> -----Original Message-----
> From: Maxime Coquelin [mailto:maxime.coque...@redhat.com]
> Sent: Thursday, September 12, 2019 5:27 PM
> To: Yu, Jin <jin...@intel.com>; dev@dpdk.org
> Cc: Liu, Changpeng <changpeng....@intel.com>; Bie, Tiwei
> <tiwei....@intel.com>; Wang, Zhihong <zhihong.w...@intel.com>; Lin Li
> <lili...@baidu.com>; Xun Ni <ni...@baidu.com>; Yu Zhang
> <zhangy...@baidu.com>
> Subject: Re: [PATCH v6 06/10] vhost: add the APIs to operate inflight ring
>
>
>
> On 8/29/19 4:12 PM, JinYu wrote:
> > This patch introduces three APIs to operate the inflight ring. Three
> > APIs are set, set last and clear. It includes split and packed ring.
> >
> > Signed-off-by: Lin Li <lili...@baidu.com>
> > Signed-off-by: Xun Ni <ni...@baidu.com>
> > Signed-off-by: Yu Zhang <zhangy...@baidu.com>
> > Signed-off-by: Jin Yu <jin...@intel.com>
> > ---
> > lib/librte_vhost/rte_vhost.h | 116 ++++++++++++++++
> > lib/librte_vhost/vhost.c | 252 +++++++++++++++++++++++++++++++++++
> > 2 files changed, 368 insertions(+)
> >
> > diff --git a/lib/librte_vhost/rte_vhost.h
> > b/lib/librte_vhost/rte_vhost.h index e090cdfee..d3b6eda21 100644
> > --- a/lib/librte_vhost/rte_vhost.h
> > +++ b/lib/librte_vhost/rte_vhost.h
> > @@ -693,6 +693,122 @@ int rte_vhost_get_mem_table(int vid, struct
> > rte_vhost_memory **mem); int rte_vhost_get_vhost_vring(int vid, uint16_t
> vring_idx,
> > struct rte_vhost_vring *vring);
> >
> > +/**
> > + * Set split inflight descriptor.
> > + *
> > + * This function save descriptors that has been comsumed in available
> > + * ring
> > + *
> > + * @param vid
> > + * vhost device ID
> > + * @param vring_idx
> > + * vring index
> > + * @param idx
> > + * inflight entry index
> > + * @return
> > + * 0 on success, -1 on failure
> > + */
> > +__rte_experimental
> > +int
> > +rte_vhost_set_inflight_desc_split(int vid, uint16_t vring_idx,
> > + uint16_t idx);
> > +
> > +/**
> > + * Set packed inflight descriptor and get corresponding inflight
> > +entry
> > + *
> > + * This function save descriptors that has been comsumed
> > + *
> > + * @param vid
> > + * vhost device ID
> > + * @param vring_idx
> > + * vring index
> > + * @param head
> > + * head of descriptors
> > + * @param last
> > + * last of descriptors
> > + * @param inflight_entry
> > + * corresponding inflight entry
> > + * @return
> > + * 0 on success, -1 on failure
> > + */
> > +__rte_experimental
> > +int
> > +rte_vhost_set_inflight_desc_packed(int vid, uint16_t vring_idx,
> > + uint16_t head, uint16_t last, uint16_t *inflight_entry);
> > +
> > +/**
> > + * Save the head of list that the last batch of used descriptors.
> > + *
> > + * @param vid
> > + * vhost device ID
> > + * @param vring_idx
> > + * vring index
> > + * @param idx
> > + * descriptor entry index
> > + * @return
> > + * 0 on success, -1 on failure
> > + */
> > +__rte_experimental
> > +int
> > +rte_vhost_set_last_inflight_io_split(int vid,
> > + uint16_t vring_idx, uint16_t idx);
> > +
> > +/**
> > + * Update the inflight free_head, used_idx and used_wrap_counter.
> > + *
> > + * This function will update status first before updating descriptors
> > + * to used
> > + *
> > + * @param vid
> > + * vhost device ID
> > + * @param vring_idx
> > + * vring index
> > + * @param head
> > + * head of descriptors
> > + * @return
> > + * 0 on success, -1 on failure
> > + */
> > +__rte_experimental
> > +int
> > +rte_vhost_set_last_inflight_io_packed(int vid,
> > + uint16_t vring_idx, uint16_t head);
> > +
> > +/**
> > + * Clear the split inflight status.
> > + *
> > + * @param vid
> > + * vhost device ID
> > + * @param vring_idx
> > + * vring index
> > + * @param last_used_idx
> > + * last used idx of used ring
> > + * @param idx
> > + * inflight entry index
> > + * @return
> > + * 0 on success, -1 on failure
> > + */
> > +__rte_experimental
> > +int
> > +rte_vhost_clr_inflight_desc_split(int vid, uint16_t vring_idx,
> > + uint16_t last_used_idx, uint16_t idx);
> > +
> > +/**
> > + * Clear the packed inflight status.
> > + *
> > + * @param vid
> > + * vhost device ID
> > + * @param vring_idx
> > + * vring index
> > + * @param head
> > + * inflight entry index
> > + * @return
> > + * 0 on success, -1 on failure
> > + */
> > +__rte_experimental
> > +int
> > +rte_vhost_clr_inflight_desc_packed(int vid, uint16_t vring_idx,
> > + uint16_t head);
> > +
> > /**
> > * Notify the guest that used descriptors have been added to the vring.
> > This
> > * function acts as a memory barrier.
> > diff --git a/lib/librte_vhost/vhost.c b/lib/librte_vhost/vhost.c index
> > 660ac2a07..58940152f 100644
> > --- a/lib/librte_vhost/vhost.c
> > +++ b/lib/librte_vhost/vhost.c
> > @@ -783,6 +783,258 @@ rte_vhost_get_vhost_vring(int vid, uint16_t
> vring_idx,
> > return 0;
> > }
> >
> > +int
> > +rte_vhost_set_inflight_desc_split(int vid, uint16_t vring_idx,
> > + uint16_t idx)
> > +{
> > + struct virtio_net *dev;
> > + struct vhost_virtqueue *vq;
> > +
> > + dev = get_device(vid);
> > + if (unlikely(!dev))
> > + return -1;
> > +
> > + if (unlikely(!(dev->protocol_features &
> > + (1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD))))
> > + return 0;
> > +
> > + if (unlikely(vq_is_packed(dev)))
> > + return -1;
> > +
> > + if (unlikely(vring_idx >= VHOST_MAX_VRING))
> > + return -1;
> > +
> > + vq = dev->virtqueue[vring_idx];
> > + if (unlikely(!vq))
> > + return -1;
> > +
> > + if (unlikely(!vq->inflight_split))
> > + return -1;
> > +
> > + vq->inflight_split->desc[idx].counter = vq->global_counter++;
> > + vq->inflight_split->desc[idx].inflight = 1;
> > + return 0;
> > +}
> > +
> > +int
> > +rte_vhost_set_inflight_desc_packed(int vid, uint16_t vring_idx,
> > + uint16_t head, uint16_t last, uint16_t *inflight_entry) {
> > + struct virtio_net *dev;
> > + struct vhost_virtqueue *vq;
> > + struct inflight_info_packed *inflight_info;
> > + struct vring_packed_desc *desc;
> > + uint16_t old_free_head, free_head;
> > +
> > + dev = get_device(vid);
> > + if (unlikely(!dev))
> > + return -1;
> > +
> > + if (unlikely(!(dev->protocol_features &
> > + (1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD))))
> > + return 0;
> > +
> > + if (unlikely(!vq_is_packed(dev)))
> > + return -1;
> > +
> > + if (unlikely(vring_idx >= VHOST_MAX_VRING))
> > + return -1;
> > +
> > + vq = dev->virtqueue[vring_idx];
> > + if (unlikely(!vq))
> > + return -1;
> > +
> > + inflight_info = vq->inflight_packed;
> > + if (unlikely(!inflight_info))
> > + return -1;
> > +
> > + desc = vq->desc_packed;
> > + old_free_head = inflight_info->old_free_head;
> > + free_head = old_free_head;
> > +
> > + /* init header descriptor */
> > + inflight_info->desc[old_free_head].num = 0;
> > + inflight_info->desc[old_free_head].counter = vq->global_counter++;
> > + inflight_info->desc[old_free_head].inflight = 1;
>
> I think head has to be validated so that it does not causes out of bounds
> accesses.
Got it.
>
> > + /* save desc entry in flight entry */
> > + while (head != ((last + 1) % vq->size)) {
> > + inflight_info->desc[old_free_head].num++;
> > + inflight_info->desc[free_head].addr = desc[head].addr;
> > + inflight_info->desc[free_head].len = desc[head].len;
> > + inflight_info->desc[free_head].flags = desc[head].flags;
> > + inflight_info->desc[free_head].id = desc[head].id;
> > +
> > + inflight_info->desc[old_free_head].last = free_head;
> > + free_head = inflight_info->desc[free_head].next;
> > + inflight_info->free_head = free_head;
> > + head = (head + 1) % vq->size;
> > + }
> > +
> > + inflight_info->old_free_head = free_head;
> > + *inflight_entry = old_free_head;
> > +
> > + return 0;
> > +}
> > +
> > +int
> > +rte_vhost_clr_inflight_desc_split(int vid, uint16_t vring_idx,
> > + uint16_t last_used_idx, uint16_t idx) {
> > + struct virtio_net *dev;
> > + struct vhost_virtqueue *vq;
> > +
> > + dev = get_device(vid);
> > + if (unlikely(!dev))
> > + return -1;
> > +
> > + if (unlikely(!(dev->protocol_features &
> > + (1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD))))
> > + return 0;
> > +
> > + if (unlikely(vq_is_packed(dev)))
> > + return -1;
> > +
> > + if (unlikely(vring_idx >= VHOST_MAX_VRING))
> > + return -1;
> > +
> > + vq = dev->virtqueue[vring_idx];
> > + if (unlikely(!vq))
> > + return -1;
> > +
> > + if (unlikely(!vq->inflight_split))
> > + return -1;
> > +
> > + rte_compiler_barrier();
> > +
> > + vq->inflight_split->desc[idx].inflight = 0;
>
>
> Maybe it would be better to check idx value, so that it does not causes
> out-of-
> bound accesses.
>
Got it.
> > +
> > + rte_compiler_barrier();
> > +
> > + vq->inflight_split->used_idx = last_used_idx;
> > + return 0;
> > +}
> > +
> > +int
> > +rte_vhost_clr_inflight_desc_packed(int vid, uint16_t vring_idx,
> > + uint16_t head)
> > +{
> > + struct virtio_net *dev;
> > + struct vhost_virtqueue *vq;
> > + struct inflight_info_packed *inflight_info;
> > +
> > + dev = get_device(vid);
> > + if (unlikely(!dev))
> > + return -1;
> > +
> > + if (unlikely(!(dev->protocol_features &
> > + (1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD))))
> > + return 0;
> > +
> > + if (unlikely(!vq_is_packed(dev)))
> > + return -1;
> > +
> > + if (unlikely(vring_idx >= VHOST_MAX_VRING))
> > + return -1;
> > +
> > + vq = dev->virtqueue[vring_idx];
> > + if (unlikely(!vq))
> > + return -1;
> > +
> > + inflight_info = vq->inflight_packed;
> > + if (unlikely(!inflight_info))
> > + return -1;
> > +
> > + rte_compiler_barrier();
> > +
> > + inflight_info->desc[head].inflight = 0;
>
>
> Maybe it would be better to check head value, so that it does not causes
> out-of-
> bound accesses.
>
Got it.
>
> > + rte_compiler_barrier();
> > +
> > + inflight_info->old_free_head = inflight_info->free_head;
> > + inflight_info->old_used_idx = inflight_info->used_idx;
> > + inflight_info->old_used_wrap_counter =
> > +inflight_info->used_wrap_counter;
> > +
> > + return 0;
> > +}
> > +
> > +int
> > +rte_vhost_set_last_inflight_io_split(int vid, uint16_t vring_idx,
> > + uint16_t idx)
> > +{
> > + struct virtio_net *dev;
> > + struct vhost_virtqueue *vq;
> > +
> > + dev = get_device(vid);
> > + if (unlikely(!dev))
> > + return -1;
> > +
> > + if (unlikely(!(dev->protocol_features &
> > + (1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD))))
> > + return 0;
> > +
> > + if (unlikely(vq_is_packed(dev)))
> > + return -1;
> > +
> > + if (unlikely(vring_idx >= VHOST_MAX_VRING))
> > + return -1;
> > +
> > + vq = dev->virtqueue[vring_idx];
> > + if (unlikely(!vq))
> > + return -1;
> > +
> > + if (unlikely(!vq->inflight_split))
> > + return -1;
> > +
> > + vq->inflight_split->last_inflight_io = idx;
> > + return 0;
> > +}
> > +
> > +int
> > +rte_vhost_set_last_inflight_io_packed(int vid, uint16_t vring_idx,
> > + uint16_t head)
> > +{
> > + struct virtio_net *dev;
> > + struct vhost_virtqueue *vq;
> > + struct inflight_info_packed *inflight_info;
> > + uint16_t last;
> > +
> > + dev = get_device(vid);
> > + if (unlikely(!dev))
> > + return -1;
> > +
> > + if (unlikely(!(dev->protocol_features &
> > + (1ULL << VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD))))
> > + return 0;
> > +
> > + if (unlikely(!vq_is_packed(dev)))
> > + return -1;
> > +
> > + if (unlikely(vring_idx >= VHOST_MAX_VRING))
> > + return -1;
> > +
> > + vq = dev->virtqueue[vring_idx];
> > + if (unlikely(!vq))
> > + return -1;
> > +
> > + inflight_info = vq->inflight_packed;
> > + if (unlikely(!inflight_info))
> > + return -1;
> > +
> > + last = inflight_info->desc[head].last;
>
> Ditto
Got it.
> > + inflight_info->desc[last].next = inflight_info->free_head;
> > + inflight_info->free_head = head;
> > + inflight_info->used_idx += inflight_info->desc[head].num;
> > + if (inflight_info->used_idx >= inflight_info->desc_num) {
> > + inflight_info->used_idx -= inflight_info->desc_num;
> > + inflight_info->used_wrap_counter =
> > + !inflight_info->used_wrap_counter;
> > + }
> > +
> > + return 0;
> > +}
> > +
> > int
> > rte_vhost_vring_call(int vid, uint16_t vring_idx) {
> >
Thanks.
Jin
