On Fri, Feb 1, 2019 at 4:34 PM Dexuan Cui <decui@xxxxxxxxxxxxx> wrote: > > > From: Dan Williams <dan.j.williams@xxxxxxxxx> > > Sent: Friday, February 1, 2019 3:47 PM > > To: Dexuan Cui <decui@xxxxxxxxxxxxx> > > > > I believe it's the same reason. Without 11189c1089da the _LSR method > > will fail, and otherwise it works and finds the label that it doesn't > > like. > Exactly. > > > I'm not seeing "invalid" data in your failure log. Could you double > > check that it's just not the success of _LSR that causes the issue? > > acpi_label_read() never fails for me. > > By "invalid", I only mean the messages in the dmesg.bad.txt I previously > attached (I'm just reading the specs to learn the details about NVDIMM > namespace's labels, so my description might be inaccurate) : > > [ 4.832367] nvdimm nmem1: nsindex0 labelsize 1 invalid > [ 4.832369] nvdimm nmem1: nsindex1 labelsize 1 invalid Oh, those are benign. They are a side effect of Linux probing for v1.2 namespace labels vs v1.1. It will always find that one of those is "invalid". > ... > [ 5.259017] nd_pmem namespace0.0: 0x0000000000000000, too small must be at least 0x1000 > > > > > The regression you are seeing is the fact that the patch enables the kernel > > to > > > > enable nvdimm-namespace-label reads. > > > Yes. > > > > > > > Those reads find a namespace index block > > > > and a label. Unfortunately the label has the LOCAL flag set and Linux > > > > explicitly ignores pmem namespace labels with that bit set. The reason > > > Can you please point out the function that ignores the flag? > > > > > > I checked where NSLABEL_FLAG_LOCAL is used, but it looks I can't find a > > > related function. > > > > scan_labels() is where the namespace label is validated relative to > > the region type: > > > > if (is_nd_blk(&nd_region->dev) > > == !!(flags & NSLABEL_FLAG_LOCAL)) > > /* pass, region matches label type */; > > else > > continue; > > > > It also has meaning for the namespace capacity allocation > > implementation that needed that flag to distinguish aliased capacity > > between Block Aperture Mode and PMEM Mode access. > Thanks for the pointer! I'm looking at this function. > > > > > for that is due to the fact that the original definition of the LOCAL > > > > bit from v1.1 of the namespace label implementation [1] explicitly > > > > limited the LOCAL flag to "block aperture" regions. If you clear that > > > > LOCAL flag I expect it will work. To my knowledge Windows pretends > > > > that the v1.1 definition never existed. > > > I'm trying to find out where the flag is used and how to clear it. > > > > Assuming Hyper-V implements _LSW, you can recreate / reinitialize the > > label area: > > I think Hyper-V only implements _LSR: > [ 4.720623] nfit ACPI0012:00: device:00: has _LSR > [ 4.723683] nfit ACPI0012:00: device:01: has _LSR That's unfortunate... > > > > > The UEFI 2.7 specification for v1.2 labels states that setting the > > > > LOCAL flag is optional when "nlabel", number of labels in the set, is > > > > 1. Linux makes that mandatory as LOCAL is redundant when nlabel is 1. > > > > > > > > That said, the Robustness Principle makes a case that Linux should > > > > tolerate the bit being set. However, it's just a non-trivial amount of > > > > work to unwind the ingrained block-aperture assumptions of that bit. > > > Can you please explain this a bit more? Sorry, I'm new to this area... > > > > The short story is that Linux enforces that LOCAL == Block Mode > > Namespaces. See section 2.2 Namespace Label Layout in the original > > spec [1]. The EFI 2.7 definition tried to allow for LOCAL to be set > > when an interleave-set was comprised of a single NVDIMM, but then also > > states its optional when Nlabel is 1. It has zero functional use for > > interleave-set based namespaces even when the interleave-set-width is > > 1. So Linux takes the option to never set it, and goes further to > > reject it if it's set and the region-type does not match, because that > > follows the v1.1 meaning of the flag. > > > > [1]: > Thanks for the link! I'll read it. > BTW, it looks Hyper-V only supports PMEM namespace, at least so far. I don't think it should bother. It only makes sense for bare metal and even then I know of no NVDIMMs that are shipping it.