Difference between revisions of "Cmep"

From Vita Development Wiki
Jump to navigation Jump to search
(39 intermediate revisions by 4 users not shown)
Line 1: Line 1:
This processor is hypothesized to perform most of the cryptography tasks including storing and handing of [[Keys|keys]]. There is little information about it though. The [[F00D Processor]] (named after the <code>e_machine</code> field of the ELF headers) is likely a custom [http://en.wikipedia.org/wiki/Media-embedded_processor Toshiba MeP] core.
+
Cmep, also named Cry, is a processor in the PS Vita responsible for most of the cryptography tasks including storing and handing of the most secure keys. Cmep was named "F00D" by Team Molecule after the <code>e_machine</code> field of the ELF headers, but its official name is CMeP (Crypto MeP). Cmep is a [http://en.wikipedia.org/wiki/Media-embedded_processor Toshiba MeP-c5] core.
  
== Communication ==
+
More information can be found in Toshiba documentations. For example: https://toshiba.semicon-storage.com/info/docget.jsp?did=14997&prodName=TMPV7504XBG.
Communication seems to go through some sort of FIFO register.
 
  
=== Write ===
+
== Devices ==
To write, put the double word into <code>0xE0000010</code>. Next read <code>0xE0000010</code> until it returns 0, which indicates the data was read by the F00D processor.
 
  
=== Read ===
+
See [[Physical_Memory#Cmep registers]] and [[Cmep registers]] for physical memory mapping.
To read, get a double word from <code>0xE0000000</code>. If it returns 0, no data is available. Otherwise, acknowledge that the data has been read by putting the same data into <code>0xE0000000</code>.
 
  
=== Extra ports ===
+
== Reset ==
In addition to the 0xE0000000 and 0xE0000010, the communication with F00D seems to happen with other ports too.
+
 
 +
Although the MeP architecture documentations specify that with EVM=0 the reset/NMI vector base is at 0x00000000, it is observed that the vector base is actually at 0x00040000. However, EVA/IVA still works as expected when EVM=1. Both [[Secure Kernel]] and [[Second Loader]] set EVM=0 at the start. This is likely a modified hardware behavior and the vector base remapping might be done when the [[First Loader]] is unmapped.
 +
 
 +
== Protocol ==
 +
 
 +
A 32-bit command buffer is defined below. The command is sent to Cmep with the method listed above.
  
 
{| class='wikitable'
 
{| class='wikitable'
 
|-
 
|-
! Port
+
! Bit End
! Read
+
! Bit Start
! Write
+
! Name
 +
! Description
 +
|-
 +
| 31
 +
| 16
 +
| Size
 +
| Size of the command buffer
 +
|-
 +
| 18
 +
| 13
 +
| -
 +
| Always 0
 +
|-
 +
| 12
 +
| 8
 +
| ID
 +
| [[#Command ID|Command ID]]
 +
|-
 +
| 7
 +
| 1
 +
| -
 +
| Always 0
 
|-
 
|-
| 0xE0000004
+
| 0
| YES
+
| 0
| ?
+
| Valid
 +
| Set to 1
 +
|-
 +
|}
 +
 
 +
=== Command ID ===
 +
 
 +
There are a total of 17 commands. Commands pass data through the 0x100 shared buffer that is setup in command 0.
 +
 
 +
==== Command 0 ====
 +
 
 +
This command is used to setup a shared buffer with Cmep. Firstly, Cmep will write in <code>0xE0000000</code> either 0x200 or 0x600 (and if an error happened, 0x400?). If the value 0x600 is written, it will setup a buffer that contains the secure_kernel.enp. It will write 1 to <code>0xE0010000</code> and after that 0x0 to <code>0xE0010000</code>, next, it will wait <code>0xE0010000</code> return 0 and <code>0xE0010004</code> return a value <= 0x0. If everything goes alright, it will clear the low 2 bits of the secure_kernel.enp address (normally at 0x1F850000) and OR it with 0x1 and then write it to <code>0xE0000010</code>. After it, the process is the same that the 0x200. If the value 0x200 is written, it will set the 0x100-sized shared buffer. The physical address of the buffer is written to <code>0xE0000010</code> and then command 0x0 is written.
 +
 
 +
==== Command 1 ====
 +
 
 +
This command is called by a subroutine of [[SceSblSmsched#sceSblSmSchedStopForTZS]] or [[SceSblSmsched]] module_start (if an error happened while [[SceSblSmsched]] was initializing). This command seems to be used to ask Cmep to remove all allocated resources (like shared memory address, loaded Secure Module, etc), and probably power off or reset it.
 +
 
 +
==== Command 2 ====
 +
 
 +
Load a [[Secure Modules|Secure Module]].
 +
 
 +
{| class='wikitable'
 
|-
 
|-
| 0xE0000008
+
! Offset
| YES
+
! Size
| ?
+
! Description
 
|-
 
|-
| 0xE000000C
+
| 0x0
| YES
+
| 0x4
| ?
+
| num_paddrs
 
|-
 
|-
| 0xE0000014
+
| 0x4
| YES
+
| 0x4
| YES
+
| paddr_list for [[Secure Modules|Secure Module]] ELF
 
|-
 
|-
| 0xE0000018
+
| 0x8
| YES
+
| 0x4
| YES
+
| buf_0x40: some 0x40 buffer
 
|-
 
|-
| 0xE000001C
+
| 0xC
| YES
+
| 0x4
| YES
+
| ctx_0x4
 
|-
 
|-
| 0xE0000054
+
| 0x10
| ?
+
| 0x4
| YES
+
| ctx_0x8
 
|-
 
|-
| 0xE0000058
+
| 0x14
| ?
+
| 0x4
| YES
+
| ctx_0xC
 
|-
 
|-
| 0xE000005C
+
| 0x1C
| ?
+
| 0x4
| YES
+
| field_60: system version (2)
 
|-
 
|-
 +
| 0x20
 +
| 0x4
 +
| partition ID, ?Media Type?
 
|}
 
|}
  
 +
==== Command 3 ====
 +
 +
Load the previously suspended [[Secure Modules|Secure Module]].
 +
 +
{| class='wikitable'
 +
|-
 +
! Offset
 +
! Size
 +
! Description
 +
|-
 +
| 0x0
 +
| 0x4
 +
| num_paddrs
 +
|-
 +
| 0x4
 +
| 0x4
 +
| paddr_list for suspend buffer
 +
|-
 +
| 0x8
 +
| 0x4
 +
| buf_0x40: some 0x40 buffer
 +
|-
 +
| 0xC
 +
| 0x4
 +
| delayed_cmd
 +
|}
  
 +
==== Command 4 ====
  
== Protocol ==
+
Suspend the current [[Secure Modules|Secure Module]].
A 32-bit command buffer is defined below. The command is sent to the F00D processor with the method listed above.
 
  
 
{| class='wikitable'
 
{| class='wikitable'
 
|-
 
|-
! Bit End
+
! Offset
! Bit Start
+
! Size
! Name
 
 
! Description
 
! Description
 
|-
 
|-
| 31
+
| 0x0
| 23
+
| 0x4
| ?
+
| num_paddrs
| ?
+
|-
 +
| 0x4
 +
| 0x4
 +
| paddr_list for suspend buffer
 
|-
 
|-
| 22
+
| 0x8
| 22
+
| 0x4
| ?
+
| buf_0x40: some 0x40 buffer
| ?
 
 
|-
 
|-
| 21
+
| 0xC
| 21
+
| 0x4
| ?
+
| delayed_cmd
| ?
+
|}
 +
 
 +
==== Command 5 ====
 +
 
 +
Kill the current [[Secure Modules|Secure Module]].
 +
 
 +
==== Command 6 ====
 +
 
 +
Force stop? Called from [[SceSblSmsched|sceSblSmSchedDeleteAll]].
 +
 
 +
==== Command 7 ====
 +
 
 +
Unused.
 +
 
 +
==== Command 8 ====
 +
 
 +
Unused.
 +
 
 +
==== Command 9 ====
 +
 
 +
Set a 0x80 sized shared buffer.
 +
 
 +
This seems like a SceKernelPARange.
 +
 
 +
{| class='wikitable'
 
|-
 
|-
| 20
+
! Offset
| 18
+
! Size
| flag?
+
! Description
| ?
 
 
|-
 
|-
| 17
+
| 0x0
| 13
+
| 0x4
| ?
+
| PA
| Always 0
 
 
|-
 
|-
| 12
+
| 0x4
| 8
+
| 0x4
| ID
+
| length
| [[F00D Processor#Command ID|Command ID]]
+
|}
 +
 
 +
==== Command 10 ====
 +
 
 +
Set the encrypted signed revoke list.
 +
 
 +
{| class='wikitable'
 
|-
 
|-
| 7
+
! Offset
| 1
+
! Size
| ?
+
! Description
| Always 0
 
 
|-
 
|-
| 0
+
| 0x0
| 0
+
| 0x4
| Valid
+
| num_paddrs
| Set 1 to indicate command is valid
 
 
|-
 
|-
 +
| 0x4
 +
| 0x4
 +
| paddr_list for signed revoke list
 
|}
 
|}
  
=== Command ID ===
+
==== Command 11 ====
Below are notes on different commands.
+
 
 +
==== Command 12 ====
 +
 
 +
==== Command 13 ====
 +
 
 +
==== Command 14 ====
 +
 
 +
==== Command 15 ====
 +
 
 +
==== Command 16 ====
 +
 
 +
=== Interrupts ===
 +
 
 +
There are 4 interrupts that Cmep sends. There are only two interrupt handles, however.
 +
 
 +
* <code>cry2arm0</code>: handles interrupt 200. This is notifications about current [[Secure Modules|Secure Module]] status changes (loaded, unloaded, suspended, etc.)
 +
* <code>cry2arm123</code>: handles interrupts 201, 202, 203. This interrupt is sent once an [[Secure Modules|Secure Module]] function has finished processing (see below).
 +
 
 +
=== Executing Secure Module functions ===
 +
 
 +
You can find an open source ARM TrustZone<->Cmep protocol implementation at https://github.com/xyzz/f00d. It allows calling [[Secure Modules|Secure Modules]] commands with a maximum of control over the sent data, received data and timing.
  
==== 0x0 ====
+
When a [[Secure Modules|Secure Module]] is loaded, you can execute [[Secure Modules Functions|functions]] it provides by writing into registers 0xE0000014, 0xE0000018, 0xE000001C the physical address of the [[Secure Modules Functions#Request_Buffer|request buffer]] OR'd with 1.
Seems to be used to set the 0x100 sized shared buffer. First the physical address of the buffer is written to <code>0xE0000010</code> and then command 0x0 is written.
 
  
==== 0x1 ====
+
Once result is available, ARM will get an interrupt 201-203. Read status code from 0xE0000004 or 0xE0000008 or 0xE000000C and confirm you received it by writing it back to the same register.
May be used to reset F00D processor.
 
  
==== 0x9 ====
+
Possible results:
Seems to be used to set a 0x80 sized shared buffer.
 
  
==== 0xA ====
+
* 1: command executed successfully
Seems to set the [[SCE]] encrypted revocation list.
+
* 3: invalid function ID
 +
* 5: invalid function buffer physical address (e.g. trying to pass secure memory)
 +
* 9: SCE_SBL_ERROR_COMMON_EIO
  
 
== Memory ==
 
== Memory ==
  
<code>kprx_auth_sm.self</code> is allowed access to <code>0x1F000000</code>, <code>0x1F840000</code>, <code>0x20000000</code>, and <code>0x40300000</code>. The address checks is likely done in software. F00D has it's own private 128KB memory from <code>0x00800000</code> to <code>0x00820000</code>. F00D SELFs are typically loaded to <code>0x0080B000</code>.
+
Cmep has its own private 128kB memory from <code>0x00800000</code> to <code>0x00820000</code>. [[Secure Kernel]] is loaded to <code>0x00800000</code>. [[Secure Modules]] are typically loaded to <code>0x0080B000</code>.
 +
 
 +
== Secure Modules Task states ==
 +
 
 +
ARM [[TrustZone]] implements a scheduler for [[Secure Modules]] tasks. Normally, only one task can run on Cmep at a time. The external [[Secure Modules]] scheduler running on ARM allows to have multiple tasks at once. The following provides an explanation of various states the task can be in.
 +
 
 +
{| class='wikitable'
 +
|-
 +
! ID
 +
! Description
 +
|-
 +
| 1
 +
| Created / Suspended (Unloaded)
 +
|-
 +
| 2
 +
| Running
 +
|-
 +
| 3
 +
| Errored (Unloaded)
 +
|-
 +
| 4
 +
| ? (Unloaded)
 +
|-
 +
| 5
 +
| ? (Unloaded)
 +
|-
 +
| 6
 +
| Start/resume requested
 +
|-
 +
| 7
 +
| Suspending
 +
|-
 +
| 8
 +
| ?
 +
|-
 +
| 9
 +
| ?
 +
|-
 +
| 10
 +
| ?
 +
|-
 +
| 11
 +
| Suspend requested
 +
|-
 +
| 12
 +
| ?
 +
|}
 +
 
 +
== Related pages ==
 +
 
 +
* [[Cmep basics]]
 +
* [[Cmep registers]]
 +
* [[F00D Key Ring Base]]
 +
* [[Secure Block]]
 +
* [[First Loader]]
 +
* [[Second Loader]]
 +
* [[Secure Kernel]]
 +
* [[Secure Modules]]
 +
* [[Secure Modules Functions]]
 +
* [[Ernie Secure]]

Revision as of 21:19, 15 February 2023

Cmep, also named Cry, is a processor in the PS Vita responsible for most of the cryptography tasks including storing and handing of the most secure keys. Cmep was named "F00D" by Team Molecule after the e_machine field of the ELF headers, but its official name is CMeP (Crypto MeP). Cmep is a Toshiba MeP-c5 core.

More information can be found in Toshiba documentations. For example: https://toshiba.semicon-storage.com/info/docget.jsp?did=14997&prodName=TMPV7504XBG.

Devices

See Physical_Memory#Cmep registers and Cmep registers for physical memory mapping.

Reset

Although the MeP architecture documentations specify that with EVM=0 the reset/NMI vector base is at 0x00000000, it is observed that the vector base is actually at 0x00040000. However, EVA/IVA still works as expected when EVM=1. Both Secure Kernel and Second Loader set EVM=0 at the start. This is likely a modified hardware behavior and the vector base remapping might be done when the First Loader is unmapped.

Protocol

A 32-bit command buffer is defined below. The command is sent to Cmep with the method listed above.

Bit End Bit Start Name Description
31 16 Size Size of the command buffer
18 13 - Always 0
12 8 ID Command ID
7 1 - Always 0
0 0 Valid Set to 1

Command ID

There are a total of 17 commands. Commands pass data through the 0x100 shared buffer that is setup in command 0.

Command 0

This command is used to setup a shared buffer with Cmep. Firstly, Cmep will write in 0xE0000000 either 0x200 or 0x600 (and if an error happened, 0x400?). If the value 0x600 is written, it will setup a buffer that contains the secure_kernel.enp. It will write 1 to 0xE0010000 and after that 0x0 to 0xE0010000, next, it will wait 0xE0010000 return 0 and 0xE0010004 return a value <= 0x0. If everything goes alright, it will clear the low 2 bits of the secure_kernel.enp address (normally at 0x1F850000) and OR it with 0x1 and then write it to 0xE0000010. After it, the process is the same that the 0x200. If the value 0x200 is written, it will set the 0x100-sized shared buffer. The physical address of the buffer is written to 0xE0000010 and then command 0x0 is written.

Command 1

This command is called by a subroutine of SceSblSmsched#sceSblSmSchedStopForTZS or SceSblSmsched module_start (if an error happened while SceSblSmsched was initializing). This command seems to be used to ask Cmep to remove all allocated resources (like shared memory address, loaded Secure Module, etc), and probably power off or reset it.

Command 2

Load a Secure Module.

Offset Size Description
0x0 0x4 num_paddrs
0x4 0x4 paddr_list for Secure Module ELF
0x8 0x4 buf_0x40: some 0x40 buffer
0xC 0x4 ctx_0x4
0x10 0x4 ctx_0x8
0x14 0x4 ctx_0xC
0x1C 0x4 field_60: system version (2)
0x20 0x4 partition ID, ?Media Type?

Command 3

Load the previously suspended Secure Module.

Offset Size Description
0x0 0x4 num_paddrs
0x4 0x4 paddr_list for suspend buffer
0x8 0x4 buf_0x40: some 0x40 buffer
0xC 0x4 delayed_cmd

Command 4

Suspend the current Secure Module.

Offset Size Description
0x0 0x4 num_paddrs
0x4 0x4 paddr_list for suspend buffer
0x8 0x4 buf_0x40: some 0x40 buffer
0xC 0x4 delayed_cmd

Command 5

Kill the current Secure Module.

Command 6

Force stop? Called from sceSblSmSchedDeleteAll.

Command 7

Unused.

Command 8

Unused.

Command 9

Set a 0x80 sized shared buffer.

This seems like a SceKernelPARange.

Offset Size Description
0x0 0x4 PA
0x4 0x4 length

Command 10

Set the encrypted signed revoke list.

Offset Size Description
0x0 0x4 num_paddrs
0x4 0x4 paddr_list for signed revoke list

Command 11

Command 12

Command 13

Command 14

Command 15

Command 16

Interrupts

There are 4 interrupts that Cmep sends. There are only two interrupt handles, however.

  • cry2arm0: handles interrupt 200. This is notifications about current Secure Module status changes (loaded, unloaded, suspended, etc.)
  • cry2arm123: handles interrupts 201, 202, 203. This interrupt is sent once an Secure Module function has finished processing (see below).

Executing Secure Module functions

You can find an open source ARM TrustZone<->Cmep protocol implementation at https://github.com/xyzz/f00d. It allows calling Secure Modules commands with a maximum of control over the sent data, received data and timing.

When a Secure Module is loaded, you can execute functions it provides by writing into registers 0xE0000014, 0xE0000018, 0xE000001C the physical address of the request buffer OR'd with 1.

Once result is available, ARM will get an interrupt 201-203. Read status code from 0xE0000004 or 0xE0000008 or 0xE000000C and confirm you received it by writing it back to the same register.

Possible results:

  • 1: command executed successfully
  • 3: invalid function ID
  • 5: invalid function buffer physical address (e.g. trying to pass secure memory)
  • 9: SCE_SBL_ERROR_COMMON_EIO

Memory

Cmep has its own private 128kB memory from 0x00800000 to 0x00820000. Secure Kernel is loaded to 0x00800000. Secure Modules are typically loaded to 0x0080B000.

Secure Modules Task states

ARM TrustZone implements a scheduler for Secure Modules tasks. Normally, only one task can run on Cmep at a time. The external Secure Modules scheduler running on ARM allows to have multiple tasks at once. The following provides an explanation of various states the task can be in.

ID Description
1 Created / Suspended (Unloaded)
2 Running
3 Errored (Unloaded)
4 ? (Unloaded)
5 ? (Unloaded)
6 Start/resume requested
7 Suspending
8 ?
9 ?
10 ?
11 Suspend requested
12 ?

Related pages