Am 28.06.2018 um 11:58 schrieb Zhang, Jerry (Junwei):

On 06/22/2018 10:11 PM, Christian König wrote:

...

The caching of SGT's done by the DRM code is actually quite harmful and should probably removed altogether in the long term.

Start by providing a separate DMA-buf export implementation in amdgpu. This is also a prerequisite of unpinned DMA-buf handling.

v2: fix unintended recursion, remove debugging leftovers v3: split out from unpinned DMA-buf work

Signed-off-by: Christian König christian.koenig@amd.com

drivers/gpu/drm/amd/amdgpu/amdgpu.h       |  1 -   drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c   |  1 -   drivers/gpu/drm/amd/amdgpu/amdgpu_prime.c | 94 +++++++++++++------------------   3 files changed, 39 insertions(+), 57 deletions(-)

diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu.h b/drivers/gpu/drm/amd/amdgpu/amdgpu.h index d8e0cc08f9db..5e71af8dd3a7 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu.h @@ -373,7 +373,6 @@ int amdgpu_gem_object_open(struct drm_gem_object *obj,   void amdgpu_gem_object_close(struct drm_gem_object *obj,                   struct drm_file *file_priv);   unsigned long amdgpu_gem_timeout(uint64_t timeout_ns); -struct sg_table *amdgpu_gem_prime_get_sg_table(struct drm_gem_object *obj);   struct drm_gem_object *   amdgpu_gem_prime_import_sg_table(struct drm_device *dev,                    struct dma_buf_attachment *attach, diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c index a549483032b0..cdf0be85d361 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_drv.c @@ -919,7 +919,6 @@ static struct drm_driver kms_driver = {       .gem_prime_export = amdgpu_gem_prime_export,       .gem_prime_import = amdgpu_gem_prime_import,       .gem_prime_res_obj = amdgpu_gem_prime_res_obj, -    .gem_prime_get_sg_table = amdgpu_gem_prime_get_sg_table,

I may miss some patches or important info.

Even applied the series of patch, I still could find below code in drm_gem_map_dma_buf(). In this case, it will cause NULL pointer access. Please help me out of here. {{{ sgt = obj->dev->driver->gem_prime_get_sg_table(obj); }}}

Take a look at the change a bit further down.

Jerry

...

.gem_prime_import_sg_table = amdgpu_gem_prime_import_sg_table,       .gem_prime_vmap = amdgpu_gem_prime_vmap,       .gem_prime_vunmap = amdgpu_gem_prime_vunmap, diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_prime.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_prime.c index b2286bc41aec..038a8c8488b7 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_prime.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_prime.c @@ -40,22 +40,6 @@

static const struct dma_buf_ops amdgpu_dmabuf_ops;

-/**

• • amdgpu_gem_prime_get_sg_table - &drm_driver.gem_prime_get_sg_table
• • implementation
• • @obj: GEM buffer object
• • Returns:
• • A scatter/gather table for the pinned pages of the buffer

object's memory.

• */

-struct sg_table *amdgpu_gem_prime_get_sg_table(struct drm_gem_object *obj) -{ -    struct amdgpu_bo *bo = gem_to_amdgpu_bo(obj); -    int npages = bo->tbo.num_pages;

-    return drm_prime_pages_to_sg(bo->tbo.ttm->pages, npages); -}

/**    * amdgpu_gem_prime_vmap - &dma_buf_ops.vmap implementation    * @obj: GEM buffer object @@ -189,35 +173,29 @@ amdgpu_gem_prime_import_sg_table(struct drm_device *dev,   }

/**

• • amdgpu_gem_map_attach - &dma_buf_ops.attach implementation
• • @dma_buf: shared DMA buffer
• • @target_dev: target device

• • amdgpu_gem_map_dma_buf - &dma_buf_ops.map_dma_buf implementation

* @attach: DMA-buf attachment

• • @dir: DMA direction

*    * Makes sure that the shared DMA buffer can be accessed by the target device.    * For now, simply pins it to the GTT domain, where it should be accessible by    * all DMA devices.    *    * Returns:

• • 0 on success or negative error code.

• • sg_table filled with the DMA addresses to use or ERR_PRT with

negative error

• • code.

*/ -static int amdgpu_gem_map_attach(struct dma_buf *dma_buf, -                 struct dma_buf_attachment *attach) +static struct sg_table * +amdgpu_gem_map_dma_buf(struct dma_buf_attachment *attach, +               enum dma_data_direction dir)   { +    struct dma_buf *dma_buf = attach->dmabuf;       struct drm_gem_object *obj = dma_buf->priv;       struct amdgpu_bo *bo = gem_to_amdgpu_bo(obj);       struct amdgpu_device *adev = amdgpu_ttm_adev(bo->tbo.bdev); +    struct sg_table *sgt;       long r;

-    r = drm_gem_map_attach(dma_buf, attach); -    if (r) -        return r;

-    r = amdgpu_bo_reserve(bo, false); -    if (unlikely(r != 0)) -        goto error_detach;

if (attach->dev->driver != adev->dev->driver) {           /*            * Wait for all shared fences to complete before we switch to future @@ -228,54 +206,62 @@ static int amdgpu_gem_map_attach(struct dma_buf *dma_buf,                               MAX_SCHEDULE_TIMEOUT);           if (unlikely(r < 0)) {               DRM_DEBUG_PRIME("Fence wait failed: %li\n", r); -            goto error_unreserve; +            return ERR_PTR(r);           }       }

/* pin buffer into GTT */       r = amdgpu_bo_pin(bo, AMDGPU_GEM_DOMAIN_GTT, NULL);       if (r) -        goto error_unreserve; +        return ERR_PTR(r);

+    sgt = drm_prime_pages_to_sg(bo->tbo.ttm->pages, bo->tbo.num_pages); +    if (IS_ERR(sgt)) +        return sgt;

+    if (!dma_map_sg_attrs(attach->dev, sgt->sgl, sgt->nents, dir, +                  DMA_ATTR_SKIP_CPU_SYNC)) +        goto error_free;

if (attach->dev->driver != adev->dev->driver)           bo->prime_shared_count++;

-error_unreserve: -    amdgpu_bo_unreserve(bo); +    return sgt;

-error_detach: -    if (r) -        drm_gem_map_detach(dma_buf, attach); -    return r; +error_free: +    sg_free_table(sgt); +    kfree(sgt); +    return ERR_PTR(-ENOMEM);   }

/**

• • amdgpu_gem_map_detach - &dma_buf_ops.detach implementation
• • @dma_buf: shared DMA buffer

• • amdgpu_gem_unmap_dma_buf - &dma_buf_ops.unmap_dma_buf implementation

* @attach: DMA-buf attachment

• • @sgt: sg_table to unmap
• • @dir: DMA direction

*    * This is called when a shared DMA buffer no longer needs to be accessible by    * the other device. For now, simply unpins the buffer from GTT.    */ -static void amdgpu_gem_map_detach(struct dma_buf *dma_buf, -                  struct dma_buf_attachment *attach) +static void amdgpu_gem_unmap_dma_buf(struct dma_buf_attachment *attach, +                     struct sg_table *sgt, +                     enum dma_data_direction dir)   { +    struct dma_buf *dma_buf = attach->dmabuf;       struct drm_gem_object *obj = dma_buf->priv;       struct amdgpu_bo *bo = gem_to_amdgpu_bo(obj);       struct amdgpu_device *adev = amdgpu_ttm_adev(bo->tbo.bdev); -    int ret = 0;

-    ret = amdgpu_bo_reserve(bo, true); -    if (unlikely(ret != 0)) -        goto error;

amdgpu_bo_unpin(bo);

if (attach->dev->driver != adev->dev->driver && bo->prime_shared_count)           bo->prime_shared_count--; -    amdgpu_bo_unreserve(bo);

-error: -    drm_gem_map_detach(dma_buf, attach); +    if (sgt) { +        dma_unmap_sg(attach->dev, sgt->sgl, sgt->nents, dir); +        sg_free_table(sgt); +        kfree(sgt); +    }   }

/** @@ -333,10 +319,8 @@ static int amdgpu_gem_begin_cpu_access(struct dma_buf *dma_buf,   }

static const struct dma_buf_ops amdgpu_dmabuf_ops = { -    .attach = amdgpu_gem_map_attach, -    .detach = amdgpu_gem_map_detach, -    .map_dma_buf = drm_gem_map_dma_buf, -    .unmap_dma_buf = drm_gem_unmap_dma_buf,

Here we are stopping to use drm_gem_map_dma_buf(), so we don't need amdgpu_gem_prime_get_sg_table() any more either.

Christian.

...

+    .map_dma_buf = amdgpu_gem_map_dma_buf, +    .unmap_dma_buf = amdgpu_gem_unmap_dma_buf,       .release = drm_gem_dmabuf_release,       .begin_cpu_access = amdgpu_gem_begin_cpu_access,       .map = drm_gem_dmabuf_kmap,