535.113.01

CHANGELOG.md

@@ -2,6 +2,12 @@
 
 ## Release 535 Entries
 
+### [535.113.01] 2023-09-21
+
+#### Fixed
+
+- Fixed building main against current centos stream 8 fails, [#550](https://github.com/NVIDIA/open-gpu-kernel-modules/issues/550) by @airlied
+
 ### [535.104.05] 2023-08-22
 
 ### [535.98] 2023-08-08

README.md

@@ -1,7 +1,7 @@
 # NVIDIA Linux Open GPU Kernel Module Source
 
 This is the source release of the NVIDIA Linux open GPU kernel modules,
-version 535.104.05.
+version 535.113.01.
 
 
 ## How to Build
@@ -17,7 +17,7 @@ as root:
 
 Note that the kernel modules built here must be used with GSP
 firmware and user-space NVIDIA GPU driver components from a corresponding
-535.104.05 driver release.  This can be achieved by installing
+535.113.01 driver release.  This can be achieved by installing
 the NVIDIA GPU driver from the .run file using the `--no-kernel-modules`
 option.  E.g.,
 
@@ -180,15 +180,15 @@ software applications.
 ## Compatible GPUs
 
 The open-gpu-kernel-modules can be used on any Turing or later GPU
-(see the table below). However, in the 535.104.05 release,
+(see the table below). However, in the 535.113.01 release,
 GeForce and Workstation support is still considered alpha-quality.
 
 To enable use of the open kernel modules on GeForce and Workstation GPUs,
 set the "NVreg_OpenRmEnableUnsupportedGpus" nvidia.ko kernel module
 parameter to 1. For more details, see the NVIDIA GPU driver end user
 README here:
 
-https://us.download.nvidia.com/XFree86/Linux-x86_64/535.104.05/README/kernel_open.html
+https://us.download.nvidia.com/XFree86/Linux-x86_64/535.113.01/README/kernel_open.html
 
 In the below table, if three IDs are listed, the first is the PCI Device 
 ID, the second is the PCI Subsystem Vendor ID, and the third is the PCI
@@ -856,6 +856,10 @@ Subsystem Device ID.
 | NVIDIA RTX 4000 SFF Ada Generation              | 27B0 103C 16FA |
 | NVIDIA RTX 4000 SFF Ada Generation              | 27B0 10DE 16FA |
 | NVIDIA RTX 4000 SFF Ada Generation              | 27B0 17AA 16FA |
+| NVIDIA RTX 4500 Ada Generation                  | 27B1 1028 180C |
+| NVIDIA RTX 4500 Ada Generation                  | 27B1 103C 180C |
+| NVIDIA RTX 4500 Ada Generation                  | 27B1 10DE 180C |
+| NVIDIA RTX 4500 Ada Generation                  | 27B1 17AA 180C |
 | NVIDIA RTX 4000 Ada Generation                  | 27B2 1028 181B |
 | NVIDIA RTX 4000 Ada Generation                  | 27B2 103C 181B |
 | NVIDIA RTX 4000 Ada Generation                  | 27B2 10DE 181B |

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-error -Wno-format-extra-args
 EXTRA_CFLAGS += -D__KERNEL__ -DMODULE -DNVRM
-EXTRA_CFLAGS += -DNV_VERSION_STRING=\"535.104.05\"
+EXTRA_CFLAGS += -DNV_VERSION_STRING=\"535.113.01\"
 
 ifneq ($(SYSSRCHOST1X),)
  EXTRA_CFLAGS += -I$(SYSSRCHOST1X)

kernel-open/common/inc/os-interface.h

@@ -207,9 +207,13 @@ enum os_pci_req_atomics_type {
     OS_INTF_PCIE_REQ_ATOMICS_128BIT
 };
 NV_STATUS   NV_API_CALL  os_enable_pci_req_atomics   (void *, enum os_pci_req_atomics_type);
+NV_STATUS   NV_API_CALL  os_get_numa_node_memory_usage (NvS32, NvU64 *, NvU64 *);
 NV_STATUS   NV_API_CALL  os_numa_add_gpu_memory      (void *, NvU64, NvU64, NvU32 *);
 NV_STATUS   NV_API_CALL  os_numa_remove_gpu_memory   (void *, NvU64, NvU64, NvU32); 
 NV_STATUS   NV_API_CALL  os_offline_page_at_address(NvU64 address);
+void*       NV_API_CALL  os_get_pid_info(void);
+void        NV_API_CALL  os_put_pid_info(void *pid_info);
+NV_STATUS   NV_API_CALL  os_find_ns_pid(void *pid_info, NvU32 *ns_pid);
 
 extern NvU32 os_page_size;
 extern NvU64 os_page_mask;

kernel-open/conftest.sh

@@ -5636,23 +5636,6 @@ compile_test() {
             compile_check_conftest "$CODE" "NV_GPIO_TO_IRQ_PRESENT" "" "functions"
         ;;
 
-        migrate_vma_setup)
-            #
-            # Determine if migrate_vma_setup() function is present
-            #
-            # migrate_vma_setup() function was added by commit
-            # a7d1f22bb74f32cf3cd93f52776007e161f1a738 ("mm: turn migrate_vma
-            # upside down) in v5.4.
-            # (2019-08-20).
-            CODE="
-            #include <linux/migrate.h>
-            int conftest_migrate_vma_setup(void) {
-                migrate_vma_setup();
-            }"
-
-            compile_check_conftest "$CODE" "NV_MIGRATE_VMA_SETUP_PRESENT" "" "functions"
-        ;;
-
         migrate_vma_added_flags)
             #
             # Determine if migrate_vma structure has flags
@@ -5743,23 +5726,25 @@ compile_test() {
             compile_check_conftest "$CODE" "NV_IOASID_GET_PRESENT" "" "functions"
         ;;
 
-        mm_pasid_set)
+        mm_pasid_drop)
             #
-            # Determine if mm_pasid_set() function is present
+            # Determine if mm_pasid_drop() function is present
+            #
+            # Added by commit 701fac40384f ("iommu/sva: Assign a PASID to mm
+            # on PASID allocation and free it on mm exit") in v5.18.
+            # Moved to linux/iommu.h in commit cd3891158a77 ("iommu/sva: Move
+            # PASID helpers to sva code") in v6.4.
             #
-            # mm_pasid_set() function was added by commit
-            # 701fac40384f07197b106136012804c3cae0b3de (iommu/sva: Assign a
-            # PASID to mm on PASID allocation and free it on mm exit) in v5.18.
-            # (2022-02-15).
             CODE="
             #if defined(NV_LINUX_SCHED_MM_H_PRESENT)
             #include <linux/sched/mm.h>
             #endif
-            void conftest_mm_pasid_set(void) {
-                mm_pasid_set();
+            #include <linux/iommu.h>
+            void conftest_mm_pasid_drop(void) {
+                mm_pasid_drop();
             }"
 
-            compile_check_conftest "$CODE" "NV_MM_PASID_SET_PRESENT" "" "functions"
+            compile_check_conftest "$CODE" "NV_MM_PASID_DROP_PRESENT" "" "functions"
         ;;
 
         drm_crtc_state_has_no_vblank)
@@ -6341,6 +6326,22 @@ compile_test() {
             compile_check_conftest "$CODE" "NV_MEMPOLICY_HAS_HOME_NODE" "" "types"
         ;;
 
+        mpol_preferred_many_present)
+            #
+            # Determine if MPOL_PREFERRED_MANY enum is present or not
+            #
+            # Added by commit b27abaccf8e8b ("mm/mempolicy: add
+            # MPOL_PREFERRED_MANY for multiple preferred nodes") in
+            # v5.15
+            #
+            CODE="
+            #include <linux/mempolicy.h>
+            int mpol_preferred_many = MPOL_PREFERRED_MANY;
+            "
+
+            compile_check_conftest "$CODE" "NV_MPOL_PREFERRED_MANY_PRESENT" "" "types"
+        ;;
+
         mmu_interval_notifier)
             #
             # Determine if mmu_interval_notifier struct is present or not

kernel-open/nvidia-uvm/nvidia-uvm.Kbuild

@@ -81,8 +81,7 @@ NV_CONFTEST_FUNCTION_COMPILE_TESTS += set_memory_uc
 NV_CONFTEST_FUNCTION_COMPILE_TESTS += set_pages_uc
 NV_CONFTEST_FUNCTION_COMPILE_TESTS += ktime_get_raw_ts64
 NV_CONFTEST_FUNCTION_COMPILE_TESTS += ioasid_get
-NV_CONFTEST_FUNCTION_COMPILE_TESTS += mm_pasid_set
-NV_CONFTEST_FUNCTION_COMPILE_TESTS += migrate_vma_setup
+NV_CONFTEST_FUNCTION_COMPILE_TESTS += mm_pasid_drop
 NV_CONFTEST_FUNCTION_COMPILE_TESTS += mmget_not_zero
 NV_CONFTEST_FUNCTION_COMPILE_TESTS += mmgrab
 NV_CONFTEST_FUNCTION_COMPILE_TESTS += iommu_sva_bind_device_has_drvdata_arg
@@ -110,6 +109,8 @@ NV_CONFTEST_TYPE_COMPILE_TESTS += handle_mm_fault_has_mm_arg
 NV_CONFTEST_TYPE_COMPILE_TESTS += handle_mm_fault_has_pt_regs_arg
 NV_CONFTEST_TYPE_COMPILE_TESTS += mempolicy_has_unified_nodes
 NV_CONFTEST_TYPE_COMPILE_TESTS += mempolicy_has_home_node
+NV_CONFTEST_TYPE_COMPILE_TESTS += mpol_preferred_many_present
 NV_CONFTEST_TYPE_COMPILE_TESTS += mmu_interval_notifier
 
 NV_CONFTEST_SYMBOL_COMPILE_TESTS += is_export_symbol_present_int_active_memcg
+NV_CONFTEST_SYMBOL_COMPILE_TESTS += is_export_symbol_present_migrate_vma_setup

kernel-open/nvidia-uvm/uvm_ats_faults.c

@@ -1,5 +1,5 @@
 /*******************************************************************************
-    Copyright (c) 2018 NVIDIA Corporation
+    Copyright (c) 2023 NVIDIA Corporation
 
     Permission is hereby granted, free of charge, to any person obtaining a copy
     of this software and associated documentation files (the "Software"), to
@@ -149,7 +149,11 @@ static void ats_batch_select_residency(uvm_gpu_va_space_t *gpu_va_space,
 
     mode = vma_policy->mode;
 
-    if ((mode == MPOL_BIND) || (mode == MPOL_PREFERRED_MANY) || (mode == MPOL_PREFERRED)) {
+    if ((mode == MPOL_BIND)
+#if defined(NV_MPOL_PREFERRED_MANY_PRESENT)
+         || (mode == MPOL_PREFERRED_MANY)
+#endif
+         || (mode == MPOL_PREFERRED)) {
         int home_node = NUMA_NO_NODE;
 
 #if defined(NV_MEMPOLICY_HAS_HOME_NODE)
@@ -467,6 +471,10 @@ NV_STATUS uvm_ats_service_faults(uvm_gpu_va_space_t *gpu_va_space,
             uvm_page_mask_and(write_fault_mask, write_fault_mask, read_fault_mask);
         else
             uvm_page_mask_zero(write_fault_mask);
+
+        // There are no pending faults beyond write faults to RO region.
+        if (uvm_page_mask_empty(read_fault_mask))
+            return status;
     }
 
     ats_batch_select_residency(gpu_va_space, vma, ats_context);

kernel-open/nvidia-uvm/uvm_ats_sva.h

@@ -32,19 +32,23 @@
 // For ATS support on aarch64, arm_smmu_sva_bind() is needed for
 // iommu_sva_bind_device() calls. Unfortunately, arm_smmu_sva_bind() is not
 // conftest-able. We instead look for the presence of ioasid_get() or
-// mm_pasid_set(). ioasid_get() was added in the same patch series as
-// arm_smmu_sva_bind() and removed in v6.0. mm_pasid_set() was added in the
+// mm_pasid_drop(). ioasid_get() was added in the same patch series as
+// arm_smmu_sva_bind() and removed in v6.0. mm_pasid_drop() was added in the
 // same patch as the removal of ioasid_get(). We assume the presence of
-// arm_smmu_sva_bind() if ioasid_get(v5.11 - v5.17) or mm_pasid_set(v5.18+) is
+// arm_smmu_sva_bind() if ioasid_get(v5.11 - v5.17) or mm_pasid_drop(v5.18+) is
 // present.
 //
 // arm_smmu_sva_bind() was added with commit
 // 32784a9562fb0518b12e9797ee2aec52214adf6f and ioasid_get() was added with
 // commit cb4789b0d19ff231ce9f73376a023341300aed96 (11/23/2020). Commit
 // 701fac40384f07197b106136012804c3cae0b3de (02/15/2022) removed ioasid_get()
-// and added mm_pasid_set().
-    #if UVM_CAN_USE_MMU_NOTIFIERS() && (defined(NV_IOASID_GET_PRESENT) || defined(NV_MM_PASID_SET_PRESENT))
-        #define UVM_ATS_SVA_SUPPORTED() 1
+// and added mm_pasid_drop().
+    #if UVM_CAN_USE_MMU_NOTIFIERS() && (defined(NV_IOASID_GET_PRESENT) || defined(NV_MM_PASID_DROP_PRESENT))
+        #if defined(CONFIG_IOMMU_SVA)
+            #define UVM_ATS_SVA_SUPPORTED() 1
+        #else
+            #define UVM_ATS_SVA_SUPPORTED() 0
+        #endif
     #else
         #define UVM_ATS_SVA_SUPPORTED() 0
     #endif

kernel-open/nvidia-uvm/uvm_ce_test.c

@@ -191,7 +191,7 @@ static NV_STATUS test_membar(uvm_gpu_t *gpu)
 
     for (i = 0; i < REDUCTIONS; ++i) {
         uvm_push_set_flag(&push, UVM_PUSH_FLAG_NEXT_MEMBAR_NONE);
-        gpu->parent->ce_hal->semaphore_reduction_inc(&push, host_mem_gpu_va, REDUCTIONS + 1);
+        gpu->parent->ce_hal->semaphore_reduction_inc(&push, host_mem_gpu_va, REDUCTIONS);
     }
 
     // Without a sys membar the channel tracking semaphore can and does complete
@@ -577,7 +577,7 @@ static NV_STATUS test_semaphore_reduction_inc(uvm_gpu_t *gpu)
 
     for (i = 0; i < REDUCTIONS; i++) {
         uvm_push_set_flag(&push, UVM_PUSH_FLAG_NEXT_MEMBAR_NONE);
-        gpu->parent->ce_hal->semaphore_reduction_inc(&push, gpu_va, i+1);
+        gpu->parent->ce_hal->semaphore_reduction_inc(&push, gpu_va, REDUCTIONS);
     }
 
     status = uvm_push_end_and_wait(&push);