TSEC Firmware

Boot Process

TSEC is configured and initialized by the first bootloader during key generation.

[6.2.0+] TSEC is now configured at the end of the first bootloader's main function.

Initialization

During this stage several clocks are programmed.

    // Program the HOST1X clock and resets
    // Uses RST_DEVICES_L, CLK_OUT_ENB_L, CLK_SOURCE_HOST1X and CLK_L_HOST1X
    enable_host1x_clkrst();
 
    // Program the TSEC clock and resets
    // Uses RST_DEVICES_U, CLK_OUT_ENB_U, CLK_SOURCE_TSEC and CLK_U_TSEC
    enable_tsec_clkrst();
 
    // Program the SOR_SAFE clock and resets
    // Uses RST_DEVICES_Y, CLK_OUT_ENB_Y and CLK_Y_SOR_SAFE
    enable_sor_safe_clkrst();
 
    // Program the SOR0 clock and resets
    // Uses RST_DEVICES_X, CLK_OUT_ENB_X and CLK_X_SOR0
    enable_sor0_clkrst();
 
    // Program the SOR1 clock and resets
    // Uses RST_DEVICES_X, CLK_OUT_ENB_X, CLK_SOURCE_SOR1 and CLK_X_SOR1
    enable_sor1_clkrst();
 
    // Program the KFUSE clock resets
    // Uses RST_DEVICES_H, CLK_OUT_ENB_H and CLK_H_KFUSE
    enable_kfuse_clkrst();

Configuration

In this stage the Falcon IRQs, interfaces and DMA engine are configured.

    // Clear the Falcon DMA control register
    *(u32 *)FALCON_DMACTL = 0;
 
    // Enable Falcon IRQs
    *(u32 *)FALCON_IRQMSET = 0xFFF2;
 
    // Enable Falcon IRQs
    *(u32 *)FALCON_IRQDEST = 0xFFF0;
 
    // Enable Falcon interfaces
    *(u32 *)FALCON_ITFEN = 0x03;
 
    // Wait for Falcon's DMA engine to be idle
    wait_flcn_dma_idle();

Firmware loading

The Falcon firmware code is stored in the first bootloader's data segment in IMEM.

    // Set DMA transfer base address to 0x40011900 >> 0x08
    *(u32 *)FALCON_DMATRFBASE = 0x400119;
 
    u32 trf_mode = 0;     // A value of 0 sets FALCON_DMATRFCMD_IMEM
    u32 dst_offset = 0;
    u32 src_offset = 0;
 
    // Load code into Falcon (0x100 bytes at a time)
    while (src_offset < 0xF00)
    {
        flcn_load_firm(trf_mode, src_offset, dst_offset);
        src_offset += 0x100;
        dst_offset += 0x100;
    }

[6.2.0+] The transfer base address and size of the Falcon firmware code changed.

    // Set DMA transfer base address to 0x40010E00 >> 0x08
    *(u32 *)FALCON_DMATRFBASE = 0x40010E;
 
    u32 trf_mode = 0;     // A value of 0 sets FALCON_DMATRFCMD_IMEM
    u32 dst_offset = 0;
    u32 src_offset = 0;
 
    // Load code into Falcon (0x100 bytes at a time)
    while (src_offset < 0x2900)
    {
        flcn_load_firm(trf_mode, src_offset, dst_offset);
        src_offset += 0x100;
        dst_offset += 0x100;
    }

Firmware booting

Falcon is booted up and the first bootloader waits for it to finish.

   // Set magic value in host1x scratch space
   *(u32 *)0x50003300 = 0x34C2E1DA;

   // Clear Falcon scratch1 MMIO
   *(u32 *)FALCON_SCRATCH1 = 0;

   // Set Falcon boot key version in scratch0 MMIO
   *(u32 *)FALCON_SCRATCH0 = 0x01;

   // Set Falcon's boot vector address
   *(u32 *)FALCON_BOOTVEC = 0;

   // Signal Falcon's CPU
   *(u32 *)FALCON_CPUCTL = 0x02;

   // Wait for Falcon's DMA engine to be idle
   wait_flcn_dma_idle();

   u32 boot_res = 0;

   // The bootloader allows the TSEC two seconds from this point to do its job
   u32 maximum_time = read_timer() + 2000000; 

   while (!boot_res)
   {
       // Read boot result from scratch1 MMIO
       boot_res = *(u32 *)FALCON_SCRATCH1;
   
       // Read from TIMERUS_CNTR_1US (microseconds from boot)
       u32 current_time = read_timer();
   
       // Booting is taking too long
       if (current_time > maximum_time)
           panic();
   }

   // Invalid boot result was returned
   if (boot_res != 0xB0B0B0B0)
       panic();

[6.2.0+] Falcon is booted up, but the first bootloader is left in an infinite loop.

    // Set magic value in host1x scratch space
    *(u32 *)0x50003300 = 0x34C2E1DA;
 
    // Clear Falcon scratch1 MMIO
    *(u32 *)FALCON_SCRATCH1 = 0;
 
    // Set Falcon boot key version in scratch0 MMIO
    *(u32 *)FALCON_SCRATCH0 = 0x01;
 
    // Set Falcon's boot vector address
    *(u32 *)FALCON_BOOTVEC = 0;
 
    // Signal Falcon's CPU
    *(u32 *)FALCON_CPUCTL = 0x02;
 
    // Infinite loop
    while (1);

TSEC key generation

The TSEC key is generated by reading SOR1 registers modified by the Falcon CPU.

    // Clear magic value in host1x scratch space
    *(u32 *)0x50003300 = 0;
 
    // Read TSEC key
    u32 tsec_key[4]; 
    tsec_key[0] = *(u32 *)NV_SOR_DP_HDCP_BKSV_LSB;
    tsec_key[1] = *(u32 *)NV_SOR_TMDS_HDCP_BKSV_LSB;
    tsec_key[2] = *(u32 *)NV_SOR_TMDS_HDCP_CN_MSB;
    tsec_key[3] = *(u32 *)NV_SOR_TMDS_HDCP_CN_LSB;
 
    // Clear SOR1 registers
    *(u32 *)NV_SOR_DP_HDCP_BKSV_LSB = 0;
    *(u32 *)NV_SOR_TMDS_HDCP_BKSV_LSB = 0;
    *(u32 *)NV_SOR_TMDS_HDCP_CN_MSB = 0;
    *(u32 *)NV_SOR_TMDS_HDCP_CN_LSB = 0;
 
    if (out_size < 0x10)
        out_size = 0x10;
 
    // Copy back the TSEC key
    memcpy(out_buf, tsec_key, out_size);

[6.2.0+] This is now done inside an encrypted TSEC payload.

Cleanup

Clocks and resets are disabled before returning.

    // Deprogram KFUSE clock and resets
    // Uses RST_DEVICES_H, CLK_OUT_ENB_H and CLK_H_KFUSE
    disable_kfuse_clkrst();
 
    // Deprogram SOR1 clock and resets
    // Uses RST_DEVICES_X, CLK_OUT_ENB_X, CLK_SOURCE_SOR1 and CLK_X_SOR1
    disable_sor1_clkrst();
 
    // Deprogram SOR0 clock and resets
    // Uses RST_DEVICES_X, CLK_OUT_ENB_X and CLK_X_SOR0
    disable_sor0_clkrst();
 
    // Deprogram SOR_SAFE clock and resets
    // Uses RST_DEVICES_Y, CLK_OUT_ENB_Y and CLK_Y_SOR_SAFE
    disable_sor_safe_clkrst();
 
    // Deprogram TSEC clock and resets
    // Uses RST_DEVICES_U, CLK_OUT_ENB_U, CLK_SOURCE_TSEC and CLK_U_TSEC
    disable_tsec_clkrst();
 
    // Deprogram HOST1X clock and resets
    // Uses RST_DEVICES_L, CLK_OUT_ENB_L, CLK_SOURCE_HOST1X and CLK_L_HOST1X
    disable_host1x_clkrst();

TSEC Firmware

The actual code loaded into TSEC is assembled in NVIDIA's proprietary fuc5 ISA using crypto extensions. Stored inside the first bootloader, this firmware binary is split into 4 blobs (names are unofficial): Boot (unencrypted and unauthenticated code), KeygenLdr (unencrypted and authenticated code), Keygen (encrypted and authenticated code) and key data.

[6.2.0+] There are now 2 new blobs (names are unofficial): SecureBootLdr (unencrypted and unauthenticated code), SecureBoot (part unencrypted and unauthenticated code, part encrypted and authenticated code).

Firmware can be disassembled with envytools' envydis:

envydis -i tsec_fw.bin -m falcon -V fuc5 -F crypt

Note that the instruction set has variable length instructions, and the disassembler is not very good at detecting locations it should start disassembling from. One needs to disassemble multiple sub-regions and join them together.

Boot

During this stage, key data is loaded and KeygenLdr is authenticated, loaded and executed. Before returning, this stage writes back to the host (using MMIO registers) and sets the key used by the first bootloader.

[6.2.0+] During this stage, key data is loaded and execution jumps to SecureBootLdr.

Initialization

Falcon sets up it's own stack pointer.

    // Read data segment size from IO space
    u32 data_seg_size = *(u32 *)UC_CAPS;
    data_seg_size >>= 0x09;
    data_seg_size &= 0x1FF;
    data_seg_size <<= 0x08;
 
    // Set the stack pointer
    *(u32 *)sp = data_seg_size;

Main

Falcon reads the key data and then authenticates, loads and executes KeygenLdr which sets the TSEC key.

    u32 boot_base_addr = 0;
    u8 key_data_buf[0x7C];
 
    // Read the key data from memory
    u32 key_data_addr = 0x300;
    u32 key_data_size = 0x7C;
    memcpy_i2d(key_data_buf, key_data_addr, key_data_size);
 
    // Read the next code segment into boot base
    u32 blob1_addr = 0x400;
    u32 blob1_size = *(u32 *)(key_data_buf + 0x74);
    memcpy_i2d(boot_base_addr, blob1_addr, blob1_size);
 
    // Upload the next code segment into Falcon's CODE region
    u32 blob1_virt_addr = 0x300;
    bool use_secret = true;
    memcpy_d2i(blob1_virt_addr, boot_base_addr, blob1_size, blob1_virt_addr, use_secret);
 
    u32 boot_res = 0;
    bool is_done = false;
    u32 time = 0;
    bool is_blob_dec = false;
 
    while (!is_done)
    {
        if (time > 4000000)
        {
            // Write boot failed (timeout) magic to FALCON_SCRATCH1
            boot_res = 0xC0C0C0C0;
            *(u32 *)FALCON_SCRATCH1 = boot_res;
       
            break;
        }
    
        // Load key version from FALCON_SCRATCH0 (bootloader sends 0x01)
        u32 key_version = *(u32 *)FALCON_SCRATCH0;
 
        if (key_version == 0x64)
        {
            // Skip all next stages
            boot_res = 0xB0B0B0B0;
            *(u32 *)FALCON_SCRATCH1 = boot_res;
       
            break;
        }
        else
        {
            if (key_version > 0x03)
                boot_res = 0xD0D0D0D0;    // Invalid key version
            else if (key_version == 0)
                boot_res = 0xB0B0B0B0;    // No keys used
            else
            {
                u32 key_buf[0x7C];
          
                // Copy key data
                memcpy(key_buf, key_data_buf, 0x7C);
 
                u32 crypto_reg_flag = 0x00060000;
                u32 blob1_hash_addr = key_buf + 0x20; 
 
                // Set auth_addr to 0x300 and auth_size to blob1_size
                $cauth = ((blob1_size << 0x10) | (0x300 >> 0x08));

                // The next 2 xfer instructions will be overridden
                // and target changes from DMA to crypto
                cxset(0x02);
          
                // Transfer data to crypto register c6
                xdst(0, (blob1_hash_addr | crypto_reg_flag));
 	  			
                // Wait for all data loads/stores to finish
                xdwait();
          
                // Jump to KeygenLdr
                u32 keygenldr_res = exec_keygenldr(key_buf, key_version, is_blob_dec);
                is_blob_dec = true;  // Set this to prevent decrypting again
 
                // Set boot finish magic on success
                if (keygenldr_res == 0)
                    boot_res = 0xB0B0B0B0
            }
       
            // Write result to FALCON_SCRATCH1
            *(u32 *)FALCON_SCRATCH1 = boot_res;
 
            if (boot_res == 0xB0B0B0B0)
                is_done = true;
        }
 
        time++;
    }
 
    // Overwrite the TSEC key in SOR1 registers
    // This has no effect because the KeygenLdr locks out the TSEC DMA engine
    tsec_set_key(key_data_buf);
 
    return boot_res;

[6.2.0+] Falcon reads the key data and jumps to SecureBootLdr.

    u8 key_data_buf[0x84];
 
    // Read the key data from memory
    u32 key_data_addr = 0x300;
    u32 key_data_size = 0x84;
    memcpy_i2d(key_data_buf, key_data_addr, key_data_size);
 
    // Calculate the next blob's address
    u32 blob4_size = *(u32 *)(key_data_buf + 0x80);
    u32 blob0_size = *(u32 *)(key_data_buf + 0x70);
    u32 blob1_size = *(u32 *)(key_data_buf + 0x74);
    u32 blob2_size = *(u32 *)(key_data_buf + 0x78);
    u32 blob3_addr = ((((blob0_size + blob1_size) + 0x100) + blob2_size) + blob4_size);
 
    // Jump to next blob
    (void *)blob3_addr();
  
    return 0;

tsec_set_key

This method takes key_data_buf (a 16 bytes buffer) as argument and writes its contents to SOR1 registers.

    // This is TSEC_MMIO + 0x1000 + (0x1C300 / 0x40)
    *(u32 *)TSEC_DMA_TIMEOUT = 0xFFF;
 
    // Read the key's words
    u32 key0 = *(u32 *)(key_data_buf + 0x00);
    u32 key1 = *(u32 *)(key_data_buf + 0x04);
    u32 key2 = *(u32 *)(key_data_buf + 0x08);
    u32 key3 = *(u32 *)(key_data_buf + 0x0C);
 
    u32 result = 0;
 
    // Write to SOR1 register
    result = tsec_dma_write(NV_SOR_DP_HDCP_BKSV_LSB, key0);
 
    // Failed to write
    if (result)
        return result;
 
    // Write to SOR1 register
    result = tsec_dma_write(NV_SOR_TMDS_HDCP_BKSV_LSB, key1);
 
    // Failed to write
    if (result)
        return result;
 
    // Write to SOR1 register
    result = tsec_dma_write(NV_SOR_TMDS_HDCP_CN_MSB, key2);
 
    // Failed to write
    if (result)
        return result;
 
    // Write to SOR1 register
    result = tsec_dma_write(NV_SOR_TMDS_HDCP_CN_LSB, key3);
 
    // Failed to write
    if (result)
        return result;
 
    return result;

tsec_dma_write

This method takes addr and value as arguments and performs a DMA write using TSEC MMIO.

    u32 result = 0;
 
    // Wait for TSEC DMA engine
    // This waits for bit 0x0C in TSEC_DMA_CMD to be 0
    result = wait_tsec_dma();
 
    // Wait failed
    if (result)
        return 1;
 
    // Set the destination address
    // This is TSEC_MMIO + 0x1000 + (0x1C100 / 0x40)
    *(u32 *)TSEC_DMA_ADDR = addr;
 
    // Set the value
    // This is TSEC_MMIO + 0x1000 + (0x1C200 / 0x40)
    *(u32 *)TSEC_DMA_VAL = value;
 
    // Start transfer
    // This is TSEC_MMIO + 0x1000 + (0x1C000 / 0x40)
    *(u32 *)TSEC_DMA_CMD = 0x800000F2;
 
    // Wait for TSEC DMA engine
    // This waits for bit 0x0C in TSEC_DMA_CMD to be 0
    result = wait_tsec_dma();
 
    // Wait failed
    if (result)
        return 1;
 
    return 0;

KeygenLdr

This stage is responsible for reconfiguring the Falcon's crypto co-processor and loading, decrypting, authenticating and executing Keygen.

Main

    // Clear interrupt flags
    $flags.ie0 = 0;
    $flags.ie1 = 0;
    $flags.ie2 = 0;
 
    // Clear overrides
    cxset(0x80);
 
    // Clear bit 0x13 in cauth
    $cauth = ($cauth & ~(1 << 0x13));
 
    // Set the target port for memory transfers
    $xtargets = 0;
 
    // Wait for all data loads/stores to finish
    xdwait();
 
    // Wait for all code loads to finish
    xcwait();
 
    // The next 2 xfer instructions will be overridden
    // and target changes from DMA to crypto
    cxset(0x02);
 
    // Transfer data to crypto register c0
    // This should clear any leftover data
    xdst(0, 0);
 
    // Wait for all data loads/stores to finish
    xdwait();
 
    // Clear all crypto registers, except c6 which is used for auth
    cxor($c0, $c0);
    cmov($c1, $c0);
    cmov($c2, $c0);
    cmov($c3, $c0);
    cmov($c4, $c0);
    cmov($c5, $c0);
    cmov($c7, $c0);
 
    // Clear TSEC_TEGRA_CTL_TKFI_KFUSE
    // This is TSEC_MMIO + 0x1000 + (0x20E00 / 0x40)
    *(u32 *)TSEC_TEGRA_CTL &= 0xEFFFF;
 
    // Set TSEC_SCP_CTL_PKEY_REQUEST_RELOAD
    // This is TSEC_MMIO + 0x1000 + (0x10600 / 0x40)
    *(u32 *)TSEC_SCP_CTL_PKEY |= 0x01;
 
    u32 is_pkey_loaded = 0;
 
    // Wait for TSEC_SCP_CTL_PKEY_LOADED
    while (!is_pkey_loaded)
        is_pkey_loaded = (*(u32 *)TSEC_SCP_CTL_PKEY & 0x02);
 
    // Read data segment size from IO space
    u32 data_seg_size = *(u32 *)FALCON_HWCFG;
    data_seg_size >>= 0x09;
    data_seg_size &= 0x1FF;
    data_seg_size <<= 0x08;
 
    // Check stack bounds
    if (($sp >= data_seg_size) || ($sp < 0x800))
        exit();
 
    // Decrypt and load Keygen stage
    load_keygen(key_buf, key_version, is_blob_dec);
 
    // Clear the cauth signature
    csigclr();
 
    // Clear all crypto registers
    cxor($c0, $c0);
    cxor($c1, $c1);
    cxor($c2, $c2);
    cxor($c3, $c3);
    cxor($c4, $c4);
    cxor($c5, $c5);
    cxor($c6, $c6);
    cxor($c7, $c7);
 
    // Take SCP out of lockdown
    // This is TSEC_MMIO + 0x1000 + (0x10300 / 0x40)
    *(u32 *)TSEC_SCP_CTL_LOCK = 0;
 
    return;

load_keygen

    u32 res = 0;
 
    u32 boot_base_addr = 0;
    u32 blob0_addr = 0;
    u32 blob0_size = *(u32 *)(key_buf + 0x70); 
 
    // Load blob0 code again
    memcpy_i2d(boot_base_addr, blob0_addr, blob0_size);
 
    // Generate "CODE_SIG_01" key into c4 crypto register
    gen_usr_key(0, 0);
 
    // Encrypt buffer with c4
    u8 sig_key[0x10];
    enc_buf(sig_key, blob0_size);
 
    u32 src_addr = boot_base_addr;
    u32 src_size = blob0_size;
    u32 iv_addr = sig_key;
    u32 dst_addr = sig_key;
    u32 mode = 0x02;   // AES-CMAC
    u32 use_imem = 0;
 
    // Do AES-CMAC over blob0 code
    do_crypto(src_addr, src_size, iv_addr, dst_addr, mode, use_imem);
 
    // Compare the hashes
    if (memcmp(dst_addr, key_buf + 0x10, 0x10))
    {
        res = 0xDEADBEEF;
        return res;
    }
 
    u32 blob1_size = *(u32 *)(key_buf + 0x74);
 
    // Decrypt Keygen blob if needed
    if (!is_blob_dec)
    {
        // Read Stage2's size from key buffer
        u32 blob2_size = *(u32 *)(key_buf + 0x78);
 
        // Check stack bounds
        if ($sp > blob2_size)
        {
            u32 boot_base_addr = 0;
            u32 blob2_virt_addr = blob0_size + blob1_size;
            u32 blob2_addr = blob2_virt_addr + 0x100;
       
            // Read Keygen encrypted blob
            memcpy_i2d(boot_base_addr, blob2_addr, blob2_size);
 
            // Generate "CODE_ENC_01" key into c4 crypto register
            gen_usr_key(0x01, 0x01);
       
            u32 src_addr = boot_base_addr;
            u32 src_size = blob2_size;
            u32 iv_addr = key_buf + 0x40;
            u32 dst_addr = boot_base_addr;
            u32 mode = 0;   // AES-128-ECB
            u32 use_imem = 0;
       
            // Decrypt Keygen blob
            do_crypto(src_addr, src_size, iv_addr, dst_addr, mode, use_imem);
       
            // Upload the next code segment into Falcon's CODE region
            bool use_secret = true;
            memcpy_d2i(blob2_virt_addr, boot_base_addr, blob2_size, blob2_virt_addr, use_secret);
 
            // Clear out the decrypted blob
            memset(boot_base_addr, 0, blob2_size);
        }
    }

    // The next 2 xfer instructions will be overridden
    // and target changes from DMA to crypto
    cxset(0x02);
 
    u32 crypto_reg_flag = 0x00060000;
    u32 blob2_hash_addr = key_buf + 0x30;
 
    // Transfer data to crypto register c6
    xdst(0, (blob2_hash_addr | crypto_reg_flag));
 				
    // Wait for all data loads/stores to finish
    xdwait();
 
    // Save previous cauth value
    u32 cauth_old = $cauth;
 
    // Set auth_addr to blob2_virt_addr and auth_size to blob2_size
    $cauth = ((blob2_virt_addr >> 0x08) | (blob2_size << 0x10));
 
    u32 hovi_key_addr = 0;
 
    // Select next stage key
    if (key_version == 0x01)		        // Use HOVI_EKS_01
        hovi_key_addr = key_buf + 0x50;
    else if (key_version == 0x02)	        // Use HOVI_COMMON_01
        hovi_key_addr = key_buf + 0x60;
    else if (key_version == 0x03)	        // Use debug key (empty)
        hovi_key_addr = key_buf + 0x00;
    else
        res = 0xD0D0D0D0
 	
    // Jump to Keygen
    if (hovi_key_addr)
        res = exec_keygen(hovi_key_addr, key_version);
          
    // Clear out key data
    memset(key_buf, 0, 0x7C);
 
    // Restore previous cauth value
    $cauth = cauth_old;
 
    return res;

gen_usr_key

This method takes type and mode as arguments and generates a key.

    u8 seed_buf[0x10];
 
    // Read a 16 bytes seed based on supplied type
    /*
        Type 0: "CODE_SIG_01" + null padding
        Type 1: "CODE_ENC_01" + null padding
    */
    get_seed(seed_buf, type);
 
    // This will write the seed into crypto register c0 
    crypto_store(0, seed_buf);
 
    // Load selected secret into crypto register c1
    csecret($c1, 0x26);
 
    // Bind c1 register as the key for enc/dec operations
    ckeyreg($c1);
 
    // Encrypt seed_buf in c0 using keyreg value as key into c1
    cenc($c1, $c0);
 
    // Encrypt the auth signature (stored in c6) with c1 and store in c1
    csigenc($c1, $c1);
 
    // Copy the result to c4 (will be used as key)
    cmov($c4, $c1);
 
    // Do key expansion for decryption if necessary
    if (mode != 0)
        ckexp($c4, $c4);
 
    return;

enc_buffer

This method takes buf (a 16 bytes buffer) and size as arguments and encrypts the supplied buffer.

    // Set first 3 words to null
    *(u32 *)(buf + 0x00) = 0;
    *(u32 *)(buf + 0x04) = 0;
    *(u32 *)(buf + 0x08) = 0;
 
    // Swap halves (b16, b32 and b16 again)
    hswap(size);
 
    // Store the size as the last word
    *(u32 *)(buf + 0x0C) = size;
 
    // This will write buf into crypto register c3 
    crypto_store(0x03, buf);
 
    // Bind c4 register as the key for enc/dec operations
    ckeyreg($c4);
 
    // Encrypt buf in c3 using keyreg value as key and store in c5
    cenc($c5, $c3);
 
    // This will read into buf from crypto register c5 
    crypto_load(0x05, buf);
 
    return;

do_crypto

This is the method responsible for all crypto operations performed during KeygenLdr. It takes src_addr, src_size, iv_addr, dst_addr, mode and use_imem as arguments.

    // Check for invalid source data size
    if (!src_size || (src_size & 0x0F))
        exit();
 
    // Check for invalid source data address
    if (src_addr & 0x0F)
        exit();
 
    // Check for invalid destination data address
    if (dst_addr & 0x0F)
        exit();
 
    // Use IV if available
    if (iv_addr)
    {
        // This will write the iv_addr into crypto register c5 
        crypto_store(0x05, iv_addr);
    }
    else
    {
        // Clear c5 register (use null IV)
        cxor($c5, $c5);
    }
 
    // Bind c4 register as the key for enc/dec operations
    ckeyreg(c4);
 
    // AES-128-CBC decrypt
    if (mode == 0x00)
    {
        // Create crypto script with 5 instructions
        cs0begin(0x05);
 	
        cxsin($c3);                   // Read 0x10 bytes from crypto stream into c3
        cdec($c2, $c3);               // Decrypt from c3 into c2
        cxor($c5, $c2);               // XOR c2 with c5 and store in c5
        cxsout($c5);                  // Write 0x10 bytes into crypto stream from c5
        cmov($c5, $c3);               // Move c3 into c5
    }
    else if (mode == 0x01)	      // AES-128-CBC encrypt
    {
        // Create crypto script with 4 instructions
        cs0begin(0x04);
 	
        cxsin($c3);                   // Read 0x10 bytes from crypto stream into c3
        cxor($c3, $c5);               // XOR c5 with c3 and store in c3
        cenc($c5, $c3);               // Encrypt from c3 into c5
        cxsout($c5);                  // Write 0x10 bytes into crypto stream from c5
    }
    else if (mode == 0x02)	      // AES-CMAC
    {
        // Create crypto script with 3 instructions
        cs0begin(0x03);
 	
        cxsin($c3);                   // Read 0x10 bytes from crypto stream into c3
        cxor($c5, $c3);               // XOR c5 with c3 and store in c3
        cenc($c5, $c5);               // Encrypt from c5 into c5
    }
    else if (mode == 0x03)	      // AES-128-ECB decrypt
    {
        // Create crypto script with 3 instructions
        cs0begin(0x03);
 	
        cxsin($c3);                   // Read 0x10 bytes from crypto stream into c3
        cdec($c5, $c3);               // Decrypt from c3 into c5
        cxsout($c5);                  // Write 0x10 bytes into crypto stream from c5
    }
    else if (mode == 0x04)	      // AES-128-ECB encrypt
    {
        // Create crypto script with 3 instructions
        cs0begin(0x03);
 	
        cxsin($c3);                   // Read 0x10 bytes from crypto stream into c3
        cenc($c5, $c3);               // Encrypt from c3 into c5
        cxsout($c5);                  // Write 0x10 bytes into crypto stream from c5
    }
    else
        return;
 
    // Main loop
    while (src_size > 0)
    {
        u32 blk_count = (src_size >> 0x04);
 	
        if (blk_count > 0x10)
            blk_count = 0x10;
   
        // Check size align
        if (blk_count & (blk_count - 0x01))
            blk_count = 0x01;
 
        u32 blk_size = (blk_count << 0x04);
   
        u32 crypto_xfer_src = 0;
        u32 crypto_xfer_dst = 0;
   
        if (block_size == 0x20)
        {
            crypto_xfer_src = (0x00030000 | src_addr);
            crypto_xfer_dst = (0x00030000 | dst_addr);
      
            // Execute crypto script 2 times (1 for each block)
            cs0exec(0x02);
        }
        else if (block_size == 0x40)
        {
            crypto_xfer_src = (0x00040000 | src_addr);
            crypto_xfer_dst = (0x00040000 | dst_addr);
      
            // Execute crypto script 4 times (1 for each block)
            cs0exec(0x04);
        }
        else if (block_size == 0x80)
        {
            crypto_xfer_src = (0x00050000 | src_addr);
            crypto_xfer_dst = (0x00050000 | dst_addr);
      
            // Execute crypto script 8 times (1 for each block)
            cs0exec(0x08);
        }
        else if (block_size == 0x100)
        {
            crypto_xfer_src = (0x00060000 | src_addr);
            crypto_xfer_dst = (0x00060000 | dst_addr);
      
            // Execute crypto script 16 times (1 for each block)
            cs0exec(0x10);
        }
        else
        {
            crypto_xfer_src = (0x00020000 | src_addr);
            crypto_xfer_dst = (0x00020000 | dst_addr);
      
            // Execute crypto script 1 time (1 for each block)
            cs0exec(0x01);
 
            // Ensure proper block size
            block_size = 0x10;
        }
 
        // The next xfer instruction will be overridden
        // and target changes from DMA to crypto input/output stream
        if (use_imem)
            cxset(0xA1);         // Flag 0xA0 is falcon imem <-> crypto input/output stream
        else
            cxset(0x21);         // Flag 0x20 is external mem <-> crypto input/output stream
 
        // Transfer data into the crypto input/output stream
        xdst(crypto_xfer_src, crypto_xfer_src);
   
        // AES-CMAC only needs one more xfer instruction
        if (mode == 0x02)
        {
            // The next xfer instruction will be overridden
            // and target changes from DMA to crypto input/output stream
            if (use_imem)
                cxset(0xA1);     // Flag 0xA0 is falcon imem <-> crypto input/output stream
            else
                cxset(0x21);     // Flag 0x20 is external mem <-> crypto input/output stream
 		
            // Wait for all data loads/stores to finish
            xdwait();
        }
        else  // AES enc/dec needs 2 more xfer instructions
        {
            // The next 2 xfer instructions will be overridden
            // and target changes from DMA to crypto input/output stream
            if (use_imem)
                cxset(0xA2);            // Flag 0xA0 is falcon imem <-> crypto input/output stream
            else
                cxset(0x22);            // Flag 0x20 is external mem <-> crypto input/output stream
 
            // Transfer data from the crypto input/output stream
            xdld(crypto_xfer_dst, crypto_xfer_dst);
 		
            // Wait for all data loads/stores to finish
            xdwait();
 
            // Increase the destination address by block size
            dst_addr += block_size;
        }
   
        // Increase the source address by block size
        src_addr += block_size;
 
        // Decrease the source size by block size
        src_size -= block_size;
    }
 
    // AES-CMAC result is in c5
    if (mode == 0x02)
    {
        // This will read into dst_addr from crypto register c5 
        crypto_load(0x05, dst_addr);
    }
 
    return;

Keygen

This stage is decrypted by KeygenLdr using a key generated by encrypting a seed with an hardware secret. It will generate the final TSEC key.

Main

The main function takes key_addr and key_type as arguments from KeygenLdr.

    u32 flcn_version = *(u32 *)FALCON_HWCFG2 & 0x0F;

    // Falcon hardware version must be 5
    if (flcn_version != 0x05)
        exit();
 
    // Clear interrupt flags
    $flags.ie0 = 0;
    $flags.ie1 = 0;
    $flags.ie2 = 0;
 
    // Set the target port for memory transfers
    $xtargets = 0;
 
    // Generate the TSEC key
    gen_tsec_key(key_addr, key_type);
 
    // Clear the cauth signature
    csigclr();

    // Clear all crypto registers
    cxor($c0, $c0);
    cxor($c1, $c1);
    cxor($c2, $c2);
    cxor($c3, $c3);
    cxor($c4, $c4);
    cxor($c5, $c5);
    cxor($c6, $c6);
    cxor($c7, $c7);

    return;

SecureBootLdr

[6.2.0+] This was introduced to try to recover the secure boot from the RCM vulnerability.

This stage starts by authenticating and executing KeygenLdr which in turn authenticates, decrypts and executes Keygen (both blobs remain unchanged from previous firmware versions). After the TSEC key has been generated, execution returns to this stage which then parses and executes SecureBoot.

Main

    u8 key_data_buf[0x84];
    u8 tmp_key_data_buf[0x84];
 
    // Read the key data from memory
    u32 key_data_addr = 0x300;
    u32 key_data_size = 0x84;
    memcpy_i2d(key_data_buf, key_data_addr, key_data_size);
 
    // Read the KeygenLdr blob from memory
    u32 boot_base_addr = 0;
    u32 blob1_addr = 0x400;
    u32 blob1_size = *(u32 *)(key_data_buf + 0x74);
    memcpy_i2d(boot_base_addr, blob1_addr, blob1_size);
  
    // Upload the next code segment into Falcon's CODE region
    u32 blob1_virt_addr = 0x300;
    bool use_secret = true;
    memcpy_d2i(blob1_virt_addr, boot_base_addr, blob1_size, blob1_virt_addr, use_secret);
 
    // Backup the key data
    memcpy(tmp_key_data_buf, key_data_buf, 0x84);
 
    // Save previous cauth value
    u32 cauth_old = $cauth;
 
    // Set auth_addr to 0x300 and auth_size to blob1_size
    $cauth = ((blob1_size << 0x10) | (0x300 >> 0x08));
 
    // The next 2 xfer instructions will be overridden
    // and target changes from DMA to crypto
    cxset(0x02);
 
    u32 crypto_reg_flag = 0x00060000;
    u32 blob1_hash_addr = tmp_key_data_buf + 0x20; 
 
    // Transfer data to crypto register c6
    xdst(0, (blob1_hash_addr | crypto_reg_flag));
 
    // Wait for all data loads/stores to finish
    xdwait();
 
    u32 key_version = 0x01;
    bool is_blob_dec = false;
 
    // Jump to KeygenLdr
    u32 keygenldr_res = exec_keygenldr(tmp_key_data_buf, key_version, is_blob_dec);
 
    // Set boot finish magic on success
    if (keygenldr_res == 0)
        keygenldr_res = 0xB0B0B0B0
       
    // Write result to FALCON_SCRATCH1
    *(u32 *)FALCON_SCRATCH1 = keygenldr_res;
 
    if (keygenldr_res != 0xB0B0B0B0)
        return keygenldr_res;
 
    // Restore previous cauth value
    $cauth = cauth_old;
 
    u8 flcn_hdr_buf[0x18];
    u8 flcn_os_hdr_buf[0x10];
 
    blob1_size = *(u32 *)(key_data_buf + 0x74);
    u32 blob2_size = *(u32 *)(key_data_buf + 0x78);
    u32 blob0_size = *(u32 *)(key_data_buf + 0x70);
 
    // Read the SecureBoot blob's Falcon header from memory
    u32 blob4_flcn_hdr_addr = (((blob0_size + blob1_size) + 0x100) + blob2_size);
    memcpy_i2d(flcn_hdr_buf, blob4_flcn_hdr_addr, 0x18);
 
    blob1_size = *(u32 *)(key_data_buf + 0x74);
    blob2_size = *(u32 *)(key_data_buf + 0x78);
    blob0_size = *(u32 *)(key_data_buf + 0x70);
    u32 flcn_hdr_size = *(u32 *)(flcn_hdr_buf + 0x0C);
 
    // Read the SecureBoot blob's Falcon OS header from memory
    u32 blob4_flcn_os_hdr_addr = ((((blob0_size + blob1_size) + 0x100) + blob2_size) + flcn_hdr_size);
    memcpy_i2d(flcn_os_hdr_buf, blob4_flcn_os_hdr_addr, 0x10);
 
    blob1_size = *(u32 *)(key_data_buf + 0x74);
    blob2_size = *(u32 *)(key_data_buf + 0x78);
    blob0_size = *(u32 *)(key_data_buf + 0x70);
    u32 flcn_code_hdr_size = *(u32 *)(flcn_hdr_buf + 0x10);
    u32 flcn_os_size = *(u32 *)(flcn_os_hdr_buf + 0x04);
 
    // Read the SecureBoot blob's Falcon OS image from memory
    u32 blob4_flcn_os_addr = ((((blob0_size + blob1_size) + 0x100) + blob2_size) + flcn_code_hdr_size);
    memcpy_i2d(boot_base_addr, blob4_flcn_os_hdr_addr, flcn_os_size);
 
    // Upload the SecureBoot's Falcon OS image boot stub code segment into Falcon's CODE region
    u32 blob4_flcn_os_boot_virt_addr = 0;
    u32 blob4_flcn_os_boot_size = 0x100;
    use_secret = false;
    memcpy_d2i(blob4_flcn_os_boot_virt_addr, boot_base_addr, blob4_flcn_os_boot_size, blob4_flcn_os_boot_virt_addr, use_secret);
 
    flcn_os_size = *(u32 *)(flcn_os_hdr_buf + 0x04); 
 
    // Upload the SecureBoot blob's Falcon OS encrypted image code segment into Falcon's CODE region
    u32 blob4_flcn_os_img_virt_addr = 0x100;
    u32 blob4_flcn_os_img_size = (flcn_os_size - 0x100);
    use_secret = true;
    memcpy_d2i(blob4_flcn_os_img_virt_addr, boot_base_addr + 0x100, blob4_flcn_os_img_size, blob4_flcn_os_img_virt_addr, use_secret);
 
    // Wait for all code loads to finish
    xcwait();
 
    blob1_size = *(u32 *)(key_data_buf + 0x74);
    blob2_size = *(u32 *)(key_data_buf + 0x78);
    blob0_size = *(u32 *)(key_data_buf + 0x70);
    flcn_code_hdr_size = *(u32 *)(flcn_hdr_buf + 0x10);
    u32 flcn_os_code_size = *(u32 *)(flcn_os_hdr_buf + 0x08);
 
    // Read the SecureBoot blob's falcon OS image's hash from memory
    u32 blob4_flcn_os_img_hash_addr = (((((blob0_size + blob1_size) + 0x100) + blob2_size) + flcn_code_hdr_size) + flcn_os_code_size);
    memcpy_i2d(0, blob4_flcn_os_img_hash_addr, 0x10);
 
    // Read data segment size from IO space
    u32 data_seg_size = *(u32 *)FALCON_HWCFG;
    data_seg_size >>= 0x03;
    data_seg_size &= 0x3FC0;
 
    u32 data_addr = 0x10;
 
    // Clear all data except the first 0x10 bytes (SecureBoot blob's Falcon OS image's hash)
    for (int data_word_count = 0x04; data_word_count < data_seg_size; data_word_count++)
    {
        *(u32 *)(data_addr) = 0; 
        data_addr += 0x04;
    }
 
    // Clear all crypto registers
    cxor($c0, $c0);
    cxor($c1, $c1);
    cxor($c2, $c2);
    cxor($c3, $c3);
    cxor($c4, $c4);
    cxor($c5, $c5);
    cxor($c6, $c6);
    cxor($c7, $c7);
 
    // Clear the cauth signature
    csigclr();
 
    // Jump to SecureBoot
    load_secboot();
 
    return 0xB0B0B0B0;

SecureBoot

[6.2.0+] This was introduced to try to recover the secure boot from the RCM vulnerability.

This stage prepares the stack then authenticates, decrypts and executes the SecureBoot blob's Falcon OS image.

Main

    // Read data segment size from IO space
    u32 data_seg_size = *(u32 *)FALCON_HWCFG;
    data_seg_size >>= 0x01;
    data_seg_size &= 0xFF00;
 
    // Set the stack pointer
    $sp = data_seg_size;
 
    // Jump to the SecureBoot blob's Falcon OS image boot stub
    init_secboot();
 
    // Halt execution
    exit();
 
    return;

init_secboot

    // Read the transfer base address from IO space
    u32 xfer_ext_base_addr = *(u32 *)FALCON_DMATRFBASE;
 
    // Copy transfer base address to data memory
    u32 scratch_data_addr = 0x300;
    *(u32 *)scratch_data_addr = xfer_ext_base_addr;
 
    // Set the transfer base address
    xcbase(xfer_ext_base_addr);
 
    // The next xfer instruction will be overridden
    // and target changes from DMA to crypto
    cxset(0x01);
 
    u32 crypto_reg_flag = 0x00060000;
    u32 blob4_flcn_os_img_hash_addr = 0; 
 
    // Transfer data to crypto register c6
    xdst(0, (blob4_flcn_os_img_hash_addr | crypto_reg_flag));
 
    // The next xfer instruction will be overridden
    // and target changes from DMA to crypto
    cxset(0x01);
 
    // Wait for all data loads/stores to finish
    xdwait();
 
    cmov($c7, $c6);
    cxor($c7, $c7);
 
    // Set auth_addr to 0x100, auth_size to 0x1300,
    // bit 16 (use_secret) and bit 17 (is_encrypted)
    $cauth = ((0x02 << 0x10) | (0x01 << 0x10) | (0x1300 << 0x10) | (0x100 >> 0x08));
 
    // Clear interrupt flags
    $flags.ie0 = 0;
    $flags.ie1 = 0;
 
    // Jump to the SecureBoot blob's Falcon OS image
    exec_secboot();
 
    return 0x0F0F0F0F;

Key data

Small buffer stored after the Boot blob and used across all stages.

Offset	Size	Description
0x00	0x10	Debug key (empty)
0x10	0x10	blob0 (Boot) auth hash
0x20	0x10	blob1 (KeygenLdr) auth hash
0x30	0x10	blob2 (Keygen) auth hash
0x40	0x10	blob2 (Keygen) AES IV
0x50	0x10	HOVI EKS seed
0x60	0x10	HOVI COMMON seed
0x70	0x04	blob0 (Boot) size
0x74	0x04	blob1 (KeygenLdr) size
0x78	0x04	blob2 (Keygen) size
0x7C	0x04	[6.2.0+] blob3 (SecureBootLdr) size
0x80	0x04	[6.2.0+] blob4 (SecureBoot) size