Comment 19 for bug 1793329

PLD linux carries a patch which uses a "naked" function instead of a toplevel asm function to allow the compiler to properly track the function.

http://git.pld-linux.org/?p=packages/kicad.git;f=lto.patch;h=ad2c2f023eb7ffeefef2c1fd4cff1fdc9f6b3156;hb=HEAD

I have tried it, and it works fine. The resulting assembly is (almost) the same as the plain assembly, you can try it with the following code:

$> cat ./test.cpp
---
void foo(int* a, char b) __attribute__ ((naked, aligned(4)));
void foo(int* a, char b)
{
asm volatile("mov $1, %eax;");
}
---
$> gcc -S ./test.cpp; gcc -c ./test.cpp; cat ./test.s; objdump -d ./test.o
---
        .file "test.cpp"
        .text
        .align 4
        .globl foo
        .type foo, @function
foo:
.LFB0:
        .cfi_startproc
#APP
# 5 "./test.cpp" 1
        mov $1, %eax;
# 0 "" 2
#NO_APP
        nop
        ud2
        .cfi_endproc
.LFE0:
        .size foo, .-foo
        .ident "GCC: (SUSE Linux) 9.1.1 20190703 [gcc-9-branch revision 273008]"
        .section .note.GNU-stack,"",@progbit
---
0000000000000000 <foo>:
   0: b8 01 00 00 00 mov $0x1,%eax
   5: 90 nop
   6: 0f 0b ud2
---

The only difference is the added nop,ud2. ud2 is the guaranteed intel illegal instruction, but as the code is not reached (last asm instruction is "hlt" respectively "jmp"), this does not matter.

The corresponding GCC BR is https://gcc.gnu.org/bugzilla/show_bug.cgi?id=57703