From 17b3222e943701c408d6e03357967829aaea0550 Mon Sep 17 00:00:00 2001 From: Yevgeny Kliteynik Date: Sun, 13 Nov 2022 00:04:17 +0200 Subject: net/mlx5: DR, Fix 'stack frame size exceeds limit' error in dr_rule If the kernel configuration asks the compiler to check frame limit of 1K, dr_rule_create_rule_nic exceed this limit: "stack frame size (1184) exceeds limit (1024)" Fixing this issue by checking configured frame limit and using the optimization STE array only for cases with the usual 2K (or larger) stack size warning. Fixes: b9b81e1e9382 ("net/mlx5: DR, For short chains of STEs, avoid allocating ste_arr dynamically") Reported-by: kernel test robot Signed-off-by: Yevgeny Kliteynik --- drivers/net/ethernet/mellanox/mlx5/core/steering/dr_rule.c | 11 +++++++---- 1 file changed, 7 insertions(+), 4 deletions(-) diff --git a/drivers/net/ethernet/mellanox/mlx5/core/steering/dr_rule.c b/drivers/net/ethernet/mellanox/mlx5/core/steering/dr_rule.c index 74cbe53ee9db..b851141e03de 100644 --- a/drivers/net/ethernet/mellanox/mlx5/core/steering/dr_rule.c +++ b/drivers/net/ethernet/mellanox/mlx5/core/steering/dr_rule.c @@ -3,7 +3,12 @@ #include "dr_types.h" +#if defined(CONFIG_FRAME_WARN) && (CONFIG_FRAME_WARN < 2048) +/* don't try to optimize STE allocation if the stack is too constaraining */ +#define DR_RULE_MAX_STES_OPTIMIZED 0 +#else #define DR_RULE_MAX_STES_OPTIMIZED 5 +#endif #define DR_RULE_MAX_STE_CHAIN_OPTIMIZED (DR_RULE_MAX_STES_OPTIMIZED + DR_ACTION_MAX_STES) static int dr_rule_append_to_miss_list(struct mlx5dr_domain *dmn, @@ -1218,10 +1223,7 @@ dr_rule_create_rule_nic(struct mlx5dr_rule *rule, mlx5dr_domain_nic_unlock(nic_dmn); - if (unlikely(!hw_ste_arr_is_opt)) - kfree(hw_ste_arr); - - return 0; + goto out; free_rule: dr_rule_clean_rule_members(rule, nic_rule); @@ -1238,6 +1240,7 @@ remove_from_nic_tbl: free_hw_ste: mlx5dr_domain_nic_unlock(nic_dmn); +out: if (unlikely(!hw_ste_arr_is_opt)) kfree(hw_ste_arr); -- cgit From e0bf81bf0d3d4747c146e0bf44774d3d881d7137 Mon Sep 17 00:00:00 2001 From: Ariel Levkovich Date: Tue, 16 Aug 2022 23:19:11 +0300 Subject: net/mlx5: check attr pointer validity before dereferencing it Fix attr pointer validity checks after it was already dereferenced. Fixes: cb0d54cbf948 ("net/mlx5e: Fix wrong source vport matching on tunnel rule") Signed-off-by: Ariel Levkovich Reported-by: kernel test robot Reported-by: Dan Carpenter Signed-off-by: Saeed Mahameed --- drivers/net/ethernet/mellanox/mlx5/core/eswitch_offloads.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/drivers/net/ethernet/mellanox/mlx5/core/eswitch_offloads.c b/drivers/net/ethernet/mellanox/mlx5/core/eswitch_offloads.c index e455b215c708..75b77dd2392b 100644 --- a/drivers/net/ethernet/mellanox/mlx5/core/eswitch_offloads.c +++ b/drivers/net/ethernet/mellanox/mlx5/core/eswitch_offloads.c @@ -143,7 +143,7 @@ mlx5_eswitch_set_rule_source_port(struct mlx5_eswitch *esw, if (mlx5_esw_indir_table_decap_vport(attr)) vport = mlx5_esw_indir_table_decap_vport(attr); - if (attr && !attr->chain && esw_attr->int_port) + if (!attr->chain && esw_attr && esw_attr->int_port) metadata = mlx5e_tc_int_port_get_metadata_for_match(esw_attr->int_port); else -- cgit From 5e72f3f1c558019082cfeedeed73748f35d780c6 Mon Sep 17 00:00:00 2001 From: Ariel Levkovich Date: Tue, 30 Aug 2022 00:32:30 +0300 Subject: net/mlx5e: TC, Keep mod hdr actions after mod hdr alloc When offloading TC NIC rule which has mod_hdr action, the mod_hdr actions list is freed upon mod_hdr allocation. In the new format of handling multi table actions and CT in particular, the mod_hdr actions list is still relevant when setting the pre and post rules and therefore, freeing the list may cause adding rules which don't set the FTE_ID. Therefore, the mod_hdr actions list needs to be kept for the pre/post flows as well and should be left for these handler to be freed. Fixes: 8300f225268b ("net/mlx5e: Create new flow attr for multi table actions") Signed-off-by: Ariel Levkovich Signed-off-by: Saeed Mahameed --- drivers/net/ethernet/mellanox/mlx5/core/en_tc.c | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-) diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_tc.c b/drivers/net/ethernet/mellanox/mlx5/core/en_tc.c index 9af2aa2922f5..dbadaf166487 100644 --- a/drivers/net/ethernet/mellanox/mlx5/core/en_tc.c +++ b/drivers/net/ethernet/mellanox/mlx5/core/en_tc.c @@ -1301,7 +1301,6 @@ mlx5e_tc_add_nic_flow(struct mlx5e_priv *priv, if (attr->action & MLX5_FLOW_CONTEXT_ACTION_MOD_HDR) { err = mlx5e_attach_mod_hdr(priv, flow, parse_attr); - mlx5e_mod_hdr_dealloc(&parse_attr->mod_hdr_acts); if (err) return err; } @@ -1359,8 +1358,10 @@ static void mlx5e_tc_del_nic_flow(struct mlx5e_priv *priv, } mutex_unlock(&tc->t_lock); - if (attr->action & MLX5_FLOW_CONTEXT_ACTION_MOD_HDR) + if (attr->action & MLX5_FLOW_CONTEXT_ACTION_MOD_HDR) { + mlx5e_mod_hdr_dealloc(&attr->parse_attr->mod_hdr_acts); mlx5e_detach_mod_hdr(priv, flow); + } if (attr->action & MLX5_FLOW_CONTEXT_ACTION_COUNT) mlx5_fc_destroy(priv->mdev, attr->counter); -- cgit From da2e552b469a0cd130ff70a88ccc4139da428a65 Mon Sep 17 00:00:00 2001 From: Moshe Shemesh Date: Mon, 28 Nov 2022 19:05:47 +0200 Subject: net/mlx5: Fix command stats access after free Command may fail while driver is reloading and can't accept FW commands till command interface is reinitialized. Such command failure is being logged to command stats. This results in NULL pointer access as command stats structure is being freed and reallocated during mlx5 devlink reload (see kernel log below). Fix it by making command stats statically allocated on driver probe. Kernel log: [ 2394.808802] BUG: unable to handle kernel paging request at 000000000002a9c0 [ 2394.810610] PGD 0 P4D 0 [ 2394.811811] Oops: 0002 [#1] SMP NOPTI ... [ 2394.815482] RIP: 0010:native_queued_spin_lock_slowpath+0x183/0x1d0 ... [ 2394.829505] Call Trace: [ 2394.830667] _raw_spin_lock_irq+0x23/0x26 [ 2394.831858] cmd_status_err+0x55/0x110 [mlx5_core] [ 2394.833020] mlx5_access_reg+0xe7/0x150 [mlx5_core] [ 2394.834175] mlx5_query_port_ptys+0x78/0xa0 [mlx5_core] [ 2394.835337] mlx5e_ethtool_get_link_ksettings+0x74/0x590 [mlx5_core] [ 2394.836454] ? kmem_cache_alloc_trace+0x140/0x1c0 [ 2394.837562] __rh_call_get_link_ksettings+0x33/0x100 [ 2394.838663] ? __rtnl_unlock+0x25/0x50 [ 2394.839755] __ethtool_get_link_ksettings+0x72/0x150 [ 2394.840862] duplex_show+0x6e/0xc0 [ 2394.841963] dev_attr_show+0x1c/0x40 [ 2394.843048] sysfs_kf_seq_show+0x9b/0x100 [ 2394.844123] seq_read+0x153/0x410 [ 2394.845187] vfs_read+0x91/0x140 [ 2394.846226] ksys_read+0x4f/0xb0 [ 2394.847234] do_syscall_64+0x5b/0x1a0 [ 2394.848228] entry_SYSCALL_64_after_hwframe+0x65/0xca Fixes: 34f46ae0d4b3 ("net/mlx5: Add command failures data to debugfs") Signed-off-by: Moshe Shemesh Reviewed-by: Shay Drory Signed-off-by: Saeed Mahameed --- drivers/net/ethernet/mellanox/mlx5/core/cmd.c | 13 ++----------- include/linux/mlx5/driver.h | 2 +- 2 files changed, 3 insertions(+), 12 deletions(-) diff --git a/drivers/net/ethernet/mellanox/mlx5/core/cmd.c b/drivers/net/ethernet/mellanox/mlx5/core/cmd.c index d3ca745d107d..c837103a9ee3 100644 --- a/drivers/net/ethernet/mellanox/mlx5/core/cmd.c +++ b/drivers/net/ethernet/mellanox/mlx5/core/cmd.c @@ -2176,15 +2176,9 @@ int mlx5_cmd_init(struct mlx5_core_dev *dev) return -EINVAL; } - cmd->stats = kvcalloc(MLX5_CMD_OP_MAX, sizeof(*cmd->stats), GFP_KERNEL); - if (!cmd->stats) - return -ENOMEM; - cmd->pool = dma_pool_create("mlx5_cmd", mlx5_core_dma_dev(dev), size, align, 0); - if (!cmd->pool) { - err = -ENOMEM; - goto dma_pool_err; - } + if (!cmd->pool) + return -ENOMEM; err = alloc_cmd_page(dev, cmd); if (err) @@ -2268,8 +2262,6 @@ err_free_page: err_free_pool: dma_pool_destroy(cmd->pool); -dma_pool_err: - kvfree(cmd->stats); return err; } @@ -2282,7 +2274,6 @@ void mlx5_cmd_cleanup(struct mlx5_core_dev *dev) destroy_msg_cache(dev); free_cmd_page(dev, cmd); dma_pool_destroy(cmd->pool); - kvfree(cmd->stats); } void mlx5_cmd_set_state(struct mlx5_core_dev *dev, diff --git a/include/linux/mlx5/driver.h b/include/linux/mlx5/driver.h index d476255c9a3f..76ef2e4fde38 100644 --- a/include/linux/mlx5/driver.h +++ b/include/linux/mlx5/driver.h @@ -315,7 +315,7 @@ struct mlx5_cmd { struct mlx5_cmd_debug dbg; struct cmd_msg_cache cache[MLX5_NUM_COMMAND_CACHES]; int checksum_disabled; - struct mlx5_cmd_stats *stats; + struct mlx5_cmd_stats stats[MLX5_CMD_OP_MAX]; }; struct mlx5_cmd_mailbox { -- cgit From ab4b01bfdaa69492fb36484026b0a0f0af02d75a Mon Sep 17 00:00:00 2001 From: Roy Novich Date: Wed, 4 Jan 2023 11:16:21 +0200 Subject: net/mlx5e: Verify dev is present for fix features ndo The native NIC port net device instance is being used as Uplink representor. While changing profiles private resources are not available, fix features ndo does not check if the netdev is present. Add driver protection to verify private resources are ready. Fixes: 7a9fb35e8c3a ("net/mlx5e: Do not reload ethernet ports when changing eswitch mode") Signed-off-by: Roy Novich Reviewed-by: Roi Dayan Signed-off-by: Saeed Mahameed --- drivers/net/ethernet/mellanox/mlx5/core/en_main.c | 3 +++ 1 file changed, 3 insertions(+) diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_main.c b/drivers/net/ethernet/mellanox/mlx5/core/en_main.c index cff5f2e29e1e..abcc614b6191 100644 --- a/drivers/net/ethernet/mellanox/mlx5/core/en_main.c +++ b/drivers/net/ethernet/mellanox/mlx5/core/en_main.c @@ -4084,6 +4084,9 @@ static netdev_features_t mlx5e_fix_features(struct net_device *netdev, struct mlx5e_vlan_table *vlan; struct mlx5e_params *params; + if (!netif_device_present(netdev)) + return features; + vlan = mlx5e_fs_get_vlan(priv->fs); mutex_lock(&priv->state_lock); params = &priv->channels.params; -- cgit From 806a8df7126a8c05d60411eeb81057c2a8bbe7a7 Mon Sep 17 00:00:00 2001 From: Dragos Tatulea Date: Thu, 15 Dec 2022 13:02:38 +0200 Subject: net/mlx5e: IPoIB, Block queue count configuration when sub interfaces are present PKEY sub interfaces share the receive queues with the parent interface. While setting the sub interface queue count is not supported, it is currently possible to change the number of queues of the parent interface. Thus we can end up with inconsistent queue sizes between the parent and its sub interfaces. This change disallows setting the queue count on the parent interface when sub interfaces are present. This is achieved by introducing an explicit reference to the parent netdev in the mlx5i_priv of the child interface. An additional counter is also required on the parent side to detect when sub interfaces are attached and for proper cleanup. The rtnl lock is taken during the ethtool op and the sub interface ndo_init/uninit ops. There is no race here around counting the sub interfaces, reading the sub interfaces and setting the number of channels. The ASSERT_RTNL was added to document that. Fixes: be98737a4faa ("net/mlx5e: Use dynamic per-channel allocations in stats") Signed-off-by: Dragos Tatulea Reviewed-by: Tariq Toukan Signed-off-by: Saeed Mahameed --- .../ethernet/mellanox/mlx5/core/ipoib/ethtool.c | 16 +++++++-- .../net/ethernet/mellanox/mlx5/core/ipoib/ipoib.c | 38 ++++++++++++++++++++++ .../net/ethernet/mellanox/mlx5/core/ipoib/ipoib.h | 6 ++++ .../ethernet/mellanox/mlx5/core/ipoib/ipoib_vlan.c | 9 +++-- 4 files changed, 62 insertions(+), 7 deletions(-) diff --git a/drivers/net/ethernet/mellanox/mlx5/core/ipoib/ethtool.c b/drivers/net/ethernet/mellanox/mlx5/core/ipoib/ethtool.c index c247cca154e9..eff92dc0927c 100644 --- a/drivers/net/ethernet/mellanox/mlx5/core/ipoib/ethtool.c +++ b/drivers/net/ethernet/mellanox/mlx5/core/ipoib/ethtool.c @@ -90,9 +90,21 @@ static void mlx5i_get_ringparam(struct net_device *dev, static int mlx5i_set_channels(struct net_device *dev, struct ethtool_channels *ch) { - struct mlx5e_priv *priv = mlx5i_epriv(dev); + struct mlx5i_priv *ipriv = netdev_priv(dev); + struct mlx5e_priv *epriv = mlx5i_epriv(dev); + + /* rtnl lock protects from race between this ethtool op and sub + * interface ndo_init/uninit. + */ + ASSERT_RTNL(); + if (ipriv->num_sub_interfaces > 0) { + mlx5_core_warn(epriv->mdev, + "can't change number of channels for interfaces with sub interfaces (%u)\n", + ipriv->num_sub_interfaces); + return -EINVAL; + } - return mlx5e_ethtool_set_channels(priv, ch); + return mlx5e_ethtool_set_channels(epriv, ch); } static void mlx5i_get_channels(struct net_device *dev, diff --git a/drivers/net/ethernet/mellanox/mlx5/core/ipoib/ipoib.c b/drivers/net/ethernet/mellanox/mlx5/core/ipoib/ipoib.c index 2c73c8445e63..911cf4d23964 100644 --- a/drivers/net/ethernet/mellanox/mlx5/core/ipoib/ipoib.c +++ b/drivers/net/ethernet/mellanox/mlx5/core/ipoib/ipoib.c @@ -160,6 +160,44 @@ void mlx5i_get_stats(struct net_device *dev, struct rtnl_link_stats64 *stats) stats->tx_dropped = sstats->tx_queue_dropped; } +struct net_device *mlx5i_parent_get(struct net_device *netdev) +{ + struct mlx5e_priv *priv = mlx5i_epriv(netdev); + struct mlx5i_priv *ipriv, *parent_ipriv; + struct net_device *parent_dev; + int parent_ifindex; + + ipriv = priv->ppriv; + + parent_ifindex = netdev->netdev_ops->ndo_get_iflink(netdev); + parent_dev = dev_get_by_index(dev_net(netdev), parent_ifindex); + if (!parent_dev) + return NULL; + + parent_ipriv = netdev_priv(parent_dev); + + ASSERT_RTNL(); + parent_ipriv->num_sub_interfaces++; + + ipriv->parent_dev = parent_dev; + + return parent_dev; +} + +void mlx5i_parent_put(struct net_device *netdev) +{ + struct mlx5e_priv *priv = mlx5i_epriv(netdev); + struct mlx5i_priv *ipriv, *parent_ipriv; + + ipriv = priv->ppriv; + parent_ipriv = netdev_priv(ipriv->parent_dev); + + ASSERT_RTNL(); + parent_ipriv->num_sub_interfaces--; + + dev_put(ipriv->parent_dev); +} + int mlx5i_init_underlay_qp(struct mlx5e_priv *priv) { struct mlx5_core_dev *mdev = priv->mdev; diff --git a/drivers/net/ethernet/mellanox/mlx5/core/ipoib/ipoib.h b/drivers/net/ethernet/mellanox/mlx5/core/ipoib/ipoib.h index 99d46fda9f82..f3f2af972020 100644 --- a/drivers/net/ethernet/mellanox/mlx5/core/ipoib/ipoib.h +++ b/drivers/net/ethernet/mellanox/mlx5/core/ipoib/ipoib.h @@ -54,9 +54,11 @@ struct mlx5i_priv { struct rdma_netdev rn; /* keep this first */ u32 qpn; bool sub_interface; + u32 num_sub_interfaces; u32 qkey; u16 pkey_index; struct mlx5i_pkey_qpn_ht *qpn_htbl; + struct net_device *parent_dev; char *mlx5e_priv[]; }; @@ -117,5 +119,9 @@ void mlx5i_sq_xmit(struct mlx5e_txqsq *sq, struct sk_buff *skb, struct mlx5_av *av, u32 dqpn, u32 dqkey, bool xmit_more); void mlx5i_get_stats(struct net_device *dev, struct rtnl_link_stats64 *stats); +/* Reference management for child to parent interfaces. */ +struct net_device *mlx5i_parent_get(struct net_device *netdev); +void mlx5i_parent_put(struct net_device *netdev); + #endif /* CONFIG_MLX5_CORE_IPOIB */ #endif /* __MLX5E_IPOB_H__ */ diff --git a/drivers/net/ethernet/mellanox/mlx5/core/ipoib/ipoib_vlan.c b/drivers/net/ethernet/mellanox/mlx5/core/ipoib/ipoib_vlan.c index 4d9c9e49645c..28795fb6bccc 100644 --- a/drivers/net/ethernet/mellanox/mlx5/core/ipoib/ipoib_vlan.c +++ b/drivers/net/ethernet/mellanox/mlx5/core/ipoib/ipoib_vlan.c @@ -158,21 +158,19 @@ static int mlx5i_pkey_dev_init(struct net_device *dev) struct mlx5e_priv *priv = mlx5i_epriv(dev); struct mlx5i_priv *ipriv, *parent_ipriv; struct net_device *parent_dev; - int parent_ifindex; ipriv = priv->ppriv; - /* Get QPN to netdevice hash table from parent */ - parent_ifindex = dev->netdev_ops->ndo_get_iflink(dev); - parent_dev = dev_get_by_index(dev_net(dev), parent_ifindex); + /* Link to parent */ + parent_dev = mlx5i_parent_get(dev); if (!parent_dev) { mlx5_core_warn(priv->mdev, "failed to get parent device\n"); return -EINVAL; } + /* Get QPN to netdevice hash table from parent */ parent_ipriv = netdev_priv(parent_dev); ipriv->qpn_htbl = parent_ipriv->qpn_htbl; - dev_put(parent_dev); return mlx5i_dev_init(dev); } @@ -184,6 +182,7 @@ static int mlx5i_pkey_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd) static void mlx5i_pkey_dev_cleanup(struct net_device *netdev) { + mlx5i_parent_put(netdev); return mlx5i_dev_cleanup(netdev); } -- cgit From 31c70bfe58ef09fe36327ddcced9143a16e9e83d Mon Sep 17 00:00:00 2001 From: Dragos Tatulea Date: Fri, 25 Nov 2022 17:51:19 +0200 Subject: net/mlx5e: IPoIB, Block PKEY interfaces with less rx queues than parent A user is able to configure an arbitrary number of rx queues when creating an interface via netlink. This doesn't work for child PKEY interfaces because the child interface uses the parent receive channels. Although the child shares the parent's receive channels, the number of rx queues is important for the channel_stats array: the parent's rx channel index is used to access the child's channel_stats. So the array has to be at least as large as the parent's rx queue size for the counting to work correctly and to prevent out of bound accesses. This patch checks for the mentioned scenario and returns an error when trying to create the interface. The error is propagated to the user. Fixes: be98737a4faa ("net/mlx5e: Use dynamic per-channel allocations in stats") Signed-off-by: Dragos Tatulea Reviewed-by: Tariq Toukan Signed-off-by: Saeed Mahameed --- drivers/net/ethernet/mellanox/mlx5/core/ipoib/ipoib_vlan.c | 9 +++++++++ 1 file changed, 9 insertions(+) diff --git a/drivers/net/ethernet/mellanox/mlx5/core/ipoib/ipoib_vlan.c b/drivers/net/ethernet/mellanox/mlx5/core/ipoib/ipoib_vlan.c index 28795fb6bccc..03e681297937 100644 --- a/drivers/net/ethernet/mellanox/mlx5/core/ipoib/ipoib_vlan.c +++ b/drivers/net/ethernet/mellanox/mlx5/core/ipoib/ipoib_vlan.c @@ -168,6 +168,15 @@ static int mlx5i_pkey_dev_init(struct net_device *dev) return -EINVAL; } + if (dev->num_rx_queues < parent_dev->real_num_rx_queues) { + mlx5_core_warn(priv->mdev, + "failed to create child device with rx queues [%d] less than parent's [%d]\n", + dev->num_rx_queues, + parent_dev->real_num_rx_queues); + mlx5i_parent_put(dev); + return -EINVAL; + } + /* Get QPN to netdevice hash table from parent */ parent_ipriv = netdev_priv(parent_dev); ipriv->qpn_htbl = parent_ipriv->qpn_htbl; -- cgit From b5e23931c45a2f99f60a2f2b98a9e4d5a62a5b13 Mon Sep 17 00:00:00 2001 From: Dragos Tatulea Date: Wed, 23 Nov 2022 16:59:13 +0200 Subject: net/mlx5e: IPoIB, Fix child PKEY interface stats on rx path The current code always does the accounting using the stats from the parent interface (linked in the rq). This doesn't work when there are child interfaces configured. Fix this behavior by always using the stats from the child interface priv. This will also work for parent only interfaces: the child (netdev) and parent netdev (rq->netdev) will point to the same thing. Fixes: be98737a4faa ("net/mlx5e: Use dynamic per-channel allocations in stats") Signed-off-by: Dragos Tatulea Reviewed-by: Tariq Toukan Signed-off-by: Saeed Mahameed --- drivers/net/ethernet/mellanox/mlx5/core/en_rx.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_rx.c b/drivers/net/ethernet/mellanox/mlx5/core/en_rx.c index c8820ab22169..3df455f6b168 100644 --- a/drivers/net/ethernet/mellanox/mlx5/core/en_rx.c +++ b/drivers/net/ethernet/mellanox/mlx5/core/en_rx.c @@ -2419,7 +2419,7 @@ static inline void mlx5i_complete_rx_cqe(struct mlx5e_rq *rq, priv = mlx5i_epriv(netdev); tstamp = &priv->tstamp; - stats = rq->stats; + stats = &priv->channel_stats[rq->ix]->rq; flags_rqpn = be32_to_cpu(cqe->flags_rqpn); g = (flags_rqpn >> 28) & 3; -- cgit From 2414c9b7a29d237c9c40abd62853dbc08b4ba7df Mon Sep 17 00:00:00 2001 From: Oz Shlomo Date: Wed, 7 Dec 2022 11:05:02 +0000 Subject: net/mlx5e: TC, ignore match level for post meter rules The post meter table only matches on reg_c5. As such, the inner/outer match levels are irrelevant for the match critieria. The cited patch only sets the outer criteria to none, thus setting the inner match level for encapsulated packets. This caused rules with police action on tunnel devices to not find an existing flow group for the match criteria, thus failing to offload the rule. Set both the inner and outer match levels to none for post_meter rules. Fixes: 0d8c38d44f33 ("net/mlx5e: TC, init post meter rules with branching attributes") Signed-off-by: Oz Shlomo Reviewed-by: Roi Dayan Signed-off-by: Saeed Mahameed --- drivers/net/ethernet/mellanox/mlx5/core/en/tc/post_meter.c | 1 + 1 file changed, 1 insertion(+) diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en/tc/post_meter.c b/drivers/net/ethernet/mellanox/mlx5/core/en/tc/post_meter.c index 8d7d761482d2..50b60fd00946 100644 --- a/drivers/net/ethernet/mellanox/mlx5/core/en/tc/post_meter.c +++ b/drivers/net/ethernet/mellanox/mlx5/core/en/tc/post_meter.c @@ -127,6 +127,7 @@ mlx5e_post_meter_add_rule(struct mlx5e_priv *priv, attr->counter = act_counter; attr->flags |= MLX5_ATTR_FLAG_NO_IN_PORT; + attr->inner_match_level = MLX5_MATCH_NONE; attr->outer_match_level = MLX5_MATCH_NONE; attr->chain = 0; attr->prio = 0; -- cgit From c09502d54dc109530ccb9a8910ab286d0745f119 Mon Sep 17 00:00:00 2001 From: Oz Shlomo Date: Thu, 8 Dec 2022 08:31:48 +0000 Subject: net/mlx5e: TC, Restore pkt rate policing support The offending commit removed the support for all packet rate metering. Restore the pkt rate metering support by removing the restriction. Fixes: 3603f26633e7 ("net/mlx5e: TC, allow meter jump control action") Signed-off-by: Oz Shlomo Reviewed-by: Roi Dayan Signed-off-by: Saeed Mahameed --- drivers/net/ethernet/mellanox/mlx5/core/en/tc/act/police.c | 6 ------ 1 file changed, 6 deletions(-) diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en/tc/act/police.c b/drivers/net/ethernet/mellanox/mlx5/core/en/tc/act/police.c index 512d43148922..c4378afdec09 100644 --- a/drivers/net/ethernet/mellanox/mlx5/core/en/tc/act/police.c +++ b/drivers/net/ethernet/mellanox/mlx5/core/en/tc/act/police.c @@ -34,12 +34,6 @@ static int police_act_validate(const struct flow_action_entry *act, return -EOPNOTSUPP; } - if (act->police.rate_pkt_ps) { - NL_SET_ERR_MSG_MOD(extack, - "QoS offload not support packets per second"); - return -EOPNOTSUPP; - } - return 0; } -- cgit From 3099d2e62f9061ed57798e68e285d35837be686f Mon Sep 17 00:00:00 2001 From: Aya Levin Date: Sun, 8 Jan 2023 12:53:02 +0200 Subject: net/mlx5e: Fix memory leak on updating vport counters When updating statistics driver queries the vport's counters. On fail, add error path releasing the allocated buffer avoiding memory leak. Fixes: 64b68e369649 ("net/mlx5: Refactor and expand rep vport stat group") Signed-off-by: Aya Levin Reviewed-by: Maor Dickman Signed-off-by: Saeed Mahameed --- drivers/net/ethernet/mellanox/mlx5/core/en_rep.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_rep.c b/drivers/net/ethernet/mellanox/mlx5/core/en_rep.c index 75b9e1528fd2..7d90e5b72854 100644 --- a/drivers/net/ethernet/mellanox/mlx5/core/en_rep.c +++ b/drivers/net/ethernet/mellanox/mlx5/core/en_rep.c @@ -191,7 +191,7 @@ static MLX5E_DECLARE_STATS_GRP_OP_UPDATE_STATS(vport_rep) if (err) { netdev_warn(priv->netdev, "vport %d error %d reading stats\n", rep->vport, err); - return; + goto out; } #define MLX5_GET_CTR(p, x) \ @@ -241,6 +241,7 @@ static MLX5E_DECLARE_STATS_GRP_OP_UPDATE_STATS(vport_rep) rep_stats->tx_vport_rdma_multicast_bytes = MLX5_GET_CTR(out, received_ib_multicast.octets); +out: kvfree(out); } -- cgit From fe91d57277eef8bb4aca05acfa337b4a51d0bba4 Mon Sep 17 00:00:00 2001 From: Rahul Rameshbabu Date: Mon, 5 Dec 2022 14:26:09 -0800 Subject: net/mlx5: Fix ptp max frequency adjustment range .max_adj of ptp_clock_info acts as an absolute value for the amount in ppb that can be set for a single call of .adjfine. This means that a single call to .getfine cannot be greater than .max_adj or less than -(.max_adj). Provides correct value for max frequency adjustment value supported by devices. Fixes: 3d8c38af1493 ("net/mlx5e: Add PTP Hardware Clock (PHC) support") Signed-off-by: Rahul Rameshbabu Reviewed-by: Gal Pressman Reviewed-by: Tariq Toukan Signed-off-by: Saeed Mahameed --- drivers/net/ethernet/mellanox/mlx5/core/lib/clock.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/drivers/net/ethernet/mellanox/mlx5/core/lib/clock.c b/drivers/net/ethernet/mellanox/mlx5/core/lib/clock.c index 69cfe60c558a..69318b143268 100644 --- a/drivers/net/ethernet/mellanox/mlx5/core/lib/clock.c +++ b/drivers/net/ethernet/mellanox/mlx5/core/lib/clock.c @@ -681,7 +681,7 @@ static int mlx5_ptp_verify(struct ptp_clock_info *ptp, unsigned int pin, static const struct ptp_clock_info mlx5_ptp_clock_info = { .owner = THIS_MODULE, .name = "mlx5_ptp", - .max_adj = 100000000, + .max_adj = 50000000, .n_alarm = 0, .n_ext_ts = 0, .n_per_out = 0, -- cgit From d515d63cae2cd186acf40deaa8ef33067bb7f637 Mon Sep 17 00:00:00 2001 From: Gavin Li Date: Tue, 27 Dec 2022 04:54:09 +0200 Subject: net/mlx5e: Don't support encap rules with gbp option Previously, encap rules with gbp option would be offloaded by mistake but driver does not support gbp option offload. To fix this issue, check if the encap rule has gbp option and don't offload the rule Fixes: d8f9dfae49ce ("net: sched: allow flower to match vxlan options") Signed-off-by: Gavin Li Reviewed-by: Maor Dickman Signed-off-by: Saeed Mahameed --- drivers/net/ethernet/mellanox/mlx5/core/en/tc_tun_vxlan.c | 2 ++ 1 file changed, 2 insertions(+) diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en/tc_tun_vxlan.c b/drivers/net/ethernet/mellanox/mlx5/core/en/tc_tun_vxlan.c index fd07c4cbfd1d..1f62c702b625 100644 --- a/drivers/net/ethernet/mellanox/mlx5/core/en/tc_tun_vxlan.c +++ b/drivers/net/ethernet/mellanox/mlx5/core/en/tc_tun_vxlan.c @@ -88,6 +88,8 @@ static int mlx5e_gen_ip_tunnel_header_vxlan(char buf[], struct udphdr *udp = (struct udphdr *)(buf); struct vxlanhdr *vxh; + if (tun_key->tun_flags & TUNNEL_VXLAN_OPT) + return -EOPNOTSUPP; vxh = (struct vxlanhdr *)((char *)udp + sizeof(struct udphdr)); *ip_proto = IPPROTO_UDP; -- cgit From cd4f186dc110a0f0b3484048b8ecd42ee006ca6d Mon Sep 17 00:00:00 2001 From: Shay Drory Date: Thu, 15 Dec 2022 15:33:38 +0200 Subject: net/mlx5: E-switch, Coverity: overlapping copy When a capability is set via port function caps callbacks, a memcpy() is performed in which the source and the target are the same address, e.g.: the copy is redundant. Hence, Remove it. Discovered by Coverity. Fixes: 7db98396ef45 ("net/mlx5: E-Switch, Implement devlink port function cmds to control RoCE") Fixes: e5b9642a33be ("net/mlx5: E-Switch, Implement devlink port function cmds to control migratable") Signed-off-by: Shay Drory Reviewed-by: Mark Bloch Signed-off-by: Saeed Mahameed --- drivers/net/ethernet/mellanox/mlx5/core/eswitch_offloads.c | 4 ---- 1 file changed, 4 deletions(-) diff --git a/drivers/net/ethernet/mellanox/mlx5/core/eswitch_offloads.c b/drivers/net/ethernet/mellanox/mlx5/core/eswitch_offloads.c index 75b77dd2392b..c981fa77f439 100644 --- a/drivers/net/ethernet/mellanox/mlx5/core/eswitch_offloads.c +++ b/drivers/net/ethernet/mellanox/mlx5/core/eswitch_offloads.c @@ -4143,8 +4143,6 @@ int mlx5_devlink_port_fn_migratable_set(struct devlink_port *port, bool enable, } hca_caps = MLX5_ADDR_OF(query_hca_cap_out, query_ctx, capability); - memcpy(hca_caps, MLX5_ADDR_OF(query_hca_cap_out, query_ctx, capability), - MLX5_UN_SZ_BYTES(hca_cap_union)); MLX5_SET(cmd_hca_cap_2, hca_caps, migratable, 1); err = mlx5_vport_set_other_func_cap(esw->dev, hca_caps, vport->vport, @@ -4236,8 +4234,6 @@ int mlx5_devlink_port_fn_roce_set(struct devlink_port *port, bool enable, } hca_caps = MLX5_ADDR_OF(query_hca_cap_out, query_ctx, capability); - memcpy(hca_caps, MLX5_ADDR_OF(query_hca_cap_out, query_ctx, capability), - MLX5_UN_SZ_BYTES(hca_cap_union)); MLX5_SET(cmd_hca_cap, hca_caps, roce, enable); err = mlx5_vport_set_other_func_cap(esw->dev, hca_caps, vport_num, -- cgit From f5e1ed04aa2ea665a796f0109091ca3f2b01024a Mon Sep 17 00:00:00 2001 From: Emeel Hakim Date: Wed, 14 Dec 2022 16:34:13 +0200 Subject: net/mlx5e: Fix macsec ssci attribute handling in offload path Currently when macsec offload is set with extended packet number (epn) enabled, the driver wrongly deduce the short secure channel identifier (ssci) from the salt instead of the stand alone ssci attribute as it should, consequently creating a mismatch between the kernel and driver's ssci values. Fix by using the ssci value from the relevant attribute. Fixes: 4411a6c0abd3 ("net/mlx5e: Support MACsec offload extended packet number (EPN)") Signed-off-by: Emeel Hakim Reviewed-by: Raed Salem Signed-off-by: Saeed Mahameed --- drivers/net/ethernet/mellanox/mlx5/core/en_accel/macsec.c | 10 +++++++--- 1 file changed, 7 insertions(+), 3 deletions(-) diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_accel/macsec.c b/drivers/net/ethernet/mellanox/mlx5/core/en_accel/macsec.c index 9369a580743e..cf7b3bb54c86 100644 --- a/drivers/net/ethernet/mellanox/mlx5/core/en_accel/macsec.c +++ b/drivers/net/ethernet/mellanox/mlx5/core/en_accel/macsec.c @@ -62,6 +62,7 @@ struct mlx5e_macsec_sa { u32 enc_key_id; u32 next_pn; sci_t sci; + ssci_t ssci; salt_t salt; struct rhash_head hash; @@ -499,10 +500,11 @@ mlx5e_macsec_get_macsec_device_context(const struct mlx5e_macsec *macsec, } static void update_macsec_epn(struct mlx5e_macsec_sa *sa, const struct macsec_key *key, - const pn_t *next_pn_halves) + const pn_t *next_pn_halves, ssci_t ssci) { struct mlx5e_macsec_epn_state *epn_state = &sa->epn_state; + sa->ssci = ssci; sa->salt = key->salt; epn_state->epn_enabled = 1; epn_state->epn_msb = next_pn_halves->upper; @@ -550,7 +552,8 @@ static int mlx5e_macsec_add_txsa(struct macsec_context *ctx) tx_sa->assoc_num = assoc_num; if (secy->xpn) - update_macsec_epn(tx_sa, &ctx_tx_sa->key, &ctx_tx_sa->next_pn_halves); + update_macsec_epn(tx_sa, &ctx_tx_sa->key, &ctx_tx_sa->next_pn_halves, + ctx_tx_sa->ssci); err = mlx5_create_encryption_key(mdev, ctx->sa.key, secy->key_len, MLX5_ACCEL_OBJ_MACSEC_KEY, @@ -945,7 +948,8 @@ static int mlx5e_macsec_add_rxsa(struct macsec_context *ctx) rx_sa->fs_id = rx_sc->sc_xarray_element->fs_id; if (ctx->secy->xpn) - update_macsec_epn(rx_sa, &ctx_rx_sa->key, &ctx_rx_sa->next_pn_halves); + update_macsec_epn(rx_sa, &ctx_rx_sa->key, &ctx_rx_sa->next_pn_halves, + ctx_rx_sa->ssci); err = mlx5_create_encryption_key(mdev, ctx->sa.key, ctx->secy->key_len, MLX5_ACCEL_OBJ_MACSEC_KEY, -- cgit From 9828994ac492e8e7de47fe66097b7e665328f348 Mon Sep 17 00:00:00 2001 From: Emeel Hakim Date: Sun, 11 Dec 2022 13:22:23 +0200 Subject: net/mlx5e: Fix macsec possible null dereference when updating MAC security entity (SecY) Upon updating MAC security entity (SecY) in hw offload path, the macsec security association (SA) initialization routine is called. In case of extended packet number (epn) is enabled the salt and ssci attributes are retrieved using the MACsec driver rx_sa context which is unavailable when updating a SecY property such as encoding-sa hence the null dereference. Fix by using the provided SA to set those attributes. Fixes: 4411a6c0abd3 ("net/mlx5e: Support MACsec offload extended packet number (EPN)") Signed-off-by: Emeel Hakim Reviewed-by: Raed Salem Signed-off-by: Saeed Mahameed --- drivers/net/ethernet/mellanox/mlx5/core/en_accel/macsec.c | 9 ++------- 1 file changed, 2 insertions(+), 7 deletions(-) diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en_accel/macsec.c b/drivers/net/ethernet/mellanox/mlx5/core/en_accel/macsec.c index cf7b3bb54c86..7f6b940830b3 100644 --- a/drivers/net/ethernet/mellanox/mlx5/core/en_accel/macsec.c +++ b/drivers/net/ethernet/mellanox/mlx5/core/en_accel/macsec.c @@ -359,7 +359,6 @@ static int mlx5e_macsec_init_sa(struct macsec_context *ctx, struct mlx5_core_dev *mdev = priv->mdev; struct mlx5_macsec_obj_attrs obj_attrs; union mlx5e_macsec_rule *macsec_rule; - struct macsec_key *key; int err; obj_attrs.next_pn = sa->next_pn; @@ -369,13 +368,9 @@ static int mlx5e_macsec_init_sa(struct macsec_context *ctx, obj_attrs.aso_pdn = macsec->aso.pdn; obj_attrs.epn_state = sa->epn_state; - key = (is_tx) ? &ctx->sa.tx_sa->key : &ctx->sa.rx_sa->key; - if (sa->epn_state.epn_enabled) { - obj_attrs.ssci = (is_tx) ? cpu_to_be32((__force u32)ctx->sa.tx_sa->ssci) : - cpu_to_be32((__force u32)ctx->sa.rx_sa->ssci); - - memcpy(&obj_attrs.salt, &key->salt, sizeof(key->salt)); + obj_attrs.ssci = cpu_to_be32((__force u32)sa->ssci); + memcpy(&obj_attrs.salt, &sa->salt, sizeof(sa->salt)); } obj_attrs.replay_window = ctx->secy->replay_window; -- cgit