[IA64] - Altix: hotplug after intr redirect can crash system

When redirecting a device interrupt on SN, not all links
between platform specific structures are being updated.
This can result in a system crash if an interrupt
redirection is followed by an unplug of that device.

The complete fix also requires a prom update. Though,
this patch is backward compatable and not dependent on
the prom patch.

Signed-off-by: John Keller <jpk@sgi.com>
Signed-off-by: Tony Luck <tony.luck@intel.com>

diff --git a/arch/ia64/sn/kernel/irq.c b/arch/ia64/sn/kernel/irq.c
index 8d2a1bf..7f6d236 100644 (file)
--- a/arch/ia64/sn/kernel/irq.c
+++ b/arch/ia64/sn/kernel/irq.c
@@ -59,6 +59,22 @@ void sn_intr_free(nasid_t local_nasid, int local_widget,
                        (u64) sn_irq_info->irq_cookie, 0, 0);
 }
 
+u64 sn_intr_redirect(nasid_t local_nasid, int local_widget,
+                     struct sn_irq_info *sn_irq_info,
+                     nasid_t req_nasid, int req_slice)
+{
+       struct ia64_sal_retval ret_stuff;
+       ret_stuff.status = 0;
+       ret_stuff.v0 = 0;
+
+       SAL_CALL_NOLOCK(ret_stuff, (u64) SN_SAL_IOIF_INTERRUPT,
+                       (u64) SAL_INTR_REDIRECT, (u64) local_nasid,
+                       (u64) local_widget, __pa(sn_irq_info),
+                       (u64) req_nasid, (u64) req_slice, 0);
+
+       return ret_stuff.status;
+}
+
 static unsigned int sn_startup_irq(unsigned int irq)
 {
        return 0;
@@ -127,15 +143,8 @@ struct sn_irq_info *sn_retarget_vector(struct sn_irq_info *sn_irq_info,
        struct sn_irq_info *new_irq_info;
        struct sn_pcibus_provider *pci_provider;
 
-       new_irq_info = kmalloc(sizeof(struct sn_irq_info), GFP_ATOMIC);
-       if (new_irq_info == NULL)
-               return NULL;
-
-       memcpy(new_irq_info, sn_irq_info, sizeof(struct sn_irq_info));
-
-       bridge = (u64) new_irq_info->irq_bridge;
+       bridge = (u64) sn_irq_info->irq_bridge;
        if (!bridge) {
-               kfree(new_irq_info);
                return NULL; /* irq is not a device interrupt */
        }
 
@@ -145,8 +154,25 @@ struct sn_irq_info *sn_retarget_vector(struct sn_irq_info *sn_irq_info,
                local_widget = TIO_SWIN_WIDGETNUM(bridge);
        else
                local_widget = SWIN_WIDGETNUM(bridge);
-
        vector = sn_irq_info->irq_irq;
+
+       /* Make use of SAL_INTR_REDIRECT if PROM supports it */
+       status = sn_intr_redirect(local_nasid, local_widget, sn_irq_info, nasid, slice);
+       if (!status) {
+               new_irq_info = sn_irq_info;
+               goto finish_up;
+       }
+
+       /*
+        * PROM does not support SAL_INTR_REDIRECT, or it failed.
+        * Revert to old method.
+        */
+       new_irq_info = kmalloc(sizeof(struct sn_irq_info), GFP_ATOMIC);
+       if (new_irq_info == NULL)
+               return NULL;
+
+       memcpy(new_irq_info, sn_irq_info, sizeof(struct sn_irq_info));
+
        /* Free the old PROM new_irq_info structure */
        sn_intr_free(local_nasid, local_widget, new_irq_info);
        unregister_intr_pda(new_irq_info);
@@ -162,11 +188,18 @@ struct sn_irq_info *sn_retarget_vector(struct sn_irq_info *sn_irq_info,
                return NULL;
        }
 
+       register_intr_pda(new_irq_info);
+       spin_lock(&sn_irq_info_lock);
+       list_replace_rcu(&sn_irq_info->list, &new_irq_info->list);
+       spin_unlock(&sn_irq_info_lock);
+       call_rcu(&sn_irq_info->rcu, sn_irq_info_free);
+
+
+finish_up:
        /* Update kernels new_irq_info with new target info */
        cpuid = nasid_slice_to_cpuid(new_irq_info->irq_nasid,
                                     new_irq_info->irq_slice);
        new_irq_info->irq_cpuid = cpuid;
-       register_intr_pda(new_irq_info);
 
        pci_provider = sn_pci_provider[new_irq_info->irq_bridge_type];
 
@@ -178,11 +211,6 @@ struct sn_irq_info *sn_retarget_vector(struct sn_irq_info *sn_irq_info,
            pci_provider && pci_provider->target_interrupt)
                (pci_provider->target_interrupt)(new_irq_info);
 
-       spin_lock(&sn_irq_info_lock);
-       list_replace_rcu(&sn_irq_info->list, &new_irq_info->list);
-       spin_unlock(&sn_irq_info_lock);
-       call_rcu(&sn_irq_info->rcu, sn_irq_info_free);
-
 #ifdef CONFIG_SMP
        cpuphys = cpu_physical_id(cpuid);
        set_irq_affinity_info((vector & 0xff), cpuphys, 0);

diff --git a/include/asm-ia64/sn/sn_sal.h b/include/asm-ia64/sn/sn_sal.h
index 2c4004e..291e8ce 100644 (file)
--- a/include/asm-ia64/sn/sn_sal.h
+++ b/include/asm-ia64/sn/sn_sal.h
@@ -106,6 +106,7 @@
 /* interrupt handling */
 #define SAL_INTR_ALLOC         1
 #define SAL_INTR_FREE          2
+#define SAL_INTR_REDIRECT      3
 
 /*
  * operations available on the generic SN_SAL_SYSCTL_OP