In article <20030123034453.3A247B0D2B@gabi-n.hauN.org>
kent@netbsd.org wrote:

> Dhoyashiki-san posted BUS_DMA_NOCACHE patch to solve the problem
> in kern/19919.  I revised the patch so that all platforms had
> BUS_DMA_NOCACHE and the bus_dma manpage had the description for
> it.
> 
> Comments?  I'll commit the following on the weekend If no objections.

bus_dmamem_mmap() should also handle BUS_DMA_NOCACHE or not?

> +.It Dv BUS_DMA_NOCACHE
> +This flag is a
> +.Em hint
> +to machine-dependent code.
> +If possible, map the unchached memory.

It's better to mention in what case BUS_DMA_NOCACHE is required/used.
(i.e. workaround for broken hardware, etc.)

>  #define	BUS_DMA_READ		0x100	/* mapping is device -> memory only */
>  #define	BUS_DMA_WRITE		0x200	/* mapping is memory -> device only */
> +#define BUS_DMA_NOCACHE		0x400	/* hint: map non-cached memory */

IMHO, we should not mix #define<space> and #define<tab>
in the same source.
---
Izumi Tsutsui
tsutsui@ceres.dti.ne.jp