[Freemint-list] struct kerinfo extensions

[Alan Hourihane]

On 27/09/17 08:48, Miro Kropáček wrote:
>
>     Not sure I understand this, why would re-ordering have any
>     implication on being able to call the function ?
>
> That was just an example what should I /not /do, of course I'm not
> going to do that (although from cosmetic point of view -- it would be
> much nicer to have the pci struct next to bios/gemdos one...)
>
>     Anyway, for XIF drivers you can just grab the PCI cookie and call
>     the functions as any other module, as realistically the plan a
>     while back was to port everything to kernel modules instead of
>     XIF's, XFS's etc.
>
> Isn't that even worse than just calling the traps directly? In both
> cases (custom traps or cookies) I have to include a not exactly short
> list of function declarations
> <https://github.com/mikrosk/freemint/blob/rtl8139/sys/sockets/xif/rtl8139.c#L249>.
> Now imagine there would be two PCI network drivers... and these very
> declarations are already visible for all kernel modules, so it would
> make sense, at least for time being, to make them available to XIFs as
> well to avoid repeating the same defines over and over again.
>

Why does calling via the cookie cause problems ? There could be lots of
TOS compatible apps that may call PCI routines directly. The kernel has
to deal with that.

Alternatively, make all the XIF's and XDD's kernel modules. We keep
needing to expose these in different ways, and ultimately moving to real
KM's will provide all the capabilities without specifics to individual
types of module.

Alan.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mail.atariforge.org/pipermail/freemint-list/attachments/20170927/4ed703f5/attachment.html