From 5149452ca66289ef33d13897ee845a2f6f5b680f Mon Sep 17 00:00:00 2001 From: Po-Wen Kao Date: Thu, 4 May 2023 23:44:49 +0800 Subject: [PATCH 1/6] scsi: ufs: core: Fix MCQ tag calculation The transfer command descriptor is allocated in ufshcd_memory_alloc() and referenced by the transfer request descriptor with stride size sizeof_utp_transfer_cmd_desc() instead of sizeof(struct utp_transfer_cmd_desc). Consequently, computing tag by address offset should also refer to the same stride. Signed-off-by: Po-Wen Kao Link: https://lore.kernel.org/r/20230504154454.26654-2-powen.kao@mediatek.com Reviewed-by: Bart Van Assche Reviewed-by: Manivannan Sadhasivam Reviewed-by: Stanley Chu Reviewed-by: Ziqi Chen Signed-off-by: Martin K. Petersen --- drivers/ufs/core/ufs-mcq.c | 2 +- drivers/ufs/core/ufshcd.c | 2 +- 2 files changed, 2 insertions(+), 2 deletions(-) diff --git a/drivers/ufs/core/ufs-mcq.c b/drivers/ufs/core/ufs-mcq.c index 202ff71e1b58..b7c5f39b50e6 100644 --- a/drivers/ufs/core/ufs-mcq.c +++ b/drivers/ufs/core/ufs-mcq.c @@ -265,7 +265,7 @@ static int ufshcd_mcq_get_tag(struct ufs_hba *hba, addr = (le64_to_cpu(cqe->command_desc_base_addr) & CQE_UCD_BA) - hba->ucdl_dma_addr; - return div_u64(addr, sizeof(struct utp_transfer_cmd_desc)); + return div_u64(addr, sizeof_utp_transfer_cmd_desc(hba)); } static void ufshcd_mcq_process_cqe(struct ufs_hba *hba, diff --git a/drivers/ufs/core/ufshcd.c b/drivers/ufs/core/ufshcd.c index 45fd374fe56c..984abbe1fb93 100644 --- a/drivers/ufs/core/ufshcd.c +++ b/drivers/ufs/core/ufshcd.c @@ -8452,7 +8452,7 @@ static void ufshcd_release_sdb_queue(struct ufs_hba *hba, int nutrs) { size_t ucdl_size, utrdl_size; - ucdl_size = sizeof(struct utp_transfer_cmd_desc) * nutrs; + ucdl_size = sizeof_utp_transfer_cmd_desc(hba) * nutrs; dmam_free_coherent(hba->dev, ucdl_size, hba->ucdl_base_addr, hba->ucdl_dma_addr); From 06caeb536b2b21668efd2d6fa97c09461957b3a7 Mon Sep 17 00:00:00 2001 From: Po-Wen Kao Date: Thu, 4 May 2023 23:44:50 +0800 Subject: [PATCH 2/6] scsi: ufs: core: Rename symbol sizeof_utp_transfer_cmd_desc() Naming the functions after standard operators like sizeof() may cause confusion. Rename it to ufshcd_get_ucd_size(). Signed-off-by: Po-Wen Kao Link: https://lore.kernel.org/r/20230504154454.26654-3-powen.kao@mediatek.com Suggested-by: Manivannan Sadhasivam Reviewed-by: Stanley Chu Reviewed-by: Ziqi Chen Reviewed-by: Bart Van Assche Signed-off-by: Martin K. Petersen --- drivers/ufs/core/ufs-mcq.c | 2 +- drivers/ufs/core/ufshcd.c | 10 +++++----- include/ufs/ufshcd.h | 2 +- 3 files changed, 7 insertions(+), 7 deletions(-) diff --git a/drivers/ufs/core/ufs-mcq.c b/drivers/ufs/core/ufs-mcq.c index b7c5f39b50e6..937933d3f77c 100644 --- a/drivers/ufs/core/ufs-mcq.c +++ b/drivers/ufs/core/ufs-mcq.c @@ -265,7 +265,7 @@ static int ufshcd_mcq_get_tag(struct ufs_hba *hba, addr = (le64_to_cpu(cqe->command_desc_base_addr) & CQE_UCD_BA) - hba->ucdl_dma_addr; - return div_u64(addr, sizeof_utp_transfer_cmd_desc(hba)); + return div_u64(addr, ufshcd_get_ucd_size(hba)); } static void ufshcd_mcq_process_cqe(struct ufs_hba *hba, diff --git a/drivers/ufs/core/ufshcd.c b/drivers/ufs/core/ufshcd.c index 984abbe1fb93..e7e79f515e14 100644 --- a/drivers/ufs/core/ufshcd.c +++ b/drivers/ufs/core/ufshcd.c @@ -2849,10 +2849,10 @@ static void ufshcd_map_queues(struct Scsi_Host *shost) static void ufshcd_init_lrb(struct ufs_hba *hba, struct ufshcd_lrb *lrb, int i) { struct utp_transfer_cmd_desc *cmd_descp = (void *)hba->ucdl_base_addr + - i * sizeof_utp_transfer_cmd_desc(hba); + i * ufshcd_get_ucd_size(hba); struct utp_transfer_req_desc *utrdlp = hba->utrdl_base_addr; dma_addr_t cmd_desc_element_addr = hba->ucdl_dma_addr + - i * sizeof_utp_transfer_cmd_desc(hba); + i * ufshcd_get_ucd_size(hba); u16 response_offset = offsetof(struct utp_transfer_cmd_desc, response_upiu); u16 prdt_offset = offsetof(struct utp_transfer_cmd_desc, prd_table); @@ -3761,7 +3761,7 @@ static int ufshcd_memory_alloc(struct ufs_hba *hba) size_t utmrdl_size, utrdl_size, ucdl_size; /* Allocate memory for UTP command descriptors */ - ucdl_size = sizeof_utp_transfer_cmd_desc(hba) * hba->nutrs; + ucdl_size = ufshcd_get_ucd_size(hba) * hba->nutrs; hba->ucdl_base_addr = dmam_alloc_coherent(hba->dev, ucdl_size, &hba->ucdl_dma_addr, @@ -3861,7 +3861,7 @@ static void ufshcd_host_memory_configure(struct ufs_hba *hba) prdt_offset = offsetof(struct utp_transfer_cmd_desc, prd_table); - cmd_desc_size = sizeof_utp_transfer_cmd_desc(hba); + cmd_desc_size = ufshcd_get_ucd_size(hba); cmd_desc_dma_addr = hba->ucdl_dma_addr; for (i = 0; i < hba->nutrs; i++) { @@ -8452,7 +8452,7 @@ static void ufshcd_release_sdb_queue(struct ufs_hba *hba, int nutrs) { size_t ucdl_size, utrdl_size; - ucdl_size = sizeof_utp_transfer_cmd_desc(hba) * nutrs; + ucdl_size = ufshcd_get_ucd_size(hba) * nutrs; dmam_free_coherent(hba->dev, ucdl_size, hba->ucdl_base_addr, hba->ucdl_dma_addr); diff --git a/include/ufs/ufshcd.h b/include/ufs/ufshcd.h index f7553293ba98..df1d04f7a542 100644 --- a/include/ufs/ufshcd.h +++ b/include/ufs/ufshcd.h @@ -1133,7 +1133,7 @@ static inline size_t ufshcd_sg_entry_size(const struct ufs_hba *hba) ({ (void)(hba); BUILD_BUG_ON(sg_entry_size != sizeof(struct ufshcd_sg_entry)); }) #endif -static inline size_t sizeof_utp_transfer_cmd_desc(const struct ufs_hba *hba) +static inline size_t ufshcd_get_ucd_size(const struct ufs_hba *hba) { return sizeof(struct utp_transfer_cmd_desc) + SG_ALL * ufshcd_sg_entry_size(hba); } From 72a81bb0b6fc9b759ac0fdaca3ec5884a8b2f304 Mon Sep 17 00:00:00 2001 From: Po-Wen Kao Date: Thu, 4 May 2023 23:44:51 +0800 Subject: [PATCH 3/6] scsi: ufs: core: Fix MCQ nr_hw_queues Since MAXQ is 0-based value, add one to obtain number of hardware queues. Signed-off-by: Po-Wen Kao Link: https://lore.kernel.org/r/20230504154454.26654-4-powen.kao@mediatek.com Reviewed-by: Bean Huo Reviewed-by: Manivannan Sadhasivam Reviewed-by: Bart Van Assche Reviewed-by: Stanley Chu Signed-off-by: Martin K. Petersen --- drivers/ufs/core/ufs-mcq.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) diff --git a/drivers/ufs/core/ufs-mcq.c b/drivers/ufs/core/ufs-mcq.c index 937933d3f77c..51b3c6ae781d 100644 --- a/drivers/ufs/core/ufs-mcq.c +++ b/drivers/ufs/core/ufs-mcq.c @@ -150,7 +150,8 @@ static int ufshcd_mcq_config_nr_queues(struct ufs_hba *hba) u32 hba_maxq, rem, tot_queues; struct Scsi_Host *host = hba->host; - hba_maxq = FIELD_GET(MAX_QUEUE_SUP, hba->mcq_capabilities); + /* maxq is 0 based value */ + hba_maxq = FIELD_GET(MAX_QUEUE_SUP, hba->mcq_capabilities) + 1; tot_queues = UFS_MCQ_NUM_DEV_CMD_QUEUES + read_queues + poll_queues + rw_queues; From 4e81a6cba517cb33584308a331f14f5e3fec369b Mon Sep 17 00:00:00 2001 From: Michael Kelley Date: Mon, 15 May 2023 10:20:41 -0700 Subject: [PATCH 4/6] scsi: storvsc: Don't pass unused PFNs to Hyper-V host In a SCSI request, storvsc pre-allocates space for up to MAX_PAGE_BUFFER_COUNT physical frame numbers to be passed to Hyper-V. If the size of the I/O request requires more PFNs, a separate memory area of exactly the correct size is dynamically allocated. But when the pre-allocated area is used, current code always passes MAX_PAGE_BUFFER_COUNT PFNs to Hyper-V, even if fewer are needed. While this doesn't break anything because the additional PFNs are always zero, more bytes than necessary are copied into the VMBus channel ring buffer. This takes CPU cycles and wastes space in the ring buffer. For a typical 4 Kbyte I/O that requires only a single PFN, 248 unnecessary bytes are copied. Fix this by setting the payload_sz based on the actual number of PFNs required, not the size of the pre-allocated space. Reported-by: John Starks Fixes: 8f43710543ef ("scsi: storvsc: Support PAGE_SIZE larger than 4K") Signed-off-by: Michael Kelley Link: https://lore.kernel.org/r/1684171241-16209-1-git-send-email-mikelley@microsoft.com Signed-off-by: Martin K. Petersen --- drivers/scsi/storvsc_drv.c | 8 ++++---- 1 file changed, 4 insertions(+), 4 deletions(-) diff --git a/drivers/scsi/storvsc_drv.c b/drivers/scsi/storvsc_drv.c index d9ce379c4d2e..e6bc622954cf 100644 --- a/drivers/scsi/storvsc_drv.c +++ b/drivers/scsi/storvsc_drv.c @@ -1780,7 +1780,7 @@ static int storvsc_queuecommand(struct Scsi_Host *host, struct scsi_cmnd *scmnd) length = scsi_bufflen(scmnd); payload = (struct vmbus_packet_mpb_array *)&cmd_request->mpb; - payload_sz = sizeof(cmd_request->mpb); + payload_sz = 0; if (scsi_sg_count(scmnd)) { unsigned long offset_in_hvpg = offset_in_hvpage(sgl->offset); @@ -1789,10 +1789,10 @@ static int storvsc_queuecommand(struct Scsi_Host *host, struct scsi_cmnd *scmnd) unsigned long hvpfn, hvpfns_to_add; int j, i = 0, sg_count; - if (hvpg_count > MAX_PAGE_BUFFER_COUNT) { + payload_sz = (hvpg_count * sizeof(u64) + + sizeof(struct vmbus_packet_mpb_array)); - payload_sz = (hvpg_count * sizeof(u64) + - sizeof(struct vmbus_packet_mpb_array)); + if (hvpg_count > MAX_PAGE_BUFFER_COUNT) { payload = kzalloc(payload_sz, GFP_ATOMIC); if (!payload) return SCSI_MLQUEUE_DEVICE_BUSY; From 6ca9818d1624e136a76ae8faedb6b6c95ca66903 Mon Sep 17 00:00:00 2001 From: Wenchao Hao Date: Mon, 15 May 2023 15:01:55 +0800 Subject: [PATCH 5/6] scsi: Revert "scsi: core: Do not increase scsi_device's iorequest_cnt if dispatch failed" The "atomic_inc(&cmd->device->iorequest_cnt)" in scsi_queue_rq() would cause kernel panic because cmd->device may be freed after returning from scsi_dispatch_cmd(). This reverts commit cfee29ffb45b1c9798011b19d454637d1b0fe87d. Signed-off-by: Wenchao Hao Reported-by: Ming Lei Closes: https://lore.kernel.org/r/ZF+zB+bB7iqe0wGd@ovpn-8-17.pek2.redhat.com Link: https://lore.kernel.org/r/20230515070156.1790181-2-haowenchao2@huawei.com Reviewed-by: Ming Lei Signed-off-by: Martin K. Petersen --- drivers/scsi/scsi_lib.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) diff --git a/drivers/scsi/scsi_lib.c b/drivers/scsi/scsi_lib.c index b7c569a42aa4..03964b26f3f2 100644 --- a/drivers/scsi/scsi_lib.c +++ b/drivers/scsi/scsi_lib.c @@ -1463,6 +1463,8 @@ static int scsi_dispatch_cmd(struct scsi_cmnd *cmd) struct Scsi_Host *host = cmd->device->host; int rtn = 0; + atomic_inc(&cmd->device->iorequest_cnt); + /* check if the device is still usable */ if (unlikely(cmd->device->sdev_state == SDEV_DEL)) { /* in SDEV_DEL we error all commands. DID_NO_CONNECT @@ -1761,7 +1763,6 @@ static blk_status_t scsi_queue_rq(struct blk_mq_hw_ctx *hctx, goto out_dec_host_busy; } - atomic_inc(&cmd->device->iorequest_cnt); return BLK_STS_OK; out_dec_host_busy: From 09e797c8641f6ad435c33ae24c223351197ea29a Mon Sep 17 00:00:00 2001 From: Wenchao Hao Date: Mon, 15 May 2023 15:01:56 +0800 Subject: [PATCH 6/6] scsi: core: Decrease scsi_device's iorequest_cnt if dispatch failed If scsi_dispatch_cmd() failed, the SCSI command was not sent to the target, scsi_queue_rq() would return BLK_STS_RESOURCE and the related request would be requeued. The timeout of this request would not fire, no one would increase iodone_cnt. The above flow would result the iodone_cnt smaller than iorequest_cnt. So decrease the iorequest_cnt if dispatch failed to workaround the issue. Signed-off-by: Wenchao Hao Reported-by: Ming Lei Closes: https://lore.kernel.org/r/ZF+zB+bB7iqe0wGd@ovpn-8-17.pek2.redhat.com Link: https://lore.kernel.org/r/20230515070156.1790181-3-haowenchao2@huawei.com Reviewed-by: Ming Lei Signed-off-by: Martin K. Petersen --- drivers/scsi/scsi_lib.c | 2 ++ 1 file changed, 2 insertions(+) diff --git a/drivers/scsi/scsi_lib.c b/drivers/scsi/scsi_lib.c index 03964b26f3f2..0226c9279cef 100644 --- a/drivers/scsi/scsi_lib.c +++ b/drivers/scsi/scsi_lib.c @@ -1485,6 +1485,7 @@ static int scsi_dispatch_cmd(struct scsi_cmnd *cmd) */ SCSI_LOG_MLQUEUE(3, scmd_printk(KERN_INFO, cmd, "queuecommand : device blocked\n")); + atomic_dec(&cmd->device->iorequest_cnt); return SCSI_MLQUEUE_DEVICE_BUSY; } @@ -1517,6 +1518,7 @@ static int scsi_dispatch_cmd(struct scsi_cmnd *cmd) trace_scsi_dispatch_cmd_start(cmd); rtn = host->hostt->queuecommand(host, cmd); if (rtn) { + atomic_dec(&cmd->device->iorequest_cnt); trace_scsi_dispatch_cmd_error(cmd, rtn); if (rtn != SCSI_MLQUEUE_DEVICE_BUSY && rtn != SCSI_MLQUEUE_TARGET_BUSY)