| Age | Commit message (Collapse) | Author |
|
Don't apply to KIQ. Seems to cause problems on KIQ
on some ARM platforms.
Closes: https://gitlab.freedesktop.org/drm/amd/-/work_items/5658
Fixes: 91cf34bc5a55 ("drm/amdgpu/gfx8: align mqd settings with KFD")
Reviewed-by: Jesse Zhang <jesse.zhang@amd.com>
Reviewed-by: Kent Russell <kent.russell@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
|
|
oversubscription_timer in MES is in micro second, so change the
timer's interval in KFD/useq to micro second.
Signed-off-by: Eric Huang <jinhuieric.huang@amd.com>
Suggested-by: Leo Lin <Leo.Lin@amd.com>
Reviewed-by: Mario Limonciello <mario.limonciello@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
|
|
why:
HDMI 2.1 Auto Low-Latency Mode (ALLM) lets a Source request the Sink's
low-latency mode through the HF-VSIF. HDMI 2.1 Section 7.6.6 requires
that when Gaming-VRR is enabled (VRR_EN=1) and the Sink advertises ALLM
in the SCDS, the Source shall transmit the HF-VSIF and set ALLM_Mode=1.
amdgpu never set ALLM_Mode, so this requirement was not met.
how:
- In update_freesync_state_on_stream(), set ALLM_Mode=1 in the HF-VSIF
when Gaming-VRR is active (vrr state ACTIVE_VARIABLE/ACTIVE_FIXED,
i.e. VRR_EN=1) and the sink advertises ALLM, per HDMI 2.1 Section
7.6.6, and push the updated HF-VSIF (vsp_infopacket) as a stream
update.
- ALLM is driven by the mandatory Gaming-VRR case.
Signed-off-by: Fangzhi Zuo <Jerry.Zuo@amd.com>
Signed-off-by: Tomasz Pakuła <tomasz.pakula.oficjalny@gmail.com>
Reviewed-by: Harry Wentland <harry.wentland@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
|
|
why:
HDMI 2.1 sinks advertise their VRR range in the HDMI Forum VSDB
(HF-VSDB), but amdgpu derived FreeSync capability only from the AMD
VSDB. Sinks that expose just the HDMI Forum VRR capability (e.g. HDMI
compliance EDIDs) were therefore reported as not VRR capable.
how:
- In amdgpu_dm_update_freesync_caps(), when the AMD VSDB does not
provide a valid FreeSync range, fall back to the HDMI 2.1 VRR range
parsed by DRM core from the HF-VSDB
(connector->display_info.hdmi.vrr_cap). VRRMAX = 0 means "up to the
Base Refresh Rate"; when the EDID provides no monitor range maximum
either, fall back to the Base Refresh Rate (the highest refresh-rate
mode of the preferred timing) so a valid VRR range is still reported
to userspace.
- Add VRR debug logging along the FreeSync capability and config paths.
Signed-off-by: Fangzhi Zuo <Jerry.Zuo@amd.com>
Signed-off-by: Tomasz Pakuła <tomasz.pakula.oficjalny@gmail.com>
Reviewed-by: Harry Wentland <harry.wentland@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
|
|
Parse the HDMI 2.1 gaming-related capabilities advertised in the HDMI
Forum VSDB (HF-VSDB) and expose them through struct drm_hdmi_info so
drivers can consume them.
Add struct drm_hdmi_vrr_cap describing the sink's VRR capabilities: Fast
VActive (Quick Frame Transport), Negative M VRR, Cinema VRR, MDelta, and
the VRRmin/VRRmax range, together with a "supported" flag derived from
that range. Add the fapa_start_location and allm (Auto Low Latency Mode)
flags to struct drm_hdmi_info.
drm_parse_hdmi_gaming_info() reads byte 8 of the HF-VSDB for the
FAPA/ALLM/FVA/CNMVRR/CinemaVRR/MDelta flags and bytes 9-10 for
VRRmin/VRRmax. Per HDMI 2.1, VRR is considered supported when VRRmin is
within 1-48 and VRRmax is either 0 (maximum based on the video mode) or
>= 100. It is invoked from drm_parse_hdmi_forum_scds(), and the parsed
values are logged for debugging.
Signed-off-by: Tomasz Pakuła <tomasz.pakula.oficjalny@gmail.com>
Signed-off-by: Fangzhi Zuo <Jerry.Zuo@amd.com>
Tested-by: Bernhard Berger <bernhard.berger@gmail.com>
Reviewed-by: Harry Wentland <harry.wentland@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
|
|
The BAR0 fallback read path was introduced as a workaround for SR-IOV VFs
where the VRAM aperture is not available during early init. Restrict this
workaround to only SR-IOV VFs where it's needed.
Reported-by: gloveless@jqluv.com
Fixes: cba4928cdffa ("drm/amdgpu: reduce early full GPU access during SR-IOV init")
Acked-by: Alex Deucher <alexander.deucher@amd.com>
Link: https://patch.msgid.link/20260826185102.2269511-1-mario.limonciello@amd.com
Signed-off-by: Mario Limonciello <mario.limonciello@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
|
|
Explicitely set use_mmio_for_reset = false in gfx_v12_1_sw_init for
using MES to perform queue/pipe reset
Signed-off-by: Amber Lin <amber.lin@amd.com>
Reviewed-by: Jesse Zhang <Jesse.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
|
|
why:
HDMI FRL sinks were not parsed for the AMD VSDB and no VTEM info packet
was emitted for them, so 2.1 FreeSync over HDMI FRL did not work. It is
backward-compatible with 2.0 FreeSync.
how:
- Accept SIGNAL_TYPE_HDMI_FRL alongside SIGNAL_TYPE_HDMI_TYPE_A when
parsing the AMD VSDB in amdgpu_dm_update_freesync_caps().
- Build and send the VTEM info packet via mod_build_infopacket_vtem()
when the stream signal is HDMI FRL during the freesync state update.
- Set the VTEM Data_Set_Length to 0 when no VTEM feature is enabled.
build_infopacket_header_vtem() hardcodes Data_Set_Length = 4, so a VTEM
with Data_Set_Length = 4 would be transmitted even when no VTEM feature
is enabled (VRR_EN = 0 and no FVA), e.g. when the sink advertises
VRRMIN = 0 and vrr_capable is false. This fails HDMI GCTS HF1-58 step
6.2.
The VTEM must keep being transmitted every MTW while VRR is enabled
(HF1-58 steps 8.1 and 8.3), so it cannot simply be suppressed per
frame. Instead, follow the MLDS option in HDMI 2.1 10.10.2.4: keep
transmitting the VTEM but set Data_Set_Length = 0 when no feature is
enabled. When VRR becomes active the full Data_Set_Length = 4 payload
with VRR_EN = 1 is sent as before.
Signed-off-by: Fangzhi Zuo <Jerry.Zuo@amd.com>
Reviewed-by: Harry Wentland <harry.wentland@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
|
|
MES (Micro Engine Scheduler) does not perform heavy-weight TLB
invalidation after unmapping queues, unlike HWS which does this
automatically. This causes a race condition where in-flight DMA
descriptors can access memory that has been unmapped, leading to page
faults and GPU queue hangs during SVM page migration.
The issue manifests as KFDSVMRangeTest.MultiThreadMigrationTest
failures on gfx1151 (Strix Point) with XNACK mode 1 enabled - the GPU
compute queue hangs with packets submitted but never consumed.
Add kfd_flush_tlb() calls after MES queue removal in two locations:
- evict_process_queues_cpsch(): after all queues removed during eviction
- suspend_queues(): after debug/criu queue suspension (with mem_fence barrier)
This ensures all in-flight memory accesses from unmapped queues are
flushed before memory is freed or migrated.
Signed-off-by: Priya Hosur <Priya.Hosur@amd.com>
Reviewed-by: Felix Kuehling <felix.kuehling@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
|
|
The id is claimed before the banks are read, so a poll that ends up
logging nothing still burns one and leaves a gap that every reader of
the log then has to walk over. Claim it when the first record of the
batch is about to go in.
Signed-off-by: Xiang Liu <xiang.liu@amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
|
|
The caller walks the log with cper_start_id and resumes at
cper_start_id + real_cper_num, but the reply only counts the ids that
held a record. A batch id that holds nothing therefore never moves the
caller forward, and every such id makes it read the next populated batch
one more time. Count the ids covered instead.
The rewind to the last populated batch when the query starts at the end
has the same effect once the caller has drained the log, and the latest
id is already reported by the snapshot query, so drop it.
Signed-off-by: Xiang Liu <xiang.liu@amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
|
|
In function amdgpu_userq_fence_signaled rptr
is a confusing variabile name as it is a fence
value written by the cp in the fence address and no
where related to the rptr.
Signed-off-by: Sunil Khatri <sunil.khatri@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
|
|
In function amdgpu_userq_fence_driver_process
rptr is a confusing variable name as it is a fence
value written by the cp in the fence address and no
where related to the rptr.
Signed-off-by: Sunil Khatri <sunil.khatri@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
|
|
A bank carrying the scrub bit was found by the background scrubber
rather than by an access, which is worth telling apart from the rest
when reading a log.
Signed-off-by: Xiang Liu <xiang.liu@amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
|
|
The prefix marks the files that back a hardware IP block. ACA is a bank
format rather than an IP, and the rest of its files are named after it
alone, so the v5 pair is the odd one out.
Signed-off-by: Xiang Liu <xiang.liu@amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
|
|
Replace spaces with hyphens in amdgpu_tlb_fence_get_driver_name() and
amdgpu_tlb_fence_get_timeline_name(). Returning string literals with
spaces makes ftrace fence events and debugfs fence entries visually
ambiguous and harder to parse using standard log analysis tools.
Signed-off-by: Prike Liang <Prike.Liang@amd.com>
Acked-by: Alex Deucher <alexander.deucher@amd.com>
Acked-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
|
|
Outside of UMC a bank with the poison bit set describes the consumption
of an error that was already reported, not a new uncorrectable one.
Injecting an uncorrectable error while a workload is running therefore
buries the real bank under a dump of GFX poison banks collected from
every other device of the hive.
Signed-off-by: Xiang Liu <xiang.liu@amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
|
|
DCE can support PREMULTI and COVERAGE blend mode depending on its
generation, however current driver implementation either doesn't expose
more than primary and cursor plane, or doesn't program registers for any
blend mode other than PIXEL_NONE. To fix the missing-blend-mode-prop
warning according to current DCE plane caps, create blend mode property
with PIXEL_NONE and PREMULTI for primary planes. As long as the
background is black and there is no overlay plane, PIXEL_NONE and
PREMULTI are equivalent, and PREMULTI has been the mandatory/default
mode for years, so keep it to avoid regressions.
Fixes: 860e748bddcc ("drm: ensure blend mode supported if pixel format with alpha exposed")
Tested-by: Viktor Jägersküpper <viktor_jaegerskuepper@freenet.de>
Tested-by: Dan Wheeler <daniel.wheeler@amd.com> #v3
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Alex Hung <alex.hung@amd.com>
Reviewed-by: Leandro Ribeiro <leandro.ribeiro@collabora.com> #v2
Signed-off-by: Melissa Wen <mwen@igalia.com>
Link: https://patch.msgid.link/20260826104143.39077-4-mwen@igalia.com
|
|
validate_blend_mode_for_alpha_formats() warns when a plane supports
formats with alpha but doesn't expose the blend mode property. Fix this
by adding the same overlay plane blend modes to primary plane, since
they are all universal planes in DCN-generation. Cursor planes support
ARGB8888 format and CURSOR_MODE_COLOR_PRE_MULTIPLIED_ALPHA is set by
default (other color formats are not implemented), so only expose
support to PREMULTI, which is the default blend mode on DRM.
Fixes: 860e748bddcc ("drm: ensure blend mode supported if pixel format with alpha exposed")
Tested-by: Mikhail Gavrilov <mikhail.v.gavrilov@gmail.com>
Tested-by: Dan Wheeler <daniel.wheeler@amd.com>
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Alex Hung <alex.hung@amd.com>
Reviewed-by: Leandro Ribeiro <leandro.ribeiro@collabora.com>
Signed-off-by: Melissa Wen <mwen@igalia.com>
Link: https://patch.msgid.link/20260826104143.39077-3-mwen@igalia.com
|
|
In __drm_atomic_helper_plane_state_init(), pixel_blend_mode is always
reset to DRM_MODE_BLEND_PREMULTI. That was consistent while
drm_plane_create_blend_mode_property() required PREMULTI in the
supported modes, but it now falls back to COVERAGE or PIXEL_NONE when
the driver doesn't support PREMULTI. The hardcoded default may therefore
not be a blend mode the hardware can do, nor one the property
advertises.
Initialize pixel_blend_mode from the blend mode property default
instead, keeping DRM_MODE_BLEND_PREMULTI for planes without the
property.
Fixes: 9813e158d13d ("drm/drm_blend: allow blend mode property without PREMULTI")
Tested-by: Mikhail Gavrilov <mikhail.v.gavrilov@gmail.com>
Tested-by: Dan Wheeler <daniel.wheeler@amd.com>
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Reviewed-by: Alex Hung <alex.hung@amd.com>
Reviewed-by: Leandro Ribeiro <leandro.ribeiro@collabora.com>
Signed-off-by: Melissa Wen <mwen@igalia.com>
Link: https://patch.msgid.link/20260826104143.39077-2-mwen@igalia.com
|
|
In prepataion to further modify the logic of this function, simplify the
error-returning code.
With this change, on any error there is an immediate 'return -<ERRNO>;'
statement, without having to carry on the return value until the end.
Additionally, clear both the panel and the bridge pointers at the
beginning. Even though this is redundant in some code paths, it allows to
have a simpler code in the rest of the function.
Reviewed-by: Maxime Ripard <mripard@kernel.org>
Link: https://patch.msgid.link/20260814-drm-bridge-every-panel-v1-2-19cd5277cc8d@bootlin.com
Signed-off-by: Luca Ceresoli <luca.ceresoli@bootlin.com>
|
|
In prepataion to further modify the logic of this function, simplify
putting the struct device_node pointer by using a cleanup action.
Reviewed-by: Maxime Ripard <mripard@kernel.org>
Link: https://patch.msgid.link/20260814-drm-bridge-every-panel-v1-1-19cd5277cc8d@bootlin.com
Signed-off-by: Luca Ceresoli <luca.ceresoli@bootlin.com>
|
|
Add the BOE 0x0b21 panel entry used in the K&D Technology
KD116N36-30NB-A001 eDP LCM, place the EDID here for subsequent
reference. Timing values come from the KD116N36-30NB-A001 spec.
00 ff ff ff ff ff ff 00 09 e5 21 0b 00 00 00 00
01 1e 01 04 a5 1a 0e 78 0a 24 10 97 59 54 8e 27
1e 50 54 00 00 00 01 01 01 01 01 01 01 01 01 01
01 01 01 01 01 01 b2 39 80 d4 70 38 4b 40 30 20
36 00 04 8c 10 00 00 1a 00 00 00 00 00 00 00 00
00 00 00 00 00 00 00 00 00 1a 00 00 00 00 00 00
00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 fe
00 51 56 31 31 36 46 48 42 2d 4e 38 31 0a 00 b3
Signed-off-by: Yongxing Mou <yongxing.mou@oss.qualcomm.com>
Reviewed-by: Douglas Anderson <dianders@chromium.org>
Signed-off-by: Douglas Anderson <dianders@chromium.org>
Link: https://patch.msgid.link/20260901-edp-id-boe-v1-1-f5e2c44255f7@oss.qualcomm.com
|
|
This driver does not use the drm_panel APIs at all, so no need to select
DRM_PANEL.
But it uses devm_drm_of_get_bridge(), which is enabled by DRM_PANEL_BRIDGE.
Cc: Douglas Anderson <dianders@chromium.org>
Cc: Jitao Shi <jitao.shi@mediatek.com>
Cc: CK Hu <ck.hu@mediatek.com>
Cc: Enric Balletbo i Serra <enric.balletbo@collabora.com>
Cc: Pin-yen Lin <treapking@chromium.org>
Reviewed-by: Douglas Anderson <dianders@chromium.org>
Link: https://patch.msgid.link/20260831-drm-bridge-drm_panel-cleanup-v3-8-50efd8168b3e@bootlin.com
Signed-off-by: Luca Ceresoli <luca.ceresoli@bootlin.com>
|
|
This driver uses the drm_panel_bridge API and not the drm_panel API so it
should select DRM_PANEL_BRIDGE and not DRM_PANEL.
Cc: Marek Vasut <marex@denx.de>
Cc: Sudarshan Shetty <tessolveupstream@gmail.com>
Cc: Alexander Stein <alexander.stein@ew.tq-group.com>
Cc: Herve Codina <herve.codina@bootlin.com>
Cc: Esben Haabendal <esben@geanix.com>
Reviewed-by: Herve Codina <herve.codina@bootlin.com>
Acked-by: Esben Haabendal <esben@geanix.com>
Link: https://patch.msgid.link/20260831-drm-bridge-drm_panel-cleanup-v3-5-50efd8168b3e@bootlin.com
Signed-off-by: Luca Ceresoli <luca.ceresoli@bootlin.com>
|
|
This driver does not use the drm_panel APIs at all, so no need to select
DRM_PANEL.
But it uses devm_drm_of_get_bridge(), which is enabled by DRM_PANEL_BRIDGE.
Cc: Douglas Anderson <dianders@chromium.org>
Cc: Franz Schnyder <franz.schnyder@toradex.com>
Cc: John Ripple <john.ripple@keysight.com>
Cc: Michael Walle <mwalle@kernel.org>
Reviewed-by: Douglas Anderson <dianders@chromium.org>
Link: https://patch.msgid.link/20260831-drm-bridge-drm_panel-cleanup-v3-4-50efd8168b3e@bootlin.com
Signed-off-by: Luca Ceresoli <luca.ceresoli@bootlin.com>
|
|
This driver does not use the drm_panel APIs at all, so no need to select
DRM_PANEL.
Cc: Peter Senna Tschudin <peter.senna@gmail.com>
Cc: Martyn Welch <martyn.welch@collabora.co.uk>
Cc: Andy Yan <andyshrk@163.com>
Reviewed-by: Ian Ray <ian.ray@gehealthcare.com>
Link: https://patch.msgid.link/20260831-drm-bridge-drm_panel-cleanup-v3-1-50efd8168b3e@bootlin.com
Signed-off-by: Luca Ceresoli <luca.ceresoli@bootlin.com>
|
|
Use the managed MIPI DSI attach and panel add helpers in display vendor
panel drivers and drop the corresponding manual detach and panel remove
calls from remove paths.
Suggested-by: Luca Ceresoli <luca.ceresoli@bootlin.com>
Signed-off-by: Osama Abdelkader <osama.abdelkader@gmail.com>
Reviewed-by: Luca Ceresoli <luca.ceresoli@bootlin.com>
Link: https://patch.msgid.link/20260812152535.82238-1-osama.abdelkader@gmail.com
Signed-off-by: Luca Ceresoli <luca.ceresoli@bootlin.com>
|
|
Use the managed MIPI DSI attach and panel add helpers in Himax and Ilitek
panel drivers and drop the corresponding manual detach and panel remove
calls from remove paths.
Suggested-by: Luca Ceresoli <luca.ceresoli@bootlin.com>
Signed-off-by: Osama Abdelkader <osama.abdelkader@gmail.com>
Reviewed-by: Luca Ceresoli <luca.ceresoli@bootlin.com>
Link: https://patch.msgid.link/20260812151000.74799-1-osama.abdelkader@gmail.com
Signed-off-by: Luca Ceresoli <luca.ceresoli@bootlin.com>
|
|
Use the managed MIPI DSI attach and panel add helpers in Samsung panel
drivers and drop the corresponding manual detach and panel remove calls
from remove paths. Also, switch to devm_backlight_device_register in
s6e3ha2 and s6e63j0x03.
Suggested-by: Luca Ceresoli <luca.ceresoli@bootlin.com>
Signed-off-by: Osama Abdelkader <osama.abdelkader@gmail.com>
Reviewed-by: Luca Ceresoli <luca.ceresoli@bootlin.com>
Link: https://patch.msgid.link/20260812144426.62074-1-osama.abdelkader@gmail.com
Signed-off-by: Luca Ceresoli <luca.ceresoli@bootlin.com>
|
|
In dw_hdmi_probe(), casting the bitwise complement of (HDMI_PHY_HPD |
HDMI_PHY_RX_SENSE) directly to u8 triggers a compiler warning
(-Wconstant-conversion).
Due to C integer promotion rules, the bitwise OR and bitwise NOT are
evaluated as 32-bit signed integers, producing 0xFFFFFF0D before being
cast down to u8.
Fix this by explicitly masking the inverted 32-bit result with 0xFF
to keep only the lower byte and prevent implicit upper-bit truncation.
Signed-off-by: Sean Chang <seanwascoding@gmail.com>
Reviewed-by: Luca Ceresoli <luca.ceresoli@bootlin.com>
Link: https://patch.msgid.link/20260819151216.56833-1-seanwascoding@gmail.com
Signed-off-by: Luca Ceresoli <luca.ceresoli@bootlin.com>
|
|
psr_get_status_and_error_status() selects the DPCD offset to read the error
status from based on whether Panel Replay is enabled:
offset = intel_dp->psr.panel_replay_enabled ?
DP_PANEL_REPLAY_ERROR_STATUS : DP_PSR_ERROR_STATUS;
but intel_psr_short_pulse() acknowledges it unconditionally to the PSR
register. Under Panel Replay the error is therefore read from
DP_PANEL_REPLAY_ERROR_STATUS (0x2020) and the acknowledgment written to
DP_PSR_ERROR_STATUS (0x2006). DP_PANEL_REPLAY_ERROR_STATUS is never written
anywhere in the tree; it appears only in the read above and in its own
The sink's Panel Replay error latch can consequently never be cleared. Once
it latches, every subsequent short pulse re-reads the same errors, so PSR
is disabled with sink_not_reliable set permanently, and until a short pulse
arrives the driver keeps Panel Replay enabled while the sink is reporting
errors it cannot see.
Observed on a Dell XPS 16 DA16260 (Panther Lake, Arc B390, display version
30.00) with the eDP Panel Replay quirk from commit cb8d155b0806
("drm/i915/psr: Disable Panel Replay on Dell XPS 16 DA16260 as a quirk")
removed locally so the feature could be exercised.
The sink reports a persistently latched error:
Sink PANEL-REPLAY status: 0x2 [active, display from RFB]
Sink PANEL-REPLAY error status: 0x1:
PANEL-REPLAY Link CRC error
which survives across reads indefinitely, while dmesg stays silent and
Panel Replay Selective Update remains enabled.
Use the same conditional offset when clearing. psr->panel_replay_enabled
cannot be used at that point because intel_psr_disable_locked() clears it
earlier in the same function whenever an error was detected, which is
exactly the case that needs the Panel Replay offset, so save it beforehand.
v2: use a copy of panel_replay_enabled taken before
intel_psr_disable_locked() clears it. In v1 the condition was evaluated
after the disable, so it selected DP_PSR_ERROR_STATUS in the error path
and the patch was a no-op there. Caught by Sashiko AI review.
Signed-off-by: Jake Steinman <j@metarealtyinc.ca>
Reviewed-by: Jouni Högander <jouni.hogander@intel.com>
Link: https://patch.msgid.link/20260901205956.130779-1-j@metarealtyinc.ca
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
[Rodrigo: adjusted commit message based on checkpatch]
|
|
In the GSP->CPU messaging path, the code reads the write pointer from GSP,
reads the response and advances the read pointer.
A LOAD->LOAD ordering is required after the write pointer read and the data
read. Add it as this is currently missing.
A LOAD->STORE ordering is required after the data read and the advance of
read pointer. Currently a Rust `SeqCst` barrier is used, which roughly maps
to `smp_mb(Full)`; this however does not order DMA operations (notably on
ARM, the generate barrier orders inner shareable and not outer shareable,
which is ordered by `dma_mb`). This ordering does not need to be in between
read pointer read and write, because it's for ordering between the ring
buffer data and the pointer; the RMW operation does not internally need a
barrier (nor it has to be atomic, as CPU pointers are updated by CPU only),
so move it to before the RMW sequence for clarity.
Signed-off-by: Gary Guo <gary@garyguo.net>
Reviewed-by: Eliot Courtney <ecourtney@nvidia.com>
Link: https://patch.msgid.link/20260901-rust-barrier-v4-2-94427f445310@garyguo.net
Signed-off-by: Danilo Krummrich <dakr@kernel.org>
|
|
In the CPU->GSP messaging path, the code reads the read pointer from GSP,
writes the command, advances the write pointer, and then notifies the GSP.
A LOAD->STORE ordering is needed after reading the read pointer from GSP
and writing the command. Control dependency exists here which provide the
needed ordering, but it's best to avoid depending on it.
A STORE->STORE ordering is needed after the command write and before the
write pointer advance. This is currently incorrectly done after the write
pointer advance (and before GSP notification), but this can cause issue if
GSP is still processing ring buffer, as it may observe the write pointer
advance before command write. Thus move this barrier to be before the write
pointer advance. Note that barriers are not needed between write pointer
advance and GSP notification, as MMIO accessors already carries the
required barrier.
Signed-off-by: Gary Guo <gary@garyguo.net>
Reviewed-by: Eliot Courtney <ecourtney@nvidia.com>
Link: https://patch.msgid.link/20260901-rust-barrier-v4-1-94427f445310@garyguo.net
Signed-off-by: Danilo Krummrich <dakr@kernel.org>
|
|
On Tegra devices, uncached maps translate to device memory, causing
unaligned accesses by userspace resulting in a SIGBUS. Instead, use
write-combined maps to ensure proper access.
This would also affect discrete cards on any Arm device. It was
determined that discrete cards regardless of cpu arch should use
write-combined maps for coherent anyways. Thus this change is made for
all gpu types.
Cc: stable@vger.kernel.org
Signed-off-by: Faith Ekstrand <faith.ekstrand@collabora.com>
Co-developed-by: Aaron Kling <webgeek1234@gmail.com>
Signed-off-by: Aaron Kling <webgeek1234@gmail.com>
Fixes: 1b4ea4c5980f ("drm/ttm: set the tt caching state at creation time")
Link: https://patch.msgid.link/20260821-tegra-coherent-wc-v2-1-2b1ddb67bf18@gmail.com
Signed-off-by: Danilo Krummrich <dakr@kernel.org>
|
|
If idr_alloc() fails after create_dynamic_oa_sysfs_entry() has
succeeded, the error path frees the OA config without removing the
metrics sysfs group.
Remove the sysfs group before releasing the config, and fix up the
misleading error message copied from the sysfs creation failure path.
Fixes: cdf02fe1a94a ("drm/xe/oa/uapi: Add/remove OA config perf ops")
Signed-off-by: Lu Yao <yaolu@kylinos.cn>
Link: https://patch.msgid.link/20260831014218.28515-1-yaolu@kylinos.cn
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
(cherry picked from commit 2c6fbda5fdde461d6dedb82a59285182720b8fef)
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
|
|
nouveau_dmem_folio_free() drops chunk->callocated once per freed folio,
while a large (compound) device-private folio is only counted once when
it is allocated. When such a folio is split, the mm core invokes
->folio_split() (nouveau_dmem_folio_split()) once for each new
sub-folio, but the hook only fixes up the sub-folio metadata and leaves
chunk->callocated unchanged.
Each resulting sub-folio is later freed separately, so after a split
the single allocation (+1) is met by N frees (-N), leaving
chunk->callocated short by N-1. On the first split/free cycle it
underflows: WARN_ON(!chunk->callocated) fires, the unsigned counter
wraps and never returns to zero, so the chunk can no longer be
reclaimed (nouveau_dmem_fini() also warns on the leaked count).
Account for the new sub-folio in the split hook, under the same lock as
nouveau_dmem_folio_free(), so the count stays balanced.
Fixes: c32287471077 ("gpu/drm/nouveau: enable THP support for GPU memory migration")
Reported-by: Yuhao Jiang <danisjiang@gmail.com>
Assisted-by: Claude:claude-opus-5
Cc: stable@vger.kernel.org
Signed-off-by: Zhenhao Wan <whi4ed0g@gmail.com>
Reviewed-by: Lyude Paul <lyude@redhat.com>
Link: https://patch.msgid.link/20260811-b4-nouveau-dmem-thp-fixes-v1-2-2cdf9860af2a@gmail.com
Signed-off-by: Danilo Krummrich <dakr@kernel.org>
|
|
Device-private THP migration maps migration buffers with page_size()
and records that length in dma_info->size. For a compound folio
page_size() is PAGE_SIZE << order, but two teardown sites still pass a
literal PAGE_SIZE to dma_unmap_page():
- nouveau_dmem_migrate_to_ram() on the success path, and
- nouveau_dmem_migrate_copy_one() on the copy-error path.
For an order > 0 folio this unmaps less than was mapped, leaking the
remainder of the IOMMU/IOVA mapping. The other unmap sites, in
nouveau_dmem_migrate_chunk() and nouveau_dmem_evict_chunk(), already
use the saved size; use it here too.
Fixes: c32287471077 ("gpu/drm/nouveau: enable THP support for GPU memory migration")
Reported-by: Yuhao Jiang <danisjiang@gmail.com>
Assisted-by: Claude:claude-opus-5
Cc: stable@vger.kernel.org
Signed-off-by: Zhenhao Wan <whi4ed0g@gmail.com>
Link: https://patch.msgid.link/20260811-b4-nouveau-dmem-thp-fixes-v1-1-2cdf9860af2a@gmail.com
Signed-off-by: Danilo Krummrich <dakr@kernel.org>
|
|
Turns out sashiko was right, and I should protect this properly
Fixes: 34e27b90552a ("nouveau/instmem: use iomapping interface for instmem handling")
Signed-off-by: Dave Airlie <airlied@redhat.com>
Link: https://patch.msgid.link/20260825030615.3464436-1-airlied@gmail.com
Signed-off-by: Danilo Krummrich <dakr@kernel.org>
|
|
The legacy cursor path forced all joiner updates to the
slow path by checking joiner_pipes.
Drop the condition so that joiner cursor updates can use
the fast path.
v4: Trimmed to only the joiner_pipes gate drop. [Chaitanya]
Assisted-by: Claude:claude-sonnet-4.6
Signed-off-by: Nemesa Garg <nemesa.garg@intel.com>
Reviewed-by: Chaitanya Kumar Borah <chaitanya.kumar.borah@intel.com>
Signed-off-by: Suraj Kandpal <suraj.kandpal@intel.com>
Link: https://patch.msgid.link/20260901072813.2596383-6-nemesa.garg@intel.com
|
|
Iterate over all joined pipes when arming/disabling the cursor plane so
secondary pipes are updated together with the primary. The pin, check
and state duplication for secondary pipes was already prepared in the
previous commit; this converts the update_arm pass to a loop over the
joined_pipe_state[] array.
Because the whole loop runs inside a single primary vblank-evade,
sample intel_crtc_get_vblank_counter() per pipe around each arm and
emit a drm_err() if the counter ticks during that pipe's own
programming.
v2: Check primary and secondary pipe together. [Ville]
v3: Use struct intel_cursor_joiner_state. [Ville]
v4: Add per-pipe vblank straddle detection around the arm loop. [Chaitanya]
v5: Move straddle check outside the loop to cover all pipes together.
Sample vblank counter from primary pipe only.
Move drm_err() after local_irq_enable(). [Chaitanya]
Assisted-by: Claude:claude-sonnet-4.6
Signed-off-by: Nemesa Garg <nemesa.garg@intel.com>
Reviewed-by: Chaitanya Kumar Borah <chaitanya.kumar.borah@intel.com>
Signed-off-by: Suraj Kandpal <suraj.kandpal@intel.com>
Link: https://patch.msgid.link/20260901072813.2596383-5-nemesa.garg@intel.com
|
|
In joiner mode the fast path cursor update must handle
secondary pipes. Iterate over all joined pipes uniformly
to duplicate plane state, run check_plane(), pin the
framebuffer and on success swap in the new plane state
for each secondary cursor.
Track every successfully prepared pipe in a joined_pipe_state[] array
of struct intel_cursor_joiner_state so that later frontbuffer, unpin
and error-cleanup paths treat primary and secondaries uniformly,
and ensures the primary's pinned framebuffer is released if a
secondary fails partway through.
Convert the primary-only vblank unpin block into a loop over the
joined_pipe_state[] array so each pipe's old cursor framebuffer is
scheduled for unpin (or unpinned inline when unchanged) independently.
All unpin work is armed on the primary crtc's vblank (&crtc->base),
consistent with the single primary vblank evasion used above.
v2: Use intel_crtc_joined_pipe_mask(). [Ville]
Add locking mechanism. [Ville]
v3: Drop the per-pipe fastpath mutex array. [sashiko]
v4: Fold parallel arrays into struct intel_cursor_joiner_state
joined_pipe_state[]. Unify primary/secondary in a single loop.
Use bare check_plane(). [Chaitanya]
v5: Fix secondary uapi.crtc to each pipe's own crtc.
Rename intel_cursor_pipe to intel_cursor_joiner_state and
joined[] to joined_pipe_state[]. [Chaitanya]
v6: Mirror primary uapi into secondary hw state, don't set uapi.crtc.
[sashiko]
v7: Squash the per-pipe cursor unpin scheduling into this patch.[Chaitanya]
v8: Resolve checkpatch errors.
Assisted-by: Claude:claude-sonnet-4.6
Signed-off-by: Nemesa Garg <nemesa.garg@intel.com>
Reviewed-by: Chaitanya Kumar Borah <chaitanya.kumar.borah@intel.com>
Signed-off-by: Suraj Kandpal <suraj.kandpal@intel.com>
Link: https://patch.msgid.link/20260901072813.2596383-4-nemesa.garg@intel.com
|
|
Move cursor fast path plane state update into helper
function. The target hw.crtc is passed as a parameter
so a later patch can reuse the helper for joiner
secondary pipes, where hw.crtc points at the secondary
being programmed while uapi.crtc stays on the primary.
v4: Drop uapi.crtc assignment from the helper. [Chaitanya]
Assisted-by: Claude:claude-sonnet-4.6
Signed-off-by: Nemesa Garg <nemesa.garg@intel.com>
Reviewed-by: Chaitanya Kumar Borah <chaitanya.kumar.borah@intel.com>
Signed-off-by: Suraj Kandpal <suraj.kandpal@intel.com>
Link: https://patch.msgid.link/20260901072813.2596383-3-nemesa.garg@intel.com
|
|
In joiner mode, secondary cursor commits may still be running
even when the primary cursor commit is done. Walking the secondary
pipes also requires holding the secondary planes modeset locks.
Add intel_cursor_lock_joined_planes() to acquire modeset locks
for all secondary cursor planes. Check all joined cursor commit
status before taking the fast path. If any commit is still pending,
fallback to slow path.
v2: Use intel_crtc_joined_pipe_mask(). [Ville]
v3: Lock secondary cursor CRTCs and planes. [sashiko]
v4: Iterate the full joined mask uniformly in both helpers, no
primary special-case.
Move the parameter-change check above the lock acquisition so
we don't grab secondary locks just to fall to slow path. [Chaitanya]
v5: Remove extra header declaration. [Chaitanya]
Assisted-by: Claude:claude-sonnet-4.6
Signed-off-by: Nemesa Garg <nemesa.garg@intel.com>
Reviewed-by: Chaitanya Kumar Borah <chaitanya.kumar.borah@intel.com>
Signed-off-by: Suraj Kandpal <suraj.kandpal@intel.com>
Link: https://patch.msgid.link/20260901072813.2596383-2-nemesa.garg@intel.com
|
|
A successful OP_UNMAP_SPARSE marks its region dirty with
nouveau_uvma_region_dirty() and defers the teardown to
nouveau_uvmm_bind_job_cleanup(); it does not remove the region from
uvmm->region_mt.
If a later op in the job fails, the unwind path never clears reg->dirty
(set in one place, cleared nowhere) and sets op->reg = NULL, so cleanup
skips the teardown. The region is left in the tree with dirty set and its
completion never signalled. Later binds over that range then fail
permanently -- -ENOENT or -EINVAL from the dirty checks, or an unkillable
wait_for_completion() in bind_validate_region() -- for the lifetime of
the uvmm.
Clear reg->dirty when the unwind reverts the sparse unmap, restoring the
region to the state it was found in.
Fixes: b88baab82871 ("drm/nouveau: implement new VM_BIND uAPI")
Reported-by: Yuhao Jiang <danisjiang@gmail.com>
Assisted-by: Claude:claude-opus-5
Cc: stable@vger.kernel.org
Signed-off-by: Zhenhao Wan <whi4ed0g@gmail.com>
Reviewed-by: Lyude Paul <lyude@redhat.com>
Link: https://patch.msgid.link/20260811-nouveau-uvmm-vmbind-fixes-v2-3-aaee4b395d04@gmail.com
Signed-off-by: Danilo Krummrich <dakr@kernel.org>
|
|
In nouveau_uvmm_bind_job_submit()'s OP_UNMAP_SPARSE arm, op->reg is set
from nouveau_uvma_region_find(), which only looks the region up and takes
no reference; a region's sole reference is its membership in
uvmm->region_mt. Two failure paths leave op->reg set: the -ENOENT check
when the region is busy, and the drm_gpuvm_sm_unmap_ops_create() failure.
The sibling nouveau_uvmm_sm_unmap_prepare() failure just below clears
op->reg; these two do not.
unwind_continue steps back one op, so the failing op is skipped by the
unwind loop and its op->reg stays set. nouveau_uvmm_bind_job_cleanup()
then enters its if (op->reg) branch and calls nouveau_uvma_region_remove()
and nouveau_uvma_region_put() on it, dropping the tree's sole reference
and freeing a region this job never created. The comment above the
cleanup loop documents the broken invariant: op->reg must be NULL on
submit failure.
This frees a live region on an unrelated failure, reachable single-job
when drm_gpuvm_sm_unmap_ops_create() returns -ENOMEM; if another job owns
the same region, its cleanup then removes and puts the freed region, a
use-after-free. Clear op->reg on both failure paths.
Fixes: b88baab82871 ("drm/nouveau: implement new VM_BIND uAPI")
Reported-by: Yuhao Jiang <danisjiang@gmail.com>
Assisted-by: Claude:claude-opus-5
Cc: stable@vger.kernel.org
Signed-off-by: Zhenhao Wan <whi4ed0g@gmail.com>
Reviewed-by: Lyude Paul <lyude@redhat.com>
Link: https://patch.msgid.link/20260811-nouveau-uvmm-vmbind-fixes-v2-2-aaee4b395d04@gmail.com
Signed-off-by: Danilo Krummrich <dakr@kernel.org>
|
|
Each bind_job_op is zeroed by kzalloc_obj() in bind_job_op_from_uop(),
and the OP_MAP_SPARSE case in nouveau_uvmm_bind_job_submit() only creates
a region, so op->ops stays NULL for a successfully processed sparse map.
If a later op in the same job fails, the reverse unwind loop revisits that
op and calls drm_gpuva_ops_free(&uvmm->base, op->ops) unconditionally.
drm_gpuva_ops_free() dereferences its argument right away
(list_for_each_entry_safe on &ops->list), so a NULL op->ops oopses. The
path is reachable by any render-node fd holder, since NOUVEAU_VM_BIND is
DRM_RENDER_ALLOW.
Guard the free with IS_ERR_OR_NULL(), as nouveau_uvmm_bind_job_cleanup()
already does for the identical free.
Fixes: b88baab82871 ("drm/nouveau: implement new VM_BIND uAPI")
Reported-by: Yuhao Jiang <danisjiang@gmail.com>
Assisted-by: Claude:claude-opus-5
Cc: stable@vger.kernel.org
Signed-off-by: Zhenhao Wan <whi4ed0g@gmail.com>
Reviewed-by: Lyude Paul <lyude@redhat.com>
Link: https://patch.msgid.link/20260811-nouveau-uvmm-vmbind-fixes-v2-1-aaee4b395d04@gmail.com
Signed-off-by: Danilo Krummrich <dakr@kernel.org>
|
|
Let's start the 7.3 drm-misc-fixes cycle.
Signed-off-by: Maxime Ripard <mripard@kernel.org>
|
|
SDP splitting for 128b/132b (UHBR) SST audio must only be enabled when
the sink advertises support for it. Previously sdp_split_enable
was set for every UHBR SST stream carrying audio, regardless of sink
capability.
In MST mode SDP splitting is inherently supported, so the sink
capability check (DP_SST_SPLIT_SDP_CAP) is applied only to the SST path.
Fixes: 8853750dbad8 ("drm/i915: Enable SDP split for DP2.0")
Assisted-by: Claude:claude-opus-4-8
Signed-off-by: Mitul Golani <mitulkumar.ajitkumar.golani@intel.com>
Reviewed-by: Suraj Kandpal <suraj.kandpal@intel.com>
Signed-off-by: Suraj Kandpal <suraj.kandpal@intel.com>
Link: https://patch.msgid.link/20260825073204.872441-1-mitulkumar.ajitkumar.golani@intel.com
|
|
The GSP-FMC firmware code lived in `firmware/fsp.rs` and described
itself as running on the FSP. This is incorrect: GSP-FMC is loaded by
the FSP, but runs on the GSP as the loader and verifier of GSP-RM. FSP
secure-boots on its own before the driver even starts.
Correct the naming and location of the GSP-FMC code by:
- Renaming `firmware/fsp.rs` to `firmware/gsp_fmc.rs` (as GSP-FMC is not
the FSP firmware),
- Changing the prefix of some declarations in that file to `FMC` or
`GspFmc` (interestingly some were already correct),
- Rewording the `fsp.rs` module documentation to describe the actual
boot flow.
Reviewed-by: Danilo Krummrich <dakr@kernel.org>
Reviewed-by: Eliot Courtney <ecourtney@nvidia.com>
Link: https://patch.msgid.link/20260825-gsp-fmc-rename-v1-1-f82526f01d7d@nvidia.com
[acourbot: drop the documentation change to be picked up by Zhi, reword
commit message a bit.]
Signed-off-by: Alexandre Courbot <acourbot@nvidia.com>
|