Re: Transport Attributes -- attempt#4

[Jeremy Higdon <jeremy@sgi.com>]

On Tue, Feb 24, 2004 at 09:02:56AM -0600, James Bottomley wrote:
> On Mon, 2004-02-23 at 23:58, Jeremy Higdon wrote:
> > James, was scsi/scsi_transport.h supposed to be patched also?
> > There is a new "size" member of the scsi_transport_template
> > structure, but no patch to the data structure itself.  Did you
> > replace attributes with size?
> 
> Well, not really.  The patch I provided is a sketch only of what I want
> done.  It compiles and works on the 53c700 which is where I tested it.
> 
> Really, I just want transport and core separated and transport
> attributes to be fixed (so there's some way of defining a transport
> private API for updating them).
> 
> Even with just offset and period as the transport attributes for SPI,
> that gives enough to move domain validation out of all the drivers into
> the SPI transport layer.

Okay, that's fine.

What I meant to say is that I get a compile error at the following
line in scsi_scan.c:

	sdev = kmalloc(sizeof(*sdev) + shost->transportt->size, GFP_ATOMIC);

In Martin's version of scsi_transport.h, there is no "size" member of
struct scsi_transport_template.

So I figured you must have left it out of your patch.  I also saw this
in your patch, which makes it look as though default_attr_values (mistakenly
called "attributes" by me above) was replaced by size:

 struct scsi_transport_template spi_transport_template = {
        .attrs = spi_transport_attrs,
        .class = &spi_transport_class,
-       .setup = &spi_alloc_transport_attrs,
-       .cleanup = &spi_destroy_transport_attrs,
-       .default_attr_values = &spi_transport_attr_defaults,
+       .setup = &spi_setup_transport_attrs,
+       .cleanup = NULL,
+       .size = sizeof(struct spi_transport_attrs) - sizeof(unsigned long),
 };

thanks

jeremy