* [RESEND NET-NEXT PATCH 08/20] igb: Introduce multiple TX queues with infrastructure
2008-07-08 22:06 [RESEND NET-NEXT PATCH 00/20] igb update for net-next Jeff Kirsher
@ 2008-07-08 22:06 ` Jeff Kirsher
2008-07-11 5:22 ` Jeff Garzik
2008-07-08 22:07 ` [RESEND NET-NEXT PATCH 09/20] igb: update ethtool stats to support multiqueue Jeff Kirsher
` (11 subsequent siblings)
12 siblings, 1 reply; 15+ messages in thread
From: Jeff Kirsher @ 2008-07-08 22:06 UTC (permalink / raw)
To: jeff, davem
Cc: netdev, linux-kernel, akpm, Peter P Waskiewicz Jr, Mitch Williams,
Auke Kok
From: Peter P Waskiewicz Jr <peter.p.waskiewicz.jr@intel.com>
This code adds multiple Tx queue infrastructure much like we
previously did in ixgbe. The MSI-X vector mapping is the bulk of
the change.
IAM can now be safely enabled and we've verified that it does
work correctly. We can also eliminate the tx ring lock.
Signed-off-by: Peter P Waskiewicz Jr <peter.p.waskiewicz.jr@intel.com>
Signed-off-by: Mitch Williams <mitch.a.williams@intel.com>
Signed-off-by: Auke Kok <auke-jan.h.kok@intel.com>
---
drivers/net/igb/igb.h | 7 +-
drivers/net/igb/igb_main.c | 160 ++++++++++++++++++++++++++++++--------------
2 files changed, 114 insertions(+), 53 deletions(-)
diff --git a/drivers/net/igb/igb.h b/drivers/net/igb/igb.h
index 2c48eec..a1431c8 100644
--- a/drivers/net/igb/igb.h
+++ b/drivers/net/igb/igb.h
@@ -62,6 +62,7 @@ struct igb_adapter;
/* Transmit and receive queues */
#define IGB_MAX_RX_QUEUES 4
+#define IGB_MAX_TX_QUEUES 4
/* RX descriptor control thresholds.
* PTHRESH - MAC will consider prefetch if it has fewer than this number of
@@ -157,8 +158,6 @@ struct igb_ring {
union {
/* TX */
struct {
- spinlock_t tx_clean_lock;
- spinlock_t tx_lock;
bool detect_tx_hung;
};
/* RX */
@@ -277,6 +276,10 @@ struct igb_adapter {
/* for ioport free */
int bars;
int need_ioport;
+
+#ifdef CONFIG_NETDEVICES_MULTIQUEUE
+ struct igb_ring *multi_tx_table[IGB_MAX_TX_QUEUES];
+#endif /* CONFIG_NETDEVICES_MULTIQUEUE */
};
enum e1000_state_t {
diff --git a/drivers/net/igb/igb_main.c b/drivers/net/igb/igb_main.c
index afd4ce3..e11a5da 100644
--- a/drivers/net/igb/igb_main.c
+++ b/drivers/net/igb/igb_main.c
@@ -103,7 +103,7 @@ static irqreturn_t igb_msix_rx(int irq, void *);
static irqreturn_t igb_msix_tx(int irq, void *);
static int igb_clean_rx_ring_msix(struct napi_struct *, int);
static bool igb_clean_tx_irq(struct igb_ring *);
-static int igb_clean(struct napi_struct *, int);
+static int igb_poll(struct napi_struct *, int);
static bool igb_clean_rx_irq_adv(struct igb_ring *, int *, int);
static void igb_alloc_rx_buffers_adv(struct igb_ring *, int);
static int igb_ioctl(struct net_device *, struct ifreq *, int cmd);
@@ -224,6 +224,11 @@ static int igb_alloc_queues(struct igb_adapter *adapter)
return -ENOMEM;
}
+ for (i = 0; i < adapter->num_tx_queues; i++) {
+ struct igb_ring *ring = &(adapter->tx_ring[i]);
+ ring->adapter = adapter;
+ ring->queue_index = i;
+ }
for (i = 0; i < adapter->num_rx_queues; i++) {
struct igb_ring *ring = &(adapter->rx_ring[i]);
ring->adapter = adapter;
@@ -231,7 +236,7 @@ static int igb_alloc_queues(struct igb_adapter *adapter)
ring->itr_register = E1000_ITR;
/* set a default napi handler for each rx_ring */
- netif_napi_add(adapter->netdev, &ring->napi, igb_clean, 64);
+ netif_napi_add(adapter->netdev, &ring->napi, igb_poll, 64);
}
return 0;
}
@@ -412,8 +417,14 @@ static void igb_set_interrupt_capability(struct igb_adapter *adapter)
/* If we can't do MSI-X, try MSI */
msi_only:
adapter->num_rx_queues = 1;
+ adapter->num_tx_queues = 1;
if (!pci_enable_msi(adapter->pdev))
adapter->msi_enabled = 1;
+
+#ifdef CONFIG_NETDEVICES_MULTIQUEUE
+ /* Notify the stack of the (possibly) reduced Tx Queue count. */
+ adapter->netdev->egress_subqueue_count = adapter->num_tx_queues;
+#endif
return;
}
@@ -693,6 +704,10 @@ void igb_down(struct igb_adapter *adapter)
/* flush and sleep below */
netif_stop_queue(netdev);
+#ifdef CONFIG_NETDEVICES_MULTIQUEUE
+ for (i = 0; i < adapter->num_tx_queues; i++)
+ netif_stop_subqueue(netdev, i);
+#endif
/* disable transmits in the hardware */
tctl = rd32(E1000_TCTL);
@@ -895,7 +910,11 @@ static int __devinit igb_probe(struct pci_dev *pdev,
pci_save_state(pdev);
err = -ENOMEM;
+#ifdef CONFIG_NETDEVICES_MULTIQUEUE
+ netdev = alloc_etherdev_mq(sizeof(struct igb_adapter), IGB_MAX_TX_QUEUES);
+#else
netdev = alloc_etherdev(sizeof(struct igb_adapter));
+#endif /* CONFIG_NETDEVICES_MULTIQUEUE */
if (!netdev)
goto err_alloc_etherdev;
@@ -997,6 +1016,10 @@ static int __devinit igb_probe(struct pci_dev *pdev,
if (pci_using_dac)
netdev->features |= NETIF_F_HIGHDMA;
+#ifdef CONFIG_NETDEVICES_MULTIQUEUE
+ netdev->features |= NETIF_F_MULTI_QUEUE;
+#endif
+
netdev->features |= NETIF_F_LLTX;
adapter->en_mng_pt = igb_enable_mng_pass_thru(&adapter->hw);
@@ -1097,6 +1120,10 @@ static int __devinit igb_probe(struct pci_dev *pdev,
/* tell the stack to leave us alone until igb_open() is called */
netif_carrier_off(netdev);
netif_stop_queue(netdev);
+#ifdef CONFIG_NETDEVICES_MULTIQUEUE
+ for (i = 0; i < adapter->num_tx_queues; i++)
+ netif_stop_subqueue(netdev, i);
+#endif
strcpy(netdev->name, "eth%d");
err = register_netdev(netdev);
@@ -1223,9 +1250,15 @@ static int __devinit igb_sw_init(struct igb_adapter *adapter)
/* Number of supported queues. */
/* Having more queues than CPUs doesn't make sense. */
+ adapter->num_rx_queues = min((u32)IGB_MAX_RX_QUEUES, (u32)num_online_cpus());
+#ifdef CONFIG_NETDEVICES_MULTIQUEUE
+ adapter->num_tx_queues = min(IGB_MAX_TX_QUEUES, num_online_cpus());
+#else
adapter->num_tx_queues = 1;
- adapter->num_rx_queues = min(IGB_MAX_RX_QUEUES, num_online_cpus());
+#endif /* CONFIG_NET_MULTI_QUEUE_DEVICE */
+ /* This call may decrease the number of queues depending on
+ * interrupt mode. */
igb_set_interrupt_capability(adapter);
if (igb_alloc_queues(adapter)) {
@@ -1386,8 +1419,6 @@ int igb_setup_tx_resources(struct igb_adapter *adapter,
tx_ring->adapter = adapter;
tx_ring->next_to_use = 0;
tx_ring->next_to_clean = 0;
- spin_lock_init(&tx_ring->tx_clean_lock);
- spin_lock_init(&tx_ring->tx_lock);
return 0;
err:
@@ -1407,6 +1438,9 @@ err:
static int igb_setup_all_tx_resources(struct igb_adapter *adapter)
{
int i, err = 0;
+#ifdef CONFIG_NETDEVICES_MULTIQUEUE
+ int r_idx;
+#endif
for (i = 0; i < adapter->num_tx_queues; i++) {
err = igb_setup_tx_resources(adapter, &adapter->tx_ring[i]);
@@ -1419,6 +1453,12 @@ static int igb_setup_all_tx_resources(struct igb_adapter *adapter)
}
}
+#ifdef CONFIG_NETDEVICES_MULTIQUEUE
+ for (i = 0; i < IGB_MAX_TX_QUEUES; i++) {
+ r_idx = i % adapter->num_tx_queues;
+ adapter->multi_tx_table[i] = &adapter->tx_ring[r_idx];
+ }
+#endif
return err;
}
@@ -2096,6 +2136,9 @@ static void igb_watchdog_task(struct work_struct *work)
struct e1000_mac_info *mac = &adapter->hw.mac;
u32 link;
s32 ret_val;
+#ifdef CONFIG_NETDEVICES_MULTIQUEUE
+ int i;
+#endif
if ((netif_carrier_ok(netdev)) &&
(rd32(E1000_STATUS) & E1000_STATUS_LU))
@@ -2152,6 +2195,10 @@ static void igb_watchdog_task(struct work_struct *work)
netif_carrier_on(netdev);
netif_wake_queue(netdev);
+#ifdef CONFIG_NETDEVICES_MULTIQUEUE
+ for (i = 0; i < adapter->num_tx_queues; i++)
+ netif_wake_subqueue(netdev, i);
+#endif
if (!test_bit(__IGB_DOWN, &adapter->state))
mod_timer(&adapter->phy_info_timer,
@@ -2164,6 +2211,10 @@ static void igb_watchdog_task(struct work_struct *work)
dev_info(&adapter->pdev->dev, "NIC Link is Down\n");
netif_carrier_off(netdev);
netif_stop_queue(netdev);
+#ifdef CONFIG_NETDEVICES_MULTIQUEUE
+ for (i = 0; i < adapter->num_tx_queues; i++)
+ netif_stop_subqueue(netdev, i);
+#endif
if (!test_bit(__IGB_DOWN, &adapter->state))
mod_timer(&adapter->phy_info_timer,
round_jiffies(jiffies + 2 * HZ));
@@ -2524,7 +2575,7 @@ static inline bool igb_tx_csum_adv(struct igb_adapter *adapter,
context_desc->type_tucmd_mlhl = cpu_to_le32(tu_cmd);
context_desc->seqnum_seed = 0;
context_desc->mss_l4len_idx =
- cpu_to_le32(tx_ring->eims_value >> 4);
+ cpu_to_le32(tx_ring->queue_index << 4);
buffer_info->time_stamp = jiffies;
buffer_info->dma = 0;
@@ -2627,7 +2678,7 @@ static inline void igb_tx_queue_adv(struct igb_adapter *adapter,
if (tx_flags & (IGB_TX_FLAGS_CSUM | IGB_TX_FLAGS_TSO |
IGB_TX_FLAGS_VLAN))
- olinfo_status |= tx_ring->eims_value >> 4;
+ olinfo_status |= tx_ring->queue_index << 4;
olinfo_status |= ((paylen - hdr_len) << E1000_ADVTXD_PAYLEN_SHIFT);
@@ -2663,7 +2714,12 @@ static int __igb_maybe_stop_tx(struct net_device *netdev,
{
struct igb_adapter *adapter = netdev_priv(netdev);
+#ifdef CONFIG_NETDEVICES_MULTIQUEUE
+ netif_stop_subqueue(netdev, tx_ring->queue_index);
+#else
netif_stop_queue(netdev);
+#endif
+
/* Herbert's original patch had:
* smp_mb__after_netif_stop_queue();
* but since that doesn't exist yet, just open code it. */
@@ -2675,7 +2731,11 @@ static int __igb_maybe_stop_tx(struct net_device *netdev,
return -EBUSY;
/* A reprieve! */
- netif_start_queue(netdev);
+#ifdef CONFIG_NETDEVICES_MULTIQUEUE
+ netif_wake_subqueue(netdev, tx_ring->queue_index);
+#else
+ netif_wake_queue(netdev);
+#endif
++adapter->restart_queue;
return 0;
}
@@ -2697,7 +2757,6 @@ static int igb_xmit_frame_ring_adv(struct sk_buff *skb,
struct igb_adapter *adapter = netdev_priv(netdev);
unsigned int tx_flags = 0;
unsigned int len;
- unsigned long irq_flags;
u8 hdr_len = 0;
int tso = 0;
@@ -2713,10 +2772,6 @@ static int igb_xmit_frame_ring_adv(struct sk_buff *skb,
return NETDEV_TX_OK;
}
- if (!spin_trylock_irqsave(&tx_ring->tx_lock, irq_flags))
- /* Collision - tell upper layer to requeue */
- return NETDEV_TX_LOCKED;
-
/* need: 1 descriptor per page,
* + 2 desc gap to keep tail from touching head,
* + 1 desc for skb->data,
@@ -2724,7 +2779,6 @@ static int igb_xmit_frame_ring_adv(struct sk_buff *skb,
* otherwise try next time */
if (igb_maybe_stop_tx(netdev, tx_ring, skb_shinfo(skb)->nr_frags + 4)) {
/* this is a hard error */
- spin_unlock_irqrestore(&tx_ring->tx_lock, irq_flags);
return NETDEV_TX_BUSY;
}
@@ -2733,12 +2787,14 @@ static int igb_xmit_frame_ring_adv(struct sk_buff *skb,
tx_flags |= (vlan_tx_tag_get(skb) << IGB_TX_FLAGS_VLAN_SHIFT);
}
+ if (skb->protocol == htons(ETH_P_IP))
+ tx_flags |= IGB_TX_FLAGS_IPV4;
+
tso = skb_is_gso(skb) ? igb_tso_adv(adapter, tx_ring, skb, tx_flags,
&hdr_len) : 0;
if (tso < 0) {
dev_kfree_skb_any(skb);
- spin_unlock_irqrestore(&tx_ring->tx_lock, irq_flags);
return NETDEV_TX_OK;
}
@@ -2748,9 +2804,6 @@ static int igb_xmit_frame_ring_adv(struct sk_buff *skb,
if (skb->ip_summed == CHECKSUM_PARTIAL)
tx_flags |= IGB_TX_FLAGS_CSUM;
- if (skb->protocol == htons(ETH_P_IP))
- tx_flags |= IGB_TX_FLAGS_IPV4;
-
igb_tx_queue_adv(adapter, tx_ring, tx_flags,
igb_tx_map_adv(adapter, tx_ring, skb),
skb->len, hdr_len);
@@ -2760,14 +2813,22 @@ static int igb_xmit_frame_ring_adv(struct sk_buff *skb,
/* Make sure there is space in the ring for the next send. */
igb_maybe_stop_tx(netdev, tx_ring, MAX_SKB_FRAGS + 4);
- spin_unlock_irqrestore(&tx_ring->tx_lock, irq_flags);
return NETDEV_TX_OK;
}
static int igb_xmit_frame_adv(struct sk_buff *skb, struct net_device *netdev)
{
struct igb_adapter *adapter = netdev_priv(netdev);
- struct igb_ring *tx_ring = &adapter->tx_ring[0];
+ struct igb_ring *tx_ring;
+
+#ifdef CONFIG_NETDEVICES_MULTIQUEUE
+ int r_idx = 0;
+ r_idx = skb->queue_mapping & (IGB_MAX_TX_QUEUES - 1);
+ tx_ring = adapter->multi_tx_table[r_idx];
+#else
+ tx_ring = &adapter->tx_ring[0];
+#endif
+
/* This goes back to the question of how to logically map a tx queue
* to a flow. Right now, performance is impacted slightly negatively
@@ -3035,7 +3096,7 @@ static irqreturn_t igb_msix_other(int irq, void *data)
/* guard against interrupt when we're going down */
if (!test_bit(__IGB_DOWN, &adapter->state))
mod_timer(&adapter->watchdog_timer, jiffies + 1);
-
+
no_link_interrupt:
wr32(E1000_IMS, E1000_IMS_LSC);
wr32(E1000_EIMS, adapter->eims_other);
@@ -3054,12 +3115,15 @@ static irqreturn_t igb_msix_tx(int irq, void *data)
tx_ring->total_bytes = 0;
tx_ring->total_packets = 0;
+
+ /* auto mask will automatically reenable the interrupt when we write
+ * EICS */
if (!igb_clean_tx_irq(tx_ring))
/* Ring was not completely cleaned, so fire another interrupt */
wr32(E1000_EICS, tx_ring->eims_value);
-
- if (!tx_ring->itr_val)
+ else
wr32(E1000_EIMS, tx_ring->eims_value);
+
return IRQ_HANDLED;
}
@@ -3163,42 +3227,24 @@ static irqreturn_t igb_intr(int irq, void *data)
}
/**
- * igb_clean - NAPI Rx polling callback
- * @adapter: board private structure
+ * igb_poll - NAPI Rx polling callback
+ * @napi: napi polling structure
+ * @budget: count of how many packets we should handle
**/
-static int igb_clean(struct napi_struct *napi, int budget)
+static int igb_poll(struct napi_struct *napi, int budget)
{
- struct igb_adapter *adapter = container_of(napi, struct igb_adapter,
- napi);
+ struct igb_ring *rx_ring = container_of(napi, struct igb_ring, napi);
+ struct igb_adapter *adapter = rx_ring->adapter;
struct net_device *netdev = adapter->netdev;
- int tx_clean_complete = 1, work_done = 0;
- int i;
+ int tx_clean_complete, work_done = 0;
- /* Must NOT use netdev_priv macro here. */
- adapter = netdev->priv;
-
- /* Keep link state information with original netdev */
- if (!netif_carrier_ok(netdev))
- goto quit_polling;
-
- /* igb_clean is called per-cpu. This lock protects tx_ring[i] from
- * being cleaned by multiple cpus simultaneously. A failure obtaining
- * the lock means tx_ring[i] is currently being cleaned anyway. */
- for (i = 0; i < adapter->num_tx_queues; i++) {
- if (spin_trylock(&adapter->tx_ring[i].tx_clean_lock)) {
- tx_clean_complete &= igb_clean_tx_irq(&adapter->tx_ring[i]);
- spin_unlock(&adapter->tx_ring[i].tx_clean_lock);
- }
- }
-
- for (i = 0; i < adapter->num_rx_queues; i++)
- igb_clean_rx_irq_adv(&adapter->rx_ring[i], &work_done,
- adapter->rx_ring[i].napi.weight);
+ /* this poll routine only supports one tx and one rx queue */
+ tx_clean_complete = igb_clean_tx_irq(&adapter->tx_ring[0]);
+ igb_clean_rx_irq_adv(&adapter->rx_ring[0], &work_done, budget);
/* If no Tx and not enough Rx work done, exit the polling mode */
if ((tx_clean_complete && (work_done < budget)) ||
!netif_running(netdev)) {
-quit_polling:
if (adapter->itr_setting & 3)
igb_set_itr(adapter, E1000_ITR, false);
netif_rx_complete(netdev, napi);
@@ -3327,11 +3373,19 @@ done_cleaning:
* sees the new next_to_clean.
*/
smp_mb();
+#ifdef CONFIG_NETDEVICES_MULTIQUEUE
+ if (__netif_subqueue_stopped(netdev, tx_ring->queue_index) &&
+ !(test_bit(__IGB_DOWN, &adapter->state))) {
+ netif_wake_subqueue(netdev, tx_ring->queue_index);
+ ++adapter->restart_queue;
+ }
+#else
if (netif_queue_stopped(netdev) &&
!(test_bit(__IGB_DOWN, &adapter->state))) {
netif_wake_queue(netdev);
++adapter->restart_queue;
}
+#endif
}
if (tx_ring->detect_tx_hung) {
@@ -3368,7 +3422,11 @@ done_cleaning:
tx_ring->buffer_info[i].time_stamp,
jiffies,
tx_desc->upper.fields.status);
+#ifdef CONFIG_NETDEVICES_MULTIQUEUE
+ netif_stop_subqueue(netdev, tx_ring->queue_index);
+#else
netif_stop_queue(netdev);
+#endif
}
}
tx_ring->total_bytes += total_bytes;
^ permalink raw reply related [flat|nested] 15+ messages in thread* Re: [RESEND NET-NEXT PATCH 08/20] igb: Introduce multiple TX queues with infrastructure
2008-07-08 22:06 ` [RESEND NET-NEXT PATCH 08/20] igb: Introduce multiple TX queues with infrastructure Jeff Kirsher
@ 2008-07-11 5:22 ` Jeff Garzik
0 siblings, 0 replies; 15+ messages in thread
From: Jeff Garzik @ 2008-07-11 5:22 UTC (permalink / raw)
To: Jeff Kirsher
Cc: davem, netdev, linux-kernel, akpm, Peter P Waskiewicz Jr,
Mitch Williams, Auke Kok
Jeff Kirsher wrote:
> From: Peter P Waskiewicz Jr <peter.p.waskiewicz.jr@intel.com>
>
> This code adds multiple Tx queue infrastructure much like we
> previously did in ixgbe. The MSI-X vector mapping is the bulk of
> the change.
>
> IAM can now be safely enabled and we've verified that it does
> work correctly. We can also eliminate the tx ring lock.
>
> Signed-off-by: Peter P Waskiewicz Jr <peter.p.waskiewicz.jr@intel.com>
> Signed-off-by: Mitch Williams <mitch.a.williams@intel.com>
> Signed-off-by: Auke Kok <auke-jan.h.kok@intel.com>
> ---
>
> drivers/net/igb/igb.h | 7 +-
> drivers/net/igb/igb_main.c | 160 ++++++++++++++++++++++++++++++--------------
> 2 files changed, 114 insertions(+), 53 deletions(-)
applied 8-20
^ permalink raw reply [flat|nested] 15+ messages in thread
* [RESEND NET-NEXT PATCH 09/20] igb: update ethtool stats to support multiqueue
2008-07-08 22:06 [RESEND NET-NEXT PATCH 00/20] igb update for net-next Jeff Kirsher
2008-07-08 22:06 ` [RESEND NET-NEXT PATCH 08/20] igb: Introduce multiple TX queues with infrastructure Jeff Kirsher
@ 2008-07-08 22:07 ` Jeff Kirsher
2008-07-08 22:07 ` [RESEND NET-NEXT PATCH 10/20] igb: add DCA support Jeff Kirsher
` (10 subsequent siblings)
12 siblings, 0 replies; 15+ messages in thread
From: Jeff Kirsher @ 2008-07-08 22:07 UTC (permalink / raw)
To: jeff, davem; +Cc: netdev, linux-kernel, akpm, Alexander Duyck, Jeff Kirsher
From: Alexander Duyck <alexander.h.duyck@intel.com>
Addesses problems seen earlier with igb driver not correctly reporting rx
and tx stats.
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
---
drivers/net/igb/igb.h | 1 +
drivers/net/igb/igb_ethtool.c | 13 +++++++++----
drivers/net/igb/igb_main.c | 2 ++
3 files changed, 12 insertions(+), 4 deletions(-)
diff --git a/drivers/net/igb/igb.h b/drivers/net/igb/igb.h
index a1431c8..5915efc 100644
--- a/drivers/net/igb/igb.h
+++ b/drivers/net/igb/igb.h
@@ -158,6 +158,7 @@ struct igb_ring {
union {
/* TX */
struct {
+ struct igb_queue_stats tx_stats;
bool detect_tx_hung;
};
/* RX */
diff --git a/drivers/net/igb/igb_ethtool.c b/drivers/net/igb/igb_ethtool.c
index 0447f9b..ed756c1 100644
--- a/drivers/net/igb/igb_ethtool.c
+++ b/drivers/net/igb/igb_ethtool.c
@@ -96,10 +96,8 @@ static const struct igb_stats igb_gstrings_stats[] = {
};
#define IGB_QUEUE_STATS_LEN \
- ((((((struct igb_adapter *)netdev->priv)->num_rx_queues > 1) ? \
- ((struct igb_adapter *)netdev->priv)->num_rx_queues : 0) + \
- (((((struct igb_adapter *)netdev->priv)->num_tx_queues > 1) ? \
- ((struct igb_adapter *)netdev->priv)->num_tx_queues : 0))) * \
+ ((((struct igb_adapter *)netdev->priv)->num_rx_queues + \
+ ((struct igb_adapter *)netdev->priv)->num_tx_queues) * \
(sizeof(struct igb_queue_stats) / sizeof(u64)))
#define IGB_GLOBAL_STATS_LEN \
sizeof(igb_gstrings_stats) / sizeof(struct igb_stats)
@@ -1842,6 +1840,13 @@ static void igb_get_ethtool_stats(struct net_device *netdev,
data[i] = (igb_gstrings_stats[i].sizeof_stat ==
sizeof(u64)) ? *(u64 *)p : *(u32 *)p;
}
+ for (j = 0; j < adapter->num_tx_queues; j++) {
+ int k;
+ queue_stat = (u64 *)&adapter->tx_ring[j].tx_stats;
+ for (k = 0; k < stat_count; k++)
+ data[i + k] = queue_stat[k];
+ i += k;
+ }
for (j = 0; j < adapter->num_rx_queues; j++) {
int k;
queue_stat = (u64 *)&adapter->rx_ring[j].rx_stats;
diff --git a/drivers/net/igb/igb_main.c b/drivers/net/igb/igb_main.c
index e11a5da..f975bfe 100644
--- a/drivers/net/igb/igb_main.c
+++ b/drivers/net/igb/igb_main.c
@@ -3431,6 +3431,8 @@ done_cleaning:
}
tx_ring->total_bytes += total_bytes;
tx_ring->total_packets += total_packets;
+ tx_ring->tx_stats.bytes += total_bytes;
+ tx_ring->tx_stats.packets += total_packets;
adapter->net_stats.tx_bytes += total_bytes;
adapter->net_stats.tx_packets += total_packets;
return retval;
^ permalink raw reply related [flat|nested] 15+ messages in thread* [RESEND NET-NEXT PATCH 10/20] igb: add DCA support
2008-07-08 22:06 [RESEND NET-NEXT PATCH 00/20] igb update for net-next Jeff Kirsher
2008-07-08 22:06 ` [RESEND NET-NEXT PATCH 08/20] igb: Introduce multiple TX queues with infrastructure Jeff Kirsher
2008-07-08 22:07 ` [RESEND NET-NEXT PATCH 09/20] igb: update ethtool stats to support multiqueue Jeff Kirsher
@ 2008-07-08 22:07 ` Jeff Kirsher
2008-07-08 22:08 ` [RESEND NET-NEXT PATCH 11/20] igb: reenable CRC stripping in hardware Jeff Kirsher
` (9 subsequent siblings)
12 siblings, 0 replies; 15+ messages in thread
From: Jeff Kirsher @ 2008-07-08 22:07 UTC (permalink / raw)
To: jeff, davem
Cc: netdev, linux-kernel, akpm, Jeb Cramer, Mitch Williams, Auke Kok,
Shannon Nelson
From: Jeb Cramer <cramerj@intel.com>
Add DCA support in the similar method that it was added to the ixgbe
driver recently. DCA allows the network device to put data in the
CPU cache and notify the chipset of that event. This reduces cache
misses during receives.
Signed-off-by: Jeb Cramer <cramerj@intel.com>
Signed-off-by: Mitch Williams <mitch.a.williams@intel.com>
Signed-off-by: Auke Kok <auke-jan.h.kok@intel.com>
Signed-off-by: Shannon Nelson <shannon.nelson@intel.com>
---
drivers/net/igb/e1000_82575.h | 11 +++
drivers/net/igb/e1000_regs.h | 2
drivers/net/igb/igb.h | 4 +
drivers/net/igb/igb_main.c | 174 ++++++++++++++++++++++++++++++++++++++++-
4 files changed, 187 insertions(+), 4 deletions(-)
diff --git a/drivers/net/igb/e1000_82575.h b/drivers/net/igb/e1000_82575.h
index d78ad33..02e57a8 100644
--- a/drivers/net/igb/e1000_82575.h
+++ b/drivers/net/igb/e1000_82575.h
@@ -144,9 +144,20 @@ struct e1000_adv_tx_context_desc {
#define E1000_RXDCTL_QUEUE_ENABLE 0x02000000 /* Enable specific Rx Queue */
/* Direct Cache Access (DCA) definitions */
+#define E1000_DCA_CTRL_DCA_ENABLE 0x00000000 /* DCA Enable */
+#define E1000_DCA_CTRL_DCA_DISABLE 0x00000001 /* DCA Disable */
+#define E1000_DCA_CTRL_DCA_MODE_CB1 0x00 /* DCA Mode CB1 */
+#define E1000_DCA_CTRL_DCA_MODE_CB2 0x02 /* DCA Mode CB2 */
+#define E1000_DCA_RXCTRL_CPUID_MASK 0x0000001F /* Rx CPUID Mask */
+#define E1000_DCA_RXCTRL_DESC_DCA_EN (1 << 5) /* DCA Rx Desc enable */
+#define E1000_DCA_RXCTRL_HEAD_DCA_EN (1 << 6) /* DCA Rx Desc header enable */
+#define E1000_DCA_RXCTRL_DATA_DCA_EN (1 << 7) /* DCA Rx Desc payload enable */
+#define E1000_DCA_TXCTRL_CPUID_MASK 0x0000001F /* Tx CPUID Mask */
+#define E1000_DCA_TXCTRL_DESC_DCA_EN (1 << 5) /* DCA Tx Desc enable */
#define E1000_DCA_TXCTRL_TX_WB_RO_EN (1 << 11) /* Tx Desc writeback RO bit */
+
#endif
diff --git a/drivers/net/igb/e1000_regs.h b/drivers/net/igb/e1000_regs.h
index ff187b7..d25e914 100644
--- a/drivers/net/igb/e1000_regs.h
+++ b/drivers/net/igb/e1000_regs.h
@@ -235,6 +235,8 @@
#define E1000_FACTPS 0x05B30 /* Function Active and Power State to MNG */
#define E1000_SWSM 0x05B50 /* SW Semaphore */
#define E1000_FWSM 0x05B54 /* FW Semaphore */
+#define E1000_DCA_ID 0x05B70 /* DCA Requester ID Information - RO */
+#define E1000_DCA_CTRL 0x05B74 /* DCA Control - RW */
#define E1000_HICR 0x08F00 /* Host Inteface Control */
/* RSS registers */
diff --git a/drivers/net/igb/igb.h b/drivers/net/igb/igb.h
index 5915efc..d4a0423 100644
--- a/drivers/net/igb/igb.h
+++ b/drivers/net/igb/igb.h
@@ -271,7 +271,9 @@ struct igb_adapter {
/* to not mess up cache alignment, always add to the bottom */
unsigned long state;
unsigned int msi_enabled;
-
+#ifdef CONFIG_DCA
+ unsigned int dca_enabled;
+#endif
u32 eeprom_wol;
/* for ioport free */
diff --git a/drivers/net/igb/igb_main.c b/drivers/net/igb/igb_main.c
index f975bfe..e8ef541 100644
--- a/drivers/net/igb/igb_main.c
+++ b/drivers/net/igb/igb_main.c
@@ -41,7 +41,9 @@
#include <linux/delay.h>
#include <linux/interrupt.h>
#include <linux/if_ether.h>
-
+#ifdef CONFIG_DCA
+#include <linux/dca.h>
+#endif
#include "igb.h"
#define DRV_VERSION "1.0.8-k2"
@@ -102,6 +104,11 @@ static irqreturn_t igb_msix_other(int irq, void *);
static irqreturn_t igb_msix_rx(int irq, void *);
static irqreturn_t igb_msix_tx(int irq, void *);
static int igb_clean_rx_ring_msix(struct napi_struct *, int);
+#ifdef CONFIG_DCA
+static void igb_update_rx_dca(struct igb_ring *);
+static void igb_update_tx_dca(struct igb_ring *);
+static void igb_setup_dca(struct igb_adapter *);
+#endif /* CONFIG_DCA */
static bool igb_clean_tx_irq(struct igb_ring *);
static int igb_poll(struct napi_struct *, int);
static bool igb_clean_rx_irq_adv(struct igb_ring *, int *, int);
@@ -119,6 +126,14 @@ static int igb_suspend(struct pci_dev *, pm_message_t);
static int igb_resume(struct pci_dev *);
#endif
static void igb_shutdown(struct pci_dev *);
+#ifdef CONFIG_DCA
+static int igb_notify_dca(struct notifier_block *, unsigned long, void *);
+static struct notifier_block dca_notifier = {
+ .notifier_call = igb_notify_dca,
+ .next = NULL,
+ .priority = 0
+};
+#endif
#ifdef CONFIG_NET_POLL_CONTROLLER
/* for netdump / net console */
@@ -183,6 +198,9 @@ static int __init igb_init_module(void)
printk(KERN_INFO "%s\n", igb_copyright);
ret = pci_register_driver(&igb_driver);
+#ifdef CONFIG_DCA
+ dca_register_notify(&dca_notifier);
+#endif
return ret;
}
@@ -196,6 +214,9 @@ module_init(igb_init_module);
**/
static void __exit igb_exit_module(void)
{
+#ifdef CONFIG_DCA
+ dca_unregister_notify(&dca_notifier);
+#endif
pci_unregister_driver(&igb_driver);
}
@@ -1130,6 +1151,17 @@ static int __devinit igb_probe(struct pci_dev *pdev,
if (err)
goto err_register;
+#ifdef CONFIG_DCA
+ if (dca_add_requester(&pdev->dev) == 0) {
+ adapter->dca_enabled = true;
+ dev_info(&pdev->dev, "DCA enabled\n");
+ /* Always use CB2 mode, difference is masked
+ * in the CB driver. */
+ wr32(E1000_DCA_CTRL, 2);
+ igb_setup_dca(adapter);
+ }
+#endif
+
dev_info(&pdev->dev, "Intel(R) Gigabit Ethernet Network Connection\n");
/* print bus type/speed/width info */
dev_info(&pdev->dev,
@@ -1193,6 +1225,7 @@ static void __devexit igb_remove(struct pci_dev *pdev)
{
struct net_device *netdev = pci_get_drvdata(pdev);
struct igb_adapter *adapter = netdev_priv(netdev);
+ struct e1000_hw *hw = &adapter->hw;
/* flush_scheduled work may reschedule our watchdog task, so
* explicitly disable watchdog tasks from being rescheduled */
@@ -1202,6 +1235,15 @@ static void __devexit igb_remove(struct pci_dev *pdev)
flush_scheduled_work();
+#ifdef CONFIG_DCA
+ if (adapter->dca_enabled) {
+ dev_info(&pdev->dev, "DCA disabled\n");
+ dca_remove_requester(&pdev->dev);
+ adapter->dca_enabled = false;
+ wr32(E1000_DCA_CTRL, 1);
+ }
+#endif
+
/* Release control of h/w to f/w. If f/w is AMT enabled, this
* would have already happened in close and is redundant. */
igb_release_hw_control(adapter);
@@ -3112,7 +3154,10 @@ static irqreturn_t igb_msix_tx(int irq, void *data)
if (!tx_ring->itr_val)
wr32(E1000_EIMC, tx_ring->eims_value);
-
+#ifdef CONFIG_DCA
+ if (adapter->dca_enabled)
+ igb_update_tx_dca(tx_ring);
+#endif
tx_ring->total_bytes = 0;
tx_ring->total_packets = 0;
@@ -3146,9 +3191,119 @@ static irqreturn_t igb_msix_rx(int irq, void *data)
if (netif_rx_schedule_prep(adapter->netdev, &rx_ring->napi))
__netif_rx_schedule(adapter->netdev, &rx_ring->napi);
- return IRQ_HANDLED;
+#ifdef CONFIG_DCA
+ if (adapter->dca_enabled)
+ igb_update_rx_dca(rx_ring);
+#endif
+ return IRQ_HANDLED;
+}
+
+#ifdef CONFIG_DCA
+static void igb_update_rx_dca(struct igb_ring *rx_ring)
+{
+ u32 dca_rxctrl;
+ struct igb_adapter *adapter = rx_ring->adapter;
+ struct e1000_hw *hw = &adapter->hw;
+ int cpu = get_cpu();
+ int q = rx_ring - adapter->rx_ring;
+
+ if (rx_ring->cpu != cpu) {
+ dca_rxctrl = rd32(E1000_DCA_RXCTRL(q));
+ dca_rxctrl &= ~E1000_DCA_RXCTRL_CPUID_MASK;
+ dca_rxctrl |= dca_get_tag(cpu);
+ dca_rxctrl |= E1000_DCA_RXCTRL_DESC_DCA_EN;
+ dca_rxctrl |= E1000_DCA_RXCTRL_HEAD_DCA_EN;
+ dca_rxctrl |= E1000_DCA_RXCTRL_DATA_DCA_EN;
+ wr32(E1000_DCA_RXCTRL(q), dca_rxctrl);
+ rx_ring->cpu = cpu;
+ }
+ put_cpu();
+}
+
+static void igb_update_tx_dca(struct igb_ring *tx_ring)
+{
+ u32 dca_txctrl;
+ struct igb_adapter *adapter = tx_ring->adapter;
+ struct e1000_hw *hw = &adapter->hw;
+ int cpu = get_cpu();
+ int q = tx_ring - adapter->tx_ring;
+
+ if (tx_ring->cpu != cpu) {
+ dca_txctrl = rd32(E1000_DCA_TXCTRL(q));
+ dca_txctrl &= ~E1000_DCA_TXCTRL_CPUID_MASK;
+ dca_txctrl |= dca_get_tag(cpu);
+ dca_txctrl |= E1000_DCA_TXCTRL_DESC_DCA_EN;
+ wr32(E1000_DCA_TXCTRL(q), dca_txctrl);
+ tx_ring->cpu = cpu;
+ }
+ put_cpu();
+}
+
+static void igb_setup_dca(struct igb_adapter *adapter)
+{
+ int i;
+
+ if (!(adapter->dca_enabled))
+ return;
+
+ for (i = 0; i < adapter->num_tx_queues; i++) {
+ adapter->tx_ring[i].cpu = -1;
+ igb_update_tx_dca(&adapter->tx_ring[i]);
+ }
+ for (i = 0; i < adapter->num_rx_queues; i++) {
+ adapter->rx_ring[i].cpu = -1;
+ igb_update_rx_dca(&adapter->rx_ring[i]);
+ }
}
+static int __igb_notify_dca(struct device *dev, void *data)
+{
+ struct net_device *netdev = dev_get_drvdata(dev);
+ struct igb_adapter *adapter = netdev_priv(netdev);
+ struct e1000_hw *hw = &adapter->hw;
+ unsigned long event = *(unsigned long *)data;
+
+ switch (event) {
+ case DCA_PROVIDER_ADD:
+ /* if already enabled, don't do it again */
+ if (adapter->dca_enabled)
+ break;
+ adapter->dca_enabled = true;
+ /* Always use CB2 mode, difference is masked
+ * in the CB driver. */
+ wr32(E1000_DCA_CTRL, 2);
+ if (dca_add_requester(dev) == 0) {
+ dev_info(&adapter->pdev->dev, "DCA enabled\n");
+ igb_setup_dca(adapter);
+ break;
+ }
+ /* Fall Through since DCA is disabled. */
+ case DCA_PROVIDER_REMOVE:
+ if (adapter->dca_enabled) {
+ /* without this a class_device is left
+ * hanging around in the sysfs model */
+ dca_remove_requester(dev);
+ dev_info(&adapter->pdev->dev, "DCA disabled\n");
+ adapter->dca_enabled = false;
+ wr32(E1000_DCA_CTRL, 1);
+ }
+ break;
+ }
+
+ return 0;
+}
+
+static int igb_notify_dca(struct notifier_block *nb, unsigned long event,
+ void *p)
+{
+ int ret_val;
+
+ ret_val = driver_for_each_device(&igb_driver.driver, NULL, &event,
+ __igb_notify_dca);
+
+ return ret_val ? NOTIFY_BAD : NOTIFY_DONE;
+}
+#endif /* CONFIG_DCA */
/**
* igb_intr_msi - Interrupt Handler
@@ -3239,7 +3394,16 @@ static int igb_poll(struct napi_struct *napi, int budget)
int tx_clean_complete, work_done = 0;
/* this poll routine only supports one tx and one rx queue */
+#ifdef CONFIG_DCA
+ if (adapter->dca_enabled)
+ igb_update_tx_dca(&adapter->tx_ring[0]);
+#endif
tx_clean_complete = igb_clean_tx_irq(&adapter->tx_ring[0]);
+
+#ifdef CONFIG_DCA
+ if (adapter->dca_enabled)
+ igb_update_rx_dca(&adapter->rx_ring[0]);
+#endif
igb_clean_rx_irq_adv(&adapter->rx_ring[0], &work_done, budget);
/* If no Tx and not enough Rx work done, exit the polling mode */
@@ -3268,6 +3432,10 @@ static int igb_clean_rx_ring_msix(struct napi_struct *napi, int budget)
if (!netif_carrier_ok(netdev))
goto quit_polling;
+#ifdef CONFIG_DCA
+ if (adapter->dca_enabled)
+ igb_update_rx_dca(rx_ring);
+#endif
igb_clean_rx_irq_adv(rx_ring, &work_done, budget);
^ permalink raw reply related [flat|nested] 15+ messages in thread* [RESEND NET-NEXT PATCH 11/20] igb: reenable CRC stripping in hardware
2008-07-08 22:06 [RESEND NET-NEXT PATCH 00/20] igb update for net-next Jeff Kirsher
` (2 preceding siblings ...)
2008-07-08 22:07 ` [RESEND NET-NEXT PATCH 10/20] igb: add DCA support Jeff Kirsher
@ 2008-07-08 22:08 ` Jeff Kirsher
2008-07-08 22:09 ` [RESEND NET-NEXT PATCH 12/20] igb: Increment driver version Jeff Kirsher
` (8 subsequent siblings)
12 siblings, 0 replies; 15+ messages in thread
From: Jeff Kirsher @ 2008-07-08 22:08 UTC (permalink / raw)
To: jeff, davem; +Cc: netdev, linux-kernel, akpm, Auke Kok, Jeff Kirsher
From: Auke Kok <auke-jan.h.kok@intel.com>
We can remove a clunky workaround for not having the hardware
strip the CRC. 82575 silicon as well as the older PCI Express
e1000e hardware all work OK in this respect.
Signed-off-by: Auke Kok <auke-jan.h.kok@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
---
drivers/net/igb/igb_main.c | 9 +++++----
1 files changed, 5 insertions(+), 4 deletions(-)
diff --git a/drivers/net/igb/igb_main.c b/drivers/net/igb/igb_main.c
index e8ef541..ae329c0 100644
--- a/drivers/net/igb/igb_main.c
+++ b/drivers/net/igb/igb_main.c
@@ -1664,10 +1664,12 @@ static void igb_setup_rctl(struct igb_adapter *adapter)
E1000_RCTL_LBM_NO | E1000_RCTL_RDMTS_HALF |
(adapter->hw.mac.mc_filter_type << E1000_RCTL_MO_SHIFT);
- /* disable the stripping of CRC because it breaks
- * BMC firmware connected over SMBUS
- rctl |= E1000_RCTL_SECRC;
+ /*
+ * enable stripping of CRC. It's unlikely this will break BMC
+ * redirection as it did with e1000. Newer features require
+ * that the HW strips the CRC.
*/
+ rctl |= E1000_RCTL_SECRC;
rctl &= ~E1000_RCTL_SBP;
@@ -3743,7 +3745,6 @@ static bool igb_clean_rx_irq_adv(struct igb_ring *rx_ring,
}
}
send_up:
- pskb_trim(skb, skb->len - 4);
i++;
if (i == rx_ring->count)
i = 0;
^ permalink raw reply related [flat|nested] 15+ messages in thread* [RESEND NET-NEXT PATCH 12/20] igb: Increment driver version
2008-07-08 22:06 [RESEND NET-NEXT PATCH 00/20] igb update for net-next Jeff Kirsher
` (3 preceding siblings ...)
2008-07-08 22:08 ` [RESEND NET-NEXT PATCH 11/20] igb: reenable CRC stripping in hardware Jeff Kirsher
@ 2008-07-08 22:09 ` Jeff Kirsher
2008-07-08 22:10 ` [RESEND NET-NEXT PATCH 13/20] igb: add 82576 MAC support Jeff Kirsher
` (7 subsequent siblings)
12 siblings, 0 replies; 15+ messages in thread
From: Jeff Kirsher @ 2008-07-08 22:09 UTC (permalink / raw)
To: jeff, davem; +Cc: netdev, linux-kernel, akpm, Auke Kok, Jeff Kirsher
From: Auke Kok <auke-jan.h.kok@intel.com>
Signed-off-by: Auke Kok <auke-jan.h.kok@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
---
drivers/net/igb/igb_main.c | 2 +-
1 files changed, 1 insertions(+), 1 deletions(-)
diff --git a/drivers/net/igb/igb_main.c b/drivers/net/igb/igb_main.c
index ae329c0..1436326 100644
--- a/drivers/net/igb/igb_main.c
+++ b/drivers/net/igb/igb_main.c
@@ -46,7 +46,7 @@
#endif
#include "igb.h"
-#define DRV_VERSION "1.0.8-k2"
+#define DRV_VERSION "1.2.45-k2"
char igb_driver_name[] = "igb";
char igb_driver_version[] = DRV_VERSION;
static const char igb_driver_string[] =
^ permalink raw reply related [flat|nested] 15+ messages in thread* [RESEND NET-NEXT PATCH 13/20] igb: add 82576 MAC support
2008-07-08 22:06 [RESEND NET-NEXT PATCH 00/20] igb update for net-next Jeff Kirsher
` (4 preceding siblings ...)
2008-07-08 22:09 ` [RESEND NET-NEXT PATCH 12/20] igb: Increment driver version Jeff Kirsher
@ 2008-07-08 22:10 ` Jeff Kirsher
2008-07-08 22:10 ` [RESEND NET-NEXT PATCH 14/20] igb: Add support for quad port WOL and feature flags Jeff Kirsher
` (6 subsequent siblings)
12 siblings, 0 replies; 15+ messages in thread
From: Jeff Kirsher @ 2008-07-08 22:10 UTC (permalink / raw)
To: jeff, davem
Cc: netdev, linux-kernel, akpm, Alexander Duyck, Auke Kok,
Jeff Kirsher
From: Alexander Duyck <alexander.h.duyck@intel.com>
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Signed-off-by: Auke Kok <auke-jan.h.kok@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
---
drivers/net/igb/e1000_82575.c | 208 +++++++++++++++++++++++++++++++++++++--
drivers/net/igb/e1000_82575.h | 9 ++
drivers/net/igb/e1000_defines.h | 10 ++
drivers/net/igb/e1000_hw.h | 8 ++
drivers/net/igb/e1000_mac.c | 3 -
drivers/net/igb/e1000_mac.h | 1
drivers/net/igb/e1000_regs.h | 7 +
drivers/net/igb/igb_ethtool.c | 146 +++++++++++++++++++++------
drivers/net/igb/igb_main.c | 170 +++++++++++++++++++++++++-------
9 files changed, 476 insertions(+), 86 deletions(-)
diff --git a/drivers/net/igb/e1000_82575.c b/drivers/net/igb/e1000_82575.c
index 2c8b910..e098f23 100644
--- a/drivers/net/igb/e1000_82575.c
+++ b/drivers/net/igb/e1000_82575.c
@@ -31,6 +31,7 @@
#include <linux/types.h>
#include <linux/slab.h>
+#include <linux/if_ether.h>
#include "e1000_mac.h"
#include "e1000_82575.h"
@@ -45,7 +46,6 @@ static s32 igb_get_cfg_done_82575(struct e1000_hw *);
static s32 igb_init_hw_82575(struct e1000_hw *);
static s32 igb_phy_hw_reset_sgmii_82575(struct e1000_hw *);
static s32 igb_read_phy_reg_sgmii_82575(struct e1000_hw *, u32, u16 *);
-static void igb_rar_set_82575(struct e1000_hw *, u8 *, u32);
static s32 igb_reset_hw_82575(struct e1000_hw *);
static s32 igb_set_d0_lplu_state_82575(struct e1000_hw *, bool);
static s32 igb_setup_copper_link_82575(struct e1000_hw *);
@@ -84,6 +84,12 @@ static s32 igb_get_invariants_82575(struct e1000_hw *hw)
case E1000_DEV_ID_82575GB_QUAD_COPPER:
mac->type = e1000_82575;
break;
+ case E1000_DEV_ID_82576:
+ case E1000_DEV_ID_82576_FIBER:
+ case E1000_DEV_ID_82576_SERDES:
+ case E1000_DEV_ID_82576_QUAD_COPPER:
+ mac->type = e1000_82576;
+ break;
default:
return -E1000_ERR_MAC_INIT;
break;
@@ -128,6 +134,8 @@ static s32 igb_get_invariants_82575(struct e1000_hw *hw)
mac->mta_reg_count = 128;
/* Set rar entry count */
mac->rar_entry_count = E1000_RAR_ENTRIES_82575;
+ if (mac->type == e1000_82576)
+ mac->rar_entry_count = E1000_RAR_ENTRIES_82576;
/* Set if part includes ASF firmware */
mac->asf_firmware_present = true;
/* Set if manageability features are enabled. */
@@ -694,13 +702,12 @@ static s32 igb_check_for_link_82575(struct e1000_hw *hw)
if ((hw->phy.media_type != e1000_media_type_copper) ||
(igb_sgmii_active_82575(hw)))
ret_val = igb_get_pcs_speed_and_duplex_82575(hw, &speed,
- &duplex);
+ &duplex);
else
ret_val = igb_check_for_copper_link(hw);
return ret_val;
}
-
/**
* igb_get_pcs_speed_and_duplex_82575 - Retrieve current speed/duplex
* @hw: pointer to the HW structure
@@ -757,18 +764,129 @@ static s32 igb_get_pcs_speed_and_duplex_82575(struct e1000_hw *hw, u16 *speed,
}
/**
- * igb_rar_set_82575 - Set receive address register
+ * igb_init_rx_addrs_82575 - Initialize receive address's
+ * @hw: pointer to the HW structure
+ * @rar_count: receive address registers
+ *
+ * Setups the receive address registers by setting the base receive address
+ * register to the devices MAC address and clearing all the other receive
+ * address registers to 0.
+ **/
+static void igb_init_rx_addrs_82575(struct e1000_hw *hw, u16 rar_count)
+{
+ u32 i;
+ u8 addr[6] = {0,0,0,0,0,0};
+ /*
+ * This function is essentially the same as that of
+ * e1000_init_rx_addrs_generic. However it also takes care
+ * of the special case where the register offset of the
+ * second set of RARs begins elsewhere. This is implicitly taken care by
+ * function e1000_rar_set_generic.
+ */
+
+ hw_dbg("e1000_init_rx_addrs_82575");
+
+ /* Setup the receive address */
+ hw_dbg("Programming MAC Address into RAR[0]\n");
+ hw->mac.ops.rar_set(hw, hw->mac.addr, 0);
+
+ /* Zero out the other (rar_entry_count - 1) receive addresses */
+ hw_dbg("Clearing RAR[1-%u]\n", rar_count-1);
+ for (i = 1; i < rar_count; i++)
+ hw->mac.ops.rar_set(hw, addr, i);
+}
+
+/**
+ * igb_update_mc_addr_list_82575 - Update Multicast addresses
+ * @hw: pointer to the HW structure
+ * @mc_addr_list: array of multicast addresses to program
+ * @mc_addr_count: number of multicast addresses to program
+ * @rar_used_count: the first RAR register free to program
+ * @rar_count: total number of supported Receive Address Registers
+ *
+ * Updates the Receive Address Registers and Multicast Table Array.
+ * The caller must have a packed mc_addr_list of multicast addresses.
+ * The parameter rar_count will usually be hw->mac.rar_entry_count
+ * unless there are workarounds that change this.
+ **/
+void igb_update_mc_addr_list_82575(struct e1000_hw *hw,
+ u8 *mc_addr_list, u32 mc_addr_count,
+ u32 rar_used_count, u32 rar_count)
+{
+ u32 hash_value;
+ u32 i;
+ u8 addr[6] = {0,0,0,0,0,0};
+ /*
+ * This function is essentially the same as that of
+ * igb_update_mc_addr_list_generic. However it also takes care
+ * of the special case where the register offset of the
+ * second set of RARs begins elsewhere. This is implicitly taken care by
+ * function e1000_rar_set_generic.
+ */
+
+ /*
+ * Load the first set of multicast addresses into the exact
+ * filters (RAR). If there are not enough to fill the RAR
+ * array, clear the filters.
+ */
+ for (i = rar_used_count; i < rar_count; i++) {
+ if (mc_addr_count) {
+ igb_rar_set(hw, mc_addr_list, i);
+ mc_addr_count--;
+ mc_addr_list += ETH_ALEN;
+ } else {
+ igb_rar_set(hw, addr, i);
+ }
+ }
+
+ /* Clear the old settings from the MTA */
+ hw_dbg("Clearing MTA\n");
+ for (i = 0; i < hw->mac.mta_reg_count; i++) {
+ array_wr32(E1000_MTA, i, 0);
+ wrfl();
+ }
+
+ /* Load any remaining multicast addresses into the hash table. */
+ for (; mc_addr_count > 0; mc_addr_count--) {
+ hash_value = igb_hash_mc_addr(hw, mc_addr_list);
+ hw_dbg("Hash value = 0x%03X\n", hash_value);
+ hw->mac.ops.mta_set(hw, hash_value);
+ mc_addr_list += ETH_ALEN;
+ }
+}
+
+/**
+ * igb_shutdown_fiber_serdes_link_82575 - Remove link during power down
* @hw: pointer to the HW structure
- * @addr: pointer to the receive address
- * @index: receive address array register
*
- * Sets the receive address array register at index to the address passed
- * in by addr.
+ * In the case of fiber serdes, shut down optics and PCS on driver unload
+ * when management pass thru is not enabled.
**/
-static void igb_rar_set_82575(struct e1000_hw *hw, u8 *addr, u32 index)
+void igb_shutdown_fiber_serdes_link_82575(struct e1000_hw *hw)
{
- if (index < E1000_RAR_ENTRIES_82575)
- igb_rar_set(hw, addr, index);
+ u32 reg;
+
+ if (hw->mac.type != e1000_82576 ||
+ (hw->phy.media_type != e1000_media_type_fiber &&
+ hw->phy.media_type != e1000_media_type_internal_serdes))
+ return;
+
+ /* if the management interface is not enabled, then power down */
+ if (!igb_enable_mng_pass_thru(hw)) {
+ /* Disable PCS to turn off link */
+ reg = rd32(E1000_PCS_CFG0);
+ reg &= ~E1000_PCS_CFG_PCS_EN;
+ wr32(E1000_PCS_CFG0, reg);
+
+ /* shutdown the laser */
+ reg = rd32(E1000_CTRL_EXT);
+ reg |= E1000_CTRL_EXT_SDP7_DATA;
+ wr32(E1000_CTRL_EXT, reg);
+
+ /* flush the write to verify completion */
+ wrfl();
+ msleep(1);
+ }
return;
}
@@ -854,7 +972,7 @@ static s32 igb_init_hw_82575(struct e1000_hw *hw)
igb_clear_vfta(hw);
/* Setup the receive address */
- igb_init_rx_addrs(hw, rar_count);
+ igb_init_rx_addrs_82575(hw, rar_count);
/* Zero out the Multicast HASH table */
hw_dbg("Zeroing the MTA\n");
for (i = 0; i < mac->mta_reg_count; i++)
@@ -1114,6 +1232,70 @@ out:
}
/**
+ * igb_translate_register_82576 - Translate the proper register offset
+ * @reg: e1000 register to be read
+ *
+ * Registers in 82576 are located in different offsets than other adapters
+ * even though they function in the same manner. This function takes in
+ * the name of the register to read and returns the correct offset for
+ * 82576 silicon.
+ **/
+u32 igb_translate_register_82576(u32 reg)
+{
+ /*
+ * Some of the Kawela registers are located at different
+ * offsets than they are in older adapters.
+ * Despite the difference in location, the registers
+ * function in the same manner.
+ */
+ switch (reg) {
+ case E1000_TDBAL(0):
+ reg = 0x0E000;
+ break;
+ case E1000_TDBAH(0):
+ reg = 0x0E004;
+ break;
+ case E1000_TDLEN(0):
+ reg = 0x0E008;
+ break;
+ case E1000_TDH(0):
+ reg = 0x0E010;
+ break;
+ case E1000_TDT(0):
+ reg = 0x0E018;
+ break;
+ case E1000_TXDCTL(0):
+ reg = 0x0E028;
+ break;
+ case E1000_RDBAL(0):
+ reg = 0x0C000;
+ break;
+ case E1000_RDBAH(0):
+ reg = 0x0C004;
+ break;
+ case E1000_RDLEN(0):
+ reg = 0x0C008;
+ break;
+ case E1000_RDH(0):
+ reg = 0x0C010;
+ break;
+ case E1000_RDT(0):
+ reg = 0x0C018;
+ break;
+ case E1000_RXDCTL(0):
+ reg = 0x0C028;
+ break;
+ case E1000_SRRCTL(0):
+ reg = 0x0C00C;
+ break;
+ default:
+ break;
+ }
+
+ return reg;
+}
+
+/**
* igb_reset_init_script_82575 - Inits HW defaults after reset
* @hw: pointer to the HW structure
*
@@ -1304,7 +1486,7 @@ static struct e1000_mac_operations e1000_mac_ops_82575 = {
.reset_hw = igb_reset_hw_82575,
.init_hw = igb_init_hw_82575,
.check_for_link = igb_check_for_link_82575,
- .rar_set = igb_rar_set_82575,
+ .rar_set = igb_rar_set,
.read_mac_addr = igb_read_mac_addr_82575,
.get_speed_and_duplex = igb_get_speed_and_duplex_copper,
};
diff --git a/drivers/net/igb/e1000_82575.h b/drivers/net/igb/e1000_82575.h
index 02e57a8..d273236 100644
--- a/drivers/net/igb/e1000_82575.h
+++ b/drivers/net/igb/e1000_82575.h
@@ -28,9 +28,13 @@
#ifndef _E1000_82575_H_
#define _E1000_82575_H_
+u32 igb_translate_register_82576(u32 reg);
+void igb_update_mc_addr_list_82575(struct e1000_hw*, u8*, u32, u32, u32);
+extern void igb_shutdown_fiber_serdes_link_82575(struct e1000_hw *hw);
extern void igb_rx_fifo_flush_82575(struct e1000_hw *hw);
#define E1000_RAR_ENTRIES_82575 16
+#define E1000_RAR_ENTRIES_82576 24
/* SRRCTL bit definitions */
#define E1000_SRRCTL_BSIZEPKT_SHIFT 10 /* Shift _right_ */
@@ -159,5 +163,10 @@ struct e1000_adv_tx_context_desc {
#define E1000_DCA_TXCTRL_DESC_DCA_EN (1 << 5) /* DCA Tx Desc enable */
#define E1000_DCA_TXCTRL_TX_WB_RO_EN (1 << 11) /* Tx Desc writeback RO bit */
+/* Additional DCA related definitions, note change in position of CPUID */
+#define E1000_DCA_TXCTRL_CPUID_MASK_82576 0xFF000000 /* Tx CPUID Mask */
+#define E1000_DCA_RXCTRL_CPUID_MASK_82576 0xFF000000 /* Rx CPUID Mask */
+#define E1000_DCA_TXCTRL_CPUID_SHIFT 24 /* Tx CPUID now in the last byte */
+#define E1000_DCA_RXCTRL_CPUID_SHIFT 24 /* Rx CPUID now in the last byte */
#endif
diff --git a/drivers/net/igb/e1000_defines.h b/drivers/net/igb/e1000_defines.h
index ed748dc..afdba3c 100644
--- a/drivers/net/igb/e1000_defines.h
+++ b/drivers/net/igb/e1000_defines.h
@@ -90,6 +90,11 @@
#define E1000_I2CCMD_ERROR 0x80000000
#define E1000_MAX_SGMII_PHY_REG_ADDR 255
#define E1000_I2CCMD_PHY_TIMEOUT 200
+#define E1000_IVAR_VALID 0x80
+#define E1000_GPIE_NSICR 0x00000001
+#define E1000_GPIE_MSIX_MODE 0x00000010
+#define E1000_GPIE_EIAME 0x40000000
+#define E1000_GPIE_PBA 0x80000000
/* Receive Descriptor bit definitions */
#define E1000_RXD_STAT_DD 0x01 /* Descriptor Done */
@@ -213,6 +218,7 @@
/* Device Control */
#define E1000_CTRL_FD 0x00000001 /* Full duplex.0=half; 1=full */
#define E1000_CTRL_GIO_MASTER_DISABLE 0x00000004 /*Blocks new Master requests */
+#define E1000_CTRL_LRST 0x00000008 /* Link reset. 0=normal,1=reset */
#define E1000_CTRL_ASDE 0x00000020 /* Auto-speed detect enable */
#define E1000_CTRL_SLU 0x00000040 /* Set link up (Force Link) */
#define E1000_CTRL_ILOS 0x00000080 /* Invert Loss-Of Signal */
@@ -244,6 +250,7 @@
*/
#define E1000_CONNSW_ENRGSRC 0x4
+#define E1000_PCS_CFG_PCS_EN 8
#define E1000_PCS_LCTL_FLV_LINK_UP 1
#define E1000_PCS_LCTL_FSV_100 2
#define E1000_PCS_LCTL_FSV_1000 4
@@ -253,6 +260,7 @@
#define E1000_PCS_LCTL_AN_ENABLE 0x10000
#define E1000_PCS_LCTL_AN_RESTART 0x20000
#define E1000_PCS_LCTL_AN_TIMEOUT 0x40000
+#define E1000_ENABLE_SERDES_LOOPBACK 0x0410
#define E1000_PCS_LSTS_LINK_OK 1
#define E1000_PCS_LSTS_SPEED_100 2
@@ -360,6 +368,7 @@
#define E1000_PBA_16K 0x0010 /* 16KB, default TX allocation */
#define E1000_PBA_24K 0x0018
#define E1000_PBA_34K 0x0022
+#define E1000_PBA_64K 0x0040 /* 64KB */
#define IFS_MAX 80
#define IFS_MIN 40
@@ -528,6 +537,7 @@
/* PHY Control Register */
#define MII_CR_FULL_DUPLEX 0x0100 /* FDX =1, half duplex =0 */
#define MII_CR_RESTART_AUTO_NEG 0x0200 /* Restart auto negotiation */
+#define MII_CR_POWER_DOWN 0x0800 /* Power down */
#define MII_CR_AUTO_NEG_EN 0x1000 /* Auto Neg Enable */
#define MII_CR_LOOPBACK 0x4000 /* 0 = normal, 1 = loopback */
#define MII_CR_RESET 0x8000 /* 0 = normal, 1 = PHY reset */
diff --git a/drivers/net/igb/e1000_hw.h b/drivers/net/igb/e1000_hw.h
index 746c3ea..19fa4ee 100644
--- a/drivers/net/igb/e1000_hw.h
+++ b/drivers/net/igb/e1000_hw.h
@@ -38,6 +38,10 @@
struct e1000_hw;
+#define E1000_DEV_ID_82576 0x10C9
+#define E1000_DEV_ID_82576_FIBER 0x10E6
+#define E1000_DEV_ID_82576_SERDES 0x10E7
+#define E1000_DEV_ID_82576_QUAD_COPPER 0x10E8
#define E1000_DEV_ID_82575EB_COPPER 0x10A7
#define E1000_DEV_ID_82575EB_FIBER_SERDES 0x10A9
#define E1000_DEV_ID_82575GB_QUAD_COPPER 0x10D6
@@ -50,6 +54,7 @@ struct e1000_hw;
enum e1000_mac_type {
e1000_undefined = 0,
e1000_82575,
+ e1000_82576,
e1000_num_macs /* List is 1-based, so subtract 1 for true count. */
};
@@ -410,14 +415,17 @@ struct e1000_mac_operations {
s32 (*check_for_link)(struct e1000_hw *);
s32 (*reset_hw)(struct e1000_hw *);
s32 (*init_hw)(struct e1000_hw *);
+ bool (*check_mng_mode)(struct e1000_hw *);
s32 (*setup_physical_interface)(struct e1000_hw *);
void (*rar_set)(struct e1000_hw *, u8 *, u32);
s32 (*read_mac_addr)(struct e1000_hw *);
s32 (*get_speed_and_duplex)(struct e1000_hw *, u16 *, u16 *);
+ void (*mta_set)(struct e1000_hw *, u32);
};
struct e1000_phy_operations {
s32 (*acquire_phy)(struct e1000_hw *);
+ s32 (*check_reset_block)(struct e1000_hw *);
s32 (*force_speed_duplex)(struct e1000_hw *);
s32 (*get_cfg_done)(struct e1000_hw *hw);
s32 (*get_cable_length)(struct e1000_hw *);
diff --git a/drivers/net/igb/e1000_mac.c b/drivers/net/igb/e1000_mac.c
index 47ad2c4..20408aa 100644
--- a/drivers/net/igb/e1000_mac.c
+++ b/drivers/net/igb/e1000_mac.c
@@ -36,7 +36,6 @@
static s32 igb_set_default_fc(struct e1000_hw *hw);
static s32 igb_set_fc_watermarks(struct e1000_hw *hw);
-static u32 igb_hash_mc_addr(struct e1000_hw *hw, u8 *mc_addr);
/**
* igb_remove_device - Free device specific structure
@@ -360,7 +359,7 @@ void igb_update_mc_addr_list(struct e1000_hw *hw,
* the multicast filter table array address and new table value. See
* igb_mta_set()
**/
-static u32 igb_hash_mc_addr(struct e1000_hw *hw, u8 *mc_addr)
+u32 igb_hash_mc_addr(struct e1000_hw *hw, u8 *mc_addr)
{
u32 hash_value, hash_mask;
u8 bit_shift = 0;
diff --git a/drivers/net/igb/e1000_mac.h b/drivers/net/igb/e1000_mac.h
index 326b659..dc2f8cc 100644
--- a/drivers/net/igb/e1000_mac.h
+++ b/drivers/net/igb/e1000_mac.h
@@ -94,5 +94,6 @@ enum e1000_mng_mode {
#define E1000_HICR_C 0x02
extern void e1000_init_function_pointers_82575(struct e1000_hw *hw);
+extern u32 igb_hash_mc_addr(struct e1000_hw *hw, u8 *mc_addr);
#endif
diff --git a/drivers/net/igb/e1000_regs.h b/drivers/net/igb/e1000_regs.h
index d25e914..b95093d 100644
--- a/drivers/net/igb/e1000_regs.h
+++ b/drivers/net/igb/e1000_regs.h
@@ -56,6 +56,9 @@
#define E1000_EIMC 0x01528 /* Ext. Interrupt Mask Clear - WO */
#define E1000_EIAC 0x0152C /* Ext. Interrupt Auto Clear - RW */
#define E1000_EIAM 0x01530 /* Ext. Interrupt Ack Auto Clear Mask - RW */
+#define E1000_GPIE 0x01514 /* General Purpose Interrupt Enable - RW */
+#define E1000_IVAR0 0x01700 /* Interrupt Vector Allocation (array) - RW */
+#define E1000_IVAR_MISC 0x01740 /* IVAR for "other" causes - RW */
#define E1000_TCTL 0x00400 /* TX Control - RW */
#define E1000_TCTL_EXT 0x00404 /* Extended TX Control - RW */
#define E1000_TIPG 0x00410 /* TX Inter-packet gap -RW */
@@ -217,6 +220,7 @@
#define E1000_RFCTL 0x05008 /* Receive Filter Control*/
#define E1000_MTA 0x05200 /* Multicast Table Array - RW Array */
#define E1000_RA 0x05400 /* Receive Address - RW Array */
+#define E1000_RA2 0x054E0 /* 2nd half of receive address array - RW Array */
#define E1000_VFTA 0x05600 /* VLAN Filter Table Array - RW Array */
#define E1000_VMD_CTL 0x0581C /* VMDq Control - RW */
#define E1000_WUC 0x05800 /* Wakeup Control - RW */
@@ -258,7 +262,8 @@
#define E1000_RETA(_i) (0x05C00 + ((_i) * 4))
#define E1000_RSSRK(_i) (0x05C80 + ((_i) * 4)) /* RSS Random Key - RW Array */
-#define E1000_REGISTER(a, reg) reg
+#define E1000_REGISTER(a, reg) (((a)->mac.type < e1000_82576) \
+ ? reg : e1000_translate_register_82576(reg))
#define wr32(reg, value) (writel(value, hw->hw_addr + reg))
#define rd32(reg) (readl(hw->hw_addr + reg))
diff --git a/drivers/net/igb/igb_ethtool.c b/drivers/net/igb/igb_ethtool.c
index ed756c1..e27d5a5 100644
--- a/drivers/net/igb/igb_ethtool.c
+++ b/drivers/net/igb/igb_ethtool.c
@@ -827,8 +827,9 @@ err_setup:
/* ethtool register test data */
struct igb_reg_test {
u16 reg;
- u8 array_len;
- u8 test_type;
+ u16 reg_offset;
+ u16 array_len;
+ u16 test_type;
u32 mask;
u32 write;
};
@@ -850,34 +851,72 @@ struct igb_reg_test {
#define TABLE64_TEST_LO 5
#define TABLE64_TEST_HI 6
-/* default register test */
+/* 82576 reg test */
+static struct igb_reg_test reg_test_82576[] = {
+ { E1000_FCAL, 0x100, 1, PATTERN_TEST, 0xFFFFFFFF, 0xFFFFFFFF },
+ { E1000_FCAH, 0x100, 1, PATTERN_TEST, 0x0000FFFF, 0xFFFFFFFF },
+ { E1000_FCT, 0x100, 1, PATTERN_TEST, 0x0000FFFF, 0xFFFFFFFF },
+ { E1000_VET, 0x100, 1, PATTERN_TEST, 0xFFFFFFFF, 0xFFFFFFFF },
+ { E1000_RDBAL(0), 0x100, 4, PATTERN_TEST, 0xFFFFFF80, 0xFFFFFFFF },
+ { E1000_RDBAH(0), 0x100, 4, PATTERN_TEST, 0xFFFFFFFF, 0xFFFFFFFF },
+ { E1000_RDLEN(0), 0x100, 4, PATTERN_TEST, 0x000FFFF0, 0x000FFFFF },
+ { E1000_RDBAL(4), 0x40, 8, PATTERN_TEST, 0xFFFFFF80, 0xFFFFFFFF },
+ { E1000_RDBAH(4), 0x40, 8, PATTERN_TEST, 0xFFFFFFFF, 0xFFFFFFFF },
+ { E1000_RDLEN(4), 0x40, 8, PATTERN_TEST, 0x000FFFF0, 0x000FFFFF },
+ /* Enable all four RX queues before testing. */
+ { E1000_RXDCTL(0), 0x100, 1, WRITE_NO_TEST, 0, E1000_RXDCTL_QUEUE_ENABLE },
+ /* RDH is read-only for 82576, only test RDT. */
+ { E1000_RDT(0), 0x100, 4, PATTERN_TEST, 0x0000FFFF, 0x0000FFFF },
+ { E1000_RXDCTL(0), 0x100, 4, WRITE_NO_TEST, 0, 0 },
+ { E1000_FCRTH, 0x100, 1, PATTERN_TEST, 0x0000FFF0, 0x0000FFF0 },
+ { E1000_FCTTV, 0x100, 1, PATTERN_TEST, 0x0000FFFF, 0x0000FFFF },
+ { E1000_TIPG, 0x100, 1, PATTERN_TEST, 0x3FFFFFFF, 0x3FFFFFFF },
+ { E1000_TDBAL(0), 0x100, 4, PATTERN_TEST, 0xFFFFFF80, 0xFFFFFFFF },
+ { E1000_TDBAH(0), 0x100, 4, PATTERN_TEST, 0xFFFFFFFF, 0xFFFFFFFF },
+ { E1000_TDLEN(0), 0x100, 4, PATTERN_TEST, 0x000FFFF0, 0x000FFFFF },
+ { E1000_TDBAL(4), 0x40, 8, PATTERN_TEST, 0xFFFFFF80, 0xFFFFFFFF },
+ { E1000_TDBAH(4), 0x40, 8, PATTERN_TEST, 0xFFFFFFFF, 0xFFFFFFFF },
+ { E1000_TDLEN(4), 0x40, 8, PATTERN_TEST, 0x000FFFF0, 0x000FFFFF },
+ { E1000_RCTL, 0x100, 1, SET_READ_TEST, 0xFFFFFFFF, 0x00000000 },
+ { E1000_RCTL, 0x100, 1, SET_READ_TEST, 0x04CFB0FE, 0x003FFFFB },
+ { E1000_RCTL, 0x100, 1, SET_READ_TEST, 0x04CFB0FE, 0xFFFFFFFF },
+ { E1000_TCTL, 0x100, 1, SET_READ_TEST, 0xFFFFFFFF, 0x00000000 },
+ { E1000_RA, 0, 16, TABLE64_TEST_LO, 0xFFFFFFFF, 0xFFFFFFFF },
+ { E1000_RA, 0, 16, TABLE64_TEST_HI, 0x83FFFFFF, 0xFFFFFFFF },
+ { E1000_RA2, 0, 8, TABLE64_TEST_LO, 0xFFFFFFFF, 0xFFFFFFFF },
+ { E1000_RA2, 0, 8, TABLE64_TEST_HI, 0x83FFFFFF, 0xFFFFFFFF },
+ { E1000_MTA, 0, 128,TABLE32_TEST, 0xFFFFFFFF, 0xFFFFFFFF },
+ { 0, 0, 0, 0 }
+};
+
+/* 82575 register test */
static struct igb_reg_test reg_test_82575[] = {
- { E1000_FCAL, 1, PATTERN_TEST, 0xFFFFFFFF, 0xFFFFFFFF },
- { E1000_FCAH, 1, PATTERN_TEST, 0x0000FFFF, 0xFFFFFFFF },
- { E1000_FCT, 1, PATTERN_TEST, 0x0000FFFF, 0xFFFFFFFF },
- { E1000_VET, 1, PATTERN_TEST, 0xFFFFFFFF, 0xFFFFFFFF },
- { E1000_RDBAL(0), 4, PATTERN_TEST, 0xFFFFFF80, 0xFFFFFFFF },
- { E1000_RDBAH(0), 4, PATTERN_TEST, 0xFFFFFFFF, 0xFFFFFFFF },
- { E1000_RDLEN(0), 4, PATTERN_TEST, 0x000FFF80, 0x000FFFFF },
+ { E1000_FCAL, 0x100, 1, PATTERN_TEST, 0xFFFFFFFF, 0xFFFFFFFF },
+ { E1000_FCAH, 0x100, 1, PATTERN_TEST, 0x0000FFFF, 0xFFFFFFFF },
+ { E1000_FCT, 0x100, 1, PATTERN_TEST, 0x0000FFFF, 0xFFFFFFFF },
+ { E1000_VET, 0x100, 1, PATTERN_TEST, 0xFFFFFFFF, 0xFFFFFFFF },
+ { E1000_RDBAL(0), 0x100, 4, PATTERN_TEST, 0xFFFFFF80, 0xFFFFFFFF },
+ { E1000_RDBAH(0), 0x100, 4, PATTERN_TEST, 0xFFFFFFFF, 0xFFFFFFFF },
+ { E1000_RDLEN(0), 0x100, 4, PATTERN_TEST, 0x000FFF80, 0x000FFFFF },
/* Enable all four RX queues before testing. */
- { E1000_RXDCTL(0), 4, WRITE_NO_TEST, 0, E1000_RXDCTL_QUEUE_ENABLE },
+ { E1000_RXDCTL(0), 0x100, 4, WRITE_NO_TEST, 0, E1000_RXDCTL_QUEUE_ENABLE },
/* RDH is read-only for 82575, only test RDT. */
- { E1000_RDT(0), 4, PATTERN_TEST, 0x0000FFFF, 0x0000FFFF },
- { E1000_RXDCTL(0), 4, WRITE_NO_TEST, 0, 0 },
- { E1000_FCRTH, 1, PATTERN_TEST, 0x0000FFF0, 0x0000FFF0 },
- { E1000_FCTTV, 1, PATTERN_TEST, 0x0000FFFF, 0x0000FFFF },
- { E1000_TIPG, 1, PATTERN_TEST, 0x3FFFFFFF, 0x3FFFFFFF },
- { E1000_TDBAL(0), 4, PATTERN_TEST, 0xFFFFFF80, 0xFFFFFFFF },
- { E1000_TDBAH(0), 4, PATTERN_TEST, 0xFFFFFFFF, 0xFFFFFFFF },
- { E1000_TDLEN(0), 4, PATTERN_TEST, 0x000FFF80, 0x000FFFFF },
- { E1000_RCTL, 1, SET_READ_TEST, 0xFFFFFFFF, 0x00000000 },
- { E1000_RCTL, 1, SET_READ_TEST, 0x04CFB3FE, 0x003FFFFB },
- { E1000_RCTL, 1, SET_READ_TEST, 0x04CFB3FE, 0xFFFFFFFF },
- { E1000_TCTL, 1, SET_READ_TEST, 0xFFFFFFFF, 0x00000000 },
- { E1000_TXCW, 1, PATTERN_TEST, 0xC000FFFF, 0x0000FFFF },
- { E1000_RA, 16, TABLE64_TEST_LO, 0xFFFFFFFF, 0xFFFFFFFF },
- { E1000_RA, 16, TABLE64_TEST_HI, 0x800FFFFF, 0xFFFFFFFF },
- { E1000_MTA, 128, TABLE32_TEST, 0xFFFFFFFF, 0xFFFFFFFF },
+ { E1000_RDT(0), 0x100, 4, PATTERN_TEST, 0x0000FFFF, 0x0000FFFF },
+ { E1000_RXDCTL(0), 0x100, 4, WRITE_NO_TEST, 0, 0 },
+ { E1000_FCRTH, 0x100, 1, PATTERN_TEST, 0x0000FFF0, 0x0000FFF0 },
+ { E1000_FCTTV, 0x100, 1, PATTERN_TEST, 0x0000FFFF, 0x0000FFFF },
+ { E1000_TIPG, 0x100, 1, PATTERN_TEST, 0x3FFFFFFF, 0x3FFFFFFF },
+ { E1000_TDBAL(0), 0x100, 4, PATTERN_TEST, 0xFFFFFF80, 0xFFFFFFFF },
+ { E1000_TDBAH(0), 0x100, 4, PATTERN_TEST, 0xFFFFFFFF, 0xFFFFFFFF },
+ { E1000_TDLEN(0), 0x100, 4, PATTERN_TEST, 0x000FFF80, 0x000FFFFF },
+ { E1000_RCTL, 0x100, 1, SET_READ_TEST, 0xFFFFFFFF, 0x00000000 },
+ { E1000_RCTL, 0x100, 1, SET_READ_TEST, 0x04CFB3FE, 0x003FFFFB },
+ { E1000_RCTL, 0x100, 1, SET_READ_TEST, 0x04CFB3FE, 0xFFFFFFFF },
+ { E1000_TCTL, 0x100, 1, SET_READ_TEST, 0xFFFFFFFF, 0x00000000 },
+ { E1000_TXCW, 0x100, 1, PATTERN_TEST, 0xC000FFFF, 0x0000FFFF },
+ { E1000_RA, 0, 16, TABLE64_TEST_LO, 0xFFFFFFFF, 0xFFFFFFFF },
+ { E1000_RA, 0, 16, TABLE64_TEST_HI, 0x800FFFFF, 0xFFFFFFFF },
+ { E1000_MTA, 0, 128, TABLE32_TEST, 0xFFFFFFFF, 0xFFFFFFFF },
{ 0, 0, 0, 0 }
};
@@ -937,7 +976,15 @@ static int igb_reg_test(struct igb_adapter *adapter, u64 *data)
u32 i, toggle;
toggle = 0x7FFFF3FF;
- test = reg_test_82575;
+
+ switch (adapter->hw.mac.type) {
+ case e1000_82576:
+ test = reg_test_82576;
+ break;
+ default:
+ test = reg_test_82575;
+ break;
+ }
/* Because the status register is such a special case,
* we handle it separately from the rest of the register
@@ -964,19 +1011,19 @@ static int igb_reg_test(struct igb_adapter *adapter, u64 *data)
for (i = 0; i < test->array_len; i++) {
switch (test->test_type) {
case PATTERN_TEST:
- REG_PATTERN_TEST(test->reg + (i * 0x100),
+ REG_PATTERN_TEST(test->reg + (i * test->reg_offset),
test->mask,
test->write);
break;
case SET_READ_TEST:
- REG_SET_AND_CHECK(test->reg + (i * 0x100),
+ REG_SET_AND_CHECK(test->reg + (i * test->reg_offset),
test->mask,
test->write);
break;
case WRITE_NO_TEST:
writel(test->write,
(adapter->hw.hw_addr + test->reg)
- + (i * 0x100));
+ + (i * test->reg_offset));
break;
case TABLE32_TEST:
REG_PATTERN_TEST(test->reg + (i * 4),
@@ -1392,13 +1439,39 @@ static int igb_set_phy_loopback(struct igb_adapter *adapter)
static int igb_setup_loopback_test(struct igb_adapter *adapter)
{
struct e1000_hw *hw = &adapter->hw;
- u32 rctl;
+ u32 reg;
if (hw->phy.media_type == e1000_media_type_fiber ||
hw->phy.media_type == e1000_media_type_internal_serdes) {
- rctl = rd32(E1000_RCTL);
- rctl |= E1000_RCTL_LBM_TCVR;
- wr32(E1000_RCTL, rctl);
+ reg = rd32(E1000_RCTL);
+ reg |= E1000_RCTL_LBM_TCVR;
+ wr32(E1000_RCTL, reg);
+
+ wr32(E1000_SCTL, E1000_ENABLE_SERDES_LOOPBACK);
+
+ reg = rd32(E1000_CTRL);
+ reg &= ~(E1000_CTRL_RFCE |
+ E1000_CTRL_TFCE |
+ E1000_CTRL_LRST);
+ reg |= E1000_CTRL_SLU |
+ E1000_CTRL_FD;
+ wr32(E1000_CTRL, reg);
+
+ /* Unset switch control to serdes energy detect */
+ reg = rd32(E1000_CONNSW);
+ reg &= ~E1000_CONNSW_ENRGSRC;
+ wr32(E1000_CONNSW, reg);
+
+ /* Set PCS register for forced speed */
+ reg = rd32(E1000_PCS_LCTL);
+ reg &= ~E1000_PCS_LCTL_AN_ENABLE; /* Disable Autoneg*/
+ reg |= E1000_PCS_LCTL_FLV_LINK_UP | /* Force link up */
+ E1000_PCS_LCTL_FSV_1000 | /* Force 1000 */
+ E1000_PCS_LCTL_FDV_FULL | /* SerDes Full duplex */
+ E1000_PCS_LCTL_FSD | /* Force Speed */
+ E1000_PCS_LCTL_FORCE_LINK; /* Force Link */
+ wr32(E1000_PCS_LCTL, reg);
+
return 0;
} else if (hw->phy.media_type == e1000_media_type_copper) {
return igb_set_phy_loopback(adapter);
@@ -1654,10 +1727,13 @@ static int igb_wol_exclusion(struct igb_adapter *adapter,
switch (hw->device_id) {
case E1000_DEV_ID_82575GB_QUAD_COPPER:
+ case E1000_DEV_ID_82576_QUAD_COPPER:
/* WoL not supported */
wol->supported = 0;
break;
case E1000_DEV_ID_82575EB_FIBER_SERDES:
+ case E1000_DEV_ID_82576_FIBER:
+ case E1000_DEV_ID_82576_SERDES:
/* Wake events not supported on port B */
if (rd32(E1000_STATUS) & E1000_STATUS_FUNC_1) {
wol->supported = 0;
diff --git a/drivers/net/igb/igb_main.c b/drivers/net/igb/igb_main.c
index 1436326..ba043c4 100644
--- a/drivers/net/igb/igb_main.c
+++ b/drivers/net/igb/igb_main.c
@@ -51,7 +51,7 @@ char igb_driver_name[] = "igb";
char igb_driver_version[] = DRV_VERSION;
static const char igb_driver_string[] =
"Intel(R) Gigabit Ethernet Network Driver";
-static const char igb_copyright[] = "Copyright (c) 2007 Intel Corporation.";
+static const char igb_copyright[] = "Copyright (c) 2008 Intel Corporation.";
static const struct e1000_info *igb_info_tbl[] = {
@@ -59,6 +59,10 @@ static const struct e1000_info *igb_info_tbl[] = {
};
static struct pci_device_id igb_pci_tbl[] = {
+ { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576), board_82575 },
+ { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_FIBER), board_82575 },
+ { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_SERDES), board_82575 },
+ { PCI_VDEVICE(INTEL, E1000_DEV_ID_82576_QUAD_COPPER), board_82575 },
{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82575EB_COPPER), board_82575 },
{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82575EB_FIBER_SERDES), board_82575 },
{ PCI_VDEVICE(INTEL, E1000_DEV_ID_82575GB_QUAD_COPPER), board_82575 },
@@ -268,6 +272,10 @@ static void igb_assign_vector(struct igb_adapter *adapter, int rx_queue,
{
u32 msixbm = 0;
struct e1000_hw *hw = &adapter->hw;
+ u32 ivar, index;
+
+ switch (hw->mac.type) {
+ case e1000_82575:
/* The 82575 assigns vectors using a bitmask, which matches the
bitmask for the EICR/EIMS/EIMC registers. To assign one
or more queues to a vector, we write the appropriate bits
@@ -282,6 +290,47 @@ static void igb_assign_vector(struct igb_adapter *adapter, int rx_queue,
E1000_EICR_TX_QUEUE0 << tx_queue;
}
array_wr32(E1000_MSIXBM(0), msix_vector, msixbm);
+ break;
+ case e1000_82576:
+ /* Kawela uses a table-based method for assigning vectors.
+ Each queue has a single entry in the table to which we write
+ a vector number along with a "valid" bit. Sadly, the layout
+ of the table is somewhat counterintuitive. */
+ if (rx_queue > IGB_N0_QUEUE) {
+ index = (rx_queue & 0x7);
+ ivar = array_rd32(E1000_IVAR0, index);
+ if (rx_queue < 8) {
+ /* vector goes into low byte of register */
+ ivar = ivar & 0xFFFFFF00;
+ ivar |= msix_vector | E1000_IVAR_VALID;
+ } else {
+ /* vector goes into third byte of register */
+ ivar = ivar & 0xFF00FFFF;
+ ivar |= (msix_vector | E1000_IVAR_VALID) << 16;
+ }
+ adapter->rx_ring[rx_queue].eims_value= 1 << msix_vector;
+ array_wr32(E1000_IVAR0, index, ivar);
+ }
+ if (tx_queue > IGB_N0_QUEUE) {
+ index = (tx_queue & 0x7);
+ ivar = array_rd32(E1000_IVAR0, index);
+ if (tx_queue < 8) {
+ /* vector goes into second byte of register */
+ ivar = ivar & 0xFFFF00FF;
+ ivar |= (msix_vector | E1000_IVAR_VALID) << 8;
+ } else {
+ /* vector goes into high byte of register */
+ ivar = ivar & 0x00FFFFFF;
+ ivar |= (msix_vector | E1000_IVAR_VALID) << 24;
+ }
+ adapter->tx_ring[tx_queue].eims_value= 1 << msix_vector;
+ array_wr32(E1000_IVAR0, index, ivar);
+ }
+ break;
+ default:
+ BUG();
+ break;
+ }
}
/**
@@ -297,6 +346,12 @@ static void igb_configure_msix(struct igb_adapter *adapter)
struct e1000_hw *hw = &adapter->hw;
adapter->eims_enable_mask = 0;
+ if (hw->mac.type == e1000_82576)
+ /* Turn on MSI-X capability first, or our settings
+ * won't stick. And it will take days to debug. */
+ wr32(E1000_GPIE, E1000_GPIE_MSIX_MODE |
+ E1000_GPIE_PBA | E1000_GPIE_EIAME |
+ E1000_GPIE_NSICR);
for (i = 0; i < adapter->num_tx_queues; i++) {
struct igb_ring *tx_ring = &adapter->tx_ring[i];
@@ -322,6 +377,8 @@ static void igb_configure_msix(struct igb_adapter *adapter)
/* set vector for other causes, i.e. link changes */
+ switch (hw->mac.type) {
+ case e1000_82575:
array_wr32(E1000_MSIXBM(0), vector++,
E1000_EIMS_OTHER);
@@ -337,6 +394,19 @@ static void igb_configure_msix(struct igb_adapter *adapter)
adapter->eims_enable_mask |= E1000_EIMS_OTHER;
adapter->eims_other = E1000_EIMS_OTHER;
+ break;
+
+ case e1000_82576:
+ tmp = (vector++ | E1000_IVAR_VALID) << 8;
+ wr32(E1000_IVAR_MISC, tmp);
+
+ adapter->eims_enable_mask = (1 << (vector)) - 1;
+ adapter->eims_other = 1 << (vector - 1);
+ break;
+ default:
+ /* do nothing, since nothing else supports MSI-X */
+ break;
+ } /* switch (hw->mac.type) */
wrfl();
}
@@ -474,8 +544,17 @@ static int igb_request_irq(struct igb_adapter *adapter)
adapter->num_rx_queues = 1;
igb_alloc_queues(adapter);
} else {
- wr32(E1000_MSIXBM(0), (E1000_EICR_RX_QUEUE0 |
- E1000_EIMS_OTHER));
+ switch (hw->mac.type) {
+ case e1000_82575:
+ wr32(E1000_MSIXBM(0),
+ (E1000_EICR_RX_QUEUE0 | E1000_EIMS_OTHER));
+ break;
+ case e1000_82576:
+ wr32(E1000_IVAR0, E1000_IVAR_VALID);
+ break;
+ default:
+ break;
+ }
}
if (adapter->msi_enabled) {
@@ -770,16 +849,23 @@ void igb_reinit_locked(struct igb_adapter *adapter)
void igb_reset(struct igb_adapter *adapter)
{
struct e1000_hw *hw = &adapter->hw;
- struct e1000_fc_info *fc = &adapter->hw.fc;
+ struct e1000_mac_info *mac = &hw->mac;
+ struct e1000_fc_info *fc = &hw->fc;
u32 pba = 0, tx_space, min_tx_space, min_rx_space;
u16 hwm;
/* Repartition Pba for greater than 9k mtu
* To take effect CTRL.RST is required.
*/
+ if (mac->type != e1000_82576) {
pba = E1000_PBA_34K;
+ }
+ else {
+ pba = E1000_PBA_64K;
+ }
- if (adapter->max_frame_size > ETH_FRAME_LEN + ETH_FCS_LEN) {
+ if ((adapter->max_frame_size > ETH_FRAME_LEN + ETH_FCS_LEN) &&
+ (mac->type < e1000_82576)) {
/* adjust PBA for jumbo frames */
wr32(E1000_PBA, pba);
@@ -818,8 +904,8 @@ void igb_reset(struct igb_adapter *adapter)
if (pba < min_rx_space)
pba = min_rx_space;
}
+ wr32(E1000_PBA, pba);
}
- wr32(E1000_PBA, pba);
/* flow control settings */
/* The high water mark must be low enough to fit one full frame
@@ -828,10 +914,15 @@ void igb_reset(struct igb_adapter *adapter)
* - 90% of the Rx FIFO size, or
* - the full Rx FIFO size minus one full frame */
hwm = min(((pba << 10) * 9 / 10),
- ((pba << 10) - adapter->max_frame_size));
+ ((pba << 10) - 2 * adapter->max_frame_size));
- fc->high_water = hwm & 0xFFF8; /* 8-byte granularity */
- fc->low_water = fc->high_water - 8;
+ if (mac->type < e1000_82576) {
+ fc->high_water = hwm & 0xFFF8; /* 8-byte granularity */
+ fc->low_water = fc->high_water - 8;
+ } else {
+ fc->high_water = hwm & 0xFFF0; /* 16-byte granularity */
+ fc->low_water = fc->high_water - 16;
+ }
fc->pause_time = 0xFFFF;
fc->send_xon = 1;
fc->type = fc->original_type;
@@ -1118,9 +1209,12 @@ static int __devinit igb_probe(struct pci_dev *pdev,
* lan on a particular port */
switch (pdev->device) {
case E1000_DEV_ID_82575GB_QUAD_COPPER:
+ case E1000_DEV_ID_82576_QUAD_COPPER:
adapter->eeprom_wol = 0;
break;
case E1000_DEV_ID_82575EB_FIBER_SERDES:
+ case E1000_DEV_ID_82576_FIBER:
+ case E1000_DEV_ID_82576_SERDES:
/* Wake events only supported on port A for dual fiber
* regardless of eeprom setting */
if (rd32(E1000_STATUS) & E1000_STATUS_FUNC_1)
@@ -1801,7 +1895,10 @@ static void igb_configure_rx(struct igb_adapter *adapter)
get_random_bytes(&random[0], 40);
- shift = 6;
+ if (hw->mac.type >= e1000_82576)
+ shift = 0;
+ else
+ shift = 6;
for (j = 0; j < (32 * 4); j++) {
reta.bytes[j & 3] =
(j % adapter->num_rx_queues) << shift;
@@ -2127,7 +2224,7 @@ static void igb_set_multi(struct net_device *netdev)
if (!netdev->mc_count) {
/* nothing to program, so clear mc list */
- igb_update_mc_addr_list(hw, NULL, 0, 1,
+ igb_update_mc_addr_list_82575(hw, NULL, 0, 1,
mac->rar_entry_count);
return;
}
@@ -2145,7 +2242,8 @@ static void igb_set_multi(struct net_device *netdev)
memcpy(mta_list + (i*ETH_ALEN), mc_ptr->dmi_addr, ETH_ALEN);
mc_ptr = mc_ptr->next;
}
- igb_update_mc_addr_list(hw, mta_list, i, 1, mac->rar_entry_count);
+ igb_update_mc_addr_list_82575(hw, mta_list, i, 1,
+ mac->rar_entry_count);
kfree(mta_list);
}
@@ -3211,8 +3309,14 @@ static void igb_update_rx_dca(struct igb_ring *rx_ring)
if (rx_ring->cpu != cpu) {
dca_rxctrl = rd32(E1000_DCA_RXCTRL(q));
- dca_rxctrl &= ~E1000_DCA_RXCTRL_CPUID_MASK;
- dca_rxctrl |= dca_get_tag(cpu);
+ if (hw->mac.type == e1000_82576) {
+ dca_rxctrl &= ~E1000_DCA_RXCTRL_CPUID_MASK_82576;
+ dca_rxctrl |= dca_get_tag(cpu) <<
+ E1000_DCA_RXCTRL_CPUID_SHIFT;
+ } else {
+ dca_rxctrl &= ~E1000_DCA_RXCTRL_CPUID_MASK;
+ dca_rxctrl |= dca_get_tag(cpu);
+ }
dca_rxctrl |= E1000_DCA_RXCTRL_DESC_DCA_EN;
dca_rxctrl |= E1000_DCA_RXCTRL_HEAD_DCA_EN;
dca_rxctrl |= E1000_DCA_RXCTRL_DATA_DCA_EN;
@@ -3232,8 +3336,14 @@ static void igb_update_tx_dca(struct igb_ring *tx_ring)
if (tx_ring->cpu != cpu) {
dca_txctrl = rd32(E1000_DCA_TXCTRL(q));
- dca_txctrl &= ~E1000_DCA_TXCTRL_CPUID_MASK;
- dca_txctrl |= dca_get_tag(cpu);
+ if (hw->mac.type == e1000_82576) {
+ dca_txctrl &= ~E1000_DCA_TXCTRL_CPUID_MASK_82576;
+ dca_txctrl |= dca_get_tag(cpu) <<
+ E1000_DCA_TXCTRL_CPUID_SHIFT;
+ } else {
+ dca_txctrl &= ~E1000_DCA_TXCTRL_CPUID_MASK;
+ dca_txctrl |= dca_get_tag(cpu);
+ }
dca_txctrl |= E1000_DCA_TXCTRL_DESC_DCA_EN;
wr32(E1000_DCA_TXCTRL(q), dca_txctrl);
tx_ring->cpu = cpu;
@@ -3572,7 +3682,7 @@ done_cleaning:
/* detected Tx unit hang */
dev_err(&adapter->pdev->dev,
"Detected Tx Unit Hang\n"
- " Tx Queue <%lu>\n"
+ " Tx Queue <%d>\n"
" TDH <%x>\n"
" TDT <%x>\n"
" next_to_use <%x>\n"
@@ -3582,8 +3692,7 @@ done_cleaning:
" time_stamp <%lx>\n"
" jiffies <%lx>\n"
" desc.status <%x>\n",
- (unsigned long)((tx_ring - adapter->tx_ring) /
- sizeof(struct igb_ring)),
+ tx_ring->queue_index,
readl(adapter->hw.hw_addr + tx_ring->head),
readl(adapter->hw.hw_addr + tx_ring->tail),
tx_ring->next_to_use,
@@ -4098,7 +4207,7 @@ static int igb_suspend(struct pci_dev *pdev, pm_message_t state)
struct net_device *netdev = pci_get_drvdata(pdev);
struct igb_adapter *adapter = netdev_priv(netdev);
struct e1000_hw *hw = &adapter->hw;
- u32 ctrl, ctrl_ext, rctl, status;
+ u32 ctrl, rctl, status;
u32 wufc = adapter->wol;
#ifdef CONFIG_PM
int retval = 0;
@@ -4141,33 +4250,24 @@ static int igb_suspend(struct pci_dev *pdev, pm_message_t state)
ctrl |= E1000_CTRL_ADVD3WUC;
wr32(E1000_CTRL, ctrl);
- if (adapter->hw.phy.media_type == e1000_media_type_fiber ||
- adapter->hw.phy.media_type ==
- e1000_media_type_internal_serdes) {
- /* keep the laser running in D3 */
- ctrl_ext = rd32(E1000_CTRL_EXT);
- ctrl_ext |= E1000_CTRL_EXT_SDP7_DATA;
- wr32(E1000_CTRL_EXT, ctrl_ext);
- }
-
/* Allow time for pending master requests to run */
igb_disable_pcie_master(&adapter->hw);
wr32(E1000_WUC, E1000_WUC_PME_EN);
wr32(E1000_WUFC, wufc);
- pci_enable_wake(pdev, PCI_D3hot, 1);
- pci_enable_wake(pdev, PCI_D3cold, 1);
} else {
wr32(E1000_WUC, 0);
wr32(E1000_WUFC, 0);
- pci_enable_wake(pdev, PCI_D3hot, 0);
- pci_enable_wake(pdev, PCI_D3cold, 0);
}
- /* make sure adapter isn't asleep if manageability is enabled */
- if (adapter->en_mng_pt) {
+ /* make sure adapter isn't asleep if manageability/wol is enabled */
+ if (wufc || adapter->en_mng_pt) {
pci_enable_wake(pdev, PCI_D3hot, 1);
pci_enable_wake(pdev, PCI_D3cold, 1);
+ } else {
+ igb_shutdown_fiber_serdes_link_82575(hw);
+ pci_enable_wake(pdev, PCI_D3hot, 0);
+ pci_enable_wake(pdev, PCI_D3cold, 0);
}
/* Release control of h/w to f/w. If f/w is AMT enabled, this
^ permalink raw reply related [flat|nested] 15+ messages in thread* [RESEND NET-NEXT PATCH 14/20] igb: Add support for quad port WOL and feature flags
2008-07-08 22:06 [RESEND NET-NEXT PATCH 00/20] igb update for net-next Jeff Kirsher
` (5 preceding siblings ...)
2008-07-08 22:10 ` [RESEND NET-NEXT PATCH 13/20] igb: add 82576 MAC support Jeff Kirsher
@ 2008-07-08 22:10 ` Jeff Kirsher
2008-07-08 22:11 ` [RESEND NET-NEXT PATCH 15/20] igb: add page recycling support Jeff Kirsher
` (5 subsequent siblings)
12 siblings, 0 replies; 15+ messages in thread
From: Jeff Kirsher @ 2008-07-08 22:10 UTC (permalink / raw)
To: jeff, davem; +Cc: netdev, linux-kernel, akpm, Alexander Duyck, Jeff Kirsher
From: Alexander Duyck <alexander.h.duyck@intel.com>
Change igb from using a series of boolean operators to using a single flags
value that contains a number of different bit flags for all the different
features of the adapter.
This patch also adds WOL support for quad port adapters.
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
---
drivers/net/igb/igb.h | 13 ++++--
drivers/net/igb/igb_ethtool.c | 12 ++++-
drivers/net/igb/igb_main.c | 92 ++++++++++++++++++++++++++++-------------
3 files changed, 81 insertions(+), 36 deletions(-)
diff --git a/drivers/net/igb/igb.h b/drivers/net/igb/igb.h
index d4a0423..ee08010 100644
--- a/drivers/net/igb/igb.h
+++ b/drivers/net/igb/igb.h
@@ -270,10 +270,7 @@ struct igb_adapter {
/* to not mess up cache alignment, always add to the bottom */
unsigned long state;
- unsigned int msi_enabled;
-#ifdef CONFIG_DCA
- unsigned int dca_enabled;
-#endif
+ unsigned int flags;
u32 eeprom_wol;
/* for ioport free */
@@ -285,6 +282,14 @@ struct igb_adapter {
#endif /* CONFIG_NETDEVICES_MULTIQUEUE */
};
+#define IGB_FLAG_HAS_MSI (1 << 0)
+#define IGB_FLAG_MSI_ENABLE (1 << 1)
+#define IGB_FLAG_HAS_DCA (1 << 2)
+#define IGB_FLAG_DCA_ENABLED (1 << 3)
+#define IGB_FLAG_IN_NETPOLL (1 << 5)
+#define IGB_FLAG_QUAD_PORT_A (1 << 6)
+#define IGB_FLAG_NEED_CTX_IDX (1 << 7)
+
enum e1000_state_t {
__IGB_TESTING,
__IGB_RESETTING,
diff --git a/drivers/net/igb/igb_ethtool.c b/drivers/net/igb/igb_ethtool.c
index e27d5a5..ef209b5 100644
--- a/drivers/net/igb/igb_ethtool.c
+++ b/drivers/net/igb/igb_ethtool.c
@@ -1097,7 +1097,7 @@ static int igb_intr_test(struct igb_adapter *adapter, u64 *data)
if (adapter->msix_entries) {
/* NOTE: we don't test MSI-X interrupts here, yet */
return 0;
- } else if (adapter->msi_enabled) {
+ } else if (adapter->flags & IGB_FLAG_HAS_MSI) {
shared_int = false;
if (request_irq(irq, &igb_test_intr, 0, netdev->name, netdev)) {
*data = 1;
@@ -1727,7 +1727,6 @@ static int igb_wol_exclusion(struct igb_adapter *adapter,
switch (hw->device_id) {
case E1000_DEV_ID_82575GB_QUAD_COPPER:
- case E1000_DEV_ID_82576_QUAD_COPPER:
/* WoL not supported */
wol->supported = 0;
break;
@@ -1742,6 +1741,15 @@ static int igb_wol_exclusion(struct igb_adapter *adapter,
/* return success for non excluded adapter ports */
retval = 0;
break;
+ case E1000_DEV_ID_82576_QUAD_COPPER:
+ /* quad port adapters only support WoL on port A */
+ if (!(adapter->flags & IGB_FLAG_QUAD_PORT_A)) {
+ wol->supported = 0;
+ break;
+ }
+ /* return success for non excluded adapter ports */
+ retval = 0;
+ break;
default:
/* dual port cards only support WoL on port A from now on
* unless it was enabled in the eeprom for port B
diff --git a/drivers/net/igb/igb_main.c b/drivers/net/igb/igb_main.c
index ba043c4..68a4fef 100644
--- a/drivers/net/igb/igb_main.c
+++ b/drivers/net/igb/igb_main.c
@@ -53,7 +53,6 @@ static const char igb_driver_string[] =
"Intel(R) Gigabit Ethernet Network Driver";
static const char igb_copyright[] = "Copyright (c) 2008 Intel Corporation.";
-
static const struct e1000_info *igb_info_tbl[] = {
[board_82575] = &e1000_82575_info,
};
@@ -170,6 +169,8 @@ static struct pci_driver igb_driver = {
.err_handler = &igb_err_handler
};
+static int global_quad_port_a; /* global quad port a indication */
+
MODULE_AUTHOR("Intel Corporation, <e1000-devel@lists.sourceforge.net>");
MODULE_DESCRIPTION("Intel(R) Gigabit Ethernet Network Driver");
MODULE_LICENSE("GPL");
@@ -201,6 +202,8 @@ static int __init igb_init_module(void)
printk(KERN_INFO "%s\n", igb_copyright);
+ global_quad_port_a = 0;
+
ret = pci_register_driver(&igb_driver);
#ifdef CONFIG_DCA
dca_register_notify(&dca_notifier);
@@ -471,7 +474,7 @@ static void igb_reset_interrupt_capability(struct igb_adapter *adapter)
pci_disable_msix(adapter->pdev);
kfree(adapter->msix_entries);
adapter->msix_entries = NULL;
- } else if (adapter->msi_enabled)
+ } else if (adapter->flags & IGB_FLAG_HAS_MSI)
pci_disable_msi(adapter->pdev);
return;
}
@@ -510,7 +513,7 @@ msi_only:
adapter->num_rx_queues = 1;
adapter->num_tx_queues = 1;
if (!pci_enable_msi(adapter->pdev))
- adapter->msi_enabled = 1;
+ adapter->flags |= IGB_FLAG_HAS_MSI;
#ifdef CONFIG_NETDEVICES_MULTIQUEUE
/* Notify the stack of the (possibly) reduced Tx Queue count. */
@@ -538,7 +541,7 @@ static int igb_request_irq(struct igb_adapter *adapter)
/* fall back to MSI */
igb_reset_interrupt_capability(adapter);
if (!pci_enable_msi(adapter->pdev))
- adapter->msi_enabled = 1;
+ adapter->flags |= IGB_FLAG_HAS_MSI;
igb_free_all_tx_resources(adapter);
igb_free_all_rx_resources(adapter);
adapter->num_rx_queues = 1;
@@ -557,14 +560,14 @@ static int igb_request_irq(struct igb_adapter *adapter)
}
}
- if (adapter->msi_enabled) {
+ if (adapter->flags & IGB_FLAG_HAS_MSI) {
err = request_irq(adapter->pdev->irq, &igb_intr_msi, 0,
netdev->name, netdev);
if (!err)
goto request_done;
/* fall back to legacy interrupts */
igb_reset_interrupt_capability(adapter);
- adapter->msi_enabled = 0;
+ adapter->flags &= ~IGB_FLAG_HAS_MSI;
}
err = request_irq(adapter->pdev->irq, &igb_intr, IRQF_SHARED,
@@ -1097,6 +1100,17 @@ static int __devinit igb_probe(struct pci_dev *pdev,
igb_get_bus_info_pcie(hw);
+ /* set flags */
+ switch (hw->mac.type) {
+ case e1000_82576:
+ case e1000_82575:
+ adapter->flags |= IGB_FLAG_HAS_DCA;
+ adapter->flags |= IGB_FLAG_NEED_CTX_IDX;
+ break;
+ default:
+ break;
+ }
+
hw->phy.autoneg_wait_to_complete = false;
hw->mac.adaptive_ifs = true;
@@ -1209,7 +1223,6 @@ static int __devinit igb_probe(struct pci_dev *pdev,
* lan on a particular port */
switch (pdev->device) {
case E1000_DEV_ID_82575GB_QUAD_COPPER:
- case E1000_DEV_ID_82576_QUAD_COPPER:
adapter->eeprom_wol = 0;
break;
case E1000_DEV_ID_82575EB_FIBER_SERDES:
@@ -1220,6 +1233,16 @@ static int __devinit igb_probe(struct pci_dev *pdev,
if (rd32(E1000_STATUS) & E1000_STATUS_FUNC_1)
adapter->eeprom_wol = 0;
break;
+ case E1000_DEV_ID_82576_QUAD_COPPER:
+ /* if quad port adapter, disable WoL on all but port A */
+ if (global_quad_port_a != 0)
+ adapter->eeprom_wol = 0;
+ else
+ adapter->flags |= IGB_FLAG_QUAD_PORT_A;
+ /* Reset for multiple quad port adapters */
+ if (++global_quad_port_a == 4)
+ global_quad_port_a = 0;
+ break;
}
/* initialize the wol settings based on the eeprom settings */
@@ -1246,8 +1269,9 @@ static int __devinit igb_probe(struct pci_dev *pdev,
goto err_register;
#ifdef CONFIG_DCA
- if (dca_add_requester(&pdev->dev) == 0) {
- adapter->dca_enabled = true;
+ if ((adapter->flags & IGB_FLAG_HAS_DCA) &&
+ (dca_add_requester(&pdev->dev) == 0)) {
+ adapter->flags |= IGB_FLAG_DCA_ENABLED;
dev_info(&pdev->dev, "DCA enabled\n");
/* Always use CB2 mode, difference is masked
* in the CB driver. */
@@ -1276,7 +1300,7 @@ static int __devinit igb_probe(struct pci_dev *pdev,
dev_info(&pdev->dev,
"Using %s interrupts. %d rx queue(s), %d tx queue(s)\n",
adapter->msix_entries ? "MSI-X" :
- adapter->msi_enabled ? "MSI" : "legacy",
+ (adapter->flags & IGB_FLAG_HAS_MSI) ? "MSI" : "legacy",
adapter->num_rx_queues, adapter->num_tx_queues);
return 0;
@@ -1330,10 +1354,10 @@ static void __devexit igb_remove(struct pci_dev *pdev)
flush_scheduled_work();
#ifdef CONFIG_DCA
- if (adapter->dca_enabled) {
+ if (adapter->flags & IGB_FLAG_DCA_ENABLED) {
dev_info(&pdev->dev, "DCA disabled\n");
dca_remove_requester(&pdev->dev);
- adapter->dca_enabled = false;
+ adapter->flags &= ~IGB_FLAG_DCA_ENABLED;
wr32(E1000_DCA_CTRL, 1);
}
#endif
@@ -2650,9 +2674,9 @@ static inline int igb_tso_adv(struct igb_adapter *adapter,
mss_l4len_idx = (skb_shinfo(skb)->gso_size << E1000_ADVTXD_MSS_SHIFT);
mss_l4len_idx |= (l4len << E1000_ADVTXD_L4LEN_SHIFT);
- /* Context index must be unique per ring. Luckily, so is the interrupt
- * mask value. */
- mss_l4len_idx |= tx_ring->eims_value >> 4;
+ /* Context index must be unique per ring. */
+ if (adapter->flags & IGB_FLAG_NEED_CTX_IDX)
+ mss_l4len_idx |= tx_ring->queue_index << 4;
context_desc->mss_l4len_idx = cpu_to_le32(mss_l4len_idx);
context_desc->seqnum_seed = 0;
@@ -2716,8 +2740,9 @@ static inline bool igb_tx_csum_adv(struct igb_adapter *adapter,
context_desc->type_tucmd_mlhl = cpu_to_le32(tu_cmd);
context_desc->seqnum_seed = 0;
- context_desc->mss_l4len_idx =
- cpu_to_le32(tx_ring->queue_index << 4);
+ if (adapter->flags & IGB_FLAG_NEED_CTX_IDX)
+ context_desc->mss_l4len_idx =
+ cpu_to_le32(tx_ring->queue_index << 4);
buffer_info->time_stamp = jiffies;
buffer_info->dma = 0;
@@ -2818,8 +2843,9 @@ static inline void igb_tx_queue_adv(struct igb_adapter *adapter,
olinfo_status |= E1000_TXD_POPTS_TXSM << 8;
}
- if (tx_flags & (IGB_TX_FLAGS_CSUM | IGB_TX_FLAGS_TSO |
- IGB_TX_FLAGS_VLAN))
+ if ((adapter->flags & IGB_FLAG_NEED_CTX_IDX) &&
+ (tx_flags & (IGB_TX_FLAGS_CSUM | IGB_TX_FLAGS_TSO |
+ IGB_TX_FLAGS_VLAN)))
olinfo_status |= tx_ring->queue_index << 4;
olinfo_status |= ((paylen - hdr_len) << E1000_ADVTXD_PAYLEN_SHIFT);
@@ -3255,7 +3281,7 @@ static irqreturn_t igb_msix_tx(int irq, void *data)
if (!tx_ring->itr_val)
wr32(E1000_EIMC, tx_ring->eims_value);
#ifdef CONFIG_DCA
- if (adapter->dca_enabled)
+ if (adapter->flags & IGB_FLAG_DCA_ENABLED)
igb_update_tx_dca(tx_ring);
#endif
tx_ring->total_bytes = 0;
@@ -3292,7 +3318,7 @@ static irqreturn_t igb_msix_rx(int irq, void *data)
__netif_rx_schedule(adapter->netdev, &rx_ring->napi);
#ifdef CONFIG_DCA
- if (adapter->dca_enabled)
+ if (adapter->flags & IGB_FLAG_DCA_ENABLED)
igb_update_rx_dca(rx_ring);
#endif
return IRQ_HANDLED;
@@ -3355,7 +3381,7 @@ static void igb_setup_dca(struct igb_adapter *adapter)
{
int i;
- if (!(adapter->dca_enabled))
+ if (!(adapter->flags & IGB_FLAG_DCA_ENABLED))
return;
for (i = 0; i < adapter->num_tx_queues; i++) {
@@ -3375,12 +3401,15 @@ static int __igb_notify_dca(struct device *dev, void *data)
struct e1000_hw *hw = &adapter->hw;
unsigned long event = *(unsigned long *)data;
+ if (!(adapter->flags & IGB_FLAG_HAS_DCA))
+ goto out;
+
switch (event) {
case DCA_PROVIDER_ADD:
/* if already enabled, don't do it again */
- if (adapter->dca_enabled)
+ if (adapter->flags & IGB_FLAG_DCA_ENABLED)
break;
- adapter->dca_enabled = true;
+ adapter->flags |= IGB_FLAG_DCA_ENABLED;
/* Always use CB2 mode, difference is masked
* in the CB driver. */
wr32(E1000_DCA_CTRL, 2);
@@ -3391,17 +3420,17 @@ static int __igb_notify_dca(struct device *dev, void *data)
}
/* Fall Through since DCA is disabled. */
case DCA_PROVIDER_REMOVE:
- if (adapter->dca_enabled) {
+ if (adapter->flags & IGB_FLAG_DCA_ENABLED) {
/* without this a class_device is left
* hanging around in the sysfs model */
dca_remove_requester(dev);
dev_info(&adapter->pdev->dev, "DCA disabled\n");
- adapter->dca_enabled = false;
+ adapter->flags &= ~IGB_FLAG_DCA_ENABLED;
wr32(E1000_DCA_CTRL, 1);
}
break;
}
-
+out:
return 0;
}
@@ -3507,13 +3536,13 @@ static int igb_poll(struct napi_struct *napi, int budget)
/* this poll routine only supports one tx and one rx queue */
#ifdef CONFIG_DCA
- if (adapter->dca_enabled)
+ if (adapter->flags & IGB_FLAG_DCA_ENABLED)
igb_update_tx_dca(&adapter->tx_ring[0]);
#endif
tx_clean_complete = igb_clean_tx_irq(&adapter->tx_ring[0]);
#ifdef CONFIG_DCA
- if (adapter->dca_enabled)
+ if (adapter->flags & IGB_FLAG_DCA_ENABLED)
igb_update_rx_dca(&adapter->rx_ring[0]);
#endif
igb_clean_rx_irq_adv(&adapter->rx_ring[0], &work_done, budget);
@@ -3545,7 +3574,7 @@ static int igb_clean_rx_ring_msix(struct napi_struct *napi, int budget)
goto quit_polling;
#ifdef CONFIG_DCA
- if (adapter->dca_enabled)
+ if (adapter->flags & IGB_FLAG_DCA_ENABLED)
igb_update_rx_dca(rx_ring);
#endif
igb_clean_rx_irq_adv(rx_ring, &work_done, budget);
@@ -4350,6 +4379,8 @@ static void igb_netpoll(struct net_device *netdev)
int work_done = 0;
igb_irq_disable(adapter);
+ adapter->flags |= IGB_FLAG_IN_NETPOLL;
+
for (i = 0; i < adapter->num_tx_queues; i++)
igb_clean_tx_irq(&adapter->tx_ring[i]);
@@ -4358,6 +4389,7 @@ static void igb_netpoll(struct net_device *netdev)
&work_done,
adapter->rx_ring[i].napi.weight);
+ adapter->flags &= ~IGB_FLAG_IN_NETPOLL;
igb_irq_enable(adapter);
}
#endif /* CONFIG_NET_POLL_CONTROLLER */
^ permalink raw reply related [flat|nested] 15+ messages in thread* [RESEND NET-NEXT PATCH 15/20] igb: add page recycling support
2008-07-08 22:06 [RESEND NET-NEXT PATCH 00/20] igb update for net-next Jeff Kirsher
` (6 preceding siblings ...)
2008-07-08 22:10 ` [RESEND NET-NEXT PATCH 14/20] igb: Add support for quad port WOL and feature flags Jeff Kirsher
@ 2008-07-08 22:11 ` Jeff Kirsher
2008-07-08 22:12 ` [RESEND NET-NEXT PATCH 16/20] igb: add support for in kernel LRO Jeff Kirsher
` (4 subsequent siblings)
12 siblings, 0 replies; 15+ messages in thread
From: Jeff Kirsher @ 2008-07-08 22:11 UTC (permalink / raw)
To: jeff, davem; +Cc: netdev, linux-kernel, akpm, Alexander Duyck, Jeff Kirsher
From: Alexander Duyck <alexander.h.duyck@intel.com>
This patch adds support for page recycling by splitting the page into two
usable portions and tracking the reference count.
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
---
drivers/net/igb/igb.h | 4 -
drivers/net/igb/igb_main.c | 138 ++++++++++++++++++++------------------------
2 files changed, 63 insertions(+), 79 deletions(-)
diff --git a/drivers/net/igb/igb.h b/drivers/net/igb/igb.h
index ee08010..f41b999 100644
--- a/drivers/net/igb/igb.h
+++ b/drivers/net/igb/igb.h
@@ -125,6 +125,7 @@ struct igb_buffer {
struct {
struct page *page;
u64 page_dma;
+ unsigned int page_offset;
};
};
};
@@ -163,9 +164,6 @@ struct igb_ring {
};
/* RX */
struct {
- /* arrays of page information for packet split */
- struct sk_buff *pending_skb;
- int pending_skb_page;
int no_itr_adjust;
struct igb_queue_stats rx_stats;
struct napi_struct napi;
diff --git a/drivers/net/igb/igb_main.c b/drivers/net/igb/igb_main.c
index 68a4fef..660a786 100644
--- a/drivers/net/igb/igb_main.c
+++ b/drivers/net/igb/igb_main.c
@@ -1725,7 +1725,6 @@ int igb_setup_rx_resources(struct igb_adapter *adapter,
rx_ring->next_to_clean = 0;
rx_ring->next_to_use = 0;
- rx_ring->pending_skb = NULL;
rx_ring->adapter = adapter;
@@ -1817,15 +1816,6 @@ static void igb_setup_rctl(struct igb_adapter *adapter)
rctl |= E1000_RCTL_SZ_2048;
rctl &= ~E1000_RCTL_BSEX;
break;
- case IGB_RXBUFFER_4096:
- rctl |= E1000_RCTL_SZ_4096;
- break;
- case IGB_RXBUFFER_8192:
- rctl |= E1000_RCTL_SZ_8192;
- break;
- case IGB_RXBUFFER_16384:
- rctl |= E1000_RCTL_SZ_16384;
- break;
}
} else {
rctl &= ~E1000_RCTL_BSEX;
@@ -1843,10 +1833,8 @@ static void igb_setup_rctl(struct igb_adapter *adapter)
* so only enable packet split for jumbo frames */
if (rctl & E1000_RCTL_LPE) {
adapter->rx_ps_hdr_size = IGB_RXBUFFER_128;
- srrctl = adapter->rx_ps_hdr_size <<
+ srrctl |= adapter->rx_ps_hdr_size <<
E1000_SRRCTL_BSIZEHDRSIZE_SHIFT;
- /* buffer size is ALWAYS one page */
- srrctl |= PAGE_SIZE >> E1000_SRRCTL_BSIZEPKT_SHIFT;
srrctl |= E1000_SRRCTL_DESCTYPE_HDR_SPLIT_ALWAYS;
} else {
adapter->rx_ps_hdr_size = 0;
@@ -2151,20 +2139,17 @@ static void igb_clean_rx_ring(struct igb_ring *rx_ring)
buffer_info->skb = NULL;
}
if (buffer_info->page) {
- pci_unmap_page(pdev, buffer_info->page_dma,
- PAGE_SIZE, PCI_DMA_FROMDEVICE);
+ if (buffer_info->page_dma)
+ pci_unmap_page(pdev, buffer_info->page_dma,
+ PAGE_SIZE / 2,
+ PCI_DMA_FROMDEVICE);
put_page(buffer_info->page);
buffer_info->page = NULL;
buffer_info->page_dma = 0;
+ buffer_info->page_offset = 0;
}
}
- /* there also may be some cached data from a chained receive */
- if (rx_ring->pending_skb) {
- dev_kfree_skb(rx_ring->pending_skb);
- rx_ring->pending_skb = NULL;
- }
-
size = sizeof(struct igb_buffer) * rx_ring->count;
memset(rx_ring->buffer_info, 0, size);
@@ -3091,7 +3076,11 @@ static int igb_change_mtu(struct net_device *netdev, int new_mtu)
else if (max_frame <= IGB_RXBUFFER_2048)
adapter->rx_buffer_len = IGB_RXBUFFER_2048;
else
- adapter->rx_buffer_len = IGB_RXBUFFER_4096;
+#if (PAGE_SIZE / 2) > IGB_RXBUFFER_16384
+ adapter->rx_buffer_len = IGB_RXBUFFER_16384;
+#else
+ adapter->rx_buffer_len = PAGE_SIZE / 2;
+#endif
/* adjust allocation if LPE protects us, and we aren't using SBP */
if ((max_frame == ETH_FRAME_LEN + ETH_FCS_LEN) ||
(max_frame == MAXIMUM_ETHERNET_VLAN_SIZE))
@@ -3796,7 +3785,7 @@ static bool igb_clean_rx_irq_adv(struct igb_ring *rx_ring,
union e1000_adv_rx_desc *rx_desc , *next_rxd;
struct igb_buffer *buffer_info , *next_buffer;
struct sk_buff *skb;
- unsigned int i, j;
+ unsigned int i;
u32 length, hlen, staterr;
bool cleaned = false;
int cleaned_count = 0;
@@ -3826,61 +3815,46 @@ static bool igb_clean_rx_irq_adv(struct igb_ring *rx_ring,
cleaned = true;
cleaned_count++;
- if (rx_ring->pending_skb != NULL) {
- skb = rx_ring->pending_skb;
- rx_ring->pending_skb = NULL;
- j = rx_ring->pending_skb_page;
- } else {
- skb = buffer_info->skb;
- prefetch(skb->data - NET_IP_ALIGN);
- buffer_info->skb = NULL;
- if (hlen) {
- pci_unmap_single(pdev, buffer_info->dma,
- adapter->rx_ps_hdr_size +
- NET_IP_ALIGN,
- PCI_DMA_FROMDEVICE);
- skb_put(skb, hlen);
- } else {
- pci_unmap_single(pdev, buffer_info->dma,
- adapter->rx_buffer_len +
- NET_IP_ALIGN,
- PCI_DMA_FROMDEVICE);
- skb_put(skb, length);
- goto send_up;
- }
- j = 0;
+ skb = buffer_info->skb;
+ prefetch(skb->data - NET_IP_ALIGN);
+ buffer_info->skb = NULL;
+ if (!adapter->rx_ps_hdr_size) {
+ pci_unmap_single(pdev, buffer_info->dma,
+ adapter->rx_buffer_len +
+ NET_IP_ALIGN,
+ PCI_DMA_FROMDEVICE);
+ skb_put(skb, length);
+ goto send_up;
+ }
+
+ if (!skb_shinfo(skb)->nr_frags) {
+ pci_unmap_single(pdev, buffer_info->dma,
+ adapter->rx_ps_hdr_size +
+ NET_IP_ALIGN,
+ PCI_DMA_FROMDEVICE);
+ skb_put(skb, hlen);
}
- while (length) {
+ if (length) {
pci_unmap_page(pdev, buffer_info->page_dma,
- PAGE_SIZE, PCI_DMA_FROMDEVICE);
+ PAGE_SIZE / 2, PCI_DMA_FROMDEVICE);
buffer_info->page_dma = 0;
- skb_fill_page_desc(skb, j, buffer_info->page,
- 0, length);
- buffer_info->page = NULL;
+
+ skb_fill_page_desc(skb, skb_shinfo(skb)->nr_frags++,
+ buffer_info->page,
+ buffer_info->page_offset,
+ length);
+
+ if ((adapter->rx_buffer_len > (PAGE_SIZE / 2)) ||
+ (page_count(buffer_info->page) != 1))
+ buffer_info->page = NULL;
+ else
+ get_page(buffer_info->page);
skb->len += length;
skb->data_len += length;
- skb->truesize += length;
- rx_desc->wb.upper.status_error = 0;
- if (staterr & E1000_RXD_STAT_EOP)
- break;
-
- j++;
- cleaned_count++;
- i++;
- if (i == rx_ring->count)
- i = 0;
- buffer_info = &rx_ring->buffer_info[i];
- rx_desc = E1000_RX_DESC_ADV(*rx_ring, i);
- staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
- length = le16_to_cpu(rx_desc->wb.upper.length);
- if (!(staterr & E1000_RXD_STAT_DD)) {
- rx_ring->pending_skb = skb;
- rx_ring->pending_skb_page = j;
- goto out;
- }
+ skb->truesize += length;
}
send_up:
i++;
@@ -3890,6 +3864,12 @@ send_up:
prefetch(next_rxd);
next_buffer = &rx_ring->buffer_info[i];
+ if (!(staterr & E1000_RXD_STAT_EOP)) {
+ buffer_info->skb = xchg(&next_buffer->skb, skb);
+ buffer_info->dma = xchg(&next_buffer->dma, 0);
+ goto next_desc;
+ }
+
if (staterr & E1000_RXDEXT_ERR_FRAME_ERR_MASK) {
dev_kfree_skb_irq(skb);
goto next_desc;
@@ -3922,7 +3902,7 @@ next_desc:
staterr = le32_to_cpu(rx_desc->wb.upper.status_error);
}
-out:
+
rx_ring->next_to_clean = i;
cleaned_count = IGB_DESC_UNUSED(rx_ring);
@@ -3960,16 +3940,22 @@ static void igb_alloc_rx_buffers_adv(struct igb_ring *rx_ring,
while (cleaned_count--) {
rx_desc = E1000_RX_DESC_ADV(*rx_ring, i);
- if (adapter->rx_ps_hdr_size && !buffer_info->page) {
- buffer_info->page = alloc_page(GFP_ATOMIC);
+ if (adapter->rx_ps_hdr_size && !buffer_info->page_dma) {
if (!buffer_info->page) {
- adapter->alloc_rx_buff_failed++;
- goto no_buffers;
+ buffer_info->page = alloc_page(GFP_ATOMIC);
+ if (!buffer_info->page) {
+ adapter->alloc_rx_buff_failed++;
+ goto no_buffers;
+ }
+ buffer_info->page_offset = 0;
+ } else {
+ buffer_info->page_offset ^= PAGE_SIZE / 2;
}
buffer_info->page_dma =
pci_map_page(pdev,
buffer_info->page,
- 0, PAGE_SIZE,
+ buffer_info->page_offset,
+ PAGE_SIZE / 2,
PCI_DMA_FROMDEVICE);
}
^ permalink raw reply related [flat|nested] 15+ messages in thread* [RESEND NET-NEXT PATCH 16/20] igb: add support for in kernel LRO
2008-07-08 22:06 [RESEND NET-NEXT PATCH 00/20] igb update for net-next Jeff Kirsher
` (7 preceding siblings ...)
2008-07-08 22:11 ` [RESEND NET-NEXT PATCH 15/20] igb: add page recycling support Jeff Kirsher
@ 2008-07-08 22:12 ` Jeff Kirsher
2008-07-08 22:13 ` [RESEND NET-NEXT PATCH 17/20] net: add netif_napi_del function to allow for removal of napistructs Jeff Kirsher
` (3 subsequent siblings)
12 siblings, 0 replies; 15+ messages in thread
From: Jeff Kirsher @ 2008-07-08 22:12 UTC (permalink / raw)
To: jeff, davem; +Cc: netdev, linux-kernel, akpm, Alexander Duyck, Jeff Kirsher
From: Alexander Duyck <alexander.h.duyck@intel.com>
This patch adds support for the use of the inet_lro module to provide
software LRO support.
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
---
drivers/net/Kconfig | 9 +++
drivers/net/igb/e1000_82575.h | 2 +
drivers/net/igb/igb.h | 16 ++++++
drivers/net/igb/igb_ethtool.c | 17 ++++++
drivers/net/igb/igb_main.c | 112 ++++++++++++++++++++++++++++++++++++++---
5 files changed, 147 insertions(+), 9 deletions(-)
diff --git a/drivers/net/Kconfig b/drivers/net/Kconfig
index ef733ab..5252514 100644
--- a/drivers/net/Kconfig
+++ b/drivers/net/Kconfig
@@ -2037,6 +2037,15 @@ config IGB
To compile this driver as a module, choose M here. The module
will be called igb.
+config IGB_LRO
+ bool "Use software LRO"
+ depends on IGB && INET
+ select INET_LRO
+ ---help---
+ Say Y here if you want to use large receive offload.
+
+ If in doubt, say N.
+
source "drivers/net/ixp2000/Kconfig"
config MYRI_SBUS
diff --git a/drivers/net/igb/e1000_82575.h b/drivers/net/igb/e1000_82575.h
index d273236..2f848e5 100644
--- a/drivers/net/igb/e1000_82575.h
+++ b/drivers/net/igb/e1000_82575.h
@@ -99,6 +99,8 @@ union e1000_adv_rx_desc {
/* RSS Hash results */
/* RSS Packet Types as indicated in the receive descriptor */
+#define E1000_RXDADV_PKTTYPE_IPV4 0x00000010 /* IPV4 hdr present */
+#define E1000_RXDADV_PKTTYPE_TCP 0x00000100 /* TCP hdr present */
/* Transmit Descriptor - Advanced */
union e1000_adv_tx_desc {
diff --git a/drivers/net/igb/igb.h b/drivers/net/igb/igb.h
index f41b999..c25ca17 100644
--- a/drivers/net/igb/igb.h
+++ b/drivers/net/igb/igb.h
@@ -36,6 +36,12 @@
struct igb_adapter;
+#ifdef CONFIG_IGB_LRO
+#include <linux/inet_lro.h>
+#define MAX_LRO_AGGR 32
+#define MAX_LRO_DESCRIPTORS 8
+#endif
+
/* Interrupt defines */
#define IGB_MAX_TX_CLEAN 72
@@ -167,6 +173,10 @@ struct igb_ring {
int no_itr_adjust;
struct igb_queue_stats rx_stats;
struct napi_struct napi;
+#ifdef CONFIG_IGB_LRO
+ struct net_lro_mgr lro_mgr;
+ bool lro_used;
+#endif
};
};
@@ -278,6 +288,12 @@ struct igb_adapter {
#ifdef CONFIG_NETDEVICES_MULTIQUEUE
struct igb_ring *multi_tx_table[IGB_MAX_TX_QUEUES];
#endif /* CONFIG_NETDEVICES_MULTIQUEUE */
+#ifdef CONFIG_IGB_LRO
+ unsigned int lro_max_aggr;
+ unsigned int lro_aggregated;
+ unsigned int lro_flushed;
+ unsigned int lro_no_desc;
+#endif
};
#define IGB_FLAG_HAS_MSI (1 << 0)
diff --git a/drivers/net/igb/igb_ethtool.c b/drivers/net/igb/igb_ethtool.c
index ef209b5..7db1830 100644
--- a/drivers/net/igb/igb_ethtool.c
+++ b/drivers/net/igb/igb_ethtool.c
@@ -93,6 +93,11 @@ static const struct igb_stats igb_gstrings_stats[] = {
{ "tx_smbus", IGB_STAT(stats.mgptc) },
{ "rx_smbus", IGB_STAT(stats.mgprc) },
{ "dropped_smbus", IGB_STAT(stats.mgpdc) },
+#ifdef CONFIG_IGB_LRO
+ { "lro_aggregated", IGB_STAT(lro_aggregated) },
+ { "lro_flushed", IGB_STAT(lro_flushed) },
+ { "lro_no_desc", IGB_STAT(lro_no_desc) },
+#endif
};
#define IGB_QUEUE_STATS_LEN \
@@ -1917,6 +1922,18 @@ static void igb_get_ethtool_stats(struct net_device *netdev,
int stat_count = sizeof(struct igb_queue_stats) / sizeof(u64);
int j;
int i;
+#ifdef CONFIG_IGB_LRO
+ int aggregated = 0, flushed = 0, no_desc = 0;
+
+ for (i = 0; i < adapter->num_rx_queues; i++) {
+ aggregated += adapter->rx_ring[i].lro_mgr.stats.aggregated;
+ flushed += adapter->rx_ring[i].lro_mgr.stats.flushed;
+ no_desc += adapter->rx_ring[i].lro_mgr.stats.no_desc;
+ }
+ adapter->lro_aggregated = aggregated;
+ adapter->lro_flushed = flushed;
+ adapter->lro_no_desc = no_desc;
+#endif
igb_update_stats(adapter);
for (i = 0; i < IGB_GLOBAL_STATS_LEN; i++) {
diff --git a/drivers/net/igb/igb_main.c b/drivers/net/igb/igb_main.c
index 660a786..89416eb 100644
--- a/drivers/net/igb/igb_main.c
+++ b/drivers/net/igb/igb_main.c
@@ -116,6 +116,9 @@ static bool igb_clean_tx_irq(struct igb_ring *);
static int igb_poll(struct napi_struct *, int);
static bool igb_clean_rx_irq_adv(struct igb_ring *, int *, int);
static void igb_alloc_rx_buffers_adv(struct igb_ring *, int);
+#ifdef CONFIG_IGB_LRO
+static int igb_get_skb_hdr(struct sk_buff *skb, void **, void **, u64 *, void *);
+#endif
static int igb_ioctl(struct net_device *, struct ifreq *, int cmd);
static void igb_tx_timeout(struct net_device *);
static void igb_reset_task(struct work_struct *);
@@ -1134,6 +1137,10 @@ static int __devinit igb_probe(struct pci_dev *pdev,
netdev->features |= NETIF_F_TSO;
netdev->features |= NETIF_F_TSO6;
+#ifdef CONFIG_IGB_LRO
+ netdev->features |= NETIF_F_LRO;
+#endif
+
netdev->vlan_features |= NETIF_F_TSO;
netdev->vlan_features |= NETIF_F_TSO6;
netdev->vlan_features |= NETIF_F_HW_CSUM;
@@ -1705,6 +1712,14 @@ int igb_setup_rx_resources(struct igb_adapter *adapter,
struct pci_dev *pdev = adapter->pdev;
int size, desc_len;
+#ifdef CONFIG_IGB_LRO
+ size = sizeof(struct net_lro_desc) * MAX_LRO_DESCRIPTORS;
+ rx_ring->lro_mgr.lro_arr = vmalloc(size);
+ if (!rx_ring->lro_mgr.lro_arr)
+ goto err;
+ memset(rx_ring->lro_mgr.lro_arr, 0, size);
+#endif
+
size = sizeof(struct igb_buffer) * rx_ring->count;
rx_ring->buffer_info = vmalloc(size);
if (!rx_ring->buffer_info)
@@ -1731,6 +1746,10 @@ int igb_setup_rx_resources(struct igb_adapter *adapter,
return 0;
err:
+#ifdef CONFIG_IGB_LRO
+ vfree(rx_ring->lro_mgr.lro_arr);
+ rx_ring->lro_mgr.lro_arr = NULL;
+#endif
vfree(rx_ring->buffer_info);
dev_err(&adapter->pdev->dev, "Unable to allocate memory for "
"the receive descriptor ring\n");
@@ -1894,6 +1913,16 @@ static void igb_configure_rx(struct igb_adapter *adapter)
rxdctl |= IGB_RX_HTHRESH << 8;
rxdctl |= IGB_RX_WTHRESH << 16;
wr32(E1000_RXDCTL(i), rxdctl);
+#ifdef CONFIG_IGB_LRO
+ /* Intitial LRO Settings */
+ ring->lro_mgr.max_aggr = MAX_LRO_AGGR;
+ ring->lro_mgr.max_desc = MAX_LRO_DESCRIPTORS;
+ ring->lro_mgr.get_skb_header = igb_get_skb_hdr;
+ ring->lro_mgr.features = LRO_F_NAPI | LRO_F_EXTRACT_VLAN_ID;
+ ring->lro_mgr.dev = adapter->netdev;
+ ring->lro_mgr.ip_summed = CHECKSUM_UNNECESSARY;
+ ring->lro_mgr.ip_summed_aggr = CHECKSUM_UNNECESSARY;
+#endif
}
if (adapter->num_rx_queues > 1) {
@@ -2085,6 +2114,11 @@ static void igb_free_rx_resources(struct igb_ring *rx_ring)
vfree(rx_ring->buffer_info);
rx_ring->buffer_info = NULL;
+#ifdef CONFIG_IGB_LRO
+ vfree(rx_ring->lro_mgr.lro_arr);
+ rx_ring->lro_mgr.lro_arr = NULL;
+#endif
+
pci_free_consistent(pdev, rx_ring->size, rx_ring->desc, rx_ring->dma);
rx_ring->desc = NULL;
@@ -3735,22 +3769,75 @@ done_cleaning:
return retval;
}
+#ifdef CONFIG_IGB_LRO
+ /**
+ * igb_get_skb_hdr - helper function for LRO header processing
+ * @skb: pointer to sk_buff to be added to LRO packet
+ * @iphdr: pointer to ip header structure
+ * @tcph: pointer to tcp header structure
+ * @hdr_flags: pointer to header flags
+ * @priv: pointer to the receive descriptor for the current sk_buff
+ **/
+static int igb_get_skb_hdr(struct sk_buff *skb, void **iphdr, void **tcph,
+ u64 *hdr_flags, void *priv)
+{
+ union e1000_adv_rx_desc *rx_desc = priv;
+ u16 pkt_type = rx_desc->wb.lower.lo_dword.pkt_info &
+ (E1000_RXDADV_PKTTYPE_IPV4 | E1000_RXDADV_PKTTYPE_TCP);
+
+ /* Verify that this is a valid IPv4 TCP packet */
+ if (pkt_type != (E1000_RXDADV_PKTTYPE_IPV4 |
+ E1000_RXDADV_PKTTYPE_TCP))
+ return -1;
+
+ /* Set network headers */
+ skb_reset_network_header(skb);
+ skb_set_transport_header(skb, ip_hdrlen(skb));
+ *iphdr = ip_hdr(skb);
+ *tcph = tcp_hdr(skb);
+ *hdr_flags = LRO_IPV4 | LRO_TCP;
+
+ return 0;
+
+}
+#endif /* CONFIG_IGB_LRO */
/**
* igb_receive_skb - helper function to handle rx indications
- * @adapter: board private structure
+ * @ring: pointer to receive ring receving this packet
* @status: descriptor status field as written by hardware
* @vlan: descriptor vlan field as written by hardware (no le/be conversion)
* @skb: pointer to sk_buff to be indicated to stack
**/
-static void igb_receive_skb(struct igb_adapter *adapter, u8 status, __le16 vlan,
- struct sk_buff *skb)
+static void igb_receive_skb(struct igb_ring *ring, u8 status,
+ union e1000_adv_rx_desc * rx_desc,
+ struct sk_buff *skb)
{
- if (adapter->vlgrp && (status & E1000_RXD_STAT_VP))
- vlan_hwaccel_receive_skb(skb, adapter->vlgrp,
- le16_to_cpu(vlan));
- else
- netif_receive_skb(skb);
+ struct igb_adapter * adapter = ring->adapter;
+ bool vlan_extracted = (adapter->vlgrp && (status & E1000_RXD_STAT_VP));
+
+#ifdef CONFIG_IGB_LRO
+ if (adapter->netdev->features & NETIF_F_LRO &&
+ skb->ip_summed == CHECKSUM_UNNECESSARY) {
+ if (vlan_extracted)
+ lro_vlan_hwaccel_receive_skb(&ring->lro_mgr, skb,
+ adapter->vlgrp,
+ le16_to_cpu(rx_desc->wb.upper.vlan),
+ rx_desc);
+ else
+ lro_receive_skb(&ring->lro_mgr,skb, rx_desc);
+ ring->lro_used = 1;
+ } else {
+#endif
+ if (vlan_extracted)
+ vlan_hwaccel_receive_skb(skb, adapter->vlgrp,
+ le16_to_cpu(rx_desc->wb.upper.vlan));
+ else
+
+ netif_receive_skb(skb);
+#ifdef CONFIG_IGB_LRO
+ }
+#endif
}
@@ -3883,7 +3970,7 @@ send_up:
skb->protocol = eth_type_trans(skb, netdev);
- igb_receive_skb(adapter, staterr, rx_desc->wb.upper.vlan, skb);
+ igb_receive_skb(rx_ring, staterr, rx_desc, skb);
netdev->last_rx = jiffies;
@@ -3906,6 +3993,13 @@ next_desc:
rx_ring->next_to_clean = i;
cleaned_count = IGB_DESC_UNUSED(rx_ring);
+#ifdef CONFIG_IGB_LRO
+ if (rx_ring->lro_used) {
+ lro_flush_all(&rx_ring->lro_mgr);
+ rx_ring->lro_used = 0;
+ }
+#endif
+
if (cleaned_count)
igb_alloc_rx_buffers_adv(rx_ring, cleaned_count);
^ permalink raw reply related [flat|nested] 15+ messages in thread* [RESEND NET-NEXT PATCH 17/20] net: add netif_napi_del function to allow for removal of napistructs
2008-07-08 22:06 [RESEND NET-NEXT PATCH 00/20] igb update for net-next Jeff Kirsher
` (8 preceding siblings ...)
2008-07-08 22:12 ` [RESEND NET-NEXT PATCH 16/20] igb: add support for in kernel LRO Jeff Kirsher
@ 2008-07-08 22:13 ` Jeff Kirsher
2008-07-08 22:13 ` [RESEND NET-NEXT PATCH 18/20] igb: update suspend resume Jeff Kirsher
` (2 subsequent siblings)
12 siblings, 0 replies; 15+ messages in thread
From: Jeff Kirsher @ 2008-07-08 22:13 UTC (permalink / raw)
To: jeff, davem; +Cc: netdev, linux-kernel, akpm, Alexander Duyck, Jeff Kirsher
From: Alexander Duyck <alexander.h.duyck@intel.com>
Adds netif_napi_del function which is used to remove the napi struct from
the netdev napi_list in cases where CONFIG_NETPOLL was enabled.
The motivation for adding this is to handle the case in which the number of
queues on a device changes due to a configuration change. Previously the
napi structs for each queue would be left in the list until the netdev was
freed.
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
---
include/linux/netdevice.h | 13 +++++++++++++
1 files changed, 13 insertions(+), 0 deletions(-)
diff --git a/include/linux/netdevice.h b/include/linux/netdevice.h
index e009c6f..60cf72e 100644
--- a/include/linux/netdevice.h
+++ b/include/linux/netdevice.h
@@ -832,6 +832,19 @@ static inline void netif_napi_add(struct net_device *dev,
set_bit(NAPI_STATE_SCHED, &napi->state);
}
+/**
+ * netif_napi_del - remove a napi context
+ * @napi: napi context
+ *
+ * netif_napi_del() removes a napi context from the network device napi list
+ */
+static inline void netif_napi_del(struct napi_struct *napi)
+{
+#ifdef CONFIG_NETPOLL
+ list_del(&napi->dev_list);
+#endif
+}
+
struct packet_type {
__be16 type; /* This is really htons(ether_type). */
struct net_device *dev; /* NULL is wildcarded here */
^ permalink raw reply related [flat|nested] 15+ messages in thread* [RESEND NET-NEXT PATCH 18/20] igb: update suspend resume
2008-07-08 22:06 [RESEND NET-NEXT PATCH 00/20] igb update for net-next Jeff Kirsher
` (9 preceding siblings ...)
2008-07-08 22:13 ` [RESEND NET-NEXT PATCH 17/20] net: add netif_napi_del function to allow for removal of napistructs Jeff Kirsher
@ 2008-07-08 22:13 ` Jeff Kirsher
2008-07-08 22:14 ` [NET-NEXT PATCH 19/20] [cosmetic] igb: unused variable warning in igb remove Jeff Kirsher
2008-07-08 22:14 ` [NET-NEXT PATCH 20/20] igb: Improve multiqueue AIM support Jeff Kirsher
12 siblings, 0 replies; 15+ messages in thread
From: Jeff Kirsher @ 2008-07-08 22:13 UTC (permalink / raw)
To: jeff, davem; +Cc: netdev, linux-kernel, akpm, Alexander Duyck, Jeff Kirsher
From: Alexander Duyck <alexander.h.duyck@intel.com>
Updates the suspend and resume to better handle the possibility of MSIX
vector changes.
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
---
drivers/net/igb/igb_main.c | 46 ++++++++++++++++++++++++++++----------------
1 files changed, 29 insertions(+), 17 deletions(-)
diff --git a/drivers/net/igb/igb_main.c b/drivers/net/igb/igb_main.c
index 89416eb..00e2e8d 100644
--- a/drivers/net/igb/igb_main.c
+++ b/drivers/net/igb/igb_main.c
@@ -272,6 +272,17 @@ static int igb_alloc_queues(struct igb_adapter *adapter)
return 0;
}
+static void igb_free_queues(struct igb_adapter *adapter)
+{
+ int i;
+
+ for (i = 0; i < adapter->num_rx_queues; i++)
+ netif_napi_del(&adapter->rx_ring[i].napi);
+
+ kfree(adapter->tx_ring);
+ kfree(adapter->rx_ring);
+}
+
#define IGB_N0_QUEUE -1
static void igb_assign_vector(struct igb_adapter *adapter, int rx_queue,
int tx_queue, int msix_vector)
@@ -1322,8 +1333,7 @@ err_eeprom:
iounmap(hw->flash_address);
igb_remove_device(hw);
- kfree(adapter->tx_ring);
- kfree(adapter->rx_ring);
+ igb_free_queues(adapter);
err_sw_init:
err_hw_init:
iounmap(hw->hw_addr);
@@ -1381,8 +1391,7 @@ static void __devexit igb_remove(struct pci_dev *pdev)
igb_remove_device(&adapter->hw);
igb_reset_interrupt_capability(adapter);
- kfree(adapter->tx_ring);
- kfree(adapter->rx_ring);
+ igb_free_queues(adapter);
iounmap(adapter->hw.hw_addr);
if (adapter->hw.flash_address)
@@ -4324,11 +4333,12 @@ static int igb_suspend(struct pci_dev *pdev, pm_message_t state)
netif_device_detach(netdev);
- if (netif_running(netdev)) {
- WARN_ON(test_bit(__IGB_RESETTING, &adapter->state));
- igb_down(adapter);
- igb_free_irq(adapter);
- }
+ if (netif_running(netdev))
+ igb_close(netdev);
+
+ igb_reset_interrupt_capability(adapter);
+
+ igb_free_queues(adapter);
#ifdef CONFIG_PM
retval = pci_save_state(pdev);
@@ -4415,10 +4425,11 @@ static int igb_resume(struct pci_dev *pdev)
pci_enable_wake(pdev, PCI_D3hot, 0);
pci_enable_wake(pdev, PCI_D3cold, 0);
- if (netif_running(netdev)) {
- err = igb_request_irq(adapter);
- if (err)
- return err;
+ igb_set_interrupt_capability(adapter);
+
+ if (igb_alloc_queues(adapter)) {
+ dev_err(&pdev->dev, "Unable to allocate memory for queues\n");
+ return -ENOMEM;
}
/* e1000_power_up_phy(adapter); */
@@ -4426,10 +4437,11 @@ static int igb_resume(struct pci_dev *pdev)
igb_reset(adapter);
wr32(E1000_WUS, ~0);
- igb_init_manageability(adapter);
-
- if (netif_running(netdev))
- igb_up(adapter);
+ if (netif_running(netdev)) {
+ err = igb_open(netdev);
+ if (err)
+ return err;
+ }
netif_device_attach(netdev);
^ permalink raw reply related [flat|nested] 15+ messages in thread* [NET-NEXT PATCH 19/20] [cosmetic] igb: unused variable warning in igb remove
2008-07-08 22:06 [RESEND NET-NEXT PATCH 00/20] igb update for net-next Jeff Kirsher
` (10 preceding siblings ...)
2008-07-08 22:13 ` [RESEND NET-NEXT PATCH 18/20] igb: update suspend resume Jeff Kirsher
@ 2008-07-08 22:14 ` Jeff Kirsher
2008-07-08 22:14 ` [NET-NEXT PATCH 20/20] igb: Improve multiqueue AIM support Jeff Kirsher
12 siblings, 0 replies; 15+ messages in thread
From: Jeff Kirsher @ 2008-07-08 22:14 UTC (permalink / raw)
To: jeff, davem; +Cc: netdev, linux-kernel, akpm, Alexander Duyck, Jeff Kirsher
From: Alexander Duyck <alexander.h.duyck@intel.com>
Wrap hw variable declaration in DCA flags to prevent unused variable
warning during compilation.
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
---
drivers/net/igb/igb_main.c | 2 ++
1 files changed, 2 insertions(+), 0 deletions(-)
diff --git a/drivers/net/igb/igb_main.c b/drivers/net/igb/igb_main.c
index 00e2e8d..2a5303c 100644
--- a/drivers/net/igb/igb_main.c
+++ b/drivers/net/igb/igb_main.c
@@ -1360,7 +1360,9 @@ static void __devexit igb_remove(struct pci_dev *pdev)
{
struct net_device *netdev = pci_get_drvdata(pdev);
struct igb_adapter *adapter = netdev_priv(netdev);
+#ifdef CONFIG_DCA
struct e1000_hw *hw = &adapter->hw;
+#endif
/* flush_scheduled work may reschedule our watchdog task, so
* explicitly disable watchdog tasks from being rescheduled */
^ permalink raw reply related [flat|nested] 15+ messages in thread* [NET-NEXT PATCH 20/20] igb: Improve multiqueue AIM support
2008-07-08 22:06 [RESEND NET-NEXT PATCH 00/20] igb update for net-next Jeff Kirsher
` (11 preceding siblings ...)
2008-07-08 22:14 ` [NET-NEXT PATCH 19/20] [cosmetic] igb: unused variable warning in igb remove Jeff Kirsher
@ 2008-07-08 22:14 ` Jeff Kirsher
12 siblings, 0 replies; 15+ messages in thread
From: Jeff Kirsher @ 2008-07-08 22:14 UTC (permalink / raw)
To: jeff, davem
Cc: netdev, linux-kernel, akpm, Alexander Duyck, Mitch Williams,
Jeff Kirsher
From: Alexander Duyck <alexander.h.duyck@intel.com>
Improve multiqueue performance
Change itr_val to reflect ITR timer value instead of ints/sec
Cleaned up AIM algorithms in general
Based on work by Mitch Williams
Signed-off-by: Alexander Duyck <alexander.h.duyck@intel.com>
Acked-by: Mitch Williams <mitch.a.williams@intel.com>
Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
---
drivers/net/igb/igb.h | 8 +-
drivers/net/igb/igb_ethtool.c | 17 +++-
drivers/net/igb/igb_main.c | 169 ++++++++++++++++++++++++-----------------
3 files changed, 113 insertions(+), 81 deletions(-)
diff --git a/drivers/net/igb/igb.h b/drivers/net/igb/igb.h
index c25ca17..56de7ec 100644
--- a/drivers/net/igb/igb.h
+++ b/drivers/net/igb/igb.h
@@ -47,7 +47,9 @@ struct igb_adapter;
#define IGB_MIN_DYN_ITR 3000
#define IGB_MAX_DYN_ITR 96000
-#define IGB_START_ITR 6000
+
+/* ((1000000000ns / (6000ints/s * 1024ns)) << 2 = 648 */
+#define IGB_START_ITR 648
#define IGB_DYN_ITR_PACKET_THRESHOLD 2
#define IGB_DYN_ITR_LENGTH_LOW 200
@@ -170,9 +172,10 @@ struct igb_ring {
};
/* RX */
struct {
- int no_itr_adjust;
struct igb_queue_stats rx_stats;
struct napi_struct napi;
+ int set_itr;
+ struct igb_ring *buddy;
#ifdef CONFIG_IGB_LRO
struct net_lro_mgr lro_mgr;
bool lro_used;
@@ -219,7 +222,6 @@ struct igb_adapter {
u32 itr_setting;
u16 tx_itr;
u16 rx_itr;
- int set_itr;
struct work_struct reset_task;
struct work_struct watchdog_task;
diff --git a/drivers/net/igb/igb_ethtool.c b/drivers/net/igb/igb_ethtool.c
index 7db1830..11aee13 100644
--- a/drivers/net/igb/igb_ethtool.c
+++ b/drivers/net/igb/igb_ethtool.c
@@ -1861,6 +1861,8 @@ static int igb_set_coalesce(struct net_device *netdev,
struct ethtool_coalesce *ec)
{
struct igb_adapter *adapter = netdev_priv(netdev);
+ struct e1000_hw *hw = &adapter->hw;
+ int i;
if ((ec->rx_coalesce_usecs > IGB_MAX_ITR_USECS) ||
((ec->rx_coalesce_usecs > 3) &&
@@ -1869,13 +1871,16 @@ static int igb_set_coalesce(struct net_device *netdev,
return -EINVAL;
/* convert to rate of irq's per second */
- if (ec->rx_coalesce_usecs <= 3)
+ if (ec->rx_coalesce_usecs && ec->rx_coalesce_usecs <= 3) {
adapter->itr_setting = ec->rx_coalesce_usecs;
- else
- adapter->itr_setting = (1000000 / ec->rx_coalesce_usecs);
+ adapter->itr = IGB_START_ITR;
+ } else {
+ adapter->itr_setting = ec->rx_coalesce_usecs << 2;
+ adapter->itr = adapter->itr_setting;
+ }
- if (netif_running(netdev))
- igb_reinit_locked(adapter);
+ for (i = 0; i < adapter->num_rx_queues; i++)
+ wr32(adapter->rx_ring[i].itr_register, adapter->itr);
return 0;
}
@@ -1888,7 +1893,7 @@ static int igb_get_coalesce(struct net_device *netdev,
if (adapter->itr_setting <= 3)
ec->rx_coalesce_usecs = adapter->itr_setting;
else
- ec->rx_coalesce_usecs = 1000000 / adapter->itr_setting;
+ ec->rx_coalesce_usecs = adapter->itr_setting >> 2;
return 0;
}
diff --git a/drivers/net/igb/igb_main.c b/drivers/net/igb/igb_main.c
index 2a5303c..aaed129 100644
--- a/drivers/net/igb/igb_main.c
+++ b/drivers/net/igb/igb_main.c
@@ -255,6 +255,8 @@ static int igb_alloc_queues(struct igb_adapter *adapter)
return -ENOMEM;
}
+ adapter->rx_ring->buddy = adapter->tx_ring;
+
for (i = 0; i < adapter->num_tx_queues; i++) {
struct igb_ring *ring = &(adapter->tx_ring[i]);
ring->adapter = adapter;
@@ -375,7 +377,7 @@ static void igb_configure_msix(struct igb_adapter *adapter)
igb_assign_vector(adapter, IGB_N0_QUEUE, i, vector++);
adapter->eims_enable_mask |= tx_ring->eims_value;
if (tx_ring->itr_val)
- writel(1000000000 / (tx_ring->itr_val * 256),
+ writel(tx_ring->itr_val,
hw->hw_addr + tx_ring->itr_register);
else
writel(1, hw->hw_addr + tx_ring->itr_register);
@@ -383,10 +385,11 @@ static void igb_configure_msix(struct igb_adapter *adapter)
for (i = 0; i < adapter->num_rx_queues; i++) {
struct igb_ring *rx_ring = &adapter->rx_ring[i];
+ rx_ring->buddy = 0;
igb_assign_vector(adapter, i, IGB_N0_QUEUE, vector++);
adapter->eims_enable_mask |= rx_ring->eims_value;
if (rx_ring->itr_val)
- writel(1000000000 / (rx_ring->itr_val * 256),
+ writel(rx_ring->itr_val,
hw->hw_addr + rx_ring->itr_register);
else
writel(1, hw->hw_addr + rx_ring->itr_register);
@@ -449,7 +452,7 @@ static int igb_request_msix(struct igb_adapter *adapter)
if (err)
goto out;
ring->itr_register = E1000_EITR(0) + (vector << 2);
- ring->itr_val = adapter->itr;
+ ring->itr_val = 976; /* ~4000 ints/sec */
vector++;
}
for (i = 0; i < adapter->num_rx_queues; i++) {
@@ -1898,8 +1901,7 @@ static void igb_configure_rx(struct igb_adapter *adapter)
mdelay(10);
if (adapter->itr_setting > 3)
- wr32(E1000_ITR,
- 1000000000 / (adapter->itr * 256));
+ wr32(E1000_ITR, adapter->itr);
/* Setup the HW Rx Head and Tail Descriptor Pointers and
* the Base and Length of the Rx Descriptor Ring */
@@ -2463,38 +2465,60 @@ enum latency_range {
};
-static void igb_lower_rx_eitr(struct igb_adapter *adapter,
- struct igb_ring *rx_ring)
+/**
+ * igb_update_ring_itr - update the dynamic ITR value based on packet size
+ *
+ * Stores a new ITR value based on strictly on packet size. This
+ * algorithm is less sophisticated than that used in igb_update_itr,
+ * due to the difficulty of synchronizing statistics across multiple
+ * receive rings. The divisors and thresholds used by this fuction
+ * were determined based on theoretical maximum wire speed and testing
+ * data, in order to minimize response time while increasing bulk
+ * throughput.
+ * This functionality is controlled by the InterruptThrottleRate module
+ * parameter (see igb_param.c)
+ * NOTE: This function is called only when operating in a multiqueue
+ * receive environment.
+ * @rx_ring: pointer to ring
+ **/
+static void igb_update_ring_itr(struct igb_ring *rx_ring)
{
- struct e1000_hw *hw = &adapter->hw;
- int new_val;
+ int new_val = rx_ring->itr_val;
+ int avg_wire_size = 0;
+ struct igb_adapter *adapter = rx_ring->adapter;
- new_val = rx_ring->itr_val / 2;
- if (new_val < IGB_MIN_DYN_ITR)
- new_val = IGB_MIN_DYN_ITR;
+ if (!rx_ring->total_packets)
+ goto clear_counts; /* no packets, so don't do anything */
- if (new_val != rx_ring->itr_val) {
- rx_ring->itr_val = new_val;
- wr32(rx_ring->itr_register,
- 1000000000 / (new_val * 256));
+ /* For non-gigabit speeds, just fix the interrupt rate at 4000
+ * ints/sec - ITR timer value of 120 ticks.
+ */
+ if (adapter->link_speed != SPEED_1000) {
+ new_val = 120;
+ goto set_itr_val;
}
-}
+ avg_wire_size = rx_ring->total_bytes / rx_ring->total_packets;
-static void igb_raise_rx_eitr(struct igb_adapter *adapter,
- struct igb_ring *rx_ring)
-{
- struct e1000_hw *hw = &adapter->hw;
- int new_val;
+ /* Add 24 bytes to size to account for CRC, preamble, and gap */
+ avg_wire_size += 24;
+
+ /* Don't starve jumbo frames */
+ avg_wire_size = min(avg_wire_size, 3000);
- new_val = rx_ring->itr_val * 2;
- if (new_val > IGB_MAX_DYN_ITR)
- new_val = IGB_MAX_DYN_ITR;
+ /* Give a little boost to mid-size frames */
+ if ((avg_wire_size > 300) && (avg_wire_size < 1200))
+ new_val = avg_wire_size / 3;
+ else
+ new_val = avg_wire_size / 2;
+set_itr_val:
if (new_val != rx_ring->itr_val) {
rx_ring->itr_val = new_val;
- wr32(rx_ring->itr_register,
- 1000000000 / (new_val * 256));
+ rx_ring->set_itr = 1;
}
+clear_counts:
+ rx_ring->total_bytes = 0;
+ rx_ring->total_packets = 0;
}
/**
@@ -2561,8 +2585,7 @@ update_itr_done:
return retval;
}
-static void igb_set_itr(struct igb_adapter *adapter, u16 itr_register,
- int rx_only)
+static void igb_set_itr(struct igb_adapter *adapter)
{
u16 current_itr;
u32 new_itr = adapter->itr;
@@ -2578,26 +2601,23 @@ static void igb_set_itr(struct igb_adapter *adapter, u16 itr_register,
adapter->rx_itr,
adapter->rx_ring->total_packets,
adapter->rx_ring->total_bytes);
- /* conservative mode (itr 3) eliminates the lowest_latency setting */
- if (adapter->itr_setting == 3 && adapter->rx_itr == lowest_latency)
- adapter->rx_itr = low_latency;
- if (!rx_only) {
+ if (adapter->rx_ring->buddy) {
adapter->tx_itr = igb_update_itr(adapter,
adapter->tx_itr,
adapter->tx_ring->total_packets,
adapter->tx_ring->total_bytes);
- /* conservative mode (itr 3) eliminates the
- * lowest_latency setting */
- if (adapter->itr_setting == 3 &&
- adapter->tx_itr == lowest_latency)
- adapter->tx_itr = low_latency;
current_itr = max(adapter->rx_itr, adapter->tx_itr);
} else {
current_itr = adapter->rx_itr;
}
+ /* conservative mode (itr 3) eliminates the lowest_latency setting */
+ if (adapter->itr_setting == 3 &&
+ current_itr == lowest_latency)
+ current_itr = low_latency;
+
switch (current_itr) {
/* counts and packets in update_itr are dependent on these numbers */
case lowest_latency:
@@ -2614,6 +2634,13 @@ static void igb_set_itr(struct igb_adapter *adapter, u16 itr_register,
}
set_itr_now:
+ adapter->rx_ring->total_bytes = 0;
+ adapter->rx_ring->total_packets = 0;
+ if (adapter->rx_ring->buddy) {
+ adapter->rx_ring->buddy->total_bytes = 0;
+ adapter->rx_ring->buddy->total_packets = 0;
+ }
+
if (new_itr != adapter->itr) {
/* this attempts to bias the interrupt rate towards Bulk
* by adding intermediate steps when interrupt rate is
@@ -2628,7 +2655,8 @@ set_itr_now:
* ends up being correct.
*/
adapter->itr = new_itr;
- adapter->set_itr = 1;
+ adapter->rx_ring->itr_val = 1000000000 / (new_itr * 256);
+ adapter->rx_ring->set_itr = 1;
}
return;
@@ -2979,6 +3007,7 @@ static int igb_xmit_frame_ring_adv(struct sk_buff *skb,
/* this is a hard error */
return NETDEV_TX_BUSY;
}
+ skb_orphan(skb);
if (adapter->vlgrp && vlan_tx_tag_present(skb)) {
tx_flags |= IGB_TX_FLAGS_VLAN;
@@ -3312,8 +3341,6 @@ static irqreturn_t igb_msix_tx(int irq, void *data)
struct igb_adapter *adapter = tx_ring->adapter;
struct e1000_hw *hw = &adapter->hw;
- if (!tx_ring->itr_val)
- wr32(E1000_EIMC, tx_ring->eims_value);
#ifdef CONFIG_DCA
if (adapter->flags & IGB_FLAG_DCA_ENABLED)
igb_update_tx_dca(tx_ring);
@@ -3332,21 +3359,36 @@ static irqreturn_t igb_msix_tx(int irq, void *data)
return IRQ_HANDLED;
}
+static void igb_write_itr(struct igb_ring *ring)
+{
+ struct e1000_hw *hw = &ring->adapter->hw;
+ if ((ring->adapter->itr_setting & 3) && ring->set_itr) {
+ switch (hw->mac.type) {
+ case e1000_82576:
+ wr32(ring->itr_register,
+ ring->itr_val |
+ 0x80000000);
+ break;
+ default:
+ wr32(ring->itr_register,
+ ring->itr_val |
+ (ring->itr_val << 16));
+ break;
+ }
+ ring->set_itr = 0;
+ }
+}
+
static irqreturn_t igb_msix_rx(int irq, void *data)
{
struct igb_ring *rx_ring = data;
struct igb_adapter *adapter = rx_ring->adapter;
- struct e1000_hw *hw = &adapter->hw;
/* Write the ITR value calculated at the end of the
* previous interrupt.
*/
- if (adapter->set_itr) {
- wr32(rx_ring->itr_register,
- 1000000000 / (rx_ring->itr_val * 256));
- adapter->set_itr = 0;
- }
+ igb_write_itr(rx_ring);
if (netif_rx_schedule_prep(adapter->netdev, &rx_ring->napi))
__netif_rx_schedule(adapter->netdev, &rx_ring->napi);
@@ -3493,13 +3535,7 @@ static irqreturn_t igb_intr_msi(int irq, void *data)
/* read ICR disables interrupts using IAM */
u32 icr = rd32(E1000_ICR);
- /* Write the ITR value calculated at the end of the
- * previous interrupt.
- */
- if (adapter->set_itr) {
- wr32(E1000_ITR, 1000000000 / (adapter->itr * 256));
- adapter->set_itr = 0;
- }
+ igb_write_itr(adapter->rx_ring);
if (icr & (E1000_ICR_RXSEQ | E1000_ICR_LSC)) {
hw->mac.get_link_status = 1;
@@ -3529,13 +3565,7 @@ static irqreturn_t igb_intr(int irq, void *data)
if (!icr)
return IRQ_NONE; /* Not our interrupt */
- /* Write the ITR value calculated at the end of the
- * previous interrupt.
- */
- if (adapter->set_itr) {
- wr32(E1000_ITR, 1000000000 / (adapter->itr * 256));
- adapter->set_itr = 0;
- }
+ igb_write_itr(adapter->rx_ring);
/* IMS will not auto-mask if INT_ASSERTED is not set, and if it is
* not set, then the adapter didn't send an interrupt */
@@ -3585,7 +3615,7 @@ static int igb_poll(struct napi_struct *napi, int budget)
if ((tx_clean_complete && (work_done < budget)) ||
!netif_running(netdev)) {
if (adapter->itr_setting & 3)
- igb_set_itr(adapter, E1000_ITR, false);
+ igb_set_itr(adapter);
netif_rx_complete(netdev, napi);
if (!test_bit(__IGB_DOWN, &adapter->state))
igb_irq_enable(adapter);
@@ -3619,15 +3649,11 @@ static int igb_clean_rx_ring_msix(struct napi_struct *napi, int budget)
quit_polling:
netif_rx_complete(netdev, napi);
- wr32(E1000_EIMS, rx_ring->eims_value);
- if ((adapter->itr_setting & 3) && !rx_ring->no_itr_adjust &&
- (rx_ring->total_packets > IGB_DYN_ITR_PACKET_THRESHOLD)) {
- int mean_size = rx_ring->total_bytes /
- rx_ring->total_packets;
- if (mean_size < IGB_DYN_ITR_LENGTH_LOW)
- igb_raise_rx_eitr(adapter, rx_ring);
- else if (mean_size > IGB_DYN_ITR_LENGTH_HIGH)
- igb_lower_rx_eitr(adapter, rx_ring);
+ if (adapter->itr_setting & 3) {
+ if (adapter->num_rx_queues == 1)
+ igb_set_itr(adapter);
+ else
+ igb_update_ring_itr(rx_ring);
}
if (!test_bit(__IGB_DOWN, &adapter->state))
@@ -3972,7 +3998,6 @@ send_up:
dev_kfree_skb_irq(skb);
goto next_desc;
}
- rx_ring->no_itr_adjust |= (staterr & E1000_RXD_STAT_DYNINT);
total_bytes += skb->len;
total_packets++;
^ permalink raw reply related [flat|nested] 15+ messages in thread