Re: [PATCH v3 8/8] vdpa_sim: add support for user VA

[Stefano Garzarella]

On Fri, Mar 24, 2023 at 11:49:32AM +0800, Jason Wang wrote:
> 在 2023/3/21 23:48, Stefano Garzarella 写道:
> > The new "use_va" module parameter (default: true) is used in
> > vdpa_alloc_device() to inform the vDPA framework that the device
> > supports VA.
> >
> > vringh is initialized to use VA only when "use_va" is true and the
> > user's mm has been bound. So, only when the bus supports user VA
> > (e.g. vhost-vdpa).
> >
> > vdpasim_mm_work_fn work is used to serialize the binding to a new
> > address space when the .bind_mm callback is invoked, and unbinding
> > when the .unbind_mm callback is invoked.
> >
> > Call mmget_not_zero()/kthread_use_mm() inside the worker function
> > to pin the address space only as long as needed, following the
> > documentation of mmget() in include/linux/sched/mm.h:
> >
> >   * Never use this function to pin this address space for an
> >   * unbounded/indefinite amount of time.
> >
> > Signed-off-by: Stefano Garzarella <sgarzare@xxxxxxxxxx>
> > ---
> >
> > Notes:
> >     v3:
> >     - called mmget_not_zero() before kthread_use_mm() [Jason]
> >       As the documentation of mmget() in include/linux/sched/mm.h says:
> >       * Never use this function to pin this address space for an
> >       * unbounded/indefinite amount of time.
> >       I moved mmget_not_zero/kthread_use_mm inside the worker function,
> >       this way we pin the address space only as long as needed.
> >       This is similar to what vfio_iommu_type1_dma_rw_chunk() does in
> >       drivers/vfio/vfio_iommu_type1.c
> >     - simplified the mm bind/unbind [Jason]
> >     - renamed vdpasim_worker_change_mm_sync() [Jason]
> >     - fix commit message (s/default: false/default: true)
> >     v2:
> >     - `use_va` set to true by default [Eugenio]
> >     - supported the new unbind_mm callback [Jason]
> >     - removed the unbind_mm call in vdpasim_do_reset() [Jason]
> >     - avoided to release the lock while call kthread_flush_work() since we
> >       are now using a mutex to protect the device state
> >
> >  drivers/vdpa/vdpa_sim/vdpa_sim.h |  1 +
> >  drivers/vdpa/vdpa_sim/vdpa_sim.c | 80 +++++++++++++++++++++++++++++++-
> >  2 files changed, 79 insertions(+), 2 deletions(-)
> >
> > diff --git a/drivers/vdpa/vdpa_sim/vdpa_sim.h b/drivers/vdpa/vdpa_sim/vdpa_sim.h
> > index 4774292fba8c..3a42887d05d9 100644
> > --- a/drivers/vdpa/vdpa_sim/vdpa_sim.h
> > +++ b/drivers/vdpa/vdpa_sim/vdpa_sim.h
> > @@ -59,6 +59,7 @@ struct vdpasim {
> >  	struct vdpasim_virtqueue *vqs;
> >  	struct kthread_worker *worker;
> >  	struct kthread_work work;
> > +	struct mm_struct *mm_bound;
> >  	struct vdpasim_dev_attr dev_attr;
> >  	/* mutex to synchronize virtqueue state */
> >  	struct mutex mutex;
> > diff --git a/drivers/vdpa/vdpa_sim/vdpa_sim.c b/drivers/vdpa/vdpa_sim/vdpa_sim.c
> > index ab4cfb82c237..23c891cdcd54 100644
> > --- a/drivers/vdpa/vdpa_sim/vdpa_sim.c
> > +++ b/drivers/vdpa/vdpa_sim/vdpa_sim.c
> > @@ -35,10 +35,44 @@ module_param(max_iotlb_entries, int, 0444);
> >  MODULE_PARM_DESC(max_iotlb_entries,
> >  		 "Maximum number of iotlb entries for each address space. 0 means unlimited. (default: 2048)");
> > +static bool use_va = true;
> > +module_param(use_va, bool, 0444);
> > +MODULE_PARM_DESC(use_va, "Enable/disable the device's ability to use VA");
> > +
> >  #define VDPASIM_QUEUE_ALIGN PAGE_SIZE
> >  #define VDPASIM_QUEUE_MAX 256
> >  #define VDPASIM_VENDOR_ID 0
> > +struct vdpasim_mm_work {
> > +	struct kthread_work work;
> > +	struct vdpasim *vdpasim;
> > +	struct mm_struct *mm_to_bind;
> > +	int ret;
> > +};
> > +
> > +static void vdpasim_mm_work_fn(struct kthread_work *work)
> > +{
> > +	struct vdpasim_mm_work *mm_work =
> > +		container_of(work, struct vdpasim_mm_work, work);
> > +	struct vdpasim *vdpasim = mm_work->vdpasim;
> > +
> > +	mm_work->ret = 0;
> > +
> > +	//TODO: should we attach the cgroup of the mm owner?
> > +	vdpasim->mm_bound = mm_work->mm_to_bind;
> > +}
> > +
> > +static void vdpasim_worker_change_mm_sync(struct vdpasim *vdpasim,
> > +					  struct vdpasim_mm_work *mm_work)
> > +{
> > +	struct kthread_work *work = &mm_work->work;
> > +
> > +	kthread_init_work(work, vdpasim_mm_work_fn);
> > +	kthread_queue_work(vdpasim->worker, work);
> > +
> > +	kthread_flush_work(work);
> > +}
> > +
> >  static struct vdpasim *vdpa_to_sim(struct vdpa_device *vdpa)
> >  {
> >  	return container_of(vdpa, struct vdpasim, vdpa);
> > @@ -59,8 +93,10 @@ static void vdpasim_queue_ready(struct vdpasim *vdpasim, unsigned int idx)
> >  {
> >  	struct vdpasim_virtqueue *vq = &vdpasim->vqs[idx];
> >  	uint16_t last_avail_idx = vq->vring.last_avail_idx;
> > +	bool va_enabled = use_va && vdpasim->mm_bound;
> > -	vringh_init_iotlb(&vq->vring, vdpasim->features, vq->num, true, false,
> > +	vringh_init_iotlb(&vq->vring, vdpasim->features, vq->num, true,
> > +			  va_enabled,
> >  			  (struct vring_desc *)(uintptr_t)vq->desc_addr,
> >  			  (struct vring_avail *)
> >  			  (uintptr_t)vq->driver_addr,
> > @@ -130,8 +166,20 @@ static const struct vdpa_config_ops vdpasim_batch_config_ops;
> >  static void vdpasim_work_fn(struct kthread_work *work)
> >  {
> >  	struct vdpasim *vdpasim = container_of(work, struct vdpasim, work);
> > +	struct mm_struct *mm = vdpasim->mm_bound;
> > +
> > +	if (mm) {
> > +		if (!mmget_not_zero(mm))
> > +			return;
>
>
> Do we need to check use_va here.

Yep, right!

> Other than this
>
> Acked-by: Jason Wang <jasowang@xxxxxxxxxx>

Thanks for the reviews,
Stefano