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> 20ff5350a8SAndy Lutomirski #include <linux/dmi.h> 2157dacad5SJay Sternberg #include <linux/init.h> 2257dacad5SJay Sternberg #include <linux/interrupt.h> 2357dacad5SJay Sternberg #include <linux/io.h> 2457dacad5SJay Sternberg #include <linux/mm.h> 2557dacad5SJay Sternberg #include <linux/module.h> 2677bf25eaSKeith Busch #include <linux/mutex.h> 2757dacad5SJay Sternberg #include <linux/pci.h> 2857dacad5SJay Sternberg #include <linux/poison.h> 2957dacad5SJay Sternberg #include <linux/t10-pi.h> 302d55cd5fSChristoph Hellwig #include <linux/timer.h> 3157dacad5SJay Sternberg #include <linux/types.h> 329cf5c095SLinus Torvalds #include <linux/io-64-nonatomic-lo-hi.h> 331d277a63SKeith Busch #include <asm/unaligned.h> 34a98e58e5SScott Bauer #include <linux/sed-opal.h> 3557dacad5SJay Sternberg 3657dacad5SJay Sternberg #include "nvme.h" 3757dacad5SJay Sternberg 3857dacad5SJay Sternberg #define NVME_Q_DEPTH 1024 3957dacad5SJay Sternberg #define NVME_AQ_DEPTH 256 4057dacad5SJay Sternberg #define SQ_SIZE(depth) (depth * sizeof(struct nvme_command)) 4157dacad5SJay Sternberg #define CQ_SIZE(depth) (depth * sizeof(struct nvme_completion)) 4257dacad5SJay Sternberg 43adf68f21SChristoph Hellwig /* 44adf68f21SChristoph Hellwig * We handle AEN commands ourselves and don't even let the 45adf68f21SChristoph Hellwig * block layer know about them. 46adf68f21SChristoph Hellwig */ 47f866fc42SChristoph Hellwig #define NVME_AQ_BLKMQ_DEPTH (NVME_AQ_DEPTH - NVME_NR_AERS) 48adf68f21SChristoph Hellwig 4957dacad5SJay Sternberg static int use_threaded_interrupts; 5057dacad5SJay Sternberg module_param(use_threaded_interrupts, int, 0); 5157dacad5SJay Sternberg 5257dacad5SJay Sternberg static bool use_cmb_sqes = true; 5357dacad5SJay Sternberg module_param(use_cmb_sqes, bool, 0644); 5457dacad5SJay Sternberg MODULE_PARM_DESC(use_cmb_sqes, "use controller's memory buffer for I/O SQes"); 5557dacad5SJay Sternberg 5687ad72a5SChristoph Hellwig static unsigned int max_host_mem_size_mb = 128; 5787ad72a5SChristoph Hellwig module_param(max_host_mem_size_mb, uint, 0444); 5887ad72a5SChristoph Hellwig MODULE_PARM_DESC(max_host_mem_size_mb, 5987ad72a5SChristoph Hellwig "Maximum Host Memory Buffer (HMB) size per controller (in MiB)"); 6087ad72a5SChristoph Hellwig 611c63dc66SChristoph Hellwig struct nvme_dev; 621c63dc66SChristoph Hellwig struct nvme_queue; 6357dacad5SJay Sternberg 64a0fa9647SJens Axboe static void nvme_process_cq(struct nvme_queue *nvmeq); 65a5cdb68cSKeith Busch static void nvme_dev_disable(struct nvme_dev *dev, bool shutdown); 6657dacad5SJay Sternberg 6757dacad5SJay Sternberg /* 681c63dc66SChristoph Hellwig * Represents an NVM Express device. Each nvme_dev is a PCI function. 691c63dc66SChristoph Hellwig */ 701c63dc66SChristoph Hellwig struct nvme_dev { 711c63dc66SChristoph Hellwig struct nvme_queue **queues; 721c63dc66SChristoph Hellwig struct blk_mq_tag_set tagset; 731c63dc66SChristoph Hellwig struct blk_mq_tag_set admin_tagset; 741c63dc66SChristoph Hellwig u32 __iomem *dbs; 751c63dc66SChristoph Hellwig struct device *dev; 761c63dc66SChristoph Hellwig struct dma_pool *prp_page_pool; 771c63dc66SChristoph Hellwig struct dma_pool *prp_small_pool; 781c63dc66SChristoph Hellwig unsigned queue_count; 791c63dc66SChristoph Hellwig unsigned online_queues; 801c63dc66SChristoph Hellwig unsigned max_qid; 811c63dc66SChristoph Hellwig int q_depth; 821c63dc66SChristoph Hellwig u32 db_stride; 831c63dc66SChristoph Hellwig void __iomem *bar; 8497f6ef64SXu Yu unsigned long bar_mapped_size; 855c8809e6SChristoph Hellwig struct work_struct remove_work; 8677bf25eaSKeith Busch struct mutex shutdown_lock; 871c63dc66SChristoph Hellwig bool subsystem; 881c63dc66SChristoph Hellwig void __iomem *cmb; 891c63dc66SChristoph Hellwig dma_addr_t cmb_dma_addr; 901c63dc66SChristoph Hellwig u64 cmb_size; 911c63dc66SChristoph Hellwig u32 cmbsz; 92202021c1SStephen Bates u32 cmbloc; 931c63dc66SChristoph Hellwig struct nvme_ctrl ctrl; 94db3cbfffSKeith Busch struct completion ioq_wait; 9587ad72a5SChristoph Hellwig 9687ad72a5SChristoph Hellwig /* shadow doorbell buffer support: */ 97f9f38e33SHelen Koike u32 *dbbuf_dbs; 98f9f38e33SHelen Koike dma_addr_t dbbuf_dbs_dma_addr; 99f9f38e33SHelen Koike u32 *dbbuf_eis; 100f9f38e33SHelen Koike dma_addr_t dbbuf_eis_dma_addr; 10187ad72a5SChristoph Hellwig 10287ad72a5SChristoph Hellwig /* host memory buffer support: */ 10387ad72a5SChristoph Hellwig u64 host_mem_size; 10487ad72a5SChristoph Hellwig u32 nr_host_mem_descs; 10587ad72a5SChristoph Hellwig struct nvme_host_mem_buf_desc *host_mem_descs; 10687ad72a5SChristoph Hellwig void **host_mem_desc_bufs; 10757dacad5SJay Sternberg }; 10857dacad5SJay Sternberg 109f9f38e33SHelen Koike static inline unsigned int sq_idx(unsigned int qid, u32 stride) 110f9f38e33SHelen Koike { 111f9f38e33SHelen Koike return qid * 2 * stride; 112f9f38e33SHelen Koike } 113f9f38e33SHelen Koike 114f9f38e33SHelen Koike static inline unsigned int cq_idx(unsigned int qid, u32 stride) 115f9f38e33SHelen Koike { 116f9f38e33SHelen Koike return (qid * 2 + 1) * stride; 117f9f38e33SHelen Koike } 118f9f38e33SHelen Koike 1191c63dc66SChristoph Hellwig static inline struct nvme_dev *to_nvme_dev(struct nvme_ctrl *ctrl) 1201c63dc66SChristoph Hellwig { 1211c63dc66SChristoph Hellwig return container_of(ctrl, struct nvme_dev, ctrl); 1221c63dc66SChristoph Hellwig } 1231c63dc66SChristoph Hellwig 12457dacad5SJay Sternberg /* 12557dacad5SJay Sternberg * An NVM Express queue. Each device has at least two (one for admin 12657dacad5SJay Sternberg * commands and one for I/O commands). 12757dacad5SJay Sternberg */ 12857dacad5SJay Sternberg struct nvme_queue { 12957dacad5SJay Sternberg struct device *q_dmadev; 13057dacad5SJay Sternberg struct nvme_dev *dev; 13157dacad5SJay Sternberg spinlock_t q_lock; 13257dacad5SJay Sternberg struct nvme_command *sq_cmds; 13357dacad5SJay Sternberg struct nvme_command __iomem *sq_cmds_io; 13457dacad5SJay Sternberg volatile struct nvme_completion *cqes; 13557dacad5SJay Sternberg struct blk_mq_tags **tags; 13657dacad5SJay Sternberg dma_addr_t sq_dma_addr; 13757dacad5SJay Sternberg dma_addr_t cq_dma_addr; 13857dacad5SJay Sternberg u32 __iomem *q_db; 13957dacad5SJay Sternberg u16 q_depth; 14057dacad5SJay Sternberg s16 cq_vector; 14157dacad5SJay Sternberg u16 sq_tail; 14257dacad5SJay Sternberg u16 cq_head; 14357dacad5SJay Sternberg u16 qid; 14457dacad5SJay Sternberg u8 cq_phase; 14557dacad5SJay Sternberg u8 cqe_seen; 146f9f38e33SHelen Koike u32 *dbbuf_sq_db; 147f9f38e33SHelen Koike u32 *dbbuf_cq_db; 148f9f38e33SHelen Koike u32 *dbbuf_sq_ei; 149f9f38e33SHelen Koike u32 *dbbuf_cq_ei; 15057dacad5SJay Sternberg }; 15157dacad5SJay Sternberg 15257dacad5SJay Sternberg /* 15371bd150cSChristoph Hellwig * The nvme_iod describes the data in an I/O, including the list of PRP 15471bd150cSChristoph Hellwig * entries. You can't see it in this data structure because C doesn't let 155f4800d6dSChristoph Hellwig * me express that. Use nvme_init_iod to ensure there's enough space 15671bd150cSChristoph Hellwig * allocated to store the PRP list. 15771bd150cSChristoph Hellwig */ 15871bd150cSChristoph Hellwig struct nvme_iod { 159d49187e9SChristoph Hellwig struct nvme_request req; 160f4800d6dSChristoph Hellwig struct nvme_queue *nvmeq; 161f4800d6dSChristoph Hellwig int aborted; 16271bd150cSChristoph Hellwig int npages; /* In the PRP list. 0 means small pool in use */ 16371bd150cSChristoph Hellwig int nents; /* Used in scatterlist */ 16471bd150cSChristoph Hellwig int length; /* Of data, in bytes */ 16571bd150cSChristoph Hellwig dma_addr_t first_dma; 166bf684057SChristoph Hellwig struct scatterlist meta_sg; /* metadata requires single contiguous buffer */ 167f4800d6dSChristoph Hellwig struct scatterlist *sg; 168f4800d6dSChristoph Hellwig struct scatterlist inline_sg[0]; 16957dacad5SJay Sternberg }; 17057dacad5SJay Sternberg 17157dacad5SJay Sternberg /* 17257dacad5SJay Sternberg * Check we didin't inadvertently grow the command struct 17357dacad5SJay Sternberg */ 17457dacad5SJay Sternberg static inline void _nvme_check_size(void) 17557dacad5SJay Sternberg { 17657dacad5SJay Sternberg BUILD_BUG_ON(sizeof(struct nvme_rw_command) != 64); 17757dacad5SJay Sternberg BUILD_BUG_ON(sizeof(struct nvme_create_cq) != 64); 17857dacad5SJay Sternberg BUILD_BUG_ON(sizeof(struct nvme_create_sq) != 64); 17957dacad5SJay Sternberg BUILD_BUG_ON(sizeof(struct nvme_delete_queue) != 64); 18057dacad5SJay Sternberg BUILD_BUG_ON(sizeof(struct nvme_features) != 64); 18157dacad5SJay Sternberg BUILD_BUG_ON(sizeof(struct nvme_format_cmd) != 64); 18257dacad5SJay Sternberg BUILD_BUG_ON(sizeof(struct nvme_abort_cmd) != 64); 18357dacad5SJay Sternberg BUILD_BUG_ON(sizeof(struct nvme_command) != 64); 1840add5e8eSJohannes Thumshirn BUILD_BUG_ON(sizeof(struct nvme_id_ctrl) != NVME_IDENTIFY_DATA_SIZE); 1850add5e8eSJohannes Thumshirn BUILD_BUG_ON(sizeof(struct nvme_id_ns) != NVME_IDENTIFY_DATA_SIZE); 18657dacad5SJay Sternberg BUILD_BUG_ON(sizeof(struct nvme_lba_range_type) != 64); 18757dacad5SJay Sternberg BUILD_BUG_ON(sizeof(struct nvme_smart_log) != 512); 188f9f38e33SHelen Koike BUILD_BUG_ON(sizeof(struct nvme_dbbuf) != 64); 189f9f38e33SHelen Koike } 190f9f38e33SHelen Koike 191f9f38e33SHelen Koike static inline unsigned int nvme_dbbuf_size(u32 stride) 192f9f38e33SHelen Koike { 193f9f38e33SHelen Koike return ((num_possible_cpus() + 1) * 8 * stride); 194f9f38e33SHelen Koike } 195f9f38e33SHelen Koike 196f9f38e33SHelen Koike static int nvme_dbbuf_dma_alloc(struct nvme_dev *dev) 197f9f38e33SHelen Koike { 198f9f38e33SHelen Koike unsigned int mem_size = nvme_dbbuf_size(dev->db_stride); 199f9f38e33SHelen Koike 200f9f38e33SHelen Koike if (dev->dbbuf_dbs) 201f9f38e33SHelen Koike return 0; 202f9f38e33SHelen Koike 203f9f38e33SHelen Koike dev->dbbuf_dbs = dma_alloc_coherent(dev->dev, mem_size, 204f9f38e33SHelen Koike &dev->dbbuf_dbs_dma_addr, 205f9f38e33SHelen Koike GFP_KERNEL); 206f9f38e33SHelen Koike if (!dev->dbbuf_dbs) 207f9f38e33SHelen Koike return -ENOMEM; 208f9f38e33SHelen Koike dev->dbbuf_eis = dma_alloc_coherent(dev->dev, mem_size, 209f9f38e33SHelen Koike &dev->dbbuf_eis_dma_addr, 210f9f38e33SHelen Koike GFP_KERNEL); 211f9f38e33SHelen Koike if (!dev->dbbuf_eis) { 212f9f38e33SHelen Koike dma_free_coherent(dev->dev, mem_size, 213f9f38e33SHelen Koike dev->dbbuf_dbs, dev->dbbuf_dbs_dma_addr); 214f9f38e33SHelen Koike dev->dbbuf_dbs = NULL; 215f9f38e33SHelen Koike return -ENOMEM; 216f9f38e33SHelen Koike } 217f9f38e33SHelen Koike 218f9f38e33SHelen Koike return 0; 219f9f38e33SHelen Koike } 220f9f38e33SHelen Koike 221f9f38e33SHelen Koike static void nvme_dbbuf_dma_free(struct nvme_dev *dev) 222f9f38e33SHelen Koike { 223f9f38e33SHelen Koike unsigned int mem_size = nvme_dbbuf_size(dev->db_stride); 224f9f38e33SHelen Koike 225f9f38e33SHelen Koike if (dev->dbbuf_dbs) { 226f9f38e33SHelen Koike dma_free_coherent(dev->dev, mem_size, 227f9f38e33SHelen Koike dev->dbbuf_dbs, dev->dbbuf_dbs_dma_addr); 228f9f38e33SHelen Koike dev->dbbuf_dbs = NULL; 229f9f38e33SHelen Koike } 230f9f38e33SHelen Koike if (dev->dbbuf_eis) { 231f9f38e33SHelen Koike dma_free_coherent(dev->dev, mem_size, 232f9f38e33SHelen Koike dev->dbbuf_eis, dev->dbbuf_eis_dma_addr); 233f9f38e33SHelen Koike dev->dbbuf_eis = NULL; 234f9f38e33SHelen Koike } 235f9f38e33SHelen Koike } 236f9f38e33SHelen Koike 237f9f38e33SHelen Koike static void nvme_dbbuf_init(struct nvme_dev *dev, 238f9f38e33SHelen Koike struct nvme_queue *nvmeq, int qid) 239f9f38e33SHelen Koike { 240f9f38e33SHelen Koike if (!dev->dbbuf_dbs || !qid) 241f9f38e33SHelen Koike return; 242f9f38e33SHelen Koike 243f9f38e33SHelen Koike nvmeq->dbbuf_sq_db = &dev->dbbuf_dbs[sq_idx(qid, dev->db_stride)]; 244f9f38e33SHelen Koike nvmeq->dbbuf_cq_db = &dev->dbbuf_dbs[cq_idx(qid, dev->db_stride)]; 245f9f38e33SHelen Koike nvmeq->dbbuf_sq_ei = &dev->dbbuf_eis[sq_idx(qid, dev->db_stride)]; 246f9f38e33SHelen Koike nvmeq->dbbuf_cq_ei = &dev->dbbuf_eis[cq_idx(qid, dev->db_stride)]; 247f9f38e33SHelen Koike } 248f9f38e33SHelen Koike 249f9f38e33SHelen Koike static void nvme_dbbuf_set(struct nvme_dev *dev) 250f9f38e33SHelen Koike { 251f9f38e33SHelen Koike struct nvme_command c; 252f9f38e33SHelen Koike 253f9f38e33SHelen Koike if (!dev->dbbuf_dbs) 254f9f38e33SHelen Koike return; 255f9f38e33SHelen Koike 256f9f38e33SHelen Koike memset(&c, 0, sizeof(c)); 257f9f38e33SHelen Koike c.dbbuf.opcode = nvme_admin_dbbuf; 258f9f38e33SHelen Koike c.dbbuf.prp1 = cpu_to_le64(dev->dbbuf_dbs_dma_addr); 259f9f38e33SHelen Koike c.dbbuf.prp2 = cpu_to_le64(dev->dbbuf_eis_dma_addr); 260f9f38e33SHelen Koike 261f9f38e33SHelen Koike if (nvme_submit_sync_cmd(dev->ctrl.admin_q, &c, NULL, 0)) { 2629bdcfb10SChristoph Hellwig dev_warn(dev->ctrl.device, "unable to set dbbuf\n"); 263f9f38e33SHelen Koike /* Free memory and continue on */ 264f9f38e33SHelen Koike nvme_dbbuf_dma_free(dev); 265f9f38e33SHelen Koike } 266f9f38e33SHelen Koike } 267f9f38e33SHelen Koike 268f9f38e33SHelen Koike static inline int nvme_dbbuf_need_event(u16 event_idx, u16 new_idx, u16 old) 269f9f38e33SHelen Koike { 270f9f38e33SHelen Koike return (u16)(new_idx - event_idx - 1) < (u16)(new_idx - old); 271f9f38e33SHelen Koike } 272f9f38e33SHelen Koike 273f9f38e33SHelen Koike /* Update dbbuf and return true if an MMIO is required */ 274f9f38e33SHelen Koike static bool nvme_dbbuf_update_and_check_event(u16 value, u32 *dbbuf_db, 275f9f38e33SHelen Koike volatile u32 *dbbuf_ei) 276f9f38e33SHelen Koike { 277f9f38e33SHelen Koike if (dbbuf_db) { 278f9f38e33SHelen Koike u16 old_value; 279f9f38e33SHelen Koike 280f9f38e33SHelen Koike /* 281f9f38e33SHelen Koike * Ensure that the queue is written before updating 282f9f38e33SHelen Koike * the doorbell in memory 283f9f38e33SHelen Koike */ 284f9f38e33SHelen Koike wmb(); 285f9f38e33SHelen Koike 286f9f38e33SHelen Koike old_value = *dbbuf_db; 287f9f38e33SHelen Koike *dbbuf_db = value; 288f9f38e33SHelen Koike 289f9f38e33SHelen Koike if (!nvme_dbbuf_need_event(*dbbuf_ei, value, old_value)) 290f9f38e33SHelen Koike return false; 291f9f38e33SHelen Koike } 292f9f38e33SHelen Koike 293f9f38e33SHelen Koike return true; 29457dacad5SJay Sternberg } 29557dacad5SJay Sternberg 29657dacad5SJay Sternberg /* 29757dacad5SJay Sternberg * Max size of iod being embedded in the request payload 29857dacad5SJay Sternberg */ 29957dacad5SJay Sternberg #define NVME_INT_PAGES 2 3005fd4ce1bSChristoph Hellwig #define NVME_INT_BYTES(dev) (NVME_INT_PAGES * (dev)->ctrl.page_size) 30157dacad5SJay Sternberg 30257dacad5SJay Sternberg /* 30357dacad5SJay Sternberg * Will slightly overestimate the number of pages needed. This is OK 30457dacad5SJay Sternberg * as it only leads to a small amount of wasted memory for the lifetime of 30557dacad5SJay Sternberg * the I/O. 30657dacad5SJay Sternberg */ 30757dacad5SJay Sternberg static int nvme_npages(unsigned size, struct nvme_dev *dev) 30857dacad5SJay Sternberg { 3095fd4ce1bSChristoph Hellwig unsigned nprps = DIV_ROUND_UP(size + dev->ctrl.page_size, 3105fd4ce1bSChristoph Hellwig dev->ctrl.page_size); 31157dacad5SJay Sternberg return DIV_ROUND_UP(8 * nprps, PAGE_SIZE - 8); 31257dacad5SJay Sternberg } 31357dacad5SJay Sternberg 314f4800d6dSChristoph Hellwig static unsigned int nvme_iod_alloc_size(struct nvme_dev *dev, 315f4800d6dSChristoph Hellwig unsigned int size, unsigned int nseg) 316f4800d6dSChristoph Hellwig { 317f4800d6dSChristoph Hellwig return sizeof(__le64 *) * nvme_npages(size, dev) + 318f4800d6dSChristoph Hellwig sizeof(struct scatterlist) * nseg; 319f4800d6dSChristoph Hellwig } 320f4800d6dSChristoph Hellwig 32157dacad5SJay Sternberg static unsigned int nvme_cmd_size(struct nvme_dev *dev) 32257dacad5SJay Sternberg { 323f4800d6dSChristoph Hellwig return sizeof(struct nvme_iod) + 324f4800d6dSChristoph Hellwig nvme_iod_alloc_size(dev, NVME_INT_BYTES(dev), NVME_INT_PAGES); 32557dacad5SJay Sternberg } 32657dacad5SJay Sternberg 32757dacad5SJay Sternberg static int nvme_admin_init_hctx(struct blk_mq_hw_ctx *hctx, void *data, 32857dacad5SJay Sternberg unsigned int hctx_idx) 32957dacad5SJay Sternberg { 33057dacad5SJay Sternberg struct nvme_dev *dev = data; 33157dacad5SJay Sternberg struct nvme_queue *nvmeq = dev->queues[0]; 33257dacad5SJay Sternberg 33357dacad5SJay Sternberg WARN_ON(hctx_idx != 0); 33457dacad5SJay Sternberg WARN_ON(dev->admin_tagset.tags[0] != hctx->tags); 33557dacad5SJay Sternberg WARN_ON(nvmeq->tags); 33657dacad5SJay Sternberg 33757dacad5SJay Sternberg hctx->driver_data = nvmeq; 33857dacad5SJay Sternberg nvmeq->tags = &dev->admin_tagset.tags[0]; 33957dacad5SJay Sternberg return 0; 34057dacad5SJay Sternberg } 34157dacad5SJay Sternberg 34257dacad5SJay Sternberg static void nvme_admin_exit_hctx(struct blk_mq_hw_ctx *hctx, unsigned int hctx_idx) 34357dacad5SJay Sternberg { 34457dacad5SJay Sternberg struct nvme_queue *nvmeq = hctx->driver_data; 34557dacad5SJay Sternberg 34657dacad5SJay Sternberg nvmeq->tags = NULL; 34757dacad5SJay Sternberg } 34857dacad5SJay Sternberg 34957dacad5SJay Sternberg static int nvme_init_hctx(struct blk_mq_hw_ctx *hctx, void *data, 35057dacad5SJay Sternberg unsigned int hctx_idx) 35157dacad5SJay Sternberg { 35257dacad5SJay Sternberg struct nvme_dev *dev = data; 35357dacad5SJay Sternberg struct nvme_queue *nvmeq = dev->queues[hctx_idx + 1]; 35457dacad5SJay Sternberg 35557dacad5SJay Sternberg if (!nvmeq->tags) 35657dacad5SJay Sternberg nvmeq->tags = &dev->tagset.tags[hctx_idx]; 35757dacad5SJay Sternberg 35857dacad5SJay Sternberg WARN_ON(dev->tagset.tags[hctx_idx] != hctx->tags); 35957dacad5SJay Sternberg hctx->driver_data = nvmeq; 36057dacad5SJay Sternberg return 0; 36157dacad5SJay Sternberg } 36257dacad5SJay Sternberg 363d6296d39SChristoph Hellwig static int nvme_init_request(struct blk_mq_tag_set *set, struct request *req, 364d6296d39SChristoph Hellwig unsigned int hctx_idx, unsigned int numa_node) 36557dacad5SJay Sternberg { 366d6296d39SChristoph Hellwig struct nvme_dev *dev = set->driver_data; 367f4800d6dSChristoph Hellwig struct nvme_iod *iod = blk_mq_rq_to_pdu(req); 3680350815aSChristoph Hellwig int queue_idx = (set == &dev->tagset) ? hctx_idx + 1 : 0; 3690350815aSChristoph Hellwig struct nvme_queue *nvmeq = dev->queues[queue_idx]; 37057dacad5SJay Sternberg 37157dacad5SJay Sternberg BUG_ON(!nvmeq); 372f4800d6dSChristoph Hellwig iod->nvmeq = nvmeq; 37357dacad5SJay Sternberg return 0; 37457dacad5SJay Sternberg } 37557dacad5SJay Sternberg 376dca51e78SChristoph Hellwig static int nvme_pci_map_queues(struct blk_mq_tag_set *set) 377dca51e78SChristoph Hellwig { 378dca51e78SChristoph Hellwig struct nvme_dev *dev = set->driver_data; 379dca51e78SChristoph Hellwig 380dca51e78SChristoph Hellwig return blk_mq_pci_map_queues(set, to_pci_dev(dev->dev)); 381dca51e78SChristoph Hellwig } 382dca51e78SChristoph Hellwig 38357dacad5SJay Sternberg /** 384adf68f21SChristoph Hellwig * __nvme_submit_cmd() - Copy a command into a queue and ring the doorbell 38557dacad5SJay Sternberg * @nvmeq: The queue to use 38657dacad5SJay Sternberg * @cmd: The command to send 38757dacad5SJay Sternberg * 38857dacad5SJay Sternberg * Safe to use from interrupt context 38957dacad5SJay Sternberg */ 39057dacad5SJay Sternberg static void __nvme_submit_cmd(struct nvme_queue *nvmeq, 39157dacad5SJay Sternberg struct nvme_command *cmd) 39257dacad5SJay Sternberg { 39357dacad5SJay Sternberg u16 tail = nvmeq->sq_tail; 39457dacad5SJay Sternberg 39557dacad5SJay Sternberg if (nvmeq->sq_cmds_io) 39657dacad5SJay Sternberg memcpy_toio(&nvmeq->sq_cmds_io[tail], cmd, sizeof(*cmd)); 39757dacad5SJay Sternberg else 39857dacad5SJay Sternberg memcpy(&nvmeq->sq_cmds[tail], cmd, sizeof(*cmd)); 39957dacad5SJay Sternberg 40057dacad5SJay Sternberg if (++tail == nvmeq->q_depth) 40157dacad5SJay Sternberg tail = 0; 402f9f38e33SHelen Koike if (nvme_dbbuf_update_and_check_event(tail, nvmeq->dbbuf_sq_db, 403f9f38e33SHelen Koike nvmeq->dbbuf_sq_ei)) 40457dacad5SJay Sternberg writel(tail, nvmeq->q_db); 40557dacad5SJay Sternberg nvmeq->sq_tail = tail; 40657dacad5SJay Sternberg } 40757dacad5SJay Sternberg 408f4800d6dSChristoph Hellwig static __le64 **iod_list(struct request *req) 40957dacad5SJay Sternberg { 410f4800d6dSChristoph Hellwig struct nvme_iod *iod = blk_mq_rq_to_pdu(req); 411f9d03f96SChristoph Hellwig return (__le64 **)(iod->sg + blk_rq_nr_phys_segments(req)); 41257dacad5SJay Sternberg } 41357dacad5SJay Sternberg 414fc17b653SChristoph Hellwig static blk_status_t nvme_init_iod(struct request *rq, struct nvme_dev *dev) 41557dacad5SJay Sternberg { 416f4800d6dSChristoph Hellwig struct nvme_iod *iod = blk_mq_rq_to_pdu(rq); 417f9d03f96SChristoph Hellwig int nseg = blk_rq_nr_phys_segments(rq); 418b131c61dSChristoph Hellwig unsigned int size = blk_rq_payload_bytes(rq); 419f4800d6dSChristoph Hellwig 420f4800d6dSChristoph Hellwig if (nseg > NVME_INT_PAGES || size > NVME_INT_BYTES(dev)) { 421f4800d6dSChristoph Hellwig iod->sg = kmalloc(nvme_iod_alloc_size(dev, size, nseg), GFP_ATOMIC); 422f4800d6dSChristoph Hellwig if (!iod->sg) 423fc17b653SChristoph Hellwig return BLK_STS_RESOURCE; 424f4800d6dSChristoph Hellwig } else { 425f4800d6dSChristoph Hellwig iod->sg = iod->inline_sg; 42657dacad5SJay Sternberg } 42757dacad5SJay Sternberg 428f4800d6dSChristoph Hellwig iod->aborted = 0; 42957dacad5SJay Sternberg iod->npages = -1; 43057dacad5SJay Sternberg iod->nents = 0; 431f4800d6dSChristoph Hellwig iod->length = size; 432f80ec966SKeith Busch 433fc17b653SChristoph Hellwig return BLK_STS_OK; 43457dacad5SJay Sternberg } 43557dacad5SJay Sternberg 436f4800d6dSChristoph Hellwig static void nvme_free_iod(struct nvme_dev *dev, struct request *req) 43757dacad5SJay Sternberg { 438f4800d6dSChristoph Hellwig struct nvme_iod *iod = blk_mq_rq_to_pdu(req); 4395fd4ce1bSChristoph Hellwig const int last_prp = dev->ctrl.page_size / 8 - 1; 44057dacad5SJay Sternberg int i; 441f4800d6dSChristoph Hellwig __le64 **list = iod_list(req); 44257dacad5SJay Sternberg dma_addr_t prp_dma = iod->first_dma; 44357dacad5SJay Sternberg 44457dacad5SJay Sternberg if (iod->npages == 0) 44557dacad5SJay Sternberg dma_pool_free(dev->prp_small_pool, list[0], prp_dma); 44657dacad5SJay Sternberg for (i = 0; i < iod->npages; i++) { 44757dacad5SJay Sternberg __le64 *prp_list = list[i]; 44857dacad5SJay Sternberg dma_addr_t next_prp_dma = le64_to_cpu(prp_list[last_prp]); 44957dacad5SJay Sternberg dma_pool_free(dev->prp_page_pool, prp_list, prp_dma); 45057dacad5SJay Sternberg prp_dma = next_prp_dma; 45157dacad5SJay Sternberg } 45257dacad5SJay Sternberg 453f4800d6dSChristoph Hellwig if (iod->sg != iod->inline_sg) 454f4800d6dSChristoph Hellwig kfree(iod->sg); 45557dacad5SJay Sternberg } 45657dacad5SJay Sternberg 45757dacad5SJay Sternberg #ifdef CONFIG_BLK_DEV_INTEGRITY 45857dacad5SJay Sternberg static void nvme_dif_prep(u32 p, u32 v, struct t10_pi_tuple *pi) 45957dacad5SJay Sternberg { 46057dacad5SJay Sternberg if (be32_to_cpu(pi->ref_tag) == v) 46157dacad5SJay Sternberg pi->ref_tag = cpu_to_be32(p); 46257dacad5SJay Sternberg } 46357dacad5SJay Sternberg 46457dacad5SJay Sternberg static void nvme_dif_complete(u32 p, u32 v, struct t10_pi_tuple *pi) 46557dacad5SJay Sternberg { 46657dacad5SJay Sternberg if (be32_to_cpu(pi->ref_tag) == p) 46757dacad5SJay Sternberg pi->ref_tag = cpu_to_be32(v); 46857dacad5SJay Sternberg } 46957dacad5SJay Sternberg 47057dacad5SJay Sternberg /** 47157dacad5SJay Sternberg * nvme_dif_remap - remaps ref tags to bip seed and physical lba 47257dacad5SJay Sternberg * 47357dacad5SJay Sternberg * The virtual start sector is the one that was originally submitted by the 47457dacad5SJay Sternberg * block layer. Due to partitioning, MD/DM cloning, etc. the actual physical 47557dacad5SJay Sternberg * start sector may be different. Remap protection information to match the 47657dacad5SJay Sternberg * physical LBA on writes, and back to the original seed on reads. 47757dacad5SJay Sternberg * 47857dacad5SJay Sternberg * Type 0 and 3 do not have a ref tag, so no remapping required. 47957dacad5SJay Sternberg */ 48057dacad5SJay Sternberg static void nvme_dif_remap(struct request *req, 48157dacad5SJay Sternberg void (*dif_swap)(u32 p, u32 v, struct t10_pi_tuple *pi)) 48257dacad5SJay Sternberg { 48357dacad5SJay Sternberg struct nvme_ns *ns = req->rq_disk->private_data; 48457dacad5SJay Sternberg struct bio_integrity_payload *bip; 48557dacad5SJay Sternberg struct t10_pi_tuple *pi; 48657dacad5SJay Sternberg void *p, *pmap; 48757dacad5SJay Sternberg u32 i, nlb, ts, phys, virt; 48857dacad5SJay Sternberg 48957dacad5SJay Sternberg if (!ns->pi_type || ns->pi_type == NVME_NS_DPS_PI_TYPE3) 49057dacad5SJay Sternberg return; 49157dacad5SJay Sternberg 49257dacad5SJay Sternberg bip = bio_integrity(req->bio); 49357dacad5SJay Sternberg if (!bip) 49457dacad5SJay Sternberg return; 49557dacad5SJay Sternberg 49657dacad5SJay Sternberg pmap = kmap_atomic(bip->bip_vec->bv_page) + bip->bip_vec->bv_offset; 49757dacad5SJay Sternberg 49857dacad5SJay Sternberg p = pmap; 49957dacad5SJay Sternberg virt = bip_get_seed(bip); 50057dacad5SJay Sternberg phys = nvme_block_nr(ns, blk_rq_pos(req)); 50157dacad5SJay Sternberg nlb = (blk_rq_bytes(req) >> ns->lba_shift); 502ac6fc48cSDan Williams ts = ns->disk->queue->integrity.tuple_size; 50357dacad5SJay Sternberg 50457dacad5SJay Sternberg for (i = 0; i < nlb; i++, virt++, phys++) { 50557dacad5SJay Sternberg pi = (struct t10_pi_tuple *)p; 50657dacad5SJay Sternberg dif_swap(phys, virt, pi); 50757dacad5SJay Sternberg p += ts; 50857dacad5SJay Sternberg } 50957dacad5SJay Sternberg kunmap_atomic(pmap); 51057dacad5SJay Sternberg } 51157dacad5SJay Sternberg #else /* CONFIG_BLK_DEV_INTEGRITY */ 51257dacad5SJay Sternberg static void nvme_dif_remap(struct request *req, 51357dacad5SJay Sternberg void (*dif_swap)(u32 p, u32 v, struct t10_pi_tuple *pi)) 51457dacad5SJay Sternberg { 51557dacad5SJay Sternberg } 51657dacad5SJay Sternberg static void nvme_dif_prep(u32 p, u32 v, struct t10_pi_tuple *pi) 51757dacad5SJay Sternberg { 51857dacad5SJay Sternberg } 51957dacad5SJay Sternberg static void nvme_dif_complete(u32 p, u32 v, struct t10_pi_tuple *pi) 52057dacad5SJay Sternberg { 52157dacad5SJay Sternberg } 52257dacad5SJay Sternberg #endif 52357dacad5SJay Sternberg 524b131c61dSChristoph Hellwig static bool nvme_setup_prps(struct nvme_dev *dev, struct request *req) 52557dacad5SJay Sternberg { 526f4800d6dSChristoph Hellwig struct nvme_iod *iod = blk_mq_rq_to_pdu(req); 52757dacad5SJay Sternberg struct dma_pool *pool; 528b131c61dSChristoph Hellwig int length = blk_rq_payload_bytes(req); 52957dacad5SJay Sternberg struct scatterlist *sg = iod->sg; 53057dacad5SJay Sternberg int dma_len = sg_dma_len(sg); 53157dacad5SJay Sternberg u64 dma_addr = sg_dma_address(sg); 5325fd4ce1bSChristoph Hellwig u32 page_size = dev->ctrl.page_size; 53357dacad5SJay Sternberg int offset = dma_addr & (page_size - 1); 53457dacad5SJay Sternberg __le64 *prp_list; 535f4800d6dSChristoph Hellwig __le64 **list = iod_list(req); 53657dacad5SJay Sternberg dma_addr_t prp_dma; 53757dacad5SJay Sternberg int nprps, i; 53857dacad5SJay Sternberg 53957dacad5SJay Sternberg length -= (page_size - offset); 54057dacad5SJay Sternberg if (length <= 0) 54169d2b571SChristoph Hellwig return true; 54257dacad5SJay Sternberg 54357dacad5SJay Sternberg dma_len -= (page_size - offset); 54457dacad5SJay Sternberg if (dma_len) { 54557dacad5SJay Sternberg dma_addr += (page_size - offset); 54657dacad5SJay Sternberg } else { 54757dacad5SJay Sternberg sg = sg_next(sg); 54857dacad5SJay Sternberg dma_addr = sg_dma_address(sg); 54957dacad5SJay Sternberg dma_len = sg_dma_len(sg); 55057dacad5SJay Sternberg } 55157dacad5SJay Sternberg 55257dacad5SJay Sternberg if (length <= page_size) { 55357dacad5SJay Sternberg iod->first_dma = dma_addr; 55469d2b571SChristoph Hellwig return true; 55557dacad5SJay Sternberg } 55657dacad5SJay Sternberg 55757dacad5SJay Sternberg nprps = DIV_ROUND_UP(length, page_size); 55857dacad5SJay Sternberg if (nprps <= (256 / 8)) { 55957dacad5SJay Sternberg pool = dev->prp_small_pool; 56057dacad5SJay Sternberg iod->npages = 0; 56157dacad5SJay Sternberg } else { 56257dacad5SJay Sternberg pool = dev->prp_page_pool; 56357dacad5SJay Sternberg iod->npages = 1; 56457dacad5SJay Sternberg } 56557dacad5SJay Sternberg 56669d2b571SChristoph Hellwig prp_list = dma_pool_alloc(pool, GFP_ATOMIC, &prp_dma); 56757dacad5SJay Sternberg if (!prp_list) { 56857dacad5SJay Sternberg iod->first_dma = dma_addr; 56957dacad5SJay Sternberg iod->npages = -1; 57069d2b571SChristoph Hellwig return false; 57157dacad5SJay Sternberg } 57257dacad5SJay Sternberg list[0] = prp_list; 57357dacad5SJay Sternberg iod->first_dma = prp_dma; 57457dacad5SJay Sternberg i = 0; 57557dacad5SJay Sternberg for (;;) { 57657dacad5SJay Sternberg if (i == page_size >> 3) { 57757dacad5SJay Sternberg __le64 *old_prp_list = prp_list; 57869d2b571SChristoph Hellwig prp_list = dma_pool_alloc(pool, GFP_ATOMIC, &prp_dma); 57957dacad5SJay Sternberg if (!prp_list) 58069d2b571SChristoph Hellwig return false; 58157dacad5SJay Sternberg list[iod->npages++] = prp_list; 58257dacad5SJay Sternberg prp_list[0] = old_prp_list[i - 1]; 58357dacad5SJay Sternberg old_prp_list[i - 1] = cpu_to_le64(prp_dma); 58457dacad5SJay Sternberg i = 1; 58557dacad5SJay Sternberg } 58657dacad5SJay Sternberg prp_list[i++] = cpu_to_le64(dma_addr); 58757dacad5SJay Sternberg dma_len -= page_size; 58857dacad5SJay Sternberg dma_addr += page_size; 58957dacad5SJay Sternberg length -= page_size; 59057dacad5SJay Sternberg if (length <= 0) 59157dacad5SJay Sternberg break; 59257dacad5SJay Sternberg if (dma_len > 0) 59357dacad5SJay Sternberg continue; 59457dacad5SJay Sternberg BUG_ON(dma_len < 0); 59557dacad5SJay Sternberg sg = sg_next(sg); 59657dacad5SJay Sternberg dma_addr = sg_dma_address(sg); 59757dacad5SJay Sternberg dma_len = sg_dma_len(sg); 59857dacad5SJay Sternberg } 59957dacad5SJay Sternberg 60069d2b571SChristoph Hellwig return true; 60157dacad5SJay Sternberg } 60257dacad5SJay Sternberg 603fc17b653SChristoph Hellwig static blk_status_t nvme_map_data(struct nvme_dev *dev, struct request *req, 604b131c61dSChristoph Hellwig struct nvme_command *cmnd) 60557dacad5SJay Sternberg { 606f4800d6dSChristoph Hellwig struct nvme_iod *iod = blk_mq_rq_to_pdu(req); 607ba1ca37eSChristoph Hellwig struct request_queue *q = req->q; 608ba1ca37eSChristoph Hellwig enum dma_data_direction dma_dir = rq_data_dir(req) ? 609ba1ca37eSChristoph Hellwig DMA_TO_DEVICE : DMA_FROM_DEVICE; 610fc17b653SChristoph Hellwig blk_status_t ret = BLK_STS_IOERR; 61157dacad5SJay Sternberg 612f9d03f96SChristoph Hellwig sg_init_table(iod->sg, blk_rq_nr_phys_segments(req)); 613ba1ca37eSChristoph Hellwig iod->nents = blk_rq_map_sg(q, req, iod->sg); 614ba1ca37eSChristoph Hellwig if (!iod->nents) 615ba1ca37eSChristoph Hellwig goto out; 616ba1ca37eSChristoph Hellwig 617fc17b653SChristoph Hellwig ret = BLK_STS_RESOURCE; 6182b6b535dSMauricio Faria de Oliveira if (!dma_map_sg_attrs(dev->dev, iod->sg, iod->nents, dma_dir, 6192b6b535dSMauricio Faria de Oliveira DMA_ATTR_NO_WARN)) 620ba1ca37eSChristoph Hellwig goto out; 621ba1ca37eSChristoph Hellwig 622b131c61dSChristoph Hellwig if (!nvme_setup_prps(dev, req)) 623ba1ca37eSChristoph Hellwig goto out_unmap; 624ba1ca37eSChristoph Hellwig 625fc17b653SChristoph Hellwig ret = BLK_STS_IOERR; 626ba1ca37eSChristoph Hellwig if (blk_integrity_rq(req)) { 627ba1ca37eSChristoph Hellwig if (blk_rq_count_integrity_sg(q, req->bio) != 1) 628ba1ca37eSChristoph Hellwig goto out_unmap; 629ba1ca37eSChristoph Hellwig 630bf684057SChristoph Hellwig sg_init_table(&iod->meta_sg, 1); 631bf684057SChristoph Hellwig if (blk_rq_map_integrity_sg(q, req->bio, &iod->meta_sg) != 1) 632ba1ca37eSChristoph Hellwig goto out_unmap; 633ba1ca37eSChristoph Hellwig 634ba1ca37eSChristoph Hellwig if (rq_data_dir(req)) 635ba1ca37eSChristoph Hellwig nvme_dif_remap(req, nvme_dif_prep); 636ba1ca37eSChristoph Hellwig 637bf684057SChristoph Hellwig if (!dma_map_sg(dev->dev, &iod->meta_sg, 1, dma_dir)) 638ba1ca37eSChristoph Hellwig goto out_unmap; 63957dacad5SJay Sternberg } 64057dacad5SJay Sternberg 641eb793e2cSChristoph Hellwig cmnd->rw.dptr.prp1 = cpu_to_le64(sg_dma_address(iod->sg)); 642eb793e2cSChristoph Hellwig cmnd->rw.dptr.prp2 = cpu_to_le64(iod->first_dma); 643ba1ca37eSChristoph Hellwig if (blk_integrity_rq(req)) 644bf684057SChristoph Hellwig cmnd->rw.metadata = cpu_to_le64(sg_dma_address(&iod->meta_sg)); 645fc17b653SChristoph Hellwig return BLK_STS_OK; 646ba1ca37eSChristoph Hellwig 647ba1ca37eSChristoph Hellwig out_unmap: 648ba1ca37eSChristoph Hellwig dma_unmap_sg(dev->dev, iod->sg, iod->nents, dma_dir); 649ba1ca37eSChristoph Hellwig out: 650ba1ca37eSChristoph Hellwig return ret; 65157dacad5SJay Sternberg } 65257dacad5SJay Sternberg 653f4800d6dSChristoph Hellwig static void nvme_unmap_data(struct nvme_dev *dev, struct request *req) 654d4f6c3abSChristoph Hellwig { 655f4800d6dSChristoph Hellwig struct nvme_iod *iod = blk_mq_rq_to_pdu(req); 656d4f6c3abSChristoph Hellwig enum dma_data_direction dma_dir = rq_data_dir(req) ? 657d4f6c3abSChristoph Hellwig DMA_TO_DEVICE : DMA_FROM_DEVICE; 658d4f6c3abSChristoph Hellwig 659d4f6c3abSChristoph Hellwig if (iod->nents) { 660d4f6c3abSChristoph Hellwig dma_unmap_sg(dev->dev, iod->sg, iod->nents, dma_dir); 661d4f6c3abSChristoph Hellwig if (blk_integrity_rq(req)) { 662d4f6c3abSChristoph Hellwig if (!rq_data_dir(req)) 663d4f6c3abSChristoph Hellwig nvme_dif_remap(req, nvme_dif_complete); 664bf684057SChristoph Hellwig dma_unmap_sg(dev->dev, &iod->meta_sg, 1, dma_dir); 665d4f6c3abSChristoph Hellwig } 666d4f6c3abSChristoph Hellwig } 667d4f6c3abSChristoph Hellwig 668f9d03f96SChristoph Hellwig nvme_cleanup_cmd(req); 669f4800d6dSChristoph Hellwig nvme_free_iod(dev, req); 67057dacad5SJay Sternberg } 67157dacad5SJay Sternberg 67257dacad5SJay Sternberg /* 67357dacad5SJay Sternberg * NOTE: ns is NULL when called on the admin queue. 67457dacad5SJay Sternberg */ 675fc17b653SChristoph Hellwig static blk_status_t nvme_queue_rq(struct blk_mq_hw_ctx *hctx, 67657dacad5SJay Sternberg const struct blk_mq_queue_data *bd) 67757dacad5SJay Sternberg { 67857dacad5SJay Sternberg struct nvme_ns *ns = hctx->queue->queuedata; 67957dacad5SJay Sternberg struct nvme_queue *nvmeq = hctx->driver_data; 68057dacad5SJay Sternberg struct nvme_dev *dev = nvmeq->dev; 68157dacad5SJay Sternberg struct request *req = bd->rq; 682ba1ca37eSChristoph Hellwig struct nvme_command cmnd; 683ebe6d874SChristoph Hellwig blk_status_t ret; 68457dacad5SJay Sternberg 685f9d03f96SChristoph Hellwig ret = nvme_setup_cmd(ns, req, &cmnd); 686fc17b653SChristoph Hellwig if (ret) 687f4800d6dSChristoph Hellwig return ret; 68857dacad5SJay Sternberg 689b131c61dSChristoph Hellwig ret = nvme_init_iod(req, dev); 690fc17b653SChristoph Hellwig if (ret) 691f9d03f96SChristoph Hellwig goto out_free_cmd; 69257dacad5SJay Sternberg 693fc17b653SChristoph Hellwig if (blk_rq_nr_phys_segments(req)) { 694b131c61dSChristoph Hellwig ret = nvme_map_data(dev, req, &cmnd); 695fc17b653SChristoph Hellwig if (ret) 696f9d03f96SChristoph Hellwig goto out_cleanup_iod; 697fc17b653SChristoph Hellwig } 698ba1ca37eSChristoph Hellwig 699aae239e1SChristoph Hellwig blk_mq_start_request(req); 700ba1ca37eSChristoph Hellwig 701ba1ca37eSChristoph Hellwig spin_lock_irq(&nvmeq->q_lock); 702ae1fba20SKeith Busch if (unlikely(nvmeq->cq_vector < 0)) { 703fc17b653SChristoph Hellwig ret = BLK_STS_IOERR; 704ae1fba20SKeith Busch spin_unlock_irq(&nvmeq->q_lock); 705f9d03f96SChristoph Hellwig goto out_cleanup_iod; 706ae1fba20SKeith Busch } 707ba1ca37eSChristoph Hellwig __nvme_submit_cmd(nvmeq, &cmnd); 70857dacad5SJay Sternberg nvme_process_cq(nvmeq); 70957dacad5SJay Sternberg spin_unlock_irq(&nvmeq->q_lock); 710fc17b653SChristoph Hellwig return BLK_STS_OK; 711f9d03f96SChristoph Hellwig out_cleanup_iod: 712f4800d6dSChristoph Hellwig nvme_free_iod(dev, req); 713f9d03f96SChristoph Hellwig out_free_cmd: 714f9d03f96SChristoph Hellwig nvme_cleanup_cmd(req); 715ba1ca37eSChristoph Hellwig return ret; 71657dacad5SJay Sternberg } 71757dacad5SJay Sternberg 71877f02a7aSChristoph Hellwig static void nvme_pci_complete_rq(struct request *req) 719eee417b0SChristoph Hellwig { 720f4800d6dSChristoph Hellwig struct nvme_iod *iod = blk_mq_rq_to_pdu(req); 721eee417b0SChristoph Hellwig 72277f02a7aSChristoph Hellwig nvme_unmap_data(iod->nvmeq->dev, req); 72377f02a7aSChristoph Hellwig nvme_complete_rq(req); 72457dacad5SJay Sternberg } 72557dacad5SJay Sternberg 726d783e0bdSMarta Rybczynska /* We read the CQE phase first to check if the rest of the entry is valid */ 727d783e0bdSMarta Rybczynska static inline bool nvme_cqe_valid(struct nvme_queue *nvmeq, u16 head, 728d783e0bdSMarta Rybczynska u16 phase) 729d783e0bdSMarta Rybczynska { 730d783e0bdSMarta Rybczynska return (le16_to_cpu(nvmeq->cqes[head].status) & 1) == phase; 731d783e0bdSMarta Rybczynska } 732d783e0bdSMarta Rybczynska 733eb281c82SSagi Grimberg static inline void nvme_ring_cq_doorbell(struct nvme_queue *nvmeq) 734eb281c82SSagi Grimberg { 735eb281c82SSagi Grimberg u16 head = nvmeq->cq_head; 736eb281c82SSagi Grimberg 737eb281c82SSagi Grimberg if (likely(nvmeq->cq_vector >= 0)) { 738eb281c82SSagi Grimberg if (nvme_dbbuf_update_and_check_event(head, nvmeq->dbbuf_cq_db, 739eb281c82SSagi Grimberg nvmeq->dbbuf_cq_ei)) 740eb281c82SSagi Grimberg writel(head, nvmeq->q_db + nvmeq->dev->db_stride); 741eb281c82SSagi Grimberg } 742eb281c82SSagi Grimberg } 743eb281c82SSagi Grimberg 74483a12fb7SSagi Grimberg static inline void nvme_handle_cqe(struct nvme_queue *nvmeq, 74583a12fb7SSagi Grimberg struct nvme_completion *cqe) 74657dacad5SJay Sternberg { 747eee417b0SChristoph Hellwig struct request *req; 748adf68f21SChristoph Hellwig 74983a12fb7SSagi Grimberg if (unlikely(cqe->command_id >= nvmeq->q_depth)) { 7501b3c47c1SSagi Grimberg dev_warn(nvmeq->dev->ctrl.device, 751aae239e1SChristoph Hellwig "invalid id %d completed on queue %d\n", 75283a12fb7SSagi Grimberg cqe->command_id, le16_to_cpu(cqe->sq_id)); 75383a12fb7SSagi Grimberg return; 754aae239e1SChristoph Hellwig } 755aae239e1SChristoph Hellwig 756adf68f21SChristoph Hellwig /* 757adf68f21SChristoph Hellwig * AEN requests are special as they don't time out and can 758adf68f21SChristoph Hellwig * survive any kind of queue freeze and often don't respond to 759adf68f21SChristoph Hellwig * aborts. We don't even bother to allocate a struct request 760adf68f21SChristoph Hellwig * for them but rather special case them here. 761adf68f21SChristoph Hellwig */ 762adf68f21SChristoph Hellwig if (unlikely(nvmeq->qid == 0 && 76383a12fb7SSagi Grimberg cqe->command_id >= NVME_AQ_BLKMQ_DEPTH)) { 7647bf58533SChristoph Hellwig nvme_complete_async_event(&nvmeq->dev->ctrl, 76583a12fb7SSagi Grimberg cqe->status, &cqe->result); 76683a12fb7SSagi Grimberg return; 767adf68f21SChristoph Hellwig } 768adf68f21SChristoph Hellwig 76983a12fb7SSagi Grimberg req = blk_mq_tag_to_rq(*nvmeq->tags, cqe->command_id); 77083a12fb7SSagi Grimberg nvme_end_request(req, cqe->status, cqe->result); 77183a12fb7SSagi Grimberg } 77283a12fb7SSagi Grimberg 773920d13a8SSagi Grimberg static inline bool nvme_read_cqe(struct nvme_queue *nvmeq, 774920d13a8SSagi Grimberg struct nvme_completion *cqe) 77583a12fb7SSagi Grimberg { 776920d13a8SSagi Grimberg if (nvme_cqe_valid(nvmeq, nvmeq->cq_head, nvmeq->cq_phase)) { 777920d13a8SSagi Grimberg *cqe = nvmeq->cqes[nvmeq->cq_head]; 77883a12fb7SSagi Grimberg 779920d13a8SSagi Grimberg if (++nvmeq->cq_head == nvmeq->q_depth) { 780920d13a8SSagi Grimberg nvmeq->cq_head = 0; 781920d13a8SSagi Grimberg nvmeq->cq_phase = !nvmeq->cq_phase; 782920d13a8SSagi Grimberg } 783920d13a8SSagi Grimberg return true; 784920d13a8SSagi Grimberg } 785920d13a8SSagi Grimberg return false; 78683a12fb7SSagi Grimberg } 78783a12fb7SSagi Grimberg 788*442e19b7SSagi Grimberg static void nvme_process_cq(struct nvme_queue *nvmeq) 789920d13a8SSagi Grimberg { 790920d13a8SSagi Grimberg struct nvme_completion cqe; 791920d13a8SSagi Grimberg int consumed = 0; 79283a12fb7SSagi Grimberg 793920d13a8SSagi Grimberg while (nvme_read_cqe(nvmeq, &cqe)) { 79483a12fb7SSagi Grimberg nvme_handle_cqe(nvmeq, &cqe); 795920d13a8SSagi Grimberg consumed++; 79657dacad5SJay Sternberg } 79757dacad5SJay Sternberg 798920d13a8SSagi Grimberg if (consumed) { 799eb281c82SSagi Grimberg nvme_ring_cq_doorbell(nvmeq); 80057dacad5SJay Sternberg nvmeq->cqe_seen = 1; 801a0fa9647SJens Axboe } 802920d13a8SSagi Grimberg } 803a0fa9647SJens Axboe 80457dacad5SJay Sternberg static irqreturn_t nvme_irq(int irq, void *data) 80557dacad5SJay Sternberg { 80657dacad5SJay Sternberg irqreturn_t result; 80757dacad5SJay Sternberg struct nvme_queue *nvmeq = data; 80857dacad5SJay Sternberg spin_lock(&nvmeq->q_lock); 80957dacad5SJay Sternberg nvme_process_cq(nvmeq); 81057dacad5SJay Sternberg result = nvmeq->cqe_seen ? IRQ_HANDLED : IRQ_NONE; 81157dacad5SJay Sternberg nvmeq->cqe_seen = 0; 81257dacad5SJay Sternberg spin_unlock(&nvmeq->q_lock); 81357dacad5SJay Sternberg return result; 81457dacad5SJay Sternberg } 81557dacad5SJay Sternberg 81657dacad5SJay Sternberg static irqreturn_t nvme_irq_check(int irq, void *data) 81757dacad5SJay Sternberg { 81857dacad5SJay Sternberg struct nvme_queue *nvmeq = data; 819d783e0bdSMarta Rybczynska if (nvme_cqe_valid(nvmeq, nvmeq->cq_head, nvmeq->cq_phase)) 82057dacad5SJay Sternberg return IRQ_WAKE_THREAD; 821d783e0bdSMarta Rybczynska return IRQ_NONE; 82257dacad5SJay Sternberg } 82357dacad5SJay Sternberg 8247776db1cSKeith Busch static int __nvme_poll(struct nvme_queue *nvmeq, unsigned int tag) 825a0fa9647SJens Axboe { 826*442e19b7SSagi Grimberg struct nvme_completion cqe; 827*442e19b7SSagi Grimberg int found = 0, consumed = 0; 828a0fa9647SJens Axboe 829*442e19b7SSagi Grimberg if (!nvme_cqe_valid(nvmeq, nvmeq->cq_head, nvmeq->cq_phase)) 830*442e19b7SSagi Grimberg return 0; 831*442e19b7SSagi Grimberg 832*442e19b7SSagi Grimberg spin_lock_irq(&nvmeq->q_lock); 833*442e19b7SSagi Grimberg while (nvme_read_cqe(nvmeq, &cqe)) { 834*442e19b7SSagi Grimberg nvme_handle_cqe(nvmeq, &cqe); 835*442e19b7SSagi Grimberg consumed++; 836*442e19b7SSagi Grimberg 837*442e19b7SSagi Grimberg if (tag == cqe.command_id) { 838*442e19b7SSagi Grimberg found = 1; 839*442e19b7SSagi Grimberg break; 840*442e19b7SSagi Grimberg } 841a0fa9647SJens Axboe } 842a0fa9647SJens Axboe 843*442e19b7SSagi Grimberg if (consumed) 844*442e19b7SSagi Grimberg nvme_ring_cq_doorbell(nvmeq); 845*442e19b7SSagi Grimberg spin_unlock_irq(&nvmeq->q_lock); 846*442e19b7SSagi Grimberg 847*442e19b7SSagi Grimberg return found; 848a0fa9647SJens Axboe } 849a0fa9647SJens Axboe 8507776db1cSKeith Busch static int nvme_poll(struct blk_mq_hw_ctx *hctx, unsigned int tag) 8517776db1cSKeith Busch { 8527776db1cSKeith Busch struct nvme_queue *nvmeq = hctx->driver_data; 8537776db1cSKeith Busch 8547776db1cSKeith Busch return __nvme_poll(nvmeq, tag); 8557776db1cSKeith Busch } 8567776db1cSKeith Busch 857f866fc42SChristoph Hellwig static void nvme_pci_submit_async_event(struct nvme_ctrl *ctrl, int aer_idx) 85857dacad5SJay Sternberg { 859f866fc42SChristoph Hellwig struct nvme_dev *dev = to_nvme_dev(ctrl); 8609396dec9SChristoph Hellwig struct nvme_queue *nvmeq = dev->queues[0]; 86157dacad5SJay Sternberg struct nvme_command c; 86257dacad5SJay Sternberg 86357dacad5SJay Sternberg memset(&c, 0, sizeof(c)); 86457dacad5SJay Sternberg c.common.opcode = nvme_admin_async_event; 865f866fc42SChristoph Hellwig c.common.command_id = NVME_AQ_BLKMQ_DEPTH + aer_idx; 86657dacad5SJay Sternberg 8679396dec9SChristoph Hellwig spin_lock_irq(&nvmeq->q_lock); 8689396dec9SChristoph Hellwig __nvme_submit_cmd(nvmeq, &c); 8699396dec9SChristoph Hellwig spin_unlock_irq(&nvmeq->q_lock); 87057dacad5SJay Sternberg } 87157dacad5SJay Sternberg 87257dacad5SJay Sternberg static int adapter_delete_queue(struct nvme_dev *dev, u8 opcode, u16 id) 87357dacad5SJay Sternberg { 87457dacad5SJay Sternberg struct nvme_command c; 87557dacad5SJay Sternberg 87657dacad5SJay Sternberg memset(&c, 0, sizeof(c)); 87757dacad5SJay Sternberg c.delete_queue.opcode = opcode; 87857dacad5SJay Sternberg c.delete_queue.qid = cpu_to_le16(id); 87957dacad5SJay Sternberg 8801c63dc66SChristoph Hellwig return nvme_submit_sync_cmd(dev->ctrl.admin_q, &c, NULL, 0); 88157dacad5SJay Sternberg } 88257dacad5SJay Sternberg 88357dacad5SJay Sternberg static int adapter_alloc_cq(struct nvme_dev *dev, u16 qid, 88457dacad5SJay Sternberg struct nvme_queue *nvmeq) 88557dacad5SJay Sternberg { 88657dacad5SJay Sternberg struct nvme_command c; 88757dacad5SJay Sternberg int flags = NVME_QUEUE_PHYS_CONTIG | NVME_CQ_IRQ_ENABLED; 88857dacad5SJay Sternberg 88957dacad5SJay Sternberg /* 89057dacad5SJay Sternberg * Note: we (ab)use the fact the the prp fields survive if no data 89157dacad5SJay Sternberg * is attached to the request. 89257dacad5SJay Sternberg */ 89357dacad5SJay Sternberg memset(&c, 0, sizeof(c)); 89457dacad5SJay Sternberg c.create_cq.opcode = nvme_admin_create_cq; 89557dacad5SJay Sternberg c.create_cq.prp1 = cpu_to_le64(nvmeq->cq_dma_addr); 89657dacad5SJay Sternberg c.create_cq.cqid = cpu_to_le16(qid); 89757dacad5SJay Sternberg c.create_cq.qsize = cpu_to_le16(nvmeq->q_depth - 1); 89857dacad5SJay Sternberg c.create_cq.cq_flags = cpu_to_le16(flags); 89957dacad5SJay Sternberg c.create_cq.irq_vector = cpu_to_le16(nvmeq->cq_vector); 90057dacad5SJay Sternberg 9011c63dc66SChristoph Hellwig return nvme_submit_sync_cmd(dev->ctrl.admin_q, &c, NULL, 0); 90257dacad5SJay Sternberg } 90357dacad5SJay Sternberg 90457dacad5SJay Sternberg static int adapter_alloc_sq(struct nvme_dev *dev, u16 qid, 90557dacad5SJay Sternberg struct nvme_queue *nvmeq) 90657dacad5SJay Sternberg { 90757dacad5SJay Sternberg struct nvme_command c; 90881c1cd98SKeith Busch int flags = NVME_QUEUE_PHYS_CONTIG; 90957dacad5SJay Sternberg 91057dacad5SJay Sternberg /* 91157dacad5SJay Sternberg * Note: we (ab)use the fact the the prp fields survive if no data 91257dacad5SJay Sternberg * is attached to the request. 91357dacad5SJay Sternberg */ 91457dacad5SJay Sternberg memset(&c, 0, sizeof(c)); 91557dacad5SJay Sternberg c.create_sq.opcode = nvme_admin_create_sq; 91657dacad5SJay Sternberg c.create_sq.prp1 = cpu_to_le64(nvmeq->sq_dma_addr); 91757dacad5SJay Sternberg c.create_sq.sqid = cpu_to_le16(qid); 91857dacad5SJay Sternberg c.create_sq.qsize = cpu_to_le16(nvmeq->q_depth - 1); 91957dacad5SJay Sternberg c.create_sq.sq_flags = cpu_to_le16(flags); 92057dacad5SJay Sternberg c.create_sq.cqid = cpu_to_le16(qid); 92157dacad5SJay Sternberg 9221c63dc66SChristoph Hellwig return nvme_submit_sync_cmd(dev->ctrl.admin_q, &c, NULL, 0); 92357dacad5SJay Sternberg } 92457dacad5SJay Sternberg 92557dacad5SJay Sternberg static int adapter_delete_cq(struct nvme_dev *dev, u16 cqid) 92657dacad5SJay Sternberg { 92757dacad5SJay Sternberg return adapter_delete_queue(dev, nvme_admin_delete_cq, cqid); 92857dacad5SJay Sternberg } 92957dacad5SJay Sternberg 93057dacad5SJay Sternberg static int adapter_delete_sq(struct nvme_dev *dev, u16 sqid) 93157dacad5SJay Sternberg { 93257dacad5SJay Sternberg return adapter_delete_queue(dev, nvme_admin_delete_sq, sqid); 93357dacad5SJay Sternberg } 93457dacad5SJay Sternberg 9352a842acaSChristoph Hellwig static void abort_endio(struct request *req, blk_status_t error) 93657dacad5SJay Sternberg { 937f4800d6dSChristoph Hellwig struct nvme_iod *iod = blk_mq_rq_to_pdu(req); 938f4800d6dSChristoph Hellwig struct nvme_queue *nvmeq = iod->nvmeq; 93957dacad5SJay Sternberg 94027fa9bc5SChristoph Hellwig dev_warn(nvmeq->dev->ctrl.device, 94127fa9bc5SChristoph Hellwig "Abort status: 0x%x", nvme_req(req)->status); 942e7a2a87dSChristoph Hellwig atomic_inc(&nvmeq->dev->ctrl.abort_limit); 943e7a2a87dSChristoph Hellwig blk_mq_free_request(req); 94457dacad5SJay Sternberg } 94557dacad5SJay Sternberg 946b2a0eb1aSKeith Busch static bool nvme_should_reset(struct nvme_dev *dev, u32 csts) 947b2a0eb1aSKeith Busch { 948b2a0eb1aSKeith Busch 949b2a0eb1aSKeith Busch /* If true, indicates loss of adapter communication, possibly by a 950b2a0eb1aSKeith Busch * NVMe Subsystem reset. 951b2a0eb1aSKeith Busch */ 952b2a0eb1aSKeith Busch bool nssro = dev->subsystem && (csts & NVME_CSTS_NSSRO); 953b2a0eb1aSKeith Busch 954b2a0eb1aSKeith Busch /* If there is a reset ongoing, we shouldn't reset again. */ 955b2a0eb1aSKeith Busch if (dev->ctrl.state == NVME_CTRL_RESETTING) 956b2a0eb1aSKeith Busch return false; 957b2a0eb1aSKeith Busch 958b2a0eb1aSKeith Busch /* We shouldn't reset unless the controller is on fatal error state 959b2a0eb1aSKeith Busch * _or_ if we lost the communication with it. 960b2a0eb1aSKeith Busch */ 961b2a0eb1aSKeith Busch if (!(csts & NVME_CSTS_CFS) && !nssro) 962b2a0eb1aSKeith Busch return false; 963b2a0eb1aSKeith Busch 964b2a0eb1aSKeith Busch /* If PCI error recovery process is happening, we cannot reset or 965b2a0eb1aSKeith Busch * the recovery mechanism will surely fail. 966b2a0eb1aSKeith Busch */ 967b2a0eb1aSKeith Busch if (pci_channel_offline(to_pci_dev(dev->dev))) 968b2a0eb1aSKeith Busch return false; 969b2a0eb1aSKeith Busch 970b2a0eb1aSKeith Busch return true; 971b2a0eb1aSKeith Busch } 972b2a0eb1aSKeith Busch 973b2a0eb1aSKeith Busch static void nvme_warn_reset(struct nvme_dev *dev, u32 csts) 974b2a0eb1aSKeith Busch { 975b2a0eb1aSKeith Busch /* Read a config register to help see what died. */ 976b2a0eb1aSKeith Busch u16 pci_status; 977b2a0eb1aSKeith Busch int result; 978b2a0eb1aSKeith Busch 979b2a0eb1aSKeith Busch result = pci_read_config_word(to_pci_dev(dev->dev), PCI_STATUS, 980b2a0eb1aSKeith Busch &pci_status); 981b2a0eb1aSKeith Busch if (result == PCIBIOS_SUCCESSFUL) 982b2a0eb1aSKeith Busch dev_warn(dev->ctrl.device, 983b2a0eb1aSKeith Busch "controller is down; will reset: CSTS=0x%x, PCI_STATUS=0x%hx\n", 984b2a0eb1aSKeith Busch csts, pci_status); 985b2a0eb1aSKeith Busch else 986b2a0eb1aSKeith Busch dev_warn(dev->ctrl.device, 987b2a0eb1aSKeith Busch "controller is down; will reset: CSTS=0x%x, PCI_STATUS read failed (%d)\n", 988b2a0eb1aSKeith Busch csts, result); 989b2a0eb1aSKeith Busch } 990b2a0eb1aSKeith Busch 99131c7c7d2SChristoph Hellwig static enum blk_eh_timer_return nvme_timeout(struct request *req, bool reserved) 99257dacad5SJay Sternberg { 993f4800d6dSChristoph Hellwig struct nvme_iod *iod = blk_mq_rq_to_pdu(req); 994f4800d6dSChristoph Hellwig struct nvme_queue *nvmeq = iod->nvmeq; 99557dacad5SJay Sternberg struct nvme_dev *dev = nvmeq->dev; 99657dacad5SJay Sternberg struct request *abort_req; 99757dacad5SJay Sternberg struct nvme_command cmd; 998b2a0eb1aSKeith Busch u32 csts = readl(dev->bar + NVME_REG_CSTS); 999b2a0eb1aSKeith Busch 1000b2a0eb1aSKeith Busch /* 1001b2a0eb1aSKeith Busch * Reset immediately if the controller is failed 1002b2a0eb1aSKeith Busch */ 1003b2a0eb1aSKeith Busch if (nvme_should_reset(dev, csts)) { 1004b2a0eb1aSKeith Busch nvme_warn_reset(dev, csts); 1005b2a0eb1aSKeith Busch nvme_dev_disable(dev, false); 1006d86c4d8eSChristoph Hellwig nvme_reset_ctrl(&dev->ctrl); 1007b2a0eb1aSKeith Busch return BLK_EH_HANDLED; 1008b2a0eb1aSKeith Busch } 100957dacad5SJay Sternberg 101031c7c7d2SChristoph Hellwig /* 10117776db1cSKeith Busch * Did we miss an interrupt? 10127776db1cSKeith Busch */ 10137776db1cSKeith Busch if (__nvme_poll(nvmeq, req->tag)) { 10147776db1cSKeith Busch dev_warn(dev->ctrl.device, 10157776db1cSKeith Busch "I/O %d QID %d timeout, completion polled\n", 10167776db1cSKeith Busch req->tag, nvmeq->qid); 10177776db1cSKeith Busch return BLK_EH_HANDLED; 10187776db1cSKeith Busch } 10197776db1cSKeith Busch 10207776db1cSKeith Busch /* 1021fd634f41SChristoph Hellwig * Shutdown immediately if controller times out while starting. The 1022fd634f41SChristoph Hellwig * reset work will see the pci device disabled when it gets the forced 1023fd634f41SChristoph Hellwig * cancellation error. All outstanding requests are completed on 1024fd634f41SChristoph Hellwig * shutdown, so we return BLK_EH_HANDLED. 1025fd634f41SChristoph Hellwig */ 1026bb8d261eSChristoph Hellwig if (dev->ctrl.state == NVME_CTRL_RESETTING) { 10271b3c47c1SSagi Grimberg dev_warn(dev->ctrl.device, 1028fd634f41SChristoph Hellwig "I/O %d QID %d timeout, disable controller\n", 1029fd634f41SChristoph Hellwig req->tag, nvmeq->qid); 1030a5cdb68cSKeith Busch nvme_dev_disable(dev, false); 103127fa9bc5SChristoph Hellwig nvme_req(req)->flags |= NVME_REQ_CANCELLED; 1032fd634f41SChristoph Hellwig return BLK_EH_HANDLED; 1033fd634f41SChristoph Hellwig } 1034fd634f41SChristoph Hellwig 1035fd634f41SChristoph Hellwig /* 1036e1569a16SKeith Busch * Shutdown the controller immediately and schedule a reset if the 1037e1569a16SKeith Busch * command was already aborted once before and still hasn't been 1038e1569a16SKeith Busch * returned to the driver, or if this is the admin queue. 103931c7c7d2SChristoph Hellwig */ 1040f4800d6dSChristoph Hellwig if (!nvmeq->qid || iod->aborted) { 10411b3c47c1SSagi Grimberg dev_warn(dev->ctrl.device, 104257dacad5SJay Sternberg "I/O %d QID %d timeout, reset controller\n", 104357dacad5SJay Sternberg req->tag, nvmeq->qid); 1044a5cdb68cSKeith Busch nvme_dev_disable(dev, false); 1045d86c4d8eSChristoph Hellwig nvme_reset_ctrl(&dev->ctrl); 1046e1569a16SKeith Busch 1047e1569a16SKeith Busch /* 1048e1569a16SKeith Busch * Mark the request as handled, since the inline shutdown 1049e1569a16SKeith Busch * forces all outstanding requests to complete. 1050e1569a16SKeith Busch */ 105127fa9bc5SChristoph Hellwig nvme_req(req)->flags |= NVME_REQ_CANCELLED; 1052e1569a16SKeith Busch return BLK_EH_HANDLED; 105357dacad5SJay Sternberg } 105457dacad5SJay Sternberg 1055e7a2a87dSChristoph Hellwig if (atomic_dec_return(&dev->ctrl.abort_limit) < 0) { 1056e7a2a87dSChristoph Hellwig atomic_inc(&dev->ctrl.abort_limit); 1057e7a2a87dSChristoph Hellwig return BLK_EH_RESET_TIMER; 1058e7a2a87dSChristoph Hellwig } 10597bf7d778SKeith Busch iod->aborted = 1; 106057dacad5SJay Sternberg 106157dacad5SJay Sternberg memset(&cmd, 0, sizeof(cmd)); 106257dacad5SJay Sternberg cmd.abort.opcode = nvme_admin_abort_cmd; 106357dacad5SJay Sternberg cmd.abort.cid = req->tag; 106457dacad5SJay Sternberg cmd.abort.sqid = cpu_to_le16(nvmeq->qid); 106557dacad5SJay Sternberg 10661b3c47c1SSagi Grimberg dev_warn(nvmeq->dev->ctrl.device, 10671b3c47c1SSagi Grimberg "I/O %d QID %d timeout, aborting\n", 106857dacad5SJay Sternberg req->tag, nvmeq->qid); 1069e7a2a87dSChristoph Hellwig 1070e7a2a87dSChristoph Hellwig abort_req = nvme_alloc_request(dev->ctrl.admin_q, &cmd, 1071eb71f435SChristoph Hellwig BLK_MQ_REQ_NOWAIT, NVME_QID_ANY); 10726bf25d16SChristoph Hellwig if (IS_ERR(abort_req)) { 10736bf25d16SChristoph Hellwig atomic_inc(&dev->ctrl.abort_limit); 107431c7c7d2SChristoph Hellwig return BLK_EH_RESET_TIMER; 107557dacad5SJay Sternberg } 107657dacad5SJay Sternberg 1077e7a2a87dSChristoph Hellwig abort_req->timeout = ADMIN_TIMEOUT; 1078e7a2a87dSChristoph Hellwig abort_req->end_io_data = NULL; 1079e7a2a87dSChristoph Hellwig blk_execute_rq_nowait(abort_req->q, NULL, abort_req, 0, abort_endio); 108057dacad5SJay Sternberg 108157dacad5SJay Sternberg /* 108257dacad5SJay Sternberg * The aborted req will be completed on receiving the abort req. 108357dacad5SJay Sternberg * We enable the timer again. If hit twice, it'll cause a device reset, 108457dacad5SJay Sternberg * as the device then is in a faulty state. 108557dacad5SJay Sternberg */ 108657dacad5SJay Sternberg return BLK_EH_RESET_TIMER; 108757dacad5SJay Sternberg } 108857dacad5SJay Sternberg 108957dacad5SJay Sternberg static void nvme_free_queue(struct nvme_queue *nvmeq) 109057dacad5SJay Sternberg { 109157dacad5SJay Sternberg dma_free_coherent(nvmeq->q_dmadev, CQ_SIZE(nvmeq->q_depth), 109257dacad5SJay Sternberg (void *)nvmeq->cqes, nvmeq->cq_dma_addr); 109357dacad5SJay Sternberg if (nvmeq->sq_cmds) 109457dacad5SJay Sternberg dma_free_coherent(nvmeq->q_dmadev, SQ_SIZE(nvmeq->q_depth), 109557dacad5SJay Sternberg nvmeq->sq_cmds, nvmeq->sq_dma_addr); 109657dacad5SJay Sternberg kfree(nvmeq); 109757dacad5SJay Sternberg } 109857dacad5SJay Sternberg 109957dacad5SJay Sternberg static void nvme_free_queues(struct nvme_dev *dev, int lowest) 110057dacad5SJay Sternberg { 110157dacad5SJay Sternberg int i; 110257dacad5SJay Sternberg 110357dacad5SJay Sternberg for (i = dev->queue_count - 1; i >= lowest; i--) { 110457dacad5SJay Sternberg struct nvme_queue *nvmeq = dev->queues[i]; 110557dacad5SJay Sternberg dev->queue_count--; 110657dacad5SJay Sternberg dev->queues[i] = NULL; 110757dacad5SJay Sternberg nvme_free_queue(nvmeq); 110857dacad5SJay Sternberg } 110957dacad5SJay Sternberg } 111057dacad5SJay Sternberg 111157dacad5SJay Sternberg /** 111257dacad5SJay Sternberg * nvme_suspend_queue - put queue into suspended state 111357dacad5SJay Sternberg * @nvmeq - queue to suspend 111457dacad5SJay Sternberg */ 111557dacad5SJay Sternberg static int nvme_suspend_queue(struct nvme_queue *nvmeq) 111657dacad5SJay Sternberg { 111757dacad5SJay Sternberg int vector; 111857dacad5SJay Sternberg 111957dacad5SJay Sternberg spin_lock_irq(&nvmeq->q_lock); 112057dacad5SJay Sternberg if (nvmeq->cq_vector == -1) { 112157dacad5SJay Sternberg spin_unlock_irq(&nvmeq->q_lock); 112257dacad5SJay Sternberg return 1; 112357dacad5SJay Sternberg } 11240ff199cbSChristoph Hellwig vector = nvmeq->cq_vector; 112557dacad5SJay Sternberg nvmeq->dev->online_queues--; 112657dacad5SJay Sternberg nvmeq->cq_vector = -1; 112757dacad5SJay Sternberg spin_unlock_irq(&nvmeq->q_lock); 112857dacad5SJay Sternberg 11291c63dc66SChristoph Hellwig if (!nvmeq->qid && nvmeq->dev->ctrl.admin_q) 113025646264SKeith Busch blk_mq_stop_hw_queues(nvmeq->dev->ctrl.admin_q); 113157dacad5SJay Sternberg 11320ff199cbSChristoph Hellwig pci_free_irq(to_pci_dev(nvmeq->dev->dev), vector, nvmeq); 113357dacad5SJay Sternberg 113457dacad5SJay Sternberg return 0; 113557dacad5SJay Sternberg } 113657dacad5SJay Sternberg 1137a5cdb68cSKeith Busch static void nvme_disable_admin_queue(struct nvme_dev *dev, bool shutdown) 113857dacad5SJay Sternberg { 1139a5cdb68cSKeith Busch struct nvme_queue *nvmeq = dev->queues[0]; 114057dacad5SJay Sternberg 114157dacad5SJay Sternberg if (!nvmeq) 114257dacad5SJay Sternberg return; 114357dacad5SJay Sternberg if (nvme_suspend_queue(nvmeq)) 114457dacad5SJay Sternberg return; 114557dacad5SJay Sternberg 1146a5cdb68cSKeith Busch if (shutdown) 1147a5cdb68cSKeith Busch nvme_shutdown_ctrl(&dev->ctrl); 1148a5cdb68cSKeith Busch else 1149a5cdb68cSKeith Busch nvme_disable_ctrl(&dev->ctrl, lo_hi_readq( 1150a5cdb68cSKeith Busch dev->bar + NVME_REG_CAP)); 115157dacad5SJay Sternberg 115257dacad5SJay Sternberg spin_lock_irq(&nvmeq->q_lock); 115357dacad5SJay Sternberg nvme_process_cq(nvmeq); 115457dacad5SJay Sternberg spin_unlock_irq(&nvmeq->q_lock); 115557dacad5SJay Sternberg } 115657dacad5SJay Sternberg 115757dacad5SJay Sternberg static int nvme_cmb_qdepth(struct nvme_dev *dev, int nr_io_queues, 115857dacad5SJay Sternberg int entry_size) 115957dacad5SJay Sternberg { 116057dacad5SJay Sternberg int q_depth = dev->q_depth; 11615fd4ce1bSChristoph Hellwig unsigned q_size_aligned = roundup(q_depth * entry_size, 11625fd4ce1bSChristoph Hellwig dev->ctrl.page_size); 116357dacad5SJay Sternberg 116457dacad5SJay Sternberg if (q_size_aligned * nr_io_queues > dev->cmb_size) { 116557dacad5SJay Sternberg u64 mem_per_q = div_u64(dev->cmb_size, nr_io_queues); 11665fd4ce1bSChristoph Hellwig mem_per_q = round_down(mem_per_q, dev->ctrl.page_size); 116757dacad5SJay Sternberg q_depth = div_u64(mem_per_q, entry_size); 116857dacad5SJay Sternberg 116957dacad5SJay Sternberg /* 117057dacad5SJay Sternberg * Ensure the reduced q_depth is above some threshold where it 117157dacad5SJay Sternberg * would be better to map queues in system memory with the 117257dacad5SJay Sternberg * original depth 117357dacad5SJay Sternberg */ 117457dacad5SJay Sternberg if (q_depth < 64) 117557dacad5SJay Sternberg return -ENOMEM; 117657dacad5SJay Sternberg } 117757dacad5SJay Sternberg 117857dacad5SJay Sternberg return q_depth; 117957dacad5SJay Sternberg } 118057dacad5SJay Sternberg 118157dacad5SJay Sternberg static int nvme_alloc_sq_cmds(struct nvme_dev *dev, struct nvme_queue *nvmeq, 118257dacad5SJay Sternberg int qid, int depth) 118357dacad5SJay Sternberg { 118457dacad5SJay Sternberg if (qid && dev->cmb && use_cmb_sqes && NVME_CMB_SQS(dev->cmbsz)) { 11855fd4ce1bSChristoph Hellwig unsigned offset = (qid - 1) * roundup(SQ_SIZE(depth), 11865fd4ce1bSChristoph Hellwig dev->ctrl.page_size); 118757dacad5SJay Sternberg nvmeq->sq_dma_addr = dev->cmb_dma_addr + offset; 118857dacad5SJay Sternberg nvmeq->sq_cmds_io = dev->cmb + offset; 118957dacad5SJay Sternberg } else { 119057dacad5SJay Sternberg nvmeq->sq_cmds = dma_alloc_coherent(dev->dev, SQ_SIZE(depth), 119157dacad5SJay Sternberg &nvmeq->sq_dma_addr, GFP_KERNEL); 119257dacad5SJay Sternberg if (!nvmeq->sq_cmds) 119357dacad5SJay Sternberg return -ENOMEM; 119457dacad5SJay Sternberg } 119557dacad5SJay Sternberg 119657dacad5SJay Sternberg return 0; 119757dacad5SJay Sternberg } 119857dacad5SJay Sternberg 119957dacad5SJay Sternberg static struct nvme_queue *nvme_alloc_queue(struct nvme_dev *dev, int qid, 1200d3af3ecdSShaohua Li int depth, int node) 120157dacad5SJay Sternberg { 1202d3af3ecdSShaohua Li struct nvme_queue *nvmeq = kzalloc_node(sizeof(*nvmeq), GFP_KERNEL, 1203d3af3ecdSShaohua Li node); 120457dacad5SJay Sternberg if (!nvmeq) 120557dacad5SJay Sternberg return NULL; 120657dacad5SJay Sternberg 120757dacad5SJay Sternberg nvmeq->cqes = dma_zalloc_coherent(dev->dev, CQ_SIZE(depth), 120857dacad5SJay Sternberg &nvmeq->cq_dma_addr, GFP_KERNEL); 120957dacad5SJay Sternberg if (!nvmeq->cqes) 121057dacad5SJay Sternberg goto free_nvmeq; 121157dacad5SJay Sternberg 121257dacad5SJay Sternberg if (nvme_alloc_sq_cmds(dev, nvmeq, qid, depth)) 121357dacad5SJay Sternberg goto free_cqdma; 121457dacad5SJay Sternberg 121557dacad5SJay Sternberg nvmeq->q_dmadev = dev->dev; 121657dacad5SJay Sternberg nvmeq->dev = dev; 121757dacad5SJay Sternberg spin_lock_init(&nvmeq->q_lock); 121857dacad5SJay Sternberg nvmeq->cq_head = 0; 121957dacad5SJay Sternberg nvmeq->cq_phase = 1; 122057dacad5SJay Sternberg nvmeq->q_db = &dev->dbs[qid * 2 * dev->db_stride]; 122157dacad5SJay Sternberg nvmeq->q_depth = depth; 122257dacad5SJay Sternberg nvmeq->qid = qid; 122357dacad5SJay Sternberg nvmeq->cq_vector = -1; 122457dacad5SJay Sternberg dev->queues[qid] = nvmeq; 122557dacad5SJay Sternberg dev->queue_count++; 122657dacad5SJay Sternberg 122757dacad5SJay Sternberg return nvmeq; 122857dacad5SJay Sternberg 122957dacad5SJay Sternberg free_cqdma: 123057dacad5SJay Sternberg dma_free_coherent(dev->dev, CQ_SIZE(depth), (void *)nvmeq->cqes, 123157dacad5SJay Sternberg nvmeq->cq_dma_addr); 123257dacad5SJay Sternberg free_nvmeq: 123357dacad5SJay Sternberg kfree(nvmeq); 123457dacad5SJay Sternberg return NULL; 123557dacad5SJay Sternberg } 123657dacad5SJay Sternberg 1237dca51e78SChristoph Hellwig static int queue_request_irq(struct nvme_queue *nvmeq) 123857dacad5SJay Sternberg { 12390ff199cbSChristoph Hellwig struct pci_dev *pdev = to_pci_dev(nvmeq->dev->dev); 12400ff199cbSChristoph Hellwig int nr = nvmeq->dev->ctrl.instance; 12410ff199cbSChristoph Hellwig 12420ff199cbSChristoph Hellwig if (use_threaded_interrupts) { 12430ff199cbSChristoph Hellwig return pci_request_irq(pdev, nvmeq->cq_vector, nvme_irq_check, 12440ff199cbSChristoph Hellwig nvme_irq, nvmeq, "nvme%dq%d", nr, nvmeq->qid); 12450ff199cbSChristoph Hellwig } else { 12460ff199cbSChristoph Hellwig return pci_request_irq(pdev, nvmeq->cq_vector, nvme_irq, 12470ff199cbSChristoph Hellwig NULL, nvmeq, "nvme%dq%d", nr, nvmeq->qid); 12480ff199cbSChristoph Hellwig } 124957dacad5SJay Sternberg } 125057dacad5SJay Sternberg 125157dacad5SJay Sternberg static void nvme_init_queue(struct nvme_queue *nvmeq, u16 qid) 125257dacad5SJay Sternberg { 125357dacad5SJay Sternberg struct nvme_dev *dev = nvmeq->dev; 125457dacad5SJay Sternberg 125557dacad5SJay Sternberg spin_lock_irq(&nvmeq->q_lock); 125657dacad5SJay Sternberg nvmeq->sq_tail = 0; 125757dacad5SJay Sternberg nvmeq->cq_head = 0; 125857dacad5SJay Sternberg nvmeq->cq_phase = 1; 125957dacad5SJay Sternberg nvmeq->q_db = &dev->dbs[qid * 2 * dev->db_stride]; 126057dacad5SJay Sternberg memset((void *)nvmeq->cqes, 0, CQ_SIZE(nvmeq->q_depth)); 1261f9f38e33SHelen Koike nvme_dbbuf_init(dev, nvmeq, qid); 126257dacad5SJay Sternberg dev->online_queues++; 126357dacad5SJay Sternberg spin_unlock_irq(&nvmeq->q_lock); 126457dacad5SJay Sternberg } 126557dacad5SJay Sternberg 126657dacad5SJay Sternberg static int nvme_create_queue(struct nvme_queue *nvmeq, int qid) 126757dacad5SJay Sternberg { 126857dacad5SJay Sternberg struct nvme_dev *dev = nvmeq->dev; 126957dacad5SJay Sternberg int result; 127057dacad5SJay Sternberg 127157dacad5SJay Sternberg nvmeq->cq_vector = qid - 1; 127257dacad5SJay Sternberg result = adapter_alloc_cq(dev, qid, nvmeq); 127357dacad5SJay Sternberg if (result < 0) 127457dacad5SJay Sternberg return result; 127557dacad5SJay Sternberg 127657dacad5SJay Sternberg result = adapter_alloc_sq(dev, qid, nvmeq); 127757dacad5SJay Sternberg if (result < 0) 127857dacad5SJay Sternberg goto release_cq; 127957dacad5SJay Sternberg 1280dca51e78SChristoph Hellwig result = queue_request_irq(nvmeq); 128157dacad5SJay Sternberg if (result < 0) 128257dacad5SJay Sternberg goto release_sq; 128357dacad5SJay Sternberg 128457dacad5SJay Sternberg nvme_init_queue(nvmeq, qid); 128557dacad5SJay Sternberg return result; 128657dacad5SJay Sternberg 128757dacad5SJay Sternberg release_sq: 128857dacad5SJay Sternberg adapter_delete_sq(dev, qid); 128957dacad5SJay Sternberg release_cq: 129057dacad5SJay Sternberg adapter_delete_cq(dev, qid); 129157dacad5SJay Sternberg return result; 129257dacad5SJay Sternberg } 129357dacad5SJay Sternberg 1294f363b089SEric Biggers static const struct blk_mq_ops nvme_mq_admin_ops = { 129557dacad5SJay Sternberg .queue_rq = nvme_queue_rq, 129677f02a7aSChristoph Hellwig .complete = nvme_pci_complete_rq, 129757dacad5SJay Sternberg .init_hctx = nvme_admin_init_hctx, 129857dacad5SJay Sternberg .exit_hctx = nvme_admin_exit_hctx, 12990350815aSChristoph Hellwig .init_request = nvme_init_request, 130057dacad5SJay Sternberg .timeout = nvme_timeout, 130157dacad5SJay Sternberg }; 130257dacad5SJay Sternberg 1303f363b089SEric Biggers static const struct blk_mq_ops nvme_mq_ops = { 130457dacad5SJay Sternberg .queue_rq = nvme_queue_rq, 130577f02a7aSChristoph Hellwig .complete = nvme_pci_complete_rq, 130657dacad5SJay Sternberg .init_hctx = nvme_init_hctx, 130757dacad5SJay Sternberg .init_request = nvme_init_request, 1308dca51e78SChristoph Hellwig .map_queues = nvme_pci_map_queues, 130957dacad5SJay Sternberg .timeout = nvme_timeout, 1310a0fa9647SJens Axboe .poll = nvme_poll, 131157dacad5SJay Sternberg }; 131257dacad5SJay Sternberg 131357dacad5SJay Sternberg static void nvme_dev_remove_admin(struct nvme_dev *dev) 131457dacad5SJay Sternberg { 13151c63dc66SChristoph Hellwig if (dev->ctrl.admin_q && !blk_queue_dying(dev->ctrl.admin_q)) { 131669d9a99cSKeith Busch /* 131769d9a99cSKeith Busch * If the controller was reset during removal, it's possible 131869d9a99cSKeith Busch * user requests may be waiting on a stopped queue. Start the 131969d9a99cSKeith Busch * queue to flush these to completion. 132069d9a99cSKeith Busch */ 132169d9a99cSKeith Busch blk_mq_start_stopped_hw_queues(dev->ctrl.admin_q, true); 13221c63dc66SChristoph Hellwig blk_cleanup_queue(dev->ctrl.admin_q); 132357dacad5SJay Sternberg blk_mq_free_tag_set(&dev->admin_tagset); 132457dacad5SJay Sternberg } 132557dacad5SJay Sternberg } 132657dacad5SJay Sternberg 132757dacad5SJay Sternberg static int nvme_alloc_admin_tags(struct nvme_dev *dev) 132857dacad5SJay Sternberg { 13291c63dc66SChristoph Hellwig if (!dev->ctrl.admin_q) { 133057dacad5SJay Sternberg dev->admin_tagset.ops = &nvme_mq_admin_ops; 133157dacad5SJay Sternberg dev->admin_tagset.nr_hw_queues = 1; 1332e3e9d50cSKeith Busch 1333e3e9d50cSKeith Busch /* 1334e3e9d50cSKeith Busch * Subtract one to leave an empty queue entry for 'Full Queue' 1335e3e9d50cSKeith Busch * condition. See NVM-Express 1.2 specification, section 4.1.2. 1336e3e9d50cSKeith Busch */ 1337e3e9d50cSKeith Busch dev->admin_tagset.queue_depth = NVME_AQ_BLKMQ_DEPTH - 1; 133857dacad5SJay Sternberg dev->admin_tagset.timeout = ADMIN_TIMEOUT; 133957dacad5SJay Sternberg dev->admin_tagset.numa_node = dev_to_node(dev->dev); 134057dacad5SJay Sternberg dev->admin_tagset.cmd_size = nvme_cmd_size(dev); 1341d3484991SJens Axboe dev->admin_tagset.flags = BLK_MQ_F_NO_SCHED; 134257dacad5SJay Sternberg dev->admin_tagset.driver_data = dev; 134357dacad5SJay Sternberg 134457dacad5SJay Sternberg if (blk_mq_alloc_tag_set(&dev->admin_tagset)) 134557dacad5SJay Sternberg return -ENOMEM; 134657dacad5SJay Sternberg 13471c63dc66SChristoph Hellwig dev->ctrl.admin_q = blk_mq_init_queue(&dev->admin_tagset); 13481c63dc66SChristoph Hellwig if (IS_ERR(dev->ctrl.admin_q)) { 134957dacad5SJay Sternberg blk_mq_free_tag_set(&dev->admin_tagset); 135057dacad5SJay Sternberg return -ENOMEM; 135157dacad5SJay Sternberg } 13521c63dc66SChristoph Hellwig if (!blk_get_queue(dev->ctrl.admin_q)) { 135357dacad5SJay Sternberg nvme_dev_remove_admin(dev); 13541c63dc66SChristoph Hellwig dev->ctrl.admin_q = NULL; 135557dacad5SJay Sternberg return -ENODEV; 135657dacad5SJay Sternberg } 135757dacad5SJay Sternberg } else 135825646264SKeith Busch blk_mq_start_stopped_hw_queues(dev->ctrl.admin_q, true); 135957dacad5SJay Sternberg 136057dacad5SJay Sternberg return 0; 136157dacad5SJay Sternberg } 136257dacad5SJay Sternberg 136397f6ef64SXu Yu static unsigned long db_bar_size(struct nvme_dev *dev, unsigned nr_io_queues) 136497f6ef64SXu Yu { 136597f6ef64SXu Yu return NVME_REG_DBS + ((nr_io_queues + 1) * 8 * dev->db_stride); 136697f6ef64SXu Yu } 136797f6ef64SXu Yu 136897f6ef64SXu Yu static int nvme_remap_bar(struct nvme_dev *dev, unsigned long size) 136997f6ef64SXu Yu { 137097f6ef64SXu Yu struct pci_dev *pdev = to_pci_dev(dev->dev); 137197f6ef64SXu Yu 137297f6ef64SXu Yu if (size <= dev->bar_mapped_size) 137397f6ef64SXu Yu return 0; 137497f6ef64SXu Yu if (size > pci_resource_len(pdev, 0)) 137597f6ef64SXu Yu return -ENOMEM; 137697f6ef64SXu Yu if (dev->bar) 137797f6ef64SXu Yu iounmap(dev->bar); 137897f6ef64SXu Yu dev->bar = ioremap(pci_resource_start(pdev, 0), size); 137997f6ef64SXu Yu if (!dev->bar) { 138097f6ef64SXu Yu dev->bar_mapped_size = 0; 138197f6ef64SXu Yu return -ENOMEM; 138297f6ef64SXu Yu } 138397f6ef64SXu Yu dev->bar_mapped_size = size; 138497f6ef64SXu Yu dev->dbs = dev->bar + NVME_REG_DBS; 138597f6ef64SXu Yu 138697f6ef64SXu Yu return 0; 138797f6ef64SXu Yu } 138897f6ef64SXu Yu 138957dacad5SJay Sternberg static int nvme_configure_admin_queue(struct nvme_dev *dev) 139057dacad5SJay Sternberg { 139157dacad5SJay Sternberg int result; 139257dacad5SJay Sternberg u32 aqa; 13937a67cbeaSChristoph Hellwig u64 cap = lo_hi_readq(dev->bar + NVME_REG_CAP); 139457dacad5SJay Sternberg struct nvme_queue *nvmeq; 139557dacad5SJay Sternberg 139697f6ef64SXu Yu result = nvme_remap_bar(dev, db_bar_size(dev, 0)); 139797f6ef64SXu Yu if (result < 0) 139897f6ef64SXu Yu return result; 139997f6ef64SXu Yu 14008ef2074dSGabriel Krisman Bertazi dev->subsystem = readl(dev->bar + NVME_REG_VS) >= NVME_VS(1, 1, 0) ? 140157dacad5SJay Sternberg NVME_CAP_NSSRC(cap) : 0; 140257dacad5SJay Sternberg 14037a67cbeaSChristoph Hellwig if (dev->subsystem && 14047a67cbeaSChristoph Hellwig (readl(dev->bar + NVME_REG_CSTS) & NVME_CSTS_NSSRO)) 14057a67cbeaSChristoph Hellwig writel(NVME_CSTS_NSSRO, dev->bar + NVME_REG_CSTS); 140657dacad5SJay Sternberg 14075fd4ce1bSChristoph Hellwig result = nvme_disable_ctrl(&dev->ctrl, cap); 140857dacad5SJay Sternberg if (result < 0) 140957dacad5SJay Sternberg return result; 141057dacad5SJay Sternberg 141157dacad5SJay Sternberg nvmeq = dev->queues[0]; 141257dacad5SJay Sternberg if (!nvmeq) { 1413d3af3ecdSShaohua Li nvmeq = nvme_alloc_queue(dev, 0, NVME_AQ_DEPTH, 1414d3af3ecdSShaohua Li dev_to_node(dev->dev)); 141557dacad5SJay Sternberg if (!nvmeq) 141657dacad5SJay Sternberg return -ENOMEM; 141757dacad5SJay Sternberg } 141857dacad5SJay Sternberg 141957dacad5SJay Sternberg aqa = nvmeq->q_depth - 1; 142057dacad5SJay Sternberg aqa |= aqa << 16; 142157dacad5SJay Sternberg 14227a67cbeaSChristoph Hellwig writel(aqa, dev->bar + NVME_REG_AQA); 14237a67cbeaSChristoph Hellwig lo_hi_writeq(nvmeq->sq_dma_addr, dev->bar + NVME_REG_ASQ); 14247a67cbeaSChristoph Hellwig lo_hi_writeq(nvmeq->cq_dma_addr, dev->bar + NVME_REG_ACQ); 142557dacad5SJay Sternberg 14265fd4ce1bSChristoph Hellwig result = nvme_enable_ctrl(&dev->ctrl, cap); 142757dacad5SJay Sternberg if (result) 1428d4875622SKeith Busch return result; 142957dacad5SJay Sternberg 143057dacad5SJay Sternberg nvmeq->cq_vector = 0; 1431dca51e78SChristoph Hellwig result = queue_request_irq(nvmeq); 143257dacad5SJay Sternberg if (result) { 143357dacad5SJay Sternberg nvmeq->cq_vector = -1; 1434d4875622SKeith Busch return result; 143557dacad5SJay Sternberg } 143657dacad5SJay Sternberg 143757dacad5SJay Sternberg return result; 143857dacad5SJay Sternberg } 143957dacad5SJay Sternberg 1440749941f2SChristoph Hellwig static int nvme_create_io_queues(struct nvme_dev *dev) 144157dacad5SJay Sternberg { 1442949928c1SKeith Busch unsigned i, max; 1443749941f2SChristoph Hellwig int ret = 0; 144457dacad5SJay Sternberg 1445749941f2SChristoph Hellwig for (i = dev->queue_count; i <= dev->max_qid; i++) { 1446d3af3ecdSShaohua Li /* vector == qid - 1, match nvme_create_queue */ 1447d3af3ecdSShaohua Li if (!nvme_alloc_queue(dev, i, dev->q_depth, 1448d3af3ecdSShaohua Li pci_irq_get_node(to_pci_dev(dev->dev), i - 1))) { 1449749941f2SChristoph Hellwig ret = -ENOMEM; 145057dacad5SJay Sternberg break; 1451749941f2SChristoph Hellwig } 1452749941f2SChristoph Hellwig } 145357dacad5SJay Sternberg 1454949928c1SKeith Busch max = min(dev->max_qid, dev->queue_count - 1); 1455949928c1SKeith Busch for (i = dev->online_queues; i <= max; i++) { 1456749941f2SChristoph Hellwig ret = nvme_create_queue(dev->queues[i], i); 1457d4875622SKeith Busch if (ret) 145857dacad5SJay Sternberg break; 145957dacad5SJay Sternberg } 146057dacad5SJay Sternberg 1461749941f2SChristoph Hellwig /* 1462749941f2SChristoph Hellwig * Ignore failing Create SQ/CQ commands, we can continue with less 1463749941f2SChristoph Hellwig * than the desired aount of queues, and even a controller without 1464749941f2SChristoph Hellwig * I/O queues an still be used to issue admin commands. This might 1465749941f2SChristoph Hellwig * be useful to upgrade a buggy firmware for example. 1466749941f2SChristoph Hellwig */ 1467749941f2SChristoph Hellwig return ret >= 0 ? 0 : ret; 146857dacad5SJay Sternberg } 146957dacad5SJay Sternberg 1470202021c1SStephen Bates static ssize_t nvme_cmb_show(struct device *dev, 1471202021c1SStephen Bates struct device_attribute *attr, 1472202021c1SStephen Bates char *buf) 1473202021c1SStephen Bates { 1474202021c1SStephen Bates struct nvme_dev *ndev = to_nvme_dev(dev_get_drvdata(dev)); 1475202021c1SStephen Bates 1476c965809cSStephen Bates return scnprintf(buf, PAGE_SIZE, "cmbloc : x%08x\ncmbsz : x%08x\n", 1477202021c1SStephen Bates ndev->cmbloc, ndev->cmbsz); 1478202021c1SStephen Bates } 1479202021c1SStephen Bates static DEVICE_ATTR(cmb, S_IRUGO, nvme_cmb_show, NULL); 1480202021c1SStephen Bates 148157dacad5SJay Sternberg static void __iomem *nvme_map_cmb(struct nvme_dev *dev) 148257dacad5SJay Sternberg { 148357dacad5SJay Sternberg u64 szu, size, offset; 148457dacad5SJay Sternberg resource_size_t bar_size; 148557dacad5SJay Sternberg struct pci_dev *pdev = to_pci_dev(dev->dev); 148657dacad5SJay Sternberg void __iomem *cmb; 148757dacad5SJay Sternberg dma_addr_t dma_addr; 148857dacad5SJay Sternberg 14897a67cbeaSChristoph Hellwig dev->cmbsz = readl(dev->bar + NVME_REG_CMBSZ); 149057dacad5SJay Sternberg if (!(NVME_CMB_SZ(dev->cmbsz))) 149157dacad5SJay Sternberg return NULL; 1492202021c1SStephen Bates dev->cmbloc = readl(dev->bar + NVME_REG_CMBLOC); 149357dacad5SJay Sternberg 1494202021c1SStephen Bates if (!use_cmb_sqes) 1495202021c1SStephen Bates return NULL; 149657dacad5SJay Sternberg 149757dacad5SJay Sternberg szu = (u64)1 << (12 + 4 * NVME_CMB_SZU(dev->cmbsz)); 149857dacad5SJay Sternberg size = szu * NVME_CMB_SZ(dev->cmbsz); 1499202021c1SStephen Bates offset = szu * NVME_CMB_OFST(dev->cmbloc); 1500202021c1SStephen Bates bar_size = pci_resource_len(pdev, NVME_CMB_BIR(dev->cmbloc)); 150157dacad5SJay Sternberg 150257dacad5SJay Sternberg if (offset > bar_size) 150357dacad5SJay Sternberg return NULL; 150457dacad5SJay Sternberg 150557dacad5SJay Sternberg /* 150657dacad5SJay Sternberg * Controllers may support a CMB size larger than their BAR, 150757dacad5SJay Sternberg * for example, due to being behind a bridge. Reduce the CMB to 150857dacad5SJay Sternberg * the reported size of the BAR 150957dacad5SJay Sternberg */ 151057dacad5SJay Sternberg if (size > bar_size - offset) 151157dacad5SJay Sternberg size = bar_size - offset; 151257dacad5SJay Sternberg 1513202021c1SStephen Bates dma_addr = pci_resource_start(pdev, NVME_CMB_BIR(dev->cmbloc)) + offset; 151457dacad5SJay Sternberg cmb = ioremap_wc(dma_addr, size); 151557dacad5SJay Sternberg if (!cmb) 151657dacad5SJay Sternberg return NULL; 151757dacad5SJay Sternberg 151857dacad5SJay Sternberg dev->cmb_dma_addr = dma_addr; 151957dacad5SJay Sternberg dev->cmb_size = size; 152057dacad5SJay Sternberg return cmb; 152157dacad5SJay Sternberg } 152257dacad5SJay Sternberg 152357dacad5SJay Sternberg static inline void nvme_release_cmb(struct nvme_dev *dev) 152457dacad5SJay Sternberg { 152557dacad5SJay Sternberg if (dev->cmb) { 152657dacad5SJay Sternberg iounmap(dev->cmb); 152757dacad5SJay Sternberg dev->cmb = NULL; 1528f63572dfSJon Derrick if (dev->cmbsz) { 1529f63572dfSJon Derrick sysfs_remove_file_from_group(&dev->ctrl.device->kobj, 1530f63572dfSJon Derrick &dev_attr_cmb.attr, NULL); 1531f63572dfSJon Derrick dev->cmbsz = 0; 1532f63572dfSJon Derrick } 153357dacad5SJay Sternberg } 153457dacad5SJay Sternberg } 153557dacad5SJay Sternberg 153687ad72a5SChristoph Hellwig static int nvme_set_host_mem(struct nvme_dev *dev, u32 bits) 153787ad72a5SChristoph Hellwig { 153887ad72a5SChristoph Hellwig size_t len = dev->nr_host_mem_descs * sizeof(*dev->host_mem_descs); 153987ad72a5SChristoph Hellwig struct nvme_command c; 154087ad72a5SChristoph Hellwig u64 dma_addr; 154187ad72a5SChristoph Hellwig int ret; 154287ad72a5SChristoph Hellwig 154387ad72a5SChristoph Hellwig dma_addr = dma_map_single(dev->dev, dev->host_mem_descs, len, 154487ad72a5SChristoph Hellwig DMA_TO_DEVICE); 154587ad72a5SChristoph Hellwig if (dma_mapping_error(dev->dev, dma_addr)) 154687ad72a5SChristoph Hellwig return -ENOMEM; 154787ad72a5SChristoph Hellwig 154887ad72a5SChristoph Hellwig memset(&c, 0, sizeof(c)); 154987ad72a5SChristoph Hellwig c.features.opcode = nvme_admin_set_features; 155087ad72a5SChristoph Hellwig c.features.fid = cpu_to_le32(NVME_FEAT_HOST_MEM_BUF); 155187ad72a5SChristoph Hellwig c.features.dword11 = cpu_to_le32(bits); 155287ad72a5SChristoph Hellwig c.features.dword12 = cpu_to_le32(dev->host_mem_size >> 155387ad72a5SChristoph Hellwig ilog2(dev->ctrl.page_size)); 155487ad72a5SChristoph Hellwig c.features.dword13 = cpu_to_le32(lower_32_bits(dma_addr)); 155587ad72a5SChristoph Hellwig c.features.dword14 = cpu_to_le32(upper_32_bits(dma_addr)); 155687ad72a5SChristoph Hellwig c.features.dword15 = cpu_to_le32(dev->nr_host_mem_descs); 155787ad72a5SChristoph Hellwig 155887ad72a5SChristoph Hellwig ret = nvme_submit_sync_cmd(dev->ctrl.admin_q, &c, NULL, 0); 155987ad72a5SChristoph Hellwig if (ret) { 156087ad72a5SChristoph Hellwig dev_warn(dev->ctrl.device, 156187ad72a5SChristoph Hellwig "failed to set host mem (err %d, flags %#x).\n", 156287ad72a5SChristoph Hellwig ret, bits); 156387ad72a5SChristoph Hellwig } 156487ad72a5SChristoph Hellwig dma_unmap_single(dev->dev, dma_addr, len, DMA_TO_DEVICE); 156587ad72a5SChristoph Hellwig return ret; 156687ad72a5SChristoph Hellwig } 156787ad72a5SChristoph Hellwig 156887ad72a5SChristoph Hellwig static void nvme_free_host_mem(struct nvme_dev *dev) 156987ad72a5SChristoph Hellwig { 157087ad72a5SChristoph Hellwig int i; 157187ad72a5SChristoph Hellwig 157287ad72a5SChristoph Hellwig for (i = 0; i < dev->nr_host_mem_descs; i++) { 157387ad72a5SChristoph Hellwig struct nvme_host_mem_buf_desc *desc = &dev->host_mem_descs[i]; 157487ad72a5SChristoph Hellwig size_t size = le32_to_cpu(desc->size) * dev->ctrl.page_size; 157587ad72a5SChristoph Hellwig 157687ad72a5SChristoph Hellwig dma_free_coherent(dev->dev, size, dev->host_mem_desc_bufs[i], 157787ad72a5SChristoph Hellwig le64_to_cpu(desc->addr)); 157887ad72a5SChristoph Hellwig } 157987ad72a5SChristoph Hellwig 158087ad72a5SChristoph Hellwig kfree(dev->host_mem_desc_bufs); 158187ad72a5SChristoph Hellwig dev->host_mem_desc_bufs = NULL; 158287ad72a5SChristoph Hellwig kfree(dev->host_mem_descs); 158387ad72a5SChristoph Hellwig dev->host_mem_descs = NULL; 158487ad72a5SChristoph Hellwig } 158587ad72a5SChristoph Hellwig 158687ad72a5SChristoph Hellwig static int nvme_alloc_host_mem(struct nvme_dev *dev, u64 min, u64 preferred) 158787ad72a5SChristoph Hellwig { 158887ad72a5SChristoph Hellwig struct nvme_host_mem_buf_desc *descs; 158987ad72a5SChristoph Hellwig u32 chunk_size, max_entries, i = 0; 159087ad72a5SChristoph Hellwig void **bufs; 159187ad72a5SChristoph Hellwig u64 size, tmp; 159287ad72a5SChristoph Hellwig 159387ad72a5SChristoph Hellwig /* start big and work our way down */ 159487ad72a5SChristoph Hellwig chunk_size = min(preferred, (u64)PAGE_SIZE << MAX_ORDER); 159587ad72a5SChristoph Hellwig retry: 159687ad72a5SChristoph Hellwig tmp = (preferred + chunk_size - 1); 159787ad72a5SChristoph Hellwig do_div(tmp, chunk_size); 159887ad72a5SChristoph Hellwig max_entries = tmp; 159987ad72a5SChristoph Hellwig descs = kcalloc(max_entries, sizeof(*descs), GFP_KERNEL); 160087ad72a5SChristoph Hellwig if (!descs) 160187ad72a5SChristoph Hellwig goto out; 160287ad72a5SChristoph Hellwig 160387ad72a5SChristoph Hellwig bufs = kcalloc(max_entries, sizeof(*bufs), GFP_KERNEL); 160487ad72a5SChristoph Hellwig if (!bufs) 160587ad72a5SChristoph Hellwig goto out_free_descs; 160687ad72a5SChristoph Hellwig 160787ad72a5SChristoph Hellwig for (size = 0; size < preferred; size += chunk_size) { 160887ad72a5SChristoph Hellwig u32 len = min_t(u64, chunk_size, preferred - size); 160987ad72a5SChristoph Hellwig dma_addr_t dma_addr; 161087ad72a5SChristoph Hellwig 161187ad72a5SChristoph Hellwig bufs[i] = dma_alloc_attrs(dev->dev, len, &dma_addr, GFP_KERNEL, 161287ad72a5SChristoph Hellwig DMA_ATTR_NO_KERNEL_MAPPING | DMA_ATTR_NO_WARN); 161387ad72a5SChristoph Hellwig if (!bufs[i]) 161487ad72a5SChristoph Hellwig break; 161587ad72a5SChristoph Hellwig 161687ad72a5SChristoph Hellwig descs[i].addr = cpu_to_le64(dma_addr); 161787ad72a5SChristoph Hellwig descs[i].size = cpu_to_le32(len / dev->ctrl.page_size); 161887ad72a5SChristoph Hellwig i++; 161987ad72a5SChristoph Hellwig } 162087ad72a5SChristoph Hellwig 162187ad72a5SChristoph Hellwig if (!size || (min && size < min)) { 162287ad72a5SChristoph Hellwig dev_warn(dev->ctrl.device, 162387ad72a5SChristoph Hellwig "failed to allocate host memory buffer.\n"); 162487ad72a5SChristoph Hellwig goto out_free_bufs; 162587ad72a5SChristoph Hellwig } 162687ad72a5SChristoph Hellwig 162787ad72a5SChristoph Hellwig dev_info(dev->ctrl.device, 162887ad72a5SChristoph Hellwig "allocated %lld MiB host memory buffer.\n", 162987ad72a5SChristoph Hellwig size >> ilog2(SZ_1M)); 163087ad72a5SChristoph Hellwig dev->nr_host_mem_descs = i; 163187ad72a5SChristoph Hellwig dev->host_mem_size = size; 163287ad72a5SChristoph Hellwig dev->host_mem_descs = descs; 163387ad72a5SChristoph Hellwig dev->host_mem_desc_bufs = bufs; 163487ad72a5SChristoph Hellwig return 0; 163587ad72a5SChristoph Hellwig 163687ad72a5SChristoph Hellwig out_free_bufs: 163787ad72a5SChristoph Hellwig while (--i >= 0) { 163887ad72a5SChristoph Hellwig size_t size = le32_to_cpu(descs[i].size) * dev->ctrl.page_size; 163987ad72a5SChristoph Hellwig 164087ad72a5SChristoph Hellwig dma_free_coherent(dev->dev, size, bufs[i], 164187ad72a5SChristoph Hellwig le64_to_cpu(descs[i].addr)); 164287ad72a5SChristoph Hellwig } 164387ad72a5SChristoph Hellwig 164487ad72a5SChristoph Hellwig kfree(bufs); 164587ad72a5SChristoph Hellwig out_free_descs: 164687ad72a5SChristoph Hellwig kfree(descs); 164787ad72a5SChristoph Hellwig out: 164887ad72a5SChristoph Hellwig /* try a smaller chunk size if we failed early */ 164987ad72a5SChristoph Hellwig if (chunk_size >= PAGE_SIZE * 2 && (i == 0 || size < min)) { 165087ad72a5SChristoph Hellwig chunk_size /= 2; 165187ad72a5SChristoph Hellwig goto retry; 165287ad72a5SChristoph Hellwig } 165387ad72a5SChristoph Hellwig dev->host_mem_descs = NULL; 165487ad72a5SChristoph Hellwig return -ENOMEM; 165587ad72a5SChristoph Hellwig } 165687ad72a5SChristoph Hellwig 165787ad72a5SChristoph Hellwig static void nvme_setup_host_mem(struct nvme_dev *dev) 165887ad72a5SChristoph Hellwig { 165987ad72a5SChristoph Hellwig u64 max = (u64)max_host_mem_size_mb * SZ_1M; 166087ad72a5SChristoph Hellwig u64 preferred = (u64)dev->ctrl.hmpre * 4096; 166187ad72a5SChristoph Hellwig u64 min = (u64)dev->ctrl.hmmin * 4096; 166287ad72a5SChristoph Hellwig u32 enable_bits = NVME_HOST_MEM_ENABLE; 166387ad72a5SChristoph Hellwig 166487ad72a5SChristoph Hellwig preferred = min(preferred, max); 166587ad72a5SChristoph Hellwig if (min > max) { 166687ad72a5SChristoph Hellwig dev_warn(dev->ctrl.device, 166787ad72a5SChristoph Hellwig "min host memory (%lld MiB) above limit (%d MiB).\n", 166887ad72a5SChristoph Hellwig min >> ilog2(SZ_1M), max_host_mem_size_mb); 166987ad72a5SChristoph Hellwig nvme_free_host_mem(dev); 167087ad72a5SChristoph Hellwig return; 167187ad72a5SChristoph Hellwig } 167287ad72a5SChristoph Hellwig 167387ad72a5SChristoph Hellwig /* 167487ad72a5SChristoph Hellwig * If we already have a buffer allocated check if we can reuse it. 167587ad72a5SChristoph Hellwig */ 167687ad72a5SChristoph Hellwig if (dev->host_mem_descs) { 167787ad72a5SChristoph Hellwig if (dev->host_mem_size >= min) 167887ad72a5SChristoph Hellwig enable_bits |= NVME_HOST_MEM_RETURN; 167987ad72a5SChristoph Hellwig else 168087ad72a5SChristoph Hellwig nvme_free_host_mem(dev); 168187ad72a5SChristoph Hellwig } 168287ad72a5SChristoph Hellwig 168387ad72a5SChristoph Hellwig if (!dev->host_mem_descs) { 168487ad72a5SChristoph Hellwig if (nvme_alloc_host_mem(dev, min, preferred)) 168587ad72a5SChristoph Hellwig return; 168687ad72a5SChristoph Hellwig } 168787ad72a5SChristoph Hellwig 168887ad72a5SChristoph Hellwig if (nvme_set_host_mem(dev, enable_bits)) 168987ad72a5SChristoph Hellwig nvme_free_host_mem(dev); 169087ad72a5SChristoph Hellwig } 169187ad72a5SChristoph Hellwig 169257dacad5SJay Sternberg static int nvme_setup_io_queues(struct nvme_dev *dev) 169357dacad5SJay Sternberg { 169457dacad5SJay Sternberg struct nvme_queue *adminq = dev->queues[0]; 169557dacad5SJay Sternberg struct pci_dev *pdev = to_pci_dev(dev->dev); 169697f6ef64SXu Yu int result, nr_io_queues; 169797f6ef64SXu Yu unsigned long size; 169857dacad5SJay Sternberg 16992800b8e7SKeith Busch nr_io_queues = num_online_cpus(); 17009a0be7abSChristoph Hellwig result = nvme_set_queue_count(&dev->ctrl, &nr_io_queues); 17019a0be7abSChristoph Hellwig if (result < 0) 170257dacad5SJay Sternberg return result; 17039a0be7abSChristoph Hellwig 1704f5fa90dcSChristoph Hellwig if (nr_io_queues == 0) 1705a5229050SKeith Busch return 0; 170657dacad5SJay Sternberg 170757dacad5SJay Sternberg if (dev->cmb && NVME_CMB_SQS(dev->cmbsz)) { 170857dacad5SJay Sternberg result = nvme_cmb_qdepth(dev, nr_io_queues, 170957dacad5SJay Sternberg sizeof(struct nvme_command)); 171057dacad5SJay Sternberg if (result > 0) 171157dacad5SJay Sternberg dev->q_depth = result; 171257dacad5SJay Sternberg else 171357dacad5SJay Sternberg nvme_release_cmb(dev); 171457dacad5SJay Sternberg } 171557dacad5SJay Sternberg 171657dacad5SJay Sternberg do { 171797f6ef64SXu Yu size = db_bar_size(dev, nr_io_queues); 171897f6ef64SXu Yu result = nvme_remap_bar(dev, size); 171997f6ef64SXu Yu if (!result) 172057dacad5SJay Sternberg break; 172157dacad5SJay Sternberg if (!--nr_io_queues) 172257dacad5SJay Sternberg return -ENOMEM; 172357dacad5SJay Sternberg } while (1); 172457dacad5SJay Sternberg adminq->q_db = dev->dbs; 172557dacad5SJay Sternberg 172657dacad5SJay Sternberg /* Deregister the admin queue's interrupt */ 17270ff199cbSChristoph Hellwig pci_free_irq(pdev, 0, adminq); 172857dacad5SJay Sternberg 172957dacad5SJay Sternberg /* 173057dacad5SJay Sternberg * If we enable msix early due to not intx, disable it again before 173157dacad5SJay Sternberg * setting up the full range we need. 173257dacad5SJay Sternberg */ 1733dca51e78SChristoph Hellwig pci_free_irq_vectors(pdev); 1734dca51e78SChristoph Hellwig nr_io_queues = pci_alloc_irq_vectors(pdev, 1, nr_io_queues, 1735dca51e78SChristoph Hellwig PCI_IRQ_ALL_TYPES | PCI_IRQ_AFFINITY); 1736dca51e78SChristoph Hellwig if (nr_io_queues <= 0) 1737dca51e78SChristoph Hellwig return -EIO; 1738dca51e78SChristoph Hellwig dev->max_qid = nr_io_queues; 173957dacad5SJay Sternberg 174057dacad5SJay Sternberg /* 174157dacad5SJay Sternberg * Should investigate if there's a performance win from allocating 174257dacad5SJay Sternberg * more queues than interrupt vectors; it might allow the submission 174357dacad5SJay Sternberg * path to scale better, even if the receive path is limited by the 174457dacad5SJay Sternberg * number of interrupts. 174557dacad5SJay Sternberg */ 174657dacad5SJay Sternberg 1747dca51e78SChristoph Hellwig result = queue_request_irq(adminq); 174857dacad5SJay Sternberg if (result) { 174957dacad5SJay Sternberg adminq->cq_vector = -1; 1750d4875622SKeith Busch return result; 175157dacad5SJay Sternberg } 1752749941f2SChristoph Hellwig return nvme_create_io_queues(dev); 175357dacad5SJay Sternberg } 175457dacad5SJay Sternberg 17552a842acaSChristoph Hellwig static void nvme_del_queue_end(struct request *req, blk_status_t error) 1756db3cbfffSKeith Busch { 1757db3cbfffSKeith Busch struct nvme_queue *nvmeq = req->end_io_data; 1758db3cbfffSKeith Busch 1759db3cbfffSKeith Busch blk_mq_free_request(req); 1760db3cbfffSKeith Busch complete(&nvmeq->dev->ioq_wait); 1761db3cbfffSKeith Busch } 1762db3cbfffSKeith Busch 17632a842acaSChristoph Hellwig static void nvme_del_cq_end(struct request *req, blk_status_t error) 1764db3cbfffSKeith Busch { 1765db3cbfffSKeith Busch struct nvme_queue *nvmeq = req->end_io_data; 1766db3cbfffSKeith Busch 1767db3cbfffSKeith Busch if (!error) { 1768db3cbfffSKeith Busch unsigned long flags; 1769db3cbfffSKeith Busch 17702e39e0f6SMing Lin /* 17712e39e0f6SMing Lin * We might be called with the AQ q_lock held 17722e39e0f6SMing Lin * and the I/O queue q_lock should always 17732e39e0f6SMing Lin * nest inside the AQ one. 17742e39e0f6SMing Lin */ 17752e39e0f6SMing Lin spin_lock_irqsave_nested(&nvmeq->q_lock, flags, 17762e39e0f6SMing Lin SINGLE_DEPTH_NESTING); 1777db3cbfffSKeith Busch nvme_process_cq(nvmeq); 1778db3cbfffSKeith Busch spin_unlock_irqrestore(&nvmeq->q_lock, flags); 1779db3cbfffSKeith Busch } 1780db3cbfffSKeith Busch 1781db3cbfffSKeith Busch nvme_del_queue_end(req, error); 1782db3cbfffSKeith Busch } 1783db3cbfffSKeith Busch 1784db3cbfffSKeith Busch static int nvme_delete_queue(struct nvme_queue *nvmeq, u8 opcode) 1785db3cbfffSKeith Busch { 1786db3cbfffSKeith Busch struct request_queue *q = nvmeq->dev->ctrl.admin_q; 1787db3cbfffSKeith Busch struct request *req; 1788db3cbfffSKeith Busch struct nvme_command cmd; 1789db3cbfffSKeith Busch 1790db3cbfffSKeith Busch memset(&cmd, 0, sizeof(cmd)); 1791db3cbfffSKeith Busch cmd.delete_queue.opcode = opcode; 1792db3cbfffSKeith Busch cmd.delete_queue.qid = cpu_to_le16(nvmeq->qid); 1793db3cbfffSKeith Busch 1794eb71f435SChristoph Hellwig req = nvme_alloc_request(q, &cmd, BLK_MQ_REQ_NOWAIT, NVME_QID_ANY); 1795db3cbfffSKeith Busch if (IS_ERR(req)) 1796db3cbfffSKeith Busch return PTR_ERR(req); 1797db3cbfffSKeith Busch 1798db3cbfffSKeith Busch req->timeout = ADMIN_TIMEOUT; 1799db3cbfffSKeith Busch req->end_io_data = nvmeq; 1800db3cbfffSKeith Busch 1801db3cbfffSKeith Busch blk_execute_rq_nowait(q, NULL, req, false, 1802db3cbfffSKeith Busch opcode == nvme_admin_delete_cq ? 1803db3cbfffSKeith Busch nvme_del_cq_end : nvme_del_queue_end); 1804db3cbfffSKeith Busch return 0; 1805db3cbfffSKeith Busch } 1806db3cbfffSKeith Busch 180770659060SKeith Busch static void nvme_disable_io_queues(struct nvme_dev *dev, int queues) 1808db3cbfffSKeith Busch { 180970659060SKeith Busch int pass; 1810db3cbfffSKeith Busch unsigned long timeout; 1811db3cbfffSKeith Busch u8 opcode = nvme_admin_delete_sq; 1812db3cbfffSKeith Busch 1813db3cbfffSKeith Busch for (pass = 0; pass < 2; pass++) { 1814014a0d60SKeith Busch int sent = 0, i = queues; 1815db3cbfffSKeith Busch 1816db3cbfffSKeith Busch reinit_completion(&dev->ioq_wait); 1817db3cbfffSKeith Busch retry: 1818db3cbfffSKeith Busch timeout = ADMIN_TIMEOUT; 1819c21377f8SGabriel Krisman Bertazi for (; i > 0; i--, sent++) 1820c21377f8SGabriel Krisman Bertazi if (nvme_delete_queue(dev->queues[i], opcode)) 1821db3cbfffSKeith Busch break; 1822c21377f8SGabriel Krisman Bertazi 1823db3cbfffSKeith Busch while (sent--) { 1824db3cbfffSKeith Busch timeout = wait_for_completion_io_timeout(&dev->ioq_wait, timeout); 1825db3cbfffSKeith Busch if (timeout == 0) 1826db3cbfffSKeith Busch return; 1827db3cbfffSKeith Busch if (i) 1828db3cbfffSKeith Busch goto retry; 1829db3cbfffSKeith Busch } 1830db3cbfffSKeith Busch opcode = nvme_admin_delete_cq; 1831db3cbfffSKeith Busch } 1832db3cbfffSKeith Busch } 1833db3cbfffSKeith Busch 183457dacad5SJay Sternberg /* 183557dacad5SJay Sternberg * Return: error value if an error occurred setting up the queues or calling 183657dacad5SJay Sternberg * Identify Device. 0 if these succeeded, even if adding some of the 183757dacad5SJay Sternberg * namespaces failed. At the moment, these failures are silent. TBD which 183857dacad5SJay Sternberg * failures should be reported. 183957dacad5SJay Sternberg */ 184057dacad5SJay Sternberg static int nvme_dev_add(struct nvme_dev *dev) 184157dacad5SJay Sternberg { 18425bae7f73SChristoph Hellwig if (!dev->ctrl.tagset) { 184357dacad5SJay Sternberg dev->tagset.ops = &nvme_mq_ops; 184457dacad5SJay Sternberg dev->tagset.nr_hw_queues = dev->online_queues - 1; 184557dacad5SJay Sternberg dev->tagset.timeout = NVME_IO_TIMEOUT; 184657dacad5SJay Sternberg dev->tagset.numa_node = dev_to_node(dev->dev); 184757dacad5SJay Sternberg dev->tagset.queue_depth = 184857dacad5SJay Sternberg min_t(int, dev->q_depth, BLK_MQ_MAX_DEPTH) - 1; 184957dacad5SJay Sternberg dev->tagset.cmd_size = nvme_cmd_size(dev); 185057dacad5SJay Sternberg dev->tagset.flags = BLK_MQ_F_SHOULD_MERGE; 185157dacad5SJay Sternberg dev->tagset.driver_data = dev; 185257dacad5SJay Sternberg 185357dacad5SJay Sternberg if (blk_mq_alloc_tag_set(&dev->tagset)) 185457dacad5SJay Sternberg return 0; 18555bae7f73SChristoph Hellwig dev->ctrl.tagset = &dev->tagset; 1856f9f38e33SHelen Koike 1857f9f38e33SHelen Koike nvme_dbbuf_set(dev); 1858949928c1SKeith Busch } else { 1859949928c1SKeith Busch blk_mq_update_nr_hw_queues(&dev->tagset, dev->online_queues - 1); 1860949928c1SKeith Busch 1861949928c1SKeith Busch /* Free previously allocated queues that are no longer usable */ 1862949928c1SKeith Busch nvme_free_queues(dev, dev->online_queues); 186357dacad5SJay Sternberg } 1864949928c1SKeith Busch 186557dacad5SJay Sternberg return 0; 186657dacad5SJay Sternberg } 186757dacad5SJay Sternberg 1868b00a726aSKeith Busch static int nvme_pci_enable(struct nvme_dev *dev) 186957dacad5SJay Sternberg { 187057dacad5SJay Sternberg u64 cap; 1871b00a726aSKeith Busch int result = -ENOMEM; 187257dacad5SJay Sternberg struct pci_dev *pdev = to_pci_dev(dev->dev); 187357dacad5SJay Sternberg 187457dacad5SJay Sternberg if (pci_enable_device_mem(pdev)) 187557dacad5SJay Sternberg return result; 187657dacad5SJay Sternberg 187757dacad5SJay Sternberg pci_set_master(pdev); 187857dacad5SJay Sternberg 187957dacad5SJay Sternberg if (dma_set_mask_and_coherent(dev->dev, DMA_BIT_MASK(64)) && 188057dacad5SJay Sternberg dma_set_mask_and_coherent(dev->dev, DMA_BIT_MASK(32))) 188157dacad5SJay Sternberg goto disable; 188257dacad5SJay Sternberg 18837a67cbeaSChristoph Hellwig if (readl(dev->bar + NVME_REG_CSTS) == -1) { 188457dacad5SJay Sternberg result = -ENODEV; 1885b00a726aSKeith Busch goto disable; 188657dacad5SJay Sternberg } 188757dacad5SJay Sternberg 188857dacad5SJay Sternberg /* 1889a5229050SKeith Busch * Some devices and/or platforms don't advertise or work with INTx 1890a5229050SKeith Busch * interrupts. Pre-enable a single MSIX or MSI vec for setup. We'll 1891a5229050SKeith Busch * adjust this later. 189257dacad5SJay Sternberg */ 1893dca51e78SChristoph Hellwig result = pci_alloc_irq_vectors(pdev, 1, 1, PCI_IRQ_ALL_TYPES); 1894dca51e78SChristoph Hellwig if (result < 0) 1895dca51e78SChristoph Hellwig return result; 189657dacad5SJay Sternberg 18977a67cbeaSChristoph Hellwig cap = lo_hi_readq(dev->bar + NVME_REG_CAP); 18987a67cbeaSChristoph Hellwig 189957dacad5SJay Sternberg dev->q_depth = min_t(int, NVME_CAP_MQES(cap) + 1, NVME_Q_DEPTH); 190057dacad5SJay Sternberg dev->db_stride = 1 << NVME_CAP_STRIDE(cap); 19017a67cbeaSChristoph Hellwig dev->dbs = dev->bar + 4096; 19021f390c1fSStephan Günther 19031f390c1fSStephan Günther /* 19041f390c1fSStephan Günther * Temporary fix for the Apple controller found in the MacBook8,1 and 19051f390c1fSStephan Günther * some MacBook7,1 to avoid controller resets and data loss. 19061f390c1fSStephan Günther */ 19071f390c1fSStephan Günther if (pdev->vendor == PCI_VENDOR_ID_APPLE && pdev->device == 0x2001) { 19081f390c1fSStephan Günther dev->q_depth = 2; 19099bdcfb10SChristoph Hellwig dev_warn(dev->ctrl.device, "detected Apple NVMe controller, " 19109bdcfb10SChristoph Hellwig "set queue depth=%u to work around controller resets\n", 19111f390c1fSStephan Günther dev->q_depth); 19121f390c1fSStephan Günther } 19131f390c1fSStephan Günther 1914202021c1SStephen Bates /* 1915202021c1SStephen Bates * CMBs can currently only exist on >=1.2 PCIe devices. We only 1916202021c1SStephen Bates * populate sysfs if a CMB is implemented. Note that we add the 1917202021c1SStephen Bates * CMB attribute to the nvme_ctrl kobj which removes the need to remove 1918202021c1SStephen Bates * it on exit. Since nvme_dev_attrs_group has no name we can pass 1919202021c1SStephen Bates * NULL as final argument to sysfs_add_file_to_group. 1920202021c1SStephen Bates */ 1921202021c1SStephen Bates 19228ef2074dSGabriel Krisman Bertazi if (readl(dev->bar + NVME_REG_VS) >= NVME_VS(1, 2, 0)) { 192357dacad5SJay Sternberg dev->cmb = nvme_map_cmb(dev); 192457dacad5SJay Sternberg 1925202021c1SStephen Bates if (dev->cmbsz) { 1926202021c1SStephen Bates if (sysfs_add_file_to_group(&dev->ctrl.device->kobj, 1927202021c1SStephen Bates &dev_attr_cmb.attr, NULL)) 19289bdcfb10SChristoph Hellwig dev_warn(dev->ctrl.device, 1929202021c1SStephen Bates "failed to add sysfs attribute for CMB\n"); 1930202021c1SStephen Bates } 1931202021c1SStephen Bates } 1932202021c1SStephen Bates 1933a0a3408eSKeith Busch pci_enable_pcie_error_reporting(pdev); 1934a0a3408eSKeith Busch pci_save_state(pdev); 193557dacad5SJay Sternberg return 0; 193657dacad5SJay Sternberg 193757dacad5SJay Sternberg disable: 193857dacad5SJay Sternberg pci_disable_device(pdev); 193957dacad5SJay Sternberg return result; 194057dacad5SJay Sternberg } 194157dacad5SJay Sternberg 194257dacad5SJay Sternberg static void nvme_dev_unmap(struct nvme_dev *dev) 194357dacad5SJay Sternberg { 1944b00a726aSKeith Busch if (dev->bar) 1945b00a726aSKeith Busch iounmap(dev->bar); 1946a1f447b3SJohannes Thumshirn pci_release_mem_regions(to_pci_dev(dev->dev)); 1947b00a726aSKeith Busch } 1948b00a726aSKeith Busch 1949b00a726aSKeith Busch static void nvme_pci_disable(struct nvme_dev *dev) 1950b00a726aSKeith Busch { 195157dacad5SJay Sternberg struct pci_dev *pdev = to_pci_dev(dev->dev); 195257dacad5SJay Sternberg 1953f63572dfSJon Derrick nvme_release_cmb(dev); 1954dca51e78SChristoph Hellwig pci_free_irq_vectors(pdev); 195557dacad5SJay Sternberg 1956a0a3408eSKeith Busch if (pci_is_enabled(pdev)) { 1957a0a3408eSKeith Busch pci_disable_pcie_error_reporting(pdev); 195857dacad5SJay Sternberg pci_disable_device(pdev); 195957dacad5SJay Sternberg } 1960a0a3408eSKeith Busch } 196157dacad5SJay Sternberg 1962a5cdb68cSKeith Busch static void nvme_dev_disable(struct nvme_dev *dev, bool shutdown) 196357dacad5SJay Sternberg { 196470659060SKeith Busch int i, queues; 1965302ad8ccSKeith Busch bool dead = true; 1966302ad8ccSKeith Busch struct pci_dev *pdev = to_pci_dev(dev->dev); 196757dacad5SJay Sternberg 196877bf25eaSKeith Busch mutex_lock(&dev->shutdown_lock); 1969302ad8ccSKeith Busch if (pci_is_enabled(pdev)) { 1970302ad8ccSKeith Busch u32 csts = readl(dev->bar + NVME_REG_CSTS); 1971302ad8ccSKeith Busch 1972302ad8ccSKeith Busch if (dev->ctrl.state == NVME_CTRL_LIVE) 1973302ad8ccSKeith Busch nvme_start_freeze(&dev->ctrl); 1974302ad8ccSKeith Busch dead = !!((csts & NVME_CSTS_CFS) || !(csts & NVME_CSTS_RDY) || 1975302ad8ccSKeith Busch pdev->error_state != pci_channel_io_normal); 197657dacad5SJay Sternberg } 1977c21377f8SGabriel Krisman Bertazi 1978302ad8ccSKeith Busch /* 1979302ad8ccSKeith Busch * Give the controller a chance to complete all entered requests if 1980302ad8ccSKeith Busch * doing a safe shutdown. 1981302ad8ccSKeith Busch */ 198287ad72a5SChristoph Hellwig if (!dead) { 198387ad72a5SChristoph Hellwig if (shutdown) 1984302ad8ccSKeith Busch nvme_wait_freeze_timeout(&dev->ctrl, NVME_IO_TIMEOUT); 198587ad72a5SChristoph Hellwig 198687ad72a5SChristoph Hellwig /* 198787ad72a5SChristoph Hellwig * If the controller is still alive tell it to stop using the 198887ad72a5SChristoph Hellwig * host memory buffer. In theory the shutdown / reset should 198987ad72a5SChristoph Hellwig * make sure that it doesn't access the host memoery anymore, 199087ad72a5SChristoph Hellwig * but I'd rather be safe than sorry.. 199187ad72a5SChristoph Hellwig */ 199287ad72a5SChristoph Hellwig if (dev->host_mem_descs) 199387ad72a5SChristoph Hellwig nvme_set_host_mem(dev, 0); 199487ad72a5SChristoph Hellwig 199587ad72a5SChristoph Hellwig } 1996302ad8ccSKeith Busch nvme_stop_queues(&dev->ctrl); 1997302ad8ccSKeith Busch 199870659060SKeith Busch queues = dev->online_queues - 1; 1999c21377f8SGabriel Krisman Bertazi for (i = dev->queue_count - 1; i > 0; i--) 2000c21377f8SGabriel Krisman Bertazi nvme_suspend_queue(dev->queues[i]); 2001c21377f8SGabriel Krisman Bertazi 2002302ad8ccSKeith Busch if (dead) { 200382469c59SGabriel Krisman Bertazi /* A device might become IO incapable very soon during 200482469c59SGabriel Krisman Bertazi * probe, before the admin queue is configured. Thus, 200582469c59SGabriel Krisman Bertazi * queue_count can be 0 here. 200682469c59SGabriel Krisman Bertazi */ 200782469c59SGabriel Krisman Bertazi if (dev->queue_count) 2008c21377f8SGabriel Krisman Bertazi nvme_suspend_queue(dev->queues[0]); 200957dacad5SJay Sternberg } else { 201070659060SKeith Busch nvme_disable_io_queues(dev, queues); 2011a5cdb68cSKeith Busch nvme_disable_admin_queue(dev, shutdown); 201257dacad5SJay Sternberg } 2013b00a726aSKeith Busch nvme_pci_disable(dev); 201457dacad5SJay Sternberg 2015e1958e65SMing Lin blk_mq_tagset_busy_iter(&dev->tagset, nvme_cancel_request, &dev->ctrl); 2016e1958e65SMing Lin blk_mq_tagset_busy_iter(&dev->admin_tagset, nvme_cancel_request, &dev->ctrl); 2017302ad8ccSKeith Busch 2018302ad8ccSKeith Busch /* 2019302ad8ccSKeith Busch * The driver will not be starting up queues again if shutting down so 2020302ad8ccSKeith Busch * must flush all entered requests to their failed completion to avoid 2021302ad8ccSKeith Busch * deadlocking blk-mq hot-cpu notifier. 2022302ad8ccSKeith Busch */ 2023302ad8ccSKeith Busch if (shutdown) 2024302ad8ccSKeith Busch nvme_start_queues(&dev->ctrl); 202577bf25eaSKeith Busch mutex_unlock(&dev->shutdown_lock); 202657dacad5SJay Sternberg } 202757dacad5SJay Sternberg 202857dacad5SJay Sternberg static int nvme_setup_prp_pools(struct nvme_dev *dev) 202957dacad5SJay Sternberg { 203057dacad5SJay Sternberg dev->prp_page_pool = dma_pool_create("prp list page", dev->dev, 203157dacad5SJay Sternberg PAGE_SIZE, PAGE_SIZE, 0); 203257dacad5SJay Sternberg if (!dev->prp_page_pool) 203357dacad5SJay Sternberg return -ENOMEM; 203457dacad5SJay Sternberg 203557dacad5SJay Sternberg /* Optimisation for I/Os between 4k and 128k */ 203657dacad5SJay Sternberg dev->prp_small_pool = dma_pool_create("prp list 256", dev->dev, 203757dacad5SJay Sternberg 256, 256, 0); 203857dacad5SJay Sternberg if (!dev->prp_small_pool) { 203957dacad5SJay Sternberg dma_pool_destroy(dev->prp_page_pool); 204057dacad5SJay Sternberg return -ENOMEM; 204157dacad5SJay Sternberg } 204257dacad5SJay Sternberg return 0; 204357dacad5SJay Sternberg } 204457dacad5SJay Sternberg 204557dacad5SJay Sternberg static void nvme_release_prp_pools(struct nvme_dev *dev) 204657dacad5SJay Sternberg { 204757dacad5SJay Sternberg dma_pool_destroy(dev->prp_page_pool); 204857dacad5SJay Sternberg dma_pool_destroy(dev->prp_small_pool); 204957dacad5SJay Sternberg } 205057dacad5SJay Sternberg 20511673f1f0SChristoph Hellwig static void nvme_pci_free_ctrl(struct nvme_ctrl *ctrl) 205257dacad5SJay Sternberg { 20531673f1f0SChristoph Hellwig struct nvme_dev *dev = to_nvme_dev(ctrl); 205457dacad5SJay Sternberg 2055f9f38e33SHelen Koike nvme_dbbuf_dma_free(dev); 205657dacad5SJay Sternberg put_device(dev->dev); 205757dacad5SJay Sternberg if (dev->tagset.tags) 205857dacad5SJay Sternberg blk_mq_free_tag_set(&dev->tagset); 20591c63dc66SChristoph Hellwig if (dev->ctrl.admin_q) 20601c63dc66SChristoph Hellwig blk_put_queue(dev->ctrl.admin_q); 206157dacad5SJay Sternberg kfree(dev->queues); 2062e286bcfcSScott Bauer free_opal_dev(dev->ctrl.opal_dev); 206357dacad5SJay Sternberg kfree(dev); 206457dacad5SJay Sternberg } 206557dacad5SJay Sternberg 2066f58944e2SKeith Busch static void nvme_remove_dead_ctrl(struct nvme_dev *dev, int status) 2067f58944e2SKeith Busch { 2068237045fcSLinus Torvalds dev_warn(dev->ctrl.device, "Removing after probe failure status: %d\n", status); 2069f58944e2SKeith Busch 2070f58944e2SKeith Busch kref_get(&dev->ctrl.kref); 207169d9a99cSKeith Busch nvme_dev_disable(dev, false); 2072f58944e2SKeith Busch if (!schedule_work(&dev->remove_work)) 2073f58944e2SKeith Busch nvme_put_ctrl(&dev->ctrl); 2074f58944e2SKeith Busch } 2075f58944e2SKeith Busch 2076fd634f41SChristoph Hellwig static void nvme_reset_work(struct work_struct *work) 207757dacad5SJay Sternberg { 2078d86c4d8eSChristoph Hellwig struct nvme_dev *dev = 2079d86c4d8eSChristoph Hellwig container_of(work, struct nvme_dev, ctrl.reset_work); 2080a98e58e5SScott Bauer bool was_suspend = !!(dev->ctrl.ctrl_config & NVME_CC_SHN_NORMAL); 2081f58944e2SKeith Busch int result = -ENODEV; 208257dacad5SJay Sternberg 208382b057caSRakesh Pandit if (WARN_ON(dev->ctrl.state != NVME_CTRL_RESETTING)) 2084fd634f41SChristoph Hellwig goto out; 2085fd634f41SChristoph Hellwig 2086fd634f41SChristoph Hellwig /* 2087fd634f41SChristoph Hellwig * If we're called to reset a live controller first shut it down before 2088fd634f41SChristoph Hellwig * moving on. 2089fd634f41SChristoph Hellwig */ 2090b00a726aSKeith Busch if (dev->ctrl.ctrl_config & NVME_CC_ENABLE) 2091a5cdb68cSKeith Busch nvme_dev_disable(dev, false); 2092fd634f41SChristoph Hellwig 2093b00a726aSKeith Busch result = nvme_pci_enable(dev); 209457dacad5SJay Sternberg if (result) 209557dacad5SJay Sternberg goto out; 209657dacad5SJay Sternberg 209757dacad5SJay Sternberg result = nvme_configure_admin_queue(dev); 209857dacad5SJay Sternberg if (result) 2099f58944e2SKeith Busch goto out; 210057dacad5SJay Sternberg 210157dacad5SJay Sternberg nvme_init_queue(dev->queues[0], 0); 210257dacad5SJay Sternberg result = nvme_alloc_admin_tags(dev); 210357dacad5SJay Sternberg if (result) 2104f58944e2SKeith Busch goto out; 210557dacad5SJay Sternberg 2106ce4541f4SChristoph Hellwig result = nvme_init_identify(&dev->ctrl); 2107ce4541f4SChristoph Hellwig if (result) 2108f58944e2SKeith Busch goto out; 2109ce4541f4SChristoph Hellwig 2110e286bcfcSScott Bauer if (dev->ctrl.oacs & NVME_CTRL_OACS_SEC_SUPP) { 2111e286bcfcSScott Bauer if (!dev->ctrl.opal_dev) 21124f1244c8SChristoph Hellwig dev->ctrl.opal_dev = 21134f1244c8SChristoph Hellwig init_opal_dev(&dev->ctrl, &nvme_sec_submit); 2114e286bcfcSScott Bauer else if (was_suspend) 21154f1244c8SChristoph Hellwig opal_unlock_from_suspend(dev->ctrl.opal_dev); 2116e286bcfcSScott Bauer } else { 2117e286bcfcSScott Bauer free_opal_dev(dev->ctrl.opal_dev); 2118e286bcfcSScott Bauer dev->ctrl.opal_dev = NULL; 2119e286bcfcSScott Bauer } 2120a98e58e5SScott Bauer 2121f9f38e33SHelen Koike if (dev->ctrl.oacs & NVME_CTRL_OACS_DBBUF_SUPP) { 2122f9f38e33SHelen Koike result = nvme_dbbuf_dma_alloc(dev); 2123f9f38e33SHelen Koike if (result) 2124f9f38e33SHelen Koike dev_warn(dev->dev, 2125f9f38e33SHelen Koike "unable to allocate dma for dbbuf\n"); 2126f9f38e33SHelen Koike } 2127f9f38e33SHelen Koike 212887ad72a5SChristoph Hellwig if (dev->ctrl.hmpre) 212987ad72a5SChristoph Hellwig nvme_setup_host_mem(dev); 213087ad72a5SChristoph Hellwig 213157dacad5SJay Sternberg result = nvme_setup_io_queues(dev); 213257dacad5SJay Sternberg if (result) 2133f58944e2SKeith Busch goto out; 213457dacad5SJay Sternberg 213521f033f7SKeith Busch /* 213621f033f7SKeith Busch * A controller that can not execute IO typically requires user 213721f033f7SKeith Busch * intervention to correct. For such degraded controllers, the driver 213821f033f7SKeith Busch * should not submit commands the user did not request, so skip 213921f033f7SKeith Busch * registering for asynchronous event notification on this condition. 214021f033f7SKeith Busch */ 2141f866fc42SChristoph Hellwig if (dev->online_queues > 1) 2142f866fc42SChristoph Hellwig nvme_queue_async_events(&dev->ctrl); 214357dacad5SJay Sternberg 214457dacad5SJay Sternberg /* 214557dacad5SJay Sternberg * Keep the controller around but remove all namespaces if we don't have 214657dacad5SJay Sternberg * any working I/O queue. 214757dacad5SJay Sternberg */ 214857dacad5SJay Sternberg if (dev->online_queues < 2) { 21491b3c47c1SSagi Grimberg dev_warn(dev->ctrl.device, "IO queues not created\n"); 21503b24774eSKeith Busch nvme_kill_queues(&dev->ctrl); 21515bae7f73SChristoph Hellwig nvme_remove_namespaces(&dev->ctrl); 215257dacad5SJay Sternberg } else { 215325646264SKeith Busch nvme_start_queues(&dev->ctrl); 2154302ad8ccSKeith Busch nvme_wait_freeze(&dev->ctrl); 215557dacad5SJay Sternberg nvme_dev_add(dev); 2156302ad8ccSKeith Busch nvme_unfreeze(&dev->ctrl); 215757dacad5SJay Sternberg } 215857dacad5SJay Sternberg 2159bb8d261eSChristoph Hellwig if (!nvme_change_ctrl_state(&dev->ctrl, NVME_CTRL_LIVE)) { 2160bb8d261eSChristoph Hellwig dev_warn(dev->ctrl.device, "failed to mark controller live\n"); 2161bb8d261eSChristoph Hellwig goto out; 2162bb8d261eSChristoph Hellwig } 216392911a55SChristoph Hellwig 216492911a55SChristoph Hellwig if (dev->online_queues > 1) 21655955be21SChristoph Hellwig nvme_queue_scan(&dev->ctrl); 216657dacad5SJay Sternberg return; 216757dacad5SJay Sternberg 216857dacad5SJay Sternberg out: 2169f58944e2SKeith Busch nvme_remove_dead_ctrl(dev, result); 217057dacad5SJay Sternberg } 217157dacad5SJay Sternberg 21725c8809e6SChristoph Hellwig static void nvme_remove_dead_ctrl_work(struct work_struct *work) 217357dacad5SJay Sternberg { 21745c8809e6SChristoph Hellwig struct nvme_dev *dev = container_of(work, struct nvme_dev, remove_work); 217557dacad5SJay Sternberg struct pci_dev *pdev = to_pci_dev(dev->dev); 217657dacad5SJay Sternberg 217769d9a99cSKeith Busch nvme_kill_queues(&dev->ctrl); 217857dacad5SJay Sternberg if (pci_get_drvdata(pdev)) 2179921920abSKeith Busch device_release_driver(&pdev->dev); 21801673f1f0SChristoph Hellwig nvme_put_ctrl(&dev->ctrl); 218157dacad5SJay Sternberg } 218257dacad5SJay Sternberg 21831c63dc66SChristoph Hellwig static int nvme_pci_reg_read32(struct nvme_ctrl *ctrl, u32 off, u32 *val) 218457dacad5SJay Sternberg { 21851c63dc66SChristoph Hellwig *val = readl(to_nvme_dev(ctrl)->bar + off); 21861c63dc66SChristoph Hellwig return 0; 218757dacad5SJay Sternberg } 21881c63dc66SChristoph Hellwig 21895fd4ce1bSChristoph Hellwig static int nvme_pci_reg_write32(struct nvme_ctrl *ctrl, u32 off, u32 val) 21905fd4ce1bSChristoph Hellwig { 21915fd4ce1bSChristoph Hellwig writel(val, to_nvme_dev(ctrl)->bar + off); 21925fd4ce1bSChristoph Hellwig return 0; 21935fd4ce1bSChristoph Hellwig } 21945fd4ce1bSChristoph Hellwig 21957fd8930fSChristoph Hellwig static int nvme_pci_reg_read64(struct nvme_ctrl *ctrl, u32 off, u64 *val) 21967fd8930fSChristoph Hellwig { 21977fd8930fSChristoph Hellwig *val = readq(to_nvme_dev(ctrl)->bar + off); 21987fd8930fSChristoph Hellwig return 0; 21997fd8930fSChristoph Hellwig } 22007fd8930fSChristoph Hellwig 22011c63dc66SChristoph Hellwig static const struct nvme_ctrl_ops nvme_pci_ctrl_ops = { 22021a353d85SMing Lin .name = "pcie", 2203e439bb12SSagi Grimberg .module = THIS_MODULE, 2204c81bfba9SChristoph Hellwig .flags = NVME_F_METADATA_SUPPORTED, 22051c63dc66SChristoph Hellwig .reg_read32 = nvme_pci_reg_read32, 22065fd4ce1bSChristoph Hellwig .reg_write32 = nvme_pci_reg_write32, 22077fd8930fSChristoph Hellwig .reg_read64 = nvme_pci_reg_read64, 22081673f1f0SChristoph Hellwig .free_ctrl = nvme_pci_free_ctrl, 2209f866fc42SChristoph Hellwig .submit_async_event = nvme_pci_submit_async_event, 22101c63dc66SChristoph Hellwig }; 221157dacad5SJay Sternberg 2212b00a726aSKeith Busch static int nvme_dev_map(struct nvme_dev *dev) 2213b00a726aSKeith Busch { 2214b00a726aSKeith Busch struct pci_dev *pdev = to_pci_dev(dev->dev); 2215b00a726aSKeith Busch 2216a1f447b3SJohannes Thumshirn if (pci_request_mem_regions(pdev, "nvme")) 2217b00a726aSKeith Busch return -ENODEV; 2218b00a726aSKeith Busch 221997f6ef64SXu Yu if (nvme_remap_bar(dev, NVME_REG_DBS + 4096)) 2220b00a726aSKeith Busch goto release; 2221b00a726aSKeith Busch 2222b00a726aSKeith Busch return 0; 2223b00a726aSKeith Busch release: 2224a1f447b3SJohannes Thumshirn pci_release_mem_regions(pdev); 2225b00a726aSKeith Busch return -ENODEV; 2226b00a726aSKeith Busch } 2227b00a726aSKeith Busch 2228ff5350a8SAndy Lutomirski static unsigned long check_dell_samsung_bug(struct pci_dev *pdev) 2229ff5350a8SAndy Lutomirski { 2230ff5350a8SAndy Lutomirski if (pdev->vendor == 0x144d && pdev->device == 0xa802) { 2231ff5350a8SAndy Lutomirski /* 2232ff5350a8SAndy Lutomirski * Several Samsung devices seem to drop off the PCIe bus 2233ff5350a8SAndy Lutomirski * randomly when APST is on and uses the deepest sleep state. 2234ff5350a8SAndy Lutomirski * This has been observed on a Samsung "SM951 NVMe SAMSUNG 2235ff5350a8SAndy Lutomirski * 256GB", a "PM951 NVMe SAMSUNG 512GB", and a "Samsung SSD 2236ff5350a8SAndy Lutomirski * 950 PRO 256GB", but it seems to be restricted to two Dell 2237ff5350a8SAndy Lutomirski * laptops. 2238ff5350a8SAndy Lutomirski */ 2239ff5350a8SAndy Lutomirski if (dmi_match(DMI_SYS_VENDOR, "Dell Inc.") && 2240ff5350a8SAndy Lutomirski (dmi_match(DMI_PRODUCT_NAME, "XPS 15 9550") || 2241ff5350a8SAndy Lutomirski dmi_match(DMI_PRODUCT_NAME, "Precision 5510"))) 2242ff5350a8SAndy Lutomirski return NVME_QUIRK_NO_DEEPEST_PS; 2243ff5350a8SAndy Lutomirski } 2244ff5350a8SAndy Lutomirski 2245ff5350a8SAndy Lutomirski return 0; 2246ff5350a8SAndy Lutomirski } 2247ff5350a8SAndy Lutomirski 224857dacad5SJay Sternberg static int nvme_probe(struct pci_dev *pdev, const struct pci_device_id *id) 224957dacad5SJay Sternberg { 225057dacad5SJay Sternberg int node, result = -ENOMEM; 225157dacad5SJay Sternberg struct nvme_dev *dev; 2252ff5350a8SAndy Lutomirski unsigned long quirks = id->driver_data; 225357dacad5SJay Sternberg 225457dacad5SJay Sternberg node = dev_to_node(&pdev->dev); 225557dacad5SJay Sternberg if (node == NUMA_NO_NODE) 22562fa84351SMasayoshi Mizuma set_dev_node(&pdev->dev, first_memory_node); 225757dacad5SJay Sternberg 225857dacad5SJay Sternberg dev = kzalloc_node(sizeof(*dev), GFP_KERNEL, node); 225957dacad5SJay Sternberg if (!dev) 226057dacad5SJay Sternberg return -ENOMEM; 226157dacad5SJay Sternberg dev->queues = kzalloc_node((num_possible_cpus() + 1) * sizeof(void *), 226257dacad5SJay Sternberg GFP_KERNEL, node); 226357dacad5SJay Sternberg if (!dev->queues) 226457dacad5SJay Sternberg goto free; 226557dacad5SJay Sternberg 226657dacad5SJay Sternberg dev->dev = get_device(&pdev->dev); 226757dacad5SJay Sternberg pci_set_drvdata(pdev, dev); 226857dacad5SJay Sternberg 2269b00a726aSKeith Busch result = nvme_dev_map(dev); 2270b00a726aSKeith Busch if (result) 2271b00a726aSKeith Busch goto free; 2272b00a726aSKeith Busch 2273d86c4d8eSChristoph Hellwig INIT_WORK(&dev->ctrl.reset_work, nvme_reset_work); 22745c8809e6SChristoph Hellwig INIT_WORK(&dev->remove_work, nvme_remove_dead_ctrl_work); 227577bf25eaSKeith Busch mutex_init(&dev->shutdown_lock); 2276db3cbfffSKeith Busch init_completion(&dev->ioq_wait); 2277f3ca80fcSChristoph Hellwig 2278f3ca80fcSChristoph Hellwig result = nvme_setup_prp_pools(dev); 2279f3ca80fcSChristoph Hellwig if (result) 2280f3ca80fcSChristoph Hellwig goto put_pci; 2281f3ca80fcSChristoph Hellwig 2282ff5350a8SAndy Lutomirski quirks |= check_dell_samsung_bug(pdev); 2283ff5350a8SAndy Lutomirski 2284f3ca80fcSChristoph Hellwig result = nvme_init_ctrl(&dev->ctrl, &pdev->dev, &nvme_pci_ctrl_ops, 2285ff5350a8SAndy Lutomirski quirks); 2286f3ca80fcSChristoph Hellwig if (result) 2287f3ca80fcSChristoph Hellwig goto release_pools; 2288f3ca80fcSChristoph Hellwig 228982b057caSRakesh Pandit nvme_change_ctrl_state(&dev->ctrl, NVME_CTRL_RESETTING); 22901b3c47c1SSagi Grimberg dev_info(dev->ctrl.device, "pci function %s\n", dev_name(&pdev->dev)); 22911b3c47c1SSagi Grimberg 2292d86c4d8eSChristoph Hellwig queue_work(nvme_wq, &dev->ctrl.reset_work); 229357dacad5SJay Sternberg return 0; 229457dacad5SJay Sternberg 229557dacad5SJay Sternberg release_pools: 229657dacad5SJay Sternberg nvme_release_prp_pools(dev); 229757dacad5SJay Sternberg put_pci: 229857dacad5SJay Sternberg put_device(dev->dev); 2299b00a726aSKeith Busch nvme_dev_unmap(dev); 230057dacad5SJay Sternberg free: 230157dacad5SJay Sternberg kfree(dev->queues); 230257dacad5SJay Sternberg kfree(dev); 230357dacad5SJay Sternberg return result; 230457dacad5SJay Sternberg } 230557dacad5SJay Sternberg 230657dacad5SJay Sternberg static void nvme_reset_notify(struct pci_dev *pdev, bool prepare) 230757dacad5SJay Sternberg { 230857dacad5SJay Sternberg struct nvme_dev *dev = pci_get_drvdata(pdev); 230957dacad5SJay Sternberg 231057dacad5SJay Sternberg if (prepare) 2311a5cdb68cSKeith Busch nvme_dev_disable(dev, false); 231257dacad5SJay Sternberg else 2313d86c4d8eSChristoph Hellwig nvme_reset_ctrl(&dev->ctrl); 231457dacad5SJay Sternberg } 231557dacad5SJay Sternberg 231657dacad5SJay Sternberg static void nvme_shutdown(struct pci_dev *pdev) 231757dacad5SJay Sternberg { 231857dacad5SJay Sternberg struct nvme_dev *dev = pci_get_drvdata(pdev); 2319a5cdb68cSKeith Busch nvme_dev_disable(dev, true); 232057dacad5SJay Sternberg } 232157dacad5SJay Sternberg 2322f58944e2SKeith Busch /* 2323f58944e2SKeith Busch * The driver's remove may be called on a device in a partially initialized 2324f58944e2SKeith Busch * state. This function must not have any dependencies on the device state in 2325f58944e2SKeith Busch * order to proceed. 2326f58944e2SKeith Busch */ 232757dacad5SJay Sternberg static void nvme_remove(struct pci_dev *pdev) 232857dacad5SJay Sternberg { 232957dacad5SJay Sternberg struct nvme_dev *dev = pci_get_drvdata(pdev); 233057dacad5SJay Sternberg 2331bb8d261eSChristoph Hellwig nvme_change_ctrl_state(&dev->ctrl, NVME_CTRL_DELETING); 2332bb8d261eSChristoph Hellwig 2333d86c4d8eSChristoph Hellwig cancel_work_sync(&dev->ctrl.reset_work); 233457dacad5SJay Sternberg pci_set_drvdata(pdev, NULL); 23350ff9d4e1SKeith Busch 23366db28edaSKeith Busch if (!pci_device_is_present(pdev)) { 23370ff9d4e1SKeith Busch nvme_change_ctrl_state(&dev->ctrl, NVME_CTRL_DEAD); 23386db28edaSKeith Busch nvme_dev_disable(dev, false); 23396db28edaSKeith Busch } 23400ff9d4e1SKeith Busch 2341d86c4d8eSChristoph Hellwig flush_work(&dev->ctrl.reset_work); 234253029b04SKeith Busch nvme_uninit_ctrl(&dev->ctrl); 2343a5cdb68cSKeith Busch nvme_dev_disable(dev, true); 234487ad72a5SChristoph Hellwig nvme_free_host_mem(dev); 234557dacad5SJay Sternberg nvme_dev_remove_admin(dev); 234657dacad5SJay Sternberg nvme_free_queues(dev, 0); 234757dacad5SJay Sternberg nvme_release_prp_pools(dev); 2348b00a726aSKeith Busch nvme_dev_unmap(dev); 23491673f1f0SChristoph Hellwig nvme_put_ctrl(&dev->ctrl); 235057dacad5SJay Sternberg } 235157dacad5SJay Sternberg 235213880f5bSKeith Busch static int nvme_pci_sriov_configure(struct pci_dev *pdev, int numvfs) 235313880f5bSKeith Busch { 235413880f5bSKeith Busch int ret = 0; 235513880f5bSKeith Busch 235613880f5bSKeith Busch if (numvfs == 0) { 235713880f5bSKeith Busch if (pci_vfs_assigned(pdev)) { 235813880f5bSKeith Busch dev_warn(&pdev->dev, 235913880f5bSKeith Busch "Cannot disable SR-IOV VFs while assigned\n"); 236013880f5bSKeith Busch return -EPERM; 236113880f5bSKeith Busch } 236213880f5bSKeith Busch pci_disable_sriov(pdev); 236313880f5bSKeith Busch return 0; 236413880f5bSKeith Busch } 236513880f5bSKeith Busch 236613880f5bSKeith Busch ret = pci_enable_sriov(pdev, numvfs); 236713880f5bSKeith Busch return ret ? ret : numvfs; 236813880f5bSKeith Busch } 236913880f5bSKeith Busch 237057dacad5SJay Sternberg #ifdef CONFIG_PM_SLEEP 237157dacad5SJay Sternberg static int nvme_suspend(struct device *dev) 237257dacad5SJay Sternberg { 237357dacad5SJay Sternberg struct pci_dev *pdev = to_pci_dev(dev); 237457dacad5SJay Sternberg struct nvme_dev *ndev = pci_get_drvdata(pdev); 237557dacad5SJay Sternberg 2376a5cdb68cSKeith Busch nvme_dev_disable(ndev, true); 237757dacad5SJay Sternberg return 0; 237857dacad5SJay Sternberg } 237957dacad5SJay Sternberg 238057dacad5SJay Sternberg static int nvme_resume(struct device *dev) 238157dacad5SJay Sternberg { 238257dacad5SJay Sternberg struct pci_dev *pdev = to_pci_dev(dev); 238357dacad5SJay Sternberg struct nvme_dev *ndev = pci_get_drvdata(pdev); 238457dacad5SJay Sternberg 2385d86c4d8eSChristoph Hellwig nvme_reset_ctrl(&ndev->ctrl); 238657dacad5SJay Sternberg return 0; 238757dacad5SJay Sternberg } 238857dacad5SJay Sternberg #endif 238957dacad5SJay Sternberg 239057dacad5SJay Sternberg static SIMPLE_DEV_PM_OPS(nvme_dev_pm_ops, nvme_suspend, nvme_resume); 239157dacad5SJay Sternberg 2392a0a3408eSKeith Busch static pci_ers_result_t nvme_error_detected(struct pci_dev *pdev, 2393a0a3408eSKeith Busch pci_channel_state_t state) 2394a0a3408eSKeith Busch { 2395a0a3408eSKeith Busch struct nvme_dev *dev = pci_get_drvdata(pdev); 2396a0a3408eSKeith Busch 2397a0a3408eSKeith Busch /* 2398a0a3408eSKeith Busch * A frozen channel requires a reset. When detected, this method will 2399a0a3408eSKeith Busch * shutdown the controller to quiesce. The controller will be restarted 2400a0a3408eSKeith Busch * after the slot reset through driver's slot_reset callback. 2401a0a3408eSKeith Busch */ 2402a0a3408eSKeith Busch switch (state) { 2403a0a3408eSKeith Busch case pci_channel_io_normal: 2404a0a3408eSKeith Busch return PCI_ERS_RESULT_CAN_RECOVER; 2405a0a3408eSKeith Busch case pci_channel_io_frozen: 2406d011fb31SKeith Busch dev_warn(dev->ctrl.device, 2407d011fb31SKeith Busch "frozen state error detected, reset controller\n"); 2408a5cdb68cSKeith Busch nvme_dev_disable(dev, false); 2409a0a3408eSKeith Busch return PCI_ERS_RESULT_NEED_RESET; 2410a0a3408eSKeith Busch case pci_channel_io_perm_failure: 2411d011fb31SKeith Busch dev_warn(dev->ctrl.device, 2412d011fb31SKeith Busch "failure state error detected, request disconnect\n"); 2413a0a3408eSKeith Busch return PCI_ERS_RESULT_DISCONNECT; 2414a0a3408eSKeith Busch } 2415a0a3408eSKeith Busch return PCI_ERS_RESULT_NEED_RESET; 2416a0a3408eSKeith Busch } 2417a0a3408eSKeith Busch 2418a0a3408eSKeith Busch static pci_ers_result_t nvme_slot_reset(struct pci_dev *pdev) 2419a0a3408eSKeith Busch { 2420a0a3408eSKeith Busch struct nvme_dev *dev = pci_get_drvdata(pdev); 2421a0a3408eSKeith Busch 24221b3c47c1SSagi Grimberg dev_info(dev->ctrl.device, "restart after slot reset\n"); 2423a0a3408eSKeith Busch pci_restore_state(pdev); 2424d86c4d8eSChristoph Hellwig nvme_reset_ctrl(&dev->ctrl); 2425a0a3408eSKeith Busch return PCI_ERS_RESULT_RECOVERED; 2426a0a3408eSKeith Busch } 2427a0a3408eSKeith Busch 2428a0a3408eSKeith Busch static void nvme_error_resume(struct pci_dev *pdev) 2429a0a3408eSKeith Busch { 2430a0a3408eSKeith Busch pci_cleanup_aer_uncorrect_error_status(pdev); 2431a0a3408eSKeith Busch } 2432a0a3408eSKeith Busch 243357dacad5SJay Sternberg static const struct pci_error_handlers nvme_err_handler = { 243457dacad5SJay Sternberg .error_detected = nvme_error_detected, 243557dacad5SJay Sternberg .slot_reset = nvme_slot_reset, 243657dacad5SJay Sternberg .resume = nvme_error_resume, 243757dacad5SJay Sternberg .reset_notify = nvme_reset_notify, 243857dacad5SJay Sternberg }; 243957dacad5SJay Sternberg 244057dacad5SJay Sternberg static const struct pci_device_id nvme_id_table[] = { 2441106198edSChristoph Hellwig { PCI_VDEVICE(INTEL, 0x0953), 244208095e70SKeith Busch .driver_data = NVME_QUIRK_STRIPE_SIZE | 2443e850fd16SChristoph Hellwig NVME_QUIRK_DEALLOCATE_ZEROES, }, 244499466e70SKeith Busch { PCI_VDEVICE(INTEL, 0x0a53), 244599466e70SKeith Busch .driver_data = NVME_QUIRK_STRIPE_SIZE | 2446e850fd16SChristoph Hellwig NVME_QUIRK_DEALLOCATE_ZEROES, }, 244799466e70SKeith Busch { PCI_VDEVICE(INTEL, 0x0a54), 244899466e70SKeith Busch .driver_data = NVME_QUIRK_STRIPE_SIZE | 2449e850fd16SChristoph Hellwig NVME_QUIRK_DEALLOCATE_ZEROES, }, 245050af47d0SAndy Lutomirski { PCI_VDEVICE(INTEL, 0xf1a5), /* Intel 600P/P3100 */ 245150af47d0SAndy Lutomirski .driver_data = NVME_QUIRK_NO_DEEPEST_PS }, 2452540c801cSKeith Busch { PCI_VDEVICE(INTEL, 0x5845), /* Qemu emulated controller */ 2453540c801cSKeith Busch .driver_data = NVME_QUIRK_IDENTIFY_CNS, }, 245454adc010SGuilherme G. Piccoli { PCI_DEVICE(0x1c58, 0x0003), /* HGST adapter */ 245554adc010SGuilherme G. Piccoli .driver_data = NVME_QUIRK_DELAY_BEFORE_CHK_RDY, }, 2456015282c9SWenbo Wang { PCI_DEVICE(0x1c5f, 0x0540), /* Memblaze Pblaze4 adapter */ 2457015282c9SWenbo Wang .driver_data = NVME_QUIRK_DELAY_BEFORE_CHK_RDY, }, 245857dacad5SJay Sternberg { PCI_DEVICE_CLASS(PCI_CLASS_STORAGE_EXPRESS, 0xffffff) }, 2459c74dc780SStephan Günther { PCI_DEVICE(PCI_VENDOR_ID_APPLE, 0x2001) }, 2460124298bdSDaniel Roschka { PCI_DEVICE(PCI_VENDOR_ID_APPLE, 0x2003) }, 246157dacad5SJay Sternberg { 0, } 246257dacad5SJay Sternberg }; 246357dacad5SJay Sternberg MODULE_DEVICE_TABLE(pci, nvme_id_table); 246457dacad5SJay Sternberg 246557dacad5SJay Sternberg static struct pci_driver nvme_driver = { 246657dacad5SJay Sternberg .name = "nvme", 246757dacad5SJay Sternberg .id_table = nvme_id_table, 246857dacad5SJay Sternberg .probe = nvme_probe, 246957dacad5SJay Sternberg .remove = nvme_remove, 247057dacad5SJay Sternberg .shutdown = nvme_shutdown, 247157dacad5SJay Sternberg .driver = { 247257dacad5SJay Sternberg .pm = &nvme_dev_pm_ops, 247357dacad5SJay Sternberg }, 247413880f5bSKeith Busch .sriov_configure = nvme_pci_sriov_configure, 247557dacad5SJay Sternberg .err_handler = &nvme_err_handler, 247657dacad5SJay Sternberg }; 247757dacad5SJay Sternberg 247857dacad5SJay Sternberg static int __init nvme_init(void) 247957dacad5SJay Sternberg { 24809a6327d2SSagi Grimberg return pci_register_driver(&nvme_driver); 248157dacad5SJay Sternberg } 248257dacad5SJay Sternberg 248357dacad5SJay Sternberg static void __exit nvme_exit(void) 248457dacad5SJay Sternberg { 248557dacad5SJay Sternberg pci_unregister_driver(&nvme_driver); 248657dacad5SJay Sternberg _nvme_check_size(); 248757dacad5SJay Sternberg } 248857dacad5SJay Sternberg 248957dacad5SJay Sternberg MODULE_AUTHOR("Matthew Wilcox <willy@linux.intel.com>"); 249057dacad5SJay Sternberg MODULE_LICENSE("GPL"); 249157dacad5SJay Sternberg MODULE_VERSION("1.0"); 249257dacad5SJay Sternberg module_init(nvme_init); 249357dacad5SJay Sternberg module_exit(nvme_exit); 2494