freemangordon | tmlind: seems there is some issue with schedule_hrtimeout(), it always waits ~300 us more than what requested | 08:06 |
---|---|---|
freemangordon | I am afraid we have way bigger issue on n900 than non-working IRTX | 08:07 |
arno11 | fremangordon: what kind of issue ? | 08:42 |
freemangordon | arno11: hard to tell how it manifests itself from user POV, but *every* schedule_hrtimeout() call (and thus every usleep_range() call and who knows what else) sleeps at least ~300 us more than the requested delay | 08:46 |
arno11 | could explain general latency issues ? | 08:51 |
freemangordon | mhm | 08:51 |
freemangordon | this is used all over the place https://elixir.bootlin.com/linux/v6.1.48/C/ident/usleep_range | 08:52 |
freemangordon | and this is seen as posix timers from userspace | 08:53 |
freemangordon | so, basically there is no way to sleep for less that 300 us, unless I am doing something stupid when measuring | 08:53 |
freemangordon | lets see if tmlind has some idea | 08:54 |
arno11 | ok | 08:55 |
arno11 | wow indeed it is used all over the place... | 08:59 |
sicelo | these are omap3 problems? just odd that it'd be N900-specific | 09:34 |
freemangordon | mhm | 10:48 |
freemangordon | lemme boot my d4 to check what the latency is | 10:48 |
uvos__ | Wizzup: tmlind: sure i know its exposed on alsa interface, the problem is this isent wired up to ofonos dtmf interface (nor is it how ofono expects to execute dtmf tones, this seams to go via qmi) | 11:00 |
uvos__ | so ideal we would add support for it in ofono, since sphone should try to avoid being the hw abstraction layer | 11:00 |
uvos__ | dtmf via qmi seams to not work on d4 | 11:01 |
freemangordon | why not use tonegend? | 11:03 |
freemangordon | https://github.com/maemo-leste/maemo-multimedia-tone-generator/blob/master/src/dtmf.c | 11:05 |
uvos__ | this is impossible | 11:05 |
freemangordon | why? | 11:05 |
uvos__ | we _can_not_ send any audio to the modem via cpu in a call | 11:05 |
freemangordon | we can't send audio? | 11:05 |
freemangordon | ah | 11:05 |
uvos__ | the modem is given the microphone | 11:05 |
uvos__ | it records itself | 11:05 |
freemangordon | ok, but what if we want to record the audio? | 11:05 |
uvos__ | we can record | 11:06 |
freemangordon | but not play? | 11:06 |
uvos__ | there a special dai | 11:06 |
uvos__ | jup | 11:06 |
freemangordon | ugh :( | 11:06 |
uvos__ | we should instead wireup tonegend to ofono really | 11:06 |
uvos__ | so the interfaces can be the same on any deivce | 11:06 |
freemangordon | mhm | 11:06 |
freemangordon | but, how would that help in d4 case? | 11:07 |
uvos__ | well it would allow sphone to not worry about if dtmf is implemented in hw or software | 11:07 |
freemangordon | ok, d4 has the same latency issue | 11:07 |
freemangordon | uvos__: ah, right | 11:07 |
freemangordon | we just ask ofono for dtmf | 11:07 |
uvos__ | yes | 11:07 |
uvos__ | "ok, d4 has the same latency issue" ugh | 11:11 |
uvos__ | thats terrible | 11:11 |
uvos__ | xD | 11:11 |
freemangordon | sudo cyclictest -s --mlockall --priority=80 | 11:13 |
freemangordon | T: 0 ( 6044) P:80 I:1000 C: 16991 Min: 7 Act: 1014 Avg: 401 Max: 1990 | 11:13 |
uvos__ | i wonder if kselftest would have caught this | 11:13 |
uvos__ | and if we should be running it | 11:13 |
freemangordon | what is kselftest? | 11:14 |
freemangordon | ah | 11:14 |
uvos__ | the kernels test suite | 11:14 |
freemangordon | ok, but can we do when cross-compile? | 11:14 |
uvos__ | hmm dont see why not, you compile it for arm32 and run the test on device | 11:15 |
freemangordon | any idea which specific test sshall be run? | 11:16 |
uvos__ | well the idea would be to run the entire suite, i dont really know if it would have caught this specific issue | 11:17 |
freemangordon | https://pastebin.com/yueUp5gN | 11:19 |
uvos__ | neat | 12:15 |
arno11 | bencoh: Wizzup: weird cases with calls happen with N900 too: sometimes ringtone is not starting at all or too late. and depending of sim card/provider sometimes tone is not working at all | 19:28 |
arno11 | i didn't notice the sms bug | 19:29 |
arno11 | *btw | 19:30 |
Wizzup | arno11: yeah this is d4 | 19:32 |
arno11 | yeah i know | 19:34 |
arno11 | just to compare | 19:34 |
sicelo | Wizzup: https://github.com/maemo-leste/libicd-network-ofono/pull/1/commits/944850b6444feb511dc97652d9987af994fb8ecc | 23:25 |
sicelo | i hope this change is correct? i think testing for dns[0] again was a typo. | 23:26 |
sicelo | anyway, the one i'm more interested in is, https://github.com/maemo-leste/libicd-network-ofono/pull/1/commits/803f91fdd9bd1e68383c93b0f52190ad4e72732f | 23:27 |
sicelo | will be happy to have your review and eventually merge. it's something that does bug me a lot | 23:28 |
sicelo | arno11: ofono in general is rock-solid on N900, so yes :-) | 23:29 |
Generated by irclog2html.py 2.17.0 by Marius Gedminas - find it at https://mg.pov.lt/irclog2html/!