Two nitpicks below. Regards, Philip
On 2026-08-21 15:33, Alex Deucher wrote:
This adds the core infrastructure for supporting UALink (Ultra Accelerator Link) connected scale up pods. I've split this into two series, one to add the core infrastructure, and one to expose the new IOCTL interface and add the documentation to avoid spamming the larger audience with the implemenation defails. This is the first series. This implements UALoE (UALink over Ethernet). An overview of the complete solution can be found here: https://www.amd.com/en/products/rackscale-solutions/helios.html Overview Connected GPUs in a pod can directly access the remove memory on another
s/remove/remote/
GPU over UALink. Unlike RMDA, there is no copy involved; it is direct loads/stores over the fabric. Shared memory can only be accessed by a remote GPU if the memory was exported and the importer has been authorized. For the memory to be shared, it must be part of a unified physical address space shared between nodes. This address space is called NPA (Nework Physical Address) space. This address space is partitioned between the GPUs so that each GPU has it's own segment of the address space in which to export its memory. Each GPU maintains a dedicated set of page tables for their NPA space similar to GPUVM. Exported memory is not pinned. The exporter may need to move memory or make it unavailable to the importers. This is handled via remote TLB shootdowns on the exporter in the case of TTM evictions or MMU notifiers and remote presense checks with retry on the importer. To access remote memory, the driver can map NPA addresses into its per process GPUVM page tables just like local memory. Applications use opaque handles to represent remote memory. GPUs in a pod communicate with eachother directly to exchange NPA addresses between importers
s/eachother/each other/
and exporters. If a node goes offline or is reset, their peers will clean up any remaining refrences that are lost when that happens. User interface Export Memory To export memory, a handle must be created for an allocation that can be shared with another node in the pod. To do this the exporter calls the GEM UALink IOCTL with the GEM handle to the buffer it wants to export. The IOCTL returns a unique handle which can be shared with the remote host. Import Memory To import the memory, the handle from the remote node must be converted from a unique handle to a local GEM object which represents the local reference to the NPA space on the importer. If the memory has already been imported, it just returns a new reference to the existing object. If not, the importer queries the exporter to get the NPA address. Once it has that the importer can create a dma-buf to represent the NPA space used by the allocation and that is returned to the application. Alex Sierra (1): drm/amdgpu: log remote memory MTYPE for GC 12.1.0 Felix Kuehling (17): drm/amdgpu: Add sysfs API for UALink information drm/amdgpu: Add sysfs API for UALink physical pod setup drm/amdgpu: Add sysfs API for UALink virtual pod config drm/amdgpu: Add sysfs API for UALink station configuration drm/amdgpu: Implement PSP cmd UAL_GET_CONFIG drm/amdgpu: Query initial UALink config from PSP drm/amdgpu: Implement PSP cmd UAL_SET_PPOD_CONFIG drm/amdgpu: Set physical pod configuration to PSP drm/amdgpu: Implement PSP cmd UAL_SET_VPOD_CONFIG drm/amdgpu: Set virtual pod configuration to PSP drm/amdgpu: Implement PSP cmd UAL_SET_STATION_CONFIG drm/amdgpu: Set UALink station config to PSP drm/amdgpu: Implement PSP cmd UAL_SET_NPA_CONFIG drm/amdgpu: Enable/disable NPA address translation using PSP drm/amdgpu: Prevent double-free of drm_exec drm/amdgpu: Pin page tables in NPA VMs drm/amdgpu: Initialize NPA PT/PDs to noretry Gangliang Xie (1): drm/amdgpu: add handler for nHT error Harish Kasiviswanathan (2): drm/amdgpu: Add helper function to check psp xgmi ta drm/amdgpu: Add ual_config_state to ual_get_config Jinzhou Su (1): drm/amdgpu: Fix NULL pointer issue during ualink init Le Ma (1): drm/amdgpu: extend PSP command polling sleep range Lijo Lazar (15): drm/amdgpu: Add psp ualink command interfaces drm/amdgpu: Fetch asp ualink interface version drm/amdgpu: Separate out ualink init sequences drm/amdgpu: Add ualink as separate ip block drm/admgpu: Seggregate ualink nht messaging drm/amdgpu: Assign accel state based on ASP config drm/amdgpu: Drop duplicate vpod check functions drm/amdgpu: Add support to send ASP completion drm/amdgpu: Add handlers for ualink notifications drm/amdgpu: Improve ualink state transitions drm/amdgpu: Use uniform logic for inband/sideband drm/amdgpu: Add name for ualink ip block drm/amdgpu: Move ualink ip version related changes drm/amdgpu: Add hw_fini for ualink drm/amdgpu: Expose ualink info under each xcp Mukul Joshi (42): drm/amdgpu: Add UALink manager core infrastructure drm/amdgpu: Add a new NPA Address space drm/amdgpu: Add address allocator for NPA addresses drm/amdgpu: Initialize VM for NPA addr management drm/amdgpu: Rework VMID reservation logic drm/amdgpu: Reserve VMID for NPA VM drm/amdgpu: Use reserved VMID for NPA VM drm/amdgpu: Enable UALink Manager when pod becomes active drm/amdgpu: Create a drm client for UALink NPA BOs drm/amdgpu: Control NPA DMA-buf importing drm/amdgpu: Add ualink handle to BOs drm/amdgpu: Implement UALink handle export drm/amdgpu: Add connection state management drm/amdgpu: Implement UALink handle import ioctl drm/amdgpu: Implement mechanism to revoke exported memory drm/amdgpu: lock UALink import invalidation via drm_exec drm/amdgpu: Cleanup exported UALink handles drm/amdgpu: Cleanup imported UALink handles drm/amdgpu: Handle connection reset drm/amdgpu: Setup PTE mappings for NPA addresses drm/amdgpu: Add handling for remote interrupts drm/amdgpu: Send TLB shootdown on exported memory unmap drm/amdgpu: Handle local GPUs in UALink import drm/amdgpu: Add debugfs to drop UALink protocol messages drm/amdgpu: Temporarily disable sending remote TLB shootdowns drm/amdgpu: Temporarily Flush TLB on NPA mapping always drm/amdgpu: fix NPA-RELEASE race in UALink exporter cleanup drm/amdgpu: initialize UALink importer node list head drm/amdgpu: Fix initialization flags for UALink XAs drm/amdgpu: fix dma_buf leak in UALink exporter cleanup drm/amdgpu: Fix uninitialized fence in UALink NPA unmap drm/amdgpu: Use vm->last_update fence in UALink NPA unmap paths drm/amdgpu: always use MTYPE_UC for remote memory on GFX 12.1 drm/amdkfd: program compute MQD coherent_aql_mtype on GFX 12.1 drm/amdgpu: Fix GART and SDMA entity leak on vPod reconfiguration drm/amdgpu: Add UALink diagnostic logging for vpod commit/activation drm/amdgpu: Handle UALink vPod reconfiguration while ACTIVE drm/amdgpu: Cleanup UALink XA entries on manager stop drm/amdgpu: Handle concurrent UALINK handle import race drm/amdgpu: create UALink NPA import BO directly in the NPA domain drm/amdgpu: add mtype_remote module parameter drm/amdgpu: Honor mtype overrides for NPA remote memory Philip Yang (15): drm/amdgpu: Fix UALink vPod double-activation drm/amdgpu: Add UALink remote state structures and API declarations drm/amdgpu: Add UALink NPA address layout helpers drm/amdgpu: Add UALink NPA address computation for ring buffers drm/amdgpu: Add UALink NPA VM mapping for ring buffers drm/amdgpu: Add UALink SDMA scheduler entities drm/amdgpu: Add UALink GART helpers for NPA address access drm/amdgpu: Add UALink ring buffer allocation and firmware init drm/amdgpu: Add UALink remote command packets and SDMA dispatch drm/amdgpu: Add UALink firmware writeback address configuration drm/amdgpu: Add UALink cross-GPU TLB shootdown and remote interrupt drm/amdgpu: Add UALink software init, teardown, and reset drm/amdgpu: Add UALink IH ring and enable interrupt drm/amdgpu: UALink use LSDMA to send remote interrupt command drm/amdgpu: Increase UALink soft ring size drivers/gpu/drm/amd/amdgpu/Makefile | 3 + drivers/gpu/drm/amd/amdgpu/amdgpu.h | 5 + drivers/gpu/drm/amd/amdgpu/amdgpu_debugfs.c | 14 + drivers/gpu/drm/amd/amdgpu/amdgpu_device.c | 2 + drivers/gpu/drm/amd/amdgpu/amdgpu_discovery.c | 21 + drivers/gpu/drm/amd/amdgpu/amdgpu_dma_buf.c | 5 + drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c | 7 + drivers/gpu/drm/amd/amdgpu/amdgpu_gfx.h | 1 + drivers/gpu/drm/amd/amdgpu/amdgpu_ids.c | 33 +- drivers/gpu/drm/amd/amdgpu/amdgpu_ids.h | 27 +- drivers/gpu/drm/amd/amdgpu/amdgpu_ih.h | 1 + drivers/gpu/drm/amd/amdgpu/amdgpu_ip.c | 1 + drivers/gpu/drm/amd/amdgpu/amdgpu_irq.c | 17 + drivers/gpu/drm/amd/amdgpu/amdgpu_irq.h | 7 +- drivers/gpu/drm/amd/amdgpu/amdgpu_object.c | 11 + drivers/gpu/drm/amd/amdgpu/amdgpu_object.h | 5 + drivers/gpu/drm/amd/amdgpu/amdgpu_psp.c | 280 +- drivers/gpu/drm/amd/amdgpu/amdgpu_psp.h | 27 + .../gpu/drm/amd/amdgpu/amdgpu_res_cursor.h | 1 + drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.c | 17 +- drivers/gpu/drm/amd/amdgpu/amdgpu_ttm.h | 4 +- drivers/gpu/drm/amd/amdgpu/amdgpu_ualink.c | 6174 +++++++++++++++++ drivers/gpu/drm/amd/amdgpu/amdgpu_ualink.h | 421 ++ drivers/gpu/drm/amd/amdgpu/amdgpu_vm.c | 44 +- drivers/gpu/drm/amd/amdgpu/amdgpu_vm.h | 4 + drivers/gpu/drm/amd/amdgpu/amdgpu_vm_pt.c | 29 +- drivers/gpu/drm/amd/amdgpu/amdgpu_xcp.c | 6 + drivers/gpu/drm/amd/amdgpu/amdgpu_xcp.h | 7 + drivers/gpu/drm/amd/amdgpu/gfx_v10_0.c | 2 +- drivers/gpu/drm/amd/amdgpu/gfx_v11_0.c | 2 +- drivers/gpu/drm/amd/amdgpu/gfx_v12_0.c | 2 +- drivers/gpu/drm/amd/amdgpu/gfx_v12_1.c | 51 +- drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c | 2 +- drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c | 2 +- drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c | 2 +- drivers/gpu/drm/amd/amdgpu/gfx_v9_4_3.c | 2 +- drivers/gpu/drm/amd/amdgpu/gmc_v10_0.c | 4 + drivers/gpu/drm/amd/amdgpu/gmc_v11_0.c | 5 + drivers/gpu/drm/amd/amdgpu/gmc_v12_0.c | 26 +- drivers/gpu/drm/amd/amdgpu/gmc_v12_1.c | 98 +- drivers/gpu/drm/amd/amdgpu/gmc_v12_1.h | 1 + drivers/gpu/drm/amd/amdgpu/gmc_v6_0.c | 6 + drivers/gpu/drm/amd/amdgpu/gmc_v7_0.c | 6 + drivers/gpu/drm/amd/amdgpu/gmc_v8_0.c | 6 + drivers/gpu/drm/amd/amdgpu/gmc_v9_0.c | 6 + drivers/gpu/drm/amd/amdgpu/ih_v7_0.c | 8 + drivers/gpu/drm/amd/amdgpu/psp_gfx_if.h | 162 + drivers/gpu/drm/amd/amdgpu/psp_v15_0_8.c | 48 +- drivers/gpu/drm/amd/amdgpu/soc_v1_0.c | 6 - drivers/gpu/drm/amd/amdgpu/ualink_v1_0.c | 143 + drivers/gpu/drm/amd/amdgpu/ualink_v1_0.h | 30 + .../drm/amd/amdkfd/kfd_mqd_manager_v12_1.c | 7 + drivers/gpu/drm/amd/amdkfd/kfd_svm.c | 8 +- drivers/gpu/drm/amd/include/amd_shared.h | 2 + .../ivsrcid/mpnht/irqsrcs_mpnht_15_0.h | 30 + drivers/gpu/drm/amd/include/v12_structs.h | 2 +- include/drm/ttm/ttm_resource.h | 2 +- include/uapi/drm/amdgpu_drm.h | 6 + 58 files changed, 7782 insertions(+), 69 deletions(-) create mode 100644 drivers/gpu/drm/amd/amdgpu/amdgpu_ualink.c create mode 100644 drivers/gpu/drm/amd/amdgpu/amdgpu_ualink.h create mode 100644 drivers/gpu/drm/amd/amdgpu/ualink_v1_0.c create mode 100644 drivers/gpu/drm/amd/amdgpu/ualink_v1_0.h create mode 100644 drivers/gpu/drm/amd/include/ivsrcid/mpnht/irqsrcs_mpnht_15_0.h
