[RFC] block: Allocate only 1 tag set for a kdump kernel

[Daniel Wagner]

Do not update nr_hw_queues again after setting it to 1 for a kdump
kernel. This avoids allocating a tag set of size nr_cpu_ids and but
then just using one tag set.

Signed-off-by: Daniel Wagner <dwagner@xxxxxxx>
---
Hi,

I stumbled across this and didn't make sense to me that we might
allocated more tag sets than we potently use. But maybe I am
not seeing the obvious thing.

Only compiled tested.

Thanks,
Daniel

 block/blk-mq.c | 15 ++++++++-------
 1 file changed, 8 insertions(+), 7 deletions(-)

diff --git a/block/blk-mq.c b/block/blk-mq.c
index 4f57d27bfa73..e32cb0217135 100644
--- a/block/blk-mq.c
+++ b/block/blk-mq.c
@@ -3291,13 +3291,14 @@ int blk_mq_alloc_tag_set(struct blk_mq_tag_set *set)
 		set->nr_hw_queues = 1;
 		set->nr_maps = 1;
 		set->queue_depth = min(64U, set->queue_depth);
+	} else {
+		/*
+		 * There is no use for more h/w queues than cpus
+		 * if we just have a single map
+		 */
+		if (set->nr_maps == 1 && set->nr_hw_queues > nr_cpu_ids)
+			set->nr_hw_queues = nr_cpu_ids;
 	}
-	/*
-	 * There is no use for more h/w queues than cpus if we just have
-	 * a single map
-	 */
-	if (set->nr_maps == 1 && set->nr_hw_queues > nr_cpu_ids)
-		set->nr_hw_queues = nr_cpu_ids;
 
 	if (blk_mq_realloc_tag_set_tags(set, 0, set->nr_hw_queues) < 0)
 		return -ENOMEM;
@@ -3309,7 +3310,7 @@ int blk_mq_alloc_tag_set(struct blk_mq_tag_set *set)
 						  GFP_KERNEL, set->numa_node);
 		if (!set->map[i].mq_map)
 			goto out_free_mq_map;
-		set->map[i].nr_queues = is_kdump_kernel() ? 1 : set->nr_hw_queues;
+		set->map[i].nr_queues = set->nr_hw_queues;
 	}
 
 	ret = blk_mq_update_queue_map(set);
-- 
2.16.4