[RFA/RFC] Fix broken user-thread debugging on x86-solaris

[Joel Brobecker <brobecker@adacore.com>]

[-- Attachment #1: Type: text/plain, Size: 2653 bytes --]

Hello,

While testing GDB on x86-solaris, I noticed that the debugger now prints
a warning during startup:

    % ./gdb
    [GDB will not be able to debug user-mode threads: ld.so.1: gdb.ref: fatal: relocation error: file /usr/lib/libthread_db.so.1: symbol ps_lgetLDT: referenced symbol not found]
    GNU gdb 6.8.50.20080311-cvs
    Copyright (C) 2008 Free Software Foundation, Inc.
    [...]

The problem was introduced when we killed the TM file, which was
defining the TM_I386SOL2_H macro that a couple of units were actually
using (ugh!).

Basically, the problem on x86-solaris is that, in order to load
libthread_db.so, we need to define ps_lgetLDT or the loading
will fail due to this symbol being undefined.  This function is
defined inside sol-thread.c but is conditionalized on TM_I386SOL2_H
being defined:

    #ifdef TM_I386SOL2_H
    
    /* Reads the local descriptor table of a LWP.  */
    
    ps_err_e
    ps_lgetLDT (gdb_ps_prochandle_t ph, lwpid_t lwpid,
                struct ssd *pldt)
    {

Similarly, a couple of other functions inside procfs.c that
this function uses are conditionalized on the same macro:
proc_get_LDT_entry and procfs_find_LDT_entry. This stuff is
entirely x86-solaris specific, so cannot be enabled on sparc-solaris.

I think the most classical way of fixing this would be to move these
to their own file, such as sol-x86-thread.c. But I'm not sure whether
this is going to be the prefered solution in this case, because
the procfs routines need access to certain fields of a structure
that is not public (struct procinfo).

In this version, I went to the simplest, which is to have configure
define a new macro (HAVE_X86_SOLARIS_USER_THREADS) when on x86-solaris
and the thread_db library is available.  I then replaced TM_I386SOL2_H
by the new macro.

If I were to move the 3 functions to their own file, I would have
to create procfs.h to provide an opaque definition of struct procinfo.
I would also add a few accessor routines (to get the pid and the ctl_fd),
and make a few routines such as find_procinfo or proc_get_gregs non-static.
I don't mind doing this work, but I want to make sure others prefer it too.

2008-03-12  Joel Brobecker  <brobecker@adacore.com>

        * configure.ac (HAVE_X86_SOLARIS_USER_THREADS): Define this macro
        if we are building a native x86-solaris debugger that has access
        to user-threads.
        * configure, config.in: Regenerate.
        * sol-thread.c, procfs.c : Use HAVE_X86_SOLARIS_USER_THREADS
        in place of TM_I386SOL2_H.

Tested on x86-solaris. No regression.
Opinions?

BTW: We should fix this in the branch as well.

Thanks,
-- 
Joel

[-- Attachment #2: x86-user-threads.diff --]
[-- Type: text/plain, Size: 3155 bytes --]

Index: configure.ac
===================================================================
--- configure.ac	(revision 34930)
+++ configure.ac	(working copy)
@@ -1221,6 +1221,11 @@ if test ${build} = ${host} -a ${host} = 
             [Define if <proc_service.h> on solaris uses int instead of
              size_t, and assorted other type changes.])
 	 fi
+         if test "${gdb_host_cpu}" = "i386" ; then
+            AC_DEFINE(HAVE_X86_SOLARIS_USER_THREADS, 1,
+            [Define if the thread layer uses the user-thread debug library
+             on an x86-solaris system])
+         fi
       else
          AC_MSG_RESULT(no)
       fi
Index: configure
===================================================================
--- configure	(revision 34930)
+++ configure	(working copy)
@@ -22865,6 +22865,13 @@ cat >>confdefs.h <<\_ACEOF
 _ACEOF
 
 	 fi
+         if test "${gdb_host_cpu}" = "i386" ; then
+
+cat >>confdefs.h <<\_ACEOF
+#define HAVE_X86_SOLARIS_USER_THREADS 1
+_ACEOF
+
+         fi
       else
          echo "$as_me:$LINENO: result: no" >&5
 echo "${ECHO_T}no" >&6
Index: config.in
===================================================================
--- config.in	(revision 34930)
+++ config.in	(working copy)
@@ -482,6 +482,10 @@
 /* Define to 1 if `vfork' works. */
 #undef HAVE_WORKING_VFORK
 
+/* Define if the thread layer uses the user-thread debug library on an
+   x86-solaris system */
+#undef HAVE_X86_SOLARIS_USER_THREADS
+
 /* Define to 1 if you have the `XML_StopParser' function. */
 #undef HAVE_XML_STOPPARSER
 
Index: sol-thread.c
===================================================================
--- sol-thread.c	(revision 34930)
+++ sol-thread.c	(working copy)
@@ -1298,7 +1298,7 @@ ps_pdmodel (gdb_ps_prochandle_t ph, int 
 }
 #endif /* PR_MODEL_LP64 */
 
-#ifdef TM_I386SOL2_H
+#ifdef HAVE_X86_SOLARIS_USER_THREADS
 
 /* Reads the local descriptor table of a LWP.  */
 
@@ -1326,7 +1326,7 @@ ps_lgetLDT (gdb_ps_prochandle_t ph, lwpi
     /* LDT not found.  */
     return PS_ERR;
 }
-#endif /* TM_I386SOL2_H */
+#endif /* HAVE_X86_SOLARIS_USER_THREADS */
 \f
 
 /* Convert PTID to printable form.  */
Index: procfs.c
===================================================================
--- procfs.c	(revision 34930)
+++ procfs.c	(working copy)
@@ -3028,7 +3028,7 @@ proc_set_watchpoint (procinfo *pi, CORE_
 #endif
 }
 
-#ifdef TM_I386SOL2_H		/* Is it hokey to use this? */
+#ifdef HAVE_X86_SOLARIS_USER_THREADS
 
 #include <sys/sysi86.h>
 
@@ -3120,7 +3120,7 @@ proc_get_LDT_entry (procinfo *pi, int ke
 #endif
 }
 
-#endif /* TM_I386SOL2_H */
+#endif /* HAVE_X86_SOLARIS_USER_THREADS */
 
 /* =============== END, non-thread part of /proc  "MODULE" =============== */
 
@@ -5549,7 +5549,7 @@ procfs_stopped_by_watchpoint (ptid_t pti
   return 0;
 }
 
-#ifdef TM_I386SOL2_H
+#ifdef HAVE_X86_SOLARIS_USER_THREADS
 /*
  * Function: procfs_find_LDT_entry
  *
@@ -5587,7 +5587,7 @@ procfs_find_LDT_entry (ptid_t ptid)
   /* Find the matching entry and return it. */
   return proc_get_LDT_entry (pi, key);
 }
-#endif /* TM_I386SOL2_H */
+#endif /* HAVE_X86_SOLARIS_USER_THREADS */
 
 /*
  * Memory Mappings Functions: