6 a.m.
On 09/09/2016 11:40 PM, Dan Williams wrote:
On Fri, Sep 9, 2016 at 1:55 AM, Xiao Guangrong
<guangrong.xiao(a)linux.intel.com> wrote:
[..]
>>
>> Whether a persistent memory mapping requires an msync/fsync is a
>> filesystem specific question. This mincore proposal is separate from
>> that. Consider device-DAX for volatile memory or mincore() called on
>> an anonymous memory range. In those cases persistence and filesystem
>> metadata are not in the picture, but it would still be useful for
>> userspace to know "is there page cache backing this mapping?" or
"what
>> is the TLB geometry of this mapping?".
>
>
> I got a question about msync/fsync which is beyond the topic of this thread
> :)
>
> Whether msync/fsync can make data persistent depends on ADR feature on
> memory
> controller, if it exists everything works well, otherwise, we need to have
> another
> interface that is why 'Flush hint table' in ACPI comes in. 'Flush hint
> table' is
> particularly useful for nvdimm virtualization if we use normal memory to
> emulate
> nvdimm with data persistent characteristic (the data will be flushed to a
> persistent storage, e.g, disk).
>
> Does current PMEM programming model fully supports 'Flush hint table'? Is
> userspace allowed to use these addresses?
If you publish flush hint addresses in the virtual NFIT the guest VM
will write to them whenever a REQ_FLUSH or REQ_FUA request is sent to
the virtual /dev/pmemX device. Yes, seems straightforward to take a
VM exit on those events and flush simulated pmem to persistent
storage.
Thank you, Dan!
However REQ_FLUSH or REQ_FUA is handled in kernel space, okay, after following
up the discussion in this thread, i understood that currently filesystems have
not supported the case that usespace itself make data be persistent without
kernel's involvement. So that works.
Hmm, Does device-DAX support this case (make data be persistent without
msync/fsync)? I guess no, but just want to confirm it. :)