blob: d2d4f47c7e28477d605980870266f16121fa374a [file] [log] [blame]
// SPDX-License-Identifier: GPL-2.0+
/* cassini.c: Sun Microsystems Cassini(+) ethernet driver.
*
* Copyright (C) 2004 Sun Microsystems Inc.
* Copyright (C) 2003 Adrian Sun (asun@darksunrising.com)
*
* This driver uses the sungem driver (c) David Miller
* (davem@redhat.com) as its basis.
*
* The cassini chip has a number of features that distinguish it from
* the gem chip:
* 4 transmit descriptor rings that are used for either QoS (VLAN) or
* load balancing (non-VLAN mode)
* batching of multiple packets
* multiple CPU dispatching
* page-based RX descriptor engine with separate completion rings
* Gigabit support (GMII and PCS interface)
* MIF link up/down detection works
*
* RX is handled by page sized buffers that are attached as fragments to
* the skb. here's what's done:
* -- driver allocates pages at a time and keeps reference counts
* on them.
* -- the upper protocol layers assume that the header is in the skb
* itself. as a result, cassini will copy a small amount (64 bytes)
* to make them happy.
* -- driver appends the rest of the data pages as frags to skbuffs
* and increments the reference count
* -- on page reclamation, the driver swaps the page with a spare page.
* if that page is still in use, it frees its reference to that page,
* and allocates a new page for use. otherwise, it just recycles the
* the page.
*
* NOTE: cassini can parse the header. however, it's not worth it
* as long as the network stack requires a header copy.
*
* TX has 4 queues. currently these queues are used in a round-robin
* fashion for load balancing. They can also be used for QoS. for that
* to work, however, QoS information needs to be exposed down to the driver
* level so that subqueues get targeted to particular transmit rings.
* alternatively, the queues can be configured via use of the all-purpose
* ioctl.
*
* RX DATA: the rx completion ring has all the info, but the rx desc
* ring has all of the data. RX can conceivably come in under multiple
* interrupts, but the INT# assignment needs to be set up properly by
* the BIOS and conveyed to the driver. PCI BIOSes don't know how to do
* that. also, the two descriptor rings are designed to distinguish between
* encrypted and non-encrypted packets, but we use them for buffering
* instead.
*
* by default, the selective clear mask is set up to process rx packets.
*/
#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
#include <linux/module.h>
#include <linux/kernel.h>
#include <linux/types.h>
#include <linux/compiler.h>
#include <linux/slab.h>
#include <linux/delay.h>
#include <linux/init.h>
#include <linux/interrupt.h>
#include <linux/vmalloc.h>
#include <linux/ioport.h>
#include <linux/pci.h>
#include <linux/mm.h>
#include <linux/highmem.h>
#include <linux/list.h>
#include <linux/dma-mapping.h>
#include <linux/netdevice.h>
#include <linux/etherdevice.h>
#include <linux/skbuff.h>
#include <linux/ethtool.h>
#include <linux/crc32.h>
#include <linux/random.h>
#include <linux/mii.h>
#include <linux/ip.h>
#include <linux/tcp.h>
#include <linux/mutex.h>
#include <linux/firmware.h>
#include <net/checksum.h>
#include <linux/atomic.h>
#include <asm/io.h>
#include <asm/byteorder.h>
#include <linux/uaccess.h>
#define cas_page_map(x) kmap_atomic((x))
#define cas_page_unmap(x) kunmap_atomic((x))
#define CAS_NCPUS num_online_cpus()
#define cas_skb_release(x) netif_rx(x)
/* select which firmware to use */
#define USE_HP_WORKAROUND
#define HP_WORKAROUND_DEFAULT /* select which firmware to use as default */
#define CAS_HP_ALT_FIRMWARE cas_prog_null /* alternate firmware */
#include "cassini.h"
#define USE_TX_COMPWB /* use completion writeback registers */
#define USE_CSMA_CD_PROTO /* standard CSMA/CD */
#define USE_RX_BLANK /* hw interrupt mitigation */
#undef USE_ENTROPY_DEV /* don't test for entropy device */
/* NOTE: these aren't useable unless PCI interrupts can be assigned.
* also, we need to make cp->lock finer-grained.
*/
#undef USE_PCI_INTB
#undef USE_PCI_INTC
#undef USE_PCI_INTD
#undef USE_QOS
#undef USE_VPD_DEBUG /* debug vpd information if defined */
/* rx processing options */
#define USE_PAGE_ORDER /* specify to allocate large rx pages */
#define RX_DONT_BATCH 0 /* if 1, don't batch flows */
#define RX_COPY_ALWAYS 0 /* if 0, use frags */
#define RX_COPY_MIN 64 /* copy a little to make upper layers happy */
#undef RX_COUNT_BUFFERS /* define to calculate RX buffer stats */
#define DRV_MODULE_NAME "cassini"
#define DRV_MODULE_VERSION "1.6"
#define DRV_MODULE_RELDATE "21 May 2008"
#define CAS_DEF_MSG_ENABLE \
(NETIF_MSG_DRV | \
NETIF_MSG_PROBE | \
NETIF_MSG_LINK | \
NETIF_MSG_TIMER | \
NETIF_MSG_IFDOWN | \
NETIF_MSG_IFUP | \
NETIF_MSG_RX_ERR | \
NETIF_MSG_TX_ERR)
/* length of time before we decide the hardware is borked,
* and dev->tx_timeout() should be called to fix the problem
*/
#define CAS_TX_TIMEOUT (HZ)
#define CAS_LINK_TIMEOUT (22*HZ/10)
#define CAS_LINK_FAST_TIMEOUT (1)
/* timeout values for state changing. these specify the number
* of 10us delays to be used before giving up.
*/
#define STOP_TRIES_PHY 1000
#define STOP_TRIES 5000
/* specify a minimum frame size to deal with some fifo issues
* max mtu == 2 * page size - ethernet header - 64 - swivel =
* 2 * page_size - 0x50
*/
#define CAS_MIN_FRAME 97
#define CAS_1000MB_MIN_FRAME 255
#define CAS_MIN_MTU 60
#define CAS_MAX_MTU min(((cp->page_size << 1) - 0x50), 9000)
#if 1
/*
* Eliminate these and use separate atomic counters for each, to
* avoid a race condition.
*/
#else
#define CAS_RESET_MTU 1
#define CAS_RESET_ALL 2
#define CAS_RESET_SPARE 3
#endif
static char version[] =
DRV_MODULE_NAME ".c:v" DRV_MODULE_VERSION " (" DRV_MODULE_RELDATE ")\n";
static int cassini_debug = -1; /* -1 == use CAS_DEF_MSG_ENABLE as value */
static int link_mode;
MODULE_AUTHOR("Adrian Sun (asun@darksunrising.com)");
MODULE_DESCRIPTION("Sun Cassini(+) ethernet driver");
MODULE_LICENSE("GPL");
MODULE_FIRMWARE("sun/cassini.bin");
module_param(cassini_debug, int, 0);
MODULE_PARM_DESC(cassini_debug, "Cassini bitmapped debugging message enable value");
module_param(link_mode, int, 0);
MODULE_PARM_DESC(link_mode, "default link mode");
/*
* Work around for a PCS bug in which the link goes down due to the chip
* being confused and never showing a link status of "up."
*/
#define DEFAULT_LINKDOWN_TIMEOUT 5
/*
* Value in seconds, for user input.
*/
static int linkdown_timeout = DEFAULT_LINKDOWN_TIMEOUT;
module_param(linkdown_timeout, int, 0);
MODULE_PARM_DESC(linkdown_timeout,
"min reset interval in sec. for PCS linkdown issue; disabled if not positive");
/*
* value in 'ticks' (units used by jiffies). Set when we init the
* module because 'HZ' in actually a function call on some flavors of
* Linux. This will default to DEFAULT_LINKDOWN_TIMEOUT * HZ.
*/
static int link_transition_timeout;
static u16 link_modes[] = {
BMCR_ANENABLE, /* 0 : autoneg */
0, /* 1 : 10bt half duplex */
BMCR_SPEED100, /* 2 : 100bt half duplex */
BMCR_FULLDPLX, /* 3 : 10bt full duplex */
BMCR_SPEED100|BMCR_FULLDPLX, /* 4 : 100bt full duplex */
CAS_BMCR_SPEED1000|BMCR_FULLDPLX /* 5 : 1000bt full duplex */
};
static const struct pci_device_id cas_pci_tbl[] = {
{ PCI_VENDOR_ID_SUN, PCI_DEVICE_ID_SUN_CASSINI,
PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0UL },
{ PCI_VENDOR_ID_NS, PCI_DEVICE_ID_NS_SATURN,
PCI_ANY_ID, PCI_ANY_ID, 0, 0, 0UL },
{ 0, }
};
MODULE_DEVICE_TABLE(pci, cas_pci_tbl);
static void cas_set_link_modes(struct cas *cp);
static inline void cas_lock_tx(struct cas *cp)
{
int i;
for (i = 0; i < N_TX_RINGS; i++)
spin_lock_nested(&cp->tx_lock[i], i);
}
/* WTZ: QA was finding deadlock problems with the previous
* versions after long test runs with multiple cards per machine.
* See if replacing cas_lock_all with safer versions helps. The
* symptoms QA is reporting match those we'd expect if interrupts
* aren't being properly restored, and we fixed a previous deadlock
* with similar symptoms by using save/restore versions in other
* places.
*/
#define cas_lock_all_save(cp, flags) \
do { \
struct cas *xxxcp = (cp); \
spin_lock_irqsave(&xxxcp->lock, flags); \
cas_lock_tx(xxxcp); \
} while (0)
static inline void cas_unlock_tx(struct cas *cp)
{
int i;
for (i = N_TX_RINGS; i > 0; i--)
spin_unlock(&cp->tx_lock[i - 1]);
}
#define cas_unlock_all_restore(cp, flags) \
do { \
struct cas *xxxcp = (cp); \
cas_unlock_tx(xxxcp); \
spin_unlock_irqrestore(&xxxcp->lock, flags); \
} while (0)
static void cas_disable_irq(struct cas *cp, const int ring)
{
/* Make sure we won't get any more interrupts */
if (ring == 0) {
writel(0xFFFFFFFF, cp->regs + REG_INTR_MASK);
return;
}
/* disable completion interrupts and selectively mask */
if (cp->cas_flags & CAS_FLAG_REG_PLUS) {
switch (ring) {
#if defined (USE_PCI_INTB) || defined(USE_PCI_INTC) || defined(USE_PCI_INTD)
#ifdef USE_PCI_INTB
case 1:
#endif
#ifdef USE_PCI_INTC
case 2:
#endif
#ifdef USE_PCI_INTD
case 3:
#endif
writel(INTRN_MASK_CLEAR_ALL | INTRN_MASK_RX_EN,
cp->regs + REG_PLUS_INTRN_MASK(ring));
break;
#endif
default:
writel(INTRN_MASK_CLEAR_ALL, cp->regs +
REG_PLUS_INTRN_MASK(ring));
break;
}
}
}
static inline void cas_mask_intr(struct cas *cp)
{
int i;
for (i = 0; i < N_RX_COMP_RINGS; i++)
cas_disable_irq(cp, i);
}
static void cas_enable_irq(struct cas *cp, const int ring)
{
if (ring == 0) { /* all but TX_DONE */
writel(INTR_TX_DONE, cp->regs + REG_INTR_MASK);
return;
}
if (cp->cas_flags & CAS_FLAG_REG_PLUS) {
switch (ring) {
#if defined (USE_PCI_INTB) || defined(USE_PCI_INTC) || defined(USE_PCI_INTD)
#ifdef USE_PCI_INTB
case 1:
#endif
#ifdef USE_PCI_INTC
case 2:
#endif
#ifdef USE_PCI_INTD
case 3:
#endif
writel(INTRN_MASK_RX_EN, cp->regs +
REG_PLUS_INTRN_MASK(ring));
break;
#endif
default:
break;
}
}
}
static inline void cas_unmask_intr(struct cas *cp)
{
int i;
for (i = 0; i < N_RX_COMP_RINGS; i++)
cas_enable_irq(cp, i);
}
static inline void cas_entropy_gather(struct cas *cp)
{
#ifdef USE_ENTROPY_DEV
if ((cp->cas_flags & CAS_FLAG_ENTROPY_DEV) == 0)
return;
batch_entropy_store(readl(cp->regs + REG_ENTROPY_IV),
readl(cp->regs + REG_ENTROPY_IV),
sizeof(uint64_t)*8);
#endif
}
static inline void cas_entropy_reset(struct cas *cp)
{
#ifdef USE_ENTROPY_DEV
if ((cp->cas_flags & CAS_FLAG_ENTROPY_DEV) == 0)
return;
writel(BIM_LOCAL_DEV_PAD | BIM_LOCAL_DEV_PROM | BIM_LOCAL_DEV_EXT,
cp->regs + REG_BIM_LOCAL_DEV_EN);
writeb(ENTROPY_RESET_STC_MODE, cp->regs + REG_ENTROPY_RESET);
writeb(0x55, cp->regs + REG_ENTROPY_RAND_REG);
/* if we read back 0x0, we don't have an entropy device */
if (readb(cp->regs + REG_ENTROPY_RAND_REG) == 0)
cp->cas_flags &= ~CAS_FLAG_ENTROPY_DEV;
#endif
}
/* access to the phy. the following assumes that we've initialized the MIF to
* be in frame rather than bit-bang mode
*/
static u16 cas_phy_read(struct cas *cp, int reg)
{
u32 cmd;
int limit = STOP_TRIES_PHY;
cmd = MIF_FRAME_ST | MIF_FRAME_OP_READ;
cmd |= CAS_BASE(MIF_FRAME_PHY_ADDR, cp->phy_addr);
cmd |= CAS_BASE(MIF_FRAME_REG_ADDR, reg);
cmd |= MIF_FRAME_TURN_AROUND_MSB;
writel(cmd, cp->regs + REG_MIF_FRAME);
/* poll for completion */
while (limit-- > 0) {
udelay(10);
cmd = readl(cp->regs + REG_MIF_FRAME);
if (cmd & MIF_FRAME_TURN_AROUND_LSB)
return cmd & MIF_FRAME_DATA_MASK;
}
return 0xFFFF; /* -1 */
}
static int cas_phy_write(struct cas *cp, int reg, u16 val)
{
int limit = STOP_TRIES_PHY;
u32 cmd;
cmd = MIF_FRAME_ST | MIF_FRAME_OP_WRITE;
cmd |= CAS_BASE(MIF_FRAME_PHY_ADDR, cp->phy_addr);
cmd |= CAS_BASE(MIF_FRAME_REG_ADDR, reg);
cmd |= MIF_FRAME_TURN_AROUND_MSB;
cmd |= val & MIF_FRAME_DATA_MASK;
writel(cmd, cp->regs + REG_MIF_FRAME);
/* poll for completion */
while (limit-- > 0) {
udelay(10);
cmd = readl(cp->regs + REG_MIF_FRAME);
if (cmd & MIF_FRAME_TURN_AROUND_LSB)
return 0;
}
return -1;
}
static void cas_phy_powerup(struct cas *cp)
{
u16 ctl = cas_phy_read(cp, MII_BMCR);
if ((ctl & BMCR_PDOWN) == 0)
return;
ctl &= ~BMCR_PDOWN;
cas_phy_write(cp, MII_BMCR, ctl);
}
static void cas_phy_powerdown(struct cas *cp)
{
u16 ctl = cas_phy_read(cp, MII_BMCR);
if (ctl & BMCR_PDOWN)
return;
ctl |= BMCR_PDOWN;
cas_phy_write(cp, MII_BMCR, ctl);
}
/* cp->lock held. note: the last put_page will free the buffer */
static int cas_page_free(struct cas *cp, cas_page_t *page)
{
dma_unmap_page(&cp->pdev->dev, page->dma_addr, cp->page_size,
DMA_FROM_DEVICE);
__free_pages(page->buffer, cp->page_order);
kfree(page);
return 0;
}
#ifdef RX_COUNT_BUFFERS
#define RX_USED_ADD(x, y) ((x)->used += (y))
#define RX_USED_SET(x, y) ((x)->used = (y))
#else
#define RX_USED_ADD(x, y) do { } while(0)
#define RX_USED_SET(x, y) do { } while(0)
#endif
/* local page allocation routines for the receive buffers. jumbo pages
* require at least 8K contiguous and 8K aligned buffers.
*/
static cas_page_t *cas_page_alloc(struct cas *cp, const gfp_t flags)
{
cas_page_t *page;
page = kmalloc(sizeof(cas_page_t), flags);
if (!page)
return NULL;
INIT_LIST_HEAD(&page->list);
RX_USED_SET(page, 0);
page->buffer = alloc_pages(flags, cp->page_order);
if (!page->buffer)
goto page_err;
page->dma_addr = dma_map_page(&cp->pdev->dev, page->buffer, 0,
cp->page_size, DMA_FROM_DEVICE);
return page;
page_err:
kfree(page);
return NULL;
}
/* initialize spare pool of rx buffers, but allocate during the open */
static void cas_spare_init(struct cas *cp)
{
spin_lock(&cp->rx_inuse_lock);
INIT_LIST_HEAD(&cp->rx_inuse_list);
spin_unlock(&cp->rx_inuse_lock);
spin_lock(&cp->rx_spare_lock);
INIT_LIST_HEAD(&cp->rx_spare_list);
cp->rx_spares_needed = RX_SPARE_COUNT;
spin_unlock(&cp->rx_spare_lock);
}
/* used on close. free all the spare buffers. */
static void cas_spare_free(struct cas *cp)
{
struct list_head list, *elem, *tmp;
/* free spare buffers */
INIT_LIST_HEAD(&list);
spin_lock(&cp->rx_spare_lock);
list_splice_init(&cp->rx_spare_list, &list);
spin_unlock(&cp->rx_spare_lock);
list_for_each_safe(elem, tmp, &list) {
cas_page_free(cp, list_entry(elem, cas_page_t, list));
}
INIT_LIST_HEAD(&list);
#if 1
/*
* Looks like Adrian had protected this with a different
* lock than used everywhere else to manipulate this list.
*/
spin_lock(&cp->rx_inuse_lock);
list_splice_init(&cp->rx_inuse_list, &list);
spin_unlock(&cp->rx_inuse_lock);
#else
spin_lock(&cp->rx_spare_lock);
list_splice_init(&cp->rx_inuse_list, &list);
spin_unlock(&cp->rx_spare_lock);
#endif
list_for_each_safe(elem, tmp, &list) {
cas_page_free(cp, list_entry(elem, cas_page_t, list));
}
}
/* replenish spares if needed */
static void cas_spare_recover(struct cas *cp, const gfp_t flags)
{
struct list_head list, *elem, *tmp;
int needed, i;
/* check inuse list. if we don't need any more free buffers,
* just free it
*/
/* make a local copy of the list */
INIT_LIST_HEAD(&list);
spin_lock(&cp->rx_inuse_lock);
list_splice_init(&cp->rx_inuse_list, &list);
spin_unlock(&cp->rx_inuse_lock);
list_for_each_safe(elem, tmp, &list) {
cas_page_t *page = list_entry(elem, cas_page_t, list);
/*
* With the lockless pagecache, cassini buffering scheme gets
* slightly less accurate: we might find that a page has an
* elevated reference count here, due to a speculative ref,
* and skip it as in-use. Ideally we would be able to reclaim
* it. However this would be such a rare case, it doesn't
* matter too much as we should pick it up the next time round.
*
* Importantly, if we find that the page has a refcount of 1
* here (our refcount), then we know it is definitely not inuse
* so we can reuse it.
*/
if (page_count(page->buffer) > 1)
continue;
list_del(elem);
spin_lock(&cp->rx_spare_lock);
if (cp->rx_spares_needed > 0) {
list_add(elem, &cp->rx_spare_list);
cp->rx_spares_needed--;
spin_unlock(&cp->rx_spare_lock);
} else {
spin_unlock(&cp->rx_spare_lock);
cas_page_free(cp, page);
}
}
/* put any inuse buffers back on the list */
if (!list_empty(&list)) {
spin_lock(&cp->rx_inuse_lock);
list_splice(&list, &cp->rx_inuse_list);
spin_unlock(&cp->rx_inuse_lock);
}
spin_lock(&cp->rx_spare_lock);
needed = cp->rx_spares_needed;
spin_unlock(&cp->rx_spare_lock);
if (!needed)
return;
/* we still need spares, so try to allocate some */
INIT_LIST_HEAD(&list);
i = 0;
while (i < needed) {
cas_page_t *spare = cas_page_alloc(cp, flags);
if (!spare)
break;
list_add(&spare->list, &list);
i++;
}
spin_lock(&cp->rx_spare_lock);
list_splice(&list, &cp->rx_spare_list);
cp->rx_spares_needed -= i;
spin_unlock(&cp->rx_spare_lock);
}
/* pull a page from the list. */
static cas_page_t *cas_page_dequeue(struct cas *cp)
{
struct list_head *entry;
int recover;
spin_lock(&cp->rx_spare_lock);
if (list_empty(&cp->rx_spare_list)) {
/* try to do a quick recovery */
spin_unlock(&cp->rx_spare_lock);
cas_spare_recover(cp, GFP_ATOMIC);
spin_lock(&cp->rx_spare_lock);
if (list_empty(&cp->rx_spare_list)) {
netif_err(cp, rx_err, cp->dev,
"no spare buffers available\n");
spin_unlock(&cp->rx_spare_lock);
return NULL;
}
}
entry = cp->rx_spare_list.next;
list_del(entry);
recover = ++cp->rx_spares_needed;
spin_unlock(&cp->rx_spare_lock);
/* trigger the timer to do the recovery */
if ((recover & (RX_SPARE_RECOVER_VAL - 1)) == 0) {
#if 1
atomic_inc(&cp->reset_task_pending);
atomic_inc(&cp->reset_task_pending_spare);
schedule_work(&cp->reset_task);
#else
atomic_set(&cp->reset_task_pending, CAS_RESET_SPARE);
schedule_work(&cp->reset_task);
#endif
}
return list_entry(entry, cas_page_t, list);
}
static void cas_mif_poll(struct cas *cp, const int enable)
{
u32 cfg;
cfg = readl(cp->regs + REG_MIF_CFG);
cfg &= (MIF_CFG_MDIO_0 | MIF_CFG_MDIO_1);
if (cp->phy_type & CAS_PHY_MII_MDIO1)
cfg |= MIF_CFG_PHY_SELECT;
/* poll and interrupt on link status change. */
if (enable) {
cfg |= MIF_CFG_POLL_EN;
cfg |= CAS_BASE(MIF_CFG_POLL_REG, MII_BMSR);
cfg |= CAS_BASE(MIF_CFG_POLL_PHY, cp->phy_addr);
}
writel((enable) ? ~(BMSR_LSTATUS | BMSR_ANEGCOMPLETE) : 0xFFFF,
cp->regs + REG_MIF_MASK);
writel(cfg, cp->regs + REG_MIF_CFG);
}
/* Must be invoked under cp->lock */
static void cas_begin_auto_negotiation(struct cas *cp,
const struct ethtool_link_ksettings *ep)
{
u16 ctl;
#if 1
int lcntl;
int changed = 0;
int oldstate = cp->lstate;
int link_was_not_down = !(oldstate == link_down);
#endif
/* Setup link parameters */
if (!ep)
goto start_aneg;
lcntl = cp->link_cntl;
if (ep->base.autoneg == AUTONEG_ENABLE) {
cp->link_cntl = BMCR_ANENABLE;
} else {
u32 speed = ep->base.speed;
cp->link_cntl = 0;
if (speed == SPEED_100)
cp->link_cntl |= BMCR_SPEED100;
else if (speed == SPEED_1000)
cp->link_cntl |= CAS_BMCR_SPEED1000;
if (ep->base.duplex == DUPLEX_FULL)
cp->link_cntl |= BMCR_FULLDPLX;
}
#if 1
changed = (lcntl != cp->link_cntl);
#endif
start_aneg:
if (cp->lstate == link_up) {
netdev_info(cp->dev, "PCS link down\n");
} else {
if (changed) {
netdev_info(cp->dev, "link configuration changed\n");
}
}
cp->lstate = link_down;
cp->link_transition = LINK_TRANSITION_LINK_DOWN;
if (!cp->hw_running)
return;
#if 1
/*
* WTZ: If the old state was link_up, we turn off the carrier
* to replicate everything we do elsewhere on a link-down
* event when we were already in a link-up state..
*/
if (oldstate == link_up)
netif_carrier_off(cp->dev);
if (changed && link_was_not_down) {
/*
* WTZ: This branch will simply schedule a full reset after
* we explicitly changed link modes in an ioctl. See if this
* fixes the link-problems we were having for forced mode.
*/
atomic_inc(&cp->reset_task_pending);
atomic_inc(&cp->reset_task_pending_all);
schedule_work(&cp->reset_task);
cp->timer_ticks = 0;
mod_timer(&cp->link_timer, jiffies + CAS_LINK_TIMEOUT);
return;
}
#endif
if (cp->phy_type & CAS_PHY_SERDES) {
u32 val = readl(cp->regs + REG_PCS_MII_CTRL);
if (cp->link_cntl & BMCR_ANENABLE) {
val |= (PCS_MII_RESTART_AUTONEG | PCS_MII_AUTONEG_EN);
cp->lstate = link_aneg;
} else {
if (cp->link_cntl & BMCR_FULLDPLX)
val |= PCS_MII_CTRL_DUPLEX;
val &= ~PCS_MII_AUTONEG_EN;
cp->lstate = link_force_ok;
}
cp->link_transition = LINK_TRANSITION_LINK_CONFIG;
writel(val, cp->regs + REG_PCS_MII_CTRL);
} else {
cas_mif_poll(cp, 0);
ctl = cas_phy_read(cp, MII_BMCR);
ctl &= ~(BMCR_FULLDPLX | BMCR_SPEED100 |
CAS_BMCR_SPEED1000 | BMCR_ANENABLE);
ctl |= cp->link_cntl;
if (ctl & BMCR_ANENABLE) {
ctl |= BMCR_ANRESTART;
cp->lstate = link_aneg;
} else {
cp->lstate = link_force_ok;
}
cp->link_transition = LINK_TRANSITION_LINK_CONFIG;
cas_phy_write(cp, MII_BMCR, ctl);
cas_mif_poll(cp, 1);
}
cp->timer_ticks = 0;
mod_timer(&cp->link_timer, jiffies + CAS_LINK_TIMEOUT);
}
/* Must be invoked under cp->lock. */
static int cas_reset_mii_phy(struct cas *cp)
{
int limit = STOP_TRIES_PHY;
u16 val;
cas_phy_write(cp, MII_BMCR, BMCR_RESET);
udelay(100);
while (--limit) {
val = cas_phy_read(cp, MII_BMCR);
if ((val & BMCR_RESET) == 0)
break;
udelay(10);
}
return limit <= 0;
}
static void cas_saturn_firmware_init(struct cas *cp)
{
const struct firmware *fw;
const char fw_name[] = "sun/cassini.bin";
int err;
if (PHY_NS_DP83065 != cp->phy_id)
return;
err = request_firmware(&fw, fw_name, &cp->pdev->dev);
if (err) {
pr_err("Failed to load firmware \"%s\"\n",
fw_name);
return;
}
if (fw->size < 2) {
pr_err("bogus length %zu in \"%s\"\n",
fw->size, fw_name);
goto out;
}
cp->fw_load_addr= fw->data[1] << 8 | fw->data[0];
cp->fw_size = fw->size - 2;
cp->fw_data = vmalloc(cp->fw_size);
if (!cp->fw_data)
goto out;
memcpy(cp->fw_data, &fw->data[2], cp->fw_size);
out:
release_firmware(fw);
}
static void cas_saturn_firmware_load(struct cas *cp)
{
int i;
if (!cp->fw_data)
return;
cas_phy_powerdown(cp);
/* expanded memory access mode */
cas_phy_write(cp, DP83065_MII_MEM, 0x0);
/* pointer configuration for new firmware */
cas_phy_write(cp, DP83065_MII_REGE, 0x8ff9);
cas_phy_write(cp, DP83065_MII_REGD, 0xbd);
cas_phy_write(cp, DP83065_MII_REGE, 0x8ffa);
cas_phy_write(cp, DP83065_MII_REGD, 0x82);
cas_phy_write(cp, DP83065_MII_REGE, 0x8ffb);
cas_phy_write(cp, DP83065_MII_REGD, 0x0);
cas_phy_write(cp, DP83065_MII_REGE, 0x8ffc);
cas_phy_write(cp, DP83065_MII_REGD, 0x39);
/* download new firmware */
cas_phy_write(cp, DP83065_MII_MEM, 0x1);
cas_phy_write(cp, DP83065_MII_REGE, cp->fw_load_addr);
for (i = 0; i < cp->fw_size; i++)
cas_phy_write(cp, DP83065_MII_REGD, cp->fw_data[i]);
/* enable firmware */
cas_phy_write(cp, DP83065_MII_REGE, 0x8ff8);
cas_phy_write(cp, DP83065_MII_REGD, 0x1);
}
/* phy initialization */
static void cas_phy_init(struct cas *cp)
{
u16 val;
/* if we're in MII/GMII mode, set up phy */
if (CAS_PHY_MII(cp->phy_type)) {
writel(PCS_DATAPATH_MODE_MII,
cp->regs + REG_PCS_DATAPATH_MODE);
cas_mif_poll(cp, 0);
cas_reset_mii_phy(cp); /* take out of isolate mode */
if (PHY_LUCENT_B0 == cp->phy_id) {
/* workaround link up/down issue with lucent */
cas_phy_write(cp, LUCENT_MII_REG, 0x8000);
cas_phy_write(cp, MII_BMCR, 0x00f1);
cas_phy_write(cp, LUCENT_MII_REG, 0x0);
} else if (PHY_BROADCOM_B0 == (cp->phy_id & 0xFFFFFFFC)) {
/* workarounds for broadcom phy */
cas_phy_write(cp, BROADCOM_MII_REG8, 0x0C20);
cas_phy_write(cp, BROADCOM_MII_REG7, 0x0012);
cas_phy_write(cp, BROADCOM_MII_REG5, 0x1804);
cas_phy_write(cp, BROADCOM_MII_REG7, 0x0013);
cas_phy_write(cp, BROADCOM_MII_REG5, 0x1204);
cas_phy_write(cp, BROADCOM_MII_REG7, 0x8006);
cas_phy_write(cp, BROADCOM_MII_REG5, 0x0132);
cas_phy_write(cp, BROADCOM_MII_REG7, 0x8006);
cas_phy_write(cp, BROADCOM_MII_REG5, 0x0232);
cas_phy_write(cp, BROADCOM_MII_REG7, 0x201F);
cas_phy_write(cp, BROADCOM_MII_REG5, 0x0A20);
} else if (PHY_BROADCOM_5411 == cp->phy_id) {
val = cas_phy_read(cp, BROADCOM_MII_REG4);
val = cas_phy_read(cp, BROADCOM_MII_REG4);
if (val & 0x0080) {
/* link workaround */
cas_phy_write(cp, BROADCOM_MII_REG4,
val & ~0x0080);
}
} else if (cp->cas_flags & CAS_FLAG_SATURN) {
writel((cp->phy_type & CAS_PHY_MII_MDIO0) ?
SATURN_PCFG_FSI : 0x0,
cp->regs + REG_SATURN_PCFG);
/* load firmware to address 10Mbps auto-negotiation
* issue. NOTE: this will need to be changed if the
* default firmware gets fixed.
*/
if (PHY_NS_DP83065 == cp->phy_id) {
cas_saturn_firmware_load(cp);
}
cas_phy_powerup(cp);
}
/* advertise capabilities */
val = cas_phy_read(cp, MII_BMCR);
val &= ~BMCR_ANENABLE;
cas_phy_write(cp, MII_BMCR, val);
udelay(10);
cas_phy_write(cp, MII_ADVERTISE,
cas_phy_read(cp, MII_ADVERTISE) |
(ADVERTISE_10HALF | ADVERTISE_10FULL |
ADVERTISE_100HALF | ADVERTISE_100FULL |
CAS_ADVERTISE_PAUSE |
CAS_ADVERTISE_ASYM_PAUSE));
if (cp->cas_flags & CAS_FLAG_1000MB_CAP) {
/* make sure that we don't advertise half
* duplex to avoid a chip issue
*/
val = cas_phy_read(cp, CAS_MII_1000_CTRL);
val &= ~CAS_ADVERTISE_1000HALF;
val |= CAS_ADVERTISE_1000FULL;
cas_phy_write(cp, CAS_MII_1000_CTRL, val);
}
} else {
/* reset pcs for serdes */
u32 val;
int limit;
writel(PCS_DATAPATH_MODE_SERDES,
cp->regs + REG_PCS_DATAPATH_MODE);
/* enable serdes pins on saturn */
if (cp->cas_flags & CAS_FLAG_SATURN)
writel(0, cp->regs + REG_SATURN_PCFG);
/* Reset PCS unit. */
val = readl(cp->regs + REG_PCS_MII_CTRL);
val |= PCS_MII_RESET;
writel(val, cp->regs + REG_PCS_MII_CTRL);
limit = STOP_TRIES;
while (--limit > 0) {
udelay(10);
if ((readl(cp->regs + REG_PCS_MII_CTRL) &
PCS_MII_RESET) == 0)
break;
}
if (limit <= 0)
netdev_warn(cp->dev, "PCS reset bit would not clear [%08x]\n",
readl(cp->regs + REG_PCS_STATE_MACHINE));
/* Make sure PCS is disabled while changing advertisement
* configuration.
*/
writel(0x0, cp->regs + REG_PCS_CFG);
/* Advertise all capabilities except half-duplex. */
val = readl(cp->regs + REG_PCS_MII_ADVERT);
val &= ~PCS_MII_ADVERT_HD;
val |= (PCS_MII_ADVERT_FD | PCS_MII_ADVERT_SYM_PAUSE |
PCS_MII_ADVERT_ASYM_PAUSE);
writel(val, cp->regs + REG_PCS_MII_ADVERT);
/* enable PCS */
writel(PCS_CFG_EN, cp->regs + REG_PCS_CFG);
/* pcs workaround: enable sync detect */
writel(PCS_SERDES_CTRL_SYNCD_EN,
cp->regs + REG_PCS_SERDES_CTRL);
}
}
static int cas_pcs_link_check(struct cas *cp)
{
u32 stat, state_machine;
int retval = 0;
/* The link status bit latches on zero, so you must
* read it twice in such a case to see a transition
* to the link being up.
*/
stat = readl(cp->regs + REG_PCS_MII_STATUS);
if ((stat & PCS_MII_STATUS_LINK_STATUS) == 0)
stat = readl(cp->regs + REG_PCS_MII_STATUS);
/* The remote-fault indication is only valid
* when autoneg has completed.
*/
if ((stat & (PCS_MII_STATUS_AUTONEG_COMP |
PCS_MII_STATUS_REMOTE_FAULT)) ==
(PCS_MII_STATUS_AUTONEG_COMP | PCS_MII_STATUS_REMOTE_FAULT))
netif_info(cp, link, cp->dev, "PCS RemoteFault\n");
/* work around link detection issue by querying the PCS state
* machine directly.
*/
state_machine = readl(cp->regs + REG_PCS_STATE_MACHINE);
if ((state_machine & PCS_SM_LINK_STATE_MASK) != SM_LINK_STATE_UP) {
stat &= ~PCS_MII_STATUS_LINK_STATUS;
} else if (state_machine & PCS_SM_WORD_SYNC_STATE_MASK) {
stat |= PCS_MII_STATUS_LINK_STATUS;
}
if (stat & PCS_MII_STATUS_LINK_STATUS) {
if (cp->lstate != link_up) {
if (cp->opened) {
cp->lstate = link_up;
cp->link_transition = LINK_TRANSITION_LINK_UP;
cas_set_link_modes(cp);
netif_carrier_on(cp->dev);
}
}
} else if (cp->lstate == link_up) {
cp->lstate = link_down;
if (link_transition_timeout != 0 &&
cp->link_transition != LINK_TRANSITION_REQUESTED_RESET &&
!cp->link_transition_jiffies_valid) {
/*
* force a reset, as a workaround for the
* link-failure problem. May want to move this to a
* point a bit earlier in the sequence. If we had
* generated a reset a short time ago, we'll wait for
* the link timer to check the status until a
* timer expires (link_transistion_jiffies_valid is
* true when the timer is running.) Instead of using
* a system timer, we just do a check whenever the
* link timer is running - this clears the flag after
* a suitable delay.
*/
retval = 1;
cp->link_transition = LINK_TRANSITION_REQUESTED_RESET;
cp->link_transition_jiffies = jiffies;
cp->link_transition_jiffies_valid = 1;
} else {
cp->link_transition = LINK_TRANSITION_ON_FAILURE;
}
netif_carrier_off(cp->dev);
if (cp->opened)
netif_info(cp, link, cp->dev, "PCS link down\n");
/* Cassini only: if you force a mode, there can be
* sync problems on link down. to fix that, the following
* things need to be checked:
* 1) read serialink state register
* 2) read pcs status register to verify link down.
* 3) if link down and serial link == 0x03, then you need
* to global reset the chip.
*/
if ((cp->cas_flags & CAS_FLAG_REG_PLUS) == 0) {
/* should check to see if we're in a forced mode */
stat = readl(cp->regs + REG_PCS_SERDES_STATE);
if (stat == 0x03)
return 1;
}
} else if (cp->lstate == link_down) {
if (link_transition_timeout != 0 &&
cp->link_transition != LINK_TRANSITION_REQUESTED_RESET &&
!cp->link_transition_jiffies_valid) {
/* force a reset, as a workaround for the
* link-failure problem. May want to move
* this to a point a bit earlier in the
* sequence.
*/
retval = 1;
cp->link_transition = LINK_TRANSITION_REQUESTED_RESET;
cp->link_transition_jiffies = jiffies;
cp->link_transition_jiffies_valid = 1;
} else {
cp->link_transition = LINK_TRANSITION_STILL_FAILED;
}
}
return retval;
}
static int cas_pcs_interrupt(struct net_device *dev,
struct cas *cp, u32 status)
{
u32 stat = readl(cp->regs + REG_PCS_INTR_STATUS);
if ((stat & PCS_INTR_STATUS_LINK_CHANGE) == 0)
return 0;
return cas_pcs_link_check(cp);
}
static int cas_txmac_interrupt(struct net_device *dev,
struct cas *cp, u32 status)
{
u32 txmac_stat = readl(cp->regs + REG_MAC_TX_STATUS);
if (!txmac_stat)
return 0;
netif_printk(cp, intr, KERN_DEBUG, cp->dev,
"txmac interrupt, txmac_stat: 0x%x\n", txmac_stat);
/* Defer timer expiration is quite normal,
* don't even log the event.
*/
if ((txmac_stat & MAC_TX_DEFER_TIMER) &&
!(txmac_stat & ~MAC_TX_DEFER_TIMER))
return 0;
spin_lock(&cp->stat_lock[0]);
if (txmac_stat & MAC_TX_UNDERRUN) {
netdev_err(dev, "TX MAC xmit underrun\n");
cp->net_stats[0].tx_fifo_errors++;
}
if (txmac_stat & MAC_TX_MAX_PACKET_ERR) {
netdev_err(dev, "TX MAC max packet size error\n");
cp->net_stats[0].tx_errors++;
}
/* The rest are all cases of one of the 16-bit TX
* counters expiring.
*/
if (txmac_stat & MAC_TX_COLL_NORMAL)
cp->net_stats[0].collisions += 0x10000;
if (txmac_stat & MAC_TX_COLL_EXCESS) {
cp->net_stats[0].tx_aborted_errors += 0x10000;
cp->net_stats[0].collisions += 0x10000;
}
if (txmac_stat & MAC_TX_COLL_LATE) {
cp->net_stats[0].tx_aborted_errors += 0x10000;
cp->net_stats[0].collisions += 0x10000;
}
spin_unlock(&cp->stat_lock[0]);
/* We do not keep track of MAC_TX_COLL_FIRST and
* MAC_TX_PEAK_ATTEMPTS events.
*/
return 0;
}
static void cas_load_firmware(struct cas *cp, cas_hp_inst_t *firmware)
{
cas_hp_inst_t *inst;
u32 val;
int i;
i = 0;
while ((inst = firmware) && inst->note) {
writel(i, cp->regs + REG_HP_INSTR_RAM_ADDR);
val = CAS_BASE(HP_INSTR_RAM_HI_VAL, inst->val);
val |= CAS_BASE(HP_INSTR_RAM_HI_MASK, inst->mask);
writel(val, cp->regs + REG_HP_INSTR_RAM_DATA_HI);
val = CAS_BASE(HP_INSTR_RAM_MID_OUTARG, inst->outarg >> 10);
val |= CAS_BASE(HP_INSTR_RAM_MID_OUTOP, inst->outop);
val |= CAS_BASE(HP_INSTR_RAM_MID_FNEXT, inst->fnext);
val |= CAS_BASE(HP_INSTR_RAM_MID_FOFF, inst->foff);
val |= CAS_BASE(HP_INSTR_RAM_MID_SNEXT, inst->snext);
val |= CAS_BASE(HP_INSTR_RAM_MID_SOFF, inst->soff);
val |= CAS_BASE(HP_INSTR_RAM_MID_OP, inst->op);
writel(val, cp->regs + REG_HP_INSTR_RAM_DATA_MID);
val = CAS_BASE(HP_INSTR_RAM_LOW_OUTMASK, inst->outmask);
val |= CAS_BASE(HP_INSTR_RAM_LOW_OUTSHIFT, inst->outshift);
val |= CAS_BASE(HP_INSTR_RAM_LOW_OUTEN, inst->outenab);
val |= CAS_BASE(HP_INSTR_RAM_LOW_OUTARG, inst->outarg);
writel(val, cp->regs + REG_HP_INSTR_RAM_DATA_LOW);
++firmware;
++i;
}
}
static void cas_init_rx_dma(struct cas *cp)
{
u64 desc_dma = cp->block_dvma;
u32 val;
int i, size;
/* rx free descriptors */
val = CAS_BASE(RX_CFG_SWIVEL, RX_SWIVEL_OFF_VAL);
val |= CAS_BASE(RX_CFG_DESC_RING, RX_DESC_RINGN_INDEX(0));
val |= CAS_BASE(RX_CFG_COMP_RING, RX_COMP_RINGN_INDEX(0));
if ((N_RX_DESC_RINGS > 1) &&
(cp->cas_flags & CAS_FLAG_REG_PLUS)) /* do desc 2 */
val |= CAS_BASE(RX_CFG_DESC_RING1, RX_DESC_RINGN_INDEX(1));
writel(val, cp->regs + REG_RX_CFG);
val = (unsigned long) cp->init_rxds[0] -
(unsigned long) cp->init_block;
writel((desc_dma + val) >> 32, cp->regs + REG_RX_DB_HI);
writel((desc_dma + val) & 0xffffffff, cp->regs + REG_RX_DB_LOW);
writel(RX_DESC_RINGN_SIZE(0) - 4, cp->regs + REG_RX_KICK);
if (cp->cas_flags & CAS_FLAG_REG_PLUS) {
/* rx desc 2 is for IPSEC packets. however,
* we don't it that for that purpose.
*/
val = (unsigned long) cp->init_rxds[1] -
(unsigned long) cp->init_block;
writel((desc_dma + val) >> 32, cp->regs + REG_PLUS_RX_DB1_HI);
writel((desc_dma + val) & 0xffffffff, cp->regs +
REG_PLUS_RX_DB1_LOW);
writel(RX_DESC_RINGN_SIZE(1) - 4, cp->regs +
REG_PLUS_RX_KICK1);
}
/* rx completion registers */
val = (unsigned long) cp->init_rxcs[0] -
(unsigned long) cp->init_block;
writel((desc_dma + val) >> 32, cp->regs + REG_RX_CB_HI);
writel((desc_dma + val) & 0xffffffff, cp->regs + REG_RX_CB_LOW);
if (cp->cas_flags & CAS_FLAG_REG_PLUS) {
/* rx comp 2-4 */
for (i = 1; i < MAX_RX_COMP_RINGS; i++) {
val = (unsigned long) cp->init_rxcs[i] -
(unsigned long) cp->init_block;
writel((desc_dma + val) >> 32, cp->regs +
REG_PLUS_RX_CBN_HI(i));
writel((desc_dma + val) & 0xffffffff, cp->regs +
REG_PLUS_RX_CBN_LOW(i));
}
}
/* read selective clear regs to prevent spurious interrupts
* on reset because complete == kick.
* selective clear set up to prevent interrupts on resets
*/
readl(cp->regs + REG_INTR_STATUS_ALIAS);
writel(INTR_RX_DONE | INTR_RX_BUF_UNAVAIL, cp->regs + REG_ALIAS_CLEAR);
if (cp->cas_flags & CAS_FLAG_REG_PLUS) {
for (i = 1; i < N_RX_COMP_RINGS; i++)
readl(cp->regs + REG_PLUS_INTRN_STATUS_ALIAS(i));
/* 2 is different from 3 and 4 */
if (N_RX_COMP_RINGS > 1)
writel(INTR_RX_DONE_ALT | INTR_RX_BUF_UNAVAIL_1,
cp->regs + REG_PLUS_ALIASN_CLEAR(1));
for (i = 2; i < N_RX_COMP_RINGS; i++)
writel(INTR_RX_DONE_ALT,
cp->regs + REG_PLUS_ALIASN_CLEAR(i));
}
/* set up pause thresholds */
val = CAS_BASE(RX_PAUSE_THRESH_OFF,
cp->rx_pause_off / RX_PAUSE_THRESH_QUANTUM);
val |= CAS_BASE(RX_PAUSE_THRESH_ON,
cp->rx_pause_on / RX_PAUSE_THRESH_QUANTUM);
writel(val, cp->regs + REG_RX_PAUSE_THRESH);
/* zero out dma reassembly buffers */
for (i = 0; i < 64; i++) {
writel(i, cp->regs + REG_RX_TABLE_ADDR);
writel(0x0, cp->regs + REG_RX_TABLE_DATA_LOW);
writel(0x0, cp->regs + REG_RX_TABLE_DATA_MID);
writel(0x0, cp->regs + REG_RX_TABLE_DATA_HI);
}
/* make sure address register is 0 for normal operation */
writel(0x0, cp->regs + REG_RX_CTRL_FIFO_ADDR);
writel(0x0, cp->regs + REG_RX_IPP_FIFO_ADDR);
/* interrupt mitigation */
#ifdef USE_RX_BLANK
val = CAS_BASE(RX_BLANK_INTR_TIME, RX_BLANK_INTR_TIME_VAL);
val |= CAS_BASE(RX_BLANK_INTR_PKT, RX_BLANK_INTR_PKT_VAL);
writel(val, cp->regs + REG_RX_BLANK);
#else
writel(0x0, cp->regs + REG_RX_BLANK);
#endif
/* interrupt generation as a function of low water marks for
* free desc and completion entries. these are used to trigger
* housekeeping for rx descs. we don't use the free interrupt
* as it's not very useful
*/
/* val = CAS_BASE(RX_AE_THRESH_FREE, RX_AE_FREEN_VAL(0)); */
val = CAS_BASE(RX_AE_THRESH_COMP, RX_AE_COMP_VAL);
writel(val, cp->regs + REG_RX_AE_THRESH);
if (cp->cas_flags & CAS_FLAG_REG_PLUS) {
val = CAS_BASE(RX_AE1_THRESH_FREE, RX_AE_FREEN_VAL(1));
writel(val, cp->regs + REG_PLUS_RX_AE1_THRESH);
}
/* Random early detect registers. useful for congestion avoidance.
* this should be tunable.
*/
writel(0x0, cp->regs + REG_RX_RED);
/* receive page sizes. default == 2K (0x800) */
val = 0;
if (cp->page_size == 0x1000)
val = 0x1;
else if (cp->page_size == 0x2000)
val = 0x2;
else if (cp->page_size == 0x4000)
val = 0x3;
/* round mtu + offset. constrain to page size. */
size = cp->dev->mtu + 64;
if (size > cp->page_size)
size = cp->page_size;
if (size <= 0x400)
i = 0x0;
else if (size <= 0x800)
i = 0x1;
else if (size <= 0x1000)
i = 0x2;
else
i = 0x3;
cp->mtu_stride = 1 << (i + 10);
val = CAS_BASE(RX_PAGE_SIZE, val);
val |= CAS_BASE(RX_PAGE_SIZE_MTU_STRIDE, i);
val |= CAS_BASE(RX_PAGE_SIZE_MTU_COUNT, cp->page_size >> (i + 10));
val |= CAS_BASE(RX_PAGE_SIZE_MTU_OFF, 0x1);
writel(val, cp->regs + REG_RX_PAGE_SIZE);
/* enable the header parser if desired */
if (CAS_HP_FIRMWARE == cas_prog_null)
return;
val = CAS_BASE(HP_CFG_NUM_CPU, CAS_NCPUS > 63 ? 0 : CAS_NCPUS);
val |= HP_CFG_PARSE_EN | HP_CFG_SYN_INC_MASK;
val |= CAS_BASE(HP_CFG_TCP_THRESH, HP_TCP_THRESH_VAL);
writel(val, cp->regs + REG_HP_CFG);
}
static inline void cas_rxc_init(struct cas_rx_comp *rxc)
{
memset(rxc, 0, sizeof(*rxc));
rxc->word4 = cpu_to_le64(RX_COMP4_ZERO);
}
/* NOTE: we use the ENC RX DESC ring for spares. the rx_page[0,1]
* flipping is protected by the fact that the chip will not
* hand back the same page index while it's being processed.
*/
static inline cas_page_t *cas_page_spare(struct cas *cp, const int index)
{
cas_page_t *page = cp->rx_pages[1][index];
cas_page_t *new;
if (page_count(page->buffer) == 1)
return page;
new = cas_page_dequeue(cp);
if (new) {
spin_lock(&cp->rx_inuse_lock);
list_add(&page->list, &cp->rx_inuse_list);
spin_unlock(&cp->rx_inuse_lock);
}
return new;
}
/* this needs to be changed if we actually use the ENC RX DESC ring */
static cas_page_t *cas_page_swap(struct cas *cp, const int ring,
const int index)
{
cas_page_t **page0 = cp->rx_pages[0];
cas_page_t **page1 = cp->rx_pages[1];
/* swap if buffer is in use */
if (page_count(page0[index]->buffer) > 1) {
cas_page_t *new = cas_page_spare(cp, index);
if (new) {
page1[index] = page0[index];
page0[index] = new;
}
}
RX_USED_SET(page0[index], 0);
return page0[index];
}
static void cas_clean_rxds(struct cas *cp)
{
/* only clean ring 0 as ring 1 is used for spare buffers */
struct cas_rx_desc *rxd = cp->init_rxds[0];
int i, size;
/* release all rx flows */
for (i = 0; i < N_RX_FLOWS; i++) {
struct sk_buff *skb;
while ((skb = __skb_dequeue(&cp->rx_flows[i]))) {
cas_skb_release(skb);
}
}
/* initialize descriptors */
size = RX_DESC_RINGN_SIZE(0);
for (i = 0; i < size; i++) {
cas_page_t *page = cas_page_swap(cp, 0, i);
rxd[i].buffer = cpu_to_le64(page->dma_addr);
rxd[i].index = cpu_to_le64(CAS_BASE(RX_INDEX_NUM, i) |
CAS_BASE(RX_INDEX_RING, 0));
}
cp->rx_old[0] = RX_DESC_RINGN_SIZE(0) - 4;
cp->rx_last[0] = 0;
cp->cas_flags &= ~CAS_FLAG_RXD_POST(0);
}
static void cas_clean_rxcs(struct cas *cp)
{
int i, j;
/* take ownership of rx comp descriptors */
memset(cp->rx_cur, 0, sizeof(*cp->rx_cur)*N_RX_COMP_RINGS);
memset(cp->rx_new, 0, sizeof(*cp->rx_new)*N_RX_COMP_RINGS);
for (i = 0; i < N_RX_COMP_RINGS; i++) {
struct cas_rx_comp *rxc = cp->init_rxcs[i];
for (j = 0; j < RX_COMP_RINGN_SIZE(i); j++) {
cas_rxc_init(rxc + j);
}
}
}
#if 0
/* When we get a RX fifo overflow, the RX unit is probably hung
* so we do the following.
*
* If any part of the reset goes wrong, we return 1 and that causes the
* whole chip to be reset.
*/
static int cas_rxmac_reset(struct cas *cp)
{
struct net_device *dev = cp->dev;
int limit;
u32 val;
/* First, reset MAC RX. */
writel(cp->mac_rx_cfg & ~MAC_RX_CFG_EN, cp->regs + REG_MAC_RX_CFG);
for (limit = 0; limit < STOP_TRIES; limit++) {
if (!(readl(cp->regs + REG_MAC_RX_CFG) & MAC_RX_CFG_EN))
break;
udelay(10);
}
if (limit == STOP_TRIES) {
netdev_err(dev, "RX MAC will not disable, resetting whole chip\n");
return 1;
}
/* Second, disable RX DMA. */
writel(0, cp->regs + REG_RX_CFG);
for (limit = 0; limit < STOP_TRIES; limit++) {
if (!(readl(cp->regs + REG_RX_CFG) & RX_CFG_DMA_EN))
break;
udelay(10);
}
if (limit == STOP_TRIES) {
netdev_err(dev, "RX DMA will not disable, resetting whole chip\n");
return 1;
}
mdelay(5);
/* Execute RX reset command. */
writel(SW_RESET_RX, cp->regs + REG_SW_RESET);
for (limit = 0; limit < STOP_TRIES; limit++) {
if (!(readl(cp->regs + REG_SW_RESET) & SW_RESET_RX))
break;
udelay(10);
}
if (limit == STOP_TRIES) {
netdev_err(dev, "RX reset command will not execute, resetting whole chip\n");
return 1;
}
/* reset driver rx state */
cas_clean_rxds(cp);
cas_clean_rxcs(cp);
/* Now, reprogram the rest of RX unit. */
cas_init_rx_dma(cp);
/* re-enable */
val = readl(cp->regs + REG_RX_CFG);
writel(val | RX_CFG_DMA_EN, cp->regs + REG_RX_CFG);
writel(MAC_RX_FRAME_RECV, cp->regs + REG_MAC_RX_MASK);
val = readl(cp->regs + REG_MAC_RX_CFG);
writel(val | MAC_RX_CFG_EN, cp->regs + REG_MAC_RX_CFG);
return 0;
}
#endif
static int cas_rxmac_interrupt(struct net_device *dev, struct cas *cp,
u32 status)
{
u32 stat = readl(cp->regs + REG_MAC_RX_STATUS);
if (!stat)
return 0;
netif_dbg(cp, intr, cp->dev, "rxmac interrupt, stat: 0x%x\n", stat);
/* these are all rollovers */
spin_lock(&cp->stat_lock[0]);
if (stat & MAC_RX_ALIGN_ERR)
cp->net_stats[0].rx_frame_errors += 0x10000;
if (stat & MAC_RX_CRC_ERR)
cp->net_stats[0].rx_crc_errors += 0x10000;
if (stat & MAC_RX_LEN_ERR)
cp->net_stats[0].rx_length_errors += 0x10000;
if (stat & MAC_RX_OVERFLOW) {
cp->net_stats[0].rx_over_errors++;
cp->net_stats[0].rx_fifo_errors++;
}
/* We do not track MAC_RX_FRAME_COUNT and MAC_RX_VIOL_ERR
* events.
*/
spin_unlock(&cp->stat_lock[0]);
return 0;
}
static int cas_mac_interrupt(struct net_device *dev, struct cas *cp,
u32 status)
{
u32 stat = readl(cp->regs + REG_MAC_CTRL_STATUS);
if (!stat)
return 0;
netif_printk(cp, intr, KERN_DEBUG, cp->dev,
"mac interrupt, stat: 0x%x\n", stat);
/* This interrupt is just for pause frame and pause
* tracking. It is useful for diagnostics and debug
* but probably by default we will mask these events.
*/
if (stat & MAC_CTRL_PAUSE_STATE)
cp->pause_entered++;
if (stat & MAC_CTRL_PAUSE_RECEIVED)
cp->pause_last_time_recvd = (stat >> 16);
return 0;
}
/* Must be invoked under cp->lock. */
static inline int cas_mdio_link_not_up(struct cas *cp)
{
u16 val;
switch (cp->lstate) {
case link_force_ret:
netif_info(cp, link, cp->dev, "Autoneg failed again, keeping forced mode\n");
cas_phy_write(cp, MII_BMCR, cp->link_fcntl);
cp->timer_ticks = 5;
cp->lstate = link_force_ok;
cp->link_transition = LINK_TRANSITION_LINK_CONFIG;
break;
case link_aneg:
val = cas_phy_read(cp, MII_BMCR);
/* Try forced modes. we try things in the following order:
* 1000 full -> 100 full/half -> 10 half
*/
val &= ~(BMCR_ANRESTART | BMCR_ANENABLE);
val |= BMCR_FULLDPLX;
val |= (cp->cas_flags & CAS_FLAG_1000MB_CAP) ?
CAS_BMCR_SPEED1000 : BMCR_SPEED100;
cas_phy_write(cp, MII_BMCR, val);
cp->timer_ticks = 5;
cp->lstate = link_force_try;
cp->link_transition = LINK_TRANSITION_LINK_CONFIG;
break;
case link_force_try:
/* Downgrade from 1000 to 100 to 10 Mbps if necessary. */
val = cas_phy_read(cp, MII_BMCR);
cp->timer_ticks = 5;
if (val & CAS_BMCR_SPEED1000) { /* gigabit */
val &= ~CAS_BMCR_SPEED1000;
val |= (BMCR_SPEED100 | BMCR_FULLDPLX);
cas_phy_write(cp, MII_BMCR, val);
break;
}
if (val & BMCR_SPEED100) {
if (val & BMCR_FULLDPLX) /* fd failed */
val &= ~BMCR_FULLDPLX;
else { /* 100Mbps failed */
val &= ~BMCR_SPEED100;
}
cas_phy_write(cp, MII_BMCR, val);
break;
}
break;
default:
break;
}
return 0;
}
/* must be invoked with cp->lock held */
static int cas_mii_link_check(struct cas *cp, const u16 bmsr)
{
int restart;
if (bmsr & BMSR_LSTATUS) {
/* Ok, here we got a link. If we had it due to a forced
* fallback, and we were configured for autoneg, we
* retry a short autoneg pass. If you know your hub is
* broken, use ethtool ;)
*/
if ((cp->lstate == link_force_try) &&
(cp->link_cntl & BMCR_ANENABLE)) {
cp->lstate = link_force_ret;
cp->link_transition = LINK_TRANSITION_LINK_CONFIG;
cas_mif_poll(cp, 0);
cp->link_fcntl = cas_phy_read(cp, MII_BMCR);
cp->timer_ticks = 5;
if (cp->opened)
netif_info(cp, link, cp->dev,
"Got link after fallback, retrying autoneg once...\n");
cas_phy_write(cp, MII_BMCR,
cp->link_fcntl | BMCR_ANENABLE |
BMCR_ANRESTART);
cas_mif_poll(cp, 1);
} else if (cp->lstate != link_up) {
cp->lstate = link_up;
cp->link_transition = LINK_TRANSITION_LINK_UP;
if (cp->opened) {
cas_set_link_modes(cp);
netif_carrier_on(cp->dev);
}
}
return 0;
}
/* link not up. if the link was previously up, we restart the
* whole process
*/
restart = 0;
if (cp->lstate == link_up) {
cp->lstate = link_down;
cp->link_transition = LINK_TRANSITION_LINK_DOWN;
netif_carrier_off(cp->dev);
if (cp->opened)
netif_info(cp, link, cp->dev, "Link down\n");
restart = 1;
} else if (++cp->timer_ticks > 10)
cas_mdio_link_not_up(cp);
return restart;
}
static int cas_mif_interrupt(struct net_device *dev, struct cas *cp,
u32 status)
{
u32 stat = readl(cp->regs + REG_MIF_STATUS);
u16 bmsr;
/* check for a link change */
if (CAS_VAL(MIF_STATUS_POLL_STATUS, stat) == 0)
return 0;
bmsr = CAS_VAL(MIF_STATUS_POLL_DATA, stat);
return cas_mii_link_check(cp, bmsr);
}
static int cas_pci_interrupt(struct net_device *dev, struct cas *cp,
u32 status)
{
u32 stat = readl(cp->regs + REG_PCI_ERR_STATUS);
if (!stat)
return 0;
netdev_err(dev, "PCI error [%04x:%04x]",
stat, readl(cp->regs + REG_BIM_DIAG));
/* cassini+ has this reserved */
if ((stat & PCI_ERR_BADACK) &&
((cp->cas_flags & CAS_FLAG_REG_PLUS) == 0))
pr_cont(" <No ACK64# during ABS64 cycle>");
if (stat & PCI_ERR_DTRTO)
pr_cont(" <Delayed transaction timeout>");
if (stat & PCI_ERR_OTHER)
pr_cont(" <other>");
if (stat & PCI_ERR_BIM_DMA_WRITE)
pr_cont(" <BIM DMA 0 write req>");
if (stat & PCI_ERR_BIM_DMA_READ)
pr_cont(" <BIM DMA 0 read req>");
pr_cont("\n");
if (stat & PCI_ERR_OTHER) {
int pci_errs;
/* Interrogate PCI config space for the
* true cause.
*/
pci_errs = pci_status_get_and_clear_errors(cp->pdev);
netdev_err(dev, "PCI status errors[%04x]\n", pci_errs);
if (pci_errs & PCI_STATUS_PARITY)
netdev_err(dev, "PCI parity error detected\n");
if (pci_errs & PCI_STATUS_SIG_TARGET_ABORT)
netdev_err(dev, "PCI target abort\n");
if (pci_errs & PCI_STATUS_REC_TARGET_ABORT)
netdev_err(dev, "PCI master acks target abort\n");
if (pci_errs & PCI_STATUS_REC_MASTER_ABORT)
netdev_err(dev, "PCI master abort\n");
if (pci_errs & PCI_STATUS_SIG_SYSTEM_ERROR)
netdev_err(dev, "PCI system error SERR#\n");
if (pci_errs & PCI_STATUS_DETECTED_PARITY)
netdev_err(dev, "PCI parity error\n");
}
/* For all PCI errors, we should reset the chip. */
return 1;
}
/* All non-normal interrupt conditions get serviced here.
* Returns non-zero if we should just exit the interrupt
* handler right now (ie. if we reset the card which invalidates
* all of the other original irq status bits).
*/
static int cas_abnormal_irq(struct net_device *dev, struct cas *cp,
u32 status)
{
if (status & INTR_RX_TAG_ERROR) {
/* corrupt RX tag framing */
netif_printk(cp, rx_err, KERN_DEBUG, cp->dev,
"corrupt rx tag framing\n");
spin_lock(&cp->stat_lock[0]);
cp->net_stats[0].rx_errors++;
spin_unlock(&cp->stat_lock[0]);
goto do_reset;
}
if (status & INTR_RX_LEN_MISMATCH) {
/* length mismatch. */
netif_printk(cp, rx_err, KERN_DEBUG, cp->dev,
"length mismatch for rx frame\n");
spin_lock(&cp->stat_lock[0]);
cp->net_stats[0].rx_errors++;
spin_unlock(&cp->stat_lock[0]);
goto do_reset;
}
if (status & INTR_PCS_STATUS) {
if (cas_pcs_interrupt(dev, cp, status))
goto do_reset;
}
if (status & INTR_TX_MAC_STATUS) {
if (cas_txmac_interrupt(dev, cp, status))
goto do_reset;
}
if (status & INTR_RX_MAC_STATUS) {
if (cas_rxmac_interrupt(dev, cp, status))
goto do_reset;
}
if (status & INTR_MAC_CTRL_STATUS) {
if (cas_mac_interrupt(dev, cp, status))
goto do_reset;
}
if (status & INTR_MIF_STATUS) {
if (cas_mif_interrupt(dev, cp, status))
goto do_reset;
}
if (status & INTR_PCI_ERROR_STATUS) {
if (cas_pci_interrupt(dev, cp, status))
goto do_reset;
}
return 0;
do_reset:
#if 1
atomic_inc(&cp->reset_task_pending);
atomic_inc(&cp->reset_task_pending_all);
netdev_err(dev, "reset called in cas_abnormal_irq [0x%x]\n", status);
schedule_work(&cp->reset_task);
#else
atomic_set(&cp->reset_task_pending, CAS_RESET_ALL);
netdev_err(dev, "reset called in cas_abnormal_irq\n");
schedule_work(&cp->reset_task);
#endif
return 1;
}
/* NOTE: CAS_TABORT returns 1 or 2 so that it can be used when
* determining whether to do a netif_stop/wakeup
*/
#define CAS_TABORT(x) (((x)->cas_flags & CAS_FLAG_TARGET_ABORT) ? 2 : 1)
#define CAS_ROUND_PAGE(x) (((x) + PAGE_SIZE - 1) & PAGE_MASK)
static inline int cas_calc_tabort(struct cas *cp, const unsigned long addr,
const int len)
{
unsigned long off = addr + len;
if (CAS_TABORT(cp) == 1)
return 0;
if ((CAS_ROUND_PAGE(off) - off) > TX_TARGET_ABORT_LEN)
return 0;
return TX_TARGET_ABORT_LEN;
}
static inline void cas_tx_ringN(struct cas *cp, int ring, int limit)
{
struct cas_tx_desc *txds;
struct sk_buff **skbs;
struct net_device *dev = cp->dev;
int entry, count;
spin_lock(&cp->tx_lock[ring]);
txds = cp->init_txds[ring];
skbs = cp->tx_skbs[ring];
entry = cp->tx_old[ring];
count = TX_BUFF_COUNT(ring, entry, limit);
while (entry != limit) {
struct sk_buff *skb = skbs[entry];
dma_addr_t daddr;
u32 dlen;
int frag;
if (!skb) {
/* this should never occur */
entry = TX_DESC_NEXT(ring, entry);
continue;
}
/* however, we might get only a partial skb release. */
count -= skb_shinfo(skb)->nr_frags +
+ cp->tx_tiny_use[ring][entry].nbufs + 1;
if (count < 0)
break;
netif_printk(cp, tx_done, KERN_DEBUG, cp->dev,
"tx[%d] done, slot %d\n", ring, entry);
skbs[entry] = NULL;
cp->tx_tiny_use[ring][entry].nbufs = 0;
for (frag = 0; frag <= skb_shinfo(skb)->nr_frags; frag++) {
struct cas_tx_desc *txd = txds + entry;
daddr = le64_to_cpu(txd->buffer);
dlen = CAS_VAL(TX_DESC_BUFLEN,
le64_to_cpu(txd->control));
dma_unmap_page(&cp->pdev->dev, daddr, dlen,
DMA_TO_DEVICE);
entry = TX_DESC_NEXT(ring, entry);
/* tiny buffer may follow */
if (cp->tx_tiny_use[ring][entry].used) {
cp->tx_tiny_use[ring][entry].used = 0;
entry = TX_DESC_NEXT(ring, entry);
}
}
spin_lock(&cp->stat_lock[ring]);
cp->net_stats[ring].tx_packets++;
cp->net_stats[ring].tx_bytes += skb->len;
spin_unlock(&cp->stat_lock[ring]);
dev_consume_skb_irq(skb);
}
cp->tx_old[ring] = entry;
/* this is wrong for multiple tx rings. the net device needs
* multiple queues for this to do the right thing. we wait
* for 2*packets to be available when using tiny buffers
*/
if (netif_queue_stopped(dev) &&
(TX_BUFFS_AVAIL(cp, ring) > CAS_TABORT(cp)*(MAX_SKB_FRAGS + 1)))
netif_wake_queue(dev);
spin_unlock(&cp->tx_lock[ring]);
}
static void cas_tx(struct net_device *dev, struct cas *cp,
u32 status)
{
int limit, ring;
#ifdef USE_TX_COMPWB
u64 compwb = le64_to_cpu(cp->init_block->tx_compwb);
#endif
netif_printk(cp, intr, KERN_DEBUG, cp->dev,
"tx interrupt, status: 0x%x, %llx\n",
status, (unsigned long long)compwb);
/* process all the rings */
for (ring = 0; ring < N_TX_RINGS; ring++) {
#ifdef USE_TX_COMPWB
/* use the completion writeback registers */
limit = (CAS_VAL(TX_COMPWB_MSB, compwb) << 8) |
CAS_VAL(TX_COMPWB_LSB, compwb);
compwb = TX_COMPWB_NEXT(compwb);
#else
limit = readl(cp->regs + REG_TX_COMPN(ring));
#endif
if (cp->tx_old[ring] != limit)
cas_tx_ringN(cp, ring, limit);
}
}
static int cas_rx_process_pkt(struct cas *cp, struct cas_rx_comp *rxc,
int entry, const u64 *words,
struct sk_buff **skbref)
{
int dlen, hlen, len, i, alloclen;
int off, swivel = RX_SWIVEL_OFF_VAL;
struct cas_page *page;
struct sk_buff *skb;
void *addr, *crcaddr;
__sum16 csum;
char *p;
hlen = CAS_VAL(RX_COMP2_HDR_SIZE, words[1]);
dlen = CAS_VAL(RX_COMP1_DATA_SIZE, words[0]);
len = hlen + dlen;
if (RX_COPY_ALWAYS || (words[2] & RX_COMP3_SMALL_PKT))
alloclen = len;
else
alloclen = max(hlen, RX_COPY_MIN);
skb = netdev_alloc_skb(cp->dev, alloclen + swivel + cp->crc_size);
if (skb == NULL)
return -1;
*skbref = skb;
skb_reserve(skb, swivel);
p = skb->data;
addr = crcaddr = NULL;
if (hlen) { /* always copy header pages */
i = CAS_VAL(RX_COMP2_HDR_INDEX, words[1]);
page = cp->rx_pages[CAS_VAL(RX_INDEX_RING, i)][CAS_VAL(RX_INDEX_NUM, i)];
off = CAS_VAL(RX_COMP2_HDR_OFF, words[1]) * 0x100 +
swivel;
i = hlen;
if (!dlen) /* attach FCS */
i += cp->crc_size;
dma_sync_single_for_cpu(&cp->pdev->dev, page->dma_addr + off,
i, DMA_FROM_DEVICE);
addr = cas_page_map(page->buffer);
memcpy(p, addr + off, i);
dma_sync_single_for_device(&cp->pdev->dev,
page->dma_addr + off, i,
DMA_FROM_DEVICE);
cas_page_unmap(addr);
RX_USED_ADD(page, 0x100);
p += hlen;
swivel = 0;
}
if (alloclen < (hlen + dlen)) {
skb_frag_t *frag = skb_shinfo(skb)->frags;
/* normal or jumbo packets. we use frags */
i = CAS_VAL(RX_COMP1_DATA_INDEX, words[0]);
page = cp->rx_pages[CAS_VAL(RX_INDEX_RING, i)][CAS_VAL(RX_INDEX_NUM, i)];
off = CAS_VAL(RX_COMP1_DATA_OFF, words[0]) + swivel;
hlen = min(cp->page_size - off, dlen);
if (hlen < 0) {
netif_printk(cp, rx_err, KERN_DEBUG, cp->dev,
"rx page overflow: %d\n", hlen);
dev_kfree_skb_irq(skb);
return -1;
}
i = hlen;
if (i == dlen) /* attach FCS */
i += cp->crc_size;
dma_sync_single_for_cpu(&cp->pdev->dev, page->dma_addr + off,
i, DMA_FROM_DEVICE);
/* make sure we always copy a header */
swivel = 0;
if (p == (char *) skb->data) { /* not split */
addr = cas_page_map(page->buffer);
memcpy(p, addr + off, RX_COPY_MIN);
dma_sync_single_for_device(&cp->pdev->dev,
page->dma_addr + off, i,
DMA_FROM_DEVICE);
cas_page_unmap(addr);
off += RX_COPY_MIN;
swivel = RX_COPY_MIN;
RX_USED_ADD(page, cp->mtu_stride);
} else {
RX_USED_ADD(page, hlen);
}
skb_put(skb, alloclen);
skb_shinfo(skb)->nr_frags++;
skb->data_len += hlen - swivel;
skb->truesize += hlen - swivel;
skb->len += hlen - swivel;
__skb_frag_set_page(frag, page->buffer);
__skb_frag_ref(frag);
skb_frag_off_set(frag, off);
skb_frag_size_set(frag, hlen - swivel);
/* any more data? */
if ((words[0] & RX_COMP1_SPLIT_PKT) && ((dlen -= hlen) > 0)) {
hlen = dlen;
off = 0;
i = CAS_VAL(RX_COMP2_NEXT_INDEX, words[1]);
page = cp->rx_pages[CAS_VAL(RX_INDEX_RING, i)][CAS_VAL(RX_INDEX_NUM, i)];
dma_sync_single_for_cpu(&cp->pdev->dev,
page->dma_addr,
hlen + cp->crc_size,
DMA_FROM_DEVICE);
dma_sync_single_for_device(&cp->pdev->dev,
page->dma_addr,
hlen + cp->crc_size,
DMA_FROM_DEVICE);
skb_shinfo(skb)->nr_frags++;
skb->data_len += hlen;
skb->len += hlen;
frag++;
__skb_frag_set_page(frag, page->buffer);
__skb_frag_ref(frag);
skb_frag_off_set(frag, 0);
skb_frag_size_set(frag, hlen);
RX_USED_ADD(page, hlen + cp->crc_size);
}
if (cp->crc_size) {
addr = cas_page_map(page->buffer);
crcaddr = addr + off + hlen;
}
} else {
/* copying packet */
if (!dlen)
goto end_copy_pkt;
i = CAS_VAL(RX_COMP1_DATA_INDEX, words[0]);
page = cp->rx_pages[CAS_VAL(RX_INDEX_RING, i)][CAS_VAL(RX_INDEX_NUM, i)];
off = CAS_VAL(RX_COMP1_DATA_OFF, words[0]) + swivel;
hlen = min(cp->page_size - off, dlen);
if (hlen < 0) {
netif_printk(cp, rx_err, KERN_DEBUG, cp->dev,
"rx page overflow: %d\n", hlen);
dev_kfree_skb_irq(skb);
return -1;
}
i = hlen;
if (i == dlen) /* attach FCS */
i += cp->crc_size;
dma_sync_single_for_cpu(&cp->pdev->dev, page->dma_addr + off,
i, DMA_FROM_DEVICE);
addr = cas_page_map(page->buffer);
memcpy(p, addr + off, i);
dma_sync_single_for_device(&cp->pdev->dev,
page->dma_addr + off, i,
DMA_FROM_DEVICE);
cas_page_unmap(addr);
if (p == (char *) skb->data) /* not split */
RX_USED_ADD(page, cp->mtu_stride);
else
RX_USED_ADD(page, i);
/* any more data? */
if ((words[0] & RX_COMP1_SPLIT_PKT) && ((dlen -= hlen) > 0)) {
p += hlen;
i = CAS_VAL(RX_COMP2_NEXT_INDEX, words[1]);
page = cp->rx_pages[CAS_VAL(RX_INDEX_RING, i)][CAS_VAL(RX_INDEX_NUM, i)];
dma_sync_single_for_cpu(&cp->pdev->dev,
page->dma_addr,
dlen + cp->crc_size,
DMA_FROM_DEVICE);
addr = cas_page_map(page->buffer);
memcpy(p, addr, dlen + cp->crc_size);
dma_sync_single_for_device(&cp->pdev->dev,
page->dma_addr,
dlen + cp->crc_size,
DMA_FROM_DEVICE);
cas_page_unmap(addr);
RX_USED_ADD(page, dlen + cp->crc_size);
}
end_copy_pkt:
if (cp->crc_size) {
addr = NULL;
crcaddr = skb->data + alloclen;
}
skb_put(skb, alloclen);
}
csum = (__force __sum16)htons(CAS_VAL(RX_COMP4_TCP_CSUM, words[3]));
if (cp->crc_size) {
/* checksum includes FCS. strip it out. */
csum = csum_fold(csum_partial(crcaddr, cp->crc_size,
csum_unfold(csum)));
if (addr)
cas_page_unmap(addr);
}
skb->protocol = eth_type_trans(skb, cp->dev);
if (skb->protocol == htons(ETH_P_IP)) {
skb->csum = csum_unfold(~csum);
skb->ip_summed = CHECKSUM_COMPLETE;
} else
skb_checksum_none_assert(skb);
return len;
}
/* we can handle up to 64 rx flows at a time. we do the same thing
* as nonreassm except that we batch up the buffers.
* NOTE: we currently just treat each flow as a bunch of packets that
* we pass up. a better way would be to coalesce the packets
* into a jumbo packet. to do that, we need to do the following:
* 1) the first packet will have a clean split between header and
* data. save both.
* 2) each time the next flow packet comes in, extend the
* data length and merge the checksums.
* 3) on flow release, fix up the header.
* 4) make sure the higher layer doesn't care.
* because packets get coalesced, we shouldn't run into fragment count
* issues.
*/
static inline void cas_rx_flow_pkt(struct cas *cp, const u64 *words,
struct sk_buff *skb)
{
int flowid = CAS_VAL(RX_COMP3_FLOWID, words[2]) & (N_RX_FLOWS - 1);
struct sk_buff_head *flow = &cp->rx_flows[flowid];
/* this is protected at a higher layer, so no need to
* do any additional locking here. stick the buffer
* at the end.
*/
__skb_queue_tail(flow, skb);
if (words[0] & RX_COMP1_RELEASE_FLOW) {
while ((skb = __skb_dequeue(flow))) {
cas_skb_release(skb);
}
}
}
/* put rx descriptor back on ring. if a buffer is in use by a higher
* layer, this will need to put in a replacement.
*/
static void cas_post_page(struct cas *cp, const int ring, const int index)
{
cas_page_t *new;
int entry;
entry = cp->rx_old[ring];
new = cas_page_swap(cp, ring, index);
cp->init_rxds[ring][entry].buffer = cpu_to_le64(new->dma_addr);
cp->init_rxds[ring][entry].index =
cpu_to_le64(CAS_BASE(RX_INDEX_NUM, index) |
CAS_BASE(RX_INDEX_RING, ring));
entry = RX_DESC_ENTRY(ring, entry + 1);
cp->rx_old[ring] = entry;
if (entry % 4)
return;
if (ring == 0)
writel(entry, cp->regs + REG_RX_KICK);
else if ((N_RX_DESC_RINGS > 1) &&
(cp->cas_flags & CAS_FLAG_REG_PLUS))
writel(entry, cp->regs + REG_PLUS_RX_KICK1);
}
/* only when things are bad */
static int cas_post_rxds_ringN(struct cas *cp, int ring, int num)
{
unsigned int entry, last, count, released;
int cluster;
cas_page_t **page = cp->rx_pages[ring];
entry = cp->rx_old[ring];
netif_printk(cp, intr, KERN_DEBUG, cp->dev,
"rxd[%d] interrupt, done: %d\n", ring, entry);
cluster = -1;
count = entry & 0x3;
last = RX_DESC_ENTRY(ring, num ? entry + num - 4: entry - 4);
released = 0;
while (entry != last) {
/* make a new buffer if it's still in use */
if (page_count(page[entry]->buffer) > 1) {
cas_page_t *new = cas_page_dequeue(cp);
if (!new) {
/* let the timer know that we need to
* do this again
*/
cp->cas_flags |= CAS_FLAG_RXD_POST(ring);
if (!timer_pending(&cp->link_timer))
mod_timer(&cp->link_timer, jiffies +
CAS_LINK_FAST_TIMEOUT);
cp->rx_old[ring] = entry;
cp->rx_last[ring] = num ? num - released : 0;
return -ENOMEM;
}
spin_lock(&cp->rx_inuse_lock);
list_add(&page[entry]->list, &cp->rx_inuse_list);
spin_unlock(&cp->rx_inuse_lock);
cp->init_rxds[ring][entry].buffer =
cpu_to_le64(new->dma_addr);
page[entry] = new;
}
if (++count == 4) {
cluster = entry;
count = 0;
}
released++;
entry = RX_DESC_ENTRY(ring, entry + 1);
}
cp->rx_old[ring] = entry;
if (cluster < 0)
return 0;
if (ring == 0)
writel(cluster, cp->regs + REG_RX_KICK);
else if ((N_RX_DESC_RINGS > 1) &&
(cp->cas_flags & CAS_FLAG_REG_PLUS))
writel(cluster, cp->regs + REG_PLUS_RX_KICK1);
return 0;
}
/* process a completion ring. packets are set up in three basic ways:
* small packets: should be copied header + data in single buffer.
* large packets: header and data in a single buffer.
* split packets: header in a separate buffer from data.
* data may be in multiple pages. data may be > 256
* bytes but in a single page.
*
* NOTE: RX page posting is done in this routine as well. while there's
* the capability of using multiple RX completion rings, it isn't
* really worthwhile due to the fact that the page posting will
* force serialization on the single descriptor ring.
*/
static int cas_rx_ringN(struct cas *cp, int ring, int budget)
{
struct cas_rx_comp *rxcs = cp->init_rxcs[ring];
int entry, drops;
int npackets = 0;
netif_printk(cp, intr, KERN_DEBUG, cp->dev,
"rx[%d] interrupt, done: %d/%d\n",
ring,
readl(cp->regs + REG_RX_COMP_HEAD), cp->rx_new[ring]);
entry = cp->rx_new[ring];
drops = 0;
while (1) {
struct cas_rx_comp *rxc = rxcs + entry;
struct sk_buff *skb;
int type, len;
u64 words[4];
int i, dring;
words[0] = le64_to_cpu(rxc->word1);
words[1] = le64_to_cpu(rxc->word2);
words[2] = le64_to_cpu(rxc->word3);
words[3] = le64_to_cpu(rxc->word4);
/* don't touch if still owned by hw */
type = CAS_VAL(RX_COMP1_TYPE, words[0]);
if (type == 0)
break;
/* hw hasn't cleared the zero bit yet */
if (words[3] & RX_COMP4_ZERO) {
break;
}
/* get info on the packet */
if (words[3] & (RX_COMP4_LEN_MISMATCH | RX_COMP4_BAD)) {
spin_lock(&cp->stat_lock[ring]);
cp->net_stats[ring].rx_errors++;
if (words[3] & RX_COMP4_LEN_MISMATCH)
cp->net_stats[ring].rx_length_errors++;
if (words[3] & RX_COMP4_BAD)
cp->net_stats[ring].rx_crc_errors++;
spin_unlock(&cp->stat_lock[ring]);
/* We'll just return it to Cassini. */
drop_it:
spin_lock(&cp->stat_lock[ring]);
++cp->net_stats[ring].rx_dropped;
spin_unlock(&cp->stat_lock[ring]);
goto next;
}
len = cas_rx_process_pkt(cp, rxc, entry, words, &skb);
if (len < 0) {
++drops;
goto drop_it;
}
/* see if it's a flow re-assembly or not. the driver
* itself handles release back up.
*/
if (RX_DONT_BATCH || (type == 0x2)) {
/* non-reassm: these always get released */
cas_skb_release(skb);
} else {
cas_rx_flow_pkt(cp, words, skb);
}
spin_lock(&cp->stat_lock[ring]);
cp->net_stats[ring].rx_packets++;
cp->net_stats[ring].rx_bytes += len;
spin_unlock(&cp->stat_lock[ring]);
next:
npackets++;
/* should it be released? */
if (words[0] & RX_COMP1_RELEASE_HDR) {
i = CAS_VAL(RX_COMP2_HDR_INDEX, words[1]);
dring = CAS_VAL(RX_INDEX_RING, i);
i = CAS_VAL(RX_INDEX_NUM, i);
cas_post_page(cp, dring, i);
}
if (words[0] & RX_COMP1_RELEASE_DATA) {
i = CAS_VAL(RX_COMP1_DATA_INDEX, words[0]);
dring = CAS_VAL(RX_INDEX_RING, i);
i = CAS_VAL(RX_INDEX_NUM, i);
cas_post_page(cp, dring, i);
}
if (words[0] & RX_COMP1_RELEASE_NEXT) {
i = CAS_VAL(RX_COMP2_NEXT_INDEX, words[1]);
dring = CAS_VAL(RX_INDEX_RING, i);
i = CAS_VAL(RX_INDEX_NUM, i);
cas_post_page(cp, dring, i);
}
/* skip to the next entry */
entry = RX_COMP_ENTRY(ring, entry + 1 +
CAS_VAL(RX_COMP1_SKIP, words[0]));
#ifdef USE_NAPI
if (budget && (npackets >= budget))
break;
#endif
}
cp->rx_new[ring] = entry;
if (drops)
netdev_info(cp->dev, "Memory squeeze, deferring packet\n");
return npackets;
}
/* put completion entries back on the ring */
static void cas_post_rxcs_ringN(struct net_device *dev,
struct cas *cp, int ring)
{
struct cas_rx_comp *rxc = cp->init_rxcs[ring];
int last, entry;
last = cp->rx_cur[ring];
entry = cp->rx_new[ring];
netif_printk(cp, intr, KERN_DEBUG, dev,
"rxc[%d] interrupt, done: %d/%d\n",
ring, readl(cp->regs + REG_RX_COMP_HEAD), entry);
/* zero and re-mark descriptors */
while (last != entry) {
cas_rxc_init(rxc + last);
last = RX_COMP_ENTRY(ring, last + 1);
}
cp->rx_cur[ring] = last;
if (ring == 0)
writel(last, cp->regs + REG_RX_COMP_TAIL);
else if (cp->cas_flags & CAS_FLAG_REG_PLUS)
writel(last, cp->regs + REG_PLUS_RX_COMPN_TAIL(ring));
}
/* cassini can use all four PCI interrupts for the completion ring.
* rings 3 and 4 are identical
*/
#if defined(USE_PCI_INTC) || defined(USE_PCI_INTD)
static inline void cas_handle_irqN(struct net_device *dev,
struct cas *cp, const u32 status,
const int ring)
{
if (status & (INTR_RX_COMP_FULL_ALT | INTR_RX_COMP_AF_ALT))
cas_post_rxcs_ringN(dev, cp, ring);
}
static irqreturn_t cas_interruptN(int irq, void *dev_id)
{
struct net_device *dev = dev_id;
struct cas *cp = netdev_priv(dev);
unsigned long flags;
int ring = (irq == cp->pci_irq_INTC) ? 2 : 3;
u32 status = readl(cp->regs + REG_PLUS_INTRN_STATUS(ring));
/* check for shared irq */
if (status == 0)
return IRQ_NONE;
spin_lock_irqsave(&cp->lock, flags);
if (status & INTR_RX_DONE_ALT) { /* handle rx separately */
#ifdef USE_NAPI
cas_mask_intr(cp);
napi_schedule(&cp->napi);
#else
cas_rx_ringN(cp, ring, 0);
#endif
status &= ~INTR_RX_DONE_ALT;
}
if (status)
cas_handle_irqN(dev, cp, status, ring);
spin_unlock_irqrestore(&cp->lock, flags);
return IRQ_HANDLED;
}
#endif
#ifdef USE_PCI_INTB
/* everything but rx packets */
static inline void cas_handle_irq1(struct cas *cp, const u32 status)
{
if (status & INTR_RX_BUF_UNAVAIL_1) {
/* Frame arrived, no free RX buffers available.
* NOTE: we can get this on a link transition. */
cas_post_rxds_ringN(cp, 1, 0);
spin_lock(&cp->stat_lock[1]);
cp->net_stats[1].rx_dropped++;
spin_unlock(&cp->stat_lock[1]);
}
if (status & INTR_RX_BUF_AE_1)
cas_post_rxds_ringN(cp, 1, RX_DESC_RINGN_SIZE(1) -
RX_AE_FREEN_VAL(1));
if (status & (INTR_RX_COMP_AF | INTR_RX_COMP_FULL))
cas_post_rxcs_ringN(cp, 1);
}
/* ring 2 handles a few more events than 3 and 4 */
static irqreturn_t cas_interrupt1(int irq, void *dev_id)
{
struct net_device *dev = dev_id;
struct cas *cp = netdev_priv(dev);
unsigned long flags;
u32 status = readl(cp->regs + REG_PLUS_INTRN_STATUS(1));
/* check for shared interrupt */
if (status == 0)
return IRQ_NONE;
spin_lock_irqsave(&cp->lock, flags);
if (status & INTR_RX_DONE_ALT) { /* handle rx separately */
#ifdef USE_NAPI
cas_mask_intr(cp);
napi_schedule(&cp->napi);
#else
cas_rx_ringN(cp, 1, 0);
#endif
status &= ~INTR_RX_DONE_ALT;
}
if (status)
cas_handle_irq1(cp, status);
spin_unlock_irqrestore(&cp->lock, flags);
return IRQ_HANDLED;
}
#endif
static inline void cas_handle_irq(struct net_device *dev,
struct cas *cp, const u32 status)
{
/* housekeeping interrupts */
if (status & INTR_ERROR_MASK)
cas_abnormal_irq(dev, cp, status);
if (status & INTR_RX_BUF_UNAVAIL) {
/* Frame arrived, no free RX buffers available.
* NOTE: we can get this on a link transition.
*/
cas_post_rxds_ringN(cp, 0, 0);
spin_lock(&cp->stat_lock[0]);
cp->net_stats[0].rx_dropped++;
spin_unlock(&cp->stat_lock[0]);
} else if (status & INTR_RX_BUF_AE) {
cas_post_rxds_ringN(cp, 0, RX_DESC_RINGN_SIZE(0) -
RX_AE_FREEN_VAL(0));
}
if (status & (INTR_RX_COMP_AF | INTR_RX_COMP_FULL))
cas_post_rxcs_ringN(dev, cp, 0);
}
static irqreturn_t cas_interrupt(int irq, void *dev_id)
{
struct net_device *dev = dev_id;
struct cas *cp = netdev_priv(dev);
unsigned long flags;
u32 status = readl(cp->regs + REG_INTR_STATUS);
if (status == 0)
return IRQ_NONE;
spin_lock_irqsave(&cp->lock, flags);
if (status & (INTR_TX_ALL | INTR_TX_INTME)) {
cas_tx(dev, cp, status);
status &= ~(INTR_TX_ALL | INTR_TX_INTME);
}
if (status & INTR_RX_DONE) {
#ifdef USE_NAPI
cas_mask_intr(cp);
napi_schedule(&cp->napi);
#else
cas_rx_ringN(cp, 0, 0);
#endif
status &= ~INTR_RX_DONE;
}
if (status)
cas_handle_irq(dev, cp, status);
spin_unlock_irqrestore(&cp->lock, flags);
return IRQ_HANDLED;
}
#ifdef USE_NAPI
static int cas_poll(struct napi_struct *napi, int budget)
{
struct cas *cp = container_of(napi, struct cas, napi);
struct net_device *dev = cp->dev;
int i, enable_intr, credits;
u32 status = readl(cp->regs + REG_INTR_STATUS);
unsigned long flags;
spin_lock_irqsave(&cp->lock, flags);
cas_tx(dev, cp, status);
spin_unlock_irqrestore(&cp->lock, flags);
/* NAPI rx packets. we spread the credits across all of the
* rxc rings
*
* to make sure we're fair with the work we loop through each
* ring N_RX_COMP_RING times with a request of
* budget / N_RX_COMP_RINGS
*/
enable_intr = 1;
credits = 0;
for (i = 0; i < N_RX_COMP_RINGS; i++) {
int j;
for (j = 0; j < N_RX_COMP_RINGS; j++) {
credits += cas_rx_ringN(cp, j, budget / N_RX_COMP_RINGS);
if (credits >= budget) {
enable_intr = 0;
goto rx_comp;
}
}
}
rx_comp:
/* final rx completion */
spin_lock_irqsave(&cp->lock, flags);
if (status)
cas_handle_irq(dev, cp, status);
#ifdef USE_PCI_INTB
if (N_RX_COMP_RINGS > 1) {
status = readl(cp->regs + REG_PLUS_INTRN_STATUS(1));
if (status)
cas_handle_irq1(dev, cp, status);
}
#endif
#ifdef USE_PCI_INTC
if (N_RX_COMP_RINGS > 2) {
status = readl(cp->regs + REG_PLUS_INTRN_STATUS(2));
if (status)
cas_handle_irqN(dev, cp, status, 2);
}
#endif
#ifdef USE_PCI_INTD
if (N_RX_COMP_RINGS > 3) {
status = readl(cp->regs + REG_PLUS_INTRN_STATUS(3));
if (status)
cas_handle_irqN(dev, cp, status, 3);
}
#endif
spin_unlock_irqrestore(&cp->lock, flags);
if (enable_intr) {
napi_complete(napi);
cas_unmask_intr(cp);
}
return credits;
}
#endif
#ifdef CONFIG_NET_POLL_CONTROLLER
static void cas_netpoll(struct net_device *dev)
{
struct cas *cp = netdev_priv(dev);
cas_disable_irq(cp, 0);
cas_interrupt(cp->pdev->irq, dev);
cas_enable_irq(cp, 0);
#ifdef USE_PCI_INTB
if (N_RX_COMP_RINGS > 1) {
/* cas_interrupt1(); */
}
#endif
#ifdef USE_PCI_INTC
if (N_RX_COMP_RINGS > 2) {
/* cas_interruptN(); */
}
#endif
#ifdef USE_PCI_INTD
if (N_RX_COMP_RINGS > 3) {
/* cas_interruptN(); */
}
#endif
}
#endif
static void cas_tx_timeout(struct net_device *dev, unsigned int txqueue)
{
struct cas *cp = netdev_priv(dev);
netdev_err(dev, "transmit timed out, resetting\n");
if (!cp->hw_running) {
netdev_err(dev, "hrm.. hw not running!\n");
return;
}
netdev_err(dev, "MIF_STATE[%08x]\n",
readl(cp->regs + REG_MIF_STATE_MACHINE));
netdev_err(dev, "MAC_STATE[%08x]\n",
readl(cp->regs + REG_MAC_STATE_MACHINE));
netdev_err(dev, "TX_STATE[%08x:%08x:%08x] FIFO[%08x:%08x:%08x] SM1[%08x] SM2[%08x]\n",
readl(cp->regs + REG_TX_CFG),
readl(cp->regs + REG_MAC_TX_STATUS),
readl(cp->regs + REG_MAC_TX_CFG),
readl(cp->regs + REG_TX_FIFO_PKT_CNT),
readl(cp->regs + REG_TX_FIFO_WRITE_PTR),
readl(cp->regs + REG_TX_FIFO_READ_PTR),
readl(cp->regs + REG_TX_SM_1),
readl(cp->regs + REG_TX_SM_2));
netdev_err(dev, "RX_STATE[%08x:%08x:%08x]\n",
readl(cp->regs + REG_RX_CFG),
readl(cp->regs + REG_MAC_RX_STATUS),
readl(cp->regs + REG_MAC_RX_CFG));
netdev_err(dev, "HP_STATE[%08x:%08x:%08x:%08x]\n",
readl(cp->regs + REG_HP_STATE_MACHINE),
readl(cp->regs + REG_HP_STATUS0),
readl(cp->regs + REG_HP_STATUS1),
readl(cp->regs + REG_HP_STATUS2));
#if 1
atomic_inc(&cp->reset_task_pending);
atomic_inc(&cp->reset_task_pending_all);
schedule_work(&cp->reset_task);
#else
atomic_set(&cp->reset_task_pending, CAS_RESET_ALL);
schedule_work(&cp->reset_task);
#endif
}
static inline int cas_intme(int ring, int entry)
{
/* Algorithm: IRQ every 1/2 of descriptors. */
if (!(entry & ((TX_DESC_RINGN_SIZE(ring) >> 1) - 1)))
return 1;
return 0;
}
static void cas_write_txd(struct cas *cp, int ring, int entry,
dma_addr_t mapping, int len, u64 ctrl, int last)
{
struct cas_tx_desc *txd = cp->init_txds[ring] + entry;
ctrl |= CAS_BASE(TX_DESC_BUFLEN, len);
if (cas_intme(ring, entry))
ctrl |= TX_DESC_INTME;
if (last)
ctrl |= TX_DESC_EOF;
txd->control = cpu_to_le64(ctrl);
txd->buffer = cpu_to_le64(mapping);
}
static inline void *tx_tiny_buf(struct cas *cp, const int ring,
const int entry)
{
return cp->tx_tiny_bufs[ring] + TX_TINY_BUF_LEN*entry;
}
static inline dma_addr_t tx_tiny_map(struct cas *cp, const int ring,
const int entry, const int tentry)
{
cp->tx_tiny_use[ring][tentry].nbufs++;
cp->tx_tiny_use[ring][entry].used = 1;
return cp->tx_tiny_dvma[ring] + TX_TINY_BUF_LEN*entry;
}
static inline int cas_xmit_tx_ringN(struct cas *cp, int ring,
struct sk_buff *skb)
{
struct net_device *dev = cp->dev;
int entry, nr_frags, frag, tabort, tentry;
dma_addr_t mapping;
unsigned long flags;
u64 ctrl;
u32 len;
spin_lock_irqsave(&cp->tx_lock[ring], flags);
/* This is a hard error, log it. */
if (TX_BUFFS_AVAIL(cp, ring) <=
CAS_TABORT(cp)*(skb_shinfo(skb)->nr_frags + 1)) {
netif_stop_queue(dev);
spin_unlock_irqrestore(&cp->tx_lock[ring], flags);
netdev_err(dev, "BUG! Tx Ring full when queue awake!\n");
return 1;
}
ctrl = 0;
if (skb->ip_summed == CHECKSUM_PARTIAL) {
const u64 csum_start_off = skb_checksum_start_offset(skb);
const u64 csum_stuff_off = csum_start_off + skb->csum_offset;
ctrl = TX_DESC_CSUM_EN |
CAS_BASE(TX_DESC_CSUM_START, csum_start_off) |
CAS_BASE(TX_DESC_CSUM_STUFF, csum_stuff_off);
}
entry = cp->tx_new[ring];
cp->tx_skbs[ring][entry] = skb;
nr_frags = skb_shinfo(skb)->nr_frags;
len = skb_headlen(skb);
mapping = dma_map_page(&cp->pdev->dev, virt_to_page(skb->data),
offset_in_page(skb->data), len, DMA_TO_DEVICE);
tentry = entry;
tabort = cas_calc_tabort(cp, (unsigned long) skb->data, len);
if (unlikely(tabort)) {
/* NOTE: len is always > tabort */
cas_write_txd(cp, ring, entry, mapping, len - tabort,
ctrl | TX_DESC_SOF, 0);
entry = TX_DESC_NEXT(ring, entry);
skb_copy_from_linear_data_offset(skb, len - tabort,
tx_tiny_buf(cp, ring, entry), tabort);
mapping = tx_tiny_map(cp, ring, entry, tentry);
cas_write_txd(cp, ring, entry, mapping, tabort, ctrl,
(nr_frags == 0));
} else {
cas_write_txd(cp, ring, entry, mapping, len, ctrl |
TX_DESC_SOF, (nr_frags == 0));
}
entry = TX_DESC_NEXT(ring, entry);
for (frag = 0; frag < nr_frags; frag++) {
const skb_frag_t *fragp = &skb_shinfo(skb)->frags[frag];
len = skb_frag_size(fragp);
mapping = skb_frag_dma_map(&cp->pdev->dev, fragp, 0, len,
DMA_TO_DEVICE);
tabort = cas_calc_tabort(cp, skb_frag_off(fragp), len);
if (unlikely(tabort)) {
void *addr;
/* NOTE: len is always > tabort */
cas_write_txd(cp, ring, entry, mapping, len - tabort,
ctrl, 0);
entry = TX_DESC_NEXT(ring, entry);
addr = cas_page_map(skb_frag_page(fragp));
memcpy(tx_tiny_buf(cp, ring, entry),
addr + skb_frag_off(fragp) + len - tabort,
tabort);
cas_page_unmap(addr);
mapping = tx_tiny_map(cp, ring, entry, tentry);
len = tabort;
}
cas_write_txd(cp, ring, entry, mapping, len, ctrl,
(frag + 1 == nr_frags));
entry = TX_DESC_NEXT(ring, entry);
}
cp->tx_new[ring] = entry;
if (TX_BUFFS_AVAIL(cp, ring) <= CAS_TABORT(cp)*(MAX_SKB_FRAGS + 1))
netif_stop_queue(dev);
netif_printk(cp, tx_queued, KERN_DEBUG, dev,
"tx[%d] queued, slot %d, skblen %d, avail %d\n",
ring, entry, skb->len, TX_BUFFS_AVAIL(cp, ring));
writel(entry, cp->regs + REG_TX_KICKN(ring));
spin_unlock_irqrestore(&cp->tx_lock[ring], flags);
return 0;
}
static netdev_tx_t cas_start_xmit(struct sk_buff *skb, struct net_device *dev)
{
struct cas *cp = netdev_priv(dev);
/* this is only used as a load-balancing hint, so it doesn't
* need to be SMP safe
*/
static int ring;
if (skb_padto(skb, cp->min_frame_size))
return NETDEV_TX_OK;
/* XXX: we need some higher-level QoS hooks to steer packets to
* individual queues.
*/
if (cas_xmit_tx_ringN(cp, ring++ & N_TX_RINGS_MASK, skb))
return NETDEV_TX_BUSY;
return NETDEV_TX_OK;
}
static void cas_init_tx_dma(struct cas *cp)
{
u64 desc_dma = cp->block_dvma;
unsigned long off;
u32 val;
int i;
/* set up tx completion writeback registers. must be 8-byte aligned */
#ifdef USE_TX_COMPWB
off = offsetof(struct cas_init_block, tx_compwb);
writel((desc_dma + off) >> 32, cp->regs + REG_TX_COMPWB_DB_HI);
writel((desc_dma + off) & 0xffffffff, cp->regs + REG_TX_COMPWB_DB_LOW);
#endif
/* enable completion writebacks, enable paced mode,
* disable read pipe, and disable pre-interrupt compwbs
*/
val = TX_CFG_COMPWB_Q1 | TX_CFG_COMPWB_Q2 |
TX_CFG_COMPWB_Q3 | TX_CFG_COMPWB_Q4 |
TX_CFG_DMA_RDPIPE_DIS | TX_CFG_PACED_MODE |
TX_CFG_INTR_COMPWB_DIS;
/* write out tx ring info and tx desc bases */
for (i = 0; i < MAX_TX_RINGS; i++) {
off = (unsigned long) cp->init_txds[i] -
(unsigned long) cp->init_block;
val |= CAS_TX_RINGN_BASE(i);
writel((desc_dma + off) >> 32, cp->regs + REG_TX_DBN_HI(i));
writel((desc_dma + off) & 0xffffffff, cp->regs +
REG_TX_DBN_LOW(i));
/* don't zero out the kick register here as the system
* will wedge
*/
}
writel(val, cp->regs + REG_TX_CFG);
/* program max burst sizes. these numbers should be different
* if doing QoS.
*/
#ifdef USE_QOS
writel(0x800, cp->regs + REG_TX_MAXBURST_0);
writel(0x1600, cp->regs + REG_TX_MAXBURST_1);
writel(0x2400, cp->regs + REG_TX_MAXBURST_2);
writel(0x4800, cp->regs + REG_TX_MAXBURST_3);
#else
writel(0x800, cp->regs + REG_TX_MAXBURST_0);
writel(0x800, cp->regs + REG_TX_MAXBURST_1);
writel(0x800, cp->regs + REG_TX_MAXBURST_2);
writel(0x800, cp->regs + REG_TX_MAXBURST_3);
#endif
}
/* Must be invoked under cp->lock. */
static inline void cas_init_dma(struct cas *cp)
{
cas_init_tx_dma(cp);
cas_init_rx_dma(cp);
}
static void cas_process_mc_list(struct cas *cp)
{
u16 hash_table[16];
u32 crc;
struct netdev_hw_addr *ha;
int i = 1;
memset(hash_table, 0, sizeof(hash_table));
netdev_for_each_mc_addr(ha, cp->dev) {
if (i <= CAS_MC_EXACT_MATCH_SIZE) {
/* use the alternate mac address registers for the
* first 15 multicast addresses
*/
writel((ha->addr[4] << 8) | ha->addr[5],
cp->regs + REG_MAC_ADDRN(i*3 + 0));
writel((ha->addr[2] << 8) | ha->addr[3],
cp->regs + REG_MAC_ADDRN(i*3 + 1));
writel((ha->addr[0] << 8) | ha->addr[1],
cp->regs + REG_MAC_ADDRN(i*3 + 2));
i++;
}
else {
/* use hw hash table for the next series of
* multicast addresses
*/
crc = ether_crc_le(ETH_ALEN, ha->addr);
crc >>= 24;
hash_table[crc >> 4] |= 1 << (15 - (crc & 0xf));
}
}
for (i = 0; i < 16; i++)
writel(hash_table[i], cp->regs + REG_MAC_HASH_TABLEN(i));
}
/* Must be invoked under cp->lock. */
static u32 cas_setup_multicast(struct cas *cp)
{
u32 rxcfg = 0;
int i;
if (cp->dev->flags & IFF_PROMISC) {
rxcfg |= MAC_RX_CFG_PROMISC_EN;
} else if (cp->dev->flags & IFF_ALLMULTI) {
for (i=0; i < 16; i++)
writel(0xFFFF, cp->regs + REG_MAC_HASH_TABLEN(i));
rxcfg |= MAC_RX_CFG_HASH_FILTER_EN;
} else {
cas_process_mc_list(cp);
rxcfg |= MAC_RX_CFG_HASH_FILTER_EN;
}
return rxcfg;
}
/* must be invoked under cp->stat_lock[N_TX_RINGS] */
static void cas_clear_mac_err(struct cas *cp)
{
writel(0, cp->regs + REG_MAC_COLL_NORMAL);
writel(0, cp->regs + REG_MAC_COLL_FIRST);
writel(0, cp->regs + REG_MAC_COLL_EXCESS);
writel(0, cp->regs + REG_MAC_COLL_LATE);
writel(0, cp->regs + REG_MAC_TIMER_DEFER);
writel(0, cp->regs + REG_MAC_ATTEMPTS_PEAK);
writel(0, cp->regs + REG_MAC_RECV_FRAME);
writel(0, cp->regs + REG_MAC_LEN_ERR);
writel(0, cp->regs + REG_MAC_ALIGN_ERR);
writel(0, cp->regs + REG_MAC_FCS_ERR);
writel(0, cp->regs + REG_MAC_RX_CODE_ERR);
}