Merge branch 'for_paulus' of master.kernel.org:/pub/scm/linux/kernel/git/galak/powerpc
[cascardo/linux.git] / drivers / net / bnx2.c
index 6fbb548..5ca99e2 100644 (file)
@@ -9,20 +9,61 @@
  * Written by: Michael Chan  (mchan@broadcom.com)
  */
 
+#include <linux/config.h>
+
+#include <linux/module.h>
+#include <linux/moduleparam.h>
+
+#include <linux/kernel.h>
+#include <linux/timer.h>
+#include <linux/errno.h>
+#include <linux/ioport.h>
+#include <linux/slab.h>
+#include <linux/vmalloc.h>
+#include <linux/interrupt.h>
+#include <linux/pci.h>
+#include <linux/init.h>
+#include <linux/netdevice.h>
+#include <linux/etherdevice.h>
+#include <linux/skbuff.h>
+#include <linux/dma-mapping.h>
+#include <asm/bitops.h>
+#include <asm/io.h>
+#include <asm/irq.h>
+#include <linux/delay.h>
+#include <asm/byteorder.h>
+#include <linux/time.h>
+#include <linux/ethtool.h>
+#include <linux/mii.h>
+#ifdef NETIF_F_HW_VLAN_TX
+#include <linux/if_vlan.h>
+#define BCM_VLAN 1
+#endif
+#ifdef NETIF_F_TSO
+#include <net/ip.h>
+#include <net/tcp.h>
+#include <net/checksum.h>
+#define BCM_TSO 1
+#endif
+#include <linux/workqueue.h>
+#include <linux/crc32.h>
+#include <linux/prefetch.h>
+#include <linux/cache.h>
+
 #include "bnx2.h"
 #include "bnx2_fw.h"
 
 #define DRV_MODULE_NAME                "bnx2"
 #define PFX DRV_MODULE_NAME    ": "
-#define DRV_MODULE_VERSION     "1.4.31"
-#define DRV_MODULE_RELDATE     "January 19, 2006"
+#define DRV_MODULE_VERSION     "1.4.39"
+#define DRV_MODULE_RELDATE     "March 22, 2006"
 
 #define RUN_AT(x) (jiffies + (x))
 
 /* Time in jiffies before concluding the transmitter is hung. */
 #define TX_TIMEOUT  (5*HZ)
 
-static char version[] __devinitdata =
+static const char version[] __devinitdata =
        "Broadcom NetXtreme II Gigabit Ethernet Driver " DRV_MODULE_NAME " v" DRV_MODULE_VERSION " (" DRV_MODULE_RELDATE ")\n";
 
 MODULE_AUTHOR("Michael Chan <mchan@broadcom.com>");
@@ -313,8 +354,6 @@ bnx2_disable_int(struct bnx2 *bp)
 static void
 bnx2_enable_int(struct bnx2 *bp)
 {
-       u32 val;
-
        REG_WR(bp, BNX2_PCICFG_INT_ACK_CMD,
               BNX2_PCICFG_INT_ACK_CMD_INDEX_VALID |
               BNX2_PCICFG_INT_ACK_CMD_MASK_INT | bp->last_status_idx);
@@ -322,8 +361,7 @@ bnx2_enable_int(struct bnx2 *bp)
        REG_WR(bp, BNX2_PCICFG_INT_ACK_CMD,
               BNX2_PCICFG_INT_ACK_CMD_INDEX_VALID | bp->last_status_idx);
 
-       val = REG_RD(bp, BNX2_HC_COMMAND);
-       REG_WR(bp, BNX2_HC_COMMAND, val | BNX2_HC_COMMAND_COAL_NOW);
+       REG_WR(bp, BNX2_HC_COMMAND, bp->hc_cmd | BNX2_HC_COMMAND_COAL_NOW);
 }
 
 static void
@@ -360,15 +398,13 @@ bnx2_netif_start(struct bnx2 *bp)
 static void
 bnx2_free_mem(struct bnx2 *bp)
 {
-       if (bp->stats_blk) {
-               pci_free_consistent(bp->pdev, sizeof(struct statistics_block),
-                                   bp->stats_blk, bp->stats_blk_mapping);
-               bp->stats_blk = NULL;
-       }
+       int i;
+
        if (bp->status_blk) {
-               pci_free_consistent(bp->pdev, sizeof(struct status_block),
+               pci_free_consistent(bp->pdev, bp->status_stats_size,
                                    bp->status_blk, bp->status_blk_mapping);
                bp->status_blk = NULL;
+               bp->stats_blk = NULL;
        }
        if (bp->tx_desc_ring) {
                pci_free_consistent(bp->pdev,
@@ -378,25 +414,28 @@ bnx2_free_mem(struct bnx2 *bp)
        }
        kfree(bp->tx_buf_ring);
        bp->tx_buf_ring = NULL;
-       if (bp->rx_desc_ring) {
-               pci_free_consistent(bp->pdev,
-                                   sizeof(struct rx_bd) * RX_DESC_CNT,
-                                   bp->rx_desc_ring, bp->rx_desc_mapping);
-               bp->rx_desc_ring = NULL;
-       }
-       kfree(bp->rx_buf_ring);
+       for (i = 0; i < bp->rx_max_ring; i++) {
+               if (bp->rx_desc_ring[i])
+                       pci_free_consistent(bp->pdev,
+                                           sizeof(struct rx_bd) * RX_DESC_CNT,
+                                           bp->rx_desc_ring[i],
+                                           bp->rx_desc_mapping[i]);
+               bp->rx_desc_ring[i] = NULL;
+       }
+       vfree(bp->rx_buf_ring);
        bp->rx_buf_ring = NULL;
 }
 
 static int
 bnx2_alloc_mem(struct bnx2 *bp)
 {
-       bp->tx_buf_ring = kmalloc(sizeof(struct sw_bd) * TX_DESC_CNT,
-                                    GFP_KERNEL);
+       int i, status_blk_size;
+
+       bp->tx_buf_ring = kzalloc(sizeof(struct sw_bd) * TX_DESC_CNT,
+                                 GFP_KERNEL);
        if (bp->tx_buf_ring == NULL)
                return -ENOMEM;
 
-       memset(bp->tx_buf_ring, 0, sizeof(struct sw_bd) * TX_DESC_CNT);
        bp->tx_desc_ring = pci_alloc_consistent(bp->pdev,
                                                sizeof(struct tx_bd) *
                                                TX_DESC_CNT,
@@ -404,34 +443,40 @@ bnx2_alloc_mem(struct bnx2 *bp)
        if (bp->tx_desc_ring == NULL)
                goto alloc_mem_err;
 
-       bp->rx_buf_ring = kmalloc(sizeof(struct sw_bd) * RX_DESC_CNT,
-                                    GFP_KERNEL);
+       bp->rx_buf_ring = vmalloc(sizeof(struct sw_bd) * RX_DESC_CNT *
+                                 bp->rx_max_ring);
        if (bp->rx_buf_ring == NULL)
                goto alloc_mem_err;
 
-       memset(bp->rx_buf_ring, 0, sizeof(struct sw_bd) * RX_DESC_CNT);
-       bp->rx_desc_ring = pci_alloc_consistent(bp->pdev,
-                                               sizeof(struct rx_bd) *
-                                               RX_DESC_CNT,
-                                               &bp->rx_desc_mapping);
-       if (bp->rx_desc_ring == NULL)
-               goto alloc_mem_err;
+       memset(bp->rx_buf_ring, 0, sizeof(struct sw_bd) * RX_DESC_CNT *
+                                  bp->rx_max_ring);
+
+       for (i = 0; i < bp->rx_max_ring; i++) {
+               bp->rx_desc_ring[i] =
+                       pci_alloc_consistent(bp->pdev,
+                                            sizeof(struct rx_bd) * RX_DESC_CNT,
+                                            &bp->rx_desc_mapping[i]);
+               if (bp->rx_desc_ring[i] == NULL)
+                       goto alloc_mem_err;
+
+       }
 
-       bp->status_blk = pci_alloc_consistent(bp->pdev,
-                                             sizeof(struct status_block),
+       /* Combine status and statistics blocks into one allocation. */
+       status_blk_size = L1_CACHE_ALIGN(sizeof(struct status_block));
+       bp->status_stats_size = status_blk_size +
+                               sizeof(struct statistics_block);
+
+       bp->status_blk = pci_alloc_consistent(bp->pdev, bp->status_stats_size,
                                              &bp->status_blk_mapping);
        if (bp->status_blk == NULL)
                goto alloc_mem_err;
 
-       memset(bp->status_blk, 0, sizeof(struct status_block));
+       memset(bp->status_blk, 0, bp->status_stats_size);
 
-       bp->stats_blk = pci_alloc_consistent(bp->pdev,
-                                            sizeof(struct statistics_block),
-                                            &bp->stats_blk_mapping);
-       if (bp->stats_blk == NULL)
-               goto alloc_mem_err;
+       bp->stats_blk = (void *) ((unsigned long) bp->status_blk +
+                                 status_blk_size);
 
-       memset(bp->stats_blk, 0, sizeof(struct statistics_block));
+       bp->stats_blk_mapping = bp->status_blk_mapping + status_blk_size;
 
        return 0;
 
@@ -1520,7 +1565,7 @@ bnx2_alloc_rx_skb(struct bnx2 *bp, u16 index)
        struct sk_buff *skb;
        struct sw_bd *rx_buf = &bp->rx_buf_ring[index];
        dma_addr_t mapping;
-       struct rx_bd *rxbd = &bp->rx_desc_ring[index];
+       struct rx_bd *rxbd = &bp->rx_desc_ring[RX_RING(index)][RX_IDX(index)];
        unsigned long align;
 
        skb = dev_alloc_skb(bp->rx_buf_size);
@@ -1656,23 +1701,30 @@ static inline void
 bnx2_reuse_rx_skb(struct bnx2 *bp, struct sk_buff *skb,
        u16 cons, u16 prod)
 {
-       struct sw_bd *cons_rx_buf = &bp->rx_buf_ring[cons];
-       struct sw_bd *prod_rx_buf = &bp->rx_buf_ring[prod];
-       struct rx_bd *cons_bd = &bp->rx_desc_ring[cons];
-       struct rx_bd *prod_bd = &bp->rx_desc_ring[prod];
+       struct sw_bd *cons_rx_buf, *prod_rx_buf;
+       struct rx_bd *cons_bd, *prod_bd;
+
+       cons_rx_buf = &bp->rx_buf_ring[cons];
+       prod_rx_buf = &bp->rx_buf_ring[prod];
 
        pci_dma_sync_single_for_device(bp->pdev,
                pci_unmap_addr(cons_rx_buf, mapping),
                bp->rx_offset + RX_COPY_THRESH, PCI_DMA_FROMDEVICE);
 
-       prod_rx_buf->skb = cons_rx_buf->skb;
-       pci_unmap_addr_set(prod_rx_buf, mapping,
-                       pci_unmap_addr(cons_rx_buf, mapping));
+       bp->rx_prod_bseq += bp->rx_buf_use_size;
 
-       memcpy(prod_bd, cons_bd, 8);
+       prod_rx_buf->skb = skb;
 
-       bp->rx_prod_bseq += bp->rx_buf_use_size;
+       if (cons == prod)
+               return;
 
+       pci_unmap_addr_set(prod_rx_buf, mapping,
+                       pci_unmap_addr(cons_rx_buf, mapping));
+
+       cons_bd = &bp->rx_desc_ring[RX_RING(cons)][RX_IDX(cons)];
+       prod_bd = &bp->rx_desc_ring[RX_RING(prod)][RX_IDX(prod)];
+       prod_bd->rx_bd_haddr_hi = cons_bd->rx_bd_haddr_hi;
+       prod_bd->rx_bd_haddr_lo = cons_bd->rx_bd_haddr_lo;
 }
 
 static int
@@ -1699,14 +1751,19 @@ bnx2_rx_int(struct bnx2 *bp, int budget)
                u32 status;
                struct sw_bd *rx_buf;
                struct sk_buff *skb;
+               dma_addr_t dma_addr;
 
                sw_ring_cons = RX_RING_IDX(sw_cons);
                sw_ring_prod = RX_RING_IDX(sw_prod);
 
                rx_buf = &bp->rx_buf_ring[sw_ring_cons];
                skb = rx_buf->skb;
-               pci_dma_sync_single_for_cpu(bp->pdev,
-                       pci_unmap_addr(rx_buf, mapping),
+
+               rx_buf->skb = NULL;
+
+               dma_addr = pci_unmap_addr(rx_buf, mapping);
+
+               pci_dma_sync_single_for_cpu(bp->pdev, dma_addr,
                        bp->rx_offset + RX_COPY_THRESH, PCI_DMA_FROMDEVICE);
 
                rx_hdr = (struct l2_fhdr *) skb->data;
@@ -1747,8 +1804,7 @@ bnx2_rx_int(struct bnx2 *bp, int budget)
                        skb = new_skb;
                }
                else if (bnx2_alloc_rx_skb(bp, sw_ring_prod) == 0) {
-                       pci_unmap_single(bp->pdev,
-                               pci_unmap_addr(rx_buf, mapping),
+                       pci_unmap_single(bp->pdev, dma_addr,
                                bp->rx_buf_use_size, PCI_DMA_FROMDEVICE);
 
                        skb_reserve(skb, bp->rx_offset);
@@ -1794,8 +1850,6 @@ reuse_rx:
                rx_pkt++;
 
 next_rx:
-               rx_buf->skb = NULL;
-
                sw_cons = NEXT_RX_BD(sw_cons);
                sw_prod = NEXT_RX_BD(sw_prod);
 
@@ -1906,6 +1960,13 @@ bnx2_poll(struct net_device *dev, int *budget)
                spin_lock(&bp->phy_lock);
                bnx2_phy_int(bp);
                spin_unlock(&bp->phy_lock);
+
+               /* This is needed to take care of transient status
+                * during link changes.
+                */
+               REG_WR(bp, BNX2_HC_COMMAND,
+                      bp->hc_cmd | BNX2_HC_COMMAND_COAL_NOW_WO_INT);
+               REG_RD(bp, BNX2_HC_COMMAND);
        }
 
        if (bp->status_blk->status_tx_quick_consumer_index0 != bp->hw_tx_cons)
@@ -3287,6 +3348,8 @@ bnx2_init_chip(struct bnx2 *bp)
 
        udelay(20);
 
+       bp->hc_cmd = REG_RD(bp, BNX2_HC_COMMAND);
+
        return rc;
 }
 
@@ -3340,27 +3403,35 @@ bnx2_init_rx_ring(struct bnx2 *bp)
        bp->hw_rx_cons = 0;
        bp->rx_prod_bseq = 0;
                
-       rxbd = &bp->rx_desc_ring[0];
-       for (i = 0; i < MAX_RX_DESC_CNT; i++, rxbd++) {
-               rxbd->rx_bd_len = bp->rx_buf_use_size;
-               rxbd->rx_bd_flags = RX_BD_FLAGS_START | RX_BD_FLAGS_END;
-       }
+       for (i = 0; i < bp->rx_max_ring; i++) {
+               int j;
 
-       rxbd->rx_bd_haddr_hi = (u64) bp->rx_desc_mapping >> 32;
-       rxbd->rx_bd_haddr_lo = (u64) bp->rx_desc_mapping & 0xffffffff;
+               rxbd = &bp->rx_desc_ring[i][0];
+               for (j = 0; j < MAX_RX_DESC_CNT; j++, rxbd++) {
+                       rxbd->rx_bd_len = bp->rx_buf_use_size;
+                       rxbd->rx_bd_flags = RX_BD_FLAGS_START | RX_BD_FLAGS_END;
+               }
+               if (i == (bp->rx_max_ring - 1))
+                       j = 0;
+               else
+                       j = i + 1;
+               rxbd->rx_bd_haddr_hi = (u64) bp->rx_desc_mapping[j] >> 32;
+               rxbd->rx_bd_haddr_lo = (u64) bp->rx_desc_mapping[j] &
+                                      0xffffffff;
+       }
 
        val = BNX2_L2CTX_CTX_TYPE_CTX_BD_CHN_TYPE_VALUE;
        val |= BNX2_L2CTX_CTX_TYPE_SIZE_L2;
        val |= 0x02 << 8;
        CTX_WR(bp, GET_CID_ADDR(RX_CID), BNX2_L2CTX_CTX_TYPE, val);
 
-       val = (u64) bp->rx_desc_mapping >> 32;
+       val = (u64) bp->rx_desc_mapping[0] >> 32;
        CTX_WR(bp, GET_CID_ADDR(RX_CID), BNX2_L2CTX_NX_BDHADDR_HI, val);
 
-       val = (u64) bp->rx_desc_mapping & 0xffffffff;
+       val = (u64) bp->rx_desc_mapping[0] & 0xffffffff;
        CTX_WR(bp, GET_CID_ADDR(RX_CID), BNX2_L2CTX_NX_BDHADDR_LO, val);
 
-       for ( ;ring_prod < bp->rx_ring_size; ) {
+       for (i = 0; i < bp->rx_ring_size; i++) {
                if (bnx2_alloc_rx_skb(bp, ring_prod) < 0) {
                        break;
                }
@@ -3374,6 +3445,29 @@ bnx2_init_rx_ring(struct bnx2 *bp)
        REG_WR(bp, MB_RX_CID_ADDR + BNX2_L2CTX_HOST_BSEQ, bp->rx_prod_bseq);
 }
 
+static void
+bnx2_set_rx_ring_size(struct bnx2 *bp, u32 size)
+{
+       u32 num_rings, max;
+
+       bp->rx_ring_size = size;
+       num_rings = 1;
+       while (size > MAX_RX_DESC_CNT) {
+               size -= MAX_RX_DESC_CNT;
+               num_rings++;
+       }
+       /* round to next power of 2 */
+       max = MAX_RX_RINGS;
+       while ((max & num_rings) == 0)
+               max >>= 1;
+
+       if (num_rings != max)
+               max <<= 1;
+
+       bp->rx_max_ring = max;
+       bp->rx_max_ring_idx = (bp->rx_max_ring * RX_DESC_CNT) - 1;
+}
+
 static void
 bnx2_free_tx_skbs(struct bnx2 *bp)
 {
@@ -3419,7 +3513,7 @@ bnx2_free_rx_skbs(struct bnx2 *bp)
        if (bp->rx_buf_ring == NULL)
                return;
 
-       for (i = 0; i < RX_DESC_CNT; i++) {
+       for (i = 0; i < bp->rx_max_ring_idx; i++) {
                struct sw_bd *rx_buf = &bp->rx_buf_ring[i];
                struct sk_buff *skb = rx_buf->skb;
 
@@ -3695,7 +3789,6 @@ bnx2_run_loopback(struct bnx2 *bp, int loopback_mode)
        struct sk_buff *skb, *rx_skb;
        unsigned char *packet;
        u16 rx_start_idx, rx_idx;
-       u32 val;
        dma_addr_t map;
        struct tx_bd *txbd;
        struct sw_bd *rx_buf;
@@ -3726,8 +3819,9 @@ bnx2_run_loopback(struct bnx2 *bp, int loopback_mode)
        map = pci_map_single(bp->pdev, skb->data, pkt_size,
                PCI_DMA_TODEVICE);
 
-       val = REG_RD(bp, BNX2_HC_COMMAND);
-       REG_WR(bp, BNX2_HC_COMMAND, val | BNX2_HC_COMMAND_COAL_NOW_WO_INT);
+       REG_WR(bp, BNX2_HC_COMMAND,
+              bp->hc_cmd | BNX2_HC_COMMAND_COAL_NOW_WO_INT);
+
        REG_RD(bp, BNX2_HC_COMMAND);
 
        udelay(5);
@@ -3751,8 +3845,9 @@ bnx2_run_loopback(struct bnx2 *bp, int loopback_mode)
 
        udelay(100);
 
-       val = REG_RD(bp, BNX2_HC_COMMAND);
-       REG_WR(bp, BNX2_HC_COMMAND, val | BNX2_HC_COMMAND_COAL_NOW_WO_INT);
+       REG_WR(bp, BNX2_HC_COMMAND,
+              bp->hc_cmd | BNX2_HC_COMMAND_COAL_NOW_WO_INT);
+
        REG_RD(bp, BNX2_HC_COMMAND);
 
        udelay(5);
@@ -3888,7 +3983,6 @@ static int
 bnx2_test_intr(struct bnx2 *bp)
 {
        int i;
-       u32 val;
        u16 status_idx;
 
        if (!netif_running(bp->dev))
@@ -3897,8 +3991,7 @@ bnx2_test_intr(struct bnx2 *bp)
        status_idx = REG_RD(bp, BNX2_PCICFG_INT_ACK_CMD) & 0xffff;
 
        /* This register is not touched during run-time. */
-       val = REG_RD(bp, BNX2_HC_COMMAND);
-       REG_WR(bp, BNX2_HC_COMMAND, val | BNX2_HC_COMMAND_COAL_NOW);
+       REG_WR(bp, BNX2_HC_COMMAND, bp->hc_cmd | BNX2_HC_COMMAND_COAL_NOW);
        REG_RD(bp, BNX2_HC_COMMAND);
 
        for (i = 0; i < 10; i++) {
@@ -4783,7 +4876,7 @@ bnx2_get_ringparam(struct net_device *dev, struct ethtool_ringparam *ering)
 {
        struct bnx2 *bp = netdev_priv(dev);
 
-       ering->rx_max_pending = MAX_RX_DESC_CNT;
+       ering->rx_max_pending = MAX_TOTAL_RX_DESC_CNT;
        ering->rx_mini_max_pending = 0;
        ering->rx_jumbo_max_pending = 0;
 
@@ -4800,17 +4893,28 @@ bnx2_set_ringparam(struct net_device *dev, struct ethtool_ringparam *ering)
 {
        struct bnx2 *bp = netdev_priv(dev);
 
-       if ((ering->rx_pending > MAX_RX_DESC_CNT) ||
+       if ((ering->rx_pending > MAX_TOTAL_RX_DESC_CNT) ||
                (ering->tx_pending > MAX_TX_DESC_CNT) ||
                (ering->tx_pending <= MAX_SKB_FRAGS)) {
 
                return -EINVAL;
        }
-       bp->rx_ring_size = ering->rx_pending;
+       if (netif_running(bp->dev)) {
+               bnx2_netif_stop(bp);
+               bnx2_reset_chip(bp, BNX2_DRV_MSG_CODE_RESET);
+               bnx2_free_skbs(bp);
+               bnx2_free_mem(bp);
+       }
+
+       bnx2_set_rx_ring_size(bp, ering->rx_pending);
        bp->tx_ring_size = ering->tx_pending;
 
        if (netif_running(bp->dev)) {
-               bnx2_netif_stop(bp);
+               int rc;
+
+               rc = bnx2_alloc_mem(bp);
+               if (rc)
+                       return rc;
                bnx2_init_nic(bp);
                bnx2_netif_start(bp);
        }
@@ -5484,7 +5588,7 @@ bnx2_init_board(struct pci_dev *pdev, struct net_device *dev)
        bp->mac_addr[5] = (u8) reg;
 
        bp->tx_ring_size = MAX_TX_DESC_CNT;
-       bp->rx_ring_size = 100;
+       bnx2_set_rx_ring_size(bp, 100);
 
        bp->rx_csum = 1;
 
@@ -5703,6 +5807,7 @@ bnx2_suspend(struct pci_dev *pdev, pm_message_t state)
        if (!netif_running(dev))
                return 0;
 
+       flush_scheduled_work();
        bnx2_netif_stop(bp);
        netif_device_detach(dev);
        del_timer_sync(&bp->timer);