Introduction to ARM64v8
Last updated
Last updated
Learn & practice AWS Hacking:HackTricks Training AWS Red Team Expert (ARTE) Learn & practice GCP Hacking: HackTricks Training GCP Red Team Expert (GRTE)
In ARMv8 architecture, execution levels, known as Exception Levels (ELs), define the privilege level and capabilities of the execution environment. There are four exception levels, ranging from EL0 to EL3, each serving a different purpose:
EL0 - User Mode:
This is the least-privileged level and is used for executing regular application code.
Applications running at EL0 are isolated from each other and from the system software, enhancing security and stability.
EL1 - Operating System Kernel Mode:
Most operating system kernels run at this level.
EL1 has more privileges than EL0 and can access system resources, but with some restrictions to ensure system integrity.
EL2 - Hypervisor Mode:
This level is used for virtualization. A hypervisor running at EL2 can manage multiple operating systems (each in its own EL1) running on the same physical hardware.
EL2 provides features for isolation and control of the virtualized environments.
EL3 - Secure Monitor Mode:
This is the most privileged level and is often used for secure booting and trusted execution environments.
EL3 can manage and control accesses between secure and non-secure states (such as secure boot, trusted OS, etc.).
The use of these levels allows for a structured and secure way to manage different aspects of the system, from user applications to the most privileged system software. ARMv8's approach to privilege levels helps in effectively isolating different system components, thereby enhancing the security and robustness of the system.
ARM64 has 31 general-purpose registers, labeled x0 through x30. Each can store a 64-bit (8-byte) value. For operations that require only 32-bit values, the same registers can be accessed in a 32-bit mode using the names w0 through w30.
x0 to x7 - These are typically used as scratch registers and for passing parameters to subroutines.
x0 also carries the return data of a function
x8 - In the Linux kernel, x8 is used as the system call number for the svc instruction. In macOS the x16 is the one used!
x9 to x15 - More temporary registers, often used for local variables.
x16 and x17 - Intra-procedural Call Registers. Temporary registers for immediate values. They are also used for indirect function calls and PLT (Procedure Linkage Table) stubs.
x16 is used as the system call number for the svc instruction in macOS.
x18 - Platform register. It can be used as a general-purpose register, but on some platforms, this register is reserved for platform-specific uses: Pointer to current thread environment block in Windows, or to point to the currently executing task structure in linux kernel.
x19 to x28 - These are callee-saved registers. A function must preserve these registers' values for its caller, so they are stored in the stack and recovered before going back to the caller.
x29 - Frame pointer to keep track of the stack frame. When a new stack frame is created because a function is called, the x29 register is stored in the stack and the new frame pointer address is (sp address) is stored in this registry.
This register can also be used as a general-purpose registry although it's usually used as reference to local variables.
x30 or lr- Link register . It holds the return address when a BL (Branch with Link) or BLR (Branch with Link to Register) instruction is executed by storing the pc value in this register.
It could also be used like any other register.
If the current function is going to call a new function and therefore overwrite lr, it will store it in the stack at the beginning, this is the epilogue (stp x29, x30 , [sp, #-48]; mov x29, sp -> Store fp and lr, generate space and get new fp) and recover it at the end, this is the prologue (ldp x29, x30, [sp], #48; ret -> Recover fp and lr and return).
sp - Stack pointer, used to keep track of the top of the stack.
the sp value should always be kept to at least a quadword alignment or a alignment exception may occur.
pc - Program counter, which points to the next instruction. This register can only be updates through exception generations, exception returns, and branches. The only ordinary instructions that can read this register are branch with link instructions (BL, BLR) to store the pc address in lr (Link Register).
xzr - Zero register. Also called wzr in it 32-bit register form. Can be used to get the zero value easily (common operation) or to perform comparisons using subs like subs XZR, Xn, #10 storing the resulting data nowhere (in xzr).
The Wn registers are the 32bit version of the Xn register.
Moreover, there are another 32 registers of 128bit length that can be used in optimized single instruction multiple data (SIMD) operations and for performing floating-point arithmetic. These are called the Vn registers although they can also operate in 64-bit, 32-bit, 16-bit and 8-bit and then they are called Qn, Dn, Sn, Hn and Bn.
There are hundreds of system registers, also called special-purpose registers (SPRs), are used for monitoring and controlling processors behaviour.
They can only be read or set using the dedicated special instruction mrs and msr.
The special registers TPIDR_EL0 and TPIDDR_EL0 are commonly found when reversing engineering. The EL0 suffix indicates the minimal exception from which the register can be accessed (in this case EL0 is the regular exception (privilege) level regular programs runs with).
They are often used to store the base address of the thread-local storage region of memory. Usually the first one is readable and writable for programs running in EL0, but the second can be read from EL0 and written from EL1 (like kernel).
mrs x0, TPIDR_EL0 ; Read TPIDR_EL0 into x0
msr TPIDR_EL0, X0 ; Write x0 into TPIDR_EL0
PSTATE contains several process components serialized into the operating-system-visible SPSR_ELx special register, being X the permission level of the triggered exception (this allows to recover the process state when the exception ends).
These are the accessible fields:
The N, Z, C and V condition flags:
N means the operation yielded a negative result
Z means the operation yielded zero
C means the operation carried
V means the operation yielded a signed overflow:
The sum of two positive numbers yields a negative result.
The sum of two negative numbers yields a positive result.
In subtraction, when a large negative number is subtracted from a smaller positive number (or vice versa), and the result cannot be represented within the range of the given bit size.
Obviously the processor doesn't now the operation is signed or not, so it will check C and V in the operations and indicate of a carry occurred in case it was signed or unsigned.
Not all the instructions update these flags. Some like CMP or TST do, and others that have an s suffix like ADDS also do it.
The current register width (nRW) flag: If the flag holds the value 0, the program will run in the AArch64 execution state once resumed.
The current Exception Level (EL): A regular program running in EL0 will have the value 0
The single stepping flag (SS): Used by debuggers to single step by setting the SS flag to 1 inside SPSR_ELx through an exception. The program will run a step and issue a single step exception.
The illegal exception state flag (IL): It's used to mark when a privileged software performs an invalid exception level transfer, this flag is set to 1 and the processor triggers an illegal state exception.
The DAIF flags: These flags allow a privileged program to selectively mask certain external exceptions.
If A is 1 it means asynchronous aborts will be triggered. The I configures to respond to external hardware Interrupts Requests (IRQs). and the F is related to Fast Interrupt Requests (FIRs).
The stack pointer select flags (SPS): Privileged programs running in EL1 and above can swap between using their own stack pointer register and the user-model one (e.g. between SP_EL1 and EL0). This switching is performed by writing to the SPSel special register. This cannot be done from EL0.
The ARM64 calling convention specifies that the first eight parameters to a function are passed in registers x0 through x7. Additional parameters are passed on the stack. The return value is passed back in register x0, or in x1 as well if its 128 bits long. The x19 to x30 and sp registers must be preserved across function calls.
When reading a function in assembly, look for the function prologue and epilogue. The prologue usually involves saving the frame pointer (x29), setting up a new frame pointer, and allocating stack space. The epilogue usually involves restoring the saved frame pointer and returning from the function.
Swift have its own calling convention that can be found in https://github.com/apple/swift/blob/main/docs/ABI/CallConvSummary.rst#arm64
ARM64 instructions generally have the format opcode dst, src1, src2, where opcode is the operation to be performed (such as add, sub, mov, etc.), dst is the destination register where the result will be stored, and src1 and src2 are the source registers. Immediate values can also be used in place of source registers.
mov: Move a value from one register to another.
Example: mov x0, x1 — This moves the value from x1 to x0.
ldr: Load a value from memory into a register.
Example: ldr x0, [x1] — This loads a value from the memory location pointed to by x1 into x0.
Offset mode: An offset affecting the orin pointer is indicated, for example:
ldr x2, [x1, #8], this will load in x2 the value from x1 + 8
ldr x2, [x0, x1, lsl #2], this will load in x2 an object from the array x0, from the position x1 (index) * 4
Pre-indexed mode: This will apply calculations to the origin, get the result and also store the new origin in the origin.
ldr x2, [x1, #8]!, this will load x1 + 8 in x2 and store in x1 the result of x1 + 8
str lr, [sp, #-4]!, Store the link register in sp and update the register sp
Post-index mode: This is like the previous one but the memory address is accessed and then the offset is calculated and stored.
ldr x0, [x1], #8, load x1 in x0 and update x1 with x1 + 8
PC-relative addressing: In this case the address to load is calculated relative to the PC register
ldr x1, =_start, This will load the address where the _start symbol starts in x1 related to the current PC.
str: Store a value from a register into memory.
Example: str x0, [x1] — This stores the value in x0 into the memory location pointed to by x1.
ldp: Load Pair of Registers. This instruction loads two registers from consecutive memory locations. The memory address is typically formed by adding an offset to the value in another register.
Example: ldp x0, x1, [x2] — This loads x0 and x1 from the memory locations at x2 and x2 + 8, respectively.
stp: Store Pair of Registers. This instruction stores two registers to consecutive memory locations. The memory address is typically formed by adding an offset to the value in another register.
Example: stp x0, x1, [sp] — This stores x0 and x1 to the memory locations at sp and sp + 8, respectively.
stp x0, x1, [sp, #16]! — This stores x0 and x1 to the memory locations at sp+16 and sp + 24, respectively, and updates sp with sp+16.
add: Add the values of two registers and store the result in a register.
Syntax: add(s) Xn1, Xn2, Xn3 | #imm, [shift #N | RRX]
Xn1 -> Destination
Xn2 -> Operand 1
Xn3 | #imm -> Operando 2 (register or immediate)
[shift #N | RRX] -> Perform a shift or call RRX
Example: add x0, x1, x2 — This adds the values in x1 and x2 together and stores the result in x0.
add x5, x5, #1, lsl #12 — This equals to 4096 (a 1 shifter 12 times) -> 1 0000 0000 0000 0000
adds This perform an add and updates the flags
sub: Subtract the values of two registers and store the result in a register.
Check add syntax.
Example: sub x0, x1, x2 — This subtracts the value in x2 from x1 and stores the result in x0.
subs This is like sub but updating the flag
mul: Multiply the values of two registers and store the result in a register.
Example: mul x0, x1, x2 — This multiplies the values in x1 and x2 and stores the result in x0.
div: Divide the value of one register by another and store the result in a register.
Example: div x0, x1, x2 — This divides the value in x1 by x2 and stores the result in x0.
lsl, lsr, asr, ror, rrx:
Logical shift left: Add 0s from the end moving the other bits forward (multiply by n-times 2)
Logical shift right: Add 1s at the beginning moving the other bits backward (divide by n-times 2 in unsigned)
Arithmetic shift right: Like lsr, but instead of adding 0s if the most significant bit is a 1, **1s are added (**divide by ntimes 2 in signed)
Rotate right: Like lsr but whatever is removed from the right it's appended to the left
Rotate Right with Extend: Like ror, but with the carry flag as the "most significant bit". So the carry flag is moved to the bit 31 and the removed bit to the carry flag.
bfm: Bit Filed Move, these operations copy bits 0...n from a value an place them in positions m..m+n. The #s specifies the leftmost bit position and #r the rotate right amount.
Bitfiled move: BFM Xd, Xn, #r
Signed Bitfield move: SBFM Xd, Xn, #r, #s
Unsigned Bitfield move: UBFM Xd, Xn, #r, #s
Bitfield Extract and Insert: Copy a bitfield from a register and copies it to another register.
BFI X1, X2, #3, #4 Insert 4 bits from X2 from the 3rd bit of X1
BFXIL X1, X2, #3, #4 Extract from the 3rd bit of X2 four bits and copy them to X1
SBFIZ X1, X2, #3, #4 Sign-extends 4 bits from X2 and inserts them into X1 starting at bit position 3 zeroing the right bits
SBFX X1, X2, #3, #4 Extracts 4 bits starting at bit 3 from X2, sign extends them, and places the result in X1
UBFIZ X1, X2, #3, #4 Zero-extends 4 bits from X2 and inserts them into X1 starting at bit position 3 zeroing the right bits
UBFX X1, X2, #3, #4 Extracts 4 bits starting at bit 3 from X2 and places the zero-extended result in X1.
Sign Extend To X: Extends the sign (or adds just 0s in the unsigned version) of a value to be able to perform operations with it:
SXTB X1, W2 Extends the sign of a byte from W2 to X1 (W2 is half of X2) to fill the 64bits
SXTH X1, W2 Extends the sign of a 16bit number from W2 to X1 to fill the 64bits
SXTW X1, W2 Extends the sign of a byte from W2 to X1 to fill the 64bits
UXTB X1, W2 Adds 0s (unsigned) to a byte from W2 to X1 to fill the 64bits
extr: Extracts bits from a specified pair of registers concatenated.
Example: EXTR W3, W2, W1, #3 This will concat W1+W2 and get from bit 3 of W2 up to bit 3 of W1 and store it in W3.
cmp: Compare two registers and set condition flags. It's an alias of subs setting the destination register to the zero register. Useful to know if m == n.
It supports the same syntax as subs
Example: cmp x0, x1 — This compares the values in x0 and x1 and sets the condition flags accordingly.
cmn: Compare negative operand. In this case it's an alias of adds and supports the same syntax. Useful to know if m == -n.
ccmp: Conditional comparison, it's a comparison that will be performed only if a previous comparison was true and will specifically set nzcv bits.
cmp x1, x2; ccmp x3, x4, 0, NE; blt _func -> if x1 != x2 and x3 < x4, jump to func
This is because ccmp will only be executed if the previous cmp was a NE, if it wasn't the bits nzcv will be set to 0 (which won't satisfy the blt comparison).
This ca also be used as ccmn (same but negative, like cmp vs cmn).
tst: It checks if any of the values of the comparison are both 1 (it works like and ANDS without storing the result anywhere). It's useful to check a registry with a value and check if any of the bits of the registry indicated in the value is 1.
Example: tst X1, #7 Check if any of the last 3 bits of X1 is 1
teq: XOR operation discarding the result
b: Unconditional Branch
Example: b myFunction
Note that this won't fill the link register with the return address (not suitable for subrutine calls that needs to return back)
bl: Branch with link, used to call a subroutine. Stores the return address in x30.
Example: bl myFunction — This calls the function myFunction and stores the return address in x30.
Note that this won't fill the link register with the return address (not suitable for subrutine calls that needs to return back)
blr: Branch with Link to Register, used to call a subroutine where the target is specified in a register. Stores the return address in x30. (This is
Example: blr x1 — This calls the function whose address is contained in x1 and stores the return address in x30.
ret: Return from subroutine, typically using the address in x30.
Example: ret — This returns from the current subroutine using the return address in x30.
b.<cond>: Conditional branches
b.eq: Branch if equal, based on the previous cmp instruction.
Example: b.eq label — If the previous cmp instruction found two equal values, this jumps to label.
b.ne: Branch if Not Equal. This instruction checks the condition flags (which were set by a previous comparison instruction), and if the compared values were not equal, it branches to a label or address.
Example: After a cmp x0, x1 instruction, b.ne label — If the values in x0 and x1 were not equal, this jumps to label.
cbz: Compare and Branch on Zero. This instruction compares a register with zero, and if they are equal, it branches to a label or address.
Example: cbz x0, label — If the value in x0 is zero, this jumps to label.
cbnz: Compare and Branch on Non-Zero. This instruction compares a register with zero, and if they are not equal, it branches to a label or address.
Example: cbnz x0, label — If the value in x0 is non-zero, this jumps to label.
tbnz: Test bit and branch on nonzero
Example: tbnz x0, #8, label
tbz: Test bit and branch on zero
Example: tbz x0, #8, label
Conditional select operations: These are operations whose behaviour varies depending on the conditional bits.
csel Xd, Xn, Xm, cond -> csel X0, X1, X2, EQ -> If true, X0 = X1, if false, X0 = X2
csinc Xd, Xn, Xm, cond -> If true, Xd = Xn, if false, Xd = Xm + 1
cinc Xd, Xn, cond -> If true, Xd = Xn + 1, if false, Xd = Xn
csinv Xd, Xn, Xm, cond -> If true, Xd = Xn, if false, Xd = NOT(Xm)
cinv Xd, Xn, cond -> If true, Xd = NOT(Xn), if false, Xd = Xn
csneg Xd, Xn, Xm, cond -> If true, Xd = Xn, if false, Xd = - Xm
cneg Xd, Xn, cond -> If true, Xd = - Xn, if false, Xd = Xn
cset Xd, Xn, Xm, cond -> If true, Xd = 1, if false, Xd = 0
csetm Xd, Xn, Xm, cond -> If true, Xd = <all 1>, if false, Xd = 0
adrp: Compute the page address of a symbol and store it in a register.
Example: adrp x0, symbol — This computes the page address of symbol and stores it in x0.
ldrsw: Load a signed 32-bit value from memory and sign-extend it to 64 bits.
Example: ldrsw x0, [x1] — This loads a signed 32-bit value from the memory location pointed to by x1, sign-extends it to 64 bits, and stores it in x0.
stur: Store a register value to a memory location, using an offset from another register.
Example: stur x0, [x1, #4] — This stores the value in x0 into the memory ddress that is 4 bytes greater than the address currently in x1.
svc : Make a system call. It stands for "Supervisor Call". When the processor executes this instruction, it switches from user mode to kernel mode and jumps to a specific location in memory where the kernel's system call handling code is located.
Example:
Save the link register and frame pointer to the stack:
Set up the new frame pointer: mov x29, sp (sets up the new frame pointer for the current function)
Allocate space on the stack for local variables (if needed): sub sp, sp, <size> (where <size> is the number of bytes needed)
Deallocate local variables (if any were allocated): add sp, sp, <size>
Restore the link register and frame pointer:
Return: ret (returns control to the caller using the address in the link register)
Armv8-A support the execution of 32-bit programs. AArch32 can run in one of two instruction sets: A32 and T32 and can switch between them via interworking.
Privileged 64-bit programs can schedule the execution of 32-bit programs by executing a exception level transfer to the lower privileged 32-bit.
Note that the transition from 64-bit to 32-bit occurs with a lower of the exception level (for example a 64-bit program in EL1 triggering a program in EL0). This is done by setting the bit 4 of SPSR_ELx special register to 1 when the AArch32 process thread is ready to be executed and the rest of SPSR_ELx stores the AArch32 programs CPSR. Then, the privileged process calls the ERET instruction so the processor transitions to AArch32 entering in A32 or T32 depending on CPSR**.**
The interworking occurs using the J and T bits of CPSR. J=0 and T=0 means A32 and J=0 and T=1 means T32. This basically traduces on setting the lowest bit to 1 to indicate the instruction set is T32.
This is set during the interworking branch instructions, but can also be set directly with other instructions when the PC is set as the destination register. Example:
Another example:
There are 16 32-bit registers (r0-r15). From r0 to r14 they can be used for any operation, however some of them are usually reserved:
r15: Program counter (always). Contains the address of the next instruction. In A32 current + 8, in T32, current + 4.
r11: Frame Pointer
r12: Intra-procedural call register
r13: Stack Pointer
r14: Link Register
Moreover, registers are backed up in banked registries. Which are places that store the registers values allowing to perform fast context switching in exception handling and privileged operations to avoid the need to manually save and restore registers every time.
This is done by saving the processor state from the CPSR to the SPSR of the processor mode to which the exception is taken. On the exception returns, the CPSR is restored from the SPSR.
In AArch32 the CPSR works similar to PSTATE in AArch64 and is also stored in SPSR_ELx when a exception is taken to restore later the execution:
The fields are divided in some groups:
Application Program Status Register (APSR): Arithmetic flags and accesible from EL0
Execution State Registers: Process behaviour (managed by the OS).
The N, Z, C, V flags (just like in AArch64)
The Q flag: It's set to 1 whenever integer saturation occurs during the execution of a specialized saturating arithmetic instruction. Once it's set to 1, it'll maintain the value until it's manually set to 0. Moreover, there isn't any instruction that checks its value implicitly, it must be done reading it manually.
GE (Greater than or equal) Flags: It's used in SIMD (Single Instruction, Multiple Data) operations, such as "parallel add" and "parallel subtract". These operations allow processing multiple data points in a single instruction.
For example, the UADD8 instruction adds four pairs of bytes (from two 32-bit operands) in parallel and stores the results in a 32-bit register. It then sets the GE flags in the APSR based on these results. Each GE flag corresponds to one of the byte additions, indicating if the addition for that byte pair overflowed.
The SEL instruction uses these GE flags to perform conditional actions.
The J and T bits: J should be 0 and if T is 0 the instruction set A32 is used, and if it's 1, the T32 is used.
IT Block State Register (ITSTATE): These are the bits from 10-15 and 25-26. They store conditions for instructions inside an IT prefixed group.
E bit: Indicates the endianness.
Mode and Exception Mask Bits (0-4): They determine the current execution state. The 5th one indicates if the program runs as 32bit (a 1) or 64bit (a 0). The other 4 represents the exception mode currently in used (when a exception occurs and it's being handled). The number set indicates the current priority in case another exception is triggered while this is being handled.
AIF: Certain exceptions can be disabled using the bits A, I, F. If A is 1 it means asynchronous aborts will be triggered. The I configures to respond to external hardware Interrupts Requests (IRQs). and the F is related to Fast Interrupt Requests (FIRs).
Check out syscalls.master. BSD syscalls will have x16 > 0.
Check out in syscall_sw.c the mach_trap_table and in mach_traps.h the prototypes. The mex number of Mach traps is MACH_TRAP_TABLE_COUNT = 128. Mach traps will have x16 < 0, so you need to call the numbers from the previous list with a minus: _kernelrpc_mach_vm_allocate_trap is -10.
You can also check libsystem_kernel.dylib in a disassembler to find how to call these (and BSD) syscalls:
Note that Ida and Ghidra can also decompile specific dylibs from the cache just by passing the cache.
Sometimes it's easier to check the decompiled code from libsystem_kernel.dylib than checking the source code because the code of several syscalls (BSD and Mach) are generated via scripts (check comments in the source code) while in the dylib you can find what is being called.
XNU supports another type of calls called machine dependent. The numbers of these calls depends on the architecture and neither the calls or numbers are guaranteed to remain constant.
This is a kernel owner memory page that is mapped into the address scape of every users process. It's meant to make the transition from user mode to kernel space faster than using syscalls for kernel services that are used so much the this transition would be vey inneficient.
For example the call gettimeofdate reads the value of timeval directly from the comm page.
It's super common to find this function used in Objective-C or Swift programs. This function allows to call a method of an objective-C object.
Parameters (more info in the docs):
x0: self -> Pointer to the instance
x1: op -> Selector of the method
x2... -> Rest of the arguments of the invoked method
So, if you put breakpoint before the branch to this function, you can easily find what is invoked in lldb with (in this example the object calls an object from NSConcreteTask that will run a command):
Setting the env variable NSObjCMessageLoggingEnabled=1 it's possible to log when this function is called in a file like /tmp/msgSends-pid.
Moreover, setting OBJC_HELP=1 and calling any binary you can see other environment variables you could use to log when certain Objc-C actions occurs.
When this function is called, it's needed to find the called method of the indicated instance, for this different searches are made:
Perform optimistic cache lookup:
If successful, done
Acquire runtimeLock (read)
If (realize && !cls->realized) realize class
If (initialize && !cls->initialized) initialize class
Try class own cache:
If successful, done
Try class method list:
If found, fill cache and done
Try superclass cache:
If successful, done
Try superclass method list:
If found, fill cache and done
If (resolver) try method resolver, and repeat from class lookup
If still here (= all else has failed) try forwarder
To compile:
To extract the bytes:
For newer macOS:
Taken from here and explained.
The goal is to execute execve("/bin/cat", ["/bin/cat", "/etc/passwd"], NULL), so the second argument (x1) is an array of params (which in memory these means a stack of the addresses).
Bind shell from https://raw.githubusercontent.com/daem0nc0re/macOS_ARM64_Shellcode/master/bindshell.s in port 4444
From https://github.com/daem0nc0re/macOS_ARM64_Shellcode/blob/master/reverseshell.s, revshell to 127.0.0.1:4444
Learn & practice AWS Hacking:HackTricks Training AWS Red Team Expert (ARTE) Learn & practice GCP Hacking: HackTricks Training GCP Red Team Expert (GRTE)
