ibmvnic: Allow multiple ltbs in rxpool ltb_set

Allow multiple LTBs in the rxpool's ltb_set. The first n-1 LTBs will all
be of the same size. The size of the last LTB in the set depends on the
number of buffers and buffer (mtu) size.

Having a set of LTBs per pool provides a couple of benefits.

First, with the current value of IBMVNIC_MAX_LTB_SIZE of 16MB, with an
MTU of 9000, we need a LTB (DMA buffer) of that size but the allocation
can fail in low memory conditions. With a set of LTBs per pool, we can
use several smaller (8MB) LTBs and hopefully have fewer allocation
failures. (See also comments in ibmvnic.h on the trade-off with smaller
LTBs)

Second since the kernel limits the size of the DMA buffer to 16MB (based
on MAX_ORDER), with a single DMA buffer per pool, the pool is also limited
to 16MB. This in turn limits the number of buffers per pool to 1763 when
MTU is 9000. With a set of LTBs per pool, we can have upto the max of 4096
buffers per pool even when MTU is 9000.

Suggested-by: Brian King <brking@linux.ibm.com>
Signed-off-by: Sukadev Bhattiprolu <sukadev@linux.ibm.com>
Signed-off-by: Dany Madden <drt@linux.ibm.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

1 files changed, 134 insertions, 18 deletions

diff --git a/drivers/net/ethernet/ibm/ibmvnic.c b/drivers/net/ethernet/ibm/ibmvnic.c
index 765a48833b3b..16fd1f1f1228 100644
--- a/drivers/net/ethernet/ibm/ibmvnic.c
+++ b/drivers/net/ethernet/ibm/ibmvnic.c
@@ -345,6 +345,14 @@ static void free_long_term_buff(struct ibmvnic_adapter *adapter,
 	ltb->map_id = 0;
 }
 
+/**
+ * free_ltb_set - free the given set of long term buffers (LTBS)
+ * @adapter: The ibmvnic adapter containing this ltb set
+ * @ltb_set: The ltb_set to be freed
+ *
+ * Free the set of LTBs in the given set.
+ */
+
 static void free_ltb_set(struct ibmvnic_adapter *adapter,
 			 struct ibmvnic_ltb_set *ltb_set)
 {
@@ -358,26 +366,117 @@ static void free_ltb_set(struct ibmvnic_adapter *adapter,
 	ltb_set->num_ltbs = 0;
 }
 
+/**
+ * alloc_ltb_set() - Allocate a set of long term buffers (LTBs)
+ *
+ * @adapter: ibmvnic adapter associated to the LTB
+ * @ltb_set: container object for the set of LTBs
+ * @num_buffs: Number of buffers in the LTB
+ * @buff_size: Size of each buffer in the LTB
+ *
+ * Allocate a set of LTBs to accommodate @num_buffs buffers of @buff_size
+ * each. We currently cap size each LTB to IBMVNIC_ONE_LTB_SIZE. If the
+ * new set of LTBs have fewer LTBs than the old set, free the excess LTBs.
+ * If new set needs more than in old set, allocate the remaining ones.
+ * Try and reuse as many LTBs as possible and avoid reallocation.
+ *
+ * Any changes to this allocation strategy must be reflected in
+ * map_rxpool_buff_to_ltb() and map_txpool_buff_to_ltb().
+ */
 static int alloc_ltb_set(struct ibmvnic_adapter *adapter,
 			 struct ibmvnic_ltb_set *ltb_set, int num_buffs,
 			 int buff_size)
 {
-	struct ibmvnic_long_term_buff *ltb;
-	int ltb_size;
-	int size;
+	struct device *dev = &adapter->vdev->dev;
+	struct ibmvnic_ltb_set old_set;
+	struct ibmvnic_ltb_set new_set;
+	int rem_size;
+	int tot_size;		/* size of all ltbs */
+	int ltb_size;		/* size of one ltb */
+	int nltbs;
+	int rc;
+	int n;
+	int i;
 
-	size = sizeof(struct ibmvnic_long_term_buff);
+	dev_dbg(dev, "%s() num_buffs %d, buff_size %d\n", __func__, num_buffs,
+		buff_size);
 
-	ltb_set->ltbs = kmalloc(size, GFP_KERNEL);
-	if (!ltb_set->ltbs)
-		return -ENOMEM;
+	ltb_size = rounddown(IBMVNIC_ONE_LTB_SIZE, buff_size);
+	tot_size = num_buffs * buff_size;
+
+	if (ltb_size > tot_size)
+		ltb_size = tot_size;
+
+	nltbs = tot_size / ltb_size;
+	if (tot_size % ltb_size)
+		nltbs++;
+
+	old_set = *ltb_set;
+
+	if (old_set.num_ltbs == nltbs) {
+		new_set = old_set;
+	} else {
+		int tmp = nltbs * sizeof(struct ibmvnic_long_term_buff);
+
+		new_set.ltbs = kzalloc(tmp, GFP_KERNEL);
+		if (!new_set.ltbs)
+			return -ENOMEM;
+
+		new_set.num_ltbs = nltbs;
+
+		/* Free any excess ltbs in old set */
+		for (i = new_set.num_ltbs; i < old_set.num_ltbs; i++)
+			free_long_term_buff(adapter, &old_set.ltbs[i]);
+
+		/* Copy remaining ltbs to new set. All LTBs except the
+		 * last one are of the same size. alloc_long_term_buff()
+		 * will realloc if the size changes.
+		 */
+		n = min(old_set.num_ltbs, new_set.num_ltbs);
+		for (i = 0; i < n; i++)
+			new_set.ltbs[i] = old_set.ltbs[i];
 
-	ltb_set->num_ltbs = 1;
-	ltb = &ltb_set->ltbs[0];
+		/* Any additional ltbs in new set will have NULL ltbs for
+		 * now and will be allocated in alloc_long_term_buff().
+		 */
+
+		/* We no longer need the old_set so free it. Note that we
+		 * may have reused some ltbs from old set and freed excess
+		 * ltbs above. So we only need to free the container now
+		 * not the LTBs themselves. (i.e. dont free_ltb_set()!)
+		 */
+		kfree(old_set.ltbs);
+		old_set.ltbs = NULL;
+		old_set.num_ltbs = 0;
+
+		/* Install the new set. If allocations fail below, we will
+		 * retry later and know what size LTBs we need.
+		 */
+		*ltb_set = new_set;
+	}
+
+	i = 0;
+	rem_size = tot_size;
+	while (rem_size) {
+		if (ltb_size > rem_size)
+			ltb_size = rem_size;
+
+		rem_size -= ltb_size;
+
+		rc = alloc_long_term_buff(adapter, &new_set.ltbs[i], ltb_size);
+		if (rc)
+			goto out;
+		i++;
+	}
 
-	ltb_size = num_buffs * buff_size;
+	WARN_ON(i != new_set.num_ltbs);
 
-	return alloc_long_term_buff(adapter, ltb, ltb_size);
+	return 0;
+out:
+	/* We may have allocated one/more LTBs before failing and we
+	 * want to try and reuse on next reset. So don't free ltb set.
+	 */
+	return rc;
 }
 
 /**
@@ -388,14 +487,30 @@ static int alloc_ltb_set(struct ibmvnic_adapter *adapter,
  * @offset: (Output) offset of buffer in the LTB from @ltbp
  *
  * Map the given buffer identified by [rxpool, bufidx] to an LTB in the
- * pool and its corresponding offset.
+ * pool and its corresponding offset. Assume for now that each LTB is of
+ * different size but could possibly be optimized based on the allocation
+ * strategy in alloc_ltb_set().
  */
 static void map_rxpool_buf_to_ltb(struct ibmvnic_rx_pool *rxpool,
 				  unsigned int bufidx,
 				  struct ibmvnic_long_term_buff **ltbp,
 				  unsigned int *offset)
 {
-	*ltbp = &rxpool->ltb_set.ltbs[0];
+	struct ibmvnic_long_term_buff *ltb;
+	int nbufs;	/* # of buffers in one ltb */
+	int i;
+
+	WARN_ON(bufidx >= rxpool->size);
+
+	for (i = 0; i < rxpool->ltb_set.num_ltbs; i++) {
+		ltb = &rxpool->ltb_set.ltbs[i];
+		nbufs = ltb->size / rxpool->buff_size;
+		if (bufidx < nbufs)
+			break;
+		bufidx -= nbufs;
+	}
+
+	*ltbp = ltb;
 	*offset = bufidx * rxpool->buff_size;
 }
 
@@ -798,8 +913,9 @@ update_ltb:
 		dev_dbg(dev, "Updating LTB for rx pool %d [%d, %d]\n",
 			i, rx_pool->size, rx_pool->buff_size);
 
-		if (alloc_ltb_set(adapter, &rx_pool->ltb_set,
-				  rx_pool->size, rx_pool->buff_size))
+		rc = alloc_ltb_set(adapter, &rx_pool->ltb_set,
+				   rx_pool->size, rx_pool->buff_size);
+		if (rc)
 			goto out;
 
 		for (j = 0; j < rx_pool->size; ++j) {
@@ -4106,16 +4222,16 @@ static void send_request_cap(struct ibmvnic_adapter *adapter, int retry)
 			adapter->desired.rx_entries =
 					adapter->max_rx_add_entries_per_subcrq;
 
-		max_entries = IBMVNIC_MAX_LTB_SIZE /
+		max_entries = IBMVNIC_LTB_SET_SIZE /
 			      (adapter->req_mtu + IBMVNIC_BUFFER_HLEN);
 
 		if ((adapter->req_mtu + IBMVNIC_BUFFER_HLEN) *
-			adapter->desired.tx_entries > IBMVNIC_MAX_LTB_SIZE) {
+			adapter->desired.tx_entries > IBMVNIC_LTB_SET_SIZE) {
 			adapter->desired.tx_entries = max_entries;
 		}
 
 		if ((adapter->req_mtu + IBMVNIC_BUFFER_HLEN) *
-			adapter->desired.rx_entries > IBMVNIC_MAX_LTB_SIZE) {
+			adapter->desired.rx_entries > IBMVNIC_LTB_SET_SIZE) {
 			adapter->desired.rx_entries = max_entries;
 		}