On Wed, Aug 13, 2025 at 03:52:44PM -0600, Keith Busch wrote: > On Wed, Aug 13, 2025 at 05:23:47PM -0400, Martin K. Petersen wrote: > > dma_alignment defines the alignment of the DMA starting address. We > > don't have a dedicated queue limit for the transfer length granularity > > described by NVMe. > > Darn, but thanks for confirming. I'll see if I can get by without > needing a new limit, or look into adding one if not. Worst case, I can > also let the device return the error, though I think we prefer not to > send an IO that we know should fail. Allowing an unprivileged user application to trivially trigger an I/O error sounds like a bad idea. But I think simply applying the dma_alignment to the length of READ/WRITE commands might be work, while ignoring it for other types of commands.
