Bit 30 being set causes the upper half of BAR2 to stay in physical mode,
mapped over the end of VRAM, even when the rest of the BAR has been set
to virtual mode.
We inherited our initial value from RM, but I'm not aware of any reason
we need to keep it that way.
This fixes severe GPU hang/lockup issues revealed by Wayland on F26.
Shout-out to NVIDIA for the quick response with the potential cause!
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Cc: stable@vger.kernel.org # 4.3+
GP102's cursors go from chan 17..20. Increase the array size to hold
their data properly.
Fixes: e50fcff15f ("drm/nouveau/disp/gp102: fix cursor/overlay immediate channel indices")
Cc: stable@vger.kernel.org # v4.10+
Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
These on()/off() calls should be done as a result of modesetting actions,
and as we shut down all heads already on unload/suspend, it's pointless
to call off() again.
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
We don't support them on G80, but we need to add them to the mapping to
avoid triggering a WARN_ON() on GPUs where the ports are present.
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Since switching the I2C-over-AUX helpers, there have been regressions on
some display combinations due to us not having support for "address only"
transactions.
This commits enables support for them for GF119 and newer.
Earlier GPUs have been reverted to a custom I2C-over-AUX algorithm.
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
A bug that I had fixed earlier just came back, with CONFIG_EXTCON=m,
the rockchip drm driver will fail to link:
drivers/gpu/drm/rockchip/cdn-dp-core.o: In function `cdn_dp_get_port_lanes':
cdn-dp-core.c:(.text.cdn_dp_get_port_lanes+0x30): undefined reference to `extcon_get_state'
cdn-dp-core.c:(.text.cdn_dp_get_port_lanes+0x6c): undefined reference to `extcon_get_property'
drivers/gpu/drm/rockchip/cdn-dp-core.o: In function `cdn_dp_check_sink_connection':
cdn-dp-core.c:(.text.cdn_dp_check_sink_connection+0x80): undefined reference to `extcon_get_state'
drivers/gpu/drm/rockchip/cdn-dp-core.o: In function `cdn_dp_enable':
cdn-dp-core.c:(.text.cdn_dp_enable+0x748): undefined reference to `extcon_get_property'
The problem is that that the sub-drivers are now all linked into the
main rockchip drm module, which breaks all the Kconfig dependencies
that are specified in the options for those sub-drivers.
This clarifies the dependency to ensure that we can only turn on the DP
driver when EXTCON is reachable. As the 'select' statements can now
cause additional options to become built-in when they should be
loadable modules, I'm moving those into the main driver config option.
The dependency on DRM_ROCKCHIP can be reduced into a single 'if'
statement here for brevity, but this has no functional effect.
Fixes: b6705157b2 ("drm/rockchip: add extcon dependency for DP")
Fixes: 8820b68bd3 ("drm/rockchip: Refactor the component match logic.")
Link: https://patchwork.kernel.org/patch/9648761/
Acked-by: Guenter Roeck <groeck@chromium.org>
Tested-by: Jeffy Chen <jeffy.chen@rock-chips.com>
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Mark Yao <mark.yao@rock-chips.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20170721211214.3386387-1-arnd@arndb.de
imx-drm: fix parallel display regression and typo in plane format list
- Fix a regression where the parallel-display driver would not probe
anymore if no panel is specified in the device tree, since the
introduction of drm_of_find_panel_or_bridge.
- Fix a typo in the plane format list: replace a duplicate BGRA8888 format
with BGRX8888, as originally intended.
* tag 'imx-drm-fixes-2017-07-18' of git://git.pengutronix.de/git/pza/linux:
drm/imx: parallel-display: Accept drm_of_find_panel_or_bridge failure
drm/imx: fix typo in ipu_plane_formats[]
Core Changes:
- fence: Introduce new fence flag to signify timestamp is populated (Chris)
- mst: Avoid processing incomplete data + fix NULL dereference (Imre)
Driver Changes:
- vc4: Avoid WARN from grabbing a ref from vblank that's not on (Boris)
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Boris Brezillon <boris.brezillon@free-electrons.com>
Cc: Imre Deak <imre.deak@intel.com>
* tag 'drm-misc-fixes-2017-07-20' of git://anongit.freedesktop.org/git/drm-misc:
drm/mst: Avoid processing partially received up/down message transactions
drm/mst: Avoid dereferencing a NULL mstb in drm_dp_mst_handle_up_req()
drm/mst: Fix error handling during MST sideband message reception
drm/vc4: Fix VBLANK handling in crtc->enable() path
dma-buf/fence: Avoid use of uninitialised timestamp
Taking the modeset locks unconditionally isn't the greatest idea,
because atm that part is still broken and times out (and then atomic
keels over). And there's really no reason to do so, the old code
didn't do that either.
To make the patch a bit simpler let's also nuke 2 cases that are only
around for the old mmioflip paths. Atomic nonblocking workers will not
die (minus bugs) when a gpu reset happens.
And of course this doesn't fix any of the gpu reset vs. modeset
deadlock fun, but it at least stop modern CI machines from keeling
over all over the place for no reason at all.
And we still have the explicit testcases to run the fake gpu reset, so
coverage isn't that much worse.
v2: Split out additional changes on top, restrict this to purely reducing
the critical section of modeset locks.
v2: Review from Maarten
- update comments
- don't oops when state is NULL in intel_finish_reset, but try to at
least still drop locks properly. The hw is going to be toast anyway.
Fixes: 7397489399 ("drm/i915: Fix modeset handling during gpu reset, v5.")
Cc: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Reviewed-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20170719125502.25696-3-daniel.vetter@ffwll.ch
Currently we may process up/down message transactions containing
uninitialized data. This can happen if there was an error during the
reception of any message in the transaction, but we happened to receive
the last message correctly with the end-of-message flag set.
To avoid this abort the reception of the transaction when the first
error is detected, rejecting any messages until a message with the
start-of-message flag is received (which will start a new transaction).
This is also what the DP 1.4 spec 2.11.8.2 calls for in this case.
In addtion this also prevents receiving bogus transactions without the
first message with the the start-of-message flag set.
v2:
- unchanged
v3:
- git add the part that actually skips messages after an error in
drm_dp_sideband_msg_build()
Cc: Dave Airlie <airlied@redhat.com>
Cc: Lyude <lyude@redhat.com>
Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Imre Deak <imre.deak@intel.com>
Reviewed-by: Lyude <lyude@redhat.com>
Cc: stable@vger.kernel.org
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Link: https://patchwork.freedesktop.org/patch/msgid/20170719134632.13366-1-imre.deak@intel.com
2nd round of 4.14 features:
- prep for deferred fbdev setup
- refactor fixed 16.16 computations and skl+ wm code (Mahesh Kumar)
- more cnl paches (Rodrigo, Imre et al)
- tighten context cleanup and handling (Chris Wilson)
- fix interlaced handling on skl+ (Mahesh Kumar)
- small bits as usual
* tag 'drm-intel-next-2017-07-17' of git://anongit.freedesktop.org/git/drm-intel: (84 commits)
drm/i915: Update DRIVER_DATE to 20170717
drm/i915: Protect against deferred fbdev setup
drm/i915/fbdev: Always forward hotplug events
drm/i915/skl+: unify cpp value in WM calculation
drm/i915/skl+: WM calculation don't require height
drm/i915: Addition wrapper for fixed16.16 operation
drm/i915: cleanup fixed-point wrappers naming
drm/i915: Always perform internal fixed16 division in 64 bits
drm/i915: take-out common clamping code of fixed16 wrappers
drm/i915/cnl: Add missing type case.
drm/i915/cnl: Add max allowed Cannonlake DC.
drm/i915: Make DP-MST connector info work
drm/i915/cnl: Get DDI clock based on PLLs.
drm/i915/cnl: Inherit RPS stuff from previous platforms.
drm/i915/cnl: Gen10 render context size.
drm/i915/cnl: Don't trust VBT's alternate pin for port D for now.
drm/i915: Fix the kernel panic when using aliasing ppgtt
drm/i915/cnl: Cannonlake color init.
drm/i915/cnl: Add force wake for gen10+.
x86/gpu: CNL uses the same GMS values as SKL
...
First thing we need to do is unregister the fbdev instance, but we
can't just go ahead and kfree it. That must wait until the hotplug and
polling work are stopped, since they can race with the with the
teardown. That means we need to split up the fbdev teardown into the
unregister part and the cleanup part.
I originally suspected that this was broken in one of the unload
shuffles, but on closer inspection the oldest sequence I've dug out
also gets this wrong. Just not quite so badly.
I've run drv_module_reload a few hundred times and it's rock solid
compared to insta-death beforehand. This bug seems to have been
uncovered by
commit 88be58be88
Author: Daniel Vetter <daniel.vetter@ffwll.ch>
Date: Thu Jul 6 15:00:19 2017 +0200
drm/i915/fbdev: Always forward hotplug events
But the effect of that seems to only be to increase the race window
enough to make it blow up easier. I'm not exactly clear on what's
going on there ...
v2: Fix whitespace and use fetch_and_zero (Chris).
Testcase: igt/drv_module_reload
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=101791
Cc: martin.peres@free.fr
Cc: chris@chris-wilson.co.uk
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20170714224656.6431-2-daniel.vetter@ffwll.ch
The legacy plane->fb pointer is refcounted by calling
drm_atomic_clean_old_fb().
In practice this isn't a real problem because:
- The caller in the i915 gpu reset code restores the original state
again, which means the plane->fb pointer won't change, hence can't
leak.
- Drivers using drm_atomic_helper_shutdown call the fbdev cleanup
first, and that usually cleans up the fb through
drm_remove_framebuffer, which does this correctly.
- Without fbdev the only framebuffers are from userspace, and those
get cleaned up (again using drm_remove_framebuffer) befor the driver
can even be unloaded.
But in i915 I've switched the cleanup sequence around so that the
_shutdown() calls happens after the drm_remove_framebuffer(), which is
how I discovered this issue.
v2: My analysis why the current code was ok for gpu reset and
suspend/resume was correct, but then I totally failed to realize that
we better keep this symmetric. Thanksfully CI noticed that for
balance, a refcounting bug must exist at 2 places if previously there
was no issue ...
v3: Don't be lazy and compute the plane_mask in
commit_duplicated_state properly too, instead of just using ~0U.
Cc: martin.peres@free.fr
Cc: chris@chris-wilson.co.uk
Acked-by: Dave Airlie <airlied@gmail.com> (v1)
Reviewed-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20170715093106.19873-1-daniel.vetter@ffwll.ch
Add HDMI CEC support to the Allwinner A10 SoC.
This SoC uses a poor-man's CEC implementation by polling the CEC pin. It is
using the CEC_PIN core implementation for such devices to do the heavy
lifting. It just provides the callbacks to read/drive the CEC pin.
Signed-off-by: Hans Verkuil <hans.verkuil@cisco.com>
Tested-by: Maxime Ripard <maxime.ripard@free-electrons.com>
Signed-off-by: Maxime Ripard <maxime.ripard@free-electrons.com>
Even though that function is defined in the TCON header, it's not defined
nor used anywhere. Remove the prototype.
Reviewed-by: Chen-Yu Tsai <wens@csie.org>
Signed-off-by: Maxime Ripard <maxime.ripard@free-electrons.com>
The atomic_check callback is optional, and we don't implement anything in
some parts of our drivers. Let's remove it.
Reviewed-by: Chen-Yu Tsai <wens@csie.org>
Signed-off-by: Maxime Ripard <maxime.ripard@free-electrons.com>
The depends on relationship is obvious, and using an if statement will
propagate it to every option without the need for each and every one of
them to define it.
Reviewed-by: Chen-Yu Tsai <wens@csie.org>
Signed-off-by: Maxime Ripard <maxime.ripard@free-electrons.com>
Turns out that just writing CURPOS isn't sufficient to move the cursor
on some platforms. My 830 works just fine, but eg. 945 and PNV don't.
On those platforms we need to arm even the CURPOS update with a
CURBASE write.
Even worse, a write to any of the cursor register apart from CURBASE
will cancel an already pending cursor update. So if we have armed a
CURCNTR/CURBASE update, a subsequent CURPOS write prior to vblank
would cancel that armed update. Thus we're left with a cursor that
doesn't appear to move, or even change shape.
Fix the problem by always performing the CURBASE write after a
CURPOS write. Bspec is somewhat unclear which platforms actually
require this CURBASE write and which don't. So to keep it simple
and to make sure we really fix the problem across all supported
devices, let's just perform the CURBASE write unconditionally.
Cc: Paul Menzel <pmenzel@molgen.mpg.de>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=101790
Fixes: 75343a44c9 ("drm/i915: Drop useless posting reads from cursor commit")
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Tested-by: Paul Menzel <paulepanter@users.sourceforge.net>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Link: https://patchwork.freedesktop.org/patch/msgid/20170714155227.6089-1-ville.syrjala@linux.intel.com
For modern drivers the DRM core doesn't use struct_mutex at all, which
means it's defacto a driver-private lock. But since we still need it
for legacy drivers we can't initialize it in drivers, which means all
the different instances share one lockdep key. Despite that they might
be placed in totally different places in the locking hierarchy.
This results in a lot of bogus lockdep splats when running stuff on
systems with multiple gpus. Partially remedy the situation by only
doing might_lock checks on drivers that do use struct_mutex still for
gem locking.
A more complete solution would be to do the mutex_init in the drm core
only for legacy drivers, plus add it to each modern driver that still
needs it, which would also give each its own lockdep key. Trying to do
that dynamically doesn't work, because lockdep requires it's keys to
be statically allocated.
v2: {} everywhere (Chris)
Cc: Hans de Goede <hdegoede@redhat.com>
Cc: Ben Skeggs <bskeggs@redhat.com>
Cc: Jiri Slaby <jirislaby@gmail.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Ingo Molnar <mingo@redhat.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20170715095328.25671-1-daniel.vetter@ffwll.ch
pci_device_id are not supposed to change at runtime. All functions
working with pci_device_id provided by <linux/pci.h> work with
const pci_device_id. So mark the non-const structs as const.
File size before:
text data bss dec hex filename
13765 800 20 14585 38f9 gpu/drm/vmwgfx/vmwgfx_drv.o
File size After adding 'const':
text data bss dec hex filename
13829 736 20 14585 38f9 gpu/drm/vmwgfx/vmwgfx_drv.o
Signed-off-by: Arvind Yadav <arvind.yadav.cs@gmail.com>
Reviewed-by: Sinclair Yeh <syeh@vmware.com>
Parts of commit <8fbf9d92a7bc> (“drm/vmwgfx: Implement the
cursor_set2 callback v2”) were not moved over when we started
atomic mode set development because at that time the DRM did
not support cursor hotspots in the fb struct.
This patch fixes what was not moved over.
Signed-off-by: Sinclair Yeh <syeh@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Tested-by: Brian Paul <brianp@vmware.com>
This was originally chosen to be an arbitrarily large number. However,
some user mode may actually try to set a 16Kx16K mode and run into other
issues.
Since 8Kx8K is the current texture limit for Mesa LLVM driver, we will
just use this limit for now.
Cc: <stable@vger.kernel.org> # 4.12.x
Signed-off-by: Sinclair Yeh <syeh@vmware.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
ttm_place are not supposed to change at runtime. All functions
working with ttm_place provided by <drm/ttm/ttm_placement.h> work
with const ttm_place. So mark the non-const structs as const.
File size before:
text data bss dec hex filename
3172 796 16 3984 f90 drivers/gpu/drm/vmwgfx/vmwgfx_buffer.o
File size After adding 'const':
text data bss dec hex filename
3456 512 16 3984 f90 drivers/gpu/drm/vmwgfx/vmwgfx_buffer.o
Signed-off-by: Arvind Yadav <arvind.yadav.cs@gmail.com>
Reviewed-by: Sinclair Yeh <syeh@vmware.com>
