Re: 53c770 driver

[<fh...@at...>]

In <200...@su...>, on 12/07/00 
   at 03:42 PM, gri...@ps... (Arno Griffioen) said:

>> > This sort of thing kills the init when CONFIG_PCI is defined:
>> > 
>> > #ifdef CONFIG_PCI
>> > static int ncr53c8xx_pci_init(Scsi_Host_Template *tpnt,
>> >              uchar bus, uchar device_fn, ncr_device *device);
>> > #else
>> > static int ncr53c770_init(Scsi_Host_Template *tpnt, unsigned long base,
>> >                           unsigned int irq, ncr_device *device);
>> > #endif
>> 
>> This doesn't make sense to me. I guess the 53c8xx code could just be removed
>> from the 53c770 code?

>Or a more elegant solution in which case there wouldn't need to be a
>separate 53c770 driver, but included in an existing 53c8xx driver.

This is the best solution IMHO.  Basically all the current LSI chips have
about the same scripts processor and share a lot of the same  features.  
Of course the very new chips began to diverge.
All of the 53c8xx drivers I have looked at have the same basic data
structures.  This is for the 53c810 chip on up.  The biggest stumbling
point I can see is that the 8xx series of chips is very PCI so the
routines that heavily depend on PCI has to have the correct logic figured
out to accommodate both.

>As this is basically development code I don't really care. Just took me 
>a while to figure out why it didn't seem tohave any effect when  I
>compiled in the driver.. 

I am playing with different driver code at the moment, the sym53c8xx.c
code.  My reason for switching is that it shares the new style data
structures and should be easier to merge into the newest versions, if that
ever becomes possible.  I have gotten the newer code to get to the cache
test, where it fails to execute the SCSI scripts code, though everything
looks "right".  For example, the correct addresses and so forth appear.

There is some very basic problem with making the 53c770 chip work with
Linux or PowerPC.  The problem has to be PPC or APUS related because those
same 53c8xx chips are used in PCI cards that work on a variety of other
architectures and Linuxes.  I  have been trying  to figure out what the
differences could be.  BTW the newer drivers  have a sort of generic mb()
support as well.  I really don't want to start  looking at assembly code,
but ultimately that may be what has to be done.   i.e. find the driver
sections that misbehave in the kernel assembler code and see if the code
looks "right."  (Any tips on how one goes about doing that would be
appreciated.)  Additionally, Richard Hirst has gotten  the 770 chip to
work on a HP system.  He had to make a few modifications  to the actual
SCSI routines in the code to take into account small  differences between
the 8xx and 770 SCSI processors.

There is another problem as well.  All the PCI based cards have ROMs or
NVRAM onboard that store the initial setting of the SCSI chip.  The
correct initial settings of the 53c770 chip has to be found.

Sorry in advance for rambling on and probably repeating myself from
previous posts.

Fred