Repository navigation
Conversation
|
Thanks Brett, this is a very complete port, and the 120 Hz numbers hold up: I ran Google's
Also, #90 changes the same DSIM compatible to a |
53151e6 to
9e2071b
Compare
|
Thanks Derek, and thanks for re-running
On #90: agreed. Whichever lands second lists
|
| Check | Result | Detail |
|---|---|---|
| boot (required) | ok | kernel 7.3.0-rc6-omamobile-husky-wifi, slot b, up 289s |
| storage guard (required) | ok | 62 partitions; only userdata and slot B's boot images writable |
| root (required) | ok | ext4 on userdata, 97G free |
| kernel health (required) | ok | no oops or BUG; 0 WARN, 0 error lines in dmesg |
| systemd units | ok | none failed |
| USB gadget (required) | ok | bound to 11210000.usb |
| display | ok | card2-DSI-1 connected, 1344x2992 |
| GPU | ok | render node on panthor |
| session (required) | ok | Hyprland and the Omarchy shell running |
| lock | ok | PAM file present, the shell can lock (passwordPam true, locked true) |
| touch | ok | SALIXP Touchscreen |
| buttons | ok | gpio-keys |
| Wi-Fi | ok | wlp1s0 on brcmfmac, link up |
| Bluetooth | skip | no controller |
| battery | skip | no battery power supply |
| sound cards | skip | none |
| sensors | skip | no sensor proxy |
| cameras | skip | no libcamera tools |
| thermal | ok | hottest zone 50 °C |
The "0 error lines" is only because I cleared the ring buffer with dmesg -C during the screen-off test on this boot. From journalctl -k -b -p err, the boot's error lines, leaving out brcmfmac's, are:
panthor: _opp_set_regulators: no regulator (mali) found, expected: ACPM sets the rail;- the HK3's one refused brightness write before
allow_tx(sending dcs data 51 02 bc failed: -1), whose level is kept and sent again; exynos-clkout: cannot match parent device, from the zuma PMU already onomamobile, and on the d3 boot too;- systemd's gpt-auto generator, and "unable to open an initial console".
|
Thanks Brett, all four points are in, and the screen-off test is exactly what I was after:
On the Pixel 7 Pro side, The merge order: Lef's #163 goes first. This PR merges cleanly on top of it and of #175, so it comes next, without another rebase. #158 then rebases once over both. If #90 lands before this one, the DSIM binding needs the one-enum form we agreed on. |
|
A correction to the order above: #96 merged first, and it carries the Pixel 9a's DSIM binding (#90's patch), so this now conflicts in |
|
Brett, since my last comment the Pixel 9a's DSIM binding landed, so #160 now conflicts in |
The Tensor G3 (zuma) DECON0 is the Tensor G4's: Google's display driver builds zuma with the same CAL (cal_9865), and the Pixel 8 Pro's stock DECON0 node has zumapro's main registers (0x19470000), DPP0 read DMA (0x19900000), interrupts (SPI 243, 242, 241) and fourteen windows. Add its compatible. Assisted-by: LLM Signed-off-by: Brett-Michael Green <brett@bmgvisualtech.com>
The Tensor G3 (zuma) DECON0 and DPP0 are the Tensor G4's: Google's display driver builds zuma with cal_9865, the CAL the zumapro support follows, and the Pixel 8 Pro's stock DECON0 sits at the same addresses with the same interrupts and fourteen windows. Match google,zuma-decon with zumapro's register layout, padded pitch and hardware trigger, and the Pixel 8 Pro's HK3 panel modes: 1344x2992 at 60 and 120 Hz, with the porches of Google's panel-gs-hk3.c and its 70 x 155 mm size. Assisted-by: LLM Signed-off-by: Brett-Michael Green <brett@bmgvisualtech.com>
Describe zuma's DECON0 with its DPP0 read DMA window and its frame start, frame done and extra interrupts, from the Pixel 8 Pro's stock node, as zumapro.dtsi does. It is disabled: a board enables it with its scanout pool and boot framebuffer. Assisted-by: LLM Signed-off-by: Brett-Michael Green <brett@bmgvisualtech.com>
Turn on DECON0 on the Pixel 8 Pro with a 64 MiB reusable scanout pool below 4 GiB, for DPP0's 32-bit DMA, in RAM that stock leaves free, and the bootloader's splash as its boot framebuffer. DPP0 already scans the splash out by its physical address (RDMA_BASEADDR_P0 reads 0xfac00000). There is no DSIM on its output port yet, so gs-decon stays at the bootloader's 60 Hz. Assisted-by: LLM Signed-off-by: Brett-Michael Green <brett@bmgvisualtech.com>
The Tensor G3 (zuma) DSIM0 has gs201's register layout for the command path: Google's cal_9865/regs-dsim.h, which its display driver builds for zuma, has cal_9845's offsets for every register gs-dsim-cmd uses, and on the Pixel 8 Pro they hold the bootloader's command-mode setup (version 0x02090100). Add google,zuma-dsim next to google,zumapro-dsim, with google,gs201-dsim as the fallback of both. Assisted-by: LLM Signed-off-by: Brett-Michael Green <brett@bmgvisualtech.com>
Describe zuma's DSIM0 with its link and D-PHY windows and its interrupt (SPI 252), from the Pixel 8 Pro's stock node, as Google's zuma tree has them. It is disabled: a board enables it with its panel. Assisted-by: LLM Signed-off-by: Brett-Michael Green <brett@bmgvisualtech.com>
Add the command-mode timers for the Pixel 8 Pro's HK3 link, so gs-decon can switch it between 60 and 120 Hz. The 60 Hz pair is the bootloader's (UNDERRUN_CTRL 0x2235, CMD_TE_CTRL1 0x34c53c1e). At 120 Hz the TE protect and timeout fields are halved, as for the other two links, and the underrun is Google's dsim_calc_underrun() for 4 lanes at 1368 Mbit/s, DSC 3:1, 1344x2992 and HK3's TE idle of 350 us with a 1 % variance: the same formula gives the bootloader's 60 Hz value exactly. Assisted-by: LLM Signed-off-by: Brett-Michael Green <brett@bmgvisualtech.com>
The HK3 is the Google Pixel 8 Pro's 1344x2992 command-mode AMOLED panel: DSC 1.2 with two 672x187 slices at 8 bits per pixel over four DSI lanes, a TE line, and 60 or 120 Hz, as in Google's panel-gs-hk3.c. Its supplies and reset line are described as on the TG4C, though the driver takes the panel over as the bootloader left it. Assisted-by: LLM Signed-off-by: Brett-Michael Green <brett@bmgvisualtech.com>
Add a driver for the Google Pixel 8 Pro's HK3 command-mode AMOLED panel, with command sequences from Google's MIT-licensed panel-gs-hk3.c. Like the TG4C driver, it takes the panel over as the bootloader left it: out of sleep, with the display on, at 60 Hz. It never resets the panel or touches its supplies. It: - puts the panel to sleep and wakes it, with Google's display on and off sequences around display on/off (the AMP type and Vreg changes, with the Vreg of panels from DVT1 on, the retail ones); - switches it between 60 and 120 Hz in high-speed (HS) manual mode, sending the whole of Google's panel feature update for that mode each time: fixed TE and early exit at 120 Hz, changeable TE and no early exit at 60 Hz, then the manual frequency, applied together by the frame update command. The HBM and IRC settings stay as the bootloader left them; - exposes its brightness as a backlight over Google's normal range, DBV 196-3307 (2-1000 nits), using pixel off at 0 as Google does. A level set while the panel is asleep, or before gs-dsim-cmd allows commands (allow_tx), is kept and sent with the next write, so a boot-time restore of the saved level does not fail. Assisted-by: LLM Signed-off-by: Brett-Michael Green <brett@bmgvisualtech.com>
Enable DSIM0 with the Pixel 8 Pro's HK3 panel and connect it to DECON0's output port, so gs-decon switches the link and the panel between 60 and 120 Hz and the panel driver handles brightness and sleep. The panel is described without its reset line and supplies: the driver takes it over as the bootloader left it, and the main PMIC's regulators are not described yet. Assisted-by: LLM Signed-off-by: Brett-Michael Green <brett@bmgvisualtech.com>
9e2071b to
d5c35c5
Compare
|
Thanks Derek, and sorry for the wait. Rebased onto today's
The other eight commits picked up unchanged (
I haven't booted this exact head. Apart from the binding, the only change is the merge with the Wi-Fi nodes, and e1 already ran that combination. For the order: #158 (rebased today as well) conflicts with this one only in |
|
A boot of the exact heads now, as promised:
What the two PRs do on this boot:
|
| Check | Result | Detail |
|---|---|---|
| boot (required) | ok | kernel 7.3.0-rc6-omamobile-push1-g54e10dcccd6b, slot b, up 57s |
| storage guard (required) | ok | 62 partitions; only userdata and slot B's boot images writable |
| root (required) | ok | ext4 on userdata, 97G free |
| kernel health (required) | ok | no oops or BUG; 0 WARN, 46 error lines in dmesg (34 known, 12 new) |
| new dmesg errors | not ok | N N: brcmf_msgbuf_rxbuf_data_post: No PKTID available !! x2; Bluetooth: N: BCM: '…' x2; exynos-clkout exynos-clkout: probe with driver exynos-clkout failed with error N x1 |
| systemd units | ok | none failed |
| USB gadget (required) | ok | bound to 11210000.usb |
| display | ok | card2-DSI-1 connected, 1344x2992 |
| GPU | ok | render node on panthor |
| session (required) | ok | Hyprland and the Omarchy shell running |
| lock | ok | PAM file present, the shell can lock (passwordPam true, locked true) |
| touch | ok | SALIXP Touchscreen |
| buttons | ok | gpio-keys |
| Wi-Fi | ok | wlp1s0 on brcmfmac, link up |
| Bluetooth | ok | hci0, powered true |
| battery | ok | 20% Charging, 22 °C |
| sound cards | skip | none |
| sensors | skip | no sensor proxy |
| cameras | skip | no libcamera tools |
| thermal | ok | hottest zone 49 °C |
The "new" dmesg lines, none from these PRs:
- brcmfmac's
No PKTID available(Somnius's one-line slack, still to land); - the BCM4398's Bluetooth ID lines, which the smoke script counts as errors;
exynos-clkout: cannot match parent device/ probe -22, the zuma PMU message already onomamobile(clk: samsung: exynos-clkout: NFC clock pad on the Pixel 8 Pro (zuma) #209 fixes it).
|
Thank you, Brett, and welcome back. Everything from my 11-01 comments is in: the DSIM binding has I checked the Pixel 7 Pro side again on today's base: its device tree is byte-identical, gs-decon The one hiccup isn't yours: GitHub's run for your 18:13Z push failed before it started any job |
|
Merged, thank you, Brett. CI's second run passed on this head (the first one never started a job, which was GitHub's side). For #158: please rebase it over this, and take the Pixel 8a point from my comment there in the same push. |
10 commits on
omamobile, rebased after #153 landed. It doesn't depend on #157 or #158.This gives the Pixel 8 Pro a real display path. gs-decon, the flip-only KMS driver, takes over from simple-framebuffer, and gs-dsim-cmd with a new HK3 panel driver adds 120 Hz, brightness and sleep. It mirrors the Pixel 9a's display work (#88, #89, #90): zuma's DECON and DSIM are the Tensor G4's.
Why zumapro's support carries over: Google's display driver builds zuma with
cal_9865(its Kbuild hasCONFIG_SOC_ZUMA→cal_9865), the CAL gs-decon's zumapro data follows. husky's stock DECON0 node has zumapro's main registers (0x19470000), DPP0 read DMA (0x19900000), interrupts (SPI 243, 242, 241) and 14 windows.cal_9865/regs-dsim.hhascal_9845's offsets for every register gs-dsim-cmd uses. DSIM0 is at0x19440000on SPI 252.The commits:
dt-bindings: display: google,gs201-decon: Add the Tensor G3drm: google: gs-decon: Add the Tensor G3: zumapro's register layout, padded pitch and hardware trigger, with the HK3 panel's modes from Google'spanel-gs-hk3.c: 1344x2992 at 60 and 120 Hz, porches 80/24/42 and 12/4/22, 70 x 155 mm.arm64: dts: exynos: google: zuma: add DECON0(disabled)arm64: dts: exynos: google: zuma-husky: enable DECON0: a 64 MiB reusable scanout pool below 4 GiB, in RAM stock leaves free, and the splash as the boot framebuffer.dt-bindings: display: samsung: google,gs201-dsim: Add the Tensor G3:google,zuma-dsimwith the gs201 fallback, as arm64: dts: exynos: google: zumapro-tegu: Enable the display #90 does for zumapro.arm64: dts: exynos: google: zuma: add DSIM0(disabled)drm: google: gs-dsim-cmd: Add the Pixel 8 Pro's link timersdt-bindings: display: panel: Add Google HK3drm/panel: Add Google HK3 panel driverarm64: dts: exynos: google: zuma-husky: put the HK3 on DSIM0The 120 Hz link timers (7):
UNDERRUN_CTRL 0x2235,CMD_TE_CTRL1 0x34c53c1e.dsim_calc_underrun()reproduces0x2235exactly for 4 lanes at 1368 Mbit/s (word clock 85.5 MHz), DSC 3:1, 1344x2992, 24 bpp and HK3's TE idle of 350 us with a 1 % variance.0x06a7. The TE protect and timeout fields are halved, as for the other two links, giving0x1a621e0f.The HK3 driver (9): command sequences from Google's MIT-licensed
panel-gs-hk3.c. Like the TG4C driver, it takes the panel over as the bootloader left it and never resets it or touches its supplies.F7 0F). HBM and IRC stay as the bootloader left them.-EPERM), is kept and sent with the next write. That way systemd-backlight's boot-time restore succeeds, and its unit stays active to save the level at shutdown.Tested on husky (tags
husky-test-20261031-d1, the first 4 commits, and-d2, all 10, each merged with #156 and #158, built and installed withmodules-push):GLOBAL_CON 0x113(command mode, running),BLD_BG_SIZE/RDMA_IMG_SIZE 0x0bb00540(1344x2992) andRDMA_BASEADDR 0xfac00000;allow_txoff, gs-dsim-cmd dumps DSIM0 and recognises the timers ("link timers for 60 Hz"), and the HK3 driver binds with its backlight. A brightness write from userspace is refused;allow_txon, brightness writes reach the panel: 1600 and then 600 dimmed the screen, and Omarchy's own brightness control works;0x6a7/0x1a621e0f;gs_dsim_cmd.allow_txbefore gs_decon as the 9a's graphics start does; tag-d3, its files installed under/etcand/usr/local):google-hk3.20800abefore gs_decon loads;husky-test-20261101-e1, withgs_decon screen_off=2, which wifi: brcmfmac: confine the Bluejay BCM4389 bring-up #228 now sets): four DPMS off/on cycles with DSI commands allowed. Each time DECON stopped and the panel went to sleep, and DPMS on woke it at 120 Hz. Over the cycles: no flip timeouts, none late, the worst flip at 10.4 ms, and no DSI or panel errors.tools/kernel-pr-check --build --mergepasses: it merges cleanly onto 51fb40d, W=1 builds of the 7 touched C files pass, 8 device trees have no new dtbs_check warnings, and dt_binding_check passes on 10 bindings. checkpatch --strict has one note: the HK3 binding's MAINTAINERS entry arrives with its driver in the next commit.Not here yet:
DRM Dumb: created a dumb allocator, 3 x XR24), so frames rendered on panthor are copied into them rather than scanned out directly. The DPU's SysMMU would remove that copy.🤖 Generated with Claude Code