Skip to content

Run a virtual iPhone on an Apple Silicon Mac.

Swift

Take the hv_vmm_present concealment out of the default patch selection

Lakrb30637a605ff
diff --git a/Research/0_binary_patch_comparison.md b/Research/0_binary_patch_comparison.md
index b07ee4ec..f0373945 100644
--- a/Research/0_binary_patch_comparison.md
+++ b/Research/0_binary_patch_comparison.md
@@ -15,8 +15,31 @@
> Two presets ship, prewritten, in
> `VPhone.bundle/Contents/Resources/patches_presets/`. Both name all nine sets and
> differ only in their selection: `standard` (the default, and what a VM gets
-> unless `--preset` says otherwise) blocks the two Frida Stalker relaxations;
-> `extended` blocks nothing. A VM records only the boxes its owner changed, in
+> unless `--preset` says otherwise) blocks the two Frida Stalker relaxations and the
+> two halves of the `hv_vmm_present` concealment; `extended` blocks nothing.
+>
+> **`hv_vmm_present` concealment is opt-in as of 2026-09-28, and it is not a
+> preference.** `kernelcache_exp.hv_vmm` (the kernel OID rename plus the
+> kernel-internal cstring mangle) and `hv_vmm_dsc` (the shared-cache mangle the JB
+> system installer runs) were made default-on by commit 228326d; on a freshly
+> restored 26.4 guest they are a brick. `bluetoothd` on 23E246 caches its
+> `sysctlbyname("kern.hv_vmm_present")` answer in a `dispatch_once`, gets ENOENT and
+> caches 0, and its chip-selection singleton then picks a transport from
+> `MGIsDeviceOneOfType` — nothing matches a virtual iPhone, the singleton stays
+> NULL, and it faults and crash-loops until launchd throttles it. `locationd` blocks
+> on a synchronous call to the throttled Bluetooth XPC service, the
+> `com.apple.locationd.migrator` datamigrator plugin hangs, and SpringBoard waits on
+> migration: black screen, no panic. The addresses and the full chain are in
+> `Research/Patches/hv_vmm_present_usermode_xrefs.md` (B.3 correction).
+>
+> Neither half is useful alone — the rename without the cache mangle breaks the
+> graphics and ML paths, the mangle without the rename does nothing — so they are
+> declared as a pair, `FirmwarePatchSetCatalog.hypervisorConcealmentPatches`, and a
+> catalogue test refuses a shipped preset that enables one without the other. Both
+> lost `bootEssential`, which they never were: a shipped preset that drops a
+> boot-essential patch warns on every run. The other former EXP patches are
+> untouched — DeviceTree identity and camera, `camera_dsc`, the watchdogd cache
+> patch, and the post-restore Preboot DeviceTree rewrite all stay on by default. A VM records only the boxes its owner changed, in
> `<vm>/PatchSelection.plist`, and `fw patch` writes what it resolved to
> `<vm>/PatchPlan.plist` for `cfw install` to reuse.
>
@@ -80,10 +103,11 @@
> repository. The native Swift JB install retains the base system patches,
> launchd jetsam guard, debugserver entitlement edit, iOS 27 Campo entitlement
> edit, GPU driver, mandatory vphoned, and the small vphone launchd hook described
-> below. The JB firmware pipeline now also runs the former EXP kernel OID and
-> DeviceTree identity/camera patches. The JB system installer runs the former
-> EXP DSC hypervisor and camera patches, watchdogd patch, and post-restore
-> Preboot DeviceTree rewrite. `SPOOF_BUILD` remains opt-in and updates the
+> below. The JB firmware pipeline also runs the former EXP DeviceTree
+> identity/camera patches. The JB system installer runs the former EXP DSC camera
+> patch, watchdogd patch, and post-restore Preboot DeviceTree rewrite. The former
+> EXP kernel OID rename and its DSC half are the exception: declared, but off in
+> `standard` since 2026-09-28 for the reason in the note above. `SPOOF_BUILD` remains opt-in and updates the
> rootfs and installed SystemOS SystemVersion.plist copies, plus the Preboot
> Cryptex copy when present. These are install-time integrations;
> successful patch dry-runs do not establish guest boot or Camera.app behavior.
diff --git a/Research/Patches/hv_vmm_present_usermode_xrefs.md b/Research/Patches/hv_vmm_present_usermode_xrefs.md
index 4bec5c1e..70d4456d 100644
--- a/Research/Patches/hv_vmm_present_usermode_xrefs.md
+++ b/Research/Patches/hv_vmm_present_usermode_xrefs.md
@@ -8,9 +8,15 @@ This document enumerates every place in user space that reads
caller appears to do with the result, and what is reasonable to expect from
patching that specific reader instead of the kernel.
-## Current shipping design — blacklist-flip + kernel rename
+## The design — blacklist-flip + kernel rename (opt-in since 2026-09-28)
-The current patch pipeline takes a **kernel-rename + user-mode-blacklist**
+**Not applied by default any more.** Both halves are blocked by the `standard`
+preset and reachable only through `--preset extended` or two per-VM checkmarks,
+because on a freshly restored 26.4 guest they mean a permanent black screen. The
+reason is `bluetoothd`, and it is written out under B.3 below. What follows
+describes the patch as it works when it *is* selected.
+
+The patch pipeline takes a **kernel-rename + user-mode-blacklist**
approach. Summary:
* The Swift kernel patch `patch_hv_vmm_rename` (JB-26, Group B) renames
@@ -37,6 +43,13 @@ approach. Summary:
rename in place, those 6 rootfs binaries fall into the "unpatched → ENOENT
→ cache 0" bucket automatically.
+ **This is the step that bricks a 26.4 guest.** "Cache 0" is not harmless for
+ every rootfs binary: on 23E246 `bluetoothd` reads the sysctl (it did not on
+ 23B85), and a cached 0 sends it down a device-type table that has no entry for a
+ virtual iPhone, so it crash-loops. The mangle covered only the 37 cstrings inside
+ the shared cache; standalone executables were never touched. See the correction
+ under B.3 for the full chain from there to a black screen.
+
The earlier whitelist-only design (no kernel patch, mangle a chosen
subset to ENOENT) is preserved for reference in the section
"Pre-blacklist whitelist enumeration" below, since the per-caller
@@ -221,7 +234,7 @@ encoding of the address (raw, image-relative, or 36-bit chained-fixup
shape) appears in any other section either.
```
-usr/sbin/bluetoothd (verified: 0 ADRPs into the page; 0 byte-pointer matches)
+usr/sbin/bluetoothd 23B85 ONLY — see the correction below (verified: 0 ADRPs into the page; 0 byte-pointer matches)
System/Library/Frameworks/MediaToolbox.framework/MediaToolbox
System/Library/Frameworks/ManagedAppDistribution.framework/Support/managedappdistributiond
System/Library/PrivateFrameworks/ApplePushService.framework/apsd
@@ -232,6 +245,39 @@ System/Library/PrivateFrameworks/Recon3D.framework/Recon3D
System/Library/PrivateFrameworks/VFX.framework/VFX
```
+> **Correction (2026-09-28): B.3 is per build, and `bluetoothd` left it.** This
+> inventory is 26.1 (23B85). On **26.4 (23E246)** `usr/sbin/bluetoothd`
+> (UUID `58F4BA7F-2E62-3E04-8ABB-03D06FD7A591`) reads the sysctl for real: the
+> `sysctlbyname("kern.hv_vmm_present", …)` is at `0x1004ac7b4`, inside a
+> `dispatch_once` at `0x1004ac798` whose flag is `0x100b51bc0`, so the answer is
+> cached for the life of the process. Its Bluetooth chip-selection singleton
+> (`0x10042fa40`) uses that flag to choose the virtual transport — the 6000 path.
+>
+> With the OID renamed the lookup returns ENOENT, 0 is cached, and the singleton
+> falls through to the `MGIsDeviceOneOfType` table instead. Nothing in the table
+> matches a virtual iPhone, the transport singleton at `0x100b50bd0` stays NULL,
+> and `bluetoothd` faults on `ldr w23, [x0, #0x31c]` at `+0x402864`
+> (EXC_BAD_ACCESS at `0x31c`). It crash-loops until launchd throttles
+> `com.apple.bluetoothd`.
+>
+> That is why the guest never draws: `locationd`'s `CLSeparationAlertsServiceSilo`
+> makes a **synchronous** call to `com.apple.server.bluetooth.general.xpc`, which is
+> now throttled, so it blocks; the `com.apple.locationd.migrator` datamigrator
+> plugin then hangs for 65+ minutes (stackshots under the guest's CrashReporter as
+> `stacks+com.apple.datamigrator-*.ips`); SpringBoard waits for migration to finish.
+> A black screen with no panic and no log line naming the cause.
+>
+> Consequence for the shipping design: both halves of the concealment — the kernel
+> OID rename and the shared-cache mangle — are **off in the `standard` preset** as of
+> this date and are opt-in only (`--preset extended`, or the two checkmarks
+> together). A freshly restored 26.4 VM does not survive them. The pair is
+> `kernelcache_exp.hv_vmm` + `hv_vmm_dsc`; see
+> `FirmwarePatchSetCatalog.hypervisorConcealmentPatches`.
+>
+> What this does not say: nothing here re-verifies B.3 for the other eight binaries
+> on 23E246, and nothing here says the 23B85 result was wrong for 23B85. Treat every
+> "unreferenced" line as a claim about the build it was measured on.
+
#### B.4 Adjacent‑string anchor (REFERENCED but not a sysctl call)
```
diff --git a/VPhoneExecutable/VPhoneCommand/FirmwarePatcher/PatchSets/FirmwareGuestIdentityPatchSet.swift b/VPhoneExecutable/VPhoneCommand/FirmwarePatcher/PatchSets/FirmwareGuestIdentityPatchSet.swift
index f52cebb2..9478e60e 100644
--- a/VPhoneExecutable/VPhoneCommand/FirmwarePatcher/PatchSets/FirmwareGuestIdentityPatchSet.swift
+++ b/VPhoneExecutable/VPhoneCommand/FirmwarePatcher/PatchSets/FirmwareGuestIdentityPatchSet.swift
@@ -4,6 +4,12 @@
// plus the camera symbols the virtual camera is published through. These pair with
// `com.vphone.patchset.kernel.hypervisor` and the device tree's camera nodes: on
// their own, each half leaves the guest inconsistent with itself.
+//
+// `hv_vmm_dsc` is therefore off in `standard` for exactly the reason its kernel
+// half is — see `FirmwareKernelHypervisorPatchSet` for what a 26.4 guest does when
+// the OID is renamed. The watchdogd patch beside it is not part of that pair and
+// stays on: it stops watchdogd acting on a cached positive answer, which is
+// harmless when the sysctl still reads true.
import Foundation
import VPhonePatchKit
@@ -21,10 +27,10 @@ public enum FirmwareGuestIdentityPatchSet {
title: "Shared cache hypervisor strings",
summary: """
Mangles the hv_vmm_present references in the shared cache, so a userland check \
- finds nothing where the kernel set renamed the sysctl.
+ finds nothing where the kernel set renamed the sysctl. Off by default, and \
+ pointless without kernelcache_exp.hv_vmm: enable the two together or neither.
""",
target: .dyldSharedCache,
- bootEssential: true,
),
VPhonePatchDeclaration(
identifier: "watchdogd.hv_vmm_cache",
diff --git a/VPhoneExecutable/VPhoneCommand/FirmwarePatcher/PatchSets/FirmwareKernelHypervisorPatchSet.swift b/VPhoneExecutable/VPhoneCommand/FirmwarePatcher/PatchSets/FirmwareKernelHypervisorPatchSet.swift
index 1e20f923..85885171 100644
--- a/VPhoneExecutable/VPhoneCommand/FirmwarePatcher/PatchSets/FirmwareKernelHypervisorPatchSet.swift
+++ b/VPhoneExecutable/VPhoneCommand/FirmwarePatcher/PatchSets/FirmwareKernelHypervisorPatchSet.swift
@@ -1,13 +1,23 @@
// FirmwareKernelHypervisorPatchSet.swift — Manifest for the hv_vmm rename.
//
// The guest runs under a hypervisor and the research kernel says so through an
-// `hv_vmm_present` sysctl. Renaming the OID and mangling its internal caller
-// hides that from the userland checks that would otherwise refuse to run, which
-// is what lets stock apps start at all.
+// `hv_vmm_present` sysctl. Renaming the OID and mangling its internal caller hides
+// that from userland checks that would otherwise refuse to run.
+//
+// **Off in `standard`, and it is not a preference.** On a freshly restored 26.4
+// guest the rename bricks the boot: `bluetoothd` caches its
+// `sysctlbyname("kern.hv_vmm_present")` answer in a `dispatch_once`, gets ENOENT,
+// caches 0, and picks its Bluetooth transport from `MGIsDeviceOneOfType` instead —
+// nothing matches, its transport singleton stays NULL, and it faults and
+// crash-loops until launchd throttles `com.apple.bluetoothd`. `locationd` then
+// blocks on a synchronous call to the throttled Bluetooth XPC service, the
+// `com.apple.locationd.migrator` data-migrator plugin hangs for over an hour, and
+// SpringBoard waits on migration — black screen, no panic. `--preset extended` or a
+// per-VM checkmark turns it back on, together with `hv_vmm_dsc`.
//
// Its own set because it is the one kernel change that is about hiding the
-// hypervisor rather than about jailbreaking, and a preset studying VM detection
-// wants it off without giving up the rest.
+// hypervisor rather than about jailbreaking, so a preset can take it or leave it
+// without giving up the rest.
import Foundation
import VPhonePatchKit
@@ -25,10 +35,11 @@ public enum FirmwareKernelHypervisorPatchSet {
title: "hv_vmm_present sysctl",
summary: """
Renames the hv_vmm_present OID and mangles its internal caller, so a userland \
- check for the hypervisor finds nothing.
+ check for the hypervisor finds nothing. Off by default: it leaves a freshly \
+ restored 26.4 guest on a black screen. Enable it together with hv_vmm_dsc, \
+ never alone.
""",
target: .firmware(.kernelcache),
- bootEssential: true,
),
],
requires: ["vphone.kernel.base"],
diff --git a/VPhoneExecutable/VPhoneCommand/FirmwarePatcher/PatchSets/FirmwarePatchSetCatalog.swift b/VPhoneExecutable/VPhoneCommand/FirmwarePatcher/PatchSets/FirmwarePatchSetCatalog.swift
index ef80f414..36a6f226 100644
--- a/VPhoneExecutable/VPhoneCommand/FirmwarePatcher/PatchSets/FirmwarePatchSetCatalog.swift
+++ b/VPhoneExecutable/VPhoneCommand/FirmwarePatcher/PatchSets/FirmwarePatchSetCatalog.swift
@@ -46,9 +46,30 @@ public enum FirmwarePatchSetCatalog {
/// The Frida relaxations widen what any process in the guest may do and
/// nothing needs them to boot, so `standard` leaves them off. They stay in the
/// catalogue, version-gated like everything else, and a VM can check them on.
- public static let manualOnlyPatches: Set<String> = Set(
- FirmwareKernelFridaPatchSet.manifest.patches.map(\.identifier),
- )
+ ///
+ /// The hypervisor concealment pair is off for a harder reason: it stops a
+ /// freshly restored 26.4 guest booting at all. See
+ /// ``hypervisorConcealmentPatches``.
+ public static let manualOnlyPatches: Set<String> =
+ Set(FirmwareKernelFridaPatchSet.manifest.patches.map(\.identifier))
+ .union(hypervisorConcealmentPatches)
+
+ /// The two halves of the `hv_vmm_present` concealment, which are one patch in
+ /// everything but name and have to be selected together.
+ ///
+ /// Renaming the OID without mangling the shared cache breaks the graphics and
+ /// ML paths; mangling the cache without the rename does nothing. Both are off
+ /// in `standard` because on a freshly restored 26.4 guest the rename makes
+ /// `bluetoothd`'s cached `kern.hv_vmm_present` lookup fail, which sends it down
+ /// the `MGIsDeviceOneOfType` path, leaves its transport singleton NULL and
+ /// crash-loops it until launchd throttles it; `locationd` then blocks on the
+ /// throttled Bluetooth XPC service, the data migrator hangs, and SpringBoard
+ /// waits on migration forever — a black screen with no panic. See
+ /// `Research/Patches/hv_vmm_present_usermode_xrefs.md`.
+ public static let hypervisorConcealmentPatches: Set<String> = [
+ "kernelcache_exp.hv_vmm",
+ "hv_vmm_dsc",
+ ]
/// The preset a VM gets when nothing else is named.
///
@@ -64,12 +85,17 @@ public enum FirmwarePatchSetCatalog {
selection: .block(manualOnlyPatches),
)
- /// Everything the bundle declares, including the Frida relaxations. Each
- /// patch's own version gate still decides whether it lands.
+ /// Everything the bundle declares, including the Frida relaxations and the
+ /// hypervisor concealment. Each patch's own version gate still decides whether
+ /// it lands.
public static let extendedPreset = VPhonePatchPreset(
identifier: "extended",
title: "Extended",
- summary: "Every patch this bundle declares, including the Frida Stalker relaxations.",
+ summary: """
+ Every patch this bundle declares, including the Frida Stalker relaxations \
+ and the hv_vmm_present concealment that a freshly restored 26.4 guest does \
+ not survive.
+ """,
patchSets: bundledReferences,
selection: .all,
)
diff --git a/VPhoneExecutable/VPhoneCommand/FirmwarePatcher/Pipeline/FirmwarePipelineComponents.swift b/VPhoneExecutable/VPhoneCommand/FirmwarePatcher/Pipeline/FirmwarePipelineComponents.swift
index 51d0ba19..bf9de108 100644
--- a/VPhoneExecutable/VPhoneCommand/FirmwarePatcher/Pipeline/FirmwarePipelineComponents.swift
+++ b/VPhoneExecutable/VPhoneCommand/FirmwarePatcher/Pipeline/FirmwarePipelineComponents.swift
@@ -45,7 +45,6 @@ extension FirmwarePipeline {
let includeBootChain = includesSet(FirmwareBootChainPatchSet.identifier)
let includeKernelBase = includesSet(FirmwareKernelBasePatchSet.identifier)
let includeKernelJailbreak = includesSet(FirmwareKernelJailbreakPatchSet.identifier)
- let includeHypervisor = includesSet(FirmwareKernelHypervisorPatchSet.identifier)
let includeDeviceTree = includesSet(FirmwareDeviceTreePatchSet.identifier)
/// Whether the plan turned a patch on. Without a plan, fall back to the
@@ -55,6 +54,13 @@ extension FirmwarePipeline {
plan?.isEnabled(identifier) ?? fallback
}
+ // The hypervisor concealment. Its set holds this one patch, so the patcher
+ // goes when the patch does — the gate would refuse the write anyway, but a
+ // patcher built to write nothing is a patcher whose log lines lie.
+ // `standard` blocks it: see FirmwareKernelHypervisorPatchSet.
+ let includeHypervisor = includesSet(FirmwareKernelHypervisorPatchSet.identifier)
+ && isEnabled("kernelcache_exp.hv_vmm", fallback: false)
+
let baseIs18 = iOSBase?.major == 18
let baseIs27 = iOSBase?.major == 27
diff --git a/VPhoneExecutable/VPhoneCommand/FirmwarePatcherTests/Core/FirmwarePatcherTests.swift b/VPhoneExecutable/VPhoneCommand/FirmwarePatcherTests/Core/FirmwarePatcherTests.swift
index f4e79970..96a92e40 100644
--- a/VPhoneExecutable/VPhoneCommand/FirmwarePatcherTests/Core/FirmwarePatcherTests.swift
+++ b/VPhoneExecutable/VPhoneCommand/FirmwarePatcherTests/Core/FirmwarePatcherTests.swift
@@ -523,16 +523,22 @@ struct IM4PPayloadParityTests {
}
struct FirmwarePipelineTests {
- @Test func `public JB includes former EXP kernel and device tree patchers`() throws {
+ @Test func `public JB includes the former EXP device tree patchers, not the hv_vmm rename`() throws {
let root = URL(fileURLWithPath: NSTemporaryDirectory())
let pipeline = FirmwarePipeline(vmDirectory: root, variant: .jb, verbose: false)
let components = pipeline.buildComponentList(
restoreDir: root,
iOSBase: VPhoneVersion("27.0"),
)
+ // Base and jailbreak, and no KernelExperimentalPatcher: the hv_vmm_present
+ // concealment is off unless a preset or a checkmark asks for it, and a
+ // patcher that would write nothing is not built. See
+ // FirmwareKernelHypervisorPatchSet for what it does to a 26.4 guest.
let kernel = try #require(components.first { $0.name == "kernelcache" })
- #expect(kernel.patcherFactories.count == 3)
- #expect(try kernel.patcherFactories[2](Data(), false) is KernelExperimentalPatcher)
+ #expect(kernel.patcherFactories.count == 2)
+ for factory in kernel.patcherFactories {
+ #expect(try !(factory(Data(), false) is KernelExperimentalPatcher))
+ }
let deviceTree = try #require(components.first { $0.name == "DeviceTree" })
let patcher = try #require(deviceTree.patcherFactories.first?(Data(), false) as? DeviceTreePatcher)
diff --git a/VPhoneExecutable/VPhoneCommand/FirmwarePatcherTests/PatchSets/FirmwarePatchSetCatalogTests.swift b/VPhoneExecutable/VPhoneCommand/FirmwarePatcherTests/PatchSets/FirmwarePatchSetCatalogTests.swift
index a317782b..6b47b75c 100644
--- a/VPhoneExecutable/VPhoneCommand/FirmwarePatcherTests/PatchSets/FirmwarePatchSetCatalogTests.swift
+++ b/VPhoneExecutable/VPhoneCommand/FirmwarePatcherTests/PatchSets/FirmwarePatchSetCatalogTests.swift
@@ -90,7 +90,7 @@ struct FirmwarePatchSetCatalogTests {
#expect(!plan.skippedByVersion.isEmpty)
}
- @Test("Standard leaves the Frida relaxations off, on every base")
+ @Test("Standard leaves every manual-only patch off, on every base")
func standardExcludesManualOnly() throws {
for base in ["18.6.2", "26.4", "27.0"] {
let plan = try VPhonePatchPlan.resolve(
@@ -105,7 +105,7 @@ struct FirmwarePatchSetCatalogTests {
}
}
- @Test("Extended is standard plus the Frida relaxations, and nothing else")
+ @Test("Extended is standard plus the manual-only patches, and nothing else")
func extendedDiffersOnlyByManualOnly() throws {
let standard = try VPhonePatchPlan.resolve(
preset: FirmwarePatchSetCatalog.standardPreset,
@@ -154,11 +154,51 @@ struct FirmwarePatchSetCatalogTests {
iOSBase: VPhoneVersion("26.4"),
cloudOS: VPhoneVersion(cloud),
)
- let on = FirmwarePatchSetCatalog.manualOnlyPatches.allSatisfy { plan.isEnabled($0) }
+ let on = FirmwareKernelFridaPatchSet.manifest.patches.allSatisfy {
+ plan.isEnabled($0.identifier)
+ }
#expect(on == (cloud == "26.4"))
}
}
+ @Test("The hv_vmm_present halves move together, and only extended has them")
+ func hypervisorConcealmentIsOptInAsAPair() throws {
+ // Two patches, one behaviour: the kernel OID rename and the shared-cache
+ // mangle. A plan holding one without the other is broken either way round —
+ // the rename alone breaks the graphics and ML paths, the mangle alone does
+ // nothing — and both together brick a freshly restored 26.4 guest, which is
+ // why standard has neither. See FirmwareKernelHypervisorPatchSet.
+ let pair = FirmwarePatchSetCatalog.hypervisorConcealmentPatches
+ #expect(pair.count == 2)
+ for identifier in pair {
+ let declaration = try #require(
+ FirmwarePatchSetCatalog.allDeclarations.first { $0.identifier == identifier },
+ "\(identifier) is not declared by any bundled set",
+ )
+ // Not boot-essential: standard drops both, and a shipped preset that
+ // drops a boot-essential patch warns on every run.
+ #expect(!declaration.bootEssential)
+ }
+ for base in ["18.6.2", "26.4", "27.0"] {
+ for cloud in ["26.1", "26.4"] {
+ for preset in FirmwarePatchSetCatalog.builtInPresets {
+ let plan = try VPhonePatchPlan.resolve(
+ preset: preset,
+ patchSets: FirmwarePatchSetCatalog.bundled,
+ iOSBase: VPhoneVersion(base),
+ cloudOS: VPhoneVersion(cloud),
+ )
+ let on = pair.filter { plan.isEnabled($0) }
+ #expect(
+ on.isEmpty || on == pair,
+ "\(preset.identifier) on \(base)/\(cloud) enabled only \(on)",
+ )
+ #expect(on.isEmpty == preset.isStandard)
+ }
+ }
+ }
+ }
+
@Test("Every set's manifest survives a plist round trip")
func manifestsRoundTrip() throws {
for manifest in FirmwarePatchSetCatalog.bundled {
diff --git a/VPhoneExecutable/VPhoneVirtualization/Resources/patches_presets/extended.plist b/VPhoneExecutable/VPhoneVirtualization/Resources/patches_presets/extended.plist
index ce6aa0b4..9e9641db 100644
--- a/VPhoneExecutable/VPhoneVirtualization/Resources/patches_presets/extended.plist
+++ b/VPhoneExecutable/VPhoneVirtualization/Resources/patches_presets/extended.plist
@@ -1,13 +1,19 @@
<?xml version="1.0" encoding="UTF-8"?>
<!DOCTYPE plist PUBLIC "-//Apple//DTD PLIST 1.0//EN" "http://www.apple.com/DTDs/PropertyList-1.0.dtd">
<!--
- Everything the bundle declares, including the Frida Stalker relaxations that
- `standard` leaves off.
+ Everything the bundle declares, including the two pairs `standard` leaves off:
+ the Frida Stalker relaxations, and the hv_vmm_present concealment.
- It names the same patch sets as standard and differs only in its selection, so
- the two presets are readable against each other: standard is this with two
- patches blocked. Each patch's own version gate still applies, so the Frida pair
- lands only on cloudOS 26.4 and newer.
+ Read this as the catalogue and standard as the safe subset — the two name the
+ same patch sets and differ only in their selection. Each patch's own version gate
+ still applies, so the Frida pair lands only on cloudOS 26.4 and newer.
+
+ The hv_vmm_present pair is here and nowhere else for a reason: on a freshly
+ restored 26.4 guest it is a brick. Renaming the OID makes bluetoothd's cached
+ sysctl lookup fail, bluetoothd crash-loops, locationd blocks on its throttled
+ XPC service, the data migrator never finishes and SpringBoard never draws.
+ Choose this preset for VM-detection work on a guest that already booted, not to
+ create one.
-->
<plist version="1.0">
<dict>
@@ -16,7 +22,7 @@
<key>Title</key>
<string>Extended</string>
<key>Summary</key>
- <string>Every patch this bundle declares, including the Frida Stalker relaxations.</string>
+ <string>Every patch this bundle declares, including the Frida Stalker relaxations and the hv_vmm_present concealment that a freshly restored 26.4 guest does not survive.</string>
<key>PatchSets</key>
<array>
<dict>
diff --git a/VPhoneExecutable/VPhoneVirtualization/Resources/patches_presets/standard.plist b/VPhoneExecutable/VPhoneVirtualization/Resources/patches_presets/standard.plist
index 0ba689ad..78f8a059 100644
--- a/VPhoneExecutable/VPhoneVirtualization/Resources/patches_presets/standard.plist
+++ b/VPhoneExecutable/VPhoneVirtualization/Resources/patches_presets/standard.plist
@@ -4,9 +4,15 @@
The preset every VM gets unless another is named.
It draws on every patch set the bundle ships and turns on all of their patches
- bar the Frida Stalker pair, which widen what any process in the guest may do and
- which nothing needs in order to boot. Each patch's own version gate still decides
- whether it lands on this pairing of iPhone base and cloudOS.
+ bar two pairs. The Frida Stalker pair widens what any process in the guest may do
+ and nothing needs it in order to boot. The hv_vmm_present pair — the kernel OID
+ rename and the shared-cache mangle — leaves a freshly restored 26.4 guest on a
+ black screen: bluetoothd caches the failed sysctl lookup, crash-loops, and the
+ data migrator SpringBoard waits on never finishes. Each pair is all-or-nothing;
+ see FirmwareKernelHypervisorPatchSet for the whole chain.
+
+ Each patch's own version gate still decides whether it lands on this pairing of
+ iPhone base and cloudOS.
Prewritten and reviewed. Nothing in the tool edits a preset: a VM records which
boxes its owner changed, and those compose on top of this.
@@ -87,6 +93,8 @@
<array>
<string>kernelcache_frida.thread_set_state_entitlement_flag</string>
<string>kernelcache_frida.vm_map_delete_immutable_code</string>
+ <string>kernelcache_exp.hv_vmm</string>
+ <string>hv_vmm_dsc</string>
</array>
</dict>
</dict>