diff options
Diffstat (limited to 'drivers/infiniband/hw/hfi1')
-rw-r--r-- | drivers/infiniband/hw/hfi1/affinity.c | 8 | ||||
-rw-r--r-- | drivers/infiniband/hw/hfi1/chip.c | 11 | ||||
-rw-r--r-- | drivers/infiniband/hw/hfi1/chip.h | 2 | ||||
-rw-r--r-- | drivers/infiniband/hw/hfi1/chip_registers.h | 1 | ||||
-rw-r--r-- | drivers/infiniband/hw/hfi1/driver.c | 1 | ||||
-rw-r--r-- | drivers/infiniband/hw/hfi1/file_ops.c | 52 | ||||
-rw-r--r-- | drivers/infiniband/hw/hfi1/hfi.h | 7 | ||||
-rw-r--r-- | drivers/infiniband/hw/hfi1/init.c | 38 | ||||
-rw-r--r-- | drivers/infiniband/hw/hfi1/iowait.c | 4 | ||||
-rw-r--r-- | drivers/infiniband/hw/hfi1/pcie.c | 4 | ||||
-rw-r--r-- | drivers/infiniband/hw/hfi1/pio.c | 3 | ||||
-rw-r--r-- | drivers/infiniband/hw/hfi1/qp.c | 5 | ||||
-rw-r--r-- | drivers/infiniband/hw/hfi1/rc.c | 16 | ||||
-rw-r--r-- | drivers/infiniband/hw/hfi1/sdma.c | 8 | ||||
-rw-r--r-- | drivers/infiniband/hw/hfi1/sysfs.c | 26 | ||||
-rw-r--r-- | drivers/infiniband/hw/hfi1/tid_rdma.c | 76 | ||||
-rw-r--r-- | drivers/infiniband/hw/hfi1/tid_rdma.h | 3 | ||||
-rw-r--r-- | drivers/infiniband/hw/hfi1/user_exp_rcv.c | 5 | ||||
-rw-r--r-- | drivers/infiniband/hw/hfi1/user_sdma.c | 46 | ||||
-rw-r--r-- | drivers/infiniband/hw/hfi1/user_sdma.h | 6 | ||||
-rw-r--r-- | drivers/infiniband/hw/hfi1/verbs.c | 14 |
21 files changed, 217 insertions, 119 deletions
diff --git a/drivers/infiniband/hw/hfi1/affinity.c b/drivers/infiniband/hw/hfi1/affinity.c index 4fe662c3bbc1..1aeea5d65c01 100644 --- a/drivers/infiniband/hw/hfi1/affinity.c +++ b/drivers/infiniband/hw/hfi1/affinity.c @@ -479,6 +479,8 @@ static int _dev_comp_vect_mappings_create(struct hfi1_devdata *dd, rvt_get_ibdev_name(&(dd)->verbs_dev.rdi), i, cpu); } + free_cpumask_var(available_cpus); + free_cpumask_var(non_intr_cpus); return 0; fail: @@ -1038,7 +1040,7 @@ int hfi1_get_proc_affinity(int node) struct hfi1_affinity_node *entry; cpumask_var_t diff, hw_thread_mask, available_mask, intrs_mask; const struct cpumask *node_mask, - *proc_mask = ¤t->cpus_allowed; + *proc_mask = current->cpus_ptr; struct hfi1_affinity_node_list *affinity = &node_affinity; struct cpu_mask_set *set = &affinity->proc; @@ -1046,7 +1048,7 @@ int hfi1_get_proc_affinity(int node) * check whether process/context affinity has already * been set */ - if (cpumask_weight(proc_mask) == 1) { + if (current->nr_cpus_allowed == 1) { hfi1_cdbg(PROC, "PID %u %s affinity set to CPU %*pbl", current->pid, current->comm, cpumask_pr_args(proc_mask)); @@ -1057,7 +1059,7 @@ int hfi1_get_proc_affinity(int node) cpu = cpumask_first(proc_mask); cpumask_set_cpu(cpu, &set->used); goto done; - } else if (cpumask_weight(proc_mask) < cpumask_weight(&set->mask)) { + } else if (current->nr_cpus_allowed < cpumask_weight(&set->mask)) { hfi1_cdbg(PROC, "PID %u %s affinity set to CPU set(s) %*pbl", current->pid, current->comm, cpumask_pr_args(proc_mask)); diff --git a/drivers/infiniband/hw/hfi1/chip.c b/drivers/infiniband/hw/hfi1/chip.c index 67052dc3100c..e510f019d712 100644 --- a/drivers/infiniband/hw/hfi1/chip.c +++ b/drivers/infiniband/hw/hfi1/chip.c @@ -1685,6 +1685,14 @@ static u64 access_sw_pio_drain(const struct cntr_entry *entry, return dd->verbs_dev.n_piodrain; } +static u64 access_sw_ctx0_seq_drop(const struct cntr_entry *entry, + void *context, int vl, int mode, u64 data) +{ + struct hfi1_devdata *dd = context; + + return dd->ctx0_seq_drop; +} + static u64 access_sw_vtx_wait(const struct cntr_entry *entry, void *context, int vl, int mode, u64 data) { @@ -4105,6 +4113,7 @@ def_access_ibp_counter(seq_naks); static struct cntr_entry dev_cntrs[DEV_CNTR_LAST] = { [C_RCV_OVF] = RXE32_DEV_CNTR_ELEM(RcvOverflow, RCV_BUF_OVFL_CNT, CNTR_SYNTH), [C_RX_LEN_ERR] = RXE32_DEV_CNTR_ELEM(RxLenErr, RCV_LENGTH_ERR_CNT, CNTR_SYNTH), +[C_RX_SHORT_ERR] = RXE32_DEV_CNTR_ELEM(RxShrErr, RCV_SHORT_ERR_CNT, CNTR_SYNTH), [C_RX_ICRC_ERR] = RXE32_DEV_CNTR_ELEM(RxICrcErr, RCV_ICRC_ERR_CNT, CNTR_SYNTH), [C_RX_EBP] = RXE32_DEV_CNTR_ELEM(RxEbpCnt, RCV_EBP_CNT, CNTR_SYNTH), [C_RX_TID_FULL] = RXE32_DEV_CNTR_ELEM(RxTIDFullEr, RCV_TID_FULL_ERR_CNT, @@ -4248,6 +4257,8 @@ static struct cntr_entry dev_cntrs[DEV_CNTR_LAST] = { access_sw_cpu_intr), [C_SW_CPU_RCV_LIM] = CNTR_ELEM("RcvLimit", 0, 0, CNTR_NORMAL, access_sw_cpu_rcv_limit), +[C_SW_CTX0_SEQ_DROP] = CNTR_ELEM("SeqDrop0", 0, 0, CNTR_NORMAL, + access_sw_ctx0_seq_drop), [C_SW_VTX_WAIT] = CNTR_ELEM("vTxWait", 0, 0, CNTR_NORMAL, access_sw_vtx_wait), [C_SW_PIO_WAIT] = CNTR_ELEM("PioWait", 0, 0, CNTR_NORMAL, diff --git a/drivers/infiniband/hw/hfi1/chip.h b/drivers/infiniband/hw/hfi1/chip.h index b76cf81f927f..39a57c35af22 100644 --- a/drivers/infiniband/hw/hfi1/chip.h +++ b/drivers/infiniband/hw/hfi1/chip.h @@ -859,6 +859,7 @@ static inline int idx_from_vl(int vl) enum { C_RCV_OVF = 0, C_RX_LEN_ERR, + C_RX_SHORT_ERR, C_RX_ICRC_ERR, C_RX_EBP, C_RX_TID_FULL, @@ -926,6 +927,7 @@ enum { C_DC_PG_STS_TX_MBE_CNT, C_SW_CPU_INTR, C_SW_CPU_RCV_LIM, + C_SW_CTX0_SEQ_DROP, C_SW_VTX_WAIT, C_SW_PIO_WAIT, C_SW_PIO_DRAIN, diff --git a/drivers/infiniband/hw/hfi1/chip_registers.h b/drivers/infiniband/hw/hfi1/chip_registers.h index ab3589d17aee..fb3ec9bff7a2 100644 --- a/drivers/infiniband/hw/hfi1/chip_registers.h +++ b/drivers/infiniband/hw/hfi1/chip_registers.h @@ -381,6 +381,7 @@ #define DC_LCB_STS_LINK_TRANSFER_ACTIVE (DC_LCB_CSRS + 0x000000000468) #define DC_LCB_STS_ROUND_TRIP_LTP_CNT (DC_LCB_CSRS + 0x0000000004B0) #define RCV_LENGTH_ERR_CNT 0 +#define RCV_SHORT_ERR_CNT 2 #define RCV_ICRC_ERR_CNT 6 #define RCV_EBP_CNT 9 #define RCV_BUF_OVFL_CNT 10 diff --git a/drivers/infiniband/hw/hfi1/driver.c b/drivers/infiniband/hw/hfi1/driver.c index 01aa1f132f55..941b465244ab 100644 --- a/drivers/infiniband/hw/hfi1/driver.c +++ b/drivers/infiniband/hw/hfi1/driver.c @@ -734,6 +734,7 @@ static noinline int skip_rcv_packet(struct hfi1_packet *packet, int thread) { int ret; + packet->rcd->dd->ctx0_seq_drop++; /* Set up for the next packet */ packet->rhqoff += packet->rsize; if (packet->rhqoff >= packet->maxcnt) diff --git a/drivers/infiniband/hw/hfi1/file_ops.c b/drivers/infiniband/hw/hfi1/file_ops.c index f9a7e9d29c8b..89e1dfd07a1b 100644 --- a/drivers/infiniband/hw/hfi1/file_ops.c +++ b/drivers/infiniband/hw/hfi1/file_ops.c @@ -200,23 +200,24 @@ static int hfi1_file_open(struct inode *inode, struct file *fp) fd = kzalloc(sizeof(*fd), GFP_KERNEL); - if (fd) { - fd->rec_cpu_num = -1; /* no cpu affinity by default */ - fd->mm = current->mm; - mmgrab(fd->mm); - fd->dd = dd; - kobject_get(&fd->dd->kobj); - fp->private_data = fd; - } else { - fp->private_data = NULL; - - if (atomic_dec_and_test(&dd->user_refcount)) - complete(&dd->user_comp); - - return -ENOMEM; - } - + if (!fd || init_srcu_struct(&fd->pq_srcu)) + goto nomem; + spin_lock_init(&fd->pq_rcu_lock); + spin_lock_init(&fd->tid_lock); + spin_lock_init(&fd->invalid_lock); + fd->rec_cpu_num = -1; /* no cpu affinity by default */ + fd->mm = current->mm; + mmgrab(fd->mm); + fd->dd = dd; + kobject_get(&fd->dd->kobj); + fp->private_data = fd; return 0; +nomem: + kfree(fd); + fp->private_data = NULL; + if (atomic_dec_and_test(&dd->user_refcount)) + complete(&dd->user_comp); + return -ENOMEM; } static long hfi1_file_ioctl(struct file *fp, unsigned int cmd, @@ -301,21 +302,30 @@ static long hfi1_file_ioctl(struct file *fp, unsigned int cmd, static ssize_t hfi1_write_iter(struct kiocb *kiocb, struct iov_iter *from) { struct hfi1_filedata *fd = kiocb->ki_filp->private_data; - struct hfi1_user_sdma_pkt_q *pq = fd->pq; + struct hfi1_user_sdma_pkt_q *pq; struct hfi1_user_sdma_comp_q *cq = fd->cq; int done = 0, reqs = 0; unsigned long dim = from->nr_segs; + int idx; - if (!cq || !pq) + idx = srcu_read_lock(&fd->pq_srcu); + pq = srcu_dereference(fd->pq, &fd->pq_srcu); + if (!cq || !pq) { + srcu_read_unlock(&fd->pq_srcu, idx); return -EIO; + } - if (!iter_is_iovec(from) || !dim) + if (!iter_is_iovec(from) || !dim) { + srcu_read_unlock(&fd->pq_srcu, idx); return -EINVAL; + } trace_hfi1_sdma_request(fd->dd, fd->uctxt->ctxt, fd->subctxt, dim); - if (atomic_read(&pq->n_reqs) == pq->n_max_reqs) + if (atomic_read(&pq->n_reqs) == pq->n_max_reqs) { + srcu_read_unlock(&fd->pq_srcu, idx); return -ENOSPC; + } while (dim) { int ret; @@ -333,6 +343,7 @@ static ssize_t hfi1_write_iter(struct kiocb *kiocb, struct iov_iter *from) reqs++; } + srcu_read_unlock(&fd->pq_srcu, idx); return reqs; } @@ -707,6 +718,7 @@ done: if (atomic_dec_and_test(&dd->user_refcount)) complete(&dd->user_comp); + cleanup_srcu_struct(&fdata->pq_srcu); kfree(fdata); return 0; } diff --git a/drivers/infiniband/hw/hfi1/hfi.h b/drivers/infiniband/hw/hfi1/hfi.h index fa45350a9a1d..b79931cc74ab 100644 --- a/drivers/infiniband/hw/hfi1/hfi.h +++ b/drivers/infiniband/hw/hfi1/hfi.h @@ -1153,6 +1153,8 @@ struct hfi1_devdata { char *boardname; /* human readable board info */ + u64 ctx0_seq_drop; + /* reset value */ u64 z_int_counter; u64 z_rcv_limit; @@ -1436,10 +1438,13 @@ struct mmu_rb_handler; /* Private data for file operations */ struct hfi1_filedata { + struct srcu_struct pq_srcu; struct hfi1_devdata *dd; struct hfi1_ctxtdata *uctxt; struct hfi1_user_sdma_comp_q *cq; - struct hfi1_user_sdma_pkt_q *pq; + /* update side lock for SRCU */ + spinlock_t pq_rcu_lock; + struct hfi1_user_sdma_pkt_q __rcu *pq; u16 subctxt; /* for cpu affinity; -1 if none */ int rec_cpu_num; diff --git a/drivers/infiniband/hw/hfi1/init.c b/drivers/infiniband/hw/hfi1/init.c index 71cb9525c074..fbff6b2f00e7 100644 --- a/drivers/infiniband/hw/hfi1/init.c +++ b/drivers/infiniband/hw/hfi1/init.c @@ -845,6 +845,29 @@ wq_error: } /** + * destroy_workqueues - destroy per port workqueues + * @dd: the hfi1_ib device + */ +static void destroy_workqueues(struct hfi1_devdata *dd) +{ + int pidx; + struct hfi1_pportdata *ppd; + + for (pidx = 0; pidx < dd->num_pports; ++pidx) { + ppd = dd->pport + pidx; + + if (ppd->hfi1_wq) { + destroy_workqueue(ppd->hfi1_wq); + ppd->hfi1_wq = NULL; + } + if (ppd->link_wq) { + destroy_workqueue(ppd->link_wq); + ppd->link_wq = NULL; + } + } +} + +/** * enable_general_intr() - Enable the IRQs that will be handled by the * general interrupt handler. * @dd: valid devdata @@ -1117,15 +1140,10 @@ static void shutdown_device(struct hfi1_devdata *dd) * We can't count on interrupts since we are stopping. */ hfi1_quiet_serdes(ppd); - - if (ppd->hfi1_wq) { - destroy_workqueue(ppd->hfi1_wq); - ppd->hfi1_wq = NULL; - } - if (ppd->link_wq) { - destroy_workqueue(ppd->link_wq); - ppd->link_wq = NULL; - } + if (ppd->hfi1_wq) + flush_workqueue(ppd->hfi1_wq); + if (ppd->link_wq) + flush_workqueue(ppd->link_wq); } sdma_exit(dd); } @@ -1489,7 +1507,6 @@ static int __init hfi1_mod_init(void) goto bail_dev; } - hfi1_compute_tid_rdma_flow_wt(); /* * These must be called before the driver is registered with * the PCI subsystem. @@ -1815,6 +1832,7 @@ static void remove_one(struct pci_dev *pdev) * clear dma engines, etc. */ shutdown_device(dd); + destroy_workqueues(dd); stop_timers(dd); diff --git a/drivers/infiniband/hw/hfi1/iowait.c b/drivers/infiniband/hw/hfi1/iowait.c index adb4a1ba921b..5836fe7b2817 100644 --- a/drivers/infiniband/hw/hfi1/iowait.c +++ b/drivers/infiniband/hw/hfi1/iowait.c @@ -81,7 +81,9 @@ void iowait_init(struct iowait *wait, u32 tx_limit, void iowait_cancel_work(struct iowait *w) { cancel_work_sync(&iowait_get_ib_work(w)->iowork); - cancel_work_sync(&iowait_get_tid_work(w)->iowork); + /* Make sure that the iowork for TID RDMA is used */ + if (iowait_get_tid_work(w)->iowork.func) + cancel_work_sync(&iowait_get_tid_work(w)->iowork); } /** diff --git a/drivers/infiniband/hw/hfi1/pcie.c b/drivers/infiniband/hw/hfi1/pcie.c index c96d193bb236..177dbeaa9d14 100644 --- a/drivers/infiniband/hw/hfi1/pcie.c +++ b/drivers/infiniband/hw/hfi1/pcie.c @@ -319,7 +319,9 @@ int pcie_speeds(struct hfi1_devdata *dd) /* * bus->max_bus_speed is set from the bridge's linkcap Max Link Speed */ - if (parent && dd->pcidev->bus->max_bus_speed != PCIE_SPEED_8_0GT) { + if (parent && + (dd->pcidev->bus->max_bus_speed == PCIE_SPEED_2_5GT || + dd->pcidev->bus->max_bus_speed == PCIE_SPEED_5_0GT)) { dd_dev_info(dd, "Parent PCIe bridge does not support Gen3\n"); dd->link_gen3_capable = 0; } diff --git a/drivers/infiniband/hw/hfi1/pio.c b/drivers/infiniband/hw/hfi1/pio.c index 4e5c2d1b8cfa..79126b2b14ab 100644 --- a/drivers/infiniband/hw/hfi1/pio.c +++ b/drivers/infiniband/hw/hfi1/pio.c @@ -1594,9 +1594,8 @@ void hfi1_sc_wantpiobuf_intr(struct send_context *sc, u32 needint) else sc_del_credit_return_intr(sc); trace_hfi1_wantpiointr(sc, needint, sc->credit_ctrl); - if (needint) { + if (needint) sc_return_credits(sc); - } } /** diff --git a/drivers/infiniband/hw/hfi1/qp.c b/drivers/infiniband/hw/hfi1/qp.c index 4e0e9fc0a777..30ba09b510fa 100644 --- a/drivers/infiniband/hw/hfi1/qp.c +++ b/drivers/infiniband/hw/hfi1/qp.c @@ -381,7 +381,10 @@ bool _hfi1_schedule_send(struct rvt_qp *qp) struct hfi1_ibport *ibp = to_iport(qp->ibqp.device, qp->port_num); struct hfi1_pportdata *ppd = ppd_from_ibp(ibp); - struct hfi1_devdata *dd = dd_from_ibdev(qp->ibqp.device); + struct hfi1_devdata *dd = ppd->dd; + + if (dd->flags & HFI1_SHUTDOWN) + return true; return iowait_schedule(&priv->s_iowait, ppd->hfi1_wq, priv->s_sde ? diff --git a/drivers/infiniband/hw/hfi1/rc.c b/drivers/infiniband/hw/hfi1/rc.c index 5c0d90418e8c..0d3e86da2784 100644 --- a/drivers/infiniband/hw/hfi1/rc.c +++ b/drivers/infiniband/hw/hfi1/rc.c @@ -2210,15 +2210,15 @@ int do_rc_ack(struct rvt_qp *qp, u32 aeth, u32 psn, int opcode, if (qp->s_flags & RVT_S_WAIT_RNR) goto bail_stop; rdi = ib_to_rvt(qp->ibqp.device); - if (qp->s_rnr_retry == 0 && - !((rdi->post_parms[wqe->wr.opcode].flags & - RVT_OPERATION_IGN_RNR_CNT) && - qp->s_rnr_retry_cnt == 0)) { - status = IB_WC_RNR_RETRY_EXC_ERR; - goto class_b; + if (!(rdi->post_parms[wqe->wr.opcode].flags & + RVT_OPERATION_IGN_RNR_CNT)) { + if (qp->s_rnr_retry == 0) { + status = IB_WC_RNR_RETRY_EXC_ERR; + goto class_b; + } + if (qp->s_rnr_retry_cnt < 7 && qp->s_rnr_retry_cnt > 0) + qp->s_rnr_retry--; } - if (qp->s_rnr_retry_cnt < 7 && qp->s_rnr_retry_cnt > 0) - qp->s_rnr_retry--; /* * The last valid PSN is the previous PSN. For TID RDMA WRITE diff --git a/drivers/infiniband/hw/hfi1/sdma.c b/drivers/infiniband/hw/hfi1/sdma.c index bc45b3b11f91..c61b6022575e 100644 --- a/drivers/infiniband/hw/hfi1/sdma.c +++ b/drivers/infiniband/hw/hfi1/sdma.c @@ -65,6 +65,7 @@ #define SDMA_DESCQ_CNT 2048 #define SDMA_DESC_INTR 64 #define INVALID_TAIL 0xffff +#define SDMA_PAD max_t(size_t, MAX_16B_PADDING, sizeof(u32)) static uint sdma_descq_cnt = SDMA_DESCQ_CNT; module_param(sdma_descq_cnt, uint, S_IRUGO); @@ -869,14 +870,13 @@ struct sdma_engine *sdma_select_user_engine(struct hfi1_devdata *dd, { struct sdma_rht_node *rht_node; struct sdma_engine *sde = NULL; - const struct cpumask *current_mask = ¤t->cpus_allowed; unsigned long cpu_id; /* * To ensure that always the same sdma engine(s) will be * selected make sure the process is pinned to this CPU only. */ - if (cpumask_weight(current_mask) != 1) + if (current->nr_cpus_allowed != 1) goto out; cpu_id = smp_processor_id(); @@ -1297,7 +1297,7 @@ void sdma_clean(struct hfi1_devdata *dd, size_t num_engines) struct sdma_engine *sde; if (dd->sdma_pad_dma) { - dma_free_coherent(&dd->pcidev->dev, 4, + dma_free_coherent(&dd->pcidev->dev, SDMA_PAD, (void *)dd->sdma_pad_dma, dd->sdma_pad_phys); dd->sdma_pad_dma = NULL; @@ -1492,7 +1492,7 @@ int sdma_init(struct hfi1_devdata *dd, u8 port) } /* Allocate memory for pad */ - dd->sdma_pad_dma = dma_alloc_coherent(&dd->pcidev->dev, sizeof(u32), + dd->sdma_pad_dma = dma_alloc_coherent(&dd->pcidev->dev, SDMA_PAD, &dd->sdma_pad_phys, GFP_KERNEL); if (!dd->sdma_pad_dma) { dd_dev_err(dd, "failed to allocate SendDMA pad memory\n"); diff --git a/drivers/infiniband/hw/hfi1/sysfs.c b/drivers/infiniband/hw/hfi1/sysfs.c index 90f62c4bddba..074ec71772d2 100644 --- a/drivers/infiniband/hw/hfi1/sysfs.c +++ b/drivers/infiniband/hw/hfi1/sysfs.c @@ -674,7 +674,11 @@ int hfi1_create_port_files(struct ib_device *ibdev, u8 port_num, dd_dev_err(dd, "Skipping sc2vl sysfs info, (err %d) port %u\n", ret, port_num); - goto bail; + /* + * Based on the documentation for kobject_init_and_add(), the + * caller should call kobject_put even if this call fails. + */ + goto bail_sc2vl; } kobject_uevent(&ppd->sc2vl_kobj, KOBJ_ADD); @@ -684,7 +688,7 @@ int hfi1_create_port_files(struct ib_device *ibdev, u8 port_num, dd_dev_err(dd, "Skipping sl2sc sysfs info, (err %d) port %u\n", ret, port_num); - goto bail_sc2vl; + goto bail_sl2sc; } kobject_uevent(&ppd->sl2sc_kobj, KOBJ_ADD); @@ -694,7 +698,7 @@ int hfi1_create_port_files(struct ib_device *ibdev, u8 port_num, dd_dev_err(dd, "Skipping vl2mtu sysfs info, (err %d) port %u\n", ret, port_num); - goto bail_sl2sc; + goto bail_vl2mtu; } kobject_uevent(&ppd->vl2mtu_kobj, KOBJ_ADD); @@ -704,7 +708,7 @@ int hfi1_create_port_files(struct ib_device *ibdev, u8 port_num, dd_dev_err(dd, "Skipping Congestion Control sysfs info, (err %d) port %u\n", ret, port_num); - goto bail_vl2mtu; + goto bail_cc; } kobject_uevent(&ppd->pport_cc_kobj, KOBJ_ADD); @@ -742,7 +746,6 @@ bail_sl2sc: kobject_put(&ppd->sl2sc_kobj); bail_sc2vl: kobject_put(&ppd->sc2vl_kobj); -bail: return ret; } @@ -853,8 +856,13 @@ int hfi1_verbs_register_sysfs(struct hfi1_devdata *dd) return 0; bail: - for (i = 0; i < dd->num_sdma; i++) - kobject_del(&dd->per_sdma[i].kobj); + /* + * The function kobject_put() will call kobject_del() if the kobject + * has been added successfully. The sysfs files created under the + * kobject directory will also be removed during the process. + */ + for (; i >= 0; i--) + kobject_put(&dd->per_sdma[i].kobj); return ret; } @@ -867,6 +875,10 @@ void hfi1_verbs_unregister_sysfs(struct hfi1_devdata *dd) struct hfi1_pportdata *ppd; int i; + /* Unwind operations in hfi1_verbs_register_sysfs() */ + for (i = 0; i < dd->num_sdma; i++) + kobject_put(&dd->per_sdma[i].kobj); + for (i = 0; i < dd->num_pports; i++) { ppd = &dd->pport[i]; diff --git a/drivers/infiniband/hw/hfi1/tid_rdma.c b/drivers/infiniband/hw/hfi1/tid_rdma.c index 7e9527ab6d64..b5bc41e10e22 100644 --- a/drivers/infiniband/hw/hfi1/tid_rdma.c +++ b/drivers/infiniband/hw/hfi1/tid_rdma.c @@ -107,8 +107,6 @@ static u32 mask_generation(u32 a) * C - Capcode */ -static u32 tid_rdma_flow_wt; - static void tid_rdma_trigger_resume(struct work_struct *work); static void hfi1_kern_exp_rcv_free_flows(struct tid_rdma_request *req); static int hfi1_kern_exp_rcv_alloc_flows(struct tid_rdma_request *req, @@ -136,6 +134,26 @@ static void update_r_next_psn_fecn(struct hfi1_packet *packet, struct tid_rdma_flow *flow, bool fecn); +static void validate_r_tid_ack(struct hfi1_qp_priv *priv) +{ + if (priv->r_tid_ack == HFI1_QP_WQE_INVALID) + priv->r_tid_ack = priv->r_tid_tail; +} + +static void tid_rdma_schedule_ack(struct rvt_qp *qp) +{ + struct hfi1_qp_priv *priv = qp->priv; + + priv->s_flags |= RVT_S_ACK_PENDING; + hfi1_schedule_tid_send(qp); +} + +static void tid_rdma_trigger_ack(struct rvt_qp *qp) +{ + validate_r_tid_ack(qp->priv); + tid_rdma_schedule_ack(qp); +} + static u64 tid_rdma_opfn_encode(struct tid_rdma_params *p) { return @@ -2730,11 +2748,6 @@ static bool handle_read_kdeth_eflags(struct hfi1_ctxtdata *rcd, diff = cmp_psn(psn, flow->flow_state.r_next_psn); if (diff > 0) { - if (!(qp->r_flags & RVT_R_RDMAR_SEQ)) - restart_tid_rdma_read_req(rcd, - qp, - wqe); - /* Drop the packet.*/ goto s_unlock; } else if (diff < 0) { @@ -3004,10 +3017,7 @@ nak_psn: qpriv->s_nak_state = IB_NAK_PSN_ERROR; /* We are NAK'ing the next expected PSN */ qpriv->s_nak_psn = mask_psn(flow->flow_state.r_next_psn); - qpriv->s_flags |= RVT_S_ACK_PENDING; - if (qpriv->r_tid_ack == HFI1_QP_WQE_INVALID) - qpriv->r_tid_ack = qpriv->r_tid_tail; - hfi1_schedule_tid_send(qp); + tid_rdma_trigger_ack(qp); } goto unlock; } @@ -3370,18 +3380,17 @@ u32 hfi1_build_tid_rdma_write_req(struct rvt_qp *qp, struct rvt_swqe *wqe, return sizeof(ohdr->u.tid_rdma.w_req) / sizeof(u32); } -void hfi1_compute_tid_rdma_flow_wt(void) +static u32 hfi1_compute_tid_rdma_flow_wt(struct rvt_qp *qp) { /* * Heuristic for computing the RNR timeout when waiting on the flow * queue. Rather than a computationaly expensive exact estimate of when * a flow will be available, we assume that if a QP is at position N in * the flow queue it has to wait approximately (N + 1) * (number of - * segments between two sync points), assuming PMTU of 4K. The rationale - * for this is that flows are released and recycled at each sync point. + * segments between two sync points). The rationale for this is that + * flows are released and recycled at each sync point. */ - tid_rdma_flow_wt = MAX_TID_FLOW_PSN * enum_to_mtu(OPA_MTU_4096) / - TID_RDMA_MAX_SEGMENT_SIZE; + return (MAX_TID_FLOW_PSN * qp->pmtu) >> TID_RDMA_SEGMENT_SHIFT; } static u32 position_in_queue(struct hfi1_qp_priv *qpriv, @@ -3504,7 +3513,7 @@ static void hfi1_tid_write_alloc_resources(struct rvt_qp *qp, bool intr_ctx) if (qpriv->flow_state.index >= RXE_NUM_TID_FLOWS) { ret = hfi1_kern_setup_hw_flow(qpriv->rcd, qp); if (ret) { - to_seg = tid_rdma_flow_wt * + to_seg = hfi1_compute_tid_rdma_flow_wt(qp) * position_in_queue(qpriv, &rcd->flow_queue); break; @@ -3525,7 +3534,7 @@ static void hfi1_tid_write_alloc_resources(struct rvt_qp *qp, bool intr_ctx) /* * If overtaking req->acked_tail, send an RNR NAK. Because the * QP is not queued in this case, and the issue can only be - * caused due a delay in scheduling the second leg which we + * caused by a delay in scheduling the second leg which we * cannot estimate, we use a rather arbitrary RNR timeout of * (MAX_FLOWS / 2) segments */ @@ -3533,8 +3542,7 @@ static void hfi1_tid_write_alloc_resources(struct rvt_qp *qp, bool intr_ctx) MAX_FLOWS)) { ret = -EAGAIN; to_seg = MAX_FLOWS >> 1; - qpriv->s_flags |= RVT_S_ACK_PENDING; - hfi1_schedule_tid_send(qp); + tid_rdma_trigger_ack(qp); break; } @@ -4334,8 +4342,7 @@ void hfi1_rc_rcv_tid_rdma_write_data(struct hfi1_packet *packet) trace_hfi1_tid_req_rcv_write_data(qp, 0, e->opcode, e->psn, e->lpsn, req); trace_hfi1_tid_write_rsp_rcv_data(qp); - if (priv->r_tid_ack == HFI1_QP_WQE_INVALID) - priv->r_tid_ack = priv->r_tid_tail; + validate_r_tid_ack(priv); if (opcode == TID_OP(WRITE_DATA_LAST)) { release_rdma_sge_mr(e); @@ -4374,8 +4381,7 @@ void hfi1_rc_rcv_tid_rdma_write_data(struct hfi1_packet *packet) } done: - priv->s_flags |= RVT_S_ACK_PENDING; - hfi1_schedule_tid_send(qp); + tid_rdma_schedule_ack(qp); exit: priv->r_next_psn_kdeth = flow->flow_state.r_next_psn; if (fecn) @@ -4387,10 +4393,7 @@ send_nak: if (!priv->s_nak_state) { priv->s_nak_state = IB_NAK_PSN_ERROR; priv->s_nak_psn = flow->flow_state.r_next_psn; - priv->s_flags |= RVT_S_ACK_PENDING; - if (priv->r_tid_ack == HFI1_QP_WQE_INVALID) - priv->r_tid_ack = priv->r_tid_tail; - hfi1_schedule_tid_send(qp); + tid_rdma_trigger_ack(qp); } goto done; } @@ -4624,6 +4627,15 @@ void hfi1_rc_rcv_tid_rdma_ack(struct hfi1_packet *packet) */ fpsn = full_flow_psn(flow, flow->flow_state.spsn); req->r_ack_psn = psn; + /* + * If resync_psn points to the last flow PSN for a + * segment and the new segment (likely from a new + * request) starts with a new generation number, we + * need to adjust resync_psn accordingly. + */ + if (flow->flow_state.generation != + (resync_psn >> HFI1_KDETH_BTH_SEQ_SHIFT)) + resync_psn = mask_psn(fpsn - 1); flow->resync_npkts += delta_psn(mask_psn(resync_psn + 1), fpsn); /* @@ -4938,8 +4950,7 @@ void hfi1_rc_rcv_tid_rdma_resync(struct hfi1_packet *packet) qpriv->resync = true; /* RESYNC request always gets a TID RDMA ACK. */ qpriv->s_nak_state = 0; - qpriv->s_flags |= RVT_S_ACK_PENDING; - hfi1_schedule_tid_send(qp); + tid_rdma_trigger_ack(qp); bail: if (fecn) qp->s_flags |= RVT_S_ECN; @@ -5389,7 +5400,10 @@ static bool _hfi1_schedule_tid_send(struct rvt_qp *qp) struct hfi1_ibport *ibp = to_iport(qp->ibqp.device, qp->port_num); struct hfi1_pportdata *ppd = ppd_from_ibp(ibp); - struct hfi1_devdata *dd = dd_from_ibdev(qp->ibqp.device); + struct hfi1_devdata *dd = ppd->dd; + + if ((dd->flags & HFI1_SHUTDOWN)) + return true; return iowait_tid_schedule(&priv->s_iowait, ppd->hfi1_wq, priv->s_sde ? diff --git a/drivers/infiniband/hw/hfi1/tid_rdma.h b/drivers/infiniband/hw/hfi1/tid_rdma.h index 1c536185261e..6e82df2190b7 100644 --- a/drivers/infiniband/hw/hfi1/tid_rdma.h +++ b/drivers/infiniband/hw/hfi1/tid_rdma.h @@ -17,6 +17,7 @@ #define TID_RDMA_MIN_SEGMENT_SIZE BIT(18) /* 256 KiB (for now) */ #define TID_RDMA_MAX_SEGMENT_SIZE BIT(18) /* 256 KiB (for now) */ #define TID_RDMA_MAX_PAGES (BIT(18) >> PAGE_SHIFT) +#define TID_RDMA_SEGMENT_SHIFT 18 /* * Bit definitions for priv->s_flags. @@ -274,8 +275,6 @@ u32 hfi1_build_tid_rdma_write_req(struct rvt_qp *qp, struct rvt_swqe *wqe, struct ib_other_headers *ohdr, u32 *bth1, u32 *bth2, u32 *len); -void hfi1_compute_tid_rdma_flow_wt(void); - void hfi1_rc_rcv_tid_rdma_write_req(struct hfi1_packet *packet); u32 hfi1_build_tid_rdma_write_resp(struct rvt_qp *qp, struct rvt_ack_entry *e, diff --git a/drivers/infiniband/hw/hfi1/user_exp_rcv.c b/drivers/infiniband/hw/hfi1/user_exp_rcv.c index 3592a9ec155e..4d732353379d 100644 --- a/drivers/infiniband/hw/hfi1/user_exp_rcv.c +++ b/drivers/infiniband/hw/hfi1/user_exp_rcv.c @@ -90,9 +90,6 @@ int hfi1_user_exp_rcv_init(struct hfi1_filedata *fd, struct hfi1_devdata *dd = uctxt->dd; int ret = 0; - spin_lock_init(&fd->tid_lock); - spin_lock_init(&fd->invalid_lock); - fd->entry_to_rb = kcalloc(uctxt->expected_count, sizeof(struct rb_node *), GFP_KERNEL); @@ -165,10 +162,12 @@ void hfi1_user_exp_rcv_free(struct hfi1_filedata *fd) if (fd->handler) { hfi1_mmu_rb_unregister(fd->handler); } else { + mutex_lock(&uctxt->exp_mutex); if (!EXP_TID_SET_EMPTY(uctxt->tid_full_list)) unlock_exp_tids(uctxt, &uctxt->tid_full_list, fd); if (!EXP_TID_SET_EMPTY(uctxt->tid_used_list)) unlock_exp_tids(uctxt, &uctxt->tid_used_list, fd); + mutex_unlock(&uctxt->exp_mutex); } kfree(fd->invalid_tids); diff --git a/drivers/infiniband/hw/hfi1/user_sdma.c b/drivers/infiniband/hw/hfi1/user_sdma.c index fd754a16475a..a92346e88628 100644 --- a/drivers/infiniband/hw/hfi1/user_sdma.c +++ b/drivers/infiniband/hw/hfi1/user_sdma.c @@ -141,6 +141,7 @@ static int defer_packet_queue( */ xchg(&pq->state, SDMA_PKT_Q_DEFERRED); if (list_empty(&pq->busy.list)) { + pq->busy.lock = &sde->waitlock; iowait_get_priority(&pq->busy); iowait_queue(pkts_sent, &pq->busy, &sde->dmawait); } @@ -155,6 +156,7 @@ static void activate_packet_queue(struct iowait *wait, int reason) { struct hfi1_user_sdma_pkt_q *pq = container_of(wait, struct hfi1_user_sdma_pkt_q, busy); + pq->busy.lock = NULL; xchg(&pq->state, SDMA_PKT_Q_ACTIVE); wake_up(&wait->wait_dma); }; @@ -179,7 +181,6 @@ int hfi1_user_sdma_alloc_queues(struct hfi1_ctxtdata *uctxt, pq = kzalloc(sizeof(*pq), GFP_KERNEL); if (!pq) return -ENOMEM; - pq->dd = dd; pq->ctxt = uctxt->ctxt; pq->subctxt = fd->subctxt; @@ -236,7 +237,7 @@ int hfi1_user_sdma_alloc_queues(struct hfi1_ctxtdata *uctxt, goto pq_mmu_fail; } - fd->pq = pq; + rcu_assign_pointer(fd->pq, pq); fd->cq = cq; return 0; @@ -257,6 +258,21 @@ pq_reqs_nomem: return ret; } +static void flush_pq_iowait(struct hfi1_user_sdma_pkt_q *pq) +{ + unsigned long flags; + seqlock_t *lock = pq->busy.lock; + + if (!lock) + return; + write_seqlock_irqsave(lock, flags); + if (!list_empty(&pq->busy.list)) { + list_del_init(&pq->busy.list); + pq->busy.lock = NULL; + } + write_sequnlock_irqrestore(lock, flags); +} + int hfi1_user_sdma_free_queues(struct hfi1_filedata *fd, struct hfi1_ctxtdata *uctxt) { @@ -264,8 +280,14 @@ int hfi1_user_sdma_free_queues(struct hfi1_filedata *fd, trace_hfi1_sdma_user_free_queues(uctxt->dd, uctxt->ctxt, fd->subctxt); - pq = fd->pq; + spin_lock(&fd->pq_rcu_lock); + pq = srcu_dereference_check(fd->pq, &fd->pq_srcu, + lockdep_is_held(&fd->pq_rcu_lock)); if (pq) { + rcu_assign_pointer(fd->pq, NULL); + spin_unlock(&fd->pq_rcu_lock); + synchronize_srcu(&fd->pq_srcu); + /* at this point there can be no more new requests */ if (pq->handler) hfi1_mmu_rb_unregister(pq->handler); iowait_sdma_drain(&pq->busy); @@ -276,8 +298,10 @@ int hfi1_user_sdma_free_queues(struct hfi1_filedata *fd, kfree(pq->reqs); kfree(pq->req_in_use); kmem_cache_destroy(pq->txreq_cache); + flush_pq_iowait(pq); kfree(pq); - fd->pq = NULL; + } else { + spin_unlock(&fd->pq_rcu_lock); } if (fd->cq) { vfree(fd->cq->comps); @@ -321,7 +345,8 @@ int hfi1_user_sdma_process_request(struct hfi1_filedata *fd, { int ret = 0, i; struct hfi1_ctxtdata *uctxt = fd->uctxt; - struct hfi1_user_sdma_pkt_q *pq = fd->pq; + struct hfi1_user_sdma_pkt_q *pq = + srcu_dereference(fd->pq, &fd->pq_srcu); struct hfi1_user_sdma_comp_q *cq = fd->cq; struct hfi1_devdata *dd = pq->dd; unsigned long idx = 0; @@ -564,10 +589,6 @@ int hfi1_user_sdma_process_request(struct hfi1_filedata *fd, set_comp_state(pq, cq, info.comp_idx, QUEUED, 0); pq->state = SDMA_PKT_Q_ACTIVE; - /* Send the first N packets in the request to buy us some time */ - ret = user_sdma_send_pkts(req, pcount); - if (unlikely(ret < 0 && ret != -EBUSY)) - goto free_req; /* * This is a somewhat blocking send implementation. @@ -580,11 +601,12 @@ int hfi1_user_sdma_process_request(struct hfi1_filedata *fd, if (ret < 0) { if (ret != -EBUSY) goto free_req; - wait_event_interruptible_timeout( + if (wait_event_interruptible_timeout( pq->busy.wait_dma, - (pq->state == SDMA_PKT_Q_ACTIVE), + pq->state == SDMA_PKT_Q_ACTIVE, msecs_to_jiffies( - SDMA_IOWAIT_TIMEOUT)); + SDMA_IOWAIT_TIMEOUT)) <= 0) + flush_pq_iowait(pq); } } *count += idx; diff --git a/drivers/infiniband/hw/hfi1/user_sdma.h b/drivers/infiniband/hw/hfi1/user_sdma.h index 4d8510b0fc38..9972e0e6545e 100644 --- a/drivers/infiniband/hw/hfi1/user_sdma.h +++ b/drivers/infiniband/hw/hfi1/user_sdma.h @@ -110,12 +110,6 @@ enum pkt_q_sdma_state { SDMA_PKT_Q_DEFERRED, }; -/* - * Maximum retry attempts to submit a TX request - * before putting the process to sleep. - */ -#define MAX_DEFER_RETRY_COUNT 1 - #define SDMA_IOWAIT_TIMEOUT 1000 /* in milliseconds */ #define SDMA_DBG(req, fmt, ...) \ diff --git a/drivers/infiniband/hw/hfi1/verbs.c b/drivers/infiniband/hw/hfi1/verbs.c index 7f9c23450579..53dd30fd2d56 100644 --- a/drivers/infiniband/hw/hfi1/verbs.c +++ b/drivers/infiniband/hw/hfi1/verbs.c @@ -147,9 +147,6 @@ static int pio_wait(struct rvt_qp *qp, /* Length of buffer to create verbs txreq cache name */ #define TXREQ_NAME_LEN 24 -/* 16B trailing buffer */ -static const u8 trail_buf[MAX_16B_PADDING]; - static uint wss_threshold = 80; module_param(wss_threshold, uint, S_IRUGO); MODULE_PARM_DESC(wss_threshold, "Percentage (1-100) of LLC to use as a threshold for a cacheless copy"); @@ -518,10 +515,11 @@ static inline void hfi1_handle_packet(struct hfi1_packet *packet, opa_get_lid(packet->dlid, 9B)); if (!mcast) goto drop; + rcu_read_lock(); list_for_each_entry_rcu(p, &mcast->qp_list, list) { packet->qp = p->qp; if (hfi1_do_pkey_check(packet)) - goto drop; + goto unlock_drop; spin_lock_irqsave(&packet->qp->r_lock, flags); packet_handler = qp_ok(packet); if (likely(packet_handler)) @@ -530,6 +528,7 @@ static inline void hfi1_handle_packet(struct hfi1_packet *packet, ibp->rvp.n_pkt_drops++; spin_unlock_irqrestore(&packet->qp->r_lock, flags); } + rcu_read_unlock(); /* * Notify rvt_multicast_detach() if it is waiting for us * to finish. @@ -820,8 +819,8 @@ static int build_verbs_tx_desc( /* add icrc, lt byte, and padding to flit */ if (extra_bytes) - ret = sdma_txadd_kvaddr(sde->dd, &tx->txreq, - (void *)trail_buf, extra_bytes); + ret = sdma_txadd_daddr(sde->dd, &tx->txreq, + sde->dd->sdma_pad_phys, extra_bytes); bail_txadd: return ret; @@ -1089,7 +1088,8 @@ int hfi1_verbs_send_pio(struct rvt_qp *qp, struct hfi1_pkt_state *ps, } /* add icrc, lt byte, and padding to flit */ if (extra_bytes) - seg_pio_copy_mid(pbuf, trail_buf, extra_bytes); + seg_pio_copy_mid(pbuf, ppd->dd->sdma_pad_dma, + extra_bytes); seg_pio_copy_end(pbuf); } |