[Bug 93718] nouveau + moveablecore => endless havoc (possibly a general drm problem)

[bugzilla-daemon at freedesktop.org]

https://bugs.freedesktop.org/show_bug.cgi?id=93718

            Bug ID: 93718
           Summary: nouveau + moveablecore => endless havoc (possibly a
                    general drm problem)
           Product: DRI
           Version: XOrg git
          Hardware: Other
                OS: All
            Status: NEW
          Severity: normal
          Priority: medium
         Component: General
          Assignee: dri-devel at lists.freedesktop.org
          Reporter: rwhite at pobox.com

NOTE: I'm actually using the gentoo x11 overlay which seems to be an xorg
git...

So reserving movable core on the linux command line seems to cause random
failure in my nouveau dual monitor desktop system. (I suspect it's happening to
a lessor extent on my radeon laptop as well).

So anyway, I've been using "moveablecore=2G" on my kernel command line for some
time in order to accommodate hugepages (etc) for transient virtual machines.
When I started playing with SDDM and kde Plasma I started getting display
hangs.

The exact moment and error of the hang/error is hard to catch, but just
starting sddm is enough to cause my dmesg to fill with faults. They all looked
memory/state related so I started playing around.

With no kernelcore= or moveablecore= options on the kernel command line the
system seems rock steady.

Using either (they both reserve a movable memory NUMA region that anonymous
mappings via mmap() "like" and the kernel is free to relocate by juggling page
tables and copying physical images) seems to steal data will-he/nil-he out from
under nouveau.

ASIDE: I say it may be happening on my radeon laptop because Chromium gets
render-hung there for odd reasons but the diagnostics are more vague.

So the work-around is to not use these options, but I suspect there's some
missing page locking or whatever creating phantom failures... particularly in
nouveau... particularly under high render pressure.

With "moveablecore=2G kernelcore=3G" I can't even get all the way through the
sddm password entry without a messy crash.

This seems to line up with a lot of can-not-duplicate type of error reports I
found with google, so I figure'd I drop my datapoint here.

-- 
You are receiving this mail because:
You are the assignee for the bug.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.freedesktop.org/archives/dri-devel/attachments/20160114/7f2e9ca4/attachment.html>