modify x64 assembly and so forth

This commit is contained in:
Niko Matsakis 2011-10-12 17:40:03 -07:00 committed by Brian Anderson
parent d3e865438f
commit 020cd360e1
4 changed files with 93 additions and 112 deletions

View File

@ -1,11 +1,39 @@
.text
/*
According to ABI documentation found at
http://www.x86-64.org/documentation.html
and Microsoft discussion at
http://msdn.microsoft.com/en-US/library/9z1stfyw%28v=VS.80%29.aspx.
BOTH CALLING CONVENTIONS
Callee save registers:
ebp, ebx, esi, edi
R12--R15, RDI, RSI, RBX, RBP, RSP
XMM0--XMM5
Caller save registers:
eax, ecx, edx
RAX, RCX, RDX, R8--R11
XMM6--XMM15
Floating point stack
MAC/AMD CALLING CONVENTIONS
Integer arguments go in registers:
rdi, rsi, rdx, rcx, r8, r9
User flags have no specified role and are not preserved
across calls, with the exception of DF in %rFLAGS,
which must be clear (set to "forward" direction)
on function entry and return.
MICROSOFT CALLING CONVENTIONS
Return value: RAX
First four arguments:
RCX, RDX, R8, R9
XMM0, XMM1, XMM2, XMM3
*/
/*
@ -17,60 +45,55 @@ Caller save registers:
// swap_registers(registers_t *oregs, registers_t *regs)
.globl swap_registers
swap_registers:
// save the old context
movl 4(%esp), %eax
//movl %eax, 0(%eax)
movl %ebx, 4(%eax)
movl %ecx, 8(%eax)
movl %edx, 12(%eax)
movl %ebp, 16(%eax)
movl %esi, 20(%eax)
movl %edi, 24(%eax)
//movl %cs, 32(%eax)
//movl %ds, 34(%eax)
//movl %ss, 36(%eax)
//movl %es, 38(%eax)
//movl %fs, 40(%eax)
//movl %gs, 42(%eax)
// n.b. when we enter, the return address is at the top of
// the stack (i.e., 0(%RSP)). We
// simply save all NV registers into oregs.
// We then restore all NV registers from regs. This restores
// the old stack pointer, which should include the proper
// return address. We can therefore just return normally to
// jump back into the old code.
// Save instruction pointer:
popl %rax
movl %rax, 104(%rdi)
// save the flags
pushf
popl %ecx
movl %ecx, 44(%eax)
// Save non-volatile integer registers:
// (including RSP)
movl %rbx, 0(%rdi)
movl %rsp, 8(%rdi)
movl %rbp, 16(%rdi)
movl %r12, 24(%rdi)
movl %r13, 32(%rdi)
movl %r14, 40(%rdi)
movl %r15, 48(%rdi)
// save the return address as the instruction pointer
// and save the stack pointer of the caller
popl %ecx
movl %esp, 28(%eax)
movl %ecx, 48(%eax)
// Save non-volatile XMM registers:
movl %xmm0, 56(%rdi)
movl %xmm1, 64(%rdi)
movl %xmm2, 72(%rdi)
movl %xmm3, 80(%rdi)
movl %xmm4, 88(%rdi)
movl %xmm5, 96(%rdi)
// restore the new context
movl 4(%esp), %eax
// Restore non-volatile integer registers:
// (including RSP)
movl 0(%rsi), %rbx
movl 8(%rsi), %rsp
movl 16(%rsi), %rbp
movl 24(%rsi), %r12
movl 32(%rsi), %r13
movl 40(%rsi), %r14
movl 48(%rsi), %r15
movl 4(%eax), %ebx
// save ecx for later...
movl 12(%eax), %edx
movl 16(%eax), %ebp
movl 20(%eax), %esi
movl 24(%eax), %edi
movl 28(%eax), %esp
// We can't actually change this...
//movl 32(%eax), %cs
//movl 34(%eax), %ds
//movl 36(%eax), %ss
//movl 38(%eax), %es
//movl 40(%eax), %fs
//movl 42(%eax), %gs
// restore the flags
movl 44(%eax), %ecx
push %ecx
popf
// ok, now we can restore ecx
movl 8(%eax), %ecx
// Return!
jmp *48(%eax)
// Restore non-volatile XMM registers:
movl 56(%rsi), %xmm0
movl 64(%rsi), %xmm1
movl 72(%rsi), %xmm2
movl 80(%rsi), %xmm3
movl 88(%rsi), %xmm4
movl 96(%rsi), %xmm5
// Jump to the instruction pointer
// found in regs:
jmp *104(%rsi)

View File

@ -11,12 +11,11 @@ _upcall_call_c_stack:
.globl upcall_call_c_stack
upcall_call_c_stack:
#endif
pushl %ebp
movl %esp,%ebp // save esp
movl 8(%esp),%eax // eax = callee
movl 12(%esp),%esp // switch stack
calll *%eax
movl %ebp,%esp // would like to use "leave" but it's slower
popl %ebp
pushl %rbp
movl %rsp,%rbp // save rsp
movl %rsi,%rsp // switch stack
calll *%rdi
movl %rbp,%rsp // would like to use "leave" but it's slower
popl %rbp
ret

View File

@ -6,9 +6,9 @@
#include <stdlib.h>
#include <assert.h>
extern "C" uint32_t CDECL swap_registers(registers_t *oregs,
registers_t *regs)
asm ("swap_registers");
extern "C" void CDECL swap_registers(registers_t *oregs,
registers_t *regs)
asm ("swap_registers");
context::context()
{
@ -26,48 +26,13 @@ void context::call(void *f, void *arg, void *stack) {
swap(*this);
// set up the trampoline frame
uint32_t *sp = (uint32_t *)stack;
uint64_t *sp = (uint64_t *)stack;
// Shift the stack pointer so the alignment works out right.
sp = align_down(sp) - 3;
*--sp = (uint32_t)arg;
*--sp = (uint64_t)arg;
*--sp = 0xdeadbeef;
regs.esp = (uint32_t)sp;
regs.eip = (uint32_t)f;
regs.regs[1] = (uint64_t)sp;
regs.ip = (uint64_t)f;
}
#if 0
// This is some useful code to check how the registers struct got
// layed out in memory.
int main() {
registers_t regs;
printf("Register offsets\n");
#define REG(r) \
printf(" %6s: +%ld\n", #r, (intptr_t)&regs.r - (intptr_t)&regs);
REG(eax);
REG(ebx);
REG(ecx);
REG(edx);
REG(ebp);
REG(esi);
REG(edi);
REG(esp);
REG(cs);
REG(ds);
REG(ss);
REG(es);
REG(fs);
REG(gs);
REG(eflags);
REG(eip);
return 0;
}
#endif

View File

@ -16,19 +16,13 @@ T align_down(T sp)
{
// There is no platform we care about that needs more than a
// 16-byte alignment.
return (T)((uint32_t)sp & ~(16 - 1));
return (T)((uint64_t)sp & ~(16 - 1));
}
struct registers_t {
// general purpose registers
uint32_t eax, ebx, ecx, edx, ebp, esi, edi, esp;
// segment registers
uint16_t cs, ds, ss, es, fs, gs;
uint32_t eflags;
uint32_t eip;
uint64_t regs[7]; // Space for the volatile regs: rbx, rsp, rbp, r12:r15
uint64_t xmms[6]; // Space for the volatile regs: xmm0:xmm5
uint64_t ip;
};
class context {