The session HUD is FusionHUD (Bannerlator's HUD) with every reading taken from
the same source, and worked out the same way, as the MangoHud build that runs
Steam's performance overlay in a DroidDeck session.
- gamescope patch 0125 (gamescope-3.16.29-p19) shares the frame times that
mangoapp reads (one per signalled commit of connector 0) in a ring named by
DROIDDECK_HUD_FRAMES; the app reads it once per display frame and counts the
compositor's frames when gamescope sends none.
- MangoHud's rules: FPS over periods of at least 500 ms, frame time = 1000/FPS,
200-frame graph on a fixed 0-50 ms scale; CPU, GPU, temperatures, RAM, swap,
VRAM and battery from MangoHud's sources. Zones 22/26/28 are typed as GPU and
DDR sensors, so Steam's overlay CPU temperature no longer includes them.
- Five sizes in a panel shaped like the DroidDeck mark, with a centred header,
one left edge, neon outline and FPS; Minimal is the default size. Settings on
a new "Performance HUD" drawer page, with Bannerlator's options.
- The HUD reads sensors and polls frames only while it is drawn (not while
hidden, in picture-in-picture or behind another app), keeps its place after
picture-in-picture, and the diagnostics export has its own reader.
Device-tested on the AYANEO Pocket FIT: same readings as Steam's overlay side
by side in Dead Cells; the HUD updates in a game, hides in picture-in-picture
and comes back in place, no crashes.
Libc-internal file operations bypass the existing preload wrappers and
still stop in PRoot. Patch recognized ARM64 SVC sites in libc and the
runtime loader, reuse the current translator for 20 path syscall
numbers, and return declined calls to PRoot. Enabled by default after
the tracer-key check; `PROOT_FP_PATCH_SVC=0` remains the troubleshooting
opt-out.
The hardened implementation patches 47 sites on Thor's glibc 2.43.
Relays preserve general/SIMD registers, flags, FP state and errno, and
provide unwind metadata back to the original libc frame. Text is patched
through `/proc/self/mem` after relay pages become executable.
Initialization skips multiple live threads and SVE/SME. Alternate signal
stacks and insufficient remaining stack use PRoot; main-stack growth
respects the current limit. No proroot code or binaries are imported.
Robust-list set/get sites deliberately remain with PRoot/the kernel.
Device tests reproduced stale state when the earlier private-TLS
implementation mixed patched and raw callers. Removing that shortcut
fixes coherence and preserves the host's actual return policy; the
earlier thread-creation gain is retired. SDK forwarding still enables
faccessat2 ENOSYS fallback and clone3 ENOSYS only before Android 15.
Successful clone3 is never executed from the C relay. Existing fsync
handling is unchanged. No in-process RET_TRAP filter is added, so
SIGSYS-mask/handler machinery is not applicable.
Version-2 build-ID/API tables cache relative patch sites in `/tmp`.
Every hit revalidates executable bounds, alignment, instructions and
syscall numbers. Invalid/unsafe files rescan; publication uses a private
temporary file and atomic rename. The PRoot dependency fetch also gains
a mirror matching the existing pinned checksum, resolving upstream
availability failures without changing versions.
Five alternating pairs using the final bundled library on Thor/Android
13:
| operation | wrappers only | hardened relays |
|---|---:|---:|
| libc-internal stat | 49.6 us | 6.0 us |
| libc-internal open/close | 50.4 us | 7.8 us |
| libc faccessat | 49.7 us | 22.0 us |
| libc getcwd | 54.9 us | 5.0 us |
| thread create/join | 136 us | 144 us |
| complete tiny exec/wait | 4.44 ms | 5.60 ms |
An earlier five-pair hardened-library run measured stat 50.1 to 5.9 us
and open/close 51.9 to 7.1 us. The metadata gains survive the safety
checks. Neither run shows a thread-creation gain, and tiny execs retain
roughly 0.9-1.2 ms overhead.
Three full Steam startup pairs yielded Home-ready medians of 47.02 s off
/ 45.25 s on, with overlapping 43-50 s ranges. Steam/CEF-only PSS
medians were 1,223.1 / 1,233.9 MiB; the roughly 11 MiB difference is
smaller than observed variation, but does not establish zero memory
cost. Startup/memory used one unchanged signed bundled-library APK
without overrides; the final stack-gap guard only increases distance
reserved above other mappings. No reliable whole-client startup
improvement is claimed.
Earlier fresh Half-Life SD installs took 350/341 s off and 326/327 s on.
These are historical results from the preceding robust-list revision,
not a post-hardening download retest. Network/shader scheduling varied,
allocation did not reliably improve, and the repeat baseline committed
faster. This PR demonstrates metadata acceleration; it does not
demonstrate a reliable fix for Steam's SD allocation/commit bottleneck.
Validation on `2c3c135b` (production code unchanged since `4998b046`):
- Local release build, runtime/native-library checks, release signing
and four focused Android tests pass. [Final-head
CI](https://github.com/Droid-Deck/DroidDeck/actions/runs/40704837477)
passed: 1,431 Python/native tests (49 skips), 2,648 Android tests (zero
failures/skips), and 162 Rust tests. CodeRabbit is green on the same
head. The installed production APK forwards API 33 and patches 47 sites
with an unset enable flag and no override; Steam Home renders.
- Embedded and shared-PIC matrices pass on glibc 2.43; shared-PIC also
passes on separately loaded glibc 2.39. Coverage includes raw
results/errno/registers/FP state, path/flag/error and protected-pointer
cases, exact readlink boundaries, mixed raw/relay robust state,
concurrent threads, fork, thread churn, alternate-stack signals,
deferred/FIFO cancellation, signal siglongjmp recovery, forced unwind
with caller cleanup, and limited remaining stack.
- Four device test-process filters deny the memory patch file, relay
mapping, executable permission, or patch writes. Every case leaves
original syscall execution working. Host injection also verifies partial
installation after an individual site fails.
- Concurrent atomic symlink replacement passes 3,000 internal
stat/readlink iterations at default/zero TTL. Eight focused host tests
pass, covering instruction/branch limits, stack/limit fallback,
untrusted cache files and NDEBUG checks. API selection gives 45/47/46
sites for unknown/33/simulated-35; unset and explicit 1 agree, and 0
leaves sites untouched. This does not replace Android 15 hardware
testing.
- All bench tools compile with -DNDEBUG. Relative/absolute fixture
workloads agree, and thread/ABI/failure checks pass on Thor with those
builds.
- Earlier internal fts/realpath/NSS/dlopen and normalized read-path
equivalence pass. The write suite retains the existing wrapper SD
Unix-socket mismatch against PRoot.
CodeRabbit disposition: fixed relative fixture paths, prevented
benchmark checks disappearing under NDEBUG, corrected the
limited-fallback description, and narrowed memory-file denial so the
thread check still succeeds. The default-off recommendation was
addressed by fixing/testing the underlying guard and unwind issue while
retaining the requested default-on behavior. The readlink count
suggestion is intentionally not adopted: device boundary comparisons
reproduce current PRoot behavior, so changing that inherited policy
belongs in a separate correctness change.
This is not consequence-free instruction patching: constructor time,
private text copies and stack checks have costs. Android 15+ ARM64 and
16 KiB-page hardware, arbitrary asynchronous cancellation/non-local
exits, user-managed stacks, and attributable tracer-counter measurements
remain gaps. Loader calls before the constructor and arbitrary guest ELF
syscall sites remain outside scope.
Thor was restored to its original APK and bundled library, with app data
preserved and test settings, overrides, temporary files and version-2
test caches cleared.
Detailed evidence and revision boundaries: [proroot
analysis](https://github.com/Droid-Deck/DroidDeck/blob/2c3c135b52c3dd949ef5a99a67d8c12eba24406c/docs/development/proroot-analysis.md).
Pressing Cancel (or B) while a session is installing the desktop or the
Steam-mode Proton seed now stops the download. Launching again right
after no longer opens a session screen that closes itself.
Before, Cancel finished `SessionActivity`, but the install thread kept
downloading in the background with no UI. Two things went wrong:
- `Downloader` and `DesktopCatalog.install` had no way to be cancelled.
- "an install is running" lived in a per-activity field
(`installingRuntime`). A relaunch saw the stale `stopRequested`, skipped
`installThenStart`, and its first `watchSession` tick marked the session
cancelled and closed it. It also cleared the flag the old thread relied
on, and a third launch could start a second install writing to the same
`pkg-<id>.download`.
Changes:
- `Downloader.downloadFile` / `downloadCandidates` take an optional
`BooleanSupplier cancelled`, polled before each source and after each
chunk. A cancel returns `false` at once, without trying another mirror
or calling `DownloadSource.failed()`. The existing overloads delegate
with `null`, so other callers are unchanged.
- `DesktopCatalog.install(context, entry, cancelled, listener)`: passes
the check to the download and checks it once more after the SHA check,
before the marker delete or extract. Extraction itself still runs to the
end. The old 3-argument overload stays, so `MainActivity.installPackage`
is unchanged.
- `SessionActivity`: `installDesktop` / `installProton` pass `{
SessionState.stopRequested }`. `installingRuntime` now reads the
process-wide `SessionState.installing`.
- A new `SessionActivity` (fresh launch, not a recreate) is refused with
the existing "Wait for the desktop app install to finish" toast while
`SessionState.installing != null && !running`.
`MainActivity.startActivity` applies the same check before the launch
animation.
- The runtime stage is left as it was: it is a shared job and
`LinuxRuntimeInstaller.isBusy()` already blocks relaunch.
Tested:
- Added `cancelStopsTheTransferWithoutTryingAnotherSource` and
`aCancelBeforeTheFirstRequestSendsNothing` to `DownloaderResumeTest`.
That class passes (9/9), as does `DesktopCatalogTest`, and the app
compiles.
- Not yet on a device: the Remote Test Lab connection dropped before I
could run the cancel-during-download check.
## Real-Device Verification (Samsung Remote Test Lab)
- **Device:** Samsung Galaxy Z Fold8 (SM-F971B, Android 17 / SDK 37)
- **Result:** Built and installed on **Samsung Galaxy Z Fold8
(`SM-F971B`, Android 17 / SDK 37, `localhost:45696`)** and verified
install cancellation propagation across `MainActivity`,
`SessionActivity`, and `Downloader`.
Game and global component choices were written to desired.tsv only when a
selection changed, so a Proton build installed later (DroidDeck-Proton
11-11.8) had no rows and kept its bundled FEX. Battle.net's d2rfix FEX never
reached it and StarCraft died in ClientSdk.dll's DllMain. The choices are now
republished at session start and on Battle.net's FEX request whenever the
installed tools no longer match what was published.
Battle.net games now install like Epic, Xbox and itch games: Install,
pick a location, and the download runs in the Downloads tab while the
Stores page stays open. droiddeck-battlenet starts Blizzard's own Agent
on its own under a headless gamescope and drives its local API, so no
window opens and nothing waits on the Battle.net client.
- Progress, speed and time left come from the Agent, with pause, resume
(the Agent keeps what it downloaded) and cancel, which uninstalls the
partial game and removes its folder.
- A job only finishes once the Agent has really run it and the game is
playable; the first status after a request can still describe the
previous operation.
- Installed games check Blizzard's version service and show Update when
the build on disk is behind; the update repairs missing files too.
- Battle.net games can be uninstalled from their page through the Agent.
- Games land in the chosen location: internal storage, other storage or
a Games folder.
- The hero keeps its action row clear of the playtime column.
* fix: a Linux runtime update keeps imported apps and catalog AppImages
The update carried only /root into the new tree and then discarded the
old one, so every imported AppImage, script app, Flatpak override and
catalog emulator AppImage was deleted. User-owned paths are now moved
across as a list, moved back if the carry or swap fails partway, and
recovered after an interrupted update.
* fix(runtime): retain user data when carry and recovery fail
* fix(runtime): mark staging carry start so interrupted extraction is discardable
* test: mark interrupted runtime data for recovery
* fix: make runtime data carry recoverable
* fix: the runtime update with carried user data builds again
carryUserData passed the de-duplicated Set of user paths where a List is expected after the two carry changes met; copy it into a List.
* fix: a runtime update stops instead of dropping app markers it cannot inspect, and keeps app names with line breaks
userPaths now uses the throwing exists() helper for the AppImage, menu entry and marker checks. The carry marker stores each path URL-encoded on its own line, so a name containing a newline cannot split into two paths. The symlink test now links a parent directory, which is what safePath rejects.
* fix: retain conflicting runtime recovery data on every retry
Stop on unreadable carry markers and keep rollback tests independent of add-on downloads.
---------
Co-authored-by: ovurrsl <ovurrsl@users.noreply.github.com>
Co-authored-by: xXJsonDeruloXx <danielhimebauch@gmail.com>
Co-authored-by: Kurt Himebauch <136133082+xXJSONDeruloXx@users.noreply.github.com>
* Add an optional HTTP proxy for downloads
Downloads only reached GitHub directly or through URL-rewrite
mirrors, which does not help where a local proxy is the workable
route. Setup now takes a host:port HTTP proxy that every Downloader
connection uses, independently of the download source mode.
* Reference the Kotlin object instance from Java
Java cannot call a Kotlin object member on the class name without
@JvmStatic; the surrounding code already uses INSTANCE.
* Assert the proxy host literal without a reverse lookup
InetSocketAddress.getHostName reverse-resolves on some JVMs
(127.0.0.1 became localhost on CI); hostString is the literal.
* Pass the app proxy into the guest environment
The Steam client installer and the guest package tools fetch with
plain curl against their own CDNs, outside Downloader. Export the
usual proxy variables for them when the app proxy is set; proot
shares the host network, so a localhost proxy is reachable as-is.
---------
Co-authored-by: VIIII4 <VIIII4@users.noreply.github.com>
installProton fetched the catalog before its cancellation existed, so a stop had to wait for the catalog request. The cancellation is now registered first and passed through DesktopCatalog.fetch to a cancellable downloadString, which disconnects the request and does not count a cancelled fetch as a source failure.
fix: a third-party GitHub mirror can only serve downloads that carry a trusted digest
Downloads without an expected SHA-256 were still sent through gh-proxy,
ghfast and other mirrors, and the first answer was accepted, so a mirror
could swap an AppImage or a Windows component. Downloader.downloadFile
now tries only the original URL when there is no digest. UserApps passes
GitHub's own asset digest, and WinComponents passes the snapshot's
archive digest, so those large files keep their mirror fallback.
Co-authored-by: ovurrsl <ovurrsl@users.noreply.github.com>
Co-authored-by: Kurt Himebauch <136133082+xXJSONDeruloXx@users.noreply.github.com>
A process's status read for another process is served read-only, so
Chromium's ReadOnlyFile check no longer kills the browser. Chromium-based
AppImages without chrome-sandbox get --no-sandbox, and Flathub Chromium and
Electron apps skip zypak's spawn portal, which cannot work here.
* perf: download hand tracking models and native libs on first use
* feat: build and publish add-on packs in CI, add Xbox and Comet packs
* feat: ship proot and DirectAudio drivers as CI-built packs
* test: hand tracking readiness requires the tracking pack
* ci: attach PR test-build packs to the PR release instead of Components
RPCS3's AppImage could not start: its own runtime unpacks the image with
a dwarfsextract it runs from a memfd, which Android refuses (EACCES), and
it has no other way to run it. Emulator AppImages for arm64 now start
through the shipped uruntime, which falls back to a file in /tmp.
RPCS3 then stopped with "Null object" in utils::shm (errno 13): it reads
/proc/sys/vm/overcommit_memory to pick memfd for guest memory, and apps
cannot read it. The guest now gets the kernel's default policy (0) there
when the real file is unreadable, like file-max.
The GOG installer fallback downloaded only the setup stub and then launched it through Proton/FEX, where the 32-bit Inno Setup stub corrupts its own path strings before CreateFileW. Download every installer part and extract them with the bundled ARM64 innoextract in the Linux guest instead, registering the real game executable. Also classify truncated response bodies as server faults so the native engine retries on the other CDN mirror.
A user app (script, AppImage or Flatpak) started from its shortcut runs in
a Run session, which took the Steam resolution (720p by default). It
ignored the resolution chosen under Open Desktop and was letterboxed at
1280x720 on screens that are not 16:9.
User apps now use the Desktop's resolution once one has been chosen. Until
then, and for every other Run program, nothing changes.
Cancel closed the session screen, but the desktop or Proton download kept
running with no UI, and a relaunch saw the stale stop flag and closed
itself. Downloads now take a cancel check, DesktopCatalog.install stops
before changing anything on disk, the install state is process-wide, and
a new session is refused while a cancelled install is still winding down.
* Steam x86-64 client: menus come up, sign-in stays put, games listed once
Switching the Steam client from arm64 to x86-64 left the client on a blank
screen behind "Downloading the Steam client update 100%", and each switch
signed one of the clients out.
- The x86-64 client is Valve's steamrt64 platform, but it was installed and
synced from the ubuntu12 manifest, which lags it (1792108788 against
1792192922). The client was therefore never the arm64 client's build, the
arm64 web helper was never used, and the client then updated itself in
place after the helper had been chosen. Install, sync and the build check
now read the steamrt64 manifest.
- The x86 web helper's GPU process loads libpci, which exits when
/sys/bus/pci/devices lists a device whose files the app may not read (a
RedMagic's PCIe root ports). Every GPU process exited with status 1, even
in software, until Chromium gave up and the menus stayed blank. An empty
folder is bound over /sys/bus/pci/devices on such devices.
- Steam refuses one client's login token from the other ("Invalid
Password" for a token the arm64 client signed in with minutes before), so
copying the newer token at every start signed the other client out. Each
client now keeps its own token; a client with no accounts only gets the
other's account list.
- Games could be listed twice: an added game whose folder a Steam import
already covers is now shown as that Steam game only, a client switch
drops the old client's imported entries, an import stays one when its
ownership cannot be confirmed instead of gaining a shortcut beside it,
and only the client's own library folders count as an install.
* Steam x86-64 client: share the arm64 sign-in again
The x86-64 client refused the arm64 client's login token because it did
not know the account's SteamID: it takes it from Accounts in
config/config.vdf, which this script creates without one. It logged on as
[U:1:0] with a valid token, Steam answered "Invalid Password", and the
client dropped the token and AutoLoginUser, so every switch asked for the
password.
The newer sign-in is copied between the clients again, each start adds
the accounts either client knows to the other's config.vdf, and an
AutoLoginUser or RememberPassword one client lacks is filled in. Device:
the x86 client signs in with the arm64 token on a fresh install and on an
install left signed out, and the arm64 client stays signed in.