Skip to content

Run a virtual iPhone on an Apple Silicon Mac.

Swift

research: record unresolved iOS 16 first-frame failure

黑曜猫d06318be7a66
diff --git a/Research/0_binary_patch_comparison.md b/Research/0_binary_patch_comparison.md
index b1681465..3bd0a839 100644
--- a/Research/0_binary_patch_comparison.md
+++ b/Research/0_binary_patch_comparison.md
@@ -6214,3 +6214,32 @@ been archived outside the source tree and removed from the final declarations,
presets and installer after successful opt-out restoration. Their unrelated
encoder/parser additions are also reverted. Only measured startup compatibility
remains in the source tree. No iOS 16 desktop display success is claimed yet.
+
+After host unlock, the running window is still black/0 fps. A native notify probe
+reports SpringBoard finishedstartup=31. Terminating the hung PreBoard temporarily
+allows SBSUndimScreen to change hasBlankedScreen from 1 to 0, but no frame appears.
+A native framebuffer probe successfully opens the real main display, reads
+1179x2556 and a nonzero IOConnect port. On 20G81 the public SwapEnd dispatch pointer
+is at connection+0x730 (not the later layout's +0x7C8); its value resolves exactly
+to _kern_SwapEnd at unslid 0x1B21CED04. Therefore a force-kern patch is not indicated.
+Temporary IOConnectCallStructMethod return tracing in the selected backboard
+library will measure method-5 submission rather than assuming a routing failure.
+
+No method-5 trace was observed through the temporary interposer; it is removed,
+since this alone cannot distinguish absent swaps from cached binding behavior.
+Backboard's native BKSystemShellSentinel has a separate in-memory completion gate.
+The earlier no-wait check-in did not set it. Static 20G81 analysis identifies
+-dataMigratorDidComplete (0x100066834, v16@0:8), which notifies local observers,
+sets _lock_migratorDidComplete at +0x50 and updates shell state; -unblockSystemApp
+(0x1000665E0, B16@0:8) locks and clears the separate +0x51 block through the native
+helper at 0x100065134. The selected lab check-in now invokes these native methods
+once, before entering its original non-waiting check-in, with both signatures
+checked. No keybag is unlocked and no migration is actually performed. This is
+an explicit fresh-lab shell-state bypass, not evidence of data migration success.
+
+Result of the local completion/unblock experiment: still black/0 fps. The native
+main IOMFB connection and mode enumeration work, but no visible desktop or guest
+screenshot is obtained. The completion/unblock code has therefore been reverted;
+no such fake completion is retained in the final source. The missing first frame
+is unresolved. iOS 16 remains an unsuccessful desktop bring-up, despite a running
+SpringBoard, and no minimum-version compatibility claim is added.
diff --git a/Research/Guest/ios16_bringup.md b/Research/Guest/ios16_bringup.md
index b150555f..32bc85f8 100644
--- a/Research/Guest/ios16_bringup.md
+++ b/Research/Guest/ios16_bringup.md
@@ -5,8 +5,8 @@ libxpc, backboardd and vphoned run. SpringBoard now reaches its main run loop;
its unavailable media controllers, widget fetch, flashlight and three media
Control Center modules have been bypassed. The latest main-thread sample is
idle in CFRunLoopRunSpecific/UIApplicationMain. Desktop display has not been
-verified: the computer-use tool reports the host Mac is locked, and the user
-has been asked to unlock it. The actual Preboot kernel contains the measured
+verified successfully: after the user unlocked the host, the VZ window remains
+black at 0 fps, and guest screenshot capture fails. The actual Preboot kernel contains the measured
syscall-filter relaxation. The original media executable is restored, and both
temporary dyld close patches are undone and removed from the final source.
@@ -89,7 +89,7 @@ The native EXC_CRASH observer captured the media failure before inserted-library
constructors execute: abort <- cc_abort <- ccrng_getentropy_generate <-
ccrng_process_init <- arc4random_buf <- libobjc initialization. Independent
getentropy/arc4random probes pass as root and mobile, including a probe signed
-with the media service's retained entitlements. The current investigation reads
+with the media service's retained entitlements. The investigation read
the faulting thread's errno through THREAD_IDENTIFIER_INFO and the measured
20G81 __error TLS layout. No random bytes are replaced or fabricated.
@@ -98,9 +98,10 @@ the faulting thread's errno through THREAD_IDENTIFIER_INFO and the measured
The active lab is `/Users/Shared/VPhoneLab16/iPhone-14-Pro-iOS16-Lab`.
The untouched external restore copy was moved out of the machine list to
`/Volumes/ZHITAI/Data/VPhone/Library/lab-backups/iPhone-14-Pro-iOS16-Lab-restored-original`.
-Use the internal library root explicitly. The current running bundle is
-2.15.0-local.eda73803; the cleaned build has not yet been deployed because
-visual verification of the running guest awaits host unlock.
+Use the internal library root explicitly. After the unsuccessful visual tests,
+the lab is stopped and bound to the cleaned 2.15.0-local.cb7e2dff bundle, with
+the speculative sentinel-completion change reverted. Its six selected startup
+patches and firmware inputs are retained for further diagnosis.
The standard preset has six explicit allows: registered Mach-port bridge,
early cache lookup, existing syscall-filter relaxation, ephemeral keybag init,
@@ -114,3 +115,29 @@ tests pass after removing discarded experiments; the native Mach-port test
passes register/lookup round-trip, empty slots and count bounds. Earlier
full-cache apply/idempotence/undo reproduced the pristine affected chunk.
The iOS 17 PR branch remains unchanged and clean. No iOS 16 work was pushed.
+
+## Result after host unlock
+
+The iOS 16.6.1 desktop has NOT been brought up. The main SpringBoard thread is
+idle in its normal event loop and finishedstartup reports its PID, but the VZ
+view stays black/0 fps, including after Home and Power events. A native screen
+wake call initially leaves hasBlankedScreen=1. Ending the hung PreBoard process
+once lets that state become 0, still without a visible frame; PreBoard respawns.
+
+The main IOMFB connection opens successfully, reports 1179x2556, has a real kernel
+port and selects _kern_SwapEnd. CoreAnimation reports a 120 Hz main display,
+which proves mode enumeration only. The driver logs IOMFB_SWAP_SUBMIT_LOST with
+zero swap IDs. The actual cause of missing presentation is not established.
+The temporary IOConnect return trace observed no calls and was removed; shared
+cache binding behavior makes that negative observation inconclusive.
+
+A further local sentinel-completion/unblock experiment did not produce a frame
+and has been reverted. No permanent keybag, passcode or migration-complete
+changes were applied. The final source retains the earlier measured startup
+work, with no temporary diagnostics or speculative media/dyld patches.
+
+Next research should measure the render-server display activation state and
+actual method-5 submission/16-to-26.1 payload semantics before proposing another
+patch. iOS 17.7 remains the lowest version verified to display a desktop in this
+session; iOS 15 was not tested. This result does not establish that iOS 16 can
+never work.