[PATCH] drm/nouveau/gsp: skip address-only I2C starts with MOT on GSP-RM

[Joe Finzel via B4 Relay]

From: Joe Finzel <joe@xxxxxxxxx>

The DRM I2C-over-AUX helper starts each I2C message with an
address-only request with MOT set. r535_dp_aux_xfer() forwards it to
GSP-RM with MOT cleared, which issues a stop on the DDC bus.

E-DDC resets the segment pointer (I2C address 0x30) on a stop. The
segment write is thus lost when the next message (offset write at 0x50)
starts, and the sink returns segment 0 again: EDID blocks 2 and 3 read
back as blocks 0 and 1. On a Dell AW2723DF connected to an RTX 3070 Ti,
block 2 is its DisplayID extension with the only 2560x1440@240Hz
timing; the kernel saw a second copy of the base block instead, and the
highest mode offered was 2560x1440@144Hz. Windows reads the correct
block from the same monitor.

Forwarding the start with MOT set instead left the display without a
signal on the same system. NVIDIA's own DisplayPort library never sends
GSP-RM an address-only request with MOT: it writes the segment and the
offset as data requests with MOT and ends with a read without it. Do
the same and skip the start; the data requests carry the address and
MOT themselves, and the address-only stop that ends a transfer is still
sent as before.

Tested on an RTX 3070 Ti (GA104, GSP-RM 570.144) with the AW2723DF over
DisplayPort: the EDID read back now matches the one Windows reads byte
for byte, and 2560x1440@240Hz is among the probed modes.

Fixes: 9e9944449023 ("drm/nouveau/disp/r535: initial support")
Cc: stable@xxxxxxxxxxxxxxx
Signed-off-by: Joe Finzel <joe@xxxxxxxxx>
---
 drivers/gpu/drm/nouveau/nvkm/subdev/gsp/rm/r535/disp.c | 14 ++++++++++++++
 1 file changed, 14 insertions(+)

diff --git a/drivers/gpu/drm/nouveau/nvkm/subdev/gsp/rm/r535/disp.c b/drivers/gpu/drm/nouveau/nvkm/subdev/gsp/rm/r535/disp.c
index f5f22173fc2..13a27476875 100644
--- a/drivers/gpu/drm/nouveau/nvkm/subdev/gsp/rm/r535/disp.c
+++ b/drivers/gpu/drm/nouveau/nvkm/subdev/gsp/rm/r535/disp.c
@@ -1072,6 +1072,20 @@ r535_dp_aux_xfer(struct nvkm_outp *outp, u8 type, u32 addr, u8 *data, u8 *psize)
 	int ret;
 	int retries;
 
+	/*
+	 * The DRM I2C-over-AUX helper starts each I2C message with an
+	 * address-only request with MOT set. GSP-RM is only ever given complete
+	 * requests (NVIDIA's DisplayPort library sends none of these), and
+	 * forwarding it as below, with MOT cleared, issues a stop: the sink then
+	 * resets the E-DDC segment pointer written just before, and EDID blocks
+	 * 2 and 3 read back as blocks 0 and 1. The data requests that follow
+	 * carry the address and MOT themselves, so the start can be skipped.
+	 */
+	if (!size &&
+	    NVDEF_TEST(type, NV0073_CTRL, DP_AUXCH_CMD, TYPE, ==, I2C) &&
+	    NVDEF_TEST(type, NV0073_CTRL, DP_AUXCH_CMD, I2C_MOT, ==, TRUE))
+		return 0; /* ACK */
+
 	for (retries = 0; retries < 3; ++retries) {
 		ctrl = nvkm_gsp_rm_ctrl_get(&disp->rm.objcom, NV0073_CTRL_CMD_DP_AUXCH_CTRL, sizeof(*ctrl));
 		if (IS_ERR(ctrl))

---
base-commit: fb26cf4f1df40a462ea65552a3fbcc8cfafab523
change-id: 20261021-nouveau-edid-segment-357805c19e7e

Best regards,
-- 
Joe Finzel <joe@xxxxxxxxx>