From f4fd0b224d60044d2da5ca02f8f2b5150c1d8731 Mon Sep 17 00:00:00 2001 From: "Michael S. Tsirkin" Date: Thu, 3 May 2007 13:48:47 +0300 Subject: IB: Add CQ comp_vector support Add a num_comp_vectors member to struct ib_device and extend ib_create_cq() to pass in a comp_vector parameter -- this parallels the userspace libibverbs API. Update all hardware drivers to set num_comp_vectors to 1 and have all ULPs pass 0 for the comp_vector value. Pass the value of num_comp_vectors to userspace rather than hard-coding a value of 1. We want multiple CQ event vector support (via MSI-X or similar for adapters that can generate multiple interrupts), but it's not clear how many vectors we want, or how we want to deal with policy issues such as how to decide which vector to use or how to set up interrupt affinity. This patch is useful for experimenting, since no core changes will be necessary when updating a driver to support multiple vectors, and we know that we want to make at least these changes anyway. Signed-off-by: Michael S. Tsirkin Signed-off-by: Roland Dreier --- include/rdma/ib_verbs.h | 7 ++++++- 1 file changed, 6 insertions(+), 1 deletion(-) (limited to 'include/rdma') diff --git a/include/rdma/ib_verbs.h b/include/rdma/ib_verbs.h index 765589f4d166..17cc309d03ef 100644 --- a/include/rdma/ib_verbs.h +++ b/include/rdma/ib_verbs.h @@ -912,6 +912,8 @@ struct ib_device { u32 flags; + int num_comp_vectors; + struct iw_cm_verbs *iwcm; int (*query_device)(struct ib_device *device, @@ -978,6 +980,7 @@ struct ib_device { struct ib_recv_wr *recv_wr, struct ib_recv_wr **bad_recv_wr); struct ib_cq * (*create_cq)(struct ib_device *device, int cqe, + int comp_vector, struct ib_ucontext *context, struct ib_udata *udata); int (*destroy_cq)(struct ib_cq *cq); @@ -1358,13 +1361,15 @@ static inline int ib_post_recv(struct ib_qp *qp, * @cq_context: Context associated with the CQ returned to the user via * the associated completion and event handlers. * @cqe: The minimum size of the CQ. + * @comp_vector - Completion vector used to signal completion events. + * Must be >= 0 and < context->num_comp_vectors. * * Users can examine the cq structure to determine the actual CQ size. */ struct ib_cq *ib_create_cq(struct ib_device *device, ib_comp_handler comp_handler, void (*event_handler)(struct ib_event *, void *), - void *cq_context, int cqe); + void *cq_context, int cqe, int comp_vector); /** * ib_resize_cq - Modifies the capacity of the CQ. -- cgit v1.2.3-58-ga151 From ed23a72778f3dbd465e55b06fe31629e7e1dd2f3 Mon Sep 17 00:00:00 2001 From: Roland Dreier Date: Sun, 6 May 2007 21:02:48 -0700 Subject: IB: Return "maybe missed event" hint from ib_req_notify_cq() The semantics defined by the InfiniBand specification say that completion events are only generated when a completions is added to a completion queue (CQ) after completion notification is requested. In other words, this means that the following race is possible: while (CQ is not empty) ib_poll_cq(CQ); // new completion is added after while loop is exited ib_req_notify_cq(CQ); // no event is generated for the existing completion To close this race, the IB spec recommends doing another poll of the CQ after requesting notification. However, it is not always possible to arrange code this way (for example, we have found that NAPI for IPoIB cannot poll after requesting notification). Also, some hardware (eg Mellanox HCAs) actually will generate an event for completions added before the call to ib_req_notify_cq() -- which is allowed by the spec, since there's no way for any upper-layer consumer to know exactly when a completion was really added -- so the extra poll of the CQ is just a waste. Motivated by this, we add a new flag "IB_CQ_REPORT_MISSED_EVENTS" for ib_req_notify_cq() so that it can return a hint about whether the a completion may have been added before the request for notification. The return value of ib_req_notify_cq() is extended so: < 0 means an error occurred while requesting notification == 0 means notification was requested successfully, and if IB_CQ_REPORT_MISSED_EVENTS was passed in, then no events were missed and it is safe to wait for another event. > 0 is only returned if IB_CQ_REPORT_MISSED_EVENTS was passed in. It means that the consumer must poll the CQ again to make sure it is empty to avoid the race described above. We add a flag to enable this behavior rather than turning it on unconditionally, because checking for missed events may incur significant overhead for some low-level drivers, and consumers that don't care about the results of this test shouldn't be forced to pay for the test. Signed-off-by: Roland Dreier --- drivers/infiniband/hw/amso1100/c2.h | 2 +- drivers/infiniband/hw/amso1100/c2_cq.c | 16 +++++++++--- drivers/infiniband/hw/cxgb3/cxio_hal.c | 3 +++ drivers/infiniband/hw/cxgb3/iwch_provider.c | 8 +++--- drivers/infiniband/hw/ehca/ehca_iverbs.h | 2 +- drivers/infiniband/hw/ehca/ehca_reqs.c | 14 +++++++--- drivers/infiniband/hw/ehca/ipz_pt_fn.h | 8 ++++++ drivers/infiniband/hw/ipath/ipath_cq.c | 15 ++++++++--- drivers/infiniband/hw/ipath/ipath_verbs.h | 2 +- drivers/infiniband/hw/mthca/mthca_cq.c | 12 +++++---- drivers/infiniband/hw/mthca/mthca_dev.h | 4 +-- include/rdma/ib_verbs.h | 40 ++++++++++++++++++++++------- 12 files changed, 93 insertions(+), 33 deletions(-) (limited to 'include/rdma') diff --git a/drivers/infiniband/hw/amso1100/c2.h b/drivers/infiniband/hw/amso1100/c2.h index 04a9db5de881..fa58200217a1 100644 --- a/drivers/infiniband/hw/amso1100/c2.h +++ b/drivers/infiniband/hw/amso1100/c2.h @@ -519,7 +519,7 @@ extern void c2_free_cq(struct c2_dev *c2dev, struct c2_cq *cq); extern void c2_cq_event(struct c2_dev *c2dev, u32 mq_index); extern void c2_cq_clean(struct c2_dev *c2dev, struct c2_qp *qp, u32 mq_index); extern int c2_poll_cq(struct ib_cq *ibcq, int num_entries, struct ib_wc *entry); -extern int c2_arm_cq(struct ib_cq *ibcq, enum ib_cq_notify notify); +extern int c2_arm_cq(struct ib_cq *ibcq, enum ib_cq_notify_flags flags); /* CM */ extern int c2_llp_connect(struct iw_cm_id *cm_id, diff --git a/drivers/infiniband/hw/amso1100/c2_cq.c b/drivers/infiniband/hw/amso1100/c2_cq.c index 5175c99ee586..d2b3366786d6 100644 --- a/drivers/infiniband/hw/amso1100/c2_cq.c +++ b/drivers/infiniband/hw/amso1100/c2_cq.c @@ -217,17 +217,19 @@ int c2_poll_cq(struct ib_cq *ibcq, int num_entries, struct ib_wc *entry) return npolled; } -int c2_arm_cq(struct ib_cq *ibcq, enum ib_cq_notify notify) +int c2_arm_cq(struct ib_cq *ibcq, enum ib_cq_notify_flags notify_flags) { struct c2_mq_shared __iomem *shared; struct c2_cq *cq; + unsigned long flags; + int ret = 0; cq = to_c2cq(ibcq); shared = cq->mq.peer; - if (notify == IB_CQ_NEXT_COMP) + if ((notify_flags & IB_CQ_SOLICITED_MASK) == IB_CQ_NEXT_COMP) writeb(C2_CQ_NOTIFICATION_TYPE_NEXT, &shared->notification_type); - else if (notify == IB_CQ_SOLICITED) + else if ((notify_flags & IB_CQ_SOLICITED_MASK) == IB_CQ_SOLICITED) writeb(C2_CQ_NOTIFICATION_TYPE_NEXT_SE, &shared->notification_type); else return -EINVAL; @@ -241,7 +243,13 @@ int c2_arm_cq(struct ib_cq *ibcq, enum ib_cq_notify notify) */ readb(&shared->armed); - return 0; + if (notify_flags & IB_CQ_REPORT_MISSED_EVENTS) { + spin_lock_irqsave(&cq->lock, flags); + ret = !c2_mq_empty(&cq->mq); + spin_unlock_irqrestore(&cq->lock, flags); + } + + return ret; } static void c2_free_cq_buf(struct c2_dev *c2dev, struct c2_mq *mq) diff --git a/drivers/infiniband/hw/cxgb3/cxio_hal.c b/drivers/infiniband/hw/cxgb3/cxio_hal.c index f5e9aeec6f6e..76049afc7655 100644 --- a/drivers/infiniband/hw/cxgb3/cxio_hal.c +++ b/drivers/infiniband/hw/cxgb3/cxio_hal.c @@ -114,7 +114,10 @@ int cxio_hal_cq_op(struct cxio_rdev *rdev_p, struct t3_cq *cq, return -EIO; } } + + return 1; } + return 0; } diff --git a/drivers/infiniband/hw/cxgb3/iwch_provider.c b/drivers/infiniband/hw/cxgb3/iwch_provider.c index 78a495f5332e..a891493fd340 100644 --- a/drivers/infiniband/hw/cxgb3/iwch_provider.c +++ b/drivers/infiniband/hw/cxgb3/iwch_provider.c @@ -292,7 +292,7 @@ static int iwch_resize_cq(struct ib_cq *cq, int cqe, struct ib_udata *udata) #endif } -static int iwch_arm_cq(struct ib_cq *ibcq, enum ib_cq_notify notify) +static int iwch_arm_cq(struct ib_cq *ibcq, enum ib_cq_notify_flags flags) { struct iwch_dev *rhp; struct iwch_cq *chp; @@ -303,7 +303,7 @@ static int iwch_arm_cq(struct ib_cq *ibcq, enum ib_cq_notify notify) chp = to_iwch_cq(ibcq); rhp = chp->rhp; - if (notify == IB_CQ_SOLICITED) + if ((flags & IB_CQ_SOLICITED_MASK) == IB_CQ_SOLICITED) cq_op = CQ_ARM_SE; else cq_op = CQ_ARM_AN; @@ -317,9 +317,11 @@ static int iwch_arm_cq(struct ib_cq *ibcq, enum ib_cq_notify notify) PDBG("%s rptr 0x%x\n", __FUNCTION__, chp->cq.rptr); err = cxio_hal_cq_op(&rhp->rdev, &chp->cq, cq_op, 0); spin_unlock_irqrestore(&chp->lock, flag); - if (err) + if (err < 0) printk(KERN_ERR MOD "Error %d rearming CQID 0x%x\n", err, chp->cq.cqid); + if (err > 0 && !(flags & IB_CQ_REPORT_MISSED_EVENTS)) + err = 0; return err; } diff --git a/drivers/infiniband/hw/ehca/ehca_iverbs.h b/drivers/infiniband/hw/ehca/ehca_iverbs.h index aff96ac4fd12..e14b029332c8 100644 --- a/drivers/infiniband/hw/ehca/ehca_iverbs.h +++ b/drivers/infiniband/hw/ehca/ehca_iverbs.h @@ -135,7 +135,7 @@ int ehca_poll_cq(struct ib_cq *cq, int num_entries, struct ib_wc *wc); int ehca_peek_cq(struct ib_cq *cq, int wc_cnt); -int ehca_req_notify_cq(struct ib_cq *cq, enum ib_cq_notify cq_notify); +int ehca_req_notify_cq(struct ib_cq *cq, enum ib_cq_notify_flags notify_flags); struct ib_qp *ehca_create_qp(struct ib_pd *pd, struct ib_qp_init_attr *init_attr, diff --git a/drivers/infiniband/hw/ehca/ehca_reqs.c b/drivers/infiniband/hw/ehca/ehca_reqs.c index 08d3f892d9f3..caec9dee09e1 100644 --- a/drivers/infiniband/hw/ehca/ehca_reqs.c +++ b/drivers/infiniband/hw/ehca/ehca_reqs.c @@ -634,11 +634,13 @@ poll_cq_exit0: return ret; } -int ehca_req_notify_cq(struct ib_cq *cq, enum ib_cq_notify cq_notify) +int ehca_req_notify_cq(struct ib_cq *cq, enum ib_cq_notify_flags notify_flags) { struct ehca_cq *my_cq = container_of(cq, struct ehca_cq, ib_cq); + unsigned long spl_flags; + int ret = 0; - switch (cq_notify) { + switch (notify_flags & IB_CQ_SOLICITED_MASK) { case IB_CQ_SOLICITED: hipz_set_cqx_n0(my_cq, 1); break; @@ -649,5 +651,11 @@ int ehca_req_notify_cq(struct ib_cq *cq, enum ib_cq_notify cq_notify) return -EINVAL; } - return 0; + if (notify_flags & IB_CQ_REPORT_MISSED_EVENTS) { + spin_lock_irqsave(&my_cq->spinlock, spl_flags); + ret = ipz_qeit_is_valid(&my_cq->ipz_queue); + spin_unlock_irqrestore(&my_cq->spinlock, spl_flags); + } + + return ret; } diff --git a/drivers/infiniband/hw/ehca/ipz_pt_fn.h b/drivers/infiniband/hw/ehca/ipz_pt_fn.h index 8199c45768a3..57f141a36bce 100644 --- a/drivers/infiniband/hw/ehca/ipz_pt_fn.h +++ b/drivers/infiniband/hw/ehca/ipz_pt_fn.h @@ -140,6 +140,14 @@ static inline void *ipz_qeit_get_inc_valid(struct ipz_queue *queue) return cqe; } +static inline int ipz_qeit_is_valid(struct ipz_queue *queue) +{ + struct ehca_cqe *cqe = ipz_qeit_get(queue); + u32 cqe_flags = cqe->cqe_flags; + + return cqe_flags >> 7 == (queue->toggle_state & 1); +} + /* * returns and resets Queue Entry iterator * returns address (kv) of first Queue Entry diff --git a/drivers/infiniband/hw/ipath/ipath_cq.c b/drivers/infiniband/hw/ipath/ipath_cq.c index 00d3eb9bc696..3e9241badba0 100644 --- a/drivers/infiniband/hw/ipath/ipath_cq.c +++ b/drivers/infiniband/hw/ipath/ipath_cq.c @@ -334,17 +334,18 @@ int ipath_destroy_cq(struct ib_cq *ibcq) /** * ipath_req_notify_cq - change the notification type for a completion queue * @ibcq: the completion queue - * @notify: the type of notification to request + * @notify_flags: the type of notification to request * * Returns 0 for success. * * This may be called from interrupt context. Also called by * ib_req_notify_cq() in the generic verbs code. */ -int ipath_req_notify_cq(struct ib_cq *ibcq, enum ib_cq_notify notify) +int ipath_req_notify_cq(struct ib_cq *ibcq, enum ib_cq_notify_flags notify_flags) { struct ipath_cq *cq = to_icq(ibcq); unsigned long flags; + int ret = 0; spin_lock_irqsave(&cq->lock, flags); /* @@ -352,9 +353,15 @@ int ipath_req_notify_cq(struct ib_cq *ibcq, enum ib_cq_notify notify) * any other transitions (see C11-31 and C11-32 in ch. 11.4.2.2). */ if (cq->notify != IB_CQ_NEXT_COMP) - cq->notify = notify; + cq->notify = notify_flags & IB_CQ_SOLICITED_MASK; + + if ((notify_flags & IB_CQ_REPORT_MISSED_EVENTS) && + cq->queue->head != cq->queue->tail) + ret = 1; + spin_unlock_irqrestore(&cq->lock, flags); - return 0; + + return ret; } /** diff --git a/drivers/infiniband/hw/ipath/ipath_verbs.h b/drivers/infiniband/hw/ipath/ipath_verbs.h index 2d734fb6eff7..7064fc222727 100644 --- a/drivers/infiniband/hw/ipath/ipath_verbs.h +++ b/drivers/infiniband/hw/ipath/ipath_verbs.h @@ -741,7 +741,7 @@ struct ib_cq *ipath_create_cq(struct ib_device *ibdev, int entries, int comp_vec int ipath_destroy_cq(struct ib_cq *ibcq); -int ipath_req_notify_cq(struct ib_cq *ibcq, enum ib_cq_notify notify); +int ipath_req_notify_cq(struct ib_cq *ibcq, enum ib_cq_notify_flags notify_flags); int ipath_resize_cq(struct ib_cq *ibcq, int cqe, struct ib_udata *udata); diff --git a/drivers/infiniband/hw/mthca/mthca_cq.c b/drivers/infiniband/hw/mthca/mthca_cq.c index efd79ef109a6..cf0868f6e965 100644 --- a/drivers/infiniband/hw/mthca/mthca_cq.c +++ b/drivers/infiniband/hw/mthca/mthca_cq.c @@ -726,11 +726,12 @@ repoll: return err == 0 || err == -EAGAIN ? npolled : err; } -int mthca_tavor_arm_cq(struct ib_cq *cq, enum ib_cq_notify notify) +int mthca_tavor_arm_cq(struct ib_cq *cq, enum ib_cq_notify_flags flags) { __be32 doorbell[2]; - doorbell[0] = cpu_to_be32((notify == IB_CQ_SOLICITED ? + doorbell[0] = cpu_to_be32(((flags & IB_CQ_SOLICITED_MASK) == + IB_CQ_SOLICITED ? MTHCA_TAVOR_CQ_DB_REQ_NOT_SOL : MTHCA_TAVOR_CQ_DB_REQ_NOT) | to_mcq(cq)->cqn); @@ -743,7 +744,7 @@ int mthca_tavor_arm_cq(struct ib_cq *cq, enum ib_cq_notify notify) return 0; } -int mthca_arbel_arm_cq(struct ib_cq *ibcq, enum ib_cq_notify notify) +int mthca_arbel_arm_cq(struct ib_cq *ibcq, enum ib_cq_notify_flags flags) { struct mthca_cq *cq = to_mcq(ibcq); __be32 doorbell[2]; @@ -755,7 +756,8 @@ int mthca_arbel_arm_cq(struct ib_cq *ibcq, enum ib_cq_notify notify) doorbell[0] = ci; doorbell[1] = cpu_to_be32((cq->cqn << 8) | (2 << 5) | (sn << 3) | - (notify == IB_CQ_SOLICITED ? 1 : 2)); + ((flags & IB_CQ_SOLICITED_MASK) == + IB_CQ_SOLICITED ? 1 : 2)); mthca_write_db_rec(doorbell, cq->arm_db); @@ -766,7 +768,7 @@ int mthca_arbel_arm_cq(struct ib_cq *ibcq, enum ib_cq_notify notify) wmb(); doorbell[0] = cpu_to_be32((sn << 28) | - (notify == IB_CQ_SOLICITED ? + ((flags & IB_CQ_SOLICITED_MASK) == IB_CQ_SOLICITED ? MTHCA_ARBEL_CQ_DB_REQ_NOT_SOL : MTHCA_ARBEL_CQ_DB_REQ_NOT) | cq->cqn); diff --git a/drivers/infiniband/hw/mthca/mthca_dev.h b/drivers/infiniband/hw/mthca/mthca_dev.h index b7e42efaf43d..9bae3cc60603 100644 --- a/drivers/infiniband/hw/mthca/mthca_dev.h +++ b/drivers/infiniband/hw/mthca/mthca_dev.h @@ -495,8 +495,8 @@ void mthca_unmap_eq_icm(struct mthca_dev *dev); int mthca_poll_cq(struct ib_cq *ibcq, int num_entries, struct ib_wc *entry); -int mthca_tavor_arm_cq(struct ib_cq *cq, enum ib_cq_notify notify); -int mthca_arbel_arm_cq(struct ib_cq *cq, enum ib_cq_notify notify); +int mthca_tavor_arm_cq(struct ib_cq *cq, enum ib_cq_notify_flags flags); +int mthca_arbel_arm_cq(struct ib_cq *cq, enum ib_cq_notify_flags flags); int mthca_init_cq(struct mthca_dev *dev, int nent, struct mthca_ucontext *ctx, u32 pdn, struct mthca_cq *cq); diff --git a/include/rdma/ib_verbs.h b/include/rdma/ib_verbs.h index 17cc309d03ef..5342ac64ed1a 100644 --- a/include/rdma/ib_verbs.h +++ b/include/rdma/ib_verbs.h @@ -431,9 +431,11 @@ struct ib_wc { u8 port_num; /* valid only for DR SMPs on switches */ }; -enum ib_cq_notify { - IB_CQ_SOLICITED, - IB_CQ_NEXT_COMP +enum ib_cq_notify_flags { + IB_CQ_SOLICITED = 1 << 0, + IB_CQ_NEXT_COMP = 1 << 1, + IB_CQ_SOLICITED_MASK = IB_CQ_SOLICITED | IB_CQ_NEXT_COMP, + IB_CQ_REPORT_MISSED_EVENTS = 1 << 2, }; enum ib_srq_attr_mask { @@ -990,7 +992,7 @@ struct ib_device { struct ib_wc *wc); int (*peek_cq)(struct ib_cq *cq, int wc_cnt); int (*req_notify_cq)(struct ib_cq *cq, - enum ib_cq_notify cq_notify); + enum ib_cq_notify_flags flags); int (*req_ncomp_notif)(struct ib_cq *cq, int wc_cnt); struct ib_mr * (*get_dma_mr)(struct ib_pd *pd, @@ -1419,14 +1421,34 @@ int ib_peek_cq(struct ib_cq *cq, int wc_cnt); /** * ib_req_notify_cq - Request completion notification on a CQ. * @cq: The CQ to generate an event for. - * @cq_notify: If set to %IB_CQ_SOLICITED, completion notification will - * occur on the next solicited event. If set to %IB_CQ_NEXT_COMP, - * notification will occur on the next completion. + * @flags: + * Must contain exactly one of %IB_CQ_SOLICITED or %IB_CQ_NEXT_COMP + * to request an event on the next solicited event or next work + * completion at any type, respectively. %IB_CQ_REPORT_MISSED_EVENTS + * may also be |ed in to request a hint about missed events, as + * described below. + * + * Return Value: + * < 0 means an error occurred while requesting notification + * == 0 means notification was requested successfully, and if + * IB_CQ_REPORT_MISSED_EVENTS was passed in, then no events + * were missed and it is safe to wait for another event. In + * this case is it guaranteed that any work completions added + * to the CQ since the last CQ poll will trigger a completion + * notification event. + * > 0 is only returned if IB_CQ_REPORT_MISSED_EVENTS was passed + * in. It means that the consumer must poll the CQ again to + * make sure it is empty to avoid missing an event because of a + * race between requesting notification and an entry being + * added to the CQ. This return value means it is possible + * (but not guaranteed) that a work completion has been added + * to the CQ since the last poll without triggering a + * completion notification event. */ static inline int ib_req_notify_cq(struct ib_cq *cq, - enum ib_cq_notify cq_notify) + enum ib_cq_notify_flags flags) { - return cq->device->req_notify_cq(cq, cq_notify); + return cq->device->req_notify_cq(cq, flags); } /** -- cgit v1.2.3-58-ga151