Re: of_iomap and request_mem_region

On Sunday 10 February 2008, Jon Smirl wrote:

On 2/9/08, Arnd Bergmann [off-list ref] wrote:

quoted

On Sunday 10 February 2008, Jon Smirl wrote:

quoted

of_iomap doesn't implicitly do a request_mem_region().  How should
request_mem_region() be handled? When using of_iomap you don't get the
length of the region back so it isn't easy to call request_mem_region.

What about adding a third param to of_iomap for the driver name? If it
is non-null also do the request region. If it is null check to make
sure the region has already be requested.

I'd prefer changing of_iomap so that it gets the of_device instead of
the device_node. That would help us do two changes at once:

1. call request_mem_region using dev->driver->name, as you mentioned.
2. use managed registrations, as lib/devres.c does e.g. in pcim_iomap,
so that we can kill all the calls to iounmap in the cleanup path
of OF drivers.

maybe we can do a new ofm_iomap interface first, and then move
drivers over to use that one.

There are only four drivers using of_iomap.

$ git-ls-files | xargs grep of_iomap | cut -f 1 -d: | uniq | wc -l
19

Unfortunately, many of them don't even use an of_device, so the
conversion is not trivial.

There are a lot of drivers using of_address_to_resource. They could
probably be converted to of_iomap if it gets changed. All of the ones
I checked follow the of_address_to_resource with a request_resource
call.

Right, but again, we are still in the progress of converting drivers
from using of_find_node_by_type() and similar to of_platform_driver.
If we don't have an of_device, the interface as I would prefer
it doesn't work.

	Arnd <><