libera/#maemo-leste/ Saturday, 2025-03-22

Wizzupsicelo: on upower, shall I just build the one from trixie (if that is new enough) for daedalus?02:34
siceloTrixie version is fine. although if there are no other concerns, I  think git master is preferable05:57
freemangordonWizzup: we use cgroups v2, right?07:49
freemangordonarno11: instead of doing sudo, perhaps you can check what cgrulesengd can do for us (in cgroups v2 you can assign nice value to a group)07:53
freemangordonso we can distribute /etc/cgrules.conf with rules for Xorg, mce, h-d, etc07:54
freemangordonand we can use ohm rules from fremantle07:54
freemangordons/use/port07:55
freemangordonhmm, at least VM is using cgroup v107:58
freemangordonarno11: see /usr/share/policy/etc/current/syspart.conf08:05
freemangordonI am really tempted to consider moving to systemd, we get all those things there with unit files08:15
freemangordonanyway...08:15
freemangordonarno11: seems we can use inotify to watch new processes creation08:16
freemangordonsee https://paste.debian.net/1364525/08:25
freemangordon'processed': https://paste.debian.net/1364526/08:28
freemangordoneverything else is done through cgroups08:28
freemangordonfor Xorg, we should be able to set nice value from the startup script08:34
freemangordonhmm, we do it, but autologin starts new pam session, which starts Xorg with nice 008:50
freemangordonso, we shall create some daemon/script/whatever that monitors processes creation and sets the appropriate nice value08:51
Wizzupno, let's not make a new daemon11:15
Wizzuphttps://github.com/OpenRC/openrc/blob/master/user-guide.md11:19
Wizzuphttps://wiki.gentoo.org/wiki/OpenRC/CGroups#Activating_cgroup_feature_support11:20
Wizzuplet's do it this way11:20
Wizzupfor xsession scripts, we can use cgrun if we must hack it in there11:20
Wizzupand ifwe do have cgrulesengd, why do we need another daemon?11:21
mkfhello.13:08
mkfi've got otg working today. :)13:09
sicelonice progress! what image is your work based on btw?13:09
mkfsunxi image in the website, but it's quite change right now.13:10
mkf*changed13:10
mkfalso updated to chimaera13:10
mkfhave wireless worked for you? i'm not sure if i'm doing something wrong or i have fired the chip... :(13:10
mkffreemangordon, Wizzup: ^13:11
freemangordonmkf: wifi works on my sunxi, but I had to patch the driver13:59
freemangordonmy tablet has some exotic realtek chip13:59
freemangordonWizzup: cgrulesengd can't control nice value14:00
freemangordonor rather - we use cgroups v1, which can't control niceness14:00
freemangordonif we can enable cgroup v2, then we don;t need another daemon14:03
freemangordonhmm...14:03
freemangordonoh, wait, seems v2 is mounted under /sys/fs/cgroup/unified14:09
freemangordonso yeah, we can go with v314:09
freemangordon*v214:09
Wizzupfreemangordon: then we rename in the xsession14:10
Wizzuprenice*14:10
freemangordonno, we can't14:10
freemangordonxsession is run as user14:11
Wizzupwhy not? sudo renice $(pidof Xorg)14:11
freemangordonno, wait14:11
Wizzuphttps://unix.stackexchange.com/questions/102945/how-can-i-allow-a-user-to-prioritize-a-process-to-negative-niceness14:11
Wizzuplooks like pam can let users set negative niceness14:11
freemangordonbut we want to ue cgroups for that14:11
freemangordon*use14:11
Wizzupok, if cgroups v2 can do it14:11
freemangordonyes, they can14:11
freemangordonthey have the so called 'sched' controller14:12
freemangordonthat seems to support niceness14:12
freemangordonat least AI said so :)14:13
mkffreemangordon: mine has that too. :)14:13
mkfwhat patches you made?14:14
Wizzupfreemangordon: I am not sure if I will believe AI, but ok14:14
freemangordon"echo 10 > /sys/fs/cgroup/my_cgroup/sched.niceness"14:15
freemangordonmkf: what exactly?14:15
Wizzupfreemangordon: ok, well, let me fix this n900 fstab issue first, but it sounds like cgroups is the answer here14:16
freemangordonmhm14:16
Wizzupfreemangordon: btw, any chance to look at git.maemo.org? if we're happy with it I might do the migration some time soon14:16
Wizzupor at least proceed further with the migration14:16
WizzupI'll check out some sign in via github kind of things, too14:16
freemangordonsorry, but I lost the context why do we need to migrate out of github14:17
freemangordoniow, what is wrong with github?14:17
Wizzupit's closed source, it's microsoft, it's US based, the CI/CD is not our own, and I don't want to say build packages on github infra14:17
Wizzupand I don't trust them to not throw us off when anyone complains14:17
freemangordonok :)14:17
freemangordonoh, so we can do builds on commit with our infra?14:18
Wizzupthe reason we went for it initially was that we thought the network effect would attract people14:18
Wizzupre: builds on commit - I plan to make that work, yes14:18
freemangordonok, cool14:18
freemangordonok, will have a look soon14:19
mkffreemangordon: rtl8723bs or 8702as or something like that...14:19
freemangordonno, it is important the *exact* chip14:19
freemangordonbecause if it is BS, it is the same as mine and I would be able to help14:20
mkfhow do i know what's the chip exactly?14:20
freemangordoncheck in fex14:20
mkfok14:21
freemangordonmkf: all the changes I have made https://paste.debian.net/1364571/14:26
freemangordonat the end is drivers/staging/rtl8723bs/os_dep/ioctl_cfg80211.c14:26
freemangordonBTW, your issues with TS might be related with wrong panel config14:26
freemangordonsee unk_qc760bg1_1024_600_72_mode14:27
freemangordonbbl14:31
mkfmine is 800 x 480 tho14:34
freemangordonhow do you know? from fex?14:35
mkffex.14:35
freemangordonok14:36
mkfhowever the image isn't the original one shiped with device, and wifi doesn't work in android, so i'm not sure if fex is to be trusted about wifi14:37
mkfsunxi wiki says i should have a realtek, and i do see a realtek chip in my board14:37
freemangordonok14:37
mkfbut fex (of incompatible?) android image says i have an rda. which i don't think is true.14:37
freemangordondid you enable the driver?14:37
freemangordonit is in staging14:38
mkfi have, but didn't applied the patches yet14:38
mkflet me recompile14:38
freemangordonwait14:38
freemangordondid it probe properly?14:38
freemangordon*does it*14:38
mkfhow do i know?14:38
freemangordonlsmod?14:38
mkfit loads it fine (added it in /etc/modules)14:39
freemangordonyou should not load it by hand14:39
mkfoh. ok.14:39
freemangordonlemme boot my device to be sure what happens14:39
mkfif i don't load it, who will load it then?14:39
freemangordonumm... who loads the other modules?14:40
freemangordonudev I would say14:40
mkfidk, i load all modules by hand. :)14:40
freemangordonoh, how's that?14:40
mkfgotcha.14:40
mkfidk, silead didnt worked properly and i thought modules must be manually loaded.14:41
freemangordonnope14:41
freemangordoneverything here is auto-probed14:41
freemangordonkeep in mind that even with that patch wifi has issues connecting every now and then14:41
mkfoh well.14:42
freemangordon[   13.172909] r8723bs: module is from the staging directory, the quality is unknown, you have been warned.14:43
mkfit's not laoded automatically.14:43
mkf*laoded14:43
mkfsilead did, so i think it's not probed.14:43
mkfor my dts is faulty. hm.14:43
freemangordonyeah, could be14:44
freemangordonwhat dts do you use? sun8i-a33-q8-tablet.dts ?14:44
mkfa23-q8-tablet14:46
freemangordonoh, yours is not a3314:46
mkfhttp://cloud9p.org/paste/cf0814:47
mkfyeah but these two are close enough.14:47
freemangordonrtl8723bs: sdio_wifi@1 {14:47
freemangordonso yeah, same chip14:47
mkf(i have modified this file)14:47
mkfhttps://elixir.bootlin.com/u-boot/v2025.04-rc4/source/arch/arm/dts/sun8i-a23-q8-tablet.dts14:49
mkforiginal file14:49
freemangordonthere is nothing about wifi there14:50
mkfyeah. :(14:51
freemangordonhmm, wait14:51
freemangordon&mmc114:51
freemangordonit is missing "status = "okay"; ,no?14:52
mkfsun8i-q8-common.dtsi does define &mmc114:52
mkfand it has status ok14:52
freemangordonok14:52
mkftho it doesn't mention what wifi model it has14:52
freemangordon(I have kernel trees here, no need to paste upnaptched files)14:52
freemangordon*unpatched14:52
mkfsorry, i'm not very fluent in device tree tounge. :)14:53
mkfokay. thanks.14:53
* freemangordon checks14:53
mkfopened the tablet, apperently it has a 87?3As14:53
freemangordonso not BS?14:54
mkfno.14:54
freemangordonwell,...14:54
mkfbut these two share the same driver14:54
freemangordonthat explains it14:54
freemangordondo they?14:54
freemangordonok, but are you sure it is on emmc bus?14:54
mkfhttps://linux-sunxi.org/Wifi#SDIO-based14:54
mkfat least here says they share the same driver14:55
freemangordonyeah14:55
mkfi'm not sure if they are on emmc bus tho.14:55
freemangordon"SDIO"14:56
mkfah.14:56
mkfwait emmc is too sdio? huh.14:56
mkfi thought they are distant standards14:56
mkfokay that explains a bit why a wifi should be on emmc14:57
freemangordonmkf: do you have anything "sdio" in dmesg?15:00
freemangordonalso, check /dev/bus/sdio15:01
gnarfacei have the same SDIO unit on a 2GB pine64+ board, and the debian kernel didn't have the right dtb patch, i had to manually add it15:01
gnarfacedevice was showing up unpowered15:02
gnarfacesame driver too, lemme know if you can get more than 20 megabits out of it in host mode15:02
gnarfaceit'll work but i can't make it as fast as it's supposed to be (150MB/s!?)15:03
freemangordongnarface: see https://paste.debian.net/1364571/, I think there is a bug in the driver15:04
freemangordon(patch at the end)15:04
freemangordonwithout that it does not even connect to my router here :)15:05
freemangordonor does it once in 20 reboots ;)15:05
gnarfacehmm15:06
gnarfaceinteresting...15:06
mkfgnarface, thank you. i think i never seen a n card go beyond 10Mb (18 Mb is theorical limit of 150MB cards)15:06
gnarfacefreemangordon: you mean just this hunk for drivers/staging/rtl8723bs/os_dep/ioctl_cfg80211.c? you think that'll fix the speed issue?15:08
gnarfaceor do i need more of this?15:08
freemangordonno idea15:08
gnarfacehmmm15:08
freemangordonbut at least it allows me to connect to my router 9 of 10 times15:08
freemangordongnarface: btw, what do you use to test speed?15:09
gnarfaceseems promising...15:09
gnarfaceusually testing with netcat15:09
gnarfacejust time and netcat of a cached 50mb file from a ram disk to /dev/null15:09
gnarface50MB i mean15:10
freemangordonyeah15:10
mkfgnarface: megabit or megabyte?15:10
gnarfaceheard some accounts of 60 megabits with wpasupplicant but they hadn't tested in host mode, so there also might be a speed discrepancy between host mode and client mode15:11
gnarfacemkf: MB == megabyte, mb == megabit15:11
freemangordonlemme see if I'll be able to run speedtest.net :)15:11
mkfaren't cards rated for mb?15:11
gnarfaceyes, but the SDIO bus is rated in MB, i thought...15:12
mkfhm.15:13
gnarfacebasically it's supposed to be good for 150MB/s, current kernel/dtb settings show it should at least be working at 40MB/s, but the best first-hand accounts i've heard can't get more than 6MB/s out of it and in host mode i can't get more than 2MB/s out of it15:13
freemangordonok, according to speedtest.net, my DL speed is 39.54, my UL is 46.7815:13
gnarfaceMB or mb?15:13
freemangordonMbps15:14
gnarfacehmm, still basically double what i'm getting15:14
gnarfaceso maybe that patch is helping15:14
freemangordonthose speeds are pretty normal, as my inet is 100Mb or something15:14
freemangordonand also, chromium uses some CPU, so that affects speets too I would say15:15
freemangordon*speeds15:15
freemangordonlemme re-test15:16
freemangordon20.57/53.0915:17
freemangordonbut web page trying to show speed meter in real-time does not help much15:19
arno11freemangordon: Wizzup: great @cgroup15:19
mkffreemangordon: are you running chromium on a33?15:20
freemangordonyes15:20
mkfdoes that work well?15:20
mkf(is it jib?)15:21
freemangordonno, *chromium*15:21
mkfoh.15:21
freemangordonworks ok15:21
mkfi see.15:21
freemangordonwith ublock origin, it is 'ok'15:21
mkfwait a sec15:22
freemangordongnarface: so, with speedtest page scrolled so meter is hidden, I get 45.11/63.9615:22
freemangordonmkf: hmm?15:23
gnarfacefreemangordon: that's in client mode though, it'd be interesting to see if it can maintain those speeds in host mode15:23
freemangordonno idea, I admit I have never used host mode :)15:24
gnarfacethat is more like what i've heard other people say they were getting out of it in client mode though15:24
mkfkernel nags about lack of regulatory db for 802.1115:24
mkfcould that might be a reason?15:24
freemangordonyeah, we knbow15:24
freemangordonno15:24
gnarfacedoesn't installing the crda package get rid of that error?15:25
mkfhm ok15:25
freemangordongnarface: no, because we build our kernels15:25
gnarfaceoh15:25
freemangordonand there is something with the keys15:25
mkfi built my own, does it effects that?15:25
freemangordonkernel?15:25
mkfyeah.15:25
freemangordonsure15:25
mkfi have another wifi card, that's usb15:26
mkfand apperently it gets detected enough so ip a can see it15:26
mkftho idk how to connect to wifi in maemo thru terminal (gui doesn't show a thing)15:26
mkfdo we use iwl in maemo?15:26
freemangordonuse wpa_supplicant15:27
freemangordonor rename to wlan015:27
freemangordonthen maemo will see it15:27
freemangordonWizzup: ok, won't work with cpu.weight.nice15:39
freemangordonthis does not change niceness, but some scheduling priority within the group15:40
Wizzupwe can leverage pam to allow user to set negative nice with renice15:40
Wizzupor maybe scheduling priority is enough15:40
freemangordonotoh, do we want to enable non-root to set niceness < 0?15:40
freemangordonI don't like the idea15:41
freemangordonI'd better invent some more secure way15:41
freemangordonor really, just create a 20-liner script (part of leste-config) that does inotifywait on /proc15:42
Wizzupit all feels like bad hacks imo15:42
Wizzupstrange that we can't just put nice -20 in front of something.15:43
freemangordonif you allow user to do that, a user-started runaway process will hang the system15:43
freemangordonso I don't think that's a good idea15:43
freemangordonbbl15:44
arno11Wizzup: freemangordon: maybe an alternative: i created a basic bash script (root) with the needed renice commands, linked to a xsession.post file (99my_script) and surprisingly works fine16:55
WizzupI'm not surprise it works16:55
Wizzupwe're just trying to figure out a way not to leverage passwordless sudo on user to work around this16:56
Wizzupas in, make it a bit more future proof, so to say16:56
arno11yeah ofc16:57
mkfyipeee \o/17:50
mkfwifi now probes17:51
freemangordonmkf: how?18:06
mkfi'm a dummy, it turns out i've compiled the incorrect driver.18:07
freemangordonheh18:07
mkfit loads this one now, and it's detected as wlan018:08
mkfalas maemo still shows no wifi networks18:08
Wizzupmaybe you need to restart icd218:33
mkfuh. how can i stop hildon from starting?18:58
mkfor xorg even.19:04
freemangordonmove dsme out of /etc/init.d/19:20
mkfokay.19:20
mkfnow got wifi in maemo working, altough i can't connect. trying to apply your patch..., which version of kernel this patch is for?19:21
freemangordon6.12?19:21
mkfoh.19:23
mkfi suppose that driver got quite some love in versions between 5.8 and 6.12.19:24
mkfhm, linux 5.8 already have done what you did roughly.19:27
mkfnvm.19:31
freemangordonmkf: better use at least 6.1220:14
freemangordonit works fine, at least here on a3320:15
arno11freemangordon: btw any idea about what's going on with Qt5 apps launch time ?20:36
arno11gpu related or ?20:37
arno11it is even a lot slower if i try to open more than one qt app.20:46
arno11if i use rasterisation (force raster surface in qt5ct), it is almost as fast as what you described for d4. and i can open 4 or 5 qt apps @the same time with no troubles20:48
arno11(i.e calculator, qalendar, hamsterfiler, conversations, tg-desktop)20:49
arno11impossible OOTB with daedalus20:49
mkffreemangordon: will do after i got mali up :)21:33
mkfhm. strange, mali is apperently loaded but there is no /dev/dri/dri021:43
Wizzupmkf: why mali and not lima22:21
mkfyeah i meant lima22:21
mkfit seems /dev/dri/card0 is there but for some reason X can't open it (no such file or directory)22:22
mkfit even exists before X runs22:22
sicelouser permissions perhaps?22:23
mkfmight be, but chown'd em to be sure and nothing changed22:24
mkfi've also ran X as root for test, and it can access /dev/dri/card0, but glxinfo still tells me i'm using llvmpipe22:24
Wizzupsicelo: re: upower, trixie is easier, but we can also do master23:44
sicelosure, should be OK23:58
siceloFYI, I currently have a build of upower that reports the following, https://paste.debian.net/1364641/  .... << freemangordon ;-)23:58

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