RDMA/cxgb4: Remove kfifo usage

Using kfifos for ID management was limiting the number of QPs and
preventing NP384 MPI jobs.  So replace it with a simple bitmap
allocator.

Remove IDs from the IDR tables before deallocating them.  This bug was
causing the BUG_ON() in insert_handle() to fire because the ID was
getting reused before being removed from the IDR table.

Signed-off-by: Vipul Pandya <vipul@chelsio.com>
Signed-off-by: Steve Wise <swise@opengridcomputing.com>
Signed-off-by: Roland Dreier <roland@purestorage.com>
diff --git a/drivers/infiniband/hw/cxgb4/resource.c b/drivers/infiniband/hw/cxgb4/resource.c
index 1b948d1..cdef4d7 100644
--- a/drivers/infiniband/hw/cxgb4/resource.c
+++ b/drivers/infiniband/hw/cxgb4/resource.c
@@ -30,96 +30,25 @@
  * SOFTWARE.
  */
 /* Crude resource management */
-#include <linux/kernel.h>
-#include <linux/random.h>
-#include <linux/slab.h>
-#include <linux/kfifo.h>
 #include <linux/spinlock.h>
-#include <linux/errno.h>
 #include <linux/genalloc.h>
 #include <linux/ratelimit.h>
 #include "iw_cxgb4.h"
 
-#define RANDOM_SIZE 16
-
-static int __c4iw_init_resource_fifo(struct kfifo *fifo,
-				   spinlock_t *fifo_lock,
-				   u32 nr, u32 skip_low,
-				   u32 skip_high,
-				   int random)
-{
-	u32 i, j, entry = 0, idx;
-	u32 random_bytes;
-	u32 rarray[16];
-	spin_lock_init(fifo_lock);
-
-	if (kfifo_alloc(fifo, nr * sizeof(u32), GFP_KERNEL))
-		return -ENOMEM;
-
-	for (i = 0; i < skip_low + skip_high; i++)
-		kfifo_in(fifo, (unsigned char *) &entry, sizeof(u32));
-	if (random) {
-		j = 0;
-		random_bytes = random32();
-		for (i = 0; i < RANDOM_SIZE; i++)
-			rarray[i] = i + skip_low;
-		for (i = skip_low + RANDOM_SIZE; i < nr - skip_high; i++) {
-			if (j >= RANDOM_SIZE) {
-				j = 0;
-				random_bytes = random32();
-			}
-			idx = (random_bytes >> (j * 2)) & 0xF;
-			kfifo_in(fifo,
-				(unsigned char *) &rarray[idx],
-				sizeof(u32));
-			rarray[idx] = i;
-			j++;
-		}
-		for (i = 0; i < RANDOM_SIZE; i++)
-			kfifo_in(fifo,
-				(unsigned char *) &rarray[i],
-				sizeof(u32));
-	} else
-		for (i = skip_low; i < nr - skip_high; i++)
-			kfifo_in(fifo, (unsigned char *) &i, sizeof(u32));
-
-	for (i = 0; i < skip_low + skip_high; i++)
-		if (kfifo_out_locked(fifo, (unsigned char *) &entry,
-				     sizeof(u32), fifo_lock))
-			break;
-	return 0;
-}
-
-static int c4iw_init_resource_fifo(struct kfifo *fifo, spinlock_t * fifo_lock,
-				   u32 nr, u32 skip_low, u32 skip_high)
-{
-	return __c4iw_init_resource_fifo(fifo, fifo_lock, nr, skip_low,
-					  skip_high, 0);
-}
-
-static int c4iw_init_resource_fifo_random(struct kfifo *fifo,
-				   spinlock_t *fifo_lock,
-				   u32 nr, u32 skip_low, u32 skip_high)
-{
-	return __c4iw_init_resource_fifo(fifo, fifo_lock, nr, skip_low,
-					  skip_high, 1);
-}
-
-static int c4iw_init_qid_fifo(struct c4iw_rdev *rdev)
+static int c4iw_init_qid_table(struct c4iw_rdev *rdev)
 {
 	u32 i;
 
-	spin_lock_init(&rdev->resource.qid_fifo_lock);
-
-	if (kfifo_alloc(&rdev->resource.qid_fifo, rdev->lldi.vr->qp.size *
-			sizeof(u32), GFP_KERNEL))
+	if (c4iw_id_table_alloc(&rdev->resource.qid_table,
+				rdev->lldi.vr->qp.start,
+				rdev->lldi.vr->qp.size,
+				rdev->lldi.vr->qp.size, 0))
 		return -ENOMEM;
 
 	for (i = rdev->lldi.vr->qp.start;
-	     i < rdev->lldi.vr->qp.start + rdev->lldi.vr->qp.size; i++)
+		i < rdev->lldi.vr->qp.start + rdev->lldi.vr->qp.size; i++)
 		if (!(i & rdev->qpmask))
-			kfifo_in(&rdev->resource.qid_fifo,
-				    (unsigned char *) &i, sizeof(u32));
+			c4iw_id_free(&rdev->resource.qid_table, i);
 	return 0;
 }
 
@@ -127,44 +56,42 @@
 int c4iw_init_resource(struct c4iw_rdev *rdev, u32 nr_tpt, u32 nr_pdid)
 {
 	int err = 0;
-	err = c4iw_init_resource_fifo_random(&rdev->resource.tpt_fifo,
-					     &rdev->resource.tpt_fifo_lock,
-					     nr_tpt, 1, 0);
+	err = c4iw_id_table_alloc(&rdev->resource.tpt_table, 0, nr_tpt, 1,
+					C4IW_ID_TABLE_F_RANDOM);
 	if (err)
 		goto tpt_err;
-	err = c4iw_init_qid_fifo(rdev);
+	err = c4iw_init_qid_table(rdev);
 	if (err)
 		goto qid_err;
-	err = c4iw_init_resource_fifo(&rdev->resource.pdid_fifo,
-				      &rdev->resource.pdid_fifo_lock,
-				      nr_pdid, 1, 0);
+	err = c4iw_id_table_alloc(&rdev->resource.pdid_table, 0,
+					nr_pdid, 1, 0);
 	if (err)
 		goto pdid_err;
 	return 0;
-pdid_err:
-	kfifo_free(&rdev->resource.qid_fifo);
-qid_err:
-	kfifo_free(&rdev->resource.tpt_fifo);
-tpt_err:
+ pdid_err:
+	c4iw_id_table_free(&rdev->resource.qid_table);
+ qid_err:
+	c4iw_id_table_free(&rdev->resource.tpt_table);
+ tpt_err:
 	return -ENOMEM;
 }
 
 /*
  * returns 0 if no resource available
  */
-u32 c4iw_get_resource(struct kfifo *fifo, spinlock_t *lock)
+u32 c4iw_get_resource(struct c4iw_id_table *id_table)
 {
 	u32 entry;
-	if (kfifo_out_locked(fifo, (unsigned char *) &entry, sizeof(u32), lock))
-		return entry;
-	else
+	entry = c4iw_id_alloc(id_table);
+	if (entry == (u32)(-1))
 		return 0;
+	return entry;
 }
 
-void c4iw_put_resource(struct kfifo *fifo, u32 entry, spinlock_t *lock)
+void c4iw_put_resource(struct c4iw_id_table *id_table, u32 entry)
 {
 	PDBG("%s entry 0x%x\n", __func__, entry);
-	kfifo_in_locked(fifo, (unsigned char *) &entry, sizeof(u32), lock);
+	c4iw_id_free(id_table, entry);
 }
 
 u32 c4iw_get_cqid(struct c4iw_rdev *rdev, struct c4iw_dev_ucontext *uctx)
@@ -181,8 +108,7 @@
 		qid = entry->qid;
 		kfree(entry);
 	} else {
-		qid = c4iw_get_resource(&rdev->resource.qid_fifo,
-					&rdev->resource.qid_fifo_lock);
+		qid = c4iw_get_resource(&rdev->resource.qid_table);
 		if (!qid)
 			goto out;
 		mutex_lock(&rdev->stats.lock);
@@ -252,8 +178,7 @@
 		qid = entry->qid;
 		kfree(entry);
 	} else {
-		qid = c4iw_get_resource(&rdev->resource.qid_fifo,
-					&rdev->resource.qid_fifo_lock);
+		qid = c4iw_get_resource(&rdev->resource.qid_table);
 		if (!qid)
 			goto out;
 		mutex_lock(&rdev->stats.lock);
@@ -311,9 +236,9 @@
 
 void c4iw_destroy_resource(struct c4iw_resource *rscp)
 {
-	kfifo_free(&rscp->tpt_fifo);
-	kfifo_free(&rscp->qid_fifo);
-	kfifo_free(&rscp->pdid_fifo);
+	c4iw_id_table_free(&rscp->tpt_table);
+	c4iw_id_table_free(&rscp->qid_table);
+	c4iw_id_table_free(&rscp->pdid_table);
 }
 
 /*
@@ -326,16 +251,14 @@
 {
 	unsigned long addr = gen_pool_alloc(rdev->pbl_pool, size);
 	PDBG("%s addr 0x%x size %d\n", __func__, (u32)addr, size);
-	if (!addr)
-		printk_ratelimited(KERN_WARNING MOD "%s: Out of PBL memory\n",
-		       pci_name(rdev->lldi.pdev));
+	mutex_lock(&rdev->stats.lock);
 	if (addr) {
-		mutex_lock(&rdev->stats.lock);
 		rdev->stats.pbl.cur += roundup(size, 1 << MIN_PBL_SHIFT);
 		if (rdev->stats.pbl.cur > rdev->stats.pbl.max)
 			rdev->stats.pbl.max = rdev->stats.pbl.cur;
-		mutex_unlock(&rdev->stats.lock);
-	}
+	} else
+		rdev->stats.pbl.fail++;
+	mutex_unlock(&rdev->stats.lock);
 	return (u32)addr;
 }
 
@@ -401,13 +324,14 @@
 	if (!addr)
 		printk_ratelimited(KERN_WARNING MOD "%s: Out of RQT memory\n",
 		       pci_name(rdev->lldi.pdev));
+	mutex_lock(&rdev->stats.lock);
 	if (addr) {
-		mutex_lock(&rdev->stats.lock);
 		rdev->stats.rqt.cur += roundup(size << 6, 1 << MIN_RQT_SHIFT);
 		if (rdev->stats.rqt.cur > rdev->stats.rqt.max)
 			rdev->stats.rqt.max = rdev->stats.rqt.cur;
-		mutex_unlock(&rdev->stats.lock);
-	}
+	} else
+		rdev->stats.rqt.fail++;
+	mutex_unlock(&rdev->stats.lock);
 	return (u32)addr;
 }