libera/#maemo-leste/ Sunday, 2025-04-27

freemangordonhmm, no matter how I tune ext4, I cannot get more than 4MB read speed (using fio) from this sandisk extreme10:08
gnarfaceyou should try xfs :)10:09
freemangordonwell, if card is bad/slow10:09
freemangordonno fs will help10:10
freemangordonit seems to have blocksize of 4k and erase size of either 512KB or 1MB10:11
gnarfacephysical block size of 4k, not just logical?10:21
freemangordonyes, according to flashbench10:21
freemangordonsee https://paste.debian.net/1371855/10:22
freemangordonlooks like blocksize of 4k and eraseblock of 512k10:22
gnarfacei uh... don't know how to read this output10:23
gnarfacei usually just check with hdparm -I and believe what it says10:23
freemangordonanyway10:23
freemangordonhmm, moving swap to a partition on sd card makes lots of difference11:30
Wizzupdidn't we determine earlier that swap on mainline was broken? or was that presumably fixed11:41
freemangordonwe did, however, I was using swapfile on sdcard12:34
freemangordonnow I made a dedicated partition12:34
arno11freemangordon: it makes a huge diff on my device as well15:14
arno11(i also use a sandisk extrem)15:15
arno11qt5 launch time is quite similar to chimaera now15:15
freemangordonhow?15:18
arno118-10 sec cold, 2-6 sec warm15:19
freemangordonthis is the same on chimaera?15:20
arno11quite the same15:20
freemangordonok15:20
arno11but with tg-desktop, that's crazy: cold was 7 min on daedalus, now 45 sec...(like chimaera)15:22
freemangordonwhat has changed?15:23
arno11what do you mean ?15:24
arno11i just use swap on partition instead of swap file15:24
arno11*dedicated partition15:25
arno11but it is still a way slower to launch qt apps than using qt5ct tweak, specially if i launch several apps15:30
Wizzupfreemangordon: is it different on 6.12 I wonder15:37
Wizzupwe're still on 6.6 right?15:37
arno11still on 6.615:39
freemangordonWizzup: no idea, but we shall plan kernel upgrade anyways16:04
uvossphone should more gracefully handle pa failing18:40
arno11uvos: it helps, yeah, ty19:31
arno11now when the call fails, sphone is immediately ready to use. but sound in next incoming or outgoing call still doesn't work (kind of 'expected')19:34
arno11then next calls are ok19:35
arno11freemangordon: really wonder what has changed @swap or kernel or whatever. sounds not normal that swap on sd partition works better than swapfile, no ? should be similar (?)19:39
uvos"now when the call fails, sphone is immediately ready to use. but sound in next incoming or outgoing call still doesn't work (kind of 'expected')"19:58
uvosnot totally expected from sphones side, sphone starting should sync its state up with pa again on the first call not the second19:59
uvosbut sphone restarts fairly fast it could be that pa has not recovered from whatever the problem was by the time sphone has started again19:59
uvoslogs would help19:59
Wizzuppa recovering means pa restarting I think20:04
Wizzupthere is no other reason you'd lose your connection to it20:04
Wizzupmaybe we just pull in latest PA and see if it's more stable20:04
WizzupI think PA is probably a dead end in the next few years as everything is moving to pipewire, and pipiwire as PA emulation (incl. modules)20:05
arno11yeah seems an easy stuff to use pipewire PA module20:05
Wizzupmaybe, it might also pose challenging20:10
arno11dealing with audio in linux is always a challenge :P20:10
arno11Wizzup: btw i wonder why it fails to switch ucm profile quickly from ringtone to call, and not if i play a song from audacious and receive a call (in this case, the switch hifi/voicecall works fine)20:16
arno11maybe something is wrong with gstreamer/ringtone ?20:17
freemangordonhmm, why dbus created processes are reniced to dbus process priority?20:19
freemangordonoh, it is inherited on fork :(20:20
uvosarno11: in the ringtone case sphone switches profile and then stops the ringtone at more or less the same time20:29
uvosarno11: these 2 things happening at once is probubly what is causing pa to trip20:29
uvosi think mapphones should more or less just work with pipewire, n900 is a different matter.20:31
arno11ok i get it, ty20:31
freemangordonuvos: is it possible to first stop the ringtoe and then to switch the profile?20:31
uvosthe call to pa is handed off to a different thread20:31
uvosso not really20:31
uvosi mean you could add syncroniztion but pa should not crash just because gst and sphone are both using the interfaces at the same time20:32
freemangordonyeah20:33
Wizzupwe can try to pull in latest PA, and if it doesn't work, pull in pipewire with PA module20:36
Wizzupalthough the latter could be more work20:36
arno11hmm fo n900 case, it is supposed to not crash ofc but supposed to remix everything (once switched to ucm voicecall profile) if the ringtone fails to stop @the right time. at least it is what's happening with any other sound20:38
arno11uvos: ^20:39
arno11(remixing is activated by default in n900 daemon.conf FYI)20:41
uvosarno11: i mean you can try moveing this https://git.maemo.org/leste/sphone/src/commit/a5ef8d4d35a37ca907e6b02a8c3afb725933e312/src/modules/manager.c#L8321:17
uvosup to the beginning of that branch, but both since audio_stop_pipe ends up sending a message to gsts thread and the call_mode_pipe ends up sending a message to the libpulse thread this only encourages the gst thread to win this race21:18
arno11_uvos: ok, ty21:24

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