Re: [build fix] Re: [GIT PATCH] SCSI part 1

[Ingo Molnar <mingo@elte.hu>]

* Matthew Wilcox <matthew@wil.cx> wrote:

> On Wed, Jul 16, 2008 at 04:46:02PM +0200, Ingo Molnar wrote:
> > yeah, that's OK - but why is scsi.h #ifdef-ed? For example we can 
> > include blkdev.h without #ifdef CONFIG_BLOCK.
> 
> Sure.  The important bit comes when deciding which bits need to be 
> available when CONFIG_BLOCK isn't set.  Obviously, there is no SCSI 
> without block (while sg is a character device, not a block device, it 
> depends on the block infrastructure to the extent that you couldn't 
> use it without CONFIG_BLOCK).
> 
> While it's not impossible that someone could want the SCSI protocol 
> and opcodes, status codes, etc. for a different character device, that 
> seems unlikely.  The four opcodes defined in the header seem to be 
> pretty SCSI-specific and not useful to use on non-scsi devices. 
> SCSI_IOCTL_GET_PCI looks like it might possibly be useful on more than 
> just SCSI, but we have better ways (ie sysfs) of determining the same 
> information in a more general way.
> 
> So I think it's fair to put an #ifdef CONFIG_BLOCK right after the 
> _SCSI_SCSI_H define and close it right at the end of the file.

yes, that's what blkdev.h does, and that's what i was suggesting we do, 
instead of #ifdef-ing around scsi.h use in fs/compat_ioctl.c.

	Ingo