From 3bba11e5c47dfc1d381a1ece26464fb7eea2d79c Mon Sep 17 00:00:00 2001 From: Ali Ayoub Date: Tue, 13 Nov 2007 15:26:57 -0800 Subject: [PATCH 1/7] mlx4_core: Fix possible bad free in mlx4_buf_free() When mlx4_buf_free() is called from the error path of mlx4_buf_alloc(), it may be passed a buffer structure that does not have all pages filled in. Add a check for NULL to mlx4_buf_free() so we avoid passing NULL to dma_free_coherent() (which will crash). Signed-off-by: Ali Ayoub Signed-off-by: Jack Morgenstein Signed-off-by: Roland Dreier --- drivers/net/mlx4/alloc.c | 7 ++++--- 1 file changed, 4 insertions(+), 3 deletions(-) diff --git a/drivers/net/mlx4/alloc.c b/drivers/net/mlx4/alloc.c index f8d63d39f592..b226e019bc8b 100644 --- a/drivers/net/mlx4/alloc.c +++ b/drivers/net/mlx4/alloc.c @@ -171,9 +171,10 @@ void mlx4_buf_free(struct mlx4_dev *dev, int size, struct mlx4_buf *buf) buf->u.direct.map); else { for (i = 0; i < buf->nbufs; ++i) - dma_free_coherent(&dev->pdev->dev, PAGE_SIZE, - buf->u.page_list[i].buf, - buf->u.page_list[i].map); + if (buf->u.page_list[i].buf) + dma_free_coherent(&dev->pdev->dev, PAGE_SIZE, + buf->u.page_list[i].buf, + buf->u.page_list[i].map); kfree(buf->u.page_list); } } From a6e7550d8f73d6b75c20afff321f0f06fe144775 Mon Sep 17 00:00:00 2001 From: Ralph Campbell Date: Wed, 24 Oct 2007 15:49:39 -0700 Subject: [PATCH 2/7] IB/ipath: Fix memory leak in ipath_resize_cq() if copy_to_user() fails Signed-off-by: Ralph Campbell Signed-off-by: Patrick Marchand Latifi Signed-off-by: Roland Dreier --- drivers/infiniband/hw/ipath/ipath_cq.c | 11 +++++++---- 1 file changed, 7 insertions(+), 4 deletions(-) diff --git a/drivers/infiniband/hw/ipath/ipath_cq.c b/drivers/infiniband/hw/ipath/ipath_cq.c index 645ed71fd797..08d8ae148cd0 100644 --- a/drivers/infiniband/hw/ipath/ipath_cq.c +++ b/drivers/infiniband/hw/ipath/ipath_cq.c @@ -404,7 +404,7 @@ int ipath_resize_cq(struct ib_cq *ibcq, int cqe, struct ib_udata *udata) ret = ib_copy_to_udata(udata, &offset, sizeof(offset)); if (ret) - goto bail; + goto bail_free; } spin_lock_irq(&cq->lock); @@ -424,10 +424,8 @@ int ipath_resize_cq(struct ib_cq *ibcq, int cqe, struct ib_udata *udata) else n = head - tail; if (unlikely((u32)cqe < n)) { - spin_unlock_irq(&cq->lock); - vfree(wc); ret = -EOVERFLOW; - goto bail; + goto bail_unlock; } for (n = 0; tail != head; n++) { if (cq->ip) @@ -459,7 +457,12 @@ int ipath_resize_cq(struct ib_cq *ibcq, int cqe, struct ib_udata *udata) } ret = 0; + goto bail; +bail_unlock: + spin_unlock_irq(&cq->lock); +bail_free: + vfree(wc); bail: return ret; } From f4ad1bcc4425a772ea584e1f24abadc64c2b839f Mon Sep 17 00:00:00 2001 From: Ralph Campbell Date: Fri, 26 Oct 2007 08:02:39 -0700 Subject: [PATCH 3/7] IB/ipath: Fix race with ACK retry timeout list management When an ACK is received, the QP is removed from the timeout list and then if there are still pending send WQEs, the QP is put back on the timeout list. It is possible that another post send has put the QP on the timeout list thus, a check needs to be made before trying to do it again or the list is corrupted. Signed-off-by: Ralph Campbell Signed-off-by: Roland Dreier --- drivers/infiniband/hw/ipath/ipath_rc.c | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-) diff --git a/drivers/infiniband/hw/ipath/ipath_rc.c b/drivers/infiniband/hw/ipath/ipath_rc.c index 5c29b2bfea17..120a61b03bc4 100644 --- a/drivers/infiniband/hw/ipath/ipath_rc.c +++ b/drivers/infiniband/hw/ipath/ipath_rc.c @@ -959,8 +959,9 @@ static int do_rc_ack(struct ipath_qp *qp, u32 aeth, u32 psn, int opcode, /* If this is a partial ACK, reset the retransmit timer. */ if (qp->s_last != qp->s_tail) { spin_lock(&dev->pending_lock); - list_add_tail(&qp->timerwait, - &dev->pending[dev->pending_index]); + if (list_empty(&qp->timerwait)) + list_add_tail(&qp->timerwait, + &dev->pending[dev->pending_index]); spin_unlock(&dev->pending_lock); /* * If we get a partial ACK for a resent operation, From 40ebb5615e2e069d3b8936b894ceff436c914003 Mon Sep 17 00:00:00 2001 From: Joachim Fenkes Date: Fri, 2 Nov 2007 15:33:51 +0200 Subject: [PATCH 4/7] IB/ehca: Return physical link information in query_port() Newer firmware versions return physical port information to the partition, so hand that information to the consumer if it's present. Signed-off-by: Joachim Fenkes Signed-off-by: Roland Dreier --- drivers/infiniband/hw/ehca/ehca_hca.c | 20 ++++++++++++++------ drivers/infiniband/hw/ehca/hipz_hw.h | 6 +++++- 2 files changed, 19 insertions(+), 7 deletions(-) diff --git a/drivers/infiniband/hw/ehca/ehca_hca.c b/drivers/infiniband/hw/ehca/ehca_hca.c index 15806d140461..5bd7b591987e 100644 --- a/drivers/infiniband/hw/ehca/ehca_hca.c +++ b/drivers/infiniband/hw/ehca/ehca_hca.c @@ -151,7 +151,6 @@ int ehca_query_port(struct ib_device *ibdev, } memset(props, 0, sizeof(struct ib_port_attr)); - props->state = rblock->state; switch (rblock->max_mtu) { case 0x1: @@ -188,11 +187,20 @@ int ehca_query_port(struct ib_device *ibdev, props->subnet_timeout = rblock->subnet_timeout; props->init_type_reply = rblock->init_type_reply; - props->active_width = IB_WIDTH_12X; - props->active_speed = 0x1; - - /* at the moment (logical) link state is always LINK_UP */ - props->phys_state = 0x5; + if (rblock->state && rblock->phys_width) { + props->phys_state = rblock->phys_pstate; + props->state = rblock->phys_state; + props->active_width = rblock->phys_width; + props->active_speed = rblock->phys_speed; + } else { + /* old firmware releases don't report physical + * port info, so use default values + */ + props->phys_state = 5; + props->state = rblock->state; + props->active_width = IB_WIDTH_12X; + props->active_speed = 0x1; + } query_port1: ehca_free_fw_ctrlblock(rblock); diff --git a/drivers/infiniband/hw/ehca/hipz_hw.h b/drivers/infiniband/hw/ehca/hipz_hw.h index d9739e554515..485b8400359e 100644 --- a/drivers/infiniband/hw/ehca/hipz_hw.h +++ b/drivers/infiniband/hw/ehca/hipz_hw.h @@ -402,7 +402,11 @@ struct hipz_query_port { u64 max_msg_sz; u32 max_mtu; u32 vl_cap; - u8 reserved2[1900]; + u32 phys_pstate; + u32 phys_state; + u32 phys_speed; + u32 phys_width; + u8 reserved2[1884]; u64 guid_entries[255]; } __attribute__ ((packed)); From 51aaa54eb9e9f01878aa5d62277fd156e458dfe1 Mon Sep 17 00:00:00 2001 From: Joachim Fenkes Date: Fri, 2 Nov 2007 15:41:49 +0200 Subject: [PATCH 5/7] IB/ehca: Fix static rate calculation The IPD (inter-packet delay) formula was a little off and assumed a fixed physical link rate; fix the formula and query the actual physical link rate, now that we can get it. Also, refactor the calculation into a common function ehca_calc_ipd() and use that instead of duplicating code. Signed-off-by: Joachim Fenkes Signed-off-by: Roland Dreier --- drivers/infiniband/hw/ehca/ehca_av.c | 48 ++++++++++++++++++----- drivers/infiniband/hw/ehca/ehca_classes.h | 1 - drivers/infiniband/hw/ehca/ehca_iverbs.h | 3 ++ drivers/infiniband/hw/ehca/ehca_main.c | 3 -- drivers/infiniband/hw/ehca/ehca_qp.c | 29 ++++++-------- 5 files changed, 54 insertions(+), 30 deletions(-) diff --git a/drivers/infiniband/hw/ehca/ehca_av.c b/drivers/infiniband/hw/ehca/ehca_av.c index 97d108634c58..453eb995c1d4 100644 --- a/drivers/infiniband/hw/ehca/ehca_av.c +++ b/drivers/infiniband/hw/ehca/ehca_av.c @@ -50,6 +50,38 @@ static struct kmem_cache *av_cache; +int ehca_calc_ipd(struct ehca_shca *shca, int port, + enum ib_rate path_rate, u32 *ipd) +{ + int path = ib_rate_to_mult(path_rate); + int link, ret; + struct ib_port_attr pa; + + if (path_rate == IB_RATE_PORT_CURRENT) { + *ipd = 0; + return 0; + } + + if (unlikely(path < 0)) { + ehca_err(&shca->ib_device, "Invalid static rate! path_rate=%x", + path_rate); + return -EINVAL; + } + + ret = ehca_query_port(&shca->ib_device, port, &pa); + if (unlikely(ret < 0)) { + ehca_err(&shca->ib_device, "Failed to query port ret=%i", ret); + return ret; + } + + link = ib_width_enum_to_int(pa.active_width) * pa.active_speed; + + /* IPD = round((link / path) - 1) */ + *ipd = ((link + (path >> 1)) / path) - 1; + + return 0; +} + struct ib_ah *ehca_create_ah(struct ib_pd *pd, struct ib_ah_attr *ah_attr) { int ret; @@ -69,15 +101,13 @@ struct ib_ah *ehca_create_ah(struct ib_pd *pd, struct ib_ah_attr *ah_attr) av->av.slid_path_bits = ah_attr->src_path_bits; if (ehca_static_rate < 0) { - int ah_mult = ib_rate_to_mult(ah_attr->static_rate); - int ehca_mult = - ib_rate_to_mult(shca->sport[ah_attr->port_num].rate ); - - if (ah_mult >= ehca_mult) - av->av.ipd = 0; - else - av->av.ipd = (ah_mult > 0) ? - ((ehca_mult - 1) / ah_mult) : 0; + u32 ipd; + if (ehca_calc_ipd(shca, ah_attr->port_num, + ah_attr->static_rate, &ipd)) { + ret = -EINVAL; + goto create_ah_exit1; + } + av->av.ipd = ipd; } else av->av.ipd = ehca_static_rate; diff --git a/drivers/infiniband/hw/ehca/ehca_classes.h b/drivers/infiniband/hw/ehca/ehca_classes.h index 2d660ae189e5..87f12d4312a7 100644 --- a/drivers/infiniband/hw/ehca/ehca_classes.h +++ b/drivers/infiniband/hw/ehca/ehca_classes.h @@ -95,7 +95,6 @@ struct ehca_sma_attr { struct ehca_sport { struct ib_cq *ibcq_aqp1; struct ib_qp *ibqp_aqp1; - enum ib_rate rate; enum ib_port_state port_state; struct ehca_sma_attr saved_attr; }; diff --git a/drivers/infiniband/hw/ehca/ehca_iverbs.h b/drivers/infiniband/hw/ehca/ehca_iverbs.h index dce503bb7d6b..5485799cdc8d 100644 --- a/drivers/infiniband/hw/ehca/ehca_iverbs.h +++ b/drivers/infiniband/hw/ehca/ehca_iverbs.h @@ -189,6 +189,9 @@ int ehca_mmap(struct ib_ucontext *context, struct vm_area_struct *vma); void ehca_poll_eqs(unsigned long data); +int ehca_calc_ipd(struct ehca_shca *shca, int port, + enum ib_rate path_rate, u32 *ipd); + #ifdef CONFIG_PPC_64K_PAGES void *ehca_alloc_fw_ctrlblock(gfp_t flags); void ehca_free_fw_ctrlblock(void *ptr); diff --git a/drivers/infiniband/hw/ehca/ehca_main.c b/drivers/infiniband/hw/ehca/ehca_main.c index c6cd38c5321f..90d4334179bf 100644 --- a/drivers/infiniband/hw/ehca/ehca_main.c +++ b/drivers/infiniband/hw/ehca/ehca_main.c @@ -327,9 +327,6 @@ static int ehca_sense_attributes(struct ehca_shca *shca) shca->hw_level = ehca_hw_level; ehca_gen_dbg(" ... hardware level=%x", shca->hw_level); - shca->sport[0].rate = IB_RATE_30_GBPS; - shca->sport[1].rate = IB_RATE_30_GBPS; - shca->hca_cap = rblock->hca_cap_indicators; ehca_gen_dbg(" ... HCA capabilities:"); for (i = 0; i < ARRAY_SIZE(hca_cap_descr); i++) diff --git a/drivers/infiniband/hw/ehca/ehca_qp.c b/drivers/infiniband/hw/ehca/ehca_qp.c index de182648b282..2e3e6547cb78 100644 --- a/drivers/infiniband/hw/ehca/ehca_qp.c +++ b/drivers/infiniband/hw/ehca/ehca_qp.c @@ -1196,10 +1196,6 @@ static int internal_modify_qp(struct ib_qp *ibqp, update_mask |= EHCA_BMASK_SET(MQPCB_MASK_QKEY, 1); } if (attr_mask & IB_QP_AV) { - int ah_mult = ib_rate_to_mult(attr->ah_attr.static_rate); - int ehca_mult = ib_rate_to_mult(shca->sport[my_qp-> - init_attr.port_num].rate); - mqpcb->dlid = attr->ah_attr.dlid; update_mask |= EHCA_BMASK_SET(MQPCB_MASK_DLID, 1); mqpcb->source_path_bits = attr->ah_attr.src_path_bits; @@ -1207,11 +1203,12 @@ static int internal_modify_qp(struct ib_qp *ibqp, mqpcb->service_level = attr->ah_attr.sl; update_mask |= EHCA_BMASK_SET(MQPCB_MASK_SERVICE_LEVEL, 1); - if (ah_mult < ehca_mult) - mqpcb->max_static_rate = (ah_mult > 0) ? - ((ehca_mult - 1) / ah_mult) : 0; - else - mqpcb->max_static_rate = 0; + if (ehca_calc_ipd(shca, my_qp->init_attr.port_num, + attr->ah_attr.static_rate, + &mqpcb->max_static_rate)) { + ret = -EINVAL; + goto modify_qp_exit2; + } update_mask |= EHCA_BMASK_SET(MQPCB_MASK_MAX_STATIC_RATE, 1); /* @@ -1280,10 +1277,6 @@ static int internal_modify_qp(struct ib_qp *ibqp, (MQPCB_MASK_RDMA_ATOMIC_OUTST_DEST_QP, 1); } if (attr_mask & IB_QP_ALT_PATH) { - int ah_mult = ib_rate_to_mult(attr->alt_ah_attr.static_rate); - int ehca_mult = ib_rate_to_mult( - shca->sport[my_qp->init_attr.port_num].rate); - if (attr->alt_port_num < 1 || attr->alt_port_num > shca->num_ports) { ret = -EINVAL; @@ -1309,10 +1302,12 @@ static int internal_modify_qp(struct ib_qp *ibqp, mqpcb->source_path_bits_al = attr->alt_ah_attr.src_path_bits; mqpcb->service_level_al = attr->alt_ah_attr.sl; - if (ah_mult > 0 && ah_mult < ehca_mult) - mqpcb->max_static_rate_al = (ehca_mult - 1) / ah_mult; - else - mqpcb->max_static_rate_al = 0; + if (ehca_calc_ipd(shca, my_qp->init_attr.port_num, + attr->alt_ah_attr.static_rate, + &mqpcb->max_static_rate_al)) { + ret = -EINVAL; + goto modify_qp_exit2; + } /* OpenIB doesn't support alternate retry counts - copy them */ mqpcb->retry_count_al = mqpcb->retry_count; From 9a7666494bac60b99d2bd7d904bd22e8c9b1e3f7 Mon Sep 17 00:00:00 2001 From: Steve Wise Date: Fri, 9 Nov 2007 09:21:58 -0600 Subject: [PATCH 6/7] RDMA/cxgb3: Set the max_qp_init_rd_atom attribute in query_device The device attribute max_qp_init_rd_atom is not getting set in cxgb3's query_device method. Version 1.0.4 of librdmacm now validates the user's requested initiator and responder resources against the max supported by the device. Since iw_cxgb3 wasn't setting this attribute (and it defaulted to 0), all rdma_connect()s fail if there are initiator resources requested by the app. Fix this by setting the correct value in iwch_query_device(). Signed-off-by: Steve Wise Signed-off-by: Roland Dreier --- drivers/infiniband/hw/cxgb3/iwch_provider.c | 1 + 1 file changed, 1 insertion(+) diff --git a/drivers/infiniband/hw/cxgb3/iwch_provider.c b/drivers/infiniband/hw/cxgb3/iwch_provider.c index f0c777589374..b5436ca92e68 100644 --- a/drivers/infiniband/hw/cxgb3/iwch_provider.c +++ b/drivers/infiniband/hw/cxgb3/iwch_provider.c @@ -1000,6 +1000,7 @@ static int iwch_query_device(struct ib_device *ibdev, props->max_sge = dev->attr.max_sge_per_wr; props->max_sge_rd = 1; props->max_qp_rd_atom = dev->attr.max_rdma_reads_per_qp; + props->max_qp_init_rd_atom = dev->attr.max_rdma_reads_per_qp; props->max_cq = dev->attr.max_cqs; props->max_cqe = dev->attr.max_cqes_per_cq; props->max_mr = dev->attr.max_mem_regs; From e383d19e90cfbbf8e00512d44194ce175b3f60a2 Mon Sep 17 00:00:00 2001 From: Jack Morgenstein Date: Wed, 14 Nov 2007 16:33:27 +0200 Subject: [PATCH 7/7] mlx4_core: Fix thinko in QP destroy (incorrect bitmap_free) Fix thinko in commit eaf559bf ("mlx4_core: Don't free special QPs in QP number bitmap"). The old commit had the logic exactly backwards and ended up freeing *only* special QPs, which not only left the original bug in place but also introduced the problem that the QP number bitmap would get full after a while. Found by Dotan Barak of Mellanox. Signed-off-by: Jack Morgenstein Signed-off-by: Roland Dreier --- drivers/net/mlx4/qp.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/drivers/net/mlx4/qp.c b/drivers/net/mlx4/qp.c index cc4b1be18219..42b47639c81c 100644 --- a/drivers/net/mlx4/qp.c +++ b/drivers/net/mlx4/qp.c @@ -240,7 +240,7 @@ void mlx4_qp_free(struct mlx4_dev *dev, struct mlx4_qp *qp) mlx4_table_put(dev, &qp_table->auxc_table, qp->qpn); mlx4_table_put(dev, &qp_table->qp_table, qp->qpn); - if (qp->qpn < dev->caps.sqp_start + 8) + if (qp->qpn >= dev->caps.sqp_start + 8) mlx4_bitmap_free(&qp_table->bitmap, qp->qpn); } EXPORT_SYMBOL_GPL(mlx4_qp_free);