This is the mail archive of the gdb-patches@sourceware.org mailing list for the GDB project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: [RFC] Replace deprecated_target_new_objfile_hook by observer?


Daniel Jacobowitz wrote:

> There Be Dragons!  Note that there is a case where the remote objfile
> hook does _not_ call the next thing on the chain.  This is somewhat
> deliberate, in that it's what prevents thread-db from being enabled
> when talking to gdbserver, and somewhat accidental, in that I'm sure it
> wasn't meant to work this way.  But we'll need to do some rearranging
> in order to keep the current state of affairs working.

I've run into this in my work on GDB support for the Cell BE (which is
getting a lot closer to a state where we can try to submit it, b.t.w.).

We're working on providing code overlay support for the Cell SPEs, and
to support this in the debugger, I needed to hook into the objfile chain;
I ran into exactly the problem you describe above.

To fix this, I'm currently using this patch, which attempts to express
more directly that linux-thread-db should not be used if the target is
remote:

diff -urN gdb-orig/gdb/linux-thread-db.c gdb-6.5/gdb/linux-thread-db.c
--- gdb-orig/gdb/linux-thread-db.c	2006-05-06 00:42:43.000000000 +0200
+++ gdb-6.5/gdb/linux-thread-db.c	2006-09-25 02:12:11.382901136 +0200
@@ -669,6 +669,10 @@
   if (!target_has_execution)
     return;
 
+  /* Don't attempt to use thread_db for remote targets.  */
+  if (!target_can_run (&current_target))
+    return;
+
   /* Initialize the structure that identifies the child process.  */
   proc_handle.pid = GET_PID (inferior_ptid);
 
diff -urN gdb-orig/gdb/remote.c gdb-6.5/gdb/remote.c
--- gdb-orig/gdb/remote.c	2006-05-05 22:08:45.000000000 +0200
+++ gdb-6.5/gdb/remote.c	2006-09-25 02:12:11.449890952 +0200
@@ -5466,8 +5466,7 @@
       remote_check_symbols (objfile);
     }
   /* Call predecessor on chain, if any.  */
-  if (remote_new_objfile_chain != 0 &&
-      remote_desc == 0)
+  if (remote_new_objfile_chain)
     remote_new_objfile_chain (objfile);
 }
 
Does this look reasonable?

Bye,
Ulrich

-- 
  Dr. Ulrich Weigand
  Linux on zSeries Development
  Ulrich.Weigand@de.ibm.com


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]