[PATCH 1/6] dma-mapping: remove the {alloc,free}_noncoherent methods

[PATCH 1/6] dma-mapping: remove the {alloc,free}_noncoherent methods

[Ricardo Ribalda]

From: Christoph Hellwig <hch@lst.de>

It turns out allowing non-contigous allocations here was a rather bad
idea, as we'll now need to define ways to get the pages for mmaping
or dma_buf sharing.  Revert this change and stick to the original
concept.  A different API for the use case of non-contigous allocations
will be added back later.

Signed-off-by: Christoph Hellwig <hch@lst.de>
---
 drivers/iommu/dma-iommu.c   | 30 ------------------------------
 include/linux/dma-map-ops.h |  5 -----
 kernel/dma/mapping.c        | 33 ++++++---------------------------
 3 files changed, 6 insertions(+), 62 deletions(-)

diff --git a/drivers/iommu/dma-iommu.c b/drivers/iommu/dma-iommu.c
index 0cbcd3fc3e7e..73249732afd3 100644
--- a/drivers/iommu/dma-iommu.c
+++ b/drivers/iommu/dma-iommu.c
@@ -1054,34 +1054,6 @@ static void *iommu_dma_alloc(struct device *dev, size_t size,
 	return cpu_addr;
 }
 
-#ifdef CONFIG_DMA_REMAP
-static void *iommu_dma_alloc_noncoherent(struct device *dev, size_t size,
-		dma_addr_t *handle, enum dma_data_direction dir, gfp_t gfp)
-{
-	if (!gfpflags_allow_blocking(gfp)) {
-		struct page *page;
-
-		page = dma_common_alloc_pages(dev, size, handle, dir, gfp);
-		if (!page)
-			return NULL;
-		return page_address(page);
-	}
-
-	return iommu_dma_alloc_remap(dev, size, handle, gfp | __GFP_ZERO,
-				     PAGE_KERNEL, 0);
-}
-
-static void iommu_dma_free_noncoherent(struct device *dev, size_t size,
-		void *cpu_addr, dma_addr_t handle, enum dma_data_direction dir)
-{
-	__iommu_dma_unmap(dev, handle, size);
-	__iommu_dma_free(dev, size, cpu_addr);
-}
-#else
-#define iommu_dma_alloc_noncoherent		NULL
-#define iommu_dma_free_noncoherent		NULL
-#endif /* CONFIG_DMA_REMAP */
-
 static int iommu_dma_mmap(struct device *dev, struct vm_area_struct *vma,
 		void *cpu_addr, dma_addr_t dma_addr, size_t size,
 		unsigned long attrs)
@@ -1152,8 +1124,6 @@ static const struct dma_map_ops iommu_dma_ops = {
 	.free			= iommu_dma_free,
 	.alloc_pages		= dma_common_alloc_pages,
 	.free_pages		= dma_common_free_pages,
-	.alloc_noncoherent	= iommu_dma_alloc_noncoherent,
-	.free_noncoherent	= iommu_dma_free_noncoherent,
 	.mmap			= iommu_dma_mmap,
 	.get_sgtable		= iommu_dma_get_sgtable,
 	.map_page		= iommu_dma_map_page,
diff --git a/include/linux/dma-map-ops.h b/include/linux/dma-map-ops.h
index a5f89fc4d6df..3d1f91464bcf 100644
--- a/include/linux/dma-map-ops.h
+++ b/include/linux/dma-map-ops.h
@@ -22,11 +22,6 @@ struct dma_map_ops {
 			gfp_t gfp);
 	void (*free_pages)(struct device *dev, size_t size, struct page *vaddr,
 			dma_addr_t dma_handle, enum dma_data_direction dir);
-	void *(*alloc_noncoherent)(struct device *dev, size_t size,
-			dma_addr_t *dma_handle, enum dma_data_direction dir,
-			gfp_t gfp);
-	void (*free_noncoherent)(struct device *dev, size_t size, void *vaddr,
-			dma_addr_t dma_handle, enum dma_data_direction dir);
 	int (*mmap)(struct device *, struct vm_area_struct *,
 			void *, dma_addr_t, size_t, unsigned long attrs);
 
diff --git a/kernel/dma/mapping.c b/kernel/dma/mapping.c
index 51bb8fa8eb89..d3032513c54b 100644
--- a/kernel/dma/mapping.c
+++ b/kernel/dma/mapping.c
@@ -514,40 +514,19 @@ EXPORT_SYMBOL_GPL(dma_free_pages);
 void *dma_alloc_noncoherent(struct device *dev, size_t size,
 		dma_addr_t *dma_handle, enum dma_data_direction dir, gfp_t gfp)
 {
-	const struct dma_map_ops *ops = get_dma_ops(dev);
-	void *vaddr;
-
-	if (!ops || !ops->alloc_noncoherent) {
-		struct page *page;
-
-		page = dma_alloc_pages(dev, size, dma_handle, dir, gfp);
-		if (!page)
-			return NULL;
-		return page_address(page);
-	}
+	struct page *page;
 
-	size = PAGE_ALIGN(size);
-	vaddr = ops->alloc_noncoherent(dev, size, dma_handle, dir, gfp);
-	if (vaddr)
-		debug_dma_map_page(dev, virt_to_page(vaddr), 0, size, dir,
-				   *dma_handle);
-	return vaddr;
+	page = dma_alloc_pages(dev, size, dma_handle, dir, gfp);
+	if (!page)
+		return NULL;
+	return page_address(page);
 }
 EXPORT_SYMBOL_GPL(dma_alloc_noncoherent);
 
 void dma_free_noncoherent(struct device *dev, size_t size, void *vaddr,
 		dma_addr_t dma_handle, enum dma_data_direction dir)
 {
-	const struct dma_map_ops *ops = get_dma_ops(dev);
-
-	if (!ops || !ops->free_noncoherent) {
-		dma_free_pages(dev, size, virt_to_page(vaddr), dma_handle, dir);
-		return;
-	}
-
-	size = PAGE_ALIGN(size);
-	debug_dma_unmap_page(dev, dma_handle, size, dir);
-	ops->free_noncoherent(dev, size, vaddr, dma_handle, dir);
+	dma_free_pages(dev, size, virt_to_page(vaddr), dma_handle, dir);
 }
 EXPORT_SYMBOL_GPL(dma_free_noncoherent);
 
-- 
2.29.2.454.gaff20da3a2-goog

[PATCH 2/6] dma-direct: use __GFP_ZERO in dma_direct_alloc_pages

[Ricardo Ribalda]

From: Christoph Hellwig <hch@lst.de>

Prepare for supporting the DMA_ATTR_NO_KERNEL_MAPPING flag in
dma_alloc_pages.

Signed-off-by: Christoph Hellwig <hch@lst.de>
---
 kernel/dma/direct.c | 7 ++-----
 1 file changed, 2 insertions(+), 5 deletions(-)

diff --git a/kernel/dma/direct.c b/kernel/dma/direct.c
index 06c111544f61..76c741e610fc 100644
--- a/kernel/dma/direct.c
+++ b/kernel/dma/direct.c
@@ -280,13 +280,12 @@ struct page *dma_direct_alloc_pages(struct device *dev, size_t size,
 		dma_addr_t *dma_handle, enum dma_data_direction dir, gfp_t gfp)
 {
 	struct page *page;
-	void *ret;
 
 	if (IS_ENABLED(CONFIG_DMA_COHERENT_POOL) &&
 	    force_dma_unencrypted(dev) && !gfpflags_allow_blocking(gfp))
 		return dma_direct_alloc_from_pool(dev, size, dma_handle, gfp);
 
-	page = __dma_direct_alloc_pages(dev, size, gfp);
+	page = __dma_direct_alloc_pages(dev, size, gfp | __GFP_ZERO);
 	if (!page)
 		return NULL;
 	if (PageHighMem(page)) {
@@ -300,13 +299,11 @@ struct page *dma_direct_alloc_pages(struct device *dev, size_t size,
 		goto out_free_pages;
 	}
 
-	ret = page_address(page);
 	if (force_dma_unencrypted(dev)) {
-		if (set_memory_decrypted((unsigned long)ret,
+		if (set_memory_decrypted((unsigned long) page_address(page),
 				1 << get_order(size)))
 			goto out_free_pages;
 	}
-	memset(ret, 0, size);
 	*dma_handle = phys_to_dma_direct(dev, page_to_phys(page));
 	return page;
 out_free_pages:
-- 
2.29.2.454.gaff20da3a2-goog

[PATCH 3/6] dma-iommu: remove __iommu_dma_mmap

[Ricardo Ribalda]

From: Christoph Hellwig <hch@lst.de>

The function has a single caller, so open code it there and take
advantage of the precalculated page count variable.

Signed-off-by: Christoph Hellwig <hch@lst.de>
---
 drivers/iommu/dma-iommu.c | 17 +----------------
 1 file changed, 1 insertion(+), 16 deletions(-)

diff --git a/drivers/iommu/dma-iommu.c b/drivers/iommu/dma-iommu.c
index 73249732afd3..a2fb92de7e3d 100644
--- a/drivers/iommu/dma-iommu.c
+++ b/drivers/iommu/dma-iommu.c
@@ -655,21 +655,6 @@ static void *iommu_dma_alloc_remap(struct device *dev, size_t size,
 	return NULL;
 }
 
-/**
- * __iommu_dma_mmap - Map a buffer into provided user VMA
- * @pages: Array representing buffer from __iommu_dma_alloc()
- * @size: Size of buffer in bytes
- * @vma: VMA describing requested userspace mapping
- *
- * Maps the pages of the buffer in @pages into @vma. The caller is responsible
- * for verifying the correct size and protection of @vma beforehand.
- */
-static int __iommu_dma_mmap(struct page **pages, size_t size,
-		struct vm_area_struct *vma)
-{
-	return vm_map_pages(vma, pages, PAGE_ALIGN(size) >> PAGE_SHIFT);
-}
-
 static void iommu_dma_sync_single_for_cpu(struct device *dev,
 		dma_addr_t dma_handle, size_t size, enum dma_data_direction dir)
 {
@@ -1074,7 +1059,7 @@ static int iommu_dma_mmap(struct device *dev, struct vm_area_struct *vma,
 		struct page **pages = dma_common_find_pages(cpu_addr);
 
 		if (pages)
-			return __iommu_dma_mmap(pages, size, vma);
+			return vm_map_pages(vma, pages, nr_pages);
 		pfn = vmalloc_to_pfn(cpu_addr);
 	} else {
 		pfn = page_to_pfn(virt_to_page(cpu_addr));
-- 
2.29.2.454.gaff20da3a2-goog

[PATCH 4/6] WIP: add a dma_alloc_contiguous API

[Ricardo Ribalda]

From: Christoph Hellwig <hch@lst.de>

Add a new API that returns a virtually non-contigous array of pages
and dma address.  This API is only implemented for dma-iommu and will
not be implemented for non-iommu DMA API instances that have to allocate
contiguous memory.  It is up to the caller to check if the API is
available.

The intent is that media drivers can use this API if either:

 - no kernel mapping or only temporary kernel mappings are required.
   That is as a better replacement for DMA_ATTR_NO_KERNEL_MAPPING
 - a kernel mapping is required for cached and DMA mapped pages, but
   the driver also needs the pages to e.g. map them to userspace.
   In that sense it is a replacement for some aspects of the recently
   removed and never fully implemented DMA_ATTR_NON_CONSISTENT

Signed-off-by: Christoph Hellwig <hch@lst.de>
---
 drivers/iommu/dma-iommu.c   | 73 +++++++++++++++++++++++++------------
 include/linux/dma-map-ops.h |  4 ++
 include/linux/dma-mapping.h |  5 +++
 kernel/dma/mapping.c        | 35 ++++++++++++++++++
 4 files changed, 93 insertions(+), 24 deletions(-)

diff --git a/drivers/iommu/dma-iommu.c b/drivers/iommu/dma-iommu.c
index a2fb92de7e3d..2e72fe1b9c3b 100644
--- a/drivers/iommu/dma-iommu.c
+++ b/drivers/iommu/dma-iommu.c
@@ -564,23 +564,12 @@ static struct page **__iommu_dma_alloc_pages(struct device *dev,
 	return pages;
 }
 
-/**
- * iommu_dma_alloc_remap - Allocate and map a buffer contiguous in IOVA space
- * @dev: Device to allocate memory for. Must be a real device
- *	 attached to an iommu_dma_domain
- * @size: Size of buffer in bytes
- * @dma_handle: Out argument for allocated DMA handle
- * @gfp: Allocation flags
- * @prot: pgprot_t to use for the remapped mapping
- * @attrs: DMA attributes for this allocation
- *
- * If @size is less than PAGE_SIZE, then a full CPU page will be allocated,
+/*
+ * If size is less than PAGE_SIZE, then a full CPU page will be allocated,
  * but an IOMMU which supports smaller pages might not map the whole thing.
- *
- * Return: Mapped virtual address, or NULL on failure.
  */
-static void *iommu_dma_alloc_remap(struct device *dev, size_t size,
-		dma_addr_t *dma_handle, gfp_t gfp, pgprot_t prot,
+static struct page **__iommu_dma_alloc_noncontiguous(struct device *dev,
+		size_t size, dma_addr_t *dma_handle, gfp_t gfp, pgprot_t prot,
 		unsigned long attrs)
 {
 	struct iommu_domain *domain = iommu_get_dma_domain(dev);
@@ -592,7 +581,6 @@ static void *iommu_dma_alloc_remap(struct device *dev, size_t size,
 	struct page **pages;
 	struct sg_table sgt;
 	dma_addr_t iova;
-	void *vaddr;
 
 	*dma_handle = DMA_MAPPING_ERROR;
 
@@ -635,17 +623,10 @@ static void *iommu_dma_alloc_remap(struct device *dev, size_t size,
 			< size)
 		goto out_free_sg;
 
-	vaddr = dma_common_pages_remap(pages, size, prot,
-			__builtin_return_address(0));
-	if (!vaddr)
-		goto out_unmap;
-
 	*dma_handle = iova;
 	sg_free_table(&sgt);
-	return vaddr;
+	return pages;
 
-out_unmap:
-	__iommu_dma_unmap(dev, iova, size);
 out_free_sg:
 	sg_free_table(&sgt);
 out_free_iova:
@@ -655,6 +636,46 @@ static void *iommu_dma_alloc_remap(struct device *dev, size_t size,
 	return NULL;
 }
 
+static void *iommu_dma_alloc_remap(struct device *dev, size_t size,
+		dma_addr_t *dma_handle, gfp_t gfp, pgprot_t prot,
+		unsigned long attrs)
+{
+	struct page **pages;
+	void *vaddr;
+
+	pages = __iommu_dma_alloc_noncontiguous(dev, size, dma_handle, gfp,
+						prot, attrs);
+	if (!pages)
+		return NULL;
+	vaddr = dma_common_pages_remap(pages, size, prot,
+			__builtin_return_address(0));
+	if (!vaddr)
+		goto out_unmap;
+	return vaddr;
+
+out_unmap:
+	__iommu_dma_unmap(dev, *dma_handle, size);
+	__iommu_dma_free_pages(pages, PAGE_ALIGN(size) >> PAGE_SHIFT);
+	return NULL;
+}
+
+#ifdef CONFIG_DMA_REMAP
+static struct page **iommu_dma_alloc_noncontiguous(struct device *dev,
+		size_t size, dma_addr_t *dma_handle, gfp_t gfp,
+		unsigned long attrs)
+{
+	return __iommu_dma_alloc_noncontiguous(dev, size, dma_handle, gfp,
+					       PAGE_KERNEL, attrs);
+}
+
+static void iommu_dma_free_noncontiguous(struct device *dev, size_t size,
+		struct page **pages, dma_addr_t dma_handle)
+{
+	__iommu_dma_unmap(dev, dma_handle, size);
+	__iommu_dma_free_pages(pages, PAGE_ALIGN(size) >> PAGE_SHIFT);
+}
+#endif
+
 static void iommu_dma_sync_single_for_cpu(struct device *dev,
 		dma_addr_t dma_handle, size_t size, enum dma_data_direction dir)
 {
@@ -1109,6 +1130,10 @@ static const struct dma_map_ops iommu_dma_ops = {
 	.free			= iommu_dma_free,
 	.alloc_pages		= dma_common_alloc_pages,
 	.free_pages		= dma_common_free_pages,
+#ifdef CONFIG_DMA_REMAP
+	.alloc_noncontiguous	= iommu_dma_alloc_noncontiguous,
+	.free_noncontiguous	= iommu_dma_free_noncontiguous,
+#endif
 	.mmap			= iommu_dma_mmap,
 	.get_sgtable		= iommu_dma_get_sgtable,
 	.map_page		= iommu_dma_map_page,
diff --git a/include/linux/dma-map-ops.h b/include/linux/dma-map-ops.h
index 3d1f91464bcf..3cc313678d42 100644
--- a/include/linux/dma-map-ops.h
+++ b/include/linux/dma-map-ops.h
@@ -22,6 +22,10 @@ struct dma_map_ops {
 			gfp_t gfp);
 	void (*free_pages)(struct device *dev, size_t size, struct page *vaddr,
 			dma_addr_t dma_handle, enum dma_data_direction dir);
+	struct page **(*alloc_noncontiguous)(struct device *dev, size_t size,
+			dma_addr_t *dma_handle, gfp_t gfp, unsigned long attrs);
+	void (*free_noncontiguous)(struct device *dev, size_t size,
+			struct page **pages, dma_addr_t dma_handle);
 	int (*mmap)(struct device *, struct vm_area_struct *,
 			void *, dma_addr_t, size_t, unsigned long attrs);
 
diff --git a/include/linux/dma-mapping.h b/include/linux/dma-mapping.h
index 956151052d45..e1b4cb1d2e55 100644
--- a/include/linux/dma-mapping.h
+++ b/include/linux/dma-mapping.h
@@ -267,6 +267,11 @@ void *dma_alloc_noncoherent(struct device *dev, size_t size,
 		dma_addr_t *dma_handle, enum dma_data_direction dir, gfp_t gfp);
 void dma_free_noncoherent(struct device *dev, size_t size, void *vaddr,
 		dma_addr_t dma_handle, enum dma_data_direction dir);
+bool dma_can_alloc_noncontiguous(struct device *dev);
+struct page **dma_alloc_noncontiguous(struct device *dev, size_t size,
+		dma_addr_t *dma_handle, gfp_t gfp, unsigned long attrs);
+void dma_free_noncontiguous(struct device *dev, size_t size,
+		struct page **pages, dma_addr_t dma_handle);
 
 static inline dma_addr_t dma_map_single_attrs(struct device *dev, void *ptr,
 		size_t size, enum dma_data_direction dir, unsigned long attrs)
diff --git a/kernel/dma/mapping.c b/kernel/dma/mapping.c
index d3032513c54b..770c2f66512d 100644
--- a/kernel/dma/mapping.c
+++ b/kernel/dma/mapping.c
@@ -530,6 +530,41 @@ void dma_free_noncoherent(struct device *dev, size_t size, void *vaddr,
 }
 EXPORT_SYMBOL_GPL(dma_free_noncoherent);
 
+bool dma_can_alloc_noncontiguous(struct device *dev)
+{
+	const struct dma_map_ops *ops = get_dma_ops(dev);
+
+	return ops && ops->free_noncontiguous;
+}
+EXPORT_SYMBOL_GPL(dma_can_alloc_noncontiguous);
+
+struct page **dma_alloc_noncontiguous(struct device *dev, size_t size,
+		dma_addr_t *dma_handle, gfp_t gfp, unsigned long attrs)
+{
+	const struct dma_map_ops *ops = get_dma_ops(dev);
+
+	if (WARN_ON_ONCE(!dma_can_alloc_noncontiguous(dev)))
+		return NULL;
+	if (attrs & ~DMA_ATTR_ALLOC_SINGLE_PAGES) {
+		dev_warn(dev, "invalid flags (0x%lx) for %s\n",
+			 attrs, __func__);
+		return NULL;
+	}
+	return ops->alloc_noncontiguous(dev, size, dma_handle, gfp, attrs);
+}
+EXPORT_SYMBOL_GPL(dma_alloc_noncontiguous);
+
+void dma_free_noncontiguous(struct device *dev, size_t size,
+		struct page **pages, dma_addr_t dma_handle)
+{
+	const struct dma_map_ops *ops = get_dma_ops(dev);
+
+	if (WARN_ON_ONCE(!dma_can_alloc_noncontiguous(dev)))
+		return;
+	ops->free_noncontiguous(dev, size, pages, dma_handle);
+}
+EXPORT_SYMBOL_GPL(dma_free_noncontiguous);
+
 int dma_supported(struct device *dev, u64 mask)
 {
 	const struct dma_map_ops *ops = get_dma_ops(dev);
-- 
2.29.2.454.gaff20da3a2-goog

[PATCH 5/6] media: uvcvideo: Use dma_alloc_noncontiguos API

[Ricardo Ribalda]

On architectures where the is no coherent caching such as ARM use the
dma_alloc_noncontiguos API and handle manually the cache flushing using
dma_sync_single().

With this patch on the affected architectures we can measure up to 20x
performance improvement in uvc_video_copy_data_work().

Signed-off-by: Ricardo Ribalda <ribalda@chromium.org>
---
 drivers/media/usb/uvc/uvc_video.c | 74 ++++++++++++++++++++++++++-----
 drivers/media/usb/uvc/uvcvideo.h  |  1 +
 2 files changed, 63 insertions(+), 12 deletions(-)

diff --git a/drivers/media/usb/uvc/uvc_video.c b/drivers/media/usb/uvc/uvc_video.c
index a6a441d92b94..9e90b261428a 100644
--- a/drivers/media/usb/uvc/uvc_video.c
+++ b/drivers/media/usb/uvc/uvc_video.c
@@ -1490,6 +1490,11 @@ static void uvc_video_encode_bulk(struct uvc_urb *uvc_urb,
 	urb->transfer_buffer_length = stream->urb_size - len;
 }
 
+static inline struct device *stream_to_dmadev(struct uvc_streaming *stream)
+{
+	return stream->dev->udev->bus->controller->parent;
+}
+
 static void uvc_video_complete(struct urb *urb)
 {
 	struct uvc_urb *uvc_urb = urb->context;
@@ -1539,6 +1544,11 @@ static void uvc_video_complete(struct urb *urb)
 	 * Process the URB headers, and optionally queue expensive memcpy tasks
 	 * to be deferred to a work queue.
 	 */
+	if (uvc_urb->pages)
+		dma_sync_single_for_cpu(stream_to_dmadev(stream),
+					urb->transfer_dma,
+					urb->transfer_buffer_length,
+					DMA_FROM_DEVICE);
 	stream->decode(uvc_urb, buf, buf_meta);
 
 	/* If no async work is needed, resubmit the URB immediately. */
@@ -1566,8 +1576,15 @@ static void uvc_free_urb_buffers(struct uvc_streaming *stream)
 			continue;
 
 #ifndef CONFIG_DMA_NONCOHERENT
-		usb_free_coherent(stream->dev->udev, stream->urb_size,
-				  uvc_urb->buffer, uvc_urb->dma);
+		if (uvc_urb->pages) {
+			vunmap(uvc_urb->buffer);
+			dma_free_noncontiguous(stream_to_dmadev(stream),
+					       stream->urb_size,
+					       uvc_urb->pages, uvc_urb->dma);
+		} else {
+			usb_free_coherent(stream->dev->udev, stream->urb_size,
+					  uvc_urb->buffer, uvc_urb->dma);
+		}
 #else
 		kfree(uvc_urb->buffer);
 #endif
@@ -1577,6 +1594,47 @@ static void uvc_free_urb_buffers(struct uvc_streaming *stream)
 	stream->urb_size = 0;
 }
 
+#ifndef CONFIG_DMA_NONCOHERENT
+static bool uvc_alloc_urb_buffer(struct uvc_streaming *stream,
+				 struct uvc_urb *uvc_urb, gfp_t gfp_flags)
+{
+	struct device *dma_dev = dma_dev = stream_to_dmadev(stream);
+
+	if (!dma_can_alloc_noncontiguous(dma_dev)) {
+		uvc_urb->buffer = usb_alloc_coherent(stream->dev->udev,
+						     stream->urb_size,
+						     gfp_flags | __GFP_NOWARN,
+						     &uvc_urb->dma);
+		return uvc_urb->buffer != NULL;
+	}
+
+	uvc_urb->pages = dma_alloc_noncontiguous(dma_dev, stream->urb_size,
+						 &uvc_urb->dma,
+						 gfp_flags | __GFP_NOWARN, 0);
+	if (!uvc_urb->pages)
+		return false;
+
+	uvc_urb->buffer = vmap(uvc_urb->pages,
+			       PAGE_ALIGN(stream->urb_size) >> PAGE_SHIFT,
+			       VM_DMA_COHERENT, PAGE_KERNEL);
+	if (!uvc_urb->buffer) {
+		dma_free_noncontiguous(dma_dev, stream->urb_size,
+				       uvc_urb->pages, uvc_urb->dma);
+		return false;
+	}
+
+	return true;
+}
+#else
+static bool uvc_alloc_urb_buffer(struct uvc_streaming *stream,
+				 struct uvc_urb *uvc_urb, gfp_t gfp_flags)
+{
+	uvc_urb->buffer = kmalloc(stream->urb_size, gfp_flags | __GFP_NOWARN);
+
+	return uvc_urb->buffer != NULL;
+}
+#endif
+
 /*
  * Allocate transfer buffers. This function can be called with buffers
  * already allocated when resuming from suspend, in which case it will
@@ -1607,19 +1665,11 @@ static int uvc_alloc_urb_buffers(struct uvc_streaming *stream,
 
 	/* Retry allocations until one succeed. */
 	for (; npackets > 1; npackets /= 2) {
+		stream->urb_size = psize * npackets;
 		for (i = 0; i < UVC_URBS; ++i) {
 			struct uvc_urb *uvc_urb = &stream->uvc_urb[i];
 
-			stream->urb_size = psize * npackets;
-#ifndef CONFIG_DMA_NONCOHERENT
-			uvc_urb->buffer = usb_alloc_coherent(
-				stream->dev->udev, stream->urb_size,
-				gfp_flags | __GFP_NOWARN, &uvc_urb->dma);
-#else
-			uvc_urb->buffer =
-			    kmalloc(stream->urb_size, gfp_flags | __GFP_NOWARN);
-#endif
-			if (!uvc_urb->buffer) {
+			if (!uvc_alloc_urb_buffer(stream, uvc_urb, gfp_flags)) {
 				uvc_free_urb_buffers(stream);
 				break;
 			}
diff --git a/drivers/media/usb/uvc/uvcvideo.h b/drivers/media/usb/uvc/uvcvideo.h
index a3dfacf069c4..3e3ef1f1daa5 100644
--- a/drivers/media/usb/uvc/uvcvideo.h
+++ b/drivers/media/usb/uvc/uvcvideo.h
@@ -532,6 +532,7 @@ struct uvc_urb {
 
 	char *buffer;
 	dma_addr_t dma;
+	struct page **pages;
 
 	unsigned int async_operations;
 	struct uvc_copy_op copy_operations[UVC_MAX_PACKETS];
-- 
2.29.2.454.gaff20da3a2-goog

[PATCH 6/6] TEST-ONLY: media: uvcvideo: Add statistics for measuring performance

[Ricardo Ribalda]

From: Shik Chen <shik@chromium.org>

Majorly based on [1], with the following tweaks:

* Use div_u64 for u64 divisions
* Calculate standard deviation
* Fix an uninitialized |min| field for header
* Apply clang-format

[1] https://git.kernel.org/pub/scm/linux/kernel/git/kbingham/rcar.git/commit/?h=uvc/async-ml&id=cebbd1b629bbe5f856ec5dc7591478c003f5a944

Signed-off-by: Shik Chen <shik@chromium.org>
---
 drivers/media/usb/uvc/uvc_video.c | 163 +++++++++++++++++++++++++++++-
 drivers/media/usb/uvc/uvcvideo.h  |  21 ++++
 2 files changed, 181 insertions(+), 3 deletions(-)

diff --git a/drivers/media/usb/uvc/uvc_video.c b/drivers/media/usb/uvc/uvc_video.c
index 9e90b261428a..d3a515015003 100644
--- a/drivers/media/usb/uvc/uvc_video.c
+++ b/drivers/media/usb/uvc/uvc_video.c
@@ -906,12 +906,61 @@ static void uvc_video_stats_update(struct uvc_streaming *stream)
 	memset(&stream->stats.frame, 0, sizeof(stream->stats.frame));
 }
 
+size_t uvc_video_dump_time_stats(char *buf, size_t size,
+				 struct uvc_stats_time *stat, const char *pfx)
+{
+	unsigned int avg = 0;
+	unsigned int std = 0;
+
+	if (stat->qty) {
+		avg = div_u64(stat->duration, stat->qty);
+		std = int_sqrt64(div_u64(stat->duration2, stat->qty) -
+				 avg * avg);
+	}
+
+	/* Stat durations are in nanoseconds, we present in micro-seconds */
+	return scnprintf(
+		buf, size,
+		"%s: %llu/%u uS/qty: %u.%03u avg %u.%03u std %u.%03u min %u.%03u max (uS)\n",
+		pfx, div_u64(stat->duration, 1000), stat->qty, avg / 1000,
+		avg % 1000, std / 1000, std % 1000, stat->min / 1000,
+		stat->min % 1000, stat->max / 1000, stat->max % 1000);
+}
+
+size_t uvc_video_dump_speed(char *buf, size_t size, const char *pfx, u64 bytes,
+			    u64 milliseconds)
+{
+	unsigned int rate = 0;
+	bool gbit = false;
+
+	if (milliseconds)
+		rate = div_u64(bytes * 8, milliseconds);
+
+	if (rate >= 1000000) {
+		gbit = true;
+		rate /= 1000;
+	}
+
+	/*
+	 * bits/milliseconds == kilobits/seconds,
+	 * presented here as Mbits/s (or Gbit/s) with 3 decimal places
+	 */
+	return scnprintf(buf, size, "%s: %d.%03d %sbits/s\n", pfx, rate / 1000,
+			 rate % 1000, gbit ? "G" : "M");
+}
+
 size_t uvc_video_stats_dump(struct uvc_streaming *stream, char *buf,
 			    size_t size)
 {
+	u64 bytes = stream->stats.stream.bytes; /* Single sample */
+	unsigned int empty_ratio = 0;
 	unsigned int scr_sof_freq;
 	unsigned int duration;
+	unsigned int fps = 0;
 	size_t count = 0;
+	u64 cpu = 0;
+	u64 cpu_q = 0;
+	u32 cpu_r = 0;
 
 	/* Compute the SCR.SOF frequency estimate. At the nominal 1kHz SOF
 	 * frequency this will not overflow before more than 1h.
@@ -924,12 +973,19 @@ size_t uvc_video_stats_dump(struct uvc_streaming *stream, char *buf,
 	else
 		scr_sof_freq = 0;
 
+	if (stream->stats.stream.nb_packets)
+		empty_ratio = stream->stats.stream.nb_empty * 100 /
+			      stream->stats.stream.nb_packets;
+
 	count += scnprintf(buf + count, size - count,
-			   "frames:  %u\npackets: %u\nempty:   %u\n"
-			   "errors:  %u\ninvalid: %u\n",
+			   "frames:  %u\n"
+			   "packets: %u\n"
+			   "empty:   %u (%u %%)\n"
+			   "errors:  %u\n"
+			   "invalid: %u\n",
 			   stream->stats.stream.nb_frames,
 			   stream->stats.stream.nb_packets,
-			   stream->stats.stream.nb_empty,
+			   stream->stats.stream.nb_empty, empty_ratio,
 			   stream->stats.stream.nb_errors,
 			   stream->stats.stream.nb_invalid);
 	count += scnprintf(buf + count, size - count,
@@ -946,6 +1002,55 @@ size_t uvc_video_stats_dump(struct uvc_streaming *stream, char *buf,
 			   stream->stats.stream.min_sof,
 			   stream->stats.stream.max_sof,
 			   scr_sof_freq / 1000, scr_sof_freq % 1000);
+	count += scnprintf(buf + count, size - count,
+			   "bytes %lld : duration %d\n", bytes, duration);
+
+	if (duration != 0) {
+		/* Duration is in milliseconds, * 100 to gain 2 dp precision */
+		fps = stream->stats.stream.nb_frames * 1000 * 100 / duration;
+		/* CPU usage as a % with 6 decimal places */
+		cpu = div_u64(stream->stats.urbstat.decode.duration, duration) *
+		      100;
+	}
+
+	count += scnprintf(buf + count, size - count, "FPS: %u.%02u\n",
+			   fps / 100, fps % 100);
+
+	/* Processing Times */
+
+	count += uvc_video_dump_time_stats(buf + count, size - count,
+					   &stream->stats.urbstat.urb, "URB");
+	count += uvc_video_dump_time_stats(buf + count, size - count,
+					   &stream->stats.urbstat.header,
+					   "header");
+	count += uvc_video_dump_time_stats(buf + count, size - count,
+					   &stream->stats.urbstat.latency,
+					   "latency");
+	count += uvc_video_dump_time_stats(buf + count, size - count,
+					   &stream->stats.urbstat.decode,
+					   "decode");
+
+	/* Processing Speeds */
+
+	/* This should be representative of the memory bus / cpu speed */
+	count += uvc_video_dump_speed(
+		buf + count, size - count, "raw decode speed", bytes,
+		div_u64(stream->stats.urbstat.decode.duration, 1000000));
+
+	/* Raw bus speed - scheduling latencies */
+	count += uvc_video_dump_speed(
+		buf + count, size - count, "raw URB handling speed", bytes,
+		div_u64(stream->stats.urbstat.urb.duration, 1000000));
+
+	/* Throughput against wall clock time, stream duration is in millis*/
+	count += uvc_video_dump_speed(buf + count, size - count, "throughput",
+				      bytes, duration);
+
+	cpu_q = div_u64_rem(cpu, 1000000, &cpu_r);
+
+	/* Determine the 'CPU Usage' of our URB processing */
+	count += scnprintf(buf + count, size - count,
+			   "URB decode CPU usage %llu.%06u %%\n", cpu_q, cpu_r);
 
 	return count;
 }
@@ -954,6 +1059,11 @@ static void uvc_video_stats_start(struct uvc_streaming *stream)
 {
 	memset(&stream->stats, 0, sizeof(stream->stats));
 	stream->stats.stream.min_sof = 2048;
+
+	stream->stats.urbstat.header.min = -1;
+	stream->stats.urbstat.latency.min = -1;
+	stream->stats.urbstat.decode.min = -1;
+	stream->stats.urbstat.urb.min = -1;
 }
 
 static void uvc_video_stats_stop(struct uvc_streaming *stream)
@@ -961,6 +1071,28 @@ static void uvc_video_stats_stop(struct uvc_streaming *stream)
 	stream->stats.stream.stop_ts = ktime_get();
 }
 
+static s64 uvc_stats_add(struct uvc_stats_time *s, const ktime_t a,
+			 const ktime_t b)
+{
+	ktime_t delta;
+	u64 duration;
+
+	delta = ktime_sub(b, a);
+	duration = ktime_to_ns(delta);
+
+	s->qty++;
+	s->duration += duration;
+	s->duration2 += duration * duration;
+
+	if (duration < s->min)
+		s->min = duration;
+
+	if (duration > s->max)
+		s->max = duration;
+
+	return duration;
+}
+
 /* ------------------------------------------------------------------------
  * Video codecs
  */
@@ -1024,6 +1156,9 @@ static int uvc_video_decode_start(struct uvc_streaming *stream,
 		stream->sequence++;
 		if (stream->sequence)
 			uvc_video_stats_update(stream);
+
+		/* Update the stream timer each frame */
+		stream->stats.stream.stop_ts = ktime_get();
 	}
 
 	uvc_video_clock_decode(stream, buf, data, len);
@@ -1106,18 +1241,34 @@ static int uvc_video_decode_start(struct uvc_streaming *stream,
 static void uvc_video_copy_data_work(struct work_struct *work)
 {
 	struct uvc_urb *uvc_urb = container_of(work, struct uvc_urb, work);
+	ktime_t now;
 	unsigned int i;
 	int ret;
 
+	/* Measure decode performance */
+	uvc_urb->decode_start = ktime_get();
+	/* Measure scheduling latency */
+	uvc_stats_add(&uvc_urb->stream->stats.urbstat.latency,
+		      uvc_urb->received, uvc_urb->decode_start);
+
 	for (i = 0; i < uvc_urb->async_operations; i++) {
 		struct uvc_copy_op *op = &uvc_urb->copy_operations[i];
 
 		memcpy(op->dst, op->src, op->len);
+		uvc_urb->stream->stats.stream.bytes += op->len;
 
 		/* Release reference taken on this buffer. */
 		uvc_queue_buffer_release(op->buf);
 	}
 
+	now = ktime_get();
+	/* measure 'memcpy time' */
+	uvc_stats_add(&uvc_urb->stream->stats.urbstat.decode,
+		      uvc_urb->decode_start, now);
+	/* measure 'full urb processing time' */
+	uvc_stats_add(&uvc_urb->stream->stats.urbstat.urb, uvc_urb->received,
+		      now);
+
 	ret = usb_submit_urb(uvc_urb->urb, GFP_KERNEL);
 	if (ret < 0)
 		uvc_printk(KERN_ERR, "Failed to resubmit video URB (%d).\n",
@@ -1507,6 +1658,9 @@ static void uvc_video_complete(struct urb *urb)
 	unsigned long flags;
 	int ret;
 
+	/* Track URB processing performance */
+	uvc_urb->received = ktime_get();
+
 	switch (urb->status) {
 	case 0:
 		break;
@@ -1562,6 +1716,9 @@ static void uvc_video_complete(struct urb *urb)
 	}
 
 	queue_work(stream->async_wq, &uvc_urb->work);
+
+	uvc_stats_add(&uvc_urb->stream->stats.urbstat.header, uvc_urb->received,
+		      ktime_get());
 }
 
 /*
diff --git a/drivers/media/usb/uvc/uvcvideo.h b/drivers/media/usb/uvc/uvcvideo.h
index 3e3ef1f1daa5..80eeeaf3cd06 100644
--- a/drivers/media/usb/uvc/uvcvideo.h
+++ b/drivers/media/usb/uvc/uvcvideo.h
@@ -475,6 +475,14 @@ struct uvc_stats_frame {
 	u32 scr_stc;			/* SCR.STC of the last packet */
 };
 
+struct uvc_stats_time {
+	u64 duration;			/* Cumulative total duration between two events */
+	u64 duration2;			/* Cumulative total duration^2 between two events */
+	unsigned int qty;		/* Number of events represented in the total */
+	unsigned int min;		/* Shortest duration */
+	unsigned int max;		/* Longest duration */
+};
+
 struct uvc_stats_stream {
 	ktime_t start_ts;		/* Stream start timestamp */
 	ktime_t stop_ts;		/* Stream stop timestamp */
@@ -496,6 +504,8 @@ struct uvc_stats_stream {
 	unsigned int scr_sof;		/* STC.SOF of the last packet */
 	unsigned int min_sof;		/* Minimum STC.SOF value */
 	unsigned int max_sof;		/* Maximum STC.SOF value */
+
+	unsigned long bytes;		/* Successfully transferred bytes */
 };
 
 #define UVC_METADATA_BUF_SIZE 1024
@@ -525,6 +535,8 @@ struct uvc_copy_op {
  * @async_operations: counter to indicate the number of copy operations
  * @copy_operations: work descriptors for asynchronous copy operations
  * @work: work queue entry for asynchronous decode
+ * @received: URB interrupt time stamp
+ * @decode_start: URB processing start time stamp
  */
 struct uvc_urb {
 	struct urb *urb;
@@ -537,6 +549,9 @@ struct uvc_urb {
 	unsigned int async_operations;
 	struct uvc_copy_op copy_operations[UVC_MAX_PACKETS];
 	struct work_struct work;
+
+	ktime_t received;
+	ktime_t decode_start;
 };
 
 struct uvc_streaming {
@@ -599,6 +614,12 @@ struct uvc_streaming {
 	struct {
 		struct uvc_stats_frame frame;
 		struct uvc_stats_stream stream;
+		struct uvc_stats_urb {
+			struct uvc_stats_time header;
+			struct uvc_stats_time latency;
+			struct uvc_stats_time decode;
+			struct uvc_stats_time urb;
+		} urbstat;
 	} stats;
 
 	/* Timestamps support. */
-- 
2.29.2.454.gaff20da3a2-goog

Re: [PATCH 5/6] media: uvcvideo: Use dma_alloc_noncontiguos API

[Robin Murphy]

On 2020-11-24 15:38, Ricardo Ribalda wrote:
> On architectures where the is no coherent caching such as ARM use the
> dma_alloc_noncontiguos API and handle manually the cache flushing using
> dma_sync_single().
> 
> With this patch on the affected architectures we can measure up to 20x
> performance improvement in uvc_video_copy_data_work().
> 
> Signed-off-by: Ricardo Ribalda <ribalda@chromium.org>
> ---
>   drivers/media/usb/uvc/uvc_video.c | 74 ++++++++++++++++++++++++++-----
>   drivers/media/usb/uvc/uvcvideo.h  |  1 +
>   2 files changed, 63 insertions(+), 12 deletions(-)
> 
> diff --git a/drivers/media/usb/uvc/uvc_video.c b/drivers/media/usb/uvc/uvc_video.c
> index a6a441d92b94..9e90b261428a 100644
> --- a/drivers/media/usb/uvc/uvc_video.c
> +++ b/drivers/media/usb/uvc/uvc_video.c
> @@ -1490,6 +1490,11 @@ static void uvc_video_encode_bulk(struct uvc_urb *uvc_urb,
>   	urb->transfer_buffer_length = stream->urb_size - len;
>   }
>   
> +static inline struct device *stream_to_dmadev(struct uvc_streaming *stream)
> +{
> +	return stream->dev->udev->bus->controller->parent;
> +}
> +
>   static void uvc_video_complete(struct urb *urb)
>   {
>   	struct uvc_urb *uvc_urb = urb->context;
> @@ -1539,6 +1544,11 @@ static void uvc_video_complete(struct urb *urb)
>   	 * Process the URB headers, and optionally queue expensive memcpy tasks
>   	 * to be deferred to a work queue.
>   	 */
> +	if (uvc_urb->pages)
> +		dma_sync_single_for_cpu(stream_to_dmadev(stream),
> +					urb->transfer_dma,
> +					urb->transfer_buffer_length,
> +					DMA_FROM_DEVICE);

This doesn't work. Even in iommu-dma, the streaming API still expects to 
work on physically-contiguous memory that could have been passed to 
dma_map_single() in the first place. As-is, this will invalidate 
transfer_buffer_length bytes from the start of the *first* physical 
page, and thus destroy random other data if lines from subsequent 
unrelated pages are dirty in caches.

The only feasible way to do a DMA sync on disjoint pages in a single 
call is with a scatterlist.

Robin.

>   	stream->decode(uvc_urb, buf, buf_meta);
>   
>   	/* If no async work is needed, resubmit the URB immediately. */
> @@ -1566,8 +1576,15 @@ static void uvc_free_urb_buffers(struct uvc_streaming *stream)
>   			continue;
>   
>   #ifndef CONFIG_DMA_NONCOHERENT
> -		usb_free_coherent(stream->dev->udev, stream->urb_size,
> -				  uvc_urb->buffer, uvc_urb->dma);
> +		if (uvc_urb->pages) {
> +			vunmap(uvc_urb->buffer);
> +			dma_free_noncontiguous(stream_to_dmadev(stream),
> +					       stream->urb_size,
> +					       uvc_urb->pages, uvc_urb->dma);
> +		} else {
> +			usb_free_coherent(stream->dev->udev, stream->urb_size,
> +					  uvc_urb->buffer, uvc_urb->dma);
> +		}
>   #else
>   		kfree(uvc_urb->buffer);
>   #endif
> @@ -1577,6 +1594,47 @@ static void uvc_free_urb_buffers(struct uvc_streaming *stream)
>   	stream->urb_size = 0;
>   }
>   
> +#ifndef CONFIG_DMA_NONCOHERENT
> +static bool uvc_alloc_urb_buffer(struct uvc_streaming *stream,
> +				 struct uvc_urb *uvc_urb, gfp_t gfp_flags)
> +{
> +	struct device *dma_dev = dma_dev = stream_to_dmadev(stream);
> +
> +	if (!dma_can_alloc_noncontiguous(dma_dev)) {
> +		uvc_urb->buffer = usb_alloc_coherent(stream->dev->udev,
> +						     stream->urb_size,
> +						     gfp_flags | __GFP_NOWARN,
> +						     &uvc_urb->dma);
> +		return uvc_urb->buffer != NULL;
> +	}
> +
> +	uvc_urb->pages = dma_alloc_noncontiguous(dma_dev, stream->urb_size,
> +						 &uvc_urb->dma,
> +						 gfp_flags | __GFP_NOWARN, 0);
> +	if (!uvc_urb->pages)
> +		return false;
> +
> +	uvc_urb->buffer = vmap(uvc_urb->pages,
> +			       PAGE_ALIGN(stream->urb_size) >> PAGE_SHIFT,
> +			       VM_DMA_COHERENT, PAGE_KERNEL);
> +	if (!uvc_urb->buffer) {
> +		dma_free_noncontiguous(dma_dev, stream->urb_size,
> +				       uvc_urb->pages, uvc_urb->dma);
> +		return false;
> +	}
> +
> +	return true;
> +}
> +#else
> +static bool uvc_alloc_urb_buffer(struct uvc_streaming *stream,
> +				 struct uvc_urb *uvc_urb, gfp_t gfp_flags)
> +{
> +	uvc_urb->buffer = kmalloc(stream->urb_size, gfp_flags | __GFP_NOWARN);
> +
> +	return uvc_urb->buffer != NULL;
> +}
> +#endif
> +
>   /*
>    * Allocate transfer buffers. This function can be called with buffers
>    * already allocated when resuming from suspend, in which case it will
> @@ -1607,19 +1665,11 @@ static int uvc_alloc_urb_buffers(struct uvc_streaming *stream,
>   
>   	/* Retry allocations until one succeed. */
>   	for (; npackets > 1; npackets /= 2) {
> +		stream->urb_size = psize * npackets;
>   		for (i = 0; i < UVC_URBS; ++i) {
>   			struct uvc_urb *uvc_urb = &stream->uvc_urb[i];
>   
> -			stream->urb_size = psize * npackets;
> -#ifndef CONFIG_DMA_NONCOHERENT
> -			uvc_urb->buffer = usb_alloc_coherent(
> -				stream->dev->udev, stream->urb_size,
> -				gfp_flags | __GFP_NOWARN, &uvc_urb->dma);
> -#else
> -			uvc_urb->buffer =
> -			    kmalloc(stream->urb_size, gfp_flags | __GFP_NOWARN);
> -#endif
> -			if (!uvc_urb->buffer) {
> +			if (!uvc_alloc_urb_buffer(stream, uvc_urb, gfp_flags)) {
>   				uvc_free_urb_buffers(stream);
>   				break;
>   			}
> diff --git a/drivers/media/usb/uvc/uvcvideo.h b/drivers/media/usb/uvc/uvcvideo.h
> index a3dfacf069c4..3e3ef1f1daa5 100644
> --- a/drivers/media/usb/uvc/uvcvideo.h
> +++ b/drivers/media/usb/uvc/uvcvideo.h
> @@ -532,6 +532,7 @@ struct uvc_urb {
>   
>   	char *buffer;
>   	dma_addr_t dma;
> +	struct page **pages;
>   
>   	unsigned int async_operations;
>   	struct uvc_copy_op copy_operations[UVC_MAX_PACKETS];
> 

Re: [PATCH 5/6] media: uvcvideo: Use dma_alloc_noncontiguos API

[Ricardo Ribalda]

Hi Robin

On Tue, Nov 24, 2020 at 5:29 PM Robin Murphy <robin.murphy@arm.com> wrote:
>
> On 2020-11-24 15:38, Ricardo Ribalda wrote:
> > On architectures where the is no coherent caching such as ARM use the
> > dma_alloc_noncontiguos API and handle manually the cache flushing using
> > dma_sync_single().
> >
> > With this patch on the affected architectures we can measure up to 20x
> > performance improvement in uvc_video_copy_data_work().
> >
> > Signed-off-by: Ricardo Ribalda <ribalda@chromium.org>
> > ---
> >   drivers/media/usb/uvc/uvc_video.c | 74 ++++++++++++++++++++++++++-----
> >   drivers/media/usb/uvc/uvcvideo.h  |  1 +
> >   2 files changed, 63 insertions(+), 12 deletions(-)
> >
> > diff --git a/drivers/media/usb/uvc/uvc_video.c b/drivers/media/usb/uvc/uvc_video.c
> > index a6a441d92b94..9e90b261428a 100644
> > --- a/drivers/media/usb/uvc/uvc_video.c
> > +++ b/drivers/media/usb/uvc/uvc_video.c
> > @@ -1490,6 +1490,11 @@ static void uvc_video_encode_bulk(struct uvc_urb *uvc_urb,
> >       urb->transfer_buffer_length = stream->urb_size - len;
> >   }
> >
> > +static inline struct device *stream_to_dmadev(struct uvc_streaming *stream)
> > +{
> > +     return stream->dev->udev->bus->controller->parent;
> > +}
> > +
> >   static void uvc_video_complete(struct urb *urb)
> >   {
> >       struct uvc_urb *uvc_urb = urb->context;
> > @@ -1539,6 +1544,11 @@ static void uvc_video_complete(struct urb *urb)
> >        * Process the URB headers, and optionally queue expensive memcpy tasks
> >        * to be deferred to a work queue.
> >        */
> > +     if (uvc_urb->pages)
> > +             dma_sync_single_for_cpu(stream_to_dmadev(stream),
> > +                                     urb->transfer_dma,
> > +                                     urb->transfer_buffer_length,
> > +                                     DMA_FROM_DEVICE);
>
> This doesn't work. Even in iommu-dma, the streaming API still expects to
> work on physically-contiguous memory that could have been passed to
> dma_map_single() in the first place. As-is, this will invalidate
> transfer_buffer_length bytes from the start of the *first* physical
> page, and thus destroy random other data if lines from subsequent
> unrelated pages are dirty in caches.
>
> The only feasible way to do a DMA sync on disjoint pages in a single
> call is with a scatterlist.

Thanks for pointing this out. I guess I was lucky on my hardware and
the areas were always  contiguous.

Will rework and send back to the list.

Thanks again.

>
> Robin.
>
> >       stream->decode(uvc_urb, buf, buf_meta);
> >
> >       /* If no async work is needed, resubmit the URB immediately. */
> > @@ -1566,8 +1576,15 @@ static void uvc_free_urb_buffers(struct uvc_streaming *stream)
> >                       continue;
> >
> >   #ifndef CONFIG_DMA_NONCOHERENT
> > -             usb_free_coherent(stream->dev->udev, stream->urb_size,
> > -                               uvc_urb->buffer, uvc_urb->dma);
> > +             if (uvc_urb->pages) {
> > +                     vunmap(uvc_urb->buffer);
> > +                     dma_free_noncontiguous(stream_to_dmadev(stream),
> > +                                            stream->urb_size,
> > +                                            uvc_urb->pages, uvc_urb->dma);
> > +             } else {
> > +                     usb_free_coherent(stream->dev->udev, stream->urb_size,
> > +                                       uvc_urb->buffer, uvc_urb->dma);
> > +             }
> >   #else
> >               kfree(uvc_urb->buffer);
> >   #endif
> > @@ -1577,6 +1594,47 @@ static void uvc_free_urb_buffers(struct uvc_streaming *stream)
> >       stream->urb_size = 0;
> >   }
> >
> > +#ifndef CONFIG_DMA_NONCOHERENT
> > +static bool uvc_alloc_urb_buffer(struct uvc_streaming *stream,
> > +                              struct uvc_urb *uvc_urb, gfp_t gfp_flags)
> > +{
> > +     struct device *dma_dev = dma_dev = stream_to_dmadev(stream);
> > +
> > +     if (!dma_can_alloc_noncontiguous(dma_dev)) {
> > +             uvc_urb->buffer = usb_alloc_coherent(stream->dev->udev,
> > +                                                  stream->urb_size,
> > +                                                  gfp_flags | __GFP_NOWARN,
> > +                                                  &uvc_urb->dma);
> > +             return uvc_urb->buffer != NULL;
> > +     }
> > +
> > +     uvc_urb->pages = dma_alloc_noncontiguous(dma_dev, stream->urb_size,
> > +                                              &uvc_urb->dma,
> > +                                              gfp_flags | __GFP_NOWARN, 0);
> > +     if (!uvc_urb->pages)
> > +             return false;
> > +
> > +     uvc_urb->buffer = vmap(uvc_urb->pages,
> > +                            PAGE_ALIGN(stream->urb_size) >> PAGE_SHIFT,
> > +                            VM_DMA_COHERENT, PAGE_KERNEL);
> > +     if (!uvc_urb->buffer) {
> > +             dma_free_noncontiguous(dma_dev, stream->urb_size,
> > +                                    uvc_urb->pages, uvc_urb->dma);
> > +             return false;
> > +     }
> > +
> > +     return true;
> > +}
> > +#else
> > +static bool uvc_alloc_urb_buffer(struct uvc_streaming *stream,
> > +                              struct uvc_urb *uvc_urb, gfp_t gfp_flags)
> > +{
> > +     uvc_urb->buffer = kmalloc(stream->urb_size, gfp_flags | __GFP_NOWARN);
> > +
> > +     return uvc_urb->buffer != NULL;
> > +}
> > +#endif
> > +
> >   /*
> >    * Allocate transfer buffers. This function can be called with buffers
> >    * already allocated when resuming from suspend, in which case it will
> > @@ -1607,19 +1665,11 @@ static int uvc_alloc_urb_buffers(struct uvc_streaming *stream,
> >
> >       /* Retry allocations until one succeed. */
> >       for (; npackets > 1; npackets /= 2) {
> > +             stream->urb_size = psize * npackets;
> >               for (i = 0; i < UVC_URBS; ++i) {
> >                       struct uvc_urb *uvc_urb = &stream->uvc_urb[i];
> >
> > -                     stream->urb_size = psize * npackets;
> > -#ifndef CONFIG_DMA_NONCOHERENT
> > -                     uvc_urb->buffer = usb_alloc_coherent(
> > -                             stream->dev->udev, stream->urb_size,
> > -                             gfp_flags | __GFP_NOWARN, &uvc_urb->dma);
> > -#else
> > -                     uvc_urb->buffer =
> > -                         kmalloc(stream->urb_size, gfp_flags | __GFP_NOWARN);
> > -#endif
> > -                     if (!uvc_urb->buffer) {
> > +                     if (!uvc_alloc_urb_buffer(stream, uvc_urb, gfp_flags)) {
> >                               uvc_free_urb_buffers(stream);
> >                               break;
> >                       }
> > diff --git a/drivers/media/usb/uvc/uvcvideo.h b/drivers/media/usb/uvc/uvcvideo.h
> > index a3dfacf069c4..3e3ef1f1daa5 100644
> > --- a/drivers/media/usb/uvc/uvcvideo.h
> > +++ b/drivers/media/usb/uvc/uvcvideo.h
> > @@ -532,6 +532,7 @@ struct uvc_urb {
> >
> >       char *buffer;
> >       dma_addr_t dma;
> > +     struct page **pages;
> >
> >       unsigned int async_operations;
> >       struct uvc_copy_op copy_operations[UVC_MAX_PACKETS];
> >



-- 
Ricardo Ribalda

Re: [PATCH 2/6] dma-direct: use __GFP_ZERO in dma_direct_alloc_pages

[Christoph Hellwig]

On Tue, Nov 24, 2020 at 04:38:41PM +0100, Ricardo Ribalda wrote:
> From: Christoph Hellwig <hch@lst.de>
> 
> Prepare for supporting the DMA_ATTR_NO_KERNEL_MAPPING flag in
> dma_alloc_pages.
> 
> Signed-off-by: Christoph Hellwig <hch@lst.de>

FYI, this patch should be dropped as I've implemented the functionality
differently.

Re: [PATCH 3/6] dma-iommu: remove __iommu_dma_mmap

[Christoph Hellwig]

On Tue, Nov 24, 2020 at 04:38:42PM +0100, Ricardo Ribalda wrote:
> From: Christoph Hellwig <hch@lst.de>
> 
> The function has a single caller, so open code it there and take
> advantage of the precalculated page count variable.
> 
> Signed-off-by: Christoph Hellwig <hch@lst.de>

It turns out this isn't really required for the series.  I think it is
a useful cleanup, but it should probably be picked up separately.

Robin, any comments?