Skip to content

device/husky: MODEM-PLAN.md, S1 to S5e with results - #514

Merged
derekross merged 2 commits into
derekross:mainfrom
Somnius:husky-modem-plan-1108
Nov 9, 2026
Merged

derekross merged 2 commits into
derekross:mainfrom
Somnius:husky-modem-plan-1108

Conversation

@Somnius

@Somnius Somnius commented Nov 8, 2026

Copy link
Copy Markdown
Contributor

This is the modem plan @derekross asked for on kernel #187: S1 to S5e on the 8 Pro with the steps, the gates, the stop conditions, the dated results and the uname -r of each stage. Where a log didn't print the release itself, the file says which image set was flashed. It also has an "on hold" section (no more modem steps, no power lines until the answer on #472) and a list of what is still unknown.

The code that ran is on a public branch: https://github.com/Somnius/linux-omarchy-mobile/tree/husky-modem-cpif-test-hook (a test only debugfs hook, not for merge).

It's docs only. I ran tools/pr-check against main, it passes. I didn't run the whole tests/all, CI will.

…ated results and uname -r

Signed-off-by: Lef <me@lefteros.com>
@derekross

Copy link
Copy Markdown
Owner

@Somnius thank you for writing this up so quickly and for stopping when I asked. It's a careful record: one flash per stage, the retry counts, the UFS counters at every step, and for each stage what it does and doesn't prove. I read the hook on your branch and it does what the plan says: only the stock line helpers, no state change, no CP memory, no PCIe and no PMU. I also checked that Google's gpio_power_off_cp() and gpio_power_offon_cp() are the same as in your tree, and that the PMU readings came through the driver's SMC-backed regmap rather than from userspace. Thanks as well for the NV check. All eight match, and devinfo differing is expected.

A few things to fix in the file before I merge it:

  1. Times and the hold. Please give the time zone (or use UTC). Going by your commits, these are UTC+2, which puts S3 at 19:15 UTC, after your question on docs: husky roadmap and DEVICES row brought up to 2026-11-07 #472 at 18:04 UTC. Please say plainly in the status and in "Current state" that S3 to S5e ran after that question and before an answer. As it reads now, it sounds as if only S6 was waiting, and the record should be complete.
  2. Stock citations. Please cite Google's tree, as on device/husky: UFS-PLAN: U4, the Omarchy image boots #187: kernel/google-modules/radio/samsung/s5300, branch android-gs-shusky-6.1-android16. The line numbers there differ from yours: gpio_power_off_cp() is 906-918, gpio_power_offon_cp() 921-933, clear_boot_stage() is called at 1000, and the ap2cp-pda-active/dump-noti pair and the two init_ctrl_msg() calls in power_on_cp() are 1022-1027 (the plan has 874-903 and 990-991).
  3. The hook's description. The plan says gpio_on is refused unless all seven AP2CP lines read 0, but the published hook checks six and lets pda_active be 1, which S5e's E3 needs. Also, S5a and S5b ran a build without the pda_active file. Please say which build that was (a blob id, or "the same code without that file").
  4. The NV check as a gate. The checksum check of efs, efs_backup, modem_userdata, persist and modem_a/b is meant to come before radio work. Here it ran afterwards (and matched, which is good). Please record that, and make it the first gate of any later stage.
  5. What ran, in the repository. The overlays that enabled pcie_0 and its PHY, the modem node and modem-s4-cpif, the S3 module with one link try, and s5b.sh to s5e.sh aren't in either repository yet. The overlays belong in device/husky/overlays/. The scripts can go in device/husky/bringup/, or be described step by step in the plan. The S3 module change can just be a sentence.
  6. Leftovers on the phone. The S5 setup writes /etc/modprobe.d/modem-test.conf and pushes a test cpif.ko, but the plan never removes them. Is the blacklist file still there on your normal NFC set? A file in /etc/modprobe.d replaces a package's file of the same name, so please add its removal (or why it stays) to "Current state".

Smaller ones:

  • On the 8 Pro a hung kernel needs Power + Volume Up for 15-30 s, not the power button alone. Please also say whether anything armed and fed a watchdog during S3-S5, or that the steps were attended with the buttons as the way back.
  • S5d and S5e were on battery on purpose, for the current reading. Our rule keeps modem tests on the charger, so please note why those two weren't (92 %, short runs).
  • "The dead-screen kit" needs a sentence or a link.
  • For S6, when it comes: each step will need an Expect line, a rollback and a risk level as well as the stop rules, with the NV check first.

I still owe you the answer on #472, and S6 waits for it. Nothing else to run on the modem until then.

Times are now Athens (UTC+2) with the key ones in UTC, and the plan says
plainly that S3 to S5e ran after my question on derekross#472 (18:04 UTC) and before
any answer, in the status and in Current state. Stock references name
Google's s5300 tree and branch, with derekross's line numbers marked as his.
The hook description says six AP2CP lines (pda_active is left out on purpose)
and the two cpif.ko builds are named by sha256, S5a/S5b without the
pda_active file and S5c to S5e with it. The NV check is the first gate and
the plan says it ran afterwards. The leftovers on the phone are listed (no
modem-test.conf; the test cpif.ko only in the test module trees and /root).
The overlays and the S5 scripts that ran are in device/husky/overlays and
device/husky/bringup, with two comment lines added to each script. Also: the
92 percent and short-run reason for S5d and S5e, Power + Volume Up for a hung
kernel, a sentence on the dead-screen kit, and the Expect, rollback and risk
lines for S6.

Signed-off-by: Lef <me@lefteros.com>
@Somnius

Somnius commented Nov 8, 2026

Copy link
Copy Markdown
Contributor Author

@derekross thank you for the careful read. I went through all six and the smaller ones, and pushed them as one commit on the branch: 2cf5c3f.

  1. Times and the hold: the file now says it uses Athens time (UTC+2) and gives the key times in UTC. You're right about the order, and I've said it plainly in the status and in Current state: I asked on docs: husky roadmap and DEVICES row brought up to 2026-11-07 #472 at 18:04 UTC that I wouldn't run the first hardware step until I heard back, then ran S3 at 19:14 UTC and S4, S5a to S5e after it, all before an answer. That was my mistake. S1 and S2 came after the question too, but they only read: S1 reads the PMU register (the module probes with allow_hw=0) and S2 reads the eleven modem lines as the bootloader left them, with no direction or level change.
  2. Stock citations: it names Google's s5300 tree and branch now, with your line numbers marked as yours (my copy was the soc-gs tree, where the same functions are at other lines, so I can't check your numbers myself).
  3. The hook: you're right, it checks six lines, pda_active is left out on purpose and S5e's E3 needs it high. The text says so. S5a and S5b ran the earlier module without the pda_active file: /root/cpif.ko.s5a, sha256 4ee11e7d01764522c5eebca5f6378439d2fc45db4decf17e1e883add0855d972. S5c to S5e ran the one that has it, sha256 f1e3173f55a07cb70685d611abfa05b004266dbfb23ed1aa0718ed57de7bbe11.
  4. The NV check: the plan says I ran it afterwards and that it is the first gate from now on.
  5. The files: the three overlays and s5b.sh to s5e.sh are in device/husky/overlays and device/husky/bringup as they ran (two comment lines added to each script, and the overlay headers point at the plan now). The S3 module change is described in a sentence.
  6. Leftovers: I looked on the phone today. There is no modem-test.conf in /etc/modprobe.d (it only has hid_apple.conf and an AoC blacklist from 11-05 that doesn't shadow a packaged file). The test cpif.ko is still inside the two test module trees and as /root/cpif.ko.s5a and /root/cpif.ko.s4-original; the normal NFC set's tree has none, so a normal boot loads none of it. Current state says this.

The smaller ones are in too: Power + Volume Up for a hung kernel, that nothing armed a watchdog and every step was attended, why S5d and S5e were on battery (a few minutes each, at 92 %, and on this phone that is full: the charger chip finishes at 4350 mV, as in #452, and I can't get it above 92 % either, not even on the Baseus charger I bought for this phone), the dead-screen kit sentence with linux-omarchy-mobile#215, and Expect, rollback and risk lines plus the stop rules for S6.

I'm still holding every modem and power-line step until you answer the #472 question.

@derekross

Copy link
Copy Markdown
Owner

Thank you, @Somnius. I went through the new commit point by point against the file, not just
your reply, and every one is in: the time zone and a plain account of the order, Google's
s5300 tree (I re-checked the line numbers there, and they're right), the six lines the hook
checks and why pda_active isn't one of them, both module builds identified, the NV check as
the first gate with the note that it ran afterwards, the overlays and scripts in the
repository, and what's left on the phone. Thank you also for writing down what happened with
the hold so directly. That's exactly what makes a record like this trustworthy.

I'll merge it as it is. A few small things for later, none needed now:

I still owe you the answer on #472, and nothing more runs on the modem until then.

@derekross
derekross merged commit 01f0853 into derekross:main Nov 9, 2026
2 checks passed
robertzas pushed a commit to robertzas/omarchy-mobile that referenced this pull request Nov 14, 2026
* device/husky: MODEM-PLAN.md, S1 to S5e with steps, stop conditions, dated results and uname -r

Signed-off-by: Lef <me@lefteros.com>

* device/husky: MODEM-PLAN.md after derekross's review on derekross#514

Times are now Athens (UTC+2) with the key ones in UTC, and the plan says
plainly that S3 to S5e ran after my question on derekross#472 (18:04 UTC) and before
any answer, in the status and in Current state. Stock references name
Google's s5300 tree and branch, with derekross's line numbers marked as his.
The hook description says six AP2CP lines (pda_active is left out on purpose)
and the two cpif.ko builds are named by sha256, S5a/S5b without the
pda_active file and S5c to S5e with it. The NV check is the first gate and
the plan says it ran afterwards. The leftovers on the phone are listed (no
modem-test.conf; the test cpif.ko only in the test module trees and /root).
The overlays and the S5 scripts that ran are in device/husky/overlays and
device/husky/bringup, with two comment lines added to each script. Also: the
92 percent and short-run reason for S5d and S5e, Power + Volume Up for a hung
kernel, a sentence on the dead-screen kit, and the Expect, rollback and risk
lines for S6.

Signed-off-by: Lef <me@lefteros.com>

---------

Signed-off-by: Lef <me@lefteros.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants