Purpose of rdi register for no argument function

Question

Consider this simple function:

struct Foo {
  int a;
  int b;
  int c;
  int d;
  int e;
  int f;
};

Foo foo() {
  Foo f;
  f.a = 1;
  f.b = 2;
  f.c = 3;
  f.d = 4;
  f.e = 5;
  f.f = 6;

  return f;
}

It generates the following assembly:

0000000000400500 <foo()>:
  400500:       48 ba 01 00 00 00 02    movabs rdx,0x200000001
  400507:       00 00 00 
  40050a:       48 b9 03 00 00 00 04    movabs rcx,0x400000003
  400511:       00 00 00 
  400514:       48 be 05 00 00 00 06    movabs rsi,0x600000005
  40051b:       00 00 00 
  40051e:       48 89 17                mov    QWORD PTR [rdi],rdx
  400521:       48 89 4f 08             mov    QWORD PTR [rdi+0x8],rcx
  400525:       48 89 77 10             mov    QWORD PTR [rdi+0x10],rsi
  400529:       48 89 f8                mov    rax,rdi
  40052c:       c3                      ret    
  40052d:       0f 1f 00                nop    DWORD PTR [rax]

Based on the assembly, I understand that the caller created space for Foo on its stack, and passed that information in rdi to the callee.

I am trying to find documentation for this convention. Calling convention in linux states that rdi contains the first integer argument. In this case, foo doesn't have any arguments.

Moreover, if I make foo take one integer argument, that is now passed as rsi (register for second argument) with rdi used for address of the return object.

Can anyone provide some documentation and clarity on how rdi is used in system V ABI?

Answer 1

See section 3.2.3 Parameter Passing in the ABI docs which says:

If the type has class MEMORY, then the caller provides space for the return value and passes the address of this storage in %rdi as if it were the first argument to the function. In effect, this address becomes a "hidden" first argument.

On return %rax will contain the address that has been passed in by the caller in %rdi.