On Aug 11, 2025 / 16:33, Anuj Gupta/Anuj Gupta wrote: > On 8/8/2025 5:10 PM, Shinichiro Kawasaki wrote: > > On Aug 05, 2025 / 11:46, Anuj Gupta wrote: > >> This test verifies end-to-end support for integrity metadata via the > >> io-uring interface. It uses the FS_IOC_GETLBMD_CAP ioctl to query the > >> logical block metadata capabilities of the device. These values are then > >> passed to fio using the md_per_io_size option. > >> > >> io_uring PI interface: https://lore.kernel.org/all/20241128112240.8867-1-anuj20.g@xxxxxxxxxxx/ > >> fio support for interface: https://lore.kernel.org/all/20250725175808.2632-2-vincent.fu@xxxxxxxxxxx/ > >> ioctl: https://git.kernel.org/pub/scm/linux/kernel/git/vfs/vfs.git/log/?h=vfs-6.17.integrity > >> Signed-off-by: Anuj Gupta <anuj20.g@xxxxxxxxxxx> > >> Signed-off-by: Vincent Fu <vincent.fu@xxxxxxxxxxx> > > > > Anuj, thank you for the patch. > > > > I wonder which test group this test case should go into, block or nvme. IIUC, > > this test case runs only for nvme devices. Said that, block group looks good for > > me since the test target ioctl interface belongs to the block layer. > > > > I tried to run the test case using QEMU NVME emulation devices with some > > ms=X,pi=Y options, but the test runs failed. The kernel reported a number of > > "protection error"s. Can we run the test case with QEMU NVME emulation device? > > If so, could you share the recommended set up of the device? > > Could you please share/paste the errors that you encountered? The errors were as follows: block/041 => nvme1n1 (io_uring read with PI metadata buffer on block device) [failed] runtime ... 0.354s --- tests/block/041.out 2025-08-13 19:56:15.487074413 +0900 +++ /home/shin/Blktests/blktests/results/nvme1n1/block/041.out.bad 2025-08-13 20:06:23.199291948 +0900 @@ -1,2 +1,12 @@ Running block/041 +fio: io_u error on file /dev/nvme1n1: Invalid or incomplete multibyte or wide character: read offset=12288, buflen=4096 +fio: io_u error on file /dev/nvme1n1: Invalid or incomplete multibyte or wide character: read offset=688128, buflen=4096 +fio: io_u error on file /dev/nvme1n1: Invalid or incomplete multibyte or wide character: read offset=630784, buflen=4096 +fio: io_u error on file /dev/nvme1n1: Invalid or incomplete multibyte or wide character: read offset=905216, buflen=4096 +fio: io_u error on file /dev/nvme1n1: Invalid or incomplete multibyte or wide character: read offset=233472, buflen=4096 +fio: io_u error on file /dev/nvme1n1: Invalid or incomplete multibyte or wide character: read offset=540672, buflen=4096 ... (Run 'diff -u tests/block/041.out /home/shin/Blktests/blktests/results/nvme1n1/block/041.out.bad' to see the entire diff) > > The issue occurs because setting ms and pi in the QEMU command line is > not enough, the namespace still needs to be formatted. Could you please > run the test again after running the nvme format command on device with > the desired LBA format (PI enabled). Ah, thanks. I ran "nvme format -i 1 /dev/nvme0n1" for the QEMU NVME device prepared with QEMU NVME option "ms=8,pi=1", then the test case passes. The test case looks working as expected :)
