diff --git a/sys/dev/nvme/nvme_private.h b/sys/dev/nvme/nvme_private.h index 02cecaf03a97..b00130910a82 100644 --- a/sys/dev/nvme/nvme_private.h +++ b/sys/dev/nvme/nvme_private.h @@ -1,572 +1,573 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (C) 2012-2014 Intel Corporation * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * $FreeBSD$ */ #ifndef __NVME_PRIVATE_H__ #define __NVME_PRIVATE_H__ #include #include #include #include #include #include #include #include #include #include #include #include #include #include "nvme.h" #define DEVICE2SOFTC(dev) ((struct nvme_controller *) device_get_softc(dev)) MALLOC_DECLARE(M_NVME); #define IDT32_PCI_ID 0x80d0111d /* 32 channel board */ #define IDT8_PCI_ID 0x80d2111d /* 8 channel board */ #define NVME_ADMIN_TRACKERS (16) #define NVME_ADMIN_ENTRIES (128) /* min and max are defined in admin queue attributes section of spec */ #define NVME_MIN_ADMIN_ENTRIES (2) #define NVME_MAX_ADMIN_ENTRIES (4096) /* * NVME_IO_ENTRIES defines the size of an I/O qpair's submission and completion * queues, while NVME_IO_TRACKERS defines the maximum number of I/O that we * will allow outstanding on an I/O qpair at any time. The only advantage in * having IO_ENTRIES > IO_TRACKERS is for debugging purposes - when dumping * the contents of the submission and completion queues, it will show a longer * history of data. */ #define NVME_IO_ENTRIES (256) #define NVME_IO_TRACKERS (128) #define NVME_MIN_IO_TRACKERS (4) #define NVME_MAX_IO_TRACKERS (1024) /* * NVME_MAX_IO_ENTRIES is not defined, since it is specified in CC.MQES * for each controller. */ #define NVME_INT_COAL_TIME (0) /* disabled */ #define NVME_INT_COAL_THRESHOLD (0) /* 0-based */ #define NVME_MAX_NAMESPACES (16) #define NVME_MAX_CONSUMERS (2) #define NVME_MAX_ASYNC_EVENTS (8) #define NVME_DEFAULT_TIMEOUT_PERIOD (30) /* in seconds */ #define NVME_MIN_TIMEOUT_PERIOD (5) #define NVME_MAX_TIMEOUT_PERIOD (120) #define NVME_DEFAULT_RETRY_COUNT (4) /* Maximum log page size to fetch for AERs. */ #define NVME_MAX_AER_LOG_SIZE (4096) /* * Define CACHE_LINE_SIZE here for older FreeBSD versions that do not define * it. */ #ifndef CACHE_LINE_SIZE #define CACHE_LINE_SIZE (64) #endif #define NVME_GONE 0xfffffffful extern int32_t nvme_retry_count; extern bool nvme_verbose_cmd_dump; struct nvme_completion_poll_status { struct nvme_completion cpl; int done; }; extern devclass_t nvme_devclass; #define NVME_REQUEST_VADDR 1 #define NVME_REQUEST_NULL 2 /* For requests with no payload. */ #define NVME_REQUEST_UIO 3 #define NVME_REQUEST_BIO 4 #define NVME_REQUEST_CCB 5 struct nvme_request { struct nvme_command cmd; struct nvme_qpair *qpair; union { void *payload; struct bio *bio; } u; uint32_t type; uint32_t payload_size; bool timeout; nvme_cb_fn_t cb_fn; void *cb_arg; int32_t retries; STAILQ_ENTRY(nvme_request) stailq; }; struct nvme_async_event_request { struct nvme_controller *ctrlr; struct nvme_request *req; struct nvme_completion cpl; uint32_t log_page_id; uint32_t log_page_size; uint8_t log_page_buffer[NVME_MAX_AER_LOG_SIZE]; }; struct nvme_tracker { TAILQ_ENTRY(nvme_tracker) tailq; struct nvme_request *req; struct nvme_qpair *qpair; sbintime_t deadline; bus_dmamap_t payload_dma_map; uint16_t cid; uint64_t *prp; bus_addr_t prp_bus_addr; }; enum nvme_recovery { RECOVERY_NONE = 0, /* Normal operations */ RECOVERY_START, /* Deadline has passed, start recovering */ RECOVERY_RESET, /* This pass, initiate reset of controller */ RECOVERY_WAITING, /* waiting for the reset to complete */ }; struct nvme_qpair { struct nvme_controller *ctrlr; uint32_t id; int domain; int cpu; uint16_t vector; int rid; struct resource *res; void *tag; struct callout timer; sbintime_t deadline; bool timer_armed; enum nvme_recovery recovery_state; uint32_t num_entries; uint32_t num_trackers; uint32_t sq_tdbl_off; uint32_t cq_hdbl_off; uint32_t phase; uint32_t sq_head; uint32_t sq_tail; uint32_t cq_head; int64_t num_cmds; int64_t num_intr_handler_calls; int64_t num_retries; int64_t num_failures; + int64_t num_ignored; struct nvme_command *cmd; struct nvme_completion *cpl; bus_dma_tag_t dma_tag; bus_dma_tag_t dma_tag_payload; bus_dmamap_t queuemem_map; uint64_t cmd_bus_addr; uint64_t cpl_bus_addr; TAILQ_HEAD(, nvme_tracker) free_tr; TAILQ_HEAD(, nvme_tracker) outstanding_tr; STAILQ_HEAD(, nvme_request) queued_req; struct nvme_tracker **act_tr; struct mtx lock __aligned(CACHE_LINE_SIZE); } __aligned(CACHE_LINE_SIZE); struct nvme_namespace { struct nvme_controller *ctrlr; struct nvme_namespace_data data; uint32_t id; uint32_t flags; struct cdev *cdev; void *cons_cookie[NVME_MAX_CONSUMERS]; uint32_t boundary; struct mtx lock; }; /* * One of these per allocated PCI device. */ struct nvme_controller { device_t dev; struct mtx lock; int domain; uint32_t ready_timeout_in_ms; uint32_t quirks; #define QUIRK_DELAY_B4_CHK_RDY 1 /* Can't touch MMIO on disable */ #define QUIRK_DISABLE_TIMEOUT 2 /* Disable broken completion timeout feature */ bus_space_tag_t bus_tag; bus_space_handle_t bus_handle; int resource_id; struct resource *resource; /* * The NVMe spec allows for the MSI-X table to be placed in BAR 4/5, * separate from the control registers which are in BAR 0/1. These * members track the mapping of BAR 4/5 for that reason. */ int bar4_resource_id; struct resource *bar4_resource; int msi_count; uint32_t enable_aborts; uint32_t num_io_queues; uint32_t max_hw_pend_io; /* Fields for tracking progress during controller initialization. */ struct intr_config_hook config_hook; uint32_t ns_identified; uint32_t queues_created; struct task reset_task; struct task fail_req_task; struct taskqueue *taskqueue; /* For shared legacy interrupt. */ int rid; struct resource *res; void *tag; /** maximum i/o size in bytes */ uint32_t max_xfer_size; /** minimum page size supported by this controller in bytes */ uint32_t min_page_size; /** interrupt coalescing time period (in microseconds) */ uint32_t int_coal_time; /** interrupt coalescing threshold */ uint32_t int_coal_threshold; /** timeout period in seconds */ uint32_t timeout_period; /** doorbell stride */ uint32_t dstrd; struct nvme_qpair adminq; struct nvme_qpair *ioq; struct nvme_registers *regs; struct nvme_controller_data cdata; struct nvme_namespace ns[NVME_MAX_NAMESPACES]; struct cdev *cdev; /** bit mask of event types currently enabled for async events */ uint32_t async_event_config; uint32_t num_aers; struct nvme_async_event_request aer[NVME_MAX_ASYNC_EVENTS]; void *cons_cookie[NVME_MAX_CONSUMERS]; uint32_t is_resetting; uint32_t is_initialized; uint32_t notification_sent; bool is_failed; bool is_dying; STAILQ_HEAD(, nvme_request) fail_req; /* Host Memory Buffer */ int hmb_nchunks; size_t hmb_chunk; bus_dma_tag_t hmb_tag; struct nvme_hmb_chunk { bus_dmamap_t hmbc_map; void *hmbc_vaddr; uint64_t hmbc_paddr; } *hmb_chunks; bus_dma_tag_t hmb_desc_tag; bus_dmamap_t hmb_desc_map; struct nvme_hmb_desc *hmb_desc_vaddr; uint64_t hmb_desc_paddr; }; #define nvme_mmio_offsetof(reg) \ offsetof(struct nvme_registers, reg) #define nvme_mmio_read_4(sc, reg) \ bus_space_read_4((sc)->bus_tag, (sc)->bus_handle, \ nvme_mmio_offsetof(reg)) #define nvme_mmio_write_4(sc, reg, val) \ bus_space_write_4((sc)->bus_tag, (sc)->bus_handle, \ nvme_mmio_offsetof(reg), val) #define nvme_mmio_write_8(sc, reg, val) \ do { \ bus_space_write_4((sc)->bus_tag, (sc)->bus_handle, \ nvme_mmio_offsetof(reg), val & 0xFFFFFFFF); \ bus_space_write_4((sc)->bus_tag, (sc)->bus_handle, \ nvme_mmio_offsetof(reg)+4, \ (val & 0xFFFFFFFF00000000ULL) >> 32); \ } while (0); #define nvme_printf(ctrlr, fmt, args...) \ device_printf(ctrlr->dev, fmt, ##args) void nvme_ns_test(struct nvme_namespace *ns, u_long cmd, caddr_t arg); void nvme_ctrlr_cmd_identify_controller(struct nvme_controller *ctrlr, void *payload, nvme_cb_fn_t cb_fn, void *cb_arg); void nvme_ctrlr_cmd_identify_namespace(struct nvme_controller *ctrlr, uint32_t nsid, void *payload, nvme_cb_fn_t cb_fn, void *cb_arg); void nvme_ctrlr_cmd_set_interrupt_coalescing(struct nvme_controller *ctrlr, uint32_t microseconds, uint32_t threshold, nvme_cb_fn_t cb_fn, void *cb_arg); void nvme_ctrlr_cmd_get_error_page(struct nvme_controller *ctrlr, struct nvme_error_information_entry *payload, uint32_t num_entries, /* 0 = max */ nvme_cb_fn_t cb_fn, void *cb_arg); void nvme_ctrlr_cmd_get_health_information_page(struct nvme_controller *ctrlr, uint32_t nsid, struct nvme_health_information_page *payload, nvme_cb_fn_t cb_fn, void *cb_arg); void nvme_ctrlr_cmd_get_firmware_page(struct nvme_controller *ctrlr, struct nvme_firmware_page *payload, nvme_cb_fn_t cb_fn, void *cb_arg); void nvme_ctrlr_cmd_create_io_cq(struct nvme_controller *ctrlr, struct nvme_qpair *io_que, nvme_cb_fn_t cb_fn, void *cb_arg); void nvme_ctrlr_cmd_create_io_sq(struct nvme_controller *ctrlr, struct nvme_qpair *io_que, nvme_cb_fn_t cb_fn, void *cb_arg); void nvme_ctrlr_cmd_delete_io_cq(struct nvme_controller *ctrlr, struct nvme_qpair *io_que, nvme_cb_fn_t cb_fn, void *cb_arg); void nvme_ctrlr_cmd_delete_io_sq(struct nvme_controller *ctrlr, struct nvme_qpair *io_que, nvme_cb_fn_t cb_fn, void *cb_arg); void nvme_ctrlr_cmd_set_num_queues(struct nvme_controller *ctrlr, uint32_t num_queues, nvme_cb_fn_t cb_fn, void *cb_arg); void nvme_ctrlr_cmd_set_async_event_config(struct nvme_controller *ctrlr, uint32_t state, nvme_cb_fn_t cb_fn, void *cb_arg); void nvme_ctrlr_cmd_abort(struct nvme_controller *ctrlr, uint16_t cid, uint16_t sqid, nvme_cb_fn_t cb_fn, void *cb_arg); void nvme_completion_poll_cb(void *arg, const struct nvme_completion *cpl); int nvme_ctrlr_construct(struct nvme_controller *ctrlr, device_t dev); void nvme_ctrlr_destruct(struct nvme_controller *ctrlr, device_t dev); void nvme_ctrlr_shutdown(struct nvme_controller *ctrlr); void nvme_ctrlr_reset(struct nvme_controller *ctrlr); /* ctrlr defined as void * to allow use with config_intrhook. */ void nvme_ctrlr_start_config_hook(void *ctrlr_arg); void nvme_ctrlr_submit_admin_request(struct nvme_controller *ctrlr, struct nvme_request *req); void nvme_ctrlr_submit_io_request(struct nvme_controller *ctrlr, struct nvme_request *req); void nvme_ctrlr_post_failed_request(struct nvme_controller *ctrlr, struct nvme_request *req); int nvme_qpair_construct(struct nvme_qpair *qpair, uint32_t num_entries, uint32_t num_trackers, struct nvme_controller *ctrlr); void nvme_qpair_submit_tracker(struct nvme_qpair *qpair, struct nvme_tracker *tr); bool nvme_qpair_process_completions(struct nvme_qpair *qpair); void nvme_qpair_submit_request(struct nvme_qpair *qpair, struct nvme_request *req); void nvme_qpair_reset(struct nvme_qpair *qpair); void nvme_qpair_fail(struct nvme_qpair *qpair); void nvme_qpair_manual_complete_request(struct nvme_qpair *qpair, struct nvme_request *req, uint32_t sct, uint32_t sc); void nvme_admin_qpair_enable(struct nvme_qpair *qpair); void nvme_admin_qpair_disable(struct nvme_qpair *qpair); void nvme_admin_qpair_destroy(struct nvme_qpair *qpair); void nvme_io_qpair_enable(struct nvme_qpair *qpair); void nvme_io_qpair_disable(struct nvme_qpair *qpair); void nvme_io_qpair_destroy(struct nvme_qpair *qpair); int nvme_ns_construct(struct nvme_namespace *ns, uint32_t id, struct nvme_controller *ctrlr); void nvme_ns_destruct(struct nvme_namespace *ns); void nvme_sysctl_initialize_ctrlr(struct nvme_controller *ctrlr); void nvme_dump_command(struct nvme_command *cmd); void nvme_dump_completion(struct nvme_completion *cpl); int nvme_attach(device_t dev); int nvme_shutdown(device_t dev); int nvme_detach(device_t dev); /* * Wait for a command to complete using the nvme_completion_poll_cb. * Used in limited contexts where the caller knows it's OK to block * briefly while the command runs. The ISR will run the callback which * will set status->done to true, usually within microseconds. If not, * then after one second timeout handler should reset the controller * and abort all outstanding requests including this polled one. If * still not after ten seconds, then something is wrong with the driver, * and panic is the only way to recover. */ static __inline void nvme_completion_poll(struct nvme_completion_poll_status *status) { int sanity = hz * 10; while (!atomic_load_acq_int(&status->done) && --sanity > 0) pause("nvme", 1); if (sanity <= 0) panic("NVME polled command failed to complete within 10s."); } static __inline void nvme_single_map(void *arg, bus_dma_segment_t *seg, int nseg, int error) { uint64_t *bus_addr = (uint64_t *)arg; KASSERT(nseg == 1, ("number of segments (%d) is not 1", nseg)); if (error != 0) printf("nvme_single_map err %d\n", error); *bus_addr = seg[0].ds_addr; } static __inline struct nvme_request * _nvme_allocate_request(nvme_cb_fn_t cb_fn, void *cb_arg) { struct nvme_request *req; req = malloc(sizeof(*req), M_NVME, M_NOWAIT | M_ZERO); if (req != NULL) { req->cb_fn = cb_fn; req->cb_arg = cb_arg; req->timeout = true; } return (req); } static __inline struct nvme_request * nvme_allocate_request_vaddr(void *payload, uint32_t payload_size, nvme_cb_fn_t cb_fn, void *cb_arg) { struct nvme_request *req; req = _nvme_allocate_request(cb_fn, cb_arg); if (req != NULL) { req->type = NVME_REQUEST_VADDR; req->u.payload = payload; req->payload_size = payload_size; } return (req); } static __inline struct nvme_request * nvme_allocate_request_null(nvme_cb_fn_t cb_fn, void *cb_arg) { struct nvme_request *req; req = _nvme_allocate_request(cb_fn, cb_arg); if (req != NULL) req->type = NVME_REQUEST_NULL; return (req); } static __inline struct nvme_request * nvme_allocate_request_bio(struct bio *bio, nvme_cb_fn_t cb_fn, void *cb_arg) { struct nvme_request *req; req = _nvme_allocate_request(cb_fn, cb_arg); if (req != NULL) { req->type = NVME_REQUEST_BIO; req->u.bio = bio; } return (req); } static __inline struct nvme_request * nvme_allocate_request_ccb(union ccb *ccb, nvme_cb_fn_t cb_fn, void *cb_arg) { struct nvme_request *req; req = _nvme_allocate_request(cb_fn, cb_arg); if (req != NULL) { req->type = NVME_REQUEST_CCB; req->u.payload = ccb; } return (req); } #define nvme_free_request(req) free(req, M_NVME) void nvme_notify_async_consumers(struct nvme_controller *ctrlr, const struct nvme_completion *async_cpl, uint32_t log_page_id, void *log_page_buffer, uint32_t log_page_size); void nvme_notify_fail_consumers(struct nvme_controller *ctrlr); void nvme_notify_new_controller(struct nvme_controller *ctrlr); void nvme_notify_ns(struct nvme_controller *ctrlr, int nsid); void nvme_ctrlr_shared_handler(void *arg); void nvme_ctrlr_poll(struct nvme_controller *ctrlr); int nvme_ctrlr_suspend(struct nvme_controller *ctrlr); int nvme_ctrlr_resume(struct nvme_controller *ctrlr); #endif /* __NVME_PRIVATE_H__ */ diff --git a/sys/dev/nvme/nvme_qpair.c b/sys/dev/nvme/nvme_qpair.c index 827054efd48e..788322092f88 100644 --- a/sys/dev/nvme/nvme_qpair.c +++ b/sys/dev/nvme/nvme_qpair.c @@ -1,1380 +1,1383 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (C) 2012-2014 Intel Corporation * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ #include __FBSDID("$FreeBSD$"); #include #include #include #include #include #include #include "nvme_private.h" typedef enum error_print { ERROR_PRINT_NONE, ERROR_PRINT_NO_RETRY, ERROR_PRINT_ALL } error_print_t; #define DO_NOT_RETRY 1 static void _nvme_qpair_submit_request(struct nvme_qpair *qpair, struct nvme_request *req); static void nvme_qpair_destroy(struct nvme_qpair *qpair); struct nvme_opcode_string { uint16_t opc; const char * str; }; static struct nvme_opcode_string admin_opcode[] = { { NVME_OPC_DELETE_IO_SQ, "DELETE IO SQ" }, { NVME_OPC_CREATE_IO_SQ, "CREATE IO SQ" }, { NVME_OPC_GET_LOG_PAGE, "GET LOG PAGE" }, { NVME_OPC_DELETE_IO_CQ, "DELETE IO CQ" }, { NVME_OPC_CREATE_IO_CQ, "CREATE IO CQ" }, { NVME_OPC_IDENTIFY, "IDENTIFY" }, { NVME_OPC_ABORT, "ABORT" }, { NVME_OPC_SET_FEATURES, "SET FEATURES" }, { NVME_OPC_GET_FEATURES, "GET FEATURES" }, { NVME_OPC_ASYNC_EVENT_REQUEST, "ASYNC EVENT REQUEST" }, { NVME_OPC_FIRMWARE_ACTIVATE, "FIRMWARE ACTIVATE" }, { NVME_OPC_FIRMWARE_IMAGE_DOWNLOAD, "FIRMWARE IMAGE DOWNLOAD" }, { NVME_OPC_DEVICE_SELF_TEST, "DEVICE SELF-TEST" }, { NVME_OPC_NAMESPACE_ATTACHMENT, "NAMESPACE ATTACHMENT" }, { NVME_OPC_KEEP_ALIVE, "KEEP ALIVE" }, { NVME_OPC_DIRECTIVE_SEND, "DIRECTIVE SEND" }, { NVME_OPC_DIRECTIVE_RECEIVE, "DIRECTIVE RECEIVE" }, { NVME_OPC_VIRTUALIZATION_MANAGEMENT, "VIRTUALIZATION MANAGEMENT" }, { NVME_OPC_NVME_MI_SEND, "NVME-MI SEND" }, { NVME_OPC_NVME_MI_RECEIVE, "NVME-MI RECEIVE" }, { NVME_OPC_DOORBELL_BUFFER_CONFIG, "DOORBELL BUFFER CONFIG" }, { NVME_OPC_FORMAT_NVM, "FORMAT NVM" }, { NVME_OPC_SECURITY_SEND, "SECURITY SEND" }, { NVME_OPC_SECURITY_RECEIVE, "SECURITY RECEIVE" }, { NVME_OPC_SANITIZE, "SANITIZE" }, { NVME_OPC_GET_LBA_STATUS, "GET LBA STATUS" }, { 0xFFFF, "ADMIN COMMAND" } }; static struct nvme_opcode_string io_opcode[] = { { NVME_OPC_FLUSH, "FLUSH" }, { NVME_OPC_WRITE, "WRITE" }, { NVME_OPC_READ, "READ" }, { NVME_OPC_WRITE_UNCORRECTABLE, "WRITE UNCORRECTABLE" }, { NVME_OPC_COMPARE, "COMPARE" }, { NVME_OPC_WRITE_ZEROES, "WRITE ZEROES" }, { NVME_OPC_DATASET_MANAGEMENT, "DATASET MANAGEMENT" }, { NVME_OPC_VERIFY, "VERIFY" }, { NVME_OPC_RESERVATION_REGISTER, "RESERVATION REGISTER" }, { NVME_OPC_RESERVATION_REPORT, "RESERVATION REPORT" }, { NVME_OPC_RESERVATION_ACQUIRE, "RESERVATION ACQUIRE" }, { NVME_OPC_RESERVATION_RELEASE, "RESERVATION RELEASE" }, { 0xFFFF, "IO COMMAND" } }; static const char * get_admin_opcode_string(uint16_t opc) { struct nvme_opcode_string *entry; entry = admin_opcode; while (entry->opc != 0xFFFF) { if (entry->opc == opc) return (entry->str); entry++; } return (entry->str); } static const char * get_io_opcode_string(uint16_t opc) { struct nvme_opcode_string *entry; entry = io_opcode; while (entry->opc != 0xFFFF) { if (entry->opc == opc) return (entry->str); entry++; } return (entry->str); } static void nvme_admin_qpair_print_command(struct nvme_qpair *qpair, struct nvme_command *cmd) { nvme_printf(qpair->ctrlr, "%s (%02x) sqid:%d cid:%d nsid:%x " "cdw10:%08x cdw11:%08x\n", get_admin_opcode_string(cmd->opc), cmd->opc, qpair->id, cmd->cid, le32toh(cmd->nsid), le32toh(cmd->cdw10), le32toh(cmd->cdw11)); } static void nvme_io_qpair_print_command(struct nvme_qpair *qpair, struct nvme_command *cmd) { switch (cmd->opc) { case NVME_OPC_WRITE: case NVME_OPC_READ: case NVME_OPC_WRITE_UNCORRECTABLE: case NVME_OPC_COMPARE: case NVME_OPC_WRITE_ZEROES: case NVME_OPC_VERIFY: nvme_printf(qpair->ctrlr, "%s sqid:%d cid:%d nsid:%d " "lba:%llu len:%d\n", get_io_opcode_string(cmd->opc), qpair->id, cmd->cid, le32toh(cmd->nsid), ((unsigned long long)le32toh(cmd->cdw11) << 32) + le32toh(cmd->cdw10), (le32toh(cmd->cdw12) & 0xFFFF) + 1); break; case NVME_OPC_FLUSH: case NVME_OPC_DATASET_MANAGEMENT: case NVME_OPC_RESERVATION_REGISTER: case NVME_OPC_RESERVATION_REPORT: case NVME_OPC_RESERVATION_ACQUIRE: case NVME_OPC_RESERVATION_RELEASE: nvme_printf(qpair->ctrlr, "%s sqid:%d cid:%d nsid:%d\n", get_io_opcode_string(cmd->opc), qpair->id, cmd->cid, le32toh(cmd->nsid)); break; default: nvme_printf(qpair->ctrlr, "%s (%02x) sqid:%d cid:%d nsid:%d\n", get_io_opcode_string(cmd->opc), cmd->opc, qpair->id, cmd->cid, le32toh(cmd->nsid)); break; } } static void nvme_qpair_print_command(struct nvme_qpair *qpair, struct nvme_command *cmd) { if (qpair->id == 0) nvme_admin_qpair_print_command(qpair, cmd); else nvme_io_qpair_print_command(qpair, cmd); if (nvme_verbose_cmd_dump) { nvme_printf(qpair->ctrlr, "nsid:%#x rsvd2:%#x rsvd3:%#x mptr:%#jx prp1:%#jx prp2:%#jx\n", cmd->nsid, cmd->rsvd2, cmd->rsvd3, (uintmax_t)cmd->mptr, (uintmax_t)cmd->prp1, (uintmax_t)cmd->prp2); nvme_printf(qpair->ctrlr, "cdw10: %#x cdw11:%#x cdw12:%#x cdw13:%#x cdw14:%#x cdw15:%#x\n", cmd->cdw10, cmd->cdw11, cmd->cdw12, cmd->cdw13, cmd->cdw14, cmd->cdw15); } } struct nvme_status_string { uint16_t sc; const char * str; }; static struct nvme_status_string generic_status[] = { { NVME_SC_SUCCESS, "SUCCESS" }, { NVME_SC_INVALID_OPCODE, "INVALID OPCODE" }, { NVME_SC_INVALID_FIELD, "INVALID_FIELD" }, { NVME_SC_COMMAND_ID_CONFLICT, "COMMAND ID CONFLICT" }, { NVME_SC_DATA_TRANSFER_ERROR, "DATA TRANSFER ERROR" }, { NVME_SC_ABORTED_POWER_LOSS, "ABORTED - POWER LOSS" }, { NVME_SC_INTERNAL_DEVICE_ERROR, "INTERNAL DEVICE ERROR" }, { NVME_SC_ABORTED_BY_REQUEST, "ABORTED - BY REQUEST" }, { NVME_SC_ABORTED_SQ_DELETION, "ABORTED - SQ DELETION" }, { NVME_SC_ABORTED_FAILED_FUSED, "ABORTED - FAILED FUSED" }, { NVME_SC_ABORTED_MISSING_FUSED, "ABORTED - MISSING FUSED" }, { NVME_SC_INVALID_NAMESPACE_OR_FORMAT, "INVALID NAMESPACE OR FORMAT" }, { NVME_SC_COMMAND_SEQUENCE_ERROR, "COMMAND SEQUENCE ERROR" }, { NVME_SC_INVALID_SGL_SEGMENT_DESCR, "INVALID SGL SEGMENT DESCRIPTOR" }, { NVME_SC_INVALID_NUMBER_OF_SGL_DESCR, "INVALID NUMBER OF SGL DESCRIPTORS" }, { NVME_SC_DATA_SGL_LENGTH_INVALID, "DATA SGL LENGTH INVALID" }, { NVME_SC_METADATA_SGL_LENGTH_INVALID, "METADATA SGL LENGTH INVALID" }, { NVME_SC_SGL_DESCRIPTOR_TYPE_INVALID, "SGL DESCRIPTOR TYPE INVALID" }, { NVME_SC_INVALID_USE_OF_CMB, "INVALID USE OF CONTROLLER MEMORY BUFFER" }, { NVME_SC_PRP_OFFET_INVALID, "PRP OFFET INVALID" }, { NVME_SC_ATOMIC_WRITE_UNIT_EXCEEDED, "ATOMIC WRITE UNIT EXCEEDED" }, { NVME_SC_OPERATION_DENIED, "OPERATION DENIED" }, { NVME_SC_SGL_OFFSET_INVALID, "SGL OFFSET INVALID" }, { NVME_SC_HOST_ID_INCONSISTENT_FORMAT, "HOST IDENTIFIER INCONSISTENT FORMAT" }, { NVME_SC_KEEP_ALIVE_TIMEOUT_EXPIRED, "KEEP ALIVE TIMEOUT EXPIRED" }, { NVME_SC_KEEP_ALIVE_TIMEOUT_INVALID, "KEEP ALIVE TIMEOUT INVALID" }, { NVME_SC_ABORTED_DUE_TO_PREEMPT, "COMMAND ABORTED DUE TO PREEMPT AND ABORT" }, { NVME_SC_SANITIZE_FAILED, "SANITIZE FAILED" }, { NVME_SC_SANITIZE_IN_PROGRESS, "SANITIZE IN PROGRESS" }, { NVME_SC_SGL_DATA_BLOCK_GRAN_INVALID, "SGL_DATA_BLOCK_GRANULARITY_INVALID" }, { NVME_SC_NOT_SUPPORTED_IN_CMB, "COMMAND NOT SUPPORTED FOR QUEUE IN CMB" }, { NVME_SC_NAMESPACE_IS_WRITE_PROTECTED, "NAMESPACE IS WRITE PROTECTED" }, { NVME_SC_COMMAND_INTERRUPTED, "COMMAND INTERRUPTED" }, { NVME_SC_TRANSIENT_TRANSPORT_ERROR, "TRANSIENT TRANSPORT ERROR" }, { NVME_SC_LBA_OUT_OF_RANGE, "LBA OUT OF RANGE" }, { NVME_SC_CAPACITY_EXCEEDED, "CAPACITY EXCEEDED" }, { NVME_SC_NAMESPACE_NOT_READY, "NAMESPACE NOT READY" }, { NVME_SC_RESERVATION_CONFLICT, "RESERVATION CONFLICT" }, { NVME_SC_FORMAT_IN_PROGRESS, "FORMAT IN PROGRESS" }, { 0xFFFF, "GENERIC" } }; static struct nvme_status_string command_specific_status[] = { { NVME_SC_COMPLETION_QUEUE_INVALID, "INVALID COMPLETION QUEUE" }, { NVME_SC_INVALID_QUEUE_IDENTIFIER, "INVALID QUEUE IDENTIFIER" }, { NVME_SC_MAXIMUM_QUEUE_SIZE_EXCEEDED, "MAX QUEUE SIZE EXCEEDED" }, { NVME_SC_ABORT_COMMAND_LIMIT_EXCEEDED, "ABORT CMD LIMIT EXCEEDED" }, { NVME_SC_ASYNC_EVENT_REQUEST_LIMIT_EXCEEDED, "ASYNC LIMIT EXCEEDED" }, { NVME_SC_INVALID_FIRMWARE_SLOT, "INVALID FIRMWARE SLOT" }, { NVME_SC_INVALID_FIRMWARE_IMAGE, "INVALID FIRMWARE IMAGE" }, { NVME_SC_INVALID_INTERRUPT_VECTOR, "INVALID INTERRUPT VECTOR" }, { NVME_SC_INVALID_LOG_PAGE, "INVALID LOG PAGE" }, { NVME_SC_INVALID_FORMAT, "INVALID FORMAT" }, { NVME_SC_FIRMWARE_REQUIRES_RESET, "FIRMWARE REQUIRES RESET" }, { NVME_SC_INVALID_QUEUE_DELETION, "INVALID QUEUE DELETION" }, { NVME_SC_FEATURE_NOT_SAVEABLE, "FEATURE IDENTIFIER NOT SAVEABLE" }, { NVME_SC_FEATURE_NOT_CHANGEABLE, "FEATURE NOT CHANGEABLE" }, { NVME_SC_FEATURE_NOT_NS_SPECIFIC, "FEATURE NOT NAMESPACE SPECIFIC" }, { NVME_SC_FW_ACT_REQUIRES_NVMS_RESET, "FIRMWARE ACTIVATION REQUIRES NVM SUBSYSTEM RESET" }, { NVME_SC_FW_ACT_REQUIRES_RESET, "FIRMWARE ACTIVATION REQUIRES RESET" }, { NVME_SC_FW_ACT_REQUIRES_TIME, "FIRMWARE ACTIVATION REQUIRES MAXIMUM TIME VIOLATION" }, { NVME_SC_FW_ACT_PROHIBITED, "FIRMWARE ACTIVATION PROHIBITED" }, { NVME_SC_OVERLAPPING_RANGE, "OVERLAPPING RANGE" }, { NVME_SC_NS_INSUFFICIENT_CAPACITY, "NAMESPACE INSUFFICIENT CAPACITY" }, { NVME_SC_NS_ID_UNAVAILABLE, "NAMESPACE IDENTIFIER UNAVAILABLE" }, { NVME_SC_NS_ALREADY_ATTACHED, "NAMESPACE ALREADY ATTACHED" }, { NVME_SC_NS_IS_PRIVATE, "NAMESPACE IS PRIVATE" }, { NVME_SC_NS_NOT_ATTACHED, "NS NOT ATTACHED" }, { NVME_SC_THIN_PROV_NOT_SUPPORTED, "THIN PROVISIONING NOT SUPPORTED" }, { NVME_SC_CTRLR_LIST_INVALID, "CONTROLLER LIST INVALID" }, { NVME_SC_SELF_TEST_IN_PROGRESS, "DEVICE SELF-TEST IN PROGRESS" }, { NVME_SC_BOOT_PART_WRITE_PROHIB, "BOOT PARTITION WRITE PROHIBITED" }, { NVME_SC_INVALID_CTRLR_ID, "INVALID CONTROLLER IDENTIFIER" }, { NVME_SC_INVALID_SEC_CTRLR_STATE, "INVALID SECONDARY CONTROLLER STATE" }, { NVME_SC_INVALID_NUM_OF_CTRLR_RESRC, "INVALID NUMBER OF CONTROLLER RESOURCES" }, { NVME_SC_INVALID_RESOURCE_ID, "INVALID RESOURCE IDENTIFIER" }, { NVME_SC_SANITIZE_PROHIBITED_WPMRE, "SANITIZE PROHIBITED WRITE PERSISTENT MEMORY REGION ENABLED" }, { NVME_SC_ANA_GROUP_ID_INVALID, "ANA GROUP IDENTIFIED INVALID" }, { NVME_SC_ANA_ATTACH_FAILED, "ANA ATTACH FAILED" }, { NVME_SC_CONFLICTING_ATTRIBUTES, "CONFLICTING ATTRIBUTES" }, { NVME_SC_INVALID_PROTECTION_INFO, "INVALID PROTECTION INFO" }, { NVME_SC_ATTEMPTED_WRITE_TO_RO_PAGE, "WRITE TO RO PAGE" }, { 0xFFFF, "COMMAND SPECIFIC" } }; static struct nvme_status_string media_error_status[] = { { NVME_SC_WRITE_FAULTS, "WRITE FAULTS" }, { NVME_SC_UNRECOVERED_READ_ERROR, "UNRECOVERED READ ERROR" }, { NVME_SC_GUARD_CHECK_ERROR, "GUARD CHECK ERROR" }, { NVME_SC_APPLICATION_TAG_CHECK_ERROR, "APPLICATION TAG CHECK ERROR" }, { NVME_SC_REFERENCE_TAG_CHECK_ERROR, "REFERENCE TAG CHECK ERROR" }, { NVME_SC_COMPARE_FAILURE, "COMPARE FAILURE" }, { NVME_SC_ACCESS_DENIED, "ACCESS DENIED" }, { NVME_SC_DEALLOCATED_OR_UNWRITTEN, "DEALLOCATED OR UNWRITTEN LOGICAL BLOCK" }, { 0xFFFF, "MEDIA ERROR" } }; static struct nvme_status_string path_related_status[] = { { NVME_SC_INTERNAL_PATH_ERROR, "INTERNAL PATH ERROR" }, { NVME_SC_ASYMMETRIC_ACCESS_PERSISTENT_LOSS, "ASYMMETRIC ACCESS PERSISTENT LOSS" }, { NVME_SC_ASYMMETRIC_ACCESS_INACCESSIBLE, "ASYMMETRIC ACCESS INACCESSIBLE" }, { NVME_SC_ASYMMETRIC_ACCESS_TRANSITION, "ASYMMETRIC ACCESS TRANSITION" }, { NVME_SC_CONTROLLER_PATHING_ERROR, "CONTROLLER PATHING ERROR" }, { NVME_SC_HOST_PATHING_ERROR, "HOST PATHING ERROR" }, { NVME_SC_COMMAND_ABOTHED_BY_HOST, "COMMAND ABOTHED BY HOST" }, { 0xFFFF, "PATH RELATED" }, }; static const char * get_status_string(uint16_t sct, uint16_t sc) { struct nvme_status_string *entry; switch (sct) { case NVME_SCT_GENERIC: entry = generic_status; break; case NVME_SCT_COMMAND_SPECIFIC: entry = command_specific_status; break; case NVME_SCT_MEDIA_ERROR: entry = media_error_status; break; case NVME_SCT_PATH_RELATED: entry = path_related_status; break; case NVME_SCT_VENDOR_SPECIFIC: return ("VENDOR SPECIFIC"); default: return ("RESERVED"); } while (entry->sc != 0xFFFF) { if (entry->sc == sc) return (entry->str); entry++; } return (entry->str); } static void nvme_qpair_print_completion(struct nvme_qpair *qpair, struct nvme_completion *cpl) { uint16_t sct, sc; sct = NVME_STATUS_GET_SCT(cpl->status); sc = NVME_STATUS_GET_SC(cpl->status); nvme_printf(qpair->ctrlr, "%s (%02x/%02x) sqid:%d cid:%d cdw0:%x\n", get_status_string(sct, sc), sct, sc, cpl->sqid, cpl->cid, cpl->cdw0); } static bool nvme_completion_is_retry(const struct nvme_completion *cpl) { uint8_t sct, sc, dnr; sct = NVME_STATUS_GET_SCT(cpl->status); sc = NVME_STATUS_GET_SC(cpl->status); dnr = NVME_STATUS_GET_DNR(cpl->status); /* Do Not Retry Bit */ /* * TODO: spec is not clear how commands that are aborted due * to TLER will be marked. So for now, it seems * NAMESPACE_NOT_READY is the only case where we should * look at the DNR bit. Requests failed with ABORTED_BY_REQUEST * set the DNR bit correctly since the driver controls that. */ switch (sct) { case NVME_SCT_GENERIC: switch (sc) { case NVME_SC_ABORTED_BY_REQUEST: case NVME_SC_NAMESPACE_NOT_READY: if (dnr) return (0); else return (1); case NVME_SC_INVALID_OPCODE: case NVME_SC_INVALID_FIELD: case NVME_SC_COMMAND_ID_CONFLICT: case NVME_SC_DATA_TRANSFER_ERROR: case NVME_SC_ABORTED_POWER_LOSS: case NVME_SC_INTERNAL_DEVICE_ERROR: case NVME_SC_ABORTED_SQ_DELETION: case NVME_SC_ABORTED_FAILED_FUSED: case NVME_SC_ABORTED_MISSING_FUSED: case NVME_SC_INVALID_NAMESPACE_OR_FORMAT: case NVME_SC_COMMAND_SEQUENCE_ERROR: case NVME_SC_LBA_OUT_OF_RANGE: case NVME_SC_CAPACITY_EXCEEDED: default: return (0); } case NVME_SCT_COMMAND_SPECIFIC: case NVME_SCT_MEDIA_ERROR: return (0); case NVME_SCT_PATH_RELATED: switch (sc) { case NVME_SC_INTERNAL_PATH_ERROR: if (dnr) return (0); else return (1); default: return (0); } case NVME_SCT_VENDOR_SPECIFIC: default: return (0); } } static void nvme_qpair_complete_tracker(struct nvme_tracker *tr, struct nvme_completion *cpl, error_print_t print_on_error) { struct nvme_qpair * qpair = tr->qpair; struct nvme_request *req; bool retry, error, retriable; req = tr->req; error = nvme_completion_is_error(cpl); retriable = nvme_completion_is_retry(cpl); retry = error && retriable && req->retries < nvme_retry_count; if (retry) qpair->num_retries++; if (error && req->retries >= nvme_retry_count && retriable) qpair->num_failures++; if (error && (print_on_error == ERROR_PRINT_ALL || (!retry && print_on_error == ERROR_PRINT_NO_RETRY))) { nvme_qpair_print_command(qpair, &req->cmd); nvme_qpair_print_completion(qpair, cpl); } qpair->act_tr[cpl->cid] = NULL; KASSERT(cpl->cid == req->cmd.cid, ("cpl cid does not match cmd cid\n")); if (!retry) { if (req->type != NVME_REQUEST_NULL) { bus_dmamap_sync(qpair->dma_tag_payload, tr->payload_dma_map, BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE); } if (req->cb_fn) req->cb_fn(req->cb_arg, cpl); } mtx_lock(&qpair->lock); if (retry) { req->retries++; nvme_qpair_submit_tracker(qpair, tr); } else { if (req->type != NVME_REQUEST_NULL) { bus_dmamap_unload(qpair->dma_tag_payload, tr->payload_dma_map); } nvme_free_request(req); tr->req = NULL; TAILQ_REMOVE(&qpair->outstanding_tr, tr, tailq); TAILQ_INSERT_HEAD(&qpair->free_tr, tr, tailq); /* * If the controller is in the middle of resetting, don't * try to submit queued requests here - let the reset logic * handle that instead. */ if (!STAILQ_EMPTY(&qpair->queued_req) && !qpair->ctrlr->is_resetting) { req = STAILQ_FIRST(&qpair->queued_req); STAILQ_REMOVE_HEAD(&qpair->queued_req, stailq); _nvme_qpair_submit_request(qpair, req); } } mtx_unlock(&qpair->lock); } static void nvme_qpair_manual_complete_tracker( struct nvme_tracker *tr, uint32_t sct, uint32_t sc, uint32_t dnr, error_print_t print_on_error) { struct nvme_completion cpl; memset(&cpl, 0, sizeof(cpl)); struct nvme_qpair * qpair = tr->qpair; cpl.sqid = qpair->id; cpl.cid = tr->cid; cpl.status |= (sct & NVME_STATUS_SCT_MASK) << NVME_STATUS_SCT_SHIFT; cpl.status |= (sc & NVME_STATUS_SC_MASK) << NVME_STATUS_SC_SHIFT; cpl.status |= (dnr & NVME_STATUS_DNR_MASK) << NVME_STATUS_DNR_SHIFT; nvme_qpair_complete_tracker(tr, &cpl, print_on_error); } void nvme_qpair_manual_complete_request(struct nvme_qpair *qpair, struct nvme_request *req, uint32_t sct, uint32_t sc) { struct nvme_completion cpl; bool error; memset(&cpl, 0, sizeof(cpl)); cpl.sqid = qpair->id; cpl.status |= (sct & NVME_STATUS_SCT_MASK) << NVME_STATUS_SCT_SHIFT; cpl.status |= (sc & NVME_STATUS_SC_MASK) << NVME_STATUS_SC_SHIFT; error = nvme_completion_is_error(&cpl); if (error) { nvme_qpair_print_command(qpair, &req->cmd); nvme_qpair_print_completion(qpair, &cpl); } if (req->cb_fn) req->cb_fn(req->cb_arg, &cpl); nvme_free_request(req); } bool nvme_qpair_process_completions(struct nvme_qpair *qpair) { struct nvme_tracker *tr; struct nvme_completion cpl; int done = 0; bool in_panic = dumping || SCHEDULER_STOPPED(); /* * qpair is not enabled, likely because a controller reset is is in * progress. Ignore the interrupt - any I/O that was associated with * this interrupt will get retried when the reset is complete. Any * pending completions for when we're in startup will be completed * as soon as initialization is complete and we start sending commands * to the device. */ - if (qpair->recovery_state != RECOVERY_NONE) + if (qpair->recovery_state != RECOVERY_NONE) { + qpair->num_ignored++; return (false); + } /* * Sanity check initialization. After we reset the hardware, the phase * is defined to be 1. So if we get here with zero prior calls and the * phase is 0, it means that we've lost a race between the * initialization and the ISR running. With the phase wrong, we'll * process a bunch of completions that aren't really completions leading * to a KASSERT below. */ KASSERT(!(qpair->num_intr_handler_calls == 0 && qpair->phase == 0), ("%s: Phase wrong for first interrupt call.", device_get_nameunit(qpair->ctrlr->dev))); qpair->num_intr_handler_calls++; bus_dmamap_sync(qpair->dma_tag, qpair->queuemem_map, BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE); /* * A panic can stop the CPU this routine is running on at any point. If * we're called during a panic, complete the sq_head wrap protocol for * the case where we are interrupted just after the increment at 1 * below, but before we can reset cq_head to zero at 2. Also cope with * the case where we do the zero at 2, but may or may not have done the * phase adjustment at step 3. The panic machinery flushes all pending * memory writes, so we can make these strong ordering assumptions * that would otherwise be unwise if we were racing in real time. */ if (__predict_false(in_panic)) { if (qpair->cq_head == qpair->num_entries) { /* * Here we know that we need to zero cq_head and then negate * the phase, which hasn't been assigned if cq_head isn't * zero due to the atomic_store_rel. */ qpair->cq_head = 0; qpair->phase = !qpair->phase; } else if (qpair->cq_head == 0) { /* * In this case, we know that the assignment at 2 * happened below, but we don't know if it 3 happened or * not. To do this, we look at the last completion * entry and set the phase to the opposite phase * that it has. This gets us back in sync */ cpl = qpair->cpl[qpair->num_entries - 1]; nvme_completion_swapbytes(&cpl); qpair->phase = !NVME_STATUS_GET_P(cpl.status); } } while (1) { uint16_t status; /* * We need to do this dance to avoid a race between the host and * the device where the device overtakes the host while the host * is reading this record, leaving the status field 'new' and * the sqhd and cid fields potentially stale. If the phase * doesn't match, that means status hasn't yet been updated and * we'll get any pending changes next time. It also means that * the phase must be the same the second time. We have to sync * before reading to ensure any bouncing completes. */ status = le16toh(qpair->cpl[qpair->cq_head].status); if (NVME_STATUS_GET_P(status) != qpair->phase) break; bus_dmamap_sync(qpair->dma_tag, qpair->queuemem_map, BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE); cpl = qpair->cpl[qpair->cq_head]; nvme_completion_swapbytes(&cpl); KASSERT( NVME_STATUS_GET_P(status) == NVME_STATUS_GET_P(cpl.status), ("Phase unexpectedly inconsistent")); tr = qpair->act_tr[cpl.cid]; if (tr != NULL) { nvme_qpair_complete_tracker(tr, &cpl, ERROR_PRINT_ALL); qpair->sq_head = cpl.sqhd; done++; } else if (!in_panic) { /* * A missing tracker is normally an error. However, a * panic can stop the CPU this routine is running on * after completing an I/O but before updating * qpair->cq_head at 1 below. Later, we re-enter this * routine to poll I/O associated with the kernel * dump. We find that the tr has been set to null before * calling the completion routine. If it hasn't * completed (or it triggers a panic), then '1' below * won't have updated cq_head. Rather than panic again, * ignore this condition because it's not unexpected. */ nvme_printf(qpair->ctrlr, "cpl does not map to outstanding cmd\n"); /* nvme_dump_completion expects device endianess */ nvme_dump_completion(&qpair->cpl[qpair->cq_head]); KASSERT(0, ("received completion for unknown cmd")); } /* * There's a number of races with the following (see above) when * the system panics. We compensate for each one of them by * using the atomic store to force strong ordering (at least when * viewed in the aftermath of a panic). */ if (++qpair->cq_head == qpair->num_entries) { /* 1 */ atomic_store_rel_int(&qpair->cq_head, 0); /* 2 */ qpair->phase = !qpair->phase; /* 3 */ } bus_space_write_4(qpair->ctrlr->bus_tag, qpair->ctrlr->bus_handle, qpair->cq_hdbl_off, qpair->cq_head); } return (done != 0); } static void nvme_qpair_msi_handler(void *arg) { struct nvme_qpair *qpair = arg; nvme_qpair_process_completions(qpair); } int nvme_qpair_construct(struct nvme_qpair *qpair, uint32_t num_entries, uint32_t num_trackers, struct nvme_controller *ctrlr) { struct nvme_tracker *tr; size_t cmdsz, cplsz, prpsz, allocsz, prpmemsz; uint64_t queuemem_phys, prpmem_phys, list_phys; uint8_t *queuemem, *prpmem, *prp_list; int i, err; qpair->vector = ctrlr->msi_count > 1 ? qpair->id : 0; qpair->num_entries = num_entries; qpair->num_trackers = num_trackers; qpair->ctrlr = ctrlr; mtx_init(&qpair->lock, "nvme qpair lock", NULL, MTX_DEF); /* Note: NVMe PRP format is restricted to 4-byte alignment. */ err = bus_dma_tag_create(bus_get_dma_tag(ctrlr->dev), 4, PAGE_SIZE, BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL, ctrlr->max_xfer_size, btoc(ctrlr->max_xfer_size) + 1, PAGE_SIZE, 0, NULL, NULL, &qpair->dma_tag_payload); if (err != 0) { nvme_printf(ctrlr, "payload tag create failed %d\n", err); goto out; } /* * Each component must be page aligned, and individual PRP lists * cannot cross a page boundary. */ cmdsz = qpair->num_entries * sizeof(struct nvme_command); cmdsz = roundup2(cmdsz, PAGE_SIZE); cplsz = qpair->num_entries * sizeof(struct nvme_completion); cplsz = roundup2(cplsz, PAGE_SIZE); /* * For commands requiring more than 2 PRP entries, one PRP will be * embedded in the command (prp1), and the rest of the PRP entries * will be in a list pointed to by the command (prp2). */ prpsz = sizeof(uint64_t) * btoc(ctrlr->max_xfer_size); prpmemsz = qpair->num_trackers * prpsz; allocsz = cmdsz + cplsz + prpmemsz; err = bus_dma_tag_create(bus_get_dma_tag(ctrlr->dev), PAGE_SIZE, 0, BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL, allocsz, 1, allocsz, 0, NULL, NULL, &qpair->dma_tag); if (err != 0) { nvme_printf(ctrlr, "tag create failed %d\n", err); goto out; } bus_dma_tag_set_domain(qpair->dma_tag, qpair->domain); if (bus_dmamem_alloc(qpair->dma_tag, (void **)&queuemem, BUS_DMA_COHERENT | BUS_DMA_NOWAIT, &qpair->queuemem_map)) { nvme_printf(ctrlr, "failed to alloc qpair memory\n"); goto out; } if (bus_dmamap_load(qpair->dma_tag, qpair->queuemem_map, queuemem, allocsz, nvme_single_map, &queuemem_phys, 0) != 0) { nvme_printf(ctrlr, "failed to load qpair memory\n"); bus_dmamem_free(qpair->dma_tag, qpair->cmd, qpair->queuemem_map); goto out; } qpair->num_cmds = 0; qpair->num_intr_handler_calls = 0; qpair->num_retries = 0; qpair->num_failures = 0; + qpair->num_ignored = 0; qpair->cmd = (struct nvme_command *)queuemem; qpair->cpl = (struct nvme_completion *)(queuemem + cmdsz); prpmem = (uint8_t *)(queuemem + cmdsz + cplsz); qpair->cmd_bus_addr = queuemem_phys; qpair->cpl_bus_addr = queuemem_phys + cmdsz; prpmem_phys = queuemem_phys + cmdsz + cplsz; callout_init(&qpair->timer, 1); qpair->timer_armed = false; qpair->recovery_state = RECOVERY_WAITING; /* * Calcuate the stride of the doorbell register. Many emulators set this * value to correspond to a cache line. However, some hardware has set * it to various small values. */ qpair->sq_tdbl_off = nvme_mmio_offsetof(doorbell[0]) + (qpair->id << (ctrlr->dstrd + 1)); qpair->cq_hdbl_off = nvme_mmio_offsetof(doorbell[0]) + (qpair->id << (ctrlr->dstrd + 1)) + (1 << ctrlr->dstrd); TAILQ_INIT(&qpair->free_tr); TAILQ_INIT(&qpair->outstanding_tr); STAILQ_INIT(&qpair->queued_req); list_phys = prpmem_phys; prp_list = prpmem; for (i = 0; i < qpair->num_trackers; i++) { if (list_phys + prpsz > prpmem_phys + prpmemsz) { qpair->num_trackers = i; break; } /* * Make sure that the PRP list for this tracker doesn't * overflow to another page. */ if (trunc_page(list_phys) != trunc_page(list_phys + prpsz - 1)) { list_phys = roundup2(list_phys, PAGE_SIZE); prp_list = (uint8_t *)roundup2((uintptr_t)prp_list, PAGE_SIZE); } tr = malloc_domainset(sizeof(*tr), M_NVME, DOMAINSET_PREF(qpair->domain), M_ZERO | M_WAITOK); bus_dmamap_create(qpair->dma_tag_payload, 0, &tr->payload_dma_map); tr->cid = i; tr->qpair = qpair; tr->prp = (uint64_t *)prp_list; tr->prp_bus_addr = list_phys; TAILQ_INSERT_HEAD(&qpair->free_tr, tr, tailq); list_phys += prpsz; prp_list += prpsz; } if (qpair->num_trackers == 0) { nvme_printf(ctrlr, "failed to allocate enough trackers\n"); goto out; } qpair->act_tr = malloc_domainset(sizeof(struct nvme_tracker *) * qpair->num_entries, M_NVME, DOMAINSET_PREF(qpair->domain), M_ZERO | M_WAITOK); if (ctrlr->msi_count > 1) { /* * MSI-X vector resource IDs start at 1, so we add one to * the queue's vector to get the corresponding rid to use. */ qpair->rid = qpair->vector + 1; qpair->res = bus_alloc_resource_any(ctrlr->dev, SYS_RES_IRQ, &qpair->rid, RF_ACTIVE); if (qpair->res == NULL) { nvme_printf(ctrlr, "unable to allocate MSI\n"); goto out; } if (bus_setup_intr(ctrlr->dev, qpair->res, INTR_TYPE_MISC | INTR_MPSAFE, NULL, nvme_qpair_msi_handler, qpair, &qpair->tag) != 0) { nvme_printf(ctrlr, "unable to setup MSI\n"); goto out; } if (qpair->id == 0) { bus_describe_intr(ctrlr->dev, qpair->res, qpair->tag, "admin"); } else { bus_describe_intr(ctrlr->dev, qpair->res, qpair->tag, "io%d", qpair->id - 1); } } return (0); out: nvme_qpair_destroy(qpair); return (ENOMEM); } static void nvme_qpair_destroy(struct nvme_qpair *qpair) { struct nvme_tracker *tr; callout_drain(&qpair->timer); if (qpair->tag) { bus_teardown_intr(qpair->ctrlr->dev, qpair->res, qpair->tag); qpair->tag = NULL; } if (qpair->act_tr) { free(qpair->act_tr, M_NVME); qpair->act_tr = NULL; } while (!TAILQ_EMPTY(&qpair->free_tr)) { tr = TAILQ_FIRST(&qpair->free_tr); TAILQ_REMOVE(&qpair->free_tr, tr, tailq); bus_dmamap_destroy(qpair->dma_tag_payload, tr->payload_dma_map); free(tr, M_NVME); } if (qpair->cmd != NULL) { bus_dmamap_unload(qpair->dma_tag, qpair->queuemem_map); bus_dmamem_free(qpair->dma_tag, qpair->cmd, qpair->queuemem_map); qpair->cmd = NULL; } if (qpair->dma_tag) { bus_dma_tag_destroy(qpair->dma_tag); qpair->dma_tag = NULL; } if (qpair->dma_tag_payload) { bus_dma_tag_destroy(qpair->dma_tag_payload); qpair->dma_tag_payload = NULL; } if (mtx_initialized(&qpair->lock)) mtx_destroy(&qpair->lock); if (qpair->res) { bus_release_resource(qpair->ctrlr->dev, SYS_RES_IRQ, rman_get_rid(qpair->res), qpair->res); qpair->res = NULL; } } static void nvme_admin_qpair_abort_aers(struct nvme_qpair *qpair) { struct nvme_tracker *tr; tr = TAILQ_FIRST(&qpair->outstanding_tr); while (tr != NULL) { if (tr->req->cmd.opc == NVME_OPC_ASYNC_EVENT_REQUEST) { nvme_qpair_manual_complete_tracker(tr, NVME_SCT_GENERIC, NVME_SC_ABORTED_SQ_DELETION, 0, ERROR_PRINT_NONE); tr = TAILQ_FIRST(&qpair->outstanding_tr); } else { tr = TAILQ_NEXT(tr, tailq); } } } void nvme_admin_qpair_destroy(struct nvme_qpair *qpair) { nvme_admin_qpair_abort_aers(qpair); nvme_qpair_destroy(qpair); } void nvme_io_qpair_destroy(struct nvme_qpair *qpair) { nvme_qpair_destroy(qpair); } static void nvme_qpair_timeout(void *arg) { struct nvme_qpair *qpair = arg; struct nvme_controller *ctrlr = qpair->ctrlr; struct nvme_tracker *tr; struct nvme_tracker *tr_temp; sbintime_t now; bool idle; uint32_t csts; uint8_t cfs; mtx_lock(&qpair->lock); idle = TAILQ_EMPTY(&qpair->outstanding_tr); again: switch (qpair->recovery_state) { case RECOVERY_NONE: if (idle) break; now = getsbinuptime(); TAILQ_FOREACH_SAFE(tr, &qpair->outstanding_tr, tailq, tr_temp) { if (now > tr->deadline && tr->deadline != 0) { /* * We're now passed our earliest deadline. We * need to do expensive things to cope, but next * time. Flag that and close the door to any * further processing. */ qpair->recovery_state = RECOVERY_START; nvme_printf(ctrlr, "RECOVERY_START %jd vs %jd\n", (uintmax_t)now, (uintmax_t)tr->deadline); break; } } break; case RECOVERY_START: /* * Read csts to get value of cfs - controller fatal status. * If no fatal status, try to call the completion routine, and * if completes transactions, report a missed interrupt and * return (this may need to be rate limited). Otherwise, if * aborts are enabled and the controller is not reporting * fatal status, abort the command. Otherwise, just reset the * controller and hope for the best. */ csts = nvme_mmio_read_4(ctrlr, csts); cfs = (csts >> NVME_CSTS_REG_CFS_SHIFT) & NVME_CSTS_REG_CFS_MASK; if (cfs) { nvme_printf(ctrlr, "Controller in fatal status, resetting\n"); qpair->recovery_state = RECOVERY_RESET; goto again; } mtx_unlock(&qpair->lock); if (nvme_qpair_process_completions(qpair)) { nvme_printf(ctrlr, "Completions present in output without an interrupt\n"); qpair->recovery_state = RECOVERY_NONE; } else { nvme_printf(ctrlr, "timeout with nothing complete, resetting\n"); qpair->recovery_state = RECOVERY_RESET; mtx_lock(&qpair->lock); goto again; } mtx_lock(&qpair->lock); break; case RECOVERY_RESET: /* * If we get here due to a possible surprise hot-unplug event, * then we let nvme_ctrlr_reset confirm and fail the * controller. */ nvme_printf(ctrlr, "Resetting controller due to a timeout%s.\n", cfs ? " and fatal error status" : ""); nvme_printf(ctrlr, "RECOVERY_WAITING\n"); qpair->recovery_state = RECOVERY_WAITING; nvme_ctrlr_reset(ctrlr); break; case RECOVERY_WAITING: nvme_printf(ctrlr, "waiting\n"); break; } /* * Rearm the timeout. */ if (!idle) { callout_schedule(&qpair->timer, hz / 2); } else { qpair->timer_armed = false; } mtx_unlock(&qpair->lock); } /* * Submit the tracker to the hardware. Must already be in the * outstanding queue when called. */ void nvme_qpair_submit_tracker(struct nvme_qpair *qpair, struct nvme_tracker *tr) { struct nvme_request *req; struct nvme_controller *ctrlr; int timeout; mtx_assert(&qpair->lock, MA_OWNED); req = tr->req; req->cmd.cid = tr->cid; qpair->act_tr[tr->cid] = tr; ctrlr = qpair->ctrlr; if (req->timeout) { if (req->cb_fn == nvme_completion_poll_cb) timeout = 1; else timeout = ctrlr->timeout_period; tr->deadline = getsbinuptime() + timeout * SBT_1S; if (!qpair->timer_armed) { qpair->timer_armed = true; callout_reset_on(&qpair->timer, hz / 2, nvme_qpair_timeout, qpair, qpair->cpu); } } else tr->deadline = SBT_MAX; /* Copy the command from the tracker to the submission queue. */ memcpy(&qpair->cmd[qpair->sq_tail], &req->cmd, sizeof(req->cmd)); if (++qpair->sq_tail == qpair->num_entries) qpair->sq_tail = 0; bus_dmamap_sync(qpair->dma_tag, qpair->queuemem_map, BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE); bus_space_write_4(qpair->ctrlr->bus_tag, qpair->ctrlr->bus_handle, qpair->sq_tdbl_off, qpair->sq_tail); qpair->num_cmds++; } static void nvme_payload_map(void *arg, bus_dma_segment_t *seg, int nseg, int error) { struct nvme_tracker *tr = arg; uint32_t cur_nseg; /* * If the mapping operation failed, return immediately. The caller * is responsible for detecting the error status and failing the * tracker manually. */ if (error != 0) { nvme_printf(tr->qpair->ctrlr, "nvme_payload_map err %d\n", error); return; } /* * Note that we specified PAGE_SIZE for alignment and max * segment size when creating the bus dma tags. So here * we can safely just transfer each segment to its * associated PRP entry. */ tr->req->cmd.prp1 = htole64(seg[0].ds_addr); if (nseg == 2) { tr->req->cmd.prp2 = htole64(seg[1].ds_addr); } else if (nseg > 2) { cur_nseg = 1; tr->req->cmd.prp2 = htole64((uint64_t)tr->prp_bus_addr); while (cur_nseg < nseg) { tr->prp[cur_nseg-1] = htole64((uint64_t)seg[cur_nseg].ds_addr); cur_nseg++; } } else { /* * prp2 should not be used by the controller * since there is only one segment, but set * to 0 just to be safe. */ tr->req->cmd.prp2 = 0; } bus_dmamap_sync(tr->qpair->dma_tag_payload, tr->payload_dma_map, BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE); nvme_qpair_submit_tracker(tr->qpair, tr); } static void _nvme_qpair_submit_request(struct nvme_qpair *qpair, struct nvme_request *req) { struct nvme_tracker *tr; int err = 0; mtx_assert(&qpair->lock, MA_OWNED); tr = TAILQ_FIRST(&qpair->free_tr); req->qpair = qpair; if (tr == NULL || qpair->recovery_state != RECOVERY_NONE) { /* * No tracker is available, or the qpair is disabled due to * an in-progress controller-level reset or controller * failure. */ if (qpair->ctrlr->is_failed) { /* * The controller has failed, so fail the request. */ nvme_qpair_manual_complete_request(qpair, req, NVME_SCT_GENERIC, NVME_SC_ABORTED_BY_REQUEST); } else { /* * Put the request on the qpair's request queue to be * processed when a tracker frees up via a command * completion or when the controller reset is * completed. */ STAILQ_INSERT_TAIL(&qpair->queued_req, req, stailq); } return; } TAILQ_REMOVE(&qpair->free_tr, tr, tailq); TAILQ_INSERT_TAIL(&qpair->outstanding_tr, tr, tailq); if (!qpair->timer_armed) tr->deadline = SBT_MAX; tr->req = req; switch (req->type) { case NVME_REQUEST_VADDR: KASSERT(req->payload_size <= qpair->ctrlr->max_xfer_size, ("payload_size (%d) exceeds max_xfer_size (%d)\n", req->payload_size, qpair->ctrlr->max_xfer_size)); err = bus_dmamap_load(tr->qpair->dma_tag_payload, tr->payload_dma_map, req->u.payload, req->payload_size, nvme_payload_map, tr, 0); if (err != 0) nvme_printf(qpair->ctrlr, "bus_dmamap_load returned 0x%x!\n", err); break; case NVME_REQUEST_NULL: nvme_qpair_submit_tracker(tr->qpair, tr); break; case NVME_REQUEST_BIO: KASSERT(req->u.bio->bio_bcount <= qpair->ctrlr->max_xfer_size, ("bio->bio_bcount (%jd) exceeds max_xfer_size (%d)\n", (intmax_t)req->u.bio->bio_bcount, qpair->ctrlr->max_xfer_size)); err = bus_dmamap_load_bio(tr->qpair->dma_tag_payload, tr->payload_dma_map, req->u.bio, nvme_payload_map, tr, 0); if (err != 0) nvme_printf(qpair->ctrlr, "bus_dmamap_load_bio returned 0x%x!\n", err); break; case NVME_REQUEST_CCB: err = bus_dmamap_load_ccb(tr->qpair->dma_tag_payload, tr->payload_dma_map, req->u.payload, nvme_payload_map, tr, 0); if (err != 0) nvme_printf(qpair->ctrlr, "bus_dmamap_load_ccb returned 0x%x!\n", err); break; default: panic("unknown nvme request type 0x%x\n", req->type); break; } if (err != 0) { /* * The dmamap operation failed, so we manually fail the * tracker here with DATA_TRANSFER_ERROR status. * * nvme_qpair_manual_complete_tracker must not be called * with the qpair lock held. */ mtx_unlock(&qpair->lock); nvme_qpair_manual_complete_tracker(tr, NVME_SCT_GENERIC, NVME_SC_DATA_TRANSFER_ERROR, DO_NOT_RETRY, ERROR_PRINT_ALL); mtx_lock(&qpair->lock); } } void nvme_qpair_submit_request(struct nvme_qpair *qpair, struct nvme_request *req) { mtx_lock(&qpair->lock); _nvme_qpair_submit_request(qpair, req); mtx_unlock(&qpair->lock); } static void nvme_qpair_enable(struct nvme_qpair *qpair) { mtx_assert(&qpair->lock, MA_OWNED); qpair->recovery_state = RECOVERY_NONE; } void nvme_qpair_reset(struct nvme_qpair *qpair) { qpair->sq_head = qpair->sq_tail = qpair->cq_head = 0; /* * First time through the completion queue, HW will set phase * bit on completions to 1. So set this to 1 here, indicating * we're looking for a 1 to know which entries have completed. * we'll toggle the bit each time when the completion queue * rolls over. */ qpair->phase = 1; memset(qpair->cmd, 0, qpair->num_entries * sizeof(struct nvme_command)); memset(qpair->cpl, 0, qpair->num_entries * sizeof(struct nvme_completion)); } void nvme_admin_qpair_enable(struct nvme_qpair *qpair) { struct nvme_tracker *tr; struct nvme_tracker *tr_temp; /* * Manually abort each outstanding admin command. Do not retry * admin commands found here, since they will be left over from * a controller reset and its likely the context in which the * command was issued no longer applies. */ TAILQ_FOREACH_SAFE(tr, &qpair->outstanding_tr, tailq, tr_temp) { nvme_printf(qpair->ctrlr, "aborting outstanding admin command\n"); nvme_qpair_manual_complete_tracker(tr, NVME_SCT_GENERIC, NVME_SC_ABORTED_BY_REQUEST, DO_NOT_RETRY, ERROR_PRINT_ALL); } mtx_lock(&qpair->lock); nvme_qpair_enable(qpair); mtx_unlock(&qpair->lock); } void nvme_io_qpair_enable(struct nvme_qpair *qpair) { STAILQ_HEAD(, nvme_request) temp; struct nvme_tracker *tr; struct nvme_tracker *tr_temp; struct nvme_request *req; /* * Manually abort each outstanding I/O. This normally results in a * retry, unless the retry count on the associated request has * reached its limit. */ TAILQ_FOREACH_SAFE(tr, &qpair->outstanding_tr, tailq, tr_temp) { nvme_printf(qpair->ctrlr, "aborting outstanding i/o\n"); nvme_qpair_manual_complete_tracker(tr, NVME_SCT_GENERIC, NVME_SC_ABORTED_BY_REQUEST, 0, ERROR_PRINT_NO_RETRY); } mtx_lock(&qpair->lock); nvme_qpair_enable(qpair); STAILQ_INIT(&temp); STAILQ_SWAP(&qpair->queued_req, &temp, nvme_request); while (!STAILQ_EMPTY(&temp)) { req = STAILQ_FIRST(&temp); STAILQ_REMOVE_HEAD(&temp, stailq); nvme_printf(qpair->ctrlr, "resubmitting queued i/o\n"); nvme_qpair_print_command(qpair, &req->cmd); _nvme_qpair_submit_request(qpair, req); } mtx_unlock(&qpair->lock); } static void nvme_qpair_disable(struct nvme_qpair *qpair) { struct nvme_tracker *tr, *tr_temp; mtx_lock(&qpair->lock); qpair->recovery_state = RECOVERY_WAITING; TAILQ_FOREACH_SAFE(tr, &qpair->outstanding_tr, tailq, tr_temp) { tr->deadline = SBT_MAX; } mtx_unlock(&qpair->lock); } void nvme_admin_qpair_disable(struct nvme_qpair *qpair) { nvme_qpair_disable(qpair); nvme_admin_qpair_abort_aers(qpair); } void nvme_io_qpair_disable(struct nvme_qpair *qpair) { nvme_qpair_disable(qpair); } void nvme_qpair_fail(struct nvme_qpair *qpair) { struct nvme_tracker *tr; struct nvme_request *req; if (!mtx_initialized(&qpair->lock)) return; mtx_lock(&qpair->lock); while (!STAILQ_EMPTY(&qpair->queued_req)) { req = STAILQ_FIRST(&qpair->queued_req); STAILQ_REMOVE_HEAD(&qpair->queued_req, stailq); nvme_printf(qpair->ctrlr, "failing queued i/o\n"); mtx_unlock(&qpair->lock); nvme_qpair_manual_complete_request(qpair, req, NVME_SCT_GENERIC, NVME_SC_ABORTED_BY_REQUEST); mtx_lock(&qpair->lock); } /* Manually abort each outstanding I/O. */ while (!TAILQ_EMPTY(&qpair->outstanding_tr)) { tr = TAILQ_FIRST(&qpair->outstanding_tr); /* * Do not remove the tracker. The abort_tracker path will * do that for us. */ nvme_printf(qpair->ctrlr, "failing outstanding i/o\n"); mtx_unlock(&qpair->lock); nvme_qpair_manual_complete_tracker(tr, NVME_SCT_GENERIC, NVME_SC_ABORTED_BY_REQUEST, DO_NOT_RETRY, ERROR_PRINT_ALL); mtx_lock(&qpair->lock); } mtx_unlock(&qpair->lock); } diff --git a/sys/dev/nvme/nvme_sysctl.c b/sys/dev/nvme/nvme_sysctl.c index 58db38dc373d..cfcc814f44d2 100644 --- a/sys/dev/nvme/nvme_sysctl.c +++ b/sys/dev/nvme/nvme_sysctl.c @@ -1,363 +1,387 @@ /*- * SPDX-License-Identifier: BSD-2-Clause-FreeBSD * * Copyright (C) 2012-2016 Intel Corporation * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ #include __FBSDID("$FreeBSD$"); #include "opt_nvme.h" #include #include #include #include "nvme_private.h" #ifndef NVME_USE_NVD #define NVME_USE_NVD 1 #endif int nvme_use_nvd = NVME_USE_NVD; bool nvme_verbose_cmd_dump = false; SYSCTL_NODE(_hw, OID_AUTO, nvme, CTLFLAG_RD | CTLFLAG_MPSAFE, 0, "NVMe sysctl tunables"); SYSCTL_INT(_hw_nvme, OID_AUTO, use_nvd, CTLFLAG_RDTUN, &nvme_use_nvd, 1, "1 = Create NVD devices, 0 = Create NDA devices"); SYSCTL_BOOL(_hw_nvme, OID_AUTO, verbose_cmd_dump, CTLFLAG_RWTUN, &nvme_verbose_cmd_dump, 0, "enable verbose command printting when a command fails"); static void nvme_dump_queue(struct nvme_qpair *qpair) { struct nvme_completion *cpl; struct nvme_command *cmd; int i; printf("id:%04Xh phase:%d\n", qpair->id, qpair->phase); printf("Completion queue:\n"); for (i = 0; i < qpair->num_entries; i++) { cpl = &qpair->cpl[i]; printf("%05d: ", i); nvme_dump_completion(cpl); } printf("Submission queue:\n"); for (i = 0; i < qpair->num_entries; i++) { cmd = &qpair->cmd[i]; printf("%05d: ", i); nvme_dump_command(cmd); } } static int nvme_sysctl_dump_debug(SYSCTL_HANDLER_ARGS) { struct nvme_qpair *qpair = arg1; uint32_t val = 0; int error = sysctl_handle_int(oidp, &val, 0, req); if (error) return (error); if (val != 0) nvme_dump_queue(qpair); return (0); } static int nvme_sysctl_int_coal_time(SYSCTL_HANDLER_ARGS) { struct nvme_controller *ctrlr = arg1; uint32_t oldval = ctrlr->int_coal_time; int error = sysctl_handle_int(oidp, &ctrlr->int_coal_time, 0, req); if (error) return (error); if (oldval != ctrlr->int_coal_time) nvme_ctrlr_cmd_set_interrupt_coalescing(ctrlr, ctrlr->int_coal_time, ctrlr->int_coal_threshold, NULL, NULL); return (0); } static int nvme_sysctl_int_coal_threshold(SYSCTL_HANDLER_ARGS) { struct nvme_controller *ctrlr = arg1; uint32_t oldval = ctrlr->int_coal_threshold; int error = sysctl_handle_int(oidp, &ctrlr->int_coal_threshold, 0, req); if (error) return (error); if (oldval != ctrlr->int_coal_threshold) nvme_ctrlr_cmd_set_interrupt_coalescing(ctrlr, ctrlr->int_coal_time, ctrlr->int_coal_threshold, NULL, NULL); return (0); } static int nvme_sysctl_timeout_period(SYSCTL_HANDLER_ARGS) { struct nvme_controller *ctrlr = arg1; uint32_t newval = ctrlr->timeout_period; int error = sysctl_handle_int(oidp, &newval, 0, req); if (error || (req->newptr == NULL)) return (error); if (newval > NVME_MAX_TIMEOUT_PERIOD || newval < NVME_MIN_TIMEOUT_PERIOD) { return (EINVAL); } else { ctrlr->timeout_period = newval; } return (0); } static void nvme_qpair_reset_stats(struct nvme_qpair *qpair) { /* * Reset the values. Due to sanity checks in * nvme_qpair_process_completions, we reset the number of interrupt * calls to 1. */ qpair->num_cmds = 0; qpair->num_intr_handler_calls = 1; qpair->num_retries = 0; qpair->num_failures = 0; + qpair->num_ignored = 0; } static int nvme_sysctl_num_cmds(SYSCTL_HANDLER_ARGS) { struct nvme_controller *ctrlr = arg1; int64_t num_cmds = 0; int i; num_cmds = ctrlr->adminq.num_cmds; for (i = 0; i < ctrlr->num_io_queues; i++) num_cmds += ctrlr->ioq[i].num_cmds; return (sysctl_handle_64(oidp, &num_cmds, 0, req)); } static int nvme_sysctl_num_intr_handler_calls(SYSCTL_HANDLER_ARGS) { struct nvme_controller *ctrlr = arg1; int64_t num_intr_handler_calls = 0; int i; num_intr_handler_calls = ctrlr->adminq.num_intr_handler_calls; for (i = 0; i < ctrlr->num_io_queues; i++) num_intr_handler_calls += ctrlr->ioq[i].num_intr_handler_calls; return (sysctl_handle_64(oidp, &num_intr_handler_calls, 0, req)); } static int nvme_sysctl_num_retries(SYSCTL_HANDLER_ARGS) { struct nvme_controller *ctrlr = arg1; int64_t num_retries = 0; int i; num_retries = ctrlr->adminq.num_retries; for (i = 0; i < ctrlr->num_io_queues; i++) num_retries += ctrlr->ioq[i].num_retries; return (sysctl_handle_64(oidp, &num_retries, 0, req)); } static int nvme_sysctl_num_failures(SYSCTL_HANDLER_ARGS) { struct nvme_controller *ctrlr = arg1; int64_t num_failures = 0; int i; num_failures = ctrlr->adminq.num_failures; for (i = 0; i < ctrlr->num_io_queues; i++) num_failures += ctrlr->ioq[i].num_failures; return (sysctl_handle_64(oidp, &num_failures, 0, req)); } +static int +nvme_sysctl_num_ignored(SYSCTL_HANDLER_ARGS) +{ + struct nvme_controller *ctrlr = arg1; + int64_t num_ignored = 0; + int i; + + num_ignored = ctrlr->adminq.num_ignored; + + for (i = 0; i < ctrlr->num_io_queues; i++) + num_ignored += ctrlr->ioq[i].num_ignored; + + return (sysctl_handle_64(oidp, &num_ignored, 0, req)); +} + static int nvme_sysctl_reset_stats(SYSCTL_HANDLER_ARGS) { struct nvme_controller *ctrlr = arg1; uint32_t i, val = 0; int error = sysctl_handle_int(oidp, &val, 0, req); if (error) return (error); if (val != 0) { nvme_qpair_reset_stats(&ctrlr->adminq); for (i = 0; i < ctrlr->num_io_queues; i++) nvme_qpair_reset_stats(&ctrlr->ioq[i]); } return (0); } static void nvme_sysctl_initialize_queue(struct nvme_qpair *qpair, struct sysctl_ctx_list *ctrlr_ctx, struct sysctl_oid *que_tree) { struct sysctl_oid_list *que_list = SYSCTL_CHILDREN(que_tree); SYSCTL_ADD_UINT(ctrlr_ctx, que_list, OID_AUTO, "num_entries", CTLFLAG_RD, &qpair->num_entries, 0, "Number of entries in hardware queue"); SYSCTL_ADD_UINT(ctrlr_ctx, que_list, OID_AUTO, "num_trackers", CTLFLAG_RD, &qpair->num_trackers, 0, "Number of trackers pre-allocated for this queue pair"); SYSCTL_ADD_UINT(ctrlr_ctx, que_list, OID_AUTO, "sq_head", CTLFLAG_RD, &qpair->sq_head, 0, "Current head of submission queue (as observed by driver)"); SYSCTL_ADD_UINT(ctrlr_ctx, que_list, OID_AUTO, "sq_tail", CTLFLAG_RD, &qpair->sq_tail, 0, "Current tail of submission queue (as observed by driver)"); SYSCTL_ADD_UINT(ctrlr_ctx, que_list, OID_AUTO, "cq_head", CTLFLAG_RD, &qpair->cq_head, 0, "Current head of completion queue (as observed by driver)"); SYSCTL_ADD_QUAD(ctrlr_ctx, que_list, OID_AUTO, "num_cmds", CTLFLAG_RD, &qpair->num_cmds, "Number of commands submitted"); SYSCTL_ADD_QUAD(ctrlr_ctx, que_list, OID_AUTO, "num_intr_handler_calls", CTLFLAG_RD, &qpair->num_intr_handler_calls, "Number of times interrupt handler was invoked (will typically be " "less than number of actual interrupts generated due to " "coalescing)"); SYSCTL_ADD_QUAD(ctrlr_ctx, que_list, OID_AUTO, "num_retries", CTLFLAG_RD, &qpair->num_retries, "Number of commands retried"); SYSCTL_ADD_QUAD(ctrlr_ctx, que_list, OID_AUTO, "num_failures", CTLFLAG_RD, &qpair->num_failures, "Number of commands ending in failure after all retries"); + SYSCTL_ADD_QUAD(ctrlr_ctx, que_list, OID_AUTO, "num_ignored", + CTLFLAG_RD, &qpair->num_ignored, + "Number of interrupts posted, but were administratively ignored"); SYSCTL_ADD_PROC(ctrlr_ctx, que_list, OID_AUTO, "dump_debug", CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_MPSAFE, qpair, 0, nvme_sysctl_dump_debug, "IU", "Dump debug data"); } void nvme_sysctl_initialize_ctrlr(struct nvme_controller *ctrlr) { struct sysctl_ctx_list *ctrlr_ctx; struct sysctl_oid *ctrlr_tree, *que_tree; struct sysctl_oid_list *ctrlr_list; #define QUEUE_NAME_LENGTH 16 char queue_name[QUEUE_NAME_LENGTH]; int i; ctrlr_ctx = device_get_sysctl_ctx(ctrlr->dev); ctrlr_tree = device_get_sysctl_tree(ctrlr->dev); ctrlr_list = SYSCTL_CHILDREN(ctrlr_tree); SYSCTL_ADD_UINT(ctrlr_ctx, ctrlr_list, OID_AUTO, "num_io_queues", CTLFLAG_RD, &ctrlr->num_io_queues, 0, "Number of I/O queue pairs"); SYSCTL_ADD_PROC(ctrlr_ctx, ctrlr_list, OID_AUTO, "int_coal_time", CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_MPSAFE, ctrlr, 0, nvme_sysctl_int_coal_time, "IU", "Interrupt coalescing timeout (in microseconds)"); SYSCTL_ADD_PROC(ctrlr_ctx, ctrlr_list, OID_AUTO, "int_coal_threshold", CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_MPSAFE, ctrlr, 0, nvme_sysctl_int_coal_threshold, "IU", "Interrupt coalescing threshold"); SYSCTL_ADD_PROC(ctrlr_ctx, ctrlr_list, OID_AUTO, "timeout_period", CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_MPSAFE, ctrlr, 0, nvme_sysctl_timeout_period, "IU", "Timeout period (in seconds)"); SYSCTL_ADD_PROC(ctrlr_ctx, ctrlr_list, OID_AUTO, "num_cmds", CTLTYPE_S64 | CTLFLAG_RD | CTLFLAG_MPSAFE, ctrlr, 0, nvme_sysctl_num_cmds, "IU", "Number of commands submitted"); SYSCTL_ADD_PROC(ctrlr_ctx, ctrlr_list, OID_AUTO, "num_intr_handler_calls", CTLTYPE_S64 | CTLFLAG_RD | CTLFLAG_MPSAFE, ctrlr, 0, nvme_sysctl_num_intr_handler_calls, "IU", "Number of times interrupt handler was invoked (will " "typically be less than number of actual interrupts " "generated due to coalescing)"); SYSCTL_ADD_PROC(ctrlr_ctx, ctrlr_list, OID_AUTO, "num_retries", CTLTYPE_S64 | CTLFLAG_RD | CTLFLAG_MPSAFE, ctrlr, 0, nvme_sysctl_num_retries, "IU", "Number of commands retried"); SYSCTL_ADD_PROC(ctrlr_ctx, ctrlr_list, OID_AUTO, "num_failures", CTLTYPE_S64 | CTLFLAG_RD | CTLFLAG_MPSAFE, ctrlr, 0, nvme_sysctl_num_failures, "IU", "Number of commands ending in failure after all retries"); + SYSCTL_ADD_PROC(ctrlr_ctx, ctrlr_list, OID_AUTO, + "num_ignored", CTLTYPE_S64 | CTLFLAG_RD | CTLFLAG_MPSAFE, + ctrlr, 0, nvme_sysctl_num_ignored, "IU", + "Number of interrupts ignored administratively"); + SYSCTL_ADD_PROC(ctrlr_ctx, ctrlr_list, OID_AUTO, "reset_stats", CTLTYPE_UINT | CTLFLAG_RW | CTLFLAG_MPSAFE, ctrlr, 0, nvme_sysctl_reset_stats, "IU", "Reset statistics to zero"); que_tree = SYSCTL_ADD_NODE(ctrlr_ctx, ctrlr_list, OID_AUTO, "adminq", CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, "Admin Queue"); nvme_sysctl_initialize_queue(&ctrlr->adminq, ctrlr_ctx, que_tree); for (i = 0; i < ctrlr->num_io_queues; i++) { snprintf(queue_name, QUEUE_NAME_LENGTH, "ioq%d", i); que_tree = SYSCTL_ADD_NODE(ctrlr_ctx, ctrlr_list, OID_AUTO, queue_name, CTLFLAG_RD | CTLFLAG_MPSAFE, NULL, "IO Queue"); nvme_sysctl_initialize_queue(&ctrlr->ioq[i], ctrlr_ctx, que_tree); } }