IB/mlx5: Use direct mkey destroy command upon UMR unreg failure

Use a direct firmware command to destroy the mkey in case the unreg UMR
operation has failed.

This prevents a case that a mkey will leak out from the cache post a
failure to be destroyed by a UMR WR.

In case the MR cache limit didn't reach a call to add another entry to the
cache instead of the destroyed one is issued.

In addition, replaced a warn message to WARN_ON() as this flow is fatal
and can't happen unless some bug around.

Link: https://lore.kernel.org/r/[email protected]
Cc: <[email protected]> # 4.10
Fixes: 49780d4 ("IB/mlx5: Expose MR cache for mlx5_ib")
Signed-off-by: Yishai Hadas <[email protected]>
Reviewed-by: Artemy Kovalyov <[email protected]>
Signed-off-by: Leon Romanovsky <[email protected]>
Reviewed-by: Jason Gunthorpe <[email protected]>
Signed-off-by: Jason Gunthorpe <[email protected]>

Author: yishaih

drivers/infiniband/hw/mlx5/mr.c

@@ -545,13 +545,16 @@ void mlx5_mr_cache_free(struct mlx5_ib_dev *dev, struct mlx5_ib_mr *mr)
 		return;
 
 	c = order2idx(dev, mr->order);
-	if (c < 0 || c >= MAX_MR_CACHE_ENTRIES) {
-		mlx5_ib_warn(dev, "order %d, cache index %d\n", mr->order, c);
-		return;
-	}
+	WARN_ON(c < 0 || c >= MAX_MR_CACHE_ENTRIES);
 
-	if (unreg_umr(dev, mr))
+	if (unreg_umr(dev, mr)) {
+		mr->allocated_from_cache = false;
+		destroy_mkey(dev, mr);
+		ent = &cache->ent[c];
+		if (ent->cur < ent->limit)
+			queue_work(cache->wq, &ent->work);
 		return;
+	}
 
 	ent = &cache->ent[c];
 	spin_lock_irq(&ent->lock);