Ville Syrjälä [Mon, 6 May 2024 12:57:10 +0000 (15:57 +0300)]
drm/i915: Split gen2 vs. gen3 .max_stride()
Plane .max_stride() is already a vfunc so having one made
up of two branches based on the display version is silly.
Split i9xx_plane_max_stride() into gen2 vs. gen3 variants
so that we get rid of said check.
Ville Syrjälä [Mon, 6 May 2024 18:33:31 +0000 (21:33 +0300)]
drm/xe: Nuke xe's copy of intel_fbdev_fb.h
For some reason xe and i915 each have an identical (fortunately)
copy of intel_fbdev_fb.h. The xe copy actually only gets included
by xe's intel_fbdev_fb.c, and the i915 copy by everyone else,
include intel_fbdev.c which is the actual caller of the
functions declared in the header.
This means the xe and i915 headers are free to define/declare
completely incompatible things and the build would still succeed
as long as the symbol names match.
That is not a good thing, so let's nuke xe's copy of the header
so that everyone will use the same header, and be forced to
agree on the same API/ABI.
Unfortunately the block has two definitions, with the cutoff
supposedly happening on ICL vs. TGL. Also according to some
notes it might be that the VBIOS (if that's still a thing)
still uses the old definition even on TGL+. Quite the mess.
Define the contents of the obsolete VBT block 55 (Compression
Parameters).
This was some early attempt at defining the compression
parameters. However the spec says:
"This block is obsolete and should not be consumed for any
compression programming."
Block 56 is the replacement that should actually be used.
So let's just name the obsolete old block but not even
bother defining the contents.
This was some easly attempt at a MIPI DSI stuff. I'm not sure
this was ever actually used (I certainly don't have any VBTs
with this block), but here's some kind of definition for it
anyway.
Define the contents of VBT block 57 (Vswing PreEmphasis Table).
The contents is highly platform specific. The columns of the
table corresponding to some set of PHY/etc registers. The rows
corresponding to all legal vswing+pre-emphasis combinations
(ie. should be 10 rows in each table). And each table
corresponds to a platform specific (mostly undocumented)
mapping based on link rate/eDP low-vswing/etc. parameters.
Define the contents of VBT block 25 (SDVO LVDS PPS).
Not 100% sure about the order of the fields as this is not
documented in the VBT spec anymore, but this order matches
what is included as part of the power sequencing SDVO commands
(struct sdvo_panel_power_sequencing). Also the real world
VBT data I have looks OK with this definition.
Define the contents of VBT block 21 (EFP List). Specs are nowhere
to be found, but real world data suggests that each entry is just
the first four bytes of the EDID PnP ID structure.
Define the contenst is VBT blocks 19,30,32 (Display Configuration
Removal Table) contents. There are three variants of this block:
pre-IVB, IVB, HSW+, with each having slightly different entries.
Curiously many HSW/BDW machines seem to have both the IVB and HSW+
variants in their VBTs simultanously. No idea why.
Define the contenst is VBT blocks 16,19,31 (Toggle List).
There are three variants of this block: pre-IVB, IVB, HSW+,
with each having slightly different entries.
Curiously many HSW/BDW machines seem to have both the IVB and
HSW+ variants in their VBTs simultanously. No idea why.
Ville Syrjälä [Fri, 3 May 2024 12:24:30 +0000 (15:24 +0300)]
drm/i915/bios: Define ALM only VBT block 9 contents
For some reason ALM VBT has two dot clock override tables.
One as the normal block 15 and a second one as block 9.
The table in block 9 has no row_size/num_rows information.
On my Fujitsu Lifebook S6010 only the block 9 table has actual
data in it. Block 15 is present but all zeroes.
Define the contents of VBT block 15 (Dot Clock Override Table)
The contents were reverse engineered by intuition. The gen2 stuff
seems solid as I can verify that against real world VBT data. The
gen3 stuff less so as all the gen3+ VBTs I have just filla the
entire block with zeroes.
Define the contents of VBT block 10 (Mode Removal Table).
There seem to be two variants:
- 8 byte entries for desktop systems
- 10 byte entries for mobile systems, with the extra
panel_flags being a bitmask of LFPs
It seems starting from HSW only the mobile variant is
used anymore.
Define the contents for VBT blocks:
- Block 6 (Extended MMIO Register Table)
- Block 7 (IO Software Flag Table)
- Block 8 (MMIO SWF Register Table)
All of these use the same basic layout, with two known variants:
- data_access_size==0xce -> offset,value tuples are u8,u8
- data_access_size==0x02 -> offset,value tuples are u32,u32
Define the contents of VBT block 5 (Generic Mode Table).
Details were mostly gleaned from some VBIOS sources.
There are apparently two variants of the block: ALM only
vs. MGM, defined here as bdb_generic_mode_table_alm
and bdb_generic_mode_table_mgm. And those are the only two
platforms where I've seen this block.
Ville Syrjälä [Fri, 3 May 2024 12:24:24 +0000 (15:24 +0300)]
drm/i915/bios: Define VBT block 4 (Mode Support List) contents
Define the contents of VBT block 4 (Mode Support List).
Slightly crazy layout with a variable length list at the start,
followed by the length of said list.
No real idea what these "Intel mode numbers" really are. What
I see in real world VBTs seems to be always the same list of
26 numbers, ranging between 0x30 and 0x84.
Ville Syrjälä [Fri, 3 May 2024 12:24:20 +0000 (15:24 +0300)]
drm/i915/bios: Define "TV" child device handle
Child device 0x2 used to be "TV" until redefined to mean
EFP5 in version 215. Add a define for the old meaning as well.
Technically it was probably deprecated a lot before version
215 since native TV encoders were last seen on CTG, and SDVO
was fully gone by HSW. So something like "???-164" might also
be a reasonable way to document this, but no real harm in
saying "???-214" since nothing else presumably occupied that
bit in the meantime.
Ville Syrjälä [Fri, 3 May 2024 12:24:16 +0000 (15:24 +0300)]
drm/i915/bios: Remove version number comment from DEVICE_HANDLE_EFP4
DEVICE_HANDLE_EFP4 has actually been in use since the very beginning,
or at least something has been occupying that bit because old
VBTs actually use it, and it definitely looks to be about external
displays given how its used. So let's ignore what the current spec
claims and remove the misleading version number comment.
Karthikeyan Ramasubramanian [Thu, 22 Feb 2024 01:06:24 +0000 (18:06 -0700)]
drm/i915/bios: Fix parsing backlight BDB data
Starting BDB version 239, hdr_dpcd_refresh_timeout is introduced to
backlight BDB data. Commit 700034566d68 ("drm/i915/bios: Define more BDB
contents") updated the backlight BDB data accordingly. This broke the
parsing of backlight BDB data in VBT for versions 236 - 238 (both
inclusive) and hence the backlight controls are not responding on units
with the concerned BDB version.
backlight_control information has been present in backlight BDB data
from at least BDB version 191 onwards, if not before. Hence this patch
extracts the backlight_control information for BDB version 191 or newer.
Tested on Chromebooks using Jasperlake SoC (reports bdb->version = 236).
Tested on Chromebooks using Raptorlake SoC (reports bdb->version = 251).
v2: removed checking the block size of the backlight BDB data
[vsyrjala: this is completely safe thanks to commit e163cfb4c96d
("drm/i915/bios: Make copies of VBT data blocks")]
Fixes: 700034566d68 ("drm/i915/bios: Define more BDB contents") Cc: stable@vger.kernel.org Cc: Jani Nikula <jani.nikula@intel.com> Cc: Ville Syrjälä <ville.syrjala@linux.intel.com> Signed-off-by: Karthikeyan Ramasubramanian <kramasub@chromium.org> Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com> Link: https://patchwork.freedesktop.org/patch/msgid/20240221180622.v2.1.I0690aa3e96a83a43b3fc33f50395d334b2981826@changeid Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Matthew Auld [Tue, 30 Apr 2024 17:28:49 +0000 (10:28 -0700)]
drm/i915/display: perform transient flush
Perform manual transient cache flush prior to flip and at the end of
frontbuffer_flush. This is needed to ensure display engine doesn't see
garbage if the surface is L3:XD dirty.
Testcase: igt@xe-pat@display-vs-wb-transient Signed-off-by: Matthew Auld <matthew.auld@intel.com> Signed-off-by: Balasubramani Vivekanandan <balasubramani.vivekanandan@intel.com> Acked-by: Nirmoy Das <nirmoy.das@intel.com> Reviewed-by: Matt Roper <matthew.d.roper@intel.com> Signed-off-by: Radhakrishna Sripada <radhakrishna.sripada@intel.com> Link: https://patchwork.freedesktop.org/patch/msgid/20240430172850.1881525-19-radhakrishna.sripada@intel.com
Nirmoy Das [Tue, 30 Apr 2024 17:28:48 +0000 (10:28 -0700)]
drm/xe/device: implement transient flush
Display surfaces can be tagged as transient by mapping it using one of
the various L3:XD PAT index modes on Xe2. The expectation is that KMD
needs to request transient data flush at the start of flip sequence to
ensure all transient data in L3 cache is flushed to memory. Add a
routine for this which we can then call from the display code.
v2: rebase(RK)
Signed-off-by: Nirmoy Das <nirmoy.das@intel.com> Co-developed-by: Matthew Auld <matthew.auld@intel.com> Signed-off-by: Matthew Auld <matthew.auld@intel.com> Signed-off-by: Balasubramani Vivekanandan <balasubramani.vivekanandan@intel.com> Reviewed-by: Matt Roper <matthew.d.roper@intel.com> Signed-off-by: Radhakrishna Sripada <radhakrishna.sripada@intel.com> Acked-by: Lucas De Marchi <lucas.demarchi@intel.com> Link: https://patchwork.freedesktop.org/patch/msgid/20240430172850.1881525-18-radhakrishna.sripada@intel.com
Revert "drm/i915/dgfx: DGFX uses direct VBT pin mapping"
This reverts commit 562f33836f519a235e5c5e71bcc723ab1faccd2f.
For BMG it seems that the VBT to DDI mapping does not follow DG1, and
DG2, but follows ADLP mapping given in Bspec:20124.
Matt Roper [Tue, 30 Apr 2024 17:28:42 +0000 (10:28 -0700)]
drm/i915/xe2hpd: Add max memory bandwidth algorithm
Unlike DG2, Xe2_HPD does support multiple GV points with different
maximum memory bandwidths, but uses a much simpler algorithm than igpu
platforms use.
drm/i915/xe2hpd: Add support for eDP PLL configuration
Tables for eDP PHY PLL configuration for different link rates added for
Xe2_HPD. Previous platforms were using C10 PHY for eDP port whereas
Xe2_HPD has C20 PHY.
Xe2_HPD has different offsets for C20 PHY SRAM configuration context
location. Use the display version to select the right address.
Note that Xe2_LPD uses the same C20 SRAM offsets used by Xe_LPDP (i.e.
MTL's display). According to the BSpec, currently, only Xe2_HPD has
different offsets, so make sure it is the only display using them in the
driver.
v2:
* Redesigned how the right offsets are selected for different display
IP versions.
v3: Fix white space error(RK)
Bspec: 67610 Cc: Clint Taylor <Clinton.A.Taylor@intel.com> Cc: Gustavo Sousa <gustavo.sousa@intel.com> Cc: Jani Nikula <jani.nikula@linux.intel.com> Signed-off-by: Balasubramani Vivekanandan <balasubramani.vivekanandan@intel.com> Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com> Signed-off-by: Radhakrishna Sripada <radhakrishna.sripada@intel.com> Reviewed-by: Radhakrishna Sripada <radhakrishna.sripada@intel.com> Link: https://patchwork.freedesktop.org/patch/msgid/20240430172850.1881525-7-radhakrishna.sripada@intel.com
Discrete cards use the Port numbers TC1-4 for the offsets. The regular
flow for type-c subsystem port initialization can be skipped. This check
is present in DG2. Extend this to future discrete products.
Display code uses IS_BATTLEMAGE macro but the platform support doesn't
exist in i915. So fake IS_BATTLEMAGE macro defined to enable building
i915 code. We should make sure the macro parameter is used in the
always-false expression so that we don't run into "unused variable"
warnings from i915 builds if the IS_BATTLEMAGE() check is the only place
the i915 pointer gets used in a function.
While we're at it, also update the IS_LUNARLAKE macro to include the
parameter in the false expression for consistency.
Mika Kahola [Thu, 2 May 2024 13:17:16 +0000 (16:17 +0300)]
drm/i915/display: Calculate crtc clock rate based on PLL parameters
With HDMI monitors we bumped up a case where the crtc clock rate
caused a mismatch on state verification. This was due to
assumption that the SW clock rate from PLL structure would match
the calculated counterpart from HW. This is not necessarily always
the case and therefore we would actually need to recalculate the
clock rate from SW PLL parameters. Then these SW and HW crtc clock
rates can be compared with each other.
The patch recalculates the crtc clock rate for SW state based on
SW PLL parameters and compares the crtc clock rate calculated
from the parameters found from the HW.
Rename need_async_flip_disable_wa to need_async_flip_toggle_wa to
better reflect the fact that we need to deal with the bad
PLANE_CTL_ASYNC_FLIP double buffering behaviour going both
ways.
Ville Syrjälä [Tue, 30 Apr 2024 09:56:38 +0000 (12:56 +0300)]
drm/i915: Eliminate extra frame from skl-glk sync->async flip change
On bdw-glk the sync->async flip change takes an extra frame due to
the double buffering behaviour of the async flip plane control bit.
Since on skl+ we are now explicitly converting the first async flip
to a sync flip (in order to allow changing the modifier and/or
ddb/watermarks) we are now taking two extra frames until async flips
are actually active. We can drop that back down to one frame by
setting the async flip bit already during the sync flip.
Note that on bdw we don't currently do the extra sync flip (see
intel_plane_do_async_flip()) so technically we wouldn't have
to deal with this in i9xx_plane_update_arm(). But I added the
relevant snippet of code there as well, just in case we ever
decide to go for the extra sync flip on pre-skl platforms as
well (we might, for example, want to change the fb stride).
Ville Syrjälä [Tue, 30 Apr 2024 09:56:37 +0000 (12:56 +0300)]
drm/i915: Allow the initial async flip to change modifier
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.
Ville Syrjälä [Tue, 30 Apr 2024 09:56:36 +0000 (12:56 +0300)]
drm/i915: Reject async flips if we need to change DDB/watermarks
DDB/watermarks are always double buffered on the vblank, so we
can't safely change them during async flips. Currently this never
happens, but we'll be making changing between sync and async
flips a bit more flexible, in which case we can actually end up
here.
Ville Syrjälä [Tue, 30 Apr 2024 09:56:35 +0000 (12:56 +0300)]
drm/i915: Align PLANE_SURF to 16k on ADL for async flips
On ADL async flips apparently generate DMAR and GGTT faults
(with accompanying visual glitches) unless PLANE_SURF is
aligned to at least 16k. Bump up the alignment to 16k.
TODO: analyze things better to figure out what is really
going on here
projects / users / jedix / linux-maple.git / log