The vm callbacks are the same as the SI ones right now
(same regs and bits). We could share the SI variants, and
I may yet do that, but I figured I would add CIK specific
ones for now in case we need to change anything.
V2: add documentation, minor fixes.
V3: integrate vram offset fixes for APUs
V4: enable 2 level VM PTs
V5: index SH_MEM_* regs properly
V6: add ib_parse()
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
v2: split soft reset into compute and gfx. Still need
to make reset more fine grained, but this should be a
start.
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
v2: tiling fixes
v3: more tiling fixes
v4: more tiling fixes
v5: additional register init
v6: rebase
v7: fix gb_addr_config for KV/KB
v8: drop wip KV bits for now, add missing config reg
v9: fix cu count on Bonaire
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
New asics support non-privileged IBs. This allows us
to skip IB checking in the driver since the hardware
will check the command buffers for us. When using
non-privileged IBs, if the CP encounters an illegal
register in the command stream, it will halt and generate
an interrupt. The CP needs to be reset to continue. For now
just do a full GPU reset when this happens.
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Drivers are allowed (actually have to) disable unrelated crtcs in
their ->set_config callback (when we steal all the connectors from
that crtc). If they do that they'll clear crtc->fb to NULL.
Which results in a refcount leak, since the drm core is keeping track
of that reference.
To fix this track the old fb of all crtcs and adjust references for
all of them. Of course, since we only hold an additional reference for
the fb for the current crtc we need to increase refcounts before we
drop the old one.
This approach has the benefit that it inches us a bit closer to an
atomic modeset world, where we want to update the config of all crtcs
in one step.
This regression has been introduce in the framebuffer refcount
conversion, specifically in
commit b0d1232589
Author: Daniel Vetter <daniel.vetter@ffwll.ch>
Date: Tue Dec 11 01:07:12 2012 +0100
drm: refcounting for crtc framebuffers
Reported-by: Russell King <linux@arm.linux.org.uk>
Cc: Russell King <linux@arm.linux.org.uk>
Cc: stable@vger.kernel.org
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Dave Airlie <airlied@redhat.com>
Historically drm lacked fb refcounting, so the updating of crtc->fb
was done by the lower levels at a point convenient to get their own
refcounting (e.g. refcounts for the underlying gem bo, pinning
refcounts) right. With the introduction of refcounted fbs the drm core
handled the fb refcounts, but still relied on drivers to update the
crtc->fb pointer (this approach required the least invasive changes in
drivers).
Enforce this contract with a WARN_ON.
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Dave Airlie <airlied@redhat.com>
Atm the crtc helper implementation of set_config has really
inconsisten semantics: If just an fb update is good enough, dpms state
will be left as-is, but if we do a full modeset we force everything to
dpms on.
This change has already been applied to the i915 modeset code in
commit e3de42b684
Author: Imre Deak <imre.deak@intel.com>
Date: Fri May 3 19:44:07 2013 +0200
drm/i915: force full modeset if the connector is in DPMS OFF mode
which according to Greg KH seems to aim for a new record in most
Bugzilla: links in a commit message.
The history of this dpms forcing is pretty interesting. This patch
here is an almost-revert of
commit 811aaa55ba
Author: Keith Packard <keithp@keithp.com>
Date: Thu Feb 3 16:57:28 2011 -0800
drm: Only set DPMS ON when actually configuring a mode
which fixed the bug of trying to dpms on disabled outputs, but
introduced the new discrepancy between an fb update only and full
modesets. The actual introduction of this goes back to
commit bf9dc102e2
Author: Keith Packard <keithp@keithp.com>
Date: Fri Nov 26 10:45:58 2010 -0800
drm: Set connector DPMS status to ON in drm_crtc_helper_set_config
And if you'd dig around in the i915 driver code there's even more fun
around forcing dpms on and losing our heads and temper of the
resulting inconsistencies. Especially the DP re-training code had tons
of funny stuff in it.
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Dave Airlie <airlied@redhat.com>
... since we already check for fb->pixel_format, which encodes all
this. The other two fields are only for backwards compat of older
drivers (and we might want to look into eventually just killing them).
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Dave Airlie <airlied@redhat.com>
There's no point in trying to clean up after driver-bugs, so just blow
up. Furthermore it's an interface abuse to set no mode but have an fb
and aslo to try to set an fb without enough connectors. These two
spefici cases of interface abuse have been committed by the fb helper,
but that's been fixed meanwhile in
commit 7e53f3a423
Author: Daniel Vetter <daniel.vetter@ffwll.ch>
Date: Mon Jan 21 10:52:17 2013 +0100
drm/fb-helper: fixup set_config semantics
The i915 driver has been shipping since a while with these BUGs with
no reports, so should be save.
Note that this drops an ugly case where we clear crtc->fb behind the
upper levels back and so cause a refcounting mayhem, which Russell
Kins spotted while trying to hunt down a drm framebuffer leak.
Reported-by: Russell King <linux@arm.linux.org.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Dave Airlie <airlied@redhat.com>
Git commit 90797e6d1e
("drm/i915: create compact dma scatter lists for gem objects") makes
certain assumptions about the under laying DMA API that are not always
correct.
On a ThinkPad X230 with an Intel HD 4000 with Xen during the bootup
I see:
[drm:intel_pipe_set_base] *ERROR* pin & fence failed
[drm:intel_crtc_set_config] *ERROR* failed to set mode on [CRTC:3], err = -28
Bit of debugging traced it down to dma_map_sg failing (in
i915_gem_gtt_prepare_object) as some of the SG entries were huge (3MB).
That unfortunately are sizes that the SWIOTLB is incapable of handling -
the maximum it can handle is a an entry of 512KB of virtual contiguous
memory for its bounce buffer. (See IO_TLB_SEGSIZE).
Previous to the above mention git commit the SG entries were of 4KB, and
the code introduced by above git commit squashed the CPU contiguous PFNs
in one big virtual address provided to DMA API.
This patch is a simple semi-revert - were we emulate the old behavior
if we detect that SWIOTLB is online. If it is not online then we continue
on with the new compact scatter gather mechanism.
An alternative solution would be for the the '.get_pages' and the
i915_gem_gtt_prepare_object to retry with smaller max gap of the
amount of PFNs that can be combined together - but with this issue
discovered during rc7 that might be too risky.
Reported-and-Tested-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
CC: Chris Wilson <chris@chris-wilson.co.uk>
CC: Imre Deak <imre.deak@intel.com>
CC: Daniel Vetter <daniel.vetter@ffwll.ch>
CC: David Airlie <airlied@linux.ie>
CC: <dri-devel@lists.freedesktop.org>
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
One remaining regression fix for i915. I've left it in -fixes for more
than a week since it's in tricky code, and it took us a few kernel
releases to notice the regression at all. The fence leak is especially
annoying on gen2/3 and will kill userspace there quickly. For extra
paranoia we've added a WARN in -next to catch this, things seem to be
solid now.
* tag 'drm-intel-fixes-2013-06-24' of git://people.freedesktop.org/~danvet/drm-intel:
drm/i915: Restore fences after resume and GPU resets
This patch merges host1x_syncpt_cpu_incr to host1x_syncpt_incr() as
they are in practise doing the same thing. host1x_syncpt_incr() is
also modified to return error codes. User space interface is modified
accordingly to pass return values.
Signed-off-by: Arto Merilainen <amerilainen@nvidia.com>
Acked-By: Terje Bergstrom <tbergstrom@nvidia.com>
Signed-off-by: Thierry Reding <thierry.reding@gmail.com>
client_managed field in syncpoint structure was defined as an
integer. The field holds, however, only a boolean value. This patch
modifies the type to boolean.
Signed-off-by: Arto Merilainen <amerilainen@nvidia.com>
Acked-By: Terje Bergstrom <tbergstrom@nvidia.com>
Signed-off-by: Thierry Reding <thierry.reding@gmail.com>
This patch fixes a bad memory access in syncpoint request code. If
no syncpoints were available, the code accessed unreserved memory
area causing unexpected behaviour.
Signed-off-by: Arto Merilainen <amerilainen@nvidia.com>
Acked-By: Terje Bergstrom <tbergstrom@nvidia.com>
Signed-off-by: Thierry Reding <thierry.reding@gmail.com>
The firewall verified gather buffers before copying them. This
allowed a malicious application to rewrite the buffer content by
timing the rewrite carefully.
This patch makes the buffer validation occur after copying the
buffers.
Signed-off-by: Arto Merilainen <amerilainen@nvidia.com>
Signed-off-by: Terje Bergstrom <tbergstrom@nvidia.com>
Signed-off-by: Thierry Reding <thierry.reding@gmail.com>
The firewall was reinitialised for each gather. Because the filter
was reinitialised, it did not track the class over gather boundaries.
This allowed the user application to set host1x class to one class
in one gather and use that class in another gather without firewall
having knowledge about that.
Signed-off-by: Terje Bergstrom <tbergstrom@nvidia.com>
Signed-off-by: Arto Merilainen <amerilainen@nvidia.com>
Signed-off-by: Thierry Reding <thierry.reding@gmail.com>
The firewall assumed that the user space always delivers a relocation
table when it is accessing address registers. If userspace did not
deliver a relocation table and tried to access the address registers,
the code performed bad memory accesses.
This patch modifies the firewall to check correctly that the firewall
table is available before accessing it. In addition, check_reloc() is
converted to use boolean return value (true when the reloc is valid,
false when invalid).
Signed-off-by: Arto Merilainen <amerilainen@nvidia.com>
Acked-By: Terje Bergstrom <tbergstrom@nvidia.com>
Signed-off-by: Thierry Reding <thierry.reding@gmail.com>
The firewall code used a wrong loop condition (pointer to a
structure) while checking INCR opcode. This patch fixes the code to
use correct loop condition (number of words remaining).
Signed-off-by: Terje Bergstrom <tbergstrom@nvidia.com>
Signed-off-by: Arto Merilainen <amerilainen@nvidia.com>
Signed-off-by: Thierry Reding <thierry.reding@gmail.com>
DRIVER_BUS_PLATFORM is not a DRM driver feature flag, it must not be set
in the driver's driver_features field.
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Signed-off-by: Thierry Reding <thierry.reding@gmail.com>
One user visible fix to stop misreport GPU hangs and subsequent resets.
* 'drm-fixes-3.10' of git://people.freedesktop.org/~agd5f/linux:
drm/radeon: update lockup tracking when scheduling in empty ring
There might be issue with lockup detection when scheduling on an
empty ring that have been sitting idle for a while. Thus update
the lockup tracking data when scheduling new work in an empty ring.
Signed-off-by: Jerome Glisse <jglisse@redhat.com>
Tested-by: Andy Lutomirski <luto@amacapital.net>
Cc: stable@vger.kernel.org
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
This simplifies cleanup paths and fixes a probe time crash in the error
path when trying to cleanup mode setting before it was initialized.
Signed-off-by: Laurent Pinchart <laurent.pinchart+renesas@ideasonboard.com>
The preferred_bpp value in currently hard-coded to 16.
This causes color corruption on the am335x-evm lcd panel which
requires 32 bpp instead. This changes attempts to use the configured
bpp value from the DT or built-in panel-info struct.
Signed-off-by: Benoit Parrot <bparrot@ti.com>
Acked-by: Rob Clark <robdclark@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
Alex writes:
Remove some harmless but confusing VM related error messages
fix a regression with suspend and UVD,
fix UVD on big endian.
* 'drm-fixes-3.10' of git://people.freedesktop.org/~agd5f/linux:
drm/radeon: fix UVD on big endian
drm/radeon: fix write back suspend regression with uvd v2
drm/radeon: do not try to uselessly update virtual memory pagetable
Now that we have this all nicely abstract into separate functions with
self-documenting names this is pointless. And as Yuly Novikov spotted
in the case of ilk-ivb also wrong since we use the pfit both for lvds
and eDP
Reported-By: Yuly Novikov <ynovikov@chromium.org>
Cc: Jesse Barnes <jbarnes@virtuousgeek.org>
Acked-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Just move the lowfreq_avail logic out of the register writing as a
prep step for the next patch, which will coalesce all the pch pll
enabling into one spot.
Note that writing the reduced clock dividers to FP1 in a few more
cases (as this patch ends up doing) isn't really relevant since the
FP1 value only matters when we enable the low lock. Which despite
can only happen if we've actually enabled the reduced dotclock and
furthermore isn't even properly implemented on ilk+: Despite claims to
the contrary in the code switching between frequencies if fully
manual.
v2: Explain matters around the FP1 change to answer a question Damien
raised in his review.
Reviewed-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Nowadays (i.e. with Valleyview) we also have edp on non-PCH_SPLIT
platforms, so just checking for LVDS is not good enough.
Secondly we have full pfit pipe config tracking, so we'll correctly
disable the pfit as part of the initial modeset.
For fastboot we need a bit of work here to correctly kill unsupported
configs (if e.g. the pfit is used on anything else than the built-in
panel). But since that's not yet supported we don't need to worry.
Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Again we don't really support different settings, so don't let the
BIOS sneak stuff through.
Since the motivation for this patch series is to ensure we have the
correct gamma table mode selected also add the required write to the
GAMMA_MODE register to select the 8bit legacy table.
And since I find lowercase letters in #defines offensive, also
bikeshed those.
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Same reasons as for the previous patch, just no bug report about
anything going wrong yet: We only support exactly the mode we program,
so don't leave any stale BIOS state behind.
Again this will be fun to properly track for fastboot.
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Dragging random garbage along from the BIOS isn't a good idea, since
we really only support exactly what we've set up.
In the specific case for the bug reporter the BIOS used the 10bit
gamma table, but since we only support an 8bit table the dark colors
ended up all wrong and the light ones all unadjusted.
Note that this has a nice implication for fastboot, it essentially
means that we have quite a bit more state to check and compare before
we can decide whether fastboot is possible.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=65593
Reported-and-Tested-by: Thomas Hebb <tommyhebb@gmail.com>
Cc: stable@vger.kernel.org
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
The omapdrm driver currently uses a string comparison to find out if the
display is a DVI display. This is not reliable, and as we now have a
specific display type for DVI, let's use that.
Signed-off-by: Tomi Valkeinen <tomi.valkeinen@ti.com>
Now that omap_dss_output has been combined into omap_dss_device, we can
add ref counting for the relevant output functions also.
This patch adds omap_dss_get_device() calls to the various find_output()
style functions. This, of course, means that the users of those
find_output functions need to do a omap_dss_put_device() after use.
Signed-off-by: Tomi Valkeinen <tomi.valkeinen@ti.com>
We currently have omap_dss_device, which represents an external display
device, sometimes an external encoder, sometimes a panel. Then we have
omap_dss_output, which represents DSS's output encoder.
In the future with new display device model, we construct a video
pipeline from the display blocks. To accomplish this, all the blocks
need to be presented by the same entity.
Thus, this patch combines omap_dss_output into omap_dss_device. Some of
the fields in omap_dss_output are already found in omap_dss_device, but
some are not. This means we'll have DSS output specific fields in
omap_dss_device, which is not very nice. However, it is easier to just
keep those output specific fields there for now, and after transition to
new display device model is made, they can be cleaned up easier than
could be done now.
Signed-off-by: Tomi Valkeinen <tomi.valkeinen@ti.com>
We currently have two steps in panel initialization and startup: probing
and enabling. After the panel has been probed, it's ready and can be
configured and later enabled.
This model is not enough with more complex display pipelines, where we
may have, for example, two panels, of which only one can be used at a
time, connected to the same video output.
To support that kind of scenarios, we need to add new step to the
initialization: connect.
This patch adds support for connecting and disconnecting panels. After
probe, but before connect, no panel ops should be called. When the
connect is called, a proper video pipeline is established, and the panel
is ready for use. If some part in the video pipeline is already
connected (by some other panel), the connect call fails.
One key difference with the old style setup is that connect() handles
also connecting to the overlay manager. This means that the omapfb (or
omapdrm) no longer needs to figure out which overlay manager to use, but
it can just call connect() on the panel, and the proper overlay manager
is connected by omapdss.
This also allows us to add back the support for dynamic switching
between two exclusive panels. However, the current panel device model is
not changed to support this, as the new device model is implemented in
the following patches and the old model will be removed. The new device
model supports dynamic switching.
Signed-off-by: Tomi Valkeinen <tomi.valkeinen@ti.com>
Currently omapdrm creates crtcs, which map directly to DSS overlay
managers, only on demand at init time. This would make it difficult to
manage connecting the display entities in the future, as the code cannot
just search for a suitable overlay manager.
We cannot fix this the sane way, which would be to create crtcs for each
overlay manager, because we need an overlay for each crtc. With limited
number of overlays, that's not possible.
So the solution for now is to detach the overlay manager from the crtc.
crtcs are still created on demand at init time, but all overlay managers
are always initialized by the omapdss.
This way we can create and connect whole display pipelines from the
overlay manager to the display, regardless of which crtcs omapdrm would
create.
Signed-off-by: Tomi Valkeinen <tomi.valkeinen@ti.com>
Add two helper functions that can be used to find either the DSS output
or the overlay manager that is connected to the given display.
This hides how the output and the manager are actually connected, making
it easier to change the connections in the future.
Signed-off-by: Tomi Valkeinen <tomi.valkeinen@ti.com>
Running mgag200_driver_unload when the driver init fails early on
causes functions like drm_mode_config_cleanup to be called. The
problem is, drm_mode_config_cleanup crashes because the corresponding
init hasn't happend yet. There really isn't anything to cleanup after
mgag200_device_init, so we can just pass the error code upwards.
Acked-by: Julia Lemire <jlemire@matrox.com>
Signed-off-by: Christopher Harvey <charvey@matrox.com>
Acked-by: Paul Menzel <paulepanter@users.sourceforge.net>
Signed-off-by: Dave Airlie <airlied@gmail.com>
G200 cards support, at best, 16 colour palleted images for the cursor
so we do a conversion in the cursor_set function, and reject cursors
with more than 16 colours, or cursors with partial transparency. Xorg
falls back gracefully to software cursors in this case.
We can't disable/enable the cursor hardware without causing momentary
corruption around the cursor. Instead, once the cursor is on we leave
it on, and simulate turning the cursor off by moving it
offscreen. This works well.
Since we can't disable -> update -> enable the cursors, we double
buffer cursor icons, then just move the base address that points to
the old cursor, to the new. This also works well, but uses an extra
page of memory.
The cursor buffers are lazily-allocated on first cursor_set. This is
to make sure they don't take priority over any framebuffers in case of
limited memory.
Here is a representation of how the bitmap for the cursor is mapped in G200 memory :
Each line of color cursor use 6 Slices of 8 bytes. Slices 0 to 3
are used for the 4bpp bitmap, slice 4 for XOR mask and slice 5 for
AND mask. Each line has the following format:
// Byte 0 Byte 1 Byte 2 Byte 3 Byte 4 Byte 5 Byte 6 Byte 7
//
// S0: P00-01 P02-03 P04-05 P06-07 P08-09 P10-11 P12-13 P14-15
// S1: P16-17 P18-19 P20-21 P22-23 P24-25 P26-27 P28-29 P30-31
// S2: P32-33 P34-35 P36-37 P38-39 P40-41 P42-43 P44-45 P46-47
// S3: P48-49 P50-51 P52-53 P54-55 P56-57 P58-59 P60-61 P62-63
// S4: X63-56 X55-48 X47-40 X39-32 X31-24 X23-16 X15-08 X07-00
// S5: A63-56 A55-48 A47-40 A39-32 A31-24 A23-16 A15-08 A07-00
//
// S0 to S5 = Slices 0 to 5
// P00 to P63 = Bitmap - pixels 0 to 63
// X00 to X63 = always 0 - pixels 0 to 63
// A00 to A63 = transparent markers - pixels 0 to 63
// 1 means colour, 0 means transparent
Signed-off-by: Christopher Harvey <charvey@matrox.com>
Signed-off-by: Mathieu Larouche <mathieu.larouche@matrox.com>
Acked-by: Julia Lemire <jlemire@matrox.com>
Tested-by: Julia Lemire <jlemire@matrox.com>
Signed-off-by: Dave Airlie <airlied@gmail.com>
