On 11.02.20 11:19, Peter Oberparleiter wrote: [...] >> but this code is still fishy: >> >> $ cat /proc/cio_ignore >> 0.0.fe00-0.0.fefe >> 0.0.ff00-0.0.ffff >> $ dd if=/proc/cio_ignore status=none >> 0.0.fe00-0.0.fefe >> 0.0.ff00-0.0.ffff >> $ dd if=/proc/cio_ignore status=none bs=10 >> 0.0.fe00-0.0.fefe >> 0.0.ff00-0.0.ff01-0.0.ff02-0.0.ff03-0.0.ff04-0.0.ff05-0.0.ff06-0.0.ff07-0.0.ff08-0.0.ffff >> $ dd if=/proc/cio_ignore status=none bs=10 skip=1 >> .0.fefe >> 0.0.ff00-0.0.ff01-0.0.ff02-0.0.ff03-0.0.ff04-0.0.ff05-0.0.ff06-0.0.ff07-0.0.ff08-0.0.ffff >> >> >> Peter, any opinions on this? > > A correct implementation of a file read operation must result in the > same data being read independently of whether the file is read in one > go, or if it is read byte-by-byte. > > It seems that the current cio_ignore seq-file implementation doesn't > meet that requirement. I don't think that this patch series is the best > way to address this problem though. > > My suggestion would be to apply this patch set as is, and then I'll take > the to-do to fix this seq file implementation at a later time. Ok, I will will Vasily patch.
