Please don't reply to lustre-devel. Instead, comment in Bugzilla by using the
following link:
https://bugzilla.lustre.org/show_bug.cgi?id=10982
(From update of attachment 8932)
>+# LN_CONFIG_PANIC_DUMPLOG
>+#
>+# check if tunable panic_dumplog is wanted
>+#
>+AC_DEFUN([LN_CONFIG_PANIC_DUMPLOG],
>+[AC_MSG_CHECKING([for tunable panic_dumplog support])
>+AC_ARG_ENABLE([panic_dumplog],
>+ AC_HELP_STRING([--enable-panic_dumplog],
>+ [enable panic_dumplog]),
>+ [],[enable_panic_dumplog='no'])
>+if test x$enable_panic_dumplog = xyes ; then
>+ AC_DEFINE(LNET_DUMP_ON_PANIC, 1, [use dumplog on panic])
>+ AC_MSG_RESULT([yes (by request)])
>+else
>+ AC_MSG_RESULT([no])
>+fi
>+])
>+
>+#
>+static int panic_notifier(struct notifier_block *self, unsigned long unused1,
>+ void *unused2)
>+{
>+#ifdef LNET_DUMP_ON_PANIC
>+ /* This is currently disabled because it spews far too much to the
>+ * console on the rare cases it is ever triggered. */
>+
>+ if (in_interrupt()) {
>+ trace_debug_print();
>+ } else {
>+ while (current->lock_depth >= 0)
>+ unlock_kernel();
>+
>+ libcfs_debug_dumplog_internal((void
>*)(long)cfs_curproc_pid());
>+ }
>+#endif
Unfortunately, the work to make this a configurable option is mostly wasted
effort. The comment is correct that this option spews too much to the console
to be useful. What is missing is that because the log format is now binary in
the kernel it is also completely useless. I would instead recommend just
removing all of the panic notifier code and the surrounding complexity, along
with the added breakage in darwin and winnt because of the lack of its
implementation.
_______________________________________________
Lustre-devel mailing list
[email protected]
https://mail.clusterfs.com/mailman/listinfo/lustre-devel
