On Wed, Jan 18, 2023 at 02:46:38PM -0800, Bart Van Assche wrote:

- The code removed by this patch negatively impacts performance of all
  SCSI hosts with two or more data LUNs and also of all NVMe controllers
  that export two or more namespaces if there are significant
  differences in the number of I/O operations per second for different
  LUNs/namespaces. This is why I think that this should be solved in the
  block layer instead of modifying each block driver individually.

I think the potential performance sacrifice was done on purpose in order
to guarantee all tagset subscribers can have at least one unblocked
access to the pool, otherwise one LUN could starve all the others by
tying up all the tags in long-running operations. Is anyone actually
benefiting from the current sharing, though? I really do not know, so
I'm having trouble weighing in on removing it.