A problem about modprobe vc4 failed is triggered with the following log
given:
[ 420.327987] Error: Driver 'vc4_hvs' is already registered, aborting...
[ 420.333904] failed to register platform driver vc4_hvs_driver [vc4]: -16
modprobe: ERROR: could not insert 'vc4': Device or resource busy
The reason is that vc4_drm_register() returns platform_driver_register()
directly without checking its return value, if platform_driver_register()
fails, it returns without unregistering all the vc4 drivers, resulting the
vc4 can never be installed later.
A simple call graph is shown as below:
vc4_drm_register()
platform_register_drivers() # all vc4 drivers are registered
platform_driver_register()
driver_register()
bus_add_driver()
priv = kzalloc(...) # OOM happened
# return without unregister drivers
Fixing this problem by checking the return value of
platform_driver_register() and do platform_unregister_drivers() if
error happened.
Fixes: c8b75bca92 ("drm/vc4: Add KMS support for Raspberry Pi.")
Signed-off-by: Yuan Can <yuancan@huawei.com>
Signed-off-by: Maxime Ripard <maxime@cerno.tech>
Link: https://patchwork.freedesktop.org/patch/msgid/20221103014705.109322-1-yuancan@huawei.com
Commit ae71ab585c ("drm/vc4: hdmi: Enforce the minimum rate at
runtime_resume") reintroduced the call to clk_set_min_rate in an attempt
to fix the boot without a monitor connected on the RaspberryPi3.
However, that introduced a regression breaking the display output
entirely (black screen but no vblank timeout) on the Pi4.
This is due to the fact that we now have in a typical modeset at boot,
in vc4_hdmi_encoder_pre_crtc_configure(), we have a first call to
clk_set_min_rate() asking for the minimum rate of the HSM clock for our
given resolution, and then a call to pm_runtime_resume_and_get(). We
will thus execute vc4_hdmi_runtime_resume() which, since the commit
mentioned above, will call clk_set_min_rate() a second time with the
absolute minimum rate we want to enforce on the HSM clock.
We're thus effectively erasing the minimum mandated by the mode we're
trying to set. The fact that only the Pi4 is affected is due to the fact
that it uses a different clock driver that tries to minimize the HSM
clock at all time. It will thus lower the HSM clock rate to 120MHz on
the second clk_set_min_rate() call.
The Pi3 doesn't use the same driver and will not change the frequency on
the second clk_set_min_rate() call since it's still within the new
boundaries and it doesn't have the code to minimize the clock rate as
needed. So even though the boundaries are still off, the clock rate is
still the right one for our given mode, so everything works.
There is a lot of moving parts, so I couldn't find any obvious
solution:
- Reverting the original is not an option, as that would break the Pi3
again.
- We can't move the clk_set_min_rate() call in _pre_crtc_configure()
since because, on the Pi3, the HSM clock has the CLK_SET_RATE_GATE
flag which prevents the clock rate from being changed after it's
been enabled. Our calls to clk_set_min_rate() can change it, so they
need to be done before clk_prepare_enable().
- We can't remove the call to clk_prepare_enable() from the
runtime_resume hook to put it into _pre_crtc_configure() either,
since we need that clock to be enabled to access the registers, and
we can't count on the fact that the display will be active in all
situations (doing any CEC operation, or listing the modes while
inactive are valid for example()).
- We can't drop the call to clk_set_min_rate() in
_pre_crtc_configure() since we would need to still enforce the
minimum rate for a given resolution, and runtime_resume doesn't have
access to the current mode, if there's any.
- We can't copy the TMDS character rate into vc4_hdmi and reuse it
since, because it's part of the KMS atomic state, it needs to be
protected by a mutex. Unfortunately, some functions (CEC operations,
mostly) can be reentrant (through the CEC framework) and still need
a pm_runtime_get.
However, we can work around this issue by leveraging the fact that the
clk_set_min_rate() calls set boundaries for its given struct clk, and
that each different clk_get() call will return a different instance of
struct clk. The clock framework will then aggregate the boundaries for
each struct clk instances linked to a given clock, plus its hardware
boundaries, and will use that.
We can thus get an extra HSM clock user for runtime_pm use only, and use
our different clock instances depending on the context: runtime_pm will
use its own to set the absolute minimum clock setup so that we never
lock the CPU waiting for a register access, and the modeset part will
set its requirement for the current resolution. And we let the CCF do
the coordination.
It's not an ideal solution, but it's fairly unintrusive and doesn't
really change any part of the logic so it looks like a rather safe fix.
Link: https://bugzilla.redhat.com/show_bug.cgi?id=2136234
Fixes: ae71ab585c ("drm/vc4: hdmi: Enforce the minimum rate at runtime_resume")
Reported-by: Peter Robinson <pbrobinson@gmail.com>
Reviewed-by: Javier Martinez Canillas <javierm@redhat.com>
Tested-by: Peter Robinson <pbrobinson@gmail.com>
Link: https://lore.kernel.org/r/20221021131339.2203291-1-maxime@cerno.tech
Signed-off-by: Maxime Ripard <maxime@cerno.tech>
A comment introduced by commit 6bed2ea3cb ("drm/vc4: hdmi: Reset link
on hotplug") mentions a drm_atomic_helper_connector_hdmi_reset_link()
function that was part of the earlier versions but got moved internally
and is now named vc4_hdmi_reset_link(). Let's fix the function name.
Fixes: 6bed2ea3cb ("drm/vc4: hdmi: Reset link on hotplug")
Reviewed-by: Javier Martinez Canillas <javierm@redhat.com>
Link: https://lore.kernel.org/r/20221024093634.118190-2-maxime@cerno.tech
Signed-off-by: Maxime Ripard <maxime@cerno.tech>
We access some fields protected by our internal mutex in
vc4_hdmi_reset_link() (saved_adjusted_mode, output_bpc, output_format)
and are calling functions that need to have that lock taken
(vc4_hdmi_supports_scrambling()).
However, the current code doesn't lock that mutex. Let's make sure it
does.
Fixes: 6bed2ea3cb ("drm/vc4: hdmi: Reset link on hotplug")
Reviewed-by: Javier Martinez Canillas <javierm@redhat.com>
Link: https://lore.kernel.org/r/20221024093634.118190-1-maxime@cerno.tech
Signed-off-by: Maxime Ripard <maxime@cerno.tech>
[Why]
Fix for some of the tool reported modes for FCLK
P-state deviations and UCLK P-state deviations that
are coming from DSC terms and/or Scaling terms
causing MinActiveFCLKChangeLatencySupported
and MaxActiveDRAMClockChangeLatencySupported
incorrectly calculated in DML for these configurations.
Reviewed-by: Chaitanya Dhere <Chaitanya.Dhere@amd.com>
Acked-by: Jasdeep Dhillon <jdhillon@amd.com>
Acked-by: Alex Hung <alex.hung@amd.com>
Signed-off-by: Nevenko Stupar <Nevenko.Stupar@amd.com>
Tested-by: Mark Broadworth <mark.broadworth@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
[Why]
Certain 4K high refresh rate modes requiring DSC are exhibiting top
of screen underflow corruption. Increasing the DSC delay by a factor
of 6 percent stops the underflow for most use cases.
[How]
Multiply DSC delay requirement in DML by a factor.
Add debug option to make this DSC delay factor configurable.
Reviewed-by: Alvin Lee <Alvin.Lee2@amd.com>
Acked-by: Alex Hung <alex.hung@amd.com>
Signed-off-by: George Shen <george.shen@amd.com>
Tested-by: Mark Broadworth <mark.broadworth@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
[Why]
DSC config is calculated separately from DML calculations.
DML should use these separately calculated DSC params. The issue is
that the calculated bpp is not properly propagated into DML.
[How]
Correctly used forced_bpp value in DML.
Reviewed-by: Alvin Lee <Alvin.Lee2@amd.com>
Acked-by: Alex Hung <alex.hung@amd.com>
Signed-off-by: George Shen <george.shen@amd.com>
Tested-by: Mark Broadworth <mark.broadworth@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
[Why]
DCN32 DSC delay calculation had an unintentional integer division,
resulting in a mismatch against the DML spreadsheet.
[How]
Cast numerator to double before performing the division.
Reviewed-by: Alvin Lee <Alvin.Lee2@amd.com>
Acked-by: Alex Hung <alex.hung@amd.com>
Signed-off-by: George Shen <george.shen@amd.com>
Tested-by: Mark Broadworth <mark.broadworth@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
The recent change brought a bug on SRIOV envrionment. It caused
unloading amdgpu failed on Guest VM. The reason is that the VF
FLR was requested while unloading amdgpu driver, but the VF FLR
of SRIOV sequence is wrong while removing PCI device.
For SRIOV, the guest driver should not trigger the whole XGMI hive
to do the reset. Host driver control how the device been reset.
Fixes: f5c7e77970 ("drm/amdgpu: Adjust removal control flow for smu v13_0_2")
Acked-by: Alex Deucher <alexander.deucher@amd.com>
Reviewed-by: Shaoyun Liu <Shaoyun.Liu@amd.com>
Signed-off-by: Gavin Wan <Gavin.Wan@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
If a system does not have swap and memory is under 100% usage,
amdgpu will fail to evict resources. Currently the suspend
carries on proceeding to reset the GPU:
```
[drm] evicting device resources failed
[drm:amdgpu_device_ip_suspend_phase2 [amdgpu]] *ERROR* suspend of IP block <vcn_v3_0> failed -12
[drm] free PSP TMR buffer
[TTM] Failed allocating page table
[drm] evicting device resources failed
amdgpu 0000:03:00.0: amdgpu: MODE1 reset
amdgpu 0000:03:00.0: amdgpu: GPU mode1 reset
amdgpu 0000:03:00.0: amdgpu: GPU smu mode1 reset
```
At this point if the suspend actually succeeded I think that amdgpu
would have recovered because the GPU would have power cut off and
restored. However the kernel fails to continue the suspend from the
memory pressure and amdgpu fails to run the "resume" from the aborted
suspend.
```
ACPI: PM: Preparing to enter system sleep state S3
SLUB: Unable to allocate memory on node -1, gfp=0xdc0(GFP_KERNEL|__GFP_ZERO)
cache: Acpi-State, object size: 80, buffer size: 80, default order: 0, min order: 0
node 0: slabs: 22, objs: 1122, free: 0
ACPI Error: AE_NO_MEMORY, Could not update object reference count (20210730/utdelete-651)
[drm:psp_hw_start [amdgpu]] *ERROR* PSP load kdb failed!
[drm:psp_resume [amdgpu]] *ERROR* PSP resume failed
[drm:amdgpu_device_fw_loading [amdgpu]] *ERROR* resume of IP block <psp> failed -62
amdgpu 0000:03:00.0: amdgpu: amdgpu_device_ip_resume failed (-62).
PM: dpm_run_callback(): pci_pm_resume+0x0/0x100 returns -62
amdgpu 0000:03:00.0: PM: failed to resume async: error -62
```
To avoid this series of unfortunate events, fail amdgpu's suspend
when the memory eviction fails. This will let the system gracefully
recover and the user can try suspend again when the memory pressure
is relieved.
Reported-by: post@davidak.de
Link: https://gitlab.freedesktop.org/drm/amd/-/issues/2223
Signed-off-by: Mario Limonciello <mario.limonciello@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Acked-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Use mes.sched_version, mes.kiq_version for debugfs as
mes.ucode_fw_version does not contain correct versioning information.
Signed-off-by: Graham Sider <Graham.Sider@amd.com>
Reviewed-by: Jack Xiao <Jack.Xiao@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
This patch to fix the gdm3 start failure with virual display:
/usr/libexec/gdm-x-session[1711]: (II) AMDGPU(0): Setting screen physical size to 270 x 203
/usr/libexec/gdm-x-session[1711]: (EE) AMDGPU(0): Failed to make import prime FD as pixmap: 22
/usr/libexec/gdm-x-session[1711]: (EE) AMDGPU(0): failed to set mode: Invalid argument
/usr/libexec/gdm-x-session[1711]: (WW) AMDGPU(0): Failed to set mode on CRTC 0
/usr/libexec/gdm-x-session[1711]: (EE) AMDGPU(0): Failed to enable any CRTC
gnome-shell[1840]: Running GNOME Shell (using mutter 42.2) as a X11 window and compositing manager
/usr/libexec/gdm-x-session[1711]: (EE) AMDGPU(0): failed to set mode: Invalid argument
vkms doesn't have modifiers support, set fb_modifiers_not_supported to bring the gdm back.
Signed-off-by: Yifan Zhang <yifan1.zhang@amd.com>
Acked-by: Guchun Chen <guchun.chen@amd.com>
Reviewed-by: Tim Huang <Tim.Huang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
swiotlb_max_segment used to return either the maximum size that swiotlb
could bounce, or for Xen PV PAGE_SIZE even if swiotlb could bounce buffer
larger mappings. This made i915 on Xen PV work as it bypasses the
coherency aspect of the DMA API and can't cope with bounce buffering
and this avoided bounce buffering for the Xen/PV case.
So instead of adding this hack back, check for Xen/PV directly in i915
for the Xen case and otherwise use the proper DMA API helper to query
the maximum mapping size.
Replace swiotlb_max_segment() calls with dma_max_mapping_size().
In i915_gem_object_get_pages_internal() no longer consider max_segment
only if CONFIG_SWIOTLB is enabled. There can be other (iommu related)
causes of specific max segment sizes.
Fixes: a2daa27c0c ("swiotlb: simplify swiotlb_max_segment")
Reported-by: Marek Marczykowski-Górecki <marmarek@invisiblethingslab.com>
Signed-off-by: Robert Beckett <bob.beckett@collabora.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
[hch: added the Xen hack, rewrote the changelog]
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20221020110308.1582518-1-hch@lst.de
(cherry picked from commit 78a07fe777)
Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Accessing the TypeC DKL PHY registers during modeset-commit,
-verification, DP link-retraining and AUX power well toggling is racy
due to these code paths being concurrent and the PHY register bank
selection register (HIP_INDEX_REG) being shared between PHY instances
(aka TC ports) and the bank selection being not atomic wrt. the actual
PHY register access.
Add the required locking around each PHY register bank selection->
register access sequence.
Kudos to Ville for noticing the race conditions.
v2:
- Add the DKL PHY register accessors to intel_dkl_phy.[ch]. (Jani)
- Make the DKL_REG_TC_PORT macro independent of PHY internals.
- Move initing the DKL PHY lock to a more logical place.
v3:
- Fix parameter reuse in the DKL_REG_TC_PORT definition.
- Document the usage of phy_lock.
v4:
- Fix adding TC_PORT_1 offset in the DKL_REG_TC_PORT definition.
Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
Cc: Jani Nikula <jani.nikula@intel.com>
Cc: <stable@vger.kernel.org> # v5.5+
Acked-by: Jani Nikula <jani.nikula@intel.com>
Signed-off-by: Imre Deak <imre.deak@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20221025114457.2191004-1-imre.deak@intel.com
(cherry picked from commit 89cb0ba4ce)
Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
drm_fb_build_fourcc_list() currently returns all emulated formats
unconditionally as long as the native format is among them, even though
not all combinations have conversion helpers. Although the list is
arguably provided to userspace in precedence order, userspace can pick
something out-of-order (and thus break when it shouldn't), or simply
only support a format that is unsupported (and thus think it can work,
which results in the appearance of a hang as FB blits fail later on,
instead of the initialization error you'd expect in this case).
Add checks to filter the list of emulated formats to only those
supported for conversion to the native format. This presumes that there
is a single native format (only the first is checked, if there are
multiple). Refactoring this API to drop the native list or support it
properly (by returning the appropriate emulated->native mapping table)
is left for a future patch.
The simpledrm driver is left as-is with a full table of emulated
formats. This keeps all currently working conversions available and
drops all the broken ones (i.e. this a strict bugfix patch, adding no
new supported formats nor removing any actually working ones). In order
to avoid proliferation of emulated formats, future drivers should
advertise only XRGB8888 as the sole emulated format (since some
userspace assumes its presence).
This fixes a real user regression where the ?RGB2101010 support commit
started advertising it unconditionally where not supported, and KWin
decided to start to use it over the native format and broke, but also
the fixes the spurious RGB565/RGB888 formats which have been wrongly
unconditionally advertised since the dawn of simpledrm.
Fixes: 6ea966fca0 ("drm/simpledrm: Add [AX]RGB2101010 formats")
Fixes: 11e8f5fd22 ("drm: Add simpledrm driver")
Cc: stable@vger.kernel.org
Signed-off-by: Hector Martin <marcan@marcan.st>
Acked-by: Pekka Paalanen <pekka.paalanen@collabora.com>
Reviewed-by: Thomas Zimmermann <tzimmermann@suse.de>
Signed-off-by: Thomas Zimmermann <tzimmermann@suse.de>
Link: https://patchwork.freedesktop.org/patch/msgid/20221027135711.24425-1-marcan@marcan.st
The vop2 driver needs to explicitly disable the planes if the crtc is
disabled. Unless the planes are explicitly disabled, the address of the
last framebuffer is kept in the registers of the VOP2. When re-enabling
the encoder after it has been disabled by the driver, the VOP2 will
start and read the framebuffer that has been freed but is still pointed
to by the register. The iommu will catch these read accesses and print
errors.
Explicitly disable the planes when the crtc is disabled to reset the
registers.
Signed-off-by: Michael Tretter <m.tretter@pengutronix.de>
Signed-off-by: Heiko Stuebner <heiko@sntech.de>
Link: https://patchwork.freedesktop.org/patch/msgid/20221028095206.2136601-3-m.tretter@pengutronix.de
If the vop2_plane_atomic_disable function is called with NULL as a
state, accessing the old_pstate runs into a null pointer exception.
However, the drm_atomic_helper_disable_planes_on_crtc function calls the
atomic_disable callback with state NULL.
Allow to disable a plane without passing a plane state by checking the
old_pstate only if a state is passed.
Signed-off-by: Michael Tretter <m.tretter@pengutronix.de>
Signed-off-by: Heiko Stuebner <heiko@sntech.de>
Link: https://patchwork.freedesktop.org/patch/msgid/20221028095206.2136601-2-m.tretter@pengutronix.de
When switching to the generic fbdev infrastructure, it was missed that
framebuffers were created with the alloc_kmap parameter to
rockchip_gem_create_object() set to true. The generic infrastructure
calls this via the .dumb_create() driver operation and thus creates a
buffer without an associated kmap.
alloc_kmap only makes a difference on devices without an IOMMU, but when
it is missing rockchip_gem_prime_vmap() fails and the framebuffer cannot
be used.
Detect the case where a buffer is being allocated for the framebuffer
and ensure a kernel mapping is created in this case.
Fixes: 24af7c34b2 ("drm/rockchip: use generic fbdev setup")
Reported-by: Johan Jonker <jbx6244@gmail.com>
Cc: Thomas Zimmermann <tzimmermann@suse.de>
Signed-off-by: John Keeping <john@metanate.com>
Signed-off-by: Heiko Stuebner <heiko@sntech.de>
Link: https://patchwork.freedesktop.org/patch/msgid/20221020181248.2497065-1-john@metanate.com
If we fail to attach the first time (especially: EPROBE_DEFER), we fail
to clean up 'usage_mode', and thus will fail to attach on any subsequent
attempts, with "dsi controller already in use".
Re-set to DW_DSI_USAGE_IDLE on attach failure.
This is especially common to hit when enabling asynchronous probe on a
duel-DSI system (such as RK3399 Gru/Scarlet), such that we're more
likely to fail dw_mipi_dsi_rockchip_find_second() the first time.
Fixes: 71f68fe7f1 ("drm/rockchip: dsi: add ability to work as a phy instead of full dsi")
Cc: <stable@vger.kernel.org>
Signed-off-by: Brian Norris <briannorris@chromium.org>
Signed-off-by: Heiko Stuebner <heiko@sntech.de>
Link: https://patchwork.freedesktop.org/patch/msgid/20221019170255.1.Ia68dfb27b835d31d22bfe23812baf366ee1c6eac@changeid
When the avdd-0v9 or avdd-1v8 supply are not yet available, EPROBE_DEFER
is returned by rockchip_hdmi_parse_dt(). This causes the following error
message to be printed multiple times:
dwhdmi-rockchip fe0a0000.hdmi: [drm:dw_hdmi_rockchip_bind [rockchipdrm]] *ERROR* Unable to parse OF data
Fix that by not printing the message when rockchip_hdmi_parse_dt()
returns -EPROBE_DEFER.
Fixes: ca80c4eb4b ("drm/rockchip: dw_hdmi: add regulator support")
Signed-off-by: Aurelien Jarno <aurelien@aurel32.net>
Reviewed-by: Dmitry Osipenko <dmitry.osipenko@collabora.com>
Signed-off-by: Heiko Stuebner <heiko@sntech.de>
Link: https://patchwork.freedesktop.org/patch/msgid/20220926203752.5430-1-aurelien@aurel32.net
In the S2idle suspend/resume phase the gfxoff is keeping functional so
some IP blocks will be likely to reinitialize at gfxoff entry and that
will result in failing to program GC registers.Therefore, let disallow
gfxoff until AMDGPU IPs reinitialized completely.
Signed-off-by: Prike Liang <Prike.Liang@amd.com>
Acked-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Cc: stable@vger.kernel.org # 5.15.x
This file was split in commit 5d945cbcd4
("drm/amd/display: Create a file dedicated to planes") and the logic in
dm_plane_format_mod_supported() function got changed by a switch logic.
That change broke drm_plane modifiers setting on series 5000 APUs
(tested on OXP mini AMD 5800U and HP Dev One 5850U PRO)
leading to Gamescope not working as reported on GitHub[1]
To reproduce the issue, enter a TTY and run:
$ gamescope -- vkcube
With said commit applied it will abort. This one restores the old logic,
fixing the issue that affects Gamescope.
[1](https://github.com/Plagman/gamescope/issues/624)
Cc: <stable@vger.kernel.org> # 6.0.x
Signed-off-by: Joaquín Ignacio Aramendía <samsagax@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
