From b47bd6ea40636362a8b6605de51207cc387ba0b8 Mon Sep 17 00:00:00 2001 From: Daniel Jurgens Date: Tue, 25 Oct 2016 18:36:24 +0300 Subject: [PATCH 01/12] {net, ib}/mlx5: Make cache line size determination at runtime. ARM 64B cache line systems have L1_CACHE_BYTES set to 128. cache_line_size() will return the correct size. Fixes: cf50b5efa2fe('net/mlx5_core/ib: New device capabilities handling.') Signed-off-by: Daniel Jurgens Signed-off-by: Saeed Mahameed Signed-off-by: David S. Miller --- drivers/infiniband/hw/mlx5/main.c | 2 +- drivers/infiniband/hw/mlx5/qp.c | 1 - .../net/ethernet/mellanox/mlx5/core/alloc.c | 31 ++++++++++++++++--- include/linux/mlx5/driver.h | 11 ------- 4 files changed, 27 insertions(+), 18 deletions(-) diff --git a/drivers/infiniband/hw/mlx5/main.c b/drivers/infiniband/hw/mlx5/main.c index 22174774dbb8..63036c731626 100644 --- a/drivers/infiniband/hw/mlx5/main.c +++ b/drivers/infiniband/hw/mlx5/main.c @@ -1019,7 +1019,7 @@ static struct ib_ucontext *mlx5_ib_alloc_ucontext(struct ib_device *ibdev, resp.qp_tab_size = 1 << MLX5_CAP_GEN(dev->mdev, log_max_qp); if (mlx5_core_is_pf(dev->mdev) && MLX5_CAP_GEN(dev->mdev, bf)) resp.bf_reg_size = 1 << MLX5_CAP_GEN(dev->mdev, log_bf_reg_size); - resp.cache_line_size = L1_CACHE_BYTES; + resp.cache_line_size = cache_line_size(); resp.max_sq_desc_sz = MLX5_CAP_GEN(dev->mdev, max_wqe_sz_sq); resp.max_rq_desc_sz = MLX5_CAP_GEN(dev->mdev, max_wqe_sz_rq); resp.max_send_wqebb = 1 << MLX5_CAP_GEN(dev->mdev, log_max_qp_sz); diff --git a/drivers/infiniband/hw/mlx5/qp.c b/drivers/infiniband/hw/mlx5/qp.c index 41f4c2afbcdd..7ce97daf26c6 100644 --- a/drivers/infiniband/hw/mlx5/qp.c +++ b/drivers/infiniband/hw/mlx5/qp.c @@ -52,7 +52,6 @@ enum { enum { MLX5_IB_SQ_STRIDE = 6, - MLX5_IB_CACHE_LINE_SIZE = 64, }; static const u32 mlx5_ib_opcode[] = { diff --git a/drivers/net/ethernet/mellanox/mlx5/core/alloc.c b/drivers/net/ethernet/mellanox/mlx5/core/alloc.c index 6cb38304669f..2c6e3c7b7417 100644 --- a/drivers/net/ethernet/mellanox/mlx5/core/alloc.c +++ b/drivers/net/ethernet/mellanox/mlx5/core/alloc.c @@ -41,6 +41,13 @@ #include "mlx5_core.h" +struct mlx5_db_pgdir { + struct list_head list; + unsigned long *bitmap; + __be32 *db_page; + dma_addr_t db_dma; +}; + /* Handling for queue buffers -- we allocate a bunch of memory and * register it in a memory region at HCA virtual address 0. */ @@ -102,17 +109,28 @@ EXPORT_SYMBOL_GPL(mlx5_buf_free); static struct mlx5_db_pgdir *mlx5_alloc_db_pgdir(struct mlx5_core_dev *dev, int node) { + u32 db_per_page = PAGE_SIZE / cache_line_size(); struct mlx5_db_pgdir *pgdir; pgdir = kzalloc(sizeof(*pgdir), GFP_KERNEL); if (!pgdir) return NULL; - bitmap_fill(pgdir->bitmap, MLX5_DB_PER_PAGE); + pgdir->bitmap = kcalloc(BITS_TO_LONGS(db_per_page), + sizeof(unsigned long), + GFP_KERNEL); + + if (!pgdir->bitmap) { + kfree(pgdir); + return NULL; + } + + bitmap_fill(pgdir->bitmap, db_per_page); pgdir->db_page = mlx5_dma_zalloc_coherent_node(dev, PAGE_SIZE, &pgdir->db_dma, node); if (!pgdir->db_page) { + kfree(pgdir->bitmap); kfree(pgdir); return NULL; } @@ -123,18 +141,19 @@ static struct mlx5_db_pgdir *mlx5_alloc_db_pgdir(struct mlx5_core_dev *dev, static int mlx5_alloc_db_from_pgdir(struct mlx5_db_pgdir *pgdir, struct mlx5_db *db) { + u32 db_per_page = PAGE_SIZE / cache_line_size(); int offset; int i; - i = find_first_bit(pgdir->bitmap, MLX5_DB_PER_PAGE); - if (i >= MLX5_DB_PER_PAGE) + i = find_first_bit(pgdir->bitmap, db_per_page); + if (i >= db_per_page) return -ENOMEM; __clear_bit(i, pgdir->bitmap); db->u.pgdir = pgdir; db->index = i; - offset = db->index * L1_CACHE_BYTES; + offset = db->index * cache_line_size(); db->db = pgdir->db_page + offset / sizeof(*pgdir->db_page); db->dma = pgdir->db_dma + offset; @@ -181,14 +200,16 @@ EXPORT_SYMBOL_GPL(mlx5_db_alloc); void mlx5_db_free(struct mlx5_core_dev *dev, struct mlx5_db *db) { + u32 db_per_page = PAGE_SIZE / cache_line_size(); mutex_lock(&dev->priv.pgdir_mutex); __set_bit(db->index, db->u.pgdir->bitmap); - if (bitmap_full(db->u.pgdir->bitmap, MLX5_DB_PER_PAGE)) { + if (bitmap_full(db->u.pgdir->bitmap, db_per_page)) { dma_free_coherent(&(dev->pdev->dev), PAGE_SIZE, db->u.pgdir->db_page, db->u.pgdir->db_dma); list_del(&db->u.pgdir->list); + kfree(db->u.pgdir->bitmap); kfree(db->u.pgdir); } diff --git a/include/linux/mlx5/driver.h b/include/linux/mlx5/driver.h index 85c4786427e4..5dbda60a09f4 100644 --- a/include/linux/mlx5/driver.h +++ b/include/linux/mlx5/driver.h @@ -625,10 +625,6 @@ struct mlx5_db { int index; }; -enum { - MLX5_DB_PER_PAGE = PAGE_SIZE / L1_CACHE_BYTES, -}; - enum { MLX5_COMP_EQ_SIZE = 1024, }; @@ -638,13 +634,6 @@ enum { MLX5_PTYS_EN = 1 << 2, }; -struct mlx5_db_pgdir { - struct list_head list; - DECLARE_BITMAP(bitmap, MLX5_DB_PER_PAGE); - __be32 *db_page; - dma_addr_t db_dma; -}; - typedef void (*mlx5_cmd_cbk_t)(int status, void *context); struct mlx5_cmd_work_ent { From bba1574c2f11d674bf343f3cf307b33d19f73d9d Mon Sep 17 00:00:00 2001 From: Daniel Jurgens Date: Tue, 25 Oct 2016 18:36:25 +0300 Subject: [PATCH 02/12] net/mlx5: Always Query HCA caps after setting them Always query the HCA caps after setting them to update the capablities data structures. Not doing so results in incorrect capabilities being reported including max_dc, max_qp and several others. Fixes: 59211bd3b632 ("net/mlx5: Split the load/unload flow into hardware and software flows") Signed-off-by: Daniel Jurgens Signed-off-by: Saeed Mahameed Signed-off-by: David S. Miller --- drivers/net/ethernet/mellanox/mlx5/core/main.c | 12 ++++++------ 1 file changed, 6 insertions(+), 6 deletions(-) diff --git a/drivers/net/ethernet/mellanox/mlx5/core/main.c b/drivers/net/ethernet/mellanox/mlx5/core/main.c index d9c3c70b29e4..8a63910bfccf 100644 --- a/drivers/net/ethernet/mellanox/mlx5/core/main.c +++ b/drivers/net/ethernet/mellanox/mlx5/core/main.c @@ -844,12 +844,6 @@ static int mlx5_init_once(struct mlx5_core_dev *dev, struct mlx5_priv *priv) struct pci_dev *pdev = dev->pdev; int err; - err = mlx5_query_hca_caps(dev); - if (err) { - dev_err(&pdev->dev, "query hca failed\n"); - goto out; - } - err = mlx5_query_board_id(dev); if (err) { dev_err(&pdev->dev, "query board id failed\n"); @@ -1023,6 +1017,12 @@ static int mlx5_load_one(struct mlx5_core_dev *dev, struct mlx5_priv *priv, mlx5_start_health_poll(dev); + err = mlx5_query_hca_caps(dev); + if (err) { + dev_err(&pdev->dev, "query hca failed\n"); + goto err_stop_poll; + } + if (boot && mlx5_init_once(dev, priv)) { dev_err(&pdev->dev, "sw objs init failed\n"); goto err_stop_poll; From eccec8da3b4e6f403040c7187338a46d2ea27c20 Mon Sep 17 00:00:00 2001 From: Paul Blakey Date: Tue, 25 Oct 2016 18:36:26 +0300 Subject: [PATCH 03/12] net/mlx5: Keep autogroups list ordered Finding a new autogroup range is done by going over a group list sorted by each group start index. The search is stopped after finding the first free range. Adding the newly created group to the list is wrongly added to the end of the list regardless of its start index as the parameter of where to insert it is ignored. This commit makes sure to use that unused parameter to insert it where requested. Fixes: f0d22d187473 ('net/mlx5_core: Introduce flow steering autogrouped flow table') Signed-off-by: Paul Blakey Signed-off-by: Saeed Mahameed Signed-off-by: David S. Miller --- drivers/net/ethernet/mellanox/mlx5/core/fs_core.c | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-) diff --git a/drivers/net/ethernet/mellanox/mlx5/core/fs_core.c b/drivers/net/ethernet/mellanox/mlx5/core/fs_core.c index 5da2cc878582..17559c842905 100644 --- a/drivers/net/ethernet/mellanox/mlx5/core/fs_core.c +++ b/drivers/net/ethernet/mellanox/mlx5/core/fs_core.c @@ -879,7 +879,7 @@ static struct mlx5_flow_group *create_flow_group_common(struct mlx5_flow_table * tree_init_node(&fg->node, !is_auto_fg, del_flow_group); tree_add_node(&fg->node, &ft->node); /* Add node to group list */ - list_add(&fg->node.list, ft->node.children.prev); + list_add(&fg->node.list, prev_fg); return fg; } @@ -893,7 +893,7 @@ struct mlx5_flow_group *mlx5_create_flow_group(struct mlx5_flow_table *ft, return ERR_PTR(-EPERM); lock_ref_node(&ft->node); - fg = create_flow_group_common(ft, fg_in, &ft->node.children, false); + fg = create_flow_group_common(ft, fg_in, ft->node.children.prev, false); unlock_ref_node(&ft->node); return fg; @@ -1012,7 +1012,7 @@ static struct mlx5_flow_group *create_autogroup(struct mlx5_flow_table *ft, u32 *match_criteria) { int inlen = MLX5_ST_SZ_BYTES(create_flow_group_in); - struct list_head *prev = &ft->node.children; + struct list_head *prev = ft->node.children.prev; unsigned int candidate_index = 0; struct mlx5_flow_group *fg; void *match_criteria_addr; From 32dba76a7a3104b2815f31b1755f4c0ba1f01a78 Mon Sep 17 00:00:00 2001 From: Paul Blakey Date: Tue, 25 Oct 2016 18:36:27 +0300 Subject: [PATCH 04/12] net/mlx5: Fix autogroups groups num not decreasing Autogroups groups num is increased when creating a new flow group, but is never decreased. Now decreasing it when deleting a flow group. Fixes: f0d22d187473 ('net/mlx5_core: Introduce flow steering autogrouped flow table') Signed-off-by: Paul Blakey Signed-off-by: Saeed Mahameed Signed-off-by: David S. Miller --- drivers/net/ethernet/mellanox/mlx5/core/fs_core.c | 3 +++ 1 file changed, 3 insertions(+) diff --git a/drivers/net/ethernet/mellanox/mlx5/core/fs_core.c b/drivers/net/ethernet/mellanox/mlx5/core/fs_core.c index 17559c842905..89696048b045 100644 --- a/drivers/net/ethernet/mellanox/mlx5/core/fs_core.c +++ b/drivers/net/ethernet/mellanox/mlx5/core/fs_core.c @@ -436,6 +436,9 @@ static void del_flow_group(struct fs_node *node) fs_get_obj(ft, fg->node.parent); dev = get_dev(&ft->node); + if (ft->autogroup.active) + ft->autogroup.num_groups--; + if (mlx5_cmd_destroy_flow_group(dev, ft, fg->id)) mlx5_core_warn(dev, "flow steering can't destroy fg %d of ft %d\n", fg->id, ft->id); From e83d6955fe422f120b0e98e4f02496af89845eef Mon Sep 17 00:00:00 2001 From: Paul Blakey Date: Tue, 25 Oct 2016 18:36:28 +0300 Subject: [PATCH 05/12] net/mlx5: Correctly initialize last use of flow counters Currently, last use timestamp is initialized to zero. This is not the expected value by higher layers such as when we do TC action offloading. To fix that, set it to the current time, e.g when the counter/rule is offloaded. This is the same behaviour of non-offloaded TC actions. Fixes: 43a335e055bb ('mlx5_core: Flow counters infrastructure') Signed-off-by: Paul Blakey Signed-off-by: Saeed Mahameed Signed-off-by: David S. Miller --- drivers/net/ethernet/mellanox/mlx5/core/fs_counters.c | 1 + 1 file changed, 1 insertion(+) diff --git a/drivers/net/ethernet/mellanox/mlx5/core/fs_counters.c b/drivers/net/ethernet/mellanox/mlx5/core/fs_counters.c index 3a9195b4169d..3b026c151cf2 100644 --- a/drivers/net/ethernet/mellanox/mlx5/core/fs_counters.c +++ b/drivers/net/ethernet/mellanox/mlx5/core/fs_counters.c @@ -218,6 +218,7 @@ struct mlx5_fc *mlx5_fc_create(struct mlx5_core_dev *dev, bool aging) goto err_out; if (aging) { + counter->cache.lastuse = jiffies; counter->aging = true; spin_lock(&fc_stats->addlist_lock); From 2b029556667a7fc1aea731c5828e501656f87653 Mon Sep 17 00:00:00 2001 From: Saeed Mahameed Date: Tue, 25 Oct 2016 18:36:29 +0300 Subject: [PATCH 06/12] net/mlx5e: Choose best nearest LRO timeout Instead of predicting the index of the wanted LRO timeout value from hardware capabilities, look for the nearest LRO timeout value. Fixes: 5c50368f3831 ('net/mlx5e: Light-weight netdev open/stop') Signed-off-by: Saeed Mahameed Signed-off-by: Mohamad Haj Yahia Signed-off-by: David S. Miller --- drivers/net/ethernet/mellanox/mlx5/core/en.h | 5 +++++ .../net/ethernet/mellanox/mlx5/core/en_main.c | 19 ++++++++++++++++--- 2 files changed, 21 insertions(+), 3 deletions(-) diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en.h b/drivers/net/ethernet/mellanox/mlx5/core/en.h index 460363b66cb1..7a43502a89cc 100644 --- a/drivers/net/ethernet/mellanox/mlx5/core/en.h +++ b/drivers/net/ethernet/mellanox/mlx5/core/en.h @@ -85,6 +85,9 @@ #define MLX5_MPWRQ_SMALL_PACKET_THRESHOLD (128) #define MLX5E_PARAMS_DEFAULT_LRO_WQE_SZ (64 * 1024) +#define MLX5E_DEFAULT_LRO_TIMEOUT 32 +#define MLX5E_LRO_TIMEOUT_ARR_SIZE 4 + #define MLX5E_PARAMS_DEFAULT_RX_CQ_MODERATION_USEC 0x10 #define MLX5E_PARAMS_DEFAULT_RX_CQ_MODERATION_USEC_FROM_CQE 0x3 #define MLX5E_PARAMS_DEFAULT_RX_CQ_MODERATION_PKTS 0x20 @@ -221,6 +224,7 @@ struct mlx5e_params { struct ieee_ets ets; #endif bool rx_am_enabled; + u32 lro_timeout; }; struct mlx5e_tstamp { @@ -888,5 +892,6 @@ int mlx5e_attach_netdev(struct mlx5_core_dev *mdev, struct net_device *netdev); void mlx5e_detach_netdev(struct mlx5_core_dev *mdev, struct net_device *netdev); struct rtnl_link_stats64 * mlx5e_get_stats(struct net_device *dev, struct rtnl_link_stats64 *stats); +u32 mlx5e_choose_lro_timeout(struct mlx5_core_dev *mdev, u32 wanted_timeout); #endif /* __MLX5_EN_H__ */ diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_main.c b/drivers/net/ethernet/mellanox/mlx5/core/en_main.c index 7eaf38020a8f..5c8ab3eabe9c 100644 --- a/drivers/net/ethernet/mellanox/mlx5/core/en_main.c +++ b/drivers/net/ethernet/mellanox/mlx5/core/en_main.c @@ -1971,9 +1971,7 @@ static void mlx5e_build_tir_ctx_lro(void *tirc, struct mlx5e_priv *priv) MLX5_SET(tirc, tirc, lro_max_ip_payload_size, (priv->params.lro_wqe_sz - ROUGH_MAX_L2_L3_HDR_SZ) >> 8); - MLX5_SET(tirc, tirc, lro_timeout_period_usecs, - MLX5_CAP_ETH(priv->mdev, - lro_timer_supported_periods[2])); + MLX5_SET(tirc, tirc, lro_timeout_period_usecs, priv->params.lro_timeout); } void mlx5e_build_tir_ctx_hash(void *tirc, struct mlx5e_priv *priv) @@ -3401,6 +3399,18 @@ static void mlx5e_query_min_inline(struct mlx5_core_dev *mdev, } } +u32 mlx5e_choose_lro_timeout(struct mlx5_core_dev *mdev, u32 wanted_timeout) +{ + int i; + + /* The supported periods are organized in ascending order */ + for (i = 0; i < MLX5E_LRO_TIMEOUT_ARR_SIZE - 1; i++) + if (MLX5_CAP_ETH(mdev, lro_timer_supported_periods[i]) >= wanted_timeout) + break; + + return MLX5_CAP_ETH(mdev, lro_timer_supported_periods[i]); +} + static void mlx5e_build_nic_netdev_priv(struct mlx5_core_dev *mdev, struct net_device *netdev, const struct mlx5e_profile *profile, @@ -3419,6 +3429,9 @@ static void mlx5e_build_nic_netdev_priv(struct mlx5_core_dev *mdev, priv->profile = profile; priv->ppriv = ppriv; + priv->params.lro_timeout = + mlx5e_choose_lro_timeout(mdev, MLX5E_DEFAULT_LRO_TIMEOUT); + priv->params.log_sq_size = MLX5E_PARAMS_DEFAULT_LOG_SQ_SIZE; /* set CQE compression */ From 5e1e93c7047381b81236f82f60c15d49c510d1a7 Mon Sep 17 00:00:00 2001 From: Mohamad Haj Yahia Date: Tue, 25 Oct 2016 18:36:30 +0300 Subject: [PATCH 07/12] net/mlx5e: Unregister netdev before detaching it Detaching the netdev before unregistering it cause some netdev cleanup ndos to fail because they check presence of the netdev, so we need to unregister the netdev first. Fixes: 26e59d8077a3 ('net/mlx5e: Implement mlx5e interface attach/detach callbacks') Signed-off-by: Mohamad Haj Yahia Signed-off-by: Saeed Mahameed Signed-off-by: David S. Miller --- drivers/net/ethernet/mellanox/mlx5/core/en_main.c | 2 +- drivers/net/ethernet/mellanox/mlx5/core/en_rep.c | 1 + 2 files changed, 2 insertions(+), 1 deletion(-) diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_main.c b/drivers/net/ethernet/mellanox/mlx5/core/en_main.c index 5c8ab3eabe9c..f4c687ce4c59 100644 --- a/drivers/net/ethernet/mellanox/mlx5/core/en_main.c +++ b/drivers/net/ethernet/mellanox/mlx5/core/en_main.c @@ -4048,7 +4048,6 @@ void mlx5e_destroy_netdev(struct mlx5_core_dev *mdev, struct mlx5e_priv *priv) const struct mlx5e_profile *profile = priv->profile; struct net_device *netdev = priv->netdev; - unregister_netdev(netdev); destroy_workqueue(priv->wq); if (profile->cleanup) profile->cleanup(priv); @@ -4065,6 +4064,7 @@ static void mlx5e_remove(struct mlx5_core_dev *mdev, void *vpriv) for (vport = 1; vport < total_vfs; vport++) mlx5_eswitch_unregister_vport_rep(esw, vport); + unregister_netdev(priv->netdev); mlx5e_detach(mdev, vpriv); mlx5e_destroy_netdev(mdev, priv); } diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_rep.c b/drivers/net/ethernet/mellanox/mlx5/core/en_rep.c index 3c97da103d30..7fe6559e4ab3 100644 --- a/drivers/net/ethernet/mellanox/mlx5/core/en_rep.c +++ b/drivers/net/ethernet/mellanox/mlx5/core/en_rep.c @@ -457,6 +457,7 @@ void mlx5e_vport_rep_unload(struct mlx5_eswitch *esw, struct mlx5e_priv *priv = rep->priv_data; struct net_device *netdev = priv->netdev; + unregister_netdev(netdev); mlx5e_detach_netdev(esw->dev, netdev); mlx5e_destroy_netdev(esw->dev, priv); } From 247f139cdae73b4f47bd348d05dff1afd40b84b6 Mon Sep 17 00:00:00 2001 From: Mohamad Haj Yahia Date: Tue, 25 Oct 2016 18:36:31 +0300 Subject: [PATCH 08/12] net/mlx5: Change the acl enable prototype to return status The Ingress/Egress ACL enable function may fail and it should return status to its caller to avoid NULL pointer dereference. Fixes: f942380c1239 ('net/mlx5: E-Switch, Vport ingress/egress ACLs rules for spoofchk') Signed-off-by: Mohamad Haj Yahia Signed-off-by: Saeed Mahameed Signed-off-by: David S. Miller --- .../net/ethernet/mellanox/mlx5/core/eswitch.c | 50 +++++++++++++------ 1 file changed, 34 insertions(+), 16 deletions(-) diff --git a/drivers/net/ethernet/mellanox/mlx5/core/eswitch.c b/drivers/net/ethernet/mellanox/mlx5/core/eswitch.c index abbf2c369923..be1f7333ab7f 100644 --- a/drivers/net/ethernet/mellanox/mlx5/core/eswitch.c +++ b/drivers/net/ethernet/mellanox/mlx5/core/eswitch.c @@ -931,8 +931,8 @@ static void esw_vport_change_handler(struct work_struct *work) mutex_unlock(&esw->state_lock); } -static void esw_vport_enable_egress_acl(struct mlx5_eswitch *esw, - struct mlx5_vport *vport) +static int esw_vport_enable_egress_acl(struct mlx5_eswitch *esw, + struct mlx5_vport *vport) { int inlen = MLX5_ST_SZ_BYTES(create_flow_group_in); struct mlx5_flow_group *vlan_grp = NULL; @@ -949,9 +949,11 @@ static void esw_vport_enable_egress_acl(struct mlx5_eswitch *esw, int table_size = 2; int err = 0; - if (!MLX5_CAP_ESW_EGRESS_ACL(dev, ft_support) || - !IS_ERR_OR_NULL(vport->egress.acl)) - return; + if (!MLX5_CAP_ESW_EGRESS_ACL(dev, ft_support)) + return -EOPNOTSUPP; + + if (!IS_ERR_OR_NULL(vport->egress.acl)) + return 0; esw_debug(dev, "Create vport[%d] egress ACL log_max_size(%d)\n", vport->vport, MLX5_CAP_ESW_EGRESS_ACL(dev, log_max_ft_size)); @@ -959,12 +961,12 @@ static void esw_vport_enable_egress_acl(struct mlx5_eswitch *esw, root_ns = mlx5_get_flow_namespace(dev, MLX5_FLOW_NAMESPACE_ESW_EGRESS); if (!root_ns) { esw_warn(dev, "Failed to get E-Switch egress flow namespace\n"); - return; + return -EIO; } flow_group_in = mlx5_vzalloc(inlen); if (!flow_group_in) - return; + return -ENOMEM; acl = mlx5_create_vport_flow_table(root_ns, 0, table_size, 0, vport->vport); if (IS_ERR(acl)) { @@ -1009,6 +1011,7 @@ out: mlx5_destroy_flow_group(vlan_grp); if (err && !IS_ERR_OR_NULL(acl)) mlx5_destroy_flow_table(acl); + return err; } static void esw_vport_cleanup_egress_rules(struct mlx5_eswitch *esw, @@ -1041,8 +1044,8 @@ static void esw_vport_disable_egress_acl(struct mlx5_eswitch *esw, vport->egress.acl = NULL; } -static void esw_vport_enable_ingress_acl(struct mlx5_eswitch *esw, - struct mlx5_vport *vport) +static int esw_vport_enable_ingress_acl(struct mlx5_eswitch *esw, + struct mlx5_vport *vport) { int inlen = MLX5_ST_SZ_BYTES(create_flow_group_in); struct mlx5_core_dev *dev = esw->dev; @@ -1063,9 +1066,11 @@ static void esw_vport_enable_ingress_acl(struct mlx5_eswitch *esw, int table_size = 4; int err = 0; - if (!MLX5_CAP_ESW_INGRESS_ACL(dev, ft_support) || - !IS_ERR_OR_NULL(vport->ingress.acl)) - return; + if (!MLX5_CAP_ESW_INGRESS_ACL(dev, ft_support)) + return -EOPNOTSUPP; + + if (!IS_ERR_OR_NULL(vport->ingress.acl)) + return 0; esw_debug(dev, "Create vport[%d] ingress ACL log_max_size(%d)\n", vport->vport, MLX5_CAP_ESW_INGRESS_ACL(dev, log_max_ft_size)); @@ -1073,12 +1078,12 @@ static void esw_vport_enable_ingress_acl(struct mlx5_eswitch *esw, root_ns = mlx5_get_flow_namespace(dev, MLX5_FLOW_NAMESPACE_ESW_INGRESS); if (!root_ns) { esw_warn(dev, "Failed to get E-Switch ingress flow namespace\n"); - return; + return -EIO; } flow_group_in = mlx5_vzalloc(inlen); if (!flow_group_in) - return; + return -ENOMEM; acl = mlx5_create_vport_flow_table(root_ns, 0, table_size, 0, vport->vport); if (IS_ERR(acl)) { @@ -1167,6 +1172,7 @@ out: } kvfree(flow_group_in); + return err; } static void esw_vport_cleanup_ingress_rules(struct mlx5_eswitch *esw, @@ -1225,7 +1231,13 @@ static int esw_vport_ingress_config(struct mlx5_eswitch *esw, return 0; } - esw_vport_enable_ingress_acl(esw, vport); + err = esw_vport_enable_ingress_acl(esw, vport); + if (err) { + mlx5_core_warn(esw->dev, + "failed to enable ingress acl (%d) on vport[%d]\n", + err, vport->vport); + return err; + } esw_debug(esw->dev, "vport[%d] configure ingress rules, vlan(%d) qos(%d)\n", @@ -1299,7 +1311,13 @@ static int esw_vport_egress_config(struct mlx5_eswitch *esw, return 0; } - esw_vport_enable_egress_acl(esw, vport); + err = esw_vport_enable_egress_acl(esw, vport); + if (err) { + mlx5_core_warn(esw->dev, + "failed to enable egress acl (%d) on vport[%d]\n", + err, vport->vport); + return err; + } esw_debug(esw->dev, "vport[%d] configure egress rules, vlan(%d) qos(%d)\n", From 2241007b3d783cbdbaa78c30bdb1994278b6f9b9 Mon Sep 17 00:00:00 2001 From: Mohamad Haj Yahia Date: Tue, 25 Oct 2016 18:36:32 +0300 Subject: [PATCH 09/12] net/mlx5: Clear health sick bit when starting health poll The health sick status should be cleared when we start the health poll. This is crucial for driver reload (unload + load) in order to behave right in case of health issue. Fixes: fd76ee4da55a ('net/mlx5_core: Fix internal error detection conditions') Signed-off-by: Mohamad Haj Yahia Signed-off-by: Saeed Mahameed Signed-off-by: David S. Miller --- drivers/net/ethernet/mellanox/mlx5/core/health.c | 1 + 1 file changed, 1 insertion(+) diff --git a/drivers/net/ethernet/mellanox/mlx5/core/health.c b/drivers/net/ethernet/mellanox/mlx5/core/health.c index 1a05fb965c8d..2cb4094c9c49 100644 --- a/drivers/net/ethernet/mellanox/mlx5/core/health.c +++ b/drivers/net/ethernet/mellanox/mlx5/core/health.c @@ -281,6 +281,7 @@ void mlx5_start_health_poll(struct mlx5_core_dev *dev) struct mlx5_core_health *health = &dev->priv.health; init_timer(&health->timer); + health->sick = 0; health->health = &dev->iseg->health; health->health_counter = &dev->iseg->health_counter; From 05ac2c0b7438ea08c5d54b48797acf9b22cb2f6f Mon Sep 17 00:00:00 2001 From: Mohamad Haj Yahia Date: Tue, 25 Oct 2016 18:36:33 +0300 Subject: [PATCH 10/12] net/mlx5: Fix race between PCI error handlers and health work Currently there is a race between the health care work and the kernel pci error handlers because both of them detect the error, the first one to be called will do the error handling. There is a chance that health care will disable the pci after resuming pci slot. Also create a separate WQ because now we will have two types of health works, one for the error detection and one for the recovery. Fixes: 89d44f0a6c73 ('net/mlx5_core: Add pci error handlers to mlx5_core driver') Signed-off-by: Mohamad Haj Yahia Signed-off-by: Saeed Mahameed Signed-off-by: David S. Miller --- .../net/ethernet/mellanox/mlx5/core/health.c | 30 +++++++++++++++++-- .../net/ethernet/mellanox/mlx5/core/main.c | 9 ++++-- include/linux/mlx5/driver.h | 4 +++ 3 files changed, 38 insertions(+), 5 deletions(-) diff --git a/drivers/net/ethernet/mellanox/mlx5/core/health.c b/drivers/net/ethernet/mellanox/mlx5/core/health.c index 2cb4094c9c49..2d00022c92d8 100644 --- a/drivers/net/ethernet/mellanox/mlx5/core/health.c +++ b/drivers/net/ethernet/mellanox/mlx5/core/health.c @@ -64,6 +64,10 @@ enum { MLX5_NIC_IFC_NO_DRAM_NIC = 2 }; +enum { + MLX5_DROP_NEW_HEALTH_WORK, +}; + static u8 get_nic_interface(struct mlx5_core_dev *dev) { return (ioread32be(&dev->iseg->cmdq_addr_l_sz) >> 8) & 3; @@ -272,7 +276,13 @@ static void poll_health(unsigned long data) if (in_fatal(dev) && !health->sick) { health->sick = true; print_health_info(dev); - schedule_work(&health->work); + spin_lock(&health->wq_lock); + if (!test_bit(MLX5_DROP_NEW_HEALTH_WORK, &health->flags)) + queue_work(health->wq, &health->work); + else + dev_err(&dev->pdev->dev, + "new health works are not permitted at this stage\n"); + spin_unlock(&health->wq_lock); } } @@ -282,6 +292,7 @@ void mlx5_start_health_poll(struct mlx5_core_dev *dev) init_timer(&health->timer); health->sick = 0; + clear_bit(MLX5_DROP_NEW_HEALTH_WORK, &health->flags); health->health = &dev->iseg->health; health->health_counter = &dev->iseg->health_counter; @@ -298,11 +309,21 @@ void mlx5_stop_health_poll(struct mlx5_core_dev *dev) del_timer_sync(&health->timer); } +void mlx5_drain_health_wq(struct mlx5_core_dev *dev) +{ + struct mlx5_core_health *health = &dev->priv.health; + + spin_lock(&health->wq_lock); + set_bit(MLX5_DROP_NEW_HEALTH_WORK, &health->flags); + spin_unlock(&health->wq_lock); + cancel_work_sync(&health->work); +} + void mlx5_health_cleanup(struct mlx5_core_dev *dev) { struct mlx5_core_health *health = &dev->priv.health; - flush_work(&health->work); + destroy_workqueue(health->wq); } int mlx5_health_init(struct mlx5_core_dev *dev) @@ -317,8 +338,11 @@ int mlx5_health_init(struct mlx5_core_dev *dev) strcpy(name, "mlx5_health"); strcat(name, dev_name(&dev->pdev->dev)); + health->wq = create_singlethread_workqueue(name); kfree(name); - + if (!health->wq) + return -ENOMEM; + spin_lock_init(&health->wq_lock); INIT_WORK(&health->work, health_care); return 0; diff --git a/drivers/net/ethernet/mellanox/mlx5/core/main.c b/drivers/net/ethernet/mellanox/mlx5/core/main.c index 8a63910bfccf..9f90226bc120 100644 --- a/drivers/net/ethernet/mellanox/mlx5/core/main.c +++ b/drivers/net/ethernet/mellanox/mlx5/core/main.c @@ -1315,8 +1315,13 @@ static pci_ers_result_t mlx5_pci_err_detected(struct pci_dev *pdev, dev_info(&pdev->dev, "%s was called\n", __func__); mlx5_enter_error_state(dev); mlx5_unload_one(dev, priv, false); - pci_save_state(pdev); - mlx5_pci_disable_device(dev); + /* In case of kernel call save the pci state and drain health wq */ + if (state) { + pci_save_state(pdev); + mlx5_drain_health_wq(dev); + mlx5_pci_disable_device(dev); + } + return state == pci_channel_io_perm_failure ? PCI_ERS_RESULT_DISCONNECT : PCI_ERS_RESULT_NEED_RESET; } diff --git a/include/linux/mlx5/driver.h b/include/linux/mlx5/driver.h index 5dbda60a09f4..7d9a5d08eb59 100644 --- a/include/linux/mlx5/driver.h +++ b/include/linux/mlx5/driver.h @@ -418,7 +418,10 @@ struct mlx5_core_health { u32 prev; int miss_counter; bool sick; + /* wq spinlock to synchronize draining */ + spinlock_t wq_lock; struct workqueue_struct *wq; + unsigned long flags; struct work_struct work; }; @@ -778,6 +781,7 @@ void mlx5_health_cleanup(struct mlx5_core_dev *dev); int mlx5_health_init(struct mlx5_core_dev *dev); void mlx5_start_health_poll(struct mlx5_core_dev *dev); void mlx5_stop_health_poll(struct mlx5_core_dev *dev); +void mlx5_drain_health_wq(struct mlx5_core_dev *dev); int mlx5_buf_alloc_node(struct mlx5_core_dev *dev, int size, struct mlx5_buf *buf, int node); int mlx5_buf_alloc(struct mlx5_core_dev *dev, int size, struct mlx5_buf *buf); From 04c0c1ab38e95105d950db5b84e727637e149ce7 Mon Sep 17 00:00:00 2001 From: Mohamad Haj Yahia Date: Tue, 25 Oct 2016 18:36:34 +0300 Subject: [PATCH 11/12] net/mlx5: PCI error recovery health care simulation In case that the kernel PCI error handlers are not called, we will trigger our own recovery flow. The health work will give priority to the kernel pci error handlers to recover the PCI by waiting for a small period, if the pci error handlers are not triggered the manual recovery flow will be executed. We don't save pci state in case of manual recovery because it will ruin the pci configuration space and we will lose dma sync. Fixes: 89d44f0a6c73 ('net/mlx5_core: Add pci error handlers to mlx5_core driver') Signed-off-by: Mohamad Haj Yahia Signed-off-by: Saeed Mahameed Signed-off-by: David S. Miller --- .../net/ethernet/mellanox/mlx5/core/health.c | 45 +++++++++++++++++-- .../net/ethernet/mellanox/mlx5/core/main.c | 18 +++++--- .../ethernet/mellanox/mlx5/core/mlx5_core.h | 1 + include/linux/mlx5/driver.h | 1 + 4 files changed, 56 insertions(+), 9 deletions(-) diff --git a/drivers/net/ethernet/mellanox/mlx5/core/health.c b/drivers/net/ethernet/mellanox/mlx5/core/health.c index 2d00022c92d8..5bcf93422ee0 100644 --- a/drivers/net/ethernet/mellanox/mlx5/core/health.c +++ b/drivers/net/ethernet/mellanox/mlx5/core/health.c @@ -61,14 +61,15 @@ enum { enum { MLX5_NIC_IFC_FULL = 0, MLX5_NIC_IFC_DISABLED = 1, - MLX5_NIC_IFC_NO_DRAM_NIC = 2 + MLX5_NIC_IFC_NO_DRAM_NIC = 2, + MLX5_NIC_IFC_INVALID = 3 }; enum { MLX5_DROP_NEW_HEALTH_WORK, }; -static u8 get_nic_interface(struct mlx5_core_dev *dev) +static u8 get_nic_state(struct mlx5_core_dev *dev) { return (ioread32be(&dev->iseg->cmdq_addr_l_sz) >> 8) & 3; } @@ -101,7 +102,7 @@ static int in_fatal(struct mlx5_core_dev *dev) struct mlx5_core_health *health = &dev->priv.health; struct health_buffer __iomem *h = health->health; - if (get_nic_interface(dev) == MLX5_NIC_IFC_DISABLED) + if (get_nic_state(dev) == MLX5_NIC_IFC_DISABLED) return 1; if (ioread32be(&h->fw_ver) == 0xffffffff) @@ -131,7 +132,7 @@ unlock: static void mlx5_handle_bad_state(struct mlx5_core_dev *dev) { - u8 nic_interface = get_nic_interface(dev); + u8 nic_interface = get_nic_state(dev); switch (nic_interface) { case MLX5_NIC_IFC_FULL: @@ -153,8 +154,34 @@ static void mlx5_handle_bad_state(struct mlx5_core_dev *dev) mlx5_disable_device(dev); } +static void health_recover(struct work_struct *work) +{ + struct mlx5_core_health *health; + struct delayed_work *dwork; + struct mlx5_core_dev *dev; + struct mlx5_priv *priv; + u8 nic_state; + + dwork = container_of(work, struct delayed_work, work); + health = container_of(dwork, struct mlx5_core_health, recover_work); + priv = container_of(health, struct mlx5_priv, health); + dev = container_of(priv, struct mlx5_core_dev, priv); + + nic_state = get_nic_state(dev); + if (nic_state == MLX5_NIC_IFC_INVALID) { + dev_err(&dev->pdev->dev, "health recovery flow aborted since the nic state is invalid\n"); + return; + } + + dev_err(&dev->pdev->dev, "starting health recovery flow\n"); + mlx5_recover_device(dev); +} + +/* How much time to wait until health resetting the driver (in msecs) */ +#define MLX5_RECOVERY_DELAY_MSECS 60000 static void health_care(struct work_struct *work) { + unsigned long recover_delay = msecs_to_jiffies(MLX5_RECOVERY_DELAY_MSECS); struct mlx5_core_health *health; struct mlx5_core_dev *dev; struct mlx5_priv *priv; @@ -164,6 +191,14 @@ static void health_care(struct work_struct *work) dev = container_of(priv, struct mlx5_core_dev, priv); mlx5_core_warn(dev, "handling bad device here\n"); mlx5_handle_bad_state(dev); + + spin_lock(&health->wq_lock); + if (!test_bit(MLX5_DROP_NEW_HEALTH_WORK, &health->flags)) + schedule_delayed_work(&health->recover_work, recover_delay); + else + dev_err(&dev->pdev->dev, + "new health works are not permitted at this stage\n"); + spin_unlock(&health->wq_lock); } static const char *hsynd_str(u8 synd) @@ -316,6 +351,7 @@ void mlx5_drain_health_wq(struct mlx5_core_dev *dev) spin_lock(&health->wq_lock); set_bit(MLX5_DROP_NEW_HEALTH_WORK, &health->flags); spin_unlock(&health->wq_lock); + cancel_delayed_work_sync(&health->recover_work); cancel_work_sync(&health->work); } @@ -344,6 +380,7 @@ int mlx5_health_init(struct mlx5_core_dev *dev) return -ENOMEM; spin_lock_init(&health->wq_lock); INIT_WORK(&health->work, health_care); + INIT_DELAYED_WORK(&health->recover_work, health_recover); return 0; } diff --git a/drivers/net/ethernet/mellanox/mlx5/core/main.c b/drivers/net/ethernet/mellanox/mlx5/core/main.c index 9f90226bc120..d5433c49b2b0 100644 --- a/drivers/net/ethernet/mellanox/mlx5/core/main.c +++ b/drivers/net/ethernet/mellanox/mlx5/core/main.c @@ -1313,6 +1313,7 @@ static pci_ers_result_t mlx5_pci_err_detected(struct pci_dev *pdev, struct mlx5_priv *priv = &dev->priv; dev_info(&pdev->dev, "%s was called\n", __func__); + mlx5_enter_error_state(dev); mlx5_unload_one(dev, priv, false); /* In case of kernel call save the pci state and drain health wq */ @@ -1378,11 +1379,6 @@ static pci_ers_result_t mlx5_pci_slot_reset(struct pci_dev *pdev) return PCI_ERS_RESULT_RECOVERED; } -void mlx5_disable_device(struct mlx5_core_dev *dev) -{ - mlx5_pci_err_detected(dev->pdev, 0); -} - static void mlx5_pci_resume(struct pci_dev *pdev) { struct mlx5_core_dev *dev = pci_get_drvdata(pdev); @@ -1432,6 +1428,18 @@ static const struct pci_device_id mlx5_core_pci_table[] = { MODULE_DEVICE_TABLE(pci, mlx5_core_pci_table); +void mlx5_disable_device(struct mlx5_core_dev *dev) +{ + mlx5_pci_err_detected(dev->pdev, 0); +} + +void mlx5_recover_device(struct mlx5_core_dev *dev) +{ + mlx5_pci_disable_device(dev); + if (mlx5_pci_slot_reset(dev->pdev) == PCI_ERS_RESULT_RECOVERED) + mlx5_pci_resume(dev->pdev); +} + static struct pci_driver mlx5_core_driver = { .name = DRIVER_NAME, .id_table = mlx5_core_pci_table, diff --git a/drivers/net/ethernet/mellanox/mlx5/core/mlx5_core.h b/drivers/net/ethernet/mellanox/mlx5/core/mlx5_core.h index 3d0cfb9f18f9..187662c8ea96 100644 --- a/drivers/net/ethernet/mellanox/mlx5/core/mlx5_core.h +++ b/drivers/net/ethernet/mellanox/mlx5/core/mlx5_core.h @@ -83,6 +83,7 @@ void mlx5_core_event(struct mlx5_core_dev *dev, enum mlx5_dev_event event, unsigned long param); void mlx5_enter_error_state(struct mlx5_core_dev *dev); void mlx5_disable_device(struct mlx5_core_dev *dev); +void mlx5_recover_device(struct mlx5_core_dev *dev); int mlx5_sriov_init(struct mlx5_core_dev *dev); void mlx5_sriov_cleanup(struct mlx5_core_dev *dev); int mlx5_sriov_attach(struct mlx5_core_dev *dev); diff --git a/include/linux/mlx5/driver.h b/include/linux/mlx5/driver.h index 7d9a5d08eb59..ecc451d89ccd 100644 --- a/include/linux/mlx5/driver.h +++ b/include/linux/mlx5/driver.h @@ -423,6 +423,7 @@ struct mlx5_core_health { struct workqueue_struct *wq; unsigned long flags; struct work_struct work; + struct delayed_work recover_work; }; struct mlx5_cq_table { From 6b276190c50a12511d889d9079ffb901ff94a822 Mon Sep 17 00:00:00 2001 From: Noa Osherovich Date: Tue, 25 Oct 2016 18:36:35 +0300 Subject: [PATCH 12/12] net/mlx5: Avoid passing dma address 0 to firmware Currently the firmware can't work with a page with dma address 0. Passing such an address to the firmware will cause the give_pages command to fail. To avoid this, in case we get a 0 dma address of a page from the dma engine, we avoid passing it to FW by remapping to get an address other than 0. Fixes: bf0bf77f6519 ('mlx5: Support communicating arbitrary host...') Signed-off-by: Noa Osherovich Signed-off-by: Saeed Mahameed Signed-off-by: David S. Miller --- .../ethernet/mellanox/mlx5/core/pagealloc.c | 26 +++++++++++++------ 1 file changed, 18 insertions(+), 8 deletions(-) diff --git a/drivers/net/ethernet/mellanox/mlx5/core/pagealloc.c b/drivers/net/ethernet/mellanox/mlx5/core/pagealloc.c index cc4fd61914d3..a57d5a81eb05 100644 --- a/drivers/net/ethernet/mellanox/mlx5/core/pagealloc.c +++ b/drivers/net/ethernet/mellanox/mlx5/core/pagealloc.c @@ -209,6 +209,7 @@ static void free_4k(struct mlx5_core_dev *dev, u64 addr) static int alloc_system_page(struct mlx5_core_dev *dev, u16 func_id) { struct page *page; + u64 zero_addr = 1; u64 addr; int err; int nid = dev_to_node(&dev->pdev->dev); @@ -218,26 +219,35 @@ static int alloc_system_page(struct mlx5_core_dev *dev, u16 func_id) mlx5_core_warn(dev, "failed to allocate page\n"); return -ENOMEM; } +map: addr = dma_map_page(&dev->pdev->dev, page, 0, PAGE_SIZE, DMA_BIDIRECTIONAL); if (dma_mapping_error(&dev->pdev->dev, addr)) { mlx5_core_warn(dev, "failed dma mapping page\n"); err = -ENOMEM; - goto out_alloc; + goto err_mapping; } + + /* Firmware doesn't support page with physical address 0 */ + if (addr == 0) { + zero_addr = addr; + goto map; + } + err = insert_page(dev, addr, page, func_id); if (err) { mlx5_core_err(dev, "failed to track allocated page\n"); - goto out_mapping; + dma_unmap_page(&dev->pdev->dev, addr, PAGE_SIZE, + DMA_BIDIRECTIONAL); } - return 0; +err_mapping: + if (err) + __free_page(page); -out_mapping: - dma_unmap_page(&dev->pdev->dev, addr, PAGE_SIZE, DMA_BIDIRECTIONAL); - -out_alloc: - __free_page(page); + if (zero_addr == 0) + dma_unmap_page(&dev->pdev->dev, zero_addr, PAGE_SIZE, + DMA_BIDIRECTIONAL); return err; }