| blob | ca16b5acbf30dc4807b1ba76ff7eed3854f8ee34 |
1 ===============================
2 LIBNVDIMM: Non-Volatile Devices
3 ===============================
5 libnvdimm - kernel / libndctl - userspace helper library
7 nvdimm@lists.linux.dev
9 Version 13
11 .. contents:
13 Glossary
14 Overview
15 Supporting Documents
16 Git Trees
17 LIBNVDIMM PMEM
18 PMEM-REGIONs, Atomic Sectors, and DAX
19 Example NVDIMM Platform
20 LIBNVDIMM Kernel Device Model and LIBNDCTL Userspace API
21 LIBNDCTL: Context
22 libndctl: instantiate a new library context example
23 LIBNVDIMM/LIBNDCTL: Bus
24 libnvdimm: control class device in /sys/class
25 libnvdimm: bus
26 libndctl: bus enumeration example
27 LIBNVDIMM/LIBNDCTL: DIMM (NMEM)
28 libnvdimm: DIMM (NMEM)
29 libndctl: DIMM enumeration example
30 LIBNVDIMM/LIBNDCTL: Region
31 libnvdimm: region
32 libndctl: region enumeration example
33 Why Not Encode the Region Type into the Region Name?
34 How Do I Determine the Major Type of a Region?
35 LIBNVDIMM/LIBNDCTL: Namespace
36 libnvdimm: namespace
37 libndctl: namespace enumeration example
38 libndctl: namespace creation example
39 Why the Term "namespace"?
40 LIBNVDIMM/LIBNDCTL: Block Translation Table "btt"
41 libnvdimm: btt layout
42 libndctl: btt creation example
43 Summary LIBNDCTL Diagram
46 Glossary
47 ========
49 PMEM:
50 A system-physical-address range where writes are persistent. A
51 block device composed of PMEM is capable of DAX. A PMEM address range
52 may span an interleave of several DIMMs.
54 DPA:
55 DIMM Physical Address, is a DIMM-relative offset. With one DIMM in
56 the system there would be a 1:1 system-physical-address:DPA association.
57 Once more DIMMs are added a memory controller interleave must be
58 decoded to determine the DPA associated with a given
59 system-physical-address.
61 DAX:
62 File system extensions to bypass the page cache and block layer to
63 mmap persistent memory, from a PMEM block device, directly into a
64 process address space.
66 DSM:
67 Device Specific Method: ACPI method to control specific
68 device - in this case the firmware.
70 DCR:
71 NVDIMM Control Region Structure defined in ACPI 6 Section 5.2.25.5.
72 It defines a vendor-id, device-id, and interface format for a given DIMM.
74 BTT:
75 Block Translation Table: Persistent memory is byte addressable.
76 Existing software may have an expectation that the power-fail-atomicity
77 of writes is at least one sector, 512 bytes. The BTT is an indirection
78 table with atomic update semantics to front a PMEM block device
79 driver and present arbitrary atomic sector sizes.
81 LABEL:
82 Metadata stored on a DIMM device that partitions and identifies
83 (persistently names) capacity allocated to different PMEM namespaces. It
84 also indicates whether an address abstraction like a BTT is applied to
85 the namespace. Note that traditional partition tables, GPT/MBR, are
86 layered on top of a PMEM namespace, or an address abstraction like BTT
87 if present, but partition support is deprecated going forward.
90 Overview
91 ========
93 The LIBNVDIMM subsystem provides support for PMEM described by platform
94 firmware or a device driver. On ACPI based systems the platform firmware
95 conveys persistent memory resource via the ACPI NFIT "NVDIMM Firmware
96 Interface Table" in ACPI 6. While the LIBNVDIMM subsystem implementation
97 is generic and supports pre-NFIT platforms, it was guided by the
98 superset of capabilities need to support this ACPI 6 definition for
99 NVDIMM resources. The original implementation supported the
100 block-window-aperture capability described in the NFIT, but that support
101 has since been abandoned and never shipped in a product.
103 Supporting Documents
104 --------------------
106 ACPI 6:
107 https://www.uefi.org/sites/default/files/resources/ACPI_6.0.pdf
108 NVDIMM Namespace:
109 https://pmem.io/documents/NVDIMM_Namespace_Spec.pdf
110 DSM Interface Example:
111 https://pmem.io/documents/NVDIMM_DSM_Interface_Example.pdf
112 Driver Writer's Guide:
113 https://pmem.io/documents/NVDIMM_Driver_Writers_Guide.pdf
115 Git Trees
116 ---------
118 LIBNVDIMM:
119 https://git.kernel.org/cgit/linux/kernel/git/nvdimm/nvdimm.git
120 LIBNDCTL:
121 https://github.com/pmem/ndctl.git
124 LIBNVDIMM PMEM
125 ==============
127 Prior to the arrival of the NFIT, non-volatile memory was described to a
128 system in various ad-hoc ways. Usually only the bare minimum was
129 provided, namely, a single system-physical-address range where writes
130 are expected to be durable after a system power loss. Now, the NFIT
131 specification standardizes not only the description of PMEM, but also
132 platform message-passing entry points for control and configuration.
134 PMEM (nd_pmem.ko): Drives a system-physical-address range. This range is
135 contiguous in system memory and may be interleaved (hardware memory controller
136 striped) across multiple DIMMs. When interleaved the platform may optionally
137 provide details of which DIMMs are participating in the interleave.
139 It is worth noting that when the labeling capability is detected (a EFI
140 namespace label index block is found), then no block device is created
141 by default as userspace needs to do at least one allocation of DPA to
142 the PMEM range. In contrast ND_NAMESPACE_IO ranges, once registered,
143 can be immediately attached to nd_pmem. This latter mode is called
144 label-less or "legacy".
146 PMEM-REGIONs, Atomic Sectors, and DAX
147 -------------------------------------
149 For the cases where an application or filesystem still needs atomic sector
150 update guarantees it can register a BTT on a PMEM device or partition. See
151 LIBNVDIMM/NDCTL: Block Translation Table "btt"
154 Example NVDIMM Platform
155 =======================
157 For the remainder of this document the following diagram will be
158 referenced for any example sysfs layouts::
161 (a) (b) DIMM
162 +-------------------+--------+--------+--------+
163 +------+ | pm0.0 | free | pm1.0 | free | 0
164 | imc0 +--+- - - region0- - - +--------+ +--------+
165 +--+---+ | pm0.0 | free | pm1.0 | free | 1
166 | +-------------------+--------v v--------+
167 +--+---+ | |
168 | cpu0 | region1
169 +--+---+ | |
170 | +----------------------------^ ^--------+
171 +--+---+ | free | pm1.0 | free | 2
172 | imc1 +--+----------------------------| +--------+
173 +------+ | free | pm1.0 | free | 3
174 +----------------------------+--------+--------+
176 In this platform we have four DIMMs and two memory controllers in one
177 socket. Each PMEM interleave set is identified by a region device with
178 a dynamically assigned id.
180 1. The first portion of DIMM0 and DIMM1 are interleaved as REGION0. A
181 single PMEM namespace is created in the REGION0-SPA-range that spans most
182 of DIMM0 and DIMM1 with a user-specified name of "pm0.0". Some of that
183 interleaved system-physical-address range is left free for
184 another PMEM namespace to be defined.
186 2. In the last portion of DIMM0 and DIMM1 we have an interleaved
187 system-physical-address range, REGION1, that spans those two DIMMs as
188 well as DIMM2 and DIMM3. Some of REGION1 is allocated to a PMEM namespace
189 named "pm1.0".
191 This bus is provided by the kernel under the device
192 /sys/devices/platform/nfit_test.0 when the nfit_test.ko module from
193 tools/testing/nvdimm is loaded. This module is a unit test for
194 LIBNVDIMM and the acpi_nfit.ko driver.
197 LIBNVDIMM Kernel Device Model and LIBNDCTL Userspace API
198 ========================================================
200 What follows is a description of the LIBNVDIMM sysfs layout and a
201 corresponding object hierarchy diagram as viewed through the LIBNDCTL
202 API. The example sysfs paths and diagrams are relative to the Example
203 NVDIMM Platform which is also the LIBNVDIMM bus used in the LIBNDCTL unit
204 test.
206 LIBNDCTL: Context
207 -----------------
209 Every API call in the LIBNDCTL library requires a context that holds the
210 logging parameters and other library instance state. The library is
211 based on the libabc template:
213 https://git.kernel.org/cgit/linux/kernel/git/kay/libabc.git
215 LIBNDCTL: instantiate a new library context example
216 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
218 ::
220 struct ndctl_ctx *ctx;
222 if (ndctl_new(&ctx) == 0)
223 return ctx;
224 else
225 return NULL;
227 LIBNVDIMM/LIBNDCTL: Bus
228 -----------------------
230 A bus has a 1:1 relationship with an NFIT. The current expectation for
231 ACPI based systems is that there is only ever one platform-global NFIT.
232 That said, it is trivial to register multiple NFITs, the specification
233 does not preclude it. The infrastructure supports multiple busses and
234 we use this capability to test multiple NFIT configurations in the unit
235 test.
237 LIBNVDIMM: control class device in /sys/class
238 ---------------------------------------------
240 This character device accepts DSM messages to be passed to DIMM
241 identified by its NFIT handle::
243 /sys/class/nd/ndctl0
244 |-- dev
245 |-- device -> ../../../ndbus0
246 |-- subsystem -> ../../../../../../../class/nd
250 LIBNVDIMM: bus
251 --------------
253 ::
255 struct nvdimm_bus *nvdimm_bus_register(struct device *parent,
256 struct nvdimm_bus_descriptor *nfit_desc);
258 ::
260 /sys/devices/platform/nfit_test.0/ndbus0
261 |-- commands
262 |-- nd
263 |-- nfit
264 |-- nmem0
265 |-- nmem1
266 |-- nmem2
267 |-- nmem3
268 |-- power
269 |-- provider
270 |-- region0
271 |-- region1
272 |-- region2
273 |-- region3
274 |-- region4
275 |-- region5
276 |-- uevent
277 `-- wait_probe
279 LIBNDCTL: bus enumeration example
280 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
282 Find the bus handle that describes the bus from Example NVDIMM Platform::
284 static struct ndctl_bus *get_bus_by_provider(struct ndctl_ctx *ctx,
285 const char *provider)
286 {
287 struct ndctl_bus *bus;
289 ndctl_bus_foreach(ctx, bus)
290 if (strcmp(provider, ndctl_bus_get_provider(bus)) == 0)
291 return bus;
293 return NULL;
294 }
296 bus = get_bus_by_provider(ctx, "nfit_test.0");
299 LIBNVDIMM/LIBNDCTL: DIMM (NMEM)
300 -------------------------------
302 The DIMM device provides a character device for sending commands to
303 hardware, and it is a container for LABELs. If the DIMM is defined by
304 NFIT then an optional 'nfit' attribute sub-directory is available to add
305 NFIT-specifics.
307 Note that the kernel device name for "DIMMs" is "nmemX". The NFIT
308 describes these devices via "Memory Device to System Physical Address
309 Range Mapping Structure", and there is no requirement that they actually
310 be physical DIMMs, so we use a more generic name.
312 LIBNVDIMM: DIMM (NMEM)
313 ^^^^^^^^^^^^^^^^^^^^^^
315 ::
317 struct nvdimm *nvdimm_create(struct nvdimm_bus *nvdimm_bus, void *provider_data,
318 const struct attribute_group **groups, unsigned long flags,
319 unsigned long *dsm_mask);
321 ::
323 /sys/devices/platform/nfit_test.0/ndbus0
324 |-- nmem0
325 | |-- available_slots
326 | |-- commands
327 | |-- dev
328 | |-- devtype
329 | |-- driver -> ../../../../../bus/nd/drivers/nvdimm
330 | |-- modalias
331 | |-- nfit
332 | | |-- device
333 | | |-- format
334 | | |-- handle
335 | | |-- phys_id
336 | | |-- rev_id
337 | | |-- serial
338 | | `-- vendor
339 | |-- state
340 | |-- subsystem -> ../../../../../bus/nd
341 | `-- uevent
342 |-- nmem1
343 [..]
346 LIBNDCTL: DIMM enumeration example
347 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
349 Note, in this example we are assuming NFIT-defined DIMMs which are
350 identified by an "nfit_handle" a 32-bit value where:
352 - Bit 3:0 DIMM number within the memory channel
353 - Bit 7:4 memory channel number
354 - Bit 11:8 memory controller ID
355 - Bit 15:12 socket ID (within scope of a Node controller if node
356 controller is present)
357 - Bit 27:16 Node Controller ID
358 - Bit 31:28 Reserved
360 ::
362 static struct ndctl_dimm *get_dimm_by_handle(struct ndctl_bus *bus,
363 unsigned int handle)
364 {
365 struct ndctl_dimm *dimm;
367 ndctl_dimm_foreach(bus, dimm)
368 if (ndctl_dimm_get_handle(dimm) == handle)
369 return dimm;
371 return NULL;
372 }
374 #define DIMM_HANDLE(n, s, i, c, d) \
375 (((n & 0xfff) << 16) | ((s & 0xf) << 12) | ((i & 0xf) << 8) \
376 | ((c & 0xf) << 4) | (d & 0xf))
378 dimm = get_dimm_by_handle(bus, DIMM_HANDLE(0, 0, 0, 0, 0));
380 LIBNVDIMM/LIBNDCTL: Region
381 --------------------------
383 A generic REGION device is registered for each PMEM interleave-set /
384 range. Per the example there are 2 PMEM regions on the "nfit_test.0"
385 bus. The primary role of regions are to be a container of "mappings". A
386 mapping is a tuple of <DIMM, DPA-start-offset, length>.
388 LIBNVDIMM provides a built-in driver for REGION devices. This driver
389 is responsible for all parsing LABELs, if present, and then emitting NAMESPACE
390 devices for the nd_pmem driver to consume.
392 In addition to the generic attributes of "mapping"s, "interleave_ways"
393 and "size" the REGION device also exports some convenience attributes.
394 "nstype" indicates the integer type of namespace-device this region
395 emits, "devtype" duplicates the DEVTYPE variable stored by udev at the
396 'add' event, "modalias" duplicates the MODALIAS variable stored by udev
397 at the 'add' event, and finally, the optional "spa_index" is provided in
398 the case where the region is defined by a SPA.
400 LIBNVDIMM: region::
402 struct nd_region *nvdimm_pmem_region_create(struct nvdimm_bus *nvdimm_bus,
403 struct nd_region_desc *ndr_desc);
405 ::
407 /sys/devices/platform/nfit_test.0/ndbus0
408 |-- region0
409 | |-- available_size
410 | |-- btt0
411 | |-- btt_seed
412 | |-- devtype
413 | |-- driver -> ../../../../../bus/nd/drivers/nd_region
414 | |-- init_namespaces
415 | |-- mapping0
416 | |-- mapping1
417 | |-- mappings
418 | |-- modalias
419 | |-- namespace0.0
420 | |-- namespace_seed
421 | |-- numa_node
422 | |-- nfit
423 | | `-- spa_index
424 | |-- nstype
425 | |-- set_cookie
426 | |-- size
427 | |-- subsystem -> ../../../../../bus/nd
428 | `-- uevent
429 |-- region1
430 [..]
432 LIBNDCTL: region enumeration example
433 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
435 Sample region retrieval routines based on NFIT-unique data like
436 "spa_index" (interleave set id).
438 ::
440 static struct ndctl_region *get_pmem_region_by_spa_index(struct ndctl_bus *bus,
441 unsigned int spa_index)
442 {
443 struct ndctl_region *region;
445 ndctl_region_foreach(bus, region) {
446 if (ndctl_region_get_type(region) != ND_DEVICE_REGION_PMEM)
447 continue;
448 if (ndctl_region_get_spa_index(region) == spa_index)
449 return region;
450 }
451 return NULL;
452 }
455 LIBNVDIMM/LIBNDCTL: Namespace
456 -----------------------------
458 A REGION, after resolving DPA aliasing and LABEL specified boundaries, surfaces
459 one or more "namespace" devices. The arrival of a "namespace" device currently
460 triggers the nd_pmem driver to load and register a disk/block device.
462 LIBNVDIMM: namespace
463 ^^^^^^^^^^^^^^^^^^^^
465 Here is a sample layout from the 2 major types of NAMESPACE where namespace0.0
466 represents DIMM-info-backed PMEM (note that it has a 'uuid' attribute), and
467 namespace1.0 represents an anonymous PMEM namespace (note that has no 'uuid'
468 attribute due to not support a LABEL)
470 ::
472 /sys/devices/platform/nfit_test.0/ndbus0/region0/namespace0.0
473 |-- alt_name
474 |-- devtype
475 |-- dpa_extents
476 |-- force_raw
477 |-- modalias
478 |-- numa_node
479 |-- resource
480 |-- size
481 |-- subsystem -> ../../../../../../bus/nd
482 |-- type
483 |-- uevent
484 `-- uuid
485 /sys/devices/platform/nfit_test.1/ndbus1/region1/namespace1.0
486 |-- block
487 | `-- pmem0
488 |-- devtype
489 |-- driver -> ../../../../../../bus/nd/drivers/pmem
490 |-- force_raw
491 |-- modalias
492 |-- numa_node
493 |-- resource
494 |-- size
495 |-- subsystem -> ../../../../../../bus/nd
496 |-- type
497 `-- uevent
499 LIBNDCTL: namespace enumeration example
500 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
501 Namespaces are indexed relative to their parent region, example below.
502 These indexes are mostly static from boot to boot, but subsystem makes
503 no guarantees in this regard. For a static namespace identifier use its
504 'uuid' attribute.
506 ::
508 static struct ndctl_namespace
509 *get_namespace_by_id(struct ndctl_region *region, unsigned int id)
510 {
511 struct ndctl_namespace *ndns;
513 ndctl_namespace_foreach(region, ndns)
514 if (ndctl_namespace_get_id(ndns) == id)
515 return ndns;
517 return NULL;
518 }
520 LIBNDCTL: namespace creation example
521 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
523 Idle namespaces are automatically created by the kernel if a given
524 region has enough available capacity to create a new namespace.
525 Namespace instantiation involves finding an idle namespace and
526 configuring it. For the most part the setting of namespace attributes
527 can occur in any order, the only constraint is that 'uuid' must be set
528 before 'size'. This enables the kernel to track DPA allocations
529 internally with a static identifier::
531 static int configure_namespace(struct ndctl_region *region,
532 struct ndctl_namespace *ndns,
533 struct namespace_parameters *parameters)
534 {
535 char devname[50];
537 snprintf(devname, sizeof(devname), "namespace%d.%d",
538 ndctl_region_get_id(region), paramaters->id);
540 ndctl_namespace_set_alt_name(ndns, devname);
541 /* 'uuid' must be set prior to setting size! */
542 ndctl_namespace_set_uuid(ndns, paramaters->uuid);
543 ndctl_namespace_set_size(ndns, paramaters->size);
544 /* unlike pmem namespaces, blk namespaces have a sector size */
545 if (parameters->lbasize)
546 ndctl_namespace_set_sector_size(ndns, parameters->lbasize);
547 ndctl_namespace_enable(ndns);
548 }
551 Why the Term "namespace"?
552 ^^^^^^^^^^^^^^^^^^^^^^^^^
554 1. Why not "volume" for instance? "volume" ran the risk of confusing
555 ND (libnvdimm subsystem) to a volume manager like device-mapper.
557 2. The term originated to describe the sub-devices that can be created
558 within a NVME controller (see the nvme specification:
559 https://www.nvmexpress.org/specifications/), and NFIT namespaces are
560 meant to parallel the capabilities and configurability of
561 NVME-namespaces.
564 LIBNVDIMM/LIBNDCTL: Block Translation Table "btt"
565 -------------------------------------------------
567 A BTT (design document: https://pmem.io/2014/09/23/btt.html) is a
568 personality driver for a namespace that fronts entire namespace as an
569 'address abstraction'.
571 LIBNVDIMM: btt layout
572 ^^^^^^^^^^^^^^^^^^^^^
574 Every region will start out with at least one BTT device which is the
575 seed device. To activate it set the "namespace", "uuid", and
576 "sector_size" attributes and then bind the device to the nd_pmem or
577 nd_blk driver depending on the region type::
579 /sys/devices/platform/nfit_test.1/ndbus0/region0/btt0/
580 |-- namespace
581 |-- delete
582 |-- devtype
583 |-- modalias
584 |-- numa_node
585 |-- sector_size
586 |-- subsystem -> ../../../../../bus/nd
587 |-- uevent
588 `-- uuid
590 LIBNDCTL: btt creation example
591 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
593 Similar to namespaces an idle BTT device is automatically created per
594 region. Each time this "seed" btt device is configured and enabled a new
595 seed is created. Creating a BTT configuration involves two steps of
596 finding and idle BTT and assigning it to consume a namespace.
598 ::
600 static struct ndctl_btt *get_idle_btt(struct ndctl_region *region)
601 {
602 struct ndctl_btt *btt;
604 ndctl_btt_foreach(region, btt)
605 if (!ndctl_btt_is_enabled(btt)
606 && !ndctl_btt_is_configured(btt))
607 return btt;
609 return NULL;
610 }
612 static int configure_btt(struct ndctl_region *region,
613 struct btt_parameters *parameters)
614 {
615 btt = get_idle_btt(region);
617 ndctl_btt_set_uuid(btt, parameters->uuid);
618 ndctl_btt_set_sector_size(btt, parameters->sector_size);
619 ndctl_btt_set_namespace(btt, parameters->ndns);
620 /* turn off raw mode device */
621 ndctl_namespace_disable(parameters->ndns);
622 /* turn on btt access */
623 ndctl_btt_enable(btt);
624 }
626 Once instantiated a new inactive btt seed device will appear underneath
627 the region.
629 Once a "namespace" is removed from a BTT that instance of the BTT device
630 will be deleted or otherwise reset to default values. This deletion is
631 only at the device model level. In order to destroy a BTT the "info
632 block" needs to be destroyed. Note, that to destroy a BTT the media
633 needs to be written in raw mode. By default, the kernel will autodetect
634 the presence of a BTT and disable raw mode. This autodetect behavior
635 can be suppressed by enabling raw mode for the namespace via the
636 ndctl_namespace_set_raw_mode() API.
639 Summary LIBNDCTL Diagram
640 ------------------------
642 For the given example above, here is the view of the objects as seen by the
643 LIBNDCTL API::
645 +---+
646 |CTX|
647 +-+-+
648 |
649 +-------+ |
650 | DIMM0 <-+ | +---------+ +--------------+ +---------------+
651 +-------+ | | +-> REGION0 +---> NAMESPACE0.0 +--> PMEM8 "pm0.0" |
652 | DIMM1 <-+ +-v--+ | +---------+ +--------------+ +---------------+
653 +-------+ +-+BUS0+-| +---------+ +--------------+ +----------------------+
654 | DIMM2 <-+ +----+ +-> REGION1 +---> NAMESPACE1.0 +--> PMEM6 "pm1.0" | BTT1 |
655 +-------+ | | +---------+ +--------------+ +---------------+------+
656 | DIMM3 <-+
657 +-------+