04.2009, 22:22 +0200 schrieb Jean Delvare: > > > > > On Sun, 05 Apr 2009 14:58:17 -0400, Andy Walls wrote: > > > > > > > > > > What can not be translated to the input system I would like to know. > > > > Andy seems to have closer looked into that. > > > > > > 1. IR blasting: sending IR codes to transmit out to a cable convertor > > > box, DTV to analog convertor box, or similar devices to change channels > > > before recording starts. An input interface doesn't work well for > > > output. > > > > On my understanding, IR output is a separate issue. AFAIK, only a very few ivtv > > devices support IR output. I'm not sure how this is currently implemented. > > For the pvrusb2 driver, MCE style 24xxx devices (2nd generation 24xxx) > and HVR-1950 devices have IR blasting capabilities. At the moment, > people have gotten this to work on the 24xxx model with the appropriate > lirc driver. In theory it should be doable for HVR-1950 as well (and > the pvrusb2 does what is needed to make it possible) but I don't think > anyone has succeeded there yet. > > Sure IR output as a concept and interface is a separate issue. But it > can be implemented in the same chip (which is the case in the two > examples I list above). So the issue is not separate; it must be dealt > with as a whole. Two drivers implementing different features but trying > to share one chip is just not fun. Yes, this should be considered by the same driver, but perhaps not using the same userspace API. I'm not sure if event interface allows such usage. > > If the driver processes correctly the IR samples, I don't see why you would > > need to pass the raw protocols to userspace. Maybe we need to add some ioctls > > at the API to allow certain controls, like, for example, ask kernel to decode > > IR using RC4 instead or RC5, on devices that supports more than one IR protocol. > > Ugh. Why should v4l-dvb get into this business when it's already solved > somewhere else? In userspace even. > > I see in so many other places people arguing for V4L functionality that > needs to be kicked out of the kernel and put into userspace. For > example, there's all that silliness over pixel formats that I'm soon > going to have to deal with... Removing kernel functionality breaks compatibility with legacy applications. > Yet in this case with IR, there already exists a subsystem that does > *more* than ir-kbd-i2c.c, AND it does all the crazy configuration / > control in userspace - and yet you argue that ir-kbd-i2c.c should be > preferred? Purely because lirc is not in-tree? Setting up lirc to work it is a way more complex that just plugging a device and having IR working. For all my usages here, I prefer to just load a different IR table than having to deal with lirc configuration stuff. It should be users option to use lirc or just rely on the existing IR support. > Well heck, lirc should be in-tree. Let's help them get there and forget ever having to deal > with IR again ourselves. Let them do it. I agree that we should help with this. IMO, the proper way for lirc drivers for media devices is that they should include linux-media oh the discussions about the inclusion of those drivers, in a way that just one driver would be used, and that the event interface will keep working by default, as currently. > I just looked at this. I freely admit I haven't noticed this before, > but having looked at it now, and having examined ir-kbd-i2c.c, I still > don't see the whole picture here: > > 1. The switch statement in ir-kbd-i2c.c:ir_attach() is apparently > implicitly trying to assume a particular type of remote based on the I2C > address of the IR receiver it's talking to. Yuck. That's really not > right at all. The IR receiver used does not automatically mean which > remote is used. What if the vendor switches remotes? That's happened > with the PVR-USB2 hardware in the past (based on photos I've seen). > Who's to say the next remote to be supplied is compatible? This is the legacy model, kept there only due to the fact that we don't really know with 100% sure what boards were using those functions. For the new model, you should take a look on a bridge driver, like, for example, on em28xx-cards: void em28xx_set_ir(struct em28xx *dev, struct IR_i2c *ir) { if (disable_ir) { ir->get_key = NULL; return ; } /* detect & configure */ switch (dev->model) { case (EM2800_BOARD_UNKNOWN): break; case (EM2820_BOARD_UNKNOWN): break; case (EM2800_BOARD_TERRATEC_CINERGY_200): case (EM2820_BOARD_TERRATEC_CINERGY_250): ir->ir_codes = ir_codes_em_terratec; ir->get_key = em28xx_get_key_terratec; snprintf(ir->c.name, sizeof(ir->c.name), "i2c IR (EM28XX Terratec)"); break; ... This function is called just after binding the IR, by i2c callback. It properly fills the IR tables, get_key and IR names based on each different board model. > > 2. Your example above is opening the video device endpoint and issuing > ioctl()s that are not part of V4L. That is supposed to work?!? I did a typo there. In fact, opens the proper /dev/event interface. The code is the same kind of code that allows userspace to change from US keymapping to a local keymapping. It should be easy to call an userspace app from udev, after a device connection, if the user wants to have a different IR keycode table for his device. > 3. A given IR remote may be described by much more than what 'scan > codes' it produces. I don't know a lot about IR, but looking at the > typical lirc definition for a remote, there's obvious timing and > protocol parameters as well. Just being able to swap scan codes around > is not always going to be enough. The IR in-kernel implementation has the proper timing and protocol definitions for each device. Having something else to take care with will just cause troubles. For example, I have two saa713x devices with different IR chips. On both, the IR protocol handling is done inside the chip. On one device, if you press a key and keep it pressed, it will produce just one scan code. So, that device has no way to support IR repetition. The other one produces scan codes to indicate a key press and a key release events. If you try to handle those cases outside kernel, you'll need to have a complex setup environment for users to specify or to detect such issues. However, if this is done on kernel, you can just use the proper event api to handle such differences. There is just one issue here that it is not addressed currently: if your device comes with an IR with, for example pulse distance protocol, but you bought a RC5 IR and wants to use it with your device, you would need to say to the driver (and maybe to the hardware) to use a different protocol for decoding. On some cases where the IR decoding is done via software, if you get the raw data, you can change the protocol easily. However, on other cases, this can be done only if you program the device with a different setup. > 4. I imagine that the input event framework in the kernel has a means > for programmatic mapping of scan codes to key codes, but looking at > ir-kbd-i2c.c, it appears to only be selecting from among a very small > set of kernel-compiled translation tables. I must be missing something > here. It sets the default key tables. In fact, on most cases, the setting of the key tables is done at the bridge driver. The ones inside ir-kbd-i2c are the exceptions. Patches are welcome to remove those by adding such setup at the bridge drivers, but extra care should be done to be sure that we won't break any IR support. > In an earlier post (from Andy?) some history was dug up about > ir-kbd-i2c.c. From what I understand, the only reason ir-kbd-i2c.c came > into existence was because lirc was late in supporting 2.6.x series > kernels and Gerd needed *something* to allow IR to work. So he created > this module, knowing full well that it didn't cover all possible cases. > Rather it covered the common cases he cared about. That was a while > ago. And we need to do all the cases - or at least not mess up what > already exists elsewhere that does handle the "uncommon" cases. The > lirc drivers do work in 2.6. And apparently they were on the scene > before ir-kbd-i2c.c, just unfortunately not in-tree. The lirc drivers > really need to get into the kernel. From where I'm sitting the long > term goal should be to get lirc into the kernel. The ir-kbd-i2c is just the top of the iceberg. It covers the minority of devices where IR is implemented via an i2c chip. Most devices are connected via GPIO. Also, after Gerd stopping work on it, several improvements and changes on the original driver, including moving the setup process to the bridge drivers. It should be noticed that there are 4 types of IRs: 1) IR chips connected via GPIO; 2) Simpler ones that just connect the IR sensor directly on some GPIO port capable of generating interrupts. This kind of setup is very common on saa7134 driver. It should be noticed that a bad implementation here will lead to machine hangups, due to IRQ troubles; 3) IR chips connected via I2C; 4) bridge chips with IR decoding capabilities. For example, em28xx has RC5 decoding internally. On several cases, the IR chip is a low-cost generic processor or ASIC, with some IR decoding software programmed there (read-only, AFAIK). I haven't look on lirc implementation lately, but I think that lirc can effectively change the decoding protocol only on the case (2) where the IR raw data is sent directly via a GPIO port. On the other cases, the IR chip will handle the protocol. So, if you have an IR chip and you want to change the protocol, you'll need to send commands via the bridge driver to set it differently. It is risky to let an external driver to do this, since this may cause troubles to the device driver, as the developers won't generally count that another driver is poking around with setup commands. Cheers, Mauro -- To unsubscribe from this list: send the line "unsubscribe linux-media" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On Sun, 05 Apr 2009 08:44:15 -0400 Andy Walls <awalls@radix.net> wrote: > The scope of a complete kernel IR infrastructure goes a bit beyond I2C > bus devices that are only input devices. > > What's the scope of what you want to tackle here? > > I certainly don't want to reinvent something that's going to look just > like the LIRC driver model: > > http://www.lirc.org/html/technical.html > > Which already has an infrastructure for IR driver module writers: > http://www.lirc.org/html/technical.html#lirc_dev As other out-of-tree drivers that have a long trip, I suspect that lirc did some assumptions, while event and v4l did different ones. Due to kernel rules to keep API's forever, we should take some care to avoid breaking the existing API in favor to another one. So, this probably means some lirc redesign, in order to get his way to kernel, on a similar path that ivtv driver did. The part of lirc that I'm concerned with are the ones that work with GPIO and I2C devices and the API. > Do we just convert lirc_dev, lirc_i2c, and lirc_zilog to a cleaned up > set of in kernel modules? We should cover also the lirc gpio module(s). > lirc_i2c can certainly be broken up into > several modules: 1 per supported device. I don't think that breaking it into one per device is the better approach. IMO, we need a common i2c glue (like what ir-kbd-i2c provides, if we remove the legacy stuff) that it is IR independent. the IR dependent parts can be part of ir-common module or eventually we can split it into sub-modules. > Should these create an input > device as well to maintain compatability with apps expecting an > ir-kbd-i2c like function? For sure. The event interface is the kernel way for input devices. There are also other IR devices (like IR mouses and keyboards) already handled via input/event interface. On a first glance, I don't see the need to exporting raw data to userspace, although I understand why lirc needs this currently. > Or do we split up ir-kbd-i2c into per device modules and in addition to > the input event interface, have it register with the lirc_dev module? > > Do we leverage LIRC's lirc_dev infrastructure module at all? (I think it > would be a waste of time not to do so.) IMO, the proper workflow would be to discuss lirc as a hole with Lirc people, linux-media and input/event people. Cheers, Mauro -- To unsubscribe from this list: send the line "unsubscribe linux-media" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On Sunday 05 April 2009 14:31:54 Janne Grunau wrote: > On Sun, Apr 05, 2009 at 01:39:33PM -0400, Andy Walls wrote: > > On Sun, 2009-04-05 at 16:37 +0200, Janne Grunau wrote: > > > > > > I would guess that it won't work. There is an effort to merge lirc. It's > > > currently stalled though. > > > > Perhaps you and Jarrod and Christopher have already discussed this, > > but... > > > > Instead of trying to push all of the LIRC kernel components through in > > one big patch set, perhaps it would be easier to just get the lirc_dev > > and any other needed infrastructure components in first. > > > > If one focuses on satisfying the LKML comments to lirc_dev and the > > Makefile to get that kernel module in the kernel, then, at least for > > video card hosted IR devices, there is an infrastructure to which to > > hook new or rewritten i2c IR driver modules. > > I guess lkml would NAK patches adding infrastructure only bits but we > will probably for the next patchset concentrate on a few lirc drivers. Yep, my thoughts exactly. > Christopher doesn't participate in the merge attempt. Christoph has been giving decent feedback and merging the git tree changes back into lirc cvs of late, but no, he's not directly participating in the effort to merge lirc into the kernel. > > > A git tree is available at > > > > > > git://git.wilsonet.com/linux-2.6-lirc.git > > > > > > Jared Wilson and I were working on it (mainly last september). Since the > > > IR on the HD PVR is also driven by the same zilog chip as on other > > > hauppauge devices I'll take of lirc_zilog. Help converting the i2c > > > drivers to the new i2c model is welcome. General cleanup of lirc to make > > > it ready for mainline is of course wellcome too. > > > > I can help with this. I'm mainly concerned with lirc_dev, lirc_i2c (for > > Rx only use of the zilog at 0x71), lirc_zilog, and lirc_mceusb2. That's > > because, of course, I have devices that use those modules. :) > > I have devices for lirc_zilog (which should probably be merged with > lirc_i2c) and lirc serial. Jarod has at least mce usb and imon devices. > That are probably the devices we'll concentrate on the next submission. Indeed, we should focus on serial, i2c/zilog, mceusb2 and imon. I think they're by far the most popular and the best maintained, and between Janne and myself, we can actually test all of them ourselves. > > lirc_dev and the API header would be my first priority, if you need > > help. Did anyone consolidate all the comments from the LKML on Jarrod's > > patch submission? > > no and I lost track which comments were already handled. I think we've got just about everything handled, but I should go back over the stack of comments before we resubmit... I was hoping to have something ready for 2.6.30, but work keeps getting in the way...
Hi Jean, Am Montag, den 06.04.2009, 10:40 +0200 schrieb Jean Delvare: > On Sun, 05 Apr 2009 23:22:03 +0200, drunk and tired hermann pitton wrote: don't tell me the French vine is always better. You likely know who introduced that currency once :) > > Hmm, I'm still "happy" with the broken DVB-T for saa7134 on 2.6.29, > > tasting some Chianti vine now and need to sleep soon, but I'm also not > > that confident that your saa7134 MSI TV@nywhere Plus i2c remote does > > work addressing it directly, since previous reports always said it > > becomes only visible at all after other devices are probed previously. > > > > Unfortunately I can't test it, but will try to reach some with such > > hardware and ask for testing, likely not on the list currently. > > Thanks for the heads up. I was curious about this as well. The original > comment said that the MSI TV@nywhere Plus IR receiver would not respond > to _probes_ before another device on the I2C bus was accessed. I didn't > know for sure if this only applied to the probe sequence or to any > attempt to access the IR receiver. As we no longer need to probe for > the device, I thought it may be OK to remove the extra code. But > probably the removal of the extra code should be delayed until we find > one tester to confirm the exact behavior. Here, done. > > Anyone out there with a MSI TV@nywhere Plus that could help with > testing? Here is a link to one of the initial reports by Henry, others are close to it. http://marc.info/?l=linux-video&m=113324147429459&w=2 There are two different variants of that MSI card, but that undocumented KS003 chip is the same on them. We still have lots of for the remote unsupported cards with KS chips, many from Kworld. Some of these chips are also seen on cx88xx cards already and other drivers may follow. Henry doesn't have this card anymore, but maybe Mark and Brian can test and Oldrich might give feedback for the Avermedia. Cheers, Hermann -- To unsubscribe from this list: send the line "unsubscribe linux-media" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On Mon, 06 Apr 2009 23:10:36 +0200, hermann pitton wrote: > Am Montag, den 06.04.2009, 10:40 +0200 schrieb Jean Delvare: > > Anyone out there with a MSI TV@nywhere Plus that could help with > > testing? > > Here is a link to one of the initial reports by Henry, others are close > to it. > > http://marc.info/?l=linux-video&m=113324147429459&w=2 > > There are two different variants of that MSI card, but that undocumented > KS003 chip is the same on them. Great, thanks for the pointer. If I understand correctly, the KS003 has a state machine flow which causes the chip to stop answering when an invalid address is used on the bus and start answering again when a valid address other than his own is used. As the old i2c model relied a lot on probing, I am not surprised that this was a problem in the past. But with the new model, probes should become infrequent, so I suspect that the workaround may no longer be needed... except when i2c_scan=1 is used. I'd rather keep the workaround in place for the time being, and only once the ir-kbd-i2c changes have settled, try to remove it if someone really cares.
Jean Delvare wrote: > On Mon, 06 Apr 2009 23:10:36 +0200, hermann pitton wrote: > >> Am Montag, den 06.04.2009, 10:40 +0200 schrieb Jean Delvare: >> >>> Anyone out there with a MSI TV@nywhere Plus that could help with >>> testing? >>> >> Here is a link to one of the initial reports by Henry, others are close >> to it. >> >> http://marc.info/?l=linux-video&m=113324147429459&w=2 >> >> There are two different variants of that MSI card, but that undocumented >> KS003 chip is the same on them. >> > > Great, thanks for the pointer. If I understand correctly, the KS003 > has a state machine flow which causes the chip to stop answering when > an invalid address is used on the bus and start answering again when a > valid address other than his own is used. As the old i2c model relied a > lot on probing, I am not surprised that this was a problem in the past. > But with the new model, probes should become infrequent, so I suspect > that the workaround may no longer be needed... except when i2c_scan=1 > is used. > > I'd rather keep the workaround in place for the time being, and only > once the ir-kbd-i2c changes have settled, try to remove it if someone > really cares. Regarding the KS003 (& KS007; the other "mystery" chip): Upon further investigation of some info from a post from last year (http://www.linuxtv.org/pipermail/linux-dvb/2008-January/022634.html), it appears that these (assuming that they are the same IC across the various MSI, Leadtek & KWorld cards; and I believe that to be true) are the "AT8PS54/S56" chip from "Feeling Technology" ... the datasheet for that part is available through a google search .... probing further (as I had never heard of FT before and so I looked them up), it looks like FT renamed and/or upgraded the chip to the "FM8PS54/S56" ... the near identical datasheet for that second version is also available: http://www.feeling-tech.com.tw/km-master/front/bin/ptdetail.phtml?Part=M1-05&Category=100018 -- To unsubscribe from this list: send the line "unsubscribe linux-media" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On Tue, 07 Apr 2009 23:02:43 -0400 CityK <cityk@rogers.com> wrote: > Regarding the KS003 (& KS007; the other "mystery" chip): > > Upon further investigation of some info from a post from last year > (http://www.linuxtv.org/pipermail/linux-dvb/2008-January/022634.html), > it appears that these (assuming that they are the same IC across the > various MSI, Leadtek & KWorld cards; and I believe that to be true) are > the "AT8PS54/S56" chip from "Feeling Technology" ... the datasheet for > that part is available through a google search .... probing further (as > I had never heard of FT before and so I looked them up), it looks like > FT renamed and/or upgraded the chip to the "FM8PS54/S56" ... the near > identical datasheet for that second version is also available: > http://www.feeling-tech.com.tw/km-master/front/bin/ptdetail.phtml?Part=M1-05&Category=100018 From what I've investigated, several of those IR chips are micro-controllers like the one you pointed. I've seen a few boards whose IR chip is not masked. On those, I always went into some micro-controller datasheet. Those IR's with a micro-controller have some software inside it to decode one IR protocol and generate scan-code sequences that can be received via GPIO or via I2C, depending on the firmware content. The datasheet of those chips are useless, since the behaviour of the device is programmed inside their ROM/EEPROM [1]. So, even being the same chip, you could have two "K007" devices with different firmwares, listening on different i2c addresses and eventually generating different scan-codes for the same IR. On the other hand, for USB devices and for bttv, saa7134 and cx88, there are some easy ways to monitor what i2c messages or GPIO pins are involved with IR. In general, the IR received messages generated by the firmware are some header, a scan code, a repeat key bit and a trailer. So, it is not hard to generate a get-key routine to get the scan code and the repeat bit from the protocol. That's why the modern ir-kbd-i2c approach is to select the proper IR parameters after binding the module, at the bridge driver. The bridge driver is the one who knows what's the IR scan code of the original IR (to set it as the default), and the proper get-key function. With the new i2c behaviour, the bridge driver can also specify the proper i2c address for each device. Cheers, Mauro [1] It doesn't seem to be practical for me to get their internal software.In general, such micro-controllers block EEPROM/ROM read of the software inside. If this is the case of this chip, the only remaining option to get the internal software would be to cut the plastic and try to see the state of each eeprom bit with the help of a good microscope. Anyway, assuming that there are some way to read the ROM content, in order to see the device behavior, one should remove the chip from the board, get the ROM/EEPROM content, write a disassembler for this processor, disassemble the code and analyse the results. This would be a real hard work, would take a lot of time, and I doubt that this would help to improve the driver, since we already know how to read scan codes from those devices. -- To unsubscribe from this list: send the line "unsubscribe linux-media" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On Monday 06 of April 2009 at 23:10:36, hermann pitton wrote: > Hi Jean, > > Am Montag, den 06.04.2009, 10:40 +0200 schrieb Jean Delvare: > > On Sun, 05 Apr 2009 23:22:03 +0200, drunk and tired hermann pitton wrote: > > don't tell me the French vine is always better. > You likely know who introduced that currency once :) > > > > Hmm, I'm still "happy" with the broken DVB-T for saa7134 on 2.6.29, > > > tasting some Chianti vine now and need to sleep soon, but I'm also not > > > that confident that your saa7134 MSI TV@nywhere Plus i2c remote does > > > work addressing it directly, since previous reports always said it > > > becomes only visible at all after other devices are probed previously. > > > > > > Unfortunately I can't test it, but will try to reach some with such > > > hardware and ask for testing, likely not on the list currently. > > > > Thanks for the heads up. I was curious about this as well. The original > > comment said that the MSI TV@nywhere Plus IR receiver would not respond > > to _probes_ before another device on the I2C bus was accessed. I didn't > > know for sure if this only applied to the probe sequence or to any > > attempt to access the IR receiver. As we no longer need to probe for > > the device, I thought it may be OK to remove the extra code. But > > probably the removal of the extra code should be delayed until we find > > one tester to confirm the exact behavior. Here, done. > > > > Anyone out there with a MSI TV@nywhere Plus that could help with > > testing? Hi Jean, I've tried your patches with AverMedia Cardbus Hybrid (E506R) and they works fine. My current experience with AverMedia's IR chip (I don't know which one is on the card) is that I2C probing didn't find anything, but it got the chip into some strange state - next operation failed (so that the autodetection on address 0x40 and "subaddress" 0x0b/0x0d failed). The chip at address 0x40 needs the write first (one byte: 0x0b or 0x0d) and immediate read, otherwise it would not respond. The saa7134's I2C 0xfd quirk (actually I would call it a hack :-)) caused failures in communication with the IR chip. The way I'm doing the IR reading is the same as the Windows driver does - I got the information through the Qemu with pci-proxy patch applied. Cheers, Oldrich. > > Here is a link to one of the initial reports by Henry, others are close > to it. > > http://marc.info/?l=linux-video&m=113324147429459&w=2 > > There are two different variants of that MSI card, but that undocumented > KS003 chip is the same on them. > > We still have lots of for the remote unsupported cards with KS chips, > many from Kworld. Some of these chips are also seen on cx88xx cards > already and other drivers may follow. > > Henry doesn't have this card anymore, but maybe Mark and Brian can test > and Oldrich might give feedback for the Avermedia. > > Cheers, > Hermann -- To unsubscribe from this list: send the line "unsubscribe linux-media" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Mauro Carvalho Chehab wrote: > On Tue, 07 Apr 2009 23:02:43 -0400 > CityK <cityk@rogers.com> wrote: > > >> Regarding the KS003 (& KS007; the other "mystery" chip): >> >> Upon further investigation of some info from a post from last year >> (http://www.linuxtv.org/pipermail/linux-dvb/2008-January/022634.html), >> it appears that these (assuming that they are the same IC across the >> various MSI, Leadtek & KWorld cards; and I believe that to be true) are >> the "AT8PS54/S56" chip from "Feeling Technology" ... the datasheet for >> that part is available through a google search .... probing further (as >> I had never heard of FT before and so I looked them up), it looks like >> FT renamed and/or upgraded the chip to the "FM8PS54/S56" ... the near >> identical datasheet for that second version is also available: >> http://www.feeling-tech.com.tw/km-master/front/bin/ptdetail.phtml?Part=M1-05&Category=100018 >> > > From what I've investigated, several of those IR chips are micro-controllers like > the one you pointed. I've seen a few boards whose IR chip is not masked. On > those, I always went into some micro-controller datasheet. > > Those IR's with a micro-controller have some software inside it to decode one IR > protocol and generate scan-code sequences that can be received via GPIO or via > I2C, depending on the firmware content. > > The datasheet of those chips are useless, since the behaviour of the > device is programmed inside their ROM/EEPROM [1]. So, even being the same chip, > you could have two "K007" devices with different firmwares, listening on > different i2c addresses and eventually generating different scan-codes for the > same IR. > > On the other hand, for USB devices and for bttv, saa7134 and cx88, there are > some easy ways to monitor what i2c messages or GPIO pins are involved with IR. > In general, the IR received messages generated by the firmware are some header, > a scan code, a repeat key bit and a trailer. So, it is not hard to generate a > get-key routine to get the scan code and the repeat bit from the protocol. > > That's why the modern ir-kbd-i2c approach is to select the proper IR parameters > after binding the module, at the bridge driver. The bridge driver is the one > who knows what's the IR scan code of the original IR (to set it as the > default), and the proper get-key function. With the new i2c behaviour, the > bridge driver can also specify the proper i2c address for each device. > > Cheers, > Mauro > > [1] It doesn't seem to be practical for me to get their internal software.In > general, such micro-controllers block EEPROM/ROM read of the software inside. > If this is the case of this chip, the only remaining option to get the internal > software would be to cut the plastic and try to see the state of each eeprom > bit with the help of a good microscope. > Anyway, assuming that there are some way to read the ROM content, in order to > see the device behavior, one should remove the chip from the board, get the > ROM/EEPROM content, write a disassembler for this processor, disassemble the > code and analyse the results. This would be a real hard work, would take a lot > of time, and I doubt that this would help to improve the driver, since we > already know how to read scan codes from those devices. Thanks for the detailed response Mauro. I've actually been wondering about whether the specific "KS00x" designation/label might refer to the embedded firmware or to a dataline, so that thought is certainly consistent with your description. -- To unsubscribe from this list: send the line "unsubscribe linux-media" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Hi! Am Sonntag, den 12.04.2009, 13:37 -0400 schrieb CityK: > Mauro Carvalho Chehab wrote: > > On Tue, 07 Apr 2009 23:02:43 -0400 > > CityK <cityk@rogers.com> wrote: > > > > > >> Regarding the KS003 (& KS007; the other "mystery" chip): > >> > >> Upon further investigation of some info from a post from last year > >> (http://www.linuxtv.org/pipermail/linux-dvb/2008-January/022634.html), > >> it appears that these (assuming that they are the same IC across the > >> various MSI, Leadtek & KWorld cards; and I believe that to be true) are > >> the "AT8PS54/S56" chip from "Feeling Technology" ... the datasheet for > >> that part is available through a google search .... probing further (as > >> I had never heard of FT before and so I looked them up), it looks like > >> FT renamed and/or upgraded the chip to the "FM8PS54/S56" ... the near > >> identical datasheet for that second version is also available: > >> http://www.feeling-tech.com.tw/km-master/front/bin/ptdetail.phtml?Part=M1-05&Category=100018 > >> > > > > From what I've investigated, several of those IR chips are micro-controllers like > > the one you pointed. I've seen a few boards whose IR chip is not masked. On > > those, I always went into some micro-controller datasheet. > > > > Those IR's with a micro-controller have some software inside it to decode one IR > > protocol and generate scan-code sequences that can be received via GPIO or via > > I2C, depending on the firmware content. > > > > The datasheet of those chips are useless, since the behaviour of the > > device is programmed inside their ROM/EEPROM [1]. So, even being the same chip, > > you could have two "K007" devices with different firmwares, listening on > > different i2c addresses and eventually generating different scan-codes for the > > same IR. > > > > On the other hand, for USB devices and for bttv, saa7134 and cx88, there are > > some easy ways to monitor what i2c messages or GPIO pins are involved with IR. > > In general, the IR received messages generated by the firmware are some header, > > a scan code, a repeat key bit and a trailer. So, it is not hard to generate a > > get-key routine to get the scan code and the repeat bit from the protocol. > > > > That's why the modern ir-kbd-i2c approach is to select the proper IR parameters > > after binding the module, at the bridge driver. The bridge driver is the one > > who knows what's the IR scan code of the original IR (to set it as the > > default), and the proper get-key function. With the new i2c behaviour, the > > bridge driver can also specify the proper i2c address for each device. > > > > Cheers, > > Mauro > > > > [1] It doesn't seem to be practical for me to get their internal software.In > > general, such micro-controllers block EEPROM/ROM read of the software inside. > > If this is the case of this chip, the only remaining option to get the internal > > software would be to cut the plastic and try to see the state of each eeprom > > bit with the help of a good microscope. > > Anyway, assuming that there are some way to read the ROM content, in order to > > see the device behavior, one should remove the chip from the board, get the > > ROM/EEPROM content, write a disassembler for this processor, disassemble the > > code and analyse the results. This would be a real hard work, would take a lot > > of time, and I doubt that this would help to improve the driver, since we > > already know how to read scan codes from those devices. > > Thanks for the detailed response Mauro. I've actually been wondering > about whether the specific "KS00x" designation/label might refer to the > embedded firmware or to a dataline, so that thought is certainly > consistent with your description. Consistent with that, as from some first seen ever, the KS007 chip remotes seem to have always more keys than the KS003 ones. Cheers, Hermann -- To unsubscribe from this list: send the line "unsubscribe linux-media" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Hi Oldrich, On Thu, 9 Apr 2009 21:15:30 +0200, Oldrich Jedlicka wrote: > I've tried your patches with AverMedia Cardbus Hybrid (E506R) and they works > fine. Thanks for testing and reporting, and sorry for the late answer. > My current experience with AverMedia's IR chip (I don't know which one is on > the card) is that I2C probing didn't find anything, but it got the chip into > some strange state - next operation failed (so that the autodetection on > address 0x40 and "subaddress" 0x0b/0x0d failed). OK, that makes sense. Many I2C devices only support a limited subset of the I2C protocol, and if you try to address them with a message format they don't support, their state machine goes into a bad state. That's probably what was happening there. This is the reason why we should always instantiate I2C devices explicitly when possible: whatever probing method you use, you have no guarantee that every device will like it. > The chip at address 0x40 needs the write first (one byte: 0x0b or 0x0d) and > immediate read, otherwise it would not respond. The saa7134's I2C 0xfd quirk > (actually I would call it a hack :-)) caused failures in communication with > the IR chip. I didn't know about this hack. The implementation choice seems wrong to me. The hack should be triggered only when needed, rather than by default with an exception for address 0x40. This goes beyond the scope of my patch though, and I don't want to touch that kind of code without hardware at hand to test my changes. > The way I'm doing the IR reading is the same as the Windows driver does - I > got the information through the Qemu with pci-proxy patch applied. Thanks,
--- v4l-dvb.orig/linux/drivers/media/video/bt8xx/bttv-i2c.c 2009-04-04 10:53:08.000000000 +0200 +++ v4l-dvb/linux/drivers/media/video/bt8xx/bttv-i2c.c 2009-04-04 10:58:36.000000000 +0200 @@ -405,6 +405,27 @@ int __devinit init_bttv_i2c(struct bttv } if (0 == btv->i2c_rc && i2c_scan) do_i2c_scan(btv->c.v4l2_dev.name, &btv->i2c_client); + + /* Instantiate the IR receiver device, if present */ + if (0 == btv->i2c_rc) { + struct i2c_board_info info; + /* The external IR receiver is at i2c address 0x34 (0x35 for + reads). Future Hauppauge cards will have an internal + receiver at 0x30 (0x31 for reads). In theory, both can be + fitted, and Hauppauge suggest an external overrides an + internal. + + That's why we probe 0x1a (~0x34) first. CB + */ + const unsigned short addr_list[] = { + 0x1a, 0x18, 0x4b, 0x64, 0x30, + I2C_CLIENT_END + }; + + memset(&info, 0, sizeof(struct i2c_board_info)); + strlcpy(info.type, "ir-kbd", I2C_NAME_SIZE); + i2c_new_probed_device(&btv->c.i2c_adap, &info, addr_list); + } return btv->i2c_rc; } --- v4l-dvb.orig/linux/drivers/media/video/cx18/cx18-i2c.c 2009-04-04 10:53:15.000000000 +0200 +++ v4l-dvb/linux/drivers/media/video/cx18/cx18-i2c.c 2009-04-04 10:58:36.000000000 +0200 @@ -211,7 +211,32 @@ static struct i2c_algo_bit_data cx18_i2c .timeout = CX18_ALGO_BIT_TIMEOUT*HZ /* jiffies */ }; -/* init + register i2c algo-bit adapter */ +static void init_cx18_i2c_ir(struct cx18 *cx) +{ + struct i2c_board_info info; + /* The external IR receiver is at i2c address 0x34 (0x35 for + reads). Future Hauppauge cards will have an internal + receiver at 0x30 (0x31 for reads). In theory, both can be + fitted, and Hauppauge suggest an external overrides an + internal. + + That's why we probe 0x1a (~0x34) first. CB + */ + const unsigned short addr_list[] = { + 0x1a, 0x18, 0x64, 0x30, + I2C_CLIENT_END + }; + + memset(&info, 0, sizeof(struct i2c_board_info)); + strlcpy(info.type, "ir-kbd", I2C_NAME_SIZE); + + /* The IR receiver device can be on either I2C bus */ + if (i2c_new_probed_device(&cx->i2c_adap[0], &info, addr_list)) + return; + i2c_new_probed_device(&cx->i2c_adap[1], &info, addr_list); +} + +/* init + register i2c adapters + instantiate IR receiver */ int init_cx18_i2c(struct cx18 *cx) { int i, err; @@ -279,6 +304,9 @@ int init_cx18_i2c(struct cx18 *cx) err = i2c_bit_add_bus(&cx->i2c_adap[1]); if (err) goto err_del_bus_0; + + /* Instantiate the IR receiver device, if present */ + init_cx18_i2c_ir(cx); return 0; err_del_bus_0: --- v4l-dvb.orig/linux/drivers/media/video/cx231xx/cx231xx-cards.c 2009-04-04 10:53:08.000000000 +0200 +++ v4l-dvb/linux/drivers/media/video/cx231xx/cx231xx-cards.c 2009-04-04 12:56:26.000000000 +0200 @@ -284,11 +284,6 @@ static void cx231xx_config_tuner(struct /* ----------------------------------------------------------------------- */ void cx231xx_set_ir(struct cx231xx *dev, struct IR_i2c *ir) { - if (disable_ir) { - ir->get_key = NULL; - return; - } - /* detect & configure */ switch (dev->model) { --- v4l-dvb.orig/linux/drivers/media/video/cx23885/cx23885-i2c.c 2009-04-04 10:53:08.000000000 +0200 +++ v4l-dvb/linux/drivers/media/video/cx23885/cx23885-i2c.c 2009-04-04 10:58:36.000000000 +0200 @@ -364,6 +364,18 @@ int cx23885_i2c_register(struct cx23885_ printk(KERN_WARNING "%s: i2c bus %d register FAILED\n", dev->name, bus->nr); + /* Instantiate the IR receiver device, if present */ + if (0 == bus->i2c_rc) { + struct i2c_board_info info; + const unsigned short addr_list[] = { + 0x6b, I2C_CLIENT_END + }; + + memset(&info, 0, sizeof(struct i2c_board_info)); + strlcpy(info.type, "ir-kbd", I2C_NAME_SIZE); + i2c_new_probed_device(&bus->i2c_adap, &info, addr_list); + } + return bus->i2c_rc; } --- v4l-dvb.orig/linux/drivers/media/video/cx88/cx88-i2c.c 2009-04-04 10:53:08.000000000 +0200 +++ v4l-dvb/linux/drivers/media/video/cx88/cx88-i2c.c 2009-04-04 10:58:36.000000000 +0200 @@ -186,6 +186,19 @@ int cx88_i2c_init(struct cx88_core *core do_i2c_scan(core->name,&core->i2c_client); } else printk("%s: i2c register FAILED\n", core->name); + + /* Instantiate the IR receiver device, if present */ + if (0 == core->i2c_rc) { + struct i2c_board_info info; + const unsigned short addr_list[] = { + 0x18, 0x6b, 0x71, + I2C_CLIENT_END + }; + + memset(&info, 0, sizeof(struct i2c_board_info)); + strlcpy(info.type, "ir-kbd", I2C_NAME_SIZE); + i2c_new_probed_device(&core->i2c_adap, &info, addr_list); + } return core->i2c_rc; } --- v4l-dvb.orig/linux/drivers/media/video/em28xx/em28xx-cards.c 2009-04-04 10:57:57.000000000 +0200 +++ v4l-dvb/linux/drivers/media/video/em28xx/em28xx-cards.c 2009-04-04 12:56:26.000000000 +0200 @@ -1904,13 +1904,23 @@ static int em28xx_hint_board(struct em28 } /* ----------------------------------------------------------------------- */ -void em28xx_set_ir(struct em28xx *dev, struct IR_i2c *ir) +void em28xx_register_i2c_ir(struct em28xx *dev) { - if (disable_ir) { - ir->get_key = NULL; - return ; - } + struct i2c_board_info info; + const unsigned short addr_list[] = { + 0x30, 0x47, I2C_CLIENT_END + }; + + if (disable_ir) + return; + + memset(&info, 0, sizeof(struct i2c_board_info)); + strlcpy(info.type, "ir-kbd", I2C_NAME_SIZE); + i2c_new_probed_device(&dev->i2c_adap, &info, addr_list); +} +void em28xx_set_ir(struct em28xx *dev, struct IR_i2c *ir) +{ /* detect & configure */ switch (dev->model) { case (EM2800_BOARD_UNKNOWN): --- v4l-dvb.orig/linux/drivers/media/video/em28xx/em28xx-i2c.c 2009-04-04 10:53:08.000000000 +0200 +++ v4l-dvb/linux/drivers/media/video/em28xx/em28xx-i2c.c 2009-04-04 12:56:26.000000000 +0200 @@ -581,6 +581,9 @@ int em28xx_i2c_register(struct em28xx *d if (i2c_scan) em28xx_do_i2c_scan(dev); + /* Instantiate the IR receiver device, if present */ + em28xx_register_i2c_ir(dev); + return 0; } --- v4l-dvb.orig/linux/drivers/media/video/em28xx/em28xx-input.c 2009-04-04 10:57:57.000000000 +0200 +++ v4l-dvb/linux/drivers/media/video/em28xx/em28xx-input.c 2009-04-04 10:58:36.000000000 +0200 @@ -86,7 +86,7 @@ int em28xx_get_key_terratec(struct IR_i2 unsigned char b; /* poll IR chip */ - if (1 != i2c_master_recv(&ir->c, &b, 1)) { + if (1 != i2c_master_recv(ir->c, &b, 1)) { i2cdprintk("read error\n"); return -EIO; } @@ -115,7 +115,7 @@ int em28xx_get_key_em_haup(struct IR_i2c unsigned char code; /* poll IR chip */ - if (2 != i2c_master_recv(&ir->c, buf, 2)) + if (2 != i2c_master_recv(ir->c, buf, 2)) return -EIO; /* Does eliminate repeated parity code */ @@ -153,7 +153,7 @@ int em28xx_get_key_pinnacle_usb_grey(str /* poll IR chip */ - if (3 != i2c_master_recv(&ir->c, buf, 3)) { + if (3 != i2c_master_recv(ir->c, buf, 3)) { i2cdprintk("read error\n"); return -EIO; } --- v4l-dvb.orig/linux/drivers/media/video/em28xx/em28xx.h 2009-04-04 10:53:08.000000000 +0200 +++ v4l-dvb/linux/drivers/media/video/em28xx/em28xx.h 2009-04-04 12:56:26.000000000 +0200 @@ -648,6 +648,7 @@ extern void em28xx_card_setup(struct em2 extern struct em28xx_board em28xx_boards[]; extern struct usb_device_id em28xx_id_table[]; extern const unsigned int em28xx_bcount; +void em28xx_register_i2c_ir(struct em28xx *dev); void em28xx_set_ir(struct em28xx *dev, struct IR_i2c *ir); int em28xx_tuner_callback(void *ptr, int component, int command, int arg); void em28xx_release_resources(struct em28xx *dev); --- v4l-dvb.orig/linux/drivers/media/video/ir-kbd-i2c.c 2009-04-04 10:57:57.000000000 +0200 +++ v4l-dvb/linux/drivers/media/video/ir-kbd-i2c.c 2009-04-04 12:56:26.000000000 +0200 @@ -75,7 +75,7 @@ static int get_key_haup_common(struct IR int start, range, toggle, dev, code, ircode; /* poll IR chip */ - if (size != i2c_master_recv(&ir->c,buf,size)) + if (size != i2c_master_recv(ir->c, buf, size)) return -EIO; /* split rc5 data block ... */ @@ -138,7 +138,7 @@ static int get_key_pixelview(struct IR_i unsigned char b; /* poll IR chip */ - if (1 != i2c_master_recv(&ir->c,&b,1)) { + if (1 != i2c_master_recv(ir->c, &b, 1)) { dprintk(1,"read error\n"); return -EIO; } @@ -152,7 +152,7 @@ static int get_key_pv951(struct IR_i2c * unsigned char b; /* poll IR chip */ - if (1 != i2c_master_recv(&ir->c,&b,1)) { + if (1 != i2c_master_recv(ir->c, &b, 1)) { dprintk(1,"read error\n"); return -EIO; } @@ -172,7 +172,7 @@ static int get_key_fusionhdtv(struct IR_ unsigned char buf[4]; /* poll IR chip */ - if (4 != i2c_master_recv(&ir->c,buf,4)) { + if (4 != i2c_master_recv(ir->c, buf, 4)) { dprintk(1,"read error\n"); return -EIO; } @@ -196,7 +196,7 @@ static int get_key_knc1(struct IR_i2c *i unsigned char b; /* poll IR chip */ - if (1 != i2c_master_recv(&ir->c,&b,1)) { + if (1 != i2c_master_recv(ir->c, &b, 1)) { dprintk(1,"read error\n"); return -EIO; } @@ -223,12 +223,12 @@ static int get_key_avermedia_cardbus(str u32 *ir_key, u32 *ir_raw) { unsigned char subaddr, key, keygroup; - struct i2c_msg msg[] = { { .addr = ir->c.addr, .flags = 0, + struct i2c_msg msg[] = { { .addr = ir->c->addr, .flags = 0, .buf = &subaddr, .len = 1}, - { .addr = ir->c.addr, .flags = I2C_M_RD, + { .addr = ir->c->addr, .flags = I2C_M_RD, .buf = &key, .len = 1} }; subaddr = 0x0d; - if (2 != i2c_transfer(ir->c.adapter, msg, 2)) { + if (2 != i2c_transfer(ir->c->adapter, msg, 2)) { dprintk(1, "read error\n"); return -EIO; } @@ -238,7 +238,7 @@ static int get_key_avermedia_cardbus(str subaddr = 0x0b; msg[1].buf = &keygroup; - if (2 != i2c_transfer(ir->c.adapter, msg, 2)) { + if (2 != i2c_transfer(ir->c->adapter, msg, 2)) { dprintk(1, "read error\n"); return -EIO; } @@ -295,7 +295,7 @@ static void ir_work(struct work_struct * /* MSI TV@nywhere Plus requires more frequent polling otherwise it will miss some keypresses */ - if (ir->c.adapter->id == I2C_HW_SAA7134 && ir->c.addr == 0x30) + if (ir->c->adapter->id == I2C_HW_SAA7134 && ir->c->addr == 0x30) polling_interval = 50; ir_key_poll(ir); @@ -304,34 +304,15 @@ static void ir_work(struct work_struct * /* ----------------------------------------------------------------------- */ -static int ir_attach(struct i2c_adapter *adap, int addr, - unsigned short flags, int kind); -static int ir_detach(struct i2c_client *client); -static int ir_probe(struct i2c_adapter *adap); - -static struct i2c_driver driver = { - .driver = { - .name = "ir-kbd-i2c", - }, - .id = I2C_DRIVERID_INFRARED, - .attach_adapter = ir_probe, - .detach_client = ir_detach, -}; - -static struct i2c_client client_template = -{ - .name = "unset", - .driver = &driver -}; - -static int ir_attach(struct i2c_adapter *adap, int addr, - unsigned short flags, int kind) +static int ir_probe(struct i2c_client *client, const struct i2c_device_id *id) { IR_KEYTAB_TYPE *ir_codes = NULL; char *name; int ir_type; struct IR_i2c *ir; struct input_dev *input_dev; + struct i2c_adapter *adap = client->adapter; + unsigned short addr = client->addr; int err; ir = kzalloc(sizeof(struct IR_i2c),GFP_KERNEL); @@ -341,14 +322,9 @@ static int ir_attach(struct i2c_adapter goto err_out_free; } - ir->c = client_template; + ir->c = client; ir->input = input_dev; - - ir->c.adapter = adap; - ir->c.addr = addr; - snprintf(ir->c.name, sizeof(ir->c.name), "ir-kbd"); - - i2c_set_clientdata(&ir->c, ir); + i2c_set_clientdata(client, ir); switch(addr) { case 0x64: @@ -423,24 +399,9 @@ static int ir_attach(struct i2c_adapter snprintf(ir->name, sizeof(ir->name), "i2c IR (%s)", name); ir->ir_codes = ir_codes; - /* register i2c device - * At device register, IR codes may be changed to be - * board dependent. - */ - err = i2c_attach_client(&ir->c); - if (err) - goto err_out_free; - - /* If IR not supported or disabled, unregisters driver */ - if (ir->get_key == NULL) { - err = -ENODEV; - goto err_out_detach; - } - - /* Phys addr can only be set after attaching (for ir->c.dev) */ snprintf(ir->phys, sizeof(ir->phys), "%s/%s/ir0", - dev_name(&ir->c.adapter->dev), - dev_name(&ir->c.dev)); + dev_name(&adap->dev), + dev_name(&client->dev)); /* init + register input device */ ir_input_init(input_dev, &ir->ir, ir_type, ir->ir_codes); @@ -450,7 +411,7 @@ static int ir_attach(struct i2c_adapter err = input_register_device(ir->input); if (err) - goto err_out_detach; + goto err_out_free; printk(DEVNAME ": %s detected at %s [%s]\n", ir->input->name, ir->input->phys, adap->name); @@ -465,135 +426,40 @@ static int ir_attach(struct i2c_adapter return 0; - err_out_detach: - i2c_detach_client(&ir->c); err_out_free: input_free_device(input_dev); kfree(ir); return err; } -static int ir_detach(struct i2c_client *client) +static int ir_remove(struct i2c_client *client) { struct IR_i2c *ir = i2c_get_clientdata(client); /* kill outstanding polls */ cancel_delayed_work_sync(&ir->work); - /* unregister devices */ + /* unregister device */ input_unregister_device(ir->input); - i2c_detach_client(&ir->c); /* free memory */ kfree(ir); return 0; } -static int ir_probe(struct i2c_adapter *adap) -{ - - /* The external IR receiver is at i2c address 0x34 (0x35 for - reads). Future Hauppauge cards will have an internal - receiver at 0x30 (0x31 for reads). In theory, both can be - fitted, and Hauppauge suggest an external overrides an - internal. - - That's why we probe 0x1a (~0x34) first. CB - */ - - static const int probe_bttv[] = { 0x1a, 0x18, 0x4b, 0x64, 0x30, -1}; - static const int probe_saa7134[] = { 0x7a, 0x47, 0x71, 0x2d, -1 }; - static const int probe_em28XX[] = { 0x30, 0x47, -1 }; - static const int probe_cx88[] = { 0x18, 0x6b, 0x71, -1 }; - static const int probe_cx23885[] = { 0x6b, -1 }; - const int *probe; - struct i2c_msg msg = { - .flags = I2C_M_RD, - .len = 0, - .buf = NULL, - }; - int i, rc; - - switch (adap->id) { - case I2C_HW_B_BT848: - probe = probe_bttv; - break; - case I2C_HW_B_CX2341X: - probe = probe_bttv; - break; - case I2C_HW_SAA7134: - probe = probe_saa7134; - break; - case I2C_HW_B_EM28XX: - probe = probe_em28XX; - break; - case I2C_HW_B_CX2388x: - probe = probe_cx88; - break; - case I2C_HW_B_CX23885: - probe = probe_cx23885; - break; - default: - return 0; - } - - for (i = 0; -1 != probe[i]; i++) { - msg.addr = probe[i]; - rc = i2c_transfer(adap, &msg, 1); - dprintk(1,"probe 0x%02x @ %s: %s\n", - probe[i], adap->name, - (1 == rc) ? "yes" : "no"); - if (1 == rc) { - ir_attach(adap, probe[i], 0, 0); - return 0; - } - } - - /* Special case for MSI TV@nywhere Plus remote */ - if (adap->id == I2C_HW_SAA7134) { - u8 temp; - - /* MSI TV@nywhere Plus controller doesn't seem to - respond to probes unless we read something from - an existing device. Weird... */ - - msg.addr = 0x50; - rc = i2c_transfer(adap, &msg, 1); - dprintk(1, "probe 0x%02x @ %s: %s\n", - msg.addr, adap->name, - (1 == rc) ? "yes" : "no"); - - /* Now do the probe. The controller does not respond - to 0-byte reads, so we use a 1-byte read instead. */ - msg.addr = 0x30; - msg.len = 1; - msg.buf = &temp; - rc = i2c_transfer(adap, &msg, 1); - dprintk(1, "probe 0x%02x @ %s: %s\n", - msg.addr, adap->name, - (1 == rc) ? "yes" : "no"); - if (1 == rc) - ir_attach(adap, msg.addr, 0, 0); - } - - /* Special case for AVerMedia Cardbus remote */ - if (adap->id == I2C_HW_SAA7134) { - unsigned char subaddr, data; - struct i2c_msg msg[] = { { .addr = 0x40, .flags = 0, - .buf = &subaddr, .len = 1}, - { .addr = 0x40, .flags = I2C_M_RD, - .buf = &data, .len = 1} }; - subaddr = 0x0d; - rc = i2c_transfer(adap, msg, 2); - dprintk(1, "probe 0x%02x/0x%02x @ %s: %s\n", - msg[0].addr, subaddr, adap->name, - (2 == rc) ? "yes" : "no"); - if (2 == rc) - ir_attach(adap, msg[0].addr, 0, 0); - } +static const struct i2c_device_id ir_kbd_id[] = { + { "ir-kbd", 0 }, + { } +}; - return 0; -} +static struct i2c_driver driver = { + .driver = { + .name = "ir-kbd-i2c", + }, + .probe = ir_probe, + .remove = ir_remove, + .id_table = ir_kbd_id, +}; /* ----------------------------------------------------------------------- */ --- v4l-dvb.orig/linux/drivers/media/video/ivtv/ivtv-i2c.c 2009-04-04 10:53:08.000000000 +0200 +++ v4l-dvb/linux/drivers/media/video/ivtv/ivtv-i2c.c 2009-04-04 10:58:36.000000000 +0200 @@ -589,9 +589,11 @@ static struct i2c_client ivtv_i2c_client .name = "ivtv internal", }; -/* init + register i2c algo-bit adapter */ +/* init + register i2c adapter + instantiate IR receiver */ int init_ivtv_i2c(struct ivtv *itv) { + int retval; + IVTV_DEBUG_I2C("i2c init\n"); /* Sanity checks for the I2C hardware arrays. They must be the @@ -631,9 +633,32 @@ int init_ivtv_i2c(struct ivtv *itv) ivtv_setsda(itv, 1); if (itv->options.newi2c > 0) - return i2c_add_adapter(&itv->i2c_adap); + retval = i2c_add_adapter(&itv->i2c_adap); else - return i2c_bit_add_bus(&itv->i2c_adap); + retval = i2c_bit_add_bus(&itv->i2c_adap); + + /* Instantiate the IR receiver device, if present */ + if (retval == 0) { + struct i2c_board_info info; + /* The external IR receiver is at i2c address 0x34 (0x35 for + reads). Future Hauppauge cards will have an internal + receiver at 0x30 (0x31 for reads). In theory, both can be + fitted, and Hauppauge suggest an external overrides an + internal. + + That's why we probe 0x1a (~0x34) first. CB + */ + const unsigned short addr_list[] = { + 0x1a, 0x18, 0x64, 0x30, + I2C_CLIENT_END + }; + + memset(&info, 0, sizeof(struct i2c_board_info)); + strlcpy(info.type, "ir-kbd", I2C_NAME_SIZE); + i2c_new_probed_device(&itv->i2c_adap, &info, addr_list); + } + + return retval; } void exit_ivtv_i2c(struct ivtv *itv) --- v4l-dvb.orig/linux/drivers/media/video/pvrusb2/pvrusb2-i2c-core.c 2009-04-04 10:53:08.000000000 +0200 +++ v4l-dvb/linux/drivers/media/video/pvrusb2/pvrusb2-i2c-core.c 2009-04-04 10:58:36.000000000 +0200 @@ -649,6 +649,27 @@ static void do_i2c_scan(struct pvr2_hdw printk(KERN_INFO "%s: i2c scan done.\n", hdw->name); } +static void pvr2_i2c_register_ir(struct i2c_adapter *i2c_adap) +{ + struct i2c_board_info info; + /* The external IR receiver is at i2c address 0x34 (0x35 for + reads). Future Hauppauge cards will have an internal + receiver at 0x30 (0x31 for reads). In theory, both can be + fitted, and Hauppauge suggest an external overrides an + internal. + + That's why we probe 0x1a (~0x34) first. CB + */ + const unsigned short addr_list[] = { + 0x1a, 0x18, 0x4b, 0x64, 0x30, + I2C_CLIENT_END + }; + + memset(&info, 0, sizeof(struct i2c_board_info)); + strlcpy(info.type, "ir-kbd", I2C_NAME_SIZE); + i2c_new_probed_device(i2c_adap, &info, addr_list); +} + void pvr2_i2c_core_init(struct pvr2_hdw *hdw) { unsigned int idx; @@ -696,6 +717,9 @@ void pvr2_i2c_core_init(struct pvr2_hdw } } if (i2c_scan) do_i2c_scan(hdw); + + /* Instantiate the IR receiver device, if present */ + pvr2_i2c_register_ir(&hdw->i2c_adap); } void pvr2_i2c_core_done(struct pvr2_hdw *hdw) --- v4l-dvb.orig/linux/drivers/media/video/saa7134/saa7134-i2c.c 2009-04-04 10:53:08.000000000 +0200 +++ v4l-dvb/linux/drivers/media/video/saa7134/saa7134-i2c.c 2009-04-04 12:56:26.000000000 +0200 @@ -444,6 +444,9 @@ int saa7134_i2c_register(struct saa7134_ saa7134_i2c_eeprom(dev,dev->eedata,sizeof(dev->eedata)); if (i2c_scan) do_i2c_scan(dev->name,&dev->i2c_client); + + /* Instantiate the IR receiver device, if present */ + saa7134_probe_i2c_ir(dev); return 0; } --- v4l-dvb.orig/linux/drivers/media/video/saa7134/saa7134-input.c 2009-04-04 10:57:57.000000000 +0200 +++ v4l-dvb/linux/drivers/media/video/saa7134/saa7134-input.c 2009-04-04 12:56:26.000000000 +0200 @@ -134,10 +134,10 @@ static int get_key_msi_tvanywhere_plus(s int gpio; /* <dev> is needed to access GPIO. Used by the saa_readl macro. */ - struct saa7134_dev *dev = ir->c.adapter->algo_data; + struct saa7134_dev *dev = ir->c->adapter->algo_data; if (dev == NULL) { dprintk("get_key_msi_tvanywhere_plus: " - "gir->c.adapter->algo_data is NULL!\n"); + "gir->c->adapter->algo_data is NULL!\n"); return -EIO; } @@ -156,7 +156,7 @@ static int get_key_msi_tvanywhere_plus(s /* GPIO says there is a button press. Get it. */ - if (1 != i2c_master_recv(&ir->c, &b, 1)) { + if (1 != i2c_master_recv(ir->c, &b, 1)) { i2cdprintk("read error\n"); return -EIO; } @@ -179,7 +179,7 @@ static int get_key_purpletv(struct IR_i2 unsigned char b; /* poll IR chip */ - if (1 != i2c_master_recv(&ir->c,&b,1)) { + if (1 != i2c_master_recv(ir->c, &b, 1)) { i2cdprintk("read error\n"); return -EIO; } @@ -202,7 +202,7 @@ static int get_key_hvr1110(struct IR_i2c unsigned char buf[5], cod4, code3, code4; /* poll IR chip */ - if (5 != i2c_master_recv(&ir->c,buf,5)) + if (5 != i2c_master_recv(ir->c, buf, 5)) return -EIO; cod4 = buf[4]; @@ -224,7 +224,7 @@ static int get_key_beholdm6xx(struct IR_ unsigned char data[12]; u32 gpio; - struct saa7134_dev *dev = ir->c.adapter->algo_data; + struct saa7134_dev *dev = ir->c->adapter->algo_data; /* rising SAA7134_GPIO_GPRESCAN reads the status */ saa_clearb(SAA7134_GPIO_GPMODE3, SAA7134_GPIO_GPRESCAN); @@ -235,9 +235,9 @@ static int get_key_beholdm6xx(struct IR_ if (0x400000 & ~gpio) return 0; /* No button press */ - ir->c.addr = 0x5a >> 1; + ir->c->addr = 0x5a >> 1; - if (12 != i2c_master_recv(&ir->c, data, 12)) { + if (12 != i2c_master_recv(ir->c, data, 12)) { i2cdprintk("read error\n"); return -EIO; } @@ -267,7 +267,7 @@ static int get_key_pinnacle(struct IR_i2 unsigned int start = 0,parity = 0,code = 0; /* poll IR chip */ - if (4 != i2c_master_recv(&ir->c, b, 4)) { + if (4 != i2c_master_recv(ir->c, b, 4)) { i2cdprintk("read error\n"); return -EIO; } @@ -682,14 +682,76 @@ void saa7134_input_fini(struct saa7134_d dev->remote = NULL; } -void saa7134_set_i2c_ir(struct saa7134_dev *dev, struct IR_i2c *ir) +void saa7134_probe_i2c_ir(struct saa7134_dev *dev) { + struct i2c_board_info info; + const unsigned short addr_list[] = { + 0x7a, 0x47, 0x71, 0x2d, + I2C_CLIENT_END + }; + + const unsigned short addr_list_msi[] = { + 0x30, I2C_CLIENT_END + }; + struct i2c_msg msg_msi = { + .addr = 0x50, + .flags = I2C_M_RD, + .len = 0, + .buf = NULL, + }; + + unsigned char subaddr, data; + struct i2c_msg msg_avermedia[] = { { + .addr = 0x40, + .flags = 0, + .len = 1, + .buf = &subaddr, + }, { + .addr = 0x40, + .flags = I2C_M_RD, + .len = 1, + .buf = &data, + } }; + + struct i2c_client *client; + int rc; + if (disable_ir) { - dprintk("Found supported i2c remote, but IR has been disabled\n"); - ir->get_key=NULL; + dprintk("IR has been disabled, not probing for i2c remote\n"); + return; + } + + memset(&info, 0, sizeof(struct i2c_board_info)); + strlcpy(info.type, "ir-kbd", I2C_NAME_SIZE); + client = i2c_new_probed_device(&dev->i2c_adap, &info, addr_list); + if (client) return; + + /* MSI TV@nywhere Plus controller doesn't seem to + respond to probes unless we read something from + an existing device. Weird... */ + rc = i2c_transfer(&dev->i2c_adap, &msg_msi, 1); + dprintk(KERN_DEBUG "probe 0x%02x @ %s: %s\n", + msg_msi.addr, dev->i2c_adap.name, + (1 == rc) ? "yes" : "no"); + client = i2c_new_probed_device(&dev->i2c_adap, &info, addr_list_msi); + if (client) + return; + + /* Special case for AVerMedia Cardbus remote */ + subaddr = 0x0d; + rc = i2c_transfer(&dev->i2c_adap, msg_avermedia, 2); + dprintk(KERN_DEBUG "probe 0x%02x/0x%02x @ %s: %s\n", + msg_avermedia[0].addr, subaddr, dev->i2c_adap.name, + (2 == rc) ? "yes" : "no"); + if (2 == rc) { + info.addr = msg_avermedia[0].addr; + i2c_new_device(&dev->i2c_adap, &info); } +} +void saa7134_set_i2c_ir(struct saa7134_dev *dev, struct IR_i2c *ir) +{ switch (dev->board) { case SAA7134_BOARD_PINNACLE_PCTV_110i: case SAA7134_BOARD_PINNACLE_PCTV_310i: --- v4l-dvb.orig/linux/drivers/media/video/saa7134/saa7134.h 2009-04-04 10:53:08.000000000 +0200 +++ v4l-dvb/linux/drivers/media/video/saa7134/saa7134.h 2009-04-04 12:56:26.000000000 +0200 @@ -791,6 +791,7 @@ void saa7134_irq_oss_done(struct saa7134 int saa7134_input_init1(struct saa7134_dev *dev); void saa7134_input_fini(struct saa7134_dev *dev); void saa7134_input_irq(struct saa7134_dev *dev); +void saa7134_probe_i2c_ir(struct saa7134_dev *dev); void saa7134_set_i2c_ir(struct saa7134_dev *dev, struct IR_i2c *ir); void saa7134_ir_start(struct saa7134_dev *dev, struct card_ir *ir); void saa7134_ir_stop(struct saa7134_dev *dev); --- v4l-dvb.orig/linux/drivers/media/video/usbvision/usbvision-i2c.c 2009-04-04 10:53:08.000000000 +0200 +++ v4l-dvb/linux/drivers/media/video/usbvision/usbvision-i2c.c 2009-04-04 10:58:36.000000000 +0200 @@ -211,6 +211,27 @@ static struct i2c_algorithm usbvision_al /* ----------------------------------------------------------------------- */ static struct i2c_adapter i2c_adap_template; +static void usbvision_i2c_register_ir(struct i2c_adapter *adap) +{ + struct i2c_board_info info; + /* The external IR receiver is at i2c address 0x34 (0x35 for + reads). Future Hauppauge cards will have an internal + receiver at 0x30 (0x31 for reads). In theory, both can be + fitted, and Hauppauge suggest an external overrides an + internal. + + That's why we probe 0x1a (~0x34) first. CB + */ + const unsigned short addr_list[] = { + 0x1a, 0x18, 0x4b, 0x64, 0x30, + I2C_CLIENT_END + }; + + memset(&info, 0, sizeof(struct i2c_board_info)); + strlcpy(info.type, "ir-kbd", I2C_NAME_SIZE); + i2c_new_probed_device(adap, &info, addr_list); +} + int usbvision_i2c_register(struct usb_usbvision *usbvision) { static unsigned short saa711x_addrs[] = { @@ -277,6 +298,9 @@ int usbvision_i2c_register(struct usb_us } } + /* Instantiate the IR receiver device, if present */ + usbvision_i2c_register_ir(&usbvision->i2c_adap); + return 0; } --- v4l-dvb.orig/linux/include/media/ir-kbd-i2c.h 2009-04-04 10:57:57.000000000 +0200 +++ v4l-dvb/linux/include/media/ir-kbd-i2c.h 2009-04-04 12:56:26.000000000 +0200 @@ -7,7 +7,7 @@ struct IR_i2c; struct IR_i2c { IR_KEYTAB_TYPE *ir_codes; - struct i2c_client c; + struct i2c_client *c; struct input_dev *input; struct ir_input_state ir;
Let card drivers probe for IR receiver devices and instantiate them if found. Ultimately it would be better if we could stop probing completely, but I suspect this won't be possible for all card types. There's certainly room for cleanups. For example, some drivers are sharing I2C adapter IDs, so they also had to share the list of I2C addresses being probed for an IR receiver. Now that each driver explicitly says which addresses should be probed, maybe some addresses can be dropped from some drivers. Also, the special cases in saa7134-i2c should probably be handled on a per-board basis. This would be more efficient and less risky than always probing extra addresses on all boards. I'll give it a try later. Signed-off-by: Jean Delvare <khali@linux-fr.org> Cc: Mauro Carvalho Chehab <mchehab@infradead.org> Cc: Hans Verkuil <hverkuil@xs4all.nl> Cc: Andy Walls <awalls@radix.net> Cc: Mike Isely <isely@pobox.com> --- linux/drivers/media/video/bt8xx/bttv-i2c.c | 21 + linux/drivers/media/video/cx18/cx18-i2c.c | 30 ++ linux/drivers/media/video/cx231xx/cx231xx-cards.c | 5 linux/drivers/media/video/cx23885/cx23885-i2c.c | 12 + linux/drivers/media/video/cx88/cx88-i2c.c | 13 + linux/drivers/media/video/em28xx/em28xx-cards.c | 20 + linux/drivers/media/video/em28xx/em28xx-i2c.c | 3 linux/drivers/media/video/em28xx/em28xx-input.c | 6 linux/drivers/media/video/em28xx/em28xx.h | 1 linux/drivers/media/video/ir-kbd-i2c.c | 198 ++---------------- linux/drivers/media/video/ivtv/ivtv-i2c.c | 31 ++ linux/drivers/media/video/pvrusb2/pvrusb2-i2c-core.c | 24 ++ linux/drivers/media/video/saa7134/saa7134-i2c.c | 3 linux/drivers/media/video/saa7134/saa7134-input.c | 86 ++++++- linux/drivers/media/video/saa7134/saa7134.h | 1 linux/drivers/media/video/usbvision/usbvision-i2c.c | 24 ++ linux/include/media/ir-kbd-i2c.h | 2 17 files changed, 284 insertions(+), 196 deletions(-)