| Message ID | 1434093500-11201-1-git-send-email-alistair@popple.id.au |
|---|---|
| State | Accepted |
| Headers | show |
diff --git a/core/lock.c b/core/lock.c index fbf2d87..53cc337 100644 --- a/core/lock.c +++ b/core/lock.c @@ -31,6 +31,8 @@ bool bust_locks = true; static void lock_error(struct lock *l, const char *reason, uint16_t err) { + bust_locks = true; + fprintf(stderr, "LOCK ERROR: %s @%p (state: 0x%016lx)\n", reason, l, l->lock_val); op_display(OP_FATAL, OP_MOD_LOCK, err); diff --git a/core/utils.c b/core/utils.c index 9c94ecc..f340b4f 100644 --- a/core/utils.c +++ b/core/utils.c @@ -38,8 +38,6 @@ void __noreturn _abort(void) for (;;) ; in_abort = true; - bust_locks = true; - op_display(OP_FATAL, OP_MOD_CORE, 0x6666); prlog(PR_EMERG, "Aborting!\n");
Currently bust_locks is set to true whenever abort() is called. This was done for the lock debugging code which calls abort() as its last step. If there is a locking error there is a good chance parts of the console code will be deadlocked (for example if it has a lock on the scom bus). However busting locks can cause problems, especially now that flash data is being read in parallel as this uses the LPC bus. Instead of getting a nice error message showing where the code aborted it is likely the message will go missing, or it will be garbled by the flash data. So instead of always busting locks make it so they're only busted if they're suspected of being broken. Signed-off-by: Alistair Popple <alistair@popple.id.au> --- core/lock.c | 2 ++ core/utils.c | 2 -- 2 files changed, 2 insertions(+), 2 deletions(-)