550.127.05

kernel-open/Kbuild

@@ -72,7 +72,7 @@ EXTRA_CFLAGS += -I$(src)/common/inc
 EXTRA_CFLAGS += -I$(src)
 EXTRA_CFLAGS += -Wall $(DEFINES) $(INCLUDES) -Wno-cast-qual -Wno-format-extra-args
 EXTRA_CFLAGS += -D__KERNEL__ -DMODULE -DNVRM
-EXTRA_CFLAGS += -DNV_VERSION_STRING=\"550.120\"
+EXTRA_CFLAGS += -DNV_VERSION_STRING=\"550.127.05\"
 
 ifneq ($(SYSSRCHOST1X),)
  EXTRA_CFLAGS += -I$(SYSSRCHOST1X)

kernel-open/nvidia-uvm/uvm.h

@@ -379,6 +379,17 @@ NV_STATUS UvmIsPageableMemoryAccessSupportedOnGpu(const NvProcessorUuid *gpuUuid
 //         OS state required to register the GPU is malformed, or the partition
 //         identified by the user handles or its configuration changed.
 //
+//     NV_ERR_NVSWITCH_FABRIC_NOT_READY:
+//         (On NvSwitch-connected system) Indicates that the fabric has not been
+//         configured yet. Caller must retry GPU registration.
+//
+//     NV_ERR_NVSWITCH_FABRIC_FAILURE:
+//         (On NvSwitch-connected systems) Indicates that the NvLink fabric
+//         failed to be configured.
+//
+//     NV_ERR_GPU_MEMORY_ONLINING_FAULURE:
+//         (On coherent systems) The GPU's memory onlining failed.
+//
 //     NV_ERR_GENERIC:
 //         Unexpected error. We try hard to avoid returning this error code,
 //         because it is not very informative.

kernel-open/nvidia-uvm/uvm_gpu.c

@@ -138,6 +138,7 @@ static NV_STATUS get_gpu_caps(uvm_gpu_t *gpu)
 
     if (gpu_caps.numaEnabled) {
         UVM_ASSERT(uvm_parent_gpu_is_coherent(gpu->parent));
+
         gpu->mem_info.numa.enabled = true;
         gpu->mem_info.numa.node_id = gpu_caps.numaNodeId;
     }
@@ -1280,7 +1281,8 @@ static NV_STATUS init_gpu(uvm_gpu_t *gpu, const UvmGpuInfo *gpu_info)
 
     status = get_gpu_caps(gpu);
     if (status != NV_OK) {
-        UVM_ERR_PRINT("Failed to get GPU caps: %s, GPU %s\n", nvstatusToString(status), uvm_gpu_name(gpu));
+        if (status != NV_ERR_NVSWITCH_FABRIC_NOT_READY)
+            UVM_ERR_PRINT("Failed to get GPU caps: %s, GPU %s\n", nvstatusToString(status), uvm_gpu_name(gpu));
         return status;
     }
 

kernel-open/nvidia/nv-pci.c

@@ -40,6 +40,9 @@
 #if !defined(NV_BUS_TYPE_HAS_IOMMU_OPS)
 #include <linux/iommu.h>
 #endif
+#if NV_IS_EXPORT_SYMBOL_GPL_pci_ats_supported
+#include <linux/pci-ats.h>
+#endif
 
 static void
 nv_check_and_exclude_gpu(
@@ -781,10 +784,15 @@ next_bar:
     // PPC64LE platform where ATS is currently supported (IBM P9).
     nv_ats_supported &= nv_platform_supports_numa(nvl);
 #else
-#if defined(NV_PCI_DEV_HAS_ATS_ENABLED)
+#if NV_IS_EXPORT_SYMBOL_GPL_pci_ats_supported
+    nv_ats_supported &= pci_ats_supported(pci_dev);
+#elif defined(NV_PCI_DEV_HAS_ATS_ENABLED)
     nv_ats_supported &= pci_dev->ats_enabled;
+#else
+    nv_ats_supported = NV_FALSE;
 #endif
 #endif
+
     if (nv_ats_supported)
     {
         NV_DEV_PRINTF(NV_DBG_INFO, nv, "ATS supported by this GPU!\n");

kernel-open/nvidia/nv.c

@@ -188,11 +188,7 @@ struct semaphore nv_linux_devices_lock;
 
 // True if all the successfully probed devices support ATS
 // Assigned at device probe (module init) time
-NvBool nv_ats_supported = NVCPU_IS_PPC64LE
-#if defined(NV_PCI_DEV_HAS_ATS_ENABLED)
-                          || NV_TRUE
-#endif
-;
+NvBool nv_ats_supported = NV_TRUE;
 
 // allow an easy way to convert all debug printfs related to events
 // back and forth between 'info' and 'errors'

kernel-open/nvidia/nvidia.Kbuild

@@ -231,6 +231,7 @@ NV_CONFTEST_SYMBOL_COMPILE_TESTS += is_export_symbol_present_tsec_comms_free_gsc
 NV_CONFTEST_SYMBOL_COMPILE_TESTS += is_export_symbol_present_memory_block_size_bytes
 NV_CONFTEST_SYMBOL_COMPILE_TESTS += crypto
 NV_CONFTEST_SYMBOL_COMPILE_TESTS += is_export_symbol_present_follow_pte
+NV_CONFTEST_SYMBOL_COMPILE_TESTS += is_export_symbol_gpl_pci_ats_supported
 
 NV_CONFTEST_TYPE_COMPILE_TESTS += dma_ops
 NV_CONFTEST_TYPE_COMPILE_TESTS += swiotlb_dma_ops