<oliv3r>
blunden: to answer your question; all community effort, zero, not even doc help, input from AW. Though one of their devs once answerd a question on the ML as far as I'm aware
hipboi has quit [Quit: Leaving]
shineworld has joined #linux-sunxi
hramrach__ has quit [*.net *.split]
hramrach__ has joined #linux-sunxi
FR^2 has joined #linux-sunxi
<oliv3r>
mripard_: i see we merged the emac driver as sun4i-emac; dind't we plan to use sunxi-emac instaed? (with sun4i-emac as DT bindings of course)?
<oliv3r>
all other drivers are currently sunxi too
<mripard_>
I had no mail telling me this, so I don't really know if it was part of your plan, but it wasn't part of mine
<mripard_>
and no, not all the drivers are named sunxi
cajg has joined #linux-sunxi
\\Mr_C\\ has quit []
<blunden>
oliv3r: ok
<wingrime>
oliv3r: how much unoptimal dram settings can casue slowness?
<wingrime>
oliv3r: ping
<wingrime>
hno: ping
<mnemoc>
moin
<hramrach__>
moin
<wingrime>
moin
jemk has joined #linux-sunxi
<rz2k>
wingrime: if you set dram clk to something like 100
<ssvb>
in my case it seems to be defined in /usr/include/xorg/dix.h
<ssvb>
mnemoc: :)
<mnemoc>
i use it always in my code too.... but defined on my own .h
shineworld has quit [Remote host closed the connection]
<PiyushVerma>
ssvb: I run some qt examples of opengl they running so slow that it looks like they not utilising acceleration any commet ?
<ssvb>
PiyushVerma: which version of qt?
<PiyushVerma>
4.8 prebuilt
<ssvb>
a much better GLES support is in Qt5
<PiyushVerma>
yes but aah need to build
<ssvb>
you need to check how Qt was configured in your case, the chances are it tries to use desktop OpenGL
<PiyushVerma>
I remember u build that before
<PiyushVerma>
ok
<PiyushVerma>
any tested steps to build qt5 ?
<mnemoc>
oh, the new server I got for sunxi includes 20TB of traffic..... nice!
<mnemoc>
no more fear about sharing images
<PiyushVerma>
mnemoc: wow
<PiyushVerma>
so huge trafic it's for sunxi ?
<hramrach__>
nice
<mnemoc>
it was near 5TB while dl.cubieboard.org was there
<hramrach__>
how much actual traffic was there?
<ssvb>
PiyushVerma: qt5 build is rather straightforward, it builds fine out of the box and no weird gymnastics is needed
<mnemoc>
but hipboi seems to have moved it into china
<hramrach__>
that does not sound like fast downloads for non-china
<mnemoc>
but anyhow, with 20TB I don't need to fear to be dropped to 10Mbps with dl.cubieboard.org or not
<PiyushVerma>
ssvb: going to put for native build let's see :) tomorrow should be ready
<ssvb>
PiyushVerma: btw, the ARRAY_SIZE issue should be now fixed in the github repo
<PiyushVerma>
ssvb: gr8
<wingrime>
ssvb: I just tested my image from a13 on a10 (replaced kernel and uboot) and have stranges
<wingrime>
htop cost 50% cpu
<ssvb>
wingrime: what exactly are you doing with it?
<wingrime>
ssvb: simply running in terminal
<wingrime>
ssvb: also just tested on fedora (without sunxifb) 35 % cpu
<ssvb>
wingrime: was it always this way, or is it a new regression?
<wingrime>
ssvb: same debian was quicker on a13
<ssvb>
I think you need to run a profiler to find what exactly is wrong
<ssvb>
there can be many possible reasons
<wingrime>
ssvb: starnge , same image was quicker on a13
<ssvb>
then it probably was not exactly the same, something could have been different (screen resolution, font size, the terminal window size where you run htop, ...)
<wingrime>
display size are same
hglm has joined #linux-sunxi
<hglm>
wingrime: Maybe it's a CPU frequency scaling issue? I had that too a few times (userland needs to properly set the CPU scaling thresholds, the default is very slow).
<ssvb>
yeah, that's a good point, running the CPU at 60MHz could behave like this
rellla has quit [Ping timeout: 264 seconds]
<wingrime>
ssvb: possible, while I used default .config
<wingrime>
ssvb: why we still not make #defconfig
<wingrime>
#define for linux and for android
Offshore has joined #linux-sunxi
<bfree>
zumbi: without the non-free mali stuff xf86-video-sunxi dosen't look hard. mali is messier, starting with needing libdri2-dev (you can just rebuild ubuntu's package), then needing to make it build all the relevant versions and then needing a sane way for it to setup the correct blob on the target (I hacked up a quick idea for a udev trigger to create the symlinks based on kernel and a config file setting for X/noX at boot)
<jemk>
ssvb: I did some research on cedarx memory limits. It's definitively possible to go beyond 256MB, decoding works there, but after some time i get kernel oops.
<bfree>
zumbi: great! I was just typing that I'd probably been a little dumb ignoring it thinking it was part of the non-free mess when it isn't really
<zumbi>
I am using mali stuff provided by hardkernel for the odroid board
<wingrime>
ssvb: thanks that was it , our default configs crap a13 have no CPUFREQ a10 have
<ssvb>
wingrime: thank hglm :)
<zumbi>
bfree: so, I am passing -lpthread in xf86-video-sunxifb configure.ac.. we'll see.. I need to package the binary part
<wingrime>
ssvb: are you tryed overclock dram ?
hglm has quit [Quit: leaving]
<ssvb>
zumbi: I'm still kinda considering different solutions for this libump pthread issue, but will probably push a fix today
<wingrime>
ssvb: I just tested 360 but than read a10-meminfo from andoid and now use 408
<ssvb>
wingrime: did you patch u-boot sources?
<techn_>
24bit modes looks good
<jemk>
wingrime: which mask, the or 0x70000000 i don't know, but have some ideas how to further investigate
shineworld has joined #linux-sunxi
<wingrime>
ssvb: current uboot have set default settings
<wingrime>
ssvb: iw16 or iw8
<wingrime>
ssvb: dram_sun4i_384_1024_iow8
<wingrime>
for example
<wingrime>
jemk: so how you use it beyond 256M?
<ssvb>
techn_: yes, but 24bpp modes have some other performance issues (this reminds me that it is a high time to write the next blog post about this stuff)
<jemk>
wingrime: looks like the highest 4 bits are moved to the lowest
tinti has joined #linux-sunxi
<jemk>
wingrime: funny way to use memory addresses, only discovered this by mistake
Black_Horseman has quit [Remote host closed the connection]
<wingrime>
jemk: you say that it must be alligned to 1k?
<jemk>
wingrime: output must be aligned to 1k, and now it seems input must be aligned to at least 16b too
<jemk>
wingrime: that was the mistake why i even discovered this, unaligned access
<wingrime>
jemk: than why OFFSET registers exist make sence
<ssvb>
jemk: lol about the address bits, that's a great find
<wingrime>
jemk: blob use that feature ?
<ssvb>
jemk: but the allwinner cedarx blobs are probably still 256MB limited
<jemk>
ssvb: don't think so, traces show similar effects, but as said, after some frames i get kernel oops
<ssvb>
jemk: simple tests show that the blob powered video players are buggy if the memory is reserved from above 256MB, so I guess we need to live with that until we get fully open source reverse engineered cedarx support
<bfree>
zumbi: I started to kinda look at it ~10 days ago after seeing my name called in vein a little here ;) I don't really care about the non-free stuff myself tbh. I ended up thinking it would need something like the following on top of the libdri2-dev: http://paste.debian.net/hidden/84b2f55d/
<ssvb>
jemk: and the memory reservation limitation can be easily removed once the drivers are ready
<wingrime>
ssvb: cedar kernel driver can be writen right after disp drm driver goes to mainline
<wingrime>
ssvb: about mem overclock, look like hi frequency also request lower CAS
<jemk>
is it possible that there is bug in kernel with reserved memory? writing to reserved mem by hand (so no cedar involved) leads to same issues as when letting cedar use mem above 256mb
<zumbi>
bfree: not sure, I need to look at it.. I'll share the results
<bfree>
zumbi: and I would happily put your sunxifb into my silly repo if you point me to the source ;) I just can't find the time+motivation to try and improve my horrible sunxi-tools package ... kernel and u-boot are ok I think (well u-boot is jumping ahead of debian which isn't ideal, but "ok" for us I think)
rellla has joined #linux-sunxi
<zumbi>
bfree: yep, that's fine
<zumbi>
bfree: I need to go an collect my computer holded by insurance company, then I can start setting up the autobuilding farm stuff
<zumbi>
I got ack from insurance I can get it back
<bfree>
zumbi: :-D I actually fell back to using qemu-user-static rather then real native compiling ;) still painful but not as bad as native
<zumbi>
bfree: qemu-user-static for compiling.. how is it? you get working binaries? iirc, you can fake gcc to use cross-gcc so you can speed up builds and env still think it is native build
<bfree>
zumbi: it just works and gives working binaries
<bfree>
zumbi: for "my" 3.4 kernel build, real times: cubie1 -j2 269m38.887s ... cubie2 -j4 164m39.431s ... qemu-user-static -j16 106m44.347s (on an i7-2600)
<zumbi>
bfree: and cross built 10m? :)
<zumbi>
bfree: but actually, qemu being not so bad
<bfree>
zumbi: no cross building ;) too much of a mess to make a clean cross builder on debian it seems (eventually multi-arch ftw I think)
<zumbi>
bfree: nod
atiti has joined #linux-sunxi
<hramrach__>
bfree: there is cross toolchain in emdebian but I doubt you can make dpkg do cross-builds even for something like kernel that has no dependencies
<ssvb>
jemk: how are you testing the memory accesses above 256MB?
<ssvb>
jemk: I could imagine the discontinuity in the physical memory address space (256MB in one place, a big hole, another 256MB), but it seems to be contiguous on my device
<jemk>
ssvb: sunxi_ve_mem_reserve=256 and manipulated libcedarx
atiti has quit [Ping timeout: 268 seconds]
<ssvb>
jemk: ok, but "writing to reserved mem by hand (so no cedar involved)" also fails?
<bfree>
hramrach__: yep, the "clean" was the critical word. maybe if emdebian was perfectly synced to sid it would be "ok", but ultimately for debian packages I'd still want to build them "natively" anyway
<jemk>
ssvb: it never fails imediatly, always little later i get kernel oops, so i think i have overwritten some kernel memory
<hramrach__>
I don't really care about clean. dpkg should save the build tools and options. if it does not they are unknow, whatever. it does not seem to have an option to cross-build, howeve
<hramrach__>
r
<ssvb>
jemk: ok, looks different from what I see (the system is alive, just no proper video playback)
<ssvb>
jemk: can you try to reserve a different address range (near the end of 512MB)? maybe it indeed happens to overlap something else?
<ssvb>
jemk: or just wait a bit, I'll push my CMA branch to github
<jemk>
ssvb: any way to do that without recompiling kernel, my crosscompiler is far away?
<ssvb>
no idea
* ssvb
is leaning to just using the first 256MB now and postponing this investigation until a later date
atiti has joined #linux-sunxi
<FR^2>
newbie-alarm ;) I've (hopefully successfully) compiled a kernel (with swap activated) using linux-sunxi/branch sunxi-3.3-cubieboard2
<FR^2>
should I just create a backup of /boot/uImage and copy the new kernel to /boot/uImage?
rellla has quit [Ping timeout: 240 seconds]
yuuki has joined #linux-sunxi
<yuuki>
trying to boot linux on A31 device (Kimdecent CS868). it boots pretty fine, but can't make to work on-board wifi module rtl8818eu. It tries to assoc and auth, log says "RTL871X: start auth", "RTL871X: auth success, start assoc", "RTL871X: assoc success", but after that I see "RTL871X: sta recv deauth reason code(2) sta:00:22:b0:f4:8a:75", "wpa_supplicant[1465]: Authentication with 00:00:00:00:00:00 timed out.", "wpa_supplicant[1465]: Association request
<yuuki>
to the driver failed"
<yuuki>
same /etc/network entry works perfectly on all other devices - hackberry, mk802 and so on
rz2k has quit [Ping timeout: 240 seconds]
<hramrach__>
yuuki: try asking the wpa_supplicant folks
<hramrach__>
it may be that the driver fails in some way but without knowing what wpa_supplicant wants and is not getting it's hard to tell
tinti has quit [Excess Flood]
<hramrach__>
also what exact kernel version?
<yuuki>
I meant wlan section in /etc/network/interfaces file
<hramrach__>
you don't really get any indication as to what failed. just something did not happen in time
tinti has joined #linux-sunxi
<hramrach__>
the driver is pretty new so it might not be exactly bug-free
rellla has joined #linux-sunxi
<hramrach__>
yuuki: are those messages from kernel or wpa-supplicant?
<shineworld>
with make ....... cubieboard2_defconfig
<bfree>
shineworld: http://linux-sunxi.org/Cubieboard2_wheezy_image mentions the problem. your u-boot and kernel probably disagree on the numbering. so "setenv machid 0xf35" in uboot (via script.bin or uenv.txt) should do it (or you can change the kernel or uboot to get them to agree)
<yuuki>
hramrach__ success messages from kernel, fail messages from wpa_supplicant
<yuuki>
hramrach__ I use same rootfs on several devices - all of them have no problems with wifi. In case of A31 I got weird problem. I think it is because of faulty rtl8818eu module or fex configuration
zeRez has joined #linux-sunxi
_BJFreeman has joined #linux-sunxi
FR^2 has quit [Quit: Connection reset by peer]
_BJFreeman is now known as BJfreeman
yuuki has quit [Quit: Leaving]
yuuki_ has joined #linux-sunxi
yuuki_ has quit [Client Quit]
rz2k has quit []
<ssvb>
has anybody tested a recent stage/sunxi-3.4 branch? looks like it got some sort of a really major ethernet performance regression :(
<ssvb>
crap, looks like I have to bisect it, because nfs root is now really unusable :(
<rm>
ssvb, did you try the usual (never reboot, always poweroff-coldstart)?
yuuki has joined #linux-sunxi
yuuki has quit [Client Quit]
<wingrime>
ssvb: Have any reason replace watchdog reboot with axp driven reboot ?
n01 has joined #linux-sunxi
<ssvb>
rm: yes, I tried a cold reboot and it did not help
<ssvb>
the old kernel never had bad ethernet performance and rebooting into it fixes the problem
<ssvb>
right now I'm in the middle of bisecting
ykchavan has joined #linux-sunxi
n01 has quit [Remote host closed the connection]
<paulk-desktop>
hi there
<paulk-desktop>
how come "mount("tmpfs", "/dev", "tmpfs", MS_NOSUID, "mode=0755");" mounts devtmpfs on some kernels?
<paulk-desktop>
mainly, that's what android does… and it does mount devtmpfs
<paulk-desktop>
however on a 3.4 kernel is just mounts an empty tmpfs
<paulk-desktop>
which I guess is the right thing to do
<paulk-desktop>
also, do you know how I can get mmcblk to show up on /dev/block instead of /dev ?
bsdfox_ has quit [Ping timeout: 264 seconds]
soul has joined #linux-sunxi
<Turl>
paulk-desktop: maybe one of the systems has an fstab where /dev is devtmpfs or something
<paulk-desktop>
that's the very early stage of init, there is no fstab yet
<paulk-desktop>
it's like the first thing the init binary does
<Turl>
paulk-desktop: I'd guess you have to tune your udev rules or such for that (/dev vs /dev/block)
<paulk-desktop>
isn't there a Kconfig option for that?
dragonn has joined #linux-sunxi
<paulk-desktop>
same for /dev/graphics/fb0 and /dev/fb0 it seems
<Turl>
paulk-desktop: I dunno, on my experience android's does it one way and the one on most distros the other way
<Turl>
I don't think it's kernel related
<paulk-desktop>
yeah right, but the way /dev is populated surely is up to the kernel
<Turl>
paulk-desktop: doesn't the kernel send events up to userspace and the daemon create the nodes?
<Turl>
(well, unless you're using devtmpfs I suppose)
<paulk-desktop>
right, here it uses devtmpfs
bsdfox_ has joined #linux-sunxi
BJfreeman has quit [Ping timeout: 252 seconds]
<ssvb>
rm: lol, I have identified the offending patch
hglm has joined #linux-sunxi
<rm>
hmm, link?
<ssvb>
just one more test round to confirm that it's the right one
<hglm>
ssvb: Interesting. I use NFS for light work (file sharing), so I didn't notice the performance regression, but I did see errors/stalls from the kernel NFS subsystem when copying large files (100MB). Not entirely sure this was on stage-3.4 or regular 3.4.
<paulk-desktop>
Turl, guess that you were right, it seems that ueventd creates the nodes on Android, so tmpfs is legit and it's not devtmpfs
<wingrime>
ssvb: I have probems with new driver updated
<wingrime>
/usr/bin/X: symbol lookup error: /usr/lib/xorg/modules/drivers/sunxifb_drv.so: undefined symbol: SunxiVideo_Init
<ssvb>
wingrime: run autogen.sh
<ssvb>
wingrime: some new files have been added, so the makefiles need to be regenerated
<wingrime>
ssvb: why not cmake for example ?
<ssvb>
because cmake sucks?
<ssvb>
:)
<wingrime>
ssvb: on a10 automake take time more than build takes
<ssvb>
yes, that's true
<wingrime>
ssvb: also plese make prefix "/usr" by default without directly setting it
<wingrime>
I once builded it without --prefix and get notworkable install
<ssvb>
that's how autotools normally work, people are used to this behavior
<hglm>
When not properly rerunning autoreconf/configure when upgrading sunxifb you can also get crashes or other unexpected behaviour, which make sense I guess because the Xorg loadable modules does little error checking.
<ssvb>
hglm: hmm, is this really so? never seen such problems
<ssvb>
hglm: you can get a missing symbol like wingrime reported, but that's because the new source file was just not compiled in
<hglm>
ssvb: Missing symbols were not flagged when compiling sunxifb, so the resulting X server crashed when running the module. Probably because of a missing new source file, yes.
<hglm>
ssvb: But it is related to the Xorg build-flags being extremely permissive and not catching errors.
<wingrime>
ssvb: that must be and compilation step, not when a try start xorg after reboot
_BJFreeman has joined #linux-sunxi
_BJFreeman is now known as BJfreeman
<ssvb>
wingrime: we are building a shared library, this library consists of a bunch of functions, gcc does not know that one of the required functions is missing
<ssvb>
wingrime: maybe something can be really improved, but that's how xorg modules are built at the moment
<hglm>
I suppose it should be possible in principle to include visibility=hidden in all local definitions, but traditionally it seems all symbols are exported.
<ssvb>
from that link: "if you were to mistype the function above into priltf – a symbol that does not exist, at least in the basic C library – the compiler will refuse to create an executable at all, even if the implicit declaration was only treated as a warning, because the symbol is, well, not defined. But this rule only applies, by default, to final executables, not to shared objects (shared libraries, dynamic libraries, .so, .dll or .dylib files)"
<ssvb>
the joy of developing shared libraries
<ssvb>
but patches are welcome if they improve the situation
<hglm>
ssvb: Maybe some cflags relating to requiring prototypes would help. The default Xorg cflags are very permissive in this area.
<wingrime>
ssvb: nice XV working(mostly)
<wingrime>
ssvb: I get once freze, but have no idea what the reason
<wingrime>
ssvb: also you support only yvy420p according mplayer
<wingrime>
ssvb: mplayer force yuv422p -> yuv420p conversion
<ykchavan>
600MB is big download for my internet speed
<ssvb>
wingrime: what are you decoding with mplayer?
<ykchavan>
I would like to know in short - what I will be able to do after installing this sdk?
<wingrime>
ssvb: simple mjpeg file
<ssvb>
wingrime: aha, thanks, I have not added yuv422 just because I did not have any media files which could be useful for testing :)
<ssvb>
wingrime: now I'll try to find some mjpegs
<ssvb>
wingrime: about freezes, I suspect that there is some problem in the kernel, I could also deadlock the system randomly when playing with these ioctls
<hglm>
ykchavan: What you can also do is follow the debootstrap method with a Debian install already running on your device. I did it to create a Debian customized unstable rootfs from Wheezy.
<wingrime>
ssvb: can cedar be used with xv or we need special videout?
<wingrime>
ssvb: all about that cedar have no any deal with mmu, but , out frame buffer not so bih
<ykchavan>
hglm, I have not created debootstrap earlier. I will go through steps and will come back if stuck anywhere.
<hglm>
ykchavan: I followed the "Debian from Debian guide". Just make sure you install proper networking connectivity before rebooting from the new rootfs.
<ykchavan>
ok
<wingrime>
ssvb: we can't make zero-copy decoding atleast we have memcopy source stream to reserved memory
<ssvb>
wingrime: only if somehow wrapping shm into ump and using mali
<wingrime>
ssvb: dma also can't be used for help as it too have no any deal with mmu (may be we can use transfers one page sized?)
<ssvb>
wingrime: that's how xv works, it was designed for video decoding done on cpu
<ssvb>
wingrime: moreover, there may be one more memcpy on the client side
<ssvb>
wingrime: mplayer has '-dr' option for avoiding this, but it does not work for all codecs
<hglm>
pacopad: I guess it's work-in-progress with not all devices supported. The more complete kernels right now for Cubieboard2 seem to be the cubie 3.3-based kernels.
<pacopad>
I know, but i need cedarx support and i didn't find it on 3.3
<hglm>
pacopad: I doubt cedarx would work on A20 though out-of-the-box. Probably different version of cedarx in A20.
<wingrime>
hglm: it need new blob as cedar blob check SoC version
<hglm>
pacopad: You may need to disable some drivers/stuff in the 3.4 wip kernel for A20...maybe there are some release notes/instructions you can find
<pacopad>
ok i'll try on that way
<pacopad>
thank you
<ssvb>
rm: at least it is easy to revert and I can continue testing stage/sunxi-3.4
wingrime has quit [Ping timeout: 240 seconds]
tinti has quit [Ping timeout: 248 seconds]
pacopad has quit [Quit: pacopad]
tinti has joined #linux-sunxi
hglm has quit [Quit: leaving]
<paulk-desktop>
it seems like my issue is because of the lack of uevent. Any idea of the Kconfig option to enable that?
tzafrir has quit [Ping timeout: 246 seconds]
<paulk-desktop>
perhaps CONFIG_DM_UEVENT ?
BJfreeman has quit [Quit: had a good time]
tinti has quit [Read error: Connection reset by peer]
zeRez has quit [Quit: zeRez]
Dreadlish is now known as DrOetker
DrOetker is now known as Dreadlish
<ssvb>
mnemoc: ping
Dreadlish is now known as DrOetker
dragonn has quit [Ping timeout: 276 seconds]
tinti has joined #linux-sunxi
tinti has quit [Remote host closed the connection]
hramrach__ has quit [Ping timeout: 240 seconds]
paulk-desktop has quit [Quit: Ex-Chat]
notmart has joined #linux-sunxi
notmart has joined #linux-sunxi
dlan has quit [Ping timeout: 240 seconds]
ykchavan has quit [Ping timeout: 240 seconds]
lkcl has quit [Ping timeout: 240 seconds]
PiyushVerma has quit [Ping timeout: 240 seconds]
dlan has joined #linux-sunxi
lkcl has joined #linux-sunxi
hramrach__ has joined #linux-sunxi
ykchavan has joined #linux-sunxi
rellla has quit [Remote host closed the connection]
hramrach__ has quit [Ping timeout: 240 seconds]
Offshore has quit [Ping timeout: 246 seconds]
grevaillot has quit [Ping timeout: 240 seconds]
DrOetker has quit [Ping timeout: 256 seconds]
zoobab_ has quit [Ping timeout: 246 seconds]
mripard_ has quit [Ping timeout: 248 seconds]
notmart has quit [Quit: notmart terminated!]
<Turl>
ssvb: do you have emac_power set to something on your script.bin?