diff --git a/sys/dev/nvme/nvme.h b/sys/dev/nvme/nvme.h index 315dbdd5a3c8..cb5e1a5f1118 100644 --- a/sys/dev/nvme/nvme.h +++ b/sys/dev/nvme/nvme.h @@ -1,2127 +1,2133 @@ /*- * SPDX-License-Identifier: BSD-2-Clause * * Copyright (C) 2012-2013 Intel Corporation * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ #ifndef __NVME_H__ #define __NVME_H__ #ifdef _KERNEL #include #endif #include #include #define NVME_PASSTHROUGH_CMD _IOWR('n', 0, struct nvme_pt_command) #define NVME_RESET_CONTROLLER _IO('n', 1) #define NVME_GET_NSID _IOR('n', 2, struct nvme_get_nsid) #define NVME_GET_MAX_XFER_SIZE _IOR('n', 3, uint64_t) #define NVME_IO_TEST _IOWR('n', 100, struct nvme_io_test) #define NVME_BIO_TEST _IOWR('n', 101, struct nvme_io_test) /* * Macros to deal with NVME revisions, as defined VS register */ #define NVME_REV(x, y) (((x) << 16) | ((y) << 8)) #define NVME_MAJOR(r) (((r) >> 16) & 0xffff) #define NVME_MINOR(r) (((r) >> 8) & 0xff) /* * Use to mark a command to apply to all namespaces, or to retrieve global * log pages. */ #define NVME_GLOBAL_NAMESPACE_TAG ((uint32_t)0xFFFFFFFF) /* Host memory buffer sizes are always in 4096 byte chunks */ #define NVME_HMB_UNITS 4096 /* Many items are expressed in terms of power of two times MPS */ #define NVME_MPS_SHIFT 12 /* Register field definitions */ #define NVME_CAP_LO_REG_MQES_SHIFT (0) #define NVME_CAP_LO_REG_MQES_MASK (0xFFFF) #define NVME_CAP_LO_REG_CQR_SHIFT (16) #define NVME_CAP_LO_REG_CQR_MASK (0x1) #define NVME_CAP_LO_REG_AMS_SHIFT (17) #define NVME_CAP_LO_REG_AMS_MASK (0x3) #define NVME_CAP_LO_REG_TO_SHIFT (24) #define NVME_CAP_LO_REG_TO_MASK (0xFF) #define NVME_CAP_LO_MQES(x) \ - (((x) >> NVME_CAP_LO_REG_MQES_SHIFT) & NVME_CAP_LO_REG_MQES_MASK) + NVMEV(NVME_CAP_LO_REG_MQES, x) #define NVME_CAP_LO_CQR(x) \ - (((x) >> NVME_CAP_LO_REG_CQR_SHIFT) & NVME_CAP_LO_REG_CQR_MASK) + NVMEV(NVME_CAP_LO_REG_CQR, x) #define NVME_CAP_LO_AMS(x) \ - (((x) >> NVME_CAP_LO_REG_AMS_SHIFT) & NVME_CAP_LO_REG_AMS_MASK) + NVMEV(NVME_CAP_LO_REG_AMS, x) #define NVME_CAP_LO_TO(x) \ - (((x) >> NVME_CAP_LO_REG_TO_SHIFT) & NVME_CAP_LO_REG_TO_MASK) + NVMEV(NVME_CAP_LO_REG_TO, x) #define NVME_CAP_HI_REG_DSTRD_SHIFT (0) #define NVME_CAP_HI_REG_DSTRD_MASK (0xF) #define NVME_CAP_HI_REG_NSSRS_SHIFT (4) #define NVME_CAP_HI_REG_NSSRS_MASK (0x1) #define NVME_CAP_HI_REG_CSS_SHIFT (5) #define NVME_CAP_HI_REG_CSS_MASK (0xff) #define NVME_CAP_HI_REG_CSS_NVM_SHIFT (5) #define NVME_CAP_HI_REG_CSS_NVM_MASK (0x1) #define NVME_CAP_HI_REG_BPS_SHIFT (13) #define NVME_CAP_HI_REG_BPS_MASK (0x1) #define NVME_CAP_HI_REG_CPS_SHIFT (14) #define NVME_CAP_HI_REG_CPS_MASK (0x3) #define NVME_CAP_HI_REG_MPSMIN_SHIFT (16) #define NVME_CAP_HI_REG_MPSMIN_MASK (0xF) #define NVME_CAP_HI_REG_MPSMAX_SHIFT (20) #define NVME_CAP_HI_REG_MPSMAX_MASK (0xF) #define NVME_CAP_HI_REG_PMRS_SHIFT (24) #define NVME_CAP_HI_REG_PMRS_MASK (0x1) #define NVME_CAP_HI_REG_CMBS_SHIFT (25) #define NVME_CAP_HI_REG_CMBS_MASK (0x1) #define NVME_CAP_HI_REG_NSSS_SHIFT (26) #define NVME_CAP_HI_REG_NSSS_MASK (0x1) #define NVME_CAP_HI_REG_CRWMS_SHIFT (27) #define NVME_CAP_HI_REG_CRWMS_MASK (0x1) #define NVME_CAP_HI_REG_CRIMS_SHIFT (28) #define NVME_CAP_HI_REG_CRIMS_MASK (0x1) #define NVME_CAP_HI_DSTRD(x) \ - (((x) >> NVME_CAP_HI_REG_DSTRD_SHIFT) & NVME_CAP_HI_REG_DSTRD_MASK) + NVMEV(NVME_CAP_HI_REG_DSTRD, x) #define NVME_CAP_HI_NSSRS(x) \ - (((x) >> NVME_CAP_HI_REG_NSSRS_SHIFT) & NVME_CAP_HI_REG_NSSRS_MASK) + NVMEV(NVME_CAP_HI_REG_NSSRS, x) #define NVME_CAP_HI_CSS(x) \ - (((x) >> NVME_CAP_HI_REG_CSS_SHIFT) & NVME_CAP_HI_REG_CSS_MASK) + NVMEV(NVME_CAP_HI_REG_CSS, x) #define NVME_CAP_HI_CSS_NVM(x) \ - (((x) >> NVME_CAP_HI_REG_CSS_NVM_SHIFT) & NVME_CAP_HI_REG_CSS_NVM_MASK) + NVMEV(NVME_CAP_HI_REG_CSS_NVM, x) #define NVME_CAP_HI_BPS(x) \ - (((x) >> NVME_CAP_HI_REG_BPS_SHIFT) & NVME_CAP_HI_REG_BPS_MASK) + NVMEV(NVME_CAP_HI_REG_BPS, x) #define NVME_CAP_HI_CPS(x) \ - (((x) >> NVME_CAP_HI_REG_CPS_SHIFT) & NVME_CAP_HI_REG_CPS_MASK) + NVMEV(NVME_CAP_HI_REG_CPS, x) #define NVME_CAP_HI_MPSMIN(x) \ - (((x) >> NVME_CAP_HI_REG_MPSMIN_SHIFT) & NVME_CAP_HI_REG_MPSMIN_MASK) + NVMEV(NVME_CAP_HI_REG_MPSMIN, x) #define NVME_CAP_HI_MPSMAX(x) \ - (((x) >> NVME_CAP_HI_REG_MPSMAX_SHIFT) & NVME_CAP_HI_REG_MPSMAX_MASK) + NVMEV(NVME_CAP_HI_REG_MPSMAX, x) #define NVME_CAP_HI_PMRS(x) \ - (((x) >> NVME_CAP_HI_REG_PMRS_SHIFT) & NVME_CAP_HI_REG_PMRS_MASK) + NVMEV(NVME_CAP_HI_REG_PMRS, x) #define NVME_CAP_HI_CMBS(x) \ - (((x) >> NVME_CAP_HI_REG_CMBS_SHIFT) & NVME_CAP_HI_REG_CMBS_MASK) + NVMEV(NVME_CAP_HI_REG_CMBS, x) #define NVME_CAP_HI_NSSS(x) \ - (((x) >> NVME_CAP_HI_REG_NSSS_SHIFT) & NVME_CAP_HI_REG_NSSS_MASK) + NVMEV(NVME_CAP_HI_REG_NSSS, x) #define NVME_CAP_HI_CRWMS(x) \ - (((x) >> NVME_CAP_HI_REG_CRWMS_SHIFT) & NVME_CAP_HI_REG_CRWMS_MASK) + NVMEV(NVME_CAP_HI_REG_CRWMS, x) #define NVME_CAP_HI_CRIMS(x) \ - (((x) >> NVME_CAP_HI_REG_CRIMS_SHIFT) & NVME_CAP_HI_REG_CRIMS_MASK) + NVMEV(NVME_CAP_HI_REG_CRIMS, x) #define NVME_CC_REG_EN_SHIFT (0) #define NVME_CC_REG_EN_MASK (0x1) #define NVME_CC_REG_CSS_SHIFT (4) #define NVME_CC_REG_CSS_MASK (0x7) #define NVME_CC_REG_MPS_SHIFT (7) #define NVME_CC_REG_MPS_MASK (0xF) #define NVME_CC_REG_AMS_SHIFT (11) #define NVME_CC_REG_AMS_MASK (0x7) #define NVME_CC_REG_SHN_SHIFT (14) #define NVME_CC_REG_SHN_MASK (0x3) #define NVME_CC_REG_IOSQES_SHIFT (16) #define NVME_CC_REG_IOSQES_MASK (0xF) #define NVME_CC_REG_IOCQES_SHIFT (20) #define NVME_CC_REG_IOCQES_MASK (0xF) #define NVME_CC_REG_CRIME_SHIFT (24) #define NVME_CC_REG_CRIME_MASK (0x1) #define NVME_CSTS_REG_RDY_SHIFT (0) #define NVME_CSTS_REG_RDY_MASK (0x1) #define NVME_CSTS_REG_CFS_SHIFT (1) #define NVME_CSTS_REG_CFS_MASK (0x1) #define NVME_CSTS_REG_SHST_SHIFT (2) #define NVME_CSTS_REG_SHST_MASK (0x3) #define NVME_CSTS_REG_NVSRO_SHIFT (4) #define NVME_CSTS_REG_NVSRO_MASK (0x1) #define NVME_CSTS_REG_PP_SHIFT (5) #define NVME_CSTS_REG_PP_MASK (0x1) #define NVME_CSTS_REG_ST_SHIFT (6) #define NVME_CSTS_REG_ST_MASK (0x1) -#define NVME_CSTS_GET_SHST(csts) (((csts) >> NVME_CSTS_REG_SHST_SHIFT) & NVME_CSTS_REG_SHST_MASK) +#define NVME_CSTS_GET_SHST(csts) \ + NVMEV(NVME_CSTS_REG_SHST, csts) #define NVME_AQA_REG_ASQS_SHIFT (0) #define NVME_AQA_REG_ASQS_MASK (0xFFF) #define NVME_AQA_REG_ACQS_SHIFT (16) #define NVME_AQA_REG_ACQS_MASK (0xFFF) #define NVME_PMRCAP_REG_RDS_SHIFT (3) #define NVME_PMRCAP_REG_RDS_MASK (0x1) #define NVME_PMRCAP_REG_WDS_SHIFT (4) #define NVME_PMRCAP_REG_WDS_MASK (0x1) #define NVME_PMRCAP_REG_BIR_SHIFT (5) #define NVME_PMRCAP_REG_BIR_MASK (0x7) #define NVME_PMRCAP_REG_PMRTU_SHIFT (8) #define NVME_PMRCAP_REG_PMRTU_MASK (0x3) #define NVME_PMRCAP_REG_PMRWBM_SHIFT (10) #define NVME_PMRCAP_REG_PMRWBM_MASK (0xf) #define NVME_PMRCAP_REG_PMRTO_SHIFT (16) #define NVME_PMRCAP_REG_PMRTO_MASK (0xff) #define NVME_PMRCAP_REG_CMSS_SHIFT (24) #define NVME_PMRCAP_REG_CMSS_MASK (0x1) #define NVME_PMRCAP_RDS(x) \ - (((x) >> NVME_PMRCAP_REG_RDS_SHIFT) & NVME_PMRCAP_REG_RDS_MASK) + NVMEV(NVME_PMRCAP_REG_RDS, x) #define NVME_PMRCAP_WDS(x) \ - (((x) >> NVME_PMRCAP_REG_WDS_SHIFT) & NVME_PMRCAP_REG_WDS_MASK) + NVMEV(NVME_PMRCAP_REG_WDS, x) #define NVME_PMRCAP_BIR(x) \ - (((x) >> NVME_PMRCAP_REG_BIR_SHIFT) & NVME_PMRCAP_REG_BIR_MASK) + NVMEV(NVME_PMRCAP_REG_BIR, x) #define NVME_PMRCAP_PMRTU(x) \ - (((x) >> NVME_PMRCAP_REG_PMRTU_SHIFT) & NVME_PMRCAP_REG_PMRTU_MASK) + NVMEV(NVME_PMRCAP_REG_PMRTU, x) #define NVME_PMRCAP_PMRWBM(x) \ - (((x) >> NVME_PMRCAP_REG_PMRWBM_SHIFT) & NVME_PMRCAP_REG_PMRWBM_MASK) + NVMEV(NVME_PMRCAP_REG_PMRWBM, x) #define NVME_PMRCAP_PMRTO(x) \ - (((x) >> NVME_PMRCAP_REG_PMRTO_SHIFT) & NVME_PMRCAP_REG_PMRTO_MASK) + NVMEV(NVME_PMRCAP_REG_PMRTO, x) #define NVME_PMRCAP_CMSS(x) \ - (((x) >> NVME_PMRCAP_REG_CMSS_SHIFT) & NVME_PMRCAP_REG_CMSS_MASK) + NVMEV(NVME_PMRCAP_REG_CMSS, x) /* Command field definitions */ #define NVME_CMD_FUSE_SHIFT (8) #define NVME_CMD_FUSE_MASK (0x3) #define NVME_STATUS_P_SHIFT (0) #define NVME_STATUS_P_MASK (0x1) #define NVME_STATUS_SC_SHIFT (1) #define NVME_STATUS_SC_MASK (0xFF) #define NVME_STATUS_SCT_SHIFT (9) #define NVME_STATUS_SCT_MASK (0x7) #define NVME_STATUS_CRD_SHIFT (12) #define NVME_STATUS_CRD_MASK (0x3) #define NVME_STATUS_M_SHIFT (14) #define NVME_STATUS_M_MASK (0x1) #define NVME_STATUS_DNR_SHIFT (15) #define NVME_STATUS_DNR_MASK (0x1) -#define NVME_STATUS_GET_P(st) (((st) >> NVME_STATUS_P_SHIFT) & NVME_STATUS_P_MASK) -#define NVME_STATUS_GET_SC(st) (((st) >> NVME_STATUS_SC_SHIFT) & NVME_STATUS_SC_MASK) -#define NVME_STATUS_GET_SCT(st) (((st) >> NVME_STATUS_SCT_SHIFT) & NVME_STATUS_SCT_MASK) -#define NVME_STATUS_GET_CRD(st) (((st) >> NVME_STATUS_CRD_SHIFT) & NVME_STATUS_CRD_MASK) -#define NVME_STATUS_GET_M(st) (((st) >> NVME_STATUS_M_SHIFT) & NVME_STATUS_M_MASK) -#define NVME_STATUS_GET_DNR(st) (((st) >> NVME_STATUS_DNR_SHIFT) & NVME_STATUS_DNR_MASK) +#define NVME_STATUS_GET_P(st) \ + NVMEV(NVME_STATUS_P, st) +#define NVME_STATUS_GET_SC(st) \ + NVMEV(NVME_STATUS_SC, st) +#define NVME_STATUS_GET_SCT(st) \ + NVMEV(NVME_STATUS_SCT, st) +#define NVME_STATUS_GET_CRD(st) \ + NVMEV(NVME_STATUS_CRD, st) +#define NVME_STATUS_GET_M(st) \ + NVMEV(NVME_STATUS_M, st) +#define NVME_STATUS_GET_DNR(st) \ + NVMEV(NVME_STATUS_DNR, st) #define NVME_PWR_ST_MPS_SHIFT (0) #define NVME_PWR_ST_MPS_MASK (0x1) #define NVME_PWR_ST_NOPS_SHIFT (1) #define NVME_PWR_ST_NOPS_MASK (0x1) #define NVME_PWR_ST_RRT_SHIFT (0) #define NVME_PWR_ST_RRT_MASK (0x1F) #define NVME_PWR_ST_RRL_SHIFT (0) #define NVME_PWR_ST_RRL_MASK (0x1F) #define NVME_PWR_ST_RWT_SHIFT (0) #define NVME_PWR_ST_RWT_MASK (0x1F) #define NVME_PWR_ST_RWL_SHIFT (0) #define NVME_PWR_ST_RWL_MASK (0x1F) #define NVME_PWR_ST_IPS_SHIFT (6) #define NVME_PWR_ST_IPS_MASK (0x3) #define NVME_PWR_ST_APW_SHIFT (0) #define NVME_PWR_ST_APW_MASK (0x7) #define NVME_PWR_ST_APS_SHIFT (6) #define NVME_PWR_ST_APS_MASK (0x3) /** Controller Multi-path I/O and Namespace Sharing Capabilities */ /* More then one port */ #define NVME_CTRLR_DATA_MIC_MPORTS_SHIFT (0) #define NVME_CTRLR_DATA_MIC_MPORTS_MASK (0x1) /* More then one controller */ #define NVME_CTRLR_DATA_MIC_MCTRLRS_SHIFT (1) #define NVME_CTRLR_DATA_MIC_MCTRLRS_MASK (0x1) /* SR-IOV Virtual Function */ #define NVME_CTRLR_DATA_MIC_SRIOVVF_SHIFT (2) #define NVME_CTRLR_DATA_MIC_SRIOVVF_MASK (0x1) /* Asymmetric Namespace Access Reporting */ #define NVME_CTRLR_DATA_MIC_ANAR_SHIFT (3) #define NVME_CTRLR_DATA_MIC_ANAR_MASK (0x1) /** OAES - Optional Asynchronous Events Supported */ /* supports Namespace Attribute Notices event */ #define NVME_CTRLR_DATA_OAES_NS_ATTR_SHIFT (8) #define NVME_CTRLR_DATA_OAES_NS_ATTR_MASK (0x1) /* supports Firmware Activation Notices event */ #define NVME_CTRLR_DATA_OAES_FW_ACTIVATE_SHIFT (9) #define NVME_CTRLR_DATA_OAES_FW_ACTIVATE_MASK (0x1) /* supports Asymmetric Namespace Access Change Notices event */ #define NVME_CTRLR_DATA_OAES_ASYM_NS_CHANGE_SHIFT (11) #define NVME_CTRLR_DATA_OAES_ASYM_NS_CHANGE_MASK (0x1) /* supports Predictable Latency Event Aggregate Log Change Notices event */ #define NVME_CTRLR_DATA_OAES_PREDICT_LATENCY_SHIFT (12) #define NVME_CTRLR_DATA_OAES_PREDICT_LATENCY_MASK (0x1) /* supports LBA Status Information Notices event */ #define NVME_CTRLR_DATA_OAES_LBA_STATUS_SHIFT (13) #define NVME_CTRLR_DATA_OAES_LBA_STATUS_MASK (0x1) /* supports Endurance Group Event Aggregate Log Page Changes Notices event */ #define NVME_CTRLR_DATA_OAES_ENDURANCE_GROUP_SHIFT (14) #define NVME_CTRLR_DATA_OAES_ENDURANCE_GROUP_MASK (0x1) /* supports Normal NVM Subsystem Shutdown event */ #define NVME_CTRLR_DATA_OAES_NORMAL_SHUTDOWN_SHIFT (15) #define NVME_CTRLR_DATA_OAES_NORMAL_SHUTDOWN_MASK (0x1) /* supports Zone Descriptor Changed Notices event */ #define NVME_CTRLR_DATA_OAES_ZONE_DESC_CHANGE_SHIFT (27) #define NVME_CTRLR_DATA_OAES_ZONE_DESC_CHANGE_MASK (0x1) /* supports Discovery Log Page Change Notification event */ #define NVME_CTRLR_DATA_OAES_LOG_PAGE_CHANGE_SHIFT (31) #define NVME_CTRLR_DATA_OAES_LOG_PAGE_CHANGE_MASK (0x1) /** OACS - optional admin command support */ /* supports security send/receive commands */ #define NVME_CTRLR_DATA_OACS_SECURITY_SHIFT (0) #define NVME_CTRLR_DATA_OACS_SECURITY_MASK (0x1) /* supports format nvm command */ #define NVME_CTRLR_DATA_OACS_FORMAT_SHIFT (1) #define NVME_CTRLR_DATA_OACS_FORMAT_MASK (0x1) /* supports firmware activate/download commands */ #define NVME_CTRLR_DATA_OACS_FIRMWARE_SHIFT (2) #define NVME_CTRLR_DATA_OACS_FIRMWARE_MASK (0x1) /* supports namespace management commands */ #define NVME_CTRLR_DATA_OACS_NSMGMT_SHIFT (3) #define NVME_CTRLR_DATA_OACS_NSMGMT_MASK (0x1) /* supports Device Self-test command */ #define NVME_CTRLR_DATA_OACS_SELFTEST_SHIFT (4) #define NVME_CTRLR_DATA_OACS_SELFTEST_MASK (0x1) /* supports Directives */ #define NVME_CTRLR_DATA_OACS_DIRECTIVES_SHIFT (5) #define NVME_CTRLR_DATA_OACS_DIRECTIVES_MASK (0x1) /* supports NVMe-MI Send/Receive */ #define NVME_CTRLR_DATA_OACS_NVMEMI_SHIFT (6) #define NVME_CTRLR_DATA_OACS_NVMEMI_MASK (0x1) /* supports Virtualization Management */ #define NVME_CTRLR_DATA_OACS_VM_SHIFT (7) #define NVME_CTRLR_DATA_OACS_VM_MASK (0x1) /* supports Doorbell Buffer Config */ #define NVME_CTRLR_DATA_OACS_DBBUFFER_SHIFT (8) #define NVME_CTRLR_DATA_OACS_DBBUFFER_MASK (0x1) /* supports Get LBA Status */ #define NVME_CTRLR_DATA_OACS_GETLBA_SHIFT (9) #define NVME_CTRLR_DATA_OACS_GETLBA_MASK (0x1) /** firmware updates */ /* first slot is read-only */ #define NVME_CTRLR_DATA_FRMW_SLOT1_RO_SHIFT (0) #define NVME_CTRLR_DATA_FRMW_SLOT1_RO_MASK (0x1) /* number of firmware slots */ #define NVME_CTRLR_DATA_FRMW_NUM_SLOTS_SHIFT (1) #define NVME_CTRLR_DATA_FRMW_NUM_SLOTS_MASK (0x7) /* firmware activation without reset */ #define NVME_CTRLR_DATA_FRMW_ACT_WO_RESET_SHIFT (4) #define NVME_CTRLR_DATA_FRMW_ACT_WO_RESET_MASK (0x1) /** log page attributes */ /* per namespace smart/health log page */ #define NVME_CTRLR_DATA_LPA_NS_SMART_SHIFT (0) #define NVME_CTRLR_DATA_LPA_NS_SMART_MASK (0x1) /** AVSCC - admin vendor specific command configuration */ /* admin vendor specific commands use spec format */ #define NVME_CTRLR_DATA_AVSCC_SPEC_FORMAT_SHIFT (0) #define NVME_CTRLR_DATA_AVSCC_SPEC_FORMAT_MASK (0x1) /** Autonomous Power State Transition Attributes */ /* Autonomous Power State Transitions supported */ #define NVME_CTRLR_DATA_APSTA_APST_SUPP_SHIFT (0) #define NVME_CTRLR_DATA_APSTA_APST_SUPP_MASK (0x1) /** Sanitize Capabilities */ /* Crypto Erase Support */ #define NVME_CTRLR_DATA_SANICAP_CES_SHIFT (0) #define NVME_CTRLR_DATA_SANICAP_CES_MASK (0x1) /* Block Erase Support */ #define NVME_CTRLR_DATA_SANICAP_BES_SHIFT (1) #define NVME_CTRLR_DATA_SANICAP_BES_MASK (0x1) /* Overwrite Support */ #define NVME_CTRLR_DATA_SANICAP_OWS_SHIFT (2) #define NVME_CTRLR_DATA_SANICAP_OWS_MASK (0x1) /* No-Deallocate Inhibited */ #define NVME_CTRLR_DATA_SANICAP_NDI_SHIFT (29) #define NVME_CTRLR_DATA_SANICAP_NDI_MASK (0x1) /* No-Deallocate Modifies Media After Sanitize */ #define NVME_CTRLR_DATA_SANICAP_NODMMAS_SHIFT (30) #define NVME_CTRLR_DATA_SANICAP_NODMMAS_MASK (0x3) #define NVME_CTRLR_DATA_SANICAP_NODMMAS_UNDEF (0) #define NVME_CTRLR_DATA_SANICAP_NODMMAS_NO (1) #define NVME_CTRLR_DATA_SANICAP_NODMMAS_YES (2) /** submission queue entry size */ #define NVME_CTRLR_DATA_SQES_MIN_SHIFT (0) #define NVME_CTRLR_DATA_SQES_MIN_MASK (0xF) #define NVME_CTRLR_DATA_SQES_MAX_SHIFT (4) #define NVME_CTRLR_DATA_SQES_MAX_MASK (0xF) /** completion queue entry size */ #define NVME_CTRLR_DATA_CQES_MIN_SHIFT (0) #define NVME_CTRLR_DATA_CQES_MIN_MASK (0xF) #define NVME_CTRLR_DATA_CQES_MAX_SHIFT (4) #define NVME_CTRLR_DATA_CQES_MAX_MASK (0xF) /** optional nvm command support */ #define NVME_CTRLR_DATA_ONCS_COMPARE_SHIFT (0) #define NVME_CTRLR_DATA_ONCS_COMPARE_MASK (0x1) #define NVME_CTRLR_DATA_ONCS_WRITE_UNC_SHIFT (1) #define NVME_CTRLR_DATA_ONCS_WRITE_UNC_MASK (0x1) #define NVME_CTRLR_DATA_ONCS_DSM_SHIFT (2) #define NVME_CTRLR_DATA_ONCS_DSM_MASK (0x1) #define NVME_CTRLR_DATA_ONCS_WRZERO_SHIFT (3) #define NVME_CTRLR_DATA_ONCS_WRZERO_MASK (0x1) #define NVME_CTRLR_DATA_ONCS_SAVEFEAT_SHIFT (4) #define NVME_CTRLR_DATA_ONCS_SAVEFEAT_MASK (0x1) #define NVME_CTRLR_DATA_ONCS_RESERV_SHIFT (5) #define NVME_CTRLR_DATA_ONCS_RESERV_MASK (0x1) #define NVME_CTRLR_DATA_ONCS_TIMESTAMP_SHIFT (6) #define NVME_CTRLR_DATA_ONCS_TIMESTAMP_MASK (0x1) #define NVME_CTRLR_DATA_ONCS_VERIFY_SHIFT (7) #define NVME_CTRLR_DATA_ONCS_VERIFY_MASK (0x1) /** Fused Operation Support */ #define NVME_CTRLR_DATA_FUSES_CNW_SHIFT (0) #define NVME_CTRLR_DATA_FUSES_CNW_MASK (0x1) /** Format NVM Attributes */ #define NVME_CTRLR_DATA_FNA_FORMAT_ALL_SHIFT (0) #define NVME_CTRLR_DATA_FNA_FORMAT_ALL_MASK (0x1) #define NVME_CTRLR_DATA_FNA_ERASE_ALL_SHIFT (1) #define NVME_CTRLR_DATA_FNA_ERASE_ALL_MASK (0x1) #define NVME_CTRLR_DATA_FNA_CRYPTO_ERASE_SHIFT (2) #define NVME_CTRLR_DATA_FNA_CRYPTO_ERASE_MASK (0x1) /** volatile write cache */ /* volatile write cache present */ #define NVME_CTRLR_DATA_VWC_PRESENT_SHIFT (0) #define NVME_CTRLR_DATA_VWC_PRESENT_MASK (0x1) /* flush all namespaces supported */ #define NVME_CTRLR_DATA_VWC_ALL_SHIFT (1) #define NVME_CTRLR_DATA_VWC_ALL_MASK (0x3) #define NVME_CTRLR_DATA_VWC_ALL_UNKNOWN (0) #define NVME_CTRLR_DATA_VWC_ALL_NO (2) #define NVME_CTRLR_DATA_VWC_ALL_YES (3) /** namespace features */ /* thin provisioning */ #define NVME_NS_DATA_NSFEAT_THIN_PROV_SHIFT (0) #define NVME_NS_DATA_NSFEAT_THIN_PROV_MASK (0x1) /* NAWUN, NAWUPF, and NACWU fields are valid */ #define NVME_NS_DATA_NSFEAT_NA_FIELDS_SHIFT (1) #define NVME_NS_DATA_NSFEAT_NA_FIELDS_MASK (0x1) /* Deallocated or Unwritten Logical Block errors supported */ #define NVME_NS_DATA_NSFEAT_DEALLOC_SHIFT (2) #define NVME_NS_DATA_NSFEAT_DEALLOC_MASK (0x1) /* NGUID and EUI64 fields are not reusable */ #define NVME_NS_DATA_NSFEAT_NO_ID_REUSE_SHIFT (3) #define NVME_NS_DATA_NSFEAT_NO_ID_REUSE_MASK (0x1) /* NPWG, NPWA, NPDG, NPDA, and NOWS are valid */ #define NVME_NS_DATA_NSFEAT_NPVALID_SHIFT (4) #define NVME_NS_DATA_NSFEAT_NPVALID_MASK (0x1) /** formatted lba size */ #define NVME_NS_DATA_FLBAS_FORMAT_SHIFT (0) #define NVME_NS_DATA_FLBAS_FORMAT_MASK (0xF) #define NVME_NS_DATA_FLBAS_EXTENDED_SHIFT (4) #define NVME_NS_DATA_FLBAS_EXTENDED_MASK (0x1) /** metadata capabilities */ /* metadata can be transferred as part of data prp list */ #define NVME_NS_DATA_MC_EXTENDED_SHIFT (0) #define NVME_NS_DATA_MC_EXTENDED_MASK (0x1) /* metadata can be transferred with separate metadata pointer */ #define NVME_NS_DATA_MC_POINTER_SHIFT (1) #define NVME_NS_DATA_MC_POINTER_MASK (0x1) /** end-to-end data protection capabilities */ /* protection information type 1 */ #define NVME_NS_DATA_DPC_PIT1_SHIFT (0) #define NVME_NS_DATA_DPC_PIT1_MASK (0x1) /* protection information type 2 */ #define NVME_NS_DATA_DPC_PIT2_SHIFT (1) #define NVME_NS_DATA_DPC_PIT2_MASK (0x1) /* protection information type 3 */ #define NVME_NS_DATA_DPC_PIT3_SHIFT (2) #define NVME_NS_DATA_DPC_PIT3_MASK (0x1) /* first eight bytes of metadata */ #define NVME_NS_DATA_DPC_MD_START_SHIFT (3) #define NVME_NS_DATA_DPC_MD_START_MASK (0x1) /* last eight bytes of metadata */ #define NVME_NS_DATA_DPC_MD_END_SHIFT (4) #define NVME_NS_DATA_DPC_MD_END_MASK (0x1) /** end-to-end data protection type settings */ /* protection information type */ #define NVME_NS_DATA_DPS_PIT_SHIFT (0) #define NVME_NS_DATA_DPS_PIT_MASK (0x7) /* 1 == protection info transferred at start of metadata */ /* 0 == protection info transferred at end of metadata */ #define NVME_NS_DATA_DPS_MD_START_SHIFT (3) #define NVME_NS_DATA_DPS_MD_START_MASK (0x1) /** Namespace Multi-path I/O and Namespace Sharing Capabilities */ /* the namespace may be attached to two or more controllers */ #define NVME_NS_DATA_NMIC_MAY_BE_SHARED_SHIFT (0) #define NVME_NS_DATA_NMIC_MAY_BE_SHARED_MASK (0x1) /** Reservation Capabilities */ /* Persist Through Power Loss */ #define NVME_NS_DATA_RESCAP_PTPL_SHIFT (0) #define NVME_NS_DATA_RESCAP_PTPL_MASK (0x1) /* supports the Write Exclusive */ #define NVME_NS_DATA_RESCAP_WR_EX_SHIFT (1) #define NVME_NS_DATA_RESCAP_WR_EX_MASK (0x1) /* supports the Exclusive Access */ #define NVME_NS_DATA_RESCAP_EX_AC_SHIFT (2) #define NVME_NS_DATA_RESCAP_EX_AC_MASK (0x1) /* supports the Write Exclusive – Registrants Only */ #define NVME_NS_DATA_RESCAP_WR_EX_RO_SHIFT (3) #define NVME_NS_DATA_RESCAP_WR_EX_RO_MASK (0x1) /* supports the Exclusive Access - Registrants Only */ #define NVME_NS_DATA_RESCAP_EX_AC_RO_SHIFT (4) #define NVME_NS_DATA_RESCAP_EX_AC_RO_MASK (0x1) /* supports the Write Exclusive – All Registrants */ #define NVME_NS_DATA_RESCAP_WR_EX_AR_SHIFT (5) #define NVME_NS_DATA_RESCAP_WR_EX_AR_MASK (0x1) /* supports the Exclusive Access - All Registrants */ #define NVME_NS_DATA_RESCAP_EX_AC_AR_SHIFT (6) #define NVME_NS_DATA_RESCAP_EX_AC_AR_MASK (0x1) /* Ignore Existing Key is used as defined in revision 1.3 or later */ #define NVME_NS_DATA_RESCAP_IEKEY13_SHIFT (7) #define NVME_NS_DATA_RESCAP_IEKEY13_MASK (0x1) /** Format Progress Indicator */ /* percentage of the Format NVM command that remains to be completed */ #define NVME_NS_DATA_FPI_PERC_SHIFT (0) #define NVME_NS_DATA_FPI_PERC_MASK (0x7f) /* namespace supports the Format Progress Indicator */ #define NVME_NS_DATA_FPI_SUPP_SHIFT (7) #define NVME_NS_DATA_FPI_SUPP_MASK (0x1) /** Deallocate Logical Block Features */ /* deallocated logical block read behavior */ #define NVME_NS_DATA_DLFEAT_READ_SHIFT (0) #define NVME_NS_DATA_DLFEAT_READ_MASK (0x07) #define NVME_NS_DATA_DLFEAT_READ_NR (0x00) #define NVME_NS_DATA_DLFEAT_READ_00 (0x01) #define NVME_NS_DATA_DLFEAT_READ_FF (0x02) /* supports the Deallocate bit in the Write Zeroes */ #define NVME_NS_DATA_DLFEAT_DWZ_SHIFT (3) #define NVME_NS_DATA_DLFEAT_DWZ_MASK (0x01) /* Guard field for deallocated logical blocks is set to the CRC */ #define NVME_NS_DATA_DLFEAT_GCRC_SHIFT (4) #define NVME_NS_DATA_DLFEAT_GCRC_MASK (0x01) /** lba format support */ /* metadata size */ #define NVME_NS_DATA_LBAF_MS_SHIFT (0) #define NVME_NS_DATA_LBAF_MS_MASK (0xFFFF) /* lba data size */ #define NVME_NS_DATA_LBAF_LBADS_SHIFT (16) #define NVME_NS_DATA_LBAF_LBADS_MASK (0xFF) /* relative performance */ #define NVME_NS_DATA_LBAF_RP_SHIFT (24) #define NVME_NS_DATA_LBAF_RP_MASK (0x3) enum nvme_critical_warning_state { NVME_CRIT_WARN_ST_AVAILABLE_SPARE = 0x1, NVME_CRIT_WARN_ST_TEMPERATURE = 0x2, NVME_CRIT_WARN_ST_DEVICE_RELIABILITY = 0x4, NVME_CRIT_WARN_ST_READ_ONLY = 0x8, NVME_CRIT_WARN_ST_VOLATILE_MEMORY_BACKUP = 0x10, }; #define NVME_CRIT_WARN_ST_RESERVED_MASK (0xE0) #define NVME_ASYNC_EVENT_NS_ATTRIBUTE (0x100) #define NVME_ASYNC_EVENT_FW_ACTIVATE (0x200) /* slot for current FW */ #define NVME_FIRMWARE_PAGE_AFI_SLOT_SHIFT (0) #define NVME_FIRMWARE_PAGE_AFI_SLOT_MASK (0x7) /* Commands Supported and Effects */ #define NVME_CE_PAGE_CSUP_SHIFT (0) #define NVME_CE_PAGE_CSUP_MASK (0x1) #define NVME_CE_PAGE_LBCC_SHIFT (1) #define NVME_CE_PAGE_LBCC_MASK (0x1) #define NVME_CE_PAGE_NCC_SHIFT (2) #define NVME_CE_PAGE_NCC_MASK (0x1) #define NVME_CE_PAGE_NIC_SHIFT (3) #define NVME_CE_PAGE_NIC_MASK (0x1) #define NVME_CE_PAGE_CCC_SHIFT (4) #define NVME_CE_PAGE_CCC_MASK (0x1) #define NVME_CE_PAGE_CSE_SHIFT (16) #define NVME_CE_PAGE_CSE_MASK (0x7) #define NVME_CE_PAGE_UUID_SHIFT (19) #define NVME_CE_PAGE_UUID_MASK (0x1) /* Sanitize Status */ #define NVME_SS_PAGE_SSTAT_STATUS_SHIFT (0) #define NVME_SS_PAGE_SSTAT_STATUS_MASK (0x7) #define NVME_SS_PAGE_SSTAT_STATUS_NEVER (0) #define NVME_SS_PAGE_SSTAT_STATUS_COMPLETED (1) #define NVME_SS_PAGE_SSTAT_STATUS_INPROG (2) #define NVME_SS_PAGE_SSTAT_STATUS_FAILED (3) #define NVME_SS_PAGE_SSTAT_STATUS_COMPLETEDWD (4) #define NVME_SS_PAGE_SSTAT_PASSES_SHIFT (3) #define NVME_SS_PAGE_SSTAT_PASSES_MASK (0x1f) #define NVME_SS_PAGE_SSTAT_GDE_SHIFT (8) #define NVME_SS_PAGE_SSTAT_GDE_MASK (0x1) /* Features */ /* Get Features */ #define NVME_FEAT_GET_SEL_SHIFT (8) #define NVME_FEAT_GET_SEL_MASK (0x7) #define NVME_FEAT_GET_FID_SHIFT (0) #define NVME_FEAT_GET_FID_MASK (0xff) /* Set Features */ #define NVME_FEAT_SET_SV_SHIFT (31) #define NVME_FEAT_SET_SV_MASK (0x1) #define NVME_FEAT_SET_FID_SHIFT (0) #define NVME_FEAT_SET_FID_MASK (0xff) /* Helper macro to combine *_MASK and *_SHIFT defines */ #define NVMEB(name) (name##_MASK << name##_SHIFT) /* Helper macro to extract value from x */ #define NVMEV(name, x) (((x) >> name##_SHIFT) & name##_MASK) /* CC register SHN field values */ enum shn_value { NVME_SHN_NORMAL = 0x1, NVME_SHN_ABRUPT = 0x2, }; /* CSTS register SHST field values */ enum shst_value { NVME_SHST_NORMAL = 0x0, NVME_SHST_OCCURRING = 0x1, NVME_SHST_COMPLETE = 0x2, }; struct nvme_registers { uint32_t cap_lo; /* controller capabilities */ uint32_t cap_hi; uint32_t vs; /* version */ uint32_t intms; /* interrupt mask set */ uint32_t intmc; /* interrupt mask clear */ uint32_t cc; /* controller configuration */ uint32_t reserved1; uint32_t csts; /* controller status */ uint32_t nssr; /* NVM Subsystem Reset */ uint32_t aqa; /* admin queue attributes */ uint64_t asq; /* admin submission queue base addr */ uint64_t acq; /* admin completion queue base addr */ uint32_t cmbloc; /* Controller Memory Buffer Location */ uint32_t cmbsz; /* Controller Memory Buffer Size */ uint32_t bpinfo; /* Boot Partition Information */ uint32_t bprsel; /* Boot Partition Read Select */ uint64_t bpmbl; /* Boot Partition Memory Buffer Location */ uint64_t cmbmsc; /* Controller Memory Buffer Memory Space Control */ uint32_t cmbsts; /* Controller Memory Buffer Status */ uint32_t cmbebs; /* Controller Memory Buffer Elasticity Buffer Size */ uint32_t cmbswtp;/* Controller Memory Buffer Sustained Write Throughput */ uint32_t nssd; /* NVM Subsystem Shutdown */ uint32_t crto; /* Controller Ready Timeouts */ uint8_t reserved3[3476]; /* 6Ch - DFFh */ uint32_t pmrcap; /* Persistent Memory Capabilities */ uint32_t pmrctl; /* Persistent Memory Region Control */ uint32_t pmrsts; /* Persistent Memory Region Status */ uint32_t pmrebs; /* Persistent Memory Region Elasticity Buffer Size */ uint32_t pmrswtp; /* Persistent Memory Region Sustained Write Throughput */ uint32_t pmrmsc_lo; /* Persistent Memory Region Controller Memory Space Control */ uint32_t pmrmsc_hi; uint8_t reserved4[484]; /* E1Ch - FFFh */ struct { uint32_t sq_tdbl; /* submission queue tail doorbell */ uint32_t cq_hdbl; /* completion queue head doorbell */ } doorbell[1]; }; _Static_assert(sizeof(struct nvme_registers) == 0x1008, "bad size for nvme_registers"); struct nvme_command { /* dword 0 */ uint8_t opc; /* opcode */ uint8_t fuse; /* fused operation */ uint16_t cid; /* command identifier */ /* dword 1 */ uint32_t nsid; /* namespace identifier */ /* dword 2-3 */ uint32_t rsvd2; uint32_t rsvd3; /* dword 4-5 */ uint64_t mptr; /* metadata pointer */ /* dword 6-7 */ uint64_t prp1; /* prp entry 1 */ /* dword 8-9 */ uint64_t prp2; /* prp entry 2 */ /* dword 10-15 */ uint32_t cdw10; /* command-specific */ uint32_t cdw11; /* command-specific */ uint32_t cdw12; /* command-specific */ uint32_t cdw13; /* command-specific */ uint32_t cdw14; /* command-specific */ uint32_t cdw15; /* command-specific */ }; _Static_assert(sizeof(struct nvme_command) == 16 * 4, "bad size for nvme_command"); struct nvme_completion { /* dword 0 */ uint32_t cdw0; /* command-specific */ /* dword 1 */ uint32_t rsvd1; /* dword 2 */ uint16_t sqhd; /* submission queue head pointer */ uint16_t sqid; /* submission queue identifier */ /* dword 3 */ uint16_t cid; /* command identifier */ uint16_t status; } __aligned(8); /* riscv: nvme_qpair_process_completions has better code gen */ _Static_assert(sizeof(struct nvme_completion) == 4 * 4, "bad size for nvme_completion"); struct nvme_dsm_range { uint32_t attributes; uint32_t length; uint64_t starting_lba; }; /* Largest DSM Trim that can be done */ #define NVME_MAX_DSM_TRIM 4096 _Static_assert(sizeof(struct nvme_dsm_range) == 16, "bad size for nvme_dsm_ranage"); /* status code types */ enum nvme_status_code_type { NVME_SCT_GENERIC = 0x0, NVME_SCT_COMMAND_SPECIFIC = 0x1, NVME_SCT_MEDIA_ERROR = 0x2, NVME_SCT_PATH_RELATED = 0x3, /* 0x3-0x6 - reserved */ NVME_SCT_VENDOR_SPECIFIC = 0x7, }; /* generic command status codes */ enum nvme_generic_command_status_code { NVME_SC_SUCCESS = 0x00, NVME_SC_INVALID_OPCODE = 0x01, NVME_SC_INVALID_FIELD = 0x02, NVME_SC_COMMAND_ID_CONFLICT = 0x03, NVME_SC_DATA_TRANSFER_ERROR = 0x04, NVME_SC_ABORTED_POWER_LOSS = 0x05, NVME_SC_INTERNAL_DEVICE_ERROR = 0x06, NVME_SC_ABORTED_BY_REQUEST = 0x07, NVME_SC_ABORTED_SQ_DELETION = 0x08, NVME_SC_ABORTED_FAILED_FUSED = 0x09, NVME_SC_ABORTED_MISSING_FUSED = 0x0a, NVME_SC_INVALID_NAMESPACE_OR_FORMAT = 0x0b, NVME_SC_COMMAND_SEQUENCE_ERROR = 0x0c, NVME_SC_INVALID_SGL_SEGMENT_DESCR = 0x0d, NVME_SC_INVALID_NUMBER_OF_SGL_DESCR = 0x0e, NVME_SC_DATA_SGL_LENGTH_INVALID = 0x0f, NVME_SC_METADATA_SGL_LENGTH_INVALID = 0x10, NVME_SC_SGL_DESCRIPTOR_TYPE_INVALID = 0x11, NVME_SC_INVALID_USE_OF_CMB = 0x12, NVME_SC_PRP_OFFET_INVALID = 0x13, NVME_SC_ATOMIC_WRITE_UNIT_EXCEEDED = 0x14, NVME_SC_OPERATION_DENIED = 0x15, NVME_SC_SGL_OFFSET_INVALID = 0x16, /* 0x17 - reserved */ NVME_SC_HOST_ID_INCONSISTENT_FORMAT = 0x18, NVME_SC_KEEP_ALIVE_TIMEOUT_EXPIRED = 0x19, NVME_SC_KEEP_ALIVE_TIMEOUT_INVALID = 0x1a, NVME_SC_ABORTED_DUE_TO_PREEMPT = 0x1b, NVME_SC_SANITIZE_FAILED = 0x1c, NVME_SC_SANITIZE_IN_PROGRESS = 0x1d, NVME_SC_SGL_DATA_BLOCK_GRAN_INVALID = 0x1e, NVME_SC_NOT_SUPPORTED_IN_CMB = 0x1f, NVME_SC_NAMESPACE_IS_WRITE_PROTECTED = 0x20, NVME_SC_COMMAND_INTERRUPTED = 0x21, NVME_SC_TRANSIENT_TRANSPORT_ERROR = 0x22, NVME_SC_LBA_OUT_OF_RANGE = 0x80, NVME_SC_CAPACITY_EXCEEDED = 0x81, NVME_SC_NAMESPACE_NOT_READY = 0x82, NVME_SC_RESERVATION_CONFLICT = 0x83, NVME_SC_FORMAT_IN_PROGRESS = 0x84, }; /* command specific status codes */ enum nvme_command_specific_status_code { NVME_SC_COMPLETION_QUEUE_INVALID = 0x00, NVME_SC_INVALID_QUEUE_IDENTIFIER = 0x01, NVME_SC_MAXIMUM_QUEUE_SIZE_EXCEEDED = 0x02, NVME_SC_ABORT_COMMAND_LIMIT_EXCEEDED = 0x03, /* 0x04 - reserved */ NVME_SC_ASYNC_EVENT_REQUEST_LIMIT_EXCEEDED = 0x05, NVME_SC_INVALID_FIRMWARE_SLOT = 0x06, NVME_SC_INVALID_FIRMWARE_IMAGE = 0x07, NVME_SC_INVALID_INTERRUPT_VECTOR = 0x08, NVME_SC_INVALID_LOG_PAGE = 0x09, NVME_SC_INVALID_FORMAT = 0x0a, NVME_SC_FIRMWARE_REQUIRES_RESET = 0x0b, NVME_SC_INVALID_QUEUE_DELETION = 0x0c, NVME_SC_FEATURE_NOT_SAVEABLE = 0x0d, NVME_SC_FEATURE_NOT_CHANGEABLE = 0x0e, NVME_SC_FEATURE_NOT_NS_SPECIFIC = 0x0f, NVME_SC_FW_ACT_REQUIRES_NVMS_RESET = 0x10, NVME_SC_FW_ACT_REQUIRES_RESET = 0x11, NVME_SC_FW_ACT_REQUIRES_TIME = 0x12, NVME_SC_FW_ACT_PROHIBITED = 0x13, NVME_SC_OVERLAPPING_RANGE = 0x14, NVME_SC_NS_INSUFFICIENT_CAPACITY = 0x15, NVME_SC_NS_ID_UNAVAILABLE = 0x16, /* 0x17 - reserved */ NVME_SC_NS_ALREADY_ATTACHED = 0x18, NVME_SC_NS_IS_PRIVATE = 0x19, NVME_SC_NS_NOT_ATTACHED = 0x1a, NVME_SC_THIN_PROV_NOT_SUPPORTED = 0x1b, NVME_SC_CTRLR_LIST_INVALID = 0x1c, NVME_SC_SELF_TEST_IN_PROGRESS = 0x1d, NVME_SC_BOOT_PART_WRITE_PROHIB = 0x1e, NVME_SC_INVALID_CTRLR_ID = 0x1f, NVME_SC_INVALID_SEC_CTRLR_STATE = 0x20, NVME_SC_INVALID_NUM_OF_CTRLR_RESRC = 0x21, NVME_SC_INVALID_RESOURCE_ID = 0x22, NVME_SC_SANITIZE_PROHIBITED_WPMRE = 0x23, NVME_SC_ANA_GROUP_ID_INVALID = 0x24, NVME_SC_ANA_ATTACH_FAILED = 0x25, NVME_SC_CONFLICTING_ATTRIBUTES = 0x80, NVME_SC_INVALID_PROTECTION_INFO = 0x81, NVME_SC_ATTEMPTED_WRITE_TO_RO_PAGE = 0x82, }; /* media error status codes */ enum nvme_media_error_status_code { NVME_SC_WRITE_FAULTS = 0x80, NVME_SC_UNRECOVERED_READ_ERROR = 0x81, NVME_SC_GUARD_CHECK_ERROR = 0x82, NVME_SC_APPLICATION_TAG_CHECK_ERROR = 0x83, NVME_SC_REFERENCE_TAG_CHECK_ERROR = 0x84, NVME_SC_COMPARE_FAILURE = 0x85, NVME_SC_ACCESS_DENIED = 0x86, NVME_SC_DEALLOCATED_OR_UNWRITTEN = 0x87, }; /* path related status codes */ enum nvme_path_related_status_code { NVME_SC_INTERNAL_PATH_ERROR = 0x00, NVME_SC_ASYMMETRIC_ACCESS_PERSISTENT_LOSS = 0x01, NVME_SC_ASYMMETRIC_ACCESS_INACCESSIBLE = 0x02, NVME_SC_ASYMMETRIC_ACCESS_TRANSITION = 0x03, NVME_SC_CONTROLLER_PATHING_ERROR = 0x60, NVME_SC_HOST_PATHING_ERROR = 0x70, NVME_SC_COMMAND_ABORTED_BY_HOST = 0x71, }; /* admin opcodes */ enum nvme_admin_opcode { NVME_OPC_DELETE_IO_SQ = 0x00, NVME_OPC_CREATE_IO_SQ = 0x01, NVME_OPC_GET_LOG_PAGE = 0x02, /* 0x03 - reserved */ NVME_OPC_DELETE_IO_CQ = 0x04, NVME_OPC_CREATE_IO_CQ = 0x05, NVME_OPC_IDENTIFY = 0x06, /* 0x07 - reserved */ NVME_OPC_ABORT = 0x08, NVME_OPC_SET_FEATURES = 0x09, NVME_OPC_GET_FEATURES = 0x0a, /* 0x0b - reserved */ NVME_OPC_ASYNC_EVENT_REQUEST = 0x0c, NVME_OPC_NAMESPACE_MANAGEMENT = 0x0d, /* 0x0e-0x0f - reserved */ NVME_OPC_FIRMWARE_ACTIVATE = 0x10, NVME_OPC_FIRMWARE_IMAGE_DOWNLOAD = 0x11, /* 0x12-0x13 - reserved */ NVME_OPC_DEVICE_SELF_TEST = 0x14, NVME_OPC_NAMESPACE_ATTACHMENT = 0x15, /* 0x16-0x17 - reserved */ NVME_OPC_KEEP_ALIVE = 0x18, NVME_OPC_DIRECTIVE_SEND = 0x19, NVME_OPC_DIRECTIVE_RECEIVE = 0x1a, /* 0x1b - reserved */ NVME_OPC_VIRTUALIZATION_MANAGEMENT = 0x1c, NVME_OPC_NVME_MI_SEND = 0x1d, NVME_OPC_NVME_MI_RECEIVE = 0x1e, /* 0x1f - reserved */ NVME_OPC_CAPACITY_MANAGEMENT = 0x20, /* 0x21-0x23 - reserved */ NVME_OPC_LOCKDOWN = 0x24, /* 0x25-0x7b - reserved */ NVME_OPC_DOORBELL_BUFFER_CONFIG = 0x7c, /* 0x7d-0x7e - reserved */ NVME_OPC_FABRICS_COMMANDS = 0x7f, NVME_OPC_FORMAT_NVM = 0x80, NVME_OPC_SECURITY_SEND = 0x81, NVME_OPC_SECURITY_RECEIVE = 0x82, /* 0x83 - reserved */ NVME_OPC_SANITIZE = 0x84, /* 0x85 - reserved */ NVME_OPC_GET_LBA_STATUS = 0x86, }; /* nvme nvm opcodes */ enum nvme_nvm_opcode { NVME_OPC_FLUSH = 0x00, NVME_OPC_WRITE = 0x01, NVME_OPC_READ = 0x02, /* 0x03 - reserved */ NVME_OPC_WRITE_UNCORRECTABLE = 0x04, NVME_OPC_COMPARE = 0x05, /* 0x06-0x07 - reserved */ NVME_OPC_WRITE_ZEROES = 0x08, NVME_OPC_DATASET_MANAGEMENT = 0x09, /* 0x0a-0x0b - reserved */ NVME_OPC_VERIFY = 0x0c, NVME_OPC_RESERVATION_REGISTER = 0x0d, NVME_OPC_RESERVATION_REPORT = 0x0e, /* 0x0f-0x10 - reserved */ NVME_OPC_RESERVATION_ACQUIRE = 0x11, /* 0x12-0x14 - reserved */ NVME_OPC_RESERVATION_RELEASE = 0x15, /* 0x16-0x18 - reserved */ NVME_OPC_COPY = 0x19, }; enum nvme_feature { /* 0x00 - reserved */ NVME_FEAT_ARBITRATION = 0x01, NVME_FEAT_POWER_MANAGEMENT = 0x02, NVME_FEAT_LBA_RANGE_TYPE = 0x03, NVME_FEAT_TEMPERATURE_THRESHOLD = 0x04, NVME_FEAT_ERROR_RECOVERY = 0x05, NVME_FEAT_VOLATILE_WRITE_CACHE = 0x06, NVME_FEAT_NUMBER_OF_QUEUES = 0x07, NVME_FEAT_INTERRUPT_COALESCING = 0x08, NVME_FEAT_INTERRUPT_VECTOR_CONFIGURATION = 0x09, NVME_FEAT_WRITE_ATOMICITY = 0x0A, NVME_FEAT_ASYNC_EVENT_CONFIGURATION = 0x0B, NVME_FEAT_AUTONOMOUS_POWER_STATE_TRANSITION = 0x0C, NVME_FEAT_HOST_MEMORY_BUFFER = 0x0D, NVME_FEAT_TIMESTAMP = 0x0E, NVME_FEAT_KEEP_ALIVE_TIMER = 0x0F, NVME_FEAT_HOST_CONTROLLED_THERMAL_MGMT = 0x10, NVME_FEAT_NON_OP_POWER_STATE_CONFIG = 0x11, NVME_FEAT_READ_RECOVERY_LEVEL_CONFIG = 0x12, NVME_FEAT_PREDICTABLE_LATENCY_MODE_CONFIG = 0x13, NVME_FEAT_PREDICTABLE_LATENCY_MODE_WINDOW = 0x14, NVME_FEAT_LBA_STATUS_INFORMATION_ATTRIBUTES = 0x15, NVME_FEAT_HOST_BEHAVIOR_SUPPORT = 0x16, NVME_FEAT_SANITIZE_CONFIG = 0x17, NVME_FEAT_ENDURANCE_GROUP_EVENT_CONFIGURATION = 0x18, /* 0x19-0x77 - reserved */ /* 0x78-0x7f - NVMe Management Interface */ NVME_FEAT_SOFTWARE_PROGRESS_MARKER = 0x80, NVME_FEAT_HOST_IDENTIFIER = 0x81, NVME_FEAT_RESERVATION_NOTIFICATION_MASK = 0x82, NVME_FEAT_RESERVATION_PERSISTENCE = 0x83, NVME_FEAT_NAMESPACE_WRITE_PROTECTION_CONFIG = 0x84, /* 0x85-0xBF - command set specific (reserved) */ /* 0xC0-0xFF - vendor specific */ }; enum nvme_dsm_attribute { NVME_DSM_ATTR_INTEGRAL_READ = 0x1, NVME_DSM_ATTR_INTEGRAL_WRITE = 0x2, NVME_DSM_ATTR_DEALLOCATE = 0x4, }; enum nvme_activate_action { NVME_AA_REPLACE_NO_ACTIVATE = 0x0, NVME_AA_REPLACE_ACTIVATE = 0x1, NVME_AA_ACTIVATE = 0x2, }; struct nvme_power_state { /** Maximum Power */ uint16_t mp; /* Maximum Power */ uint8_t ps_rsvd1; uint8_t mps_nops; /* Max Power Scale, Non-Operational State */ uint32_t enlat; /* Entry Latency */ uint32_t exlat; /* Exit Latency */ uint8_t rrt; /* Relative Read Throughput */ uint8_t rrl; /* Relative Read Latency */ uint8_t rwt; /* Relative Write Throughput */ uint8_t rwl; /* Relative Write Latency */ uint16_t idlp; /* Idle Power */ uint8_t ips; /* Idle Power Scale */ uint8_t ps_rsvd8; uint16_t actp; /* Active Power */ uint8_t apw_aps; /* Active Power Workload, Active Power Scale */ uint8_t ps_rsvd10[9]; } __packed; _Static_assert(sizeof(struct nvme_power_state) == 32, "bad size for nvme_power_state"); #define NVME_SERIAL_NUMBER_LENGTH 20 #define NVME_MODEL_NUMBER_LENGTH 40 #define NVME_FIRMWARE_REVISION_LENGTH 8 struct nvme_controller_data { /* bytes 0-255: controller capabilities and features */ /** pci vendor id */ uint16_t vid; /** pci subsystem vendor id */ uint16_t ssvid; /** serial number */ uint8_t sn[NVME_SERIAL_NUMBER_LENGTH]; /** model number */ uint8_t mn[NVME_MODEL_NUMBER_LENGTH]; /** firmware revision */ uint8_t fr[NVME_FIRMWARE_REVISION_LENGTH]; /** recommended arbitration burst */ uint8_t rab; /** ieee oui identifier */ uint8_t ieee[3]; /** multi-interface capabilities */ uint8_t mic; /** maximum data transfer size */ uint8_t mdts; /** Controller ID */ uint16_t ctrlr_id; /** Version */ uint32_t ver; /** RTD3 Resume Latency */ uint32_t rtd3r; /** RTD3 Enter Latency */ uint32_t rtd3e; /** Optional Asynchronous Events Supported */ uint32_t oaes; /* bitfield really */ /** Controller Attributes */ uint32_t ctratt; /* bitfield really */ /** Read Recovery Levels Supported */ uint16_t rrls; uint8_t reserved1[9]; /** Controller Type */ uint8_t cntrltype; /** FRU Globally Unique Identifier */ uint8_t fguid[16]; /** Command Retry Delay Time 1 */ uint16_t crdt1; /** Command Retry Delay Time 2 */ uint16_t crdt2; /** Command Retry Delay Time 3 */ uint16_t crdt3; uint8_t reserved2[122]; /* bytes 256-511: admin command set attributes */ /** optional admin command support */ uint16_t oacs; /** abort command limit */ uint8_t acl; /** asynchronous event request limit */ uint8_t aerl; /** firmware updates */ uint8_t frmw; /** log page attributes */ uint8_t lpa; /** error log page entries */ uint8_t elpe; /** number of power states supported */ uint8_t npss; /** admin vendor specific command configuration */ uint8_t avscc; /** Autonomous Power State Transition Attributes */ uint8_t apsta; /** Warning Composite Temperature Threshold */ uint16_t wctemp; /** Critical Composite Temperature Threshold */ uint16_t cctemp; /** Maximum Time for Firmware Activation */ uint16_t mtfa; /** Host Memory Buffer Preferred Size */ uint32_t hmpre; /** Host Memory Buffer Minimum Size */ uint32_t hmmin; /** Name space capabilities */ struct { /* if nsmgmt, report tnvmcap and unvmcap */ uint8_t tnvmcap[16]; uint8_t unvmcap[16]; } __packed untncap; /** Replay Protected Memory Block Support */ uint32_t rpmbs; /* Really a bitfield */ /** Extended Device Self-test Time */ uint16_t edstt; /** Device Self-test Options */ uint8_t dsto; /* Really a bitfield */ /** Firmware Update Granularity */ uint8_t fwug; /** Keep Alive Support */ uint16_t kas; /** Host Controlled Thermal Management Attributes */ uint16_t hctma; /* Really a bitfield */ /** Minimum Thermal Management Temperature */ uint16_t mntmt; /** Maximum Thermal Management Temperature */ uint16_t mxtmt; /** Sanitize Capabilities */ uint32_t sanicap; /* Really a bitfield */ /** Host Memory Buffer Minimum Descriptor Entry Size */ uint32_t hmminds; /** Host Memory Maximum Descriptors Entries */ uint16_t hmmaxd; /** NVM Set Identifier Maximum */ uint16_t nsetidmax; /** Endurance Group Identifier Maximum */ uint16_t endgidmax; /** ANA Transition Time */ uint8_t anatt; /** Asymmetric Namespace Access Capabilities */ uint8_t anacap; /** ANA Group Identifier Maximum */ uint32_t anagrpmax; /** Number of ANA Group Identifiers */ uint32_t nanagrpid; /** Persistent Event Log Size */ uint32_t pels; uint8_t reserved3[156]; /* bytes 512-703: nvm command set attributes */ /** submission queue entry size */ uint8_t sqes; /** completion queue entry size */ uint8_t cqes; /** Maximum Outstanding Commands */ uint16_t maxcmd; /** number of namespaces */ uint32_t nn; /** optional nvm command support */ uint16_t oncs; /** fused operation support */ uint16_t fuses; /** format nvm attributes */ uint8_t fna; /** volatile write cache */ uint8_t vwc; /** Atomic Write Unit Normal */ uint16_t awun; /** Atomic Write Unit Power Fail */ uint16_t awupf; /** NVM Vendor Specific Command Configuration */ uint8_t nvscc; /** Namespace Write Protection Capabilities */ uint8_t nwpc; /** Atomic Compare & Write Unit */ uint16_t acwu; uint16_t reserved6; /** SGL Support */ uint32_t sgls; /** Maximum Number of Allowed Namespaces */ uint32_t mnan; /* bytes 540-767: Reserved */ uint8_t reserved7[224]; /** NVM Subsystem NVMe Qualified Name */ uint8_t subnqn[256]; /* bytes 1024-1791: Reserved */ uint8_t reserved8[768]; /* bytes 1792-2047: NVMe over Fabrics specification */ uint8_t reserved9[256]; /* bytes 2048-3071: power state descriptors */ struct nvme_power_state power_state[32]; /* bytes 3072-4095: vendor specific */ uint8_t vs[1024]; } __packed __aligned(4); _Static_assert(sizeof(struct nvme_controller_data) == 4096, "bad size for nvme_controller_data"); struct nvme_namespace_data { /** namespace size */ uint64_t nsze; /** namespace capacity */ uint64_t ncap; /** namespace utilization */ uint64_t nuse; /** namespace features */ uint8_t nsfeat; /** number of lba formats */ uint8_t nlbaf; /** formatted lba size */ uint8_t flbas; /** metadata capabilities */ uint8_t mc; /** end-to-end data protection capabilities */ uint8_t dpc; /** end-to-end data protection type settings */ uint8_t dps; /** Namespace Multi-path I/O and Namespace Sharing Capabilities */ uint8_t nmic; /** Reservation Capabilities */ uint8_t rescap; /** Format Progress Indicator */ uint8_t fpi; /** Deallocate Logical Block Features */ uint8_t dlfeat; /** Namespace Atomic Write Unit Normal */ uint16_t nawun; /** Namespace Atomic Write Unit Power Fail */ uint16_t nawupf; /** Namespace Atomic Compare & Write Unit */ uint16_t nacwu; /** Namespace Atomic Boundary Size Normal */ uint16_t nabsn; /** Namespace Atomic Boundary Offset */ uint16_t nabo; /** Namespace Atomic Boundary Size Power Fail */ uint16_t nabspf; /** Namespace Optimal IO Boundary */ uint16_t noiob; /** NVM Capacity */ uint8_t nvmcap[16]; /** Namespace Preferred Write Granularity */ uint16_t npwg; /** Namespace Preferred Write Alignment */ uint16_t npwa; /** Namespace Preferred Deallocate Granularity */ uint16_t npdg; /** Namespace Preferred Deallocate Alignment */ uint16_t npda; /** Namespace Optimal Write Size */ uint16_t nows; /* bytes 74-91: Reserved */ uint8_t reserved5[18]; /** ANA Group Identifier */ uint32_t anagrpid; /* bytes 96-98: Reserved */ uint8_t reserved6[3]; /** Namespace Attributes */ uint8_t nsattr; /** NVM Set Identifier */ uint16_t nvmsetid; /** Endurance Group Identifier */ uint16_t endgid; /** Namespace Globally Unique Identifier */ uint8_t nguid[16]; /** IEEE Extended Unique Identifier */ uint8_t eui64[8]; /** lba format support */ uint32_t lbaf[16]; uint8_t reserved7[192]; uint8_t vendor_specific[3712]; } __packed __aligned(4); _Static_assert(sizeof(struct nvme_namespace_data) == 4096, "bad size for nvme_namepsace_data"); enum nvme_log_page { /* 0x00 - reserved */ NVME_LOG_ERROR = 0x01, NVME_LOG_HEALTH_INFORMATION = 0x02, NVME_LOG_FIRMWARE_SLOT = 0x03, NVME_LOG_CHANGED_NAMESPACE = 0x04, NVME_LOG_COMMAND_EFFECT = 0x05, NVME_LOG_DEVICE_SELF_TEST = 0x06, NVME_LOG_TELEMETRY_HOST_INITIATED = 0x07, NVME_LOG_TELEMETRY_CONTROLLER_INITIATED = 0x08, NVME_LOG_ENDURANCE_GROUP_INFORMATION = 0x09, NVME_LOG_PREDICTABLE_LATENCY_PER_NVM_SET = 0x0a, NVME_LOG_PREDICTABLE_LATENCY_EVENT_AGGREGATE = 0x0b, NVME_LOG_ASYMMETRIC_NAMESPACE_ACCESS = 0x0c, NVME_LOG_PERSISTENT_EVENT_LOG = 0x0d, NVME_LOG_LBA_STATUS_INFORMATION = 0x0e, NVME_LOG_ENDURANCE_GROUP_EVENT_AGGREGATE = 0x0f, /* 0x06-0x7F - reserved */ /* 0x80-0xBF - I/O command set specific */ NVME_LOG_RES_NOTIFICATION = 0x80, NVME_LOG_SANITIZE_STATUS = 0x81, /* 0x82-0xBF - reserved */ /* 0xC0-0xFF - vendor specific */ /* * The following are Intel Specific log pages, but they seem * to be widely implemented. */ INTEL_LOG_READ_LAT_LOG = 0xc1, INTEL_LOG_WRITE_LAT_LOG = 0xc2, INTEL_LOG_TEMP_STATS = 0xc5, INTEL_LOG_ADD_SMART = 0xca, INTEL_LOG_DRIVE_MKT_NAME = 0xdd, /* * HGST log page, with lots ofs sub pages. */ HGST_INFO_LOG = 0xc1, }; struct nvme_error_information_entry { uint64_t error_count; uint16_t sqid; uint16_t cid; uint16_t status; uint16_t error_location; uint64_t lba; uint32_t nsid; uint8_t vendor_specific; uint8_t trtype; uint16_t reserved30; uint64_t csi; uint16_t ttsi; uint8_t reserved[22]; } __packed __aligned(4); _Static_assert(sizeof(struct nvme_error_information_entry) == 64, "bad size for nvme_error_information_entry"); struct nvme_health_information_page { uint8_t critical_warning; uint16_t temperature; uint8_t available_spare; uint8_t available_spare_threshold; uint8_t percentage_used; uint8_t reserved[26]; /* * Note that the following are 128-bit values, but are * defined as an array of 2 64-bit values. */ /* Data Units Read is always in 512-byte units. */ uint64_t data_units_read[2]; /* Data Units Written is always in 512-byte units. */ uint64_t data_units_written[2]; /* For NVM command set, this includes Compare commands. */ uint64_t host_read_commands[2]; uint64_t host_write_commands[2]; /* Controller Busy Time is reported in minutes. */ uint64_t controller_busy_time[2]; uint64_t power_cycles[2]; uint64_t power_on_hours[2]; uint64_t unsafe_shutdowns[2]; uint64_t media_errors[2]; uint64_t num_error_info_log_entries[2]; uint32_t warning_temp_time; uint32_t error_temp_time; uint16_t temp_sensor[8]; /* Thermal Management Temperature 1 Transition Count */ uint32_t tmt1tc; /* Thermal Management Temperature 2 Transition Count */ uint32_t tmt2tc; /* Total Time For Thermal Management Temperature 1 */ uint32_t ttftmt1; /* Total Time For Thermal Management Temperature 2 */ uint32_t ttftmt2; uint8_t reserved2[280]; } __packed __aligned(4); _Static_assert(sizeof(struct nvme_health_information_page) == 512, "bad size for nvme_health_information_page"); struct nvme_firmware_page { uint8_t afi; uint8_t reserved[7]; /* revisions for 7 slots */ uint8_t revision[7][NVME_FIRMWARE_REVISION_LENGTH]; uint8_t reserved2[448]; } __packed __aligned(4); _Static_assert(sizeof(struct nvme_firmware_page) == 512, "bad size for nvme_firmware_page"); struct nvme_ns_list { uint32_t ns[1024]; } __packed __aligned(4); _Static_assert(sizeof(struct nvme_ns_list) == 4096, "bad size for nvme_ns_list"); struct nvme_command_effects_page { uint32_t acs[256]; uint32_t iocs[256]; uint8_t reserved[2048]; } __packed __aligned(4); _Static_assert(sizeof(struct nvme_command_effects_page) == 4096, "bad size for nvme_command_effects_page"); struct nvme_device_self_test_page { uint8_t curr_operation; uint8_t curr_compl; uint8_t rsvd2[2]; struct { uint8_t status; uint8_t segment_num; uint8_t valid_diag_info; uint8_t rsvd3; uint64_t poh; uint32_t nsid; /* Define as an array to simplify alignment issues */ uint8_t failing_lba[8]; uint8_t status_code_type; uint8_t status_code; uint8_t vendor_specific[2]; } __packed result[20]; } __packed __aligned(4); _Static_assert(sizeof(struct nvme_device_self_test_page) == 564, "bad size for nvme_device_self_test_page"); struct nvme_res_notification_page { uint64_t log_page_count; uint8_t log_page_type; uint8_t available_log_pages; uint8_t reserved2; uint32_t nsid; uint8_t reserved[48]; } __packed __aligned(4); _Static_assert(sizeof(struct nvme_res_notification_page) == 64, "bad size for nvme_res_notification_page"); struct nvme_sanitize_status_page { uint16_t sprog; uint16_t sstat; uint32_t scdw10; uint32_t etfo; uint32_t etfbe; uint32_t etfce; uint32_t etfownd; uint32_t etfbewnd; uint32_t etfcewnd; uint8_t reserved[480]; } __packed __aligned(4); _Static_assert(sizeof(struct nvme_sanitize_status_page) == 512, "bad size for nvme_sanitize_status_page"); struct intel_log_temp_stats { uint64_t current; uint64_t overtemp_flag_last; uint64_t overtemp_flag_life; uint64_t max_temp; uint64_t min_temp; uint64_t _rsvd[5]; uint64_t max_oper_temp; uint64_t min_oper_temp; uint64_t est_offset; } __packed __aligned(4); _Static_assert(sizeof(struct intel_log_temp_stats) == 13 * 8, "bad size for intel_log_temp_stats"); struct nvme_resv_reg_ctrlr { uint16_t ctrlr_id; /* Controller ID */ uint8_t rcsts; /* Reservation Status */ uint8_t reserved3[5]; uint64_t hostid; /* Host Identifier */ uint64_t rkey; /* Reservation Key */ } __packed __aligned(4); _Static_assert(sizeof(struct nvme_resv_reg_ctrlr) == 24, "bad size for nvme_resv_reg_ctrlr"); struct nvme_resv_reg_ctrlr_ext { uint16_t ctrlr_id; /* Controller ID */ uint8_t rcsts; /* Reservation Status */ uint8_t reserved3[5]; uint64_t rkey; /* Reservation Key */ uint64_t hostid[2]; /* Host Identifier */ uint8_t reserved32[32]; } __packed __aligned(4); _Static_assert(sizeof(struct nvme_resv_reg_ctrlr_ext) == 64, "bad size for nvme_resv_reg_ctrlr_ext"); struct nvme_resv_status { uint32_t gen; /* Generation */ uint8_t rtype; /* Reservation Type */ uint8_t regctl[2]; /* Number of Registered Controllers */ uint8_t reserved7[2]; uint8_t ptpls; /* Persist Through Power Loss State */ uint8_t reserved10[14]; struct nvme_resv_reg_ctrlr ctrlr[0]; } __packed __aligned(4); _Static_assert(sizeof(struct nvme_resv_status) == 24, "bad size for nvme_resv_status"); struct nvme_resv_status_ext { uint32_t gen; /* Generation */ uint8_t rtype; /* Reservation Type */ uint8_t regctl[2]; /* Number of Registered Controllers */ uint8_t reserved7[2]; uint8_t ptpls; /* Persist Through Power Loss State */ uint8_t reserved10[14]; uint8_t reserved24[40]; struct nvme_resv_reg_ctrlr_ext ctrlr[0]; } __packed __aligned(4); _Static_assert(sizeof(struct nvme_resv_status_ext) == 64, "bad size for nvme_resv_status_ext"); #define NVME_TEST_MAX_THREADS 128 struct nvme_io_test { enum nvme_nvm_opcode opc; uint32_t size; uint32_t time; /* in seconds */ uint32_t num_threads; uint32_t flags; uint64_t io_completed[NVME_TEST_MAX_THREADS]; }; enum nvme_io_test_flags { /* * Specifies whether dev_refthread/dev_relthread should be * called during NVME_BIO_TEST. Ignored for other test * types. */ NVME_TEST_FLAG_REFTHREAD = 0x1, }; struct nvme_pt_command { /* * cmd is used to specify a passthrough command to a controller or * namespace. * * The following fields from cmd may be specified by the caller: * * opc (opcode) * * nsid (namespace id) - for admin commands only * * cdw10-cdw15 * * Remaining fields must be set to 0 by the caller. */ struct nvme_command cmd; /* * cpl returns completion status for the passthrough command * specified by cmd. * * The following fields will be filled out by the driver, for * consumption by the caller: * * cdw0 * * status (except for phase) * * Remaining fields will be set to 0 by the driver. */ struct nvme_completion cpl; /* buf is the data buffer associated with this passthrough command. */ void * buf; /* * len is the length of the data buffer associated with this * passthrough command. */ uint32_t len; /* * is_read = 1 if the passthrough command will read data into the * supplied buffer from the controller. * * is_read = 0 if the passthrough command will write data from the * supplied buffer to the controller. */ uint32_t is_read; /* * driver_lock is used by the driver only. It must be set to 0 * by the caller. */ struct mtx * driver_lock; }; struct nvme_get_nsid { char cdev[SPECNAMELEN + 1]; uint32_t nsid; }; struct nvme_hmb_desc { uint64_t addr; uint32_t size; uint32_t reserved; }; #define nvme_completion_is_error(cpl) \ (NVME_STATUS_GET_SC((cpl)->status) != 0 || NVME_STATUS_GET_SCT((cpl)->status) != 0) void nvme_strvis(uint8_t *dst, const uint8_t *src, int dstlen, int srclen); #ifdef _KERNEL struct bio; struct thread; struct nvme_namespace; struct nvme_controller; struct nvme_consumer; typedef void (*nvme_cb_fn_t)(void *, const struct nvme_completion *); typedef void *(*nvme_cons_ns_fn_t)(struct nvme_namespace *, void *); typedef void *(*nvme_cons_ctrlr_fn_t)(struct nvme_controller *); typedef void (*nvme_cons_async_fn_t)(void *, const struct nvme_completion *, uint32_t, void *, uint32_t); typedef void (*nvme_cons_fail_fn_t)(void *); enum nvme_namespace_flags { NVME_NS_DEALLOCATE_SUPPORTED = 0x1, NVME_NS_FLUSH_SUPPORTED = 0x2, }; int nvme_ctrlr_passthrough_cmd(struct nvme_controller *ctrlr, struct nvme_pt_command *pt, uint32_t nsid, int is_user_buffer, int is_admin_cmd); /* Admin functions */ void nvme_ctrlr_cmd_set_feature(struct nvme_controller *ctrlr, uint8_t feature, uint32_t cdw11, uint32_t cdw12, uint32_t cdw13, uint32_t cdw14, uint32_t cdw15, void *payload, uint32_t payload_size, nvme_cb_fn_t cb_fn, void *cb_arg); void nvme_ctrlr_cmd_get_feature(struct nvme_controller *ctrlr, uint8_t feature, uint32_t cdw11, void *payload, uint32_t payload_size, nvme_cb_fn_t cb_fn, void *cb_arg); void nvme_ctrlr_cmd_get_log_page(struct nvme_controller *ctrlr, uint8_t log_page, uint32_t nsid, void *payload, uint32_t payload_size, nvme_cb_fn_t cb_fn, void *cb_arg); /* NVM I/O functions */ int nvme_ns_cmd_write(struct nvme_namespace *ns, void *payload, uint64_t lba, uint32_t lba_count, nvme_cb_fn_t cb_fn, void *cb_arg); int nvme_ns_cmd_write_bio(struct nvme_namespace *ns, struct bio *bp, nvme_cb_fn_t cb_fn, void *cb_arg); int nvme_ns_cmd_read(struct nvme_namespace *ns, void *payload, uint64_t lba, uint32_t lba_count, nvme_cb_fn_t cb_fn, void *cb_arg); int nvme_ns_cmd_read_bio(struct nvme_namespace *ns, struct bio *bp, nvme_cb_fn_t cb_fn, void *cb_arg); int nvme_ns_cmd_deallocate(struct nvme_namespace *ns, void *payload, uint8_t num_ranges, nvme_cb_fn_t cb_fn, void *cb_arg); int nvme_ns_cmd_flush(struct nvme_namespace *ns, nvme_cb_fn_t cb_fn, void *cb_arg); int nvme_ns_dump(struct nvme_namespace *ns, void *virt, off_t offset, size_t len); /* Registration functions */ struct nvme_consumer * nvme_register_consumer(nvme_cons_ns_fn_t ns_fn, nvme_cons_ctrlr_fn_t ctrlr_fn, nvme_cons_async_fn_t async_fn, nvme_cons_fail_fn_t fail_fn); void nvme_unregister_consumer(struct nvme_consumer *consumer); /* Controller helper functions */ device_t nvme_ctrlr_get_device(struct nvme_controller *ctrlr); const struct nvme_controller_data * nvme_ctrlr_get_data(struct nvme_controller *ctrlr); static inline bool nvme_ctrlr_has_dataset_mgmt(const struct nvme_controller_data *cd) { /* Assumes cd was byte swapped by nvme_controller_data_swapbytes() */ - return ((cd->oncs >> NVME_CTRLR_DATA_ONCS_DSM_SHIFT) & - NVME_CTRLR_DATA_ONCS_DSM_MASK); + return (NVMEV(NVME_CTRLR_DATA_ONCS_DSM, cd->oncs) != 0); } /* Namespace helper functions */ uint32_t nvme_ns_get_max_io_xfer_size(struct nvme_namespace *ns); uint32_t nvme_ns_get_sector_size(struct nvme_namespace *ns); uint64_t nvme_ns_get_num_sectors(struct nvme_namespace *ns); uint64_t nvme_ns_get_size(struct nvme_namespace *ns); uint32_t nvme_ns_get_flags(struct nvme_namespace *ns); const char * nvme_ns_get_serial_number(struct nvme_namespace *ns); const char * nvme_ns_get_model_number(struct nvme_namespace *ns); const struct nvme_namespace_data * nvme_ns_get_data(struct nvme_namespace *ns); uint32_t nvme_ns_get_stripesize(struct nvme_namespace *ns); int nvme_ns_bio_process(struct nvme_namespace *ns, struct bio *bp, nvme_cb_fn_t cb_fn); int nvme_ns_ioctl_process(struct nvme_namespace *ns, u_long cmd, caddr_t arg, int flag, struct thread *td); /* * Command building helper functions -- shared with CAM * These functions assume allocator zeros out cmd structure * CAM's xpt_get_ccb and the request allocator for nvme both * do zero'd allocations. */ static inline void nvme_ns_flush_cmd(struct nvme_command *cmd, uint32_t nsid) { cmd->opc = NVME_OPC_FLUSH; cmd->nsid = htole32(nsid); } static inline void nvme_ns_rw_cmd(struct nvme_command *cmd, uint32_t rwcmd, uint32_t nsid, uint64_t lba, uint32_t count) { cmd->opc = rwcmd; cmd->nsid = htole32(nsid); cmd->cdw10 = htole32(lba & 0xffffffffu); cmd->cdw11 = htole32(lba >> 32); cmd->cdw12 = htole32(count-1); } static inline void nvme_ns_write_cmd(struct nvme_command *cmd, uint32_t nsid, uint64_t lba, uint32_t count) { nvme_ns_rw_cmd(cmd, NVME_OPC_WRITE, nsid, lba, count); } static inline void nvme_ns_read_cmd(struct nvme_command *cmd, uint32_t nsid, uint64_t lba, uint32_t count) { nvme_ns_rw_cmd(cmd, NVME_OPC_READ, nsid, lba, count); } static inline void nvme_ns_trim_cmd(struct nvme_command *cmd, uint32_t nsid, uint32_t num_ranges) { cmd->opc = NVME_OPC_DATASET_MANAGEMENT; cmd->nsid = htole32(nsid); cmd->cdw10 = htole32(num_ranges - 1); cmd->cdw11 = htole32(NVME_DSM_ATTR_DEALLOCATE); } extern int nvme_use_nvd; #endif /* _KERNEL */ /* Endianess conversion functions for NVMe structs */ static inline void nvme_completion_swapbytes(struct nvme_completion *s __unused) { #if _BYTE_ORDER != _LITTLE_ENDIAN s->cdw0 = le32toh(s->cdw0); /* omit rsvd1 */ s->sqhd = le16toh(s->sqhd); s->sqid = le16toh(s->sqid); /* omit cid */ s->status = le16toh(s->status); #endif } static inline void nvme_power_state_swapbytes(struct nvme_power_state *s __unused) { #if _BYTE_ORDER != _LITTLE_ENDIAN s->mp = le16toh(s->mp); s->enlat = le32toh(s->enlat); s->exlat = le32toh(s->exlat); s->idlp = le16toh(s->idlp); s->actp = le16toh(s->actp); #endif } static inline void nvme_controller_data_swapbytes(struct nvme_controller_data *s __unused) { #if _BYTE_ORDER != _LITTLE_ENDIAN int i; s->vid = le16toh(s->vid); s->ssvid = le16toh(s->ssvid); s->ctrlr_id = le16toh(s->ctrlr_id); s->ver = le32toh(s->ver); s->rtd3r = le32toh(s->rtd3r); s->rtd3e = le32toh(s->rtd3e); s->oaes = le32toh(s->oaes); s->ctratt = le32toh(s->ctratt); s->rrls = le16toh(s->rrls); s->crdt1 = le16toh(s->crdt1); s->crdt2 = le16toh(s->crdt2); s->crdt3 = le16toh(s->crdt3); s->oacs = le16toh(s->oacs); s->wctemp = le16toh(s->wctemp); s->cctemp = le16toh(s->cctemp); s->mtfa = le16toh(s->mtfa); s->hmpre = le32toh(s->hmpre); s->hmmin = le32toh(s->hmmin); s->rpmbs = le32toh(s->rpmbs); s->edstt = le16toh(s->edstt); s->kas = le16toh(s->kas); s->hctma = le16toh(s->hctma); s->mntmt = le16toh(s->mntmt); s->mxtmt = le16toh(s->mxtmt); s->sanicap = le32toh(s->sanicap); s->hmminds = le32toh(s->hmminds); s->hmmaxd = le16toh(s->hmmaxd); s->nsetidmax = le16toh(s->nsetidmax); s->endgidmax = le16toh(s->endgidmax); s->anagrpmax = le32toh(s->anagrpmax); s->nanagrpid = le32toh(s->nanagrpid); s->pels = le32toh(s->pels); s->maxcmd = le16toh(s->maxcmd); s->nn = le32toh(s->nn); s->oncs = le16toh(s->oncs); s->fuses = le16toh(s->fuses); s->awun = le16toh(s->awun); s->awupf = le16toh(s->awupf); s->acwu = le16toh(s->acwu); s->sgls = le32toh(s->sgls); s->mnan = le32toh(s->mnan); for (i = 0; i < 32; i++) nvme_power_state_swapbytes(&s->power_state[i]); #endif } static inline void nvme_namespace_data_swapbytes(struct nvme_namespace_data *s __unused) { #if _BYTE_ORDER != _LITTLE_ENDIAN int i; s->nsze = le64toh(s->nsze); s->ncap = le64toh(s->ncap); s->nuse = le64toh(s->nuse); s->nawun = le16toh(s->nawun); s->nawupf = le16toh(s->nawupf); s->nacwu = le16toh(s->nacwu); s->nabsn = le16toh(s->nabsn); s->nabo = le16toh(s->nabo); s->nabspf = le16toh(s->nabspf); s->noiob = le16toh(s->noiob); s->npwg = le16toh(s->npwg); s->npwa = le16toh(s->npwa); s->npdg = le16toh(s->npdg); s->npda = le16toh(s->npda); s->nows = le16toh(s->nows); s->anagrpid = le32toh(s->anagrpid); s->nvmsetid = le16toh(s->nvmsetid); s->endgid = le16toh(s->endgid); for (i = 0; i < 16; i++) s->lbaf[i] = le32toh(s->lbaf[i]); #endif } static inline void nvme_error_information_entry_swapbytes( struct nvme_error_information_entry *s __unused) { #if _BYTE_ORDER != _LITTLE_ENDIAN s->error_count = le64toh(s->error_count); s->sqid = le16toh(s->sqid); s->cid = le16toh(s->cid); s->status = le16toh(s->status); s->error_location = le16toh(s->error_location); s->lba = le64toh(s->lba); s->nsid = le32toh(s->nsid); s->csi = le64toh(s->csi); s->ttsi = le16toh(s->ttsi); #endif } static inline void nvme_le128toh(void *p __unused) { #if _BYTE_ORDER != _LITTLE_ENDIAN /* Swap 16 bytes in place */ char *tmp = (char*)p; char b; int i; for (i = 0; i < 8; i++) { b = tmp[i]; tmp[i] = tmp[15-i]; tmp[15-i] = b; } #endif } static inline void nvme_health_information_page_swapbytes( struct nvme_health_information_page *s __unused) { #if _BYTE_ORDER != _LITTLE_ENDIAN int i; s->temperature = le16toh(s->temperature); nvme_le128toh((void *)s->data_units_read); nvme_le128toh((void *)s->data_units_written); nvme_le128toh((void *)s->host_read_commands); nvme_le128toh((void *)s->host_write_commands); nvme_le128toh((void *)s->controller_busy_time); nvme_le128toh((void *)s->power_cycles); nvme_le128toh((void *)s->power_on_hours); nvme_le128toh((void *)s->unsafe_shutdowns); nvme_le128toh((void *)s->media_errors); nvme_le128toh((void *)s->num_error_info_log_entries); s->warning_temp_time = le32toh(s->warning_temp_time); s->error_temp_time = le32toh(s->error_temp_time); for (i = 0; i < 8; i++) s->temp_sensor[i] = le16toh(s->temp_sensor[i]); s->tmt1tc = le32toh(s->tmt1tc); s->tmt2tc = le32toh(s->tmt2tc); s->ttftmt1 = le32toh(s->ttftmt1); s->ttftmt2 = le32toh(s->ttftmt2); #endif } static inline void nvme_ns_list_swapbytes(struct nvme_ns_list *s __unused) { #if _BYTE_ORDER != _LITTLE_ENDIAN int i; for (i = 0; i < 1024; i++) s->ns[i] = le32toh(s->ns[i]); #endif } static inline void nvme_command_effects_page_swapbytes( struct nvme_command_effects_page *s __unused) { #if _BYTE_ORDER != _LITTLE_ENDIAN int i; for (i = 0; i < 256; i++) s->acs[i] = le32toh(s->acs[i]); for (i = 0; i < 256; i++) s->iocs[i] = le32toh(s->iocs[i]); #endif } static inline void nvme_res_notification_page_swapbytes( struct nvme_res_notification_page *s __unused) { #if _BYTE_ORDER != _LITTLE_ENDIAN s->log_page_count = le64toh(s->log_page_count); s->nsid = le32toh(s->nsid); #endif } static inline void nvme_sanitize_status_page_swapbytes( struct nvme_sanitize_status_page *s __unused) { #if _BYTE_ORDER != _LITTLE_ENDIAN s->sprog = le16toh(s->sprog); s->sstat = le16toh(s->sstat); s->scdw10 = le32toh(s->scdw10); s->etfo = le32toh(s->etfo); s->etfbe = le32toh(s->etfbe); s->etfce = le32toh(s->etfce); s->etfownd = le32toh(s->etfownd); s->etfbewnd = le32toh(s->etfbewnd); s->etfcewnd = le32toh(s->etfcewnd); #endif } static inline void intel_log_temp_stats_swapbytes(struct intel_log_temp_stats *s __unused) { #if _BYTE_ORDER != _LITTLE_ENDIAN s->current = le64toh(s->current); s->overtemp_flag_last = le64toh(s->overtemp_flag_last); s->overtemp_flag_life = le64toh(s->overtemp_flag_life); s->max_temp = le64toh(s->max_temp); s->min_temp = le64toh(s->min_temp); /* omit _rsvd[] */ s->max_oper_temp = le64toh(s->max_oper_temp); s->min_oper_temp = le64toh(s->min_oper_temp); s->est_offset = le64toh(s->est_offset); #endif } static inline void nvme_resv_status_swapbytes(struct nvme_resv_status *s __unused, size_t size __unused) { #if _BYTE_ORDER != _LITTLE_ENDIAN size_t i, n; s->gen = le32toh(s->gen); n = (s->regctl[1] << 8) | s->regctl[0]; n = MIN(n, (size - sizeof(s)) / sizeof(s->ctrlr[0])); for (i = 0; i < n; i++) { s->ctrlr[i].ctrlr_id = le16toh(s->ctrlr[i].ctrlr_id); s->ctrlr[i].hostid = le64toh(s->ctrlr[i].hostid); s->ctrlr[i].rkey = le64toh(s->ctrlr[i].rkey); } #endif } static inline void nvme_resv_status_ext_swapbytes(struct nvme_resv_status_ext *s __unused, size_t size __unused) { #if _BYTE_ORDER != _LITTLE_ENDIAN size_t i, n; s->gen = le32toh(s->gen); n = (s->regctl[1] << 8) | s->regctl[0]; n = MIN(n, (size - sizeof(s)) / sizeof(s->ctrlr[0])); for (i = 0; i < n; i++) { s->ctrlr[i].ctrlr_id = le16toh(s->ctrlr[i].ctrlr_id); s->ctrlr[i].rkey = le64toh(s->ctrlr[i].rkey); nvme_le128toh((void *)s->ctrlr[i].hostid); } #endif } static inline void nvme_device_self_test_swapbytes(struct nvme_device_self_test_page *s __unused) { #if _BYTE_ORDER != _LITTLE_ENDIAN uint8_t *tmp; uint32_t r, i; uint8_t b; for (r = 0; r < 20; r++) { s->result[r].poh = le64toh(s->result[r].poh); s->result[r].nsid = le32toh(s->result[r].nsid); /* Unaligned 64-bit loads fail on some architectures */ tmp = s->result[r].failing_lba; for (i = 0; i < 4; i++) { b = tmp[i]; tmp[i] = tmp[7-i]; tmp[7-i] = b; } } #endif } #endif /* __NVME_H__ */ diff --git a/sys/dev/nvme/nvme_ctrlr.c b/sys/dev/nvme/nvme_ctrlr.c index e659a3776957..7096f0443f1d 100644 --- a/sys/dev/nvme/nvme_ctrlr.c +++ b/sys/dev/nvme/nvme_ctrlr.c @@ -1,1732 +1,1731 @@ /*- * SPDX-License-Identifier: BSD-2-Clause * * Copyright (C) 2012-2016 Intel Corporation * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ #include #include "opt_cam.h" #include "opt_nvme.h" #include #include #include #include #include #include #include #include #include #include #include #include #include #include "nvme_private.h" #define B4_CHK_RDY_DELAY_MS 2300 /* work around controller bug */ static void nvme_ctrlr_construct_and_submit_aer(struct nvme_controller *ctrlr, struct nvme_async_event_request *aer); static void nvme_ctrlr_barrier(struct nvme_controller *ctrlr, int flags) { bus_barrier(ctrlr->resource, 0, rman_get_size(ctrlr->resource), flags); } static void nvme_ctrlr_devctl_log(struct nvme_controller *ctrlr, const char *type, const char *msg, ...) { struct sbuf sb; va_list ap; int error; if (sbuf_new(&sb, NULL, 0, SBUF_AUTOEXTEND | SBUF_NOWAIT) == NULL) return; sbuf_printf(&sb, "%s: ", device_get_nameunit(ctrlr->dev)); va_start(ap, msg); sbuf_vprintf(&sb, msg, ap); va_end(ap); error = sbuf_finish(&sb); if (error == 0) printf("%s\n", sbuf_data(&sb)); sbuf_clear(&sb); sbuf_printf(&sb, "name=\"%s\" reason=\"", device_get_nameunit(ctrlr->dev)); va_start(ap, msg); sbuf_vprintf(&sb, msg, ap); va_end(ap); sbuf_printf(&sb, "\""); error = sbuf_finish(&sb); if (error == 0) devctl_notify("nvme", "controller", type, sbuf_data(&sb)); sbuf_delete(&sb); } static int nvme_ctrlr_construct_admin_qpair(struct nvme_controller *ctrlr) { struct nvme_qpair *qpair; uint32_t num_entries; int error; qpair = &ctrlr->adminq; qpair->id = 0; qpair->cpu = CPU_FFS(&cpuset_domain[ctrlr->domain]) - 1; qpair->domain = ctrlr->domain; num_entries = NVME_ADMIN_ENTRIES; TUNABLE_INT_FETCH("hw.nvme.admin_entries", &num_entries); /* * If admin_entries was overridden to an invalid value, revert it * back to our default value. */ if (num_entries < NVME_MIN_ADMIN_ENTRIES || num_entries > NVME_MAX_ADMIN_ENTRIES) { nvme_printf(ctrlr, "invalid hw.nvme.admin_entries=%d " "specified\n", num_entries); num_entries = NVME_ADMIN_ENTRIES; } /* * The admin queue's max xfer size is treated differently than the * max I/O xfer size. 16KB is sufficient here - maybe even less? */ error = nvme_qpair_construct(qpair, num_entries, NVME_ADMIN_TRACKERS, ctrlr); return (error); } #define QP(ctrlr, c) ((c) * (ctrlr)->num_io_queues / mp_ncpus) static int nvme_ctrlr_construct_io_qpairs(struct nvme_controller *ctrlr) { struct nvme_qpair *qpair; uint32_t cap_lo; uint16_t mqes; int c, error, i, n; int num_entries, num_trackers, max_entries; /* * NVMe spec sets a hard limit of 64K max entries, but devices may * specify a smaller limit, so we need to check the MQES field in the * capabilities register. We have to cap the number of entries to the * current stride allows for in BAR 0/1, otherwise the remainder entries * are inaccessible. MQES should reflect this, and this is just a * fail-safe. */ max_entries = (rman_get_size(ctrlr->resource) - nvme_mmio_offsetof(doorbell[0])) / (1 << (ctrlr->dstrd + 1)); num_entries = NVME_IO_ENTRIES; TUNABLE_INT_FETCH("hw.nvme.io_entries", &num_entries); cap_lo = nvme_mmio_read_4(ctrlr, cap_lo); mqes = NVME_CAP_LO_MQES(cap_lo); num_entries = min(num_entries, mqes + 1); num_entries = min(num_entries, max_entries); num_trackers = NVME_IO_TRACKERS; TUNABLE_INT_FETCH("hw.nvme.io_trackers", &num_trackers); num_trackers = max(num_trackers, NVME_MIN_IO_TRACKERS); num_trackers = min(num_trackers, NVME_MAX_IO_TRACKERS); /* * No need to have more trackers than entries in the submit queue. Note * also that for a queue size of N, we can only have (N-1) commands * outstanding, hence the "-1" here. */ num_trackers = min(num_trackers, (num_entries-1)); /* * Our best estimate for the maximum number of I/Os that we should * normally have in flight at one time. This should be viewed as a hint, * not a hard limit and will need to be revisited when the upper layers * of the storage system grows multi-queue support. */ ctrlr->max_hw_pend_io = num_trackers * ctrlr->num_io_queues * 3 / 4; ctrlr->ioq = malloc(ctrlr->num_io_queues * sizeof(struct nvme_qpair), M_NVME, M_ZERO | M_WAITOK); for (i = c = n = 0; i < ctrlr->num_io_queues; i++, c += n) { qpair = &ctrlr->ioq[i]; /* * Admin queue has ID=0. IO queues start at ID=1 - * hence the 'i+1' here. */ qpair->id = i + 1; if (ctrlr->num_io_queues > 1) { /* Find number of CPUs served by this queue. */ for (n = 1; QP(ctrlr, c + n) == i; n++) ; /* Shuffle multiple NVMe devices between CPUs. */ qpair->cpu = c + (device_get_unit(ctrlr->dev)+n/2) % n; qpair->domain = pcpu_find(qpair->cpu)->pc_domain; } else { qpair->cpu = CPU_FFS(&cpuset_domain[ctrlr->domain]) - 1; qpair->domain = ctrlr->domain; } /* * For I/O queues, use the controller-wide max_xfer_size * calculated in nvme_attach(). */ error = nvme_qpair_construct(qpair, num_entries, num_trackers, ctrlr); if (error) return (error); /* * Do not bother binding interrupts if we only have one I/O * interrupt thread for this controller. */ if (ctrlr->num_io_queues > 1) bus_bind_intr(ctrlr->dev, qpair->res, qpair->cpu); } return (0); } static void nvme_ctrlr_fail(struct nvme_controller *ctrlr) { int i; /* * No need to disable queues before failing them. Failing is a superet * of disabling (though pedantically we'd abort the AERs silently with * a different error, though when we fail, that hardly matters). */ ctrlr->is_failed = true; nvme_qpair_fail(&ctrlr->adminq); if (ctrlr->ioq != NULL) { for (i = 0; i < ctrlr->num_io_queues; i++) { nvme_qpair_fail(&ctrlr->ioq[i]); } } nvme_notify_fail_consumers(ctrlr); } void nvme_ctrlr_post_failed_request(struct nvme_controller *ctrlr, struct nvme_request *req) { mtx_lock(&ctrlr->lock); STAILQ_INSERT_TAIL(&ctrlr->fail_req, req, stailq); mtx_unlock(&ctrlr->lock); if (!ctrlr->is_dying) taskqueue_enqueue(ctrlr->taskqueue, &ctrlr->fail_req_task); } static void nvme_ctrlr_fail_req_task(void *arg, int pending) { struct nvme_controller *ctrlr = arg; struct nvme_request *req; mtx_lock(&ctrlr->lock); while ((req = STAILQ_FIRST(&ctrlr->fail_req)) != NULL) { STAILQ_REMOVE_HEAD(&ctrlr->fail_req, stailq); mtx_unlock(&ctrlr->lock); nvme_qpair_manual_complete_request(req->qpair, req, NVME_SCT_GENERIC, NVME_SC_ABORTED_BY_REQUEST); mtx_lock(&ctrlr->lock); } mtx_unlock(&ctrlr->lock); } /* * Wait for RDY to change. * * Starts sleeping for 1us and geometrically increases it the longer we wait, * capped at 1ms. */ static int nvme_ctrlr_wait_for_ready(struct nvme_controller *ctrlr, int desired_val) { int timeout = ticks + MSEC_2_TICKS(ctrlr->ready_timeout_in_ms); sbintime_t delta_t = SBT_1US; uint32_t csts; while (1) { csts = nvme_mmio_read_4(ctrlr, csts); if (csts == NVME_GONE) /* Hot unplug. */ return (ENXIO); - if (((csts >> NVME_CSTS_REG_RDY_SHIFT) & NVME_CSTS_REG_RDY_MASK) - == desired_val) + if (NVMEV(NVME_CSTS_REG_RDY, csts) == desired_val) break; if (timeout - ticks < 0) { nvme_printf(ctrlr, "controller ready did not become %d " "within %d ms\n", desired_val, ctrlr->ready_timeout_in_ms); return (ENXIO); } pause_sbt("nvmerdy", delta_t, 0, C_PREL(1)); delta_t = min(SBT_1MS, delta_t * 3 / 2); } return (0); } static int nvme_ctrlr_disable(struct nvme_controller *ctrlr) { uint32_t cc; uint32_t csts; uint8_t en, rdy; int err; cc = nvme_mmio_read_4(ctrlr, cc); csts = nvme_mmio_read_4(ctrlr, csts); - en = (cc >> NVME_CC_REG_EN_SHIFT) & NVME_CC_REG_EN_MASK; - rdy = (csts >> NVME_CSTS_REG_RDY_SHIFT) & NVME_CSTS_REG_RDY_MASK; + en = NVMEV(NVME_CC_REG_EN, cc); + rdy = NVMEV(NVME_CSTS_REG_RDY, csts); /* * Per 3.1.5 in NVME 1.3 spec, transitioning CC.EN from 0 to 1 * when CSTS.RDY is 1 or transitioning CC.EN from 1 to 0 when * CSTS.RDY is 0 "has undefined results" So make sure that CSTS.RDY * isn't the desired value. Short circuit if we're already disabled. */ if (en == 0) { /* Wait for RDY == 0 or timeout & fail */ if (rdy == 0) return (0); return (nvme_ctrlr_wait_for_ready(ctrlr, 0)); } if (rdy == 0) { /* EN == 1, wait for RDY == 1 or timeout & fail */ err = nvme_ctrlr_wait_for_ready(ctrlr, 1); if (err != 0) return (err); } cc &= ~NVME_CC_REG_EN_MASK; nvme_mmio_write_4(ctrlr, cc, cc); /* * A few drives have firmware bugs that freeze the drive if we access * the mmio too soon after we disable. */ if (ctrlr->quirks & QUIRK_DELAY_B4_CHK_RDY) pause("nvmeR", MSEC_2_TICKS(B4_CHK_RDY_DELAY_MS)); return (nvme_ctrlr_wait_for_ready(ctrlr, 0)); } static int nvme_ctrlr_enable(struct nvme_controller *ctrlr) { uint32_t cc; uint32_t csts; uint32_t aqa; uint32_t qsize; uint8_t en, rdy; int err; cc = nvme_mmio_read_4(ctrlr, cc); csts = nvme_mmio_read_4(ctrlr, csts); - en = (cc >> NVME_CC_REG_EN_SHIFT) & NVME_CC_REG_EN_MASK; - rdy = (csts >> NVME_CSTS_REG_RDY_SHIFT) & NVME_CSTS_REG_RDY_MASK; + en = NVMEV(NVME_CC_REG_EN, cc); + rdy = NVMEV(NVME_CSTS_REG_RDY, csts); /* * See note in nvme_ctrlr_disable. Short circuit if we're already enabled. */ if (en == 1) { if (rdy == 1) return (0); return (nvme_ctrlr_wait_for_ready(ctrlr, 1)); } /* EN == 0 already wait for RDY == 0 or timeout & fail */ err = nvme_ctrlr_wait_for_ready(ctrlr, 0); if (err != 0) return (err); nvme_mmio_write_8(ctrlr, asq, ctrlr->adminq.cmd_bus_addr); nvme_mmio_write_8(ctrlr, acq, ctrlr->adminq.cpl_bus_addr); /* acqs and asqs are 0-based. */ qsize = ctrlr->adminq.num_entries - 1; aqa = 0; aqa = (qsize & NVME_AQA_REG_ACQS_MASK) << NVME_AQA_REG_ACQS_SHIFT; aqa |= (qsize & NVME_AQA_REG_ASQS_MASK) << NVME_AQA_REG_ASQS_SHIFT; nvme_mmio_write_4(ctrlr, aqa, aqa); /* Initialization values for CC */ cc = 0; cc |= 1 << NVME_CC_REG_EN_SHIFT; cc |= 0 << NVME_CC_REG_CSS_SHIFT; cc |= 0 << NVME_CC_REG_AMS_SHIFT; cc |= 0 << NVME_CC_REG_SHN_SHIFT; cc |= 6 << NVME_CC_REG_IOSQES_SHIFT; /* SQ entry size == 64 == 2^6 */ cc |= 4 << NVME_CC_REG_IOCQES_SHIFT; /* CQ entry size == 16 == 2^4 */ /* * Use the Memory Page Size selected during device initialization. Note * that value stored in mps is suitable to use here without adjusting by * NVME_MPS_SHIFT. */ cc |= ctrlr->mps << NVME_CC_REG_MPS_SHIFT; nvme_ctrlr_barrier(ctrlr, BUS_SPACE_BARRIER_WRITE); nvme_mmio_write_4(ctrlr, cc, cc); return (nvme_ctrlr_wait_for_ready(ctrlr, 1)); } static void nvme_ctrlr_disable_qpairs(struct nvme_controller *ctrlr) { int i; nvme_admin_qpair_disable(&ctrlr->adminq); /* * I/O queues are not allocated before the initial HW * reset, so do not try to disable them. Use is_initialized * to determine if this is the initial HW reset. */ if (ctrlr->is_initialized) { for (i = 0; i < ctrlr->num_io_queues; i++) nvme_io_qpair_disable(&ctrlr->ioq[i]); } } static int nvme_ctrlr_hw_reset(struct nvme_controller *ctrlr) { int err; TSENTER(); nvme_ctrlr_disable_qpairs(ctrlr); err = nvme_ctrlr_disable(ctrlr); if (err != 0) goto out; err = nvme_ctrlr_enable(ctrlr); out: TSEXIT(); return (err); } void nvme_ctrlr_reset(struct nvme_controller *ctrlr) { int cmpset; cmpset = atomic_cmpset_32(&ctrlr->is_resetting, 0, 1); if (cmpset == 0 || ctrlr->is_failed) /* * Controller is already resetting or has failed. Return * immediately since there is no need to kick off another * reset in these cases. */ return; if (!ctrlr->is_dying) taskqueue_enqueue(ctrlr->taskqueue, &ctrlr->reset_task); } static int nvme_ctrlr_identify(struct nvme_controller *ctrlr) { struct nvme_completion_poll_status status; status.done = 0; nvme_ctrlr_cmd_identify_controller(ctrlr, &ctrlr->cdata, nvme_completion_poll_cb, &status); nvme_completion_poll(&status); if (nvme_completion_is_error(&status.cpl)) { nvme_printf(ctrlr, "nvme_identify_controller failed!\n"); return (ENXIO); } /* Convert data to host endian */ nvme_controller_data_swapbytes(&ctrlr->cdata); /* * Use MDTS to ensure our default max_xfer_size doesn't exceed what the * controller supports. */ if (ctrlr->cdata.mdts > 0) ctrlr->max_xfer_size = min(ctrlr->max_xfer_size, 1 << (ctrlr->cdata.mdts + NVME_MPS_SHIFT + NVME_CAP_HI_MPSMIN(ctrlr->cap_hi))); return (0); } static int nvme_ctrlr_set_num_qpairs(struct nvme_controller *ctrlr) { struct nvme_completion_poll_status status; int cq_allocated, sq_allocated; status.done = 0; nvme_ctrlr_cmd_set_num_queues(ctrlr, ctrlr->num_io_queues, nvme_completion_poll_cb, &status); nvme_completion_poll(&status); if (nvme_completion_is_error(&status.cpl)) { nvme_printf(ctrlr, "nvme_ctrlr_set_num_qpairs failed!\n"); return (ENXIO); } /* * Data in cdw0 is 0-based. * Lower 16-bits indicate number of submission queues allocated. * Upper 16-bits indicate number of completion queues allocated. */ sq_allocated = (status.cpl.cdw0 & 0xFFFF) + 1; cq_allocated = (status.cpl.cdw0 >> 16) + 1; /* * Controller may allocate more queues than we requested, * so use the minimum of the number requested and what was * actually allocated. */ ctrlr->num_io_queues = min(ctrlr->num_io_queues, sq_allocated); ctrlr->num_io_queues = min(ctrlr->num_io_queues, cq_allocated); if (ctrlr->num_io_queues > vm_ndomains) ctrlr->num_io_queues -= ctrlr->num_io_queues % vm_ndomains; return (0); } static int nvme_ctrlr_create_qpairs(struct nvme_controller *ctrlr) { struct nvme_completion_poll_status status; struct nvme_qpair *qpair; int i; for (i = 0; i < ctrlr->num_io_queues; i++) { qpair = &ctrlr->ioq[i]; status.done = 0; nvme_ctrlr_cmd_create_io_cq(ctrlr, qpair, nvme_completion_poll_cb, &status); nvme_completion_poll(&status); if (nvme_completion_is_error(&status.cpl)) { nvme_printf(ctrlr, "nvme_create_io_cq failed!\n"); return (ENXIO); } status.done = 0; nvme_ctrlr_cmd_create_io_sq(ctrlr, qpair, nvme_completion_poll_cb, &status); nvme_completion_poll(&status); if (nvme_completion_is_error(&status.cpl)) { nvme_printf(ctrlr, "nvme_create_io_sq failed!\n"); return (ENXIO); } } return (0); } static int nvme_ctrlr_delete_qpairs(struct nvme_controller *ctrlr) { struct nvme_completion_poll_status status; struct nvme_qpair *qpair; for (int i = 0; i < ctrlr->num_io_queues; i++) { qpair = &ctrlr->ioq[i]; status.done = 0; nvme_ctrlr_cmd_delete_io_sq(ctrlr, qpair, nvme_completion_poll_cb, &status); nvme_completion_poll(&status); if (nvme_completion_is_error(&status.cpl)) { nvme_printf(ctrlr, "nvme_destroy_io_sq failed!\n"); return (ENXIO); } status.done = 0; nvme_ctrlr_cmd_delete_io_cq(ctrlr, qpair, nvme_completion_poll_cb, &status); nvme_completion_poll(&status); if (nvme_completion_is_error(&status.cpl)) { nvme_printf(ctrlr, "nvme_destroy_io_cq failed!\n"); return (ENXIO); } } return (0); } static int nvme_ctrlr_construct_namespaces(struct nvme_controller *ctrlr) { struct nvme_namespace *ns; uint32_t i; for (i = 0; i < min(ctrlr->cdata.nn, NVME_MAX_NAMESPACES); i++) { ns = &ctrlr->ns[i]; nvme_ns_construct(ns, i+1, ctrlr); } return (0); } static bool is_log_page_id_valid(uint8_t page_id) { switch (page_id) { case NVME_LOG_ERROR: case NVME_LOG_HEALTH_INFORMATION: case NVME_LOG_FIRMWARE_SLOT: case NVME_LOG_CHANGED_NAMESPACE: case NVME_LOG_COMMAND_EFFECT: case NVME_LOG_RES_NOTIFICATION: case NVME_LOG_SANITIZE_STATUS: return (true); } return (false); } static uint32_t nvme_ctrlr_get_log_page_size(struct nvme_controller *ctrlr, uint8_t page_id) { uint32_t log_page_size; switch (page_id) { case NVME_LOG_ERROR: log_page_size = min( sizeof(struct nvme_error_information_entry) * (ctrlr->cdata.elpe + 1), NVME_MAX_AER_LOG_SIZE); break; case NVME_LOG_HEALTH_INFORMATION: log_page_size = sizeof(struct nvme_health_information_page); break; case NVME_LOG_FIRMWARE_SLOT: log_page_size = sizeof(struct nvme_firmware_page); break; case NVME_LOG_CHANGED_NAMESPACE: log_page_size = sizeof(struct nvme_ns_list); break; case NVME_LOG_COMMAND_EFFECT: log_page_size = sizeof(struct nvme_command_effects_page); break; case NVME_LOG_RES_NOTIFICATION: log_page_size = sizeof(struct nvme_res_notification_page); break; case NVME_LOG_SANITIZE_STATUS: log_page_size = sizeof(struct nvme_sanitize_status_page); break; default: log_page_size = 0; break; } return (log_page_size); } static void nvme_ctrlr_log_critical_warnings(struct nvme_controller *ctrlr, uint8_t state) { if (state & NVME_CRIT_WARN_ST_AVAILABLE_SPARE) nvme_ctrlr_devctl_log(ctrlr, "critical", "available spare space below threshold"); if (state & NVME_CRIT_WARN_ST_TEMPERATURE) nvme_ctrlr_devctl_log(ctrlr, "critical", "temperature above threshold"); if (state & NVME_CRIT_WARN_ST_DEVICE_RELIABILITY) nvme_ctrlr_devctl_log(ctrlr, "critical", "device reliability degraded"); if (state & NVME_CRIT_WARN_ST_READ_ONLY) nvme_ctrlr_devctl_log(ctrlr, "critical", "media placed in read only mode"); if (state & NVME_CRIT_WARN_ST_VOLATILE_MEMORY_BACKUP) nvme_ctrlr_devctl_log(ctrlr, "critical", "volatile memory backup device failed"); if (state & NVME_CRIT_WARN_ST_RESERVED_MASK) nvme_ctrlr_devctl_log(ctrlr, "critical", "unknown critical warning(s): state = 0x%02x", state); } static void nvme_ctrlr_async_event_log_page_cb(void *arg, const struct nvme_completion *cpl) { struct nvme_async_event_request *aer = arg; struct nvme_health_information_page *health_info; struct nvme_ns_list *nsl; struct nvme_error_information_entry *err; int i; /* * If the log page fetch for some reason completed with an error, * don't pass log page data to the consumers. In practice, this case * should never happen. */ if (nvme_completion_is_error(cpl)) nvme_notify_async_consumers(aer->ctrlr, &aer->cpl, aer->log_page_id, NULL, 0); else { /* Convert data to host endian */ switch (aer->log_page_id) { case NVME_LOG_ERROR: err = (struct nvme_error_information_entry *)aer->log_page_buffer; for (i = 0; i < (aer->ctrlr->cdata.elpe + 1); i++) nvme_error_information_entry_swapbytes(err++); break; case NVME_LOG_HEALTH_INFORMATION: nvme_health_information_page_swapbytes( (struct nvme_health_information_page *)aer->log_page_buffer); break; case NVME_LOG_CHANGED_NAMESPACE: nvme_ns_list_swapbytes( (struct nvme_ns_list *)aer->log_page_buffer); break; case NVME_LOG_COMMAND_EFFECT: nvme_command_effects_page_swapbytes( (struct nvme_command_effects_page *)aer->log_page_buffer); break; case NVME_LOG_RES_NOTIFICATION: nvme_res_notification_page_swapbytes( (struct nvme_res_notification_page *)aer->log_page_buffer); break; case NVME_LOG_SANITIZE_STATUS: nvme_sanitize_status_page_swapbytes( (struct nvme_sanitize_status_page *)aer->log_page_buffer); break; case INTEL_LOG_TEMP_STATS: intel_log_temp_stats_swapbytes( (struct intel_log_temp_stats *)aer->log_page_buffer); break; default: break; } if (aer->log_page_id == NVME_LOG_HEALTH_INFORMATION) { health_info = (struct nvme_health_information_page *) aer->log_page_buffer; nvme_ctrlr_log_critical_warnings(aer->ctrlr, health_info->critical_warning); /* * Critical warnings reported through the * SMART/health log page are persistent, so * clear the associated bits in the async event * config so that we do not receive repeated * notifications for the same event. */ aer->ctrlr->async_event_config &= ~health_info->critical_warning; nvme_ctrlr_cmd_set_async_event_config(aer->ctrlr, aer->ctrlr->async_event_config, NULL, NULL); } else if (aer->log_page_id == NVME_LOG_CHANGED_NAMESPACE && !nvme_use_nvd) { nsl = (struct nvme_ns_list *)aer->log_page_buffer; for (i = 0; i < nitems(nsl->ns) && nsl->ns[i] != 0; i++) { if (nsl->ns[i] > NVME_MAX_NAMESPACES) break; nvme_notify_ns(aer->ctrlr, nsl->ns[i]); } } /* * Pass the cpl data from the original async event completion, * not the log page fetch. */ nvme_notify_async_consumers(aer->ctrlr, &aer->cpl, aer->log_page_id, aer->log_page_buffer, aer->log_page_size); } /* * Repost another asynchronous event request to replace the one * that just completed. */ nvme_ctrlr_construct_and_submit_aer(aer->ctrlr, aer); } static void nvme_ctrlr_async_event_cb(void *arg, const struct nvme_completion *cpl) { struct nvme_async_event_request *aer = arg; if (nvme_completion_is_error(cpl)) { /* * Do not retry failed async event requests. This avoids * infinite loops where a new async event request is submitted * to replace the one just failed, only to fail again and * perpetuate the loop. */ return; } /* Associated log page is in bits 23:16 of completion entry dw0. */ aer->log_page_id = (cpl->cdw0 & 0xFF0000) >> 16; nvme_printf(aer->ctrlr, "async event occurred (type 0x%x, info 0x%02x," " page 0x%02x)\n", (cpl->cdw0 & 0x07), (cpl->cdw0 & 0xFF00) >> 8, aer->log_page_id); if (is_log_page_id_valid(aer->log_page_id)) { aer->log_page_size = nvme_ctrlr_get_log_page_size(aer->ctrlr, aer->log_page_id); memcpy(&aer->cpl, cpl, sizeof(*cpl)); nvme_ctrlr_cmd_get_log_page(aer->ctrlr, aer->log_page_id, NVME_GLOBAL_NAMESPACE_TAG, aer->log_page_buffer, aer->log_page_size, nvme_ctrlr_async_event_log_page_cb, aer); /* Wait to notify consumers until after log page is fetched. */ } else { nvme_notify_async_consumers(aer->ctrlr, cpl, aer->log_page_id, NULL, 0); /* * Repost another asynchronous event request to replace the one * that just completed. */ nvme_ctrlr_construct_and_submit_aer(aer->ctrlr, aer); } } static void nvme_ctrlr_construct_and_submit_aer(struct nvme_controller *ctrlr, struct nvme_async_event_request *aer) { struct nvme_request *req; aer->ctrlr = ctrlr; req = nvme_allocate_request_null(nvme_ctrlr_async_event_cb, aer); aer->req = req; /* * Disable timeout here, since asynchronous event requests should by * nature never be timed out. */ req->timeout = false; req->cmd.opc = NVME_OPC_ASYNC_EVENT_REQUEST; nvme_ctrlr_submit_admin_request(ctrlr, req); } static void nvme_ctrlr_configure_aer(struct nvme_controller *ctrlr) { struct nvme_completion_poll_status status; struct nvme_async_event_request *aer; uint32_t i; ctrlr->async_event_config = NVME_CRIT_WARN_ST_AVAILABLE_SPARE | NVME_CRIT_WARN_ST_DEVICE_RELIABILITY | NVME_CRIT_WARN_ST_READ_ONLY | NVME_CRIT_WARN_ST_VOLATILE_MEMORY_BACKUP; if (ctrlr->cdata.ver >= NVME_REV(1, 2)) ctrlr->async_event_config |= NVME_ASYNC_EVENT_NS_ATTRIBUTE | NVME_ASYNC_EVENT_FW_ACTIVATE; status.done = 0; nvme_ctrlr_cmd_get_feature(ctrlr, NVME_FEAT_TEMPERATURE_THRESHOLD, 0, NULL, 0, nvme_completion_poll_cb, &status); nvme_completion_poll(&status); if (nvme_completion_is_error(&status.cpl) || (status.cpl.cdw0 & 0xFFFF) == 0xFFFF || (status.cpl.cdw0 & 0xFFFF) == 0x0000) { nvme_printf(ctrlr, "temperature threshold not supported\n"); } else ctrlr->async_event_config |= NVME_CRIT_WARN_ST_TEMPERATURE; nvme_ctrlr_cmd_set_async_event_config(ctrlr, ctrlr->async_event_config, NULL, NULL); /* aerl is a zero-based value, so we need to add 1 here. */ ctrlr->num_aers = min(NVME_MAX_ASYNC_EVENTS, (ctrlr->cdata.aerl+1)); for (i = 0; i < ctrlr->num_aers; i++) { aer = &ctrlr->aer[i]; nvme_ctrlr_construct_and_submit_aer(ctrlr, aer); } } static void nvme_ctrlr_configure_int_coalescing(struct nvme_controller *ctrlr) { ctrlr->int_coal_time = 0; TUNABLE_INT_FETCH("hw.nvme.int_coal_time", &ctrlr->int_coal_time); ctrlr->int_coal_threshold = 0; TUNABLE_INT_FETCH("hw.nvme.int_coal_threshold", &ctrlr->int_coal_threshold); nvme_ctrlr_cmd_set_interrupt_coalescing(ctrlr, ctrlr->int_coal_time, ctrlr->int_coal_threshold, NULL, NULL); } static void nvme_ctrlr_hmb_free(struct nvme_controller *ctrlr) { struct nvme_hmb_chunk *hmbc; int i; if (ctrlr->hmb_desc_paddr) { bus_dmamap_unload(ctrlr->hmb_desc_tag, ctrlr->hmb_desc_map); bus_dmamem_free(ctrlr->hmb_desc_tag, ctrlr->hmb_desc_vaddr, ctrlr->hmb_desc_map); ctrlr->hmb_desc_paddr = 0; } if (ctrlr->hmb_desc_tag) { bus_dma_tag_destroy(ctrlr->hmb_desc_tag); ctrlr->hmb_desc_tag = NULL; } for (i = 0; i < ctrlr->hmb_nchunks; i++) { hmbc = &ctrlr->hmb_chunks[i]; bus_dmamap_unload(ctrlr->hmb_tag, hmbc->hmbc_map); bus_dmamem_free(ctrlr->hmb_tag, hmbc->hmbc_vaddr, hmbc->hmbc_map); } ctrlr->hmb_nchunks = 0; if (ctrlr->hmb_tag) { bus_dma_tag_destroy(ctrlr->hmb_tag); ctrlr->hmb_tag = NULL; } if (ctrlr->hmb_chunks) { free(ctrlr->hmb_chunks, M_NVME); ctrlr->hmb_chunks = NULL; } } static void nvme_ctrlr_hmb_alloc(struct nvme_controller *ctrlr) { struct nvme_hmb_chunk *hmbc; size_t pref, min, minc, size; int err, i; uint64_t max; /* Limit HMB to 5% of RAM size per device by default. */ max = (uint64_t)physmem * PAGE_SIZE / 20; TUNABLE_UINT64_FETCH("hw.nvme.hmb_max", &max); /* * Units of Host Memory Buffer in the Identify info are always in terms * of 4k units. */ min = (long long unsigned)ctrlr->cdata.hmmin * NVME_HMB_UNITS; if (max == 0 || max < min) return; pref = MIN((long long unsigned)ctrlr->cdata.hmpre * NVME_HMB_UNITS, max); minc = MAX(ctrlr->cdata.hmminds * NVME_HMB_UNITS, ctrlr->page_size); if (min > 0 && ctrlr->cdata.hmmaxd > 0) minc = MAX(minc, min / ctrlr->cdata.hmmaxd); ctrlr->hmb_chunk = pref; again: /* * However, the chunk sizes, number of chunks, and alignment of chunks * are all based on the current MPS (ctrlr->page_size). */ ctrlr->hmb_chunk = roundup2(ctrlr->hmb_chunk, ctrlr->page_size); ctrlr->hmb_nchunks = howmany(pref, ctrlr->hmb_chunk); if (ctrlr->cdata.hmmaxd > 0 && ctrlr->hmb_nchunks > ctrlr->cdata.hmmaxd) ctrlr->hmb_nchunks = ctrlr->cdata.hmmaxd; ctrlr->hmb_chunks = malloc(sizeof(struct nvme_hmb_chunk) * ctrlr->hmb_nchunks, M_NVME, M_WAITOK); err = bus_dma_tag_create(bus_get_dma_tag(ctrlr->dev), ctrlr->page_size, 0, BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL, ctrlr->hmb_chunk, 1, ctrlr->hmb_chunk, 0, NULL, NULL, &ctrlr->hmb_tag); if (err != 0) { nvme_printf(ctrlr, "HMB tag create failed %d\n", err); nvme_ctrlr_hmb_free(ctrlr); return; } for (i = 0; i < ctrlr->hmb_nchunks; i++) { hmbc = &ctrlr->hmb_chunks[i]; if (bus_dmamem_alloc(ctrlr->hmb_tag, (void **)&hmbc->hmbc_vaddr, BUS_DMA_NOWAIT, &hmbc->hmbc_map)) { nvme_printf(ctrlr, "failed to alloc HMB\n"); break; } if (bus_dmamap_load(ctrlr->hmb_tag, hmbc->hmbc_map, hmbc->hmbc_vaddr, ctrlr->hmb_chunk, nvme_single_map, &hmbc->hmbc_paddr, BUS_DMA_NOWAIT) != 0) { bus_dmamem_free(ctrlr->hmb_tag, hmbc->hmbc_vaddr, hmbc->hmbc_map); nvme_printf(ctrlr, "failed to load HMB\n"); break; } bus_dmamap_sync(ctrlr->hmb_tag, hmbc->hmbc_map, BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE); } if (i < ctrlr->hmb_nchunks && i * ctrlr->hmb_chunk < min && ctrlr->hmb_chunk / 2 >= minc) { ctrlr->hmb_nchunks = i; nvme_ctrlr_hmb_free(ctrlr); ctrlr->hmb_chunk /= 2; goto again; } ctrlr->hmb_nchunks = i; if (ctrlr->hmb_nchunks * ctrlr->hmb_chunk < min) { nvme_ctrlr_hmb_free(ctrlr); return; } size = sizeof(struct nvme_hmb_desc) * ctrlr->hmb_nchunks; err = bus_dma_tag_create(bus_get_dma_tag(ctrlr->dev), 16, 0, BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL, size, 1, size, 0, NULL, NULL, &ctrlr->hmb_desc_tag); if (err != 0) { nvme_printf(ctrlr, "HMB desc tag create failed %d\n", err); nvme_ctrlr_hmb_free(ctrlr); return; } if (bus_dmamem_alloc(ctrlr->hmb_desc_tag, (void **)&ctrlr->hmb_desc_vaddr, BUS_DMA_WAITOK, &ctrlr->hmb_desc_map)) { nvme_printf(ctrlr, "failed to alloc HMB desc\n"); nvme_ctrlr_hmb_free(ctrlr); return; } if (bus_dmamap_load(ctrlr->hmb_desc_tag, ctrlr->hmb_desc_map, ctrlr->hmb_desc_vaddr, size, nvme_single_map, &ctrlr->hmb_desc_paddr, BUS_DMA_NOWAIT) != 0) { bus_dmamem_free(ctrlr->hmb_desc_tag, ctrlr->hmb_desc_vaddr, ctrlr->hmb_desc_map); nvme_printf(ctrlr, "failed to load HMB desc\n"); nvme_ctrlr_hmb_free(ctrlr); return; } for (i = 0; i < ctrlr->hmb_nchunks; i++) { memset(&ctrlr->hmb_desc_vaddr[i], 0, sizeof(struct nvme_hmb_desc)); ctrlr->hmb_desc_vaddr[i].addr = htole64(ctrlr->hmb_chunks[i].hmbc_paddr); ctrlr->hmb_desc_vaddr[i].size = htole32(ctrlr->hmb_chunk / ctrlr->page_size); } bus_dmamap_sync(ctrlr->hmb_desc_tag, ctrlr->hmb_desc_map, BUS_DMASYNC_PREWRITE); nvme_printf(ctrlr, "Allocated %lluMB host memory buffer\n", (long long unsigned)ctrlr->hmb_nchunks * ctrlr->hmb_chunk / 1024 / 1024); } static void nvme_ctrlr_hmb_enable(struct nvme_controller *ctrlr, bool enable, bool memret) { struct nvme_completion_poll_status status; uint32_t cdw11; cdw11 = 0; if (enable) cdw11 |= 1; if (memret) cdw11 |= 2; status.done = 0; nvme_ctrlr_cmd_set_feature(ctrlr, NVME_FEAT_HOST_MEMORY_BUFFER, cdw11, ctrlr->hmb_nchunks * ctrlr->hmb_chunk / ctrlr->page_size, ctrlr->hmb_desc_paddr, ctrlr->hmb_desc_paddr >> 32, ctrlr->hmb_nchunks, NULL, 0, nvme_completion_poll_cb, &status); nvme_completion_poll(&status); if (nvme_completion_is_error(&status.cpl)) nvme_printf(ctrlr, "nvme_ctrlr_hmb_enable failed!\n"); } static void nvme_ctrlr_start(void *ctrlr_arg, bool resetting) { struct nvme_controller *ctrlr = ctrlr_arg; uint32_t old_num_io_queues; int i; TSENTER(); /* * Only reset adminq here when we are restarting the * controller after a reset. During initialization, * we have already submitted admin commands to get * the number of I/O queues supported, so cannot reset * the adminq again here. */ if (resetting) { nvme_qpair_reset(&ctrlr->adminq); nvme_admin_qpair_enable(&ctrlr->adminq); } if (ctrlr->ioq != NULL) { for (i = 0; i < ctrlr->num_io_queues; i++) nvme_qpair_reset(&ctrlr->ioq[i]); } /* * If it was a reset on initialization command timeout, just * return here, letting initialization code fail gracefully. */ if (resetting && !ctrlr->is_initialized) return; if (resetting && nvme_ctrlr_identify(ctrlr) != 0) { nvme_ctrlr_fail(ctrlr); return; } /* * The number of qpairs are determined during controller initialization, * including using NVMe SET_FEATURES/NUMBER_OF_QUEUES to determine the * HW limit. We call SET_FEATURES again here so that it gets called * after any reset for controllers that depend on the driver to * explicit specify how many queues it will use. This value should * never change between resets, so panic if somehow that does happen. */ if (resetting) { old_num_io_queues = ctrlr->num_io_queues; if (nvme_ctrlr_set_num_qpairs(ctrlr) != 0) { nvme_ctrlr_fail(ctrlr); return; } if (old_num_io_queues != ctrlr->num_io_queues) { panic("num_io_queues changed from %u to %u", old_num_io_queues, ctrlr->num_io_queues); } } if (ctrlr->cdata.hmpre > 0 && ctrlr->hmb_nchunks == 0) { nvme_ctrlr_hmb_alloc(ctrlr); if (ctrlr->hmb_nchunks > 0) nvme_ctrlr_hmb_enable(ctrlr, true, false); } else if (ctrlr->hmb_nchunks > 0) nvme_ctrlr_hmb_enable(ctrlr, true, true); if (nvme_ctrlr_create_qpairs(ctrlr) != 0) { nvme_ctrlr_fail(ctrlr); return; } if (nvme_ctrlr_construct_namespaces(ctrlr) != 0) { nvme_ctrlr_fail(ctrlr); return; } nvme_ctrlr_configure_aer(ctrlr); nvme_ctrlr_configure_int_coalescing(ctrlr); for (i = 0; i < ctrlr->num_io_queues; i++) nvme_io_qpair_enable(&ctrlr->ioq[i]); TSEXIT(); } void nvme_ctrlr_start_config_hook(void *arg) { struct nvme_controller *ctrlr = arg; TSENTER(); /* * Don't call pre/post reset here. We've not yet created the qpairs, * haven't setup the ISRs, so there's no need to 'drain' them or * 'exclude' them. */ if (nvme_ctrlr_hw_reset(ctrlr) != 0) { fail: nvme_ctrlr_fail(ctrlr); config_intrhook_disestablish(&ctrlr->config_hook); return; } #ifdef NVME_2X_RESET /* * Reset controller twice to ensure we do a transition from cc.en==1 to * cc.en==0. This is because we don't really know what status the * controller was left in when boot handed off to OS. Linux doesn't do * this, however, and when the controller is in state cc.en == 0, no * I/O can happen. */ if (nvme_ctrlr_hw_reset(ctrlr) != 0) goto fail; #endif nvme_qpair_reset(&ctrlr->adminq); nvme_admin_qpair_enable(&ctrlr->adminq); if (nvme_ctrlr_identify(ctrlr) == 0 && nvme_ctrlr_set_num_qpairs(ctrlr) == 0 && nvme_ctrlr_construct_io_qpairs(ctrlr) == 0) nvme_ctrlr_start(ctrlr, false); else goto fail; nvme_sysctl_initialize_ctrlr(ctrlr); config_intrhook_disestablish(&ctrlr->config_hook); ctrlr->is_initialized = 1; nvme_notify_new_controller(ctrlr); TSEXIT(); } static void nvme_ctrlr_reset_task(void *arg, int pending) { struct nvme_controller *ctrlr = arg; int status; nvme_ctrlr_devctl_log(ctrlr, "RESET", "resetting controller"); status = nvme_ctrlr_hw_reset(ctrlr); if (status == 0) nvme_ctrlr_start(ctrlr, true); else nvme_ctrlr_fail(ctrlr); atomic_cmpset_32(&ctrlr->is_resetting, 1, 0); } /* * Poll all the queues enabled on the device for completion. */ void nvme_ctrlr_poll(struct nvme_controller *ctrlr) { int i; nvme_qpair_process_completions(&ctrlr->adminq); for (i = 0; i < ctrlr->num_io_queues; i++) if (ctrlr->ioq && ctrlr->ioq[i].cpl) nvme_qpair_process_completions(&ctrlr->ioq[i]); } /* * Poll the single-vector interrupt case: num_io_queues will be 1 and * there's only a single vector. While we're polling, we mask further * interrupts in the controller. */ void nvme_ctrlr_shared_handler(void *arg) { struct nvme_controller *ctrlr = arg; nvme_mmio_write_4(ctrlr, intms, 1); nvme_ctrlr_poll(ctrlr); nvme_mmio_write_4(ctrlr, intmc, 1); } static void nvme_pt_done(void *arg, const struct nvme_completion *cpl) { struct nvme_pt_command *pt = arg; struct mtx *mtx = pt->driver_lock; uint16_t status; bzero(&pt->cpl, sizeof(pt->cpl)); pt->cpl.cdw0 = cpl->cdw0; status = cpl->status; status &= ~NVME_STATUS_P_MASK; pt->cpl.status = status; mtx_lock(mtx); pt->driver_lock = NULL; wakeup(pt); mtx_unlock(mtx); } int nvme_ctrlr_passthrough_cmd(struct nvme_controller *ctrlr, struct nvme_pt_command *pt, uint32_t nsid, int is_user_buffer, int is_admin_cmd) { struct nvme_request *req; struct mtx *mtx; struct buf *buf = NULL; int ret = 0; if (pt->len > 0) { if (pt->len > ctrlr->max_xfer_size) { nvme_printf(ctrlr, "pt->len (%d) " "exceeds max_xfer_size (%d)\n", pt->len, ctrlr->max_xfer_size); return EIO; } if (is_user_buffer) { /* * Ensure the user buffer is wired for the duration of * this pass-through command. */ PHOLD(curproc); buf = uma_zalloc(pbuf_zone, M_WAITOK); buf->b_iocmd = pt->is_read ? BIO_READ : BIO_WRITE; if (vmapbuf(buf, pt->buf, pt->len, 1) < 0) { ret = EFAULT; goto err; } req = nvme_allocate_request_vaddr(buf->b_data, pt->len, nvme_pt_done, pt); } else req = nvme_allocate_request_vaddr(pt->buf, pt->len, nvme_pt_done, pt); } else req = nvme_allocate_request_null(nvme_pt_done, pt); /* Assume user space already converted to little-endian */ req->cmd.opc = pt->cmd.opc; req->cmd.fuse = pt->cmd.fuse; req->cmd.rsvd2 = pt->cmd.rsvd2; req->cmd.rsvd3 = pt->cmd.rsvd3; req->cmd.cdw10 = pt->cmd.cdw10; req->cmd.cdw11 = pt->cmd.cdw11; req->cmd.cdw12 = pt->cmd.cdw12; req->cmd.cdw13 = pt->cmd.cdw13; req->cmd.cdw14 = pt->cmd.cdw14; req->cmd.cdw15 = pt->cmd.cdw15; req->cmd.nsid = htole32(nsid); mtx = mtx_pool_find(mtxpool_sleep, pt); pt->driver_lock = mtx; if (is_admin_cmd) nvme_ctrlr_submit_admin_request(ctrlr, req); else nvme_ctrlr_submit_io_request(ctrlr, req); mtx_lock(mtx); while (pt->driver_lock != NULL) mtx_sleep(pt, mtx, PRIBIO, "nvme_pt", 0); mtx_unlock(mtx); if (buf != NULL) { vunmapbuf(buf); err: uma_zfree(pbuf_zone, buf); PRELE(curproc); } return (ret); } static int nvme_ctrlr_ioctl(struct cdev *cdev, u_long cmd, caddr_t arg, int flag, struct thread *td) { struct nvme_controller *ctrlr; struct nvme_pt_command *pt; ctrlr = cdev->si_drv1; switch (cmd) { case NVME_RESET_CONTROLLER: nvme_ctrlr_reset(ctrlr); break; case NVME_PASSTHROUGH_CMD: pt = (struct nvme_pt_command *)arg; return (nvme_ctrlr_passthrough_cmd(ctrlr, pt, le32toh(pt->cmd.nsid), 1 /* is_user_buffer */, 1 /* is_admin_cmd */)); case NVME_GET_NSID: { struct nvme_get_nsid *gnsid = (struct nvme_get_nsid *)arg; strncpy(gnsid->cdev, device_get_nameunit(ctrlr->dev), sizeof(gnsid->cdev)); gnsid->cdev[sizeof(gnsid->cdev) - 1] = '\0'; gnsid->nsid = 0; break; } case NVME_GET_MAX_XFER_SIZE: *(uint64_t *)arg = ctrlr->max_xfer_size; break; default: return (ENOTTY); } return (0); } static struct cdevsw nvme_ctrlr_cdevsw = { .d_version = D_VERSION, .d_flags = 0, .d_ioctl = nvme_ctrlr_ioctl }; int nvme_ctrlr_construct(struct nvme_controller *ctrlr, device_t dev) { struct make_dev_args md_args; uint32_t cap_lo; uint32_t cap_hi; uint32_t to, vs, pmrcap; int status, timeout_period; ctrlr->dev = dev; mtx_init(&ctrlr->lock, "nvme ctrlr lock", NULL, MTX_DEF); if (bus_get_domain(dev, &ctrlr->domain) != 0) ctrlr->domain = 0; ctrlr->cap_lo = cap_lo = nvme_mmio_read_4(ctrlr, cap_lo); if (bootverbose) { device_printf(dev, "CapLo: 0x%08x: MQES %u%s%s%s%s, TO %u\n", cap_lo, NVME_CAP_LO_MQES(cap_lo), NVME_CAP_LO_CQR(cap_lo) ? ", CQR" : "", NVME_CAP_LO_AMS(cap_lo) ? ", AMS" : "", (NVME_CAP_LO_AMS(cap_lo) & 0x1) ? " WRRwUPC" : "", (NVME_CAP_LO_AMS(cap_lo) & 0x2) ? " VS" : "", NVME_CAP_LO_TO(cap_lo)); } ctrlr->cap_hi = cap_hi = nvme_mmio_read_4(ctrlr, cap_hi); if (bootverbose) { device_printf(dev, "CapHi: 0x%08x: DSTRD %u%s, CSS %x%s, " "CPS %x, MPSMIN %u, MPSMAX %u%s%s%s%s%s\n", cap_hi, NVME_CAP_HI_DSTRD(cap_hi), NVME_CAP_HI_NSSRS(cap_hi) ? ", NSSRS" : "", NVME_CAP_HI_CSS(cap_hi), NVME_CAP_HI_BPS(cap_hi) ? ", BPS" : "", NVME_CAP_HI_CPS(cap_hi), NVME_CAP_HI_MPSMIN(cap_hi), NVME_CAP_HI_MPSMAX(cap_hi), NVME_CAP_HI_PMRS(cap_hi) ? ", PMRS" : "", NVME_CAP_HI_CMBS(cap_hi) ? ", CMBS" : "", NVME_CAP_HI_NSSS(cap_hi) ? ", NSSS" : "", NVME_CAP_HI_CRWMS(cap_hi) ? ", CRWMS" : "", NVME_CAP_HI_CRIMS(cap_hi) ? ", CRIMS" : ""); } if (bootverbose) { vs = nvme_mmio_read_4(ctrlr, vs); device_printf(dev, "Version: 0x%08x: %d.%d\n", vs, NVME_MAJOR(vs), NVME_MINOR(vs)); } if (bootverbose && NVME_CAP_HI_PMRS(cap_hi)) { pmrcap = nvme_mmio_read_4(ctrlr, pmrcap); device_printf(dev, "PMRCap: 0x%08x: BIR %u%s%s, PMRTU %u, " "PMRWBM %x, PMRTO %u%s\n", pmrcap, NVME_PMRCAP_BIR(pmrcap), NVME_PMRCAP_RDS(pmrcap) ? ", RDS" : "", NVME_PMRCAP_WDS(pmrcap) ? ", WDS" : "", NVME_PMRCAP_PMRTU(pmrcap), NVME_PMRCAP_PMRWBM(pmrcap), NVME_PMRCAP_PMRTO(pmrcap), NVME_PMRCAP_CMSS(pmrcap) ? ", CMSS" : ""); } ctrlr->dstrd = NVME_CAP_HI_DSTRD(cap_hi) + 2; ctrlr->mps = NVME_CAP_HI_MPSMIN(cap_hi); ctrlr->page_size = 1 << (NVME_MPS_SHIFT + ctrlr->mps); /* Get ready timeout value from controller, in units of 500ms. */ to = NVME_CAP_LO_TO(cap_lo) + 1; ctrlr->ready_timeout_in_ms = to * 500; timeout_period = NVME_ADMIN_TIMEOUT_PERIOD; TUNABLE_INT_FETCH("hw.nvme.admin_timeout_period", &timeout_period); timeout_period = min(timeout_period, NVME_MAX_TIMEOUT_PERIOD); timeout_period = max(timeout_period, NVME_MIN_TIMEOUT_PERIOD); ctrlr->admin_timeout_period = timeout_period; timeout_period = NVME_DEFAULT_TIMEOUT_PERIOD; TUNABLE_INT_FETCH("hw.nvme.timeout_period", &timeout_period); timeout_period = min(timeout_period, NVME_MAX_TIMEOUT_PERIOD); timeout_period = max(timeout_period, NVME_MIN_TIMEOUT_PERIOD); ctrlr->timeout_period = timeout_period; nvme_retry_count = NVME_DEFAULT_RETRY_COUNT; TUNABLE_INT_FETCH("hw.nvme.retry_count", &nvme_retry_count); ctrlr->enable_aborts = 0; TUNABLE_INT_FETCH("hw.nvme.enable_aborts", &ctrlr->enable_aborts); /* Cap transfers by the maximum addressable by page-sized PRP (4KB pages -> 2MB). */ ctrlr->max_xfer_size = MIN(maxphys, (ctrlr->page_size / 8 * ctrlr->page_size)); if (nvme_ctrlr_construct_admin_qpair(ctrlr) != 0) return (ENXIO); /* * Create 2 threads for the taskqueue. The reset thread will block when * it detects that the controller has failed until all I/O has been * failed up the stack. The fail_req task needs to be able to run in * this case to finish the request failure for some cases. * * We could partially solve this race by draining the failed requeust * queue before proceding to free the sim, though nothing would stop * new I/O from coming in after we do that drain, but before we reach * cam_sim_free, so this big hammer is used instead. */ ctrlr->taskqueue = taskqueue_create("nvme_taskq", M_WAITOK, taskqueue_thread_enqueue, &ctrlr->taskqueue); taskqueue_start_threads(&ctrlr->taskqueue, 2, PI_DISK, "nvme taskq"); ctrlr->is_resetting = 0; ctrlr->is_initialized = 0; ctrlr->notification_sent = 0; TASK_INIT(&ctrlr->reset_task, 0, nvme_ctrlr_reset_task, ctrlr); TASK_INIT(&ctrlr->fail_req_task, 0, nvme_ctrlr_fail_req_task, ctrlr); STAILQ_INIT(&ctrlr->fail_req); ctrlr->is_failed = false; make_dev_args_init(&md_args); md_args.mda_devsw = &nvme_ctrlr_cdevsw; md_args.mda_uid = UID_ROOT; md_args.mda_gid = GID_WHEEL; md_args.mda_mode = 0600; md_args.mda_unit = device_get_unit(dev); md_args.mda_si_drv1 = (void *)ctrlr; status = make_dev_s(&md_args, &ctrlr->cdev, "nvme%d", device_get_unit(dev)); if (status != 0) return (ENXIO); return (0); } void nvme_ctrlr_destruct(struct nvme_controller *ctrlr, device_t dev) { int gone, i; ctrlr->is_dying = true; if (ctrlr->resource == NULL) goto nores; if (!mtx_initialized(&ctrlr->adminq.lock)) goto noadminq; /* * Check whether it is a hot unplug or a clean driver detach. * If device is not there any more, skip any shutdown commands. */ gone = (nvme_mmio_read_4(ctrlr, csts) == NVME_GONE); if (gone) nvme_ctrlr_fail(ctrlr); else nvme_notify_fail_consumers(ctrlr); for (i = 0; i < NVME_MAX_NAMESPACES; i++) nvme_ns_destruct(&ctrlr->ns[i]); if (ctrlr->cdev) destroy_dev(ctrlr->cdev); if (ctrlr->is_initialized) { if (!gone) { if (ctrlr->hmb_nchunks > 0) nvme_ctrlr_hmb_enable(ctrlr, false, false); nvme_ctrlr_delete_qpairs(ctrlr); } nvme_ctrlr_hmb_free(ctrlr); } if (ctrlr->ioq != NULL) { for (i = 0; i < ctrlr->num_io_queues; i++) nvme_io_qpair_destroy(&ctrlr->ioq[i]); free(ctrlr->ioq, M_NVME); } nvme_admin_qpair_destroy(&ctrlr->adminq); /* * Notify the controller of a shutdown, even though this is due to * a driver unload, not a system shutdown (this path is not invoked * during shutdown). This ensures the controller receives a * shutdown notification in case the system is shutdown before * reloading the driver. */ if (!gone) nvme_ctrlr_shutdown(ctrlr); if (!gone) nvme_ctrlr_disable(ctrlr); noadminq: if (ctrlr->taskqueue) taskqueue_free(ctrlr->taskqueue); if (ctrlr->tag) bus_teardown_intr(ctrlr->dev, ctrlr->res, ctrlr->tag); if (ctrlr->res) bus_release_resource(ctrlr->dev, SYS_RES_IRQ, rman_get_rid(ctrlr->res), ctrlr->res); if (ctrlr->bar4_resource != NULL) { bus_release_resource(dev, SYS_RES_MEMORY, ctrlr->bar4_resource_id, ctrlr->bar4_resource); } bus_release_resource(dev, SYS_RES_MEMORY, ctrlr->resource_id, ctrlr->resource); nores: mtx_destroy(&ctrlr->lock); } void nvme_ctrlr_shutdown(struct nvme_controller *ctrlr) { uint32_t cc; uint32_t csts; int timeout; cc = nvme_mmio_read_4(ctrlr, cc); cc &= ~(NVME_CC_REG_SHN_MASK << NVME_CC_REG_SHN_SHIFT); cc |= NVME_SHN_NORMAL << NVME_CC_REG_SHN_SHIFT; nvme_mmio_write_4(ctrlr, cc, cc); timeout = ticks + (ctrlr->cdata.rtd3e == 0 ? 5 * hz : ((uint64_t)ctrlr->cdata.rtd3e * hz + 999999) / 1000000); while (1) { csts = nvme_mmio_read_4(ctrlr, csts); if (csts == NVME_GONE) /* Hot unplug. */ break; if (NVME_CSTS_GET_SHST(csts) == NVME_SHST_COMPLETE) break; if (timeout - ticks < 0) { nvme_printf(ctrlr, "shutdown timeout\n"); break; } pause("nvmeshut", 1); } } void nvme_ctrlr_submit_admin_request(struct nvme_controller *ctrlr, struct nvme_request *req) { nvme_qpair_submit_request(&ctrlr->adminq, req); } void nvme_ctrlr_submit_io_request(struct nvme_controller *ctrlr, struct nvme_request *req) { struct nvme_qpair *qpair; qpair = &ctrlr->ioq[QP(ctrlr, curcpu)]; nvme_qpair_submit_request(qpair, req); } device_t nvme_ctrlr_get_device(struct nvme_controller *ctrlr) { return (ctrlr->dev); } const struct nvme_controller_data * nvme_ctrlr_get_data(struct nvme_controller *ctrlr) { return (&ctrlr->cdata); } int nvme_ctrlr_suspend(struct nvme_controller *ctrlr) { int to = hz; /* * Can't touch failed controllers, so it's already suspended. */ if (ctrlr->is_failed) return (0); /* * We don't want the reset taskqueue running, since it does similar * things, so prevent it from running after we start. Wait for any reset * that may have been started to complete. The reset process we follow * will ensure that any new I/O will queue and be given to the hardware * after we resume (though there should be none). */ while (atomic_cmpset_32(&ctrlr->is_resetting, 0, 1) == 0 && to-- > 0) pause("nvmesusp", 1); if (to <= 0) { nvme_printf(ctrlr, "Competing reset task didn't finish. Try again later.\n"); return (EWOULDBLOCK); } if (ctrlr->hmb_nchunks > 0) nvme_ctrlr_hmb_enable(ctrlr, false, false); /* * Per Section 7.6.2 of NVMe spec 1.4, to properly suspend, we need to * delete the hardware I/O queues, and then shutdown. This properly * flushes any metadata the drive may have stored so it can survive * having its power removed and prevents the unsafe shutdown count from * incriminating. Once we delete the qpairs, we have to disable them * before shutting down. */ nvme_ctrlr_delete_qpairs(ctrlr); nvme_ctrlr_disable_qpairs(ctrlr); nvme_ctrlr_shutdown(ctrlr); return (0); } int nvme_ctrlr_resume(struct nvme_controller *ctrlr) { /* * Can't touch failed controllers, so nothing to do to resume. */ if (ctrlr->is_failed) return (0); if (nvme_ctrlr_hw_reset(ctrlr) != 0) goto fail; /* * Now that we've reset the hardware, we can restart the controller. Any * I/O that was pending is requeued. Any admin commands are aborted with * an error. Once we've restarted, take the controller out of reset. */ nvme_ctrlr_start(ctrlr, true); (void)atomic_cmpset_32(&ctrlr->is_resetting, 1, 0); return (0); fail: /* * Since we can't bring the controller out of reset, announce and fail * the controller. However, we have to return success for the resume * itself, due to questionable APIs. */ nvme_printf(ctrlr, "Failed to reset on resume, failing.\n"); nvme_ctrlr_fail(ctrlr); (void)atomic_cmpset_32(&ctrlr->is_resetting, 1, 0); return (0); } diff --git a/sys/dev/nvme/nvme_ns.c b/sys/dev/nvme/nvme_ns.c index 55f607c96e59..329b8d583898 100644 --- a/sys/dev/nvme/nvme_ns.c +++ b/sys/dev/nvme/nvme_ns.c @@ -1,627 +1,623 @@ /*- * SPDX-License-Identifier: BSD-2-Clause * * Copyright (C) 2012-2013 Intel Corporation * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ #include #include #include #include #include #include #include #include #include #include #include #include #include #include #include "nvme_private.h" static void nvme_bio_child_inbed(struct bio *parent, int bio_error); static void nvme_bio_child_done(void *arg, const struct nvme_completion *cpl); static uint32_t nvme_get_num_segments(uint64_t addr, uint64_t size, uint32_t alignment); static void nvme_free_child_bios(int num_bios, struct bio **child_bios); static struct bio ** nvme_allocate_child_bios(int num_bios); static struct bio ** nvme_construct_child_bios(struct bio *bp, uint32_t alignment, int *num_bios); static int nvme_ns_split_bio(struct nvme_namespace *ns, struct bio *bp, uint32_t alignment); static int nvme_ns_ioctl(struct cdev *cdev, u_long cmd, caddr_t arg, int flag, struct thread *td) { struct nvme_namespace *ns; struct nvme_controller *ctrlr; struct nvme_pt_command *pt; ns = cdev->si_drv1; ctrlr = ns->ctrlr; switch (cmd) { case NVME_IO_TEST: case NVME_BIO_TEST: nvme_ns_test(ns, cmd, arg); break; case NVME_PASSTHROUGH_CMD: pt = (struct nvme_pt_command *)arg; return (nvme_ctrlr_passthrough_cmd(ctrlr, pt, ns->id, 1 /* is_user_buffer */, 0 /* is_admin_cmd */)); case NVME_GET_NSID: { struct nvme_get_nsid *gnsid = (struct nvme_get_nsid *)arg; strncpy(gnsid->cdev, device_get_nameunit(ctrlr->dev), sizeof(gnsid->cdev)); gnsid->cdev[sizeof(gnsid->cdev) - 1] = '\0'; gnsid->nsid = ns->id; break; } case DIOCGMEDIASIZE: *(off_t *)arg = (off_t)nvme_ns_get_size(ns); break; case DIOCGSECTORSIZE: *(u_int *)arg = nvme_ns_get_sector_size(ns); break; default: return (ENOTTY); } return (0); } static int nvme_ns_open(struct cdev *dev __unused, int flags, int fmt __unused, struct thread *td) { int error = 0; if (flags & FWRITE) error = securelevel_gt(td->td_ucred, 0); return (error); } static int nvme_ns_close(struct cdev *dev __unused, int flags, int fmt __unused, struct thread *td) { return (0); } static void nvme_ns_strategy_done(void *arg, const struct nvme_completion *cpl) { struct bio *bp = arg; /* * TODO: add more extensive translation of NVMe status codes * to different bio error codes (i.e. EIO, EINVAL, etc.) */ if (nvme_completion_is_error(cpl)) { bp->bio_error = EIO; bp->bio_flags |= BIO_ERROR; bp->bio_resid = bp->bio_bcount; } else bp->bio_resid = 0; biodone(bp); } static void nvme_ns_strategy(struct bio *bp) { struct nvme_namespace *ns; int err; ns = bp->bio_dev->si_drv1; err = nvme_ns_bio_process(ns, bp, nvme_ns_strategy_done); if (err) { bp->bio_error = err; bp->bio_flags |= BIO_ERROR; bp->bio_resid = bp->bio_bcount; biodone(bp); } } static struct cdevsw nvme_ns_cdevsw = { .d_version = D_VERSION, .d_flags = D_DISK, .d_read = physread, .d_write = physwrite, .d_open = nvme_ns_open, .d_close = nvme_ns_close, .d_strategy = nvme_ns_strategy, .d_ioctl = nvme_ns_ioctl }; uint32_t nvme_ns_get_max_io_xfer_size(struct nvme_namespace *ns) { return ns->ctrlr->max_xfer_size; } uint32_t nvme_ns_get_sector_size(struct nvme_namespace *ns) { uint8_t flbas_fmt, lbads; - flbas_fmt = (ns->data.flbas >> NVME_NS_DATA_FLBAS_FORMAT_SHIFT) & - NVME_NS_DATA_FLBAS_FORMAT_MASK; - lbads = (ns->data.lbaf[flbas_fmt] >> NVME_NS_DATA_LBAF_LBADS_SHIFT) & - NVME_NS_DATA_LBAF_LBADS_MASK; + flbas_fmt = NVMEV(NVME_NS_DATA_FLBAS_FORMAT, ns->data.flbas); + lbads = NVMEV(NVME_NS_DATA_LBAF_LBADS, ns->data.lbaf[flbas_fmt]); return (1 << lbads); } uint64_t nvme_ns_get_num_sectors(struct nvme_namespace *ns) { return (ns->data.nsze); } uint64_t nvme_ns_get_size(struct nvme_namespace *ns) { return (nvme_ns_get_num_sectors(ns) * nvme_ns_get_sector_size(ns)); } uint32_t nvme_ns_get_flags(struct nvme_namespace *ns) { return (ns->flags); } const char * nvme_ns_get_serial_number(struct nvme_namespace *ns) { return ((const char *)ns->ctrlr->cdata.sn); } const char * nvme_ns_get_model_number(struct nvme_namespace *ns) { return ((const char *)ns->ctrlr->cdata.mn); } const struct nvme_namespace_data * nvme_ns_get_data(struct nvme_namespace *ns) { return (&ns->data); } uint32_t nvme_ns_get_stripesize(struct nvme_namespace *ns) { uint32_t ss; - if (((ns->data.nsfeat >> NVME_NS_DATA_NSFEAT_NPVALID_SHIFT) & - NVME_NS_DATA_NSFEAT_NPVALID_MASK) != 0) { + if (NVMEV(NVME_NS_DATA_NSFEAT_NPVALID, ns->data.nsfeat) != 0) { ss = nvme_ns_get_sector_size(ns); if (ns->data.npwa != 0) return ((ns->data.npwa + 1) * ss); else if (ns->data.npwg != 0) return ((ns->data.npwg + 1) * ss); } return (ns->boundary); } static void nvme_ns_bio_done(void *arg, const struct nvme_completion *status) { struct bio *bp = arg; nvme_cb_fn_t bp_cb_fn; bp_cb_fn = bp->bio_driver1; if (bp->bio_driver2) free(bp->bio_driver2, M_NVME); if (nvme_completion_is_error(status)) { bp->bio_flags |= BIO_ERROR; if (bp->bio_error == 0) bp->bio_error = EIO; } if ((bp->bio_flags & BIO_ERROR) == 0) bp->bio_resid = 0; else bp->bio_resid = bp->bio_bcount; bp_cb_fn(bp, status); } static void nvme_bio_child_inbed(struct bio *parent, int bio_error) { struct nvme_completion parent_cpl; int children, inbed; if (bio_error != 0) { parent->bio_flags |= BIO_ERROR; parent->bio_error = bio_error; } /* * atomic_fetchadd will return value before adding 1, so we still * must add 1 to get the updated inbed number. Save bio_children * before incrementing to guard against race conditions when * two children bios complete on different queues. */ children = atomic_load_acq_int(&parent->bio_children); inbed = atomic_fetchadd_int(&parent->bio_inbed, 1) + 1; if (inbed == children) { bzero(&parent_cpl, sizeof(parent_cpl)); if (parent->bio_flags & BIO_ERROR) { parent_cpl.status &= ~(NVME_STATUS_SC_MASK << NVME_STATUS_SC_SHIFT); parent_cpl.status |= (NVME_SC_DATA_TRANSFER_ERROR) << NVME_STATUS_SC_SHIFT; } nvme_ns_bio_done(parent, &parent_cpl); } } static void nvme_bio_child_done(void *arg, const struct nvme_completion *cpl) { struct bio *child = arg; struct bio *parent; int bio_error; parent = child->bio_parent; g_destroy_bio(child); bio_error = nvme_completion_is_error(cpl) ? EIO : 0; nvme_bio_child_inbed(parent, bio_error); } static uint32_t nvme_get_num_segments(uint64_t addr, uint64_t size, uint32_t align) { uint32_t num_segs, offset, remainder; if (align == 0) return (1); KASSERT((align & (align - 1)) == 0, ("alignment not power of 2\n")); num_segs = size / align; remainder = size & (align - 1); offset = addr & (align - 1); if (remainder > 0 || offset > 0) num_segs += 1 + (remainder + offset - 1) / align; return (num_segs); } static void nvme_free_child_bios(int num_bios, struct bio **child_bios) { int i; for (i = 0; i < num_bios; i++) { if (child_bios[i] != NULL) g_destroy_bio(child_bios[i]); } free(child_bios, M_NVME); } static struct bio ** nvme_allocate_child_bios(int num_bios) { struct bio **child_bios; int err = 0, i; child_bios = malloc(num_bios * sizeof(struct bio *), M_NVME, M_NOWAIT); if (child_bios == NULL) return (NULL); for (i = 0; i < num_bios; i++) { child_bios[i] = g_new_bio(); if (child_bios[i] == NULL) err = ENOMEM; } if (err == ENOMEM) { nvme_free_child_bios(num_bios, child_bios); return (NULL); } return (child_bios); } static struct bio ** nvme_construct_child_bios(struct bio *bp, uint32_t alignment, int *num_bios) { struct bio **child_bios; struct bio *child; uint64_t cur_offset; caddr_t data; uint32_t rem_bcount; int i; struct vm_page **ma; uint32_t ma_offset; *num_bios = nvme_get_num_segments(bp->bio_offset, bp->bio_bcount, alignment); child_bios = nvme_allocate_child_bios(*num_bios); if (child_bios == NULL) return (NULL); bp->bio_children = *num_bios; bp->bio_inbed = 0; cur_offset = bp->bio_offset; rem_bcount = bp->bio_bcount; data = bp->bio_data; ma_offset = bp->bio_ma_offset; ma = bp->bio_ma; for (i = 0; i < *num_bios; i++) { child = child_bios[i]; child->bio_parent = bp; child->bio_cmd = bp->bio_cmd; child->bio_offset = cur_offset; child->bio_bcount = min(rem_bcount, alignment - (cur_offset & (alignment - 1))); child->bio_flags = bp->bio_flags; if (bp->bio_flags & BIO_UNMAPPED) { child->bio_ma_offset = ma_offset; child->bio_ma = ma; child->bio_ma_n = nvme_get_num_segments(child->bio_ma_offset, child->bio_bcount, PAGE_SIZE); ma_offset = (ma_offset + child->bio_bcount) & PAGE_MASK; ma += child->bio_ma_n; if (ma_offset != 0) ma -= 1; } else { child->bio_data = data; data += child->bio_bcount; } cur_offset += child->bio_bcount; rem_bcount -= child->bio_bcount; } return (child_bios); } static int nvme_ns_split_bio(struct nvme_namespace *ns, struct bio *bp, uint32_t alignment) { struct bio *child; struct bio **child_bios; int err, i, num_bios; child_bios = nvme_construct_child_bios(bp, alignment, &num_bios); if (child_bios == NULL) return (ENOMEM); for (i = 0; i < num_bios; i++) { child = child_bios[i]; err = nvme_ns_bio_process(ns, child, nvme_bio_child_done); if (err != 0) { nvme_bio_child_inbed(bp, err); g_destroy_bio(child); } } free(child_bios, M_NVME); return (0); } int nvme_ns_bio_process(struct nvme_namespace *ns, struct bio *bp, nvme_cb_fn_t cb_fn) { struct nvme_dsm_range *dsm_range; uint32_t num_bios; int err; bp->bio_driver1 = cb_fn; if (ns->boundary > 0 && (bp->bio_cmd == BIO_READ || bp->bio_cmd == BIO_WRITE)) { num_bios = nvme_get_num_segments(bp->bio_offset, bp->bio_bcount, ns->boundary); if (num_bios > 1) return (nvme_ns_split_bio(ns, bp, ns->boundary)); } switch (bp->bio_cmd) { case BIO_READ: err = nvme_ns_cmd_read_bio(ns, bp, nvme_ns_bio_done, bp); break; case BIO_WRITE: err = nvme_ns_cmd_write_bio(ns, bp, nvme_ns_bio_done, bp); break; case BIO_FLUSH: err = nvme_ns_cmd_flush(ns, nvme_ns_bio_done, bp); break; case BIO_DELETE: dsm_range = malloc(sizeof(struct nvme_dsm_range), M_NVME, M_ZERO | M_NOWAIT); if (!dsm_range) { err = ENOMEM; break; } dsm_range->length = htole32(bp->bio_bcount/nvme_ns_get_sector_size(ns)); dsm_range->starting_lba = htole64(bp->bio_offset/nvme_ns_get_sector_size(ns)); bp->bio_driver2 = dsm_range; err = nvme_ns_cmd_deallocate(ns, dsm_range, 1, nvme_ns_bio_done, bp); if (err != 0) free(dsm_range, M_NVME); break; default: err = EOPNOTSUPP; break; } return (err); } int nvme_ns_ioctl_process(struct nvme_namespace *ns, u_long cmd, caddr_t arg, int flag, struct thread *td) { return (nvme_ns_ioctl(ns->cdev, cmd, arg, flag, td)); } int nvme_ns_construct(struct nvme_namespace *ns, uint32_t id, struct nvme_controller *ctrlr) { struct make_dev_args md_args; struct nvme_completion_poll_status status; int res; int unit; uint8_t flbas_fmt; uint8_t vwc_present; ns->ctrlr = ctrlr; ns->id = id; /* * Namespaces are reconstructed after a controller reset, so check * to make sure we only call mtx_init once on each mtx. * * TODO: Move this somewhere where it gets called at controller * construction time, which is not invoked as part of each * controller reset. */ if (!mtx_initialized(&ns->lock)) mtx_init(&ns->lock, "nvme ns lock", NULL, MTX_DEF); status.done = 0; nvme_ctrlr_cmd_identify_namespace(ctrlr, id, &ns->data, nvme_completion_poll_cb, &status); nvme_completion_poll(&status); if (nvme_completion_is_error(&status.cpl)) { nvme_printf(ctrlr, "nvme_identify_namespace failed\n"); return (ENXIO); } /* Convert data to host endian */ nvme_namespace_data_swapbytes(&ns->data); /* * If the size of is zero, chances are this isn't a valid * namespace (eg one that's not been configured yet). The * standard says the entire id will be zeros, so this is a * cheap way to test for that. */ if (ns->data.nsze == 0) return (ENXIO); - flbas_fmt = (ns->data.flbas >> NVME_NS_DATA_FLBAS_FORMAT_SHIFT) & - NVME_NS_DATA_FLBAS_FORMAT_MASK; + flbas_fmt = NVMEV(NVME_NS_DATA_FLBAS_FORMAT, ns->data.flbas); + /* * Note: format is a 0-based value, so > is appropriate here, * not >=. */ if (flbas_fmt > ns->data.nlbaf) { nvme_printf(ctrlr, "lba format %d exceeds number supported (%d)\n", flbas_fmt, ns->data.nlbaf + 1); return (ENXIO); } /* * Older Intel devices (like the PC35xxx and P45xx series) advertise in * vendor specific space an alignment that improves performance. If * present use for the stripe size. NVMe 1.3 standardized this as * NOIOB, and newer Intel drives use that. */ if ((ctrlr->quirks & QUIRK_INTEL_ALIGNMENT) != 0) { if (ctrlr->cdata.vs[3] != 0) ns->boundary = 1 << (ctrlr->cdata.vs[3] + NVME_MPS_SHIFT + NVME_CAP_HI_MPSMIN(ctrlr->cap_hi)); else ns->boundary = 0; } else { ns->boundary = ns->data.noiob * nvme_ns_get_sector_size(ns); } if (nvme_ctrlr_has_dataset_mgmt(&ctrlr->cdata)) ns->flags |= NVME_NS_DEALLOCATE_SUPPORTED; - vwc_present = (ctrlr->cdata.vwc >> NVME_CTRLR_DATA_VWC_PRESENT_SHIFT) & - NVME_CTRLR_DATA_VWC_PRESENT_MASK; + vwc_present = NVMEV(NVME_CTRLR_DATA_VWC_PRESENT, ctrlr->cdata.vwc); if (vwc_present) ns->flags |= NVME_NS_FLUSH_SUPPORTED; /* * cdev may have already been created, if we are reconstructing the * namespace after a controller-level reset. */ if (ns->cdev != NULL) return (0); /* * Namespace IDs start at 1, so we need to subtract 1 to create a * correct unit number. */ unit = device_get_unit(ctrlr->dev) * NVME_MAX_NAMESPACES + ns->id - 1; make_dev_args_init(&md_args); md_args.mda_devsw = &nvme_ns_cdevsw; md_args.mda_unit = unit; md_args.mda_mode = 0600; md_args.mda_si_drv1 = ns; res = make_dev_s(&md_args, &ns->cdev, "nvme%dns%d", device_get_unit(ctrlr->dev), ns->id); if (res != 0) return (ENXIO); ns->cdev->si_flags |= SI_UNMAPPED; return (0); } void nvme_ns_destruct(struct nvme_namespace *ns) { if (ns->cdev != NULL) destroy_dev(ns->cdev); } diff --git a/sys/dev/nvme/nvme_qpair.c b/sys/dev/nvme/nvme_qpair.c index 2e9af4310a15..e087abcb2654 100644 --- a/sys/dev/nvme/nvme_qpair.c +++ b/sys/dev/nvme/nvme_qpair.c @@ -1,1511 +1,1511 @@ /*- * SPDX-License-Identifier: BSD-2-Clause * * Copyright (C) 2012-2014 Intel Corporation * All rights reserved. * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. */ #include #include #include #include #include #include #include #include "nvme_private.h" typedef enum error_print { ERROR_PRINT_NONE, ERROR_PRINT_NO_RETRY, ERROR_PRINT_ALL } error_print_t; #define DO_NOT_RETRY 1 static void _nvme_qpair_submit_request(struct nvme_qpair *qpair, struct nvme_request *req); static void nvme_qpair_destroy(struct nvme_qpair *qpair); #define DEFAULT_INDEX 256 #define DEFAULT_ENTRY(x) [DEFAULT_INDEX] = x #define OPC_ENTRY(x) [NVME_OPC_ ## x] = #x static const char *admin_opcode[DEFAULT_INDEX + 1] = { OPC_ENTRY(DELETE_IO_SQ), OPC_ENTRY(CREATE_IO_SQ), OPC_ENTRY(GET_LOG_PAGE), OPC_ENTRY(DELETE_IO_CQ), OPC_ENTRY(CREATE_IO_CQ), OPC_ENTRY(IDENTIFY), OPC_ENTRY(ABORT), OPC_ENTRY(SET_FEATURES), OPC_ENTRY(GET_FEATURES), OPC_ENTRY(ASYNC_EVENT_REQUEST), OPC_ENTRY(NAMESPACE_MANAGEMENT), OPC_ENTRY(FIRMWARE_ACTIVATE), OPC_ENTRY(FIRMWARE_IMAGE_DOWNLOAD), OPC_ENTRY(DEVICE_SELF_TEST), OPC_ENTRY(NAMESPACE_ATTACHMENT), OPC_ENTRY(KEEP_ALIVE), OPC_ENTRY(DIRECTIVE_SEND), OPC_ENTRY(DIRECTIVE_RECEIVE), OPC_ENTRY(VIRTUALIZATION_MANAGEMENT), OPC_ENTRY(NVME_MI_SEND), OPC_ENTRY(NVME_MI_RECEIVE), OPC_ENTRY(CAPACITY_MANAGEMENT), OPC_ENTRY(LOCKDOWN), OPC_ENTRY(DOORBELL_BUFFER_CONFIG), OPC_ENTRY(FABRICS_COMMANDS), OPC_ENTRY(FORMAT_NVM), OPC_ENTRY(SECURITY_SEND), OPC_ENTRY(SECURITY_RECEIVE), OPC_ENTRY(SANITIZE), OPC_ENTRY(GET_LBA_STATUS), DEFAULT_ENTRY("ADMIN COMMAND"), }; static const char *io_opcode[DEFAULT_INDEX + 1] = { OPC_ENTRY(FLUSH), OPC_ENTRY(WRITE), OPC_ENTRY(READ), OPC_ENTRY(WRITE_UNCORRECTABLE), OPC_ENTRY(COMPARE), OPC_ENTRY(WRITE_ZEROES), OPC_ENTRY(DATASET_MANAGEMENT), OPC_ENTRY(VERIFY), OPC_ENTRY(RESERVATION_REGISTER), OPC_ENTRY(RESERVATION_REPORT), OPC_ENTRY(RESERVATION_ACQUIRE), OPC_ENTRY(RESERVATION_RELEASE), OPC_ENTRY(COPY), DEFAULT_ENTRY("IO COMMAND"), }; static const char * get_opcode_string(const char *op[DEFAULT_INDEX + 1], uint16_t opc) { const char *nm = opc < DEFAULT_INDEX ? op[opc] : op[DEFAULT_INDEX]; return (nm != NULL ? nm : op[DEFAULT_INDEX]); } static const char * get_admin_opcode_string(uint16_t opc) { return (get_opcode_string(admin_opcode, opc)); } static const char * get_io_opcode_string(uint16_t opc) { return (get_opcode_string(io_opcode, opc)); } static void nvme_admin_qpair_print_command(struct nvme_qpair *qpair, struct nvme_command *cmd) { nvme_printf(qpair->ctrlr, "%s (%02x) sqid:%d cid:%d nsid:%x " "cdw10:%08x cdw11:%08x\n", get_admin_opcode_string(cmd->opc), cmd->opc, qpair->id, cmd->cid, le32toh(cmd->nsid), le32toh(cmd->cdw10), le32toh(cmd->cdw11)); } static void nvme_io_qpair_print_command(struct nvme_qpair *qpair, struct nvme_command *cmd) { switch (cmd->opc) { case NVME_OPC_WRITE: case NVME_OPC_READ: case NVME_OPC_WRITE_UNCORRECTABLE: case NVME_OPC_COMPARE: case NVME_OPC_WRITE_ZEROES: case NVME_OPC_VERIFY: nvme_printf(qpair->ctrlr, "%s sqid:%d cid:%d nsid:%d " "lba:%llu len:%d\n", get_io_opcode_string(cmd->opc), qpair->id, cmd->cid, le32toh(cmd->nsid), ((unsigned long long)le32toh(cmd->cdw11) << 32) + le32toh(cmd->cdw10), (le32toh(cmd->cdw12) & 0xFFFF) + 1); break; case NVME_OPC_FLUSH: case NVME_OPC_DATASET_MANAGEMENT: case NVME_OPC_RESERVATION_REGISTER: case NVME_OPC_RESERVATION_REPORT: case NVME_OPC_RESERVATION_ACQUIRE: case NVME_OPC_RESERVATION_RELEASE: nvme_printf(qpair->ctrlr, "%s sqid:%d cid:%d nsid:%d\n", get_io_opcode_string(cmd->opc), qpair->id, cmd->cid, le32toh(cmd->nsid)); break; default: nvme_printf(qpair->ctrlr, "%s (%02x) sqid:%d cid:%d nsid:%d\n", get_io_opcode_string(cmd->opc), cmd->opc, qpair->id, cmd->cid, le32toh(cmd->nsid)); break; } } void nvme_qpair_print_command(struct nvme_qpair *qpair, struct nvme_command *cmd) { if (qpair->id == 0) nvme_admin_qpair_print_command(qpair, cmd); else nvme_io_qpair_print_command(qpair, cmd); if (nvme_verbose_cmd_dump) { nvme_printf(qpair->ctrlr, "nsid:%#x rsvd2:%#x rsvd3:%#x mptr:%#jx prp1:%#jx prp2:%#jx\n", cmd->nsid, cmd->rsvd2, cmd->rsvd3, (uintmax_t)cmd->mptr, (uintmax_t)cmd->prp1, (uintmax_t)cmd->prp2); nvme_printf(qpair->ctrlr, "cdw10: %#x cdw11:%#x cdw12:%#x cdw13:%#x cdw14:%#x cdw15:%#x\n", cmd->cdw10, cmd->cdw11, cmd->cdw12, cmd->cdw13, cmd->cdw14, cmd->cdw15); } } struct nvme_status_string { uint16_t sc; const char * str; }; static struct nvme_status_string generic_status[] = { { NVME_SC_SUCCESS, "SUCCESS" }, { NVME_SC_INVALID_OPCODE, "INVALID OPCODE" }, { NVME_SC_INVALID_FIELD, "INVALID_FIELD" }, { NVME_SC_COMMAND_ID_CONFLICT, "COMMAND ID CONFLICT" }, { NVME_SC_DATA_TRANSFER_ERROR, "DATA TRANSFER ERROR" }, { NVME_SC_ABORTED_POWER_LOSS, "ABORTED - POWER LOSS" }, { NVME_SC_INTERNAL_DEVICE_ERROR, "INTERNAL DEVICE ERROR" }, { NVME_SC_ABORTED_BY_REQUEST, "ABORTED - BY REQUEST" }, { NVME_SC_ABORTED_SQ_DELETION, "ABORTED - SQ DELETION" }, { NVME_SC_ABORTED_FAILED_FUSED, "ABORTED - FAILED FUSED" }, { NVME_SC_ABORTED_MISSING_FUSED, "ABORTED - MISSING FUSED" }, { NVME_SC_INVALID_NAMESPACE_OR_FORMAT, "INVALID NAMESPACE OR FORMAT" }, { NVME_SC_COMMAND_SEQUENCE_ERROR, "COMMAND SEQUENCE ERROR" }, { NVME_SC_INVALID_SGL_SEGMENT_DESCR, "INVALID SGL SEGMENT DESCRIPTOR" }, { NVME_SC_INVALID_NUMBER_OF_SGL_DESCR, "INVALID NUMBER OF SGL DESCRIPTORS" }, { NVME_SC_DATA_SGL_LENGTH_INVALID, "DATA SGL LENGTH INVALID" }, { NVME_SC_METADATA_SGL_LENGTH_INVALID, "METADATA SGL LENGTH INVALID" }, { NVME_SC_SGL_DESCRIPTOR_TYPE_INVALID, "SGL DESCRIPTOR TYPE INVALID" }, { NVME_SC_INVALID_USE_OF_CMB, "INVALID USE OF CONTROLLER MEMORY BUFFER" }, { NVME_SC_PRP_OFFET_INVALID, "PRP OFFET INVALID" }, { NVME_SC_ATOMIC_WRITE_UNIT_EXCEEDED, "ATOMIC WRITE UNIT EXCEEDED" }, { NVME_SC_OPERATION_DENIED, "OPERATION DENIED" }, { NVME_SC_SGL_OFFSET_INVALID, "SGL OFFSET INVALID" }, { NVME_SC_HOST_ID_INCONSISTENT_FORMAT, "HOST IDENTIFIER INCONSISTENT FORMAT" }, { NVME_SC_KEEP_ALIVE_TIMEOUT_EXPIRED, "KEEP ALIVE TIMEOUT EXPIRED" }, { NVME_SC_KEEP_ALIVE_TIMEOUT_INVALID, "KEEP ALIVE TIMEOUT INVALID" }, { NVME_SC_ABORTED_DUE_TO_PREEMPT, "COMMAND ABORTED DUE TO PREEMPT AND ABORT" }, { NVME_SC_SANITIZE_FAILED, "SANITIZE FAILED" }, { NVME_SC_SANITIZE_IN_PROGRESS, "SANITIZE IN PROGRESS" }, { NVME_SC_SGL_DATA_BLOCK_GRAN_INVALID, "SGL_DATA_BLOCK_GRANULARITY_INVALID" }, { NVME_SC_NOT_SUPPORTED_IN_CMB, "COMMAND NOT SUPPORTED FOR QUEUE IN CMB" }, { NVME_SC_NAMESPACE_IS_WRITE_PROTECTED, "NAMESPACE IS WRITE PROTECTED" }, { NVME_SC_COMMAND_INTERRUPTED, "COMMAND INTERRUPTED" }, { NVME_SC_TRANSIENT_TRANSPORT_ERROR, "TRANSIENT TRANSPORT ERROR" }, { NVME_SC_LBA_OUT_OF_RANGE, "LBA OUT OF RANGE" }, { NVME_SC_CAPACITY_EXCEEDED, "CAPACITY EXCEEDED" }, { NVME_SC_NAMESPACE_NOT_READY, "NAMESPACE NOT READY" }, { NVME_SC_RESERVATION_CONFLICT, "RESERVATION CONFLICT" }, { NVME_SC_FORMAT_IN_PROGRESS, "FORMAT IN PROGRESS" }, { 0xFFFF, "GENERIC" } }; static struct nvme_status_string command_specific_status[] = { { NVME_SC_COMPLETION_QUEUE_INVALID, "INVALID COMPLETION QUEUE" }, { NVME_SC_INVALID_QUEUE_IDENTIFIER, "INVALID QUEUE IDENTIFIER" }, { NVME_SC_MAXIMUM_QUEUE_SIZE_EXCEEDED, "MAX QUEUE SIZE EXCEEDED" }, { NVME_SC_ABORT_COMMAND_LIMIT_EXCEEDED, "ABORT CMD LIMIT EXCEEDED" }, { NVME_SC_ASYNC_EVENT_REQUEST_LIMIT_EXCEEDED, "ASYNC LIMIT EXCEEDED" }, { NVME_SC_INVALID_FIRMWARE_SLOT, "INVALID FIRMWARE SLOT" }, { NVME_SC_INVALID_FIRMWARE_IMAGE, "INVALID FIRMWARE IMAGE" }, { NVME_SC_INVALID_INTERRUPT_VECTOR, "INVALID INTERRUPT VECTOR" }, { NVME_SC_INVALID_LOG_PAGE, "INVALID LOG PAGE" }, { NVME_SC_INVALID_FORMAT, "INVALID FORMAT" }, { NVME_SC_FIRMWARE_REQUIRES_RESET, "FIRMWARE REQUIRES RESET" }, { NVME_SC_INVALID_QUEUE_DELETION, "INVALID QUEUE DELETION" }, { NVME_SC_FEATURE_NOT_SAVEABLE, "FEATURE IDENTIFIER NOT SAVEABLE" }, { NVME_SC_FEATURE_NOT_CHANGEABLE, "FEATURE NOT CHANGEABLE" }, { NVME_SC_FEATURE_NOT_NS_SPECIFIC, "FEATURE NOT NAMESPACE SPECIFIC" }, { NVME_SC_FW_ACT_REQUIRES_NVMS_RESET, "FIRMWARE ACTIVATION REQUIRES NVM SUBSYSTEM RESET" }, { NVME_SC_FW_ACT_REQUIRES_RESET, "FIRMWARE ACTIVATION REQUIRES RESET" }, { NVME_SC_FW_ACT_REQUIRES_TIME, "FIRMWARE ACTIVATION REQUIRES MAXIMUM TIME VIOLATION" }, { NVME_SC_FW_ACT_PROHIBITED, "FIRMWARE ACTIVATION PROHIBITED" }, { NVME_SC_OVERLAPPING_RANGE, "OVERLAPPING RANGE" }, { NVME_SC_NS_INSUFFICIENT_CAPACITY, "NAMESPACE INSUFFICIENT CAPACITY" }, { NVME_SC_NS_ID_UNAVAILABLE, "NAMESPACE IDENTIFIER UNAVAILABLE" }, { NVME_SC_NS_ALREADY_ATTACHED, "NAMESPACE ALREADY ATTACHED" }, { NVME_SC_NS_IS_PRIVATE, "NAMESPACE IS PRIVATE" }, { NVME_SC_NS_NOT_ATTACHED, "NS NOT ATTACHED" }, { NVME_SC_THIN_PROV_NOT_SUPPORTED, "THIN PROVISIONING NOT SUPPORTED" }, { NVME_SC_CTRLR_LIST_INVALID, "CONTROLLER LIST INVALID" }, { NVME_SC_SELF_TEST_IN_PROGRESS, "DEVICE SELF-TEST IN PROGRESS" }, { NVME_SC_BOOT_PART_WRITE_PROHIB, "BOOT PARTITION WRITE PROHIBITED" }, { NVME_SC_INVALID_CTRLR_ID, "INVALID CONTROLLER IDENTIFIER" }, { NVME_SC_INVALID_SEC_CTRLR_STATE, "INVALID SECONDARY CONTROLLER STATE" }, { NVME_SC_INVALID_NUM_OF_CTRLR_RESRC, "INVALID NUMBER OF CONTROLLER RESOURCES" }, { NVME_SC_INVALID_RESOURCE_ID, "INVALID RESOURCE IDENTIFIER" }, { NVME_SC_SANITIZE_PROHIBITED_WPMRE, "SANITIZE PROHIBITED WRITE PERSISTENT MEMORY REGION ENABLED" }, { NVME_SC_ANA_GROUP_ID_INVALID, "ANA GROUP IDENTIFIED INVALID" }, { NVME_SC_ANA_ATTACH_FAILED, "ANA ATTACH FAILED" }, { NVME_SC_CONFLICTING_ATTRIBUTES, "CONFLICTING ATTRIBUTES" }, { NVME_SC_INVALID_PROTECTION_INFO, "INVALID PROTECTION INFO" }, { NVME_SC_ATTEMPTED_WRITE_TO_RO_PAGE, "WRITE TO RO PAGE" }, { 0xFFFF, "COMMAND SPECIFIC" } }; static struct nvme_status_string media_error_status[] = { { NVME_SC_WRITE_FAULTS, "WRITE FAULTS" }, { NVME_SC_UNRECOVERED_READ_ERROR, "UNRECOVERED READ ERROR" }, { NVME_SC_GUARD_CHECK_ERROR, "GUARD CHECK ERROR" }, { NVME_SC_APPLICATION_TAG_CHECK_ERROR, "APPLICATION TAG CHECK ERROR" }, { NVME_SC_REFERENCE_TAG_CHECK_ERROR, "REFERENCE TAG CHECK ERROR" }, { NVME_SC_COMPARE_FAILURE, "COMPARE FAILURE" }, { NVME_SC_ACCESS_DENIED, "ACCESS DENIED" }, { NVME_SC_DEALLOCATED_OR_UNWRITTEN, "DEALLOCATED OR UNWRITTEN LOGICAL BLOCK" }, { 0xFFFF, "MEDIA ERROR" } }; static struct nvme_status_string path_related_status[] = { { NVME_SC_INTERNAL_PATH_ERROR, "INTERNAL PATH ERROR" }, { NVME_SC_ASYMMETRIC_ACCESS_PERSISTENT_LOSS, "ASYMMETRIC ACCESS PERSISTENT LOSS" }, { NVME_SC_ASYMMETRIC_ACCESS_INACCESSIBLE, "ASYMMETRIC ACCESS INACCESSIBLE" }, { NVME_SC_ASYMMETRIC_ACCESS_TRANSITION, "ASYMMETRIC ACCESS TRANSITION" }, { NVME_SC_CONTROLLER_PATHING_ERROR, "CONTROLLER PATHING ERROR" }, { NVME_SC_HOST_PATHING_ERROR, "HOST PATHING ERROR" }, { NVME_SC_COMMAND_ABORTED_BY_HOST, "COMMAND ABORTED BY HOST" }, { 0xFFFF, "PATH RELATED" }, }; static const char * get_status_string(uint16_t sct, uint16_t sc) { struct nvme_status_string *entry; switch (sct) { case NVME_SCT_GENERIC: entry = generic_status; break; case NVME_SCT_COMMAND_SPECIFIC: entry = command_specific_status; break; case NVME_SCT_MEDIA_ERROR: entry = media_error_status; break; case NVME_SCT_PATH_RELATED: entry = path_related_status; break; case NVME_SCT_VENDOR_SPECIFIC: return ("VENDOR SPECIFIC"); default: return ("RESERVED"); } while (entry->sc != 0xFFFF) { if (entry->sc == sc) return (entry->str); entry++; } return (entry->str); } void nvme_qpair_print_completion(struct nvme_qpair *qpair, struct nvme_completion *cpl) { uint8_t sct, sc, crd, m, dnr, p; sct = NVME_STATUS_GET_SCT(cpl->status); sc = NVME_STATUS_GET_SC(cpl->status); crd = NVME_STATUS_GET_CRD(cpl->status); m = NVME_STATUS_GET_M(cpl->status); dnr = NVME_STATUS_GET_DNR(cpl->status); p = NVME_STATUS_GET_P(cpl->status); nvme_printf(qpair->ctrlr, "%s (%02x/%02x) crd:%x m:%x dnr:%x p:%d " "sqid:%d cid:%d cdw0:%x\n", get_status_string(sct, sc), sct, sc, crd, m, dnr, p, cpl->sqid, cpl->cid, cpl->cdw0); } static bool nvme_completion_is_retry(const struct nvme_completion *cpl) { uint8_t sct, sc, dnr; sct = NVME_STATUS_GET_SCT(cpl->status); sc = NVME_STATUS_GET_SC(cpl->status); dnr = NVME_STATUS_GET_DNR(cpl->status); /* Do Not Retry Bit */ /* * TODO: spec is not clear how commands that are aborted due * to TLER will be marked. So for now, it seems * NAMESPACE_NOT_READY is the only case where we should * look at the DNR bit. Requests failed with ABORTED_BY_REQUEST * set the DNR bit correctly since the driver controls that. */ switch (sct) { case NVME_SCT_GENERIC: switch (sc) { case NVME_SC_ABORTED_BY_REQUEST: case NVME_SC_NAMESPACE_NOT_READY: if (dnr) return (0); else return (1); case NVME_SC_INVALID_OPCODE: case NVME_SC_INVALID_FIELD: case NVME_SC_COMMAND_ID_CONFLICT: case NVME_SC_DATA_TRANSFER_ERROR: case NVME_SC_ABORTED_POWER_LOSS: case NVME_SC_INTERNAL_DEVICE_ERROR: case NVME_SC_ABORTED_SQ_DELETION: case NVME_SC_ABORTED_FAILED_FUSED: case NVME_SC_ABORTED_MISSING_FUSED: case NVME_SC_INVALID_NAMESPACE_OR_FORMAT: case NVME_SC_COMMAND_SEQUENCE_ERROR: case NVME_SC_LBA_OUT_OF_RANGE: case NVME_SC_CAPACITY_EXCEEDED: default: return (0); } case NVME_SCT_COMMAND_SPECIFIC: case NVME_SCT_MEDIA_ERROR: return (0); case NVME_SCT_PATH_RELATED: switch (sc) { case NVME_SC_INTERNAL_PATH_ERROR: if (dnr) return (0); else return (1); default: return (0); } case NVME_SCT_VENDOR_SPECIFIC: default: return (0); } } static void nvme_qpair_complete_tracker(struct nvme_tracker *tr, struct nvme_completion *cpl, error_print_t print_on_error) { struct nvme_qpair * qpair = tr->qpair; struct nvme_request *req; bool retry, error, retriable; req = tr->req; error = nvme_completion_is_error(cpl); retriable = nvme_completion_is_retry(cpl); retry = error && retriable && req->retries < nvme_retry_count; if (retry) qpair->num_retries++; if (error && req->retries >= nvme_retry_count && retriable) qpair->num_failures++; if (error && (print_on_error == ERROR_PRINT_ALL || (!retry && print_on_error == ERROR_PRINT_NO_RETRY))) { nvme_qpair_print_command(qpair, &req->cmd); nvme_qpair_print_completion(qpair, cpl); } qpair->act_tr[cpl->cid] = NULL; KASSERT(cpl->cid == req->cmd.cid, ("cpl cid does not match cmd cid\n")); if (!retry) { if (req->payload_valid) { bus_dmamap_sync(qpair->dma_tag_payload, tr->payload_dma_map, BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE); } if (req->cb_fn) req->cb_fn(req->cb_arg, cpl); } mtx_lock(&qpair->lock); if (retry) { req->retries++; nvme_qpair_submit_tracker(qpair, tr); } else { if (req->payload_valid) { bus_dmamap_unload(qpair->dma_tag_payload, tr->payload_dma_map); } nvme_free_request(req); tr->req = NULL; TAILQ_REMOVE(&qpair->outstanding_tr, tr, tailq); TAILQ_INSERT_HEAD(&qpair->free_tr, tr, tailq); /* * If the controller is in the middle of resetting, don't * try to submit queued requests here - let the reset logic * handle that instead. */ if (!STAILQ_EMPTY(&qpair->queued_req) && !qpair->ctrlr->is_resetting) { req = STAILQ_FIRST(&qpair->queued_req); STAILQ_REMOVE_HEAD(&qpair->queued_req, stailq); _nvme_qpair_submit_request(qpair, req); } } mtx_unlock(&qpair->lock); } static void nvme_qpair_manual_complete_tracker( struct nvme_tracker *tr, uint32_t sct, uint32_t sc, uint32_t dnr, error_print_t print_on_error) { struct nvme_completion cpl; memset(&cpl, 0, sizeof(cpl)); struct nvme_qpair * qpair = tr->qpair; cpl.sqid = qpair->id; cpl.cid = tr->cid; cpl.status |= (sct & NVME_STATUS_SCT_MASK) << NVME_STATUS_SCT_SHIFT; cpl.status |= (sc & NVME_STATUS_SC_MASK) << NVME_STATUS_SC_SHIFT; cpl.status |= (dnr & NVME_STATUS_DNR_MASK) << NVME_STATUS_DNR_SHIFT; /* M=0 : this is artificial so no data in error log page */ /* CRD=0 : this is artificial and no delayed retry support anyway */ /* P=0 : phase not checked */ nvme_qpair_complete_tracker(tr, &cpl, print_on_error); } void nvme_qpair_manual_complete_request(struct nvme_qpair *qpair, struct nvme_request *req, uint32_t sct, uint32_t sc) { struct nvme_completion cpl; bool error; memset(&cpl, 0, sizeof(cpl)); cpl.sqid = qpair->id; cpl.status |= (sct & NVME_STATUS_SCT_MASK) << NVME_STATUS_SCT_SHIFT; cpl.status |= (sc & NVME_STATUS_SC_MASK) << NVME_STATUS_SC_SHIFT; error = nvme_completion_is_error(&cpl); if (error) { nvme_qpair_print_command(qpair, &req->cmd); nvme_qpair_print_completion(qpair, &cpl); } if (req->cb_fn) req->cb_fn(req->cb_arg, &cpl); nvme_free_request(req); } /* Locked version of completion processor */ static bool _nvme_qpair_process_completions(struct nvme_qpair *qpair) { struct nvme_tracker *tr; struct nvme_completion cpl; bool done = false; bool in_panic = dumping || SCHEDULER_STOPPED(); mtx_assert(&qpair->recovery, MA_OWNED); /* * qpair is not enabled, likely because a controller reset is in * progress. Ignore the interrupt - any I/O that was associated with * this interrupt will get retried when the reset is complete. Any * pending completions for when we're in startup will be completed * as soon as initialization is complete and we start sending commands * to the device. */ if (qpair->recovery_state != RECOVERY_NONE) { qpair->num_ignored++; return (false); } /* * Sanity check initialization. After we reset the hardware, the phase * is defined to be 1. So if we get here with zero prior calls and the * phase is 0, it means that we've lost a race between the * initialization and the ISR running. With the phase wrong, we'll * process a bunch of completions that aren't really completions leading * to a KASSERT below. */ KASSERT(!(qpair->num_intr_handler_calls == 0 && qpair->phase == 0), ("%s: Phase wrong for first interrupt call.", device_get_nameunit(qpair->ctrlr->dev))); qpair->num_intr_handler_calls++; bus_dmamap_sync(qpair->dma_tag, qpair->queuemem_map, BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE); /* * A panic can stop the CPU this routine is running on at any point. If * we're called during a panic, complete the sq_head wrap protocol for * the case where we are interrupted just after the increment at 1 * below, but before we can reset cq_head to zero at 2. Also cope with * the case where we do the zero at 2, but may or may not have done the * phase adjustment at step 3. The panic machinery flushes all pending * memory writes, so we can make these strong ordering assumptions * that would otherwise be unwise if we were racing in real time. */ if (__predict_false(in_panic)) { if (qpair->cq_head == qpair->num_entries) { /* * Here we know that we need to zero cq_head and then negate * the phase, which hasn't been assigned if cq_head isn't * zero due to the atomic_store_rel. */ qpair->cq_head = 0; qpair->phase = !qpair->phase; } else if (qpair->cq_head == 0) { /* * In this case, we know that the assignment at 2 * happened below, but we don't know if it 3 happened or * not. To do this, we look at the last completion * entry and set the phase to the opposite phase * that it has. This gets us back in sync */ cpl = qpair->cpl[qpair->num_entries - 1]; nvme_completion_swapbytes(&cpl); qpair->phase = !NVME_STATUS_GET_P(cpl.status); } } while (1) { uint16_t status; /* * We need to do this dance to avoid a race between the host and * the device where the device overtakes the host while the host * is reading this record, leaving the status field 'new' and * the sqhd and cid fields potentially stale. If the phase * doesn't match, that means status hasn't yet been updated and * we'll get any pending changes next time. It also means that * the phase must be the same the second time. We have to sync * before reading to ensure any bouncing completes. */ status = le16toh(qpair->cpl[qpair->cq_head].status); if (NVME_STATUS_GET_P(status) != qpair->phase) break; bus_dmamap_sync(qpair->dma_tag, qpair->queuemem_map, BUS_DMASYNC_POSTREAD | BUS_DMASYNC_POSTWRITE); cpl = qpair->cpl[qpair->cq_head]; nvme_completion_swapbytes(&cpl); KASSERT( NVME_STATUS_GET_P(status) == NVME_STATUS_GET_P(cpl.status), ("Phase unexpectedly inconsistent")); if (cpl.cid < qpair->num_trackers) tr = qpair->act_tr[cpl.cid]; else tr = NULL; done = true; if (tr != NULL) { nvme_qpair_complete_tracker(tr, &cpl, ERROR_PRINT_ALL); qpair->sq_head = cpl.sqhd; } else if (!in_panic) { /* * A missing tracker is normally an error. However, a * panic can stop the CPU this routine is running on * after completing an I/O but before updating * qpair->cq_head at 1 below. Later, we re-enter this * routine to poll I/O associated with the kernel * dump. We find that the tr has been set to null before * calling the completion routine. If it hasn't * completed (or it triggers a panic), then '1' below * won't have updated cq_head. Rather than panic again, * ignore this condition because it's not unexpected. */ nvme_printf(qpair->ctrlr, "cpl (cid = %u) does not map to outstanding cmd\n", cpl.cid); nvme_qpair_print_completion(qpair, &qpair->cpl[qpair->cq_head]); KASSERT(0, ("received completion for unknown cmd")); } /* * There's a number of races with the following (see above) when * the system panics. We compensate for each one of them by * using the atomic store to force strong ordering (at least when * viewed in the aftermath of a panic). */ if (++qpair->cq_head == qpair->num_entries) { /* 1 */ atomic_store_rel_int(&qpair->cq_head, 0); /* 2 */ qpair->phase = !qpair->phase; /* 3 */ } } if (done) { bus_space_write_4(qpair->ctrlr->bus_tag, qpair->ctrlr->bus_handle, qpair->cq_hdbl_off, qpair->cq_head); } return (done); } bool nvme_qpair_process_completions(struct nvme_qpair *qpair) { bool done; /* * Interlock with reset / recovery code. This is an usually uncontended * to make sure that we drain out of the ISRs before we reset the card * and to prevent races with the recovery process called from a timeout * context. */ if (!mtx_trylock(&qpair->recovery)) { qpair->num_recovery_nolock++; return (false); } done = _nvme_qpair_process_completions(qpair); mtx_unlock(&qpair->recovery); return (done); } static void nvme_qpair_msi_handler(void *arg) { struct nvme_qpair *qpair = arg; nvme_qpair_process_completions(qpair); } int nvme_qpair_construct(struct nvme_qpair *qpair, uint32_t num_entries, uint32_t num_trackers, struct nvme_controller *ctrlr) { struct nvme_tracker *tr; size_t cmdsz, cplsz, prpsz, allocsz, prpmemsz; uint64_t queuemem_phys, prpmem_phys, list_phys; uint8_t *queuemem, *prpmem, *prp_list; int i, err; qpair->vector = ctrlr->msi_count > 1 ? qpair->id : 0; qpair->num_entries = num_entries; qpair->num_trackers = num_trackers; qpair->ctrlr = ctrlr; mtx_init(&qpair->lock, "nvme qpair lock", NULL, MTX_DEF); mtx_init(&qpair->recovery, "nvme qpair recovery", NULL, MTX_DEF); /* Note: NVMe PRP format is restricted to 4-byte alignment. */ err = bus_dma_tag_create(bus_get_dma_tag(ctrlr->dev), 4, ctrlr->page_size, BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL, ctrlr->max_xfer_size, howmany(ctrlr->max_xfer_size, ctrlr->page_size) + 1, ctrlr->page_size, 0, NULL, NULL, &qpair->dma_tag_payload); if (err != 0) { nvme_printf(ctrlr, "payload tag create failed %d\n", err); goto out; } /* * Each component must be page aligned, and individual PRP lists * cannot cross a page boundary. */ cmdsz = qpair->num_entries * sizeof(struct nvme_command); cmdsz = roundup2(cmdsz, ctrlr->page_size); cplsz = qpair->num_entries * sizeof(struct nvme_completion); cplsz = roundup2(cplsz, ctrlr->page_size); /* * For commands requiring more than 2 PRP entries, one PRP will be * embedded in the command (prp1), and the rest of the PRP entries * will be in a list pointed to by the command (prp2). */ prpsz = sizeof(uint64_t) * howmany(ctrlr->max_xfer_size, ctrlr->page_size); prpmemsz = qpair->num_trackers * prpsz; allocsz = cmdsz + cplsz + prpmemsz; err = bus_dma_tag_create(bus_get_dma_tag(ctrlr->dev), ctrlr->page_size, 0, BUS_SPACE_MAXADDR, BUS_SPACE_MAXADDR, NULL, NULL, allocsz, 1, allocsz, 0, NULL, NULL, &qpair->dma_tag); if (err != 0) { nvme_printf(ctrlr, "tag create failed %d\n", err); goto out; } bus_dma_tag_set_domain(qpair->dma_tag, qpair->domain); if (bus_dmamem_alloc(qpair->dma_tag, (void **)&queuemem, BUS_DMA_COHERENT | BUS_DMA_NOWAIT, &qpair->queuemem_map)) { nvme_printf(ctrlr, "failed to alloc qpair memory\n"); goto out; } if (bus_dmamap_load(qpair->dma_tag, qpair->queuemem_map, queuemem, allocsz, nvme_single_map, &queuemem_phys, 0) != 0) { nvme_printf(ctrlr, "failed to load qpair memory\n"); bus_dmamem_free(qpair->dma_tag, qpair->cmd, qpair->queuemem_map); goto out; } qpair->num_cmds = 0; qpair->num_intr_handler_calls = 0; qpair->num_retries = 0; qpair->num_failures = 0; qpair->num_ignored = 0; qpair->cmd = (struct nvme_command *)queuemem; qpair->cpl = (struct nvme_completion *)(queuemem + cmdsz); prpmem = (uint8_t *)(queuemem + cmdsz + cplsz); qpair->cmd_bus_addr = queuemem_phys; qpair->cpl_bus_addr = queuemem_phys + cmdsz; prpmem_phys = queuemem_phys + cmdsz + cplsz; callout_init_mtx(&qpair->timer, &qpair->recovery, 0); qpair->timer_armed = false; qpair->recovery_state = RECOVERY_WAITING; /* * Calcuate the stride of the doorbell register. Many emulators set this * value to correspond to a cache line. However, some hardware has set * it to various small values. */ qpair->sq_tdbl_off = nvme_mmio_offsetof(doorbell[0]) + (qpair->id << (ctrlr->dstrd + 1)); qpair->cq_hdbl_off = nvme_mmio_offsetof(doorbell[0]) + (qpair->id << (ctrlr->dstrd + 1)) + (1 << ctrlr->dstrd); TAILQ_INIT(&qpair->free_tr); TAILQ_INIT(&qpair->outstanding_tr); STAILQ_INIT(&qpair->queued_req); list_phys = prpmem_phys; prp_list = prpmem; for (i = 0; i < qpair->num_trackers; i++) { if (list_phys + prpsz > prpmem_phys + prpmemsz) { qpair->num_trackers = i; break; } /* * Make sure that the PRP list for this tracker doesn't * overflow to another nvme page. */ if (trunc_page(list_phys) != trunc_page(list_phys + prpsz - 1)) { list_phys = roundup2(list_phys, ctrlr->page_size); prp_list = (uint8_t *)roundup2((uintptr_t)prp_list, ctrlr->page_size); } tr = malloc_domainset(sizeof(*tr), M_NVME, DOMAINSET_PREF(qpair->domain), M_ZERO | M_WAITOK); bus_dmamap_create(qpair->dma_tag_payload, 0, &tr->payload_dma_map); tr->cid = i; tr->qpair = qpair; tr->prp = (uint64_t *)prp_list; tr->prp_bus_addr = list_phys; TAILQ_INSERT_HEAD(&qpair->free_tr, tr, tailq); list_phys += prpsz; prp_list += prpsz; } if (qpair->num_trackers == 0) { nvme_printf(ctrlr, "failed to allocate enough trackers\n"); goto out; } qpair->act_tr = malloc_domainset(sizeof(struct nvme_tracker *) * qpair->num_entries, M_NVME, DOMAINSET_PREF(qpair->domain), M_ZERO | M_WAITOK); if (ctrlr->msi_count > 1) { /* * MSI-X vector resource IDs start at 1, so we add one to * the queue's vector to get the corresponding rid to use. */ qpair->rid = qpair->vector + 1; qpair->res = bus_alloc_resource_any(ctrlr->dev, SYS_RES_IRQ, &qpair->rid, RF_ACTIVE); if (qpair->res == NULL) { nvme_printf(ctrlr, "unable to allocate MSI\n"); goto out; } if (bus_setup_intr(ctrlr->dev, qpair->res, INTR_TYPE_MISC | INTR_MPSAFE, NULL, nvme_qpair_msi_handler, qpair, &qpair->tag) != 0) { nvme_printf(ctrlr, "unable to setup MSI\n"); goto out; } if (qpair->id == 0) { bus_describe_intr(ctrlr->dev, qpair->res, qpair->tag, "admin"); } else { bus_describe_intr(ctrlr->dev, qpair->res, qpair->tag, "io%d", qpair->id - 1); } } return (0); out: nvme_qpair_destroy(qpair); return (ENOMEM); } static void nvme_qpair_destroy(struct nvme_qpair *qpair) { struct nvme_tracker *tr; callout_drain(&qpair->timer); if (qpair->tag) { bus_teardown_intr(qpair->ctrlr->dev, qpair->res, qpair->tag); qpair->tag = NULL; } if (qpair->act_tr) { free(qpair->act_tr, M_NVME); qpair->act_tr = NULL; } while (!TAILQ_EMPTY(&qpair->free_tr)) { tr = TAILQ_FIRST(&qpair->free_tr); TAILQ_REMOVE(&qpair->free_tr, tr, tailq); bus_dmamap_destroy(qpair->dma_tag_payload, tr->payload_dma_map); free(tr, M_NVME); } if (qpair->cmd != NULL) { bus_dmamap_unload(qpair->dma_tag, qpair->queuemem_map); bus_dmamem_free(qpair->dma_tag, qpair->cmd, qpair->queuemem_map); qpair->cmd = NULL; } if (qpair->dma_tag) { bus_dma_tag_destroy(qpair->dma_tag); qpair->dma_tag = NULL; } if (qpair->dma_tag_payload) { bus_dma_tag_destroy(qpair->dma_tag_payload); qpair->dma_tag_payload = NULL; } if (mtx_initialized(&qpair->lock)) mtx_destroy(&qpair->lock); if (mtx_initialized(&qpair->recovery)) mtx_destroy(&qpair->recovery); if (qpair->res) { bus_release_resource(qpair->ctrlr->dev, SYS_RES_IRQ, rman_get_rid(qpair->res), qpair->res); qpair->res = NULL; } } static void nvme_admin_qpair_abort_aers(struct nvme_qpair *qpair) { struct nvme_tracker *tr; /* * nvme_complete_tracker must be called without the qpair lock held. It * takes the lock to adjust outstanding_tr list, so make sure we don't * have it yet (since this is a general purpose routine). We take the * lock to make the list traverse safe, but have to drop the lock to * complete any AER. We restart the list scan when we do this to make * this safe. There's interlock with the ISR so we know this tracker * won't be completed twice. */ mtx_assert(&qpair->lock, MA_NOTOWNED); mtx_lock(&qpair->lock); tr = TAILQ_FIRST(&qpair->outstanding_tr); while (tr != NULL) { if (tr->req->cmd.opc == NVME_OPC_ASYNC_EVENT_REQUEST) { mtx_unlock(&qpair->lock); nvme_qpair_manual_complete_tracker(tr, NVME_SCT_GENERIC, NVME_SC_ABORTED_SQ_DELETION, 0, ERROR_PRINT_NONE); mtx_lock(&qpair->lock); tr = TAILQ_FIRST(&qpair->outstanding_tr); } else { tr = TAILQ_NEXT(tr, tailq); } } mtx_unlock(&qpair->lock); } void nvme_admin_qpair_destroy(struct nvme_qpair *qpair) { mtx_assert(&qpair->lock, MA_NOTOWNED); nvme_admin_qpair_abort_aers(qpair); nvme_qpair_destroy(qpair); } void nvme_io_qpair_destroy(struct nvme_qpair *qpair) { nvme_qpair_destroy(qpair); } static void nvme_abort_complete(void *arg, const struct nvme_completion *status) { struct nvme_tracker *tr = arg; /* * If cdw0 == 1, the controller was not able to abort the command * we requested. We still need to check the active tracker array, * to cover race where I/O timed out at same time controller was * completing the I/O. */ if (status->cdw0 == 1 && tr->qpair->act_tr[tr->cid] != NULL) { /* * An I/O has timed out, and the controller was unable to * abort it for some reason. Construct a fake completion * status, and then complete the I/O's tracker manually. */ nvme_printf(tr->qpair->ctrlr, "abort command failed, aborting command manually\n"); nvme_qpair_manual_complete_tracker(tr, NVME_SCT_GENERIC, NVME_SC_ABORTED_BY_REQUEST, 0, ERROR_PRINT_ALL); } } static void nvme_qpair_timeout(void *arg) { struct nvme_qpair *qpair = arg; struct nvme_controller *ctrlr = qpair->ctrlr; struct nvme_tracker *tr; sbintime_t now; bool idle = false; bool needs_reset; uint32_t csts; uint8_t cfs; mtx_assert(&qpair->recovery, MA_OWNED); switch (qpair->recovery_state) { case RECOVERY_NONE: /* * Read csts to get value of cfs - controller fatal status. If * we are in the hot-plug or controller failed status proceed * directly to reset. We also bail early if the status reads all * 1's or the control fatal status bit is now 1. The latter is * always true when the former is true, but not vice versa. The * intent of the code is that if the card is gone (all 1's) or * we've failed, then try to do a reset (which someitmes * unwedges a card reading all 1's that's not gone away, but * usually doesn't). */ csts = nvme_mmio_read_4(ctrlr, csts); - cfs = (csts >> NVME_CSTS_REG_CFS_SHIFT) & NVME_CSTS_REG_CFS_MASK; + cfs = NVMEV(NVME_CSTS_REG_CFS, csts); if (csts == NVME_GONE || cfs == 1) goto do_reset; /* * Process completions. We already have the recovery lock, so * call the locked version. */ _nvme_qpair_process_completions(qpair); /* * Check to see if we need to timeout any commands. If we do, then * we also enter a recovery phase. */ now = getsbinuptime(); needs_reset = false; idle = true; mtx_lock(&qpair->lock); TAILQ_FOREACH(tr, &qpair->outstanding_tr, tailq) { /* * Skip async commands, they are posted to the card for * an indefinite amount of time and have no deadline. */ if (tr->deadline == SBT_MAX) continue; if (now > tr->deadline) { if (tr->req->cb_fn != nvme_abort_complete && ctrlr->enable_aborts) { /* * This isn't an abort command, ask * for a hardware abort. */ nvme_ctrlr_cmd_abort(ctrlr, tr->cid, qpair->id, nvme_abort_complete, tr); } else { /* * Otherwise we have a live command in * the card (either one we couldn't * abort, or aborts weren't enabled). * The only safe way to proceed is to do * a reset. */ needs_reset = true; } } else { idle = false; } } mtx_unlock(&qpair->lock); if (!needs_reset) break; /* * We've had a command timeout that we weren't able to abort * * If we get here due to a possible surprise hot-unplug event, * then we let nvme_ctrlr_reset confirm and fail the * controller. */ do_reset: nvme_printf(ctrlr, "Resetting controller due to a timeout%s.\n", (csts == 0xffffffff) ? " and possible hot unplug" : (cfs ? " and fatal error status" : "")); qpair->recovery_state = RECOVERY_WAITING; nvme_ctrlr_reset(ctrlr); idle = false; /* We want to keep polling */ break; case RECOVERY_WAITING: /* * These messages aren't interesting while we're suspended. We * put the queues into waiting state while * suspending. Suspending takes a while, so we'll see these * during that time and they aren't diagnostic. At other times, * they indicate a problem that's worth complaining about. */ if (!device_is_suspended(ctrlr->dev)) nvme_printf(ctrlr, "Waiting for reset to complete\n"); idle = false; /* We want to keep polling */ break; case RECOVERY_FAILED: KASSERT(qpair->ctrlr->is_failed, ("Recovery state failed w/o failed controller\n")); idle = true; /* nothing to monitor */ break; } /* * Rearm the timeout. */ if (!idle) { callout_schedule_sbt(&qpair->timer, SBT_1S / 2, SBT_1S / 2, 0); } else { qpair->timer_armed = false; } } /* * Submit the tracker to the hardware. Must already be in the * outstanding queue when called. */ void nvme_qpair_submit_tracker(struct nvme_qpair *qpair, struct nvme_tracker *tr) { struct nvme_request *req; struct nvme_controller *ctrlr; int timeout; mtx_assert(&qpair->lock, MA_OWNED); req = tr->req; req->cmd.cid = tr->cid; qpair->act_tr[tr->cid] = tr; ctrlr = qpair->ctrlr; if (req->timeout) { if (req->cb_fn == nvme_completion_poll_cb) timeout = 1; else if (qpair->id == 0) timeout = ctrlr->admin_timeout_period; else timeout = ctrlr->timeout_period; tr->deadline = getsbinuptime() + timeout * SBT_1S; if (!qpair->timer_armed) { qpair->timer_armed = true; callout_reset_sbt_on(&qpair->timer, SBT_1S / 2, SBT_1S / 2, nvme_qpair_timeout, qpair, qpair->cpu, 0); } } else tr->deadline = SBT_MAX; /* Copy the command from the tracker to the submission queue. */ memcpy(&qpair->cmd[qpair->sq_tail], &req->cmd, sizeof(req->cmd)); if (++qpair->sq_tail == qpair->num_entries) qpair->sq_tail = 0; bus_dmamap_sync(qpair->dma_tag, qpair->queuemem_map, BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE); bus_space_write_4(qpair->ctrlr->bus_tag, qpair->ctrlr->bus_handle, qpair->sq_tdbl_off, qpair->sq_tail); qpair->num_cmds++; } static void nvme_payload_map(void *arg, bus_dma_segment_t *seg, int nseg, int error) { struct nvme_tracker *tr = arg; uint32_t cur_nseg; /* * If the mapping operation failed, return immediately. The caller * is responsible for detecting the error status and failing the * tracker manually. */ if (error != 0) { nvme_printf(tr->qpair->ctrlr, "nvme_payload_map err %d\n", error); return; } /* * Note that we specified ctrlr->page_size for alignment and max * segment size when creating the bus dma tags. So here we can safely * just transfer each segment to its associated PRP entry. */ tr->req->cmd.prp1 = htole64(seg[0].ds_addr); if (nseg == 2) { tr->req->cmd.prp2 = htole64(seg[1].ds_addr); } else if (nseg > 2) { cur_nseg = 1; tr->req->cmd.prp2 = htole64((uint64_t)tr->prp_bus_addr); while (cur_nseg < nseg) { tr->prp[cur_nseg-1] = htole64((uint64_t)seg[cur_nseg].ds_addr); cur_nseg++; } } else { /* * prp2 should not be used by the controller * since there is only one segment, but set * to 0 just to be safe. */ tr->req->cmd.prp2 = 0; } bus_dmamap_sync(tr->qpair->dma_tag_payload, tr->payload_dma_map, BUS_DMASYNC_PREREAD | BUS_DMASYNC_PREWRITE); nvme_qpair_submit_tracker(tr->qpair, tr); } static void _nvme_qpair_submit_request(struct nvme_qpair *qpair, struct nvme_request *req) { struct nvme_tracker *tr; int err = 0; mtx_assert(&qpair->lock, MA_OWNED); tr = TAILQ_FIRST(&qpair->free_tr); req->qpair = qpair; if (tr == NULL || qpair->recovery_state != RECOVERY_NONE) { /* * No tracker is available, or the qpair is disabled due to an * in-progress controller-level reset or controller failure. If * we lose the race with recovery_state, then we may add an * extra request to the queue which will be resubmitted later. * We only set recovery_state to NONE with qpair->lock also * held. */ if (qpair->ctrlr->is_failed) { /* * The controller has failed, so fail the request. */ nvme_qpair_manual_complete_request(qpair, req, NVME_SCT_GENERIC, NVME_SC_ABORTED_BY_REQUEST); } else { /* * Put the request on the qpair's request queue to be * processed when a tracker frees up via a command * completion or when the controller reset is * completed. */ STAILQ_INSERT_TAIL(&qpair->queued_req, req, stailq); } return; } TAILQ_REMOVE(&qpair->free_tr, tr, tailq); TAILQ_INSERT_TAIL(&qpair->outstanding_tr, tr, tailq); tr->deadline = SBT_MAX; tr->req = req; if (!req->payload_valid) { nvme_qpair_submit_tracker(tr->qpair, tr); return; } err = bus_dmamap_load_mem(tr->qpair->dma_tag_payload, tr->payload_dma_map, &req->payload, nvme_payload_map, tr, 0); if (err != 0) { /* * The dmamap operation failed, so we manually fail the * tracker here with DATA_TRANSFER_ERROR status. * * nvme_qpair_manual_complete_tracker must not be called * with the qpair lock held. */ nvme_printf(qpair->ctrlr, "bus_dmamap_load_mem returned 0x%x!\n", err); mtx_unlock(&qpair->lock); nvme_qpair_manual_complete_tracker(tr, NVME_SCT_GENERIC, NVME_SC_DATA_TRANSFER_ERROR, DO_NOT_RETRY, ERROR_PRINT_ALL); mtx_lock(&qpair->lock); } } void nvme_qpair_submit_request(struct nvme_qpair *qpair, struct nvme_request *req) { mtx_lock(&qpair->lock); _nvme_qpair_submit_request(qpair, req); mtx_unlock(&qpair->lock); } static void nvme_qpair_enable(struct nvme_qpair *qpair) { if (mtx_initialized(&qpair->recovery)) mtx_assert(&qpair->recovery, MA_OWNED); if (mtx_initialized(&qpair->lock)) mtx_assert(&qpair->lock, MA_OWNED); KASSERT(qpair->recovery_state != RECOVERY_FAILED, ("Enabling a failed qpair\n")); qpair->recovery_state = RECOVERY_NONE; } void nvme_qpair_reset(struct nvme_qpair *qpair) { qpair->sq_head = qpair->sq_tail = qpair->cq_head = 0; /* * First time through the completion queue, HW will set phase * bit on completions to 1. So set this to 1 here, indicating * we're looking for a 1 to know which entries have completed. * we'll toggle the bit each time when the completion queue * rolls over. */ qpair->phase = 1; memset(qpair->cmd, 0, qpair->num_entries * sizeof(struct nvme_command)); memset(qpair->cpl, 0, qpair->num_entries * sizeof(struct nvme_completion)); } void nvme_admin_qpair_enable(struct nvme_qpair *qpair) { struct nvme_tracker *tr; struct nvme_tracker *tr_temp; bool rpt; /* * Manually abort each outstanding admin command. Do not retry * admin commands found here, since they will be left over from * a controller reset and its likely the context in which the * command was issued no longer applies. */ rpt = !TAILQ_EMPTY(&qpair->outstanding_tr); if (rpt) nvme_printf(qpair->ctrlr, "aborting outstanding admin command\n"); TAILQ_FOREACH_SAFE(tr, &qpair->outstanding_tr, tailq, tr_temp) { nvme_qpair_manual_complete_tracker(tr, NVME_SCT_GENERIC, NVME_SC_ABORTED_BY_REQUEST, DO_NOT_RETRY, ERROR_PRINT_ALL); } if (rpt) nvme_printf(qpair->ctrlr, "done aborting outstanding admin\n"); mtx_lock(&qpair->recovery); mtx_lock(&qpair->lock); nvme_qpair_enable(qpair); mtx_unlock(&qpair->lock); mtx_unlock(&qpair->recovery); } void nvme_io_qpair_enable(struct nvme_qpair *qpair) { STAILQ_HEAD(, nvme_request) temp; struct nvme_tracker *tr; struct nvme_tracker *tr_temp; struct nvme_request *req; bool report; /* * Manually abort each outstanding I/O. This normally results in a * retry, unless the retry count on the associated request has * reached its limit. */ report = !TAILQ_EMPTY(&qpair->outstanding_tr); if (report) nvme_printf(qpair->ctrlr, "aborting outstanding i/o\n"); TAILQ_FOREACH_SAFE(tr, &qpair->outstanding_tr, tailq, tr_temp) { nvme_qpair_manual_complete_tracker(tr, NVME_SCT_GENERIC, NVME_SC_ABORTED_BY_REQUEST, 0, ERROR_PRINT_NO_RETRY); } if (report) nvme_printf(qpair->ctrlr, "done aborting outstanding i/o\n"); mtx_lock(&qpair->recovery); mtx_lock(&qpair->lock); nvme_qpair_enable(qpair); STAILQ_INIT(&temp); STAILQ_SWAP(&qpair->queued_req, &temp, nvme_request); report = !STAILQ_EMPTY(&temp); if (report) nvme_printf(qpair->ctrlr, "resubmitting queued i/o\n"); while (!STAILQ_EMPTY(&temp)) { req = STAILQ_FIRST(&temp); STAILQ_REMOVE_HEAD(&temp, stailq); nvme_qpair_print_command(qpair, &req->cmd); _nvme_qpair_submit_request(qpair, req); } if (report) nvme_printf(qpair->ctrlr, "done resubmitting i/o\n"); mtx_unlock(&qpair->lock); mtx_unlock(&qpair->recovery); } static void nvme_qpair_disable(struct nvme_qpair *qpair) { struct nvme_tracker *tr, *tr_temp; if (mtx_initialized(&qpair->recovery)) mtx_assert(&qpair->recovery, MA_OWNED); if (mtx_initialized(&qpair->lock)) mtx_assert(&qpair->lock, MA_OWNED); qpair->recovery_state = RECOVERY_WAITING; TAILQ_FOREACH_SAFE(tr, &qpair->outstanding_tr, tailq, tr_temp) { tr->deadline = SBT_MAX; } } void nvme_admin_qpair_disable(struct nvme_qpair *qpair) { mtx_lock(&qpair->recovery); mtx_lock(&qpair->lock); nvme_qpair_disable(qpair); mtx_unlock(&qpair->lock); nvme_admin_qpair_abort_aers(qpair); mtx_unlock(&qpair->recovery); } void nvme_io_qpair_disable(struct nvme_qpair *qpair) { mtx_lock(&qpair->recovery); mtx_lock(&qpair->lock); nvme_qpair_disable(qpair); mtx_unlock(&qpair->lock); mtx_unlock(&qpair->recovery); } void nvme_qpair_fail(struct nvme_qpair *qpair) { struct nvme_tracker *tr; struct nvme_request *req; if (!mtx_initialized(&qpair->lock)) return; mtx_lock(&qpair->recovery); qpair->recovery_state = RECOVERY_FAILED; mtx_unlock(&qpair->recovery); mtx_lock(&qpair->lock); if (!STAILQ_EMPTY(&qpair->queued_req)) { nvme_printf(qpair->ctrlr, "failing queued i/o\n"); } while (!STAILQ_EMPTY(&qpair->queued_req)) { req = STAILQ_FIRST(&qpair->queued_req); STAILQ_REMOVE_HEAD(&qpair->queued_req, stailq); mtx_unlock(&qpair->lock); nvme_qpair_manual_complete_request(qpair, req, NVME_SCT_GENERIC, NVME_SC_ABORTED_BY_REQUEST); mtx_lock(&qpair->lock); } if (!TAILQ_EMPTY(&qpair->outstanding_tr)) { nvme_printf(qpair->ctrlr, "failing outstanding i/o\n"); } /* Manually abort each outstanding I/O. */ while (!TAILQ_EMPTY(&qpair->outstanding_tr)) { tr = TAILQ_FIRST(&qpair->outstanding_tr); /* * Do not remove the tracker. The abort_tracker path will * do that for us. */ mtx_unlock(&qpair->lock); nvme_qpair_manual_complete_tracker(tr, NVME_SCT_GENERIC, NVME_SC_ABORTED_BY_REQUEST, DO_NOT_RETRY, ERROR_PRINT_ALL); mtx_lock(&qpair->lock); } mtx_unlock(&qpair->lock); }