]> err.no Git - linux-2.6/commitdiff
[PATCH] forcedeth: add recoverable error support
authorAyaz Abdulla <aabdulla@nvidia.com>
Mon, 30 Oct 2006 22:32:01 +0000 (17:32 -0500)
committerJeff Garzik <jeff@garzik.org>
Sat, 2 Dec 2006 05:12:01 +0000 (00:12 -0500)
This patch adds support to recover from a previously fatal MAC error. In
the past the MAC would be hung on an internal fatal error. On new
chipsets, the MAC has the ability to enter a non-fatal state and allow
the driver to re-init it.

Signed-Off-By: Ayaz Abdulla <aabdulla@nvidia.com>
Signed-off-by: Jeff Garzik <jeff@garzik.org>
drivers/net/forcedeth.c

index 5472d12122b4345b448ebffd959549a0c8304fcb..3b8087159b6d4c3339d4c2f395bbb6dc63274616 100644 (file)
  *     0.56: 22 Mar 2006: Additional ethtool config and moduleparam support.
  *     0.57: 14 May 2006: Mac address set in probe/remove and order corrections.
  *     0.58: 30 Oct 2006: Added support for sideband management unit.
+ *     0.59: 30 Oct 2006: Added support for recoverable error.
  *
  * Known bugs:
  * We suspect that on some hardware no TX done interrupts are generated.
 #else
 #define DRIVERNAPI
 #endif
-#define FORCEDETH_VERSION              "0.58"
+#define FORCEDETH_VERSION              "0.59"
 #define DRV_NAME                       "forcedeth"
 
 #include <linux/module.h>
 enum {
        NvRegIrqStatus = 0x000,
 #define NVREG_IRQSTAT_MIIEVENT 0x040
-#define NVREG_IRQSTAT_MASK             0x1ff
+#define NVREG_IRQSTAT_MASK             0x81ff
        NvRegIrqMask = 0x004,
 #define NVREG_IRQ_RX_ERROR             0x0001
 #define NVREG_IRQ_RX                   0x0002
@@ -191,15 +192,16 @@ enum {
 #define NVREG_IRQ_LINK                 0x0040
 #define NVREG_IRQ_RX_FORCED            0x0080
 #define NVREG_IRQ_TX_FORCED            0x0100
+#define NVREG_IRQ_RECOVER_ERROR                0x8000
 #define NVREG_IRQMASK_THROUGHPUT       0x00df
 #define NVREG_IRQMASK_CPU              0x0040
 #define NVREG_IRQ_TX_ALL               (NVREG_IRQ_TX_ERR|NVREG_IRQ_TX_OK|NVREG_IRQ_TX_FORCED)
 #define NVREG_IRQ_RX_ALL               (NVREG_IRQ_RX_ERROR|NVREG_IRQ_RX|NVREG_IRQ_RX_NOBUF|NVREG_IRQ_RX_FORCED)
-#define NVREG_IRQ_OTHER                        (NVREG_IRQ_TIMER|NVREG_IRQ_LINK)
+#define NVREG_IRQ_OTHER                        (NVREG_IRQ_TIMER|NVREG_IRQ_LINK|NVREG_IRQ_RECOVER_ERROR)
 
 #define NVREG_IRQ_UNKNOWN      (~(NVREG_IRQ_RX_ERROR|NVREG_IRQ_RX|NVREG_IRQ_RX_NOBUF|NVREG_IRQ_TX_ERR| \
                                        NVREG_IRQ_TX_OK|NVREG_IRQ_TIMER|NVREG_IRQ_LINK|NVREG_IRQ_RX_FORCED| \
-                                       NVREG_IRQ_TX_FORCED))
+                                       NVREG_IRQ_TX_FORCED|NVREG_IRQ_RECOVER_ERROR))
 
        NvRegUnknownSetupReg6 = 0x008,
 #define NVREG_UNKSETUP6_VAL            3
@@ -718,6 +720,7 @@ struct fe_priv {
        unsigned int phy_model;
        u16 gigabit;
        int intr_test;
+       int recover_error;
 
        /* General data: RO fields */
        dma_addr_t ring_addr;
@@ -2455,6 +2458,23 @@ static irqreturn_t nv_nic_irq(int foo, void *data)
                        printk(KERN_DEBUG "%s: received irq with unknown events 0x%x. Please report\n",
                                                dev->name, events);
                }
+               if (unlikely(events & NVREG_IRQ_RECOVER_ERROR)) {
+                       spin_lock(&np->lock);
+                       /* disable interrupts on the nic */
+                       if (!(np->msi_flags & NV_MSI_X_ENABLED))
+                               writel(0, base + NvRegIrqMask);
+                       else
+                               writel(np->irqmask, base + NvRegIrqMask);
+                       pci_push(base);
+
+                       if (!np->in_shutdown) {
+                               np->nic_poll_irq = np->irqmask;
+                               np->recover_error = 1;
+                               mod_timer(&np->nic_poll, jiffies + POLL_WAIT);
+                       }
+                       spin_unlock(&np->lock);
+                       break;
+               }
 #ifdef CONFIG_FORCEDETH_NAPI
                if (events & NVREG_IRQ_RX_ALL) {
                        netif_rx_schedule(dev);
@@ -2685,6 +2705,20 @@ static irqreturn_t nv_nic_irq_other(int foo, void *data)
                        spin_unlock_irqrestore(&np->lock, flags);
                        np->link_timeout = jiffies + LINK_TIMEOUT;
                }
+               if (events & NVREG_IRQ_RECOVER_ERROR) {
+                       spin_lock_irq(&np->lock);
+                       /* disable interrupts on the nic */
+                       writel(NVREG_IRQ_OTHER, base + NvRegIrqMask);
+                       pci_push(base);
+
+                       if (!np->in_shutdown) {
+                               np->nic_poll_irq |= NVREG_IRQ_OTHER;
+                               np->recover_error = 1;
+                               mod_timer(&np->nic_poll, jiffies + POLL_WAIT);
+                       }
+                       spin_unlock_irq(&np->lock);
+                       break;
+               }
                if (events & (NVREG_IRQ_UNKNOWN)) {
                        printk(KERN_DEBUG "%s: received irq with unknown events 0x%x. Please report\n",
                                                dev->name, events);
@@ -2914,6 +2948,42 @@ static void nv_do_nic_poll(unsigned long data)
        }
        np->nic_poll_irq = 0;
 
+       if (np->recover_error) {
+               np->recover_error = 0;
+               printk(KERN_INFO "forcedeth: MAC in recoverable error state\n");
+               if (netif_running(dev)) {
+                       netif_tx_lock_bh(dev);
+                       spin_lock(&np->lock);
+                       /* stop engines */
+                       nv_stop_rx(dev);
+                       nv_stop_tx(dev);
+                       nv_txrx_reset(dev);
+                       /* drain rx queue */
+                       nv_drain_rx(dev);
+                       nv_drain_tx(dev);
+                       /* reinit driver view of the rx queue */
+                       set_bufsize(dev);
+                       if (nv_init_ring(dev)) {
+                               if (!np->in_shutdown)
+                                       mod_timer(&np->oom_kick, jiffies + OOM_REFILL);
+                       }
+                       /* reinit nic view of the rx queue */
+                       writel(np->rx_buf_sz, base + NvRegOffloadConfig);
+                       setup_hw_rings(dev, NV_SETUP_RX_RING | NV_SETUP_TX_RING);
+                       writel( ((np->rx_ring_size-1) << NVREG_RINGSZ_RXSHIFT) + ((np->tx_ring_size-1) << NVREG_RINGSZ_TXSHIFT),
+                               base + NvRegRingSizes);
+                       pci_push(base);
+                       writel(NVREG_TXRXCTL_KICK|np->txrxctl_bits, get_hwbase(dev) + NvRegTxRxControl);
+                       pci_push(base);
+
+                       /* restart rx engine */
+                       nv_start_rx(dev);
+                       nv_start_tx(dev);
+                       spin_unlock(&np->lock);
+                       netif_tx_unlock_bh(dev);
+               }
+       }
+
        /* FIXME: Do we need synchronize_irq(dev->irq) here? */
 
        writel(mask, base + NvRegIrqMask);