This looks workable to me, and is fine with me from the NVMe side. I could see arguments for passing the flag to the actual suspend method, but as that would require much bigger changes I'm not going to ask for it just because. Acked-by: Christoph Hellwig <hch at lst.de>