FreeChainXenon/binutils-gdb
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

Fix gdbserver regression exposed by gdb.threads/multi-create-ns-info-thr.exp

Browse source 
Commit 8629910955 ("Add thread_db_notice_clone to gdbserver")
introduced calls into libthread_db without making sure that the
current thread is pointing to a know-stopped thread.  This resulted in
sometimes thread_db_notice_clone failing->find_one_thread failing like
this, as seen when running gdb.threads/multi-create-ns-info-thr.exp:

~~~
  Thread <6> executing
  Thread <7> executing
  gdbserver: PID mismatch!  Expected 27472, got 27471
  gdbserver: Cannot find thread after clone.

  Thread <1000> executing
  Thread <1001> executing
~~~

Things go south from here and sometimes that ends up resulting in
gdbserver crashing and the test failing.

gdb/gdbserver/ChangeLog:
2017-09-29  Pedro Alves  <palves@redhat.com>

	* linux-low.c (handle_extended_wait): Pass parent thread instead
	of process to thread_db_notice_clone.
	* linux-low.h (thread_db_notice_clone): Replace parent process
	parameter with parent thread parameter.
	* thread-db.c (find_one_thread): Add comment.
	(thread_db_notice_clone): Replace parent process parameter with
	parent thread parameter.  Temporarily switch to the parent thread.
This commit is contained in:
Pedro Alves 2017-09-29 13:06:34 +01:00
parent f6ac8c52c9
commit 94c207e097
4 changed files with 26 additions and 5 deletions
Download patch file Download diff file Expand all files Collapse all files

2
gdb/gdbserver/linux-low.h
View file

@ -417,7 +417,7 @@ int thread_db_look_up_one_symbol (const char *name, CORE_ADDR *addrp);
both the clone and the parent should be stopped. This function does
whatever is required have the clone under thread_db's control. */
void thread_db_notice_clone (struct process_info *proc, ptid_t lwp);
void thread_db_notice_clone (struct thread_info *parent_thr, ptid_t child_ptid);
bool thread_db_thread_handle (ptid_t ptid, gdb_byte **handle, int *handle_len);