Difference between revisions of "SCECAF"

From Vita Development Wiki
Jump to navigation Jump to search
Line 84: Line 84:
 
== Kernel Coredump Encrypted ELF ==
 
== Kernel Coredump Encrypted ELF ==
  
On FW 2.12 and greater, a SceKernelProcess coredump can be created when a critical error (such as a DABT or PABT) occurs in a syscall or kernel thread. On reboot, the data is encrypted and compressed prior to being wrapped in the SCECAF container.
+
On FW 2.12 and greater, a SceKernelProcess coredump can be created when a critical error (such as a DABT or PABT) occurs in a syscall or kernel thread. On reboot, the data is encrypted and compressed prior to being wrapped in a SCECAF container.
  
 
=== Process ===
 
=== Process ===
  
# The exception handler in the kernel calls [[SMC]] 0x122: int smc_intr_crash(int type, void *pTTBR0, void *pVBAR, void *pSysbase);
+
# The exception handler in the kernel ([[SceExcpmgr]]) calls [[SMC]] 0x122.
# [[TrustZone]] configures Pervasive device and SceGrab device. Need why still?
+
# [[TrustZone]] configures [[Pervasive]] device and [[SceGrab]] device. Need why still?
 
# Reset SceDmac5Reg Device
 
# Reset SceDmac5Reg Device
 
# Create a SceCrashDumpHeader at PA 0x58000000 on retail and 0x60000000 on DevKit (maybe only in DevKit Memory Size mode)
 
# Create a SceCrashDumpHeader at PA 0x58000000 on retail and 0x60000000 on DevKit (maybe only in DevKit Memory Size mode)
Line 96: Line 96:
 
# Configure SceDmac5Reg and loop through all of DRAM at PA 0x40200000 in 0x10000 byte chunks
 
# Configure SceDmac5Reg and loop through all of DRAM at PA 0x40200000 in 0x10000 byte chunks
 
# Call [[SMC]] 0x11A to initiate a reboot in requested coredump mode
 
# Call [[SMC]] 0x11A to initiate a reboot in requested coredump mode
# Upon reboot, coredump handler will create a SCECAF container of the data from PA 0x58000000 or 0x60000000
+
# Upon reboot, coredump handler ([[SceCrashDump]]) will create a SCECAF container of the data from PA 0x58000000 or 0x60000000

Revision as of 18:55, 12 February 2021

Files from coredumps or crash reports are in SCECAF format when encrypted. The file names will be either crash_report.caf or a coredump with a filetype of .spsp2dmp. All data is in little endian format.

SCECAF Container File Structure

Structure:

  • Header
  • Repeating Section Meta Blocks
  • Repeating Section Hash Blocks
  • Header Hash
  • Sections

Header

Offset Size Description
0x0 0x8 0x5343454341460000 magic SCECAF\0\0
0x8 0x8 Type/Version? (always 2)
0x10 0x8 HMAC Key ID (always 1)
0x18 0x8 Number of sections
0x20 0x8 Size of Header
0x28 0x8 Total Sections Size

Section Meta Block

Offset Size Description
0x0 0x8 Section ID (starts with 0)
0x8 0x8 Section Start Offset
0x10 0x8 Section Length
0x18 0x8 HMAC Key ID (Always 1)
0x20 0x8 Encryption Key ID (Always 2)
0x28 0x10 Encryption IV
0x38 0x8 Filler

Section Hash Block

Offset Size Description
0x0 0x8 Section ID (starts with 0)
0x8 0x20 HMAC Hash
0x28 0x8 Filler

Header Hash Block

Offset Size Description
0x0 0x20 HMAC Hash

Sections

Sections are data (normally Gzipped ELF core or CPAD padding) that is encrypted. The encryption is AES 128 in CBC mode. The IV comes from the Section Meta Header. The HMAC is the SCE-modified HMAC SHA256.

The current theory is that the key to use is specified in the Section Meta Header, though since there are no sample that vary (same ID and same key), it is just a guess.

Kernel Coredump Encrypted ELF

On FW 2.12 and greater, a SceKernelProcess coredump can be created when a critical error (such as a DABT or PABT) occurs in a syscall or kernel thread. On reboot, the data is encrypted and compressed prior to being wrapped in a SCECAF container.

Process

  1. The exception handler in the kernel (SceExcpmgr) calls SMC 0x122.
  2. TrustZone configures Pervasive device and SceGrab device. Need why still?
  3. Reset SceDmac5Reg Device
  4. Create a SceCrashDumpHeader at PA 0x58000000 on retail and 0x60000000 on DevKit (maybe only in DevKit Memory Size mode)
  5. Map SceCrashDumpZero to start of DRAM at PA 0x40000000
  6. Configure SceDmac5Reg and loop through all of VRAM at PA 0x20000000 in 0x1000 byte chunks
  7. Configure SceDmac5Reg and loop through all of DRAM at PA 0x40200000 in 0x10000 byte chunks
  8. Call SMC 0x11A to initiate a reboot in requested coredump mode
  9. Upon reboot, coredump handler (SceCrashDump) will create a SCECAF container of the data from PA 0x58000000 or 0x60000000