Refactor the logic to get the number of joined pipes. Start with a single
pipe and incrementally try additional pipes only if needed. While DSC
overhead is not yet computed here, this restructuring prepares the code to
support that in follow-up changes.
v2:
- Remove fallback in case force-joiner configuration fails. (Imre)
- Drop redundant MODE_OK assignment (Imre)
v3:
- Align with the changes in intel_dp_mode_valid(). (Imre)
v4:
- Set MODE_CLOCK_HIGH on DSC/rate failures aligning with SST case. (Imre)
Signed-off-by: Ankit Nautiyal <ankit.k.nautiyal@intel.com>
Reviewed-by: Imre Deak <imre.deak@intel.com>
Link: https://patch.msgid.link/20260202103731.357416-9-ankit.k.nautiyal@intel.com
Move the calculation of num_joined_pipes and other constraints that
depend on it, into a single block in mst_stream_compute_config().
This groups all joiner-dependent logic together, preparing the code for a
future loop-based evaluation of multiple joiner configurations.
Signed-off-by: Ankit Nautiyal <ankit.k.nautiyal@intel.com>
Reviewed-by: Imre Deak <imre.deak@intel.com>
Link: https://patch.msgid.link/20260202103731.357416-8-ankit.k.nautiyal@intel.com
Refactor the mode_valid to have all joiner dependent stuff together and
place the check for dotclock limit at the very end.
This will help in the following refactor to iterate over the joiner
candidates and find the best joiner candidate that satisfy all checks
and limits.
v2: Update status to MODE_CLOCK_HIGH if max_dotclock check fails. (Imre)
v3: Drop redundant MODE_OK assignment and return. (Imre)
Signed-off-by: Ankit Nautiyal <ankit.k.nautiyal@intel.com>
Reviewed-by: Imre Deak <imre.deak@intel.com>
Link: https://patch.msgid.link/20260202103731.357416-7-ankit.k.nautiyal@intel.com
Currently, the number of joined pipes are determined early in the flow,
which limits flexibility for accounting DSC slice overhead. To address
this, recompute the joined pipe count during DSC configuration.
Refactor intel_dp_dsc_compute_config() to iterate over joiner candidates
and select the minimal joiner configuration that satisfies the mode
requirements. This prepares the logic for future changes that will
consider DSC slice overhead.
v2:
- Rename helper to intel_dp_compute_link_for_joined_pipes(). (Imre)
- Move the check for max dotclock inside the helper so that if dotclock
check fails for non DSC case for a given number of joined pipes, we
are able to fallback to the DSC mode. (Imre)
v3:
- Drop fallback to other joiner configurations, if the force joiner
configuration fails. (Imre)
- Check for maxdotclock limit for non-DSC case first and fall back to
DSC if the check fails. (Imre)
- Initialize ret to -EINVAL to handle case where we bail out early.
(Imre)
Signed-off-by: Ankit Nautiyal <ankit.k.nautiyal@intel.com>
Reviewed-by: Imre Deak <imre.deak@intel.com>
Link: https://patch.msgid.link/20260202103731.357416-6-ankit.k.nautiyal@intel.com
Currently in intel_dp_mode_valid(), we compute the number of joined pipes
required before deciding whether DSC is needed. This ordering prevents us
from accounting for DSC-related overhead when determining pipe
requirements.
It is not possible to first decide whether DSC is needed and then compute
the required number of joined pipes, because the two depend on each other:
- DSC need is a function of the pipe count (e.g., 4‑pipe always requires
DSC; 2‑pipe may require it if uncompressed joiner is unavailable).
- Whether a given pipe‑join configuration is sufficient depends on
effective bandwidth, which itself changes when DSC is used.
As a result, the only correct approach is to iterate candidate pipe counts.
So, refactor the logic to start with a single pipe and incrementally try
additional pipes only if needed. While DSC overhead is not yet computed
here, this restructuring prepares the code to support that in a follow-up
changes.
If a forced joiner configuration is present, we just check for that
configuration. If it fails, we bailout and return instead of trying with
other joiner configurations.
v2:
- Iterate over number of pipes to be joined instead of joiner
candidates. (Jani)
- Document the rationale of iterating over number of joined pipes.
(Imre)
v3:
- In case the force joiner configuration doesn't work, do not fallback
to the normal routine, bailout instead of trying other joiner
configurations. (Imre)
v4:
- Use num_joined_pipes instead of num_pipes. (Imre)
- Inititialize status before the loops starts. (Imre)
v5:
- Fix status handling by setting MODE_CLOCK_HIGH on DSC/rate failures.
(Imre)
Signed-off-by: Ankit Nautiyal <ankit.k.nautiyal@intel.com>
Reviewed-by: Imre Deak <imre.deak@intel.com>
Link: https://patch.msgid.link/20260202103731.357416-5-ankit.k.nautiyal@intel.com
Separate out function to get the hdisplay limit for a given platform.
v2: Rename the helper to intel_dp_max_hdisplay_per_pipe(). (Imre)
Signed-off-by: Ankit Nautiyal <ankit.k.nautiyal@intel.com>
Reviewed-by: Imre Deak <imre.deak@intel.com>
Link: https://patch.msgid.link/20260202103731.357416-4-ankit.k.nautiyal@intel.com
Move the calculation of `num_joined_pipes` and other constraints that
depend on it, into a single block.
This groups all joiner-dependent logic together, preparing the code for a
future loop-based evaluation of multiple joiner configurations.
Signed-off-by: Ankit Nautiyal <ankit.k.nautiyal@intel.com>
Reviewed-by: Imre Deak <imre.deak@intel.com>
Link: https://patch.msgid.link/20260202103731.357416-3-ankit.k.nautiyal@intel.com
Add a meaningful return to intel_dp_read_dsc_dpcd. This is to avoid
unwanted DPCD reads which are not needed once we know DSC DPCD
read fails. To do this convert drm_dp_dpcd_read to drm_dp_dpcd_read_data
which returns a meaningful error which can be propogated up when all
bits are not read.
While we are at it convert the drm_err in intel_dp_read_dsc_dpcd to
drm_dbg_kms. This is because we do not want a hard ERROR when we
call this function, during the intel_dp_detect phase since AUX may not
be up, it is expected to fail but we do not expect a failure in read
when we call intel_dp_dsc_get_sink_cap so we move the drm_err there.
Signed-off-by: Suraj Kandpal <suraj.kandpal@intel.com>
Reviewed-by: Ankit Nautiyal <ankit.k.nautiyal@intel.com>
Link: https://patch.msgid.link/20260202081801.1400708-1-suraj.kandpal@intel.com
Pixel normalizer is enabled with normalization factor as 1.0 for
FP16 formats in order to support FBC for those formats in xe3p_lpd.
Previously pixel normalizer gets disabled during the plane disable
routine. But there could be plane format settings without explicitly
calling the plane disable in-between and we could endup keeping the
pixel normalizer enabled for formats which we don't require that.
This is causing crc mismatches in yuv formats and FIFO underruns in
planar formats like NV12. Fix this by updating the pixel normalizer
configuration based on the pixel formats explicitly during the plane
settings arm calls itself - enable it for FP16 and disable it for
other formats in HDR capable planes.
v2: avoid redundant pixel normalization setting updates
v3: moved the normalization factor definition to intel_fbc.c and some
updates to comments
v4: simplified the pixel normalizer setting handling
Fixes: 5298eea7ed ("drm/i915/xe3p_lpd: use pixel normalizer for fp16 formats for FBC")
Signed-off-by: Vinod Govindapillai <vinod.govindapillai@intel.com>
Reviewed-by: Uma Shankar <uma.shankar@intel.com>
Link: https://patch.msgid.link/20260130095919.107805-1-vinod.govindapillai@intel.com
Now we have everything in place for triggering PSR "frame change" event
using TRANS_PUSH: use TRANS_PUSH for LunarLake and onwards.
v4:
- Added call to intel_vrr_psr_frame_change_enable call
- added setting LNL_TRANS_PUSH_PSR_PR_EN into intel_vrr_send_push
v3: use HAS_PSR_FRAME_CHANGE macro
v2: use AND instead of OR in intel_psr_use_trans_push
Signed-off-by: Jouni Högander <jouni.hogander@intel.com>
Reviewed-by: Ankit Nautiyal <ankit.k.nautiyal@intel.com>
Link: https://patch.msgid.link/20260127124120.1069026-12-jouni.hogander@intel.com
Add a macro indicating that the platform supports triggering a Frame
Change event for the PSR HW using the TRANS PUSH mechanism.
Signed-off-by: Jouni Högander <jouni.hogander@intel.com>
Reviewed-by: Ankit Nautiyal <ankit.k.nautiyal@intel.com>
Link: https://patch.msgid.link/20260127124120.1069026-11-jouni.hogander@intel.com
We need to use intel_psr_exit in frontbuffer flush on LunarLake and
onwards if we want to move using trans push mechanism to trigger Frame
Change event.
Keep PSR1 and PSR2 HW tracking as it is for older platforms as this was
seen causing problems there.
v2: typo fixed
Signed-off-by: Jouni Högander <jouni.hogander@intel.com>
Reviewed-by: Ankit Nautiyal <ankit.k.nautiyal@intel.com>
Link: https://patch.msgid.link/20260127124120.1069026-10-jouni.hogander@intel.com
We are planning to move using trans push mechanism to trigger the Frame
Change event. In that case we can't wait PSR to idle before send push
happens. Due to this move wait for idle to be done after possible send push
is done.
This should be ok for Frame Change event triggered by register write as
well. Wait for idle is needed only for corner case where PSR is
transitioning into DEEP_SLEEP when Frame Change event is triggered. It just
has to be before wait for vblank. Otherwise we may have vblank before PSR
enters DEEP_SLEEP and still using old frame buffers for first frame after
wake up.
v2: some typos fixed
Signed-off-by: Jouni Högander <jouni.hogander@intel.com>
Reviewed-by: Ankit Nautiyal <ankit.k.nautiyal@intel.com>
Link: https://patch.msgid.link/20260127124120.1069026-9-jouni.hogander@intel.com
In case PSR uses trans push as a "frame change" event and we need to wait
vblank after triggering PSR "frame change" event. Otherwise we may miss
selective updates.
DSB skips all waits while PSR is active. Check push send is skipped as well
because trans push send bit is not reset by the HW if VRR is not enabled ->
we may start configuring new selective update while previous is not
complete. Avoid this by waiting for vblank after sending trans push.
v2: some typos fixed
Signed-off-by: Jouni Högander <jouni.hogander@intel.com>
Reviewed-by: Ankit Nautiyal <ankit.k.nautiyal@intel.com>
Link: https://patch.msgid.link/20260127124120.1069026-8-jouni.hogander@intel.com
On LunarLake we are using TRANS_PUSH mechanism to trigger "Frame Change"
event. This way we have more control on when PSR HW is woken up. I.e. not
every display register write is triggering sending update. This allows us
setting DSB_SKIP_WAITS_EN chicken bit as well.
Signed-off-by: Jouni Högander <jouni.hogander@intel.com>
Reviewed-by: Ankit Nautiyal <ankit.k.nautiyal@intel.com>
Link: https://patch.msgid.link/20260127124120.1069026-7-jouni.hogander@intel.com
On Lunarlake and onward it is possible to generate PSR Frame Change
event using TRANS_PUSH mechanism:
Ensure push is performed on flip if needed by PSR.
Don't use intel_psr_trigger_frame_change_event if PSR uses trans push as
Frame Change event: When not using trans push to generate Frame Change
event plane/pipe register writes are triggering the event. This is causing
problems when transiting to PSR active/idle and the event is coming in. To
tackle this we are waiting PSR to idle and using
intel_psr_trigger_frame_change_event to ensure PSR exit. If the Frame
change event is triggered after everything is configured (on trans push) we
don't need to worry about all this.
Ensure possible intel_vrr_tg_disable call is not overwriting
LNL_TRANS_PUSH_PSR_PR_EN.
v8:
- split adding intel_vrr_psr_frame_change_enable as a separate change
- update commit message
v7:
- HAS_PSR_FRAME_CHANGE macro moved to separate patch and renamed as
HAS_PSR_TRANS_PUSH_FRAME_CHANGE
- use intel_psr_use_trans_push instead of HAS_PSR_FRAME_CHANGE in
intel_psr_trigger_frame_change
- moved calling intel_vrr_psr_frame_change_enable away from this patch
v6:
- add HAS_PSR_FRAME_CHANGE macro
- use TRANS_PUSH in instead of TRAN_VRR_CTL
v5: use intel_psr_use_trans_push for intel_vrr_psr_frame_change_enable
v4:
- use rmw when enabling/disabling transcoder
- set TRANS_PUSH_EN conditionally in intel_vrr_send_push
- do not call intel_vrr_send_push from intel_psr_trigger_frame_change
- do not enable using TRANS_PUSH mechanism for PSR "Frame Change"
v3:
- use rmw when enabling/disabling
- keep LNL_TRANS_PUSH_PSR_PR_EN set always on LunarLake and onwards
v2: use intel_vrr_trans_push_enabled_set_clear instead of rmw
Signed-off-by: Jouni Högander <jouni.hogander@intel.com>
Reviewed-by: Ankit Nautiyal <ankit.k.nautiyal@intel.com>
Link: https://patch.msgid.link/20260127124120.1069026-6-jouni.hogander@intel.com
Add new helper to enable PSR Frame Change event generation using trans push
mechanism. This helper is to be used by PSR code on PSR enable.
Signed-off-by: Jouni Högander <jouni.hogander@intel.com>
Reviewed-by: Ankit Nautiyal <ankit.k.nautiyal@intel.com>
Link: https://patch.msgid.link/20260127124120.1069026-5-jouni.hogander@intel.com
On Lunarlake and onwards it is possible to generate PSR "frame change"
event using TRANS_PUSH mechanism. As a preparation add new helper to parse
value to be written into TRANS_PUSH register. Setting
LNL_TRANS_PUSH_PSR_PR_EN is done in upcoming change.
Signed-off-by: Jouni Högander <jouni.hogander@intel.com>
Reviewed-by: Ankit Nautiyal <ankit.k.nautiyal@intel.com>
Link: https://patch.msgid.link/20260127124120.1069026-4-jouni.hogander@intel.com
This is a preparation to start using trans push as a PSR "Frame Change"
event. It adds intel_psr_use_trans_push placeholder which return false for
now until we have everything in place.
v2:
- modify commit message
- add TODO
Signed-off-by: Jouni Högander <jouni.hogander@intel.com>
Reviewed-by: Ankit Nautiyal <ankit.k.nautiyal@intel.com>
Link: https://patch.msgid.link/20260127124120.1069026-3-jouni.hogander@intel.com
Both initialisation and removal of GVT happen at different abstraction
levels. Hence caller of i915_driver_hw_probe() has no way of knowing
status of intel_gvt_init(). This can lead to an unbalanced number of
calls of intel_gvt_init() and intel_gvt_driver_remove() since GVT error
path is currently handled in i915_driver_probe(). One such scenario has
been seen with i915_driver_hw_probe() fault injection, which caused
double entry deletion and list corruption.
Move intel_gvt_init() up to i915_driver_probe(). Add out_cleanup_gvt
error path for removing gvt. Trigger it only after intel_gvt_init()
succeeded.
In case intel_gvt_init() failed, theoretically we should follow err_msi
error path. That is actually impossible since call to intel_gvt_init()
unconditionally returns 0, although it claims to return negative error
code on failure. Thus follow standard out_cleanup_hw error path on a
hypothetical future intel_gvt_init() failure. Remove err_msi label from
i915_driver_hw_probe() since intel_gvt_init() was the only user of it.
Changelog:
v1->v2
- don't move err_msi error path from i915_driver_hw_probe (Jani)
- rewrite commit message
Closes: https://gitlab.freedesktop.org/drm/i915/kernel/-/issues/15481
Signed-off-by: Michał Grzelak <michal.grzelak@intel.com>
Reviewed-by: Andi Shyti <andi.shyti@linux.intel.com>
Link: https://patch.msgid.link/20260129090129.2601661-2-michal.grzelak@intel.com
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
Avoid even accidental use of display details by dropping the include of
intel_display_types.h. We'll still have to include intel_dpll_mgr.h for
the DPLL IDs, but at least the surface is smaller.
Add duplicate defines of pipe_name() and port_name() to avoid depending
on display. They're trivial enough to be acceptable.
Cc: Ankit Nautiyal <ankit.k.nautiyal@intel.com>
Reviewed-by: Ankit Nautiyal <ankit.k.nautiyal@intel.com>
Link: https://patch.msgid.link/2fa5677f5ff3dbeaa75a7984d74fb9855a4ba3d2.1769612208.git.jani.nikula@intel.com
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
Move enum aux_ch from intel_display.h to intel_display_limits.h to avoid
having to include the former from gvt. This currently happens via
intel_display_types.h, which we can also drop in favour of
intel_display_limits.h in intel_gvt_mmio_table.c.
Cc: Ankit Nautiyal <ankit.k.nautiyal@intel.com>
Reviewed-by: Ankit Nautiyal <ankit.k.nautiyal@intel.com>
Reviewed-by: Suraj Kandpal <suraj.kandpal@intel.com>
Link: https://patch.msgid.link/86ccc91ee1ca96fa145b750e2f1935918ec288d0.1769612208.git.jani.nikula@intel.com
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
Xe3p_LPD also needs Wa_13012396614. Update the conditions for that
workaround accordingly.
Signed-off-by: Gustavo Sousa <gustavo.sousa@intel.com>
Reviewed-by: Mika Kahola <mika.kahola@intel.com>
Link: https://patch.msgid.link/20260122-wa_13012396614-xe3p_lpd-v1-1-39d34a3a5b77@intel.com
Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
Static analysis issue:
intel_conn_to_vcpi currently checks and asserts vcpi >= 0, as grabbed
from payload->vcpi. But payload->vcpi is a u8, so this is always true.
Remove this unnecessary check.
While we're here, the jump statements in this function are largely
defunct now, so remove those as well.
Signed-off-by: Jonathan Cavitt <jonathan.cavitt@intel.com>
Cc: Matt Roper <matthew.d.roper@intel.com>
Reviewed-by: Matt Roper <matthew.d.roper@intel.com>
Link: https://patch.msgid.link/20260123170914.64548-2-jonathan.cavitt@intel.com
Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
Call the parent driver pcode functions through the parent interface
function pointers instead of expecting both to have functions of the
same name.
In i915, add the interface to existing intel_pcode.[ch], while in xe
move them to new display/xe_display_pcode.[ch] and build it only for
CONFIG_DRM_XE_DISPLAY=y.
Do not add separate write and write_timeout calls in the
interface. Instead, handle the default 1 ms timeout in the
intel_parent.c glue layer.
This drops the last intel_pcode.h includes from display, and allows us
to remove the corresponding xe compat header.
v2: initialize .pcode in i915
Reviewed-by: Michał Grzelak <michal.grzelak@intel.com>
Link: https://patch.msgid.link/20260126112925.2452171-1-jani.nikula@intel.com
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
Rename the non static intel_clear_response_ready_flag to
intel_cx0_clear_response_ready_flag so that we follow the
naming standards of non static function.
Signed-off-by: Suraj Kandpal <suraj.kandpal@intel.com>
Reviewed-by: Nemesa Garg <nemesa.garg@intel.com>
Link: https://patch.msgid.link/20260122044859.753682-2-suraj.kandpal@intel.com
Clear the response ready and error bit of PORT_P2M_MESSAGE_BUS_STATUS
before writing the transaction pending bit of
PORT_M2P_MSGBUS_CTL as that is a hard requirement. If not done
we find that the PHY hangs since it ends up in a weird state if left
idle for more than 1 hour.
Bspec: 65101
Signed-off-by: Suraj Kandpal <suraj.kandpal@intel.com>
Reviewed-by: Gustavo Sousa <gustavo.sousa@intel.com>
Reviewed-by: Michał Grzelak <michal.grzelak@intel.com>
Link: https://patch.msgid.link/20260122044859.753682-1-suraj.kandpal@intel.com
Move the DSB buffer handling to the display parent interface, making
display more independent of i915 and xe driver implementations.
Since the DSB parent interface is only called from intel_dsb.c, add the
wrappers there with smaller visibility instead of the usual
intel_parent.[ch], and using struct intel_dsb as the context parameter
for convenience.
Unfortunately, memset() being a macro in linux/fortify-string.h, we
can't use that as the function pointer name. dsb->memset() would be
using the macro and leading to build failures. Therefore, use .fill()
for the memset() functionality.
v2: s/memset/fill/
Reviewed-by: Michał Grzelak <michal.grzelak@intel.com>
Link: https://patch.msgid.link/df117c862a6d34dae340e4a85c2482b4e29c8884.1768923917.git.jani.nikula@intel.com
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
The error interrupt register GEN7_ERR_INT is a display irq
register. Move its GPU error capture to display irq snapshot.
Cc: Uma Shankar <uma.shankar@intel.com>
Reviewed-by: Uma Shankar <uma.shankar@intel.com>
Link: https://patch.msgid.link/20260122123759.1701492-1-jani.nikula@intel.com
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
The actual GMCH_CRTL lives in the host bridge aka. device 0,
but device 2 has a read-only mirror on i85x/i865+. Document
that fact.
Also remove the ancient tales about where the defines are used.
Those haven't been true in a long time.
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patch.msgid.link/20251208182637.334-20-ville.syrjala@linux.intel.com
Reviewed-by: Jani Nikula <jani.nikula@intel.com>
The PCI config space register defines in i915_drm.h are
a bit of a mess; Whitespace is all over the place, register
masks and values are defined in inconsistent ways.
Clean it up a bit.
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patch.msgid.link/20251208182637.334-19-ville.syrjala@linux.intel.com
Reviewed-by: Jani Nikula <jani.nikula@intel.com>
Use the proper IS0_R name for the VGA input status register 0, instead
of using the VGA_MIS_W alias which is meant for write accesses to the
same address. Yes, VGA registers are weird.
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patch.msgid.link/20251208182637.334-16-ville.syrjala@linux.intel.com
Reviewed-by: Jani Nikula <jani.nikula@intel.com>
Add a proper name for the "Input status register 0" IO address.
Currently we have some code that does reads using the aliasing
VGA_MSR_W define, making it unclear what register we're
actually reading.
v2: Remove stray '?'
Cc: Helge Deller <deller@gmx.de>
Cc: linux-fbdev@vger.kernel.org
Cc: dri-devel@lists.freedesktop.org
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patch.msgid.link/20251209075549.14051-1-ville.syrjala@linux.intel.com
Reviewed-by: Jani Nikula <jani.nikula@intel.com>
Acked-by: Helge Deller <deller@gmx.de>
On pre-g4x VGA registers are accessible via MMIO. Make use of
it so that we can avoid dealing with the VGA arbiter.
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patch.msgid.link/20251208182637.334-14-ville.syrjala@linux.intel.com
Reviewed-by: Jani Nikula <jani.nikula@intel.com>
VGA register are rather special since they either get accessed
via the global IO addresses, or possibly through MMIO on
pre-g4x platforms. Wrap all VGA register accesses in
intel_vga_{read,write}() to make it obvious where they get
accessed.
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patch.msgid.link/20251208182637.334-13-ville.syrjala@linux.intel.com
Reviewed-by: Jani Nikula <jani.nikula@intel.com>
intel_de_read8() is only needed for VGA register MMIO access
by the CRT code on gen2/3. Remove the redundant wakelock stuff,
and add a platform check to make sure this won't get used on
any platform where MMIO VGA register accesses don't work.
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patch.msgid.link/20251208182637.334-11-ville.syrjala@linux.intel.com
Reviewed-by: Jani Nikula <jani.nikula@intel.com>
We want our VGA register accesses to land on the correct GPU.
Check that the VGA routing is appropriately configured.
For the iGPU this just means the IO decode enable on the GPU, but
for dGPUs we also need the entire chain of bridges to forward the
VGA accesses.
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patch.msgid.link/20251208182637.334-10-ville.syrjala@linux.intel.com
Acked-by: Jani Nikula <jani.nikula@intel.com>
intel_gmch_vga_set_state() is a complete lie on ILK+ because
the GMCH_CTRL register is locked and can't actually be written.
But we still need to remove the iGPU from the VGA arbitration
on iGPU+dGPU systems, or else Xorg performance will tank due
to the constant VGA arbiter accesses.
For VGA memory decode we can't turn off the PCI_COMMAND
memory deocde as that would disable even normal MMIO.
Instead we can disable just the VGA memory decode via
the VGA MSR register. And we can do that just once
when disabling the VGA plane. That way we don't have
to touch VGA registers anywhere else.
We can also inform the arbiter that we're no longer decoding
VGA memory. This will stop the arbiter from disabling all
memory decode for the iGPU via PCI_COMMAND (and thus breaking
everything) whenever some other GPU wants to own the VGA memory
accesses.
For IO we can disable all IO decode via the PCI_COMMAND
register, except around the few VGA register accesses that
we need to do in intel_vga_disable(). Unfortunately we can't
disable IO decode permanently as it makes some laptops (eg.
Dell Latitude E5400) hang during reboot/shutdown. One option
would be to re-enable IO decode from the poweroff hooks, but
that won't help the sysrq emergency reboot/shutdown since it
won't call said hooks. So let's try to keep IO decode in its
original setting unless we really need to disable it to
exclude the GPU from VGA arbitration.
I suppose we could keep frobbing GMCH_CTRL on pre-ILK, but
it seems better to not do it since it has other side effects
such as changing the class code of the PCI device.
For discrete GPUs we'll rely on the bridge control instead.
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patch.msgid.link/20251208182637.334-9-ville.syrjala@linux.intel.com
Acked-by: Jani Nikula <jani.nikula@intel.com>
Avoid using the VGA arbiter during intel_vga_get() for iGPUs because
that will clobber the VGA routing for whatever external GPU is the
current VGA device. That will cause all reads from VGA memory to
come back as 0xff/white, and thus we get a white rectangle on screen
when the external GPU switches from vgacon to fbcon.
The iGPU has the highest VGA decode priority so it will steal all
VGA register accesses whenever its IO decoding is enabled. We'll only
keep the IO decode enabled for a short time so hopefully we don't
end up eating too many unrelated VGA register accesses.
For discrete GPUs we need all the bridges to have their VGA forwarding
bits correctly configured so we can't really avoid the VGA arbiter
there. Although we only do this stuff on dGPUs when the VGA plane was
actaully enabled, so the dGPU should be the current VGA device
and thus have VGA routed to it already anyway.
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patch.msgid.link/20251208182637.334-8-ville.syrjala@linux.intel.com
Acked-by: Jani Nikula <jani.nikula@intel.com>
Turns out at least some systems (eg. HSW Lenovo ThinkCentre E73)
configure the VGA registers even when booting in UEFI mode. So
in order to avoid any issues with the MSR register we should
clean up the VGA registers anyway.
For now this mostly avoids the potential for unclaimed register
accesses due to the power well vs. MDA/CGA selection. But this
will become more important soon as we'll start to rely on the
MSR register to control VGA memory decode as well.
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patch.msgid.link/20251208182637.334-7-ville.syrjala@linux.intel.com
Acked-by: Jani Nikula <jani.nikula@intel.com>
On some systems the BIOS will disable the VGA decode logic in the
iGPU (via GMCH_CTRL) when an external GPU is used as the primary
VGA device. In that case the iGPU will never claim any VGA register
accesses, and any access we do will in fact end up on the external
GPU. Don't go poking around in the other GPUs' registers.
Note that (at least on the g4x board where I tested this) the BIOS
forgets to set the VGACNTR VGA_DISP_DISABLE bit, and the reset
value for said bit is 0. That apparently prevents the pipes from
running, so we must still remember to set the bit, despite the VGA
plane was never actually enabled. On more modern platforms (hsw+
maybe?) the reset value for VGACNTR was changed to have
VGA_DISP_DISABLE already set.
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patch.msgid.link/20251208182637.334-6-ville.syrjala@linux.intel.com
Acked-by: Jani Nikula <jani.nikula@intel.com>