On Thu, Mar 7, 2024 at 4:15 PM Xuan Zhuo <xuanzhuo@xxxxxxxxxxxxxxxxx> wrote: > > On Thu, 7 Mar 2024 13:28:27 +0800, Jason Wang <jasowang@xxxxxxxxxx> wrote: > > On Wed, Mar 6, 2024 at 6:01 PM Xuan Zhuo <xuanzhuo@xxxxxxxxxxxxxxxxx> wrote: > > > > > > On Thu, 29 Feb 2024 04:34:20 -0500, "Michael S. Tsirkin" <mst@xxxxxxxxxx> wrote: > > > > On Thu, Feb 29, 2024 at 05:02:37PM +0800, Xuan Zhuo wrote: > > > > > On Thu, 29 Feb 2024 03:21:14 -0500, "Michael S. Tsirkin" <mst@xxxxxxxxxx> wrote: > > > > > > On Thu, Feb 29, 2024 at 03:20:25PM +0800, Xuan Zhuo wrote: > > > > > > > As discussed: > > > > > > > http://lore.kernel.org/all/CACGkMEvq0No8QGC46U4mGsMtuD44fD_cfLcPaVmJ3rHYqRZxYg@xxxxxxxxxxxxxx > > > > > > > > > > > > > > If the virtio is premapped mode, the driver should manage the dma info by self. > > > > > > > So the virtio core should not store the dma info. > > > > > > > So we can release the memory used to store the dma info. > > > > > > > > > > > > > > But if the desc_extra has not dma info, we face a new question, > > > > > > > it is hard to get the dma info of the desc with indirect flag. > > > > > > > For split mode, that is easy from desc, but for the packed mode, > > > > > > > it is hard to get the dma info from the desc. And for hardening > > > > > > > the dma unmap is saft, we should store the dma info of indirect > > > > > > > descs. > > > > > > > > > > > > > > So I introduce the "structure the indirect desc table" to > > > > > > > allocate space to store dma info with the desc table. > > > > > > > > > > > > > > On the other side, we mix the descs with indirect flag > > > > > > > with other descs together to share the unmap api. That > > > > > > > is complex. I found if we we distinguish the descs with > > > > > > > VRING_DESC_F_INDIRECT before unmap, thing will be clearer. > > > > > > > > > > > > > > Because of the dma array is allocated in the find_vqs(), > > > > > > > so I introduce a new parameter to find_vqs(). > > > > > > > > > > > > > > Note: > > > > > > > this is on the top of > > > > > > > [PATCH vhost v1] virtio: packed: fix unmap leak for indirect desc table > > > > > > > http://lore.kernel.org/all/20240223071833.26095-1-xuanzhuo@xxxxxxxxxxxxxxxxx > > > > > > > > > > > > > > Please review. > > > > > > > > > > > > > > Thanks > > > > > > > > > > > > > > v3: > > > > > > > 1. fix the conflict with the vp_modern_create_avq(). > > > > > > > > > > > > Okay but are you going to address huge memory waste all this is causing for > > > > > > - people who never do zero copy > > > > > > - systems where dma unmap is a nop > > > > > > > > > > > > ? > > > > > > > > > > > > You should address all comments when you post a new version, not just > > > > > > what was expedient, or alternatively tag patch as RFC and explain > > > > > > in commit log that you plan to do it later. > > > > > > > > > > > > > > > Do you miss this one? > > > > > http://lore.kernel.org/all/1708997579.5613105-1-xuanzhuo@xxxxxxxxxxxxxxxxx > > > > > > > > > > > > I did. The answer is that no, you don't get to regress memory usage > > > > for lots of people then fix it up. > > > > So the patchset is big, I guess it will take a couple of cycles to > > > > merge gradually. > > > > > > Hi @Michael > > > > > > So, how about this patch set? > > > > > > I do not think they (dma maintainers) will agree the API dma_can_skip_unmap(). > > > > > > If you think sq wastes too much memory using pre-mapped dma mode, how about > > > we only enable it when xsk is bond? > > > > > > Could you give me some advice? > > > > I think we have some discussion, one possible solution is: > > > > when pre mapping is enabled, virtio core won't store dma metadatas. > > > > Then it makes virtio-net align with other NIC. > > > YES. > > This patch set works as this. > > But the virtio-net must allocate too much memory to store dma and len. > > num = queue size * 19 > > Michael thinks that waste too much memory. > http://lore.kernel.org/all/20240225032330-mutt-send-email-mst@xxxxxxxxxx > > So we try this: > http://lore.kernel.org/all/20240301071918.64631-1-xuanzhuo@xxxxxxxxxxxxxxxxx > > But I think that is difficult to be accepted by the DMA maintainers. > > So I have two advices: > > 1. virtio-net sq works without indirect. > - that more like other NIC > - the num of the memory to store the dma info is queue_size This requires benchmarks. > > 2. The default mode of virtio-net sq is no-premapped > - we just switch the mode when binding xsk This could be one step. We can hear from Michael. Thanks > > Thanks. > > > > > > Thanks > > > > > > > > Thanks. > > > > > > > > > > > > > > > I asked you. But I didnot recv your answer. > > > > > > > > > > Thanks. > > > > > > > > > > > > > > > > > > > > > > > v2: > > > > > > > 1. change the dma item of virtio-net, every item have MAX_SKB_FRAGS + 2 > > > > > > > addr + len pairs. > > > > > > > 2. introduce virtnet_sq_free_stats for __free_old_xmit > > > > > > > > > > > > > > v1: > > > > > > > 1. rename transport_vq_config to vq_transport_config > > > > > > > 2. virtio-net set dma meta number to (ring-size + 1)(MAX_SKB_FRGAS +2) > > > > > > > 3. introduce virtqueue_dma_map_sg_attrs > > > > > > > 4. separate vring_create_virtqueue to an independent commit > > > > > > > > > > > > > > > > > > > > > > > > > > > > Xuan Zhuo (19): > > > > > > > virtio_ring: introduce vring_need_unmap_buffer > > > > > > > virtio_ring: packed: remove double check of the unmap ops > > > > > > > virtio_ring: packed: structure the indirect desc table > > > > > > > virtio_ring: split: remove double check of the unmap ops > > > > > > > virtio_ring: split: structure the indirect desc table > > > > > > > virtio_ring: no store dma info when unmap is not needed > > > > > > > virtio: find_vqs: pass struct instead of multi parameters > > > > > > > virtio: vring_create_virtqueue: pass struct instead of multi > > > > > > > parameters > > > > > > > virtio: vring_new_virtqueue(): pass struct instead of multi parameters > > > > > > > virtio_ring: simplify the parameters of the funcs related to > > > > > > > vring_create/new_virtqueue() > > > > > > > virtio: find_vqs: add new parameter premapped > > > > > > > virtio_ring: export premapped to driver by struct virtqueue > > > > > > > virtio_net: set premapped mode by find_vqs() > > > > > > > virtio_ring: remove api of setting vq premapped > > > > > > > virtio_ring: introduce dma map api for page > > > > > > > virtio_ring: introduce virtqueue_dma_map_sg_attrs > > > > > > > virtio_net: unify the code for recycling the xmit ptr > > > > > > > virtio_net: rename free_old_xmit_skbs to free_old_xmit > > > > > > > virtio_net: sq support premapped mode > > > > > > > > > > > > > > arch/um/drivers/virtio_uml.c | 31 +- > > > > > > > drivers/net/virtio_net.c | 283 ++++++--- > > > > > > > drivers/platform/mellanox/mlxbf-tmfifo.c | 24 +- > > > > > > > drivers/remoteproc/remoteproc_virtio.c | 31 +- > > > > > > > drivers/s390/virtio/virtio_ccw.c | 33 +- > > > > > > > drivers/virtio/virtio_mmio.c | 30 +- > > > > > > > drivers/virtio/virtio_pci_common.c | 59 +- > > > > > > > drivers/virtio/virtio_pci_common.h | 9 +- > > > > > > > drivers/virtio/virtio_pci_legacy.c | 16 +- > > > > > > > drivers/virtio/virtio_pci_modern.c | 38 +- > > > > > > > drivers/virtio/virtio_ring.c | 698 ++++++++++++----------- > > > > > > > drivers/virtio/virtio_vdpa.c | 45 +- > > > > > > > include/linux/virtio.h | 13 +- > > > > > > > include/linux/virtio_config.h | 48 +- > > > > > > > include/linux/virtio_ring.h | 82 +-- > > > > > > > tools/virtio/virtio_test.c | 4 +- > > > > > > > tools/virtio/vringh_test.c | 28 +- > > > > > > > 17 files changed, 847 insertions(+), 625 deletions(-) > > > > > > > > > > > > > > -- > > > > > > > 2.32.0.3.g01195cf9f > > > > > > > > > > > > > > > >