| Message ID | 20211028104919.3393-2-xuanzhuo@linux.alibaba.com (mailing list archive) |
|---|---|
| State | Superseded |
| Delegated to: | Netdev Maintainers |
| Headers | show |
| Series | virtio support cache indirect desc | expand |
| Context | Check | Description |
|---|---|---|
| netdev/tree_selection | success | Not a local patch |
On Thu, Oct 28, 2021 at 6:49 PM Xuan Zhuo <xuanzhuo@linux.alibaba.com> wrote: > > In the case of using indirect, indirect desc must be allocated and > released each time, which increases a lot of cpu overhead. > > Here, a cache is added for indirect. If the number of indirect desc to be > applied for is less than VIRT_QUEUE_CACHE_DESC_NUM, the desc array with > the size of VIRT_QUEUE_CACHE_DESC_NUM is fixed and cached for reuse. The commit log looks out of date? It's actually desc_cache_thr now. > > Signed-off-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com> > --- > drivers/virtio/virtio.c | 6 +++ > drivers/virtio/virtio_ring.c | 77 ++++++++++++++++++++++++++++++++---- > include/linux/virtio.h | 14 +++++++ > 3 files changed, 89 insertions(+), 8 deletions(-) > > diff --git a/drivers/virtio/virtio.c b/drivers/virtio/virtio.c > index 0a5b54034d4b..1047149ac2a4 100644 > --- a/drivers/virtio/virtio.c > +++ b/drivers/virtio/virtio.c > @@ -431,6 +431,12 @@ bool is_virtio_device(struct device *dev) > } > EXPORT_SYMBOL_GPL(is_virtio_device); > > +void virtio_set_desc_cache(struct virtio_device *dev, u32 thr) > +{ > + dev->desc_cache_thr = thr; > +} > +EXPORT_SYMBOL_GPL(virtio_set_desc_cache); > + > void unregister_virtio_device(struct virtio_device *dev) > { > int index = dev->index; /* save for after device release */ > diff --git a/drivers/virtio/virtio_ring.c b/drivers/virtio/virtio_ring.c > index dd95dfd85e98..0ebcd4f12d3b 100644 > --- a/drivers/virtio/virtio_ring.c > +++ b/drivers/virtio/virtio_ring.c > @@ -117,6 +117,15 @@ struct vring_virtqueue { > /* Hint for event idx: already triggered no need to disable. */ > bool event_triggered; > > + /* desc cache threshold > + * 0 - disable desc cache > + * > 0 - enable desc cache. As the threshold of the desc cache. > + */ > + u32 desc_cache_thr; > + > + /* desc cache chain */ > + struct list_head desc_cache; > + > union { > /* Available for split ring */ > struct { > @@ -423,7 +432,53 @@ static unsigned int vring_unmap_one_split(const struct vring_virtqueue *vq, > return extra[i].next; > } > > -static struct vring_desc *alloc_indirect_split(struct virtqueue *_vq, > +static void desc_cache_free(struct list_head *head) > +{ > + struct list_head *n, *pos; > + > + BUILD_BUG_ON(sizeof(struct list_head) > sizeof(struct vring_desc)); > + BUILD_BUG_ON(sizeof(struct list_head) > sizeof(struct vring_packed_desc)); > + > + list_for_each_prev_safe(pos, n, head) > + kfree(pos); > +} > + > +static void __desc_cache_put(struct vring_virtqueue *vq, > + struct list_head *node, int n) > +{ > + if (n <= vq->desc_cache_thr) > + list_add(node, &vq->desc_cache); > + else > + kfree(node); > +} > + > +#define desc_cache_put(vq, desc, n) \ > + __desc_cache_put(vq, (struct list_head *)desc, n) > + > +static void *desc_cache_get(struct vring_virtqueue *vq, > + int size, int n, gfp_t gfp) > +{ > + struct list_head *node; > + > + if (n > vq->desc_cache_thr) > + return kmalloc_array(n, size, gfp); > + > + if (!list_empty(&vq->desc_cache)) { > + node = vq->desc_cache.next; > + list_del(node); > + return node; > + } > + > + return kmalloc_array(vq->desc_cache_thr, size, gfp); Instead of doing the if .. kmalloc_array here. I wonder if we can simply pre-allocate per buffer indirect descriptors array. E.g if we use MAX_SKB_FRAGS + 1, it will be ~64KB for 256 queue size and ~272KB for 1024 queue size. Then we avoid list manipulation and kmalloc in datapath with much more simpler codes. Thanks > +} > + > +#define _desc_cache_get(vq, n, gfp, tp) \ > + ((tp *)desc_cache_get(vq, (sizeof(tp)), n, gfp)) > + > +#define desc_cache_get_split(vq, n, gfp) \ > + _desc_cache_get(vq, n, gfp, struct vring_desc) > + > +static struct vring_desc *alloc_indirect_split(struct vring_virtqueue *vq, > unsigned int total_sg, > gfp_t gfp) > { > @@ -437,12 +492,12 @@ static struct vring_desc *alloc_indirect_split(struct virtqueue *_vq, > */ > gfp &= ~__GFP_HIGHMEM; > > - desc = kmalloc_array(total_sg, sizeof(struct vring_desc), gfp); > + desc = desc_cache_get_split(vq, total_sg, gfp); > if (!desc) > return NULL; > > for (i = 0; i < total_sg; i++) > - desc[i].next = cpu_to_virtio16(_vq->vdev, i + 1); > + desc[i].next = cpu_to_virtio16(vq->vq.vdev, i + 1); > return desc; > } > > @@ -508,7 +563,7 @@ static inline int virtqueue_add_split(struct virtqueue *_vq, > head = vq->free_head; > > if (virtqueue_use_indirect(_vq, total_sg)) > - desc = alloc_indirect_split(_vq, total_sg, gfp); > + desc = alloc_indirect_split(vq, total_sg, gfp); > else { > desc = NULL; > WARN_ON_ONCE(total_sg > vq->split.vring.num && !vq->indirect); > @@ -652,7 +707,7 @@ static inline int virtqueue_add_split(struct virtqueue *_vq, > } > > if (indirect) > - kfree(desc); > + desc_cache_put(vq, desc, total_sg); > > END_USE(vq); > return -ENOMEM; > @@ -717,7 +772,7 @@ static void detach_buf_split(struct vring_virtqueue *vq, unsigned int head, > if (vq->indirect) { > struct vring_desc *indir_desc = > vq->split.desc_state[head].indir_desc; > - u32 len; > + u32 len, n; > > /* Free the indirect table, if any, now that it's unmapped. */ > if (!indir_desc) > @@ -729,10 +784,12 @@ static void detach_buf_split(struct vring_virtqueue *vq, unsigned int head, > VRING_DESC_F_INDIRECT)); > BUG_ON(len == 0 || len % sizeof(struct vring_desc)); > > - for (j = 0; j < len / sizeof(struct vring_desc); j++) > + n = len / sizeof(struct vring_desc); > + > + for (j = 0; j < n; j++) > vring_unmap_one_split_indirect(vq, &indir_desc[j]); > > - kfree(indir_desc); > + desc_cache_put(vq, indir_desc, n); > vq->split.desc_state[head].indir_desc = NULL; > } else if (ctx) { > *ctx = vq->split.desc_state[head].indir_desc; > @@ -2199,6 +2256,9 @@ struct virtqueue *__vring_new_virtqueue(unsigned int index, > vq->indirect = virtio_has_feature(vdev, VIRTIO_RING_F_INDIRECT_DESC) && > !context; > vq->event = virtio_has_feature(vdev, VIRTIO_RING_F_EVENT_IDX); > + vq->desc_cache_thr = vdev->desc_cache_thr; > + > + INIT_LIST_HEAD(&vq->desc_cache); > > if (virtio_has_feature(vdev, VIRTIO_F_ORDER_PLATFORM)) > vq->weak_barriers = false; > @@ -2329,6 +2389,7 @@ void vring_del_virtqueue(struct virtqueue *_vq) > if (!vq->packed_ring) { > kfree(vq->split.desc_state); > kfree(vq->split.desc_extra); > + desc_cache_free(&vq->desc_cache); > } > kfree(vq); > } > diff --git a/include/linux/virtio.h b/include/linux/virtio.h > index 41edbc01ffa4..bda6f9853e97 100644 > --- a/include/linux/virtio.h > +++ b/include/linux/virtio.h > @@ -118,6 +118,7 @@ struct virtio_device { > struct list_head vqs; > u64 features; > void *priv; > + u32 desc_cache_thr; > }; > > static inline struct virtio_device *dev_to_virtio(struct device *_dev) > @@ -130,6 +131,19 @@ int register_virtio_device(struct virtio_device *dev); > void unregister_virtio_device(struct virtio_device *dev); > bool is_virtio_device(struct device *dev); > > +/** > + * virtio_set_desc_cache - set virtio ring desc cache threshold > + * > + * virtio will cache the allocated indirect desc. > + * > + * This function must be called before find_vqs. > + * > + * @thr: > + * 0 - disable desc cache > + * > 0 - enable desc cache. As the threshold of the desc cache. > + */ > +void virtio_set_desc_cache(struct virtio_device *dev, u32 thr); > + > void virtio_break_device(struct virtio_device *dev); > > void virtio_config_changed(struct virtio_device *dev); > -- > 2.31.0 >
On Thu, Oct 28, 2021 at 06:49:17PM +0800, Xuan Zhuo wrote: > In the case of using indirect, indirect desc must be allocated and > released each time, which increases a lot of cpu overhead. > > Here, a cache is added for indirect. If the number of indirect desc to be > applied for is less than VIRT_QUEUE_CACHE_DESC_NUM, the desc array with > the size of VIRT_QUEUE_CACHE_DESC_NUM is fixed and cached for reuse. > > Signed-off-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com> What bothers me here is what happens if cache gets filled on one numa node, then used on another? > --- > drivers/virtio/virtio.c | 6 +++ > drivers/virtio/virtio_ring.c | 77 ++++++++++++++++++++++++++++++++---- > include/linux/virtio.h | 14 +++++++ > 3 files changed, 89 insertions(+), 8 deletions(-) > > diff --git a/drivers/virtio/virtio.c b/drivers/virtio/virtio.c > index 0a5b54034d4b..1047149ac2a4 100644 > --- a/drivers/virtio/virtio.c > +++ b/drivers/virtio/virtio.c > @@ -431,6 +431,12 @@ bool is_virtio_device(struct device *dev) > } > EXPORT_SYMBOL_GPL(is_virtio_device); > > +void virtio_set_desc_cache(struct virtio_device *dev, u32 thr) > +{ > + dev->desc_cache_thr = thr; > +} > +EXPORT_SYMBOL_GPL(virtio_set_desc_cache); > + > void unregister_virtio_device(struct virtio_device *dev) > { > int index = dev->index; /* save for after device release */ > diff --git a/drivers/virtio/virtio_ring.c b/drivers/virtio/virtio_ring.c > index dd95dfd85e98..0ebcd4f12d3b 100644 > --- a/drivers/virtio/virtio_ring.c > +++ b/drivers/virtio/virtio_ring.c > @@ -117,6 +117,15 @@ struct vring_virtqueue { > /* Hint for event idx: already triggered no need to disable. */ > bool event_triggered; > > + /* desc cache threshold > + * 0 - disable desc cache > + * > 0 - enable desc cache. As the threshold of the desc cache. > + */ > + u32 desc_cache_thr; not really descriptive. also pls eschew abbreviation. > + > + /* desc cache chain */ > + struct list_head desc_cache; hmm this puts extra pressure on cache. you never need to drop things in the middle. llist_head would be better I think ... no? > + > union { > /* Available for split ring */ > struct { > @@ -423,7 +432,53 @@ static unsigned int vring_unmap_one_split(const struct vring_virtqueue *vq, > return extra[i].next; > } > > -static struct vring_desc *alloc_indirect_split(struct virtqueue *_vq, > +static void desc_cache_free(struct list_head *head) > +{ > + struct list_head *n, *pos; > + > + BUILD_BUG_ON(sizeof(struct list_head) > sizeof(struct vring_desc)); > + BUILD_BUG_ON(sizeof(struct list_head) > sizeof(struct vring_packed_desc)); > + > + list_for_each_prev_safe(pos, n, head) > + kfree(pos); > +} > + > +static void __desc_cache_put(struct vring_virtqueue *vq, > + struct list_head *node, int n) > +{ > + if (n <= vq->desc_cache_thr) > + list_add(node, &vq->desc_cache); > + else > + kfree(node); this bothers me. Do we really need a full VQ's worth of indirect descriptors? Can't we set a limit on how many are used? > +} > + > +#define desc_cache_put(vq, desc, n) \ > + __desc_cache_put(vq, (struct list_head *)desc, n) replace with an inline function pls. in fact we dont need __desc_cache_put at all. > + > +static void *desc_cache_get(struct vring_virtqueue *vq, > + int size, int n, gfp_t gfp) > +{ > + struct list_head *node; > + > + if (n > vq->desc_cache_thr) > + return kmalloc_array(n, size, gfp); > + > + if (!list_empty(&vq->desc_cache)) { > + node = vq->desc_cache.next; > + list_del(node); > + return node; > + } > + > + return kmalloc_array(vq->desc_cache_thr, size, gfp); > +} > + > +#define _desc_cache_get(vq, n, gfp, tp) \ > + ((tp *)desc_cache_get(vq, (sizeof(tp)), n, gfp)) > + > +#define desc_cache_get_split(vq, n, gfp) \ > + _desc_cache_get(vq, n, gfp, struct vring_desc) > + same thing here. > +static struct vring_desc *alloc_indirect_split(struct vring_virtqueue *vq, > unsigned int total_sg, > gfp_t gfp) > { > @@ -437,12 +492,12 @@ static struct vring_desc *alloc_indirect_split(struct virtqueue *_vq, > */ > gfp &= ~__GFP_HIGHMEM; > > - desc = kmalloc_array(total_sg, sizeof(struct vring_desc), gfp); > + desc = desc_cache_get_split(vq, total_sg, gfp); > if (!desc) > return NULL; > > for (i = 0; i < total_sg; i++) > - desc[i].next = cpu_to_virtio16(_vq->vdev, i + 1); > + desc[i].next = cpu_to_virtio16(vq->vq.vdev, i + 1); > return desc; > } > > @@ -508,7 +563,7 @@ static inline int virtqueue_add_split(struct virtqueue *_vq, > head = vq->free_head; > > if (virtqueue_use_indirect(_vq, total_sg)) > - desc = alloc_indirect_split(_vq, total_sg, gfp); > + desc = alloc_indirect_split(vq, total_sg, gfp); > else { > desc = NULL; > WARN_ON_ONCE(total_sg > vq->split.vring.num && !vq->indirect); > @@ -652,7 +707,7 @@ static inline int virtqueue_add_split(struct virtqueue *_vq, > } > > if (indirect) > - kfree(desc); > + desc_cache_put(vq, desc, total_sg); > > END_USE(vq); > return -ENOMEM; > @@ -717,7 +772,7 @@ static void detach_buf_split(struct vring_virtqueue *vq, unsigned int head, > if (vq->indirect) { > struct vring_desc *indir_desc = > vq->split.desc_state[head].indir_desc; > - u32 len; > + u32 len, n; > > /* Free the indirect table, if any, now that it's unmapped. */ > if (!indir_desc) > @@ -729,10 +784,12 @@ static void detach_buf_split(struct vring_virtqueue *vq, unsigned int head, > VRING_DESC_F_INDIRECT)); > BUG_ON(len == 0 || len % sizeof(struct vring_desc)); > > - for (j = 0; j < len / sizeof(struct vring_desc); j++) > + n = len / sizeof(struct vring_desc); > + > + for (j = 0; j < n; j++) > vring_unmap_one_split_indirect(vq, &indir_desc[j]); > > - kfree(indir_desc); > + desc_cache_put(vq, indir_desc, n); > vq->split.desc_state[head].indir_desc = NULL; > } else if (ctx) { > *ctx = vq->split.desc_state[head].indir_desc; > @@ -2199,6 +2256,9 @@ struct virtqueue *__vring_new_virtqueue(unsigned int index, > vq->indirect = virtio_has_feature(vdev, VIRTIO_RING_F_INDIRECT_DESC) && > !context; > vq->event = virtio_has_feature(vdev, VIRTIO_RING_F_EVENT_IDX); > + vq->desc_cache_thr = vdev->desc_cache_thr; > + > + INIT_LIST_HEAD(&vq->desc_cache); > > if (virtio_has_feature(vdev, VIRTIO_F_ORDER_PLATFORM)) > vq->weak_barriers = false; So e.g. for rx, we are wasting memory since indirect isn't used. > @@ -2329,6 +2389,7 @@ void vring_del_virtqueue(struct virtqueue *_vq) > if (!vq->packed_ring) { > kfree(vq->split.desc_state); > kfree(vq->split.desc_extra); > + desc_cache_free(&vq->desc_cache); > } > kfree(vq); > } > diff --git a/include/linux/virtio.h b/include/linux/virtio.h > index 41edbc01ffa4..bda6f9853e97 100644 > --- a/include/linux/virtio.h > +++ b/include/linux/virtio.h > @@ -118,6 +118,7 @@ struct virtio_device { > struct list_head vqs; > u64 features; > void *priv; > + u32 desc_cache_thr; > }; > > static inline struct virtio_device *dev_to_virtio(struct device *_dev) > @@ -130,6 +131,19 @@ int register_virtio_device(struct virtio_device *dev); > void unregister_virtio_device(struct virtio_device *dev); > bool is_virtio_device(struct device *dev); > > +/** > + * virtio_set_desc_cache - set virtio ring desc cache threshold > + * > + * virtio will cache the allocated indirect desc. > + * > + * This function must be called before find_vqs. > + * > + * @thr: > + * 0 - disable desc cache > + * > 0 - enable desc cache. As the threshold of the desc cache. > + */ > +void virtio_set_desc_cache(struct virtio_device *dev, u32 thr); > + > void virtio_break_device(struct virtio_device *dev); > > void virtio_config_changed(struct virtio_device *dev); > -- > 2.31.0
On Thu, Oct 28, 2021 at 06:49:17PM +0800, Xuan Zhuo wrote: > In the case of using indirect, indirect desc must be allocated and > released each time, which increases a lot of cpu overhead. > > Here, a cache is added for indirect. If the number of indirect desc to be > applied for is less than VIRT_QUEUE_CACHE_DESC_NUM, the desc array with > the size of VIRT_QUEUE_CACHE_DESC_NUM is fixed and cached for reuse. > > Signed-off-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com> > --- > drivers/virtio/virtio.c | 6 +++ > drivers/virtio/virtio_ring.c | 77 ++++++++++++++++++++++++++++++++---- > include/linux/virtio.h | 14 +++++++ > 3 files changed, 89 insertions(+), 8 deletions(-) > > diff --git a/drivers/virtio/virtio.c b/drivers/virtio/virtio.c > index 0a5b54034d4b..1047149ac2a4 100644 > --- a/drivers/virtio/virtio.c > +++ b/drivers/virtio/virtio.c > @@ -431,6 +431,12 @@ bool is_virtio_device(struct device *dev) > } > EXPORT_SYMBOL_GPL(is_virtio_device); > > +void virtio_set_desc_cache(struct virtio_device *dev, u32 thr) > +{ > + dev->desc_cache_thr = thr; > +} > +EXPORT_SYMBOL_GPL(virtio_set_desc_cache); > + > void unregister_virtio_device(struct virtio_device *dev) > { > int index = dev->index; /* save for after device release */ > diff --git a/drivers/virtio/virtio_ring.c b/drivers/virtio/virtio_ring.c > index dd95dfd85e98..0ebcd4f12d3b 100644 > --- a/drivers/virtio/virtio_ring.c > +++ b/drivers/virtio/virtio_ring.c > @@ -117,6 +117,15 @@ struct vring_virtqueue { > /* Hint for event idx: already triggered no need to disable. */ > bool event_triggered; > > + /* desc cache threshold > + * 0 - disable desc cache > + * > 0 - enable desc cache. As the threshold of the desc cache. > + */ > + u32 desc_cache_thr; > + > + /* desc cache chain */ > + struct list_head desc_cache; > + > union { > /* Available for split ring */ > struct { > @@ -423,7 +432,53 @@ static unsigned int vring_unmap_one_split(const struct vring_virtqueue *vq, > return extra[i].next; > } > > -static struct vring_desc *alloc_indirect_split(struct virtqueue *_vq, > +static void desc_cache_free(struct list_head *head) > +{ > + struct list_head *n, *pos; > + > + BUILD_BUG_ON(sizeof(struct list_head) > sizeof(struct vring_desc)); > + BUILD_BUG_ON(sizeof(struct list_head) > sizeof(struct vring_packed_desc)); > + > + list_for_each_prev_safe(pos, n, head) > + kfree(pos); > +} > + > +static void __desc_cache_put(struct vring_virtqueue *vq, > + struct list_head *node, int n) > +{ > + if (n <= vq->desc_cache_thr) > + list_add(node, &vq->desc_cache); > + else > + kfree(node); > +} > + > +#define desc_cache_put(vq, desc, n) \ > + __desc_cache_put(vq, (struct list_head *)desc, n) > + > +static void *desc_cache_get(struct vring_virtqueue *vq, > + int size, int n, gfp_t gfp) > +{ > + struct list_head *node; > + > + if (n > vq->desc_cache_thr) > + return kmalloc_array(n, size, gfp); > + > + if (!list_empty(&vq->desc_cache)) { > + node = vq->desc_cache.next; > + list_del(node); > + return node; > + } > + > + return kmalloc_array(vq->desc_cache_thr, size, gfp); > +} > + > +#define _desc_cache_get(vq, n, gfp, tp) \ > + ((tp *)desc_cache_get(vq, (sizeof(tp)), n, gfp)) > + > +#define desc_cache_get_split(vq, n, gfp) \ > + _desc_cache_get(vq, n, gfp, struct vring_desc) > + > +static struct vring_desc *alloc_indirect_split(struct vring_virtqueue *vq, > unsigned int total_sg, > gfp_t gfp) > { > @@ -437,12 +492,12 @@ static struct vring_desc *alloc_indirect_split(struct virtqueue *_vq, > */ > gfp &= ~__GFP_HIGHMEM; > > - desc = kmalloc_array(total_sg, sizeof(struct vring_desc), gfp); > + desc = desc_cache_get_split(vq, total_sg, gfp); > if (!desc) > return NULL; > > for (i = 0; i < total_sg; i++) > - desc[i].next = cpu_to_virtio16(_vq->vdev, i + 1); > + desc[i].next = cpu_to_virtio16(vq->vq.vdev, i + 1); > return desc; > } > > @@ -508,7 +563,7 @@ static inline int virtqueue_add_split(struct virtqueue *_vq, > head = vq->free_head; > > if (virtqueue_use_indirect(_vq, total_sg)) > - desc = alloc_indirect_split(_vq, total_sg, gfp); > + desc = alloc_indirect_split(vq, total_sg, gfp); > else { > desc = NULL; > WARN_ON_ONCE(total_sg > vq->split.vring.num && !vq->indirect); > @@ -652,7 +707,7 @@ static inline int virtqueue_add_split(struct virtqueue *_vq, > } > > if (indirect) > - kfree(desc); > + desc_cache_put(vq, desc, total_sg); > > END_USE(vq); > return -ENOMEM; > @@ -717,7 +772,7 @@ static void detach_buf_split(struct vring_virtqueue *vq, unsigned int head, > if (vq->indirect) { > struct vring_desc *indir_desc = > vq->split.desc_state[head].indir_desc; > - u32 len; > + u32 len, n; > > /* Free the indirect table, if any, now that it's unmapped. */ > if (!indir_desc) > @@ -729,10 +784,12 @@ static void detach_buf_split(struct vring_virtqueue *vq, unsigned int head, > VRING_DESC_F_INDIRECT)); > BUG_ON(len == 0 || len % sizeof(struct vring_desc)); > > - for (j = 0; j < len / sizeof(struct vring_desc); j++) > + n = len / sizeof(struct vring_desc); > + > + for (j = 0; j < n; j++) > vring_unmap_one_split_indirect(vq, &indir_desc[j]); > > - kfree(indir_desc); > + desc_cache_put(vq, indir_desc, n); > vq->split.desc_state[head].indir_desc = NULL; > } else if (ctx) { > *ctx = vq->split.desc_state[head].indir_desc; > @@ -2199,6 +2256,9 @@ struct virtqueue *__vring_new_virtqueue(unsigned int index, > vq->indirect = virtio_has_feature(vdev, VIRTIO_RING_F_INDIRECT_DESC) && > !context; > vq->event = virtio_has_feature(vdev, VIRTIO_RING_F_EVENT_IDX); > + vq->desc_cache_thr = vdev->desc_cache_thr; > + > + INIT_LIST_HEAD(&vq->desc_cache); > > if (virtio_has_feature(vdev, VIRTIO_F_ORDER_PLATFORM)) > vq->weak_barriers = false; > @@ -2329,6 +2389,7 @@ void vring_del_virtqueue(struct virtqueue *_vq) > if (!vq->packed_ring) { > kfree(vq->split.desc_state); > kfree(vq->split.desc_extra); > + desc_cache_free(&vq->desc_cache); > } > kfree(vq); > } > diff --git a/include/linux/virtio.h b/include/linux/virtio.h > index 41edbc01ffa4..bda6f9853e97 100644 > --- a/include/linux/virtio.h > +++ b/include/linux/virtio.h > @@ -118,6 +118,7 @@ struct virtio_device { > struct list_head vqs; > u64 features; > void *priv; > + u32 desc_cache_thr; > }; > > static inline struct virtio_device *dev_to_virtio(struct device *_dev) > @@ -130,6 +131,19 @@ int register_virtio_device(struct virtio_device *dev); > void unregister_virtio_device(struct virtio_device *dev); > bool is_virtio_device(struct device *dev); > > +/** > + * virtio_set_desc_cache - set virtio ring desc cache threshold > + * > + * virtio will cache the allocated indirect desc. > + * > + * This function must be called before find_vqs. > + * > + * @thr: > + * 0 - disable desc cache > + * > 0 - enable desc cache. As the threshold of the desc cache. > + */ > +void virtio_set_desc_cache(struct virtio_device *dev, u32 thr); > + Would a better API be a cache size in bytes? This controls how much memory is spent after all. > void virtio_break_device(struct virtio_device *dev); > > void virtio_config_changed(struct virtio_device *dev); > -- > 2.31.0
diff --git a/drivers/virtio/virtio.c b/drivers/virtio/virtio.c index 0a5b54034d4b..1047149ac2a4 100644 --- a/drivers/virtio/virtio.c +++ b/drivers/virtio/virtio.c @@ -431,6 +431,12 @@ bool is_virtio_device(struct device *dev) } EXPORT_SYMBOL_GPL(is_virtio_device); +void virtio_set_desc_cache(struct virtio_device *dev, u32 thr) +{ + dev->desc_cache_thr = thr; +} +EXPORT_SYMBOL_GPL(virtio_set_desc_cache); + void unregister_virtio_device(struct virtio_device *dev) { int index = dev->index; /* save for after device release */ diff --git a/drivers/virtio/virtio_ring.c b/drivers/virtio/virtio_ring.c index dd95dfd85e98..0ebcd4f12d3b 100644 --- a/drivers/virtio/virtio_ring.c +++ b/drivers/virtio/virtio_ring.c @@ -117,6 +117,15 @@ struct vring_virtqueue { /* Hint for event idx: already triggered no need to disable. */ bool event_triggered; + /* desc cache threshold + * 0 - disable desc cache + * > 0 - enable desc cache. As the threshold of the desc cache. + */ + u32 desc_cache_thr; + + /* desc cache chain */ + struct list_head desc_cache; + union { /* Available for split ring */ struct { @@ -423,7 +432,53 @@ static unsigned int vring_unmap_one_split(const struct vring_virtqueue *vq, return extra[i].next; } -static struct vring_desc *alloc_indirect_split(struct virtqueue *_vq, +static void desc_cache_free(struct list_head *head) +{ + struct list_head *n, *pos; + + BUILD_BUG_ON(sizeof(struct list_head) > sizeof(struct vring_desc)); + BUILD_BUG_ON(sizeof(struct list_head) > sizeof(struct vring_packed_desc)); + + list_for_each_prev_safe(pos, n, head) + kfree(pos); +} + +static void __desc_cache_put(struct vring_virtqueue *vq, + struct list_head *node, int n) +{ + if (n <= vq->desc_cache_thr) + list_add(node, &vq->desc_cache); + else + kfree(node); +} + +#define desc_cache_put(vq, desc, n) \ + __desc_cache_put(vq, (struct list_head *)desc, n) + +static void *desc_cache_get(struct vring_virtqueue *vq, + int size, int n, gfp_t gfp) +{ + struct list_head *node; + + if (n > vq->desc_cache_thr) + return kmalloc_array(n, size, gfp); + + if (!list_empty(&vq->desc_cache)) { + node = vq->desc_cache.next; + list_del(node); + return node; + } + + return kmalloc_array(vq->desc_cache_thr, size, gfp); +} + +#define _desc_cache_get(vq, n, gfp, tp) \ + ((tp *)desc_cache_get(vq, (sizeof(tp)), n, gfp)) + +#define desc_cache_get_split(vq, n, gfp) \ + _desc_cache_get(vq, n, gfp, struct vring_desc) + +static struct vring_desc *alloc_indirect_split(struct vring_virtqueue *vq, unsigned int total_sg, gfp_t gfp) { @@ -437,12 +492,12 @@ static struct vring_desc *alloc_indirect_split(struct virtqueue *_vq, */ gfp &= ~__GFP_HIGHMEM; - desc = kmalloc_array(total_sg, sizeof(struct vring_desc), gfp); + desc = desc_cache_get_split(vq, total_sg, gfp); if (!desc) return NULL; for (i = 0; i < total_sg; i++) - desc[i].next = cpu_to_virtio16(_vq->vdev, i + 1); + desc[i].next = cpu_to_virtio16(vq->vq.vdev, i + 1); return desc; } @@ -508,7 +563,7 @@ static inline int virtqueue_add_split(struct virtqueue *_vq, head = vq->free_head; if (virtqueue_use_indirect(_vq, total_sg)) - desc = alloc_indirect_split(_vq, total_sg, gfp); + desc = alloc_indirect_split(vq, total_sg, gfp); else { desc = NULL; WARN_ON_ONCE(total_sg > vq->split.vring.num && !vq->indirect); @@ -652,7 +707,7 @@ static inline int virtqueue_add_split(struct virtqueue *_vq, } if (indirect) - kfree(desc); + desc_cache_put(vq, desc, total_sg); END_USE(vq); return -ENOMEM; @@ -717,7 +772,7 @@ static void detach_buf_split(struct vring_virtqueue *vq, unsigned int head, if (vq->indirect) { struct vring_desc *indir_desc = vq->split.desc_state[head].indir_desc; - u32 len; + u32 len, n; /* Free the indirect table, if any, now that it's unmapped. */ if (!indir_desc) @@ -729,10 +784,12 @@ static void detach_buf_split(struct vring_virtqueue *vq, unsigned int head, VRING_DESC_F_INDIRECT)); BUG_ON(len == 0 || len % sizeof(struct vring_desc)); - for (j = 0; j < len / sizeof(struct vring_desc); j++) + n = len / sizeof(struct vring_desc); + + for (j = 0; j < n; j++) vring_unmap_one_split_indirect(vq, &indir_desc[j]); - kfree(indir_desc); + desc_cache_put(vq, indir_desc, n); vq->split.desc_state[head].indir_desc = NULL; } else if (ctx) { *ctx = vq->split.desc_state[head].indir_desc; @@ -2199,6 +2256,9 @@ struct virtqueue *__vring_new_virtqueue(unsigned int index, vq->indirect = virtio_has_feature(vdev, VIRTIO_RING_F_INDIRECT_DESC) && !context; vq->event = virtio_has_feature(vdev, VIRTIO_RING_F_EVENT_IDX); + vq->desc_cache_thr = vdev->desc_cache_thr; + + INIT_LIST_HEAD(&vq->desc_cache); if (virtio_has_feature(vdev, VIRTIO_F_ORDER_PLATFORM)) vq->weak_barriers = false; @@ -2329,6 +2389,7 @@ void vring_del_virtqueue(struct virtqueue *_vq) if (!vq->packed_ring) { kfree(vq->split.desc_state); kfree(vq->split.desc_extra); + desc_cache_free(&vq->desc_cache); } kfree(vq); } diff --git a/include/linux/virtio.h b/include/linux/virtio.h index 41edbc01ffa4..bda6f9853e97 100644 --- a/include/linux/virtio.h +++ b/include/linux/virtio.h @@ -118,6 +118,7 @@ struct virtio_device { struct list_head vqs; u64 features; void *priv; + u32 desc_cache_thr; }; static inline struct virtio_device *dev_to_virtio(struct device *_dev) @@ -130,6 +131,19 @@ int register_virtio_device(struct virtio_device *dev); void unregister_virtio_device(struct virtio_device *dev); bool is_virtio_device(struct device *dev); +/** + * virtio_set_desc_cache - set virtio ring desc cache threshold + * + * virtio will cache the allocated indirect desc. + * + * This function must be called before find_vqs. + * + * @thr: + * 0 - disable desc cache + * > 0 - enable desc cache. As the threshold of the desc cache. + */ +void virtio_set_desc_cache(struct virtio_device *dev, u32 thr); + void virtio_break_device(struct virtio_device *dev); void virtio_config_changed(struct virtio_device *dev);
In the case of using indirect, indirect desc must be allocated and released each time, which increases a lot of cpu overhead. Here, a cache is added for indirect. If the number of indirect desc to be applied for is less than VIRT_QUEUE_CACHE_DESC_NUM, the desc array with the size of VIRT_QUEUE_CACHE_DESC_NUM is fixed and cached for reuse. Signed-off-by: Xuan Zhuo <xuanzhuo@linux.alibaba.com> --- drivers/virtio/virtio.c | 6 +++ drivers/virtio/virtio_ring.c | 77 ++++++++++++++++++++++++++++++++---- include/linux/virtio.h | 14 +++++++ 3 files changed, 89 insertions(+), 8 deletions(-)