Re: [RFC PATCH 5/7] libnvdimm, e820: make CONFIG_X86_PMEM_LEGACY a tristate option

[Dan Williams <dan.j.williams@intel.com>]

On Sat, Aug 15, 2015 at 2:06 AM, Christoph Hellwig <hch@lst.de> wrote:
> On Wed, Aug 12, 2015 at 11:50:29PM -0400, Dan Williams wrote:
>> Purely for ease of testing, with this in place we can run the unit test
>> alongside any tests that depend on the memmap=ss!nn kernel parameter.
>> The unit test mocking implementation requires that libnvdimm be a module
>> and not built-in.
>>
>> A nice side effect is the implementation is a bit more generic as it no
>> longer depends on <asm/e820.h>.
>
> I really don't like this artifical split, and I also don't like how
> your weird "unit tests" force even more ugliness on the kernel.  Almost
> reminds of the python projects spending more effort on getting their
> class mockable than actually producing results..

Well, the minute you see a 'struct DeviceFactory' appear in the kernel
source you can delete all the unit tests and come take away my
keyboard.  Until then can we please push probing platform resources to
a device driver ->probe() method where it belongs?  Also given the
type-7 type-12 confusion I'm just waiting for some firmware to
describe persistent memory with type-12 at the e820 level and expect
an ACPI-NFIT to be able to sub-divide it.  In that case you'd want to
blacklist either 'nd_e820.ko' or 'nfit.ko'  to resolve the conflict.
I'm not grokking the argument against allowing this functionality to
be modular.

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>