Skip to content

device/dipper: record stage-0 facts and first light - #330

Merged
derekross merged 22 commits into
derekross:mainfrom
quatrixone:dipper-support
Nov 11, 2026
Merged

derekross merged 22 commits into
derekross:mainfrom
quatrixone:dipper-support

Conversation

@quatrixone

@quatrixone quatrixone commented Nov 2, 2026 •

Copy link
Copy Markdown
Contributor

Summary

Record the Xiaomi Mi 8 (dipper) stock hardware facts, the historical 2026-10-29 first-light result, and the additional storage and modem activity that can be established from the available notes. This is a notes-only change.

The note contains no captures, serials, MAC addresses, Wi-Fi credentials, or private host/path details. Device support code and shared changes remain in their own PRs.

The current RAM-only image and storage guard keep UFS read-only by design. The historical recovery image checksum, logdump backup coverage, exact intervening boot sequence and first-light mount options still need source/backup verification before these notes can serve as a recovery record. No phone action was performed for this revision.

Checks

  • git diff --check — passed
  • No automated tests run; this PR changes documentation only.

Add Dipper device metadata, support package and firmware staging.

Record non-A/B safety rules and hardware evidence without owner network identities.

Keep local SSH settings out of the public tree.
@derekross

Copy link
Copy Markdown
Owner

Thank you for all this work, and for writing down so carefully what the hardware said. The first-light notes, the read-only stock backup with checksums on the phone, and keeping NV read-only throughout are exactly the habits this project runs on.

Phones without A/B slots are welcome. Keeping stock on slot A is how we avoid bricking phones during development, not a permanent requirement, so a Mi 8 port can become a supported device. #383 adds a section to CONTRIBUTING §3.1 for phones without a second slot. To get there, a few changes:

  • The rules live in CONTRIBUTING, not in this PR. Please drop the AGENTS.md changes (including the rtk line) and add the Mi 8's entry under CONTRIBUTING, install-phone: phones without A/B slots, and never flash a bootloader to match slots #383's section instead: the boot partitions the installer writes, and how to reach fastboot by the buttons.
  • tools/dipper/push-verified --to-partition can write any named partition as root from stock Android, with no allowlist. Please drop it or limit it to boot through our tools.
  • omamobile-dipper-identity mounts persist at boot. Even read-only, an ext4 mount can replay the journal. On the Pixel 7 Pro we read the files we need with debugfs and never mount it. Could it do the same?
  • The shared changes each need their own PR, with a reason that holds on every phone: freedreno in our Mesa (worth doing for every Qualcomm phone, but it rebuilds the Mesa every Pixel installs), the NFS root in the shared init, ROOTFS_DIST_PACKAGES, the build-packages order and the phone-ssh timeout.
  • The initramfs's no-USB-host panic would restart into the same image on a phone with no fallback slot; the descriptor and plan should say what happens instead.
  • The kernel side (the DTS and the panel driver) needs to be public, on a branch of yours or as a PR to linux-omarchy-mobile, before the descriptor can point at it.
  • In NOTES.md, please drop the sentence about how another distribution lists the touchscreen. We don't describe other projects' status in our docs. Citing a kernel tree for a fact, with a commit link, is fine.
  • The Tailscale join unit and the HANDOFF.md/STATE.md session notes don't belong in the repository.

A notes-only PR with the stage-0 facts and your first-light record is something I can take straight away while the rest comes together.

@quatrixone quatrixone changed the title device/dipper: add Xiaomi Mi 8 support device/dipper: record stage-0 facts and first light Nov 7, 2026
@quatrixone
quatrixone marked this pull request as ready for review November 7, 2026 22:28
Keep the early-probe lesson without linking to the kernel config removed from this notes-only PR.
The USB endpoint is project-standard and not needed to record first light; keep the public note free of private-range IPs.
@quatrixone

Copy link
Copy Markdown
Contributor Author

Follow-up: the Dipper device bring-up now has a separate ready PR, #505. It adds a RAM-only descriptor and build path; it refuses flash/push/install and keeps UFS read-only. Kernel prerequisites are public in linux-omarchy-mobile PRs #218–#221; the button recovery entry is in #499. Wi-Fi and per-device identity remain disabled until the identity path reads persist with debugfs without mounting it. The kernel build and hardware probe are still pending.

@derekross

Copy link
Copy Markdown
Owner

Thank you for cutting this down to the notes, and for working through every point from my last comment. I checked them one by one against the branch:

The identifier and blob checks pass on the branch. CI is waiting for my approval, and I'll approve it.

One thing has to change before I can merge this as a record, because it's about what was written on the phone.

The first-light record and the storage guard. Line 51 says "The storage guard works as designed: boot, recovery and persist read-only, userdata and logdump writable." Our guard can't leave logdump writable. The allowlist accepts only slot B's boot images and userdata, and the init refuses anything else. The first version of this branch explains why it was writable: tools/dipper/bringup-patch-init.py widened the allowlist to add logdump and copied the kernel log into that partition with dd every 2 s. Line 58's "two blind attempts from recovery" were that diagnostic image, written into the stock recovery partition from rooted Android (the build's note said "Write it to recovery only", and push-verified --to-partition did the writing). This all happened before CONTRIBUTING had any rules for phones without slots (those landed on 2026-11-03), so I'm not blaming anyone. These were still writes to two stock partitions outside the guard, though, and the notes are the place that says so. Could you replace that sentence with a short, plain record?

  • which partitions were written (recovery, logdump), on which dates, by what (the diagnostic image's log mirror, and dd from rooted Android);
  • what recovery holds now: stock's image again (its sha256 against the backup you took before the first write), or the diagnostic one;
  • that logdump now holds our log, and whether your backup covers it (the backup-stock list in the first version didn't include it);
  • and that the guard as it is now, and the RAM-only image in device/dipper: add a RAM-only first-light build path #505, leave every partition read-only.

Before the phone's next boot, either here or in a follow-up notes PR: the first version's state file records a lot more than first light, and the notes should hold it too. That means the NFS-root Omarchy sessions on 10-30, 10-31 and 11-02; the modem starts under rmtfs -r, with the XPU violation crash and the remoteproc recovery loops; and the 10-30 partition compare, where persist, modemst1 and frp changed and you put that down to stock Android's boots. For that compare, please list the boots between the two backups (date, what booted, what ran). A changed persist is only explained once a stock boot in that interval accounts for it. Then a later compare can tell a change of ours from stock's.

Smaller things:

  • Please give the first-light boot's kernel release (uname -r). Your state file has 7.3.0-rc6-omamobile-push1-g2a318a96ae9d-dirty. Please also say that it came from a tree that wasn't published then. The kernel PRs are its published successor, not the tested code.
  • "UFS up (all six LUNs, ext4 mounts)": which partitions were mounted, and with which options? If persist was ever mounted, even read-only, please say so, and check it against your backup.
  • The FTS5 section says the driver is "not yet built or hardware-tested", but the state file records a physical tap that produced ABS_MT_POSITION_X events on 10-30. Please record that result with its kernel, or say why it doesn't count. The section's paths (linux/drivers/..., "the DTS now declares") point at your private tree. Please point them at kernel omamobile-audio 0.5.6: earpiece mode, a Voice Call verb, and no stored Voice Call profile at startup #219 and [Device support request] Samsung Galaxy S22 Ultra #220 instead.
  • "Five defects in the recovered MR !71 work, three of them fatal" lists six. With the prior-art section removed, "MR !71" also no longer says what it is. A one-line citation with a link to the sdm845-mainline merge request is welcome. Citing a kernel tree for the code you started from is exactly right.
  • Line 5's "(private off-device reference directory (path omitted)/)" reads like a template left behind. "Captures live outside the repository because …" is enough.

When I merge, I'll write the squash message myself. The branch's earlier commits ("add Xiaomi Mi 8 support … Record non-A/B safety rules", the bluez and panel commits, the merge of main) describe code that isn't in it any more. The title is fine as it is.

@derekross

Copy link
Copy Markdown
Owner

Thank you for rewriting the first-light record. It now says plainly that the boot ran an unpublished tree, that a temporary init widened the allowlist for logdump, and that the diagnostic image went into stock recovery from rooted Android. That's the honest record I asked for, and the MR !71 link, the six-item count and the template leftover are all fixed too.

I checked each point from my last comment against the new head:

  • Kernel release and tree: done.
  • The guard sentence: mostly done. Two facts are still missing, and one of them I can answer from your own branch:
    • Dates. Please give the date of each write: the two recovery writes, and the boots whose init mirrored the log into logdump. As far as I can tell from your first version, that's 2026-10-29.
    • logdump and the backup. The notes say the backup's coverage of logdump is unknown. Your first version's tools/dipper/backup-stock had PARTS=(boot recovery dtbo vbmeta persist persistbak modemst1 modemst2 fsg fsc devinfo misc frp logo splash), so logdump isn't in it. Please say that, so nobody later reads "unknown" as "maybe covered".
  • What recovery holds now: recorded as not compared. That's an honest answer. The compare is a read-only check, though: the sha256 of recovery against the copy in your first backup, if that backup came before the 10-29 writes, or else against the recovery.img in Xiaomi's fastboot package for the MIUI build on the phone. I need it before I can decide on CONTRIBUTING: add Dipper single-slot recovery entry #499, so please do it before the phone's next boot, or say why it can't be done.
  • The later activity and the partition compare: the paragraph lists the sessions, but it now gives less than your first version did. That version recorded a verified 15-partition snapshot before the radio runs and a compare after them: boot, recovery, dtbo, vbmeta, fsc, devinfo, misc, logo, splash, persistbak, modemst2 and fsg byte-identical, only persist, modemst1 and frp changed, and all five NV partitions at ro=1 before and after your runs. Those are facts, and they belong in the record. Please put them back, together with "the cause is not established" as you have it now. The list of boots between the two snapshots (date, what booted: stock Android or which RAM image, what ran) is still what turns that into an answer. Until then the later snapshot shouldn't be called a baseline. The paragraph's last sentence already says this, which is right.
  • Mount options and persist: recorded as not known. Fine as a record. If you still have the 10-29 console or dmesg, the EXT4-fs (sdXN): mounted lines answer it.
  • FTS5: done, with the result and its kernel. One correction: the FTS5 binding is in kernel #219, not [Device support request] Samsung Galaxy S22 Ultra #220. [Device support request] Samsung Galaxy S22 Ultra #220 is the board.
  • Links: "kernel PRs device-dodge: follow-ups from #111 (fastboot refuse, install bind trap, park deliberate panic) #218–omamobile-device-husky 0.1.2: let the GPU clock scale to 890 MHz #221" in the first-light paragraph links to this repository's PRs. Please write them as links to linux-omarchy-mobile, as the FTS5 section does.

One correction of my own: on the kernel PR I doubted that the upstream FTS5 series had been tested on the Mi 8. David's cover letter says it was ("Also, tested on Xiaomi Mi 8"), so your sentence is right. If you link the cover letter instead of patch 4/5, the citation will say so directly.

When I merge, I'll write the squash message myself, as I said last time. The title is fine.

@quatrixone

Copy link
Copy Markdown
Contributor Author

2026-11-09 update on the outstanding checks: the current recovery partition SHA-256 matches the recovery image in the verified 2026-10-29 19:01 snapshot. The historical order of that snapshot and the two diagnostic recovery writes is still unknown, so the notes state the limit rather than claiming restoration timing. The current userdata filesystem now has a verified private file-level archive; it is live, not a raw or atomic partition image.

On this handset, an attended hold of Power alone for more than 10 seconds restarted it. Volume Down + Power then entered fastboot; read-only variables confirmed product dipper and unlocked=yes. No serial is recorded. This is a restart test from the running system, not a simulated hard hang.

No PR #505 image has been booted and no partition has been written. The installed dirty kernel still has the root mounted read-write and logged failure to set the qcom watchdog timeout. The updated notes distinguish those old-kernel results from the pending PR-image probe.

@derekross

Copy link
Copy Markdown
Owner

Thank you for this round. Everything from my last list is in: the dates of the recovery writes and the logdump mirror, the plain statement that logdump wasn't in the backup, the 15-partition compare and the NV partitions' read-only state put back with "cause not established", the kernel links pointing at the right repository, and the cover letter cited for the Mi 8 test.

Two things in the new 2026-11-09 entry need another look.

The entry and your replies disagree. The entry (committed at 08:52 UTC) says recovery hasn't been compared, the userdata backup isn't complete, and a later SSH attempt got no reply, with neither ADB nor fastboot seeing the phone. Your replies on #499 and #505 a minute later say recovery matches the saved stock image, the userdata archive is complete and checked, and the phone answers over USB SSH. The notes are what stays, so please bring the entry up to date with the later results, and say:

What the phone was running. The entry shows the old first-light kernel with the root on /dev/sda21 read-write and /sys/block/sda/ro at 0. With the whole-disk node writable, the per-partition read-only flags don't protect the other partitions on that LUN (which ones are on sda is worth adding to the notes too), and our guard removes /dev/sda for that reason. Please add how that image was started (a fastboot boot, on which date, and the uptime when you checked), whether /dev/sda existed, and whether the logdump mirror from the diagnostic init was running. Until it ends, please use that session only for reads, as you've done, and please don't boot that image again. The next boot should be #505's image, after its gates.

One more small thing: please quote the exact journal line about setting the qcom_wdt timeout. #505's NFS-root note relies on RuntimeWatchdogSec=60s, so a failure there matters for that path.

And a nit: #221 is closed, so it can come out of the list of successor PRs.

When I merge, I'll write the squash message myself. The title is fine.

@derekross

Copy link
Copy Markdown
Owner

Thank you, this is a big step: the first boot of #505's own image, recorded with the checks the runbook asked for. /dev/sda removed, all 85 partitions read-only, the watchdog bound and fed, USB and SSH up, simpledrm on the panel, the PMIC keys, and the /proc/iomem lines. Every point from my last comment is answered too. Saying plainly what the surviving record doesn't hold (how the old image was started, the uptime, the mirror, the journal line) is the right way to write it.

Three things before I merge:

1. Please keep each boot. The commits show three RAM boots today: 93484a91… (no PMIC keys), f2f5cb2e… (qcom-pon built in, where Volume Down and Volume Up produced events), and e2cdb765… (rebuilt on kernel #220 b288f8e after #219 merged). The file now names only the last hash, and the key results come from an image whose hash isn't in it anymore. Could you list them, one line per boot: the time, the boot.img sha256, the omarchy-mobile and kernel commits it was built from, what changed from the boot before, and what you checked? #505's status line has the same issue: it credits the Volume Down result to the current image.

2. What recovery holds. Thank you for the compare. Since the 10-29 19:01 snapshot may have been taken after the two diagnostic writes, a match tells us recovery hasn't changed since then, but not whether it holds stock's recovery or the diagnostic image. I'm also wondering how the old first-light kernel was running this morning: boot holds Android (your Power-hold test restarted into it), and the notes have no fastboot boot for it. Was it started from recovery? Either way, one read-only check on the saved copy answers it, off the phone: unpack its header and look at the kernel version string (a 4.9 MIUI kernel or 7.3.0-rc6-omamobile…), or compare it with recovery.img from Xiaomi's fastboot package for the MIUI build on your phone. Until we know, please don't use reboot-to recovery from the RAM image.

3. One wording fix. "FTS5 PR #219 is now merged upstream": #219 was merged into this project's kernel branch (omamobile) today, not into upstream Linux. "merged into linux-omarchy-mobile" says it right. #505's status line says the same thing.

And one suggestion for the next RAM boot, not a condition: you've tried the Power-hold restart from Android. On another phone, Power alone did nothing under our kernel, so it's worth one attended try while the RAM image is running and healthy, with Volume Down + Power ready for fastboot.

When I merge, I'll write the squash message myself. The title is fine.

@quatrixone

Copy link
Copy Markdown
Contributor Author

I updated the notes in 281ce5bdd118e8478cb0b67ab9c37f2d060c9746.

  • The current recovery partition matches the saved 10-29 image and its decompressed kernel identifies as 4.9.337-perf-g0b3e38ef6c40, so it contains the stock recovery image rather than the 7.3 diagnostic kernel. The snapshot/write order is still unknown. The 10-29 first-light image is recorded as a fastboot boot RAM image; the separate 08:53 old-kernel observation's launch method is still not in the retained record.
  • The omamobile-audio 0.5.6: earpiece mode, a Voice Call verb, and no stored Voice Call profile at startup #219 wording now says the driver merged into linux-omarchy-mobile.
  • The EA8074 panel test is recorded: kernel 7.3.0-rc6-omamobile-push1, image SHA-256 0f0ec9d62ebe57bef9ae8a4804bb23728f624daab824a0bf7dee18f63c2a3d59, visible console, brightness 0→511, and DPMS Off/On. It also records the missing DSI vdda supply and unavailable GPU firmware.
  • I recorded the three images you identified. Only e2cdb765… has a retained full hash and manifest; the first two hashes, source commits and boot times are absent from the retained artifacts. The third manifest records its build time, not its boot time or Git source commit, so I left those fields explicitly unknown rather than infer them.

The exact stock recovery package is not in the retained local artifacts, but the recovery header check matches the 4.9 MIUI kernel you suggested.

@quatrixone

Copy link
Copy Markdown
Contributor Author

Updated the device notes: holding Power for 10 seconds restarted the phone back into Android under both the installed system and the Omarchy Mobile RAM kernel. The earlier “not tested under our kernel” qualification has been removed.

@quatrixone

Copy link
Copy Markdown
Contributor Author

The notes now include the partial NFS-root run from #503: image hash, NFS/systemd handover, persistent guard log confirmations for 85 partitions and six UFS disks, generic Arch ARM root limitation, and the missing direct sysfs capture/post-#484 check.

@derekross

Copy link
Copy Markdown
Owner

Thank you for all of this. The record now answers what I asked: recovery doesn't hold the diagnostic image, the 10-29 first light was a fastboot boot, and the Power hold works under our kernel too. On top of that the day brought the panel test, the full backup and the first NFS hand-over, each written with its limits. Leaving fields "unknown" where the artifacts don't say is exactly right.

Two fixes to the record, and then I'll merge. Neither needs another boot.

1. The list of boots, and the two hashes aren't lost. You wrote that the first two image hashes are no longer in your artifacts. They are in this branch's own history:

  • 599edd07: 93484a91103c294f6c73799ae6cb7161019129b9f89ab331222ae04f10ecf3e1 (the first image, before the PMIC keys)
  • 47052cc4: f2f5cb2e3da8fe992d366dba36e059c0d9106cf14a7d1c8872c61ee144a4474b (qcom-pon built in, where Volume Down and Volume Up produced events)

Could you add one table of the day's RAM sessions to the notes? By my count there are seven now: those two, e2cdb765… (rebuilt on kernel #220 b288f8e), the panel image 0f0ec9d6…, the session the backup was read from, the session of the Power-hold test, and the NFS image 2244d74a…. One row each: the image's sha256, what it was built from as far as you know, what changed, what you checked, and "unknown" where it is. For the backup and the Power-hold test, the image's hash or "same boot as the panel test" is enough. For the panel image, please name the kernel tree (it needs the EA8074 driver from #218).

2. "Stock". Your 10-29 section says the phone runs a custom Android 14 ROM with Magisk on a 4.9.337 kernel, not MIUI. So the 4.9.337-perf-g0b3e38ef6c40 string shows that recovery holds a recovery built on the installed ROM's kernel, not Xiaomi's recovery. That still answers my question, and my own suggestion was off: I asked you to look for "a 4.9 MIUI kernel" on a phone that, by your notes, has no MIUI. Could you word it as "the installed ROM's recovery", and add one more read from the saved copies: the kernel string in the backup's boot.img? If it carries the same g0b3e38ef6c40, we know recovery and boot come from the same build.

Two smaller things for the same push:

  • The backup and the last Android boot. The backup was taken before the Power-hold test, which restarted into Android, and Android rewrites the modem's partitions. Nothing is wrong with the backup; please add that the checksums get taken again after the last Android boot and before any flash.
  • 85, 84 and 92. The 10-29 section counts 92 names under /dev/block/by-name/, our kernel shows 85 partitions, and the backup has 84. What are the other seven names?

I'll write the squash message myself; the title will be device/dipper: stage-0 facts, first light and the 2026-11-09 RAM boots.

@quatrixone

quatrixone commented Nov 10, 2026 •

Copy link
Copy Markdown
Contributor Author

I rechecked the exact local image with SHA-256 e2cdb765de21a5869a1db314c19f1fddb668219a51d3c9ce81a1a083536d711e over the USB serial shell. It reports 7.3.0-rc6-omamobile-push1, but /dev/sda exists with /sys/block/sda/ro=0; sda21 is ro=1. No UFS write was attempted. I had incorrectly labeled this file the PR #505 image; I corrected the notes in commit cac9ad7e2714850df626cee51edf077f0d7ae573 on this branch. I cannot tie the earlier /dev/sda-absent result to another retained image, so please do not count it as verification of PR #505 until the exact build is identified and rerun. The phone is back in fastboot.

@quatrixone

Copy link
Copy Markdown
Contributor Author

Added the final exact-image hardware record in the latest PR head 453d9f49ae9e6adcc91b5515c3354f7d8b9c2cca. Image SHA-256: 0ee4f11d1973085c434e741a2e9dd73af5a9cd157296c25801ed5a92a6b9e9d9, RAM booted with #505 head 994944d7ecbacd69444221e6951153cb329a374e and kernel/DTS #220 head 055a4019998f22cc0e2a8afef7650274a6ce0657. All 85 UFS partition flags were read-only; sda sysfs was ro=0 with its whole-disk node absent, and sdb–sdf were read-only with nodes present. Watchdog feeder, simpledrm/fb0, USB/SSH, reserved TZ range, and return to fastboot were confirmed. No UFS write was attempted. Physical button events remain unverified on this exact image.

@derekross

Copy link
Copy Markdown
Owner

Thank you for the three pushes since my last comment, and above all for the correction in cac9ad7e. You booted e2cdb765… again, saw /dev/sda present and writable, said so here within minutes, took back the "#505 image" label and wrote "do not count it as verification". My earlier "the first boot of #505's own image, /dev/sda removed" rested on that label, so I take it back too. The record that replaces it is solid: 0ee4f11d… built from #505 at 994944d7 on kernel #220 055a4019, the NFS repeat a2d435a7… and the combined 540b5d31…, each with uname -r, the hash, and every ro value.

The four points from my last comment are still open. I think they crossed with your pushes, so here they are again, with what changed since. None needs a boot.

1. One table of the RAM sessions. It matters more now than yesterday, because the file's paragraphs switch between images and one image turned out to be something else. By my reading there are about eleven sessions:

One row each: date, sha256, what it was built from as far as you know, the state of /dev/sda and the partitions, what you checked, and "unknown" where it is. For the backup and the Power-hold test, "same boot as …" or the image's hash is enough.

2. What e2cdb765… is. You still have the file, so this is an offline read: unpack its ramdisk and compare its init with initramfs/init. Does it remove the whole-disk node at all? Is it the October diagnostic init? One line with the answer, or "unknown", goes in the table. Please also fix the sentence it left behind in the 10-29 section, "The 2026-11-09 PR #505 RAM boot checked every UFS partition flag": by your own correction, no 11-09 boot can be tied to #505's image.

3. "Stock". The file still says "the stock recovery image", "stock 4.9.337-perf…", "stock snapshot" and "stock partition checksums". Your 10-29 section says the phone runs a custom Android 14 ROM with Magisk, so these are "the installed ROM's recovery" and "the backup of the installed system". One more offline read would tie it together: the kernel string in the backup's boot.img. If it carries the same g0b3e38ef6c40, recovery and boot come from the same build.

4. Two short additions.

  • The backup was taken before the Power-hold test, which restarted into Android, and Android rewrites the modem's partitions. Please add that the checksums are taken again after the last Android boot and before any flash.
  • 92 names under /dev/block/by-name/ on 10-29, 85 partitions under our kernel, 84 in the backup plus userdata. What are the other seven names?

Small things for the same push: "Hence BOOT_METHOD=android-v1-appended-dtb" isn't a method in the tree (#505's descriptor says ram-only), so "header v1 with the DTB appended" is enough; and the two 11-10 sections would read better above the 11-09 one, newest first.

I'll write the squash message myself; the title will be device/dipper: stage-0 facts, first light and the 2026-11-09 and 11-10 RAM boots.

@quatrixone

quatrixone commented Nov 11, 2026 •

Copy link
Copy Markdown
Contributor Author

Updated the notes in commit b0c5132. The change adds the newest-first 11-session RAM boot table and corrects the 10-29 paragraph, recovery/backup wording, header-v1 description, and pre-flash checksum reminder. I inspected e2cdb765… offline: its ramdisk init differs from the shared init and contains neither the October logdump mirror nor whole-disk-node removal. It is not the October diagnostic init and does not verify #505. The final #505 image verification is recorded separately. I left explicit unknowns where no retained artifact provides the answer: the backup boot.img kernel string, the other seven /dev/block/by-name names, and the exact image/build for the Power-hold session. tools/pr-check origin/main and git diff --check pass.

@derekross

Copy link
Copy Markdown
Owner

Thank you, @quatrixone. b0c5132 does what I asked: the table of the eleven RAM sessions is there, newest first, with the two early hashes back in; the e2cdb765… image is now identified by its init (not the October diagnostic init, not #505's image); the wording says "the installed ROM's recovery" throughout; and the reminder to take the checksums again after the last Android boot is in. Leaving the boot.img kernel string, the seven other names and the Power-hold session's image as "unknown" is right where the artifacts don't say.

I'm merging this. One thing for your next push to the notes, not for this one: the table's last two rows say /dev/sda was absent on 93484a91… and f2f5cb2e…, which is what your commits recorded on the day, but a paragraph further down still says the earlier removal result "cannot be tied" to an image. I think the table is right and the paragraph is left over from the e2cdb765… correction; a sentence saying so would close it. If you later get at the backup's boot.img, its kernel string would be good to add too.

@derekross
derekross merged commit a595a85 into derekross:main Nov 11, 2026
2 checks passed
robertzas pushed a commit to robertzas/omarchy-mobile that referenced this pull request Nov 14, 2026
…0 RAM boots (derekross#330)

The Xiaomi Mi 8's installed-system facts read on 2026-10-29 (SDM845, single boot partition, EA8074 panel, ST FTS5 touch), the first light on an unpublished kernel with its limits, and a table of the eleven RAM-only sessions of 2026-11-09 and 11-10 with each image's sha256, source and storage-guard state. Notes only; no captures or identifiers, and nothing here is a recovery procedure yet.
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