powerpc/pseries: Enable RAS hotplug events later
authorSam Bobroff <sam.bobroff@au1.ibm.com>
Mon, 12 Feb 2018 00:19:29 +0000 (11:19 +1100)
committerMichael Ellerman <mpe@ellerman.id.au>
Tue, 13 Feb 2018 11:37:46 +0000 (22:37 +1100)
Currently if the kernel receives a memory hot-unplug event early
enough, it may get stuck in an infinite loop in
dissolve_free_huge_pages(). This appears as a stall just after:

  pseries-hotplug-mem: Attempting to hot-remove XX LMB(s) at YYYYYYYY

It appears to be caused by "minimum_order" being uninitialized, due to
init_ras_IRQ() executing before hugetlb_init().

To correct this, extract the part of init_ras_IRQ() that enables
hotplug event processing and place it in the machine_late_initcall
phase, which is guaranteed to be after hugetlb_init() is called.

Signed-off-by: Sam Bobroff <sam.bobroff@au1.ibm.com>
Acked-by: Balbir Singh <bsingharora@gmail.com>
[mpe: Reorder the functions to make the diff readable]
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
arch/powerpc/platforms/pseries/ras.c

index 81d8614e73790b1923a3c8cfd336a2531fee76ce..5e1ef915018208c3511ef0e91c0064c8c9474389 100644 (file)
@@ -48,6 +48,28 @@ static irqreturn_t ras_epow_interrupt(int irq, void *dev_id);
 static irqreturn_t ras_error_interrupt(int irq, void *dev_id);
 
 
+/*
+ * Enable the hotplug interrupt late because processing them may touch other
+ * devices or systems (e.g. hugepages) that have not been initialized at the
+ * subsys stage.
+ */
+int __init init_ras_hotplug_IRQ(void)
+{
+       struct device_node *np;
+
+       /* Hotplug Events */
+       np = of_find_node_by_path("/event-sources/hot-plug-events");
+       if (np != NULL) {
+               if (dlpar_workqueue_init() == 0)
+                       request_event_sources_irqs(np, ras_hotplug_interrupt,
+                                                  "RAS_HOTPLUG");
+               of_node_put(np);
+       }
+
+       return 0;
+}
+machine_late_initcall(pseries, init_ras_hotplug_IRQ);
+
 /*
  * Initialize handlers for the set of interrupts caused by hardware errors
  * and power system events.
@@ -66,15 +88,6 @@ static int __init init_ras_IRQ(void)
                of_node_put(np);
        }
 
-       /* Hotplug Events */
-       np = of_find_node_by_path("/event-sources/hot-plug-events");
-       if (np != NULL) {
-               if (dlpar_workqueue_init() == 0)
-                       request_event_sources_irqs(np, ras_hotplug_interrupt,
-                                          "RAS_HOTPLUG");
-               of_node_put(np);
-       }
-
        /* EPOW Events */
        np = of_find_node_by_path("/event-sources/epow-events");
        if (np != NULL) {