Skip to contentSkip to conversation

lib/scatterlist: Provide a DMA page iterator

PATCH Applied Jason Gunthorpe 4 participants 5 messages Feb 7, 2019 – Feb 11, 2019

Subsystems (8)
lore
Share & tools
lore
#0 lore raw

PATCH v2 1/1 lib/scatterlist: Provide a DMA page iterator

Commit 2db76d7c3c6d ("lib/scatterlist: sg_page_iter: support sg lists w/o backing pages") introduced the sg_page_iter_dma_address() function without providing a way to use it in the general case. If the sg_dma_len() is not equal to the sg length callers cannot safely use the for_each_sg_page/sg_page_iter_dma_address combination.

Resolve this API mistake by providing a DMA specific iterator, for_each_sg_dma_page(), that uses the right length so sg_page_iter_dma_address() works as expected with all sglists.

A new iterator type is introduced to provide compile-time safety against wrongly mixing accessors and iterators.

---
 .clang-format                              |  1 +
 drivers/gpu/drm/vmwgfx/vmwgfx_ttm_buffer.c |  8 +++-
 drivers/media/pci/intel/ipu3/ipu3-cio2.c   |  4 +-
 include/linux/scatterlist.h                | 49 ++++++++++++++++++----
 lib/scatterlist.c                          | 26 ++++++++++++
 5 files changed, 76 insertions(+), 12 deletions(-)

v2:
- Drop the vmwgfx fix in favor of keeping it unchanged as there is no reviewer. Use an ugly case with a comment instead.

I'd like to apply this to the RDMA tree next week, a large series from Shiraz is waiting on it:

https://patchwork.kernel.org/project/linux-rdma/list/?series=71841&state=*

Regards,
Jason

.clang-format +1 -0
index bc2ffb2a0b5366..335ce29ab8132c 100644
@@ -240,6 +240,7 @@ ForEachMacros:
   - 'for_each_set_bit'
   - 'for_each_set_bit_from'
   - 'for_each_sg'
+  - 'for_each_sg_dma_page'
   - 'for_each_sg_page'
   - 'for_each_sibling_event'
   - '__for_each_thread'
drivers/gpu/drm/vmwgfx/vmwgfx_ttm_buffer.c +7 -1
index 31786b200afc47..e84f6aaee778f0 100644
@@ -311,7 +311,13 @@ static dma_addr_t __vmw_piter_dma_addr(struct vmw_piter *viter)
 
 static dma_addr_t __vmw_piter_sg_addr(struct vmw_piter *viter)
 {
-	return sg_page_iter_dma_address(&viter->iter);
+	/*
+	 * FIXME: This driver wrongly mixes DMA and CPU SG list iteration and
+	 * needs revision. See
+	 * https://lore.kernel.org/lkml/[email protected]/
+	 */
+	return sg_page_iter_dma_address(
+		(struct sg_dma_page_iter *)&viter->iter);
 }
 
 
drivers/media/pci/intel/ipu3/ipu3-cio2.c +2 -2
index cdb79ae2d8dc72..c0a5ce1d13b0bc 100644
@@ -846,7 +846,7 @@ static int cio2_vb2_buf_init(struct vb2_buffer *vb)
 	unsigned int pages = DIV_ROUND_UP(vb->planes[0].length, CIO2_PAGE_SIZE);
 	unsigned int lops = DIV_ROUND_UP(pages + 1, entries_per_page);
 	struct sg_table *sg;
-	struct sg_page_iter sg_iter;
+	struct sg_dma_page_iter sg_iter;
 	int i, j;
 
 	if (lops <= 0 || lops > CIO2_MAX_LOPS) {
@@ -873,7 +873,7 @@ static int cio2_vb2_buf_init(struct vb2_buffer *vb)
 		b->offset = sg->sgl->offset;
 
 	i = j = 0;
-	for_each_sg_page(sg->sgl, &sg_iter, sg->nents, 0) {
+	for_each_sg_dma_page(sg->sgl, &sg_iter, sg->nents, 0) {
 		if (!pages--)
 			break;
 		b->lop[i][j] = sg_page_iter_dma_address(&sg_iter) >> PAGE_SHIFT;
include/linux/scatterlist.h +40 -9

81 lines not on this page · raw message

lib/scatterlist.c +26 -0
index 9ba349e775ef08..739dc9fe2c55ec 100644
@@ -625,6 +625,32 @@ bool __sg_page_iter_next(struct sg_page_iter *piter)
 }
 EXPORT_SYMBOL(__sg_page_iter_next);
 
+static int sg_dma_page_count(struct scatterlist *sg)
+{
+	return PAGE_ALIGN(sg->offset + sg_dma_len(sg)) >> PAGE_SHIFT;
+}
+
+bool __sg_page_iter_dma_next(struct sg_dma_page_iter *dma_iter)
+{
+	struct sg_page_iter *piter = &dma_iter->base;
+
+	if (!piter->__nents || !piter->sg)
+		return false;
+
+	piter->sg_pgoffset += piter->__pg_advance;
+	piter->__pg_advance = 1;
+
+	while (piter->sg_pgoffset >= sg_dma_page_count(piter->sg)) {
+		piter->sg_pgoffset -= sg_dma_page_count(piter->sg);
+		piter->sg = sg_next(piter->sg);
+		if (!--piter->__nents || !piter->sg)
+			return false;
+	}
+
+	return true;
+}
+EXPORT_SYMBOL(__sg_page_iter_dma_next);
+
 /**
  * sg_miter_start - start mapping iteration over a sg list
  * @miter: sg mapping iter to be started
Signature · 1 line
2.20.1
#1 lore raw
Quoted 11 lines · Jason Gunthorpe

Commit 2db76d7c3c6d ("lib/scatterlist: sg_page_iter: support sg lists w/o backing pages") introduced the sg_page_iter_dma_address() function without providing a way to use it in the general case. If the sg_dma_len() is not equal to the sg length callers cannot safely use the for_each_sg_page/sg_page_iter_dma_address combination.

Resolve this API mistake by providing a DMA specific iterator, for_each_sg_dma_page(), that uses the right length so sg_page_iter_dma_address() works as expected with all sglists.

Quoted 7 lines

A new iterator type is introduced to provide compile-time safety against wrongly mixing accessors and iterators.

---
 .clang-format                              |  1 +

Thanks for updating the .clang-format, Jason! :-)

Cheers,
Miguel

Acked-by on v2 1/1
#2 lore raw
In reply to Jason Gunthorpe
Quoted 13 lines · Jason Gunthorpe

Commit 2db76d7c3c6d ("lib/scatterlist: sg_page_iter: support sg lists w/o
backing pages") introduced the sg_page_iter_dma_address() function without
providing a way to use it in the general case. If the sg_dma_len() is not
equal to the sg length callers cannot safely use the for_each_sg_page/sg_page_iter_dma_address combination.

Resolve this API mistake by providing a DMA specific iterator, for_each_sg_dma_page(), that uses the right length so sg_page_iter_dma_address() works as expected with all sglists.

Quoted 7 lines

A new iterator type is introduced to provide compile-time safety against
wrongly mixing accessors and iterators.

For the vmwgfx part,

I'll take a deeper look to provide a vmwgfx fix as a follow up.

Thanks,
Thomas

_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel

#3 lore raw
In reply to Jason Gunthorpe
Quoted 5 lines · Jason Gunthorpe
drivers/gpu/drm/vmwgfx/vmwgfx_ttm_buffer.c +1 -0
index 31786b200afc47..e84f6aaee778f0 100644
+++ b/drivers/gpu/drm/vmwgfx/vmwgfx_ttm_buffer.c
@@ -311,7 +311,13 @@ static dma_addr_t __vmw_piter_dma_addr(struct vmw_piter *viter)
 

drivers/gpu/drm/vmwgfx/vmwgfx_ttm_buffer.c static dma_addr_t __vmw_piter_dma_addr(struct vmw_piter *viter) · line 320

Quoted 10 lines
drivers/gpu/drm/vmwgfx/vmwgfx_ttm_buffer.c +7 -1
 static dma_addr_t __vmw_piter_sg_addr(struct vmw_piter *viter)
 {
-	return sg_page_iter_dma_address(&viter->iter);
+	/*
+	 * FIXME: This driver wrongly mixes DMA and CPU SG list iteration and
+	 * needs revision. See
+	 * https://lore.kernel.org/lkml/[email protected]/
+	 */
+	return sg_page_iter_dma_address(
+		(struct sg_dma_page_iter *)&viter->iter);

Occured to me this would be better written as:

	return sg_page_iter_dma_address(
		container_of(&viter->iter, struct sg_dma_page_iter, base));

Since I think we are done with this now I'll fix it when I apply this patch

Thanks for all the acks everyone

Regards,
Jason

#4 lore raw
In reply to Jason Gunthorpe
Quoted 16 lines · Jason Gunthorpe

Commit 2db76d7c3c6d ("lib/scatterlist: sg_page_iter: support sg lists w/o backing pages") introduced the sg_page_iter_dma_address() function without providing a way to use it in the general case. If the sg_dma_len() is not equal to the sg length callers cannot safely use the for_each_sg_page/sg_page_iter_dma_address combination.

Resolve this API mistake by providing a DMA specific iterator, for_each_sg_dma_page(), that uses the right length so sg_page_iter_dma_address() works as expected with all sglists.

A new iterator type is introduced to provide compile-time safety against wrongly mixing accessors and iterators.

Quoted 7 lines
---
 .clang-format                              |  1 +
 drivers/gpu/drm/vmwgfx/vmwgfx_ttm_buffer.c |  8 +++-
 drivers/media/pci/intel/ipu3/ipu3-cio2.c   |  4 +-
 include/linux/scatterlist.h                | 49 ++++++++++++++++++----
 lib/scatterlist.c                          | 26 ++++++++++++
 5 files changed, 76 insertions(+), 12 deletions(-)

Applied to rdma.git's for-next, thanks everyone.

Jason