host_entry.S

#include "sgx_arch.h"

#include "asm-offsets.h"

    .extern tcs_base
    .extern g_in_aex_profiling
    .extern dump_and_reset_stats

    .global sgx_ecall
    .type sgx_ecall, @function

sgx_ecall:
    .cfi_startproc

    # put entry address in RDX
    leaq .Lsgx_entry(%rip), %rdx

    # other arguments: RDI - code, RSI - ms

.Ldo_ecall_callee_save:
    pushq %rbx
    .cfi_adjust_cfa_offset 8
    pushq %rbp
    .cfi_adjust_cfa_offset 8
    pushq %r12
    .cfi_adjust_cfa_offset 8
    pushq %r13
    .cfi_adjust_cfa_offset 8
    pushq %r14
    .cfi_adjust_cfa_offset 8
    pushq %r15
    .cfi_adjust_cfa_offset 8

.Ldo_ecall:
    # increment per-thread EENTER counter for stats
    lock incq %gs:PAL_HOST_TCB_EENTER_CNT

    # RBX has to be the TCS of the thread
    movq %gs:PAL_HOST_TCB_TCS, %rbx

    # RCX has to be the AEP (Asynchronous Exit Pointer)
    leaq async_exit_pointer(%rip), %rcx

    movq $EENTER, %rax
    enclu

    # currently only ECALL_THREAD_RESET returns
.Lafter_resume:
    popq %r15
    .cfi_adjust_cfa_offset -8
    popq %r14
    .cfi_adjust_cfa_offset -8
    popq %r13
    .cfi_adjust_cfa_offset -8
    popq %r12
    .cfi_adjust_cfa_offset -8
    popq %rbp
    .cfi_adjust_cfa_offset -8
    popq %rbx
    .cfi_adjust_cfa_offset -8
    retq
    .cfi_endproc

    .global async_exit_pointer
    .type async_exit_pointer, @function

async_exit_pointer:
    .cfi_startproc
    .cfi_undefined %rip

    # increment per-thread AEX counter for stats
    lock incq %gs:PAL_HOST_TCB_AEX_CNT

    # Inform that we are in AEX profiling code
    movb $1, %gs:PAL_HOST_TCB_IN_AEX_PROF

    subq $RED_ZONE_SIZE, %rsp
    .cfi_adjust_cfa_offset RED_ZONE_SIZE
    # FIXME: This second RSP adjustment is because Exitless re-uses the same stack to create
    #        RPC request objects, as the AEX flow here. It just so happens that the total size of
    #        allocated RPC request objects is less than 128B (red zone size), so I reuse this const.
    #        This double-stack-usage needs a proper fix.
    subq $RED_ZONE_SIZE, %rsp
    .cfi_adjust_cfa_offset RED_ZONE_SIZE

    # Save ERESUME parameters
    pushq %rax
    .cfi_adjust_cfa_offset 8
    pushq %rbx
    .cfi_adjust_cfa_offset 8
    pushq %rcx
    .cfi_adjust_cfa_offset 8

    pushq %rbp
    .cfi_adjust_cfa_offset 8
    movq %rsp, %rbp
    .cfi_offset %rbp, -16
    .cfi_def_cfa_register %rbp
    andq $~0xF, %rsp  # Required by System V AMD64 ABI.

#ifdef DEBUG
    # Call sgx_profile_sample_aex with %rdi = TCS
    movq %rbx, %rdi
    call sgx_profile_sample_aex
#endif

    call dump_and_reset_stats

    # Restore stack
    movq %rbp, %rsp
    popq %rbp
    .cfi_def_cfa %rsp, 8

    # Restore ERESUME parameters
    popq %rcx
    .cfi_adjust_cfa_offset -8
    popq %rbx
    .cfi_adjust_cfa_offset -8
    popq %rax
    .cfi_adjust_cfa_offset -8
    addq $RED_ZONE_SIZE, %rsp
    .cfi_adjust_cfa_offset -RED_ZONE_SIZE
    # FIXME: See above for explanation for this second RSP adjustment.
    addq $RED_ZONE_SIZE, %rsp
    .cfi_adjust_cfa_offset -RED_ZONE_SIZE

    movb $0, %gs:PAL_HOST_TCB_IN_AEX_PROF
    .cfi_endproc

    # fall-through to ERESUME

    .global eresume_pointer
    .type eresume_pointer, @function

eresume_pointer:
    enclu   # perform ERESUME

    .global async_exit_pointer_end
    .type async_exit_pointer_end, @function

async_exit_pointer_end:

    .global sgx_raise
    .type sgx_raise, @function

sgx_raise:
    leaq .Lafter_resume(%rip), %rdx
    jmp .Ldo_ecall_callee_save

.Lsgx_entry:
    # arguments: RDI - code, RSI - ms
    .cfi_startproc

    # increment per-thread EEXIT counter for stats
    lock incq %gs:PAL_HOST_TCB_EEXIT_CNT

    leaq ocall_table(%rip), %rbx
    movq (%rbx,%rdi,8), %rbx
    movq %rsi, %rdi

    pushq %rbp
    .cfi_adjust_cfa_offset 8
    movq %rsp, %rbp
    .cfi_offset %rbp, -16
    .cfi_def_cfa_register %rbp

    # Adjust stack and save RDI
    subq $8, %rsp
    andq $~0xF, %rsp  # Required by System V AMD64 ABI.
    movq %rdi, -8(%rbp)

#if DEBUG
    # Call sgx_profile_sample_ocall_outer with RBX (ocall handler)
    movq %rbx, %rdi
    call sgx_profile_sample_ocall_outer

    # Call sgx_profile_sample_ocall_inner with RDX (pointer to in-enclave context)
    movq %rdx, %rdi
    call sgx_profile_sample_ocall_inner
#endif

    call dump_and_reset_stats

    # Restore RDI
    movq -8(%rbp), %rdi

    callq *%rbx

    movq %rbp, %rsp
    popq %rbp
    .cfi_def_cfa %rsp, 8

    movq %rax, %rdi
    movq $PAL_EVENT_NO_EVENT, %rsi
    lock xchgl %esi, %gs:PAL_HOST_TCB_LAST_ASYNC_EVENT

    # return to enclave, arguments:
    # RDI - return value
    # RSI - external event
    jmp .Ldo_ecall
    .cfi_endproc