powerpc/eeh: Fix stale cached primary bus
authorGavin Shan <gwshan@linux.vnet.ibm.com>
Tue, 9 Feb 2016 04:50:21 +0000 (15:50 +1100)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Thu, 25 Feb 2016 20:01:18 +0000 (12:01 -0800)
commit 05ba75f848647135f063199dc0e9f40fee769724 upstream.

When PE is created, its primary bus is cached to pe->bus. At later
point, the cached primary bus is returned from eeh_pe_bus_get().
However, we could get stale cached primary bus and run into kernel
crash in one case: full hotplug as part of fenced PHB error recovery
releases all PCI busses under the PHB at unplugging time and recreate
them at plugging time. pe->bus is still dereferencing the PCI bus
that was released.

This adds another PE flag (EEH_PE_PRI_BUS) to represent the validity
of pe->bus. pe->bus is updated when its first child EEH device is
online and the flag is set. Before unplugging in full hotplug for
error recovery, the flag is cleared.

Fixes: 8cdb2833 ("powerpc/eeh: Trace PCI bus from PE")
Reported-by: Andrew Donnellan <andrew.donnellan@au1.ibm.com>
Reported-by: Pradipta Ghosh <pradghos@in.ibm.com>
Signed-off-by: Gavin Shan <gwshan@linux.vnet.ibm.com>
Tested-by: Andrew Donnellan <andrew.donnellan@au1.ibm.com>
Signed-off-by: Michael Ellerman <mpe@ellerman.id.au>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
arch/powerpc/include/asm/eeh.h
arch/powerpc/kernel/eeh_driver.c
arch/powerpc/kernel/eeh_pe.c
arch/powerpc/platforms/powernv/eeh-powernv.c

index c5eb86f..867c39b 100644 (file)
@@ -81,6 +81,7 @@ struct pci_dn;
 #define EEH_PE_KEEP            (1 << 8)        /* Keep PE on hotplug   */
 #define EEH_PE_CFG_RESTRICTED  (1 << 9)        /* Block config on error */
 #define EEH_PE_REMOVED         (1 << 10)       /* Removed permanently  */
+#define EEH_PE_PRI_BUS         (1 << 11)       /* Cached primary bus   */
 
 struct eeh_pe {
        int type;                       /* PE type: PHB/Bus/Device      */
index 8d14feb..f69ecaa 100644 (file)
@@ -564,6 +564,7 @@ static int eeh_reset_device(struct eeh_pe *pe, struct pci_bus *bus)
         */
        eeh_pe_state_mark(pe, EEH_PE_KEEP);
        if (bus) {
+               eeh_pe_state_clear(pe, EEH_PE_PRI_BUS);
                pci_lock_rescan_remove();
                pcibios_remove_pci_devices(bus);
                pci_unlock_rescan_remove();
@@ -803,6 +804,7 @@ perm_error:
         * the their PCI config any more.
         */
        if (frozen_bus) {
+               eeh_pe_state_clear(pe, EEH_PE_PRI_BUS);
                eeh_pe_dev_mode_mark(pe, EEH_DEV_REMOVED);
 
                pci_lock_rescan_remove();
@@ -886,6 +888,7 @@ static void eeh_handle_special_event(void)
                                        continue;
 
                                /* Notify all devices to be down */
+                               eeh_pe_state_clear(pe, EEH_PE_PRI_BUS);
                                bus = eeh_pe_bus_get(phb_pe);
                                eeh_pe_dev_traverse(pe,
                                        eeh_report_failure, NULL);
index ca9e537..98f8180 100644 (file)
@@ -928,7 +928,7 @@ struct pci_bus *eeh_pe_bus_get(struct eeh_pe *pe)
                bus = pe->phb->bus;
        } else if (pe->type & EEH_PE_BUS ||
                   pe->type & EEH_PE_DEVICE) {
-               if (pe->bus) {
+               if (pe->state & EEH_PE_PRI_BUS) {
                        bus = pe->bus;
                        goto out;
                }
index e1c9072..2ba6025 100644 (file)
@@ -444,9 +444,12 @@ static void *pnv_eeh_probe(struct pci_dn *pdn, void *data)
         * PCI devices of the PE are expected to be removed prior
         * to PE reset.
         */
-       if (!edev->pe->bus)
+       if (!(edev->pe->state & EEH_PE_PRI_BUS)) {
                edev->pe->bus = pci_find_bus(hose->global_number,
                                             pdn->busno);
+               if (edev->pe->bus)
+                       edev->pe->state |= EEH_PE_PRI_BUS;
+       }
 
        /*
         * Enable EEH explicitly so that we will do EEH check