This series adds HDMI support for MT8188 based boards, such as the Genio
510 and Genio 700 EVKs, in order to display a boot logo over HDMI. It
builds on top of the HDMI PHY driver [1], the DDC (I2C) driver [2], the
clock series [3] and a clock fix [4] already sent earlier.
The drivers here were inspired by the corresponding kernel drivers, but
were largely modified, mainly because U-Boot has no DRM framework and
the goal is only to display a framebuffer. The HDMI driver locates each
pipeline component through the device-tree aliases the SoC dtsi already
provides and programs them directly. The framebuffer lives in a
dedicated reserved-memory region.
The video pipeline looks like this:
reserved-memory (framebuffer, in DRAM)
|
+------------+------------+
v v
mdp_rdma4 mdp_rdma5 two DMA read engines,
(left half) (right half) each fetching one frame half
| |
padding4 padding5 (bypass)
| |
+------------+------------+
v
vpp merge (merge3) recombine the two halves
|
v
ethdr mixer (ethdr0)
|
v
hdr back-end (bypass)
|
v
vpp merge (merge5)
|
v
dpi1 timing generator -> HDMI TX
|
v
ddc/i2c ---> hdmi tx controller
(read EDID) |
v
hdmi phy
|
v
hdmi connector ---> sink
mutex1 : hardware SOF synchronization across the pipeline
vdosys1 : routing muxes (SEL_IN/SOUT) and the sizes of the
clock-domain-crossing wrappers between the blocks above
larb2/3 : SMI local arbiters gating the RDMAs' access to DRAM
The series adds the HDMI TX controller (UCLASS_VIDEO), a set of shared
display-component helpers and the individual pipeline components (ethdr,
merge, mutex, padding, rdma, dpi1), a minimal SMI local arbiter driver,
and the device-tree glue and config that enable the pipeline.
Tested on a Genio 700 EVK: the U-Boot logo shows up on an HDMI display.
[1]:
https://patchwork.ozlabs.org/project/uboot/patch/20260717-mt8188-add-hdmi-phy-driver-v1-1-4278457fe...@baylibre.com/
[2]:
https://patchwork.ozlabs.org/project/uboot/cover/20260717-mt8188-add-ddc-i2c-driver-v1-0-e89ccd173...@baylibre.com/
[3]:
https://patchwork.ozlabs.org/project/uboot/patch/20260715-mt8188-add-clk-add-display-related-clocks-v2-1-0aaf7acca...@baylibre.com/
[4]:
https://patchwork.ozlabs.org/project/uboot/patch/20260728-mt8188-clk-fix-top0-gate-polarity-v1-1-56bc92ba6...@baylibre.com/
Signed-off-by: Julien Stephan <[email protected]>
---
Changes in v5:
Across the whole series
- Dropped the dead colorspace/colorimetry handling: only RGB444 8bpc is ever
produced, so set_csp_depth/csp, the YUV
cases of the AVI InfoFrame, the YUV420 downsample path and the DPI CSC are
gone. That removed the only user of the
include/linux/hdmi.h added in v2, so that patch is dropped too.
- Programmed the resolution- and DRAM-latency-dependent registers the mainline
kernel writes and this series skipped;
their reset values only work at the timing they happened to be tested at.
Retested on a Genio 700 EVK.
- of_match tables consistently use "{ }".
- Include hygiene: <linux/bitops.h>, <errno.h> and <fdtdec.h> where needed,
unused <video.h>/<asm/io.h> dropped, #endif
guards named, redundant "struct udevice;" forward declarations removed.
Per patch
- memory: new preparatory patch sorting drivers/memory/Makefile, so the larb
entry lands in an ordered list.
- SMI larb: moved to drivers/memory/mtk-smi-larb.c with its own
CONFIG_MTK_SMI_LARB and UCLASS_MEMORY
- component helpers: added mtk_disp_comp_update(), the read-modify-write
counterpart of mtk_disp_comp_write().
- ethdr: put the HDR video back-end in bypass like Linux' mtk_ethdr_config()
does; "vdo_be" mapped by reg-name.
- merge: mtk_disp_merge_config() reshaped to the kernel's (l_w, r_w, h) and
CFG_1/25/27 written unconditionally;
documented that the single-input no-fifo mode is unreachable from this
pipeline, hence untested.
- padding: clear all four registers in bypass mode, not only the picture size
- rdma:
- program MDP_RDMA_GMCIF_CON and clear MDP_RDMA_COMP_CON and FLD_OUTPUT_ARGB
- reject a stride too wide for FLD_MF_BKGD_WB;
-print the framebuffer base with "%#lx" (it is a ulong);
- resolve the larb in UCLASS_MEMORY.
- rdma: kept the datasheet name BGRA8888 over the kernel's
RDMA_INPUT_FORMAT_RGBA8888 (same byte order, i.e.
DRM_FORMAT_XRGB8888) and TRANSFORM_0 bit 15 over MDP_RDMA_CON's FLD_OUTPUT_10B
(that bit is in neither kernel driver, it
came with the downstream U-Boot code). Both documented in comments.
- dpi: the CSC/matrix path went with the colorspace handling, and
MATRIX_SEL_RGB_TO_BT709 with it; mtk_dpi_hw_disable()
is now used, in the HDMI probe unwind; mtk_dpi_mask() replaced by
mtk_disp_comp_update(); local types renamed
mtk_dpintf_* -> mtk_dpi_* (this node is mediatek,mt8195-dpi, not the dp-intf
variant) and the config parameters const.
- get_unaligned_le24(): moved between the le16 and le32 helpers to keep the
file ordered, and split into
__get_unaligned_le24()/get_unaligned_le24() as Linux has it.
- hdmi driver: above 340 MHz, tell the sink to scramble and use the 1/40 bit
clock ratio through its SCDC TMDS_Config
(offset 0x20 at I2C 0x54).
- hdmi driver: sink type taken from the EDID (hdmi_monitor) instead of always
assuming HDMI, and no InfoFrames sent to a
DVI sink.
- hdmi driver: zero the EDID extension block when reading it fails, so the
parser cannot pick a timing out of
uninitialised stack
- hdmi driver: propagate the generic_phy_configure()/_power_on() failures
instead of logging and carrying on.
- hdmi driver: clear the GCP AVMUTE, in case an earlier boot stage left the
sink muted.
- hdmi driver: the VIC is now looked up in a table keyed on the timing's exact
rounded refresh rate; an unknown timing
reports VIC 0 and no aspect ratio rather than a wrong one, and the AVI
colorimetry is "No Data" as CTA-861 wants for
RGB. Also set A0 in the AVI InfoFrame, the active format R3..R0 is only valid
with it.
- hdmi driver: mtk_hdmi_mode_valid() also rejects interlaced timings, odd
widths and pixel clocks outside 25-594 MHz.
The pipeline reads the frame as two halves and feeds the DPI two pixels per
clock, and the PHY will not lock outside
that range; rejecting them here lets the EDID parser fall back to the next
detailed timing instead of failing.
- hdmi driver: no data islands towards a DVI sink either, the GCP and null
packets are gated on hdmi_monitor like the
InfoFrames.
- hdmi driver: dropped the infoframe buffer size checks, which cannot fail, and
collapsed the three-way HPD state into
a bool.
- hdmi driver: program the vdosys1 clock-domain-crossing wrapper sizes and the
mixer input padding. The alpha half of
the kernel's mtk_mmsys_mixer_in_config() is left alone: the mixer layer runs
with MIXER_ALPHA_AEN clear.
- hdmi driver: zero the framebuffer before the RDMAs start and enable the DPI
last, so a stale reserved region is never
scanned out; video_post_probe() only clears the framebuffer after probe returns.
- hdmi driver: dev_remap_addr() and a void __iomem * for the register window;
extension_flag instead of a raw 0x7e
offset; the 150 us settle delay moved back where the kernel has it, so it does
not fire on the disable path; "display is
not connected" demoted to debug; the cached struct display_timing dropped,
timing and buffer parameters const, "inline"
dropped from the two static helpers, includes sorted.
- hdmi driver: retitled "video: mediatek: add hdmi driver".
- mtk_hdmi_regs_v2.h is a verbatim copy of the kernel header except
VID_DOWNSAMPLE_CONFIG, at 0x8f0 per the MT8390/MT8188 datasheet and not the
0x8d0 Linux uses (a patch was sent to fix this on kernel side)
- dts: the reserved-memory and the hdmi "memory-region" phandle are a single
patch now; SPDX (GPL-2.0 OR MIT) ->
(GPL-2.0-only OR MIT), node renamed framebuffer@72000000, 32 MiB size
documented.
- configs: drop CONFIG_CMD_I2C, the DDC transfers go through DM_I2C inside the
driver; the commit message notes that
CONFIG_VIDEO_LOGO is "default y if !SPLASH_SCREEN", so CONFIG_VIDEO=y already
brings it in.
- Link to v4:
https://patch.msgid.link/20260902-add-genio-510-and-genio-700-hdmi-v4-0-7c3b4f068...@baylibre.com
Changes in v4:
- refresh config fragment with savedefconfig: remove CONFIG_VIDEO_LOGO,
correctly place CONFIG_WATCHDOG_AUTOSTART
that is enabled by default
- reword commit message on patch 13: remove useless mentions to larb/vdosys1
aliases that were removed in v3
- Link to v3:
https://patch.msgid.link/20260901-add-genio-510-and-genio-700-hdmi-v3-0-38194a531...@baylibre.com
Changes in v3:
Across the whole series
- Re-based every register definition on the mainline Linux kernel definitions
(upstream names, values and
BIT()/GENMASK()/FIELD_PREP()) instead of the downstream U-Boot ones.
- SPDX fixes: GPL-2.0 / GPL-2.0+ -> GPL-2.0-only on all new driver files,
Makefile and Kconfig; relicensed
include/linux/hdmi.h to MIT (it's a partial port of Linux' MIT-licensed header)
and added the full MIT permission text.
- Bind on the generic mediatek,mt8195-* fallback compatible (shared IP) rather
than the SoC-specific mediatek,mt8188-*:
dpi, ethdr, merge, rdma.
- Makefile/Kconfig header cleanup: copyright set to BayLibre and the obj- list
sorted alphabetically.
Per patch
- display pipeline component helpers: reworded the commit message (probe and
enable sequence).
- ethdr: use the kernel's parameterized MIX_L_SRC_CON(n)/MIX_L_SRC_SIZE(n)
macros, drop the unused *_CK_EN defines, add
BACKGROUND_RELAY to the datapath write; keep the downstream-only L2 select bits
behind a comment.
- merge: rename registers to the kernel's DISP_REG_MERGE_* / MERGE_EN; keep the
downstream-only CFG_30 behind a comment.
- mutex: make the register offsets index-based (DISP_MUTEX_*(id)) and rename
the helper to mtk_disp_mutex_config_hdmi()
so a future DSI user can reuse it.
- padding: rename registers to the kernel's PADDING_CONTROL_REG /
PADDING_ENABLE / PADDING_PIC_SIZE_REG.
- rdma: use the kernel's FLD_* fields with FIELD_PREP(); take the stride from
the framebuffer line_length instead of
recomputing it; error out if the framebuffer base is above 4 GB (the
source-base register is 32-bit), add larb configuration.
- dpi: import mtk_dpi_regs.h verbatim from the mainline kernel and adjust the
driver to the upstream names (DPI_RET,
DPI_INPUT_2P_EN, DPI_OUTPUT_1T1P_EN); keep the one downstream-only matrix value
(MATRIX_SEL_RGB_TO_BT709) locally.
- include/linux/hdmi.h: MIT relicense (see above).
- hdmi driver: move the register defines into a new mtk_hdmi_regs_v2.h imported
verbatim from the mainline kernel;
correct VID_DOWNSAMPLE_CONFIG to 0x8f0 per the MT8390 datasheet (mainline has a
wrong 0x8d0 — a separate kernel fix was
sent); remove the dead deep-color code path (the driver only ever drives 8
bpc); drop the unused length argument of the
infoframe helpers; check the RDMA config return value; resolve the vdosys1
syscon via ofnode_by_compatible() instead of
a U-Boot-invented alias.
- dts (reserved-memory): drop the unused splashscreen carveout, keep only the
framebuffer region; retitle the patch to
"add framebuffer reserved-memory".
- dts (hdmi override): drop the U-Boot-invented vdosys1 alias (resolved by
compatible now) and larb2/larb3 aliases
now resolved using iommus and DT directly inside rdma component.
- configs: use CONFIG_VIDEO_LOGO (show the U-Boot logo) instead of
CONFIG_SPLASH_SCREEN, since upstream lacks the
plumbing to make the splash actually work — this also keeps CONFIG_EXPO enabled
and drops the now-unneeded CMD_BMP /
BMP_24BPP.
- Link to v2:
https://patch.msgid.link/20260826-add-genio-510-and-genio-700-hdmi-v2-0-68ebb1bf1...@baylibre.com
Changes in v2:
- Add a common `get_unaligned_le24()` helper for reading 24-bit little-endian
values, ported from kernel
- Add a common `linux/hdmi.h` header with HDMI colorspace and colorimetry
enums, based on the corresponding Linux header.
- Refactor the HDMI driver to use the new common HDMI definitions and unaligned
access helper.
- Rename the mixer component to ETHDR and configure the ETHDR before the
remaining display pipeline components.
- Add the SMI LARB components and configure the VDOSYS1 routing directly from
the HDMI driver.
- Switch display component lookup from phandles to DT aliases and add proper
error unwinding for component clocks, HDMI PHY power, and clock resources.
- Refactor HDMI register access and InfoFrame handling.
- Improve HDMI mode handling, YUV420 downsampling, TMDS scrambling, deep-color
handling, and InfoFrame generation.
- Move framebuffer/splash reserved-memory definitions to the Genio common
U-Boot DTSI and add the corresponding HDMI override node for MT8390/Genio EVKs.
- Drop the dedicated MT8188 VDOSYS1 pipeline component and its corresponding
DTS node, to avoid having 2 nodes with same reg address.
- Remove the unnecessary `dm_scan_fdt_dev` bind hook from the HDMI driver.
- Link to v1:
https://patch.msgid.link/20260728-add-genio-510-and-genio-700-hdmi-v1-0-ac54bd75f...@baylibre.com
To: [email protected]
Cc: GSS_MTK_Uboot_upstream <[email protected]>
Cc: Tom Rini <[email protected]>
Cc: Anatolij Gustschin <[email protected]>
Cc: Ryder Lee <[email protected]>
Cc: Weijie Gao <[email protected]>
Cc: Chunfeng Yun <[email protected]>
Cc: Igor Belwon <[email protected]>
Cc: David Lechner <[email protected]>
Cc: Julien Stephan <[email protected]>
Cc: Dario Binacchi <[email protected]>
Cc: Johan Jonker <[email protected]>
Cc: Patrice Chotard <[email protected]>
Cc: Simon Glass <[email protected]>
Cc: Pavlo Yadvychuk <[email protected]>
Cc: Macpaul Lin <[email protected]>
Cc: Ilias Apalodimas <[email protected]>
Cc: Mattijs Korpershoek <[email protected]>
Cc: Carlo Caione <[email protected]>
Cc: Anshul Dalal <[email protected]>
Cc: Rasmus Villemoes <[email protected]>
---
Julien Stephan (13):
memory: sort the Makefile alphabetically
memory: mediatek: add SMI larb driver
video: mediatek: add display pipeline component helpers
video: mediatek: add ethdr component
video: mediatek: add merge component
video: mediatek: add mutex component
video: mediatek: add padding component
video: mediatek: add rdma component
video: mediatek: add dpi1 component
asm-generic: unaligned: add get_unaligned_le24()
video: mediatek: add hdmi driver
arm: dts: mt8390: add framebuffer reserved-memory
configs: mediatek: mt8188: enable hdmi pipeline
arch/arm/dts/mt8370-genio-510-evk-u-boot.dtsi | 1 +
arch/arm/dts/mt8390-genio-700-evk-u-boot.dtsi | 1 +
arch/arm/dts/mt8390-genio-common-u-boot.dtsi | 26 +
configs/mt8188.config | 4 +
drivers/memory/Kconfig | 10 +
drivers/memory/Makefile | 3 +-
drivers/memory/mtk-smi-larb.c | 71 ++
drivers/video/Kconfig | 1 +
drivers/video/Makefile | 1 +
drivers/video/mediatek/Kconfig | 22 +
drivers/video/mediatek/Makefile | 15 +
drivers/video/mediatek/mtk_disp_comp.c | 90 +++
drivers/video/mediatek/mtk_disp_comp.h | 34 +
drivers/video/mediatek/mtk_disp_merge.c | 162 +++++
drivers/video/mediatek/mtk_disp_merge.h | 25 +
drivers/video/mediatek/mtk_disp_mutex.c | 68 ++
drivers/video/mediatek/mtk_disp_mutex.h | 14 +
drivers/video/mediatek/mtk_disp_padding.c | 49 ++
drivers/video/mediatek/mtk_disp_padding.h | 16 +
drivers/video/mediatek/mtk_dpi.c | 359 ++++++++++
drivers/video/mediatek/mtk_dpi.h | 19 +
drivers/video/mediatek/mtk_dpi_regs.h | 247 +++++++
drivers/video/mediatek/mtk_ethdr.c | 104 +++
drivers/video/mediatek/mtk_ethdr.h | 27 +
drivers/video/mediatek/mtk_hdmi.c | 997 ++++++++++++++++++++++++++
drivers/video/mediatek/mtk_hdmi_regs_v2.h | 267 +++++++
drivers/video/mediatek/mtk_mdp_rdma.c | 213 ++++++
drivers/video/mediatek/mtk_mdp_rdma.h | 20 +
include/asm-generic/unaligned.h | 10 +
include/mtk_smi.h | 17 +
30 files changed, 2892 insertions(+), 1 deletion(-)
---
base-commit: 677a768c91b2baaefdf09633d1fdb5c9f758674f
change-id: 20260701-add-genio-510-and-genio-700-hdmi-ca62c6b8210c
Best regards,
--
Julien Stephan <[email protected]>
