Re: [PATCH] Return unconditionally ptid.pid () in get_ptrace_pid() for NetBSD

[Simon Marchi <simark@simark.ca>]

On 2020-03-17 12:30 p.m., Kamil Rytarowski wrote:
> NetBSD tracks the PID and LWP pair separately and both values are
> needed and meaningful.
> ---
>  gdb/inf-ptrace.c | 6 +++++-
>  1 file changed, 5 insertions(+), 1 deletion(-)
> 
> diff --git a/gdb/inf-ptrace.c b/gdb/inf-ptrace.c
> index db17a76d946..6a6cb554ba7 100644
> --- a/gdb/inf-ptrace.c
> +++ b/gdb/inf-ptrace.c
> @@ -321,10 +321,14 @@ get_ptrace_pid (ptid_t ptid)
>  {
>    pid_t pid;
> 
> +#if !defined(__NetBSD__)
>    /* If we have an LWPID to work with, use it.  Otherwise, we're
> -     dealing with a non-threaded program/target.  */
> +     dealing with a non-threaded program/target.
> +
> +     NetBSD tracks the PID and LWP pair separately. */
>    pid = ptid.lwp ();
>    if (pid == 0)
> +#endif
>      pid = ptid.pid ();
>    return pid;
>  }
> --
> 2.25.0
> 

I think you should just avoid using get_ptrace_pid on NetBSD altogether, since
it is meant for OSes that require passing a single thread identifier to ptrace
(whereas NetBSD requires the (pid, lwp) pair).

Even with this modification in get_ptrace_pid, you need to change all the ptrace
call sites to pass the lwp on top of it.

I would suggest to instead #ifdef out get_ptrace_pid entirely on NetBSD, to avoid
using it by mistake, and just replace all ptrace call sites possibly used on BSD
to be

  ptrace (request, ptid.pid (), addr, ptid.lwp ());

This matches what I suggested in:

  https://sourceware.org/pipermail/gdb-patches/2020-March/166735.html

Simon