kvm_ipi_create() registers the IPI device on the IOCSR bus and jumps to the error label when that fails. The error path then discards the error code that kvm_io_bus_register_dev() reported and returns -EFAULT instead: ret = kvm_io_bus_register_dev(kvm, KVM_IOCSR_BUS, IOCSR_IPI_BASE, IOCSR_IPI_SIZE, device); mutex_unlock(&kvm->slots_lock); if (ret < 0) { kvm_pr_unimpl("%s: Initialize IOCSR dev failed, ret = %d\n", __func__, ret); goto err; } ... err: kfree(s); return -EFAULT; kvm_io_bus_register_dev() fails with -ENOMEM or -ENOSPC, so userspace creating a KVM_DEV_TYPE_LOONGARCH_IPI device is told that it passed a bad address when the real problem is that the host is out of memory or out of IOCSR bus slots. -EFAULT is reserved for faulting user addresses, and the remaining -EFAULT returns in this file are get_user() failures, where it is correct. Return ret instead. It is guaranteed to hold a valid negative errno because the ret < 0 branch above is the only path that reaches the error label. This also matches kvm_eiointc_create(), which already propagates the kvm_io_bus_register_dev() error code unchanged. Only compile-tested; the fix is a straight propagation of an error code that the adjacent kvm_pr_unimpl() already logs. Fixes: c532de5a67a7 ("LoongArch: KVM: Add IPI device support") Assisted-by: Cursor:claude-opus-5 Signed-off-by: Chaithanya Lagisetty --- arch/loongarch/kvm/intc/ipi.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/arch/loongarch/kvm/intc/ipi.c b/arch/loongarch/kvm/intc/ipi.c index 7b333a4a0430..672ec053a997 100644 --- a/arch/loongarch/kvm/intc/ipi.c +++ b/arch/loongarch/kvm/intc/ipi.c @@ -431,7 +431,7 @@ static int kvm_ipi_create(struct kvm_device *dev, u32 type) err: kfree(s); - return -EFAULT; + return ret; } static void kvm_ipi_destroy(struct kvm_device *dev) -- 2.43.0