scsi: mpi3mr: Create operational request and reply queue pair
authorKashyap Desai <kashyap.desai@broadcom.com>
Thu, 20 May 2021 15:25:24 +0000 (20:55 +0530)
committerMartin K. Petersen <martin.petersen@oracle.com>
Wed, 2 Jun 2021 04:56:16 +0000 (00:56 -0400)
Create operational request and reply queue pair.

The MPI3 transport interface consists of an Administrative Request Queue,
an Administrative Reply Queue, and Operational Messaging Queues.  The
Operational Messaging Queues are the primary communication mechanism
between the host and the I/O Controller (IOC).  Request messages, allocated
in host memory, identify I/O operations to be performed by the IOC. These
operations are queued on an Operational Request Queue by the host driver.
Reply descriptors track I/O operations as they complete.  The IOC queues
these completions in an Operational Reply Queue.

To fulfil large contiguous memory requirement, driver creates multiple
segments and provide the list of segments. Each segment size should be 4K
which is a hardware requirement. An element array is contiguous or
segmented.  A contiguous element array is located in contiguous physical
memory.  A contiguous element array must be aligned on an element size
boundary.  An element's physical address within the array may be directly
calculated from the base address, the Producer/Consumer index, and the
element size.

Expected phased identifier bit is used to find out valid entry on reply
queue. Driver sets <ephase> bit and IOC inverts the value of this bit on
each pass.

Link: https://lore.kernel.org/r/20210520152545.2710479-4-kashyap.desai@broadcom.com
Cc: sathya.prakash@broadcom.com
Reviewed-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: Tomas Henzl <thenzl@redhat.com>
Reviewed-by: Himanshu Madhani <himanshu.madhani@oracle.com>
Signed-off-by: Kashyap Desai <kashyap.desai@broadcom.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
drivers/scsi/mpi3mr/mpi3mr.h
drivers/scsi/mpi3mr/mpi3mr_fw.c
drivers/scsi/mpi3mr/mpi3mr_os.c

index babb5c5..9582ef2 100644 (file)
@@ -74,6 +74,12 @@ extern struct list_head mrioc_list;
 #define MPI3MR_ADMIN_REQ_FRAME_SZ      128
 #define MPI3MR_ADMIN_REPLY_FRAME_SZ    16
 
+/* Operational queue management definitions */
+#define MPI3MR_OP_REQ_Q_QD             512
+#define MPI3MR_OP_REP_Q_QD             4096
+#define MPI3MR_OP_REQ_Q_SEG_SIZE       4096
+#define MPI3MR_OP_REP_Q_SEG_SIZE       4096
+#define MPI3MR_MAX_SEG_LIST_SIZE       4096
 
 /* Reserved Host Tag definitions */
 #define MPI3MR_HOSTTAG_INVALID         0xFFFF
@@ -135,6 +141,9 @@ extern struct list_head mrioc_list;
        (MPI3_SGE_FLAGS_ELEMENT_TYPE_SIMPLE | MPI3_SGE_FLAGS_DLAS_SYSTEM | \
        MPI3_SGE_FLAGS_END_OF_LIST)
 
+/* MSI Index from Reply Queue Index */
+#define REPLY_QUEUE_IDX_TO_MSIX_IDX(qidx, offset)      (qidx + offset)
+
 /* IOC State definitions */
 enum mpi3mr_iocstate {
        MRIOC_STATE_READY = 1,
@@ -226,14 +235,44 @@ struct mpi3mr_ioc_facts {
 };
 
 /**
+ * struct segments - memory descriptor structure to store
+ * virtual and dma addresses for operational queue segments.
+ *
+ * @segment: virtual address
+ * @segment_dma: dma address
+ */
+struct segments {
+       void *segment;
+       dma_addr_t segment_dma;
+};
+
+/**
  * struct op_req_qinfo -  Operational Request Queue Information
  *
  * @ci: consumer index
  * @pi: producer index
+ * @num_request: Maximum number of entries in the queue
+ * @qid: Queue Id starting from 1
+ * @reply_qid: Associated reply queue Id
+ * @num_segments: Number of discontiguous memory segments
+ * @segment_qd: Depth of each segments
+ * @q_lock: Concurrent queue access lock
+ * @q_segments: Segment descriptor pointer
+ * @q_segment_list: Segment list base virtual address
+ * @q_segment_list_dma: Segment list base DMA address
  */
 struct op_req_qinfo {
        u16 ci;
        u16 pi;
+       u16 num_requests;
+       u16 qid;
+       u16 reply_qid;
+       u16 num_segments;
+       u16 segment_qd;
+       spinlock_t q_lock;
+       struct segments *q_segments;
+       void *q_segment_list;
+       dma_addr_t q_segment_list_dma;
 };
 
 /**
@@ -241,10 +280,24 @@ struct op_req_qinfo {
  *
  * @ci: consumer index
  * @qid: Queue Id starting from 1
+ * @num_replies: Maximum number of entries in the queue
+ * @num_segments: Number of discontiguous memory segments
+ * @segment_qd: Depth of each segments
+ * @q_segments: Segment descriptor pointer
+ * @q_segment_list: Segment list base virtual address
+ * @q_segment_list_dma: Segment list base DMA address
+ * @ephase: Expected phased identifier for the reply queue
  */
 struct op_reply_qinfo {
        u16 ci;
        u16 qid;
+       u16 num_replies;
+       u16 num_segments;
+       u16 segment_qd;
+       struct segments *q_segments;
+       void *q_segment_list;
+       dma_addr_t q_segment_list_dma;
+       u8 ephase;
 };
 
 /**
@@ -404,6 +457,7 @@ struct scmd_priv {
  * @current_event: Firmware event currently in process
  * @driver_info: Driver, Kernel, OS information to firmware
  * @change_count: Topology change count
+ * @op_reply_q_offset: Operational reply queue offset with MSIx
  */
 struct mpi3mr_ioc {
        struct list_head list;
@@ -411,6 +465,7 @@ struct mpi3mr_ioc {
        struct Scsi_Host *shost;
        u8 id;
        int cpu_count;
+       bool enable_segqueue;
 
        char name[MPI3MR_NAME_LENGTH];
        char driver_name[MPI3MR_NAME_LENGTH];
@@ -497,6 +552,7 @@ struct mpi3mr_ioc {
        struct mpi3mr_fwevt *current_event;
        struct mpi3_driver_info_layout driver_info;
        u16 change_count;
+       u16 op_reply_q_offset;
 };
 
 int mpi3mr_setup_resources(struct mpi3mr_ioc *mrioc);
index 40d67b8..353bbf4 100644 (file)
@@ -404,6 +404,7 @@ static int mpi3mr_setup_isr(struct mpi3mr_ioc *mrioc, u8 setup_one)
 
        irq_flags |= PCI_IRQ_AFFINITY | PCI_IRQ_ALL_TYPES;
 
+       mrioc->op_reply_q_offset = (max_vectors > 1) ? 1 : 0;
        i = pci_alloc_irq_vectors_affinity(mrioc->pdev,
            1, max_vectors, irq_flags, &desc);
        if (i <= 0) {
@@ -414,6 +415,12 @@ static int mpi3mr_setup_isr(struct mpi3mr_ioc *mrioc, u8 setup_one)
                ioc_info(mrioc,
                    "allocated vectors (%d) are less than configured (%d)\n",
                    i, max_vectors);
+               /*
+                * If only one MSI-x is allocated, then MSI-x 0 will be shared
+                * between Admin queue and operational queue
+                */
+               if (i == 1)
+                       mrioc->op_reply_q_offset = 0;
 
                max_vectors = i;
        }
@@ -720,6 +727,582 @@ out:
 }
 
 /**
+ * mpi3mr_free_op_req_q_segments - free request memory segments
+ * @mrioc: Adapter instance reference
+ * @q_idx: operational request queue index
+ *
+ * Free memory segments allocated for operational request queue
+ *
+ * Return: Nothing.
+ */
+static void mpi3mr_free_op_req_q_segments(struct mpi3mr_ioc *mrioc, u16 q_idx)
+{
+       u16 j;
+       int size;
+       struct segments *segments;
+
+       segments = mrioc->req_qinfo[q_idx].q_segments;
+       if (!segments)
+               return;
+
+       if (mrioc->enable_segqueue) {
+               size = MPI3MR_OP_REQ_Q_SEG_SIZE;
+               if (mrioc->req_qinfo[q_idx].q_segment_list) {
+                       dma_free_coherent(&mrioc->pdev->dev,
+                           MPI3MR_MAX_SEG_LIST_SIZE,
+                           mrioc->req_qinfo[q_idx].q_segment_list,
+                           mrioc->req_qinfo[q_idx].q_segment_list_dma);
+                       mrioc->op_reply_qinfo[q_idx].q_segment_list = NULL;
+               }
+       } else
+               size = mrioc->req_qinfo[q_idx].num_requests *
+                   mrioc->facts.op_req_sz;
+
+       for (j = 0; j < mrioc->req_qinfo[q_idx].num_segments; j++) {
+               if (!segments[j].segment)
+                       continue;
+               dma_free_coherent(&mrioc->pdev->dev,
+                   size, segments[j].segment, segments[j].segment_dma);
+               segments[j].segment = NULL;
+       }
+       kfree(mrioc->req_qinfo[q_idx].q_segments);
+       mrioc->req_qinfo[q_idx].q_segments = NULL;
+       mrioc->req_qinfo[q_idx].qid = 0;
+}
+
+/**
+ * mpi3mr_free_op_reply_q_segments - free reply memory segments
+ * @mrioc: Adapter instance reference
+ * @q_idx: operational reply queue index
+ *
+ * Free memory segments allocated for operational reply queue
+ *
+ * Return: Nothing.
+ */
+static void mpi3mr_free_op_reply_q_segments(struct mpi3mr_ioc *mrioc, u16 q_idx)
+{
+       u16 j;
+       int size;
+       struct segments *segments;
+
+       segments = mrioc->op_reply_qinfo[q_idx].q_segments;
+       if (!segments)
+               return;
+
+       if (mrioc->enable_segqueue) {
+               size = MPI3MR_OP_REP_Q_SEG_SIZE;
+               if (mrioc->op_reply_qinfo[q_idx].q_segment_list) {
+                       dma_free_coherent(&mrioc->pdev->dev,
+                           MPI3MR_MAX_SEG_LIST_SIZE,
+                           mrioc->op_reply_qinfo[q_idx].q_segment_list,
+                           mrioc->op_reply_qinfo[q_idx].q_segment_list_dma);
+                       mrioc->op_reply_qinfo[q_idx].q_segment_list = NULL;
+               }
+       } else
+               size = mrioc->op_reply_qinfo[q_idx].segment_qd *
+                   mrioc->op_reply_desc_sz;
+
+       for (j = 0; j < mrioc->op_reply_qinfo[q_idx].num_segments; j++) {
+               if (!segments[j].segment)
+                       continue;
+               dma_free_coherent(&mrioc->pdev->dev,
+                   size, segments[j].segment, segments[j].segment_dma);
+               segments[j].segment = NULL;
+       }
+
+       kfree(mrioc->op_reply_qinfo[q_idx].q_segments);
+       mrioc->op_reply_qinfo[q_idx].q_segments = NULL;
+       mrioc->op_reply_qinfo[q_idx].qid = 0;
+}
+
+/**
+ * mpi3mr_delete_op_reply_q - delete operational reply queue
+ * @mrioc: Adapter instance reference
+ * @qidx: operational reply queue index
+ *
+ * Delete operatinal reply queue by issuing MPI request
+ * through admin queue.
+ *
+ * Return:  0 on success, non-zero on failure.
+ */
+static int mpi3mr_delete_op_reply_q(struct mpi3mr_ioc *mrioc, u16 qidx)
+{
+       struct mpi3_delete_reply_queue_request delq_req;
+       int retval = 0;
+       u16 reply_qid = 0, midx;
+
+       reply_qid = mrioc->op_reply_qinfo[qidx].qid;
+
+       midx = REPLY_QUEUE_IDX_TO_MSIX_IDX(qidx, mrioc->op_reply_q_offset);
+
+       if (!reply_qid) {
+               retval = -1;
+               ioc_err(mrioc, "Issue DelRepQ: called with invalid ReqQID\n");
+               goto out;
+       }
+
+       memset(&delq_req, 0, sizeof(delq_req));
+       mutex_lock(&mrioc->init_cmds.mutex);
+       if (mrioc->init_cmds.state & MPI3MR_CMD_PENDING) {
+               retval = -1;
+               ioc_err(mrioc, "Issue DelRepQ: Init command is in use\n");
+               mutex_unlock(&mrioc->init_cmds.mutex);
+               goto out;
+       }
+       mrioc->init_cmds.state = MPI3MR_CMD_PENDING;
+       mrioc->init_cmds.is_waiting = 1;
+       mrioc->init_cmds.callback = NULL;
+       delq_req.host_tag = cpu_to_le16(MPI3MR_HOSTTAG_INITCMDS);
+       delq_req.function = MPI3_FUNCTION_DELETE_REPLY_QUEUE;
+       delq_req.queue_id = cpu_to_le16(reply_qid);
+
+       init_completion(&mrioc->init_cmds.done);
+       retval = mpi3mr_admin_request_post(mrioc, &delq_req, sizeof(delq_req),
+           1);
+       if (retval) {
+               ioc_err(mrioc, "Issue DelRepQ: Admin Post failed\n");
+               goto out_unlock;
+       }
+       wait_for_completion_timeout(&mrioc->init_cmds.done,
+           (MPI3MR_INTADMCMD_TIMEOUT * HZ));
+       if (!(mrioc->init_cmds.state & MPI3MR_CMD_COMPLETE)) {
+               ioc_err(mrioc, "Issue DelRepQ: command timed out\n");
+               mpi3mr_set_diagsave(mrioc);
+               mpi3mr_issue_reset(mrioc,
+                   MPI3_SYSIF_HOST_DIAG_RESET_ACTION_DIAG_FAULT,
+                   MPI3MR_RESET_FROM_DELREPQ_TIMEOUT);
+               mrioc->unrecoverable = 1;
+
+               retval = -1;
+               goto out_unlock;
+       }
+       if ((mrioc->init_cmds.ioc_status & MPI3_IOCSTATUS_STATUS_MASK)
+           != MPI3_IOCSTATUS_SUCCESS) {
+               ioc_err(mrioc,
+                   "Issue DelRepQ: Failed ioc_status(0x%04x) Loginfo(0x%08x)\n",
+                   (mrioc->init_cmds.ioc_status & MPI3_IOCSTATUS_STATUS_MASK),
+                   mrioc->init_cmds.ioc_loginfo);
+               retval = -1;
+               goto out_unlock;
+       }
+       mrioc->intr_info[midx].op_reply_q = NULL;
+
+       mpi3mr_free_op_reply_q_segments(mrioc, qidx);
+out_unlock:
+       mrioc->init_cmds.state = MPI3MR_CMD_NOTUSED;
+       mutex_unlock(&mrioc->init_cmds.mutex);
+out:
+
+       return retval;
+}
+
+/**
+ * mpi3mr_alloc_op_reply_q_segments -Alloc segmented reply pool
+ * @mrioc: Adapter instance reference
+ * @qidx: request queue index
+ *
+ * Allocate segmented memory pools for operational reply
+ * queue.
+ *
+ * Return: 0 on success, non-zero on failure.
+ */
+static int mpi3mr_alloc_op_reply_q_segments(struct mpi3mr_ioc *mrioc, u16 qidx)
+{
+       struct op_reply_qinfo *op_reply_q = mrioc->op_reply_qinfo + qidx;
+       int i, size;
+       u64 *q_segment_list_entry = NULL;
+       struct segments *segments;
+
+       if (mrioc->enable_segqueue) {
+               op_reply_q->segment_qd =
+                   MPI3MR_OP_REP_Q_SEG_SIZE / mrioc->op_reply_desc_sz;
+
+               size = MPI3MR_OP_REP_Q_SEG_SIZE;
+
+               op_reply_q->q_segment_list = dma_alloc_coherent(&mrioc->pdev->dev,
+                   MPI3MR_MAX_SEG_LIST_SIZE, &op_reply_q->q_segment_list_dma,
+                   GFP_KERNEL);
+               if (!op_reply_q->q_segment_list)
+                       return -ENOMEM;
+               q_segment_list_entry = (u64 *)op_reply_q->q_segment_list;
+       } else {
+               op_reply_q->segment_qd = op_reply_q->num_replies;
+               size = op_reply_q->num_replies * mrioc->op_reply_desc_sz;
+       }
+
+       op_reply_q->num_segments = DIV_ROUND_UP(op_reply_q->num_replies,
+           op_reply_q->segment_qd);
+
+       op_reply_q->q_segments = kcalloc(op_reply_q->num_segments,
+           sizeof(struct segments), GFP_KERNEL);
+       if (!op_reply_q->q_segments)
+               return -ENOMEM;
+
+       segments = op_reply_q->q_segments;
+       for (i = 0; i < op_reply_q->num_segments; i++) {
+               segments[i].segment =
+                   dma_alloc_coherent(&mrioc->pdev->dev,
+                   size, &segments[i].segment_dma, GFP_KERNEL);
+               if (!segments[i].segment)
+                       return -ENOMEM;
+               if (mrioc->enable_segqueue)
+                       q_segment_list_entry[i] =
+                           (unsigned long)segments[i].segment_dma;
+       }
+
+       return 0;
+}
+
+/**
+ * mpi3mr_alloc_op_req_q_segments - Alloc segmented req pool.
+ * @mrioc: Adapter instance reference
+ * @qidx: request queue index
+ *
+ * Allocate segmented memory pools for operational request
+ * queue.
+ *
+ * Return: 0 on success, non-zero on failure.
+ */
+static int mpi3mr_alloc_op_req_q_segments(struct mpi3mr_ioc *mrioc, u16 qidx)
+{
+       struct op_req_qinfo *op_req_q = mrioc->req_qinfo + qidx;
+       int i, size;
+       u64 *q_segment_list_entry = NULL;
+       struct segments *segments;
+
+       if (mrioc->enable_segqueue) {
+               op_req_q->segment_qd =
+                   MPI3MR_OP_REQ_Q_SEG_SIZE / mrioc->facts.op_req_sz;
+
+               size = MPI3MR_OP_REQ_Q_SEG_SIZE;
+
+               op_req_q->q_segment_list = dma_alloc_coherent(&mrioc->pdev->dev,
+                   MPI3MR_MAX_SEG_LIST_SIZE, &op_req_q->q_segment_list_dma,
+                   GFP_KERNEL);
+               if (!op_req_q->q_segment_list)
+                       return -ENOMEM;
+               q_segment_list_entry = (u64 *)op_req_q->q_segment_list;
+
+       } else {
+               op_req_q->segment_qd = op_req_q->num_requests;
+               size = op_req_q->num_requests * mrioc->facts.op_req_sz;
+       }
+
+       op_req_q->num_segments = DIV_ROUND_UP(op_req_q->num_requests,
+           op_req_q->segment_qd);
+
+       op_req_q->q_segments = kcalloc(op_req_q->num_segments,
+           sizeof(struct segments), GFP_KERNEL);
+       if (!op_req_q->q_segments)
+               return -ENOMEM;
+
+       segments = op_req_q->q_segments;
+       for (i = 0; i < op_req_q->num_segments; i++) {
+               segments[i].segment =
+                   dma_alloc_coherent(&mrioc->pdev->dev,
+                   size, &segments[i].segment_dma, GFP_KERNEL);
+               if (!segments[i].segment)
+                       return -ENOMEM;
+               if (mrioc->enable_segqueue)
+                       q_segment_list_entry[i] =
+                           (unsigned long)segments[i].segment_dma;
+       }
+
+       return 0;
+}
+
+/**
+ * mpi3mr_create_op_reply_q - create operational reply queue
+ * @mrioc: Adapter instance reference
+ * @qidx: operational reply queue index
+ *
+ * Create operatinal reply queue by issuing MPI request
+ * through admin queue.
+ *
+ * Return:  0 on success, non-zero on failure.
+ */
+static int mpi3mr_create_op_reply_q(struct mpi3mr_ioc *mrioc, u16 qidx)
+{
+       struct mpi3_create_reply_queue_request create_req;
+       struct op_reply_qinfo *op_reply_q = mrioc->op_reply_qinfo + qidx;
+       int retval = 0;
+       u16 reply_qid = 0, midx;
+
+       reply_qid = op_reply_q->qid;
+
+       midx = REPLY_QUEUE_IDX_TO_MSIX_IDX(qidx, mrioc->op_reply_q_offset);
+
+       if (reply_qid) {
+               retval = -1;
+               ioc_err(mrioc, "CreateRepQ: called for duplicate qid %d\n",
+                   reply_qid);
+
+               return retval;
+       }
+
+       reply_qid = qidx + 1;
+       op_reply_q->num_replies = MPI3MR_OP_REP_Q_QD;
+       op_reply_q->ci = 0;
+       op_reply_q->ephase = 1;
+
+       if (!op_reply_q->q_segments) {
+               retval = mpi3mr_alloc_op_reply_q_segments(mrioc, qidx);
+               if (retval) {
+                       mpi3mr_free_op_reply_q_segments(mrioc, qidx);
+                       goto out;
+               }
+       }
+
+       memset(&create_req, 0, sizeof(create_req));
+       mutex_lock(&mrioc->init_cmds.mutex);
+       if (mrioc->init_cmds.state & MPI3MR_CMD_PENDING) {
+               retval = -1;
+               ioc_err(mrioc, "CreateRepQ: Init command is in use\n");
+               goto out;
+       }
+       mrioc->init_cmds.state = MPI3MR_CMD_PENDING;
+       mrioc->init_cmds.is_waiting = 1;
+       mrioc->init_cmds.callback = NULL;
+       create_req.host_tag = cpu_to_le16(MPI3MR_HOSTTAG_INITCMDS);
+       create_req.function = MPI3_FUNCTION_CREATE_REPLY_QUEUE;
+       create_req.queue_id = cpu_to_le16(reply_qid);
+       create_req.flags = MPI3_CREATE_REPLY_QUEUE_FLAGS_INT_ENABLE_ENABLE;
+       create_req.msix_index = cpu_to_le16(mrioc->intr_info[midx].msix_index);
+       if (mrioc->enable_segqueue) {
+               create_req.flags |=
+                   MPI3_CREATE_REQUEST_QUEUE_FLAGS_SEGMENTED_SEGMENTED;
+               create_req.base_address = cpu_to_le64(
+                   op_reply_q->q_segment_list_dma);
+       } else
+               create_req.base_address = cpu_to_le64(
+                   op_reply_q->q_segments[0].segment_dma);
+
+       create_req.size = cpu_to_le16(op_reply_q->num_replies);
+
+       init_completion(&mrioc->init_cmds.done);
+       retval = mpi3mr_admin_request_post(mrioc, &create_req,
+           sizeof(create_req), 1);
+       if (retval) {
+               ioc_err(mrioc, "CreateRepQ: Admin Post failed\n");
+               goto out_unlock;
+       }
+       wait_for_completion_timeout(&mrioc->init_cmds.done,
+           (MPI3MR_INTADMCMD_TIMEOUT * HZ));
+       if (!(mrioc->init_cmds.state & MPI3MR_CMD_COMPLETE)) {
+               ioc_err(mrioc, "CreateRepQ: command timed out\n");
+               mpi3mr_set_diagsave(mrioc);
+               mpi3mr_issue_reset(mrioc,
+                   MPI3_SYSIF_HOST_DIAG_RESET_ACTION_DIAG_FAULT,
+                   MPI3MR_RESET_FROM_CREATEREPQ_TIMEOUT);
+               mrioc->unrecoverable = 1;
+               retval = -1;
+               goto out_unlock;
+       }
+       if ((mrioc->init_cmds.ioc_status & MPI3_IOCSTATUS_STATUS_MASK)
+           != MPI3_IOCSTATUS_SUCCESS) {
+               ioc_err(mrioc,
+                   "CreateRepQ: Failed ioc_status(0x%04x) Loginfo(0x%08x)\n",
+                   (mrioc->init_cmds.ioc_status & MPI3_IOCSTATUS_STATUS_MASK),
+                   mrioc->init_cmds.ioc_loginfo);
+               retval = -1;
+               goto out_unlock;
+       }
+       op_reply_q->qid = reply_qid;
+       mrioc->intr_info[midx].op_reply_q = op_reply_q;
+
+out_unlock:
+       mrioc->init_cmds.state = MPI3MR_CMD_NOTUSED;
+       mutex_unlock(&mrioc->init_cmds.mutex);
+out:
+
+       return retval;
+}
+
+/**
+ * mpi3mr_create_op_req_q - create operational request queue
+ * @mrioc: Adapter instance reference
+ * @idx: operational request queue index
+ * @reply_qid: Reply queue ID
+ *
+ * Create operatinal request queue by issuing MPI request
+ * through admin queue.
+ *
+ * Return:  0 on success, non-zero on failure.
+ */
+static int mpi3mr_create_op_req_q(struct mpi3mr_ioc *mrioc, u16 idx,
+       u16 reply_qid)
+{
+       struct mpi3_create_request_queue_request create_req;
+       struct op_req_qinfo *op_req_q = mrioc->req_qinfo + idx;
+       int retval = 0;
+       u16 req_qid = 0;
+
+       req_qid = op_req_q->qid;
+
+       if (req_qid) {
+               retval = -1;
+               ioc_err(mrioc, "CreateReqQ: called for duplicate qid %d\n",
+                   req_qid);
+
+               return retval;
+       }
+       req_qid = idx + 1;
+
+       op_req_q->num_requests = MPI3MR_OP_REQ_Q_QD;
+       op_req_q->ci = 0;
+       op_req_q->pi = 0;
+       op_req_q->reply_qid = reply_qid;
+       spin_lock_init(&op_req_q->q_lock);
+
+       if (!op_req_q->q_segments) {
+               retval = mpi3mr_alloc_op_req_q_segments(mrioc, idx);
+               if (retval) {
+                       mpi3mr_free_op_req_q_segments(mrioc, idx);
+                       goto out;
+               }
+       }
+
+       memset(&create_req, 0, sizeof(create_req));
+       mutex_lock(&mrioc->init_cmds.mutex);
+       if (mrioc->init_cmds.state & MPI3MR_CMD_PENDING) {
+               retval = -1;
+               ioc_err(mrioc, "CreateReqQ: Init command is in use\n");
+               goto out;
+       }
+       mrioc->init_cmds.state = MPI3MR_CMD_PENDING;
+       mrioc->init_cmds.is_waiting = 1;
+       mrioc->init_cmds.callback = NULL;
+       create_req.host_tag = cpu_to_le16(MPI3MR_HOSTTAG_INITCMDS);
+       create_req.function = MPI3_FUNCTION_CREATE_REQUEST_QUEUE;
+       create_req.queue_id = cpu_to_le16(req_qid);
+       if (mrioc->enable_segqueue) {
+               create_req.flags =
+                   MPI3_CREATE_REQUEST_QUEUE_FLAGS_SEGMENTED_SEGMENTED;
+               create_req.base_address = cpu_to_le64(
+                   op_req_q->q_segment_list_dma);
+       } else
+               create_req.base_address = cpu_to_le64(
+                   op_req_q->q_segments[0].segment_dma);
+       create_req.reply_queue_id = cpu_to_le16(reply_qid);
+       create_req.size = cpu_to_le16(op_req_q->num_requests);
+
+       init_completion(&mrioc->init_cmds.done);
+       retval = mpi3mr_admin_request_post(mrioc, &create_req,
+           sizeof(create_req), 1);
+       if (retval) {
+               ioc_err(mrioc, "CreateReqQ: Admin Post failed\n");
+               goto out_unlock;
+       }
+       wait_for_completion_timeout(&mrioc->init_cmds.done,
+           (MPI3MR_INTADMCMD_TIMEOUT * HZ));
+       if (!(mrioc->init_cmds.state & MPI3MR_CMD_COMPLETE)) {
+               ioc_err(mrioc, "CreateReqQ: command timed out\n");
+               mpi3mr_set_diagsave(mrioc);
+               if (mpi3mr_issue_reset(mrioc,
+                   MPI3_SYSIF_HOST_DIAG_RESET_ACTION_DIAG_FAULT,
+                   MPI3MR_RESET_FROM_CREATEREQQ_TIMEOUT))
+                       mrioc->unrecoverable = 1;
+               retval = -1;
+               goto out_unlock;
+       }
+       if ((mrioc->init_cmds.ioc_status & MPI3_IOCSTATUS_STATUS_MASK)
+           != MPI3_IOCSTATUS_SUCCESS) {
+               ioc_err(mrioc,
+                   "CreateReqQ: Failed ioc_status(0x%04x) Loginfo(0x%08x)\n",
+                   (mrioc->init_cmds.ioc_status & MPI3_IOCSTATUS_STATUS_MASK),
+                   mrioc->init_cmds.ioc_loginfo);
+               retval = -1;
+               goto out_unlock;
+       }
+       op_req_q->qid = req_qid;
+
+out_unlock:
+       mrioc->init_cmds.state = MPI3MR_CMD_NOTUSED;
+       mutex_unlock(&mrioc->init_cmds.mutex);
+out:
+
+       return retval;
+}
+
+/**
+ * mpi3mr_create_op_queues - create operational queue pairs
+ * @mrioc: Adapter instance reference
+ *
+ * Allocate memory for operational queue meta data and call
+ * create request and reply queue functions.
+ *
+ * Return: 0 on success, non-zero on failures.
+ */
+static int mpi3mr_create_op_queues(struct mpi3mr_ioc *mrioc)
+{
+       int retval = 0;
+       u16 num_queues = 0, i = 0, msix_count_op_q = 1;
+
+       num_queues = min_t(int, mrioc->facts.max_op_reply_q,
+           mrioc->facts.max_op_req_q);
+
+       msix_count_op_q =
+           mrioc->intr_info_count - mrioc->op_reply_q_offset;
+       if (!mrioc->num_queues)
+               mrioc->num_queues = min_t(int, num_queues, msix_count_op_q);
+       num_queues = mrioc->num_queues;
+       ioc_info(mrioc, "Trying to create %d Operational Q pairs\n",
+           num_queues);
+
+       if (!mrioc->req_qinfo) {
+               mrioc->req_qinfo = kcalloc(num_queues,
+                   sizeof(struct op_req_qinfo), GFP_KERNEL);
+               if (!mrioc->req_qinfo) {
+                       retval = -1;
+                       goto out_failed;
+               }
+
+               mrioc->op_reply_qinfo = kzalloc(sizeof(struct op_reply_qinfo) *
+                   num_queues, GFP_KERNEL);
+               if (!mrioc->op_reply_qinfo) {
+                       retval = -1;
+                       goto out_failed;
+               }
+       }
+
+       if (mrioc->enable_segqueue)
+               ioc_info(mrioc,
+                   "allocating operational queues through segmented queues\n");
+
+       for (i = 0; i < num_queues; i++) {
+               if (mpi3mr_create_op_reply_q(mrioc, i)) {
+                       ioc_err(mrioc, "Cannot create OP RepQ %d\n", i);
+                       break;
+               }
+               if (mpi3mr_create_op_req_q(mrioc, i,
+                   mrioc->op_reply_qinfo[i].qid)) {
+                       ioc_err(mrioc, "Cannot create OP ReqQ %d\n", i);
+                       mpi3mr_delete_op_reply_q(mrioc, i);
+                       break;
+               }
+       }
+
+       if (i == 0) {
+               /* Not even one queue is created successfully*/
+               retval = -1;
+               goto out_failed;
+       }
+       mrioc->num_op_reply_q = mrioc->num_op_req_q = i;
+       ioc_info(mrioc, "Successfully created %d Operational Q pairs\n",
+           mrioc->num_op_reply_q);
+
+       return retval;
+out_failed:
+       kfree(mrioc->req_qinfo);
+       mrioc->req_qinfo = NULL;
+
+       kfree(mrioc->op_reply_qinfo);
+       mrioc->op_reply_qinfo = NULL;
+
+       return retval;
+}
+
+/**
  * mpi3mr_setup_admin_qpair - Setup admin queue pair
  * @mrioc: Adapter instance reference
  *
@@ -1589,6 +2172,13 @@ int mpi3mr_init_ioc(struct mpi3mr_ioc *mrioc)
                goto out_failed;
        }
 
+       retval = mpi3mr_create_op_queues(mrioc);
+       if (retval) {
+               ioc_err(mrioc, "Failed to create OpQueues error %d\n",
+                   retval);
+               goto out_failed;
+       }
+
        return retval;
 
 out_failed:
@@ -1644,6 +2234,12 @@ static void mpi3mr_free_mem(struct mpi3mr_ioc *mrioc)
                mrioc->reply_free_q_pool = NULL;
        }
 
+       for (i = 0; i < mrioc->num_op_req_q; i++)
+               mpi3mr_free_op_req_q_segments(mrioc, i);
+
+       for (i = 0; i < mrioc->num_op_reply_q; i++)
+               mpi3mr_free_op_reply_q_segments(mrioc, i);
+
        for (i = 0; i < mrioc->intr_info_count; i++) {
                intr_info = mrioc->intr_info + i;
                if (intr_info)
index 3751750..bda5312 100644 (file)
@@ -40,7 +40,7 @@ static int mpi3mr_map_queues(struct Scsi_Host *shost)
        struct mpi3mr_ioc *mrioc = shost_priv(shost);
 
        return blk_mq_pci_map_queues(&shost->tag_set.map[HCTX_TYPE_DEFAULT],
-           mrioc->pdev, 0);
+           mrioc->pdev, mrioc->op_reply_q_offset);
 }
 
 /**
@@ -218,6 +218,8 @@ mpi3mr_probe(struct pci_dev *pdev, const struct pci_device_id *id)
        spin_lock_init(&mrioc->sbq_lock);
 
        mpi3mr_init_drv_cmd(&mrioc->init_cmds, MPI3MR_HOSTTAG_INITCMDS);
+       if (pdev->revision)
+               mrioc->enable_segqueue = true;
 
        mrioc->logging_level = logging_level;
        mrioc->shost = shost;