Repository navigation
T6050 sleep: select drm/apple v27 transport and asahi g17 provider by hardware, not board - #2442
RyanTheTide wants to merge 3 commits into
Conversation
Several drivers select their T6050 (M5 Pro and Max) deep-sleep handling with a list of boards, the ones the sleep has run on (J714s, J716c). What those boards have in common is in the device tree: an available sleep-controller node, compatible "apple,t6050-sleep". No other board in the tree has one; a board gains it with its sleep description. Add apple_sleep_described(compatible), true when an available node with that compatible exists, and apple_t6050_sleep_described() for "apple,t6050-sleep". Both are static inline and independent of CONFIG_APPLE_SOC_SLEEP, as the board checks they replace are. Signed-off-by: Ryan Murray <ryan@aurorasilicon.org>
Besides the J714s's own firmware 27.0 path, the generic 27.0 sleep transport was taken on a machine that is "apple,j716c" and "apple,t6050". It belongs to the T6050 sleep, so take it where the device tree describes that sleep (apple,t6050-sleep). Nothing changes in the tree today: the J716c is as before; the J714s takes its own path first when the firmware is 27.0, and with any other firmware the generic transport's own 27.0 check refuses it, as the board check did; no other board describes the sleep. The J714s-only rules (dcp_j714s_ext(), v27_j714s, the fabric's sticky pipeline and the peer limit) are left as they are: the J716c runs a different method table and panel profile and has not run them. Fixes: fd5cc59 ("drm/apple: v27: the T6050 sleep transport on the J716c") Signed-off-by: Ryan Murray <ryan@aurorasilicon.org>
…iption The wrapped provider translation for the GPU's commit stop was used on a machine that is "apple,t6050" and "apple,j714s" or "apple,j716c". It belongs to the T6050 deep sleep, so use it for an "apple,agx-g17s" where the device tree describes that sleep (apple,t6050-sleep): the same two boards in the tree today, so nothing changes. Fixes: 541d9be ("drm/asahi: g17: the M5 Pro's commit stop on the J716c too") Signed-off-by: Ryan Murray <ryan@aurorasilicon.org>
|
Note 🤖 Claude Opus 5.5 responding on behalf of Isaac-Duarte J714s deep-sleep row for this head (b9165ce). Setup: same tree as my row on #2440: J714s, firmware 27.0,
Firmware messages from the external DCP around the sleep ( Not run: J716c (I only have a J714s). On the J714s with 27.0, the transport change in |
Problem: the drm/apple firmware 27 sleep transport for the J716c and the drm/asahi G17 provider translation are switched on by board name. That is where they were run, not a hardware property.
Change: both now use
apple_t6050_sleep_described(), which is true when the DT has an availableapple,t6050-sleepnode. Only the J714s and J716c trees have one, so on every board in the tree today nothing changes. Builds on the helper commit at the base of #2440.Testing: W=1 compile of drm/apple dcp.o and drm/asahi g17/sleep.o (release config, clang) at base and with this branch: both build, no code warnings before or after. The only drm/asahi file touched is C; the asahi Rust objects are unchanged and were not rebuilt.
Still to run: @DjDeveloperr @erikwb, could you give a J714s and J716c row (deep sleep with the panel on and with a USB-C display, then resume)? Draft until then.
Related: #2384 also edits dcp.c (the J813 system-sleep stop, around :9405, and the s2r_untested text). The hunks do not overlap, and
git merge-treewith its head 8407d40 merges cleanly.Details
apple,t6050-sleepapple,t6050-sleepand agx-g17sdcp.c:5791 on the J714s: with firmware 27.0,
v27_j714sreturns first, as before. With any other firmware, the generic branch's ownfw_compat == 27.0test refuses it, as the board check did. So the result is the same on every board.Kept (J714s only):
dcp_j714s_ext()(dcp.c:473),v27_j714s(dcp.c:7003), the fabric's sticky pipeline (dcp-fabric.c:3363) and the peer limit (dcp-peer-v27.inc:72). The J716c runs a different IOMFB method table and panel profile and has not run these. Widening them would change the J716c, so they need a J716c row first.Hardware rows: J714s/J716c,
echo mem > /sys/power/statewith deep sleep. The panel comes back, and the GPU commit-stop lines in dmesg match the current tip.Original work: DjDeveloperr (T6050 display/GPU sleep), Erik Bourget (J716c).
Assisted-by: LLM (Claude:Opus 5.5)