Ville Syrjälä [Wed, 7 Sep 2022 09:10:55 +0000 (12:10 +0300)]
drm/i915: Allow M/N change during fastset on bdw+
On BDW+ M/N are double buffered and so we can easily reprogram them
during a fastset. So for eDP panels that support seamless DRRS we
can just change these without a full modeset.
For earlier platforms we'd need to play tricks with M1/N1 vs.
M2/N2 during the fastset to make sure we do the switch atomically.
Not sure the added complexity is worth the hassle, so leave it
alone for now.
The slight downside is that we have to keep the link running at
a link rate capable of supporting the highest refresh rate we
want to use. For the moment we just pick the highest mode the
panel reports and calculate the link based on that. This might
need further refinement (eg. if we run into bandwidth
restrictions)...
v2: Only use the high link rate if the platform really supports
the seamless M/N change uring fastset (ie. bdw+)
v3: Rebase due to HAS_DOUBLE_BUFFERED_M_N()
Ville Syrjälä [Wed, 7 Sep 2022 09:10:54 +0000 (12:10 +0300)]
drm/i915: Add intel_panel_highest_mode()
Add a function to get the fixed_mode with the highest clock.
The plan is to use this for the link bw calculation on seamless
DRRS panels so that we alwasy end up with the same link params
regardless of the requested refresh rate. This will allow fastset
to do seamless refresh rate changes based on userspace request
instead of having to go for a full modeset.
Ville Syrjälä [Wed, 7 Sep 2022 09:10:51 +0000 (12:10 +0300)]
drm/i915: Set active dpll early for icl+
To make the fastboot checks at least somewhat sensible let's mark
the expected DPLL as the active one right after we finished the
state computation. Otherwise intel_pipe_config_compare() will
always be comparing things against NULL/0.
TODO: This is still not really right. If the previous commit
had to fall back to the other PLL then the comparisong will
now fail. I guess intel_pipe_config_compare() should rather
be comparing port_dplls[] instead. But to do that we really
should just unify every platform to use the port_dplls[]
approach whether they have any need for PLL fallbacks or not.
Ville Syrjälä [Wed, 7 Sep 2022 09:10:49 +0000 (12:10 +0300)]
drm/i915: Make M/N checks non-fuzzy
Now that we no longer fuzz M/N during fastset these should
match exctly.
In order to get a match with what the BIOS does we need to round
M/N down. And we do the opposite rounding when doing the readback.
That gets us pretty much the same thing back.
There can still be slight rounding differences between FDI M/N
vs. the DPLL output so we allow for tiny deviation in
intel_pipe_config_sanity_check().
Ville Syrjälä [Wed, 7 Sep 2022 09:10:48 +0000 (12:10 +0300)]
drm/i915: Compute clocks earlier
Do the DPLL computation before fastset checks. This should
allow us to get rid of all that horrible fuzzy clock handling
for fastsets. Who knows how many bugs there are caused by our
state not actually matching what the hardware will generate.
Ville Syrjälä [Wed, 7 Sep 2022 09:10:47 +0000 (12:10 +0300)]
drm/i915: Feed the DPLL output freq back into crtc_state
Fill port_clock and hw.adjusted_mode.crtc_clock with the actual
frequency we're going to be getting from the hardware. This will
let us accurately compute all derived state that depends on those.
v2: Reintroduce iCLKIP WARN
v3: Try to deal with VLV/BXT DSI PLL as well
Ville Syrjälä [Wed, 7 Sep 2022 09:10:46 +0000 (12:10 +0300)]
drm/i915: Reassign DPLLs only for crtcs going throug .compute_config()
Only reassign the pipe's DPLL if it's going through a full
.compute_config() cycle. If OTOH it's just getting modeset
eg. in order to change cdclk there doesn't seem much point in
picking a new DPLL for it.
This should also prevent .get_dplls() from seeing a funky port_clock
for DP even in cases where the readout produces a non-standard
clock and we (for some reason) have decided to not fully recompute
the state to remedy the situation.
Ville Syrjälä [Wed, 7 Sep 2022 09:10:45 +0000 (12:10 +0300)]
drm/i915: Do .crtc_compute_clock() earlier
Currently we calculate a lot of things (pixel rate, watermarks,
cdclk) trusting that the DPLL can generate the exact frequency
we ask it. In practice that is not true and there can be
certain amount of rounding involved.
To allow us to eventually get accurate numbers for all our
DPLL clock derived state we need to move the DPLL calculation
to hapen much earlier. To that end we hoist it up to the just
after the fastset checks. For now we just do the easy code
motion, and the actual back feeding of the final DPLL clock
into the state will come later.
A slight change here is that now .crtc_compute_clock()
can get called while the shared_dpll is still assigned.
But since .crtc_compute_clock() no longer assignes new
shared_dplls this is perfectly fine.
TODO: I'd actually like to do this before the fastset check
so that if the DPLL state should change we actually do the
modeset. Which I think is what the video aficionados want,
but it might not be what the fans of fastboot want. Not yet
sure how to reconcile those conflicting requirements...
drm/i915/vdsc: Set VDSC PIC_HEIGHT before using for DP DSC
Currently, pic_height of vdsc_cfg structure is being used to calculate
slice_height, before it is set for DP.
So taking out the lines to set pic_height from the helper
intel_dp_dsc_compute_params() to individual encoders, and setting
pic_height, before it is used to calculate slice_height for DP.
Fixes: 5a6d866f8e1b ("drm/i915: Get slice height before computing rc params") Cc: Manasi Navare <manasi.d.navare@intel.com> Cc: Vandita Kulkarni <vandita.kulkarni@intel.com> Cc: Matt Roper <matthew.d.roper@intel.com> Signed-off-by: Ankit Nautiyal <ankit.k.nautiyal@intel.com> Reviewed-by: Vandita Kulkarni <vandita.kulkarni@intel.com> Signed-off-by: Matt Roper <matthew.d.roper@intel.com> Link: https://patchwork.freedesktop.org/patch/msgid/20220902103219.1168781-1-ankit.k.nautiyal@intel.com
Ville Syrjälä [Wed, 7 Sep 2022 09:10:43 +0000 (12:10 +0300)]
drm/i915: Extract HAS_DOUBLE_BUFFERED_M_N()
We have a couple of places that want to make distinction between
double buffered M/N registers vs. the split M1/N1+M2/N2 registers.
Add a helper for that.
Part of a series where patches were modified while applying to resolve
conflicts, leading to further conflicts between drm-misc-next and
drm-intel-next, resulting in build failures in drm-tip. To be applied
again on a baseline with drm-misc-next and drm-intel-next in sync.
Acked-by: Rodrigo Vivi <rodrigo.vivi@intel.com> Signed-off-by: Jani Nikula <jani.nikula@intel.com>
Part of a series where patches were modified while applying to resolve
conflicts, leading to further conflicts between drm-misc-next and
drm-intel-next, resulting in build failures in drm-tip. To be applied
again on a baseline with drm-misc-next and drm-intel-next in sync.
Acked-by: Rodrigo Vivi <rodrigo.vivi@intel.com> Signed-off-by: Jani Nikula <jani.nikula@intel.com>
Part of a series where patches were modified while applying to resolve
conflicts, leading to further conflicts between drm-misc-next and
drm-intel-next, resulting in build failures in drm-tip. To be applied
again on a baseline with drm-misc-next and drm-intel-next in sync.
Acked-by: Rodrigo Vivi <rodrigo.vivi@intel.com> Signed-off-by: Jani Nikula <jani.nikula@intel.com>
Part of a series where patches were modified while applying to resolve
conflicts, leading to further conflicts between drm-misc-next and
drm-intel-next, resulting in build failures in drm-tip. To be applied
again on a baseline with drm-misc-next and drm-intel-next in sync.
Acked-by: Rodrigo Vivi <rodrigo.vivi@intel.com> Signed-off-by: Jani Nikula <jani.nikula@intel.com>
The function was removed four years ago in commit 6faf5916e6be
("drm/i915: Remove HW semaphores for gen7 inter-engine
synchronisation"). Finish the job.
Whenever we are not able to get enough timeslots
for required PBN, let's try to allocate those
using DSC, just same way as we do for SST.
v2: Removed intel_dp_mst_dsc_compute_config and refactored
intel_dp_dsc_compute_config to support timeslots as a
parameter(Ville Syrjälä)
v3: - Rebased
- Added a debug to see that we at least try reserving
VCPI slots using DSC, because currently its not visible
from the logs, thus making debugging more tricky.
- Moved timeslots to numerator, where it should be.
v4: - Call drm_dp_mst_atomic_check already during link
config computation, because we need to know already
by this moment if uncompressed amount of VCPI slots
needed can fit, otherwise we need to use DSC.
(thanks to Vinod Govindapillai for pointing this out)
v5: - Put pipe_config->bigjoiner_pipes back to original
condition in intel_dp_dsc_compute_config
(don't remember when I lost it)
v6: - Removed unnecessary drm_dp_mst_atomic_check as it is
now always called in a newly introduced
intel_dp_mst_find_vcpi_slots_for_bpp function
(Vinod Govindapillai)
drm/i915: Extract drm_dp_atomic_find_vcpi_slots cycle to separate function
We would be using almost same code to loop through bpps while calling
drm_dp_atomic_find_vcpi_slots - lets remove this duplication by
introducing a new function intel_dp_mst_find_vcpi_slots_for_bpp
v2: Fix pbn_div calculation - shouldn't matter if its DSC or not.
We currently always exit that bpp loop because
drm_dp_atomic_find_vcpi_slots doesn't care if we actually
can fit those or not.
I think that wasn't the initial intention here, especially when
we keep trying with lower bpps, we are supposed to keep trying
until we actually find some _working_ configuration, which isn't the
case here.
So added that drm_dp_mst_check here, so that we can make sure
that try all the bpps before we fail.
Andrzej Hajda [Fri, 26 Aug 2022 14:19:28 +0000 (16:19 +0200)]
drm/i915/fbdev: suspend HPD before fbdev unregistration
HPD event after fbdev unregistration can cause registration of deferred
fbdev which will not be unregistered later, causing use-after-free.
To avoid it HPD handling should be suspended before fbdev unregistration.
It should fix following GPF:
[272.634530] general protection fault, probably for non-canonical address 0x6b6b6b6b6b6b6b6b: 0000 [#1] PREEMPT SMP NOPTI
[272.634536] CPU: 0 PID: 6030 Comm: i915_selftest Tainted: G U 5.18.0-rc5-CI_DRM_11603-g12dccf4f5eef+ #1
[272.634541] Hardware name: Intel Corporation Raptor Lake Client Platform/RPL-S ADP-S DDR5 UDIMM CRB, BIOS RPLSFWI1.R00.2397.A01.2109300731 09/30/2021
[272.634545] RIP: 0010:fb_do_apertures_overlap.part.14+0x26/0x60
...
[272.634582] Call Trace:
[272.634583] <TASK>
[272.634585] do_remove_conflicting_framebuffers+0x59/0xa0
[272.634589] remove_conflicting_framebuffers+0x2d/0xc0
[272.634592] remove_conflicting_pci_framebuffers+0xc8/0x110
[272.634595] drm_aperture_remove_conflicting_pci_framebuffers+0x52/0x70
[272.634604] i915_driver_probe+0x63a/0xdd0 [i915]
Closes: https://gitlab.freedesktop.org/drm/intel/-/issues/5329 Closes: https://gitlab.freedesktop.org/drm/intel/-/issues/5510 Signed-off-by: Andrzej Hajda <andrzej.hajda@intel.com> Reviewed-by: Arun R Murthy <arun.r.murthy@intel.com> Reviewed-by: Imre Deak <imre.deak@intel.com> Signed-off-by: Imre Deak <imre.deak@intel.com> Link: https://patchwork.freedesktop.org/patch/msgid/20220826141929.189681-3-andrzej.hajda@intel.com
Andrzej Hajda [Fri, 26 Aug 2022 14:19:27 +0000 (16:19 +0200)]
drm/i915/hpd: suspend MST at the end of intel_modeset_driver_remove
i915->hotplug.dig_port_work can be queued from intel_hpd_irq_handler
called by IRQ handler or by intel_hpd_trigger_irq called from dp_mst.
Since dp_mst is suspended after irq handler uninstall, a cleaner approach
is to cancel hpd work after intel_dp_mst_suspend, otherwise we risk
use-after-free.
Ville Syrjälä [Fri, 15 Jul 2022 20:20:38 +0000 (23:20 +0300)]
drm/i915: Define VBT max HDMI FRL rate bits
The VBT gained some bits to inidicate the max FRL rate for
HDMI 2.1, define them.
These just outright replaced the slave_port bits for ganged eDP.
Apparently that feature was never actually used so someone decided
that reusing the bits is fine. Although the actual ganged eDP
enable bit was still left defined elsewhere for some reason.
Ville Syrjälä [Fri, 15 Jul 2022 20:20:33 +0000 (23:20 +0300)]
drm/i915: Unify VBT version number comments
Use a more standard form for the VT version number comments.
One slight oddball case is the dp_max_link_rate that had two
version numbers (216/230) and a platform name (GLK). The
story goes that the field was introduced in the spec in
version 216, along with a note that it's used on CNL+. Later
in version 230 the definition of the bit was changed in
bacakwards incompatible ways and the CNL note disappeard.
For us the original CNL+ note in the header got changed to
to GLK+ when all CNL support was dropped from the codebase.
We do still need (and have) handling for both the 216+ and
the 230+ defintions (parse_bdb_216_dp_max_link_rate() vs.
parse_bdb_230_dp_max_link_rate()).
With the Parade PS8461E MUX workaround (WaEdpLinkRateDataReload)
implemented we can get finally rid of the is_low_voltage_sku()
check that incorrectly prevents many machines from using the
8.1Gpbs link rate.
Cc: Jason A. Donenfeld <Jason@zx2c4.com> Cc: Ankit Nautiyal <ankit.k.nautiyal@intel.com> Cc: Jani Nikula <jani.nikula@intel.com> Closes: https://gitlab.freedesktop.org/drm/intel/-/issues/5272 Closes: https://gitlab.freedesktop.org/drm/intel/-/issues/6323 Closes: https://gitlab.freedesktop.org/drm/intel/-/issues/6205 Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com> Link: https://patchwork.freedesktop.org/patch/msgid/20220902070319.15395-2-ville.syrjala@linux.intel.com Tested-by: Aaron Ma <aaron.ma@canonical.com> Tested-by: Jason A. Donenfeld <Jason@zx2c4.com> Reviewed-by: Jani Nikula <jani.nikula@intel.com>
Ville Syrjälä [Fri, 2 Sep 2022 07:03:18 +0000 (10:03 +0300)]
drm/i915: Implement WaEdpLinkRateDataReload
A lot of modern laptops use the Parade PS8461E MUX for eDP
switching. The MUX can operate in jitter cleaning mode or
redriver mode, the first one resulting in higher link
quality. The jitter cleaning mode needs to know the link
rate used and the MUX achieves this by snooping the
LINK_BW_SET, LINK_RATE_SELECT and SUPPORTED_LINK_RATES
DPCD accesses.
When the MUX is powered down (seems this can happen whenever
the display is turned off) it loses track of the snooped
link rates so when we do the LINK_RATE_SELECT write it no
longer knowns which link rate we're selecting, and thus it
falls back to the lower quality redriver mode. This results
in unstable high link rates (eg. usually 8.1Gbps link rate
no longer works correctly).
In order to avoid all that let's re-snoop SUPPORTED_LINK_RATES
from the sink at the start of every link training.
Unfortunately we don't have a way to detect the presence of
the MUX. It looks like the set of laptops equipped with this
MUX is fairly large and contains devices from multiple
manufacturers. It may also still be growing with new models.
So a quirk doesn't seem like a very easily maintainable
option, thus we shall attempt to do this unconditionally on
all machines that use LINK_RATE_SELECT. Hopefully this extra
DPCD read doesn't cause issues for any unaffected machine.
If that turns out to be the case we'll need to convert this
into a quirk in the future.
Cc: stable@vger.kernel.org Cc: Jason A. Donenfeld <Jason@zx2c4.com> Cc: Ankit Nautiyal <ankit.k.nautiyal@intel.com> Cc: Jani Nikula <jani.nikula@intel.com> Closes: https://gitlab.freedesktop.org/drm/intel/-/issues/6205 Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com> Link: https://patchwork.freedesktop.org/patch/msgid/20220902070319.15395-1-ville.syrjala@linux.intel.com Tested-by: Aaron Ma <aaron.ma@canonical.com> Tested-by: Jason A. Donenfeld <Jason@zx2c4.com> Reviewed-by: Jani Nikula <jani.nikula@intel.com>
Ville Syrjälä [Thu, 18 Aug 2022 19:22:23 +0000 (22:22 +0300)]
drm/i915/bios: Use hardcoded fp_timing size for generating LFP data pointers
The current scheme for generating the LFP data table pointers
(when the block including them is missing from the VBT) expects
the 0xffff sequence to only appear in the fp_timing terminator
entries. However some VBTs also have extra 0xffff sequences
elsewhere in the LFP data. When looking for the terminators
we may end up finding those extra sequeneces insted, which means
we deduce the wrong size for the fp_timing table. The code
then notices the inconsistent looking values and gives up on
the generated data table pointers, preventing us from parsing
the LFP data table entirely.
Let's give up on the "search for the terminators" approach
and instead just hardcode the expected size for the fp_timing
table.
We have enough sanity checks in place to make sure we
shouldn't end up parsing total garbage even if that size
should change in the future (although that seems unlikely
as the fp_timing and dvo_timing tables have been declared
obsolete as of VBT version 229).
Mitul Golani [Tue, 30 Aug 2022 08:51:58 +0000 (14:21 +0530)]
drm/i915/display: Fix warning callstack for imbalance wakeref
While executing i915_selftest, wakeref imbalance warning is seen
with i915_selftest failure.
Currently when Driver is suspended, while doing unregister
it is taking wakeref without resuming the device.
This patch is resuming the device, if driver is already suspended
and doing unregister process.
Ville Syrjälä [Tue, 30 Aug 2022 21:24:36 +0000 (00:24 +0300)]
drm/i915: Allow more varied alternate fixed modes for panels
On some systems the panel reports alternate modes with
different blanking periods. If the EDID reports them and VBT
doesn't tell us otherwise then I can't really see why they
should be rejected. So allow their use for the purposes of
static DRRS.
For seamless DRRS we still require a much more exact match
of course. But that logic only kicks in when selecting the
downclock mode (or in the future when determining whether
we can do a seamless refresh rate change due to a user
modeset).
Ville Syrjälä [Mon, 29 Aug 2022 13:58:34 +0000 (16:58 +0300)]
drm/i915/bios: Copy the whole MIPI sequence block
Turns out the MIPI sequence block version number and
new block size fields are considered part of the block
header and are not included in the reported new block size
field itself. Bump up the block size appropriately so that
we'll copy over the last five bytes of the block as well.
For this particular machine those last five bytes included
parts of the GPIO op for the backlight on sequence, causing
the backlight no longer to turn back on:
Sequence 6 - MIPI_SEQ_BACKLIGHT_ON
Delay: 20000 us
- GPIO index 0, number 0, set 0 (0x00)
+ GPIO index 1, number 70, set 1 (0x01)
Jani Nikula [Tue, 30 Aug 2022 10:28:00 +0000 (13:28 +0300)]
drm/i915/gmbus: stop using implicit dev_priv in register definitions
Since the beginning of time, we've implicitly assumed dev_priv is
present as a local variable in many places. We've gone a long way in
removing many of them, but the register macro definitions are the last
holdout. Remove them from the gmbus macros.
Jani Nikula [Mon, 29 Aug 2022 13:18:22 +0000 (16:18 +0300)]
drm/i915/quirks: abstract quirks further by making quirk ids an enum
Turn the quirk ids to enums instead of bits, and hide the masking inside
intel_quirks.c. Define the enums in intel_quirks.h to declutter
i915_drv.h while at it.
Jani Nikula [Mon, 29 Aug 2022 11:44:38 +0000 (14:44 +0300)]
Merge drm/drm-next into drm-intel-next
Sync drm-intel-next with v6.0-rc as well as recent drm-intel-gt-next.
Since drm-next does not have commit f0c70d41e4e8 ("drm/i915/guc: remove
runtime info printing from time stamp logging") yet, only
drm-intel-gt-next, will need to do that as part of the merge here to
build.
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
Jani Nikula [Wed, 24 Aug 2022 13:15:52 +0000 (16:15 +0300)]
drm/i915/vrr: drop window2_delay member from i915
The window2_delay member has been functionally unused (always set to 0)
since it was added in commit bb265dbdf38d ("drm/i915/xelpd: Add VRR
guardband for VRR CTL"). Replace it with a FIXME comment.
projects / thirdparty / linux.git / log
