Original Source: 

https://github.com/torvalds/linux/blob/master/Documentation/nvdimm/nvdimm.txt

 LIBNVDIMM: Non-Volatile Devices

     libnvdimm - kernel / libndctl - userspace helper library

• Glossary
• Overview
•    Supporting Documents
• LIBNVDIMM PMEM and BLK

• Why BLK?
•    PMEM vs BLK
•        BLK-REGIONs, PMEM-REGIONs, Atomic Sectors, and DAX
• Example NVDIMM Platform
• LIBNVDIMM Kernel Device Model and LIBNDCTL Userspace API
•    LIBNDCTL: Context
•        libndctl: instantiate a new library context example
•    LIBNVDIMM/LIBNDCTL: Bus
•        libnvdimm: control class device in /sys/class
•        libnvdimm: bus
•        libndctl: bus enumeration example
•    LIBNVDIMM/LIBNDCTL: DIMM (NMEM)
•        libnvdimm: DIMM (NMEM)
•        libndctl: DIMM enumeration example
•    LIBNVDIMM/LIBNDCTL: Region
•        libnvdimm: region
•        libndctl: region enumeration example
•        Why Not Encode the Region Type into the Region Name?
•        How Do I Determine the Major Type of a Region?
•    LIBNVDIMM/LIBNDCTL: Namespace
•        libnvdimm: namespace
•        libndctl: namespace enumeration example
•        libndctl: namespace creation example
•        Why the Term "namespace"?
•    LIBNVDIMM/LIBNDCTL: Block Translation Table "btt"
•        libnvdimm: btt layout
•        libndctl: btt creation example
• Summary LIBNDCTL Diagram

Glossary

• PMEM: A system-physical-address range where writes are persistent. A block device composed of PMEM is capable of DAX. A PMEM address range may span an interleave of several DIMMs.

• BLK: A set of one or more programmable memory mapped apertures provided by a DIMM to access its media. This indirection precludes the performance benefit of interleaving, but enables DIMM-bounded failure modes.
• DPA: DIMM Physical Address, is a DIMM-relative offset.  With one DIMM in the system there would be a 1:1 system-physical-address: DPA association. Once more DIMMs are added a memory controller interleave must be decoded to determine the DPA associated with a given system-physical-address.  BLK capacity always has a 1:1 relationship with a single-DIMM's DPA range.
• DAX: File system extensions to bypass the page cache and block layer to mmap persistent memory, from a PMEM block device, directly into a process address space.
• DSM: Device Specific Method: ACPI method to to control specific device - in this case the firmware.

• DCR: NVDIMM Control Region Structure defined in ACPI 6 Section 5.2.25.5. It defines a vendor-id, device-id, and interface format for a given DIMM.
• BTT: Block Translation Table: Persistent memory is byte addressable. Existing software may have an expectation that the power-fail-atomicity of writes is at least one sector, 512 bytes.  The BTT is an indirection table with atomic update semantics to front a PMEM/BLK block device driver and present arbitrary atomic sector sizes.
• LABEL: Metadata stored on a DIMM device that partitions and identifies (persistently names) storage between PMEM and BLK.  It also partitions BLK storage to host BTTs with different parameters per BLK-partition.
• Note that traditional partition tables, GPT/MBR, are layered on top of a BLK or PMEM device.

Overview

The LIBNVDIMM subsystem provides support for three types of NVDIMMs, namely, PMEM, BLK, and NVDIMM devices that can simultaneously support both PMEM and BLK mode access.  These three modes of operation are described by the "NVDIMM Firmware Interface Table" (NFIT) in ACPI 6.  While the LIBNVDIMM implementation is generic and supports pre-NFIT platforms, it was guided by the superset of capabilities need to support this ACPI 6 definition for NVDIMM resources. The bulk of the kernel implementation is in place to handle the case where DPA accessible via PMEM is aliased with DPA accessible via BLK.  When that occurs a LABEL is needed to reserve DPAfor exclusive access via one mode a time.

Supporting Documents

ACPI 6: http://www.uefi.org/sites/default/files/resources/ACPI_6.0.pdf

NVDIMM Namespace: http://pmem.io/documents/NVDIMM_Namespace_Spec.pdf

DSM Interface Example: http://pmem.io/documents/NVDIMM_DSM_Interface_Example.pdf

Driver Writer's Guide: http://pmem.io/documents/NVDIMM_Driver_Writers_Guide.pdf

Git Trees

LIBNVDIMM: https://git.kernel.org/cgit/linux/kernel/git/djbw/nvdimm.git

LIBNDCTL: https://github.com/pmem/ndctl.git

PMEM: https://github.com/01org/prd

LIBNVDIMM PMEM and BLK

Prior to the arrival of the NFIT, non-volatile memory was described to a system in various ad-hoc ways.  Usually only the bare minimum was provided, namely, a single system-physical-address range where writes are expected to be durable after a system power loss.  Now, the NFIT specification standardizes not only the description of PMEM, but also BLK and platform message-passing entry points for control and configuration.

For each NVDIMM access method (PMEM, BLK), LIBNVDIMM provides a block device driver:

1. PMEM (nd_pmem.ko): Drives a system-physical-address range. This range is contiguous in system memory and may be interleaved (hardware memory controller striped) across multiple DIMMs.  When interleaved the platform may optionally provide details of which DIMMs are participating in the interleave.

Note that while LIBNVDIMM describes system-physical-address ranges that may alias with BLK access as ND_NAMESPACE_PMEM ranges and those without alias as ND_NAMESPACE_IO ranges, to the nd_pmem driver there is no distinction.  The different device-types are an implementation detail that userspace can exploit to implement policies like "only interface with address ranges from certain DIMMs".  It is worth noting that when aliasing is present and a DIMM lacks a label, then no block device can be created by default as userspace needs to do at least one allocation of DPA to the PMEM range. In contrast ND_NAMESPACE_IO ranges, once registered, can be immediately attached to nd_pmem.

2. BLK (nd_blk.ko): This driver performs I/O using a set of platform defined apertures.  A set of apertures will access just one DIMM. Multiple windows (apertures) allow multiple concurrent accesses, much like tagged-command-queuing, and would likely be used by different threads or different CPUs.

The NFIT specification defines a standard format for a BLK-aperture, but the spec also allows for vendor specific layouts, and non-NFIT BLK implementations may have other designs for BLK I/O.  For this reason "nd_blk" calls back into platform-specific code to perform the I/O. One such implementation is defined in the "Driver Writer's Guide" and "DSM Interface Example".

Why BLK?

While PMEM provides direct byte-addressable CPU-load/store access to NVDIMM storage, it does not provide the best system RAS (recovery, availability, and serviceability) model.  An access to a corrupted system-physical-address address causes a CPU exception while an access to a corrupted address through an BLK-aperture causes that block window to raise an error status in a register. The latter is more aligned withthe standard error model that host-bus-adapter attached disks present. Also, if an administrator ever wants to replace a memory it is easier to service a system at DIMM module boundaries.  Compare this to PMEM where data could be interleaved in an opaque hardware specific manner across several DIMMs.

PMEM vs BLK

BLK-apertures solve these RAS problems, but their presence is also the major contributing factor to the complexity of the ND subsystem. They complicate the implementation because PMEM and BLK alias in DPA space. Any given DIMM's DPA-range may contribute to one or more system-physical-address sets of interleaved DIMMs, *and* may also be accessed in its entirety through its BLK-aperture. Accessing a DPA through a system-physical-address while simultaneously accessing the same DPA through a BLK-aperture has undefined results. For this reason, DIMMs with this dual interface configuration include a DSM function to store/retrieve a LABEL.  The LABEL effectively partitions the DPA-space into exclusive system-physical-address and BLK-aperture accessible regions.  For simplicity a DIMM is allowed a PMEM "region" per each interleave set in which it is a member. The remaining DPA space can be carved into an arbitrary number of BLK devices with discontiguous extents.

BLK-REGIONs, PMEM-REGIONs, Atomic Sectors, and DAX

One of the few reasons to allow multiple BLK namespaces per REGION is so that each BLK-namespace can be configured with a BTT with unique atomic sector sizes.  While a PMEM device can host a BTT the LABEL specification does not provide for a sector size to be specified for a PMEM namespace. This is due to the expectation that the primary usage model for PMEM is via DAX, and the BTT is incompatible with DAX.  However, for the cases where an application or filesystem still needs atomic sector updateguarantees it can register a BTT on a PMEM device or partition. See LIBNVDIMM/NDCTL: Block Translation Table "btt"

LIBNVDIMM Kernel Device Model and LIBNDCTL Userspace API

What follows is a description of the LIBNVDIMM sysfs layout and a corresponding object hierarchy diagram as viewed through the LIBNDCTL API.  The example sysfs paths and diagrams are relative to the Example NVDIMM Platform which is also the LIBNVDIMM bus used in the LIBNDCTL unit test.

LIBNDCTL: Context Every API call in the LIBNDCTL library requires a context that holds the logging parameters and other library instance state.  The library is based on the libabc template:

https://git.kernel.org/cgit/linux/kernel/git/kay/libabc.git

LIBNDCTL: instantiate a new library context example

struct ndctl_ctx *ctx;

if (ndctl_new(&ctx) == 0)

return ctx;

else

return NULL;

LIBNVDIMM/LIBNDCTL: Bus

A bus has a 1:1 relationship with an NFIT.  The current expectation for ACPI based systems is that there is only ever one platform-global NFIT. That said, it is trivial to register multiple NFITs, the specification does not preclude it.  The infrastructure supports multiple buses and we we use this capability to test multiple NFIT configurations in the unit test.

LIBNVDIMM: control class device in /sys/class

This character device accepts DSM messages to be passed to DIMM identified by its NFIT handle.

/sys/class/nd/ndctl0

|-- dev

|-- device -> ../../../ndbus0

|-- subsystem -> ../../../../../../../class/nd

LIBNVDIMM: bus

struct nvdimm_bus *nvdimm_bus_register(struct device *parent,

      struct nvdimm_bus_descriptor *nfit_desc);

LIBNDCTL: bus enumeration example. Find the bus handle that describes the bus from Example NVDIMM Platform

static struct ndctl_bus *get_bus_by_provider(struct ndctl_ctx *ctx,

const char *provider)

{

struct ndctl_bus *bus;

ndctl_bus_foreach(ctx, bus)

if (strcmp(provider, ndctl_bus_get_provider(bus)) == 0)

return bus;

return NULL;

}

bus = get_bus_by_provider(ctx, "nfit_test.0");

LIBNVDIMM/LIBNDCTL: DIMM (NMEM)

The DIMM device provides a character device for sending commands to hardware, and it is a container for LABELs.  If the DIMM is defined by NFIT then an optional 'nfit' attribute sub-directory is available to add NFIT-specifics.

Note that the kernel device name for "DIMMs" is "nmemX".  The NFIT describes these devices via "Memory Device to System Physical Address Range Mapping Structure", and there is no requirement that they actually be physical DIMMs, so we use a more generic name.

LIBNVDIMM: DIMM (NMEM)

struct nvdimm *nvdimm_create(struct nvdimm_bus *nvdimm_bus, void *provider_data,

const struct attribute_group **groups, unsigned long flags,

unsigned long *dsm_mask);

LIBNDCTL: DIMM enumeration example

Note, in this example we are assuming NFIT-defined DIMMs which are identified by an "nfit_handle" a 32-bit value where:

• Bit 3:0 DIMM number within the memory channel
• Bit 7:4 memory channel number
• Bit 11:8 memory controller ID
• Bit 15:12 socket ID (within scope of a Node controller if node controller is present)
• Bit 27:16 Node Controller ID
• Bit 31:28 Reserved

static struct ndctl_dimm *get_dimm_by_handle(struct ndctl_bus *bus,

      unsigned int handle)

{

struct ndctl_dimm *dimm;

ndctl_dimm_foreach(bus, dimm)

if (ndctl_dimm_get_handle(dimm) == handle)

return dimm;

return NULL;

}

#define DIMM_HANDLE(n, s, i, c, d) \

(((n & 0xfff) << 16) | ((s & 0xf) << 12) | ((i & 0xf) << 8) \

| ((c & 0xf) << 4) | (d & 0xf))

dimm = get_dimm_by_handle(bus, DIMM_HANDLE(0, 0, 0, 0, 0));

LIBNVDIMM/LIBNDCTL: Region

A generic REGION device is registered for each PMEM range or BLK-apertureset. Per the example there are 6 regions: 2 PMEM and 4 BLK-aperture sets on the "nfit_test.0" bus.  The primary role of regions are to be a container of "mappings".  A mapping is a tuple of <DIMM, DPA-start-offset, length>.

LIBNVDIMM provides a built-in driver for these REGION devices.  This driver is responsible for reconciling the aliased DPA mappings across all regions, parsing the LABEL, if present, and then emitting NAMESPACE devices with the resolved/exclusive DPA-boundaries for the nd_pmem or nd_blk device driver to consume.

In addition to the generic attributes of "mapping"s, "interleave_ways" and "size" the REGION device also exports some convenience attributes."nstype" indicates the integer type of namespace-device this region emits, "devtype" duplicates the DEVTYPE variable stored by udev at the 'add' event, "modalias" duplicates the MODALIAS variable stored by udev at the 'add' event, and finally, the optional "spa_index" is provided in the case where the region is defined by a SPA.

LIBNVDIMM: region

struct nd_region *nvdimm_pmem_region_create(struct nvdimm_bus *nvdimm_bus,

struct nd_region_desc *ndr_desc);

struct nd_region *nvdimm_blk_region_create(struct nvdimm_bus *nvdimm_bus,

struct nd_region_desc *ndr_desc);

LIBNDCTL: region enumeration example

Sample region retrieval routines based on NFIT-unique data like "spa_index" (interleave set id) for PMEM and "nfit_handle" (dimm id) for BLK.

static struct ndctl_region *get_pmem_region_by_spa_index(struct ndctl_bus *bus,

unsigned int spa_index)

{

struct ndctl_region *region;

ndctl_region_foreach(bus, region) {

if (ndctl_region_get_type(region) != ND_DEVICE_REGION_PMEM)

continue;

if (ndctl_region_get_spa_index(region) == spa_index)

return region;

}

return NULL;

}

static struct ndctl_region *get_blk_region_by_dimm_handle(struct ndctl_bus *bus,

unsigned int handle)

{

struct ndctl_region *region;

ndctl_region_foreach(bus, region) {

struct ndctl_mapping *map;

if (ndctl_region_get_type(region) != ND_DEVICE_REGION_BLOCK)

continue;

ndctl_mapping_foreach(region, map) {

struct ndctl_dimm *dimm = ndctl_mapping_get_dimm(map);

if (ndctl_dimm_get_handle(dimm) == handle)

return region;

}

}

return NULL;

}

Why Not Encode the Region Type into the Region Name?

At first glance it seems since NFIT defines just PMEM and BLK interface types that we should simply name REGION devices with something derived from those type names.  However, the ND subsystem explicitly keeps the REGION name generic and expects userspace to always consider the region-attributes for four reasons:

1. There are already more than two REGION and "namespace" types.  For PMEM there are two subtypes.  As mentioned previously we have PMEM where the constituent DIMM devices are known and anonymous PMEM. For BLK regions the NFIT specification already anticipates vendor specific implementations.  The exact distinction of what a region contains is in the region-attributes not the region-name or the region-devtype.

2. A region with zero child-namespaces is a possible configuration.  For example, the NFIT allows for a DCR to be published without a corresponding BLK-aperture.  This equates to a DIMM that can only accept control/configuration messages, but no i/o through a descendant block device.  Again, this "type" is advertised in the attributes ('mappings'== 0) and the name does not tell you much.

3. What if a third major interface type arises in the future?  Outside of vendor specific implementations, it's not difficult to envision a third class of interface type beyond BLK and PMEM.  With a generic name for the REGION level of the device-hierarchy old userspace    implementations can still make sense of new kernel advertised region-types.  Userspace can always rely on the generic region attributes like "mappings", "size", etc and the expected child devices named "namespace".  This generic format of the device-model hierarchy allows the LIBNVDIMM and LIBNDCTL implementations to be more uniform and future-proof.

4. There are more robust mechanisms for determining the major type of a region than a device name.  See the next section, How Do I Determine the Major Type of a Region?

How Do I Determine the Major Type of a Region?

Outside of the blanket recommendation of "use libndctl", or simply looking at the kernel header (/usr/include/linux/ndctl.h) to decode the "nstype" integer attribute, here are some other options.

1. module alias lookup: The whole point of region/namespace device type differentiation is to     decide which block-device driver will attach to a given LIBNVDIMM namespace. One can simply use the modalias to lookup the resulting module.  It's important to note that this method is robust in the presence of a vendor-specific driver down the road.  If a vendor-specific    implementation wants to supplant the standard nd_blk driver it can with minimal impact to the rest of LIBNVDIMM.

In fact, a vendor may also want to have a vendor-specific region-driver (outside of nd_region).  For example, if a vendor defined its own LABEL format it would need its own region driver to parse that LABEL and emit the resulting namespaces.  The output from module resolution is more accurate than a region-name or region-devtype.

2. udev: The kernel "devtype" is registered in the udev database

    # udevadm info --path=/devices/platform/nfit_test.0/ndbus0/region0

    P: /devices/platform/nfit_test.0/ndbus0/region0

    E: DEVPATH=/devices/platform/nfit_test.0/ndbus0/region0

    E: DEVTYPE=nd_pmem

    E: MODALIAS=nd:t2

    E: SUBSYSTEM=nd

    # udevadm info --path=/devices/platform/nfit_test.0/ndbus0/region4

    P: /devices/platform/nfit_test.0/ndbus0/region4

    E: DEVPATH=/devices/platform/nfit_test.0/ndbus0/region4

    E: DEVTYPE=nd_blk

    E: MODALIAS=nd:t3

    E: SUBSYSTEM=nd

...and is available as a region attribute, but keep in mind that the "devtype" does not indicate sub-type variations and scripts should really be understanding the other attributes.

3. type specific attributes: As it currently stands a BLK-aperture region will never have a    "nfit/spa_index" attribute, but neither will a non-NFIT PMEM region.  A BLK region with a "mappings" value of 0 is, as mentioned above, a DIMM that does not allow I/O.  A PMEM region with a "mappings" value of zero is a simple system-physical-address range.

LIBNVDIMM/LIBNDCTL: Namespace

A REGION, after resolving DPA aliasing and LABEL specified boundaries, surfaces one or more "namespace" devices.  The arrival of a "namespace" device currently triggers either the nd_blk or nd_pmem driver to load and register a disk/block device.

LIBNVDIMM: namespace

Here is a sample layout from the three major types of NAMESPACE where namespace0.0 represents DIMM-info-backed PMEM (note that it has a 'uuid' attribute), namespace2.0 represents a BLK namespace (note it has a 'sector_size' attribute) that, and namespace6.0 represents an anonymous PMEM namespace (note that has no 'uuid' attribute due to not support a LABEL).

(See the Original Source)

LIBNDCTL: namespace enumeration example

Namespaces are indexed relative to their parent region, example below. These indexes are mostly static from boot to boot, but subsystem makesno guarantees in this regard.  For a static namespace identifier use its 'uuid' attribute.

static struct ndctl_namespace *get_namespace_by_id(struct ndctl_region *region,

                unsigned int id)

{

        struct ndctl_namespace *ndns;

        ndctl_namespace_foreach(region, ndns)

                if (ndctl_namespace_get_id(ndns) == id)

                        return ndns;

        return NULL;

}

LIBNDCTL: namespace creation example

Idle namespaces are automatically created by the kernel if a given region has enough available capacity to create a new namespace. Namespace instantiation involves finding an idle namespace and configuring it.  For the most part the setting of namespace attributes can occur in any order, the only constraint is that 'uuid' must be set before 'size'.  This enables the kernel to track DPA allocations internally with a static identifier.

static int configure_namespace(struct ndctl_region *region,

                struct ndctl_namespace *ndns,

                struct namespace_parameters *parameters)

{

        char devname[50];

        snprintf(devname, sizeof(devname), "namespace%d.%d",

                        ndctl_region_get_id(region), paramaters->id);

        ndctl_namespace_set_alt_name(ndns, devname);

        /* 'uuid' must be set prior to setting size! */

        ndctl_namespace_set_uuid(ndns, paramaters->uuid);

        ndctl_namespace_set_size(ndns, paramaters->size);

        /* unlike pmem namespaces, blk namespaces have a sector size */

        if (parameters->lbasize)

                ndctl_namespace_set_sector_size(ndns, parameters->lbasize);

        ndctl_namespace_enable(ndns);

}

Why the Term "namespace"?

1. Why not "volume" for instance?  "volume" ran the risk of confusing ND (libnvdimm subsystem) to a volume manager like device-mapper.

2. The term originated to describe the sub-devices that can be created within a NVME controller (see the nvme specification: http://www.nvmexpress.org/specifications/), and NFIT namespaces are meant to parallel the capabilities and configurability of NVME-namespaces.

LIBNVDIMM/LIBNDCTL: Block Translation Table "btt"

A BTT (design document: http://pmem.io/2014/09/23/btt.html) is a stacked block device driver that fronts either the whole block device or a partition of a block device emitted by either a PMEM or BLK NAMESPACE.

LIBNVDIMM: btt layout Every region will start out with at least one BTT device which is the seed device. To activate it set the "namespace", "uuid", and "sector_size" attributes and then bind the device to the nd_pmem or nd_blk driver depending on the region type.

LIBNDCTL: btt creation example

Similar to namespaces an idle BTT device is automatically created per region.  Each time this seed" btt device is configured and enabled a new seed is created.  Creating a BTT configuration involves two steps of finding and idle BTT and assigning it to consume a PMEM or BLK namespace.

static struct ndctl_btt *get_idle_btt(struct ndctl_region *region)

{

struct ndctl_btt *btt;

ndctl_btt_foreach(region, btt)

if (!ndctl_btt_is_enabled(btt)

&& !ndctl_btt_is_configured(btt))

return btt;

return NULL;

}

static int configure_btt(struct ndctl_region *region,

struct btt_parameters *parameters)

{

btt = get_idle_btt(region);

ndctl_btt_set_uuid(btt, parameters->uuid);

ndctl_btt_set_sector_size(btt, parameters->sector_size);

ndctl_btt_set_namespace(btt, parameters->ndns);

/* turn off raw mode device */

ndctl_namespace_disable(parameters->ndns);

/* turn on btt access */

ndctl_btt_enable(btt);

}

Once instantiated a new inactive btt seed device will appear underneath the region. Once a "namespace" is removed from a BTT that instance of the BTT device will be deleted or otherwise reset to default values.  This deletion is only at the device model level.  In order to destroy a BTT the "info block" needs to be destroyed.  Note, that to destroy a BTT the media needs to be written in raw mode.  By default, the kernel will autodetect the presence of a BTT and disable raw mode.  This autodetect behavior can be suppressed by enabling raw mode for the namespace via the ndctl_namespace_set_raw_mode() API.