Re: [PATCH v4 10/45] libnvdimm/pfn_dev: increase MAX_STRUCT_PAGE_SIZE

[Dan Williams]

Alexander Potapenko wrote:
[..]
> >
> > diff --git a/drivers/nvdimm/Kconfig b/drivers/nvdimm/Kconfig
> > index 79d93126453d..5693869b720b 100644
> > --- a/drivers/nvdimm/Kconfig
> > +++ b/drivers/nvdimm/Kconfig
> > @@ -63,6 +63,7 @@ config NVDIMM_PFN
> >         bool "PFN: Map persistent (device) memory"
> >         default LIBNVDIMM
> >         depends on ZONE_DEVICE
> > +       depends on !KMSAN
> >         select ND_CLAIM
> >         help
> >           Map persistent memory, i.e. advertise it to the memory
> >
> 
> Looks like we still don't have a resolution for this problem.
> I have the following options in mind:
> 
> 1. Set MAX_STRUCT_PAGE_SIZE to 80 (i.e. increase it by 2*sizeof(struct
> page *) added by KMSAN) instead of 128.
> 2. Disable storing of struct pages on device for KMSAN builds.
> 
> , but if those are infeasible, we can always go for:
> 
> 3. Disable KMSAN for NVDIMM and reflect it in Documentation. I am
> happy to send the patch if we decide this is the best option.

I copied you on the new proposal here:

https://lore.kernel.org/nvdimm/167467815773.463042.7022545814443036382.stgit@xxxxxxxxxxxxxxxxxxxxxxxxx/

It disables PMEM namespace creation with page-array reservations when
sizeof(struct page) > 64.

Note, it was pre-existing behavior for PMEM namespaces with too small of
a reservation to fail to enable. That gives me confidence that the
restriction to lose some PMEM namespace access with these memory debug
facilities is acceptable.