| Message ID | 20240830070351.2855919-2-jens.wiklander@linaro.org |
|---|---|
| State | New |
| Headers | show |
| Series | [RFC,1/4] dma-buf: heaps: restricted_heap: add no_map attribute | expand |
On Fri, Aug 30, 2024 at 10:47 AM Christian König <christian.koenig@amd.com> wrote: > > Am 30.08.24 um 09:03 schrieb Jens Wiklander: > > Add a no_map attribute to struct restricted_heap_attachment and struct > > restricted_heap to skip the call to dma_map_sgtable() if set. This > > avoids trying to map a dma-buf that doens't refer to memory accessible > > by the kernel. > > You seem to have a misunderstanding here dma_map_sgtable() is called to > map a table into IOMMU and not any kernel address space. > > So please explain why you need that? You're right, I had misunderstood dma_map_sgtable(). There's no need for the no_map attribute, so I'll remove it. Perhaps you have a suggestion on how to fix a problem when using dma_map_sgtable()? Without it, I'll have to assign a pointer to teedev->dev.dma_mask when using the restricted heap with the OP-TEE driver or there will be a warning in __dma_map_sg_attrs() ending with a failure when trying to register the dma-buf fd. OP-TEE is probed with a platform device, and taking the dma_mask pointer from that device works. Is that a good approach or is there a better way of assigning dma_mask? Thanks, Jens > > Regards, > Christian. > > > > > Signed-off-by: Jens Wiklander <jens.wiklander@linaro.org> > > --- > > drivers/dma-buf/heaps/restricted_heap.c | 17 +++++++++++++---- > > drivers/dma-buf/heaps/restricted_heap.h | 2 ++ > > 2 files changed, 15 insertions(+), 4 deletions(-) > > > > diff --git a/drivers/dma-buf/heaps/restricted_heap.c b/drivers/dma-buf/heaps/restricted_heap.c > > index 8bc8a5e3f969..4bf28e3727ca 100644 > > --- a/drivers/dma-buf/heaps/restricted_heap.c > > +++ b/drivers/dma-buf/heaps/restricted_heap.c > > @@ -16,6 +16,7 @@ > > struct restricted_heap_attachment { > > struct sg_table *table; > > struct device *dev; > > + bool no_map; > > }; > > > > static int > > @@ -54,6 +55,8 @@ restricted_heap_memory_free(struct restricted_heap *rheap, struct restricted_buf > > static int restricted_heap_attach(struct dma_buf *dmabuf, struct dma_buf_attachment *attachment) > > { > > struct restricted_buffer *restricted_buf = dmabuf->priv; > > + struct dma_heap *heap = restricted_buf->heap; > > + struct restricted_heap *rheap = dma_heap_get_drvdata(heap); > > struct restricted_heap_attachment *a; > > struct sg_table *table; > > > > @@ -70,6 +73,7 @@ static int restricted_heap_attach(struct dma_buf *dmabuf, struct dma_buf_attachm > > sg_dma_mark_restricted(table->sgl); > > a->table = table; > > a->dev = attachment->dev; > > + a->no_map = rheap->no_map; > > attachment->priv = a; > > > > return 0; > > @@ -92,9 +96,12 @@ restricted_heap_map_dma_buf(struct dma_buf_attachment *attachment, > > struct sg_table *table = a->table; > > int ret; > > > > - ret = dma_map_sgtable(attachment->dev, table, direction, DMA_ATTR_SKIP_CPU_SYNC); > > - if (ret) > > - return ERR_PTR(ret); > > + if (!a->no_map) { > > + ret = dma_map_sgtable(attachment->dev, table, direction, > > + DMA_ATTR_SKIP_CPU_SYNC); > > + if (ret) > > + return ERR_PTR(ret); > > + } > > return table; > > } > > > > @@ -106,7 +113,9 @@ restricted_heap_unmap_dma_buf(struct dma_buf_attachment *attachment, struct sg_t > > > > WARN_ON(a->table != table); > > > > - dma_unmap_sgtable(attachment->dev, table, direction, DMA_ATTR_SKIP_CPU_SYNC); > > + if (!a->no_map) > > + dma_unmap_sgtable(attachment->dev, table, direction, > > + DMA_ATTR_SKIP_CPU_SYNC); > > } > > > > static int > > diff --git a/drivers/dma-buf/heaps/restricted_heap.h b/drivers/dma-buf/heaps/restricted_heap.h > > index 7dec4b8a471b..94cc0842f70d 100644 > > --- a/drivers/dma-buf/heaps/restricted_heap.h > > +++ b/drivers/dma-buf/heaps/restricted_heap.h > > @@ -27,6 +27,8 @@ struct restricted_heap { > > unsigned long cma_paddr; > > unsigned long cma_size; > > > > + bool no_map; > > + > > void *priv_data; > > }; > > >
Am 05.09.24 um 08:56 schrieb Jens Wiklander: > On Fri, Aug 30, 2024 at 10:47 AM Christian König > <christian.koenig@amd.com> wrote: >> Am 30.08.24 um 09:03 schrieb Jens Wiklander: >>> Add a no_map attribute to struct restricted_heap_attachment and struct >>> restricted_heap to skip the call to dma_map_sgtable() if set. This >>> avoids trying to map a dma-buf that doens't refer to memory accessible >>> by the kernel. >> You seem to have a misunderstanding here dma_map_sgtable() is called to >> map a table into IOMMU and not any kernel address space. >> >> So please explain why you need that? > You're right, I had misunderstood dma_map_sgtable(). There's no need > for the no_map attribute, so I'll remove it. > > Perhaps you have a suggestion on how to fix a problem when using > dma_map_sgtable()? > > Without it, I'll have to assign a pointer to teedev->dev.dma_mask when > using the restricted heap with the OP-TEE driver or there will be a > warning in __dma_map_sg_attrs() ending with a failure when trying to > register the dma-buf fd. OP-TEE is probed with a platform device, and > taking the dma_mask pointer from that device works. Is that a good > approach or is there a better way of assigning dma_mask? Mhm, I don't know the full picture so I have to make some assumptions. The teedev is just a virtual device which represents the restricted memory access paths of a real device? If that is true then taking the dma-mask from the real device is most likely the right thing to do. Regards, Christian. > > Thanks, > Jens > >> Regards, >> Christian. >> >>> Signed-off-by: Jens Wiklander <jens.wiklander@linaro.org> >>> --- >>> drivers/dma-buf/heaps/restricted_heap.c | 17 +++++++++++++---- >>> drivers/dma-buf/heaps/restricted_heap.h | 2 ++ >>> 2 files changed, 15 insertions(+), 4 deletions(-) >>> >>> diff --git a/drivers/dma-buf/heaps/restricted_heap.c b/drivers/dma-buf/heaps/restricted_heap.c >>> index 8bc8a5e3f969..4bf28e3727ca 100644 >>> --- a/drivers/dma-buf/heaps/restricted_heap.c >>> +++ b/drivers/dma-buf/heaps/restricted_heap.c >>> @@ -16,6 +16,7 @@ >>> struct restricted_heap_attachment { >>> struct sg_table *table; >>> struct device *dev; >>> + bool no_map; >>> }; >>> >>> static int >>> @@ -54,6 +55,8 @@ restricted_heap_memory_free(struct restricted_heap *rheap, struct restricted_buf >>> static int restricted_heap_attach(struct dma_buf *dmabuf, struct dma_buf_attachment *attachment) >>> { >>> struct restricted_buffer *restricted_buf = dmabuf->priv; >>> + struct dma_heap *heap = restricted_buf->heap; >>> + struct restricted_heap *rheap = dma_heap_get_drvdata(heap); >>> struct restricted_heap_attachment *a; >>> struct sg_table *table; >>> >>> @@ -70,6 +73,7 @@ static int restricted_heap_attach(struct dma_buf *dmabuf, struct dma_buf_attachm >>> sg_dma_mark_restricted(table->sgl); >>> a->table = table; >>> a->dev = attachment->dev; >>> + a->no_map = rheap->no_map; >>> attachment->priv = a; >>> >>> return 0; >>> @@ -92,9 +96,12 @@ restricted_heap_map_dma_buf(struct dma_buf_attachment *attachment, >>> struct sg_table *table = a->table; >>> int ret; >>> >>> - ret = dma_map_sgtable(attachment->dev, table, direction, DMA_ATTR_SKIP_CPU_SYNC); >>> - if (ret) >>> - return ERR_PTR(ret); >>> + if (!a->no_map) { >>> + ret = dma_map_sgtable(attachment->dev, table, direction, >>> + DMA_ATTR_SKIP_CPU_SYNC); >>> + if (ret) >>> + return ERR_PTR(ret); >>> + } >>> return table; >>> } >>> >>> @@ -106,7 +113,9 @@ restricted_heap_unmap_dma_buf(struct dma_buf_attachment *attachment, struct sg_t >>> >>> WARN_ON(a->table != table); >>> >>> - dma_unmap_sgtable(attachment->dev, table, direction, DMA_ATTR_SKIP_CPU_SYNC); >>> + if (!a->no_map) >>> + dma_unmap_sgtable(attachment->dev, table, direction, >>> + DMA_ATTR_SKIP_CPU_SYNC); >>> } >>> >>> static int >>> diff --git a/drivers/dma-buf/heaps/restricted_heap.h b/drivers/dma-buf/heaps/restricted_heap.h >>> index 7dec4b8a471b..94cc0842f70d 100644 >>> --- a/drivers/dma-buf/heaps/restricted_heap.h >>> +++ b/drivers/dma-buf/heaps/restricted_heap.h >>> @@ -27,6 +27,8 @@ struct restricted_heap { >>> unsigned long cma_paddr; >>> unsigned long cma_size; >>> >>> + bool no_map; >>> + >>> void *priv_data; >>> }; >>>
diff --git a/drivers/dma-buf/heaps/restricted_heap.c b/drivers/dma-buf/heaps/restricted_heap.c index 8bc8a5e3f969..4bf28e3727ca 100644 --- a/drivers/dma-buf/heaps/restricted_heap.c +++ b/drivers/dma-buf/heaps/restricted_heap.c @@ -16,6 +16,7 @@ struct restricted_heap_attachment { struct sg_table *table; struct device *dev; + bool no_map; }; static int @@ -54,6 +55,8 @@ restricted_heap_memory_free(struct restricted_heap *rheap, struct restricted_buf static int restricted_heap_attach(struct dma_buf *dmabuf, struct dma_buf_attachment *attachment) { struct restricted_buffer *restricted_buf = dmabuf->priv; + struct dma_heap *heap = restricted_buf->heap; + struct restricted_heap *rheap = dma_heap_get_drvdata(heap); struct restricted_heap_attachment *a; struct sg_table *table; @@ -70,6 +73,7 @@ static int restricted_heap_attach(struct dma_buf *dmabuf, struct dma_buf_attachm sg_dma_mark_restricted(table->sgl); a->table = table; a->dev = attachment->dev; + a->no_map = rheap->no_map; attachment->priv = a; return 0; @@ -92,9 +96,12 @@ restricted_heap_map_dma_buf(struct dma_buf_attachment *attachment, struct sg_table *table = a->table; int ret; - ret = dma_map_sgtable(attachment->dev, table, direction, DMA_ATTR_SKIP_CPU_SYNC); - if (ret) - return ERR_PTR(ret); + if (!a->no_map) { + ret = dma_map_sgtable(attachment->dev, table, direction, + DMA_ATTR_SKIP_CPU_SYNC); + if (ret) + return ERR_PTR(ret); + } return table; } @@ -106,7 +113,9 @@ restricted_heap_unmap_dma_buf(struct dma_buf_attachment *attachment, struct sg_t WARN_ON(a->table != table); - dma_unmap_sgtable(attachment->dev, table, direction, DMA_ATTR_SKIP_CPU_SYNC); + if (!a->no_map) + dma_unmap_sgtable(attachment->dev, table, direction, + DMA_ATTR_SKIP_CPU_SYNC); } static int diff --git a/drivers/dma-buf/heaps/restricted_heap.h b/drivers/dma-buf/heaps/restricted_heap.h index 7dec4b8a471b..94cc0842f70d 100644 --- a/drivers/dma-buf/heaps/restricted_heap.h +++ b/drivers/dma-buf/heaps/restricted_heap.h @@ -27,6 +27,8 @@ struct restricted_heap { unsigned long cma_paddr; unsigned long cma_size; + bool no_map; + void *priv_data; };
Add a no_map attribute to struct restricted_heap_attachment and struct restricted_heap to skip the call to dma_map_sgtable() if set. This avoids trying to map a dma-buf that doens't refer to memory accessible by the kernel. Signed-off-by: Jens Wiklander <jens.wiklander@linaro.org> --- drivers/dma-buf/heaps/restricted_heap.c | 17 +++++++++++++---- drivers/dma-buf/heaps/restricted_heap.h | 2 ++ 2 files changed, 15 insertions(+), 4 deletions(-)