scsi: megaraid_sas: SAS3.5 Generic Megaraid Controllers Stream Detection and IO Coalescing

Detect sequential Write IOs and pass the hint that it is part of sequential
stream to help HBA Firmware do the Full Stripe Writes. For read IOs on
certain RAID volumes like Read Ahead volumes,this will help driver to
send it to Firmware even if the IOs can potentially be sent to
hardware directly (called fast path) bypassing firmware.

Design: 8 streams are maintained per RAID volume as per the combined
firmware/driver design. When there is no stream detected the LRU stream
is used for next potential stream and LRU/MRU map is updated to make this
as MRU stream. Every time a stream is detected the MRU map
is updated to make the current stream as MRU stream.
Signed-off-by: Sasikumar Chandrasekaran <sasikumar.pc@broadcom.com>
Reviewed-by: Tomas Henzl <thenzl@redhat.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

drivers/scsi/megaraid/megaraid_sas.h

@@ -2070,6 +2070,7 @@ struct megasas_instance {
 	/* used to sync fire the cmd to fw */
 	spinlock_t hba_lock;
 	/* used to synch producer, consumer ptrs in dpc */
+	spinlock_t stream_lock;
 	spinlock_t completion_lock;
 	struct dma_pool *frame_dma_pool;
 	struct dma_pool *sense_dma_pool;

drivers/scsi/megaraid/megaraid_sas_base.c

@@ -5001,7 +5001,7 @@ static int megasas_init_fw(struct megasas_instance *instance)
 	struct megasas_register_set __iomem *reg_set;
 	struct megasas_ctrl_info *ctrl_info = NULL;
 	unsigned long bar_list;
-	int i, loop, fw_msix_count = 0;
+	int i, j, loop, fw_msix_count = 0;
 	struct IOV_111 *iovPtr;
 	struct fusion_context *fusion;
 
@@ -5194,6 +5194,36 @@ static int megasas_init_fw(struct megasas_instance *instance)
 	}
 
 	memset(instance->ld_ids, 0xff, MEGASAS_MAX_LD_IDS);
+
+	/* stream detection initialization */
+	if (instance->is_ventura) {
+		fusion->stream_detect_by_ld =
+		kzalloc(sizeof(struct LD_STREAM_DETECT *)
+		* MAX_LOGICAL_DRIVES_EXT,
+		GFP_KERNEL);
+		if (!fusion->stream_detect_by_ld) {
+			dev_err(&instance->pdev->dev,
+					"unable to allocate stream detection for pool of LDs\n");
+			goto fail_get_ld_pd_list;
+		}
+		for (i = 0; i < MAX_LOGICAL_DRIVES_EXT; ++i) {
+			fusion->stream_detect_by_ld[i] =
+				kmalloc(sizeof(struct LD_STREAM_DETECT),
+				GFP_KERNEL);
+			if (!fusion->stream_detect_by_ld[i]) {
+				dev_err(&instance->pdev->dev,
+					"unable to allocate stream detect by LD\n ");
+				for (j = 0; j < i; ++j)
+					kfree(fusion->stream_detect_by_ld[j]);
+				kfree(fusion->stream_detect_by_ld);
+				fusion->stream_detect_by_ld = NULL;
+				goto fail_get_ld_pd_list;
+			}
+			fusion->stream_detect_by_ld[i]->mru_bit_map
+				= MR_STREAM_BITMAP;
+		}
+	}
+
 	if (megasas_ld_list_query(instance,
 				  MR_LD_QUERY_TYPE_EXPOSED_TO_HOST))
 		megasas_get_ld_list(instance);
@@ -5313,6 +5343,8 @@ static int megasas_init_fw(struct megasas_instance *instance)
 
 	return 0;
 
+fail_get_ld_pd_list:
+	instance->instancet->disable_intr(instance);
 fail_get_pd_list:
 	instance->instancet->disable_intr(instance);
 fail_init_adapter:
@@ -5846,6 +5878,7 @@ static int megasas_probe_one(struct pci_dev *pdev,
 
 	spin_lock_init(&instance->mfi_pool_lock);
 	spin_lock_init(&instance->hba_lock);
+	spin_lock_init(&instance->stream_lock);
 	spin_lock_init(&instance->completion_lock);
 
 	mutex_init(&instance->reset_mutex);
@@ -6353,6 +6386,14 @@ static void megasas_detach_one(struct pci_dev *pdev)
 	if (instance->msix_vectors)
 		pci_free_irq_vectors(instance->pdev);
 
+	if (instance->is_ventura) {
+		for (i = 0; i < MAX_LOGICAL_DRIVES_EXT; ++i)
+			kfree(fusion->stream_detect_by_ld[i]);
+		kfree(fusion->stream_detect_by_ld);
+		fusion->stream_detect_by_ld = NULL;
+	}
+
+
 	if (instance->ctrl_context) {
 		megasas_release_fusion(instance);
 			pd_seq_map_sz = sizeof(struct MR_PD_CFG_SEQ_NUM_SYNC) +

drivers/scsi/megaraid/megaraid_sas_fp.c

@@ -935,6 +935,8 @@ MR_BuildRaidContext(struct megasas_instance *instance,
 
 	ld = MR_TargetIdToLdGet(ldTgtId, map);
 	raid = MR_LdRaidGet(ld, map);
+	/*check read ahead bit*/
+	io_info->ra_capable = raid->capability.ra_capable;
 
 	/*
 	 * if rowDataSize @RAID map and spanRowDataSize @SPAN INFO are zero

drivers/scsi/megaraid/megaraid_sas_fusion.h

@@ -133,12 +133,95 @@ struct RAID_CONTEXT {
 	u8      resvd2;
 };
 
+/*
+ * Raid Context structure which describes ventura MegaRAID specific
+ * IO Paramenters ,This resides at offset 0x60 where the SGL normally
+ * starts in MPT IO Frames
+ */
+struct RAID_CONTEXT_G35 {
+#if   defined(__BIG_ENDIAN_BITFIELD)
+	u16	resvd0:8;
+	u16	nseg:4;
+	u16	type:4;
+#else
+	u16	type:4;		    /* 0x00 */
+	u16	nseg:4;		    /* 0x00 */
+	u16 resvd0:8;
+#endif
+	u16 timeout_value; /* 0x02 -0x03 */
+	union {
+		struct {
+#if	defined(__BIG_ENDIAN_BITFIELD)
+		u16	set_divert:4;
+		u16	cpu_sel:4;
+		u16	log:1;
+		u16	rw:1;
+		u16	sbs:1;
+		u16	sqn:1;
+		u16	fwn:1;
+		u16	c2f:1;
+		u16	sld:1;
+		u16	reserved:1;
+#else
+		u16	reserved:1;
+		u16	sld:1;
+		u16	c2f:1;
+		u16	fwn:1;
+		u16	sqn:1;
+		u16	sbs:1;
+		u16	rw:1;
+		u16	log:1;
+		u16	cpu_sel:4;
+		u16	set_divert:4;
+#endif
+			} bits;
+		u16 s;
+	} routing_flags;	/* 0x04 -0x05 routing flags */
+	u16 virtual_disk_tgt_id;   /* 0x06 -0x07 */
+	u64 reg_lock_row_lba;      /* 0x08 - 0x0F */
+	u32 reg_lock_length;      /* 0x10 - 0x13 */
+	union {
+		u16 next_lmid; /* 0x14 - 0x15 */
+		u16	peer_smid;	/* used for the raid 1/10 fp writes */
+	} smid;
+	u8 ex_status;       /* 0x16 : OUT */
+	u8 status;          /* 0x17 status */
+	u8 RAIDFlags;		/* 0x18 resvd[7:6], ioSubType[5:4],
+				 * resvd[3:1], preferredCpu[0]
+				 */
+	u8 span_arm;            /* 0x1C span[7:5], arm[4:0] */
+	u16	config_seq_num;           /* 0x1A -0x1B */
+#if   defined(__BIG_ENDIAN_BITFIELD) /* 0x1C - 0x1D */
+	u16 stream_detected:1;
+	u16 reserved:3;
+	u16 num_sge:12;
+#else
+	u16 num_sge:12;
+	u16 reserved:3;
+	u16 stream_detected:1;
+#endif
+	u8 resvd2[2];          /* 0x1E-0x1F */
+};
+
+union RAID_CONTEXT_UNION {
+	struct RAID_CONTEXT raid_context;
+	struct RAID_CONTEXT_G35 raid_context_g35;
+};
+
 #define RAID_CTX_SPANARM_ARM_SHIFT	(0)
 #define RAID_CTX_SPANARM_ARM_MASK	(0x1f)
 
 #define RAID_CTX_SPANARM_SPAN_SHIFT	(5)
 #define RAID_CTX_SPANARM_SPAN_MASK	(0xE0)
 
+/* number of bits per index in U32 TrackStream */
+#define BITS_PER_INDEX_STREAM		4
+#define INVALID_STREAM_NUM              16
+#define MR_STREAM_BITMAP		0x76543210
+#define STREAM_MASK			((1 << BITS_PER_INDEX_STREAM) - 1)
+#define ZERO_LAST_STREAM		0x0fffffff
+#define MAX_STREAMS_TRACKED		8
+
 /*
  * define region lock types
  */
@@ -409,7 +492,7 @@ struct MPI2_RAID_SCSI_IO_REQUEST {
 	u8                      LUN[8];                         /* 0x34 */
 	__le32			Control;                        /* 0x3C */
 	union MPI2_SCSI_IO_CDB_UNION  CDB;			/* 0x40 */
-	struct RAID_CONTEXT	RaidContext;                    /* 0x60 */
+	union RAID_CONTEXT_UNION RaidContext;  /* 0x60 */
 	union MPI2_SGE_IO_UNION       SGL;			/* 0x80 */
 };
 
@@ -656,11 +739,13 @@ struct MR_LD_RAID {
 		u32     encryptionType:8;
 		u32     pdPiMode:4;
 		u32     ldPiMode:4;
-		u32     reserved5:3;
+		u32 reserved5:2;
+		u32 ra_capable:1;
 		u32     fpCapable:1;
 #else
 		u32     fpCapable:1;
-		u32     reserved5:3;
+		u32 ra_capable:1;
+		u32 reserved5:2;
 		u32     ldPiMode:4;
 		u32     pdPiMode:4;
 		u32     encryptionType:8;
@@ -745,6 +830,7 @@ struct IO_REQUEST_INFO {
 	u64 start_row;
 	u8  span_arm;	/* span[7:5], arm[4:0] */
 	u8  pd_after_lb;
+	bool ra_capable;
 };
 
 struct MR_LD_TARGET_SYNC {
@@ -930,6 +1016,30 @@ struct MR_PD_CFG_SEQ_NUM_SYNC {
 	struct MR_PD_CFG_SEQ seq[1];
 } __packed;
 
+/* stream detection */
+struct STREAM_DETECT {
+	u64 next_seq_lba; /* next LBA to match sequential access */
+	struct megasas_cmd_fusion *first_cmd_fusion; /* first cmd in group */
+	struct megasas_cmd_fusion *last_cmd_fusion; /* last cmd in group */
+	u32 count_cmds_in_stream; /* count of host commands in this stream */
+	u16 num_sges_in_group; /* total number of SGEs in grouped IOs */
+	u8 is_read; /* SCSI OpCode for this stream */
+	u8 group_depth; /* total number of host commands in group */
+	/* TRUE if cannot add any more commands to this group */
+	bool group_flush;
+	u8 reserved[7]; /* pad to 64-bit alignment */
+};
+
+struct LD_STREAM_DETECT {
+	bool write_back; /* TRUE if WB, FALSE if WT */
+	bool fp_write_enabled;
+	bool members_ssds;
+	bool fp_cache_bypass_capable;
+	u32 mru_bit_map; /* bitmap used to track MRU and LRU stream indicies */
+	/* this is the array of stream detect structures (one per stream) */
+	struct STREAM_DETECT stream_track[MAX_STREAMS_TRACKED];
+};
+
 struct MPI2_IOC_INIT_RDPQ_ARRAY_ENTRY {
 	u64 RDPQBaseAddress;
 	u32 Reserved1;
@@ -983,6 +1093,7 @@ struct fusion_context {
 	struct LD_LOAD_BALANCE_INFO load_balance_info[MAX_LOGICAL_DRIVES_EXT];
 	LD_SPAN_INFO log_to_span[MAX_LOGICAL_DRIVES_EXT];
 	u8 adapter_type;
+	struct LD_STREAM_DETECT **stream_detect_by_ld;
 };
 
 union desc_value {