netxen: support pci error handlers

o Support pci error detection and recovery.
o Refactor suspend and resume code, to share with io_error_detected,
  and slot_reset callbacks
o NX_NEED_AER device state added, to synchronize with firmware
  recovery.

Signed-off-by: Amit Kumar Salecha <amit.salecha@qlogic.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

diff --git a/drivers/net/netxen/netxen_nic_hdr.h b/drivers/net/netxen/netxen_nic_hdr.h
index d138fc22927afee3281e0d7137d01c2b4eb59309..638369024908831e8761d7d9d4cf63df631b8ff2 100644 (file)
--- a/drivers/net/netxen/netxen_nic_hdr.h
+++ b/drivers/net/netxen/netxen_nic_hdr.h
@@ -969,7 +969,8 @@ enum {
 #define NX_DEV_READY           3
 #define NX_DEV_NEED_RESET      4
 #define NX_DEV_NEED_QUISCENT   5
-#define NX_DEV_FAILED          6
+#define NX_DEV_NEED_AER        6
+#define NX_DEV_FAILED          7
 
 #define NX_RCODE_DRIVER_INFO           0x20000000
 #define NX_RCODE_DRIVER_CAN_RELOAD     0x40000000

diff --git a/drivers/net/netxen/netxen_nic_main.c b/drivers/net/netxen/netxen_nic_main.c
index 1aca6260909aecb6686ab0f5a7288493dff4754d..f13a07f717c70002809ceff0ad5fb678e0e4a5f9 100644 (file)
--- a/drivers/net/netxen/netxen_nic_main.c
+++ b/drivers/net/netxen/netxen_nic_main.c
@@ -35,6 +35,7 @@
 #include <linux/ipv6.h>
 #include <linux/inetdevice.h>
 #include <linux/sysfs.h>
+#include <linux/aer.h>
 
 MODULE_DESCRIPTION("QLogic/NetXen (1/10) GbE Converged Ethernet Driver");
 MODULE_LICENSE("GPL");
@@ -84,6 +85,7 @@ static void netxen_remove_sysfs_entries(struct netxen_adapter *adapter);
 static void netxen_create_diag_entries(struct netxen_adapter *adapter);
 static void netxen_remove_diag_entries(struct netxen_adapter *adapter);
 
+static int nx_dev_request_aer(struct netxen_adapter *adapter);
 static int nx_decr_dev_ref_cnt(struct netxen_adapter *adapter);
 static int netxen_can_start_firmware(struct netxen_adapter *adapter);
 
@@ -1262,6 +1264,9 @@ netxen_nic_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
        if ((err = pci_request_regions(pdev, netxen_nic_driver_name)))
                goto err_out_disable_pdev;
 
+       if (NX_IS_REVISION_P3(pdev->revision))
+               pci_enable_pcie_error_reporting(pdev);
+
        pci_set_master(pdev);
 
        netdev = alloc_etherdev(sizeof(struct netxen_adapter));
@@ -1409,17 +1414,19 @@ static void __devexit netxen_nic_remove(struct pci_dev *pdev)
 
        netxen_release_firmware(adapter);
 
+       if (NX_IS_REVISION_P3(pdev->revision))
+               pci_disable_pcie_error_reporting(pdev);
+
        pci_release_regions(pdev);
        pci_disable_device(pdev);
        pci_set_drvdata(pdev, NULL);
 
        free_netdev(netdev);
 }
-static int __netxen_nic_shutdown(struct pci_dev *pdev)
+
+static void netxen_nic_detach_func(struct netxen_adapter *adapter)
 {
-       struct netxen_adapter *adapter = pci_get_drvdata(pdev);
        struct net_device *netdev = adapter->netdev;
-       int retval;
 
        netif_device_detach(netdev);
 
@@ -1438,43 +1445,9 @@ static int __netxen_nic_shutdown(struct pci_dev *pdev)
        nx_decr_dev_ref_cnt(adapter);
 
        clear_bit(__NX_RESETTING, &adapter->state);
-
-       retval = pci_save_state(pdev);
-       if (retval)
-               return retval;
-
-       if (netxen_nic_wol_supported(adapter)) {
-               pci_enable_wake(pdev, PCI_D3cold, 1);
-               pci_enable_wake(pdev, PCI_D3hot, 1);
-       }
-
-       return 0;
-}
-static void netxen_nic_shutdown(struct pci_dev *pdev)
-{
-       if (__netxen_nic_shutdown(pdev))
-               return;
-
-       pci_disable_device(pdev);
-}
-#ifdef CONFIG_PM
-static int
-netxen_nic_suspend(struct pci_dev *pdev, pm_message_t state)
-{
-       int retval;
-
-       retval = __netxen_nic_shutdown(pdev);
-       if (retval)
-               return retval;
-
-       pci_set_power_state(pdev, pci_choose_state(pdev, state));
-
-       pci_disable_device(pdev);
-       return 0;
 }
 
-static int
-netxen_nic_resume(struct pci_dev *pdev)
+static int netxen_nic_attach_func(struct pci_dev *pdev)
 {
        struct netxen_adapter *adapter = pci_get_drvdata(pdev);
        struct net_device *netdev = adapter->netdev;
@@ -1519,6 +1492,85 @@ err_out:
        nx_decr_dev_ref_cnt(adapter);
        return err;
 }
+
+static pci_ers_result_t netxen_io_error_detected(struct pci_dev *pdev,
+                                               pci_channel_state_t state)
+{
+       struct netxen_adapter *adapter = pci_get_drvdata(pdev);
+
+       if (state == pci_channel_io_perm_failure)
+               return PCI_ERS_RESULT_DISCONNECT;
+
+       if (nx_dev_request_aer(adapter))
+               return PCI_ERS_RESULT_RECOVERED;
+
+       netxen_nic_detach_func(adapter);
+
+       pci_disable_device(pdev);
+
+       return PCI_ERS_RESULT_NEED_RESET;
+}
+
+static pci_ers_result_t netxen_io_slot_reset(struct pci_dev *pdev)
+{
+       int err = 0;
+
+       err = netxen_nic_attach_func(pdev);
+
+       return err ? PCI_ERS_RESULT_DISCONNECT : PCI_ERS_RESULT_RECOVERED;
+}
+
+static void netxen_io_resume(struct pci_dev *pdev)
+{
+       pci_cleanup_aer_uncorrect_error_status(pdev);
+}
+
+static void netxen_nic_shutdown(struct pci_dev *pdev)
+{
+       struct netxen_adapter *adapter = pci_get_drvdata(pdev);
+
+       netxen_nic_detach_func(adapter);
+
+       if (pci_save_state(pdev))
+               return;
+
+       if (netxen_nic_wol_supported(adapter)) {
+               pci_enable_wake(pdev, PCI_D3cold, 1);
+               pci_enable_wake(pdev, PCI_D3hot, 1);
+       }
+
+       pci_disable_device(pdev);
+}
+
+#ifdef CONFIG_PM
+static int
+netxen_nic_suspend(struct pci_dev *pdev, pm_message_t state)
+{
+       struct netxen_adapter *adapter = pci_get_drvdata(pdev);
+       int retval;
+
+       netxen_nic_detach_func(adapter);
+
+       retval = pci_save_state(pdev);
+       if (retval)
+               return retval;
+
+       if (netxen_nic_wol_supported(adapter)) {
+               pci_enable_wake(pdev, PCI_D3cold, 1);
+               pci_enable_wake(pdev, PCI_D3hot, 1);
+       }
+
+       pci_disable_device(pdev);
+       pci_set_power_state(pdev, pci_choose_state(pdev, state));
+
+       return 0;
+}
+
+static int
+netxen_nic_resume(struct pci_dev *pdev)
+{
+       return netxen_nic_attach_func(pdev);
+}
 #endif
 
 static int netxen_nic_open(struct net_device *netdev)
@@ -2110,20 +2162,49 @@ nx_decr_dev_ref_cnt(struct netxen_adapter *adapter)
        return count;
 }
 
-static void
+static int
+nx_dev_request_aer(struct netxen_adapter *adapter)
+{
+       u32 state;
+       int ret = -EINVAL;
+
+       if (netxen_api_lock(adapter))
+               return ret;
+
+       state = NXRD32(adapter, NX_CRB_DEV_STATE);
+
+       if (state == NX_DEV_NEED_AER)
+               ret = 0;
+       else if (state == NX_DEV_READY) {
+               NXWR32(adapter, NX_CRB_DEV_STATE, NX_DEV_NEED_AER);
+               ret = 0;
+       }
+
+       netxen_api_unlock(adapter);
+       return ret;
+}
+
+static int
 nx_dev_request_reset(struct netxen_adapter *adapter)
 {
        u32 state;
+       int ret = -EINVAL;
 
        if (netxen_api_lock(adapter))
-               return;
+               return ret;
 
        state = NXRD32(adapter, NX_CRB_DEV_STATE);
 
-       if (state != NX_DEV_INITALIZING)
+       if (state == NX_DEV_NEED_RESET)
+               ret = 0;
+       else if (state != NX_DEV_INITALIZING && state != NX_DEV_NEED_AER) {
                NXWR32(adapter, NX_CRB_DEV_STATE, NX_DEV_NEED_RESET);
+               ret = 0;
+       }
 
        netxen_api_unlock(adapter);
+
+       return ret;
 }
 
 static int
@@ -2275,18 +2356,28 @@ netxen_check_health(struct netxen_adapter *adapter)
        u32 state, heartbit;
        struct net_device *netdev = adapter->netdev;
 
+       state = NXRD32(adapter, NX_CRB_DEV_STATE);
+       if (state == NX_DEV_NEED_AER)
+               return 0;
+
        if (netxen_nic_check_temp(adapter))
                goto detach;
 
        if (adapter->need_fw_reset) {
-               nx_dev_request_reset(adapter);
+               if (nx_dev_request_reset(adapter))
+                       return 0;
                goto detach;
        }
 
-       state = NXRD32(adapter, NX_CRB_DEV_STATE);
+       /* NX_DEV_NEED_RESET, this state can be marked in two cases
+        * 1. Tx timeout 2. Fw hang
+        * Send request to destroy context in case of tx timeout only
+        * and doesn't required in case of Fw hang
+        */
        if (state == NX_DEV_NEED_RESET) {
                adapter->need_fw_reset = 1;
-               goto detach;
+               if (NX_IS_REVISION_P2(adapter->ahw.revision_id))
+                       goto detach;
        }
 
        if (NX_IS_REVISION_P2(adapter->ahw.revision_id))
@@ -2296,12 +2387,17 @@ netxen_check_health(struct netxen_adapter *adapter)
        if (heartbit != adapter->heartbit) {
                adapter->heartbit = heartbit;
                adapter->fw_fail_cnt = 0;
+               if (adapter->need_fw_reset)
+                       goto detach;
                return 0;
        }
 
        if (++adapter->fw_fail_cnt < FW_FAIL_THRESH)
                return 0;
 
+       if (nx_dev_request_reset(adapter))
+               return 0;
+
        clear_bit(__NX_FW_ATTACHED, &adapter->state);
 
        dev_info(&netdev->dev, "firmware hang detected\n");
@@ -2731,6 +2827,12 @@ netxen_config_indev_addr(struct net_device *dev, unsigned long event)
 { }
 #endif
 
+static struct pci_error_handlers netxen_err_handler = {
+       .error_detected = netxen_io_error_detected,
+       .slot_reset = netxen_io_slot_reset,
+       .resume = netxen_io_resume,
+};
+
 static struct pci_driver netxen_driver = {
        .name = netxen_nic_driver_name,
        .id_table = netxen_pci_tbl,
@@ -2740,7 +2842,8 @@ static struct pci_driver netxen_driver = {
        .suspend = netxen_nic_suspend,
        .resume = netxen_nic_resume,
 #endif
-       .shutdown = netxen_nic_shutdown
+       .shutdown = netxen_nic_shutdown,
+       .err_handler = &netxen_err_handler
 };
 
 static int __init netxen_init_module(void)