[PATCH 0/2] PR gdb/19676: problems when /proc is not mounted

[PATCH 0/2] PR gdb/19676: problems when /proc is not mounted

[Pedro Alves]

I've just finished setting up a few machine, and what better to test
that "make check" works than fix a couple bugs...

This series fixes a couple problems/regressions when /proc is not
mounted, both introduced by yours truly.

Users better mount /proc, but at least this keeps GDB limping along...

Regression tested on x86-64 Fedora 23.

Pedro Alves (2):
  Fix PR gdb/19676: Disable displaced stepping if /proc not mounted
  Fix PR gdb/19676: Internal error in linux-thread.db.c if /proc not
    mounted

 gdb/infrun.c          |  3 ++-
 gdb/linux-tdep.c      |  3 ++-
 gdb/linux-thread-db.c | 16 ++++++++++------
 3 files changed, 14 insertions(+), 8 deletions(-)

-- 
2.5.0

[PATCH 1/2] Fix PR gdb/19676: Disable displaced stepping if /proc not mounted

[Pedro Alves]

On GNU/Linux archs that support displaced stepping, if /proc is not
mounted, GDB gets stuck not able to step past breakpoints:

 (gdb) c
 Continuing.
 dl_main (phdr=<optimized out>, phnum=<optimized out>, user_entry=<optimized out>, auxv=<optimized out>) at rtld.c:2163
 2163      LIBC_PROBE (init_complete, 2, LM_ID_BASE, r);
 Cannot find AT_ENTRY auxiliary vector entry.
 (gdb) c
 Continuing.
 dl_main (phdr=<optimized out>, phnum=<optimized out>, user_entry=<optimized out>, auxv=<optimized out>) at rtld.c:2163
 2163      LIBC_PROBE (init_complete, 2, LM_ID_BASE, r);
 Cannot find AT_ENTRY auxiliary vector entry.
 (gdb)

That's because GDB can't figure out where the scratch pad is.

This is a regression introduced by the earlier changes to make the
Linux native target always work in non-stop mode.

This commit makes GDB detect the case and fallback to stepping over
breakpoints in-line.

gdb/ChangeLog:
2016-03-01  Pedro Alves  <pedro@cascais.lan>

	PR gdb/19676
	* infrun.c (displaced_step_prepare): Also disable displaced
	stepping on NOT_SUPPORTED_ERROR.
	* linux-tdep.c (linux_displaced_step_location): If reading auxv
	fails, throw NOT_SUPPORTED_ERROR instead of generic error.
---
 gdb/infrun.c     | 3 ++-
 gdb/linux-tdep.c | 3 ++-
 2 files changed, 4 insertions(+), 2 deletions(-)

diff --git a/gdb/infrun.c b/gdb/infrun.c
index 3e8c9e0..696105d 100644
--- a/gdb/infrun.c
+++ b/gdb/infrun.c
@@ -1894,7 +1894,8 @@ displaced_step_prepare (ptid_t ptid)
     {
       struct displaced_step_inferior_state *displaced_state;
 
-      if (ex.error != MEMORY_ERROR)
+      if (ex.error != MEMORY_ERROR
+	  && ex.error != NOT_SUPPORTED_ERROR)
 	throw_exception (ex);
 
       if (debug_infrun)
diff --git a/gdb/linux-tdep.c b/gdb/linux-tdep.c
index 555c302..f197aa7 100644
--- a/gdb/linux-tdep.c
+++ b/gdb/linux-tdep.c
@@ -2426,7 +2426,8 @@ linux_displaced_step_location (struct gdbarch *gdbarch)
      location.  The auxiliary vector gets us the PowerPC-side entry
      point address instead.  */
   if (target_auxv_search (&current_target, AT_ENTRY, &addr) <= 0)
-    error (_("Cannot find AT_ENTRY auxiliary vector entry."));
+    throw_error (NOT_SUPPORTED_ERROR,
+		 _("Cannot find AT_ENTRY auxiliary vector entry."));
 
   /* Make certain that the address points at real code, and not a
      function descriptor.  */
-- 
2.5.0

Re: [PATCH 1/2] Fix PR gdb/19676: Disable displaced stepping if /proc not mounted

[Simon Marchi]

On 16-03-01 11:45 AM, Pedro Alves wrote:
> On GNU/Linux archs that support displaced stepping, if /proc is not
> mounted, GDB gets stuck not able to step past breakpoints:
> 
>  (gdb) c
>  Continuing.
>  dl_main (phdr=<optimized out>, phnum=<optimized out>, user_entry=<optimized out>, auxv=<optimized out>) at rtld.c:2163
>  2163      LIBC_PROBE (init_complete, 2, LM_ID_BASE, r);
>  Cannot find AT_ENTRY auxiliary vector entry.
>  (gdb) c
>  Continuing.
>  dl_main (phdr=<optimized out>, phnum=<optimized out>, user_entry=<optimized out>, auxv=<optimized out>) at rtld.c:2163
>  2163      LIBC_PROBE (init_complete, 2, LM_ID_BASE, r);
>  Cannot find AT_ENTRY auxiliary vector entry.
>  (gdb)
> 
> That's because GDB can't figure out where the scratch pad is.
> 
> This is a regression introduced by the earlier changes to make the
> Linux native target always work in non-stop mode.
> 
> This commit makes GDB detect the case and fallback to stepping over
> breakpoints in-line.
> 
> gdb/ChangeLog:
> 2016-03-01  Pedro Alves  <pedro@cascais.lan>
> 
> 	PR gdb/19676
> 	* infrun.c (displaced_step_prepare): Also disable displaced
> 	stepping on NOT_SUPPORTED_ERROR.
> 	* linux-tdep.c (linux_displaced_step_location): If reading auxv
> 	fails, throw NOT_SUPPORTED_ERROR instead of generic error.
> ---
>  gdb/infrun.c     | 3 ++-
>  gdb/linux-tdep.c | 3 ++-
>  2 files changed, 4 insertions(+), 2 deletions(-)
> 
> diff --git a/gdb/infrun.c b/gdb/infrun.c
> index 3e8c9e0..696105d 100644
> --- a/gdb/infrun.c
> +++ b/gdb/infrun.c
> @@ -1894,7 +1894,8 @@ displaced_step_prepare (ptid_t ptid)
>      {
>        struct displaced_step_inferior_state *displaced_state;
>  
> -      if (ex.error != MEMORY_ERROR)
> +      if (ex.error != MEMORY_ERROR
> +	  && ex.error != NOT_SUPPORTED_ERROR)
>  	throw_exception (ex);
>  
>        if (debug_infrun)
> diff --git a/gdb/linux-tdep.c b/gdb/linux-tdep.c
> index 555c302..f197aa7 100644
> --- a/gdb/linux-tdep.c
> +++ b/gdb/linux-tdep.c
> @@ -2426,7 +2426,8 @@ linux_displaced_step_location (struct gdbarch *gdbarch)
>       location.  The auxiliary vector gets us the PowerPC-side entry
>       point address instead.  */
>    if (target_auxv_search (&current_target, AT_ENTRY, &addr) <= 0)
> -    error (_("Cannot find AT_ENTRY auxiliary vector entry."));
> +    throw_error (NOT_SUPPORTED_ERROR,
> +		 _("Cannot find AT_ENTRY auxiliary vector entry."));
>  
>    /* Make certain that the address points at real code, and not a
>       function descriptor.  */
> 

Whoa, I didn't even know you could run without proc.

You would probably have seen it anyway, but the email address in your
ChangeLog entries has a local host name.

Simon

Re: [PATCH 1/2] Fix PR gdb/19676: Disable displaced stepping if /proc not mounted

[Pedro Alves]

On 03/01/2016 05:00 PM, Simon Marchi wrote:

> You would probably have seen it anyway, but the email address in your
> ChangeLog entries has a local host name.

Whoops, thanks.  :-)

-- 
Pedro Alves

[PATCH 2/2] Fix PR gdb/19676: Internal error in linux-thread.db.c if /proc not mounted

[Pedro Alves]

If /proc is not mounted, GDB fails an assertion in find_new_threads_once:

 Continuing.
 /home/pedro/gdb/mygit/src/gdb/linux-thread-db.c:1249: internal-error: find_new_threads_once: Assertion `!target_has_execution' failed.
 A problem internal to GDB has been detected,
 further debugging may prove unreliable.
 Quit this debugging session? (y or n)

That was supposed to catch misuses of td_ta_thr_iter, which is unsafe
for live debugging.  However, if /proc is not mounted, we still
fallback to using it.

I didn't bother with a warning, because GDB already prints several
others related to failing to open /proc files.

gdb/ChangeLog:
2016-03-01  Pedro Alves  <pedro@cascais.lan>

	PR gdb/19676
	* linux-thread-db.c (try_thread_db_load_1): Leave
	info->td_ta_thr_iter_p NULL iff debugging a live process and we
	have /proc access.
	(find_new_threads_once): Assert that we have a non-NULL
	info->td_ta_thr_iter_p instead of checking whether the target has
	execution.
---
 gdb/linux-thread-db.c | 16 ++++++++++------
 1 file changed, 10 insertions(+), 6 deletions(-)

diff --git a/gdb/linux-thread-db.c b/gdb/linux-thread-db.c
index 1eb457d..ce60beb 100644
--- a/gdb/linux-thread-db.c
+++ b/gdb/linux-thread-db.c
@@ -564,7 +564,6 @@ try_thread_db_load_1 (struct thread_db_info *info)
 
   /* These are essential.  */
   CHK (TDB_VERBOSE_DLSYM (info, td_ta_map_lwp2thr));
-  CHK (TDB_VERBOSE_DLSYM (info, td_ta_thr_iter));
   CHK (TDB_VERBOSE_DLSYM (info, td_thr_validate));
   CHK (TDB_VERBOSE_DLSYM (info, td_thr_get_info));
 
@@ -572,10 +571,6 @@ try_thread_db_load_1 (struct thread_db_info *info)
   TDB_DLSYM (info, td_thr_tls_get_addr);
   TDB_DLSYM (info, td_thr_tlsbase);
 
-#undef TDB_VERBOSE_DLSYM
-#undef TDB_DLSYM
-#undef CHK
-
   /* It's best to avoid td_ta_thr_iter if possible.  That walks data
      structures in the inferior's address space that may be corrupted,
      or, if the target is running, may change while we walk them.  If
@@ -587,6 +582,15 @@ try_thread_db_load_1 (struct thread_db_info *info)
      currently on core targets, as it uses ptrace directly.  */
   if (target_has_execution
       && linux_proc_task_list_dir_exists (ptid_get_pid (inferior_ptid)))
+    info->td_ta_thr_iter_p = NULL;
+  else
+    CHK (TDB_VERBOSE_DLSYM (info, td_ta_thr_iter));
+
+#undef TDB_VERBOSE_DLSYM
+#undef TDB_DLSYM
+#undef CHK
+
+  if (info->td_ta_thr_iter_p == NULL)
     {
       struct lwp_info *lp;
       int pid = ptid_get_pid (inferior_ptid);
@@ -1246,7 +1250,7 @@ find_new_threads_once (struct thread_db_info *info, int iteration,
   data.new_threads = 0;
 
   /* See comment in thread_db_update_thread_list.  */
-  gdb_assert (!target_has_execution);
+  gdb_assert (info->td_ta_thr_iter_p != NULL);
 
   TRY
     {
-- 
2.5.0

Re: [PATCH 0/2] PR gdb/19676: problems when /proc is not mounted

[Pedro Alves]

On 03/01/2016 04:45 PM, Pedro Alves wrote:
> I've just finished setting up a few machine, and what better to test
> that "make check" works than fix a couple bugs...
>
> This series fixes a couple problems/regressions when /proc is not
> mounted, both introduced by yours truly.
>
> Users better mount /proc, but at least this keeps GDB limping along...
>
> Regression tested on x86-64 Fedora 23.
>
> Pedro Alves (2):
>    Fix PR gdb/19676: Disable displaced stepping if /proc not mounted
>    Fix PR gdb/19676: Internal error in linux-thread.db.c if /proc not
>      mounted

Now pushed.

Thanks,
Pedro Alves

Re: [PATCH 0/2] PR gdb/19676: problems when /proc is not mounted

[Pedro Alves]

On 03/15/2016 04:34 PM, Pedro Alves wrote:
> On 03/01/2016 04:45 PM, Pedro Alves wrote:
>> I've just finished setting up a few machine, and what better to test
>> that "make check" works than fix a couple bugs...
>>
>> This series fixes a couple problems/regressions when /proc is not
>> mounted, both introduced by yours truly.
>>
>> Users better mount /proc, but at least this keeps GDB limping along...
>>
>> Regression tested on x86-64 Fedora 23.
>>
>> Pedro Alves (2):
>>    Fix PR gdb/19676: Disable displaced stepping if /proc not mounted
>>    Fix PR gdb/19676: Internal error in linux-thread.db.c if /proc not
>>      mounted
>
> Now pushed.

Also pushed to the 7.11 branch.

Thanks,
Pedro Alves