From 4139e5680bcf3016561c8830d914ee0b0ab741d1 Mon Sep 17 00:00:00 2001 From: Neil Armstrong Date: Mon, 30 Sep 2024 14:44:25 +0200 Subject: [PATCH] ufs: split flush and invalidate to only invalidate when required There is no need to flush and invalidate all data updated by the driver, mainly because on ARM platforms flush also invalidates the cachelines. Split the function in two and add the appropriate cacheline invalidates after the UFS DMA operation finishes to make sure we read from memory. Flushing then invalidating cacheline unaligned data causes data corruption issues on Qualcomm platforms, and is largely unnecessary anyway, so let's cleanup the cache operations. Reviewed-by: Neha Malcom Francis Tested-by: Venkatesh Yadav Abbarapu Tested-by: Julius Lehmann Link: https://lore.kernel.org/r/20240930-topic-ufs-enhancements-v3-3-58234f84ab89@linaro.org Signed-off-by: Neil Armstrong --- drivers/ufs/ufs.c | 45 ++++++++++++++++++++++++++++++--------------- 1 file changed, 30 insertions(+), 15 deletions(-) diff --git a/drivers/ufs/ufs.c b/drivers/ufs/ufs.c index 3d9a7d7ee1..5845fd694d 100644 --- a/drivers/ufs/ufs.c +++ b/drivers/ufs/ufs.c @@ -696,17 +696,28 @@ static inline u8 ufshcd_get_upmcrs(struct ufs_hba *hba) } /** - * ufshcd_cache_flush_and_invalidate - Flush and invalidate cache + * ufshcd_cache_flush - Flush cache * - * Flush and invalidate cache in aligned address..address+size range. - * The invalidation is in place to avoid stale data in cache. + * Flush cache in aligned address..address+size range. */ -static void ufshcd_cache_flush_and_invalidate(void *addr, unsigned long size) +static void ufshcd_cache_flush(void *addr, unsigned long size) { uintptr_t start_addr = (uintptr_t)addr & ~(ARCH_DMA_MINALIGN - 1); uintptr_t end_addr = ALIGN((uintptr_t)addr + size, ARCH_DMA_MINALIGN); flush_dcache_range(start_addr, end_addr); +} + +/** + * ufshcd_cache_invalidate - Invalidate cache + * + * Invalidate cache in aligned address..address+size range. + */ +static void ufshcd_cache_invalidate(void *addr, unsigned long size) +{ + uintptr_t start_addr = (uintptr_t)addr & ~(ARCH_DMA_MINALIGN - 1); + uintptr_t end_addr = ALIGN((uintptr_t)addr + size, ARCH_DMA_MINALIGN); + invalidate_dcache_range(start_addr, end_addr); } @@ -754,7 +765,7 @@ static void ufshcd_prepare_req_desc_hdr(struct ufs_hba *hba, req_desc->prd_table_length = 0; - ufshcd_cache_flush_and_invalidate(req_desc, sizeof(*req_desc)); + ufshcd_cache_flush(req_desc, sizeof(*req_desc)); } static void ufshcd_prepare_utp_query_req_upiu(struct ufs_hba *hba, @@ -785,13 +796,13 @@ static void ufshcd_prepare_utp_query_req_upiu(struct ufs_hba *hba, /* Copy the Descriptor */ if (query->request.upiu_req.opcode == UPIU_QUERY_OPCODE_WRITE_DESC) { memcpy(ucd_req_ptr + 1, query->descriptor, len); - ufshcd_cache_flush_and_invalidate(ucd_req_ptr, 2 * sizeof(*ucd_req_ptr)); + ufshcd_cache_flush(ucd_req_ptr, 2 * sizeof(*ucd_req_ptr)); } else { - ufshcd_cache_flush_and_invalidate(ucd_req_ptr, sizeof(*ucd_req_ptr)); + ufshcd_cache_flush(ucd_req_ptr, sizeof(*ucd_req_ptr)); } memset(hba->ucd_rsp_ptr, 0, sizeof(struct utp_upiu_rsp)); - ufshcd_cache_flush_and_invalidate(hba->ucd_rsp_ptr, sizeof(*hba->ucd_rsp_ptr)); + ufshcd_cache_flush(hba->ucd_rsp_ptr, sizeof(*hba->ucd_rsp_ptr)); } static inline void ufshcd_prepare_utp_nop_upiu(struct ufs_hba *hba) @@ -809,8 +820,8 @@ static inline void ufshcd_prepare_utp_nop_upiu(struct ufs_hba *hba) memset(hba->ucd_rsp_ptr, 0, sizeof(struct utp_upiu_rsp)); - ufshcd_cache_flush_and_invalidate(ucd_req_ptr, sizeof(*ucd_req_ptr)); - ufshcd_cache_flush_and_invalidate(hba->ucd_rsp_ptr, sizeof(*hba->ucd_rsp_ptr)); + ufshcd_cache_flush(ucd_req_ptr, sizeof(*ucd_req_ptr)); + ufshcd_cache_flush(hba->ucd_rsp_ptr, sizeof(*hba->ucd_rsp_ptr)); } /** @@ -877,6 +888,8 @@ static int ufshcd_send_command(struct ufs_hba *hba, unsigned int task_tag) */ static inline int ufshcd_get_req_rsp(struct utp_upiu_rsp *ucd_rsp_ptr) { + ufshcd_cache_invalidate(ucd_rsp_ptr, sizeof(*ucd_rsp_ptr)); + return be32_to_cpu(ucd_rsp_ptr->header.dword_0) >> 24; } @@ -888,6 +901,8 @@ static inline int ufshcd_get_tr_ocs(struct ufs_hba *hba) { struct utp_transfer_req_desc *req_desc = hba->utrdl; + ufshcd_cache_invalidate(req_desc, sizeof(*req_desc)); + return le32_to_cpu(req_desc->header.dword_2) & MASK_OCS; } @@ -1437,8 +1452,8 @@ void ufshcd_prepare_utp_scsi_cmd_upiu(struct ufs_hba *hba, memcpy(ucd_req_ptr->sc.cdb, pccb->cmd, cdb_len); memset(hba->ucd_rsp_ptr, 0, sizeof(struct utp_upiu_rsp)); - ufshcd_cache_flush_and_invalidate(ucd_req_ptr, sizeof(*ucd_req_ptr)); - ufshcd_cache_flush_and_invalidate(hba->ucd_rsp_ptr, sizeof(*hba->ucd_rsp_ptr)); + ufshcd_cache_flush(ucd_req_ptr, sizeof(*ucd_req_ptr)); + ufshcd_cache_flush(hba->ucd_rsp_ptr, sizeof(*hba->ucd_rsp_ptr)); } static inline void prepare_prdt_desc(struct ufshcd_sg_entry *entry, @@ -1461,7 +1476,7 @@ static void prepare_prdt_table(struct ufs_hba *hba, struct scsi_cmd *pccb) if (!datalen) { req_desc->prd_table_length = 0; - ufshcd_cache_flush_and_invalidate(req_desc, sizeof(*req_desc)); + ufshcd_cache_flush(req_desc, sizeof(*req_desc)); return; } @@ -1487,8 +1502,8 @@ static void prepare_prdt_table(struct ufs_hba *hba, struct scsi_cmd *pccb) prepare_prdt_desc(&prd_table[table_length - i - 1], buf, datalen - 1); req_desc->prd_table_length = table_length; - ufshcd_cache_flush_and_invalidate(prd_table, sizeof(*prd_table) * table_length); - ufshcd_cache_flush_and_invalidate(req_desc, sizeof(*req_desc)); + ufshcd_cache_flush(prd_table, sizeof(*prd_table) * table_length); + ufshcd_cache_flush(req_desc, sizeof(*req_desc)); } static int ufs_scsi_exec(struct udevice *scsi_dev, struct scsi_cmd *pccb) -- 2.39.5