Skip to content

Run a virtual iPhone on an Apple Silicon Mac.

Swift

Boot OpenAVB and OpenSEP machines with less than 7.5 GiB of memory

Jacky52bbf8059c87
diff --git a/Research/Firmware/openavb_boot_rom.md b/Research/Firmware/openavb_boot_rom.md
index be579fd8..5c015cf9 100644
--- a/Research/Firmware/openavb_boot_rom.md
+++ b/Research/Firmware/openavb_boot_rom.md
@@ -133,6 +133,21 @@ was found on a live machine by the first thing that stopped, in this order:
slot, and AVPBooter's "not full" send predicate were each tried and are
not needed.
+7. **SIDP block on a small guest** (carried as
+ `0001-Publish-the-iLLB-manifest-digest-on-t8110-only.patch`). Before LLB
+ runs, `source_image_publish_boot_manifest` writes the LLB manifest's
+ SHA-384 into the t8110 SIDP block (`0x2_3D2E_0044`, `+0x74`, `+0x84`).
+ Virtualization.framework maps nothing there unless the guest has about
+ 7.5 GiB of memory or more, the same threshold as OpenSEP's warm-boot mark
+ and DART (`opensep_sep_rom.md`). On a 6144 MB iPhone17,3 iOS 27.0.1
+ machine the first read of `+0x84` aborted right after `iLLB container
+ accepted`; the AP sat in `pvpanic_trigger` with x19 = `0x2_3D2E_0084`, the
+ SEP waited in its mailbox, and `vm create`'s first boot check timed out.
+ The 2026-10-18 validation ran on 8192 MB machines, where the block is
+ guest RAM. The vresearch101 iBoot boots on Apple's ROM, which publishes
+ nothing there, so the patch skips the publish when the chip ID register
+ reads 0, as it does under Virtualization.framework (change 2).
+
**How these were found.** The VM's GDB stub (the lower of the two ports
`lsof -iTCP -sTCP:LISTEN` shows for the `com.apple.Virtualization.VirtualMachine`
XPC process) reads registers and memory: the AP's PC, ESR/ELR/FAR, OpenAVB's
@@ -145,6 +160,11 @@ reach the VM's serial log.
## Validation
+- 6144 MB (2026-10-19, change 7 and OpenSEP's two patches): an iPhone17,3
+ iOS 27.0.1 machine boots on OpenAVB and OpenSEP, vphoned answers 13 s after
+ launch, and three guest reboots come back in 9 to 10 s
+ (`opensep_sep_rom.md` has the run).
+
- Normal boot: an iPhone17,3 iOS 26.6.2 machine restored on Apple's ROM, with
`romImages.avpBooter` pointed at OpenAVB, prints `OpenBootROM: SEP bootstrap
complete`, `Loaded LLB` and iBoot, and vphoned answers `device.info`.
@@ -174,8 +194,8 @@ log show `OpenBootROM:` for OpenAVB.
## Upstream
-All six changes are upstream: 1 to 5 in owngoal-dev/OpenAVB#1 (folded into
-`115169e`), 6 in owngoal-dev/OpenAVB#2. vphone carries no patch. The pin is
+Changes 1 to 6 are upstream: 1 to 5 in owngoal-dev/OpenAVB#1 (folded into
+`115169e`), 6 in owngoal-dev/OpenAVB#2. Change 7 is carried as a patch until owngoal-dev/OpenAVB#3 is merged. The pin is
#2's own commit, `3c7e7f6`, not its merge `63c1ead`: both have the same tree
and `main` reaches both, but upstream has rewritten `main` once, and GitHub
keeps a PR's head commit reachable (`refs/pull/2/head`) after a rewrite. If
diff --git a/Research/Firmware/opensep_sep_rom.md b/Research/Firmware/opensep_sep_rom.md
index 42ce1f3e..2f5f1d81 100644
--- a/Research/Firmware/opensep_sep_rom.md
+++ b/Research/Firmware/opensep_sep_rom.md
@@ -37,8 +37,9 @@ next start; the ROM it replaces stays in the folder. `vm info` prints
## The build
-`make PLATFORM=vresearch1 build/OpenSEP.bin` at the pinned commit, with no
-patches. `vresearch1` is OpenSEP's default and the row of its handoff table
+`make PLATFORM=vresearch1 build/OpenSEP.bin` at the pinned commit, with the
+patches in `Resources/OpenSEP/patches/` applied (two: the warm-boot check
+and the DART walk, below). `vresearch1` is OpenSEP's default and the row of its handoff table
that stands in for Virtualization.framework's AVPSEPBooter: the SEPOS runtime
is entered at the runtime region's base (`0x06000000`) with the handoff pointer
in `x0`, `x1` zero, and the firmware building its own translation tables. The
@@ -47,7 +48,7 @@ OpenSEP's `docs/platforms.md` cites the AVPSEPBooter addresses each value of
the `vresearch1` row was read from (macOS 26's image, SHA-256 `86d008eb…`, the
one this Mac's Virtualization.framework ships).
-The image is 147496 bytes and builds byte for byte the same from a fresh clone
+The image is 147512 bytes and builds byte for byte the same from a fresh clone
(`UpdateOpenSEP.sh` with no argument).
## What changes on the boot chain
@@ -75,14 +76,45 @@ and loaded by iBoot at boot, `sepfw-load-at-boot = 1`) stays as it was.
`OPEN_SECURE_ROM_ENFORCE_IMAGE_TRUST`. The `vphone600` firmware is not
encrypted, so OpenSEP's table of iPhone14,4 firmware keys is never used here.
- **Warm boot.** OpenSEP's reset path checks a warm-boot word at
- `0x2_3D2E_4804` that only the t8110 runtime writes. OpenSEP's own docs say
- the path is unexercised on `vresearch1`; guest reboots, which restart the SEP
- with the AP, come back (see Validation), so the SEP starts cold there.
+ `0x2_3D2E_4804` that only the t8110 runtime writes, and the pinned commit
+ does it on `vresearch1` too. Virtualization.framework maps nothing at that
+ IPA unless the guest has about 7.5 GiB of memory or more (measured on macOS
+ 27.0.1 26A434: DFU came up at 7680 and 8192 MB, never at 4096, 6144 or
+ 7168 MB). Below that the load is the SEP core's first fault: VBAR_EL1 is
+ still 0, it lands at `0x200`, and the framework's SEP process ends with
+ `Unhandled exception code` about 130 ms after it starts (a crash report of
+ `com.apple.Virtualization.VirtualMachine` with only `cpu-0` and
+ `virtual-clock` threads). The AP's ROM, OpenAVB or Apple's, then spins in
+ its mailbox acquire, `vm launch --dfu` says the VM started, and
+ `recovery-probe` never finds it. Single-stepping the SEP core through its
+ GDB stub stops on that load. The carried patch
+ (`0001-Check-the-warm-boot-mark-on-t8110-only.patch`) leaves the check out
+ of the `vresearch1` build, which then always starts cold; guest reboots,
+ which restart the SEP with the AP, come back cold anyway (see Validation).
+ Above the threshold the IPA is guest RAM, so the unpatched check could also
+ read a stale `1` and take the warm path.
+- **DART.** OpenSEP translated every address that arrives over the mailbox
+ through dart-sep's stream zero, reading its TCR and TTBR at IPA
+ `0x2_5D2C_0000`, the t8110 DART. That IPA has the same threshold. With the
+ warm-boot patch alone a 6144 MB iPhone17,3 iOS 27.0.1 machine came up in
+ DFU, took iBSS, iBEC, the ramdisk and `RestoreSEP`, entered restore mode,
+ and lost restored right after NORData (`Could not read data (-256)`), with
+ no crash report from the framework (by then OpenSEP's own vector table is
+ set, so a fault there would end in OpenSEP's panic path; that was not
+ observed directly). With the second patch the same restore finishes. On
+ the 8 GiB machines of the first validation the read was guest RAM that
+ still read zero, so the walk took the stream as unprogrammed and used the
+ address as it is. The second patch
+ (`0002-Walk-the-DART-on-t8110-only.patch`) makes that the `vresearch1`
+ behaviour.
## Validation
All on 2026-10-18, iPhone17,3 iOS 26.6.2 / cloudOS 26.4 (23E5207q), on a
-bundle with this change installed beside the default (`--keep-default`).
+bundle with this change installed beside the default (`--keep-default`), on
+machines with the default 8192 MB of memory and the unpatched image. That is
+above the warm-boot threshold, which is why none of it saw the failure; the
+Mac fleet's 6144 MB staging machines did.
- Restore: a machine created with `vm create --no-template` (`OpenAVB.bin` and
`OpenSEP.bin` in the folder) restores through OpenAVB's DFU:
@@ -110,8 +142,53 @@ To check the image after an update: `UpdateOpenSEP.sh` with no argument
rebuilds the pinned commit and compares byte for byte. To check a machine:
`vphone-cli vm info <vm>` prints `seprom:`.
+### 6144 MB, with both patches (2026-10-19)
+
+On this Mac (M5 Pro, macOS 27.0.1 26A434), iPhone17,3 iOS 27.0.1 (24A446) /
+cloudOS 26.4 (23E5207q), 6 CPU, 6144 MB, `vm create --no-template` from a
+build of upstream `61b4efb2` with both patches (image `92690b9c…`):
+
+- DFU comes up through OpenAVB, the restore ends `Restore Finished`, CFW
+ installs.
+- The first boot check timed out. The SEP core was waiting in
+ `sep_mailbox_receive_slot`; the AP was parked in OpenAVB's
+ `pvpanic_trigger` after `prepare_virtual_firmware_illb`, with x19 at
+ `0x2_3D2E_0084`. OpenAVB's `source_image_publish_boot_manifest` reads and
+ writes the t8110 SIDP window at `0x2_3D2E_0044`…`0x84`, in the same unbacked
+ range. That is OpenAVB's, not OpenSEP's; vphone carries a patch for it
+ (`openavb_boot_rom.md`, change 7).
+- The same machine after `vm config --rom apple` (Apple's AVPBooter, OpenSEP
+ kept): iOS boots, AppleSEPKeyStore loads the keybag and the data volume
+ unlocks, vphoned answers `device.info` (`ios_version` 27.0.1,
+ `memory_bytes` 6418317312), and `system.reboot` (`force: true`) comes back
+ in about 10 s with no `SEP RX timeout` or SEP panic on the console.
+
+- The same machine back on OpenAVB with its patch (image `bbcabe73…`) and
+ OpenSEP: `iLLB container accepted`, `virtual firmware prepared`, iBoot and
+ iOS boot, vphoned answers `device.info` 13 s after launch, and three
+ `system.reboot` (`force: true`) come back in 10, 9 and 10 s, with no
+ `SEP RX timeout` or panic on the console.
+
+- A template the way the Mac fleet builds one (`plan_template.go`), on the
+ final images (OpenAVB `bbcabe73…`, OpenSEP `92690b9c…`): `vm new
+ template-… --cpu 6 --memory 6144`, `fw prepare`, `fw patch --preset
+ standard`, `vm launch --dfu` (the first `recovery-probe` answered),
+ `restore` (`Restore Finished`), `cfw install`, `vm template trim --tier
+ standard`, `vm template setup --slim on --strict` (settle 64 s, its reboot
+ 13 s), `vm template adopt` (template `99538dfad74f`, not stale). A clone,
+ `vm create --template … --cpu 6 --memory 6144`, boots on OpenAVB and
+ OpenSEP: vphoned answers 15 s after launch, three `system.reboot` come back
+ in 11, 13 and 15 s, with no `SEP RX timeout` or panic. `vm create` with a
+ template does not reproduce the fleet here: it builds the template's staging
+ machine at 8 CPU and 8192 MB whatever `--memory` says
+ (`VPhoneVirtualMachineCreator.buildTemplate`), above the threshold.
+
+Not covered: iPad guests, other iOS versions, and the Mac fleet's own hosts
+(M4, the fleet's `tunnel` network).
+
## Upstream
-No patches are carried. If one is needed, it goes in
-`Resources/OpenSEP/patches/` as a `git format-patch` file until upstream takes
-it; `UpdateOpenSEP.sh` refuses a patch that no longer applies.
+Two patches are carried in `Resources/OpenSEP/patches/` as `git format-patch`
+files until owngoal-dev/OpenSEP#1 is merged: the warm-boot check and the DART walk
+above. `UpdateOpenSEP.sh`
+refuses a patch that no longer applies; delete it once the pin moves past it.
diff --git a/VPhoneExecutable/VPhoneVirtualization/Resources/OpenAVB/OpenAVB.bin b/VPhoneExecutable/VPhoneVirtualization/Resources/OpenAVB/OpenAVB.bin
index 08a4f977..2c370ca3 100644
Binary files a/VPhoneExecutable/VPhoneVirtualization/Resources/OpenAVB/OpenAVB.bin and b/VPhoneExecutable/VPhoneVirtualization/Resources/OpenAVB/OpenAVB.bin differ
diff --git a/VPhoneExecutable/VPhoneVirtualization/Resources/OpenAVB/README.md b/VPhoneExecutable/VPhoneVirtualization/Resources/OpenAVB/README.md
index 9f68e0f8..6e96b0d5 100644
--- a/VPhoneExecutable/VPhoneVirtualization/Resources/OpenAVB/README.md
+++ b/VPhoneExecutable/VPhoneVirtualization/Resources/OpenAVB/README.md
@@ -11,10 +11,10 @@ the image into each new machine folder.
| --- | --- |
| Source | `https://github.com/owngoal-dev/OpenAVB` |
| Commit | `3c7e7f602d2d35af67f50454a8216337f35b11bb` |
-| Patches | none (`patches/*.patch`, applied in order, when there are any) |
+| Patches | `patches/*.patch`, applied in order (one, see below) |
| Target | `make build/OpenAVB.bin` |
| Toolchain | Homebrew LLVM 23.1.3 (`clang`, `llvm-objcopy`), Homebrew LLD 23.1.3 |
-| SHA-256 | `d63ea62226d4383288a5a36c440aa2d1df8cccabe0118cac0ab2e0c23ba4c3d2` |
+| SHA-256 | `bbcabe73fd39fb92b96fbb4c661c1a15b0f9e5d9411468f7954c4f433911cb9c` |
| Size | 460057 bytes |
The image is checked in rather than built by the `VPhone` scheme, so building
@@ -26,7 +26,17 @@ header). It clones and builds outside the repository and needs
## Patches
-None. OpenAVB's first commit, `54b67bf`, did not run under
+- `0001-Publish-the-iLLB-manifest-digest-on-t8110-only.patch`. Before LLB runs,
+ OpenAVB publishes the LLB manifest's SHA-384 into the t8110 SIDP block at
+ IPA `0x2_3D2E_0044`…`0x84`. Virtualization.framework maps nothing there for
+ a guest with less than about 7.5 GiB of memory: the first read aborts right
+ after `iLLB container accepted` and the ROM parks in `pvpanic_trigger`, so a
+ 6144 MB machine never boots. The patch skips the publish when the chip ID
+ register reads 0, which is how Virtualization.framework presents it.
+ Proposed upstream in
+ [owngoal-dev/OpenAVB#3](https://github.com/owngoal-dev/OpenAVB/pull/3).
+
+Earlier changes are all upstream. OpenAVB's first commit, `54b67bf`, did not run under
Virtualization.framework. Five changes found on live machines made it boot and
restore, and upstream took them in
[owngoal-dev/OpenAVB#1](https://github.com/owngoal-dev/OpenAVB/pull/1), now
diff --git a/VPhoneExecutable/VPhoneVirtualization/Resources/OpenAVB/patches/0001-Publish-the-iLLB-manifest-digest-on-t8110-only.patch b/VPhoneExecutable/VPhoneVirtualization/Resources/OpenAVB/patches/0001-Publish-the-iLLB-manifest-digest-on-t8110-only.patch
new file mode 100644
index 00000000..7a49b04b
--- /dev/null
+++ b/VPhoneExecutable/VPhoneVirtualization/Resources/OpenAVB/patches/0001-Publish-the-iLLB-manifest-digest-on-t8110-only.patch
@@ -0,0 +1,98 @@
+From 0000000000000000000000000000000000000000 Mon Sep 17 00:00:00 2001
+From: Jacky <[email protected]>
+Date: Mon, 19 Oct 2026 11:06:52 +0800
+Subject: [PATCH] Publish the iLLB manifest digest on t8110 only
+
+source_image_publish_boot_manifest reads and writes the t8110 SIDP block at
+0x2_3D2E_0044..0x84. Virtualization.framework maps nothing there for a guest
+with less than about 7.5 GiB of memory: on a 6144 MB iPhone17,3 iOS 27.0.1
+guest the read of +0x84 aborts right after "iLLB container accepted" and the
+ROM parks in pvpanic_trigger before LLB runs (x19 = 0x2_3D2E_0084). With more
+memory the block is guest RAM. The vresearch101 iBoot boots on Apple's ROM,
+which publishes nothing there.
+
+The chip ID register, which Virtualization.framework leaves at 0 and
+platform_read_identity already falls back from, now names the machine: the
+publish runs only when it reads non-zero.
+
+Co-Authored-By: Claude Opus 5.5 <[email protected]>
+---
+ include/avp/platform/platform.h | 5 +++++
+ src/image/boot/source_image.c | 10 ++++++++++
+ src/platform/platform.c | 5 +----
+ 3 files changed, 16 insertions(+), 4 deletions(-)
+
+diff --git a/include/avp/platform/platform.h b/include/avp/platform/platform.h
+index 036ce25..764eec3 100644
+--- a/include/avp/platform/platform.h
++++ b/include/avp/platform/platform.h
+@@ -7,6 +7,11 @@
+
+ enum {
+ AVP_PLATFORM_BASE = 0x00400000,
++ /*
++ * Virtualization.framework leaves this register at 0 (platform.c), which
++ * is also how the ROM tells it from the t8110 machine it was written for.
++ */
++ AVP_PLATFORM_CHIP_ID = AVP_PLATFORM_BASE + 0x6f0,
+ AVP_PLATFORM_VARIANT_CAPACITY = 64,
+ AVP_AP_NONCE_SIZE = 32,
+ AVP_SEP_NONCE_SIZE = 20,
+diff --git a/src/image/boot/source_image.c b/src/image/boot/source_image.c
+index f747f1e..174b10e 100644
+--- a/src/image/boot/source_image.c
++++ b/src/image/boot/source_image.c
+@@ -4,6 +4,7 @@
+ #include "avp/crypto/hash/corecrypto_sha512_compat.h"
+ #include "avp/crypto/hash/sha256.h"
+ #include "avp/platform/mmio.h"
++#include "avp/platform/platform.h"
+ #include "avp/runtime/memory/runtime_memory.h"
+ #include "lzfse.h"
+
+@@ -332,6 +333,12 @@ static bool prepare_source_fixture(const uint8_t *image, uint32_t image_size) {
+ * panics on finding it already locked -- so this runs once, here, before the
+ * firmware starts. Bit 16 of `+0x40` is the factory lock on the ROM manifest
+ * and is not ours to touch; the board seeds it.
++ *
++ * Only the t8110 machine has the block. Virtualization.framework maps nothing
++ * at `0x2_3D2E_0000` unless the guest has about 7.5 GiB of memory or more:
++ * below that the first read of `+0x84` aborts and the ROM parks in pvpanic
++ * before LLB runs, and above it the window is guest RAM. Its iBoot boots on
++ * Apple's ROM, which publishes nothing here, so there the ROM leaves it alone.
+ */
+ enum {
+ SOURCE_IMAGE_SIDP_BOOT = 0x23d2e0044ull,
+@@ -349,6 +356,9 @@ static void source_image_publish_boot_manifest(
+ AVPCCSHA512Context context;
+ uint8_t digest[AVP_SHA384_DIGEST_SIZE];
+
++ if (mmio_read32(AVP_PLATFORM_CHIP_ID) == 0) {
++ return;
++ }
+ if (image->encoded_manifest_bytes == 0 ||
+ image->encoded_manifest_size == 0) {
+ source_image_diagnostic(
+diff --git a/src/platform/platform.c b/src/platform/platform.c
+index bf18fb7..6891e6a 100644
+--- a/src/platform/platform.c
++++ b/src/platform/platform.c
+@@ -13,7 +13,6 @@ enum {
+ PLATFORM_ENTROPY_OFFSET = 0x38,
+ PLATFORM_BOOT_CONTEXT_OFFSET = 0x300,
+ PLATFORM_VARIANT_OFFSET = 0x5c8,
+- PLATFORM_CHIP_ID_OFFSET = 0x6f0,
+ /*
+ * The matching vphone600 machine profile carries these values outside
+ * the readable identity registers.
+@@ -91,9 +90,7 @@ void platform_publish_boot_context(const AVPBootContext *context) {
+
+ void platform_read_identity(AVPPlatformIdentity *identity) {
+ identity->ecid = platform_ecid();
+- identity->chip_id = mmio_read32(
+- AVP_PLATFORM_BASE + PLATFORM_CHIP_ID_OFFSET
+- );
++ identity->chip_id = mmio_read32(AVP_PLATFORM_CHIP_ID);
+ if (identity->chip_id == 0) {
+ identity->chip_id = VPHONE600_CHIP_ID;
+ }
diff --git a/VPhoneExecutable/VPhoneVirtualization/Resources/OpenSEP/OpenSEP.bin b/VPhoneExecutable/VPhoneVirtualization/Resources/OpenSEP/OpenSEP.bin
index 70a69774..052d0b17 100644
Binary files a/VPhoneExecutable/VPhoneVirtualization/Resources/OpenSEP/OpenSEP.bin and b/VPhoneExecutable/VPhoneVirtualization/Resources/OpenSEP/OpenSEP.bin differ
diff --git a/VPhoneExecutable/VPhoneVirtualization/Resources/OpenSEP/README.md b/VPhoneExecutable/VPhoneVirtualization/Resources/OpenSEP/README.md
index f1519f72..c6cc3b3e 100644
--- a/VPhoneExecutable/VPhoneVirtualization/Resources/OpenSEP/README.md
+++ b/VPhoneExecutable/VPhoneVirtualization/Resources/OpenSEP/README.md
@@ -12,11 +12,11 @@ each new machine folder.
| --- | --- |
| Source | `https://github.com/owngoal-dev/OpenSEP` |
| Commit | `39ff3d7572059668077388dd907b63308ddf645e` |
-| Patches | `patches/*.patch`, applied in order (none) |
+| Patches | `patches/*.patch`, applied in order (two, see below) |
| Target | `make PLATFORM=vresearch1 build/OpenSEP.bin` |
| Toolchain | Homebrew LLVM 23.1.3 (`clang`, `llvm-objcopy`), Homebrew LLD 23.1.3 |
-| SHA-256 | `fe3f8de4334dd00ba6862bd95cd86d9bea7534e47592432a601d815edfb911d6` |
-| Size | 147496 bytes (the SEP ROM window is `0x29000`, 167936) |
+| SHA-256 | `92690b9c7e900ec0054271dbbc0444ee2d8567c8d41d3a2e491ba011e2b08b12` |
+| Size | 147512 bytes (the SEP ROM window is `0x29000`, 167936) |
The image is checked in rather than built by the `VPhone` scheme, so building
the bundle needs no cross toolchain. To move to another commit, or to check
@@ -27,7 +27,23 @@ header). It clones and builds outside the repository and needs
## Patches
-None. If vphone needs a change upstream has not taken yet, put it in
+- `0001-Check-the-warm-boot-mark-on-t8110-only.patch`. The reset path reads
+ a warm-boot mark at IPA `0x2_3D2E_4804` that only the t8110 runtime writes.
+ Virtualization.framework maps nothing there for a guest with less than about
+ 7.5 GiB of memory, so the load aborts before the vector table is set, the
+ framework ends the SEP core (`Unhandled exception code`) and the AP's ROM
+ waits for the mailbox forever: DFU never comes up. The patch leaves the
+ check out of the `vresearch1` build. Proposed upstream in
+ [owngoal-dev/OpenSEP#1](https://github.com/owngoal-dev/OpenSEP/pull/1).
+- `0002-Walk-the-DART-on-t8110-only.patch`. Mailbox addresses were translated
+ through the t8110 DART, whose registers OpenSEP reads at IPA
+ `0x2_5D2C_0000`. Under the same memory threshold that read aborts inside
+ OpenSEP's own vector table during the first image copy of a restore, the
+ SEP panics, and restored drops the connection right after NORData. The
+ `vresearch1` build now uses the addresses as they are. Proposed upstream in
+ [owngoal-dev/OpenSEP#1](https://github.com/owngoal-dev/OpenSEP/pull/1).
+
+If vphone needs a change upstream has not taken yet, put it in
`patches/` as a `git format-patch` file. `UpdateOpenSEP.sh` applies patches in
order and refuses one that no longer applies. Delete a patch once the pin
moves past it.
diff --git a/VPhoneExecutable/VPhoneVirtualization/Resources/OpenSEP/patches/0001-Check-the-warm-boot-mark-on-t8110-only.patch b/VPhoneExecutable/VPhoneVirtualization/Resources/OpenSEP/patches/0001-Check-the-warm-boot-mark-on-t8110-only.patch
new file mode 100644
index 00000000..91c60094
--- /dev/null
+++ b/VPhoneExecutable/VPhoneVirtualization/Resources/OpenSEP/patches/0001-Check-the-warm-boot-mark-on-t8110-only.patch
@@ -0,0 +1,109 @@
+From 0000000000000000000000000000000000000000 Mon Sep 17 00:00:00 2001
+From: Jacky <[email protected]>
+Date: Mon, 19 Oct 2026 08:36:57 +0800
+Subject: [PATCH 1/2] Check the warm-boot mark on t8110 only
+
+The reset path reads the warm-boot mark at 0x2_3D2E_4804, which only the
+t8110 runtime's deep-sleep path writes. Under Virtualization.framework
+(vresearch1) nothing is mapped at that IPA for a guest with less than about
+7.5 GiB of memory: the load takes a synchronous abort before VBAR_EL1 is set,
+the framework ends the SEP core with "Unhandled exception code", and the AP's
+ROM waits for the mailbox forever. With more memory the IPA is guest RAM, so
+a stale 1 there would send a cold boot into the warm path.
+
+Measured on macOS 27.0.1 (26A434) with a vresearch1 build: DFU comes up at
+7680 and 8192 MB and never at 4096, 6144 or 7168 MB; with this change it
+comes up at 4096 and 6144 MB. Single-stepping the SEP core through the GDB
+stub stops at the load, then at the vector offset 0x200.
+
+The vresearch1 build now always starts cold. The t8110 build is unchanged.
+
+Co-Authored-By: Claude Opus 5.5 <[email protected]>
+---
+ Makefile | 3 ++-
+ docs/external-contract.md | 9 +++++----
+ docs/platforms.md | 9 ++++++---
+ src/runtime/boot/start.S | 8 ++++++++
+ 4 files changed, 21 insertions(+), 8 deletions(-)
+
+diff --git a/Makefile b/Makefile
+index b26b822..472133c 100644
+--- a/Makefile
++++ b/Makefile
+@@ -26,7 +26,8 @@ CFLAGS := --target=aarch64-none-elf -std=c17 -O3 -ffast-math -g \
+ -mbranch-protection=bti -fno-pic -fno-pie -fno-common \
+ -fno-vectorize -fno-slp-vectorize -Wall -Wextra -Werror $(INCLUDES) \
+ -DOPEN_SECURE_ROM_PLATFORM=$(PLATFORM_ID)
+-ASFLAGS := --target=aarch64-none-elf -g -c $(INCLUDES)
++ASFLAGS := --target=aarch64-none-elf -g -c $(INCLUDES) \
++ $(if $(filter t8110,$(PLATFORM)),-DOPEN_SECURE_ROM_WARM_BOOT)
+ LDFLAGS := -T linker.ld --no-undefined --build-id=none \
+ --fatal-warnings -nostdlib -static
+
+diff --git a/docs/external-contract.md b/docs/external-contract.md
+index 6144164..c187118 100644
+--- a/docs/external-contract.md
++++ b/docs/external-contract.md
+@@ -23,10 +23,11 @@ per-platform differences in the handoff are in [platforms.md](platforms.md).
+ pointer in x0. On `t8110` it is IPA `0x3_4000_0000`, where the d16
+ firmware expects to run (it checks its own PC), so the emulator must alias
+ that IPA onto the same memory. See [docs/platforms.md](platforms.md).
+-- **Warm boot**: on reset the ROM reads the word at `0x2_3D2E_4804`. If it is
+- `1` (the t8110 runtime's deep-sleep path writes it), the ROM clears it and
+- jumps straight back into the runtime at `0x3_4000_0000`. On a cold boot it
+- must read as anything other than `1`.
++- **Warm boot** (`t8110` only): on reset the ROM reads the word at
++ `0x2_3D2E_4804`. If it is `1` (the t8110 runtime's deep-sleep path writes
++ it), the ROM clears it and jumps straight back into the runtime at
++ `0x3_4000_0000`. On a cold boot it must read as anything other than `1`.
++ The `vresearch1` build does not read it.
+ - **DART**: addresses that arrive over the mailbox are dart-sep IOVAs. The ROM
+ walks the DART itself (stream 0 at IPA `0x2_5D2C_0000`: TCR at `+0x1000`,
+ TTBR at `+0x1400`, two levels of 11 index bits over 16 KiB pages). A
+diff --git a/docs/platforms.md b/docs/platforms.md
+index 041c956..96cd1d8 100644
+--- a/docs/platforms.md
++++ b/docs/platforms.md
+@@ -83,9 +83,12 @@ them for this yet (see the end of this page).
+
+ ## Not yet covered by the table
+
+-- **Warm boot.** On every platform, `start.S` re-enters the runtime at
+- `0x3_4000_0000` when the warm mark at `0x2_3D2E_4804` reads `1`. That is the
+- t8110 behaviour. On `vresearch1` the path has not been exercised.
++- **Warm boot.** Only the `t8110` build checks the warm mark at
++ `0x2_3D2E_4804` and re-enters the runtime at `0x3_4000_0000` when it reads
++ `1`. The `vresearch1` build always starts cold: Virtualization.framework
++ backs nothing at that IPA for a guest with less than about 7.5 GiB of
++ memory, where the load aborts before the vector table is set and the
++ framework stops the SEP core, and for a larger guest it is guest RAM.
+ - **Selection at run time.** The row is fixed when the ROM is built. Two ways
+ to choose it at run time instead:
+ - read the manifest's CHIP/BORD, which the ROM already evaluates;
+diff --git a/src/runtime/boot/start.S b/src/runtime/boot/start.S
+index eaa1171..469589e 100644
+--- a/src/runtime/boot/start.S
++++ b/src/runtime/boot/start.S
+@@ -53,7 +53,14 @@ sep_reset_entry:
+ * image+0x20000 and returns into the thread that went down (measured on
+ * the 22G100 SEP runtime). A cold boot never has the word
+ * set -- only that path writes it -- so nothing else here changes.
++ *
++ * Only t8110 has that path. Virtualization.framework backs nothing at
++ * this IPA on vresearch1 unless the guest has about 7.5 GiB of memory or
++ * more: below that the load takes a synchronous abort before
++ * VBAR_EL1 is set and the framework ends the SEP core, and above it the
++ * word is guest RAM, which can read 1.
+ */
++#if defined(OPEN_SECURE_ROM_WARM_BOOT)
+ ldr x10, =SEP_BOOT_WARM_MARK
+ ldr w11, [x10]
+ cmp w11, #1
+@@ -63,6 +70,7 @@ sep_reset_entry:
+ isb
+ ldr x10, =SEP_BOOT_RUNTIME_ENTRY
+ br x10
++#endif
+ .Lbootstrap_cold:
+
+ adr x10, sep_vector_table
diff --git a/VPhoneExecutable/VPhoneVirtualization/Resources/OpenSEP/patches/0002-Walk-the-DART-on-t8110-only.patch b/VPhoneExecutable/VPhoneVirtualization/Resources/OpenSEP/patches/0002-Walk-the-DART-on-t8110-only.patch
new file mode 100644
index 00000000..8767b8be
--- /dev/null
+++ b/VPhoneExecutable/VPhoneVirtualization/Resources/OpenSEP/patches/0002-Walk-the-DART-on-t8110-only.patch
@@ -0,0 +1,145 @@
+From 0000000000000000000000000000000000000000 Mon Sep 17 00:00:00 2001
+From: Jacky <[email protected]>
+Date: Mon, 19 Oct 2026 09:14:21 +0800
+Subject: [PATCH 2/2] Walk the DART on t8110 only
+
+Addresses that arrive over the mailbox are translated through dart-sep's
+stream zero, whose registers are read at 0x2_5D2C_0000, the t8110 DART.
+Virtualization.framework (vresearch1) maps nothing there for a guest with
+less than about 7.5 GiB of memory. With only the warm-boot fix, a 6144 MB
+guest entered restore mode and lost restored right after NORData ("Could not
+read data (-256)") with no crash report from the framework; with this change
+the same restore finishes, and the guest then boots with its SEP on this ROM
+and comes back from a reboot. With more memory the read is guest RAM, which
+only works while that RAM still reads zero.
+
+The platform row gains walks_dart; vresearch1 uses the address as it is,
+which is what those reads amounted to on the 8 GiB guests the platform was
+validated on. t8110 is unchanged.
+
+Co-Authored-By: Claude Opus 5.5 <[email protected]>
+---
+ docs/external-contract.md | 3 ++-
+ docs/platforms.md | 5 +++++
+ .../runtime/boot/boot_platform.h | 6 ++++++
+ src/image/shared/shared_region.c | 17 +++++++++++------
+ src/runtime/boot/boot_platform.c | 7 +++++++
+ 5 files changed, 31 insertions(+), 7 deletions(-)
+
+diff --git a/docs/external-contract.md b/docs/external-contract.md
+index c187118..153148b 100644
+--- a/docs/external-contract.md
++++ b/docs/external-contract.md
+@@ -28,7 +28,8 @@ per-platform differences in the handoff are in [platforms.md](platforms.md).
+ it), the ROM clears it and jumps straight back into the runtime at
+ `0x3_4000_0000`. On a cold boot it must read as anything other than `1`.
+ The `vresearch1` build does not read it.
+-- **DART**: addresses that arrive over the mailbox are dart-sep IOVAs. The ROM
++- **DART** (`t8110` only): addresses that arrive over the mailbox are dart-sep
++ IOVAs. The `vresearch1` build uses them as they are. The `t8110` ROM
+ walks the DART itself (stream 0 at IPA `0x2_5D2C_0000`: TCR at `+0x1000`,
+ TTBR at `+0x1400`, two levels of 11 index bits over 16 KiB pages). A
+ bypassed or unprogrammed stream means the address is used as is.
+diff --git a/docs/platforms.md b/docs/platforms.md
+index 96cd1d8..cafe97f 100644
+--- a/docs/platforms.md
++++ b/docs/platforms.md
+@@ -89,6 +89,11 @@ them for this yet (see the end of this page).
+ backs nothing at that IPA for a guest with less than about 7.5 GiB of
+ memory, where the load aborts before the vector table is set and the
+ framework stops the SEP core, and for a larger guest it is guest RAM.
++- **DART.** Only the `t8110` build translates mailbox addresses through
++ dart-sep's stream zero at `0x2_5D2C_0000`. The `vresearch1` build uses them
++ as they are: under Virtualization.framework that IPA is unbacked for a
++ guest with less than about 7.5 GiB of memory, where the read aborts and the
++ image copy panics mid-restore, and guest RAM for a larger one.
+ - **Selection at run time.** The row is fixed when the ROM is built. Two ways
+ to choose it at run time instead:
+ - read the manifest's CHIP/BORD, which the ROM already evaluates;
+diff --git a/include/open_secure_rom/runtime/boot/boot_platform.h b/include/open_secure_rom/runtime/boot/boot_platform.h
+index 51e82d1..48e048d 100644
+--- a/include/open_secure_rom/runtime/boot/boot_platform.h
++++ b/include/open_secure_rom/runtime/boot/boot_platform.h
+@@ -48,6 +48,12 @@ typedef struct {
+ * with; otherwise the runtime builds its own.
+ */
+ sep_bool rom_builds_translation_tables;
++ /*
++ * Addresses that arrive over the mailbox are translated through
++ * dart-sep's stream zero, read at SEP_DART_BASE (shared_region.c);
++ * otherwise they are used as they are.
++ */
++ sep_bool walks_dart;
+ } sep_boot_platform_t;
+
+ enum {
+diff --git a/src/image/shared/shared_region.c b/src/image/shared/shared_region.c
+index e6816f1..af1842c 100644
+--- a/src/image/shared/shared_region.c
++++ b/src/image/shared/shared_region.c
+@@ -1,5 +1,7 @@
+ #include "open_secure_rom/system_closure.h"
+
++#include "open_secure_rom/boot_platform.h"
++
+ #include "open_secure_rom/der.h"
+ #include "open_secure_rom/event_log.h"
+ #include "open_secure_rom/panic.h"
+@@ -89,16 +91,19 @@ static sep_u64 sep_peek64(sep_u64 physical_address, sep_u32 attribute) {
+ }
+
+ sep_u64 sep_shared_region_translate(sep_u64 iova) {
+- const sep_u32 tcr =
+- sep_peek32(SEP_DART_BASE + SEP_DART_TCR + SEP_DART_STREAM * 4,
+- SEP_ATTR_DEVICE);
+- const sep_u32 ttbr =
+- sep_peek32(SEP_DART_BASE + SEP_DART_TTBR + SEP_DART_STREAM * 4,
+- SEP_ATTR_DEVICE);
++ sep_u32 tcr;
++ sep_u32 ttbr;
+ sep_u64 table;
+ sep_u64 entry = 0;
+ int level;
+
++ if (!sep_boot_platform()->walks_dart) {
++ return iova;
++ }
++ tcr = sep_peek32(SEP_DART_BASE + SEP_DART_TCR + SEP_DART_STREAM * 4,
++ SEP_ATTR_DEVICE);
++ ttbr = sep_peek32(SEP_DART_BASE + SEP_DART_TTBR + SEP_DART_STREAM * 4,
++ SEP_ATTR_DEVICE);
+ if ((tcr & SEP_DART_TCR_BYPASS) != 0) {
+ return iova;
+ }
+diff --git a/src/runtime/boot/boot_platform.c b/src/runtime/boot/boot_platform.c
+index 5d3f9e3..4fe45cf 100644
+--- a/src/runtime/boot/boot_platform.c
++++ b/src/runtime/boot/boot_platform.c
+@@ -20,6 +20,11 @@ static const sep_boot_platform_t boot_platforms[] = {
+ * 0x18000 (0x2001D00); and builds no translation tables. The
+ * vphone600 SEPOS keeps x0, branches on it being zero, and reads the
+ * magic through it (0x4008, 0x404C, 0x6328 in 23E5207q).
++ *
++ * It walks no DART. Virtualization.framework backs nothing at the
++ * t8110 DART's address unless the guest has about 7.5 GiB of memory or
++ * more; below that the read aborts and the copy panics mid-restore, and
++ * above it the read is guest RAM.
+ */
+ [SEP_BOOT_PLATFORM_VRESEARCH1] = {
+ .runtime_entry = 0,
+@@ -28,6 +33,7 @@ static const sep_boot_platform_t boot_platforms[] = {
+ .handoff_in_power_down_register = 0,
+ .cold_input_offset = 0,
+ .rom_builds_translation_tables = 0,
++ .walks_dart = 0,
+ },
+ /*
+ * What the d16 SEPOS needs from the t8110 SEPROM, measured on the
+@@ -66,6 +72,7 @@ static const sep_boot_platform_t boot_platforms[] = {
+ .handoff_in_power_down_register = 1,
+ .cold_input_offset = 0x680000ULL,
+ .rom_builds_translation_tables = 1,
++ .walks_dart = 1,
+ },
+ };
+