Boot Order

Boot Sequence

Power on: syscon boots from its internal (non-encrypted / dual banked) ROM *1 *2

+ syscon powers up various power subsystems
+ syscon powers up cell and checks status
+ syscon sends Cell configuration ring to Cell It is either sent during or before bootldr. The config ring is checked within bootldr (ch64).
+ syscon pulls the reset of Cell high -> Cell INIT Not done till end of bootldr. Even after, LV0 is still running at 400MHz. (LV0 boots frequency to 3GHz).

Cell INIT: CELL boots from its internal ROM *2

+ fetches encrypted bootldr off NAND  (at address 0x000000) /NOR flash (at address 0xFC0000) and boots in isolated SPU.

Bootldr Running: (Which SPU?)

+ Initialises I/O (IOIF0/IOIF1)
+ Initialises RAM (And verifies with memtest elf - On SPU 0 - It's hardcoded to load there).
+ bootldr decrypts lv0 which runs on PPU -> loaders INIT
  NEW consoles only: metadata lv0.2 (signed with nonrandomfail key) is used to check lv0 integrity

loaders INIT: lv0 loads metldr (SPE2)

+ passes lv1ldr (which loads lv1) to metldr
+ passes lv2ldr (which loads lv2) to metldr
+ passes appldr (which loads vsh) to metldr
+ passes isoldr (which loads *.iso_spu_module) to metldr
+ passes rvkldr (which loads rvkprg / rvklist) to metldr

• 1) Read/Writeable with undocumented / should also be read/writeable through serial port and possible to switch it to the backup bank1 with backup_mode pulled high
• 2) CEX/Retail consoles go to standby with red light. SEX/SHOP/SECH will not standby, but instead boot through without waiting for powerbutton. Also check is done on all models if update is flagged to set it into firmware updating procedure
• 3) Partialy Read/Writeable

about the disabled SPE: syscon reads it’s internal (non-encrypted) eeprom @ 0x48C30 which is value 0×06 on all CEX/Retail consoles and will set the cell config ring accordingly for 7 SPE’s. SPE0 and SPE2 are reserved for bootldr and metldr for isolation respectively. Setting the value to a nonworking state (e.g. 0×00, 0xFF, enabling a defective SPE or disabling a needed SPE for proper boot) might brick the console, locking you out from restoring the correct value to the syscon eeprom. Config ring is checked against the known one in bootldr. If you were to modify syscon and the config ring, it still wouldn't boot and would panic as the config ring does not match the expected one.

References

• Secure power-on reset engine
  • US7895426.pdf
  • US20090055637.pdf
• CellBE_Handbook_v1.12_3Apr09_pub.pdf
• Cell_Broadband_Engine_processor_vault_security_architecture.pdf
• CellBEBootprocess.pdf (177.69 KB)) //
• CBE_Secure_SDK_Guide_v3.0.pdf
• CellBE_HIG_65nm_v1.01_8Jun2007.pdf)
• CellBE_HIG_90nm_v1.5_30Nov2007_pub.pdf)
• BE_Hardwar_Init_Guide_v1.3_31March2006.pdf

Chain of Trust

Name	Location	Processor	Encryption	Updateable	Revokable	Usage	Exploited
Runtime Secure Boot	Hardware based	Cell	Hardware Based	no	no	Verification of images loaded into isolated SPE	no
bootldr (Boot Loader)	NAND (0x000000) / NOR (0xFC0000)	SPE(0)	Per Console Encrypted at factory	No *	No	Setup Primairy Hardware + load lv0	No
lv0 (Level 0)	NAND/NOR (COREOS)	PPU	Static Encryption / Signed	Yes	No	Setup Hardware	Yes
metldr (asecure_loader)	NAND (0x0040810) / NOR (0x000810)	SPE(2)	Per Console Encrypted at factory	No *	No	load loaders (Meta Loader)	Yes
lv1ldr (Level 1 (Hypervisor) Loader)	NAND/NOR (COREOS)	SPE(2)	Static Encryption / Signed	Yes	No	Decrypt lv1 (Hypervisor)	Yes
lv2ldr (Level 2 (GameOS) Loader)	NAND/NOR (COREOS)	SPE(2)	Static Encryption / Signed	Yes	No	Decrypt lv2 (GameOS)	Yes
appldr (Application Loader)	NAND/NOR (COREOS)	SPE(2)	Static Encryption / Signed	Yes	Yes	Decrypt games	Yes
isoldr (Isolation Loader)	NAND/NOR (COREOS)	SPE(2)	Static Encryption / Signed	Yes	No	Decrypt modules	Yes
rvkldr (Revokation Loader) (Discarded after 0.8)	NAND/NOR (COREOS)	SPE(2)	Static Encryption / Signed	Yes	No	Decrypt revoke list	Yes

* : ofcourse with new hardware revisions, it is updated in factory. See Flash#new_metldr.2

Chain of trust Diagram

Runtime Secure

This runtime secure boot, in fact, is tightly coupled with an SPE entering isolation mode. An application must go through the hardware authentication step before it can execute on an isolated SPE. When isolation mode is requested, first, the previous thread is stopped and cancelled. Then, the hardware will automatically start fetching the application into the LS, and the hardware will verify the integrity of the application. If the integrity check fails, the application will not be executed. The check can fail for one of two reasons. The application might have been modified within memory or storage. Then, the assumption is that the functionality might have changed and it cannot be trusted anymore. Or, the writer of the application does not know the cryptographic secret that is needed for a successful authentication. Otherwise, if the authentication check is successful, the hardware will automatically kick-start the application's execution in isolation mode. Because the hardware controls all of these steps, the verification of the application's integrity cannot be skipped or manipulated and will happen consistently and correctly.

• http://www.ibm.com/developerworks/power/library/pa-cellsecurity/

Changes in firmware 3.60

Lv0 has now been changed, LV0 now appears to encapsulate all of the Loaders (appldr, isoldr, lv1ldr, lv2ldr). Now in order to break the chain of trust we need to be able to decrypt/exploit LV0 (or bootldr which loads LV0) -> done! see http://www.ps3devwiki.com/wiki/Keys#lv0

Chain of trust Diagram 3.60++

LV0 with encapsulated loaders (appldr, isoldr, lv1ldr, lv2ldr).)

not in this diagram: the added .2 metadata