IRC log of #cubox of Fri 23 Nov 2012. All times are in CET < Back to index

00:46 ptl does anyone know where I can get the source for the kernel used in CuBox, with all patches applied and .config set?
00:55 _rmk 00:55 * _rmk_ chucks out the sdhci patches
01:05 _rmk_ I think that's about it for the v3.8 merge window... I guess the next easy one should be the hwmon stuff, but that needs sorting properly
01:05 _rmk_ that will leave vmeta and sorting the video driver one way or another, and galcore
01:08 _rmk_ once 3.7 is out, I'll rebase my copies of these patches onto v3.7 and make that available (which should be the same as folk have queued for the merge window)
01:08 _rmk_ the difference will be, mine will be directly on top of v3.7, rather than some other point
02:03 ptl does anyone know where I can get the source for the kernel used in CuBox, with all patches applied and .config set?
07:37 jnettlet _rmk_, I am working on getting galcore 4.6.4 and suite released for use right now...it is pending Legal okay at Marvell.
07:37 jnettlet of course userland will still be binary HAL, and EGL/GLESv2
10:00 shesselba _rmk_: You want me to add any acked-by or tested-by to the sdhci patches? BTW it's spelled "Hesselbarth".
10:02 _rmk 10:02 * _rmk_ wonders where that typo came from :) sorry about that
10:02 _rmk_ lets wait for Sascha to respond to my reply first
10:03 shesselba IRC, rabeeh used it once or twice.. no big deal - it's a long name ;) ok, I ll wait
10:04 _rmk_ "Code done by Sebastian Hasselbrath" - yep, that's where I got it from
10:16 rabeeh my fault?
10:16 _rmk_ the above is what's in your git tree :(
10:17 rabeeh jnettlet: do you have android GPU drivers too?
10:19 _rmk_ jnettlet: I wonder if they're going to obfuscate gc_hal_kernel_hardware.c in the same way
10:20 _rmk_ ... and still provide the macros that were used for that obfuscation
10:24 _rmk_ jnettlet: also... will it be an arhmf or armel release of userspace?
10:24 _rmk 10:24 * _rmk_ hopes both.
10:24 _rmk_ or failing that, armhf
11:17 jnettlet rabeeh, yes android as well...although I haven't explicitely asked to release them. We are dabbling in an Android release for our hardware, so I should probably do that now. It takes forever.
11:19 jnettlet _rmk_, do you mean those horrible long register writes? Yeah those are the same. Haven't looked for the macros, can check...yes I have them build everything for hardfp
11:19 jnettlet Everything is compiled for hardfp and iwmmxt
11:21 rabeeh jnettlet: there is interest in getting JellyBeans on CuBox. that would be very helpful.
11:23 jnettlet rabeeh, We only have ICS right now. Although the newest framebuffer drivers expose the vsync uevents needed for Jellybean's project butter
11:26 rabeeh oh; so probably the same binaries can run as-is on CuBox?
11:26 rabeeh have a pointer for code drop?
11:26 rabeeh using r3184 i assume
11:27 jnettlet it is using vivante's 4.6.4 with Marvell specific tweaks. I am still in Legal limbo getting the okay to release that. It is making our C1 board release very difficult
11:27 Kiranos jnettlet: what is your project?
11:27 jnettlet their's might be 4.6.2.
11:27 jnettlet Kiranos, I work for OLPC (One Laptop Per Child)
11:28 ralix damn damn, I can not drive by a usb boot;-)
11:28 ralix Sorry Good Morning @ alldamn damn, I can not drive by a usb boot;-)
11:28 ralix Sorry Good Morning @ all
11:28 ralix copy and paste error ;-)
11:29 rabeeh hi ralix.
11:30 Kiranos hope the rest of the day dosnt continue in the same vein :P
11:31 dotarray hi ralix :)
11:34 ralix I wanted to boot from a usb drive. I made it according to the instructions: "http://archlinuxarm.org/platforms/armv7/cubox"
11:34 ralix Uncompressing Linux... done, booting the kernel.
11:34 ralix m25p80 spi0.0: unrecognized JEDEC id 20ba16
11:34 ralix mmc0: Card removed during transfer!
11:34 ralix mmc0: Resetting controller.
11:34 ralix Kernel panic - not syncing: VFS: Unable to mount root fs on unknown-block(0,0)
11:34 ralix :(
11:36 Kiranos [11:34:36] mmc0: Card removed during transfer!
11:36 Kiranos that doesnt sound to good
11:36 ralix rabeeh, The arrangement of the interfaces and the serial-usb at the rear Cubox is nice as my first :)
11:36 Kiranos have you tried another sd-card?
11:37 ralix I booted completely without sdcarte, only one hard drive
11:52 jnettlet It still amazes me that I can get a nano-usb plug that provides 300Mbs 802.11
12:09 RandomPixels hello
12:09 ralix New try :) booting from mmc and root on sda1:
12:09 ralix Uncompressing Linux ... done, booting the kernel.
12:09 ralix M25P80 spi0.0: unrecognized JEDEC id 20ba16
12:09 ralix Kernel panic - not syncing: VFS: Unable to mount root fs on unknown-block (0.0)
12:09 ralix hello RandomPixels
12:09 RandomPixels E: Package git has no installation candidate
12:09 RandomPixels any ideas? :)
12:10 ralix I do not understand
12:10 RandomPixels trying to apt-get install git
12:11 Kiranos git-core
12:11 RandomPixels actually, i'm trying to set up a XBMC using rabeeh's fork
12:12 RandomPixels thanks Kiranos
12:51 ralix f*ck
13:00 ralix I'm too stupid ;-)
14:03 RandomPixels trying geexbox :)
14:11 ralix no ;-)
14:12 ralix I like archlinux ... geexbox is quite nice, I had already tested.
15:15 rabeeh I did send Rabeeh a few patches which sort out the memory grabbing for vmeta/bmm/gpu stuff but I don't think he ever merged them
15:15 rabeeh rmk_: the ioctl callback to the vmeta driver somehow didn't reach to the point where it really shuts down the power of vmeta when unused
15:15 rabeeh that hack was somehow to force it to do that.
15:16 rabeeh (ugly unclean hack).
15:16 rabeeh to explain that, Dove has few power domans -
15:16 rabeeh to explain that, Dove has few power domains -
15:16 rabeeh oh; rmk_ is not online. anyhow i will continue the speech and point him out on the logs
15:16 rabeeh 1. cpu domain
15:16 rabeeh 2. gpu domain
15:16 rabeeh 3. vmeta domain
15:17 rabeeh 4. peripherals (or sometimes called core)
15:17 rabeeh 5. rtc
15:17 rabeeh for the gpu/vmeta you can control the clock (on/off) and power (on/off). Somehow the ioctls back from the vmeta user space driver didn't reach a point where vmeta power down is reached
15:18 rabeeh so constant power was drawn.
15:49 ralix rabeeh, Are you working yet on Cubox-installer, so that one can not only install on mmc but also on usb / sata?
15:50 rabeeh ralix: it's in the queue
15:51 rabeeh install from hdmi is first
15:51 rabeeh patches welcomed :)
15:51 ralix ok
15:51 rabeeh https://github.com/rabeeh/cubox-installer
15:51 ralix I have difficulty reading archlinux install a usb hda, and I just wanted to get the sources ;)
15:52 ralix I've already found;-) , I'll checkout when I'm at home .
15:52 rabeeh i think the changes would be root=/dev/sda1 or something like that
15:52 rabeeh maybe rootdelay=5 is also needed to wait for usb stick to connect to subsystem.
15:53 ralix I'll try, but it loads the kernel already on the hdd and then Kernel panic - not syncing: VFS: Unable to mount root fs on unknown-block (0.0)
15:53 rabeeh do you have the 'root=...' set?
15:56 ralix yes,
15:56 ralix ox systemd top usb boot scrip{echo ======== Setting bootargs ========
15:56 ralix setenv bootargs 'console=ttyS0,115200n8 console=tty1,115200n8 init=/bin/systemd root=/dev/sda1 rw vmalloc=384M video=dovefb:lcd0:1280x720-32@60-edid clcd.lcd0_enable=1 clcd.lcd1_enable=0 usb0Mode=host usb1Mode=host'
15:56 ralix echo ======== Loading kernel ========
15:56 ralix ext4load usb 0:1 0x00200000 /boot/uImage
15:56 ralix echo ======== Booting kernel ========
15:56 ralix bootm
15:59 dung ralix, try bootarg rootwait
15:59 dung http://www.solid-run.com/mw/index.php/Kernel_parameters
16:00 ralix ok i try ... 5 Minutes
16:06 ralix There it is going! YOU have my degree was real happy!
16:11 ralix Ok, I can now happily at home after.
16:12 _rmk_ umm, console=tty1 doesn't take any other arguments
16:12 _rmk_ and if you have systemd installed, it should run just as 'init', the init= argument shouldn't be required
16:13 ralix ok thx :) !
16:13 dung isn't just one console?
16:13 _rmk_ dung: you can have multiple consoles, but only the first is the primary one
16:13 _rmk_ the output is duplicated on the others you list
16:14 _rmk_ rabeeh: right, I've noticed that libvmeta doesn't get the appropriate callbacks from the closed source part for the power control
16:30 jnettlet _rmk_, do you mean the vdec_os_api_clock_on clock_off power_on power_off calls?
16:30 _rmk_ yup
16:31 jnettlet those don't get called by the UIO binary...they should be called by whatever is using libvmeta to decode the video. So the gstreamer plugin implements those calls
16:32 jnettlet or should. Just like the xbmc implementation should also implement those calls if it wants to clock things up or down.
16:33 jnettlet in most the reference code the clocking should actually be based on the resolution of the video to be decoded.
16:34 _rmk_ it looked to me like libcodecvmetadec makes the clock calls
16:34 _rmk_ and it makes some of the power calls; certainly when the codec library is initialized
16:39 _rmk_ in any case, the uio kernel folk have rejected vmeta outright due to the ioctl stuff, so either it stays out of mainline or we switch to my misc device version (which I of course have the appropriate libvmeta for)
16:39 jnettlet I actually have a test version that switches the ioctl version to using sysfs
16:39 _rmk_ why bother
16:39 jnettlet my hope was to submit it to show the UIO folks that their response was ridiculous and it has basically made uio an almost useless subsystem.
16:40 jnettlet it is most ridiculous to create something and tell vendors that have to provide binary drivers...use this, but then completely limit the functionality.
16:41 _rmk_ no; I'm on the side of the UIO folks on this
16:41 _rmk_ and quite honestly the vmeta uio driver is rather crappy
16:42 _rmk_ there's a whole bunch of things it can do better, and all that user ID management is rather broken
16:44 _rmk_ http://ftp.arm.linux.org.uk/git/gitweb.cgi?p=libvmeta.git
16:44 _rmk_ that's my version of libvmeta
16:44 _rmk_ the kernel side isn't fully debugged; there's still horrid bugs in the user id management code
16:44 _rmk_ especially with the "lets wipe out all state on force-init" thing which breaks any other concurrent user
16:46 _rmk_ as for the locking in the original kernel driver, that was total crap (which is actually putting it in a nice way)
16:46 _rmk_ and the clk API handling was just as bad
16:49 jnettlet yes but that stuff was never going to get fixed because when Marvell first tried to get the driver included they got shot down on the ioctl's
16:49 jnettlet and they had no real incentive to make it better because all the hardware was basically single user systems
16:50 _rmk_ well, strangely the ioctls are just fine as a misc device
16:50 jnettlet sure, but I thought misc devices that were only used by binaries wouldn't be accepted upstream ?
16:50 _rmk_ oh, and... btw.. the irq handling in libvmeta is broken; after the first IRQ it just returns immediately because nothing ever does the required UIO read.
16:52 jnettlet I never noticed that problem. Could be fixed in my version of the driver.
16:52 _rmk_ when I discussed with gregkh, he basically agreed that this as a miscdevice would be fine
16:52 _rmk_ that was after I went through the licensing battle with him
16:53 _rmk_ I have 52 commits against that vmeta kernel driver
16:54 jnettlet If it is getting converted to a miscdevice why not dump bmm and just do the dma directly from the vmeta driver?
16:58 _rmk_ do we want the X server opening the vmeta device? I don't think we do; it makes development against it harder because you'd have to shut the X server down to remove vmeta as a kernel module.
16:58 _rmk_ afaics bmm is better off being separate at the moment
17:08 jnettlet sorry wife called. The X server shouldn't open the vmeta device. The userspace program should open the vmeta device causing vmeta to request a dma-buf then it can generate a fd which can be passed up to the accessing program.
17:10 jnettlet the userspace program, should then be able to use dri2 to render the decoded frame on screen.
17:11 _rmk_ what you're suggesting means getting rid of the overlay completely, and involve having to have the GPU blit the image into the screen
17:12 _rmk_ I'd rather keep Xv with its overlay support
17:12 jnettlet do you have kms overlay support in your driver?
17:12 _rmk_ of course :)
17:14 _rmk_ supporting not only the common YUV formats but also the ones vmeta likes to spit out, both in BMM mode and as DRM buffer objects passed via named buffers (which my bmmxvimagesink makes use of)
17:16 _rmk_ you can run software-decoding vlc against my X server and it works; try that with the standard dove backend driver and it'll grind because vlc has inefficient format conversion
17:19 jnettlet I believe you would just need to change your bmmxvimagesink to hook directly to the kms-overlay instead of going through XV....basically allowing you to decode video via gstreamer without needing an X server running.
17:19 jnettle 17:19 * jnettlet has to run. back in an hour or so.
17:26 _rmk_ except there's a huge can of worms surrounding colourkeying, cropping and in multi-head setups which crtc to use, and with virtual X resolutions where on the screen the image should be placed...
17:40 RandomPixels hello guys
17:40 RandomPixels i ran into some trouble trying to install geexbox for cubox
17:40 RandomPixels anyone did it successfully ?
20:29 RandomPixels hello
20:30 RandomPixels anyone around ?