Pedigree

The x86_64 port now #PFs somewhere within either the TUI application, the dynamic linker, or a usermode stub within the kernel. Still debugging.

#PF is now mostly reproduced.

The following code, at the start of the TUI main() function, works:

void posix_syscall1(long func, long p1)
{
  long eax = ((1&0xFFFF) << 16) | (func&0xFFFF);
  long ret;
  asm volatile("syscall" : "=a" (ret), "=b" (errno) : "0" (eax), "1" (p1) : "rcx");
}

int main (int argc, char **argv)
{
    posix_syscall1(4321, 0xdeadbeef);

    ....

However, the following page faults:

int main (int argc, char **argv)
{
    __pedigree_test_posix_syscalls();

    ....

__pedigree_test_posix_syscalls is defined in libc.so.

The log file for both is as follows:

(NN) Syscall: at start, will be returning to 0x804fe58.
(NN) POSIX: returning to 0x804fe58.
(NN) POSIX: [0x2] : 4321
(EE) PosixSyscallManager: invalid syscall received: 4321
(NN) Syscall: at end, returning to 0x804fe58.

(NN) Syscall: at start, will be returning to 0x4000201c.
(NN) Syscall: at end, returning to 0x4000201c.

(NN) Syscall: at start, will be returning to 0x600676.
(NN) POSIX: returning to 0x600676.
(NN) POSIX: [0x2] : 1234
(EE) PosixSyscallManager: invalid syscall received: 1234
(NN) Syscall: at end, returning to 0x600676.
(EE) Page Fault Exception at 0x00005678, error code 0x00000006
(EE) Details: NOT PRESENT | WRITE | USER MODE | 

As can be seen, the libc-based system call fails. The code directly after this test is a while loop. Seems like a dynamic linker issue to me.

Turns out that #PF was due to an register not being added to the clobber list. New #PF somewhere in libc (inside _svfprintf_r), but I'm not convinced it's a libc issue.