staging/rdma/hfi1: Adding support for hfi counters via sysfs
authorSebastian Sanchez <sebastian.sanchez@intel.com>
Wed, 3 Feb 2016 22:38:07 +0000 (14:38 -0800)
committerDoug Ledford <dledford@redhat.com>
Fri, 11 Mar 2016 01:38:01 +0000 (20:38 -0500)
It enables access to counters in
/sys/class/infiniband/hfi1_0/ports/1/counters
by providing infrastructure when PMA queries occur. Counters symbol_error
and VL15_dropped are not supported in OPA, therefore, 0 will always be
returned. In addition, two common routines (pma_get_opa_port_dctrs,
pma_get_opa_port_ectrs) were created to query counters to avoid code
duplication.

Reviewed-by: Mike Marciniszyn <mike.marciniszyn@intel.com>
Signed-off-by: Sebastian Sanchez <sebastian.sanchez@intel.com>
Signed-off-by: Doug Ledford <dledford@redhat.com>
drivers/staging/rdma/hfi1/mad.c

index a31557911f54b36e6a56248632c70246be9132b8..2fcc9f3290d7a712e028222ccbbc50eaf301d28c 100644 (file)
@@ -2524,6 +2524,27 @@ static void a0_datacounters(struct hfi1_pportdata *ppd, struct _port_dctrs *rsp,
        }
 }
 
+static void pma_get_opa_port_dctrs(struct ib_device *ibdev,
+                                  struct _port_dctrs *rsp)
+{
+       struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
+
+       rsp->port_xmit_data = cpu_to_be64(read_dev_cntr(dd, C_DC_XMIT_FLITS,
+                                               CNTR_INVALID_VL));
+       rsp->port_rcv_data = cpu_to_be64(read_dev_cntr(dd, C_DC_RCV_FLITS,
+                                               CNTR_INVALID_VL));
+       rsp->port_xmit_pkts = cpu_to_be64(read_dev_cntr(dd, C_DC_XMIT_PKTS,
+                                               CNTR_INVALID_VL));
+       rsp->port_rcv_pkts = cpu_to_be64(read_dev_cntr(dd, C_DC_RCV_PKTS,
+                                               CNTR_INVALID_VL));
+       rsp->port_multicast_xmit_pkts =
+               cpu_to_be64(read_dev_cntr(dd, C_DC_MC_XMIT_PKTS,
+                                         CNTR_INVALID_VL));
+       rsp->port_multicast_rcv_pkts =
+               cpu_to_be64(read_dev_cntr(dd, C_DC_MC_RCV_PKTS,
+                                         CNTR_INVALID_VL));
+}
+
 static int pma_get_opa_datacounters(struct opa_pma_mad *pmp,
                        struct ib_device *ibdev, u8 port, u32 *resp_len)
 {
@@ -2592,34 +2613,14 @@ static int pma_get_opa_datacounters(struct opa_pma_mad *pmp,
         */
        hfi1_read_link_quality(dd, &lq);
        rsp->link_quality_indicator = cpu_to_be32((u32)lq);
+       pma_get_opa_port_dctrs(ibdev, rsp);
 
-       /* rsp->sw_port_congestion is 0 for HFIs */
-       /* rsp->port_xmit_time_cong is 0 for HFIs */
-       /* rsp->port_xmit_wasted_bw ??? */
-       /* rsp->port_xmit_wait_data ??? */
-       /* rsp->port_mark_fecn is 0 for HFIs */
-
-       rsp->port_xmit_data = cpu_to_be64(read_dev_cntr(dd, C_DC_XMIT_FLITS,
-                                               CNTR_INVALID_VL));
-       rsp->port_rcv_data = cpu_to_be64(read_dev_cntr(dd, C_DC_RCV_FLITS,
-                                               CNTR_INVALID_VL));
-       rsp->port_xmit_pkts = cpu_to_be64(read_dev_cntr(dd, C_DC_XMIT_PKTS,
-                                               CNTR_INVALID_VL));
-       rsp->port_rcv_pkts = cpu_to_be64(read_dev_cntr(dd, C_DC_RCV_PKTS,
-                                               CNTR_INVALID_VL));
-       rsp->port_multicast_xmit_pkts =
-               cpu_to_be64(read_dev_cntr(dd, C_DC_MC_XMIT_PKTS,
-                                               CNTR_INVALID_VL));
-       rsp->port_multicast_rcv_pkts =
-               cpu_to_be64(read_dev_cntr(dd, C_DC_MC_RCV_PKTS,
-                                               CNTR_INVALID_VL));
        rsp->port_xmit_wait =
                cpu_to_be64(read_port_cntr(ppd, C_TX_WAIT, CNTR_INVALID_VL));
        rsp->port_rcv_fecn =
                cpu_to_be64(read_dev_cntr(dd, C_DC_RCV_FCN, CNTR_INVALID_VL));
        rsp->port_rcv_becn =
                cpu_to_be64(read_dev_cntr(dd, C_DC_RCV_BCN, CNTR_INVALID_VL));
-
        rsp->port_error_counter_summary =
                cpu_to_be64(get_error_counter_summary(ibdev, port,
                                                      res_lli, res_ler));
@@ -2682,6 +2683,81 @@ static int pma_get_opa_datacounters(struct opa_pma_mad *pmp,
        return reply((struct ib_mad_hdr *)pmp);
 }
 
+static int pma_get_ib_portcounters_ext(struct ib_pma_mad *pmp,
+                                      struct ib_device *ibdev, u8 port)
+{
+       struct ib_pma_portcounters_ext *p = (struct ib_pma_portcounters_ext *)
+                                               pmp->data;
+       struct _port_dctrs rsp;
+
+       if (pmp->mad_hdr.attr_mod != 0 || p->port_select != port) {
+               pmp->mad_hdr.status |= IB_SMP_INVALID_FIELD;
+               goto bail;
+       }
+
+       memset(&rsp, 0, sizeof(rsp));
+       pma_get_opa_port_dctrs(ibdev, &rsp);
+
+       p->port_xmit_data = rsp.port_xmit_data;
+       p->port_rcv_data = rsp.port_rcv_data;
+       p->port_xmit_packets = rsp.port_xmit_pkts;
+       p->port_rcv_packets = rsp.port_rcv_pkts;
+       p->port_unicast_xmit_packets = 0;
+       p->port_unicast_rcv_packets =  0;
+       p->port_multicast_xmit_packets = rsp.port_multicast_xmit_pkts;
+       p->port_multicast_rcv_packets = rsp.port_multicast_rcv_pkts;
+
+bail:
+       return reply((struct ib_mad_hdr *)pmp);
+}
+
+static void pma_get_opa_port_ectrs(struct ib_device *ibdev,
+                                  struct _port_ectrs *rsp, u8 port)
+{
+       u64 tmp, tmp2;
+       struct hfi1_devdata *dd = dd_from_ibdev(ibdev);
+       struct hfi1_ibport *ibp = to_iport(ibdev, port);
+       struct hfi1_pportdata *ppd = ppd_from_ibp(ibp);
+
+       tmp = read_dev_cntr(dd, C_DC_SEQ_CRC_CNT, CNTR_INVALID_VL);
+       tmp2 = tmp + read_dev_cntr(dd, C_DC_REINIT_FROM_PEER_CNT,
+                                       CNTR_INVALID_VL);
+       if (tmp2 > (u32)UINT_MAX || tmp2 < tmp) {
+               /* overflow/wrapped */
+               rsp->link_error_recovery = cpu_to_be32(~0);
+       } else {
+               rsp->link_error_recovery = cpu_to_be32(tmp2);
+       }
+
+       rsp->link_downed = cpu_to_be32(read_port_cntr(ppd, C_SW_LINK_DOWN,
+                                               CNTR_INVALID_VL));
+       rsp->port_rcv_errors =
+               cpu_to_be64(read_dev_cntr(dd, C_DC_RCV_ERR, CNTR_INVALID_VL));
+       rsp->port_rcv_remote_physical_errors =
+               cpu_to_be64(read_dev_cntr(dd, C_DC_RMT_PHY_ERR,
+                                         CNTR_INVALID_VL));
+       rsp->port_rcv_switch_relay_errors = 0;
+       rsp->port_xmit_discards =
+               cpu_to_be64(read_port_cntr(ppd, C_SW_XMIT_DSCD,
+                                          CNTR_INVALID_VL));
+       rsp->port_xmit_constraint_errors =
+               cpu_to_be64(read_port_cntr(ppd, C_SW_XMIT_CSTR_ERR,
+                                          CNTR_INVALID_VL));
+       rsp->port_rcv_constraint_errors =
+               cpu_to_be64(read_port_cntr(ppd, C_SW_RCV_CSTR_ERR,
+                                          CNTR_INVALID_VL));
+       tmp = read_dev_cntr(dd, C_DC_RX_REPLAY, CNTR_INVALID_VL);
+       tmp2 = tmp + read_dev_cntr(dd, C_DC_TX_REPLAY, CNTR_INVALID_VL);
+       if (tmp2 < tmp) {
+               /* overflow/wrapped */
+               rsp->local_link_integrity_errors = cpu_to_be64(~0);
+       } else {
+               rsp->local_link_integrity_errors = cpu_to_be64(tmp2);
+       }
+       rsp->excessive_buffer_overruns =
+               cpu_to_be64(read_dev_cntr(dd, C_RCV_OVF, CNTR_INVALID_VL));
+}
+
 static int pma_get_opa_porterrors(struct opa_pma_mad *pmp,
                        struct ib_device *ibdev, u8 port, u32 *resp_len)
 {
@@ -2697,7 +2773,7 @@ static int pma_get_opa_porterrors(struct opa_pma_mad *pmp,
        struct hfi1_pportdata *ppd;
        struct _vls_ectrs *vlinfo;
        unsigned long vl;
-       u64 port_mask, tmp, tmp2;
+       u64 port_mask, tmp;
        u32 vl_select_mask;
        int vfi;
 
@@ -2741,44 +2817,16 @@ static int pma_get_opa_porterrors(struct opa_pma_mad *pmp,
        memset(rsp, 0, sizeof(*rsp));
        rsp->port_number = port_num;
 
-       rsp->port_rcv_constraint_errors =
-               cpu_to_be64(read_port_cntr(ppd, C_SW_RCV_CSTR_ERR,
-                                          CNTR_INVALID_VL));
-       /* port_rcv_switch_relay_errors is 0 for HFIs */
-       rsp->port_xmit_discards =
-               cpu_to_be64(read_port_cntr(ppd, C_SW_XMIT_DSCD,
-                                               CNTR_INVALID_VL));
+       pma_get_opa_port_ectrs(ibdev, rsp, port_num);
+
        rsp->port_rcv_remote_physical_errors =
                cpu_to_be64(read_dev_cntr(dd, C_DC_RMT_PHY_ERR,
-                                               CNTR_INVALID_VL));
-       tmp = read_dev_cntr(dd, C_DC_RX_REPLAY, CNTR_INVALID_VL);
-       tmp2 = tmp + read_dev_cntr(dd, C_DC_TX_REPLAY, CNTR_INVALID_VL);
-       if (tmp2 < tmp) {
-               /* overflow/wrapped */
-               rsp->local_link_integrity_errors = cpu_to_be64(~0);
-       } else {
-               rsp->local_link_integrity_errors = cpu_to_be64(tmp2);
-       }
-       tmp = read_dev_cntr(dd, C_DC_SEQ_CRC_CNT, CNTR_INVALID_VL);
-       tmp2 = tmp + read_dev_cntr(dd, C_DC_REINIT_FROM_PEER_CNT,
-                                       CNTR_INVALID_VL);
-       if (tmp2 > (u32)UINT_MAX || tmp2 < tmp) {
-               /* overflow/wrapped */
-               rsp->link_error_recovery = cpu_to_be32(~0);
-       } else {
-               rsp->link_error_recovery = cpu_to_be32(tmp2);
-       }
-       rsp->port_xmit_constraint_errors =
-               cpu_to_be64(read_port_cntr(ppd, C_SW_XMIT_CSTR_ERR,
-                                          CNTR_INVALID_VL));
-       rsp->excessive_buffer_overruns =
-               cpu_to_be64(read_dev_cntr(dd, C_RCV_OVF, CNTR_INVALID_VL));
+                                         CNTR_INVALID_VL));
        rsp->fm_config_errors =
                cpu_to_be64(read_dev_cntr(dd, C_DC_FM_CFG_ERR,
                                                CNTR_INVALID_VL));
-       rsp->link_downed = cpu_to_be32(read_port_cntr(ppd, C_SW_LINK_DOWN,
-                                               CNTR_INVALID_VL));
        tmp = read_dev_cntr(dd, C_DC_UNC_ERR, CNTR_INVALID_VL);
+
        rsp->uncorrectable_errors = tmp < 0x100 ? (tmp & 0xff) : 0xff;
 
        vlinfo = (struct _vls_ectrs *)&(rsp->vls[0]);
@@ -2798,6 +2846,91 @@ static int pma_get_opa_porterrors(struct opa_pma_mad *pmp,
        return reply((struct ib_mad_hdr *)pmp);
 }
 
+static int pma_get_ib_portcounters(struct ib_pma_mad *pmp,
+                                  struct ib_device *ibdev, u8 port)
+{
+       struct ib_pma_portcounters *p = (struct ib_pma_portcounters *)
+               pmp->data;
+       struct _port_ectrs rsp;
+       u64 temp_link_overrun_errors;
+       u64 temp_64;
+       u32 temp_32;
+
+       memset(&rsp, 0, sizeof(rsp));
+       pma_get_opa_port_ectrs(ibdev, &rsp, port);
+
+       if (pmp->mad_hdr.attr_mod != 0 || p->port_select != port) {
+               pmp->mad_hdr.status |= IB_SMP_INVALID_FIELD;
+               goto bail;
+       }
+
+       p->symbol_error_counter = 0; /* N/A for OPA */
+
+       temp_32 = be32_to_cpu(rsp.link_error_recovery);
+       if (temp_32 > 0xFFUL)
+               p->link_error_recovery_counter = 0xFF;
+       else
+               p->link_error_recovery_counter = (u8)temp_32;
+
+       temp_32 = be32_to_cpu(rsp.link_downed);
+       if (temp_32 > 0xFFUL)
+               p->link_downed_counter = 0xFF;
+       else
+               p->link_downed_counter = (u8)temp_32;
+
+       temp_64 = be64_to_cpu(rsp.port_rcv_errors);
+       if (temp_64 > 0xFFFFUL)
+               p->port_rcv_errors = cpu_to_be16(0xFFFF);
+       else
+               p->port_rcv_errors = cpu_to_be16((u16)temp_64);
+
+       temp_64 = be64_to_cpu(rsp.port_rcv_remote_physical_errors);
+       if (temp_64 > 0xFFFFUL)
+               p->port_rcv_remphys_errors = cpu_to_be16(0xFFFF);
+       else
+               p->port_rcv_remphys_errors = cpu_to_be16((u16)temp_64);
+
+       temp_64 = be64_to_cpu(rsp.port_rcv_switch_relay_errors);
+       p->port_rcv_switch_relay_errors = cpu_to_be16((u16)temp_64);
+
+       temp_64 = be64_to_cpu(rsp.port_xmit_discards);
+       if (temp_64 > 0xFFFFUL)
+               p->port_xmit_discards = cpu_to_be16(0xFFFF);
+       else
+               p->port_xmit_discards = cpu_to_be16((u16)temp_64);
+
+       temp_64 = be64_to_cpu(rsp.port_xmit_constraint_errors);
+       if (temp_64 > 0xFFUL)
+               p->port_xmit_constraint_errors = 0xFF;
+       else
+               p->port_xmit_constraint_errors = (u8)temp_64;
+
+       temp_64 = be64_to_cpu(rsp.port_rcv_constraint_errors);
+       if (temp_64 > 0xFFUL)
+               p->port_rcv_constraint_errors = 0xFFUL;
+       else
+               p->port_rcv_constraint_errors = (u8)temp_64;
+
+       /* LocalLink: 7:4, BufferOverrun: 3:0 */
+       temp_64 = be64_to_cpu(rsp.local_link_integrity_errors);
+       if (temp_64 > 0xFUL)
+               temp_64 = 0xFUL;
+
+       temp_link_overrun_errors = temp_64 << 4;
+
+       temp_64 = be64_to_cpu(rsp.excessive_buffer_overruns);
+       if (temp_64 > 0xFUL)
+               temp_64 = 0xFUL;
+       temp_link_overrun_errors |= temp_64;
+
+       p->link_overrun_errors = (u8)temp_link_overrun_errors;
+
+       p->vl15_dropped = 0; /* N/A for OPA */
+
+bail:
+       return reply((struct ib_mad_hdr *)pmp);
+}
+
 static int pma_get_opa_errorinfo(struct opa_pma_mad *pmp,
                        struct ib_device *ibdev, u8 port, u32 *resp_len)
 {
@@ -3964,6 +4097,68 @@ static int process_subn(struct ib_device *ibdev, int mad_flags,
        return ret;
 }
 
+static int process_perf(struct ib_device *ibdev, u8 port,
+                       const struct ib_mad *in_mad,
+                       struct ib_mad *out_mad)
+{
+       struct ib_pma_mad *pmp = (struct ib_pma_mad *)out_mad;
+       struct ib_class_port_info *cpi = (struct ib_class_port_info *)
+                                               &pmp->data;
+       int ret = IB_MAD_RESULT_FAILURE;
+
+       *out_mad = *in_mad;
+       if (pmp->mad_hdr.class_version != 1) {
+               pmp->mad_hdr.status |= IB_SMP_UNSUP_VERSION;
+               ret = reply((struct ib_mad_hdr *)pmp);
+               return ret;
+       }
+
+       switch (pmp->mad_hdr.method) {
+       case IB_MGMT_METHOD_GET:
+               switch (pmp->mad_hdr.attr_id) {
+               case IB_PMA_PORT_COUNTERS:
+                       ret = pma_get_ib_portcounters(pmp, ibdev, port);
+                       break;
+               case IB_PMA_PORT_COUNTERS_EXT:
+                       ret = pma_get_ib_portcounters_ext(pmp, ibdev, port);
+                       break;
+               case IB_PMA_CLASS_PORT_INFO:
+                       cpi->capability_mask = IB_PMA_CLASS_CAP_EXT_WIDTH;
+                       ret = reply((struct ib_mad_hdr *)pmp);
+                       break;
+               default:
+                       pmp->mad_hdr.status |= IB_SMP_UNSUP_METH_ATTR;
+                       ret = reply((struct ib_mad_hdr *)pmp);
+                       break;
+               }
+               break;
+
+       case IB_MGMT_METHOD_SET:
+               if (pmp->mad_hdr.attr_id) {
+                       pmp->mad_hdr.status |= IB_SMP_UNSUP_METH_ATTR;
+                       ret = reply((struct ib_mad_hdr *)pmp);
+               }
+               break;
+
+       case IB_MGMT_METHOD_TRAP:
+       case IB_MGMT_METHOD_GET_RESP:
+               /*
+                * The ib_mad module will call us to process responses
+                * before checking for other consumers.
+                * Just tell the caller to process it normally.
+                */
+               ret = IB_MAD_RESULT_SUCCESS;
+               break;
+
+       default:
+               pmp->mad_hdr.status |= IB_SMP_UNSUP_METHOD;
+               ret = reply((struct ib_mad_hdr *)pmp);
+               break;
+       }
+
+       return ret;
+}
+
 static int process_perf_opa(struct ib_device *ibdev, u8 port,
                            const struct opa_mad *in_mad,
                            struct opa_mad *out_mad, u32 *resp_len)
@@ -4107,6 +4302,9 @@ static int hfi1_process_ib_mad(struct ib_device *ibdev, int mad_flags, u8 port,
        case IB_MGMT_CLASS_SUBN_LID_ROUTED:
                ret = process_subn(ibdev, mad_flags, port, in_mad, out_mad);
                break;
+       case IB_MGMT_CLASS_PERF_MGMT:
+               ret = process_perf(ibdev, port, in_mad, out_mad);
+               break;
        default:
                ret = IB_MAD_RESULT_SUCCESS;
                break;