Using the current state of the page directory registers, we can
determine which of our address spaces was active when the hang occurred.
This allows us to scan through all the address spaces to identify the
"active" one during error capture.
v2: Rebased for BDW error detection. BDW error detection is similar
except instead of PP_DIR_BASE, we can use the PDP registers.
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
[danvet: Add FIXME about global gtt misuse.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
The existing check was insufficient to determine whether we can use the
GTT mapping to read out the object during error capture.
The previous condition was, if the object has a GGTT mapping, and the
reloc is in the GTT range... the can happen with opjects mapped into
multiple vms (one of which being the GTT).
There are two solutions to this problem:
1. This patch, which avoid reading the io mapping
2. Use the GGTT offset with the io mapping.
Since error capture is about recording the most accurate possible error
state, and the error was caused by the object not in the GGTT - I opted
for the former.
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
This came from a patch called, "drm/i915: Move active to vma"
When moving an object to the inactive list, we do it for all VMs for
which the object is bound.
The primary difference from that patch is this time around we don't not
track 'active' per vma, but rather by object. Therefore, we only need
one unref.
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
This was found by code inspection. If the GTT setup fails then we are
left without properly tearing down the drm_mm.
Hopefully this never happens.
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
To be able to effectively use the GGTT object lookup function, we don't
want to warn when there is no GGTT mapping. Let the caller deal with it
instead.
Originally, I had intended to have this behavior, and has not
introduced the WARN. It was introduced during review with the addition
of the follow commit
commit 5c2abbeab7
Author: Ben Widawsky <benjamin.widawsky@intel.com>
Date: Tue Sep 24 09:57:57 2013 -0700
drm/i915: Provide a cheap ggtt vma lookup
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Since the beginning, the functions which try to properly reference the
aliasing PPGTT have deferences a potentially null aliasing_ppgtt member.
Since the accessors are meant to be global, this will not do.
Introduced originally in:
commit a70a3148b0
Author: Ben Widawsky <ben@bwidawsk.net>
Date: Wed Jul 31 16:59:56 2013 -0700
drm/i915: Make proper functions for VMs
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
The initial implementation of this function used MMIO to write the PDPs.
Upon review it was determined (correctly) that the docs say to use LRI.
The issue is there are times where we want to do a synchronous write
(GPU reset).
I've tested this, and it works. I've verified with as many people as
possible that it should work.
This should fix the failing reset problems.
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.15 (GNU/Linux)
iQEcBAABAgAGBQJSogqUAAoJEHm+PkMAQRiGM2MIAJrr5KEXEWuuAR4+JkkWBK7A
+dVT4n1MM4wP/aCIyriSlq7kgT03Wxk4Q4wKsj2wZvDQkNgEQjrctgIihc75jqi5
126nmT3YXJLwgDpFA3RHZUWve3j3vfUG53rRuk7K9Xx1sGWU3Ls7BuInvQZ//+QS
6UB4UuEAalmose5U8ToXQfMqZhjwreZKeb64TEZwFvu2klv4cnka1L/zHbmQGgRg
2Pfv+aUrjsYE8s9lkEKX8MIQsDn28Q5Lsv7XIEQwo2at4rYbJaxX6usuC1OI0MQ5
BLUn1GgtvOidq6FzSg6kXiA/MJYH3J0S+p4uULWAprxA+KeJRbWNRroM94W1qAk=
=1Wcq
-----END PGP SIGNATURE-----
Merge tag 'v3.13-rc3' into drm-intel-next-queued
Linux 3.13-rc3
I need a backmerge for two reasons:
- For merging the ppgtt patches from Ben I need to pull in the bdw
support.
- We now have duplicated calls to intel_uncore_forcewake_reset in the
setup code to due 2 different patches merged into -next and 3.13.
The conflict is silen so I need the merge to be able to apply
Deepak's fixup patch.
Conflicts:
drivers/gpu/drm/i915/intel_display.c
Trivial conflict, it doesn't even show up in the merge diff.
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Currently, PC8 is enabled at modeset_global_resources, which is called
after intel_modeset_update_state. Due to this, there's a small race
condition on the case where we start enabling PC8, then do a modeset
while PC8 is still being enabled. The racing condition triggers a WARN
because intel_modeset_update_state will mark the CRTC as enabled, then
the thread that's still enabling PC8 might look at the data structure
and think that PC8 is being enabled while a pipe is enabled. Despite
the WARN, this is not really a bug since we'll wait for the
PC8-enabling thread to finish when we call modeset_global_resources.
The spec says the CRTC cannot be enabled when we disable LCPLL, so we
had a check for crtc->base.enabled. If we change to crtc->active we
will still prevent disabling LCPLL while the CRTC is enabled, and we
will also prevent the WARN above.
This is a replacement for the previous patch named
"drm/i915: get/put PC8 when we get/put a CRTC"
Testcase: igt/pm_pc8/modeset-lpsp-stress-no-wait
Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
If we force the hw to idle as our first step during unload, we can abort
the unload upon failure. Later we can probe whether the hardware remain
active even after we try to shut it down.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@gmail.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@gmail.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Doing it early prevents moving and relocating objects in vain
for contexts that won't get any GPU time.
Reported-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Mika Kuoppala <mika.kuoppala@intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
It is useful to assert that if the object is bound, then it must have
its pages pinned to prevent the shrinker from reaping its backing store.
This is even more useful with the introduction of real-ppgtt whereupon
we may have the object bound into several vma, with each instance
pinning the backing store. This assertion breaks down during unbind
where we unpinned the backing store before decoupling the vma binding.
This can be fixed with a trivial reording of the unbind sequence, which
reinforces the
pin pages
bind to vma
...
unbind from vma
unpin pages
concept.
v2: Bonus comment
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Ben Widawsky <ben@bwidawsk.net>
Reviewed-by: Ben Widawsky <ben@bwidawsk.net>
Tested-by: Ben Widawsky <ben@bwidawsk.net>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
On VLV, FIFO will be shared by both SW and HW. So, we read the
free entries through register and update dev_priv variable
and wait for only 20 entries to be free
From Deepak's follow-up mail explaining why vlv is special:
"On SB, Out of 64 FIFO Entries, 20 Entries will be used by HW and
remaining 44 will be used by the SW,. I think due to this reason, we
have a threshold of 20 Entries."
"On VLV, HW and SW can access all 64 fifo entries, I don't think
having a threshold of 20 Entries is mandatory on VLV. Also, since both
SW and HW can access all 64 Entries. I think on VLV, we need to update
the fifo_count before waiting for the FIFO."
v2: Apply mask when we read the number of free FIFO entries (Ville).
v3: Mask applied after reading the register (Deepak).
Signed-off-by: Deepak S <deepak.s@intel.com>
[danvet: Add further explanation from Deepak to commit message.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Only plane A is FBC capable on gen2 (like gen3), but the panel fitter
is hooked up to pipe B, so we want to prefer pipe B + plane A.
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
[danvet: Add the code comment Chris requested in his review.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Initialize the FBC vfuncs on gen2 and gen3 chipsets. Also make
a clean split for gen7+ vs. gen5+ vfunc initialization.
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
On gen2 and gen3 chipsets FBC is supported only on plane A. Fix (and
simplify) the plane checks in intel_update_fbc() accordingly.
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Chris Wilons <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Add a REG_WRITE_FOOTER macro as a counterpart to the REG_WRITE_HEADER.
The current code has the spin_lock() in the HEADER, but the
spin_unlock() is open coded, which looks rather confusing on the first
glance. A bit of additional symmetry might help.
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
When inspecting reports that boot/suspend/resume times are unusual it
would be useful to clearly identify the time we must spend waiting for
the hardware to complete its task. In this case we have a notification
before we start waiting for the panel to change state, but none
afterwards - which would be useful.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Checkpatch tells me
WARNING: __packed is preferred over __attribute__((packed))
so switch over to __packed across the driver before adding new packed
structs.
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@gmail.com>
Reviewed-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Check that the N and P dividers don't cause a divide by zero.
This shouldn't happen under normal circumstances, but can
happen eg. under simulation.
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
It's a pain for two reasons:
- The vga plane redisablign requires actual legacy vgao i/o to pull
of. The hw engineers really botched this one here :(
- There seem to be some BIOS out there which send out lid events when
unplugging. Together with our broken DP code, which disables the
port when the cable is lost, this results in an immediate modeset
call, which can hang on the wait for outstanding flips.
- Also we don't want to force a modeset on machines where it's not
really needed, see the referenced bug.
We might want to extend this in general to also all machines that
support opregion, since there the BIOS supposedly should manage the
gfx hardware more cooperatively.
v2: Pimp commit message a bit.
Cc: Roland Dreier <roland@kernel.org>
References: https://bugs.freedesktop.org/show_bug.cgi?id=65486
Acked-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
We're currently misprinting the port name when vlv_wait_port_ready()
times out. Fix it by using port_name().
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
wow no idea how I got this far without seeing this,
leaking the entries in the list makes kmalloc-64 slab grow.
References: https://bugzilla.kernel.org/show_bug.cgi?id=65121
Cc: stable@vger.kernel.org
Reported-by: Matthew Stapleton <matthew4196@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
We had some mode_valid() vfuncs returning an int, others the enum. Let's
use the latter everywhere.
Signed-off-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Atm we call intel_display_power_enabled() from
i915_capture_error_state() in IRQ context and then take a mutex. To fix
this add a new intel_display_power_enabled_sw() which returns the domain
state based on software tracking as opposed to reading the actual HW
state.
Since we use domain_use_count for this without locking on the reader
side make sure we increase the counter only after enabling all required
power wells and decrease it before disabling any of these power wells.
Regression introduced in
commit 1b02383464b4a915627ef3b8fd0ad7f07168c54c
Author: Imre Deak <imre.deak@intel.com>
Date: Tue Sep 24 16:17:09 2013 +0300
drm/i915: support for multiple power wells
Note that atm we depend on the value returned by
intel_display_power_enabled_sw() in i915_capture_error_state() to avoid
unclaimed register access reports. This was never guaranteed though,
since another thread can disable the power concurrently. If this is a
problem we need another explicit way to disable the reporting during
error captures.
v2:
- remove barriers as the caller can't depend on the value
returned from i915_capture_error_state_sw() anyway (Ville)
- dump the state of pipe/transcoder power domain state (Daniel)
Reported-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Imre Deak <imre.deak@intel.com>
Reviewed-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
This should just be a debug. Add another debug msg to the inherit path
while we're at it.
Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Reviewed-by: Damien Lespiau <damien.lespiau@intel.com>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=72098
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Setting this bit restores all ring contexts in parallel rather than
serially. Matches current BWG recommendations.
Tested-by: "Meng, Mengmeng" <mengmeng.meng@intel.com>
Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Reviewed-by: Deepak S <deepak.s@inel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
We use timeout mode, and we need to lower the timeout to get good RC6
residency when loads are running. This gets me from 0% residency during
glxgears to 77%, which is a pretty good improvement. This value also
matches the current BWG recommentations.
Tested-by: "Meng, Mengmeng" <mengmeng.meng@intel.com>
Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Reviewed-by: Deepak S <deepak.s@inel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
It leads to a big mess when stuff interleaves. Especially with the new
patch I've submitted for the drm core to no longer artificially split
up debug messages.
v2: The size parameter to snprintf includes the terminating 0, but the
return value does not. Adjust the logic accordingly. Spotted by Mika.
Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
It doesn't like that we assign 0 to a pointer, it wants the real NULL.
On closer look that initialization is actually bogus, and the compiler
can easily see that we never use it unitialized. So let's just drop
this.
Cc: Deepak S <deepak.s@intel.com>
Cc: Jesse Barnes <jbarnes@virtuousgeek.org>
Reported-by: kbuild test robot <fengguang.wu@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
This was fumbled in the conversion to per-engine forcewake.
Cc: Deepak S <deepak.s@intel.com>
Cc: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
On VLV the GTFIFOCTL register has other bits besides the number of free
entries in the GT wake FIFO. Apply a mask when we read th register to
make sure we don't misinterpret the number of free FIFO entries.
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Jesse Barnes <jbarnes@virtuousgeek.org>
[danvet: There's some unclarity about hsw, but brushed off as todays'
Bspec just acting up a bit.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
On VLV GTFIFODBG has more bits. Just report them all.
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Forcewake counts for valleyview are not exposed throgh DebugFS.
Exposing with this change.
Signed-off-by: Deepak S <deepak.s@intel.com>
Reviewed-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Split vlv force wake routines to help individually control Media/Render
well based on the register access.
We've seen power savings in the lower sub-1W range on workloads that
only need on of the power wells, e.g. glbenchmark, media playback
Note: The same split isn't there for the forcewake queue, only the
forcwake domains are split.
Signed-off-by: Deepak S <deepak.s@intel.com>
Reviewed-by: Jesse Barnes <jbarnes@virtuousgeek.org>
[danvet: Rebase on top of the removed forcewake hack in the ring irq
get/put code and add a note to add Deepak's answer to Chris question.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Added power well arguments to all the force wake routines
to help us individually control power well based on the
scenario.
Signed-off-by: Deepak S <deepak.s@intel.com>
Reviewed-by: Jesse Barnes <jbarnes@virtuousgeek.org>
[danvet: Resolve conflict with the removed forcewake hack and drop one
spurious hunk Jesse noticed.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Reported-by: Jim Davis <jim.epost@gmail.com>
Tested-by: Jim Davis <jim.epost@gmail.com>
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Signed-off-by: Dave Airlie <airlied@redhat.com>
Copy/Paste typo.. we need to test for ->kdev instead of ->dev.
Reported-by: Juha Leppänen <juha_efku@dnainternet.net>
Signed-off-by: David Herrmann <dh.herrmann@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
Due to user fudging (for instance using video=VGA-1:e with FBDEV=n) we can
attempt to reset an inconsistent CRTC that is marked as active but has
no assigned fb. It would be wise to fix this earlier, but the long
term plan is to have primary and secondary planes associated with a
CRTC, in which crtc->fb being NULL will be expected. So for a quick
short term fix with pretensions of grandeur, just check for a NULL fb
during GPU reset and ignore the plane restoration.
This fixes a potential hard hang (a panic in the panic handler)
following a GPU hang.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
[danvet: Add a corresponding fixme comment.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Add a debugfs entry showing the use-count for all power domains of each
power well.
v3: address comments from Paulo:
- simplify power_domain_str() by using a switch table
- move power_well::domain_count to power_domains
- WARN_ON decrementing a 0 refcount
Signed-off-by: Imre Deak <imre.deak@intel.com>
Reviewed-by: Paulo Zanoni <paulo.zanoni@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
So far we distinguished platforms without a dynamic power well with
the HAS_POWER_WELL macro and for such platforms we didn't call any power
domain functions. Instead of doing this check we can add an always-on
power well for these platforms and call the power domain functions
unconditionally. For always-on power wells we only increase/decrease
their refcounts, otherwise they are nop.
This makes high level driver code more readable and as a bonus provides
some idea of the current power domains state for all platforms (once
the relevant debugfs entry is added).
v3: rename intel_power_wells to i9xx_always_on_power_well (Paulo)
Signed-off-by: Imre Deak <imre.deak@intel.com>
Reviewed-by: Paulo Zanoni <paulo.zanoni@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Imre Deak <imre.deak@intel.com>
Reviewed-by: Paulo Zanoni <paulo.zanoni@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
This may need work if other platforms do the same thing, but in the
meantime we should avoid looking at HSW specific bits in this generic
function.
Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Reviewed-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
[added IS_BROADWELL too as that needs the same handling (Imre)]
Signed-off-by: Imre Deak <imre.deak@intel.com>
[danvet: Add Imre's missing sob.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
In intel_display_capture_error_state we use HAS_POWER_WELL to check if
we are running on Haswell/Broadwell when accessing HSW_PWR_WELL_DRIVER
which is specific to these platforms. Future platforms with power wells
don't have this register, so HAS_POWER_WELL won't work there any more.
Use IS_HASWELL/IS_BROADWELL instead.
v3: fix using logical || instead of bitwise | (Paulo)
Signed-off-by: Imre Deak <imre.deak@intel.com>
Reviewed-by: Paulo Zanoni <paulo.zanoni@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Instead of using a separate function to check whether a power domain is
is always on, add an always-on power well covering all these power
domains and do the usual get/put on these unconditionally. Since we
don't assign a .set handler for these the get/put won't have any effect
besides the adjusted refcount.
This makes the code more readable and provides debug info also on the
use of always-on power wells (once the relevant debugfs entry is added.)
v3: make is_always_on to be bool instead of a bit field (Paulo)
Signed-off-by: Imre Deak <imre.deak@intel.com>
Reviewed-by: Paulo Zanoni <paulo.zanoni@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
HW generations so far had only one always-on power well and optionally
one dynamic power well. Upcoming HW gens may have multiple dynamic power
wells, so add some infrastructure to support them.
The idea is to keep the existing power domain API used by the rest of
the driver and create a mapping between these power domains and the
underlying power wells. This mapping can differ from one HW to another
but high level driver code doesn't need to know about this. Through the
existing get/put API it would just ask for a given power domain and the
power domain framework would make sure the relevant power wells get
enabled in the right order.
Signed-off-by: Imre Deak <imre.deak@intel.com>
Reviewed-by: Paulo Zanoni <paulo.zanoni@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
