[openib-general] 2.6.18-rc5-mm1: drivers/infiniband/hw/amso1100/c2.c compile error

[Roland Dreier]

    Andrew> It depends on how many other devices have (or are expected
    Andrew> to have) mthca-like requirements.  If the answer is "very
    Andrew> few, maybe none" then perhaps we don't need to go
    Andrew> designing generic interfaces to support such things.

I actually don't know of any others -- not that I'm an expert on the
range of devices that exist...

What's your feeling about drivers like amso1100, which don't
particularly care about atomicity, but just want to write a 64-bit
quantity conveniently?  Should we require writeq()/__raw_writeq() for
all archs, and then define CONFIG_ARCH_HAS_64BIT_ATOMIC_MMIO_WRITES as
appropriate?

I see stuff like this is drivers/dma/ioatdma.c:

#if (BITS_PER_LONG == 64)
	ioatdma_chan_write64(ioat_chan, IOAT_CHAINADDR_OFFSET, desc->phys);
#else
	ioatdma_chan_write32(ioat_chan,
	                     IOAT_CHAINADDR_OFFSET_LOW,
	                     (u32) desc->phys);
	ioatdma_chan_write32(ioat_chan, IOAT_CHAINADDR_OFFSET_HIGH, 0);
#endif

and drivers/char/hpet.c:

#ifndef readq
static inline unsigned long long readq(void __iomem *addr)
{
	return readl(addr) | (((unsigned long long)readl(addr + 4)) << 32LL);
}
#endif

#ifndef writeq
static inline void writeq(unsigned long long v, void __iomem *addr)
{
	writel(v & 0xffffffff, addr);
	writel(v >> 32, addr + 4);
}
#endif

and so on...

 - R.