570.169

kernel-open/nvidia/nv-caps.c

@@ -1,5 +1,5 @@
 /*
- * SPDX-FileCopyrightText: Copyright (c) 2019-2024 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+ * SPDX-FileCopyrightText: Copyright (c) 2019-2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
  * SPDX-License-Identifier: MIT
  *
  * Permission is hereby granted, free of charge, to any person obtaining a
@@ -582,6 +582,9 @@ err:
 void NV_API_CALL nv_cap_close_fd(int fd)
 {
 #if NV_FILESYSTEM_ACCESS_AVAILABLE
+    struct file *file;
+    NvBool is_nv_cap_fd;
+
     if (fd == -1)
     {
         return;
@@ -600,6 +603,30 @@ void NV_API_CALL nv_cap_close_fd(int fd)
         return;
     }
 
+    file = fget(fd);
+    if (file == NULL)
+    {
+        task_unlock(current);
+        return;
+    }
+
+    /* Make sure the fd belongs to the nv-cap-drv */
+    is_nv_cap_fd = (file->f_op == &g_nv_cap_drv_fops);
+
+    fput(file);
+
+    /*
+     * In some cases, we may be in shutdown path and execute
+     * in context of unrelated process. In that case we should
+     * not access any 'current' state, but instead let kernel
+     * clean up capability files on its own.
+     */
+    if (!is_nv_cap_fd)
+    {
+        task_unlock(current);
+        return;
+    }
+
 /*
  * From v4.17-rc1 (to v5.10.8) kernels have stopped exporting sys_close(fd)
  * and started exporting __close_fd, as of this commit:

kernel-open/nvidia/nv-mmap.c

@@ -1,5 +1,5 @@
 /*
- * SPDX-FileCopyrightText: Copyright (c) 1999-2024 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+ * SPDX-FileCopyrightText: Copyright (c) 1999-2025 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
  * SPDX-License-Identifier: MIT
  *
  * Permission is hereby granted, free of charge, to any person obtaining a
@@ -867,3 +867,75 @@ void NV_API_CALL nv_set_safe_to_mmap_locked(
 
     nvl->safe_to_mmap = safe_to_mmap;
 }
+
+#if !NV_CAN_CALL_VMA_START_WRITE
+static NvBool nv_vma_enter_locked(struct vm_area_struct *vma, NvBool detaching)
+{
+    NvU32 tgt_refcnt = VMA_LOCK_OFFSET;
+    NvBool interrupted = NV_FALSE;
+    if (!detaching)
+    {
+        tgt_refcnt++;
+    }
+    if (!refcount_add_not_zero(VMA_LOCK_OFFSET, &vma->vm_refcnt))
+    {
+        return NV_FALSE;
+    }
+
+    rwsem_acquire(&vma->vmlock_dep_map, 0, 0, _RET_IP_);
+    prepare_to_rcuwait(&vma->vm_mm->vma_writer_wait);
+
+    for (;;)
+    {
+        set_current_state(TASK_UNINTERRUPTIBLE);
+        if (refcount_read(&vma->vm_refcnt) == tgt_refcnt)
+            break;
+
+        if (signal_pending_state(TASK_UNINTERRUPTIBLE, current))
+        {
+            interrupted = NV_TRUE;
+            break;
+        }
+
+        schedule();
+    }
+
+    // This is an open-coded version of finish_rcuwait().
+    rcu_assign_pointer(vma->vm_mm->vma_writer_wait.task, NULL);
+    __set_current_state(TASK_RUNNING);
+
+    if (interrupted)
+    {
+        // Clean up on error: release refcount and dep_map
+        refcount_sub_and_test(VMA_LOCK_OFFSET, &vma->vm_refcnt);
+        rwsem_release(&vma->vmlock_dep_map, _RET_IP_);
+        return NV_FALSE;
+    }
+
+    lock_acquired(&vma->vmlock_dep_map, _RET_IP_);
+    return NV_TRUE;
+}
+
+/*
+ * Helper function to handle VMA locking and refcount management.
+ */
+void nv_vma_start_write(struct vm_area_struct *vma)
+{
+    NvU32 mm_lock_seq;
+    NvBool locked;
+    if (__is_vma_write_locked(vma, &mm_lock_seq))
+        return;
+
+    locked = nv_vma_enter_locked(vma, NV_FALSE);
+
+    WRITE_ONCE(vma->vm_lock_seq, mm_lock_seq);
+    if (locked)
+    {
+        NvBool detached;
+        detached = refcount_sub_and_test(VMA_LOCK_OFFSET, &vma->vm_refcnt);
+        rwsem_release(&vma->vmlock_dep_map, _RET_IP_);
+        WARN_ON_ONCE(detached);
+    }
+}
+EXPORT_SYMBOL(nv_vma_start_write);
+#endif // !NV_CAN_CALL_VMA_START_WRITE

kernel-open/nvidia/nvidia.Kbuild

@@ -236,6 +236,9 @@ NV_CONFTEST_SYMBOL_COMPILE_TESTS += is_export_symbol_gpl_pci_ats_supported
 NV_CONFTEST_SYMBOL_COMPILE_TESTS += ecc_digits_from_bytes
 NV_CONFTEST_SYMBOL_COMPILE_TESTS += is_export_symbol_present_timer_delete_sync
 NV_CONFTEST_SYMBOL_COMPILE_TESTS += is_export_symbol_present_hrtimer_setup
+NV_CONFTEST_SYMBOL_COMPILE_TESTS += is_export_symbol_gpl___vma_start_write
+NV_CONFTEST_SYMBOL_COMPILE_TESTS += is_export_symbol_gpl_iommu_dev_enable_feature
+NV_CONFTEST_SYMBOL_COMPILE_TESTS += is_export_symbol_gpl_iommu_dev_disable_feature
 
 NV_CONFTEST_TYPE_COMPILE_TESTS += dma_ops
 NV_CONFTEST_TYPE_COMPILE_TESTS += swiotlb_dma_ops
@@ -257,7 +260,7 @@ NV_CONFTEST_TYPE_COMPILE_TESTS += add_memory_driver_managed_has_mhp_flags_arg
 NV_CONFTEST_TYPE_COMPILE_TESTS += num_registered_fb
 NV_CONFTEST_TYPE_COMPILE_TESTS += pci_driver_has_driver_managed_dma
 NV_CONFTEST_TYPE_COMPILE_TESTS += vm_area_struct_has_const_vm_flags
-NV_CONFTEST_TYPE_COMPILE_TESTS += memory_failure_has_trapno_arg
+NV_CONFTEST_TYPE_COMPILE_TESTS += memory_failure_queue_has_trapno_arg
 NV_CONFTEST_TYPE_COMPILE_TESTS += foll_longterm_present
 NV_CONFTEST_TYPE_COMPILE_TESTS += bus_type_has_iommu_ops
 NV_CONFTEST_TYPE_COMPILE_TESTS += class_create_has_no_owner_arg

kernel-open/nvidia/os-interface.c

@@ -2596,7 +2596,6 @@ NV_STATUS NV_API_CALL os_offline_page_at_address
 {
 #if defined(CONFIG_MEMORY_FAILURE)
     int flags = 0;
-    int ret;
     NvU64 pfn;
     struct page *page = NV_GET_PAGE_STRUCT(address);
 
@@ -2621,22 +2620,18 @@ NV_STATUS NV_API_CALL os_offline_page_at_address
     flags |= MF_SW_SIMULATED;
 #endif
 
-#ifdef NV_MEMORY_FAILURE_HAS_TRAPNO_ARG
-    ret = memory_failure(pfn, 0, flags);
-#else
-    ret = memory_failure(pfn, flags);
-#endif
+    nv_printf(NV_DBG_INFO, "NVRM: offlining page at address: 0x%llx pfn: 0x%llx\n",
+              address, pfn);
 
-    if (ret != 0)
-    {
-        nv_printf(NV_DBG_ERRORS, "NVRM: page offlining failed. address: 0x%llx pfn: 0x%llx ret: %d\n",
-                  address, pfn, ret);
-        return NV_ERR_OPERATING_SYSTEM;
-    }
+#ifdef NV_MEMORY_FAILURE_QUEUE_HAS_TRAPNO_ARG
+    memory_failure_queue(pfn, 0, flags);
+#else
+    memory_failure_queue(pfn, flags);
+#endif
 
     return NV_OK;
 #else // !defined(CONFIG_MEMORY_FAILURE)
-    nv_printf(NV_DBG_ERRORS, "NVRM: memory_failure() not supported by kernel. page offlining failed. address: 0x%llx\n",
+    nv_printf(NV_DBG_ERRORS, "NVRM: memory_failure_queue() not supported by kernel. page offlining failed. address: 0x%llx\n",
               address);
     return NV_ERR_NOT_SUPPORTED;
 #endif