s390/scm_block: use mempool to manage aidaw requests

We currently use one preallocated page per HW request to store
aidaws. With this patch we use mempool to allocate an aidaw page
whenever we need it.

Signed-off-by: Sebastian Ott <sebott@linux.vnet.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>

3 files changed, 55 insertions, 12 deletions

diff --git a/drivers/s390/block/scm_blk.c b/drivers/s390/block/scm_blk.c
index 56046ab..5b2abad 100644
--- a/drivers/s390/block/scm_blk.c
+++ b/drivers/s390/block/scm_blk.c
@@ -10,6 +10,7 @@
 
 #include <linux/interrupt.h>
 #include <linux/spinlock.h>
+#include <linux/mempool.h>
 #include <linux/module.h>
 #include <linux/blkdev.h>
 #include <linux/genhd.h>
@@ -20,6 +21,7 @@
 
 debug_info_t *scm_debug;
 static int scm_major;
+static mempool_t *aidaw_pool;
 static DEFINE_SPINLOCK(list_lock);
 static LIST_HEAD(inactive_requests);
 static unsigned int nr_requests = 64;
@@ -36,7 +38,6 @@ static void __scm_free_rq(struct scm_request *scmrq)
 	struct aob_rq_header *aobrq = to_aobrq(scmrq);
 
 	free_page((unsigned long) scmrq->aob);
-	free_page((unsigned long) scmrq->aidaw);
 	__scm_free_rq_cluster(scmrq);
 	kfree(aobrq);
 }
@@ -53,6 +54,8 @@ static void scm_free_rqs(void)
 		__scm_free_rq(scmrq);
 	}
 	spin_unlock_irq(&list_lock);
+
+	mempool_destroy(aidaw_pool);
 }
 
 static int __scm_alloc_rq(void)
@@ -65,9 +68,8 @@ static int __scm_alloc_rq(void)
 		return -ENOMEM;
 
 	scmrq = (void *) aobrq->data;
-	scmrq->aidaw = (void *) get_zeroed_page(GFP_DMA);
 	scmrq->aob = (void *) get_zeroed_page(GFP_DMA);
-	if (!scmrq->aob || !scmrq->aidaw) {
+	if (!scmrq->aob) {
 		__scm_free_rq(scmrq);
 		return -ENOMEM;
 	}
@@ -89,6 +91,10 @@ static int scm_alloc_rqs(unsigned int nrqs)
 {
 	int ret = 0;
 
+	aidaw_pool = mempool_create_page_pool(max(nrqs/8, 1U), 0);
+	if (!aidaw_pool)
+		return -ENOMEM;
+
 	while (nrqs-- && !ret)
 		ret = __scm_alloc_rq();
 
@@ -111,8 +117,13 @@ out:
 
 static void scm_request_done(struct scm_request *scmrq)
 {
+	struct msb *msb = &scmrq->aob->msb[0];
+	u64 aidaw = msb->data_addr;
 	unsigned long flags;
 
+	if ((msb->flags & MSB_FLAG_IDA) && aidaw)
+		mempool_free(virt_to_page(aidaw), aidaw_pool);
+
 	spin_lock_irqsave(&list_lock, flags);
 	list_add(&scmrq->list, &inactive_requests);
 	spin_unlock_irqrestore(&list_lock, flags);
@@ -123,15 +134,26 @@ static bool scm_permit_request(struct scm_blk_dev *bdev, struct request *req)
 	return rq_data_dir(req) != WRITE || bdev->state != SCM_WR_PROHIBIT;
 }
 
-static void scm_request_prepare(struct scm_request *scmrq)
+struct aidaw *scm_aidaw_alloc(void)
+{
+	struct page *page = mempool_alloc(aidaw_pool, GFP_ATOMIC);
+
+	return page ? page_address(page) : NULL;
+}
+
+static int scm_request_prepare(struct scm_request *scmrq)
 {
 	struct scm_blk_dev *bdev = scmrq->bdev;
 	struct scm_device *scmdev = bdev->gendisk->private_data;
-	struct aidaw *aidaw = scmrq->aidaw;
+	struct aidaw *aidaw = scm_aidaw_alloc();
 	struct msb *msb = &scmrq->aob->msb[0];
 	struct req_iterator iter;
 	struct bio_vec bv;
 
+	if (!aidaw)
+		return -ENOMEM;
+
+	memset(aidaw, 0, PAGE_SIZE);
 	msb->bs = MSB_BS_4K;
 	scmrq->aob->request.msb_count = 1;
 	msb->scm_addr = scmdev->address +
@@ -147,6 +169,8 @@ static void scm_request_prepare(struct scm_request *scmrq)
 		aidaw->data_addr = (u64) page_address(bv.bv_page);
 		aidaw++;
 	}
+
+	return 0;
 }
 
 static inline void scm_request_init(struct scm_blk_dev *bdev,
@@ -157,7 +181,6 @@ static inline void scm_request_init(struct scm_blk_dev *bdev,
 	struct aob *aob = scmrq->aob;
 
 	memset(aob, 0, sizeof(*aob));
-	memset(scmrq->aidaw, 0, PAGE_SIZE);
 	aobrq->scmdev = bdev->scmdev;
 	aob->request.cmd_code = ARQB_CMD_MOVE;
 	aob->request.data = (u64) aobrq;
@@ -236,7 +259,15 @@ static void scm_blk_request(struct request_queue *rq)
 			scm_initiate_cluster_request(scmrq);
 			return;
 		}
-		scm_request_prepare(scmrq);
+
+		if (scm_request_prepare(scmrq)) {
+			SCM_LOG(5, "no aidaw");
+			scm_release_cluster(scmrq);
+			scm_request_done(scmrq);
+			scm_ensure_queue_restart(bdev);
+			return;
+		}
+
 		atomic_inc(&bdev->queued_reqs);
 		blk_start_request(req);
 
diff --git a/drivers/s390/block/scm_blk.h b/drivers/s390/block/scm_blk.h
index e59331e..a315ef0 100644
--- a/drivers/s390/block/scm_blk.h
+++ b/drivers/s390/block/scm_blk.h
@@ -31,7 +31,6 @@ struct scm_blk_dev {
 struct scm_request {
 	struct scm_blk_dev *bdev;
 	struct request *request;
-	struct aidaw *aidaw;
 	struct aob *aob;
 	struct list_head list;
 	u8 retries;
@@ -55,6 +54,8 @@ void scm_blk_irq(struct scm_device *, void *, int);
 void scm_request_finish(struct scm_request *);
 void scm_request_requeue(struct scm_request *);
 
+struct aidaw *scm_aidaw_alloc(void);
+
 int scm_drv_init(void);
 void scm_drv_cleanup(void);
 
diff --git a/drivers/s390/block/scm_blk_cluster.c b/drivers/s390/block/scm_blk_cluster.c
index 9aae909..4787f80 100644
--- a/drivers/s390/block/scm_blk_cluster.c
+++ b/drivers/s390/block/scm_blk_cluster.c
@@ -114,14 +114,14 @@ void scm_blk_dev_cluster_setup(struct scm_blk_dev *bdev)
 	blk_queue_io_opt(bdev->rq, CLUSTER_SIZE);
 }
 
-static void scm_prepare_cluster_request(struct scm_request *scmrq)
+static int scm_prepare_cluster_request(struct scm_request *scmrq)
 {
 	struct scm_blk_dev *bdev = scmrq->bdev;
 	struct scm_device *scmdev = bdev->gendisk->private_data;
 	struct request *req = scmrq->request;
-	struct aidaw *aidaw = scmrq->aidaw;
 	struct msb *msb = &scmrq->aob->msb[0];
 	struct req_iterator iter;
+	struct aidaw *aidaw;
 	struct bio_vec bv;
 	int i = 0;
 	u64 addr;
@@ -131,6 +131,11 @@ static void scm_prepare_cluster_request(struct scm_request *scmrq)
 		scmrq->cluster.state = CLUSTER_READ;
 		/* fall through */
 	case CLUSTER_READ:
+		aidaw = scm_aidaw_alloc();
+		if (!aidaw)
+			return -ENOMEM;
+
+		memset(aidaw, 0, PAGE_SIZE);
 		scmrq->aob->request.msb_count = 1;
 		msb->bs = MSB_BS_4K;
 		msb->oc = MSB_OC_READ;
@@ -153,6 +158,7 @@ static void scm_prepare_cluster_request(struct scm_request *scmrq)
 
 		break;
 	case CLUSTER_WRITE:
+		aidaw = (void *) msb->data_addr;
 		msb->oc = MSB_OC_WRITE;
 
 		for (addr = msb->scm_addr;
@@ -173,6 +179,7 @@ static void scm_prepare_cluster_request(struct scm_request *scmrq)
 		}
 		break;
 	}
+	return 0;
 }
 
 bool scm_need_cluster_request(struct scm_request *scmrq)
@@ -186,9 +193,13 @@ bool scm_need_cluster_request(struct scm_request *scmrq)
 /* Called with queue lock held. */
 void scm_initiate_cluster_request(struct scm_request *scmrq)
 {
-	scm_prepare_cluster_request(scmrq);
+	if (scm_prepare_cluster_request(scmrq))
+		goto requeue;
 	if (eadm_start_aob(scmrq->aob))
-		scm_request_requeue(scmrq);
+		goto requeue;
+	return;
+requeue:
+	scm_request_requeue(scmrq);
 }
 
 bool scm_test_cluster_request(struct scm_request *scmrq)