diff --git a/drivers/infiniband/hw/cxgb3/cxio_hal.c b/drivers/infiniband/hw/cxgb3/cxio_hal.c index 0677fc7dfd51..a28e862f2d68 100644 --- a/drivers/infiniband/hw/cxgb3/cxio_hal.c +++ b/drivers/infiniband/hw/cxgb3/cxio_hal.c @@ -109,7 +109,6 @@ int cxio_hal_cq_op(struct cxio_rdev *rdev_p, struct t3_cq *cq, while (!CQ_VLD_ENTRY(rptr, cq->size_log2, cqe)) { udelay(1); if (i++ > 1000000) { - BUG_ON(1); printk(KERN_ERR "%s: stalled rnic\n", rdev_p->dev_name); return -EIO; @@ -155,7 +154,7 @@ static int cxio_hal_clear_qp_ctx(struct cxio_rdev *rdev_p, u32 qpid) return iwch_cxgb3_ofld_send(rdev_p->t3cdev_p, skb); } -int cxio_create_cq(struct cxio_rdev *rdev_p, struct t3_cq *cq) +int cxio_create_cq(struct cxio_rdev *rdev_p, struct t3_cq *cq, int kernel) { struct rdma_cq_setup setup; int size = (1UL << (cq->size_log2)) * sizeof(struct t3_cqe); @@ -163,12 +162,12 @@ int cxio_create_cq(struct cxio_rdev *rdev_p, struct t3_cq *cq) cq->cqid = cxio_hal_get_cqid(rdev_p->rscp); if (!cq->cqid) return -ENOMEM; - cq->sw_queue = kzalloc(size, GFP_KERNEL); - if (!cq->sw_queue) - return -ENOMEM; - cq->queue = dma_alloc_coherent(&(rdev_p->rnic_info.pdev->dev), - (1UL << (cq->size_log2)) * - sizeof(struct t3_cqe), + if (kernel) { + cq->sw_queue = kzalloc(size, GFP_KERNEL); + if (!cq->sw_queue) + return -ENOMEM; + } + cq->queue = dma_alloc_coherent(&(rdev_p->rnic_info.pdev->dev), size, &(cq->dma_addr), GFP_KERNEL); if (!cq->queue) { kfree(cq->sw_queue); diff --git a/drivers/infiniband/hw/cxgb3/cxio_hal.h b/drivers/infiniband/hw/cxgb3/cxio_hal.h index f3d440cc68f2..073373c2c560 100644 --- a/drivers/infiniband/hw/cxgb3/cxio_hal.h +++ b/drivers/infiniband/hw/cxgb3/cxio_hal.h @@ -53,7 +53,7 @@ #define T3_MAX_PBL_SIZE 256 #define T3_MAX_RQ_SIZE 1024 #define T3_MAX_QP_DEPTH (T3_MAX_RQ_SIZE-1) -#define T3_MAX_CQ_DEPTH 8192 +#define T3_MAX_CQ_DEPTH 262144 #define T3_MAX_NUM_STAG (1<<15) #define T3_MAX_MR_SIZE 0x100000000ULL #define T3_PAGESIZE_MASK 0xffff000 /* 4KB-128MB */ @@ -157,7 +157,7 @@ int cxio_rdev_open(struct cxio_rdev *rdev); void cxio_rdev_close(struct cxio_rdev *rdev); int cxio_hal_cq_op(struct cxio_rdev *rdev, struct t3_cq *cq, enum t3_cq_opcode op, u32 credit); -int cxio_create_cq(struct cxio_rdev *rdev, struct t3_cq *cq); +int cxio_create_cq(struct cxio_rdev *rdev, struct t3_cq *cq, int kernel); int cxio_destroy_cq(struct cxio_rdev *rdev, struct t3_cq *cq); int cxio_resize_cq(struct cxio_rdev *rdev, struct t3_cq *cq); void cxio_release_ucontext(struct cxio_rdev *rdev, struct cxio_ucontext *uctx); diff --git a/drivers/infiniband/hw/cxgb3/cxio_wr.h b/drivers/infiniband/hw/cxgb3/cxio_wr.h index a197a5b7ac7f..15073b2da1c5 100644 --- a/drivers/infiniband/hw/cxgb3/cxio_wr.h +++ b/drivers/infiniband/hw/cxgb3/cxio_wr.h @@ -730,7 +730,22 @@ struct t3_cq { static inline void cxio_set_wq_in_error(struct t3_wq *wq) { - wq->queue->wq_in_err.err = 1; + wq->queue->wq_in_err.err |= 1; +} + +static inline void cxio_disable_wq_db(struct t3_wq *wq) +{ + wq->queue->wq_in_err.err |= 2; +} + +static inline void cxio_enable_wq_db(struct t3_wq *wq) +{ + wq->queue->wq_in_err.err &= ~2; +} + +static inline int cxio_wq_db_enabled(struct t3_wq *wq) +{ + return !(wq->queue->wq_in_err.err & 2); } static inline struct t3_cqe *cxio_next_hw_cqe(struct t3_cq *cq) diff --git a/drivers/infiniband/hw/cxgb3/iwch.c b/drivers/infiniband/hw/cxgb3/iwch.c index b0ea0105ddf6..ee1d8b4d4541 100644 --- a/drivers/infiniband/hw/cxgb3/iwch.c +++ b/drivers/infiniband/hw/cxgb3/iwch.c @@ -65,6 +65,46 @@ struct cxgb3_client t3c_client = { static LIST_HEAD(dev_list); static DEFINE_MUTEX(dev_mutex); +static int disable_qp_db(int id, void *p, void *data) +{ + struct iwch_qp *qhp = p; + + cxio_disable_wq_db(&qhp->wq); + return 0; +} + +static int enable_qp_db(int id, void *p, void *data) +{ + struct iwch_qp *qhp = p; + + if (data) + ring_doorbell(qhp->rhp->rdev.ctrl_qp.doorbell, qhp->wq.qpid); + cxio_enable_wq_db(&qhp->wq); + return 0; +} + +static void disable_dbs(struct iwch_dev *rnicp) +{ + spin_lock_irq(&rnicp->lock); + idr_for_each(&rnicp->qpidr, disable_qp_db, NULL); + spin_unlock_irq(&rnicp->lock); +} + +static void enable_dbs(struct iwch_dev *rnicp, int ring_db) +{ + spin_lock_irq(&rnicp->lock); + idr_for_each(&rnicp->qpidr, enable_qp_db, + (void *)(unsigned long)ring_db); + spin_unlock_irq(&rnicp->lock); +} + +static void iwch_db_drop_task(struct work_struct *work) +{ + struct iwch_dev *rnicp = container_of(work, struct iwch_dev, + db_drop_task.work); + enable_dbs(rnicp, 1); +} + static void rnic_init(struct iwch_dev *rnicp) { PDBG("%s iwch_dev %p\n", __func__, rnicp); @@ -72,6 +112,7 @@ static void rnic_init(struct iwch_dev *rnicp) idr_init(&rnicp->qpidr); idr_init(&rnicp->mmidr); spin_lock_init(&rnicp->lock); + INIT_DELAYED_WORK(&rnicp->db_drop_task, iwch_db_drop_task); rnicp->attr.max_qps = T3_MAX_NUM_QP - 32; rnicp->attr.max_wrs = T3_MAX_QP_DEPTH; @@ -147,6 +188,8 @@ static void close_rnic_dev(struct t3cdev *tdev) mutex_lock(&dev_mutex); list_for_each_entry_safe(dev, tmp, &dev_list, entry) { if (dev->rdev.t3cdev_p == tdev) { + dev->rdev.flags = CXIO_ERROR_FATAL; + cancel_delayed_work_sync(&dev->db_drop_task); list_del(&dev->entry); iwch_unregister_device(dev); cxio_rdev_close(&dev->rdev); @@ -165,7 +208,8 @@ static void iwch_event_handler(struct t3cdev *tdev, u32 evt, u32 port_id) struct cxio_rdev *rdev = tdev->ulp; struct iwch_dev *rnicp; struct ib_event event; - u32 portnum = port_id + 1; + u32 portnum = port_id + 1; + int dispatch = 0; if (!rdev) return; @@ -174,21 +218,49 @@ static void iwch_event_handler(struct t3cdev *tdev, u32 evt, u32 port_id) case OFFLOAD_STATUS_DOWN: { rdev->flags = CXIO_ERROR_FATAL; event.event = IB_EVENT_DEVICE_FATAL; + dispatch = 1; break; } case OFFLOAD_PORT_DOWN: { event.event = IB_EVENT_PORT_ERR; + dispatch = 1; break; } case OFFLOAD_PORT_UP: { event.event = IB_EVENT_PORT_ACTIVE; + dispatch = 1; + break; + } + case OFFLOAD_DB_FULL: { + disable_dbs(rnicp); + break; + } + case OFFLOAD_DB_EMPTY: { + enable_dbs(rnicp, 1); + break; + } + case OFFLOAD_DB_DROP: { + unsigned long delay = 1000; + unsigned short r; + + disable_dbs(rnicp); + get_random_bytes(&r, 2); + delay += r & 1023; + + /* + * delay is between 1000-2023 usecs. + */ + schedule_delayed_work(&rnicp->db_drop_task, + usecs_to_jiffies(delay)); break; } } - event.device = &rnicp->ibdev; - event.element.port_num = portnum; - ib_dispatch_event(&event); + if (dispatch) { + event.device = &rnicp->ibdev; + event.element.port_num = portnum; + ib_dispatch_event(&event); + } return; } diff --git a/drivers/infiniband/hw/cxgb3/iwch.h b/drivers/infiniband/hw/cxgb3/iwch.h index 84735506333f..a1c44578e039 100644 --- a/drivers/infiniband/hw/cxgb3/iwch.h +++ b/drivers/infiniband/hw/cxgb3/iwch.h @@ -36,6 +36,7 @@ #include #include #include +#include #include @@ -110,6 +111,7 @@ struct iwch_dev { struct idr mmidr; spinlock_t lock; struct list_head entry; + struct delayed_work db_drop_task; }; static inline struct iwch_dev *to_iwch_dev(struct ib_device *ibdev) diff --git a/drivers/infiniband/hw/cxgb3/iwch_provider.c b/drivers/infiniband/hw/cxgb3/iwch_provider.c index ed7175549ebd..47b35c6608d2 100644 --- a/drivers/infiniband/hw/cxgb3/iwch_provider.c +++ b/drivers/infiniband/hw/cxgb3/iwch_provider.c @@ -187,7 +187,7 @@ static struct ib_cq *iwch_create_cq(struct ib_device *ibdev, int entries, int ve entries = roundup_pow_of_two(entries); chp->cq.size_log2 = ilog2(entries); - if (cxio_create_cq(&rhp->rdev, &chp->cq)) { + if (cxio_create_cq(&rhp->rdev, &chp->cq, !ucontext)) { kfree(chp); return ERR_PTR(-ENOMEM); } diff --git a/drivers/infiniband/hw/cxgb3/iwch_qp.c b/drivers/infiniband/hw/cxgb3/iwch_qp.c index 3eb8cecf81d7..b4d893de3650 100644 --- a/drivers/infiniband/hw/cxgb3/iwch_qp.c +++ b/drivers/infiniband/hw/cxgb3/iwch_qp.c @@ -452,7 +452,8 @@ int iwch_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr, ++(qhp->wq.sq_wptr); } spin_unlock_irqrestore(&qhp->lock, flag); - ring_doorbell(qhp->wq.doorbell, qhp->wq.qpid); + if (cxio_wq_db_enabled(&qhp->wq)) + ring_doorbell(qhp->wq.doorbell, qhp->wq.qpid); out: if (err) @@ -514,7 +515,8 @@ int iwch_post_receive(struct ib_qp *ibqp, struct ib_recv_wr *wr, num_wrs--; } spin_unlock_irqrestore(&qhp->lock, flag); - ring_doorbell(qhp->wq.doorbell, qhp->wq.qpid); + if (cxio_wq_db_enabled(&qhp->wq)) + ring_doorbell(qhp->wq.doorbell, qhp->wq.qpid); out: if (err) @@ -597,7 +599,8 @@ int iwch_bind_mw(struct ib_qp *qp, ++(qhp->wq.sq_wptr); spin_unlock_irqrestore(&qhp->lock, flag); - ring_doorbell(qhp->wq.doorbell, qhp->wq.qpid); + if (cxio_wq_db_enabled(&qhp->wq)) + ring_doorbell(qhp->wq.doorbell, qhp->wq.qpid); return err; } diff --git a/drivers/net/cxgb3/adapter.h b/drivers/net/cxgb3/adapter.h index 3e8618b4efbc..4cd7f420766a 100644 --- a/drivers/net/cxgb3/adapter.h +++ b/drivers/net/cxgb3/adapter.h @@ -264,6 +264,10 @@ struct adapter { struct work_struct fatal_error_handler_task; struct work_struct link_fault_handler_task; + struct work_struct db_full_task; + struct work_struct db_empty_task; + struct work_struct db_drop_task; + struct dentry *debugfs_root; struct mutex mdio_lock; @@ -335,6 +339,7 @@ int t3_sge_alloc_qset(struct adapter *adapter, unsigned int id, int nports, int t3_get_desc(const struct sge_qset *qs, unsigned int qnum, unsigned int idx, unsigned char *data); irqreturn_t t3_sge_intr_msix(int irq, void *cookie); +extern struct workqueue_struct *cxgb3_wq; int t3_get_edc_fw(struct cphy *phy, int edc_idx, int size); diff --git a/drivers/net/cxgb3/cxgb3_main.c b/drivers/net/cxgb3/cxgb3_main.c index 89bec9c3c141..37945fce7fa5 100644 --- a/drivers/net/cxgb3/cxgb3_main.c +++ b/drivers/net/cxgb3/cxgb3_main.c @@ -45,6 +45,7 @@ #include #include #include +#include #include #include "common.h" @@ -140,7 +141,7 @@ MODULE_PARM_DESC(ofld_disable, "whether to enable offload at init time or not"); * will block keventd as it needs the rtnl lock, and we'll deadlock waiting * for our work to complete. Get our own work queue to solve this. */ -static struct workqueue_struct *cxgb3_wq; +struct workqueue_struct *cxgb3_wq; /** * link_report - show link status and link speed/duplex @@ -590,6 +591,19 @@ static void setup_rss(struct adapter *adap) V_RRCPLCPUSIZE(6) | F_HASHTOEPLITZ, cpus, rspq_map); } +static void ring_dbs(struct adapter *adap) +{ + int i, j; + + for (i = 0; i < SGE_QSETS; i++) { + struct sge_qset *qs = &adap->sge.qs[i]; + + if (qs->adap) + for (j = 0; j < SGE_TXQ_PER_SET; j++) + t3_write_reg(adap, A_SG_KDOORBELL, F_SELEGRCNTX | V_EGRCNTX(qs->txq[j].cntxt_id)); + } +} + static void init_napi(struct adapter *adap) { int i; @@ -2754,6 +2768,42 @@ static void t3_adap_check_task(struct work_struct *work) spin_unlock_irq(&adapter->work_lock); } +static void db_full_task(struct work_struct *work) +{ + struct adapter *adapter = container_of(work, struct adapter, + db_full_task); + + cxgb3_event_notify(&adapter->tdev, OFFLOAD_DB_FULL, 0); +} + +static void db_empty_task(struct work_struct *work) +{ + struct adapter *adapter = container_of(work, struct adapter, + db_empty_task); + + cxgb3_event_notify(&adapter->tdev, OFFLOAD_DB_EMPTY, 0); +} + +static void db_drop_task(struct work_struct *work) +{ + struct adapter *adapter = container_of(work, struct adapter, + db_drop_task); + unsigned long delay = 1000; + unsigned short r; + + cxgb3_event_notify(&adapter->tdev, OFFLOAD_DB_DROP, 0); + + /* + * Sleep a while before ringing the driver qset dbs. + * The delay is between 1000-2023 usecs. + */ + get_random_bytes(&r, 2); + delay += r & 1023; + set_current_state(TASK_UNINTERRUPTIBLE); + schedule_timeout(usecs_to_jiffies(delay)); + ring_dbs(adapter); +} + /* * Processes external (PHY) interrupts in process context. */ @@ -3222,6 +3272,11 @@ static int __devinit init_one(struct pci_dev *pdev, INIT_LIST_HEAD(&adapter->adapter_list); INIT_WORK(&adapter->ext_intr_handler_task, ext_intr_task); INIT_WORK(&adapter->fatal_error_handler_task, fatal_error_task); + + INIT_WORK(&adapter->db_full_task, db_full_task); + INIT_WORK(&adapter->db_empty_task, db_empty_task); + INIT_WORK(&adapter->db_drop_task, db_drop_task); + INIT_DELAYED_WORK(&adapter->adap_check_task, t3_adap_check_task); for (i = 0; i < ai->nports0 + ai->nports1; ++i) { diff --git a/drivers/net/cxgb3/cxgb3_offload.h b/drivers/net/cxgb3/cxgb3_offload.h index 670aa62042da..929c298115ca 100644 --- a/drivers/net/cxgb3/cxgb3_offload.h +++ b/drivers/net/cxgb3/cxgb3_offload.h @@ -73,7 +73,10 @@ enum { OFFLOAD_STATUS_UP, OFFLOAD_STATUS_DOWN, OFFLOAD_PORT_DOWN, - OFFLOAD_PORT_UP + OFFLOAD_PORT_UP, + OFFLOAD_DB_FULL, + OFFLOAD_DB_EMPTY, + OFFLOAD_DB_DROP }; struct cxgb3_client { diff --git a/drivers/net/cxgb3/regs.h b/drivers/net/cxgb3/regs.h index 1b5327b5a965..cb42353c9fdd 100644 --- a/drivers/net/cxgb3/regs.h +++ b/drivers/net/cxgb3/regs.h @@ -254,6 +254,22 @@ #define V_LOPIODRBDROPERR(x) ((x) << S_LOPIODRBDROPERR) #define F_LOPIODRBDROPERR V_LOPIODRBDROPERR(1U) +#define S_HIPRIORITYDBFULL 7 +#define V_HIPRIORITYDBFULL(x) ((x) << S_HIPRIORITYDBFULL) +#define F_HIPRIORITYDBFULL V_HIPRIORITYDBFULL(1U) + +#define S_HIPRIORITYDBEMPTY 6 +#define V_HIPRIORITYDBEMPTY(x) ((x) << S_HIPRIORITYDBEMPTY) +#define F_HIPRIORITYDBEMPTY V_HIPRIORITYDBEMPTY(1U) + +#define S_LOPRIORITYDBFULL 5 +#define V_LOPRIORITYDBFULL(x) ((x) << S_LOPRIORITYDBFULL) +#define F_LOPRIORITYDBFULL V_LOPRIORITYDBFULL(1U) + +#define S_LOPRIORITYDBEMPTY 4 +#define V_LOPRIORITYDBEMPTY(x) ((x) << S_LOPRIORITYDBEMPTY) +#define F_LOPRIORITYDBEMPTY V_LOPRIORITYDBEMPTY(1U) + #define S_RSPQDISABLED 3 #define V_RSPQDISABLED(x) ((x) << S_RSPQDISABLED) #define F_RSPQDISABLED V_RSPQDISABLED(1U) diff --git a/drivers/net/cxgb3/sge.c b/drivers/net/cxgb3/sge.c index 318a018ca7c5..9b434461c4f1 100644 --- a/drivers/net/cxgb3/sge.c +++ b/drivers/net/cxgb3/sge.c @@ -42,6 +42,7 @@ #include "sge_defs.h" #include "t3_cpl.h" #include "firmware_exports.h" +#include "cxgb3_offload.h" #define USE_GTS 0 @@ -2833,8 +2834,13 @@ void t3_sge_err_intr_handler(struct adapter *adapter) } if (status & (F_HIPIODRBDROPERR | F_LOPIODRBDROPERR)) - CH_ALERT(adapter, "SGE dropped %s priority doorbell\n", - status & F_HIPIODRBDROPERR ? "high" : "lo"); + queue_work(cxgb3_wq, &adapter->db_drop_task); + + if (status & (F_HIPRIORITYDBFULL | F_LOPRIORITYDBFULL)) + queue_work(cxgb3_wq, &adapter->db_full_task); + + if (status & (F_HIPRIORITYDBEMPTY | F_LOPRIORITYDBEMPTY)) + queue_work(cxgb3_wq, &adapter->db_empty_task); t3_write_reg(adapter, A_SG_INT_CAUSE, status); if (status & SGE_FATALERR) diff --git a/drivers/net/cxgb3/t3_hw.c b/drivers/net/cxgb3/t3_hw.c index 032cfe065570..c38fc717a0d1 100644 --- a/drivers/net/cxgb3/t3_hw.c +++ b/drivers/net/cxgb3/t3_hw.c @@ -1432,7 +1432,10 @@ static int t3_handle_intr_status(struct adapter *adapter, unsigned int reg, F_IRPARITYERROR | V_ITPARITYERROR(M_ITPARITYERROR) | \ V_FLPARITYERROR(M_FLPARITYERROR) | F_LODRBPARITYERROR | \ F_HIDRBPARITYERROR | F_LORCQPARITYERROR | \ - F_HIRCQPARITYERROR) + F_HIRCQPARITYERROR | F_LOPRIORITYDBFULL | \ + F_HIPRIORITYDBFULL | F_LOPRIORITYDBEMPTY | \ + F_HIPRIORITYDBEMPTY | F_HIPIODRBDROPERR | \ + F_LOPIODRBDROPERR) #define MC5_INTR_MASK (F_PARITYERR | F_ACTRGNFULL | F_UNKNOWNCMD | \ F_REQQPARERR | F_DISPQPARERR | F_DELACTEMPTY | \ F_NFASRCHFAIL)