bnx2x: MSI support
authorEilon Greenstein <eilong@broadcom.com>
Thu, 12 Feb 2009 08:36:15 +0000 (08:36 +0000)
committerDavid S. Miller <davem@davemloft.net>
Mon, 16 Feb 2009 07:31:05 +0000 (23:31 -0800)
Enabling MSI on top of MSI-X and INTA. Also changing the module parameter to
allow choosing INTA or MSI even when MSI-X is available. The default status
block should not be reversed for endianity. Since MSI can issue
re-configuration, the interrupt disable function now requires mmiowb

Signed-off-by: Eilon Greenstein <eilong@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
drivers/net/bnx2x.h
drivers/net/bnx2x_hsi.h
drivers/net/bnx2x_main.c
drivers/net/bnx2x_reg.h

index 7edad1f9b3309b555b76c73b0429c4bdbb305326..153cc9564b4cc380ce088b3c8fbd81d8f766fb48 100644 (file)
@@ -740,6 +740,9 @@ struct bnx2x {
 
        atomic_t                intr_sem;
        struct msix_entry       msix_table[MAX_CONTEXT+1];
+#define INT_MODE_INTx                  1
+#define INT_MODE_MSI                   2
+#define INT_MODE_MSIX                  3
 
        int                     tx_ring_size;
 
@@ -794,7 +797,7 @@ struct bnx2x {
 #define NO_WOL_FLAG                    8
 #define USING_DAC_FLAG                 0x10
 #define USING_MSIX_FLAG                        0x20
-#define ASF_ENABLE_FLAG                        0x40
+#define USING_MSI_FLAG                 0x40
 #define TPA_ENABLE_FLAG                        0x80
 #define NO_MCP_FLAG                    0x100
 #define BP_NOMCP(bp)                   (bp->flags & NO_MCP_FLAG)
index c41682353050ef19358bb7de289d46c2f974d9a4..25751a2b7a24d130c6867d0ad1153e517ed306c9 100644 (file)
@@ -1224,15 +1224,9 @@ struct host_func_stats {
 struct atten_def_status_block {
        u32 attn_bits;
        u32 attn_bits_ack;
-#if defined(__BIG_ENDIAN)
-       u16 attn_bits_index;
-       u8 reserved0;
-       u8 status_block_id;
-#elif defined(__LITTLE_ENDIAN)
        u8 status_block_id;
        u8 reserved0;
        u16 attn_bits_index;
-#endif
        u32 reserved1;
 };
 
index 60762f769efc038a3b8a1f04e2cd0ec53019e844..db8506d08fa520267a14f9fba0bfe56abb188f0f 100644 (file)
@@ -77,17 +77,19 @@ static int multi_mode = 1;
 module_param(multi_mode, int, 0);
 
 static int disable_tpa;
-static int use_inta;
 static int poll;
 static int debug;
 static int load_count[3]; /* 0-common, 1-port0, 2-port1 */
 
 module_param(disable_tpa, int, 0);
-module_param(use_inta, int, 0);
+
+static int int_mode;
+module_param(int_mode, int, 0);
+MODULE_PARM_DESC(int_mode, " Force interrupt mode (1 INT#x; 2 MSI)");
+
 module_param(poll, int, 0);
 module_param(debug, int, 0);
 MODULE_PARM_DESC(disable_tpa, "disable the TPA (LRO) feature");
-MODULE_PARM_DESC(use_inta, "use INT#A instead of MSI-X");
 MODULE_PARM_DESC(poll, "use polling (for debug)");
 MODULE_PARM_DESC(debug, "default debug msglevel");
 
@@ -588,34 +590,41 @@ static void bnx2x_int_enable(struct bnx2x *bp)
        u32 addr = port ? HC_REG_CONFIG_1 : HC_REG_CONFIG_0;
        u32 val = REG_RD(bp, addr);
        int msix = (bp->flags & USING_MSIX_FLAG) ? 1 : 0;
+       int msi = (bp->flags & USING_MSI_FLAG) ? 1 : 0;
 
        if (msix) {
-               val &= ~HC_CONFIG_0_REG_SINGLE_ISR_EN_0;
+               val &= ~(HC_CONFIG_0_REG_SINGLE_ISR_EN_0 |
+                        HC_CONFIG_0_REG_INT_LINE_EN_0);
                val |= (HC_CONFIG_0_REG_MSI_MSIX_INT_EN_0 |
                        HC_CONFIG_0_REG_ATTN_BIT_EN_0);
+       } else if (msi) {
+               val &= ~HC_CONFIG_0_REG_INT_LINE_EN_0;
+               val |= (HC_CONFIG_0_REG_SINGLE_ISR_EN_0 |
+                       HC_CONFIG_0_REG_MSI_MSIX_INT_EN_0 |
+                       HC_CONFIG_0_REG_ATTN_BIT_EN_0);
        } else {
                val |= (HC_CONFIG_0_REG_SINGLE_ISR_EN_0 |
                        HC_CONFIG_0_REG_MSI_MSIX_INT_EN_0 |
                        HC_CONFIG_0_REG_INT_LINE_EN_0 |
                        HC_CONFIG_0_REG_ATTN_BIT_EN_0);
 
-               DP(NETIF_MSG_INTR, "write %x to HC %d (addr 0x%x)  MSI-X %d\n",
-                  val, port, addr, msix);
+               DP(NETIF_MSG_INTR, "write %x to HC %d (addr 0x%x)\n",
+                  val, port, addr);
 
                REG_WR(bp, addr, val);
 
                val &= ~HC_CONFIG_0_REG_MSI_MSIX_INT_EN_0;
        }
 
-       DP(NETIF_MSG_INTR, "write %x to HC %d (addr 0x%x)  MSI-X %d\n",
-          val, port, addr, msix);
+       DP(NETIF_MSG_INTR, "write %x to HC %d (addr 0x%x)  mode %s\n",
+          val, port, addr, (msix ? "MSI-X" : (msi ? "MSI" : "INTx")));
 
        REG_WR(bp, addr, val);
 
        if (CHIP_IS_E1H(bp)) {
                /* init leading/trailing edge */
                if (IS_E1HMF(bp)) {
-                       val = (0xfe0f | (1 << (BP_E1HVN(bp) + 4)));
+                       val = (0xee0f | (1 << (BP_E1HVN(bp) + 4)));
                        if (bp->port.pmf)
                                /* enable nig attention */
                                val |= 0x0100;
@@ -641,6 +650,9 @@ static void bnx2x_int_disable(struct bnx2x *bp)
        DP(NETIF_MSG_INTR, "write %x to HC %d (addr 0x%x)\n",
           val, port, addr);
 
+       /* flush all outstanding writes */
+       mmiowb();
+
        REG_WR(bp, addr, val);
        if (REG_RD(bp, addr) != val)
                BNX2X_ERR("BUG! proper val not read from IGU!\n");
@@ -649,7 +661,7 @@ static void bnx2x_int_disable(struct bnx2x *bp)
 static void bnx2x_int_disable_sync(struct bnx2x *bp, int disable_hw)
 {
        int msix = (bp->flags & USING_MSIX_FLAG) ? 1 : 0;
-       int i;
+       int i, offset;
 
        /* disable interrupt handling */
        atomic_inc(&bp->intr_sem);
@@ -659,11 +671,10 @@ static void bnx2x_int_disable_sync(struct bnx2x *bp, int disable_hw)
 
        /* make sure all ISRs are done */
        if (msix) {
+               synchronize_irq(bp->msix_table[0].vector);
+               offset = 1;
                for_each_queue(bp, i)
-                       synchronize_irq(bp->msix_table[i].vector);
-
-               /* one more for the Slow Path IRQ */
-               synchronize_irq(bp->msix_table[i].vector);
+                       synchronize_irq(bp->msix_table[i + offset].vector);
        } else
                synchronize_irq(bp->pdev->irq);
 
@@ -5198,6 +5209,8 @@ static int bnx2x_init_common(struct bnx2x *bp)
        REG_WR(bp, PXP2_REG_RQ_SRC_ENDIAN_M, 1);
        REG_WR(bp, PXP2_REG_RQ_CDU_ENDIAN_M, 1);
        REG_WR(bp, PXP2_REG_RQ_DBG_ENDIAN_M, 1);
+       /* make sure this value is 0 */
+       REG_WR(bp, PXP2_REG_RQ_HC_ENDIAN_M, 0);
 
 /*     REG_WR(bp, PXP2_REG_RD_PBF_SWAP_MODE, 1); */
        REG_WR(bp, PXP2_REG_RD_QM_SWAP_MODE, 1);
@@ -5648,10 +5661,17 @@ static int bnx2x_init_func(struct bnx2x *bp)
 {
        int port = BP_PORT(bp);
        int func = BP_FUNC(bp);
+       u32 addr, val;
        int i;
 
        DP(BNX2X_MSG_MCP, "starting func init  func %x\n", func);
 
+       /* set MSI reconfigure capability */
+       addr = (port ? HC_REG_CONFIG_1 : HC_REG_CONFIG_0);
+       val = REG_RD(bp, addr);
+       val |= HC_CONFIG_0_REG_MSI_ATTN_EN_0;
+       REG_WR(bp, addr, val);
+
        i = FUNC_ILT_BASE(func);
 
        bnx2x_ilt_wr(bp, i, bnx2x_sp_mapping(bp, context));
@@ -6053,10 +6073,6 @@ static void bnx2x_free_msix_irqs(struct bnx2x *bp)
                   "state %x\n", i, bp->msix_table[i + offset].vector,
                   bnx2x_fp(bp, i, state));
 
-               if (bnx2x_fp(bp, i, state) != BNX2X_FP_STATE_CLOSED)
-                       BNX2X_ERR("IRQ of fp #%d being freed while "
-                                 "state != closed\n", i);
-
                free_irq(bp->msix_table[i + offset].vector, &bp->fp[i]);
        }
 }
@@ -6068,21 +6084,25 @@ static void bnx2x_free_irq(struct bnx2x *bp)
                pci_disable_msix(bp->pdev);
                bp->flags &= ~USING_MSIX_FLAG;
 
+       } else if (bp->flags & USING_MSI_FLAG) {
+               free_irq(bp->pdev->irq, bp->dev);
+               pci_disable_msi(bp->pdev);
+               bp->flags &= ~USING_MSI_FLAG;
+
        } else
                free_irq(bp->pdev->irq, bp->dev);
 }
 
 static int bnx2x_enable_msix(struct bnx2x *bp)
 {
-       int i, rc, offset;
+       int i, rc, offset = 1;
+       int igu_vec = 0;
 
-       bp->msix_table[0].entry = 0;
-       offset = 1;
-       DP(NETIF_MSG_IFUP, "msix_table[0].entry = 0 (slowpath)\n");
+       bp->msix_table[0].entry = igu_vec;
+       DP(NETIF_MSG_IFUP, "msix_table[0].entry = %d (slowpath)\n", igu_vec);
 
        for_each_queue(bp, i) {
-               int igu_vec = offset + i + BP_L_ID(bp);
-
+               igu_vec = BP_L_ID(bp) + offset + i;
                bp->msix_table[i + offset].entry = igu_vec;
                DP(NETIF_MSG_IFUP, "msix_table[%d].entry = %d "
                   "(fastpath #%u)\n", i + offset, igu_vec, i);
@@ -6091,9 +6111,10 @@ static int bnx2x_enable_msix(struct bnx2x *bp)
        rc = pci_enable_msix(bp->pdev, &bp->msix_table[0],
                             BNX2X_NUM_QUEUES(bp) + offset);
        if (rc) {
-               DP(NETIF_MSG_IFUP, "MSI-X is not attainable\n");
-               return -1;
+               DP(NETIF_MSG_IFUP, "MSI-X is not attainable  rc %d\n", rc);
+               return rc;
        }
+
        bp->flags |= USING_MSIX_FLAG;
 
        return 0;
@@ -6140,11 +6161,31 @@ static int bnx2x_req_msix_irqs(struct bnx2x *bp)
        return 0;
 }
 
+static int bnx2x_enable_msi(struct bnx2x *bp)
+{
+       int rc;
+
+       rc = pci_enable_msi(bp->pdev);
+       if (rc) {
+               DP(NETIF_MSG_IFUP, "MSI is not attainable\n");
+               return -1;
+       }
+       bp->flags |= USING_MSI_FLAG;
+
+       return 0;
+}
+
 static int bnx2x_req_irq(struct bnx2x *bp)
 {
+       unsigned long flags;
        int rc;
 
-       rc = request_irq(bp->pdev->irq, bnx2x_interrupt, IRQF_SHARED,
+       if (bp->flags & USING_MSI_FLAG)
+               flags = 0;
+       else
+               flags = IRQF_SHARED;
+
+       rc = request_irq(bp->pdev->irq, bnx2x_interrupt, flags,
                         bp->dev->name, bp->dev);
        if (!rc)
                bnx2x_fp(bp, 0, state) = BNX2X_FP_STATE_IRQ;
@@ -6365,28 +6406,23 @@ static int bnx2x_setup_multi(struct bnx2x *bp, int index)
 }
 
 static int bnx2x_poll(struct napi_struct *napi, int budget);
-static void bnx2x_set_rx_mode(struct net_device *dev);
 
-/* must be called with rtnl_lock */
-static int bnx2x_nic_load(struct bnx2x *bp, int load_mode)
+static void bnx2x_set_int_mode(struct bnx2x *bp)
 {
-       u32 load_code;
-       int i, rc = 0;
        int num_queues;
-#ifdef BNX2X_STOP_ON_ERROR
-       if (unlikely(bp->panic))
-               return -EPERM;
-#endif
-
-       bp->state = BNX2X_STATE_OPENING_WAIT4_LOAD;
 
-       if (use_inta) {
+       switch (int_mode) {
+       case INT_MODE_INTx:
+       case INT_MODE_MSI:
                num_queues = 1;
                bp->num_rx_queues = num_queues;
                bp->num_tx_queues = num_queues;
                DP(NETIF_MSG_IFUP,
                   "set number of queues to %d\n", num_queues);
-       } else {
+               break;
+
+       case INT_MODE_MSIX:
+       default:
                if (bp->multi_mode == ETH_RSS_MODE_REGULAR)
                        num_queues = min_t(u32, num_online_cpus(),
                                           BNX2X_MAX_QUEUES(bp));
@@ -6401,8 +6437,7 @@ static int bnx2x_nic_load(struct bnx2x *bp, int load_mode)
                 * so try to enable MSI-X with the requested number of fp's
                 * and fallback to MSI or legacy INTx with one fp
                 */
-               rc = bnx2x_enable_msix(bp);
-               if (rc) {
+               if (bnx2x_enable_msix(bp)) {
                        /* failed to enable MSI-X */
                        num_queues = 1;
                        bp->num_rx_queues = num_queues;
@@ -6412,8 +6447,27 @@ static int bnx2x_nic_load(struct bnx2x *bp, int load_mode)
                                          "enable MSI-X  set number of "
                                          "queues to %d\n", num_queues);
                }
+               break;
        }
        bp->dev->real_num_tx_queues = bp->num_tx_queues;
+}
+
+static void bnx2x_set_rx_mode(struct net_device *dev);
+
+/* must be called with rtnl_lock */
+static int bnx2x_nic_load(struct bnx2x *bp, int load_mode)
+{
+       u32 load_code;
+       int i, rc = 0;
+#ifdef BNX2X_STOP_ON_ERROR
+       DP(NETIF_MSG_IFUP, "enter  load_mode %d\n", load_mode);
+       if (unlikely(bp->panic))
+               return -EPERM;
+#endif
+
+       bp->state = BNX2X_STATE_OPENING_WAIT4_LOAD;
+
+       bnx2x_set_int_mode(bp);
 
        if (bnx2x_alloc_mem(bp))
                return -ENOMEM;
@@ -6445,14 +6499,22 @@ static int bnx2x_nic_load(struct bnx2x *bp, int load_mode)
                        pci_disable_msix(bp->pdev);
                        goto load_error1;
                }
-               printk(KERN_INFO PFX "%s: using MSI-X\n", bp->dev->name);
        } else {
+               if ((rc != -ENOMEM) && (int_mode != INT_MODE_INTx))
+                       bnx2x_enable_msi(bp);
                bnx2x_ack_int(bp);
                rc = bnx2x_req_irq(bp);
                if (rc) {
                        BNX2X_ERR("IRQ request failed  rc %d, aborting\n", rc);
+                       if (bp->flags & USING_MSI_FLAG)
+                               pci_disable_msi(bp->pdev);
                        goto load_error1;
                }
+               if (bp->flags & USING_MSI_FLAG) {
+                       bp->dev->irq = bp->pdev->irq;
+                       printk(KERN_INFO PFX "%s: using MSI  IRQ %d\n",
+                              bp->dev->name, bp->pdev->irq);
+               }
        }
 
        /* Send LOAD_REQUEST command to MCP
@@ -6689,8 +6751,6 @@ static void bnx2x_reset_func(struct bnx2x *bp)
        REG_WR(bp, HC_REG_LEADING_EDGE_0 + port*8, 0);
        REG_WR(bp, HC_REG_TRAILING_EDGE_0 + port*8, 0);
 
-       REG_WR(bp, HC_REG_CONFIG_0 + port*4, 0x1000);
-
        /* Clear ILT */
        base = FUNC_ILT_BASE(func);
        for (i = base; i < base + ILT_PER_FUNC; i++)
@@ -7636,9 +7696,10 @@ static int __devinit bnx2x_init_bp(struct bnx2x *bp)
                       "MCP disabled, must load devices in order!\n");
 
        /* Set multi queue mode */
-       if ((multi_mode != ETH_RSS_MODE_DISABLED) && (!use_inta)) {
+       if ((multi_mode != ETH_RSS_MODE_DISABLED) &&
+           ((int_mode == INT_MODE_INTx) || (int_mode == INT_MODE_MSI))) {
                printk(KERN_ERR PFX
-                     "Multi disabled since INTA is requested\n");
+                     "Multi disabled since int_mode requested is not MSI-X\n");
                multi_mode = ETH_RSS_MODE_DISABLED;
        }
        bp->multi_mode = multi_mode;
index fc957fa5c24978601a84b9a36a8ebd8f5a67eafb..0be77c5bcefa9a6c69f31d5e853b4063170f6250 100644 (file)
 #define DORQ_REG_SHRT_CMHEAD                                    0x170054
 #define HC_CONFIG_0_REG_ATTN_BIT_EN_0                           (0x1<<4)
 #define HC_CONFIG_0_REG_INT_LINE_EN_0                           (0x1<<3)
+#define HC_CONFIG_0_REG_MSI_ATTN_EN_0                           (0x1<<7)
 #define HC_CONFIG_0_REG_MSI_MSIX_INT_EN_0                       (0x1<<2)
 #define HC_CONFIG_0_REG_SINGLE_ISR_EN_0                         (0x1<<1)
 #define HC_REG_AGG_INT_0                                        0x108050
 #define PCICFG_PM_CSR_STATE                    (0x3<<0)
 #define PCICFG_PM_CSR_PME_ENABLE               (1<<8)
 #define PCICFG_PM_CSR_PME_STATUS               (1<<15)
+#define PCICFG_MSI_CAP_ID                              0x58
+#define PCICFG_MSI_CONTROL_ENABLE              (0x1<<16)
+#define PCICFG_MSI_CONTROL_MCAP                (0x7<<17)
+#define PCICFG_MSI_CONTROL_MENA                (0x7<<20)
+#define PCICFG_MSI_CONTROL_64_BIT_ADDR_CAP     (0x1<<23)
+#define PCICFG_MSI_CONTROL_MSI_PVMASK_CAPABLE  (0x1<<24)
 #define PCICFG_GRC_ADDRESS                             0x78
 #define PCICFG_GRC_DATA                                0x80
+#define PCICFG_MSIX_CAP_ID                             0xa0
+#define PCICFG_MSIX_CONTROL_TABLE_SIZE         (0x7ff<<16)
+#define PCICFG_MSIX_CONTROL_RESERVED           (0x7<<27)
+#define PCICFG_MSIX_CONTROL_FUNC_MASK          (0x1<<30)
+#define PCICFG_MSIX_CONTROL_MSIX_ENABLE        (0x1<<31)
+
 #define PCICFG_DEVICE_CONTROL                          0xb4
+#define PCICFG_DEVICE_STATUS                           0xb6
+#define PCICFG_DEVICE_STATUS_CORR_ERR_DET      (1<<0)
+#define PCICFG_DEVICE_STATUS_NON_FATAL_ERR_DET (1<<1)
+#define PCICFG_DEVICE_STATUS_FATAL_ERR_DET     (1<<2)
+#define PCICFG_DEVICE_STATUS_UNSUP_REQ_DET     (1<<3)
+#define PCICFG_DEVICE_STATUS_AUX_PWR_DET       (1<<4)
+#define PCICFG_DEVICE_STATUS_NO_PEND           (1<<5)
 #define PCICFG_LINK_CONTROL                            0xbc