Re: [PATCH v7 2/3] uacce: add uacce driver

[Jean-Philippe Brucker <jean-philippe@linaro.org>]

Hi Zhangfei,

Thanks for simplifying this, it's a lot easier to review. I have some
additional comments.

On Tue, Oct 29, 2019 at 02:40:15PM +0800, Zhangfei Gao wrote:
> +static int uacce_sva_exit(struct device *dev, struct iommu_sva *handle,
> +			  void *data)
> +{
> +	struct uacce_device *uacce = data;
> +	struct uacce_queue *q;
> +
> +	mutex_lock(&uacce->q_lock);
> +	list_for_each_entry(q, &uacce->qs, list) {
> +		if (q->pid == task_pid_nr(current))
> +			uacce_put_queue(q);

This won't work in some cases, because any thread can call __mmput() and
end up here. For example a sibling thread that inherited the queue, or a
workqueue that's executing mmput_async_fn(). In addition I think comparing
PID values is unsafe (see comment in pid.h), we'd need to use the struct
pid if we wanted to do it this way.

But I still believe it would be better to create an uacce_mm structure
that tracks all queues bound to this mm, and pass that to uacce_sva_exit
instead of the uacce_device.

The queue isn't bound to a task, but its address space. With clone() the
address space can be shared between tasks. In addition, whoever has a
queue fd also gets access to this address space. So after a fork() the
child may be able to program the queue to DMA into the parent's address
space, even without CLONE_VM. Users must be aware of this and I think it's
important to explain it very clearly in the UAPI.

[...]
> +static struct uacce_qfile_region *
> +uacce_create_region(struct uacce_queue *q, struct vm_area_struct *vma,
> +		    enum uacce_qfrt type, unsigned int flags)
> +{
> +	struct uacce_device *uacce = q->uacce;
> +	struct uacce_qfile_region *qfr;
> +	int ret = -ENOMEM;
> +
> +	qfr = kzalloc(sizeof(*qfr), GFP_KERNEL);
> +	if (!qfr)
> +		return ERR_PTR(-ENOMEM);
> +
> +	qfr->type = type;
> +	qfr->flags = flags;
> +
> +	if (vma->vm_flags & VM_READ)
> +		qfr->prot |= IOMMU_READ;

qfr->prot and qfr->flags aren't used at the moment, you could remove them.

> +
> +	if (vma->vm_flags & VM_WRITE)
> +		qfr->prot |= IOMMU_WRITE;
> +
> +	if (flags & UACCE_QFRF_SELFMT) {
> +		if (!uacce->ops->mmap) {
> +			ret = -EINVAL;
> +			goto err_with_qfr;
> +		}
> +
> +		ret = uacce->ops->mmap(q, vma, qfr);
> +		if (ret)
> +			goto err_with_qfr;
> +		return qfr;
> +	}
> +
> +	return qfr;
> +
> +err_with_qfr:
> +	kfree(qfr);
> +	return ERR_PTR(ret);
> +}
> +
> +static int uacce_fops_mmap(struct file *filep, struct vm_area_struct *vma)
> +{
> +	struct uacce_queue *q = filep->private_data;
> +	struct uacce_device *uacce = q->uacce;
> +	struct uacce_qfile_region *qfr;
> +	enum uacce_qfrt type = 0;
> +	unsigned int flags = 0;
> +	int ret;
> +
> +	if (vma->vm_pgoff < UACCE_QFRT_MAX)
> +		type = vma->vm_pgoff;

Otherwise return -EINVAL?  type probably shouldn't default to MMIO if it
wasn't explicitly requested by the user.

> +
> +	vma->vm_flags |= VM_DONTCOPY | VM_DONTEXPAND | VM_WIPEONFORK;
> +	vma->vm_ops = &uacce_vm_ops;
> +	vma->vm_private_data = q;
> +
> +	mutex_lock(&uacce_mutex);
> +
> +	if (q->qfrs[type]) {
> +		ret = -EEXIST;
> +		goto out_with_lock;
> +	}
> +
> +	switch (type) {
> +	case UACCE_QFRT_MMIO:
> +		flags = UACCE_QFRF_SELFMT;
> +		break;
> +
> +	case UACCE_QFRT_DUS:
> +		if (uacce->flags & UACCE_DEV_SVA) {
> +			flags = UACCE_QFRF_SELFMT;

I'd simplify this even further by getting rid of the SELFMT flag. It's the
only possibility at the moment.

> +			break;
> +		}
> +		break;
> +
> +	default:
> +		WARN_ON(&uacce->dev);

WARN_ON(uacce->dev). But shouldn't we instead return -EINVAL here?
UACCE_QFRT_MAX is currently 16, so users can easily trigger this WARN by
passing an invalid value.

[...]
> +void uacce_unregister(struct uacce_device *uacce)
> +{
> +	if (!uacce)
> +		return;
> +
> +	mutex_lock(&uacce->q_lock);
> +	if (!list_empty(&uacce->qs)) {
> +		struct uacce_queue *q;
> +
> +		list_for_each_entry(q, &uacce->qs, list) {
> +			uacce_put_queue(q);

The open file descriptor will still exist after this function returns.
Can all fops can be called with a stale queue?

Thanks,
Jean