Re: [PATCH v3 1/6] libnvdimm/namespace: Make namespace size validation arch dependent

["Aneesh Kumar K.V" <aneesh.kumar@linux.ibm.com>]

Dan Williams <dan.j.williams@intel.com> writes:

> On Fri, Jan 10, 2020 at 8:33 PM Aneesh Kumar K.V
> <aneesh.kumar@linux.ibm.com> wrote:
>>
>> On 1/11/20 2:08 AM, Jeff Moyer wrote:
>> > Hi, Aneesh,
>
>> "namespace0.2" not having a 2MB aligned size which cause namespace 0.1
>> start addr to be not aligned. Hence both the namespace are marked disabled.
>>
>
> 2 observations:
>
> - It's ok if the namespace start address is not subsection aligned as
> long as the mapped portion for data access is subsection aligned, at
> least on x86.
>
> - "sector" mode namespaces are not mapped by devm_memremap_pages() so
> there should be no restriction there. If powerpc can't map them that's
> a separate concern.

Does that mean the `supported_size_align` attribute should be a property
of pfn and dax seed device? Considering we don't want to apply this
restrictions for blk, raw namespace, and btt mode namespace should we
make the attribute a seed device property rather than a namespace
property?


>
> So, cross arch compatible namespaces is a goal, but not regressing
> existing namespaces takes precedence. I'd be happy if newly created
> namespaces tried to account for all the arch quirks, but if libnvdimm
> can enable a namespace it should try.

Ok. So that means we apply the alignment rules when creating new
namespaces irrespective of its type/mode. But when initializing via scan
label we only apply them for devdax and fsdax namespace?

Something like

static bool nvdimm_valid_namespace(struct device *dev,
		struct nd_namespace_common *ndns, resource_size_t size)
{
	struct nd_region *nd_region = to_nd_region(ndns->dev.parent);
	unsigned long align_size = arch_namespace_align_size();
	struct resource *res;
	u32 remainder;

	/*
	 * Don't validate the start and size for blk namespace type
	 */
	if (is_namespace_blk(&ndns->dev))
		return true;

	/*
	 * For btt and raw namespace we use ioremap. Assume both can work
	 * with PAGE_SIZE alignment.
	 */
	if (is_nd_btt(dev) || is_namespace_io(dev))
		return true;

	div_u64_rem(size, align_size * nd_region->ndr_mappings, &remainder);
	if (remainder)
		return false;

	if (is_namespace_pmem(&ndns->dev)) {
		struct nd_namespace_pmem *nspm = to_nd_namespace_pmem(&ndns->dev);

		res = &nspm->nsio.res;
	} else
		/* cannot reach */
		return false;

	div_u64_rem(res->start, align_size * nd_region->ndr_mappings, &remainder);
	if (remainder)
		return false;

	return true;
}

-aneesh
_______________________________________________
Linux-nvdimm mailing list -- linux-nvdimm@lists.01.org
To unsubscribe send an email to linux-nvdimm-leave@lists.01.org