On 2020/6/2 ??9:06, Michael S. Tsirkin wrote:> With this patch applied, new and old code perform identically. > > Lots of extra optimizations are now possible, e.g. > we can fetch multiple heads with copy_from/to_user now. > We can get rid of maintaining the log array. Etc etc. > > Signed-off-by: Michael S. Tsirkin <mst at redhat.com> > Signed-off-by: Eugenio P?rez <eperezma at redhat.com> > Link: https://lore.kernel.org/r/20200401183118.8334-4-eperezma at redhat.com > Signed-off-by: Michael S. Tsirkin <mst at redhat.com> > --- > drivers/vhost/test.c | 2 +- > drivers/vhost/vhost.c | 47 ++++++++++++++++++++++++++++++++++++++----- > drivers/vhost/vhost.h | 5 ++++- > 3 files changed, 47 insertions(+), 7 deletions(-) > > diff --git a/drivers/vhost/test.c b/drivers/vhost/test.c > index 9a3a09005e03..02806d6f84ef 100644 > --- a/drivers/vhost/test.c > +++ b/drivers/vhost/test.c > @@ -119,7 +119,7 @@ static int vhost_test_open(struct inode *inode, struct file *f) > dev = &n->dev; > vqs[VHOST_TEST_VQ] = &n->vqs[VHOST_TEST_VQ]; > n->vqs[VHOST_TEST_VQ].handle_kick = handle_vq_kick; > - vhost_dev_init(dev, vqs, VHOST_TEST_VQ_MAX, UIO_MAXIOV, > + vhost_dev_init(dev, vqs, VHOST_TEST_VQ_MAX, UIO_MAXIOV + 64, > VHOST_TEST_PKT_WEIGHT, VHOST_TEST_WEIGHT, NULL); > > f->private_data = n; > diff --git a/drivers/vhost/vhost.c b/drivers/vhost/vhost.c > index 8f9a07282625..aca2a5b0d078 100644 > --- a/drivers/vhost/vhost.c > +++ b/drivers/vhost/vhost.c > @@ -299,6 +299,7 @@ static void vhost_vq_reset(struct vhost_dev *dev, > { > vq->num = 1; > vq->ndescs = 0; > + vq->first_desc = 0; > vq->desc = NULL; > vq->avail = NULL; > vq->used = NULL; > @@ -367,6 +368,11 @@ static int vhost_worker(void *data) > return 0; > } > > +static int vhost_vq_num_batch_descs(struct vhost_virtqueue *vq) > +{ > + return vq->max_descs - UIO_MAXIOV; > +}1 descriptor does not mean 1 iov, e.g userspace may pass several 1 byte length memory regions for us to translate.> + > static void vhost_vq_free_iovecs(struct vhost_virtqueue *vq) > { > kfree(vq->descs); > @@ -389,6 +395,9 @@ static long vhost_dev_alloc_iovecs(struct vhost_dev *dev) > for (i = 0; i < dev->nvqs; ++i) { > vq = dev->vqs[i]; > vq->max_descs = dev->iov_limit; > + if (vhost_vq_num_batch_descs(vq) < 0) { > + return -EINVAL; > + } > vq->descs = kmalloc_array(vq->max_descs, > sizeof(*vq->descs), > GFP_KERNEL); > @@ -1570,6 +1579,7 @@ long vhost_vring_ioctl(struct vhost_dev *d, unsigned int ioctl, void __user *arg > vq->last_avail_idx = s.num; > /* Forget the cached index value. */ > vq->avail_idx = vq->last_avail_idx; > + vq->ndescs = vq->first_desc = 0; > break; > case VHOST_GET_VRING_BASE: > s.index = idx; > @@ -2136,7 +2146,7 @@ static int fetch_indirect_descs(struct vhost_virtqueue *vq, > return 0; > } > > -static int fetch_descs(struct vhost_virtqueue *vq) > +static int fetch_buf(struct vhost_virtqueue *vq) > { > unsigned int i, head, found = 0; > struct vhost_desc *last; > @@ -2149,7 +2159,11 @@ static int fetch_descs(struct vhost_virtqueue *vq) > /* Check it isn't doing very strange things with descriptor numbers. */ > last_avail_idx = vq->last_avail_idx; > > - if (vq->avail_idx == vq->last_avail_idx) { > + if (unlikely(vq->avail_idx == vq->last_avail_idx)) { > + /* If we already have work to do, don't bother re-checking. */ > + if (likely(vq->ndescs)) > + return vq->num; > + > if (unlikely(vhost_get_avail_idx(vq, &avail_idx))) { > vq_err(vq, "Failed to access avail idx at %p\n", > &vq->avail->idx); > @@ -2240,6 +2254,24 @@ static int fetch_descs(struct vhost_virtqueue *vq) > return 0; > } > > +static int fetch_descs(struct vhost_virtqueue *vq) > +{ > + int ret = 0; > + > + if (unlikely(vq->first_desc >= vq->ndescs)) { > + vq->first_desc = 0; > + vq->ndescs = 0; > + } > + > + if (vq->ndescs) > + return 0; > + > + while (!ret && vq->ndescs <= vhost_vq_num_batch_descs(vq)) > + ret = fetch_buf(vq); > + > + return vq->ndescs ? 0 : ret; > +} > + > /* This looks in the virtqueue and for the first available buffer, and converts > * it to an iovec for convenient access. Since descriptors consist of some > * number of output then some number of input descriptors, it's actually two > @@ -2265,7 +2297,7 @@ int vhost_get_vq_desc(struct vhost_virtqueue *vq, > if (unlikely(log)) > *log_num = 0; > > - for (i = 0; i < vq->ndescs; ++i) { > + for (i = vq->first_desc; i < vq->ndescs; ++i) { > unsigned iov_count = *in_num + *out_num; > struct vhost_desc *desc = &vq->descs[i]; > int access; > @@ -2311,14 +2343,19 @@ int vhost_get_vq_desc(struct vhost_virtqueue *vq, > } > > ret = desc->id; > + > + if (!(desc->flags & VRING_DESC_F_NEXT)) > + break; > } > > - vq->ndescs = 0; > + vq->first_desc = i + 1; > > return ret; > > err: > - vhost_discard_vq_desc(vq, 1); > + for (i = vq->first_desc; i < vq->ndescs; ++i) > + if (!(vq->descs[i].flags & VRING_DESC_F_NEXT)) > + vhost_discard_vq_desc(vq, 1); > vq->ndescs = 0; > > return ret; > diff --git a/drivers/vhost/vhost.h b/drivers/vhost/vhost.h > index 76356edee8e5..a67bda9792ec 100644 > --- a/drivers/vhost/vhost.h > +++ b/drivers/vhost/vhost.h > @@ -81,6 +81,7 @@ struct vhost_virtqueue { > > struct vhost_desc *descs; > int ndescs; > + int first_desc; > int max_descs; > > struct file *kick; > @@ -229,7 +230,7 @@ void vhost_iotlb_map_free(struct vhost_iotlb *iotlb, > struct vhost_iotlb_map *map); > > #define vq_err(vq, fmt, ...) do { \ > - pr_debug(pr_fmt(fmt), ##__VA_ARGS__); \ > + pr_err(pr_fmt(fmt), ##__VA_ARGS__); \Need a separate patch for this? Thanks> if ((vq)->error_ctx) \ > eventfd_signal((vq)->error_ctx, 1);\ > } while (0) > @@ -255,6 +256,8 @@ static inline void vhost_vq_set_backend(struct vhost_virtqueue *vq, > void *private_data) > { > vq->private_data = private_data; > + vq->ndescs = 0; > + vq->first_desc = 0; > } > > /**
On Wed, Jun 03, 2020 at 03:27:39PM +0800, Jason Wang wrote:> > On 2020/6/2 ??9:06, Michael S. Tsirkin wrote: > > With this patch applied, new and old code perform identically. > > > > Lots of extra optimizations are now possible, e.g. > > we can fetch multiple heads with copy_from/to_user now. > > We can get rid of maintaining the log array. Etc etc. > > > > Signed-off-by: Michael S. Tsirkin <mst at redhat.com> > > Signed-off-by: Eugenio P?rez <eperezma at redhat.com> > > Link: https://lore.kernel.org/r/20200401183118.8334-4-eperezma at redhat.com > > Signed-off-by: Michael S. Tsirkin <mst at redhat.com> > > --- > > drivers/vhost/test.c | 2 +- > > drivers/vhost/vhost.c | 47 ++++++++++++++++++++++++++++++++++++++----- > > drivers/vhost/vhost.h | 5 ++++- > > 3 files changed, 47 insertions(+), 7 deletions(-) > > > > diff --git a/drivers/vhost/test.c b/drivers/vhost/test.c > > index 9a3a09005e03..02806d6f84ef 100644 > > --- a/drivers/vhost/test.c > > +++ b/drivers/vhost/test.c > > @@ -119,7 +119,7 @@ static int vhost_test_open(struct inode *inode, struct file *f) > > dev = &n->dev; > > vqs[VHOST_TEST_VQ] = &n->vqs[VHOST_TEST_VQ]; > > n->vqs[VHOST_TEST_VQ].handle_kick = handle_vq_kick; > > - vhost_dev_init(dev, vqs, VHOST_TEST_VQ_MAX, UIO_MAXIOV, > > + vhost_dev_init(dev, vqs, VHOST_TEST_VQ_MAX, UIO_MAXIOV + 64, > > VHOST_TEST_PKT_WEIGHT, VHOST_TEST_WEIGHT, NULL); > > f->private_data = n; > > diff --git a/drivers/vhost/vhost.c b/drivers/vhost/vhost.c > > index 8f9a07282625..aca2a5b0d078 100644 > > --- a/drivers/vhost/vhost.c > > +++ b/drivers/vhost/vhost.c > > @@ -299,6 +299,7 @@ static void vhost_vq_reset(struct vhost_dev *dev, > > { > > vq->num = 1; > > vq->ndescs = 0; > > + vq->first_desc = 0; > > vq->desc = NULL; > > vq->avail = NULL; > > vq->used = NULL; > > @@ -367,6 +368,11 @@ static int vhost_worker(void *data) > > return 0; > > } > > +static int vhost_vq_num_batch_descs(struct vhost_virtqueue *vq) > > +{ > > + return vq->max_descs - UIO_MAXIOV; > > +} > > > 1 descriptor does not mean 1 iov, e.g userspace may pass several 1 byte > length memory regions for us to translate. >Yes but I don't see the relevance. This tells us how many descriptors to batch, not how many IOVs.> > + > > static void vhost_vq_free_iovecs(struct vhost_virtqueue *vq) > > { > > kfree(vq->descs); > > @@ -389,6 +395,9 @@ static long vhost_dev_alloc_iovecs(struct vhost_dev *dev) > > for (i = 0; i < dev->nvqs; ++i) { > > vq = dev->vqs[i]; > > vq->max_descs = dev->iov_limit; > > + if (vhost_vq_num_batch_descs(vq) < 0) { > > + return -EINVAL; > > + } > > vq->descs = kmalloc_array(vq->max_descs, > > sizeof(*vq->descs), > > GFP_KERNEL); > > @@ -1570,6 +1579,7 @@ long vhost_vring_ioctl(struct vhost_dev *d, unsigned int ioctl, void __user *arg > > vq->last_avail_idx = s.num; > > /* Forget the cached index value. */ > > vq->avail_idx = vq->last_avail_idx; > > + vq->ndescs = vq->first_desc = 0; > > break; > > case VHOST_GET_VRING_BASE: > > s.index = idx; > > @@ -2136,7 +2146,7 @@ static int fetch_indirect_descs(struct vhost_virtqueue *vq, > > return 0; > > } > > -static int fetch_descs(struct vhost_virtqueue *vq) > > +static int fetch_buf(struct vhost_virtqueue *vq) > > { > > unsigned int i, head, found = 0; > > struct vhost_desc *last; > > @@ -2149,7 +2159,11 @@ static int fetch_descs(struct vhost_virtqueue *vq) > > /* Check it isn't doing very strange things with descriptor numbers. */ > > last_avail_idx = vq->last_avail_idx; > > - if (vq->avail_idx == vq->last_avail_idx) { > > + if (unlikely(vq->avail_idx == vq->last_avail_idx)) { > > + /* If we already have work to do, don't bother re-checking. */ > > + if (likely(vq->ndescs)) > > + return vq->num; > > + > > if (unlikely(vhost_get_avail_idx(vq, &avail_idx))) { > > vq_err(vq, "Failed to access avail idx at %p\n", > > &vq->avail->idx); > > @@ -2240,6 +2254,24 @@ static int fetch_descs(struct vhost_virtqueue *vq) > > return 0; > > } > > +static int fetch_descs(struct vhost_virtqueue *vq) > > +{ > > + int ret = 0; > > + > > + if (unlikely(vq->first_desc >= vq->ndescs)) { > > + vq->first_desc = 0; > > + vq->ndescs = 0; > > + } > > + > > + if (vq->ndescs) > > + return 0; > > + > > + while (!ret && vq->ndescs <= vhost_vq_num_batch_descs(vq)) > > + ret = fetch_buf(vq); > > + > > + return vq->ndescs ? 0 : ret; > > +} > > + > > /* This looks in the virtqueue and for the first available buffer, and converts > > * it to an iovec for convenient access. Since descriptors consist of some > > * number of output then some number of input descriptors, it's actually two > > @@ -2265,7 +2297,7 @@ int vhost_get_vq_desc(struct vhost_virtqueue *vq, > > if (unlikely(log)) > > *log_num = 0; > > - for (i = 0; i < vq->ndescs; ++i) { > > + for (i = vq->first_desc; i < vq->ndescs; ++i) { > > unsigned iov_count = *in_num + *out_num; > > struct vhost_desc *desc = &vq->descs[i]; > > int access; > > @@ -2311,14 +2343,19 @@ int vhost_get_vq_desc(struct vhost_virtqueue *vq, > > } > > ret = desc->id; > > + > > + if (!(desc->flags & VRING_DESC_F_NEXT)) > > + break; > > } > > - vq->ndescs = 0; > > + vq->first_desc = i + 1; > > return ret; > > err: > > - vhost_discard_vq_desc(vq, 1); > > + for (i = vq->first_desc; i < vq->ndescs; ++i) > > + if (!(vq->descs[i].flags & VRING_DESC_F_NEXT)) > > + vhost_discard_vq_desc(vq, 1); > > vq->ndescs = 0; > > return ret; > > diff --git a/drivers/vhost/vhost.h b/drivers/vhost/vhost.h > > index 76356edee8e5..a67bda9792ec 100644 > > --- a/drivers/vhost/vhost.h > > +++ b/drivers/vhost/vhost.h > > @@ -81,6 +81,7 @@ struct vhost_virtqueue { > > struct vhost_desc *descs; > > int ndescs; > > + int first_desc; > > int max_descs; > > struct file *kick; > > @@ -229,7 +230,7 @@ void vhost_iotlb_map_free(struct vhost_iotlb *iotlb, > > struct vhost_iotlb_map *map); > > #define vq_err(vq, fmt, ...) do { \ > > - pr_debug(pr_fmt(fmt), ##__VA_ARGS__); \ > > + pr_err(pr_fmt(fmt), ##__VA_ARGS__); \ > > > Need a separate patch for this? > > ThanksOh that's a debugging thing. I will drop it.> > > if ((vq)->error_ctx) \ > > eventfd_signal((vq)->error_ctx, 1);\ > > } while (0) > > @@ -255,6 +256,8 @@ static inline void vhost_vq_set_backend(struct vhost_virtqueue *vq, > > void *private_data) > > { > > vq->private_data = private_data; > > + vq->ndescs = 0; > > + vq->first_desc = 0; > > } > > /**
On 2020/6/4 ??4:59, Michael S. Tsirkin wrote:> On Wed, Jun 03, 2020 at 03:27:39PM +0800, Jason Wang wrote: >> On 2020/6/2 ??9:06, Michael S. Tsirkin wrote: >>> With this patch applied, new and old code perform identically. >>> >>> Lots of extra optimizations are now possible, e.g. >>> we can fetch multiple heads with copy_from/to_user now. >>> We can get rid of maintaining the log array. Etc etc. >>> >>> Signed-off-by: Michael S. Tsirkin<mst at redhat.com> >>> Signed-off-by: Eugenio P?rez<eperezma at redhat.com> >>> Link:https://lore.kernel.org/r/20200401183118.8334-4-eperezma at redhat.com >>> Signed-off-by: Michael S. Tsirkin<mst at redhat.com> >>> --- >>> drivers/vhost/test.c | 2 +- >>> drivers/vhost/vhost.c | 47 ++++++++++++++++++++++++++++++++++++++----- >>> drivers/vhost/vhost.h | 5 ++++- >>> 3 files changed, 47 insertions(+), 7 deletions(-) >>> >>> diff --git a/drivers/vhost/test.c b/drivers/vhost/test.c >>> index 9a3a09005e03..02806d6f84ef 100644 >>> --- a/drivers/vhost/test.c >>> +++ b/drivers/vhost/test.c >>> @@ -119,7 +119,7 @@ static int vhost_test_open(struct inode *inode, struct file *f) >>> dev = &n->dev; >>> vqs[VHOST_TEST_VQ] = &n->vqs[VHOST_TEST_VQ]; >>> n->vqs[VHOST_TEST_VQ].handle_kick = handle_vq_kick; >>> - vhost_dev_init(dev, vqs, VHOST_TEST_VQ_MAX, UIO_MAXIOV, >>> + vhost_dev_init(dev, vqs, VHOST_TEST_VQ_MAX, UIO_MAXIOV + 64, >>> VHOST_TEST_PKT_WEIGHT, VHOST_TEST_WEIGHT, NULL); >>> f->private_data = n; >>> diff --git a/drivers/vhost/vhost.c b/drivers/vhost/vhost.c >>> index 8f9a07282625..aca2a5b0d078 100644 >>> --- a/drivers/vhost/vhost.c >>> +++ b/drivers/vhost/vhost.c >>> @@ -299,6 +299,7 @@ static void vhost_vq_reset(struct vhost_dev *dev, >>> { >>> vq->num = 1; >>> vq->ndescs = 0; >>> + vq->first_desc = 0; >>> vq->desc = NULL; >>> vq->avail = NULL; >>> vq->used = NULL; >>> @@ -367,6 +368,11 @@ static int vhost_worker(void *data) >>> return 0; >>> } >>> +static int vhost_vq_num_batch_descs(struct vhost_virtqueue *vq) >>> +{ >>> + return vq->max_descs - UIO_MAXIOV; >>> +} >> 1 descriptor does not mean 1 iov, e.g userspace may pass several 1 byte >> length memory regions for us to translate. >> > Yes but I don't see the relevance. This tells us how many descriptors to > batch, not how many IOVs.Yes, but questions are: - this introduce another obstacle to support more than 1K queue size - if we support 1K queue size, does it mean we need to cache 1K descriptors, which seems a large stress on the cache Thanks>