blob: 2a70dcafa171b96416562259fa63b4facadba7ff (
plain) (
blame)
1
2
3
4
5
6
7
8
9
10
11
12
|
accel/hvf: Instance size not properly declared
Description of problem:
In [`include/sysemu/hvf.h`](https://gitlab.com/qemu-project/qemu/-/blob/master/include/sysemu/hvf.h#L36), `HVFState` is declared to have the QOM type `TYPE_HVF_ACCEL`;
However, when the type is registered, proper `instance_size` for it was [not declared](https://gitlab.com/qemu-project/qemu/-/blob/master/accel/hvf/hvf-accel-ops.c#L351).
As a result, a bad workaround was introduced. That is, when [`hvf_accel_init`](https://gitlab.com/qemu-project/qemu/-/blob/master/accel/hvf/hvf-accel-ops.c#L329) is called from [`accel_init_machine`](https://gitlab.com/qemu-project/qemu/-/blob/master/accel/accel-softmmu.c#L33), an new instance of `HVFState` is allocated while we should have used the pre-allocated instance in `ms->accelerator` similar to [what KVM does](https://gitlab.com/qemu-project/qemu/-/blob/master/accel/kvm/kvm-all.c#L2381) (the code didn't do so since the allocated ([using `object_new_with_class`](https://gitlab.com/qemu-project/qemu/-/blob/master/softmmu/vl.c#L2218)) instance didn't allocate enough memory for `HVFState`).
Eventhough the code wouldn't crash nor have any serious implication, this would leak an `AccelState` and attempts to manually manage accelerators would cause a buffer-overflow.
Steps to reproduce:
1. Run a HVF-accelerated VM
Additional information:
|