Merge to Fedora kernel-2.6.18-1.2224_FC5 patched with stable patch-2.6.18.1-vs2.0...
[linux-2.6.git] / drivers / infiniband / hw / ipath / ipath_verbs.c
index 28fdbda..d70a9b6 100644 (file)
@@ -1,4 +1,5 @@
 /*
+ * Copyright (c) 2006 QLogic, Inc. All rights reserved.
  * Copyright (c) 2005, 2006 PathScale, Inc. All rights reserved.
  *
  * This software is available to you under a choice of one of two
@@ -36,7 +37,7 @@
 
 #include "ipath_kernel.h"
 #include "ipath_verbs.h"
-#include "ips_common.h"
+#include "ipath_common.h"
 
 /* Not static, because we don't want the compiler removing it */
 const char ipath_verbs_version[] = "ipath_verbs " IPATH_IDSTR;
@@ -55,9 +56,62 @@ unsigned int ib_ipath_debug; /* debug mask */
 module_param_named(debug, ib_ipath_debug, uint, S_IWUSR | S_IRUGO);
 MODULE_PARM_DESC(debug, "Verbs debug mask");
 
+static unsigned int ib_ipath_max_pds = 0xFFFF;
+module_param_named(max_pds, ib_ipath_max_pds, uint, S_IWUSR | S_IRUGO);
+MODULE_PARM_DESC(max_pds,
+                "Maximum number of protection domains to support");
+
+static unsigned int ib_ipath_max_ahs = 0xFFFF;
+module_param_named(max_ahs, ib_ipath_max_ahs, uint, S_IWUSR | S_IRUGO);
+MODULE_PARM_DESC(max_ahs, "Maximum number of address handles to support");
+
+unsigned int ib_ipath_max_cqes = 0x2FFFF;
+module_param_named(max_cqes, ib_ipath_max_cqes, uint, S_IWUSR | S_IRUGO);
+MODULE_PARM_DESC(max_cqes,
+                "Maximum number of completion queue entries to support");
+
+unsigned int ib_ipath_max_cqs = 0x1FFFF;
+module_param_named(max_cqs, ib_ipath_max_cqs, uint, S_IWUSR | S_IRUGO);
+MODULE_PARM_DESC(max_cqs, "Maximum number of completion queues to support");
+
+unsigned int ib_ipath_max_qp_wrs = 0x3FFF;
+module_param_named(max_qp_wrs, ib_ipath_max_qp_wrs, uint,
+                  S_IWUSR | S_IRUGO);
+MODULE_PARM_DESC(max_qp_wrs, "Maximum number of QP WRs to support");
+
+unsigned int ib_ipath_max_sges = 0x60;
+module_param_named(max_sges, ib_ipath_max_sges, uint, S_IWUSR | S_IRUGO);
+MODULE_PARM_DESC(max_sges, "Maximum number of SGEs to support");
+
+unsigned int ib_ipath_max_mcast_grps = 16384;
+module_param_named(max_mcast_grps, ib_ipath_max_mcast_grps, uint,
+                  S_IWUSR | S_IRUGO);
+MODULE_PARM_DESC(max_mcast_grps,
+                "Maximum number of multicast groups to support");
+
+unsigned int ib_ipath_max_mcast_qp_attached = 16;
+module_param_named(max_mcast_qp_attached, ib_ipath_max_mcast_qp_attached,
+                  uint, S_IWUSR | S_IRUGO);
+MODULE_PARM_DESC(max_mcast_qp_attached,
+                "Maximum number of attached QPs to support");
+
+unsigned int ib_ipath_max_srqs = 1024;
+module_param_named(max_srqs, ib_ipath_max_srqs, uint, S_IWUSR | S_IRUGO);
+MODULE_PARM_DESC(max_srqs, "Maximum number of SRQs to support");
+
+unsigned int ib_ipath_max_srq_sges = 128;
+module_param_named(max_srq_sges, ib_ipath_max_srq_sges,
+                  uint, S_IWUSR | S_IRUGO);
+MODULE_PARM_DESC(max_srq_sges, "Maximum number of SRQ SGEs to support");
+
+unsigned int ib_ipath_max_srq_wrs = 0x1FFFF;
+module_param_named(max_srq_wrs, ib_ipath_max_srq_wrs,
+                  uint, S_IWUSR | S_IRUGO);
+MODULE_PARM_DESC(max_srq_wrs, "Maximum number of SRQ WRs support");
+
 MODULE_LICENSE("GPL");
-MODULE_AUTHOR("PathScale <support@pathscale.com>");
-MODULE_DESCRIPTION("Pathscale InfiniPath driver");
+MODULE_AUTHOR("QLogic <support@pathscale.com>");
+MODULE_DESCRIPTION("QLogic InfiniPath driver");
 
 const int ib_ipath_state_ops[IB_QPS_ERR + 1] = {
        [IB_QPS_RESET] = 0,
@@ -137,10 +191,6 @@ void ipath_skip_sge(struct ipath_sge_state *ss, u32 length)
 {
        struct ipath_sge *sge = &ss->sge;
 
-       while (length > sge->sge_length) {
-               length -= sge->sge_length;
-               ss->sge = *ss->sg_list++;
-       }
        while (length) {
                u32 len = sge->length;
 
@@ -193,7 +243,7 @@ static int ipath_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
                switch (qp->ibqp.qp_type) {
                case IB_QPT_UC:
                case IB_QPT_RC:
-                       err = ipath_post_rc_send(qp, wr);
+                       err = ipath_post_ruc_send(qp, wr);
                        break;
 
                case IB_QPT_SMI:
@@ -375,7 +425,7 @@ static void ipath_ib_rcv(void *arg, void *rhdr, void *data, u32 tlen)
 
        /* Check for a valid destination LID (see ch. 7.11.1). */
        lid = be16_to_cpu(hdr->lrh[1]);
-       if (lid < IPS_MULTICAST_LID_BASE) {
+       if (lid < IPATH_MULTICAST_LID_BASE) {
                lid &= ~((1 << (dev->mkeyprot_resv_lmc & 7)) - 1);
                if (unlikely(lid != ipath_layer_get_lid(dev->dd))) {
                        dev->rcv_errors++;
@@ -385,9 +435,9 @@ static void ipath_ib_rcv(void *arg, void *rhdr, void *data, u32 tlen)
 
        /* Check for GRH */
        lnh = be16_to_cpu(hdr->lrh[0]) & 3;
-       if (lnh == IPS_LRH_BTH)
+       if (lnh == IPATH_LRH_BTH)
                ohdr = &hdr->u.oth;
-       else if (lnh == IPS_LRH_GRH)
+       else if (lnh == IPATH_LRH_GRH)
                ohdr = &hdr->u.l.oth;
        else {
                dev->rcv_errors++;
@@ -399,8 +449,8 @@ static void ipath_ib_rcv(void *arg, void *rhdr, void *data, u32 tlen)
        dev->opstats[opcode].n_packets++;
 
        /* Get the destination QP number. */
-       qp_num = be32_to_cpu(ohdr->bth[1]) & IPS_QPN_MASK;
-       if (qp_num == IPS_MULTICAST_QPN) {
+       qp_num = be32_to_cpu(ohdr->bth[1]) & IPATH_QPN_MASK;
+       if (qp_num == IPATH_MULTICAST_QPN) {
                struct ipath_mcast *mcast;
                struct ipath_mcast_qp *p;
 
@@ -411,7 +461,7 @@ static void ipath_ib_rcv(void *arg, void *rhdr, void *data, u32 tlen)
                }
                dev->n_multicast_rcv++;
                list_for_each_entry_rcu(p, &mcast->qp_list, list)
-                       ipath_qp_rcv(dev, hdr, lnh == IPS_LRH_GRH, data,
+                       ipath_qp_rcv(dev, hdr, lnh == IPATH_LRH_GRH, data,
                                     tlen, p->qp);
                /*
                 * Notify ipath_multicast_detach() if it is waiting for us
@@ -423,7 +473,7 @@ static void ipath_ib_rcv(void *arg, void *rhdr, void *data, u32 tlen)
                qp = ipath_lookup_qpn(&dev->qp_table, qp_num);
                if (qp) {
                        dev->n_unicast_rcv++;
-                       ipath_qp_rcv(dev, hdr, lnh == IPS_LRH_GRH, data,
+                       ipath_qp_rcv(dev, hdr, lnh == IPATH_LRH_GRH, data,
                                     tlen, qp);
                        /*
                         * Notify ipath_destroy_qp() if it is waiting
@@ -567,40 +617,39 @@ static int ipath_query_device(struct ib_device *ibdev,
                              struct ib_device_attr *props)
 {
        struct ipath_ibdev *dev = to_idev(ibdev);
-       u32 vendor, boardrev, majrev, minrev;
 
        memset(props, 0, sizeof(*props));
 
        props->device_cap_flags = IB_DEVICE_BAD_PKEY_CNTR |
                IB_DEVICE_BAD_QKEY_CNTR | IB_DEVICE_SHUTDOWN_PORT |
                IB_DEVICE_SYS_IMAGE_GUID;
-       ipath_layer_query_device(dev->dd, &vendor, &boardrev,
-                                &majrev, &minrev);
-       props->vendor_id = vendor;
-       props->vendor_part_id = boardrev;
-       props->hw_ver = boardrev << 16 | majrev << 8 | minrev;
+       props->page_size_cap = PAGE_SIZE;
+       props->vendor_id = ipath_layer_get_vendorid(dev->dd);
+       props->vendor_part_id = ipath_layer_get_deviceid(dev->dd);
+       props->hw_ver = ipath_layer_get_pcirev(dev->dd);
 
        props->sys_image_guid = dev->sys_image_guid;
 
        props->max_mr_size = ~0ull;
-       props->max_qp = 0xffff;
-       props->max_qp_wr = 0xffff;
-       props->max_sge = 255;
-       props->max_cq = 0xffff;
-       props->max_cqe = 0xffff;
-       props->max_mr = 0xffff;
-       props->max_pd = 0xffff;
+       props->max_qp = dev->qp_table.max;
+       props->max_qp_wr = ib_ipath_max_qp_wrs;
+       props->max_sge = ib_ipath_max_sges;
+       props->max_cq = ib_ipath_max_cqs;
+       props->max_ah = ib_ipath_max_ahs;
+       props->max_cqe = ib_ipath_max_cqes;
+       props->max_mr = dev->lk_table.max;
+       props->max_pd = ib_ipath_max_pds;
        props->max_qp_rd_atom = 1;
        props->max_qp_init_rd_atom = 1;
        /* props->max_res_rd_atom */
-       props->max_srq = 0xffff;
-       props->max_srq_wr = 0xffff;
-       props->max_srq_sge = 255;
+       props->max_srq = ib_ipath_max_srqs;
+       props->max_srq_wr = ib_ipath_max_srq_wrs;
+       props->max_srq_sge = ib_ipath_max_srq_sges;
        /* props->local_ca_ack_delay */
        props->atomic_cap = IB_ATOMIC_HCA;
        props->max_pkeys = ipath_layer_get_npkeys(dev->dd);
-       props->max_mcast_grp = 0xffff;
-       props->max_mcast_qp_attach = 0xffff;
+       props->max_mcast_grp = ib_ipath_max_mcast_grps;
+       props->max_mcast_qp_attach = ib_ipath_max_mcast_qp_attached;
        props->max_total_mcast_qp_attach = props->max_mcast_qp_attach *
                props->max_mcast_grp;
 
@@ -643,10 +692,10 @@ static int ipath_query_port(struct ib_device *ibdev,
                ipath_layer_get_lastibcstat(dev->dd) & 0xf];
        props->port_cap_flags = dev->port_cap_flags;
        props->gid_tbl_len = 1;
-       props->max_msg_sz = 4096;
+       props->max_msg_sz = 0x80000000;
        props->pkey_tbl_len = ipath_layer_get_npkeys(dev->dd);
        props->bad_pkey_cntr = ipath_layer_get_cr_errpkey(dev->dd) -
-               dev->n_pkey_violations;
+               dev->z_pkey_violations;
        props->qkey_viol_cntr = dev->qkey_violations;
        props->active_width = IB_WIDTH_4X;
        /* See rate_show() */
@@ -743,15 +792,30 @@ static struct ib_pd *ipath_alloc_pd(struct ib_device *ibdev,
                                    struct ib_ucontext *context,
                                    struct ib_udata *udata)
 {
+       struct ipath_ibdev *dev = to_idev(ibdev);
        struct ipath_pd *pd;
        struct ib_pd *ret;
 
+       /*
+        * This is actually totally arbitrary.  Some correctness tests
+        * assume there's a maximum number of PDs that can be allocated.
+        * We don't actually have this limit, but we fail the test if
+        * we allow allocations of more than we report for this value.
+        */
+
+       if (dev->n_pds_allocated == ib_ipath_max_pds) {
+               ret = ERR_PTR(-ENOMEM);
+               goto bail;
+       }
+
        pd = kmalloc(sizeof *pd, GFP_KERNEL);
        if (!pd) {
                ret = ERR_PTR(-ENOMEM);
                goto bail;
        }
 
+       dev->n_pds_allocated++;
+
        /* ib_alloc_pd() will initialize pd->ibpd. */
        pd->user = udata != NULL;
 
@@ -764,6 +828,9 @@ bail:
 static int ipath_dealloc_pd(struct ib_pd *ibpd)
 {
        struct ipath_pd *pd = to_ipd(ibpd);
+       struct ipath_ibdev *dev = to_idev(ibpd->device);
+
+       dev->n_pds_allocated--;
 
        kfree(pd);
 
@@ -782,21 +849,40 @@ static struct ib_ah *ipath_create_ah(struct ib_pd *pd,
 {
        struct ipath_ah *ah;
        struct ib_ah *ret;
+       struct ipath_ibdev *dev = to_idev(pd->device);
+
+       if (dev->n_ahs_allocated == ib_ipath_max_ahs) {
+               ret = ERR_PTR(-ENOMEM);
+               goto bail;
+       }
 
        /* A multicast address requires a GRH (see ch. 8.4.1). */
-       if (ah_attr->dlid >= IPS_MULTICAST_LID_BASE &&
-           ah_attr->dlid != IPS_PERMISSIVE_LID &&
+       if (ah_attr->dlid >= IPATH_MULTICAST_LID_BASE &&
+           ah_attr->dlid != IPATH_PERMISSIVE_LID &&
            !(ah_attr->ah_flags & IB_AH_GRH)) {
                ret = ERR_PTR(-EINVAL);
                goto bail;
        }
 
+       if (ah_attr->dlid == 0) {
+               ret = ERR_PTR(-EINVAL);
+               goto bail;
+       }
+
+       if (ah_attr->port_num < 1 ||
+           ah_attr->port_num > pd->device->phys_port_cnt) {
+               ret = ERR_PTR(-EINVAL);
+               goto bail;
+       }
+
        ah = kmalloc(sizeof *ah, GFP_ATOMIC);
        if (!ah) {
                ret = ERR_PTR(-ENOMEM);
                goto bail;
        }
 
+       dev->n_ahs_allocated++;
+
        /* ib_create_ah() will initialize ah->ibah. */
        ah->attr = *ah_attr;
 
@@ -814,8 +900,11 @@ bail:
  */
 static int ipath_destroy_ah(struct ib_ah *ibah)
 {
+       struct ipath_ibdev *dev = to_idev(ibah->device);
        struct ipath_ah *ah = to_iah(ibah);
 
+       dev->n_ahs_allocated--;
+
        kfree(ah);
 
        return 0;
@@ -889,6 +978,7 @@ static int ipath_verbs_register_sysfs(struct ib_device *dev);
  */
 static void *ipath_register_ib_device(int unit, struct ipath_devdata *dd)
 {
+       struct ipath_layer_counters cntrs;
        struct ipath_ibdev *idev;
        struct ib_device *dev;
        int ret;
@@ -939,6 +1029,25 @@ static void *ipath_register_ib_device(int unit, struct ipath_devdata *dd)
        idev->pma_counter_select[5] = IB_PMA_PORT_XMIT_WAIT;
        idev->link_width_enabled = 3;   /* 1x or 4x */
 
+       /* Snapshot current HW counters to "clear" them. */
+       ipath_layer_get_counters(dd, &cntrs);
+       idev->z_symbol_error_counter = cntrs.symbol_error_counter;
+       idev->z_link_error_recovery_counter =
+               cntrs.link_error_recovery_counter;
+       idev->z_link_downed_counter = cntrs.link_downed_counter;
+       idev->z_port_rcv_errors = cntrs.port_rcv_errors;
+       idev->z_port_rcv_remphys_errors =
+               cntrs.port_rcv_remphys_errors;
+       idev->z_port_xmit_discards = cntrs.port_xmit_discards;
+       idev->z_port_xmit_data = cntrs.port_xmit_data;
+       idev->z_port_rcv_data = cntrs.port_rcv_data;
+       idev->z_port_xmit_packets = cntrs.port_xmit_packets;
+       idev->z_port_rcv_packets = cntrs.port_rcv_packets;
+       idev->z_local_link_integrity_errors =
+               cntrs.local_link_integrity_errors;
+       idev->z_excessive_buffer_overrun_errors =
+               cntrs.excessive_buffer_overrun_errors;
+
        /*
         * The system image GUID is supposed to be the same for all
         * IB HCAs in a single system but since there can be other
@@ -1109,11 +1218,8 @@ static ssize_t show_rev(struct class_device *cdev, char *buf)
 {
        struct ipath_ibdev *dev =
                container_of(cdev, struct ipath_ibdev, ibdev.class_dev);
-       int vendor, boardrev, majrev, minrev;
 
-       ipath_layer_query_device(dev->dd, &vendor, &boardrev,
-                                &majrev, &minrev);
-       return sprintf(buf, "%d.%d\n", majrev, minrev);
+       return sprintf(buf, "%x\n", ipath_layer_get_pcirev(dev->dd));
 }
 
 static ssize_t show_hca(struct class_device *cdev, char *buf)