Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
Subject: Re: [PATCH 0/7] ACPI HMAT memory sysfs representation
To:     Anshuman Khandual <anshuman.khandual@arm.com>,
        Keith Busch <keith.busch@intel.com>,
        linux-kernel@vger.kernel.org, linux-acpi@vger.kernel.org,
        linux-mm@kvack.org
Cc:     Greg Kroah-Hartman <gregkh@linuxfoundation.org>,
        Rafael Wysocki <rafael@kernel.org>,
        Dan Williams <dan.j.williams@intel.com>
References: <20181114224902.12082-1-keith.busch@intel.com>
 <1ed406b2-b85f-8e02-1df0-7c39aa21eca9@arm.com>
 <4ea6e80f-80ba-6992-8aa0-5c2d88996af7@intel.com>
 <b79804b0-32ee-03f9-fa62-a89684d46be6@arm.com>
 <c6abb754-0d82-8739-fe08-24e9402bae75@intel.com>
 <aae34dde-fa70-870a-9b74-fff9e385bfc9@arm.com>
 <f5315662-5c1a-68a3-4d04-21b4b5ca94b1@intel.com>
 <ac942498-8966-6a9b-0e55-c79ae167c679@arm.com>
 <9015e51a-3584-7bb2-cc5e-25b0ec8e5494@intel.com>
 <1a9e887b-8087-e897-6195-e8df325bd458@arm.com>
From:   Dave Hansen <dave.hansen@intel.com>
Openpgp: preference=signencrypt
Autocrypt: addr=dave.hansen@intel.com; keydata=
 xsFNBE6HMP0BEADIMA3XYkQfF3dwHlj58Yjsc4E5y5G67cfbt8dvaUq2fx1lR0K9h1bOI6fC
 oAiUXvGAOxPDsB/P6UEOISPpLl5IuYsSwAeZGkdQ5g6m1xq7AlDJQZddhr/1DC/nMVa/2BoY
 2UnKuZuSBu7lgOE193+7Uks3416N2hTkyKUSNkduyoZ9F5twiBhxPJwPtn/wnch6n5RsoXsb
 ygOEDxLEsSk/7eyFycjE+btUtAWZtx+HseyaGfqkZK0Z9bT1lsaHecmB203xShwCPT49Blxz
 VOab8668QpaEOdLGhtvrVYVK7x4skyT3nGWcgDCl5/Vp3TWA4K+IofwvXzX2ON/Mj7aQwf5W
 iC+3nWC7q0uxKwwsddJ0Nu+dpA/UORQWa1NiAftEoSpk5+nUUi0WE+5DRm0H+TXKBWMGNCFn
 c6+EKg5zQaa8KqymHcOrSXNPmzJuXvDQ8uj2J8XuzCZfK4uy1+YdIr0yyEMI7mdh4KX50LO1
 pmowEqDh7dLShTOif/7UtQYrzYq9cPnjU2ZW4qd5Qz2joSGTG9eCXLz5PRe5SqHxv6ljk8mb
 ApNuY7bOXO/A7T2j5RwXIlcmssqIjBcxsRRoIbpCwWWGjkYjzYCjgsNFL6rt4OL11OUF37wL
 QcTl7fbCGv53KfKPdYD5hcbguLKi/aCccJK18ZwNjFhqr4MliQARAQABzShEYXZpZCBDaHJp
 c3RvcGhlciBIYW5zZW4gPGRhdmVAc3I3MS5uZXQ+wsF7BBMBAgAlAhsDBgsJCAcDAgYVCAIJ
 CgsEFgIDAQIeAQIXgAUCTo3k0QIZAQAKCRBoNZUwcMmSsMO2D/421Xg8pimb9mPzM5N7khT0
 2MCnaGssU1T59YPE25kYdx2HntwdO0JA27Wn9xx5zYijOe6B21ufrvsyv42auCO85+oFJWfE
 K2R/IpLle09GDx5tcEmMAHX6KSxpHmGuJmUPibHVbfep2aCh9lKaDqQR07gXXWK5/yU1Dx0r
 VVFRaHTasp9fZ9AmY4K9/BSA3VkQ8v3OrxNty3OdsrmTTzO91YszpdbjjEFZK53zXy6tUD2d
 e1i0kBBS6NLAAsqEtneplz88T/v7MpLmpY30N9gQU3QyRC50jJ7LU9RazMjUQY1WohVsR56d
 ORqFxS8ChhyJs7BI34vQusYHDTp6PnZHUppb9WIzjeWlC7Jc8lSBDlEWodmqQQgp5+6AfhTD
 kDv1a+W5+ncq+Uo63WHRiCPuyt4di4/0zo28RVcjtzlGBZtmz2EIC3vUfmoZbO/Gn6EKbYAn
 rzz3iU/JWV8DwQ+sZSGu0HmvYMt6t5SmqWQo/hyHtA7uF5Wxtu1lCgolSQw4t49ZuOyOnQi5
 f8R3nE7lpVCSF1TT+h8kMvFPv3VG7KunyjHr3sEptYxQs4VRxqeirSuyBv1TyxT+LdTm6j4a
 mulOWf+YtFRAgIYyyN5YOepDEBv4LUM8Tz98lZiNMlFyRMNrsLV6Pv6SxhrMxbT6TNVS5D+6
 UorTLotDZKp5+M7BTQRUY85qARAAsgMW71BIXRgxjYNCYQ3Xs8k3TfAvQRbHccky50h99TUY
 sqdULbsb3KhmY29raw1bgmyM0a4DGS1YKN7qazCDsdQlxIJp9t2YYdBKXVRzPCCsfWe1dK/q
 66UVhRPP8EGZ4CmFYuPTxqGY+dGRInxCeap/xzbKdvmPm01Iw3YFjAE4PQ4hTMr/H76KoDbD
 cq62U50oKC83ca/PRRh2QqEqACvIH4BR7jueAZSPEDnzwxvVgzyeuhwqHY05QRK/wsKuhq7s
 UuYtmN92Fasbxbw2tbVLZfoidklikvZAmotg0dwcFTjSRGEg0Gr3p/xBzJWNavFZZ95Rj7Et
 db0lCt0HDSY5q4GMR+SrFbH+jzUY/ZqfGdZCBqo0cdPPp58krVgtIGR+ja2Mkva6ah94/oQN
 lnCOw3udS+Eb/aRcM6detZr7XOngvxsWolBrhwTQFT9D2NH6ryAuvKd6yyAFt3/e7r+HHtkU
 kOy27D7IpjngqP+b4EumELI/NxPgIqT69PQmo9IZaI/oRaKorYnDaZrMXViqDrFdD37XELwQ
 gmLoSm2VfbOYY7fap/AhPOgOYOSqg3/Nxcapv71yoBzRRxOc4FxmZ65mn+q3rEM27yRztBW9
 AnCKIc66T2i92HqXCw6AgoBJRjBkI3QnEkPgohQkZdAb8o9WGVKpfmZKbYBo4pEAEQEAAcLB
 XwQYAQIACQUCVGPOagIbDAAKCRBoNZUwcMmSsJeCEACCh7P/aaOLKWQxcnw47p4phIVR6pVL
 e4IEdR7Jf7ZL00s3vKSNT+nRqdl1ugJx9Ymsp8kXKMk9GSfmZpuMQB9c6io1qZc6nW/3TtvK
 pNGz7KPPtaDzvKA4S5tfrWPnDr7n15AU5vsIZvgMjU42gkbemkjJwP0B1RkifIK60yQqAAlT
 YZ14P0dIPdIPIlfEPiAWcg5BtLQU4Wg3cNQdpWrCJ1E3m/RIlXy/2Y3YOVVohfSy+4kvvYU3
 lXUdPb04UPw4VWwjcVZPg7cgR7Izion61bGHqVqURgSALt2yvHl7cr68NYoFkzbNsGsye9ft
 M9ozM23JSgMkRylPSXTeh5JIK9pz2+etco3AfLCKtaRVysjvpysukmWMTrx8QnI5Nn5MOlJj
 1Ov4/50JY9pXzgIDVSrgy6LYSMc4vKZ3QfCY7ipLRORyalFDF3j5AGCMRENJjHPD6O7bl3Xo
 4DzMID+8eucbXxKiNEbs21IqBZbbKdY1GkcEGTE7AnkA3Y6YB7I/j9mQ3hCgm5muJuhM/2Fr
 OPsw5tV/LmQ5GXH0JQ/TZXWygyRFyyI2FqNTx4WHqUn3yFj8rwTAU1tluRUYyeLy0ayUlKBH
 ybj0N71vWO936MqP6haFERzuPAIpxj2ezwu0xb1GjTk4ynna6h5GjnKgdfOWoRtoWndMZxbA
 z5cecg==
Message-ID: <3b86c5c5-53f2-29bf-48e7-5749c7287dca@intel.com>
Date:   Mon, 26 Nov 2018 09:20:04 -0800
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:60.0) Gecko/20100101
 Thunderbird/60.2.1
MIME-Version: 1.0
In-Reply-To: <1a9e887b-8087-e897-6195-e8df325bd458@arm.com>
Content-Type: text/plain; charset=utf-8
Content-Language: en-US
Content-Transfer-Encoding: 8bit
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On 11/26/18 7:38 AM, Anshuman Khandual wrote:
> On 11/24/2018 12:51 AM, Dave Hansen wrote:
>> On 11/22/18 10:42 PM, Anshuman Khandual wrote:
>>> Are we willing to go in the direction for inclusion of a new system
>>> call, subset of it appears on sysfs etc ? My primary concern is not
>>> how the attribute information appears on the sysfs but lack of it's
>>> completeness.
>>
>> 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.
> 
> Okay if we agree on the need for a new specific system call extracting
> the superset attribute information MAX_NUMNODES * MAX_NUMNODES * U64
> (u64 packs 8 bit values for 8 attributes or something like that) as we
> had discussed before, it makes sense to export a subset of it which can
> be faster but useful for the user space without going through a system
> call. 

The information that needs to be exported is a bit more than that.  It's
not just a binary attribute.

The information we have from the new ACPI table, for instance, is the
read and write bandwidth and latency between two nodes.  They are, IIRC,
two-byte values in the ACPI table[1], each.  That's 8 bytes worth of
data right there, which wouldn't fit *anything* else.

The list of things we want to export will certainly grow.  That means we
need a syscall something like this:

int get_mem_attribute(unsigned long attribute_nr,
		      unsigned long __user * initiator_nmask,
		      unsigned long __user * target_nmask,
		      unsigned long maxnode,
		      unsigned long *attributes_out);

#define MEM_ATTR_READ_BANDWIDTH		1
#define MEM_ATTR_WRITE_BANDWIDTH	2
#define MEM_ATTR_READ_LATENCY		3
#define MEM_ATTR_WRITE_LATENCTY		4
#define MEM_ATTR_ENCRYPTION		5

If you want to know the read latency between nodes 4 and 8, you do:

	ret = get_mem_attr(MEM_ATTR_READ_LATENCY,
			   (1<<4), (1<<8), max, &array);

And the answer shows up at array[0] in this example.  If you had more
than one bit set in the two nmasks, you would have a longer array.

The length of the array is the number of bits set in initiator_nmask *
the number of bits set in target_nmask * sizeof(ulong).

This has the advantage of supporting ULONG_MAX attributes, and scales
from asking for one attribute at a time all the way up to dumping the
entire system worth of data for a single attribute.  The only downside
is that it's one syscall per attribute instead of packing them all
together.  But, if we have a small enough number to pack them in one
ulong, then I think we can make 64 syscalls without too much trouble.

> Do you agree on a (system call + sysfs) approach in principle ?
> Also sysfs exported information has to be derived from whats available
> through the system call not the other way round. Hence the starting
> point has to be the system call definition.

Both the sysfs information *and* what will be exported in any future
interfaces are derived from platform-specific information.  They are not
derived from one _interface_ or the other.

They obviously need to be consistent, though.

1. See "Table 5-142 System Locality Latency and Bandwidth Information
Structure" here:
http://www.uefi.org/sites/default/files/resources/ACPI_6_2.pdf