On Thu, Mar 21, 2019 at 10:09 PM Dexuan Cui <decui(a)microsoft.com> wrote:
> From: Dan Williams <dan.j.williams(a)intel.com>
> Sent: Thursday, March 21, 2019 9:13 PM
> > ...
> > Actually, this _is_ an issue for NVDIMM_FAMILY_HYPERV (and the other
> > families except for NVDIMM_FAMILY_INTEL) : see the kernel function
> > acpi_nfit_register_dimms(), where ND_CMD_SMART is set in the
> > "cmd_mask" only for NVDIMM_FAMILY_INTEL.
> >
> > So, on Hyper-V, ndctl_dimm_is_cmd_supported(dimm, ND_CMD_SMART)
> > is always false, and "ndctl monitor" exits with "no smart
support".
>
> Can the Hyper-V implementation emulate those commands? That's the
> expectation, i.e. that the implementation can return they required
> payloads, but it's fine if the payloads disclaim support for certain
> fields.
Hyper-V Virtual NVDIMM doesn't emulate ND_CMD_SMART(1).
The _DSM Function 1 on Hyper-V is "Get Health Information" [1], which is
incomptabile with NVDIMM_FAMILY_INTEL's ND_CMD_SMART(1).
Even if Hyper-V could emulate ND_CMD_SMART, the current "ndctl monitor"
code still wouldn't work for Hyper-V (and the other families): the essence of the
issue is that:
1. the kernel only sets ND_CMD_SMART flag for NVDIMM_FAMILY_INTEL.
2. "ndctl monitor" assumes the ND_CMD_SMART should be set by checking
/sys/class/nd/ndctl0/device/nmem0/commands.
This means ndctl_dimm_is_cmd_supported(dimm, ND_CMD_SMART) can
only be true on NVDIMM_FAMILY_INTEL.
My patch skips the assumption of ndctl on Hyper-V, so "ndctl monitor"
can work on Hyper-V.
It looks we do need an ops->monitor_supported() in ndctl?
No, I think you misunderstand. Hyper-V implements "function-1",
"command-1" support can be emulated. The request is to translate the
Hyper-V function-1 payload into the command-1 payload format.