scsi: mpi3mr: Avoid reply queue full condition

To avoid reply queue full condition, update the driver to check IOCFacts
capabilities for qfull.

Update the operational reply queue's Consumer Index after processing 100
replies. If pending I/Os on a reply queue exceeds a threshold
(reply_queue_depth - 200), then return I/O back to OS to retry.

Also increase default admin reply queue size to 2K.

Signed-off-by: Sumit Saxena <sumit.saxena@broadcom.com>
Signed-off-by: Ranjan Kumar <ranjan.kumar@broadcom.com>
Link: https://lore.kernel.org/r/20250129100850.25430-2-ranjan.kumar@broadcom.com
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>

3 files changed, 63 insertions, 5 deletions

diff --git a/drivers/scsi/mpi3mr/mpi3mr.h b/drivers/scsi/mpi3mr/mpi3mr.h
index 0d72b5f1b69d..9ed20ed581be 100644
--- a/drivers/scsi/mpi3mr/mpi3mr.h
+++ b/drivers/scsi/mpi3mr/mpi3mr.h
@@ -80,13 +80,14 @@ extern atomic64_t event_counter;
 
 /* Admin queue management definitions */
 #define MPI3MR_ADMIN_REQ_Q_SIZE		(2 * MPI3MR_PAGE_SIZE_4K)
-#define MPI3MR_ADMIN_REPLY_Q_SIZE	(4 * MPI3MR_PAGE_SIZE_4K)
+#define MPI3MR_ADMIN_REPLY_Q_SIZE	(8 * MPI3MR_PAGE_SIZE_4K)
 #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		1024
+#define MPI3MR_OP_REP_Q_QD2K		2048
 #define MPI3MR_OP_REP_Q_QD4K		4096
 #define MPI3MR_OP_REQ_Q_SEG_SIZE	4096
 #define MPI3MR_OP_REP_Q_SEG_SIZE	4096
@@ -328,6 +329,7 @@ enum mpi3mr_reset_reason {
 #define MPI3MR_RESET_REASON_OSTYPE_SHIFT	28
 #define MPI3MR_RESET_REASON_IOCNUM_SHIFT	20
 
+
 /* Queue type definitions */
 enum queue_type {
 	MPI3MR_DEFAULT_QUEUE = 0,
@@ -387,6 +389,7 @@ struct mpi3mr_ioc_facts {
 	u16 max_msix_vectors;
 	u8 personality;
 	u8 dma_mask;
+	bool max_req_limit;
 	u8 protocol_flags;
 	u8 sge_mod_mask;
 	u8 sge_mod_value;
@@ -456,6 +459,8 @@ struct op_req_qinfo {
  * @enable_irq_poll: Flag to indicate polling is enabled
  * @in_use: Queue is handled by poll/ISR
  * @qtype: Type of queue (types defined in enum queue_type)
+ * @qfull_watermark: Watermark defined in reply queue to avoid
+ *                    reply queue full
  */
 struct op_reply_qinfo {
 	u16 ci;
@@ -471,6 +476,7 @@ struct op_reply_qinfo {
 	bool enable_irq_poll;
 	atomic_t in_use;
 	enum queue_type qtype;
+	u16 qfull_watermark;
 };
 
 /**
@@ -1153,6 +1159,8 @@ struct scmd_priv {
  * @snapdump_trigger_active: Snapdump trigger active flag
  * @pci_err_recovery: PCI error recovery in progress
  * @block_on_pci_err: Block IO during PCI error recovery
+ * @reply_qfull_count: Occurences of reply queue full avoidance kicking-in
+ * @prevent_reply_qfull: Enable reply queue prevention
  */
 struct mpi3mr_ioc {
 	struct list_head list;
@@ -1351,6 +1359,8 @@ struct mpi3mr_ioc {
 	bool fw_release_trigger_active;
 	bool pci_err_recovery;
 	bool block_on_pci_err;
+	atomic_t reply_qfull_count;
+	bool prevent_reply_qfull;
 };
 
 /**
diff --git a/drivers/scsi/mpi3mr/mpi3mr_app.c b/drivers/scsi/mpi3mr/mpi3mr_app.c
index 7589f48aebc8..1532436f0f3a 100644
--- a/drivers/scsi/mpi3mr/mpi3mr_app.c
+++ b/drivers/scsi/mpi3mr/mpi3mr_app.c
@@ -3061,6 +3061,29 @@ reply_queue_count_show(struct device *dev, struct device_attribute *attr,
 static DEVICE_ATTR_RO(reply_queue_count);
 
 /**
+ * reply_qfull_count_show - Show reply qfull count
+ * @dev: class device
+ * @attr: Device attributes
+ * @buf: Buffer to copy
+ *
+ * Retrieves the current value of the reply_qfull_count from the mrioc structure and
+ * formats it as a string for display.
+ *
+ * Return: sysfs_emit() return
+ */
+static ssize_t
+reply_qfull_count_show(struct device *dev, struct device_attribute *attr,
+			char *buf)
+{
+	struct Scsi_Host *shost = class_to_shost(dev);
+	struct mpi3mr_ioc *mrioc = shost_priv(shost);
+
+	return sysfs_emit(buf, "%u\n", atomic_read(&mrioc->reply_qfull_count));
+}
+
+static DEVICE_ATTR_RO(reply_qfull_count);
+
+/**
  * logging_level_show - Show controller debug level
  * @dev: class device
  * @attr: Device attributes
@@ -3152,6 +3175,7 @@ static struct attribute *mpi3mr_host_attrs[] = {
 	&dev_attr_fw_queue_depth.attr,
 	&dev_attr_op_req_q_count.attr,
 	&dev_attr_reply_queue_count.attr,
+	&dev_attr_reply_qfull_count.attr,
 	&dev_attr_logging_level.attr,
 	&dev_attr_adp_state.attr,
 	NULL,
diff --git a/drivers/scsi/mpi3mr/mpi3mr_fw.c b/drivers/scsi/mpi3mr/mpi3mr_fw.c
index 5ed31fe57474..656108dd2ee3 100644
--- a/drivers/scsi/mpi3mr/mpi3mr_fw.c
+++ b/drivers/scsi/mpi3mr/mpi3mr_fw.c
@@ -2104,15 +2104,22 @@ static int mpi3mr_create_op_reply_q(struct mpi3mr_ioc *mrioc, u16 qidx)
 	}
 
 	reply_qid = qidx + 1;
-	op_reply_q->num_replies = MPI3MR_OP_REP_Q_QD;
-	if ((mrioc->pdev->device == MPI3_MFGPAGE_DEVID_SAS4116) &&
-		!mrioc->pdev->revision)
-		op_reply_q->num_replies = MPI3MR_OP_REP_Q_QD4K;
+
+	if (mrioc->pdev->device == MPI3_MFGPAGE_DEVID_SAS4116) {
+		if (mrioc->pdev->revision)
+			op_reply_q->num_replies = MPI3MR_OP_REP_Q_QD;
+		else
+			op_reply_q->num_replies = MPI3MR_OP_REP_Q_QD4K;
+	} else
+		op_reply_q->num_replies = MPI3MR_OP_REP_Q_QD2K;
+
 	op_reply_q->ci = 0;
 	op_reply_q->ephase = 1;
 	atomic_set(&op_reply_q->pend_ios, 0);
 	atomic_set(&op_reply_q->in_use, 0);
 	op_reply_q->enable_irq_poll = false;
+	op_reply_q->qfull_watermark =
+		op_reply_q->num_replies - (MPI3MR_THRESHOLD_REPLY_COUNT * 2);
 
 	if (!op_reply_q->q_segments) {
 		retval = mpi3mr_alloc_op_reply_q_segments(mrioc, qidx);
@@ -2416,8 +2423,10 @@ int mpi3mr_op_request_post(struct mpi3mr_ioc *mrioc,
 	void *segment_base_addr;
 	u16 req_sz = mrioc->facts.op_req_sz;
 	struct segments *segments = op_req_q->q_segments;
+	struct op_reply_qinfo *op_reply_q = NULL;
 
 	reply_qidx = op_req_q->reply_qid - 1;
+	op_reply_q = mrioc->op_reply_qinfo + reply_qidx;
 
 	if (mrioc->unrecoverable)
 		return -EFAULT;
@@ -2448,6 +2457,15 @@ int mpi3mr_op_request_post(struct mpi3mr_ioc *mrioc,
 		goto out;
 	}
 
+	/* Reply queue is nearing to get full, push back IOs to SML */
+	if ((mrioc->prevent_reply_qfull == true) &&
+		(atomic_read(&op_reply_q->pend_ios) >
+	     (op_reply_q->qfull_watermark))) {
+		atomic_inc(&mrioc->reply_qfull_count);
+		retval = -EAGAIN;
+		goto out;
+	}
+
 	segment_base_addr = segments[pi / op_req_q->segment_qd].segment;
 	req_entry = (u8 *)segment_base_addr +
 	    ((pi % op_req_q->segment_qd) * req_sz);
@@ -3091,6 +3109,9 @@ static void mpi3mr_process_factsdata(struct mpi3mr_ioc *mrioc,
 	mrioc->facts.dma_mask = (facts_flags &
 	    MPI3_IOCFACTS_FLAGS_DMA_ADDRESS_WIDTH_MASK) >>
 	    MPI3_IOCFACTS_FLAGS_DMA_ADDRESS_WIDTH_SHIFT;
+	mrioc->facts.dma_mask = (facts_flags &
+	    MPI3_IOCFACTS_FLAGS_DMA_ADDRESS_WIDTH_MASK) >>
+	    MPI3_IOCFACTS_FLAGS_DMA_ADDRESS_WIDTH_SHIFT;
 	mrioc->facts.protocol_flags = facts_data->protocol_flags;
 	mrioc->facts.mpi_version = le32_to_cpu(facts_data->mpi_version.word);
 	mrioc->facts.max_reqs = le16_to_cpu(facts_data->max_outstanding_requests);
@@ -4214,6 +4235,9 @@ retry_init:
 		mrioc->shost->transportt = mpi3mr_transport_template;
 	}
 
+	if (mrioc->facts.max_req_limit)
+		mrioc->prevent_reply_qfull = true;
+
 	mrioc->reply_sz = mrioc->facts.reply_sz;
 
 	retval = mpi3mr_check_reset_dma_mask(mrioc);