libera/#maemo-leste/ Sunday, 2021-10-24

tmlindinteresting, so what's the minimal test case not working with chromeos mesa that is working with the ti blobs?08:47
tmlindnot sure i've seen this issue with my use case08:48
tmlindfreemangordon: hoping test/revert/fix the extra patches in droid4-pending-pvr-omapdrm-v5.15 branch today09:15
tmlindfreemangordon: i can only test on d4 right now, if you can test the patch i sent at some point on n900 that would be great09:15
tmlindi only tested n900 so sgx loads and initializes09:16
freemangordontmlind: ok, will do09:16
freemangordonthe minimum test is glmark2-es2-drm09:17
freemangordonworks with blobs, crashes SGX with chromeos09:17
tmlindok09:22
tmlindcrashes on d4 or also on n900?09:23
freemangordononly on d409:23
tmlindok09:23
freemangordonon n900 works perfect, giving glmar of 3709:23
freemangordon*glscore09:24
tmlindnice :)09:24
uvosfreemangordon:  did you run over how to use libebook while working on abook? specificly how do you get the default address book? old libebook has e_book_new_default_addressbook10:15
uvosbut this is gone in newer versions10:15
uvosand you need to use e_book_client_connect(ESource *source)10:15
uvosno sure where the source is supposed to come from10:15
uvossomething that points to eds somehow i gues10:16
freemangordonuvos: yes, sec10:25
freemangordonWTF? where did all the online libebook documentation go?10:30
freemangordonwhat's wrong with gnome/freedesktop guys?!?10:30
Wizzupyeah it all disappeared10:31
freemangordonbut...but.. why?10:32
Wizzupwhere was it before?10:33
Wizzuphttps://developer-old.gnome.org/libebook-contacts/stable/10:34
freemangordoneverywhere :)10:34
Wizzupfreemangordon: I think developer-old. helps10:34
freemangordonuvos: https://github.com/maemo-leste/osso-abook/blob/84e785a82737fd2b7238df87c6942bacada786b2/lib/osso-abook-util.c#L16210:35
freemangordonI think this is what you need10:35
Wizzupso next time you find a dead link, just add 'developer-old.' before gnome.org10:35
freemangordonI simply cannot find any link10:36
freemangordonfor example - search google for E_SOURCE_EXTENSION_ADDRESS_BOOK10:36
freemangordonor e_source_registry_new_sync10:37
freemangordonnothing!10:37
Wizzupfreemangordon: what about the link I shared?10:37
freemangordonyes, but google knows nothing about it10:37
WizzupI searhed for 'libebook-contacts documentation'10:38
freemangordonbut, how do you know it is libebook-contacts?10:39
uvosyes this is excatly what i discoverd10:39
freemangordonthe point is - a month or so ago it was enough to search for function name10:39
uvoseven the link to documentation on gnome mainpage just goses to 40410:40
freemangordonnot it is like those libs have never existed10:40
freemangordonuvos: :nod:10:40
freemangordonsomething weird is going on10:40
freemangordon*now it is like10:40
Wizzupwell they always liked to break things every few years and reinvent it10:41
Wizzupprobably just an accident with the docs10:41
Wizzup'they look prettier' 'but all the links are broken' 'oh, we will make it a new domain' 'the new domain is not visible in search engines' 'oh well'10:42
uvosthe new documentation pages are way less usable too10:42
uvoseveything is closed by default10:42
uvosi cant even ctrl-f to find a symbol10:42
uvosfreemangordon: btw e_book_* is depricated10:45
uvosfreemangordon: replaced by e_book_client_*10:46
uvosaccording to headers10:46
freemangordonuvos: yes, it is10:51
freemangordonbut...10:52
Wizzupbetter finish the RE before switching api eh :)11:04
tmlindack i see segfault on d4 with glmark2-es2-drm11:09
freemangordonhmm11:12
freemangordonI don't see segfault but SGX HW recovery triggered11:12
freemangordonWizzup: exactly :)11:13
_uvos_sure just ensureing awarenis wrt deprication11:32
_uvos_sphone has some basic ebook support now :) shows contact names on incomeing calls and sutch11:33
_uvos_anyone here know sql querys well?11:33
_uvos_i could use some help (next week) with sphone-store11:34
siceloHas its own db?11:35
_uvos_yeah for recent/missed calls/sms11:36
tmlindfreemangordon: yeah you're right sgx hw recovery with glmark-es2-drm with chromeos mesa11:44
tmlindfreemangordon: PVR_LINUX_MEM_AREA_USE_VMAP is no longer needed so reverting that one, seems it already got fixed somewhere in upstream, probably mainline linux11:53
freemangordongood11:57
freemangordonany difference?11:58
tmlindno, dropping PVR_LINUX_MEM_AREA_USE_VMAP does not seem to affect any of my test cases11:58
freemangordon:(11:58
Wizzupuvos: I can help with bql12:19
tmlindfreemangordon: ok pushed out updated droid4-pending-pvr-omapdrm-v5.15 with all the mystery patches reverted and the ocp patch applied12:19
Wizzupalthough we have osso rtcom for it, with sqlite format12:20
Wizzupcould be a good db format maybe12:20
tmlindlooks like reverting 33bc438d6d88 ("drm/omap: Fix page fault handling and flush what framebuffe wants flushed") causes almost constant trails on termite on sway at least on the left side of the lcd12:21
tmlindbbl12:21
Wizzupuvos: with sql12:22
tmlindheh termite is almost unusable with 33bc reverted.. some sgx flush or command mode lcd update is needed somewhere for sure12:42
tmlindis n900 updating the screen just fine or also showing old data on the lcd?12:43
freemangordonupdates fine12:48
freemangordonactually it hangs without 33bc reverted :)12:48
tmlindok, on droid4 hdmi seems to behave too so must be a command mode lcd issue12:48
freemangordonmhm12:49
freemangordonI also have 'hangs' in glamor on d412:49
freemangordonwhich do not happen on n90012:49
freemangordonvsync issue that is12:49
freemangordonas we already discussed back then - modesetting waits for vsyn to present next buffer, but it never comes12:50
freemangordonbecause there is no vsync12:50
tmlindi also see corrupt window titles on hdmi on droid4 with wlroots fyi12:51
tmlindseems to be some wlroots issue12:51
tmlindfreemangordon: if you have an idea where d4 needs the command mode update please let me know12:57
freemangordontmlind: do you remember my idea for fake vsync back then?12:58
tmlindhmm no, we carry some patch for omapdrm right now for that though12:58
freemangordonlemme try to dig it12:59
tmlindok12:59
freemangordontmlind: "Re: xpresent/vsync and omapdrm", you're on CC13:00
freemangordonand then I hit 1.17 breakage on n900 and never sent this RFC patch :)13:05
tmlindheh ok13:06
freemangordonbut, I think I described my idea pretty well, should be doable13:06
freemangordonor, we can do the same as WL and create a timer "if there is no vblank event in 20 ms, screw it and start presenting" in MS13:07
freemangordonbut, this is really a nasty hack13:08
tmlindyeah event triggered variable refresh would be ideal13:08
freemangordonbut, who triggers the event?13:09
tmlindright, no idea :)13:09
freemangordonI think this belongs to omapdrm13:09
tmlindseems like that's a whack a mole game for the triggering from various places13:09
freemangordonbecause the driver knows about details - is that a manual update display, is there TF interrupt, etc13:10
tmlindyeah seems like omapdrm should know it13:10
tmlindbbl13:21
freemangordontmlind: what is the easiest way to increase CMA size?14:30
bencohon many (most?) platforms you just either change the .config file, or define cma= at boot14:31
bencoh(assuming you're really using the CMA and not some custom allocator with its own pool)14:32
freemangordonit is CMO14:34
freemangordon*CMA14:34
freemangordoncma: cma_alloc: reserved: alloc failed, req-size: 375 pages, ret: -1214:34
bencohthen what I mentionned should work afaict, unless it changed in recent kernels14:35
bencohkernel should report cma size at boot btw14:36
bencoh([    0.000000] cma: Reserved 400 MiB at 0x00000000a1800000 on some unrelated board)14:36
freemangordon16MiB14:39
freemangordonthis is on n90014:39
freemangordontmlind: just tested droid4-pending-pvr-omapdrm-v5.15 on n900, works like a charm. I have your prm patch too, on top15:19
tmlindfreemangordon: ok great, good to hear & thanks for testing15:21
freemangordonjust a sec to check if pvr is idle when it has to be15:21
freemangordonand also if pvr interrupts increase15:22
tmlindnot sure if we need the ocp interrupts enabled on omap4 and later, but that's easy to disable by dropping the ocp area for the selected compatible in pvr-drv15:22
tmlindmaybe let's consider dropping the omap4 ocp interrupts after we have the refresh issues fixed..15:23
freemangordonsgx_pwrdm (INA),OFF:1,RET:0,INA:10,ON:11,RET-LOGIC-OFF:0,RET-MEMBANK1-OFF:015:25
freemangordonhmm, I was expecting OFF there. or, this is ok?15:26
freemangordontmlind: ^^^15:26
tmlindit does hit off with fremantle kernel, right?15:26
freemangordonno idea15:27
freemangordonbut I guess yes15:27
tmlindfreemangordon: maybe try adding this for the sgx domain in omap_prm.c driver: .flags = OMAP_PRM_RET_WHEN_IDLE15:29
freemangordonok, will do15:30
freemangordontmlind: for some reason performance with glmark-es has increased15:31
sicelo:-)15:31
tmlindactually that flag may not help unless the sgx device is configured for autoidle which we don't know because of the broken ocp.. worth a try though15:32
freemangordontmlind: cuold it be that missing workaround?15:32
tmlindwhat was the missing workaround again?15:33
freemangordonwhich you said is applicable only for boards with reset button :)15:33
tmlindweird that it show ina instead of ret or off though, maybe some bits are different15:33
tmlindoh that one, i think that's only for development boards with reset button15:34
freemangordonalso, in genpd_debug (or somesuch)  it shows off-015:34
tmlindmaybe the prm bits are swapped for sgx domain or something15:34
freemangordonanyway, this is not such an issue now15:34
tmlindyeah15:35
tmlindthere's a comment about sgx not supporting retention in powerdomains3xxx_data.c15:36
tmlindbest to check the related values on fremantle kernel15:36
freemangordonok15:36
tmlindwith grep sgx /sys/kernel/debug/pm_debug/count15:37
freemangordonwill do, when it comes to it15:37
tmlindok15:37
freemangordonglmark2 Score: 2215:37
freemangordona bit better15:37
tmlindfreemangordon: so does n900 hang for you if you make omap_gem_is_cached_coherent() always return false?15:51
tmlindor comment out the calls for omap_gem_is_cached_coherent() in both places or just one place?15:52
freemangordontmlind: will try later on16:02
tmlindok16:44
freemangordontmlind: hmm: https://gitlab.freedesktop.org/xorg/xserver/-/commit/db9e9d45e8ba73510f11eb9e534c176102f6623e17:30
Wizzuplooks like we want that17:33
freemangordonъеах17:33
freemangordonyeah17:33
freemangordonI am going to take the whole ms directory into our xserver17:34
freemangordonto see how/if it will work17:34
Wizzupcan we not use latest xorg?17:35
Wizzupor did that have the dependency hell17:35
freemangordonabi has changed17:36
freemangordonI guess we can just pick whatever patches are needed17:36
freemangordonif it makes sense17:36
Wizzupok17:36
freemangordonhmm, actually we need only modesetting driver17:43
freemangordonwell, we will need stuff from chimaera when and if it comes to building it in the repos17:44
freemangordonbut, I am not sure how close we are to that :)17:44
Wizzupright18:05
Wizzupyou probably know better than we do18:05
freemangordonhmm, this one seems to support 16 bpp as well, at least by looking at the commits18:31
Wizzupfun18:41
freemangordonupstream master xorg build finished, lets see if there will be any difference19:06
_uvos_freemangordon: so a drm ioctl exists that updates the display19:30
_uvos_and xorg for sure can use it with radion/amdgpu19:30
_uvos_i have a vrr display that shows update frequency19:30
_uvos_and it works fine there19:30
_uvos_so maybe investigating how it works there might be prudent19:31
_uvos_the missing tiles is a different issue19:32
_uvos_at least i think so19:32
_uvos_modesetting has an optin to turn this on19:33
_uvos_cant check rn19:33
_uvos_google it or wait untill i can look at my xorg config19:34
freemangordonwe're getting there :)19:46
freemangordonwith upstream xserver I no longer see excessive CPU usage on n90019:47
Wizzupyay19:48
freemangordonlets see what glmark will say19:51
freemangordonbut better do that with traces disbled :D19:52
freemangordonhmm, after I disabled the traces, high CPU usage is back hhere19:58
freemangordonstill:19:59
freemangordonglmark2 Score: 2519:59
freemangordonthat's better19:59
freemangordonoh, ok, we need that FlushBehaviour set to 220:01
Wizzupand then high cpu is gone?20:08
freemangordonmhm20:08
Wizzupneat20:11
uvos.BI "Option \*qVariableRefresh\*q \*q" boolean \*q20:43
uvosEnables support for enabling variable refresh on the Screen's CRTCs when an suitable application is flipping via the Present extension.20:43
uvoshttps://gitlab.freedesktop.org/xorg/xserver/-/blob/master/hw/xfree86/drivers/modesetting/modesetting.man20:43
freemangordonglmark2 Score: 26 :)20:45
uvosthats pretty bad compeared to on drm no?20:45
freemangordon3720:46
freemangordonnot *that* bad20:46
uvosi gues20:46
uvossurvivable20:46
freemangordonI think I can improve it a bit20:46
Wizzupsorry for asking, but I have to ask, does h-d work with a sample app, say osso-xterm ?20:46
Wizzup:D20:46
uvosim more concernd with it working that it having full perf anyhow20:46
Wizzupjust excited20:46
freemangordonyes20:46
freemangordonbut, there are terrible rendering artifacts20:47
Wizzupaha20:47
freemangordonI suspect because my glamor replacement doesnt implement TFP yet20:47
uvosWizzup: so sql was about https://github.com/maemo-leste/sphone/blob/master/src/utils/store.c20:47
uvosWizzup: i need to make this a module and add some new proparties to it20:47
uvos(eg what backend produced an event)20:47
freemangordonalso, all the 2d rendering is done through mmaped VRAM20:48
freemangordonttyl20:48
Wizzupfreemangordon: ttyl20:48
uvoswe can add a rtcom backend too20:48
uvosno problem20:48
Wizzupyeah, I need to look at rtcom, but I think it abstracts the sql away mostly20:48
WizzupI can get a schema of the db in a bit from my n90020:48
uvosrtcom is closed source?20:49
uvosanyhow indipendant sphone path needs to continue to work20:49
uvosbut having 2 modules for logging is no issue ofc20:49
Wizzupuvos: well the libraries are not, but the phone and sms app are also 'rtcom'20:52
uvosWizzup: ok so are the librarys that specifcl do communications event logging closed source?20:53
freemangordonno20:54
Wizzupright they are not, but the handlers are, but sphone can be a handler20:54
uvoswhats a handler in this context?20:55
Wizzupa program that listens to ofono sms notifications and stores them rtcom20:55
uvosright20:55
uvossure sphone should be that20:55
uvos:P20:55
Wizzup(in the fremantle case they also have a ui)20:55
uvoswell for phone events anyhow (not nesscarly sms)20:55
uvosbtw since everything in sphone is just events on datapipes20:55
uvosthats very gui toolkit neutral20:56
uvoswe could add another sms ui in qt20:56
uvosas a plugin20:56
uvosnot sure if that makes sense20:56
WizzupI am hoping to get that going soon (will have to)20:57
Wizzupqt plugin in gtk is probably tricky20:57
uvosWizzup: its not gtk anymore its just glib20:57
uvosand its not very ticky at that point20:58
uvosi can wirte a quick example plugin20:58
uvosmostly its not even glib anymore just plain c20:58
uvoswhat you interact with20:58
WizzupI mean gtk and qt in the same progress20:59
Wizzuptomorrow around 1300 I am switching ISP, so leste.maemo.org will be down for an unknown time (same for jenkins), but hopefully less than an hour or so21:11
uvosWizzup: https://github.com/maemo-leste/sphone/tree/sphone-qt23:31
Wizzupheh.... I wish and you make it happen eh23:32
uvosWizzup: slightly cursed version of sphone that uses the fact that on linux QAplication is just glib mainloop in a trenchcoat to allow qt and gtk to interoperate23:32
uvosif you load the test module23:32
uvoswhich i dident push23:33
uvosupps sec23:33
uvosok23:34
uvosso if you load test https://github.com/maemo-leste/sphone/blob/sphone-qt/src/modules/test.cpp23:34
Wizzupbut, so this has gtk and qt live in the same application?23:34
uvosyeah23:34
uvossphone will show you a qt widget when you click on "contacts" in the gtk dialer window23:34
Wizzupok23:35
uvosso now you can go create a sphone-sms ui that is in qt or we can slowly port it or whatever23:35
uvosport it (the ui) to qt that is23:35
Wizzupok23:35
uvosso anyhow i wont merge this into mainline sphone - unless you want to go this route in some way23:36
uvos(since it forces a link to both qt and gtk wich is strange and scary :P)23:36
WizzupI need a few more days before I start hacking on yappari and turn it into conversations stuff23:37
WizzupI want to start on it now, but I need to finish other stuff first23:37
Wizzupthen I'll look at what makes sense integration wise23:37
uvosok23:37
Wizzupgreat that you hacked this up already, though :)23:37
Wizzupas in, I think we need a news post first now :P23:37
WizzupI'm travelling on tuesday, but after that I'm on my own for ~2 weeks so should have lots of leste time23:38
uvosgreat :)23:38
uvosanyhow ttyl getting some sleep23:38
Wizzupttyl!23:38

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