Re: [PATCH v8 1/1] mm: report per-page metadata information

[Pasha Tatashin]

On Thu, Feb 15, 2024 at 7:40 PM Suren Baghdasaryan <surenb@xxxxxxxxxx> wrote:
>
> On Thu, Feb 15, 2024 at 4:14 PM Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx> wrote:
> >
> > On Wed, 14 Feb 2024 14:57:40 -0800 Sourav Panda <souravpanda@xxxxxxxxxx> wrote:
> >
> > > Adds two new per-node fields, namely nr_memmap and nr_memmap_boot,
> > > to /sys/devices/system/node/nodeN/vmstat and a global Memmap field
> > > to /proc/meminfo. This information can be used by users to see how
> > > much memory is being used by per-page metadata, which can vary
> > > depending on build configuration, machine architecture, and system
> > > use.
> >
> > Would this information be available by the proposed memory
> > allocation profiling?
>
> Well, not without jumping through the hoops. You would need to find
> the places in the source code where all this matadata is allocated and
> find the appropriate records inside /proc/allocinfo file.

Another difference is that memory allocation profiling does not report
memblock allocated memory, this patch reports the amount of per-page
metadata allocated by the memblock allocator in addition to various
paths that are used to allocate per-page metdata after the buddy
allocator is initialized. Also, the memory profiling has page_ext
overhead, so it is not practical to unconditionally enable it
throughout the fleet.

Pasha