On Tue, 2020-07-21 at 22:55 -0400, Martin K. Petersen wrote: > Christoph, > > > Hmm, I wonder if we should simply add the check and warning to > > blk_queue_logical_block_size and add an error in that case. Then > > drivers only have to check the error return, which might add a lot > > less boiler plate code. > > Yep, I agree. > I also agree that this would be cleaner (I actually tried to implement this the way you suggest), but let me explain my reasoning for doing it this way. The problem is that most current users of blk_queue_logical_block_size (43 uses in the tree, out of which only 9 use constant block size) check for the block size relatively early, often store it in some internal struct etc, prior to calling blk_queue_logical_block_size thus making them only to rely on blk_queue_logical_block_size as the check for block size validity will need non-trivial changes in their code. Instead of this adding blk_is_valid_logical_block_size allowed me to trivially convert most of the uses. For RFC I converted only some drivers that I am more familiar with and/or can test but I can remove the driver's own checks in most other drivers with low chance of introducing a bug, even if I can't test the driver. What do you think? I can also both make blk_queue_logical_block_size return an error value, and have blk_is_valid_logical_block_size and use either of these checks, depending on the driver with eventual goal of un-exporting blk_is_valid_logical_block_size. Also note that I did add WARN_ON to blk_queue_logical_block_size. Best regards, Maxim Levitsky
