Message-Id: <200707142138.36619.mb@bu3sch.de>
Date:	Sat, 14 Jul 2007 21:38:36 +0200
From:	Michael Buesch <mb@...sch.de>
To:	bryan.wu@...log.com
Cc:	Jeff Garzik <jeff@...zik.org>, LKML <linux-kernel@...r.kernel.org>,
	netdev@...r.kernel.org, Andrew Morton <akpm@...ux-foundation.org>
Subject: Re: [PATCH 1/3] Blackfin ethernet driver: on chip ethernet MAC controller driver

On Saturday 14 July 2007 20:49:53 Bryan Wu wrote:
> +#if defined(CONFIG_BFIN_MAC_USE_L1)
> +# define bfin_mac_alloc(dma_handle, size)  l1_data_sram_zalloc(size)
> +# define bfin_mac_free(dma_handle, ptr)    l1_data_sram_free(ptr)
> +#else
> +# define bfin_mac_alloc(dma_handle, size) \
> +	 dma_alloc_coherent(NULL, size, dma_handle, GFP_DMA)

You shouldn't be using GFP_DMA here (that's only for legacy ISA-DMA
memory), but one of GFP_KERNEL or GFP_ATOMIC.

> +# define bfin_mac_free(dma_handle, ptr) \
> +	dma_free_coherent(NULL, sizeof(*ptr), ptr, dma_handle)
> +#endif
> +

> +/* pointers to maintain transmit list */
> +static struct net_dma_desc_tx *tx_list_head;
> +static struct net_dma_desc_tx *tx_list_tail;
> +static struct net_dma_desc_rx *rx_list_head;
> +static struct net_dma_desc_rx *rx_list_tail;
> +static struct net_dma_desc_rx *current_rx_ptr;
> +static struct net_dma_desc_tx *current_tx_ptr;
> +static struct net_dma_desc_tx *tx_desc;
> +static struct net_dma_desc_rx *rx_desc;

Hm, from this I guess only one of these devices is possible
at a time in the system?

> +static int desc_list_init(void)
> +{
> +	struct net_dma_desc_tx *tmp_desc_tx;
> +	struct net_dma_desc_rx *tmp_desc_rx;
> +	int i;
> +	struct sk_buff *new_skb;
> +#if !defined(CONFIG_BFIN_MAC_USE_L1)
> +	dma_addr_t dma_handle;
> +#endif
> +
> +	tx_desc =
> +	    bfin_mac_alloc(&dma_handle,
> +			   sizeof(struct net_dma_desc_tx) *
> +			   CONFIG_BFIN_TX_DESC_NUM);
> +	if (tx_desc == NULL)
> +		goto init_error;
> +
> +	rx_desc =
> +	    bfin_mac_alloc(&dma_handle,
> +			   sizeof(struct net_dma_desc_rx) *
> +			   CONFIG_BFIN_RX_DESC_NUM);

You are overriding and throwing dma_handle away, here.
You might need it later for freeing. (and maybe for other
purposes, too, like writing the address to a device register?)

> +	if (rx_desc == NULL)
> +		goto init_error;
> +
> +	/* init tx_list */
> +	for (i = 0; i < CONFIG_BFIN_TX_DESC_NUM; i++) {
> +
> +		tmp_desc_tx = tx_desc + i;
> +
> +		if (i == 0) {
> +			tx_list_head = tmp_desc_tx;
> +			tx_list_tail = tmp_desc_tx;
> +		}
> +
> +		tmp_desc_tx->desc_a.start_addr =
> +		    (unsigned long)tmp_desc_tx->packet;
> +		tmp_desc_tx->desc_a.x_count = 0;
> +		/* disabled */
> +		tmp_desc_tx->desc_a.config.b_DMA_EN = 0;
> +		/* read from memory */
> +		tmp_desc_tx->desc_a.config.b_WNR = 0;
> +		/* wordsize is 32 bits */
> +		tmp_desc_tx->desc_a.config.b_WDSIZE = 2;
> +		/* 6 half words is desc size. */
> +		tmp_desc_tx->desc_a.config.b_NDSIZE = 6;
> +		/* large desc flow */
> +		tmp_desc_tx->desc_a.config.b_FLOW = 7;
> +		tmp_desc_tx->desc_a.next_dma_desc = &(tmp_desc_tx->desc_b);
> +
> +		tmp_desc_tx->desc_b.start_addr =
> +		    (unsigned long)(&(tmp_desc_tx->status));
> +		tmp_desc_tx->desc_b.x_count = 0;
> +		/* enabled */
> +		tmp_desc_tx->desc_b.config.b_DMA_EN = 1;
> +		/* write to memory */
> +		tmp_desc_tx->desc_b.config.b_WNR = 1;
> +		/* wordsize is 32 bits */
> +		tmp_desc_tx->desc_b.config.b_WDSIZE = 2;
> +		/* disable interrupt */
> +		tmp_desc_tx->desc_b.config.b_DI_EN = 0;
> +		tmp_desc_tx->desc_b.config.b_NDSIZE = 6;
> +		/* stop mode */
> +		tmp_desc_tx->desc_b.config.b_FLOW = 7;
> +		tmp_desc_tx->skb = NULL;
> +		tx_list_tail->desc_b.next_dma_desc = &(tmp_desc_tx->desc_a);
> +		tx_list_tail->next = tmp_desc_tx;
> +
> +		tx_list_tail = tmp_desc_tx;
> +	}
> +	tx_list_tail->next = tx_list_head;	/* tx_list is a circle */
> +	tx_list_tail->desc_b.next_dma_desc = &(tx_list_head->desc_a);
> +	current_tx_ptr = tx_list_head;
> +
> +	/* init rx_list */
> +	for (i = 0; i < CONFIG_BFIN_RX_DESC_NUM; i++) {
> +
> +		tmp_desc_rx = rx_desc + i;
> +
> +		if (i == 0) {
> +			rx_list_head = tmp_desc_rx;
> +			rx_list_tail = tmp_desc_rx;
> +		}
> +
> +		/* allocat a new skb for next time receive */
> +		new_skb = dev_alloc_skb(PKT_BUF_SZ + 2);
> +		if (!new_skb) {
> +			printk(KERN_NOTICE CARDNAME
> +			       ": init: low on mem - packet dropped\n");
> +			goto init_error;
> +		}
> +		skb_reserve(new_skb, 2);
> +		tmp_desc_rx->skb = new_skb;
> +		/* since RXDWA is enabled */
> +		tmp_desc_rx->desc_a.start_addr =
> +		    (unsigned long)new_skb->data - 2;
> +		tmp_desc_rx->desc_a.x_count = 0;
> +		/* enabled */
> +		tmp_desc_rx->desc_a.config.b_DMA_EN = 1;
> +		/* Write to memory */
> +		tmp_desc_rx->desc_a.config.b_WNR = 1;
> +		/* wordsize is 32 bits */
> +		tmp_desc_rx->desc_a.config.b_WDSIZE = 2;
> +		/* 6 half words is desc size. */
> +		tmp_desc_rx->desc_a.config.b_NDSIZE = 6;
> +		/* large desc flow */
> +		tmp_desc_rx->desc_a.config.b_FLOW = 7;
> +		tmp_desc_rx->desc_a.next_dma_desc = &(tmp_desc_rx->desc_b);
> +
> +		tmp_desc_rx->desc_b.start_addr =
> +		    (unsigned long)(&(tmp_desc_rx->status));
> +		tmp_desc_rx->desc_b.x_count = 0;
> +		/* enabled */
> +		tmp_desc_rx->desc_b.config.b_DMA_EN = 1;
> +		/* Write to memory */
> +		tmp_desc_rx->desc_b.config.b_WNR = 1;
> +		/* wordsize is 32 bits */
> +		tmp_desc_rx->desc_b.config.b_WDSIZE = 2;
> +		tmp_desc_rx->desc_b.config.b_NDSIZE = 6;
> +		/* enable interrupt */
> +		tmp_desc_rx->desc_b.config.b_DI_EN = 1;
> +		/* large mode */
> +		tmp_desc_rx->desc_b.config.b_FLOW = 7;
> +		rx_list_tail->desc_b.next_dma_desc = &(tmp_desc_rx->desc_a);
> +
> +		rx_list_tail->next = tmp_desc_rx;
> +		rx_list_tail = tmp_desc_rx;
> +	}
> +	rx_list_tail->next = rx_list_head;	/* rx_list is a circle */
> +	rx_list_tail->desc_b.next_dma_desc = &(rx_list_head->desc_a);
> +	current_rx_ptr = rx_list_head;
> +
> +	return 0;
> +
> +init_error:
> +	desc_list_free();
> +	printk(KERN_ERR CARDNAME ": kmalloc failed\n");
> +	return -ENOMEM;
> +}
> +
> +static void desc_list_free(void)
> +{
> +	struct net_dma_desc_rx *tmp_desc_rx;
> +	struct net_dma_desc_tx *tmp_desc_tx;
> +	int i;
> +#if !defined(CONFIG_BFIN_MAC_USE_L1)
> +	dma_addr_t dma_handle = 0;
> +#endif
> +
> +	if (tx_desc != NULL) {
> +		tmp_desc_tx = tx_list_head;
> +		for (i = 0; i < CONFIG_BFIN_TX_DESC_NUM; i++) {
> +			if (tmp_desc_tx != NULL) {
> +				if (tmp_desc_tx->skb) {
> +					dev_kfree_skb(tmp_desc_tx->skb);
> +					tmp_desc_tx->skb = NULL;
> +				}
> +
> +			}
> +			tmp_desc_tx = tmp_desc_tx->next;
> +		}
> +		bfin_mac_free(dma_handle, tx_desc);

dma_handle will always be 0 here. I guess that's a bug.
Also see my comment on the alloc code, above.

> +	}
> +
> +	if (rx_desc != NULL) {
> +		tmp_desc_rx = rx_list_head;
> +		for (i = 0; i < CONFIG_BFIN_RX_DESC_NUM; i++) {
> +			if (tmp_desc_rx != NULL) {
> +				if (tmp_desc_rx->skb) {
> +					dev_kfree_skb(tmp_desc_rx->skb);
> +					tmp_desc_rx->skb = NULL;
> +				}
> +			}
> +			tmp_desc_rx = tmp_desc_rx->next;
> +		}
> +		bfin_mac_free(dma_handle, rx_desc);

Same here.

> +/* Wait until the previous MDC/MDIO transaction has completed */
> +static void poll_mdc_done(void)
> +{
> +	/* poll the STABUSY bit */
> +	while ((bfin_read_EMAC_STAADD()) & STABUSY) {
> +	};

Please add a timeout here, as this could loop forever in case of
slightly faulty hardware (or firmware, or whatever).
It's good practice to add a timeout of say 5 seconds to such loops.
You also might want to sleep between checks (dunno if this is called
in atomic, though), so an mdelay(10) or something might be desired.



> +/* set up the phy */
> +static void bf537mac_setphy(struct net_device *dev)
> +{
> +	u16 phydat;
> +	u32 sysctl;
> +	struct bf537mac_local *lp = netdev_priv(dev);
> +
> +	pr_debug("start settting up phy\n");

settting :)

> +	/* Program PHY registers */
> +	phydat = 0;

No need to zero phydat out, here.

> +	/* issue a reset */
> +	raw_write_phy_reg(lp->PhyAddr, PHYREG_MODECTL, 0x8000);
> +
> +	/* wait half a second */
> +	udelay(500);

First:  500usec != 1/2sec
Second: 500usec is a lot. You might want to use msleep(1)
        or msleep(500), if you _really_ want to sleep half a sec.

> +	phydat = read_phy_reg(lp->PhyAddr, PHYREG_MODECTL);
> +
> +	/* advertise flow control supported */
> +	phydat = read_phy_reg(lp->PhyAddr, PHYREG_ANAR);
> +	phydat |= (1 << 10);
> +	write_phy_reg(lp->PhyAddr, PHYREG_ANAR, phydat);
> +
> +	phydat = 0;
> +	if (lp->Negotiate) {
> +		phydat |= 0x1000;	/* enable auto negotiation */
> +	} else {
> +		if (lp->FullDuplex) {
> +			phydat |= (1 << 8);	/* full duplex */
> +		} else {
> +			phydat &= (~(1 << 8));	/* half duplex */
> +		}
> +		if (!lp->Port10) {
> +			phydat |= (1 << 13);	/* 100 Mbps */
> +		} else {
> +			phydat &= (~(1 << 13));	/* 10 Mbps */
> +		}
> +	}
> +
> +	if (lp->Loopback) {
> +		phydat |= (1 << 14);	/* enable TX->RX loopback */
> +#ifdef DRV_DEBUG
> +		write_phy_reg(lp->PhyAddr, PHYREG_MODECTL, phydat);
> +#endif
> +	}
> +
> +	write_phy_reg(lp->PhyAddr, PHYREG_MODECTL, phydat);
> +	udelay(500);

See my comment above, about the delay vs sleep.

> +	phydat = read_phy_reg(lp->PhyAddr, PHYREG_MODECTL);
> +	/* check for SMSC PHY */
> +	if ((read_phy_reg(lp->PhyAddr, PHYREG_PHYID1) == 0x7) &&
> +	((read_phy_reg(lp->PhyAddr, PHYREG_PHYID2) & 0xfff0) == 0xC0A0)) {

Hm, lots of magic values in this function. You should probably
use #define for better readability.

> +		/*
> +		 * we have SMSC PHY so reqest interrupt
> +		 * on link down condition
> +		 */
> +
> +		/* enable interrupts */
> +		write_phy_reg(lp->PhyAddr, 30, 0x0ff);

-
> +		/* enable PHY_INT */
> +		sysctl = bfin_read_EMAC_SYSCTL();
> +		sysctl |= 0x1;
> +#ifdef DRV_DEBUG

Seems like you can move this #ifdef three lines up.

> +		bfin_write_EMAC_SYSCTL(sysctl);
> +#endif
> +	}
> +}
> +
> +/**************************************************************************/
> +void setup_system_regs(struct net_device *dev)
> +{
> +	int PHYADDR;

Lowercase, please.

> +	unsigned short sysctl, phydat;
> +	u32 opmode;
> +	struct bf537mac_local *lp = netdev_priv(dev);
> +	int count = 0;
> +
> +	PHYADDR = lp->PhyAddr;
> +
> +	/* Enable PHY output */
> +	if (!(bfin_read_VR_CTL() & PHYCLKOE))
> +		bfin_write_VR_CTL(bfin_read_VR_CTL() | PHYCLKOE);
> +
> +	/* MDC  = 2.5 MHz */
> +	sysctl = SET_MDCDIV(24);
> +	/* Odd word alignment for Receive Frame DMA word */
> +	/* Configure checksum support and rcve frame word alignment */
> +#if defined(BFIN_MAC_CSUM_OFFLOAD)
> +	sysctl |= RXDWA | RXCKS;
> +#else
> +	sysctl |= RXDWA;
> +#endif
> +	bfin_write_EMAC_SYSCTL(sysctl);
> +	/* auto negotiation on  */
> +	/* full duplex          */
> +	/* 100 Mbps             */
> +	phydat = PHY_ANEG_EN | PHY_DUPLEX | PHY_SPD_SET;
> +	write_phy_reg(PHYADDR, PHYREG_MODECTL, phydat);
> +
> +	/* test if full duplex supported */
> +	do {
> +		msleep(100);
> +		phydat = read_phy_reg(PHYADDR, PHYREG_MODESTAT);
> +		if (count > 30) {
> +			printk(KERN_NOTICE CARDNAME ": Link is down\n");
> +			printk(KERN_NOTICE CARDNAME
> +				 "please check your network connection\n");
> +			break;
> +		}
> +		count++;
> +	} while (!(phydat & 0x0004));
> +
> +	phydat = read_phy_reg(PHYADDR, PHYREG_ANLPAR);
> +
> +	if ((phydat & 0x0100) || (phydat & 0x0040)) {
> +		opmode = FDMODE;
> +	} else {
> +		opmode = 0;
> +		printk(KERN_INFO CARDNAME
> +			": Network is set to half duplex\n");
> +	}
> +
> +#if defined(CONFIG_BFIN_MAC_RMII)
> +	opmode |= RMII; /* For Now only 100MBit are supported */
> +#endif
> +
> +	bfin_write_EMAC_OPMODE(opmode);
> +
> +#ifdef DRV_DEBUG
> +	bfin_write_EMAC_MMC_CTL(RSTC | CROLL | MMCE);
> +#endif
> +	bfin_write_EMAC_MMC_CTL(RSTC | CROLL);

Is it desired to overwrite it here, or is this a missing #else?

> +void setup_mac_addr(u8 * mac_addr)
> +{
> +	/* this depends on a little-endian machine */
> +	bfin_write_EMAC_ADDRLO(*(u32 *) & mac_addr[0]);
> +	bfin_write_EMAC_ADDRHI(*(u16 *) & mac_addr[4]);

So why not properly code it with leXX_to_cpu?
It's for free, if it's not needed on your platform.

> +}
> +
> +static void adjust_tx_list(void)
> +{
> +	if (tx_list_head->status.status_word != 0
> +	    && current_tx_ptr != tx_list_head) {
> +		goto adjust_head;	/* released something, just return; */
> +	}
> +
> +	/*
> +	 * if nothing released, check wait condition
> +	 * current's next can not be the head,
> +	 * otherwise the dma will not stop as we want
> +	 */
> +	if (current_tx_ptr->next->next == tx_list_head) {
> +		while (tx_list_head->status.status_word == 0) {
> +			udelay(10);
> +			if (tx_list_head->status.status_word != 0
> +			    || !(bfin_read_DMA2_IRQ_STATUS() & 0x08)) {
> +				goto adjust_head;
> +			}

Timeout?

> +		}
> +		if (tx_list_head->status.status_word != 0) {
> +			goto adjust_head;
> +		}
> +	}

To me it looks like you are accessing DMA memory without any
DMA syncing here. Maybe that's not needed on your platform
and you can access DMA memory directly there, but it's ugly
and code that properly uses the syncing functions is _much_
easier to understand.

> +	return;
> +
> +adjust_head:
> +	do {
> +		tx_list_head->desc_a.config.b_DMA_EN = 0;
> +		tx_list_head->status.status_word = 0;
> +		if (tx_list_head->skb) {
> +			dev_kfree_skb(tx_list_head->skb);
> +			tx_list_head->skb = NULL;
> +		} else {
> +			printk(KERN_ERR CARDNAME
> +			       ": no sk_buff in a transmitted frame!\n");
> +		}
> +		tx_list_head = tx_list_head->next;
> +	} while (tx_list_head->status.status_word != 0
> +		 && current_tx_ptr != tx_list_head);
> +	return;
> +
> +}
> +

> +static void bf537mac_rx(struct net_device *dev)
> +{
> +	struct sk_buff *skb, *new_skb;
> +	struct bf537mac_local *lp = netdev_priv(dev);
> +	unsigned short len;
> +
> +	/* allocat a new skb for next time receive */

Typo, "allocat"

> +	skb = current_rx_ptr->skb;
> +	new_skb = dev_alloc_skb(PKT_BUF_SZ + 2);
> +	if (!new_skb) {
> +		printk(KERN_NOTICE CARDNAME
> +		       ": rx: low on mem - packet dropped\n");
> +		lp->stats.rx_dropped++;
> +		goto out;
> +	}
> +	/* reserve 2 bytes for RXDWA padding */
> +	skb_reserve(new_skb, 2);
> +	current_rx_ptr->skb = new_skb;
> +	current_rx_ptr->desc_a.start_addr = (unsigned long)new_skb->data - 2;
> +
> +#ifdef DRV_DEBUG
> +	int i;

This generates a GCC warning.

> +	if (len >= 64) {
> +		for (i = 0; i < len; i++) {
> +			printk(KERN_DEBUG "%.2x-", ((unsigned char *)pkt)[i]);

Does this even compile? I don't see where "pkt" is defined.

> +			if (((i % 8) == 0) && (i != 0))
> +				printk(KERN_DEBUG "\n");
> +		}
> +	printk(KERN_DEBUG"\n");

One more tab indention and a space between KERN_DEBUG and "\n", please.

> +static void bf537mac_reset(void)
> +{
> +	unsigned int opmode;
> +
> +	opmode = bfin_read_EMAC_OPMODE();
> +	opmode &= (~RE);
> +	opmode &= (~TE);
> +	/* Turn off the EMAC */
> +	bfin_write_EMAC_OPMODE(bfin_read_EMAC_OPMODE() & opmode);

bfin_write_EMAC_OPMODE(opmode);
would have the same effect, no?

> +static int __init bf537mac_probe(struct net_device *dev)
> +{
> +	struct bf537mac_local *lp = netdev_priv(dev);
> +	int retval;
> +
> +	/* Grab the MAC address in the MAC */
> +	*(u32 *) (&(dev->dev_addr[0])) = bfin_read_EMAC_ADDRLO();
> +	*(u16 *) (&(dev->dev_addr[4])) = (u16) bfin_read_EMAC_ADDRHI();

Endianess broken.

> +	setup_mac_addr(dev->dev_addr);
> +
> +	/* Fill in the fields of the device structure with ethernet values. */
> +	ether_setup(dev);
> +
> +	dev->open = bf537mac_open;
> +	dev->stop = bf537mac_close;
> +	dev->hard_start_xmit = bf537mac_hard_start_xmit;
> +	dev->tx_timeout = bf537mac_timeout;
> +	dev->get_stats = bf537mac_query_statistics;
> +	dev->set_multicast_list = bf537mac_set_multicast_list;
> +#ifdef DVR_DEBUG

Typo, "DVR".
But why only use ethtool when debugging?

> +	dev->ethtool_ops = &bf537mac_ethtool_ops;
> +#endif
> +#ifdef CONFIG_NET_POLL_CONTROLLER
> +	dev->poll_controller = bf537mac_poll;
> +#endif

> +	/* Enable PHY output early */
> +	if (!(bfin_read_VR_CTL() & PHYCLKOE))
> +		bfin_write_VR_CTL(bfin_read_VR_CTL() | PHYCLKOE);
> +
> +	retval = register_netdev(dev);
> +	if (retval == 0) {
> +		/* now, print out the card info, in a short format.. */
> +		printk(KERN_INFO "Blackfin mac net device registered\n");
> +	}

Unwind the allocations above, if registering fails.

> +
> +err_out:
> +	return retval;
> +}


> +struct dma_config_reg {
> +	unsigned short b_DMA_EN:1;  /* Bit 0 : DMA Enable */
> +	unsigned short b_WNR:1;     /* Bit 1 : DMA Direction */
> +	unsigned short b_WDSIZE:2;  /* Bit 2 & 3 : DMA Tranfer Word size */
> +	unsigned short b_DMA2D:1;   /* Bit 4 : DMA Mode 2D or 1D */
> +	unsigned short b_RESTART:1; /* Bit 5 : Retain the FIFO */
> +	unsigned short b_DI_SEL:1;  /* Bit 6 : Data Interrupt Timing Select */
> +	unsigned short b_DI_EN:1;   /* Bit 7 : Data Interrupt Enable */
> +	unsigned short b_NDSIZE:4;  /* Bit 8 to 11 : Flex descriptor Size */
> +	unsigned short b_FLOW:3;    /* Bit 12 to 14 : FLOW */
> +};

This is most likely not endianess safe.

-- 
Greetings Michael.
-
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Powered by blists - more mailing lists