acenic_np.c

该软件根据网络数据生成NetFlow记录。NetFlow可用于网络规划、负载均衡、安全监控等

		 RX_JUMBO_RING_ENTRIES +
		 RX_MINI_RING_ENTRIES +
		 RX_RETURN_RING_ENTRIES));

	ap->rx_std_ring = pci_alloc_consistent(ap->pdev, size,
					       &ap->rx_ring_base_dma);
	if (ap->rx_std_ring == NULL)
		goto fail;

	ap->rx_jumbo_ring = ap->rx_std_ring + RX_STD_RING_ENTRIES;
	ap->rx_mini_ring = ap->rx_jumbo_ring + RX_JUMBO_RING_ENTRIES;
	ap->rx_return_ring = ap->rx_mini_ring + RX_MINI_RING_ENTRIES;

	size = (sizeof(struct event) * EVT_RING_ENTRIES);

	ap->evt_ring = pci_alloc_consistent(ap->pdev, size, &ap->evt_ring_dma);

	if (ap->evt_ring == NULL)
		goto fail;

	size = (sizeof(struct tx_desc) * TX_RING_ENTRIES);

	ap->tx_ring = pci_alloc_consistent(ap->pdev, size, &ap->tx_ring_dma);

	if (ap->tx_ring == NULL)
		goto fail;

	ap->evt_prd = pci_alloc_consistent(ap->pdev, sizeof(u32),
					   &ap->evt_prd_dma);
	if (ap->evt_prd == NULL)
		goto fail;

	ap->rx_ret_prd = pci_alloc_consistent(ap->pdev, sizeof(u32),
					      &ap->rx_ret_prd_dma);
	if (ap->rx_ret_prd == NULL)
		goto fail;

	ap->tx_csm = pci_alloc_consistent(ap->pdev, sizeof(u32),
					  &ap->tx_csm_dma);
	if (ap->tx_csm == NULL)
		goto fail;

	return 0;

fail:
	/* Clean up. */
	ace_init_cleanup(dev);
	return 1;
}


/*
 * Generic cleanup handling data allocated during init. Used when the
 * module is unloaded or if an error occurs during initialization
 */
static void ace_init_cleanup(struct net_device *dev)
{
	struct ace_private *ap;

	ap = dev->priv;

	ace_free_descriptors(dev);

	if (ap->info)
		pci_free_consistent(ap->pdev, sizeof(struct ace_info),
				    ap->info, ap->info_dma);
	if (ap->skb)
		kfree(ap->skb);
	if (ap->trace_buf)
		kfree(ap->trace_buf);

	if (dev->irq)
		free_irq(dev->irq, dev);

	unregister_netdev(dev);
	iounmap(ap->regs);
}


/*
 * Commands are considered to be slow.
 */
static inline void ace_issue_cmd(struct ace_regs *regs, struct cmd *cmd)
{
	u32 idx;

	idx = readl(&regs->CmdPrd);

	writel(*(u32 *)(cmd), &regs->CmdRng[idx]);
	idx = (idx + 1) % CMD_RING_ENTRIES;

	writel(idx, &regs->CmdPrd);
}


static int __init ace_init(struct net_device *dev)
{
	struct ace_private *ap;
	struct ace_regs *regs;
	struct ace_info *info = NULL;
	unsigned long tmp_ptr, myjif;
	u32 tig_ver, mac1, mac2, tmp, pci_state;
	int board_idx, ecode = 0;
	short i;
	unsigned char cache;

	ap = dev->priv;
	regs = ap->regs;

	board_idx = ap->board_idx;

	/*
	 * [e-mail] - its useful to do a NIC reset here to
	 * address the `Firmware not running' problem subsequent
	 * to any crashes involving the NIC
	 */
	writel(HW_RESET | (HW_RESET << 24), &regs->HostCtrl);
	wmb();

	/*
	 * Don't access any other registes before this point!
	 */
#ifdef __BIG_ENDIAN
	/*
	 * This will most likely need BYTE_SWAP once we switch
	 * to using __raw_writel()
	 */
#ifdef __parisc__
	writel((WORD_SWAP | BYTE_SWAP | CLR_INT |
		((WORD_SWAP | BYTE_SWAP | CLR_INT) << 24)),
	       &regs->HostCtrl);
#else
	writel((WORD_SWAP | CLR_INT | ((WORD_SWAP | CLR_INT) << 24)),
	       &regs->HostCtrl);
#endif
#else
	writel((CLR_INT | WORD_SWAP | ((CLR_INT | WORD_SWAP) << 24)),
	       &regs->HostCtrl);
#endif
	mb();

	/*
	 * Stop the NIC CPU and clear pending interrupts
	 */
	writel(readl(&regs->CpuCtrl) | CPU_HALT, &regs->CpuCtrl);
	writel(0, &regs->Mb0Lo);

	tig_ver = readl(&regs->HostCtrl) >> 28;

	switch(tig_ver){
#ifndef CONFIG_ACENIC_OMIT_TIGON_I
	case 4:
		printk(KERN_INFO "  Tigon I  (Rev. 4), Firmware: %i.%i.%i, ",
		       tigonFwReleaseMajor, tigonFwReleaseMinor,
		       tigonFwReleaseFix);
		writel(0, &regs->LocalCtrl);
		ap->version = 1;
		break;
#endif
	case 6:
		printk(KERN_INFO "  Tigon II (Rev. %i), Firmware: %i.%i.%i, ",
		       tig_ver, tigon2FwReleaseMajor, tigon2FwReleaseMinor,
		       tigon2FwReleaseFix);
		writel(readl(&regs->CpuBCtrl) | CPU_HALT, &regs->CpuBCtrl);
		/*
		 * The SRAM bank size does _not_ indicate the amount
		 * of memory on the card, it controls the _bank_ size!
		 * Ie. a 1MB AceNIC will have two banks of 512KB.
		 */
		writel(SRAM_BANK_512K, &regs->LocalCtrl);
		writel(SYNC_SRAM_TIMING, &regs->MiscCfg);
		ap->version = 2;
		break;
	default:
		printk(KERN_WARNING "  Unsupported Tigon version detected "
		       "(%i), ", tig_ver);
		ecode = -ENODEV;
		goto init_error;
	}

	/*
	 * ModeStat _must_ be set after the SRAM settings as this change
	 * seems to corrupt the ModeStat and possible other registers.
	 * The SRAM settings survive resets and setting it to the same
	 * value a second time works as well. This is what caused the
	 * `Firmware not running' problem on the Tigon II.
	 */
#ifdef __BIG_ENDIAN
	writel(ACE_BYTE_SWAP_DMA | ACE_WARN | ACE_FATAL | ACE_BYTE_SWAP_BD |
	       ACE_WORD_SWAP_BD | ACE_NO_JUMBO_FRAG, &regs->ModeStat);
#else
	writel(ACE_BYTE_SWAP_DMA | ACE_WARN | ACE_FATAL |
	       ACE_WORD_SWAP_BD | ACE_NO_JUMBO_FRAG, &regs->ModeStat);
#endif
	mb();

	mac1 = 0;
	for(i = 0; i < 4; i++) {
		mac1 = mac1 << 8;
		tmp = read_eeprom_byte(dev, 0x8c+i);
		if (tmp < 0) {
			ecode = -EIO;
			goto init_error;
		} else
			mac1 |= (tmp & 0xff);
	}
	mac2 = 0;
	for(i = 4; i < 8; i++) {
		mac2 = mac2 << 8;
		tmp = read_eeprom_byte(dev, 0x8c+i);
		if (tmp < 0) {
			ecode = -EIO;
			goto init_error;
		} else
			mac2 |= (tmp & 0xff);
	}

	writel(mac1, &regs->MacAddrHi);
	writel(mac2, &regs->MacAddrLo);

	printk("MAC: %02x:%02x:%02x:%02x:%02x:%02x\n",
	       (mac1 >> 8) & 0xff, mac1 & 0xff, (mac2 >> 24) &0xff,
	       (mac2 >> 16) & 0xff, (mac2 >> 8) & 0xff, mac2 & 0xff);

	dev->dev_addr[0] = (mac1 >> 8) & 0xff;
	dev->dev_addr[1] = mac1 & 0xff;
	dev->dev_addr[2] = (mac2 >> 24) & 0xff;
	dev->dev_addr[3] = (mac2 >> 16) & 0xff;
	dev->dev_addr[4] = (mac2 >> 8) & 0xff;
	dev->dev_addr[5] = mac2 & 0xff;

	/*
	 * Looks like this is necessary to deal with on all architectures,
	 * even this %$#%$# N440BX Intel based thing doesn't get it right.
	 * Ie. having two NICs in the machine, one will have the cache
	 * line set at boot time, the other will not.
	 */
	pci_read_config_byte(ap->pdev, PCI_CACHE_LINE_SIZE, &cache);
	if ((cache << 2) != SMP_CACHE_BYTES) {
		printk(KERN_INFO "  PCI cache line size set incorrectly "
		       "(%i bytes) by BIOS/FW, correcting to %i\n",
		       (cache << 2), SMP_CACHE_BYTES);
		pci_write_config_byte(ap->pdev, PCI_CACHE_LINE_SIZE,
				      SMP_CACHE_BYTES >> 2);
	}

	pci_state = readl(&regs->PciState);
	printk(KERN_INFO "  PCI bus width: %i bits, speed: %iMHz, "
	       "latency: %i clks\n",
	       	(pci_state & PCI_32BIT) ? 32 : 64,
		(pci_state & PCI_66MHZ) ? 66 : 33, 
		ap->pci_latency);

	/*
	 * Set the max DMA transfer size. Seems that for most systems
	 * the performance is better when no MAX parameter is
	 * set. However for systems enabling PCI write and invalidate,
	 * DMA writes must be set to the L1 cache line size to get
	 * optimal performance.
	 *
	 * The default is now to turn the PCI write and invalidate off
	 * - that is what Alteon does for NT.
	 */
	tmp = READ_CMD_MEM | WRITE_CMD_MEM;
	if (ap->version >= 2) {
		tmp |= (MEM_READ_MULTIPLE | (pci_state & PCI_66MHZ));
		/*
		 * Tuning parameters only supported for 8 cards
		 */
		if (board_idx == BOARD_IDX_OVERFLOW ||
		    dis_pci_mem_inval[board_idx]) {
			if (ap->pci_command & PCI_COMMAND_INVALIDATE) {
				ap->pci_command &= ~PCI_COMMAND_INVALIDATE;
				pci_write_config_word(ap->pdev, PCI_COMMAND,
						      ap->pci_command);
				printk(KERN_INFO "  Disabling PCI memory "
				       "write and invalidate\n");
			}
		} else if (ap->pci_command & PCI_COMMAND_INVALIDATE) {
			printk(KERN_INFO "  PCI memory write & invalidate "
			       "enabled by BIOS, enabling counter measures\n");

			switch(SMP_CACHE_BYTES) {
			case 16:
				tmp |= DMA_WRITE_MAX_16;
				break;
			case 32:
				tmp |= DMA_WRITE_MAX_32;
				break;
			case 64:
				tmp |= DMA_WRITE_MAX_64;
				break;
			default:
				printk(KERN_INFO "  Cache line size %i not "
				       "supported, PCI write and invalidate "
				       "disabled\n", SMP_CACHE_BYTES);
				ap->pci_command &= ~PCI_COMMAND_INVALIDATE;
				pci_write_config_word(ap->pdev, PCI_COMMAND,
						      ap->pci_command);
			}
		}
	}

#ifdef __sparc__
	/*
	 * On this platform, we know what the best dma settings
	 * are.  We use 64-byte maximum bursts, because if we
	 * burst larger than the cache line size (or even cross
	 * a 64byte boundry in a single burst) the UltraSparc
	 * PCI controller will disconnect at 64-byte multiples.
	 *
	 * Read-multiple will be properly enabled above, and when
	 * set will give the PCI controller proper hints about
	 * prefetching.
	 */
	tmp = tmp & ~DMA_READ_WRITE_MASK;
	tmp |= DMA_READ_MAX_64;
	tmp |= DMA_WRITE_MAX_64;
#endif
	writel(tmp, &regs->PciState);

#if 0
	/*
	 * I have received reports from people having problems when this
	 * bit is enabled.
	 */
	if (!(ap->pci_command & PCI_COMMAND_FAST_BACK)) {
		printk(KERN_INFO "  Enabling PCI Fast Back to Back\n");
		ap->pci_command |= PCI_COMMAND_FAST_BACK;
		pci_write_config_word(ap->pdev, PCI_COMMAND, ap->pci_command);
	}
#endif
		
	/*
	 * Initialize the generic info block and the command+event rings
	 * and the control blocks for the transmit and receive rings
	 * as they need to be setup once and for all.
	 */
	if (!(info = pci_alloc_consistent(ap->pdev, sizeof(struct ace_info),
				    &ap->info_dma))) {
		ecode = -EAGAIN;
		goto init_error;
	}
	ap->info = info;

	/*
	 * Get the memory for the skb rings.
	 */
	if (!(ap->skb = kmalloc(sizeof(struct ace_skb), GFP_KERNEL))) {
		ecode = -EAGAIN;
		goto init_error;
	}

	ecode = request_irq(dev->irq, ace_interrupt, SA_SHIRQ, dev->name, dev);
	if (ecode) {
		printk(KERN_WARNING "%s: Requested IRQ %d is busy\n",
		       dev->name, dev->irq);
		goto init_error;
	}

	/*
	 * Register the device here to be able to catch allocated
	 * interrupt handlers in case the firmware doesn't come up.
	 */
	ap->next = root_dev;
	root_dev = dev;

#ifdef INDEX_DEBUG
	spin_lock_init(&ap->debug_lock);
	ap->last_tx = TX_RING_ENTRIES - 1;
	ap->last_std_rx = 0;
	ap->last_mini_rx = 0;
#endif

	memset(ap->info, 0, sizeof(struct ace_info));
	memset(ap->skb, 0, sizeof(struct ace_skb));

	ace_load_firmware(dev);
	ap->fw_running = 0;

	tmp_ptr = (unsigned long) ap->info_dma;
#ifdef ACE_64BIT_PTR
	writel(tmp_ptr >> 32, &regs->InfoPtrHi);
#else
	writel(0, &regs->InfoPtrHi);
#endif
	writel(tmp_ptr & 0xffffffff, &regs->InfoPtrLo);

	memset(ap->evt_ring, 0, EVT_RING_ENTRIES * sizeof(struct event));

	set_aceaddr(&info->evt_ctrl.rngptr, ap->evt_ring_dma);
	info->evt_ctrl.flags = 0;

	set_aceaddr(&info->evt_prd_ptr, ap->evt_prd_dma);
	*(ap->evt_prd) = 0;
	wmb();
	writel(0, &regs->EvtCsm);

	set_aceaddr(&info->cmd_ctrl.rngptr, 0x100);
	info->cmd_ctrl.flags = 0;
	info->cmd_ctrl.max_len = 0;

	for (i = 0; i < CMD_RING_ENTRIES; i++)
		writel(0, &regs->CmdRng[i]);

	writel(0, &regs->CmdPrd);
	writel(0, &regs->CmdCsm);

	tmp_ptr = ap->info_dma;
	tmp_ptr += (unsigned long) &(((struct ace_info *)0)->s.stats);
	set_aceaddr(&info->stats2_ptr, (dma_addr_t) tmp_ptr);

	set_aceaddr(&info->rx_std_ctrl.rngptr, ap->rx_ring_base_dma);
	info->rx_std_ctrl.max_len = ACE_STD_MTU + ETH_HLEN + 4;
	info->rx_std_ctrl.flags = RCB_FLG_TCP_UDP_SUM;

	memset(ap->rx_std_ring, 0,
	       RX_STD_RING_ENTRIES * sizeof(struct rx_desc));

	for (i = 0; i < RX_STD_RING_ENTRIES; i++)
		ap->rx_std_ring[i].flags = BD_FLG_TCP_UDP_SUM;

	ap->rx_std_skbprd = 0;
	atomic_set(&ap->cur_rx_bufs, 0);

	set_aceaddr(&info->rx_jumbo_ctrl.rngptr,
		    (ap->rx_ring_base_dma +
		     (sizeof(struct rx_desc) * RX_STD_RING_ENTRIES)));
	info->rx_jumbo_ctrl.max_len = 0;
	info->rx_jumbo_ctrl.flags = RCB_FLG_TCP_UDP_SUM;

	memset(ap->rx_jumbo_ring, 0,
	       RX_JUMBO_RING_ENTRIES * sizeof(struct rx_desc));

	for (i = 0; i < RX_JUMBO_RING_ENTRIES; i++)
		ap->rx_jumbo_ring[i].flags = BD_FLG_TCP_UDP_SUM | BD_FLG_JUMBO;

	ap->rx_jumbo_skbprd = 0;
	atomic_set(&ap->cur_jumbo_bufs, 0);

	memset(ap->rx_mini_ring, 0,
	       RX_MINI_RING_ENTRIES * sizeof(struct rx_desc));

	if (ap->version >= 2) {
		set_aceaddr(&info->rx_mini_ctrl.rngptr,
			    (ap->rx_ring_base_dma +
			     (sizeof(struct rx_desc) *
			      (RX_STD_RING_ENTRIES +
			       RX_JUMBO_RING_ENTRIES))));
		info->rx_mini_ctrl.max_len = ACE_MINI_SIZE;
		info->rx_mini_ctrl.flags = RCB_FLG_TCP_UDP_SUM;

		for (i = 0; i < RX_MINI_RING_ENTRIES; i++)
			ap->rx_mini_ring[i].flags =
				BD_FLG_TCP_UDP_SUM | BD_FLG_MINI;
	} else {
		set_aceaddr(&info->rx_mini_ctrl.rngptr, 0);
		info->rx_mini_ctrl.flags = RCB_FLG_RNG_DISABLE;
		info->rx_mini_ctrl.max_len = 0;
	}

	ap->rx_mini_skbprd = 0;
	atomic_set(&ap->cur_mini_bufs, 0);

	set_aceaddr(&info->rx_return_ctrl.rngptr,
		    (ap->rx_ring_base_dma +
		     (sizeof(struct rx_desc) *
		      (RX_STD_RING_ENTRIES +
		       RX_JUMBO_RING_ENTRIES +