Re: [PATCH v11 1/8] mm/demotion: Add support for explicit memory tiers

[Huang, Ying]

Aneesh Kumar K V <aneesh.kumar@xxxxxxxxxxxxx> writes:

> On 8/4/22 6:26 AM, Huang, Ying wrote:
>> Aneesh Kumar K V <aneesh.kumar@xxxxxxxxxxxxx> writes:
>> 
>>> On 8/2/22 12:27 PM, Huang, Ying wrote:
>>>> Dan Williams <dan.j.williams@xxxxxxxxx> writes:
>>>>
>>>>> Huang, Ying wrote:
>>>>>> Dan Williams <dan.j.williams@xxxxxxxxx> writes:
>>>>>>
>>>>>>> Aneesh Kumar K.V wrote:
>>>>>>>> In the current kernel, memory tiers are defined implicitly via a demotion path
>>>>>>>> relationship between NUMA nodes, which is created during the kernel
>>>>>>>> initialization and updated when a NUMA node is hot-added or hot-removed. The
>>>>>>>> current implementation puts all nodes with CPU into the highest tier, and builds
>>>>>>>> the tier hierarchy tier-by-tier by establishing the per-node demotion targets
>>>>>>>> based on the distances between nodes.
>>>>>>>>
>>>>>>>> This current memory tier kernel implementation needs to be improved for several
>>>>>>>> important use cases,
>>>>>>>>
>>>>>>>> The current tier initialization code always initializes each memory-only NUMA
>>>>>>>> node into a lower tier. But a memory-only NUMA node may have a high performance
>>>>>>>> memory device (e.g. a DRAM-backed memory-only node on a virtual machine) that
>>>>>>>> should be put into a higher tier.
>>>>>>>>
>>>>>>>> The current tier hierarchy always puts CPU nodes into the top tier. But on a
>>>>>>>> system with HBM or GPU devices, the memory-only NUMA nodes mapping these devices
>>>>>>>> should be in the top tier, and DRAM nodes with CPUs are better to be placed into
>>>>>>>> the next lower tier.
>>>>>>>>
>>>>>>>> With current kernel higher tier node can only be demoted to nodes with shortest
>>>>>>>> distance on the next lower tier as defined by the demotion path, not any other
>>>>>>>> node from any lower tier. This strict, demotion order does not work in all use
>>>>>>>> cases (e.g. some use cases may want to allow cross-socket demotion to another
>>>>>>>> node in the same demotion tier as a fallback when the preferred demotion node is
>>>>>>>> out of space), This demotion order is also inconsistent with the page allocation
>>>>>>>> fallback order when all the nodes in a higher tier are out of space: The page
>>>>>>>> allocation can fall back to any node from any lower tier, whereas the demotion
>>>>>>>> order doesn't allow that.
>>>>>>>>
>>>>>>>> This patch series address the above by defining memory tiers explicitly.
>>>>>>>>
>>>>>>>> Linux kernel presents memory devices as NUMA nodes and each memory device is of
>>>>>>>> a specific type. The memory type of a device is represented by its abstract
>>>>>>>> distance. A memory tier corresponds to a range of abstract distance. This allows
>>>>>>>> for classifying memory devices with a specific performance range into a memory
>>>>>>>> tier.
>>>>>>>>
>>>>>>>> This patch configures the range/chunk size to be 128. The default DRAM
>>>>>>>> abstract distance is 512. We can have 4 memory tiers below the default DRAM
>>>>>>>> abstract distance which cover the range 0 - 127, 127 - 255, 256- 383, 384 - 511.
>>>>>>>> Slower memory devices like persistent memory will have abstract distance below
>>>>>>>> the default DRAM level and hence will be placed in these 4 lower tiers.
>>>>>>>>
>>>>>>>> A kernel parameter is provided to override the default memory tier.
>>>>>>>>
>>>>>>>> Link: https://lore.kernel.org/linux-mm/CAAPL-u9Wv+nH1VOZTj=9p9S70Y3Qz3+63EkqncRDdHfubsrjfw@xxxxxxxxxxxxxx
>>>>>>>> Link: https://lore.kernel.org/linux-mm/7b72ccf4-f4ae-cb4e-f411-74d055482026@xxxxxxxxxxxxx
>>>>>>>>
>>>>>>>> Signed-off-by: Jagdish Gediya <jvgediya@xxxxxxxxxxxxx>
>>>>>>>> Signed-off-by: Aneesh Kumar K.V <aneesh.kumar@xxxxxxxxxxxxx>
>>>>>>>> ---
>>>>>>>>  include/linux/memory-tiers.h |  17 ++++++
>>>>>>>>  mm/Makefile                  |   1 +
>>>>>>>>  mm/memory-tiers.c            | 102 +++++++++++++++++++++++++++++++++++
>>>>>>>>  3 files changed, 120 insertions(+)
>>>>>>>>  create mode 100644 include/linux/memory-tiers.h
>>>>>>>>  create mode 100644 mm/memory-tiers.c
>>>>>>>>
>>>>>>>> diff --git a/include/linux/memory-tiers.h b/include/linux/memory-tiers.h
>>>>>>>> new file mode 100644
>>>>>>>> index 000000000000..8d7884b7a3f0
>>>>>>>> --- /dev/null
>>>>>>>> +++ b/include/linux/memory-tiers.h
>>>>>>>> @@ -0,0 +1,17 @@
>>>>>>>> +/* SPDX-License-Identifier: GPL-2.0 */
>>>>>>>> +#ifndef _LINUX_MEMORY_TIERS_H
>>>>>>>> +#define _LINUX_MEMORY_TIERS_H
>>>>>>>> +
>>>>>>>> +/*
>>>>>>>> + * Each tier cover a abstrace distance chunk size of 128
>>>>>>>> + */
>>>>>>>> +#define MEMTIER_CHUNK_BITS	7
>>>>>>>> +#define MEMTIER_CHUNK_SIZE	(1 << MEMTIER_CHUNK_BITS)
>>>>>>>> +/*
>>>>>>>> + * For now let's have 4 memory tier below default DRAM tier.
>>>>>>>> + */
>>>>>>>> +#define MEMTIER_ADISTANCE_DRAM	(1 << (MEMTIER_CHUNK_BITS + 2))
>>>>>>>> +/* leave one tier below this slow pmem */
>>>>>>>> +#define MEMTIER_ADISTANCE_PMEM	(1 << MEMTIER_CHUNK_BITS)
>>>>>>>
>>>>>>> Why is memory type encoded in these values? There is no reason to
>>>>>>> believe that PMEM is of a lower performance tier than DRAM. Consider
>>>>>>> high performance energy backed DRAM that makes it "PMEM", consider CXL
>>>>>>> attached DRAM over a switch topology and constrained links that makes it
>>>>>>> a lower performance tier than locally attached DRAM. The names should be
>>>>>>> associated with tiers that indicate their usage. Something like HOT,
>>>>>>> GENERAL, and COLD. Where, for example, HOT is low capacity high
>>>>>>> performance compared to the general purpose pool, and COLD is high
>>>>>>> capacity low performance intended to offload the general purpose tier.
>>>>>>>
>>>>>>> It does not need to be exactly that ontology, but please try to not
>>>>>>> encode policy meaning behind memory types. There has been explicit
>>>>>>> effort to avoid that to date because types are fraught for declaring
>>>>>>> relative performance characteristics, and the relative performance
>>>>>>> changes based on what memory types are assembled in a given system.
>>>>>>
>>>>>> Yes.  MEMTIER_ADISTANCE_PMEM is something over simplified.  That is only
>>>>>> used in this very first version to make it as simple as possible.  
>>>>>
>>>>> I am failing to see the simplicity of using names that convey a
>>>>> performance contract that are invalid depending on the system.
>>>>>
>>>>>> I think we can come up with something better in the later version.
>>>>>> For example, identify the abstract distance of a PMEM device based on
>>>>>> HMAT, etc. 
>>>>>
>>>>> Memory tiering has nothing to do with persistence why is PMEM in the
>>>>> name at all?
>>>>>
>>>>>>  And even in this first version, we should put MEMTIER_ADISTANCE_PMEM
>>>>>>  in dax/kmem.c.  Because it's just for that specific type of memory
>>>>>>  used now, not for all PMEM.
>>>>>
>>>>> dax/kmem.c also handles HBM and "soft reserved" memory in general. There
>>>>> is also nothing PMEM specific about the device-dax subsystem.
>>>>
>>>> Ah... I see the issue here.  For the systems in our hand, dax/kmem.c is
>>>> used to online PMEM only.  Even the "soft reserved" memory is used for
>>>> PMEM or simulating PMEM too.  So to make the code as simple as possible,
>>>> we treat all memory devices onlined by dax/kmem as PMEM in the first
>>>> version.  And plan to support more memory types in the future versions.
>>>>
>>>> But from your above words, our assumption are wrong here.  dax/kmem.c
>>>> can online HBM and other memory devices already.  If so, how do we
>>>> distinguish between them and how to get the performance character of
>>>> these devices?  We can start with SLIT?
>>>>
>>>
>>> We would let low level driver register memory_dev_types for the NUMA nodes
>>> that will be mapped to these devices. ie, a papr_scm, ACPI NFIT or CXL
>>> can register different memory_dev_type based on device tree, HMAT or CDAT. 
>> 
>> I didn't find ACPI NFIT can provide any performance information, just
>> whether it's non-volatile.  HMAT or CDAT should help here, but it's not
>> available always.  For now, what we have is just SLIT at least for quite
>> some machines.
>> 
>
>
> The lower level driver that is creating the nvdimm regions can assign a
> memory type to the numa node which it associates with the region. For now,
> drivers like papr_scm do that on ppc64. When it associates a numa node to
> nvdimm regions, it can query every detail available (device tree
> in case of papr_scm, can be HMAT/SLIT or CDAT) to associate the NUMA node
> to a memory type. 

If we have only one information source, it's OK to create all memory
type with this source.  But if we have multiple sources, we need a
mechanism to coordinate among these sources.  It gives us good
flexibility to create memory types in driver.  Because drivers can
use any information sources.

Best Regards,
Huang, Ying

>> I prefer to create memory_dev_type in high level driver like dax/kmem.
>> And it may query low level driver like SLIT, HMAT, CDAT, etc for more
>> information based on availability etc.
>> 
>> Best Regards,
>> Huang, Ying