Date: Wed, 24 Jun 2026 19:51:17 +0000 From: John Baldwin <jhb@FreeBSD.org> To: src-committers@FreeBSD.org, dev-commits-src-all@FreeBSD.org, dev-commits-src-main@FreeBSD.org Subject: git: cd567a22e28a - main - OFED: Various changes from Linux 4.18 Message-ID: <6a3c3535.43de0.3d971eb8@gitrepo.freebsd.org>
index | next in thread | raw e-mail
The branch main has been updated by jhb: URL: https://cgit.FreeBSD.org/src/commit/?id=cd567a22e28a68aebf12cc320d83648b5b68c274 commit cd567a22e28a68aebf12cc320d83648b5b68c274 Author: John Baldwin <jhb@FreeBSD.org> AuthorDate: 2026-06-24 19:30:56 +0000 Commit: John Baldwin <jhb@FreeBSD.org> CommitDate: 2026-06-24 19:30:56 +0000 OFED: Various changes from Linux 4.18 This contains changes from the following Linux commits: e822ff213fe6 IB/cm: Store and restore ah_attr during CM message processing 0e225dcb7681 IB/cm: Store and restore ah_attr during LAP msg processing a5c57d327272 IB/cm: Avoid AV ah_attr overwriting during LAP message handling 724631a9c6e9 IB/core: Introduce and use rdma_gid_table() 25e62655c793 IB/core: Reduce the places that use zgid 08bb558ac11a IB/core: Make testing MR flags for writability a static inline function Commit 0e225dcb7681 was partially merged previously as cm_init_av_for_lap() already existed in FreeBSD matching the version from the Linux commit. Tested by: Wafa Hamzah <wafah@nvidia.com> (mlx5_ib) Tested by: John Baldwin <jhb@FreeBSD.org> (iw_cxgbe) Sponsored by: Chelsio Communications --- sys/dev/mlx4/mlx4_ib/mlx4_ib_main.c | 5 ++- sys/dev/mlx4/mlx4_ib/mlx4_ib_qp.c | 2 +- sys/ofed/drivers/infiniband/core/ib_cache.c | 44 ++++++++++++++-------- sys/ofed/drivers/infiniband/core/ib_cm.c | 35 +++++++++++------ .../drivers/infiniband/core/ib_roce_gid_mgmt.c | 2 +- sys/ofed/drivers/infiniband/core/ib_umem.c | 11 +----- sys/ofed/include/rdma/ib_cache.h | 1 + 7 files changed, 60 insertions(+), 40 deletions(-) diff --git a/sys/dev/mlx4/mlx4_ib/mlx4_ib_main.c b/sys/dev/mlx4/mlx4_ib/mlx4_ib_main.c index 9f51bac12d51..e7375d422abf 100644 --- a/sys/dev/mlx4/mlx4_ib/mlx4_ib_main.c +++ b/sys/dev/mlx4/mlx4_ib/mlx4_ib_main.c @@ -282,7 +282,7 @@ static int mlx4_ib_add_gid(const union ib_gid *gid, found = i; break; } - if (free < 0 && !memcmp(&port_gid_table->gids[i].gid, &zgid, sizeof(*gid))) + if (free < 0 && rdma_is_zero_gid(&port_gid_table->gids[i].gid)) free = i; /* HW has space */ } @@ -351,7 +351,8 @@ static int mlx4_ib_del_gid(const struct ib_gid_attr *attr, void **context) if (!ctx->refcount) { unsigned int real_index = ctx->real_index; - memcpy(&port_gid_table->gids[real_index].gid, &zgid, sizeof(zgid)); + memset(&port_gid_table->gids[real_index].gid, 0, + sizeof(port_gid_table->gids[real_index].gid)); kfree(port_gid_table->gids[real_index].ctx); port_gid_table->gids[real_index].ctx = NULL; hw_update = 1; diff --git a/sys/dev/mlx4/mlx4_ib/mlx4_ib_qp.c b/sys/dev/mlx4/mlx4_ib/mlx4_ib_qp.c index 42606907a2e3..985980d206fe 100644 --- a/sys/dev/mlx4/mlx4_ib/mlx4_ib_qp.c +++ b/sys/dev/mlx4/mlx4_ib/mlx4_ib_qp.c @@ -2435,7 +2435,7 @@ static int fill_gid_by_hw_index(struct mlx4_ib_dev *ibdev, u8 port_num, memcpy(gid, &port_gid_table->gids[index].gid, sizeof(*gid)); *gid_type = port_gid_table->gids[index].gid_type; spin_unlock_irqrestore(&iboe->lock, flags); - if (!memcmp(gid, &zgid, sizeof(*gid))) + if (rdma_is_zero_gid(gid)) return -ENOENT; return 0; diff --git a/sys/ofed/drivers/infiniband/core/ib_cache.c b/sys/ofed/drivers/infiniband/core/ib_cache.c index 1b7f0dfd4761..bcc0a4315ee7 100644 --- a/sys/ofed/drivers/infiniband/core/ib_cache.c +++ b/sys/ofed/drivers/infiniband/core/ib_cache.c @@ -128,6 +128,16 @@ const char *ib_cache_gid_type_str(enum ib_gid_type gid_type) } EXPORT_SYMBOL(ib_cache_gid_type_str); +/** rdma_is_zero_gid - Check if given GID is zero or not. + * @gid: GID to check + * Returns true if given GID is zero, returns false otherwise. + */ +bool rdma_is_zero_gid(const union ib_gid *gid) +{ + return !memcmp(gid, &zgid, sizeof(*gid)); +} +EXPORT_SYMBOL(rdma_is_zero_gid); + int ib_cache_gid_parse_type_str(const char *buf) { unsigned int i; @@ -152,6 +162,11 @@ int ib_cache_gid_parse_type_str(const char *buf) } EXPORT_SYMBOL(ib_cache_gid_parse_type_str); +static struct ib_gid_table *rdma_gid_table(struct ib_device *device, u8 port) +{ + return device->cache.ports[port - rdma_start_port(device)].gid; +} + static void del_roce_gid(struct ib_device *device, u8 port_num, struct ib_gid_table *table, int ix) { @@ -234,7 +249,7 @@ static int add_modify_gid(struct ib_gid_table *table, * So ignore such behavior for IB link layer and don't * fail the call, but don't add such entry to GID cache. */ - if (!memcmp(gid, &zgid, sizeof(*gid))) + if (rdma_is_zero_gid(gid)) return 0; } @@ -267,7 +282,7 @@ static void del_gid(struct ib_device *ib_dev, u8 port, if (rdma_protocol_roce(ib_dev, port)) del_roce_gid(ib_dev, port, table, ix); - memcpy(&table->data_vec[ix].gid, &zgid, sizeof(zgid)); + memset(&table->data_vec[ix].gid, 0, sizeof(table->data_vec[ix].gid)); memset(&table->data_vec[ix].attr, 0, sizeof(table->data_vec[ix].attr)); table->data_vec[ix].context = NULL; } @@ -380,10 +395,10 @@ static int __ib_cache_gid_add(struct ib_device *ib_dev, u8 port, * IB spec version 1.3 section 4.1.1 point (6) and * section 12.7.10 and section 12.7.20 */ - if (!memcmp(gid, &zgid, sizeof(*gid))) + if (rdma_is_zero_gid(gid)) return -EINVAL; - table = ib_dev->cache.ports[port - rdma_start_port(ib_dev)].gid; + table = rdma_gid_table(ib_dev, port); mutex_lock(&table->lock); @@ -433,7 +448,7 @@ _ib_cache_gid_del(struct ib_device *ib_dev, u8 port, int ret = 0; int ix; - table = ib_dev->cache.ports[port - rdma_start_port(ib_dev)].gid; + table = rdma_gid_table(ib_dev, port); mutex_lock(&table->lock); @@ -472,7 +487,7 @@ int ib_cache_gid_del_all_netdev_gids(struct ib_device *ib_dev, u8 port, int ix; bool deleted = false; - table = ib_dev->cache.ports[port - rdma_start_port(ib_dev)].gid; + table = rdma_gid_table(ib_dev, port); mutex_lock(&table->lock); @@ -496,7 +511,7 @@ static int __ib_cache_gid_get(struct ib_device *ib_dev, u8 port, int index, { struct ib_gid_table *table; - table = ib_dev->cache.ports[port - rdma_start_port(ib_dev)].gid; + table = rdma_gid_table(ib_dev, port); if (index < 0 || index >= table->sz) return -EINVAL; @@ -594,7 +609,7 @@ int ib_find_cached_gid_by_port(struct ib_device *ib_dev, if (!rdma_is_port_valid(ib_dev, port)) return -ENOENT; - table = ib_dev->cache.ports[port - rdma_start_port(ib_dev)].gid; + table = rdma_gid_table(ib_dev, port); if (ndev) mask |= GID_ATTR_FIND_MASK_NETDEV; @@ -653,7 +668,7 @@ static int ib_cache_gid_find_by_filter(struct ib_device *ib_dev, if (!rdma_protocol_roce(ib_dev, port)) return -EPROTONOSUPPORT; - table = ib_dev->cache.ports[port - rdma_start_port(ib_dev)].gid; + table = rdma_gid_table(ib_dev, port); read_lock_irqsave(&table->rwlock, flags); for (i = 0; i < table->sz; i++) { @@ -730,8 +745,7 @@ static void cleanup_gid_table_port(struct ib_device *ib_dev, u8 port, mutex_lock(&table->lock); for (i = 0; i < table->sz; ++i) { - if (memcmp(&table->data_vec[i].gid, &zgid, - sizeof(table->data_vec[i].gid))) { + if (!rdma_is_zero_gid(&table->data_vec[i].gid)) { del_gid(ib_dev, port, table, i); deleted = true; } @@ -753,7 +767,7 @@ void ib_cache_gid_set_default_gid(struct ib_device *ib_dev, u8 port, unsigned int gid_type; unsigned long mask; - table = ib_dev->cache.ports[port - rdma_start_port(ib_dev)].gid; + table = rdma_gid_table(ib_dev, port); mask = GID_ATTR_FIND_MASK_GID_TYPE | GID_ATTR_FIND_MASK_DEFAULT | @@ -907,7 +921,7 @@ int ib_get_cached_gid(struct ib_device *device, if (!rdma_is_port_valid(device, port_num)) return -EINVAL; - table = device->cache.ports[port_num - rdma_start_port(device)].gid; + table = rdma_gid_table(device, port_num); read_lock_irqsave(&table->rwlock, flags); res = __ib_cache_gid_get(device, port_num, index, gid, gid_attr); read_unlock_irqrestore(&table->rwlock, flags); @@ -1106,7 +1120,7 @@ static int config_non_roce_gid_cache(struct ib_device *device, gid_attr.device = device; gid_attr.port_num = port; - table = device->cache.ports[port - rdma_start_port(device)].gid; + table = rdma_gid_table(device, port); mutex_lock(&table->lock); for (i = 0; i < gid_tbl_len; ++i) { @@ -1138,7 +1152,7 @@ static void ib_cache_update(struct ib_device *device, if (!rdma_is_port_valid(device, port)) return; - table = device->cache.ports[port - rdma_start_port(device)].gid; + table = rdma_gid_table(device, port); tprops = kmalloc(sizeof *tprops, GFP_KERNEL); if (!tprops) diff --git a/sys/ofed/drivers/infiniband/core/ib_cm.c b/sys/ofed/drivers/infiniband/core/ib_cm.c index 05d2992a3dd3..4cd36910f030 100644 --- a/sys/ofed/drivers/infiniband/core/ib_cm.c +++ b/sys/ofed/drivers/infiniband/core/ib_cm.c @@ -541,6 +541,7 @@ static struct cm_port *get_cm_port_from_path(struct sa_path_rec *path) static int cm_init_av_by_path(struct sa_path_rec *path, struct cm_av *av, struct cm_id_private *cm_id_priv) { + struct rdma_ah_attr new_ah_attr; struct cm_device *cm_dev; struct cm_port *port; int ret; @@ -556,15 +557,26 @@ static int cm_init_av_by_path(struct sa_path_rec *path, struct cm_av *av, return ret; av->port = port; - ret = ib_init_ah_attr_from_path(cm_dev->ib_device, port->port_num, - path, &av->ah_attr); + + /* + * av->ah_attr might be initialized based on wc or during + * request processing time. So initialize a new ah_attr on stack. + * If initialization fails, old ah_attr is used for sending any + * responses. If initialization is successful, than new ah_attr + * is used by overwriting the old one. + */ + ret = ib_init_ah_attr_from_path(cm_dev->ib_device, port->port_num, path, + &new_ah_attr); if (ret) return ret; av->timeout = path->packet_life_time + 1; ret = add_cm_id_to_port_list(cm_id_priv, av, port); - return ret; + if (ret) + return ret; + memcpy(&av->ah_attr, &new_ah_attr, sizeof(new_ah_attr)); + return 0; } static int cm_alloc_id(struct cm_id_private *cm_id_priv) @@ -3177,12 +3189,6 @@ static int cm_lap_handler(struct cm_work *work) if (!cm_id_priv) return -EINVAL; - ret = cm_init_av_for_response(work->port, work->mad_recv_wc->wc, - work->mad_recv_wc->recv_buf.grh, - &cm_id_priv->av); - if (ret) - goto deref; - param = &work->cm_event.param.lap_rcvd; memset(&work->path[0], 0, sizeof(work->path[1])); cm_path_set_rec_type(work->port->cm_dev->ib_device, @@ -3227,12 +3233,19 @@ static int cm_lap_handler(struct cm_work *work) goto unlock; } - cm_id_priv->id.lap_state = IB_CM_LAP_RCVD; - cm_id_priv->tid = lap_msg->hdr.tid; + ret = cm_init_av_for_lap(work->port, work->mad_recv_wc->wc, + work->mad_recv_wc->recv_buf.grh, + &cm_id_priv->av); + if (ret) + goto unlock; + ret = cm_init_av_by_path(param->alternate_path, &cm_id_priv->alt_av, cm_id_priv); if (ret) goto unlock; + + cm_id_priv->id.lap_state = IB_CM_LAP_RCVD; + cm_id_priv->tid = lap_msg->hdr.tid; ret = atomic_inc_and_test(&cm_id_priv->work_count); if (!ret) list_add_tail(&work->list, &cm_id_priv->work_list); diff --git a/sys/ofed/drivers/infiniband/core/ib_roce_gid_mgmt.c b/sys/ofed/drivers/infiniband/core/ib_roce_gid_mgmt.c index 50932d6ec675..8b119efad5f6 100644 --- a/sys/ofed/drivers/infiniband/core/ib_roce_gid_mgmt.c +++ b/sys/ofed/drivers/infiniband/core/ib_roce_gid_mgmt.c @@ -280,7 +280,7 @@ roce_gid_update_addr_callback(struct ib_device *device, u8 port, dev_put(ndev); /* don't delete empty entries */ - if (memcmp(&gid, &zgid, sizeof(zgid)) == 0) + if (rdma_is_zero_gid(&gid)) continue; /* zero default */ diff --git a/sys/ofed/drivers/infiniband/core/ib_umem.c b/sys/ofed/drivers/infiniband/core/ib_umem.c index a72047075257..87325c480057 100644 --- a/sys/ofed/drivers/infiniband/core/ib_umem.c +++ b/sys/ofed/drivers/infiniband/core/ib_umem.c @@ -119,16 +119,7 @@ struct ib_umem *ib_umem_get(struct ib_ucontext *context, unsigned long addr, umem->address = addr; umem->page_shift = PAGE_SHIFT; umem->pid = get_pid(task_pid(current)); - /* - * We ask for writable memory if any of the following - * access flags are set. "Local write" and "remote write" - * obviously require write access. "Remote atomic" can do - * things like fetch and add, which will modify memory, and - * "MW bind" can change permissions by binding a window. - */ - umem->writable = !!(access & - (IB_ACCESS_LOCAL_WRITE | IB_ACCESS_REMOTE_WRITE | - IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_MW_BIND)); + umem->writable = ib_access_writable(access); if (access & IB_ACCESS_ON_DEMAND) { put_pid(umem->pid); diff --git a/sys/ofed/include/rdma/ib_cache.h b/sys/ofed/include/rdma/ib_cache.h index 07ebea90be01..fbd9135ddab7 100644 --- a/sys/ofed/include/rdma/ib_cache.h +++ b/sys/ofed/include/rdma/ib_cache.h @@ -151,4 +151,5 @@ int ib_get_cached_port_state(struct ib_device *device, u8 port_num, enum ib_port_state *port_active); +bool rdma_is_zero_gid(const union ib_gid *gid); #endif /* _IB_CACHE_H */home | help
Want to link to this message? Use this
URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?6a3c3535.43de0.3d971eb8>