From: zhanghc <zhanghc>
Date: Fri, 22 May 2009 05:42:51 +0000 (+0000)
Subject: b=11505
X-Git-Tag: v1_9_0_200~88
X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=commitdiff_plain;h=e12bc46a7167bc0ca43963338889eb12a2b8a375

b=11505

1, don't call "find_task_by_pid" in lcw_dump
   for this function was dropped after 2.6.27

2, add "reliable" argument in "dump_trace"

i=shadow
i=sheng.yang
---

diff --git a/libcfs/libcfs/linux/linux-debug.c b/libcfs/libcfs/linux/linux-debug.c
index b8a4982..bdcd432 100644
--- a/libcfs/libcfs/linux/linux-debug.c
+++ b/libcfs/libcfs/linux/linux-debug.c
@@ -205,6 +205,54 @@ void lbug_with_loc(const char *file, const char *func, const int line)
 
 #ifdef __KERNEL__
 
+#ifdef HAVE_DUMP_TRACE
+#include <linux/nmi.h>
+#include <asm/stacktrace.h>
+
+static void
+print_trace_warning_symbol(void *data, char *msg, unsigned long symbol)
+{
+	printk(data);
+	print_symbol(msg, symbol);
+	printk("\n");
+}
+
+static void print_trace_warning(void *data, char *msg)
+{
+	printk("%s%s\n", (char *)data, msg);
+}
+
+static int print_trace_stack(void *data, char *name)
+{
+	printk(" <%s> ", name);
+	return 0;
+}
+
+#ifdef HAVE_TRACE_ADDRESS_RELIABLE
+# define RELIABLE reliable
+# define DUMP_TRACE_CONST const
+static void print_trace_address(void *data, unsigned long addr, int reliable)
+#else
+/* before 2.6.24 there was no reliable arg */
+# define RELIABLE 1
+# define DUMP_TRACE_CONST
+static void print_trace_address(void *data, unsigned long addr)
+#endif
+{
+        char fmt[32];
+	touch_nmi_watchdog();
+        sprintf(fmt, " [<%016lx>] %s%%s\n", addr, RELIABLE ? "": "? ");
+	__print_symbol(fmt, addr);
+}
+
+static DUMP_TRACE_CONST struct stacktrace_ops print_trace_ops = {
+	.warning = print_trace_warning,
+	.warning_symbol = print_trace_warning_symbol,
+	.stack = print_trace_stack,
+	.address = print_trace_address,
+};
+#endif
+
 void libcfs_debug_dumpstack(struct task_struct *tsk)
 {
 #if defined(__arch_um__)
@@ -212,6 +260,18 @@ void libcfs_debug_dumpstack(struct task_struct *tsk)
                 CWARN("stack dump for pid %d (%d) requested; wake up gdb.\n",
                       tsk->pid, UML_PID(tsk));
         //asm("int $3");
+#elif defined(HAVE_DUMP_TRACE)
+        /* dump_stack() */
+        /* show_trace() */
+	printk("Pid: %d, comm: %.20s\n", current->pid, current->comm);
+        /* show_trace_log_lvl() */
+	printk("\nCall Trace:\n");
+	dump_trace(tsk, NULL, NULL,
+#ifdef HAVE_TRACE_ADDRESS_RELIABLE
+                   0,
+#endif /* HAVE_TRACE_ADDRESS_RELIABLE */
+                   &print_trace_ops, NULL);
+	printk("\n");
 #elif defined(HAVE_SHOW_TASK)
         /* this is exported by lustre kernel version 42 */
         extern void show_task(struct task_struct *);
diff --git a/libcfs/libcfs/watchdog.c b/libcfs/libcfs/watchdog.c
index 24fb8ee..e3fd951 100644
--- a/libcfs/libcfs/watchdog.c
+++ b/libcfs/libcfs/watchdog.c
@@ -103,27 +103,21 @@ static spinlock_t lcw_last_watchdog_lock = SPIN_LOCK_UNLOCKED;
 static void
 lcw_dump(struct lc_watchdog *lcw)
 {
-        cfs_task_t *tsk;
+        ENTRY;
 #if defined(HAVE_TASKLIST_LOCK)
         read_lock(&tasklist_lock);
 #elif defined(HAVE_TASK_RCU)
         rcu_read_lock();
 #else
         CERROR("unable to dump stack because of missing export\n"); 
-        return;
+        RETURN_EXIT;
 #endif
-        ENTRY;
-
-        tsk = find_task_by_pid(lcw->lcw_pid);
-
-        if (tsk == NULL) {
+       if (lcw->lcw_task == NULL) { 
                 CWARN("Process " LPPID " was not found in the task list; "
-                      "watchdog callback may be incomplete\n", lcw->lcw_pid);
-        } else if (tsk != lcw->lcw_task) {
-                CWARN("The current process " LPPID " did not set the watchdog; "
-                      "watchdog callback may be incomplete\n", lcw->lcw_pid);
+                      "watchdog callback may be incomplete\n", 
+                      (int)lcw->lcw_pid);
         } else {
-                libcfs_debug_dumpstack(tsk);
+                libcfs_debug_dumpstack(lcw->lcw_task);
         }
 
 #if defined(HAVE_TASKLIST_LOCK)
diff --git a/lnet/autoconf/lustre-lnet.m4 b/lnet/autoconf/lustre-lnet.m4
index 1d67f1a..997c635 100644
--- a/lnet/autoconf/lustre-lnet.m4
+++ b/lnet/autoconf/lustre-lnet.m4
@@ -892,6 +892,39 @@ AC_SUBST(RALND)
 ])
 
 
+#
+# LN_FUNC_DUMP_TRACE
+#
+# 2.6.23 exports dump_trace() so we can dump_stack() on any task
+# 2.6.24 has stacktrace_ops.address with "reliable" parameter
+#
+AC_DEFUN([LN_FUNC_DUMP_TRACE],
+[LB_CHECK_SYMBOL_EXPORT([dump_trace],
+[kernel/ksyms.c arch/${LINUX_ARCH%_64}/kernel/traps_64.c],[
+	AC_DEFINE(HAVE_DUMP_TRACE, 1, [dump_trace is exported])
+	AC_MSG_CHECKING([whether print_trace_address has reliable argument])
+	tmp_flags="$EXTRA_KCFLAGS"
+	EXTRA_KCFLAGS="-Werror"
+	LB_LINUX_TRY_COMPILE([
+		struct task_struct;
+		struct pt_regs;
+		void print_addr(void *data, unsigned long addr, int reliable);
+		#include <asm/stacktrace.h>
+	],[
+		struct stacktrace_ops ops;
+
+		ops.address = print_addr;
+	],[
+		AC_MSG_RESULT(yes)
+		AC_DEFINE(HAVE_TRACE_ADDRESS_RELIABLE, 1,
+			  [print_trace_address has reliable argument])
+	],[
+		AC_MSG_RESULT(no)
+	],[
+	])
+EXTRA_KCFLAGS="$tmp_flags"
+])
+])
 
 #
 # LN_PROG_LINUX
@@ -912,6 +945,7 @@ LN_CONFIG_O2IB
 LN_CONFIG_RALND
 LN_CONFIG_PTLLND
 LN_CONFIG_MX
+LN_FUNC_DUMP_TRACE
 ])
 
 #