myri10ge: add multislices support
Add multi-slice/MSI-X support. By default, a single slice
(and the normal firmware) are used. To enable msi-x, multi-slice
mode, one must load the driver with myri10ge_max_slices set to
either -1, or something larger than 1.
Signed-off-by: Brice Goglin <brice@myri.com>
Signed-off-by: Andrew Gallatin <gallatin@myri.com>
Signed-off-by: Jeff Garzik <jgarzik@redhat.com>
diff --git a/drivers/net/myri10ge/myri10ge.c b/drivers/net/myri10ge/myri10ge.c
index f3ef9e3..eddcee3 100644
--- a/drivers/net/myri10ge/myri10ge.c
+++ b/drivers/net/myri10ge/myri10ge.c
@@ -185,11 +185,13 @@
dma_addr_t fw_stats_bus;
int watchdog_tx_done;
int watchdog_tx_req;
+ char irq_desc[32];
};
struct myri10ge_priv {
- struct myri10ge_slice_state ss;
+ struct myri10ge_slice_state *ss;
int tx_boundary; /* boundary transmits cannot cross */
+ int num_slices;
int running; /* running? */
int csum_flag; /* rx_csums? */
int small_bytes;
@@ -208,6 +210,8 @@
dma_addr_t cmd_bus;
struct pci_dev *pdev;
int msi_enabled;
+ int msix_enabled;
+ struct msix_entry *msix_vectors;
u32 link_state;
unsigned int rdma_tags_available;
int intr_coal_delay;
@@ -244,6 +248,8 @@
static char *myri10ge_fw_unaligned = "myri10ge_ethp_z8e.dat";
static char *myri10ge_fw_aligned = "myri10ge_eth_z8e.dat";
+static char *myri10ge_fw_rss_unaligned = "myri10ge_rss_ethp_z8e.dat";
+static char *myri10ge_fw_rss_aligned = "myri10ge_rss_eth_z8e.dat";
static char *myri10ge_fw_name = NULL;
module_param(myri10ge_fw_name, charp, S_IRUGO | S_IWUSR);
@@ -321,6 +327,14 @@
module_param(myri10ge_wcfifo, int, S_IRUGO);
MODULE_PARM_DESC(myri10ge_wcfifo, "Enable WC Fifo when WC is enabled");
+static int myri10ge_max_slices = 1;
+module_param(myri10ge_max_slices, int, S_IRUGO);
+MODULE_PARM_DESC(myri10ge_max_slices, "Max tx/rx queues");
+
+static int myri10ge_rss_hash = MXGEFW_RSS_HASH_TYPE_SRC_PORT;
+module_param(myri10ge_rss_hash, int, S_IRUGO);
+MODULE_PARM_DESC(myri10ge_rss_hash, "Type of RSS hashing to do");
+
#define MYRI10GE_FW_OFFSET 1024*1024
#define MYRI10GE_HIGHPART_TO_U32(X) \
(sizeof (X) == 8) ? ((u32)((u64)(X) >> 32)) : (0)
@@ -657,7 +671,7 @@
return 0;
}
-static int myri10ge_load_firmware(struct myri10ge_priv *mgp)
+static int myri10ge_load_firmware(struct myri10ge_priv *mgp, int adopt)
{
char __iomem *submit;
__be32 buf[16] __attribute__ ((__aligned__(8)));
@@ -667,6 +681,8 @@
size = 0;
status = myri10ge_load_hotplug_firmware(mgp, &size);
if (status) {
+ if (!adopt)
+ return status;
dev_warn(&mgp->pdev->dev, "hotplug firmware loading failed\n");
/* Do not attempt to adopt firmware if there
@@ -859,7 +875,8 @@
static int myri10ge_reset(struct myri10ge_priv *mgp)
{
struct myri10ge_cmd cmd;
- int status;
+ struct myri10ge_slice_state *ss;
+ int i, status;
size_t bytes;
/* try to send a reset command to the card to see if it
@@ -872,20 +889,74 @@
}
(void)myri10ge_dma_test(mgp, MXGEFW_DMA_TEST);
+ /*
+ * Use non-ndis mcp_slot (eg, 4 bytes total,
+ * no toeplitz hash value returned. Older firmware will
+ * not understand this command, but will use the correct
+ * sized mcp_slot, so we ignore error returns
+ */
+ cmd.data0 = MXGEFW_RSS_MCP_SLOT_TYPE_MIN;
+ (void)myri10ge_send_cmd(mgp, MXGEFW_CMD_SET_RSS_MCP_SLOT_TYPE, &cmd, 0);
/* Now exchange information about interrupts */
- bytes = mgp->max_intr_slots * sizeof(*mgp->ss.rx_done.entry);
- memset(mgp->ss.rx_done.entry, 0, bytes);
+ bytes = mgp->max_intr_slots * sizeof(*mgp->ss[0].rx_done.entry);
cmd.data0 = (u32) bytes;
status = myri10ge_send_cmd(mgp, MXGEFW_CMD_SET_INTRQ_SIZE, &cmd, 0);
- cmd.data0 = MYRI10GE_LOWPART_TO_U32(mgp->ss.rx_done.bus);
- cmd.data1 = MYRI10GE_HIGHPART_TO_U32(mgp->ss.rx_done.bus);
- status |= myri10ge_send_cmd(mgp, MXGEFW_CMD_SET_INTRQ_DMA, &cmd, 0);
+
+ /*
+ * Even though we already know how many slices are supported
+ * via myri10ge_probe_slices() MXGEFW_CMD_GET_MAX_RSS_QUEUES
+ * has magic side effects, and must be called after a reset.
+ * It must be called prior to calling any RSS related cmds,
+ * including assigning an interrupt queue for anything but
+ * slice 0. It must also be called *after*
+ * MXGEFW_CMD_SET_INTRQ_SIZE, since the intrq size is used by
+ * the firmware to compute offsets.
+ */
+
+ if (mgp->num_slices > 1) {
+
+ /* ask the maximum number of slices it supports */
+ status = myri10ge_send_cmd(mgp, MXGEFW_CMD_GET_MAX_RSS_QUEUES,
+ &cmd, 0);
+ if (status != 0) {
+ dev_err(&mgp->pdev->dev,
+ "failed to get number of slices\n");
+ }
+
+ /*
+ * MXGEFW_CMD_ENABLE_RSS_QUEUES must be called prior
+ * to setting up the interrupt queue DMA
+ */
+
+ cmd.data0 = mgp->num_slices;
+ cmd.data1 = 1; /* use MSI-X */
+ status = myri10ge_send_cmd(mgp, MXGEFW_CMD_ENABLE_RSS_QUEUES,
+ &cmd, 0);
+ if (status != 0) {
+ dev_err(&mgp->pdev->dev,
+ "failed to set number of slices\n");
+
+ return status;
+ }
+ }
+ for (i = 0; i < mgp->num_slices; i++) {
+ ss = &mgp->ss[i];
+ cmd.data0 = MYRI10GE_LOWPART_TO_U32(ss->rx_done.bus);
+ cmd.data1 = MYRI10GE_HIGHPART_TO_U32(ss->rx_done.bus);
+ cmd.data2 = i;
+ status |= myri10ge_send_cmd(mgp, MXGEFW_CMD_SET_INTRQ_DMA,
+ &cmd, 0);
+ };
status |=
myri10ge_send_cmd(mgp, MXGEFW_CMD_GET_IRQ_ACK_OFFSET, &cmd, 0);
- mgp->ss.irq_claim = (__iomem __be32 *) (mgp->sram + cmd.data0);
+ for (i = 0; i < mgp->num_slices; i++) {
+ ss = &mgp->ss[i];
+ ss->irq_claim =
+ (__iomem __be32 *) (mgp->sram + cmd.data0 + 8 * i);
+ }
status |= myri10ge_send_cmd(mgp, MXGEFW_CMD_GET_IRQ_DEASSERT_OFFSET,
&cmd, 0);
mgp->irq_deassert = (__iomem __be32 *) (mgp->sram + cmd.data0);
@@ -899,18 +970,25 @@
}
put_be32(htonl(mgp->intr_coal_delay), mgp->intr_coal_delay_ptr);
- memset(mgp->ss.rx_done.entry, 0, bytes);
-
/* reset mcp/driver shared state back to 0 */
- mgp->ss.tx.req = 0;
- mgp->ss.tx.done = 0;
- mgp->ss.tx.pkt_start = 0;
- mgp->ss.tx.pkt_done = 0;
- mgp->ss.rx_big.cnt = 0;
- mgp->ss.rx_small.cnt = 0;
- mgp->ss.rx_done.idx = 0;
- mgp->ss.rx_done.cnt = 0;
+
mgp->link_changes = 0;
+ for (i = 0; i < mgp->num_slices; i++) {
+ ss = &mgp->ss[i];
+
+ memset(ss->rx_done.entry, 0, bytes);
+ ss->tx.req = 0;
+ ss->tx.done = 0;
+ ss->tx.pkt_start = 0;
+ ss->tx.pkt_done = 0;
+ ss->rx_big.cnt = 0;
+ ss->rx_small.cnt = 0;
+ ss->rx_done.idx = 0;
+ ss->rx_done.cnt = 0;
+ ss->tx.wake_queue = 0;
+ ss->tx.stop_queue = 0;
+ }
+
status = myri10ge_update_mac_address(mgp, mgp->dev->dev_addr);
myri10ge_change_pause(mgp, mgp->pause);
myri10ge_set_multicast_list(mgp->dev);
@@ -1095,9 +1173,10 @@
rx_frags[0].size -= MXGEFW_PAD;
len -= MXGEFW_PAD;
lro_receive_frags(&ss->rx_done.lro_mgr, rx_frags,
- len, len,
/* opaque, will come back in get_frag_header */
+ len, len,
(void *)(__force unsigned long)csum, csum);
+
return 1;
}
@@ -1236,7 +1315,7 @@
static inline void myri10ge_check_statblock(struct myri10ge_priv *mgp)
{
- struct mcp_irq_data *stats = mgp->ss.fw_stats;
+ struct mcp_irq_data *stats = mgp->ss[0].fw_stats;
if (unlikely(stats->stats_updated)) {
unsigned link_up = ntohl(stats->link_up);
@@ -1302,6 +1381,13 @@
u32 send_done_count;
int i;
+ /* an interrupt on a non-zero slice is implicitly valid
+ * since MSI-X irqs are not shared */
+ if (ss != mgp->ss) {
+ netif_rx_schedule(ss->dev, &ss->napi);
+ return (IRQ_HANDLED);
+ }
+
/* make sure it is our IRQ, and that the DMA has finished */
if (unlikely(!stats->valid))
return (IRQ_NONE);
@@ -1311,7 +1397,7 @@
if (stats->valid & 1)
netif_rx_schedule(ss->dev, &ss->napi);
- if (!mgp->msi_enabled) {
+ if (!mgp->msi_enabled && !mgp->msix_enabled) {
put_be32(0, mgp->irq_deassert);
if (!myri10ge_deassert_wait)
stats->valid = 0;
@@ -1446,10 +1532,10 @@
{
struct myri10ge_priv *mgp = netdev_priv(netdev);
- ring->rx_mini_max_pending = mgp->ss.rx_small.mask + 1;
- ring->rx_max_pending = mgp->ss.rx_big.mask + 1;
+ ring->rx_mini_max_pending = mgp->ss[0].rx_small.mask + 1;
+ ring->rx_max_pending = mgp->ss[0].rx_big.mask + 1;
ring->rx_jumbo_max_pending = 0;
- ring->tx_max_pending = mgp->ss.rx_small.mask + 1;
+ ring->tx_max_pending = mgp->ss[0].rx_small.mask + 1;
ring->rx_mini_pending = ring->rx_mini_max_pending;
ring->rx_pending = ring->rx_max_pending;
ring->rx_jumbo_pending = ring->rx_jumbo_max_pending;
@@ -1497,7 +1583,7 @@
"tx_aborted_errors", "tx_carrier_errors", "tx_fifo_errors",
"tx_heartbeat_errors", "tx_window_errors",
/* device-specific stats */
- "tx_boundary", "WC", "irq", "MSI",
+ "tx_boundary", "WC", "irq", "MSI", "MSIX",
"read_dma_bw_MBs", "write_dma_bw_MBs", "read_write_dma_bw_MBs",
"serial_number", "watchdog_resets",
"link_changes", "link_up", "dropped_link_overflow",
@@ -1524,23 +1610,31 @@
static void
myri10ge_get_strings(struct net_device *netdev, u32 stringset, u8 * data)
{
+ struct myri10ge_priv *mgp = netdev_priv(netdev);
+ int i;
+
switch (stringset) {
case ETH_SS_STATS:
memcpy(data, *myri10ge_gstrings_main_stats,
sizeof(myri10ge_gstrings_main_stats));
data += sizeof(myri10ge_gstrings_main_stats);
- memcpy(data, *myri10ge_gstrings_slice_stats,
- sizeof(myri10ge_gstrings_slice_stats));
- data += sizeof(myri10ge_gstrings_slice_stats);
+ for (i = 0; i < mgp->num_slices; i++) {
+ memcpy(data, *myri10ge_gstrings_slice_stats,
+ sizeof(myri10ge_gstrings_slice_stats));
+ data += sizeof(myri10ge_gstrings_slice_stats);
+ }
break;
}
}
static int myri10ge_get_sset_count(struct net_device *netdev, int sset)
{
+ struct myri10ge_priv *mgp = netdev_priv(netdev);
+
switch (sset) {
case ETH_SS_STATS:
- return MYRI10GE_MAIN_STATS_LEN + MYRI10GE_SLICE_STATS_LEN;
+ return MYRI10GE_MAIN_STATS_LEN +
+ mgp->num_slices * MYRI10GE_SLICE_STATS_LEN;
default:
return -EOPNOTSUPP;
}
@@ -1552,6 +1646,7 @@
{
struct myri10ge_priv *mgp = netdev_priv(netdev);
struct myri10ge_slice_state *ss;
+ int slice;
int i;
for (i = 0; i < MYRI10GE_NET_STATS_LEN; i++)
@@ -1561,6 +1656,7 @@
data[i++] = (unsigned int)mgp->wc_enabled;
data[i++] = (unsigned int)mgp->pdev->irq;
data[i++] = (unsigned int)mgp->msi_enabled;
+ data[i++] = (unsigned int)mgp->msix_enabled;
data[i++] = (unsigned int)mgp->read_dma;
data[i++] = (unsigned int)mgp->write_dma;
data[i++] = (unsigned int)mgp->read_write_dma;
@@ -1569,7 +1665,7 @@
data[i++] = (unsigned int)mgp->link_changes;
/* firmware stats are useful only in the first slice */
- ss = &mgp->ss;
+ ss = &mgp->ss[0];
data[i++] = (unsigned int)ntohl(ss->fw_stats->link_up);
data[i++] = (unsigned int)ntohl(ss->fw_stats->dropped_link_overflow);
data[i++] =
@@ -1585,24 +1681,27 @@
data[i++] = (unsigned int)ntohl(ss->fw_stats->dropped_no_small_buffer);
data[i++] = (unsigned int)ntohl(ss->fw_stats->dropped_no_big_buffer);
- data[i++] = 0;
- data[i++] = (unsigned int)ss->tx.pkt_start;
- data[i++] = (unsigned int)ss->tx.pkt_done;
- data[i++] = (unsigned int)ss->tx.req;
- data[i++] = (unsigned int)ss->tx.done;
- data[i++] = (unsigned int)ss->rx_small.cnt;
- data[i++] = (unsigned int)ss->rx_big.cnt;
- data[i++] = (unsigned int)ss->tx.wake_queue;
- data[i++] = (unsigned int)ss->tx.stop_queue;
- data[i++] = (unsigned int)ss->tx.linearized;
- data[i++] = ss->rx_done.lro_mgr.stats.aggregated;
- data[i++] = ss->rx_done.lro_mgr.stats.flushed;
- if (ss->rx_done.lro_mgr.stats.flushed)
- data[i++] = ss->rx_done.lro_mgr.stats.aggregated /
- ss->rx_done.lro_mgr.stats.flushed;
- else
- data[i++] = 0;
- data[i++] = ss->rx_done.lro_mgr.stats.no_desc;
+ for (slice = 0; slice < mgp->num_slices; slice++) {
+ ss = &mgp->ss[slice];
+ data[i++] = slice;
+ data[i++] = (unsigned int)ss->tx.pkt_start;
+ data[i++] = (unsigned int)ss->tx.pkt_done;
+ data[i++] = (unsigned int)ss->tx.req;
+ data[i++] = (unsigned int)ss->tx.done;
+ data[i++] = (unsigned int)ss->rx_small.cnt;
+ data[i++] = (unsigned int)ss->rx_big.cnt;
+ data[i++] = (unsigned int)ss->tx.wake_queue;
+ data[i++] = (unsigned int)ss->tx.stop_queue;
+ data[i++] = (unsigned int)ss->tx.linearized;
+ data[i++] = ss->rx_done.lro_mgr.stats.aggregated;
+ data[i++] = ss->rx_done.lro_mgr.stats.flushed;
+ if (ss->rx_done.lro_mgr.stats.flushed)
+ data[i++] = ss->rx_done.lro_mgr.stats.aggregated /
+ ss->rx_done.lro_mgr.stats.flushed;
+ else
+ data[i++] = 0;
+ data[i++] = ss->rx_done.lro_mgr.stats.no_desc;
+ }
}
static void myri10ge_set_msglevel(struct net_device *netdev, u32 value)
@@ -1645,12 +1744,15 @@
struct net_device *dev = mgp->dev;
int tx_ring_size, rx_ring_size;
int tx_ring_entries, rx_ring_entries;
- int i, status;
+ int i, slice, status;
size_t bytes;
/* get ring sizes */
+ slice = ss - mgp->ss;
+ cmd.data0 = slice;
status = myri10ge_send_cmd(mgp, MXGEFW_CMD_GET_SEND_RING_SIZE, &cmd, 0);
tx_ring_size = cmd.data0;
+ cmd.data0 = slice;
status |= myri10ge_send_cmd(mgp, MXGEFW_CMD_GET_RX_RING_SIZE, &cmd, 0);
if (status != 0)
return status;
@@ -1715,15 +1817,17 @@
mgp->small_bytes + MXGEFW_PAD, 0);
if (ss->rx_small.fill_cnt < ss->rx_small.mask + 1) {
- printk(KERN_ERR "myri10ge: %s: alloced only %d small bufs\n",
- dev->name, ss->rx_small.fill_cnt);
+ printk(KERN_ERR
+ "myri10ge: %s:slice-%d: alloced only %d small bufs\n",
+ dev->name, slice, ss->rx_small.fill_cnt);
goto abort_with_rx_small_ring;
}
myri10ge_alloc_rx_pages(mgp, &ss->rx_big, mgp->big_bytes, 0);
if (ss->rx_big.fill_cnt < ss->rx_big.mask + 1) {
- printk(KERN_ERR "myri10ge: %s: alloced only %d big bufs\n",
- dev->name, ss->rx_big.fill_cnt);
+ printk(KERN_ERR
+ "myri10ge: %s:slice-%d: alloced only %d big bufs\n",
+ dev->name, slice, ss->rx_big.fill_cnt);
goto abort_with_rx_big_ring;
}
@@ -1775,6 +1879,10 @@
struct myri10ge_tx_buf *tx;
int i, len, idx;
+ /* If not allocated, skip it */
+ if (ss->tx.req_list == NULL)
+ return;
+
for (i = ss->rx_big.cnt; i < ss->rx_big.fill_cnt; i++) {
idx = i & ss->rx_big.mask;
if (i == ss->rx_big.fill_cnt - 1)
@@ -1837,25 +1945,67 @@
static int myri10ge_request_irq(struct myri10ge_priv *mgp)
{
struct pci_dev *pdev = mgp->pdev;
+ struct myri10ge_slice_state *ss;
+ struct net_device *netdev = mgp->dev;
+ int i;
int status;
+ mgp->msi_enabled = 0;
+ mgp->msix_enabled = 0;
+ status = 0;
if (myri10ge_msi) {
- status = pci_enable_msi(pdev);
- if (status != 0)
- dev_err(&pdev->dev,
- "Error %d setting up MSI; falling back to xPIC\n",
- status);
- else
- mgp->msi_enabled = 1;
- } else {
- mgp->msi_enabled = 0;
+ if (mgp->num_slices > 1) {
+ status =
+ pci_enable_msix(pdev, mgp->msix_vectors,
+ mgp->num_slices);
+ if (status == 0) {
+ mgp->msix_enabled = 1;
+ } else {
+ dev_err(&pdev->dev,
+ "Error %d setting up MSI-X\n", status);
+ return status;
+ }
+ }
+ if (mgp->msix_enabled == 0) {
+ status = pci_enable_msi(pdev);
+ if (status != 0) {
+ dev_err(&pdev->dev,
+ "Error %d setting up MSI; falling back to xPIC\n",
+ status);
+ } else {
+ mgp->msi_enabled = 1;
+ }
+ }
}
- status = request_irq(pdev->irq, myri10ge_intr, IRQF_SHARED,
- mgp->dev->name, mgp);
- if (status != 0) {
- dev_err(&pdev->dev, "failed to allocate IRQ\n");
- if (mgp->msi_enabled)
- pci_disable_msi(pdev);
+ if (mgp->msix_enabled) {
+ for (i = 0; i < mgp->num_slices; i++) {
+ ss = &mgp->ss[i];
+ snprintf(ss->irq_desc, sizeof(ss->irq_desc),
+ "%s:slice-%d", netdev->name, i);
+ status = request_irq(mgp->msix_vectors[i].vector,
+ myri10ge_intr, 0, ss->irq_desc,
+ ss);
+ if (status != 0) {
+ dev_err(&pdev->dev,
+ "slice %d failed to allocate IRQ\n", i);
+ i--;
+ while (i >= 0) {
+ free_irq(mgp->msix_vectors[i].vector,
+ &mgp->ss[i]);
+ i--;
+ }
+ pci_disable_msix(pdev);
+ return status;
+ }
+ }
+ } else {
+ status = request_irq(pdev->irq, myri10ge_intr, IRQF_SHARED,
+ mgp->dev->name, &mgp->ss[0]);
+ if (status != 0) {
+ dev_err(&pdev->dev, "failed to allocate IRQ\n");
+ if (mgp->msi_enabled)
+ pci_disable_msi(pdev);
+ }
}
return status;
}
@@ -1863,10 +2013,18 @@
static void myri10ge_free_irq(struct myri10ge_priv *mgp)
{
struct pci_dev *pdev = mgp->pdev;
+ int i;
- free_irq(pdev->irq, mgp);
+ if (mgp->msix_enabled) {
+ for (i = 0; i < mgp->num_slices; i++)
+ free_irq(mgp->msix_vectors[i].vector, &mgp->ss[i]);
+ } else {
+ free_irq(pdev->irq, &mgp->ss[0]);
+ }
if (mgp->msi_enabled)
pci_disable_msi(pdev);
+ if (mgp->msix_enabled)
+ pci_disable_msix(pdev);
}
static int
@@ -1928,7 +2086,6 @@
return 0;
}
-#if 0
static int myri10ge_get_txrx(struct myri10ge_priv *mgp, int slice)
{
struct myri10ge_cmd cmd;
@@ -1996,14 +2153,15 @@
}
return 0;
}
-#endif
static int myri10ge_open(struct net_device *dev)
{
+ struct myri10ge_slice_state *ss;
struct myri10ge_priv *mgp = netdev_priv(dev);
struct myri10ge_cmd cmd;
+ int i, status, big_pow2, slice;
+ u8 *itable;
struct net_lro_mgr *lro_mgr;
- int status, big_pow2;
if (mgp->running != MYRI10GE_ETH_STOPPED)
return -EBUSY;
@@ -2015,6 +2173,48 @@
goto abort_with_nothing;
}
+ if (mgp->num_slices > 1) {
+ cmd.data0 = mgp->num_slices;
+ cmd.data1 = 1; /* use MSI-X */
+ status = myri10ge_send_cmd(mgp, MXGEFW_CMD_ENABLE_RSS_QUEUES,
+ &cmd, 0);
+ if (status != 0) {
+ printk(KERN_ERR
+ "myri10ge: %s: failed to set number of slices\n",
+ dev->name);
+ goto abort_with_nothing;
+ }
+ /* setup the indirection table */
+ cmd.data0 = mgp->num_slices;
+ status = myri10ge_send_cmd(mgp, MXGEFW_CMD_SET_RSS_TABLE_SIZE,
+ &cmd, 0);
+
+ status |= myri10ge_send_cmd(mgp,
+ MXGEFW_CMD_GET_RSS_TABLE_OFFSET,
+ &cmd, 0);
+ if (status != 0) {
+ printk(KERN_ERR
+ "myri10ge: %s: failed to setup rss tables\n",
+ dev->name);
+ }
+
+ /* just enable an identity mapping */
+ itable = mgp->sram + cmd.data0;
+ for (i = 0; i < mgp->num_slices; i++)
+ __raw_writeb(i, &itable[i]);
+
+ cmd.data0 = 1;
+ cmd.data1 = myri10ge_rss_hash;
+ status = myri10ge_send_cmd(mgp, MXGEFW_CMD_SET_RSS_ENABLE,
+ &cmd, 0);
+ if (status != 0) {
+ printk(KERN_ERR
+ "myri10ge: %s: failed to enable slices\n",
+ dev->name);
+ goto abort_with_nothing;
+ }
+ }
+
status = myri10ge_request_irq(mgp);
if (status != 0)
goto abort_with_nothing;
@@ -2038,41 +2238,6 @@
if (myri10ge_small_bytes > 0)
mgp->small_bytes = myri10ge_small_bytes;
- /* get the lanai pointers to the send and receive rings */
-
- status |= myri10ge_send_cmd(mgp, MXGEFW_CMD_GET_SEND_OFFSET, &cmd, 0);
- mgp->ss.tx.lanai =
- (struct mcp_kreq_ether_send __iomem *)(mgp->sram + cmd.data0);
-
- status |=
- myri10ge_send_cmd(mgp, MXGEFW_CMD_GET_SMALL_RX_OFFSET, &cmd, 0);
- mgp->ss.rx_small.lanai =
- (struct mcp_kreq_ether_recv __iomem *)(mgp->sram + cmd.data0);
-
- status |= myri10ge_send_cmd(mgp, MXGEFW_CMD_GET_BIG_RX_OFFSET, &cmd, 0);
- mgp->ss.rx_big.lanai =
- (struct mcp_kreq_ether_recv __iomem *)(mgp->sram + cmd.data0);
-
- if (status != 0) {
- printk(KERN_ERR
- "myri10ge: %s: failed to get ring sizes or locations\n",
- dev->name);
- mgp->running = MYRI10GE_ETH_STOPPED;
- goto abort_with_irq;
- }
-
- if (myri10ge_wcfifo && mgp->wc_enabled) {
- mgp->ss.tx.wc_fifo = (u8 __iomem *) mgp->sram + MXGEFW_ETH_SEND_4;
- mgp->ss.rx_small.wc_fifo =
- (u8 __iomem *) mgp->sram + MXGEFW_ETH_RECV_SMALL;
- mgp->ss.rx_big.wc_fifo =
- (u8 __iomem *) mgp->sram + MXGEFW_ETH_RECV_BIG;
- } else {
- mgp->ss.tx.wc_fifo = NULL;
- mgp->ss.rx_small.wc_fifo = NULL;
- mgp->ss.rx_big.wc_fifo = NULL;
- }
-
/* Firmware needs the big buff size as a power of 2. Lie and
* tell him the buffer is larger, because we only use 1
* buffer/pkt, and the mtu will prevent overruns.
@@ -2087,9 +2252,44 @@
mgp->big_bytes = big_pow2;
}
- status = myri10ge_allocate_rings(&mgp->ss);
- if (status != 0)
- goto abort_with_irq;
+ /* setup the per-slice data structures */
+ for (slice = 0; slice < mgp->num_slices; slice++) {
+ ss = &mgp->ss[slice];
+
+ status = myri10ge_get_txrx(mgp, slice);
+ if (status != 0) {
+ printk(KERN_ERR
+ "myri10ge: %s: failed to get ring sizes or locations\n",
+ dev->name);
+ goto abort_with_rings;
+ }
+ status = myri10ge_allocate_rings(ss);
+ if (status != 0)
+ goto abort_with_rings;
+ if (slice == 0)
+ status = myri10ge_set_stats(mgp, slice);
+ if (status) {
+ printk(KERN_ERR
+ "myri10ge: %s: Couldn't set stats DMA\n",
+ dev->name);
+ goto abort_with_rings;
+ }
+
+ lro_mgr = &ss->rx_done.lro_mgr;
+ lro_mgr->dev = dev;
+ lro_mgr->features = LRO_F_NAPI;
+ lro_mgr->ip_summed = CHECKSUM_COMPLETE;
+ lro_mgr->ip_summed_aggr = CHECKSUM_UNNECESSARY;
+ lro_mgr->max_desc = MYRI10GE_MAX_LRO_DESCRIPTORS;
+ lro_mgr->lro_arr = ss->rx_done.lro_desc;
+ lro_mgr->get_frag_header = myri10ge_get_frag_header;
+ lro_mgr->max_aggr = myri10ge_lro_max_pkts;
+ if (lro_mgr->max_aggr > MAX_SKB_FRAGS)
+ lro_mgr->max_aggr = MAX_SKB_FRAGS;
+
+ /* must happen prior to any irq */
+ napi_enable(&(ss)->napi);
+ }
/* now give firmware buffers sizes, and MTU */
cmd.data0 = dev->mtu + ETH_HLEN + VLAN_HLEN;
@@ -2106,25 +2306,15 @@
goto abort_with_rings;
}
- cmd.data0 = MYRI10GE_LOWPART_TO_U32(mgp->ss.fw_stats_bus);
- cmd.data1 = MYRI10GE_HIGHPART_TO_U32(mgp->ss.fw_stats_bus);
- cmd.data2 = sizeof(struct mcp_irq_data);
- status = myri10ge_send_cmd(mgp, MXGEFW_CMD_SET_STATS_DMA_V2, &cmd, 0);
- if (status == -ENOSYS) {
- dma_addr_t bus = mgp->ss.fw_stats_bus;
- bus += offsetof(struct mcp_irq_data, send_done_count);
- cmd.data0 = MYRI10GE_LOWPART_TO_U32(bus);
- cmd.data1 = MYRI10GE_HIGHPART_TO_U32(bus);
- status = myri10ge_send_cmd(mgp,
- MXGEFW_CMD_SET_STATS_DMA_OBSOLETE,
- &cmd, 0);
- /* Firmware cannot support multicast without STATS_DMA_V2 */
- mgp->fw_multicast_support = 0;
- } else {
- mgp->fw_multicast_support = 1;
- }
- if (status) {
- printk(KERN_ERR "myri10ge: %s: Couldn't set stats DMA\n",
+ /*
+ * Set Linux style TSO mode; this is needed only on newer
+ * firmware versions. Older versions default to Linux
+ * style TSO
+ */
+ cmd.data0 = 0;
+ status = myri10ge_send_cmd(mgp, MXGEFW_CMD_SET_TSO_MODE, &cmd, 0);
+ if (status && status != -ENOSYS) {
+ printk(KERN_ERR "myri10ge: %s: Couldn't set TSO mode\n",
dev->name);
goto abort_with_rings;
}
@@ -2132,21 +2322,6 @@
mgp->link_state = ~0U;
mgp->rdma_tags_available = 15;
- lro_mgr = &mgp->ss.rx_done.lro_mgr;
- lro_mgr->dev = dev;
- lro_mgr->features = LRO_F_NAPI;
- lro_mgr->ip_summed = CHECKSUM_COMPLETE;
- lro_mgr->ip_summed_aggr = CHECKSUM_UNNECESSARY;
- lro_mgr->max_desc = MYRI10GE_MAX_LRO_DESCRIPTORS;
- lro_mgr->lro_arr = mgp->ss.rx_done.lro_desc;
- lro_mgr->get_frag_header = myri10ge_get_frag_header;
- lro_mgr->max_aggr = myri10ge_lro_max_pkts;
- lro_mgr->frag_align_pad = 2;
- if (lro_mgr->max_aggr > MAX_SKB_FRAGS)
- lro_mgr->max_aggr = MAX_SKB_FRAGS;
-
- napi_enable(&mgp->ss.napi); /* must happen prior to any irq */
-
status = myri10ge_send_cmd(mgp, MXGEFW_CMD_ETHERNET_UP, &cmd, 0);
if (status) {
printk(KERN_ERR "myri10ge: %s: Couldn't bring up link\n",
@@ -2154,8 +2329,6 @@
goto abort_with_rings;
}
- mgp->ss.tx.wake_queue = 0;
- mgp->ss.tx.stop_queue = 0;
mgp->running = MYRI10GE_ETH_RUNNING;
mgp->watchdog_timer.expires = jiffies + myri10ge_watchdog_timeout * HZ;
add_timer(&mgp->watchdog_timer);
@@ -2163,9 +2336,9 @@
return 0;
abort_with_rings:
- myri10ge_free_rings(&mgp->ss);
+ for (i = 0; i < mgp->num_slices; i++)
+ myri10ge_free_rings(&mgp->ss[i]);
-abort_with_irq:
myri10ge_free_irq(mgp);
abort_with_nothing:
@@ -2178,16 +2351,19 @@
struct myri10ge_priv *mgp = netdev_priv(dev);
struct myri10ge_cmd cmd;
int status, old_down_cnt;
+ int i;
if (mgp->running != MYRI10GE_ETH_RUNNING)
return 0;
- if (mgp->ss.tx.req_bytes == NULL)
+ if (mgp->ss[0].tx.req_bytes == NULL)
return 0;
del_timer_sync(&mgp->watchdog_timer);
mgp->running = MYRI10GE_ETH_STOPPING;
- napi_disable(&mgp->ss.napi);
+ for (i = 0; i < mgp->num_slices; i++) {
+ napi_disable(&mgp->ss[i].napi);
+ }
netif_carrier_off(dev);
netif_stop_queue(dev);
old_down_cnt = mgp->down_cnt;
@@ -2203,7 +2379,8 @@
netif_tx_disable(dev);
myri10ge_free_irq(mgp);
- myri10ge_free_rings(&mgp->ss);
+ for (i = 0; i < mgp->num_slices; i++)
+ myri10ge_free_rings(&mgp->ss[i]);
mgp->running = MYRI10GE_ETH_STOPPED;
return 0;
@@ -2324,7 +2501,7 @@
u8 flags, odd_flag;
/* always transmit through slot 0 */
- ss = &mgp->ss;
+ ss = mgp->ss;
tx = &ss->tx;
again:
req = tx->req_list;
@@ -2629,7 +2806,21 @@
static struct net_device_stats *myri10ge_get_stats(struct net_device *dev)
{
struct myri10ge_priv *mgp = netdev_priv(dev);
- return &mgp->stats;
+ struct myri10ge_slice_netstats *slice_stats;
+ struct net_device_stats *stats = &mgp->stats;
+ int i;
+
+ memset(stats, 0, sizeof(*stats));
+ for (i = 0; i < mgp->num_slices; i++) {
+ slice_stats = &mgp->ss[i].stats;
+ stats->rx_packets += slice_stats->rx_packets;
+ stats->tx_packets += slice_stats->tx_packets;
+ stats->rx_bytes += slice_stats->rx_bytes;
+ stats->tx_bytes += slice_stats->tx_bytes;
+ stats->rx_dropped += slice_stats->rx_dropped;
+ stats->tx_dropped += slice_stats->tx_dropped;
+ }
+ return stats;
}
static void myri10ge_set_multicast_list(struct net_device *dev)
@@ -2840,10 +3031,10 @@
*
* If the driver can neither enable ECRC nor verify that it has
* already been enabled, then it must use a firmware image which works
- * around unaligned completion packets (myri10ge_ethp_z8e.dat), and it
+ * around unaligned completion packets (myri10ge_rss_ethp_z8e.dat), and it
* should also ensure that it never gives the device a Read-DMA which is
* larger than 2KB by setting the tx_boundary to 2KB. If ECRC is
- * enabled, then the driver should use the aligned (myri10ge_eth_z8e.dat)
+ * enabled, then the driver should use the aligned (myri10ge_rss_eth_z8e.dat)
* firmware image, and set tx_boundary to 4KB.
*/
@@ -2872,7 +3063,7 @@
* completions) in order to see if it works on this host.
*/
mgp->fw_name = myri10ge_fw_aligned;
- status = myri10ge_load_firmware(mgp);
+ status = myri10ge_load_firmware(mgp, 1);
if (status != 0) {
goto abort;
}
@@ -3053,6 +3244,7 @@
struct myri10ge_tx_buf *tx;
u32 reboot;
int status;
+ int i;
u16 cmd, vendor;
mgp->watchdog_resets++;
@@ -3100,20 +3292,26 @@
printk(KERN_ERR "myri10ge: %s: device timeout, resetting\n",
mgp->dev->name);
- tx = &mgp->ss.tx;
- printk(KERN_INFO "myri10ge: %s: %d %d %d %d %d\n",
- mgp->dev->name, tx->req, tx->done,
- tx->pkt_start, tx->pkt_done,
- (int)ntohl(mgp->ss.fw_stats->send_done_count));
- msleep(2000);
- printk(KERN_INFO "myri10ge: %s: %d %d %d %d %d\n",
- mgp->dev->name, tx->req, tx->done,
- tx->pkt_start, tx->pkt_done,
- (int)ntohl(mgp->ss.fw_stats->send_done_count));
+ for (i = 0; i < mgp->num_slices; i++) {
+ tx = &mgp->ss[i].tx;
+ printk(KERN_INFO
+ "myri10ge: %s: (%d): %d %d %d %d %d\n",
+ mgp->dev->name, i, tx->req, tx->done,
+ tx->pkt_start, tx->pkt_done,
+ (int)ntohl(mgp->ss[i].fw_stats->
+ send_done_count));
+ msleep(2000);
+ printk(KERN_INFO
+ "myri10ge: %s: (%d): %d %d %d %d %d\n",
+ mgp->dev->name, i, tx->req, tx->done,
+ tx->pkt_start, tx->pkt_done,
+ (int)ntohl(mgp->ss[i].fw_stats->
+ send_done_count));
+ }
}
rtnl_lock();
myri10ge_close(mgp->dev);
- status = myri10ge_load_firmware(mgp);
+ status = myri10ge_load_firmware(mgp, 1);
if (status != 0)
printk(KERN_ERR "myri10ge: %s: failed to load firmware\n",
mgp->dev->name);
@@ -3133,50 +3331,59 @@
{
struct myri10ge_priv *mgp;
struct myri10ge_slice_state *ss;
+ int i, reset_needed;
u32 rx_pause_cnt;
mgp = (struct myri10ge_priv *)arg;
- rx_pause_cnt = ntohl(mgp->ss.fw_stats->dropped_pause);
+ rx_pause_cnt = ntohl(mgp->ss[0].fw_stats->dropped_pause);
+ for (i = 0, reset_needed = 0;
+ i < mgp->num_slices && reset_needed == 0; ++i) {
- ss = &mgp->ss;
- if (ss->rx_small.watchdog_needed) {
- myri10ge_alloc_rx_pages(mgp, &ss->rx_small,
- mgp->small_bytes + MXGEFW_PAD, 1);
- if (ss->rx_small.fill_cnt - ss->rx_small.cnt >=
- myri10ge_fill_thresh)
- ss->rx_small.watchdog_needed = 0;
- }
- if (ss->rx_big.watchdog_needed) {
- myri10ge_alloc_rx_pages(mgp, &ss->rx_big, mgp->big_bytes, 1);
- if (ss->rx_big.fill_cnt - ss->rx_big.cnt >=
- myri10ge_fill_thresh)
- ss->rx_big.watchdog_needed = 0;
- }
-
- if (ss->tx.req != ss->tx.done &&
- ss->tx.done == ss->watchdog_tx_done &&
- ss->watchdog_tx_req != ss->watchdog_tx_done) {
- /* nic seems like it might be stuck.. */
- if (rx_pause_cnt != mgp->watchdog_pause) {
- if (net_ratelimit())
- printk(KERN_WARNING "myri10ge %s:"
- "TX paused, check link partner\n",
- mgp->dev->name);
- } else {
- schedule_work(&mgp->watchdog_work);
- return;
+ ss = &mgp->ss[i];
+ if (ss->rx_small.watchdog_needed) {
+ myri10ge_alloc_rx_pages(mgp, &ss->rx_small,
+ mgp->small_bytes + MXGEFW_PAD,
+ 1);
+ if (ss->rx_small.fill_cnt - ss->rx_small.cnt >=
+ myri10ge_fill_thresh)
+ ss->rx_small.watchdog_needed = 0;
}
+ if (ss->rx_big.watchdog_needed) {
+ myri10ge_alloc_rx_pages(mgp, &ss->rx_big,
+ mgp->big_bytes, 1);
+ if (ss->rx_big.fill_cnt - ss->rx_big.cnt >=
+ myri10ge_fill_thresh)
+ ss->rx_big.watchdog_needed = 0;
+ }
+
+ if (ss->tx.req != ss->tx.done &&
+ ss->tx.done == ss->watchdog_tx_done &&
+ ss->watchdog_tx_req != ss->watchdog_tx_done) {
+ /* nic seems like it might be stuck.. */
+ if (rx_pause_cnt != mgp->watchdog_pause) {
+ if (net_ratelimit())
+ printk(KERN_WARNING "myri10ge %s:"
+ "TX paused, check link partner\n",
+ mgp->dev->name);
+ } else {
+ reset_needed = 1;
+ }
+ }
+ ss->watchdog_tx_done = ss->tx.done;
+ ss->watchdog_tx_req = ss->tx.req;
}
- /* rearm timer */
- mod_timer(&mgp->watchdog_timer,
- jiffies + myri10ge_watchdog_timeout * HZ);
- ss->watchdog_tx_done = ss->tx.done;
- ss->watchdog_tx_req = ss->tx.req;
mgp->watchdog_pause = rx_pause_cnt;
+
+ if (reset_needed) {
+ schedule_work(&mgp->watchdog_work);
+ } else {
+ /* rearm timer */
+ mod_timer(&mgp->watchdog_timer,
+ jiffies + myri10ge_watchdog_timeout * HZ);
+ }
}
-#if 0
static void myri10ge_free_slices(struct myri10ge_priv *mgp)
{
struct myri10ge_slice_state *ss;
@@ -3360,14 +3567,12 @@
mgp->fw_name = old_fw;
myri10ge_load_firmware(mgp, 0);
}
-#endif
static int myri10ge_probe(struct pci_dev *pdev, const struct pci_device_id *ent)
{
struct net_device *netdev;
struct myri10ge_priv *mgp;
struct device *dev = &pdev->dev;
- size_t bytes;
int i;
int status = -ENXIO;
int dac_enabled;
@@ -3382,7 +3587,6 @@
mgp = netdev_priv(netdev);
mgp->dev = netdev;
- netif_napi_add(netdev, &mgp->ss.napi, myri10ge_poll, myri10ge_napi_weight);
mgp->pdev = pdev;
mgp->csum_flag = MXGEFW_FLAGS_CKSUM;
mgp->pause = myri10ge_flow_control;
@@ -3428,11 +3632,6 @@
if (mgp->cmd == NULL)
goto abort_with_netdev;
- mgp->ss.fw_stats = dma_alloc_coherent(&pdev->dev, sizeof(*mgp->ss.fw_stats),
- &mgp->ss.fw_stats_bus, GFP_KERNEL);
- if (mgp->ss.fw_stats == NULL)
- goto abort_with_cmd;
-
mgp->board_span = pci_resource_len(pdev, 0);
mgp->iomem_base = pci_resource_start(pdev, 0);
mgp->mtrr = -1;
@@ -3469,26 +3668,24 @@
for (i = 0; i < ETH_ALEN; i++)
netdev->dev_addr[i] = mgp->mac_addr[i];
- /* allocate rx done ring */
- bytes = mgp->max_intr_slots * sizeof(*mgp->ss.rx_done.entry);
- mgp->ss.rx_done.entry = dma_alloc_coherent(&pdev->dev, bytes,
- &mgp->ss.rx_done.bus, GFP_KERNEL);
- if (mgp->ss.rx_done.entry == NULL)
- goto abort_with_ioremap;
- memset(mgp->ss.rx_done.entry, 0, bytes);
-
myri10ge_select_firmware(mgp);
- status = myri10ge_load_firmware(mgp);
+ status = myri10ge_load_firmware(mgp, 1);
if (status != 0) {
dev_err(&pdev->dev, "failed to load firmware\n");
- goto abort_with_rx_done;
+ goto abort_with_ioremap;
+ }
+ myri10ge_probe_slices(mgp);
+ status = myri10ge_alloc_slices(mgp);
+ if (status != 0) {
+ dev_err(&pdev->dev, "failed to alloc slice state\n");
+ goto abort_with_firmware;
}
status = myri10ge_reset(mgp);
if (status != 0) {
dev_err(&pdev->dev, "failed reset\n");
- goto abort_with_firmware;
+ goto abort_with_slices;
}
pci_set_drvdata(pdev, mgp);
@@ -3533,24 +3730,27 @@
dev_err(&pdev->dev, "register_netdev failed: %d\n", status);
goto abort_with_state;
}
- dev_info(dev, "%s IRQ %d, tx bndry %d, fw %s, WC %s\n",
- (mgp->msi_enabled ? "MSI" : "xPIC"),
- netdev->irq, mgp->tx_boundary, mgp->fw_name,
- (mgp->wc_enabled ? "Enabled" : "Disabled"));
+ if (mgp->msix_enabled)
+ dev_info(dev, "%d MSI-X IRQs, tx bndry %d, fw %s, WC %s\n",
+ mgp->num_slices, mgp->tx_boundary, mgp->fw_name,
+ (mgp->wc_enabled ? "Enabled" : "Disabled"));
+ else
+ dev_info(dev, "%s IRQ %d, tx bndry %d, fw %s, WC %s\n",
+ mgp->msi_enabled ? "MSI" : "xPIC",
+ netdev->irq, mgp->tx_boundary, mgp->fw_name,
+ (mgp->wc_enabled ? "Enabled" : "Disabled"));
return 0;
abort_with_state:
pci_restore_state(pdev);
+abort_with_slices:
+ myri10ge_free_slices(mgp);
+
abort_with_firmware:
myri10ge_dummy_rdma(mgp, 0);
-abort_with_rx_done:
- bytes = mgp->max_intr_slots * sizeof(*mgp->ss.rx_done.entry);
- dma_free_coherent(&pdev->dev, bytes,
- mgp->ss.rx_done.entry, mgp->ss.rx_done.bus);
-
abort_with_ioremap:
iounmap(mgp->sram);
@@ -3559,10 +3759,6 @@
if (mgp->mtrr >= 0)
mtrr_del(mgp->mtrr, mgp->iomem_base, mgp->board_span);
#endif
- dma_free_coherent(&pdev->dev, sizeof(*mgp->ss.fw_stats),
- mgp->ss.fw_stats, mgp->ss.fw_stats_bus);
-
-abort_with_cmd:
dma_free_coherent(&pdev->dev, sizeof(*mgp->cmd),
mgp->cmd, mgp->cmd_bus);
@@ -3583,7 +3779,6 @@
{
struct myri10ge_priv *mgp;
struct net_device *netdev;
- size_t bytes;
mgp = pci_get_drvdata(pdev);
if (mgp == NULL)
@@ -3598,19 +3793,15 @@
/* avoid a memory leak */
pci_restore_state(pdev);
- bytes = mgp->max_intr_slots * sizeof(*mgp->ss.rx_done.entry);
- dma_free_coherent(&pdev->dev, bytes,
- mgp->ss.rx_done.entry, mgp->ss.rx_done.bus);
-
iounmap(mgp->sram);
#ifdef CONFIG_MTRR
if (mgp->mtrr >= 0)
mtrr_del(mgp->mtrr, mgp->iomem_base, mgp->board_span);
#endif
- dma_free_coherent(&pdev->dev, sizeof(*mgp->ss.fw_stats),
- mgp->ss.fw_stats, mgp->ss.fw_stats_bus);
-
+ myri10ge_free_slices(mgp);
+ if (mgp->msix_vectors != NULL)
+ kfree(mgp->msix_vectors);
dma_free_coherent(&pdev->dev, sizeof(*mgp->cmd),
mgp->cmd, mgp->cmd_bus);
@@ -3643,6 +3834,15 @@
{
printk(KERN_INFO "%s: Version %s\n", myri10ge_driver.name,
MYRI10GE_VERSION_STR);
+
+ if (myri10ge_rss_hash > MXGEFW_RSS_HASH_TYPE_SRC_PORT ||
+ myri10ge_rss_hash < MXGEFW_RSS_HASH_TYPE_IPV4) {
+ printk(KERN_ERR
+ "%s: Illegal rssh hash type %d, defaulting to source port\n",
+ myri10ge_driver.name, myri10ge_rss_hash);
+ myri10ge_rss_hash = MXGEFW_RSS_HASH_TYPE_SRC_PORT;
+ }
+
return pci_register_driver(&myri10ge_driver);
}