ptrace asm wrapper for aarch64

[kib]

Summary

Align aarch64 with other arches, clear errno before calling ptrace syscall in the libc wrapper.

Test Plan

I was unable to test this, after I installed freshly build world/kernel from HEAD into 11.0 RC2 vm disk, I got

>> FreeBSD EFI boot block
   Loader path: /boot/loader.efi

   Initializing modules: ZFS UFS
   Probing 5 block devices.......* done
    ZFS found no pools
    UFS found 1 partition


Synchronous Exception at 0x294AD3FA2A0003EC

[andrew]

Do you have any specific test, or would single stepping with lldb be enough?

lib/libc/aarch64/sys/ptrace.S

39–42

I think this could be rewritten as:

stp x0, x1, [sp, #-48]!
stp x2, x3, [sp, #16]
stp x29, lr, [sp, #32]
add x29, sp, #32

It will then comply with the ABI.

45–48

This would then be:

ldp x29, lr, [sp, #32]
ldp x2, x3, [sp, #16]
ldp x0, x1, [sp], #48

[kib]

Update according to the recommendations.

BTW, from my reading of the IHI 0055C_beta, it is allowed to skip frame construction for some functions:

It may require the frame pointer to address a valid frame record at all times, except that any subroutine may
elect not to create a frame record

For testing, just running truss /bin/ls with the patched libc would give good enough coverage.

[kib]

And e.g. cerror does not build the frame when calling into __error.

[andrew]

In D7674#159465, @kib wrote:

Update according to the recommendations.

BTW, from my reading of the IHI 0055C_beta, it is allowed to skip frame construction for some functions:

It may require the frame pointer to address a valid frame record at all times, except that any subroutine may
elect not to create a frame record

The main advantage with creating it is the debugger can then perform a backtrace through the function. I added this to the kernel exception frame for this purpose.

For testing, just running truss /bin/ls with the patched libc would give good enough coverage.

Ok, I'll test it when I get a chance to update one of the netperf Caviums.

In D7674#159471, @kib wrote:

And e.g. cerror does not build the frame when calling into __error.

That should also be fixed at some stage. There are also a few places within the rtld code that also create bad frames.

[kib]

Superseded by D7694.