libera/#maemo-leste/ Friday, 2024-07-12

uvosis dapm simply broken?00:06
WizzupI doubt that it is broken in 6.600:07
Wizzuptodays log has my and fmgs discussion on it00:07
Wizzuphe's using c2c and I think it's never being toggled off00:07
uvosthat would break dapm for that chain yes00:08
uvosall the devices in that chain would then never be powerd down00:08
WizzupI mean, c2c clearly has to work with dapm somehow00:08
Wizzupwe just need to figure out how00:08
uvossure00:09
uvosi mean i dont have time right now00:09
uvosbut its easy to check00:09
uvosomapconf shows you what dai are active (on the omap side ofc) and debugfs shows you the snd-soc widgets and the dapm state for those00:10
uvosomap dai blocks ret however so i dont think its that since fmg had some ret00:11
WizzupI think we checked dapm and saw it's active even outside of a call00:18
Wizzupand it's clearly related because pm was fine before this change on 6.600:18
WizzupI think we just need to figure out how to tell alsa/kernel to not have it on - through some widget or otherwise00:18
Wizzupor maybe our routing is off00:18
uvosi maen as a hack you could break the dapm chain by insterting a fake switch00:20
uvosbut i dont think this is the right way00:20
Wizzupso I think having a switch to enable all the call bits would be a great way00:21
Wizzupwe could easily switch this with UCM too00:21
Wizzup"Call mode" "On" / "Off"00:21
uvosits not the right way to do it from alsas perpsective00:21
Wizzupso what is the right way to do it for streams that bypass cpu/alsa entirely?00:22
Wizzuphttps://kernel.org/doc/html/latest/sound/soc/codec-to-codec.html00:22
uvosdapm turns on the nodes on the route when it thinks the source is active and the route is unbroken00:22
uvosthe source being a dai00:22
uvosi gues the kernel thinks the dai of the modem is active, we need to figure out how to make it know its not00:23
Wizzupswitch seems like a great way imo, since we can toggle it from userspace with ucm without having to parse AT in kernel00:24
Wizzupwell, -some- AT at least00:24
uvosdoing that is def wrong form a framework perspective00:24
uvosnomater how nice it looks from userspace00:25
uvosbut anyhow00:25
WizzupI suppose this is relevant:00:26
WizzupMake sure to name your corresponding cpu and codec playback and capture dai names ending with “Playback” and “Capture” respectively as dapm core will link and power those dais based on the name.00:26
uvosalso i dont think you can shutdown cpcaps dai this way, at least thats before you could insert a switch in the chain is after the cpcap dai00:26
uvoslooking at alsa/snd-soc stuff allways makes me think: this is really compilcated and this lacks features i need :P00:29
uvosalso: i dont get how this bit works00:30
uvosskill issue i gues00:30
Wizzupit's not just you :)00:30
Wizzupfreemangordon: actually I don't see a pm regression10:23
Wizzupmy phone still said 2 days before I went to sleep10:23
Wizzupnow it's at 83%10:23
Wizzupso maybe you just solved it :)10:29
Wizzuparno11: ok I see the problem @ version / headrs11:50
uvosWizzup: is ci fixed?11:52
Wizzupuvos: give it a try, I am still trying to figure out arm64 but I restarted jenkins and it might just work11:53
uvoswonlt the build fail reglardless if one architecture dosent work11:54
uvosie the repo part wont be triggered?11:54
uvosanyhow i kicked sphone11:54
Wizzupno, others will pass11:54
uvos(pending—Waiting for next available executor)11:55
uvosseams stuck11:55
Wizzuplet me check11:58
Wizzupnot sure what it is waiting for :D11:59
uvosyeah all the workers are idle11:59
Wizzupthey all seem reachable too12:01
uvosmysterious :(12:01
WizzupI will just kick them all12:01
uvosseams to not have done sphone any good12:05
Wizzupmhm12:06
uvosbtw12:06
uvosi hope to implement the framework level changes to sphone to decouple the display name tomorrow12:07
dsc_PSA jib now has built-in adblock with rules from ublock origin12:08
uvosneat12:09
uvoswould you mind if i implement a method to pin websites to the recents thingy12:09
dsc_sure12:11
Wizzuprebooting the jenkins vm just in case12:11
WizzupI saw some SEVERE  hudson.triggers.SafeTimerTask#run: Timer task hudson.model.Queue$MaintainTask failed12:12
dsc_uvos: https://github.com/maemo-leste-extras/jib/blob/master/src/popularitymodel.cpp#L28 <== "model" holds these items, the source is SQL but you could inject something here at your discretion12:12
dsc_but maybe you want a 2nd row for such pins12:13
Wizzupjenkins is ging now12:13
Wizzuptry jib in 10-150 mins dsc12:13
dsc_actually we're talking about bookmarks right now, its already half implemented. There is bookmarkmodel.cpp12:14
dsc_but not used12:14
Wizzupafter that I will build another kernel12:14
dsc_Wizzup: thanks12:15
Wizzupuvos: so I think today I would like to work on the rtcom logging stuff we discussed, I'll dig through the logs to find the most recent approach we agreed upon, unless you have some other ideas now?12:15
uvosno but that exactly needs the framework changes i would like to do myself12:16
uvosso please work on something today else ill do this tomorrow12:16
uvos*else today12:16
uvosdsc_: if you are implementing bookmarks soon i wolnt bother, i was thinking of pining things as a bookmarks-lite12:17
dsc_i need half a day to implement bookmarks12:18
dsc_maybe ill do it tomorrow12:18
Wizzupuvos: ok, ty12:18
Wizzupuvos: no need to do it assuming you're busy, but I think wifi hotspot started to work sometime in the stable 6.1.y patches12:25
Wizzupso I would wager that it doesn't work in stable, but it works on devel kernel12:25
Wizzupmaybe we can try to identify that fix commit and pull it in - since 6.6.y doesn't work again12:26
uvoshmm12:32
uvosi kinda doubt there is a commit that fixes this persay since i would expect them to apply this to 6.6 stable too12:33
uvosbut bisecting might give a clue why it sometimes works and sometimes dosent12:33
uvosi remember it not working 5.9 or something and then suddenly working in 5.10 again12:33
uvosso this has been flip floping for a while12:34
Wizzupit did not work in 6.1 earlier but does now in -devel12:40
uvosi get that12:42
uvosbut i expect fixes to be backported to 6.1 to also be backported to 6.612:43
uvosso if a obvious fix patch exists, i expect it would be applied, so i suspect that its not a fix persay but that something is racy, which would explain why it breaks often in random kernel version ranges12:43
Wizzupmaybe, I also blame hostapd version changes12:50
Wizzuplike maybe it tries to enable wpa3 or something silly12:50
Wizzuparno11: building new kernel with likely header fix12:56
arno11cool12:57
arno11i still wonder where slowdown is comming from: conversations or other qt stuff take 25 sec to launch sometimes ! instead of 1 or 2 sec on 6.113:00
arno11how kernel could affect mostly/only qt5 ?13:01
uvosWizzup: could be13:01
uvosarno11: since the n900 is esstanly oom when launching qt stuf the kernels choices can have a big impact13:02
arno11it really doesn't seem oom honestly13:03
uvosi mean its fairly deep into swap no?13:03
uvosit was last time i tried13:03
uvoswhich pages the kenrel decides to ejct makes a big difference here, ofc im just thinking out loud i dont really know why it would change that mutch13:04
arno11i didn't noticed any diff with swap in 6.6 atm. it usually only uses around 100MB of swap13:07
arno11and issue is really random13:07
arno11sometimes it works fine13:07
uvosi maen 100mb of swap is alot on a 256mb machine13:08
uvosbut yeah i have no idea about the root cause13:08
Wizzupfreemangordon: btw on fremantle I can get xmpp calls ti work with my xmpp server to a certain degree: my n900 can hear my laptop sound, but not vice versa (probably stun rtelated)13:10
arno11uvos: yes indeed that's a lot but perfectly fine with 6.113:11
arno11Wizzup: yes probably @stun or maybe the mic ?13:14
arno11ah cool build seems ok, let's try to upgrade again13:19
siceloarno11: what qt programs give you problems?13:23
siceloi generally stay away from qt applications anyway (even on desktop), so i haven't really seen the excessive slow downs you mention13:25
sicelo(or I'm just too patient 😃)13:25
Wizzupa lot of maemo leste is qt so..13:26
siceloyeah, I'm aware (which is a pity ... Fremantle was mostly gtk)13:27
WizzupI don't think it's a pity, gtk3 and gtk4 have gotten significantly more memory heavy and sluggish afaict13:28
arno11sicelo: all qt5 apps but only on 6.6, quite fast on 6.113:28
Wizzupand we can't stay on gtk2 forever13:28
Wizzuparno11: try new kernel wrt dkms13:29
arno11i just finished upgrade, iphb-dkms is ok :)13:29
arno11need to reboot now13:29
sicelonot really (@gtk3/4 getting worse). at least going by the fact that plasma mobile is still heavier than phosh (gtk3) ... which is funny since plamo is fully accelerated while large parts of phosh/gtk aren't13:30
siceloarno11: will retest soon on mine13:30
uvosthats qtquick13:30
uvosi find qt widgets alot better than gtk4 and generally lighter too13:30
uvosno suprise given how gtk4 works13:31
uvosbut yes quick is extreamly slow13:31
arno11sicelo: ok13:31
sicelobtw, yesterday evening i was playing with alarms on the d4 ...  the face-down-to-silence gesture did not work. known issue?13:34
uvosyes i think so13:34
uvosi dimly remember something missing in mce13:35
sicelommm, maybe I'll look into it over the weekend13:35
uvosif you do please add some way to disable this13:36
uvosas i hate this feature on android13:36
uvos(after haveing being bitten by it with my gfs phone being in her handbag a couple of times)13:37
siceloi suppose accelerometer and proximity are hooked and working via mce, so sounds strange why it shouldn't be working already (i forget if it's based on prox or accel)13:37
uvosiirc its both13:37
uvosand mce sends a special signal for it13:38
uvosbut i may be wrong this is all a long time ago13:38
siceloI'll check.13:38
uvoshttps://github.com/maemo-leste/mce/blob/23f40478c86cb14c39e5bcf972cef574054f0598/src/modules/iio-accelerometer.c#L5513:43
sicelonow that we dropped actdead, alarm from powered off state takes much longer to notify user13:43
uvossicelo: MCE_ORIENTATION_FACE_DOWN is not supported13:44
uvosprobubly because it dident translate well to a iio-s-p state13:44
siceloah13:44
uvosso you would have to do something more fancy13:44
dsc_conversations is 50mb resident mem on startup and idle in the background13:45
dsc_qtquick is only used for chat windows13:45
dsc_and its +10mb per window13:46
dsc_</Qt propaganda>13:46
sicelo:-)13:47
siceloyeah iio-s-p is limited (or limiting)13:48
sicelothe new iio-s-p maintainer is also a maintainer pmos-side. maybe there could be interest in adding more accel states13:49
siceloactually it's only just face up and face down that's missing  ... sounds like something i can add too13:57
Wizzupsicelo: yes please add, I really miss the putting the phone face down\14:02
uvosare two stucts with the same members except one as some more members at the end guarenteed to have the same padding?14:09
Wizzupmaybe, but I wouldn't rely on it14:10
Wizzupok, calls work on the bionic too14:13
Wizzupwell, maybe this is when I switch back to using leste as main phone :)14:13
sicelosilly, but interesting nonetheless ... https://gitlab.freedesktop.org/hadess/iio-sensor-proxy/-/blob/master/src/orientation.c?ref_type=heads#L517:07
siceloin fact, https://gitlab.freedesktop.org/hadess/iio-sensor-proxy/-/blob/master/src/orientation.c?ref_type=heads#L5-917:07
inkyI solved hildon rotation in pascal. Wanted to do with x11 functions but it didnt work out for some reason, ut worked via gtk/gdk calls.18:08
arno11Wizzup: ok so i tried again 6.6 and...swap seems totaly broken18:10
arno11i mean vm.swappiness=0 solves the issue18:11
arno11ofc multitasking is difficult but extrem slowness disappeared entirely18:12
arno11i really don't know what's the diff with 6.1 excepting known swap bugs18:12
siceloinky: what hildon rotation you mean?18:13
arno11and really no idea to solve the issue18:14
arno11*how to solve18:14
arno11will investigate more. that's easier with a well responsive device now :P18:15
uvosio might be slow for some reason18:15
uvosi think we had that before18:15
arno11ah ok18:16
uvosmaybe try benchmarking the sdcard between 6.1 and 6.618:16
arno11ah good idea, i have already results for 6.118:17
arno11i'll do that18:17
uvosor i gues emmc rather18:17
uvossince swap is on emmc18:17
uvosunless you changed it?18:17
arno11nope i use emmc18:17
arno11and works fine with 6.118:18
arno11but i can try swap on sd + 6.618:18
arno11to check the diff18:18
uvosrather more interesting is to see if io benchmarks are slower on 6.618:19
uvossdcard is on the same interface as emmc anyhow18:19
uvoshard to belive one would be affected but not the other18:19
arno11yes indeed18:19
uvosswaping on sdcard should be faster in general, as modern sd cards have mutch faster random reads/writes than n900s emmc18:19
uvosbut thats a seperate issue18:20
uvoswell good modern sdcards anyhow18:20
arno11yes18:20
arno11btw i hit sd bus limits with my sd and 6.118:21
arno11let me check 6.618:21
uvosfor sequential access any sdcard will hit the bus limits18:23
uvossmall randoms are more interesting18:23
uvosbut my sdcard still hits the bus limit until below 1k blocks18:23
arno11ah ok18:24
uvossmall randoms is also closer to what swap needs18:24
uvosswap tends to need a page here, a page there etc18:25
arno11results are similar with what i see on 6.1, at least using dd18:31
arno11*for the sd card18:32
arno11uvos: with small blocks it seems very slow. what command do you use to bench btw ?18:36
arno11using hdparm, read seems slow on emmc18:50
arno11only 5MB/sec18:50
Wizzupis that different on 6.1?18:55
gnarfacewith emmc it's usually safe to use a larger block size18:59
gnarfacethat should speed it up, though i'd try to still make it a multiple of the physical block size18:59
arno11ok18:59
gnarfacetry something between 1MB and 8MB just make sure it's an even multiple of the physical block size18:59
arno11ok19:00
siceloiio-s-p folk say they'd love to have the face-down/up support19:03
arno11i'll LYK later, need time to compare with 6.119:04
arno11bbl19:05
Wizzupsicelo: great!21:29
sicelogetting unexpected behavior with iio-s-p ... https://paste.debian.net/1323081/21:58
sicelocode used is https://github.com/maemo-leste-upstream-forks/iio-sensor-proxy/21:58
siceloi'm not expecting the values to be different at this point :-/22:07
arno11Wizzup: uvos: to make it short, the swap issue is due to emmc itself (with 6.6, not with 6.1)22:09
arno11i got a well working stuff using sd swap file22:09
arno11now it is stable22:09
arno11and almost as fast as 6.122:10
arno11but it took several minutes to swapoff emmc22:10
arno11maybe something is wrong on boot ?22:10
sicelo sounds all good to me :-)22:12
siceloit's correct for swapoff to take some time22:13
gnarfacewhy not use zram for swap? just curious22:13
arno11according to irc logs it was problematic on N90022:14
gnarfaceoh, didn't know that22:14
arno11but i don't know exactly why22:14
arno11fmg should know22:14
sicelotime to build latest/upstream iio-s-p i guess22:14
gnarfacehow much ram does it have? maybe it's not enough22:14
sicelo256MB22:15
gnarfaceoh. yea, maybe that's why. also, now that i realize you're using a N900, i revise the parameters for optimal dd block size use; 8MB may be too much. some multiple of the physical emmc block size for sure, but maybe start with 2x or 4x and work your way up doing some benchmarks22:16
arno11with 1MB i got similar results with both 6.1 and 6.622:22
arno11btw, not related but calls works fine with no more hangup issue, boost mode is ok, conversations ok, i still need to kill hsm as usual but apparently no other issues22:26
arno11so excepting swap, everything seems similar to 6.1 now so i'm happy :)22:27

Generated by irclog2html.py 2.17.0 by Marius Gedminas - find it at https://mg.pov.lt/irclog2html/!