[PATCH] e1000: Fix adapter structure and prepare for multique fix

Fix adapter structure to handle multiple queues and prepping the driver for full multiple queue support, some changes are ifdef'd our unless you define CONFIG_E1000_MQ.

Signed-off-by: Jeff Kirsher <jeffrey.t.kirsher@intel.com>
Signed-off-by: Jesse Brandeburg <jesse.brandeburg@intel.com>
Signed-off-by: John Ronciak <john.ronciak@intel.com>
Signed-off-by: Jeff Garzik <jgarzik@pobox.com>

diff --git a/drivers/net/e1000/e1000.h b/drivers/net/e1000/e1000.h
index 4dd1326dd04fe8d3009e830a8a970d5638fc20e9..5940f7a223bc7e4fbccf841ebcffb63cdb7d826d 100644 (file)
--- a/drivers/net/e1000/e1000.h
+++ b/drivers/net/e1000/e1000.h
@@ -268,6 +268,7 @@ struct e1000_adapter {
 #ifdef CONFIG_E1000_MQ
        struct e1000_tx_ring **cpu_tx_ring; /* per-cpu */
 #endif
+       unsigned long tx_queue_len;
        uint32_t txd_cmd;
        uint32_t tx_int_delay;
        uint32_t tx_abs_int_delay;
@@ -303,7 +304,8 @@ struct e1000_adapter {
        struct call_async_data_struct rx_sched_call_data;
        int cpu_for_queue[4];
 #endif
-       int num_queues;
+       int num_tx_queues;
+       int num_rx_queues;
 
        uint64_t hw_csum_err;
        uint64_t hw_csum_good;
@@ -336,6 +338,7 @@ struct e1000_adapter {
        struct e1000_rx_ring test_rx_ring;
 
 
+       u32 *config_space;
        int msg_enable;
 #ifdef CONFIG_PCI_MSI
        boolean_t have_msi;

diff --git a/drivers/net/e1000/e1000_ethtool.c b/drivers/net/e1000/e1000_ethtool.c
index 54ae8805e79f95593da117f4d336a954127178a1..fa9a4659369af08c875b71db66102826d25f99f0 100644 (file)
--- a/drivers/net/e1000/e1000_ethtool.c
+++ b/drivers/net/e1000/e1000_ethtool.c
@@ -614,8 +614,8 @@ e1000_set_ringparam(struct net_device *netdev,
        struct e1000_rx_ring *rxdr, *rx_old, *rx_new;
        int i, err, tx_ring_size, rx_ring_size;
 
-       tx_ring_size = sizeof(struct e1000_tx_ring) * adapter->num_queues;
-       rx_ring_size = sizeof(struct e1000_rx_ring) * adapter->num_queues;
+       tx_ring_size = sizeof(struct e1000_tx_ring) * adapter->num_tx_queues;
+       rx_ring_size = sizeof(struct e1000_rx_ring) * adapter->num_rx_queues;
 
        if (netif_running(adapter->netdev))
                e1000_down(adapter);
@@ -654,10 +654,10 @@ e1000_set_ringparam(struct net_device *netdev,
                E1000_MAX_TXD : E1000_MAX_82544_TXD));
        E1000_ROUNDUP(txdr->count, REQ_TX_DESCRIPTOR_MULTIPLE); 
 
-       for (i = 0; i < adapter->num_queues; i++) {
+       for (i = 0; i < adapter->num_tx_queues; i++)
                txdr[i].count = txdr->count;
+       for (i = 0; i < adapter->num_rx_queues; i++)
                rxdr[i].count = rxdr->count;
-       }
 
        if(netif_running(adapter->netdev)) {
                /* Try to get new resources before deleting old */

diff --git a/drivers/net/e1000/e1000_hw.h b/drivers/net/e1000/e1000_hw.h
index 7caa35748ceac9c8236d1510f9335b2e57f9073b..1ddfd56fc5df31453bf6ce67c87fae19207b45be 100644 (file)
--- a/drivers/net/e1000/e1000_hw.h
+++ b/drivers/net/e1000/e1000_hw.h
@@ -1497,6 +1497,7 @@ struct e1000_hw {
 #define E1000_CTRL_EXT_EE_RST    0x00002000 /* Reinitialize from EEPROM */
 #define E1000_CTRL_EXT_IPS       0x00004000 /* Invert Power State */
 #define E1000_CTRL_EXT_SPD_BYPS  0x00008000 /* Speed Select Bypass */
+#define E1000_CTRL_EXT_RO_DIS    0x00020000 /* Relaxed Ordering disable */
 #define E1000_CTRL_EXT_LINK_MODE_MASK 0x00C00000
 #define E1000_CTRL_EXT_LINK_MODE_GMII 0x00000000
 #define E1000_CTRL_EXT_LINK_MODE_TBI  0x00C00000

diff --git a/drivers/net/e1000/e1000_main.c b/drivers/net/e1000/e1000_main.c
index 5f848b3a3a4d6a04c512689fab40173f7687d224..540c8561fc246c44bf15d76c61a54211103ee67a 100644 (file)
--- a/drivers/net/e1000/e1000_main.c
+++ b/drivers/net/e1000/e1000_main.c
@@ -411,8 +411,9 @@ e1000_up(struct e1000_adapter *adapter)
        e1000_configure_tx(adapter);
        e1000_setup_rctl(adapter);
        e1000_configure_rx(adapter);
-       for (i = 0; i < adapter->num_queues; i++)
+       for (i = 0; i < adapter->num_rx_queues; i++) {
                adapter->alloc_rx_buf(adapter, &adapter->rx_ring[i]);
+       }
 
 #ifdef CONFIG_PCI_MSI
        if(adapter->hw.mac_type > e1000_82547_rev_2) {
@@ -867,7 +868,7 @@ e1000_remove(struct pci_dev *pdev)
 
        unregister_netdev(netdev);
 #ifdef CONFIG_E1000_NAPI
-       for (i = 0; i < adapter->num_queues; i++)
+       for (i = 0; i < adapter->num_rx_queues; i++)
                __dev_put(&adapter->polling_netdev[i]);
 #endif
 
@@ -972,15 +973,25 @@ e1000_sw_init(struct e1000_adapter *adapter)
        switch (hw->mac_type) {
        case e1000_82571:
        case e1000_82572:
-               adapter->num_queues = 2;
+               /* These controllers support 2 tx queues, but with a single
+                * qdisc implementation, multiple tx queues aren't quite as
+                * interesting.  If we can find a logical way of mapping
+                * flows to a queue, then perhaps we can up the num_tx_queue
+                * count back to its default.  Until then, we run the risk of
+                * terrible performance due to SACK overload. */
+               adapter->num_tx_queues = 1;
+               adapter->num_rx_queues = 2;
                break;
        default:
-               adapter->num_queues = 1;
+               adapter->num_tx_queues = 1;
+               adapter->num_rx_queues = 1;
                break;
        }
-       adapter->num_queues = min(adapter->num_queues, num_online_cpus());
+       adapter->num_rx_queues = min(adapter->num_rx_queues, num_online_cpus());
+       adapter->num_tx_queues = min(adapter->num_tx_queues, num_online_cpus());
 #else
-       adapter->num_queues = 1;
+       adapter->num_tx_queues = 1;
+       adapter->num_rx_queues = 1;
 #endif
 
        if (e1000_alloc_queues(adapter)) {
@@ -989,7 +1000,7 @@ e1000_sw_init(struct e1000_adapter *adapter)
        }
 
 #ifdef CONFIG_E1000_NAPI
-       for (i = 0; i < adapter->num_queues; i++) {
+       for (i = 0; i < adapter->num_rx_queues; i++) {
                adapter->polling_netdev[i].priv = adapter;
                adapter->polling_netdev[i].poll = &e1000_clean;
                adapter->polling_netdev[i].weight = 64;
@@ -1022,13 +1033,13 @@ e1000_alloc_queues(struct e1000_adapter *adapter)
 {
        int size;
 
-       size = sizeof(struct e1000_tx_ring) * adapter->num_queues;
+       size = sizeof(struct e1000_tx_ring) * adapter->num_tx_queues;
        adapter->tx_ring = kmalloc(size, GFP_KERNEL);
        if (!adapter->tx_ring)
                return -ENOMEM;
        memset(adapter->tx_ring, 0, size);
 
-       size = sizeof(struct e1000_rx_ring) * adapter->num_queues;
+       size = sizeof(struct e1000_rx_ring) * adapter->num_rx_queues;
        adapter->rx_ring = kmalloc(size, GFP_KERNEL);
        if (!adapter->rx_ring) {
                kfree(adapter->tx_ring);
@@ -1037,7 +1048,7 @@ e1000_alloc_queues(struct e1000_adapter *adapter)
        memset(adapter->rx_ring, 0, size);
 
 #ifdef CONFIG_E1000_NAPI
-       size = sizeof(struct net_device) * adapter->num_queues;
+       size = sizeof(struct net_device) * adapter->num_rx_queues;
        adapter->polling_netdev = kmalloc(size, GFP_KERNEL);
        if (!adapter->polling_netdev) {
                kfree(adapter->tx_ring);
@@ -1066,12 +1077,12 @@ e1000_setup_queue_mapping(struct e1000_adapter *adapter)
        lock_cpu_hotplug();
        i = 0;
        for_each_online_cpu(cpu) {
-               *per_cpu_ptr(adapter->cpu_tx_ring, cpu) = &adapter->tx_ring[i % adapter->num_queues];
+               *per_cpu_ptr(adapter->cpu_tx_ring, cpu) = &adapter->tx_ring[i % adapter->num_tx_queues];
                /* This is incomplete because we'd like to assign separate
                 * physical cpus to these netdev polling structures and
                 * avoid saturating a subset of cpus.
                 */
-               if (i < adapter->num_queues) {
+               if (i < adapter->num_rx_queues) {
                        *per_cpu_ptr(adapter->cpu_netdev, cpu) = &adapter->polling_netdev[i];
                        adapter->cpu_for_queue[i] = cpu;
                } else
@@ -1291,7 +1302,7 @@ e1000_setup_all_tx_resources(struct e1000_adapter *adapter)
 {
        int i, err = 0;
 
-       for (i = 0; i < adapter->num_queues; i++) {
+       for (i = 0; i < adapter->num_tx_queues; i++) {
                err = e1000_setup_tx_resources(adapter, &adapter->tx_ring[i]);
                if (err) {
                        DPRINTK(PROBE, ERR,
@@ -1319,7 +1330,7 @@ e1000_configure_tx(struct e1000_adapter *adapter)
 
        /* Setup the HW Tx Head and Tail descriptor pointers */
 
-       switch (adapter->num_queues) {
+       switch (adapter->num_tx_queues) {
        case 2:
                tdba = adapter->tx_ring[1].dma;
                tdlen = adapter->tx_ring[1].count *
@@ -1537,7 +1548,7 @@ e1000_setup_all_rx_resources(struct e1000_adapter *adapter)
 {
        int i, err = 0;
 
-       for (i = 0; i < adapter->num_queues; i++) {
+       for (i = 0; i < adapter->num_rx_queues; i++) {
                err = e1000_setup_rx_resources(adapter, &adapter->rx_ring[i]);
                if (err) {
                        DPRINTK(PROBE, ERR,
@@ -1709,7 +1720,7 @@ e1000_configure_rx(struct e1000_adapter *adapter)
 
        /* Setup the HW Rx Head and Tail Descriptor Pointers and
         * the Base and Length of the Rx Descriptor Ring */
-       switch (adapter->num_queues) {
+       switch (adapter->num_rx_queues) {
 #ifdef CONFIG_E1000_MQ
        case 2:
                rdba = adapter->rx_ring[1].dma;
@@ -1736,7 +1747,7 @@ e1000_configure_rx(struct e1000_adapter *adapter)
        }
 
 #ifdef CONFIG_E1000_MQ
-       if (adapter->num_queues > 1) {
+       if (adapter->num_rx_queues > 1) {
                uint32_t random[10];
 
                get_random_bytes(&random[0], 40);
@@ -1746,7 +1757,7 @@ e1000_configure_rx(struct e1000_adapter *adapter)
                        E1000_WRITE_REG(hw, RSSIM, 0);
                }
 
-               switch (adapter->num_queues) {
+               switch (adapter->num_rx_queues) {
                case 2:
                default:
                        reta = 0x00800080;
@@ -1838,7 +1849,7 @@ e1000_free_all_tx_resources(struct e1000_adapter *adapter)
 {
        int i;
 
-       for (i = 0; i < adapter->num_queues; i++)
+       for (i = 0; i < adapter->num_tx_queues; i++)
                e1000_free_tx_resources(adapter, &adapter->tx_ring[i]);
 }
 
@@ -1905,7 +1916,7 @@ e1000_clean_all_tx_rings(struct e1000_adapter *adapter)
 {
        int i;
 
-       for (i = 0; i < adapter->num_queues; i++)
+       for (i = 0; i < adapter->num_tx_queues; i++)
                e1000_clean_tx_ring(adapter, &adapter->tx_ring[i]);
 }
 
@@ -1949,7 +1960,7 @@ e1000_free_all_rx_resources(struct e1000_adapter *adapter)
 {
        int i;
 
-       for (i = 0; i < adapter->num_queues; i++)
+       for (i = 0; i < adapter->num_rx_queues; i++)
                e1000_free_rx_resources(adapter, &adapter->rx_ring[i]);
 }
 
@@ -2025,7 +2036,7 @@ e1000_clean_all_rx_rings(struct e1000_adapter *adapter)
 {
        int i;
 
-       for (i = 0; i < adapter->num_queues; i++)
+       for (i = 0; i < adapter->num_rx_queues; i++)
                e1000_clean_rx_ring(adapter, &adapter->rx_ring[i]);
 }
 
@@ -2325,7 +2336,10 @@ e1000_watchdog_task(struct e1000_adapter *adapter)
 
        e1000_update_adaptive(&adapter->hw);
 
-       if (adapter->num_queues == 1 && !netif_carrier_ok(netdev)) {
+#ifdef CONFIG_E1000_MQ
+       txdr = *per_cpu_ptr(adapter->cpu_tx_ring, smp_processor_id());
+#endif
+       if (!netif_carrier_ok(netdev)) {
                if (E1000_DESC_UNUSED(txdr) + 1 < txdr->count) {
                        /* We've lost link, so the controller stops DMA,
                         * but we've got queued Tx work that's never going
@@ -3197,14 +3211,12 @@ e1000_intr(int irq, void *data, struct pt_regs *regs)
        E1000_WRITE_FLUSH(hw);
 #ifdef CONFIG_E1000_MQ
        if (atomic_read(&adapter->rx_sched_call_data.count) == 0) {
-               cpu_set(adapter->cpu_for_queue[0],
-                       adapter->rx_sched_call_data.cpumask);
-               for (i = 1; i < adapter->num_queues; i++) {
-                       cpu_set(adapter->cpu_for_queue[i],
-                               adapter->rx_sched_call_data.cpumask);
-                       atomic_inc(&adapter->irq_sem);
-               }
-               atomic_set(&adapter->rx_sched_call_data.count, i);
+               /* We must setup the cpumask once count == 0 since
+                * each cpu bit is cleared when the work is done. */
+               adapter->rx_sched_call_data.cpumask = adapter->cpumask;
+               atomic_add(adapter->num_rx_queues - 1, &adapter->irq_sem);
+               atomic_set(&adapter->rx_sched_call_data.count,
+                          adapter->num_rx_queues);
                smp_call_async_mask(&adapter->rx_sched_call_data);
        } else {
                printk("call_data.count == %u\n", atomic_read(&adapter->rx_sched_call_data.count));
@@ -3267,7 +3279,7 @@ e1000_clean(struct net_device *poll_dev, int *budget)
 
        while (poll_dev != &adapter->polling_netdev[i]) {
                i++;
-               if (unlikely(i == adapter->num_queues))
+               if (unlikely(i == adapter->num_rx_queues))
                        BUG();
        }
 

diff --git a/drivers/net/e1000/e1000_param.c b/drivers/net/e1000/e1000_param.c
index 852841f12fb9faa1c939699c7fe5e19f678f1fd9..b0be7d69af34493231cb91af5613807c705b3c5b 100644 (file)
--- a/drivers/net/e1000/e1000_param.c
+++ b/drivers/net/e1000/e1000_param.c
@@ -320,7 +320,7 @@ e1000_check_options(struct e1000_adapter *adapter)
                } else {
                        tx_ring->count = opt.def;
                }
-               for (i = 0; i < adapter->num_queues; i++)
+               for (i = 0; i < adapter->num_tx_queues; i++)
                        tx_ring[i].count = tx_ring->count;
        }
        { /* Receive Descriptor Count */
@@ -346,7 +346,7 @@ e1000_check_options(struct e1000_adapter *adapter)
                } else {
                        rx_ring->count = opt.def;
                }
-               for (i = 0; i < adapter->num_queues; i++)
+               for (i = 0; i < adapter->num_rx_queues; i++)
                        rx_ring[i].count = rx_ring->count;
        }
        { /* Checksum Offload Enable/Disable */