Am 05.01.2013 16:06, schrieb Mauro Carvalho Chehab: > Em Sat, 05 Jan 2013 14:22:08 +0100 > Frank Schäfer <fschaefer.oss@xxxxxxxxxxxxxx> escreveu: > >> Am 04.01.2013 22:15, schrieb Mauro Carvalho Chehab: >>> Frank pointed that IR was not working with I2C devices. So, I took some >>> time to fix them. >>> >>> Tested with Hauppauge WinTV USB2. >>> >>> Mauro Carvalho Chehab (4): >>> [media] em28xx: initialize button/I2C IR earlier >>> [media] em28xx: autoload em28xx-rc if the device has an I2C IR >>> [media] em28xx: simplify IR names on I2C devices >>> [media] em28xx: tell ir-kbd-i2c that WinTV uses an RC5 protocol >>> >>> drivers/media/usb/em28xx/em28xx-cards.c | 2 +- >>> drivers/media/usb/em28xx/em28xx-input.c | 29 ++++++++++++++++------------- >>> 2 files changed, 17 insertions(+), 14 deletions(-) >>> >> While these patches make I2C IR remote controls working again, they >> leave several issues unaddressed which should really be fixed: >> 1) the i2c client isn't unregistered on module unload. This was the >> reason for patch 2 in my series. There is also a FIXME comment about >> this in em28xx_release_resources() (although this is the wrong place to >> do it). > AFAIKT, this is not really needed, as the I2C clients are unregistered > when the I2C bus is unregistered. > > So, a device disconnect will release it. Also, an em28xx driver unload. > > The only difference might be if just ir-kbd-i2c and em28xx-rc are > unloaded, but em28xx is still loaded, but I think that, even on this > case, calling the .release code for an I2C bus will release it. > > So, I don't see any need for such patch. I might be wrong, of course, but, > in order to proof that a release code is needed, you'll need to check if > some memory are lost after module load/unload. Mauro, just because code luckily 'works' in the current constellation, it isn't necessarily good code. It's this kind of issues that can easily cause regressions if the code changes somewhere else. em28xx-input registers the i2c device, so it should unregister it on uninit/close/unload, too. Pretty simple and easy to fix (+5 - 2 = 3 additional lines of code). >> 2) there is no error checking in em28xx_register_i2c_ir(). >> em28xx_ir_init should really bail out if no i2c device is found. > A failure to initialize IR should not be fatal for the driver, as the > rest of the hardware still works. I'm talking about em28xx-input and the RC part only. Of course not the whole driver. Do you really want to load module ir-kbd-i2c even though there is no device ? > Also, there's no way to warrant that the I2C code is actually running, > as ir-i2c-kbd may not even be compiled. > > So, returning 0 there doesn't mean that IR is working. You can check the success of request_module. The whole thing is really easy to fix, I fail to see why you don't want to do it. >> 3) All RC maps should be assigned at the same place, no matter if the >> receiver/demodulator is built in or external. Spreading them over the >> code is inconsistent and makes the code bug prone. > I don't agree. It is better to keep RC maps for those devices together > with the RC protocol setting, get_key config, etc. At boards config, > it is very easy to identify I2C IR's, as there's an special field there > to mark those devices (has_ir_i2c). So, if the board has_ir_i2c, the > IR config is inside em28xx-input. ... which is exactly what made it so easy to cause this regression !!! It's not obvious for programmers that no RC map has to be specified for i2c RCs in the board data. It's also not obvious that em28xx-input silently overwrites the rc-map assigned at board level. In general, it's not obvious that two completely different code areas have to be touched for these devices. That's why we really should avoid those board specific code parts spread all over the driver as much as possible. In case of the RC map it's really easy. I also fail to see what you would loose in em28xx-input. We would still assign the RC map to dev->init_data. If you prefer seeing the used RC map in the em28xx-input code directly, then the same should apply for devices with built-in IR recceiver/decoder (which means moving all rc-map assignments to em28xx-input). You could also get rid of field ir_codes this way. > That's the same logic that it is > there for has_dvb: if this field is true, the DVB specifics is inside > em28xx-dvb. Different case. Avoiding board specific code parts there likely isn't possible (and reasonable), although it should be the goal. >> 4) the list of known i2c devices in em28xx-i2c.c misses client address >> 0x3e >> 1 = 0x1f. See client list in em28xx_register_i2c_ir(). > Ok. Separate patch, please. Coming soon. >> 5) there should be a warning message for the case that we call >> ir-kbd-i2c with an unknown rc device. > Why? All boards with has_ir_i2c have entries there. I double-checked. > Adding will just bloat the code with no reason. We just need to take > care if we get a patch adding I2C IR support for an old card, to be > sure that data is filled on both places. > > Considering that we don't receive any IR I2C code for several years, > and that newer devices won't use that part of the code, it seems highly > unlikely that such code would be ever used. Yes, this is indeed more a theoretical thing. I think good code should have such a warning, but that's also matter of taste. Talking about code bloating seems a bit exaggerated to me, we are talking about 1 or 2 extra lines of code. >> 6) because we use our own key polling functions with ir-kbd-i2c, we >> should also select the polling interval value manually. That makes >> things consistent and avoids confusion. > I disagree. The polling interval is mainly dictated by the RC protocol > used (e. g. the minimal time for a repeat code) and by the speed that > users can type things. It is typically ~100 ms everywhere, except when > there are some exceptional cases, like GPIO polling. Yepp, and that's the responsibility of the driver. em28xx-input and not ir-kbd-i2c ! The driver that supplies the key polling functions. > Regards, > Mauro Mauro, I would like to repeat what I've already said above: just because code 'works' in a specific constellation, it isn't necessarily good code. The fact that this code has been suffering from a big fat regression for a long time should really draw our attention ! So let's try to avoid this happening again by fixing the issues the code has, so that it becomes much more robust. I'm offering to send patches. Regards, Frank -- To unsubscribe from this list: send the line "unsubscribe linux-media" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
