From: "Michael S. Tsirkin" <mst@redhat.com> To: Guo Zhi <qtxuning1999@sjtu.edu.cn> Cc: jasowang@redhat.com, eperezma@redhat.com, virtualization@lists.linux-foundation.org, linux-kernel@vger.kernel.org, sgarzare@redhat.com Subject: Re: [PATCH v2 4/4] virtio_test: enable indirection feature Date: Thu, 7 Jul 2022 01:59:15 -0400 [thread overview] Message-ID: <20220707015717-mutt-send-email-mst@kernel.org> (raw) In-Reply-To: <2ff418ad-0ab8-1474-4308-2d163c1e6652@sjtu.edu.cn> On Thu, Jul 07, 2022 at 01:56:37PM +0800, Guo Zhi wrote: > On 2022/7/7 12:59, Michael S. Tsirkin wrote: > > On Thu, Jul 07, 2022 at 10:44:09AM +0800, Guo Zhi wrote: > > > Prior implementation don't use indirection feature because there is only > > > one descriptor for every io event, actually prior implementation don't > > > support indirection because vhost can't translate and find the indirect > > > descriptors. This commit enable virtio_test malloc indirect descriptors > > > in a indirect buffer and map this buffer to vhost, thus resolve this > > > problem. > > > > > > Signed-off-by: Guo Zhi <qtxuning1999@sjtu.edu.cn> > > > --- > > > tools/virtio/virtio_test.c | 50 ++++++++++++++++++++++++++++++++------ > > > 1 file changed, 42 insertions(+), 8 deletions(-) > > > > > > diff --git a/tools/virtio/virtio_test.c b/tools/virtio/virtio_test.c > > > index 363695b33..dca408a5c 100644 > > > --- a/tools/virtio/virtio_test.c > > > +++ b/tools/virtio/virtio_test.c > > > @@ -25,7 +25,7 @@ > > > #define RINGSIZE 256 > > > #define TEST_BUF_NUM 0x100000 > > > #define BUF_SIZE 1024 > > > -/* Unused */ > > > +#define INDIRECTS_SIZE (RINGSIZE * sizeof(struct vring_desc) * 8) > > > void *__kmalloc_fake, *__kfree_ignore_start, *__kfree_ignore_end; > > > struct vq_info { > > > @@ -47,6 +47,8 @@ struct vdev_info { > > > int nvqs; > > > void *buf; > > > size_t buf_size; > > > + void *indirects; > > > + size_t indirects_size; > > What are these exactly? > The buffer is used to put indirect descriptors, and the buffer will be added > in the vhost iotlb through VHOST_SET_MEM_TABLE, so that vhost can translate > the descriptors in user mode correctly. Pls come up with a better name. indirect_buf and indirect_buf_size? And add a comment. > > > > > struct vhost_memory *mem; > > > }; > > > @@ -131,6 +133,8 @@ static void vq_info_add(struct vdev_info *dev, int num) > > > static void vdev_info_init(struct vdev_info* dev, unsigned long long features) > > > { > > > int r; > > > + int nregions = 2; > > > + > > > memset(dev, 0, sizeof *dev); > > > dev->vdev.features = features; > > > INIT_LIST_HEAD(&dev->vdev.vqs); > > > @@ -138,19 +142,25 @@ static void vdev_info_init(struct vdev_info* dev, unsigned long long features) > > > dev->buf_size = BUF_SIZE; > > > dev->buf = malloc(dev->buf_size); > > > assert(dev->buf); > > > - dev->control = open("/dev/vhost-test", O_RDWR); > > > + dev->indirects_size = INDIRECTS_SIZE; > > > + dev->indirects = malloc(dev->indirects_size); > > > + assert(dev->indirects); > > > + dev->control = open("/dev/vhost-test", O_RDWR); > > > assert(dev->control >= 0); > > > r = ioctl(dev->control, VHOST_SET_OWNER, NULL); > > > assert(r >= 0); > > > dev->mem = malloc(offsetof(struct vhost_memory, regions) + > > > - sizeof dev->mem->regions[0]); > > > + (sizeof(dev->mem->regions[0])) * nregions); > > > assert(dev->mem); > > > memset(dev->mem, 0, offsetof(struct vhost_memory, regions) + > > > - sizeof dev->mem->regions[0]); > > > - dev->mem->nregions = 1; > > > + (sizeof(dev->mem->regions[0])) * nregions); > > > + dev->mem->nregions = nregions; > > > dev->mem->regions[0].guest_phys_addr = (long)dev->buf; > > > dev->mem->regions[0].userspace_addr = (long)dev->buf; > > > dev->mem->regions[0].memory_size = dev->buf_size; > > > + dev->mem->regions[1].guest_phys_addr = (long)dev->indirects; > > > + dev->mem->regions[1].userspace_addr = (long)dev->indirects; > > > + dev->mem->regions[1].memory_size = dev->indirects_size; > > > r = ioctl(dev->control, VHOST_SET_MEM_TABLE, dev->mem); > > > assert(r >= 0); > > > } > > > @@ -170,6 +180,19 @@ static void wait_for_interrupt(struct vdev_info *dev) > > > } > > > } > > > +static int test_virtqueue_add_outbuf(struct virtqueue *vq, > > > + struct scatterlist *sg, unsigned int num, > > > + void *data, void *indirects) > > > +{ > > > + int r; > > > + > > > + __kmalloc_fake = indirects; > > > + r = virtqueue_add_outbuf(vq, sg, num, data, > > > + GFP_ATOMIC); > > > + __kmalloc_fake = NULL; > > > + return r; > > > +} > > > + > > Quite a hack. Please add comments with documentation. Also - no way to > > avoid hacks? > > The __kmalloc_fake is refered from vringh_test. > > If not using __kmalloc_fake here, the vhost doesn't know how to translate > the indirects descriptors(user address). > > We may could register a single region as large as the whole virtual space in > the vhost iotlb using 1:1 mapping. > > But since they are tests, IMHO, better here to use VHOST_SET_MEM_TABLE with > more regions. But why do we need to do this on each buffer add? > > > > > static void run_test(struct vdev_info *dev, struct vq_info *vq, > > > bool delayed, int batch, int reset_n, int bufs) > > > { > > > @@ -181,6 +204,7 @@ static void run_test(struct vdev_info *dev, struct vq_info *vq, > > > unsigned len; > > > long long spurious = 0; > > > const bool random_batch = batch == RANDOM_BATCH; > > > + void *indirects; > > > r = ioctl(dev->control, VHOST_TEST_RUN, &test); > > > assert(r >= 0); > > > @@ -188,10 +212,15 @@ static void run_test(struct vdev_info *dev, struct vq_info *vq, > > > next_reset = INT_MAX; > > > } > > > + /* Don't kfree indirects. */ > > > + __kfree_ignore_start = dev->indirects; > > > + __kfree_ignore_end = dev->indirects + dev->indirects_size; > > > + > > > for (;;) { > > > virtqueue_disable_cb(vq->vq); > > > completed_before = completed; > > > started_before = started; > > > + indirects = dev->indirects; > > > do { > > > const bool reset = completed > next_reset; > > > if (random_batch) > > > @@ -203,9 +232,13 @@ static void run_test(struct vdev_info *dev, struct vq_info *vq, > > > sg_init_table(sg, sg_size); > > > for (int i = 0; i < sg_size; ++i) > > > sg_set_buf(&sg[i], dev->buf + i, 0x1); > > > - r = virtqueue_add_outbuf(vq->vq, sg, sg_size, > > > - dev->buf + started, > > > - GFP_ATOMIC); > > > + > > > + // use indirects buffer repeatedly > > > > C style comments please. > It will be modified. > > > > > + if (indirects + sg_size * sizeof(struct vring_desc) > > > > + dev->indirects + dev->indirects_size) > > > + indirects = dev->indirects; > > > + r = test_virtqueue_add_outbuf(vq->vq, sg, sg_size, > > > + dev->buf + started, indirects); > > > if (unlikely(r != 0)) { > > > if (r == -ENOSPC && > > > started > started_before) > > > @@ -216,6 +249,7 @@ static void run_test(struct vdev_info *dev, struct vq_info *vq, > > > } > > > ++started; > > > + indirects += sg_size * sizeof(struct vring_desc); > > > } > > > if (unlikely(!virtqueue_kick(vq->vq))) { > > > r = -1; > > > -- > > > 2.17.1 >
WARNING: multiple messages have this Message-ID (diff)
From: "Michael S. Tsirkin" <mst@redhat.com> To: Guo Zhi <qtxuning1999@sjtu.edu.cn> Cc: eperezma@redhat.com, linux-kernel@vger.kernel.org, virtualization@lists.linux-foundation.org Subject: Re: [PATCH v2 4/4] virtio_test: enable indirection feature Date: Thu, 7 Jul 2022 01:59:15 -0400 [thread overview] Message-ID: <20220707015717-mutt-send-email-mst@kernel.org> (raw) In-Reply-To: <2ff418ad-0ab8-1474-4308-2d163c1e6652@sjtu.edu.cn> On Thu, Jul 07, 2022 at 01:56:37PM +0800, Guo Zhi wrote: > On 2022/7/7 12:59, Michael S. Tsirkin wrote: > > On Thu, Jul 07, 2022 at 10:44:09AM +0800, Guo Zhi wrote: > > > Prior implementation don't use indirection feature because there is only > > > one descriptor for every io event, actually prior implementation don't > > > support indirection because vhost can't translate and find the indirect > > > descriptors. This commit enable virtio_test malloc indirect descriptors > > > in a indirect buffer and map this buffer to vhost, thus resolve this > > > problem. > > > > > > Signed-off-by: Guo Zhi <qtxuning1999@sjtu.edu.cn> > > > --- > > > tools/virtio/virtio_test.c | 50 ++++++++++++++++++++++++++++++++------ > > > 1 file changed, 42 insertions(+), 8 deletions(-) > > > > > > diff --git a/tools/virtio/virtio_test.c b/tools/virtio/virtio_test.c > > > index 363695b33..dca408a5c 100644 > > > --- a/tools/virtio/virtio_test.c > > > +++ b/tools/virtio/virtio_test.c > > > @@ -25,7 +25,7 @@ > > > #define RINGSIZE 256 > > > #define TEST_BUF_NUM 0x100000 > > > #define BUF_SIZE 1024 > > > -/* Unused */ > > > +#define INDIRECTS_SIZE (RINGSIZE * sizeof(struct vring_desc) * 8) > > > void *__kmalloc_fake, *__kfree_ignore_start, *__kfree_ignore_end; > > > struct vq_info { > > > @@ -47,6 +47,8 @@ struct vdev_info { > > > int nvqs; > > > void *buf; > > > size_t buf_size; > > > + void *indirects; > > > + size_t indirects_size; > > What are these exactly? > The buffer is used to put indirect descriptors, and the buffer will be added > in the vhost iotlb through VHOST_SET_MEM_TABLE, so that vhost can translate > the descriptors in user mode correctly. Pls come up with a better name. indirect_buf and indirect_buf_size? And add a comment. > > > > > struct vhost_memory *mem; > > > }; > > > @@ -131,6 +133,8 @@ static void vq_info_add(struct vdev_info *dev, int num) > > > static void vdev_info_init(struct vdev_info* dev, unsigned long long features) > > > { > > > int r; > > > + int nregions = 2; > > > + > > > memset(dev, 0, sizeof *dev); > > > dev->vdev.features = features; > > > INIT_LIST_HEAD(&dev->vdev.vqs); > > > @@ -138,19 +142,25 @@ static void vdev_info_init(struct vdev_info* dev, unsigned long long features) > > > dev->buf_size = BUF_SIZE; > > > dev->buf = malloc(dev->buf_size); > > > assert(dev->buf); > > > - dev->control = open("/dev/vhost-test", O_RDWR); > > > + dev->indirects_size = INDIRECTS_SIZE; > > > + dev->indirects = malloc(dev->indirects_size); > > > + assert(dev->indirects); > > > + dev->control = open("/dev/vhost-test", O_RDWR); > > > assert(dev->control >= 0); > > > r = ioctl(dev->control, VHOST_SET_OWNER, NULL); > > > assert(r >= 0); > > > dev->mem = malloc(offsetof(struct vhost_memory, regions) + > > > - sizeof dev->mem->regions[0]); > > > + (sizeof(dev->mem->regions[0])) * nregions); > > > assert(dev->mem); > > > memset(dev->mem, 0, offsetof(struct vhost_memory, regions) + > > > - sizeof dev->mem->regions[0]); > > > - dev->mem->nregions = 1; > > > + (sizeof(dev->mem->regions[0])) * nregions); > > > + dev->mem->nregions = nregions; > > > dev->mem->regions[0].guest_phys_addr = (long)dev->buf; > > > dev->mem->regions[0].userspace_addr = (long)dev->buf; > > > dev->mem->regions[0].memory_size = dev->buf_size; > > > + dev->mem->regions[1].guest_phys_addr = (long)dev->indirects; > > > + dev->mem->regions[1].userspace_addr = (long)dev->indirects; > > > + dev->mem->regions[1].memory_size = dev->indirects_size; > > > r = ioctl(dev->control, VHOST_SET_MEM_TABLE, dev->mem); > > > assert(r >= 0); > > > } > > > @@ -170,6 +180,19 @@ static void wait_for_interrupt(struct vdev_info *dev) > > > } > > > } > > > +static int test_virtqueue_add_outbuf(struct virtqueue *vq, > > > + struct scatterlist *sg, unsigned int num, > > > + void *data, void *indirects) > > > +{ > > > + int r; > > > + > > > + __kmalloc_fake = indirects; > > > + r = virtqueue_add_outbuf(vq, sg, num, data, > > > + GFP_ATOMIC); > > > + __kmalloc_fake = NULL; > > > + return r; > > > +} > > > + > > Quite a hack. Please add comments with documentation. Also - no way to > > avoid hacks? > > The __kmalloc_fake is refered from vringh_test. > > If not using __kmalloc_fake here, the vhost doesn't know how to translate > the indirects descriptors(user address). > > We may could register a single region as large as the whole virtual space in > the vhost iotlb using 1:1 mapping. > > But since they are tests, IMHO, better here to use VHOST_SET_MEM_TABLE with > more regions. But why do we need to do this on each buffer add? > > > > > static void run_test(struct vdev_info *dev, struct vq_info *vq, > > > bool delayed, int batch, int reset_n, int bufs) > > > { > > > @@ -181,6 +204,7 @@ static void run_test(struct vdev_info *dev, struct vq_info *vq, > > > unsigned len; > > > long long spurious = 0; > > > const bool random_batch = batch == RANDOM_BATCH; > > > + void *indirects; > > > r = ioctl(dev->control, VHOST_TEST_RUN, &test); > > > assert(r >= 0); > > > @@ -188,10 +212,15 @@ static void run_test(struct vdev_info *dev, struct vq_info *vq, > > > next_reset = INT_MAX; > > > } > > > + /* Don't kfree indirects. */ > > > + __kfree_ignore_start = dev->indirects; > > > + __kfree_ignore_end = dev->indirects + dev->indirects_size; > > > + > > > for (;;) { > > > virtqueue_disable_cb(vq->vq); > > > completed_before = completed; > > > started_before = started; > > > + indirects = dev->indirects; > > > do { > > > const bool reset = completed > next_reset; > > > if (random_batch) > > > @@ -203,9 +232,13 @@ static void run_test(struct vdev_info *dev, struct vq_info *vq, > > > sg_init_table(sg, sg_size); > > > for (int i = 0; i < sg_size; ++i) > > > sg_set_buf(&sg[i], dev->buf + i, 0x1); > > > - r = virtqueue_add_outbuf(vq->vq, sg, sg_size, > > > - dev->buf + started, > > > - GFP_ATOMIC); > > > + > > > + // use indirects buffer repeatedly > > > > C style comments please. > It will be modified. > > > > > + if (indirects + sg_size * sizeof(struct vring_desc) > > > > + dev->indirects + dev->indirects_size) > > > + indirects = dev->indirects; > > > + r = test_virtqueue_add_outbuf(vq->vq, sg, sg_size, > > > + dev->buf + started, indirects); > > > if (unlikely(r != 0)) { > > > if (r == -ENOSPC && > > > started > started_before) > > > @@ -216,6 +249,7 @@ static void run_test(struct vdev_info *dev, struct vq_info *vq, > > > } > > > ++started; > > > + indirects += sg_size * sizeof(struct vring_desc); > > > } > > > if (unlikely(!virtqueue_kick(vq->vq))) { > > > r = -1; > > > -- > > > 2.17.1 > _______________________________________________ Virtualization mailing list Virtualization@lists.linux-foundation.org https://lists.linuxfoundation.org/mailman/listinfo/virtualization
next prev parent reply other threads:[~2022-07-07 5:59 UTC|newest] Thread overview: 21+ messages / expand[flat|nested] mbox.gz Atom feed top 2022-07-07 2:44 [PATCH v2 0/4] virtio/virtio_test Guo Zhi 2022-07-07 2:44 ` [PATCH v2 1/4] virtio_test: kick vhost for a batch of descriptors Guo Zhi 2022-07-07 2:44 ` [PATCH v2 2/4] virtio_test: move magic number in code as defined constant Guo Zhi 2022-07-07 5:09 ` Michael S. Tsirkin 2022-07-07 5:09 ` Michael S. Tsirkin 2022-07-07 6:17 ` Guo Zhi 2022-07-07 7:36 ` Michael S. Tsirkin 2022-07-07 7:36 ` Michael S. Tsirkin 2022-07-07 12:15 ` Guo Zhi 2022-07-07 12:19 ` Guo Zhi 2022-07-07 2:44 ` [PATCH v2 3/4] virtio_test: use random length scatterlists to test descriptor chain Guo Zhi 2022-07-07 5:16 ` Michael S. Tsirkin 2022-07-07 5:16 ` Michael S. Tsirkin 2022-07-07 6:24 ` Guo Zhi 2022-07-07 2:44 ` [PATCH v2 4/4] virtio_test: enable indirection feature Guo Zhi 2022-07-07 4:59 ` Michael S. Tsirkin 2022-07-07 4:59 ` Michael S. Tsirkin 2022-07-07 5:56 ` Guo Zhi 2022-07-07 5:59 ` Michael S. Tsirkin [this message] 2022-07-07 5:59 ` Michael S. Tsirkin 2022-07-07 6:05 ` Guo Zhi
Reply instructions: You may reply publicly to this message via plain-text email using any one of the following methods: * Save the following mbox file, import it into your mail client, and reply-to-all from there: mbox Avoid top-posting and favor interleaved quoting: https://en.wikipedia.org/wiki/Posting_style#Interleaved_style * Reply using the --to, --cc, and --in-reply-to switches of git-send-email(1): git send-email \ --in-reply-to=20220707015717-mutt-send-email-mst@kernel.org \ --to=mst@redhat.com \ --cc=eperezma@redhat.com \ --cc=jasowang@redhat.com \ --cc=linux-kernel@vger.kernel.org \ --cc=qtxuning1999@sjtu.edu.cn \ --cc=sgarzare@redhat.com \ --cc=virtualization@lists.linux-foundation.org \ /path/to/YOUR_REPLY https://kernel.org/pub/software/scm/git/docs/git-send-email.html * If your mail client supports setting the In-Reply-To header via mailto: links, try the mailto: linkBe sure your reply has a Subject: header at the top and a blank line before the message body.
This is an external index of several public inboxes, see mirroring instructions on how to clone and mirror all data and code used by this external index.