On Tue, Jun 7, 2016 at 7:57 AM, Keith Busch [off-list ref] wrote:

On Mon, Jun 06, 2016 at 11:21:52PM +0200, Christoph Hellwig wrote:

quoted

+struct request *blk_mq_alloc_request_hctx(struct request_queue *q, int rw,
+             unsigned int flags, unsigned int hctx_idx)
+{
+     struct blk_mq_hw_ctx *hctx;
+     struct blk_mq_ctx *ctx;
+     struct request *rq;
+     struct blk_mq_alloc_data alloc_data;
+     int ret;
+
+     ret = blk_queue_enter(q, flags & BLK_MQ_REQ_NOWAIT);
+     if (ret)
+             return ERR_PTR(ret);
+
+     hctx = q->queue_hw_ctx[hctx_idx];

We probably want to check 'if (hctx_idx < q->nr_hw_queues)' before
getting the hctx. Even if hctx_idx was origially valid, it's possible
(though unlikely) blk_queue_enter waits on reallocating h/w contexts,
which can make hctx_idx invalid.

Yes, I'll update it.