**** BEGIN LOGGING AT Mon Aug 20 03:00:01 2018 Aug 20 07:23:12 Someone's working on mainlining pinctrl for MT6797: http://lists.infradead.org/pipermail/linux-mediatek/2018-August/014707.html Aug 20 07:23:41 meanwhile, something still overwrites memory in my kernel :/ Aug 20 07:31:16 How is your kernel different from the stock gemini kernel then? Aug 20 07:32:12 it's plain mainline, with nothing working except for busybox over TTY Aug 20 07:32:32 TTY being a serial port terminal, no graphics yet Aug 20 07:32:53 Ah starting from scratch, that's very brave! Aug 20 07:34:00 yeah, I sorta bought the gemini for that kind of a project Aug 20 09:21:02 Jasu_M: single bits, bytes or bigger areas? Aug 20 09:21:35 can you compare the memory map from Gemini's original kernel and mainline? Aug 20 09:22:10 does it overwrite in the whole memory range or only a certain area? Aug 20 09:22:49 if later you could simple forbid access to it for now (the linux kernel has options to disable memory aread - mostly used for disabling broken memory Aug 20 09:57:38 mifritscher: I've compared the memory map, it's identical except for the dynamically reserved (alloc-ranges) regions, which should not matter Aug 20 09:58:00 I'm getting single-bit flips and maybe others. Aug 20 09:58:30 I've checked that any non-reserved regions are not protected by MediaTek's memory protection unit Aug 20 09:59:18 I have no clue about this platform, but I've seen similar problems on a TI SOC with faulty clock configurations. Aug 20 09:59:50 clocks are one suspect. The Android kernel turns on spread-spectrum for CPU clocks, this doesn't exist in the mainline Aug 20 10:00:30 another suspect is M4U (the IOMMU), it is not initialized, but it might still be active and write fault information to memory or actuallly read pagetables from 0x0000000 and do something wrong Aug 20 10:00:31 from what i've seen spread spectrum usually does not make things better or more stable Aug 20 10:01:04 but you never know, every propriatary platform has tons of these little details that should be just-right to make everything work Aug 20 10:02:22 yeah... this is the first time I've tried to port linux kernel to anything, but I've been following sunxi efforts and it didn't seem particularly easy :P Aug 20 10:03:20 if the M4U thing doesn't work, I'm going to turn off clocks one by one and hope it starts working at some point Aug 20 10:04:22 the IOMMU system is super confusing on MediaTek, so I'm going to have a hard time wrapping my head around it Aug 20 10:05:14 Jasu_M: there are a couple of other people who've been working on something similar, so far I've not see anyone post a link to their source so everyone seems to be starting a fresh... (this is just me encouraging you to put something up on github or similar) Aug 20 10:05:46 like there's the M4U which is the IOMMU which sits between memory and SMI (smart multimedia interface) which has multiple LARBs (local arbiter) which are connected to different subsystems and request memory I/O through them... Aug 20 10:06:17 adam_b: my plan is to put the thing to GitHub when it works stably, if I can't get it to work soon I'll likely just dump it there anyway Aug 20 10:07:01 yea but if everyone waits till its stable then you'll all be doing the same work and you could have helped each other, but yea if you dump it before taking a break then thats at least something Aug 20 10:07:31 that's true. TBH I was expecting to make it work sooner so that wouldn't be an issue Aug 20 10:08:03 I'll dump it this weekend if it still doesn't work, that's a promise :P Aug 20 10:08:04 Oh yeah, i feel like the mediatek clocks are a bit wonky even on the official android kernel. Aug 20 10:08:19 Because I'm getting problems when i'm trying to measure time over a longer distance. Aug 20 10:09:04 Like, I had a bash timer that took the clockcount and tried to use it. I got like two minutes offshoot in a 20min measurement. Aug 20 10:09:15 I did not examine what causes it. Aug 20 10:09:48 how did you measure the time? did you do something like `sleep 2000` or something else+ Aug 20 10:10:09 with sleep yeah. Aug 20 10:10:34 ok. then I think it has to do with clocks rather than the real-time clock, I think at least... Aug 20 10:10:56 Yeah, i'd suppose so. Aug 20 10:15:35 It's of course possible that android causes it on termux, but still pretty odd behaviour. Aug 20 10:15:50 razumihin: was the timer late or early? I'm betting on late Aug 20 10:16:04 Late, yep. Aug 20 10:16:41 I'd hypothesize that it drops the clockspeeds after it runs the first part of the script. Aug 20 10:17:16 that's one possibility - another is that it does something that makes it lose timer IRQs Aug 20 10:18:42 Oh for the disclaimer, the kernel is not entirely "official" as i'm running Zevv's github version. Aug 20 10:19:42 But still though. Aug 20 10:41:42 I've sent a mail to planet requesting if they are willing to share some tech specs about the camera connector, so I can use this as extension for various twiddelings like a display Aug 20 10:42:03 I wonder what the answer will be - I wouldn't be suprised if Planet does not even has access to the schematics theirselves Aug 20 10:42:48 nice :) let's hope they answer Aug 20 10:44:45 Zevv I thought someone had already got that, thats how we know there is i2c pins Aug 20 10:45:00 there's always I2C pins on a camera :) Aug 20 10:45:12 and DuncanT has ordered some connectors Aug 20 10:45:22 sweet Aug 20 10:45:44 Zevv: what are you wanting to do with the camera connector? Aug 20 10:45:45 I thought I read above an offer to post you one Aug 20 10:46:13 True, but I didn't know the pinning was known Aug 20 10:46:41 its stuff they should really just put online for all to see Aug 20 10:47:04 like, are you going to physically replace the camera? or do something with software Aug 20 10:47:28 Yeah, but how much do they actually know and have? I feel they just have everything designed by a chinese partner, who might just have kept all the technical designs in house Aug 20 10:47:46 Jasu_M: from reading above Zevv and DuncanT are doing an external screen Aug 20 10:47:54 Jasu_M: it's the obvious extension to get power and a serial bus to custom peripherals Aug 20 10:49:12 nice, is there some I2C-only or something display you're going to hook there? Aug 20 10:49:18 but it is not a fun connector at all for prototyping at all Aug 20 10:52:21 I can imagine :P Aug 20 12:54:19 perhaps one can squeete in a bluetooth device, so only power is needed :-D Aug 20 12:54:32 *display Aug 20 14:13:07 hehe, i think "fitting" is the biggest problem here, not the connection Aug 20 14:13:44 I made some pictures the other day with the plastic cover removed: Aug 20 14:13:50 https://i.imgur.com/FgAbTPE.jpg Aug 20 14:14:11 The biggest advantage of your solution is that fitting in is easy: directly on the soldered battery leads Aug 20 14:14:21 everyone with basic soldering skills could do that Aug 20 14:14:56 The device is remarkably non hackable, as it is with all those shielded areas Aug 20 14:29:26 >i2c pins Aug 20 14:29:50 so when will someone make the necessary gemini-to-jolla adapter? ;) Aug 20 18:00:53 Am I back? **** ENDING LOGGING AT Mon Aug 20 18:09:39 2018 **** BEGIN LOGGING AT Mon Aug 27 01:24:04 2018 Aug 27 01:26:27 Every once in a while, my Gemini decides my SD card is "corrupt" and wants me to repurpose it. I just take it out of that silly slot that doesn't have a proper retention mechanism - and reseat it. Boom, works fine. But I'm getting tired of the SD Card dance....why can't this thing just work lol Aug 27 01:26:49 I just might use some kind of fixative next, its annoying as hell. **** ENDING LOGGING AT Mon Aug 27 03:00:00 2018