platform/kernel/linux-rpi.git
4 months agofbdev: Allow client to request a particular /dev/fbN node
Dave Stevenson [Mon, 9 Oct 2023 15:32:45 +0000 (16:32 +0100)]
fbdev: Allow client to request a particular /dev/fbN node

Add a flag custom_fb_num to denote that the client has
requested a specific fbdev node number via node.

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
4 months agodts: bcm2712: Use the new model name
Phil Elwell [Tue, 17 Oct 2023 16:28:11 +0000 (17:28 +0100)]
dts: bcm2712: Use the new model name

"Model B" is no more - "Raspberry Pi 5" is the official name.

Signed-off-by: Phil Elwell <phil@raspberrypi.com>
4 months agodrivers: media: rp1_cfe: Fix link validate test for pixel format
Naushir Patuck [Tue, 17 Oct 2023 08:35:44 +0000 (09:35 +0100)]
drivers: media: rp1_cfe: Fix link validate test for pixel format

Now that we have removed unique PISP media bus codes, the cfe format
table has multiple entries with the same media bus code for 16-bit
formats. The test in cfe_video_link_validate() did not account for this.
Fix it by testing the media bus code and the V4L2 pixelformat 4cc
together.

As a drive-by, ensure we have a valid CSI2 datatype id when programming
the hardware block.

Signed-off-by: Naushir Patuck <naush@raspberrypi.com>
4 months agodrm/vc4: Correct address offset for planes with src_[xy] offsets
Dave Stevenson [Mon, 16 Oct 2023 11:13:38 +0000 (12:13 +0100)]
drm/vc4: Correct address offset for planes with src_[xy] offsets

11cf37e741b4 switched to using drm_fb_dma_get_gem_addr instead of
drm_fb_dma_get_gem_obj and adding fb->offset[].

However the tiled formats need to compute the offset in a more
involved manner than drm_fb_dma_get_gem_addr applies, and we
were ending up with the offset for src_[xy] being applied twice.

Switch back to using drm_fb_dma_get_gem_obj and fully computing
the offsets ourselves.

Fixes: 11cf37e741b4 ("drm/vc4: Move the buffer offset out of the vc4_plane_state")
Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
4 months agovc_mem: Add the DMA memcpy support from bcm2708_fb
Phil Elwell [Sat, 14 Oct 2023 13:57:49 +0000 (14:57 +0100)]
vc_mem: Add the DMA memcpy support from bcm2708_fb

bcm2708_fb is disabled by the vc4-kms-v3d overlay, which means that the
DMA memcpy support it provides is not available to allow vclog to read
the VC logs from the top 16MB on Pi 2 and Pi 3. Add the code to the
vc_mem driver, which will still be enabled.

It ought to be possible to do a proper DMA_MEM_TO_MEM copy via the
generic DMA customer API, but that can be a later step.

Signed-off-by: Phil Elwell <phil@raspberrypi.com>
4 months agodts: bcm2712: Set default I2C baudrates to 100kHz
Phil Elwell [Mon, 16 Oct 2023 08:06:25 +0000 (09:06 +0100)]
dts: bcm2712: Set default I2C baudrates to 100kHz

The RP1 I2C interfaces were being left with their default clock rates,
apparently 400kHz.

Signed-off-by: Phil Elwell <phil@raspberrypi.com>
4 months agooverlays: mcp23017: allow specification of the i2c bus
Janis Streib [Sun, 15 Oct 2023 19:08:40 +0000 (21:08 +0200)]
overlays: mcp23017: allow specification of the i2c bus

Analogous to i2c-rtc-overlay.dts

See: https://github.com/raspberrypi/linux/pull/5650

4 months agoRevert "bcm2708_fb: Hack out dma support"
Dom Cobley [Thu, 12 Oct 2023 14:52:38 +0000 (15:52 +0100)]
Revert "bcm2708_fb: Hack out dma support"

This reverts commit d60956f30586ced37e2152f2ba338f547e6c64f7.

4 months agoRevert "bcm2708_fb: Fix more build warnings"
Dom Cobley [Thu, 12 Oct 2023 14:52:24 +0000 (15:52 +0100)]
Revert "bcm2708_fb: Fix more build warnings"

This reverts commit 7b11991e91e65ebeb315986fae05d363663d24c4.

4 months agoinput: touchscreen: edt-ft5x06: Suppress bogus data on startup
Nick Hollinghurst [Wed, 11 Oct 2023 14:14:59 +0000 (15:14 +0100)]
input: touchscreen: edt-ft5x06: Suppress bogus data on startup

When polled without the use of IRQ, FT5x06 registers may return
undefined initial data, causing unwanted touches or event spamming.
A simple way to filter this out is to suppress touches until the
TD_STATUS register changes for the first time.

Increase the delay before first polling to 300ms, to avoid
transient I2C read flakiness that seems to occur after reset.

Signed-off-by: Nick Hollinghurst <nick.hollinghurst@raspberrypi.com>
4 months agoconfigs: Enable more sound devices
Phil Elwell [Fri, 6 Oct 2023 13:30:50 +0000 (14:30 +0100)]
configs: Enable more sound devices

See: https://github.com/raspberrypi/linux/issues/4179

Signed-off-by: Phil Elwell <phil@raspberrypi.com>
4 months agovc04_services: bcm2835-codec: Correct alignment requirements for YUYV
Dave Stevenson [Wed, 11 Oct 2023 14:05:38 +0000 (15:05 +0100)]
vc04_services: bcm2835-codec: Correct alignment requirements for YUYV

The firmware wants the YUYV format stride alignment to be to a multiple
of 32pixels / 64 bytes. The kernel driver was configuring it to a multiple
of 16 pixels / 32 bytes, which then failed when it tried starting to
stream.

Correct the alignment requirements.

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
4 months agodrivers: media: rp1_cfe: Remove PISP specific MBUS formats
Naushir Patuck [Tue, 10 Oct 2023 11:41:15 +0000 (12:41 +0100)]
drivers: media: rp1_cfe: Remove PISP specific MBUS formats

Remove the MEDIA_BUS_FMT_PISP* format codcs entirely. For the image
pad formats, use the 16-bit Bayer format mbus codes instead. For the
config and stats pad formats, use MEDIA_BUS_FMT_FIXED.

Signed-off-by: Naushir Patuck <naush@raspberrypi.com>
4 months agodts/bcm2712: Add iommu to rpivid
John Cox [Mon, 2 Oct 2023 14:15:13 +0000 (15:15 +0100)]
dts/bcm2712: Add iommu to rpivid

Add iommu to rpivid so it can cope with scatter/gather

Signed-off-by: John Cox <jc@kynesim.co.uk>
4 months agomedia/rpivid: Allow use of iommu in rpivid
John Cox [Mon, 2 Oct 2023 14:12:52 +0000 (15:12 +0100)]
media/rpivid: Allow use of iommu in rpivid

In order to use iommu on hevc set dma mask_and_coherent in probe.
I am assured dma_set_mask_and_coherent is benign on Pi4 (which has
no iommu) and it seems to be so in practice.
Also adds a bit of debug to make internal buffer allocation failure
easier to spot in future

Signed-off-by: John Cox <jc@kynesim.co.uk>
4 months agomedia: i2c: ov9282: Read chip ID via 2 reads
Dave Stevenson [Wed, 11 Oct 2023 10:12:41 +0000 (11:12 +0100)]
media: i2c: ov9282: Read chip ID via 2 reads

Vision Components have made an OV9281 module which blocks reading
back the majority of registers to comply with NDAs, and in doing
so doesn't allow auto-increment register reading as used when
reading the chip ID.

Use two reads and manually combine the results.

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
4 months agoRevert "pinctrl-bcm2835: Set base to 0 give expected gpio numbering"
Phil Elwell [Fri, 6 Oct 2023 16:35:43 +0000 (17:35 +0100)]
Revert "pinctrl-bcm2835: Set base to 0 give expected gpio numbering"

This reverts commit 2afa8798ada2b03fd6ae3ef5c5dfcb39a5877ad8.

4 months agogpio: brcmstb: Use dynamic GPIO base numbers
Phil Elwell [Fri, 6 Oct 2023 15:30:35 +0000 (16:30 +0100)]
gpio: brcmstb: Use dynamic GPIO base numbers

Forcing a gpiochip to have a fixed base number now leads to a warning
message. Remove the need to do so by calculating hwirq numbers based
on bank numbers.

Signed-off-by: Phil Elwell <phil@raspberrypi.com>
Fixes: 3b0213d56eb7 ("gpio: Add GPIO support for Broadcom STB SoCs")

4 months agoconfig: Regenerate defconfigs
Phil Elwell [Fri, 6 Oct 2023 07:59:42 +0000 (08:59 +0100)]
config: Regenerate defconfigs

Signed-off-by: Phil Elwell <phil@raspberrypi.com>
4 months agoconfig: 2712: Update defconfig
Dom Cobley [Tue, 19 Sep 2023 14:56:20 +0000 (15:56 +0100)]
config: 2712: Update defconfig

4 months agodts: 2712: Update for device tree
Phil Elwell [Thu, 5 Oct 2023 09:06:18 +0000 (10:06 +0100)]
dts: 2712: Update for device tree

dtoverlays: Fix up edt5406 entries to match with vc4-kms-dsi-7inch

vc4-kms-dsi-7inch expects the touch fragment to be named ts_i2c_frag,
but edt5406 didn't do this.

dt: Add DSI1 and DSI2 aliases to 2712

In order to keep the DRM names consistent as DSI-1 and DSI-2, add
aliases to the Pi5 DT.

Signed-off-by: Phil Elwell <phil@raspberrypi.com>
4 months agoREADME: Show rpi-6.5.y build status
Phil Elwell [Fri, 29 Sep 2023 11:21:29 +0000 (12:21 +0100)]
README: Show rpi-6.5.y build status

Signed-off-by: Phil Elwell <phil@raspberrypi.com>
4 months ago.github/workflows: Add a bcm2712 build configuration
Phil Elwell [Thu, 28 Sep 2023 14:45:05 +0000 (15:45 +0100)]
.github/workflows: Add a bcm2712 build configuration

Signed-off-by: Phil Elwell <phil@raspberrypi.com>
4 months agovc4/drm: Remove the clear of SCALER_DISPBKGND_FILL
Dom Cobley [Wed, 4 Oct 2023 15:02:39 +0000 (16:02 +0100)]
vc4/drm: Remove the clear of SCALER_DISPBKGND_FILL

Since "drm/vc4: hvs: Support BCM2712 HVS" booting Pi4
with dual 4kp30 displays connected fails with:
vc4-drm gpu: [drm] *ERROR* [CRTC:107:pixelvalve-4] flip_done timed out

It has been tracked down to the referenced commit adding a
path to clear the SCALER_DISPBKGND_FILL when not required.

Dual 4kp30 works with a core clock of 297MHz when background fill
is enabled, but requires a higher value with it disabled.
320MHz still fails, while 330MHz seems okay.

Lets always enable background fill for Pi0-4.

Fixes: e84da235223d ("drm/vc4: hvs: Support BCM2712 HVS")

Signed-off-by: Dom Cobley <popcornmix@gmail.com>
4 months agodrm: Look for an alias for the displays to use as the DRM device name
Dave Stevenson [Thu, 28 Sep 2023 17:27:09 +0000 (18:27 +0100)]
drm: Look for an alias for the displays to use as the DRM device name

Allow DT aliases of eg DSI2 to force make DRM allocate the
display with the requested name.

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
4 months agomedia: i2c: Move Kconfig entry for IMX477 to the camera sensor section
Naushir Patuck [Wed, 4 Oct 2023 10:09:10 +0000 (11:09 +0100)]
media: i2c: Move Kconfig entry for IMX477 to the camera sensor section

It was accidentally placed in the audio decoder section.

Signed-off-by: Naushir Patuck <naush@raspberrypi.com>
4 months agomedia: rp1: cfe: Fail streaming if FE_CONFIG node is not enabled
Naushir Patuck [Wed, 4 Oct 2023 08:39:59 +0000 (09:39 +0100)]
media: rp1: cfe: Fail streaming if FE_CONFIG node is not enabled

When the FE is enabled, ensure that the FE_CONFIG node is enabled.
Otherwise fail cfe_start_streaming() entirely.

Signed-off-by: Naushir Patuck <naush@raspberrypi.com>
4 months agomedia: rp1: cfe: Fix default meta format's field
Tomi Valkeinen [Mon, 2 Oct 2023 11:38:07 +0000 (14:38 +0300)]
media: rp1: cfe: Fix default meta format's field

Set default meta format's field to V4L2_FIELD_NONE, instead of zeroing
it which indicates V4L2_FIELD_ANY. Metadata doesn't have fields, so NONE
makes sense, and furthermore the default v4l2 link validation will check
for matching fields, or that the sink field is NONE.

Signed-off-by: Tomi Valkeinen <tomi.valkeinen@ideasonboard.com>
4 months agomedia: rp1: fe: Fix default mbus code
Tomi Valkeinen [Fri, 29 Sep 2023 10:29:15 +0000 (13:29 +0300)]
media: rp1: fe: Fix default mbus code

When pisp_fe_pad_set_fmt() is given an mbus code that CFE does not
support, it currently defaults to MEDIA_BUS_FMT_SBGGR10_1X10. This is
not correct, as FE does not support SBGGR10.

Set the default to MEDIA_BUS_FMT_SRGGB16_1X16 instead.

Signed-off-by: Tomi Valkeinen <tomi.valkeinen@ideasonboard.com>
4 months agomedia: rp1: csi2: Set values for enum csi2_mode
Tomi Valkeinen [Fri, 29 Sep 2023 09:57:23 +0000 (12:57 +0300)]
media: rp1: csi2: Set values for enum csi2_mode

Set hardcoded values for enum csi2_mode, as the values will be
programmed to HW registers.

Signed-off-by: Tomi Valkeinen <tomi.valkeinen@ideasonboard.com>
4 months agomedia: rp1: cfe: Drop unused field
Tomi Valkeinen [Thu, 28 Sep 2023 14:04:07 +0000 (17:04 +0300)]
media: rp1: cfe: Drop unused field

Drop 'sensor_embedded_data' field, as it is unused.

Signed-off-by: Tomi Valkeinen <tomi.valkeinen@ideasonboard.com>
4 months agomedia: rp1: csi2: Track CSI-2 errors
Tomi Valkeinen [Thu, 21 Sep 2023 15:18:53 +0000 (18:18 +0300)]
media: rp1: csi2: Track CSI-2 errors

Track the errors from the CSI-2 receiver: overflows and discards. These
are recorded in a table which can be read by the userspace via debugfs.

As tracking the errors may cause much more interrupt load, the tracking
needs to be enabled with a module parameter.

Note that the recording is not perfect: we only record the last
discarded DT for each discard type, instead of recording all of them.
This means that e.g. if the device is discarding two unmatched DTs, the
debugfs file only shows the last one recorded. Recording all of them
would need a more sophisticated recording system to avoid the need of a
very large table, or dynamic allocation.

Signed-off-by: Tomi Valkeinen <tomi.valkeinen@ideasonboard.com>
4 months agomedia: rp1: cfe: Add verbose debug module parameter
Tomi Valkeinen [Wed, 4 Oct 2023 07:19:47 +0000 (10:19 +0300)]
media: rp1: cfe: Add verbose debug module parameter

Expose the verbose debug flag as a module parameter.

Signed-off-by: Tomi Valkeinen <tomi.valkeinen@ideasonboard.com>
4 months agomedia: rp1: Add back reg write debug prints
Tomi Valkeinen [Fri, 22 Sep 2023 09:39:33 +0000 (12:39 +0300)]
media: rp1: Add back reg write debug prints

Add back debug prints in csi2 and pisp_fe reg_write() functions, but use
the 'irq' variants to avoid spamming in normal situation.

Signed-off-by: Tomi Valkeinen <tomi.valkeinen@ideasonboard.com>
4 months agomedia: rp1: cfe: Rename xxx_dbg_irq() to xxx_dbg_verbose()
Tomi Valkeinen [Wed, 4 Oct 2023 07:12:37 +0000 (10:12 +0300)]
media: rp1: cfe: Rename xxx_dbg_irq() to xxx_dbg_verbose()

Rename the xxx_dbg_irq() macros to xxx_dbg_verbose(), as they can be
used to verbose debugs outside irq context too.

Signed-off-by: Tomi Valkeinen <tomi.valkeinen@ideasonboard.com>
4 months agomedia: rp1: cfe: Fix verbose debug print
Tomi Valkeinen [Fri, 22 Sep 2023 09:41:35 +0000 (12:41 +0300)]
media: rp1: cfe: Fix verbose debug print

The debug print in cfe_schedule_next_csi2_job() is printed every frame,
and should thus use cfe_dbg_irq() to avoid spamming, rather than cfe_dbg().

Signed-off-by: Tomi Valkeinen <tomi.valkeinen@ideasonboard.com>
4 months agomedia: rp1: fe: Use ~0, not -1, when working with unsigned values
Tomi Valkeinen [Thu, 21 Sep 2023 13:03:07 +0000 (16:03 +0300)]
media: rp1: fe: Use ~0, not -1, when working with unsigned values

Use ~0, not -1, when working with unsigned values (-1 is not unsigned).

Signed-off-by: Tomi Valkeinen <tomi.valkeinen@ideasonboard.com>
4 months agomedia: rp1: csi2: Fix missing reg writes
Tomi Valkeinen [Thu, 28 Sep 2023 07:42:22 +0000 (10:42 +0300)]
media: rp1: csi2: Fix missing reg writes

The driver has two places where it writes a register based on a
condition, and when that condition is false, the driver presumes that
the register has the reset value. This is not a good idea, so fix those
places to always write the register.

Signed-off-by: Tomi Valkeinen <tomi.valkeinen@ideasonboard.com>
4 months agomedia: rp1: cfe: Fix width & height in cfe_start_channel()
Tomi Valkeinen [Wed, 27 Sep 2023 13:00:39 +0000 (16:00 +0300)]
media: rp1: cfe: Fix width & height in cfe_start_channel()

The logic for handling width & height in cfe_start_channel() is somewhat
odd and, afaics, broken. The code reads:

bool start_fe = is_fe_enabled(cfe) &&
                test_all_nodes(cfe, NODE_ENABLED, NODE_STREAMING);

if (start_fe || is_image_output_node(node)) {
        width = node->fmt.fmt.pix.width;
        height = node->fmt.fmt.pix.height;
}

cfe_start_channel() is called for all video nodes that will be used. So
this means that if, say, fe_stats is enabled as the last node, start_fe
will be true, and width and height will be taken from fe_stats' node.
The width and height will thus contain garbage, which then gets
programmed to the csi2 registers.

It seems that this often still works fine, though, probably if the width
& height are large enough.

Drop the above code, and instead get the width & height from the csi2
subdev's sink pad for the csi2 channel that is used. For metadata the
width & height will be 0 as before.

Signed-off-by: Tomi Valkeinen <tomi.valkeinen@ideasonboard.com>
4 months agomedia: rp1: cfe: Fix use of freed memory on errors
Tomi Valkeinen [Thu, 28 Sep 2023 08:33:53 +0000 (11:33 +0300)]
media: rp1: cfe: Fix use of freed memory on errors

cfe_probe_complete() calls cfe_put() on both success and fail code paths.
This works for the success path, but causes the cfe_device struct to be
freed, even if it will be used later in the teardown code.

Fix this by making the ref handling a bit saner: Let the video nodes
have the refs as they do now, but also keep a ref in the "main" driver,
released only at cfe_remove() time. This way the driver does not depend
on the video nodes keeping the refs.

Signed-off-by: Tomi Valkeinen <tomi.valkeinen@ideasonboard.com>
4 months agomedia: i2c: imx477: Fix locking in imx477_init_controls()
Tomi Valkeinen [Mon, 11 Sep 2023 09:17:25 +0000 (12:17 +0300)]
media: i2c: imx477: Fix locking in imx477_init_controls()

The driver does not lock the imx477 mutex when calling
imx477_set_framing_limits(), leading to:

WARNING: CPU: 3 PID: 426 at drivers/media/v4l2-core/v4l2-ctrls-api.c:934 __v4l2_ctrl_modify_range+0x1a0/0x210 [
videodev]

Fix this by taking the lock.

Signed-off-by: Tomi Valkeinen <tomi.valkeinen@ideasonboard.com>
4 months agodrm/panel: waveshare: Fix up timings for 10.1" panel
Dave Stevenson [Fri, 28 Jul 2023 17:10:53 +0000 (18:10 +0100)]
drm/panel: waveshare: Fix up timings for 10.1" panel

The 10.1" panel doesn't work with the timings defined. vc4
will always have been fixing up the timing due to the limited
integer divider, so compute the fixed up mode and use it
directly.

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
4 months agodrm/panel: simple: Alter the timing for the Pi 7" DSI display
Dave Stevenson [Fri, 28 Jul 2023 16:40:27 +0000 (17:40 +0100)]
drm/panel: simple: Alter the timing for the Pi 7" DSI display

vc4 has always fixed up the timing, so the values defined have
never actually appeared on the wire.
The display appears to want a slightly longer HFP, so extend
the timings and recompute the clock to give the same frame rate.

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
4 months agodrm/vc4: Assign LBM memory during atomic_flush.
Dave Stevenson [Thu, 31 Aug 2023 10:45:38 +0000 (11:45 +0100)]
drm/vc4: Assign LBM memory during atomic_flush.

Avoid double buffering LBM allocations by making the
allocation a single alloc per crtc at atomic_flush.

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
4 months agodrm/vc4: kms: Avoid setting core and disp clocks for hdmi modes
Dom Cobley [Mon, 17 Jul 2023 16:45:32 +0000 (17:45 +0100)]
drm/vc4: kms: Avoid setting core and disp clocks for hdmi modes

On 2712, the firmware always runs these clock at a speed sufficient
for dual 4kp60.

The requests here prevent the gpu from going into its lowest voltage
mode, so just skip the clock requests.

With this applied the idle voltage on my pi 5 reduces from 0.7424V
to 0.72V.

Signed-off-by: Dom Cobley <popcornmix@gmail.com>
4 months agodrm/vc4: tests: Introduce a test for LBM buffer size
Maxime Ripard [Fri, 14 Apr 2023 11:43:32 +0000 (13:43 +0200)]
drm/vc4: tests: Introduce a test for LBM buffer size

The BCM2712 comes with a different LBM size computation than the
previous generations, so let's add the few examples provided as kunit
tests to make sure we always satisfy those requirements.

Signed-off-by: Maxime Ripard <maxime@cerno.tech>
4 months agodrm/vc4: tests: Support a few more plane formats
Maxime Ripard [Fri, 14 Apr 2023 10:59:05 +0000 (12:59 +0200)]
drm/vc4: tests: Support a few more plane formats

We'll start testing our planes code in situations where we will use more
than XRGB8888, so let's add a few common pixel formats.

Signed-off-by: Maxime Ripard <maxime@cerno.tech>
4 months agodrm/vc4: tests: Add helper to add a new plane to a state
Maxime Ripard [Fri, 14 Apr 2023 10:57:53 +0000 (12:57 +0200)]
drm/vc4: tests: Add helper to add a new plane to a state

We'll start to add some tests for the plane state logic, so let's create
a helper to add a plane to an existing atomic state.

Signed-off-by: Maxime Ripard <maxime@cerno.tech>
4 months agodrm/vc4: tests: Add function to lookup a plane for a CRTC
Maxime Ripard [Fri, 14 Apr 2023 09:26:58 +0000 (11:26 +0200)]
drm/vc4: tests: Add function to lookup a plane for a CRTC

Some tests will need to find a plane to run a test on for a given CRTC.
Let's create a small helper to do that.

Signed-off-by: Maxime Ripard <maxime@cerno.tech>
4 months agodrm/vc4: tests: Use custom plane state for mock
Maxime Ripard [Fri, 14 Apr 2023 09:24:37 +0000 (11:24 +0200)]
drm/vc4: tests: Use custom plane state for mock

The current mock planes were just using the regular drm_plane_state,
while the driver expect struct vc4_plane_state that subclasses
drm_plane_state.

Hook the proper implementations of reset, duplicate_state, destroy and
atomic_check to create vc4_plane_state.

Signed-off-by: Maxime Ripard <maxime@cerno.tech>
4 months agodrm/vc4: fkms: Rename plane related functions
Maxime Ripard [Fri, 14 Apr 2023 09:21:34 +0000 (11:21 +0200)]
drm/vc4: fkms: Rename plane related functions

The name collide with the Full KMS functions that are going to be made
public.

Signed-off-by: Maxime Ripard <maxime@cerno.tech>
4 months agodrm/vc4: tests: Add tests for BCM2712 PixelValve Muxing
Maxime Ripard [Fri, 14 Apr 2023 09:14:22 +0000 (11:14 +0200)]
drm/vc4: tests: Add tests for BCM2712 PixelValve Muxing

The BCM2712 has a simpler pipeline than the BCM2711, and thus the muxing
requirements are different. Create some tests to make sure we get proper
muxing decisions.

Signed-off-by: Maxime Ripard <maxime@cerno.tech>
4 months agodrm/vc4: tests: Add BCM2712 mock driver
Maxime Ripard [Fri, 17 Feb 2023 12:38:10 +0000 (13:38 +0100)]
drm/vc4: tests: Add BCM2712 mock driver

The BCM2712 has a simpler pipeline that can only output to a writeback
connector and two HDMI controllers.

Let's allow our kunit tests to create a mock of that pipeline.

Signed-off-by: Maxime Ripard <maxime@cerno.tech>
4 months agodrm/vc4: tests: Return the allocated output
Maxime Ripard [Fri, 24 Mar 2023 09:02:59 +0000 (10:02 +0100)]
drm/vc4: tests: Return the allocated output

Some tests will need to retrieve the output that was just allocated by
vc4_mock_atomic_add_output().

Instead of making them look them up in the DRM device, we can simply
make vc4_mock_atomic_add_output() return an error pointer that holds the
allocated output instead of the error code.

Signed-off-by: Maxime Ripard <maxime@cerno.tech>
4 months agodrm/vc4: tests: Drop drm parameter for vc4_find_crtc_for_encoder
Maxime Ripard [Fri, 24 Mar 2023 08:58:15 +0000 (09:58 +0100)]
drm/vc4: tests: Drop drm parameter for vc4_find_crtc_for_encoder

The DRM device pointer and the DRM encoder pointer are redundant, since
the latter is attached to the former and we can just follow the
drm_encoder->dev pointer.

Let's remove the drm_device pointer argument.

Signed-off-by: Maxime Ripard <maxime@cerno.tech>
4 months agodrm/vc4: tests: Switch generation mockup to a switch
Maxime Ripard [Fri, 17 Feb 2023 12:35:16 +0000 (13:35 +0100)]
drm/vc4: tests: Switch generation mockup to a switch

Testing whether the VideoCore generation we want to mock is vc5 or vc4
worked so far, but will be difficult to extend to support BCM2712 (VC6).

Convert to a switch.

Signed-off-by: Maxime Ripard <maxime@cerno.tech>
4 months agodrm/vc4: Add additional warn_on
Maxime Ripard [Tue, 21 Feb 2023 13:38:32 +0000 (14:38 +0100)]
drm/vc4: Add additional warn_on

Some code path in vc4 are conditional to a generation and cannot be
executed on others. Let's put a WARN_ON if that ever happens.

Signed-off-by: Maxime Ripard <maxime@cerno.tech>
4 months agodrm/vc4: txp: Add BCM2712 MOPLET support
Maxime Ripard [Mon, 20 Feb 2023 16:16:01 +0000 (17:16 +0100)]
drm/vc4: txp: Add BCM2712 MOPLET support

The BCM2712 features a simpler TXP called MOPLET. Let's add support for
it.

Signed-off-by: Maxime Ripard <maxime@cerno.tech>
4 months agodrm/vc4: txp: Add support for BCM2712 MOP
Maxime Ripard [Thu, 27 Apr 2023 07:30:49 +0000 (09:30 +0200)]
drm/vc4: txp: Add support for BCM2712 MOP

The BCM2712 has an evolution of what used to be called TXP in the
earlier SoCs, but is now called MOP.

There's a few differences still, so we can add a new compatible to deal
with them easily.

Signed-off-by: Maxime Ripard <maxime@cerno.tech>
4 months agodrm/vc4: txp: Add a new TXP encoder type
Maxime Ripard [Thu, 27 Apr 2023 09:49:28 +0000 (11:49 +0200)]
drm/vc4: txp: Add a new TXP encoder type

Starting with BCM2712, we'll have a two TXP. Let's follow the HDMI
example and add two encoder types for TXP: TXP0 and TXP1.

Signed-off-by: Maxime Ripard <maxime@cerno.tech>
4 months agodrm/vc4: txp: Move the encoder type in the variant structure
Maxime Ripard [Thu, 27 Apr 2023 09:26:10 +0000 (11:26 +0200)]
drm/vc4: txp: Move the encoder type in the variant structure

We'll have multiple TXP instances in the BCM2712, so we can't use a
single encoder type anymore. Let's tie the encoder type to the
compatible.

Signed-off-by: Maxime Ripard <maxime@cerno.tech>
4 months agodrm/vc4: txp: Handle 40-bits DMA Addresses
Maxime Ripard [Fri, 14 Apr 2023 15:47:11 +0000 (17:47 +0200)]
drm/vc4: txp: Handle 40-bits DMA Addresses

The BCM2712 MOP and MOPLET can handle addresses larger than 32bits
through an extra register. We can easily support it and make it
conditional based on the compatible through a boolean in our variant
structure.

Signed-off-by: Maxime Ripard <maxime@cerno.tech>
4 months agodrm/vc4: txp: Add horizontal and vertical size offset toggle bit
Maxime Ripard [Thu, 27 Apr 2023 07:47:54 +0000 (09:47 +0200)]
drm/vc4: txp: Add horizontal and vertical size offset toggle bit

The new writeback controllers that can be found on the BCM2712 require
to have their horizontal and vertical size reduced by one.

Let's tie that behaviour to the compatible so we can support both the
new and old controllers.

Signed-off-by: Maxime Ripard <maxime@cerno.tech>
4 months agodrm/vc4: txp: Add byte enable toggle bit
Maxime Ripard [Thu, 27 Apr 2023 07:30:33 +0000 (09:30 +0200)]
drm/vc4: txp: Add byte enable toggle bit

The MOPLET doesn't have the BYTE_ENABLE field to set, but the TXP and
MOP do, so let's add a boolean to control whether or not we need to set
it.

Signed-off-by: Maxime Ripard <maxime@cerno.tech>
4 months agodrm/vc4: txp: Rename TXP data structure
Maxime Ripard [Tue, 25 Apr 2023 08:21:53 +0000 (10:21 +0200)]
drm/vc4: txp: Rename TXP data structure

The TXP data structure has a name too generic for the multiple variants
we'll have to support. Let's rename it to mention the SoC it applies to.

Signed-off-by: Maxime Ripard <maxime@cerno.tech>
4 months agodrm/vc4: txp: Introduce structure to deal with revision differences
Maxime Ripard [Tue, 25 Apr 2023 08:12:32 +0000 (10:12 +0200)]
drm/vc4: txp: Introduce structure to deal with revision differences

The BCM2712 will have several TXP with small differences. Let's add a
structure tied to the compatible to deal with those differences.

Signed-off-by: Maxime Ripard <maxime@cerno.tech>
4 months agodrm/vc4: hdmi: Add support for BCM2712 HDMI controllers
Maxime Ripard [Fri, 17 Feb 2023 14:34:30 +0000 (15:34 +0100)]
drm/vc4: hdmi: Add support for BCM2712 HDMI controllers

The HDMI controllers found in the BCM2712 are largely the ones found in
the BCM2711 with a different PHY.

There's some difference with how timings are split between registers,
and HDMI1 is now able to run at 4k/60Hz.

Signed-off-by: Maxime Ripard <maxime@cerno.tech>
4 months agodrm/vc4: crtc: Add support for BCM2712 PixelValves
Maxime Ripard [Fri, 17 Feb 2023 14:33:23 +0000 (15:33 +0100)]
drm/vc4: crtc: Add support for BCM2712 PixelValves

The PixelValves found on the BCM2712 are similar to the ones found in
the previous generation.

Compared to BCM2711, the pixelvalves only drive one HDMI controller each
and HDMI1 PixelValve has a FIFO long enough to support 4k at 60Hz.

Signed-off-by: Maxime Ripard <maxime@cerno.tech>
4 months agodrm/vc4: hvs: Support BCM2712 HVS
Maxime Ripard [Fri, 17 Feb 2023 14:25:16 +0000 (15:25 +0100)]
drm/vc4: hvs: Support BCM2712 HVS

The HVS found in the BCM2712, while having a similar role, is very
different from the one found in the previous SoCs. Indeed, the register
layout is fairly different, and the DLIST format is new as well.

Let's introduce the needed functions to support the new HVS.

Signed-off-by: Maxime Ripard <maxime@cerno.tech>
4 months agodrm/vc4: drv: Support BCM2712
Maxime Ripard [Fri, 17 Feb 2023 14:29:52 +0000 (15:29 +0100)]
drm/vc4: drv: Support BCM2712

The BCM2712 has an improved display pipeline, most notably with a
different HVS and only HDMI and writeback outputs.

Let's introduce it as a new VideoCore generation and compatible.

Signed-off-by: Maxime Ripard <maxime@cerno.tech>
4 months agodt-bindings: display: Add BCM2712 KMS driver bindings
Maxime Ripard [Fri, 17 Feb 2023 12:36:51 +0000 (13:36 +0100)]
dt-bindings: display: Add BCM2712 KMS driver bindings

The BCM2712 SoC comes with a new variation of the videocore display
pipeline. Let's create a new compatible for it.

Signed-off-by: Maxime Ripard <maxime@cerno.tech>
4 months agodt-bindings: display: Add BCM2712 MOPLET bindings
Maxime Ripard [Fri, 17 Feb 2023 12:36:36 +0000 (13:36 +0100)]
dt-bindings: display: Add BCM2712 MOPLET bindings

The BCM2712 has a MOPLET controller which is basically a TXP without the
transpose feature.

Express that by adding a new compatible for it.

Signed-off-by: Maxime Ripard <maxime@cerno.tech>
4 months agodt-bindings: display: Add BCM2712 MOP bindings
Maxime Ripard [Fri, 17 Feb 2023 12:36:36 +0000 (13:36 +0100)]
dt-bindings: display: Add BCM2712 MOP bindings

The BCM2712 has a MOP controller which is basically a new revision of
the TXP.

Express that by adding a new compatible for it.

Signed-off-by: Maxime Ripard <maxime@cerno.tech>
4 months agodt-bindings: display: Add BCM2712 PixelValve bindings
Maxime Ripard [Fri, 17 Feb 2023 12:36:27 +0000 (13:36 +0100)]
dt-bindings: display: Add BCM2712 PixelValve bindings

The BCM2712 has 3 different pixelvalves that are similar to the ones
found in the previous generations but with slightly different
capabilities.

Express that using a new set of compatibles.

Signed-off-by: Maxime Ripard <maxime@cerno.tech>
4 months agodt-bindings: display: Add BCM2712 HVS bindings
Maxime Ripard [Fri, 17 Feb 2023 12:36:14 +0000 (13:36 +0100)]
dt-bindings: display: Add BCM2712 HVS bindings

The BCM2712 has a completely different HVS than the previous
generations, so let's add a new compatible for it.

Signed-off-by: Maxime Ripard <maxime@cerno.tech>
4 months agodt-bindings: display: Add BCM2712 HDMI bindings
Maxime Ripard [Fri, 17 Feb 2023 12:36:02 +0000 (13:36 +0100)]
dt-bindings: display: Add BCM2712 HDMI bindings

The BCM2712 HDMI controller uses a slightly different HDMI controller
than the BCM2711, and a completely different PHY.

Let's introduce a new compatible for it.

Signed-off-by: Maxime Ripard <maxime@cerno.tech>
4 months agodrm/vc4: Enable SCALER_CONTROL early in HVS init
Tim Gover [Thu, 13 Jul 2023 16:47:22 +0000 (17:47 +0100)]
drm/vc4: Enable SCALER_CONTROL early in HVS init

Always enable SCALER_CONTROL before attempting other HVS
operations. It's safe to write to some parts of the HVS but
in general it's dangerous to do this because it can cause bus
lockups.

Signed-off-by: Tim Gover <tim.gover@raspberrypi.com>
4 months agodrm/vc4: hvs: Remove ABORT_ON_EMPTY flag
Dom Cobley [Thu, 22 Jun 2023 13:06:40 +0000 (14:06 +0100)]
drm/vc4: hvs: Remove ABORT_ON_EMPTY flag

ABORT_ON_EMPTY chooses whether the HVS abandons the current frame
when it experiences an underflow, or attempts to continue.

In theory the frame should be black from the point of underflow,
compared to a shift of sebsequent pixels to the left.

Unfortunately it seems to put the HVS is a bad state where it is not
possible to recover simply. This typically requires a reboot
following the 'flip done timed out message'.

Discussion with Broadcom has suggested we don't use this flag.
All their testing is done with it disabled.

Additionally setting BLANK_INSERT_EN causes the HDMI to output
blank pixels on an underflow which avoids it losing sync.

After this change a 'flip done timed out' due to sdram bandwidth
starvation or too low a clock is recoverable once the situation improves.

Signed-off-by: Dom Cobley <popcornmix@gmail.com>
4 months agodrm: vc4: Remove incorrect limit from hvs_dlist debugfs function
Dave Stevenson [Fri, 1 Sep 2023 12:45:08 +0000 (13:45 +0100)]
drm: vc4: Remove incorrect limit from hvs_dlist debugfs function

The debugfs function to dump dlists aborted at 256 bytes,
when actually the dlist memory is generally significantly
larger but varies based on SoC.

We already have the correct limit in __vc4_hvs_alloc, so
store it for use in the debugfs dlist function.

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
4 months agodrm/vc4: Fix dlist debug not resetting the next entry pointer
Dave Stevenson [Thu, 24 Aug 2023 14:36:21 +0000 (15:36 +0100)]
drm/vc4: Fix dlist debug not resetting the next entry pointer

The debug function to display the dlists didn't reset next_entry_start
when starting each display, so resulting in not stopping the
list at the correct place.

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
4 months agodrm/vc4: Move the buffer offset out of the vc4_plane_state
Dave Stevenson [Mon, 25 Sep 2023 15:57:07 +0000 (16:57 +0100)]
drm/vc4: Move the buffer offset out of the vc4_plane_state

The offset fields in vc4_plane_state are described as being
the offset for each buffer in the bo, however it is used to
store the complete DMA address that is then written into the
register.

The DMA address including the fb ofset  can be retrieved
using drm_fb_dma_get_gem_addr, and the offset adjustment due to
clipping is local to vc4_plane_mode_set.
Drop the offset field from the state, and compute the complete
DMA address in vc4_plane_mode_set.

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
4 months agodrm/vc4: hdmi: Avoid hang with debug registers when suspended
Dom Cobley [Tue, 5 Sep 2023 18:38:24 +0000 (19:38 +0100)]
drm/vc4: hdmi: Avoid hang with debug registers when suspended

Trying to read /sys/kernel/debug/dri/1/hdmi1_regs
when the hdmi is disconnected results in a fatal system hang.

This is due to the pm suspend code disabling the dvp clock.
That is just a gate of the 108MHz clock in DVP_HT_RPI_MISC_CONFIG,
which results in accesses hanging AXI bus.

Protect against this.

Signed-off-by: Dom Cobley <popcornmix@gmail.com>
4 months agodrm/vc4: UV planes vertical scaling must always be enabled
Dave Stevenson [Wed, 23 Aug 2023 16:48:23 +0000 (17:48 +0100)]
drm/vc4: UV planes vertical scaling must always be enabled

It has been observed that a YUV422 unity scaled plane isn't displayed.
Enabling vertical scaling on the UV planes solves this. There is
already a similar clause to always enable horizontal scaling on the
UV planes.

Signed-off-by: Dave Stevenson <dave.stevenson@raspberrypi.com>
4 months agodrm/vc4: hvs: Change prototype of __vc4_hvs_alloc to pass registers
Maxime Ripard [Fri, 24 Mar 2023 14:45:50 +0000 (15:45 +0100)]
drm/vc4: hvs: Change prototype of __vc4_hvs_alloc to pass registers

The BCM2712 HVS has registers to report the size of the various SRAM the
driver uses, and their size actually differ depending on the stepping.

The initialisation of the memory pools happen in the __vc4_hvs_alloc()
function that also allocates the main HVS structure, that will then hold
the pointer to the memory mapping of the registers.

This creates some kind of circular dependency that we can break by
passing the mapping pointer as an argument for __vc4_hvs_alloc() to use
to query to get the SRAM sizes and initialise the memory pools
accordingly.

Signed-off-by: Maxime Ripard <maxime@cerno.tech>
4 months agodrm/vc4: hvs: Rework LBM alignment
Maxime Ripard [Thu, 13 Apr 2023 08:12:19 +0000 (10:12 +0200)]
drm/vc4: hvs: Rework LBM alignment

With the introduction of the support for BCM2712, the check of whether
we're running on vc5 or not to compute the LBM alignment requirement
doesn't work anymore.

Moreover, the LBM size will need to be computed in words for the
BCM2712, while we've had sizes in bytes so far.

Aligning on either 64 or 32 words is thus fairly harmful on BCM2712, so
let's just explicitly align the size when needed, and then call
drm_mm_insert_node_generic() with an alignment of 1.

Signed-off-by: Maxime Ripard <maxime@cerno.tech>
4 months agodrm/vc4: plane: Change ptr0_offset to an array
Maxime Ripard [Fri, 24 Mar 2023 08:56:31 +0000 (09:56 +0100)]
drm/vc4: plane: Change ptr0_offset to an array

The BCM2712 will have a fairly different dlist, that will feature one
Pointer 0 word for each plane.

Let's prepare by changing the ptr0_offset variable that holds the offset
in a dlist of the pointer 0 word to an array.

Signed-off-by: Maxime Ripard <maxime@cerno.tech>
4 months agodrm/vc4: hvs: Rename hvs_regs list
Maxime Ripard [Fri, 17 Feb 2023 12:41:59 +0000 (13:41 +0100)]
drm/vc4: hvs: Rename hvs_regs list

The HVS register set has been heavily modified in the BCM2712, and we'll
thus need a separate debugfs_reg32 array for it.

The name hvs_regs is thus a bit too generic, so let's rename it to
something more specific.

Signed-off-by: Maxime Ripard <maxime@cerno.tech>
4 months agodrm/vc4: hvs: Create cob_init function
Maxime Ripard [Fri, 17 Feb 2023 14:14:55 +0000 (15:14 +0100)]
drm/vc4: hvs: Create cob_init function

Just like the HVS itself, the COB parameters will be fairly different in
the BCM2712.

Let's move the COB parameters computation and its initialisation to a
separate function that will be easier to extend in the future.

Signed-off-by: Maxime Ripard <maxime@cerno.tech>
4 months agodrm/vc4: hvs: Create hw_init function
Maxime Ripard [Fri, 17 Feb 2023 13:36:28 +0000 (14:36 +0100)]
drm/vc4: hvs: Create hw_init function

Since the BCM2712 will feature a significantly different HVS, let's move
the hardware initialisation part of our bind function into a separate
function.

That way, it will be easier to extend in the future.

Signed-off-by: Maxime Ripard <maxime@cerno.tech>
4 months agodrm/vc4: hvs: Test if the EOF interrupts are enabled
Maxime Ripard [Thu, 27 Apr 2023 11:46:53 +0000 (13:46 +0200)]
drm/vc4: hvs: Test if the EOF interrupts are enabled

We currently enable the EOF interrupts through the CRTC destroy_state
implementation.

However, nothing guarantees that we can't call destroy_state multiple
times in a row, and therefore before the EOF interrupt even happens.

This means we would enable the interrupt multiple times but disable it
only once. It wasn't an issue so far since the interrupts were only
enabled by setting a bit in a register, but with BCM2712 we will use an
external interrupt controller, with a refcounted interrupt.

Signed-off-by: Maxime Ripard <maxime@cerno.tech>
4 months agodrm/vc4: hvs: Use switch statement to simplify enabling/disabling irq
Maxime Ripard [Fri, 17 Feb 2023 14:07:29 +0000 (15:07 +0100)]
drm/vc4: hvs: Use switch statement to simplify enabling/disabling irq

Since we'll support BCM2712 soon, let's move the logic to enable and
disable the end-of-frame interrupts to a switch to extend it more
easily.

Signed-off-by: Maxime Ripard <maxime@cerno.tech>
4 months agodrm/vc4: hvs: Use switch statement to simplify vc4_hvs_get_fifo_from_output
Maxime Ripard [Fri, 17 Feb 2023 14:07:29 +0000 (15:07 +0100)]
drm/vc4: hvs: Use switch statement to simplify vc4_hvs_get_fifo_from_output

Since we'll support BCM2712 soon, let's move the logic behind
vc4_hvs_get_fifo_from_output() to a switch to extend it more easily.

Signed-off-by: Maxime Ripard <maxime@cerno.tech>
4 months agodrm/vc4: Make v3d paths unavailable on any generation newer than vc4
Maxime Ripard [Fri, 17 Feb 2023 12:29:27 +0000 (13:29 +0100)]
drm/vc4: Make v3d paths unavailable on any generation newer than vc4

The V3D IP has been separate since BCM2711, so let's make sure we issue
a WARN if we're running not only on BCM2711, but also anything newer.

Signed-off-by: Maxime Ripard <maxime@cerno.tech>
4 months agodrm/vc4: Introduce generation number enum
Maxime Ripard [Fri, 17 Feb 2023 12:07:36 +0000 (13:07 +0100)]
drm/vc4: Introduce generation number enum

With the introduction of the BCM2712 support, we will get yet another
generation of display engine to support.

The binary check of whether it's VC5 or not thus doesn't work anymore,
especially since some parts of the driver will have changed with BCM2711,
and some others with BCM2712.

Let's introduce an enum to store the generation the driver is running
on, which should provide more flexibility.

Signed-off-by: Maxime Ripard <maxime@cerno.tech>
4 months agodrm/vc4: crtc: Move assigned_channel to a variable
Maxime Ripard [Fri, 17 Feb 2023 12:39:13 +0000 (13:39 +0100)]
drm/vc4: crtc: Move assigned_channel to a variable

We access multiple times the vc4_crtc_state->assigned_channel variable
in the vc4_crtc_get_scanout_position() function, so let's store it in a
local variable.

Signed-off-by: Maxime Ripard <maxime@cerno.tech>
4 months agodrm/vc4: plane: Use return variable in atomic_check
Maxime Ripard [Fri, 17 Feb 2023 13:26:44 +0000 (14:26 +0100)]
drm/vc4: plane: Use return variable in atomic_check

The vc4_plane_atomic_check() directly returns the result of the final
function it calls.

Using the already defined ret variable to check its content on error,
and a separate return 0 on success, makes it easier to extend.

Signed-off-by: Maxime Ripard <maxime@cerno.tech>
4 months agodrm/vc4: plane: Add more debugging for LBM allocation
Maxime Ripard [Wed, 22 Mar 2023 15:17:57 +0000 (16:17 +0100)]
drm/vc4: plane: Add more debugging for LBM allocation

LBM allocations need a different size depending on the line length,
format, etc.

This can get tricky, and fail. Let's add some more prints to ease the
debugging when it does.

Signed-off-by: Maxime Ripard <maxime@cerno.tech>
4 months agodrm/vc4: hvs: Print error if we fail an allocation
Maxime Ripard [Wed, 22 Mar 2023 08:53:17 +0000 (09:53 +0100)]
drm/vc4: hvs: Print error if we fail an allocation

We need to allocate a few additional structures when checking our
atomic_state, especially related to hardware SRAM that will hold the
plane descriptors (DLIST) and the current line context (LBM) during
composition.

Since those allocation can fail, let's add some error message in that
case to help debug what goes wrong.

Signed-off-by: Maxime Ripard <maxime@cerno.tech>
4 months agodrm/vc4: hvs: More logging for dlist generation
Maxime Ripard [Wed, 22 Mar 2023 08:51:51 +0000 (09:51 +0100)]
drm/vc4: hvs: More logging for dlist generation

DLIST generation can get pretty tricky and there's not a lot of debug in
the driver to help. Let's add a few more to track the generated DLIST
size.

Signed-off-by: Maxime Ripard <maxime@cerno.tech>
4 months agodrm/vc4: hdmi: Warn if writing to an unknown HDMI register
Maxime Ripard [Thu, 23 Feb 2023 18:44:32 +0000 (19:44 +0100)]
drm/vc4: hdmi: Warn if writing to an unknown HDMI register

The VC4 HDMI driver has a bunch of accessors to read from a register.
The read accessor was warning when accessing an unknown register, but
the write one was just returning silently.

Let's make sure we warn also when writing to an unknown register.

Signed-off-by: Maxime Ripard <maxime@cerno.tech>