drm/fabric: vendor-neutral topology infrastructure for scale-up accelerator interconnects

new file mode 100644
index 000000000000..99b189dc614e
@@ -0,0 +1,142 @@
+.. SPDX-License-Identifier: (GPL-2.0+ OR MIT)
+
+===============================
+DRM Fabric over Generic Netlink
+===============================
+
+Modern GPUs and dedicated AI accelerators are increasingly connected through
+scale-up interconnect fabrics such as AMD xGMI and
+`UALink <https://ualinkconsortium.org/specification/>`__.
+
+DRM Fabric is a registry and dispatcher. It does not discover routes, compute
+reachability, program switch forwarding, manage device memory, or provide a
+data path; those remain with the vendor driver and the fabric controller.
+
+Key Goals:
+
+* Provide a standardized topology model for GPU and accelerator interconnects
+  (xGMI, UALink and similar), enabling data-center discovery and monitoring.
+* Support read-only enumeration, monitoring and state queries for
+  provider-owned topology.
+* Offer a flexible, future-proof interface that can be extended with new fabric
+  types and attributes without breaking the uAPI.
+* Allow multiple endpoints and ports per provider, so drivers can model
+  accelerator attachments, links and their peers.
+
+.. contents::
+
+Object model
+============
+
+DRM Fabric models interconnect topology with four object types::
+
+    fabric
+      `-- endpoint
+            `-- port
+                  `-- peer (optional value descriptor)
+
+A *fabric* is a membership group of *endpoints*; each endpoint represents one
+accelerator attachment and owns a fixed set of *ports*; and a *port* may carry a
+*peer* describing the device directly adjacent at the far end of its link, which
+is either another accelerator or a fabric switch.
+
+Identity is layered. The core assigns kernel-local ``fabric-id`` and
+``endpoint-id`` values that address live registry objects for the duration of
+their registration. Providers supply an endpoint ``fabric-ep-id`` -- an
+accelerator's identity within its fabric's identity domain. A peer instead
+carries a type-qualified ``peer-id``: for ``peer-type = accel`` it is the
+far-end accelerator's ``fabric-ep-id``, and for ``peer-type = switch`` it is an
+opaque provider-defined switch identity that names no local object. Accelerator
+and switch identities occupy separate namespaces selected by ``peer-type``, so
+the same numeric value may name different objects under each type.
+
+Fabric membership does not imply end-to-end reachability, and the topology is
+not necessarily a tree. The registered shape reflects the direct adjacency the
+provider reports: for example, it may be a full mesh with no root, a linear
+chain, or a switch-based topology in which ports terminate at opaque switch
+peers rather than locally registered endpoints.
+
+A *peer* is a value descriptor, not a reference to a live kernel object: its
+``peer-id`` may name a remote accelerator managed by another OS or an opaque
+switch in another trust domain, and need not resolve in the local registry. The
+core stores one directed half-edge and does not require the reverse half-edge to
+exist, so removing an endpoint does not retract peer descriptors held by other
+endpoints. ``peer-type = switch`` only describes the kind of far end; it does
+not create a first-class switch object.
+
+.. kernel-doc:: drivers/gpu/drm/fabric/drm_fabric.c
+   :doc: DRM Fabric core
+
+Peer semantics
+--------------
+
+A peer is an identity rather than a reference to a live object, and the
+difference decides what the core reports. A ``peer-id`` that resolves in the
+local registry today may stop resolving later because the endpoint it named
+unregistered, and no event is emitted on the port still carrying it: a peer
+is recorded on one local half-edge, and peer disappearance does not retract
+that half-edge. A peer is therefore topology as last set, not proof of live
+connectivity; liveness belongs to the fabric controller.
+
+The core never retracts a half-edge on its own. Failing to resolve a peer
+locally is not the same as the link going away -- the far end may be a switch,
+an accelerator on another node, or a local endpoint that merely unregistered
+-- so only the provider knows when a port's physical adjacency actually
+changed, and only the provider retracts or replaces the descriptor.
+
+Endpoint teardown removes the endpoint's owned half-edges without generating
+a separate event for each port: the delete already describes the transition,
+so removing an endpoint advances the topology generation once rather than
+once per child port.
+
+Driver API
+----------
+
+.. kernel-doc:: include/drm/drm_fabric.h
+   :internal:
+
+.. kernel-doc:: drivers/gpu/drm/fabric/drm_fabric.c
+   :export:
+
+Design scope and boundaries
+===========================
+
+Vendor drivers retain hardware discovery, firmware interaction and the
+load/store data path; DRM Fabric represents only the topology and
+provider-reported state of DRM-managed accelerators, which is why it
+belongs in DRM. The interface does not define MMU programming, switch
+policy, key management, live migration, or any required user space daemon.
+
+DRM Fabric does not define in-network collective operations or how an
+endpoint or switch executes them. Such capabilities belong to the
+interconnect implementation and its provider. Adding capability later is not
+foreclosed: for an endpoint it is a new attribute, while
+``peer-type = switch`` is a value descriptor rather than a registered object,
+so there is nowhere to attach one today. ``peer-id`` is already an opaque
+switch identity that need not resolve locally, so making one resolvable later
+strengthens the contract rather than breaking it.
+
+Object lifetime and locking
+===========================
+
+All registry and object state is protected by ``drm_fabric_lock``. A fabric and
+its endpoints are created and torn down through the provider API; endpoint
+unregister removes the endpoint from the registry and frees its fixed set of
+ports. Fabric membership is tracked, so a provider must remove all member
+endpoints before unregistering a provider-owned fabric: drm_fabric_unregister()
+returns ``-EBUSY`` and leaves the fabric registered if any remain, so the
+provider must retry after removing them rather than treat the fabric as gone.
+
+Objects are reference counted and a port is pinned through its owning endpoint.
+Endpoint unregister drops the registration reference and waits for outstanding
+pins before freeing the ports; fabric membership holds a fabric reference.
+
+Providers own object lifetime, so a provider must serialise endpoint
+registration against unregistration of the containing fabric. The unregister
+entry points compare the supplied pointer against the registry before
+dereferencing it, so a stale or repeated teardown is rejected: fabric
+unregistration returns ``-ENODEV``, and endpoint unregistration, having no
+error return, warns and performs no teardown. Endpoint registration rejects a
+departed parent the same way. These checks prove current address membership
+only: they cannot tell an earlier incarnation from another object registered
+later at the same address.

index 65bf3b26e4f4..7d99c47ddbe7 100644
@@ -16,6 +16,7 @@ GPU Driver Developer's Guide
    driver-uapi
    drm-client
    drm-compute
+   drm-fabric
    drivers
    backlight
    vga-switcheroo

index 3c508bda61d5..2ab63bd763b3 100644
@@ -8966,6 +8966,20 @@ F:	Documentation/devicetree/bindings/display/xlnx/
 F:	Documentation/gpu/zynqmp.rst
 F:	drivers/gpu/drm/xlnx/
 
+DRM FABRIC
+M:	Konstantin Sinyuk <[email protected]>
+M:	Maarten Lankhorst <[email protected]>
+R:	Francois Dugast <[email protected]>
+L:	[email protected]
+S:	Maintained
+T:	git https://gitlab.freedesktop.org/drm/misc/kernel.git
+F:	Documentation/gpu/drm-fabric.rst
+F:	Documentation/netlink/specs/drm_fabric.yaml
+F:	drivers/gpu/drm/fabric/
+F:	include/drm/drm_fabric.h
+F:	include/uapi/drm/drm_fabric.h
+F:	tools/testing/selftests/drivers/gpu/drm_fabric/
+
 DRM GPU SCHEDULER
 M:	Matthew Brost <[email protected]>
 M:	Danilo Krummrich <[email protected]>

index 323422861e8f..fc66146a603b 100644
@@ -309,6 +309,7 @@ source "drivers/gpu/drm/atmel-hlcdc/Kconfig"
 source "drivers/gpu/drm/bridge/Kconfig"
 source "drivers/gpu/drm/etnaviv/Kconfig"
 source "drivers/gpu/drm/exynos/Kconfig"
+source "drivers/gpu/drm/fabric/Kconfig"
 source "drivers/gpu/drm/fsl-dcu/Kconfig"
 source "drivers/gpu/drm/gma500/Kconfig"
 source "drivers/gpu/drm/gud/Kconfig"

index e635fcffd379..bdf67dd3aab9 100644
@@ -213,6 +213,7 @@ obj-y			+= panel/
 obj-y			+= bridge/
 obj-$(CONFIG_DRM_FSL_DCU) += fsl-dcu/
 obj-$(CONFIG_DRM_ETNAVIV) += etnaviv/
+obj-$(CONFIG_DRM_FABRIC) += fabric/
 obj-y			+= hisilicon/
 obj-y			+= mxsfb/
 obj-y			+= sysfb/

new file mode 100644
index 000000000000..21fbfae9863d
@@ -0,0 +1,14 @@
+# SPDX-License-Identifier: GPL-2.0
+
+config DRM_FABRIC
+	tristate "DRM fabric support"
+	depends on DRM && NET
+	help
+	  Enable DRM fabric support. This infrastructure provides the
+	  core object model and provider API for registered accelerator
+	  interconnect topologies.
+
+	  To compile this as a module, choose M here: the module will be
+	  called drm-fabric.
+
+	  If in doubt, say N.

new file mode 100644
index 000000000000..3a76f31f1e83
@@ -0,0 +1,4 @@
+# SPDX-License-Identifier: GPL-2.0
+
+obj-$(CONFIG_DRM_FABRIC) += drm-fabric.o
+drm-fabric-y := drm_fabric.o

index 99b189dc614e..d2c9e694b67f 100644
@@ -140,3 +140,38 @@ error return, warns and performs no teardown. Endpoint registration rejects a
 departed parent the same way. These checks prove current address membership
 only: they cannot tell an earlier incarnation from another object registered
 later at the same address.
+
+Generic Netlink family
+======================
+
+DRM Fabric is exposed through the ``drm-fabric`` Generic Netlink family.
+Dump enumeration and asynchronous notifications fit this multi-object model
+better than one-value-per-file sysfs. The family follows the YAML/ynl
+discipline used by DRM RAS; devlink's device hierarchy does not represent a
+fabric spanning multiple DRM devices.
+
+YAML specification
+------------------
+
+The interface is described in a YAML specification
+``Documentation/netlink/specs/drm_fabric.yaml``, which is the source of truth for
+the wire format. It auto-generates the uAPI header
+(``include/uapi/drm/drm_fabric.h``) and the kernel glue via
+``tools/net/ynl/pyynl/ynl_gen_c.py``. Generated files must never be edited by
+hand; regenerate them with ``tools/net/ynl/ynl-regen.sh`` after any spec change.
+
+uAPI stability
+--------------
+
+The YAML specification is the contract. New attributes, commands and enum values
+are added append-only; existing attribute numbers, command numbers and meanings
+are never reused. Requests are strictly validated, so an unknown attribute in a
+request is rejected; user space should ignore attributes it does not recognize
+in replies and notifications. The family is versioned through
+``DRM_FABRIC_FAMILY_VERSION``.
+
+Kernel-local ``fabric-id`` and ``endpoint-id`` values identify live registry
+objects and are not persistent hardware identities: they remain valid for the
+lifetime of the registered object, but may disappear or be reused after the
+object is unregistered. ``instance-id`` and ``fabric-ep-id`` carry
+provider-defined identity, whose scope is described by the containing object.

427 lines not on this page · raw message

index d2c9e694b67f..a8b33d6cd618 100644
@@ -18,8 +18,8 @@ Key Goals:
   (xGMI, UALink and similar), enabling data-center discovery and monitoring.
 * Support read-only enumeration, monitoring and state queries for
   provider-owned topology.
-* Offer a flexible, future-proof interface that can be extended with new fabric
-  types and attributes without breaking the uAPI.
+* Allow new attributes and fabric types to be added without reusing existing
+  wire identifiers, so the uAPI extends without breaking existing consumers.
 * Allow multiple endpoints and ports per provider, so drivers can model
   accelerator attachments, links and their peers.
 

@@ -84,10 +84,14 @@ an accelerator on another node, or a local endpoint that merely unregistered
 -- so only the provider knows when a port's physical adjacency actually
 changed, and only the provider retracts or replaces the descriptor.
 
-Endpoint teardown removes the endpoint's owned half-edges without generating
-a separate event for each port: the delete already describes the transition,
-so removing an endpoint advances the topology generation once rather than
-once per child port.
+``port-peer-delete-ntf`` reports an explicitly retracted half-edge; it is
+not emitted when a peer merely becomes locally unresolvable, so its absence
+is not evidence the far end is still reachable.
+
+Endpoint teardown removes the endpoint's owned half-edges without
+generating a separate event for each port: the delete already describes
+the transition, so removing an endpoint advances the topology generation
+once rather than once per child port.
 
 Driver API
 ----------

@@ -175,3 +179,135 @@ objects and are not persistent hardware identities: they remain valid for the
 lifetime of the registered object, but may disappear or be reused after the
 object is unregistered. ``instance-id`` and ``fabric-ep-id`` carry
 provider-defined identity, whose scope is described by the containing object.
+
+Query operations
+================
+
+User space enumerates topology with four read-only commands, each supporting a
+single lookup (``do``) and a bulk dump (``dump``):
+
+* ``fabric-get`` -- enumerate fabrics (``do`` by ``fabric-id``, ``dump`` for all).
+* ``endpoint-get`` -- enumerate endpoints, optionally filtered by ``fabric-id``,
+  or resolve one by ``endpoint-id`` or backing ``dev-name``/``bus-name``.
+* ``port-get`` -- enumerate ports, filtered by ``endpoint-id``. Ports may report
+  the provider's maximum capability as ``max-lane-count`` and
+  ``max-lane-signaling-rate-mbps`` (zero means unknown); these are maxima, not
+  the currently negotiated width or rate.
+* ``port-stats-get`` -- per-port statistics, filtered by ``endpoint-id``. A
+  targeted request for a port whose provider does not implement
+  ``port_stats_get`` returns ``-EOPNOTSUPP``. During a dump, ports without
+  statistics support are omitted and enumeration continues with later ports;
+  any other provider error ends the dump.
+
+Notifications
+-------------
+
+Subscribe to the ``monitor`` multicast group to receive asynchronous change
+notifications. Full-object notifications reuse the shape of their matching
+``get`` reply and are declared with ``notify:``; peer-link notifications carry
+a partial payload and are declared with ``event:``.
+
+.. list-table::
+   :header-rows: 1
+
+   * - Notification
+     - Trigger
+     - Payload
+   * - ``fabric-create-ntf``
+     - a fabric is registered by a provider
+     - reuses ``fabric-get``
+   * - ``fabric-delete-ntf``
+     - a provider-owned fabric is unregistered
+     - reuses ``fabric-get``
+   * - ``endpoint-create-ntf``
+     - an endpoint is registered
+     - reuses ``endpoint-get``
+   * - ``endpoint-delete-ntf``
+     - an endpoint is unregistered
+     - reuses ``endpoint-get``
+   * - ``port-change-ntf``
+     - a port's operational state changes
+     - reuses ``port-get``
+   * - ``port-peer-create-ntf``
+     - a port's peer descriptor is set by its provider
+     - partial (``event:``)
+   * - ``port-peer-delete-ntf``
+     - a port's peer descriptor is explicitly unset by its provider; never
+       emitted for an implicit half-edge loss (see `Peer semantics`_)
+     - partial (``event:``)
+
+Notifications are best-effort. A listener that detects loss, restarts, or
+receives an interrupted dump must rebuild state with the query commands.
+
+Topology generation and dump consistency
+----------------------------------------
+
+The core keeps a nonzero generation counter and advances it whenever topology or
+exposed state changes. It is surfaced as the ``topology-generation`` attribute on
+``fabric-get``, ``endpoint-get`` and ``port-get`` replies and on the topology
+notifications. A provider-reported change advances the generation before its
+notification is serialized, so an event carries the post-change value that a
+later ``get``/``dump`` will also report.
+
+``topology-generation`` is a change token, not a timestamp, liveness counter or
+event count: a changed value means topology changed, but the delta between two
+values has no defined meaning and the counter may wrap (it skips zero). Statistics
+reads do not advance it.
+
+The same value backs dump consistency. Every multipart dump samples it and calls
+``genl_dump_check_consistent()``; if it changes between dump batches, Generic
+Netlink marks the dump with ``NLM_F_DUMP_INTR``, meaning the snapshot may be torn.
+User space must then discard the partial result and retry the complete dump. The
+generation does not apply to statistics reads.
+
+Only a dump that serialized at least one entry can carry ``NLM_F_DUMP_INTR``,
+because Generic Netlink arms the consistency check on the first entry it emits.
+A dump that yields no entries at all cannot report interruption, so user space
+should treat an empty result as advisory and re-read ``topology-generation``
+before concluding that the topology is empty.
+
+Namespaces
+----------
+
+DRM Fabric objects describe host-global hardware and are not scoped per network
+namespace. Query and dump operations (``*-get``) are therefore accepted only from
+the initial network namespace; a request from any other network namespace fails
+with ``-EPERM``. The Generic Netlink family is registered ``netnsok`` (so it
+resolves in any network namespace and can return that policy error rather than
+being invisible), but the operations themselves remain confined to ``init_net``.
+Monitor notifications are likewise emitted only into ``init_net``, so a listener
+that joins the multicast group from another network namespace never receives
+them.
+
+Examples
+========
+
+Query the topology with the in-tree YNL tool, pointing it at the spec:
+
+.. code-block:: bash
+
+    # List all fabrics
+    ./tools/net/ynl/pyynl/cli.py \
+        --spec Documentation/netlink/specs/drm_fabric.yaml \
+        --dump fabric-get
+
+Replies follow the shapes described above; a provider must be registered for
+the topology to be non-empty.
+
+List the endpoints of a fabric:
+
+.. code-block:: bash
+
+    ./tools/net/ynl/pyynl/cli.py \
+        --spec Documentation/netlink/specs/drm_fabric.yaml \
+        --dump endpoint-get --json '{"fabric-id": 1}'
+
+Query a single port:
+
+.. code-block:: bash
+
+    ./tools/net/ynl/pyynl/cli.py \
+        --spec Documentation/netlink/specs/drm_fabric.yaml \
+        --do port-get --json '{"endpoint-id": 1, "port-index": 0}'
+
+The family name on the wire is ``drm-fabric``.

index 21fbfae9863d..7e9c569fd258 100644
@@ -4,9 +4,9 @@ config DRM_FABRIC
 	tristate "DRM fabric support"
 	depends on DRM && NET
 	help
-	  Enable DRM fabric support. This infrastructure provides the
-	  core object model and provider API for registered accelerator
-	  interconnect topologies.
+	  Enable DRM fabric support. This infrastructure exposes
+	  registered accelerator interconnect topologies to userspace
+	  through the drm-fabric generic netlink family.
 
 	  To compile this as a module, choose M here: the module will be
 	  called drm-fabric.

index 3a76f31f1e83..cf9d9d6be3d3 100644
@@ -1,4 +1,4 @@
 # SPDX-License-Identifier: GPL-2.0
 
 obj-$(CONFIG_DRM_FABRIC) += drm-fabric.o
-drm-fabric-y := drm_fabric.o
+drm-fabric-y := drm_fabric.o drm_fabric_netlink.o drm_fabric_nl.o

index a8b33d6cd618..8bd3633d41be 100644
@@ -291,10 +291,16 @@ Query the topology with the in-tree YNL tool, pointing it at the spec:
         --spec Documentation/netlink/specs/drm_fabric.yaml \
         --dump fabric-get
 
-Replies follow the shapes described above; a provider must be registered for
-the topology to be non-empty.
+Against drm_fabric_sim loaded with ``topology=linear num_endpoints=2``
+(both non-default, for a minimal example), this returns:
 
-List the endpoints of a fabric:
+.. code-block:: text
+
+    [{'fabric': {'fabric-id': 1, 'instance-id': 2156317438,
+                 'name': 'fabricsim', 'type': 'synthetic'},
+      'topology-generation': 18}]
+
+List the endpoints of that fabric:
 
 .. code-block:: bash
 

@@ -302,6 +308,15 @@ List the endpoints of a fabric:
         --spec Documentation/netlink/specs/drm_fabric.yaml \
         --dump endpoint-get --json '{"fabric-id": 1}'
 
+.. code-block:: text
+
+    [{'endpoint': {'bus-name': 'platform', 'dev-name': 'fabricsim.0',
+                   'endpoint-id': 0, 'fabric-ep-id': 256, 'fabric-id': 1,
+                   'name': 'sim-ep0'}, 'topology-generation': 18},
+     {'endpoint': {'bus-name': 'platform', 'dev-name': 'fabricsim.1',
+                   'endpoint-id': 1, 'fabric-ep-id': 257, 'fabric-id': 1,
+                   'name': 'sim-ep1'}, 'topology-generation': 18}]
+
 Query a single port:
 
 .. code-block:: bash

@@ -310,4 +325,29 @@ Query a single port:
         --spec Documentation/netlink/specs/drm_fabric.yaml \
         --do port-get --json '{"endpoint-id": 1, "port-index": 0}'
 
+.. code-block:: text
+
+    {'port': {'endpoint-id': 1, 'max-lane-count': 4,
+              'max-lane-signaling-rate-mbps': 200000, 'oper-state': 'active',
+              'peer': {'peer-id': 256, 'port-index': 0, 'type': 'accel'},
+              'port-index': 0},
+     'topology-generation': 18}
+
 The family name on the wire is ``drm-fabric``.
+
+Synthetic provider
+==================
+
+``CONFIG_DRM_FABRIC_SIM`` builds ``drm-fabric-sim.ko``, a software-only provider
+modeled on netdevsim (Documentation/networking/devlink/netdevsim.rst) that drives
+the object model and uAPI without real hardware. Module parameters select a
+linear, mesh or switch-shaped topology and bound the number of endpoints and
+ports. The switch shape links every endpoint to an opaque switch peer
+(``peer-type = switch``) whose id does not resolve to an endpoint, exercising the
+directed half-edge model without a first-class switch object.
+
+Its debugfs knobs stimulate synthetic counter activity, operational-state changes
+and runtime endpoint add/remove. These files are unstable test controls and are
+not part of the uAPI; the stable, reviewed interface is the YAML-described
+Generic Netlink family. Tests mutate simulator state through debugfs and observe
+the result over Generic Netlink.

index 7e9c569fd258..87115356baca 100644
@@ -12,3 +12,13 @@ config DRM_FABRIC
 	  called drm-fabric.
 
 	  If in doubt, say N.
+
+config DRM_FABRIC_SIM
+	tristate "DRM fabric synthetic provider (test-only)"
+	depends on DRM_FABRIC
+	help
+	  Synthetic provider for testing drm_fabric topology and ABI.
+	  Provides debugfs-only hooks for synthetic activity generation,
+	  error injection, and port-state transitions.  These hooks are
+	  NOT part of the drm_fabric uAPI and are used only by selftests.
+	  Does not model UALink protocol traffic or memory semantics.

index cf9d9d6be3d3..bc0a6c742164 100644
@@ -2,3 +2,6 @@
 
 obj-$(CONFIG_DRM_FABRIC) += drm-fabric.o
 drm-fabric-y := drm_fabric.o drm_fabric_netlink.o drm_fabric_nl.o
+
+obj-$(CONFIG_DRM_FABRIC_SIM) += drm-fabric-sim.o
+drm-fabric-sim-y := drm_fabric_sim.o

index 8bd3633d41be..cd98a97e1662 100644
@@ -351,3 +351,9 @@ and runtime endpoint add/remove. These files are unstable test controls and are
 not part of the uAPI; the stable, reviewed interface is the YAML-described
 Generic Netlink family. Tests mutate simulator state through debugfs and observe
 the result over Generic Netlink.
+
+Testing
+=======
+
+The object model is covered by KUnit when ``CONFIG_DRM_FABRIC_KUNIT_TEST`` is
+enabled. The test source is folded into the core translation unit.

new file mode 100644
index 000000000000..d6087f45f2ac
@@ -0,0 +1,5 @@
+CONFIG_KUNIT=y
+CONFIG_NET=y
+CONFIG_DRM=y
+CONFIG_DRM_FABRIC=y
+CONFIG_DRM_FABRIC_KUNIT_TEST=y

new file mode 100644
index 000000000000..8add027fc766
@@ -0,0 +1,16 @@
+CONFIG_KUNIT=y
+CONFIG_NET=y
+CONFIG_DRM=y
+CONFIG_DRM_FABRIC=y
+CONFIG_DRM_FABRIC_KUNIT_TEST=y
+CONFIG_DEBUG_KERNEL=y
+CONFIG_KASAN=y
+CONFIG_KASAN_GENERIC=y
+CONFIG_DEBUG_KMEMLEAK=y
+CONFIG_PROVE_LOCKING=y
+CONFIG_DEBUG_ATOMIC_SLEEP=y
+CONFIG_UBSAN=y
+CONFIG_UBSAN_BOUNDS=y
+CONFIG_UBSAN_SHIFT=y
+CONFIG_UBSAN_ENUM=y
+CONFIG_UBSAN_BOOL=y

index 87115356baca..2a70cac85b1d 100644
@@ -22,3 +22,19 @@ config DRM_FABRIC_SIM
 	  error injection, and port-state transitions.  These hooks are
 	  NOT part of the drm_fabric uAPI and are used only by selftests.
 	  Does not model UALink protocol traffic or memory semantics.
+
+config DRM_FABRIC_KUNIT_TEST
+	bool "DRM fabric object-model KUnit tests" if !KUNIT_ALL_TESTS
+	depends on DRM_FABRIC && KUNIT
+	depends on KUNIT=y || DRM_FABRIC=m
+	default KUNIT_ALL_TESTS
+	help
+	  Enable KUnit coverage for the drm_fabric object model.
+
+	  The tests are built into drm_fabric itself, so they need no exported
+	  symbols or test-only accessors in the production source.
+
+	  For more information on KUnit and unit tests in general, please refer
+	  to the KUnit documentation in Documentation/dev-tools/kunit/.
+
+	  If unsure, say N.

index bc0a6c742164..05f1c96bb8de 100644
@@ -5,3 +5,5 @@ drm-fabric-y := drm_fabric.o drm_fabric_netlink.o drm_fabric_nl.o
 
 obj-$(CONFIG_DRM_FABRIC_SIM) += drm-fabric-sim.o
 drm-fabric-sim-y := drm_fabric_sim.o
+
+# The tests are included in drm_fabric.o, not built as a separate object.

index cd98a97e1662..bc7b87c766cc 100644
@@ -357,3 +357,10 @@ Testing
 
 The object model is covered by KUnit when ``CONFIG_DRM_FABRIC_KUNIT_TEST`` is
 enabled. The test source is folded into the core translation unit.
+
+Generic Netlink ABI tests live under
+``tools/testing/selftests/drivers/gpu/drm_fabric``. They cover the YNL query
+paths, malformed policy input, generated-header synchronization, dump-cursor
+correctness across endpoint removal, ``NLM_F_DUMP_INTR`` handling, the opaque
+switch half-edge, and provider fault handling. See that directory's ``README.rst``
+for build and execution commands.

index b622052ec3e9..c90eb5d33ec6 100644
@@ -18,6 +18,7 @@ TARGETS += devices/error_logs
 TARGETS += devices/probe
 TARGETS += dmabuf-heaps
 TARGETS += drivers/dma-buf
+TARGETS += drivers/gpu/drm_fabric
 TARGETS += drivers/ntsync
 TARGETS += drivers/s390x/uvdevice
 TARGETS += drivers/net

41 lines not on this page · raw message

137 lines not on this page · raw message

114 lines not on this page · raw message

new file mode 100644
index 000000000000..6eaab8a7d771
@@ -0,0 +1,8 @@
+# SPDX-License-Identifier: GPL-2.0
+# Kernel config fragment required to run the drm_fabric kselftests.
+# Merge with: scripts/kconfig/merge_config.sh or make kselftest-merge.
+CONFIG_NET=y
+CONFIG_DRM=y
+CONFIG_DEBUG_FS=y
+CONFIG_DRM_FABRIC=m
+CONFIG_DRM_FABRIC_SIM=m

361 lines not on this page · raw message

178 lines not on this page · raw message

401 lines not on this page · raw message

105 lines not on this page · raw message

113 lines not on this page · raw message

204 lines not on this page · raw message

481 lines not on this page · raw message

485 lines not on this page · raw message

355 lines not on this page · raw message

374 lines not on this page · raw message

new file mode 100644
index 000000000000..694d70710ff0

Lines of this file are not shown. View the raw message.

108 lines not on this page · raw message

index bc7b87c766cc..61e2c18bf0b5 100644
@@ -18,6 +18,8 @@ Key Goals:
   (xGMI, UALink and similar), enabling data-center discovery and monitoring.
 * Support read-only enumeration, monitoring and state queries for
   provider-owned topology.
+* Support topology provisioning as an optional provider capability, covering
+  fabric and endpoint lifecycle, port administration and peer management.
 * Allow new attributes and fabric types to be added without reusing existing
   wire identifiers, so the uAPI extends without breaking existing consumers.
 * Allow multiple endpoints and ports per provider, so drivers can model

@@ -56,13 +58,14 @@ provider reports: for example, it may be a full mesh with no root, a linear
 chain, or a switch-based topology in which ports terminate at opaque switch
 peers rather than locally registered endpoints.
 
-A *peer* is a value descriptor, not a reference to a live kernel object: its
-``peer-id`` may name a remote accelerator managed by another OS or an opaque
-switch in another trust domain, and need not resolve in the local registry. The
-core stores one directed half-edge and does not require the reverse half-edge to
-exist, so removing an endpoint does not retract peer descriptors held by other
-endpoints. ``peer-type = switch`` only describes the kind of far end; it does
-not create a first-class switch object.
+An endpoint may be registered without a fabric. Such an endpoint is an *orphan*
+and reports ``fabric-id`` 0. A *peer* is a value descriptor, not a reference to
+a live kernel object: its ``peer-id`` may name a remote accelerator managed by
+another OS or an opaque switch in another trust domain, and need not resolve in
+the local registry. The core stores one directed half-edge and does not require
+the reverse half-edge to exist, so removing an endpoint does not retract peer
+descriptors held by other endpoints. ``peer-type = switch`` only describes the
+kind of far end; it does not create a first-class switch object.
 
 .. kernel-doc:: drivers/gpu/drm/fabric/drm_fabric.c
    :doc: DRM Fabric core

@@ -79,10 +82,10 @@ that half-edge. A peer is therefore topology as last set, not proof of live
 connectivity; liveness belongs to the fabric controller.
 
 The core never retracts a half-edge on its own. Failing to resolve a peer
-locally is not the same as the link going away -- the far end may be a switch,
-an accelerator on another node, or a local endpoint that merely unregistered
--- so only the provider knows when a port's physical adjacency actually
-changed, and only the provider retracts or replaces the descriptor.
+locally is not the same as the link going away -- the far end may be a
+switch, an accelerator on another node, or a local endpoint that merely
+unregistered -- so only the provider knows when a port's physical adjacency
+actually changed, and only the provider retracts or replaces the descriptor.
 
 ``port-peer-delete-ntf`` reports an explicitly retracted half-edge; it is
 not emitted when a peer merely becomes locally unresolvable, so its absence

@@ -105,11 +108,15 @@ Driver API
 Design scope and boundaries
 ===========================
 
-Vendor drivers retain hardware discovery, firmware interaction and the
-load/store data path; DRM Fabric represents only the topology and
-provider-reported state of DRM-managed accelerators, which is why it
-belongs in DRM. The interface does not define MMU programming, switch
-policy, key management, live migration, or any required user space daemon.
+The core records direct adjacency only, not end-to-end reachability or switch
+forwarding, which remain with the fabric controller. Vendor drivers retain
+hardware discovery, firmware interaction, memory semantics and the
+hardware-carried data path; DRM Fabric represents only the topology and
+control state of DRM-managed accelerators, which is why it belongs in DRM.
+It does not create a network device or own route computation, switch
+forwarding, transport or congestion control. The interface also does not
+define MMU programming, switch policy, key management, live migration, or
+any required user space daemon.
 
 DRM Fabric does not define in-network collective operations or how an
 endpoint or switch executes them. Such capabilities belong to the

@@ -123,27 +130,49 @@ strengthens the contract rather than breaking it.
 Object lifetime and locking
 ===========================
 
-All registry and object state is protected by ``drm_fabric_lock``. A fabric and
-its endpoints are created and torn down through the provider API; endpoint
-unregister removes the endpoint from the registry and frees its fixed set of
-ports. Fabric membership is tracked, so a provider must remove all member
-endpoints before unregistering a provider-owned fabric: drm_fabric_unregister()
-returns ``-EBUSY`` and leaves the fabric registered if any remain, so the
-provider must retry after removing them rather than treat the fabric as gone.
+Registry membership and object state are protected by ``drm_fabric_lock``;
+topology mutation is additionally serialised by ``drm_fabric_mutation_lock``,
+described below. A fabric and its endpoints are created and torn down through
+the provider API; endpoint unregister removes the endpoint from the registry
+and frees its fixed set of ports. Fabric membership is tracked, so a provider
+must remove all member endpoints before unregistering a provider-owned fabric:
+drm_fabric_unregister() returns ``-EBUSY`` and leaves the fabric registered if
+any remain, so the provider must retry after removing them rather than treat
+the fabric as gone.
 
 Objects are reference counted and a port is pinned through its owning endpoint.
 Endpoint unregister drops the registration reference and waits for outstanding
 pins before freeing the ports; fabric membership holds a fabric reference.
 
-Providers own object lifetime, so a provider must serialise endpoint
-registration against unregistration of the containing fabric. The unregister
-entry points compare the supplied pointer against the registry before
-dereferencing it, so a stale or repeated teardown is rejected: fabric
-unregistration returns ``-ENODEV``, and endpoint unregistration, having no
-error return, warns and performs no teardown. Endpoint registration rejects a
-departed parent the same way. These checks prove current address membership
-only: they cannot tell an earlier incarnation from another object registered
-later at the same address.
+Providers own object lifetime. The unregister entry points compare the supplied
+pointer against the registry before dereferencing it, so a stale or repeated
+teardown is rejected: fabric unregistration returns ``-ENODEV``, and endpoint
+unregistration, having no error return, warns and performs no teardown.
+Endpoint registration rejects a departed parent the same way. These checks
+prove current address membership only: they cannot tell an earlier incarnation
+from another object registered later at the same address. Only the provider
+knows its own object lifecycle.
+
+Netlink mutation commands additionally hold ``drm_fabric_mutation_lock`` across
+target resolution, the provider callback and the core commit. The lock order
+is::
+
+    drm_fabric_mutation_lock -> drm_fabric_lock
+
+Provider lifecycle operations that can invalidate a prepared mutation take the
+same mutation lock: endpoint registration as well as endpoint and fabric
+unregister. A provider-driven teardown therefore cannot race an in-flight
+mutation on the same object, and a registration cannot claim a fabric-scoped
+``fabric-ep-id`` after an attach has validated the identifier but before its
+provider callback completes. A provider must not invoke any of these lifecycle
+operations from one of its own mutation callbacks: the callback already holds
+the mutation lock, so the nested acquisition would self-deadlock. A successful
+callback therefore cannot be invalidated before commit, and the post-callback
+checks are invariant assertions only.
+
+A fabric records whether it was created by a provider or by userspace. Only an
+empty userspace-created fabric may be deleted through the provisioning core;
+provider-owned and non-empty fabrics are rejected.
 
 Generic Netlink family
 ======================

index 61e2c18bf0b5..4ee7cb7ef6ac 100644
@@ -87,9 +87,11 @@ switch, an accelerator on another node, or a local endpoint that merely
 unregistered -- so only the provider knows when a port's physical adjacency
 actually changed, and only the provider retracts or replaces the descriptor.
 
-``port-peer-delete-ntf`` reports an explicitly retracted half-edge; it is
-not emitted when a peer merely becomes locally unresolvable, so its absence
-is not evidence the far end is still reachable.
+``port-peer-delete-ntf`` reports an explicitly retracted half-edge -- through
+the provider's own report or, on a userspace-managed port, a userspace
+``port-peer-del`` request -- and is never emitted when a peer merely becomes
+locally unresolvable, so its absence is not evidence the far end is still
+reachable.
 
 Endpoint teardown removes the endpoint's owned half-edges without
 generating a separate event for each port: the delete already describes

@@ -170,9 +172,9 @@ the mutation lock, so the nested acquisition would self-deadlock. A successful
 callback therefore cannot be invalidated before commit, and the post-callback
 checks are invariant assertions only.
 
-A fabric records whether it was created by a provider or by userspace. Only an
-empty userspace-created fabric may be deleted through the provisioning core;
-provider-owned and non-empty fabrics are rejected.
+A fabric records whether it was created by a provider
+(drm_fabric_register()) or by userspace (``fabric-new``); see
+`Provisioning`_ for the deletion rules.
 
 Generic Netlink family
 ======================

@@ -216,8 +218,9 @@ User space enumerates topology with four read-only commands, each supporting a
 single lookup (``do``) and a bulk dump (``dump``):
 
 * ``fabric-get`` -- enumerate fabrics (``do`` by ``fabric-id``, ``dump`` for all).
-* ``endpoint-get`` -- enumerate endpoints, optionally filtered by ``fabric-id``,
-  or resolve one by ``endpoint-id`` or backing ``dev-name``/``bus-name``.
+* ``endpoint-get`` -- enumerate endpoints, optionally filtered by ``fabric-id``
+  (0 selects orphans), or resolve one by ``endpoint-id`` or by its backing
+  ``dev-name`` with an optional ``bus-name``.
 * ``port-get`` -- enumerate ports, filtered by ``endpoint-id``. Ports may report
   the provider's maximum capability as ``max-lane-count`` and
   ``max-lane-signaling-rate-mbps`` (zero means unknown); these are maxima, not

@@ -243,10 +246,10 @@ a partial payload and are declared with ``event:``.
      - Trigger
      - Payload
    * - ``fabric-create-ntf``
-     - a fabric is registered by a provider
+     - a fabric is registered (provider or ``fabric-new``)
      - reuses ``fabric-get``
    * - ``fabric-delete-ntf``
-     - a provider-owned fabric is unregistered
+     - a fabric is unregistered or removed via ``fabric-del``
      - reuses ``fabric-get``
    * - ``endpoint-create-ntf``
      - an endpoint is registered

@@ -254,15 +257,19 @@ a partial payload and are declared with ``event:``.
    * - ``endpoint-delete-ntf``
      - an endpoint is unregistered
      - reuses ``endpoint-get``
+   * - ``endpoint-change-ntf``
+     - an endpoint's fabric attachment or administrative state changes
+     - reuses ``endpoint-get``
    * - ``port-change-ntf``
-     - a port's operational state changes
+     - a port's operational or administrative state changes
      - reuses ``port-get``
    * - ``port-peer-create-ntf``
-     - a port's peer descriptor is set by its provider
+     - a port's peer descriptor is set (provider or ``port-peer-new``)
      - partial (``event:``)
    * - ``port-peer-delete-ntf``
-     - a port's peer descriptor is explicitly unset by its provider; never
-       emitted for an implicit half-edge loss (see `Peer semantics`_)
+     - a port's peer descriptor is explicitly unset (provider or
+       ``port-peer-del``); never emitted for an implicit half-edge loss
+       (see `Peer semantics`_)
      - partial (``event:``)
 
 Notifications are best-effort. A listener that detects loss, restarts, or

@@ -274,9 +281,9 @@ Topology generation and dump consistency
 The core keeps a nonzero generation counter and advances it whenever topology or
 exposed state changes. It is surfaced as the ``topology-generation`` attribute on
 ``fabric-get``, ``endpoint-get`` and ``port-get`` replies and on the topology
-notifications. A provider-reported change advances the generation before its
-notification is serialized, so an event carries the post-change value that a
-later ``get``/``dump`` will also report.
+notifications. A committed change -- provider-reported or userspace mutation
+-- advances the generation before its notification is serialized, so an event
+carries the post-change value that a later ``get``/``dump`` will also report.
 
 ``topology-generation`` is a change token, not a timestamp, liveness counter or
 event count: a changed value means topology changed, but the delta between two

@@ -308,8 +315,8 @@ Monitor notifications are likewise emitted only into ``init_net``, so a listener
 that joins the multicast group from another network namespace never receives
 them.
 
-Examples
-========
+Query examples
+--------------
 
 Query the topology with the in-tree YNL tool, pointing it at the spec:
 

@@ -364,6 +371,78 @@ Query a single port:
 
 The family name on the wire is ``drm-fabric``.
 
+Provisioning
+============
+
+The provisioning interface requires ``CAP_NET_ADMIN``. ``fabric-new`` and
+``fabric-del`` are core-owned; the rest are serviced by the provider and
+return ``-EOPNOTSUPP`` where it does not implement the matching callback:
+
+* ``fabric-new`` -- create an empty, userspace-owned fabric. The request
+  carries a ``fabric-new-params`` nest with the fabric type, an optional name
+  and a provider-defined instance ID.
+* ``fabric-del`` -- delete an empty, userspace-owned fabric. Refuses a
+  provider-owned fabric (``-EPERM``) and a non-empty one (``-EBUSY``).
+* ``endpoint-set`` -- attach an orphan endpoint, selected by ``endpoint-id`` or
+  by its backing ``dev-name`` with an optional ``bus-name``, to a fabric;
+  detach it with ``fabric-id`` 0; and/or set its administrative state.
+* ``port-set`` -- set a port's administrative state.
+* ``port-peer-new`` / ``port-peer-del`` -- set or unset a port's neighbor
+  half-edge. Valid only on a userspace-managed port; a provider-managed port
+  returns ``-EOPNOTSUPP`` (see `Peer management mode`_).
+
+Mutation commands carry ``GENL_ADMIN_PERM`` and are additionally confined to
+``init_net``. ``GENL_ADMIN_PERM`` only checks ``CAP_NET_ADMIN`` in the caller's
+user namespace, which a process in a nested user and network namespace may
+hold; restricting mutators to ``init_net`` prevents container-local privilege
+from reconfiguring host-global topology.
+
+A userspace-created fabric additionally pins this module for as long as it
+remains registered: ``fabric-new`` takes the pin, ``fabric-del`` releases it.
+
+Peer management mode
+--------------------
+
+A port uses one peer-management mode, fixed by the provider at registration.
+
+In provider-managed mode, the provider reports peer creation and removal
+(drm_fabric_port_set_peer() / drm_fabric_port_unset_peer()) and userspace peer
+provisioning is unsupported (``port-peer-new`` / ``port-peer-del`` return
+``-EOPNOTSUPP``).
+
+In userspace-managed mode, peer creation and removal are requested through the
+provisioning API and the provider does not independently replace or retract the
+peer (drm_fabric_port_set_peer() / drm_fabric_port_unset_peer() return
+``-EOPNOTSUPP``).
+
+One authority per descriptor prevents concurrent updates from both sources.
+Setting an occupied port returns ``-EEXIST``; unsetting an empty port returns
+``-ENOENT``.
+
+Provisioning examples
+---------------------
+
+Provision with the in-tree YNL tool (root, initial namespace):
+
+.. code-block:: bash
+
+    # Create an empty fabric
+    ./tools/net/ynl/pyynl/cli.py \
+        --spec Documentation/netlink/specs/drm_fabric.yaml \
+        --do fabric-new \
+        --json '{"fabric-new-params": {"type": "synthetic", "instance-id": 42}}'
+
+The command returns a core-assigned ``fabric-id``. Use that value in the
+following requests.
+
+.. code-block:: bash
+
+    # Attach an orphan endpoint to the returned fabric
+    ./tools/net/ynl/pyynl/cli.py \
+        --spec Documentation/netlink/specs/drm_fabric.yaml \
+        --do endpoint-set \
+        --json '{"endpoint-id": 5, "fabric-id": <returned-fabric-id>}'
+
 Synthetic provider
 ==================
 

203 lines not on this page · raw message

index 4ee7cb7ef6ac..1fd48027eee3 100644
@@ -454,6 +454,10 @@ ports. The switch shape links every endpoint to an opaque switch peer
 (``peer-type = switch``) whose id does not resolve to an endpoint, exercising the
 directed half-edge model without a first-class switch object.
 
+It implements the provisioning callbacks, so it is also the reference provider
+for the mutation commands. It reserves each endpoint's last port for userspace
+peer management and wires topology on the ports below it.
+
 Its debugfs knobs stimulate synthetic counter activity, operational-state changes
 and runtime endpoint add/remove. These files are unstable test controls and are
 not part of the uAPI; the stable, reviewed interface is the YAML-described

index 2a70cac85b1d..7f427ac42a27 100644
@@ -29,7 +29,8 @@ config DRM_FABRIC_KUNIT_TEST
 	depends on KUNIT=y || DRM_FABRIC=m
 	default KUNIT_ALL_TESTS
 	help
-	  Enable KUnit coverage for the drm_fabric object model.
+	  KUnit tests for the drm_fabric object model and topology-mutation
+	  paths.
 
 	  The tests are built into drm_fabric itself, so they need no exported
 	  symbols or test-only accessors in the production source.

index 1fd48027eee3..e400c44d32b8 100644
@@ -474,5 +474,6 @@ Generic Netlink ABI tests live under
 ``tools/testing/selftests/drivers/gpu/drm_fabric``. They cover the YNL query
 paths, malformed policy input, generated-header synchronization, dump-cursor
 correctness across endpoint removal, ``NLM_F_DUMP_INTR`` handling, the opaque
-switch half-edge, and provider fault handling. See that directory's ``README.rst``
+switch half-edge, ``CAP_NET_ADMIN`` gating and provisioning rejects, and
+provider fault-injection failure atomicity. See that directory's ``README.rst``
 for build and execution commands.

index 54d756979d97..6cdf14c44a5a 100644
@@ -18,7 +18,10 @@ TEST_PROGS := \
 	hotplug_abi.py \
 	dump_scale_abi.py \
 	switch_abi.py \
+	cap_netadmin.py \
+	netns_abi.py \
 	fault_abi.py \
+	provisioning_scenarios_abi.py \
 	harness_reset_abi.py
 
 TEST_FILES := lib_drm_fabric.py

index 6c24581db1f2..33d9618b31d5 100644
@@ -4,7 +4,7 @@
 drm_fabric selftests
 ====================
 
-These selftests exercise the ``drm-fabric`` query uAPI against
+These selftests exercise the ``drm-fabric`` query and mutation uAPI against
 ``drm_fabric_sim`` using the in-tree YNL library. KUnit covers the core object
 model.
 

@@ -45,7 +45,7 @@ Suites
   ends the dump.
 
 ``hotplug_abi.py``
-  Endpoint hotplug: CREATE/DELETE notifications.
+  Endpoint hotplug: CREATE/DELETE NTFs and mutation round-trips.
 
 ``dump_scale_abi.py``
   Dump resume under many endpoints (``bulk_add``).

@@ -54,10 +54,19 @@ Suites
   Opaque switch peers whose identifiers do not resolve to an endpoint
   (``topology=switch``).
 
+``cap_netadmin.py``
+  ``CAP_NET_ADMIN`` enforcement for mutation commands.
+
+``netns_abi.py``
+  Rejects commands outside ``init_net``, including with ``CAP_NET_ADMIN``.
+
 ``fault_abi.py``
-  Provider fault injection: errno propagation and no leaked endpoint
+  Provider failures: errno propagation, rollback and no notification
   (``fail_*``).
 
+``provisioning_scenarios_abi.py``
+  Endpoint, port and peer provisioning scenarios.
+
 ``harness_reset_abi.py``
   Recovery after a SIGKILL-terminated predecessor.
 

@@ -71,6 +80,7 @@ A SKIP means a required precondition was unavailable.
 
 Environment
   ``check-spec-regen.sh`` needs PyYAML and writable temporary storage.
+  ``netns_abi.py`` needs ``CONFIG_NET_NS``.
 
 Per case
   A case skips when a required control, parameter or family capability is

@@ -81,7 +91,7 @@ Whole suite
 
 Timing
   The two ``dump_intr_abi.py`` boundary cases may skip if the concurrent
-  topology change misses the required dump boundary.
+  mutation misses the required dump boundary.
 
 KUnit
 -----

@@ -133,4 +143,4 @@ Build out-of-tree, boot with ``vng`` and run the same target in the guest:
        make -C tools/testing/selftests TARGETS=drivers/gpu/drm_fabric run_tests
 
 Dependencies (Debian/Ubuntu): ``python3``, ``python3-yaml``,
-``qemu-system-x86``, ``virtme-ng`` (``pip install --user virtme-ng``).
\ No newline at end of file
+``qemu-system-x86``, ``virtme-ng`` (``pip install --user virtme-ng``).

314 lines not on this page · raw message

index 6eaab8a7d771..f7b38c11a5a9 100644
@@ -2,6 +2,11 @@
 # Kernel config fragment required to run the drm_fabric kselftests.
 # Merge with: scripts/kconfig/merge_config.sh or make kselftest-merge.
 CONFIG_NET=y
+# netns_abi.py drives the init_net restriction from a non-initial namespace.
+# USER_NS lets it model container root; without it the suite still runs, using
+# a network namespace alone.
+CONFIG_NET_NS=y
+CONFIG_USER_NS=y
 CONFIG_DRM=y
 CONFIG_DEBUG_FS=y
 CONFIG_DRM_FABRIC=m

194 lines not on this page · raw message

250 lines not on this page · raw message

185 lines not on this page · raw message

index 30fb0edb02b9..9e5dfd6a6ba9 100644
@@ -267,6 +267,16 @@ def family_has_op(fab, name):
     return name in getattr(fab, "ops", {})
 
 
+def select_cases(fab, cases, mutation_cases, probe="fabric-new"):
+    """Return @cases, dropping @mutation_cases when @probe (a representative
+    mutation op) is absent from the family.
+    """
+    if family_has_op(fab, probe):
+        return tuple(cases)
+    drop = set(mutation_cases)
+    return tuple(c for c in cases if c not in drop)
+
+
 # System helpers (kselftest runs as root)
 
 def is_root():

@@ -462,6 +472,13 @@ def fabricsim(ksft, topology=None, need_debugfs=False, need_control=None,
         if not insmod("drm-fabric.ko") or not insmod("drm-fabric-sim.ko"):
             ksft.skip_all("could not load drm_fabric + drm_fabric_sim modules")
         wait_until(lambda: module_loaded("drm_fabric_sim"))
+    else:
+        # Running against providers somebody else loaded (--no-load, or a
+        # previous suite that restored the sim but kept the core). There is no
+        # module to unwind, but the suite can still add endpoints and peers,
+        # and without a teardown that state would leak into the next suite and
+        # survive the timeout killer's SIGTERM. Restore the default shape.
+        on_teardown(sim_restore_default)
 
     if not module_loaded("drm_fabric_sim"):
         ksft.skip_all("drm_fabric_sim not loaded")

294 lines not on this page · raw message

395 lines not on this page · raw message

324 lines not on this page · raw message

index 775be4ac2160..152d16cdc464 100755
@@ -4,7 +4,9 @@
 """
 fabricsim's "switch" shape links each leaf's first port to an opaque
 switch that is not a registered endpoint: asserts half-edge serialization
-and peer-id non-resolution, not leaf-switch-leaf reachability.
+and peer-id non-resolution, not leaf-switch-leaf reachability. A final case
+reloads at a one-port-per-endpoint request, where the port reserved for
+userspace peers would otherwise consume the only half-edge.
 
 --no-load is ignored (needs a fresh insmod). Run as root.
 """

@@ -78,11 +80,38 @@ def test_switch_id_does_not_resolve(ksft, cfg):
                % sorted(leaked))
 
 
+def test_minimum_request_preserves_switch_wiring(ksft, cfg):
+    """A one-port-per-endpoint request must still leave the switch wired:
+    setup raises the count so the reserved userspace port does not consume
+    the only half-edge.
+
+    Runs last: it reloads the sim, invalidating the snapshot above.
+    """
+    fab = cfg.fab
+    L.rmmod("drm_fabric_sim")
+    if not L.insmod("drm-fabric-sim.ko", "topology=switch", "ports_per_ep=1"):
+        ksft.skip("switch-minimum-request-wired",
+                  "could not load sim with ports_per_ep=1")
+        return
+    if not L.wait_until(lambda: L.module_loaded("drm_fabric_sim")):
+        ksft.skip("switch-minimum-request-wired", "sim did not reappear")
+        return
+
+    eps = [e["endpoint"] for e in fab.dump("endpoint-get", {})]
+    sim_eps = [e for e in eps if e["name"].startswith("sim-ep")]
+    peers = switch_peers(fab, sim_eps)
+    ksft.check(bool(sim_eps) and len(peers) == len(sim_eps),
+               "switch-minimum-request-wired",
+               "ports_per_ep=1: switch-peers=%d leaves=%d"
+               % (len(peers), len(sim_eps)))
+
+
 CASES = (
     test_every_leaf_has_switch_peer,
     test_half_edge_fully_serialized,
     test_single_opaque_switch_id,
     test_switch_id_does_not_resolve,
+    test_minimum_request_preserves_switch_wiring,
 )