From: Shukun Tan <tanshukun1@huawei.com>
Date: Thu, 5 Mar 2020 10:06:21 +0800
Subject: crypto: hisilicon - Use one workqueue per qm instead of per qp
Git-commit: 57ca81245f4db4a0222d545f8f5d4709544c26cf
Patch-mainline: v5.7-rc1
References: jsc#SLE-16507 jsc#SLE-15835

Since SEC need not so many workqueues as our test, we just use
one workqueue created by the device driver of QM if necessary,
which will also reduce CPU waste without any throughput decreasing.

Signed-off-by: Shukun Tan <tanshukun1@huawei.com>
Signed-off-by: Zaibo Xu <xuzaibo@huawei.com>
Reviewed-by: Jonathan Cameron <Jonathan.Cameron@huawei.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
Signed-off-by: Matthias Brugger <mbrugger@suse.com>
---
 drivers/crypto/hisilicon/qm.c |   39 ++++++++++++++++-----------------------
 drivers/crypto/hisilicon/qm.h |    5 +++--
 2 files changed, 19 insertions(+), 25 deletions(-)

--- a/drivers/crypto/hisilicon/qm.c
+++ b/drivers/crypto/hisilicon/qm.c
@@ -486,17 +486,9 @@ static void qm_poll_qp(struct hisi_qp *q
 	}
 }
 
-static void qm_qp_work_func(struct work_struct *work)
+static void qm_work_process(struct work_struct *work)
 {
-	struct hisi_qp *qp;
-
-	qp = container_of(work, struct hisi_qp, work);
-	qm_poll_qp(qp, qp->qm);
-}
-
-static irqreturn_t qm_irq_handler(int irq, void *data)
-{
-	struct hisi_qm *qm = data;
+	struct hisi_qm *qm = container_of(work, struct hisi_qm, work);
 	struct qm_eqe *eqe = qm->eqe + qm->status.eq_head;
 	struct hisi_qp *qp;
 	int eqe_num = 0;
@@ -505,7 +497,7 @@ static irqreturn_t qm_irq_handler(int ir
 		eqe_num++;
 		qp = qm_to_hisi_qp(qm, eqe);
 		if (qp)
-			queue_work(qp->wq, &qp->work);
+			qm_poll_qp(qp, qm);
 
 		if (qm->status.eq_head == QM_Q_DEPTH - 1) {
 			qm->status.eqc_phase = !qm->status.eqc_phase;
@@ -523,6 +515,17 @@ static irqreturn_t qm_irq_handler(int ir
 	}
 
 	qm_db(qm, 0, QM_DOORBELL_CMD_EQ, qm->status.eq_head, 0);
+}
+
+static irqreturn_t do_qm_irq(int irq, void *data)
+{
+	struct hisi_qm *qm = (struct hisi_qm *)data;
+
+	/* the workqueue created by device driver of QM */
+	if (qm->wq)
+		queue_work(qm->wq, &qm->work);
+	else
+		schedule_work(&qm->work);
 
 	return IRQ_HANDLED;
 }
@@ -532,7 +535,7 @@ static irqreturn_t qm_irq(int irq, void
 	struct hisi_qm *qm = data;
 
 	if (readl(qm->io_base + QM_VF_EQ_INT_SOURCE))
-		return qm_irq_handler(irq, data);
+		return do_qm_irq(irq, data);
 
 	dev_err(&qm->pdev->dev, "invalid int source\n");
 	qm_db(qm, 0, QM_DOORBELL_CMD_EQ, qm->status.eq_head, 0);
@@ -1151,20 +1154,9 @@ struct hisi_qp *hisi_qm_create_qp(struct
 
 	qp->qp_id = qp_id;
 	qp->alg_type = alg_type;
-	INIT_WORK(&qp->work, qm_qp_work_func);
-	qp->wq = alloc_workqueue("hisi_qm", WQ_UNBOUND | WQ_HIGHPRI |
-				 WQ_CPU_INTENSIVE | WQ_MEM_RECLAIM, 0);
-	if (!qp->wq) {
-		ret = -EFAULT;
-		goto err_free_qp_mem;
-	}
 
 	return qp;
 
-err_free_qp_mem:
-	if (qm->use_dma_api)
-		dma_free_coherent(dev, qp->qdma.size, qp->qdma.va,
-				  qp->qdma.dma);
 err_clear_bit:
 	write_lock(&qm->qps_lock);
 	qm->qp_array[qp_id] = NULL;
@@ -1483,6 +1475,7 @@ int hisi_qm_init(struct hisi_qm *qm)
 	qm->qp_in_used = 0;
 	mutex_init(&qm->mailbox_lock);
 	rwlock_init(&qm->qps_lock);
+	INIT_WORK(&qm->work, qm_work_process);
 
 	dev_dbg(dev, "init qm %s with %s\n", pdev->is_physfn ? "pf" : "vf",
 		qm->use_dma_api ? "dma api" : "iommu api");
--- a/drivers/crypto/hisilicon/qm.h
+++ b/drivers/crypto/hisilicon/qm.h
@@ -180,6 +180,9 @@ struct hisi_qm {
 	u32 error_mask;
 	u32 msi_mask;
 
+	struct workqueue_struct *wq;
+	struct work_struct work;
+
 	bool use_dma_api;
 };
 
@@ -210,8 +213,6 @@ struct hisi_qp {
 	struct hisi_qp_ops *hw_ops;
 	void *qp_ctx;
 	void (*req_cb)(struct hisi_qp *qp, void *data);
-	struct work_struct work;
-	struct workqueue_struct *wq;
 
 	struct hisi_qm *qm;
 };