CVE-2024-53195

In the Linux kernel, the following vulnerability has been resolved:<br /> <br /> KVM: arm64: Get rid of userspace_irqchip_in_use<br /> <br /> Improper use of userspace_irqchip_in_use led to syzbot hitting the<br /> following WARN_ON() in kvm_timer_update_irq():<br /> <br /> WARNING: CPU: 0 PID: 3281 at arch/arm64/kvm/arch_timer.c:459<br /> kvm_timer_update_irq+0x21c/0x394<br /> Call trace:<br /> kvm_timer_update_irq+0x21c/0x394 arch/arm64/kvm/arch_timer.c:459<br /> kvm_timer_vcpu_reset+0x158/0x684 arch/arm64/kvm/arch_timer.c:968<br /> kvm_reset_vcpu+0x3b4/0x560 arch/arm64/kvm/reset.c:264<br /> kvm_vcpu_set_target arch/arm64/kvm/arm.c:1553 [inline]<br /> kvm_arch_vcpu_ioctl_vcpu_init arch/arm64/kvm/arm.c:1573 [inline]<br /> kvm_arch_vcpu_ioctl+0x112c/0x1b3c arch/arm64/kvm/arm.c:1695<br /> kvm_vcpu_ioctl+0x4ec/0xf74 virt/kvm/kvm_main.c:4658<br /> vfs_ioctl fs/ioctl.c:51 [inline]<br /> __do_sys_ioctl fs/ioctl.c:907 [inline]<br /> __se_sys_ioctl fs/ioctl.c:893 [inline]<br /> __arm64_sys_ioctl+0x108/0x184 fs/ioctl.c:893<br /> __invoke_syscall arch/arm64/kernel/syscall.c:35 [inline]<br /> invoke_syscall+0x78/0x1b8 arch/arm64/kernel/syscall.c:49<br /> el0_svc_common+0xe8/0x1b0 arch/arm64/kernel/syscall.c:132<br /> do_el0_svc+0x40/0x50 arch/arm64/kernel/syscall.c:151<br /> el0_svc+0x54/0x14c arch/arm64/kernel/entry-common.c:712<br /> el0t_64_sync_handler+0x84/0xfc arch/arm64/kernel/entry-common.c:730<br /> el0t_64_sync+0x190/0x194 arch/arm64/kernel/entry.S:598<br /> <br /> The following sequence led to the scenario:<br /> - Userspace creates a VM and a vCPU.<br /> - The vCPU is initialized with KVM_ARM_VCPU_PMU_V3 during<br /> KVM_ARM_VCPU_INIT.<br /> - Without any other setup, such as vGIC or vPMU, userspace issues<br /> KVM_RUN on the vCPU. Since the vPMU is requested, but not setup,<br /> kvm_arm_pmu_v3_enable() fails in kvm_arch_vcpu_run_pid_change().<br /> As a result, KVM_RUN returns after enabling the timer, but before<br /> incrementing &amp;#39;userspace_irqchip_in_use&amp;#39;:<br /> kvm_arch_vcpu_run_pid_change()<br /> ret = kvm_arm_pmu_v3_enable()<br /> if (!vcpu-&gt;arch.pmu.created)<br /> return -EINVAL;<br /> if (ret)<br /> return ret;<br /> [...]<br /> if (!irqchip_in_kernel(kvm))<br /> static_branch_inc(&amp;userspace_irqchip_in_use);<br /> - Userspace ignores the error and issues KVM_ARM_VCPU_INIT again.<br /> Since the timer is already enabled, control moves through the<br /> following flow, ultimately hitting the WARN_ON():<br /> kvm_timer_vcpu_reset()<br /> if (timer-&gt;enabled)<br /> kvm_timer_update_irq()<br /> if (!userspace_irqchip())<br /> ret = kvm_vgic_inject_irq()<br /> ret = vgic_lazy_init()<br /> if (unlikely(!vgic_initialized(kvm)))<br /> if (kvm-&gt;arch.vgic.vgic_model !=<br /> KVM_DEV_TYPE_ARM_VGIC_V2)<br /> return -EBUSY;<br /> WARN_ON(ret);<br /> <br /> Theoretically, since userspace_irqchip_in_use&amp;#39;s functionality can be<br /> simply replaced by &amp;#39;!irqchip_in_kernel()&amp;#39;, get rid of the static key<br /> to avoid the mismanagement, which also helps with the syzbot issue.