Re: [PATCH v11 00/21] Add support for NV-DIMMs to ext4

[Matthew Wilcox <willy@linux.intel.com>]

On Thu, Oct 16, 2014 at 09:39:08AM +0200, Mathieu Desnoyers wrote:
> First of all, thanks a lot for this patchset! Secondly, I must voice out
> that you really need to work on your marketing skills. What your
> changelog does not show is that this feature is tremendously useful
> *today* in the following use-case:
> 
> - On *any* platform for which you can teach the BIOS not to clear memory
>   on soft reboot,
> - Use a kernel argument to restrain it to portion of memory at boot
>   (e.g. 15GB out of 16GB),
> - Create an ext4 or ext2 filesystem in this available memory area,
> - Mount it with DAX flags,

Yes, I definitely suck at technical marketing.  I was thinking that
"NV-DIMMs" were the new hotness, and definitely available today, and
so advertising support for them was the best way to go.  I personally
do use your use case for testing DAX, but it didn't occur to me that
it would have real-world usages.

> >From there, you can do lots of interesting stuff. In my use-case, I
> would love to use it to mmap LTTng kernel/userspace tracer buffers, so
> we can extract them after a soft reboot and analyze a system crash.
> 
> My recommendation would be to rename this patchset as e.g.
> 
> "DAX: Page cache bypass for in-memory persistent filesystems"
> 
> which might attract more interest from reviewers and maintainers, since
> they can try it out today on commodity hardware. Also, pointing out to
> ext4 specifically in the patchset introduction title does not reflect
> the content accurately, since there is also ext2 implementation within
> the series.

Well ... ext2 already has the 'xip' implementation which probably works
well enough for enough of the time.  Most people probably won't hit the
races it has.

> One thing I would really like to see is a Documentation file that
> explains how to setup the kernel so it leaves a memory area free at the
> end of the physical address space, and how to setup a filesystem into
> it. Perhaps it already exists, in this case, pointing to it in the
> patchset introduction changelog would be helpful. (IOW, answering the
> question: how can someone test this today on commodity hardware ?).
> Also, if there are ways to setup pstore or such to achieve something
> similar of a wider range of systems, it would be nice to see
> documentation (or links to doc) explaining how to configure this.

I think that documentation properly belongs to the 'pmem' block driver that
Ross has been posting.  Here's 1/4, which contains some documentation,
but I think you're after something more detailed:

http://marc.info/?l=linux-fsdevel&m=140917398012020&w=2

> I'll try to review your patchset soon, however keeping in mind that it
> would be best to have mm experts having a look into it.

Yes, mm experts have many demands on their time, unfortunately :-(

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>