> > Testing has been done by modifying fio: > > https://github.com/SamsungDS/fio/tree/priv/feat/pi-test-v11 > > It looks like this never got into upstream fio. Do you plan to submit > it? It would also be extremely useful to have a testing using it in > blktests, because it seems like we don't have any test coverage for the > read/write with metadata code at the moment. > > Just bringing this up because I want to be able to properly test the > metadata side of the nvme/block support for the new DMA mapping API > and I'm ѕtruggling to come up with good test coverage. > Hi Christoph, The fio plumbing I had done for testing was pretty hacky (e.g., using NVMe ioctls directly to query PI capabilities), so I didn’t send it upstream. I plan to submit a liburing test. While working on it, I realized that writing generic userspace tests is tricky without a way to query the device’s integrity capabilities. The current sysfs interface is limited — it doesn't expose key fields like pi_size or metadata_size, which are necessary to correctly prepare protection information in userspace. That’s what motivated the ioctl RFC I sent earlier — to make it feasible for userspace to construct metadata buffers correctly. Once it gets settled, I can write some tests using it. Do you see this differently?
