Am Mo, den 26.01.2004 schrieb Mike Christie um 22:22: > > I don't think that will happen. The values are already the allowed > > maximum values. Or am I wrong? > > I don't think they are the max values. SCSI will override them depending > on the HBA settings. Hmm. Then I think there is something wrong anyway. Why are the max constants not maximum values... > > If the table isn't backed by a real device (like if it only contains an > > error target) the resulting queue will use those "unrealstic" values > > that exceed the system-wide limits. And a hardsect size of 0 bytes... > > If you do not have a device, should you set sane IO restrictions in the > target accroding to your target implementation limits. Interestingly the code that sets these defaults limits isn't very old. It was a simplified version of my combine_default_restrictions function that did what you are proposing. See the dm-devel archices around November 2003. The default limits were all zero before (the min_not_zero function took care of everything) so if you think this is wrong I suggest that we should either look into why the maximum values are not maximum values or that we just revert the io_restrictions patch.