Add support for Multi-stream transport for MSM chipsets that allow a single instance of DP controller to send multiple streams.
This series has been validated on sa8775p ride platform using multiple MST dongles and also daisy chain method on both DP0 and DP1 upto 1080P. With 4x4K monitors, due to lack of layer mixers that combination will not work but this can be supported as well after some rework on the DPU side. In addition, SST was re-validated with all these changes to ensure there were no regressions. This series depends on Dmitry's "drm: handle IRQ_HPD events correctly" series [1], which adds the IRQ_HPD propagation infrastructure required for Type-C MST support. Overall, the patch series has been organized in the following way: 1) MST controller support (patches 01-07): low-level additions for multi-stream operation — ACT/VCPF packet transmission, payload slot allocation, and mainlink MST mode. 2) Driver refactors (patches 08-11): replace the power_on flag with a per-stream bitmask, factor out panel-accepting bridge helpers, and introduce link_ready for explicit link-level state tracking. 3) MST DRM infrastructure (patches 12-18): new dp_mst_drm module for topology management, DPU per-stream encoder support, and atomic stream enable/disable handling. 4) SST/MST connector state and HPD fixes (patches 19-29): SST/MST connector state management, HPD event routing through the MST manager, per-stream bpp selection, and Type-C retimer guard. Note: - Validation for this series has been done on the latest linux-next on LeMans (sa8775p) and Glymur, covering both FB console and Weston. Type-C MST has been validated on Glymur with MST dongles. - AI review comments on patch 13 and subsequent patches will be reviewed and addressed in the next revision of this series. [1] https://lore.kernel.org/all/[email protected]/ --- Changes in v9: - Patch 02: Turn off the pixel clock if sending ACT or waiting for video ready fails.[Dmitry] - Patch 07: Drop the unused DP_MAX_TIME_SLOTS define.[Dmitry] - Patch 10: Fix the debug message printed when the link is already disabled.[Dmitry] - Link to v8: https://lore.kernel.org/r/[email protected] Changes in v8: - Rebased onto linux-next-20261102; dropped patches 01-07 of v7. - Dropped patch 32/35 of v7 since it not needed. - Renamed the controller's mst_active to mst_mode so that it no longer shadows struct msm_dp::mst_active). - Dropped the driver-side slot bookkeeping; the MST helpers now pass a msm_dp_mst_slot_info table built from the topology state (patches 07, 09, 18). - Patch 01 (new): Skip sink requests while the mainlink clocks are off. - Patch 02: Use DIV_ROUND_UP_ULL() for the 64-bit division in the ACT wait time. - Patch 06: Mask the STATUS5/STATUS7 reads down to their own interrupt bits. - Patch 07: Take the slot table from the caller; simplify the rate governor's fixed-point rounding. Rewrite commit message. - Patch 08: Replace active_stream_cnt with active_stream_mask so that a single stream can be queried. - Patch 09: Thread the slot table through the helpers; key the disable guard on the stream's own bit. - Patch 10: Release the runtime PM reference on every prepare_link() failure path, and check sink_count before taking it. - Patch 12: Skip the MST manager cleanup when dpu_kms->dev is NULL after an early probe failure. - Patch 13: Use the DP controller's own device for runtime PM; return early when MST is not active. - Patch 18: Report enable failures and stop the stream enable sequence; collect the payload slots of all streams. - Patch 19: Read the old connector status in the same critical section as detect(). - Patch 20: Drop the explanatory comment. - Patch 21: Drop the plug guard in msm_dp_hpd_plug_handle(); hpd v8 has a stronger one. - Patch 23: Check for MST before taking plugged_lock. - Patch 27: Drop the hunk that handled plug before IRQ_HPD, now done by hpd v8; retitle accordingly. - Patch 28 (new): Select the MST stream bpp during the atomic check instead of failing the modeset. - Link to v7: https://lore.kernel.org/r/[email protected] Changes in v7: - Rebased onto linux-next-20260918. - Incorporate Type-C IRQ_HPD support: new patches 33-35 handle IRQ_HPD events from Type-C controllers. - Patch 02: Rewrite commit message. - Patch 05: Handle optional pixel resources without spurious errors and track each register block's size separately. - Patch 07: Use optional-resource mapping for MST2/MST3 registers. - Patch 12: Handle stream 2/3 VCPF completion through STATUS7 and reuse the existing idle helper for SST. - Patch 15: Preserve the disable guard and adapt it to active_stream_cnt. - Patch 18: Add explicit MST manager cleanup during KMS teardown. - Patch 23 (new): Move link capabilities from msm_dp_panel to msm_dp_link. - Patch 24: Propagate enable errors, update remaining payload slots on removal, skip disable hardware operations after disconnect, and use shared link capabilities. - Patch 25 (new): drm/bridge_connector: suppress hotplug for IRQ_HPD without status changes. - Patch 26 (new): drm/bridge_connector: avoid detect-based HPD notifications for DisplayPort. - Patch 27: Route IRQ_HPD through bridge events, drain pending ESI events, and synchronize the IRQ before manager destruction. - Patch 30: Use shared link capabilities and explicitly check for MST. - Patch 31: Preserve SST operation for SST-sideband-only sinks. - Patch 33 (new): Fix threaded HPD handler not woken on controller IRQ; Fixes: 5043357e75ba ("drm/msm/dp: rework HPD handling"); Cc: stable. - Patch 34 (new): Order IRQ HPD handling: process plug before IRQ_HPD; discard IRQ_HPD on unplug. - Patch 35 (new): soc: qcom: pmic-glink-altmode: skip retimer reset on DP IRQ when state is unchanged. - Link to v6: https://lore.kernel.org/r/[email protected] Changes in v6: - Rebased onto linux-next 20260828; drop already-merged prerequisite cleanup patches. - Reordered patches for better logical flow. - Patch 01: Also move power_on=false into display_unprepare. - Patch 02 (new): Split from old patch 04. - Patch 03 (new): Split from old patch 04. - Patch 04: Squash num_pixel_clks into this patch. - Patch 05: Use loop for snapshot; skip inactive streams before adding. - Patch 06 (new): Split register definitions out of old patch 04. - Patch 07: Rewrite old patch 04; replace stream_id special-casing with transparent msm_dp_read/write_link() helpers. - Patch 08: Fix REG_DP_MST_ACT register placement. - Patch 13: Drop defensive start_slot bounds check; rename channel_alloc -> update_timeslot. - Patch 14: Guard link training with active_stream_cnt == 0. - Patch 16: Add locking annotation; group link_ready with status variables. - Patch 17: Drop redundant stream_id >= DP_STREAM_MAX check. - Patch 18: Consolidate msm_dp_mst_attach_encoder() and msm_dp_get_mst_max_stream() definition, declaration, and usage into this patch. - Patch 19 (new): Split connector/topology callbacks out of old patch 23+24. - Patch 20 (new): Split dpu_encoder_get_intf() refactor out of old patch 21. - Patch 21: Squash stream_id condition; reorder before MST encoder patch. - Patch 22: Use thin wrappers instead of renaming/exporting core dpu_encoder functions; use suggested for-loop style. - Patch 23 (new): Split atomic stream enable/disable/check out of old patch 23. - Patch 24: Add hpd_irq_lock; use %*ph for ESI debug prints. - Patch 26: Defer SST-disconnect-on-MST-cap to new patches 28-29 to avoid bisect breakage. - Patch 27: Add comment for clear-before-write DPCD sequence. - Patch 28 (new): Split from old patch 14; handle boot-with-cable SST bridge detection. - Patch 29 (new): Handle SST connector status when MST is enabled. - Link to v5: https://lore.kernel.org/r/[email protected] Changes in v5(fix comments from V4): - Dropped early refactoring patches (v4 01-12); restructured as "move link-level teardown", "factor out _helper variants", etc. - Dropped v4 39/39 (platform MST stream list); use DT pixel-clock count instead per Dmitry's request. - Patch 02: Remove artificial max-stream limitation check. unify register naming (REG_DP1 vs MMSS_DP1). - Patch 05: Calculate ACT wait time dynamically from mode parameters instead of hardcoded 20 ms. - Patch 08: Replace start-slot loop with direct math; fix commit message. - Patch 09: Add locking annotation "Must be called with msm_dp_mst::mst_lock held" to msm_dp_ctrl_push_vcpf(). - Patch 17: Rename prepared flag to link_ready. - Patch 22: Move drm_dp_mst_atomic_check() call into msm_atomic.c. - Patch 23: Replace bridge-based MST DRM model with encoder-based approach; remove redundant bridge layer between encoder and connector. - Patch 24: Add drm_edid_free(); add FIXME for bpp negotiation. - Patch 25: Use dp->plugged flag instead of link status for MST plug path routing. - Link to v4: https://lore.kernel.org/r/[email protected] Changes in v4: - Fixed most comments from V3. - Rebase onto HPD refactor V5. - Fixed casing/formatting issues, for example: "mst". - Drop .mode_set() and set_mode in .atomic_enable(). - Rewrite commit messages that are unclear. - Use the same API for MST link and SST link writes. - Use the new drm_dp_dpcd_read_byte() / drm_dp_dpcd_write_byte() interfaces. - Remove some unnecessary payload fields from the MST bridge. - Remove some defensive NULL pointer checks. - Reworked the patch order to make the series easier to follow. - Add support for more platforms. - Link to v3: https://lore.kernel.org/r/[email protected] Changes in v3: Fixed review comments from Dmitry - Fixed lots of comments from series V1/V2. - Rebased onto next-20250808. - Rebased onto Jessica's HPD-refactor branch. - Fixed formatting issues in commit messages under changes. - Removed unnecessary one-line wrappers. - Relocated MST-related .atomic_check() calls to their appropriate positions. - Removed the logic related to slot checking in .mode_valid(). - Link to v2: https://lore.kernel.org/r/[email protected] Changes in v2: Fixed review comments from Dmitry - Rebase on top of next-20250606 - Add all 4 streams pixel clks support and MST2/MST3 Link clk support - Address the formatting issues mentioned in the review comments - Drop the cache of msm_dp_panel->drm_edid cached - Remove the one-line wrapper funtion and redundant conditional check - Fixed the commit messgae descriptions of some patches - Reordered the patches and renamed some functions and variables - Link to v1: https://lore.kernel.org/all/20241205-dp_mst-v1-0-f [email protected]/ Signed-off-by: Yongxing Mou <[email protected]> --- Abhinav Kumar (13): drm/msm/dp: add support to send ACT packets for MST drm/msm/dp: add support to enable MST in mainlink control drm/msm/dp: no need to update tu calculation for mst drm/msm/dp: always program MST_FIFO_CONSTANT_FILL for MST use cases drm/msm/dp: add support for sending VCPF packets in DP controller drm/msm/dp: add support for MST channel slot allocation drm/msm/dp: replace power_on with active_stream_mask drm/msm/dp: factor out _helper variants of bridge ops accepting a panel drm/msm/dp: add msm_dp_display_get_panel() to initialize DP panel drm/msm/dp: add HPD callback for dp MST drm/msm/dp: mark the SST bridge disconnected when mst is active drm/msm/dp: enable MST on capable sinks drm/msm/dp: mark the SST bridge disconnected when an MST-capable sink is present Yongxing Mou (16): drm/msm/dp: skip sink requests while the mainlink clocks are off drm/msm/dp: add link_ready to manage link-level operations drm/msm/dp: introduce dp_mst_drm module drm/msm/dp: add MST connector creation and topology callbacks drm/msm/dpu: pass msm_display_info to dpu_encoder_get_intf() drm/msm/dpu: use stream_id to select MST interfaces drm/msm/dpu: add per-stream MST encoders drm/msm/dp: move link capabilities to dp_link drm/msm/dp: add atomic stream handling for MST drm/bridge_connector: suppress hotplug for IRQ_HPD without status changes drm/bridge_connector: avoid detect-based HPD notifications for DisplayPort drm/msm/dp: wire MST helpers into atomic check and commit paths drm/msm/dp: wake threaded handler for HPD IRQs drm/msm/dp: ignore IRQ HPD latched together with an unplug drm/msm/dp: select MST stream bpp during atomic check soc: qcom: pmic-glink-altmode: skip retimer reset on DP IRQ drivers/gpu/drm/display/drm_bridge_connector.c | 23 +- drivers/gpu/drm/msm/Makefile | 3 +- drivers/gpu/drm/msm/disp/dpu1/dpu_encoder.c | 51 +- drivers/gpu/drm/msm/disp/dpu1/dpu_encoder.h | 2 + drivers/gpu/drm/msm/disp/dpu1/dpu_kms.c | 37 +- drivers/gpu/drm/msm/dp/dp_audio.c | 2 +- drivers/gpu/drm/msm/dp/dp_ctrl.c | 325 +++++++++++- drivers/gpu/drm/msm/dp/dp_ctrl.h | 9 +- drivers/gpu/drm/msm/dp/dp_debug.c | 6 +- drivers/gpu/drm/msm/dp/dp_display.c | 388 ++++++++++---- drivers/gpu/drm/msm/dp/dp_display.h | 27 +- drivers/gpu/drm/msm/dp/dp_link.h | 2 + drivers/gpu/drm/msm/dp/dp_mst_drm.c | 679 +++++++++++++++++++++++++ drivers/gpu/drm/msm/dp/dp_mst_drm.h | 18 + drivers/gpu/drm/msm/dp/dp_panel.c | 34 +- drivers/gpu/drm/msm/dp/dp_panel.h | 10 +- drivers/gpu/drm/msm/dp/dp_reg.h | 17 + drivers/gpu/drm/msm/msm_atomic.c | 14 +- drivers/gpu/drm/msm/msm_drv.h | 47 ++ drivers/gpu/drm/msm/msm_kms.c | 1 + drivers/soc/qcom/pmic_glink_altmode.c | 38 +- 21 files changed, 1603 insertions(+), 130 deletions(-) --- base-commit: 2637bdec0535a2044cd830eba9a79363651d9043 change-id: 20260410-msm-dp-mst-35130b6e8b84 prerequisite-message-id: <[email protected]> prerequisite-patch-id: 38205d2013335920df343c240fa0afc17b987763 prerequisite-patch-id: 358e5f35d442041b88e52d8e39a564a053a0f990 prerequisite-patch-id: 9860b9acb12b1c2652e4e9f4860968a2ca3361f0 prerequisite-patch-id: 53486e695198f785401740f7f4372bdf335fb620 prerequisite-patch-id: c8e551a07553f7dc107063a1dbec1d415184f97b prerequisite-patch-id: 1a6251919897ca5e8b12a63774a833fd8460c67a prerequisite-patch-id: 846ab694795c6bec123d1dc46d7a53097bf594bf prerequisite-patch-id: 2a058aa8309016f424b1d4ca19a9b20e92f4d8b6 prerequisite-patch-id: 76ea1ecf4e1d83ce2e5730aa36ed1a4524f15e56 Best regards, -- Yongxing Mou <[email protected]>
