acpi: report NUMA nodes for device memory using GI
v9 cover acpi: report NUMA nodes for device memory using GI
From: Ankit Agrawal <[email protected]>
There are upcoming devices which allow CPU to cache coherently access their memory. It is sensible to expose such memory as NUMA nodes separate from the sysmem node to the OS. The ACPI spec provides a scheme in SRAT called Generic Initiator Affinity Structure [1] to allow an association between a Proximity Domain (PXM) and a Generic Initiator (GI) (e.g. heterogeneous processors and accelerators, GPUs, and I/O devices with integrated compute or DMA engines).
While a single node per device may cover several use cases, it is however insufficient for a full utilization of the NVIDIA GPUs MIG (Mult-Instance GPUs) [2] feature. The feature allows partitioning of the GPU device resources (including device memory) into several (upto 8) isolated instances. Each of the partitioned memory requires a dedicated NUMA node to operate. The partitions are not fixed and they can be created/deleted at runtime.
Linux OS does not provide a means to dynamically create/destroy NUMA nodes and such feature implementation is expected to be non-trivial. The nodes that OS discovers at the boot time while parsing SRAT remains fixed. So we utilize the GI Affinity structures that allows association between nodes and devices. Multiple GI structures per device/BDF is possible, allowing creation of multiple nodes in the VM by exposing unique PXM in each of these structures.
Implement the mechanism to build the GI affinity structures as Qemu currently does not. Introduce a new acpi-generic-initiator object to allow host admin link a device with an associated NUMA node. Qemu maintains this association and use this object to build the requisite GI Affinity Structure.
When multiple NUMA nodes are associated with a device, it is required to create those many number of acpi-generic-initiator objects, each representing a unique device:node association.
Following is one of a decoded GI affinity structure in VM ACPI SRAT.
[0C8h 0200 1] Subtable Type : 05 [Generic Initiator Affinity]
[0C9h 0201 1] Length : 20
[0CAh 0202 1] Reserved1 : 00 [0CBh 0203 1] Device Handle Type : 01 [0CCh 0204 4] Proximity Domain : 00000007
[0D0h 0208 16] Device Handle : 00 00 20 00 00 00 00 00 00 00 00
00 00 00 00 00
[0E0h 0224 4] Flags (decoded below) : 00000001
Enabled : 1
[0E4h 0228 4] Reserved2 : 00000000
[0E8h 0232 1] Subtable Type : 05 [Generic Initiator Affinity]
[0E9h 0233 1] Length : 20
On Grace Hopper systems, an admin will create a range of 8 nodes and associate them with the device using the acpi-generic-initiator object. While a configuration of less than 8 nodes per device is allowed, such configuration will prevent utilization of the feature to the fullest. This setting is applicable to all the Grace+Hopper systems. The following is an example of the Qemu command line arguments to create 8 nodes and link them to the device 'dev0':
-numa node,nodeid=2 -numa node,nodeid=3 -numa node,nodeid=4 \
-numa node,nodeid=5 -numa node,nodeid=6 -numa node,nodeid=7 \
-numa node,nodeid=8 -numa node,nodeid=9 \
-device vfio-pci-nohotplug,host=0009:01:00.0,bus=pcie.0,addr=04.0,rombar=0,id=dev0 \
-object acpi-generic-initiator,id=gi0,pci-dev=dev0,node=2 \
-object acpi-generic-initiator,id=gi1,pci-dev=dev0,node=3 \
-object acpi-generic-initiator,id=gi2,pci-dev=dev0,node=4 \
-object acpi-generic-initiator,id=gi3,pci-dev=dev0,node=5 \
-object acpi-generic-initiator,id=gi4,pci-dev=dev0,node=6 \
-object acpi-generic-initiator,id=gi5,pci-dev=dev0,node=7 \
-object acpi-generic-initiator,id=gi6,pci-dev=dev0,node=8 \
-object acpi-generic-initiator,id=gi7,pci-dev=dev0,node=9 \
The performance benefits can be realized by providing the NUMA node distances appropriately (through libvirt tags or Qemu params). The admin can get the distance among nodes in hardware using numactl -H.
This series goes along with the recenty added vfio-pci variant driver [3].
Applied over v8.2.2
base commit: 11aa0b1ff115b86160c4d37e7c37e6a6b13b77ea
[1] ACPI Spec 6.3, Section 5.2.16.6
Link for v8:
v8 -> v9
- Removed unused included headers based on Jonathan's suggestion.
- Collected Reviewed-by from Jonathan.
- Added acpi-generic-initiator support for i386
- Moved HMAT change from patch 1/2 to 2/3.
- Fixed nits.
v7 -> v8
- Replaced the code to collect the acpi-generic-initiator objects with the code to use recursive helper object_child_foreach_recursive based on suggestion from Jonathan Cameron.
- Added sanity check for the node id passed to the acpi-generic-initiator object.
- Added change to use GI as HMAT initiator as per Jonathan's suggestion.
- Fixed nits pointed by Marcus and Jonathan.
- Collected Marcus' Acked-by.
- Rebased to v8.2.2.
v6 -> v7
- Updated code and the commit message to make acpi-generic-initiator define a 1:1 relationship between device and node based on Jonathan Cameron's suggestion.
- Updated commit message to include the decoded GI entry in the SRAT.
- Rebased to v8.2.1.
v5 -> v6
- Updated commit message for the [1/2] and the cover letter.
- Updated the acpi-generic-initiator object comment description for clarity on the input host-nodes.
- Rebased to v8.2.0-rc4.
v4 -> v5
- Removed acpi-dev option until full support.
- The NUMA nodes are saved as bitmap instead of uint16List.
- Replaced asserts to exit calls.
- Addressed other miscellaneous comments.
v3 -> v4
- changed the ':' delimited way to a uint16 array to communicate the nodes associated with the device.
- added asserts to handle invalid inputs.
- addressed other miscellaneous v3 comments.
v2 -> v3
- changed param to accept a ':' delimited list of NUMA nodes, instead of a range.
- Removed nvidia-acpi-generic-initiator object.
- Addressed miscellaneous comments in v2.
v1 -> v2
- Removed dependency on sysfs to communicate the feature with variant module.
- Use GI Affinity SRAT structure instead of Memory Affinity.
- No DSDT entries needed to communicate the PXM for the device. SRAT GI structure is used instead.
- New objects introduced to establish link between device and nodes.
Ankit Agrawal (3):
qom: new object to associate device to NUMA node
hw/acpi: Implement the SRAT GI affinity structure hw/i386/acpi-build: Add support for SRAT Generic Initiator structures
hw/acpi/acpi_generic_initiator.c | 148 +++++++++++++++++++++++ hw/acpi/hmat.c | 2 +- hw/acpi/meson.build | 1 + hw/arm/virt-acpi-build.c | 3 + hw/core/numa.c | 3 +- hw/i386/acpi-build.c | 3 + include/hw/acpi/acpi_generic_initiator.h | 47 +++++++ include/sysemu/numa.h | 1 + qapi/qom.json | 17 +++ 9 files changed, 223 insertions(+), 2 deletions(-) create mode 100644 hw/acpi/acpi_generic_initiator.c create mode 100644 include/hw/acpi/acpi_generic_initiator.h
Signature · 1 line
2.34.1
PATCH v9 1/3 qom: new object to associate device to NUMA node
From: Ankit Agrawal <[email protected]>
NVIDIA GPU's support MIG (Mult-Instance GPUs) feature [1], which allows partitioning of the GPU device resources (including device memory) into several (upto 8) isolated instances. Each of the partitioned memory needs a dedicated NUMA node to operate. The partitions are not fixed and they can be created/deleted at runtime.
Unfortunately Linux OS does not provide a means to dynamically create/destroy NUMA nodes and such feature implementation is not expected to be trivial. The nodes that OS discovers at the boot time while parsing SRAT remains fixed. So we utilize the Generic Initiator (GI) Affinity structures that allows association between nodes and devices. Multiple GI structures per BDF is possible, allowing creation of multiple nodes by exposing unique PXM in each of these structures.
Implement the mechanism to build the GI affinity structures as Qemu currently does not. Introduce a new acpi-generic-initiator object to allow host admin link a device with an associated NUMA node. Qemu maintains this association and use this object to build the requisite GI Affinity Structure.
When multiple NUMA nodes are associated with a device, it is required to create those many number of acpi-generic-initiator objects, each representing a unique device:node association.
Following is one of a decoded GI affinity structure in VM ACPI SRAT.
[0C8h 0200 1] Subtable Type : 05 [Generic Initiator Affinity]
[0C9h 0201 1] Length : 20
[0CAh 0202 1] Reserved1 : 00 [0CBh 0203 1] Device Handle Type : 01 [0CCh 0204 4] Proximity Domain : 00000007
[0D0h 0208 16] Device Handle : 00 00 20 00 00 00 00 00 00 00 00
00 00 00 00 00
[0E0h 0224 4] Flags (decoded below) : 00000001
Enabled : 1
[0E4h 0228 4] Reserved2 : 00000000
[0E8h 0232 1] Subtable Type : 05 [Generic Initiator Affinity]
[0E9h 0233 1] Length : 20
An admin can provide a range of acpi-generic-initiator objects, each associating a device (by providing the id through pci-dev argument) to the desired NUMA node (using the node argument). Currently, only PCI device is supported.
For the grace hopper system, create a range of 8 nodes and associate that with the device using the acpi-generic-initiator object. While a configuration of less than 8 nodes per device is allowed, such configuration will prevent utilization of the feature to the fullest. The following sample creates 8 nodes per PCI device for a VM with 2 PCI devices and link them to the respecitve PCI device using acpi-generic-initiator objects:
-numa node,nodeid=2 -numa node,nodeid=3 -numa node,nodeid=4 \
-numa node,nodeid=5 -numa node,nodeid=6 -numa node,nodeid=7 \
-numa node,nodeid=8 -numa node,nodeid=9 \
-device vfio-pci-nohotplug,host=0009:01:00.0,bus=pcie.0,addr=04.0,rombar=0,id=dev0 \
-object acpi-generic-initiator,id=gi0,pci-dev=dev0,node=2 \
-object acpi-generic-initiator,id=gi1,pci-dev=dev0,node=3 \
-object acpi-generic-initiator,id=gi2,pci-dev=dev0,node=4 \
-object acpi-generic-initiator,id=gi3,pci-dev=dev0,node=5 \
-object acpi-generic-initiator,id=gi4,pci-dev=dev0,node=6 \
-object acpi-generic-initiator,id=gi5,pci-dev=dev0,node=7 \
-object acpi-generic-initiator,id=gi6,pci-dev=dev0,node=8 \
-object acpi-generic-initiator,id=gi7,pci-dev=dev0,node=9 \
-numa node,nodeid=10 -numa node,nodeid=11 -numa node,nodeid=12 \
-numa node,nodeid=13 -numa node,nodeid=14 -numa node,nodeid=15 \
-numa node,nodeid=16 -numa node,nodeid=17 \
-device vfio-pci-nohotplug,host=0009:01:01.0,bus=pcie.0,addr=05.0,rombar=0,id=dev1 \
-object acpi-generic-initiator,id=gi8,pci-dev=dev1,node=10 \
-object acpi-generic-initiator,id=gi9,pci-dev=dev1,node=11 \
-object acpi-generic-initiator,id=gi10,pci-dev=dev1,node=12 \
-object acpi-generic-initiator,id=gi11,pci-dev=dev1,node=13 \
-object acpi-generic-initiator,id=gi12,pci-dev=dev1,node=14 \
-object acpi-generic-initiator,id=gi13,pci-dev=dev1,node=15 \
-object acpi-generic-initiator,id=gi14,pci-dev=dev1,node=16 \
-object acpi-generic-initiator,id=gi15,pci-dev=dev1,node=17 \
--- hw/acpi/acpi_generic_initiator.c | 71 ++++++++++++++++++++++++ hw/acpi/meson.build | 1 + include/hw/acpi/acpi_generic_initiator.h | 22 ++++++++ qapi/qom.json | 17 ++++++ 4 files changed, 111 insertions(+) create mode 100644 hw/acpi/acpi_generic_initiator.c create mode 100644 include/hw/acpi/acpi_generic_initiator.h
hw/acpi/acpi_generic_initiator.c new file +71 -0
@@ -0,0 +1,71 @@ +// SPDX-License-Identifier: GPL-2.0-only +/* +* Copyright (c) 2024, NVIDIA CORPORATION & AFFILIATES. All rights reserved +*/ + +#include "qemu/osdep.h" +#include "hw/acpi/acpi_generic_initiator.h" +#include "hw/boards.h" +#include "qemu/error-report.h" + +typedef struct AcpiGenericInitiatorClass { +ObjectClass parent_class; +}AcpiGenericInitiatorClass ; + +OBJECT_DEFINE_TYPE_WITH_INTERFACES (AcpiGenericInitiator, acpi_generic_initiator, +ACPI_GENERIC_INITIATOR ,OBJECT , + { TYPE_USER_CREATABLE }, + { NULL }) + +OBJECT_DECLARE_SIMPLE_TYPE(AcpiGenericInitiator, ACPI_GENERIC_INITIATOR) + +static void acpi_generic_initiator_init (Object *obj) +{ +AcpiGenericInitiator *gi =ACPI_GENERIC_INITIATOR (obj); + + gi->node =MAX_NODES ; + gi->pci_dev =NULL ; +} + +static void acpi_generic_initiator_finalize (Object *obj) +{ +AcpiGenericInitiator *gi =ACPI_GENERIC_INITIATOR (obj); + +g_free (gi->pci_dev); +} + +static void acpi_generic_initiator_set_pci_device (Object *obj,const char *val, +Error **errp) +{ +AcpiGenericInitiator *gi =ACPI_GENERIC_INITIATOR (obj); + + gi->pci_dev =g_strdup (val); +} + +static void acpi_generic_initiator_set_node (Object *obj,Visitor *v, +const char *name,void *opaque, +Error **errp) +{ +AcpiGenericInitiator *gi =ACPI_GENERIC_INITIATOR (obj); +uint32_t value; + +if (!visit_type_uint32 (v, name, &value, errp)) { +return ; + } + +if (value >=MAX_NODES ) { +error_printf ("%s: Invalid NUMA node specified\n" , +TYPE_ACPI_GENERIC_INITIATOR ); +exit (1 ); + } + + gi->node = value; +} + +static void acpi_generic_initiator_class_init (ObjectClass *oc,void *data) +{ +object_class_property_add_str (oc,"pci-dev" ,NULL , + acpi_generic_initiator_set_pci_device); +object_class_property_add (oc,"node" ,"int" ,NULL , + acpi_generic_initiator_set_node,NULL ,NULL ); +}
hw/acpi/meson.build +1 -0
@@ -1,5 +1,6 @@ acpi_ss = ss.source_set() acpi_ss.add(files( + 'acpi_generic_initiator.c', 'acpi_interface.c', 'aml-build.c', 'bios-linker-loader.c',
include/hw/acpi/acpi_generic_initiator.h new file +22 -0
@@ -0,0 +1,22 @@ +// SPDX-License-Identifier: GPL-2.0-only +/* +* Copyright (c) 2024, NVIDIA CORPORATION & AFFILIATES. All rights reserved +*/ + +#ifndef ACPI_GENERIC_INITIATOR_H +#define ACPI_GENERIC_INITIATOR_H + +#include "qom/object_interfaces.h" + +#define TYPE_ACPI_GENERIC_INITIATOR "acpi-generic-initiator" + +typedef struct AcpiGenericInitiator { +/* private */ +Object parent; + +/* public */ +char *pci_dev; +uint16_t node; +}AcpiGenericInitiator ; + +#endif
qapi/qom.json +17 -0
@@ -794,6 +794,21 @@ { 'struct': 'VfioUserServerProperties', 'data': { 'socket': 'SocketAddress', 'device': 'str' } } +## +# @AcpiGenericInitiatorProperties: +# +# Properties for acpi-generic-initiator objects. +# +# @pci-dev: PCI device ID to be associated with the node +# +# @node: NUMA node associated with the PCI device +# +# Since: 9.0 +## +{ 'struct': 'AcpiGenericInitiatorProperties', + 'data': { 'pci-dev': 'str', + 'node': 'uint32' } } + ## # @RngProperties: #
@@ -911,6 +926,7 @@ ## { 'enum': 'ObjectType', 'data': [ + 'acpi-generic-initiator', 'authz-list', 'authz-listfile', 'authz-pam',
@@ -981,6 +997,7 @@ 'id': 'str' }, 'discriminator': 'qom-type', 'data': { + 'acpi-generic-initiator': 'AcpiGenericInitiatorProperties', 'authz-list': 'AuthZListProperties', 'authz-listfile': 'AuthZListFileProperties', 'authz-pam': 'AuthZPAMProperties',
Signature · 1 line
2.34.1
PATCH v9 2/3 hw/acpi: Implement the SRAT GI affinity structure
In reply to [email protected]From: Ankit Agrawal <[email protected]>
ACPI spec provides a scheme to associate "Generic Initiators" [1] (e.g. heterogeneous processors and accelerators, GPUs, and I/O devices with integrated compute or DMA engines GPUs) with Proximity Domains. This is achieved using Generic Initiator Affinity Structure in SRAT. During bootup, Linux kernel parse the ACPI SRAT to determine the PXM ids and create a NUMA node for each unique PXM ID encountered. Qemu currently do not implement these structures while building SRAT.
Add GI structures while building VM ACPI SRAT. The association between device and node are stored using acpi-generic-initiator object. Lookup presence of all such objects and use them to build these structures.
The structure needs a PCI device handle [2] that consists of the device BDF. The vfio-pci device corresponding to the acpi-generic-initiator object is located to determine the BDF.
[1] ACPI Spec 6.3, Section 5.2.16.6
[2] ACPI Spec 6.3, Table 5.80
--- hw/acpi/acpi_generic_initiator.c | 77 ++++++++++++++++++++++++ hw/acpi/hmat.c | 2 +- hw/arm/virt-acpi-build.c | 3 + hw/core/numa.c | 3 +- include/hw/acpi/acpi_generic_initiator.h | 25 ++++++++ include/sysemu/numa.h | 1 + 6 files changed, 109 insertions(+), 2 deletions(-)
hw/acpi/acpi_generic_initiator.c +77 -0
@@ -5,7 +5,9 @@#include "qemu/osdep.h" #include "hw/acpi/acpi_generic_initiator.h" +#include "hw/acpi/aml-build.h" #include "hw/boards.h" +#include "hw/pci/pci_device.h" #include "qemu/error-report.h" typedef struct AcpiGenericInitiatorClass {
@@ -47,6 +49,7 @@ static void acpi_generic_initiator_set_node(Object *obj, Visitor *v, Error **errp) {AcpiGenericInitiator *gi =ACPI_GENERIC_INITIATOR (obj); +MachineState *ms =MACHINE (qdev_get_machine ());uint32_t value;if (!visit_type_uint32 (v, name, &value, errp)) {
@@ -60,6 +63,7 @@ static void acpi_generic_initiator_set_node(Object *obj, Visitor *v, } gi->node = value; + ms->numa_state->nodes[gi->node].has_gi = true; }static void acpi_generic_initiator_class_init (ObjectClass *oc,void *data)
@@ -69,3 +73,76 @@ static void acpi_generic_initiator_class_init(ObjectClass *oc, void *data)object_class_property_add (oc,"node" ,"int" ,NULL , acpi_generic_initiator_set_node,NULL ,NULL ); } + +/* +* ACPI 6.3: +* Table 5-78 Generic Initiator Affinity Structure +*/ +static void +build_srat_generic_pci_initiator_affinity (GArray *table_data,int node, +PCIDeviceHandle *handle) +{ +uint8_t index; + +build_append_int_noprefix (table_data,5 ,1 );/* Type */ +build_append_int_noprefix (table_data,32 ,1 );/* Length */ +build_append_int_noprefix (table_data,0 ,1 );/* Reserved */ +build_append_int_noprefix (table_data,1 ,1 );/* Device Handle Type: PCI */ +build_append_int_noprefix (table_data, node,4 );/* Proximity Domain */ + +/* Device Handle - PCI */ +build_append_int_noprefix (table_data, handle->segment,2 ); +build_append_int_noprefix (table_data, handle->bdf,2 ); +for (index =0 ; index <12 ; index++) { +build_append_int_noprefix (table_data,0 ,1 ); + } + +build_append_int_noprefix (table_data,GEN_AFFINITY_ENABLED ,4 );/* Flags */ +build_append_int_noprefix (table_data,0 ,4 );/* Reserved */ +} + +static int build_all_acpi_generic_initiators (Object *obj,void *opaque) +{ +MachineState *ms =MACHINE (qdev_get_machine ()); +AcpiGenericInitiator *gi; +GArray *table_data = opaque; +PCIDeviceHandle dev_handle; +PCIDevice *pci_dev; +Object *o; + +if (!object_dynamic_cast (obj,TYPE_ACPI_GENERIC_INITIATOR )) { +return 0 ; + } + + gi =ACPI_GENERIC_INITIATOR (obj); +if (gi->node >= ms->numa_state->num_nodes) { +error_printf ("%s: Specified node %d is invalid.\n" , +TYPE_ACPI_GENERIC_INITIATOR , gi->node); +exit (1 ); + } + + o =object_resolve_path_type (gi->pci_dev,TYPE_PCI_DEVICE ,NULL ); +if (!o) { +error_printf ("%s: Specified device must be a PCI device.\n" , +TYPE_ACPI_GENERIC_INITIATOR ); +exit (1 ); + } + + pci_dev =PCI_DEVICE (o); + + dev_handle.segment =0 ; + dev_handle.bdf =PCI_BUILD_BDF (pci_bus_num (pci_get_bus (pci_dev)), + pci_dev->devfn); + +build_srat_generic_pci_initiator_affinity (table_data, + gi->node, &dev_handle); + +return 0 ; +} + +void build_srat_generic_pci_initiator (GArray *table_data) +{ +object_child_foreach_recursive (object_get_root (), + build_all_acpi_generic_initiators, + table_data); +}
hw/acpi/hmat.c +1 -1
@@ -214,7 +214,7 @@ static void hmat_build_table_structs(GArray *table_data, NumaState *numa_state) }for (i =0 ; i < numa_state->num_nodes; i++) { -if (numa_state->nodes[i].has_cpu) { +if (numa_state->nodes[i].has_cpu || numa_state->nodes[i].has_gi) { initiator_list[num_initiator++] = i; } }
hw/arm/virt-acpi-build.c +3 -0
@@ -58,6 +58,7 @@#include "migration/vmstate.h" #include "hw/acpi/ghes.h" #include "hw/acpi/viot.h" +#include "hw/acpi/acpi_generic_initiator.h" #define ARM_SPI_BASE 32
@@ -558,6 +559,8 @@ build_srat(GArray *table_data, BIOSLinker *linker, VirtMachineState *vms) } } +build_srat_generic_pci_initiator (table_data); +if (ms->nvdimms_state->is_enabled) {nvdimm_build_srat (table_data); }
hw/core/numa.c +2 -1
@@ -229,7 +229,8 @@ void parse_numa_hmat_lb(NumaState *numa_state, NumaHmatLBOptions *node, node->target, numa_state->num_nodes);return ; } -if (!numa_info[node->initiator].has_cpu) { +if (!numa_info[node->initiator].has_cpu && + !numa_info[node->initiator].has_gi) {error_setg (errp,"Invalid initiator=%d, it isn't an " "initiator proximity domain" , node->initiator);return ;
include/hw/acpi/acpi_generic_initiator.h +25 -0
@@ -19,4 +19,29 @@ typedef struct AcpiGenericInitiator { uint16_t node; } AcpiGenericInitiator; +/* +* ACPI 6.3: +* Table 5-81 Flags – Generic Initiator Affinity Structure +*/ +typedef enum { +/* +* If clear, the OSPM ignores the contents of the Generic +* Initiator/Port Affinity Structure. This allows system firmware +* to populate the SRAT with a static number of structures, but only +* enable them as necessary. +*/ +GEN_AFFINITY_ENABLED = (1 <<0 ), +}GenericAffinityFlags ; + +/* +* ACPI 6.3: +* Table 5-80 Device Handle - PCI +*/ +typedef struct PCIDeviceHandle { +uint16_t segment; +uint16_t bdf; +}PCIDeviceHandle ; + +void build_srat_generic_pci_initiator (GArray *table_data); +#endif
include/sysemu/numa.h +1 -0
@@ -41,6 +41,7 @@ struct NodeInfo {struct HostMemoryBackend *node_memdev;bool present;bool has_cpu; +bool has_gi;uint8_t lb_info_provided;uint16_t initiator;uint8_t distance[MAX_NODES ];
Signature · 1 line
2.34.1
PATCH v9 3/3 hw/i386/acpi-build: Add support for SRAT Generic Initiator structures
In reply to [email protected]From: Ankit Agrawal <[email protected]>
The acpi-generic-initiator object is added to allow a host device to be linked with a NUMA node. Qemu use it to build the SRAT Generic Initiator Affinity structure [1]. Add support for i386.
[1] ACPI Spec 6.3, Section 5.2.16.6
--- hw/i386/acpi-build.c | 3 +++ 1 file changed, 3 insertions(+)
hw/i386/acpi-build.c +3 -0
@@ -68,6 +68,7 @@#include "hw/acpi/utils.h" #include "hw/acpi/pci.h" #include "hw/acpi/cxl.h" +#include "hw/acpi/acpi_generic_initiator.h" #include "qom/qom-qobject.h" #include "hw/i386/amd_iommu.h"
@@ -2056,6 +2057,8 @@ build_srat(GArray *table_data, BIOSLinker *linker, MachineState *machine)build_srat_memory (table_data,0 ,0 ,0 ,MEM_AFFINITY_NOFLAGS ); } +build_srat_generic_pci_initiator (table_data); +/* * Entry is required for Windows to enable memory hotplug in OS * and for Linux to enable SWIOTLB when booted with less than
Signature · 1 line
2.34.1
Re: [PATCH v9 3/3] hw/i386/acpi-build: Add support for SRAT Generic Initiator structures
Quoted 9 lines · On Fri, 8 Mar 2024 14:55:25 +0000 <[email protected]> wrote:
From: Ankit Agrawal <[email protected]>
The acpi-generic-initiator object is added to allow a host device to be linked with a NUMA node. Qemu use it to build the SRAT Generic Initiator Affinity structure [1]. Add support for i386.
[1] ACPI Spec 6.3, Section 5.2.16.6
Suggested-by Jonathan Cameron <[email protected]>
Quoted 26 lines
Signed-off-by Ankit Agrawal <[email protected]>--- hw/i386/acpi-build.c | 3 +++ 1 file changed, 3 insertions(+)hw/i386/acpi-build.c +3
-0@@ -68,6 +68,7 @@#include "hw/acpi/utils.h" #include "hw/acpi/pci.h" #include "hw/acpi/cxl.h" +#include "hw/acpi/acpi_generic_initiator.h" #include "qom/qom-qobject.h" #include "hw/i386/amd_iommu.h" @@ -2056,6 +2057,8 @@ build_srat(GArray *table_data, BIOSLinker *linker, MachineState *machine)build_srat_memory (table_data,0 ,0 ,0 ,MEM_AFFINITY_NOFLAGS ); } +build_srat_generic_pci_initiator (table_data); +/* * Entry is required for Windows to enable memory hotplug in OS * and for Linux to enable SWIOTLB when booted with less than
Quoted 81 lines · On 3/8/24 15:55, [email protected] wrote:
From: Ankit Agrawal <[email protected]>
There are upcoming devices which allow CPU to cache coherently access their memory. It is sensible to expose such memory as NUMA nodes separate from the sysmem node to the OS. The ACPI spec provides a scheme in SRAT called Generic Initiator Affinity Structure [1] to allow an association between a Proximity Domain (PXM) and a Generic Initiator (GI) (e.g. heterogeneous processors and accelerators, GPUs, and I/O devices with integrated compute or DMA engines).
While a single node per device may cover several use cases, it is however insufficient for a full utilization of the NVIDIA GPUs MIG (Mult-Instance GPUs) [2] feature. The feature allows partitioning of the GPU device resources (including device memory) into several (upto 8) isolated instances. Each of the partitioned memory requires a dedicated NUMA node to operate. The partitions are not fixed and they can be created/deleted at runtime.
Linux OS does not provide a means to dynamically create/destroy NUMA nodes and such feature implementation is expected to be non-trivial. The nodes that OS discovers at the boot time while parsing SRAT remains fixed. So we utilize the GI Affinity structures that allows association between nodes and devices. Multiple GI structures per device/BDF is possible, allowing creation of multiple nodes in the VM by exposing unique PXM in each of these structures.
Implement the mechanism to build the GI affinity structures as Qemu currently does not. Introduce a new acpi-generic-initiator object to allow host admin link a device with an associated NUMA node. Qemu maintains this association and use this object to build the requisite GI Affinity Structure.
When multiple NUMA nodes are associated with a device, it is required to create those many number of acpi-generic-initiator objects, each representing a unique device:node association.
Following is one of a decoded GI affinity structure in VM ACPI SRAT.
[0C8h 0200 1] Subtable Type : 05 [Generic Initiator Affinity]
[0C9h 0201 1] Length : 20[0CAh 0202 1] Reserved1 : 00 [0CBh 0203 1] Device Handle Type : 01 [0CCh 0204 4] Proximity Domain : 00000007[0D0h 0208 16] Device Handle : 00 00 20 00 00 00 00 00 00 00 00
00 00 00 00 00
[0E0h 0224 4] Flags (decoded below) : 00000001
Enabled : 1
[0E4h 0228 4] Reserved2 : 00000000[0E8h 0232 1] Subtable Type : 05 [Generic Initiator Affinity]
[0E9h 0233 1] Length : 20On Grace Hopper systems, an admin will create a range of 8 nodes and associate them with the device using the acpi-generic-initiator object. While a configuration of less than 8 nodes per device is allowed, such configuration will prevent utilization of the feature to the fullest. This setting is applicable to all the Grace+Hopper systems. The following is an example of the Qemu command line arguments to create 8 nodes and link them to the device 'dev0':
(quoted diff) +0
-12-numa node,nodeid=2 -numa node,nodeid=3 -numa node,nodeid=4 \ -numa node,nodeid=5 -numa node,nodeid=6 -numa node,nodeid=7 \ -numa node,nodeid=8 -numa node,nodeid=9 \ -device vfio-pci-nohotplug,host=0009:01:00.0,bus=pcie.0,addr=04.0,rombar=0,id=dev0 \ -object acpi-generic-initiator,id=gi0,pci-dev=dev0,node=2 \ -object acpi-generic-initiator,id=gi1,pci-dev=dev0,node=3 \ -object acpi-generic-initiator,id=gi2,pci-dev=dev0,node=4 \ -object acpi-generic-initiator,id=gi3,pci-dev=dev0,node=5 \ -object acpi-generic-initiator,id=gi4,pci-dev=dev0,node=6 \ -object acpi-generic-initiator,id=gi5,pci-dev=dev0,node=7 \ -object acpi-generic-initiator,id=gi6,pci-dev=dev0,node=8 \ -object acpi-generic-initiator,id=gi7,pci-dev=dev0,node=9 \The performance benefits can be realized by providing the NUMA node distances appropriately (through libvirt tags or Qemu params). The admin can get the distance among nodes in hardware using
numactl -H.This series goes along with the recenty added vfio-pci variant driver [3].
Applied over v8.2.2
base commit:11aa0b1ff115b86160c4d37e7c37e6a6b13b77ea
Quoted 6 lines
[1] ACPI Spec 6.3, Section 5.2.16.6
Link for v8:
v9 looks ready for QEMU 9.0. An Ack from the ACPI supporters is missing though.
Michal, Igor, Ani,
Did you have time to take a look ?
Thanks
C.
Quoted 74 lines
v8 -> v9
- Removed unused included headers based on Jonathan's suggestion.
- Collected Reviewed-by from Jonathan.
- Added acpi-generic-initiator support for i386
- Moved HMAT change from patch 1/2 to 2/3.
- Fixed nits.v7 -> v8
- Replaced the code to collect the acpi-generic-initiator objects with the code to use recursive helperobject_child_foreach_recursivebased on suggestion from Jonathan Cameron.
- Added sanity check for the node id passed to the acpi-generic-initiator object.
- Added change to use GI as HMAT initiator as per Jonathan's suggestion.
- Fixed nits pointed by Marcus and Jonathan.
- Collected Marcus' Acked-by.
- Rebased to v8.2.2.v6 -> v7
- Updated code and the commit message to make acpi-generic-initiator define a 1:1 relationship between device and node based on Jonathan Cameron's suggestion.
- Updated commit message to include the decoded GI entry in the SRAT.
- Rebased to v8.2.1.v5 -> v6
- Updated commit message for the [1/2] and the cover letter.
- Updated the acpi-generic-initiator object comment description for clarity on the input host-nodes.
- Rebased to v8.2.0-rc4.v4 -> v5
- Removed acpi-dev option until full support.
- The NUMA nodes are saved as bitmap instead of uint16List.
- Replaced asserts to exit calls.
- Addressed other miscellaneous comments.v3 -> v4
- changed the ':' delimited way to a uint16 array to communicate the nodes associated with the device.
- added asserts to handle invalid inputs.
- addressed other miscellaneous v3 comments.v2 -> v3
- changed param to accept a ':' delimited list of NUMA nodes, instead of a range.
- Removed nvidia-acpi-generic-initiator object.
- Addressed miscellaneous comments in v2.v1 -> v2
- Removed dependency on sysfs to communicate the feature with variant module.
- Use GI Affinity SRAT structure instead of Memory Affinity.
- No DSDT entries needed to communicate the PXM for the device. SRAT GI structure is used instead.
- New objects introduced to establish link between device and nodes.Ankit Agrawal (3):
qom: new object to associate device to NUMA node
hw/acpi: Implement the SRAT GI affinity structure hw/i386/acpi-build: Add support for SRAT Generic Initiator structureshw/acpi/acpi_generic_initiator.c | 148 +++++++++++++++++++++++ hw/acpi/hmat.c | 2 +- hw/acpi/meson.build | 1 + hw/arm/virt-acpi-build.c | 3 + hw/core/numa.c | 3 +- hw/i386/acpi-build.c | 3 + include/hw/acpi/acpi_generic_initiator.h | 47 +++++++ include/sysemu/numa.h | 1 + qapi/qom.json | 17 +++ 9 files changed, 223 insertions(+), 2 deletions(-)create mode 100644
hw/acpi/acpi_generic_initiator.ccreate mode 100644include/hw/acpi/acpi_generic_initiator.h
Quoted 92 lines · Cédric Le Goater
Quoted 81 lines · On 3/8/24 15:55, [email protected] wrote:
From: Ankit Agrawal <[email protected]>
There are upcoming devices which allow CPU to cache coherently access their memory. It is sensible to expose such memory as NUMA nodes separate from the sysmem node to the OS. The ACPI spec provides a scheme in SRAT called Generic Initiator Affinity Structure [1] to allow an association between a Proximity Domain (PXM) and a Generic Initiator (GI) (e.g. heterogeneous processors and accelerators, GPUs, and I/O devices with integrated compute or DMA engines).
While a single node per device may cover several use cases, it is however insufficient for a full utilization of the NVIDIA GPUs MIG (Mult-Instance GPUs) [2] feature. The feature allows partitioning of the GPU device resources (including device memory) into several (upto 8) isolated instances. Each of the partitioned memory requires a dedicated NUMA node to operate. The partitions are not fixed and they can be created/deleted at runtime.
Linux OS does not provide a means to dynamically create/destroy NUMA nodes and such feature implementation is expected to be non-trivial. The nodes that OS discovers at the boot time while parsing SRAT remains fixed. So we utilize the GI Affinity structures that allows association between nodes and devices. Multiple GI structures per device/BDF is possible, allowing creation of multiple nodes in the VM by exposing unique PXM in each of these structures.
Implement the mechanism to build the GI affinity structures as Qemu currently does not. Introduce a new acpi-generic-initiator object to allow host admin link a device with an associated NUMA node. Qemu maintains this association and use this object to build the requisite GI Affinity Structure.
When multiple NUMA nodes are associated with a device, it is required to create those many number of acpi-generic-initiator objects, each representing a unique device:node association.
Following is one of a decoded GI affinity structure in VM ACPI SRAT.
[0C8h 0200 1] Subtable Type : 05 [Generic Initiator Affinity]
[0C9h 0201 1] Length : 20[0CAh 0202 1] Reserved1 : 00 [0CBh 0203 1] Device Handle Type : 01 [0CCh 0204 4] Proximity Domain : 00000007[0D0h 0208 16] Device Handle : 00 00 20 00 00 00 00 00 00 00 00
00 00 00 00 00
[0E0h 0224 4] Flags (decoded below) : 00000001
Enabled : 1
[0E4h 0228 4] Reserved2 : 00000000[0E8h 0232 1] Subtable Type : 05 [Generic Initiator Affinity]
[0E9h 0233 1] Length : 20On Grace Hopper systems, an admin will create a range of 8 nodes and associate them with the device using the acpi-generic-initiator object. While a configuration of less than 8 nodes per device is allowed, such configuration will prevent utilization of the feature to the fullest. This setting is applicable to all the Grace+Hopper systems. The following is an example of the Qemu command line arguments to create 8 nodes and link them to the device 'dev0':
(quoted diff) +0
-12-numa node,nodeid=2 -numa node,nodeid=3 -numa node,nodeid=4 \ -numa node,nodeid=5 -numa node,nodeid=6 -numa node,nodeid=7 \ -numa node,nodeid=8 -numa node,nodeid=9 \ -device vfio-pci-nohotplug,host=0009:01:00.0,bus=pcie.0,addr=04.0,rombar=0,id=dev0 \ -object acpi-generic-initiator,id=gi0,pci-dev=dev0,node=2 \ -object acpi-generic-initiator,id=gi1,pci-dev=dev0,node=3 \ -object acpi-generic-initiator,id=gi2,pci-dev=dev0,node=4 \ -object acpi-generic-initiator,id=gi3,pci-dev=dev0,node=5 \ -object acpi-generic-initiator,id=gi4,pci-dev=dev0,node=6 \ -object acpi-generic-initiator,id=gi5,pci-dev=dev0,node=7 \ -object acpi-generic-initiator,id=gi6,pci-dev=dev0,node=8 \ -object acpi-generic-initiator,id=gi7,pci-dev=dev0,node=9 \The performance benefits can be realized by providing the NUMA node distances appropriately (through libvirt tags or Qemu params). The admin can get the distance among nodes in hardware using
numactl -H.This series goes along with the recenty added vfio-pci variant driver [3].
Applied over v8.2.2
base commit:11aa0b1ff115b86160c4d37e7c37e6a6b13b77eaQuoted 6 lines
[1] ACPI Spec 6.3, Section 5.2.16.6
Link for v8:
v9 looks ready for QEMU 9.0. An Ack from the ACPI supporters is missing though.
Quoted 7 lines
Michal, Igor, Ani,
Did you have time to take a look ?
Thanks
C.
I tagged it already.
Quoted 76 lines
Quoted 76 lines
v8 -> v9
- Removed unused included headers based on Jonathan's suggestion.
- Collected Reviewed-by from Jonathan.
- Added acpi-generic-initiator support for i386
- Moved HMAT change from patch 1/2 to 2/3.
- Fixed nits.v7 -> v8
- Replaced the code to collect the acpi-generic-initiator objects with the code to use recursive helperobject_child_foreach_recursivebased on suggestion from Jonathan Cameron.
- Added sanity check for the node id passed to the acpi-generic-initiator object.
- Added change to use GI as HMAT initiator as per Jonathan's suggestion.
- Fixed nits pointed by Marcus and Jonathan.
- Collected Marcus' Acked-by.
- Rebased to v8.2.2.v6 -> v7
- Updated code and the commit message to make acpi-generic-initiator define a 1:1 relationship between device and node based on Jonathan Cameron's suggestion.
- Updated commit message to include the decoded GI entry in the SRAT.
- Rebased to v8.2.1.v5 -> v6
- Updated commit message for the [1/2] and the cover letter.
- Updated the acpi-generic-initiator object comment description for clarity on the input host-nodes.
- Rebased to v8.2.0-rc4.v4 -> v5
- Removed acpi-dev option until full support.
- The NUMA nodes are saved as bitmap instead of uint16List.
- Replaced asserts to exit calls.
- Addressed other miscellaneous comments.v3 -> v4
- changed the ':' delimited way to a uint16 array to communicate the nodes associated with the device.
- added asserts to handle invalid inputs.
- addressed other miscellaneous v3 comments.v2 -> v3
- changed param to accept a ':' delimited list of NUMA nodes, instead of a range.
- Removed nvidia-acpi-generic-initiator object.
- Addressed miscellaneous comments in v2.v1 -> v2
- Removed dependency on sysfs to communicate the feature with variant module.
- Use GI Affinity SRAT structure instead of Memory Affinity.
- No DSDT entries needed to communicate the PXM for the device. SRAT GI structure is used instead.
- New objects introduced to establish link between device and nodes.Ankit Agrawal (3):
qom: new object to associate device to NUMA node
hw/acpi: Implement the SRAT GI affinity structure hw/i386/acpi-build: Add support for SRAT Generic Initiator structureshw/acpi/acpi_generic_initiator.c | 148 +++++++++++++++++++++++ hw/acpi/hmat.c | 2 +- hw/acpi/meson.build | 1 + hw/arm/virt-acpi-build.c | 3 + hw/core/numa.c | 3 +- hw/i386/acpi-build.c | 3 + include/hw/acpi/acpi_generic_initiator.h | 47 +++++++ include/sysemu/numa.h | 1 + qapi/qom.json | 17 +++ 9 files changed, 223 insertions(+), 2 deletions(-)create mode 100644
hw/acpi/acpi_generic_initiator.ccreate mode 100644include/hw/acpi/acpi_generic_initiator.h