Merge remote-tracking branch 'pci/pci/gavin-window-alignment' into next
[firefly-linux-kernel-4.4.55.git] / arch / powerpc / platforms / pseries / eeh.c
index ecd394cf34e604b24b71138d481f570cd0d926ac..18c168b752da2e24785ec1d5a6c0a879577df798 100644 (file)
@@ -92,6 +92,20 @@ struct eeh_ops *eeh_ops = NULL;
 int eeh_subsystem_enabled;
 EXPORT_SYMBOL(eeh_subsystem_enabled);
 
+/*
+ * EEH probe mode support. The intention is to support multiple
+ * platforms for EEH. Some platforms like pSeries do PCI emunation
+ * based on device tree. However, other platforms like powernv probe
+ * PCI devices from hardware. The flag is used to distinguish that.
+ * In addition, struct eeh_ops::probe would be invoked for particular
+ * OF node or PCI device so that the corresponding PE would be created
+ * there.
+ */
+int eeh_probe_mode;
+
+/* Global EEH mutex */
+DEFINE_MUTEX(eeh_mutex);
+
 /* Lock to avoid races due to multiple reports of an error */
 static DEFINE_RAW_SPINLOCK(confirm_error_lock);
 
@@ -204,22 +218,12 @@ static size_t eeh_gather_pci_data(struct eeh_dev *edev, char * buf, size_t len)
                }
        }
 
-       /* Gather status on devices under the bridge */
-       if (dev->class >> 16 == PCI_BASE_CLASS_BRIDGE) {
-               struct device_node *child;
-
-               for_each_child_of_node(dn, child) {
-                       if (of_node_to_eeh_dev(child))
-                               n += eeh_gather_pci_data(of_node_to_eeh_dev(child), buf+n, len-n);
-               }
-       }
-
        return n;
 }
 
 /**
  * eeh_slot_error_detail - Generate combined log including driver log and error log
- * @edev: device to report error log for
+ * @pe: EEH PE
  * @severity: temporary or permanent error log
  *
  * This routine should be called to generate the combined log, which
@@ -227,17 +231,22 @@ static size_t eeh_gather_pci_data(struct eeh_dev *edev, char * buf, size_t len)
  * out from the config space of the corresponding PCI device, while
  * the error log is fetched through platform dependent function call.
  */
-void eeh_slot_error_detail(struct eeh_dev *edev, int severity)
+void eeh_slot_error_detail(struct eeh_pe *pe, int severity)
 {
        size_t loglen = 0;
-       pci_regs_buf[0] = 0;
+       struct eeh_dev *edev;
 
-       eeh_pci_enable(edev, EEH_OPT_THAW_MMIO);
-       eeh_ops->configure_bridge(eeh_dev_to_of_node(edev));
-       eeh_restore_bars(edev);
-       loglen = eeh_gather_pci_data(edev, pci_regs_buf, EEH_PCI_REGS_LOG_LEN);
+       eeh_pci_enable(pe, EEH_OPT_THAW_MMIO);
+       eeh_ops->configure_bridge(pe);
+       eeh_pe_restore_bars(pe);
 
-       eeh_ops->get_log(eeh_dev_to_of_node(edev), severity, pci_regs_buf, loglen);
+       pci_regs_buf[0] = 0;
+       eeh_pe_for_each_dev(pe, edev) {
+               loglen += eeh_gather_pci_data(edev, pci_regs_buf,
+                               EEH_PCI_REGS_LOG_LEN);
+        }
+
+       eeh_ops->get_log(pe, severity, pci_regs_buf, loglen);
 }
 
 /**
@@ -261,126 +270,8 @@ static inline unsigned long eeh_token_to_phys(unsigned long token)
 }
 
 /**
- * eeh_find_device_pe - Retrieve the PE for the given device
- * @dn: device node
- *
- * Return the PE under which this device lies
- */
-struct device_node *eeh_find_device_pe(struct device_node *dn)
-{
-       while (dn->parent && of_node_to_eeh_dev(dn->parent) &&
-              (of_node_to_eeh_dev(dn->parent)->mode & EEH_MODE_SUPPORTED)) {
-               dn = dn->parent;
-       }
-       return dn;
-}
-
-/**
- * __eeh_mark_slot - Mark all child devices as failed
- * @parent: parent device
- * @mode_flag: failure flag
- *
- * Mark all devices that are children of this device as failed.
- * Mark the device driver too, so that it can see the failure
- * immediately; this is critical, since some drivers poll
- * status registers in interrupts ... If a driver is polling,
- * and the slot is frozen, then the driver can deadlock in
- * an interrupt context, which is bad.
- */
-static void __eeh_mark_slot(struct device_node *parent, int mode_flag)
-{
-       struct device_node *dn;
-
-       for_each_child_of_node(parent, dn) {
-               if (of_node_to_eeh_dev(dn)) {
-                       /* Mark the pci device driver too */
-                       struct pci_dev *dev = of_node_to_eeh_dev(dn)->pdev;
-
-                       of_node_to_eeh_dev(dn)->mode |= mode_flag;
-
-                       if (dev && dev->driver)
-                               dev->error_state = pci_channel_io_frozen;
-
-                       __eeh_mark_slot(dn, mode_flag);
-               }
-       }
-}
-
-/**
- * eeh_mark_slot - Mark the indicated device and its children as failed
- * @dn: parent device
- * @mode_flag: failure flag
- *
- * Mark the indicated device and its child devices as failed.
- * The device drivers are marked as failed as well.
- */
-void eeh_mark_slot(struct device_node *dn, int mode_flag)
-{
-       struct pci_dev *dev;
-       dn = eeh_find_device_pe(dn);
-
-       /* Back up one, since config addrs might be shared */
-       if (!pcibios_find_pci_bus(dn) && of_node_to_eeh_dev(dn->parent))
-               dn = dn->parent;
-
-       of_node_to_eeh_dev(dn)->mode |= mode_flag;
-
-       /* Mark the pci device too */
-       dev = of_node_to_eeh_dev(dn)->pdev;
-       if (dev)
-               dev->error_state = pci_channel_io_frozen;
-
-       __eeh_mark_slot(dn, mode_flag);
-}
-
-/**
- * __eeh_clear_slot - Clear failure flag for the child devices
- * @parent: parent device
- * @mode_flag: flag to be cleared
- *
- * Clear failure flag for the child devices.
- */
-static void __eeh_clear_slot(struct device_node *parent, int mode_flag)
-{
-       struct device_node *dn;
-
-       for_each_child_of_node(parent, dn) {
-               if (of_node_to_eeh_dev(dn)) {
-                       of_node_to_eeh_dev(dn)->mode &= ~mode_flag;
-                       of_node_to_eeh_dev(dn)->check_count = 0;
-                       __eeh_clear_slot(dn, mode_flag);
-               }
-       }
-}
-
-/**
- * eeh_clear_slot - Clear failure flag for the indicated device and its children
- * @dn: parent device
- * @mode_flag: flag to be cleared
- *
- * Clear failure flag for the indicated device and its children.
- */
-void eeh_clear_slot(struct device_node *dn, int mode_flag)
-{
-       unsigned long flags;
-       raw_spin_lock_irqsave(&confirm_error_lock, flags);
-       
-       dn = eeh_find_device_pe(dn);
-       
-       /* Back up one, since config addrs might be shared */
-       if (!pcibios_find_pci_bus(dn) && of_node_to_eeh_dev(dn->parent))
-               dn = dn->parent;
-
-       of_node_to_eeh_dev(dn)->mode &= ~mode_flag;
-       of_node_to_eeh_dev(dn)->check_count = 0;
-       __eeh_clear_slot(dn, mode_flag);
-       raw_spin_unlock_irqrestore(&confirm_error_lock, flags);
-}
-
-/**
- * eeh_dn_check_failure - Check if all 1's data is due to EEH slot freeze
- * @dn: device node
- * @dev: pci device, if known
+ * eeh_dev_check_failure - Check if all 1's data is due to EEH slot freeze
+ * @edev: eeh device
  *
  * Check for an EEH failure for the given device node.  Call this
  * routine if the result of a read was all 0xff's and you want to
@@ -392,11 +283,13 @@ void eeh_clear_slot(struct device_node *dn, int mode_flag)
  *
  * It is safe to call this routine in an interrupt context.
  */
-int eeh_dn_check_failure(struct device_node *dn, struct pci_dev *dev)
+int eeh_dev_check_failure(struct eeh_dev *edev)
 {
        int ret;
        unsigned long flags;
-       struct eeh_dev *edev;
+       struct device_node *dn;
+       struct pci_dev *dev;
+       struct eeh_pe *pe;
        int rc = 0;
        const char *location;
 
@@ -405,23 +298,23 @@ int eeh_dn_check_failure(struct device_node *dn, struct pci_dev *dev)
        if (!eeh_subsystem_enabled)
                return 0;
 
-       if (!dn) {
+       if (!edev) {
                eeh_stats.no_dn++;
                return 0;
        }
-       dn = eeh_find_device_pe(dn);
-       edev = of_node_to_eeh_dev(dn);
+       dn = eeh_dev_to_of_node(edev);
+       dev = eeh_dev_to_pci_dev(edev);
+       pe = edev->pe;
 
        /* Access to IO BARs might get this far and still not want checking. */
-       if (!(edev->mode & EEH_MODE_SUPPORTED) ||
-           edev->mode & EEH_MODE_NOCHECK) {
+       if (!pe) {
                eeh_stats.ignored_check++;
-               pr_debug("EEH: Ignored check (%x) for %s %s\n",
-                       edev->mode, eeh_pci_name(dev), dn->full_name);
+               pr_debug("EEH: Ignored check for %s %s\n",
+                       eeh_pci_name(dev), dn->full_name);
                return 0;
        }
 
-       if (!edev->config_addr && !edev->pe_config_addr) {
+       if (!pe->addr && !pe->config_addr) {
                eeh_stats.no_cfg_addr++;
                return 0;
        }
@@ -434,13 +327,13 @@ int eeh_dn_check_failure(struct device_node *dn, struct pci_dev *dev)
         */
        raw_spin_lock_irqsave(&confirm_error_lock, flags);
        rc = 1;
-       if (edev->mode & EEH_MODE_ISOLATED) {
-               edev->check_count++;
-               if (edev->check_count % EEH_MAX_FAILS == 0) {
+       if (pe->state & EEH_PE_ISOLATED) {
+               pe->check_count++;
+               if (pe->check_count % EEH_MAX_FAILS == 0) {
                        location = of_get_property(dn, "ibm,loc-code", NULL);
                        printk(KERN_ERR "EEH: %d reads ignored for recovering device at "
                                "location=%s driver=%s pci addr=%s\n",
-                               edev->check_count, location,
+                               pe->check_count, location,
                                eeh_driver_name(dev), eeh_pci_name(dev));
                        printk(KERN_ERR "EEH: Might be infinite loop in %s driver\n",
                                eeh_driver_name(dev));
@@ -456,7 +349,7 @@ int eeh_dn_check_failure(struct device_node *dn, struct pci_dev *dev)
         * function zero of a multi-function device.
         * In any case they must share a common PHB.
         */
-       ret = eeh_ops->get_state(dn, NULL);
+       ret = eeh_ops->get_state(pe, NULL);
 
        /* Note that config-io to empty slots may fail;
         * they are empty when they don't have children.
@@ -469,7 +362,7 @@ int eeh_dn_check_failure(struct device_node *dn, struct pci_dev *dev)
            (ret & (EEH_STATE_MMIO_ACTIVE | EEH_STATE_DMA_ACTIVE)) ==
            (EEH_STATE_MMIO_ACTIVE | EEH_STATE_DMA_ACTIVE)) {
                eeh_stats.false_positives++;
-               edev->false_positives ++;
+               pe->false_positives++;
                rc = 0;
                goto dn_unlock;
        }
@@ -480,10 +373,10 @@ int eeh_dn_check_failure(struct device_node *dn, struct pci_dev *dev)
         * with other functions on this device, and functions under
         * bridges.
         */
-       eeh_mark_slot(dn, EEH_MODE_ISOLATED);
+       eeh_pe_state_mark(pe, EEH_PE_ISOLATED);
        raw_spin_unlock_irqrestore(&confirm_error_lock, flags);
 
-       eeh_send_failure_event(edev);
+       eeh_send_failure_event(pe);
 
        /* Most EEH events are due to device driver bugs.  Having
         * a stack trace will help the device-driver authors figure
@@ -497,7 +390,7 @@ dn_unlock:
        return rc;
 }
 
-EXPORT_SYMBOL_GPL(eeh_dn_check_failure);
+EXPORT_SYMBOL_GPL(eeh_dev_check_failure);
 
 /**
  * eeh_check_failure - Check if all 1's data is due to EEH slot freeze
@@ -514,21 +407,19 @@ EXPORT_SYMBOL_GPL(eeh_dn_check_failure);
 unsigned long eeh_check_failure(const volatile void __iomem *token, unsigned long val)
 {
        unsigned long addr;
-       struct pci_dev *dev;
-       struct device_node *dn;
+       struct eeh_dev *edev;
 
        /* Finding the phys addr + pci device; this is pretty quick. */
        addr = eeh_token_to_phys((unsigned long __force) token);
-       dev = pci_addr_cache_get_device(addr);
-       if (!dev) {
+       edev = eeh_addr_cache_get_dev(addr);
+       if (!edev) {
                eeh_stats.no_device++;
                return val;
        }
 
-       dn = pci_device_to_OF_node(dev);
-       eeh_dn_check_failure(dn, dev);
+       eeh_dev_check_failure(edev);
 
-       pci_dev_put(dev);
+       pci_dev_put(eeh_dev_to_pci_dev(edev));
        return val;
 }
 
@@ -537,23 +428,22 @@ EXPORT_SYMBOL(eeh_check_failure);
 
 /**
  * eeh_pci_enable - Enable MMIO or DMA transfers for this slot
- * @edev: pci device node
+ * @pe: EEH PE
  *
  * This routine should be called to reenable frozen MMIO or DMA
  * so that it would work correctly again. It's useful while doing
  * recovery or log collection on the indicated device.
  */
-int eeh_pci_enable(struct eeh_dev *edev, int function)
+int eeh_pci_enable(struct eeh_pe *pe, int function)
 {
        int rc;
-       struct device_node *dn = eeh_dev_to_of_node(edev);
 
-       rc = eeh_ops->set_option(dn, function);
+       rc = eeh_ops->set_option(pe, function);
        if (rc)
-               printk(KERN_WARNING "EEH: Unexpected state change %d, err=%d dn=%s\n",
-                       function, rc, dn->full_name);
+               pr_warning("%s: Unexpected state change %d on PHB#%d-PE#%x, err=%d\n",
+                       __func__, function, pe->phb->global_number, pe->addr, rc);
 
-       rc = eeh_ops->wait_state(dn, PCI_BUS_RESET_WAIT_MSEC);
+       rc = eeh_ops->wait_state(pe, PCI_BUS_RESET_WAIT_MSEC);
        if (rc > 0 && (rc & EEH_STATE_MMIO_ENABLED) &&
           (function == EEH_OPT_THAW_MMIO))
                return 0;
@@ -571,17 +461,24 @@ int eeh_pci_enable(struct eeh_dev *edev, int function)
  */
 int pcibios_set_pcie_reset_state(struct pci_dev *dev, enum pcie_reset_state state)
 {
-       struct device_node *dn = pci_device_to_OF_node(dev);
+       struct eeh_dev *edev = pci_dev_to_eeh_dev(dev);
+       struct eeh_pe *pe = edev->pe;
+
+       if (!pe) {
+               pr_err("%s: No PE found on PCI device %s\n",
+                       __func__, pci_name(dev));
+               return -EINVAL;
+       }
 
        switch (state) {
        case pcie_deassert_reset:
-               eeh_ops->reset(dn, EEH_RESET_DEACTIVATE);
+               eeh_ops->reset(pe, EEH_RESET_DEACTIVATE);
                break;
        case pcie_hot_reset:
-               eeh_ops->reset(dn, EEH_RESET_HOT);
+               eeh_ops->reset(pe, EEH_RESET_HOT);
                break;
        case pcie_warm_reset:
-               eeh_ops->reset(dn, EEH_RESET_FUNDAMENTAL);
+               eeh_ops->reset(pe, EEH_RESET_FUNDAMENTAL);
                break;
        default:
                return -EINVAL;
@@ -591,66 +488,37 @@ int pcibios_set_pcie_reset_state(struct pci_dev *dev, enum pcie_reset_state stat
 }
 
 /**
- * __eeh_set_pe_freset - Check the required reset for child devices
- * @parent: parent device
- * @freset: return value
- *
- * Each device might have its preferred reset type: fundamental or
- * hot reset. The routine is used to collect the information from
- * the child devices so that they could be reset accordingly.
- */
-void __eeh_set_pe_freset(struct device_node *parent, unsigned int *freset)
-{
-       struct device_node *dn;
-
-       for_each_child_of_node(parent, dn) {
-               if (of_node_to_eeh_dev(dn)) {
-                       struct pci_dev *dev = of_node_to_eeh_dev(dn)->pdev;
-
-                       if (dev && dev->driver)
-                               *freset |= dev->needs_freset;
-
-                       __eeh_set_pe_freset(dn, freset);
-               }
-       }
-}
-
-/**
- * eeh_set_pe_freset - Check the required reset for the indicated device and its children
- * @dn: parent device
- * @freset: return value
+ * eeh_set_pe_freset - Check the required reset for the indicated device
+ * @data: EEH device
+ * @flag: return value
  *
  * Each device might have its preferred reset type: fundamental or
  * hot reset. The routine is used to collected the information for
  * the indicated device and its children so that the bunch of the
  * devices could be reset properly.
  */
-void eeh_set_pe_freset(struct device_node *dn, unsigned int *freset)
+static void *eeh_set_dev_freset(void *data, void *flag)
 {
        struct pci_dev *dev;
-       dn = eeh_find_device_pe(dn);
-
-       /* Back up one, since config addrs might be shared */
-       if (!pcibios_find_pci_bus(dn) && of_node_to_eeh_dev(dn->parent))
-               dn = dn->parent;
+       unsigned int *freset = (unsigned int *)flag;
+       struct eeh_dev *edev = (struct eeh_dev *)data;
 
-       dev = of_node_to_eeh_dev(dn)->pdev;
+       dev = eeh_dev_to_pci_dev(edev);
        if (dev)
                *freset |= dev->needs_freset;
 
-       __eeh_set_pe_freset(dn, freset);
+       return NULL;
 }
 
 /**
  * eeh_reset_pe_once - Assert the pci #RST line for 1/4 second
- * @edev: pci device node to be reset.
+ * @pe: EEH PE
  *
  * Assert the PCI #RST line for 1/4 second.
  */
-static void eeh_reset_pe_once(struct eeh_dev *edev)
+static void eeh_reset_pe_once(struct eeh_pe *pe)
 {
        unsigned int freset = 0;
-       struct device_node *dn = eeh_dev_to_of_node(edev);
 
        /* Determine type of EEH reset required for
         * Partitionable Endpoint, a hot-reset (1)
@@ -658,12 +526,12 @@ static void eeh_reset_pe_once(struct eeh_dev *edev)
         * A fundamental reset required by any device under
         * Partitionable Endpoint trumps hot-reset.
         */
-       eeh_set_pe_freset(dn, &freset);
+       eeh_pe_dev_traverse(pe, eeh_set_dev_freset, &freset);
 
        if (freset)
-               eeh_ops->reset(dn, EEH_RESET_FUNDAMENTAL);
+               eeh_ops->reset(pe, EEH_RESET_FUNDAMENTAL);
        else
-               eeh_ops->reset(dn, EEH_RESET_HOT);
+               eeh_ops->reset(pe, EEH_RESET_HOT);
 
        /* The PCI bus requires that the reset be held high for at least
         * a 100 milliseconds. We wait a bit longer 'just in case'.
@@ -675,9 +543,9 @@ static void eeh_reset_pe_once(struct eeh_dev *edev)
         * pci slot reset line is dropped. Make sure we don't miss
         * these, and clear the flag now.
         */
-       eeh_clear_slot(dn, EEH_MODE_ISOLATED);
+       eeh_pe_state_clear(pe, EEH_PE_ISOLATED);
 
-       eeh_ops->reset(dn, EEH_RESET_DEACTIVATE);
+       eeh_ops->reset(pe, EEH_RESET_DEACTIVATE);
 
        /* After a PCI slot has been reset, the PCI Express spec requires
         * a 1.5 second idle time for the bus to stabilize, before starting
@@ -689,116 +557,36 @@ static void eeh_reset_pe_once(struct eeh_dev *edev)
 
 /**
  * eeh_reset_pe - Reset the indicated PE
- * @edev: PCI device associated EEH device
+ * @pe: EEH PE
  *
  * This routine should be called to reset indicated device, including
  * PE. A PE might include multiple PCI devices and sometimes PCI bridges
  * might be involved as well.
  */
-int eeh_reset_pe(struct eeh_dev *edev)
+int eeh_reset_pe(struct eeh_pe *pe)
 {
        int i, rc;
-       struct device_node *dn = eeh_dev_to_of_node(edev);
 
        /* Take three shots at resetting the bus */
        for (i=0; i<3; i++) {
-               eeh_reset_pe_once(edev);
+               eeh_reset_pe_once(pe);
 
-               rc = eeh_ops->wait_state(dn, PCI_BUS_RESET_WAIT_MSEC);
+               rc = eeh_ops->wait_state(pe, PCI_BUS_RESET_WAIT_MSEC);
                if (rc == (EEH_STATE_MMIO_ACTIVE | EEH_STATE_DMA_ACTIVE))
                        return 0;
 
                if (rc < 0) {
-                       printk(KERN_ERR "EEH: unrecoverable slot failure %s\n",
-                              dn->full_name);
+                       pr_err("%s: Unrecoverable slot failure on PHB#%d-PE#%x",
+                               __func__, pe->phb->global_number, pe->addr);
                        return -1;
                }
-               printk(KERN_ERR "EEH: bus reset %d failed on slot %s, rc=%d\n",
-                      i+1, dn->full_name, rc);
+               pr_err("EEH: bus reset %d failed on PHB#%d-PE#%x, rc=%d\n",
+                       i+1, pe->phb->global_number, pe->addr, rc);
        }
 
        return -1;
 }
 
-/** Save and restore of PCI BARs
- *
- * Although firmware will set up BARs during boot, it doesn't
- * set up device BAR's after a device reset, although it will,
- * if requested, set up bridge configuration. Thus, we need to
- * configure the PCI devices ourselves.  
- */
-
-/**
- * eeh_restore_one_device_bars - Restore the Base Address Registers for one device
- * @edev: PCI device associated EEH device
- *
- * Loads the PCI configuration space base address registers,
- * the expansion ROM base address, the latency timer, and etc.
- * from the saved values in the device node.
- */
-static inline void eeh_restore_one_device_bars(struct eeh_dev *edev)
-{
-       int i;
-       u32 cmd;
-       struct device_node *dn = eeh_dev_to_of_node(edev);
-
-       if (!edev->phb)
-               return;
-
-       for (i=4; i<10; i++) {
-               eeh_ops->write_config(dn, i*4, 4, edev->config_space[i]);
-       }
-
-       /* 12 == Expansion ROM Address */
-       eeh_ops->write_config(dn, 12*4, 4, edev->config_space[12]);
-
-#define BYTE_SWAP(OFF) (8*((OFF)/4)+3-(OFF))
-#define SAVED_BYTE(OFF) (((u8 *)(edev->config_space))[BYTE_SWAP(OFF)])
-
-       eeh_ops->write_config(dn, PCI_CACHE_LINE_SIZE, 1,
-                   SAVED_BYTE(PCI_CACHE_LINE_SIZE));
-
-       eeh_ops->write_config(dn, PCI_LATENCY_TIMER, 1,
-                   SAVED_BYTE(PCI_LATENCY_TIMER));
-
-       /* max latency, min grant, interrupt pin and line */
-       eeh_ops->write_config(dn, 15*4, 4, edev->config_space[15]);
-
-       /* Restore PERR & SERR bits, some devices require it,
-        * don't touch the other command bits
-        */
-       eeh_ops->read_config(dn, PCI_COMMAND, 4, &cmd);
-       if (edev->config_space[1] & PCI_COMMAND_PARITY)
-               cmd |= PCI_COMMAND_PARITY;
-       else
-               cmd &= ~PCI_COMMAND_PARITY;
-       if (edev->config_space[1] & PCI_COMMAND_SERR)
-               cmd |= PCI_COMMAND_SERR;
-       else
-               cmd &= ~PCI_COMMAND_SERR;
-       eeh_ops->write_config(dn, PCI_COMMAND, 4, cmd);
-}
-
-/**
- * eeh_restore_bars - Restore the PCI config space info
- * @edev: EEH device
- *
- * This routine performs a recursive walk to the children
- * of this device as well.
- */
-void eeh_restore_bars(struct eeh_dev *edev)
-{
-       struct device_node *dn;
-       if (!edev)
-               return;
-       
-       if ((edev->mode & EEH_MODE_SUPPORTED) && !IS_BRIDGE(edev->class_code))
-               eeh_restore_one_device_bars(edev);
-
-       for_each_child_of_node(eeh_dev_to_of_node(edev), dn)
-               eeh_restore_bars(of_node_to_eeh_dev(dn));
-}
-
 /**
  * eeh_save_bars - Save device bars
  * @edev: PCI device associated EEH device
@@ -808,7 +596,7 @@ void eeh_restore_bars(struct eeh_dev *edev)
  * PCI devices are added individually; but, for the restore,
  * an entire slot is reset at a time.
  */
-static void eeh_save_bars(struct eeh_dev *edev)
+void eeh_save_bars(struct eeh_dev *edev)
 {
        int i;
        struct device_node *dn;
@@ -821,102 +609,6 @@ static void eeh_save_bars(struct eeh_dev *edev)
                eeh_ops->read_config(dn, i * 4, 4, &edev->config_space[i]);
 }
 
-/**
- * eeh_early_enable - Early enable EEH on the indicated device
- * @dn: device node
- * @data: BUID
- *
- * Enable EEH functionality on the specified PCI device. The function
- * is expected to be called before real PCI probing is done. However,
- * the PHBs have been initialized at this point.
- */
-static void *eeh_early_enable(struct device_node *dn, void *data)
-{
-       int ret;
-       const u32 *class_code = of_get_property(dn, "class-code", NULL);
-       const u32 *vendor_id = of_get_property(dn, "vendor-id", NULL);
-       const u32 *device_id = of_get_property(dn, "device-id", NULL);
-       const u32 *regs;
-       int enable;
-       struct eeh_dev *edev = of_node_to_eeh_dev(dn);
-
-       edev->class_code = 0;
-       edev->mode = 0;
-       edev->check_count = 0;
-       edev->freeze_count = 0;
-       edev->false_positives = 0;
-
-       if (!of_device_is_available(dn))
-               return NULL;
-
-       /* Ignore bad nodes. */
-       if (!class_code || !vendor_id || !device_id)
-               return NULL;
-
-       /* There is nothing to check on PCI to ISA bridges */
-       if (dn->type && !strcmp(dn->type, "isa")) {
-               edev->mode |= EEH_MODE_NOCHECK;
-               return NULL;
-       }
-       edev->class_code = *class_code;
-
-       /* Ok... see if this device supports EEH.  Some do, some don't,
-        * and the only way to find out is to check each and every one.
-        */
-       regs = of_get_property(dn, "reg", NULL);
-       if (regs) {
-               /* First register entry is addr (00BBSS00)  */
-               /* Try to enable eeh */
-               ret = eeh_ops->set_option(dn, EEH_OPT_ENABLE);
-
-               enable = 0;
-               if (ret == 0) {
-                       edev->config_addr = regs[0];
-
-                       /* If the newer, better, ibm,get-config-addr-info is supported, 
-                        * then use that instead.
-                        */
-                       edev->pe_config_addr = eeh_ops->get_pe_addr(dn);
-
-                       /* Some older systems (Power4) allow the
-                        * ibm,set-eeh-option call to succeed even on nodes
-                        * where EEH is not supported. Verify support
-                        * explicitly.
-                        */
-                       ret = eeh_ops->get_state(dn, NULL);
-                       if (ret > 0 && ret != EEH_STATE_NOT_SUPPORT)
-                               enable = 1;
-               }
-
-               if (enable) {
-                       eeh_subsystem_enabled = 1;
-                       edev->mode |= EEH_MODE_SUPPORTED;
-
-                       pr_debug("EEH: %s: eeh enabled, config=%x pe_config=%x\n",
-                                dn->full_name, edev->config_addr,
-                                edev->pe_config_addr);
-               } else {
-
-                       /* This device doesn't support EEH, but it may have an
-                        * EEH parent, in which case we mark it as supported.
-                        */
-                       if (dn->parent && of_node_to_eeh_dev(dn->parent) &&
-                           (of_node_to_eeh_dev(dn->parent)->mode & EEH_MODE_SUPPORTED)) {
-                               /* Parent supports EEH. */
-                               edev->mode |= EEH_MODE_SUPPORTED;
-                               edev->config_addr = of_node_to_eeh_dev(dn->parent)->config_addr;
-                               return NULL;
-                       }
-               }
-       } else {
-               printk(KERN_WARNING "EEH: %s: unable to get reg property.\n",
-                      dn->full_name);
-       }
-
-       eeh_save_bars(edev);
-       return NULL;
-}
-
 /**
  * eeh_ops_register - Register platform dependent EEH operations
  * @ops: platform dependent EEH operations
@@ -982,7 +674,7 @@ int __exit eeh_ops_unregister(const char *name)
  * Even if force-off is set, the EEH hardware is still enabled, so that
  * newer systems can boot.
  */
-void __init eeh_init(void)
+static int __init eeh_init(void)
 {
        struct pci_controller *hose, *tmp;
        struct device_node *phb;
@@ -992,27 +684,34 @@ void __init eeh_init(void)
        if (!eeh_ops) {
                pr_warning("%s: Platform EEH operation not found\n",
                        __func__);
-               return;
+               return -EEXIST;
        } else if ((ret = eeh_ops->init())) {
                pr_warning("%s: Failed to call platform init function (%d)\n",
                        __func__, ret);
-               return;
+               return ret;
        }
 
        raw_spin_lock_init(&confirm_error_lock);
 
        /* Enable EEH for all adapters */
-       list_for_each_entry_safe(hose, tmp, &hose_list, list_node) {
-               phb = hose->dn;
-               traverse_pci_devices(phb, eeh_early_enable, NULL);
+       if (eeh_probe_mode_devtree()) {
+               list_for_each_entry_safe(hose, tmp,
+                       &hose_list, list_node) {
+                       phb = hose->dn;
+                       traverse_pci_devices(phb, eeh_ops->of_probe, NULL);
+               }
        }
 
        if (eeh_subsystem_enabled)
-               printk(KERN_INFO "EEH: PCI Enhanced I/O Error Handling Enabled\n");
+               pr_info("EEH: PCI Enhanced I/O Error Handling Enabled\n");
        else
-               printk(KERN_WARNING "EEH: No capable adapters found\n");
+               pr_warning("EEH: No capable adapters found\n");
+
+       return ret;
 }
 
+core_initcall_sync(eeh_init);
+
 /**
  * eeh_add_device_early - Enable EEH for the indicated device_node
  * @dn: device node for which to set up EEH
@@ -1037,7 +736,8 @@ static void eeh_add_device_early(struct device_node *dn)
        if (NULL == phb || 0 == phb->buid)
                return;
 
-       eeh_early_enable(dn, NULL);
+       /* FIXME: hotplug support on POWERNV */
+       eeh_ops->of_probe(dn, NULL);
 }
 
 /**
@@ -1087,7 +787,7 @@ static void eeh_add_device_late(struct pci_dev *dev)
        edev->pdev = dev;
        dev->dev.archdata.edev = edev;
 
-       pci_addr_cache_insert_device(dev);
+       eeh_addr_cache_insert_dev(dev);
        eeh_sysfs_add_device(dev);
 }
 
@@ -1143,7 +843,8 @@ static void eeh_remove_device(struct pci_dev *dev)
        dev->dev.archdata.edev = NULL;
        pci_dev_put(dev);
 
-       pci_addr_cache_remove_device(dev);
+       eeh_rmv_from_parent_pe(edev);
+       eeh_addr_cache_rmv_dev(dev);
        eeh_sysfs_remove_device(dev);
 }