[IA64] MCA/INIT: avoid reading INIT record during INIT event

Reading the INIT record from SAL during the INIT event has proved to be
unreliable, and a source of hangs during INIT processing.  The new
MCA/INIT handlers remove the need to get the INIT record from SAL.
Change salinfo.c so mca.c can just flag that a new record is available,
without having to read the record during INIT processing.  This patch
can be applied without the new MCA/INIT handlers.

Also clean up some usage of NR_CPUS which should have been using
cpu_online().
Signed-off-by: Keith Owens <kaos@sgi.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>

arch/ia64/kernel/salinfo.c

@@ -22,6 +22,11 @@
  *
  * Dec  5 2004	kaos@sgi.com
  *   Standardize which records are cleared automatically.
+ *
+ * Aug 18 2005	kaos@sgi.com
+ *   mca.c may not pass a buffer, a NULL buffer just indicates that a new
+ *   record is available in SAL.
+ *   Replace some NR_CPUS by cpus_online, for hotplug cpu.
  */
 
 #include <linux/types.h>
@@ -193,7 +198,7 @@ shift1_data_saved (struct salinfo_data *data, int shift)
  * The buffer passed from mca.c points to the output from ia64_log_get. This is
  * a persistent buffer but its contents can change between the interrupt and
  * when user space processes the record.  Save the record id to identify
- * changes.
+ * changes.  If the buffer is NULL then just update the bitmap.
  */
 void
 salinfo_log_wakeup(int type, u8 *buffer, u64 size, int irqsafe)
@@ -206,6 +211,7 @@ salinfo_log_wakeup(int type, u8 *buffer, u64 size, int irqsafe)
 
 	BUG_ON(type >= ARRAY_SIZE(salinfo_log_name));
 
+	if (buffer) {
 		if (irqsafe)
 			spin_lock_irqsave(&data_saved_lock, flags);
 		for (i = 0, data_saved = data->data_saved; i < saved_size; ++i, ++data_saved) {
@@ -227,6 +233,7 @@ salinfo_log_wakeup(int type, u8 *buffer, u64 size, int irqsafe)
 		}
 		if (irqsafe)
 			spin_unlock_irqrestore(&data_saved_lock, flags);
+	}
 
 	if (!test_and_set_bit(smp_processor_id(), &data->cpu_event)) {
 		if (irqsafe)
@@ -244,7 +251,7 @@ salinfo_timeout_check(struct salinfo_data *data)
 	int i;
 	if (!data->open)
 		return;
-	for (i = 0; i < NR_CPUS; ++i) {
+	for_each_online_cpu(i) {
 		if (test_bit(i, &data->cpu_event)) {
 			/* double up() is not a problem, user space will see no
 			 * records for the additional "events".
@@ -291,7 +298,7 @@ retry:
 
 	n = data->cpu_check;
 	for (i = 0; i < NR_CPUS; i++) {
-		if (test_bit(n, &data->cpu_event)) {
+		if (test_bit(n, &data->cpu_event) && cpu_online(n)) {
 			cpu = n;
 			break;
 		}
@@ -585,8 +592,7 @@ salinfo_init(void)
 
 		/* we missed any events before now */
 		online = 0;
-		for (j = 0; j < NR_CPUS; j++)
-			if (cpu_online(j)) {
+		for_each_online_cpu(j) {
 			set_bit(j, &data->cpu_event);
 			++online;
 		}