[PATCH v2 0/2] usb: dwc3: support two connectors sharing one HS data bus

[Qianhui Xu]

v1: <20261013102927.1141696-1-xiaqinghuan@xxxxxxxxxxxxxxxx>

Changes in v2, following the review of v1:

- Drop "dt-bindings: usb: snps,dwc3: document usb-psy-name" entirely.
  Krzysztof is right: the property is not a DT convention -- a supply is
  referenced by a phandle, not by a name.  The driver does read the string
  (drivers/usb/dwc3/core.c: dwc3_vbus_draw_work() ->
  device_property_read_string(dwc->dev, "usb-psy-name", ...)), but that is a
  legacy/ACPI-shaped lookup that the board can avoid, so documenting it as a
  generic DWC3 binding property was the wrong call.  Nothing in this series
  needs it.  (For reference: on the affected board the same current limit is
  reachable through the standard "power-supplies" phandle on the connector,
  and I will follow up separately if a binding is really warranted.)

- binding: no longer touches usb-psy-name at all.  In v1 the patch that
  documented the property and the patch that described the mux raced in
  ordering, so the second patch read as deleting what the first one added.
  Krzysztof's "No, you just added it!" was a correct reading of that diff.

- driver: a failed GPIO write no longer aborts __dwc3_set_mode() half-way.
  Reported by sashiko-ai: the mode change tears down the previous host or
  gadget instance before programming the new role, so bailing out between the
  two left a torn-down role behind with a stale current_dr_role, and the next
  request would tear it down a second time (double free / use-after-free).

- driver: no more flush_work() on the freezable workqueue.  Reported by
  sashiko-ai: flushing system_freezable_wq from the role-switch path can
  deadlock against system suspend.  The mux now keeps exactly the same
  asynchronous semantics as the existing single-connector role switch --
  dwc3_set_mode() queues the work, __dwc3_set_mode() is the only place that
  programs hardware.

- driver: dropped the "mux_role" variable that was set but never used,
  reported by kernel test robot (arm-defconfig, clang, W=1).

- driver: "oldest request wins" is now actually implemented, with a
timestamp
  per connector taken at the USB_ROLE_NONE -> role transition.  v1 selected
  port 0 first, i.e. static priority, which contradicted its own commit
  message.  Reported by sashiko-ai.

- driver: the runtime-PM reference is no longer taken on a path that can
jump
  to the error label before the reference is acquired (sashiko-ai).

Konrad: to answer your question about the wiring -- this is a real board
topology, not a hub.  Lenovo Legion Y700 (4th gen, TB322FC, SM8750P) has
two
USB-C receptacles driven by a single dwc3 instance (a600000.usb).  The
Super
Speed pairs are not shared: each receptacle has its own SS PHY pair.  Only
the
High Speed D+/D- lines are routed through one 2:1 mux, selected by TLMM
GPIO
29, and each receptacle keeps its own Type-C PD/role input (both connectors
report data-role = "dual" through pmic-glink).  So the controller can be a
host on one receptacle while the other is charging-only, and plugging a
device
into the other receptacle has to be able to steal the controller.  Without
the
mux support, the second receptacle's role switch is simply not registered
and
nothing happens.

Relevant excerpt from the board DTS ("ports" endpoints carry the mux-state
level of their own connector):

  &usb {
          dr_mode = "otg";
          usb-role-switch;
          port-select-gpios = <&tlmm 29 GPIO_ACTIVE_HIGH>;
          ports {
                  port@0 {                /* long-edge receptacle */
                          reg = <0>;
                          usb_dwc3_hs: endpoint {
                                  remote-endpoint = <&pmic_glink_hs_in>;
                                  usb-role-switch;
                                  mux-state = <0>;
                          };
                  };
                  port@2 {                /* short-edge receptacle */
                          reg = <2>;
                          usb_dwc3_hs2: endpoint {
                                  remote-endpoint = <&pmic_glink_hs_c2_in>;
                                  usb-role-switch;
                                  mux-state = <1>;
                          };
                  };
          };
  };

Testing (all on the board, 7.3-rc7 based tree):
 - two receptacles: host mode on either receptacle works, and moving the
   cable from one to the other hands the controller over;
 - charging-only on one receptacle while the other has a device attached;
 - single-connector boards: the mux path is only taken when
   "port-select-gpios" is present, otherwise the existing role switch is used
   unchanged.

Qianhui Xu (2):
  dt-bindings: usb: snps,dwc3: describe the two-connector HS data path
  usb: dwc3: support two connectors sharing one HS data bus

 .../bindings/usb/snps,dwc3-common.yaml        |  63 +++++-
 drivers/usb/dwc3/core.c                       |  29 ++-
 drivers/usb/dwc3/core.h                       |  38 ++++
 drivers/usb/dwc3/debugfs.c                    |   4 +
 drivers/usb/dwc3/drd.c                        | 199 +++++++++++++++++-
 5 files changed, 330 insertions(+), 3 deletions(-)

-- 
2.53.0