Re: [PATCH] Clarify pci_iomap() usage for MMIO-only devices

[Linus Torvalds]

On Tue, 18 Sep 2007, Luis R. Rodriguez wrote:
> 
> An extra branch is created on MMIO-only devices on read/writes on the
> IO_COND macro using this interface -- or is this optimized out?

Umm. Does anybody actually have any performance numbers? 

The thing is, those things are *cheap* compared to the IO. And any 
high-performance device will be using DMA for the real IO, so we're not 
generally even talking about any performance-critical stuff. 

Quite frankly, if performance is a _real_ reason to avoid 
ioread*/iowrite*, I'll happily accept read*/write*, but it would be needed 
to be backed up by real numbers. Can you even measure it?

I would definitely *not* encourage the notion that people should use 
readl/writel because of "performance reasons". That may be valid for some 
fbcon driver, but those drivers go to other extremes (ie they use 
"__raw_writel()" etc, and MTRR's etc).

If you don't use write-combining memory regions, the performance argument 
is not really valid.

			Linus
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/