| 07:55 | topi` | does anyone happen to have a working .dts file for the hummingboard2 variants for the 4.x kernels? |
| 08:02 | topi` | found the patch on spinics.net |
| 08:46 | vpeter | topi`: you can also look here https://github.com/xbianonpi/xbian-sources-kernel/tree/imx6-4.4.y |
| 08:49 | topi` | is this your source tree ) |
| 08:59 | vpeter | no, from mk01. |
| 09:59 | jnettlet | Its official. https://blog.nxp.com/iot/step-up-your-next-audio-video-and-voice-iot-development-with-i-mx-8m-ecosystem-partners |
| 10:04 | vpeter | familiar name there :-) |
| 10:05 | topi` | jnettlet: so basically you get early access to the first IMX8 products? |
| 10:05 | vpeter | jnettlet: Is this next big thing? |
| 10:05 | topi` | the post is full of marketing speak, so is this the correct summary? |
| 10:05 | jnettlet | yes |
| 10:06 | jnettlet | it is the next thing :) |
| 10:06 | topi` | vpeter: I seriously doubt NXP+freescale has the resources what e.g. Qualcomm has to actually produce bleeding edge products |
| 10:06 | topi` | "innovative", next thing stuff |
| 10:06 | topi` | QC can throw 1000 engineers at a problem |
| 10:06 | topi` | NXP has their hands full in managing the freescale acquisition |
| 10:06 | topi` | but probably doesn't affect the day-to-day work of engineers |
| 10:07 | topi` | but honestly, I expected to get some Cortex-A73 cores instead of the lackluster A53 :/ |
| 10:08 | topi` | the A53 is just basically as weak as the A9, but an in-order core |
| 10:08 | topi` | of course it can probably be clocked higher, like the Mediatek P20 at 2.2ghz |
| 10:08 | suihkulokki | pft, A53 is just fine unless you want to power a sauna :P |
| 10:08 | topi` | it is fine unless you're doing some compute intensive stuff |
| 10:09 | jnettlet | topi`, this is the iMX8M the iMX8 QuadPlus and QuadMax have A72s |
| 10:10 | vpeter | 10:06 QC can throw 1000 engineers at a problem <<< this reminds me on a problem few years back. In our company there were 5 people working on a problem at some conformance testing. At the same time Huawei has full room of them. But we won anyway. So it is not quantity but quality :-) |
| 10:10 | jnettlet | the A53s fit our customers better. All about thermals and powermanagement. |
| 10:10 | suihkulokki | more interesting is if the vaguely defined "4 shader opengl es 3.1 / vulcan gpu" is still vivante or not |
| 10:11 | jnettlet | it is. a GC3000 |
| 10:11 | jnettlet | it might be a slightly higher rev, but that family |
| 10:11 | suihkulokki | so it's actually worth working on mainlining the graphics stack |
| 10:12 | suihkulokki | unlike things with mali/powervr which can essentially be treated as headless hw :P |
| 10:12 | jnettlet | I believe wumpus already has basic working GC3000 support. |
| 10:13 | suihkulokki | ! |
| 10:13 | jnettlet | I am most interested to see how Vivante's Vulkan support is |
| 10:13 | wumpus | yes, I have done some work on GC3000, not all of it pushed yet |
| 10:14 | topi` | vpeter: I agree, but I was just pointing out that I have doubts ;) |
| 10:14 | wumpus | especially integer ALU througput in the shader became much faster, as the integer operations are 4-wide now |
| 10:14 | topi` | sounds good for any OpenCL worloads |
| 10:15 | wumpus | and it can finally render to a single buffer again instead of the multi pipe split buffer madness |
| 10:15 | wumpus | which helps rendering to textures as well as getting the image on the screen... |
| 10:16 | wumpus | apart from that it's quite similar to GC2000 |
| 10:16 | wumpus | oh it can load shaders from memory instead of registers so supports much larger shaders |
| 10:16 | jnettlet | yes very important |
| 10:17 | topi` | it seems Vivante wants to stay relevant, after all ;) |
| 10:19 | wumpus | I also wonder about Vivante's vulkan support. I also wonder about Mesa's vulkan support. I haven't done anything with vulkan yet tbh |
| 10:20 | jnettlet | wumpus, I know Vulkan crashes quite nicely on our Braswell platform, and my Skylake laptop for that matter. |
| 10:20 | jnettlet | it seems very rough. Only place where I have had success is with NVidia |
| 10:22 | jnettlet | for those interested. http://www.vivantecorp.com/en/technology/3d.html |
| 10:23 | wumpus | with the blob drivers or nouveau? |
| 10:23 | jnettlet | on Android...so blob |
| 10:23 | topi` | I just hope the new imx 8M will learn from the mistakes in i.mx6... like the gigabit ethernet bandwidth, memory bandwidth... |
| 10:24 | jnettlet | well it does 4K so we have to assume the memory bandwidth issues have been fixed. Not to mention support for LPDDR4 |
| 10:24 | jnettlet | and 4K 60...proper HDMI 2.0a and HEVC |
| 10:24 | topi` | any idea on which process they will build it? Probably 28nm |
| 10:24 | topi` | i.mx6 was 40nm |
| 10:27 | wumpus | jnettlet: ok that makes sense, nvidia is usually a bit ahead in the drivers game with new APIs and such |
| 10:28 | jnettlet | The SoCs are fabricated with a 28nm FDSOI (fully depleted silicon on insulator) process touted for improved MTBF and decreased soft error rates. |
| 10:29 | topi` | yes, the newest 28nm processes also leak less current than the older ones |
| 10:29 | topi` | witness the excellent power characterstics of the Snapdragon 652, for example. Quad-core A72 and made on a 28nm process |
| 10:30 | topi` | though it's more about how often stuff gets to run on the A53 cores ;) |
| 10:46 | topi` | what I'd like to see in NXP's chips, would be the same kind of ability to rapidly change clocks, like Intel does, to achieve higher perf for short operations |
| 10:47 | topi` | but I guess it's intel's secret sauce and nobody else can do it |
| 10:49 | jnettlet | well it is also more needed because of the long instruction pipelines that intel chips have. |
| 10:51 | topi` | the A53 has a fairly short pipeline |
| 10:51 | jnettlet | yep |
| 10:51 | topi` | it did surprisingly well on kernel compilation, (I benched it on the ODROID-C2) |
| 10:51 | jnettlet | they are very good low power chips |
| 10:52 | topi` | I was especially surprised to see that the memory bandwidth against cortex-A9 systems was essentially double |
| 10:53 | topi` | it's like those Alpha cores of the last millennium, they were wide in-order designs that were very dependent on correct instruction layout by the compiler |
| 10:53 | suihkulokki | I wonder if SOMs built with i.MX8 and i.IMX8-M can share same carrierboard |
| 10:53 | topi` | but had heaps of mem bandwidth to compensate |
| 10:54 | topi` | when I was young, I optimized the mpg123 decoder in hand assembly for the Alpha A21064 chip (removing pipeline stalls) |
| 10:55 | topi` | the gcc output was full of data dependencies, it was essentially calculating everything by using just 3 registers :) a great way to stall a long pipeline |
| 10:55 | topi` | nowadays the gcc 4 and newer can automatically unroll these kinds of loops |
| 10:56 | topi` | (our student guild had one Alpha machine donated by DEC :) |
| 10:57 | jnettlet | We used to have an alpha machine running NT back in the day |
| 11:55 | topi` | is there any rational explanation for the behavior of UBOOT on the HB? sometimes it works as expected, sometimes I just see the solid-run logo and nothing else for a while, seemingly some operation is blocking |
| 11:57 | topi` | it also does not react to my keyboard presses, even though this basic Microsoft keyboard is known to work with UBOOT |
| 11:57 | topi` | and I mostly use it to hack uboot variables |
| 12:07 | jnettlet | this is not a behaviour I am aware of. If you get the solid-run logo then you should get a command prompt. |
| 12:08 | jnettlet | wireless keyboards can cause issues due to the usb implementation in u-boot |
| 12:10 | jnettlet | that is better |
| 12:33 | topi` | is rabeeh at work? we need some kind of certification documents for the Hummingboard |
| 12:34 | topi` | we're probably going to need some certification about the EMC characteristics, etc |
| 12:34 | topi` | lots of paperwork |
| 12:56 | topi` | Artox: what's the proper way of re-building the initrd image? |
| 12:56 | topi` | I unpacked it with gzip -dc | cpio -i |
| 12:56 | topi` | modified stuff |
| 12:56 | topi` | and then rebuilt with find . | cpio -o -c | gzip -9c >initrd.img |
| 12:57 | topi` | but the result is 100MB of size! |
| 12:57 | topi` | 211458 blocks, reports cpio. Sounds odd because cpio -i said "34714 blocks" |
| 13:04 | topi` | update-initramfs (from debian pkg) seems to do a reasonably sized initrd of a few megabytes |
| 13:05 | jnettlet | topi`, just email support for the docs. We just finished emf certification for the HB2 |
| 13:06 | jnettlet | but our SOMs have been certified to go into outerspace, so we have lots of documentation now |
| 13:20 | topi` | :D |
| 13:22 | topi` | what, are you going to build a satellite with i.mx6? :) |
| 13:35 | Artox | topi`: |
| 13:35 | Artox | sudo update-initramfs -u |
| 13:35 | jnettlet | topi`, http://lunar.xprize.org/teams/hakuto |
| 13:35 | jnettlet | powered by SolidRun |
| 13:36 | topi` | Artox: the damn update-initramfs didn't pick my firmware files in /lib/firmware |
| 13:36 | Artox | if you want to modify its content, check /etc/initramfs-tools |
| 13:36 | Artox | for firmware, have a look at /etc/initramfs-tools/hooks/imx-sdma script |
| 13:37 | Artox | I hear there is a newer way to do this, but I haven't looked into it |
| 13:38 | Artox | its been a long time since I wrote that hook script |
| 13:38 | topi` | what is imx-sdma? sometimes I get the message that it failed to load firmware |
| 13:38 | Artox | its a kernel module that requires firmware as early as initramfs |
| 13:39 | Artox | which is why that script makes rue teh firmware is included |
| 13:39 | topi` | so that's why it has to go into the initrd? |
| 13:39 | Artox | when that firmware faisl to load, you are booting eitehr without initrd, or with one thats missing the firmware |
| 13:39 | Artox | yep |
| 13:39 | topi` | what kind of subsystem is that imx-sdma anyways? |
| 13:39 | Artox | apparently that module can't load the firmware later |
| 13:40 | Artox | that infromation is not in my archive |
| 13:40 | Artox | its some kind of memory controller |
| 13:41 | Artox | oh, and you had trouble with connman right? |
| 13:41 | Artox | updated connman is in the repos. |
| 13:41 | jnettlet | it is the dma engine. it works without the firmware, but the firmware does some optimizations for things like hdmi audio etc |
| 13:51 | topi` | I think everyone is having trouble with connman :) |
| 13:51 | topi` | so why do you insist on it, instead of the tried and tested /etc/network/interfaces ;) |
| 13:52 | topi` | granted, scripting with the legacy interfaces file gets ugly... especially when having to use wifi/wpa |
| 13:52 | topi` | but nothing beats it for a quick "iface eth0 inet dhcp" kind of thing :) |
| 13:55 | jnettlet | topi`, nothing stops you from using that. But for up and running quick connman works. The issues have been because debian updated systemd on their stable repos to something that broke compatibility |
| 13:55 | jnettlet | this is a debian failure not a connman failure |
| 13:55 | topi` | you have the right to sympathise for connman :) |
| 13:56 | topi` | somebody said that hackers are highly opinionated folk. Maybe it is so :) |
| 13:56 | suihkulokki | I note i.MX8 series is completly missing from current mainline |
| 13:56 | jnettlet | how much screwing around do you have to do in order to get tethering working using /etc/network/interfaces and hostapd.conf etc. connmanctl tethering wifi on essid password done |
| 13:57 | suihkulokki | is there an evil vendor kernel publicly around? |
| 13:57 | jnettlet | suihkulokki, the soc isn't even released yet. |
| 13:57 | topi` | jnettlet: I guess ppl just lack basic knowledge on how to use connman |
| 13:57 | topi` | everything is *ctl nowadays :) |
| 13:58 | jnettlet | it is easy enough to install NetworkManager if preferred. The problem is that Debian's version is so old it is missing a lot of useful things. |
| 13:58 | topi` | and networkmanager sucks anyawys |
| 13:58 | topi` | I just kill it :) |
| 13:59 | jnettlet | not at all, you won't find a more featurely networkmanager |
| 13:59 | jnettlet | and now it is all modular so you can pick and choose what you want. |
| 14:00 | jnettlet | everything just takes time |
| 14:01 | jnettle | 14:01 * jnettlet can't wait to replace wpa_supplicant with intel's new wifi manager |
| 14:14 | topi` | I just want to replace the brcmfmac with some saner wifi :) |
| 14:15 | topi` | the brcmfmac is somewhat unreliable with hostapd |
| 14:15 | topi` | I wish this isn't the case with Wilink8 |
| 14:19 | jnettlet | even with the most recent kernels? I just backported a lot of more recent patches that fix AP oriented functionality. |
| 14:19 | topi` | I think I've been running 3.14.60 something |
| 14:19 | topi` | or was that 3.14.54 |
| 14:20 | jnettlet | but even hostapd I would make sure you have the latest version. I have run AP mode using connmanctl which uses wpa_supplicant and it has run for months without a hiccup |
| 14:21 | topi` | the debian/jessie version of hostapd might be a bit old :) |
| 14:26 | topi` | uboot 2013.10-rc4, isn't this a very old uboot? |
| 14:39 | jnettlet | yes, but we prioritize software built for industrial deployments. They don't want things upgraded to the latest software release so they can fight off more bugs. |
| 14:40 | jnettlet | I have at least 6 bugs patched in our u-boot that still exist in mainline and will cause a boot failure on iMX6 |
| 14:42 | vpeter | Here is another one :) https://github.com/SolidRun/u-boot-imx6/pull/9/commits/c33d8da2a0cea60c85f2455a5c8328648a18f4fc |
| 14:51 | jnettlet | vpeter, that is actually not a bug. It is C sementics. in C++ that is a bug. |
| 14:58 | vpeter | How is that? dt_prefix is constant string length and you just append something (and overwrite whatever is after this string. |
| 15:16 | vpeter | But maybe u-boot works different way. |
| 16:29 | topi` | jnettlet: we were able to have a stable picture on the 4.11-rc1 kernel. it seems to have working drm_kms_helper.edid_firmware |
| 16:29 | topi` | and my colleague converted fb timings into an EDID firmware blob |
| 16:30 | topi` | now the only pain point is the lack of hw acceleration in DRM :) |
| 16:30 | topi` | and we have to have 12 Hummingboards somehow from somewhere ASAP |
| 16:31 | topi` | otherwise the bosses say they're going to put some PCs instead |
| 16:36 | topi` | ah, Ilya says 12 pcs are in stock right now :) |
| 16:39 | topi` | what's the best procedure to switch the HB2 to boot from the eMMC instead of SD? we'd have to write UBOOT to the eMMC in some fashion, probably manually. |
| 16:40 | topi` | and for that, you'd need to boot from the SD card, so doing the eFuses beforehand is out of the question |
| 17:17 | jnettlet | vpeter, I will look into it, but that is actually functionality that is generally unused. |
| 17:18 | jnettlet | wumpus, topi`, okay just looked at the press releases...looks like the i.MX 8M has a GC7000Lite |
| 17:19 | jnettlet | this is Vega 8X series. |
| 17:21 | jnettlet | it also comes with a Cortex- M4F core on the SOC, this is aimed at ultra low power always on audio processing like the Amazon echo, or Google Home. |
| 18:48 | vpeter | jnettlet: Could be and error never happen. |