IRC log of #cubox of Fri 13 Dec 2013. All times are in CET < Back to index

10:58 jnettle 10:58 * jnettlet found the performance problems with 3.10 after not looking for a day or two. 3.10 has cpu-freq scaling working and it was broken.
10:59 jnettlet :-\
11:03 Wizzup Silly question, is some support for cuboxes going in mainline?
11:03 Wizzup Silly as in, I haven't been active here for a bit
11:04 jnettlet Wizzup, the original cubox?
11:04 Wizzup jnettlet: both, preferrably
11:05 Wizzup Although I fear for the cubox-i, since it's freescal
11:05 Wizzup e
11:05 Wizzup (or are both?)
11:05 jnettlet Wizzup, _rmk_ has device-tree support for Carrier-1 that should hopefully be in for 3.14
11:06 jnettlet that will allow basic boot functionality and KMS but not graphics acceleration or VPU.
11:06 Wizzup Cool.
11:06 Wizzup Ah, I don't need vpu/graphics
11:06 jnettlet I believe the same support is already in mainline for the original Cubox
11:06 Wizzup just sata,usb,ethernet
11:06 Wizzup Awesome.
11:06 Wizzup Thanks!
11:06 jnettlet except for KMS that should be in the next release
11:07 _rmk_ KMS on Dove Cubox... the code is there but there's no DT support for it
11:09 Wizzup Now I'll definitely pick up Gentoo on (orig) cubox again.
11:09 jnettlet we should probably release a u-boot update that has a new device-tree
11:09 Wizzup The wiki page seems lonely without me
11:10 _rmk_ it needs my componentised device thing which I've sorted out for imx-drm to have dt support working there
11:11 jnettlet right
11:12 jnettlet _rmk_, I have all your relevant patches ported to the V4 Vivante codebase now.
11:59 hste_ jnettlet: did u fix or just find the broken cpu-freq scaling or just disable it?
12:23 jnettlet hste_, well that is a good question. So it was defaulting to interactive which would load, but then didn't scale the frequency. That is probably due to me running things from an ssh session, but I haven't looked at that.
12:24 jnettlet Then when I selected either performance, or ondemand random crashes would happen on boot. I fixed that problem...pretty trivial
12:25 jnettlet but then I also had to go in because the stupid Ubuntu image has both debian cpufreq init script as well as it's ondemand initscript enabled. Ubuntu's ondemand initscript just looks at the list of governors provided and chooses either interactive or ondemand depending what it finds first. Well it was then finding interactive and resetting my default....grrrrr
12:26 jnettlet But with a kernel fix, and some initscript changes I have the ondemand scaling working properly
12:36 hste_ :) So it was a mixed rootfs problem also
12:37 jnettlet well the kernel was completely broken
12:38 jnettlet do you have 3.10 running on anything you could check on your rootfs.
12:38 hste_ nope only 4.1.0
12:39 hste_ there isn't uboot supporting dts for gk802
12:39 jnettlet now performance is fixed for graphics but I am getting corruption using the 3d core
12:39 jnettlet hste_, oh you can just append the dtb to the kernel image for testing.
12:40 hste_ a corruption only on 3.10 ?
12:41 jnettlet yep
12:41 jnettlet could be a change in the dri interface. Need to check that now.
12:42 jnettlet and I broke glesv2. It went from running really well to not running. Never a dull moment.
12:42 _rmk_ :)
14:57 MikeSeth snow in Tel Aviv
14:57 MikeSeth expect Israeli post office to screw you
14:57 MikeSeth (those two statements are unrelated)
15:30 xraxor MikeSeth: really? shit
15:31 xraxor so no hope to receive my i4Pro till next year? this sucks
17:05 nikotime http://www.haaretz.com/news/national/.premium-1.563057 on the tel aviv weather (sorry if already been covered - continuing from the log of this irc). sounds pretty nasty
18:44 xraxor great news, someone have received his cubox-i today :D
18:44 xraxor its on the forum
18:52 jnettlet how does that help me?
18:52 jnettlet :-) just kidding, that is great news.
18:53 jnettlet I am just pissed at Vivante again.
18:53 _rmk_ heh
18:53 jnettlet I have reduced the latency on the driver enough that I have created more races :-(
18:53 _rmk_ why does "vivante" and "races" always seem to go together :p
18:54 jnettlet actually I think _rmk_'s patchset is partially to blame
18:54 _rmk_ that depends what you've done to my patches :)
18:54 jnettlet they are working perfectly. All the event handling is seamless.
18:55 jnettlet and don't worry you will get to review them soon enough.
18:55 jnettlet This has to do with the power handling. A race between the GPU becoming idle and the commit atom decrementing it's lock.
18:55 _rmk_ I think this evening I'll make all the fencing optional (defaulting to off) in my xorg driver
18:56 jnettlet we are processing events so efficiently now that they are processed before the code can exit the commit :-P
18:57 jnettlet processed by the hardware that is.
18:58 jnettlet so poor power management is angry because it said we should be able to suspend but there are still commit atoms ready to go.
19:05 _rmk_ there we go, one xorg driver with batching removed
19:07 _rmk_ fastest to connect and then disconnect? :)
19:07 jnettlet as long as userspace can keep up with the commits
19:07 jnettlet the irc quickdraw
19:07 _rmk_ that doesn't matter... since commit with stall=true should now properly wait for the previously submitted operations to complete before returning
19:08 jnettlet true.
19:09 jnettlet I am thinking if it would help performance. Especially since Xorg loves lots of small little copies and fills.
19:10 _rmk_ well, it means once less GPU operation per batch
19:10 _rmk_ even though it is a single pixel fill :)
19:13 jnettlet If the numbers I have found posted for the gc2000 don't lie, then I have the gc880 running at over 1/3 the performance the gc2000 is doing. which if the hardware is supposed to be 1/4 as fast it is a pretty decent bump
19:14 jnettlet let me amend that....we have the gc880
19:14 jnettlet :-)
19:15 jnettlet can't wait to test this on the Marvell hardware
19:15 jnettle 19:15 * jnettlet has to go install a new faucet now.
19:15 jnettlet ttyl
19:15 _rmk_ that's the 3d engine, the 2d engine appears to be gc320 if the hex number can be believed