| *** raymonddull has quit IRC | 00:06 | |
| *** raymonddull has joined #xdandroid | 00:06 | |
| *** virsys has quit IRC | 00:09 | |
| *** virsys has joined #xdandroid | 00:13 | |
| *** bzo has quit IRC | 00:32 | |
| *** rpierce99 has quit IRC | 00:36 | |
| *** d3tul3 has quit IRC | 00:44 | |
| *** kalemas has joined #xdandroid | 02:03 | |
| *** corein has joined #xdandroid | 03:28 | |
| *** programmer8922 has joined #xdandroid | 04:49 | |
| *** hardwalker has quit IRC | 05:22 | |
| *** programmer8922 has quit IRC | 06:21 | |
| *** d3tul3 has joined #xdandroid | 06:55 | |
| *** raymonddull has quit IRC | 07:38 | |
| *** d3tul3 has quit IRC | 08:08 | |
| *** mgross029 has joined #xdandroid | 08:24 | |
| *** corein has quit IRC | 08:24 | |
| *** Kraln has quit IRC | 08:26 | |
| *** Kraln has joined #xdandroid | 08:36 | |
| *** dario_ has joined #xdandroid | 09:25 | |
| *** rpierce99 has joined #xdandroid | 09:34 | |
| *** helicopter88 has joined #xdandroid | 09:35 | |
| *** dario_ has quit IRC | 10:14 | |
| *** Tandyman100 has joined #xdandroid | 10:18 | |
| *** corein has joined #xdandroid | 10:34 | |
| *** vinceweis has joined #xdandroid | 11:25 | |
| *** raymonddull has joined #xdandroid | 11:40 | |
| *** ray|yar has joined #xdandroid | 12:01 | |
| *** raymonddull has quit IRC | 12:02 | |
| *** ray|yar is now known as raymonddull | 12:03 | |
| *** newbie has joined #xdandroid | 12:09 | |
| *** newbie is now known as Guest42698 | 12:09 | |
| *** helicopter88 has quit IRC | 12:11 | |
| *** Guest42698 is now known as helicopter88 | 12:23 | |
| *** programmer8922 has joined #xdandroid | 12:39 | |
| *** programmer8922 has quit IRC | 12:44 | |
| *** |Jeroen| has joined #xdandroid | 12:52 | |
| mgross029 | Quiet in here today | 13:19 |
|---|---|---|
| * mgross029 looks around and listens to the crickets | 13:19 | |
| rpierce99 | solved all of the worlds problems yesterday | 13:19 |
| mgross029 | ha | 13:19 |
| mgross029 | gee I must have missed something in the logs. :p | 13:19 |
| mgross029 | rpierce99, which rhod you on? | 13:20 |
| rpierce99 | the best | 13:21 |
| rpierce99 | the only one that matters | 13:21 |
| * mgross029 wonders why he even bothered to ask. :p | 13:22 | |
| rpierce99 | oh you don't know which one that is? | 13:22 |
| mgross029 | I'm guessing at rhod400, but wasn't sure | 13:22 |
| rpierce99 | you can be sure, rhod400 is the best and the only one that matters | 13:23 |
| mgross029 | Ummm Ok... If you say so. | 13:24 |
| rpierce99 | my tongue-in-cheek self-centeredness is not coming through very well i think | 13:25 |
| mgross029 | It sure is but my aloof reponses to your tongue-in-cheek self-centeredness must have been mistaken | 13:26 |
| mgross029 | Anyway, so saw a lot of .39 commits yesterday wistilt2 was pretty busy. | 13:28 |
| rpierce99 | i didn't look at them but i believe he was pulling in detules changes, so that they were both working from a shared base | 13:28 |
| mgross029 | Ahh that's right remember reading that in the logs | 13:29 |
| rpierce99 | he may have snuck in the change that allows non-rhodw to use it without losing the incall screen | 13:30 |
| arrrghhh | i'm sure he snuck that in | 13:30 |
| arrrghhh | considering it benefits him :P | 13:31 |
| mgross029 | cool | 13:31 |
| arrrghhh | damn you nmap | 13:31 |
| mgross029 | He will have to change his git repo name. :p | 13:31 |
| arrrghhh | ? | 13:31 |
| mgross029 | linux-msm-rhod | 13:32 |
| rpierce99 | it's still rhod, just not only 400/500 (W) now | 13:32 |
| arrrghhh | went flying over my head | 13:32 |
| arrrghhh | yea... i don't get it | 13:32 |
| arrrghhh | still RHOD... | 13:32 |
| arrrghhh | and his repo was always for all RHOD's... it's just the prox GPIO poop that needs tweaking for GSM folk | 13:32 |
| mgross029 | Ahhh.. That flew over my head then... Thought you were talking NON-Rhod period | 13:33 |
| rpierce99 | no, snuck the w in on the end there | 13:33 |
| rpierce99 | worldphones | 13:33 |
| rpierce99 | the gsm only folk couldn't use prox | 13:33 |
| arrrghhh | yea those heathens don't test anything anymoar | 13:33 |
| mgross029 | Interesting did not know that | 13:33 |
| rpierce99 | no one did, until someone (wistilt2) tested it | 13:34 |
| arrrghhh | indeed | 13:35 |
| arrrghhh | emwe's GSM, but his RHOD is the good one. | 13:35 |
| mgross029 | Someone with a working prox that is. :p | 13:35 |
| arrrghhh | he's got a RHOD400 dude. | 13:35 |
| arrrghhh | so even if his prox did work... | 13:35 |
| arrrghhh | he wouldn't have known the problem with GSM devices. | 13:35 |
| mgross029 | Oh that's tru. duh | 13:36 |
| *** GlemSom has joined #xdandroid | 13:38 | |
| *** Tandyman100 has quit IRC | 13:43 | |
| *** Kraln- has joined #xdandroid | 13:44 | |
| *** rpierce99 has quit IRC | 13:45 | |
| *** Kraln has quit IRC | 13:45 | |
| *** Entropy512 has quit IRC | 13:45 | |
| *** rpierce99 has joined #xdandroid | 13:46 | |
| *** Entropy512 has joined #xdandroid | 13:46 | |
| rpierce99 | that was fun, i had my very own #xdandroid all to myself there for a bit | 13:47 |
| arrrghhh | lol | 13:47 |
| arrrghhh | were you OP? | 13:47 |
| rpierce99 | don't think so | 13:47 |
| *** Entropy512 has joined #xdandroid | 13:47 | |
| *** AndersG has joined #xdandroid | 13:48 | |
| arrrghhh | bummer | 13:48 |
| rpierce99 | maybe if i had left and rejoined | 13:48 |
| rpierce99 | thursday night football is back already? that sucks | 13:50 |
| *** Tandyman100 has joined #xdandroid | 13:51 | |
| detule | alright i am fairly certain that the OOM killer in .39 is not working as it should | 13:59 |
| detule | or at all | 13:59 |
| rpierce99 | the kernel oom killer? do we really want it to run? | 14:00 |
| detule | i thought android's runs on top of it | 14:00 |
| rpierce99 | i have no idea how androids works, I've hypothesized that the kernel one is why we get reboots when downloading with the market though | 14:00 |
| detule | need emwe up in here so i can bisect him with questions | 14:01 |
| rpierce99 | because we have memory overallocation enabled | 14:01 |
| rpierce99 | yeah it looks like the android modification was to allow userspace to define the settings for the kernel oom handler | 14:04 |
| *** helicopter88 is now known as helicAWAY | 14:25 | |
| detule | first off, emwe's kernel has some sort of a swap of its own | 14:28 |
| detule | second, i am using some sort of a memory widget, and on .39 i can easily drop the available memory to under 20MB | 14:28 |
| detule | at which point stuff starts breaking left and right | 14:28 |
| detule | this just by openning maps, market, kindle app, browser | 14:28 |
| detule | my launcher has an open tasks widget, so i can see they remain persistent in memory and are not closed | 14:29 |
| detule | otoh, with .35 i can not drop available memory under 40MB | 14:29 |
| rpierce99 | what is your lmk value? | 14:29 |
| detule | brb | 14:29 |
| *** |Jeroen| has quit IRC | 14:31 | |
| arrrghhh | i'm betting the lmk stuff isn't working on .39 like he said. | 14:33 |
| arrrghhh | the rootfs sets that value, and assuming he's comparing apples to apples... | 14:33 |
| *** helicAWAY is now known as helicopter88 | 14:46 | |
| detule | on .35 i notice items disappearing out of the task list also | 15:03 |
| detule | as in the oom killer is actually working | 15:03 |
| detule | on .39 not so much they stay there until i manually hit the big 'close all programs' button | 15:03 |
| *** bzo has joined #xdandroid | 15:21 | |
| arrrghhh | detule, that's really odd | 15:22 |
| arrrghhh | but i guess i don't really grasp what in the kernel those OOM/LMK values actually do | 15:22 |
| detule | neither do i | 15:22 |
| arrrghhh | so perhaps .39 doesn't have the correct 'hooks' for userland to make any adjustments? | 15:22 |
| arrrghhh | lol | 15:23 |
| arrrghhh | guessing is fun :D | 15:23 |
| bzo | there is something really screwed in 39 as far as android accessing memory | 15:23 |
| bzo | right when I boot up, android thinks it only has about 50mb to use | 15:23 |
| arrrghhh | bzo! what's up. | 15:23 |
| arrrghhh | lol | 15:23 |
| bzo | hey arrrghhh | 15:24 |
| bzo | and after doing stuff the number keeps going lower until everything grinds to a halt | 15:24 |
| arrrghhh | hm | 15:24 |
| bzo | you would expected used+available to stay the same | 15:24 |
| bzo | but that sum keeps going down | 15:24 |
| arrrghhh | odd.. | 15:24 |
| bzo | when I boot .35, android thinks it has about 130mb to use | 15:24 |
| arrrghhh | you're not removing RAM physically from the phone are you bzo ? :P | 15:25 |
| bzo | I think this is the heart of the slowdown problem in 39 | 15:25 |
| bzo | detule - when you go to settings->apps->manage->running, what used+avail numbers do you see? | 15:26 |
| detule | bzo currently in a state of sod with .35 + usb | 15:27 |
| detule | i am going through the mm/Makefile | 15:27 |
| detule | wth is MEMBLOCK? | 15:27 |
| arrrghhh | heh | 15:27 |
| detule | bzo you are reading this about oom and .39 -> i can't get it to kill anything | 15:28 |
| detule | if i let it, it will drop to <20mb available | 15:28 |
| detule | .35 and .39 boot with approx the same avail memory for me.....aprox 90mb | 15:28 |
| bzo | doesn't work regardless of v6 super script, right? | 15:29 |
| detule | right same system same data as .35......in .35 i can't bring available memory below 40MB | 15:30 |
| detule | it starts killing things off | 15:30 |
| detule | in .39 it doesn't seem to kill anything off the list | 15:30 |
| bzo | I think that may be a symptom of a bigger problem | 15:32 |
| bzo | I don't understand how used+free keeps going down, you would think it should be constant | 15:32 |
| detule | android could be doing stuff in the background | 15:33 |
| bzo | sure, but why would the total pool of memory shrink like that? | 15:34 |
| bzo | memory use seems fairly constant on the kernel side | 15:34 |
| bzo | free is showing 10-20mb available even when android is pegged | 15:34 |
| bzo | pegged meaning something like 30mb used + 1mb free | 15:35 |
| bzo | maybe the memory free mechanism is not working between the kernel and dalvik, causing a huge memory leak | 15:36 |
| detule | hm what's in the kernel in terms of connecting with dalvik | 15:36 |
| bzo | dunno, though the memory mechanism is probably not specific to android/dalvik | 15:37 |
| *** infidelus has joined #xdandroid | 15:40 | |
| detule | this memblock is really suspicious | 15:51 |
| bzo | doesn't seem to be set in .35 | 15:53 |
| *** kalemas has left #xdandroid | 15:54 | |
| detule | or in .27 for that matter | 15:55 |
| detule | everything else is more or less the same except for this percpu_up vs percpu in comparison to .35 | 15:55 |
| detule | also in .39 we are not setting VMALLOC_RESERVE | 15:56 |
| bzo | I see some other android devices that have memblock enabled, so perhaps that is not a red flag | 16:00 |
| bzo | does seem like vmalloc_reserve should be set though | 16:01 |
| *** Tandyman100 has quit IRC | 16:02 | |
| detule | i am just going to start listing things i think we should change | 16:03 |
| detule | CONFIG_DEFAULT_MMAP_MIN_ADDR=32768 | 16:03 |
| detule | er scratch that cyano has it 32K, .35 is 32K, .39 and .27 at 4K | 16:06 |
| *** raymonddull has quit IRC | 16:09 | |
| *** Tandyman100 has joined #xdandroid | 16:11 | |
| detule | bzo, how come we are hardcoding this mem value CONFIG_CMDLINE="mem=64M console=ttyMSM,115200n8" | 16:12 |
| arrrghhh | make it 1gb | 16:12 |
| bzo | no clue... | 16:12 |
| Tandyman100 | <Picard> Make is so. | 16:12 |
| Tandyman100 | it* | 16:12 |
| arrrghhh | /fail | 16:12 |
| Tandyman100 | That's just like me; try to make a bad joke, screw it up with a typo. | 16:12 |
| arrrghhh | lol | 16:13 |
| bzo | I do vaguely recall jonpry having problems with the 2nd memory bank | 16:13 |
| bzo | hopefully he figured it out, and this is just a remnant of his testing | 16:13 |
| detule | i wonder if that default cmdline is needed for kexec | 16:17 |
| *** GlemSom has quit IRC | 16:20 | |
| *** mgross029 has quit IRC | 16:30 | |
| *** AndersG has quit IRC | 16:32 | |
| *** helicopter88 has quit IRC | 17:01 | |
| *** programmer8922 has joined #xdandroid | 17:32 | |
| *** rpierce99 has quit IRC | 17:48 | |
| *** fishhead2567 has quit IRC | 17:58 | |
| *** ImCoKeMaN has quit IRC | 18:24 | |
| *** raymonddull has joined #xdandroid | 18:32 | |
| *** corein has quit IRC | 18:52 | |
| *** programmer8922 has quit IRC | 19:03 | |
| *** rpierce99 has joined #xdandroid | 19:14 | |
| *** d3tul3 has joined #xdandroid | 19:18 | |
| d3tul3 | ok looks like the oom_killer was in fact rewritten in .39 | 19:24 |
| d3tul3 | oom_adj is deprecated in favor for oom_score_adj | 19:24 |
| rpierce99 | is there a simple place in userland to change that, or in the kernel to change it back? | 19:25 |
| d3tul3 | they didn't leave us a way to simply revert back to oom_adj | 19:25 |
| bzo | d3tul3: sounds like you may have found the problem | 19:28 |
| bzo | cat oom_score_adj = -1000 | 19:28 |
| bzo | OOM_SCORE_ADJ_MIN | 19:28 |
| d3tul3 | yeah something is not right with those values | 19:28 |
| d3tul3 | ok so i need to understand this lowmemorykiller is sitting on top of oom_kill | 19:29 |
| d3tul3 | lowmemorykiller has sysfs parameters adj and minfree | 19:30 |
| d3tul3 | is adj relevant to the old oom_adj | 19:30 |
| d3tul3 | or to the new oom_score_adj | 19:30 |
| d3tul3 | either way actually it won't work | 19:30 |
| d3tul3 | adj currently has values ranging 0 - 8 | 19:30 |
| d3tul3 | if relevant to oom_score_adj it should have values -1000 to 1000 or something like that | 19:30 |
| bzo | maybe they are independent | 19:32 |
| bzo | the oom stuff may be handling stuff on the kernel side | 19:32 |
| bzo | and lmk is purely an android process killing mechanism | 19:33 |
| bzo | when I echo 0 to oom_score_adj, linux free memory goes up to 30mb | 19:33 |
| bzo | but the android mem didn't seem to change | 19:33 |
| d3tul3 | hm i thought lmk sits on top of oom | 19:33 |
| d3tul3 | perhaps it only uses the oom score and kills independently | 19:34 |
| d3tul3 | i think though the moral of the story is that in .39 i don't think oom_adj is actually updated | 19:34 |
| d3tul3 | so if lmk looks at oom_adj to figure out what to kill | 19:34 |
| bzo | where is the lmk source anyways? I'm not seeing it at the expected location of /drivers/misc | 19:35 |
| d3tul3 | drivers staging android | 19:35 |
| d3tul3 | "The lowmemorykiller driver lets user-space specify a set of memory thresholds | 19:35 |
| d3tul3 | * where processes with a range of oom_adj values will get killed." | 19:35 |
| d3tul3 | ok since oom_kill was rewritten by google in .39 it seems, we just need to find an up to date lowmemorykiller | 19:36 |
| bzo | I seem to recall that jonpry started from the mainline 39 code | 19:37 |
| bzo | if so, it is entirely possible an old lmk was integrated | 19:37 |
| d3tul3 | i can't find one based on oom_score_adj.... | 19:42 |
| d3tul3 | hm i guess we can just scale | 19:42 |
| d3tul3 | 0-17 -> 0-100? | 19:42 |
| bzo | lol, the fun of being on the bleeding edge | 19:43 |
| bzo | I'm not sure there are any other .39 android kernels, are there? | 19:43 |
| d3tul3 | i haven't seen any | 19:43 |
| *** larryone has quit IRC | 19:46 | |
| rpierce99 | https://lkml.org/lkml/2010/2/15/322 decent description of the difference between the two | 19:49 |
| d3tul3 | dinnertime bbl | 19:49 |
| *** larryone has joined #xdandroid | 19:59 | |
| rpierce99 | https://lkml.org/lkml/2010/2/15/315 says that oom_adj may still be used with the old range but it is then scaled to oom_score_adj units for a rough linear approximation. | 20:01 |
| d3tul3 | only the write handle scales | 20:08 |
| d3tul3 | the read handle (which probably lowmemorykiller) doesn't scale (i think) | 20:08 |
| bzo | I think the old stuff may be deprecated but functional | 20:08 |
| bzo | cm has a .37 kernel that uses the same lmk driver | 20:09 |
| bzo | new oom stuff started in .36 | 20:09 |
| d3tul3 | look at oom_adjust_read and *_write here https://gitorious.org/~detule/linux-msm-rhod/detules-linux-msm-rhod/blobs/9c8f8e3867cca493dcb2887592b03d3c9a06979a/fs/proc/base.c | 20:09 |
| d3tul3 | they even give us the scale formula, we just use that formula in the _read handle and we're in business | 20:10 |
| d3tul3 | oh bzo, didn't realize that, perhaps this oom business is not it after all | 20:11 |
| bzo | yeah, I'm starting to think this is not it | 20:11 |
| bzo | did you try any of the kernel config changes yet? | 20:13 |
| d3tul3 | which ones? memblock didn't help | 20:14 |
| bzo | how about the vmalloc setting? | 20:14 |
| d3tul3 | on the to-do-list for later tonight bbl family time | 20:18 |
| bzo | can probably skip that, looks like the VMALLOC_RESERVE is not part of the 39 tree | 20:23 |
| bzo | and in the .35 tree, it is being set to the default of 128m anyways (which is what it is hardcoded to in .39) | 20:24 |
| *** hardwalker has joined #xdandroid | 20:29 | |
| d3tul3 | there's no way lowmemorykiller is working as it should when every single process returns oom_adj -16 | 20:42 |
| d3tul3 | need to boot 35 to see what kind of values processes report there | 20:42 |
| d3tul3 | 35 reports 4 for gapps, 10 for maps | 20:52 |
| d3tul3 | (both -16 on .39) | 20:53 |
| d3tul3 | echo "6" > debug_level in lowmemorykiller/parameters spits out all the lmk activity in dmesg (dmesg | grep adj) | 21:00 |
| *** infidelus has quit IRC | 21:00 | |
| *** BulitPruf has joined #xdandroid | 21:38 | |
| *** BulitPruf has left #xdandroid | 21:49 | |
| *** [7] has quit IRC | 22:01 | |
| *** TheSeven has joined #xdandroid | 22:01 | |
| *** raymonddull has quit IRC | 22:02 | |
| *** programmer8922 has joined #xdandroid | 22:48 | |
| *** TheSeven has quit IRC | 22:59 | |
| *** [7] has joined #xdandroid | 22:59 | |
| *** programmer8922 has quit IRC | 23:45 | |
Generated by irclog2html.py 2.7 by Marius Gedminas - find it at mg.pov.lt!