x86, UV: Fix NMI handler for UV platforms

This fixes problems seen on UV systems handling NMIs from the
node controller.

I isolated the "dazed..." messages that I saw earlier to a bug in
the BMC on our platform. It was sending NMIs w/o properly setting
a register that indicated the source of NMI.

So rather than _assuming_ any unhandled NMI came from the UV system
maintenance console (SMC), add a check to verify that the SMC actually
sent the NMI.

Signed-off-by: Jack Steiner <steiner@sgi.com>
Cc: gorcunov@gmail.com
Cc: dzickus@redhat.com
Signed-off-by: Ingo Molnar <mingo@elte.hu>

1 files changed, 43 insertions, 5 deletions

diff --git a/arch/x86/kernel/apic/x2apic_uv_x.c b/arch/x86/kernel/apic/x2apic_uv_x.c
index 33b10a0fc09..7acd2d2ac96 100644
--- a/arch/x86/kernel/apic/x2apic_uv_x.c
+++ b/arch/x86/kernel/apic/x2apic_uv_x.c
@@ -37,6 +37,13 @@
 #include <asm/smp.h>
 #include <asm/x86_init.h>
 #include <asm/emergency-restart.h>
+#include <asm/nmi.h>
+
+/* BMC sets a bit this MMR non-zero before sending an NMI */
+#define UVH_NMI_MMR				UVH_SCRATCH5
+#define UVH_NMI_MMR_CLEAR			(UVH_NMI_MMR + 8)
+#define UV_NMI_PENDING_MASK			(1UL << 63)
+DEFINE_PER_CPU(unsigned long, cpu_last_nmi_count);
 
 DEFINE_PER_CPU(int, x2apic_extra_bits);
 
@@ -642,18 +649,46 @@ void __cpuinit uv_cpu_init(void)
  */
 int uv_handle_nmi(struct notifier_block *self, unsigned long reason, void *data)
 {
+	unsigned long real_uv_nmi;
+	int bid;
+
 	if (reason != DIE_NMIUNKNOWN)
 		return NOTIFY_OK;
 
 	if (in_crash_kexec)
 		/* do nothing if entering the crash kernel */
 		return NOTIFY_OK;
+
 	/*
-	 * Use a lock so only one cpu prints at a time
-	 * to prevent intermixed output.
+	 * Each blade has an MMR that indicates when an NMI has been sent
+	 * to cpus on the blade. If an NMI is detected, atomically
+	 * clear the MMR and update a per-blade NMI count used to
+	 * cause each cpu on the blade to notice a new NMI.
+	 */
+	bid = uv_numa_blade_id();
+	real_uv_nmi = (uv_read_local_mmr(UVH_NMI_MMR) & UV_NMI_PENDING_MASK);
+
+	if (unlikely(real_uv_nmi)) {
+		spin_lock(&uv_blade_info[bid].nmi_lock);
+		real_uv_nmi = (uv_read_local_mmr(UVH_NMI_MMR) & UV_NMI_PENDING_MASK);
+		if (real_uv_nmi) {
+			uv_blade_info[bid].nmi_count++;
+			uv_write_local_mmr(UVH_NMI_MMR_CLEAR, UV_NMI_PENDING_MASK);
+		}
+		spin_unlock(&uv_blade_info[bid].nmi_lock);
+	}
+
+	if (likely(__get_cpu_var(cpu_last_nmi_count) == uv_blade_info[bid].nmi_count))
+		return NOTIFY_DONE;
+
+	__get_cpu_var(cpu_last_nmi_count) = uv_blade_info[bid].nmi_count;
+
+	/*
+	 * Use a lock so only one cpu prints at a time.
+	 * This prevents intermixed output.
 	 */
 	spin_lock(&uv_nmi_lock);
-	pr_info("NMI stack dump cpu %u:\n", smp_processor_id());
+	pr_info("UV NMI stack dump cpu %u:\n", smp_processor_id());
 	dump_stack();
 	spin_unlock(&uv_nmi_lock);
 
@@ -661,7 +696,8 @@ int uv_handle_nmi(struct notifier_block *self, unsigned long reason, void *data)
 }
 
 static struct notifier_block uv_dump_stack_nmi_nb = {
-	.notifier_call	= uv_handle_nmi
+	.notifier_call	= uv_handle_nmi,
+	.priority = NMI_LOCAL_LOW_PRIOR - 1,
 };
 
 void uv_register_nmi_notifier(void)
@@ -720,8 +756,9 @@ void __init uv_system_init(void)
 	printk(KERN_DEBUG "UV: Found %d blades\n", uv_num_possible_blades());
 
 	bytes = sizeof(struct uv_blade_info) * uv_num_possible_blades();
-	uv_blade_info = kmalloc(bytes, GFP_KERNEL);
+	uv_blade_info = kzalloc(bytes, GFP_KERNEL);
 	BUG_ON(!uv_blade_info);
+
 	for (blade = 0; blade < uv_num_possible_blades(); blade++)
 		uv_blade_info[blade].memory_nid = -1;
 
@@ -747,6 +784,7 @@ void __init uv_system_init(void)
 			uv_blade_info[blade].pnode = pnode;
 			uv_blade_info[blade].nr_possible_cpus = 0;
 			uv_blade_info[blade].nr_online_cpus = 0;
+			spin_lock_init(&uv_blade_info[blade].nmi_lock);
 			max_pnode = max(pnode, max_pnode);
 			blade++;
 		}