Currently, we may bump into pll mismatch errors during the
state verification stage. This happens when we try to use
fastset instead of full modeset. Hence, we would need to add
a check for pipe configuration to ensure that the sw and the
hw configuration will match. In case of hw and sw mismatch,
we would need to disable fastset and use full modeset instead.
v2: Fix C10 error on PLL comparison (BAT)
Use memcmp instead of fixed loops for pll config
comparison (Jani)
Clean up and use intel_cx0pll_dump_hw_state() to dump
pll information (Jani)
Reviewed-by: Jani Nikula <jani.nikula@intel.com>
Signed-off-by: Mika Kahola <mika.kahola@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240523134649.31452-3-mika.kahola@intel.com
Bunch of stuff in intel_color_check() needs to look at both the
old and new crtc states. Currently we do that by digging the
full atomic state via the crtc_state->state pointer. That thing
is a total footgun if I ever saw one, as it's only valid during
specific parts of the atomic flow. A lot of people have been
bitten by this thing in the past when trying to use it after
it's no longer valid.
Take a small step towards elimination of the footgun by not
using it in the inte_color_check(). Instead we plumb in the
entire atomic state all the way from the top.
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240523182818.15382-2-ville.syrjala@linux.intel.com
Reviewed-by: Jani Nikula <jani.nikula@intel.com>
With Xorg+modesetting on skl+ we see the following behaviour:
1. root pixmap is X-tiled
2. client submitted buffers can be Y-tiled (w/ 'Option "dmabuf_capable"')
3. we try to switch from the X-tiled buffer to the Y-tiled buffer
using an async flip (when vsync is disabled).
4. the async flip will be rejected by i915 due to the modifier change
Relax the rules a bit by turning the first async flip into a sync
flip so that we can change the modifier if necessary. Note that
we already convert the first async flip into a sync flip on adl+
in order to reprogram the watermarks.
Reviewed-by: Vandita Kulkarni <vandita.kulkarni@intel.com>
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240430095639.26390-4-ville.syrjala@linux.intel.com
Replace all code that initializes or releases fbdev emulation
throughout the driver. Instead initialize the fbdev client by a
single call to intel_fbdev_setup() after i915 has registered its
DRM device. Just like similar code in other drivers, i915 fbdev
emulation now acts like a regular DRM client. Do the same for xe.
The fbdev client setup consists of the initial preparation and the
hot-plugging of the display. The latter creates the fbdev device
and sets up the fbdev framebuffer. The setup performs display
hot-plugging once. If no display can be detected, DRM probe helpers
re-run the detection on each hotplug event.
A call to drm_client_dev_unregister() releases all in-kernel clients
automatically. No further action is required within i915. If the fbdev
framebuffer has been fully set up, struct fb_ops.fb_destroy implements
the release. For partially initialized emulation, the fbdev client
reverts the initial setup. Do the same for xe and remove its call to
intel_fbdev_fini().
v8:
- setup client in intel_display_driver_register (Jouni)
- mention xe in commit message
v7:
- update xe driver
- reword commit message
v6:
- use 'i915' for i915 device (Jouni)
- remove unnecessary code for non-atomic mode setting (Jouni, Ville)
- fix function name in commit message (Jouni)
v3:
- as before, silently ignore devices without displays
v2:
- let drm_client_register() handle initial hotplug
- fix driver name in error message (Jani)
- fix non-fbdev build (kernel test robot)
Signed-off-by: Thomas Zimmermann <tzimmermann@suse.de>
Reviewed-by: Jouni Högander <jouni.hogander@intel.com>
Acked-by: Lucas De Marchi <lucas.demarchi@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240409081029.17843-7-tzimmermann@suse.de
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
struct intel_dpll_hw_state has a spot for all possible
PLL registers across all platforms (well, apart from
cx0/snps). This makes it rather confusing when trying to
figure out which members belong to which platform(s).
Split the struct up into five different platform specific
sub-structures. For now this will actually increase the size
a little bit as we have to duplicate a few members from
skl to icl, but that will be remedied soon when we turn
the thing into a union.
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240412182703.19916-17-ville.syrjala@linux.intel.com
Reviewed-by: Jani Nikula <jani.nikula@intel.com>
Handle only bigjoiner masters in skl_commit_modeset_enables/disables,
slave crtcs should be handled by master hooks. Same for encoders.
That way we can also remove a bunch of checks like intel_crtc_is_bigjoiner_slave.
v2: - Moved skl_pfit_enable, intel_dsc_enable, intel_crtc_vblank_on to intel_enable_ddi,
so that it is now finally symmetrical with the disable case, because currently
for some weird reason we are calling those from skl_commit_modeset_enables, while
for the disable case those are called from the ddi disable hooks.
v3: - Create intel_ddi_enable_hdmi_or_sst symmetrical to
intel_ddi_post_disable_hdmi_or_sst and move it also under non-mst check.
v4: - Fix intel_enable_ddi sequence
- Call intel_crtc_update_active_timings for slave pipes as well
[v5: vsyrjala: Use the name 'pipe_crtc' for the per-pipe crtc pointer
Use consistent style and naming
Protect macro arguments properly
Drop superfluous changes to the modeset sequence,
this now follows the old non-joiner sequence 100%
apart from just looping in places]
Signed-off-by: Stanislav Lisovskiy <stanislav.lisovskiy@intel.com>
Tested-by: Vidya Srinivas <vidya.srinivas@intel.com>
Reviewed-by: Manasi Navare <navaremanasi@chromium.org> #v4?
Co-developed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240409163502.29633-5-ville.syrjala@linux.intel.com
With bigjoiner the master crtc is the one that will send out the
uapi event/etc. We want that to happen after all the slaves are
done, so let's try to do the commits in reverse order so that
the master comes last.
Even worse, the modeset helper will simply complete the commit
on the slave pipe immediately as it consider the crtc to be inactive
(it can't see our crtc_state->hw.active/etc.).
With regular sync updates this generally doesn't matter all that
much as the slave pipe should typically finish its work during the
same frame as the master pipe. However in case the slave pipe's commit
slips into the next frame we end up in a bit of trouble. This is most
visible with either async flips (currently disabled with bigjoiner
exactly for this reason), and DSB gamma updates. With DSB the problem
happens because the DSB itself will wait until the next start vblank
before starting to execute. So if the master pipe already finished its
commit and the DSB on the slave pipe is still waiting for the next
vblank we will assume the DSB as gotten stuck and terminate it.
Reversing the commit order should ameliarate this for the most part
as the master pipe is guaranteed to start its commit after the slave
pipe started. The one thing that can still screw us over is the fact
that we aren't necessarily going to commit the pipes in the reverse
order as the actual order is dictated by the DDB overlap avoidance.
But that can only happen while other pipes are being enabled/disabled,
and so in the normal steady state we should be safe.
The full fix will involve making the commit machinery aware of the
slave pipes and not finish their commits prematurely. But that
will involve a bit more work than this. And this commit order
reversal will still be beneficial to avoid userspace getting an
-EBUSY from the following page flip if the second pipe's commit
does stretch into the next frame.
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240404213441.17637-2-ville.syrjala@linux.intel.com
Reviewed-by: Arun R Murthy <arun.r.murthy@intel.com>
Reviewed-by: Vandita Kulkarni <vandita.kulkarni@intel.com>
Currently intel_modeset_pipe_config_late() is called after the
bigjoiner state copy, and it will actually not do anything for
bigjoiner slaves. This can lead to a mismatched state between
the master and slave.
The two things that we do in the encoder .compute_config_late()
hook are mst master transcoder and port sync master transcoder
elections. So if either of either MST or port sync is combined
with bigjoiner then we can see the mismatch.
Currently this problem is more or less theoretical; MST+bigjoiner
has not been implemented yet, and port sync+bigjoiner would
require a tiled display with >5k tiles (or a very high
dotclock per tile). Although we do have kms_tiled_display in
igt which can fake a tiled display, and we can now force bigjoiner
via debugfs, so it is possible to trigger this if you try hard
enough.
Reorder the code such that intel_modeset_pipe_config_late()
will be called before the bigjoiner state copy happens so
that both pipes will end up with the same state.
Tested-by: Vidya Srinivas <vidya.srinivas@intel.com>
Reviewed-by: Vandita Kulkarni <vandita.kulkarni@intel.com>
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240404213441.17637-7-ville.syrjala@linux.intel.com
Currently we can't change MBUS join status without doing a modeset,
because we are lacking mechanism to synchronize those with vblank.
However then this means that we can't do a fastset, if there is a need
to change MBUS join state. Fix that by implementing such change.
We already call correspondent check and update at pre_plane dbuf update,
so the only thing left is to have a non-modeset version of that.
If active pipes stay the same then fastset is possible and only MBUS
join state/ddb allocation updates would be committed.
The full mbus/cdclk sequence will look as follows:
1. disable pipes
2. increase cdclk if necessary
2.1 reprogram cdclk
2.2 update dbuf tracker value
3. enable mbus joining if necessary
3.1 update mbus_ctl
3.2 update dbuf tracker value
4. reallocate dbuf for planes on active pipes
5. disable mbus joining if necessary
5.1 update dbuf tracker value
5.2 update mbus_ctl
6. enable pipes
7. decrease cdclk if necessary
7.1 update dbuf tracker value
7.2 reprogram cdclk
And in order to keep things in sync we need:
Step 2:
- mbus_join == old
- mdclk/cdclk ratio == new
Step 3:
- mbus_join == new
- mdclk/cdclk ratio == old when cdclk is changing in step 7
- mdclk/cdclk ratio == new when cdclk is changing in step 2
Step 5:
- mbus_join == new
- mdclk/cdclk ratio == old when cdclk is changing in step 7
- mdclk/cdclk ratio == new when cdclk is changing in step 2
Step 7:
- mbus_join == new
- mdclk/cdclk ratio == new
v2: - Removed redundant parentheses(Ville Syrjälä)
- Constified new_crtc_state in intel_mbus_joined_pipe(Ville Syrjälä)
- Removed pipe_select variable(Ville Syrjälä)
[v3: vsyrjala: Correctly sequence vs. cdclk updates,
properly describe the full sequence,
shuffle code around to make the diff more legible,
streamline a few things]
[v4: vsyrjala: Move the intel_cdclk_is_decreasing_later() stuff
to a separate patch]
Cc: Gustavo Sousa <gustavo.sousa@intel.com>
Reviewed-by: Uma Shankar <uma.shankar@intel.com> #v3
Signed-off-by: Stanislav Lisovskiy <stanislav.lisovskiy@intel.com>
Co-developed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20240402155016.13733-13-ville.syrjala@linux.intel.com
