Repository navigation
drm/apple: K116s DCP: PMU temperature for every panel, syslog filter, 27.2 D590 - #1646
Conversation
81d1f3d to
d30426f
Compare
RyanTheTide
left a comment
There was a problem hiding this comment.
@sithglan host-tests fails in apple-dcp because boot.py (around line 97) asserts that the internal DCP rtkit_ops has no syslog_is_debug hook. 4e7352f adds that hook, so please update the selftest in the same commit to match the new policy.
Also, the PMUS change answers the firmware on every panel now, so the firmware gets a temperature where it used to get "false". Please describe that in the commit message as a behaviour change. We will run it on our LCD panels once CI is green.
d30426f to
907bb7e
Compare
|
@RyanTheTide both done, force-pushed as 907bb7e (was d30426f).
Both edited commits carry a |
… fallback The firmware asks the AP for the PMU temperature property at boot (callback D401 "PMUS"/"Temperature") for its panel compensation's initial guess; the driver answered it only for mini-LED panels, so the OLED K116s logged "set_PMU_service: failed to read PMU temperature property!" twice. Answer with the battery temperature as the H17P and v14.7 paths do. The DCP also reads eleven SMC temperature keys itself for its 2D virtual temperature model and logs once a second that TAOC is -127.0, the SMC's no-sample value, and that it uses the 2DVT fallback. TAOC and its siblings are AOP-sourced sensors the SMC has no sample for until the AOP's sensor servers run, so route that record to the debug level; another sensor, value or verdict stays at info. This is a behaviour change for every panel that is not mini-LED, on every firmware this template drives: the firmware's PMUS "Temperature" request used to get "false" there and now gets the battery temperature, which its panel compensation then takes as the panel's initial temperature. Mini-LED panels got the temperature before and still do. Signed-off-by: Jens Axboe <axboe@kernel.dk> [thomas: describe the PMUS reply as a behaviour change] Signed-off-by: Thomas Glanzmann <thomas@glanzmann.de>
The filter that demotes known-benign firmware syslog records was only in the external H17P RTKit ops, so the TAOC fallback record still logged once a second on the internal panel. The apple-dcp host test's syslog check expects the hook on both RTKit ops now. Signed-off-by: Jens Axboe <axboe@kernel.dk> [thomas: update the apple-dcp host test to the new policy] Signed-off-by: Thomas Glanzmann <thomas@glanzmann.de>
The 27.2 firmware sends a 20-byte D590 notification with no reply after swaps (the macOS handler next to swap_info is the buffer-release notification). Nothing here consumes it, so take it as a no-op instead of warning about an unknown callback. Signed-off-by: Jens Axboe <axboe@kernel.dk>
907bb7e to
bfc0858
Compare
|
@RyanTheTide rebased onto integration/aurora 514ae1f. It is now bfc0858 (was 907bb7e); the old branch no longer merged.
Testing at 514ae1f:
Both points of your 11-12 review were addressed in 907bb7e (my reply above), and they are unchanged here. Could you take another look? |
|
Both points are addressed on bfc0858. Since PMUS now answers the firmware on every panel, the next step on our side is a boot, brightness and DPMS row on the internal panels we have (T8112, T6040, T8140). |
DjDeveloperr
left a comment
There was a problem hiding this comment.
Thanks Thomas, and thanks to Jens for the commits. Ryan's two points from 11-12 are done: the apple-dcp selftest now expects the hook on both transports, and the PMUS commit message calls out the behaviour change.
What I checked against the current integration/aurora (458f759):
- The six files merge cleanly, and
[590]isn't defined anywhere else in either table iniomfb_v27.c. - The syslog match is exact: another sensor, value or verdict stays at info, as the new KUnit cases show.
- On a Mac without a battery,
dcp_pmu_temperature()keeps the 302.9 K fallback, so those panels get the same value the mini-LED path always sent.
Authorship: the three commits are Jens Axboe's, with his sign-off and, per the description, his OK. A small nit for next time: the third commit carries only Jens's Signed-off-by. When you pass on someone else's patch, add yours after theirs, as you did on the first two.
Post-merge verification: Ryan's boot, brightness and DPMS rows on the T8112, T6040 and T8140 internal panels. main has moved in dcp.c since your base, so I'm updating the branch and will merge once host-tests pass on the new head.
d91db38
into
aurora-silicon:integration/aurora
|
Merged in d91db38. Post-merge verification: boot, brightness and DPMS rows on the T8112, T6040 and T8140 internal panels. A revert is ready if any of them regresses. |
Base: integration/aurora (514ae1f). Three commits, all in drm/apple. None of them changes behaviour, only what reaches the log:
Not included: Jens's "drm/apple: drive firmware 27.2 with the 27.0 interface and its new method numbers". Its 27.2 tags are exactly the ones in the 27.2 table of the open #1174 (Jeff's table from #894, choosing the table by
apple,firmware-compat). Jens's K116s confirms that table on hardware: AppleDCP-1317.40.16 with the +50 IOMobileFramebufferAP numbering, A389 to A390, and unchanged callbacks.Tested by Jens on the K116s (27.2, 26B8125), internal OLED panel through DCP.
Build: arm64, LLVM 22;
drivers/gpu/drm/apple/, no warnings.Authorship. Jens Axboe wrote these commits and runs them on his K116s (16-inch MacBook Pro, M5 Pro, OLED; Mac17,12; stub macOS 27.2 26B8125). They come from his K116s bundle of 2026-11-11 (integration/aurora 254c9bd plus 34 commits, kernel 7.3.0 running on the machine). I rebased them onto integration/aurora 514ae1f and removed the LLM co-author and session trailers. Jens has given his OK: the commits carry his Signed-off-by.