On Wed, Dec 17, 2025 at 4:14 AM <[email protected]> wrote: > > Michał Cłapiński wrote: > [..] > > > Sure, then make it 1280K of label space. There's no practical limit in > > > the implementation. > > > > Hi Dan, > > I just had the time to try this out. So I modified the code to > > increase the label space to 2M and I was able to create the > > namespaces. It put the metadata in volatile memory. > > > > But the infoblocks are still within the namespaces, right? If I try to > > create a 3G namespace with alignment set to 1G, its actual usable size > > is 2G. So I can't divide the whole pmem into 1G devices with 1G > > alignment. > > Ugh, yes, I failed to predict that outcome. > > > If I modify the code to remove the infoblocks, the namespace mode > > won't be persistent, right? In my solution I get that information from > > the kernel command line, so I don't need the infoblocks. > > So, I dislike the command line option ABI expansion proposal enough to > invest some time to find an alternative. One observation is that the > label is able to indicate the namespace mode independent of an > info-block. The info-block is only really needed when deciding whether > and how much space to reserve to allocate 'struct page' metadata. > > -- 8< -- > From 4f44cbb6e3bd4cac9481bdd4caf28975a4f1e471 Mon Sep 17 00:00:00 2001 > From: Dan Williams <[email protected]> > Date: Mon, 15 Dec 2025 17:10:04 -0800 > Subject: [PATCH] nvdimm: Allow fsdax and devdax namespace modes without > info-blocks > MIME-Version: 1.0 > Content-Type: text/plain; charset=UTF-8 > Content-Transfer-Encoding: 8bit > > Michał reports that the new ramdax facility does not meet his needs which > is to carve large reservations of memory into multiple 1GB aligned > namespaces/volumes. While ramdax solves the problem of in-memory > description of the volume layout, the nvdimm "infoblocks" eat capacity and > destroy alignment properties. > > The infoblock serves 2 purposes, it indicates whether the namespace should > operate in fsdax or devdax mode, Michał needs this, and it optionally > reserves namespace capacity for storing 'struct page' metadata, Michał does > not need this. It turns out the mode information is already recorded in the > namespace label, and if no reservation is needed for 'struct page' metadata > then infoblock settings can just be hard coded. > > Introduce a new ND_REGION_VIRT_INFOBLOCK flag for ramdax to indicate that > all infoblocks be synthesized and not consume any capacity from the > namespace. > > With that ramdax can create a full sized namespace: > > $ ndctl create-namespace -r region0 -s 1G -a 1G -M mem > { > "dev":"namespace0.0", > "mode":"fsdax", > "map":"mem", > "size":"1024.00 MiB (1073.74 MB)", > "uuid":"c48c4991-86af-4de1-8c7c-8919358df1f9", > "sector_size":512, > "align":1073741824, > "blockdev":"pmem0" > }
Thank you for working on this. I tried it an indeed it works with fsdax. It doesn't seem to work with devdax though. $ ndctl create-namespace -v -r region1 -m devdax -a 1G -M mem -s 1G libndctl: ndctl_dax_enable: dax1.0: failed to enable Error: namespace1.1: failed to enable failed to create namespace: No such device or address $ dmesg | grep dax [...] [ 29.504763] dax_pmem dax1.0: could not reserve metadata [ 29.504766] dax_pmem dax1.0: probe with driver dax_pmem failed with error -16 [ 29.506553] probe of dax1.0 returned 16 after 1815 usecs [...] I think the dax_pmem driver needs to be modified too.
