157dacad5SJay Sternberg /* 257dacad5SJay Sternberg * NVM Express device driver 357dacad5SJay Sternberg * Copyright (c) 2011-2014, Intel Corporation. 457dacad5SJay Sternberg * 557dacad5SJay Sternberg * This program is free software; you can redistribute it and/or modify it 657dacad5SJay Sternberg * under the terms and conditions of the GNU General Public License, 757dacad5SJay Sternberg * version 2, as published by the Free Software Foundation. 857dacad5SJay Sternberg * 957dacad5SJay Sternberg * This program is distributed in the hope it will be useful, but WITHOUT 1057dacad5SJay Sternberg * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or 1157dacad5SJay Sternberg * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for 1257dacad5SJay Sternberg * more details. 1357dacad5SJay Sternberg */ 1457dacad5SJay Sternberg 15a0a3408eSKeith Busch #include <linux/aer.h> 1657dacad5SJay Sternberg #include <linux/bitops.h> 1757dacad5SJay Sternberg #include <linux/blkdev.h> 1857dacad5SJay Sternberg #include <linux/blk-mq.h> 19dca51e78SChristoph Hellwig #include <linux/blk-mq-pci.h> 2057dacad5SJay Sternberg #include <linux/cpu.h> 2157dacad5SJay Sternberg #include <linux/delay.h> 2257dacad5SJay Sternberg #include <linux/errno.h> 2357dacad5SJay Sternberg #include <linux/fs.h> 2457dacad5SJay Sternberg #include <linux/genhd.h> 2557dacad5SJay Sternberg #include <linux/hdreg.h> 2657dacad5SJay Sternberg #include <linux/idr.h> 2757dacad5SJay Sternberg #include <linux/init.h> 2857dacad5SJay Sternberg #include <linux/interrupt.h> 2957dacad5SJay Sternberg #include <linux/io.h> 3057dacad5SJay Sternberg #include <linux/kdev_t.h> 3157dacad5SJay Sternberg #include <linux/kernel.h> 3257dacad5SJay Sternberg #include <linux/mm.h> 3357dacad5SJay Sternberg #include <linux/module.h> 3457dacad5SJay Sternberg #include <linux/moduleparam.h> 3577bf25eaSKeith Busch #include <linux/mutex.h> 3657dacad5SJay Sternberg #include <linux/pci.h> 3757dacad5SJay Sternberg #include <linux/poison.h> 3857dacad5SJay Sternberg #include <linux/ptrace.h> 3957dacad5SJay Sternberg #include <linux/sched.h> 4057dacad5SJay Sternberg #include <linux/slab.h> 4157dacad5SJay Sternberg #include <linux/t10-pi.h> 422d55cd5fSChristoph Hellwig #include <linux/timer.h> 4357dacad5SJay Sternberg #include <linux/types.h> 449cf5c095SLinus Torvalds #include <linux/io-64-nonatomic-lo-hi.h> 451d277a63SKeith Busch #include <asm/unaligned.h> 46a98e58e5SScott Bauer #include <linux/sed-opal.h> 4757dacad5SJay Sternberg 4857dacad5SJay Sternberg #include "nvme.h" 4957dacad5SJay Sternberg 5057dacad5SJay Sternberg #define NVME_Q_DEPTH 1024 5157dacad5SJay Sternberg #define NVME_AQ_DEPTH 256 5257dacad5SJay Sternberg #define SQ_SIZE(depth) (depth * sizeof(struct nvme_command)) 5357dacad5SJay Sternberg #define CQ_SIZE(depth) (depth * sizeof(struct nvme_completion)) 5457dacad5SJay Sternberg 55adf68f21SChristoph Hellwig /* 56adf68f21SChristoph Hellwig * We handle AEN commands ourselves and don't even let the 57adf68f21SChristoph Hellwig * block layer know about them. 58adf68f21SChristoph Hellwig */ 59f866fc42SChristoph Hellwig #define NVME_AQ_BLKMQ_DEPTH (NVME_AQ_DEPTH - NVME_NR_AERS) 60adf68f21SChristoph Hellwig 6157dacad5SJay Sternberg static int use_threaded_interrupts; 6257dacad5SJay Sternberg module_param(use_threaded_interrupts, int, 0); 6357dacad5SJay Sternberg 6457dacad5SJay Sternberg static bool use_cmb_sqes = true; 6557dacad5SJay Sternberg module_param(use_cmb_sqes, bool, 0644); 6657dacad5SJay Sternberg MODULE_PARM_DESC(use_cmb_sqes, "use controller's memory buffer for I/O SQes"); 6757dacad5SJay Sternberg 6857dacad5SJay Sternberg static struct workqueue_struct *nvme_workq; 6957dacad5SJay Sternberg 701c63dc66SChristoph Hellwig struct nvme_dev; 711c63dc66SChristoph Hellwig struct nvme_queue; 7257dacad5SJay Sternberg 7357dacad5SJay Sternberg static int nvme_reset(struct nvme_dev *dev); 74a0fa9647SJens Axboe static void nvme_process_cq(struct nvme_queue *nvmeq); 75a5cdb68cSKeith Busch static void nvme_dev_disable(struct nvme_dev *dev, bool shutdown); 7657dacad5SJay Sternberg 7757dacad5SJay Sternberg /* 781c63dc66SChristoph Hellwig * Represents an NVM Express device. Each nvme_dev is a PCI function. 791c63dc66SChristoph Hellwig */ 801c63dc66SChristoph Hellwig struct nvme_dev { 811c63dc66SChristoph Hellwig struct nvme_queue **queues; 821c63dc66SChristoph Hellwig struct blk_mq_tag_set tagset; 831c63dc66SChristoph Hellwig struct blk_mq_tag_set admin_tagset; 841c63dc66SChristoph Hellwig u32 __iomem *dbs; 851c63dc66SChristoph Hellwig struct device *dev; 861c63dc66SChristoph Hellwig struct dma_pool *prp_page_pool; 871c63dc66SChristoph Hellwig struct dma_pool *prp_small_pool; 881c63dc66SChristoph Hellwig unsigned queue_count; 891c63dc66SChristoph Hellwig unsigned online_queues; 901c63dc66SChristoph Hellwig unsigned max_qid; 911c63dc66SChristoph Hellwig int q_depth; 921c63dc66SChristoph Hellwig u32 db_stride; 931c63dc66SChristoph Hellwig void __iomem *bar; 941c63dc66SChristoph Hellwig struct work_struct reset_work; 955c8809e6SChristoph Hellwig struct work_struct remove_work; 962d55cd5fSChristoph Hellwig struct timer_list watchdog_timer; 9777bf25eaSKeith Busch struct mutex shutdown_lock; 981c63dc66SChristoph Hellwig bool subsystem; 991c63dc66SChristoph Hellwig void __iomem *cmb; 1001c63dc66SChristoph Hellwig dma_addr_t cmb_dma_addr; 1011c63dc66SChristoph Hellwig u64 cmb_size; 1021c63dc66SChristoph Hellwig u32 cmbsz; 103202021c1SStephen Bates u32 cmbloc; 1041c63dc66SChristoph Hellwig struct nvme_ctrl ctrl; 105db3cbfffSKeith Busch struct completion ioq_wait; 10657dacad5SJay Sternberg }; 10757dacad5SJay Sternberg 1081c63dc66SChristoph Hellwig static inline struct nvme_dev *to_nvme_dev(struct nvme_ctrl *ctrl) 1091c63dc66SChristoph Hellwig { 1101c63dc66SChristoph Hellwig return container_of(ctrl, struct nvme_dev, ctrl); 1111c63dc66SChristoph Hellwig } 1121c63dc66SChristoph Hellwig 11357dacad5SJay Sternberg /* 11457dacad5SJay Sternberg * An NVM Express queue. Each device has at least two (one for admin 11557dacad5SJay Sternberg * commands and one for I/O commands). 11657dacad5SJay Sternberg */ 11757dacad5SJay Sternberg struct nvme_queue { 11857dacad5SJay Sternberg struct device *q_dmadev; 11957dacad5SJay Sternberg struct nvme_dev *dev; 12057dacad5SJay Sternberg char irqname[24]; /* nvme4294967295-65535\0 */ 12157dacad5SJay Sternberg spinlock_t q_lock; 12257dacad5SJay Sternberg struct nvme_command *sq_cmds; 12357dacad5SJay Sternberg struct nvme_command __iomem *sq_cmds_io; 12457dacad5SJay Sternberg volatile struct nvme_completion *cqes; 12557dacad5SJay Sternberg struct blk_mq_tags **tags; 12657dacad5SJay Sternberg dma_addr_t sq_dma_addr; 12757dacad5SJay Sternberg dma_addr_t cq_dma_addr; 12857dacad5SJay Sternberg u32 __iomem *q_db; 12957dacad5SJay Sternberg u16 q_depth; 13057dacad5SJay Sternberg s16 cq_vector; 13157dacad5SJay Sternberg u16 sq_tail; 13257dacad5SJay Sternberg u16 cq_head; 13357dacad5SJay Sternberg u16 qid; 13457dacad5SJay Sternberg u8 cq_phase; 13557dacad5SJay Sternberg u8 cqe_seen; 13657dacad5SJay Sternberg }; 13757dacad5SJay Sternberg 13857dacad5SJay Sternberg /* 13971bd150cSChristoph Hellwig * The nvme_iod describes the data in an I/O, including the list of PRP 14071bd150cSChristoph Hellwig * entries. You can't see it in this data structure because C doesn't let 141f4800d6dSChristoph Hellwig * me express that. Use nvme_init_iod to ensure there's enough space 14271bd150cSChristoph Hellwig * allocated to store the PRP list. 14371bd150cSChristoph Hellwig */ 14471bd150cSChristoph Hellwig struct nvme_iod { 145d49187e9SChristoph Hellwig struct nvme_request req; 146f4800d6dSChristoph Hellwig struct nvme_queue *nvmeq; 147f4800d6dSChristoph Hellwig int aborted; 14871bd150cSChristoph Hellwig int npages; /* In the PRP list. 0 means small pool in use */ 14971bd150cSChristoph Hellwig int nents; /* Used in scatterlist */ 15071bd150cSChristoph Hellwig int length; /* Of data, in bytes */ 15171bd150cSChristoph Hellwig dma_addr_t first_dma; 152bf684057SChristoph Hellwig struct scatterlist meta_sg; /* metadata requires single contiguous buffer */ 153f4800d6dSChristoph Hellwig struct scatterlist *sg; 154f4800d6dSChristoph Hellwig struct scatterlist inline_sg[0]; 15557dacad5SJay Sternberg }; 15657dacad5SJay Sternberg 15757dacad5SJay Sternberg /* 15857dacad5SJay Sternberg * Check we didin't inadvertently grow the command struct 15957dacad5SJay Sternberg */ 16057dacad5SJay Sternberg static inline void _nvme_check_size(void) 16157dacad5SJay Sternberg { 16257dacad5SJay Sternberg BUILD_BUG_ON(sizeof(struct nvme_rw_command) != 64); 16357dacad5SJay Sternberg BUILD_BUG_ON(sizeof(struct nvme_create_cq) != 64); 16457dacad5SJay Sternberg BUILD_BUG_ON(sizeof(struct nvme_create_sq) != 64); 16557dacad5SJay Sternberg BUILD_BUG_ON(sizeof(struct nvme_delete_queue) != 64); 16657dacad5SJay Sternberg BUILD_BUG_ON(sizeof(struct nvme_features) != 64); 16757dacad5SJay Sternberg BUILD_BUG_ON(sizeof(struct nvme_format_cmd) != 64); 16857dacad5SJay Sternberg BUILD_BUG_ON(sizeof(struct nvme_abort_cmd) != 64); 16957dacad5SJay Sternberg BUILD_BUG_ON(sizeof(struct nvme_command) != 64); 17057dacad5SJay Sternberg BUILD_BUG_ON(sizeof(struct nvme_id_ctrl) != 4096); 17157dacad5SJay Sternberg BUILD_BUG_ON(sizeof(struct nvme_id_ns) != 4096); 17257dacad5SJay Sternberg BUILD_BUG_ON(sizeof(struct nvme_lba_range_type) != 64); 17357dacad5SJay Sternberg BUILD_BUG_ON(sizeof(struct nvme_smart_log) != 512); 17457dacad5SJay Sternberg } 17557dacad5SJay Sternberg 17657dacad5SJay Sternberg /* 17757dacad5SJay Sternberg * Max size of iod being embedded in the request payload 17857dacad5SJay Sternberg */ 17957dacad5SJay Sternberg #define NVME_INT_PAGES 2 1805fd4ce1bSChristoph Hellwig #define NVME_INT_BYTES(dev) (NVME_INT_PAGES * (dev)->ctrl.page_size) 18157dacad5SJay Sternberg 18257dacad5SJay Sternberg /* 18357dacad5SJay Sternberg * Will slightly overestimate the number of pages needed. This is OK 18457dacad5SJay Sternberg * as it only leads to a small amount of wasted memory for the lifetime of 18557dacad5SJay Sternberg * the I/O. 18657dacad5SJay Sternberg */ 18757dacad5SJay Sternberg static int nvme_npages(unsigned size, struct nvme_dev *dev) 18857dacad5SJay Sternberg { 1895fd4ce1bSChristoph Hellwig unsigned nprps = DIV_ROUND_UP(size + dev->ctrl.page_size, 1905fd4ce1bSChristoph Hellwig dev->ctrl.page_size); 19157dacad5SJay Sternberg return DIV_ROUND_UP(8 * nprps, PAGE_SIZE - 8); 19257dacad5SJay Sternberg } 19357dacad5SJay Sternberg 194f4800d6dSChristoph Hellwig static unsigned int nvme_iod_alloc_size(struct nvme_dev *dev, 195f4800d6dSChristoph Hellwig unsigned int size, unsigned int nseg) 196f4800d6dSChristoph Hellwig { 197f4800d6dSChristoph Hellwig return sizeof(__le64 *) * nvme_npages(size, dev) + 198f4800d6dSChristoph Hellwig sizeof(struct scatterlist) * nseg; 199f4800d6dSChristoph Hellwig } 200f4800d6dSChristoph Hellwig 20157dacad5SJay Sternberg static unsigned int nvme_cmd_size(struct nvme_dev *dev) 20257dacad5SJay Sternberg { 203f4800d6dSChristoph Hellwig return sizeof(struct nvme_iod) + 204f4800d6dSChristoph Hellwig nvme_iod_alloc_size(dev, NVME_INT_BYTES(dev), NVME_INT_PAGES); 20557dacad5SJay Sternberg } 20657dacad5SJay Sternberg 207dca51e78SChristoph Hellwig static int nvmeq_irq(struct nvme_queue *nvmeq) 208dca51e78SChristoph Hellwig { 209dca51e78SChristoph Hellwig return pci_irq_vector(to_pci_dev(nvmeq->dev->dev), nvmeq->cq_vector); 210dca51e78SChristoph Hellwig } 211dca51e78SChristoph Hellwig 21257dacad5SJay Sternberg static int nvme_admin_init_hctx(struct blk_mq_hw_ctx *hctx, void *data, 21357dacad5SJay Sternberg unsigned int hctx_idx) 21457dacad5SJay Sternberg { 21557dacad5SJay Sternberg struct nvme_dev *dev = data; 21657dacad5SJay Sternberg struct nvme_queue *nvmeq = dev->queues[0]; 21757dacad5SJay Sternberg 21857dacad5SJay Sternberg WARN_ON(hctx_idx != 0); 21957dacad5SJay Sternberg WARN_ON(dev->admin_tagset.tags[0] != hctx->tags); 22057dacad5SJay Sternberg WARN_ON(nvmeq->tags); 22157dacad5SJay Sternberg 22257dacad5SJay Sternberg hctx->driver_data = nvmeq; 22357dacad5SJay Sternberg nvmeq->tags = &dev->admin_tagset.tags[0]; 22457dacad5SJay Sternberg return 0; 22557dacad5SJay Sternberg } 22657dacad5SJay Sternberg 22757dacad5SJay Sternberg static void nvme_admin_exit_hctx(struct blk_mq_hw_ctx *hctx, unsigned int hctx_idx) 22857dacad5SJay Sternberg { 22957dacad5SJay Sternberg struct nvme_queue *nvmeq = hctx->driver_data; 23057dacad5SJay Sternberg 23157dacad5SJay Sternberg nvmeq->tags = NULL; 23257dacad5SJay Sternberg } 23357dacad5SJay Sternberg 23457dacad5SJay Sternberg static int nvme_admin_init_request(void *data, struct request *req, 23557dacad5SJay Sternberg unsigned int hctx_idx, unsigned int rq_idx, 23657dacad5SJay Sternberg unsigned int numa_node) 23757dacad5SJay Sternberg { 23857dacad5SJay Sternberg struct nvme_dev *dev = data; 239f4800d6dSChristoph Hellwig struct nvme_iod *iod = blk_mq_rq_to_pdu(req); 24057dacad5SJay Sternberg struct nvme_queue *nvmeq = dev->queues[0]; 24157dacad5SJay Sternberg 24257dacad5SJay Sternberg BUG_ON(!nvmeq); 243f4800d6dSChristoph Hellwig iod->nvmeq = nvmeq; 24457dacad5SJay Sternberg return 0; 24557dacad5SJay Sternberg } 24657dacad5SJay Sternberg 24757dacad5SJay Sternberg static int nvme_init_hctx(struct blk_mq_hw_ctx *hctx, void *data, 24857dacad5SJay Sternberg unsigned int hctx_idx) 24957dacad5SJay Sternberg { 25057dacad5SJay Sternberg struct nvme_dev *dev = data; 25157dacad5SJay Sternberg struct nvme_queue *nvmeq = dev->queues[hctx_idx + 1]; 25257dacad5SJay Sternberg 25357dacad5SJay Sternberg if (!nvmeq->tags) 25457dacad5SJay Sternberg nvmeq->tags = &dev->tagset.tags[hctx_idx]; 25557dacad5SJay Sternberg 25657dacad5SJay Sternberg WARN_ON(dev->tagset.tags[hctx_idx] != hctx->tags); 25757dacad5SJay Sternberg hctx->driver_data = nvmeq; 25857dacad5SJay Sternberg return 0; 25957dacad5SJay Sternberg } 26057dacad5SJay Sternberg 26157dacad5SJay Sternberg static int nvme_init_request(void *data, struct request *req, 26257dacad5SJay Sternberg unsigned int hctx_idx, unsigned int rq_idx, 26357dacad5SJay Sternberg unsigned int numa_node) 26457dacad5SJay Sternberg { 26557dacad5SJay Sternberg struct nvme_dev *dev = data; 266f4800d6dSChristoph Hellwig struct nvme_iod *iod = blk_mq_rq_to_pdu(req); 26757dacad5SJay Sternberg struct nvme_queue *nvmeq = dev->queues[hctx_idx + 1]; 26857dacad5SJay Sternberg 26957dacad5SJay Sternberg BUG_ON(!nvmeq); 270f4800d6dSChristoph Hellwig iod->nvmeq = nvmeq; 27157dacad5SJay Sternberg return 0; 27257dacad5SJay Sternberg } 27357dacad5SJay Sternberg 274dca51e78SChristoph Hellwig static int nvme_pci_map_queues(struct blk_mq_tag_set *set) 275dca51e78SChristoph Hellwig { 276dca51e78SChristoph Hellwig struct nvme_dev *dev = set->driver_data; 277dca51e78SChristoph Hellwig 278dca51e78SChristoph Hellwig return blk_mq_pci_map_queues(set, to_pci_dev(dev->dev)); 279dca51e78SChristoph Hellwig } 280dca51e78SChristoph Hellwig 28157dacad5SJay Sternberg /** 282adf68f21SChristoph Hellwig * __nvme_submit_cmd() - Copy a command into a queue and ring the doorbell 28357dacad5SJay Sternberg * @nvmeq: The queue to use 28457dacad5SJay Sternberg * @cmd: The command to send 28557dacad5SJay Sternberg * 28657dacad5SJay Sternberg * Safe to use from interrupt context 28757dacad5SJay Sternberg */ 28857dacad5SJay Sternberg static void __nvme_submit_cmd(struct nvme_queue *nvmeq, 28957dacad5SJay Sternberg struct nvme_command *cmd) 29057dacad5SJay Sternberg { 29157dacad5SJay Sternberg u16 tail = nvmeq->sq_tail; 29257dacad5SJay Sternberg 29357dacad5SJay Sternberg if (nvmeq->sq_cmds_io) 29457dacad5SJay Sternberg memcpy_toio(&nvmeq->sq_cmds_io[tail], cmd, sizeof(*cmd)); 29557dacad5SJay Sternberg else 29657dacad5SJay Sternberg memcpy(&nvmeq->sq_cmds[tail], cmd, sizeof(*cmd)); 29757dacad5SJay Sternberg 29857dacad5SJay Sternberg if (++tail == nvmeq->q_depth) 29957dacad5SJay Sternberg tail = 0; 30057dacad5SJay Sternberg writel(tail, nvmeq->q_db); 30157dacad5SJay Sternberg nvmeq->sq_tail = tail; 30257dacad5SJay Sternberg } 30357dacad5SJay Sternberg 304f4800d6dSChristoph Hellwig static __le64 **iod_list(struct request *req) 30557dacad5SJay Sternberg { 306f4800d6dSChristoph Hellwig struct nvme_iod *iod = blk_mq_rq_to_pdu(req); 307f9d03f96SChristoph Hellwig return (__le64 **)(iod->sg + blk_rq_nr_phys_segments(req)); 30857dacad5SJay Sternberg } 30957dacad5SJay Sternberg 31058b45602SMing Lin static int nvme_init_iod(struct request *rq, unsigned size, 31158b45602SMing Lin struct nvme_dev *dev) 31257dacad5SJay Sternberg { 313f4800d6dSChristoph Hellwig struct nvme_iod *iod = blk_mq_rq_to_pdu(rq); 314f9d03f96SChristoph Hellwig int nseg = blk_rq_nr_phys_segments(rq); 315f4800d6dSChristoph Hellwig 316f4800d6dSChristoph Hellwig if (nseg > NVME_INT_PAGES || size > NVME_INT_BYTES(dev)) { 317f4800d6dSChristoph Hellwig iod->sg = kmalloc(nvme_iod_alloc_size(dev, size, nseg), GFP_ATOMIC); 318f4800d6dSChristoph Hellwig if (!iod->sg) 319f4800d6dSChristoph Hellwig return BLK_MQ_RQ_QUEUE_BUSY; 320f4800d6dSChristoph Hellwig } else { 321f4800d6dSChristoph Hellwig iod->sg = iod->inline_sg; 32257dacad5SJay Sternberg } 32357dacad5SJay Sternberg 324f4800d6dSChristoph Hellwig iod->aborted = 0; 32557dacad5SJay Sternberg iod->npages = -1; 32657dacad5SJay Sternberg iod->nents = 0; 327f4800d6dSChristoph Hellwig iod->length = size; 328f80ec966SKeith Busch 329e8064021SChristoph Hellwig if (!(rq->rq_flags & RQF_DONTPREP)) { 330f80ec966SKeith Busch rq->retries = 0; 331e8064021SChristoph Hellwig rq->rq_flags |= RQF_DONTPREP; 332f80ec966SKeith Busch } 333bac0000aSOmar Sandoval return BLK_MQ_RQ_QUEUE_OK; 33457dacad5SJay Sternberg } 33557dacad5SJay Sternberg 336f4800d6dSChristoph Hellwig static void nvme_free_iod(struct nvme_dev *dev, struct request *req) 33757dacad5SJay Sternberg { 338f4800d6dSChristoph Hellwig struct nvme_iod *iod = blk_mq_rq_to_pdu(req); 3395fd4ce1bSChristoph Hellwig const int last_prp = dev->ctrl.page_size / 8 - 1; 34057dacad5SJay Sternberg int i; 341f4800d6dSChristoph Hellwig __le64 **list = iod_list(req); 34257dacad5SJay Sternberg dma_addr_t prp_dma = iod->first_dma; 34357dacad5SJay Sternberg 34457dacad5SJay Sternberg if (iod->npages == 0) 34557dacad5SJay Sternberg dma_pool_free(dev->prp_small_pool, list[0], prp_dma); 34657dacad5SJay Sternberg for (i = 0; i < iod->npages; i++) { 34757dacad5SJay Sternberg __le64 *prp_list = list[i]; 34857dacad5SJay Sternberg dma_addr_t next_prp_dma = le64_to_cpu(prp_list[last_prp]); 34957dacad5SJay Sternberg dma_pool_free(dev->prp_page_pool, prp_list, prp_dma); 35057dacad5SJay Sternberg prp_dma = next_prp_dma; 35157dacad5SJay Sternberg } 35257dacad5SJay Sternberg 353f4800d6dSChristoph Hellwig if (iod->sg != iod->inline_sg) 354f4800d6dSChristoph Hellwig kfree(iod->sg); 35557dacad5SJay Sternberg } 35657dacad5SJay Sternberg 35757dacad5SJay Sternberg #ifdef CONFIG_BLK_DEV_INTEGRITY 35857dacad5SJay Sternberg static void nvme_dif_prep(u32 p, u32 v, struct t10_pi_tuple *pi) 35957dacad5SJay Sternberg { 36057dacad5SJay Sternberg if (be32_to_cpu(pi->ref_tag) == v) 36157dacad5SJay Sternberg pi->ref_tag = cpu_to_be32(p); 36257dacad5SJay Sternberg } 36357dacad5SJay Sternberg 36457dacad5SJay Sternberg static void nvme_dif_complete(u32 p, u32 v, struct t10_pi_tuple *pi) 36557dacad5SJay Sternberg { 36657dacad5SJay Sternberg if (be32_to_cpu(pi->ref_tag) == p) 36757dacad5SJay Sternberg pi->ref_tag = cpu_to_be32(v); 36857dacad5SJay Sternberg } 36957dacad5SJay Sternberg 37057dacad5SJay Sternberg /** 37157dacad5SJay Sternberg * nvme_dif_remap - remaps ref tags to bip seed and physical lba 37257dacad5SJay Sternberg * 37357dacad5SJay Sternberg * The virtual start sector is the one that was originally submitted by the 37457dacad5SJay Sternberg * block layer. Due to partitioning, MD/DM cloning, etc. the actual physical 37557dacad5SJay Sternberg * start sector may be different. Remap protection information to match the 37657dacad5SJay Sternberg * physical LBA on writes, and back to the original seed on reads. 37757dacad5SJay Sternberg * 37857dacad5SJay Sternberg * Type 0 and 3 do not have a ref tag, so no remapping required. 37957dacad5SJay Sternberg */ 38057dacad5SJay Sternberg static void nvme_dif_remap(struct request *req, 38157dacad5SJay Sternberg void (*dif_swap)(u32 p, u32 v, struct t10_pi_tuple *pi)) 38257dacad5SJay Sternberg { 38357dacad5SJay Sternberg struct nvme_ns *ns = req->rq_disk->private_data; 38457dacad5SJay Sternberg struct bio_integrity_payload *bip; 38557dacad5SJay Sternberg struct t10_pi_tuple *pi; 38657dacad5SJay Sternberg void *p, *pmap; 38757dacad5SJay Sternberg u32 i, nlb, ts, phys, virt; 38857dacad5SJay Sternberg 38957dacad5SJay Sternberg if (!ns->pi_type || ns->pi_type == NVME_NS_DPS_PI_TYPE3) 39057dacad5SJay Sternberg return; 39157dacad5SJay Sternberg 39257dacad5SJay Sternberg bip = bio_integrity(req->bio); 39357dacad5SJay Sternberg if (!bip) 39457dacad5SJay Sternberg return; 39557dacad5SJay Sternberg 39657dacad5SJay Sternberg pmap = kmap_atomic(bip->bip_vec->bv_page) + bip->bip_vec->bv_offset; 39757dacad5SJay Sternberg 39857dacad5SJay Sternberg p = pmap; 39957dacad5SJay Sternberg virt = bip_get_seed(bip); 40057dacad5SJay Sternberg phys = nvme_block_nr(ns, blk_rq_pos(req)); 40157dacad5SJay Sternberg nlb = (blk_rq_bytes(req) >> ns->lba_shift); 402ac6fc48cSDan Williams ts = ns->disk->queue->integrity.tuple_size; 40357dacad5SJay Sternberg 40457dacad5SJay Sternberg for (i = 0; i < nlb; i++, virt++, phys++) { 40557dacad5SJay Sternberg pi = (struct t10_pi_tuple *)p; 40657dacad5SJay Sternberg dif_swap(phys, virt, pi); 40757dacad5SJay Sternberg p += ts; 40857dacad5SJay Sternberg } 40957dacad5SJay Sternberg kunmap_atomic(pmap); 41057dacad5SJay Sternberg } 41157dacad5SJay Sternberg #else /* CONFIG_BLK_DEV_INTEGRITY */ 41257dacad5SJay Sternberg static void nvme_dif_remap(struct request *req, 41357dacad5SJay Sternberg void (*dif_swap)(u32 p, u32 v, struct t10_pi_tuple *pi)) 41457dacad5SJay Sternberg { 41557dacad5SJay Sternberg } 41657dacad5SJay Sternberg static void nvme_dif_prep(u32 p, u32 v, struct t10_pi_tuple *pi) 41757dacad5SJay Sternberg { 41857dacad5SJay Sternberg } 41957dacad5SJay Sternberg static void nvme_dif_complete(u32 p, u32 v, struct t10_pi_tuple *pi) 42057dacad5SJay Sternberg { 42157dacad5SJay Sternberg } 42257dacad5SJay Sternberg #endif 42357dacad5SJay Sternberg 424f4800d6dSChristoph Hellwig static bool nvme_setup_prps(struct nvme_dev *dev, struct request *req, 42569d2b571SChristoph Hellwig int total_len) 42657dacad5SJay Sternberg { 427f4800d6dSChristoph Hellwig struct nvme_iod *iod = blk_mq_rq_to_pdu(req); 42857dacad5SJay Sternberg struct dma_pool *pool; 42957dacad5SJay Sternberg int length = total_len; 43057dacad5SJay Sternberg struct scatterlist *sg = iod->sg; 43157dacad5SJay Sternberg int dma_len = sg_dma_len(sg); 43257dacad5SJay Sternberg u64 dma_addr = sg_dma_address(sg); 4335fd4ce1bSChristoph Hellwig u32 page_size = dev->ctrl.page_size; 43457dacad5SJay Sternberg int offset = dma_addr & (page_size - 1); 43557dacad5SJay Sternberg __le64 *prp_list; 436f4800d6dSChristoph Hellwig __le64 **list = iod_list(req); 43757dacad5SJay Sternberg dma_addr_t prp_dma; 43857dacad5SJay Sternberg int nprps, i; 43957dacad5SJay Sternberg 44057dacad5SJay Sternberg length -= (page_size - offset); 44157dacad5SJay Sternberg if (length <= 0) 44269d2b571SChristoph Hellwig return true; 44357dacad5SJay Sternberg 44457dacad5SJay Sternberg dma_len -= (page_size - offset); 44557dacad5SJay Sternberg if (dma_len) { 44657dacad5SJay Sternberg dma_addr += (page_size - offset); 44757dacad5SJay Sternberg } else { 44857dacad5SJay Sternberg sg = sg_next(sg); 44957dacad5SJay Sternberg dma_addr = sg_dma_address(sg); 45057dacad5SJay Sternberg dma_len = sg_dma_len(sg); 45157dacad5SJay Sternberg } 45257dacad5SJay Sternberg 45357dacad5SJay Sternberg if (length <= page_size) { 45457dacad5SJay Sternberg iod->first_dma = dma_addr; 45569d2b571SChristoph Hellwig return true; 45657dacad5SJay Sternberg } 45757dacad5SJay Sternberg 45857dacad5SJay Sternberg nprps = DIV_ROUND_UP(length, page_size); 45957dacad5SJay Sternberg if (nprps <= (256 / 8)) { 46057dacad5SJay Sternberg pool = dev->prp_small_pool; 46157dacad5SJay Sternberg iod->npages = 0; 46257dacad5SJay Sternberg } else { 46357dacad5SJay Sternberg pool = dev->prp_page_pool; 46457dacad5SJay Sternberg iod->npages = 1; 46557dacad5SJay Sternberg } 46657dacad5SJay Sternberg 46769d2b571SChristoph Hellwig prp_list = dma_pool_alloc(pool, GFP_ATOMIC, &prp_dma); 46857dacad5SJay Sternberg if (!prp_list) { 46957dacad5SJay Sternberg iod->first_dma = dma_addr; 47057dacad5SJay Sternberg iod->npages = -1; 47169d2b571SChristoph Hellwig return false; 47257dacad5SJay Sternberg } 47357dacad5SJay Sternberg list[0] = prp_list; 47457dacad5SJay Sternberg iod->first_dma = prp_dma; 47557dacad5SJay Sternberg i = 0; 47657dacad5SJay Sternberg for (;;) { 47757dacad5SJay Sternberg if (i == page_size >> 3) { 47857dacad5SJay Sternberg __le64 *old_prp_list = prp_list; 47969d2b571SChristoph Hellwig prp_list = dma_pool_alloc(pool, GFP_ATOMIC, &prp_dma); 48057dacad5SJay Sternberg if (!prp_list) 48169d2b571SChristoph Hellwig return false; 48257dacad5SJay Sternberg list[iod->npages++] = prp_list; 48357dacad5SJay Sternberg prp_list[0] = old_prp_list[i - 1]; 48457dacad5SJay Sternberg old_prp_list[i - 1] = cpu_to_le64(prp_dma); 48557dacad5SJay Sternberg i = 1; 48657dacad5SJay Sternberg } 48757dacad5SJay Sternberg prp_list[i++] = cpu_to_le64(dma_addr); 48857dacad5SJay Sternberg dma_len -= page_size; 48957dacad5SJay Sternberg dma_addr += page_size; 49057dacad5SJay Sternberg length -= page_size; 49157dacad5SJay Sternberg if (length <= 0) 49257dacad5SJay Sternberg break; 49357dacad5SJay Sternberg if (dma_len > 0) 49457dacad5SJay Sternberg continue; 49557dacad5SJay Sternberg BUG_ON(dma_len < 0); 49657dacad5SJay Sternberg sg = sg_next(sg); 49757dacad5SJay Sternberg dma_addr = sg_dma_address(sg); 49857dacad5SJay Sternberg dma_len = sg_dma_len(sg); 49957dacad5SJay Sternberg } 50057dacad5SJay Sternberg 50169d2b571SChristoph Hellwig return true; 50257dacad5SJay Sternberg } 50357dacad5SJay Sternberg 504f4800d6dSChristoph Hellwig static int nvme_map_data(struct nvme_dev *dev, struct request *req, 50503b5929eSMing Lin unsigned size, struct nvme_command *cmnd) 50657dacad5SJay Sternberg { 507f4800d6dSChristoph Hellwig struct nvme_iod *iod = blk_mq_rq_to_pdu(req); 508ba1ca37eSChristoph Hellwig struct request_queue *q = req->q; 509ba1ca37eSChristoph Hellwig enum dma_data_direction dma_dir = rq_data_dir(req) ? 510ba1ca37eSChristoph Hellwig DMA_TO_DEVICE : DMA_FROM_DEVICE; 511ba1ca37eSChristoph Hellwig int ret = BLK_MQ_RQ_QUEUE_ERROR; 51257dacad5SJay Sternberg 513f9d03f96SChristoph Hellwig sg_init_table(iod->sg, blk_rq_nr_phys_segments(req)); 514ba1ca37eSChristoph Hellwig iod->nents = blk_rq_map_sg(q, req, iod->sg); 515ba1ca37eSChristoph Hellwig if (!iod->nents) 516ba1ca37eSChristoph Hellwig goto out; 517ba1ca37eSChristoph Hellwig 518ba1ca37eSChristoph Hellwig ret = BLK_MQ_RQ_QUEUE_BUSY; 5192b6b535dSMauricio Faria de Oliveira if (!dma_map_sg_attrs(dev->dev, iod->sg, iod->nents, dma_dir, 5202b6b535dSMauricio Faria de Oliveira DMA_ATTR_NO_WARN)) 521ba1ca37eSChristoph Hellwig goto out; 522ba1ca37eSChristoph Hellwig 52303b5929eSMing Lin if (!nvme_setup_prps(dev, req, size)) 524ba1ca37eSChristoph Hellwig goto out_unmap; 525ba1ca37eSChristoph Hellwig 526ba1ca37eSChristoph Hellwig ret = BLK_MQ_RQ_QUEUE_ERROR; 527ba1ca37eSChristoph Hellwig if (blk_integrity_rq(req)) { 528ba1ca37eSChristoph Hellwig if (blk_rq_count_integrity_sg(q, req->bio) != 1) 529ba1ca37eSChristoph Hellwig goto out_unmap; 530ba1ca37eSChristoph Hellwig 531bf684057SChristoph Hellwig sg_init_table(&iod->meta_sg, 1); 532bf684057SChristoph Hellwig if (blk_rq_map_integrity_sg(q, req->bio, &iod->meta_sg) != 1) 533ba1ca37eSChristoph Hellwig goto out_unmap; 534ba1ca37eSChristoph Hellwig 535ba1ca37eSChristoph Hellwig if (rq_data_dir(req)) 536ba1ca37eSChristoph Hellwig nvme_dif_remap(req, nvme_dif_prep); 537ba1ca37eSChristoph Hellwig 538bf684057SChristoph Hellwig if (!dma_map_sg(dev->dev, &iod->meta_sg, 1, dma_dir)) 539ba1ca37eSChristoph Hellwig goto out_unmap; 54057dacad5SJay Sternberg } 54157dacad5SJay Sternberg 542eb793e2cSChristoph Hellwig cmnd->rw.dptr.prp1 = cpu_to_le64(sg_dma_address(iod->sg)); 543eb793e2cSChristoph Hellwig cmnd->rw.dptr.prp2 = cpu_to_le64(iod->first_dma); 544ba1ca37eSChristoph Hellwig if (blk_integrity_rq(req)) 545bf684057SChristoph Hellwig cmnd->rw.metadata = cpu_to_le64(sg_dma_address(&iod->meta_sg)); 546ba1ca37eSChristoph Hellwig return BLK_MQ_RQ_QUEUE_OK; 547ba1ca37eSChristoph Hellwig 548ba1ca37eSChristoph Hellwig out_unmap: 549ba1ca37eSChristoph Hellwig dma_unmap_sg(dev->dev, iod->sg, iod->nents, dma_dir); 550ba1ca37eSChristoph Hellwig out: 551ba1ca37eSChristoph Hellwig return ret; 55257dacad5SJay Sternberg } 55357dacad5SJay Sternberg 554f4800d6dSChristoph Hellwig static void nvme_unmap_data(struct nvme_dev *dev, struct request *req) 555d4f6c3abSChristoph Hellwig { 556f4800d6dSChristoph Hellwig struct nvme_iod *iod = blk_mq_rq_to_pdu(req); 557d4f6c3abSChristoph Hellwig enum dma_data_direction dma_dir = rq_data_dir(req) ? 558d4f6c3abSChristoph Hellwig DMA_TO_DEVICE : DMA_FROM_DEVICE; 559d4f6c3abSChristoph Hellwig 560d4f6c3abSChristoph Hellwig if (iod->nents) { 561d4f6c3abSChristoph Hellwig dma_unmap_sg(dev->dev, iod->sg, iod->nents, dma_dir); 562d4f6c3abSChristoph Hellwig if (blk_integrity_rq(req)) { 563d4f6c3abSChristoph Hellwig if (!rq_data_dir(req)) 564d4f6c3abSChristoph Hellwig nvme_dif_remap(req, nvme_dif_complete); 565bf684057SChristoph Hellwig dma_unmap_sg(dev->dev, &iod->meta_sg, 1, dma_dir); 566d4f6c3abSChristoph Hellwig } 567d4f6c3abSChristoph Hellwig } 568d4f6c3abSChristoph Hellwig 569f9d03f96SChristoph Hellwig nvme_cleanup_cmd(req); 570f4800d6dSChristoph Hellwig nvme_free_iod(dev, req); 57157dacad5SJay Sternberg } 57257dacad5SJay Sternberg 57357dacad5SJay Sternberg /* 57457dacad5SJay Sternberg * NOTE: ns is NULL when called on the admin queue. 57557dacad5SJay Sternberg */ 57657dacad5SJay Sternberg static int nvme_queue_rq(struct blk_mq_hw_ctx *hctx, 57757dacad5SJay Sternberg const struct blk_mq_queue_data *bd) 57857dacad5SJay Sternberg { 57957dacad5SJay Sternberg struct nvme_ns *ns = hctx->queue->queuedata; 58057dacad5SJay Sternberg struct nvme_queue *nvmeq = hctx->driver_data; 58157dacad5SJay Sternberg struct nvme_dev *dev = nvmeq->dev; 58257dacad5SJay Sternberg struct request *req = bd->rq; 583ba1ca37eSChristoph Hellwig struct nvme_command cmnd; 58458b45602SMing Lin unsigned map_len; 585ba1ca37eSChristoph Hellwig int ret = BLK_MQ_RQ_QUEUE_OK; 58657dacad5SJay Sternberg 58757dacad5SJay Sternberg /* 58857dacad5SJay Sternberg * If formated with metadata, require the block layer provide a buffer 58957dacad5SJay Sternberg * unless this namespace is formated such that the metadata can be 59057dacad5SJay Sternberg * stripped/generated by the controller with PRACT=1. 59157dacad5SJay Sternberg */ 59257dacad5SJay Sternberg if (ns && ns->ms && !blk_integrity_rq(req)) { 59357dacad5SJay Sternberg if (!(ns->pi_type && ns->ms == 8) && 59457dacad5SJay Sternberg req->cmd_type != REQ_TYPE_DRV_PRIV) { 595eee417b0SChristoph Hellwig blk_mq_end_request(req, -EFAULT); 59657dacad5SJay Sternberg return BLK_MQ_RQ_QUEUE_OK; 59757dacad5SJay Sternberg } 59857dacad5SJay Sternberg } 59957dacad5SJay Sternberg 600f9d03f96SChristoph Hellwig ret = nvme_setup_cmd(ns, req, &cmnd); 601bac0000aSOmar Sandoval if (ret != BLK_MQ_RQ_QUEUE_OK) 602f4800d6dSChristoph Hellwig return ret; 60357dacad5SJay Sternberg 604f9d03f96SChristoph Hellwig map_len = nvme_map_len(req); 605f9d03f96SChristoph Hellwig ret = nvme_init_iod(req, map_len, dev); 606bac0000aSOmar Sandoval if (ret != BLK_MQ_RQ_QUEUE_OK) 607f9d03f96SChristoph Hellwig goto out_free_cmd; 60857dacad5SJay Sternberg 609f9d03f96SChristoph Hellwig if (blk_rq_nr_phys_segments(req)) 61003b5929eSMing Lin ret = nvme_map_data(dev, req, map_len, &cmnd); 611ba1ca37eSChristoph Hellwig 612bac0000aSOmar Sandoval if (ret != BLK_MQ_RQ_QUEUE_OK) 613f9d03f96SChristoph Hellwig goto out_cleanup_iod; 614ba1ca37eSChristoph Hellwig 615aae239e1SChristoph Hellwig blk_mq_start_request(req); 616ba1ca37eSChristoph Hellwig 617ba1ca37eSChristoph Hellwig spin_lock_irq(&nvmeq->q_lock); 618ae1fba20SKeith Busch if (unlikely(nvmeq->cq_vector < 0)) { 61969d9a99cSKeith Busch if (ns && !test_bit(NVME_NS_DEAD, &ns->flags)) 620ae1fba20SKeith Busch ret = BLK_MQ_RQ_QUEUE_BUSY; 62169d9a99cSKeith Busch else 62269d9a99cSKeith Busch ret = BLK_MQ_RQ_QUEUE_ERROR; 623ae1fba20SKeith Busch spin_unlock_irq(&nvmeq->q_lock); 624f9d03f96SChristoph Hellwig goto out_cleanup_iod; 625ae1fba20SKeith Busch } 626ba1ca37eSChristoph Hellwig __nvme_submit_cmd(nvmeq, &cmnd); 62757dacad5SJay Sternberg nvme_process_cq(nvmeq); 62857dacad5SJay Sternberg spin_unlock_irq(&nvmeq->q_lock); 62957dacad5SJay Sternberg return BLK_MQ_RQ_QUEUE_OK; 630f9d03f96SChristoph Hellwig out_cleanup_iod: 631f4800d6dSChristoph Hellwig nvme_free_iod(dev, req); 632f9d03f96SChristoph Hellwig out_free_cmd: 633f9d03f96SChristoph Hellwig nvme_cleanup_cmd(req); 634ba1ca37eSChristoph Hellwig return ret; 63557dacad5SJay Sternberg } 63657dacad5SJay Sternberg 637eee417b0SChristoph Hellwig static void nvme_complete_rq(struct request *req) 638eee417b0SChristoph Hellwig { 639f4800d6dSChristoph Hellwig struct nvme_iod *iod = blk_mq_rq_to_pdu(req); 640f4800d6dSChristoph Hellwig struct nvme_dev *dev = iod->nvmeq->dev; 641eee417b0SChristoph Hellwig int error = 0; 642eee417b0SChristoph Hellwig 643f4800d6dSChristoph Hellwig nvme_unmap_data(dev, req); 644eee417b0SChristoph Hellwig 645eee417b0SChristoph Hellwig if (unlikely(req->errors)) { 646eee417b0SChristoph Hellwig if (nvme_req_needs_retry(req, req->errors)) { 647f80ec966SKeith Busch req->retries++; 648eee417b0SChristoph Hellwig nvme_requeue_req(req); 649eee417b0SChristoph Hellwig return; 650eee417b0SChristoph Hellwig } 651eee417b0SChristoph Hellwig 652eee417b0SChristoph Hellwig if (req->cmd_type == REQ_TYPE_DRV_PRIV) 653eee417b0SChristoph Hellwig error = req->errors; 654eee417b0SChristoph Hellwig else 655eee417b0SChristoph Hellwig error = nvme_error_status(req->errors); 656eee417b0SChristoph Hellwig } 657eee417b0SChristoph Hellwig 658f4800d6dSChristoph Hellwig if (unlikely(iod->aborted)) { 6591b3c47c1SSagi Grimberg dev_warn(dev->ctrl.device, 660eee417b0SChristoph Hellwig "completing aborted command with status: %04x\n", 661eee417b0SChristoph Hellwig req->errors); 662eee417b0SChristoph Hellwig } 663eee417b0SChristoph Hellwig 664eee417b0SChristoph Hellwig blk_mq_end_request(req, error); 66557dacad5SJay Sternberg } 66657dacad5SJay Sternberg 667d783e0bdSMarta Rybczynska /* We read the CQE phase first to check if the rest of the entry is valid */ 668d783e0bdSMarta Rybczynska static inline bool nvme_cqe_valid(struct nvme_queue *nvmeq, u16 head, 669d783e0bdSMarta Rybczynska u16 phase) 670d783e0bdSMarta Rybczynska { 671d783e0bdSMarta Rybczynska return (le16_to_cpu(nvmeq->cqes[head].status) & 1) == phase; 672d783e0bdSMarta Rybczynska } 673d783e0bdSMarta Rybczynska 674a0fa9647SJens Axboe static void __nvme_process_cq(struct nvme_queue *nvmeq, unsigned int *tag) 67557dacad5SJay Sternberg { 67657dacad5SJay Sternberg u16 head, phase; 67757dacad5SJay Sternberg 67857dacad5SJay Sternberg head = nvmeq->cq_head; 67957dacad5SJay Sternberg phase = nvmeq->cq_phase; 68057dacad5SJay Sternberg 681d783e0bdSMarta Rybczynska while (nvme_cqe_valid(nvmeq, head, phase)) { 68257dacad5SJay Sternberg struct nvme_completion cqe = nvmeq->cqes[head]; 683eee417b0SChristoph Hellwig struct request *req; 684adf68f21SChristoph Hellwig 68557dacad5SJay Sternberg if (++head == nvmeq->q_depth) { 68657dacad5SJay Sternberg head = 0; 68757dacad5SJay Sternberg phase = !phase; 68857dacad5SJay Sternberg } 689adf68f21SChristoph Hellwig 690a0fa9647SJens Axboe if (tag && *tag == cqe.command_id) 691a0fa9647SJens Axboe *tag = -1; 692adf68f21SChristoph Hellwig 693aae239e1SChristoph Hellwig if (unlikely(cqe.command_id >= nvmeq->q_depth)) { 6941b3c47c1SSagi Grimberg dev_warn(nvmeq->dev->ctrl.device, 695aae239e1SChristoph Hellwig "invalid id %d completed on queue %d\n", 696aae239e1SChristoph Hellwig cqe.command_id, le16_to_cpu(cqe.sq_id)); 697aae239e1SChristoph Hellwig continue; 698aae239e1SChristoph Hellwig } 699aae239e1SChristoph Hellwig 700adf68f21SChristoph Hellwig /* 701adf68f21SChristoph Hellwig * AEN requests are special as they don't time out and can 702adf68f21SChristoph Hellwig * survive any kind of queue freeze and often don't respond to 703adf68f21SChristoph Hellwig * aborts. We don't even bother to allocate a struct request 704adf68f21SChristoph Hellwig * for them but rather special case them here. 705adf68f21SChristoph Hellwig */ 706adf68f21SChristoph Hellwig if (unlikely(nvmeq->qid == 0 && 707adf68f21SChristoph Hellwig cqe.command_id >= NVME_AQ_BLKMQ_DEPTH)) { 7087bf58533SChristoph Hellwig nvme_complete_async_event(&nvmeq->dev->ctrl, 7097bf58533SChristoph Hellwig cqe.status, &cqe.result); 710adf68f21SChristoph Hellwig continue; 711adf68f21SChristoph Hellwig } 712adf68f21SChristoph Hellwig 713eee417b0SChristoph Hellwig req = blk_mq_tag_to_rq(*nvmeq->tags, cqe.command_id); 714d49187e9SChristoph Hellwig nvme_req(req)->result = cqe.result; 715d783e0bdSMarta Rybczynska blk_mq_complete_request(req, le16_to_cpu(cqe.status) >> 1); 71657dacad5SJay Sternberg } 71757dacad5SJay Sternberg 71857dacad5SJay Sternberg if (head == nvmeq->cq_head && phase == nvmeq->cq_phase) 719a0fa9647SJens Axboe return; 72057dacad5SJay Sternberg 721604e8c8dSKeith Busch if (likely(nvmeq->cq_vector >= 0)) 72257dacad5SJay Sternberg writel(head, nvmeq->q_db + nvmeq->dev->db_stride); 72357dacad5SJay Sternberg nvmeq->cq_head = head; 72457dacad5SJay Sternberg nvmeq->cq_phase = phase; 72557dacad5SJay Sternberg 72657dacad5SJay Sternberg nvmeq->cqe_seen = 1; 727a0fa9647SJens Axboe } 728a0fa9647SJens Axboe 729a0fa9647SJens Axboe static void nvme_process_cq(struct nvme_queue *nvmeq) 730a0fa9647SJens Axboe { 731a0fa9647SJens Axboe __nvme_process_cq(nvmeq, NULL); 73257dacad5SJay Sternberg } 73357dacad5SJay Sternberg 73457dacad5SJay Sternberg static irqreturn_t nvme_irq(int irq, void *data) 73557dacad5SJay Sternberg { 73657dacad5SJay Sternberg irqreturn_t result; 73757dacad5SJay Sternberg struct nvme_queue *nvmeq = data; 73857dacad5SJay Sternberg spin_lock(&nvmeq->q_lock); 73957dacad5SJay Sternberg nvme_process_cq(nvmeq); 74057dacad5SJay Sternberg result = nvmeq->cqe_seen ? IRQ_HANDLED : IRQ_NONE; 74157dacad5SJay Sternberg nvmeq->cqe_seen = 0; 74257dacad5SJay Sternberg spin_unlock(&nvmeq->q_lock); 74357dacad5SJay Sternberg return result; 74457dacad5SJay Sternberg } 74557dacad5SJay Sternberg 74657dacad5SJay Sternberg static irqreturn_t nvme_irq_check(int irq, void *data) 74757dacad5SJay Sternberg { 74857dacad5SJay Sternberg struct nvme_queue *nvmeq = data; 749d783e0bdSMarta Rybczynska if (nvme_cqe_valid(nvmeq, nvmeq->cq_head, nvmeq->cq_phase)) 75057dacad5SJay Sternberg return IRQ_WAKE_THREAD; 751d783e0bdSMarta Rybczynska return IRQ_NONE; 75257dacad5SJay Sternberg } 75357dacad5SJay Sternberg 754a0fa9647SJens Axboe static int nvme_poll(struct blk_mq_hw_ctx *hctx, unsigned int tag) 755a0fa9647SJens Axboe { 756a0fa9647SJens Axboe struct nvme_queue *nvmeq = hctx->driver_data; 757a0fa9647SJens Axboe 758d783e0bdSMarta Rybczynska if (nvme_cqe_valid(nvmeq, nvmeq->cq_head, nvmeq->cq_phase)) { 759a0fa9647SJens Axboe spin_lock_irq(&nvmeq->q_lock); 760a0fa9647SJens Axboe __nvme_process_cq(nvmeq, &tag); 761a0fa9647SJens Axboe spin_unlock_irq(&nvmeq->q_lock); 762a0fa9647SJens Axboe 763a0fa9647SJens Axboe if (tag == -1) 764a0fa9647SJens Axboe return 1; 765a0fa9647SJens Axboe } 766a0fa9647SJens Axboe 767a0fa9647SJens Axboe return 0; 768a0fa9647SJens Axboe } 769a0fa9647SJens Axboe 770f866fc42SChristoph Hellwig static void nvme_pci_submit_async_event(struct nvme_ctrl *ctrl, int aer_idx) 77157dacad5SJay Sternberg { 772f866fc42SChristoph Hellwig struct nvme_dev *dev = to_nvme_dev(ctrl); 7739396dec9SChristoph Hellwig struct nvme_queue *nvmeq = dev->queues[0]; 77457dacad5SJay Sternberg struct nvme_command c; 77557dacad5SJay Sternberg 77657dacad5SJay Sternberg memset(&c, 0, sizeof(c)); 77757dacad5SJay Sternberg c.common.opcode = nvme_admin_async_event; 778f866fc42SChristoph Hellwig c.common.command_id = NVME_AQ_BLKMQ_DEPTH + aer_idx; 77957dacad5SJay Sternberg 7809396dec9SChristoph Hellwig spin_lock_irq(&nvmeq->q_lock); 7819396dec9SChristoph Hellwig __nvme_submit_cmd(nvmeq, &c); 7829396dec9SChristoph Hellwig spin_unlock_irq(&nvmeq->q_lock); 78357dacad5SJay Sternberg } 78457dacad5SJay Sternberg 78557dacad5SJay Sternberg static int adapter_delete_queue(struct nvme_dev *dev, u8 opcode, u16 id) 78657dacad5SJay Sternberg { 78757dacad5SJay Sternberg struct nvme_command c; 78857dacad5SJay Sternberg 78957dacad5SJay Sternberg memset(&c, 0, sizeof(c)); 79057dacad5SJay Sternberg c.delete_queue.opcode = opcode; 79157dacad5SJay Sternberg c.delete_queue.qid = cpu_to_le16(id); 79257dacad5SJay Sternberg 7931c63dc66SChristoph Hellwig return nvme_submit_sync_cmd(dev->ctrl.admin_q, &c, NULL, 0); 79457dacad5SJay Sternberg } 79557dacad5SJay Sternberg 79657dacad5SJay Sternberg static int adapter_alloc_cq(struct nvme_dev *dev, u16 qid, 79757dacad5SJay Sternberg struct nvme_queue *nvmeq) 79857dacad5SJay Sternberg { 79957dacad5SJay Sternberg struct nvme_command c; 80057dacad5SJay Sternberg int flags = NVME_QUEUE_PHYS_CONTIG | NVME_CQ_IRQ_ENABLED; 80157dacad5SJay Sternberg 80257dacad5SJay Sternberg /* 80357dacad5SJay Sternberg * Note: we (ab)use the fact the the prp fields survive if no data 80457dacad5SJay Sternberg * is attached to the request. 80557dacad5SJay Sternberg */ 80657dacad5SJay Sternberg memset(&c, 0, sizeof(c)); 80757dacad5SJay Sternberg c.create_cq.opcode = nvme_admin_create_cq; 80857dacad5SJay Sternberg c.create_cq.prp1 = cpu_to_le64(nvmeq->cq_dma_addr); 80957dacad5SJay Sternberg c.create_cq.cqid = cpu_to_le16(qid); 81057dacad5SJay Sternberg c.create_cq.qsize = cpu_to_le16(nvmeq->q_depth - 1); 81157dacad5SJay Sternberg c.create_cq.cq_flags = cpu_to_le16(flags); 81257dacad5SJay Sternberg c.create_cq.irq_vector = cpu_to_le16(nvmeq->cq_vector); 81357dacad5SJay Sternberg 8141c63dc66SChristoph Hellwig return nvme_submit_sync_cmd(dev->ctrl.admin_q, &c, NULL, 0); 81557dacad5SJay Sternberg } 81657dacad5SJay Sternberg 81757dacad5SJay Sternberg static int adapter_alloc_sq(struct nvme_dev *dev, u16 qid, 81857dacad5SJay Sternberg struct nvme_queue *nvmeq) 81957dacad5SJay Sternberg { 82057dacad5SJay Sternberg struct nvme_command c; 82157dacad5SJay Sternberg int flags = NVME_QUEUE_PHYS_CONTIG | NVME_SQ_PRIO_MEDIUM; 82257dacad5SJay Sternberg 82357dacad5SJay Sternberg /* 82457dacad5SJay Sternberg * Note: we (ab)use the fact the the prp fields survive if no data 82557dacad5SJay Sternberg * is attached to the request. 82657dacad5SJay Sternberg */ 82757dacad5SJay Sternberg memset(&c, 0, sizeof(c)); 82857dacad5SJay Sternberg c.create_sq.opcode = nvme_admin_create_sq; 82957dacad5SJay Sternberg c.create_sq.prp1 = cpu_to_le64(nvmeq->sq_dma_addr); 83057dacad5SJay Sternberg c.create_sq.sqid = cpu_to_le16(qid); 83157dacad5SJay Sternberg c.create_sq.qsize = cpu_to_le16(nvmeq->q_depth - 1); 83257dacad5SJay Sternberg c.create_sq.sq_flags = cpu_to_le16(flags); 83357dacad5SJay Sternberg c.create_sq.cqid = cpu_to_le16(qid); 83457dacad5SJay Sternberg 8351c63dc66SChristoph Hellwig return nvme_submit_sync_cmd(dev->ctrl.admin_q, &c, NULL, 0); 83657dacad5SJay Sternberg } 83757dacad5SJay Sternberg 83857dacad5SJay Sternberg static int adapter_delete_cq(struct nvme_dev *dev, u16 cqid) 83957dacad5SJay Sternberg { 84057dacad5SJay Sternberg return adapter_delete_queue(dev, nvme_admin_delete_cq, cqid); 84157dacad5SJay Sternberg } 84257dacad5SJay Sternberg 84357dacad5SJay Sternberg static int adapter_delete_sq(struct nvme_dev *dev, u16 sqid) 84457dacad5SJay Sternberg { 84557dacad5SJay Sternberg return adapter_delete_queue(dev, nvme_admin_delete_sq, sqid); 84657dacad5SJay Sternberg } 84757dacad5SJay Sternberg 848e7a2a87dSChristoph Hellwig static void abort_endio(struct request *req, int error) 84957dacad5SJay Sternberg { 850f4800d6dSChristoph Hellwig struct nvme_iod *iod = blk_mq_rq_to_pdu(req); 851f4800d6dSChristoph Hellwig struct nvme_queue *nvmeq = iod->nvmeq; 852e7a2a87dSChristoph Hellwig u16 status = req->errors; 85357dacad5SJay Sternberg 8541cb3cce5SChristoph Hellwig dev_warn(nvmeq->dev->ctrl.device, "Abort status: 0x%x", status); 855e7a2a87dSChristoph Hellwig atomic_inc(&nvmeq->dev->ctrl.abort_limit); 856e7a2a87dSChristoph Hellwig blk_mq_free_request(req); 85757dacad5SJay Sternberg } 85857dacad5SJay Sternberg 85931c7c7d2SChristoph Hellwig static enum blk_eh_timer_return nvme_timeout(struct request *req, bool reserved) 86057dacad5SJay Sternberg { 861f4800d6dSChristoph Hellwig struct nvme_iod *iod = blk_mq_rq_to_pdu(req); 862f4800d6dSChristoph Hellwig struct nvme_queue *nvmeq = iod->nvmeq; 86357dacad5SJay Sternberg struct nvme_dev *dev = nvmeq->dev; 86457dacad5SJay Sternberg struct request *abort_req; 86557dacad5SJay Sternberg struct nvme_command cmd; 86657dacad5SJay Sternberg 86731c7c7d2SChristoph Hellwig /* 868fd634f41SChristoph Hellwig * Shutdown immediately if controller times out while starting. The 869fd634f41SChristoph Hellwig * reset work will see the pci device disabled when it gets the forced 870fd634f41SChristoph Hellwig * cancellation error. All outstanding requests are completed on 871fd634f41SChristoph Hellwig * shutdown, so we return BLK_EH_HANDLED. 872fd634f41SChristoph Hellwig */ 873bb8d261eSChristoph Hellwig if (dev->ctrl.state == NVME_CTRL_RESETTING) { 8741b3c47c1SSagi Grimberg dev_warn(dev->ctrl.device, 875fd634f41SChristoph Hellwig "I/O %d QID %d timeout, disable controller\n", 876fd634f41SChristoph Hellwig req->tag, nvmeq->qid); 877a5cdb68cSKeith Busch nvme_dev_disable(dev, false); 878fd634f41SChristoph Hellwig req->errors = NVME_SC_CANCELLED; 879fd634f41SChristoph Hellwig return BLK_EH_HANDLED; 880fd634f41SChristoph Hellwig } 881fd634f41SChristoph Hellwig 882fd634f41SChristoph Hellwig /* 883e1569a16SKeith Busch * Shutdown the controller immediately and schedule a reset if the 884e1569a16SKeith Busch * command was already aborted once before and still hasn't been 885e1569a16SKeith Busch * returned to the driver, or if this is the admin queue. 88631c7c7d2SChristoph Hellwig */ 887f4800d6dSChristoph Hellwig if (!nvmeq->qid || iod->aborted) { 8881b3c47c1SSagi Grimberg dev_warn(dev->ctrl.device, 88957dacad5SJay Sternberg "I/O %d QID %d timeout, reset controller\n", 89057dacad5SJay Sternberg req->tag, nvmeq->qid); 891a5cdb68cSKeith Busch nvme_dev_disable(dev, false); 892c5f6ce97SKeith Busch nvme_reset(dev); 893e1569a16SKeith Busch 894e1569a16SKeith Busch /* 895e1569a16SKeith Busch * Mark the request as handled, since the inline shutdown 896e1569a16SKeith Busch * forces all outstanding requests to complete. 897e1569a16SKeith Busch */ 898e1569a16SKeith Busch req->errors = NVME_SC_CANCELLED; 899e1569a16SKeith Busch return BLK_EH_HANDLED; 90057dacad5SJay Sternberg } 90157dacad5SJay Sternberg 902e7a2a87dSChristoph Hellwig if (atomic_dec_return(&dev->ctrl.abort_limit) < 0) { 903e7a2a87dSChristoph Hellwig atomic_inc(&dev->ctrl.abort_limit); 904e7a2a87dSChristoph Hellwig return BLK_EH_RESET_TIMER; 905e7a2a87dSChristoph Hellwig } 9067bf7d778SKeith Busch iod->aborted = 1; 90757dacad5SJay Sternberg 90857dacad5SJay Sternberg memset(&cmd, 0, sizeof(cmd)); 90957dacad5SJay Sternberg cmd.abort.opcode = nvme_admin_abort_cmd; 91057dacad5SJay Sternberg cmd.abort.cid = req->tag; 91157dacad5SJay Sternberg cmd.abort.sqid = cpu_to_le16(nvmeq->qid); 91257dacad5SJay Sternberg 9131b3c47c1SSagi Grimberg dev_warn(nvmeq->dev->ctrl.device, 9141b3c47c1SSagi Grimberg "I/O %d QID %d timeout, aborting\n", 91557dacad5SJay Sternberg req->tag, nvmeq->qid); 916e7a2a87dSChristoph Hellwig 917e7a2a87dSChristoph Hellwig abort_req = nvme_alloc_request(dev->ctrl.admin_q, &cmd, 918eb71f435SChristoph Hellwig BLK_MQ_REQ_NOWAIT, NVME_QID_ANY); 9196bf25d16SChristoph Hellwig if (IS_ERR(abort_req)) { 9206bf25d16SChristoph Hellwig atomic_inc(&dev->ctrl.abort_limit); 92131c7c7d2SChristoph Hellwig return BLK_EH_RESET_TIMER; 92257dacad5SJay Sternberg } 92357dacad5SJay Sternberg 924e7a2a87dSChristoph Hellwig abort_req->timeout = ADMIN_TIMEOUT; 925e7a2a87dSChristoph Hellwig abort_req->end_io_data = NULL; 926e7a2a87dSChristoph Hellwig blk_execute_rq_nowait(abort_req->q, NULL, abort_req, 0, abort_endio); 92757dacad5SJay Sternberg 92857dacad5SJay Sternberg /* 92957dacad5SJay Sternberg * The aborted req will be completed on receiving the abort req. 93057dacad5SJay Sternberg * We enable the timer again. If hit twice, it'll cause a device reset, 93157dacad5SJay Sternberg * as the device then is in a faulty state. 93257dacad5SJay Sternberg */ 93357dacad5SJay Sternberg return BLK_EH_RESET_TIMER; 93457dacad5SJay Sternberg } 93557dacad5SJay Sternberg 93657dacad5SJay Sternberg static void nvme_free_queue(struct nvme_queue *nvmeq) 93757dacad5SJay Sternberg { 93857dacad5SJay Sternberg dma_free_coherent(nvmeq->q_dmadev, CQ_SIZE(nvmeq->q_depth), 93957dacad5SJay Sternberg (void *)nvmeq->cqes, nvmeq->cq_dma_addr); 94057dacad5SJay Sternberg if (nvmeq->sq_cmds) 94157dacad5SJay Sternberg dma_free_coherent(nvmeq->q_dmadev, SQ_SIZE(nvmeq->q_depth), 94257dacad5SJay Sternberg nvmeq->sq_cmds, nvmeq->sq_dma_addr); 94357dacad5SJay Sternberg kfree(nvmeq); 94457dacad5SJay Sternberg } 94557dacad5SJay Sternberg 94657dacad5SJay Sternberg static void nvme_free_queues(struct nvme_dev *dev, int lowest) 94757dacad5SJay Sternberg { 94857dacad5SJay Sternberg int i; 94957dacad5SJay Sternberg 95057dacad5SJay Sternberg for (i = dev->queue_count - 1; i >= lowest; i--) { 95157dacad5SJay Sternberg struct nvme_queue *nvmeq = dev->queues[i]; 95257dacad5SJay Sternberg dev->queue_count--; 95357dacad5SJay Sternberg dev->queues[i] = NULL; 95457dacad5SJay Sternberg nvme_free_queue(nvmeq); 95557dacad5SJay Sternberg } 95657dacad5SJay Sternberg } 95757dacad5SJay Sternberg 95857dacad5SJay Sternberg /** 95957dacad5SJay Sternberg * nvme_suspend_queue - put queue into suspended state 96057dacad5SJay Sternberg * @nvmeq - queue to suspend 96157dacad5SJay Sternberg */ 96257dacad5SJay Sternberg static int nvme_suspend_queue(struct nvme_queue *nvmeq) 96357dacad5SJay Sternberg { 96457dacad5SJay Sternberg int vector; 96557dacad5SJay Sternberg 96657dacad5SJay Sternberg spin_lock_irq(&nvmeq->q_lock); 96757dacad5SJay Sternberg if (nvmeq->cq_vector == -1) { 96857dacad5SJay Sternberg spin_unlock_irq(&nvmeq->q_lock); 96957dacad5SJay Sternberg return 1; 97057dacad5SJay Sternberg } 971dca51e78SChristoph Hellwig vector = nvmeq_irq(nvmeq); 97257dacad5SJay Sternberg nvmeq->dev->online_queues--; 97357dacad5SJay Sternberg nvmeq->cq_vector = -1; 97457dacad5SJay Sternberg spin_unlock_irq(&nvmeq->q_lock); 97557dacad5SJay Sternberg 9761c63dc66SChristoph Hellwig if (!nvmeq->qid && nvmeq->dev->ctrl.admin_q) 97725646264SKeith Busch blk_mq_stop_hw_queues(nvmeq->dev->ctrl.admin_q); 97857dacad5SJay Sternberg 97957dacad5SJay Sternberg free_irq(vector, nvmeq); 98057dacad5SJay Sternberg 98157dacad5SJay Sternberg return 0; 98257dacad5SJay Sternberg } 98357dacad5SJay Sternberg 984a5cdb68cSKeith Busch static void nvme_disable_admin_queue(struct nvme_dev *dev, bool shutdown) 98557dacad5SJay Sternberg { 986a5cdb68cSKeith Busch struct nvme_queue *nvmeq = dev->queues[0]; 98757dacad5SJay Sternberg 98857dacad5SJay Sternberg if (!nvmeq) 98957dacad5SJay Sternberg return; 99057dacad5SJay Sternberg if (nvme_suspend_queue(nvmeq)) 99157dacad5SJay Sternberg return; 99257dacad5SJay Sternberg 993a5cdb68cSKeith Busch if (shutdown) 994a5cdb68cSKeith Busch nvme_shutdown_ctrl(&dev->ctrl); 995a5cdb68cSKeith Busch else 996a5cdb68cSKeith Busch nvme_disable_ctrl(&dev->ctrl, lo_hi_readq( 997a5cdb68cSKeith Busch dev->bar + NVME_REG_CAP)); 99857dacad5SJay Sternberg 99957dacad5SJay Sternberg spin_lock_irq(&nvmeq->q_lock); 100057dacad5SJay Sternberg nvme_process_cq(nvmeq); 100157dacad5SJay Sternberg spin_unlock_irq(&nvmeq->q_lock); 100257dacad5SJay Sternberg } 100357dacad5SJay Sternberg 100457dacad5SJay Sternberg static int nvme_cmb_qdepth(struct nvme_dev *dev, int nr_io_queues, 100557dacad5SJay Sternberg int entry_size) 100657dacad5SJay Sternberg { 100757dacad5SJay Sternberg int q_depth = dev->q_depth; 10085fd4ce1bSChristoph Hellwig unsigned q_size_aligned = roundup(q_depth * entry_size, 10095fd4ce1bSChristoph Hellwig dev->ctrl.page_size); 101057dacad5SJay Sternberg 101157dacad5SJay Sternberg if (q_size_aligned * nr_io_queues > dev->cmb_size) { 101257dacad5SJay Sternberg u64 mem_per_q = div_u64(dev->cmb_size, nr_io_queues); 10135fd4ce1bSChristoph Hellwig mem_per_q = round_down(mem_per_q, dev->ctrl.page_size); 101457dacad5SJay Sternberg q_depth = div_u64(mem_per_q, entry_size); 101557dacad5SJay Sternberg 101657dacad5SJay Sternberg /* 101757dacad5SJay Sternberg * Ensure the reduced q_depth is above some threshold where it 101857dacad5SJay Sternberg * would be better to map queues in system memory with the 101957dacad5SJay Sternberg * original depth 102057dacad5SJay Sternberg */ 102157dacad5SJay Sternberg if (q_depth < 64) 102257dacad5SJay Sternberg return -ENOMEM; 102357dacad5SJay Sternberg } 102457dacad5SJay Sternberg 102557dacad5SJay Sternberg return q_depth; 102657dacad5SJay Sternberg } 102757dacad5SJay Sternberg 102857dacad5SJay Sternberg static int nvme_alloc_sq_cmds(struct nvme_dev *dev, struct nvme_queue *nvmeq, 102957dacad5SJay Sternberg int qid, int depth) 103057dacad5SJay Sternberg { 103157dacad5SJay Sternberg if (qid && dev->cmb && use_cmb_sqes && NVME_CMB_SQS(dev->cmbsz)) { 10325fd4ce1bSChristoph Hellwig unsigned offset = (qid - 1) * roundup(SQ_SIZE(depth), 10335fd4ce1bSChristoph Hellwig dev->ctrl.page_size); 103457dacad5SJay Sternberg nvmeq->sq_dma_addr = dev->cmb_dma_addr + offset; 103557dacad5SJay Sternberg nvmeq->sq_cmds_io = dev->cmb + offset; 103657dacad5SJay Sternberg } else { 103757dacad5SJay Sternberg nvmeq->sq_cmds = dma_alloc_coherent(dev->dev, SQ_SIZE(depth), 103857dacad5SJay Sternberg &nvmeq->sq_dma_addr, GFP_KERNEL); 103957dacad5SJay Sternberg if (!nvmeq->sq_cmds) 104057dacad5SJay Sternberg return -ENOMEM; 104157dacad5SJay Sternberg } 104257dacad5SJay Sternberg 104357dacad5SJay Sternberg return 0; 104457dacad5SJay Sternberg } 104557dacad5SJay Sternberg 104657dacad5SJay Sternberg static struct nvme_queue *nvme_alloc_queue(struct nvme_dev *dev, int qid, 104757dacad5SJay Sternberg int depth) 104857dacad5SJay Sternberg { 104957dacad5SJay Sternberg struct nvme_queue *nvmeq = kzalloc(sizeof(*nvmeq), GFP_KERNEL); 105057dacad5SJay Sternberg if (!nvmeq) 105157dacad5SJay Sternberg return NULL; 105257dacad5SJay Sternberg 105357dacad5SJay Sternberg nvmeq->cqes = dma_zalloc_coherent(dev->dev, CQ_SIZE(depth), 105457dacad5SJay Sternberg &nvmeq->cq_dma_addr, GFP_KERNEL); 105557dacad5SJay Sternberg if (!nvmeq->cqes) 105657dacad5SJay Sternberg goto free_nvmeq; 105757dacad5SJay Sternberg 105857dacad5SJay Sternberg if (nvme_alloc_sq_cmds(dev, nvmeq, qid, depth)) 105957dacad5SJay Sternberg goto free_cqdma; 106057dacad5SJay Sternberg 106157dacad5SJay Sternberg nvmeq->q_dmadev = dev->dev; 106257dacad5SJay Sternberg nvmeq->dev = dev; 106357dacad5SJay Sternberg snprintf(nvmeq->irqname, sizeof(nvmeq->irqname), "nvme%dq%d", 10641c63dc66SChristoph Hellwig dev->ctrl.instance, qid); 106557dacad5SJay Sternberg spin_lock_init(&nvmeq->q_lock); 106657dacad5SJay Sternberg nvmeq->cq_head = 0; 106757dacad5SJay Sternberg nvmeq->cq_phase = 1; 106857dacad5SJay Sternberg nvmeq->q_db = &dev->dbs[qid * 2 * dev->db_stride]; 106957dacad5SJay Sternberg nvmeq->q_depth = depth; 107057dacad5SJay Sternberg nvmeq->qid = qid; 107157dacad5SJay Sternberg nvmeq->cq_vector = -1; 107257dacad5SJay Sternberg dev->queues[qid] = nvmeq; 107357dacad5SJay Sternberg dev->queue_count++; 107457dacad5SJay Sternberg 107557dacad5SJay Sternberg return nvmeq; 107657dacad5SJay Sternberg 107757dacad5SJay Sternberg free_cqdma: 107857dacad5SJay Sternberg dma_free_coherent(dev->dev, CQ_SIZE(depth), (void *)nvmeq->cqes, 107957dacad5SJay Sternberg nvmeq->cq_dma_addr); 108057dacad5SJay Sternberg free_nvmeq: 108157dacad5SJay Sternberg kfree(nvmeq); 108257dacad5SJay Sternberg return NULL; 108357dacad5SJay Sternberg } 108457dacad5SJay Sternberg 1085dca51e78SChristoph Hellwig static int queue_request_irq(struct nvme_queue *nvmeq) 108657dacad5SJay Sternberg { 108757dacad5SJay Sternberg if (use_threaded_interrupts) 1088dca51e78SChristoph Hellwig return request_threaded_irq(nvmeq_irq(nvmeq), nvme_irq_check, 1089dca51e78SChristoph Hellwig nvme_irq, IRQF_SHARED, nvmeq->irqname, nvmeq); 1090dca51e78SChristoph Hellwig else 1091dca51e78SChristoph Hellwig return request_irq(nvmeq_irq(nvmeq), nvme_irq, IRQF_SHARED, 1092dca51e78SChristoph Hellwig nvmeq->irqname, nvmeq); 109357dacad5SJay Sternberg } 109457dacad5SJay Sternberg 109557dacad5SJay Sternberg static void nvme_init_queue(struct nvme_queue *nvmeq, u16 qid) 109657dacad5SJay Sternberg { 109757dacad5SJay Sternberg struct nvme_dev *dev = nvmeq->dev; 109857dacad5SJay Sternberg 109957dacad5SJay Sternberg spin_lock_irq(&nvmeq->q_lock); 110057dacad5SJay Sternberg nvmeq->sq_tail = 0; 110157dacad5SJay Sternberg nvmeq->cq_head = 0; 110257dacad5SJay Sternberg nvmeq->cq_phase = 1; 110357dacad5SJay Sternberg nvmeq->q_db = &dev->dbs[qid * 2 * dev->db_stride]; 110457dacad5SJay Sternberg memset((void *)nvmeq->cqes, 0, CQ_SIZE(nvmeq->q_depth)); 110557dacad5SJay Sternberg dev->online_queues++; 110657dacad5SJay Sternberg spin_unlock_irq(&nvmeq->q_lock); 110757dacad5SJay Sternberg } 110857dacad5SJay Sternberg 110957dacad5SJay Sternberg static int nvme_create_queue(struct nvme_queue *nvmeq, int qid) 111057dacad5SJay Sternberg { 111157dacad5SJay Sternberg struct nvme_dev *dev = nvmeq->dev; 111257dacad5SJay Sternberg int result; 111357dacad5SJay Sternberg 111457dacad5SJay Sternberg nvmeq->cq_vector = qid - 1; 111557dacad5SJay Sternberg result = adapter_alloc_cq(dev, qid, nvmeq); 111657dacad5SJay Sternberg if (result < 0) 111757dacad5SJay Sternberg return result; 111857dacad5SJay Sternberg 111957dacad5SJay Sternberg result = adapter_alloc_sq(dev, qid, nvmeq); 112057dacad5SJay Sternberg if (result < 0) 112157dacad5SJay Sternberg goto release_cq; 112257dacad5SJay Sternberg 1123dca51e78SChristoph Hellwig result = queue_request_irq(nvmeq); 112457dacad5SJay Sternberg if (result < 0) 112557dacad5SJay Sternberg goto release_sq; 112657dacad5SJay Sternberg 112757dacad5SJay Sternberg nvme_init_queue(nvmeq, qid); 112857dacad5SJay Sternberg return result; 112957dacad5SJay Sternberg 113057dacad5SJay Sternberg release_sq: 113157dacad5SJay Sternberg adapter_delete_sq(dev, qid); 113257dacad5SJay Sternberg release_cq: 113357dacad5SJay Sternberg adapter_delete_cq(dev, qid); 113457dacad5SJay Sternberg return result; 113557dacad5SJay Sternberg } 113657dacad5SJay Sternberg 113757dacad5SJay Sternberg static struct blk_mq_ops nvme_mq_admin_ops = { 113857dacad5SJay Sternberg .queue_rq = nvme_queue_rq, 1139eee417b0SChristoph Hellwig .complete = nvme_complete_rq, 114057dacad5SJay Sternberg .init_hctx = nvme_admin_init_hctx, 114157dacad5SJay Sternberg .exit_hctx = nvme_admin_exit_hctx, 114257dacad5SJay Sternberg .init_request = nvme_admin_init_request, 114357dacad5SJay Sternberg .timeout = nvme_timeout, 114457dacad5SJay Sternberg }; 114557dacad5SJay Sternberg 114657dacad5SJay Sternberg static struct blk_mq_ops nvme_mq_ops = { 114757dacad5SJay Sternberg .queue_rq = nvme_queue_rq, 1148eee417b0SChristoph Hellwig .complete = nvme_complete_rq, 114957dacad5SJay Sternberg .init_hctx = nvme_init_hctx, 115057dacad5SJay Sternberg .init_request = nvme_init_request, 1151dca51e78SChristoph Hellwig .map_queues = nvme_pci_map_queues, 115257dacad5SJay Sternberg .timeout = nvme_timeout, 1153a0fa9647SJens Axboe .poll = nvme_poll, 115457dacad5SJay Sternberg }; 115557dacad5SJay Sternberg 115657dacad5SJay Sternberg static void nvme_dev_remove_admin(struct nvme_dev *dev) 115757dacad5SJay Sternberg { 11581c63dc66SChristoph Hellwig if (dev->ctrl.admin_q && !blk_queue_dying(dev->ctrl.admin_q)) { 115969d9a99cSKeith Busch /* 116069d9a99cSKeith Busch * If the controller was reset during removal, it's possible 116169d9a99cSKeith Busch * user requests may be waiting on a stopped queue. Start the 116269d9a99cSKeith Busch * queue to flush these to completion. 116369d9a99cSKeith Busch */ 116469d9a99cSKeith Busch blk_mq_start_stopped_hw_queues(dev->ctrl.admin_q, true); 11651c63dc66SChristoph Hellwig blk_cleanup_queue(dev->ctrl.admin_q); 116657dacad5SJay Sternberg blk_mq_free_tag_set(&dev->admin_tagset); 116757dacad5SJay Sternberg } 116857dacad5SJay Sternberg } 116957dacad5SJay Sternberg 117057dacad5SJay Sternberg static int nvme_alloc_admin_tags(struct nvme_dev *dev) 117157dacad5SJay Sternberg { 11721c63dc66SChristoph Hellwig if (!dev->ctrl.admin_q) { 117357dacad5SJay Sternberg dev->admin_tagset.ops = &nvme_mq_admin_ops; 117457dacad5SJay Sternberg dev->admin_tagset.nr_hw_queues = 1; 1175e3e9d50cSKeith Busch 1176e3e9d50cSKeith Busch /* 1177e3e9d50cSKeith Busch * Subtract one to leave an empty queue entry for 'Full Queue' 1178e3e9d50cSKeith Busch * condition. See NVM-Express 1.2 specification, section 4.1.2. 1179e3e9d50cSKeith Busch */ 1180e3e9d50cSKeith Busch dev->admin_tagset.queue_depth = NVME_AQ_BLKMQ_DEPTH - 1; 118157dacad5SJay Sternberg dev->admin_tagset.timeout = ADMIN_TIMEOUT; 118257dacad5SJay Sternberg dev->admin_tagset.numa_node = dev_to_node(dev->dev); 118357dacad5SJay Sternberg dev->admin_tagset.cmd_size = nvme_cmd_size(dev); 1184d3484991SJens Axboe dev->admin_tagset.flags = BLK_MQ_F_NO_SCHED; 118557dacad5SJay Sternberg dev->admin_tagset.driver_data = dev; 118657dacad5SJay Sternberg 118757dacad5SJay Sternberg if (blk_mq_alloc_tag_set(&dev->admin_tagset)) 118857dacad5SJay Sternberg return -ENOMEM; 118957dacad5SJay Sternberg 11901c63dc66SChristoph Hellwig dev->ctrl.admin_q = blk_mq_init_queue(&dev->admin_tagset); 11911c63dc66SChristoph Hellwig if (IS_ERR(dev->ctrl.admin_q)) { 119257dacad5SJay Sternberg blk_mq_free_tag_set(&dev->admin_tagset); 119357dacad5SJay Sternberg return -ENOMEM; 119457dacad5SJay Sternberg } 11951c63dc66SChristoph Hellwig if (!blk_get_queue(dev->ctrl.admin_q)) { 119657dacad5SJay Sternberg nvme_dev_remove_admin(dev); 11971c63dc66SChristoph Hellwig dev->ctrl.admin_q = NULL; 119857dacad5SJay Sternberg return -ENODEV; 119957dacad5SJay Sternberg } 120057dacad5SJay Sternberg } else 120125646264SKeith Busch blk_mq_start_stopped_hw_queues(dev->ctrl.admin_q, true); 120257dacad5SJay Sternberg 120357dacad5SJay Sternberg return 0; 120457dacad5SJay Sternberg } 120557dacad5SJay Sternberg 120657dacad5SJay Sternberg static int nvme_configure_admin_queue(struct nvme_dev *dev) 120757dacad5SJay Sternberg { 120857dacad5SJay Sternberg int result; 120957dacad5SJay Sternberg u32 aqa; 12107a67cbeaSChristoph Hellwig u64 cap = lo_hi_readq(dev->bar + NVME_REG_CAP); 121157dacad5SJay Sternberg struct nvme_queue *nvmeq; 121257dacad5SJay Sternberg 12138ef2074dSGabriel Krisman Bertazi dev->subsystem = readl(dev->bar + NVME_REG_VS) >= NVME_VS(1, 1, 0) ? 121457dacad5SJay Sternberg NVME_CAP_NSSRC(cap) : 0; 121557dacad5SJay Sternberg 12167a67cbeaSChristoph Hellwig if (dev->subsystem && 12177a67cbeaSChristoph Hellwig (readl(dev->bar + NVME_REG_CSTS) & NVME_CSTS_NSSRO)) 12187a67cbeaSChristoph Hellwig writel(NVME_CSTS_NSSRO, dev->bar + NVME_REG_CSTS); 121957dacad5SJay Sternberg 12205fd4ce1bSChristoph Hellwig result = nvme_disable_ctrl(&dev->ctrl, cap); 122157dacad5SJay Sternberg if (result < 0) 122257dacad5SJay Sternberg return result; 122357dacad5SJay Sternberg 122457dacad5SJay Sternberg nvmeq = dev->queues[0]; 122557dacad5SJay Sternberg if (!nvmeq) { 122657dacad5SJay Sternberg nvmeq = nvme_alloc_queue(dev, 0, NVME_AQ_DEPTH); 122757dacad5SJay Sternberg if (!nvmeq) 122857dacad5SJay Sternberg return -ENOMEM; 122957dacad5SJay Sternberg } 123057dacad5SJay Sternberg 123157dacad5SJay Sternberg aqa = nvmeq->q_depth - 1; 123257dacad5SJay Sternberg aqa |= aqa << 16; 123357dacad5SJay Sternberg 12347a67cbeaSChristoph Hellwig writel(aqa, dev->bar + NVME_REG_AQA); 12357a67cbeaSChristoph Hellwig lo_hi_writeq(nvmeq->sq_dma_addr, dev->bar + NVME_REG_ASQ); 12367a67cbeaSChristoph Hellwig lo_hi_writeq(nvmeq->cq_dma_addr, dev->bar + NVME_REG_ACQ); 123757dacad5SJay Sternberg 12385fd4ce1bSChristoph Hellwig result = nvme_enable_ctrl(&dev->ctrl, cap); 123957dacad5SJay Sternberg if (result) 1240d4875622SKeith Busch return result; 124157dacad5SJay Sternberg 124257dacad5SJay Sternberg nvmeq->cq_vector = 0; 1243dca51e78SChristoph Hellwig result = queue_request_irq(nvmeq); 124457dacad5SJay Sternberg if (result) { 124557dacad5SJay Sternberg nvmeq->cq_vector = -1; 1246d4875622SKeith Busch return result; 124757dacad5SJay Sternberg } 124857dacad5SJay Sternberg 124957dacad5SJay Sternberg return result; 125057dacad5SJay Sternberg } 125157dacad5SJay Sternberg 1252c875a709SGuilherme G. Piccoli static bool nvme_should_reset(struct nvme_dev *dev, u32 csts) 1253c875a709SGuilherme G. Piccoli { 1254c875a709SGuilherme G. Piccoli 1255c875a709SGuilherme G. Piccoli /* If true, indicates loss of adapter communication, possibly by a 1256c875a709SGuilherme G. Piccoli * NVMe Subsystem reset. 1257c875a709SGuilherme G. Piccoli */ 1258c875a709SGuilherme G. Piccoli bool nssro = dev->subsystem && (csts & NVME_CSTS_NSSRO); 1259c875a709SGuilherme G. Piccoli 1260c875a709SGuilherme G. Piccoli /* If there is a reset ongoing, we shouldn't reset again. */ 1261c875a709SGuilherme G. Piccoli if (work_busy(&dev->reset_work)) 1262c875a709SGuilherme G. Piccoli return false; 1263c875a709SGuilherme G. Piccoli 1264c875a709SGuilherme G. Piccoli /* We shouldn't reset unless the controller is on fatal error state 1265c875a709SGuilherme G. Piccoli * _or_ if we lost the communication with it. 1266c875a709SGuilherme G. Piccoli */ 1267c875a709SGuilherme G. Piccoli if (!(csts & NVME_CSTS_CFS) && !nssro) 1268c875a709SGuilherme G. Piccoli return false; 1269c875a709SGuilherme G. Piccoli 1270c875a709SGuilherme G. Piccoli /* If PCI error recovery process is happening, we cannot reset or 1271c875a709SGuilherme G. Piccoli * the recovery mechanism will surely fail. 1272c875a709SGuilherme G. Piccoli */ 1273c875a709SGuilherme G. Piccoli if (pci_channel_offline(to_pci_dev(dev->dev))) 1274c875a709SGuilherme G. Piccoli return false; 1275c875a709SGuilherme G. Piccoli 1276c875a709SGuilherme G. Piccoli return true; 1277c875a709SGuilherme G. Piccoli } 1278c875a709SGuilherme G. Piccoli 1279d2a61918SAndy Lutomirski static void nvme_warn_reset(struct nvme_dev *dev, u32 csts) 1280d2a61918SAndy Lutomirski { 1281d2a61918SAndy Lutomirski /* Read a config register to help see what died. */ 1282d2a61918SAndy Lutomirski u16 pci_status; 1283d2a61918SAndy Lutomirski int result; 1284d2a61918SAndy Lutomirski 1285d2a61918SAndy Lutomirski result = pci_read_config_word(to_pci_dev(dev->dev), PCI_STATUS, 1286d2a61918SAndy Lutomirski &pci_status); 1287d2a61918SAndy Lutomirski if (result == PCIBIOS_SUCCESSFUL) 1288d2a61918SAndy Lutomirski dev_warn(dev->dev, 1289d2a61918SAndy Lutomirski "controller is down; will reset: CSTS=0x%x, PCI_STATUS=0x%hx\n", 1290d2a61918SAndy Lutomirski csts, pci_status); 1291d2a61918SAndy Lutomirski else 1292d2a61918SAndy Lutomirski dev_warn(dev->dev, 1293d2a61918SAndy Lutomirski "controller is down; will reset: CSTS=0x%x, PCI_STATUS read failed (%d)\n", 1294d2a61918SAndy Lutomirski csts, result); 1295d2a61918SAndy Lutomirski } 1296d2a61918SAndy Lutomirski 12972d55cd5fSChristoph Hellwig static void nvme_watchdog_timer(unsigned long data) 129857dacad5SJay Sternberg { 12992d55cd5fSChristoph Hellwig struct nvme_dev *dev = (struct nvme_dev *)data; 13007a67cbeaSChristoph Hellwig u32 csts = readl(dev->bar + NVME_REG_CSTS); 130157dacad5SJay Sternberg 1302c875a709SGuilherme G. Piccoli /* Skip controllers under certain specific conditions. */ 1303c875a709SGuilherme G. Piccoli if (nvme_should_reset(dev, csts)) { 1304c5f6ce97SKeith Busch if (!nvme_reset(dev)) 1305d2a61918SAndy Lutomirski nvme_warn_reset(dev, csts); 13062d55cd5fSChristoph Hellwig return; 130757dacad5SJay Sternberg } 130857dacad5SJay Sternberg 13092d55cd5fSChristoph Hellwig mod_timer(&dev->watchdog_timer, round_jiffies(jiffies + HZ)); 131057dacad5SJay Sternberg } 131157dacad5SJay Sternberg 1312749941f2SChristoph Hellwig static int nvme_create_io_queues(struct nvme_dev *dev) 131357dacad5SJay Sternberg { 1314949928c1SKeith Busch unsigned i, max; 1315749941f2SChristoph Hellwig int ret = 0; 131657dacad5SJay Sternberg 1317749941f2SChristoph Hellwig for (i = dev->queue_count; i <= dev->max_qid; i++) { 1318749941f2SChristoph Hellwig if (!nvme_alloc_queue(dev, i, dev->q_depth)) { 1319749941f2SChristoph Hellwig ret = -ENOMEM; 132057dacad5SJay Sternberg break; 1321749941f2SChristoph Hellwig } 1322749941f2SChristoph Hellwig } 132357dacad5SJay Sternberg 1324949928c1SKeith Busch max = min(dev->max_qid, dev->queue_count - 1); 1325949928c1SKeith Busch for (i = dev->online_queues; i <= max; i++) { 1326749941f2SChristoph Hellwig ret = nvme_create_queue(dev->queues[i], i); 1327d4875622SKeith Busch if (ret) 132857dacad5SJay Sternberg break; 132957dacad5SJay Sternberg } 133057dacad5SJay Sternberg 1331749941f2SChristoph Hellwig /* 1332749941f2SChristoph Hellwig * Ignore failing Create SQ/CQ commands, we can continue with less 1333749941f2SChristoph Hellwig * than the desired aount of queues, and even a controller without 1334749941f2SChristoph Hellwig * I/O queues an still be used to issue admin commands. This might 1335749941f2SChristoph Hellwig * be useful to upgrade a buggy firmware for example. 1336749941f2SChristoph Hellwig */ 1337749941f2SChristoph Hellwig return ret >= 0 ? 0 : ret; 133857dacad5SJay Sternberg } 133957dacad5SJay Sternberg 1340202021c1SStephen Bates static ssize_t nvme_cmb_show(struct device *dev, 1341202021c1SStephen Bates struct device_attribute *attr, 1342202021c1SStephen Bates char *buf) 1343202021c1SStephen Bates { 1344202021c1SStephen Bates struct nvme_dev *ndev = to_nvme_dev(dev_get_drvdata(dev)); 1345202021c1SStephen Bates 1346c965809cSStephen Bates return scnprintf(buf, PAGE_SIZE, "cmbloc : x%08x\ncmbsz : x%08x\n", 1347202021c1SStephen Bates ndev->cmbloc, ndev->cmbsz); 1348202021c1SStephen Bates } 1349202021c1SStephen Bates static DEVICE_ATTR(cmb, S_IRUGO, nvme_cmb_show, NULL); 1350202021c1SStephen Bates 135157dacad5SJay Sternberg static void __iomem *nvme_map_cmb(struct nvme_dev *dev) 135257dacad5SJay Sternberg { 135357dacad5SJay Sternberg u64 szu, size, offset; 135457dacad5SJay Sternberg resource_size_t bar_size; 135557dacad5SJay Sternberg struct pci_dev *pdev = to_pci_dev(dev->dev); 135657dacad5SJay Sternberg void __iomem *cmb; 135757dacad5SJay Sternberg dma_addr_t dma_addr; 135857dacad5SJay Sternberg 13597a67cbeaSChristoph Hellwig dev->cmbsz = readl(dev->bar + NVME_REG_CMBSZ); 136057dacad5SJay Sternberg if (!(NVME_CMB_SZ(dev->cmbsz))) 136157dacad5SJay Sternberg return NULL; 1362202021c1SStephen Bates dev->cmbloc = readl(dev->bar + NVME_REG_CMBLOC); 136357dacad5SJay Sternberg 1364202021c1SStephen Bates if (!use_cmb_sqes) 1365202021c1SStephen Bates return NULL; 136657dacad5SJay Sternberg 136757dacad5SJay Sternberg szu = (u64)1 << (12 + 4 * NVME_CMB_SZU(dev->cmbsz)); 136857dacad5SJay Sternberg size = szu * NVME_CMB_SZ(dev->cmbsz); 1369202021c1SStephen Bates offset = szu * NVME_CMB_OFST(dev->cmbloc); 1370202021c1SStephen Bates bar_size = pci_resource_len(pdev, NVME_CMB_BIR(dev->cmbloc)); 137157dacad5SJay Sternberg 137257dacad5SJay Sternberg if (offset > bar_size) 137357dacad5SJay Sternberg return NULL; 137457dacad5SJay Sternberg 137557dacad5SJay Sternberg /* 137657dacad5SJay Sternberg * Controllers may support a CMB size larger than their BAR, 137757dacad5SJay Sternberg * for example, due to being behind a bridge. Reduce the CMB to 137857dacad5SJay Sternberg * the reported size of the BAR 137957dacad5SJay Sternberg */ 138057dacad5SJay Sternberg if (size > bar_size - offset) 138157dacad5SJay Sternberg size = bar_size - offset; 138257dacad5SJay Sternberg 1383202021c1SStephen Bates dma_addr = pci_resource_start(pdev, NVME_CMB_BIR(dev->cmbloc)) + offset; 138457dacad5SJay Sternberg cmb = ioremap_wc(dma_addr, size); 138557dacad5SJay Sternberg if (!cmb) 138657dacad5SJay Sternberg return NULL; 138757dacad5SJay Sternberg 138857dacad5SJay Sternberg dev->cmb_dma_addr = dma_addr; 138957dacad5SJay Sternberg dev->cmb_size = size; 139057dacad5SJay Sternberg return cmb; 139157dacad5SJay Sternberg } 139257dacad5SJay Sternberg 139357dacad5SJay Sternberg static inline void nvme_release_cmb(struct nvme_dev *dev) 139457dacad5SJay Sternberg { 139557dacad5SJay Sternberg if (dev->cmb) { 139657dacad5SJay Sternberg iounmap(dev->cmb); 139757dacad5SJay Sternberg dev->cmb = NULL; 139857dacad5SJay Sternberg } 139957dacad5SJay Sternberg } 140057dacad5SJay Sternberg 140157dacad5SJay Sternberg static size_t db_bar_size(struct nvme_dev *dev, unsigned nr_io_queues) 140257dacad5SJay Sternberg { 140357dacad5SJay Sternberg return 4096 + ((nr_io_queues + 1) * 8 * dev->db_stride); 140457dacad5SJay Sternberg } 140557dacad5SJay Sternberg 140657dacad5SJay Sternberg static int nvme_setup_io_queues(struct nvme_dev *dev) 140757dacad5SJay Sternberg { 140857dacad5SJay Sternberg struct nvme_queue *adminq = dev->queues[0]; 140957dacad5SJay Sternberg struct pci_dev *pdev = to_pci_dev(dev->dev); 1410dca51e78SChristoph Hellwig int result, nr_io_queues, size; 141157dacad5SJay Sternberg 14122800b8e7SKeith Busch nr_io_queues = num_online_cpus(); 14139a0be7abSChristoph Hellwig result = nvme_set_queue_count(&dev->ctrl, &nr_io_queues); 14149a0be7abSChristoph Hellwig if (result < 0) 141557dacad5SJay Sternberg return result; 14169a0be7abSChristoph Hellwig 1417f5fa90dcSChristoph Hellwig if (nr_io_queues == 0) 1418a5229050SKeith Busch return 0; 141957dacad5SJay Sternberg 142057dacad5SJay Sternberg if (dev->cmb && NVME_CMB_SQS(dev->cmbsz)) { 142157dacad5SJay Sternberg result = nvme_cmb_qdepth(dev, nr_io_queues, 142257dacad5SJay Sternberg sizeof(struct nvme_command)); 142357dacad5SJay Sternberg if (result > 0) 142457dacad5SJay Sternberg dev->q_depth = result; 142557dacad5SJay Sternberg else 142657dacad5SJay Sternberg nvme_release_cmb(dev); 142757dacad5SJay Sternberg } 142857dacad5SJay Sternberg 142957dacad5SJay Sternberg size = db_bar_size(dev, nr_io_queues); 143057dacad5SJay Sternberg if (size > 8192) { 143157dacad5SJay Sternberg iounmap(dev->bar); 143257dacad5SJay Sternberg do { 143357dacad5SJay Sternberg dev->bar = ioremap(pci_resource_start(pdev, 0), size); 143457dacad5SJay Sternberg if (dev->bar) 143557dacad5SJay Sternberg break; 143657dacad5SJay Sternberg if (!--nr_io_queues) 143757dacad5SJay Sternberg return -ENOMEM; 143857dacad5SJay Sternberg size = db_bar_size(dev, nr_io_queues); 143957dacad5SJay Sternberg } while (1); 14407a67cbeaSChristoph Hellwig dev->dbs = dev->bar + 4096; 144157dacad5SJay Sternberg adminq->q_db = dev->dbs; 144257dacad5SJay Sternberg } 144357dacad5SJay Sternberg 144457dacad5SJay Sternberg /* Deregister the admin queue's interrupt */ 1445dca51e78SChristoph Hellwig free_irq(pci_irq_vector(pdev, 0), adminq); 144657dacad5SJay Sternberg 144757dacad5SJay Sternberg /* 144857dacad5SJay Sternberg * If we enable msix early due to not intx, disable it again before 144957dacad5SJay Sternberg * setting up the full range we need. 145057dacad5SJay Sternberg */ 1451dca51e78SChristoph Hellwig pci_free_irq_vectors(pdev); 1452dca51e78SChristoph Hellwig nr_io_queues = pci_alloc_irq_vectors(pdev, 1, nr_io_queues, 1453dca51e78SChristoph Hellwig PCI_IRQ_ALL_TYPES | PCI_IRQ_AFFINITY); 1454dca51e78SChristoph Hellwig if (nr_io_queues <= 0) 1455dca51e78SChristoph Hellwig return -EIO; 1456dca51e78SChristoph Hellwig dev->max_qid = nr_io_queues; 145757dacad5SJay Sternberg 145857dacad5SJay Sternberg /* 145957dacad5SJay Sternberg * Should investigate if there's a performance win from allocating 146057dacad5SJay Sternberg * more queues than interrupt vectors; it might allow the submission 146157dacad5SJay Sternberg * path to scale better, even if the receive path is limited by the 146257dacad5SJay Sternberg * number of interrupts. 146357dacad5SJay Sternberg */ 146457dacad5SJay Sternberg 1465dca51e78SChristoph Hellwig result = queue_request_irq(adminq); 146657dacad5SJay Sternberg if (result) { 146757dacad5SJay Sternberg adminq->cq_vector = -1; 1468d4875622SKeith Busch return result; 146957dacad5SJay Sternberg } 1470749941f2SChristoph Hellwig return nvme_create_io_queues(dev); 147157dacad5SJay Sternberg } 147257dacad5SJay Sternberg 1473db3cbfffSKeith Busch static void nvme_del_queue_end(struct request *req, int error) 1474db3cbfffSKeith Busch { 1475db3cbfffSKeith Busch struct nvme_queue *nvmeq = req->end_io_data; 1476db3cbfffSKeith Busch 1477db3cbfffSKeith Busch blk_mq_free_request(req); 1478db3cbfffSKeith Busch complete(&nvmeq->dev->ioq_wait); 1479db3cbfffSKeith Busch } 1480db3cbfffSKeith Busch 1481db3cbfffSKeith Busch static void nvme_del_cq_end(struct request *req, int error) 1482db3cbfffSKeith Busch { 1483db3cbfffSKeith Busch struct nvme_queue *nvmeq = req->end_io_data; 1484db3cbfffSKeith Busch 1485db3cbfffSKeith Busch if (!error) { 1486db3cbfffSKeith Busch unsigned long flags; 1487db3cbfffSKeith Busch 14882e39e0f6SMing Lin /* 14892e39e0f6SMing Lin * We might be called with the AQ q_lock held 14902e39e0f6SMing Lin * and the I/O queue q_lock should always 14912e39e0f6SMing Lin * nest inside the AQ one. 14922e39e0f6SMing Lin */ 14932e39e0f6SMing Lin spin_lock_irqsave_nested(&nvmeq->q_lock, flags, 14942e39e0f6SMing Lin SINGLE_DEPTH_NESTING); 1495db3cbfffSKeith Busch nvme_process_cq(nvmeq); 1496db3cbfffSKeith Busch spin_unlock_irqrestore(&nvmeq->q_lock, flags); 1497db3cbfffSKeith Busch } 1498db3cbfffSKeith Busch 1499db3cbfffSKeith Busch nvme_del_queue_end(req, error); 1500db3cbfffSKeith Busch } 1501db3cbfffSKeith Busch 1502db3cbfffSKeith Busch static int nvme_delete_queue(struct nvme_queue *nvmeq, u8 opcode) 1503db3cbfffSKeith Busch { 1504db3cbfffSKeith Busch struct request_queue *q = nvmeq->dev->ctrl.admin_q; 1505db3cbfffSKeith Busch struct request *req; 1506db3cbfffSKeith Busch struct nvme_command cmd; 1507db3cbfffSKeith Busch 1508db3cbfffSKeith Busch memset(&cmd, 0, sizeof(cmd)); 1509db3cbfffSKeith Busch cmd.delete_queue.opcode = opcode; 1510db3cbfffSKeith Busch cmd.delete_queue.qid = cpu_to_le16(nvmeq->qid); 1511db3cbfffSKeith Busch 1512eb71f435SChristoph Hellwig req = nvme_alloc_request(q, &cmd, BLK_MQ_REQ_NOWAIT, NVME_QID_ANY); 1513db3cbfffSKeith Busch if (IS_ERR(req)) 1514db3cbfffSKeith Busch return PTR_ERR(req); 1515db3cbfffSKeith Busch 1516db3cbfffSKeith Busch req->timeout = ADMIN_TIMEOUT; 1517db3cbfffSKeith Busch req->end_io_data = nvmeq; 1518db3cbfffSKeith Busch 1519db3cbfffSKeith Busch blk_execute_rq_nowait(q, NULL, req, false, 1520db3cbfffSKeith Busch opcode == nvme_admin_delete_cq ? 1521db3cbfffSKeith Busch nvme_del_cq_end : nvme_del_queue_end); 1522db3cbfffSKeith Busch return 0; 1523db3cbfffSKeith Busch } 1524db3cbfffSKeith Busch 152570659060SKeith Busch static void nvme_disable_io_queues(struct nvme_dev *dev, int queues) 1526db3cbfffSKeith Busch { 152770659060SKeith Busch int pass; 1528db3cbfffSKeith Busch unsigned long timeout; 1529db3cbfffSKeith Busch u8 opcode = nvme_admin_delete_sq; 1530db3cbfffSKeith Busch 1531db3cbfffSKeith Busch for (pass = 0; pass < 2; pass++) { 1532014a0d60SKeith Busch int sent = 0, i = queues; 1533db3cbfffSKeith Busch 1534db3cbfffSKeith Busch reinit_completion(&dev->ioq_wait); 1535db3cbfffSKeith Busch retry: 1536db3cbfffSKeith Busch timeout = ADMIN_TIMEOUT; 1537c21377f8SGabriel Krisman Bertazi for (; i > 0; i--, sent++) 1538c21377f8SGabriel Krisman Bertazi if (nvme_delete_queue(dev->queues[i], opcode)) 1539db3cbfffSKeith Busch break; 1540c21377f8SGabriel Krisman Bertazi 1541db3cbfffSKeith Busch while (sent--) { 1542db3cbfffSKeith Busch timeout = wait_for_completion_io_timeout(&dev->ioq_wait, timeout); 1543db3cbfffSKeith Busch if (timeout == 0) 1544db3cbfffSKeith Busch return; 1545db3cbfffSKeith Busch if (i) 1546db3cbfffSKeith Busch goto retry; 1547db3cbfffSKeith Busch } 1548db3cbfffSKeith Busch opcode = nvme_admin_delete_cq; 1549db3cbfffSKeith Busch } 1550db3cbfffSKeith Busch } 1551db3cbfffSKeith Busch 155257dacad5SJay Sternberg /* 155357dacad5SJay Sternberg * Return: error value if an error occurred setting up the queues or calling 155457dacad5SJay Sternberg * Identify Device. 0 if these succeeded, even if adding some of the 155557dacad5SJay Sternberg * namespaces failed. At the moment, these failures are silent. TBD which 155657dacad5SJay Sternberg * failures should be reported. 155757dacad5SJay Sternberg */ 155857dacad5SJay Sternberg static int nvme_dev_add(struct nvme_dev *dev) 155957dacad5SJay Sternberg { 15605bae7f73SChristoph Hellwig if (!dev->ctrl.tagset) { 156157dacad5SJay Sternberg dev->tagset.ops = &nvme_mq_ops; 156257dacad5SJay Sternberg dev->tagset.nr_hw_queues = dev->online_queues - 1; 156357dacad5SJay Sternberg dev->tagset.timeout = NVME_IO_TIMEOUT; 156457dacad5SJay Sternberg dev->tagset.numa_node = dev_to_node(dev->dev); 156557dacad5SJay Sternberg dev->tagset.queue_depth = 156657dacad5SJay Sternberg min_t(int, dev->q_depth, BLK_MQ_MAX_DEPTH) - 1; 156757dacad5SJay Sternberg dev->tagset.cmd_size = nvme_cmd_size(dev); 156857dacad5SJay Sternberg dev->tagset.flags = BLK_MQ_F_SHOULD_MERGE; 156957dacad5SJay Sternberg dev->tagset.driver_data = dev; 157057dacad5SJay Sternberg 157157dacad5SJay Sternberg if (blk_mq_alloc_tag_set(&dev->tagset)) 157257dacad5SJay Sternberg return 0; 15735bae7f73SChristoph Hellwig dev->ctrl.tagset = &dev->tagset; 1574949928c1SKeith Busch } else { 1575949928c1SKeith Busch blk_mq_update_nr_hw_queues(&dev->tagset, dev->online_queues - 1); 1576949928c1SKeith Busch 1577949928c1SKeith Busch /* Free previously allocated queues that are no longer usable */ 1578949928c1SKeith Busch nvme_free_queues(dev, dev->online_queues); 157957dacad5SJay Sternberg } 1580949928c1SKeith Busch 158157dacad5SJay Sternberg return 0; 158257dacad5SJay Sternberg } 158357dacad5SJay Sternberg 1584b00a726aSKeith Busch static int nvme_pci_enable(struct nvme_dev *dev) 158557dacad5SJay Sternberg { 158657dacad5SJay Sternberg u64 cap; 1587b00a726aSKeith Busch int result = -ENOMEM; 158857dacad5SJay Sternberg struct pci_dev *pdev = to_pci_dev(dev->dev); 158957dacad5SJay Sternberg 159057dacad5SJay Sternberg if (pci_enable_device_mem(pdev)) 159157dacad5SJay Sternberg return result; 159257dacad5SJay Sternberg 159357dacad5SJay Sternberg pci_set_master(pdev); 159457dacad5SJay Sternberg 159557dacad5SJay Sternberg if (dma_set_mask_and_coherent(dev->dev, DMA_BIT_MASK(64)) && 159657dacad5SJay Sternberg dma_set_mask_and_coherent(dev->dev, DMA_BIT_MASK(32))) 159757dacad5SJay Sternberg goto disable; 159857dacad5SJay Sternberg 15997a67cbeaSChristoph Hellwig if (readl(dev->bar + NVME_REG_CSTS) == -1) { 160057dacad5SJay Sternberg result = -ENODEV; 1601b00a726aSKeith Busch goto disable; 160257dacad5SJay Sternberg } 160357dacad5SJay Sternberg 160457dacad5SJay Sternberg /* 1605a5229050SKeith Busch * Some devices and/or platforms don't advertise or work with INTx 1606a5229050SKeith Busch * interrupts. Pre-enable a single MSIX or MSI vec for setup. We'll 1607a5229050SKeith Busch * adjust this later. 160857dacad5SJay Sternberg */ 1609dca51e78SChristoph Hellwig result = pci_alloc_irq_vectors(pdev, 1, 1, PCI_IRQ_ALL_TYPES); 1610dca51e78SChristoph Hellwig if (result < 0) 1611dca51e78SChristoph Hellwig return result; 161257dacad5SJay Sternberg 16137a67cbeaSChristoph Hellwig cap = lo_hi_readq(dev->bar + NVME_REG_CAP); 16147a67cbeaSChristoph Hellwig 161557dacad5SJay Sternberg dev->q_depth = min_t(int, NVME_CAP_MQES(cap) + 1, NVME_Q_DEPTH); 161657dacad5SJay Sternberg dev->db_stride = 1 << NVME_CAP_STRIDE(cap); 16177a67cbeaSChristoph Hellwig dev->dbs = dev->bar + 4096; 16181f390c1fSStephan Günther 16191f390c1fSStephan Günther /* 16201f390c1fSStephan Günther * Temporary fix for the Apple controller found in the MacBook8,1 and 16211f390c1fSStephan Günther * some MacBook7,1 to avoid controller resets and data loss. 16221f390c1fSStephan Günther */ 16231f390c1fSStephan Günther if (pdev->vendor == PCI_VENDOR_ID_APPLE && pdev->device == 0x2001) { 16241f390c1fSStephan Günther dev->q_depth = 2; 16251f390c1fSStephan Günther dev_warn(dev->dev, "detected Apple NVMe controller, set " 16261f390c1fSStephan Günther "queue depth=%u to work around controller resets\n", 16271f390c1fSStephan Günther dev->q_depth); 16281f390c1fSStephan Günther } 16291f390c1fSStephan Günther 1630202021c1SStephen Bates /* 1631202021c1SStephen Bates * CMBs can currently only exist on >=1.2 PCIe devices. We only 1632202021c1SStephen Bates * populate sysfs if a CMB is implemented. Note that we add the 1633202021c1SStephen Bates * CMB attribute to the nvme_ctrl kobj which removes the need to remove 1634202021c1SStephen Bates * it on exit. Since nvme_dev_attrs_group has no name we can pass 1635202021c1SStephen Bates * NULL as final argument to sysfs_add_file_to_group. 1636202021c1SStephen Bates */ 1637202021c1SStephen Bates 16388ef2074dSGabriel Krisman Bertazi if (readl(dev->bar + NVME_REG_VS) >= NVME_VS(1, 2, 0)) { 163957dacad5SJay Sternberg dev->cmb = nvme_map_cmb(dev); 164057dacad5SJay Sternberg 1641202021c1SStephen Bates if (dev->cmbsz) { 1642202021c1SStephen Bates if (sysfs_add_file_to_group(&dev->ctrl.device->kobj, 1643202021c1SStephen Bates &dev_attr_cmb.attr, NULL)) 1644202021c1SStephen Bates dev_warn(dev->dev, 1645202021c1SStephen Bates "failed to add sysfs attribute for CMB\n"); 1646202021c1SStephen Bates } 1647202021c1SStephen Bates } 1648202021c1SStephen Bates 1649a0a3408eSKeith Busch pci_enable_pcie_error_reporting(pdev); 1650a0a3408eSKeith Busch pci_save_state(pdev); 165157dacad5SJay Sternberg return 0; 165257dacad5SJay Sternberg 165357dacad5SJay Sternberg disable: 165457dacad5SJay Sternberg pci_disable_device(pdev); 165557dacad5SJay Sternberg return result; 165657dacad5SJay Sternberg } 165757dacad5SJay Sternberg 165857dacad5SJay Sternberg static void nvme_dev_unmap(struct nvme_dev *dev) 165957dacad5SJay Sternberg { 1660b00a726aSKeith Busch if (dev->bar) 1661b00a726aSKeith Busch iounmap(dev->bar); 1662a1f447b3SJohannes Thumshirn pci_release_mem_regions(to_pci_dev(dev->dev)); 1663b00a726aSKeith Busch } 1664b00a726aSKeith Busch 1665b00a726aSKeith Busch static void nvme_pci_disable(struct nvme_dev *dev) 1666b00a726aSKeith Busch { 166757dacad5SJay Sternberg struct pci_dev *pdev = to_pci_dev(dev->dev); 166857dacad5SJay Sternberg 1669dca51e78SChristoph Hellwig pci_free_irq_vectors(pdev); 167057dacad5SJay Sternberg 1671a0a3408eSKeith Busch if (pci_is_enabled(pdev)) { 1672a0a3408eSKeith Busch pci_disable_pcie_error_reporting(pdev); 167357dacad5SJay Sternberg pci_disable_device(pdev); 167457dacad5SJay Sternberg } 1675a0a3408eSKeith Busch } 167657dacad5SJay Sternberg 1677a5cdb68cSKeith Busch static void nvme_dev_disable(struct nvme_dev *dev, bool shutdown) 167857dacad5SJay Sternberg { 167970659060SKeith Busch int i, queues; 168057dacad5SJay Sternberg u32 csts = -1; 168157dacad5SJay Sternberg 16822d55cd5fSChristoph Hellwig del_timer_sync(&dev->watchdog_timer); 168357dacad5SJay Sternberg 168477bf25eaSKeith Busch mutex_lock(&dev->shutdown_lock); 1685b00a726aSKeith Busch if (pci_is_enabled(to_pci_dev(dev->dev))) { 168625646264SKeith Busch nvme_stop_queues(&dev->ctrl); 16877a67cbeaSChristoph Hellwig csts = readl(dev->bar + NVME_REG_CSTS); 168857dacad5SJay Sternberg } 1689c21377f8SGabriel Krisman Bertazi 169070659060SKeith Busch queues = dev->online_queues - 1; 1691c21377f8SGabriel Krisman Bertazi for (i = dev->queue_count - 1; i > 0; i--) 1692c21377f8SGabriel Krisman Bertazi nvme_suspend_queue(dev->queues[i]); 1693c21377f8SGabriel Krisman Bertazi 169457dacad5SJay Sternberg if (csts & NVME_CSTS_CFS || !(csts & NVME_CSTS_RDY)) { 169582469c59SGabriel Krisman Bertazi /* A device might become IO incapable very soon during 169682469c59SGabriel Krisman Bertazi * probe, before the admin queue is configured. Thus, 169782469c59SGabriel Krisman Bertazi * queue_count can be 0 here. 169882469c59SGabriel Krisman Bertazi */ 169982469c59SGabriel Krisman Bertazi if (dev->queue_count) 1700c21377f8SGabriel Krisman Bertazi nvme_suspend_queue(dev->queues[0]); 170157dacad5SJay Sternberg } else { 170270659060SKeith Busch nvme_disable_io_queues(dev, queues); 1703a5cdb68cSKeith Busch nvme_disable_admin_queue(dev, shutdown); 170457dacad5SJay Sternberg } 1705b00a726aSKeith Busch nvme_pci_disable(dev); 170657dacad5SJay Sternberg 1707e1958e65SMing Lin blk_mq_tagset_busy_iter(&dev->tagset, nvme_cancel_request, &dev->ctrl); 1708e1958e65SMing Lin blk_mq_tagset_busy_iter(&dev->admin_tagset, nvme_cancel_request, &dev->ctrl); 170977bf25eaSKeith Busch mutex_unlock(&dev->shutdown_lock); 171057dacad5SJay Sternberg } 171157dacad5SJay Sternberg 171257dacad5SJay Sternberg static int nvme_setup_prp_pools(struct nvme_dev *dev) 171357dacad5SJay Sternberg { 171457dacad5SJay Sternberg dev->prp_page_pool = dma_pool_create("prp list page", dev->dev, 171557dacad5SJay Sternberg PAGE_SIZE, PAGE_SIZE, 0); 171657dacad5SJay Sternberg if (!dev->prp_page_pool) 171757dacad5SJay Sternberg return -ENOMEM; 171857dacad5SJay Sternberg 171957dacad5SJay Sternberg /* Optimisation for I/Os between 4k and 128k */ 172057dacad5SJay Sternberg dev->prp_small_pool = dma_pool_create("prp list 256", dev->dev, 172157dacad5SJay Sternberg 256, 256, 0); 172257dacad5SJay Sternberg if (!dev->prp_small_pool) { 172357dacad5SJay Sternberg dma_pool_destroy(dev->prp_page_pool); 172457dacad5SJay Sternberg return -ENOMEM; 172557dacad5SJay Sternberg } 172657dacad5SJay Sternberg return 0; 172757dacad5SJay Sternberg } 172857dacad5SJay Sternberg 172957dacad5SJay Sternberg static void nvme_release_prp_pools(struct nvme_dev *dev) 173057dacad5SJay Sternberg { 173157dacad5SJay Sternberg dma_pool_destroy(dev->prp_page_pool); 173257dacad5SJay Sternberg dma_pool_destroy(dev->prp_small_pool); 173357dacad5SJay Sternberg } 173457dacad5SJay Sternberg 17351673f1f0SChristoph Hellwig static void nvme_pci_free_ctrl(struct nvme_ctrl *ctrl) 173657dacad5SJay Sternberg { 17371673f1f0SChristoph Hellwig struct nvme_dev *dev = to_nvme_dev(ctrl); 173857dacad5SJay Sternberg 173957dacad5SJay Sternberg put_device(dev->dev); 174057dacad5SJay Sternberg if (dev->tagset.tags) 174157dacad5SJay Sternberg blk_mq_free_tag_set(&dev->tagset); 17421c63dc66SChristoph Hellwig if (dev->ctrl.admin_q) 17431c63dc66SChristoph Hellwig blk_put_queue(dev->ctrl.admin_q); 174457dacad5SJay Sternberg kfree(dev->queues); 1745*4f1244c8SChristoph Hellwig kfree(dev->ctrl.opal_dev); 174657dacad5SJay Sternberg kfree(dev); 174757dacad5SJay Sternberg } 174857dacad5SJay Sternberg 1749f58944e2SKeith Busch static void nvme_remove_dead_ctrl(struct nvme_dev *dev, int status) 1750f58944e2SKeith Busch { 1751237045fcSLinus Torvalds dev_warn(dev->ctrl.device, "Removing after probe failure status: %d\n", status); 1752f58944e2SKeith Busch 1753f58944e2SKeith Busch kref_get(&dev->ctrl.kref); 175469d9a99cSKeith Busch nvme_dev_disable(dev, false); 1755f58944e2SKeith Busch if (!schedule_work(&dev->remove_work)) 1756f58944e2SKeith Busch nvme_put_ctrl(&dev->ctrl); 1757f58944e2SKeith Busch } 1758f58944e2SKeith Busch 1759fd634f41SChristoph Hellwig static void nvme_reset_work(struct work_struct *work) 176057dacad5SJay Sternberg { 1761fd634f41SChristoph Hellwig struct nvme_dev *dev = container_of(work, struct nvme_dev, reset_work); 1762a98e58e5SScott Bauer bool was_suspend = !!(dev->ctrl.ctrl_config & NVME_CC_SHN_NORMAL); 1763f58944e2SKeith Busch int result = -ENODEV; 176457dacad5SJay Sternberg 1765bb8d261eSChristoph Hellwig if (WARN_ON(dev->ctrl.state == NVME_CTRL_RESETTING)) 1766fd634f41SChristoph Hellwig goto out; 1767fd634f41SChristoph Hellwig 1768fd634f41SChristoph Hellwig /* 1769fd634f41SChristoph Hellwig * If we're called to reset a live controller first shut it down before 1770fd634f41SChristoph Hellwig * moving on. 1771fd634f41SChristoph Hellwig */ 1772b00a726aSKeith Busch if (dev->ctrl.ctrl_config & NVME_CC_ENABLE) 1773a5cdb68cSKeith Busch nvme_dev_disable(dev, false); 1774fd634f41SChristoph Hellwig 1775bb8d261eSChristoph Hellwig if (!nvme_change_ctrl_state(&dev->ctrl, NVME_CTRL_RESETTING)) 17769bf2b972SKeith Busch goto out; 17779bf2b972SKeith Busch 1778b00a726aSKeith Busch result = nvme_pci_enable(dev); 177957dacad5SJay Sternberg if (result) 178057dacad5SJay Sternberg goto out; 178157dacad5SJay Sternberg 178257dacad5SJay Sternberg result = nvme_configure_admin_queue(dev); 178357dacad5SJay Sternberg if (result) 1784f58944e2SKeith Busch goto out; 178557dacad5SJay Sternberg 178657dacad5SJay Sternberg nvme_init_queue(dev->queues[0], 0); 178757dacad5SJay Sternberg result = nvme_alloc_admin_tags(dev); 178857dacad5SJay Sternberg if (result) 1789f58944e2SKeith Busch goto out; 179057dacad5SJay Sternberg 1791ce4541f4SChristoph Hellwig result = nvme_init_identify(&dev->ctrl); 1792ce4541f4SChristoph Hellwig if (result) 1793f58944e2SKeith Busch goto out; 1794ce4541f4SChristoph Hellwig 1795*4f1244c8SChristoph Hellwig if (!dev->ctrl.opal_dev) { 1796*4f1244c8SChristoph Hellwig dev->ctrl.opal_dev = 1797*4f1244c8SChristoph Hellwig init_opal_dev(&dev->ctrl, &nvme_sec_submit); 1798*4f1244c8SChristoph Hellwig } 1799a98e58e5SScott Bauer 1800a98e58e5SScott Bauer if (was_suspend) 1801*4f1244c8SChristoph Hellwig opal_unlock_from_suspend(dev->ctrl.opal_dev); 1802a98e58e5SScott Bauer 180357dacad5SJay Sternberg result = nvme_setup_io_queues(dev); 180457dacad5SJay Sternberg if (result) 1805f58944e2SKeith Busch goto out; 180657dacad5SJay Sternberg 180721f033f7SKeith Busch /* 180821f033f7SKeith Busch * A controller that can not execute IO typically requires user 180921f033f7SKeith Busch * intervention to correct. For such degraded controllers, the driver 181021f033f7SKeith Busch * should not submit commands the user did not request, so skip 181121f033f7SKeith Busch * registering for asynchronous event notification on this condition. 181221f033f7SKeith Busch */ 1813f866fc42SChristoph Hellwig if (dev->online_queues > 1) 1814f866fc42SChristoph Hellwig nvme_queue_async_events(&dev->ctrl); 181557dacad5SJay Sternberg 18162d55cd5fSChristoph Hellwig mod_timer(&dev->watchdog_timer, round_jiffies(jiffies + HZ)); 181757dacad5SJay Sternberg 181857dacad5SJay Sternberg /* 181957dacad5SJay Sternberg * Keep the controller around but remove all namespaces if we don't have 182057dacad5SJay Sternberg * any working I/O queue. 182157dacad5SJay Sternberg */ 182257dacad5SJay Sternberg if (dev->online_queues < 2) { 18231b3c47c1SSagi Grimberg dev_warn(dev->ctrl.device, "IO queues not created\n"); 18243b24774eSKeith Busch nvme_kill_queues(&dev->ctrl); 18255bae7f73SChristoph Hellwig nvme_remove_namespaces(&dev->ctrl); 182657dacad5SJay Sternberg } else { 182725646264SKeith Busch nvme_start_queues(&dev->ctrl); 182857dacad5SJay Sternberg nvme_dev_add(dev); 182957dacad5SJay Sternberg } 183057dacad5SJay Sternberg 1831bb8d261eSChristoph Hellwig if (!nvme_change_ctrl_state(&dev->ctrl, NVME_CTRL_LIVE)) { 1832bb8d261eSChristoph Hellwig dev_warn(dev->ctrl.device, "failed to mark controller live\n"); 1833bb8d261eSChristoph Hellwig goto out; 1834bb8d261eSChristoph Hellwig } 183592911a55SChristoph Hellwig 183692911a55SChristoph Hellwig if (dev->online_queues > 1) 18375955be21SChristoph Hellwig nvme_queue_scan(&dev->ctrl); 183857dacad5SJay Sternberg return; 183957dacad5SJay Sternberg 184057dacad5SJay Sternberg out: 1841f58944e2SKeith Busch nvme_remove_dead_ctrl(dev, result); 184257dacad5SJay Sternberg } 184357dacad5SJay Sternberg 18445c8809e6SChristoph Hellwig static void nvme_remove_dead_ctrl_work(struct work_struct *work) 184557dacad5SJay Sternberg { 18465c8809e6SChristoph Hellwig struct nvme_dev *dev = container_of(work, struct nvme_dev, remove_work); 184757dacad5SJay Sternberg struct pci_dev *pdev = to_pci_dev(dev->dev); 184857dacad5SJay Sternberg 184969d9a99cSKeith Busch nvme_kill_queues(&dev->ctrl); 185057dacad5SJay Sternberg if (pci_get_drvdata(pdev)) 1851921920abSKeith Busch device_release_driver(&pdev->dev); 18521673f1f0SChristoph Hellwig nvme_put_ctrl(&dev->ctrl); 185357dacad5SJay Sternberg } 185457dacad5SJay Sternberg 185557dacad5SJay Sternberg static int nvme_reset(struct nvme_dev *dev) 185657dacad5SJay Sternberg { 18571c63dc66SChristoph Hellwig if (!dev->ctrl.admin_q || blk_queue_dying(dev->ctrl.admin_q)) 185857dacad5SJay Sternberg return -ENODEV; 1859c5f6ce97SKeith Busch if (work_busy(&dev->reset_work)) 1860c5f6ce97SKeith Busch return -ENODEV; 1861846cc05fSChristoph Hellwig if (!queue_work(nvme_workq, &dev->reset_work)) 1862846cc05fSChristoph Hellwig return -EBUSY; 186357dacad5SJay Sternberg return 0; 186457dacad5SJay Sternberg } 186557dacad5SJay Sternberg 18661c63dc66SChristoph Hellwig static int nvme_pci_reg_read32(struct nvme_ctrl *ctrl, u32 off, u32 *val) 186757dacad5SJay Sternberg { 18681c63dc66SChristoph Hellwig *val = readl(to_nvme_dev(ctrl)->bar + off); 18691c63dc66SChristoph Hellwig return 0; 187057dacad5SJay Sternberg } 18711c63dc66SChristoph Hellwig 18725fd4ce1bSChristoph Hellwig static int nvme_pci_reg_write32(struct nvme_ctrl *ctrl, u32 off, u32 val) 18735fd4ce1bSChristoph Hellwig { 18745fd4ce1bSChristoph Hellwig writel(val, to_nvme_dev(ctrl)->bar + off); 18755fd4ce1bSChristoph Hellwig return 0; 18765fd4ce1bSChristoph Hellwig } 18775fd4ce1bSChristoph Hellwig 18787fd8930fSChristoph Hellwig static int nvme_pci_reg_read64(struct nvme_ctrl *ctrl, u32 off, u64 *val) 18797fd8930fSChristoph Hellwig { 18807fd8930fSChristoph Hellwig *val = readq(to_nvme_dev(ctrl)->bar + off); 18817fd8930fSChristoph Hellwig return 0; 18827fd8930fSChristoph Hellwig } 18837fd8930fSChristoph Hellwig 1884f3ca80fcSChristoph Hellwig static int nvme_pci_reset_ctrl(struct nvme_ctrl *ctrl) 1885f3ca80fcSChristoph Hellwig { 1886c5f6ce97SKeith Busch struct nvme_dev *dev = to_nvme_dev(ctrl); 1887c5f6ce97SKeith Busch int ret = nvme_reset(dev); 1888c5f6ce97SKeith Busch 1889c5f6ce97SKeith Busch if (!ret) 1890c5f6ce97SKeith Busch flush_work(&dev->reset_work); 1891c5f6ce97SKeith Busch return ret; 1892f3ca80fcSChristoph Hellwig } 1893f3ca80fcSChristoph Hellwig 18941c63dc66SChristoph Hellwig static const struct nvme_ctrl_ops nvme_pci_ctrl_ops = { 18951a353d85SMing Lin .name = "pcie", 1896e439bb12SSagi Grimberg .module = THIS_MODULE, 18971c63dc66SChristoph Hellwig .reg_read32 = nvme_pci_reg_read32, 18985fd4ce1bSChristoph Hellwig .reg_write32 = nvme_pci_reg_write32, 18997fd8930fSChristoph Hellwig .reg_read64 = nvme_pci_reg_read64, 1900f3ca80fcSChristoph Hellwig .reset_ctrl = nvme_pci_reset_ctrl, 19011673f1f0SChristoph Hellwig .free_ctrl = nvme_pci_free_ctrl, 1902f866fc42SChristoph Hellwig .submit_async_event = nvme_pci_submit_async_event, 19031c63dc66SChristoph Hellwig }; 190457dacad5SJay Sternberg 1905b00a726aSKeith Busch static int nvme_dev_map(struct nvme_dev *dev) 1906b00a726aSKeith Busch { 1907b00a726aSKeith Busch struct pci_dev *pdev = to_pci_dev(dev->dev); 1908b00a726aSKeith Busch 1909a1f447b3SJohannes Thumshirn if (pci_request_mem_regions(pdev, "nvme")) 1910b00a726aSKeith Busch return -ENODEV; 1911b00a726aSKeith Busch 1912b00a726aSKeith Busch dev->bar = ioremap(pci_resource_start(pdev, 0), 8192); 1913b00a726aSKeith Busch if (!dev->bar) 1914b00a726aSKeith Busch goto release; 1915b00a726aSKeith Busch 1916b00a726aSKeith Busch return 0; 1917b00a726aSKeith Busch release: 1918a1f447b3SJohannes Thumshirn pci_release_mem_regions(pdev); 1919b00a726aSKeith Busch return -ENODEV; 1920b00a726aSKeith Busch } 1921b00a726aSKeith Busch 192257dacad5SJay Sternberg static int nvme_probe(struct pci_dev *pdev, const struct pci_device_id *id) 192357dacad5SJay Sternberg { 192457dacad5SJay Sternberg int node, result = -ENOMEM; 192557dacad5SJay Sternberg struct nvme_dev *dev; 192657dacad5SJay Sternberg 192757dacad5SJay Sternberg node = dev_to_node(&pdev->dev); 192857dacad5SJay Sternberg if (node == NUMA_NO_NODE) 19292fa84351SMasayoshi Mizuma set_dev_node(&pdev->dev, first_memory_node); 193057dacad5SJay Sternberg 193157dacad5SJay Sternberg dev = kzalloc_node(sizeof(*dev), GFP_KERNEL, node); 193257dacad5SJay Sternberg if (!dev) 193357dacad5SJay Sternberg return -ENOMEM; 193457dacad5SJay Sternberg dev->queues = kzalloc_node((num_possible_cpus() + 1) * sizeof(void *), 193557dacad5SJay Sternberg GFP_KERNEL, node); 193657dacad5SJay Sternberg if (!dev->queues) 193757dacad5SJay Sternberg goto free; 193857dacad5SJay Sternberg 193957dacad5SJay Sternberg dev->dev = get_device(&pdev->dev); 194057dacad5SJay Sternberg pci_set_drvdata(pdev, dev); 194157dacad5SJay Sternberg 1942b00a726aSKeith Busch result = nvme_dev_map(dev); 1943b00a726aSKeith Busch if (result) 1944b00a726aSKeith Busch goto free; 1945b00a726aSKeith Busch 1946f3ca80fcSChristoph Hellwig INIT_WORK(&dev->reset_work, nvme_reset_work); 19475c8809e6SChristoph Hellwig INIT_WORK(&dev->remove_work, nvme_remove_dead_ctrl_work); 19482d55cd5fSChristoph Hellwig setup_timer(&dev->watchdog_timer, nvme_watchdog_timer, 19492d55cd5fSChristoph Hellwig (unsigned long)dev); 195077bf25eaSKeith Busch mutex_init(&dev->shutdown_lock); 1951db3cbfffSKeith Busch init_completion(&dev->ioq_wait); 1952f3ca80fcSChristoph Hellwig 1953f3ca80fcSChristoph Hellwig result = nvme_setup_prp_pools(dev); 1954f3ca80fcSChristoph Hellwig if (result) 1955f3ca80fcSChristoph Hellwig goto put_pci; 1956f3ca80fcSChristoph Hellwig 1957f3ca80fcSChristoph Hellwig result = nvme_init_ctrl(&dev->ctrl, &pdev->dev, &nvme_pci_ctrl_ops, 1958f3ca80fcSChristoph Hellwig id->driver_data); 1959f3ca80fcSChristoph Hellwig if (result) 1960f3ca80fcSChristoph Hellwig goto release_pools; 1961f3ca80fcSChristoph Hellwig 19621b3c47c1SSagi Grimberg dev_info(dev->ctrl.device, "pci function %s\n", dev_name(&pdev->dev)); 19631b3c47c1SSagi Grimberg 196492f7a162SKeith Busch queue_work(nvme_workq, &dev->reset_work); 196557dacad5SJay Sternberg return 0; 196657dacad5SJay Sternberg 196757dacad5SJay Sternberg release_pools: 196857dacad5SJay Sternberg nvme_release_prp_pools(dev); 196957dacad5SJay Sternberg put_pci: 197057dacad5SJay Sternberg put_device(dev->dev); 1971b00a726aSKeith Busch nvme_dev_unmap(dev); 197257dacad5SJay Sternberg free: 197357dacad5SJay Sternberg kfree(dev->queues); 197457dacad5SJay Sternberg kfree(dev); 197557dacad5SJay Sternberg return result; 197657dacad5SJay Sternberg } 197757dacad5SJay Sternberg 197857dacad5SJay Sternberg static void nvme_reset_notify(struct pci_dev *pdev, bool prepare) 197957dacad5SJay Sternberg { 198057dacad5SJay Sternberg struct nvme_dev *dev = pci_get_drvdata(pdev); 198157dacad5SJay Sternberg 198257dacad5SJay Sternberg if (prepare) 1983a5cdb68cSKeith Busch nvme_dev_disable(dev, false); 198457dacad5SJay Sternberg else 1985c5f6ce97SKeith Busch nvme_reset(dev); 198657dacad5SJay Sternberg } 198757dacad5SJay Sternberg 198857dacad5SJay Sternberg static void nvme_shutdown(struct pci_dev *pdev) 198957dacad5SJay Sternberg { 199057dacad5SJay Sternberg struct nvme_dev *dev = pci_get_drvdata(pdev); 1991a5cdb68cSKeith Busch nvme_dev_disable(dev, true); 199257dacad5SJay Sternberg } 199357dacad5SJay Sternberg 1994f58944e2SKeith Busch /* 1995f58944e2SKeith Busch * The driver's remove may be called on a device in a partially initialized 1996f58944e2SKeith Busch * state. This function must not have any dependencies on the device state in 1997f58944e2SKeith Busch * order to proceed. 1998f58944e2SKeith Busch */ 199957dacad5SJay Sternberg static void nvme_remove(struct pci_dev *pdev) 200057dacad5SJay Sternberg { 200157dacad5SJay Sternberg struct nvme_dev *dev = pci_get_drvdata(pdev); 200257dacad5SJay Sternberg 2003bb8d261eSChristoph Hellwig nvme_change_ctrl_state(&dev->ctrl, NVME_CTRL_DELETING); 2004bb8d261eSChristoph Hellwig 200557dacad5SJay Sternberg pci_set_drvdata(pdev, NULL); 20060ff9d4e1SKeith Busch 20070ff9d4e1SKeith Busch if (!pci_device_is_present(pdev)) 20080ff9d4e1SKeith Busch nvme_change_ctrl_state(&dev->ctrl, NVME_CTRL_DEAD); 20090ff9d4e1SKeith Busch 20109bf2b972SKeith Busch flush_work(&dev->reset_work); 201153029b04SKeith Busch nvme_uninit_ctrl(&dev->ctrl); 2012a5cdb68cSKeith Busch nvme_dev_disable(dev, true); 201357dacad5SJay Sternberg nvme_dev_remove_admin(dev); 201457dacad5SJay Sternberg nvme_free_queues(dev, 0); 201557dacad5SJay Sternberg nvme_release_cmb(dev); 201657dacad5SJay Sternberg nvme_release_prp_pools(dev); 2017b00a726aSKeith Busch nvme_dev_unmap(dev); 20181673f1f0SChristoph Hellwig nvme_put_ctrl(&dev->ctrl); 201957dacad5SJay Sternberg } 202057dacad5SJay Sternberg 202113880f5bSKeith Busch static int nvme_pci_sriov_configure(struct pci_dev *pdev, int numvfs) 202213880f5bSKeith Busch { 202313880f5bSKeith Busch int ret = 0; 202413880f5bSKeith Busch 202513880f5bSKeith Busch if (numvfs == 0) { 202613880f5bSKeith Busch if (pci_vfs_assigned(pdev)) { 202713880f5bSKeith Busch dev_warn(&pdev->dev, 202813880f5bSKeith Busch "Cannot disable SR-IOV VFs while assigned\n"); 202913880f5bSKeith Busch return -EPERM; 203013880f5bSKeith Busch } 203113880f5bSKeith Busch pci_disable_sriov(pdev); 203213880f5bSKeith Busch return 0; 203313880f5bSKeith Busch } 203413880f5bSKeith Busch 203513880f5bSKeith Busch ret = pci_enable_sriov(pdev, numvfs); 203613880f5bSKeith Busch return ret ? ret : numvfs; 203713880f5bSKeith Busch } 203813880f5bSKeith Busch 203957dacad5SJay Sternberg #ifdef CONFIG_PM_SLEEP 204057dacad5SJay Sternberg static int nvme_suspend(struct device *dev) 204157dacad5SJay Sternberg { 204257dacad5SJay Sternberg struct pci_dev *pdev = to_pci_dev(dev); 204357dacad5SJay Sternberg struct nvme_dev *ndev = pci_get_drvdata(pdev); 204457dacad5SJay Sternberg 2045a5cdb68cSKeith Busch nvme_dev_disable(ndev, true); 204657dacad5SJay Sternberg return 0; 204757dacad5SJay Sternberg } 204857dacad5SJay Sternberg 204957dacad5SJay Sternberg static int nvme_resume(struct device *dev) 205057dacad5SJay Sternberg { 205157dacad5SJay Sternberg struct pci_dev *pdev = to_pci_dev(dev); 205257dacad5SJay Sternberg struct nvme_dev *ndev = pci_get_drvdata(pdev); 205357dacad5SJay Sternberg 2054c5f6ce97SKeith Busch nvme_reset(ndev); 205557dacad5SJay Sternberg return 0; 205657dacad5SJay Sternberg } 205757dacad5SJay Sternberg #endif 205857dacad5SJay Sternberg 205957dacad5SJay Sternberg static SIMPLE_DEV_PM_OPS(nvme_dev_pm_ops, nvme_suspend, nvme_resume); 206057dacad5SJay Sternberg 2061a0a3408eSKeith Busch static pci_ers_result_t nvme_error_detected(struct pci_dev *pdev, 2062a0a3408eSKeith Busch pci_channel_state_t state) 2063a0a3408eSKeith Busch { 2064a0a3408eSKeith Busch struct nvme_dev *dev = pci_get_drvdata(pdev); 2065a0a3408eSKeith Busch 2066a0a3408eSKeith Busch /* 2067a0a3408eSKeith Busch * A frozen channel requires a reset. When detected, this method will 2068a0a3408eSKeith Busch * shutdown the controller to quiesce. The controller will be restarted 2069a0a3408eSKeith Busch * after the slot reset through driver's slot_reset callback. 2070a0a3408eSKeith Busch */ 2071a0a3408eSKeith Busch switch (state) { 2072a0a3408eSKeith Busch case pci_channel_io_normal: 2073a0a3408eSKeith Busch return PCI_ERS_RESULT_CAN_RECOVER; 2074a0a3408eSKeith Busch case pci_channel_io_frozen: 2075d011fb31SKeith Busch dev_warn(dev->ctrl.device, 2076d011fb31SKeith Busch "frozen state error detected, reset controller\n"); 2077a5cdb68cSKeith Busch nvme_dev_disable(dev, false); 2078a0a3408eSKeith Busch return PCI_ERS_RESULT_NEED_RESET; 2079a0a3408eSKeith Busch case pci_channel_io_perm_failure: 2080d011fb31SKeith Busch dev_warn(dev->ctrl.device, 2081d011fb31SKeith Busch "failure state error detected, request disconnect\n"); 2082a0a3408eSKeith Busch return PCI_ERS_RESULT_DISCONNECT; 2083a0a3408eSKeith Busch } 2084a0a3408eSKeith Busch return PCI_ERS_RESULT_NEED_RESET; 2085a0a3408eSKeith Busch } 2086a0a3408eSKeith Busch 2087a0a3408eSKeith Busch static pci_ers_result_t nvme_slot_reset(struct pci_dev *pdev) 2088a0a3408eSKeith Busch { 2089a0a3408eSKeith Busch struct nvme_dev *dev = pci_get_drvdata(pdev); 2090a0a3408eSKeith Busch 20911b3c47c1SSagi Grimberg dev_info(dev->ctrl.device, "restart after slot reset\n"); 2092a0a3408eSKeith Busch pci_restore_state(pdev); 2093c5f6ce97SKeith Busch nvme_reset(dev); 2094a0a3408eSKeith Busch return PCI_ERS_RESULT_RECOVERED; 2095a0a3408eSKeith Busch } 2096a0a3408eSKeith Busch 2097a0a3408eSKeith Busch static void nvme_error_resume(struct pci_dev *pdev) 2098a0a3408eSKeith Busch { 2099a0a3408eSKeith Busch pci_cleanup_aer_uncorrect_error_status(pdev); 2100a0a3408eSKeith Busch } 2101a0a3408eSKeith Busch 210257dacad5SJay Sternberg static const struct pci_error_handlers nvme_err_handler = { 210357dacad5SJay Sternberg .error_detected = nvme_error_detected, 210457dacad5SJay Sternberg .slot_reset = nvme_slot_reset, 210557dacad5SJay Sternberg .resume = nvme_error_resume, 210657dacad5SJay Sternberg .reset_notify = nvme_reset_notify, 210757dacad5SJay Sternberg }; 210857dacad5SJay Sternberg 210957dacad5SJay Sternberg static const struct pci_device_id nvme_id_table[] = { 2110106198edSChristoph Hellwig { PCI_VDEVICE(INTEL, 0x0953), 211108095e70SKeith Busch .driver_data = NVME_QUIRK_STRIPE_SIZE | 211208095e70SKeith Busch NVME_QUIRK_DISCARD_ZEROES, }, 211399466e70SKeith Busch { PCI_VDEVICE(INTEL, 0x0a53), 211499466e70SKeith Busch .driver_data = NVME_QUIRK_STRIPE_SIZE | 211599466e70SKeith Busch NVME_QUIRK_DISCARD_ZEROES, }, 211699466e70SKeith Busch { PCI_VDEVICE(INTEL, 0x0a54), 211799466e70SKeith Busch .driver_data = NVME_QUIRK_STRIPE_SIZE | 211899466e70SKeith Busch NVME_QUIRK_DISCARD_ZEROES, }, 2119540c801cSKeith Busch { PCI_VDEVICE(INTEL, 0x5845), /* Qemu emulated controller */ 2120540c801cSKeith Busch .driver_data = NVME_QUIRK_IDENTIFY_CNS, }, 212154adc010SGuilherme G. Piccoli { PCI_DEVICE(0x1c58, 0x0003), /* HGST adapter */ 212254adc010SGuilherme G. Piccoli .driver_data = NVME_QUIRK_DELAY_BEFORE_CHK_RDY, }, 2123015282c9SWenbo Wang { PCI_DEVICE(0x1c5f, 0x0540), /* Memblaze Pblaze4 adapter */ 2124015282c9SWenbo Wang .driver_data = NVME_QUIRK_DELAY_BEFORE_CHK_RDY, }, 212557dacad5SJay Sternberg { PCI_DEVICE_CLASS(PCI_CLASS_STORAGE_EXPRESS, 0xffffff) }, 2126c74dc780SStephan Günther { PCI_DEVICE(PCI_VENDOR_ID_APPLE, 0x2001) }, 212757dacad5SJay Sternberg { 0, } 212857dacad5SJay Sternberg }; 212957dacad5SJay Sternberg MODULE_DEVICE_TABLE(pci, nvme_id_table); 213057dacad5SJay Sternberg 213157dacad5SJay Sternberg static struct pci_driver nvme_driver = { 213257dacad5SJay Sternberg .name = "nvme", 213357dacad5SJay Sternberg .id_table = nvme_id_table, 213457dacad5SJay Sternberg .probe = nvme_probe, 213557dacad5SJay Sternberg .remove = nvme_remove, 213657dacad5SJay Sternberg .shutdown = nvme_shutdown, 213757dacad5SJay Sternberg .driver = { 213857dacad5SJay Sternberg .pm = &nvme_dev_pm_ops, 213957dacad5SJay Sternberg }, 214013880f5bSKeith Busch .sriov_configure = nvme_pci_sriov_configure, 214157dacad5SJay Sternberg .err_handler = &nvme_err_handler, 214257dacad5SJay Sternberg }; 214357dacad5SJay Sternberg 214457dacad5SJay Sternberg static int __init nvme_init(void) 214557dacad5SJay Sternberg { 214657dacad5SJay Sternberg int result; 214757dacad5SJay Sternberg 214892f7a162SKeith Busch nvme_workq = alloc_workqueue("nvme", WQ_UNBOUND | WQ_MEM_RECLAIM, 0); 214957dacad5SJay Sternberg if (!nvme_workq) 215057dacad5SJay Sternberg return -ENOMEM; 215157dacad5SJay Sternberg 215257dacad5SJay Sternberg result = pci_register_driver(&nvme_driver); 215357dacad5SJay Sternberg if (result) 215457dacad5SJay Sternberg destroy_workqueue(nvme_workq); 215557dacad5SJay Sternberg return result; 215657dacad5SJay Sternberg } 215757dacad5SJay Sternberg 215857dacad5SJay Sternberg static void __exit nvme_exit(void) 215957dacad5SJay Sternberg { 216057dacad5SJay Sternberg pci_unregister_driver(&nvme_driver); 216157dacad5SJay Sternberg destroy_workqueue(nvme_workq); 216257dacad5SJay Sternberg _nvme_check_size(); 216357dacad5SJay Sternberg } 216457dacad5SJay Sternberg 216557dacad5SJay Sternberg MODULE_AUTHOR("Matthew Wilcox <willy@linux.intel.com>"); 216657dacad5SJay Sternberg MODULE_LICENSE("GPL"); 216757dacad5SJay Sternberg MODULE_VERSION("1.0"); 216857dacad5SJay Sternberg module_init(nvme_init); 216957dacad5SJay Sternberg module_exit(nvme_exit); 2170