Boot Process

From Xenon Wiki
Revision as of 17:25, 23 October 2017 by imported>TEIR1plus2 (→‎CF)
Jump to navigation Jump to search

This page details the boot process used in the Xbox 360. The process is slightly different between Devkit/Retail and certain boxes which have a secondary CB loader (trinity/some jaspers)

Retail

1BL (Inside CPU)

Loads and decrypts CB(_A) into ram, it computes a RotSumSha1 of CB(_A) and uses it to verify its RSA signature. If the signature is valid, 1BL jumps to CB(_A).

CB

SLIMS ONLY: CB_A loads and decrypts CB_B into ram, it computes a RotSumSha1 of CB_B and checks it against the known hash. If it matches CB_A jumps to CB_B.

CB(_B) starts up a virtual machine, which does:

- initializes the PCI Bridge
- disables the GPU PCIE JTAG test port
- initializes the serial port
- talks to the SMC to clear the "handshake" bit
- initializes memory
- generates RROD if memory init fails

CB(_B) loads and decrypts CD into ram, it computes a RotSumSha1 of CD and checks it against the known hash. If it matches CB(_B) jumps to CD.

Methods to dump the CB bootloaders and examples of them reversed can be found here: CB Code

CD

Loads and decrypts CE into ram, computes a RotSumSha1 of CE. If it matches the known hash, it decompresses the LZX compressed base kernel.
It then checks for patch slots. If it finds one, it will load and decrypt the corresponding CF bootloader for that patch. It then computes a RotSumSha1 of CF and uses it to verify the CF loaders signature. If the signature is valid, CD stays in memory but jumps to CF.

There can be up to 2 CF/CG pairs for patching the kernel.

CF

Loads CG into memory by reading the CG data in the NAND header and then reads the rest of the CG data from the CG blocks specified in CF header. It then decrypts the CG in memory using the key generated from decrypting CF. To verify it, it computes a RotSumSha1 and checks it against the known hash. If it matches, it uses LZX delta decompress to apply the patch to the base kernel in memory. Once its done, it jumps back to CD and once CD has finished up, it jumps to the reset vector in the Hypervisor.

hypervisor

no booting details known changes between beta hardware and final: alpha hardware = macintosh beta = ? looks like retail, but no encryption second beta =! retail

tried to dump RAM could only dump virtual memory ram is at 8000_0000 southbridge: pci config space, mapped to VM, accessible by user apps memory at bottom looks random/encrypted, might be hypervisor 256 KB

8040_0000 xbox kernel starts, MZ header

read memory using debug interface: everything is in plaintext, you can read kernel + app (dashboard etc.), i.e. virtual memory is not encrypted

kernel interesting to disassemble communication with hypervisor using syscalls

hypervisor does interrupts/exceptions

syscalls:

final:

SC 00: GetVersionCode (e.g. r3=072F8002)

SC 01: KeStartupProcessors

SC 02: unknown KiQuiesce

SC 03: KeFlushEntireTb

SC 04: called in FlushMultipleTb

SC 05: ??

SC 06: KeGetSpecialPurposeRegister (r3=0x3F5)

SC 07: KeSetSpecialPurposeRegister

SC 08: KeGetSocRegister(r3=???)/KeGetPWMRegister(r3=60000)/ KeGetPRVRegister(r3=61000)


SC 09: KeSetSocRegister

SC 0A: KeStartupProcessors

SC 0B: called in ReserveKernelPtes

SC 0C: called from MmAllocatePhysicalMemoryEx

SC 0D: setAD16

SC 0E: KeEnablePPUPerformanceMonitor

SC 0F: called from MmGetPhysicalAddress

SC 10: called from MmDbgReleaseAddress


SC 11: XexpLoadFile calls it, seems to get privkey

r4 = phys addr (of header?) offset: +8

r5 = region

r6 = ?? offset: +4

r7 = ?? size?


SC 12: called from MmAllocateImageMemory

SC 13: called from MmAllocateImageMemory


SC 14: called in XexpLoadFile

SC 15: called in XexpLoadFile


SC 16: called in XexpCompleteImageLoad

SC 17: called in XexpCompleteImageLoad


SC 18: called in XexpLoadFile, XexpCompleteImageLoad

SC 19: unload?

SC 1A: unload?

SC 1B: unload?

SC 1c: called on XexpTitleTerminateNotification


SC 1d: KeCreateUserMode

SC 1e: KeDeleteUserMode

SC 1f: Flush TLB

SC 20: set power

SC 21: shadow boot

SC 22: f*** fuses

SC 23: FSB interrupt related

SC 24: KeLockL2


SC 25:

SC 26

SC 27

SC 28

SC 29

SC 2A

SC 2B

SC 2C: SataCdRomHvVerifyLBA

SC 2D

SC 2E: XeKeysInitialize (r3, r4 = address)

SC 2F: XeKeysGetKeyProperties

SC 30: XeKeysGetStatus

SC 31: XeKeysGenerateRandomKey

SC 32: XeKeysGetFactoryChallenge

SC 33: XeKeysSetFactoryResponse

SC 34: XeKeysSaveBootLoader

SC 35: XeKeysSaveKeyVault

SC 36: XeKeysSetKey

SC 37: XeKeysGetKey

SC 38: XeKeysGetDigest

SC 39: XeKeysQwNeRsaPrvCrypt

SC 3D: XeKeysDesCbc. r6: address, r5: context


SC 3F: XeKeysSaveSystemUpdate

SC 40: XeKeysExecute


SC 22 =

tested on 2 kernels

first: SC "access fuses"

second: "burn fuses"

(rumour has it that this is used to make retail boxes out of debug boxes)


memory management 0F/10: perhaps page table access code in hypervisor, all high level code in kernel


you can't map memory as you like

network adapter in the southbridge

debug code dumps registers with names

it is possible to dump physical memory using network adapter DMA accesses

not perfect dump


reading physical memory = encrypted

data segments are not encrypted, but nearly all code segments


older recovery cd (early 2005), worked on first beta developer kits, without security enabled:

cd included kernel which included stuff that is encrypted in retail version

includes hypervisor code! it is old, but...

getspr: SC 6

setspr: SC 7 -> possible to see implementation of basic syscall handling


function in hypervisor to chain-run a new kernel from the old kernel


hypervisor: sign with private key etc.

hypervisor can only do physical memory

hashing: load into register base address, length, destination of hash buffer, call syscode, hypervisor will hash -> attack: hash 1 byte, *itself*, -> hangs


hypervisor lies at 0 in VM and physical mem