Repository navigation
arm64: dts: exynos: google: zumapro-tegu: Add the Kepler GNSS link - #237
Conversation
The Pixel 9a's GNSS receiver is the Kepler core in its S5300 modem. Its data go over an SPI bus of the application processor, USI3 of HSI0 (spi21), in 64-byte blocks paced by two handshake lines; gnssif_spi drives both. Add the SPI device and the handshake node, and enable the USI. The bus keeps zumapro's port configuration, with which spi-s3c64xx moves a 64-byte block in one piece. With gs101's compatible alone (the controller's automatic chip select, as stock uses for this device) the driver cut each block into 60 and 4 bytes, and in two sessions on the phone 12 of those reads timed out (-110) in 13 minutes; with zumapro's, two more sessions had none. The feedback delay is stock's 0. The bus has no pin group: it runs on the bootloader's pin state, which the data were clean with. Nothing probes until the modules load, and the core answers only while the modem is up and the GNSS daemon has loaded its firmware. Signed-off-by: W4HO <87742609+W4HO-ham@users.noreply.github.com>
|
Thank you, @W4HO-ham. It's good to see GNSS move from the bring-up overlay into the board file, and I appreciate that you tested both bus configurations and wrote down why the one you kept is the right one. I checked the wiring against the file tegu's stock DT actually includes, Three things for afterwards, none of which hold this:
On my side: the drivers aren't built by the shared config yet, so this is inert until omarchy-mobile#594. When that goes in I'll make sure device-tegu with the blacklist is published no later than the first kernel with the option, so nobody's 9a autoloads the GNSS modules without the modem. |
|
Thank you for checking the wiring against the file tegu really includes.
|
|
The two handshake pins, read on 2026-11-10 on a build of the published tree (plus #224) with the gpa5-4 is an input with the pull-down, as stock has it. gpp2-3 is an input with a pull-down too, where stock's group has an output with no pull: the driver makes it an output when it loads, and nothing in the tree sets its pull, so the GNSS sessions most likely ran with that pull-down under a driven line. I haven't read the pins with the drivers loaded. The readings are in tegu's MODEM-PLAN, section P, with omarchy-mobile#590 (commit 05943847). |
The Pixel 9a's GNSS receiver is the Kepler core in its S5300 modem; its data go over
spi21(USI3 of HSI0) in 64-byte blocks, paced by two handshake lines. gnssif_spi (#185) drives both, but the board's device tree had neither node, so the driver could only be tried from a bring-up overlay. This adds the SPI device, the handshake node, and enables the USI.The bus keeps zumapro's SPI port configuration. My first version gave the bus gs101's compatible alone, for the controller's automatic chip select, as stock has it for this device. On the palmtop that worked, but spi-s3c64xx then cut each 64-byte block into 60 and 4 bytes (
poll_full_fifois zumapro's), and 12 of those reads timed out in 13 minutes over two GNSS sessions (gnss_spi spi1.0: I/O Error: rx-1 tx-0 rx-f tx-p len-4 ... res-(-110), thenlen-60). With zumapro's compatible kept (a block in one piece, polled; chip select held per message) two more sessions and 10 minutes of link cycling logged none. The feedback delay is stock's 0 in both. The bus has no pin group: it runs on the bootloader's pin state, as #151's closing found sufficient.Tested on the palmtop (2026-11-09,
7.3.0-rc6-omamobile-push1-g773b4102f5ec:omamobileatabcd6aac51cfplus #224, #236 and this change's first version, then this version's DTB on the same Image; defconfig, the shared config, omarchy-mobile#274's PCIe PHY andCONFIG_EXYNOS_GNSS_IF_SPI=m; the modem from the in-tree CPIF with omarchy-mobile#594's overlay):CHECK_DTBS=yforzumapro-tegu.dtbis quiet, checkpatch is quiet. Nothing changes for other boards: the nodes are inzumapro-tegu.dtsonly.