Date: Mon, 22 Sep 2025 13:39:46 +0200 From: Paul Floyd <paulf2718@gmail.com> To: freebsd-hackers@FreeBSD.org Subject: ptrace PT_SETREGS issue Message-ID: <6e71483f-e0af-4933-b200-55e660b5748d@gmail.com>
next in thread | raw e-mail | index | archive | help
Hi
The second issue that I'm seeing with FreeBSD 15 is with ALPHA2 on aarch64.
The problem is with ptrace PT_SETREGS.
The context is when Valgrind is running with a connection to vgdb/gdb.
In this configuration, if the user wants to interrupt the running
process with control C the following sequence of events happens
1. vgdb uses ptrace to attach to the Valgrind process
2. It brute-force modifies the PC so that Valgrind will execute a
function that will interrupt the guest process.
This has been working for a bit over a year since aarch64 support in
Valgrind was added. With FreeBSD there is an issue. As an additional
check the ptrace code writes a special value, 0x8BADF00D in x0 (first
function call argument). On the receiving end there's an assert. That
assert is now firing (I'm seeing a value of 4 in the first argument). If
I uncomment the assert then the tests all work. That means that PC is
getting set OK but not X0.
The code that does this is
static
Bool setregs (pid_t pid, struct reg *regs)
{
if (ptrace(PT_SETREGS, pid, (caddr_t)regs, 0) < 0) {
return False;
}
return True;
}
This is going to be fairly tricky to debug. I can't run Valgrind under
gdb since that will prevent vgdb using ptrace to attach to it. I'll see
if I can put in a pause after vgdb has detached and then attach gdb to
Valgrind.
In the meantime, any ideas what could have changed with ptrace?
A+
Paul
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?6e71483f-e0af-4933-b200-55e660b5748d>