diff --git a/sys/dev/mlx5/driver.h b/sys/dev/mlx5/driver.h index d21a4aa075c0..614a308e8e2f 100644 --- a/sys/dev/mlx5/driver.h +++ b/sys/dev/mlx5/driver.h @@ -1,1227 +1,1229 @@ /*- * Copyright (c) 2013-2019, Mellanox Technologies, Ltd. All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY AUTHOR AND CONTRIBUTORS `AS IS' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * $FreeBSD$ */ #ifndef MLX5_DRIVER_H #define MLX5_DRIVER_H #include "opt_ratelimit.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #define MLX5_QCOUNTER_SETS_NETDEV 64 #define MLX5_MAX_NUMBER_OF_VFS 128 enum { MLX5_BOARD_ID_LEN = 64, MLX5_MAX_NAME_LEN = 16, }; enum { MLX5_CMD_TIMEOUT_MSEC = 60 * 1000, }; enum { CMD_OWNER_SW = 0x0, CMD_OWNER_HW = 0x1, CMD_STATUS_SUCCESS = 0, }; enum mlx5_sqp_t { MLX5_SQP_SMI = 0, MLX5_SQP_GSI = 1, MLX5_SQP_IEEE_1588 = 2, MLX5_SQP_SNIFFER = 3, MLX5_SQP_SYNC_UMR = 4, }; enum { MLX5_MAX_PORTS = 2, }; enum { MLX5_EQ_VEC_PAGES = 0, MLX5_EQ_VEC_CMD = 1, MLX5_EQ_VEC_ASYNC = 2, MLX5_EQ_VEC_COMP_BASE, }; enum { MLX5_ATOMIC_MODE_OFF = 16, MLX5_ATOMIC_MODE_NONE = 0 << MLX5_ATOMIC_MODE_OFF, MLX5_ATOMIC_MODE_IB_COMP = 1 << MLX5_ATOMIC_MODE_OFF, MLX5_ATOMIC_MODE_CX = 2 << MLX5_ATOMIC_MODE_OFF, MLX5_ATOMIC_MODE_8B = 3 << MLX5_ATOMIC_MODE_OFF, MLX5_ATOMIC_MODE_16B = 4 << MLX5_ATOMIC_MODE_OFF, MLX5_ATOMIC_MODE_32B = 5 << MLX5_ATOMIC_MODE_OFF, MLX5_ATOMIC_MODE_64B = 6 << MLX5_ATOMIC_MODE_OFF, MLX5_ATOMIC_MODE_128B = 7 << MLX5_ATOMIC_MODE_OFF, MLX5_ATOMIC_MODE_256B = 8 << MLX5_ATOMIC_MODE_OFF, }; enum { MLX5_ATOMIC_MODE_DCT_OFF = 20, MLX5_ATOMIC_MODE_DCT_NONE = 0 << MLX5_ATOMIC_MODE_DCT_OFF, MLX5_ATOMIC_MODE_DCT_IB_COMP = 1 << MLX5_ATOMIC_MODE_DCT_OFF, MLX5_ATOMIC_MODE_DCT_CX = 2 << MLX5_ATOMIC_MODE_DCT_OFF, MLX5_ATOMIC_MODE_DCT_8B = 3 << MLX5_ATOMIC_MODE_DCT_OFF, MLX5_ATOMIC_MODE_DCT_16B = 4 << MLX5_ATOMIC_MODE_DCT_OFF, MLX5_ATOMIC_MODE_DCT_32B = 5 << MLX5_ATOMIC_MODE_DCT_OFF, MLX5_ATOMIC_MODE_DCT_64B = 6 << MLX5_ATOMIC_MODE_DCT_OFF, MLX5_ATOMIC_MODE_DCT_128B = 7 << MLX5_ATOMIC_MODE_DCT_OFF, MLX5_ATOMIC_MODE_DCT_256B = 8 << MLX5_ATOMIC_MODE_DCT_OFF, }; enum { MLX5_ATOMIC_OPS_CMP_SWAP = 1 << 0, MLX5_ATOMIC_OPS_FETCH_ADD = 1 << 1, MLX5_ATOMIC_OPS_MASKED_CMP_SWAP = 1 << 2, MLX5_ATOMIC_OPS_MASKED_FETCH_ADD = 1 << 3, }; enum { MLX5_REG_QPTS = 0x4002, MLX5_REG_QETCR = 0x4005, MLX5_REG_QPDP = 0x4007, MLX5_REG_QTCT = 0x400A, MLX5_REG_QPDPM = 0x4013, MLX5_REG_QHLL = 0x4016, MLX5_REG_QCAM = 0x4019, MLX5_REG_DCBX_PARAM = 0x4020, MLX5_REG_DCBX_APP = 0x4021, MLX5_REG_FPGA_CAP = 0x4022, MLX5_REG_FPGA_CTRL = 0x4023, MLX5_REG_FPGA_ACCESS_REG = 0x4024, MLX5_REG_FPGA_SHELL_CNTR = 0x4025, MLX5_REG_PCAP = 0x5001, MLX5_REG_PMLP = 0x5002, MLX5_REG_PMTU = 0x5003, MLX5_REG_PTYS = 0x5004, MLX5_REG_PAOS = 0x5006, MLX5_REG_PFCC = 0x5007, MLX5_REG_PPCNT = 0x5008, MLX5_REG_PUDE = 0x5009, MLX5_REG_PPTB = 0x500B, MLX5_REG_PBMC = 0x500C, MLX5_REG_PELC = 0x500E, MLX5_REG_PVLC = 0x500F, MLX5_REG_PMPE = 0x5010, MLX5_REG_PMAOS = 0x5012, MLX5_REG_PPLM = 0x5023, MLX5_REG_PDDR = 0x5031, MLX5_REG_PBSR = 0x5038, MLX5_REG_PCAM = 0x507f, MLX5_REG_NODE_DESC = 0x6001, MLX5_REG_HOST_ENDIANNESS = 0x7004, MLX5_REG_MTMP = 0x900a, MLX5_REG_MCIA = 0x9014, MLX5_REG_MFRL = 0x9028, MLX5_REG_MPCNT = 0x9051, MLX5_REG_MCQI = 0x9061, MLX5_REG_MCC = 0x9062, MLX5_REG_MCDA = 0x9063, MLX5_REG_MCAM = 0x907f, }; enum dbg_rsc_type { MLX5_DBG_RSC_QP, MLX5_DBG_RSC_EQ, MLX5_DBG_RSC_CQ, }; enum { MLX5_INTERFACE_PROTOCOL_IB = 0, MLX5_INTERFACE_PROTOCOL_ETH = 1, MLX5_INTERFACE_NUMBER = 2, }; struct mlx5_field_desc { struct dentry *dent; int i; }; struct mlx5_rsc_debug { struct mlx5_core_dev *dev; void *object; enum dbg_rsc_type type; struct dentry *root; struct mlx5_field_desc fields[0]; }; enum mlx5_dev_event { MLX5_DEV_EVENT_SYS_ERROR, MLX5_DEV_EVENT_PORT_UP, MLX5_DEV_EVENT_PORT_DOWN, MLX5_DEV_EVENT_PORT_INITIALIZED, MLX5_DEV_EVENT_LID_CHANGE, MLX5_DEV_EVENT_PKEY_CHANGE, MLX5_DEV_EVENT_GUID_CHANGE, MLX5_DEV_EVENT_CLIENT_REREG, MLX5_DEV_EVENT_VPORT_CHANGE, MLX5_DEV_EVENT_ERROR_STATE_DCBX, MLX5_DEV_EVENT_REMOTE_CONFIG_CHANGE, MLX5_DEV_EVENT_LOCAL_OPER_CHANGE, MLX5_DEV_EVENT_REMOTE_CONFIG_APPLICATION_PRIORITY_CHANGE, }; enum mlx5_port_status { MLX5_PORT_UP = 1 << 0, MLX5_PORT_DOWN = 1 << 1, }; enum { MLX5_VSC_SPACE_SUPPORTED = 0x1, MLX5_VSC_SPACE_OFFSET = 0x4, MLX5_VSC_COUNTER_OFFSET = 0x8, MLX5_VSC_SEMA_OFFSET = 0xC, MLX5_VSC_ADDR_OFFSET = 0x10, MLX5_VSC_DATA_OFFSET = 0x14, MLX5_VSC_MAX_RETRIES = 0x1000, }; #define MLX5_PROT_MASK(link_mode) (1 << link_mode) struct mlx5_cmd_first { __be32 data[4]; }; struct cache_ent; struct mlx5_fw_page { union { struct rb_node rb_node; struct list_head list; }; struct mlx5_cmd_first first; struct mlx5_core_dev *dev; bus_dmamap_t dma_map; bus_addr_t dma_addr; void *virt_addr; struct cache_ent *cache; u32 numpages; u16 load_done; #define MLX5_LOAD_ST_NONE 0 #define MLX5_LOAD_ST_SUCCESS 1 #define MLX5_LOAD_ST_FAILURE 2 u16 func_id; }; #define mlx5_cmd_msg mlx5_fw_page struct mlx5_cmd_debug { struct dentry *dbg_root; struct dentry *dbg_in; struct dentry *dbg_out; struct dentry *dbg_outlen; struct dentry *dbg_status; struct dentry *dbg_run; void *in_msg; void *out_msg; u8 status; u16 inlen; u16 outlen; }; struct cache_ent { /* protect block chain allocations */ spinlock_t lock; struct list_head head; }; struct cmd_msg_cache { struct cache_ent large; struct cache_ent med; }; struct mlx5_traffic_counter { u64 packets; u64 octets; }; enum mlx5_cmd_mode { MLX5_CMD_MODE_POLLING, MLX5_CMD_MODE_EVENTS }; struct mlx5_cmd_stats { u64 sum; u64 n; struct dentry *root; struct dentry *avg; struct dentry *count; /* protect command average calculations */ spinlock_t lock; }; struct mlx5_cmd { struct mlx5_fw_page *cmd_page; bus_dma_tag_t dma_tag; struct sx dma_sx; struct mtx dma_mtx; #define MLX5_DMA_OWNED(dev) mtx_owned(&(dev)->cmd.dma_mtx) #define MLX5_DMA_LOCK(dev) mtx_lock(&(dev)->cmd.dma_mtx) #define MLX5_DMA_UNLOCK(dev) mtx_unlock(&(dev)->cmd.dma_mtx) struct cv dma_cv; #define MLX5_DMA_DONE(dev) cv_broadcast(&(dev)->cmd.dma_cv) #define MLX5_DMA_WAIT(dev) cv_wait(&(dev)->cmd.dma_cv, &(dev)->cmd.dma_mtx) void *cmd_buf; dma_addr_t dma; u16 cmdif_rev; u8 log_sz; u8 log_stride; int max_reg_cmds; int events; u32 __iomem *vector; /* protect command queue allocations */ spinlock_t alloc_lock; /* protect token allocations */ spinlock_t token_lock; u8 token; unsigned long bitmask; struct semaphore sem; struct semaphore pages_sem; enum mlx5_cmd_mode mode; struct mlx5_cmd_work_ent * volatile ent_arr[MLX5_MAX_COMMANDS]; volatile enum mlx5_cmd_mode ent_mode[MLX5_MAX_COMMANDS]; struct mlx5_cmd_debug dbg; struct cmd_msg_cache cache; int checksum_disabled; struct mlx5_cmd_stats stats[MLX5_CMD_OP_MAX]; }; struct mlx5_port_caps { int gid_table_len; int pkey_table_len; u8 ext_port_cap; }; struct mlx5_buf { bus_dma_tag_t dma_tag; bus_dmamap_t dma_map; struct mlx5_core_dev *dev; struct { void *buf; } direct; u64 *page_list; int npages; int size; u8 page_shift; u8 load_done; }; struct mlx5_frag_buf { struct mlx5_buf_list *frags; int npages; int size; u8 page_shift; }; struct mlx5_eq { struct mlx5_core_dev *dev; __be32 __iomem *doorbell; u32 cons_index; struct mlx5_buf buf; int size; u8 irqn; u8 eqn; int nent; u64 mask; struct list_head list; int index; struct mlx5_rsc_debug *dbg; }; struct mlx5_core_psv { u32 psv_idx; struct psv_layout { u32 pd; u16 syndrome; u16 reserved; u16 bg; u16 app_tag; u32 ref_tag; } psv; }; struct mlx5_core_sig_ctx { struct mlx5_core_psv psv_memory; struct mlx5_core_psv psv_wire; #if (__FreeBSD_version >= 1100000) struct ib_sig_err err_item; #endif bool sig_status_checked; bool sig_err_exists; u32 sigerr_count; }; enum { MLX5_MKEY_MR = 1, MLX5_MKEY_MW, MLX5_MKEY_MR_USER, }; struct mlx5_core_mkey { u64 iova; u64 size; u32 key; u32 pd; u32 type; }; struct mlx5_core_mr { u64 iova; u64 size; u32 key; u32 pd; }; enum mlx5_res_type { MLX5_RES_QP = MLX5_EVENT_QUEUE_TYPE_QP, MLX5_RES_RQ = MLX5_EVENT_QUEUE_TYPE_RQ, MLX5_RES_SQ = MLX5_EVENT_QUEUE_TYPE_SQ, MLX5_RES_SRQ = 3, MLX5_RES_XSRQ = 4, MLX5_RES_DCT = 5, }; struct mlx5_core_rsc_common { enum mlx5_res_type res; atomic_t refcount; struct completion free; }; struct mlx5_uars_page { void __iomem *map; bool wc; u32 index; struct list_head list; unsigned int bfregs; unsigned long *reg_bitmap; /* for non fast path bf regs */ unsigned long *fp_bitmap; unsigned int reg_avail; unsigned int fp_avail; struct kref ref_count; struct mlx5_core_dev *mdev; }; struct mlx5_bfreg_head { /* protect blue flame registers allocations */ struct mutex lock; struct list_head list; }; struct mlx5_bfreg_data { struct mlx5_bfreg_head reg_head; struct mlx5_bfreg_head wc_head; }; struct mlx5_sq_bfreg { void __iomem *map; struct mlx5_uars_page *up; bool wc; u32 index; unsigned int offset; }; struct mlx5_core_srq { struct mlx5_core_rsc_common common; /* must be first */ u32 srqn; int max; size_t max_gs; size_t max_avail_gather; int wqe_shift; void (*event)(struct mlx5_core_srq *, int); atomic_t refcount; struct completion free; }; struct mlx5_eq_table { void __iomem *update_ci; void __iomem *update_arm_ci; struct list_head comp_eqs_list; struct mlx5_eq pages_eq; struct mlx5_eq async_eq; struct mlx5_eq cmd_eq; int num_comp_vectors; /* protect EQs list */ spinlock_t lock; }; struct mlx5_core_health { struct mlx5_health_buffer __iomem *health; __be32 __iomem *health_counter; struct timer_list timer; u32 prev; int miss_counter; u32 fatal_error; struct workqueue_struct *wq_watchdog; struct work_struct work_watchdog; /* wq spinlock to synchronize draining */ spinlock_t wq_lock; struct workqueue_struct *wq; unsigned long flags; struct work_struct work; struct delayed_work recover_work; unsigned int last_reset_req; struct work_struct work_cmd_completion; struct workqueue_struct *wq_cmd; }; #define MLX5_CQ_LINEAR_ARRAY_SIZE 1024 struct mlx5_cq_linear_array_entry { struct mlx5_core_cq * volatile cq; }; struct mlx5_cq_table { /* protect radix tree */ spinlock_t writerlock; atomic_t writercount; struct radix_tree_root tree; struct mlx5_cq_linear_array_entry linear_array[MLX5_CQ_LINEAR_ARRAY_SIZE]; }; struct mlx5_qp_table { /* protect radix tree */ spinlock_t lock; struct radix_tree_root tree; }; struct mlx5_srq_table { /* protect radix tree */ spinlock_t lock; struct radix_tree_root tree; }; struct mlx5_mr_table { /* protect radix tree */ spinlock_t lock; struct radix_tree_root tree; }; #ifdef RATELIMIT struct mlx5_rl_entry { u32 rate; u16 burst; u16 index; u32 refcount; }; struct mlx5_rl_table { struct mutex rl_lock; u16 max_size; u32 max_rate; u32 min_rate; struct mlx5_rl_entry *rl_entry; }; #endif struct mlx5_pme_stats { u64 status_counters[MLX5_MODULE_STATUS_NUM]; u64 error_counters[MLX5_MODULE_EVENT_ERROR_NUM]; }; struct mlx5_priv { char name[MLX5_MAX_NAME_LEN]; struct mlx5_eq_table eq_table; struct msix_entry *msix_arr; MLX5_DECLARE_DOORBELL_LOCK(cq_uar_lock); int disable_irqs; /* pages stuff */ struct workqueue_struct *pg_wq; struct rb_root page_root; s64 fw_pages; atomic_t reg_pages; s64 pages_per_func[MLX5_MAX_NUMBER_OF_VFS]; struct mlx5_core_health health; struct mlx5_srq_table srq_table; /* start: qp staff */ struct mlx5_qp_table qp_table; struct dentry *qp_debugfs; struct dentry *eq_debugfs; struct dentry *cq_debugfs; struct dentry *cmdif_debugfs; /* end: qp staff */ /* start: cq staff */ struct mlx5_cq_table cq_table; /* end: cq staff */ /* start: mr staff */ struct mlx5_mr_table mr_table; /* end: mr staff */ /* start: alloc staff */ int numa_node; struct mutex pgdir_mutex; struct list_head pgdir_list; /* end: alloc staff */ struct dentry *dbg_root; /* protect mkey key part */ spinlock_t mkey_lock; u8 mkey_key; struct list_head dev_list; struct list_head ctx_list; spinlock_t ctx_lock; unsigned long pci_dev_data; #ifdef RATELIMIT struct mlx5_rl_table rl_table; #endif struct mlx5_pme_stats pme_stats; struct mlx5_eswitch *eswitch; struct mlx5_bfreg_data bfregs; struct mlx5_uars_page *uar; }; enum mlx5_device_state { MLX5_DEVICE_STATE_UP, MLX5_DEVICE_STATE_INTERNAL_ERROR, }; enum mlx5_interface_state { MLX5_INTERFACE_STATE_UP = 0x1, MLX5_INTERFACE_STATE_TEARDOWN = 0x2, }; enum mlx5_pci_status { MLX5_PCI_STATUS_DISABLED, MLX5_PCI_STATUS_ENABLED, }; #define MLX5_MAX_RESERVED_GIDS 8 struct mlx5_rsvd_gids { unsigned int start; unsigned int count; struct ida ida; }; struct mlx5_special_contexts { int resd_lkey; }; struct mlx5_flow_root_namespace; struct mlx5_core_dev { struct pci_dev *pdev; /* sync pci state */ struct mutex pci_status_mutex; enum mlx5_pci_status pci_status; char board_id[MLX5_BOARD_ID_LEN]; struct mlx5_cmd cmd; struct mlx5_port_caps port_caps[MLX5_MAX_PORTS]; u32 hca_caps_cur[MLX5_CAP_NUM][MLX5_UN_SZ_DW(hca_cap_union)]; u32 hca_caps_max[MLX5_CAP_NUM][MLX5_UN_SZ_DW(hca_cap_union)]; struct { u32 pcam[MLX5_ST_SZ_DW(pcam_reg)]; u32 mcam[MLX5_ST_SZ_DW(mcam_reg)]; u32 qcam[MLX5_ST_SZ_DW(qcam_reg)]; u32 fpga[MLX5_ST_SZ_DW(fpga_cap)]; } caps; phys_addr_t iseg_base; struct mlx5_init_seg __iomem *iseg; enum mlx5_device_state state; /* sync interface state */ struct mutex intf_state_mutex; unsigned long intf_state; void (*event) (struct mlx5_core_dev *dev, enum mlx5_dev_event event, unsigned long param); struct mlx5_priv priv; struct mlx5_profile *profile; atomic_t num_qps; u32 vsc_addr; u32 issi; struct mlx5_special_contexts special_contexts; unsigned int module_status[MLX5_MAX_PORTS]; struct mlx5_flow_root_namespace *root_ns; struct mlx5_flow_root_namespace *fdb_root_ns; struct mlx5_flow_root_namespace *esw_egress_root_ns; struct mlx5_flow_root_namespace *esw_ingress_root_ns; struct mlx5_flow_root_namespace *sniffer_rx_root_ns; struct mlx5_flow_root_namespace *sniffer_tx_root_ns; u32 num_q_counter_allocated[MLX5_INTERFACE_NUMBER]; struct mlx5_crspace_regmap *dump_rege; uint32_t *dump_data; unsigned dump_size; bool dump_valid; bool dump_copyout; struct mtx dump_lock; struct sysctl_ctx_list sysctl_ctx; int msix_eqvec; int pwr_status; int pwr_value; struct { struct mlx5_rsvd_gids reserved_gids; atomic_t roce_en; } roce; struct { spinlock_t spinlock; #define MLX5_MPFS_TABLE_MAX 32 long bitmap[BITS_TO_LONGS(MLX5_MPFS_TABLE_MAX)]; } mpfs; #ifdef CONFIG_MLX5_FPGA struct mlx5_fpga_device *fpga; #endif }; enum { MLX5_WOL_DISABLE = 0, MLX5_WOL_SECURED_MAGIC = 1 << 1, MLX5_WOL_MAGIC = 1 << 2, MLX5_WOL_ARP = 1 << 3, MLX5_WOL_BROADCAST = 1 << 4, MLX5_WOL_MULTICAST = 1 << 5, MLX5_WOL_UNICAST = 1 << 6, MLX5_WOL_PHY_ACTIVITY = 1 << 7, }; struct mlx5_db { __be32 *db; union { struct mlx5_db_pgdir *pgdir; struct mlx5_ib_user_db_page *user_page; } u; dma_addr_t dma; int index; }; struct mlx5_net_counters { u64 packets; u64 octets; }; struct mlx5_ptys_reg { u8 an_dis_admin; u8 an_dis_ap; u8 local_port; u8 proto_mask; u32 eth_proto_cap; u16 ib_link_width_cap; u16 ib_proto_cap; u32 eth_proto_admin; u16 ib_link_width_admin; u16 ib_proto_admin; u32 eth_proto_oper; u16 ib_link_width_oper; u16 ib_proto_oper; u32 eth_proto_lp_advertise; }; struct mlx5_pvlc_reg { u8 local_port; u8 vl_hw_cap; u8 vl_admin; u8 vl_operational; }; struct mlx5_pmtu_reg { u8 local_port; u16 max_mtu; u16 admin_mtu; u16 oper_mtu; }; struct mlx5_vport_counters { struct mlx5_net_counters received_errors; struct mlx5_net_counters transmit_errors; struct mlx5_net_counters received_ib_unicast; struct mlx5_net_counters transmitted_ib_unicast; struct mlx5_net_counters received_ib_multicast; struct mlx5_net_counters transmitted_ib_multicast; struct mlx5_net_counters received_eth_broadcast; struct mlx5_net_counters transmitted_eth_broadcast; struct mlx5_net_counters received_eth_unicast; struct mlx5_net_counters transmitted_eth_unicast; struct mlx5_net_counters received_eth_multicast; struct mlx5_net_counters transmitted_eth_multicast; }; enum { MLX5_DB_PER_PAGE = MLX5_ADAPTER_PAGE_SIZE / L1_CACHE_BYTES, }; struct mlx5_core_dct { struct mlx5_core_rsc_common common; /* must be first */ void (*event)(struct mlx5_core_dct *, int); int dctn; struct completion drained; struct mlx5_rsc_debug *dbg; int pid; u16 uid; }; enum { MLX5_COMP_EQ_SIZE = 1024, }; enum { MLX5_PTYS_IB = 1 << 0, MLX5_PTYS_EN = 1 << 2, }; struct mlx5_db_pgdir { struct list_head list; DECLARE_BITMAP(bitmap, MLX5_DB_PER_PAGE); struct mlx5_fw_page *fw_page; __be32 *db_page; dma_addr_t db_dma; }; typedef void (*mlx5_cmd_cbk_t)(int status, void *context); struct mlx5_cmd_work_ent { struct mlx5_cmd_msg *in; struct mlx5_cmd_msg *out; int uin_size; void *uout; int uout_size; mlx5_cmd_cbk_t callback; struct delayed_work cb_timeout_work; void *context; int idx; struct completion done; struct mlx5_cmd *cmd; struct work_struct work; struct mlx5_cmd_layout *lay; int ret; int page_queue; u8 status; u8 token; u64 ts1; u64 ts2; u16 op; u8 busy; bool polling; }; struct mlx5_pas { u64 pa; u8 log_sz; }; enum port_state_policy { MLX5_POLICY_DOWN = 0, MLX5_POLICY_UP = 1, MLX5_POLICY_FOLLOW = 2, MLX5_POLICY_INVALID = 0xffffffff }; static inline void * mlx5_buf_offset(struct mlx5_buf *buf, int offset) { return ((char *)buf->direct.buf + offset); } extern struct workqueue_struct *mlx5_core_wq; #define STRUCT_FIELD(header, field) \ .struct_offset_bytes = offsetof(struct ib_unpacked_ ## header, field), \ .struct_size_bytes = sizeof((struct ib_unpacked_ ## header *)0)->field static inline struct mlx5_core_dev *pci2mlx5_core_dev(struct pci_dev *pdev) { return pci_get_drvdata(pdev); } extern struct dentry *mlx5_debugfs_root; static inline u16 fw_rev_maj(struct mlx5_core_dev *dev) { return ioread32be(&dev->iseg->fw_rev) & 0xffff; } static inline u16 fw_rev_min(struct mlx5_core_dev *dev) { return ioread32be(&dev->iseg->fw_rev) >> 16; } static inline u16 fw_rev_sub(struct mlx5_core_dev *dev) { return ioread32be(&dev->iseg->cmdif_rev_fw_sub) & 0xffff; } static inline u16 cmdif_rev_get(struct mlx5_core_dev *dev) { return ioread32be(&dev->iseg->cmdif_rev_fw_sub) >> 16; } static inline int mlx5_get_gid_table_len(u16 param) { if (param > 4) { printf("M4_CORE_DRV_NAME: WARN: ""gid table length is zero\n"); return 0; } return 8 * (1 << param); } static inline void *mlx5_vzalloc(unsigned long size) { void *rtn; rtn = kzalloc(size, GFP_KERNEL | __GFP_NOWARN); return rtn; } static inline void *mlx5_vmalloc(unsigned long size) { void *rtn; rtn = kmalloc(size, GFP_KERNEL | __GFP_NOWARN); if (!rtn) rtn = vmalloc(size); return rtn; } static inline u32 mlx5_base_mkey(const u32 key) { return key & 0xffffff00u; } int mlx5_cmd_init(struct mlx5_core_dev *dev); void mlx5_cmd_cleanup(struct mlx5_core_dev *dev); void mlx5_cmd_use_events(struct mlx5_core_dev *dev); void mlx5_cmd_use_polling(struct mlx5_core_dev *dev); void mlx5_cmd_mbox_status(void *out, u8 *status, u32 *syndrome); int mlx5_core_get_caps(struct mlx5_core_dev *dev, enum mlx5_cap_type cap_type); struct mlx5_async_ctx { struct mlx5_core_dev *dev; atomic_t num_inflight; struct wait_queue_head wait; }; struct mlx5_async_work; typedef void (*mlx5_async_cbk_t)(int status, struct mlx5_async_work *context); struct mlx5_async_work { struct mlx5_async_ctx *ctx; mlx5_async_cbk_t user_callback; }; void mlx5_cmd_init_async_ctx(struct mlx5_core_dev *dev, struct mlx5_async_ctx *ctx); void mlx5_cmd_cleanup_async_ctx(struct mlx5_async_ctx *ctx); int mlx5_cmd_exec_cb(struct mlx5_async_ctx *ctx, void *in, int in_size, void *out, int out_size, mlx5_async_cbk_t callback, struct mlx5_async_work *work); int mlx5_cmd_exec(struct mlx5_core_dev *dev, void *in, int in_size, void *out, int out_size); int mlx5_cmd_exec_polling(struct mlx5_core_dev *dev, void *in, int in_size, void *out, int out_size); int mlx5_cmd_alloc_uar(struct mlx5_core_dev *dev, u32 *uarn); int mlx5_cmd_free_uar(struct mlx5_core_dev *dev, u32 uarn); int mlx5_alloc_bfreg(struct mlx5_core_dev *mdev, struct mlx5_sq_bfreg *bfreg, bool map_wc, bool fast_path); void mlx5_free_bfreg(struct mlx5_core_dev *mdev, struct mlx5_sq_bfreg *bfreg); struct mlx5_uars_page *mlx5_get_uars_page(struct mlx5_core_dev *mdev); void mlx5_put_uars_page(struct mlx5_core_dev *mdev, struct mlx5_uars_page *up); 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, bool disable_health); void mlx5_drain_health_wq(struct mlx5_core_dev *dev); void mlx5_drain_health_recovery(struct mlx5_core_dev *dev); void mlx5_trigger_health_work(struct mlx5_core_dev *dev); void mlx5_trigger_health_watchdog(struct mlx5_core_dev *dev); -#define mlx5_buf_alloc_node(dev, size, direct, buf, node) \ - mlx5_buf_alloc(dev, size, direct, buf) int mlx5_buf_alloc(struct mlx5_core_dev *dev, int size, int max_direct, struct mlx5_buf *buf); void mlx5_buf_free(struct mlx5_core_dev *dev, struct mlx5_buf *buf); int mlx5_core_create_srq(struct mlx5_core_dev *dev, struct mlx5_core_srq *srq, struct mlx5_srq_attr *in); int mlx5_core_destroy_srq(struct mlx5_core_dev *dev, struct mlx5_core_srq *srq); int mlx5_core_query_srq(struct mlx5_core_dev *dev, struct mlx5_core_srq *srq, struct mlx5_srq_attr *out); int mlx5_core_query_vendor_id(struct mlx5_core_dev *mdev, u32 *vendor_id); int mlx5_core_arm_srq(struct mlx5_core_dev *dev, struct mlx5_core_srq *srq, u16 lwm, int is_srq); void mlx5_init_mr_table(struct mlx5_core_dev *dev); void mlx5_cleanup_mr_table(struct mlx5_core_dev *dev); int mlx5_core_create_mkey_cb(struct mlx5_core_dev *dev, struct mlx5_core_mr *mkey, struct mlx5_async_ctx *async_ctx, u32 *in, int inlen, u32 *out, int outlen, mlx5_async_cbk_t callback, struct mlx5_async_work *context); int mlx5_core_create_mkey(struct mlx5_core_dev *dev, struct mlx5_core_mr *mr, u32 *in, int inlen); int mlx5_core_destroy_mkey(struct mlx5_core_dev *dev, struct mlx5_core_mr *mkey); int mlx5_core_query_mkey(struct mlx5_core_dev *dev, struct mlx5_core_mr *mkey, u32 *out, int outlen); int mlx5_core_dump_fill_mkey(struct mlx5_core_dev *dev, struct mlx5_core_mr *mr, u32 *mkey); int mlx5_core_alloc_pd(struct mlx5_core_dev *dev, u32 *pdn); int mlx5_core_dealloc_pd(struct mlx5_core_dev *dev, u32 pdn); int mlx5_core_mad_ifc(struct mlx5_core_dev *dev, const void *inb, void *outb, u16 opmod, u8 port); void mlx5_fwp_flush(struct mlx5_fw_page *fwp); void mlx5_fwp_invalidate(struct mlx5_fw_page *fwp); struct mlx5_fw_page *mlx5_fwp_alloc(struct mlx5_core_dev *dev, gfp_t flags, unsigned num); void mlx5_fwp_free(struct mlx5_fw_page *fwp); u64 mlx5_fwp_get_dma(struct mlx5_fw_page *fwp, size_t offset); void *mlx5_fwp_get_virt(struct mlx5_fw_page *fwp, size_t offset); void mlx5_pagealloc_init(struct mlx5_core_dev *dev); void mlx5_pagealloc_cleanup(struct mlx5_core_dev *dev); int mlx5_pagealloc_start(struct mlx5_core_dev *dev); void mlx5_pagealloc_stop(struct mlx5_core_dev *dev); void mlx5_core_req_pages_handler(struct mlx5_core_dev *dev, u16 func_id, s32 npages); int mlx5_satisfy_startup_pages(struct mlx5_core_dev *dev, int boot); int mlx5_reclaim_startup_pages(struct mlx5_core_dev *dev); s64 mlx5_wait_for_reclaim_vfs_pages(struct mlx5_core_dev *dev); void mlx5_register_debugfs(void); void mlx5_unregister_debugfs(void); int mlx5_eq_init(struct mlx5_core_dev *dev); void mlx5_eq_cleanup(struct mlx5_core_dev *dev); void mlx5_fill_page_array(struct mlx5_buf *buf, __be64 *pas); void mlx5_cq_completion(struct mlx5_core_dev *dev, struct mlx5_eqe *eqe); void mlx5_rsc_event(struct mlx5_core_dev *dev, u32 rsn, int event_type); void mlx5_srq_event(struct mlx5_core_dev *dev, u32 srqn, int event_type); struct mlx5_core_srq *mlx5_core_get_srq(struct mlx5_core_dev *dev, u32 srqn); void mlx5_cmd_comp_handler(struct mlx5_core_dev *dev, u64 vector, enum mlx5_cmd_mode mode); void mlx5_cq_event(struct mlx5_core_dev *dev, u32 cqn, int event_type); int mlx5_create_map_eq(struct mlx5_core_dev *dev, struct mlx5_eq *eq, u8 vecidx, int nent, u64 mask); int mlx5_destroy_unmap_eq(struct mlx5_core_dev *dev, struct mlx5_eq *eq); int mlx5_start_eqs(struct mlx5_core_dev *dev); int mlx5_stop_eqs(struct mlx5_core_dev *dev); int mlx5_vector2eqn(struct mlx5_core_dev *dev, int vector, int *eqn, int *irqn); int mlx5_core_attach_mcg(struct mlx5_core_dev *dev, union ib_gid *mgid, u32 qpn); int mlx5_core_detach_mcg(struct mlx5_core_dev *dev, union ib_gid *mgid, u32 qpn); int mlx5_core_set_dc_cnak_trace(struct mlx5_core_dev *dev, int enable, u64 addr); int mlx5_qp_debugfs_init(struct mlx5_core_dev *dev); void mlx5_qp_debugfs_cleanup(struct mlx5_core_dev *dev); int mlx5_core_access_reg(struct mlx5_core_dev *dev, void *data_in, int size_in, void *data_out, int size_out, u16 reg_num, int arg, int write); void mlx5_toggle_port_link(struct mlx5_core_dev *dev); int mlx5_debug_eq_add(struct mlx5_core_dev *dev, struct mlx5_eq *eq); void mlx5_debug_eq_remove(struct mlx5_core_dev *dev, struct mlx5_eq *eq); int mlx5_core_eq_query(struct mlx5_core_dev *dev, struct mlx5_eq *eq, u32 *out, int outlen); int mlx5_eq_debugfs_init(struct mlx5_core_dev *dev); void mlx5_eq_debugfs_cleanup(struct mlx5_core_dev *dev); int mlx5_cq_debugfs_init(struct mlx5_core_dev *dev); void mlx5_cq_debugfs_cleanup(struct mlx5_core_dev *dev); int mlx5_db_alloc(struct mlx5_core_dev *dev, struct mlx5_db *db); -int mlx5_db_alloc_node(struct mlx5_core_dev *dev, struct mlx5_db *db, - int node); void mlx5_db_free(struct mlx5_core_dev *dev, struct mlx5_db *db); +static inline struct domainset * +mlx5_dev_domainset(struct mlx5_core_dev *mdev) +{ + return (linux_get_vm_domain_set(mdev->priv.numa_node)); +} + const char *mlx5_command_str(int command); int mlx5_cmdif_debugfs_init(struct mlx5_core_dev *dev); void mlx5_cmdif_debugfs_cleanup(struct mlx5_core_dev *dev); int mlx5_core_create_psv(struct mlx5_core_dev *dev, u32 pdn, int npsvs, u32 *sig_index); int mlx5_core_destroy_psv(struct mlx5_core_dev *dev, int psv_num); void mlx5_core_put_rsc(struct mlx5_core_rsc_common *common); u8 mlx5_is_wol_supported(struct mlx5_core_dev *dev); int mlx5_set_wol(struct mlx5_core_dev *dev, u8 wol_mode); int mlx5_set_dropless_mode(struct mlx5_core_dev *dev, u16 timeout); int mlx5_query_dropless_mode(struct mlx5_core_dev *dev, u16 *timeout); int mlx5_query_wol(struct mlx5_core_dev *dev, u8 *wol_mode); int mlx5_core_access_pvlc(struct mlx5_core_dev *dev, struct mlx5_pvlc_reg *pvlc, int write); int mlx5_core_access_ptys(struct mlx5_core_dev *dev, struct mlx5_ptys_reg *ptys, int write); int mlx5_core_access_pmtu(struct mlx5_core_dev *dev, struct mlx5_pmtu_reg *pmtu, int write); int mlx5_vxlan_udp_port_add(struct mlx5_core_dev *dev, u16 port); int mlx5_vxlan_udp_port_delete(struct mlx5_core_dev *dev, u16 port); int mlx5_query_port_cong_status(struct mlx5_core_dev *mdev, int protocol, int priority, int *is_enable); int mlx5_modify_port_cong_status(struct mlx5_core_dev *mdev, int protocol, int priority, int enable); int mlx5_query_port_cong_params(struct mlx5_core_dev *mdev, int protocol, void *out, int out_size); int mlx5_modify_port_cong_params(struct mlx5_core_dev *mdev, void *in, int in_size); int mlx5_query_port_cong_statistics(struct mlx5_core_dev *mdev, int clear, void *out, int out_size); int mlx5_set_diagnostic_params(struct mlx5_core_dev *mdev, void *in, int in_size); int mlx5_query_diagnostic_counters(struct mlx5_core_dev *mdev, u8 num_of_samples, u16 sample_index, void *out, int out_size); int mlx5_vsc_find_cap(struct mlx5_core_dev *mdev); int mlx5_vsc_lock(struct mlx5_core_dev *mdev); void mlx5_vsc_unlock(struct mlx5_core_dev *mdev); int mlx5_vsc_set_space(struct mlx5_core_dev *mdev, u16 space); int mlx5_vsc_wait_on_flag(struct mlx5_core_dev *mdev, u32 expected); int mlx5_vsc_write(struct mlx5_core_dev *mdev, u32 addr, const u32 *data); int mlx5_vsc_read(struct mlx5_core_dev *mdev, u32 addr, u32 *data); int mlx5_vsc_lock_addr_space(struct mlx5_core_dev *mdev, u32 addr); int mlx5_vsc_unlock_addr_space(struct mlx5_core_dev *mdev, u32 addr); int mlx5_pci_read_power_status(struct mlx5_core_dev *mdev, u16 *p_power, u8 *p_status); static inline u32 mlx5_mkey_to_idx(u32 mkey) { return mkey >> 8; } static inline u32 mlx5_idx_to_mkey(u32 mkey_idx) { return mkey_idx << 8; } static inline u8 mlx5_mkey_variant(u32 mkey) { return mkey & 0xff; } enum { MLX5_PROF_MASK_QP_SIZE = (u64)1 << 0, MLX5_PROF_MASK_MR_CACHE = (u64)1 << 1, }; enum { MAX_MR_CACHE_ENTRIES = 15, }; struct mlx5_interface { void * (*add)(struct mlx5_core_dev *dev); void (*remove)(struct mlx5_core_dev *dev, void *context); void (*event)(struct mlx5_core_dev *dev, void *context, enum mlx5_dev_event event, unsigned long param); void * (*get_dev)(void *context); int protocol; struct list_head list; }; void *mlx5_get_protocol_dev(struct mlx5_core_dev *mdev, int protocol); int mlx5_register_interface(struct mlx5_interface *intf); void mlx5_unregister_interface(struct mlx5_interface *intf); unsigned int mlx5_core_reserved_gids_count(struct mlx5_core_dev *dev); int mlx5_core_roce_gid_set(struct mlx5_core_dev *dev, unsigned int index, u8 roce_version, u8 roce_l3_type, const u8 *gid, const u8 *mac, bool vlan, u16 vlan_id); struct mlx5_profile { u64 mask; u8 log_max_qp; struct { int size; int limit; } mr_cache[MAX_MR_CACHE_ENTRIES]; }; enum { MLX5_PCI_DEV_IS_VF = 1 << 0, }; enum { MLX5_TRIGGERED_CMD_COMP = (u64)1 << 32, }; static inline int mlx5_core_is_pf(struct mlx5_core_dev *dev) { return !(dev->priv.pci_dev_data & MLX5_PCI_DEV_IS_VF); } #ifdef RATELIMIT int mlx5_init_rl_table(struct mlx5_core_dev *dev); void mlx5_cleanup_rl_table(struct mlx5_core_dev *dev); int mlx5_rl_add_rate(struct mlx5_core_dev *dev, u32 rate, u32 burst, u16 *index); void mlx5_rl_remove_rate(struct mlx5_core_dev *dev, u32 rate, u32 burst); bool mlx5_rl_is_in_range(const struct mlx5_core_dev *dev, u32 rate, u32 burst); static inline bool mlx5_rl_is_supported(struct mlx5_core_dev *dev) { return !!(dev->priv.rl_table.max_size); } #endif void mlx5_disable_interrupts(struct mlx5_core_dev *); void mlx5_poll_interrupts(struct mlx5_core_dev *); static inline int mlx5_get_qp_default_ts(struct mlx5_core_dev *dev) { return !MLX5_CAP_ROCE(dev, qp_ts_format) ? MLX5_QPC_TIMESTAMP_FORMAT_FREE_RUNNING : MLX5_QPC_TIMESTAMP_FORMAT_DEFAULT; } static inline int mlx5_get_rq_default_ts(struct mlx5_core_dev *dev) { return !MLX5_CAP_GEN(dev, rq_ts_format) ? MLX5_RQC_TIMESTAMP_FORMAT_FREE_RUNNING : MLX5_RQC_TIMESTAMP_FORMAT_DEFAULT; } static inline int mlx5_get_sq_default_ts(struct mlx5_core_dev *dev) { return !MLX5_CAP_GEN(dev, sq_ts_format) ? MLX5_SQC_TIMESTAMP_FORMAT_FREE_RUNNING : MLX5_SQC_TIMESTAMP_FORMAT_DEFAULT; } #endif /* MLX5_DRIVER_H */ diff --git a/sys/dev/mlx5/mlx5_core/mlx5_alloc.c b/sys/dev/mlx5/mlx5_core/mlx5_alloc.c index 1dabbd9f586f..99e39b80a4ec 100644 --- a/sys/dev/mlx5/mlx5_core/mlx5_alloc.c +++ b/sys/dev/mlx5/mlx5_core/mlx5_alloc.c @@ -1,261 +1,254 @@ /*- * Copyright (c) 2013-2017, Mellanox Technologies, Ltd. All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY AUTHOR AND CONTRIBUTORS `AS IS' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * $FreeBSD$ */ #include #include #include #include #include #include #include "mlx5_core.h" /* Handling for queue buffers -- we allocate a bunch of memory and * register it in a memory region at HCA virtual address 0. If the * requested size is > max_direct, we split the allocation into * multiple pages, so we don't require too much contiguous memory. */ static void mlx5_buf_load_mem_cb(void *arg, bus_dma_segment_t *segs, int nseg, int error) { struct mlx5_buf *buf; uint8_t owned; int x; buf = (struct mlx5_buf *)arg; owned = MLX5_DMA_OWNED(buf->dev); if (!owned) MLX5_DMA_LOCK(buf->dev); if (error == 0) { for (x = 0; x != nseg; x++) { buf->page_list[x] = segs[x].ds_addr; KASSERT(segs[x].ds_len == PAGE_SIZE, ("Invalid segment size")); } buf->load_done = MLX5_LOAD_ST_SUCCESS; } else { buf->load_done = MLX5_LOAD_ST_FAILURE; } MLX5_DMA_DONE(buf->dev); if (!owned) MLX5_DMA_UNLOCK(buf->dev); } int mlx5_buf_alloc(struct mlx5_core_dev *dev, int size, int max_direct, struct mlx5_buf *buf) { int err; buf->npages = howmany(size, PAGE_SIZE); buf->page_shift = PAGE_SHIFT; buf->load_done = MLX5_LOAD_ST_NONE; buf->dev = dev; buf->page_list = kcalloc(buf->npages, sizeof(*buf->page_list), GFP_KERNEL); err = -bus_dma_tag_create( bus_get_dma_tag(dev->pdev->dev.bsddev), PAGE_SIZE, /* alignment */ 0, /* no boundary */ BUS_SPACE_MAXADDR, /* lowaddr */ BUS_SPACE_MAXADDR, /* highaddr */ NULL, NULL, /* filter, filterarg */ PAGE_SIZE * buf->npages, /* maxsize */ buf->npages, /* nsegments */ PAGE_SIZE, /* maxsegsize */ 0, /* flags */ NULL, NULL, /* lockfunc, lockfuncarg */ &buf->dma_tag); if (err != 0) goto err_dma_tag; /* allocate memory */ err = -bus_dmamem_alloc(buf->dma_tag, &buf->direct.buf, BUS_DMA_WAITOK | BUS_DMA_COHERENT, &buf->dma_map); if (err != 0) goto err_dma_alloc; /* load memory into DMA */ MLX5_DMA_LOCK(dev); err = bus_dmamap_load( buf->dma_tag, buf->dma_map, buf->direct.buf, PAGE_SIZE * buf->npages, &mlx5_buf_load_mem_cb, buf, BUS_DMA_WAITOK | BUS_DMA_COHERENT); while (buf->load_done == MLX5_LOAD_ST_NONE) MLX5_DMA_WAIT(dev); MLX5_DMA_UNLOCK(dev); /* check for error */ if (buf->load_done != MLX5_LOAD_ST_SUCCESS) { err = -ENOMEM; goto err_dma_load; } /* clean memory */ memset(buf->direct.buf, 0, PAGE_SIZE * buf->npages); /* flush memory to RAM */ bus_dmamap_sync(buf->dev->cmd.dma_tag, buf->dma_map, BUS_DMASYNC_PREWRITE); return (0); err_dma_load: bus_dmamem_free(buf->dma_tag, buf->direct.buf, buf->dma_map); err_dma_alloc: bus_dma_tag_destroy(buf->dma_tag); err_dma_tag: kfree(buf->page_list); return (err); } void mlx5_buf_free(struct mlx5_core_dev *dev, struct mlx5_buf *buf) { bus_dmamap_unload(buf->dma_tag, buf->dma_map); bus_dmamem_free(buf->dma_tag, buf->direct.buf, buf->dma_map); bus_dma_tag_destroy(buf->dma_tag); kfree(buf->page_list); } EXPORT_SYMBOL_GPL(mlx5_buf_free); -static struct mlx5_db_pgdir *mlx5_alloc_db_pgdir(struct mlx5_core_dev *dev, - int node) +static struct mlx5_db_pgdir *mlx5_alloc_db_pgdir(struct mlx5_core_dev *dev) { struct mlx5_db_pgdir *pgdir; pgdir = kzalloc(sizeof(*pgdir), GFP_KERNEL); bitmap_fill(pgdir->bitmap, MLX5_DB_PER_PAGE); pgdir->fw_page = mlx5_fwp_alloc(dev, GFP_KERNEL, 1); if (pgdir->fw_page != NULL) { pgdir->db_page = pgdir->fw_page->virt_addr; pgdir->db_dma = pgdir->fw_page->dma_addr; /* clean allocated memory */ memset(pgdir->db_page, 0, MLX5_ADAPTER_PAGE_SIZE); /* flush memory to RAM */ mlx5_fwp_flush(pgdir->fw_page); } if (!pgdir->db_page) { kfree(pgdir); return NULL; } return pgdir; } static int mlx5_alloc_db_from_pgdir(struct mlx5_db_pgdir *pgdir, struct mlx5_db *db) { int offset; int i; i = find_first_bit(pgdir->bitmap, MLX5_DB_PER_PAGE); if (i >= MLX5_DB_PER_PAGE) return -ENOMEM; __clear_bit(i, pgdir->bitmap); db->u.pgdir = pgdir; db->index = i; offset = db->index * L1_CACHE_BYTES; db->db = pgdir->db_page + offset / sizeof(*pgdir->db_page); db->dma = pgdir->db_dma + offset; db->db[0] = 0; db->db[1] = 0; return 0; } -int mlx5_db_alloc_node(struct mlx5_core_dev *dev, struct mlx5_db *db, int node) +int mlx5_db_alloc(struct mlx5_core_dev *dev, struct mlx5_db *db) { struct mlx5_db_pgdir *pgdir; int ret = 0; mutex_lock(&dev->priv.pgdir_mutex); list_for_each_entry(pgdir, &dev->priv.pgdir_list, list) if (!mlx5_alloc_db_from_pgdir(pgdir, db)) goto out; - pgdir = mlx5_alloc_db_pgdir(dev, node); + pgdir = mlx5_alloc_db_pgdir(dev); if (!pgdir) { ret = -ENOMEM; goto out; } list_add(&pgdir->list, &dev->priv.pgdir_list); /* This should never fail -- we just allocated an empty page: */ WARN_ON(mlx5_alloc_db_from_pgdir(pgdir, db)); out: mutex_unlock(&dev->priv.pgdir_mutex); return ret; } -EXPORT_SYMBOL_GPL(mlx5_db_alloc_node); - -int mlx5_db_alloc(struct mlx5_core_dev *dev, struct mlx5_db *db) -{ - return mlx5_db_alloc_node(dev, db, dev->priv.numa_node); -} EXPORT_SYMBOL_GPL(mlx5_db_alloc); void mlx5_db_free(struct mlx5_core_dev *dev, struct mlx5_db *db) { 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)) { mlx5_fwp_free(db->u.pgdir->fw_page); list_del(&db->u.pgdir->list); kfree(db->u.pgdir); } mutex_unlock(&dev->priv.pgdir_mutex); } EXPORT_SYMBOL_GPL(mlx5_db_free); void mlx5_fill_page_array(struct mlx5_buf *buf, __be64 *pas) { int i; for (i = 0; i != buf->npages; i++) pas[i] = cpu_to_be64(buf->page_list[i]); } EXPORT_SYMBOL_GPL(mlx5_fill_page_array); diff --git a/sys/dev/mlx5/mlx5_core/mlx5_main.c b/sys/dev/mlx5/mlx5_core/mlx5_main.c index 755e8844a892..7527af524456 100644 --- a/sys/dev/mlx5/mlx5_core/mlx5_main.c +++ b/sys/dev/mlx5/mlx5_core/mlx5_main.c @@ -1,2062 +1,2064 @@ /*- * Copyright (c) 2013-2020, Mellanox Technologies, Ltd. All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY AUTHOR AND CONTRIBUTORS `AS IS' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * $FreeBSD$ */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include "mlx5_core.h" #include "eswitch.h" #include "fs_core.h" #ifdef PCI_IOV #include #include #include #endif static const char mlx5_version[] = "Mellanox Core driver " DRIVER_VERSION " (" DRIVER_RELDATE ")"; MODULE_AUTHOR("Eli Cohen "); MODULE_DESCRIPTION("Mellanox Connect-IB, ConnectX-4 core driver"); MODULE_LICENSE("Dual BSD/GPL"); MODULE_DEPEND(mlx5, linuxkpi, 1, 1, 1); MODULE_DEPEND(mlx5, mlxfw, 1, 1, 1); MODULE_DEPEND(mlx5, firmware, 1, 1, 1); MODULE_VERSION(mlx5, 1); SYSCTL_NODE(_hw, OID_AUTO, mlx5, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, "mlx5 hardware controls"); int mlx5_core_debug_mask; SYSCTL_INT(_hw_mlx5, OID_AUTO, debug_mask, CTLFLAG_RWTUN, &mlx5_core_debug_mask, 0, "debug mask: 1 = dump cmd data, 2 = dump cmd exec time, 3 = both. Default=0"); #define MLX5_DEFAULT_PROF 2 static int mlx5_prof_sel = MLX5_DEFAULT_PROF; SYSCTL_INT(_hw_mlx5, OID_AUTO, prof_sel, CTLFLAG_RWTUN, &mlx5_prof_sel, 0, "profile selector. Valid range 0 - 2"); static int mlx5_fast_unload_enabled = 1; SYSCTL_INT(_hw_mlx5, OID_AUTO, fast_unload_enabled, CTLFLAG_RWTUN, &mlx5_fast_unload_enabled, 0, "Set to enable fast unload. Clear to disable."); -#define NUMA_NO_NODE -1 - static LIST_HEAD(intf_list); static LIST_HEAD(dev_list); static DEFINE_MUTEX(intf_mutex); struct mlx5_device_context { struct list_head list; struct mlx5_interface *intf; void *context; }; enum { MLX5_ATOMIC_REQ_MODE_BE = 0x0, MLX5_ATOMIC_REQ_MODE_HOST_ENDIANNESS = 0x1, }; static struct mlx5_profile profiles[] = { [0] = { .mask = 0, }, [1] = { .mask = MLX5_PROF_MASK_QP_SIZE, .log_max_qp = 12, }, [2] = { .mask = MLX5_PROF_MASK_QP_SIZE | MLX5_PROF_MASK_MR_CACHE, .log_max_qp = 17, .mr_cache[0] = { .size = 500, .limit = 250 }, .mr_cache[1] = { .size = 500, .limit = 250 }, .mr_cache[2] = { .size = 500, .limit = 250 }, .mr_cache[3] = { .size = 500, .limit = 250 }, .mr_cache[4] = { .size = 500, .limit = 250 }, .mr_cache[5] = { .size = 500, .limit = 250 }, .mr_cache[6] = { .size = 500, .limit = 250 }, .mr_cache[7] = { .size = 500, .limit = 250 }, .mr_cache[8] = { .size = 500, .limit = 250 }, .mr_cache[9] = { .size = 500, .limit = 250 }, .mr_cache[10] = { .size = 500, .limit = 250 }, .mr_cache[11] = { .size = 500, .limit = 250 }, .mr_cache[12] = { .size = 64, .limit = 32 }, .mr_cache[13] = { .size = 32, .limit = 16 }, .mr_cache[14] = { .size = 16, .limit = 8 }, }, [3] = { .mask = MLX5_PROF_MASK_QP_SIZE, .log_max_qp = 17, }, }; #ifdef PCI_IOV static const char iov_mac_addr_name[] = "mac-addr"; static const char iov_node_guid_name[] = "node-guid"; static const char iov_port_guid_name[] = "port-guid"; #endif static int set_dma_caps(struct pci_dev *pdev) { struct mlx5_core_dev *dev = pci_get_drvdata(pdev); int err; err = pci_set_dma_mask(pdev, DMA_BIT_MASK(64)); if (err) { mlx5_core_warn(dev, "couldn't set 64-bit PCI DMA mask\n"); err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32)); if (err) { mlx5_core_err(dev, "Can't set PCI DMA mask, aborting\n"); return err; } } err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64)); if (err) { mlx5_core_warn(dev, "couldn't set 64-bit consistent PCI DMA mask\n"); err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32)); if (err) { mlx5_core_err(dev, "Can't set consistent PCI DMA mask, aborting\n"); return err; } } dma_set_max_seg_size(&pdev->dev, 2u * 1024 * 1024 * 1024); return err; } int mlx5_pci_read_power_status(struct mlx5_core_dev *dev, u16 *p_power, u8 *p_status) { u32 in[MLX5_ST_SZ_DW(mpein_reg)] = {}; u32 out[MLX5_ST_SZ_DW(mpein_reg)] = {}; int err; err = mlx5_core_access_reg(dev, in, sizeof(in), out, sizeof(out), MLX5_ACCESS_REG_SUMMARY_CTRL_ID_MPEIN, 0, 0); *p_status = MLX5_GET(mpein_reg, out, pwr_status); *p_power = MLX5_GET(mpein_reg, out, pci_power); return err; } static int mlx5_pci_enable_device(struct mlx5_core_dev *dev) { struct pci_dev *pdev = dev->pdev; int err = 0; mutex_lock(&dev->pci_status_mutex); if (dev->pci_status == MLX5_PCI_STATUS_DISABLED) { err = pci_enable_device(pdev); if (!err) dev->pci_status = MLX5_PCI_STATUS_ENABLED; } mutex_unlock(&dev->pci_status_mutex); return err; } static void mlx5_pci_disable_device(struct mlx5_core_dev *dev) { struct pci_dev *pdev = dev->pdev; mutex_lock(&dev->pci_status_mutex); if (dev->pci_status == MLX5_PCI_STATUS_ENABLED) { pci_disable_device(pdev); dev->pci_status = MLX5_PCI_STATUS_DISABLED; } mutex_unlock(&dev->pci_status_mutex); } static int request_bar(struct pci_dev *pdev) { struct mlx5_core_dev *dev = pci_get_drvdata(pdev); int err = 0; if (!(pci_resource_flags(pdev, 0) & IORESOURCE_MEM)) { mlx5_core_err(dev, "Missing registers BAR, aborting\n"); return -ENODEV; } err = pci_request_regions(pdev, DRIVER_NAME); if (err) mlx5_core_err(dev, "Couldn't get PCI resources, aborting\n"); return err; } static void release_bar(struct pci_dev *pdev) { pci_release_regions(pdev); } static int mlx5_enable_msix(struct mlx5_core_dev *dev) { struct mlx5_priv *priv = &dev->priv; struct mlx5_eq_table *table = &priv->eq_table; int num_eqs = 1 << MLX5_CAP_GEN(dev, log_max_eq); int limit = dev->msix_eqvec; int nvec = MLX5_EQ_VEC_COMP_BASE; int i; if (limit > 0) nvec += limit; else nvec += MLX5_CAP_GEN(dev, num_ports) * num_online_cpus(); if (nvec > num_eqs) nvec = num_eqs; if (nvec > 256) nvec = 256; /* limit of firmware API */ if (nvec <= MLX5_EQ_VEC_COMP_BASE) return -ENOMEM; priv->msix_arr = kzalloc(nvec * sizeof(*priv->msix_arr), GFP_KERNEL); for (i = 0; i < nvec; i++) priv->msix_arr[i].entry = i; nvec = pci_enable_msix_range(dev->pdev, priv->msix_arr, MLX5_EQ_VEC_COMP_BASE + 1, nvec); if (nvec < 0) return nvec; table->num_comp_vectors = nvec - MLX5_EQ_VEC_COMP_BASE; return 0; } static void mlx5_disable_msix(struct mlx5_core_dev *dev) { struct mlx5_priv *priv = &dev->priv; pci_disable_msix(dev->pdev); kfree(priv->msix_arr); } struct mlx5_reg_host_endianess { u8 he; u8 rsvd[15]; }; #define CAP_MASK(pos, size) ((u64)((1 << (size)) - 1) << (pos)) enum { MLX5_CAP_BITS_RW_MASK = CAP_MASK(MLX5_CAP_OFF_CMDIF_CSUM, 2) | MLX5_DEV_CAP_FLAG_DCT | MLX5_DEV_CAP_FLAG_DRAIN_SIGERR, }; static u16 to_fw_pkey_sz(struct mlx5_core_dev *dev, u32 size) { switch (size) { case 128: return 0; case 256: return 1; case 512: return 2; case 1024: return 3; case 2048: return 4; case 4096: return 5; default: mlx5_core_warn(dev, "invalid pkey table size %d\n", size); return 0; } } static int mlx5_core_get_caps_mode(struct mlx5_core_dev *dev, enum mlx5_cap_type cap_type, enum mlx5_cap_mode cap_mode) { u8 in[MLX5_ST_SZ_BYTES(query_hca_cap_in)]; int out_sz = MLX5_ST_SZ_BYTES(query_hca_cap_out); void *out, *hca_caps; u16 opmod = (cap_type << 1) | (cap_mode & 0x01); int err; memset(in, 0, sizeof(in)); out = kzalloc(out_sz, GFP_KERNEL); MLX5_SET(query_hca_cap_in, in, opcode, MLX5_CMD_OP_QUERY_HCA_CAP); MLX5_SET(query_hca_cap_in, in, op_mod, opmod); err = mlx5_cmd_exec(dev, in, sizeof(in), out, out_sz); if (err) { mlx5_core_warn(dev, "QUERY_HCA_CAP : type(%x) opmode(%x) Failed(%d)\n", cap_type, cap_mode, err); goto query_ex; } hca_caps = MLX5_ADDR_OF(query_hca_cap_out, out, capability); switch (cap_mode) { case HCA_CAP_OPMOD_GET_MAX: memcpy(dev->hca_caps_max[cap_type], hca_caps, MLX5_UN_SZ_BYTES(hca_cap_union)); break; case HCA_CAP_OPMOD_GET_CUR: memcpy(dev->hca_caps_cur[cap_type], hca_caps, MLX5_UN_SZ_BYTES(hca_cap_union)); break; default: mlx5_core_warn(dev, "Tried to query dev cap type(%x) with wrong opmode(%x)\n", cap_type, cap_mode); err = -EINVAL; break; } query_ex: kfree(out); return err; } int mlx5_core_get_caps(struct mlx5_core_dev *dev, enum mlx5_cap_type cap_type) { int ret; ret = mlx5_core_get_caps_mode(dev, cap_type, HCA_CAP_OPMOD_GET_CUR); if (ret) return ret; return mlx5_core_get_caps_mode(dev, cap_type, HCA_CAP_OPMOD_GET_MAX); } static int set_caps(struct mlx5_core_dev *dev, void *in, int in_sz) { u32 out[MLX5_ST_SZ_DW(set_hca_cap_out)] = {0}; MLX5_SET(set_hca_cap_in, in, opcode, MLX5_CMD_OP_SET_HCA_CAP); return mlx5_cmd_exec(dev, in, in_sz, out, sizeof(out)); } static int handle_hca_cap(struct mlx5_core_dev *dev) { void *set_ctx = NULL; struct mlx5_profile *prof = dev->profile; int err = -ENOMEM; int set_sz = MLX5_ST_SZ_BYTES(set_hca_cap_in); void *set_hca_cap; set_ctx = kzalloc(set_sz, GFP_KERNEL); err = mlx5_core_get_caps(dev, MLX5_CAP_GENERAL); if (err) goto query_ex; set_hca_cap = MLX5_ADDR_OF(set_hca_cap_in, set_ctx, capability); memcpy(set_hca_cap, dev->hca_caps_cur[MLX5_CAP_GENERAL], MLX5_ST_SZ_BYTES(cmd_hca_cap)); mlx5_core_dbg(dev, "Current Pkey table size %d Setting new size %d\n", mlx5_to_sw_pkey_sz(MLX5_CAP_GEN(dev, pkey_table_size)), 128); /* we limit the size of the pkey table to 128 entries for now */ MLX5_SET(cmd_hca_cap, set_hca_cap, pkey_table_size, to_fw_pkey_sz(dev, 128)); if (prof->mask & MLX5_PROF_MASK_QP_SIZE) MLX5_SET(cmd_hca_cap, set_hca_cap, log_max_qp, prof->log_max_qp); /* disable cmdif checksum */ MLX5_SET(cmd_hca_cap, set_hca_cap, cmdif_checksum, 0); /* Enable 4K UAR only when HCA supports it and page size is bigger * than 4K. */ if (MLX5_CAP_GEN_MAX(dev, uar_4k) && PAGE_SIZE > 4096) MLX5_SET(cmd_hca_cap, set_hca_cap, uar_4k, 1); /* enable drain sigerr */ MLX5_SET(cmd_hca_cap, set_hca_cap, drain_sigerr, 1); MLX5_SET(cmd_hca_cap, set_hca_cap, log_uar_page_sz, PAGE_SHIFT - 12); err = set_caps(dev, set_ctx, set_sz); query_ex: kfree(set_ctx); return err; } static int handle_hca_cap_atomic(struct mlx5_core_dev *dev) { void *set_ctx; void *set_hca_cap; int set_sz = MLX5_ST_SZ_BYTES(set_hca_cap_in); int req_endianness; int err; if (MLX5_CAP_GEN(dev, atomic)) { err = mlx5_core_get_caps(dev, MLX5_CAP_ATOMIC); if (err) return err; } else { return 0; } req_endianness = MLX5_CAP_ATOMIC(dev, supported_atomic_req_8B_endianess_mode_1); if (req_endianness != MLX5_ATOMIC_REQ_MODE_HOST_ENDIANNESS) return 0; set_ctx = kzalloc(set_sz, GFP_KERNEL); if (!set_ctx) return -ENOMEM; MLX5_SET(set_hca_cap_in, set_ctx, op_mod, MLX5_SET_HCA_CAP_OP_MOD_ATOMIC << 1); set_hca_cap = MLX5_ADDR_OF(set_hca_cap_in, set_ctx, capability); /* Set requestor to host endianness */ MLX5_SET(atomic_caps, set_hca_cap, atomic_req_8B_endianess_mode, MLX5_ATOMIC_REQ_MODE_HOST_ENDIANNESS); err = set_caps(dev, set_ctx, set_sz); kfree(set_ctx); return err; } static int set_hca_ctrl(struct mlx5_core_dev *dev) { struct mlx5_reg_host_endianess he_in; struct mlx5_reg_host_endianess he_out; int err; if (MLX5_CAP_GEN(dev, port_type) == MLX5_CAP_PORT_TYPE_ETH && !MLX5_CAP_GEN(dev, roce)) return 0; memset(&he_in, 0, sizeof(he_in)); he_in.he = MLX5_SET_HOST_ENDIANNESS; err = mlx5_core_access_reg(dev, &he_in, sizeof(he_in), &he_out, sizeof(he_out), MLX5_REG_HOST_ENDIANNESS, 0, 1); return err; } static int mlx5_core_enable_hca(struct mlx5_core_dev *dev, u16 func_id) { u32 out[MLX5_ST_SZ_DW(enable_hca_out)] = {0}; u32 in[MLX5_ST_SZ_DW(enable_hca_in)] = {0}; MLX5_SET(enable_hca_in, in, opcode, MLX5_CMD_OP_ENABLE_HCA); MLX5_SET(enable_hca_in, in, function_id, func_id); return mlx5_cmd_exec(dev, &in, sizeof(in), &out, sizeof(out)); } static int mlx5_core_disable_hca(struct mlx5_core_dev *dev) { u32 out[MLX5_ST_SZ_DW(disable_hca_out)] = {0}; u32 in[MLX5_ST_SZ_DW(disable_hca_in)] = {0}; MLX5_SET(disable_hca_in, in, opcode, MLX5_CMD_OP_DISABLE_HCA); return mlx5_cmd_exec(dev, in, sizeof(in), out, sizeof(out)); } static int mlx5_core_set_issi(struct mlx5_core_dev *dev) { u32 query_in[MLX5_ST_SZ_DW(query_issi_in)] = {0}; u32 query_out[MLX5_ST_SZ_DW(query_issi_out)] = {0}; u32 sup_issi; int err; MLX5_SET(query_issi_in, query_in, opcode, MLX5_CMD_OP_QUERY_ISSI); err = mlx5_cmd_exec(dev, query_in, sizeof(query_in), query_out, sizeof(query_out)); if (err) { u32 syndrome; u8 status; mlx5_cmd_mbox_status(query_out, &status, &syndrome); if (status == MLX5_CMD_STAT_BAD_OP_ERR) { mlx5_core_dbg(dev, "Only ISSI 0 is supported\n"); return 0; } mlx5_core_err(dev, "failed to query ISSI\n"); return err; } sup_issi = MLX5_GET(query_issi_out, query_out, supported_issi_dw0); if (sup_issi & (1 << 1)) { u32 set_in[MLX5_ST_SZ_DW(set_issi_in)] = {0}; u32 set_out[MLX5_ST_SZ_DW(set_issi_out)] = {0}; MLX5_SET(set_issi_in, set_in, opcode, MLX5_CMD_OP_SET_ISSI); MLX5_SET(set_issi_in, set_in, current_issi, 1); err = mlx5_cmd_exec(dev, set_in, sizeof(set_in), set_out, sizeof(set_out)); if (err) { mlx5_core_err(dev, "failed to set ISSI=1 err(%d)\n", err); return err; } dev->issi = 1; return 0; } else if (sup_issi & (1 << 0)) { return 0; } return -ENOTSUPP; } int mlx5_vector2eqn(struct mlx5_core_dev *dev, int vector, int *eqn, int *irqn) { struct mlx5_eq_table *table = &dev->priv.eq_table; struct mlx5_eq *eq; int err = -ENOENT; spin_lock(&table->lock); list_for_each_entry(eq, &table->comp_eqs_list, list) { if (eq->index == vector) { *eqn = eq->eqn; *irqn = eq->irqn; err = 0; break; } } spin_unlock(&table->lock); return err; } EXPORT_SYMBOL(mlx5_vector2eqn); static void free_comp_eqs(struct mlx5_core_dev *dev) { struct mlx5_eq_table *table = &dev->priv.eq_table; struct mlx5_eq *eq, *n; spin_lock(&table->lock); list_for_each_entry_safe(eq, n, &table->comp_eqs_list, list) { list_del(&eq->list); spin_unlock(&table->lock); if (mlx5_destroy_unmap_eq(dev, eq)) mlx5_core_warn(dev, "failed to destroy EQ 0x%x\n", eq->eqn); kfree(eq); spin_lock(&table->lock); } spin_unlock(&table->lock); } static int alloc_comp_eqs(struct mlx5_core_dev *dev) { struct mlx5_eq_table *table = &dev->priv.eq_table; struct mlx5_eq *eq; int ncomp_vec; int nent; int err; int i; INIT_LIST_HEAD(&table->comp_eqs_list); ncomp_vec = table->num_comp_vectors; nent = MLX5_COMP_EQ_SIZE; for (i = 0; i < ncomp_vec; i++) { - eq = kzalloc(sizeof(*eq), GFP_KERNEL); + eq = kzalloc_node(sizeof(*eq), GFP_KERNEL, dev->priv.numa_node); err = mlx5_create_map_eq(dev, eq, i + MLX5_EQ_VEC_COMP_BASE, nent, 0); if (err) { kfree(eq); goto clean; } mlx5_core_dbg(dev, "allocated completion EQN %d\n", eq->eqn); eq->index = i; spin_lock(&table->lock); list_add_tail(&eq->list, &table->comp_eqs_list); spin_unlock(&table->lock); } return 0; clean: free_comp_eqs(dev); return err; } static inline int fw_initializing(struct mlx5_core_dev *dev) { return ioread32be(&dev->iseg->initializing) >> 31; } static int wait_fw_init(struct mlx5_core_dev *dev, u32 max_wait_mili, u32 warn_time_mili) { int warn = jiffies + msecs_to_jiffies(warn_time_mili); int end = jiffies + msecs_to_jiffies(max_wait_mili); int err = 0; MPASS(max_wait_mili > warn_time_mili); while (fw_initializing(dev) == 1) { if (time_after(jiffies, end)) { err = -EBUSY; break; } if (warn_time_mili && time_after(jiffies, warn)) { mlx5_core_warn(dev, "Waiting for FW initialization, timeout abort in %u s\n", (unsigned)(jiffies_to_msecs(end - warn) / 1000)); warn = jiffies + msecs_to_jiffies(warn_time_mili); } msleep(FW_INIT_WAIT_MS); } if (err != 0) mlx5_core_dbg(dev, "Full initializing bit dword = 0x%x\n", ioread32be(&dev->iseg->initializing)); return err; } static void mlx5_add_device(struct mlx5_interface *intf, struct mlx5_priv *priv) { struct mlx5_device_context *dev_ctx; struct mlx5_core_dev *dev = container_of(priv, struct mlx5_core_dev, priv); - dev_ctx = kzalloc(sizeof(*dev_ctx), GFP_KERNEL); + dev_ctx = kzalloc_node(sizeof(*dev_ctx), GFP_KERNEL, priv->numa_node); if (!dev_ctx) return; dev_ctx->intf = intf; CURVNET_SET_QUIET(vnet0); dev_ctx->context = intf->add(dev); CURVNET_RESTORE(); if (dev_ctx->context) { spin_lock_irq(&priv->ctx_lock); list_add_tail(&dev_ctx->list, &priv->ctx_list); spin_unlock_irq(&priv->ctx_lock); } else { kfree(dev_ctx); } } static void mlx5_remove_device(struct mlx5_interface *intf, struct mlx5_priv *priv) { struct mlx5_device_context *dev_ctx; struct mlx5_core_dev *dev = container_of(priv, struct mlx5_core_dev, priv); list_for_each_entry(dev_ctx, &priv->ctx_list, list) if (dev_ctx->intf == intf) { spin_lock_irq(&priv->ctx_lock); list_del(&dev_ctx->list); spin_unlock_irq(&priv->ctx_lock); intf->remove(dev, dev_ctx->context); kfree(dev_ctx); return; } } int mlx5_register_device(struct mlx5_core_dev *dev) { struct mlx5_priv *priv = &dev->priv; struct mlx5_interface *intf; mutex_lock(&intf_mutex); list_add_tail(&priv->dev_list, &dev_list); list_for_each_entry(intf, &intf_list, list) mlx5_add_device(intf, priv); mutex_unlock(&intf_mutex); return 0; } void mlx5_unregister_device(struct mlx5_core_dev *dev) { struct mlx5_priv *priv = &dev->priv; struct mlx5_interface *intf; mutex_lock(&intf_mutex); list_for_each_entry(intf, &intf_list, list) mlx5_remove_device(intf, priv); list_del(&priv->dev_list); mutex_unlock(&intf_mutex); } int mlx5_register_interface(struct mlx5_interface *intf) { struct mlx5_priv *priv; if (!intf->add || !intf->remove) return -EINVAL; mutex_lock(&intf_mutex); list_add_tail(&intf->list, &intf_list); list_for_each_entry(priv, &dev_list, dev_list) mlx5_add_device(intf, priv); mutex_unlock(&intf_mutex); return 0; } EXPORT_SYMBOL(mlx5_register_interface); void mlx5_unregister_interface(struct mlx5_interface *intf) { struct mlx5_priv *priv; mutex_lock(&intf_mutex); list_for_each_entry(priv, &dev_list, dev_list) mlx5_remove_device(intf, priv); list_del(&intf->list); mutex_unlock(&intf_mutex); } EXPORT_SYMBOL(mlx5_unregister_interface); void *mlx5_get_protocol_dev(struct mlx5_core_dev *mdev, int protocol) { struct mlx5_priv *priv = &mdev->priv; struct mlx5_device_context *dev_ctx; unsigned long flags; void *result = NULL; spin_lock_irqsave(&priv->ctx_lock, flags); list_for_each_entry(dev_ctx, &mdev->priv.ctx_list, list) if ((dev_ctx->intf->protocol == protocol) && dev_ctx->intf->get_dev) { result = dev_ctx->intf->get_dev(dev_ctx->context); break; } spin_unlock_irqrestore(&priv->ctx_lock, flags); return result; } EXPORT_SYMBOL(mlx5_get_protocol_dev); static int mlx5_auto_fw_update; SYSCTL_INT(_hw_mlx5, OID_AUTO, auto_fw_update, CTLFLAG_RDTUN | CTLFLAG_NOFETCH, &mlx5_auto_fw_update, 0, "Allow automatic firmware update on driver start"); static int mlx5_firmware_update(struct mlx5_core_dev *dev) { const struct firmware *fw; int err; TUNABLE_INT_FETCH("hw.mlx5.auto_fw_update", &mlx5_auto_fw_update); if (!mlx5_auto_fw_update) return (0); fw = firmware_get("mlx5fw_mfa"); if (fw) { err = mlx5_firmware_flash(dev, fw); firmware_put(fw, FIRMWARE_UNLOAD); } else return (-ENOENT); return err; } static int mlx5_pci_init(struct mlx5_core_dev *dev, struct mlx5_priv *priv) { struct pci_dev *pdev = dev->pdev; device_t bsddev; int err; pdev = dev->pdev; bsddev = pdev->dev.bsddev; pci_set_drvdata(dev->pdev, dev); strncpy(priv->name, dev_name(&pdev->dev), MLX5_MAX_NAME_LEN); priv->name[MLX5_MAX_NAME_LEN - 1] = 0; mutex_init(&priv->pgdir_mutex); INIT_LIST_HEAD(&priv->pgdir_list); spin_lock_init(&priv->mkey_lock); - priv->numa_node = NUMA_NO_NODE; - err = mlx5_pci_enable_device(dev); if (err) { mlx5_core_err(dev, "Cannot enable PCI device, aborting\n"); goto err_dbg; } err = request_bar(pdev); if (err) { mlx5_core_err(dev, "error requesting BARs, aborting\n"); goto err_disable; } pci_set_master(pdev); err = set_dma_caps(pdev); if (err) { mlx5_core_err(dev, "Failed setting DMA capabilities mask, aborting\n"); goto err_clr_master; } dev->iseg_base = pci_resource_start(dev->pdev, 0); dev->iseg = ioremap(dev->iseg_base, sizeof(*dev->iseg)); if (!dev->iseg) { err = -ENOMEM; mlx5_core_err(dev, "Failed mapping initialization segment, aborting\n"); goto err_clr_master; } return 0; err_clr_master: release_bar(dev->pdev); err_disable: mlx5_pci_disable_device(dev); err_dbg: return err; } static void mlx5_pci_close(struct mlx5_core_dev *dev, struct mlx5_priv *priv) { #ifdef PCI_IOV if (MLX5_CAP_GEN(dev, eswitch_flow_table)) pci_iov_detach(dev->pdev->dev.bsddev); #endif iounmap(dev->iseg); release_bar(dev->pdev); mlx5_pci_disable_device(dev); } static int mlx5_init_once(struct mlx5_core_dev *dev, struct mlx5_priv *priv) { int err; err = mlx5_vsc_find_cap(dev); if (err) mlx5_core_warn(dev, "Unable to find vendor specific capabilities\n"); err = mlx5_query_hca_caps(dev); if (err) { mlx5_core_err(dev, "query hca failed\n"); goto out; } err = mlx5_query_board_id(dev); if (err) { mlx5_core_err(dev, "query board id failed\n"); goto out; } err = mlx5_eq_init(dev); if (err) { mlx5_core_err(dev, "failed to initialize eq\n"); goto out; } MLX5_INIT_DOORBELL_LOCK(&priv->cq_uar_lock); err = mlx5_init_cq_table(dev); if (err) { mlx5_core_err(dev, "failed to initialize cq table\n"); goto err_eq_cleanup; } mlx5_init_qp_table(dev); mlx5_init_srq_table(dev); mlx5_init_mr_table(dev); mlx5_init_reserved_gids(dev); mlx5_fpga_init(dev); #ifdef RATELIMIT err = mlx5_init_rl_table(dev); if (err) { mlx5_core_err(dev, "Failed to init rate limiting\n"); goto err_tables_cleanup; } #endif return 0; #ifdef RATELIMIT err_tables_cleanup: mlx5_cleanup_mr_table(dev); mlx5_cleanup_srq_table(dev); mlx5_cleanup_qp_table(dev); mlx5_cleanup_cq_table(dev); #endif err_eq_cleanup: mlx5_eq_cleanup(dev); out: return err; } static void mlx5_cleanup_once(struct mlx5_core_dev *dev) { #ifdef RATELIMIT mlx5_cleanup_rl_table(dev); #endif mlx5_fpga_cleanup(dev); mlx5_cleanup_reserved_gids(dev); mlx5_cleanup_mr_table(dev); mlx5_cleanup_srq_table(dev); mlx5_cleanup_qp_table(dev); mlx5_cleanup_cq_table(dev); mlx5_eq_cleanup(dev); } static int mlx5_load_one(struct mlx5_core_dev *dev, struct mlx5_priv *priv, bool boot) { int err; mutex_lock(&dev->intf_state_mutex); if (test_bit(MLX5_INTERFACE_STATE_UP, &dev->intf_state)) { mlx5_core_warn(dev, "interface is up, NOP\n"); goto out; } mlx5_core_dbg(dev, "firmware version: %d.%d.%d\n", fw_rev_maj(dev), fw_rev_min(dev), fw_rev_sub(dev)); /* * On load removing any previous indication of internal error, * device is up */ dev->state = MLX5_DEVICE_STATE_UP; /* wait for firmware to accept initialization segments configurations */ err = wait_fw_init(dev, FW_PRE_INIT_TIMEOUT_MILI, FW_INIT_WARN_MESSAGE_INTERVAL); if (err) { dev_err(&dev->pdev->dev, "Firmware over %d MS in pre-initializing state, aborting\n", FW_PRE_INIT_TIMEOUT_MILI); goto out_err; } err = mlx5_cmd_init(dev); if (err) { mlx5_core_err(dev, "Failed initializing command interface, aborting\n"); goto out_err; } err = wait_fw_init(dev, FW_INIT_TIMEOUT_MILI, 0); if (err) { mlx5_core_err(dev, "Firmware over %d MS in initializing state, aborting\n", FW_INIT_TIMEOUT_MILI); goto err_cmd_cleanup; } err = mlx5_core_enable_hca(dev, 0); if (err) { mlx5_core_err(dev, "enable hca failed\n"); goto err_cmd_cleanup; } err = mlx5_core_set_issi(dev); if (err) { mlx5_core_err(dev, "failed to set issi\n"); goto err_disable_hca; } err = mlx5_pagealloc_start(dev); if (err) { mlx5_core_err(dev, "mlx5_pagealloc_start failed\n"); goto err_disable_hca; } err = mlx5_satisfy_startup_pages(dev, 1); if (err) { mlx5_core_err(dev, "failed to allocate boot pages\n"); goto err_pagealloc_stop; } err = set_hca_ctrl(dev); if (err) { mlx5_core_err(dev, "set_hca_ctrl failed\n"); goto reclaim_boot_pages; } err = handle_hca_cap(dev); if (err) { mlx5_core_err(dev, "handle_hca_cap failed\n"); goto reclaim_boot_pages; } err = handle_hca_cap_atomic(dev); if (err) { mlx5_core_err(dev, "handle_hca_cap_atomic failed\n"); goto reclaim_boot_pages; } err = mlx5_satisfy_startup_pages(dev, 0); if (err) { mlx5_core_err(dev, "failed to allocate init pages\n"); goto reclaim_boot_pages; } err = mlx5_cmd_init_hca(dev); if (err) { mlx5_core_err(dev, "init hca failed\n"); goto reclaim_boot_pages; } mlx5_start_health_poll(dev); if (boot && (err = mlx5_init_once(dev, priv))) { mlx5_core_err(dev, "sw objs init failed\n"); goto err_stop_poll; } dev->priv.uar = mlx5_get_uars_page(dev); if (IS_ERR(dev->priv.uar)) { mlx5_core_err(dev, "Failed allocating uar, aborting\n"); err = PTR_ERR(dev->priv.uar); goto err_cleanup_once; } err = mlx5_enable_msix(dev); if (err) { mlx5_core_err(dev, "enable msix failed\n"); goto err_cleanup_uar; } err = mlx5_start_eqs(dev); if (err) { mlx5_core_err(dev, "Failed to start pages and async EQs\n"); goto err_disable_msix; } err = alloc_comp_eqs(dev); if (err) { mlx5_core_err(dev, "Failed to alloc completion EQs\n"); goto err_stop_eqs; } err = mlx5_init_fs(dev); if (err) { mlx5_core_err(dev, "flow steering init %d\n", err); goto err_free_comp_eqs; } err = mlx5_mpfs_init(dev); if (err) { mlx5_core_err(dev, "mpfs init failed %d\n", err); goto err_fs; } err = mlx5_fpga_device_start(dev); if (err) { mlx5_core_err(dev, "fpga device start failed %d\n", err); goto err_mpfs; } err = mlx5_register_device(dev); if (err) { mlx5_core_err(dev, "mlx5_register_device failed %d\n", err); goto err_fpga; } set_bit(MLX5_INTERFACE_STATE_UP, &dev->intf_state); out: mutex_unlock(&dev->intf_state_mutex); return 0; err_fpga: mlx5_fpga_device_stop(dev); err_mpfs: mlx5_mpfs_destroy(dev); err_fs: mlx5_cleanup_fs(dev); err_free_comp_eqs: free_comp_eqs(dev); err_stop_eqs: mlx5_stop_eqs(dev); err_disable_msix: mlx5_disable_msix(dev); err_cleanup_uar: mlx5_put_uars_page(dev, dev->priv.uar); err_cleanup_once: if (boot) mlx5_cleanup_once(dev); err_stop_poll: mlx5_stop_health_poll(dev, boot); if (mlx5_cmd_teardown_hca(dev)) { mlx5_core_err(dev, "tear_down_hca failed, skip cleanup\n"); goto out_err; } reclaim_boot_pages: mlx5_reclaim_startup_pages(dev); err_pagealloc_stop: mlx5_pagealloc_stop(dev); err_disable_hca: mlx5_core_disable_hca(dev); err_cmd_cleanup: mlx5_cmd_cleanup(dev); out_err: dev->state = MLX5_DEVICE_STATE_INTERNAL_ERROR; mutex_unlock(&dev->intf_state_mutex); return err; } static int mlx5_unload_one(struct mlx5_core_dev *dev, struct mlx5_priv *priv, bool cleanup) { int err = 0; if (cleanup) mlx5_drain_health_recovery(dev); mutex_lock(&dev->intf_state_mutex); if (!test_bit(MLX5_INTERFACE_STATE_UP, &dev->intf_state)) { mlx5_core_warn(dev, "%s: interface is down, NOP\n", __func__); if (cleanup) mlx5_cleanup_once(dev); goto out; } mlx5_unregister_device(dev); mlx5_eswitch_cleanup(dev->priv.eswitch); mlx5_fpga_device_stop(dev); mlx5_mpfs_destroy(dev); mlx5_cleanup_fs(dev); mlx5_wait_for_reclaim_vfs_pages(dev); free_comp_eqs(dev); mlx5_stop_eqs(dev); mlx5_disable_msix(dev); mlx5_put_uars_page(dev, dev->priv.uar); if (cleanup) mlx5_cleanup_once(dev); mlx5_stop_health_poll(dev, cleanup); err = mlx5_cmd_teardown_hca(dev); if (err) { mlx5_core_err(dev, "tear_down_hca failed, skip cleanup\n"); goto out; } mlx5_pagealloc_stop(dev); mlx5_reclaim_startup_pages(dev); mlx5_core_disable_hca(dev); mlx5_cmd_cleanup(dev); out: clear_bit(MLX5_INTERFACE_STATE_UP, &dev->intf_state); mutex_unlock(&dev->intf_state_mutex); return err; } void mlx5_core_event(struct mlx5_core_dev *dev, enum mlx5_dev_event event, unsigned long param) { struct mlx5_priv *priv = &dev->priv; struct mlx5_device_context *dev_ctx; unsigned long flags; spin_lock_irqsave(&priv->ctx_lock, flags); list_for_each_entry(dev_ctx, &priv->ctx_list, list) if (dev_ctx->intf->event) dev_ctx->intf->event(dev, dev_ctx->context, event, param); spin_unlock_irqrestore(&priv->ctx_lock, flags); } struct mlx5_core_event_handler { void (*event)(struct mlx5_core_dev *dev, enum mlx5_dev_event event, void *data); }; #define MLX5_STATS_DESC(a, b, c, d, e, ...) d, e, #define MLX5_PORT_MODULE_ERROR_STATS(m) \ m(+1, u64, power_budget_exceeded, "power_budget", "Module Power Budget Exceeded") \ m(+1, u64, long_range, "long_range", "Module Long Range for non MLNX cable/module") \ m(+1, u64, bus_stuck, "bus_stuck", "Module Bus stuck(I2C or data shorted)") \ m(+1, u64, no_eeprom, "no_eeprom", "No EEPROM/retry timeout") \ m(+1, u64, enforce_part_number, "enforce_part_number", "Module Enforce part number list") \ m(+1, u64, unknown_id, "unknown_id", "Module Unknown identifier") \ m(+1, u64, high_temp, "high_temp", "Module High Temperature") \ m(+1, u64, cable_shorted, "cable_shorted", "Module Cable is shorted") \ m(+1, u64, pmd_type_not_enabled, "pmd_type_not_enabled", "PMD type is not enabled") \ m(+1, u64, laster_tec_failure, "laster_tec_failure", "Laster TEC failure") \ m(+1, u64, high_current, "high_current", "High current") \ m(+1, u64, high_voltage, "high_voltage", "High voltage") \ m(+1, u64, pcie_sys_power_slot_exceeded, "pcie_sys_power_slot_exceeded", "PCIe system power slot Exceeded") \ m(+1, u64, high_power, "high_power", "High power") \ m(+1, u64, module_state_machine_fault, "module_state_machine_fault", "Module State Machine fault") static const char *mlx5_pme_err_desc[] = { MLX5_PORT_MODULE_ERROR_STATS(MLX5_STATS_DESC) }; static int init_one(struct pci_dev *pdev, const struct pci_device_id *id) { struct mlx5_core_dev *dev; struct mlx5_priv *priv; device_t bsddev = pdev->dev.bsddev; #ifdef PCI_IOV nvlist_t *pf_schema, *vf_schema; int num_vfs, sriov_pos; #endif int i,err; + int numa_node; struct sysctl_oid *pme_sysctl_node; struct sysctl_oid *pme_err_sysctl_node; struct sysctl_oid *cap_sysctl_node; struct sysctl_oid *current_cap_sysctl_node; struct sysctl_oid *max_cap_sysctl_node; - dev = kzalloc(sizeof(*dev), GFP_KERNEL); + numa_node = dev_to_node(&pdev->dev); + + dev = kzalloc_node(sizeof(*dev), GFP_KERNEL, numa_node); + priv = &dev->priv; + priv->numa_node = numa_node; + if (id) priv->pci_dev_data = id->driver_data; if (mlx5_prof_sel < 0 || mlx5_prof_sel >= ARRAY_SIZE(profiles)) { device_printf(bsddev, "WARN: selected profile out of range, selecting default (%d)\n", MLX5_DEFAULT_PROF); mlx5_prof_sel = MLX5_DEFAULT_PROF; } dev->profile = &profiles[mlx5_prof_sel]; dev->pdev = pdev; dev->event = mlx5_core_event; /* Set desc */ device_set_desc(bsddev, mlx5_version); sysctl_ctx_init(&dev->sysctl_ctx); SYSCTL_ADD_INT(&dev->sysctl_ctx, SYSCTL_CHILDREN(device_get_sysctl_tree(bsddev)), OID_AUTO, "msix_eqvec", CTLFLAG_RDTUN, &dev->msix_eqvec, 0, "Maximum number of MSIX event queue vectors, if set"); SYSCTL_ADD_INT(&dev->sysctl_ctx, SYSCTL_CHILDREN(device_get_sysctl_tree(bsddev)), OID_AUTO, "power_status", CTLFLAG_RD, &dev->pwr_status, 0, "0:Invalid 1:Sufficient 2:Insufficient"); SYSCTL_ADD_INT(&dev->sysctl_ctx, SYSCTL_CHILDREN(device_get_sysctl_tree(bsddev)), OID_AUTO, "power_value", CTLFLAG_RD, &dev->pwr_value, 0, "Current power value in Watts"); pme_sysctl_node = SYSCTL_ADD_NODE(&dev->sysctl_ctx, SYSCTL_CHILDREN(device_get_sysctl_tree(bsddev)), OID_AUTO, "pme_stats", CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, "Port module event statistics"); if (pme_sysctl_node == NULL) { err = -ENOMEM; goto clean_sysctl_ctx; } pme_err_sysctl_node = SYSCTL_ADD_NODE(&dev->sysctl_ctx, SYSCTL_CHILDREN(pme_sysctl_node), OID_AUTO, "errors", CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, "Port module event error statistics"); if (pme_err_sysctl_node == NULL) { err = -ENOMEM; goto clean_sysctl_ctx; } SYSCTL_ADD_U64(&dev->sysctl_ctx, SYSCTL_CHILDREN(pme_sysctl_node), OID_AUTO, "module_plug", CTLFLAG_RD | CTLFLAG_MPSAFE, &dev->priv.pme_stats.status_counters[MLX5_MODULE_STATUS_PLUGGED_ENABLED], 0, "Number of time module plugged"); SYSCTL_ADD_U64(&dev->sysctl_ctx, SYSCTL_CHILDREN(pme_sysctl_node), OID_AUTO, "module_unplug", CTLFLAG_RD | CTLFLAG_MPSAFE, &dev->priv.pme_stats.status_counters[MLX5_MODULE_STATUS_UNPLUGGED], 0, "Number of time module unplugged"); for (i = 0 ; i < MLX5_MODULE_EVENT_ERROR_NUM; i++) { SYSCTL_ADD_U64(&dev->sysctl_ctx, SYSCTL_CHILDREN(pme_err_sysctl_node), OID_AUTO, mlx5_pme_err_desc[2 * i], CTLFLAG_RD | CTLFLAG_MPSAFE, &dev->priv.pme_stats.error_counters[i], 0, mlx5_pme_err_desc[2 * i + 1]); } cap_sysctl_node = SYSCTL_ADD_NODE(&dev->sysctl_ctx, SYSCTL_CHILDREN(device_get_sysctl_tree(bsddev)), OID_AUTO, "caps", CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, "hardware capabilities raw bitstrings"); if (cap_sysctl_node == NULL) { err = -ENOMEM; goto clean_sysctl_ctx; } current_cap_sysctl_node = SYSCTL_ADD_NODE(&dev->sysctl_ctx, SYSCTL_CHILDREN(cap_sysctl_node), OID_AUTO, "current", CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, ""); if (current_cap_sysctl_node == NULL) { err = -ENOMEM; goto clean_sysctl_ctx; } max_cap_sysctl_node = SYSCTL_ADD_NODE(&dev->sysctl_ctx, SYSCTL_CHILDREN(cap_sysctl_node), OID_AUTO, "max", CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, ""); if (max_cap_sysctl_node == NULL) { err = -ENOMEM; goto clean_sysctl_ctx; } SYSCTL_ADD_OPAQUE(&dev->sysctl_ctx, SYSCTL_CHILDREN(current_cap_sysctl_node), OID_AUTO, "general", CTLFLAG_RD | CTLFLAG_MPSAFE, &dev->hca_caps_cur[MLX5_CAP_GENERAL], MLX5_UN_SZ_DW(hca_cap_union) * sizeof(u32), "IU", ""); SYSCTL_ADD_OPAQUE(&dev->sysctl_ctx, SYSCTL_CHILDREN(max_cap_sysctl_node), OID_AUTO, "general", CTLFLAG_RD | CTLFLAG_MPSAFE, &dev->hca_caps_max[MLX5_CAP_GENERAL], MLX5_UN_SZ_DW(hca_cap_union) * sizeof(u32), "IU", ""); SYSCTL_ADD_OPAQUE(&dev->sysctl_ctx, SYSCTL_CHILDREN(current_cap_sysctl_node), OID_AUTO, "ether", CTLFLAG_RD | CTLFLAG_MPSAFE, &dev->hca_caps_cur[MLX5_CAP_ETHERNET_OFFLOADS], MLX5_UN_SZ_DW(hca_cap_union) * sizeof(u32), "IU", ""); SYSCTL_ADD_OPAQUE(&dev->sysctl_ctx, SYSCTL_CHILDREN(max_cap_sysctl_node), OID_AUTO, "ether", CTLFLAG_RD | CTLFLAG_MPSAFE, &dev->hca_caps_max[MLX5_CAP_ETHERNET_OFFLOADS], MLX5_UN_SZ_DW(hca_cap_union) * sizeof(u32), "IU", ""); SYSCTL_ADD_OPAQUE(&dev->sysctl_ctx, SYSCTL_CHILDREN(current_cap_sysctl_node), OID_AUTO, "odp", CTLFLAG_RD | CTLFLAG_MPSAFE, &dev->hca_caps_cur[MLX5_CAP_ODP], MLX5_UN_SZ_DW(hca_cap_union) * sizeof(u32), "IU", ""); SYSCTL_ADD_OPAQUE(&dev->sysctl_ctx, SYSCTL_CHILDREN(max_cap_sysctl_node), OID_AUTO, "odp", CTLFLAG_RD | CTLFLAG_MPSAFE, &dev->hca_caps_max[MLX5_CAP_ODP], MLX5_UN_SZ_DW(hca_cap_union) * sizeof(u32), "IU", ""); SYSCTL_ADD_OPAQUE(&dev->sysctl_ctx, SYSCTL_CHILDREN(current_cap_sysctl_node), OID_AUTO, "atomic", CTLFLAG_RD | CTLFLAG_MPSAFE, &dev->hca_caps_cur[MLX5_CAP_ATOMIC], MLX5_UN_SZ_DW(hca_cap_union) * sizeof(u32), "IU", ""); SYSCTL_ADD_OPAQUE(&dev->sysctl_ctx, SYSCTL_CHILDREN(max_cap_sysctl_node), OID_AUTO, "atomic", CTLFLAG_RD | CTLFLAG_MPSAFE, &dev->hca_caps_max[MLX5_CAP_ATOMIC], MLX5_UN_SZ_DW(hca_cap_union) * sizeof(u32), "IU", ""); SYSCTL_ADD_OPAQUE(&dev->sysctl_ctx, SYSCTL_CHILDREN(current_cap_sysctl_node), OID_AUTO, "roce", CTLFLAG_RD | CTLFLAG_MPSAFE, &dev->hca_caps_cur[MLX5_CAP_ROCE], MLX5_UN_SZ_DW(hca_cap_union) * sizeof(u32), "IU", ""); SYSCTL_ADD_OPAQUE(&dev->sysctl_ctx, SYSCTL_CHILDREN(max_cap_sysctl_node), OID_AUTO, "roce", CTLFLAG_RD | CTLFLAG_MPSAFE, &dev->hca_caps_max[MLX5_CAP_ROCE], MLX5_UN_SZ_DW(hca_cap_union) * sizeof(u32), "IU", ""); SYSCTL_ADD_OPAQUE(&dev->sysctl_ctx, SYSCTL_CHILDREN(current_cap_sysctl_node), OID_AUTO, "ipoib", CTLFLAG_RD | CTLFLAG_MPSAFE, &dev->hca_caps_cur[MLX5_CAP_IPOIB_OFFLOADS], MLX5_UN_SZ_DW(hca_cap_union) * sizeof(u32), "IU", ""); SYSCTL_ADD_OPAQUE(&dev->sysctl_ctx, SYSCTL_CHILDREN(max_cap_sysctl_node), OID_AUTO, "ipoib", CTLFLAG_RD | CTLFLAG_MPSAFE, &dev->hca_caps_max[MLX5_CAP_IPOIB_OFFLOADS], MLX5_UN_SZ_DW(hca_cap_union) * sizeof(u32), "IU", ""); SYSCTL_ADD_OPAQUE(&dev->sysctl_ctx, SYSCTL_CHILDREN(current_cap_sysctl_node), OID_AUTO, "eoib", CTLFLAG_RD | CTLFLAG_MPSAFE, &dev->hca_caps_cur[MLX5_CAP_EOIB_OFFLOADS], MLX5_UN_SZ_DW(hca_cap_union) * sizeof(u32), "IU", ""); SYSCTL_ADD_OPAQUE(&dev->sysctl_ctx, SYSCTL_CHILDREN(max_cap_sysctl_node), OID_AUTO, "eoib", CTLFLAG_RD | CTLFLAG_MPSAFE, &dev->hca_caps_max[MLX5_CAP_EOIB_OFFLOADS], MLX5_UN_SZ_DW(hca_cap_union) * sizeof(u32), "IU", ""); SYSCTL_ADD_OPAQUE(&dev->sysctl_ctx, SYSCTL_CHILDREN(current_cap_sysctl_node), OID_AUTO, "flow_table", CTLFLAG_RD | CTLFLAG_MPSAFE, &dev->hca_caps_cur[MLX5_CAP_FLOW_TABLE], MLX5_UN_SZ_DW(hca_cap_union) * sizeof(u32), "IU", ""); SYSCTL_ADD_OPAQUE(&dev->sysctl_ctx, SYSCTL_CHILDREN(max_cap_sysctl_node), OID_AUTO, "flow_table", CTLFLAG_RD | CTLFLAG_MPSAFE, &dev->hca_caps_max[MLX5_CAP_FLOW_TABLE], MLX5_UN_SZ_DW(hca_cap_union) * sizeof(u32), "IU", ""); SYSCTL_ADD_OPAQUE(&dev->sysctl_ctx, SYSCTL_CHILDREN(current_cap_sysctl_node), OID_AUTO, "eswitch_flow_table", CTLFLAG_RD | CTLFLAG_MPSAFE, &dev->hca_caps_cur[MLX5_CAP_ESWITCH_FLOW_TABLE], MLX5_UN_SZ_DW(hca_cap_union) * sizeof(u32), "IU", ""); SYSCTL_ADD_OPAQUE(&dev->sysctl_ctx, SYSCTL_CHILDREN(max_cap_sysctl_node), OID_AUTO, "eswitch_flow_table", CTLFLAG_RD | CTLFLAG_MPSAFE, &dev->hca_caps_max[MLX5_CAP_ESWITCH_FLOW_TABLE], MLX5_UN_SZ_DW(hca_cap_union) * sizeof(u32), "IU", ""); SYSCTL_ADD_OPAQUE(&dev->sysctl_ctx, SYSCTL_CHILDREN(current_cap_sysctl_node), OID_AUTO, "eswitch", CTLFLAG_RD | CTLFLAG_MPSAFE, &dev->hca_caps_cur[MLX5_CAP_ESWITCH], MLX5_UN_SZ_DW(hca_cap_union) * sizeof(u32), "IU", ""); SYSCTL_ADD_OPAQUE(&dev->sysctl_ctx, SYSCTL_CHILDREN(max_cap_sysctl_node), OID_AUTO, "eswitch", CTLFLAG_RD | CTLFLAG_MPSAFE, &dev->hca_caps_max[MLX5_CAP_ESWITCH], MLX5_UN_SZ_DW(hca_cap_union) * sizeof(u32), "IU", ""); SYSCTL_ADD_OPAQUE(&dev->sysctl_ctx, SYSCTL_CHILDREN(current_cap_sysctl_node), OID_AUTO, "snapshot", CTLFLAG_RD | CTLFLAG_MPSAFE, &dev->hca_caps_cur[MLX5_CAP_SNAPSHOT], MLX5_UN_SZ_DW(hca_cap_union) * sizeof(u32), "IU", ""); SYSCTL_ADD_OPAQUE(&dev->sysctl_ctx, SYSCTL_CHILDREN(max_cap_sysctl_node), OID_AUTO, "snapshot", CTLFLAG_RD | CTLFLAG_MPSAFE, &dev->hca_caps_max[MLX5_CAP_SNAPSHOT], MLX5_UN_SZ_DW(hca_cap_union) * sizeof(u32), "IU", ""); SYSCTL_ADD_OPAQUE(&dev->sysctl_ctx, SYSCTL_CHILDREN(current_cap_sysctl_node), OID_AUTO, "vector_calc", CTLFLAG_RD | CTLFLAG_MPSAFE, &dev->hca_caps_cur[MLX5_CAP_VECTOR_CALC], MLX5_UN_SZ_DW(hca_cap_union) * sizeof(u32), "IU", ""); SYSCTL_ADD_OPAQUE(&dev->sysctl_ctx, SYSCTL_CHILDREN(max_cap_sysctl_node), OID_AUTO, "vector_calc", CTLFLAG_RD | CTLFLAG_MPSAFE, &dev->hca_caps_max[MLX5_CAP_VECTOR_CALC], MLX5_UN_SZ_DW(hca_cap_union) * sizeof(u32), "IU", ""); SYSCTL_ADD_OPAQUE(&dev->sysctl_ctx, SYSCTL_CHILDREN(current_cap_sysctl_node), OID_AUTO, "qos", CTLFLAG_RD | CTLFLAG_MPSAFE, &dev->hca_caps_cur[MLX5_CAP_QOS], MLX5_UN_SZ_DW(hca_cap_union) * sizeof(u32), "IU", ""); SYSCTL_ADD_OPAQUE(&dev->sysctl_ctx, SYSCTL_CHILDREN(max_cap_sysctl_node), OID_AUTO, "qos", CTLFLAG_RD | CTLFLAG_MPSAFE, &dev->hca_caps_max[MLX5_CAP_QOS], MLX5_UN_SZ_DW(hca_cap_union) * sizeof(u32), "IU", ""); SYSCTL_ADD_OPAQUE(&dev->sysctl_ctx, SYSCTL_CHILDREN(current_cap_sysctl_node), OID_AUTO, "debug", CTLFLAG_RD | CTLFLAG_MPSAFE, &dev->hca_caps_cur[MLX5_CAP_DEBUG], MLX5_UN_SZ_DW(hca_cap_union) * sizeof(u32), "IU", ""); SYSCTL_ADD_OPAQUE(&dev->sysctl_ctx, SYSCTL_CHILDREN(max_cap_sysctl_node), OID_AUTO, "debug", CTLFLAG_RD | CTLFLAG_MPSAFE, &dev->hca_caps_max[MLX5_CAP_DEBUG], MLX5_UN_SZ_DW(hca_cap_union) * sizeof(u32), "IU", ""); SYSCTL_ADD_OPAQUE(&dev->sysctl_ctx, SYSCTL_CHILDREN(cap_sysctl_node), OID_AUTO, "pcam", CTLFLAG_RD | CTLFLAG_MPSAFE, &dev->caps.pcam, sizeof(dev->caps.pcam), "IU", ""); SYSCTL_ADD_OPAQUE(&dev->sysctl_ctx, SYSCTL_CHILDREN(cap_sysctl_node), OID_AUTO, "mcam", CTLFLAG_RD | CTLFLAG_MPSAFE, &dev->caps.mcam, sizeof(dev->caps.mcam), "IU", ""); SYSCTL_ADD_OPAQUE(&dev->sysctl_ctx, SYSCTL_CHILDREN(cap_sysctl_node), OID_AUTO, "qcam", CTLFLAG_RD | CTLFLAG_MPSAFE, &dev->caps.qcam, sizeof(dev->caps.qcam), "IU", ""); SYSCTL_ADD_OPAQUE(&dev->sysctl_ctx, SYSCTL_CHILDREN(cap_sysctl_node), OID_AUTO, "fpga", CTLFLAG_RD | CTLFLAG_MPSAFE, &dev->caps.fpga, sizeof(dev->caps.fpga), "IU", ""); INIT_LIST_HEAD(&priv->ctx_list); spin_lock_init(&priv->ctx_lock); mutex_init(&dev->pci_status_mutex); mutex_init(&dev->intf_state_mutex); mutex_init(&priv->bfregs.reg_head.lock); mutex_init(&priv->bfregs.wc_head.lock); INIT_LIST_HEAD(&priv->bfregs.reg_head.list); INIT_LIST_HEAD(&priv->bfregs.wc_head.list); mtx_init(&dev->dump_lock, "mlx5dmp", NULL, MTX_DEF | MTX_NEW); err = mlx5_pci_init(dev, priv); if (err) { mlx5_core_err(dev, "mlx5_pci_init failed %d\n", err); goto clean_dev; } err = mlx5_health_init(dev); if (err) { mlx5_core_err(dev, "mlx5_health_init failed %d\n", err); goto close_pci; } mlx5_pagealloc_init(dev); err = mlx5_load_one(dev, priv, true); if (err) { mlx5_core_err(dev, "mlx5_load_one failed %d\n", err); goto clean_health; } mlx5_fwdump_prep(dev); mlx5_firmware_update(dev); #ifdef PCI_IOV if (MLX5_CAP_GEN(dev, vport_group_manager)) { if (pci_find_extcap(bsddev, PCIZ_SRIOV, &sriov_pos) == 0) { num_vfs = pci_read_config(bsddev, sriov_pos + PCIR_SRIOV_TOTAL_VFS, 2); } else { mlx5_core_info(dev, "cannot find SR-IOV PCIe cap\n"); num_vfs = 0; } err = mlx5_eswitch_init(dev, 1 + num_vfs); if (err == 0) { pf_schema = pci_iov_schema_alloc_node(); vf_schema = pci_iov_schema_alloc_node(); pci_iov_schema_add_unicast_mac(vf_schema, iov_mac_addr_name, 0, NULL); pci_iov_schema_add_uint64(vf_schema, iov_node_guid_name, 0, 0); pci_iov_schema_add_uint64(vf_schema, iov_port_guid_name, 0, 0); err = pci_iov_attach(bsddev, pf_schema, vf_schema); if (err != 0) { device_printf(bsddev, "Failed to initialize SR-IOV support, error %d\n", err); } } else { mlx5_core_err(dev, "eswitch init failed, error %d\n", err); } } #endif pci_save_state(pdev); return 0; clean_health: mlx5_pagealloc_cleanup(dev); mlx5_health_cleanup(dev); close_pci: mlx5_pci_close(dev, priv); clean_dev: mtx_destroy(&dev->dump_lock); clean_sysctl_ctx: sysctl_ctx_free(&dev->sysctl_ctx); kfree(dev); return err; } static void remove_one(struct pci_dev *pdev) { struct mlx5_core_dev *dev = pci_get_drvdata(pdev); struct mlx5_priv *priv = &dev->priv; #ifdef PCI_IOV pci_iov_detach(pdev->dev.bsddev); mlx5_eswitch_disable_sriov(priv->eswitch); #endif if (mlx5_unload_one(dev, priv, true)) { mlx5_core_err(dev, "mlx5_unload_one() failed, leaked %lld bytes\n", (long long)(dev->priv.fw_pages * MLX5_ADAPTER_PAGE_SIZE)); } mlx5_pagealloc_cleanup(dev); mlx5_health_cleanup(dev); mlx5_fwdump_clean(dev); mlx5_pci_close(dev, priv); mtx_destroy(&dev->dump_lock); pci_set_drvdata(pdev, NULL); sysctl_ctx_free(&dev->sysctl_ctx); kfree(dev); } static pci_ers_result_t mlx5_pci_err_detected(struct pci_dev *pdev, pci_channel_state_t state) { struct mlx5_core_dev *dev = pci_get_drvdata(pdev); struct mlx5_priv *priv = &dev->priv; mlx5_core_info(dev, "%s was called\n", __func__); mlx5_enter_error_state(dev, false); mlx5_unload_one(dev, priv, false); if (state) { 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; } static pci_ers_result_t mlx5_pci_slot_reset(struct pci_dev *pdev) { struct mlx5_core_dev *dev = pci_get_drvdata(pdev); int err = 0; mlx5_core_info(dev,"%s was called\n", __func__); err = mlx5_pci_enable_device(dev); if (err) { mlx5_core_err(dev, "mlx5_pci_enable_device failed with error code: %d\n" ,err); return PCI_ERS_RESULT_DISCONNECT; } pci_set_master(pdev); pci_set_powerstate(pdev->dev.bsddev, PCI_POWERSTATE_D0); pci_restore_state(pdev); pci_save_state(pdev); return err ? PCI_ERS_RESULT_DISCONNECT : PCI_ERS_RESULT_RECOVERED; } /* wait for the device to show vital signs. For now we check * that we can read the device ID and that the health buffer * shows a non zero value which is different than 0xffffffff */ static void wait_vital(struct pci_dev *pdev) { struct mlx5_core_dev *dev = pci_get_drvdata(pdev); struct mlx5_core_health *health = &dev->priv.health; const int niter = 100; u32 count; u16 did; int i; /* Wait for firmware to be ready after reset */ msleep(1000); for (i = 0; i < niter; i++) { if (pci_read_config_word(pdev, 2, &did)) { mlx5_core_warn(dev, "failed reading config word\n"); break; } if (did == pdev->device) { mlx5_core_info(dev, "device ID correctly read after %d iterations\n", i); break; } msleep(50); } if (i == niter) mlx5_core_warn(dev, "could not read device ID\n"); for (i = 0; i < niter; i++) { count = ioread32be(health->health_counter); if (count && count != 0xffffffff) { mlx5_core_info(dev, "Counter value 0x%x after %d iterations\n", count, i); break; } msleep(50); } if (i == niter) mlx5_core_warn(dev, "could not read device ID\n"); } static void mlx5_pci_resume(struct pci_dev *pdev) { struct mlx5_core_dev *dev = pci_get_drvdata(pdev); struct mlx5_priv *priv = &dev->priv; int err; mlx5_core_info(dev,"%s was called\n", __func__); wait_vital(pdev); err = mlx5_load_one(dev, priv, false); if (err) mlx5_core_err(dev, "mlx5_load_one failed with error code: %d\n" ,err); else mlx5_core_info(dev,"device recovered\n"); } static const struct pci_error_handlers mlx5_err_handler = { .error_detected = mlx5_pci_err_detected, .slot_reset = mlx5_pci_slot_reset, .resume = mlx5_pci_resume }; #ifdef PCI_IOV static int mlx5_iov_init(device_t dev, uint16_t num_vfs, const nvlist_t *pf_config) { struct pci_dev *pdev; struct mlx5_core_dev *core_dev; struct mlx5_priv *priv; int err; pdev = device_get_softc(dev); core_dev = pci_get_drvdata(pdev); priv = &core_dev->priv; if (priv->eswitch == NULL) return (ENXIO); if (priv->eswitch->total_vports < num_vfs + 1) num_vfs = priv->eswitch->total_vports - 1; err = mlx5_eswitch_enable_sriov(priv->eswitch, num_vfs); return (-err); } static void mlx5_iov_uninit(device_t dev) { struct pci_dev *pdev; struct mlx5_core_dev *core_dev; struct mlx5_priv *priv; pdev = device_get_softc(dev); core_dev = pci_get_drvdata(pdev); priv = &core_dev->priv; mlx5_eswitch_disable_sriov(priv->eswitch); } static int mlx5_iov_add_vf(device_t dev, uint16_t vfnum, const nvlist_t *vf_config) { struct pci_dev *pdev; struct mlx5_core_dev *core_dev; struct mlx5_priv *priv; const void *mac; size_t mac_size; uint64_t node_guid, port_guid; int error; pdev = device_get_softc(dev); core_dev = pci_get_drvdata(pdev); priv = &core_dev->priv; if (vfnum + 1 >= priv->eswitch->total_vports) return (ENXIO); if (nvlist_exists_binary(vf_config, iov_mac_addr_name)) { mac = nvlist_get_binary(vf_config, iov_mac_addr_name, &mac_size); error = -mlx5_eswitch_set_vport_mac(priv->eswitch, vfnum + 1, __DECONST(u8 *, mac)); if (error != 0) { mlx5_core_err(core_dev, "setting MAC for VF %d failed, error %d\n", vfnum + 1, error); } } if (nvlist_exists_number(vf_config, iov_node_guid_name)) { node_guid = nvlist_get_number(vf_config, iov_node_guid_name); error = -mlx5_modify_nic_vport_node_guid(core_dev, vfnum + 1, node_guid); if (error != 0) { mlx5_core_err(core_dev, "modifying node GUID for VF %d failed, error %d\n", vfnum + 1, error); } } if (nvlist_exists_number(vf_config, iov_port_guid_name)) { port_guid = nvlist_get_number(vf_config, iov_port_guid_name); error = -mlx5_modify_nic_vport_port_guid(core_dev, vfnum + 1, port_guid); if (error != 0) { mlx5_core_err(core_dev, "modifying port GUID for VF %d failed, error %d\n", vfnum + 1, error); } } error = -mlx5_eswitch_set_vport_state(priv->eswitch, vfnum + 1, VPORT_STATE_FOLLOW); if (error != 0) { mlx5_core_err(core_dev, "upping vport for VF %d failed, error %d\n", vfnum + 1, error); } error = -mlx5_core_enable_hca(core_dev, vfnum + 1); if (error != 0) { mlx5_core_err(core_dev, "enabling VF %d failed, error %d\n", vfnum + 1, error); } return (error); } #endif static int mlx5_try_fast_unload(struct mlx5_core_dev *dev) { bool fast_teardown, force_teardown; int err; if (!mlx5_fast_unload_enabled) { mlx5_core_dbg(dev, "fast unload is disabled by user\n"); return -EOPNOTSUPP; } fast_teardown = MLX5_CAP_GEN(dev, fast_teardown); force_teardown = MLX5_CAP_GEN(dev, force_teardown); mlx5_core_dbg(dev, "force teardown firmware support=%d\n", force_teardown); mlx5_core_dbg(dev, "fast teardown firmware support=%d\n", fast_teardown); if (!fast_teardown && !force_teardown) return -EOPNOTSUPP; if (dev->state == MLX5_DEVICE_STATE_INTERNAL_ERROR) { mlx5_core_dbg(dev, "Device in internal error state, giving up\n"); return -EAGAIN; } /* Panic tear down fw command will stop the PCI bus communication * with the HCA, so the health polll is no longer needed. */ mlx5_drain_health_wq(dev); mlx5_stop_health_poll(dev, false); err = mlx5_cmd_fast_teardown_hca(dev); if (!err) goto done; err = mlx5_cmd_force_teardown_hca(dev); if (!err) goto done; mlx5_core_dbg(dev, "Firmware couldn't do fast unload error: %d\n", err); mlx5_start_health_poll(dev); return err; done: mlx5_enter_error_state(dev, true); return 0; } static void mlx5_shutdown_disable_interrupts(struct mlx5_core_dev *mdev) { int nvec = mdev->priv.eq_table.num_comp_vectors + MLX5_EQ_VEC_COMP_BASE; int x; mdev->priv.disable_irqs = 1; /* wait for all IRQ handlers to finish processing */ for (x = 0; x != nvec; x++) synchronize_irq(mdev->priv.msix_arr[x].vector); } static void shutdown_one(struct pci_dev *pdev) { struct mlx5_core_dev *dev = pci_get_drvdata(pdev); struct mlx5_priv *priv = &dev->priv; int err; /* enter polling mode */ mlx5_cmd_use_polling(dev); set_bit(MLX5_INTERFACE_STATE_TEARDOWN, &dev->intf_state); /* disable all interrupts */ mlx5_shutdown_disable_interrupts(dev); err = mlx5_try_fast_unload(dev); if (err) mlx5_unload_one(dev, priv, false); mlx5_pci_disable_device(dev); } static const struct pci_device_id mlx5_core_pci_table[] = { { PCI_VDEVICE(MELLANOX, 4113) }, /* Connect-IB */ { PCI_VDEVICE(MELLANOX, 4114) }, /* Connect-IB VF */ { PCI_VDEVICE(MELLANOX, 4115) }, /* ConnectX-4 */ { PCI_VDEVICE(MELLANOX, 4116) }, /* ConnectX-4 VF */ { PCI_VDEVICE(MELLANOX, 4117) }, /* ConnectX-4LX */ { PCI_VDEVICE(MELLANOX, 4118) }, /* ConnectX-4LX VF */ { PCI_VDEVICE(MELLANOX, 4119) }, /* ConnectX-5, PCIe 3.0 */ { PCI_VDEVICE(MELLANOX, 4120) }, /* ConnectX-5 VF */ { PCI_VDEVICE(MELLANOX, 4121) }, /* ConnectX-5 Ex */ { PCI_VDEVICE(MELLANOX, 4122) }, /* ConnectX-5 Ex VF */ { PCI_VDEVICE(MELLANOX, 4123) }, /* ConnectX-6 */ { PCI_VDEVICE(MELLANOX, 4124) }, /* ConnectX-6 VF */ { PCI_VDEVICE(MELLANOX, 4125) }, /* ConnectX-6 Dx */ { PCI_VDEVICE(MELLANOX, 4126) }, /* ConnectX Family mlx5Gen Virtual Function */ { PCI_VDEVICE(MELLANOX, 4127) }, /* ConnectX-6 LX */ { PCI_VDEVICE(MELLANOX, 4128) }, { PCI_VDEVICE(MELLANOX, 4129) }, { PCI_VDEVICE(MELLANOX, 4130) }, { PCI_VDEVICE(MELLANOX, 4131) }, { PCI_VDEVICE(MELLANOX, 4132) }, { PCI_VDEVICE(MELLANOX, 4133) }, { PCI_VDEVICE(MELLANOX, 4134) }, { PCI_VDEVICE(MELLANOX, 4135) }, { PCI_VDEVICE(MELLANOX, 4136) }, { PCI_VDEVICE(MELLANOX, 4137) }, { PCI_VDEVICE(MELLANOX, 4138) }, { PCI_VDEVICE(MELLANOX, 4139) }, { PCI_VDEVICE(MELLANOX, 4140) }, { PCI_VDEVICE(MELLANOX, 4141) }, { PCI_VDEVICE(MELLANOX, 4142) }, { PCI_VDEVICE(MELLANOX, 4143) }, { PCI_VDEVICE(MELLANOX, 4144) }, { PCI_VDEVICE(MELLANOX, 0xa2d2) }, /* BlueField integrated ConnectX-5 network controller */ { PCI_VDEVICE(MELLANOX, 0xa2d3) }, /* BlueField integrated ConnectX-5 network controller VF */ { PCI_VDEVICE(MELLANOX, 0xa2d6) }, /* BlueField-2 integrated ConnectX-6 Dx network controller */ { } }; 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); } struct pci_driver mlx5_core_driver = { .name = DRIVER_NAME, .id_table = mlx5_core_pci_table, .shutdown = shutdown_one, .probe = init_one, .remove = remove_one, .err_handler = &mlx5_err_handler, #ifdef PCI_IOV .bsd_iov_init = mlx5_iov_init, .bsd_iov_uninit = mlx5_iov_uninit, .bsd_iov_add_vf = mlx5_iov_add_vf, #endif }; static int __init init(void) { int err; err = pci_register_driver(&mlx5_core_driver); if (err) goto err_debug; err = mlx5_ctl_init(); if (err) goto err_ctl; return 0; err_ctl: pci_unregister_driver(&mlx5_core_driver); err_debug: return err; } static void __exit cleanup(void) { mlx5_ctl_fini(); pci_unregister_driver(&mlx5_core_driver); } module_init_order(init, SI_ORDER_FIRST); module_exit_order(cleanup, SI_ORDER_FIRST); diff --git a/sys/dev/mlx5/mlx5_core/mlx5_wq.c b/sys/dev/mlx5/mlx5_core/mlx5_wq.c index 6cbb7c0000ae..131ed9bfce2a 100644 --- a/sys/dev/mlx5/mlx5_core/mlx5_wq.c +++ b/sys/dev/mlx5/mlx5_core/mlx5_wq.c @@ -1,187 +1,184 @@ /*- * Copyright (c) 2013-2017, Mellanox Technologies, Ltd. All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY AUTHOR AND CONTRIBUTORS `AS IS' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * $FreeBSD$ */ #include #include "wq.h" #include "mlx5_core.h" u32 mlx5_wq_cyc_get_size(struct mlx5_wq_cyc *wq) { return (u32)wq->sz_m1 + 1; } u32 mlx5_cqwq_get_size(struct mlx5_cqwq *wq) { return wq->sz_m1 + 1; } u32 mlx5_wq_ll_get_size(struct mlx5_wq_ll *wq) { return (u32)wq->sz_m1 + 1; } static u32 mlx5_wq_cyc_get_byte_size(struct mlx5_wq_cyc *wq) { return mlx5_wq_cyc_get_size(wq) << wq->log_stride; } static u32 mlx5_cqwq_get_byte_size(struct mlx5_cqwq *wq) { return mlx5_cqwq_get_size(wq) << wq->log_stride; } static u32 mlx5_wq_ll_get_byte_size(struct mlx5_wq_ll *wq) { return mlx5_wq_ll_get_size(wq) << wq->log_stride; } int mlx5_wq_cyc_create(struct mlx5_core_dev *mdev, struct mlx5_wq_param *param, void *wqc, struct mlx5_wq_cyc *wq, struct mlx5_wq_ctrl *wq_ctrl) { int max_direct = param->linear ? INT_MAX : 0; int err; wq->log_stride = MLX5_GET(wq, wqc, log_wq_stride); wq->sz_m1 = (1 << MLX5_GET(wq, wqc, log_wq_sz)) - 1; - err = mlx5_db_alloc_node(mdev, &wq_ctrl->db, param->db_numa_node); + err = mlx5_db_alloc(mdev, &wq_ctrl->db); if (err) { mlx5_core_warn(mdev, "mlx5_db_alloc() failed, %d\n", err); return err; } - err = mlx5_buf_alloc_node(mdev, mlx5_wq_cyc_get_byte_size(wq), - max_direct, &wq_ctrl->buf, - param->buf_numa_node); + err = mlx5_buf_alloc(mdev, mlx5_wq_cyc_get_byte_size(wq), + max_direct, &wq_ctrl->buf); if (err) { mlx5_core_warn(mdev, "mlx5_buf_alloc() failed, %d\n", err); goto err_db_free; } wq->buf = wq_ctrl->buf.direct.buf; wq->db = wq_ctrl->db.db; wq_ctrl->mdev = mdev; return 0; err_db_free: mlx5_db_free(mdev, &wq_ctrl->db); return err; } int mlx5_cqwq_create(struct mlx5_core_dev *mdev, struct mlx5_wq_param *param, void *cqc, struct mlx5_cqwq *wq, struct mlx5_wq_ctrl *wq_ctrl) { int max_direct = param->linear ? INT_MAX : 0; int err; wq->log_stride = 6 + MLX5_GET(cqc, cqc, cqe_sz); wq->log_sz = MLX5_GET(cqc, cqc, log_cq_size); wq->sz_m1 = (1 << wq->log_sz) - 1; - err = mlx5_db_alloc_node(mdev, &wq_ctrl->db, param->db_numa_node); + err = mlx5_db_alloc(mdev, &wq_ctrl->db); if (err) { mlx5_core_warn(mdev, "mlx5_db_alloc() failed, %d\n", err); return err; } - err = mlx5_buf_alloc_node(mdev, mlx5_cqwq_get_byte_size(wq), - max_direct, &wq_ctrl->buf, - param->buf_numa_node); + err = mlx5_buf_alloc(mdev, mlx5_cqwq_get_byte_size(wq), + max_direct, &wq_ctrl->buf); if (err) { mlx5_core_warn(mdev, "mlx5_buf_alloc() failed, %d\n", err); goto err_db_free; } wq->buf = wq_ctrl->buf.direct.buf; wq->db = wq_ctrl->db.db; wq_ctrl->mdev = mdev; return 0; err_db_free: mlx5_db_free(mdev, &wq_ctrl->db); return err; } int mlx5_wq_ll_create(struct mlx5_core_dev *mdev, struct mlx5_wq_param *param, void *wqc, struct mlx5_wq_ll *wq, struct mlx5_wq_ctrl *wq_ctrl) { struct mlx5_wqe_srq_next_seg *next_seg; int max_direct = param->linear ? INT_MAX : 0; int err; int i; wq->log_stride = MLX5_GET(wq, wqc, log_wq_stride); wq->sz_m1 = (1 << MLX5_GET(wq, wqc, log_wq_sz)) - 1; - err = mlx5_db_alloc_node(mdev, &wq_ctrl->db, param->db_numa_node); + err = mlx5_db_alloc(mdev, &wq_ctrl->db); if (err) { mlx5_core_warn(mdev, "mlx5_db_alloc() failed, %d\n", err); return err; } - err = mlx5_buf_alloc_node(mdev, mlx5_wq_ll_get_byte_size(wq), - max_direct, &wq_ctrl->buf, - param->buf_numa_node); + err = mlx5_buf_alloc(mdev, mlx5_wq_ll_get_byte_size(wq), + max_direct, &wq_ctrl->buf); if (err) { mlx5_core_warn(mdev, "mlx5_buf_alloc() failed, %d\n", err); goto err_db_free; } wq->buf = wq_ctrl->buf.direct.buf; wq->db = wq_ctrl->db.db; for (i = 0; i < wq->sz_m1; i++) { next_seg = mlx5_wq_ll_get_wqe(wq, i); next_seg->next_wqe_index = cpu_to_be16(i + 1); } next_seg = mlx5_wq_ll_get_wqe(wq, i); wq->tail_next = &next_seg->next_wqe_index; wq_ctrl->mdev = mdev; return 0; err_db_free: mlx5_db_free(mdev, &wq_ctrl->db); return err; } void mlx5_wq_destroy(struct mlx5_wq_ctrl *wq_ctrl) { mlx5_buf_free(wq_ctrl->mdev, &wq_ctrl->buf); mlx5_db_free(wq_ctrl->mdev, &wq_ctrl->db); } diff --git a/sys/dev/mlx5/mlx5_core/wq.h b/sys/dev/mlx5/mlx5_core/wq.h index 13f26b0dd9a1..f513e4d7f6b4 100644 --- a/sys/dev/mlx5/mlx5_core/wq.h +++ b/sys/dev/mlx5/mlx5_core/wq.h @@ -1,177 +1,175 @@ /*- * Copyright (c) 2013-2017, Mellanox Technologies, Ltd. All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY AUTHOR AND CONTRIBUTORS `AS IS' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * $FreeBSD$ */ #ifndef __MLX5_WQ_H__ #define __MLX5_WQ_H__ #include struct mlx5_wq_param { int linear; - int buf_numa_node; - int db_numa_node; }; struct mlx5_wq_ctrl { struct mlx5_core_dev *mdev; struct mlx5_buf buf; struct mlx5_db db; }; struct mlx5_frag_wq_ctrl { struct mlx5_core_dev *mdev; struct mlx5_frag_buf frag_buf; struct mlx5_db db; }; struct mlx5_wq_cyc { void *buf; __be32 *db; u16 sz_m1; u8 log_stride; }; struct mlx5_wq_qp { struct mlx5_wq_cyc rq; struct mlx5_wq_cyc sq; }; struct mlx5_cqwq { void *buf; __be32 *db; u32 sz_m1; u32 cc; /* consumer counter */ u8 log_sz; u8 log_stride; }; struct mlx5_wq_ll { void *buf; __be32 *db; __be16 *tail_next; u16 sz_m1; u16 head; u16 wqe_ctr; u16 cur_sz; u8 log_stride; }; int mlx5_wq_cyc_create(struct mlx5_core_dev *mdev, struct mlx5_wq_param *param, void *wqc, struct mlx5_wq_cyc *wq, struct mlx5_wq_ctrl *wq_ctrl); u32 mlx5_wq_cyc_get_size(struct mlx5_wq_cyc *wq); int mlx5_cqwq_create(struct mlx5_core_dev *mdev, struct mlx5_wq_param *param, void *cqc, struct mlx5_cqwq *wq, struct mlx5_wq_ctrl *wq_ctrl); u32 mlx5_cqwq_get_size(struct mlx5_cqwq *wq); int mlx5_wq_ll_create(struct mlx5_core_dev *mdev, struct mlx5_wq_param *param, void *wqc, struct mlx5_wq_ll *wq, struct mlx5_wq_ctrl *wq_ctrl); u32 mlx5_wq_ll_get_size(struct mlx5_wq_ll *wq); void mlx5_wq_destroy(struct mlx5_wq_ctrl *wq_ctrl); static inline u16 mlx5_wq_cyc_ctr2ix(struct mlx5_wq_cyc *wq, u16 ctr) { return ctr & wq->sz_m1; } static inline void *mlx5_wq_cyc_get_wqe(struct mlx5_wq_cyc *wq, u16 ix) { return wq->buf + (ix << wq->log_stride); } static inline int mlx5_wq_cyc_cc_bigger(u16 cc1, u16 cc2) { int equal = (cc1 == cc2); int smaller = 0x8000 & (cc1 - cc2); return !equal && !smaller; } static inline u32 mlx5_cqwq_get_ci(struct mlx5_cqwq *wq) { return wq->cc & wq->sz_m1; } static inline void *mlx5_cqwq_get_wqe(struct mlx5_cqwq *wq, u32 ix) { return wq->buf + (ix << wq->log_stride); } static inline u32 mlx5_cqwq_get_wrap_cnt(struct mlx5_cqwq *wq) { return wq->cc >> wq->log_sz; } static inline void mlx5_cqwq_pop(struct mlx5_cqwq *wq) { wq->cc++; } static inline void mlx5_cqwq_update_db_record(struct mlx5_cqwq *wq) { *wq->db = cpu_to_be32(wq->cc & 0xffffff); } static inline int mlx5_wq_ll_is_full(struct mlx5_wq_ll *wq) { return wq->cur_sz == wq->sz_m1; } static inline int mlx5_wq_ll_is_empty(struct mlx5_wq_ll *wq) { return !wq->cur_sz; } static inline void mlx5_wq_ll_push(struct mlx5_wq_ll *wq, u16 head_next) { wq->head = head_next; wq->wqe_ctr++; wq->cur_sz++; } static inline void mlx5_wq_ll_pop(struct mlx5_wq_ll *wq, __be16 ix, __be16 *next_tail_next) { *wq->tail_next = ix; wq->tail_next = next_tail_next; wq->cur_sz--; } static inline void mlx5_wq_ll_update_db_record(struct mlx5_wq_ll *wq) { *wq->db = cpu_to_be32(wq->wqe_ctr); } static inline void *mlx5_wq_ll_get_wqe(struct mlx5_wq_ll *wq, u16 ix) { return wq->buf + (ix << wq->log_stride); } #endif /* __MLX5_WQ_H__ */ diff --git a/sys/dev/mlx5/mlx5_en/mlx5_en_main.c b/sys/dev/mlx5/mlx5_en/mlx5_en_main.c index 9345798006f2..170fb03be968 100644 --- a/sys/dev/mlx5/mlx5_en/mlx5_en_main.c +++ b/sys/dev/mlx5/mlx5_en/mlx5_en_main.c @@ -1,4932 +1,4928 @@ /*- * Copyright (c) 2015-2021 Mellanox Technologies. All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY AUTHOR AND CONTRIBUTORS `AS IS' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * $FreeBSD$ */ #include "opt_kern_tls.h" #include "en.h" #include #include #include #include #ifndef ETH_DRIVER_VERSION #define ETH_DRIVER_VERSION "3.6.0" #endif #define DRIVER_RELDATE "December 2020" static const char mlx5e_version[] = "mlx5en: Mellanox Ethernet driver " ETH_DRIVER_VERSION " (" DRIVER_RELDATE ")\n"; static int mlx5e_get_wqe_sz(struct mlx5e_priv *priv, u32 *wqe_sz, u32 *nsegs); struct mlx5e_channel_param { struct mlx5e_rq_param rq; struct mlx5e_sq_param sq; struct mlx5e_cq_param rx_cq; struct mlx5e_cq_param tx_cq; }; struct media { u32 subtype; u64 baudrate; }; static const struct media mlx5e_mode_table[MLX5E_LINK_SPEEDS_NUMBER] = { [MLX5E_1000BASE_CX_SGMII] = { .subtype = IFM_1000_CX_SGMII, .baudrate = IF_Mbps(1000ULL), }, [MLX5E_1000BASE_KX] = { .subtype = IFM_1000_KX, .baudrate = IF_Mbps(1000ULL), }, [MLX5E_10GBASE_CX4] = { .subtype = IFM_10G_CX4, .baudrate = IF_Gbps(10ULL), }, [MLX5E_10GBASE_KX4] = { .subtype = IFM_10G_KX4, .baudrate = IF_Gbps(10ULL), }, [MLX5E_10GBASE_KR] = { .subtype = IFM_10G_KR, .baudrate = IF_Gbps(10ULL), }, [MLX5E_20GBASE_KR2] = { .subtype = IFM_20G_KR2, .baudrate = IF_Gbps(20ULL), }, [MLX5E_40GBASE_CR4] = { .subtype = IFM_40G_CR4, .baudrate = IF_Gbps(40ULL), }, [MLX5E_40GBASE_KR4] = { .subtype = IFM_40G_KR4, .baudrate = IF_Gbps(40ULL), }, [MLX5E_56GBASE_R4] = { .subtype = IFM_56G_R4, .baudrate = IF_Gbps(56ULL), }, [MLX5E_10GBASE_CR] = { .subtype = IFM_10G_CR1, .baudrate = IF_Gbps(10ULL), }, [MLX5E_10GBASE_SR] = { .subtype = IFM_10G_SR, .baudrate = IF_Gbps(10ULL), }, [MLX5E_10GBASE_ER_LR] = { .subtype = IFM_10G_ER, .baudrate = IF_Gbps(10ULL), }, [MLX5E_40GBASE_SR4] = { .subtype = IFM_40G_SR4, .baudrate = IF_Gbps(40ULL), }, [MLX5E_40GBASE_LR4_ER4] = { .subtype = IFM_40G_LR4, .baudrate = IF_Gbps(40ULL), }, [MLX5E_100GBASE_CR4] = { .subtype = IFM_100G_CR4, .baudrate = IF_Gbps(100ULL), }, [MLX5E_100GBASE_SR4] = { .subtype = IFM_100G_SR4, .baudrate = IF_Gbps(100ULL), }, [MLX5E_100GBASE_KR4] = { .subtype = IFM_100G_KR4, .baudrate = IF_Gbps(100ULL), }, [MLX5E_100GBASE_LR4] = { .subtype = IFM_100G_LR4, .baudrate = IF_Gbps(100ULL), }, [MLX5E_100BASE_TX] = { .subtype = IFM_100_TX, .baudrate = IF_Mbps(100ULL), }, [MLX5E_1000BASE_T] = { .subtype = IFM_1000_T, .baudrate = IF_Mbps(1000ULL), }, [MLX5E_10GBASE_T] = { .subtype = IFM_10G_T, .baudrate = IF_Gbps(10ULL), }, [MLX5E_25GBASE_CR] = { .subtype = IFM_25G_CR, .baudrate = IF_Gbps(25ULL), }, [MLX5E_25GBASE_KR] = { .subtype = IFM_25G_KR, .baudrate = IF_Gbps(25ULL), }, [MLX5E_25GBASE_SR] = { .subtype = IFM_25G_SR, .baudrate = IF_Gbps(25ULL), }, [MLX5E_50GBASE_CR2] = { .subtype = IFM_50G_CR2, .baudrate = IF_Gbps(50ULL), }, [MLX5E_50GBASE_KR2] = { .subtype = IFM_50G_KR2, .baudrate = IF_Gbps(50ULL), }, [MLX5E_50GBASE_KR4] = { .subtype = IFM_50G_KR4, .baudrate = IF_Gbps(50ULL), }, }; static const struct media mlx5e_ext_mode_table[MLX5E_EXT_LINK_SPEEDS_NUMBER][MLX5E_CABLE_TYPE_NUMBER] = { /**/ [MLX5E_SGMII_100M][MLX5E_CABLE_TYPE_UNKNOWN] = { .subtype = IFM_100_SGMII, .baudrate = IF_Mbps(100), }, /**/ [MLX5E_1000BASE_X_SGMII][MLX5E_CABLE_TYPE_UNKNOWN] = { .subtype = IFM_1000_CX, .baudrate = IF_Mbps(1000), }, [MLX5E_1000BASE_X_SGMII][MLX5E_CABLE_TYPE_OPTICAL_MODULE] = { .subtype = IFM_1000_SX, .baudrate = IF_Mbps(1000), }, /**/ [MLX5E_5GBASE_R][MLX5E_CABLE_TYPE_UNKNOWN] = { .subtype = IFM_5000_KR, .baudrate = IF_Mbps(5000), }, [MLX5E_5GBASE_R][MLX5E_CABLE_TYPE_TWISTED_PAIR] = { .subtype = IFM_5000_T, .baudrate = IF_Mbps(5000), }, /**/ [MLX5E_10GBASE_XFI_XAUI_1][MLX5E_CABLE_TYPE_UNKNOWN] = { .subtype = IFM_10G_KR, .baudrate = IF_Gbps(10ULL), }, [MLX5E_10GBASE_XFI_XAUI_1][MLX5E_CABLE_TYPE_PASSIVE_COPPER] = { .subtype = IFM_10G_CR1, .baudrate = IF_Gbps(10ULL), }, [MLX5E_10GBASE_XFI_XAUI_1][MLX5E_CABLE_TYPE_OPTICAL_MODULE] = { .subtype = IFM_10G_SR, .baudrate = IF_Gbps(10ULL), }, /**/ [MLX5E_40GBASE_XLAUI_4_XLPPI_4][MLX5E_CABLE_TYPE_UNKNOWN] = { .subtype = IFM_40G_KR4, .baudrate = IF_Gbps(40ULL), }, [MLX5E_40GBASE_XLAUI_4_XLPPI_4][MLX5E_CABLE_TYPE_PASSIVE_COPPER] = { .subtype = IFM_40G_CR4, .baudrate = IF_Gbps(40ULL), }, [MLX5E_40GBASE_XLAUI_4_XLPPI_4][MLX5E_CABLE_TYPE_OPTICAL_MODULE] = { .subtype = IFM_40G_SR4, .baudrate = IF_Gbps(40ULL), }, /**/ [MLX5E_25GAUI_1_25GBASE_CR_KR][MLX5E_CABLE_TYPE_UNKNOWN] = { .subtype = IFM_25G_KR, .baudrate = IF_Gbps(25ULL), }, [MLX5E_25GAUI_1_25GBASE_CR_KR][MLX5E_CABLE_TYPE_PASSIVE_COPPER] = { .subtype = IFM_25G_CR, .baudrate = IF_Gbps(25ULL), }, [MLX5E_25GAUI_1_25GBASE_CR_KR][MLX5E_CABLE_TYPE_OPTICAL_MODULE] = { .subtype = IFM_25G_SR, .baudrate = IF_Gbps(25ULL), }, [MLX5E_25GAUI_1_25GBASE_CR_KR][MLX5E_CABLE_TYPE_TWISTED_PAIR] = { .subtype = IFM_25G_T, .baudrate = IF_Gbps(25ULL), }, /**/ [MLX5E_50GAUI_2_LAUI_2_50GBASE_CR2_KR2][MLX5E_CABLE_TYPE_UNKNOWN] = { .subtype = IFM_50G_KR2, .baudrate = IF_Gbps(50ULL), }, [MLX5E_50GAUI_2_LAUI_2_50GBASE_CR2_KR2][MLX5E_CABLE_TYPE_PASSIVE_COPPER] = { .subtype = IFM_50G_CR2, .baudrate = IF_Gbps(50ULL), }, [MLX5E_50GAUI_2_LAUI_2_50GBASE_CR2_KR2][MLX5E_CABLE_TYPE_OPTICAL_MODULE] = { .subtype = IFM_50G_SR2, .baudrate = IF_Gbps(50ULL), }, /**/ [MLX5E_50GAUI_1_LAUI_1_50GBASE_CR_KR][MLX5E_CABLE_TYPE_UNKNOWN] = { .subtype = IFM_50G_KR_PAM4, .baudrate = IF_Gbps(50ULL), }, [MLX5E_50GAUI_1_LAUI_1_50GBASE_CR_KR][MLX5E_CABLE_TYPE_PASSIVE_COPPER] = { .subtype = IFM_50G_CP, .baudrate = IF_Gbps(50ULL), }, [MLX5E_50GAUI_1_LAUI_1_50GBASE_CR_KR][MLX5E_CABLE_TYPE_OPTICAL_MODULE] = { .subtype = IFM_50G_SR, .baudrate = IF_Gbps(50ULL), }, /**/ [MLX5E_CAUI_4_100GBASE_CR4_KR4][MLX5E_CABLE_TYPE_UNKNOWN] = { .subtype = IFM_100G_KR4, .baudrate = IF_Gbps(100ULL), }, [MLX5E_CAUI_4_100GBASE_CR4_KR4][MLX5E_CABLE_TYPE_PASSIVE_COPPER] = { .subtype = IFM_100G_CR4, .baudrate = IF_Gbps(100ULL), }, [MLX5E_CAUI_4_100GBASE_CR4_KR4][MLX5E_CABLE_TYPE_OPTICAL_MODULE] = { .subtype = IFM_100G_SR4, .baudrate = IF_Gbps(100ULL), }, /**/ [MLX5E_100GAUI_1_100GBASE_CR_KR][MLX5E_CABLE_TYPE_UNKNOWN] = { .subtype = IFM_100G_KR_PAM4, .baudrate = IF_Gbps(100ULL), }, [MLX5E_100GAUI_1_100GBASE_CR_KR][MLX5E_CABLE_TYPE_PASSIVE_COPPER] = { .subtype = IFM_100G_CR_PAM4, .baudrate = IF_Gbps(100ULL), }, [MLX5E_100GAUI_1_100GBASE_CR_KR][MLX5E_CABLE_TYPE_OPTICAL_MODULE] = { .subtype = IFM_100G_SR2, /* XXX */ .baudrate = IF_Gbps(100ULL), }, /**/ [MLX5E_100GAUI_2_100GBASE_CR2_KR2][MLX5E_CABLE_TYPE_UNKNOWN] = { .subtype = IFM_100G_KR4, .baudrate = IF_Gbps(100ULL), }, [MLX5E_100GAUI_2_100GBASE_CR2_KR2][MLX5E_CABLE_TYPE_PASSIVE_COPPER] = { .subtype = IFM_100G_CP2, .baudrate = IF_Gbps(100ULL), }, [MLX5E_100GAUI_2_100GBASE_CR2_KR2][MLX5E_CABLE_TYPE_OPTICAL_MODULE] = { .subtype = IFM_100G_SR2, .baudrate = IF_Gbps(100ULL), }, /**/ [MLX5E_200GAUI_2_200GBASE_CR2_KR2][MLX5E_CABLE_TYPE_UNKNOWN] = { .subtype = IFM_200G_KR4_PAM4, /* XXX */ .baudrate = IF_Gbps(200ULL), }, [MLX5E_200GAUI_2_200GBASE_CR2_KR2][MLX5E_CABLE_TYPE_PASSIVE_COPPER] = { .subtype = IFM_200G_CR4_PAM4, /* XXX */ .baudrate = IF_Gbps(200ULL), }, [MLX5E_200GAUI_2_200GBASE_CR2_KR2][MLX5E_CABLE_TYPE_OPTICAL_MODULE] = { .subtype = IFM_200G_SR4, /* XXX */ .baudrate = IF_Gbps(200ULL), }, /**/ [MLX5E_200GAUI_4_200GBASE_CR4_KR4][MLX5E_CABLE_TYPE_UNKNOWN] = { .subtype = IFM_200G_KR4_PAM4, .baudrate = IF_Gbps(200ULL), }, [MLX5E_200GAUI_4_200GBASE_CR4_KR4][MLX5E_CABLE_TYPE_PASSIVE_COPPER] = { .subtype = IFM_200G_CR4_PAM4, .baudrate = IF_Gbps(200ULL), }, [MLX5E_200GAUI_4_200GBASE_CR4_KR4][MLX5E_CABLE_TYPE_OPTICAL_MODULE] = { .subtype = IFM_200G_SR4, .baudrate = IF_Gbps(200ULL), }, /**/ [MLX5E_400GAUI_8][MLX5E_CABLE_TYPE_UNKNOWN] = { .subtype = IFM_400G_LR8, /* XXX */ .baudrate = IF_Gbps(400ULL), }, /**/ [MLX5E_400GAUI_4_400GBASE_CR4_KR4][MLX5E_CABLE_TYPE_UNKNOWN] = { .subtype = IFM_400G_LR8, /* XXX */ .baudrate = IF_Gbps(400ULL), }, }; DEBUGNET_DEFINE(mlx5_en); MALLOC_DEFINE(M_MLX5EN, "MLX5EN", "MLX5 Ethernet"); static void mlx5e_update_carrier(struct mlx5e_priv *priv) { struct mlx5_core_dev *mdev = priv->mdev; u32 out[MLX5_ST_SZ_DW(ptys_reg)]; u32 eth_proto_oper; int error; u8 i; u8 cable_type; u8 port_state; u8 is_er_type; bool ext; struct media media_entry = {}; port_state = mlx5_query_vport_state(mdev, MLX5_QUERY_VPORT_STATE_IN_OP_MOD_VNIC_VPORT, 0); if (port_state == VPORT_STATE_UP) { priv->media_status_last |= IFM_ACTIVE; } else { priv->media_status_last &= ~IFM_ACTIVE; priv->media_active_last = IFM_ETHER; if_link_state_change(priv->ifp, LINK_STATE_DOWN); return; } error = mlx5_query_port_ptys(mdev, out, sizeof(out), MLX5_PTYS_EN, 1); if (error) { priv->media_active_last = IFM_ETHER; priv->ifp->if_baudrate = 1; mlx5_en_err(priv->ifp, "query port ptys failed: 0x%x\n", error); return; } ext = MLX5_CAP_PCAM_FEATURE(mdev, ptys_extended_ethernet); eth_proto_oper = MLX5_GET_ETH_PROTO(ptys_reg, out, ext, eth_proto_oper); i = ilog2(eth_proto_oper); if (ext) { error = mlx5_query_pddr_cable_type(mdev, 1, &cable_type); if (error != 0) { /* use fallback entry */ media_entry = mlx5e_ext_mode_table[i][MLX5E_CABLE_TYPE_UNKNOWN]; mlx5_en_err(priv->ifp, "query port pddr failed: %d\n", error); } else { media_entry = mlx5e_ext_mode_table[i][cable_type]; /* check if we should use fallback entry */ if (media_entry.subtype == 0) media_entry = mlx5e_ext_mode_table[i][MLX5E_CABLE_TYPE_UNKNOWN]; } } else { media_entry = mlx5e_mode_table[i]; } if (media_entry.subtype == 0) { mlx5_en_err(priv->ifp, "Could not find operational media subtype\n"); return; } switch (media_entry.subtype) { case IFM_10G_ER: error = mlx5_query_pddr_range_info(mdev, 1, &is_er_type); if (error != 0) { mlx5_en_err(priv->ifp, "query port pddr failed: %d\n", error); } if (error != 0 || is_er_type == 0) media_entry.subtype = IFM_10G_LR; break; case IFM_40G_LR4: error = mlx5_query_pddr_range_info(mdev, 1, &is_er_type); if (error != 0) { mlx5_en_err(priv->ifp, "query port pddr failed: %d\n", error); } if (error == 0 && is_er_type != 0) media_entry.subtype = IFM_40G_ER4; break; } priv->media_active_last = media_entry.subtype | IFM_ETHER | IFM_FDX; priv->ifp->if_baudrate = media_entry.baudrate; if_link_state_change(priv->ifp, LINK_STATE_UP); } static void mlx5e_media_status(struct ifnet *dev, struct ifmediareq *ifmr) { struct mlx5e_priv *priv = dev->if_softc; ifmr->ifm_status = priv->media_status_last; ifmr->ifm_current = ifmr->ifm_active = priv->media_active_last | (priv->params.rx_pauseframe_control ? IFM_ETH_RXPAUSE : 0) | (priv->params.tx_pauseframe_control ? IFM_ETH_TXPAUSE : 0); } static u32 mlx5e_find_link_mode(u32 subtype, bool ext) { u32 link_mode = 0; switch (subtype) { case 0: goto done; case IFM_10G_LR: subtype = IFM_10G_ER; break; case IFM_40G_ER4: subtype = IFM_40G_LR4; break; default: break; } if (ext) { for (unsigned i = 0; i != MLX5E_EXT_LINK_SPEEDS_NUMBER; i++) { for (unsigned j = 0; j != MLX5E_CABLE_TYPE_NUMBER; j++) { if (mlx5e_ext_mode_table[i][j].subtype == subtype) link_mode |= MLX5E_PROT_MASK(i); } } } else { for (unsigned i = 0; i != MLX5E_LINK_SPEEDS_NUMBER; i++) { if (mlx5e_mode_table[i].subtype == subtype) link_mode |= MLX5E_PROT_MASK(i); } } done: return (link_mode); } static int mlx5e_set_port_pause_and_pfc(struct mlx5e_priv *priv) { return (mlx5_set_port_pause_and_pfc(priv->mdev, 1, priv->params.rx_pauseframe_control, priv->params.tx_pauseframe_control, priv->params.rx_priority_flow_control, priv->params.tx_priority_flow_control)); } static int mlx5e_set_port_pfc(struct mlx5e_priv *priv) { int error; if (priv->gone != 0) { error = -ENXIO; } else if (priv->params.rx_pauseframe_control || priv->params.tx_pauseframe_control) { mlx5_en_err(priv->ifp, "Global pauseframes must be disabled before enabling PFC.\n"); error = -EINVAL; } else { error = mlx5e_set_port_pause_and_pfc(priv); } return (error); } static int mlx5e_media_change(struct ifnet *dev) { struct mlx5e_priv *priv = dev->if_softc; struct mlx5_core_dev *mdev = priv->mdev; u32 eth_proto_cap; u32 link_mode; u32 out[MLX5_ST_SZ_DW(ptys_reg)]; int was_opened; int locked; int error; bool ext; locked = PRIV_LOCKED(priv); if (!locked) PRIV_LOCK(priv); if (IFM_TYPE(priv->media.ifm_media) != IFM_ETHER) { error = EINVAL; goto done; } error = mlx5_query_port_ptys(mdev, out, sizeof(out), MLX5_PTYS_EN, 1); if (error != 0) { mlx5_en_err(dev, "Query port media capability failed\n"); goto done; } ext = MLX5_CAP_PCAM_FEATURE(mdev, ptys_extended_ethernet); link_mode = mlx5e_find_link_mode(IFM_SUBTYPE(priv->media.ifm_media), ext); /* query supported capabilities */ eth_proto_cap = MLX5_GET_ETH_PROTO(ptys_reg, out, ext, eth_proto_capability); /* check for autoselect */ if (IFM_SUBTYPE(priv->media.ifm_media) == IFM_AUTO) { link_mode = eth_proto_cap; if (link_mode == 0) { mlx5_en_err(dev, "Port media capability is zero\n"); error = EINVAL; goto done; } } else { link_mode = link_mode & eth_proto_cap; if (link_mode == 0) { mlx5_en_err(dev, "Not supported link mode requested\n"); error = EINVAL; goto done; } } if (priv->media.ifm_media & (IFM_ETH_RXPAUSE | IFM_ETH_TXPAUSE)) { /* check if PFC is enabled */ if (priv->params.rx_priority_flow_control || priv->params.tx_priority_flow_control) { mlx5_en_err(dev, "PFC must be disabled before enabling global pauseframes.\n"); error = EINVAL; goto done; } } /* update pauseframe control bits */ priv->params.rx_pauseframe_control = (priv->media.ifm_media & IFM_ETH_RXPAUSE) ? 1 : 0; priv->params.tx_pauseframe_control = (priv->media.ifm_media & IFM_ETH_TXPAUSE) ? 1 : 0; /* check if device is opened */ was_opened = test_bit(MLX5E_STATE_OPENED, &priv->state); /* reconfigure the hardware */ mlx5_set_port_status(mdev, MLX5_PORT_DOWN); mlx5_set_port_proto(mdev, link_mode, MLX5_PTYS_EN, ext); error = -mlx5e_set_port_pause_and_pfc(priv); if (was_opened) mlx5_set_port_status(mdev, MLX5_PORT_UP); done: if (!locked) PRIV_UNLOCK(priv); return (error); } static void mlx5e_update_carrier_work(struct work_struct *work) { struct mlx5e_priv *priv = container_of(work, struct mlx5e_priv, update_carrier_work); PRIV_LOCK(priv); if (test_bit(MLX5E_STATE_OPENED, &priv->state)) mlx5e_update_carrier(priv); PRIV_UNLOCK(priv); } #define MLX5E_PCIE_PERF_GET_64(a,b,c,d,e,f) \ s_debug->c = MLX5_GET64(mpcnt_reg, out, counter_set.f.c); #define MLX5E_PCIE_PERF_GET_32(a,b,c,d,e,f) \ s_debug->c = MLX5_GET(mpcnt_reg, out, counter_set.f.c); static void mlx5e_update_pcie_counters(struct mlx5e_priv *priv) { struct mlx5_core_dev *mdev = priv->mdev; struct mlx5e_port_stats_debug *s_debug = &priv->stats.port_stats_debug; const unsigned sz = MLX5_ST_SZ_BYTES(mpcnt_reg); void *out; void *in; int err; /* allocate firmware request structures */ in = mlx5_vzalloc(sz); out = mlx5_vzalloc(sz); if (in == NULL || out == NULL) goto free_out; MLX5_SET(mpcnt_reg, in, grp, MLX5_PCIE_PERFORMANCE_COUNTERS_GROUP); err = mlx5_core_access_reg(mdev, in, sz, out, sz, MLX5_REG_MPCNT, 0, 0); if (err != 0) goto free_out; MLX5E_PCIE_PERFORMANCE_COUNTERS_64(MLX5E_PCIE_PERF_GET_64) MLX5E_PCIE_PERFORMANCE_COUNTERS_32(MLX5E_PCIE_PERF_GET_32) MLX5_SET(mpcnt_reg, in, grp, MLX5_PCIE_TIMERS_AND_STATES_COUNTERS_GROUP); err = mlx5_core_access_reg(mdev, in, sz, out, sz, MLX5_REG_MPCNT, 0, 0); if (err != 0) goto free_out; MLX5E_PCIE_TIMERS_AND_STATES_COUNTERS_32(MLX5E_PCIE_PERF_GET_32) MLX5_SET(mpcnt_reg, in, grp, MLX5_PCIE_LANE_COUNTERS_GROUP); err = mlx5_core_access_reg(mdev, in, sz, out, sz, MLX5_REG_MPCNT, 0, 0); if (err != 0) goto free_out; MLX5E_PCIE_LANE_COUNTERS_32(MLX5E_PCIE_PERF_GET_32) free_out: /* free firmware request structures */ kvfree(in); kvfree(out); } /* * This function reads the physical port counters from the firmware * using a pre-defined layout defined by various MLX5E_PPORT_XXX() * macros. The output is converted from big-endian 64-bit values into * host endian ones and stored in the "priv->stats.pport" structure. */ static void mlx5e_update_pport_counters(struct mlx5e_priv *priv) { struct mlx5_core_dev *mdev = priv->mdev; struct mlx5e_pport_stats *s = &priv->stats.pport; struct mlx5e_port_stats_debug *s_debug = &priv->stats.port_stats_debug; u32 *in; u32 *out; const u64 *ptr; unsigned sz = MLX5_ST_SZ_BYTES(ppcnt_reg); unsigned x; unsigned y; unsigned z; /* allocate firmware request structures */ in = mlx5_vzalloc(sz); out = mlx5_vzalloc(sz); if (in == NULL || out == NULL) goto free_out; /* * Get pointer to the 64-bit counter set which is located at a * fixed offset in the output firmware request structure: */ ptr = (const uint64_t *)MLX5_ADDR_OF(ppcnt_reg, out, counter_set); MLX5_SET(ppcnt_reg, in, local_port, 1); /* read IEEE802_3 counter group using predefined counter layout */ MLX5_SET(ppcnt_reg, in, grp, MLX5_IEEE_802_3_COUNTERS_GROUP); mlx5_core_access_reg(mdev, in, sz, out, sz, MLX5_REG_PPCNT, 0, 0); for (x = 0, y = MLX5E_PPORT_PER_PRIO_STATS_NUM; x != MLX5E_PPORT_IEEE802_3_STATS_NUM; x++, y++) s->arg[y] = be64toh(ptr[x]); /* read RFC2819 counter group using predefined counter layout */ MLX5_SET(ppcnt_reg, in, grp, MLX5_RFC_2819_COUNTERS_GROUP); mlx5_core_access_reg(mdev, in, sz, out, sz, MLX5_REG_PPCNT, 0, 0); for (x = 0; x != MLX5E_PPORT_RFC2819_STATS_NUM; x++, y++) s->arg[y] = be64toh(ptr[x]); for (y = 0; x != MLX5E_PPORT_RFC2819_STATS_NUM + MLX5E_PPORT_RFC2819_STATS_DEBUG_NUM; x++, y++) s_debug->arg[y] = be64toh(ptr[x]); /* read RFC2863 counter group using predefined counter layout */ MLX5_SET(ppcnt_reg, in, grp, MLX5_RFC_2863_COUNTERS_GROUP); mlx5_core_access_reg(mdev, in, sz, out, sz, MLX5_REG_PPCNT, 0, 0); for (x = 0; x != MLX5E_PPORT_RFC2863_STATS_DEBUG_NUM; x++, y++) s_debug->arg[y] = be64toh(ptr[x]); /* read physical layer stats counter group using predefined counter layout */ MLX5_SET(ppcnt_reg, in, grp, MLX5_PHYSICAL_LAYER_COUNTERS_GROUP); mlx5_core_access_reg(mdev, in, sz, out, sz, MLX5_REG_PPCNT, 0, 0); for (x = 0; x != MLX5E_PPORT_PHYSICAL_LAYER_STATS_DEBUG_NUM; x++, y++) s_debug->arg[y] = be64toh(ptr[x]); /* read Extended Ethernet counter group using predefined counter layout */ MLX5_SET(ppcnt_reg, in, grp, MLX5_ETHERNET_EXTENDED_COUNTERS_GROUP); mlx5_core_access_reg(mdev, in, sz, out, sz, MLX5_REG_PPCNT, 0, 0); for (x = 0; x != MLX5E_PPORT_ETHERNET_EXTENDED_STATS_DEBUG_NUM; x++, y++) s_debug->arg[y] = be64toh(ptr[x]); /* read Extended Statistical Group */ if (MLX5_CAP_GEN(mdev, pcam_reg) && MLX5_CAP_PCAM_FEATURE(mdev, ppcnt_statistical_group) && MLX5_CAP_PCAM_FEATURE(mdev, per_lane_error_counters)) { /* read Extended Statistical counter group using predefined counter layout */ MLX5_SET(ppcnt_reg, in, grp, MLX5_PHYSICAL_LAYER_STATISTICAL_GROUP); mlx5_core_access_reg(mdev, in, sz, out, sz, MLX5_REG_PPCNT, 0, 0); for (x = 0; x != MLX5E_PPORT_STATISTICAL_DEBUG_NUM; x++, y++) s_debug->arg[y] = be64toh(ptr[x]); } /* read PCIE counters */ mlx5e_update_pcie_counters(priv); /* read per-priority counters */ MLX5_SET(ppcnt_reg, in, grp, MLX5_PER_PRIORITY_COUNTERS_GROUP); /* iterate all the priorities */ for (y = z = 0; z != MLX5E_PPORT_PER_PRIO_STATS_NUM_PRIO; z++) { MLX5_SET(ppcnt_reg, in, prio_tc, z); mlx5_core_access_reg(mdev, in, sz, out, sz, MLX5_REG_PPCNT, 0, 0); /* read per priority stats counter group using predefined counter layout */ for (x = 0; x != (MLX5E_PPORT_PER_PRIO_STATS_NUM / MLX5E_PPORT_PER_PRIO_STATS_NUM_PRIO); x++, y++) s->arg[y] = be64toh(ptr[x]); } free_out: /* free firmware request structures */ kvfree(in); kvfree(out); } static void mlx5e_grp_vnic_env_update_stats(struct mlx5e_priv *priv) { u32 out[MLX5_ST_SZ_DW(query_vnic_env_out)] = {}; u32 in[MLX5_ST_SZ_DW(query_vnic_env_in)] = {}; if (!MLX5_CAP_GEN(priv->mdev, nic_receive_steering_discard)) return; MLX5_SET(query_vnic_env_in, in, opcode, MLX5_CMD_OP_QUERY_VNIC_ENV); MLX5_SET(query_vnic_env_in, in, op_mod, 0); MLX5_SET(query_vnic_env_in, in, other_vport, 0); if (mlx5_cmd_exec(priv->mdev, in, sizeof(in), out, sizeof(out)) != 0) return; priv->stats.vport.rx_steer_missed_packets = MLX5_GET64(query_vnic_env_out, out, vport_env.nic_receive_steering_discard); } /* * This function is called regularly to collect all statistics * counters from the firmware. The values can be viewed through the * sysctl interface. Execution is serialized using the priv's global * configuration lock. */ static void mlx5e_update_stats_locked(struct mlx5e_priv *priv) { struct mlx5_core_dev *mdev = priv->mdev; struct mlx5e_vport_stats *s = &priv->stats.vport; struct mlx5e_sq_stats *sq_stats; #if (__FreeBSD_version < 1100000) struct ifnet *ifp = priv->ifp; #endif u32 in[MLX5_ST_SZ_DW(query_vport_counter_in)]; u32 *out; int outlen = MLX5_ST_SZ_BYTES(query_vport_counter_out); u64 tso_packets = 0; u64 tso_bytes = 0; u64 tx_queue_dropped = 0; u64 tx_defragged = 0; u64 tx_offload_none = 0; u64 lro_packets = 0; u64 lro_bytes = 0; u64 sw_lro_queued = 0; u64 sw_lro_flushed = 0; u64 rx_csum_none = 0; u64 rx_wqe_err = 0; u64 rx_packets = 0; u64 rx_bytes = 0; u32 rx_out_of_buffer = 0; int error; int i; int j; out = mlx5_vzalloc(outlen); if (out == NULL) goto free_out; /* Collect firts the SW counters and then HW for consistency */ for (i = 0; i < priv->params.num_channels; i++) { struct mlx5e_channel *pch = priv->channel + i; struct mlx5e_rq *rq = &pch->rq; struct mlx5e_rq_stats *rq_stats = &pch->rq.stats; /* collect stats from LRO */ rq_stats->sw_lro_queued = rq->lro.lro_queued; rq_stats->sw_lro_flushed = rq->lro.lro_flushed; sw_lro_queued += rq_stats->sw_lro_queued; sw_lro_flushed += rq_stats->sw_lro_flushed; lro_packets += rq_stats->lro_packets; lro_bytes += rq_stats->lro_bytes; rx_csum_none += rq_stats->csum_none; rx_wqe_err += rq_stats->wqe_err; rx_packets += rq_stats->packets; rx_bytes += rq_stats->bytes; for (j = 0; j < priv->num_tc; j++) { sq_stats = &pch->sq[j].stats; tso_packets += sq_stats->tso_packets; tso_bytes += sq_stats->tso_bytes; tx_queue_dropped += sq_stats->dropped; tx_queue_dropped += sq_stats->enobuf; tx_defragged += sq_stats->defragged; tx_offload_none += sq_stats->csum_offload_none; } } #ifdef RATELIMIT /* Collect statistics from all rate-limit queues */ for (j = 0; j < priv->rl.param.tx_worker_threads_def; j++) { struct mlx5e_rl_worker *rlw = priv->rl.workers + j; for (i = 0; i < priv->rl.param.tx_channels_per_worker_def; i++) { struct mlx5e_rl_channel *channel = rlw->channels + i; struct mlx5e_sq *sq = channel->sq; if (sq == NULL) continue; sq_stats = &sq->stats; tso_packets += sq_stats->tso_packets; tso_bytes += sq_stats->tso_bytes; tx_queue_dropped += sq_stats->dropped; tx_queue_dropped += sq_stats->enobuf; tx_defragged += sq_stats->defragged; tx_offload_none += sq_stats->csum_offload_none; } } #endif /* update counters */ s->tso_packets = tso_packets; s->tso_bytes = tso_bytes; s->tx_queue_dropped = tx_queue_dropped; s->tx_defragged = tx_defragged; s->lro_packets = lro_packets; s->lro_bytes = lro_bytes; s->sw_lro_queued = sw_lro_queued; s->sw_lro_flushed = sw_lro_flushed; s->rx_csum_none = rx_csum_none; s->rx_wqe_err = rx_wqe_err; s->rx_packets = rx_packets; s->rx_bytes = rx_bytes; mlx5e_grp_vnic_env_update_stats(priv); /* HW counters */ memset(in, 0, sizeof(in)); MLX5_SET(query_vport_counter_in, in, opcode, MLX5_CMD_OP_QUERY_VPORT_COUNTER); MLX5_SET(query_vport_counter_in, in, op_mod, 0); MLX5_SET(query_vport_counter_in, in, other_vport, 0); memset(out, 0, outlen); /* get number of out-of-buffer drops first */ if (test_bit(MLX5E_STATE_OPENED, &priv->state) != 0 && mlx5_vport_query_out_of_rx_buffer(mdev, priv->counter_set_id, &rx_out_of_buffer) == 0) { s->rx_out_of_buffer = rx_out_of_buffer; } /* get port statistics */ if (mlx5_cmd_exec(mdev, in, sizeof(in), out, outlen) == 0) { #define MLX5_GET_CTR(out, x) \ MLX5_GET64(query_vport_counter_out, out, x) s->rx_error_packets = MLX5_GET_CTR(out, received_errors.packets); s->rx_error_bytes = MLX5_GET_CTR(out, received_errors.octets); s->tx_error_packets = MLX5_GET_CTR(out, transmit_errors.packets); s->tx_error_bytes = MLX5_GET_CTR(out, transmit_errors.octets); s->rx_unicast_packets = MLX5_GET_CTR(out, received_eth_unicast.packets); s->rx_unicast_bytes = MLX5_GET_CTR(out, received_eth_unicast.octets); s->tx_unicast_packets = MLX5_GET_CTR(out, transmitted_eth_unicast.packets); s->tx_unicast_bytes = MLX5_GET_CTR(out, transmitted_eth_unicast.octets); s->rx_multicast_packets = MLX5_GET_CTR(out, received_eth_multicast.packets); s->rx_multicast_bytes = MLX5_GET_CTR(out, received_eth_multicast.octets); s->tx_multicast_packets = MLX5_GET_CTR(out, transmitted_eth_multicast.packets); s->tx_multicast_bytes = MLX5_GET_CTR(out, transmitted_eth_multicast.octets); s->rx_broadcast_packets = MLX5_GET_CTR(out, received_eth_broadcast.packets); s->rx_broadcast_bytes = MLX5_GET_CTR(out, received_eth_broadcast.octets); s->tx_broadcast_packets = MLX5_GET_CTR(out, transmitted_eth_broadcast.packets); s->tx_broadcast_bytes = MLX5_GET_CTR(out, transmitted_eth_broadcast.octets); s->tx_packets = s->tx_unicast_packets + s->tx_multicast_packets + s->tx_broadcast_packets; s->tx_bytes = s->tx_unicast_bytes + s->tx_multicast_bytes + s->tx_broadcast_bytes; /* Update calculated offload counters */ s->tx_csum_offload = s->tx_packets - tx_offload_none; s->rx_csum_good = s->rx_packets - s->rx_csum_none; } /* Get physical port counters */ mlx5e_update_pport_counters(priv); s->tx_jumbo_packets = priv->stats.port_stats_debug.tx_stat_p1519to2047octets + priv->stats.port_stats_debug.tx_stat_p2048to4095octets + priv->stats.port_stats_debug.tx_stat_p4096to8191octets + priv->stats.port_stats_debug.tx_stat_p8192to10239octets; #if (__FreeBSD_version < 1100000) /* no get_counters interface in fbsd 10 */ ifp->if_ipackets = s->rx_packets; ifp->if_ierrors = priv->stats.pport.in_range_len_errors + priv->stats.pport.out_of_range_len + priv->stats.pport.too_long_errors + priv->stats.pport.check_seq_err + priv->stats.pport.alignment_err; ifp->if_iqdrops = s->rx_out_of_buffer; ifp->if_opackets = s->tx_packets; ifp->if_oerrors = priv->stats.port_stats_debug.out_discards; ifp->if_snd.ifq_drops = s->tx_queue_dropped; ifp->if_ibytes = s->rx_bytes; ifp->if_obytes = s->tx_bytes; ifp->if_collisions = priv->stats.pport.collisions; #endif free_out: kvfree(out); /* Update diagnostics, if any */ if (priv->params_ethtool.diag_pci_enable || priv->params_ethtool.diag_general_enable) { error = mlx5_core_get_diagnostics_full(mdev, priv->params_ethtool.diag_pci_enable ? &priv->params_pci : NULL, priv->params_ethtool.diag_general_enable ? &priv->params_general : NULL); if (error != 0) mlx5_en_err(priv->ifp, "Failed reading diagnostics: %d\n", error); } /* Update FEC, if any */ error = mlx5e_fec_update(priv); if (error != 0 && error != EOPNOTSUPP) { mlx5_en_err(priv->ifp, "Updating FEC failed: %d\n", error); } /* Update temperature, if any */ if (priv->params_ethtool.hw_num_temp != 0) { error = mlx5e_hw_temperature_update(priv); if (error != 0 && error != EOPNOTSUPP) { mlx5_en_err(priv->ifp, "Updating temperature failed: %d\n", error); } } } static void mlx5e_update_stats_work(struct work_struct *work) { struct mlx5e_priv *priv; priv = container_of(work, struct mlx5e_priv, update_stats_work); PRIV_LOCK(priv); if (test_bit(MLX5E_STATE_OPENED, &priv->state) != 0 && !test_bit(MLX5_INTERFACE_STATE_TEARDOWN, &priv->mdev->intf_state)) mlx5e_update_stats_locked(priv); PRIV_UNLOCK(priv); } static void mlx5e_update_stats(void *arg) { struct mlx5e_priv *priv = arg; queue_work(priv->wq, &priv->update_stats_work); callout_reset(&priv->watchdog, hz / 4, &mlx5e_update_stats, priv); } static void mlx5e_async_event_sub(struct mlx5e_priv *priv, enum mlx5_dev_event event) { switch (event) { case MLX5_DEV_EVENT_PORT_UP: case MLX5_DEV_EVENT_PORT_DOWN: queue_work(priv->wq, &priv->update_carrier_work); break; default: break; } } static void mlx5e_async_event(struct mlx5_core_dev *mdev, void *vpriv, enum mlx5_dev_event event, unsigned long param) { struct mlx5e_priv *priv = vpriv; mtx_lock(&priv->async_events_mtx); if (test_bit(MLX5E_STATE_ASYNC_EVENTS_ENABLE, &priv->state)) mlx5e_async_event_sub(priv, event); mtx_unlock(&priv->async_events_mtx); } static void mlx5e_enable_async_events(struct mlx5e_priv *priv) { set_bit(MLX5E_STATE_ASYNC_EVENTS_ENABLE, &priv->state); } static void mlx5e_disable_async_events(struct mlx5e_priv *priv) { mtx_lock(&priv->async_events_mtx); clear_bit(MLX5E_STATE_ASYNC_EVENTS_ENABLE, &priv->state); mtx_unlock(&priv->async_events_mtx); } static void mlx5e_calibration_callout(void *arg); static int mlx5e_calibration_duration = 20; static int mlx5e_fast_calibration = 1; static int mlx5e_normal_calibration = 30; static SYSCTL_NODE(_hw_mlx5, OID_AUTO, calibr, CTLFLAG_RW | CTLFLAG_MPSAFE, 0, "MLX5 timestamp calibration parameteres"); SYSCTL_INT(_hw_mlx5_calibr, OID_AUTO, duration, CTLFLAG_RWTUN, &mlx5e_calibration_duration, 0, "Duration of initial calibration"); SYSCTL_INT(_hw_mlx5_calibr, OID_AUTO, fast, CTLFLAG_RWTUN, &mlx5e_fast_calibration, 0, "Recalibration interval during initial calibration"); SYSCTL_INT(_hw_mlx5_calibr, OID_AUTO, normal, CTLFLAG_RWTUN, &mlx5e_normal_calibration, 0, "Recalibration interval during normal operations"); /* * Ignites the calibration process. */ static void mlx5e_reset_calibration_callout(struct mlx5e_priv *priv) { if (priv->clbr_done == 0) mlx5e_calibration_callout(priv); else callout_reset_curcpu(&priv->tstmp_clbr, (priv->clbr_done < mlx5e_calibration_duration ? mlx5e_fast_calibration : mlx5e_normal_calibration) * hz, mlx5e_calibration_callout, priv); } static uint64_t mlx5e_timespec2usec(const struct timespec *ts) { return ((uint64_t)ts->tv_sec * 1000000000 + ts->tv_nsec); } static uint64_t mlx5e_hw_clock(struct mlx5e_priv *priv) { struct mlx5_init_seg *iseg; uint32_t hw_h, hw_h1, hw_l; iseg = priv->mdev->iseg; do { hw_h = ioread32be(&iseg->internal_timer_h); hw_l = ioread32be(&iseg->internal_timer_l); hw_h1 = ioread32be(&iseg->internal_timer_h); } while (hw_h1 != hw_h); return (((uint64_t)hw_h << 32) | hw_l); } /* * The calibration callout, it runs either in the context of the * thread which enables calibration, or in callout. It takes the * snapshot of system and adapter clocks, then advances the pointers to * the calibration point to allow rx path to read the consistent data * lockless. */ static void mlx5e_calibration_callout(void *arg) { struct mlx5e_priv *priv; struct mlx5e_clbr_point *next, *curr; struct timespec ts; int clbr_curr_next; priv = arg; curr = &priv->clbr_points[priv->clbr_curr]; clbr_curr_next = priv->clbr_curr + 1; if (clbr_curr_next >= nitems(priv->clbr_points)) clbr_curr_next = 0; next = &priv->clbr_points[clbr_curr_next]; next->base_prev = curr->base_curr; next->clbr_hw_prev = curr->clbr_hw_curr; next->clbr_hw_curr = mlx5e_hw_clock(priv); if (((next->clbr_hw_curr - curr->clbr_hw_curr) >> MLX5E_TSTMP_PREC) == 0) { if (priv->clbr_done != 0) { mlx5_en_err(priv->ifp, "HW failed tstmp frozen %#jx %#jx, disabling\n", next->clbr_hw_curr, curr->clbr_hw_prev); priv->clbr_done = 0; } atomic_store_rel_int(&curr->clbr_gen, 0); return; } nanouptime(&ts); next->base_curr = mlx5e_timespec2usec(&ts); curr->clbr_gen = 0; atomic_thread_fence_rel(); priv->clbr_curr = clbr_curr_next; atomic_store_rel_int(&next->clbr_gen, ++(priv->clbr_gen)); if (priv->clbr_done < mlx5e_calibration_duration) priv->clbr_done++; mlx5e_reset_calibration_callout(priv); } static const char *mlx5e_rq_stats_desc[] = { MLX5E_RQ_STATS(MLX5E_STATS_DESC) }; static int mlx5e_create_rq(struct mlx5e_channel *c, struct mlx5e_rq_param *param, struct mlx5e_rq *rq) { struct mlx5e_priv *priv = c->priv; struct mlx5_core_dev *mdev = priv->mdev; char buffer[16]; void *rqc = param->rqc; void *rqc_wq = MLX5_ADDR_OF(rqc, rqc, wq); int wq_sz; int err; int i; u32 nsegs, wqe_sz; err = mlx5e_get_wqe_sz(priv, &wqe_sz, &nsegs); if (err != 0) goto done; /* Create DMA descriptor TAG */ if ((err = -bus_dma_tag_create( bus_get_dma_tag(mdev->pdev->dev.bsddev), 1, /* any alignment */ 0, /* no boundary */ BUS_SPACE_MAXADDR, /* lowaddr */ BUS_SPACE_MAXADDR, /* highaddr */ NULL, NULL, /* filter, filterarg */ nsegs * MLX5E_MAX_RX_BYTES, /* maxsize */ nsegs, /* nsegments */ nsegs * MLX5E_MAX_RX_BYTES, /* maxsegsize */ 0, /* flags */ NULL, NULL, /* lockfunc, lockfuncarg */ &rq->dma_tag))) goto done; err = mlx5_wq_ll_create(mdev, ¶m->wq, rqc_wq, &rq->wq, &rq->wq_ctrl); if (err) goto err_free_dma_tag; rq->wq.db = &rq->wq.db[MLX5_RCV_DBR]; err = mlx5e_get_wqe_sz(priv, &rq->wqe_sz, &rq->nsegs); if (err != 0) goto err_rq_wq_destroy; wq_sz = mlx5_wq_ll_get_size(&rq->wq); err = -tcp_lro_init_args(&rq->lro, priv->ifp, TCP_LRO_ENTRIES, wq_sz); if (err) goto err_rq_wq_destroy; - rq->mbuf = malloc(wq_sz * sizeof(rq->mbuf[0]), M_MLX5EN, M_WAITOK | M_ZERO); + rq->mbuf = malloc_domainset(wq_sz * sizeof(rq->mbuf[0]), M_MLX5EN, + mlx5_dev_domainset(mdev), M_WAITOK | M_ZERO); for (i = 0; i != wq_sz; i++) { struct mlx5e_rx_wqe *wqe = mlx5_wq_ll_get_wqe(&rq->wq, i); int j; err = -bus_dmamap_create(rq->dma_tag, 0, &rq->mbuf[i].dma_map); if (err != 0) { while (i--) bus_dmamap_destroy(rq->dma_tag, rq->mbuf[i].dma_map); goto err_rq_mbuf_free; } /* set value for constant fields */ for (j = 0; j < rq->nsegs; j++) wqe->data[j].lkey = cpu_to_be32(priv->mr.key); } INIT_WORK(&rq->dim.work, mlx5e_dim_work); if (priv->params.rx_cq_moderation_mode < 2) { rq->dim.mode = NET_DIM_CQ_PERIOD_MODE_DISABLED; } else { void *cqc = container_of(param, struct mlx5e_channel_param, rq)->rx_cq.cqc; switch (MLX5_GET(cqc, cqc, cq_period_mode)) { case MLX5_CQ_PERIOD_MODE_START_FROM_EQE: rq->dim.mode = NET_DIM_CQ_PERIOD_MODE_START_FROM_EQE; break; case MLX5_CQ_PERIOD_MODE_START_FROM_CQE: rq->dim.mode = NET_DIM_CQ_PERIOD_MODE_START_FROM_CQE; break; default: rq->dim.mode = NET_DIM_CQ_PERIOD_MODE_DISABLED; break; } } rq->ifp = priv->ifp; rq->channel = c; rq->ix = c->ix; snprintf(buffer, sizeof(buffer), "rxstat%d", c->ix); mlx5e_create_stats(&rq->stats.ctx, SYSCTL_CHILDREN(priv->sysctl_ifnet), buffer, mlx5e_rq_stats_desc, MLX5E_RQ_STATS_NUM, rq->stats.arg); return (0); err_rq_mbuf_free: free(rq->mbuf, M_MLX5EN); tcp_lro_free(&rq->lro); err_rq_wq_destroy: mlx5_wq_destroy(&rq->wq_ctrl); err_free_dma_tag: bus_dma_tag_destroy(rq->dma_tag); done: return (err); } static void mlx5e_destroy_rq(struct mlx5e_rq *rq) { int wq_sz; int i; /* destroy all sysctl nodes */ sysctl_ctx_free(&rq->stats.ctx); /* free leftover LRO packets, if any */ tcp_lro_free(&rq->lro); wq_sz = mlx5_wq_ll_get_size(&rq->wq); for (i = 0; i != wq_sz; i++) { if (rq->mbuf[i].mbuf != NULL) { bus_dmamap_unload(rq->dma_tag, rq->mbuf[i].dma_map); m_freem(rq->mbuf[i].mbuf); } bus_dmamap_destroy(rq->dma_tag, rq->mbuf[i].dma_map); } free(rq->mbuf, M_MLX5EN); mlx5_wq_destroy(&rq->wq_ctrl); bus_dma_tag_destroy(rq->dma_tag); } static int mlx5e_enable_rq(struct mlx5e_rq *rq, struct mlx5e_rq_param *param) { struct mlx5e_channel *c = rq->channel; struct mlx5e_priv *priv = c->priv; struct mlx5_core_dev *mdev = priv->mdev; void *in; void *rqc; void *wq; int inlen; int err; u8 ts_format; inlen = MLX5_ST_SZ_BYTES(create_rq_in) + sizeof(u64) * rq->wq_ctrl.buf.npages; in = mlx5_vzalloc(inlen); if (in == NULL) return (-ENOMEM); ts_format = mlx5_get_rq_default_ts(mdev); rqc = MLX5_ADDR_OF(create_rq_in, in, ctx); wq = MLX5_ADDR_OF(rqc, rqc, wq); memcpy(rqc, param->rqc, sizeof(param->rqc)); MLX5_SET(rqc, rqc, cqn, c->rq.cq.mcq.cqn); MLX5_SET(rqc, rqc, state, MLX5_RQC_STATE_RST); MLX5_SET(rqc, rqc, ts_format, ts_format); MLX5_SET(rqc, rqc, flush_in_error_en, 1); if (priv->counter_set_id >= 0) MLX5_SET(rqc, rqc, counter_set_id, priv->counter_set_id); MLX5_SET(wq, wq, log_wq_pg_sz, rq->wq_ctrl.buf.page_shift - PAGE_SHIFT); MLX5_SET64(wq, wq, dbr_addr, rq->wq_ctrl.db.dma); mlx5_fill_page_array(&rq->wq_ctrl.buf, (__be64 *) MLX5_ADDR_OF(wq, wq, pas)); err = mlx5_core_create_rq(mdev, in, inlen, &rq->rqn); kvfree(in); return (err); } static int mlx5e_modify_rq(struct mlx5e_rq *rq, int curr_state, int next_state) { struct mlx5e_channel *c = rq->channel; struct mlx5e_priv *priv = c->priv; struct mlx5_core_dev *mdev = priv->mdev; void *in; void *rqc; int inlen; int err; inlen = MLX5_ST_SZ_BYTES(modify_rq_in); in = mlx5_vzalloc(inlen); if (in == NULL) return (-ENOMEM); rqc = MLX5_ADDR_OF(modify_rq_in, in, ctx); MLX5_SET(modify_rq_in, in, rqn, rq->rqn); MLX5_SET(modify_rq_in, in, rq_state, curr_state); MLX5_SET(rqc, rqc, state, next_state); err = mlx5_core_modify_rq(mdev, in, inlen); kvfree(in); return (err); } static void mlx5e_disable_rq(struct mlx5e_rq *rq) { struct mlx5e_channel *c = rq->channel; struct mlx5e_priv *priv = c->priv; struct mlx5_core_dev *mdev = priv->mdev; mlx5_core_destroy_rq(mdev, rq->rqn); } static int mlx5e_wait_for_min_rx_wqes(struct mlx5e_rq *rq) { struct mlx5e_channel *c = rq->channel; struct mlx5e_priv *priv = c->priv; struct mlx5_wq_ll *wq = &rq->wq; int i; for (i = 0; i < 1000; i++) { if (wq->cur_sz >= priv->params.min_rx_wqes) return (0); msleep(4); } return (-ETIMEDOUT); } static int mlx5e_open_rq(struct mlx5e_channel *c, struct mlx5e_rq_param *param, struct mlx5e_rq *rq) { int err; err = mlx5e_create_rq(c, param, rq); if (err) return (err); err = mlx5e_enable_rq(rq, param); if (err) goto err_destroy_rq; err = mlx5e_modify_rq(rq, MLX5_RQC_STATE_RST, MLX5_RQC_STATE_RDY); if (err) goto err_disable_rq; c->rq.enabled = 1; return (0); err_disable_rq: mlx5e_disable_rq(rq); err_destroy_rq: mlx5e_destroy_rq(rq); return (err); } static void mlx5e_close_rq(struct mlx5e_rq *rq) { mtx_lock(&rq->mtx); rq->enabled = 0; callout_stop(&rq->watchdog); mtx_unlock(&rq->mtx); mlx5e_modify_rq(rq, MLX5_RQC_STATE_RDY, MLX5_RQC_STATE_ERR); } static void mlx5e_close_rq_wait(struct mlx5e_rq *rq) { mlx5e_disable_rq(rq); mlx5e_close_cq(&rq->cq); cancel_work_sync(&rq->dim.work); mlx5e_destroy_rq(rq); } void mlx5e_free_sq_db(struct mlx5e_sq *sq) { int wq_sz = mlx5_wq_cyc_get_size(&sq->wq); int x; for (x = 0; x != wq_sz; x++) { if (unlikely(sq->mbuf[x].p_refcount != NULL)) { atomic_add_int(sq->mbuf[x].p_refcount, -1); sq->mbuf[x].p_refcount = NULL; } if (sq->mbuf[x].mbuf != NULL) { bus_dmamap_unload(sq->dma_tag, sq->mbuf[x].dma_map); m_freem(sq->mbuf[x].mbuf); } bus_dmamap_destroy(sq->dma_tag, sq->mbuf[x].dma_map); } free(sq->mbuf, M_MLX5EN); } int mlx5e_alloc_sq_db(struct mlx5e_sq *sq) { int wq_sz = mlx5_wq_cyc_get_size(&sq->wq); int err; int x; - sq->mbuf = malloc(wq_sz * sizeof(sq->mbuf[0]), M_MLX5EN, M_WAITOK | M_ZERO); + sq->mbuf = malloc_domainset(wq_sz * sizeof(sq->mbuf[0]), M_MLX5EN, + mlx5_dev_domainset(sq->priv->mdev), M_WAITOK | M_ZERO); /* Create DMA descriptor MAPs */ for (x = 0; x != wq_sz; x++) { err = -bus_dmamap_create(sq->dma_tag, 0, &sq->mbuf[x].dma_map); if (err != 0) { while (x--) bus_dmamap_destroy(sq->dma_tag, sq->mbuf[x].dma_map); free(sq->mbuf, M_MLX5EN); return (err); } } return (0); } static const char *mlx5e_sq_stats_desc[] = { MLX5E_SQ_STATS(MLX5E_STATS_DESC) }; void mlx5e_update_sq_inline(struct mlx5e_sq *sq) { sq->max_inline = sq->priv->params.tx_max_inline; sq->min_inline_mode = sq->priv->params.tx_min_inline_mode; /* * Check if trust state is DSCP or if inline mode is NONE which * indicates CX-5 or newer hardware. */ if (sq->priv->params_ethtool.trust_state != MLX5_QPTS_TRUST_PCP || sq->min_inline_mode == MLX5_INLINE_MODE_NONE) { if (MLX5_CAP_ETH(sq->priv->mdev, wqe_vlan_insert)) sq->min_insert_caps = MLX5E_INSERT_VLAN | MLX5E_INSERT_NON_VLAN; else sq->min_insert_caps = MLX5E_INSERT_NON_VLAN; } else { sq->min_insert_caps = 0; } } static void mlx5e_refresh_sq_inline_sub(struct mlx5e_priv *priv, struct mlx5e_channel *c) { int i; for (i = 0; i != priv->num_tc; i++) { mtx_lock(&c->sq[i].lock); mlx5e_update_sq_inline(&c->sq[i]); mtx_unlock(&c->sq[i].lock); } } void mlx5e_refresh_sq_inline(struct mlx5e_priv *priv) { int i; /* check if channels are closed */ if (test_bit(MLX5E_STATE_OPENED, &priv->state) == 0) return; for (i = 0; i < priv->params.num_channels; i++) mlx5e_refresh_sq_inline_sub(priv, &priv->channel[i]); } static int mlx5e_create_sq(struct mlx5e_channel *c, int tc, struct mlx5e_sq_param *param, struct mlx5e_sq *sq) { struct mlx5e_priv *priv = c->priv; struct mlx5_core_dev *mdev = priv->mdev; char buffer[16]; void *sqc = param->sqc; void *sqc_wq = MLX5_ADDR_OF(sqc, sqc, wq); int err; /* Create DMA descriptor TAG */ if ((err = -bus_dma_tag_create( bus_get_dma_tag(mdev->pdev->dev.bsddev), 1, /* any alignment */ 0, /* no boundary */ BUS_SPACE_MAXADDR, /* lowaddr */ BUS_SPACE_MAXADDR, /* highaddr */ NULL, NULL, /* filter, filterarg */ MLX5E_MAX_TX_PAYLOAD_SIZE, /* maxsize */ MLX5E_MAX_TX_MBUF_FRAGS, /* nsegments */ MLX5E_MAX_TX_MBUF_SIZE, /* maxsegsize */ 0, /* flags */ NULL, NULL, /* lockfunc, lockfuncarg */ &sq->dma_tag))) goto done; + sq->mkey_be = cpu_to_be32(priv->mr.key); + sq->ifp = priv->ifp; + sq->priv = priv; + sq->tc = tc; + err = mlx5_wq_cyc_create(mdev, ¶m->wq, sqc_wq, &sq->wq, &sq->wq_ctrl); if (err) goto err_free_dma_tag; sq->wq.db = &sq->wq.db[MLX5_SND_DBR]; err = mlx5e_alloc_sq_db(sq); if (err) goto err_sq_wq_destroy; - sq->mkey_be = cpu_to_be32(priv->mr.key); - sq->ifp = priv->ifp; - sq->priv = priv; - sq->tc = tc; - mlx5e_update_sq_inline(sq); snprintf(buffer, sizeof(buffer), "txstat%dtc%d", c->ix, tc); mlx5e_create_stats(&sq->stats.ctx, SYSCTL_CHILDREN(priv->sysctl_ifnet), buffer, mlx5e_sq_stats_desc, MLX5E_SQ_STATS_NUM, sq->stats.arg); return (0); err_sq_wq_destroy: mlx5_wq_destroy(&sq->wq_ctrl); err_free_dma_tag: bus_dma_tag_destroy(sq->dma_tag); done: return (err); } static void mlx5e_destroy_sq(struct mlx5e_sq *sq) { /* destroy all sysctl nodes */ sysctl_ctx_free(&sq->stats.ctx); mlx5e_free_sq_db(sq); mlx5_wq_destroy(&sq->wq_ctrl); bus_dma_tag_destroy(sq->dma_tag); } int mlx5e_enable_sq(struct mlx5e_sq *sq, struct mlx5e_sq_param *param, const struct mlx5_sq_bfreg *bfreg, int tis_num) { void *in; void *sqc; void *wq; int inlen; int err; u8 ts_format; inlen = MLX5_ST_SZ_BYTES(create_sq_in) + sizeof(u64) * sq->wq_ctrl.buf.npages; in = mlx5_vzalloc(inlen); if (in == NULL) return (-ENOMEM); sq->uar_map = bfreg->map; ts_format = mlx5_get_sq_default_ts(sq->priv->mdev); sqc = MLX5_ADDR_OF(create_sq_in, in, ctx); wq = MLX5_ADDR_OF(sqc, sqc, wq); memcpy(sqc, param->sqc, sizeof(param->sqc)); MLX5_SET(sqc, sqc, tis_num_0, tis_num); MLX5_SET(sqc, sqc, cqn, sq->cq.mcq.cqn); MLX5_SET(sqc, sqc, state, MLX5_SQC_STATE_RST); MLX5_SET(sqc, sqc, ts_format, ts_format); MLX5_SET(sqc, sqc, tis_lst_sz, 1); MLX5_SET(sqc, sqc, flush_in_error_en, 1); MLX5_SET(sqc, sqc, allow_swp, 1); MLX5_SET(wq, wq, wq_type, MLX5_WQ_TYPE_CYCLIC); MLX5_SET(wq, wq, uar_page, bfreg->index); MLX5_SET(wq, wq, log_wq_pg_sz, sq->wq_ctrl.buf.page_shift - PAGE_SHIFT); MLX5_SET64(wq, wq, dbr_addr, sq->wq_ctrl.db.dma); mlx5_fill_page_array(&sq->wq_ctrl.buf, (__be64 *) MLX5_ADDR_OF(wq, wq, pas)); err = mlx5_core_create_sq(sq->priv->mdev, in, inlen, &sq->sqn); kvfree(in); return (err); } int mlx5e_modify_sq(struct mlx5e_sq *sq, int curr_state, int next_state) { void *in; void *sqc; int inlen; int err; inlen = MLX5_ST_SZ_BYTES(modify_sq_in); in = mlx5_vzalloc(inlen); if (in == NULL) return (-ENOMEM); sqc = MLX5_ADDR_OF(modify_sq_in, in, ctx); MLX5_SET(modify_sq_in, in, sqn, sq->sqn); MLX5_SET(modify_sq_in, in, sq_state, curr_state); MLX5_SET(sqc, sqc, state, next_state); err = mlx5_core_modify_sq(sq->priv->mdev, in, inlen); kvfree(in); return (err); } void mlx5e_disable_sq(struct mlx5e_sq *sq) { mlx5_core_destroy_sq(sq->priv->mdev, sq->sqn); } static int mlx5e_open_sq(struct mlx5e_channel *c, int tc, struct mlx5e_sq_param *param, struct mlx5e_sq *sq) { int err; sq->cev_factor = c->priv->params_ethtool.tx_completion_fact; /* ensure the TX completion event factor is not zero */ if (sq->cev_factor == 0) sq->cev_factor = 1; err = mlx5e_create_sq(c, tc, param, sq); if (err) return (err); err = mlx5e_enable_sq(sq, param, &c->bfreg, c->priv->tisn[tc]); if (err) goto err_destroy_sq; err = mlx5e_modify_sq(sq, MLX5_SQC_STATE_RST, MLX5_SQC_STATE_RDY); if (err) goto err_disable_sq; WRITE_ONCE(sq->running, 1); return (0); err_disable_sq: mlx5e_disable_sq(sq); err_destroy_sq: mlx5e_destroy_sq(sq); return (err); } static void mlx5e_sq_send_nops_locked(struct mlx5e_sq *sq, int can_sleep) { /* fill up remainder with NOPs */ while (sq->cev_counter != 0) { while (!mlx5e_sq_has_room_for(sq, 1)) { if (can_sleep != 0) { mtx_unlock(&sq->lock); msleep(4); mtx_lock(&sq->lock); } else { goto done; } } /* send a single NOP */ mlx5e_send_nop(sq, 1); atomic_thread_fence_rel(); } done: /* Check if we need to write the doorbell */ if (likely(sq->doorbell.d64 != 0)) { mlx5e_tx_notify_hw(sq, sq->doorbell.d32); sq->doorbell.d64 = 0; } } void mlx5e_sq_cev_timeout(void *arg) { struct mlx5e_sq *sq = arg; mtx_assert(&sq->lock, MA_OWNED); /* check next state */ switch (sq->cev_next_state) { case MLX5E_CEV_STATE_SEND_NOPS: /* fill TX ring with NOPs, if any */ mlx5e_sq_send_nops_locked(sq, 0); /* check if completed */ if (sq->cev_counter == 0) { sq->cev_next_state = MLX5E_CEV_STATE_INITIAL; return; } break; default: /* send NOPs on next timeout */ sq->cev_next_state = MLX5E_CEV_STATE_SEND_NOPS; break; } /* restart timer */ callout_reset_curcpu(&sq->cev_callout, hz, mlx5e_sq_cev_timeout, sq); } void mlx5e_drain_sq(struct mlx5e_sq *sq) { int error; struct mlx5_core_dev *mdev= sq->priv->mdev; /* * Check if already stopped. * * NOTE: Serialization of this function is managed by the * caller ensuring the priv's state lock is locked or in case * of rate limit support, a single thread manages drain and * resume of SQs. The "running" variable can therefore safely * be read without any locks. */ if (READ_ONCE(sq->running) == 0) return; /* don't put more packets into the SQ */ WRITE_ONCE(sq->running, 0); /* serialize access to DMA rings */ mtx_lock(&sq->lock); /* teardown event factor timer, if any */ sq->cev_next_state = MLX5E_CEV_STATE_HOLD_NOPS; callout_stop(&sq->cev_callout); /* send dummy NOPs in order to flush the transmit ring */ mlx5e_sq_send_nops_locked(sq, 1); mtx_unlock(&sq->lock); /* wait till SQ is empty or link is down */ mtx_lock(&sq->lock); while (sq->cc != sq->pc && (sq->priv->media_status_last & IFM_ACTIVE) != 0 && mdev->state != MLX5_DEVICE_STATE_INTERNAL_ERROR && pci_channel_offline(mdev->pdev) == 0) { mtx_unlock(&sq->lock); msleep(1); sq->cq.mcq.comp(&sq->cq.mcq, NULL); mtx_lock(&sq->lock); } mtx_unlock(&sq->lock); /* error out remaining requests */ error = mlx5e_modify_sq(sq, MLX5_SQC_STATE_RDY, MLX5_SQC_STATE_ERR); if (error != 0) { mlx5_en_err(sq->ifp, "mlx5e_modify_sq() from RDY to ERR failed: %d\n", error); } /* wait till SQ is empty */ mtx_lock(&sq->lock); while (sq->cc != sq->pc && mdev->state != MLX5_DEVICE_STATE_INTERNAL_ERROR && pci_channel_offline(mdev->pdev) == 0) { mtx_unlock(&sq->lock); msleep(1); sq->cq.mcq.comp(&sq->cq.mcq, NULL); mtx_lock(&sq->lock); } mtx_unlock(&sq->lock); } static void mlx5e_close_sq_wait(struct mlx5e_sq *sq) { mlx5e_drain_sq(sq); mlx5e_disable_sq(sq); mlx5e_destroy_sq(sq); } static int mlx5e_create_cq(struct mlx5e_priv *priv, struct mlx5e_cq_param *param, struct mlx5e_cq *cq, mlx5e_cq_comp_t *comp, int eq_ix) { struct mlx5_core_dev *mdev = priv->mdev; struct mlx5_core_cq *mcq = &cq->mcq; int eqn_not_used; int irqn; int err; u32 i; - param->wq.buf_numa_node = 0; - param->wq.db_numa_node = 0; - err = mlx5_vector2eqn(mdev, eq_ix, &eqn_not_used, &irqn); if (err) return (err); err = mlx5_cqwq_create(mdev, ¶m->wq, param->cqc, &cq->wq, &cq->wq_ctrl); if (err) return (err); mcq->cqe_sz = 64; mcq->set_ci_db = cq->wq_ctrl.db.db; mcq->arm_db = cq->wq_ctrl.db.db + 1; *mcq->set_ci_db = 0; *mcq->arm_db = 0; mcq->vector = eq_ix; mcq->comp = comp; mcq->event = mlx5e_cq_error_event; mcq->irqn = irqn; for (i = 0; i < mlx5_cqwq_get_size(&cq->wq); i++) { struct mlx5_cqe64 *cqe = mlx5_cqwq_get_wqe(&cq->wq, i); cqe->op_own = 0xf1; } cq->priv = priv; return (0); } static void mlx5e_destroy_cq(struct mlx5e_cq *cq) { mlx5_wq_destroy(&cq->wq_ctrl); } static int mlx5e_enable_cq(struct mlx5e_cq *cq, struct mlx5e_cq_param *param, int eq_ix) { struct mlx5_core_cq *mcq = &cq->mcq; u32 out[MLX5_ST_SZ_DW(create_cq_out)]; void *in; void *cqc; int inlen; int irqn_not_used; int eqn; int err; inlen = MLX5_ST_SZ_BYTES(create_cq_in) + sizeof(u64) * cq->wq_ctrl.buf.npages; in = mlx5_vzalloc(inlen); if (in == NULL) return (-ENOMEM); cqc = MLX5_ADDR_OF(create_cq_in, in, cq_context); memcpy(cqc, param->cqc, sizeof(param->cqc)); mlx5_fill_page_array(&cq->wq_ctrl.buf, (__be64 *) MLX5_ADDR_OF(create_cq_in, in, pas)); mlx5_vector2eqn(cq->priv->mdev, eq_ix, &eqn, &irqn_not_used); MLX5_SET(cqc, cqc, c_eqn, eqn); MLX5_SET(cqc, cqc, log_page_size, cq->wq_ctrl.buf.page_shift - PAGE_SHIFT); MLX5_SET64(cqc, cqc, dbr_addr, cq->wq_ctrl.db.dma); err = mlx5_core_create_cq(cq->priv->mdev, mcq, in, inlen, out, sizeof(out)); kvfree(in); if (err) return (err); mlx5e_cq_arm(cq, MLX5_GET_DOORBELL_LOCK(&cq->priv->doorbell_lock)); return (0); } static void mlx5e_disable_cq(struct mlx5e_cq *cq) { mlx5_core_destroy_cq(cq->priv->mdev, &cq->mcq); } int mlx5e_open_cq(struct mlx5e_priv *priv, struct mlx5e_cq_param *param, struct mlx5e_cq *cq, mlx5e_cq_comp_t *comp, int eq_ix) { int err; err = mlx5e_create_cq(priv, param, cq, comp, eq_ix); if (err) return (err); err = mlx5e_enable_cq(cq, param, eq_ix); if (err) goto err_destroy_cq; return (0); err_destroy_cq: mlx5e_destroy_cq(cq); return (err); } void mlx5e_close_cq(struct mlx5e_cq *cq) { mlx5e_disable_cq(cq); mlx5e_destroy_cq(cq); } static int mlx5e_open_tx_cqs(struct mlx5e_channel *c, struct mlx5e_channel_param *cparam) { int err; int tc; for (tc = 0; tc < c->priv->num_tc; tc++) { /* open completion queue */ err = mlx5e_open_cq(c->priv, &cparam->tx_cq, &c->sq[tc].cq, &mlx5e_tx_cq_comp, c->ix); if (err) goto err_close_tx_cqs; } return (0); err_close_tx_cqs: for (tc--; tc >= 0; tc--) mlx5e_close_cq(&c->sq[tc].cq); return (err); } static void mlx5e_close_tx_cqs(struct mlx5e_channel *c) { int tc; for (tc = 0; tc < c->priv->num_tc; tc++) mlx5e_close_cq(&c->sq[tc].cq); } static int mlx5e_open_sqs(struct mlx5e_channel *c, struct mlx5e_channel_param *cparam) { int err; int tc; for (tc = 0; tc < c->priv->num_tc; tc++) { err = mlx5e_open_sq(c, tc, &cparam->sq, &c->sq[tc]); if (err) goto err_close_sqs; } return (0); err_close_sqs: for (tc--; tc >= 0; tc--) mlx5e_close_sq_wait(&c->sq[tc]); return (err); } static void mlx5e_close_sqs_wait(struct mlx5e_channel *c) { int tc; for (tc = 0; tc < c->priv->num_tc; tc++) mlx5e_close_sq_wait(&c->sq[tc]); } static void mlx5e_chan_static_init(struct mlx5e_priv *priv, struct mlx5e_channel *c, int ix) { int tc; /* setup priv and channel number */ c->priv = priv; c->ix = ix; /* setup send tag */ m_snd_tag_init(&c->tag, c->priv->ifp, IF_SND_TAG_TYPE_UNLIMITED); init_completion(&c->completion); mtx_init(&c->rq.mtx, "mlx5rx", MTX_NETWORK_LOCK, MTX_DEF); callout_init_mtx(&c->rq.watchdog, &c->rq.mtx, 0); for (tc = 0; tc != MLX5E_MAX_TX_NUM_TC; tc++) { struct mlx5e_sq *sq = c->sq + tc; mtx_init(&sq->lock, "mlx5tx", MTX_NETWORK_LOCK " TX", MTX_DEF); mtx_init(&sq->comp_lock, "mlx5comp", MTX_NETWORK_LOCK " TX", MTX_DEF); callout_init_mtx(&sq->cev_callout, &sq->lock, 0); } } static void mlx5e_chan_wait_for_completion(struct mlx5e_channel *c) { m_snd_tag_rele(&c->tag); wait_for_completion(&c->completion); } static void mlx5e_priv_wait_for_completion(struct mlx5e_priv *priv, const uint32_t channels) { uint32_t x; for (x = 0; x != channels; x++) mlx5e_chan_wait_for_completion(&priv->channel[x]); } static void mlx5e_chan_static_destroy(struct mlx5e_channel *c) { int tc; callout_drain(&c->rq.watchdog); mtx_destroy(&c->rq.mtx); for (tc = 0; tc != MLX5E_MAX_TX_NUM_TC; tc++) { callout_drain(&c->sq[tc].cev_callout); mtx_destroy(&c->sq[tc].lock); mtx_destroy(&c->sq[tc].comp_lock); } } static int mlx5e_open_channel(struct mlx5e_priv *priv, struct mlx5e_channel_param *cparam, struct mlx5e_channel *c) { struct epoch_tracker et; int i, err; /* zero non-persistant data */ MLX5E_ZERO(&c->rq, mlx5e_rq_zero_start); for (i = 0; i != priv->num_tc; i++) MLX5E_ZERO(&c->sq[i], mlx5e_sq_zero_start); /* open transmit completion queue */ err = mlx5e_open_tx_cqs(c, cparam); if (err) goto err_free; /* open receive completion queue */ err = mlx5e_open_cq(c->priv, &cparam->rx_cq, &c->rq.cq, &mlx5e_rx_cq_comp, c->ix); if (err) goto err_close_tx_cqs; err = mlx5e_open_sqs(c, cparam); if (err) goto err_close_rx_cq; err = mlx5e_open_rq(c, &cparam->rq, &c->rq); if (err) goto err_close_sqs; /* poll receive queue initially */ NET_EPOCH_ENTER(et); c->rq.cq.mcq.comp(&c->rq.cq.mcq, NULL); NET_EPOCH_EXIT(et); return (0); err_close_sqs: mlx5e_close_sqs_wait(c); err_close_rx_cq: mlx5e_close_cq(&c->rq.cq); err_close_tx_cqs: mlx5e_close_tx_cqs(c); err_free: return (err); } static void mlx5e_close_channel(struct mlx5e_channel *c) { mlx5e_close_rq(&c->rq); } static void mlx5e_close_channel_wait(struct mlx5e_channel *c) { mlx5e_close_rq_wait(&c->rq); mlx5e_close_sqs_wait(c); mlx5e_close_tx_cqs(c); } static int mlx5e_get_wqe_sz(struct mlx5e_priv *priv, u32 *wqe_sz, u32 *nsegs) { u32 r, n; r = priv->params.hw_lro_en ? priv->params.lro_wqe_sz : MLX5E_SW2MB_MTU(priv->ifp->if_mtu); if (r > MJUM16BYTES) return (-ENOMEM); if (r > MJUM9BYTES) r = MJUM16BYTES; else if (r > MJUMPAGESIZE) r = MJUM9BYTES; else if (r > MCLBYTES) r = MJUMPAGESIZE; else r = MCLBYTES; /* * n + 1 must be a power of two, because stride size must be. * Stride size is 16 * (n + 1), as the first segment is * control. */ for (n = howmany(r, MLX5E_MAX_RX_BYTES); !powerof2(n + 1); n++) ; if (n > MLX5E_MAX_BUSDMA_RX_SEGS) return (-ENOMEM); *wqe_sz = r; *nsegs = n; return (0); } static void mlx5e_build_rq_param(struct mlx5e_priv *priv, struct mlx5e_rq_param *param) { void *rqc = param->rqc; void *wq = MLX5_ADDR_OF(rqc, rqc, wq); u32 wqe_sz, nsegs; mlx5e_get_wqe_sz(priv, &wqe_sz, &nsegs); MLX5_SET(wq, wq, wq_type, MLX5_WQ_TYPE_LINKED_LIST); MLX5_SET(wq, wq, end_padding_mode, MLX5_WQ_END_PAD_MODE_ALIGN); MLX5_SET(wq, wq, log_wq_stride, ilog2(sizeof(struct mlx5e_rx_wqe) + nsegs * sizeof(struct mlx5_wqe_data_seg))); MLX5_SET(wq, wq, log_wq_sz, priv->params.log_rq_size); MLX5_SET(wq, wq, pd, priv->pdn); - param->wq.buf_numa_node = 0; - param->wq.db_numa_node = 0; param->wq.linear = 1; } static void mlx5e_build_sq_param(struct mlx5e_priv *priv, struct mlx5e_sq_param *param) { void *sqc = param->sqc; void *wq = MLX5_ADDR_OF(sqc, sqc, wq); MLX5_SET(wq, wq, log_wq_sz, priv->params.log_sq_size); MLX5_SET(wq, wq, log_wq_stride, ilog2(MLX5_SEND_WQE_BB)); MLX5_SET(wq, wq, pd, priv->pdn); - param->wq.buf_numa_node = 0; - param->wq.db_numa_node = 0; param->wq.linear = 1; } static void mlx5e_build_common_cq_param(struct mlx5e_priv *priv, struct mlx5e_cq_param *param) { void *cqc = param->cqc; MLX5_SET(cqc, cqc, uar_page, priv->mdev->priv.uar->index); } static void mlx5e_get_default_profile(struct mlx5e_priv *priv, int mode, struct net_dim_cq_moder *ptr) { *ptr = net_dim_get_profile(mode, MLX5E_DIM_DEFAULT_PROFILE); /* apply LRO restrictions */ if (priv->params.hw_lro_en && ptr->pkts > MLX5E_DIM_MAX_RX_CQ_MODERATION_PKTS_WITH_LRO) { ptr->pkts = MLX5E_DIM_MAX_RX_CQ_MODERATION_PKTS_WITH_LRO; } } static void mlx5e_build_rx_cq_param(struct mlx5e_priv *priv, struct mlx5e_cq_param *param) { struct net_dim_cq_moder curr; void *cqc = param->cqc; /* * We use MLX5_CQE_FORMAT_HASH because the RX hash mini CQE * format is more beneficial for FreeBSD use case. * * Adding support for MLX5_CQE_FORMAT_CSUM will require changes * in mlx5e_decompress_cqe. */ if (priv->params.cqe_zipping_en) { MLX5_SET(cqc, cqc, mini_cqe_res_format, MLX5_CQE_FORMAT_HASH); MLX5_SET(cqc, cqc, cqe_compression_en, 1); } MLX5_SET(cqc, cqc, log_cq_size, priv->params.log_rq_size); switch (priv->params.rx_cq_moderation_mode) { case 0: MLX5_SET(cqc, cqc, cq_period, priv->params.rx_cq_moderation_usec); MLX5_SET(cqc, cqc, cq_max_count, priv->params.rx_cq_moderation_pkts); MLX5_SET(cqc, cqc, cq_period_mode, MLX5_CQ_PERIOD_MODE_START_FROM_EQE); break; case 1: MLX5_SET(cqc, cqc, cq_period, priv->params.rx_cq_moderation_usec); MLX5_SET(cqc, cqc, cq_max_count, priv->params.rx_cq_moderation_pkts); if (MLX5_CAP_GEN(priv->mdev, cq_period_start_from_cqe)) MLX5_SET(cqc, cqc, cq_period_mode, MLX5_CQ_PERIOD_MODE_START_FROM_CQE); else MLX5_SET(cqc, cqc, cq_period_mode, MLX5_CQ_PERIOD_MODE_START_FROM_EQE); break; case 2: mlx5e_get_default_profile(priv, NET_DIM_CQ_PERIOD_MODE_START_FROM_EQE, &curr); MLX5_SET(cqc, cqc, cq_period, curr.usec); MLX5_SET(cqc, cqc, cq_max_count, curr.pkts); MLX5_SET(cqc, cqc, cq_period_mode, MLX5_CQ_PERIOD_MODE_START_FROM_EQE); break; case 3: mlx5e_get_default_profile(priv, NET_DIM_CQ_PERIOD_MODE_START_FROM_CQE, &curr); MLX5_SET(cqc, cqc, cq_period, curr.usec); MLX5_SET(cqc, cqc, cq_max_count, curr.pkts); if (MLX5_CAP_GEN(priv->mdev, cq_period_start_from_cqe)) MLX5_SET(cqc, cqc, cq_period_mode, MLX5_CQ_PERIOD_MODE_START_FROM_CQE); else MLX5_SET(cqc, cqc, cq_period_mode, MLX5_CQ_PERIOD_MODE_START_FROM_EQE); break; default: break; } mlx5e_dim_build_cq_param(priv, param); mlx5e_build_common_cq_param(priv, param); } static void mlx5e_build_tx_cq_param(struct mlx5e_priv *priv, struct mlx5e_cq_param *param) { void *cqc = param->cqc; MLX5_SET(cqc, cqc, log_cq_size, priv->params.log_sq_size); MLX5_SET(cqc, cqc, cq_period, priv->params.tx_cq_moderation_usec); MLX5_SET(cqc, cqc, cq_max_count, priv->params.tx_cq_moderation_pkts); switch (priv->params.tx_cq_moderation_mode) { case 0: MLX5_SET(cqc, cqc, cq_period_mode, MLX5_CQ_PERIOD_MODE_START_FROM_EQE); break; default: if (MLX5_CAP_GEN(priv->mdev, cq_period_start_from_cqe)) MLX5_SET(cqc, cqc, cq_period_mode, MLX5_CQ_PERIOD_MODE_START_FROM_CQE); else MLX5_SET(cqc, cqc, cq_period_mode, MLX5_CQ_PERIOD_MODE_START_FROM_EQE); break; } mlx5e_build_common_cq_param(priv, param); } static void mlx5e_build_channel_param(struct mlx5e_priv *priv, struct mlx5e_channel_param *cparam) { memset(cparam, 0, sizeof(*cparam)); mlx5e_build_rq_param(priv, &cparam->rq); mlx5e_build_sq_param(priv, &cparam->sq); mlx5e_build_rx_cq_param(priv, &cparam->rx_cq); mlx5e_build_tx_cq_param(priv, &cparam->tx_cq); } static int mlx5e_open_channels(struct mlx5e_priv *priv) { struct mlx5e_channel_param *cparam; int err; int i; int j; cparam = malloc(sizeof(*cparam), M_MLX5EN, M_WAITOK); mlx5e_build_channel_param(priv, cparam); for (i = 0; i < priv->params.num_channels; i++) { err = mlx5e_open_channel(priv, cparam, &priv->channel[i]); if (err) goto err_close_channels; /* Bind interrupt vectors, if any. */ if (priv->params_ethtool.irq_cpu_base > -1) { cpuset_t cpuset; int cpu; int irq; int eqn; int nirq; err = mlx5_vector2eqn(priv->mdev, i, &eqn, &nirq); /* error here is non-fatal */ if (err != 0) continue; irq = priv->mdev->priv.msix_arr[nirq].vector; cpu = (unsigned)(priv->params_ethtool.irq_cpu_base + i * priv->params_ethtool.irq_cpu_stride) % (unsigned)mp_ncpus; CPU_ZERO(&cpuset); CPU_SET(cpu, &cpuset); intr_setaffinity(irq, CPU_WHICH_INTRHANDLER, &cpuset); } } for (j = 0; j < priv->params.num_channels; j++) { err = mlx5e_wait_for_min_rx_wqes(&priv->channel[j].rq); if (err) goto err_close_channels; } free(cparam, M_MLX5EN); return (0); err_close_channels: while (i--) { mlx5e_close_channel(&priv->channel[i]); mlx5e_close_channel_wait(&priv->channel[i]); } free(cparam, M_MLX5EN); return (err); } static void mlx5e_close_channels(struct mlx5e_priv *priv) { int i; for (i = 0; i < priv->params.num_channels; i++) mlx5e_close_channel(&priv->channel[i]); for (i = 0; i < priv->params.num_channels; i++) mlx5e_close_channel_wait(&priv->channel[i]); } static int mlx5e_refresh_sq_params(struct mlx5e_priv *priv, struct mlx5e_sq *sq) { if (MLX5_CAP_GEN(priv->mdev, cq_period_mode_modify)) { uint8_t cq_mode; switch (priv->params.tx_cq_moderation_mode) { case 0: case 2: cq_mode = MLX5_CQ_PERIOD_MODE_START_FROM_EQE; break; default: cq_mode = MLX5_CQ_PERIOD_MODE_START_FROM_CQE; break; } return (mlx5_core_modify_cq_moderation_mode(priv->mdev, &sq->cq.mcq, priv->params.tx_cq_moderation_usec, priv->params.tx_cq_moderation_pkts, cq_mode)); } return (mlx5_core_modify_cq_moderation(priv->mdev, &sq->cq.mcq, priv->params.tx_cq_moderation_usec, priv->params.tx_cq_moderation_pkts)); } static int mlx5e_refresh_rq_params(struct mlx5e_priv *priv, struct mlx5e_rq *rq) { if (MLX5_CAP_GEN(priv->mdev, cq_period_mode_modify)) { uint8_t cq_mode; uint8_t dim_mode; int retval; switch (priv->params.rx_cq_moderation_mode) { case 0: case 2: cq_mode = MLX5_CQ_PERIOD_MODE_START_FROM_EQE; dim_mode = NET_DIM_CQ_PERIOD_MODE_START_FROM_EQE; break; default: cq_mode = MLX5_CQ_PERIOD_MODE_START_FROM_CQE; dim_mode = NET_DIM_CQ_PERIOD_MODE_START_FROM_CQE; break; } /* tear down dynamic interrupt moderation */ mtx_lock(&rq->mtx); rq->dim.mode = NET_DIM_CQ_PERIOD_MODE_DISABLED; mtx_unlock(&rq->mtx); /* wait for dynamic interrupt moderation work task, if any */ cancel_work_sync(&rq->dim.work); if (priv->params.rx_cq_moderation_mode >= 2) { struct net_dim_cq_moder curr; mlx5e_get_default_profile(priv, dim_mode, &curr); retval = mlx5_core_modify_cq_moderation_mode(priv->mdev, &rq->cq.mcq, curr.usec, curr.pkts, cq_mode); /* set dynamic interrupt moderation mode and zero defaults */ mtx_lock(&rq->mtx); rq->dim.mode = dim_mode; rq->dim.state = 0; rq->dim.profile_ix = MLX5E_DIM_DEFAULT_PROFILE; mtx_unlock(&rq->mtx); } else { retval = mlx5_core_modify_cq_moderation_mode(priv->mdev, &rq->cq.mcq, priv->params.rx_cq_moderation_usec, priv->params.rx_cq_moderation_pkts, cq_mode); } return (retval); } return (mlx5_core_modify_cq_moderation(priv->mdev, &rq->cq.mcq, priv->params.rx_cq_moderation_usec, priv->params.rx_cq_moderation_pkts)); } static int mlx5e_refresh_channel_params_sub(struct mlx5e_priv *priv, struct mlx5e_channel *c) { int err; int i; err = mlx5e_refresh_rq_params(priv, &c->rq); if (err) goto done; for (i = 0; i != priv->num_tc; i++) { err = mlx5e_refresh_sq_params(priv, &c->sq[i]); if (err) goto done; } done: return (err); } int mlx5e_refresh_channel_params(struct mlx5e_priv *priv) { int i; /* check if channels are closed */ if (test_bit(MLX5E_STATE_OPENED, &priv->state) == 0) return (EINVAL); for (i = 0; i < priv->params.num_channels; i++) { int err; err = mlx5e_refresh_channel_params_sub(priv, &priv->channel[i]); if (err) return (err); } return (0); } static int mlx5e_open_tis(struct mlx5e_priv *priv, int tc) { struct mlx5_core_dev *mdev = priv->mdev; u32 in[MLX5_ST_SZ_DW(create_tis_in)]; void *tisc = MLX5_ADDR_OF(create_tis_in, in, ctx); memset(in, 0, sizeof(in)); MLX5_SET(tisc, tisc, prio, tc); MLX5_SET(tisc, tisc, transport_domain, priv->tdn); return (mlx5_core_create_tis(mdev, in, sizeof(in), &priv->tisn[tc])); } static void mlx5e_close_tis(struct mlx5e_priv *priv, int tc) { mlx5_core_destroy_tis(priv->mdev, priv->tisn[tc]); } static int mlx5e_open_tises(struct mlx5e_priv *priv) { int num_tc = priv->num_tc; int err; int tc; for (tc = 0; tc < num_tc; tc++) { err = mlx5e_open_tis(priv, tc); if (err) goto err_close_tises; } return (0); err_close_tises: for (tc--; tc >= 0; tc--) mlx5e_close_tis(priv, tc); return (err); } static void mlx5e_close_tises(struct mlx5e_priv *priv) { int num_tc = priv->num_tc; int tc; for (tc = 0; tc < num_tc; tc++) mlx5e_close_tis(priv, tc); } static int mlx5e_open_rqt(struct mlx5e_priv *priv) { struct mlx5_core_dev *mdev = priv->mdev; u32 *in; u32 out[MLX5_ST_SZ_DW(create_rqt_out)] = {0}; void *rqtc; int inlen; int err; int sz; int i; sz = 1 << priv->params.rx_hash_log_tbl_sz; inlen = MLX5_ST_SZ_BYTES(create_rqt_in) + sizeof(u32) * sz; in = mlx5_vzalloc(inlen); if (in == NULL) return (-ENOMEM); rqtc = MLX5_ADDR_OF(create_rqt_in, in, rqt_context); MLX5_SET(rqtc, rqtc, rqt_actual_size, sz); MLX5_SET(rqtc, rqtc, rqt_max_size, sz); for (i = 0; i < sz; i++) { int ix = i; #ifdef RSS ix = rss_get_indirection_to_bucket(ix); #endif /* ensure we don't overflow */ ix %= priv->params.num_channels; /* apply receive side scaling stride, if any */ ix -= ix % (int)priv->params.channels_rsss; MLX5_SET(rqtc, rqtc, rq_num[i], priv->channel[ix].rq.rqn); } MLX5_SET(create_rqt_in, in, opcode, MLX5_CMD_OP_CREATE_RQT); err = mlx5_cmd_exec(mdev, in, inlen, out, sizeof(out)); if (!err) priv->rqtn = MLX5_GET(create_rqt_out, out, rqtn); kvfree(in); return (err); } static void mlx5e_close_rqt(struct mlx5e_priv *priv) { u32 in[MLX5_ST_SZ_DW(destroy_rqt_in)] = {0}; u32 out[MLX5_ST_SZ_DW(destroy_rqt_out)] = {0}; MLX5_SET(destroy_rqt_in, in, opcode, MLX5_CMD_OP_DESTROY_RQT); MLX5_SET(destroy_rqt_in, in, rqtn, priv->rqtn); mlx5_cmd_exec(priv->mdev, in, sizeof(in), out, sizeof(out)); } #define MLX5E_RSS_KEY_SIZE (10 * 4) /* bytes */ static void mlx5e_get_rss_key(void *key_ptr) { #ifdef RSS rss_getkey(key_ptr); #else static const u32 rsskey[] = { cpu_to_be32(0xD181C62C), cpu_to_be32(0xF7F4DB5B), cpu_to_be32(0x1983A2FC), cpu_to_be32(0x943E1ADB), cpu_to_be32(0xD9389E6B), cpu_to_be32(0xD1039C2C), cpu_to_be32(0xA74499AD), cpu_to_be32(0x593D56D9), cpu_to_be32(0xF3253C06), cpu_to_be32(0x2ADC1FFC), }; CTASSERT(sizeof(rsskey) == MLX5E_RSS_KEY_SIZE); memcpy(key_ptr, rsskey, MLX5E_RSS_KEY_SIZE); #endif } static void mlx5e_build_tir_ctx(struct mlx5e_priv *priv, u32 * tirc, int tt, bool inner_vxlan) { void *hfso = MLX5_ADDR_OF(tirc, tirc, rx_hash_field_selector_outer); void *hfsi = MLX5_ADDR_OF(tirc, tirc, rx_hash_field_selector_inner); void *hfs = inner_vxlan ? hfsi : hfso; __be32 *hkey; MLX5_SET(tirc, tirc, transport_domain, priv->tdn); #define ROUGH_MAX_L2_L3_HDR_SZ 256 #define MLX5_HASH_IP (MLX5_HASH_FIELD_SEL_SRC_IP |\ MLX5_HASH_FIELD_SEL_DST_IP) #define MLX5_HASH_ALL (MLX5_HASH_FIELD_SEL_SRC_IP |\ MLX5_HASH_FIELD_SEL_DST_IP |\ MLX5_HASH_FIELD_SEL_L4_SPORT |\ MLX5_HASH_FIELD_SEL_L4_DPORT) #define MLX5_HASH_IP_IPSEC_SPI (MLX5_HASH_FIELD_SEL_SRC_IP |\ MLX5_HASH_FIELD_SEL_DST_IP |\ MLX5_HASH_FIELD_SEL_IPSEC_SPI) if (priv->params.hw_lro_en) { MLX5_SET(tirc, tirc, lro_enable_mask, MLX5_TIRC_LRO_ENABLE_MASK_IPV4_LRO | MLX5_TIRC_LRO_ENABLE_MASK_IPV6_LRO); MLX5_SET(tirc, tirc, lro_max_msg_sz, (priv->params.lro_wqe_sz - ROUGH_MAX_L2_L3_HDR_SZ) >> 8); /* TODO: add the option to choose timer value dynamically */ MLX5_SET(tirc, tirc, lro_timeout_period_usecs, MLX5_CAP_ETH(priv->mdev, lro_timer_supported_periods[2])); } if (inner_vxlan) MLX5_SET(tirc, tirc, tunneled_offload_en, 1); /* setup parameters for hashing TIR type, if any */ switch (tt) { case MLX5E_TT_ANY: MLX5_SET(tirc, tirc, disp_type, MLX5_TIRC_DISP_TYPE_DIRECT); MLX5_SET(tirc, tirc, inline_rqn, priv->channel[0].rq.rqn); break; default: MLX5_SET(tirc, tirc, disp_type, MLX5_TIRC_DISP_TYPE_INDIRECT); MLX5_SET(tirc, tirc, indirect_table, priv->rqtn); MLX5_SET(tirc, tirc, rx_hash_fn, MLX5_TIRC_RX_HASH_FN_HASH_TOEPLITZ); hkey = (__be32 *) MLX5_ADDR_OF(tirc, tirc, rx_hash_toeplitz_key); CTASSERT(MLX5_FLD_SZ_BYTES(tirc, rx_hash_toeplitz_key) >= MLX5E_RSS_KEY_SIZE); #ifdef RSS /* * The FreeBSD RSS implementation does currently not * support symmetric Toeplitz hashes: */ MLX5_SET(tirc, tirc, rx_hash_symmetric, 0); #else MLX5_SET(tirc, tirc, rx_hash_symmetric, 1); #endif mlx5e_get_rss_key(hkey); break; } switch (tt) { case MLX5E_TT_IPV4_TCP: MLX5_SET(rx_hash_field_select, hfs, l3_prot_type, MLX5_L3_PROT_TYPE_IPV4); MLX5_SET(rx_hash_field_select, hfs, l4_prot_type, MLX5_L4_PROT_TYPE_TCP); #ifdef RSS if (!(rss_gethashconfig() & RSS_HASHTYPE_RSS_TCP_IPV4)) { MLX5_SET(rx_hash_field_select, hfs, selected_fields, MLX5_HASH_IP); } else #endif MLX5_SET(rx_hash_field_select, hfs, selected_fields, MLX5_HASH_ALL); break; case MLX5E_TT_IPV6_TCP: MLX5_SET(rx_hash_field_select, hfs, l3_prot_type, MLX5_L3_PROT_TYPE_IPV6); MLX5_SET(rx_hash_field_select, hfs, l4_prot_type, MLX5_L4_PROT_TYPE_TCP); #ifdef RSS if (!(rss_gethashconfig() & RSS_HASHTYPE_RSS_TCP_IPV6)) { MLX5_SET(rx_hash_field_select, hfs, selected_fields, MLX5_HASH_IP); } else #endif MLX5_SET(rx_hash_field_select, hfs, selected_fields, MLX5_HASH_ALL); break; case MLX5E_TT_IPV4_UDP: MLX5_SET(rx_hash_field_select, hfs, l3_prot_type, MLX5_L3_PROT_TYPE_IPV4); MLX5_SET(rx_hash_field_select, hfs, l4_prot_type, MLX5_L4_PROT_TYPE_UDP); #ifdef RSS if (!(rss_gethashconfig() & RSS_HASHTYPE_RSS_UDP_IPV4)) { MLX5_SET(rx_hash_field_select, hfs, selected_fields, MLX5_HASH_IP); } else #endif MLX5_SET(rx_hash_field_select, hfs, selected_fields, MLX5_HASH_ALL); break; case MLX5E_TT_IPV6_UDP: MLX5_SET(rx_hash_field_select, hfs, l3_prot_type, MLX5_L3_PROT_TYPE_IPV6); MLX5_SET(rx_hash_field_select, hfs, l4_prot_type, MLX5_L4_PROT_TYPE_UDP); #ifdef RSS if (!(rss_gethashconfig() & RSS_HASHTYPE_RSS_UDP_IPV6)) { MLX5_SET(rx_hash_field_select, hfs, selected_fields, MLX5_HASH_IP); } else #endif MLX5_SET(rx_hash_field_select, hfs, selected_fields, MLX5_HASH_ALL); break; case MLX5E_TT_IPV4_IPSEC_AH: MLX5_SET(rx_hash_field_select, hfs, l3_prot_type, MLX5_L3_PROT_TYPE_IPV4); MLX5_SET(rx_hash_field_select, hfs, selected_fields, MLX5_HASH_IP_IPSEC_SPI); break; case MLX5E_TT_IPV6_IPSEC_AH: MLX5_SET(rx_hash_field_select, hfs, l3_prot_type, MLX5_L3_PROT_TYPE_IPV6); MLX5_SET(rx_hash_field_select, hfs, selected_fields, MLX5_HASH_IP_IPSEC_SPI); break; case MLX5E_TT_IPV4_IPSEC_ESP: MLX5_SET(rx_hash_field_select, hfs, l3_prot_type, MLX5_L3_PROT_TYPE_IPV4); MLX5_SET(rx_hash_field_select, hfs, selected_fields, MLX5_HASH_IP_IPSEC_SPI); break; case MLX5E_TT_IPV6_IPSEC_ESP: MLX5_SET(rx_hash_field_select, hfs, l3_prot_type, MLX5_L3_PROT_TYPE_IPV6); MLX5_SET(rx_hash_field_select, hfs, selected_fields, MLX5_HASH_IP_IPSEC_SPI); break; case MLX5E_TT_IPV4: MLX5_SET(rx_hash_field_select, hfs, l3_prot_type, MLX5_L3_PROT_TYPE_IPV4); MLX5_SET(rx_hash_field_select, hfs, selected_fields, MLX5_HASH_IP); break; case MLX5E_TT_IPV6: MLX5_SET(rx_hash_field_select, hfs, l3_prot_type, MLX5_L3_PROT_TYPE_IPV6); MLX5_SET(rx_hash_field_select, hfs, selected_fields, MLX5_HASH_IP); break; default: break; } } static int mlx5e_open_tir(struct mlx5e_priv *priv, int tt, bool inner_vxlan) { struct mlx5_core_dev *mdev = priv->mdev; u32 *in; void *tirc; int inlen; int err; inlen = MLX5_ST_SZ_BYTES(create_tir_in); in = mlx5_vzalloc(inlen); if (in == NULL) return (-ENOMEM); tirc = MLX5_ADDR_OF(create_tir_in, in, tir_context); mlx5e_build_tir_ctx(priv, tirc, tt, inner_vxlan); err = mlx5_core_create_tir(mdev, in, inlen, inner_vxlan ? &priv->tirn_inner_vxlan[tt] : &priv->tirn[tt]); kvfree(in); return (err); } static void mlx5e_close_tir(struct mlx5e_priv *priv, int tt, bool inner_vxlan) { mlx5_core_destroy_tir(priv->mdev, inner_vxlan ? priv->tirn_inner_vxlan[tt] : priv->tirn[tt]); } static int mlx5e_open_tirs(struct mlx5e_priv *priv, bool inner_vxlan) { int err; int i; for (i = 0; i < MLX5E_NUM_TT; i++) { err = mlx5e_open_tir(priv, i, inner_vxlan); if (err) goto err_close_tirs; } return (0); err_close_tirs: for (i--; i >= 0; i--) mlx5e_close_tir(priv, i, inner_vxlan); return (err); } static void mlx5e_close_tirs(struct mlx5e_priv *priv, bool inner_vxlan) { int i; for (i = 0; i < MLX5E_NUM_TT; i++) mlx5e_close_tir(priv, i, inner_vxlan); } /* * SW MTU does not include headers, * HW MTU includes all headers and checksums. */ static int mlx5e_set_dev_port_mtu(struct ifnet *ifp, int sw_mtu) { struct mlx5e_priv *priv = ifp->if_softc; struct mlx5_core_dev *mdev = priv->mdev; int hw_mtu; int err; hw_mtu = MLX5E_SW2HW_MTU(sw_mtu); err = mlx5_set_port_mtu(mdev, hw_mtu); if (err) { mlx5_en_err(ifp, "mlx5_set_port_mtu failed setting %d, err=%d\n", sw_mtu, err); return (err); } /* Update vport context MTU */ err = mlx5_set_vport_mtu(mdev, hw_mtu); if (err) { mlx5_en_err(ifp, "Failed updating vport context with MTU size, err=%d\n", err); } ifp->if_mtu = sw_mtu; err = mlx5_query_vport_mtu(mdev, &hw_mtu); if (err || !hw_mtu) { /* fallback to port oper mtu */ err = mlx5_query_port_oper_mtu(mdev, &hw_mtu); } if (err) { mlx5_en_err(ifp, "Query port MTU, after setting new MTU value, failed\n"); return (err); } else if (MLX5E_HW2SW_MTU(hw_mtu) < sw_mtu) { err = -E2BIG, mlx5_en_err(ifp, "Port MTU %d is smaller than ifp mtu %d\n", hw_mtu, sw_mtu); } else if (MLX5E_HW2SW_MTU(hw_mtu) > sw_mtu) { err = -EINVAL; mlx5_en_err(ifp, "Port MTU %d is bigger than ifp mtu %d\n", hw_mtu, sw_mtu); } priv->params_ethtool.hw_mtu = hw_mtu; /* compute MSB */ while (hw_mtu & (hw_mtu - 1)) hw_mtu &= (hw_mtu - 1); priv->params_ethtool.hw_mtu_msb = hw_mtu; return (err); } int mlx5e_open_locked(struct ifnet *ifp) { struct mlx5e_priv *priv = ifp->if_softc; int err; u16 set_id; /* check if already opened */ if (test_bit(MLX5E_STATE_OPENED, &priv->state) != 0) return (0); #ifdef RSS if (rss_getnumbuckets() > priv->params.num_channels) { mlx5_en_info(ifp, "NOTE: There are more RSS buckets(%u) than channels(%u) available\n", rss_getnumbuckets(), priv->params.num_channels); } #endif err = mlx5e_open_tises(priv); if (err) { mlx5_en_err(ifp, "mlx5e_open_tises failed, %d\n", err); return (err); } err = mlx5_vport_alloc_q_counter(priv->mdev, MLX5_INTERFACE_PROTOCOL_ETH, &set_id); if (err) { mlx5_en_err(priv->ifp, "mlx5_vport_alloc_q_counter failed: %d\n", err); goto err_close_tises; } /* store counter set ID */ priv->counter_set_id = set_id; err = mlx5e_open_channels(priv); if (err) { mlx5_en_err(ifp, "mlx5e_open_channels failed, %d\n", err); goto err_dalloc_q_counter; } err = mlx5e_open_rqt(priv); if (err) { mlx5_en_err(ifp, "mlx5e_open_rqt failed, %d\n", err); goto err_close_channels; } err = mlx5e_open_tirs(priv, false); if (err) { mlx5_en_err(ifp, "mlx5e_open_tir(main) failed, %d\n", err); goto err_close_rqls; } if ((ifp->if_capenable & IFCAP_VXLAN_HWCSUM) != 0) { err = mlx5e_open_tirs(priv, true); if (err) { mlx5_en_err(ifp, "mlx5e_open_tir(inner) failed, %d\n", err); goto err_close_tirs; } } err = mlx5e_open_flow_table(priv); if (err) { mlx5_en_err(ifp, "mlx5e_open_flow_table failed, %d\n", err); goto err_close_tirs_inner; } err = mlx5e_add_all_vlan_rules(priv); if (err) { mlx5_en_err(ifp, "mlx5e_add_all_vlan_rules failed, %d\n", err); goto err_close_flow_table; } if ((ifp->if_capenable & IFCAP_VXLAN_HWCSUM) != 0) { err = mlx5e_add_all_vxlan_rules(priv); if (err) { mlx5_en_err(ifp, "mlx5e_add_all_vxlan_rules failed, %d\n", err); goto err_del_vlan_rules; } } set_bit(MLX5E_STATE_OPENED, &priv->state); mlx5e_update_carrier(priv); mlx5e_set_rx_mode_core(priv); return (0); err_del_vlan_rules: mlx5e_del_all_vlan_rules(priv); err_close_flow_table: mlx5e_close_flow_table(priv); err_close_tirs_inner: if ((ifp->if_capenable & IFCAP_VXLAN_HWCSUM) != 0) mlx5e_close_tirs(priv, true); err_close_tirs: mlx5e_close_tirs(priv, false); err_close_rqls: mlx5e_close_rqt(priv); err_close_channels: mlx5e_close_channels(priv); err_dalloc_q_counter: mlx5_vport_dealloc_q_counter(priv->mdev, MLX5_INTERFACE_PROTOCOL_ETH, priv->counter_set_id); err_close_tises: mlx5e_close_tises(priv); return (err); } static void mlx5e_open(void *arg) { struct mlx5e_priv *priv = arg; PRIV_LOCK(priv); if (mlx5_set_port_status(priv->mdev, MLX5_PORT_UP)) mlx5_en_err(priv->ifp, "Setting port status to up failed\n"); mlx5e_open_locked(priv->ifp); priv->ifp->if_drv_flags |= IFF_DRV_RUNNING; PRIV_UNLOCK(priv); } int mlx5e_close_locked(struct ifnet *ifp) { struct mlx5e_priv *priv = ifp->if_softc; /* check if already closed */ if (test_bit(MLX5E_STATE_OPENED, &priv->state) == 0) return (0); clear_bit(MLX5E_STATE_OPENED, &priv->state); mlx5e_set_rx_mode_core(priv); mlx5e_del_all_vlan_rules(priv); if ((ifp->if_capenable & IFCAP_VXLAN_HWCSUM) != 0) mlx5e_del_all_vxlan_rules(priv); if_link_state_change(priv->ifp, LINK_STATE_DOWN); mlx5e_close_flow_table(priv); if ((ifp->if_capenable & IFCAP_VXLAN_HWCSUM) != 0) mlx5e_close_tirs(priv, true); mlx5e_close_tirs(priv, false); mlx5e_close_rqt(priv); mlx5e_close_channels(priv); mlx5_vport_dealloc_q_counter(priv->mdev, MLX5_INTERFACE_PROTOCOL_ETH, priv->counter_set_id); mlx5e_close_tises(priv); return (0); } #if (__FreeBSD_version >= 1100000) static uint64_t mlx5e_get_counter(struct ifnet *ifp, ift_counter cnt) { struct mlx5e_priv *priv = ifp->if_softc; u64 retval; /* PRIV_LOCK(priv); XXX not allowed */ switch (cnt) { case IFCOUNTER_IPACKETS: retval = priv->stats.vport.rx_packets; break; case IFCOUNTER_IERRORS: retval = priv->stats.pport.in_range_len_errors + priv->stats.pport.out_of_range_len + priv->stats.pport.too_long_errors + priv->stats.pport.check_seq_err + priv->stats.pport.alignment_err; break; case IFCOUNTER_IQDROPS: retval = priv->stats.vport.rx_out_of_buffer; break; case IFCOUNTER_OPACKETS: retval = priv->stats.vport.tx_packets; break; case IFCOUNTER_OERRORS: retval = priv->stats.port_stats_debug.out_discards; break; case IFCOUNTER_IBYTES: retval = priv->stats.vport.rx_bytes; break; case IFCOUNTER_OBYTES: retval = priv->stats.vport.tx_bytes; break; case IFCOUNTER_IMCASTS: retval = priv->stats.vport.rx_multicast_packets; break; case IFCOUNTER_OMCASTS: retval = priv->stats.vport.tx_multicast_packets; break; case IFCOUNTER_OQDROPS: retval = priv->stats.vport.tx_queue_dropped; break; case IFCOUNTER_COLLISIONS: retval = priv->stats.pport.collisions; break; default: retval = if_get_counter_default(ifp, cnt); break; } /* PRIV_UNLOCK(priv); XXX not allowed */ return (retval); } #endif static void mlx5e_set_rx_mode(struct ifnet *ifp) { struct mlx5e_priv *priv = ifp->if_softc; queue_work(priv->wq, &priv->set_rx_mode_work); } static int mlx5e_ioctl(struct ifnet *ifp, u_long command, caddr_t data) { struct mlx5e_priv *priv; struct ifreq *ifr; struct ifdownreason *ifdr; struct ifi2creq i2c; struct ifrsskey *ifrk; struct ifrsshash *ifrh; int error = 0; int mask = 0; int size_read = 0; int module_status; int module_num; int max_mtu; uint8_t read_addr; priv = ifp->if_softc; /* check if detaching */ if (priv == NULL || priv->gone != 0) return (ENXIO); switch (command) { case SIOCSIFMTU: ifr = (struct ifreq *)data; PRIV_LOCK(priv); mlx5_query_port_max_mtu(priv->mdev, &max_mtu); if (ifr->ifr_mtu >= MLX5E_MTU_MIN && ifr->ifr_mtu <= MIN(MLX5E_MTU_MAX, max_mtu)) { int was_opened; was_opened = test_bit(MLX5E_STATE_OPENED, &priv->state); if (was_opened) mlx5e_close_locked(ifp); /* set new MTU */ mlx5e_set_dev_port_mtu(ifp, ifr->ifr_mtu); if (was_opened) mlx5e_open_locked(ifp); } else { error = EINVAL; mlx5_en_err(ifp, "Invalid MTU value. Min val: %d, Max val: %d\n", MLX5E_MTU_MIN, MIN(MLX5E_MTU_MAX, max_mtu)); } PRIV_UNLOCK(priv); break; case SIOCSIFFLAGS: if ((ifp->if_flags & IFF_UP) && (ifp->if_drv_flags & IFF_DRV_RUNNING)) { mlx5e_set_rx_mode(ifp); break; } PRIV_LOCK(priv); if (ifp->if_flags & IFF_UP) { if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0) { if (test_bit(MLX5E_STATE_OPENED, &priv->state) == 0) mlx5e_open_locked(ifp); ifp->if_drv_flags |= IFF_DRV_RUNNING; mlx5_set_port_status(priv->mdev, MLX5_PORT_UP); } } else { if (ifp->if_drv_flags & IFF_DRV_RUNNING) { mlx5_set_port_status(priv->mdev, MLX5_PORT_DOWN); if (test_bit(MLX5E_STATE_OPENED, &priv->state) != 0) mlx5e_close_locked(ifp); mlx5e_update_carrier(priv); ifp->if_drv_flags &= ~IFF_DRV_RUNNING; } } PRIV_UNLOCK(priv); break; case SIOCADDMULTI: case SIOCDELMULTI: mlx5e_set_rx_mode(ifp); break; case SIOCSIFMEDIA: case SIOCGIFMEDIA: case SIOCGIFXMEDIA: ifr = (struct ifreq *)data; error = ifmedia_ioctl(ifp, ifr, &priv->media, command); break; case SIOCSIFCAP: ifr = (struct ifreq *)data; PRIV_LOCK(priv); mask = ifr->ifr_reqcap ^ ifp->if_capenable; if (mask & IFCAP_TXCSUM) { ifp->if_capenable ^= IFCAP_TXCSUM; ifp->if_hwassist ^= (CSUM_TCP | CSUM_UDP | CSUM_IP); if (IFCAP_TSO4 & ifp->if_capenable && !(IFCAP_TXCSUM & ifp->if_capenable)) { mask &= ~IFCAP_TSO4; ifp->if_capenable &= ~IFCAP_TSO4; ifp->if_hwassist &= ~CSUM_IP_TSO; mlx5_en_err(ifp, "tso4 disabled due to -txcsum.\n"); } } if (mask & IFCAP_TXCSUM_IPV6) { ifp->if_capenable ^= IFCAP_TXCSUM_IPV6; ifp->if_hwassist ^= (CSUM_UDP_IPV6 | CSUM_TCP_IPV6); if (IFCAP_TSO6 & ifp->if_capenable && !(IFCAP_TXCSUM_IPV6 & ifp->if_capenable)) { mask &= ~IFCAP_TSO6; ifp->if_capenable &= ~IFCAP_TSO6; ifp->if_hwassist &= ~CSUM_IP6_TSO; mlx5_en_err(ifp, "tso6 disabled due to -txcsum6.\n"); } } if (mask & IFCAP_MEXTPG) ifp->if_capenable ^= IFCAP_MEXTPG; if (mask & IFCAP_TXTLS4) ifp->if_capenable ^= IFCAP_TXTLS4; if (mask & IFCAP_TXTLS6) ifp->if_capenable ^= IFCAP_TXTLS6; #ifdef RATELIMIT if (mask & IFCAP_TXTLS_RTLMT) ifp->if_capenable ^= IFCAP_TXTLS_RTLMT; #endif if (mask & IFCAP_RXCSUM) ifp->if_capenable ^= IFCAP_RXCSUM; if (mask & IFCAP_RXCSUM_IPV6) ifp->if_capenable ^= IFCAP_RXCSUM_IPV6; if (mask & IFCAP_TSO4) { if (!(IFCAP_TSO4 & ifp->if_capenable) && !(IFCAP_TXCSUM & ifp->if_capenable)) { mlx5_en_err(ifp, "enable txcsum first.\n"); error = EAGAIN; goto out; } ifp->if_capenable ^= IFCAP_TSO4; ifp->if_hwassist ^= CSUM_IP_TSO; } if (mask & IFCAP_TSO6) { if (!(IFCAP_TSO6 & ifp->if_capenable) && !(IFCAP_TXCSUM_IPV6 & ifp->if_capenable)) { mlx5_en_err(ifp, "enable txcsum6 first.\n"); error = EAGAIN; goto out; } ifp->if_capenable ^= IFCAP_TSO6; ifp->if_hwassist ^= CSUM_IP6_TSO; } if (mask & IFCAP_VLAN_HWTSO) ifp->if_capenable ^= IFCAP_VLAN_HWTSO; if (mask & IFCAP_VLAN_HWFILTER) { if (ifp->if_capenable & IFCAP_VLAN_HWFILTER) mlx5e_disable_vlan_filter(priv); else mlx5e_enable_vlan_filter(priv); ifp->if_capenable ^= IFCAP_VLAN_HWFILTER; } if (mask & IFCAP_VLAN_HWTAGGING) ifp->if_capenable ^= IFCAP_VLAN_HWTAGGING; if (mask & IFCAP_WOL_MAGIC) ifp->if_capenable ^= IFCAP_WOL_MAGIC; if (mask & IFCAP_VXLAN_HWCSUM) { int was_opened = test_bit(MLX5E_STATE_OPENED, &priv->state); if (was_opened) mlx5e_close_locked(ifp); ifp->if_capenable ^= IFCAP_VXLAN_HWCSUM; ifp->if_hwassist ^= CSUM_INNER_IP | CSUM_INNER_IP_UDP | CSUM_INNER_IP_TCP | CSUM_INNER_IP6_UDP | CSUM_INNER_IP6_TCP; if (was_opened) mlx5e_open_locked(ifp); } if (mask & IFCAP_VXLAN_HWTSO) { ifp->if_capenable ^= IFCAP_VXLAN_HWTSO; ifp->if_hwassist ^= CSUM_INNER_IP_TSO | CSUM_INNER_IP6_TSO; } VLAN_CAPABILITIES(ifp); /* turn off LRO means also turn of HW LRO - if it's on */ if (mask & IFCAP_LRO) { int was_opened = test_bit(MLX5E_STATE_OPENED, &priv->state); bool need_restart = false; ifp->if_capenable ^= IFCAP_LRO; /* figure out if updating HW LRO is needed */ if (!(ifp->if_capenable & IFCAP_LRO)) { if (priv->params.hw_lro_en) { priv->params.hw_lro_en = false; need_restart = true; } } else { if (priv->params.hw_lro_en == false && priv->params_ethtool.hw_lro != 0) { priv->params.hw_lro_en = true; need_restart = true; } } if (was_opened && need_restart) { mlx5e_close_locked(ifp); mlx5e_open_locked(ifp); } } if (mask & IFCAP_HWRXTSTMP) { ifp->if_capenable ^= IFCAP_HWRXTSTMP; if (ifp->if_capenable & IFCAP_HWRXTSTMP) { if (priv->clbr_done == 0) mlx5e_reset_calibration_callout(priv); } else { callout_drain(&priv->tstmp_clbr); priv->clbr_done = 0; } } out: PRIV_UNLOCK(priv); break; case SIOCGI2C: ifr = (struct ifreq *)data; /* * Copy from the user-space address ifr_data to the * kernel-space address i2c */ error = copyin(ifr_data_get_ptr(ifr), &i2c, sizeof(i2c)); if (error) break; if (i2c.len > sizeof(i2c.data)) { error = EINVAL; break; } PRIV_LOCK(priv); /* Get module_num which is required for the query_eeprom */ error = mlx5_query_module_num(priv->mdev, &module_num); if (error) { mlx5_en_err(ifp, "Query module num failed, eeprom reading is not supported\n"); error = EINVAL; goto err_i2c; } /* Check if module is present before doing an access */ module_status = mlx5_query_module_status(priv->mdev, module_num); if (module_status != MLX5_MODULE_STATUS_PLUGGED_ENABLED) { error = EINVAL; goto err_i2c; } /* * Currently 0XA0 and 0xA2 are the only addresses permitted. * The internal conversion is as follows: */ if (i2c.dev_addr == 0xA0) read_addr = MLX5_I2C_ADDR_LOW; else if (i2c.dev_addr == 0xA2) read_addr = MLX5_I2C_ADDR_HIGH; else { mlx5_en_err(ifp, "Query eeprom failed, Invalid Address: %X\n", i2c.dev_addr); error = EINVAL; goto err_i2c; } error = mlx5_query_eeprom(priv->mdev, read_addr, MLX5_EEPROM_LOW_PAGE, (uint32_t)i2c.offset, (uint32_t)i2c.len, module_num, (uint32_t *)i2c.data, &size_read); if (error) { mlx5_en_err(ifp, "Query eeprom failed, eeprom reading is not supported\n"); error = EINVAL; goto err_i2c; } if (i2c.len > MLX5_EEPROM_MAX_BYTES) { error = mlx5_query_eeprom(priv->mdev, read_addr, MLX5_EEPROM_LOW_PAGE, (uint32_t)(i2c.offset + size_read), (uint32_t)(i2c.len - size_read), module_num, (uint32_t *)(i2c.data + size_read), &size_read); } if (error) { mlx5_en_err(ifp, "Query eeprom failed, eeprom reading is not supported\n"); error = EINVAL; goto err_i2c; } error = copyout(&i2c, ifr_data_get_ptr(ifr), sizeof(i2c)); err_i2c: PRIV_UNLOCK(priv); break; case SIOCGIFDOWNREASON: ifdr = (struct ifdownreason *)data; bzero(ifdr->ifdr_msg, sizeof(ifdr->ifdr_msg)); PRIV_LOCK(priv); error = -mlx5_query_pddr_troubleshooting_info(priv->mdev, NULL, ifdr->ifdr_msg, sizeof(ifdr->ifdr_msg)); PRIV_UNLOCK(priv); if (error == 0) ifdr->ifdr_reason = IFDR_REASON_MSG; break; case SIOCGIFRSSKEY: ifrk = (struct ifrsskey *)data; ifrk->ifrk_func = RSS_FUNC_TOEPLITZ; ifrk->ifrk_keylen = MLX5E_RSS_KEY_SIZE; CTASSERT(sizeof(ifrk->ifrk_key) >= MLX5E_RSS_KEY_SIZE); mlx5e_get_rss_key(ifrk->ifrk_key); break; case SIOCGIFRSSHASH: ifrh = (struct ifrsshash *)data; ifrh->ifrh_func = RSS_FUNC_TOEPLITZ; ifrh->ifrh_types = RSS_TYPE_IPV4 | RSS_TYPE_TCP_IPV4 | RSS_TYPE_UDP_IPV4 | RSS_TYPE_IPV6 | RSS_TYPE_TCP_IPV6 | RSS_TYPE_UDP_IPV6; break; default: error = ether_ioctl(ifp, command, data); break; } return (error); } static int mlx5e_check_required_hca_cap(struct mlx5_core_dev *mdev) { /* * TODO: uncoment once FW really sets all these bits if * (!mdev->caps.eth.rss_ind_tbl_cap || !mdev->caps.eth.csum_cap || * !mdev->caps.eth.max_lso_cap || !mdev->caps.eth.vlan_cap || * !(mdev->caps.gen.flags & MLX5_DEV_CAP_FLAG_SCQE_BRK_MOD)) return * -ENOTSUPP; */ /* TODO: add more must-to-have features */ if (MLX5_CAP_GEN(mdev, port_type) != MLX5_CAP_PORT_TYPE_ETH) return (-ENODEV); return (0); } static u16 mlx5e_get_max_inline_cap(struct mlx5_core_dev *mdev) { const int min_size = ETHER_VLAN_ENCAP_LEN + ETHER_HDR_LEN; const int max_size = MLX5E_MAX_TX_INLINE; const int bf_buf_size = ((1U << MLX5_CAP_GEN(mdev, log_bf_reg_size)) / 2U) - (sizeof(struct mlx5e_tx_wqe) - 2); /* verify against driver limits */ if (bf_buf_size > max_size) return (max_size); else if (bf_buf_size < min_size) return (min_size); else return (bf_buf_size); } static int mlx5e_build_ifp_priv(struct mlx5_core_dev *mdev, struct mlx5e_priv *priv, int num_comp_vectors) { int err; /* * TODO: Consider link speed for setting "log_sq_size", * "log_rq_size" and "cq_moderation_xxx": */ priv->params.log_sq_size = MLX5E_PARAMS_DEFAULT_LOG_SQ_SIZE; priv->params.log_rq_size = MLX5E_PARAMS_DEFAULT_LOG_RQ_SIZE; priv->params.rx_cq_moderation_usec = MLX5_CAP_GEN(mdev, cq_period_start_from_cqe) ? MLX5E_PARAMS_DEFAULT_RX_CQ_MODERATION_USEC_FROM_CQE : MLX5E_PARAMS_DEFAULT_RX_CQ_MODERATION_USEC; priv->params.rx_cq_moderation_mode = MLX5_CAP_GEN(mdev, cq_period_start_from_cqe) ? 1 : 0; priv->params.rx_cq_moderation_pkts = MLX5E_PARAMS_DEFAULT_RX_CQ_MODERATION_PKTS; priv->params.tx_cq_moderation_usec = MLX5E_PARAMS_DEFAULT_TX_CQ_MODERATION_USEC; priv->params.tx_cq_moderation_pkts = MLX5E_PARAMS_DEFAULT_TX_CQ_MODERATION_PKTS; priv->params.min_rx_wqes = MLX5E_PARAMS_DEFAULT_MIN_RX_WQES; priv->params.rx_hash_log_tbl_sz = (order_base_2(num_comp_vectors) > MLX5E_PARAMS_DEFAULT_RX_HASH_LOG_TBL_SZ) ? order_base_2(num_comp_vectors) : MLX5E_PARAMS_DEFAULT_RX_HASH_LOG_TBL_SZ; priv->params.num_tc = 1; priv->params.default_vlan_prio = 0; priv->counter_set_id = -1; priv->params.tx_max_inline = mlx5e_get_max_inline_cap(mdev); err = mlx5_query_min_inline(mdev, &priv->params.tx_min_inline_mode); if (err) return (err); /* * hw lro is currently defaulted to off. when it won't anymore we * will consider the HW capability: "!!MLX5_CAP_ETH(mdev, lro_cap)" */ priv->params.hw_lro_en = false; priv->params.lro_wqe_sz = MLX5E_PARAMS_DEFAULT_LRO_WQE_SZ; /* * CQE zipping is currently defaulted to off. when it won't * anymore we will consider the HW capability: * "!!MLX5_CAP_GEN(mdev, cqe_compression)" */ priv->params.cqe_zipping_en = false; priv->mdev = mdev; priv->params.num_channels = num_comp_vectors; priv->params.channels_rsss = 1; priv->order_base_2_num_channels = order_base_2(num_comp_vectors); priv->queue_mapping_channel_mask = roundup_pow_of_two(num_comp_vectors) - 1; priv->num_tc = priv->params.num_tc; priv->default_vlan_prio = priv->params.default_vlan_prio; INIT_WORK(&priv->update_stats_work, mlx5e_update_stats_work); INIT_WORK(&priv->update_carrier_work, mlx5e_update_carrier_work); INIT_WORK(&priv->set_rx_mode_work, mlx5e_set_rx_mode_work); return (0); } static void mlx5e_mkey_set_relaxed_ordering(struct mlx5_core_dev *mdev, void *mkc) { bool ro_pci_enable = pci_get_relaxed_ordering_enabled(mdev->pdev->dev.bsddev); bool ro_write = MLX5_CAP_GEN(mdev, relaxed_ordering_write); bool ro_read = MLX5_CAP_GEN(mdev, relaxed_ordering_read); MLX5_SET(mkc, mkc, relaxed_ordering_read, ro_pci_enable && ro_read); MLX5_SET(mkc, mkc, relaxed_ordering_write, ro_pci_enable && ro_write); } static int mlx5e_create_mkey(struct mlx5e_priv *priv, u32 pdn, struct mlx5_core_mr *mkey) { struct ifnet *ifp = priv->ifp; struct mlx5_core_dev *mdev = priv->mdev; int inlen = MLX5_ST_SZ_BYTES(create_mkey_in); void *mkc; u32 *in; int err; in = mlx5_vzalloc(inlen); if (in == NULL) { mlx5_en_err(ifp, "failed to allocate inbox\n"); return (-ENOMEM); } mkc = MLX5_ADDR_OF(create_mkey_in, in, memory_key_mkey_entry); MLX5_SET(mkc, mkc, access_mode, MLX5_ACCESS_MODE_PA); MLX5_SET(mkc, mkc, umr_en, 1); /* used by HW TLS */ MLX5_SET(mkc, mkc, lw, 1); MLX5_SET(mkc, mkc, lr, 1); mlx5e_mkey_set_relaxed_ordering(mdev, mkc); MLX5_SET(mkc, mkc, pd, pdn); MLX5_SET(mkc, mkc, length64, 1); MLX5_SET(mkc, mkc, qpn, 0xffffff); err = mlx5_core_create_mkey(mdev, mkey, in, inlen); if (err) mlx5_en_err(ifp, "mlx5_core_create_mkey failed, %d\n", err); kvfree(in); return (err); } static const char *mlx5e_vport_stats_desc[] = { MLX5E_VPORT_STATS(MLX5E_STATS_DESC) }; static const char *mlx5e_pport_stats_desc[] = { MLX5E_PPORT_STATS(MLX5E_STATS_DESC) }; static int mlx5e_priv_static_init(struct mlx5e_priv *priv, struct mlx5_core_dev *mdev, const uint32_t channels) { uint32_t x; int err; mtx_init(&priv->async_events_mtx, "mlx5async", MTX_NETWORK_LOCK, MTX_DEF); sx_init(&priv->state_lock, "mlx5state"); callout_init_mtx(&priv->watchdog, &priv->async_events_mtx, 0); MLX5_INIT_DOORBELL_LOCK(&priv->doorbell_lock); for (x = 0; x != channels; x++) mlx5e_chan_static_init(priv, &priv->channel[x], x); for (x = 0; x != channels; x++) { err = mlx5_alloc_bfreg(mdev, &priv->channel[x].bfreg, false, false); if (err) goto err_alloc_bfreg; } return (0); err_alloc_bfreg: while (x--) mlx5_free_bfreg(mdev, &priv->channel[x].bfreg); for (x = 0; x != channels; x++) mlx5e_chan_static_destroy(&priv->channel[x]); callout_drain(&priv->watchdog); mtx_destroy(&priv->async_events_mtx); sx_destroy(&priv->state_lock); return (err); } static void mlx5e_priv_static_destroy(struct mlx5e_priv *priv, struct mlx5_core_dev *mdev, const uint32_t channels) { uint32_t x; for (x = 0; x != channels; x++) mlx5_free_bfreg(mdev, &priv->channel[x].bfreg); for (x = 0; x != channels; x++) mlx5e_chan_static_destroy(&priv->channel[x]); callout_drain(&priv->watchdog); mtx_destroy(&priv->async_events_mtx); sx_destroy(&priv->state_lock); } static int sysctl_firmware(SYSCTL_HANDLER_ARGS) { /* * %d.%d%.d the string format. * fw_rev_{maj,min,sub} return u16, 2^16 = 65536. * We need at most 5 chars to store that. * It also has: two "." and NULL at the end, which means we need 18 * (5*3 + 3) chars at most. */ char fw[18]; struct mlx5e_priv *priv = arg1; int error; snprintf(fw, sizeof(fw), "%d.%d.%d", fw_rev_maj(priv->mdev), fw_rev_min(priv->mdev), fw_rev_sub(priv->mdev)); error = sysctl_handle_string(oidp, fw, sizeof(fw), req); return (error); } static void mlx5e_disable_tx_dma(struct mlx5e_channel *ch) { int i; for (i = 0; i < ch->priv->num_tc; i++) mlx5e_drain_sq(&ch->sq[i]); } static void mlx5e_reset_sq_doorbell_record(struct mlx5e_sq *sq) { sq->doorbell.d32[0] = cpu_to_be32(MLX5_OPCODE_NOP); sq->doorbell.d32[1] = cpu_to_be32(sq->sqn << 8); mlx5e_tx_notify_hw(sq, sq->doorbell.d32); sq->doorbell.d64 = 0; } void mlx5e_resume_sq(struct mlx5e_sq *sq) { int err; /* check if already enabled */ if (READ_ONCE(sq->running) != 0) return; err = mlx5e_modify_sq(sq, MLX5_SQC_STATE_ERR, MLX5_SQC_STATE_RST); if (err != 0) { mlx5_en_err(sq->ifp, "mlx5e_modify_sq() from ERR to RST failed: %d\n", err); } sq->cc = 0; sq->pc = 0; /* reset doorbell prior to moving from RST to RDY */ mlx5e_reset_sq_doorbell_record(sq); err = mlx5e_modify_sq(sq, MLX5_SQC_STATE_RST, MLX5_SQC_STATE_RDY); if (err != 0) { mlx5_en_err(sq->ifp, "mlx5e_modify_sq() from RST to RDY failed: %d\n", err); } sq->cev_next_state = MLX5E_CEV_STATE_INITIAL; WRITE_ONCE(sq->running, 1); } static void mlx5e_enable_tx_dma(struct mlx5e_channel *ch) { int i; for (i = 0; i < ch->priv->num_tc; i++) mlx5e_resume_sq(&ch->sq[i]); } static void mlx5e_disable_rx_dma(struct mlx5e_channel *ch) { struct mlx5e_rq *rq = &ch->rq; struct epoch_tracker et; int err; mtx_lock(&rq->mtx); rq->enabled = 0; callout_stop(&rq->watchdog); mtx_unlock(&rq->mtx); err = mlx5e_modify_rq(rq, MLX5_RQC_STATE_RDY, MLX5_RQC_STATE_ERR); if (err != 0) { mlx5_en_err(rq->ifp, "mlx5e_modify_rq() from RDY to RST failed: %d\n", err); } while (!mlx5_wq_ll_is_empty(&rq->wq)) { msleep(1); NET_EPOCH_ENTER(et); rq->cq.mcq.comp(&rq->cq.mcq, NULL); NET_EPOCH_EXIT(et); } /* * Transitioning into RST state will allow the FW to track less ERR state queues, * thus reducing the recv queue flushing time */ err = mlx5e_modify_rq(rq, MLX5_RQC_STATE_ERR, MLX5_RQC_STATE_RST); if (err != 0) { mlx5_en_err(rq->ifp, "mlx5e_modify_rq() from ERR to RST failed: %d\n", err); } } static void mlx5e_enable_rx_dma(struct mlx5e_channel *ch) { struct mlx5e_rq *rq = &ch->rq; struct epoch_tracker et; int err; rq->wq.wqe_ctr = 0; mlx5_wq_ll_update_db_record(&rq->wq); err = mlx5e_modify_rq(rq, MLX5_RQC_STATE_RST, MLX5_RQC_STATE_RDY); if (err != 0) { mlx5_en_err(rq->ifp, "mlx5e_modify_rq() from RST to RDY failed: %d\n", err); } rq->enabled = 1; NET_EPOCH_ENTER(et); rq->cq.mcq.comp(&rq->cq.mcq, NULL); NET_EPOCH_EXIT(et); } void mlx5e_modify_tx_dma(struct mlx5e_priv *priv, uint8_t value) { int i; if (test_bit(MLX5E_STATE_OPENED, &priv->state) == 0) return; for (i = 0; i < priv->params.num_channels; i++) { if (value) mlx5e_disable_tx_dma(&priv->channel[i]); else mlx5e_enable_tx_dma(&priv->channel[i]); } } void mlx5e_modify_rx_dma(struct mlx5e_priv *priv, uint8_t value) { int i; if (test_bit(MLX5E_STATE_OPENED, &priv->state) == 0) return; for (i = 0; i < priv->params.num_channels; i++) { if (value) mlx5e_disable_rx_dma(&priv->channel[i]); else mlx5e_enable_rx_dma(&priv->channel[i]); } } static void mlx5e_add_hw_stats(struct mlx5e_priv *priv) { SYSCTL_ADD_PROC(&priv->sysctl_ctx, SYSCTL_CHILDREN(priv->sysctl_hw), OID_AUTO, "fw_version", CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, priv, 0, sysctl_firmware, "A", "HCA firmware version"); SYSCTL_ADD_STRING(&priv->sysctl_ctx, SYSCTL_CHILDREN(priv->sysctl_hw), OID_AUTO, "board_id", CTLFLAG_RD, priv->mdev->board_id, 0, "Board ID"); } static int mlx5e_sysctl_tx_priority_flow_control(SYSCTL_HANDLER_ARGS) { struct mlx5e_priv *priv = arg1; uint8_t temp[MLX5E_MAX_PRIORITY]; uint32_t tx_pfc; int err; int i; PRIV_LOCK(priv); tx_pfc = priv->params.tx_priority_flow_control; for (i = 0; i != MLX5E_MAX_PRIORITY; i++) temp[i] = (tx_pfc >> i) & 1; err = SYSCTL_OUT(req, temp, MLX5E_MAX_PRIORITY); if (err || !req->newptr) goto done; err = SYSCTL_IN(req, temp, MLX5E_MAX_PRIORITY); if (err) goto done; priv->params.tx_priority_flow_control = 0; /* range check input value */ for (i = 0; i != MLX5E_MAX_PRIORITY; i++) { if (temp[i] > 1) { err = ERANGE; goto done; } priv->params.tx_priority_flow_control |= (temp[i] << i); } /* check if update is required */ if (tx_pfc != priv->params.tx_priority_flow_control) err = -mlx5e_set_port_pfc(priv); done: if (err != 0) priv->params.tx_priority_flow_control= tx_pfc; PRIV_UNLOCK(priv); return (err); } static int mlx5e_sysctl_rx_priority_flow_control(SYSCTL_HANDLER_ARGS) { struct mlx5e_priv *priv = arg1; uint8_t temp[MLX5E_MAX_PRIORITY]; uint32_t rx_pfc; int err; int i; PRIV_LOCK(priv); rx_pfc = priv->params.rx_priority_flow_control; for (i = 0; i != MLX5E_MAX_PRIORITY; i++) temp[i] = (rx_pfc >> i) & 1; err = SYSCTL_OUT(req, temp, MLX5E_MAX_PRIORITY); if (err || !req->newptr) goto done; err = SYSCTL_IN(req, temp, MLX5E_MAX_PRIORITY); if (err) goto done; priv->params.rx_priority_flow_control = 0; /* range check input value */ for (i = 0; i != MLX5E_MAX_PRIORITY; i++) { if (temp[i] > 1) { err = ERANGE; goto done; } priv->params.rx_priority_flow_control |= (temp[i] << i); } /* check if update is required */ if (rx_pfc != priv->params.rx_priority_flow_control) { err = -mlx5e_set_port_pfc(priv); if (err == 0 && priv->sw_is_port_buf_owner) err = mlx5e_update_buf_lossy(priv); } done: if (err != 0) priv->params.rx_priority_flow_control= rx_pfc; PRIV_UNLOCK(priv); return (err); } static void mlx5e_setup_pauseframes(struct mlx5e_priv *priv) { #if (__FreeBSD_version < 1100000) char path[96]; #endif int error; /* enable pauseframes by default */ priv->params.tx_pauseframe_control = 1; priv->params.rx_pauseframe_control = 1; /* disable ports flow control, PFC, by default */ priv->params.tx_priority_flow_control = 0; priv->params.rx_priority_flow_control = 0; #if (__FreeBSD_version < 1100000) /* compute path for sysctl */ snprintf(path, sizeof(path), "dev.mce.%d.tx_pauseframe_control", device_get_unit(priv->mdev->pdev->dev.bsddev)); /* try to fetch tunable, if any */ TUNABLE_INT_FETCH(path, &priv->params.tx_pauseframe_control); /* compute path for sysctl */ snprintf(path, sizeof(path), "dev.mce.%d.rx_pauseframe_control", device_get_unit(priv->mdev->pdev->dev.bsddev)); /* try to fetch tunable, if any */ TUNABLE_INT_FETCH(path, &priv->params.rx_pauseframe_control); #endif /* register pauseframe SYSCTLs */ SYSCTL_ADD_INT(&priv->sysctl_ctx, SYSCTL_CHILDREN(priv->sysctl_ifnet), OID_AUTO, "tx_pauseframe_control", CTLFLAG_RDTUN, &priv->params.tx_pauseframe_control, 0, "Set to enable TX pause frames. Clear to disable."); SYSCTL_ADD_INT(&priv->sysctl_ctx, SYSCTL_CHILDREN(priv->sysctl_ifnet), OID_AUTO, "rx_pauseframe_control", CTLFLAG_RDTUN, &priv->params.rx_pauseframe_control, 0, "Set to enable RX pause frames. Clear to disable."); /* register priority flow control, PFC, SYSCTLs */ SYSCTL_ADD_PROC(&priv->sysctl_ctx, SYSCTL_CHILDREN(priv->sysctl_ifnet), OID_AUTO, "tx_priority_flow_control", CTLTYPE_U8 | CTLFLAG_RWTUN | CTLFLAG_MPSAFE, priv, 0, &mlx5e_sysctl_tx_priority_flow_control, "CU", "Set to enable TX ports flow control frames for priorities 0..7. Clear to disable."); SYSCTL_ADD_PROC(&priv->sysctl_ctx, SYSCTL_CHILDREN(priv->sysctl_ifnet), OID_AUTO, "rx_priority_flow_control", CTLTYPE_U8 | CTLFLAG_RWTUN | CTLFLAG_MPSAFE, priv, 0, &mlx5e_sysctl_rx_priority_flow_control, "CU", "Set to enable RX ports flow control frames for priorities 0..7. Clear to disable."); PRIV_LOCK(priv); /* range check */ priv->params.tx_pauseframe_control = priv->params.tx_pauseframe_control ? 1 : 0; priv->params.rx_pauseframe_control = priv->params.rx_pauseframe_control ? 1 : 0; /* update firmware */ error = mlx5e_set_port_pause_and_pfc(priv); if (error == -EINVAL) { mlx5_en_err(priv->ifp, "Global pauseframes must be disabled before enabling PFC.\n"); priv->params.rx_priority_flow_control = 0; priv->params.tx_priority_flow_control = 0; /* update firmware */ (void) mlx5e_set_port_pause_and_pfc(priv); } PRIV_UNLOCK(priv); } int mlx5e_ul_snd_tag_alloc(struct ifnet *ifp, union if_snd_tag_alloc_params *params, struct m_snd_tag **ppmt) { struct mlx5e_priv *priv; struct mlx5e_channel *pch; priv = ifp->if_softc; if (unlikely(priv->gone || params->hdr.flowtype == M_HASHTYPE_NONE)) { return (EOPNOTSUPP); } else { /* keep this code synced with mlx5e_select_queue() */ u32 ch = priv->params.num_channels; #ifdef RSS u32 temp; if (rss_hash2bucket(params->hdr.flowid, params->hdr.flowtype, &temp) == 0) ch = temp % ch; else #endif ch = (params->hdr.flowid % 128) % ch; /* * NOTE: The channels array is only freed at detach * and it safe to return a pointer to the send tag * inside the channels structure as long as we * reference the priv. */ pch = priv->channel + ch; /* check if send queue is not running */ if (unlikely(pch->sq[0].running == 0)) return (ENXIO); m_snd_tag_ref(&pch->tag); *ppmt = &pch->tag; return (0); } } int mlx5e_ul_snd_tag_query(struct m_snd_tag *pmt, union if_snd_tag_query_params *params) { struct mlx5e_channel *pch = container_of(pmt, struct mlx5e_channel, tag); params->unlimited.max_rate = -1ULL; params->unlimited.queue_level = mlx5e_sq_queue_level(&pch->sq[0]); return (0); } void mlx5e_ul_snd_tag_free(struct m_snd_tag *pmt) { struct mlx5e_channel *pch = container_of(pmt, struct mlx5e_channel, tag); complete(&pch->completion); } static int mlx5e_snd_tag_alloc(struct ifnet *ifp, union if_snd_tag_alloc_params *params, struct m_snd_tag **ppmt) { switch (params->hdr.type) { #ifdef RATELIMIT case IF_SND_TAG_TYPE_RATE_LIMIT: return (mlx5e_rl_snd_tag_alloc(ifp, params, ppmt)); #ifdef KERN_TLS case IF_SND_TAG_TYPE_TLS_RATE_LIMIT: return (mlx5e_tls_snd_tag_alloc(ifp, params, ppmt)); #endif #endif case IF_SND_TAG_TYPE_UNLIMITED: return (mlx5e_ul_snd_tag_alloc(ifp, params, ppmt)); #ifdef KERN_TLS case IF_SND_TAG_TYPE_TLS: return (mlx5e_tls_snd_tag_alloc(ifp, params, ppmt)); #endif default: return (EOPNOTSUPP); } } static int mlx5e_snd_tag_modify(struct m_snd_tag *pmt, union if_snd_tag_modify_params *params) { switch (pmt->type) { #ifdef RATELIMIT case IF_SND_TAG_TYPE_RATE_LIMIT: return (mlx5e_rl_snd_tag_modify(pmt, params)); #ifdef KERN_TLS case IF_SND_TAG_TYPE_TLS_RATE_LIMIT: return (mlx5e_tls_snd_tag_modify(pmt, params)); #endif #endif case IF_SND_TAG_TYPE_UNLIMITED: #ifdef KERN_TLS case IF_SND_TAG_TYPE_TLS: #endif default: return (EOPNOTSUPP); } } static int mlx5e_snd_tag_query(struct m_snd_tag *pmt, union if_snd_tag_query_params *params) { switch (pmt->type) { #ifdef RATELIMIT case IF_SND_TAG_TYPE_RATE_LIMIT: return (mlx5e_rl_snd_tag_query(pmt, params)); #ifdef KERN_TLS case IF_SND_TAG_TYPE_TLS_RATE_LIMIT: return (mlx5e_tls_snd_tag_query(pmt, params)); #endif #endif case IF_SND_TAG_TYPE_UNLIMITED: return (mlx5e_ul_snd_tag_query(pmt, params)); #ifdef KERN_TLS case IF_SND_TAG_TYPE_TLS: return (mlx5e_tls_snd_tag_query(pmt, params)); #endif default: return (EOPNOTSUPP); } } #ifdef RATELIMIT #define NUM_HDWR_RATES_MLX 13 static const uint64_t adapter_rates_mlx[NUM_HDWR_RATES_MLX] = { 135375, /* 1,083,000 */ 180500, /* 1,444,000 */ 270750, /* 2,166,000 */ 361000, /* 2,888,000 */ 541500, /* 4,332,000 */ 721875, /* 5,775,000 */ 1082875, /* 8,663,000 */ 1443875, /* 11,551,000 */ 2165750, /* 17,326,000 */ 2887750, /* 23,102,000 */ 4331625, /* 34,653,000 */ 5775500, /* 46,204,000 */ 8663125 /* 69,305,000 */ }; static void mlx5e_ratelimit_query(struct ifnet *ifp __unused, struct if_ratelimit_query_results *q) { /* * This function needs updating by the driver maintainer! * For the MLX card there are currently (ConectX-4?) 13 * pre-set rates and others i.e. ConnectX-5, 6, 7?? * * This will change based on later adapters * and this code should be updated to look at ifp * and figure out the specific adapter type * settings i.e. how many rates as well * as if they are fixed (as is shown here) or * if they are dynamic (example chelsio t4). Also if there * is a maximum number of flows that the adapter * can handle that too needs to be updated in * the max_flows field. */ q->rate_table = adapter_rates_mlx; q->flags = RT_IS_FIXED_TABLE; q->max_flows = 0; /* mlx has no limit */ q->number_of_rates = NUM_HDWR_RATES_MLX; q->min_segment_burst = 1; } #endif static void mlx5e_snd_tag_free(struct m_snd_tag *pmt) { switch (pmt->type) { #ifdef RATELIMIT case IF_SND_TAG_TYPE_RATE_LIMIT: mlx5e_rl_snd_tag_free(pmt); break; #ifdef KERN_TLS case IF_SND_TAG_TYPE_TLS_RATE_LIMIT: mlx5e_tls_snd_tag_free(pmt); break; #endif #endif case IF_SND_TAG_TYPE_UNLIMITED: mlx5e_ul_snd_tag_free(pmt); break; #ifdef KERN_TLS case IF_SND_TAG_TYPE_TLS: mlx5e_tls_snd_tag_free(pmt); break; #endif default: break; } } static void mlx5e_ifm_add(struct mlx5e_priv *priv, int type) { ifmedia_add(&priv->media, type | IFM_ETHER, 0, NULL); ifmedia_add(&priv->media, type | IFM_ETHER | IFM_ETH_RXPAUSE | IFM_ETH_TXPAUSE, 0, NULL); ifmedia_add(&priv->media, type | IFM_ETHER | IFM_ETH_RXPAUSE, 0, NULL); ifmedia_add(&priv->media, type | IFM_ETHER | IFM_ETH_TXPAUSE, 0, NULL); ifmedia_add(&priv->media, type | IFM_ETHER | IFM_FDX, 0, NULL); ifmedia_add(&priv->media, type | IFM_ETHER | IFM_FDX | IFM_ETH_RXPAUSE, 0, NULL); ifmedia_add(&priv->media, type | IFM_ETHER | IFM_FDX | IFM_ETH_TXPAUSE, 0, NULL); ifmedia_add(&priv->media, type | IFM_ETHER | IFM_FDX | IFM_ETH_RXPAUSE | IFM_ETH_TXPAUSE, 0, NULL); } static void * mlx5e_create_ifp(struct mlx5_core_dev *mdev) { struct ifnet *ifp; struct mlx5e_priv *priv; u8 dev_addr[ETHER_ADDR_LEN] __aligned(4); struct sysctl_oid_list *child; int ncv = mdev->priv.eq_table.num_comp_vectors; char unit[16]; struct pfil_head_args pa; int err; u32 eth_proto_cap; u32 out[MLX5_ST_SZ_DW(ptys_reg)]; bool ext; struct media media_entry = {}; if (mlx5e_check_required_hca_cap(mdev)) { mlx5_core_dbg(mdev, "mlx5e_check_required_hca_cap() failed\n"); return (NULL); } + /* * Try to allocate the priv and make room for worst-case * number of channel structures: */ - priv = malloc(sizeof(*priv) + + priv = malloc_domainset(sizeof(*priv) + (sizeof(priv->channel[0]) * mdev->priv.eq_table.num_comp_vectors), - M_MLX5EN, M_WAITOK | M_ZERO); + M_MLX5EN, mlx5_dev_domainset(mdev), M_WAITOK | M_ZERO); ifp = priv->ifp = if_alloc_dev(IFT_ETHER, mdev->pdev->dev.bsddev); if (ifp == NULL) { mlx5_core_err(mdev, "if_alloc() failed\n"); goto err_free_priv; } /* setup all static fields */ if (mlx5e_priv_static_init(priv, mdev, mdev->priv.eq_table.num_comp_vectors)) { mlx5_core_err(mdev, "mlx5e_priv_static_init() failed\n"); goto err_free_ifp; } ifp->if_softc = priv; if_initname(ifp, "mce", device_get_unit(mdev->pdev->dev.bsddev)); ifp->if_mtu = ETHERMTU; ifp->if_init = mlx5e_open; ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST | IFF_KNOWSEPOCH; ifp->if_ioctl = mlx5e_ioctl; ifp->if_transmit = mlx5e_xmit; ifp->if_qflush = if_qflush; #if (__FreeBSD_version >= 1100000) ifp->if_get_counter = mlx5e_get_counter; #endif ifp->if_snd.ifq_maxlen = ifqmaxlen; /* * Set driver features */ ifp->if_capabilities |= IFCAP_HWCSUM | IFCAP_HWCSUM_IPV6; ifp->if_capabilities |= IFCAP_VLAN_MTU | IFCAP_VLAN_HWTAGGING; ifp->if_capabilities |= IFCAP_VLAN_HWCSUM | IFCAP_VLAN_HWFILTER; ifp->if_capabilities |= IFCAP_LINKSTATE | IFCAP_JUMBO_MTU; ifp->if_capabilities |= IFCAP_LRO; ifp->if_capabilities |= IFCAP_TSO | IFCAP_VLAN_HWTSO; ifp->if_capabilities |= IFCAP_HWSTATS | IFCAP_HWRXTSTMP; ifp->if_capabilities |= IFCAP_MEXTPG; ifp->if_capabilities |= IFCAP_TXTLS4 | IFCAP_TXTLS6; #ifdef RATELIMIT ifp->if_capabilities |= IFCAP_TXRTLMT | IFCAP_TXTLS_RTLMT; #endif ifp->if_capabilities |= IFCAP_VXLAN_HWCSUM | IFCAP_VXLAN_HWTSO; ifp->if_snd_tag_alloc = mlx5e_snd_tag_alloc; ifp->if_snd_tag_free = mlx5e_snd_tag_free; ifp->if_snd_tag_modify = mlx5e_snd_tag_modify; ifp->if_snd_tag_query = mlx5e_snd_tag_query; #ifdef RATELIMIT ifp->if_ratelimit_query = mlx5e_ratelimit_query; #endif /* set TSO limits so that we don't have to drop TX packets */ ifp->if_hw_tsomax = MLX5E_MAX_TX_PAYLOAD_SIZE - (ETHER_HDR_LEN + ETHER_VLAN_ENCAP_LEN); ifp->if_hw_tsomaxsegcount = MLX5E_MAX_TX_MBUF_FRAGS - 1 /* hdr */; ifp->if_hw_tsomaxsegsize = MLX5E_MAX_TX_MBUF_SIZE; ifp->if_capenable = ifp->if_capabilities; ifp->if_hwassist = 0; if (ifp->if_capenable & IFCAP_TSO) ifp->if_hwassist |= CSUM_TSO; if (ifp->if_capenable & IFCAP_TXCSUM) ifp->if_hwassist |= (CSUM_TCP | CSUM_UDP | CSUM_IP); if (ifp->if_capenable & IFCAP_TXCSUM_IPV6) ifp->if_hwassist |= (CSUM_UDP_IPV6 | CSUM_TCP_IPV6); if (ifp->if_capabilities & IFCAP_VXLAN_HWCSUM) ifp->if_hwassist |= CSUM_INNER_IP6_UDP | CSUM_INNER_IP6_TCP | CSUM_INNER_IP | CSUM_INNER_IP_UDP | CSUM_INNER_IP_TCP | CSUM_ENCAP_VXLAN; if (ifp->if_capabilities & IFCAP_VXLAN_HWTSO) ifp->if_hwassist |= CSUM_INNER_IP6_TSO | CSUM_INNER_IP_TSO; /* ifnet sysctl tree */ sysctl_ctx_init(&priv->sysctl_ctx); priv->sysctl_ifnet = SYSCTL_ADD_NODE(&priv->sysctl_ctx, SYSCTL_STATIC_CHILDREN(_dev), OID_AUTO, ifp->if_dname, CTLFLAG_RD | CTLFLAG_MPSAFE, 0, "MLX5 ethernet - interface name"); if (priv->sysctl_ifnet == NULL) { mlx5_core_err(mdev, "SYSCTL_ADD_NODE() failed\n"); goto err_free_sysctl; } snprintf(unit, sizeof(unit), "%d", ifp->if_dunit); priv->sysctl_ifnet = SYSCTL_ADD_NODE(&priv->sysctl_ctx, SYSCTL_CHILDREN(priv->sysctl_ifnet), OID_AUTO, unit, CTLFLAG_RD | CTLFLAG_MPSAFE, 0, "MLX5 ethernet - interface unit"); if (priv->sysctl_ifnet == NULL) { mlx5_core_err(mdev, "SYSCTL_ADD_NODE() failed\n"); goto err_free_sysctl; } /* HW sysctl tree */ child = SYSCTL_CHILDREN(device_get_sysctl_tree(mdev->pdev->dev.bsddev)); priv->sysctl_hw = SYSCTL_ADD_NODE(&priv->sysctl_ctx, child, OID_AUTO, "hw", CTLFLAG_RD | CTLFLAG_MPSAFE, 0, "MLX5 ethernet dev hw"); if (priv->sysctl_hw == NULL) { mlx5_core_err(mdev, "SYSCTL_ADD_NODE() failed\n"); goto err_free_sysctl; } err = mlx5e_build_ifp_priv(mdev, priv, ncv); if (err) { mlx5_core_err(mdev, "mlx5e_build_ifp_priv() failed (%d)\n", err); goto err_free_sysctl; } /* reuse mlx5core's watchdog workqueue */ priv->wq = mdev->priv.health.wq_watchdog; err = mlx5_core_alloc_pd(mdev, &priv->pdn); if (err) { mlx5_en_err(ifp, "mlx5_core_alloc_pd failed, %d\n", err); goto err_free_wq; } err = mlx5_alloc_transport_domain(mdev, &priv->tdn); if (err) { mlx5_en_err(ifp, "mlx5_alloc_transport_domain failed, %d\n", err); goto err_dealloc_pd; } err = mlx5e_create_mkey(priv, priv->pdn, &priv->mr); if (err) { mlx5_en_err(ifp, "mlx5e_create_mkey failed, %d\n", err); goto err_dealloc_transport_domain; } mlx5_query_nic_vport_mac_address(priv->mdev, 0, dev_addr); /* check if we should generate a random MAC address */ if (MLX5_CAP_GEN(priv->mdev, vport_group_manager) == 0 && is_zero_ether_addr(dev_addr)) { random_ether_addr(dev_addr); mlx5_en_err(ifp, "Assigned random MAC address\n"); } err = mlx5e_rl_init(priv); if (err) { mlx5_en_err(ifp, "mlx5e_rl_init failed, %d\n", err); goto err_create_mkey; } err = mlx5e_tls_init(priv); if (err) { if_printf(ifp, "%s: mlx5e_tls_init failed\n", __func__); goto err_rl_init; } /* set default MTU */ mlx5e_set_dev_port_mtu(ifp, ifp->if_mtu); /* Set default media status */ priv->media_status_last = IFM_AVALID; priv->media_active_last = IFM_ETHER | IFM_AUTO | IFM_FDX; /* setup default pauseframes configuration */ mlx5e_setup_pauseframes(priv); /* Setup supported medias */ if (!mlx5_query_port_ptys(mdev, out, sizeof(out), MLX5_PTYS_EN, 1)) { ext = MLX5_CAP_PCAM_FEATURE(mdev, ptys_extended_ethernet); eth_proto_cap = MLX5_GET_ETH_PROTO(ptys_reg, out, ext, eth_proto_capability); } else { ext = false; eth_proto_cap = 0; mlx5_en_err(ifp, "Query port media capability failed, %d\n", err); } ifmedia_init(&priv->media, IFM_IMASK, mlx5e_media_change, mlx5e_media_status); if (ext) { for (unsigned i = 0; i != MLX5E_EXT_LINK_SPEEDS_NUMBER; i++) { /* check if hardware has the right capability */ if (MLX5E_PROT_MASK(i) & ~eth_proto_cap) continue; for (unsigned j = 0; j != MLX5E_CABLE_TYPE_NUMBER; j++) { media_entry = mlx5e_ext_mode_table[i][j]; if (media_entry.subtype == 0) continue; /* check if this subtype was already added */ for (unsigned k = 0; k != i; k++) { /* check if hardware has the right capability */ if (MLX5E_PROT_MASK(k) & ~eth_proto_cap) continue; for (unsigned m = 0; m != MLX5E_CABLE_TYPE_NUMBER; m++) { if (media_entry.subtype == mlx5e_ext_mode_table[k][m].subtype) goto skip_ext_media; } } mlx5e_ifm_add(priv, media_entry.subtype); skip_ext_media:; } } } else { for (unsigned i = 0; i != MLX5E_LINK_SPEEDS_NUMBER; i++) { media_entry = mlx5e_mode_table[i]; if (media_entry.subtype == 0) continue; if (MLX5E_PROT_MASK(i) & ~eth_proto_cap) continue; /* check if this subtype was already added */ for (unsigned k = 0; k != i; k++) { if (media_entry.subtype == mlx5e_mode_table[k].subtype) goto skip_media; } mlx5e_ifm_add(priv, media_entry.subtype); /* NOTE: 10G ER and LR shares the same entry */ if (media_entry.subtype == IFM_10G_ER) mlx5e_ifm_add(priv, IFM_10G_LR); skip_media:; } } mlx5e_ifm_add(priv, IFM_AUTO); /* Set autoselect by default */ ifmedia_set(&priv->media, IFM_ETHER | IFM_AUTO | IFM_FDX | IFM_ETH_RXPAUSE | IFM_ETH_TXPAUSE); DEBUGNET_SET(ifp, mlx5_en); ether_ifattach(ifp, dev_addr); /* Register for VLAN events */ priv->vlan_attach = EVENTHANDLER_REGISTER(vlan_config, mlx5e_vlan_rx_add_vid, priv, EVENTHANDLER_PRI_FIRST); priv->vlan_detach = EVENTHANDLER_REGISTER(vlan_unconfig, mlx5e_vlan_rx_kill_vid, priv, EVENTHANDLER_PRI_FIRST); /* Register for VxLAN events */ priv->vxlan_start = EVENTHANDLER_REGISTER(vxlan_start, mlx5e_vxlan_start, priv, EVENTHANDLER_PRI_ANY); priv->vxlan_stop = EVENTHANDLER_REGISTER(vxlan_stop, mlx5e_vxlan_stop, priv, EVENTHANDLER_PRI_ANY); /* Link is down by default */ if_link_state_change(ifp, LINK_STATE_DOWN); mlx5e_enable_async_events(priv); mlx5e_add_hw_stats(priv); mlx5e_create_stats(&priv->stats.vport.ctx, SYSCTL_CHILDREN(priv->sysctl_ifnet), "vstats", mlx5e_vport_stats_desc, MLX5E_VPORT_STATS_NUM, priv->stats.vport.arg); mlx5e_create_stats(&priv->stats.pport.ctx, SYSCTL_CHILDREN(priv->sysctl_ifnet), "pstats", mlx5e_pport_stats_desc, MLX5E_PPORT_STATS_NUM, priv->stats.pport.arg); mlx5e_create_ethtool(priv); mtx_lock(&priv->async_events_mtx); mlx5e_update_stats(priv); mtx_unlock(&priv->async_events_mtx); SYSCTL_ADD_INT(&priv->sysctl_ctx, SYSCTL_CHILDREN(priv->sysctl_ifnet), OID_AUTO, "rx_clbr_done", CTLFLAG_RD, &priv->clbr_done, 0, "RX timestamps calibration state"); callout_init(&priv->tstmp_clbr, CALLOUT_DIRECT); mlx5e_reset_calibration_callout(priv); pa.pa_version = PFIL_VERSION; pa.pa_flags = PFIL_IN; pa.pa_type = PFIL_TYPE_ETHERNET; pa.pa_headname = ifp->if_xname; priv->pfil = pfil_head_register(&pa); return (priv); err_rl_init: mlx5e_rl_cleanup(priv); err_create_mkey: mlx5_core_destroy_mkey(priv->mdev, &priv->mr); err_dealloc_transport_domain: mlx5_dealloc_transport_domain(mdev, priv->tdn); err_dealloc_pd: mlx5_core_dealloc_pd(mdev, priv->pdn); err_free_wq: flush_workqueue(priv->wq); err_free_sysctl: sysctl_ctx_free(&priv->sysctl_ctx); if (priv->sysctl_debug) sysctl_ctx_free(&priv->stats.port_stats_debug.ctx); mlx5e_priv_static_destroy(priv, mdev, mdev->priv.eq_table.num_comp_vectors); err_free_ifp: if_free(ifp); err_free_priv: free(priv, M_MLX5EN); return (NULL); } static void mlx5e_destroy_ifp(struct mlx5_core_dev *mdev, void *vpriv) { struct mlx5e_priv *priv = vpriv; struct ifnet *ifp = priv->ifp; /* don't allow more IOCTLs */ priv->gone = 1; /* XXX wait a bit to allow IOCTL handlers to complete */ pause("W", hz); #ifdef RATELIMIT /* * The kernel can have reference(s) via the m_snd_tag's into * the ratelimit channels, and these must go away before * detaching: */ while (READ_ONCE(priv->rl.stats.tx_active_connections) != 0) { mlx5_en_err(priv->ifp, "Waiting for all ratelimit connections to terminate\n"); pause("W", hz); } #endif #ifdef KERN_TLS /* wait for all TLS tags to get freed */ while (priv->tls.init != 0 && uma_zone_get_cur(priv->tls.zone) != 0) { mlx5_en_err(priv->ifp, "Waiting for all TLS connections to terminate\n"); pause("W", hz); } #endif /* wait for all unlimited send tags to complete */ mlx5e_priv_wait_for_completion(priv, mdev->priv.eq_table.num_comp_vectors); /* stop watchdog timer */ callout_drain(&priv->watchdog); callout_drain(&priv->tstmp_clbr); if (priv->vlan_attach != NULL) EVENTHANDLER_DEREGISTER(vlan_config, priv->vlan_attach); if (priv->vlan_detach != NULL) EVENTHANDLER_DEREGISTER(vlan_unconfig, priv->vlan_detach); if (priv->vxlan_start != NULL) EVENTHANDLER_DEREGISTER(vxlan_start, priv->vxlan_start); if (priv->vxlan_stop != NULL) EVENTHANDLER_DEREGISTER(vxlan_stop, priv->vxlan_stop); /* make sure device gets closed */ PRIV_LOCK(priv); mlx5e_close_locked(ifp); PRIV_UNLOCK(priv); /* deregister pfil */ if (priv->pfil != NULL) { pfil_head_unregister(priv->pfil); priv->pfil = NULL; } /* unregister device */ ifmedia_removeall(&priv->media); ether_ifdetach(ifp); mlx5e_tls_cleanup(priv); mlx5e_rl_cleanup(priv); /* destroy all remaining sysctl nodes */ sysctl_ctx_free(&priv->stats.vport.ctx); sysctl_ctx_free(&priv->stats.pport.ctx); if (priv->sysctl_debug) sysctl_ctx_free(&priv->stats.port_stats_debug.ctx); sysctl_ctx_free(&priv->sysctl_ctx); mlx5_core_destroy_mkey(priv->mdev, &priv->mr); mlx5_dealloc_transport_domain(priv->mdev, priv->tdn); mlx5_core_dealloc_pd(priv->mdev, priv->pdn); mlx5e_disable_async_events(priv); flush_workqueue(priv->wq); mlx5e_priv_static_destroy(priv, mdev, mdev->priv.eq_table.num_comp_vectors); if_free(ifp); free(priv, M_MLX5EN); } #ifdef DEBUGNET static void mlx5_en_debugnet_init(struct ifnet *dev, int *nrxr, int *ncl, int *clsize) { struct mlx5e_priv *priv = if_getsoftc(dev); PRIV_LOCK(priv); *nrxr = priv->params.num_channels; *ncl = DEBUGNET_MAX_IN_FLIGHT; *clsize = MLX5E_MAX_RX_BYTES; PRIV_UNLOCK(priv); } static void mlx5_en_debugnet_event(struct ifnet *dev, enum debugnet_ev event) { } static int mlx5_en_debugnet_transmit(struct ifnet *dev, struct mbuf *m) { struct mlx5e_priv *priv = if_getsoftc(dev); struct mlx5e_sq *sq; int err; if ((if_getdrvflags(dev) & (IFF_DRV_RUNNING | IFF_DRV_OACTIVE)) != IFF_DRV_RUNNING || (priv->media_status_last & IFM_ACTIVE) == 0) return (ENOENT); sq = &priv->channel[0].sq[0]; if (sq->running == 0) { m_freem(m); return (ENOENT); } if (mlx5e_sq_xmit(sq, &m) != 0) { m_freem(m); err = ENOBUFS; } else { err = 0; } if (likely(sq->doorbell.d64 != 0)) { mlx5e_tx_notify_hw(sq, sq->doorbell.d32); sq->doorbell.d64 = 0; } return (err); } static int mlx5_en_debugnet_poll(struct ifnet *dev, int count) { struct mlx5e_priv *priv = if_getsoftc(dev); if ((if_getdrvflags(dev) & IFF_DRV_RUNNING) == 0 || (priv->media_status_last & IFM_ACTIVE) == 0) return (ENOENT); mlx5_poll_interrupts(priv->mdev); return (0); } #endif /* DEBUGNET */ static void * mlx5e_get_ifp(void *vpriv) { struct mlx5e_priv *priv = vpriv; return (priv->ifp); } static struct mlx5_interface mlx5e_interface = { .add = mlx5e_create_ifp, .remove = mlx5e_destroy_ifp, .event = mlx5e_async_event, .protocol = MLX5_INTERFACE_PROTOCOL_ETH, .get_dev = mlx5e_get_ifp, }; void mlx5e_init(void) { mlx5_register_interface(&mlx5e_interface); } void mlx5e_cleanup(void) { mlx5_unregister_interface(&mlx5e_interface); } static void mlx5e_show_version(void __unused *arg) { printf("%s", mlx5e_version); } SYSINIT(mlx5e_show_version, SI_SUB_DRIVERS, SI_ORDER_ANY, mlx5e_show_version, NULL); module_init_order(mlx5e_init, SI_ORDER_SIXTH); module_exit_order(mlx5e_cleanup, SI_ORDER_SIXTH); #if (__FreeBSD_version >= 1100000) MODULE_DEPEND(mlx5en, linuxkpi, 1, 1, 1); #endif MODULE_DEPEND(mlx5en, mlx5, 1, 1, 1); MODULE_VERSION(mlx5en, 1); diff --git a/sys/dev/mlx5/mlx5_en/mlx5_en_rl.c b/sys/dev/mlx5/mlx5_en/mlx5_en_rl.c index bd66b8ecd725..fa7b812fbdf1 100644 --- a/sys/dev/mlx5/mlx5_en/mlx5_en_rl.c +++ b/sys/dev/mlx5/mlx5_en/mlx5_en_rl.c @@ -1,1574 +1,1572 @@ /*- * Copyright (c) 2016-2020 Mellanox Technologies. All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY AUTHOR AND CONTRIBUTORS `AS IS' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * $FreeBSD$ */ #include "en.h" #ifdef RATELIMIT static int mlx5e_rl_open_workers(struct mlx5e_priv *); static void mlx5e_rl_close_workers(struct mlx5e_priv *); static int mlx5e_rl_sysctl_show_rate_table(SYSCTL_HANDLER_ARGS); static void mlx5e_rl_sysctl_add_u64_oid(struct mlx5e_rl_priv_data *, unsigned x, struct sysctl_oid *, const char *name, const char *desc); static void mlx5e_rl_sysctl_add_stats_u64_oid(struct mlx5e_rl_priv_data *rl, unsigned x, struct sysctl_oid *node, const char *name, const char *desc); static int mlx5e_rl_tx_limit_add(struct mlx5e_rl_priv_data *, uint64_t value); static int mlx5e_rl_tx_limit_clr(struct mlx5e_rl_priv_data *, uint64_t value); static void mlx5e_rl_build_sq_param(struct mlx5e_rl_priv_data *rl, struct mlx5e_sq_param *param) { void *sqc = param->sqc; void *wq = MLX5_ADDR_OF(sqc, sqc, wq); uint8_t log_sq_size = order_base_2(rl->param.tx_queue_size); MLX5_SET(wq, wq, log_wq_sz, log_sq_size); MLX5_SET(wq, wq, log_wq_stride, ilog2(MLX5_SEND_WQE_BB)); MLX5_SET(wq, wq, pd, rl->priv->pdn); - param->wq.buf_numa_node = 0; - param->wq.db_numa_node = 0; param->wq.linear = 1; } static void mlx5e_rl_build_cq_param(struct mlx5e_rl_priv_data *rl, struct mlx5e_cq_param *param) { void *cqc = param->cqc; uint8_t log_sq_size = order_base_2(rl->param.tx_queue_size); MLX5_SET(cqc, cqc, log_cq_size, log_sq_size); MLX5_SET(cqc, cqc, cq_period, rl->param.tx_coalesce_usecs); MLX5_SET(cqc, cqc, cq_max_count, rl->param.tx_coalesce_pkts); MLX5_SET(cqc, cqc, uar_page, rl->priv->mdev->priv.uar->index); switch (rl->param.tx_coalesce_mode) { case 0: MLX5_SET(cqc, cqc, cq_period_mode, MLX5_CQ_PERIOD_MODE_START_FROM_EQE); break; default: if (MLX5_CAP_GEN(rl->priv->mdev, cq_period_start_from_cqe)) MLX5_SET(cqc, cqc, cq_period_mode, MLX5_CQ_PERIOD_MODE_START_FROM_CQE); else MLX5_SET(cqc, cqc, cq_period_mode, MLX5_CQ_PERIOD_MODE_START_FROM_EQE); break; } } static void mlx5e_rl_build_channel_param(struct mlx5e_rl_priv_data *rl, struct mlx5e_rl_channel_param *cparam) { memset(cparam, 0, sizeof(*cparam)); mlx5e_rl_build_sq_param(rl, &cparam->sq); mlx5e_rl_build_cq_param(rl, &cparam->cq); } static int mlx5e_rl_create_sq(struct mlx5e_priv *priv, struct mlx5e_sq *sq, struct mlx5e_sq_param *param, int ix) { struct mlx5_core_dev *mdev = priv->mdev; void *sqc = param->sqc; void *sqc_wq = MLX5_ADDR_OF(sqc, sqc, wq); int err; /* Create DMA descriptor TAG */ if ((err = -bus_dma_tag_create( bus_get_dma_tag(mdev->pdev->dev.bsddev), 1, /* any alignment */ 0, /* no boundary */ BUS_SPACE_MAXADDR, /* lowaddr */ BUS_SPACE_MAXADDR, /* highaddr */ NULL, NULL, /* filter, filterarg */ MLX5E_MAX_TX_PAYLOAD_SIZE, /* maxsize */ MLX5E_MAX_TX_MBUF_FRAGS, /* nsegments */ MLX5E_MAX_TX_MBUF_SIZE, /* maxsegsize */ 0, /* flags */ NULL, NULL, /* lockfunc, lockfuncarg */ &sq->dma_tag))) goto done; + sq->mkey_be = cpu_to_be32(priv->mr.key); + sq->ifp = priv->ifp; + sq->priv = priv; + err = mlx5_wq_cyc_create(mdev, ¶m->wq, sqc_wq, &sq->wq, &sq->wq_ctrl); if (err) goto err_free_dma_tag; sq->wq.db = &sq->wq.db[MLX5_SND_DBR]; err = mlx5e_alloc_sq_db(sq); if (err) goto err_sq_wq_destroy; - sq->mkey_be = cpu_to_be32(priv->mr.key); - sq->ifp = priv->ifp; - sq->priv = priv; - mlx5e_update_sq_inline(sq); return (0); err_sq_wq_destroy: mlx5_wq_destroy(&sq->wq_ctrl); err_free_dma_tag: bus_dma_tag_destroy(sq->dma_tag); done: return (err); } static void mlx5e_rl_destroy_sq(struct mlx5e_sq *sq) { mlx5e_free_sq_db(sq); mlx5_wq_destroy(&sq->wq_ctrl); bus_dma_tag_destroy(sq->dma_tag); } static int mlx5e_rl_open_sq(struct mlx5e_priv *priv, struct mlx5e_sq *sq, struct mlx5e_sq_param *param, int ix) { int err; err = mlx5e_rl_create_sq(priv, sq, param, ix); if (err) return (err); err = mlx5e_enable_sq(sq, param, &priv->channel[ix].bfreg, priv->rl.tisn); if (err) goto err_destroy_sq; err = mlx5e_modify_sq(sq, MLX5_SQC_STATE_RST, MLX5_SQC_STATE_RDY); if (err) goto err_disable_sq; WRITE_ONCE(sq->running, 1); return (0); err_disable_sq: mlx5e_disable_sq(sq); err_destroy_sq: mlx5e_rl_destroy_sq(sq); return (err); } static void mlx5e_rl_chan_mtx_init(struct mlx5e_priv *priv, struct mlx5e_sq *sq) { mtx_init(&sq->lock, "mlx5tx-rl", NULL, MTX_DEF); mtx_init(&sq->comp_lock, "mlx5comp-rl", NULL, MTX_DEF); callout_init_mtx(&sq->cev_callout, &sq->lock, 0); sq->cev_factor = priv->rl.param.tx_completion_fact; /* ensure the TX completion event factor is not zero */ if (sq->cev_factor == 0) sq->cev_factor = 1; } static int mlx5e_rl_open_channel(struct mlx5e_rl_worker *rlw, int eq_ix, struct mlx5e_rl_channel_param *cparam, struct mlx5e_sq *volatile *ppsq) { struct mlx5e_priv *priv = rlw->priv; struct mlx5e_sq *sq; int err; sq = malloc(sizeof(*sq), M_MLX5EN, M_WAITOK | M_ZERO); /* init mutexes */ mlx5e_rl_chan_mtx_init(priv, sq); /* open TX completion queue */ err = mlx5e_open_cq(priv, &cparam->cq, &sq->cq, &mlx5e_tx_cq_comp, eq_ix); if (err) goto err_free; err = mlx5e_rl_open_sq(priv, sq, &cparam->sq, eq_ix); if (err) goto err_close_tx_cq; /* store TX channel pointer */ *ppsq = sq; /* poll TX queue initially */ sq->cq.mcq.comp(&sq->cq.mcq, NULL); return (0); err_close_tx_cq: mlx5e_close_cq(&sq->cq); err_free: /* destroy mutexes */ mtx_destroy(&sq->lock); mtx_destroy(&sq->comp_lock); free(sq, M_MLX5EN); atomic_add_64(&priv->rl.stats.tx_allocate_resource_failure, 1ULL); return (err); } static void mlx5e_rl_close_channel(struct mlx5e_sq *volatile *ppsq) { struct mlx5e_sq *sq = *ppsq; /* check if channel is already closed */ if (sq == NULL) return; /* ensure channel pointer is no longer used */ *ppsq = NULL; /* teardown and destroy SQ */ mlx5e_drain_sq(sq); mlx5e_disable_sq(sq); mlx5e_rl_destroy_sq(sq); /* close CQ */ mlx5e_close_cq(&sq->cq); /* destroy mutexes */ mtx_destroy(&sq->lock); mtx_destroy(&sq->comp_lock); free(sq, M_MLX5EN); } static void mlx5e_rl_sync_tx_completion_fact(struct mlx5e_rl_priv_data *rl) { /* * Limit the maximum distance between completion events to * half of the currently set TX queue size. * * The maximum number of queue entries a single IP packet can * consume is given by MLX5_SEND_WQE_MAX_WQEBBS. * * The worst case max value is then given as below: */ uint64_t max = rl->param.tx_queue_size / (2 * MLX5_SEND_WQE_MAX_WQEBBS); /* * Update the maximum completion factor value in case the * tx_queue_size field changed. Ensure we don't overflow * 16-bits. */ if (max < 1) max = 1; else if (max > 65535) max = 65535; rl->param.tx_completion_fact_max = max; /* * Verify that the current TX completion factor is within the * given limits: */ if (rl->param.tx_completion_fact < 1) rl->param.tx_completion_fact = 1; else if (rl->param.tx_completion_fact > max) rl->param.tx_completion_fact = max; } static int mlx5e_rl_modify_sq(struct mlx5e_sq *sq, uint16_t rl_index) { struct mlx5e_priv *priv = sq->priv; struct mlx5_core_dev *mdev = priv->mdev; void *in; void *sqc; int inlen; int err; inlen = MLX5_ST_SZ_BYTES(modify_sq_in); in = mlx5_vzalloc(inlen); if (in == NULL) return (-ENOMEM); sqc = MLX5_ADDR_OF(modify_sq_in, in, ctx); MLX5_SET(modify_sq_in, in, sqn, sq->sqn); MLX5_SET(modify_sq_in, in, sq_state, MLX5_SQC_STATE_RDY); MLX5_SET64(modify_sq_in, in, modify_bitmask, 1); MLX5_SET(sqc, sqc, state, MLX5_SQC_STATE_RDY); MLX5_SET(sqc, sqc, packet_pacing_rate_limit_index, rl_index); err = mlx5_core_modify_sq(mdev, in, inlen); kvfree(in); return (err); } /* * This function will search the configured rate limit table for the * best match to avoid that a single socket based application can * allocate all the available hardware rates. If the user selected * rate deviates too much from the closes rate available in the rate * limit table, unlimited rate will be selected. */ static uint64_t mlx5e_rl_find_best_rate_locked(struct mlx5e_rl_priv_data *rl, uint64_t user_rate) { uint64_t distance = -1ULL; uint64_t diff; uint64_t retval = 0; /* unlimited */ uint64_t x; /* search for closest rate */ for (x = 0; x != rl->param.tx_rates_def; x++) { uint64_t rate = rl->rate_limit_table[x]; if (rate == 0) continue; if (rate > user_rate) diff = rate - user_rate; else diff = user_rate - rate; /* check if distance is smaller than previous rate */ if (diff < distance) { distance = diff; retval = rate; } } /* range check for multiplication below */ if (user_rate > rl->param.tx_limit_max) user_rate = rl->param.tx_limit_max; /* fallback to unlimited, if rate deviates too much */ if (distance > howmany(user_rate * rl->param.tx_allowed_deviation, 1000ULL)) retval = 0; return (retval); } /* * This function sets the requested rate for a rate limit channel, in * bits per second. The requested rate will be filtered through the * find best rate function above. */ static int mlx5e_rlw_channel_set_rate_locked(struct mlx5e_rl_worker *rlw, struct mlx5e_rl_channel *channel, uint64_t rate) { struct mlx5e_rl_priv_data *rl = &rlw->priv->rl; struct mlx5e_sq *sq; uint64_t temp; uint16_t index; uint16_t burst; int error; if (rate != 0) { MLX5E_RL_WORKER_UNLOCK(rlw); MLX5E_RL_RLOCK(rl); /* get current burst size in bytes */ temp = rl->param.tx_burst_size * MLX5E_SW2HW_MTU(rlw->priv->ifp->if_mtu); /* limit burst size to 64K currently */ if (temp > 65535) temp = 65535; burst = temp; /* find best rate */ rate = mlx5e_rl_find_best_rate_locked(rl, rate); MLX5E_RL_RUNLOCK(rl); if (rate == 0) { /* rate doesn't exist, fallback to unlimited */ index = 0; rate = 0; atomic_add_64(&rlw->priv->rl.stats.tx_modify_rate_failure, 1ULL); } else { /* get a reference on the new rate */ error = -mlx5_rl_add_rate(rlw->priv->mdev, howmany(rate, 1000), burst, &index); if (error != 0) { /* adding rate failed, fallback to unlimited */ index = 0; rate = 0; atomic_add_64(&rlw->priv->rl.stats.tx_add_new_rate_failure, 1ULL); } } MLX5E_RL_WORKER_LOCK(rlw); } else { index = 0; burst = 0; /* default */ } /* atomically swap rates */ temp = channel->last_rate; channel->last_rate = rate; rate = temp; /* atomically swap burst size */ temp = channel->last_burst; channel->last_burst = burst; burst = temp; MLX5E_RL_WORKER_UNLOCK(rlw); /* put reference on the old rate, if any */ if (rate != 0) { mlx5_rl_remove_rate(rlw->priv->mdev, howmany(rate, 1000), burst); } /* set new rate, if SQ is running */ sq = channel->sq; if (sq != NULL && READ_ONCE(sq->running) != 0) { error = mlx5e_rl_modify_sq(sq, index); if (error != 0) atomic_add_64(&rlw->priv->rl.stats.tx_modify_rate_failure, 1ULL); } else error = 0; MLX5E_RL_WORKER_LOCK(rlw); return (-error); } static void mlx5e_rl_worker(void *arg) { struct thread *td; struct mlx5e_rl_worker *rlw = arg; struct mlx5e_rl_channel *channel; struct mlx5e_priv *priv; unsigned ix; uint64_t x; int error; /* set thread priority */ td = curthread; thread_lock(td); sched_prio(td, PI_SWI(SWI_NET)); thread_unlock(td); priv = rlw->priv; /* compute completion vector */ ix = (rlw - priv->rl.workers) % priv->mdev->priv.eq_table.num_comp_vectors; /* TODO bind to CPU */ /* open all the SQs */ MLX5E_RL_WORKER_LOCK(rlw); for (x = 0; x < priv->rl.param.tx_channels_per_worker_def; x++) { struct mlx5e_rl_channel *channel = rlw->channels + x; #if !defined(HAVE_RL_PRE_ALLOCATE_CHANNELS) if (channel->state == MLX5E_RL_ST_FREE) continue; #endif MLX5E_RL_WORKER_UNLOCK(rlw); MLX5E_RL_RLOCK(&priv->rl); error = mlx5e_rl_open_channel(rlw, ix, &priv->rl.chan_param, &channel->sq); MLX5E_RL_RUNLOCK(&priv->rl); MLX5E_RL_WORKER_LOCK(rlw); if (error != 0) { mlx5_en_err(priv->ifp, "mlx5e_rl_open_channel failed: %d\n", error); break; } mlx5e_rlw_channel_set_rate_locked(rlw, channel, channel->init_rate); } while (1) { if (STAILQ_FIRST(&rlw->process_head) == NULL) { /* check if we are tearing down */ if (rlw->worker_done != 0) break; cv_wait(&rlw->cv, &rlw->mtx); } /* check if we are tearing down */ if (rlw->worker_done != 0) break; channel = STAILQ_FIRST(&rlw->process_head); if (channel != NULL) { STAILQ_REMOVE_HEAD(&rlw->process_head, entry); switch (channel->state) { case MLX5E_RL_ST_MODIFY: channel->state = MLX5E_RL_ST_USED; MLX5E_RL_WORKER_UNLOCK(rlw); /* create channel by demand */ if (channel->sq == NULL) { MLX5E_RL_RLOCK(&priv->rl); error = mlx5e_rl_open_channel(rlw, ix, &priv->rl.chan_param, &channel->sq); MLX5E_RL_RUNLOCK(&priv->rl); if (error != 0) { mlx5_en_err(priv->ifp, "mlx5e_rl_open_channel failed: %d\n", error); } else { atomic_add_64(&rlw->priv->rl.stats.tx_open_queues, 1ULL); } } else { mlx5e_resume_sq(channel->sq); } MLX5E_RL_WORKER_LOCK(rlw); /* convert from bytes/s to bits/s and set new rate */ error = mlx5e_rlw_channel_set_rate_locked(rlw, channel, channel->new_rate * 8ULL); if (error != 0) { mlx5_en_err(priv->ifp, "mlx5e_rlw_channel_set_rate_locked failed: %d\n", error); } break; case MLX5E_RL_ST_DESTROY: error = mlx5e_rlw_channel_set_rate_locked(rlw, channel, 0); if (error != 0) { mlx5_en_err(priv->ifp, "mlx5e_rlw_channel_set_rate_locked failed: %d\n", error); } if (channel->sq != NULL) { /* * Make sure all packets are * transmitted before SQ is * returned to free list: */ MLX5E_RL_WORKER_UNLOCK(rlw); mlx5e_drain_sq(channel->sq); MLX5E_RL_WORKER_LOCK(rlw); } /* put the channel back into the free list */ STAILQ_INSERT_HEAD(&rlw->index_list_head, channel, entry); channel->state = MLX5E_RL_ST_FREE; atomic_add_64(&priv->rl.stats.tx_active_connections, -1ULL); break; default: /* NOP */ break; } } } /* close all the SQs */ for (x = 0; x < priv->rl.param.tx_channels_per_worker_def; x++) { struct mlx5e_rl_channel *channel = rlw->channels + x; /* update the initial rate */ channel->init_rate = channel->last_rate; /* make sure we free up the rate resource */ mlx5e_rlw_channel_set_rate_locked(rlw, channel, 0); if (channel->sq != NULL) { MLX5E_RL_WORKER_UNLOCK(rlw); mlx5e_rl_close_channel(&channel->sq); atomic_add_64(&rlw->priv->rl.stats.tx_open_queues, -1ULL); MLX5E_RL_WORKER_LOCK(rlw); } } rlw->worker_done = 0; cv_broadcast(&rlw->cv); MLX5E_RL_WORKER_UNLOCK(rlw); kthread_exit(); } static int mlx5e_rl_open_tis(struct mlx5e_priv *priv) { struct mlx5_core_dev *mdev = priv->mdev; u32 in[MLX5_ST_SZ_DW(create_tis_in)]; void *tisc = MLX5_ADDR_OF(create_tis_in, in, ctx); memset(in, 0, sizeof(in)); MLX5_SET(tisc, tisc, prio, 0); MLX5_SET(tisc, tisc, transport_domain, priv->tdn); return (mlx5_core_create_tis(mdev, in, sizeof(in), &priv->rl.tisn)); } static void mlx5e_rl_close_tis(struct mlx5e_priv *priv) { mlx5_core_destroy_tis(priv->mdev, priv->rl.tisn); } static void mlx5e_rl_set_default_params(struct mlx5e_rl_params *param, struct mlx5_core_dev *mdev) { /* ratelimit workers */ param->tx_worker_threads_def = mdev->priv.eq_table.num_comp_vectors; param->tx_worker_threads_max = MLX5E_RL_MAX_WORKERS; /* range check */ if (param->tx_worker_threads_def == 0 || param->tx_worker_threads_def > param->tx_worker_threads_max) param->tx_worker_threads_def = param->tx_worker_threads_max; /* ratelimit channels */ param->tx_channels_per_worker_def = MLX5E_RL_MAX_SQS / param->tx_worker_threads_def; param->tx_channels_per_worker_max = MLX5E_RL_MAX_SQS; /* range check */ if (param->tx_channels_per_worker_def > MLX5E_RL_DEF_SQ_PER_WORKER) param->tx_channels_per_worker_def = MLX5E_RL_DEF_SQ_PER_WORKER; /* set default burst size */ param->tx_burst_size = 4; /* MTUs */ /* * Set maximum burst size * * The burst size is multiplied by the MTU and clamped to the * range 0 ... 65535 bytes inclusivly before fed into the * firmware. * * NOTE: If the burst size or MTU is changed only ratelimit * connections made after the change will use the new burst * size. */ param->tx_burst_size_max = 255; /* get firmware rate limits in 1000bit/s and convert them to bit/s */ param->tx_limit_min = mdev->priv.rl_table.min_rate * 1000ULL; param->tx_limit_max = mdev->priv.rl_table.max_rate * 1000ULL; /* ratelimit table size */ param->tx_rates_max = mdev->priv.rl_table.max_size; /* range check */ if (param->tx_rates_max > MLX5E_RL_MAX_TX_RATES) param->tx_rates_max = MLX5E_RL_MAX_TX_RATES; /* set default number of rates */ param->tx_rates_def = param->tx_rates_max; /* set maximum allowed rate deviation */ if (param->tx_limit_max != 0) { /* * Make sure the deviation multiplication doesn't * overflow unsigned 64-bit: */ param->tx_allowed_deviation_max = -1ULL / param->tx_limit_max; } /* set default rate deviation */ param->tx_allowed_deviation = 50; /* 5.0% */ /* channel parameters */ param->tx_queue_size = (1 << MLX5E_PARAMS_MINIMUM_LOG_SQ_SIZE); param->tx_coalesce_usecs = MLX5E_RL_TX_COAL_USEC_DEFAULT; param->tx_coalesce_pkts = MLX5E_RL_TX_COAL_PKTS_DEFAULT; param->tx_coalesce_mode = MLX5E_RL_TX_COAL_MODE_DEFAULT; param->tx_completion_fact = MLX5E_RL_TX_COMP_FACT_DEFAULT; } static const char *mlx5e_rl_params_desc[] = { MLX5E_RL_PARAMS(MLX5E_STATS_DESC) }; static const char *mlx5e_rl_table_params_desc[] = { MLX5E_RL_TABLE_PARAMS(MLX5E_STATS_DESC) }; static const char *mlx5e_rl_stats_desc[] = { MLX5E_RL_STATS(MLX5E_STATS_DESC) }; int mlx5e_rl_init(struct mlx5e_priv *priv) { struct mlx5e_rl_priv_data *rl = &priv->rl; struct sysctl_oid *node; struct sysctl_oid *stats; char buf[64]; uint64_t i; uint64_t j; int error; /* check if there is support for packet pacing */ if (!MLX5_CAP_GEN(priv->mdev, qos) || !MLX5_CAP_QOS(priv->mdev, packet_pacing)) return (0); rl->priv = priv; sysctl_ctx_init(&rl->ctx); sx_init(&rl->rl_sxlock, "ratelimit-sxlock"); /* open own TIS domain for ratelimit SQs */ error = mlx5e_rl_open_tis(priv); if (error) goto done; /* setup default value for parameters */ mlx5e_rl_set_default_params(&rl->param, priv->mdev); /* update the completion factor */ mlx5e_rl_sync_tx_completion_fact(rl); /* create root node */ node = SYSCTL_ADD_NODE(&rl->ctx, SYSCTL_CHILDREN(priv->sysctl_ifnet), OID_AUTO, "rate_limit", CTLFLAG_RW | CTLFLAG_MPSAFE, NULL, "Rate limiting support"); if (node != NULL) { /* create SYSCTLs */ for (i = 0; i != MLX5E_RL_PARAMS_NUM; i++) { mlx5e_rl_sysctl_add_u64_oid(rl, MLX5E_RL_PARAMS_INDEX(arg[i]), node, mlx5e_rl_params_desc[2 * i], mlx5e_rl_params_desc[2 * i + 1]); } stats = SYSCTL_ADD_NODE(&rl->ctx, SYSCTL_CHILDREN(node), OID_AUTO, "stats", CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, "Rate limiting statistics"); if (stats != NULL) { /* create SYSCTLs */ for (i = 0; i != MLX5E_RL_STATS_NUM; i++) { mlx5e_rl_sysctl_add_stats_u64_oid(rl, i, stats, mlx5e_rl_stats_desc[2 * i], mlx5e_rl_stats_desc[2 * i + 1]); } } } /* allocate workers array */ rl->workers = malloc(sizeof(rl->workers[0]) * rl->param.tx_worker_threads_def, M_MLX5EN, M_WAITOK | M_ZERO); /* allocate rate limit array */ rl->rate_limit_table = malloc(sizeof(rl->rate_limit_table[0]) * rl->param.tx_rates_def, M_MLX5EN, M_WAITOK | M_ZERO); if (node != NULL) { /* create more SYSCTls */ SYSCTL_ADD_PROC(&rl->ctx, SYSCTL_CHILDREN(node), OID_AUTO, "tx_rate_show", CTLTYPE_STRING | CTLFLAG_RD | CTLFLAG_MPSAFE, rl, 0, &mlx5e_rl_sysctl_show_rate_table, "A", "Show table of all configured TX rates"); /* try to fetch rate table from kernel environment */ for (i = 0; i != rl->param.tx_rates_def; i++) { /* compute path for tunable */ snprintf(buf, sizeof(buf), "dev.mce.%d.rate_limit.tx_rate_add_%d", device_get_unit(priv->mdev->pdev->dev.bsddev), (int)i); if (TUNABLE_QUAD_FETCH(buf, &j)) mlx5e_rl_tx_limit_add(rl, j); } /* setup rate table sysctls */ for (i = 0; i != MLX5E_RL_TABLE_PARAMS_NUM; i++) { mlx5e_rl_sysctl_add_u64_oid(rl, MLX5E_RL_PARAMS_INDEX(table_arg[i]), node, mlx5e_rl_table_params_desc[2 * i], mlx5e_rl_table_params_desc[2 * i + 1]); } } for (j = 0; j < rl->param.tx_worker_threads_def; j++) { struct mlx5e_rl_worker *rlw = rl->workers + j; rlw->priv = priv; cv_init(&rlw->cv, "mlx5-worker-cv"); mtx_init(&rlw->mtx, "mlx5-worker-mtx", NULL, MTX_DEF); STAILQ_INIT(&rlw->index_list_head); STAILQ_INIT(&rlw->process_head); rlw->channels = malloc(sizeof(rlw->channels[0]) * rl->param.tx_channels_per_worker_def, M_MLX5EN, M_WAITOK | M_ZERO); MLX5E_RL_WORKER_LOCK(rlw); for (i = 0; i < rl->param.tx_channels_per_worker_def; i++) { struct mlx5e_rl_channel *channel = rlw->channels + i; channel->worker = rlw; channel->tag.type = IF_SND_TAG_TYPE_RATE_LIMIT; STAILQ_INSERT_TAIL(&rlw->index_list_head, channel, entry); } MLX5E_RL_WORKER_UNLOCK(rlw); } PRIV_LOCK(priv); error = mlx5e_rl_open_workers(priv); PRIV_UNLOCK(priv); if (error != 0) { mlx5_en_err(priv->ifp, "mlx5e_rl_open_workers failed: %d\n", error); } return (0); done: sysctl_ctx_free(&rl->ctx); sx_destroy(&rl->rl_sxlock); return (error); } static int mlx5e_rl_open_workers(struct mlx5e_priv *priv) { struct mlx5e_rl_priv_data *rl = &priv->rl; struct thread *rl_thread = NULL; struct proc *rl_proc = NULL; uint64_t j; int error; if (priv->gone || rl->opened) return (-EINVAL); MLX5E_RL_WLOCK(rl); /* compute channel parameters once */ mlx5e_rl_build_channel_param(rl, &rl->chan_param); MLX5E_RL_WUNLOCK(rl); for (j = 0; j < rl->param.tx_worker_threads_def; j++) { struct mlx5e_rl_worker *rlw = rl->workers + j; /* start worker thread */ error = kproc_kthread_add(mlx5e_rl_worker, rlw, &rl_proc, &rl_thread, RFHIGHPID, 0, "mlx5-ratelimit", "mlx5-rl-worker-thread-%d", (int)j); if (error != 0) { mlx5_en_err(rl->priv->ifp, "kproc_kthread_add failed: %d\n", error); rlw->worker_done = 1; } } rl->opened = 1; return (0); } static void mlx5e_rl_close_workers(struct mlx5e_priv *priv) { struct mlx5e_rl_priv_data *rl = &priv->rl; uint64_t y; if (rl->opened == 0) return; /* tear down worker threads simultaneously */ for (y = 0; y < rl->param.tx_worker_threads_def; y++) { struct mlx5e_rl_worker *rlw = rl->workers + y; /* tear down worker before freeing SQs */ MLX5E_RL_WORKER_LOCK(rlw); if (rlw->worker_done == 0) { rlw->worker_done = 1; cv_broadcast(&rlw->cv); } else { /* XXX thread not started */ rlw->worker_done = 0; } MLX5E_RL_WORKER_UNLOCK(rlw); } /* wait for worker threads to exit */ for (y = 0; y < rl->param.tx_worker_threads_def; y++) { struct mlx5e_rl_worker *rlw = rl->workers + y; /* tear down worker before freeing SQs */ MLX5E_RL_WORKER_LOCK(rlw); while (rlw->worker_done != 0) cv_wait(&rlw->cv, &rlw->mtx); MLX5E_RL_WORKER_UNLOCK(rlw); } rl->opened = 0; } static void mlx5e_rl_reset_rates(struct mlx5e_rl_priv_data *rl) { unsigned x; MLX5E_RL_WLOCK(rl); for (x = 0; x != rl->param.tx_rates_def; x++) rl->rate_limit_table[x] = 0; MLX5E_RL_WUNLOCK(rl); } void mlx5e_rl_cleanup(struct mlx5e_priv *priv) { struct mlx5e_rl_priv_data *rl = &priv->rl; uint64_t y; /* check if there is support for packet pacing */ if (!MLX5_CAP_GEN(priv->mdev, qos) || !MLX5_CAP_QOS(priv->mdev, packet_pacing)) return; /* TODO check if there is support for packet pacing */ sysctl_ctx_free(&rl->ctx); PRIV_LOCK(priv); mlx5e_rl_close_workers(priv); PRIV_UNLOCK(priv); mlx5e_rl_reset_rates(rl); /* close TIS domain */ mlx5e_rl_close_tis(priv); for (y = 0; y < rl->param.tx_worker_threads_def; y++) { struct mlx5e_rl_worker *rlw = rl->workers + y; cv_destroy(&rlw->cv); mtx_destroy(&rlw->mtx); free(rlw->channels, M_MLX5EN); } free(rl->rate_limit_table, M_MLX5EN); free(rl->workers, M_MLX5EN); sx_destroy(&rl->rl_sxlock); } static void mlx5e_rlw_queue_channel_locked(struct mlx5e_rl_worker *rlw, struct mlx5e_rl_channel *channel) { STAILQ_INSERT_TAIL(&rlw->process_head, channel, entry); cv_broadcast(&rlw->cv); } static void mlx5e_rl_free(struct mlx5e_rl_worker *rlw, struct mlx5e_rl_channel *channel) { if (channel == NULL) return; MLX5E_RL_WORKER_LOCK(rlw); switch (channel->state) { case MLX5E_RL_ST_MODIFY: channel->state = MLX5E_RL_ST_DESTROY; break; case MLX5E_RL_ST_USED: channel->state = MLX5E_RL_ST_DESTROY; mlx5e_rlw_queue_channel_locked(rlw, channel); break; default: break; } MLX5E_RL_WORKER_UNLOCK(rlw); } static int mlx5e_rl_modify(struct mlx5e_rl_worker *rlw, struct mlx5e_rl_channel *channel, uint64_t rate) { MLX5E_RL_WORKER_LOCK(rlw); channel->new_rate = rate; switch (channel->state) { case MLX5E_RL_ST_USED: channel->state = MLX5E_RL_ST_MODIFY; mlx5e_rlw_queue_channel_locked(rlw, channel); break; default: break; } MLX5E_RL_WORKER_UNLOCK(rlw); return (0); } static int mlx5e_rl_query(struct mlx5e_rl_worker *rlw, struct mlx5e_rl_channel *channel, union if_snd_tag_query_params *params) { int retval; MLX5E_RL_WORKER_LOCK(rlw); switch (channel->state) { case MLX5E_RL_ST_USED: params->rate_limit.max_rate = channel->last_rate; params->rate_limit.queue_level = mlx5e_sq_queue_level(channel->sq); retval = 0; break; case MLX5E_RL_ST_MODIFY: params->rate_limit.max_rate = channel->last_rate; params->rate_limit.queue_level = mlx5e_sq_queue_level(channel->sq); retval = EBUSY; break; default: retval = EINVAL; break; } MLX5E_RL_WORKER_UNLOCK(rlw); return (retval); } static int mlx5e_find_available_tx_ring_index(struct mlx5e_rl_worker *rlw, struct mlx5e_rl_channel **pchannel) { struct mlx5e_rl_channel *channel; int retval = ENOMEM; MLX5E_RL_WORKER_LOCK(rlw); /* Check for available channel in free list */ if ((channel = STAILQ_FIRST(&rlw->index_list_head)) != NULL) { retval = 0; /* Remove head index from available list */ STAILQ_REMOVE_HEAD(&rlw->index_list_head, entry); channel->state = MLX5E_RL_ST_USED; atomic_add_64(&rlw->priv->rl.stats.tx_active_connections, 1ULL); } else { atomic_add_64(&rlw->priv->rl.stats.tx_available_resource_failure, 1ULL); } MLX5E_RL_WORKER_UNLOCK(rlw); *pchannel = channel; #ifdef RATELIMIT_DEBUG mlx5_en_info(rlw->priv->ifp, "Channel pointer for rate limit connection is %p\n", channel); #endif return (retval); } int mlx5e_rl_snd_tag_alloc(struct ifnet *ifp, union if_snd_tag_alloc_params *params, struct m_snd_tag **ppmt) { struct mlx5e_rl_channel *channel; struct mlx5e_rl_worker *rlw; struct mlx5e_priv *priv; int error; priv = ifp->if_softc; /* check if there is support for packet pacing or if device is going away */ if (!MLX5_CAP_GEN(priv->mdev, qos) || !MLX5_CAP_QOS(priv->mdev, packet_pacing) || priv->gone || params->rate_limit.hdr.type != IF_SND_TAG_TYPE_RATE_LIMIT) return (EOPNOTSUPP); /* compute worker thread this TCP connection belongs to */ rlw = priv->rl.workers + ((params->rate_limit.hdr.flowid % 128) % priv->rl.param.tx_worker_threads_def); error = mlx5e_find_available_tx_ring_index(rlw, &channel); if (error != 0) goto done; error = mlx5e_rl_modify(rlw, channel, params->rate_limit.max_rate); if (error != 0) { mlx5e_rl_free(rlw, channel); goto done; } /* store pointer to mbuf tag */ MPASS(channel->tag.refcount == 0); m_snd_tag_init(&channel->tag, ifp, IF_SND_TAG_TYPE_RATE_LIMIT); *ppmt = &channel->tag; done: return (error); } int mlx5e_rl_snd_tag_modify(struct m_snd_tag *pmt, union if_snd_tag_modify_params *params) { struct mlx5e_rl_channel *channel = container_of(pmt, struct mlx5e_rl_channel, tag); return (mlx5e_rl_modify(channel->worker, channel, params->rate_limit.max_rate)); } int mlx5e_rl_snd_tag_query(struct m_snd_tag *pmt, union if_snd_tag_query_params *params) { struct mlx5e_rl_channel *channel = container_of(pmt, struct mlx5e_rl_channel, tag); return (mlx5e_rl_query(channel->worker, channel, params)); } void mlx5e_rl_snd_tag_free(struct m_snd_tag *pmt) { struct mlx5e_rl_channel *channel = container_of(pmt, struct mlx5e_rl_channel, tag); mlx5e_rl_free(channel->worker, channel); } static int mlx5e_rl_sysctl_show_rate_table(SYSCTL_HANDLER_ARGS) { struct mlx5e_rl_priv_data *rl = arg1; struct mlx5e_priv *priv = rl->priv; struct sbuf sbuf; unsigned x; int error; error = sysctl_wire_old_buffer(req, 0); if (error != 0) return (error); PRIV_LOCK(priv); sbuf_new_for_sysctl(&sbuf, NULL, 128 * rl->param.tx_rates_def, req); sbuf_printf(&sbuf, "\n\n" "\t" "ENTRY" "\t" "BURST" "\t" "RATE [bit/s]\n" "\t" "--------------------------------------------\n"); MLX5E_RL_RLOCK(rl); for (x = 0; x != rl->param.tx_rates_def; x++) { if (rl->rate_limit_table[x] == 0) continue; sbuf_printf(&sbuf, "\t" "%3u" "\t" "%3u" "\t" "%lld\n", x, (unsigned)rl->param.tx_burst_size, (long long)rl->rate_limit_table[x]); } MLX5E_RL_RUNLOCK(rl); error = sbuf_finish(&sbuf); sbuf_delete(&sbuf); PRIV_UNLOCK(priv); return (error); } static int mlx5e_rl_refresh_channel_params(struct mlx5e_rl_priv_data *rl) { uint64_t x; uint64_t y; MLX5E_RL_WLOCK(rl); /* compute channel parameters once */ mlx5e_rl_build_channel_param(rl, &rl->chan_param); MLX5E_RL_WUNLOCK(rl); for (y = 0; y != rl->param.tx_worker_threads_def; y++) { struct mlx5e_rl_worker *rlw = rl->workers + y; for (x = 0; x != rl->param.tx_channels_per_worker_def; x++) { struct mlx5e_rl_channel *channel; struct mlx5e_sq *sq; channel = rlw->channels + x; sq = channel->sq; if (sq == NULL) continue; if (MLX5_CAP_GEN(rl->priv->mdev, cq_period_mode_modify)) { mlx5_core_modify_cq_moderation_mode(rl->priv->mdev, &sq->cq.mcq, rl->param.tx_coalesce_usecs, rl->param.tx_coalesce_pkts, rl->param.tx_coalesce_mode); } else { mlx5_core_modify_cq_moderation(rl->priv->mdev, &sq->cq.mcq, rl->param.tx_coalesce_usecs, rl->param.tx_coalesce_pkts); } } } return (0); } void mlx5e_rl_refresh_sq_inline(struct mlx5e_rl_priv_data *rl) { uint64_t x; uint64_t y; for (y = 0; y != rl->param.tx_worker_threads_def; y++) { struct mlx5e_rl_worker *rlw = rl->workers + y; for (x = 0; x != rl->param.tx_channels_per_worker_def; x++) { struct mlx5e_rl_channel *channel; struct mlx5e_sq *sq; channel = rlw->channels + x; sq = channel->sq; if (sq == NULL) continue; mtx_lock(&sq->lock); mlx5e_update_sq_inline(sq); mtx_unlock(&sq->lock); } } } static int mlx5e_rl_tx_limit_add(struct mlx5e_rl_priv_data *rl, uint64_t value) { unsigned x; int error; if (value < 1000 || mlx5_rl_is_in_range(rl->priv->mdev, howmany(value, 1000), 0) == 0) return (EINVAL); MLX5E_RL_WLOCK(rl); error = ENOMEM; /* check if rate already exists */ for (x = 0; x != rl->param.tx_rates_def; x++) { if (rl->rate_limit_table[x] != value) continue; error = EEXIST; break; } /* check if there is a free rate entry */ if (x == rl->param.tx_rates_def) { for (x = 0; x != rl->param.tx_rates_def; x++) { if (rl->rate_limit_table[x] != 0) continue; rl->rate_limit_table[x] = value; error = 0; break; } } MLX5E_RL_WUNLOCK(rl); return (error); } static int mlx5e_rl_tx_limit_clr(struct mlx5e_rl_priv_data *rl, uint64_t value) { unsigned x; int error; if (value == 0) return (EINVAL); MLX5E_RL_WLOCK(rl); /* check if rate already exists */ for (x = 0; x != rl->param.tx_rates_def; x++) { if (rl->rate_limit_table[x] != value) continue; /* free up rate */ rl->rate_limit_table[x] = 0; break; } /* check if there is a free rate entry */ if (x == rl->param.tx_rates_def) error = ENOENT; else error = 0; MLX5E_RL_WUNLOCK(rl); return (error); } static int mlx5e_rl_sysctl_handler(SYSCTL_HANDLER_ARGS) { struct mlx5e_rl_priv_data *rl = arg1; struct mlx5e_priv *priv = rl->priv; unsigned mode_modify; unsigned was_opened; uint64_t value; uint64_t old; int error; PRIV_LOCK(priv); MLX5E_RL_RLOCK(rl); value = rl->param.arg[arg2]; MLX5E_RL_RUNLOCK(rl); if (req != NULL) { old = value; error = sysctl_handle_64(oidp, &value, 0, req); if (error || req->newptr == NULL || value == rl->param.arg[arg2]) goto done; } else { old = 0; error = 0; } /* check if device is gone */ if (priv->gone) { error = ENXIO; goto done; } was_opened = rl->opened; mode_modify = MLX5_CAP_GEN(priv->mdev, cq_period_mode_modify); switch (MLX5E_RL_PARAMS_INDEX(arg[arg2])) { case MLX5E_RL_PARAMS_INDEX(tx_worker_threads_def): if (value > rl->param.tx_worker_threads_max) value = rl->param.tx_worker_threads_max; else if (value < 1) value = 1; /* store new value */ rl->param.arg[arg2] = value; break; case MLX5E_RL_PARAMS_INDEX(tx_channels_per_worker_def): if (value > rl->param.tx_channels_per_worker_max) value = rl->param.tx_channels_per_worker_max; else if (value < 1) value = 1; /* store new value */ rl->param.arg[arg2] = value; break; case MLX5E_RL_PARAMS_INDEX(tx_rates_def): if (value > rl->param.tx_rates_max) value = rl->param.tx_rates_max; else if (value < 1) value = 1; /* store new value */ rl->param.arg[arg2] = value; break; case MLX5E_RL_PARAMS_INDEX(tx_coalesce_usecs): /* range check */ if (value < 1) value = 0; else if (value > MLX5E_FLD_MAX(cqc, cq_period)) value = MLX5E_FLD_MAX(cqc, cq_period); /* store new value */ rl->param.arg[arg2] = value; /* check to avoid down and up the network interface */ if (was_opened) error = mlx5e_rl_refresh_channel_params(rl); break; case MLX5E_RL_PARAMS_INDEX(tx_coalesce_pkts): /* import TX coal pkts */ if (value < 1) value = 0; else if (value > MLX5E_FLD_MAX(cqc, cq_max_count)) value = MLX5E_FLD_MAX(cqc, cq_max_count); /* store new value */ rl->param.arg[arg2] = value; /* check to avoid down and up the network interface */ if (was_opened) error = mlx5e_rl_refresh_channel_params(rl); break; case MLX5E_RL_PARAMS_INDEX(tx_coalesce_mode): /* network interface must be down */ if (was_opened != 0 && mode_modify == 0) mlx5e_rl_close_workers(priv); /* import TX coalesce mode */ if (value != 0) value = 1; /* store new value */ rl->param.arg[arg2] = value; /* restart network interface, if any */ if (was_opened != 0) { if (mode_modify == 0) mlx5e_rl_open_workers(priv); else error = mlx5e_rl_refresh_channel_params(rl); } break; case MLX5E_RL_PARAMS_INDEX(tx_queue_size): /* network interface must be down */ if (was_opened) mlx5e_rl_close_workers(priv); /* import TX queue size */ if (value < (1 << MLX5E_PARAMS_MINIMUM_LOG_SQ_SIZE)) value = (1 << MLX5E_PARAMS_MINIMUM_LOG_SQ_SIZE); else if (value > priv->params_ethtool.tx_queue_size_max) value = priv->params_ethtool.tx_queue_size_max; /* store actual TX queue size */ value = 1ULL << order_base_2(value); /* store new value */ rl->param.arg[arg2] = value; /* verify TX completion factor */ mlx5e_rl_sync_tx_completion_fact(rl); /* restart network interface, if any */ if (was_opened) mlx5e_rl_open_workers(priv); break; case MLX5E_RL_PARAMS_INDEX(tx_completion_fact): /* network interface must be down */ if (was_opened) mlx5e_rl_close_workers(priv); /* store new value */ rl->param.arg[arg2] = value; /* verify parameter */ mlx5e_rl_sync_tx_completion_fact(rl); /* restart network interface, if any */ if (was_opened) mlx5e_rl_open_workers(priv); break; case MLX5E_RL_PARAMS_INDEX(tx_limit_add): error = mlx5e_rl_tx_limit_add(rl, value); break; case MLX5E_RL_PARAMS_INDEX(tx_limit_clr): error = mlx5e_rl_tx_limit_clr(rl, value); break; case MLX5E_RL_PARAMS_INDEX(tx_allowed_deviation): /* range check */ if (value > rl->param.tx_allowed_deviation_max) value = rl->param.tx_allowed_deviation_max; else if (value < rl->param.tx_allowed_deviation_min) value = rl->param.tx_allowed_deviation_min; MLX5E_RL_WLOCK(rl); rl->param.arg[arg2] = value; MLX5E_RL_WUNLOCK(rl); break; case MLX5E_RL_PARAMS_INDEX(tx_burst_size): /* range check */ if (value > rl->param.tx_burst_size_max) value = rl->param.tx_burst_size_max; else if (value < rl->param.tx_burst_size_min) value = rl->param.tx_burst_size_min; MLX5E_RL_WLOCK(rl); rl->param.arg[arg2] = value; MLX5E_RL_WUNLOCK(rl); break; default: break; } done: PRIV_UNLOCK(priv); return (error); } static void mlx5e_rl_sysctl_add_u64_oid(struct mlx5e_rl_priv_data *rl, unsigned x, struct sysctl_oid *node, const char *name, const char *desc) { /* * NOTE: In FreeBSD-11 and newer the CTLFLAG_RWTUN flag will * take care of loading default sysctl value from the kernel * environment, if any: */ if (strstr(name, "_max") != 0 || strstr(name, "_min") != 0) { /* read-only SYSCTLs */ SYSCTL_ADD_PROC(&rl->ctx, SYSCTL_CHILDREN(node), OID_AUTO, name, CTLTYPE_U64 | CTLFLAG_RD | CTLFLAG_MPSAFE, rl, x, &mlx5e_rl_sysctl_handler, "QU", desc); } else { if (strstr(name, "_def") != 0) { #ifdef RATELIMIT_DEBUG /* tunable read-only advanced SYSCTLs */ SYSCTL_ADD_PROC(&rl->ctx, SYSCTL_CHILDREN(node), OID_AUTO, name, CTLTYPE_U64 | CTLFLAG_RDTUN | CTLFLAG_MPSAFE, rl, x, &mlx5e_rl_sysctl_handler, "QU", desc); #endif } else { /* read-write SYSCTLs */ SYSCTL_ADD_PROC(&rl->ctx, SYSCTL_CHILDREN(node), OID_AUTO, name, CTLTYPE_U64 | CTLFLAG_RWTUN | CTLFLAG_MPSAFE, rl, x, &mlx5e_rl_sysctl_handler, "QU", desc); } } } static void mlx5e_rl_sysctl_add_stats_u64_oid(struct mlx5e_rl_priv_data *rl, unsigned x, struct sysctl_oid *node, const char *name, const char *desc) { /* read-only SYSCTLs */ SYSCTL_ADD_U64(&rl->ctx, SYSCTL_CHILDREN(node), OID_AUTO, name, CTLFLAG_RD, &rl->stats.arg[x], 0, desc); } #else int mlx5e_rl_init(struct mlx5e_priv *priv) { return (0); } void mlx5e_rl_cleanup(struct mlx5e_priv *priv) { /* NOP */ } #endif /* RATELIMIT */