Hi, I have here a dual radeon RV790 gpu, three screen setup. Tried this out with dpm=1 locks up the machine. The lockup happens either when X starts and lightdm starts to paint the login widget or, if the machine survives this, it usually occurs once the mouse crosses the borders of the two gpu's screens on the X display. I run fedora 19 with a recent patch level. The kernel that runs in dpm mode is 7dee8dff47734ee52da2cd8b8ea9736e42c5062f which is todays linus master tree and includes all of agd5f/drm-fixes-3.12. Attached are, dmesg with and without dpm enabled, Xorg logs with these cases, the xorg configuraton file for these three screens, and the video bioses of these two only similar cards. The dpm variants are acquired logged in via ssh on a session where lightdm did not manage to lock up the machine at the first time but locked it when I moved the mouse across the screen borders. Some observations when digging somehow into this problem: 1. The attached patch shows a minimal change to rv770_dpm_set_power_state that makes the machine run as normal. I think that already narrows down the problem a bit. 2. The pi->clk_regs.rv770.mpll_ss{1,2} values are set into the table that aims to be activated above, but are never initialized like the other values beneath on rv770. May be they need to be read initially from some gpu register? 3. Disabling aspm does not help. 4. Unplugging the second monitor on the second gpu makes the problem appear way more seldom if it stays unplugged. Hotplugging this screen results in the same reproducible lockup. Greetings Mathias
Created attachment 87153 [details] xorg.conf
Created attachment 87154 [details] dmesg with dpm enabled
Created attachment 87155 [details] dmesg without dpm enabled
Created attachment 87156 [details] Xorg.log with dpm enabled
Created attachment 87157 [details] Xorg.log without dpm enabled
Created attachment 87158 [details] Vidoe bios of one card.
Created attachment 87159 [details] Vidoe biso of the other card
Created attachment 87160 [details] [review] Patch hopefully narrowing down the problem.
-- GitLab Migration Automatic Message -- This bug has been migrated to freedesktop.org's GitLab instance and has been closed from further activity. You can subscribe and participate further through the new bug through this link to our GitLab instance: https://gitlab.freedesktop.org/drm/amd/issues/389.
Use of freedesktop.org services, including Bugzilla, is subject to our Code of Conduct. How we collect and use information is described in our Privacy Policy.