typhoon.c

来自「Linux Kernel 2.6.9 for OMAP1710」· C语言 代码 · 共 2,314 行 · 第 1/5 页

	 * process of issuing a command, so we don't need to respond.
	 */
	if(spin_trylock(&tp->command_lock)) {
		cmd = (struct cmd_desc *)(ring->ringBase + ring->lastWrite);
		typhoon_inc_cmd_index(&ring->lastWrite, 1);

		INIT_COMMAND_NO_RESPONSE(cmd, TYPHOON_CMD_HELLO_RESP);
		smp_wmb();
		writel(ring->lastWrite, tp->ioaddr + TYPHOON_REG_CMD_READY);
		spin_unlock(&tp->command_lock);
	}
}

static int
typhoon_process_response(struct typhoon *tp, int resp_size,
				struct resp_desc *resp_save)
{
	struct typhoon_indexes *indexes = tp->indexes;
	struct resp_desc *resp;
	u8 *base = tp->respRing.ringBase;
	int count, len, wrap_len;
	u32 cleared;
	u32 ready;

	cleared = le32_to_cpu(indexes->respCleared);
	ready = le32_to_cpu(indexes->respReady);
	while(cleared != ready) {
		resp = (struct resp_desc *)(base + cleared);
		count = resp->numDesc + 1;
		if(resp_save && resp->seqNo) {
			if(count > resp_size) {
				resp_save->flags = TYPHOON_RESP_ERROR;
				goto cleanup;
			}

			wrap_len = 0;
			len = count * sizeof(*resp);
			if(unlikely(cleared + len > RESPONSE_RING_SIZE)) {
				wrap_len = cleared + len - RESPONSE_RING_SIZE;
				len = RESPONSE_RING_SIZE - cleared;
			}

			memcpy(resp_save, resp, len);
			if(unlikely(wrap_len)) {
				resp_save += len / sizeof(*resp);
				memcpy(resp_save, base, wrap_len);
			}

			resp_save = NULL;
		} else if(resp->cmd == TYPHOON_CMD_READ_MEDIA_STATUS) {
			typhoon_media_status(tp->dev, resp);
		} else if(resp->cmd == TYPHOON_CMD_HELLO_RESP) {
			typhoon_hello(tp);
		} else {
			printk(KERN_ERR "%s: dumping unexpected response "
			       "0x%04x:%d:0x%02x:0x%04x:%08x:%08x\n",
			       tp->name, le16_to_cpu(resp->cmd),
			       resp->numDesc, resp->flags,
			       le16_to_cpu(resp->parm1),
			       le32_to_cpu(resp->parm2),
			       le32_to_cpu(resp->parm3));
		}

cleanup:
		typhoon_inc_resp_index(&cleared, count);
	}

	indexes->respCleared = cpu_to_le32(cleared);
	wmb();
	return (resp_save == NULL);
}

static inline int
typhoon_num_free(int lastWrite, int lastRead, int ringSize)
{
	/* this works for all descriptors but rx_desc, as they are a
	 * different size than the cmd_desc -- everyone else is the same
	 */
	lastWrite /= sizeof(struct cmd_desc);
	lastRead /= sizeof(struct cmd_desc);
	return (ringSize + lastRead - lastWrite - 1) % ringSize;
}

static inline int
typhoon_num_free_cmd(struct typhoon *tp)
{
	int lastWrite = tp->cmdRing.lastWrite;
	int cmdCleared = le32_to_cpu(tp->indexes->cmdCleared);

	return typhoon_num_free(lastWrite, cmdCleared, COMMAND_ENTRIES);
}

static inline int
typhoon_num_free_resp(struct typhoon *tp)
{
	int respReady = le32_to_cpu(tp->indexes->respReady);
	int respCleared = le32_to_cpu(tp->indexes->respCleared);

	return typhoon_num_free(respReady, respCleared, RESPONSE_ENTRIES);
}

static inline int
typhoon_num_free_tx(struct transmit_ring *ring)
{
	/* if we start using the Hi Tx ring, this needs updating */
	return typhoon_num_free(ring->lastWrite, ring->lastRead, TXLO_ENTRIES);
}

static int
typhoon_issue_command(struct typhoon *tp, int num_cmd, struct cmd_desc *cmd,
		      int num_resp, struct resp_desc *resp)
{
	struct typhoon_indexes *indexes = tp->indexes;
	struct basic_ring *ring = &tp->cmdRing;
	struct resp_desc local_resp;
	int i, err = 0;
	int got_resp;
	int freeCmd, freeResp;
	int len, wrap_len;

	spin_lock(&tp->command_lock);

	freeCmd = typhoon_num_free_cmd(tp);
	freeResp = typhoon_num_free_resp(tp);

	if(freeCmd < num_cmd || freeResp < num_resp) {
		printk("%s: no descs for cmd, had (needed) %d (%d) cmd, "
			"%d (%d) resp\n", tp->name, freeCmd, num_cmd,
			freeResp, num_resp);
		err = -ENOMEM;
		goto out;
	}

	if(cmd->flags & TYPHOON_CMD_RESPOND) {
		/* If we're expecting a response, but the caller hasn't given
		 * us a place to put it, we'll provide one.
		 */
		tp->awaiting_resp = 1;
		if(resp == NULL) {
			resp = &local_resp;
			num_resp = 1;
		}
	}

	wrap_len = 0;
	len = num_cmd * sizeof(*cmd);
	if(unlikely(ring->lastWrite + len > COMMAND_RING_SIZE)) {
		wrap_len = ring->lastWrite + len - COMMAND_RING_SIZE;
		len = COMMAND_RING_SIZE - ring->lastWrite;
	}

	memcpy(ring->ringBase + ring->lastWrite, cmd, len);
	if(unlikely(wrap_len)) {
		struct cmd_desc *wrap_ptr = cmd;
		wrap_ptr += len / sizeof(*cmd);
		memcpy(ring->ringBase, wrap_ptr, wrap_len);
	}

	typhoon_inc_cmd_index(&ring->lastWrite, num_cmd);

	/* "I feel a presence... another warrior is on the the mesa."
	 */
	wmb();
	writel(ring->lastWrite, tp->ioaddr + TYPHOON_REG_CMD_READY);
	typhoon_post_pci_writes(tp->ioaddr);

	if((cmd->flags & TYPHOON_CMD_RESPOND) == 0)
		goto out;

	/* Ugh. We'll be here about 8ms, spinning our thumbs, unable to
	 * preempt or do anything other than take interrupts. So, don't
	 * wait for a response unless you have to.
	 *
	 * I've thought about trying to sleep here, but we're called
	 * from many contexts that don't allow that. Also, given the way
	 * 3Com has implemented irq coalescing, we would likely timeout --
	 * this has been observed in real life!
	 *
	 * The big killer is we have to wait to get stats from the card,
	 * though we could go to a periodic refresh of those if we don't
	 * mind them getting somewhat stale. The rest of the waiting
	 * commands occur during open/close/suspend/resume, so they aren't
	 * time critical. Creating SAs in the future will also have to
	 * wait here.
	 */
	got_resp = 0;
	for(i = 0; i < TYPHOON_WAIT_TIMEOUT && !got_resp; i++) {
		if(indexes->respCleared != indexes->respReady)
			got_resp = typhoon_process_response(tp, num_resp,
								resp);
		udelay(TYPHOON_UDELAY);
	}

	if(!got_resp) {
		err = -ETIMEDOUT;
		goto out;
	}

	/* Collect the error response even if we don't care about the
	 * rest of the response
	 */
	if(resp->flags & TYPHOON_RESP_ERROR)
		err = -EIO;

out:
	if(tp->awaiting_resp) {
		tp->awaiting_resp = 0;
		smp_wmb();

		/* Ugh. If a response was added to the ring between
		 * the call to typhoon_process_response() and the clearing
		 * of tp->awaiting_resp, we could have missed the interrupt
		 * and it could hang in the ring an indeterminate amount of
		 * time. So, check for it, and interrupt ourselves if this
		 * is the case.
		 */
		if(indexes->respCleared != indexes->respReady)
			writel(1, tp->ioaddr + TYPHOON_REG_SELF_INTERRUPT);
	}

	spin_unlock(&tp->command_lock);
	return err;
}

static void
typhoon_vlan_rx_register(struct net_device *dev, struct vlan_group *grp)
{
	struct typhoon *tp = (struct typhoon *) dev->priv;
	struct cmd_desc xp_cmd;
	int err;

	spin_lock_bh(&tp->state_lock);
	if(!tp->vlgrp != !grp) {
		/* We've either been turned on for the first time, or we've
		 * been turned off. Update the 3XP.
		 */
		if(grp)
			tp->offload |= TYPHOON_OFFLOAD_VLAN;
		else
			tp->offload &= ~TYPHOON_OFFLOAD_VLAN;

		/* If the interface is up, the runtime is running -- and we
		 * must be up for the vlan core to call us.
		 *
		 * Do the command outside of the spin lock, as it is slow.
		 */
		INIT_COMMAND_WITH_RESPONSE(&xp_cmd,
					TYPHOON_CMD_SET_OFFLOAD_TASKS);
		xp_cmd.parm2 = tp->offload;
		xp_cmd.parm3 = tp->offload;
		spin_unlock_bh(&tp->state_lock);
		err = typhoon_issue_command(tp, 1, &xp_cmd, 0, NULL);
		if(err < 0)
			printk("%s: vlan offload error %d\n", tp->name, -err);
		spin_lock_bh(&tp->state_lock);
	}

	/* now make the change visible */
	tp->vlgrp = grp;
	spin_unlock_bh(&tp->state_lock);
}

static void
typhoon_vlan_rx_kill_vid(struct net_device *dev, unsigned short vid)
{
	struct typhoon *tp = (struct typhoon *) dev->priv;
	spin_lock_bh(&tp->state_lock);
	if(tp->vlgrp)
		tp->vlgrp->vlan_devices[vid] = NULL;
	spin_unlock_bh(&tp->state_lock);
}

static inline void
typhoon_tso_fill(struct sk_buff *skb, struct transmit_ring *txRing,
			u32 ring_dma)
{
	struct tcpopt_desc *tcpd;
	u32 tcpd_offset = ring_dma;

	tcpd = (struct tcpopt_desc *) (txRing->ringBase + txRing->lastWrite);
	tcpd_offset += txRing->lastWrite;
	tcpd_offset += offsetof(struct tcpopt_desc, bytesTx);
	typhoon_inc_tx_index(&txRing->lastWrite, 1);

	tcpd->flags = TYPHOON_OPT_DESC | TYPHOON_OPT_TCP_SEG;
	tcpd->numDesc = 1;
	tcpd->mss_flags = cpu_to_le16(skb_tso_size(skb));
	tcpd->mss_flags |= TYPHOON_TSO_FIRST | TYPHOON_TSO_LAST;
	tcpd->respAddrLo = cpu_to_le32(tcpd_offset);
	tcpd->bytesTx = cpu_to_le32(skb->len);
	tcpd->status = 0;
}

static int
typhoon_start_tx(struct sk_buff *skb, struct net_device *dev)
{
	struct typhoon *tp = (struct typhoon *) dev->priv;
	struct transmit_ring *txRing;
	struct tx_desc *txd, *first_txd;
	dma_addr_t skb_dma;
	int numDesc;

	/* we have two rings to choose from, but we only use txLo for now
	 * If we start using the Hi ring as well, we'll need to update
	 * typhoon_stop_runtime(), typhoon_interrupt(), typhoon_num_free_tx(),
	 * and TXHI_ENTIRES to match, as well as update the TSO code below
	 * to get the right DMA address
	 */
	txRing = &tp->txLoRing;

	/* We need one descriptor for each fragment of the sk_buff, plus the
	 * one for the ->data area of it.
	 *
	 * The docs say a maximum of 16 fragment descriptors per TCP option
	 * descriptor, then make a new packet descriptor and option descriptor
	 * for the next 16 fragments. The engineers say just an option
	 * descriptor is needed. I've tested up to 26 fragments with a single
	 * packet descriptor/option descriptor combo, so I use that for now.
	 *
	 * If problems develop with TSO, check this first.
	 */
	numDesc = skb_shinfo(skb)->nr_frags + 1;
	if(skb_tso_size(skb))
		numDesc++;

	/* When checking for free space in the ring, we need to also
	 * account for the initial Tx descriptor, and we always must leave
	 * at least one descriptor unused in the ring so that it doesn't
	 * wrap and look empty.
	 *
	 * The only time we should loop here is when we hit the race
	 * between marking the queue awake and updating the cleared index.
	 * Just loop and it will appear. This comes from the acenic driver.
	 */
	while(unlikely(typhoon_num_free_tx(txRing) < (numDesc + 2)))
		smp_rmb();

	first_txd = (struct tx_desc *) (txRing->ringBase + txRing->lastWrite);
	typhoon_inc_tx_index(&txRing->lastWrite, 1);

	first_txd->flags = TYPHOON_TX_DESC | TYPHOON_DESC_VALID;
	first_txd->numDesc = 0;
	first_txd->len = 0;
	first_txd->addr = (u64)((unsigned long) skb) & 0xffffffff;
	first_txd->addrHi = (u64)((unsigned long) skb) >> 32;
	first_txd->processFlags = 0;

	if(skb->ip_summed == CHECKSUM_HW) {
		/* The 3XP will figure out if this is UDP/TCP */
		first_txd->processFlags |= TYPHOON_TX_PF_TCP_CHKSUM;
		first_txd->processFlags |= TYPHOON_TX_PF_UDP_CHKSUM;
		first_txd->processFlags |= TYPHOON_TX_PF_IP_CHKSUM;
	}

	if(vlan_tx_tag_present(skb)) {
		first_txd->processFlags |=
		    TYPHOON_TX_PF_INSERT_VLAN | TYPHOON_TX_PF_VLAN_PRIORITY;
		first_txd->processFlags |=
		    cpu_to_le32(htons(vlan_tx_tag_get(skb)) <<
				TYPHOON_TX_PF_VLAN_TAG_SHIFT);
	}

	if(skb_tso_size(skb)) {
		first_txd->processFlags |= TYPHOON_TX_PF_TCP_SEGMENT;
		first_txd->numDesc++;

		typhoon_tso_fill(skb, txRing, tp->txlo_dma_addr);
	}

	txd = (struct tx_desc *) (txRing->ringBase + txRing->lastWrite);
	typhoon_inc_tx_index(&txRing->lastWrite, 1);

	/* No need to worry about padding packet -- the firmware pads
	 * it with zeros to ETH_ZLEN for us.
	 */
	if(skb_shinfo(skb)->nr_frags == 0) {
		skb_dma = pci_map_single(tp->tx_pdev, skb->data, skb->len,
				       PCI_DMA_TODEVICE);
		txd->flags = TYPHOON_FRAG_DESC | TYPHOON_DESC_VALID;
		txd->len = cpu_to_le16(skb->len);
		txd->addr = cpu_to_le32(skb_dma);
		txd->addrHi = 0;
		first_txd->numDesc++;
	} else {
		int i, len;

		len = skb_headlen(skb);
		skb_dma = pci_map_single(tp->tx_pdev, skb->data, len,
				         PCI_DMA_TODEVICE);
		txd->flags = TYPHOON_FRAG_DESC | TYPHOON_DESC_VALID;
		txd->len = cpu_to_le16(len);
		txd->addr = cpu_to_le32(skb_dma);
		txd->addrHi = 0;
		first_txd->numDesc++;

		for(i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
			skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
			void *frag_addr;

			txd = (struct tx_desc *) (txRing->ringBase +
						txRing->lastWrite);
			typhoon_inc_tx_index(&txRing->lastWrite, 1);

			len = frag->size;
			frag_addr = (void *) page_address(frag->page) +
						frag->page_offset;
			skb_dma = pci_map_single(tp->tx_pdev, frag_addr, len,
					 PCI_DMA_TODEVICE);
			txd->flags = TYPHOON_FRAG_DESC | TYPHOON_DESC_VALID;
			txd->len = cpu_to_le16(len);
			txd->addr = cpu_to_le32(skb_dma);
			txd->addrHi = 0;
			first_txd->numDesc++;
		}
	}

	/* Kick the 3XP
	 */
	wmb();
	writel(txRing->lastWrite, tp->tx_ioaddr + txRing->writeRegister);

	dev->trans_start = jiffies;

	/* If we don't have room to put the worst case packet on the
	 * queue, then we must stop the queue. We need 2 extra
	 * descriptors -- one to prevent ring wrap, and one for the
	 * Tx header.
	 */
	numDesc = MAX_SKB_FRAGS + TSO_NUM_DESCRIPTORS + 1;

	if(typhoon_num_free_tx(txRing) < (numDesc + 2)) {
		netif_stop_queue(dev);

		/* A Tx complete IRQ could have gotten inbetween, making
		 * the ring free again. Only need to recheck here, since
		 * Tx is serialized.
		 */
		if(typhoon_num_free_tx(txRing) >= (numDesc + 2))
			netif_wake_queue(dev);
	}

	return 0;
}

static void
typhoon_set_rx_mode(struct net_device *dev)
{
	struct typhoon *tp = (struct typhoon *) dev->priv;
	struct cmd_desc xp_cmd;
	u32 mc_filter[2];
	u16 filter;

	filter = TYPHOON_RX_FILTER_DIRECTED | TYPHOON_RX_FILTER_BROADCAST;
	if(dev->flags & IFF_PROMISC) {
		printk(KERN_NOTICE "%s: Promiscuous mode enabled.\n",
		       dev->name);
		filter |= TYPHOON_RX_FILTER_PROMISCOUS;
	} else if((dev->mc_count > multicast_filter_limit) ||
		  (dev->flags & IFF_ALLMULTI)) {
		/* Too many to match, or accept all multicasts. */
		filter |= TYPHOON_RX_FILTER_ALL_MCAST;
	} else if(dev->mc_count) {
		struct dev_mc_list *mclist;