| Wizzup | happy new year! | 00:44 |
|---|---|---|
| Wizzup | this year will be great :) | 00:44 |
| r3boot | It will :) Enjoy! :) | 00:48 |
| sicelo | kexecboot starts on droid 4 without a battery in the device. Wonder what's missing to allow an OS to start too | 11:34 |
| freemangordon | Wizzup: booting with non-cma buffers patch :) | 11:57 |
| freemangordon | first try that is | 11:57 |
| Wizzup | freemangordon: great | 12:35 |
| freemangordon | Wizzup: seems to wokr | 12:53 |
| freemangordon | *work | 12:53 |
| freemangordon | Wizzup: do you have old n900 image around? with old kernel/pvr? | 12:56 |
| freemangordon | I need to know what FPS achieve that when swiping hildon-dekstop | 12:56 |
| freemangordon | also, gears fps | 12:57 |
| freemangordon | also, what is bpp of Xorg | 12:57 |
| Wizzup | freemangordon: I can dd some old(er) image | 13:03 |
| Wizzup | freemangordon: pretty sure we're on 24bit regardless | 13:03 |
| freemangordon | ok | 13:03 |
| freemangordon | but I would like to have fps | 13:03 |
| freemangordon | because current seems a bit low to me | 13:03 |
| Wizzup | ok | 13:04 |
| Wizzup | wunderw - still around? | 13:04 |
| Wizzup | or is that maybe wunderwungiel[m] | 13:04 |
| Wizzup | I wonder what the latest image is that still worked | 13:04 |
| Wizzup | maybe maemo-leste-1.0-armhf-n900-20211031.img.xz | 13:04 |
| Wizzup | dd'ing | 13:10 |
| freemangordon | Wizzup: how do you want to test the patch? I think it would be easier to give you .patch file,no? | 13:12 |
| Wizzup | yes | 13:16 |
| freemangordon | ok | 13:16 |
| freemangordon | Wizzup: final test and will provide you the patches | 13:20 |
| Wizzup | ok | 13:22 |
| freemangordon | hmm... | 13:25 |
| freemangordon | maybe it doesn't work | 13:25 |
| freemangordon | I cannot open any application | 13:25 |
| freemangordon | or, it is something else | 13:26 |
| Wizzup | freemangordon: do you want test with or without clutter | 13:26 |
| freemangordon | anyway, I'll provide the patches to you | 13:26 |
| freemangordon | shoudl not matter | 13:27 |
| freemangordon | Wizzup: sent over the email | 13:28 |
| freemangordon | hmm, I don;t see any memory allocation errors, so it must be something else | 13:32 |
| freemangordon | Wizzup: you have to patch both kernel and omap driver | 13:34 |
| Wizzup | ok | 13:36 |
| freemangordon | what?!? | 13:38 |
| freemangordon | died loading invoked application: '/usr/bin/controlpanel.launch: cannot dynamically load position-independent executable' | 13:38 |
| freemangordon | why we hit that again? | 13:39 |
| Wizzup | glibc stuff? | 13:39 |
| freemangordon | hmm | 13:39 |
| freemangordon | oh | 13:39 |
| freemangordon | I guess I have new glibc on that device | 13:39 |
| freemangordon | yeah 2.31-4 | 13:40 |
| freemangordon | ok, I think I have to flash an image | 13:40 |
| freemangordon | not sure I can revert | 13:40 |
| freemangordon | lets try though | 13:40 |
| freemangordon | Wizzup: but wait, how hildon-desktop starts then? | 13:41 |
| freemangordon | maybe we didn;t fix all the applications | 13:41 |
| freemangordon | yeah, looks like that | 13:41 |
| Wizzup | weird, I don't have this problem | 13:42 |
| Wizzup | are you sure you're up to date/ | 13:42 |
| freemangordon | I installed libc from chimaera back then | 13:42 |
| Wizzup | freemangordon: what is clutter fps again | 13:44 |
| Wizzup | env | 13:44 |
| freemangordon | CLUTTER_SHOW_FPS=1 | 13:44 |
| Wizzup | max fps I get is 49 | 13:45 |
| Wizzup | but most is lower than that | 13:45 |
| Wizzup | between 25-35 | 13:46 |
| freemangordon | yeah, max is what I wanted to know | 13:46 |
| freemangordon | here we hit max 40 | 13:47 |
| freemangordon | I know why is that, but this should be fixed in the kernel | 13:47 |
| freemangordon | the p[oint is now omap driver waits for rendering to complete before queuing page flip | 13:48 |
| Wizzup | ok | 13:48 |
| freemangordon | this is again that missing fence | 13:48 |
| Wizzup | freemangordon: so I should also disable cma now? | 13:49 |
| Wizzup | in .config | 13:49 |
| freemangordon | no, you cannot | 13:49 |
| freemangordon | just lower the value to 16MB | 13:49 |
| Wizzup | ok | 13:50 |
| Wizzup | building kernel | 13:52 |
| Wizzup | freemangordon: so apply your xf86-video-omap on top of your triple buffering commit? | 13:53 |
| freemangordon | yes | 13:54 |
| Wizzup | ok | 13:54 |
| Wizzup | freemangordon: I get like 18 fps | 14:09 |
| freemangordon | go to launcher | 14:09 |
| freemangordon | then back | 14:09 |
| Wizzup | wait | 14:09 |
| Wizzup | forgot omap ddx | 14:10 |
| freemangordon | Wizzup: hmm? | 14:16 |
| Wizzup | sec | 14:16 |
| Wizzup | freemangordon: well I forgot to dpkg -i the ddx I built | 14:16 |
| freemangordon | ah | 14:16 |
| Wizzup | freemangordon: but now with the new ddx it basically crashed almost right away | 14:17 |
| freemangordon | weird | 14:17 |
| Wizzup | with: | 14:17 |
| Wizzup | [ 138.991149] cma: cma_alloc: reserved: alloc failed, req-size: 375 pages, ret: -16 | 14:17 |
| Wizzup | maybe I installed the wrong ddx? | 14:17 |
| Wizzup | hmm | 14:17 |
| freemangordon | see, maybe my patch does not work | 14:17 |
| freemangordon | but yeah, you must have the correct ddx | 14:17 |
| Wizzup | I think I have | 14:17 |
| Wizzup | let me try once more | 14:17 |
| Wizzup | but it's weird since it seemed fine before | 14:18 |
| freemangordon | here it goes locale generation, will take a while it seems :( | 14:18 |
| freemangordon | but at least I think I will be able to downgrade | 14:19 |
| sicelo | Wizzup: did sre's patch for acx565akm help? | 14:30 |
| Wizzup | the panel? yes, I replied that on ml | 14:30 |
| sicelo | oh nice. for some reason i don't get all the messages anymore. but nice if that helps | 14:31 |
| freemangordon | Wizzup: any progress? | 14:33 |
| Wizzup | checking | 14:39 |
| Wizzup | max fps is like 6 | 14:40 |
| Wizzup | 26 | 14:40 |
| Wizzup | no crash this time at least | 14:41 |
| Wizzup | well not yet | 14:41 |
| freemangordon | enter and exit the launcher | 14:41 |
| freemangordon | and check fps then | 14:41 |
| Wizzup | yes I did that | 14:42 |
| freemangordon | still 26? | 14:42 |
| Wizzup | X crashes again | 14:42 |
| freemangordon | hmm | 14:42 |
| Wizzup | [ 131.519] (WW) OMAP(0): WARNING: get vblank counter failed: Invalid argument | 14:42 |
| Wizzup | [ 131.521] (WW) OMAP(0): WARNING: flip queue failed: Invalid argument | 14:42 |
| Wizzup | [ 639.147] (EE) | 14:42 |
| Wizzup | [ 639.153] (EE) Backtrace: | 14:42 |
| Wizzup | [ 639.243] (EE) | 14:42 |
| Wizzup | [ 639.244] (EE) Segmentation fault at address 0x10 | 14:42 |
| Wizzup | [ 116.882995] omapdrm omapdrm.0: Failed to setup plane gfx | 14:42 |
| Wizzup | [ 539.129241] omapdrm omapdrm.0: Failed to setup plane gfx | 14:42 |
| Wizzup | [ 639.150726] cma: cma_alloc: reserved: alloc failed, req-size: 375 pages, ret: -16 | 14:42 |
| freemangordon | ugh | 14:42 |
| freemangordon | well, not sure what's going on | 14:43 |
| freemangordon | trying to bring my n900 back | 14:43 |
| Wizzup | you don't get this on n900? | 14:43 |
| Wizzup | ah ok | 14:43 |
| freemangordon | no, but I didn;t have most of the things runnung | 14:43 |
| Wizzup | yeah it usually happens for me when I start apps | 14:43 |
| freemangordon | hmm , maybe 16MiB is too small | 14:44 |
| freemangordon | could you try with 32? | 14:44 |
| Wizzup | could be, but I also got this even with 64MiB some times, but that was before your patch | 14:44 |
| Wizzup | sure | 14:44 |
| freemangordon | yeah, now the patch should change the thing | 14:44 |
| Wizzup | building kernel | 14:45 |
| Wizzup | trying now | 15:11 |
| freemangordon | oh,ok, seems I will have to dd | 15:13 |
| freemangordon | but what? | 15:13 |
| Wizzup | you mean what iamge? | 15:14 |
| freemangordon | mhm | 15:14 |
| Wizzup | https://maedevu.maemo.org/images/n900/20211031/ this boots ok | 15:14 |
| freemangordon | will flash latest | 15:14 |
| Wizzup | ok, you just need to fix mime and hope other stuff wasn't broken | 15:14 |
| Wizzup | but for kernel you will need to do some more stuff I guess | 15:14 |
| freemangordon | hmm yeah, other stuff | 15:14 |
| Wizzup | with 32MB I got the same failure with just h-d scrolling | 15:15 |
| freemangordon | ok, lemme dd and boot and test | 15:15 |
| freemangordon | I wonder why it continues to use CMA | 15:16 |
| Wizzup | mhm | 15:23 |
| Wizzup | anything you want me to check | 15:23 |
| freemangordon | no | 15:24 |
| freemangordon | maybe my patch is inclomplete | 15:25 |
| freemangordon | *incomplete | 15:25 |
| freemangordon | but I need to boot here to verify | 15:26 |
| freemangordon | ugh, that one tears like hell | 15:30 |
| freemangordon | I guess that's why it is so fast | 15:30 |
| freemangordon | Wizzup: I shall upgrade to -experimental and then put my kernel, right? | 15:31 |
| Wizzup | freemangordon: yes | 15:31 |
| freemangordon | ok | 15:31 |
| Wizzup | freemangordon: -devel might be ok too | 15:31 |
| freemangordon | ok | 15:32 |
| Wizzup | freemangordon: make sure to have the right libs | 15:32 |
| Wizzup | I haven't set up the n900 meta yet | 15:32 |
| Wizzup | freemangordon: as in ti343x | 15:32 |
| freemangordon | yeah | 15:32 |
| Wizzup | I still need to try dtb boot | 15:33 |
| Wizzup | (without appended I mean) | 15:34 |
| freemangordon | mhm | 15:34 |
| Wizzup | shouldn't be too hard | 15:35 |
| freemangordon | Wizzup: what is the kernel package name/version? | 15:37 |
| freemangordon | or, we don;t have one yet? | 15:39 |
| Wizzup | freemangordon: add droid4 component and insall linux-image-droid4 iirc | 15:39 |
| Wizzup | or linux-droid4-image | 15:39 |
| Wizzup | let me check | 15:39 |
| freemangordon | ah, ok | 15:39 |
| Wizzup | linux-image-droid4 | 15:39 |
| freemangordon | ok | 15:39 |
| freemangordon | Wizzup: how many applications to open? | 16:18 |
| freemangordon | Wizzup: also, any clue why wifi does not work? | 16:19 |
| Wizzup | freemangordon: wifi should work with our tree | 16:21 |
| Wizzup | freemangordon: xterm, cpa, and just click around is enough | 16:21 |
| freemangordon | hmm, cannot repro here | 16:22 |
| freemangordon | but, with *my* tree | 16:22 |
| Wizzup | freemangordon: I took your patch and applied it on top of this https://github.com/maemo-leste/droid4-linux/commits/wip/n900/maemo-5.15-cleaned-up | 16:22 |
| Wizzup | and then also decrease cma from 64 to 32 | 16:22 |
| freemangordon | I opened xterm, cpa, clock, pdf,qalendar | 16:22 |
| Wizzup | hmmm | 16:22 |
| freemangordon | and that's with xorg from the repo | 16:22 |
| freemangordon | lemme check cma size | 16:22 |
| Wizzup | let me update the n900 I have | 16:22 |
| Wizzup | I maybe forgot to dist-upgrade it | 16:23 |
| Wizzup | although I don't think it matters | 16:23 |
| Wizzup | 52 packages can be upgraded. Run 'apt list --upgradable' to see them. | 16:23 |
| Wizzup | The following NEW packages will be installed: ti-omap3-sgx xserver-xorg-video-pvrsgx | 16:23 |
| freemangordon | 65536K cma-reserved | 16:23 |
| Wizzup | oof | 16:24 |
| Wizzup | 64MB then | 16:24 |
| freemangordon | mhm | 16:24 |
| freemangordon | lemme reduce that | 16:24 |
| Wizzup | it helps but you still run into it | 16:25 |
| freemangordon | mhm | 16:25 |
| freemangordon | yep, got that with 16mb | 16:41 |
| Wizzup | ok | 16:42 |
| freemangordon | hmm, maybe I shall take https://github.com/maemo-leste/xf86-video-omap/blob/master/src/omap_dri2.c#L237 into account | 16:43 |
| freemangordon | pfff | 16:43 |
| freemangordon | Wizzup: hmm, I think we shall do some calculations first | 16:56 |
| freemangordon | like, if we want to do page flipping, what is the needed CMA per application | 16:56 |
| freemangordon | maybe we shall not use page flipping on n900 | 16:57 |
| freemangordon | and instead do blits, the same way Nokia did it | 16:57 |
| freemangordon | I guess they had reason to do it like that | 16:57 |
| freemangordon | so, each application is 2-buffered in clutter - if we want to flip, those buffers must be CMA allocated | 17:02 |
| freemangordon | so we have ~2MiB per application for those buffers | 17:03 |
| freemangordon | and 1MiB for the 3rd buffer | 17:04 |
| freemangordon | thus, we need 2xnum_apps+1 MiB CMA just for the buffers, IIUC | 17:04 |
| freemangordon | hmm, what the? even after Xorg crashed, CMA free is still 0?!? | 17:08 |
| freemangordon | looks like someone leaks cma memory | 17:10 |
| Wizzup | freemangordon: I think so too @ memory leak | 17:29 |
| Wizzup | freemangordon: does cma show you what allocs what | 18:17 |
| freemangordon | I don't think so | 18:20 |
| freemangordon | in proc/meminfo there is something about it | 18:20 |
| freemangordon | but I don;t know what it means | 18:20 |
| freemangordon | like, what cmafree is | 18:20 |
| freemangordon | hmm, with 24MB CMA I have now 13 apps open | 18:21 |
| freemangordon | 8 of them are xterm, vbut still | 18:21 |
| freemangordon | Wizzup: could you try with 32MB, to see if you can trigger the crash? | 18:22 |
| freemangordon | Wizzup: with 24MiB CMA I can start 11 xterms (one of them running htop)+settings+calendar+calculator+pdf+ham | 18:38 |
| freemangordon | I don;t think it makes sense to torture the device anymore than that :) | 18:39 |
| sicelo | fmg the exorcist :-p | 18:51 |
| Wizzup | freemangordon: I had the crash with 32MB almost right after boot | 19:13 |
| Wizzup | freemangordon: well, we still don't want X to crash if it runs out of vram | 19:13 |
| freemangordon | Wizzup: agree (crash), but why it is not crashing for me? | 19:44 |
| freemangordon | Wizzup: hmm compaction is disabled in the kernel you use, right? | 19:46 |
| Wizzup | freemangordon: yes | 19:49 |
| freemangordon | that might explain it | 19:49 |
| Wizzup | maybe try to disable it | 19:49 |
| Wizzup | we don't want it on | 19:49 |
| freemangordon | yes, but who is going to compact CMA? | 19:49 |
| freemangordon | but ok | 19:49 |
| Wizzup | this kernel feature is new | 19:50 |
| Wizzup | like 5.9 new or something | 19:50 |
| Wizzup | clearly it worked before as well | 19:50 |
| freemangordon | did it? I remember having hard times allocating CMA memory back then | 19:50 |
| freemangordon | to the point I was fixing some drivers to have private CMA pool | 19:51 |
| freemangordon | omapfb especially ;) | 19:51 |
| freemangordon | Wizzup: also, what do you think is the correct way to handle vram OOM? | 19:51 |
| freemangordon | shall we fix the kernel to pass __GFP_NOFAIL (or whatever the flag is)? | 19:52 |
| freemangordon | kernel == omapdrm | 19:52 |
| freemangordon | Wizzup: it is CONFIG_COMPACTION, right? | 19:57 |
| Wizzup | freemangordon: yes | 19:57 |
| freemangordon | I don't think we can disable that | 19:58 |
| freemangordon | https://cateee.net/lkddb/web-lkddb/COMPACTION.html | 19:58 |
| Wizzup | freemangordon: maybe killing the X window | 19:58 |
| freemangordon | we cannot disable compaction, IIUC without it there is no way CMA to work | 19:58 |
| Wizzup | it is new since 5.9 | 19:58 |
| Wizzup | so unless CMA never worked | 19:58 |
| Wizzup | let me re-check | 19:58 |
| freemangordon | found in Linux kernels: 3.7–3.19, 4.0–4.20, 5.0–5.15, 5.16-rc+HEAD | 19:59 |
| Wizzup | freemangordon: maybe we enable it but revert facdaa917c4d5a376d09d25865f5a863f906234a | 19:59 |
| freemangordon | I see | 20:00 |
| freemangordon | lemme read that huge commi message | 20:00 |
| Wizzup | freemangordon: no nee | 20:00 |
| Wizzup | tl;dr is the guy measured some jvm improvements when he wakes up the system twice a second | 20:00 |
| Wizzup | and he's ignored all our of emails so far | 20:00 |
| Wizzup | or might be on a very long vacation | 20:00 |
| Wizzup | this kills off mode | 20:00 |
| Wizzup | and also is a pm nightmare | 20:00 |
| Wizzup | the commit doesn't provide a way to stop the wake ups or even tweak the interval | 20:01 |
| Wizzup | freemangordon: I'll revert that commit and re-enable the config | 20:01 |
| Wizzup | and test | 20:01 |
| freemangordon | Wizzup: just re-enable compaction and test | 20:02 |
| freemangordon | to see if this is the one that leads to almost instant crashes on you device | 20:02 |
| freemangordon | we'll think about PM if we know this is the problematic feature | 20:03 |
| Wizzup | freemangordon: no, I'm going to revert that commit in any case | 20:04 |
| Wizzup | we don't want it | 20:04 |
| Wizzup | and I already reverted it anyway | 20:04 |
| freemangordon | Wizzup: wait a second | 20:04 |
| freemangordon | ah, ok :) | 20:04 |
| freemangordon | wouldn;t it be easier to just replace HPAGE_FRAG_CHECK_INTERVAL_MSEC with some tunable? | 20:05 |
| freemangordon | like 30s or something | 20:06 |
| freemangordon | as this looks like a useful feature to me | 20:06 |
| Wizzup | maybe, but they did not respond to any of my emails so I'm not particularly interested in working with them | 20:06 |
| freemangordon | am I in CC? I think I saw 2 mails so far? | 20:07 |
| freemangordon | I'll ping tomorrow | 20:07 |
| freemangordon | we may ask tmlind to pester them too | 20:07 |
| freemangordon | also, gta04 guys should be bitten by that one too | 20:08 |
| freemangordon | basically, this is every baterry-operated device | 20:09 |
| freemangordon | *battery | 20:09 |
| freemangordon | Wizzup: how often is n900 woken-up without this? | 20:09 |
| freemangordon | who the hell thought that "static const int HPAGE_FRAG_CHECK_INTERVAL_MSEC = 500;" is a good idea? | 20:12 |
| Wizzup | freemangordon: not very often at all wrt wakeup | 20:16 |
| Wizzup | freemangordon: tmlind also asked | 20:16 |
| freemangordon | yeah, I know | 20:16 |
| Wizzup | freemangordon: like less than 0.5 wakeups/s | 20:16 |
| Wizzup | when I did some measurements | 20:16 |
| Wizzup | sometimes way less because it stays in off mode for 30+ seconds | 20:17 |
| freemangordon | ok, so waking up once every 30 seconds should not affect PM, right? | 20:17 |
| freemangordon | but yeah, maybe we don;t really need that | 20:17 |
| freemangordon | so, setting the tunable to 0 shall disable that feature, iiuc | 20:18 |
| freemangordon | but right, lets simply revert until we have a better solution | 20:18 |
| Wizzup | freemangordon: no, setting that via sysctl does not stop wakeups | 20:30 |
| Wizzup | freemangordon: I treid my hand at reverting it, if this doesn't work I'll try to use it as is and then see if the bug still occurs | 20:32 |
| freemangordon | yes, I see that setting it to 0 does not stop wakeups, I meant that setting ti to zero should stop wakeups :) | 20:35 |
| freemangordon | sorry for not being clear | 20:35 |
| freemangordon | the patch to stop wakeups for 0 seems like one-liner though | 20:36 |
| Wizzup | maybe | 20:36 |
| Wizzup | don't think it will be a one liner since you need to change the interval probably disable the old timer | 20:36 |
| Wizzup | but fwiw yes, I don't see crashes now | 20:36 |
| freemangordon | good | 20:37 |
| Wizzup | so I suggest we do a careful revert of that commit | 20:37 |
| Wizzup | (it conflicts in ~6 places) | 20:37 |
| Wizzup | and also another commit uses that sysctl | 20:38 |
| Wizzup | either that or we really up that interval value | 20:38 |
| Wizzup | but I think really upping that value probably defeats the point in the measurements this guy did | 20:38 |
| freemangordon | I think it is easier to just increase the wakeup interval to 60 seconds or something | 20:38 |
| Wizzup | k | 20:38 |
| Wizzup | I'll do that and try | 20:38 |
| freemangordon | ttyl | 20:39 |
| Wizzup | I'll make it two mins | 20:39 |
| freemangordon | ok | 20:40 |
| Wizzup | btw I still don't see like 40fps I think | 20:40 |
| Wizzup | but I will try again once this new kernel hits | 20:40 |
| Wizzup | maybe cma makes a diff | 20:40 |
| Wizzup | err compaction | 20:40 |
| Wizzup | freemangordon: ok so with these patches I will start with omap-linux | 20:51 |
| Wizzup | agreed? | 21:04 |
| Wizzup | freemangordon: I pushed to wip/n900/maemo-5.15-cleaned-up and confirmed that with much higher interval we are ok with off mode | 21:12 |
| freemangordon | Wizzup: yep, I am fine. I will have to create formal patches in xorg omap and linux trees | 22:04 |
| freemangordon | but not now | 22:04 |
| Wizzup | freemangordon: sure | 23:08 |
Generated by irclog2html.py 2.17.0 by Marius Gedminas - find it at https://mg.pov.lt/irclog2html/!