On Tue, Sep 10, 2024 at 05:45:26PM +0200, Christoph Hellwig wrote: > On Wed, Sep 04, 2024 at 08:26:03AM -0700, Keith Busch wrote: > > From: Keith Busch <kbusch@xxxxxxxxxx> > > > > The queue limits for block access are not the same as metadata access. > > Delete these. > > While for NVMe-PCIe there isn't any metadata mapping using PRPs, for RDMA > the PRP-like scheme is used for anything, so the same virtual boundary > limits apply for all mappings. Oh shoot. The end goal here is to support NVMe META SGL mode, and that virt boundary doesn't work there. I was hoping to not introduce another queue limit for the metadata virt_boundary_mask. We could remove the virt_boundary from the NVMe PCI driver if it supports SGL's and then we're fine. But we're commiting to that data format for all IO even if PRP might have been more efficient. That might be alright, though.
