Re: block layer sg, bsg

[Douglas Gilbert <dougg@torque.net>]

Jens Axboe wrote:
> On Wed, Aug 04 2004, Jeff Garzik wrote:
> 
>>On Wed, Aug 04, 2004 at 08:44:29AM -0700, David S. Miller wrote:
>>
>>>Or use a more portable well-defined type which does not change
>>>size nor layout between 32-bit and 64-bit environments.
>>
>>IMO if this (the above) is not done, the interface needs work.
>>
>>For interfaces that replace ioctl(2) with read(2)/write(2), for passing
>>data structures to/from the kernel, Al has rightly suggested that these
>>structures be not only fixed size (as David described above), but also
>>fixed-endian.
> 
> 
> I completely agree with that, we need a different structure for other
> devices as well. Show me what you'd like for libata, for instance.

If a new structure with anything but 'S' (0x53) in the first 4
bytes was chosen then the bsg driver could handle the new
structure and sg_io_hdr.

For example:
struct bsg_io_hdr {
     int8_t interface_id[4];   /* [i] 'B' in each element (required) */
     uint8_t dxfer_direction;
     .............
};

SCSI commands are all big endian and that seems to work
fine. Could pointers be passed as 8 byte big endian
unsigned integers?

Doug Gilbert