On 20/03/2026 20:48, Cory Keitz wrote:
On Sun, Mar 01, 2026 at 01:51:28AM +0100, David Heidelberg via B4 Relay wrote:-static int csid_set_clock_rates(struct csid_device *csid) +static int csid_set_clock_rates(struct v4l2_subdev *sd, struct csid_device *csid) { struct device *dev = csid->camss->dev; + struct csiphy_device *csiphy = v4l2_get_subdevdata(sd); + struct csiphy_lanes_cfg *lane_cfg = &csiphy->cfg.csi2->lane_cfg;I believe there's a regression here from v3. I've been testing v4 of this series on SA8775P (kernel 6.18) with a GMSL2 camera pipeline and hit an invalid pointer dereference in csid_set_clock_rates() during pipeline power-up: pc : camss_get_link_freq+0x1c/0xc4 [qcom_camss] lr : csid_set_power+0xc8/0x404 [qcom_camss] Code: 910003fd a90153f3 aa0003f4 a9025bf5 (b9400040) The sd passed here from csid_set_power() is the csid subdev, so the subdevdata is a csid_device rather than a csiphy_device. The csid_device stores its linked csiphy_id during csid_link_setup(), so the fix is to look it up through the camss device array: static int csid_set_clock_rates(struct csid_device *csid) { struct csiphy_device *csiphy = &csid->camss->csiphy[csid->phy.csiphy_id]; struct csiphy_lanes_cfg *lane_cfg = &csiphy->cfg.csi2->lane_cfg; This also lets us drop the v4l2_subdev parameter entirely since it's no longer needed. With the above change my pipeline powers on and streams correctly.
Thank you for catching this, until I'll sent next revision, it's fixed in the git branch [1].
David [1] https://codeberg.org/sdm845/linux/commits/branch/b4/qcom-cphy
Tested-by: Cory Keitz <[email protected]> --- Regards, Cory
-- David Heidelberg
