Add the rt linux 4.1.3-rt3 as base
[kvmfornfv.git] / kernel / drivers / pci / pcie / aer / aerdrv_core.c
diff --git a/kernel/drivers/pci/pcie/aer/aerdrv_core.c b/kernel/drivers/pci/pcie/aer/aerdrv_core.c
new file mode 100644 (file)
index 0000000..5653ea9
--- /dev/null
@@ -0,0 +1,805 @@
+/*
+ * drivers/pci/pcie/aer/aerdrv_core.c
+ *
+ * This file is subject to the terms and conditions of the GNU General Public
+ * License.  See the file "COPYING" in the main directory of this archive
+ * for more details.
+ *
+ * This file implements the core part of PCI-Express AER. When an pci-express
+ * error is delivered, an error message will be collected and printed to
+ * console, then, an error recovery procedure will be executed by following
+ * the pci error recovery rules.
+ *
+ * Copyright (C) 2006 Intel Corp.
+ *     Tom Long Nguyen (tom.l.nguyen@intel.com)
+ *     Zhang Yanmin (yanmin.zhang@intel.com)
+ *
+ */
+
+#include <linux/module.h>
+#include <linux/pci.h>
+#include <linux/kernel.h>
+#include <linux/errno.h>
+#include <linux/pm.h>
+#include <linux/suspend.h>
+#include <linux/delay.h>
+#include <linux/slab.h>
+#include <linux/kfifo.h>
+#include "aerdrv.h"
+
+static bool forceload;
+static bool nosourceid;
+module_param(forceload, bool, 0);
+module_param(nosourceid, bool, 0);
+
+#define        PCI_EXP_AER_FLAGS       (PCI_EXP_DEVCTL_CERE | PCI_EXP_DEVCTL_NFERE | \
+                                PCI_EXP_DEVCTL_FERE | PCI_EXP_DEVCTL_URRE)
+
+int pci_enable_pcie_error_reporting(struct pci_dev *dev)
+{
+       if (pcie_aer_get_firmware_first(dev))
+               return -EIO;
+
+       if (!pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ERR))
+               return -EIO;
+
+       return pcie_capability_set_word(dev, PCI_EXP_DEVCTL, PCI_EXP_AER_FLAGS);
+}
+EXPORT_SYMBOL_GPL(pci_enable_pcie_error_reporting);
+
+int pci_disable_pcie_error_reporting(struct pci_dev *dev)
+{
+       if (pcie_aer_get_firmware_first(dev))
+               return -EIO;
+
+       return pcie_capability_clear_word(dev, PCI_EXP_DEVCTL,
+                                         PCI_EXP_AER_FLAGS);
+}
+EXPORT_SYMBOL_GPL(pci_disable_pcie_error_reporting);
+
+int pci_cleanup_aer_uncorrect_error_status(struct pci_dev *dev)
+{
+       int pos;
+       u32 status;
+
+       pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ERR);
+       if (!pos)
+               return -EIO;
+
+       pci_read_config_dword(dev, pos + PCI_ERR_UNCOR_STATUS, &status);
+       if (status)
+               pci_write_config_dword(dev, pos + PCI_ERR_UNCOR_STATUS, status);
+
+       return 0;
+}
+EXPORT_SYMBOL_GPL(pci_cleanup_aer_uncorrect_error_status);
+
+/**
+ * add_error_device - list device to be handled
+ * @e_info: pointer to error info
+ * @dev: pointer to pci_dev to be added
+ */
+static int add_error_device(struct aer_err_info *e_info, struct pci_dev *dev)
+{
+       if (e_info->error_dev_num < AER_MAX_MULTI_ERR_DEVICES) {
+               e_info->dev[e_info->error_dev_num] = dev;
+               e_info->error_dev_num++;
+               return 0;
+       }
+       return -ENOSPC;
+}
+
+/**
+ * is_error_source - check whether the device is source of reported error
+ * @dev: pointer to pci_dev to be checked
+ * @e_info: pointer to reported error info
+ */
+static bool is_error_source(struct pci_dev *dev, struct aer_err_info *e_info)
+{
+       int pos;
+       u32 status, mask;
+       u16 reg16;
+
+       /*
+        * When bus id is equal to 0, it might be a bad id
+        * reported by root port.
+        */
+       if (!nosourceid && (PCI_BUS_NUM(e_info->id) != 0)) {
+               /* Device ID match? */
+               if (e_info->id == ((dev->bus->number << 8) | dev->devfn))
+                       return true;
+
+               /* Continue id comparing if there is no multiple error */
+               if (!e_info->multi_error_valid)
+                       return false;
+       }
+
+       /*
+        * When either
+        *      1) nosourceid==y;
+        *      2) bus id is equal to 0. Some ports might lose the bus
+        *              id of error source id;
+        *      3) There are multiple errors and prior id comparing fails;
+        * We check AER status registers to find possible reporter.
+        */
+       if (atomic_read(&dev->enable_cnt) == 0)
+               return false;
+
+       /* Check if AER is enabled */
+       pcie_capability_read_word(dev, PCI_EXP_DEVCTL, &reg16);
+       if (!(reg16 & PCI_EXP_AER_FLAGS))
+               return false;
+
+       pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ERR);
+       if (!pos)
+               return false;
+
+       /* Check if error is recorded */
+       if (e_info->severity == AER_CORRECTABLE) {
+               pci_read_config_dword(dev, pos + PCI_ERR_COR_STATUS, &status);
+               pci_read_config_dword(dev, pos + PCI_ERR_COR_MASK, &mask);
+       } else {
+               pci_read_config_dword(dev, pos + PCI_ERR_UNCOR_STATUS, &status);
+               pci_read_config_dword(dev, pos + PCI_ERR_UNCOR_MASK, &mask);
+       }
+       if (status & ~mask)
+               return true;
+
+       return false;
+}
+
+static int find_device_iter(struct pci_dev *dev, void *data)
+{
+       struct aer_err_info *e_info = (struct aer_err_info *)data;
+
+       if (is_error_source(dev, e_info)) {
+               /* List this device */
+               if (add_error_device(e_info, dev)) {
+                       /* We cannot handle more... Stop iteration */
+                       /* TODO: Should print error message here? */
+                       return 1;
+               }
+
+               /* If there is only a single error, stop iteration */
+               if (!e_info->multi_error_valid)
+                       return 1;
+       }
+       return 0;
+}
+
+/**
+ * find_source_device - search through device hierarchy for source device
+ * @parent: pointer to Root Port pci_dev data structure
+ * @e_info: including detailed error information such like id
+ *
+ * Return true if found.
+ *
+ * Invoked by DPC when error is detected at the Root Port.
+ * Caller of this function must set id, severity, and multi_error_valid of
+ * struct aer_err_info pointed by @e_info properly.  This function must fill
+ * e_info->error_dev_num and e_info->dev[], based on the given information.
+ */
+static bool find_source_device(struct pci_dev *parent,
+               struct aer_err_info *e_info)
+{
+       struct pci_dev *dev = parent;
+       int result;
+
+       /* Must reset in this function */
+       e_info->error_dev_num = 0;
+
+       /* Is Root Port an agent that sends error message? */
+       result = find_device_iter(dev, e_info);
+       if (result)
+               return true;
+
+       pci_walk_bus(parent->subordinate, find_device_iter, e_info);
+
+       if (!e_info->error_dev_num) {
+               dev_printk(KERN_DEBUG, &parent->dev,
+                               "can't find device of ID%04x\n",
+                               e_info->id);
+               return false;
+       }
+       return true;
+}
+
+static int report_error_detected(struct pci_dev *dev, void *data)
+{
+       pci_ers_result_t vote;
+       const struct pci_error_handlers *err_handler;
+       struct aer_broadcast_data *result_data;
+       result_data = (struct aer_broadcast_data *) data;
+
+       device_lock(&dev->dev);
+       dev->error_state = result_data->state;
+
+       if (!dev->driver ||
+               !dev->driver->err_handler ||
+               !dev->driver->err_handler->error_detected) {
+               if (result_data->state == pci_channel_io_frozen &&
+                       !(dev->hdr_type & PCI_HEADER_TYPE_BRIDGE)) {
+                       /*
+                        * In case of fatal recovery, if one of down-
+                        * stream device has no driver. We might be
+                        * unable to recover because a later insmod
+                        * of a driver for this device is unaware of
+                        * its hw state.
+                        */
+                       dev_printk(KERN_DEBUG, &dev->dev, "device has %s\n",
+                                  dev->driver ?
+                                  "no AER-aware driver" : "no driver");
+               }
+
+               /*
+                * If there's any device in the subtree that does not
+                * have an error_detected callback, returning
+                * PCI_ERS_RESULT_NO_AER_DRIVER prevents calling of
+                * the subsequent mmio_enabled/slot_reset/resume
+                * callbacks of "any" device in the subtree. All the
+                * devices in the subtree are left in the error state
+                * without recovery.
+                */
+
+               if (!(dev->hdr_type & PCI_HEADER_TYPE_BRIDGE))
+                       vote = PCI_ERS_RESULT_NO_AER_DRIVER;
+               else
+                       vote = PCI_ERS_RESULT_NONE;
+       } else {
+               err_handler = dev->driver->err_handler;
+               vote = err_handler->error_detected(dev, result_data->state);
+       }
+
+       result_data->result = merge_result(result_data->result, vote);
+       device_unlock(&dev->dev);
+       return 0;
+}
+
+static int report_mmio_enabled(struct pci_dev *dev, void *data)
+{
+       pci_ers_result_t vote;
+       const struct pci_error_handlers *err_handler;
+       struct aer_broadcast_data *result_data;
+       result_data = (struct aer_broadcast_data *) data;
+
+       device_lock(&dev->dev);
+       if (!dev->driver ||
+               !dev->driver->err_handler ||
+               !dev->driver->err_handler->mmio_enabled)
+               goto out;
+
+       err_handler = dev->driver->err_handler;
+       vote = err_handler->mmio_enabled(dev);
+       result_data->result = merge_result(result_data->result, vote);
+out:
+       device_unlock(&dev->dev);
+       return 0;
+}
+
+static int report_slot_reset(struct pci_dev *dev, void *data)
+{
+       pci_ers_result_t vote;
+       const struct pci_error_handlers *err_handler;
+       struct aer_broadcast_data *result_data;
+       result_data = (struct aer_broadcast_data *) data;
+
+       device_lock(&dev->dev);
+       if (!dev->driver ||
+               !dev->driver->err_handler ||
+               !dev->driver->err_handler->slot_reset)
+               goto out;
+
+       err_handler = dev->driver->err_handler;
+       vote = err_handler->slot_reset(dev);
+       result_data->result = merge_result(result_data->result, vote);
+out:
+       device_unlock(&dev->dev);
+       return 0;
+}
+
+static int report_resume(struct pci_dev *dev, void *data)
+{
+       const struct pci_error_handlers *err_handler;
+
+       device_lock(&dev->dev);
+       dev->error_state = pci_channel_io_normal;
+
+       if (!dev->driver ||
+               !dev->driver->err_handler ||
+               !dev->driver->err_handler->resume)
+               goto out;
+
+       err_handler = dev->driver->err_handler;
+       err_handler->resume(dev);
+out:
+       device_unlock(&dev->dev);
+       return 0;
+}
+
+/**
+ * broadcast_error_message - handle message broadcast to downstream drivers
+ * @dev: pointer to from where in a hierarchy message is broadcasted down
+ * @state: error state
+ * @error_mesg: message to print
+ * @cb: callback to be broadcasted
+ *
+ * Invoked during error recovery process. Once being invoked, the content
+ * of error severity will be broadcasted to all downstream drivers in a
+ * hierarchy in question.
+ */
+static pci_ers_result_t broadcast_error_message(struct pci_dev *dev,
+       enum pci_channel_state state,
+       char *error_mesg,
+       int (*cb)(struct pci_dev *, void *))
+{
+       struct aer_broadcast_data result_data;
+
+       dev_printk(KERN_DEBUG, &dev->dev, "broadcast %s message\n", error_mesg);
+       result_data.state = state;
+       if (cb == report_error_detected)
+               result_data.result = PCI_ERS_RESULT_CAN_RECOVER;
+       else
+               result_data.result = PCI_ERS_RESULT_RECOVERED;
+
+       if (dev->hdr_type & PCI_HEADER_TYPE_BRIDGE) {
+               /*
+                * If the error is reported by a bridge, we think this error
+                * is related to the downstream link of the bridge, so we
+                * do error recovery on all subordinates of the bridge instead
+                * of the bridge and clear the error status of the bridge.
+                */
+               if (cb == report_error_detected)
+                       dev->error_state = state;
+               pci_walk_bus(dev->subordinate, cb, &result_data);
+               if (cb == report_resume) {
+                       pci_cleanup_aer_uncorrect_error_status(dev);
+                       dev->error_state = pci_channel_io_normal;
+               }
+       } else {
+               /*
+                * If the error is reported by an end point, we think this
+                * error is related to the upstream link of the end point.
+                */
+               pci_walk_bus(dev->bus, cb, &result_data);
+       }
+
+       return result_data.result;
+}
+
+/**
+ * default_reset_link - default reset function
+ * @dev: pointer to pci_dev data structure
+ *
+ * Invoked when performing link reset on a Downstream Port or a
+ * Root Port with no aer driver.
+ */
+static pci_ers_result_t default_reset_link(struct pci_dev *dev)
+{
+       pci_reset_bridge_secondary_bus(dev);
+       dev_printk(KERN_DEBUG, &dev->dev, "downstream link has been reset\n");
+       return PCI_ERS_RESULT_RECOVERED;
+}
+
+static int find_aer_service_iter(struct device *device, void *data)
+{
+       struct pcie_port_service_driver *service_driver, **drv;
+
+       drv = (struct pcie_port_service_driver **) data;
+
+       if (device->bus == &pcie_port_bus_type && device->driver) {
+               service_driver = to_service_driver(device->driver);
+               if (service_driver->service == PCIE_PORT_SERVICE_AER) {
+                       *drv = service_driver;
+                       return 1;
+               }
+       }
+
+       return 0;
+}
+
+static struct pcie_port_service_driver *find_aer_service(struct pci_dev *dev)
+{
+       struct pcie_port_service_driver *drv = NULL;
+
+       device_for_each_child(&dev->dev, &drv, find_aer_service_iter);
+
+       return drv;
+}
+
+static pci_ers_result_t reset_link(struct pci_dev *dev)
+{
+       struct pci_dev *udev;
+       pci_ers_result_t status;
+       struct pcie_port_service_driver *driver;
+
+       if (dev->hdr_type & PCI_HEADER_TYPE_BRIDGE) {
+               /* Reset this port for all subordinates */
+               udev = dev;
+       } else {
+               /* Reset the upstream component (likely downstream port) */
+               udev = dev->bus->self;
+       }
+
+       /* Use the aer driver of the component firstly */
+       driver = find_aer_service(udev);
+
+       if (driver && driver->reset_link) {
+               status = driver->reset_link(udev);
+       } else if (pci_pcie_type(udev) == PCI_EXP_TYPE_DOWNSTREAM ||
+               pci_pcie_type(udev) == PCI_EXP_TYPE_ROOT_PORT) {
+               status = default_reset_link(udev);
+       } else {
+               dev_printk(KERN_DEBUG, &dev->dev,
+                       "no link-reset support at upstream device %s\n",
+                       pci_name(udev));
+               return PCI_ERS_RESULT_DISCONNECT;
+       }
+
+       if (status != PCI_ERS_RESULT_RECOVERED) {
+               dev_printk(KERN_DEBUG, &dev->dev,
+                       "link reset at upstream device %s failed\n",
+                       pci_name(udev));
+               return PCI_ERS_RESULT_DISCONNECT;
+       }
+
+       return status;
+}
+
+/**
+ * do_recovery - handle nonfatal/fatal error recovery process
+ * @dev: pointer to a pci_dev data structure of agent detecting an error
+ * @severity: error severity type
+ *
+ * Invoked when an error is nonfatal/fatal. Once being invoked, broadcast
+ * error detected message to all downstream drivers within a hierarchy in
+ * question and return the returned code.
+ */
+static void do_recovery(struct pci_dev *dev, int severity)
+{
+       pci_ers_result_t status, result = PCI_ERS_RESULT_RECOVERED;
+       enum pci_channel_state state;
+
+       if (severity == AER_FATAL)
+               state = pci_channel_io_frozen;
+       else
+               state = pci_channel_io_normal;
+
+       status = broadcast_error_message(dev,
+                       state,
+                       "error_detected",
+                       report_error_detected);
+
+       if (severity == AER_FATAL) {
+               result = reset_link(dev);
+               if (result != PCI_ERS_RESULT_RECOVERED)
+                       goto failed;
+       }
+
+       if (status == PCI_ERS_RESULT_CAN_RECOVER)
+               status = broadcast_error_message(dev,
+                               state,
+                               "mmio_enabled",
+                               report_mmio_enabled);
+
+       if (status == PCI_ERS_RESULT_NEED_RESET) {
+               /*
+                * TODO: Should call platform-specific
+                * functions to reset slot before calling
+                * drivers' slot_reset callbacks?
+                */
+               status = broadcast_error_message(dev,
+                               state,
+                               "slot_reset",
+                               report_slot_reset);
+       }
+
+       if (status != PCI_ERS_RESULT_RECOVERED)
+               goto failed;
+
+       broadcast_error_message(dev,
+                               state,
+                               "resume",
+                               report_resume);
+
+       dev_info(&dev->dev, "AER: Device recovery successful\n");
+       return;
+
+failed:
+       /* TODO: Should kernel panic here? */
+       dev_info(&dev->dev, "AER: Device recovery failed\n");
+}
+
+/**
+ * handle_error_source - handle logging error into an event log
+ * @aerdev: pointer to pcie_device data structure of the root port
+ * @dev: pointer to pci_dev data structure of error source device
+ * @info: comprehensive error information
+ *
+ * Invoked when an error being detected by Root Port.
+ */
+static void handle_error_source(struct pcie_device *aerdev,
+       struct pci_dev *dev,
+       struct aer_err_info *info)
+{
+       int pos;
+
+       if (info->severity == AER_CORRECTABLE) {
+               /*
+                * Correctable error does not need software intervention.
+                * No need to go through error recovery process.
+                */
+               pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ERR);
+               if (pos)
+                       pci_write_config_dword(dev, pos + PCI_ERR_COR_STATUS,
+                                       info->status);
+       } else
+               do_recovery(dev, info->severity);
+}
+
+#ifdef CONFIG_ACPI_APEI_PCIEAER
+static void aer_recover_work_func(struct work_struct *work);
+
+#define AER_RECOVER_RING_ORDER         4
+#define AER_RECOVER_RING_SIZE          (1 << AER_RECOVER_RING_ORDER)
+
+struct aer_recover_entry {
+       u8      bus;
+       u8      devfn;
+       u16     domain;
+       int     severity;
+       struct aer_capability_regs *regs;
+};
+
+static DEFINE_KFIFO(aer_recover_ring, struct aer_recover_entry,
+                   AER_RECOVER_RING_SIZE);
+/*
+ * Mutual exclusion for writers of aer_recover_ring, reader side don't
+ * need lock, because there is only one reader and lock is not needed
+ * between reader and writer.
+ */
+static DEFINE_SPINLOCK(aer_recover_ring_lock);
+static DECLARE_WORK(aer_recover_work, aer_recover_work_func);
+
+void aer_recover_queue(int domain, unsigned int bus, unsigned int devfn,
+                      int severity, struct aer_capability_regs *aer_regs)
+{
+       unsigned long flags;
+       struct aer_recover_entry entry = {
+               .bus            = bus,
+               .devfn          = devfn,
+               .domain         = domain,
+               .severity       = severity,
+               .regs           = aer_regs,
+       };
+
+       spin_lock_irqsave(&aer_recover_ring_lock, flags);
+       if (kfifo_put(&aer_recover_ring, entry))
+               schedule_work(&aer_recover_work);
+       else
+               pr_err("AER recover: Buffer overflow when recovering AER for %04x:%02x:%02x:%x\n",
+                      domain, bus, PCI_SLOT(devfn), PCI_FUNC(devfn));
+       spin_unlock_irqrestore(&aer_recover_ring_lock, flags);
+}
+EXPORT_SYMBOL_GPL(aer_recover_queue);
+
+static void aer_recover_work_func(struct work_struct *work)
+{
+       struct aer_recover_entry entry;
+       struct pci_dev *pdev;
+
+       while (kfifo_get(&aer_recover_ring, &entry)) {
+               pdev = pci_get_domain_bus_and_slot(entry.domain, entry.bus,
+                                                  entry.devfn);
+               if (!pdev) {
+                       pr_err("AER recover: Can not find pci_dev for %04x:%02x:%02x:%x\n",
+                              entry.domain, entry.bus,
+                              PCI_SLOT(entry.devfn), PCI_FUNC(entry.devfn));
+                       continue;
+               }
+               cper_print_aer(pdev, entry.severity, entry.regs);
+               do_recovery(pdev, entry.severity);
+               pci_dev_put(pdev);
+       }
+}
+#endif
+
+/**
+ * get_device_error_info - read error status from dev and store it to info
+ * @dev: pointer to the device expected to have a error record
+ * @info: pointer to structure to store the error record
+ *
+ * Return 1 on success, 0 on error.
+ *
+ * Note that @info is reused among all error devices. Clear fields properly.
+ */
+static int get_device_error_info(struct pci_dev *dev, struct aer_err_info *info)
+{
+       int pos, temp;
+
+       /* Must reset in this function */
+       info->status = 0;
+       info->tlp_header_valid = 0;
+
+       pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ERR);
+
+       /* The device might not support AER */
+       if (!pos)
+               return 1;
+
+       if (info->severity == AER_CORRECTABLE) {
+               pci_read_config_dword(dev, pos + PCI_ERR_COR_STATUS,
+                       &info->status);
+               pci_read_config_dword(dev, pos + PCI_ERR_COR_MASK,
+                       &info->mask);
+               if (!(info->status & ~info->mask))
+                       return 0;
+       } else if (dev->hdr_type & PCI_HEADER_TYPE_BRIDGE ||
+               info->severity == AER_NONFATAL) {
+
+               /* Link is still healthy for IO reads */
+               pci_read_config_dword(dev, pos + PCI_ERR_UNCOR_STATUS,
+                       &info->status);
+               pci_read_config_dword(dev, pos + PCI_ERR_UNCOR_MASK,
+                       &info->mask);
+               if (!(info->status & ~info->mask))
+                       return 0;
+
+               /* Get First Error Pointer */
+               pci_read_config_dword(dev, pos + PCI_ERR_CAP, &temp);
+               info->first_error = PCI_ERR_CAP_FEP(temp);
+
+               if (info->status & AER_LOG_TLP_MASKS) {
+                       info->tlp_header_valid = 1;
+                       pci_read_config_dword(dev,
+                               pos + PCI_ERR_HEADER_LOG, &info->tlp.dw0);
+                       pci_read_config_dword(dev,
+                               pos + PCI_ERR_HEADER_LOG + 4, &info->tlp.dw1);
+                       pci_read_config_dword(dev,
+                               pos + PCI_ERR_HEADER_LOG + 8, &info->tlp.dw2);
+                       pci_read_config_dword(dev,
+                               pos + PCI_ERR_HEADER_LOG + 12, &info->tlp.dw3);
+               }
+       }
+
+       return 1;
+}
+
+static inline void aer_process_err_devices(struct pcie_device *p_device,
+                       struct aer_err_info *e_info)
+{
+       int i;
+
+       /* Report all before handle them, not to lost records by reset etc. */
+       for (i = 0; i < e_info->error_dev_num && e_info->dev[i]; i++) {
+               if (get_device_error_info(e_info->dev[i], e_info))
+                       aer_print_error(e_info->dev[i], e_info);
+       }
+       for (i = 0; i < e_info->error_dev_num && e_info->dev[i]; i++) {
+               if (get_device_error_info(e_info->dev[i], e_info))
+                       handle_error_source(p_device, e_info->dev[i], e_info);
+       }
+}
+
+/**
+ * aer_isr_one_error - consume an error detected by root port
+ * @p_device: pointer to error root port service device
+ * @e_src: pointer to an error source
+ */
+static void aer_isr_one_error(struct pcie_device *p_device,
+               struct aer_err_source *e_src)
+{
+       struct aer_err_info *e_info;
+
+       /* struct aer_err_info might be big, so we allocate it with slab */
+       e_info = kmalloc(sizeof(struct aer_err_info), GFP_KERNEL);
+       if (!e_info) {
+               dev_printk(KERN_DEBUG, &p_device->port->dev,
+                       "Can't allocate mem when processing AER errors\n");
+               return;
+       }
+
+       /*
+        * There is a possibility that both correctable error and
+        * uncorrectable error being logged. Report correctable error first.
+        */
+       if (e_src->status & PCI_ERR_ROOT_COR_RCV) {
+               e_info->id = ERR_COR_ID(e_src->id);
+               e_info->severity = AER_CORRECTABLE;
+
+               if (e_src->status & PCI_ERR_ROOT_MULTI_COR_RCV)
+                       e_info->multi_error_valid = 1;
+               else
+                       e_info->multi_error_valid = 0;
+
+               aer_print_port_info(p_device->port, e_info);
+
+               if (find_source_device(p_device->port, e_info))
+                       aer_process_err_devices(p_device, e_info);
+       }
+
+       if (e_src->status & PCI_ERR_ROOT_UNCOR_RCV) {
+               e_info->id = ERR_UNCOR_ID(e_src->id);
+
+               if (e_src->status & PCI_ERR_ROOT_FATAL_RCV)
+                       e_info->severity = AER_FATAL;
+               else
+                       e_info->severity = AER_NONFATAL;
+
+               if (e_src->status & PCI_ERR_ROOT_MULTI_UNCOR_RCV)
+                       e_info->multi_error_valid = 1;
+               else
+                       e_info->multi_error_valid = 0;
+
+               aer_print_port_info(p_device->port, e_info);
+
+               if (find_source_device(p_device->port, e_info))
+                       aer_process_err_devices(p_device, e_info);
+       }
+
+       kfree(e_info);
+}
+
+/**
+ * get_e_source - retrieve an error source
+ * @rpc: pointer to the root port which holds an error
+ * @e_src: pointer to store retrieved error source
+ *
+ * Return 1 if an error source is retrieved, otherwise 0.
+ *
+ * Invoked by DPC handler to consume an error.
+ */
+static int get_e_source(struct aer_rpc *rpc, struct aer_err_source *e_src)
+{
+       unsigned long flags;
+
+       /* Lock access to Root error producer/consumer index */
+       spin_lock_irqsave(&rpc->e_lock, flags);
+       if (rpc->prod_idx == rpc->cons_idx) {
+               spin_unlock_irqrestore(&rpc->e_lock, flags);
+               return 0;
+       }
+
+       *e_src = rpc->e_sources[rpc->cons_idx];
+       rpc->cons_idx++;
+       if (rpc->cons_idx == AER_ERROR_SOURCES_MAX)
+               rpc->cons_idx = 0;
+       spin_unlock_irqrestore(&rpc->e_lock, flags);
+
+       return 1;
+}
+
+/**
+ * aer_isr - consume errors detected by root port
+ * @work: definition of this work item
+ *
+ * Invoked, as DPC, when root port records new detected error
+ */
+void aer_isr(struct work_struct *work)
+{
+       struct aer_rpc *rpc = container_of(work, struct aer_rpc, dpc_handler);
+       struct pcie_device *p_device = rpc->rpd;
+       struct aer_err_source uninitialized_var(e_src);
+
+       mutex_lock(&rpc->rpc_mutex);
+       while (get_e_source(rpc, &e_src))
+               aer_isr_one_error(p_device, &e_src);
+       mutex_unlock(&rpc->rpc_mutex);
+
+       wake_up(&rpc->wait_release);
+}
+
+/**
+ * aer_init - provide AER initialization
+ * @dev: pointer to AER pcie device
+ *
+ * Invoked when AER service driver is loaded.
+ */
+int aer_init(struct pcie_device *dev)
+{
+       if (forceload) {
+               dev_printk(KERN_DEBUG, &dev->device,
+                          "aerdrv forceload requested.\n");
+               pcie_aer_force_firmware_first(dev->port, 0);
+       }
+       return 0;
+}