e100_main.c

Linux* Base Driver for the Intel(R) PRO/100 Family of Adapters

			       "Half" : "Full");

			e100_config_fc(bdp);
			e100_config(bdp);  

		} else {
			printk(KERN_ERR "e100: %s NIC Link is Down\n",
			       bdp->device->name);
		}
	}

	// toggle the tx queue according to link status
	// this also resolves a race condition between tx & non-cu cmd flows
	if (netif_carrier_ok(dev)) {
		if (netif_running(dev))
			netif_wake_queue(dev);
	} else {
		if (netif_running(dev))
			netif_stop_queue(dev);
		/* When changing to non-autoneg, device may lose  */
		/* link with some switches. e100 will try to      */
		/* revover link by sending command to PHY layer   */
		if (bdp->params.e100_speed_duplex != E100_AUTONEG)
			e100_force_speed_duplex_to_phy(bdp);
	}

	rmb();

	if (e100_update_stats(bdp)) {

		/* Check if a change in the IFS parameter is needed,
		   and configure the device accordingly */
		if (bdp->params.b_params & PRM_IFS)
			e100_manage_adaptive_ifs(bdp);

		/* Now adjust our dynamic tx threshold value */
		e100_refresh_txthld(bdp);

		/* Now if we are on a 557 and we havn't received any frames then we
		 * should issue a multicast command to reset the RU */
		if (bdp->rev_id < D101A4_REV_ID) {
			if (!(bdp->stats_counters->basic_stats.rcv_gd_frames)) {
				e100_set_multi(dev);
			}
		}
	}
	/* Issue command to dump statistics from device.        */
	/* Check for command completion on next watchdog timer. */
	e100_dump_stats_cntrs(bdp);

	wmb();

	/* relaunch watchdog timer in 2 sec */
	mod_timer(&(bdp->watchdog_timer), jiffies + (2 * HZ));

	if (list_empty(&bdp->active_rx_list))
		e100_trigger_SWI(bdp);
}

/**
 * e100_manage_adaptive_ifs
 * @bdp: atapter's private data struct
 *
 * This routine manages the adaptive Inter-Frame Spacing algorithm
 * using a state machine.
 */
void
e100_manage_adaptive_ifs(struct e100_private *bdp)
{
	static u16 state_table[9][4] = {	// rows are states
		{2, 0, 0, 0},	// state0   // column0: next state if increasing
		{2, 0, 5, 30},	// state1   // column1: next state if decreasing
		{5, 1, 5, 30},	// state2   // column2: IFS value for 100 mbit
		{5, 3, 0, 0},	// state3   // column3: IFS value for 10 mbit
		{5, 3, 10, 60},	// state4
		{8, 4, 10, 60},	// state5
		{8, 6, 0, 0},	// state6
		{8, 6, 20, 60},	// state7
		{8, 7, 20, 60}	// state8
	};

	u32 transmits =
		le32_to_cpu(bdp->stats_counters->basic_stats.xmt_gd_frames);
	u32 collisions =
		le32_to_cpu(bdp->stats_counters->basic_stats.xmt_ttl_coll);
	u32 state = bdp->ifs_state;
	u32 old_value = bdp->ifs_value;
	int next_col;
	u32 min_transmits;

	if (bdp->cur_dplx_mode == FULL_DUPLEX) {
		bdp->ifs_state = 0;
		bdp->ifs_value = 0;

	} else {		/* Half Duplex */
		/* Set speed specific parameters */
		if (bdp->cur_line_speed == 100) {
			next_col = 2;
			min_transmits = MIN_NUMBER_OF_TRANSMITS_100;

		} else {	/* 10 Mbps */
			next_col = 3;
			min_transmits = MIN_NUMBER_OF_TRANSMITS_10;
		}

		if ((transmits / 32 < collisions)
		    && (transmits > min_transmits)) {
			state = state_table[state][0];	/* increment */

		} else if (transmits < min_transmits) {
			state = state_table[state][1];	/* decrement */
		}

		bdp->ifs_value = state_table[state][next_col];
		bdp->ifs_state = state;
	}

	/* If the IFS value has changed, configure the device */
	if (bdp->ifs_value != old_value) {
		e100_config_ifs(bdp);
		e100_config(bdp);
	}
}

/**
 * e100intr - interrupt handler
 * @irq: the IRQ number
 * @dev_inst: the net_device struct
 * @regs: registers (unused)
 *
 * This routine is the ISR for the e100 board. It services
 * the RX & TX queues & starts the RU if it has stopped due
 * to no resources.
 */
irqreturn_t
e100intr(int irq, void *dev_inst, struct pt_regs *regs)
{
	struct net_device *dev;
	struct e100_private *bdp;
	u16 intr_status;

	dev = dev_inst;
	bdp = dev->priv;

	intr_status = readw(&bdp->scb->scb_status);
	/* If not my interrupt, just return */
	if (!(intr_status & SCB_STATUS_ACK_MASK) || (intr_status == 0xffff)) {
		return IRQ_NONE;
	}

	/* disable and ack intr */
	e100_disable_clear_intr(bdp);

	/* the device is closed, don't continue or else bad things may happen. */
	if (!netif_running(dev)) {
		e100_set_intr_mask(bdp);
		return IRQ_NONE;
	}

	/* SWI intr (triggered by watchdog) is signal to allocate new skb buffers */
	if (intr_status & SCB_STATUS_ACK_SWI) {
		e100_alloc_skbs(bdp);
	}

	/* do recv work if any */
	if (intr_status &
	    (SCB_STATUS_ACK_FR | SCB_STATUS_ACK_RNR | SCB_STATUS_ACK_SWI)) 
		bdp->drv_stats.rx_intr_pkts += e100_rx_srv(bdp);

	/* clean up after tx'ed packets */
	if (intr_status & (SCB_STATUS_ACK_CNA | SCB_STATUS_ACK_CX))
		e100_tx_srv(bdp);

	e100_set_intr_mask(bdp);
	return IRQ_HANDLED;
}

/**
 * e100_tx_skb_free - free TX skbs resources
 * @bdp: atapter's private data struct
 * @tcb: associated tcb of the freed skb
 *
 * This routine frees resources of TX skbs.
 */
static inline void
e100_tx_skb_free(struct e100_private *bdp, tcb_t *tcb)
{
	if (tcb->tcb_skb) {
#ifdef MAX_SKB_FRAGS
		int i;
		tbd_t *tbd_arr = tcb->tbd_ptr;
		int frags = skb_shinfo(tcb->tcb_skb)->nr_frags;

		for (i = 0; i <= frags; i++, tbd_arr++) {
			pci_unmap_single(bdp->pdev,
					 le32_to_cpu(tbd_arr->tbd_buf_addr),
					 le16_to_cpu(tbd_arr->tbd_buf_cnt),
					 PCI_DMA_TODEVICE);
		}
#else
		pci_unmap_single(bdp->pdev,
				 le32_to_cpu((tcb->tbd_ptr)->tbd_buf_addr),
				 tcb->tcb_skb->len, PCI_DMA_TODEVICE);
#endif
		dev_kfree_skb_irq(tcb->tcb_skb);
		tcb->tcb_skb = NULL;
	}
}

/**
 * e100_tx_srv - service TX queues
 * @bdp: atapter's private data struct
 *
 * This routine services the TX queues. It reclaims the TCB's & TBD's & other
 * resources used during the transmit of this buffer. It is called from the ISR.
 * We don't need a tx_lock since we always access buffers which were already
 * prepared.
 */
void
e100_tx_srv(struct e100_private *bdp)
{
	tcb_t *tcb;
	int i;

	/* go over at most TxDescriptors buffers */
	for (i = 0; i < bdp->params.TxDescriptors; i++) {
		tcb = bdp->tcb_pool.data;
		tcb += bdp->tcb_pool.head;

		rmb();

		/* if the buffer at 'head' is not complete, break */
		if (!(tcb->tcb_hdr.cb_status &
		      __constant_cpu_to_le16(CB_STATUS_COMPLETE)))
			break;

		/* service next buffer, clear the out of resource condition */
		e100_tx_skb_free(bdp, tcb);

		if (netif_running(bdp->device))
			netif_wake_queue(bdp->device);

		/* if we've caught up with 'tail', break */
		if (NEXT_TCB_TOUSE(bdp->tcb_pool.head) == bdp->tcb_pool.tail) {
			break;
		}

		bdp->tcb_pool.head = NEXT_TCB_TOUSE(bdp->tcb_pool.head);
	}
}

/**
 * e100_rx_srv - service RX queue
 * @bdp: atapter's private data struct
 * @max_number_of_rfds: max number of RFDs to process
 * @rx_congestion: flag pointer, to inform the calling function of congestion.
 *
 * This routine processes the RX interrupt & services the RX queues.
 * For each successful RFD, it allocates a new msg block, links that
 * into the RFD list, and sends the old msg upstream.
 * The new RFD is then put at the end of the free list of RFD's.
 * It returns the number of serviced RFDs.
 */
u32
e100_rx_srv(struct e100_private *bdp)
{
	rfd_t *rfd;		/* new rfd, received rfd */
	int i;
	u16 rfd_status;
	struct sk_buff *skb;
	struct net_device *dev;
	unsigned int data_sz;
	struct rx_list_elem *rx_struct;
	u32 rfd_cnt = 0;

	dev = bdp->device;

	/* current design of rx is as following:
	 * 1. socket buffer (skb) used to pass network packet to upper layer
	 * 2. all HW host memory structures (like RFDs, RBDs and data buffers)
	 *    are placed in a skb's data room
	 * 3. when rx process is complete, we change skb internal pointers to exclude
	 *    from data area all unrelated things (RFD, RDB) and to leave
	 *    just rx'ed packet netto
	 * 4. for each skb passed to upper layer, new one is allocated instead.
	 * 5. if no skb left, in 2 sec another atempt to allocate skbs will be made
	 *    (watchdog trigger SWI intr and isr should allocate new skbs)
	 */
	for (i = 0; i < bdp->params.RxDescriptors; i++) {
		if (list_empty(&(bdp->active_rx_list))) {
			break;
		}

		rx_struct = list_entry(bdp->active_rx_list.next,
				       struct rx_list_elem, list_elem);
		skb = rx_struct->skb;

		rfd = RFD_POINTER(skb, bdp);	/* locate RFD within skb */

		// sync only the RFD header
		pci_dma_sync_single(bdp->pdev, rx_struct->dma_addr,
				    bdp->rfd_size, PCI_DMA_FROMDEVICE);
		rfd_status = le16_to_cpu(rfd->rfd_header.cb_status);	/* get RFD's status */
		if (!(rfd_status & RFD_STATUS_COMPLETE))	/* does not contains data yet - exit */
			break;

		/* to allow manipulation with current skb we need to unlink it */
		list_del(&(rx_struct->list_elem));

		/* do not free & unmap badly received packet.
		 * move it to the end of skb list for reuse */
		if (!(rfd_status & RFD_STATUS_OK)) {
			e100_add_skb_to_end(bdp, rx_struct);
			continue;
		}

		data_sz = min_t(u16, (le16_to_cpu(rfd->rfd_act_cnt) & 0x3fff),
				(sizeof (rfd_t) - bdp->rfd_size));

		/* now sync all the data */
		pci_dma_sync_single(bdp->pdev, rx_struct->dma_addr,
				    (data_sz + bdp->rfd_size),
				    PCI_DMA_FROMDEVICE);

		pci_unmap_single(bdp->pdev, rx_struct->dma_addr,
				 sizeof (rfd_t), PCI_DMA_FROMDEVICE);

		list_add(&(rx_struct->list_elem), &(bdp->rx_struct_pool));

		/* end of dma access to rfd */
		bdp->skb_req++;	/* incr number of requested skbs */
		e100_alloc_skbs(bdp);	/* and get them */

		/* set packet size, excluding checksum (2 last bytes) if it is present */
		if ((bdp->flags & DF_CSUM_OFFLOAD)
		    && (bdp->rev_id < D102_REV_ID))
			skb_put(skb, (int) data_sz - 2);
		else
			skb_put(skb, (int) data_sz);

		/* set the protocol */
		skb->protocol = eth_type_trans(skb, dev);

		/* set the checksum info */
		if (bdp->flags & DF_CSUM_OFFLOAD) {
			if (bdp->rev_id >= D102_REV_ID) {
				skb->ip_summed = e100_D102_check_checksum(rfd);
			} else {
				skb->ip_summed = e100_D101M_checksum(bdp, skb);
			}
		} else {
			skb->ip_summed = CHECKSUM_NONE;
		}
#ifdef E100_IA64_DMA_FIX
		//Free low-memory skb buffer without passing it up to the IP stack
		if (non_DMA32_memory_present) {
			skb_linearize(skb, GFP_ATOMIC);
		}
#endif

		bdp->drv_stats.net_stats.rx_bytes += skb->len;

#ifdef NETIF_F_HW_VLAN_TX
		if(bdp->vlgrp && (rfd_status & CB_STATUS_VLAN)) {
			vlan_hwaccel_rx(skb, bdp->vlgrp, be16_to_cpu(rfd->vlanid));
		} else {
			netif_rx(skb);
		}
#else
		netif_rx(skb);
#endif
		dev->last_rx = jiffies;
		
		rfd_cnt++;
	}			/* end of rfd loop */

	/* restart the RU if it has stopped */
	if ((readw(&bdp->scb->scb_status) & SCB_RUS_MASK) != SCB_RUS_READY) {
		e100_start_ru(bdp);
	}

	return rfd_cnt;
}

void
e100_refresh_txthld(struct e100_private *bdp)
{
	basic_cntr_t *pstat = &(bdp->stats_counters->basic_stats);

	/* as long as tx_per_underrun is not 0, we can go about dynamically *
	 * adjusting the xmit threshold. we stop doing that & resort to defaults
	 * * once the adjustments become meaningless. the value is adjusted by *
	 * dumping the error counters & checking the # of xmit underrun errors *
	 * we've had. */
	if (bdp->tx_per_underrun) {
		/* We are going to last values dumped from the dump statistics
		 * command */
		if (le32_to_cpu(pstat->xmt_gd_frames)) {
			if (le32_to_cpu(pstat->xmt_uruns)) {
				/* 
				 * if we have had more than one underrun per "DEFAULT #
				 * OF XMITS ALLOWED PER UNDERRUN" good xmits, raise the
				 * THRESHOLD.
				 */
				if ((le32_to_cpu(pstat->xmt_gd_frames) /
				     le32_to_cpu(pstat->xmt_uruns)) <
				    bdp->tx_per_underrun) {
					bdp->tx_thld += 3;
				}
			}

			/* 
			 * if we've had less than one underrun per the DEFAULT number of
			 * of good xmits allowed, lower the THOLD but not less than 0 
			 */
			if (le32_to_cpu(pstat->xmt_gd_frames) >
			    bdp->tx_per_underrun) {
				bdp->tx_thld--;

				if (bdp->tx_thld < 6)
					bdp->tx_thld = 6;

			}
		}

		/* end good xmits */
		/* 
		 * * if our adjustments are becoming unresonable, stop adjusting &
		 * resort * to defaults & pray. A THOLD value > 190 means that the
		 * adapter will * wait for 190*8=1520 bytes in TX FIFO before it
		 * starts xmit. Since * MTU is 1514, it doesn't make any sense for
		 * further increase. */
		if (bdp->tx_thld >= 190) {
			bdp->tx_per_underrun = 0;
			bdp->tx_thld = 189;
		}
	}			/* end underrun check */
}

/**
 * e100_prepare_xmit_buff - prepare a buffer for transmission
 * @bdp: atapter's private data struct
 * @skb: skb to send
 *
 * This routine prepare a buffer for transmission. It checks
 * the message length for the appropiate size. It picks up a
 * free tcb from the TCB pool and sets up the corresponding
 * TBD's. If the number of fragments are more than the number
 * of TBD/TCB it copies all the fragments in a coalesce buffer.
 * It returns a pointer to the prepared TCB.
 */
static inline tcb_t *
e100_prepare_xmit_buff(struct e100_private *bdp, struct sk_buff *skb)
{
	tcb_t *tcb, *prev_tcb;

	tcb = bdp->t