Re: [PATCH] Use current_inferior ()->pid for AIX

[Simon Marchi <simon.marchi@polymtl.ca>]

Hi Aditya,

I don't think that using current_inferior throughout is a good solution
to your problems.  It will be case by case, unfortunately there's no
easy way out of this but to dig in GDB and understand its internals a
bit.

> From eb10e0ebc422d01f5ce96786f0bcf78a4f57d7ef Mon Sep 17 00:00:00 2001
> From: Aditya Vidyadhar Kamath
>  <adityakamath@li-ee942a4f-4ab2-47e9-8494-fb53eafbb9e1.ibm.com>
> Date: Wed, 30 Mar 2022 11:19:14 +0530
> Subject: [PATCH] Use current_inferior ()->pid for AIX.
> 
> ---
>  gdb/aix-thread.c     | 15 ++++++++-------
>  gdb/rs6000-aix-nat.c |  8 ++++----
>  2 files changed, 12 insertions(+), 11 deletions(-)
> 
> diff --git a/gdb/aix-thread.c b/gdb/aix-thread.c
> index 85be4c15f1c..9331de3beb2 100644
> --- a/gdb/aix-thread.c
> +++ b/gdb/aix-thread.c
> @@ -708,7 +708,7 @@ get_signaled_thread (void)
>  
>    while (1)
>      {
> -      if (getthrds (inferior_ptid.pid (), &thrinf,
> +      if (getthrds (current_inferior ()->pid, &thrinf,
>  		    sizeof (thrinf), &ktid, 1) != 1)

This function doesn't seem to depend on the current inferior/thread, so
it could easily be changed to receive the pid as a parameter.

>  	break;
>  
> @@ -791,7 +791,7 @@ sync_threadlists (void)
>  
>    /* Apply differences between the two arrays to GDB's thread list.  */
>  
> -  infpid = inferior_ptid.pid ();
> +  infpid = current_inferior ()->pid;
>    for (pi = gi = 0; pi < pcount || gi < gcount;)
>      {
>        if (pi == pcount)
> @@ -883,11 +883,11 @@ pd_update (int set_infpid)
>    struct thread_info *thread = NULL;
>  
>    if (!pd_active)
> -    return inferior_ptid;
> +    return ptid_t (current_inferior ()->pid);
>  
>    status = pthdb_session_update (pd_session);
>    if (status != PTHDB_SUCCESS)
> -    return inferior_ptid;
> +    return ptid_t (current_inferior ()->pid);

Here, the pd_update function seems to want to return the event ptid
unmodified.  So, perhaps that ptid could be passed by the caller and
returned here.  Or, pd_update could be changed to return an
optional<ptid_t> and the caller deals with returning the right ptid.

>  
>    sync_threadlists ();
>  
> @@ -897,7 +897,7 @@ pd_update (int set_infpid)
>    if (tid != 0)
>      thread = iterate_over_threads (iter_tid, &tid);
>    if (!thread)
> -    ptid = inferior_ptid;
> +    ptid = ptid_t (current_inferior ()->pid);
>    else
>      {
>        ptid = thread->ptid;
> @@ -921,7 +921,7 @@ pd_activate (int set_infpid)
>  			       &pd_session);
>    if (status != PTHDB_SUCCESS)
>      {
> -      return inferior_ptid;
> +      return ptid_t (current_inferior ()->pid);

This is the same idea as in pd_update.

>      }
>    pd_active = 1;
>    return pd_update (set_infpid);
> @@ -932,11 +932,12 @@ pd_activate (int set_infpid)
>  static void
>  pd_deactivate (void)
>  {
> +  ptid_t ptdrtn = ptid_t (current_inferior ()->pid)	
>    if (!pd_active)
>      return;
>    pthdb_session_destroy (pd_session);
>    
> -  pid_to_prc (&inferior_ptid);
> +  pid_to_prc (&ptdrtn);

Since pid_to_prc only writes to ptdrtn, it essentially a no-op.  I would
try just removing that call.  It was used to modify inferior_ptid after
the thread layer was disabled, which does not seem useful to me.

> diff --git a/gdb/rs6000-aix-nat.c b/gdb/rs6000-aix-nat.c
> index 8563aea313a..0b8e4d2c687 100644
> --- a/gdb/rs6000-aix-nat.c
> +++ b/gdb/rs6000-aix-nat.c
> @@ -397,7 +397,7 @@ rs6000_nat_target::xfer_partial (enum target_object object,
>  				 ULONGEST offset, ULONGEST len,
>  				 ULONGEST *xfered_len)
>  {
> -  pid_t pid = inferior_ptid.pid ();
> +  pid_t pid = current_inferior ()->pid;
>    int arch64 = ARCH64 ();
>  
>    switch (object)

Here, inferior_ptid should be valid, this change should not be
necessary.

> @@ -525,11 +525,11 @@ rs6000_nat_target::wait (ptid_t ptid, struct target_waitstatus *ourstatus,
>  
>  	  /* Claim it exited with unknown signal.  */
>  	  ourstatus->set_signalled (GDB_SIGNAL_UNKNOWN);
> -	  return inferior_ptid;
> +	  return ptid_t (current_inferior ()->pid);
>  	}
>  
>        /* Ignore terminated detached child processes.  */
> -      if (!WIFSTOPPED (status) && pid != inferior_ptid.pid ())
> +      if (!WIFSTOPPED (status) && pid != pid_t(current_inferior ()->pid))
>  	pid = -1;
>      }
>    while (pid == -1);

The target_ops::wait method implemenattions should ne rely on the
current inferior either.  I don't think that this target supports
debugging multiple processes at the same time, but imagine that it does,
eventually.  When rs6000_nat_target::wait gets called, the current
inferior is one of the inferiors managed by that target, at random.  So
what the wait method needs to do is fetch an event from the OS, and
returning, regardless of which inferior it applies to (except if a ptid
argument is given).  So, you should never need to use either
inferior_ptid nor current_inferior in wait.

> @@ -658,7 +658,7 @@ rs6000_nat_target::xfer_shared_libraries
>    if (writebuf)
>      return TARGET_XFER_E_IO;
>  
> -  gdb::byte_vector ldi_buf = rs6000_ptrace_ldinfo (inferior_ptid);
> +  gdb::byte_vector ldi_buf = rs6000_ptrace_ldinfo (ptid_t(current_inferior ()->pid));
>    result = rs6000_aix_ld_info_to_xml (target_gdbarch (), ldi_buf.data (),
>  				      readbuf, offset, len, 1);

This is used by xfer_partial, where inferior_ptid should be valid, so
this change shouldn't be necessary.

Simon