Date: Mon, 25 Jul 2016 09:36:41 +0000 From: Natasha Kerensikova <natbsd@instinctive.eu> To: freebsd-x11@freebsd.org Subject: 4.6 DM/i915 test report on Bay Trail (Celeron J1800) Message-ID: <20160725093641.GA98977@nat.rebma.instinctive.eu>
next in thread | raw e-mail | index | archive | help
Hello, I have already bothered you with my Celeron J1800, but I thought I would try to write a formal test report. I'm a bit late to the party, but I hope it's still somewhat relevant. If it's not but you still have some time to read me, I would welcome criticism on my report to make it more efficient on the next CFT round. Also I thouht I read some uncertainty about the code on SandyBridge, and I happen to have one ready to (show its power to) serve if I manage to update the BIOS settings on the UEFI/MBR/bootstrap issue. As I'm weary to mess with BIOS settings that work for now, is three a need for more SandyBrdige tasting? To be clear, I downloaded the image at http://www.bsddesktop.com/images/cftdisk_nodebug_2016062423.img.xz I couldn't use it directly, lacking a large enough USB stick, so I copied the zfs on my production disk and updated bootfs accordingly. I think the only relevant thing it changes is that the bootloader is still the one from 10.3-RELEASE, which might impact the console initial state. To describe the machine, the sysctl hw.model is Intel(R) Celeron(R) CPU J1800 @ 2.41GHz I first tried with SNA acceleration. Something weird happened the only time I tried `startx` without first `kldload i915kms`, with `startx` takings tens of seconds (maybe even more than a minute) to fail, and unable to change mode. With `kldload i915kms` first though X starts correctly with XFCE. I saw then some "healing" artifacts whenever a part of the screen gets redraw in block. It's as if the wrong pixels are immediately updated, but then in the next few seconds the display is stochastically refreshed, eventually forcing all pixels to their intended values. It's so fast and smooth that at first I took it for fancy WM eye-candy, before I realized it made the terminal barely usable. Some redraws seem immune from these artifacts, most notably the background picture and the 3D screensaver. I couldn't get mplayer or vlc to display anything but a black window, while sound played correctly. VLC was quite unhelpful in diagnosing the issue, while mplayer kept spitting: X11 error: BadAlloc (insufficient resources for operation) On the other hand, youtube on firefox showed nice and smooth videos but no sound. Then I tested with UXA, which mostly worked fine except outside of web browser. I watched full HD video with mplayer, but I didn't check whether the decoding was CPU or GPU based. XCFE looked fine, but after a while I noticed that the shadow of windows, instead of being translucent, was a comb of horizontal lines. While the artifact doesn't affect the usability at all, I guess it implies something wrong with alpha-compositing, which I don't I have otherwise the opportunity to test. The main window of both firefox and chromium starts fine, but after a random time becomes completely and irreversibly corrupt, exactly like mishurov's screenshot of May 31st in https://github.com/FreeBSDDesktop/freebsd-base-graphics/issues/6 During an instance with an unusually long time before the corruption, I managed to start a youtube video, which was fine and smooth but still without sound. surf-browser is not affected though, and I think I would use the image daily if it could handle pages with HTML5-sound. If there is anything more I should test, please let me know, I can still boot on the CFT fs. Hoping this meager contribution helps, Natasha
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20160725093641.GA98977>