On Thursday 17 July 2014 01:20:54 Peter Chen wrote:

Thanks, arnd.

For chipidea generic glue layer case, if there are three devices who use this
driver, and all devices have 32-bit bus, some devices have less 32 address lines.
For example:

- the device_a doesn't need to use dma_mask
- the device_b needs dma_mask as 0xfffffffff
- the device_c needs dma_mask as 0xfffffff0, assume it has only 28 address lines

This makes no sense. You always need a dma mask, so the first case doesn't exist,
and the second one is the default.

In the third case, I assume you mean 0x0fffffff, which is a 28-bit mask.

My questions are:
- Can we not set dma_mask at driver, and only set dma-ranges at dts for device_b
and device_c as a solution to cover this different dma mask use case?

try to understand my earlier reply. What is the problem with device_b?
Is that a limitation of the bus it is connected to, or the version of the
chipidea hardware?

- If we can't use this solution, would you suggest one?

It depends on what the requirement of the hardware is, as I explained
now for three times.

- If we can use this solution, for device_b and device_c, how can we write dma-ranges?
I can't find any arm platforms use it, only some powerpc platform use it.
According to the definition from Power_ePAPR_APPROVED_v1.1.pdf, it is
dma-ranges = <child-bus-address, parent-bus-address, length>
but I find the powerpc has different way for using dma-ranges. 

It's now handled by of_dma_configure() in drivers/of/platform.c for all
architectures.

	Arnd
--
To unsubscribe from this list: send the line "unsubscribe devicetree" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html