]> git.ipfire.org Git - thirdparty/kernel/linux.git/commitdiff
RISC-V: KVM: Fix null pointer dereference in kvm_riscv_vcpu_aia_rmw_topei()
authorJiakai Xu <xujiakai2025@iscas.ac.cn>
Thu, 26 Feb 2026 08:51:19 +0000 (08:51 +0000)
committerAnup Patel <anup@brainfault.org>
Fri, 6 Mar 2026 05:50:30 +0000 (11:20 +0530)
kvm_riscv_vcpu_aia_rmw_topei() assumes that the per-vCPU IMSIC state has
been initialized once AIA is reported as available and initialized at
the VM level. This assumption does not always hold.

Under fuzzed ioctl sequences, a guest may access the IMSIC TOPEI CSR
before the vCPU IMSIC state is set up. In this case,
vcpu->arch.aia_context.imsic_state is still NULL, and the TOPEI RMW path
dereferences it unconditionally, leading to a host kernel crash.

The crash manifests as:
  Unable to handle kernel paging request at virtual address
  dfffffff0000000e
  ...
  kvm_riscv_vcpu_aia_imsic_rmw arch/riscv/kvm/aia_imsic.c:909
  kvm_riscv_vcpu_aia_rmw_topei arch/riscv/kvm/aia.c:231
  csr_insn arch/riscv/kvm/vcpu_insn.c:208
  system_opcode_insn arch/riscv/kvm/vcpu_insn.c:281
  kvm_riscv_vcpu_virtual_insn arch/riscv/kvm/vcpu_insn.c:355
  kvm_riscv_vcpu_exit arch/riscv/kvm/vcpu_exit.c:230
  kvm_arch_vcpu_ioctl_run arch/riscv/kvm/vcpu.c:1008
  ...

Fix this by explicitly checking whether the vCPU IMSIC state has been
initialized before handling TOPEI CSR accesses. If not, forward the CSR
emulation to user space.

Fixes: db8b7e97d6137 ("RISC-V: KVM: Add in-kernel virtualization of AIA IMSIC")
Signed-off-by: Jiakai Xu <xujiakai2025@iscas.ac.cn>
Signed-off-by: Jiakai Xu <jiakaiPeanut@gmail.com>
Reviewed-by: Nutty Liu <nutty.liu@hotmail.com>
Reviewed-by: Anup Patel <anup@brainfault.org>
Link: https://lore.kernel.org/r/20260226085119.643295-1-xujiakai2025@iscas.ac.cn
Signed-off-by: Anup Patel <anup@brainfault.org>
arch/riscv/kvm/aia_imsic.c

index 06752fa247987da75e8509d799cdf2a8e37b3d85..8786f52cf65a2c6570901afe66c57df8ddaf0b02 100644 (file)
@@ -908,6 +908,10 @@ int kvm_riscv_vcpu_aia_imsic_rmw(struct kvm_vcpu *vcpu, unsigned long isel,
        int r, rc = KVM_INSN_CONTINUE_NEXT_SEPC;
        struct imsic *imsic = vcpu->arch.aia_context.imsic_state;
 
+       /* If IMSIC vCPU state not initialized then forward to user space */
+       if (!imsic)
+               return KVM_INSN_EXIT_TO_USER_SPACE;
+
        if (isel == KVM_RISCV_AIA_IMSIC_TOPEI) {
                /* Read pending and enabled interrupt with highest priority */
                topei = imsic_mrif_topei(imsic->swfile, imsic->nr_eix,