Darrick, > On May 3, 2021, at 5:21 PM, Darrick J. Wong <djwong@xxxxxxxxxx> wrote: > > In conclusion, the drive firmware is broken. > > Question: Should we be doing /some/ kind of re-read after a zeroing the > log to detect these sh*tty firmwares and fall back to a pwrite()? As pointed out by Christoph and Dave, this maybe a broken write zeroes issues for NVMe device(not entirely sure). We have seen such controllers and we add a quirk so that kernel can fallback to issuing writes instead of offloading write zeroes to controller so that usrspace don't have to deal with it. Please send the report to Linux NVMe list that is mentioned in the Christoph's reply.
