Re: [patch] ioremap + check_mem_region for 8390 drivers

Donald Becker (becker@cesdis1.gsfc.nasa.gov)
Tue, 28 Dec 1999 17:15:36 -0500 (EST)


On Tue, 28 Dec 1999, Alan Cox wrote:

> > Yes, there are now-obsolete cards that are reconfigurable. The only current
> > production Ethernet card that has reconfigurable IRQs (in an obvious way
> > e.g. non-ISA-PnP) is the 3c509 driver.
>
> And we still support them. So we still need the intefaces

We still need AN interface, perhaps not this specific one. If we are going
to change the driver interface (and that's what is happening), we should do
it all at once instead of dribbling the changes in over time.

The capability to change the IRQ is useful mainly (only?) for ISA cards.
Changing the IRQ is very similar to changing the ISA shared memory base --
it can only done on a small number of ISA card types, and can only
reasonably be done when the interface is down.

> > The dev->dma parameter is a good example of something that should be
> > removed. Only a few drivers use it, and one (znet.c) need *two* DMA
> > channels. It think the only current production chip that uses it is the
> > ISA-bus-master PCnet/ISA+,
>
> Drivers using it:
>
> 3c505, 3c515, hp100, lance, ltpc, ni5010, ni65, sdla_chdlc, sdla_fr,
> sdla_ppp, sdla_x25, sktr, znet.
>
> So we can't go randomly changing it pre 2.4. Its too much work to fix and test
> all those drivers yet again.

Hmmm, yes, I think that you can still buy the 3c515.
The lance driver is the PCnet/ISA chip.
The 3c505, hp100, ni5010, ni65, znet cards are not in current production.
I don't know if the sdla_ hardware is still availabe, but the sdla* drivers
you list above don't really use dev->dma, they just (effectively) copy 0 to
dev->dma.

My point isn't that we should abandon support for old hardware. It's that
if we are changing the drivers anyway, we should move the dev->dma and
dev->mem_start info to e.g. dev->priv->dma.

> > and ISA bus masters will no longer legal with PCI
> > v2.0+.
>
> PCI 2.0 doesn't make ISA bus masters illegal. In fact its irrelevant because
> your ISA bus interfaces don't even have to overlay your PCI space. Thats just
> a cheap PC compatibility hack.

Find a post-486 PC that doesn't implement ISA as a PCI-ISA bus bridge, with
a sideband IRQ mapping hack. It's not the DMA request lines that are the
problem, it's a length of time and semantics of an ISA bus master holding
the bus that's the problem.

Donald Becker
Scyld Computing Corporation, and
USRA-CESDIS, becker@cesdis.gsfc.nasa.gov

-
To unsubscribe from this list: send the line "unsubscribe linux-net" in
the body of a message to majordomo@vger.rutgers.edu