diff options
Diffstat (limited to 'drivers/infiniband/hw/efa')
| -rw-r--r-- | drivers/infiniband/hw/efa/Kconfig | 15 | ||||
| -rw-r--r-- | drivers/infiniband/hw/efa/Makefile | 9 | ||||
| -rw-r--r-- | drivers/infiniband/hw/efa/efa.h | 163 | ||||
| -rw-r--r-- | drivers/infiniband/hw/efa/efa_admin_cmds_defs.h | 794 | ||||
| -rw-r--r-- | drivers/infiniband/hw/efa/efa_admin_defs.h | 136 | ||||
| -rw-r--r-- | drivers/infiniband/hw/efa/efa_com.c | 1160 | ||||
| -rw-r--r-- | drivers/infiniband/hw/efa/efa_com.h | 144 | ||||
| -rw-r--r-- | drivers/infiniband/hw/efa/efa_com_cmd.c | 692 | ||||
| -rw-r--r-- | drivers/infiniband/hw/efa/efa_com_cmd.h | 270 | ||||
| -rw-r--r-- | drivers/infiniband/hw/efa/efa_common_defs.h | 18 | ||||
| -rw-r--r-- | drivers/infiniband/hw/efa/efa_main.c | 533 | ||||
| -rw-r--r-- | drivers/infiniband/hw/efa/efa_regs_defs.h | 113 | ||||
| -rw-r--r-- | drivers/infiniband/hw/efa/efa_verbs.c | 1825 |
13 files changed, 5872 insertions, 0 deletions
diff --git a/drivers/infiniband/hw/efa/Kconfig b/drivers/infiniband/hw/efa/Kconfig new file mode 100644 index 000000000000..457e18ba1d57 --- /dev/null +++ b/drivers/infiniband/hw/efa/Kconfig @@ -0,0 +1,15 @@ +# SPDX-License-Identifier: GPL-2.0 OR BSD-2-Clause +# Copyright 2018-2019 Amazon.com, Inc. or its affiliates. All rights reserved. +# +# Amazon fabric device configuration +# + +config INFINIBAND_EFA + tristate "Amazon Elastic Fabric Adapter (EFA) support" + depends on PCI_MSI && 64BIT && !CPU_BIG_ENDIAN + depends on INFINIBAND_USER_ACCESS + help + This driver supports Amazon Elastic Fabric Adapter (EFA). + + To compile this driver as a module, choose M here. + The module will be called efa. diff --git a/drivers/infiniband/hw/efa/Makefile b/drivers/infiniband/hw/efa/Makefile new file mode 100644 index 000000000000..6e83083af0bc --- /dev/null +++ b/drivers/infiniband/hw/efa/Makefile @@ -0,0 +1,9 @@ +# SPDX-License-Identifier: GPL-2.0 OR BSD-2-Clause +# Copyright 2018-2019 Amazon.com, Inc. or its affiliates. All rights reserved. +# +# Makefile for Amazon Elastic Fabric Adapter (EFA) device driver. +# + +obj-$(CONFIG_INFINIBAND_EFA) += efa.o + +efa-y := efa_com_cmd.o efa_com.o efa_main.o efa_verbs.o diff --git a/drivers/infiniband/hw/efa/efa.h b/drivers/infiniband/hw/efa/efa.h new file mode 100644 index 000000000000..9e3cc3239c13 --- /dev/null +++ b/drivers/infiniband/hw/efa/efa.h @@ -0,0 +1,163 @@ +/* SPDX-License-Identifier: GPL-2.0 OR BSD-2-Clause */ +/* + * Copyright 2018-2019 Amazon.com, Inc. or its affiliates. All rights reserved. + */ + +#ifndef _EFA_H_ +#define _EFA_H_ + +#include <linux/bitops.h> +#include <linux/idr.h> +#include <linux/interrupt.h> +#include <linux/pci.h> +#include <linux/sched.h> + +#include <rdma/efa-abi.h> +#include <rdma/ib_verbs.h> + +#include "efa_com_cmd.h" + +#define DRV_MODULE_NAME "efa" +#define DEVICE_NAME "Elastic Fabric Adapter (EFA)" + +#define EFA_IRQNAME_SIZE 40 + +/* 1 for AENQ + ADMIN */ +#define EFA_NUM_MSIX_VEC 1 +#define EFA_MGMNT_MSIX_VEC_IDX 0 + +struct efa_irq { + irq_handler_t handler; + void *data; + int cpu; + u32 vector; + cpumask_t affinity_hint_mask; + char name[EFA_IRQNAME_SIZE]; +}; + +struct efa_sw_stats { + atomic64_t alloc_pd_err; + atomic64_t create_qp_err; + atomic64_t create_cq_err; + atomic64_t reg_mr_err; + atomic64_t alloc_ucontext_err; + atomic64_t create_ah_err; +}; + +/* Don't use anything other than atomic64 */ +struct efa_stats { + struct efa_sw_stats sw_stats; + atomic64_t keep_alive_rcvd; +}; + +struct efa_dev { + struct ib_device ibdev; + struct efa_com_dev edev; + struct pci_dev *pdev; + struct efa_com_get_device_attr_result dev_attr; + + u64 reg_bar_addr; + u64 reg_bar_len; + u64 mem_bar_addr; + u64 mem_bar_len; + u64 db_bar_addr; + u64 db_bar_len; + u8 addr[EFA_GID_SIZE]; + u32 mtu; + + int admin_msix_vector_idx; + struct efa_irq admin_irq; + + struct efa_stats stats; +}; + +struct efa_ucontext { + struct ib_ucontext ibucontext; + struct xarray mmap_xa; + u32 mmap_xa_page; + u16 uarn; +}; + +struct efa_pd { + struct ib_pd ibpd; + u16 pdn; +}; + +struct efa_mr { + struct ib_mr ibmr; + struct ib_umem *umem; +}; + +struct efa_cq { + struct ib_cq ibcq; + struct efa_ucontext *ucontext; + dma_addr_t dma_addr; + void *cpu_addr; + size_t size; + u16 cq_idx; +}; + +struct efa_qp { + struct ib_qp ibqp; + dma_addr_t rq_dma_addr; + void *rq_cpu_addr; + size_t rq_size; + enum ib_qp_state state; + u32 qp_handle; + u32 max_send_wr; + u32 max_recv_wr; + u32 max_send_sge; + u32 max_recv_sge; + u32 max_inline_data; +}; + +struct efa_ah { + struct ib_ah ibah; + u16 ah; + /* dest_addr */ + u8 id[EFA_GID_SIZE]; +}; + +int efa_query_device(struct ib_device *ibdev, + struct ib_device_attr *props, + struct ib_udata *udata); +int efa_query_port(struct ib_device *ibdev, u8 port, + struct ib_port_attr *props); +int efa_query_qp(struct ib_qp *ibqp, struct ib_qp_attr *qp_attr, + int qp_attr_mask, + struct ib_qp_init_attr *qp_init_attr); +int efa_query_gid(struct ib_device *ibdev, u8 port, int index, + union ib_gid *gid); +int efa_query_pkey(struct ib_device *ibdev, u8 port, u16 index, + u16 *pkey); +int efa_alloc_pd(struct ib_pd *ibpd, struct ib_udata *udata); +void efa_dealloc_pd(struct ib_pd *ibpd, struct ib_udata *udata); +int efa_destroy_qp(struct ib_qp *ibqp, struct ib_udata *udata); +struct ib_qp *efa_create_qp(struct ib_pd *ibpd, + struct ib_qp_init_attr *init_attr, + struct ib_udata *udata); +int efa_destroy_cq(struct ib_cq *ibcq, struct ib_udata *udata); +struct ib_cq *efa_create_cq(struct ib_device *ibdev, + const struct ib_cq_init_attr *attr, + struct ib_udata *udata); +struct ib_mr *efa_reg_mr(struct ib_pd *ibpd, u64 start, u64 length, + u64 virt_addr, int access_flags, + struct ib_udata *udata); +int efa_dereg_mr(struct ib_mr *ibmr, struct ib_udata *udata); +int efa_get_port_immutable(struct ib_device *ibdev, u8 port_num, + struct ib_port_immutable *immutable); +int efa_alloc_ucontext(struct ib_ucontext *ibucontext, struct ib_udata *udata); +void efa_dealloc_ucontext(struct ib_ucontext *ibucontext); +int efa_mmap(struct ib_ucontext *ibucontext, + struct vm_area_struct *vma); +int efa_create_ah(struct ib_ah *ibah, + struct rdma_ah_attr *ah_attr, + u32 flags, + struct ib_udata *udata); +void efa_destroy_ah(struct ib_ah *ibah, u32 flags); +int efa_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *qp_attr, + int qp_attr_mask, struct ib_udata *udata); +enum rdma_link_layer efa_port_link_layer(struct ib_device *ibdev, + u8 port_num); + +#endif /* _EFA_H_ */ diff --git a/drivers/infiniband/hw/efa/efa_admin_cmds_defs.h b/drivers/infiniband/hw/efa/efa_admin_cmds_defs.h new file mode 100644 index 000000000000..2be0469d545f --- /dev/null +++ b/drivers/infiniband/hw/efa/efa_admin_cmds_defs.h @@ -0,0 +1,794 @@ +/* SPDX-License-Identifier: GPL-2.0 OR BSD-2-Clause */ +/* + * Copyright 2018-2019 Amazon.com, Inc. or its affiliates. All rights reserved. + */ + +#ifndef _EFA_ADMIN_CMDS_H_ +#define _EFA_ADMIN_CMDS_H_ + +#define EFA_ADMIN_API_VERSION_MAJOR 0 +#define EFA_ADMIN_API_VERSION_MINOR 1 + +/* EFA admin queue opcodes */ +enum efa_admin_aq_opcode { + EFA_ADMIN_CREATE_QP = 1, + EFA_ADMIN_MODIFY_QP = 2, + EFA_ADMIN_QUERY_QP = 3, + EFA_ADMIN_DESTROY_QP = 4, + EFA_ADMIN_CREATE_AH = 5, + EFA_ADMIN_DESTROY_AH = 6, + EFA_ADMIN_REG_MR = 7, + EFA_ADMIN_DEREG_MR = 8, + EFA_ADMIN_CREATE_CQ = 9, + EFA_ADMIN_DESTROY_CQ = 10, + EFA_ADMIN_GET_FEATURE = 11, + EFA_ADMIN_SET_FEATURE = 12, + EFA_ADMIN_GET_STATS = 13, + EFA_ADMIN_ALLOC_PD = 14, + EFA_ADMIN_DEALLOC_PD = 15, + EFA_ADMIN_ALLOC_UAR = 16, + EFA_ADMIN_DEALLOC_UAR = 17, + EFA_ADMIN_MAX_OPCODE = 17, +}; + +enum efa_admin_aq_feature_id { + EFA_ADMIN_DEVICE_ATTR = 1, + EFA_ADMIN_AENQ_CONFIG = 2, + EFA_ADMIN_NETWORK_ATTR = 3, + EFA_ADMIN_QUEUE_ATTR = 4, + EFA_ADMIN_HW_HINTS = 5, + EFA_ADMIN_FEATURES_OPCODE_NUM = 8, +}; + +/* QP transport type */ +enum efa_admin_qp_type { + /* Unreliable Datagram */ + EFA_ADMIN_QP_TYPE_UD = 1, + /* Scalable Reliable Datagram */ + EFA_ADMIN_QP_TYPE_SRD = 2, +}; + +/* QP state */ +enum efa_admin_qp_state { + EFA_ADMIN_QP_STATE_RESET = 0, + EFA_ADMIN_QP_STATE_INIT = 1, + EFA_ADMIN_QP_STATE_RTR = 2, + EFA_ADMIN_QP_STATE_RTS = 3, + EFA_ADMIN_QP_STATE_SQD = 4, + EFA_ADMIN_QP_STATE_SQE = 5, + EFA_ADMIN_QP_STATE_ERR = 6, +}; + +enum efa_admin_get_stats_type { + EFA_ADMIN_GET_STATS_TYPE_BASIC = 0, +}; + +enum efa_admin_get_stats_scope { + EFA_ADMIN_GET_STATS_SCOPE_ALL = 0, + EFA_ADMIN_GET_STATS_SCOPE_QUEUE = 1, +}; + +enum efa_admin_modify_qp_mask_bits { + EFA_ADMIN_QP_STATE_BIT = 0, + EFA_ADMIN_CUR_QP_STATE_BIT = 1, + EFA_ADMIN_QKEY_BIT = 2, + EFA_ADMIN_SQ_PSN_BIT = 3, + EFA_ADMIN_SQ_DRAINED_ASYNC_NOTIFY_BIT = 4, +}; + +/* + * QP allocation sizes, converted by fabric QueuePair (QP) create command + * from QP capabilities. + */ +struct efa_admin_qp_alloc_size { + /* Send descriptor ring size in bytes */ + u32 send_queue_ring_size; + + /* Max number of WQEs that can be outstanding on send queue. */ + u32 send_queue_depth; + + /* + * Recv descriptor ring size in bytes, sufficient for user-provided + * number of WQEs + */ + u32 recv_queue_ring_size; + + /* Max number of WQEs that can be outstanding on recv queue */ + u32 recv_queue_depth; +}; + +struct efa_admin_create_qp_cmd { + /* Common Admin Queue descriptor */ + struct efa_admin_aq_common_desc aq_common_desc; + + /* Protection Domain associated with this QP */ + u16 pd; + + /* QP type */ + u8 qp_type; + + /* + * 0 : sq_virt - If set, SQ ring base address is + * virtual (IOVA returned by MR registration) + * 1 : rq_virt - If set, RQ ring base address is + * virtual (IOVA returned by MR registration) + * 7:2 : reserved - MBZ + */ + u8 flags; + + /* + * Send queue (SQ) ring base physical address. This field is not + * used if this is a Low Latency Queue(LLQ). + */ + u64 sq_base_addr; + + /* Receive queue (RQ) ring base address. */ + u64 rq_base_addr; + + /* Index of CQ to be associated with Send Queue completions */ + u32 send_cq_idx; + + /* Index of CQ to be associated with Recv Queue completions */ + u32 recv_cq_idx; + + /* + * Memory registration key for the SQ ring, used only when not in + * LLQ mode and base address is virtual + */ + u32 sq_l_key; + + /* + * Memory registration key for the RQ ring, used only when base + * address is virtual + */ + u32 rq_l_key; + + /* Requested QP allocation sizes */ + struct efa_admin_qp_alloc_size qp_alloc_size; + + /* UAR number */ + u16 uar; + + /* MBZ */ + u16 reserved; + + /* MBZ */ + u32 reserved2; +}; + +struct efa_admin_create_qp_resp { + /* Common Admin Queue completion descriptor */ + struct efa_admin_acq_common_desc acq_common_desc; + + /* Opaque handle to be used for consequent operations on the QP */ + u32 qp_handle; + + /* QP number in the given EFA virtual device */ + u16 qp_num; + + /* MBZ */ + u16 reserved; + + /* Index of sub-CQ for Send Queue completions */ + u16 send_sub_cq_idx; + + /* Index of sub-CQ for Receive Queue completions */ + u16 recv_sub_cq_idx; + + /* SQ doorbell address, as offset to PCIe DB BAR */ + u32 sq_db_offset; + + /* RQ doorbell address, as offset to PCIe DB BAR */ + u32 rq_db_offset; + + /* + * low latency send queue ring base address as an offset to PCIe + * MMIO LLQ_MEM BAR + */ + u32 llq_descriptors_offset; +}; + +struct efa_admin_modify_qp_cmd { + /* Common Admin Queue descriptor */ + struct efa_admin_aq_common_desc aq_common_desc; + + /* + * Mask indicating which fields should be updated see enum + * efa_admin_modify_qp_mask_bits + */ + u32 modify_mask; + + /* QP handle returned by create_qp command */ + u32 qp_handle; + + /* QP state */ + u32 qp_state; + + /* Override current QP state (before applying the transition) */ + u32 cur_qp_state; + + /* QKey */ + u32 qkey; + + /* SQ PSN */ + u32 sq_psn; + + /* Enable async notification when SQ is drained */ + u8 sq_drained_async_notify; + + /* MBZ */ + u8 reserved1; + + /* MBZ */ + u16 reserved2; +}; + +struct efa_admin_modify_qp_resp { + /* Common Admin Queue completion descriptor */ + struct efa_admin_acq_common_desc acq_common_desc; +}; + +struct efa_admin_query_qp_cmd { + /* Common Admin Queue descriptor */ + struct efa_admin_aq_common_desc aq_common_desc; + + /* QP handle returned by create_qp command */ + u32 qp_handle; +}; + +struct efa_admin_query_qp_resp { + /* Common Admin Queue completion descriptor */ + struct efa_admin_acq_common_desc acq_common_desc; + + /* QP state */ + u32 qp_state; + + /* QKey */ + u32 qkey; + + /* SQ PSN */ + u32 sq_psn; + + /* Indicates that draining is in progress */ + u8 sq_draining; + + /* MBZ */ + u8 reserved1; + + /* MBZ */ + u16 reserved2; +}; + +struct efa_admin_destroy_qp_cmd { + /* Common Admin Queue descriptor */ + struct efa_admin_aq_common_desc aq_common_desc; + + /* QP handle returned by create_qp command */ + u32 qp_handle; +}; + +struct efa_admin_destroy_qp_resp { + /* Common Admin Queue completion descriptor */ + struct efa_admin_acq_common_desc acq_common_desc; +}; + +/* + * Create Address Handle command parameters. Must not be called more than + * once for the same destination + */ +struct efa_admin_create_ah_cmd { + /* Common Admin Queue descriptor */ + struct efa_admin_aq_common_desc aq_common_desc; + + /* Destination address in network byte order */ + u8 dest_addr[16]; + + /* PD number */ + u16 pd; + + u16 reserved; +}; + +struct efa_admin_create_ah_resp { + /* Common Admin Queue completion descriptor */ + struct efa_admin_acq_common_desc acq_common_desc; + + /* Target interface address handle (opaque) */ + u16 ah; + + u16 reserved; +}; + +struct efa_admin_destroy_ah_cmd { + /* Common Admin Queue descriptor */ + struct efa_admin_aq_common_desc aq_common_desc; + + /* Target interface address handle (opaque) */ + u16 ah; + + /* PD number */ + u16 pd; +}; + +struct efa_admin_destroy_ah_resp { + /* Common Admin Queue completion descriptor */ + struct efa_admin_acq_common_desc acq_common_desc; +}; + +/* + * Registration of MemoryRegion, required for QP working with Virtual + * Addresses. In standard verbs semantics, region length is limited to 2GB + * space, but EFA offers larger MR support for large memory space, to ease + * on users working with very large datasets (i.e. full GPU memory mapping). + */ +struct efa_admin_reg_mr_cmd { + /* Common Admin Queue descriptor */ + struct efa_admin_aq_common_desc aq_common_desc; + + /* Protection Domain */ + u16 pd; + + /* MBZ */ + u16 reserved16_w1; + + /* Physical Buffer List, each element is page-aligned. */ + union { + /* + * Inline array of guest-physical page addresses of user + * memory pages (optimization for short region + * registrations) + */ + u64 inline_pbl_array[4]; + + /* points to PBL (direct or indirect, chained if needed) */ + struct efa_admin_ctrl_buff_info pbl; + } pbl; + + /* Memory region length, in bytes. */ + u64 mr_length; + + /* + * flags and page size + * 4:0 : phys_page_size_shift - page size is (1 << + * phys_page_size_shift). Page size is used for + * building the Virtual to Physical address mapping + * 6:5 : reserved - MBZ + * 7 : mem_addr_phy_mode_en - Enable bit for physical + * memory registration (no translation), can be used + * only by privileged clients. If set, PBL must + * contain a single entry. + */ + u8 flags; + + /* + * permissions + * 0 : local_write_enable - Write permissions: value + * of 1 needed for RQ buffers and for RDMA write + * 7:1 : reserved1 - remote access flags, etc + */ + u8 permissions; + + u16 reserved16_w5; + + /* number of pages in PBL (redundant, could be calculated) */ + u32 page_num; + + /* + * IO Virtual Address associated with this MR. If + * mem_addr_phy_mode_en is set, contains the physical address of + * the region. + */ + u64 iova; +}; + +struct efa_admin_reg_mr_resp { + /* Common Admin Queue completion descriptor */ + struct efa_admin_acq_common_desc acq_common_desc; + + /* + * L_Key, to be used in conjunction with local buffer references in + * SQ and RQ WQE, or with virtual RQ/CQ rings + */ + u32 l_key; + + /* + * R_Key, to be used in RDMA messages to refer to remotely accessed + * memory region + */ + u32 r_key; +}; + +struct efa_admin_dereg_mr_cmd { + /* Common Admin Queue descriptor */ + struct efa_admin_aq_common_desc aq_common_desc; + + /* L_Key, memory region's l_key */ + u32 l_key; +}; + +struct efa_admin_dereg_mr_resp { + /* Common Admin Queue completion descriptor */ + struct efa_admin_acq_common_desc acq_common_desc; +}; + +struct efa_admin_create_cq_cmd { + struct efa_admin_aq_common_desc aq_common_desc; + + /* + * 4:0 : reserved5 + * 5 : interrupt_mode_enabled - if set, cq operates + * in interrupt mode (i.e. CQ events and MSI-X are + * generated), otherwise - polling + * 6 : virt - If set, ring base address is virtual + * (IOVA returned by MR registration) + * 7 : reserved6 + */ + u8 cq_caps_1; + + /* + * 4:0 : cq_entry_size_words - size of CQ entry in + * 32-bit words, valid values: 4, 8. + * 7:5 : reserved7 + */ + u8 cq_caps_2; + + /* completion queue depth in # of entries. must be power of 2 */ + u16 cq_depth; + + /* msix vector assigned to this cq */ + u32 msix_vector_idx; + + /* + * CQ ring base address, virtual or physical depending on 'virt' + * flag + */ + struct efa_common_mem_addr cq_ba; + + /* + * Memory registration key for the ring, used only when base + * address is virtual + */ + u32 l_key; + + /* + * number of sub cqs - must be equal to sub_cqs_per_cq of queue + * attributes. + */ + u16 num_sub_cqs; + + /* UAR number */ + u16 uar; +}; + +struct efa_admin_create_cq_resp { + struct efa_admin_acq_common_desc acq_common_desc; + + u16 cq_idx; + + /* actual cq depth in number of entries */ + u16 cq_actual_depth; +}; + +struct efa_admin_destroy_cq_cmd { + struct efa_admin_aq_common_desc aq_common_desc; + + u16 cq_idx; + + u16 reserved1; +}; + +struct efa_admin_destroy_cq_resp { + struct efa_admin_acq_common_desc acq_common_desc; +}; + +/* + * EFA AQ Get Statistics command. Extended statistics are placed in control + * buffer pointed by AQ entry + */ +struct efa_admin_aq_get_stats_cmd { + struct efa_admin_aq_common_desc aq_common_descriptor; + + union { + /* command specific inline data */ + u32 inline_data_w1[3]; + + struct efa_admin_ctrl_buff_info control_buffer; + } u; + + /* stats type as defined in enum efa_admin_get_stats_type */ + u8 type; + + /* stats scope defined in enum efa_admin_get_stats_scope */ + u8 scope; + + u16 scope_modifier; +}; + +struct efa_admin_basic_stats { + u64 tx_bytes; + + u64 tx_pkts; + + u64 rx_bytes; + + u64 rx_pkts; + + u64 rx_drops; +}; + +struct efa_admin_acq_get_stats_resp { + struct efa_admin_acq_common_desc acq_common_desc; + + struct efa_admin_basic_stats basic_stats; +}; + +struct efa_admin_get_set_feature_common_desc { + /* + * 1:0 : select - 0x1 - current value; 0x3 - default + * value + * 7:3 : reserved3 + */ + u8 flags; + + /* as appears in efa_admin_aq_feature_id */ + u8 feature_id; + + /* MBZ */ + u16 reserved16; +}; + +struct efa_admin_feature_device_attr_desc { + /* Bitmap of efa_admin_aq_feature_id */ + u64 supported_features; + + /* Bitmap of supported page sizes in MR registrations */ + u64 page_size_cap; + + u32 fw_version; + + u32 admin_api_version; + + u32 device_version; + + /* Bar used for SQ and RQ doorbells */ + u16 db_bar; + + /* Indicates how many bits are used physical address access */ + u8 phys_addr_width; + + /* Indicates how many bits are used virtual address access */ + u8 virt_addr_width; +}; + +struct efa_admin_feature_queue_attr_desc { + /* The maximum number of queue pairs supported */ + u32 max_qp; + + u32 max_sq_depth; + + /* max send wr used in inline-buf */ + u32 inline_buf_size; + + u32 max_rq_depth; + + /* The maximum number of completion queues supported per VF */ + u32 max_cq; + + u32 max_cq_depth; + + /* Number of sub-CQs to be created for each CQ */ + u16 sub_cqs_per_cq; + + u16 reserved; + + /* + * Maximum number of SGEs (buffs) allowed for a single send work + * queue element (WQE) + */ + u16 max_wr_send_sges; + + /* Maximum number of SGEs allowed for a single recv WQE */ + u16 max_wr_recv_sges; + + /* The maximum number of memory regions supported */ + u32 max_mr; + + /* The maximum number of pages can be registered */ + u32 max_mr_pages; + + /* The maximum number of protection domains supported */ + u32 max_pd; + + /* The maximum number of address handles supported */ + u32 max_ah; + + /* The maximum size of LLQ in bytes */ + u32 max_llq_size; +}; + +struct efa_admin_feature_aenq_desc { + /* bitmask for AENQ groups the device can report */ + u32 supported_groups; + + /* bitmask for AENQ groups to report */ + u32 enabled_groups; +}; + +struct efa_admin_feature_network_attr_desc { + /* Raw address data in network byte order */ + u8 addr[16]; + + u32 mtu; +}; + +/* + * When hint value is 0, hints capabilities are not supported or driver + * should use its own predefined value + */ +struct efa_admin_hw_hints { + /* value in ms */ + u16 mmio_read_timeout; + + /* value in ms */ + u16 driver_watchdog_timeout; + + /* value in ms */ + u16 admin_completion_timeout; + + /* poll interval in ms */ + u16 poll_interval; +}; + +struct efa_admin_get_feature_cmd { + struct efa_admin_aq_common_desc aq_common_descriptor; + + struct efa_admin_ctrl_buff_info control_buffer; + + struct efa_admin_get_set_feature_common_desc feature_common; + + u32 raw[11]; +}; + +struct efa_admin_get_feature_resp { + struct efa_admin_acq_common_desc acq_common_desc; + + union { + u32 raw[14]; + + struct efa_admin_feature_device_attr_desc device_attr; + + struct efa_admin_feature_aenq_desc aenq; + + struct efa_admin_feature_network_attr_desc network_attr; + + struct efa_admin_feature_queue_attr_desc queue_attr; + + struct efa_admin_hw_hints hw_hints; + } u; +}; + +struct efa_admin_set_feature_cmd { + struct efa_admin_aq_common_desc aq_common_descriptor; + + struct efa_admin_ctrl_buff_info control_buffer; + + struct efa_admin_get_set_feature_common_desc feature_common; + + union { + u32 raw[11]; + + /* AENQ configuration */ + struct efa_admin_feature_aenq_desc aenq; + } u; +}; + +struct efa_admin_set_feature_resp { + struct efa_admin_acq_common_desc acq_common_desc; + + union { + u32 raw[14]; + } u; +}; + +struct efa_admin_alloc_pd_cmd { + struct efa_admin_aq_common_desc aq_common_descriptor; +}; + +struct efa_admin_alloc_pd_resp { + struct efa_admin_acq_common_desc acq_common_desc; + + /* PD number */ + u16 pd; + + /* MBZ */ + u16 reserved; +}; + +struct efa_admin_dealloc_pd_cmd { + struct efa_admin_aq_common_desc aq_common_descriptor; + + /* PD number */ + u16 pd; + + /* MBZ */ + u16 reserved; +}; + +struct efa_admin_dealloc_pd_resp { + struct efa_admin_acq_common_desc acq_common_desc; +}; + +struct efa_admin_alloc_uar_cmd { + struct efa_admin_aq_common_desc aq_common_descriptor; +}; + +struct efa_admin_alloc_uar_resp { + struct efa_admin_acq_common_desc acq_common_desc; + + /* UAR number */ + u16 uar; + + /* MBZ */ + u16 reserved; +}; + +struct efa_admin_dealloc_uar_cmd { + struct efa_admin_aq_common_desc aq_common_descriptor; + + /* UAR number */ + u16 uar; + + /* MBZ */ + u16 reserved; +}; + +struct efa_admin_dealloc_uar_resp { + struct efa_admin_acq_common_desc acq_common_desc; +}; + +/* asynchronous event notification groups */ +enum efa_admin_aenq_group { + EFA_ADMIN_FATAL_ERROR = 1, + EFA_ADMIN_WARNING = 2, + EFA_ADMIN_NOTIFICATION = 3, + EFA_ADMIN_KEEP_ALIVE = 4, + EFA_ADMIN_AENQ_GROUPS_NUM = 5, +}; + +enum efa_admin_aenq_notification_syndrom { + EFA_ADMIN_SUSPEND = 0, + EFA_ADMIN_RESUME = 1, + EFA_ADMIN_UPDATE_HINTS = 2, +}; + +struct efa_admin_mmio_req_read_less_resp { + u16 req_id; + + u16 reg_off; + + /* value is valid when poll is cleared */ + u32 reg_val; +}; + +/* create_qp_cmd */ +#define EFA_ADMIN_CREATE_QP_CMD_SQ_VIRT_MASK BIT(0) +#define EFA_ADMIN_CREATE_QP_CMD_RQ_VIRT_SHIFT 1 +#define EFA_ADMIN_CREATE_QP_CMD_RQ_VIRT_MASK BIT(1) + +/* reg_mr_cmd */ +#define EFA_ADMIN_REG_MR_CMD_PHYS_PAGE_SIZE_SHIFT_MASK GENMASK(4, 0) +#define EFA_ADMIN_REG_MR_CMD_MEM_ADDR_PHY_MODE_EN_SHIFT 7 +#define EFA_ADMIN_REG_MR_CMD_MEM_ADDR_PHY_MODE_EN_MASK BIT(7) +#define EFA_ADMIN_REG_MR_CMD_LOCAL_WRITE_ENABLE_MASK BIT(0) + +/* create_cq_cmd */ +#define EFA_ADMIN_CREATE_CQ_CMD_INTERRUPT_MODE_ENABLED_SHIFT 5 +#define EFA_ADMIN_CREATE_CQ_CMD_INTERRUPT_MODE_ENABLED_MASK BIT(5) +#define EFA_ADMIN_CREATE_CQ_CMD_VIRT_SHIFT 6 +#define EFA_ADMIN_CREATE_CQ_CMD_VIRT_MASK BIT(6) +#define EFA_ADMIN_CREATE_CQ_CMD_CQ_ENTRY_SIZE_WORDS_MASK GENMASK(4, 0) + +/* get_set_feature_common_desc */ +#define EFA_ADMIN_GET_SET_FEATURE_COMMON_DESC_SELECT_MASK GENMASK(1, 0) + +#endif /* _EFA_ADMIN_CMDS_H_ */ diff --git a/drivers/infiniband/hw/efa/efa_admin_defs.h b/drivers/infiniband/hw/efa/efa_admin_defs.h new file mode 100644 index 000000000000..c8e0c8b905be --- /dev/null +++ b/drivers/infiniband/hw/efa/efa_admin_defs.h @@ -0,0 +1,136 @@ +/* SPDX-License-Identifier: GPL-2.0 OR BSD-2-Clause */ +/* + * Copyright 2018-2019 Amazon.com, Inc. or its affiliates. All rights reserved. + */ + +#ifndef _EFA_ADMIN_H_ +#define _EFA_ADMIN_H_ + +enum efa_admin_aq_completion_status { + EFA_ADMIN_SUCCESS = 0, + EFA_ADMIN_RESOURCE_ALLOCATION_FAILURE = 1, + EFA_ADMIN_BAD_OPCODE = 2, + EFA_ADMIN_UNSUPPORTED_OPCODE = 3, + EFA_ADMIN_MALFORMED_REQUEST = 4, + /* Additional status is provided in ACQ entry extended_status */ + EFA_ADMIN_ILLEGAL_PARAMETER = 5, + EFA_ADMIN_UNKNOWN_ERROR = 6, + EFA_ADMIN_RESOURCE_BUSY = 7, +}; + +struct efa_admin_aq_common_desc { + /* + * 11:0 : command_id + * 15:12 : reserved12 + */ + u16 command_id; + + /* as appears in efa_admin_aq_opcode */ + u8 opcode; + + /* + * 0 : phase + * 1 : ctrl_data - control buffer address valid + * 2 : ctrl_data_indirect - control buffer address + * points to list of pages with addresses of control + * buffers + * 7:3 : reserved3 + */ + u8 flags; +}; + +/* + * used in efa_admin_aq_entry. Can point directly to control data, or to a + * page list chunk. Used also at the end of indirect mode page list chunks, + * for chaining. + */ +struct efa_admin_ctrl_buff_info { + u32 length; + + struct efa_common_mem_addr address; +}; + +struct efa_admin_aq_entry { + struct efa_admin_aq_common_desc aq_common_descriptor; + + union { + u32 inline_data_w1[3]; + + struct efa_admin_ctrl_buff_info control_buffer; + } u; + + u32 inline_data_w4[12]; +}; + +struct efa_admin_acq_common_desc { + /* + * command identifier to associate it with the aq descriptor + * 11:0 : command_id + * 15:12 : reserved12 + */ + u16 command; + + u8 status; + + /* + * 0 : phase + * 7:1 : reserved1 + */ + u8 flags; + + u16 extended_status; + + /* + * indicates to the driver which AQ entry has been consumed by the + * device and could be reused + */ + u16 sq_head_indx; +}; + +struct efa_admin_acq_entry { + struct efa_admin_acq_common_desc acq_common_descriptor; + + u32 response_specific_data[14]; +}; + +struct efa_admin_aenq_common_desc { + u16 group; + + u16 syndrom; + + /* + * 0 : phase + * 7:1 : reserved - MBZ + */ + u8 flags; + + u8 reserved1[3]; + + u32 timestamp_low; + + u32 timestamp_high; +}; + +struct efa_admin_aenq_entry { + struct efa_admin_aenq_common_desc aenq_common_desc; + + /* command specific inline data */ + u32 inline_data_w4[12]; +}; + +/* aq_common_desc */ +#define EFA_ADMIN_AQ_COMMON_DESC_COMMAND_ID_MASK GENMASK(11, 0) +#define EFA_ADMIN_AQ_COMMON_DESC_PHASE_MASK BIT(0) +#define EFA_ADMIN_AQ_COMMON_DESC_CTRL_DATA_SHIFT 1 +#define EFA_ADMIN_AQ_COMMON_DESC_CTRL_DATA_MASK BIT(1) +#define EFA_ADMIN_AQ_COMMON_DESC_CTRL_DATA_INDIRECT_SHIFT 2 +#define EFA_ADMIN_AQ_COMMON_DESC_CTRL_DATA_INDIRECT_MASK BIT(2) + +/* acq_common_desc */ +#define EFA_ADMIN_ACQ_COMMON_DESC_COMMAND_ID_MASK GENMASK(11, 0) +#define EFA_ADMIN_ACQ_COMMON_DESC_PHASE_MASK BIT(0) + +/* aenq_common_desc */ +#define EFA_ADMIN_AENQ_COMMON_DESC_PHASE_MASK BIT(0) + +#endif /* _EFA_ADMIN_H_ */ diff --git a/drivers/infiniband/hw/efa/efa_com.c b/drivers/infiniband/hw/efa/efa_com.c new file mode 100644 index 000000000000..a5c788741a04 --- /dev/null +++ b/drivers/infiniband/hw/efa/efa_com.c @@ -0,0 +1,1160 @@ +// SPDX-License-Identifier: GPL-2.0 OR BSD-2-Clause +/* + * Copyright 2018-2019 Amazon.com, Inc. or its affiliates. All rights reserved. + */ + +#include "efa_com.h" +#include "efa_regs_defs.h" + +#define ADMIN_CMD_TIMEOUT_US 30000000 /* usecs */ + +#define EFA_REG_READ_TIMEOUT_US 50000 /* usecs */ +#define EFA_MMIO_READ_INVALID 0xffffffff + +#define EFA_POLL_INTERVAL_MS 100 /* msecs */ + +#define EFA_ASYNC_QUEUE_DEPTH 16 +#define EFA_ADMIN_QUEUE_DEPTH 32 + +#define MIN_EFA_VER\ + ((EFA_ADMIN_API_VERSION_MAJOR << EFA_REGS_VERSION_MAJOR_VERSION_SHIFT) | \ + (EFA_ADMIN_API_VERSION_MINOR & EFA_REGS_VERSION_MINOR_VERSION_MASK)) + +#define EFA_CTRL_MAJOR 0 +#define EFA_CTRL_MINOR 0 +#define EFA_CTRL_SUB_MINOR 1 + +#define MIN_EFA_CTRL_VER \ + (((EFA_CTRL_MAJOR) << \ + (EFA_REGS_CONTROLLER_VERSION_MAJOR_VERSION_SHIFT)) | \ + ((EFA_CTRL_MINOR) << \ + (EFA_REGS_CONTROLLER_VERSION_MINOR_VERSION_SHIFT)) | \ + (EFA_CTRL_SUB_MINOR)) + +#define EFA_DMA_ADDR_TO_UINT32_LOW(x) ((u32)((u64)(x))) +#define EFA_DMA_ADDR_TO_UINT32_HIGH(x) ((u32)(((u64)(x)) >> 32)) + +#define EFA_REGS_ADMIN_INTR_MASK 1 + +enum efa_cmd_status { + EFA_CMD_SUBMITTED, + EFA_CMD_COMPLETED, + /* Abort - canceled by the driver */ + EFA_CMD_ABORTED, +}; + +struct efa_comp_ctx { + struct completion wait_event; + struct efa_admin_acq_entry *user_cqe; + u32 comp_size; + enum efa_cmd_status status; + /* status from the device */ + u8 comp_status; + u8 cmd_opcode; + u8 occupied; +}; + +static const char *efa_com_cmd_str(u8 cmd) +{ +#define EFA_CMD_STR_CASE(_cmd) case EFA_ADMIN_##_cmd: return #_cmd + + switch (cmd) { + EFA_CMD_STR_CASE(CREATE_QP); + EFA_CMD_STR_CASE(MODIFY_QP); + EFA_CMD_STR_CASE(QUERY_QP); + EFA_CMD_STR_CASE(DESTROY_QP); + EFA_CMD_STR_CASE(CREATE_AH); + EFA_CMD_STR_CASE(DESTROY_AH); + EFA_CMD_STR_CASE(REG_MR); + EFA_CMD_STR_CASE(DEREG_MR); + EFA_CMD_STR_CASE(CREATE_CQ); + EFA_CMD_STR_CASE(DESTROY_CQ); + EFA_CMD_STR_CASE(GET_FEATURE); + EFA_CMD_STR_CASE(SET_FEATURE); + EFA_CMD_STR_CASE(GET_STATS); + EFA_CMD_STR_CASE(ALLOC_PD); + EFA_CMD_STR_CASE(DEALLOC_PD); + EFA_CMD_STR_CASE(ALLOC_UAR); + EFA_CMD_STR_CASE(DEALLOC_UAR); + default: return "unknown command opcode"; + } +#undef EFA_CMD_STR_CASE +} + +static u32 efa_com_reg_read32(struct efa_com_dev *edev, u16 offset) +{ + struct efa_com_mmio_read *mmio_read = &edev->mmio_read; + struct efa_admin_mmio_req_read_less_resp *read_resp; + unsigned long exp_time; + u32 mmio_read_reg; + u32 err; + + read_resp = mmio_read->read_resp; + + spin_lock(&mmio_read->lock); + mmio_read->seq_num++; + + /* trash DMA req_id to identify when hardware is done */ + read_resp->req_id = mmio_read->seq_num + 0x9aL; + mmio_read_reg = (offset << EFA_REGS_MMIO_REG_READ_REG_OFF_SHIFT) & + EFA_REGS_MMIO_REG_READ_REG_OFF_MASK; + mmio_read_reg |= mmio_read->seq_num & + EFA_REGS_MMIO_REG_READ_REQ_ID_MASK; + + writel(mmio_read_reg, edev->reg_bar + EFA_REGS_MMIO_REG_READ_OFF); + + exp_time = jiffies + usecs_to_jiffies(mmio_read->mmio_read_timeout); + do { + if (READ_ONCE(read_resp->req_id) == mmio_read->seq_num) + break; + udelay(1); + } while (time_is_after_jiffies(exp_time)); + + if (read_resp->req_id != mmio_read->seq_num) { + ibdev_err(edev->efa_dev, + "Reading register timed out. expected: req id[%u] offset[%#x] actual: req id[%u] offset[%#x]\n", + mmio_read->seq_num, offset, read_resp->req_id, + read_resp->reg_off); + err = EFA_MMIO_READ_INVALID; + goto out; + } + + if (read_resp->reg_off != offset) { + ibdev_err(edev->efa_dev, + "Reading register failed: wrong offset provided\n"); + err = EFA_MMIO_READ_INVALID; + goto out; + } + + err = read_resp->reg_val; +out: + spin_unlock(&mmio_read->lock); + return err; +} + +static int efa_com_admin_init_sq(struct efa_com_dev *edev) +{ + struct efa_com_admin_queue *aq = &edev->aq; + struct efa_com_admin_sq *sq = &aq->sq; + u16 size = aq->depth * sizeof(*sq->entries); + u32 addr_high; + u32 addr_low; + u32 aq_caps; + + sq->entries = + dma_alloc_coherent(aq->dmadev, size, &sq->dma_addr, GFP_KERNEL); + if (!sq->entries) + return -ENOMEM; + + spin_lock_init(&sq->lock); + + sq->cc = 0; + sq->pc = 0; + sq->phase = 1; + + sq->db_addr = (u32 __iomem *)(edev->reg_bar + EFA_REGS_AQ_PROD_DB_OFF); + + addr_high = EFA_DMA_ADDR_TO_UINT32_HIGH(sq->dma_addr); + addr_low = EFA_DMA_ADDR_TO_UINT32_LOW(sq->dma_addr); + + writel(addr_low, edev->reg_bar + EFA_REGS_AQ_BASE_LO_OFF); + writel(addr_high, edev->reg_bar + EFA_REGS_AQ_BASE_HI_OFF); + + aq_caps = aq->depth & EFA_REGS_AQ_CAPS_AQ_DEPTH_MASK; + aq_caps |= (sizeof(struct efa_admin_aq_entry) << + EFA_REGS_AQ_CAPS_AQ_ENTRY_SIZE_SHIFT) & + EFA_REGS_AQ_CAPS_AQ_ENTRY_SIZE_MASK; + + writel(aq_caps, edev->reg_bar + EFA_REGS_AQ_CAPS_OFF); + + return 0; +} + +static int efa_com_admin_init_cq(struct efa_com_dev *edev) +{ + struct efa_com_admin_queue *aq = &edev->aq; + struct efa_com_admin_cq *cq = &aq->cq; + u16 size = aq->depth * sizeof(*cq->entries); + u32 addr_high; + u32 addr_low; + u32 acq_caps; + + cq->entries = + dma_alloc_coherent(aq->dmadev, size, &cq->dma_addr, GFP_KERNEL); + if (!cq->entries) + return -ENOMEM; + + spin_lock_init(&cq->lock); + + cq->cc = 0; + cq->phase = 1; + + addr_high = EFA_DMA_ADDR_TO_UINT32_HIGH(cq->dma_addr); + addr_low = EFA_DMA_ADDR_TO_UINT32_LOW(cq->dma_addr); + + writel(addr_low, edev->reg_bar + EFA_REGS_ACQ_BASE_LO_OFF); + writel(addr_high, edev->reg_bar + EFA_REGS_ACQ_BASE_HI_OFF); + + acq_caps = aq->depth & EFA_REGS_ACQ_CAPS_ACQ_DEPTH_MASK; + acq_caps |= (sizeof(struct efa_admin_acq_entry) << + EFA_REGS_ACQ_CAPS_ACQ_ENTRY_SIZE_SHIFT) & + EFA_REGS_ACQ_CAPS_ACQ_ENTRY_SIZE_MASK; + acq_caps |= (aq->msix_vector_idx << + EFA_REGS_ACQ_CAPS_ACQ_MSIX_VECTOR_SHIFT) & + EFA_REGS_ACQ_CAPS_ACQ_MSIX_VECTOR_MASK; + + writel(acq_caps, edev->reg_bar + EFA_REGS_ACQ_CAPS_OFF); + + return 0; +} + +static int efa_com_admin_init_aenq(struct efa_com_dev *edev, + struct efa_aenq_handlers *aenq_handlers) +{ + struct efa_com_aenq *aenq = &edev->aenq; + u32 addr_low, addr_high, aenq_caps; + u16 size; + + if (!aenq_handlers) { + ibdev_err(edev->efa_dev, "aenq handlers pointer is NULL\n"); + return -EINVAL; + } + + size = EFA_ASYNC_QUEUE_DEPTH * sizeof(*aenq->entries); + aenq->entries = dma_alloc_coherent(edev->dmadev, size, &aenq->dma_addr, + GFP_KERNEL); + if (!aenq->entries) + return -ENOMEM; + + aenq->aenq_handlers = aenq_handlers; + aenq->depth = EFA_ASYNC_QUEUE_DEPTH; + aenq->cc = 0; + aenq->phase = 1; + + addr_low = EFA_DMA_ADDR_TO_UINT32_LOW(aenq->dma_addr); + addr_high = EFA_DMA_ADDR_TO_UINT32_HIGH(aenq->dma_addr); + + writel(addr_low, edev->reg_bar + EFA_REGS_AENQ_BASE_LO_OFF); + writel(addr_high, edev->reg_bar + EFA_REGS_AENQ_BASE_HI_OFF); + + aenq_caps = aenq->depth & EFA_REGS_AENQ_CAPS_AENQ_DEPTH_MASK; + aenq_caps |= (sizeof(struct efa_admin_aenq_entry) << + EFA_REGS_AENQ_CAPS_AENQ_ENTRY_SIZE_SHIFT) & + EFA_REGS_AENQ_CAPS_AENQ_ENTRY_SIZE_MASK; + aenq_caps |= (aenq->msix_vector_idx + << EFA_REGS_AENQ_CAPS_AENQ_MSIX_VECTOR_SHIFT) & + EFA_REGS_AENQ_CAPS_AENQ_MSIX_VECTOR_MASK; + writel(aenq_caps, edev->reg_bar + EFA_REGS_AENQ_CAPS_OFF); + + /* + * Init cons_db to mark that all entries in the queue + * are initially available + */ + writel(edev->aenq.cc, edev->reg_bar + EFA_REGS_AENQ_CONS_DB_OFF); + + return 0; +} + +/* ID to be used with efa_com_get_comp_ctx */ +static u16 efa_com_alloc_ctx_id(struct efa_com_admin_queue *aq) +{ + u16 ctx_id; + + spin_lock(&aq->comp_ctx_lock); + ctx_id = aq->comp_ctx_pool[aq->comp_ctx_pool_next]; + aq->comp_ctx_pool_next++; + spin_unlock(&aq->comp_ctx_lock); + + return ctx_id; +} + +static void efa_com_dealloc_ctx_id(struct efa_com_admin_queue *aq, + u16 ctx_id) +{ + spin_lock(&aq->comp_ctx_lock); + aq->comp_ctx_pool_next--; + aq->comp_ctx_pool[aq->comp_ctx_pool_next] = ctx_id; + spin_unlock(&aq->comp_ctx_lock); +} + +static inline void efa_com_put_comp_ctx(struct efa_com_admin_queue *aq, + struct efa_comp_ctx *comp_ctx) +{ + u16 comp_id = comp_ctx->user_cqe->acq_common_descriptor.command & + EFA_ADMIN_ACQ_COMMON_DESC_COMMAND_ID_MASK; + + ibdev_dbg(aq->efa_dev, "Putting completion command_id %d\n", comp_id); + comp_ctx->occupied = 0; + efa_com_dealloc_ctx_id(aq, comp_id); +} + +static struct efa_comp_ctx *efa_com_get_comp_ctx(struct efa_com_admin_queue *aq, + u16 command_id, bool capture) +{ + if (command_id >= aq->depth) { + ibdev_err(aq->efa_dev, + "command id is larger than the queue size. cmd_id: %u queue size %d\n", + command_id, aq->depth); + return NULL; + } + + if (aq->comp_ctx[command_id].occupied && capture) { + ibdev_err(aq->efa_dev, "Completion context is occupied\n"); + return NULL; + } + + if (capture) { + aq->comp_ctx[command_id].occupied = 1; + ibdev_dbg(aq->efa_dev, "Taking completion ctxt command_id %d\n", + command_id); + } + + return &aq->comp_ctx[command_id]; +} + +static struct efa_comp_ctx *__efa_com_submit_admin_cmd(struct efa_com_admin_queue *aq, + struct efa_admin_aq_entry *cmd, + size_t cmd_size_in_bytes, + struct efa_admin_acq_entry *comp, + size_t comp_size_in_bytes) +{ + struct efa_comp_ctx *comp_ctx; + u16 queue_size_mask; + u16 ctx_id; + u16 pi; + + queue_size_mask = aq->depth - 1; + pi = aq->sq.pc & queue_size_mask; + + ctx_id = efa_com_alloc_ctx_id(aq); + + cmd->aq_common_descriptor.flags |= aq->sq.phase & + EFA_ADMIN_AQ_COMMON_DESC_PHASE_MASK; + + cmd->aq_common_descriptor.command_id |= ctx_id & + EFA_ADMIN_AQ_COMMON_DESC_COMMAND_ID_MASK; + + comp_ctx = efa_com_get_comp_ctx(aq, ctx_id, true); + if (!comp_ctx) { + efa_com_dealloc_ctx_id(aq, ctx_id); + return ERR_PTR(-EINVAL); + } + + comp_ctx->status = EFA_CMD_SUBMITTED; + comp_ctx->comp_size = comp_size_in_bytes; + comp_ctx->user_cqe = comp; + comp_ctx->cmd_opcode = cmd->aq_common_descriptor.opcode; + + reinit_completion(&comp_ctx->wait_event); + + memcpy(&aq->sq.entries[pi], cmd, cmd_size_in_bytes); + + aq->sq.pc++; + atomic64_inc(&aq->stats.submitted_cmd); + + if ((aq->sq.pc & queue_size_mask) == 0) + aq->sq.phase = !aq->sq.phase; + + /* barrier not needed in case of writel */ + writel(aq->sq.pc, aq->sq.db_addr); + + return comp_ctx; +} + +static inline int efa_com_init_comp_ctxt(struct efa_com_admin_queue *aq) +{ + size_t pool_size = aq->depth * sizeof(*aq->comp_ctx_pool); + size_t size = aq->depth * sizeof(struct efa_comp_ctx); + struct efa_comp_ctx *comp_ctx; + u16 i; + + aq->comp_ctx = devm_kzalloc(aq->dmadev, size, GFP_KERNEL); + aq->comp_ctx_pool = devm_kzalloc(aq->dmadev, pool_size, GFP_KERNEL); + if (!aq->comp_ctx || !aq->comp_ctx_pool) { + devm_kfree(aq->dmadev, aq->comp_ctx_pool); + devm_kfree(aq->dmadev, aq->comp_ctx); + return -ENOMEM; + } + + for (i = 0; i < aq->depth; i++) { + comp_ctx = efa_com_get_comp_ctx(aq, i, false); + if (comp_ctx) + init_completion(&comp_ctx->wait_event); + + aq->comp_ctx_pool[i] = i; + } + + spin_lock_init(&aq->comp_ctx_lock); + + aq->comp_ctx_pool_next = 0; + + return 0; +} + +static struct efa_comp_ctx *efa_com_submit_admin_cmd(struct efa_com_admin_queue *aq, + struct efa_admin_aq_entry *cmd, + size_t cmd_size_in_bytes, + struct efa_admin_acq_entry *comp, + size_t comp_size_in_bytes) +{ + struct efa_comp_ctx *comp_ctx; + + spin_lock(&aq->sq.lock); + if (!test_bit(EFA_AQ_STATE_RUNNING_BIT, &aq->state)) { + ibdev_err(aq->efa_dev, "Admin queue is closed\n"); + spin_unlock(&aq->sq.lock); + return ERR_PTR(-ENODEV); + } + + comp_ctx = __efa_com_submit_admin_cmd(aq, cmd, cmd_size_in_bytes, comp, + comp_size_in_bytes); + spin_unlock(&aq->sq.lock); + if (IS_ERR(comp_ctx)) + clear_bit(EFA_AQ_STATE_RUNNING_BIT, &aq->state); + + return comp_ctx; +} + +static void efa_com_handle_single_admin_completion(struct efa_com_admin_queue *aq, + struct efa_admin_acq_entry *cqe) +{ + struct efa_comp_ctx *comp_ctx; + u16 cmd_id; + + cmd_id = cqe->acq_common_descriptor.command & + EFA_ADMIN_ACQ_COMMON_DESC_COMMAND_ID_MASK; + + comp_ctx = efa_com_get_comp_ctx(aq, cmd_id, false); + if (!comp_ctx) { + ibdev_err(aq->efa_dev, + "comp_ctx is NULL. Changing the admin queue running state\n"); + clear_bit(EFA_AQ_STATE_RUNNING_BIT, &aq->state); + return; + } + + comp_ctx->status = EFA_CMD_COMPLETED; + comp_ctx->comp_status = cqe->acq_common_descriptor.status; + if (comp_ctx->user_cqe) + memcpy(comp_ctx->user_cqe, cqe, comp_ctx->comp_size); + + if (!test_bit(EFA_AQ_STATE_POLLING_BIT, &aq->state)) + complete(&comp_ctx->wait_event); +} + +static void efa_com_handle_admin_completion(struct efa_com_admin_queue *aq) +{ + struct efa_admin_acq_entry *cqe; + u16 queue_size_mask; + u16 comp_num = 0; + u8 phase; + u16 ci; + + queue_size_mask = aq->depth - 1; + + ci = aq->cq.cc & queue_size_mask; + phase = aq->cq.phase; + + cqe = &aq->cq.entries[ci]; + + /* Go over all the completions */ + while ((READ_ONCE(cqe->acq_common_descriptor.flags) & + EFA_ADMIN_ACQ_COMMON_DESC_PHASE_MASK) == phase) { + /* + * Do not read the rest of the completion entry before the + * phase bit was validated + */ + dma_rmb(); + efa_com_handle_single_admin_completion(aq, cqe); + + ci++; + comp_num++; + if (ci == aq->depth) { + ci = 0; + phase = !phase; + } + + cqe = &aq->cq.entries[ci]; + } + + aq->cq.cc += comp_num; + aq->cq.phase = phase; + aq->sq.cc += comp_num; + atomic64_add(comp_num, &aq->stats.completed_cmd); +} + +static int efa_com_comp_status_to_errno(u8 comp_status) +{ + switch (comp_status) { + case EFA_ADMIN_SUCCESS: + return 0; + case EFA_ADMIN_RESOURCE_ALLOCATION_FAILURE: + return -ENOMEM; + case EFA_ADMIN_UNSUPPORTED_OPCODE: + return -EOPNOTSUPP; + case EFA_ADMIN_BAD_OPCODE: + case EFA_ADMIN_MALFORMED_REQUEST: + case EFA_ADMIN_ILLEGAL_PARAMETER: + case EFA_ADMIN_UNKNOWN_ERROR: + return -EINVAL; + default: + return -EINVAL; + } +} + +static int efa_com_wait_and_process_admin_cq_polling(struct efa_comp_ctx *comp_ctx, + struct efa_com_admin_queue *aq) +{ + unsigned long timeout; + unsigned long flags; + int err; + + timeout = jiffies + usecs_to_jiffies(aq->completion_timeout); + + while (1) { + spin_lock_irqsave(&aq->cq.lock, flags); + efa_com_handle_admin_completion(aq); + spin_unlock_irqrestore(&aq->cq.lock, flags); + + if (comp_ctx->status != EFA_CMD_SUBMITTED) + break; + + if (time_is_before_jiffies(timeout)) { + ibdev_err(aq->efa_dev, + "Wait for completion (polling) timeout\n"); + /* EFA didn't have any completion */ + atomic64_inc(&aq->stats.no_completion); + + clear_bit(EFA_AQ_STATE_RUNNING_BIT, &aq->state); + err = -ETIME; + goto out; + } + + msleep(aq->poll_interval); + } + + if (comp_ctx->status == EFA_CMD_ABORTED) { + ibdev_err(aq->efa_dev, "Command was aborted\n"); + atomic64_inc(&aq->stats.aborted_cmd); + err = -ENODEV; + goto out; + } + + WARN_ONCE(comp_ctx->status != EFA_CMD_COMPLETED, + "Invalid completion status %d\n", comp_ctx->status); + + err = efa_com_comp_status_to_errno(comp_ctx->comp_status); +out: + efa_com_put_comp_ctx(aq, comp_ctx); + return err; +} + +static int efa_com_wait_and_process_admin_cq_interrupts(struct efa_comp_ctx *comp_ctx, + struct efa_com_admin_queue *aq) +{ + unsigned long flags; + int err; + + wait_for_completion_timeout(&comp_ctx->wait_event, + usecs_to_jiffies(aq->completion_timeout)); + + /* + * In case the command wasn't completed find out the root cause. + * There might be 2 kinds of errors + * 1) No completion (timeout reached) + * 2) There is completion but the device didn't get any msi-x interrupt. + */ + if (comp_ctx->status == EFA_CMD_SUBMITTED) { + spin_lock_irqsave(&aq->cq.lock, flags); + efa_com_handle_admin_completion(aq); + spin_unlock_irqrestore(&aq->cq.lock, flags); + + atomic64_inc(&aq->stats.no_completion); + + if (comp_ctx->status == EFA_CMD_COMPLETED) + ibdev_err(aq->efa_dev, + "The device sent a completion but the driver didn't receive any MSI-X interrupt for admin cmd %s(%d) status %d (ctx: 0x%p, sq producer: %d, sq consumer: %d, cq consumer: %d)\n", + efa_com_cmd_str(comp_ctx->cmd_opcode), + comp_ctx->cmd_opcode, comp_ctx->status, + comp_ctx, aq->sq.pc, aq->sq.cc, aq->cq.cc); + else + ibdev_err(aq->efa_dev, + "The device didn't send any completion for admin cmd %s(%d) status %d (ctx 0x%p, sq producer: %d, sq consumer: %d, cq consumer: %d)\n", + efa_com_cmd_str(comp_ctx->cmd_opcode), + comp_ctx->cmd_opcode, comp_ctx->status, + comp_ctx, aq->sq.pc, aq->sq.cc, aq->cq.cc); + + clear_bit(EFA_AQ_STATE_RUNNING_BIT, &aq->state); + err = -ETIME; + goto out; + } + + err = efa_com_comp_status_to_errno(comp_ctx->comp_status); +out: + efa_com_put_comp_ctx(aq, comp_ctx); + return err; +} + +/* + * There are two types to wait for completion. + * Polling mode - wait until the completion is available. + * Async mode - wait on wait queue until the completion is ready + * (or the timeout expired). + * It is expected that the IRQ called efa_com_handle_admin_completion + * to mark the completions. + */ +static int efa_com_wait_and_process_admin_cq(struct efa_comp_ctx *comp_ctx, + struct efa_com_admin_queue *aq) +{ + if (test_bit(EFA_AQ_STATE_POLLING_BIT, &aq->state)) + return efa_com_wait_and_process_admin_cq_polling(comp_ctx, aq); + + return efa_com_wait_and_process_admin_cq_interrupts(comp_ctx, aq); +} + +/** + * efa_com_cmd_exec - Execute admin command + * @aq: admin queue. + * @cmd: the admin command to execute. + * @cmd_size: the command size. + * @comp: command completion return entry. + * @comp_size: command completion size. + * Submit an admin command and then wait until the device will return a + * completion. + * The completion will be copied into comp. + * + * @return - 0 on success, negative value on failure. + */ +int efa_com_cmd_exec(struct efa_com_admin_queue *aq, + struct efa_admin_aq_entry *cmd, + size_t cmd_size, + struct efa_admin_acq_entry *comp, + size_t comp_size) +{ + struct efa_comp_ctx *comp_ctx; + int err; + + might_sleep(); + + /* In case of queue FULL */ + down(&aq->avail_cmds); + + ibdev_dbg(aq->efa_dev, "%s (opcode %d)\n", + efa_com_cmd_str(cmd->aq_common_descriptor.opcode), + cmd->aq_common_descriptor.opcode); + comp_ctx = efa_com_submit_admin_cmd(aq, cmd, cmd_size, comp, comp_size); + if (IS_ERR(comp_ctx)) { + ibdev_err(aq->efa_dev, + "Failed to submit command %s (opcode %u) err %ld\n", + efa_com_cmd_str(cmd->aq_common_descriptor.opcode), + cmd->aq_common_descriptor.opcode, PTR_ERR(comp_ctx)); + + up(&aq->avail_cmds); + return PTR_ERR(comp_ctx); + } + + err = efa_com_wait_and_process_admin_cq(comp_ctx, aq); + if (err) + ibdev_err(aq->efa_dev, + "Failed to process command %s (opcode %u) comp_status %d err %d\n", + efa_com_cmd_str(cmd->aq_common_descriptor.opcode), + cmd->aq_common_descriptor.opcode, + comp_ctx->comp_status, err); + + up(&aq->avail_cmds); + + return err; +} + +/** + * efa_com_abort_admin_commands - Abort all the outstanding admin commands. + * @edev: EFA communication layer struct + * + * This method aborts all the outstanding admin commands. + * The caller should then call efa_com_wait_for_abort_completion to make sure + * all the commands were completed. + */ +static void efa_com_abort_admin_commands(struct efa_com_dev *edev) +{ + struct efa_com_admin_queue *aq = &edev->aq; + struct efa_comp_ctx *comp_ctx; + unsigned long flags; + u16 i; + + spin_lock(&aq->sq.lock); + spin_lock_irqsave(&aq->cq.lock, flags); + for (i = 0; i < aq->depth; i++) { + comp_ctx = efa_com_get_comp_ctx(aq, i, false); + if (!comp_ctx) + break; + + comp_ctx->status = EFA_CMD_ABORTED; + + complete(&comp_ctx->wait_event); + } + spin_unlock_irqrestore(&aq->cq.lock, flags); + spin_unlock(&aq->sq.lock); +} + +/** + * efa_com_wait_for_abort_completion - Wait for admin commands abort. + * @edev: EFA communication layer struct + * + * This method wait until all the outstanding admin commands will be completed. + */ +static void efa_com_wait_for_abort_completion(struct efa_com_dev *edev) +{ + struct efa_com_admin_queue *aq = &edev->aq; + int i; + + /* all mine */ + for (i = 0; i < aq->depth; i++) + down(&aq->avail_cmds); + + /* let it go */ + for (i = 0; i < aq->depth; i++) + up(&aq->avail_cmds); +} + +static void efa_com_admin_flush(struct efa_com_dev *edev) +{ + struct efa_com_admin_queue *aq = &edev->aq; + + clear_bit(EFA_AQ_STATE_RUNNING_BIT, &aq->state); + + efa_com_abort_admin_commands(edev); + efa_com_wait_for_abort_completion(edev); +} + +/** + * efa_com_admin_destroy - Destroy the admin and the async events queues. + * @edev: EFA communication layer struct + */ +void efa_com_admin_destroy(struct efa_com_dev *edev) +{ + struct efa_com_admin_queue *aq = &edev->aq; + struct efa_com_aenq *aenq = &edev->aenq; + struct efa_com_admin_cq *cq = &aq->cq; + struct efa_com_admin_sq *sq = &aq->sq; + u16 size; + + efa_com_admin_flush(edev); + + devm_kfree(edev->dmadev, aq->comp_ctx_pool); + devm_kfree(edev->dmadev, aq->comp_ctx); + + size = aq->depth * sizeof(*sq->entries); + dma_free_coherent(edev->dmadev, size, sq->entries, sq->dma_addr); + + size = aq->depth * sizeof(*cq->entries); + dma_free_coherent(edev->dmadev, size, cq->entries, cq->dma_addr); + + size = aenq->depth * sizeof(*aenq->entries); + dma_free_coherent(edev->dmadev, size, aenq->entries, aenq->dma_addr); +} + +/** + * efa_com_set_admin_polling_mode - Set the admin completion queue polling mode + * @edev: EFA communication layer struct + * @polling: Enable/Disable polling mode + * + * Set the admin completion mode. + */ +void efa_com_set_admin_polling_mode(struct efa_com_dev *edev, bool polling) +{ + u32 mask_value = 0; + + if (polling) + mask_value = EFA_REGS_ADMIN_INTR_MASK; + + writel(mask_value, edev->reg_bar + EFA_REGS_INTR_MASK_OFF); + if (polling) + set_bit(EFA_AQ_STATE_POLLING_BIT, &edev->aq.state); + else + clear_bit(EFA_AQ_STATE_POLLING_BIT, &edev->aq.state); +} + +static void efa_com_stats_init(struct efa_com_dev *edev) +{ + atomic64_t *s = (atomic64_t *)&edev->aq.stats; + int i; + + for (i = 0; i < sizeof(edev->aq.stats) / sizeof(*s); i++, s++) + atomic64_set(s, 0); +} + +/** + * efa_com_admin_init - Init the admin and the async queues + * @edev: EFA communication layer struct + * @aenq_handlers: Those handlers to be called upon event. + * + * Initialize the admin submission and completion queues. + * Initialize the asynchronous events notification queues. + * + * @return - 0 on success, negative value on failure. + */ +int efa_com_admin_init(struct efa_com_dev *edev, + struct efa_aenq_handlers *aenq_handlers) +{ + struct efa_com_admin_queue *aq = &edev->aq; + u32 timeout; + u32 dev_sts; + u32 cap; + int err; + + dev_sts = efa_com_reg_read32(edev, EFA_REGS_DEV_STS_OFF); + if (!(dev_sts & EFA_REGS_DEV_STS_READY_MASK)) { + ibdev_err(edev->efa_dev, + "Device isn't ready, abort com init %#x\n", dev_sts); + return -ENODEV; + } + + aq->depth = EFA_ADMIN_QUEUE_DEPTH; + + aq->dmadev = edev->dmadev; + aq->efa_dev = edev->efa_dev; + set_bit(EFA_AQ_STATE_POLLING_BIT, &aq->state); + + sema_init(&aq->avail_cmds, aq->depth); + + efa_com_stats_init(edev); + + err = efa_com_init_comp_ctxt(aq); + if (err) + return err; + + err = efa_com_admin_init_sq(edev); + if (err) + goto err_destroy_comp_ctxt; + + err = efa_com_admin_init_cq(edev); + if (err) + goto err_destroy_sq; + + efa_com_set_admin_polling_mode(edev, false); + + err = efa_com_admin_init_aenq(edev, aenq_handlers); + if (err) + goto err_destroy_cq; + + cap = efa_com_reg_read32(edev, EFA_REGS_CAPS_OFF); + timeout = (cap & EFA_REGS_CAPS_ADMIN_CMD_TO_MASK) >> + EFA_REGS_CAPS_ADMIN_CMD_TO_SHIFT; + if (timeout) + /* the resolution of timeout reg is 100ms */ + aq->completion_timeout = timeout * 100000; + else + aq->completion_timeout = ADMIN_CMD_TIMEOUT_US; + + aq->poll_interval = EFA_POLL_INTERVAL_MS; + + set_bit(EFA_AQ_STATE_RUNNING_BIT, &aq->state); + + return 0; + +err_destroy_cq: + dma_free_coherent(edev->dmadev, aq->depth * sizeof(*aq->cq.entries), + aq->cq.entries, aq->cq.dma_addr); +err_destroy_sq: + dma_free_coherent(edev->dmadev, aq->depth * sizeof(*aq->sq.entries), + aq->sq.entries, aq->sq.dma_addr); +err_destroy_comp_ctxt: + devm_kfree(edev->dmadev, aq->comp_ctx); + + return err; +} + +/** + * efa_com_admin_q_comp_intr_handler - admin queue interrupt handler + * @edev: EFA communication layer struct + * + * This method goes over the admin completion queue and wakes up + * all the pending threads that wait on the commands wait event. + * + * @note: Should be called after MSI-X interrupt. + */ +void efa_com_admin_q_comp_intr_handler(struct efa_com_dev *edev) +{ + unsigned long flags; + + spin_lock_irqsave(&edev->aq.cq.lock, flags); + efa_com_handle_admin_completion(&edev->aq); + spin_unlock_irqrestore(&edev->aq.cq.lock, flags); +} + +/* + * efa_handle_specific_aenq_event: + * return the handler that is relevant to the specific event group + */ +static efa_aenq_handler efa_com_get_specific_aenq_cb(struct efa_com_dev *edev, + u16 group) +{ + struct efa_aenq_handlers *aenq_handlers = edev->aenq.aenq_handlers; + + if (group < EFA_MAX_HANDLERS && aenq_handlers->handlers[group]) + return aenq_handlers->handlers[group]; + + return aenq_handlers->unimplemented_handler; +} + +/** + * efa_com_aenq_intr_handler - AENQ interrupt handler + * @edev: EFA communication layer struct + * @data: Data of interrupt handler. + * + * Go over the async event notification queue and call the proper aenq handler. + */ +void efa_com_aenq_intr_handler(struct efa_com_dev *edev, void *data) +{ + struct efa_admin_aenq_common_desc *aenq_common; + struct efa_com_aenq *aenq = &edev->aenq; + struct efa_admin_aenq_entry *aenq_e; + efa_aenq_handler handler_cb; + u32 processed = 0; + u8 phase; + u32 ci; + + ci = aenq->cc & (aenq->depth - 1); + phase = aenq->phase; + aenq_e = &aenq->entries[ci]; /* Get first entry */ + aenq_common = &aenq_e->aenq_common_desc; + + /* Go over all the events */ + while ((READ_ONCE(aenq_common->flags) & + EFA_ADMIN_AENQ_COMMON_DESC_PHASE_MASK) == phase) { + /* + * Do not read the rest of the completion entry before the + * phase bit was validated + */ + dma_rmb(); + + /* Handle specific event*/ + handler_cb = efa_com_get_specific_aenq_cb(edev, + aenq_common->group); + handler_cb(data, aenq_e); /* call the actual event handler*/ + + /* Get next event entry */ + ci++; + processed++; + + if (ci == aenq->depth) { + ci = 0; + phase = !phase; + } + aenq_e = &aenq->entries[ci]; + aenq_common = &aenq_e->aenq_common_desc; + } + + aenq->cc += processed; + aenq->phase = phase; + + /* Don't update aenq doorbell if there weren't any processed events */ + if (!processed) + return; + + /* barrier not needed in case of writel */ + writel(aenq->cc, edev->reg_bar + EFA_REGS_AENQ_CONS_DB_OFF); +} + +static void efa_com_mmio_reg_read_resp_addr_init(struct efa_com_dev *edev) +{ + struct efa_com_mmio_read *mmio_read = &edev->mmio_read; + u32 addr_high; + u32 addr_low; + + /* dma_addr_bits is unknown at this point */ + addr_high = (mmio_read->read_resp_dma_addr >> 32) & GENMASK(31, 0); + addr_low = mmio_read->read_resp_dma_addr & GENMASK(31, 0); + + writel(addr_high, edev->reg_bar + EFA_REGS_MMIO_RESP_HI_OFF); + writel(addr_low, edev->reg_bar + EFA_REGS_MMIO_RESP_LO_OFF); +} + +int efa_com_mmio_reg_read_init(struct efa_com_dev *edev) +{ + struct efa_com_mmio_read *mmio_read = &edev->mmio_read; + + spin_lock_init(&mmio_read->lock); + mmio_read->read_resp = + dma_alloc_coherent(edev->dmadev, sizeof(*mmio_read->read_resp), + &mmio_read->read_resp_dma_addr, GFP_KERNEL); + if (!mmio_read->read_resp) + return -ENOMEM; + + efa_com_mmio_reg_read_resp_addr_init(edev); + + mmio_read->read_resp->req_id = 0; + mmio_read->seq_num = 0; + mmio_read->mmio_read_timeout = EFA_REG_READ_TIMEOUT_US; + + return 0; +} + +void efa_com_mmio_reg_read_destroy(struct efa_com_dev *edev) +{ + struct efa_com_mmio_read *mmio_read = &edev->mmio_read; + + dma_free_coherent(edev->dmadev, sizeof(*mmio_read->read_resp), + mmio_read->read_resp, mmio_read->read_resp_dma_addr); +} + +int efa_com_validate_version(struct efa_com_dev *edev) +{ + u32 ctrl_ver_masked; + u32 ctrl_ver; + u32 ver; + + /* + * Make sure the EFA version and the controller version are at least + * as the driver expects + */ + ver = efa_com_reg_read32(edev, EFA_REGS_VERSION_OFF); + ctrl_ver = efa_com_reg_read32(edev, + EFA_REGS_CONTROLLER_VERSION_OFF); + + ibdev_dbg(edev->efa_dev, "efa device version: %d.%d\n", + (ver & EFA_REGS_VERSION_MAJOR_VERSION_MASK) >> + EFA_REGS_VERSION_MAJOR_VERSION_SHIFT, + ver & EFA_REGS_VERSION_MINOR_VERSION_MASK); + + if (ver < MIN_EFA_VER) { + ibdev_err(edev->efa_dev, + "EFA version is lower than the minimal version the driver supports\n"); + return -EOPNOTSUPP; + } + + ibdev_dbg(edev->efa_dev, + "efa controller version: %d.%d.%d implementation version %d\n", + (ctrl_ver & EFA_REGS_CONTROLLER_VERSION_MAJOR_VERSION_MASK) >> + EFA_REGS_CONTROLLER_VERSION_MAJOR_VERSION_SHIFT, + (ctrl_ver & EFA_REGS_CONTROLLER_VERSION_MINOR_VERSION_MASK) >> + EFA_REGS_CONTROLLER_VERSION_MINOR_VERSION_SHIFT, + (ctrl_ver & EFA_REGS_CONTROLLER_VERSION_SUBMINOR_VERSION_MASK), + (ctrl_ver & EFA_REGS_CONTROLLER_VERSION_IMPL_ID_MASK) >> + EFA_REGS_CONTROLLER_VERSION_IMPL_ID_SHIFT); + + ctrl_ver_masked = + (ctrl_ver & EFA_REGS_CONTROLLER_VERSION_MAJOR_VERSION_MASK) | + (ctrl_ver & EFA_REGS_CONTROLLER_VERSION_MINOR_VERSION_MASK) | + (ctrl_ver & EFA_REGS_CONTROLLER_VERSION_SUBMINOR_VERSION_MASK); + + /* Validate the ctrl version without the implementation ID */ + if (ctrl_ver_masked < MIN_EFA_CTRL_VER) { + ibdev_err(edev->efa_dev, + "EFA ctrl version is lower than the minimal ctrl version the driver supports\n"); + return -EOPNOTSUPP; + } + + return 0; +} + +/** + * efa_com_get_dma_width - Retrieve physical dma address width the device + * supports. + * @edev: EFA communication layer struct + * + * Retrieve the maximum physical address bits the device can handle. + * + * @return: > 0 on Success and negative value otherwise. + */ +int efa_com_get_dma_width(struct efa_com_dev *edev) +{ + u32 caps = efa_com_reg_read32(edev, EFA_REGS_CAPS_OFF); + int width; + + width = (caps & EFA_REGS_CAPS_DMA_ADDR_WIDTH_MASK) >> + EFA_REGS_CAPS_DMA_ADDR_WIDTH_SHIFT; + + ibdev_dbg(edev->efa_dev, "DMA width: %d\n", width); + + if (width < 32 || width > 64) { + ibdev_err(edev->efa_dev, "DMA width illegal value: %d\n", width); + return -EINVAL; + } + + edev->dma_addr_bits = width; + + return width; +} + +static int wait_for_reset_state(struct efa_com_dev *edev, u32 timeout, + u16 exp_state) +{ + u32 val, i; + + for (i = 0; i < timeout; i++) { + val = efa_com_reg_read32(edev, EFA_REGS_DEV_STS_OFF); + + if ((val & EFA_REGS_DEV_STS_RESET_IN_PROGRESS_MASK) == + exp_state) + return 0; + + ibdev_dbg(edev->efa_dev, "Reset indication val %d\n", val); + msleep(EFA_POLL_INTERVAL_MS); + } + + return -ETIME; +} + +/** + * efa_com_dev_reset - Perform device FLR to the device. + * @edev: EFA communication layer struct + * @reset_reason: Specify what is the trigger for the reset in case of an error. + * + * @return - 0 on success, negative value on failure. + */ +int efa_com_dev_reset(struct efa_com_dev *edev, + enum efa_regs_reset_reason_types reset_reason) +{ + u32 stat, timeout, cap, reset_val; + int err; + + stat = efa_com_reg_read32(edev, EFA_REGS_DEV_STS_OFF); + cap = efa_com_reg_read32(edev, EFA_REGS_CAPS_OFF); + + if (!(stat & EFA_REGS_DEV_STS_READY_MASK)) { + ibdev_err(edev->efa_dev, + "Device isn't ready, can't reset device\n"); + return -EINVAL; + } + + timeout = (cap & EFA_REGS_CAPS_RESET_TIMEOUT_MASK) >> + EFA_REGS_CAPS_RESET_TIMEOUT_SHIFT; + if (!timeout) { + ibdev_err(edev->efa_dev, "Invalid timeout value\n"); + return -EINVAL; + } + + /* start reset */ + reset_val = EFA_REGS_DEV_CTL_DEV_RESET_MASK; + reset_val |= (reset_reason << EFA_REGS_DEV_CTL_RESET_REASON_SHIFT) & + EFA_REGS_DEV_CTL_RESET_REASON_MASK; + writel(reset_val, edev->reg_bar + EFA_REGS_DEV_CTL_OFF); + + /* reset clears the mmio readless address, restore it */ + efa_com_mmio_reg_read_resp_addr_init(edev); + + err = wait_for_reset_state(edev, timeout, + EFA_REGS_DEV_STS_RESET_IN_PROGRESS_MASK); + if (err) { + ibdev_err(edev->efa_dev, "Reset indication didn't turn on\n"); + return err; + } + + /* reset done */ + writel(0, edev->reg_bar + EFA_REGS_DEV_CTL_OFF); + err = wait_for_reset_state(edev, timeout, 0); + if (err) { + ibdev_err(edev->efa_dev, "Reset indication didn't turn off\n"); + return err; + } + + timeout = (cap & EFA_REGS_CAPS_ADMIN_CMD_TO_MASK) >> + EFA_REGS_CAPS_ADMIN_CMD_TO_SHIFT; + if (timeout) + /* the resolution of timeout reg is 100ms */ + edev->aq.completion_timeout = timeout * 100000; + else + edev->aq.completion_timeout = ADMIN_CMD_TIMEOUT_US; + + return 0; +} diff --git a/drivers/infiniband/hw/efa/efa_com.h b/drivers/infiniband/hw/efa/efa_com.h new file mode 100644 index 000000000000..84d96724a74b --- /dev/null +++ b/drivers/infiniband/hw/efa/efa_com.h @@ -0,0 +1,144 @@ +/* SPDX-License-Identifier: GPL-2.0 OR BSD-2-Clause */ +/* + * Copyright 2018-2019 Amazon.com, Inc. or its affiliates. All rights reserved. + */ + +#ifndef _EFA_COM_H_ +#define _EFA_COM_H_ + +#include <linux/delay.h> +#include <linux/device.h> +#include <linux/dma-mapping.h> +#include <linux/semaphore.h> +#include <linux/sched.h> + +#include <rdma/ib_verbs.h> + +#include "efa_common_defs.h" +#include "efa_admin_defs.h" +#include "efa_admin_cmds_defs.h" +#include "efa_regs_defs.h" + +#define EFA_MAX_HANDLERS 256 + +struct efa_com_admin_cq { + struct efa_admin_acq_entry *entries; + dma_addr_t dma_addr; + spinlock_t lock; /* Protects ACQ */ + + u16 cc; /* consumer counter */ + u8 phase; +}; + +struct efa_com_admin_sq { + struct efa_admin_aq_entry *entries; + dma_addr_t dma_addr; + spinlock_t lock; /* Protects ASQ */ + + u32 __iomem *db_addr; + + u16 cc; /* consumer counter */ + u16 pc; /* producer counter */ + u8 phase; + +}; + +/* Don't use anything other than atomic64 */ +struct efa_com_stats_admin { + atomic64_t aborted_cmd; + atomic64_t submitted_cmd; + atomic64_t completed_cmd; + atomic64_t no_completion; +}; + +enum { + EFA_AQ_STATE_RUNNING_BIT = 0, + EFA_AQ_STATE_POLLING_BIT = 1, +}; + +struct efa_com_admin_queue { + void *dmadev; + void *efa_dev; + struct efa_comp_ctx *comp_ctx; + u32 completion_timeout; /* usecs */ + u16 poll_interval; /* msecs */ + u16 depth; + struct efa_com_admin_cq cq; + struct efa_com_admin_sq sq; + u16 msix_vector_idx; + + unsigned long state; + + /* Count the number of available admin commands */ + struct semaphore avail_cmds; + + struct efa_com_stats_admin stats; + + spinlock_t comp_ctx_lock; /* Protects completion context pool */ + u32 *comp_ctx_pool; + u16 comp_ctx_pool_next; +}; + +struct efa_aenq_handlers; + +struct efa_com_aenq { + struct efa_admin_aenq_entry *entries; + struct efa_aenq_handlers *aenq_handlers; + dma_addr_t dma_addr; + u32 cc; /* consumer counter */ + u16 msix_vector_idx; + u16 depth; + u8 phase; +}; + +struct efa_com_mmio_read { + struct efa_admin_mmio_req_read_less_resp *read_resp; + dma_addr_t read_resp_dma_addr; + u16 seq_num; + u16 mmio_read_timeout; /* usecs */ + /* serializes mmio reads */ + spinlock_t lock; +}; + +struct efa_com_dev { + struct efa_com_admin_queue aq; + struct efa_com_aenq aenq; + u8 __iomem *reg_bar; + void *dmadev; + void *efa_dev; + u32 supported_features; + u32 dma_addr_bits; + + struct efa_com_mmio_read mmio_read; +}; + +typedef void (*efa_aenq_handler)(void *data, + struct efa_admin_aenq_entry *aenq_e); + +/* Holds aenq handlers. Indexed by AENQ event group */ +struct efa_aenq_handlers { + efa_aenq_handler handlers[EFA_MAX_HANDLERS]; + efa_aenq_handler unimplemented_handler; +}; + +int efa_com_admin_init(struct efa_com_dev *edev, + struct efa_aenq_handlers *aenq_handlers); +void efa_com_admin_destroy(struct efa_com_dev *edev); +int efa_com_dev_reset(struct efa_com_dev *edev, + enum efa_regs_reset_reason_types reset_reason); +void efa_com_set_admin_polling_mode(struct efa_com_dev *edev, bool polling); +void efa_com_admin_q_comp_intr_handler(struct efa_com_dev *edev); +int efa_com_mmio_reg_read_init(struct efa_com_dev *edev); +void efa_com_mmio_reg_read_destroy(struct efa_com_dev *edev); + +int efa_com_validate_version(struct efa_com_dev *edev); +int efa_com_get_dma_width(struct efa_com_dev *edev); + +int efa_com_cmd_exec(struct efa_com_admin_queue *aq, + struct efa_admin_aq_entry *cmd, + size_t cmd_size, + struct efa_admin_acq_entry *comp, + size_t comp_size); +void efa_com_aenq_intr_handler(struct efa_com_dev *edev, void *data); + +#endif /* _EFA_COM_H_ */ diff --git a/drivers/infiniband/hw/efa/efa_com_cmd.c b/drivers/infiniband/hw/efa/efa_com_cmd.c new file mode 100644 index 000000000000..14227725521c --- /dev/null +++ b/drivers/infiniband/hw/efa/efa_com_cmd.c @@ -0,0 +1,692 @@ +// SPDX-License-Identifier: GPL-2.0 OR BSD-2-Clause +/* + * Copyright 2018-2019 Amazon.com, Inc. or its affiliates. All rights reserved. + */ + +#include "efa.h" +#include "efa_com.h" +#include "efa_com_cmd.h" + +void efa_com_set_dma_addr(dma_addr_t addr, u32 *addr_high, u32 *addr_low) +{ + *addr_low = lower_32_bits(addr); + *addr_high = upper_32_bits(addr); +} + +int efa_com_create_qp(struct efa_com_dev *edev, + struct efa_com_create_qp_params *params, + struct efa_com_create_qp_result *res) +{ + struct efa_admin_create_qp_cmd create_qp_cmd = {}; + struct efa_admin_create_qp_resp cmd_completion; + struct efa_com_admin_queue *aq = &edev->aq; + int err; + + create_qp_cmd.aq_common_desc.opcode = EFA_ADMIN_CREATE_QP; + + create_qp_cmd.pd = params->pd; + create_qp_cmd.qp_type = params->qp_type; + create_qp_cmd.rq_base_addr = params->rq_base_addr; + create_qp_cmd.send_cq_idx = params->send_cq_idx; + create_qp_cmd.recv_cq_idx = params->recv_cq_idx; + create_qp_cmd.qp_alloc_size.send_queue_ring_size = + params->sq_ring_size_in_bytes; + create_qp_cmd.qp_alloc_size.send_queue_depth = + params->sq_depth; + create_qp_cmd.qp_alloc_size.recv_queue_ring_size = + params->rq_ring_size_in_bytes; + create_qp_cmd.qp_alloc_size.recv_queue_depth = + params->rq_depth; + create_qp_cmd.uar = params->uarn; + + err = efa_com_cmd_exec(aq, + (struct efa_admin_aq_entry *)&create_qp_cmd, + sizeof(create_qp_cmd), + (struct efa_admin_acq_entry *)&cmd_completion, + sizeof(cmd_completion)); + if (err) { + ibdev_err(edev->efa_dev, "Failed to create qp [%d]\n", err); + return err; + } + + res->qp_handle = cmd_completion.qp_handle; + res->qp_num = cmd_completion.qp_num; + res->sq_db_offset = cmd_completion.sq_db_offset; + res->rq_db_offset = cmd_completion.rq_db_offset; + res->llq_descriptors_offset = cmd_completion.llq_descriptors_offset; + res->send_sub_cq_idx = cmd_completion.send_sub_cq_idx; + res->recv_sub_cq_idx = cmd_completion.recv_sub_cq_idx; + + return err; +} + +int efa_com_modify_qp(struct efa_com_dev *edev, + struct efa_com_modify_qp_params *params) +{ + struct efa_com_admin_queue *aq = &edev->aq; + struct efa_admin_modify_qp_cmd cmd = {}; + struct efa_admin_modify_qp_resp resp; + int err; + + cmd.aq_common_desc.opcode = EFA_ADMIN_MODIFY_QP; + cmd.modify_mask = params->modify_mask; + cmd.qp_handle = params->qp_handle; + cmd.qp_state = params->qp_state; + cmd.cur_qp_state = params->cur_qp_state; + cmd.qkey = params->qkey; + cmd.sq_psn = params->sq_psn; + cmd.sq_drained_async_notify = params->sq_drained_async_notify; + + err = efa_com_cmd_exec(aq, + (struct efa_admin_aq_entry *)&cmd, + sizeof(cmd), + (struct efa_admin_acq_entry *)&resp, + sizeof(resp)); + if (err) { + ibdev_err(edev->efa_dev, + "Failed to modify qp-%u modify_mask[%#x] [%d]\n", + cmd.qp_handle, cmd.modify_mask, err); + return err; + } + + return 0; +} + +int efa_com_query_qp(struct efa_com_dev *edev, + struct efa_com_query_qp_params *params, + struct efa_com_query_qp_result *result) +{ + struct efa_com_admin_queue *aq = &edev->aq; + struct efa_admin_query_qp_cmd cmd = {}; + struct efa_admin_query_qp_resp resp; + int err; + + cmd.aq_common_desc.opcode = EFA_ADMIN_QUERY_QP; + cmd.qp_handle = params->qp_handle; + + err = efa_com_cmd_exec(aq, + (struct efa_admin_aq_entry *)&cmd, + sizeof(cmd), + (struct efa_admin_acq_entry *)&resp, + sizeof(resp)); + if (err) { + ibdev_err(edev->efa_dev, "Failed to query qp-%u [%d]\n", + cmd.qp_handle, err); + return err; + } + + result->qp_state = resp.qp_state; + result->qkey = resp.qkey; + result->sq_draining = resp.sq_draining; + result->sq_psn = resp.sq_psn; + + return 0; +} + +int efa_com_destroy_qp(struct efa_com_dev *edev, + struct efa_com_destroy_qp_params *params) +{ + struct efa_admin_destroy_qp_resp cmd_completion; + struct efa_admin_destroy_qp_cmd qp_cmd = {}; + struct efa_com_admin_queue *aq = &edev->aq; + int err; + + qp_cmd.aq_common_desc.opcode = EFA_ADMIN_DESTROY_QP; + qp_cmd.qp_handle = params->qp_handle; + + err = efa_com_cmd_exec(aq, + (struct efa_admin_aq_entry *)&qp_cmd, + sizeof(qp_cmd), + (struct efa_admin_acq_entry *)&cmd_completion, + sizeof(cmd_completion)); + if (err) + ibdev_err(edev->efa_dev, "Failed to destroy qp-%u [%d]\n", + qp_cmd.qp_handle, err); + + return 0; +} + +int efa_com_create_cq(struct efa_com_dev *edev, + struct efa_com_create_cq_params *params, + struct efa_com_create_cq_result *result) +{ + struct efa_admin_create_cq_resp cmd_completion; + struct efa_admin_create_cq_cmd create_cmd = {}; + struct efa_com_admin_queue *aq = &edev->aq; + int err; + + create_cmd.aq_common_desc.opcode = EFA_ADMIN_CREATE_CQ; + create_cmd.cq_caps_2 = (params->entry_size_in_bytes / 4) & + EFA_ADMIN_CREATE_CQ_CMD_CQ_ENTRY_SIZE_WORDS_MASK; + create_cmd.cq_depth = params->cq_depth; + create_cmd.num_sub_cqs = params->num_sub_cqs; + create_cmd.uar = params->uarn; + + efa_com_set_dma_addr(params->dma_addr, + &create_cmd.cq_ba.mem_addr_high, + &create_cmd.cq_ba.mem_addr_low); + + err = efa_com_cmd_exec(aq, + (struct efa_admin_aq_entry *)&create_cmd, + sizeof(create_cmd), + (struct efa_admin_acq_entry *)&cmd_completion, + sizeof(cmd_completion)); + if (err) { + ibdev_err(edev->efa_dev, "Failed to create cq[%d]\n", err); + return err; + } + + result->cq_idx = cmd_completion.cq_idx; + result->actual_depth = params->cq_depth; + + return err; +} + +int efa_com_destroy_cq(struct efa_com_dev *edev, + struct efa_com_destroy_cq_params *params) +{ + struct efa_admin_destroy_cq_cmd destroy_cmd = {}; + struct efa_admin_destroy_cq_resp destroy_resp; + struct efa_com_admin_queue *aq = &edev->aq; + int err; + + destroy_cmd.cq_idx = params->cq_idx; + destroy_cmd.aq_common_desc.opcode = EFA_ADMIN_DESTROY_CQ; + + err = efa_com_cmd_exec(aq, + (struct efa_admin_aq_entry *)&destroy_cmd, + sizeof(destroy_cmd), + (struct efa_admin_acq_entry *)&destroy_resp, + sizeof(destroy_resp)); + + if (err) + ibdev_err(edev->efa_dev, "Failed to destroy CQ-%u [%d]\n", + params->cq_idx, err); + + return 0; +} + +int efa_com_register_mr(struct efa_com_dev *edev, + struct efa_com_reg_mr_params *params, + struct efa_com_reg_mr_result *result) +{ + struct efa_admin_reg_mr_resp cmd_completion; + struct efa_com_admin_queue *aq = &edev->aq; + struct efa_admin_reg_mr_cmd mr_cmd = {}; + int err; + + mr_cmd.aq_common_desc.opcode = EFA_ADMIN_REG_MR; + mr_cmd.pd = params->pd; + mr_cmd.mr_length = params->mr_length_in_bytes; + mr_cmd.flags |= params->page_shift & + EFA_ADMIN_REG_MR_CMD_PHYS_PAGE_SIZE_SHIFT_MASK; + mr_cmd.iova = params->iova; + mr_cmd.permissions |= params->permissions & + EFA_ADMIN_REG_MR_CMD_LOCAL_WRITE_ENABLE_MASK; + + if (params->inline_pbl) { + memcpy(mr_cmd.pbl.inline_pbl_array, + params->pbl.inline_pbl_array, + sizeof(mr_cmd.pbl.inline_pbl_array)); + } else { + mr_cmd.pbl.pbl.length = params->pbl.pbl.length; + mr_cmd.pbl.pbl.address.mem_addr_low = + params->pbl.pbl.address.mem_addr_low; + mr_cmd.pbl.pbl.address.mem_addr_high = + params->pbl.pbl.address.mem_addr_high; + mr_cmd.aq_common_desc.flags |= + EFA_ADMIN_AQ_COMMON_DESC_CTRL_DATA_MASK; + if (params->indirect) + mr_cmd.aq_common_desc.flags |= + EFA_ADMIN_AQ_COMMON_DESC_CTRL_DATA_INDIRECT_MASK; + } + + err = efa_com_cmd_exec(aq, + (struct efa_admin_aq_entry *)&mr_cmd, + sizeof(mr_cmd), + (struct efa_admin_acq_entry *)&cmd_completion, + sizeof(cmd_completion)); + if (err) { + ibdev_err(edev->efa_dev, "Failed to register mr [%d]\n", err); + return err; + } + + result->l_key = cmd_completion.l_key; + result->r_key = cmd_completion.r_key; + + return 0; +} + +int efa_com_dereg_mr(struct efa_com_dev *edev, + struct efa_com_dereg_mr_params *params) +{ + struct efa_admin_dereg_mr_resp cmd_completion; + struct efa_com_admin_queue *aq = &edev->aq; + struct efa_admin_dereg_mr_cmd mr_cmd = {}; + int err; + + mr_cmd.aq_common_desc.opcode = EFA_ADMIN_DEREG_MR; + mr_cmd.l_key = params->l_key; + + err = efa_com_cmd_exec(aq, + (struct efa_admin_aq_entry *)&mr_cmd, + sizeof(mr_cmd), + (struct efa_admin_acq_entry *)&cmd_completion, + sizeof(cmd_completion)); + if (err) + ibdev_err(edev->efa_dev, + "Failed to de-register mr(lkey-%u) [%d]\n", + mr_cmd.l_key, err); + + return 0; +} + +int efa_com_create_ah(struct efa_com_dev *edev, + struct efa_com_create_ah_params *params, + struct efa_com_create_ah_result *result) +{ + struct efa_admin_create_ah_resp cmd_completion; + struct efa_com_admin_queue *aq = &edev->aq; + struct efa_admin_create_ah_cmd ah_cmd = {}; + int err; + + ah_cmd.aq_common_desc.opcode = EFA_ADMIN_CREATE_AH; + + memcpy(ah_cmd.dest_addr, params->dest_addr, sizeof(ah_cmd.dest_addr)); + ah_cmd.pd = params->pdn; + + err = efa_com_cmd_exec(aq, + (struct efa_admin_aq_entry *)&ah_cmd, + sizeof(ah_cmd), + (struct efa_admin_acq_entry *)&cmd_completion, + sizeof(cmd_completion)); + if (err) { + ibdev_err(edev->efa_dev, "Failed to create ah [%d]\n", err); + return err; + } + + result->ah = cmd_completion.ah; + + return 0; +} + +int efa_com_destroy_ah(struct efa_com_dev *edev, + struct efa_com_destroy_ah_params *params) +{ + struct efa_admin_destroy_ah_resp cmd_completion; + struct efa_admin_destroy_ah_cmd ah_cmd = {}; + struct efa_com_admin_queue *aq = &edev->aq; + int err; + + ah_cmd.aq_common_desc.opcode = EFA_ADMIN_DESTROY_AH; + ah_cmd.ah = params->ah; + ah_cmd.pd = params->pdn; + + err = efa_com_cmd_exec(aq, + (struct efa_admin_aq_entry *)&ah_cmd, + sizeof(ah_cmd), + (struct efa_admin_acq_entry *)&cmd_completion, + sizeof(cmd_completion)); + if (err) + ibdev_err(edev->efa_dev, "Failed to destroy ah-%d pd-%d [%d]\n", + ah_cmd.ah, ah_cmd.pd, err); + + return 0; +} + +static bool +efa_com_check_supported_feature_id(struct efa_com_dev *edev, + enum efa_admin_aq_feature_id feature_id) +{ + u32 feature_mask = 1 << feature_id; + + /* Device attributes is always supported */ + if (feature_id != EFA_ADMIN_DEVICE_ATTR && + !(edev->supported_features & feature_mask)) + return false; + + return true; +} + +static int efa_com_get_feature_ex(struct efa_com_dev *edev, + struct efa_admin_get_feature_resp *get_resp, + enum efa_admin_aq_feature_id feature_id, + dma_addr_t control_buf_dma_addr, + u32 control_buff_size) +{ + struct efa_admin_get_feature_cmd get_cmd = {}; + struct efa_com_admin_queue *aq; + int err; + + if (!efa_com_check_supported_feature_id(edev, feature_id)) { + ibdev_err(edev->efa_dev, "Feature %d isn't supported\n", + feature_id); + return -EOPNOTSUPP; + } + + aq = &edev->aq; + + get_cmd.aq_common_descriptor.opcode = EFA_ADMIN_GET_FEATURE; + + if (control_buff_size) + get_cmd.aq_common_descriptor.flags = + EFA_ADMIN_AQ_COMMON_DESC_CTRL_DATA_INDIRECT_MASK; + + + efa_com_set_dma_addr(control_buf_dma_addr, + &get_cmd.control_buffer.address.mem_addr_high, + &get_cmd.control_buffer.address.mem_addr_low); + + get_cmd.control_buffer.length = control_buff_size; + get_cmd.feature_common.feature_id = feature_id; + err = efa_com_cmd_exec(aq, + (struct efa_admin_aq_entry *) + &get_cmd, + sizeof(get_cmd), + (struct efa_admin_acq_entry *) + get_resp, + sizeof(*get_resp)); + + if (err) + ibdev_err(edev->efa_dev, + "Failed to submit get_feature command %d [%d]\n", + feature_id, err); + + return 0; +} + +static int efa_com_get_feature(struct efa_com_dev *edev, + struct efa_admin_get_feature_resp *get_resp, + enum efa_admin_aq_feature_id feature_id) +{ + return efa_com_get_feature_ex(edev, get_resp, feature_id, 0, 0); +} + +int efa_com_get_network_attr(struct efa_com_dev *edev, + struct efa_com_get_network_attr_result *result) +{ + struct efa_admin_get_feature_resp resp; + int err; + + err = efa_com_get_feature(edev, &resp, + EFA_ADMIN_NETWORK_ATTR); + if (err) { + ibdev_err(edev->efa_dev, + "Failed to get network attributes %d\n", err); + return err; + } + + memcpy(result->addr, resp.u.network_attr.addr, + sizeof(resp.u.network_attr.addr)); + result->mtu = resp.u.network_attr.mtu; + + return 0; +} + +int efa_com_get_device_attr(struct efa_com_dev *edev, + struct efa_com_get_device_attr_result *result) +{ + struct efa_admin_get_feature_resp resp; + int err; + + err = efa_com_get_feature(edev, &resp, EFA_ADMIN_DEVICE_ATTR); + if (err) { + ibdev_err(edev->efa_dev, "Failed to get device attributes %d\n", + err); + return err; + } + + result->page_size_cap = resp.u.device_attr.page_size_cap; + result->fw_version = resp.u.device_attr.fw_version; + result->admin_api_version = resp.u.device_attr.admin_api_version; + result->device_version = resp.u.device_attr.device_version; + result->supported_features = resp.u.device_attr.supported_features; + result->phys_addr_width = resp.u.device_attr.phys_addr_width; + result->virt_addr_width = resp.u.device_attr.virt_addr_width; + result->db_bar = resp.u.device_attr.db_bar; + + if (result->admin_api_version < 1) { + ibdev_err(edev->efa_dev, + "Failed to get device attr api version [%u < 1]\n", + result->admin_api_version); + return -EINVAL; + } + + edev->supported_features = resp.u.device_attr.supported_features; + err = efa_com_get_feature(edev, &resp, + EFA_ADMIN_QUEUE_ATTR); + if (err) { + ibdev_err(edev->efa_dev, + "Failed to get network attributes %d\n", err); + return err; + } + + result->max_qp = resp.u.queue_attr.max_qp; + result->max_sq_depth = resp.u.queue_attr.max_sq_depth; + result->max_rq_depth = resp.u.queue_attr.max_rq_depth; + result->max_cq = resp.u.queue_attr.max_cq; + result->max_cq_depth = resp.u.queue_attr.max_cq_depth; + result->inline_buf_size = resp.u.queue_attr.inline_buf_size; + result->max_sq_sge = resp.u.queue_attr.max_wr_send_sges; + result->max_rq_sge = resp.u.queue_attr.max_wr_recv_sges; + result->max_mr = resp.u.queue_attr.max_mr; + result->max_mr_pages = resp.u.queue_attr.max_mr_pages; + result->max_pd = resp.u.queue_attr.max_pd; + result->max_ah = resp.u.queue_attr.max_ah; + result->max_llq_size = resp.u.queue_attr.max_llq_size; + result->sub_cqs_per_cq = resp.u.queue_attr.sub_cqs_per_cq; + + return 0; +} + +int efa_com_get_hw_hints(struct efa_com_dev *edev, + struct efa_com_get_hw_hints_result *result) +{ + struct efa_admin_get_feature_resp resp; + int err; + + err = efa_com_get_feature(edev, &resp, EFA_ADMIN_HW_HINTS); + if (err) { + ibdev_err(edev->efa_dev, "Failed to get hw hints %d\n", err); + return err; + } + + result->admin_completion_timeout = resp.u.hw_hints.admin_completion_timeout; + result->driver_watchdog_timeout = resp.u.hw_hints.driver_watchdog_timeout; + result->mmio_read_timeout = resp.u.hw_hints.mmio_read_timeout; + result->poll_interval = resp.u.hw_hints.poll_interval; + + return 0; +} + +static int efa_com_set_feature_ex(struct efa_com_dev *edev, + struct efa_admin_set_feature_resp *set_resp, + struct efa_admin_set_feature_cmd *set_cmd, + enum efa_admin_aq_feature_id feature_id, + dma_addr_t control_buf_dma_addr, + u32 control_buff_size) +{ + struct efa_com_admin_queue *aq; + int err; + + if (!efa_com_check_supported_feature_id(edev, feature_id)) { + ibdev_err(edev->efa_dev, "Feature %d isn't supported\n", + feature_id); + return -EOPNOTSUPP; + } + + aq = &edev->aq; + + set_cmd->aq_common_descriptor.opcode = EFA_ADMIN_SET_FEATURE; + if (control_buff_size) { + set_cmd->aq_common_descriptor.flags = + EFA_ADMIN_AQ_COMMON_DESC_CTRL_DATA_INDIRECT_MASK; + efa_com_set_dma_addr(control_buf_dma_addr, + &set_cmd->control_buffer.address.mem_addr_high, + &set_cmd->control_buffer.address.mem_addr_low); + } + + set_cmd->control_buffer.length = control_buff_size; + set_cmd->feature_common.feature_id = feature_id; + err = efa_com_cmd_exec(aq, + (struct efa_admin_aq_entry *)set_cmd, + sizeof(*set_cmd), + (struct efa_admin_acq_entry *)set_resp, + sizeof(*set_resp)); + + if (err) + ibdev_err(edev->efa_dev, + "Failed to submit set_feature command %d error: %d\n", + feature_id, err); + + return 0; +} + +static int efa_com_set_feature(struct efa_com_dev *edev, + struct efa_admin_set_feature_resp *set_resp, + struct efa_admin_set_feature_cmd *set_cmd, + enum efa_admin_aq_feature_id feature_id) +{ + return efa_com_set_feature_ex(edev, set_resp, set_cmd, feature_id, + 0, 0); +} + +int efa_com_set_aenq_config(struct efa_com_dev *edev, u32 groups) +{ + struct efa_admin_get_feature_resp get_resp; + struct efa_admin_set_feature_resp set_resp; + struct efa_admin_set_feature_cmd cmd = {}; + int err; + + ibdev_dbg(edev->efa_dev, "Configuring aenq with groups[%#x]\n", groups); + + err = efa_com_get_feature(edev, &get_resp, EFA_ADMIN_AENQ_CONFIG); + if (err) { + ibdev_err(edev->efa_dev, "Failed to get aenq attributes: %d\n", + err); + return err; + } + + ibdev_dbg(edev->efa_dev, + "Get aenq groups: supported[%#x] enabled[%#x]\n", + get_resp.u.aenq.supported_groups, + get_resp.u.aenq.enabled_groups); + + if ((get_resp.u.aenq.supported_groups & groups) != groups) { + ibdev_err(edev->efa_dev, + "Trying to set unsupported aenq groups[%#x] supported[%#x]\n", + groups, get_resp.u.aenq.supported_groups); + return -EOPNOTSUPP; + } + + cmd.u.aenq.enabled_groups = groups; + err = efa_com_set_feature(edev, &set_resp, &cmd, + EFA_ADMIN_AENQ_CONFIG); + if (err) { + ibdev_err(edev->efa_dev, "Failed to set aenq attributes: %d\n", + err); + return err; + } + + return 0; +} + +int efa_com_alloc_pd(struct efa_com_dev *edev, + struct efa_com_alloc_pd_result *result) +{ + struct efa_com_admin_queue *aq = &edev->aq; + struct efa_admin_alloc_pd_cmd cmd = {}; + struct efa_admin_alloc_pd_resp resp; + int err; + + cmd.aq_common_descriptor.opcode = EFA_ADMIN_ALLOC_PD; + + err = efa_com_cmd_exec(aq, + (struct efa_admin_aq_entry *)&cmd, + sizeof(cmd), + (struct efa_admin_acq_entry *)&resp, + sizeof(resp)); + if (err) { + ibdev_err(edev->efa_dev, "Failed to allocate pd[%d]\n", err); + return err; + } + + result->pdn = resp.pd; + + return 0; +} + +int efa_com_dealloc_pd(struct efa_com_dev *edev, + struct efa_com_dealloc_pd_params *params) +{ + struct efa_com_admin_queue *aq = &edev->aq; + struct efa_admin_dealloc_pd_cmd cmd = {}; + struct efa_admin_dealloc_pd_resp resp; + int err; + + cmd.aq_common_descriptor.opcode = EFA_ADMIN_DEALLOC_PD; + cmd.pd = params->pdn; + + err = efa_com_cmd_exec(aq, + (struct efa_admin_aq_entry *)&cmd, + sizeof(cmd), + (struct efa_admin_acq_entry *)&resp, + sizeof(resp)); + if (err) { + ibdev_err(edev->efa_dev, "Failed to deallocate pd-%u [%d]\n", + cmd.pd, err); + return err; + } + + return 0; +} + +int efa_com_alloc_uar(struct efa_com_dev *edev, + struct efa_com_alloc_uar_result *result) +{ + struct efa_com_admin_queue *aq = &edev->aq; + struct efa_admin_alloc_uar_cmd cmd = {}; + struct efa_admin_alloc_uar_resp resp; + int err; + + cmd.aq_common_descriptor.opcode = EFA_ADMIN_ALLOC_UAR; + + err = efa_com_cmd_exec(aq, + (struct efa_admin_aq_entry *)&cmd, + sizeof(cmd), + (struct efa_admin_acq_entry *)&resp, + sizeof(resp)); + if (err) { + ibdev_err(edev->efa_dev, "Failed to allocate uar[%d]\n", err); + return err; + } + + result->uarn = resp.uar; + + return 0; +} + +int efa_com_dealloc_uar(struct efa_com_dev *edev, + struct efa_com_dealloc_uar_params *params) +{ + struct efa_com_admin_queue *aq = &edev->aq; + struct efa_admin_dealloc_uar_cmd cmd = {}; + struct efa_admin_dealloc_uar_resp resp; + int err; + + cmd.aq_common_descriptor.opcode = EFA_ADMIN_DEALLOC_UAR; + cmd.uar = params->uarn; + + err = efa_com_cmd_exec(aq, + (struct efa_admin_aq_entry *)&cmd, + sizeof(cmd), + (struct efa_admin_acq_entry *)&resp, + sizeof(resp)); + if (err) { + ibdev_err(edev->efa_dev, "Failed to deallocate uar-%u [%d]\n", + cmd.uar, err); + return err; + } + + return 0; +} diff --git a/drivers/infiniband/hw/efa/efa_com_cmd.h b/drivers/infiniband/hw/efa/efa_com_cmd.h new file mode 100644 index 000000000000..a1174380462c --- /dev/null +++ b/drivers/infiniband/hw/efa/efa_com_cmd.h @@ -0,0 +1,270 @@ +/* SPDX-License-Identifier: GPL-2.0 OR BSD-2-Clause */ +/* + * Copyright 2018-2019 Amazon.com, Inc. or its affiliates. All rights reserved. + */ + +#ifndef _EFA_COM_CMD_H_ +#define _EFA_COM_CMD_H_ + +#include "efa_com.h" + +#define EFA_GID_SIZE 16 + +struct efa_com_create_qp_params { + u64 rq_base_addr; + u32 send_cq_idx; + u32 recv_cq_idx; + /* + * Send descriptor ring size in bytes, + * sufficient for user-provided number of WQEs and SGL size + */ + u32 sq_ring_size_in_bytes; + /* Max number of WQEs that will be posted on send queue */ + u32 sq_depth; + /* Recv descriptor ring size in bytes */ + u32 rq_ring_size_in_bytes; + u32 rq_depth; + u16 pd; + u16 uarn; + u8 qp_type; +}; + +struct efa_com_create_qp_result { + u32 qp_handle; + u32 qp_num; + u32 sq_db_offset; + u32 rq_db_offset; + u32 llq_descriptors_offset; + u16 send_sub_cq_idx; + u16 recv_sub_cq_idx; +}; + +struct efa_com_modify_qp_params { + u32 modify_mask; + u32 qp_handle; + u32 qp_state; + u32 cur_qp_state; + u32 qkey; + u32 sq_psn; + u8 sq_drained_async_notify; +}; + +struct efa_com_query_qp_params { + u32 qp_handle; +}; + +struct efa_com_query_qp_result { + u32 qp_state; + u32 qkey; + u32 sq_draining; + u32 sq_psn; +}; + +struct efa_com_destroy_qp_params { + u32 qp_handle; +}; + +struct efa_com_create_cq_params { + /* cq physical base address in OS memory */ + dma_addr_t dma_addr; + /* completion queue depth in # of entries */ + u16 cq_depth; + u16 num_sub_cqs; + u16 uarn; + u8 entry_size_in_bytes; +}; + +struct efa_com_create_cq_result { + /* cq identifier */ + u16 cq_idx; + /* actual cq depth in # of entries */ + u16 actual_depth; +}; + +struct efa_com_destroy_cq_params { + u16 cq_idx; +}; + +struct efa_com_create_ah_params { + u16 pdn; + /* Destination address in network byte order */ + u8 dest_addr[EFA_GID_SIZE]; +}; + +struct efa_com_create_ah_result { + u16 ah; +}; + +struct efa_com_destroy_ah_params { + u16 ah; + u16 pdn; +}; + +struct efa_com_get_network_attr_result { + u8 addr[EFA_GID_SIZE]; + u32 mtu; +}; + +struct efa_com_get_device_attr_result { + u64 page_size_cap; + u64 max_mr_pages; + u32 fw_version; + u32 admin_api_version; + u32 device_version; + u32 supported_features; + u32 phys_addr_width; + u32 virt_addr_width; + u32 max_qp; + u32 max_sq_depth; /* wqes */ + u32 max_rq_depth; /* wqes */ + u32 max_cq; + u32 max_cq_depth; /* cqes */ + u32 inline_buf_size; + u32 max_mr; + u32 max_pd; + u32 max_ah; + u32 max_llq_size; + u16 sub_cqs_per_cq; + u16 max_sq_sge; + u16 max_rq_sge; + u8 db_bar; +}; + +struct efa_com_get_hw_hints_result { + u16 mmio_read_timeout; + u16 driver_watchdog_timeout; + u16 admin_completion_timeout; + u16 poll_interval; + u32 reserved[4]; +}; + +struct efa_com_mem_addr { + u32 mem_addr_low; + u32 mem_addr_high; +}; + +/* Used at indirect mode page list chunks for chaining */ +struct efa_com_ctrl_buff_info { + /* indicates length of the buffer pointed by control_buffer_address. */ + u32 length; + /* points to control buffer (direct or indirect) */ + struct efa_com_mem_addr address; +}; + +struct efa_com_reg_mr_params { + /* Memory region length, in bytes. */ + u64 mr_length_in_bytes; + /* IO Virtual Address associated with this MR. */ + u64 iova; + /* words 8:15: Physical Buffer List, each element is page-aligned. */ + union { + /* + * Inline array of physical addresses of app pages + * (optimization for short region reservations) + */ + u64 inline_pbl_array[4]; + /* + * Describes the next physically contiguous chunk of indirect + * page list. A page list contains physical addresses of command + * data pages. Data pages are 4KB; page list chunks are + * variable-sized. + */ + struct efa_com_ctrl_buff_info pbl; + } pbl; + /* number of pages in PBL (redundant, could be calculated) */ + u32 page_num; + /* Protection Domain */ + u16 pd; + /* + * phys_page_size_shift - page size is (1 << phys_page_size_shift) + * Page size is used for building the Virtual to Physical + * address mapping + */ + u8 page_shift; + /* + * permissions + * 0: local_write_enable - Write permissions: value of 1 needed + * for RQ buffers and for RDMA write:1: reserved1 - remote + * access flags, etc + */ + u8 permissions; + u8 inline_pbl; + u8 indirect; +}; + +struct efa_com_reg_mr_result { + /* + * To be used in conjunction with local buffers references in SQ and + * RQ WQE + */ + u32 l_key; + /* + * To be used in incoming RDMA semantics messages to refer to remotely + * accessed memory region + */ + u32 r_key; +}; + +struct efa_com_dereg_mr_params { + u32 l_key; +}; + +struct efa_com_alloc_pd_result { + u16 pdn; +}; + +struct efa_com_dealloc_pd_params { + u16 pdn; +}; + +struct efa_com_alloc_uar_result { + u16 uarn; +}; + +struct efa_com_dealloc_uar_params { + u16 uarn; +}; + +void efa_com_set_dma_addr(dma_addr_t addr, u32 *addr_high, u32 *addr_low); +int efa_com_create_qp(struct efa_com_dev *edev, + struct efa_com_create_qp_params *params, + struct efa_com_create_qp_result *res); +int efa_com_modify_qp(struct efa_com_dev *edev, + struct efa_com_modify_qp_params *params); +int efa_com_query_qp(struct efa_com_dev *edev, + struct efa_com_query_qp_params *params, + struct efa_com_query_qp_result *result); +int efa_com_destroy_qp(struct efa_com_dev *edev, + struct efa_com_destroy_qp_params *params); +int efa_com_create_cq(struct efa_com_dev *edev, + struct efa_com_create_cq_params *params, + struct efa_com_create_cq_result *result); +int efa_com_destroy_cq(struct efa_com_dev *edev, + struct efa_com_destroy_cq_params *params); +int efa_com_register_mr(struct efa_com_dev *edev, + struct efa_com_reg_mr_params *params, + struct efa_com_reg_mr_result *result); +int efa_com_dereg_mr(struct efa_com_dev *edev, + struct efa_com_dereg_mr_params *params); +int efa_com_create_ah(struct efa_com_dev *edev, + struct efa_com_create_ah_params *params, + struct efa_com_create_ah_result *result); +int efa_com_destroy_ah(struct efa_com_dev *edev, + struct efa_com_destroy_ah_params *params); +int efa_com_get_network_attr(struct efa_com_dev *edev, + struct efa_com_get_network_attr_result *result); +int efa_com_get_device_attr(struct efa_com_dev *edev, + struct efa_com_get_device_attr_result *result); +int efa_com_get_hw_hints(struct efa_com_dev *edev, + struct efa_com_get_hw_hints_result *result); +int efa_com_set_aenq_config(struct efa_com_dev *edev, u32 groups); +int efa_com_alloc_pd(struct efa_com_dev *edev, + struct efa_com_alloc_pd_result *result); +int efa_com_dealloc_pd(struct efa_com_dev *edev, + struct efa_com_dealloc_pd_params *params); +int efa_com_alloc_uar(struct efa_com_dev *edev, + struct efa_com_alloc_uar_result *result); +int efa_com_dealloc_uar(struct efa_com_dev *edev, + struct efa_com_dealloc_uar_params *params); + +#endif /* _EFA_COM_CMD_H_ */ diff --git a/drivers/infiniband/hw/efa/efa_common_defs.h b/drivers/infiniband/hw/efa/efa_common_defs.h new file mode 100644 index 000000000000..c559ec08898e --- /dev/null +++ b/drivers/infiniband/hw/efa/efa_common_defs.h @@ -0,0 +1,18 @@ +/* SPDX-License-Identifier: GPL-2.0 OR BSD-2-Clause */ +/* + * Copyright 2018-2019 Amazon.com, Inc. or its affiliates. All rights reserved. + */ + +#ifndef _EFA_COMMON_H_ +#define _EFA_COMMON_H_ + +#define EFA_COMMON_SPEC_VERSION_MAJOR 2 +#define EFA_COMMON_SPEC_VERSION_MINOR 0 + +struct efa_common_mem_addr { + u32 mem_addr_low; + + u32 mem_addr_high; +}; + +#endif /* _EFA_COMMON_H_ */ diff --git a/drivers/infiniband/hw/efa/efa_main.c b/drivers/infiniband/hw/efa/efa_main.c new file mode 100644 index 000000000000..db974caf1eb1 --- /dev/null +++ b/drivers/infiniband/hw/efa/efa_main.c @@ -0,0 +1,533 @@ +// SPDX-License-Identifier: GPL-2.0 OR BSD-2-Clause +/* + * Copyright 2018-2019 Amazon.com, Inc. or its affiliates. All rights reserved. + */ + +#include <linux/module.h> +#include <linux/pci.h> + +#include <rdma/ib_user_verbs.h> + +#include "efa.h" + +#define PCI_DEV_ID_EFA_VF 0xefa0 + +static const struct pci_device_id efa_pci_tbl[] = { + { PCI_VDEVICE(AMAZON, PCI_DEV_ID_EFA_VF) }, + { } +}; + +MODULE_AUTHOR("Amazon.com, Inc. or its affiliates"); +MODULE_LICENSE("Dual BSD/GPL"); +MODULE_DESCRIPTION(DEVICE_NAME); +MODULE_DEVICE_TABLE(pci, efa_pci_tbl); + +#define EFA_REG_BAR 0 +#define EFA_MEM_BAR 2 +#define EFA_BASE_BAR_MASK (BIT(EFA_REG_BAR) | BIT(EFA_MEM_BAR)) + +#define EFA_AENQ_ENABLED_GROUPS \ + (BIT(EFA_ADMIN_FATAL_ERROR) | BIT(EFA_ADMIN_WARNING) | \ + BIT(EFA_ADMIN_NOTIFICATION) | BIT(EFA_ADMIN_KEEP_ALIVE)) + +static void efa_update_network_attr(struct efa_dev *dev, + struct efa_com_get_network_attr_result *network_attr) +{ + memcpy(dev->addr, network_attr->addr, sizeof(network_attr->addr)); + dev->mtu = network_attr->mtu; + + dev_dbg(&dev->pdev->dev, "Full address %pI6\n", dev->addr); +} + +/* This handler will called for unknown event group or unimplemented handlers */ +static void unimplemented_aenq_handler(void *data, + struct efa_admin_aenq_entry *aenq_e) +{ + struct efa_dev *dev = (struct efa_dev *)data; + + ibdev_err(&dev->ibdev, + "Unknown event was received or event with unimplemented handler\n"); +} + +static void efa_keep_alive(void *data, struct efa_admin_aenq_entry *aenq_e) +{ + struct efa_dev *dev = (struct efa_dev *)data; + + atomic64_inc(&dev->stats.keep_alive_rcvd); +} + +static struct efa_aenq_handlers aenq_handlers = { + .handlers = { + [EFA_ADMIN_KEEP_ALIVE] = efa_keep_alive, + }, + .unimplemented_handler = unimplemented_aenq_handler +}; + +static void efa_release_bars(struct efa_dev *dev, int bars_mask) +{ + struct pci_dev *pdev = dev->pdev; + int release_bars; + + release_bars = pci_select_bars(pdev, IORESOURCE_MEM) & bars_mask; + pci_release_selected_regions(pdev, release_bars); +} + +static irqreturn_t efa_intr_msix_mgmnt(int irq, void *data) +{ + struct efa_dev *dev = data; + + efa_com_admin_q_comp_intr_handler(&dev->edev); + efa_com_aenq_intr_handler(&dev->edev, data); + + return IRQ_HANDLED; +} + +static int efa_request_mgmnt_irq(struct efa_dev *dev) +{ + struct efa_irq *irq; + int err; + + irq = &dev->admin_irq; + err = request_irq(irq->vector, irq->handler, 0, irq->name, + irq->data); + if (err) { + dev_err(&dev->pdev->dev, "Failed to request admin irq (%d)\n", + err); + return err; + } + + dev_dbg(&dev->pdev->dev, "Set affinity hint of mgmnt irq to %*pbl (irq vector: %d)\n", + nr_cpumask_bits, &irq->affinity_hint_mask, irq->vector); + irq_set_affinity_hint(irq->vector, &irq->affinity_hint_mask); + + return err; +} + +static void efa_setup_mgmnt_irq(struct efa_dev *dev) +{ + u32 cpu; + + snprintf(dev->admin_irq.name, EFA_IRQNAME_SIZE, + "efa-mgmnt@pci:%s", pci_name(dev->pdev)); + dev->admin_irq.handler = efa_intr_msix_mgmnt; + dev->admin_irq.data = dev; + dev->admin_irq.vector = + pci_irq_vector(dev->pdev, dev->admin_msix_vector_idx); + cpu = cpumask_first(cpu_online_mask); + dev->admin_irq.cpu = cpu; + cpumask_set_cpu(cpu, + &dev->admin_irq.affinity_hint_mask); + dev_info(&dev->pdev->dev, "Setup irq:0x%p vector:%d name:%s\n", + &dev->admin_irq, + dev->admin_irq.vector, + dev->admin_irq.name); +} + +static void efa_free_mgmnt_irq(struct efa_dev *dev) +{ + struct efa_irq *irq; + + irq = &dev->admin_irq; + irq_set_affinity_hint(irq->vector, NULL); + free_irq(irq->vector, irq->data); +} + +static int efa_set_mgmnt_irq(struct efa_dev *dev) +{ + efa_setup_mgmnt_irq(dev); + + return efa_request_mgmnt_irq(dev); +} + +static int efa_request_doorbell_bar(struct efa_dev *dev) +{ + u8 db_bar_idx = dev->dev_attr.db_bar; + struct pci_dev *pdev = dev->pdev; + int bars; + int err; + + if (!(BIT(db_bar_idx) & EFA_BASE_BAR_MASK)) { + bars = pci_select_bars(pdev, IORESOURCE_MEM) & BIT(db_bar_idx); + + err = pci_request_selected_regions(pdev, bars, DRV_MODULE_NAME); + if (err) { + dev_err(&dev->pdev->dev, + "pci_request_selected_regions for bar %d failed %d\n", + db_bar_idx, err); + return err; + } + } + + dev->db_bar_addr = pci_resource_start(dev->pdev, db_bar_idx); + dev->db_bar_len = pci_resource_len(dev->pdev, db_bar_idx); + + return 0; +} + +static void efa_release_doorbell_bar(struct efa_dev *dev) +{ + if (!(BIT(dev->dev_attr.db_bar) & EFA_BASE_BAR_MASK)) + efa_release_bars(dev, BIT(dev->dev_attr.db_bar)); +} + +static void efa_update_hw_hints(struct efa_dev *dev, + struct efa_com_get_hw_hints_result *hw_hints) +{ + struct efa_com_dev *edev = &dev->edev; + + if (hw_hints->mmio_read_timeout) + edev->mmio_read.mmio_read_timeout = + hw_hints->mmio_read_timeout * 1000; + + if (hw_hints->poll_interval) + edev->aq.poll_interval = hw_hints->poll_interval; + + if (hw_hints->admin_completion_timeout) + edev->aq.completion_timeout = + hw_hints->admin_completion_timeout; +} + +static void efa_stats_init(struct efa_dev *dev) +{ + atomic64_t *s = (atomic64_t *)&dev->stats; + int i; + + for (i = 0; i < sizeof(dev->stats) / sizeof(*s); i++, s++) + atomic64_set(s, 0); +} + +static const struct ib_device_ops efa_dev_ops = { + .alloc_pd = efa_alloc_pd, + .alloc_ucontext = efa_alloc_ucontext, + .create_ah = efa_create_ah, + .create_cq = efa_create_cq, + .create_qp = efa_create_qp, + .dealloc_pd = efa_dealloc_pd, + .dealloc_ucontext = efa_dealloc_ucontext, + .dereg_mr = efa_dereg_mr, + .destroy_ah = efa_destroy_ah, + .destroy_cq = efa_destroy_cq, + .destroy_qp = efa_destroy_qp, + .get_link_layer = efa_port_link_layer, + .get_port_immutable = efa_get_port_immutable, + .mmap = efa_mmap, + .modify_qp = efa_modify_qp, + .query_device = efa_query_device, + .query_gid = efa_query_gid, + .query_pkey = efa_query_pkey, + .query_port = efa_query_port, + .query_qp = efa_query_qp, + .reg_user_mr = efa_reg_mr, + + INIT_RDMA_OBJ_SIZE(ib_ah, efa_ah, ibah), + INIT_RDMA_OBJ_SIZE(ib_pd, efa_pd, ibpd), + INIT_RDMA_OBJ_SIZE(ib_ucontext, efa_ucontext, ibucontext), +}; + +static int efa_ib_device_add(struct efa_dev *dev) +{ + struct efa_com_get_network_attr_result network_attr; + struct efa_com_get_hw_hints_result hw_hints; + struct pci_dev *pdev = dev->pdev; + int err; + + efa_stats_init(dev); + + err = efa_com_get_device_attr(&dev->edev, &dev->dev_attr); + if (err) + return err; + + dev_dbg(&dev->pdev->dev, "Doorbells bar (%d)\n", dev->dev_attr.db_bar); + err = efa_request_doorbell_bar(dev); + if (err) + return err; + + err = efa_com_get_network_attr(&dev->edev, &network_attr); + if (err) + goto err_release_doorbell_bar; + + efa_update_network_attr(dev, &network_attr); + + err = efa_com_get_hw_hints(&dev->edev, &hw_hints); + if (err) + goto err_release_doorbell_bar; + + efa_update_hw_hints(dev, &hw_hints); + + /* Try to enable all the available aenq groups */ + err = efa_com_set_aenq_config(&dev->edev, EFA_AENQ_ENABLED_GROUPS); + if (err) + goto err_release_doorbell_bar; + + dev->ibdev.owner = THIS_MODULE; + dev->ibdev.node_type = RDMA_NODE_UNSPECIFIED; + dev->ibdev.phys_port_cnt = 1; + dev->ibdev.num_comp_vectors = 1; + dev->ibdev.dev.parent = &pdev->dev; + dev->ibdev.uverbs_abi_ver = EFA_UVERBS_ABI_VERSION; + + dev->ibdev.uverbs_cmd_mask = + (1ull << IB_USER_VERBS_CMD_GET_CONTEXT) | + (1ull << IB_USER_VERBS_CMD_QUERY_DEVICE) | + (1ull << IB_USER_VERBS_CMD_QUERY_PORT) | + (1ull << IB_USER_VERBS_CMD_ALLOC_PD) | + (1ull << IB_USER_VERBS_CMD_DEALLOC_PD) | + (1ull << IB_USER_VERBS_CMD_REG_MR) | + (1ull << IB_USER_VERBS_CMD_DEREG_MR) | + (1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL) | + (1ull << IB_USER_VERBS_CMD_CREATE_CQ) | + (1ull << IB_USER_VERBS_CMD_DESTROY_CQ) | + (1ull << IB_USER_VERBS_CMD_CREATE_QP) | + (1ull << IB_USER_VERBS_CMD_MODIFY_QP) | + (1ull << IB_USER_VERBS_CMD_QUERY_QP) | + (1ull << IB_USER_VERBS_CMD_DESTROY_QP) | + (1ull << IB_USER_VERBS_CMD_CREATE_AH) | + (1ull << IB_USER_VERBS_CMD_DESTROY_AH); + + dev->ibdev.uverbs_ex_cmd_mask = + (1ull << IB_USER_VERBS_EX_CMD_QUERY_DEVICE); + + dev->ibdev.driver_id = RDMA_DRIVER_EFA; + ib_set_device_ops(&dev->ibdev, &efa_dev_ops); + + err = ib_register_device(&dev->ibdev, "efa_%d"); + if (err) + goto err_release_doorbell_bar; + + ibdev_info(&dev->ibdev, "IB device registered\n"); + + return 0; + +err_release_doorbell_bar: + efa_release_doorbell_bar(dev); + return err; +} + +static void efa_ib_device_remove(struct efa_dev *dev) +{ + efa_com_dev_reset(&dev->edev, EFA_REGS_RESET_NORMAL); + ibdev_info(&dev->ibdev, "Unregister ib device\n"); + ib_unregister_device(&dev->ibdev); + efa_release_doorbell_bar(dev); +} + +static void efa_disable_msix(struct efa_dev *dev) +{ + pci_free_irq_vectors(dev->pdev); +} + +static int efa_enable_msix(struct efa_dev *dev) +{ + int msix_vecs, irq_num; + + /* Reserve the max msix vectors we might need */ + msix_vecs = EFA_NUM_MSIX_VEC; + dev_dbg(&dev->pdev->dev, "Trying to enable MSI-X, vectors %d\n", + msix_vecs); + + dev->admin_msix_vector_idx = EFA_MGMNT_MSIX_VEC_IDX; + irq_num = pci_alloc_irq_vectors(dev->pdev, msix_vecs, + msix_vecs, PCI_IRQ_MSIX); + + if (irq_num < 0) { + dev_err(&dev->pdev->dev, "Failed to enable MSI-X. irq_num %d\n", + irq_num); + return -ENOSPC; + } + + if (irq_num != msix_vecs) { + dev_err(&dev->pdev->dev, + "Allocated %d MSI-X (out of %d requested)\n", + irq_num, msix_vecs); + return -ENOSPC; + } + + return 0; +} + +static int efa_device_init(struct efa_com_dev *edev, struct pci_dev *pdev) +{ + int dma_width; + int err; + + err = efa_com_dev_reset(edev, EFA_REGS_RESET_NORMAL); + if (err) + return err; + + err = efa_com_validate_version(edev); + if (err) + return err; + + dma_width = efa_com_get_dma_width(edev); + if (dma_width < 0) { + err = dma_width; + return err; + } + + err = pci_set_dma_mask(pdev, DMA_BIT_MASK(dma_width)); + if (err) { + dev_err(&pdev->dev, "pci_set_dma_mask failed %d\n", err); + return err; + } + + err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(dma_width)); + if (err) { + dev_err(&pdev->dev, + "err_pci_set_consistent_dma_mask failed %d\n", + err); + return err; + } + + return 0; +} + +static struct efa_dev *efa_probe_device(struct pci_dev *pdev) +{ + struct efa_com_dev *edev; + struct efa_dev *dev; + int bars; + int err; + + err = pci_enable_device_mem(pdev); + if (err) { + dev_err(&pdev->dev, "pci_enable_device_mem() failed!\n"); + return ERR_PTR(err); + } + + pci_set_master(pdev); + + dev = ib_alloc_device(efa_dev, ibdev); + if (!dev) { + dev_err(&pdev->dev, "Device alloc failed\n"); + err = -ENOMEM; + goto err_disable_device; + } + + pci_set_drvdata(pdev, dev); + edev = &dev->edev; + edev->efa_dev = dev; + edev->dmadev = &pdev->dev; + dev->pdev = pdev; + + bars = pci_select_bars(pdev, IORESOURCE_MEM) & EFA_BASE_BAR_MASK; + err = pci_request_selected_regions(pdev, bars, DRV_MODULE_NAME); + if (err) { + dev_err(&pdev->dev, "pci_request_selected_regions failed %d\n", + err); + goto err_ibdev_destroy; + } + + dev->reg_bar_addr = pci_resource_start(pdev, EFA_REG_BAR); + dev->reg_bar_len = pci_resource_len(pdev, EFA_REG_BAR); + dev->mem_bar_addr = pci_resource_start(pdev, EFA_MEM_BAR); + dev->mem_bar_len = pci_resource_len(pdev, EFA_MEM_BAR); + + edev->reg_bar = devm_ioremap(&pdev->dev, + dev->reg_bar_addr, + dev->reg_bar_len); + if (!edev->reg_bar) { + dev_err(&pdev->dev, "Failed to remap register bar\n"); + err = -EFAULT; + goto err_release_bars; + } + + err = efa_com_mmio_reg_read_init(edev); + if (err) { + dev_err(&pdev->dev, "Failed to init readless MMIO\n"); + goto err_iounmap; + } + + err = efa_device_init(edev, pdev); + if (err) { + dev_err(&pdev->dev, "EFA device init failed\n"); + if (err == -ETIME) + err = -EPROBE_DEFER; + goto err_reg_read_destroy; + } + + err = efa_enable_msix(dev); + if (err) + goto err_reg_read_destroy; + + edev->aq.msix_vector_idx = dev->admin_msix_vector_idx; + edev->aenq.msix_vector_idx = dev->admin_msix_vector_idx; + + err = efa_set_mgmnt_irq(dev); + if (err) + goto err_disable_msix; + + err = efa_com_admin_init(edev, &aenq_handlers); + if (err) + goto err_free_mgmnt_irq; + + return dev; + +err_free_mgmnt_irq: + efa_free_mgmnt_irq(dev); +err_disable_msix: + efa_disable_msix(dev); +err_reg_read_destroy: + efa_com_mmio_reg_read_destroy(edev); +err_iounmap: + devm_iounmap(&pdev->dev, edev->reg_bar); +err_release_bars: + efa_release_bars(dev, EFA_BASE_BAR_MASK); +err_ibdev_destroy: + ib_dealloc_device(&dev->ibdev); +err_disable_device: + pci_disable_device(pdev); + return ERR_PTR(err); +} + +static void efa_remove_device(struct pci_dev *pdev) +{ + struct efa_dev *dev = pci_get_drvdata(pdev); + struct efa_com_dev *edev; + + edev = &dev->edev; + efa_com_admin_destroy(edev); + efa_free_mgmnt_irq(dev); + efa_disable_msix(dev); + efa_com_mmio_reg_read_destroy(edev); + devm_iounmap(&pdev->dev, edev->reg_bar); + efa_release_bars(dev, EFA_BASE_BAR_MASK); + ib_dealloc_device(&dev->ibdev); + pci_disable_device(pdev); +} + +static int efa_probe(struct pci_dev *pdev, const struct pci_device_id *ent) +{ + struct efa_dev *dev; + int err; + + dev = efa_probe_device(pdev); + if (IS_ERR(dev)) + return PTR_ERR(dev); + + err = efa_ib_device_add(dev); + if (err) + goto err_remove_device; + + return 0; + +err_remove_device: + efa_remove_device(pdev); + return err; +} + +static void efa_remove(struct pci_dev *pdev) +{ + struct efa_dev *dev = pci_get_drvdata(pdev); + + efa_ib_device_remove(dev); + efa_remove_device(pdev); +} + +static struct pci_driver efa_pci_driver = { + .name = DRV_MODULE_NAME, + .id_table = efa_pci_tbl, + .probe = efa_probe, + .remove = efa_remove, +}; + +module_pci_driver(efa_pci_driver); diff --git a/drivers/infiniband/hw/efa/efa_regs_defs.h b/drivers/infiniband/hw/efa/efa_regs_defs.h new file mode 100644 index 000000000000..bb9cad3d6a15 --- /dev/null +++ b/drivers/infiniband/hw/efa/efa_regs_defs.h @@ -0,0 +1,113 @@ +/* SPDX-License-Identifier: GPL-2.0 OR BSD-2-Clause */ +/* + * Copyright 2018-2019 Amazon.com, Inc. or its affiliates. All rights reserved. + */ + +#ifndef _EFA_REGS_H_ +#define _EFA_REGS_H_ + +enum efa_regs_reset_reason_types { + EFA_REGS_RESET_NORMAL = 0, + /* Keep alive timeout */ + EFA_REGS_RESET_KEEP_ALIVE_TO = 1, + EFA_REGS_RESET_ADMIN_TO = 2, + EFA_REGS_RESET_INIT_ERR = 3, + EFA_REGS_RESET_DRIVER_INVALID_STATE = 4, + EFA_REGS_RESET_OS_TRIGGER = 5, + EFA_REGS_RESET_SHUTDOWN = 6, + EFA_REGS_RESET_USER_TRIGGER = 7, + EFA_REGS_RESET_GENERIC = 8, +}; + +/* efa_registers offsets */ + +/* 0 base */ +#define EFA_REGS_VERSION_OFF 0x0 +#define EFA_REGS_CONTROLLER_VERSION_OFF 0x4 +#define EFA_REGS_CAPS_OFF 0x8 +#define EFA_REGS_AQ_BASE_LO_OFF 0x10 +#define EFA_REGS_AQ_BASE_HI_OFF 0x14 +#define EFA_REGS_AQ_CAPS_OFF 0x18 +#define EFA_REGS_ACQ_BASE_LO_OFF 0x20 +#define EFA_REGS_ACQ_BASE_HI_OFF 0x24 +#define EFA_REGS_ACQ_CAPS_OFF 0x28 +#define EFA_REGS_AQ_PROD_DB_OFF 0x2c +#define EFA_REGS_AENQ_CAPS_OFF 0x34 +#define EFA_REGS_AENQ_BASE_LO_OFF 0x38 +#define EFA_REGS_AENQ_BASE_HI_OFF 0x3c +#define EFA_REGS_AENQ_CONS_DB_OFF 0x40 +#define EFA_REGS_INTR_MASK_OFF 0x4c +#define EFA_REGS_DEV_CTL_OFF 0x54 +#define EFA_REGS_DEV_STS_OFF 0x58 +#define EFA_REGS_MMIO_REG_READ_OFF 0x5c +#define EFA_REGS_MMIO_RESP_LO_OFF 0x60 +#define EFA_REGS_MMIO_RESP_HI_OFF 0x64 + +/* version register */ +#define EFA_REGS_VERSION_MINOR_VERSION_MASK 0xff +#define EFA_REGS_VERSION_MAJOR_VERSION_SHIFT 8 +#define EFA_REGS_VERSION_MAJOR_VERSION_MASK 0xff00 + +/* controller_version register */ +#define EFA_REGS_CONTROLLER_VERSION_SUBMINOR_VERSION_MASK 0xff +#define EFA_REGS_CONTROLLER_VERSION_MINOR_VERSION_SHIFT 8 +#define EFA_REGS_CONTROLLER_VERSION_MINOR_VERSION_MASK 0xff00 +#define EFA_REGS_CONTROLLER_VERSION_MAJOR_VERSION_SHIFT 16 +#define EFA_REGS_CONTROLLER_VERSION_MAJOR_VERSION_MASK 0xff0000 +#define EFA_REGS_CONTROLLER_VERSION_IMPL_ID_SHIFT 24 +#define EFA_REGS_CONTROLLER_VERSION_IMPL_ID_MASK 0xff000000 + +/* caps register */ +#define EFA_REGS_CAPS_CONTIGUOUS_QUEUE_REQUIRED_MASK 0x1 +#define EFA_REGS_CAPS_RESET_TIMEOUT_SHIFT 1 +#define EFA_REGS_CAPS_RESET_TIMEOUT_MASK 0x3e +#define EFA_REGS_CAPS_DMA_ADDR_WIDTH_SHIFT 8 +#define EFA_REGS_CAPS_DMA_ADDR_WIDTH_MASK 0xff00 +#define EFA_REGS_CAPS_ADMIN_CMD_TO_SHIFT 16 +#define EFA_REGS_CAPS_ADMIN_CMD_TO_MASK 0xf0000 + +/* aq_caps register */ +#define EFA_REGS_AQ_CAPS_AQ_DEPTH_MASK 0xffff +#define EFA_REGS_AQ_CAPS_AQ_ENTRY_SIZE_SHIFT 16 +#define EFA_REGS_AQ_CAPS_AQ_ENTRY_SIZE_MASK 0xffff0000 + +/* acq_caps register */ +#define EFA_REGS_ACQ_CAPS_ACQ_DEPTH_MASK 0xffff +#define EFA_REGS_ACQ_CAPS_ACQ_ENTRY_SIZE_SHIFT 16 +#define EFA_REGS_ACQ_CAPS_ACQ_ENTRY_SIZE_MASK 0xff0000 +#define EFA_REGS_ACQ_CAPS_ACQ_MSIX_VECTOR_SHIFT 24 +#define EFA_REGS_ACQ_CAPS_ACQ_MSIX_VECTOR_MASK 0xff000000 + +/* aenq_caps register */ +#define EFA_REGS_AENQ_CAPS_AENQ_DEPTH_MASK 0xffff +#define EFA_REGS_AENQ_CAPS_AENQ_ENTRY_SIZE_SHIFT 16 +#define EFA_REGS_AENQ_CAPS_AENQ_ENTRY_SIZE_MASK 0xff0000 +#define EFA_REGS_AENQ_CAPS_AENQ_MSIX_VECTOR_SHIFT 24 +#define EFA_REGS_AENQ_CAPS_AENQ_MSIX_VECTOR_MASK 0xff000000 + +/* dev_ctl register */ +#define EFA_REGS_DEV_CTL_DEV_RESET_MASK 0x1 +#define EFA_REGS_DEV_CTL_AQ_RESTART_SHIFT 1 +#define EFA_REGS_DEV_CTL_AQ_RESTART_MASK 0x2 +#define EFA_REGS_DEV_CTL_RESET_REASON_SHIFT 28 +#define EFA_REGS_DEV_CTL_RESET_REASON_MASK 0xf0000000 + +/* dev_sts register */ +#define EFA_REGS_DEV_STS_READY_MASK 0x1 +#define EFA_REGS_DEV_STS_AQ_RESTART_IN_PROGRESS_SHIFT 1 +#define EFA_REGS_DEV_STS_AQ_RESTART_IN_PROGRESS_MASK 0x2 +#define EFA_REGS_DEV_STS_AQ_RESTART_FINISHED_SHIFT 2 +#define EFA_REGS_DEV_STS_AQ_RESTART_FINISHED_MASK 0x4 +#define EFA_REGS_DEV_STS_RESET_IN_PROGRESS_SHIFT 3 +#define EFA_REGS_DEV_STS_RESET_IN_PROGRESS_MASK 0x8 +#define EFA_REGS_DEV_STS_RESET_FINISHED_SHIFT 4 +#define EFA_REGS_DEV_STS_RESET_FINISHED_MASK 0x10 +#define EFA_REGS_DEV_STS_FATAL_ERROR_SHIFT 5 +#define EFA_REGS_DEV_STS_FATAL_ERROR_MASK 0x20 + +/* mmio_reg_read register */ +#define EFA_REGS_MMIO_REG_READ_REQ_ID_MASK 0xffff +#define EFA_REGS_MMIO_REG_READ_REG_OFF_SHIFT 16 +#define EFA_REGS_MMIO_REG_READ_REG_OFF_MASK 0xffff0000 + +#endif /* _EFA_REGS_H_ */ diff --git a/drivers/infiniband/hw/efa/efa_verbs.c b/drivers/infiniband/hw/efa/efa_verbs.c new file mode 100644 index 000000000000..6d6886c9009f --- /dev/null +++ b/drivers/infiniband/hw/efa/efa_verbs.c @@ -0,0 +1,1825 @@ +// SPDX-License-Identifier: GPL-2.0 OR Linux-OpenIB +/* + * Copyright 2018-2019 Amazon.com, Inc. or its affiliates. All rights reserved. + */ + +#include <linux/vmalloc.h> + +#include <rdma/ib_addr.h> +#include <rdma/ib_umem.h> +#include <rdma/ib_user_verbs.h> +#include <rdma/ib_verbs.h> +#include <rdma/uverbs_ioctl.h> + +#include "efa.h" + +#define EFA_MMAP_FLAG_SHIFT 56 +#define EFA_MMAP_PAGE_MASK GENMASK(EFA_MMAP_FLAG_SHIFT - 1, 0) +#define EFA_MMAP_INVALID U64_MAX + +enum { + EFA_MMAP_DMA_PAGE = 0, + EFA_MMAP_IO_WC, + EFA_MMAP_IO_NC, +}; + +#define EFA_AENQ_ENABLED_GROUPS \ + (BIT(EFA_ADMIN_FATAL_ERROR) | BIT(EFA_ADMIN_WARNING) | \ + BIT(EFA_ADMIN_NOTIFICATION) | BIT(EFA_ADMIN_KEEP_ALIVE)) + +struct efa_mmap_entry { + void *obj; + u64 address; + u64 length; + u32 mmap_page; + u8 mmap_flag; +}; + +static inline u64 get_mmap_key(const struct efa_mmap_entry *efa) +{ + return ((u64)efa->mmap_flag << EFA_MMAP_FLAG_SHIFT) | + ((u64)efa->mmap_page << PAGE_SHIFT); +} + +#define EFA_CHUNK_PAYLOAD_SHIFT 12 +#define EFA_CHUNK_PAYLOAD_SIZE BIT(EFA_CHUNK_PAYLOAD_SHIFT) +#define EFA_CHUNK_PAYLOAD_PTR_SIZE 8 + +#define EFA_CHUNK_SHIFT 12 +#define EFA_CHUNK_SIZE BIT(EFA_CHUNK_SHIFT) +#define EFA_CHUNK_PTR_SIZE sizeof(struct efa_com_ctrl_buff_info) + +#define EFA_PTRS_PER_CHUNK \ + ((EFA_CHUNK_SIZE - EFA_CHUNK_PTR_SIZE) / EFA_CHUNK_PAYLOAD_PTR_SIZE) + +#define EFA_CHUNK_USED_SIZE \ + ((EFA_PTRS_PER_CHUNK * EFA_CHUNK_PAYLOAD_PTR_SIZE) + EFA_CHUNK_PTR_SIZE) + +#define EFA_SUPPORTED_ACCESS_FLAGS IB_ACCESS_LOCAL_WRITE + +struct pbl_chunk { + dma_addr_t dma_addr; + u64 *buf; + u32 length; +}; + +struct pbl_chunk_list { + struct pbl_chunk *chunks; + unsigned int size; +}; + +struct pbl_context { + union { + struct { + dma_addr_t dma_addr; + } continuous; + struct { + u32 pbl_buf_size_in_pages; + struct scatterlist *sgl; + int sg_dma_cnt; + struct pbl_chunk_list chunk_list; + } indirect; + } phys; + u64 *pbl_buf; + u32 pbl_buf_size_in_bytes; + u8 physically_continuous; +}; + +static inline struct efa_dev *to_edev(struct ib_device *ibdev) +{ + return container_of(ibdev, struct efa_dev, ibdev); +} + +static inline struct efa_ucontext *to_eucontext(struct ib_ucontext *ibucontext) +{ + return container_of(ibucontext, struct efa_ucontext, ibucontext); +} + +static inline struct efa_pd *to_epd(struct ib_pd *ibpd) +{ + return container_of(ibpd, struct efa_pd, ibpd); +} + +static inline struct efa_mr *to_emr(struct ib_mr *ibmr) +{ + return container_of(ibmr, struct efa_mr, ibmr); +} + +static inline struct efa_qp *to_eqp(struct ib_qp *ibqp) +{ + return container_of(ibqp, struct efa_qp, ibqp); +} + +static inline struct efa_cq *to_ecq(struct ib_cq *ibcq) +{ + return container_of(ibcq, struct efa_cq, ibcq); +} + +static inline struct efa_ah *to_eah(struct ib_ah *ibah) +{ + return container_of(ibah, struct efa_ah, ibah); +} + +#define field_avail(x, fld, sz) (offsetof(typeof(x), fld) + \ + sizeof(((typeof(x) *)0)->fld) <= (sz)) + +#define is_reserved_cleared(reserved) \ + !memchr_inv(reserved, 0, sizeof(reserved)) + +static void *efa_zalloc_mapped(struct efa_dev *dev, dma_addr_t *dma_addr, + size_t size, enum dma_data_direction dir) +{ + void *addr; + + addr = alloc_pages_exact(size, GFP_KERNEL | __GFP_ZERO); + if (!addr) + return NULL; + + *dma_addr = dma_map_single(&dev->pdev->dev, addr, size, dir); + if (dma_mapping_error(&dev->pdev->dev, *dma_addr)) { + ibdev_err(&dev->ibdev, "Failed to map DMA address\n"); + free_pages_exact(addr, size); + return NULL; + } + + return addr; +} + +/* + * This is only called when the ucontext is destroyed and there can be no + * concurrent query via mmap or allocate on the xarray, thus we can be sure no + * other thread is using the entry pointer. We also know that all the BAR + * pages have either been zap'd or munmaped at this point. Normal pages are + * refcounted and will be freed at the proper time. + */ +static void mmap_entries_remove_free(struct efa_dev *dev, + struct efa_ucontext *ucontext) +{ + struct efa_mmap_entry *entry; + unsigned long mmap_page; + + xa_for_each(&ucontext->mmap_xa, mmap_page, entry) { + xa_erase(&ucontext->mmap_xa, mmap_page); + + ibdev_dbg( + &dev->ibdev, + "mmap: obj[0x%p] key[%#llx] addr[%#llx] len[%#llx] removed\n", + entry->obj, get_mmap_key(entry), entry->address, + entry->length); + if (entry->mmap_flag == EFA_MMAP_DMA_PAGE) + /* DMA mapping is already gone, now free the pages */ + free_pages_exact(phys_to_virt(entry->address), + entry->length); + kfree(entry); + } +} + +static struct efa_mmap_entry *mmap_entry_get(struct efa_dev *dev, + struct efa_ucontext *ucontext, + u64 key, u64 len) +{ + struct efa_mmap_entry *entry; + u64 mmap_page; + + mmap_page = (key & EFA_MMAP_PAGE_MASK) >> PAGE_SHIFT; + if (mmap_page > U32_MAX) + return NULL; + + entry = xa_load(&ucontext->mmap_xa, mmap_page); + if (!entry || get_mmap_key(entry) != key || entry->length != len) + return NULL; + + ibdev_dbg(&dev->ibdev, + "mmap: obj[0x%p] key[%#llx] addr[%#llx] len[%#llx] removed\n", + entry->obj, key, entry->address, entry->length); + + return entry; +} + +/* + * Note this locking scheme cannot support removal of entries, except during + * ucontext destruction when the core code guarentees no concurrency. + */ +static u64 mmap_entry_insert(struct efa_dev *dev, struct efa_ucontext *ucontext, + void *obj, u64 address, u64 length, u8 mmap_flag) +{ + struct efa_mmap_entry *entry; + int err; + + entry = kmalloc(sizeof(*entry), GFP_KERNEL); + if (!entry) + return EFA_MMAP_INVALID; + + entry->obj = obj; + entry->address = address; + entry->length = length; + entry->mmap_flag = mmap_flag; + + xa_lock(&ucontext->mmap_xa); + entry->mmap_page = ucontext->mmap_xa_page; + ucontext->mmap_xa_page += DIV_ROUND_UP(length, PAGE_SIZE); + err = __xa_insert(&ucontext->mmap_xa, entry->mmap_page, entry, + GFP_KERNEL); + xa_unlock(&ucontext->mmap_xa); + if (err){ + kfree(entry); + return EFA_MMAP_INVALID; + } + + ibdev_dbg( + &dev->ibdev, + "mmap: obj[0x%p] addr[%#llx], len[%#llx], key[%#llx] inserted\n", + entry->obj, entry->address, entry->length, get_mmap_key(entry)); + + return get_mmap_key(entry); +} + +int efa_query_device(struct ib_device *ibdev, + struct ib_device_attr *props, + struct ib_udata *udata) +{ + struct efa_com_get_device_attr_result *dev_attr; + struct efa_ibv_ex_query_device_resp resp = {}; + struct efa_dev *dev = to_edev(ibdev); + int err; + + if (udata && udata->inlen && + !ib_is_udata_cleared(udata, 0, udata->inlen)) { + ibdev_dbg(ibdev, + "Incompatible ABI params, udata not cleared\n"); + return -EINVAL; + } + + dev_attr = &dev->dev_attr; + + memset(props, 0, sizeof(*props)); + props->max_mr_size = dev_attr->max_mr_pages * PAGE_SIZE; + props->page_size_cap = dev_attr->page_size_cap; + props->vendor_id = dev->pdev->vendor; + props->vendor_part_id = dev->pdev->device; + props->hw_ver = dev->pdev->subsystem_device; + props->max_qp = dev_attr->max_qp; + props->max_cq = dev_attr->max_cq; + props->max_pd = dev_attr->max_pd; + props->max_mr = dev_attr->max_mr; + props->max_ah = dev_attr->max_ah; + props->max_cqe = dev_attr->max_cq_depth; + props->max_qp_wr = min_t(u32, dev_attr->max_sq_depth, + dev_attr->max_rq_depth); + props->max_send_sge = dev_attr->max_sq_sge; + props->max_recv_sge = dev_attr->max_rq_sge; + + if (udata && udata->outlen) { + resp.max_sq_sge = dev_attr->max_sq_sge; + resp.max_rq_sge = dev_attr->max_rq_sge; + resp.max_sq_wr = dev_attr->max_sq_depth; + resp.max_rq_wr = dev_attr->max_rq_depth; + + err = ib_copy_to_udata(udata, &resp, + min(sizeof(resp), udata->outlen)); + if (err) { + ibdev_dbg(ibdev, + "Failed to copy udata for query_device\n"); + return err; + } + } + + return 0; +} + +int efa_query_port(struct ib_device *ibdev, u8 port, + struct ib_port_attr *props) +{ + struct efa_dev *dev = to_edev(ibdev); + + props->lmc = 1; + + props->state = IB_PORT_ACTIVE; + props->phys_state = 5; + props->gid_tbl_len = 1; + props->pkey_tbl_len = 1; + props->active_speed = IB_SPEED_EDR; + props->active_width = IB_WIDTH_4X; + props->max_mtu = ib_mtu_int_to_enum(dev->mtu); + props->active_mtu = ib_mtu_int_to_enum(dev->mtu); + props->max_msg_sz = dev->mtu; + props->max_vl_num = 1; + + return 0; +} + +int efa_query_qp(struct ib_qp *ibqp, struct ib_qp_attr *qp_attr, + int qp_attr_mask, + struct ib_qp_init_attr *qp_init_attr) +{ + struct efa_dev *dev = to_edev(ibqp->device); + struct efa_com_query_qp_params params = {}; + struct efa_com_query_qp_result result; + struct efa_qp *qp = to_eqp(ibqp); + int err; + +#define EFA_QUERY_QP_SUPP_MASK \ + (IB_QP_STATE | IB_QP_PKEY_INDEX | IB_QP_PORT | \ + IB_QP_QKEY | IB_QP_SQ_PSN | IB_QP_CAP) + + if (qp_attr_mask & ~EFA_QUERY_QP_SUPP_MASK) { + ibdev_dbg(&dev->ibdev, + "Unsupported qp_attr_mask[%#x] supported[%#x]\n", + qp_attr_mask, EFA_QUERY_QP_SUPP_MASK); + return -EOPNOTSUPP; + } + + memset(qp_attr, 0, sizeof(*qp_attr)); + memset(qp_init_attr, 0, sizeof(*qp_init_attr)); + + params.qp_handle = qp->qp_handle; + err = efa_com_query_qp(&dev->edev, ¶ms, &result); + if (err) + return err; + + qp_attr->qp_state = result.qp_state; + qp_attr->qkey = result.qkey; + qp_attr->sq_psn = result.sq_psn; + qp_attr->sq_draining = result.sq_draining; + qp_attr->port_num = 1; + + qp_attr->cap.max_send_wr = qp->max_send_wr; + qp_attr->cap.max_recv_wr = qp->max_recv_wr; + qp_attr->cap.max_send_sge = qp->max_send_sge; + qp_attr->cap.max_recv_sge = qp->max_recv_sge; + qp_attr->cap.max_inline_data = qp->max_inline_data; + + qp_init_attr->qp_type = ibqp->qp_type; + qp_init_attr->recv_cq = ibqp->recv_cq; + qp_init_attr->send_cq = ibqp->send_cq; + qp_init_attr->qp_context = ibqp->qp_context; + qp_init_attr->cap = qp_attr->cap; + + return 0; +} + +int efa_query_gid(struct ib_device *ibdev, u8 port, int index, + union ib_gid *gid) +{ + struct efa_dev *dev = to_edev(ibdev); + + memcpy(gid->raw, dev->addr, sizeof(dev->addr)); + + return 0; +} + +int efa_query_pkey(struct ib_device *ibdev, u8 port, u16 index, + u16 *pkey) +{ + if (index > 0) + return -EINVAL; + + *pkey = 0xffff; + return 0; +} + +static int efa_pd_dealloc(struct efa_dev *dev, u16 pdn) +{ + struct efa_com_dealloc_pd_params params = { + .pdn = pdn, + }; + + return efa_com_dealloc_pd(&dev->edev, ¶ms); +} + +int efa_alloc_pd(struct ib_pd *ibpd, struct ib_udata *udata) +{ + struct efa_dev *dev = to_edev(ibpd->device); + struct efa_ibv_alloc_pd_resp resp = {}; + struct efa_com_alloc_pd_result result; + struct efa_pd *pd = to_epd(ibpd); + int err; + + if (udata->inlen && + !ib_is_udata_cleared(udata, 0, udata->inlen)) { + ibdev_dbg(&dev->ibdev, + "Incompatible ABI params, udata not cleared\n"); + err = -EINVAL; + goto err_out; + } + + err = efa_com_alloc_pd(&dev->edev, &result); + if (err) + goto err_out; + + pd->pdn = result.pdn; + resp.pdn = result.pdn; + + if (udata->outlen) { + err = ib_copy_to_udata(udata, &resp, + min(sizeof(resp), udata->outlen)); + if (err) { + ibdev_dbg(&dev->ibdev, + "Failed to copy udata for alloc_pd\n"); + goto err_dealloc_pd; + } + } + + ibdev_dbg(&dev->ibdev, "Allocated pd[%d]\n", pd->pdn); + + return 0; + +err_dealloc_pd: + efa_pd_dealloc(dev, result.pdn); +err_out: + atomic64_inc(&dev->stats.sw_stats.alloc_pd_err); + return err; +} + +void efa_dealloc_pd(struct ib_pd *ibpd, struct ib_udata *udata) +{ + struct efa_dev *dev = to_edev(ibpd->device); + struct efa_pd *pd = to_epd(ibpd); + + if (udata->inlen && + !ib_is_udata_cleared(udata, 0, udata->inlen)) { + ibdev_dbg(&dev->ibdev, "Incompatible ABI params\n"); + return; + } + + ibdev_dbg(&dev->ibdev, "Dealloc pd[%d]\n", pd->pdn); + efa_pd_dealloc(dev, pd->pdn); +} + +static int efa_destroy_qp_handle(struct efa_dev *dev, u32 qp_handle) +{ + struct efa_com_destroy_qp_params params = { .qp_handle = qp_handle }; + + return efa_com_destroy_qp(&dev->edev, ¶ms); +} + +int efa_destroy_qp(struct ib_qp *ibqp, struct ib_udata *udata) +{ + struct efa_dev *dev = to_edev(ibqp->pd->device); + struct efa_qp *qp = to_eqp(ibqp); + int err; + + if (udata->inlen && + !ib_is_udata_cleared(udata, 0, udata->inlen)) { + ibdev_dbg(&dev->ibdev, "Incompatible ABI params\n"); + return -EINVAL; + } + + ibdev_dbg(&dev->ibdev, "Destroy qp[%u]\n", ibqp->qp_num); + err = efa_destroy_qp_handle(dev, qp->qp_handle); + if (err) + return err; + + if (qp->rq_cpu_addr) { + ibdev_dbg(&dev->ibdev, + "qp->cpu_addr[0x%p] freed: size[%lu], dma[%pad]\n", + qp->rq_cpu_addr, qp->rq_size, + &qp->rq_dma_addr); + dma_unmap_single(&dev->pdev->dev, qp->rq_dma_addr, qp->rq_size, + DMA_TO_DEVICE); + } + + kfree(qp); + return 0; +} + +static int qp_mmap_entries_setup(struct efa_qp *qp, + struct efa_dev *dev, + struct efa_ucontext *ucontext, + struct efa_com_create_qp_params *params, + struct efa_ibv_create_qp_resp *resp) +{ + /* + * Once an entry is inserted it might be mmapped, hence cannot be + * cleaned up until dealloc_ucontext. + */ + resp->sq_db_mmap_key = + mmap_entry_insert(dev, ucontext, qp, + dev->db_bar_addr + resp->sq_db_offset, + PAGE_SIZE, EFA_MMAP_IO_NC); + if (resp->sq_db_mmap_key == EFA_MMAP_INVALID) + return -ENOMEM; + + resp->sq_db_offset &= ~PAGE_MASK; + + resp->llq_desc_mmap_key = + mmap_entry_insert(dev, ucontext, qp, + dev->mem_bar_addr + resp->llq_desc_offset, + PAGE_ALIGN(params->sq_ring_size_in_bytes + + (resp->llq_desc_offset & ~PAGE_MASK)), + EFA_MMAP_IO_WC); + if (resp->llq_desc_mmap_key == EFA_MMAP_INVALID) + return -ENOMEM; + + resp->llq_desc_offset &= ~PAGE_MASK; + + if (qp->rq_size) { + resp->rq_db_mmap_key = + mmap_entry_insert(dev, ucontext, qp, + dev->db_bar_addr + resp->rq_db_offset, + PAGE_SIZE, EFA_MMAP_IO_NC); + if (resp->rq_db_mmap_key == EFA_MMAP_INVALID) + return -ENOMEM; + + resp->rq_db_offset &= ~PAGE_MASK; + + resp->rq_mmap_key = + mmap_entry_insert(dev, ucontext, qp, + virt_to_phys(qp->rq_cpu_addr), + qp->rq_size, EFA_MMAP_DMA_PAGE); + if (resp->rq_mmap_key == EFA_MMAP_INVALID) + return -ENOMEM; + + resp->rq_mmap_size = qp->rq_size; + } + + return 0; +} + +static int efa_qp_validate_cap(struct efa_dev *dev, + struct ib_qp_init_attr *init_attr) +{ + if (init_attr->cap.max_send_wr > dev->dev_attr.max_sq_depth) { + ibdev_dbg(&dev->ibdev, + "qp: requested send wr[%u] exceeds the max[%u]\n", + init_attr->cap.max_send_wr, + dev->dev_attr.max_sq_depth); + return -EINVAL; + } + if (init_attr->cap.max_recv_wr > dev->dev_attr.max_rq_depth) { + ibdev_dbg(&dev->ibdev, + "qp: requested receive wr[%u] exceeds the max[%u]\n", + init_attr->cap.max_recv_wr, + dev->dev_attr.max_rq_depth); + return -EINVAL; + } + if (init_attr->cap.max_send_sge > dev->dev_attr.max_sq_sge) { + ibdev_dbg(&dev->ibdev, + "qp: requested sge send[%u] exceeds the max[%u]\n", + init_attr->cap.max_send_sge, dev->dev_attr.max_sq_sge); + return -EINVAL; + } + if (init_attr->cap.max_recv_sge > dev->dev_attr.max_rq_sge) { + ibdev_dbg(&dev->ibdev, + "qp: requested sge recv[%u] exceeds the max[%u]\n", + init_attr->cap.max_recv_sge, dev->dev_attr.max_rq_sge); + return -EINVAL; + } + if (init_attr->cap.max_inline_data > dev->dev_attr.inline_buf_size) { + ibdev_dbg(&dev->ibdev, + "qp: requested inline data[%u] exceeds the max[%u]\n", + init_attr->cap.max_inline_data, + dev->dev_attr.inline_buf_size); + return -EINVAL; + } + + return 0; +} + +static int efa_qp_validate_attr(struct efa_dev *dev, + struct ib_qp_init_attr *init_attr) +{ + if (init_attr->qp_type != IB_QPT_DRIVER && + init_attr->qp_type != IB_QPT_UD) { + ibdev_dbg(&dev->ibdev, + "Unsupported qp type %d\n", init_attr->qp_type); + return -EOPNOTSUPP; + } + + if (init_attr->srq) { + ibdev_dbg(&dev->ibdev, "SRQ is not supported\n"); + return -EOPNOTSUPP; + } + + if (init_attr->create_flags) { + ibdev_dbg(&dev->ibdev, "Unsupported create flags\n"); + return -EOPNOTSUPP; + } + + return 0; +} + +struct ib_qp *efa_create_qp(struct ib_pd *ibpd, + struct ib_qp_init_attr *init_attr, + struct ib_udata *udata) +{ + struct efa_com_create_qp_params create_qp_params = {}; + struct efa_com_create_qp_result create_qp_resp; + struct efa_dev *dev = to_edev(ibpd->device); + struct efa_ibv_create_qp_resp resp = {}; + struct efa_ibv_create_qp cmd = {}; + bool rq_entry_inserted = false; + struct efa_ucontext *ucontext; + struct efa_qp *qp; + int err; + + ucontext = rdma_udata_to_drv_context(udata, struct efa_ucontext, + ibucontext); + + err = efa_qp_validate_cap(dev, init_attr); + if (err) + goto err_out; + + err = efa_qp_validate_attr(dev, init_attr); + if (err) + goto err_out; + + if (!field_avail(cmd, driver_qp_type, udata->inlen)) { + ibdev_dbg(&dev->ibdev, + "Incompatible ABI params, no input udata\n"); + err = -EINVAL; + goto err_out; + } + + if (udata->inlen > sizeof(cmd) && + !ib_is_udata_cleared(udata, sizeof(cmd), + udata->inlen - sizeof(cmd))) { + ibdev_dbg(&dev->ibdev, + "Incompatible ABI params, unknown fields in udata\n"); + err = -EINVAL; + goto err_out; + } + + err = ib_copy_from_udata(&cmd, udata, + min(sizeof(cmd), udata->inlen)); + if (err) { + ibdev_dbg(&dev->ibdev, + "Cannot copy udata for create_qp\n"); + goto err_out; + } + + if (cmd.comp_mask) { + ibdev_dbg(&dev->ibdev, + "Incompatible ABI params, unknown fields in udata\n"); + err = -EINVAL; + goto err_out; + } + + qp = kzalloc(sizeof(*qp), GFP_KERNEL); + if (!qp) { + err = -ENOMEM; + goto err_out; + } + + create_qp_params.uarn = ucontext->uarn; + create_qp_params.pd = to_epd(ibpd)->pdn; + + if (init_attr->qp_type == IB_QPT_UD) { + create_qp_params.qp_type = EFA_ADMIN_QP_TYPE_UD; + } else if (cmd.driver_qp_type == EFA_QP_DRIVER_TYPE_SRD) { + create_qp_params.qp_type = EFA_ADMIN_QP_TYPE_SRD; + } else { + ibdev_dbg(&dev->ibdev, + "Unsupported qp type %d driver qp type %d\n", + init_attr->qp_type, cmd.driver_qp_type); + err = -EOPNOTSUPP; + goto err_free_qp; + } + + ibdev_dbg(&dev->ibdev, "Create QP: qp type %d driver qp type %#x\n", + init_attr->qp_type, cmd.driver_qp_type); + create_qp_params.send_cq_idx = to_ecq(init_attr->send_cq)->cq_idx; + create_qp_params.recv_cq_idx = to_ecq(init_attr->recv_cq)->cq_idx; + create_qp_params.sq_depth = init_attr->cap.max_send_wr; + create_qp_params.sq_ring_size_in_bytes = cmd.sq_ring_size; + + create_qp_params.rq_depth = init_attr->cap.max_recv_wr; + create_qp_params.rq_ring_size_in_bytes = cmd.rq_ring_size; + qp->rq_size = PAGE_ALIGN(create_qp_params.rq_ring_size_in_bytes); + if (qp->rq_size) { + qp->rq_cpu_addr = efa_zalloc_mapped(dev, &qp->rq_dma_addr, + qp->rq_size, DMA_TO_DEVICE); + if (!qp->rq_cpu_addr) { + err = -ENOMEM; + goto err_free_qp; + } + + ibdev_dbg(&dev->ibdev, + "qp->cpu_addr[0x%p] allocated: size[%lu], dma[%pad]\n", + qp->rq_cpu_addr, qp->rq_size, &qp->rq_dma_addr); + create_qp_params.rq_base_addr = qp->rq_dma_addr; + } + + err = efa_com_create_qp(&dev->edev, &create_qp_params, + &create_qp_resp); + if (err) + goto err_free_mapped; + + resp.sq_db_offset = create_qp_resp.sq_db_offset; + resp.rq_db_offset = create_qp_resp.rq_db_offset; + resp.llq_desc_offset = create_qp_resp.llq_descriptors_offset; + resp.send_sub_cq_idx = create_qp_resp.send_sub_cq_idx; + resp.recv_sub_cq_idx = create_qp_resp.recv_sub_cq_idx; + + err = qp_mmap_entries_setup(qp, dev, ucontext, &create_qp_params, + &resp); + if (err) + goto err_destroy_qp; + + rq_entry_inserted = true; + qp->qp_handle = create_qp_resp.qp_handle; + qp->ibqp.qp_num = create_qp_resp.qp_num; + qp->ibqp.qp_type = init_attr->qp_type; + qp->max_send_wr = init_attr->cap.max_send_wr; + qp->max_recv_wr = init_attr->cap.max_recv_wr; + qp->max_send_sge = init_attr->cap.max_send_sge; + qp->max_recv_sge = init_attr->cap.max_recv_sge; + qp->max_inline_data = init_attr->cap.max_inline_data; + + if (udata->outlen) { + err = ib_copy_to_udata(udata, &resp, + min(sizeof(resp), udata->outlen)); + if (err) { + ibdev_dbg(&dev->ibdev, + "Failed to copy udata for qp[%u]\n", + create_qp_resp.qp_num); + goto err_destroy_qp; + } + } + + ibdev_dbg(&dev->ibdev, "Created qp[%d]\n", qp->ibqp.qp_num); + + return &qp->ibqp; + +err_destroy_qp: + efa_destroy_qp_handle(dev, create_qp_resp.qp_handle); +err_free_mapped: + if (qp->rq_size) { + dma_unmap_single(&dev->pdev->dev, qp->rq_dma_addr, qp->rq_size, + DMA_TO_DEVICE); + if (!rq_entry_inserted) + free_pages_exact(qp->rq_cpu_addr, qp->rq_size); + } +err_free_qp: + kfree(qp); +err_out: + atomic64_inc(&dev->stats.sw_stats.create_qp_err); + return ERR_PTR(err); +} + +static int efa_modify_qp_validate(struct efa_dev *dev, struct efa_qp *qp, + struct ib_qp_attr *qp_attr, int qp_attr_mask, + enum ib_qp_state cur_state, + enum ib_qp_state new_state) +{ +#define EFA_MODIFY_QP_SUPP_MASK \ + (IB_QP_STATE | IB_QP_CUR_STATE | IB_QP_EN_SQD_ASYNC_NOTIFY | \ + IB_QP_PKEY_INDEX | IB_QP_PORT | IB_QP_QKEY | IB_QP_SQ_PSN) + + if (qp_attr_mask & ~EFA_MODIFY_QP_SUPP_MASK) { + ibdev_dbg(&dev->ibdev, + "Unsupported qp_attr_mask[%#x] supported[%#x]\n", + qp_attr_mask, EFA_MODIFY_QP_SUPP_MASK); + return -EOPNOTSUPP; + } + + if (!ib_modify_qp_is_ok(cur_state, new_state, IB_QPT_UD, + qp_attr_mask)) { + ibdev_dbg(&dev->ibdev, "Invalid modify QP parameters\n"); + return -EINVAL; + } + + if ((qp_attr_mask & IB_QP_PORT) && qp_attr->port_num != 1) { + ibdev_dbg(&dev->ibdev, "Can't change port num\n"); + return -EOPNOTSUPP; + } + + if ((qp_attr_mask & IB_QP_PKEY_INDEX) && qp_attr->pkey_index) { + ibdev_dbg(&dev->ibdev, "Can't change pkey index\n"); + return -EOPNOTSUPP; + } + + return 0; +} + +int efa_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *qp_attr, + int qp_attr_mask, struct ib_udata *udata) +{ + struct efa_dev *dev = to_edev(ibqp->device); + struct efa_com_modify_qp_params params = {}; + struct efa_qp *qp = to_eqp(ibqp); + enum ib_qp_state cur_state; + enum ib_qp_state new_state; + int err; + + if (udata->inlen && + !ib_is_udata_cleared(udata, 0, udata->inlen)) { + ibdev_dbg(&dev->ibdev, + "Incompatible ABI params, udata not cleared\n"); + return -EINVAL; + } + + cur_state = qp_attr_mask & IB_QP_CUR_STATE ? qp_attr->cur_qp_state : + qp->state; + new_state = qp_attr_mask & IB_QP_STATE ? qp_attr->qp_state : cur_state; + + err = efa_modify_qp_validate(dev, qp, qp_attr, qp_attr_mask, cur_state, + new_state); + if (err) + return err; + + params.qp_handle = qp->qp_handle; + + if (qp_attr_mask & IB_QP_STATE) { + params.modify_mask |= BIT(EFA_ADMIN_QP_STATE_BIT) | + BIT(EFA_ADMIN_CUR_QP_STATE_BIT); + params.cur_qp_state = qp_attr->cur_qp_state; + params.qp_state = qp_attr->qp_state; + } + + if (qp_attr_mask & IB_QP_EN_SQD_ASYNC_NOTIFY) { + params.modify_mask |= + BIT(EFA_ADMIN_SQ_DRAINED_ASYNC_NOTIFY_BIT); + params.sq_drained_async_notify = qp_attr->en_sqd_async_notify; + } + + if (qp_attr_mask & IB_QP_QKEY) { + params.modify_mask |= BIT(EFA_ADMIN_QKEY_BIT); + params.qkey = qp_attr->qkey; + } + + if (qp_attr_mask & IB_QP_SQ_PSN) { + params.modify_mask |= BIT(EFA_ADMIN_SQ_PSN_BIT); + params.sq_psn = qp_attr->sq_psn; + } + + err = efa_com_modify_qp(&dev->edev, ¶ms); + if (err) + return err; + + qp->state = new_state; + + return 0; +} + +static int efa_destroy_cq_idx(struct efa_dev *dev, int cq_idx) +{ + struct efa_com_destroy_cq_params params = { .cq_idx = cq_idx }; + + return efa_com_destroy_cq(&dev->edev, ¶ms); +} + +int efa_destroy_cq(struct ib_cq *ibcq, struct ib_udata *udata) +{ + struct efa_dev *dev = to_edev(ibcq->device); + struct efa_cq *cq = to_ecq(ibcq); + int err; + + if (udata->inlen && + !ib_is_udata_cleared(udata, 0, udata->inlen)) { + ibdev_dbg(&dev->ibdev, "Incompatible ABI params\n"); + return -EINVAL; + } + + ibdev_dbg(&dev->ibdev, + "Destroy cq[%d] virt[0x%p] freed: size[%lu], dma[%pad]\n", + cq->cq_idx, cq->cpu_addr, cq->size, &cq->dma_addr); + + err = efa_destroy_cq_idx(dev, cq->cq_idx); + if (err) + return err; + + dma_unmap_single(&dev->pdev->dev, cq->dma_addr, cq->size, + DMA_FROM_DEVICE); + + kfree(cq); + return 0; +} + +static int cq_mmap_entries_setup(struct efa_dev *dev, struct efa_cq *cq, + struct efa_ibv_create_cq_resp *resp) +{ + resp->q_mmap_size = cq->size; + resp->q_mmap_key = mmap_entry_insert(dev, cq->ucontext, cq, + virt_to_phys(cq->cpu_addr), + cq->size, EFA_MMAP_DMA_PAGE); + if (resp->q_mmap_key == EFA_MMAP_INVALID) + return -ENOMEM; + + return 0; +} + +static struct ib_cq *do_create_cq(struct ib_device *ibdev, int entries, + int vector, struct ib_ucontext *ibucontext, + struct ib_udata *udata) +{ + struct efa_ibv_create_cq_resp resp = {}; + struct efa_com_create_cq_params params; + struct efa_com_create_cq_result result; + struct efa_dev *dev = to_edev(ibdev); + struct efa_ibv_create_cq cmd = {}; + bool cq_entry_inserted = false; + struct efa_cq *cq; + int err; + + ibdev_dbg(ibdev, "create_cq entries %d\n", entries); + + if (entries < 1 || entries > dev->dev_attr.max_cq_depth) { + ibdev_dbg(ibdev, + "cq: requested entries[%u] non-positive or greater than max[%u]\n", + entries, dev->dev_attr.max_cq_depth); + err = -EINVAL; + goto err_out; + } + + if (!field_avail(cmd, num_sub_cqs, udata->inlen)) { + ibdev_dbg(ibdev, + "Incompatible ABI params, no input udata\n"); + err = -EINVAL; + goto err_out; + } + + if (udata->inlen > sizeof(cmd) && + !ib_is_udata_cleared(udata, sizeof(cmd), + udata->inlen - sizeof(cmd))) { + ibdev_dbg(ibdev, + "Incompatible ABI params, unknown fields in udata\n"); + err = -EINVAL; + goto err_out; + } + + err = ib_copy_from_udata(&cmd, udata, + min(sizeof(cmd), udata->inlen)); + if (err) { + ibdev_dbg(ibdev, "Cannot copy udata for create_cq\n"); + goto err_out; + } + + if (cmd.comp_mask || !is_reserved_cleared(cmd.reserved_50)) { + ibdev_dbg(ibdev, + "Incompatible ABI params, unknown fields in udata\n"); + err = -EINVAL; + goto err_out; + } + + if (!cmd.cq_entry_size) { + ibdev_dbg(ibdev, + "Invalid entry size [%u]\n", cmd.cq_entry_size); + err = -EINVAL; + goto err_out; + } + + if (cmd.num_sub_cqs != dev->dev_attr.sub_cqs_per_cq) { + ibdev_dbg(ibdev, + "Invalid number of sub cqs[%u] expected[%u]\n", + cmd.num_sub_cqs, dev->dev_attr.sub_cqs_per_cq); + err = -EINVAL; + goto err_out; + } + + cq = kzalloc(sizeof(*cq), GFP_KERNEL); + if (!cq) { + err = -ENOMEM; + goto err_out; + } + + cq->ucontext = to_eucontext(ibucontext); + cq->size = PAGE_ALIGN(cmd.cq_entry_size * entries * cmd.num_sub_cqs); + cq->cpu_addr = efa_zalloc_mapped(dev, &cq->dma_addr, cq->size, + DMA_FROM_DEVICE); + if (!cq->cpu_addr) { + err = -ENOMEM; + goto err_free_cq; + } + + params.uarn = cq->ucontext->uarn; + params.cq_depth = entries; + params.dma_addr = cq->dma_addr; + params.entry_size_in_bytes = cmd.cq_entry_size; + params.num_sub_cqs = cmd.num_sub_cqs; + err = efa_com_create_cq(&dev->edev, ¶ms, &result); + if (err) + goto err_free_mapped; + + resp.cq_idx = result.cq_idx; + cq->cq_idx = result.cq_idx; + cq->ibcq.cqe = result.actual_depth; + WARN_ON_ONCE(entries != result.actual_depth); + + err = cq_mmap_entries_setup(dev, cq, &resp); + if (err) { + ibdev_dbg(ibdev, + "Could not setup cq[%u] mmap entries\n", cq->cq_idx); + goto err_destroy_cq; + } + + cq_entry_inserted = true; + + if (udata->outlen) { + err = ib_copy_to_udata(udata, &resp, + min(sizeof(resp), udata->outlen)); + if (err) { + ibdev_dbg(ibdev, + "Failed to copy udata for create_cq\n"); + goto err_destroy_cq; + } + } + + ibdev_dbg(ibdev, + "Created cq[%d], cq depth[%u]. dma[%pad] virt[0x%p]\n", + cq->cq_idx, result.actual_depth, &cq->dma_addr, cq->cpu_addr); + + return &cq->ibcq; + +err_destroy_cq: + efa_destroy_cq_idx(dev, cq->cq_idx); +err_free_mapped: + dma_unmap_single(&dev->pdev->dev, cq->dma_addr, cq->size, + DMA_FROM_DEVICE); + if (!cq_entry_inserted) + free_pages_exact(cq->cpu_addr, cq->size); +err_free_cq: + kfree(cq); +err_out: + atomic64_inc(&dev->stats.sw_stats.create_cq_err); + return ERR_PTR(err); +} + +struct ib_cq *efa_create_cq(struct ib_device *ibdev, + const struct ib_cq_init_attr *attr, + struct ib_udata *udata) +{ + struct efa_ucontext *ucontext = rdma_udata_to_drv_context(udata, + struct efa_ucontext, + ibucontext); + + return do_create_cq(ibdev, attr->cqe, attr->comp_vector, + &ucontext->ibucontext, udata); +} + +static int umem_to_page_list(struct efa_dev *dev, + struct ib_umem *umem, + u64 *page_list, + u32 hp_cnt, + u8 hp_shift) +{ + u32 pages_in_hp = BIT(hp_shift - PAGE_SHIFT); + struct sg_dma_page_iter sg_iter; + unsigned int page_idx = 0; + unsigned int hp_idx = 0; + + ibdev_dbg(&dev->ibdev, "hp_cnt[%u], pages_in_hp[%u]\n", + hp_cnt, pages_in_hp); + + for_each_sg_dma_page(umem->sg_head.sgl, &sg_iter, umem->nmap, 0) { + if (page_idx % pages_in_hp == 0) { + page_list[hp_idx] = sg_page_iter_dma_address(&sg_iter); + hp_idx++; + } + + page_idx++; + } + + return 0; +} + +static struct scatterlist *efa_vmalloc_buf_to_sg(u64 *buf, int page_cnt) +{ + struct scatterlist *sglist; + struct page *pg; + int i; + + sglist = kcalloc(page_cnt, sizeof(*sglist), GFP_KERNEL); + if (!sglist) + return NULL; + sg_init_table(sglist, page_cnt); + for (i = 0; i < page_cnt; i++) { + pg = vmalloc_to_page(buf); + if (!pg) + goto err; + sg_set_page(&sglist[i], pg, PAGE_SIZE, 0); + buf += PAGE_SIZE / sizeof(*buf); + } + return sglist; + +err: + kfree(sglist); + return NULL; +} + +/* + * create a chunk list of physical pages dma addresses from the supplied + * scatter gather list + */ +static int pbl_chunk_list_create(struct efa_dev *dev, struct pbl_context *pbl) +{ + unsigned int entry, payloads_in_sg, chunk_list_size, chunk_idx, payload_idx; + struct pbl_chunk_list *chunk_list = &pbl->phys.indirect.chunk_list; + int page_cnt = pbl->phys.indirect.pbl_buf_size_in_pages; + struct scatterlist *pages_sgl = pbl->phys.indirect.sgl; + int sg_dma_cnt = pbl->phys.indirect.sg_dma_cnt; + struct efa_com_ctrl_buff_info *ctrl_buf; + u64 *cur_chunk_buf, *prev_chunk_buf; + struct scatterlist *sg; + dma_addr_t dma_addr; + int i; + + /* allocate a chunk list that consists of 4KB chunks */ + chunk_list_size = DIV_ROUND_UP(page_cnt, EFA_PTRS_PER_CHUNK); + + chunk_list->size = chunk_list_size; + chunk_list->chunks = kcalloc(chunk_list_size, + sizeof(*chunk_list->chunks), + GFP_KERNEL); + if (!chunk_list->chunks) + return -ENOMEM; + + ibdev_dbg(&dev->ibdev, + "chunk_list_size[%u] - pages[%u]\n", chunk_list_size, + page_cnt); + + /* allocate chunk buffers: */ + for (i = 0; i < chunk_list_size; i++) { + chunk_list->chunks[i].buf = kzalloc(EFA_CHUNK_SIZE, GFP_KERNEL); + if (!chunk_list->chunks[i].buf) + goto chunk_list_dealloc; + + chunk_list->chunks[i].length = EFA_CHUNK_USED_SIZE; + } + chunk_list->chunks[chunk_list_size - 1].length = + ((page_cnt % EFA_PTRS_PER_CHUNK) * EFA_CHUNK_PAYLOAD_PTR_SIZE) + + EFA_CHUNK_PTR_SIZE; + + /* fill the dma addresses of sg list pages to chunks: */ + chunk_idx = 0; + payload_idx = 0; + cur_chunk_buf = chunk_list->chunks[0].buf; + for_each_sg(pages_sgl, sg, sg_dma_cnt, entry) { + payloads_in_sg = sg_dma_len(sg) >> EFA_CHUNK_PAYLOAD_SHIFT; + for (i = 0; i < payloads_in_sg; i++) { + cur_chunk_buf[payload_idx++] = + (sg_dma_address(sg) & ~(EFA_CHUNK_PAYLOAD_SIZE - 1)) + + (EFA_CHUNK_PAYLOAD_SIZE * i); + + if (payload_idx == EFA_PTRS_PER_CHUNK) { + chunk_idx++; + cur_chunk_buf = chunk_list->chunks[chunk_idx].buf; + payload_idx = 0; + } + } + } + + /* map chunks to dma and fill chunks next ptrs */ + for (i = chunk_list_size - 1; i >= 0; i--) { + dma_addr = dma_map_single(&dev->pdev->dev, + chunk_list->chunks[i].buf, + chunk_list->chunks[i].length, + DMA_TO_DEVICE); + if (dma_mapping_error(&dev->pdev->dev, dma_addr)) { + ibdev_err(&dev->ibdev, + "chunk[%u] dma_map_failed\n", i); + goto chunk_list_unmap; + } + + chunk_list->chunks[i].dma_addr = dma_addr; + ibdev_dbg(&dev->ibdev, + "chunk[%u] mapped at [%pad]\n", i, &dma_addr); + + if (!i) + break; + + prev_chunk_buf = chunk_list->chunks[i - 1].buf; + + ctrl_buf = (struct efa_com_ctrl_buff_info *) + &prev_chunk_buf[EFA_PTRS_PER_CHUNK]; + ctrl_buf->length = chunk_list->chunks[i].length; + + efa_com_set_dma_addr(dma_addr, + &ctrl_buf->address.mem_addr_high, + &ctrl_buf->address.mem_addr_low); + } + + return 0; + +chunk_list_unmap: + for (; i < chunk_list_size; i++) { + dma_unmap_single(&dev->pdev->dev, chunk_list->chunks[i].dma_addr, + chunk_list->chunks[i].length, DMA_TO_DEVICE); + } +chunk_list_dealloc: + for (i = 0; i < chunk_list_size; i++) + kfree(chunk_list->chunks[i].buf); + + kfree(chunk_list->chunks); + return -ENOMEM; +} + +static void pbl_chunk_list_destroy(struct efa_dev *dev, struct pbl_context *pbl) +{ + struct pbl_chunk_list *chunk_list = &pbl->phys.indirect.chunk_list; + int i; + + for (i = 0; i < chunk_list->size; i++) { + dma_unmap_single(&dev->pdev->dev, chunk_list->chunks[i].dma_addr, + chunk_list->chunks[i].length, DMA_TO_DEVICE); + kfree(chunk_list->chunks[i].buf); + } + + kfree(chunk_list->chunks); +} + +/* initialize pbl continuous mode: map pbl buffer to a dma address. */ +static int pbl_continuous_initialize(struct efa_dev *dev, + struct pbl_context *pbl) +{ + dma_addr_t dma_addr; + + dma_addr = dma_map_single(&dev->pdev->dev, pbl->pbl_buf, + pbl->pbl_buf_size_in_bytes, DMA_TO_DEVICE); + if (dma_mapping_error(&dev->pdev->dev, dma_addr)) { + ibdev_err(&dev->ibdev, "Unable to map pbl to DMA address\n"); + return -ENOMEM; + } + + pbl->phys.continuous.dma_addr = dma_addr; + ibdev_dbg(&dev->ibdev, + "pbl continuous - dma_addr = %pad, size[%u]\n", + &dma_addr, pbl->pbl_buf_size_in_bytes); + + return 0; +} + +/* + * initialize pbl indirect mode: + * create a chunk list out of the dma addresses of the physical pages of + * pbl buffer. + */ +static int pbl_indirect_initialize(struct efa_dev *dev, struct pbl_context *pbl) +{ + u32 size_in_pages = DIV_ROUND_UP(pbl->pbl_buf_size_in_bytes, PAGE_SIZE); + struct scatterlist *sgl; + int sg_dma_cnt, err; + + BUILD_BUG_ON(EFA_CHUNK_PAYLOAD_SIZE > PAGE_SIZE); + sgl = efa_vmalloc_buf_to_sg(pbl->pbl_buf, size_in_pages); + if (!sgl) + return -ENOMEM; + + sg_dma_cnt = dma_map_sg(&dev->pdev->dev, sgl, size_in_pages, DMA_TO_DEVICE); + if (!sg_dma_cnt) { + err = -EINVAL; + goto err_map; + } + + pbl->phys.indirect.pbl_buf_size_in_pages = size_in_pages; + pbl->phys.indirect.sgl = sgl; + pbl->phys.indirect.sg_dma_cnt = sg_dma_cnt; + err = pbl_chunk_list_create(dev, pbl); + if (err) { + ibdev_dbg(&dev->ibdev, + "chunk_list creation failed[%d]\n", err); + goto err_chunk; + } + + ibdev_dbg(&dev->ibdev, + "pbl indirect - size[%u], chunks[%u]\n", + pbl->pbl_buf_size_in_bytes, + pbl->phys.indirect.chunk_list.size); + + return 0; + +err_chunk: + dma_unmap_sg(&dev->pdev->dev, sgl, size_in_pages, DMA_TO_DEVICE); +err_map: + kfree(sgl); + return err; +} + +static void pbl_indirect_terminate(struct efa_dev *dev, struct pbl_context *pbl) +{ + pbl_chunk_list_destroy(dev, pbl); + dma_unmap_sg(&dev->pdev->dev, pbl->phys.indirect.sgl, + pbl->phys.indirect.pbl_buf_size_in_pages, DMA_TO_DEVICE); + kfree(pbl->phys.indirect.sgl); +} + +/* create a page buffer list from a mapped user memory region */ +static int pbl_create(struct efa_dev *dev, + struct pbl_context *pbl, + struct ib_umem *umem, + int hp_cnt, + u8 hp_shift) +{ + int err; + + pbl->pbl_buf_size_in_bytes = hp_cnt * EFA_CHUNK_PAYLOAD_PTR_SIZE; + pbl->pbl_buf = kzalloc(pbl->pbl_buf_size_in_bytes, + GFP_KERNEL | __GFP_NOWARN); + if (pbl->pbl_buf) { + pbl->physically_continuous = 1; + err = umem_to_page_list(dev, umem, pbl->pbl_buf, hp_cnt, + hp_shift); + if (err) + goto err_continuous; + err = pbl_continuous_initialize(dev, pbl); + if (err) + goto err_continuous; + } else { + pbl->physically_continuous = 0; + pbl->pbl_buf = vzalloc(pbl->pbl_buf_size_in_bytes); + if (!pbl->pbl_buf) + return -ENOMEM; + + err = umem_to_page_list(dev, umem, pbl->pbl_buf, hp_cnt, + hp_shift); + if (err) + goto err_indirect; + err = pbl_indirect_initialize(dev, pbl); + if (err) + goto err_indirect; + } + + ibdev_dbg(&dev->ibdev, + "user_pbl_created: user_pages[%u], continuous[%u]\n", + hp_cnt, pbl->physically_continuous); + + return 0; + +err_continuous: + kfree(pbl->pbl_buf); + return err; +err_indirect: + vfree(pbl->pbl_buf); + return err; +} + +static void pbl_destroy(struct efa_dev *dev, struct pbl_context *pbl) +{ + if (pbl->physically_continuous) { + dma_unmap_single(&dev->pdev->dev, pbl->phys.continuous.dma_addr, + pbl->pbl_buf_size_in_bytes, DMA_TO_DEVICE); + kfree(pbl->pbl_buf); + } else { + pbl_indirect_terminate(dev, pbl); + vfree(pbl->pbl_buf); + } +} + +static int efa_create_inline_pbl(struct efa_dev *dev, struct efa_mr *mr, + struct efa_com_reg_mr_params *params) +{ + int err; + + params->inline_pbl = 1; + err = umem_to_page_list(dev, mr->umem, params->pbl.inline_pbl_array, + params->page_num, params->page_shift); + if (err) + return err; + + ibdev_dbg(&dev->ibdev, + "inline_pbl_array - pages[%u]\n", params->page_num); + + return 0; +} + +static int efa_create_pbl(struct efa_dev *dev, + struct pbl_context *pbl, + struct efa_mr *mr, + struct efa_com_reg_mr_params *params) +{ + int err; + + err = pbl_create(dev, pbl, mr->umem, params->page_num, + params->page_shift); + if (err) { + ibdev_dbg(&dev->ibdev, "Failed to create pbl[%d]\n", err); + return err; + } + + params->inline_pbl = 0; + params->indirect = !pbl->physically_continuous; + if (pbl->physically_continuous) { + params->pbl.pbl.length = pbl->pbl_buf_size_in_bytes; + + efa_com_set_dma_addr(pbl->phys.continuous.dma_addr, + ¶ms->pbl.pbl.address.mem_addr_high, + ¶ms->pbl.pbl.address.mem_addr_low); + } else { + params->pbl.pbl.length = + pbl->phys.indirect.chunk_list.chunks[0].length; + + efa_com_set_dma_addr(pbl->phys.indirect.chunk_list.chunks[0].dma_addr, + ¶ms->pbl.pbl.address.mem_addr_high, + ¶ms->pbl.pbl.address.mem_addr_low); + } + + return 0; +} + +static void efa_cont_pages(struct ib_umem *umem, u64 addr, + unsigned long max_page_shift, + int *count, u8 *shift, u32 *ncont) +{ + struct scatterlist *sg; + u64 base = ~0, p = 0; + unsigned long tmp; + unsigned long m; + u64 len, pfn; + int i = 0; + int entry; + + addr = addr >> PAGE_SHIFT; + tmp = (unsigned long)addr; + m = find_first_bit(&tmp, BITS_PER_LONG); + if (max_page_shift) + m = min_t(unsigned long, max_page_shift - PAGE_SHIFT, m); + + for_each_sg(umem->sg_head.sgl, sg, umem->nmap, entry) { + len = DIV_ROUND_UP(sg_dma_len(sg), PAGE_SIZE); + pfn = sg_dma_address(sg) >> PAGE_SHIFT; + if (base + p != pfn) { + /* + * If either the offset or the new + * base are unaligned update m + */ + tmp = (unsigned long)(pfn | p); + if (!IS_ALIGNED(tmp, 1 << m)) + m = find_first_bit(&tmp, BITS_PER_LONG); + + base = pfn; + p = 0; + } + + p += len; + i += len; + } + + if (i) { + m = min_t(unsigned long, ilog2(roundup_pow_of_two(i)), m); + *ncont = DIV_ROUND_UP(i, (1 << m)); + } else { + m = 0; + *ncont = 0; + } + + *shift = PAGE_SHIFT + m; + *count = i; +} + +struct ib_mr *efa_reg_mr(struct ib_pd *ibpd, u64 start, u64 length, + u64 virt_addr, int access_flags, + struct ib_udata *udata) +{ + struct efa_dev *dev = to_edev(ibpd->device); + struct efa_com_reg_mr_params params = {}; + struct efa_com_reg_mr_result result = {}; + unsigned long max_page_shift; + struct pbl_context pbl; + struct efa_mr *mr; + int inline_size; + int npages; + int err; + + if (udata->inlen && + !ib_is_udata_cleared(udata, 0, sizeof(udata->inlen))) { + ibdev_dbg(&dev->ibdev, + "Incompatible ABI params, udata not cleared\n"); + err = -EINVAL; + goto err_out; + } + + if (access_flags & ~EFA_SUPPORTED_ACCESS_FLAGS) { + ibdev_dbg(&dev->ibdev, + "Unsupported access flags[%#x], supported[%#x]\n", + access_flags, EFA_SUPPORTED_ACCESS_FLAGS); + err = -EOPNOTSUPP; + goto err_out; + } + + mr = kzalloc(sizeof(*mr), GFP_KERNEL); + if (!mr) { + err = -ENOMEM; + goto err_out; + } + + mr->umem = ib_umem_get(udata, start, length, access_flags, 0); + if (IS_ERR(mr->umem)) { + err = PTR_ERR(mr->umem); + ibdev_dbg(&dev->ibdev, + "Failed to pin and map user space memory[%d]\n", err); + goto err_free; + } + + params.pd = to_epd(ibpd)->pdn; + params.iova = virt_addr; + params.mr_length_in_bytes = length; + params.permissions = access_flags & 0x1; + max_page_shift = fls64(dev->dev_attr.page_size_cap); + + efa_cont_pages(mr->umem, start, max_page_shift, &npages, + ¶ms.page_shift, ¶ms.page_num); + ibdev_dbg(&dev->ibdev, + "start %#llx length %#llx npages %d params.page_shift %u params.page_num %u\n", + start, length, npages, params.page_shift, params.page_num); + + inline_size = ARRAY_SIZE(params.pbl.inline_pbl_array); + if (params.page_num <= inline_size) { + err = efa_create_inline_pbl(dev, mr, ¶ms); + if (err) + goto err_unmap; + + err = efa_com_register_mr(&dev->edev, ¶ms, &result); + if (err) + goto err_unmap; + } else { + err = efa_create_pbl(dev, &pbl, mr, ¶ms); + if (err) + goto err_unmap; + + err = efa_com_register_mr(&dev->edev, ¶ms, &result); + pbl_destroy(dev, &pbl); + + if (err) + goto err_unmap; + } + + mr->ibmr.lkey = result.l_key; + mr->ibmr.rkey = result.r_key; + mr->ibmr.length = length; + ibdev_dbg(&dev->ibdev, "Registered mr[%d]\n", mr->ibmr.lkey); + + return &mr->ibmr; + +err_unmap: + ib_umem_release(mr->umem); +err_free: + kfree(mr); +err_out: + atomic64_inc(&dev->stats.sw_stats.reg_mr_err); + return ERR_PTR(err); +} + +int efa_dereg_mr(struct ib_mr *ibmr, struct ib_udata *udata) +{ + struct efa_dev *dev = to_edev(ibmr->device); + struct efa_com_dereg_mr_params params; + struct efa_mr *mr = to_emr(ibmr); + int err; + + if (udata->inlen && + !ib_is_udata_cleared(udata, 0, udata->inlen)) { + ibdev_dbg(&dev->ibdev, "Incompatible ABI params\n"); + return -EINVAL; + } + + ibdev_dbg(&dev->ibdev, "Deregister mr[%d]\n", ibmr->lkey); + + if (mr->umem) { + params.l_key = mr->ibmr.lkey; + err = efa_com_dereg_mr(&dev->edev, ¶ms); + if (err) + return err; + ib_umem_release(mr->umem); + } + + kfree(mr); + + return 0; +} + +int efa_get_port_immutable(struct ib_device *ibdev, u8 port_num, + struct ib_port_immutable *immutable) +{ + struct ib_port_attr attr; + int err; + + err = ib_query_port(ibdev, port_num, &attr); + if (err) { + ibdev_dbg(ibdev, "Couldn't query port err[%d]\n", err); + return err; + } + + immutable->pkey_tbl_len = attr.pkey_tbl_len; + immutable->gid_tbl_len = attr.gid_tbl_len; + + return 0; +} + +static int efa_dealloc_uar(struct efa_dev *dev, u16 uarn) +{ + struct efa_com_dealloc_uar_params params = { + .uarn = uarn, + }; + + return efa_com_dealloc_uar(&dev->edev, ¶ms); +} + +int efa_alloc_ucontext(struct ib_ucontext *ibucontext, struct ib_udata *udata) +{ + struct efa_ucontext *ucontext = to_eucontext(ibucontext); + struct efa_dev *dev = to_edev(ibucontext->device); + struct efa_ibv_alloc_ucontext_resp resp = {}; + struct efa_com_alloc_uar_result result; + int err; + + /* + * it's fine if the driver does not know all request fields, + * we will ack input fields in our response. + */ + + err = efa_com_alloc_uar(&dev->edev, &result); + if (err) + goto err_out; + + ucontext->uarn = result.uarn; + xa_init(&ucontext->mmap_xa); + + resp.cmds_supp_udata_mask |= EFA_USER_CMDS_SUPP_UDATA_QUERY_DEVICE; + resp.cmds_supp_udata_mask |= EFA_USER_CMDS_SUPP_UDATA_CREATE_AH; + resp.sub_cqs_per_cq = dev->dev_attr.sub_cqs_per_cq; + resp.inline_buf_size = dev->dev_attr.inline_buf_size; + resp.max_llq_size = dev->dev_attr.max_llq_size; + + if (udata && udata->outlen) { + err = ib_copy_to_udata(udata, &resp, + min(sizeof(resp), udata->outlen)); + if (err) + goto err_dealloc_uar; + } + + return 0; + +err_dealloc_uar: + efa_dealloc_uar(dev, result.uarn); +err_out: + atomic64_inc(&dev->stats.sw_stats.alloc_ucontext_err); + return err; +} + +void efa_dealloc_ucontext(struct ib_ucontext *ibucontext) +{ + struct efa_ucontext *ucontext = to_eucontext(ibucontext); + struct efa_dev *dev = to_edev(ibucontext->device); + + mmap_entries_remove_free(dev, ucontext); + efa_dealloc_uar(dev, ucontext->uarn); +} + +static int __efa_mmap(struct efa_dev *dev, struct efa_ucontext *ucontext, + struct vm_area_struct *vma, u64 key, u64 length) +{ + struct efa_mmap_entry *entry; + unsigned long va; + u64 pfn; + int err; + + entry = mmap_entry_get(dev, ucontext, key, length); + if (!entry) { + ibdev_dbg(&dev->ibdev, "key[%#llx] does not have valid entry\n", + key); + return -EINVAL; + } + + ibdev_dbg(&dev->ibdev, + "Mapping address[%#llx], length[%#llx], mmap_flag[%d]\n", + entry->address, length, entry->mmap_flag); + + pfn = entry->address >> PAGE_SHIFT; + switch (entry->mmap_flag) { + case EFA_MMAP_IO_NC: + err = rdma_user_mmap_io(&ucontext->ibucontext, vma, pfn, length, + pgprot_noncached(vma->vm_page_prot)); + break; + case EFA_MMAP_IO_WC: + err = rdma_user_mmap_io(&ucontext->ibucontext, vma, pfn, length, + pgprot_writecombine(vma->vm_page_prot)); + break; + case EFA_MMAP_DMA_PAGE: + for (va = vma->vm_start; va < vma->vm_end; + va += PAGE_SIZE, pfn++) { + err = vm_insert_page(vma, va, pfn_to_page(pfn)); + if (err) + break; + } + break; + default: + err = -EINVAL; + } + + if (err) + ibdev_dbg( + &dev->ibdev, + "Couldn't mmap address[%#llx] length[%#llx] mmap_flag[%d] err[%d]\n", + entry->address, length, entry->mmap_flag, err); + + return err; +} + +int efa_mmap(struct ib_ucontext *ibucontext, + struct vm_area_struct *vma) +{ + struct efa_ucontext *ucontext = to_eucontext(ibucontext); + struct efa_dev *dev = to_edev(ibucontext->device); + u64 length = vma->vm_end - vma->vm_start; + u64 key = vma->vm_pgoff << PAGE_SHIFT; + + ibdev_dbg(&dev->ibdev, + "start %#lx, end %#lx, length = %#llx, key = %#llx\n", + vma->vm_start, vma->vm_end, length, key); + + if (length % PAGE_SIZE != 0 || !(vma->vm_flags & VM_SHARED)) { + ibdev_dbg(&dev->ibdev, + "length[%#llx] is not page size aligned[%#lx] or VM_SHARED is not set [%#lx]\n", + length, PAGE_SIZE, vma->vm_flags); + return -EINVAL; + } + + if (vma->vm_flags & VM_EXEC) { + ibdev_dbg(&dev->ibdev, "Mapping executable pages is not permitted\n"); + return -EPERM; + } + vma->vm_flags &= ~VM_MAYEXEC; + + return __efa_mmap(dev, ucontext, vma, key, length); +} + +static int efa_ah_destroy(struct efa_dev *dev, struct efa_ah *ah) +{ + struct efa_com_destroy_ah_params params = { + .ah = ah->ah, + .pdn = to_epd(ah->ibah.pd)->pdn, + }; + + return efa_com_destroy_ah(&dev->edev, ¶ms); +} + +int efa_create_ah(struct ib_ah *ibah, + struct rdma_ah_attr *ah_attr, + u32 flags, + struct ib_udata *udata) +{ + struct efa_dev *dev = to_edev(ibah->device); + struct efa_com_create_ah_params params = {}; + struct efa_ibv_create_ah_resp resp = {}; + struct efa_com_create_ah_result result; + struct efa_ah *ah = to_eah(ibah); + int err; + + if (!(flags & RDMA_CREATE_AH_SLEEPABLE)) { + ibdev_dbg(&dev->ibdev, + "Create address handle is not supported in atomic context\n"); + err = -EOPNOTSUPP; + goto err_out; + } + + if (udata->inlen && + !ib_is_udata_cleared(udata, 0, udata->inlen)) { + ibdev_dbg(&dev->ibdev, "Incompatible ABI params\n"); + err = -EINVAL; + goto err_out; + } + + memcpy(params.dest_addr, ah_attr->grh.dgid.raw, + sizeof(params.dest_addr)); + params.pdn = to_epd(ibah->pd)->pdn; + err = efa_com_create_ah(&dev->edev, ¶ms, &result); + if (err) + goto err_out; + + memcpy(ah->id, ah_attr->grh.dgid.raw, sizeof(ah->id)); + ah->ah = result.ah; + + resp.efa_address_handle = result.ah; + + if (udata->outlen) { + err = ib_copy_to_udata(udata, &resp, + min(sizeof(resp), udata->outlen)); + if (err) { + ibdev_dbg(&dev->ibdev, + "Failed to copy udata for create_ah response\n"); + goto err_destroy_ah; + } + } + ibdev_dbg(&dev->ibdev, "Created ah[%d]\n", ah->ah); + + return 0; + +err_destroy_ah: + efa_ah_destroy(dev, ah); +err_out: + atomic64_inc(&dev->stats.sw_stats.create_ah_err); + return err; +} + +void efa_destroy_ah(struct ib_ah *ibah, u32 flags) +{ + struct efa_dev *dev = to_edev(ibah->pd->device); + struct efa_ah *ah = to_eah(ibah); + + ibdev_dbg(&dev->ibdev, "Destroy ah[%d]\n", ah->ah); + + if (!(flags & RDMA_DESTROY_AH_SLEEPABLE)) { + ibdev_dbg(&dev->ibdev, + "Destroy address handle is not supported in atomic context\n"); + return; + } + + efa_ah_destroy(dev, ah); +} + +enum rdma_link_layer efa_port_link_layer(struct ib_device *ibdev, + u8 port_num) +{ + return IB_LINK_LAYER_UNSPECIFIED; +} + |