Messages in this thread | | | Subject | Re: [PATCH v3 8/9] xen/gntdev: Implement dma-buf export functionality | From | Oleksandr Andrushchenko <> | Date | Wed, 13 Jun 2018 14:57:13 +0300 |
| |
On 06/13/2018 05:58 AM, Boris Ostrovsky wrote: > > > On 06/12/2018 09:41 AM, Oleksandr Andrushchenko wrote: >> From: Oleksandr Andrushchenko <oleksandr_andrushchenko@epam.com> >> >> 1. Create a dma-buf from grant references provided by the foreign >> domain. By default dma-buf is backed by system memory pages, but >> by providing GNTDEV_DMA_FLAG_XXX flags it can also be created >> as a DMA write-combine/coherent buffer, e.g. allocated with >> corresponding dma_alloc_xxx API. >> Export the resulting buffer as a new dma-buf. >> >> 2. Implement waiting for the dma-buf to be released: block until the >> dma-buf with the file descriptor provided is released. >> If within the time-out provided the buffer is not released then >> -ETIMEDOUT error is returned. If the buffer with the file descriptor >> does not exist or has already been released, then -ENOENT is >> returned. For valid file descriptors this must not be treated as >> error. >> >> 3. Make gntdev's common code and structures available to dma-buf. >> >> Signed-off-by: Oleksandr Andrushchenko >> <oleksandr_andrushchenko@epam.com> >> --- >> drivers/xen/gntdev-common.h | 4 + >> drivers/xen/gntdev-dmabuf.c | 470 +++++++++++++++++++++++++++++++++++- >> drivers/xen/gntdev.c | 10 + >> 3 files changed, 482 insertions(+), 2 deletions(-) >> >> diff --git a/drivers/xen/gntdev-common.h b/drivers/xen/gntdev-common.h >> index a3408fd39b07..72f80dbce861 100644 >> --- a/drivers/xen/gntdev-common.h >> +++ b/drivers/xen/gntdev-common.h >> @@ -89,4 +89,8 @@ bool gntdev_account_mapped_pages(int count); >> int gntdev_map_grant_pages(struct gntdev_grant_map *map); >> +#ifdef CONFIG_XEN_GNTDEV_DMABUF >> +void gntdev_remove_map(struct gntdev_priv *priv, struct >> gntdev_grant_map *map); >> +#endif >> + >> #endif >> diff --git a/drivers/xen/gntdev-dmabuf.c b/drivers/xen/gntdev-dmabuf.c >> index dc57c6a25525..84cba67c6ad7 100644 >> --- a/drivers/xen/gntdev-dmabuf.c >> +++ b/drivers/xen/gntdev-dmabuf.c >> @@ -3,13 +3,53 @@ >> /* >> * Xen dma-buf functionality for gntdev. >> * >> + * DMA buffer implementation is based on drivers/gpu/drm/drm_prime.c. >> + * >> * Copyright (c) 2018 Oleksandr Andrushchenko, EPAM Systems Inc. >> */ >> +#include <linux/dma-buf.h> >> #include <linux/slab.h> >> +#include <xen/grant_table.h> >> +#include <xen/gntdev.h> >> + >> +#include "gntdev-common.h" >> #include "gntdev-dmabuf.h" >> +struct gntdev_dmabuf { >> + struct gntdev_dmabuf_priv *priv; >> + struct dma_buf *dmabuf; >> + struct list_head next; >> + int fd; >> + >> + union { >> + struct { >> + /* Exported buffers are reference counted. */ >> + struct kref refcount; >> + >> + struct gntdev_priv *priv; >> + struct gntdev_grant_map *map; >> + } exp; >> + } u; >> + >> + /* Number of pages this buffer has. */ >> + int nr_pages; >> + /* Pages of this buffer. */ >> + struct page **pages; >> +}; >> + >> +struct gntdev_dmabuf_wait_obj { >> + struct list_head next; >> + struct gntdev_dmabuf *gntdev_dmabuf; >> + struct completion completion; >> +}; >> + >> +struct gntdev_dmabuf_attachment { >> + struct sg_table *sgt; >> + enum dma_data_direction dir; >> +}; >> + >> struct gntdev_dmabuf_priv { >> /* List of exported DMA buffers. */ >> struct list_head exp_list; >> @@ -23,17 +63,439 @@ struct gntdev_dmabuf_priv { >> /* Implementation of wait for exported DMA buffer to be released. */ >> +static void dmabuf_exp_release(struct kref *kref); >> + >> +static struct gntdev_dmabuf_wait_obj * >> +dmabuf_exp_wait_obj_new(struct gntdev_dmabuf_priv *priv, >> + struct gntdev_dmabuf *gntdev_dmabuf) >> +{ >> + struct gntdev_dmabuf_wait_obj *obj; >> + >> + obj = kzalloc(sizeof(*obj), GFP_KERNEL); >> + if (!obj) >> + return ERR_PTR(-ENOMEM); >> + >> + init_completion(&obj->completion); >> + obj->gntdev_dmabuf = gntdev_dmabuf; >> + >> + mutex_lock(&priv->lock); >> + list_add(&obj->next, &priv->exp_wait_list); >> + /* Put our reference and wait for gntdev_dmabuf's release to >> fire. */ >> + kref_put(&gntdev_dmabuf->u.exp.refcount, dmabuf_exp_release); >> + mutex_unlock(&priv->lock); >> + return obj; >> +} >> + >> +static void dmabuf_exp_wait_obj_free(struct gntdev_dmabuf_priv *priv, >> + struct gntdev_dmabuf_wait_obj *obj) >> +{ >> + struct gntdev_dmabuf_wait_obj *cur_obj, *q; >> + >> + mutex_lock(&priv->lock); >> + list_for_each_entry_safe(cur_obj, q, &priv->exp_wait_list, next) >> + if (cur_obj == obj) { >> + list_del(&obj->next); >> + kfree(obj); >> + break; >> + } >> + mutex_unlock(&priv->lock); > > > Do we really need to walk the list? > It can be deleted without walking the list and no reason to do that walk. Just an example of over-engineering here, thank you for spotting this. > And if we do, do we need the safe variant of the walk? We are holding > the lock. Here and elsewhere. > You are perfectly right. I will not use safe variant of the walk, no need for that > >> +} >> + >> +static int dmabuf_exp_wait_obj_wait(struct gntdev_dmabuf_wait_obj *obj, >> + u32 wait_to_ms) >> +{ >> + if (wait_for_completion_timeout(&obj->completion, >> + msecs_to_jiffies(wait_to_ms)) <= 0) >> + return -ETIMEDOUT; >> + >> + return 0; >> +} >> + >> +static void dmabuf_exp_wait_obj_signal(struct gntdev_dmabuf_priv *priv, >> + struct gntdev_dmabuf *gntdev_dmabuf) >> +{ >> + struct gntdev_dmabuf_wait_obj *obj, *q; >> + >> + list_for_each_entry_safe(obj, q, &priv->exp_wait_list, next) >> + if (obj->gntdev_dmabuf == gntdev_dmabuf) { >> + pr_debug("Found gntdev_dmabuf in the wait list, wake\n"); >> + complete_all(&obj->completion); >> + break; >> + } >> +} >> + >> +static struct gntdev_dmabuf * >> +dmabuf_exp_wait_obj_get_by_fd(struct gntdev_dmabuf_priv *priv, int fd) > > > The name of this routine implies (to me) that we are getting a wait > object but IIUIC we are getting a gntdev_dmabuf that we are going to > later associate with a wait object. > How about dmabuf_exp_wait_obj_get_dmabuf_by_fd? I would like to keep function prefixes, e.g. dmabuf_exp_wait_obj_ just to show to which functionality a routine belongs. > >> +{ >> + struct gntdev_dmabuf *q, *gntdev_dmabuf, *ret = ERR_PTR(-ENOENT); >> + >> + mutex_lock(&priv->lock); >> + list_for_each_entry_safe(gntdev_dmabuf, q, &priv->exp_list, next) >> + if (gntdev_dmabuf->fd == fd) { >> + pr_debug("Found gntdev_dmabuf in the wait list\n"); >> + kref_get(&gntdev_dmabuf->u.exp.refcount); >> + ret = gntdev_dmabuf; >> + break; >> + } >> + mutex_unlock(&priv->lock); >> + return ret; >> +} >> + >> int gntdev_dmabuf_exp_wait_released(struct gntdev_dmabuf_priv >> *priv, int fd, >> int wait_to_ms) >> { >> - return -EINVAL; >> + struct gntdev_dmabuf *gntdev_dmabuf; >> + struct gntdev_dmabuf_wait_obj *obj; >> + int ret; >> + >> + pr_debug("Will wait for dma-buf with fd %d\n", fd); >> + /* >> + * Try to find the DMA buffer: if not found means that >> + * either the buffer has already been released or file descriptor >> + * provided is wrong. >> + */ >> + gntdev_dmabuf = dmabuf_exp_wait_obj_get_by_fd(priv, fd); >> + if (IS_ERR(gntdev_dmabuf)) >> + return PTR_ERR(gntdev_dmabuf); >> + >> + /* >> + * gntdev_dmabuf still exists and is reference count locked by >> us now, >> + * so prepare to wait: allocate wait object and add it to the >> wait list, >> + * so we can find it on release. >> + */ >> + obj = dmabuf_exp_wait_obj_new(priv, gntdev_dmabuf); >> + if (IS_ERR(obj)) >> + return PTR_ERR(obj); >> + >> + ret = dmabuf_exp_wait_obj_wait(obj, wait_to_ms); >> + dmabuf_exp_wait_obj_free(priv, obj); >> + return ret; >> +} >> + >> +/* DMA buffer export support. */ >> + >> +static struct sg_table * >> +dmabuf_pages_to_sgt(struct page **pages, unsigned int nr_pages) >> +{ >> + struct sg_table *sgt; >> + int ret; >> + >> + sgt = kmalloc(sizeof(*sgt), GFP_KERNEL); >> + if (!sgt) { >> + ret = -ENOMEM; >> + goto out; >> + } >> + >> + ret = sg_alloc_table_from_pages(sgt, pages, nr_pages, 0, >> + nr_pages << PAGE_SHIFT, >> + GFP_KERNEL); >> + if (ret) >> + goto out; >> + >> + return sgt; >> + >> +out: >> + kfree(sgt); >> + return ERR_PTR(ret); >> +} >> + >> +static int dmabuf_exp_ops_attach(struct dma_buf *dma_buf, >> + struct device *target_dev, >> + struct dma_buf_attachment *attach) >> +{ >> + struct gntdev_dmabuf_attachment *gntdev_dmabuf_attach; >> + >> + gntdev_dmabuf_attach = kzalloc(sizeof(*gntdev_dmabuf_attach), >> + GFP_KERNEL); >> + if (!gntdev_dmabuf_attach) >> + return -ENOMEM; >> + >> + gntdev_dmabuf_attach->dir = DMA_NONE; >> + attach->priv = gntdev_dmabuf_attach; >> + return 0; >> +} >> + >> +static void dmabuf_exp_ops_detach(struct dma_buf *dma_buf, >> + struct dma_buf_attachment *attach) >> +{ >> + struct gntdev_dmabuf_attachment *gntdev_dmabuf_attach = >> attach->priv; >> + >> + if (gntdev_dmabuf_attach) { >> + struct sg_table *sgt = gntdev_dmabuf_attach->sgt; >> + >> + if (sgt) { >> + if (gntdev_dmabuf_attach->dir != DMA_NONE) >> + dma_unmap_sg_attrs(attach->dev, sgt->sgl, >> + sgt->nents, >> + gntdev_dmabuf_attach->dir, >> + DMA_ATTR_SKIP_CPU_SYNC); >> + sg_free_table(sgt); >> + } >> + >> + kfree(sgt); >> + kfree(gntdev_dmabuf_attach); >> + attach->priv = NULL; >> + } >> +} >> + >> +static struct sg_table * >> +dmabuf_exp_ops_map_dma_buf(struct dma_buf_attachment *attach, >> + enum dma_data_direction dir) >> +{ >> + struct gntdev_dmabuf_attachment *gntdev_dmabuf_attach = >> attach->priv; >> + struct gntdev_dmabuf *gntdev_dmabuf = attach->dmabuf->priv; >> + struct sg_table *sgt; >> + >> + pr_debug("Mapping %d pages for dev %p\n", gntdev_dmabuf->nr_pages, >> + attach->dev); >> + >> + if (dir == DMA_NONE || !gntdev_dmabuf_attach) >> + return ERR_PTR(-EINVAL); >> + >> + /* Return the cached mapping when possible. */ >> + if (gntdev_dmabuf_attach->dir == dir) >> + return gntdev_dmabuf_attach->sgt; >> + >> + /* >> + * Two mappings with different directions for the same >> attachment are >> + * not allowed. >> + */ >> + if (gntdev_dmabuf_attach->dir != DMA_NONE) >> + return ERR_PTR(-EBUSY); >> + >> + sgt = dmabuf_pages_to_sgt(gntdev_dmabuf->pages, >> + gntdev_dmabuf->nr_pages); >> + if (!IS_ERR(sgt)) { >> + if (!dma_map_sg_attrs(attach->dev, sgt->sgl, sgt->nents, dir, >> + DMA_ATTR_SKIP_CPU_SYNC)) { >> + sg_free_table(sgt); >> + kfree(sgt); >> + sgt = ERR_PTR(-ENOMEM); >> + } else { >> + gntdev_dmabuf_attach->sgt = sgt; >> + gntdev_dmabuf_attach->dir = dir; >> + } >> + } >> + if (IS_ERR(sgt)) >> + pr_debug("Failed to map sg table for dev %p\n", attach->dev); >> + return sgt; >> +} >> + >> +static void dmabuf_exp_ops_unmap_dma_buf(struct dma_buf_attachment >> *attach, >> + struct sg_table *sgt, >> + enum dma_data_direction dir) >> +{ >> + /* Not implemented. The unmap is done at >> dmabuf_exp_ops_detach(). */ >> +} >> + >> +static void dmabuf_exp_release(struct kref *kref) >> +{ >> + struct gntdev_dmabuf *gntdev_dmabuf = >> + container_of(kref, struct gntdev_dmabuf, u.exp.refcount); >> + >> + dmabuf_exp_wait_obj_signal(gntdev_dmabuf->priv, gntdev_dmabuf); >> + list_del(&gntdev_dmabuf->next); >> + kfree(gntdev_dmabuf); >> +} >> + >> +static void dmabuf_exp_ops_release(struct dma_buf *dma_buf) >> +{ >> + struct gntdev_dmabuf *gntdev_dmabuf = dma_buf->priv; >> + struct gntdev_dmabuf_priv *priv = gntdev_dmabuf->priv; >> + >> + gntdev_remove_map(gntdev_dmabuf->u.exp.priv, >> gntdev_dmabuf->u.exp.map); >> + mutex_lock(&priv->lock); >> + kref_put(&gntdev_dmabuf->u.exp.refcount, dmabuf_exp_release); >> + mutex_unlock(&priv->lock); >> +} >> + >> +static void *dmabuf_exp_ops_kmap_atomic(struct dma_buf *dma_buf, >> + unsigned long page_num) >> +{ >> + /* Not implemented. */ >> + return NULL; >> +} >> + >> +static void dmabuf_exp_ops_kunmap_atomic(struct dma_buf *dma_buf, >> + unsigned long page_num, void *addr) >> +{ >> + /* Not implemented. */ >> +} >> + >> +static void *dmabuf_exp_ops_kmap(struct dma_buf *dma_buf, >> + unsigned long page_num) >> +{ >> + /* Not implemented. */ >> + return NULL; >> +} >> + >> +static void dmabuf_exp_ops_kunmap(struct dma_buf *dma_buf, >> + unsigned long page_num, void *addr) >> +{ >> + /* Not implemented. */ >> +} >> + >> +static int dmabuf_exp_ops_mmap(struct dma_buf *dma_buf, >> + struct vm_area_struct *vma) >> +{ >> + /* Not implemented. */ >> + return 0; >> +} >> + >> +static const struct dma_buf_ops dmabuf_exp_ops = { >> + .attach = dmabuf_exp_ops_attach, >> + .detach = dmabuf_exp_ops_detach, >> + .map_dma_buf = dmabuf_exp_ops_map_dma_buf, >> + .unmap_dma_buf = dmabuf_exp_ops_unmap_dma_buf, >> + .release = dmabuf_exp_ops_release, >> + .map = dmabuf_exp_ops_kmap, >> + .map_atomic = dmabuf_exp_ops_kmap_atomic, >> + .unmap = dmabuf_exp_ops_kunmap, >> + .unmap_atomic = dmabuf_exp_ops_kunmap_atomic, >> + .mmap = dmabuf_exp_ops_mmap, >> +}; >> + >> +struct gntdev_dmabuf_export_args { >> + struct gntdev_priv *priv; >> + struct gntdev_grant_map *map; >> + struct gntdev_dmabuf_priv *dmabuf_priv; >> + struct device *dev; >> + int count; >> + struct page **pages; >> + u32 fd; >> +}; >> + >> +static int dmabuf_exp_from_pages(struct gntdev_dmabuf_export_args >> *args) >> +{ >> + DEFINE_DMA_BUF_EXPORT_INFO(exp_info); >> + struct gntdev_dmabuf *gntdev_dmabuf; >> + int ret = 0; > > > Not necessary. > Will remove =0; >> + >> + gntdev_dmabuf = kzalloc(sizeof(*gntdev_dmabuf), GFP_KERNEL); >> + if (!gntdev_dmabuf) >> + return -ENOMEM; >> + >> + kref_init(&gntdev_dmabuf->u.exp.refcount); >> + >> + gntdev_dmabuf->priv = args->dmabuf_priv; >> + gntdev_dmabuf->nr_pages = args->count; >> + gntdev_dmabuf->pages = args->pages; >> + gntdev_dmabuf->u.exp.priv = args->priv; >> + gntdev_dmabuf->u.exp.map = args->map; >> + >> + exp_info.exp_name = KBUILD_MODNAME; >> + if (args->dev->driver && args->dev->driver->owner) >> + exp_info.owner = args->dev->driver->owner; >> + else >> + exp_info.owner = THIS_MODULE; >> + exp_info.ops = &dmabuf_exp_ops; >> + exp_info.size = args->count << PAGE_SHIFT; >> + exp_info.flags = O_RDWR; >> + exp_info.priv = gntdev_dmabuf; >> + >> + gntdev_dmabuf->dmabuf = dma_buf_export(&exp_info); >> + if (IS_ERR(gntdev_dmabuf->dmabuf)) { >> + ret = PTR_ERR(gntdev_dmabuf->dmabuf); >> + gntdev_dmabuf->dmabuf = NULL; >> + goto fail; >> + } >> + >> + ret = dma_buf_fd(gntdev_dmabuf->dmabuf, O_CLOEXEC); >> + if (ret < 0) >> + goto fail; >> + >> + gntdev_dmabuf->fd = ret; >> + args->fd = ret; >> + >> + pr_debug("Exporting DMA buffer with fd %d\n", ret); >> + >> + mutex_lock(&args->dmabuf_priv->lock); >> + list_add(&gntdev_dmabuf->next, &args->dmabuf_priv->exp_list); >> + mutex_unlock(&args->dmabuf_priv->lock); >> + return 0; >> + >> +fail: >> + if (gntdev_dmabuf->dmabuf) >> + dma_buf_put(gntdev_dmabuf->dmabuf); >> + kfree(gntdev_dmabuf); >> + return ret; >> +} >> + >> +static struct gntdev_grant_map * >> +dmabuf_exp_alloc_backing_storage(struct gntdev_priv *priv, int >> dmabuf_flags, >> + int count) >> +{ >> + struct gntdev_grant_map *map; >> + >> + if (unlikely(count <= 0)) >> + return ERR_PTR(-EINVAL); >> + >> + if ((dmabuf_flags & GNTDEV_DMA_FLAG_WC) && >> + (dmabuf_flags & GNTDEV_DMA_FLAG_COHERENT)) { >> + pr_debug("Wrong dma-buf flags: either WC or coherent, not >> both\n"); > > Why not just print the value of the flags? Will print hex value of the flags > >> + return ERR_PTR(-EINVAL); >> + } >> + >> + map = gntdev_alloc_map(priv, count, dmabuf_flags); >> + if (!map) >> + return ERR_PTR(-ENOMEM); >> + >> + if (unlikely(gntdev_account_mapped_pages(count))) { >> + pr_debug("can't map: over limit\n"); > > > I think printing @count value here would be useful. > Will add > >> + gntdev_put_map(NULL, map); >> + return ERR_PTR(-ENOMEM); >> + } >> + return map; >> } >> int gntdev_dmabuf_exp_from_refs(struct gntdev_priv *priv, int flags, >> int count, u32 domid, u32 *refs, u32 *fd) >> { >> + struct gntdev_grant_map *map; >> + struct gntdev_dmabuf_export_args args; >> + int i, ret; >> + >> *fd = -1; > > > Is this still needed? No, will remove. I was thinking here about if user-space ignores IOCTL return value and tries to use the fd so it fails on -1. But, ok, no reason to fix user-space bugs in the kernel > >> - return -EINVAL; >> + >> + map = dmabuf_exp_alloc_backing_storage(priv, flags, count); >> + if (IS_ERR(map)) >> + return PTR_ERR(map); >> + >> + for (i = 0; i < count; i++) { >> + map->grants[i].domid = domid; >> + map->grants[i].ref = refs[i]; >> + } >> + >> + mutex_lock(&priv->lock); >> + gntdev_add_map(priv, map); >> + mutex_unlock(&priv->lock); >> + >> + map->flags |= GNTMAP_host_map; >> +#if defined(CONFIG_X86) >> + map->flags |= GNTMAP_device_map; >> +#endif >> + >> + ret = gntdev_map_grant_pages(map); >> + if (ret < 0) >> + goto out; >> + >> + args.priv = priv; >> + args.map = map; >> + args.dev = priv->dma_dev; >> + args.dmabuf_priv = priv->dmabuf_priv; >> + args.count = map->count; >> + args.pages = map->pages; >> + >> + ret = dmabuf_exp_from_pages(&args); >> + if (ret < 0) >> + goto out; >> + >> + *fd = args.fd; >> + return 0; >> + >> +out: >> + gntdev_remove_map(priv, map); >> + return ret; >> } >> /* DMA buffer import support. */ >> @@ -63,6 +525,10 @@ struct gntdev_dmabuf_priv *gntdev_dmabuf_init(void) >> if (!priv) >> return ERR_PTR(-ENOMEM); >> + mutex_init(&priv->lock); >> + INIT_LIST_HEAD(&priv->exp_list); >> + INIT_LIST_HEAD(&priv->exp_wait_list); >> + >> return priv; >> } >> diff --git a/drivers/xen/gntdev.c b/drivers/xen/gntdev.c >> index e82660d81d7e..5f93cd534840 100644 >> --- a/drivers/xen/gntdev.c >> +++ b/drivers/xen/gntdev.c >> @@ -262,6 +262,16 @@ void gntdev_put_map(struct gntdev_priv *priv, >> struct gntdev_grant_map *map) >> gntdev_free_map(map); >> } >> +#ifdef CONFIG_XEN_GNTDEV_DMABUF >> +void gntdev_remove_map(struct gntdev_priv *priv, struct >> gntdev_grant_map *map) >> +{ >> + mutex_lock(&priv->lock); >> + list_del(&map->next); >> + gntdev_put_map(NULL /* already removed */, map); > > > Why not pass call gntdev_put_map(priv, map) and then not have this > routine at all? > Well, I wish I could, but the main difference when calling gntdev_put_map(priv, map) with priv != NULL and my code is that:
void gntdev_put_map(struct gntdev_priv *priv, struct gntdev_grant_map *map) { [...] if (populate_freeable_maps && priv) { ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ mutex_lock(&priv->lock); list_del(&map->next); mutex_unlock(&priv->lock); } [...] }
and
#define populate_freeable_maps use_ptemod ^^^^^^^^^^ which means the map will never be removed from the list in my case because use_ptemod == false for dma-buf. This is why I do that by hand, e.g. remove the item from the list and then pass NULL for priv.
Also, I will remove gntdev_remove_map as I can now access priv->lock and gntdev_put_map directly form gntdev-dmabuf.c
> I really dislike the fact that we are taking a lock here that > gntdev_put_map() takes as well, although not with NULL argument. (And > yes, I see that gntdev_release() does it too.) > This can be re-factored later I guess? > > -boris > Thank you, Oleksandr > >> + mutex_unlock(&priv->lock); >> +} >> +#endif >> + >> /* >> ------------------------------------------------------------------ */ >> static int find_grant_ptes(pte_t *pte, pgtable_t token, >>
| |