> I didn't write that I didn't want MMIO regions to be memory-mapped from
> user space, nor I forget this need, and I will probably _never_ write
> that. It is its memory-like based design that makes PCI so simple and
> pleasant. I already wrote about what I think about hardwares that do not
> have a unified memory addressing for all BUSES and host memory that break
> the simplicity of PCI and introduced the complexity with addressing we
> have to deal with.
I share the same opinion, but since most PCI cards support only
32-bit addressing and we certainly want to support both PCI and >4G
of memory on 64-bit architectures, we probably need to extend
the interface a bit to cover IOMMU-like translation as well as
pure all-in-the-same-space cases.
> I was only speaking about the base_addresses provided by the pci_dev
> structure that has been presented as some input parameters for ioremap()
> and told that this is _not_, in my opinion, what a PCI device driver
> expects as kernel service, or at least _not_ enough.
This is certainly not enough and I it's probably my fault my explanation
sounded like "these are just ioremap() cookies" -- I meant to say that
these are physical addresses, suitable for example for ioremap()
or mmap() on /dev/mem.
Have a nice fortnight
-- Martin `MJ' Mares <mj@ucw.cz> http://atrey.karlin.mff.cuni.cz/~mj/ Faculty of Math and Physics, Charles University, Prague, Czech Rep., Earth "DEC = Delay in Error Correction"- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.rutgers.edu Please read the FAQ at http://www.tux.org/lkml/