Re: [PATCH 0/7] ACPI HMAT memory sysfs representation

From: Dan Williams <dan.j.williams@intel.com>
To: anshuman.khandual@arm.com
Cc: Dave Hansen <dave.hansen@intel.com>,
	Keith Busch <keith.busch@intel.com>,
	Linux Kernel Mailing List <linux-kernel@vger.kernel.org>,
	Linux ACPI <linux-acpi@vger.kernel.org>,
	Linux MM <linux-mm@kvack.org>,
	Greg KH <gregkh@linuxfoundation.org>,
	"Rafael J. Wysocki" <rafael@kernel.org>
Subject: Re: [PATCH 0/7] ACPI HMAT memory sysfs representation
Date: Tue, 27 Nov 2018 08:56:09 -0800	[thread overview]
Message-ID: <CAPcyv4iA+yA_OZMUSbJTV=c9=QsRAcLjzLK94uuVsMTaPu6eTg@mail.gmail.com> (raw)
In-Reply-To: <325d0e69-053a-ae9c-eede-7cdf28b1dbd6@arm.com>

On Tue, Nov 27, 2018 at 2:15 AM Anshuman Khandual
<anshuman.khandual@arm.com> wrote:
>
>
>
> On 11/26/2018 11:38 PM, Dan Williams wrote:
> > On Mon, Nov 26, 2018 at 8:42 AM Dave Hansen <dave.hansen@intel.com> wrote:
> >>
> >> On 11/23/18 1:13 PM, Dan Williams wrote:
> >>>> A new system call makes total sense to me.  I have the same concern
> >>>> about the completeness of what's exposed in sysfs, I just don't see a
> >>>> _route_ to completeness with sysfs itself.  Thus, the minimalist
> >>>> approach as a first step.
> >>> Outside of platform-firmware-id to Linux-numa-node-id what other
> >>> userspace API infrastructure does the kernel need to provide? It seems
> >>> userspace enumeration of memory attributes is fully enabled once the
> >>> firmware-to-Linux identification is established.
> >>
> >> It would be nice not to have each app need to know about each specific
> >> platform's firmware.
> >
> > The app wouldn't need to know if it uses a common library. Whether the
> > library calls into the kernel or not is an implementation detail. If
> > it is information that only the app cares about and the kernel does
> > not consume, why have a syscall?
>
> If we just care about platform-firmware-id <--> Linux-numa-node-id mapping
> and fetching memory attribute from the platform (and hiding implementation
> details in a library) then the following interface should be sufficient.
>
> /sys/devices/system/node/nodeX/platform_id
>
> But as the series proposes (and rightly so) kernel needs to start providing
> ABI interfaces for memory attributes instead of hiding them in libraries.

Yes, I can get on board with sysfs providing a subset of the
performance description for administrators to discover the common case
via scripting and leave the exhaustive attribute description to a
separate interface. I was pushing back on the notion that sysfs must
be that exhaustive interface... we're making progress.

I still think we need /sys/devices/system/node/nodeX/platform_id to
enable higher order platform enumeration tooling, but that need not be
the end of the kernel interface description.