scsi: ufs: Add Host Performance Booster Support
A message that is not in the archive (sent off-list or not yet received): CGME20200623010201epcms2p11aebdf1fbc719b409968cba997507114@epcms2p1
v3 cover scsi: ufs: Add Host Performance Booster Support
Changelog:
v2 -> v3
1. Add checking input module parameter value.
2. Change base commit from 5.8/scsi-queue to 5.9/scsi-queue.
3. Cleanup for unused variables and label.
v1 -> v2
1. Change the full boilerplate text to SPDX style.
2. Adopt dynamic allocation for sub-region data structure.
3. Cleanup.
NAND flash memory-based storage devices use Flash Translation Layer (FTL) to translate logical addresses of I/O requests to corresponding flash memory addresses. Mobile storage devices typically have RAM with constrained size, thus lack in memory to keep the whole mapping table. Therefore, mapping tables are partially retrieved from NAND flash on demand, causing random-read performance degradation.
To improve random read performance, JESD220-3 (HPB v1.0) proposes HPB (Host Performance Booster) which uses host system memory as a cache for the FTL mapping table. By using HPB, FTL data can be read from host memory faster than from NAND flash memory.
The current version only supports the DCM (device control mode). This patch consists of 4 parts to support HPB feature.
1) UFS-feature layer
2) HPB probe and initialization process
3) READ -> HPB READ using cached map information
4) L2P (logical to physical) map management
The UFS-feature is an additional layer to avoid the structure in which the UFS-core driver and the UFS-feature are entangled with each other in a single module.
By adding the layer, UFS-features composed of various combinations can be supported. Also, even if a new feature is added, modification of the UFS-core driver can be minimized.
In the HPB probe and init process, the device information of the UFS is queried. After checking supported features, the data structure for the HPB is initialized according to the device information.
A read I/O in the active sub-region where the map is cached is changed to HPB READ by the HPB module.
The HPB module manages the L2P map using information received from the device. For active sub-region, the HPB module caches through ufshpb_map request. For the in-active region, the HPB module discards the L2P map. When a write I/O occurs in an active sub-region area, associated dirty bitmap checked as dirty for preventing stale read.
HPB is shown to have a performance improvement of 58 - 67% for random read workload. [1]
This series patches are based on the 5.9/scsi-queue branch.
[1]:
https://www.usenix.org/conference/hotstorage17/program/presentation/jeong
Daejun park (5):
scsi: ufs: Add UFS feature related parameter
scsi: ufs: Add UFS feature layer
scsi: ufs: Introduce HPB module
scsi: ufs: L2P map management for HPB read
scsi: ufs: Prepare HPB read for cached sub-region
drivers/scsi/ufs/Kconfig | 9 + drivers/scsi/ufs/Makefile | 3 +- drivers/scsi/ufs/ufs.h | 12 + drivers/scsi/ufs/ufsfeature.c | 148 +++ drivers/scsi/ufs/ufsfeature.h | 69 ++ drivers/scsi/ufs/ufshcd.c | 23 +- drivers/scsi/ufs/ufshcd.h | 3 + drivers/scsi/ufs/ufshpb.c | 1996 ++++++++++++++++++++++++++++++++++++ drivers/scsi/ufs/ufshpb.h | 234 +++++ 9 files changed, 2494 insertions(+), 3 deletions(-)
created mode 100644 drivers/scsi/ufs/ufsfeature.c created mode 100644 drivers/scsi/ufs/ufsfeature.h created mode 100644 drivers/scsi/ufs/ufshpb.c created mode 100644 drivers/scsi/ufs/ufshpb.h
RFC PATCH v3 1/5 scsi: ufs: Add UFS feature related parameter
This is a patch for parameters to be used for UFS features layer and HPB module.
--- drivers/scsi/ufs/ufs.h | 12 ++++++++++++ 1 file changed, 12 insertions(+)
drivers/scsi/ufs/ufs.h +12 -0
@@ -122,6 +122,7 @@ enum flag_idn {QUERY_FLAG_IDN_WB_EN =0x0E ,QUERY_FLAG_IDN_WB_BUFF_FLUSH_EN =0x0F ,QUERY_FLAG_IDN_WB_BUFF_FLUSH_DURING_HIBERN8 =0x10 , + QUERY_FLAG_IDN_HPB_RESET = 0x11, };/* Attribute idn for Query requests */
@@ -195,6 +196,9 @@ enum unit_desc_param {UNIT_DESC_PARAM_PHY_MEM_RSRC_CNT =0x18 ,UNIT_DESC_PARAM_CTX_CAPABILITIES =0x20 ,UNIT_DESC_PARAM_LARGE_UNIT_SIZE_M1 =0x22 , +UNIT_DESC_HPB_LU_MAX_ACTIVE_REGIONS =0x23 , +UNIT_DESC_HPB_LU_PIN_REGION_START_OFFSET =0x25 , +UNIT_DESC_HPB_LU_NUM_PIN_REGIONS =0x27 , UNIT_DESC_PARAM_WB_BUF_ALLOC_UNITS = 0x29, };
@@ -235,6 +239,8 @@ enum device_desc_param {DEVICE_DESC_PARAM_PSA_MAX_DATA =0x25 ,DEVICE_DESC_PARAM_PSA_TMT =0x29 ,DEVICE_DESC_PARAM_PRDCT_REV =0x2A , +DEVICE_DESC_PARAM_HPB_VER =0x40 , +DEVICE_DESC_PARAM_HPB_CONTROL =0x42 ,DEVICE_DESC_PARAM_EXT_UFS_FEATURE_SUP =0x4F ,DEVICE_DESC_PARAM_WB_PRESRV_USRSPC_EN =0x53 , DEVICE_DESC_PARAM_WB_TYPE = 0x54,
@@ -283,6 +289,10 @@ enum geometry_desc_param {GEOMETRY_DESC_PARAM_ENM4_MAX_NUM_UNITS =0x3E ,GEOMETRY_DESC_PARAM_ENM4_CAP_ADJ_FCTR =0x42 ,GEOMETRY_DESC_PARAM_OPT_LOG_BLK_SIZE =0x44 , +GEOMETRY_DESC_HPB_REGION_SIZE =0x48 , +GEOMETRY_DESC_HPB_NUMBER_LU =0x49 , +GEOMETRY_DESC_HPB_SUBREGION_SIZE =0x4A , +GEOMETRY_DESC_HPB_DEVICE_MAX_ACTIVE_REGIONS =0x4B ,GEOMETRY_DESC_PARAM_WB_MAX_ALLOC_UNITS =0x4F ,GEOMETRY_DESC_PARAM_WB_MAX_WB_LUNS =0x53 , GEOMETRY_DESC_PARAM_WB_BUFF_CAP_ADJ = 0x54,
@@ -327,6 +337,7 @@ enum {/* Possible values for dExtendedUFSFeaturesSupport */ enum { +UFS_DEV_HPB_SUPPORT =BIT (7 ),UFS_DEV_WRITE_BOOSTER_SUP =BIT (8 ), };
@@ -537,6 +548,7 @@ struct ufs_dev_info {u8 *model;u16 wspecversion;u32 clk_gating_wait_us; +u8 b_ufs_feature_sup;u32 d_ext_ufs_feature_sup;u8 b_wb_buffer_type;u32 d_wb_alloc_units;
base-commit: 3145550a7f8b08356c8ff29feaa6c56aca12901d
Signature · 1 line
2.17.1
RFC PATCH v3 2/5 scsi: ufs: Add UFS-feature layer
This patch is adding UFS feature layer to UFS core driver.
UFS Driver data structure (struct ufs_hba)
│ ┌--------------┐ │ UFS feature │ <-- HPB module │ layer │ <-- other extended feature module └--------------┘
Each extended UFS-Feature module has a bus of ufs-ext feature type. The UFS feature layer manages common APIs used by each extended feature module. The APIs are set of UFS Query requests and UFS Vendor commands related to each extended feature module.
Other extended features can also be implemented using the proposed APIs. For example, in Write Booster, "prep_fn" can be used to guarantee the lifetime of UFS by updating the amount of write IO.
And reset/reset_host/suspend/resume can be used to manage the kernel task for checking lifetime of UFS.
The following 6 callback functions have been added to "ufshcd.c".prep_fn: called after construct upiu structure
reset: called after proving hbareset_host: called before ufshcd_host_reset_and_restore
suspend: called before ufshcd_suspend
resume: called after ufshcd_resumersp_upiu: called in ufshcd_transfer_rsp_status with SAM_STAT_GOOD state
--- drivers/scsi/ufs/Makefile | 2 +- drivers/scsi/ufs/ufsfeature.c | 148 ++++++++++++++++++++++++++++++++++ drivers/scsi/ufs/ufsfeature.h | 69 ++++++++++++++++ drivers/scsi/ufs/ufshcd.c | 17 ++++ drivers/scsi/ufs/ufshcd.h | 3 + 5 files changed, 238 insertions(+), 1 deletion(-) create mode 100644 drivers/scsi/ufs/ufsfeature.c create mode 100644 drivers/scsi/ufs/ufsfeature.h
drivers/scsi/ufs/Makefile +1 -1
@@ -6,7 +6,7 @@ obj-$(CONFIG_SCSI_UFS_CDNS_PLATFORM) += cdns-pltfrm.o obj-$(CONFIG_SCSI_UFS_QCOM) += ufs-qcom.o obj-$(CONFIG_SCSI_UFS_EXYNOS) += ufs-exynos.o obj-$(CONFIG_SCSI_UFSHCD) += ufshcd-core.o -ufshcd-core-y += ufshcd.o ufs-sysfs.o +ufshcd-core-y += ufshcd.o ufs-sysfs.o ufsfeature.o ufshcd-core-$(CONFIG_SCSI_UFS_BSG) += ufs_bsg.o obj-$(CONFIG_SCSI_UFSHCD_PCI) += ufshcd-pci.o obj-$(CONFIG_SCSI_UFSHCD_PLATFORM) += ufshcd-pltfrm.o
drivers/scsi/ufs/ufsfeature.c new file +148 -0
148 lines not on this page · raw message
drivers/scsi/ufs/ufsfeature.h new file +69 -0
69 lines not on this page · raw message
drivers/scsi/ufs/ufshcd.c +17 -0
@@ -2533,6 +2533,8 @@ static int ufshcd_queuecommand(struct Scsi_Host *host, struct scsi_cmnd *cmd)ufshcd_comp_scsi_upiu (hba, lrbp); +ufsf_ops_prep_fn (hba, lrbp); + err =ufshcd_map_sg (hba, lrbp);if (err) { lrbp->cmd =NULL ;
@@ -4665,6 +4667,8 @@ static int ufshcd_slave_configure(struct scsi_device *sdev)struct ufs_hba *hba =shost_priv (sdev->host);struct request_queue *q = sdev->request_queue; +ufsf_slave_configure (hba, sdev); +blk_queue_update_dma_pad (q,PRDT_DATA_BYTE_COUNT_PAD -1 );if (ufshcd_is_rpm_autosuspend_allowed (hba))
@@ -4791,6 +4795,9 @@ ufshcd_transfer_rsp_status(struct ufs_hba *hba, struct ufshcd_lrb *lrbp)*/ pm_runtime_get_noresume (hba->dev); } + +if (scsi_status ==SAM_STAT_GOOD ) +ufsf_ops_rsp_upiu (hba, lrbp);break ;case UPIU_TRANSACTION_REJECT_UPIU :/* TODO: handle Reject UPIU Response */
@@ -6539,6 +6546,8 @@ static int ufshcd_host_reset_and_restore(struct ufs_hba *hba)* Stop the host controller and complete the requests * cleared by h/w */ +ufsf_ops_reset_host (hba); +ufshcd_hba_stop (hba);spin_lock_irqsave (hba->host->host_lock, flags);
@@ -6963,6 +6972,7 @@ static int ufs_get_device_desc(struct ufs_hba *hba)/* getting Specification Version in big endian format */ dev_info->wspecversion = desc_buf[DEVICE_DESC_PARAM_SPEC_VER ] <<8 | desc_buf[DEVICE_DESC_PARAM_SPEC_VER +1 ]; + dev_info->b_ufs_feature_sup = desc_buf[DEVICE_DESC_PARAM_UFS_FEAT ]; model_index = desc_buf[DEVICE_DESC_PARAM_PRDCT_NAME ];
@@ -7340,6 +7350,7 @@ static int ufshcd_add_lus(struct ufs_hba *hba) }ufs_bsg_probe (hba); +ufsf_scan_features (hba);scsi_scan_host (hba->host);pm_runtime_put_sync (hba->dev);
@@ -7428,6 +7439,7 @@ static int ufshcd_probe_hba(struct ufs_hba *hba, bool async)/* Enable Auto-Hibernate if configured */ ufshcd_auto_hibern8_enable (hba); +ufsf_ops_reset (hba); out: trace_ufshcd_init(dev_name(hba->dev), ret,
@@ -8185,6 +8197,8 @@ static int ufshcd_suspend(struct ufs_hba *hba, enum ufs_pm_op pm_op) req_link_state =UIC_LINK_OFF_STATE ; } +ufsf_ops_suspend (hba); +/* * If we can't transition into any of the low power modes * just gate the clocks.
@@ -8306,6 +8320,7 @@ static int ufshcd_suspend(struct ufs_hba *hba, enum ufs_pm_op pm_op) hba->clk_gating.is_suspended = false; hba->dev_info.b_rpm_dev_flush_capable = false;ufshcd_release (hba); +ufsf_ops_resume (hba); out:if (hba->dev_info.b_rpm_dev_flush_capable) { schedule_delayed_work(&hba->rpm_dev_flush_recheck_work,
@@ -8402,6 +8417,8 @@ static int ufshcd_resume(struct ufs_hba *hba, enum ufs_pm_op pm_op)/* Enable Auto-Hibernate if configured */ ufshcd_auto_hibern8_enable (hba); +ufsf_ops_resume (hba); +if (hba->dev_info.b_rpm_dev_flush_capable) { hba->dev_info.b_rpm_dev_flush_capable = false;cancel_delayed_work (&hba->rpm_dev_flush_recheck_work);
drivers/scsi/ufs/ufshcd.h +3 -0
@@ -46,6 +46,7 @@#include "ufs.h" #include "ufs_quirks.h" #include "ufshci.h" +#include "ufsfeature.h" #define UFSHCD "ufshcd"#define UFSHCD_DRIVER_VERSION "0.2"
@@ -736,6 +737,8 @@ struct ufs_hba {bool wb_buf_flush_enabled;bool wb_enabled;struct delayed_work rpm_dev_flush_recheck_work; + +struct ufsf_feature_info ufsf; };/* Returns true if clocks can be gated. Otherwise false */
Signature · 1 line
2.17.1
RFC PATCH v3 3/5 scsi: ufs: Introduce HPB module
This is a patch for the HPB module.
The HPB module queries UFS for device information during initialization. We added the export symbol to two functions in ufshcd.c to initialize the HPB module.
The HPB module can be loaded or built-in as needed.
The mininum size of the memory pool used in the HPB module is implemented as a module parameter, so that it can be configurable by the user.
To gurantee a minimum memory pool size of 4MB:
$ insmod ufshpb.ko ufshpb_host_map_kbytes=4096
--- drivers/scsi/ufs/Kconfig | 9 + drivers/scsi/ufs/Makefile | 1 + drivers/scsi/ufs/ufshcd.c | 2 + drivers/scsi/ufs/ufshpb.c | 777 ++++++++++++++++++++++++++++++++++++++ drivers/scsi/ufs/ufshpb.h | 162 ++++++++ 5 files changed, 951 insertions(+) create mode 100644 drivers/scsi/ufs/ufshpb.c create mode 100644 drivers/scsi/ufs/ufshpb.h
drivers/scsi/ufs/Kconfig +9 -0
@@ -172,3 +172,12 @@ config SCSI_UFS_EXYNOS Select this if you have UFS host controller on EXYNOS chipset. If unsure, say N. + +config UFSHPB + tristate "Support UFS Host Performance Booster" + depends on SCSI_UFSHCD + help + A UFS HPB Feature improves random read performance. It caches + L2P map of UFS to host DRAM. The driver uses HPB read command + by piggybacking physical page number for bypassing FTL's L2P address + translation.
drivers/scsi/ufs/Makefile +1 -0
@@ -8,6 +8,7 @@ obj-$(CONFIG_SCSI_UFS_EXYNOS) += ufs-exynos.o obj-$(CONFIG_SCSI_UFSHCD) += ufshcd-core.o ufshcd-core-y += ufshcd.o ufs-sysfs.o ufsfeature.o ufshcd-core-$(CONFIG_SCSI_UFS_BSG) += ufs_bsg.o +obj-$(CONFIG_UFSHPB) += ufshpb.o obj-$(CONFIG_SCSI_UFSHCD_PCI) += ufshcd-pci.o obj-$(CONFIG_SCSI_UFSHCD_PLATFORM) += ufshcd-pltfrm.o obj-$(CONFIG_SCSI_UFS_HISI) += ufs-hisi.o
drivers/scsi/ufs/ufshcd.c +2 -0
@@ -2863,6 +2863,7 @@ int ufshcd_query_flag(struct ufs_hba *hba, enum query_opcode opcode,ufshcd_release (hba);return err; } +EXPORT_SYMBOL (ufshcd_query_flag);/** * ufshcd_query_attr - API function for sending attribute requests
@@ -3061,6 +3062,7 @@ int ufshcd_query_descriptor_retry(struct ufs_hba *hba,return err; } +EXPORT_SYMBOL (ufshcd_query_descriptor_retry);/** * ufshcd_map_desc_id_to_length - map descriptor IDN to its length
drivers/scsi/ufs/ufshpb.c new file +777 -0
777 lines not on this page · raw message
drivers/scsi/ufs/ufshpb.h new file +162 -0
162 lines not on this page · raw message
Signature · 1 line
2.17.1
RFC PATCH v3 4/5 scsi: ufs: L2P map management for HPB read
This is a patch for managing L2P map in HPB module.
The HPB divides logical addresses into several regions. A region consists of several sub-regions. The sub-region is a basic unit where L2P mapping is managed. The driver loads L2P mapping data of each sub-region. The loaded sub-region is called active-state. The HPB driver unloads L2P mapping data as region unit. The unloaded region is called inactive-state.
Sub-region/region candidates to be loaded and unloaded are delivered from the UFS device. The UFS device delivers the recommended active sub-region and inactivate region to the driver using sensedata.
The HPB module performs L2P mapping management on the host through the delivered information.
A pinned region is a pre-set regions on the UFS device that is always activate-state.
The data structure for map data request and L2P map uses mempool API, minimizing allocation overhead while avoiding static allocation.
The map_work manages active/inactive by 2 "to-do" lists. Each hpb lun maintains 2 "to-do" lists:
hpb->lh_inact_rgn - regions to be inactivated, and hpb->lh_act_srgn - subregions to be activated
Those lists are maintained on IO completion.
--- drivers/scsi/ufs/ufshpb.c | 992 +++++++++++++++++++++++++++++++++++++- drivers/scsi/ufs/ufshpb.h | 72 +++ 2 files changed, 1060 insertions(+), 4 deletions(-)
drivers/scsi/ufs/ufshpb.c +988 -4
@@ -15,6 +15,7 @@#include "ufshpb.h" static struct ufshpb_driver ufshpb_drv; +unsigned int ufshpb_host_map_kbytes =1 *1024 ;static int ufshpb_create_sysfs (struct ufs_hba *hba,struct ufshpb_lu *hpb);
@@ -25,6 +26,63 @@ static inline int ufshpb_is_valid_srgn(struct ufshpb_region *rgn, srgn->srgn_state ==HPB_SRGN_VALID ; } +static inline bool ufshpb_is_general_lun (int lun) +{ +return lun <UFS_UPIU_MAX_UNIT_NUM_ID ; +} + +static inline bool +ufshpb_is_pinned_region (struct ufshpb_lu *hpb,int rgn_idx) +{ +if (hpb->lu_pinned_end !=PINNED_NOT_SET && + rgn_idx >= hpb->lu_pinned_start && + rgn_idx <= hpb->lu_pinned_end) +return true; + +return false; +} + +static bool ufshpb_is_empty_rsp_lists (struct ufshpb_lu *hpb) +{ +bool ret = true; +unsigned long flags; + +spin_lock_irqsave (&hpb->rsp_list_lock, flags); +if (!list_empty (&hpb->lh_inact_rgn) || !list_empty (&hpb->lh_act_srgn)) + ret = false; +spin_unlock_irqrestore (&hpb->rsp_list_lock, flags); + +return ret; +} + +static inline int ufshpb_may_field_valid (struct ufs_hba *hba, +struct ufshcd_lrb *lrbp, +struct ufshpb_rsp_field *rsp_field) +{ +if (be16_to_cpu (rsp_field->sense_data_len) !=DEV_SENSE_SEG_LEN || + rsp_field->desc_type !=DEV_DES_TYPE || + rsp_field->additional_len !=DEV_ADDITIONAL_LEN || + rsp_field->hpb_type ==HPB_RSP_NONE || + rsp_field->active_rgn_cnt >MAX_ACTIVE_NUM || + rsp_field->inactive_rgn_cnt >MAX_INACTIVE_NUM || + (!rsp_field->active_rgn_cnt && !rsp_field->inactive_rgn_cnt)) +return -EINVAL ; + +if (!ufshpb_is_general_lun (lrbp->lun)) { +dev_warn (hba->dev,"ufshpb: lun(%d) not supported\n" , + lrbp->lun); +return -EINVAL ; + } + +return 0 ; +} + + +static inline struct ufshpb_lu *ufshpb_get_hpb_data (struct scsi_cmnd *cmd) +{ +return cmd->device->hostdata; +} +static inline int ufshpb_get_state (struct ufshpb_lu *hpb) {return atomic_read (&hpb->hpb_state);
@@ -59,6 +117,765 @@ static inline void ufshpb_lu_put(struct ufshpb_lu *hpb)put_device (&hpb->hpb_lu_dev); } +static struct ufshpb_req *ufshpb_get_map_req (struct ufshpb_lu *hpb, +struct ufshpb_subregion *srgn) +{ +struct ufshpb_req *map_req; +struct request *req; +struct bio *bio; + + map_req =kmem_cache_alloc (hpb->map_req_cache,GFP_KERNEL ); +if (!map_req) +return NULL ; + + req =blk_get_request (hpb->sdev_ufs_lu->request_queue, +REQ_OP_SCSI_IN ,BLK_MQ_REQ_PREEMPT ); +if (IS_ERR (req)) + goto free_map_req; + + bio =bio_alloc (GFP_KERNEL , hpb->pages_per_srgn); +if (!bio) { +blk_put_request (req); + goto free_map_req; + } + + map_req->hpb = hpb; + map_req->req = req; + map_req->bio = bio; + + map_req->rgn_idx = srgn->rgn_idx; + map_req->srgn_idx = srgn->srgn_idx; + map_req->mctx = srgn->mctx; + map_req->lun = hpb->lun; + +return map_req; + +free_map_req: +kmem_cache_free (hpb->map_req_cache, map_req); +return NULL ; +} + +static inline void ufshpb_put_map_req (struct ufshpb_lu *hpb, +struct ufshpb_req *map_req) +{ +bio_put (map_req->bio); +blk_put_request (map_req->req); +kmem_cache_free (hpb->map_req_cache, map_req); +} + + +static inline int ufshpb_clear_dirty_bitmap (struct ufshpb_lu *hpb, +struct ufshpb_subregion *srgn) +{ +WARN_ON (!srgn->mctx); +bitmap_zero (srgn->mctx->ppn_dirty, hpb->entries_per_srgn); +return 0 ; +} + +static void ufshpb_update_active_info (struct ufshpb_lu *hpb,int rgn_idx, +int srgn_idx) +{ +struct ufshpb_region *rgn; +struct ufshpb_subregion *srgn; + + rgn = hpb->rgn_tbl + rgn_idx; + srgn = rgn->srgn_tbl + srgn_idx; + +list_del_init (&rgn->list_inact_rgn); + +if (list_empty (&srgn->list_act_srgn)) +list_add_tail (&srgn->list_act_srgn, &hpb->lh_act_srgn); +} + +static void ufshpb_update_inactive_info (struct ufshpb_lu *hpb,int rgn_idx) +{ +struct ufshpb_region *rgn; +struct ufshpb_subregion *srgn; +int srgn_idx; + + rgn = hpb->rgn_tbl + rgn_idx; + +for (srgn_idx =0 ; srgn_idx < rgn->srgn_cnt; srgn_idx++) { + srgn = rgn->srgn_tbl + srgn_idx; + +list_del_init (&srgn->list_act_srgn); + } + +if (list_empty (&rgn->list_inact_rgn)) +list_add_tail (&rgn->list_inact_rgn, &hpb->lh_inact_rgn); +} + +static void ufshpb_activate_subregion (struct ufshpb_lu *hpb, +struct ufshpb_subregion *srgn) +{ +struct ufshpb_region *rgn; + +/* +* If there is no mctx in subregion +* after I/O progress for HPB_READ_BUFFER, the region to which the +* subregion belongs was evicted. +* Mask sure the the region must not evict in I/O progress +*/ +WARN_ON (!srgn->mctx); + + rgn = hpb->rgn_tbl + srgn->rgn_idx; + +if (unlikely (rgn->rgn_state ==HPB_RGN_INACTIVE )) { +dev_err (&hpb->hpb_lu_dev, +"region %d subregion %d evicted\n" , + srgn->rgn_idx, srgn->srgn_idx); +return ; + } + srgn->srgn_state =HPB_SRGN_VALID ; +} + +static void ufshpb_map_req_compl_fn (struct request *req,blk_status_t error) +{ +struct ufshpb_req *map_req = (struct ufshpb_req *) req->end_io_data; +struct ufshpb_lu *hpb = map_req->hpb; +struct ufshpb_subregion *srgn; +unsigned long flags; + + srgn = hpb->rgn_tbl[map_req->rgn_idx].srgn_tbl + + map_req->srgn_idx; + +spin_lock_irqsave (&hpb->hpb_state_lock, flags); +ufshpb_activate_subregion (hpb, srgn); +spin_unlock_irqrestore (&hpb->hpb_state_lock, flags); + +ufshpb_put_map_req (map_req->hpb, map_req); +ufshpb_lu_put (hpb); +} + +static inline void ufshpb_set_read_buf_cmd (unsigned char *cdb,int rgn_idx, +int srgn_idx,int srgn_mem_size) +{ + cdb[0 ] =UFSHPB_READ_BUFFER ; + cdb[1 ] =UFSHPB_READ_BUFFER_ID ; + +put_unaligned_be16 (rgn_idx, &cdb[2 ]); +put_unaligned_be16 (srgn_idx, &cdb[4 ]); +put_unaligned_be24 (srgn_mem_size, &cdb[6 ]); + + cdb[9 ] =0x00 ; +} + +static int ufshpb_map_req_add_bio_page (struct ufshpb_lu *hpb, +struct request_queue *q,struct bio *bio, +struct ufshpb_map_ctx *mctx) +{ +int i, ret =0 ; + +for (i =0 ; i < hpb->pages_per_srgn; i++) { + ret =bio_add_pc_page (q, bio, mctx->m_page[i],PAGE_SIZE ,0 ); +if (ret !=PAGE_SIZE ) { +dev_notice (&hpb->hpb_lu_dev, +"bio_add_pc_page fail %d\n" , ret); +return -ENOMEM ; + } + } + +return 0 ; +} + +static int ufshpb_execute_map_req (struct ufshpb_lu *hpb, +struct ufshpb_req *map_req) +{ +struct request_queue *q; +struct request *req; +struct scsi_request *rq; +int ret =0 ; + + q = hpb->sdev_ufs_lu->request_queue; + ret =ufshpb_map_req_add_bio_page (hpb, q, map_req->bio, + map_req->mctx); +if (ret) { +dev_notice (&hpb->hpb_lu_dev, +"map_req_add_bio_page fail %d - %d\n" , + map_req->rgn_idx, map_req->srgn_idx); +return ret; + } + + req = map_req->req; + +blk_rq_append_bio (req, &map_req->bio); + + req->timeout =0 ; + req->end_io_data = (void *)map_req; + + rq =scsi_req (req); +ufshpb_set_read_buf_cmd (rq->cmd, map_req->rgn_idx, + map_req->srgn_idx, hpb->srgn_mem_size); + rq->cmd_len =HPB_READ_BUFFER_CMD_LENGTH ; + +blk_execute_rq_nowait (q,NULL , req,1 , ufshpb_map_req_compl_fn); + +atomic_inc (&hpb->stats.map_req_cnt); +return 0 ; +} + +static struct ufshpb_map_ctx *ufshpb_get_map_ctx (struct ufshpb_lu *hpb) +{ +struct ufshpb_map_ctx *mctx; +int i, j; + + mctx =mempool_alloc (ufshpb_drv.ufshpb_mctx_pool,GFP_KERNEL ); +if (!mctx) +return NULL ; + + mctx->m_page =kmem_cache_alloc (hpb->m_page_cache,GFP_KERNEL ); +if (!mctx->m_page) + goto release_mctx; + + mctx->ppn_dirty =bitmap_zalloc (hpb->entries_per_srgn,GFP_KERNEL ); +if (!mctx->ppn_dirty) + goto release_m_page; + +for (i =0 ; i < hpb->pages_per_srgn; i++) { + mctx->m_page[i] =mempool_alloc (ufshpb_drv.ufshpb_page_pool, +GFP_KERNEL ); +if (!mctx->m_page[i]) { +for (j =0 ; j < i; j++) +mempool_free (mctx->m_page[j], + ufshpb_drv.ufshpb_page_pool); + goto release_ppn_dirty; + } +clear_page (page_address (mctx->m_page[i])); + } + +return mctx; +release_ppn_dirty: +bitmap_free (mctx->ppn_dirty); +release_m_page: +kmem_cache_free (hpb->m_page_cache, mctx->m_page); +release_mctx: +mempool_free (mctx, ufshpb_drv.ufshpb_mctx_pool); +return NULL ; +} + +static inline void ufshpb_put_map_ctx (struct ufshpb_lu *hpb, +struct ufshpb_map_ctx *mctx) +{ +int i; + +for (i =0 ; i < hpb->pages_per_srgn; i++) +mempool_free (mctx->m_page[i], + ufshpb_drv.ufshpb_page_pool); + +bitmap_free (mctx->ppn_dirty); +kmem_cache_free (hpb->m_page_cache, mctx->m_page); +mempool_free (mctx, ufshpb_drv.ufshpb_mctx_pool); +} + +static int ufshpb_check_issue_state_srgns (struct ufshpb_lu *hpb, +struct ufshpb_region *rgn) +{ +struct ufshpb_subregion *srgn; +int srgn_idx; + +for (srgn_idx =0 ; srgn_idx < rgn->srgn_cnt; srgn_idx++) { + srgn = rgn->srgn_tbl + srgn_idx; + +if (srgn->srgn_state ==HPB_SRGN_ISSUED ) +return -EPERM ; + } +return 0 ; +} + +static inline void ufshpb_add_lru_info (struct victim_select_info *lru_info, +struct ufshpb_region *rgn) +{ + rgn->rgn_state =HPB_RGN_ACTIVE ; +list_add_tail (&rgn->list_lru_rgn, &lru_info->lh_lru_rgn); +atomic_inc (&lru_info->active_cnt); +} + +static inline void ufshpb_hit_lru_info (struct victim_select_info *lru_info, +struct ufshpb_region *rgn) +{ +list_move_tail (&rgn->list_lru_rgn, &lru_info->lh_lru_rgn); +} + +static struct ufshpb_region *ufshpb_victim_lru_info (struct ufshpb_lu *hpb) +{ +struct victim_select_info *lru_info = &hpb->lru_info; +struct ufshpb_region *rgn, *victim_rgn =NULL ; + +list_for_each_entry (rgn, &lru_info->lh_lru_rgn, list_lru_rgn) { +WARN_ON (!rgn); +if (ufshpb_check_issue_state_srgns (hpb, rgn)) +continue ; + + victim_rgn = rgn; +break ; + } + +return victim_rgn; +} + +static inline void ufshpb_cleanup_lru_info (struct victim_select_info *lru_info, +struct ufshpb_region *rgn) +{ +list_del_init (&rgn->list_lru_rgn); + rgn->rgn_state =HPB_RGN_INACTIVE ; +atomic_dec (&lru_info->active_cnt); +} + +static inline void ufshpb_purge_active_subregion (struct ufshpb_lu *hpb, +struct ufshpb_subregion *srgn) +{ +if (srgn->srgn_state !=HPB_SRGN_UNUSED ) { +ufshpb_put_map_ctx (hpb, srgn->mctx); + srgn->srgn_state =HPB_SRGN_UNUSED ; + srgn->mctx =NULL ; + } +} + +static void __ufshpb_evict_region (struct ufshpb_lu *hpb, +struct ufshpb_region *rgn) +{ +struct victim_select_info *lru_info; +struct ufshpb_subregion *srgn; +int srgn_idx; + + lru_info = &hpb->lru_info; + +dev_dbg (&hpb->hpb_lu_dev,"evict region %d\n" , rgn->rgn_idx); + +ufshpb_cleanup_lru_info (lru_info, rgn); + +for (srgn_idx =0 ; srgn_idx < rgn->srgn_cnt; srgn_idx++) { + srgn = rgn->srgn_tbl + srgn_idx; + +ufshpb_purge_active_subregion (hpb, srgn); + } +} + +static int ufshpb_evict_region (struct ufshpb_lu *hpb,struct ufshpb_region *rgn) +{ +unsigned long flags; +int ret =0 ; + +spin_lock_irqsave (&hpb->hpb_state_lock, flags); +if (rgn->rgn_state ==HPB_RGN_PINNED ) { +dev_warn (&hpb->hpb_lu_dev, +"pinned region cannot drop-out. region %d\n" , + rgn->rgn_idx); + goto out; + } +if (!list_empty (&rgn->list_lru_rgn)) { +if (ufshpb_check_issue_state_srgns (hpb, rgn)) { + ret = -EBUSY ; + goto out; + } + +__ufshpb_evict_region (hpb, rgn); + } +out: +spin_unlock_irqrestore (&hpb->hpb_state_lock, flags); +return ret; +} + +static inline struct +ufshpb_rsp_field *ufshpb_get_hpb_rsp (struct ufshcd_lrb *lrbp) +{ +return (struct ufshpb_rsp_field *)&lrbp->ucd_rsp_ptr->sr.sense_data_len; +} + +static int ufshpb_issue_map_req (struct ufshpb_lu *hpb, +struct ufshpb_region *rgn, +struct ufshpb_subregion *srgn) +{ +struct ufshpb_req *map_req; +unsigned long flags; +int ret; +int err = -EAGAIN ; +bool alloc_required = false; +enum HPB_RGN_STATE state =HPB_SRGN_INVALID ; + +spin_lock_irqsave (&hpb->hpb_state_lock, flags); +/* +* Since the region state change occurs only in the map_work, +* the state of the region cannot HPB_RGN_INACTIVE at this point. +* The region state must be changed in the map_work +*/ +WARN_ON (rgn->rgn_state ==HPB_RGN_INACTIVE ); + +if (srgn->srgn_state ==HPB_SRGN_UNUSED ) + alloc_required = true; + +/* +* If the subregion is already ISSUED state, +* a specific event (e.g., GC or wear-leveling, etc.) occurs in +* the device and HPB response for map loading is received. +* In this case, after finishing the HPB_READ_BUFFER, +* the next HPB_READ_BUFFER is performed again to obtain the latest +* map data. +*/ +if (srgn->srgn_state ==HPB_SRGN_ISSUED ) + goto unlock_out; + + srgn->srgn_state =HPB_SRGN_ISSUED ; +spin_unlock_irqrestore (&hpb->hpb_state_lock, flags); + +if (alloc_required) { +WARN_ON (srgn->mctx); + srgn->mctx =ufshpb_get_map_ctx (hpb); +if (!srgn->mctx) { +dev_notice (&hpb->hpb_lu_dev, +"get map_ctx failed. region %d - %d\n" , + rgn->rgn_idx, srgn->srgn_idx); + state =HPB_SRGN_UNUSED ; + goto change_srgn_state; + } + } + +ufshpb_clear_dirty_bitmap (hpb, srgn); + map_req =ufshpb_get_map_req (hpb, srgn); +if (!map_req) + goto change_srgn_state; + + ret =ufshpb_lu_get (hpb); +if (unlikely (ret)) { +dev_notice (&hpb->hpb_lu_dev, +"%s: ufshpb_lu_get failed: %d" , __func__, ret); + goto free_map_req; + } + + ret =ufshpb_execute_map_req (hpb, map_req); +if (ret) { +dev_notice (&hpb->hpb_lu_dev, +"%s: issue map_req failed: %d, region %d - %d\n" , + __func__, ret, srgn->rgn_idx, srgn->srgn_idx); +ufshpb_lu_put (hpb); + goto free_map_req; + } +return 0 ; + +free_map_req: +ufshpb_put_map_req (hpb, map_req); +change_srgn_state: +spin_lock_irqsave (&hpb->hpb_state_lock, flags); + srgn->srgn_state = state; +unlock_out: +spin_unlock_irqrestore (&hpb->hpb_state_lock, flags); +return err; +} + +static int ufshpb_add_region (struct ufshpb_lu *hpb,struct ufshpb_region *rgn) +{ +struct ufshpb_region *victim_rgn; +struct victim_select_info *lru_info = &hpb->lru_info; +unsigned long flags; +int ret =0 ; + +spin_lock_irqsave (&hpb->hpb_state_lock, flags); +/* +* If region belongs to lru_list, just move the region +* to the front of lru list. because the state of the region +* is already active-state +*/ +if (!list_empty (&rgn->list_lru_rgn)) { +ufshpb_hit_lru_info (lru_info, rgn); + goto out; + } + +if (rgn->rgn_state ==HPB_RGN_INACTIVE ) { +if (atomic_read (&lru_info->active_cnt) + == lru_info->max_lru_active_cnt) { +/* +* If the maximum number of active regions +* is exceeded, evict the least recently used region. +* This case may occur when the device responds +* to the eviction information late. +* It is okay to evict the least recently used region, +* because the device could detect this region +* by not issuing HPB_READ +*/ + victim_rgn =ufshpb_victim_lru_info (hpb); +if (!victim_rgn) { +dev_warn (&hpb->hpb_lu_dev, +"cannot get victim region error\n" ); + ret = -ENOMEM ; + goto out; + } + +dev_dbg (&hpb->hpb_lu_dev, +"LRU full (%d), choost victim %d\n" , +atomic_read (&lru_info->active_cnt), + victim_rgn->rgn_idx); +__ufshpb_evict_region (hpb, victim_rgn); + } + +/* +* When a region is added to lru_info list_head, +* it is guaranteed that the subregion has been +* assigned all mctx. If failed, try to receive mctx again +* without being added to lru_info list_head +*/ +ufshpb_add_lru_info (lru_info, rgn); + } +out: +spin_unlock_irqrestore (&hpb->hpb_state_lock, flags); +return ret; +} + +static void ufshpb_rsp_req_region_update (struct ufshpb_lu *hpb, +struct ufshpb_rsp_field *rsp_field) +{ +int i, rgn_idx, srgn_idx; + +/* +* If the active region and the inactive region are the same, +* we will inactivate this region. +* The device could check this (region inactivated) and +* will response the proper active region information +*/ +spin_lock (&hpb->rsp_list_lock); +for (i =0 ; i < rsp_field->active_rgn_cnt; i++) { + rgn_idx = +be16_to_cpu (rsp_field->hpb_active_field[i].active_rgn); + srgn_idx = +be16_to_cpu (rsp_field->hpb_active_field[i].active_srgn); + +dev_dbg (&hpb->hpb_lu_dev,"activate(%d) region %d - %d\n" , + i, rgn_idx, srgn_idx); +ufshpb_update_active_info (hpb, rgn_idx, srgn_idx); +atomic_inc (&hpb->stats.rb_active_cnt); + } + +for (i =0 ; i < rsp_field->inactive_rgn_cnt; i++) { + rgn_idx =be16_to_cpu (rsp_field->hpb_inactive_field[i]); +dev_dbg (&hpb->hpb_lu_dev,"inactivate(%d) region %d\n" , + i, rgn_idx); +ufshpb_update_inactive_info (hpb, rgn_idx); +atomic_inc (&hpb->stats.rb_inactive_cnt); + } +spin_unlock (&hpb->rsp_list_lock); + +dev_dbg (&hpb->hpb_lu_dev,"Noti: #ACT %u #INACT %u\n" , + rsp_field->active_rgn_cnt, rsp_field->inactive_rgn_cnt); + +queue_work (ufshpb_drv.ufshpb_wq, &hpb->map_work); +} + +/* routine : isr (ufs) */ +static void ufshpb_rsp_upiu (struct ufs_hba *hba,struct ufshcd_lrb *lrbp) +{ +struct ufshpb_lu *hpb; +struct ufshpb_rsp_field *rsp_field; +int data_seg_len, ret; + + data_seg_len =be32_to_cpu (lrbp->ucd_rsp_ptr->header.dword_2) + &MASK_RSP_UPIU_DATA_SEG_LEN ; + +/* To flush remained rsp_list, we queue the map_work task */ +if (!data_seg_len) { +if (!ufshpb_is_general_lun (lrbp->lun)) +return ; + + hpb =ufshpb_get_hpb_data (lrbp->cmd); + ret =ufshpb_lu_get (hpb); +if (ret) +return ; + +if (!ufshpb_is_empty_rsp_lists (hpb)) +queue_work (ufshpb_drv.ufshpb_wq, &hpb->map_work); + + goto put_hpb; + } + +/* Check HPB_UPDATE_ALERT */ +if (!(lrbp->ucd_rsp_ptr->header.dword_2 & +UPIU_HEADER_DWORD (0 ,2 ,0 ,0 ))) +return ; + + rsp_field =ufshpb_get_hpb_rsp (lrbp); +if (ufshpb_may_field_valid (hba, lrbp, rsp_field)) +return ; + + hpb =ufshpb_get_hpb_data (lrbp->cmd); + ret =ufshpb_lu_get (hpb); +if (ret) +return ; + +atomic_inc (&hpb->stats.rb_noti_cnt); + +switch (rsp_field->hpb_type) { +case HPB_RSP_REQ_REGION_UPDATE : +WARN_ON (data_seg_len !=DEV_DATA_SEG_LEN ); +ufshpb_rsp_req_region_update (hpb, rsp_field); +break ; +case HPB_RSP_DEV_RESET : +dev_warn (&hpb->hpb_lu_dev, +"UFS device lost HPB information during PM.\n" ); +break ; +default : +dev_notice (&hpb->hpb_lu_dev,"hpb_type is not available: %d\n" , + rsp_field->hpb_type); +break ; + } +put_hpb: +ufshpb_lu_put (hpb); +} + +static void ufshpb_add_active_list (struct ufshpb_lu *hpb, +struct ufshpb_region *rgn, +struct ufshpb_subregion *srgn) +{ +if (!list_empty (&rgn->list_inact_rgn)) +return ; + +if (!list_empty (&srgn->list_act_srgn)) { +list_move (&srgn->list_act_srgn, &hpb->lh_act_srgn); +return ; + } + +list_add (&srgn->list_act_srgn, &hpb->lh_act_srgn); +} + +static void ufshpb_add_pending_evict_list (struct ufshpb_lu *hpb, +struct ufshpb_region *rgn, +struct list_head *pending_list) +{ +struct ufshpb_subregion *srgn; +int srgn_idx; + +if (!list_empty (&rgn->list_inact_rgn)) +return ; + +for (srgn_idx =0 ; srgn_idx < rgn->srgn_cnt; srgn_idx++) { + srgn = rgn->srgn_tbl + srgn_idx; + +if (!list_empty (&srgn->list_act_srgn)) +return ; + } + +list_add_tail (&rgn->list_inact_rgn, pending_list); +} + +static void ufshpb_run_active_subregion_list (struct ufshpb_lu *hpb) +{ +struct ufshpb_region *rgn; +struct ufshpb_subregion *srgn; +unsigned long flags; +int ret =0 ; + +spin_lock_irqsave (&hpb->rsp_list_lock, flags); +while ((srgn =list_first_entry_or_null (&hpb->lh_act_srgn, + struct ufshpb_subregion, + list_act_srgn))) { +list_del_init (&srgn->list_act_srgn); +spin_unlock_irqrestore (&hpb->rsp_list_lock, flags); + + rgn = hpb->rgn_tbl + srgn->rgn_idx; + ret =ufshpb_add_region (hpb, rgn); +if (ret) +break ; + + ret =ufshpb_issue_map_req (hpb, rgn, srgn); +if (ret) { +dev_notice (&hpb->hpb_lu_dev, +"issue map_req failed. ret %d, region %d - %d\n" , + ret, rgn->rgn_idx, srgn->srgn_idx); +break ; + } +spin_lock_irqsave (&hpb->rsp_list_lock, flags); + } + +if (ret) { +dev_notice (&hpb->hpb_lu_dev,"region %d - %d, will retry\n" , + rgn->rgn_idx, srgn->srgn_idx); +spin_lock_irqsave (&hpb->rsp_list_lock, flags); +ufshpb_add_active_list (hpb, rgn, srgn); + } +spin_unlock_irqrestore (&hpb->rsp_list_lock, flags); +} + +static void ufshpb_run_inactive_region_list (struct ufshpb_lu *hpb) +{ +struct ufshpb_region *rgn; +unsigned long flags; +int ret; +LIST_HEAD (pending_list); + +spin_lock_irqsave (&hpb->rsp_list_lock, flags); +while ((rgn =list_first_entry_or_null (&hpb->lh_inact_rgn, + struct ufshpb_region, + list_inact_rgn))) { +list_del_init (&rgn->list_inact_rgn); +spin_unlock_irqrestore (&hpb->rsp_list_lock, flags); + + ret =ufshpb_evict_region (hpb, rgn); +if (ret) { +spin_lock_irqsave (&hpb->rsp_list_lock, flags); +ufshpb_add_pending_evict_list (hpb, rgn, &pending_list); +spin_unlock_irqrestore (&hpb->rsp_list_lock, flags); + } + +spin_lock_irqsave (&hpb->rsp_list_lock, flags); + } + +list_splice (&pending_list, &hpb->lh_inact_rgn); +spin_unlock_irqrestore (&hpb->rsp_list_lock, flags); +} + +static void ufshpb_map_work_handler (struct work_struct *work) +{ +struct ufshpb_lu *hpb; +int ret; + + hpb = container_of(work, struct ufshpb_lu, map_work); + ret =ufshpb_lu_get (hpb); +if (ret) { +dev_info (&hpb->hpb_lu_dev,"%s: exit, state %d\n" , + __func__,ufshpb_get_state (hpb)); +return ; + } + +ufshpb_run_inactive_region_list (hpb); +ufshpb_run_active_subregion_list (hpb); + +ufshpb_lu_put (hpb); +} + +/* +* this function doesn't need to hold lock due to be called in init. +* (hpb_state_lock, rsp_list_lock, etc..) +*/ +static int ufshpb_init_pinned_active_region (struct ufs_hba *hba, +struct ufshpb_lu *hpb, +struct ufshpb_region *rgn) +{ +struct ufshpb_subregion *srgn; +int srgn_idx, i; +int err =0 ; + +for (srgn_idx =0 ; srgn_idx < rgn->srgn_cnt; srgn_idx++) { + srgn = rgn->srgn_tbl + srgn_idx; + + srgn->mctx =ufshpb_get_map_ctx (hpb); + srgn->srgn_state =HPB_SRGN_INVALID ; +if (!srgn->mctx) { +dev_err (hba->dev, +"alloc mctx for pinned region failed\n" ); + goto release; + } + +list_add_tail (&srgn->list_act_srgn, &hpb->lh_act_srgn); + } + + rgn->rgn_state =HPB_RGN_PINNED ; +return 0 ; + +release: +for (i =0 ; i < srgn_idx; i++) { + srgn = rgn->srgn_tbl + i; +ufshpb_put_map_ctx (hpb, srgn->mctx); + } +return err; +} +static void ufshpb_init_subregion_tbl (struct ufshpb_lu *hpb,struct ufshpb_region *rgn) {
@@ -67,6 +884,8 @@ static void ufshpb_init_subregion_tbl(struct ufshpb_lu *hpb,for (srgn_idx =0 ; srgn_idx < rgn->srgn_cnt; srgn_idx++) {struct ufshpb_subregion *srgn = rgn->srgn_tbl + srgn_idx; +INIT_LIST_HEAD (&srgn->list_act_srgn); + srgn->rgn_idx = rgn->rgn_idx; srgn->srgn_idx = srgn_idx; srgn->srgn_state =HPB_SRGN_UNUSED ;
@@ -99,6 +918,8 @@ static void ufshpb_init_lu_parameter(struct ufs_hba *hba, hpb->lu_pinned_end = hpb_lu_info->num_pinned ? (hpb_lu_info->pinned_start + hpb_lu_info->num_pinned -1 ) :PINNED_NOT_SET ; + hpb->lru_info.max_lru_active_cnt = + hpb_lu_info->max_active_rgns - hpb_lu_info->num_pinned; rgn_mem_size = (1ULL << hpb_dev_info->rgn_size) *HPB_RGN_SIZE_UNIT /HPB_ENTRY_BLOCK_SIZE *HPB_ENTRY_SIZE ;
@@ -152,6 +973,9 @@ static int ufshpb_alloc_region_tbl(struct ufs_hba *hba, struct ufshpb_lu *hpb) rgn = rgn_table + rgn_idx; rgn->rgn_idx = rgn_idx; +INIT_LIST_HEAD (&rgn->list_inact_rgn); +INIT_LIST_HEAD (&rgn->list_lru_rgn); +if (rgn_idx == hpb->rgns_per_lu -1 ) srgn_cnt = ((hpb->srgns_per_lu -1 ) % hpb->srgns_per_rgn) +1 ;
@@ -161,7 +985,13 @@ static int ufshpb_alloc_region_tbl(struct ufs_hba *hba, struct ufshpb_lu *hpb) goto release_srgn_table;ufshpb_init_subregion_tbl (hpb, rgn); - rgn->rgn_state =HPB_RGN_INACTIVE ; +if (ufshpb_is_pinned_region (hpb, rgn_idx)) { + ret =ufshpb_init_pinned_active_region (hba, hpb, rgn); +if (ret) + goto release_srgn_table; + }else { + rgn->rgn_state =HPB_RGN_INACTIVE ; + } }return 0 ;
@@ -185,7 +1015,10 @@ static void ufshpb_destroy_subregion_tbl(struct ufshpb_lu *hpb,struct ufshpb_subregion *srgn; srgn = rgn->srgn_tbl + srgn_idx; - srgn->srgn_state =HPB_SRGN_UNUSED ; +if (srgn->srgn_state !=HPB_SRGN_UNUSED ) { + srgn->srgn_state =HPB_SRGN_UNUSED ; +ufshpb_put_map_ctx (hpb, srgn->mctx); + } } }
@@ -215,8 +1048,36 @@ static int ufshpb_lu_hpb_init(struct ufs_hba *hba, struct ufshpb_lu *hpb,int ret;spin_lock_init (&hpb->hpb_state_lock); +spin_lock_init (&hpb->rsp_list_lock); + +INIT_LIST_HEAD (&hpb->lru_info.lh_lru_rgn); +INIT_LIST_HEAD (&hpb->lh_act_srgn); +INIT_LIST_HEAD (&hpb->lh_inact_rgn); +INIT_LIST_HEAD (&hpb->list_hpb_lu); + +INIT_WORK (&hpb->map_work, ufshpb_map_work_handler); + + hpb->map_req_cache =kmem_cache_create ("ufshpb_req_cache" , +sizeof (struct ufshpb_req ),0 ,0 ,NULL ); +if (!hpb->map_req_cache) { +dev_err (hba->dev,"ufshpb(%d) ufshpb_req_cache create fail" , + hpb->lun); +return -ENOMEM ; + } + + hpb->m_page_cache =kmem_cache_create ("ufshpb_m_page_cache" , +sizeof (struct page *) * hpb->pages_per_srgn, +0 ,0 ,NULL ); +if (!hpb->m_page_cache) { +dev_err (hba->dev,"ufshpb(%d) ufshpb_m_page_cache create fail" , + hpb->lun); + ret = -ENOMEM ; + goto release_req_cache; + } ret =ufshpb_alloc_region_tbl (hba, hpb); +if (ret) + goto release_m_page_cache; ret =ufshpb_create_sysfs (hba, hpb);if (ret)
@@ -226,6 +1087,10 @@ static int ufshpb_lu_hpb_init(struct ufs_hba *hba, struct ufshpb_lu *hpb, release_rgn_table:ufshpb_destroy_region_tbl (hpb); +release_m_page_cache: +kmem_cache_destroy (hpb->m_page_cache); +release_req_cache: +kmem_cache_destroy (hpb->map_req_cache);return ret; }
@@ -257,10 +1122,42 @@ static struct ufshpb_lu *ufshpb_alloc_hpb_lu(struct ufs_hba *hba, int lun,return NULL ; } +static void ufshpb_discard_rsp_lists (struct ufshpb_lu *hpb) +{ +struct ufshpb_region *rgn, *next_rgn; +struct ufshpb_subregion *srgn, *next_srgn; +unsigned long flags; + +/* +* If the device reset occurred, the remained HPB region information +* may be stale. Therefore, by dicarding the lists of HPB response +* that remained after reset, it prevents unnecessary work. +*/ +spin_lock_irqsave (&hpb->rsp_list_lock, flags); +list_for_each_entry_safe (rgn, next_rgn, &hpb->lh_inact_rgn, + list_inact_rgn) +list_del_init (&rgn->list_inact_rgn); + +list_for_each_entry_safe (srgn, next_srgn, &hpb->lh_act_srgn, + list_act_srgn) +list_del_init (&srgn->list_act_srgn); +spin_unlock_irqrestore (&hpb->rsp_list_lock, flags); +} + +static inline void ufshpb_cancel_jobs (struct ufshpb_lu *hpb) +{ +cancel_work_sync (&hpb->map_work); +} +static void ufshpb_lu_release (struct ufshpb_lu *hpb) { +ufshpb_cancel_jobs (hpb); +ufshpb_destroy_region_tbl (hpb); +kmem_cache_destroy (hpb->map_req_cache); +kmem_cache_destroy (hpb->m_page_cache); +list_del_init (&hpb->list_hpb_lu); }
@@ -347,6 +1244,9 @@ static void ufshpb_reset_host(struct ufs_hba *hba)dev_info (&hpb->hpb_lu_dev,"ufshpb run reset_host" );ufshpb_set_state (hpb,HPB_RESET ); +ufshpb_cancel_jobs (hpb); +ufshpb_discard_rsp_lists (hpb); +ufshpb_lu_put (hpb); } }
@@ -361,6 +1261,7 @@ static void ufshpb_suspend(struct ufs_hba *hba)dev_info (&hpb->hpb_lu_dev,"ufshpb goto suspend" );ufshpb_set_state (hpb,HPB_SUSPEND ); +ufshpb_cancel_jobs (hpb);ufshpb_lu_put (hpb); }
@@ -376,6 +1277,9 @@ static void ufshpb_resume(struct ufs_hba *hba)dev_info (&hpb->hpb_lu_dev,"ufshpb resume" );ufshpb_set_state (hpb,HPB_PRESENT ); +if (!ufshpb_is_empty_rsp_lists (hpb)) +queue_work (ufshpb_drv.ufshpb_wq, &hpb->map_work); +ufshpb_lu_put (hpb); } }
@@ -617,6 +1521,8 @@ static void ufshpb_scan_hpb_lu(struct ufs_hba *hba,struct scsi_device *sdev;struct ufshpb_lu *hpb;int find_hpb_lu =0 ; +int tot_active_srgn_pages =0 ; +int pool_size;int ret;INIT_LIST_HEAD (&ufshpb_drv.lh_hpb_lu);
@@ -637,6 +1543,9 @@ static void ufshpb_scan_hpb_lu(struct ufs_hba *hba,if (!hpb)continue ; + tot_active_srgn_pages += hpb_lu_info.max_active_rgns * + hpb->srgns_per_rgn * hpb->pages_per_srgn; + hpb->sdev_ufs_lu = sdev; sdev->hostdata = hpb;
@@ -648,11 +1557,31 @@ static void ufshpb_scan_hpb_lu(struct ufs_hba *hba,return ;ufshpb_check_hpb_reset_query (hba); + + pool_size =DIV_ROUND_UP (ufshpb_host_map_kbytes *1024 ,PAGE_SIZE ); +if (pool_size > tot_active_srgn_pages) { +dev_info (hba->dev, +"reset pool_size to %lu KB.\n" , + tot_active_srgn_pages *PAGE_SIZE /1024 ); +mempool_resize (ufshpb_drv.ufshpb_mctx_pool, + tot_active_srgn_pages); +mempool_resize (ufshpb_drv.ufshpb_page_pool, + tot_active_srgn_pages); + } +dev_set_drvdata (&hba->ufsf.hpb_dev, &ufshpb_drv);list_for_each_entry (hpb, &ufshpb_drv.lh_hpb_lu, list_hpb_lu) {dev_info (&hpb->hpb_lu_dev,"set state to present\n" );ufshpb_set_state (hpb,HPB_PRESENT ); + +if ((hpb->lu_pinned_end - hpb->lu_pinned_start) >0 ) { +dev_info (&hpb->hpb_lu_dev, +"loading pinned regions %d - %d\n" , + hpb->lu_pinned_start, hpb->lu_pinned_end); +queue_work (ufshpb_drv.ufshpb_wq, + &hpb->map_work); + } } }
@@ -719,6 +1648,8 @@ static int ufshpb_remove(struct device *dev) sdev = hpb->sdev_ufs_lu; sdev->hostdata =NULL ; +ufshpb_cancel_jobs (hpb); +device_del (&hpb->hpb_lu_dev); dev_info(&hpb->hpb_lu_dev, "hpb_lu_dev refcnt %d\n",
@@ -744,10 +1675,10 @@ static struct ufshpb_driver ufshpb_drv = { .reset_host = ufshpb_reset_host, .suspend = ufshpb_suspend, .resume = ufshpb_resume, + .rsp_upiu = ufshpb_rsp_upiu, }, }; -unsigned int ufshpb_host_map_kbytes =1 *1024 ;module_param (ufshpb_host_map_kbytes, uint,0644 );MODULE_PARM_DESC (ufshpb_host_map_kbytes,"ufshpb host mapping memory kilo-bytes for ufshpb memory-pool" );
@@ -755,17 +1686,70 @@ MODULE_PARM_DESC(ufshpb_host_map_kbytes,static int __init ufshpb_init (void ) {int ret; +unsigned int pool_size; + + ufshpb_drv.ufshpb_mctx_cache =kmem_cache_create ("ufshpb_mctx_cache" , +sizeof (struct ufshpb_map_ctx ), +0 ,0 ,NULL ); +if (!ufshpb_drv.ufshpb_mctx_cache) { +pr_err ("ufshpb: cannot init mctx cache\n" ); +return -ENOMEM ; + } + + pool_size =DIV_ROUND_UP (ufshpb_host_map_kbytes *1024 ,PAGE_SIZE ); +pr_info ("%s:%d ufshpb_host_map_kbytes %u pool_size %u\n" , + __func__, __LINE__, ufshpb_host_map_kbytes, pool_size); + + ufshpb_drv.ufshpb_mctx_pool =mempool_create_slab_pool ( + pool_size, ufshpb_drv.ufshpb_mctx_cache); +if (!ufshpb_drv.ufshpb_mctx_pool) { +pr_err ("ufshpb: cannot init mctx pool\n" ); + ret = -ENOMEM ; + goto release_mctx_cache; + } + + ufshpb_drv.ufshpb_page_pool =mempool_create_page_pool (pool_size,0 ); +if (!ufshpb_drv.ufshpb_page_pool) { +pr_err ("ufshpb: cannot init page pool\n" ); + ret = -ENOMEM ; + goto release_mctx_pool; + } + + ufshpb_drv.ufshpb_wq =alloc_workqueue ("ufshpb-wq" , +WQ_UNBOUND |WQ_MEM_RECLAIM ,0 ); +if (!ufshpb_drv.ufshpb_wq) { +pr_err ("ufshpb: alloc workqueue failed\n" ); + ret = -ENOMEM ; + goto release_page_pool; + } ret =driver_register (&ufshpb_drv.drv); -if (ret) +if (ret) {pr_err ("ufshpb: driver register failed\n" ); + goto release_wq; + } +return 0 ; +release_wq: +destroy_workqueue (ufshpb_drv.ufshpb_wq); +release_page_pool: +mempool_destroy (ufshpb_drv.ufshpb_page_pool); +release_mctx_pool: +mempool_destroy (ufshpb_drv.ufshpb_mctx_pool); +release_mctx_cache: +kmem_cache_destroy (ufshpb_drv.ufshpb_mctx_cache);return ret; } static void __exit ufshpb_exit(void) {driver_unregister (&ufshpb_drv.drv); + +mempool_destroy (ufshpb_drv.ufshpb_page_pool); +mempool_destroy (ufshpb_drv.ufshpb_mctx_pool); +kmem_cache_destroy (ufshpb_drv.ufshpb_mctx_cache); + +destroy_workqueue (ufshpb_drv.ufshpb_wq); }MODULE_AUTHOR ("Yongmyong Lee <[email protected]>" );
drivers/scsi/ufs/ufshpb.h +72 -0
100 lines not on this page · raw message
Signature · 1 line
2.17.1
RFC PATCH v3 5/5 scsi: ufs: Prepare HPB read for cached sub-region
This patch changes the read I/O to the HPB read I/O.
If the logical address of the read I/O belongs to active sub-region, the HPB driver modifies the read I/O command to HPB read. It modifies the upiu command of UFS instead of modifying the existing SCSI command.
In the HPB version 1.0, the maximum read I/O size that can be converted to HPB read is 4KB.
The dirty map of the active sub-region prevents an incorrect HPB read that has stale physical page number which is updated by previous write I/O.
--- drivers/scsi/ufs/ufshpb.c | 235 ++++++++++++++++++++++++++++++++++++++ 1 file changed, 235 insertions(+)
drivers/scsi/ufs/ufshpb.c +235 -0
@@ -26,6 +26,22 @@ static inline int ufshpb_is_valid_srgn(struct ufshpb_region *rgn, srgn->srgn_state ==HPB_SRGN_VALID ; } +static inline bool ufshpb_is_read_cmd (struct scsi_cmnd *cmd) +{ +return req_op (cmd->request) ==REQ_OP_READ ; +} + +static inline bool ufshpb_is_write_discard_cmd (struct scsi_cmnd *cmd) +{ +return op_is_write (req_op (cmd->request)) || +op_is_discard (req_op (cmd->request)); +} + +static inline bool ufshpb_is_support_chunk (int transfer_len) +{ +return transfer_len <=HPB_MULTI_CHUNK_HIGH ; +} +static inline bool ufshpb_is_general_lun (int lun) {return lun <UFS_UPIU_MAX_UNIT_NUM_ID ;
@@ -117,6 +133,224 @@ static inline void ufshpb_lu_put(struct ufshpb_lu *hpb)put_device (&hpb->hpb_lu_dev); } +static inline u32 ufshpb_get_lpn (struct scsi_cmnd *cmnd) +{ +return blk_rq_pos (cmnd->request) >> + (ilog2 (cmnd->device->sector_size) -9 ); +} + +static inline unsigned int ufshpb_get_len (struct scsi_cmnd *cmnd) +{ +return blk_rq_sectors (cmnd->request) >> + (ilog2 (cmnd->device->sector_size) -9 ); +} + +static void ufshpb_set_ppn_dirty (struct ufshpb_lu *hpb,int rgn_idx, +int srgn_idx,int srgn_offset,int cnt) +{ +struct ufshpb_region *rgn; +struct ufshpb_subregion *srgn; +int set_bit_len; +int bitmap_len = hpb->entries_per_srgn; + +next_srgn: + rgn = hpb->rgn_tbl + rgn_idx; + srgn = rgn->srgn_tbl + srgn_idx; + +if ((srgn_offset + cnt) > bitmap_len) + set_bit_len = bitmap_len - srgn_offset; +else + set_bit_len = cnt; + +if (rgn->rgn_state !=HPB_RGN_INACTIVE && + srgn->srgn_state ==HPB_SRGN_VALID ) +bitmap_set (srgn->mctx->ppn_dirty, srgn_offset, set_bit_len); + + srgn_offset =0 ; +if (++srgn_idx == hpb->srgns_per_rgn) { + srgn_idx =0 ; + rgn_idx++; + } + + cnt -= set_bit_len; +if (cnt >0 ) + goto next_srgn; + +WARN_ON (cnt <0 ); +} + +static bool ufshpb_test_ppn_dirty (struct ufshpb_lu *hpb,int rgn_idx, +int srgn_idx,int srgn_offset,int cnt) +{ +struct ufshpb_region *rgn; +struct ufshpb_subregion *srgn; +int bitmap_len = hpb->entries_per_srgn; +int bit_len; + +next_srgn: + rgn = hpb->rgn_tbl + rgn_idx; + srgn = rgn->srgn_tbl + srgn_idx; + +if (!ufshpb_is_valid_srgn (rgn, srgn)) +return true; + +/* +* If the region state is active, mctx must be allocated. +* In this case, check whether the region is evicted or +* mctx allcation fail. +*/ +WARN_ON (!srgn->mctx); + +if ((srgn_offset + cnt) > bitmap_len) + bit_len = bitmap_len - srgn_offset; +else + bit_len = cnt; + +if (find_next_bit (srgn->mctx->ppn_dirty, + bit_len, srgn_offset) >= srgn_offset) +return true; + + srgn_offset =0 ; +if (++srgn_idx == hpb->srgns_per_rgn) { + srgn_idx =0 ; + rgn_idx++; + } + + cnt -= bit_len; +if (cnt >0 ) + goto next_srgn; + +return false; +} + +static u64 ufshpb_get_ppn (struct ufshpb_lu *hpb, +struct ufshpb_map_ctx *mctx,int pos,int *error) +{ +u64 *ppn_table; +struct page *page; +int index, offset; + + index = pos / (PAGE_SIZE /HPB_ENTRY_SIZE ); + offset = pos % (PAGE_SIZE /HPB_ENTRY_SIZE ); + + page = mctx->m_page[index]; +if (unlikely (!page)) { + *error = -ENOMEM ; +dev_err (&hpb->hpb_lu_dev, +"error. cannot find page in mctx\n" ); +return 0 ; + } + + ppn_table =page_address (page); +if (unlikely (!ppn_table)) { + *error = -ENOMEM ; +dev_err (&hpb->hpb_lu_dev,"error. cannot get ppn_table\n" ); +return 0 ; + } + +return ppn_table[offset]; +} + +static inline void +ufshpb_get_pos_from_lpn (struct ufshpb_lu *hpb,unsigned long lpn,int *rgn_idx, +int *srgn_idx,int *offset) +{ +int rgn_offset; + + *rgn_idx = lpn >> hpb->entries_per_rgn_shift; + rgn_offset = lpn & hpb->entries_per_rgn_mask; + *srgn_idx = rgn_offset >> hpb->entries_per_srgn_shift; + *offset = rgn_offset & hpb->entries_per_srgn_mask; +} + +static void +ufshpb_set_hpb_read_to_upiu (struct ufshpb_lu *hpb,struct ufshcd_lrb *lrbp, +u32 lpn,u64 ppn,unsigned int transfer_len) +{ +unsigned char *cdb = lrbp->ucd_req_ptr->sc.cdb; + + cdb[0 ] =UFSHPB_READ ; + +put_unaligned_be64 (ppn, &cdb[6 ]); + cdb[14 ] = transfer_len; +} + +/* routine : READ10 -> HPB_READ */ +static void ufshpb_prep_fn (struct ufs_hba *hba,struct ufshcd_lrb *lrbp) +{ +struct ufshpb_lu *hpb; +struct ufshpb_region *rgn; +struct ufshpb_subregion *srgn; +struct scsi_cmnd *cmd = lrbp->cmd; +u32 lpn; +u64 ppn; +unsigned long flags; +int transfer_len, rgn_idx, srgn_idx, srgn_offset; +int err =0 ; + + hpb =ufshpb_get_hpb_data (cmd); + err =ufshpb_lu_get (hpb); +if (unlikely (err)) +return ; + +WARN_ON (hpb->lun != cmd->device->lun); +if (!ufshpb_is_write_discard_cmd (cmd) && + !ufshpb_is_read_cmd (cmd)) + goto put_hpb; + + transfer_len =ufshpb_get_len (cmd); +if (unlikely (!transfer_len)) + goto put_hpb; + + lpn =ufshpb_get_lpn (cmd); +ufshpb_get_pos_from_lpn (hpb, lpn, &rgn_idx, &srgn_idx, &srgn_offset); + rgn = hpb->rgn_tbl + rgn_idx; + srgn = rgn->srgn_tbl + srgn_idx; + +/* If command type is WRITE or DISCARD, set bitmap as drity */ +if (ufshpb_is_write_discard_cmd (cmd)) { +spin_lock_irqsave (&hpb->hpb_state_lock, flags); +ufshpb_set_ppn_dirty (hpb, rgn_idx, srgn_idx, srgn_offset, + transfer_len); +spin_unlock_irqrestore (&hpb->hpb_state_lock, flags); + goto put_hpb; + } + +WARN_ON (!ufshpb_is_read_cmd (cmd)); + +if (!ufshpb_is_support_chunk (transfer_len)) + goto put_hpb; + +spin_lock_irqsave (&hpb->hpb_state_lock, flags); +if (ufshpb_test_ppn_dirty (hpb, rgn_idx, srgn_idx, srgn_offset, + transfer_len)) { +atomic_inc (&hpb->stats.miss_cnt); +spin_unlock_irqrestore (&hpb->hpb_state_lock, flags); + goto put_hpb; + } + + ppn =ufshpb_get_ppn (hpb, srgn->mctx, srgn_offset, &err); +spin_unlock_irqrestore (&hpb->hpb_state_lock, flags); +if (unlikely (err)) { +/* +* In this case, the region state is active, +* but the ppn table is not allocated. +* Make sure that ppn table must be allocated on +* active state. +*/ +WARN_ON (true); +dev_err (&hpb->hpb_lu_dev, +"ufshpb_get_ppn failed. err %d\n" , err); + goto put_hpb; + } + +ufshpb_set_hpb_read_to_upiu (hpb, lrbp, lpn, ppn, transfer_len); + +atomic_inc (&hpb->stats.hit_cnt); +put_hpb: +ufshpb_lu_put (hpb); +} +static struct ufshpb_req *ufshpb_get_map_req (struct ufshpb_lu *hpb,struct ufshpb_subregion *srgn) {
@@ -1671,6 +1905,7 @@ static struct ufshpb_driver ufshpb_drv = { .probe_type = PROBE_PREFER_ASYNCHRONOUS, }, .ufshpb_ops = { + .prep_fn = ufshpb_prep_fn, .reset = ufshpb_reset, .reset_host = ufshpb_reset_host, .suspend = ufshpb_suspend,
Signature · 1 line
2.17.1
RE: [RFC PATCH v3 4/5] scsi: ufs: L2P map management for HPB read
In reply to Daejun ParkQuoted 2 lines
static struct
ufshpb_driverufshpb_drv;
+unsigned intufshpb_host_map_kbytes= 1 * 1024;
I think you've already declared this as a module parameter in 3/5.
No need to fix this now, unless there will be some more comments, And you'll issue a v4.
Thanks,
Avri
RE: [RFC PATCH v3 4/5] scsi: ufs: L2P map management for HPB read
Quoted 6 lines
static struct
ufshpb_driverufshpb_drv;Quoted 1 line
+unsigned int
ufshpb_host_map_kbytes= 1 * 1024;I think you've already declared this as a module parameter in 3/5.
No need to fix this now, unless there will be some more comments, And you'll issue a v4.
OK, thanks!
Hi Daejun
Seems you intentionally ignored to give you comments on my suggestion. let me provide the reason.
Before submitting your next version patch, please check your L2P mapping HPB reqeust submission logical algorithem. I have did performance comparison testing on 4KB, there are about 13% performance drop. Also the hit count is lower. I don't know if this is related to your current work queue scheduling, since you didn't add the timer for each HPB request.
Thanks,
Bean
Quoted 66 lines · Daejun Park
Changelog:
v2 -> v3
1. Add checking input module parameter value.
2. Change base commit from 5.8/scsi-queue to 5.9/scsi-queue.
3. Cleanup for unused variables and label.v1 -> v2
1. Change the full boilerplate text to SPDX style.
2. Adopt dynamic allocation for sub-region data structure.
3. Cleanup.NAND flash memory-based storage devices use Flash Translation Layer (FTL)
to translate logical addresses of I/O requests to corresponding flash memory addresses. Mobile storage devices typically have RAM with constrained size, thus lack in memory to keep the whole mapping table.
Therefore, mapping tables are partially retrieved from NAND flash on demand, causing random-read performance degradation.To improve random read performance, JESD220-3 (HPB v1.0) proposes HPB (Host Performance Booster) which uses host system memory as a cache for the
FTL mapping table. By using HPB, FTL data can be read from host memory
faster than from NAND flash memory.The current version only supports the DCM (device control mode). This patch consists of 4 parts to support HPB feature.
1) UFS-feature layer
2) HPB probe and initialization process
3) READ -> HPB READ using cached map information
4) L2P (logical to physical) map managementThe UFS-feature is an additional layer to avoid the structure in which the
UFS-core driver and the UFS-feature are entangled with each other in a
single module.
By adding the layer, UFS-features composed of various combinations can be
supported. Also, even if a new feature is added, modification of the UFS-core driver can be minimized.In the HPB probe and init process, the device information of the UFS is
queried. After checking supported features, the data structure for the HPB
is initialized according to the device information.A read I/O in the active sub-region where the map is cached is changed to
HPB READ by the HPB module.The HPB module manages the L2P map using information received from the
device. For active sub-region, the HPB module caches throughufshpb_map
request. For the in-active region, the HPB module discards the L2P map.
When a write I/O occurs in an active sub-region area, associated dirty
bitmap checked as dirty for preventing stale read.
Quoted 8 lines
HPB is shown to have a performance improvement of 58 - 67% for random read
workload. [1]This series patches are based on the 5.9/scsi-queue branch.
[1]:
https://www.usenix.org/conference/hotstorage17/program/presentation/jeong
Quoted 23 lines
Daejun park (5):
scsi: ufs: Add UFS feature related parameter
scsi: ufs: Add UFS feature layer
scsi: ufs: Introduce HPB module
scsi: ufs: L2P map management for HPB read
scsi: ufs: Prepare HPB read for cached sub-regiondrivers/scsi/ufs/Kconfig | 9 + drivers/scsi/ufs/Makefile | 3 +- drivers/scsi/ufs/ufs.h | 12 + drivers/scsi/ufs/ufsfeature.c | 148 +++ drivers/scsi/ufs/ufsfeature.h | 69 ++ drivers/scsi/ufs/ufshcd.c | 23 +- drivers/scsi/ufs/ufshcd.h | 3 + drivers/scsi/ufs/ufshpb.c | 1996 ++++++++++++++++++++++++++++++++++++ drivers/scsi/ufs/ufshpb.h | 234 +++++ 9 files changed, 2494 insertions(+), 3 deletions(-)created mode 100644
drivers/scsi/ufs/ufsfeature.ccreated mode 100644drivers/scsi/ufs/ufsfeature.hcreated mode 100644drivers/scsi/ufs/ufshpb.ccreated mode 100644drivers/scsi/ufs/ufshpb.h
Hi Bean,
Quoted 6 lines
Hi Daejun
Seems you intentionally ignored to give you comments on my suggestion. let me provide the reason.
Quoted 6 lines
Before submitting your next version patch, please check your L2P mapping HPB reqeust submission logical algorithem. I have did performance comparison testing on 4KB, there are about 13% performance drop. Also the hit count is lower. I don't know if this is related to your current work queue scheduling, since you didn't add the timer for each HPB request.
In device control mode, the various decisions,
and specifically those that are causing repetitive evictions, are made by the device.
Is this the issue that you are referring to?
As for the driver, do you see any issue that is causing unnecessary latency?
Thanks,
Avri
Hi Avri
Quoted 18 lines · Avri Altman
Hi Bean,
Quoted 7 lines
Hi Daejun
Seems you intentionally ignored to give you comments on my suggestion.
let me provide the reason.Quoted 9 lines
Before submitting your next version patch, please check your L2P mapping HPB reqeust submission logical algorithem. I have did performance comparison testing on 4KB, there are about 13% performance
drop. Also the hit count is lower. I don't know if this is related to
your current work queue scheduling, since you didn't add the timer for
each HPB request.
Quoted 5 lines
In device control mode, the various decisions,
and specifically those that are causing repetitive evictions, are made by the device.
Is this the issue that you are referring to?
For this device mode, if HPB mapping table of the active region becomes dirty in the UFS device side, there is repetitive inactive rsp, but it is not the reason for the condition I mentioned here.
Quoted 3 lines
As for the driver, do you see any issue that is causing unnecessary latency?
In Daejun's patch, it now uses work_queue, and as long there is new RSP of thesubregion to be activated, the driver will queue "work" to this work queue, actually, this is deferred work. we don't know when it will be
scheduled/finished. we need to optimize it.
Thanks,
Bean
Quoted 34 lines
Hi Avri
Quoted 18 lines · Avri Altman
Hi Bean,
Quoted 7 lines
Hi Daejun
Seems you intentionally ignored to give you comments on my suggestion.
let me provide the reason.Quoted 9 lines
Before submitting your next version patch, please check your L2P mapping HPB reqeust submission logical algorithem. I have did performance comparison testing on 4KB, there are about 13% performance
drop. Also the hit count is lower. I don't know if this is related to
your current work queue scheduling, since you didn't add the timer for
each HPB request.Quoted 5 lines
In device control mode, the various decisions,
and specifically those that are causing repetitive evictions, are made by the device.
Is this the issue that you are referring to?For this device mode, if HPB mapping table of the active region becomes dirty in the UFS device side, there is repetitive inactive rsp, but it is not the reason for the condition I mentioned here.
Quoted 8 lines
Quoted 3 lines
As for the driver, do you see any issue that is causing unnecessary latency?
In Daejun's patch, it now uses
work_queue, and as long there is new RSP of thesubregion to be activated, the driver will queue "work" to this work queue, actually, this is deferred work. we don't know when it will be scheduled/finished. we need to optimize it.
But those "to-do" lists are checked on every completion interrupt and on every resume. Do you see any scenario in which the "to-be-activated" or "to-be-inactivate" work is getting starved?
Quoted 50 lines · Avri Altman
Quoted 42 lines
Hi Avri
Quoted 21 lines · Avri Altman
Hi Bean,
Quoted 7 lines
Hi Daejun
Seems you intentionally ignored to give you comments on my suggestion.
let me provide the reason.Quoted 12 lines
Before submitting your next version patch, please check your L2P
mapping HPB reqeust submission logical algorithem. I have did performance comparison testing on 4KB, there are about 13% performance
drop. Also the hit count is lower. I don't know if this is related
to
your current work queue scheduling, since you didn't add the timer
for
each HPB request.Quoted 5 lines
In device control mode, the various decisions,
and specifically those that are causing repetitive evictions, are made by the device.
Is this the issue that you are referring to?For this device mode, if HPB mapping table of the active region becomes
dirty in the UFS device side, there is repetitive inactive rsp, but it
is not the reason for the condition I mentioned here.Quoted 4 lines
As for the driver, do you see any issue that is causing unnecessary
latency?Quoted 7 lines
In Daejun's patch, it now uses
work_queue, and as long there is new RSP of
thesubregion to be activated, the driver will queue "work" to this work
queue, actually, this is deferred work. we don't know when it will be
scheduled/finished. we need to optimize it.
Quoted 5 lines
But those "to-do" lists are checked on every completion interrupt and on every resume.
Do you see any scenario in which the "to-be-activated" or "to-be- inactivate" work is getting starved?
let me run more testing cases, will back to you if there is new updates.
Thanks,
Bean
Quoted 2 lines
Seems you intentionally ignored to give you comments on my suggestion. let me provide the reason.
Sorry! I replied to your comment (https://lkml.org/lkml/2020/6/15/1492),
but you didn't reply on that. I thought you agreed because you didn't send any more comments.
Quoted 2 lines
Before submitting your next version patch, please check your L2P mapping HPB reqeust submission logical algorithem. I have did
We are also reviewing the code that you submitted before.
It seems to be a performance improvement as it sends a map request directly.
Quoted 2 lines
performance comparison testing on 4KB, there are about 13% performance drop. Also the hit count is lower. I don't know if this is related to
It is interesting that there is actually a performance improvement. Could you share the test environment, please? However, I think stability is important to HPB driver. We have tested our method with the real products and the HPB 1.0 driver is based on that.
After this patch, your approach can be done as an incremental patch? I would like to test the patch that you submitted and verify it.
Quoted 2 lines
your current work queue scheduling, since you didn't add the timer for each HPB request.
There was Bart's comment that it was not good add an arbitrary timeout value to the request. (please refer to: https://lkml.org/lkml/2020/6/11/1043)
When no timer is added to the request, the SD timout will be set as default timeout at the block layer.
Thanks,
Daejun
Hi Daejun
Quoted 23 lines · Daejun Park
Quoted 3 lines
Seems you intentionally ignored to give you comments on my suggestion.
let me provide the reason.Sorry! I replied to your comment (
https://lkml.org/lkml/2020/6/15/1492),
but you didn't reply on that. I thought you agreed because you didn't send
any more comments.Quoted 2 lines
Before submitting your next version patch, please check your L2P mapping HPB reqeust submission logical algorithem. I have did
We are also reviewing the code that you submitted before. It seems to be a performance improvement as it sends a map request directly.
Quoted 4 lines
performance comparison testing on 4KB, there are about 13% performance
drop. Also the hit count is lower. I don't know if this is related to
Quoted 6 lines
It is interesting that there is actually a performance improvement. Could you share the test environment, please? However, I think stability is
important to HPB driver. We have tested our method with the real products and
the HPB 1.0 driver is based on that.
I just run fio benchmark tool with --rw=randread, --bs=4kb, -- size=8G/10G/64G/100G. and see what performance diff with the direct submission approach.
Quoted 8 lines
After this patch, your approach can be done as an incremental patch? I would
like to test the patch that you submitted and verify it.Quoted 3 lines
your current work queue scheduling, since you didn't add the timer for
each HPB request.
Taking into consideration of the HPB 2.0, can we submit the HPB write request to the SCSI layer? if not, it will be a direct submission way. why not directly use direct way? or maybe you have a more advisable approach to work around this. would you please share with us. appreciate.
Quoted 8 lines
There was Bart's comment that it was not good add an arbitrary timeout value
to the request. (please refer to:
https://lkml.org/lkml/2020/6/11/1043)
When no timer is added to the request, the SD timout will be set as default
timeout at the block layer.
I saw that, so I should add a timer in order to optimise HPB reqeust scheduling/completition. this is ok so far.
Quoted 2 lines
Thanks,
Daejun
Thanks,
Bean
Hi Bean,
Quoted 31 lines · Daejun Park
Quoted 23 lines
Quoted 3 lines
Seems you intentionally ignored to give you comments on my suggestion.
let me provide the reason.Sorry! I replied to your comment (
https://protect2.fireeye.com/url?k=be575021-e3854728-be56db6e-0cc47a31cdf8-6c7d0e1e42762b92&q=1&u=https%3A%2F%2Flkml.org%2Flkml%2F2020%2F6%2F15%2F1492),
but you didn't reply on that. I thought you agreed because you didn't send
any more comments.Quoted 2 lines
Before submitting your next version patch, please check your L2P mapping HPB reqeust submission logical algorithem. I have did
We are also reviewing the code that you submitted before. It seems to be a performance improvement as it sends a map request directly.
Quoted 4 lines
performance comparison testing on 4KB, there are about 13% performance
drop. Also the hit count is lower. I don't know if this is related toQuoted 6 lines
It is interesting that there is actually a performance improvement. Could you share the test environment, please? However, I think stability is
important to HPB driver. We have tested our method with the real products and
the HPB 1.0 driver is based on that.
Quoted 3 lines
I just run fio benchmark tool with --rw=randread, --bs=4kb, -- size=8G/10G/64G/100G. and see what performance diff with the direct submission approach.
Thanks!
Quoted 9 lines
Quoted 8 lines
After this patch, your approach can be done as an incremental patch? I would
like to test the patch that you submitted and verify it.Quoted 3 lines
your current work queue scheduling, since you didn't add the timer for
each HPB request.
Quoted 5 lines
Taking into consideration of the HPB 2.0, can we submit the HPB write request to the SCSI layer? if not, it will be a direct submission way. why not directly use direct way? or maybe you have a more advisable approach to work around this. would you please share with us. appreciate.
I am considering a direct submission way for the next version.
We will implement the write buffer command of HPB 2.0, after patching HPB 1.0.
As for the direct submission of HPB releated command including HPB write buffer, I think we'd better discuss the right approach in depth before moving on to the next step.
Thanks,
Daejun
Hi,
Quoted 57 lines
Hi Bean,
Quoted 7 lines · Daejun Park
Quoted 6 lines
Quoted 3 lines
Seems you intentionally ignored to give you comments on my suggestion.
let me provide the reason.Sorry! I replied to your comment (
https://protect2.fireeye.com/url?k=be575021-e3854728-be56db6e-0cc47a31cdf8- 6c7d0e1e42762b92&q=1&u=https%3A%2F%2Flkml.org%2Flkml%2F2020%2F6% 2F15%2F1492),
Quoted 24 lines
Quoted 17 lines
but you didn't reply on that. I thought you agreed because you didn't send
any more comments.Quoted 2 lines
Before submitting your next version patch, please check your L2P mapping HPB reqeust submission logical algorithem. I have did
We are also reviewing the code that you submitted before. It seems to be a performance improvement as it sends a map request directly.
Quoted 4 lines
performance comparison testing on 4KB, there are about 13% performance
drop. Also the hit count is lower. I don't know if this is related toQuoted 6 lines
It is interesting that there is actually a performance improvement. Could you share the test environment, please? However, I think stability is
important to HPB driver. We have tested our method with the real products and
the HPB 1.0 driver is based on that.Quoted 3 lines
I just run fio benchmark tool with --rw=randread, --bs=4kb, -- size=8G/10G/64G/100G. and see what performance diff with the direct submission approach.
Thanks!
Quoted 9 lines
Quoted 8 lines
After this patch, your approach can be done as an incremental patch? I would
like to test the patch that you submitted and verify it.Quoted 3 lines
your current work queue scheduling, since you didn't add the timer for
each HPB request.Quoted 5 lines
Taking into consideration of the HPB 2.0, can we submit the HPB write request to the SCSI layer? if not, it will be a direct submission way. why not directly use direct way? or maybe you have a more advisable approach to work around this. would you please share with us. appreciate.
Quoted 7 lines
I am considering a direct submission way for the next version. We will implement the write buffer command of HPB 2.0, after patching HPB 1.0.
As for the direct submission of HPB releated command including HPB write buffer, I think we'd better discuss the right approach in depth before moving on to the next step.
I vote to stay with the current implementation because:
1) Bean is probably right about 2.0, but it's out of scope for now -
there is a long way to go before we'll need to worry about it
2) For now, we should focus on the functional flows.
Performance issues, should such issues indeed exists, can be dealt with later. And,
3) The current code base is running in production for more than 3 years now. I am not so eager to dump a robust, well debugged code unless it absolutely necessary.
Thanks,
Avri
Quoted 93 lines · Avri Altman
Hi,
Quoted 74 lines
Hi Bean,
Quoted 7 lines · Daejun Park
Quoted 6 lines
Quoted 3 lines
Seems you intentionally ignored to give you comments on my suggestion.
let me provide the reason.Sorry! I replied to your comment (
https://protect2.fireeye.com/url?k=be575021-e3854728-be56db6e-0cc47a31cdf8- 6c7d0e1e42762b92&q=1&u=https%3A%2F%2Flkml.org%2Flkml%2F2020%2F6% 2F15%2F1492),
Quoted 30 lines
Quoted 21 lines
but you didn't reply on that. I thought you agreed because you didn't
send
any more comments.Quoted 3 lines
Before submitting your next version patch, please check your L2P
mapping HPB reqeust submission logical algorithem. I have didWe are also reviewing the code that you submitted before. It seems to be a performance improvement as it sends a map request
directly.Quoted 5 lines
performance comparison testing on 4KB, there are about 13% performance
drop. Also the hit count is lower. I don't know if this is related
toQuoted 8 lines
It is interesting that there is actually a performance improvement.
Could you share the test environment, please? However, I think stability is
important to HPB driver. We have tested our method with the real
products and
the HPB 1.0 driver is based on that.Quoted 4 lines
I just run fio benchmark tool with --rw=randread, --bs=4kb, -- size=8G/10G/64G/100G. and see what performance diff with the direct
submission approach.Thanks!
Quoted 10 lines
Quoted 9 lines
After this patch, your approach can be done as an incremental patch?
I would
like to test the patch that you submitted and verify it.Quoted 4 lines
your current work queue scheduling, since you didn't add the timer
for
each HPB request.Quoted 8 lines
Taking into consideration of the HPB 2.0, can we submit the HPB write
request to the SCSI layer? if not, it will be a direct submission way.
why not directly use direct way? or maybe you have a more advisable
approach to work around this. would you please share with us. appreciate.I am considering a direct submission way for the next version. We will implement the write buffer command of HPB 2.0, after patching HPB
1.0.Quoted 5 lines
As for the direct submission of HPB releated command including HPB write
buffer, I think we'd better discuss the right approach in depth before
moving on to the next step.I vote to stay with the current implementation because:
1) Bean is probably right about 2.0, but it's out of scope for now - there is a long way to go before we'll need to worry about it
2) For now, we should focus on the functional flows. Performance issues, should such issues indeed exists, can be
dealt with later. And,
3) The current code base is running in production for more than 3 years now.
I am not so eager to dump a robust, well debugged code unless it
absolutely necessary.
Quoted 4 lines
Thanks,
Avri
Hi Avri
Thanks, appreciate you shared your position on this topic. I don't know how I can convince you to change your opinion.
Let me try.
1. HPB 2.0 is not out of scope.
HPB 1.0 only supports 4KB read length, which is useless. I don't know if there will be users who want to use HPB driver only supports 4KB chunk size. I think, we all know that some smartphone vendors have already use HPB 2.0, even HPB 2.0 has not been released yet. you mentioned this in your before emails. HPB 1.0 is just a transition(limited) version, we need to think about the HPB 2.0 support when we develop the HPB 1.0 driver.
To say the least, if we don't think about HPB 2.0 support, and just focus HPB 1.0, in the end, after HPB 2.0 releasing, we need to return original point, re-do lots thing, why we cannot fix it now and think one step further.
2. The major goal of the HPB feature is to improve random read performance, and HPB device mode implementing flow is now already very clear enough. I don't know what the functional flows you mentioned. if it is HPB host mode, no, this is another big topic, I think we'd better not add in current driver until we all have a final approach.
3. Regarding the Daejun's HPB driver used age, I can't easily jump to a conclusion. But for sure, before he disclosed his HPB driver and submitted to the community, he did lots of changes and deletions. That means it still needs lots of tests.
I didn't mean to disrupt Daejun's patch upstreaming. If Daejun can consider HPB 2.0 support while developing HPB 1.0 patch, that is super. Thus we can quickly add HPB 2.0 support once HPB 2.0 Spec released. Think about that who is now using HPB 1.0?
Thanks,
Bean
Quoted 80 lines
-----Original Message-----
From: Avri Altman <[email protected]>
Sent: 30 June 2020 12:09
To: [email protected]; Bean Huo <[email protected]>; [email protected]; [email protected]; [email protected]; [email protected]; [email protected]; [email protected]; [email protected]; ALIM AKHTAR <[email protected]>
Cc: [email protected]; [email protected]; Sang-yoon Oh <[email protected]>; Sung-Jun Park <[email protected]>; yongmyung lee <[email protected]>; Jinyoung CHOI <[email protected]>; Adel Choi <[email protected]>; BoRam Shin <[email protected]>
Subject: RE: [RFC PATCH v3 0/5] scsi: ufs: Add Host Performance Booster SupportHi,
Quoted 11 lines
Hi Bean,
Quoted 7 lines · Daejun Park
Quoted 6 lines
Quoted 3 lines
Seems you intentionally ignored to give you comments on my suggestion.
let me provide the reason.Sorry! I replied to your comment (
https://protect2.fireeye.com/url?k=be575021-e3854728-be56db6e-0cc47a31cdf8-
6c7d0e1e42762b92&q=1&u=https%3A%2F%2Flkml.org%2Flkml%2F2020%2F6%
Quoted 41 lines
2F15%2F1492),
Quoted 15 lines
Quoted 15 lines
but you didn't reply on that. I thought you agreed because you didn't send any more comments.
Quoted 2 lines
Before submitting your next version patch, please check your L2P mapping HPB reqeust submission logical algorithem. I have did
We are also reviewing the code that you submitted before. It seems to be a performance improvement as it sends a map request directly.
Quoted 3 lines
performance comparison testing on 4KB, there are about 13% performance drop. Also the hit count is lower. I don't know if this is related to
Quoted 8 lines
Quoted 4 lines
It is interesting that there is actually a performance improvement. Could you share the test environment, please? However, I think stability is important to HPB driver. We have tested our method with the real products and the HPB 1.0 driver is based on that.
I just run fio benchmark tool with --rw=randread, --bs=4kb, -- size=8G/10G/64G/100G. and see what performance diff with the direct submission approach.
Thanks!
Quoted 8 lines
Quoted 7 lines
After this patch, your approach can be done as an incremental patch? I would
like to test the patch that you submitted and verify it.Quoted 2 lines
your current work queue scheduling, since you didn't add the timer for each HPB request.
Quoted 5 lines
Taking into consideration of the HPB 2.0, can we submit the HPB write request to the SCSI layer? if not, it will be a direct submission way. why not directly use direct way? or maybe you have a more advisable approach to work around this. would you please share with us. appreciate.
Quoted 7 lines
I am considering a direct submission way for the next version. We will implement the write buffer command of HPB 2.0, after patching HPB 1.0.
As for the direct submission of HPB releated command including HPB write buffer, I think we'd better discuss the right approach in depth before moving on to the next step.
Quoted 10 lines
I vote to stay with the current implementation because:
1) Bean is probably right about 2.0, but it's out of scope for now - there is a long way to go before we'll need to worry about it
2) For now, we should focus on the functional flows.
Performance issues, should such issues indeed exists, can be dealt with later.
And,
3) The current code base is running in production for more than 3 years now. I am not so eager to dump a robust, well debugged code unless it absolutely
necessary.
Avri and Bean,
I think this is good approach to take, and let us add incremental patches to add future specification enhancements.
Quoted 3 lines
Thanks,
Avri
Quoted 8 lines · Daejun Park
Hi Bean,
Quoted 7 lines · Daejun Park
Quoted 6 lines
Quoted 3 lines
Seems you intentionally ignored to give you comments on my suggestion.
let me provide the reason.Sorry! I replied to your comment (
Quoted 58 lines
Quoted 30 lines
Quoted 22 lines
),
but you didn't reply on that. I thought you agreed because you didn't
send
any more comments.Quoted 3 lines
Before submitting your next version patch, please check your L2P
mapping HPB reqeust submission logical algorithem. I have didWe are also reviewing the code that you submitted before. It seems to be a performance improvement as it sends a map request
directly.Quoted 5 lines
performance comparison testing on 4KB, there are about 13% performance
drop. Also the hit count is lower. I don't know if this is related
toQuoted 7 lines
It is interesting that there is actually a performance improvement.
Could you share the test environment, please? However, I think stability is
important to HPB driver. We have tested our method with the real products and
the HPB 1.0 driver is based on that.Quoted 3 lines
I just run fio benchmark tool with --rw=randread, --bs=4kb, -- size=8G/10G/64G/100G. and see what performance diff with the direct submission approach.
Thanks!
Quoted 10 lines
Quoted 9 lines
After this patch, your approach can be done as an incremental patch?
I would
like to test the patch that you submitted and verify it.Quoted 4 lines
your current work queue scheduling, since you didn't add the timer
for
each HPB request.Quoted 7 lines
Taking into consideration of the HPB 2.0, can we submit the HPB write
request to the SCSI layer? if not, it will be a direct submission way.
why not directly use direct way? or maybe you have a more advisable approach to work around this. would you please share with us. appreciate.I am considering a direct submission way for the next version. We will implement the write buffer command of HPB 2.0, after patching HPB 1.0.
Quoted 6 lines
As for the direct submission of HPB releated command including HPB write
buffer, I think we'd better discuss the right approach in depth before
moving on to the next step.
Hi Daejun
If you need reference code, you can freely copy my code from my RFC v3 patchset. or if you need my side testing support, just let me, I can help you test your code.
Thanks,
Bean
Quoted 74 lines · Daejun Park
Quoted 8 lines
Hi Bean,
Quoted 7 lines · Daejun Park
Quoted 6 lines
Quoted 3 lines
Seems you intentionally ignored to give you comments on my suggestion.
let me provide the reason.Sorry! I replied to your comment (
Quoted 58 lines
Quoted 30 lines
Quoted 22 lines
),
but you didn't reply on that. I thought you agreed because you didn't
send
any more comments.Quoted 3 lines
Before submitting your next version patch, please check your L2P
mapping HPB reqeust submission logical algorithem. I have didWe are also reviewing the code that you submitted before. It seems to be a performance improvement as it sends a map request
directly.Quoted 5 lines
performance comparison testing on 4KB, there are about 13% performance
drop. Also the hit count is lower. I don't know if this is related
toQuoted 7 lines
It is interesting that there is actually a performance improvement.
Could you share the test environment, please? However, I think stability is
important to HPB driver. We have tested our method with the real products and
the HPB 1.0 driver is based on that.Quoted 3 lines
I just run fio benchmark tool with --rw=randread, --bs=4kb, -- size=8G/10G/64G/100G. and see what performance diff with the direct submission approach.
Thanks!
Quoted 10 lines
Quoted 9 lines
After this patch, your approach can be done as an incremental patch?
I would
like to test the patch that you submitted and verify it.Quoted 4 lines
your current work queue scheduling, since you didn't add the timer
for
each HPB request.Quoted 7 lines
Taking into consideration of the HPB 2.0, can we submit the HPB write
request to the SCSI layer? if not, it will be a direct submission way.
why not directly use direct way? or maybe you have a more advisable approach to work around this. would you please share with us. appreciate.I am considering a direct submission way for the next version. We will implement the write buffer command of HPB 2.0, after patching HPB 1.0.
Quoted 6 lines
As for the direct submission of HPB releated command including HPB write
buffer, I think we'd better discuss the right approach in depth before
moving on to the next step.
Quoted 5 lines
Hi Daejun
If you need reference code, you can freely copy my code from my RFC v3 patchset. or if you need my side testing support, just let me, I can help you test your code.
It will be good example code for developing HPB 2.0.
Thanks,
Daejun
If no-one else objects, maybe you can submit your patches as non-RFC for review?
Thanks,
Avri
Quoted 105 lines
-----Original Message-----
From: Daejun Park <[email protected]>
Sent: Tuesday, June 23, 2020 4:02 AM
To: Avri Altman <[email protected]>; [email protected]; [email protected]; [email protected]; [email protected]; [email protected]; [email protected]; [email protected]; [email protected]; ALIM AKHTAR <[email protected]>; Daejun Park <[email protected]>
Cc: [email protected]; [email protected]; Sang-yoon Oh <[email protected]>; Sung-Jun Park <[email protected]>; yongmyung lee <[email protected]>; Jinyoung CHOI <j- [email protected]>; Adel Choi <[email protected]>; BoRam Shin <[email protected]>
Subject: [RFC PATCH v3 0/5] scsi: ufs: Add Host Performance Booster SupportCAUTION: This email originated from outside of Western Digital. Do not click on links or open attachments unless you recognize the sender and know that the content is safe.
Changelog:
v2 -> v3
1. Add checking input module parameter value.
2. Change base commit from 5.8/scsi-queue to 5.9/scsi-queue.
3. Cleanup for unused variables and label.v1 -> v2
1. Change the full boilerplate text to SPDX style.
2. Adopt dynamic allocation for sub-region data structure.
3. Cleanup.NAND flash memory-based storage devices use Flash Translation Layer (FTL) to translate logical addresses of I/O requests to corresponding flash memory addresses. Mobile storage devices typically have RAM with constrained size, thus lack in memory to keep the whole mapping table. Therefore, mapping tables are partially retrieved from NAND flash on demand, causing random-read performance degradation.
To improve random read performance, JESD220-3 (HPB v1.0) proposes HPB (Host Performance Booster) which uses host system memory as a cache for the
FTL mapping table. By using HPB, FTL data can be read from host memory faster than from NAND flash memory.The current version only supports the DCM (device control mode). This patch consists of 4 parts to support HPB feature.
1) UFS-feature layer
2) HPB probe and initialization process
3) READ -> HPB READ using cached map information
4) L2P (logical to physical) map managementThe UFS-feature is an additional layer to avoid the structure in which the UFS-core driver and the UFS-feature are entangled with each other in a single module.
By adding the layer, UFS-features composed of various combinations can be supported. Also, even if a new feature is added, modification of the UFS-core driver can be minimized.In the HPB probe and init process, the device information of the UFS is queried. After checking supported features, the data structure for the HPB is initialized according to the device information.
A read I/O in the active sub-region where the map is cached is changed to HPB READ by the HPB module.
The HPB module manages the L2P map using information received from the device. For active sub-region, the HPB module caches through
ufshpb_maprequest. For the in-active region, the HPB module discards the L2P map. When a write I/O occurs in an active sub-region area, associated dirty bitmap checked as dirty for preventing stale read.HPB is shown to have a performance improvement of 58 - 67% for random read
workload. [1]This series patches are based on the 5.9/scsi-queue branch.
[1]:
https://www.usenix.org/conference/hotstorage17/program/presentation/jeo
ngDaejun park (5):
scsi: ufs: Add UFS feature related parameter
scsi: ufs: Add UFS feature layer
scsi: ufs: Introduce HPB module
scsi: ufs: L2P map management for HPB read
scsi: ufs: Prepare HPB read for cached sub-regiondrivers/scsi/ufs/Kconfig | 9 + drivers/scsi/ufs/Makefile | 3 +- drivers/scsi/ufs/ufs.h | 12 + drivers/scsi/ufs/ufsfeature.c | 148 +++ drivers/scsi/ufs/ufsfeature.h | 69 ++ drivers/scsi/ufs/ufshcd.c | 23 +- drivers/scsi/ufs/ufshcd.h | 3 + drivers/scsi/ufs/ufshpb.c | 1996 ++++++++++++++++++++++++++++++++++++ drivers/scsi/ufs/ufshpb.h | 234 +++++ 9 files changed, 2494 insertions(+), 3 deletions(-)created mode 100644
drivers/scsi/ufs/ufsfeature.ccreated mode 100644drivers/scsi/ufs/ufsfeature.hcreated mode 100644drivers/scsi/ufs/ufshpb.ccreated mode 100644drivers/scsi/ufs/ufshpb.h