diff --git a/drivers/infiniband/core/Makefile b/drivers/infiniband/core/Makefile index f22555f982e2..2b86a514c9bf 100644 --- a/drivers/infiniband/core/Makefile +++ b/drivers/infiniband/core/Makefile @@ -11,7 +11,8 @@ ib_core-y := packer.o ud_header.o verbs.o cq.o rw.o sysfs.o \ device.o fmr_pool.o cache.o netlink.o \ roce_gid_mgmt.o mr_pool.o addr.o sa_query.o \ multicast.o mad.o smi.o agent.o mad_rmpp.o \ - nldev.o restrack.o counters.o ib_core_uverbs.o + nldev.o restrack.o counters.o ib_core_uverbs.o \ + trace.o ib_core-$(CONFIG_SECURITY_INFINIBAND) += security.o ib_core-$(CONFIG_CGROUP_RDMA) += cgroup.o diff --git a/drivers/infiniband/core/cq.c b/drivers/infiniband/core/cq.c index bbfded6d5d3d..4f25b2400694 100644 --- a/drivers/infiniband/core/cq.c +++ b/drivers/infiniband/core/cq.c @@ -7,6 +7,8 @@ #include #include +#include + /* # of WCs to poll for with a single call to ib_poll_cq */ #define IB_POLL_BATCH 16 #define IB_POLL_BATCH_DIRECT 8 @@ -41,6 +43,7 @@ static void ib_cq_rdma_dim_work(struct work_struct *w) dim->state = DIM_START_MEASURE; + trace_cq_modify(cq, comps, usec); cq->device->ops.modify_cq(cq, comps, usec); } @@ -65,18 +68,29 @@ static void rdma_dim_init(struct ib_cq *cq) INIT_WORK(&dim->work, ib_cq_rdma_dim_work); } +static int __poll_cq(struct ib_cq *cq, int num_entries, struct ib_wc *wc) +{ + int rc; + + rc = ib_poll_cq(cq, num_entries, wc); + trace_cq_poll(cq, num_entries, rc); + return rc; +} + static int __ib_process_cq(struct ib_cq *cq, int budget, struct ib_wc *wcs, int batch) { int i, n, completed = 0; + trace_cq_process(cq); + /* * budget might be (-1) if the caller does not * want to bound this call, thus we need unsigned * minimum here. */ - while ((n = ib_poll_cq(cq, min_t(u32, batch, - budget - completed), wcs)) > 0) { + while ((n = __poll_cq(cq, min_t(u32, batch, + budget - completed), wcs)) > 0) { for (i = 0; i < n; i++) { struct ib_wc *wc = &wcs[i]; @@ -131,8 +145,10 @@ static int ib_poll_handler(struct irq_poll *iop, int budget) completed = __ib_process_cq(cq, budget, cq->wc, IB_POLL_BATCH); if (completed < budget) { irq_poll_complete(&cq->iop); - if (ib_req_notify_cq(cq, IB_POLL_FLAGS) > 0) + if (ib_req_notify_cq(cq, IB_POLL_FLAGS) > 0) { + trace_cq_reschedule(cq); irq_poll_sched(&cq->iop); + } } if (dim) @@ -143,6 +159,7 @@ static int ib_poll_handler(struct irq_poll *iop, int budget) static void ib_cq_completion_softirq(struct ib_cq *cq, void *private) { + trace_cq_schedule(cq); irq_poll_sched(&cq->iop); } @@ -162,6 +179,7 @@ static void ib_cq_poll_work(struct work_struct *work) static void ib_cq_completion_workqueue(struct ib_cq *cq, void *private) { + trace_cq_schedule(cq); queue_work(cq->comp_wq, &cq->work); } @@ -239,6 +257,7 @@ struct ib_cq *__ib_alloc_cq_user(struct ib_device *dev, void *private, goto out_destroy_cq; } + trace_cq_alloc(cq, nr_cqe, comp_vector, poll_ctx); return cq; out_destroy_cq: @@ -248,6 +267,7 @@ struct ib_cq *__ib_alloc_cq_user(struct ib_device *dev, void *private, kfree(cq->wc); out_free_cq: kfree(cq); + trace_cq_alloc_error(nr_cqe, comp_vector, poll_ctx, ret); return ERR_PTR(ret); } EXPORT_SYMBOL(__ib_alloc_cq_user); @@ -304,6 +324,7 @@ void ib_free_cq_user(struct ib_cq *cq, struct ib_udata *udata) WARN_ON_ONCE(1); } + trace_cq_free(cq); rdma_restrack_del(&cq->res); cq->device->ops.destroy_cq(cq, udata); if (cq->dim) diff --git a/drivers/infiniband/core/trace.c b/drivers/infiniband/core/trace.c new file mode 100644 index 000000000000..6c3514beac4d --- /dev/null +++ b/drivers/infiniband/core/trace.c @@ -0,0 +1,14 @@ +// SPDX-License-Identifier: GPL-2.0-only +/* + * Trace points for core RDMA functions. + * + * Author: Chuck Lever + * + * Copyright (c) 2019, Oracle and/or its affiliates. All rights reserved. + */ + +#define CREATE_TRACE_POINTS + +#include + +#include diff --git a/drivers/infiniband/core/verbs.c b/drivers/infiniband/core/verbs.c index dd765e176cdd..289b2f7a9d5e 100644 --- a/drivers/infiniband/core/verbs.c +++ b/drivers/infiniband/core/verbs.c @@ -53,6 +53,8 @@ #include "core_priv.h" +#include + static int ib_resolve_eth_dmac(struct ib_device *device, struct rdma_ah_attr *ah_attr); @@ -2744,6 +2746,7 @@ void ib_drain_sq(struct ib_qp *qp) qp->device->ops.drain_sq(qp); else __ib_drain_sq(qp); + trace_cq_drain_complete(qp->send_cq); } EXPORT_SYMBOL(ib_drain_sq); @@ -2772,6 +2775,7 @@ void ib_drain_rq(struct ib_qp *qp) qp->device->ops.drain_rq(qp); else __ib_drain_rq(qp); + trace_cq_drain_complete(qp->recv_cq); } EXPORT_SYMBOL(ib_drain_rq); diff --git a/include/rdma/ib_verbs.h b/include/rdma/ib_verbs.h index 5608e14e3aad..42f28d39f28c 100644 --- a/include/rdma/ib_verbs.h +++ b/include/rdma/ib_verbs.h @@ -1558,6 +1558,11 @@ struct ib_cq { }; struct workqueue_struct *comp_wq; struct dim *dim; + + /* updated only by trace points */ + ktime_t timestamp; + bool interrupt; + /* * Implementation details of the RDMA core, don't use in drivers: */ diff --git a/include/trace/events/rdma_core.h b/include/trace/events/rdma_core.h new file mode 100644 index 000000000000..08f481554e7f --- /dev/null +++ b/include/trace/events/rdma_core.h @@ -0,0 +1,271 @@ +/* SPDX-License-Identifier: GPL-2.0-only */ +/* + * Trace point definitions for core RDMA functions. + * + * Author: Chuck Lever + * + * Copyright (c) 2019, Oracle and/or its affiliates. All rights reserved. + */ + +#undef TRACE_SYSTEM +#define TRACE_SYSTEM rdma_core + +#if !defined(_TRACE_RDMA_CORE_H) || defined(TRACE_HEADER_MULTI_READ) +#define _TRACE_RDMA_CORE_H + +#include +#include + +/* + * enum ib_poll_context, from include/rdma/ib_verbs.h + */ +#define IB_POLL_CTX_LIST \ + ib_poll_ctx(DIRECT) \ + ib_poll_ctx(SOFTIRQ) \ + ib_poll_ctx(WORKQUEUE) \ + ib_poll_ctx_end(UNBOUND_WORKQUEUE) + +#undef ib_poll_ctx +#undef ib_poll_ctx_end + +#define ib_poll_ctx(x) TRACE_DEFINE_ENUM(IB_POLL_##x); +#define ib_poll_ctx_end(x) TRACE_DEFINE_ENUM(IB_POLL_##x); + +IB_POLL_CTX_LIST + +#undef ib_poll_ctx +#undef ib_poll_ctx_end + +#define ib_poll_ctx(x) { IB_POLL_##x, #x }, +#define ib_poll_ctx_end(x) { IB_POLL_##x, #x } + +#define rdma_show_ib_poll_ctx(x) \ + __print_symbolic(x, IB_POLL_CTX_LIST) + +/** + ** Completion Queue events + **/ + +TRACE_EVENT(cq_schedule, + TP_PROTO( + struct ib_cq *cq + ), + + TP_ARGS(cq), + + TP_STRUCT__entry( + __field(u32, cq_id) + ), + + TP_fast_assign( + cq->timestamp = ktime_get(); + cq->interrupt = true; + + __entry->cq_id = cq->res.id; + ), + + TP_printk("cq.id=%u", __entry->cq_id) +); + +TRACE_EVENT(cq_reschedule, + TP_PROTO( + struct ib_cq *cq + ), + + TP_ARGS(cq), + + TP_STRUCT__entry( + __field(u32, cq_id) + ), + + TP_fast_assign( + cq->timestamp = ktime_get(); + cq->interrupt = false; + + __entry->cq_id = cq->res.id; + ), + + TP_printk("cq.id=%u", __entry->cq_id) +); + +TRACE_EVENT(cq_process, + TP_PROTO( + const struct ib_cq *cq + ), + + TP_ARGS(cq), + + TP_STRUCT__entry( + __field(u32, cq_id) + __field(bool, interrupt) + __field(s64, latency) + ), + + TP_fast_assign( + ktime_t latency = ktime_sub(ktime_get(), cq->timestamp); + + __entry->cq_id = cq->res.id; + __entry->latency = ktime_to_us(latency); + __entry->interrupt = cq->interrupt; + ), + + TP_printk("cq.id=%u wake-up took %lld [us] from %s", + __entry->cq_id, __entry->latency, + __entry->interrupt ? "interrupt" : "reschedule" + ) +); + +TRACE_EVENT(cq_poll, + TP_PROTO( + const struct ib_cq *cq, + int requested, + int rc + ), + + TP_ARGS(cq, requested, rc), + + TP_STRUCT__entry( + __field(u32, cq_id) + __field(int, requested) + __field(int, rc) + ), + + TP_fast_assign( + __entry->cq_id = cq->res.id; + __entry->requested = requested; + __entry->rc = rc; + ), + + TP_printk("cq.id=%u requested %d, returned %d", + __entry->cq_id, __entry->requested, __entry->rc + ) +); + +TRACE_EVENT(cq_drain_complete, + TP_PROTO( + const struct ib_cq *cq + ), + + TP_ARGS(cq), + + TP_STRUCT__entry( + __field(u32, cq_id) + ), + + TP_fast_assign( + __entry->cq_id = cq->res.id; + ), + + TP_printk("cq.id=%u", + __entry->cq_id + ) +); + + +TRACE_EVENT(cq_modify, + TP_PROTO( + const struct ib_cq *cq, + u16 comps, + u16 usec + ), + + TP_ARGS(cq, comps, usec), + + TP_STRUCT__entry( + __field(u32, cq_id) + __field(unsigned int, comps) + __field(unsigned int, usec) + ), + + TP_fast_assign( + __entry->cq_id = cq->res.id; + __entry->comps = comps; + __entry->usec = usec; + ), + + TP_printk("cq.id=%u comps=%u usec=%u", + __entry->cq_id, __entry->comps, __entry->usec + ) +); + +TRACE_EVENT(cq_alloc, + TP_PROTO( + const struct ib_cq *cq, + int nr_cqe, + int comp_vector, + enum ib_poll_context poll_ctx + ), + + TP_ARGS(cq, nr_cqe, comp_vector, poll_ctx), + + TP_STRUCT__entry( + __field(u32, cq_id) + __field(int, nr_cqe) + __field(int, comp_vector) + __field(unsigned long, poll_ctx) + ), + + TP_fast_assign( + __entry->cq_id = cq->res.id; + __entry->nr_cqe = nr_cqe; + __entry->comp_vector = comp_vector; + __entry->poll_ctx = poll_ctx; + ), + + TP_printk("cq.id=%u nr_cqe=%d comp_vector=%d poll_ctx=%s", + __entry->cq_id, __entry->nr_cqe, __entry->comp_vector, + rdma_show_ib_poll_ctx(__entry->poll_ctx) + ) +); + +TRACE_EVENT(cq_alloc_error, + TP_PROTO( + int nr_cqe, + int comp_vector, + enum ib_poll_context poll_ctx, + int rc + ), + + TP_ARGS(nr_cqe, comp_vector, poll_ctx, rc), + + TP_STRUCT__entry( + __field(int, rc) + __field(int, nr_cqe) + __field(int, comp_vector) + __field(unsigned long, poll_ctx) + ), + + TP_fast_assign( + __entry->rc = rc; + __entry->nr_cqe = nr_cqe; + __entry->comp_vector = comp_vector; + __entry->poll_ctx = poll_ctx; + ), + + TP_printk("nr_cqe=%d comp_vector=%d poll_ctx=%s rc=%d", + __entry->nr_cqe, __entry->comp_vector, + rdma_show_ib_poll_ctx(__entry->poll_ctx), __entry->rc + ) +); + +TRACE_EVENT(cq_free, + TP_PROTO( + const struct ib_cq *cq + ), + + TP_ARGS(cq), + + TP_STRUCT__entry( + __field(u32, cq_id) + ), + + TP_fast_assign( + __entry->cq_id = cq->res.id; + ), + + TP_printk("cq.id=%u", __entry->cq_id) +); + +#endif /* _TRACE_RDMA_CORE_H */ + +#include