CVE-2025-10865

Software installed and run as a non-privileged user may conduct improper GPU system calls to cause mismanagement of resources and reference counting, creating a potential write use-after-free scenario.

Software installed and run as a non-privileged user may conduct GPU system calls which cause GPU memory leaks and possible kernel heap corruption. This is caused by memory free paths not maintaining state data of upgraded higher order allocations, potentially leading to memory leak or double free.

Software installed and run as a non-privileged user may conduct improper GPU system calls to pass invalid log2 page size when allocating physical pages, leading to OOB read and/or write due to improper validation. This could lead to 4K pages being treated as higher order pages and allowing access beyond the 4K threshold.

Software running as a non-privileged user may perform improper GPU system calls, leading to out-of-bounds access and kernel null pointer dereference in an error path. The null pointer dereference occurs in a function running in a kernel thread, causing kernel exceptions, platform instability, and denial of service.

Kernel software running inside a Guest VM can send improper commands to the GPU firmware, causing out-of-bounds memory writes outside the guest's virtualized GPU memory.

Kernel software running inside a Guest VM can send improper commands to the GPU firmware, triggering out-of-bounds memory accesses. This can be used for privilege escalation escaping virtualization boundaries.

Software run by a non-privileged user can issue improper GPU system calls to manipulate the lifetimes of synchronization objects in the kernel, leading to read/write use-after-free (UAF) conditions.

A vulnerability in the GPU driver allows a non-privileged user to execute a sequence of improper GPU system calls, leading to use-after-free. This facilitates unauthorized memory access from shader code.

A vulnerability in GPU software allows a non-privileged user to perform improper GPU system calls, causing an integer overflow and mapping two GPU virtual addresses to the same physical address. After one virtual mapping is freed, the physical page can be reallocated, and the second mapping enables access to freed memory (use-after-free).

Software run by a non-privileged user can perform improper GPU system calls, causing mismanagement of a mapping state for a sparse memory allocation. The product accidentally refers to the wrong memory due to the semantics of how math operations are implicitly scaled across buffers of different sizes.