Re: First draft list of 2.3.x "Things to fix"

[Douglas Gilbert]

On Tue, 4 Jan 2000, David Weinehall wrote:
> On Tue, 4 Jan 2000, Manfred Spraul wrote:
> 
> > Alan Cox wrote:
> > > 
> > > The order is approximate priority. It may also not be totally accurate.
> > > Suggestions/fixes welcome.
> > > 
> > 
> > * either add the 32bit-uids, or undo the size_t changes in shm.
> > * add scsi_allocate_device()/scsi_get_host_dev() [Eric Youngdale],
> > required for gdth.
> 
> Is the scsi-error handling (for sg) in a state where Juerg Schilling would
> stop complaining yet? From what I can make out of his homepage, there are
> only a few changes that needs to be done, but without them, sg is quite
> useless. Any clues (I've never used anything but sd, so...)

Joerg Schilling has made the following criticisms of sg (summarized):
 a) cannot get SCSI status byte
 b) cannot send SCSI commands with vendor specific lengths
 c) sense buffer restricted to 16 bytes (SCSI 2 requires >= 18 bytes)
 d) can't get DMA residual count

There are 3 major versions of the sg driver:
 1) original from 1992 until lk 2.2.5
 2) 2.2.6 to 2.2.14 and 2.3 series up to 2.3.36
 3) "version 3" that seems too late for 2.3/2.4

All Joerg Schilling's criticisms apply to version 1. Version 2
fixes a) and b) amongst other things. Criticisms c) and d)
needed additions to the scsi mid level which have been made
in 2.3.19 and onwards. Version 3 (when applied to latter
2.3 series kernels) addresses all four criticisms. The DMA
residual count requires adapter support which has already
been done in some adapters.

Joerg Schilling was complaining more 8 months ago than he
has been recently. Still, he hasn't withdawn his material, 
some of which was used by slashdot.com to kick off a lively 
discussion some weeks back about the Linux scsi subsystem.

Doug Gilbert