| Message ID | 20150609162710.21910.57295.stgit@dwillia2-desk3.amr.corp.intel.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
> -----Original Message----- > From: linux-kernel-owner@vger.kernel.org [mailto:linux-kernel- > owner@vger.kernel.org] On Behalf Of Dan Williams > Sent: Tuesday, June 09, 2015 10:27 AM > Subject: [PATCH 1/2] scatterlist: use sg_phys() > ... > diff --git a/arch/microblaze/kernel/dma.c b/arch/microblaze/kernel/dma.c > index ed7ba8a11822..dcb3c594d626 100644 > --- a/arch/microblaze/kernel/dma.c > +++ b/arch/microblaze/kernel/dma.c > @@ -61,7 +61,7 @@ static int dma_direct_map_sg(struct device *dev, struct > scatterlist *sgl, > /* FIXME this part of code is untested */ > for_each_sg(sgl, sg, nents, i) { > sg->dma_address = sg_phys(sg); > - __dma_sync(page_to_phys(sg_page(sg)) + sg->offset, > + __dma_sync(sg_phys(sg), > sg->length, direction); > } That one ends up with weird indentation.
On Tue, Jun 09, 2015 at 12:27:10PM -0400, Dan Williams wrote: > diff --git a/arch/arm/mm/dma-mapping.c b/arch/arm/mm/dma-mapping.c > index 7e7583ddd607..9f6ff6671f01 100644 > --- a/arch/arm/mm/dma-mapping.c > +++ b/arch/arm/mm/dma-mapping.c > @@ -1502,7 +1502,7 @@ static int __map_sg_chunk(struct device *dev, struct scatterlist *sg, > return -ENOMEM; > > for (count = 0, s = sg; count < (size >> PAGE_SHIFT); s = sg_next(s)) { > - phys_addr_t phys = page_to_phys(sg_page(s)); > + phys_addr_t phys = sg_phys(s) - s->offset; So sg_phys() turns out to be 'page_to_phys(sg_page(s)) + s->offset', which makes the above statement to: page_to_phys(sg_page(s)) + s->offset - s->offset; The compiler will probably optimize that away, but it still doesn't look like an improvement. > unsigned int len = PAGE_ALIGN(s->offset + s->length); > > if (!is_coherent && > diff --git a/arch/microblaze/kernel/dma.c b/arch/microblaze/kernel/dma.c > index ed7ba8a11822..dcb3c594d626 100644 > --- a/arch/microblaze/kernel/dma.c > +++ b/arch/microblaze/kernel/dma.c > @@ -61,7 +61,7 @@ static int dma_direct_map_sg(struct device *dev, struct scatterlist *sgl, > /* FIXME this part of code is untested */ > for_each_sg(sgl, sg, nents, i) { > sg->dma_address = sg_phys(sg); > - __dma_sync(page_to_phys(sg_page(sg)) + sg->offset, > + __dma_sync(sg_phys(sg), > sg->length, direction); Here the replacement makes sense, but weird indendation. Could all be moved to one line, I guess. > } > > diff --git a/drivers/iommu/intel-iommu.c b/drivers/iommu/intel-iommu.c > index 68d43beccb7e..9b9ada71e0d3 100644 > --- a/drivers/iommu/intel-iommu.c > +++ b/drivers/iommu/intel-iommu.c > @@ -1998,7 +1998,7 @@ static int __domain_mapping(struct dmar_domain *domain, unsigned long iov_pfn, > sg_res = aligned_nrpages(sg->offset, sg->length); > sg->dma_address = ((dma_addr_t)iov_pfn << VTD_PAGE_SHIFT) + sg->offset; > sg->dma_length = sg->length; > - pteval = page_to_phys(sg_page(sg)) | prot; > + pteval = (sg_phys(sg) - sg->offset) | prot; Here it doesn't make sense too. In general, please remove the cases where you have to subtract sg->offset after the conversion. Joerg
On Wed, Jun 10, 2015 at 2:32 AM, Joerg Roedel <joro@8bytes.org> wrote: > On Tue, Jun 09, 2015 at 12:27:10PM -0400, Dan Williams wrote: >> diff --git a/arch/arm/mm/dma-mapping.c b/arch/arm/mm/dma-mapping.c >> index 7e7583ddd607..9f6ff6671f01 100644 >> --- a/arch/arm/mm/dma-mapping.c >> +++ b/arch/arm/mm/dma-mapping.c >> @@ -1502,7 +1502,7 @@ static int __map_sg_chunk(struct device *dev, struct scatterlist *sg, >> return -ENOMEM; >> >> for (count = 0, s = sg; count < (size >> PAGE_SHIFT); s = sg_next(s)) { >> - phys_addr_t phys = page_to_phys(sg_page(s)); >> + phys_addr_t phys = sg_phys(s) - s->offset; > > So sg_phys() turns out to be 'page_to_phys(sg_page(s)) + s->offset', > which makes the above statement to: > > page_to_phys(sg_page(s)) + s->offset - s->offset; > > The compiler will probably optimize that away, but it still doesn't look > like an improvement. The goal is to eventually stop leaking struct page deep into the i/o stack. Anything that relies on being able to retrieve a struct page out of an sg entry needs to be converted. I think we need a new helper for this case "sg_phys_aligned()?".
On Wed, Jun 10, 2015 at 09:00:31AM -0700, Dan Williams wrote: > On Wed, Jun 10, 2015 at 2:32 AM, Joerg Roedel <joro@8bytes.org> wrote: > > On Tue, Jun 09, 2015 at 12:27:10PM -0400, Dan Williams wrote: > >> diff --git a/arch/arm/mm/dma-mapping.c b/arch/arm/mm/dma-mapping.c > >> index 7e7583ddd607..9f6ff6671f01 100644 > >> --- a/arch/arm/mm/dma-mapping.c > >> +++ b/arch/arm/mm/dma-mapping.c > >> @@ -1502,7 +1502,7 @@ static int __map_sg_chunk(struct device *dev, struct scatterlist *sg, > >> return -ENOMEM; > >> > >> for (count = 0, s = sg; count < (size >> PAGE_SHIFT); s = sg_next(s)) { > >> - phys_addr_t phys = page_to_phys(sg_page(s)); > >> + phys_addr_t phys = sg_phys(s) - s->offset; > > > > So sg_phys() turns out to be 'page_to_phys(sg_page(s)) + s->offset', > > which makes the above statement to: > > > > page_to_phys(sg_page(s)) + s->offset - s->offset; > > > > The compiler will probably optimize that away, but it still doesn't look > > like an improvement. > > The goal is to eventually stop leaking struct page deep into the i/o > stack. Anything that relies on being able to retrieve a struct page > out of an sg entry needs to be converted. I think we need a new > helper for this case "sg_phys_aligned()?". Why? The aim of the code is not to detect whether the address is aligned to a page (if it were, it'd be testing for a zero s->offset, or it would be testing for an s->offset being a multiple of the page size. Let's first understand the code that's being modified (which seems to be something which isn't done very much today - people seem to just like changing code for the hell of it.) for (count = 0, s = sg; count < (size >> PAGE_SHIFT); s = sg_next(s)) { phys_addr_t phys = page_to_phys(sg_page(s)); unsigned int len = PAGE_ALIGN(s->offset + s->length); if (!is_coherent && !dma_get_attr(DMA_ATTR_SKIP_CPU_SYNC, attrs)) __dma_page_cpu_to_dev(sg_page(s), s->offset, s->length, dir); prot = __dma_direction_to_prot(dir); ret = iommu_map(mapping->domain, iova, phys, len, prot); if (ret < 0) goto fail; count += len >> PAGE_SHIFT; iova += len; } What it's doing is trying to map each scatterlist entry via an IOMMU. Unsurprisingly, IOMMUs are page based - you can't map a partial IOMMU page. However, what says that the IOMMU page size is the same as the host CPU page size - it may not be... so the above code is wrong for a completely different reason. What we _should_ be doing is finding out what the IOMMU minimum page size is, and using that in conjunction with the sg_phys() of the scatterlist entry to determine the start and length of the mapping such that the IOMMU mapping covers the range described by the scatterlist entry. (iommu_map() takes arguments which must be aligned to the IOMMU minimum page size.) However, what we can also see from the above is that we have other code here using sg_page() - which is a necessity to be able to perform the required DMA cache maintanence to ensure that the data is visible to the DMA device. We need to kmap_atomic() these in order to flush them, and there's no other way other than struct page to represent a highmem page. So, I think your intent to want to remove struct page from the I/O subsystem is a false hope, unless you want to end up rewriting lots of architecture code, and you can come up with an alternative method to represent highmem pages.
On Wed, Jun 10, 2015 at 9:31 AM, Russell King - ARM Linux <linux@arm.linux.org.uk> wrote: > On Wed, Jun 10, 2015 at 09:00:31AM -0700, Dan Williams wrote: >> On Wed, Jun 10, 2015 at 2:32 AM, Joerg Roedel <joro@8bytes.org> wrote: >> > On Tue, Jun 09, 2015 at 12:27:10PM -0400, Dan Williams wrote: >> >> diff --git a/arch/arm/mm/dma-mapping.c b/arch/arm/mm/dma-mapping.c >> >> index 7e7583ddd607..9f6ff6671f01 100644 >> >> --- a/arch/arm/mm/dma-mapping.c >> >> +++ b/arch/arm/mm/dma-mapping.c >> >> @@ -1502,7 +1502,7 @@ static int __map_sg_chunk(struct device *dev, struct scatterlist *sg, >> >> return -ENOMEM; >> >> >> >> for (count = 0, s = sg; count < (size >> PAGE_SHIFT); s = sg_next(s)) { >> >> - phys_addr_t phys = page_to_phys(sg_page(s)); >> >> + phys_addr_t phys = sg_phys(s) - s->offset; >> > >> > So sg_phys() turns out to be 'page_to_phys(sg_page(s)) + s->offset', >> > which makes the above statement to: >> > >> > page_to_phys(sg_page(s)) + s->offset - s->offset; >> > >> > The compiler will probably optimize that away, but it still doesn't look >> > like an improvement. >> >> The goal is to eventually stop leaking struct page deep into the i/o >> stack. Anything that relies on being able to retrieve a struct page >> out of an sg entry needs to be converted. I think we need a new >> helper for this case "sg_phys_aligned()?". > > Why? The aim of the code is not to detect whether the address is aligned > to a page (if it were, it'd be testing for a zero s->offset, or it would > be testing for an s->offset being a multiple of the page size. > > Let's first understand the code that's being modified (which seems to be > something which isn't done very much today - people seem to just like > changing code for the hell of it.) > > for (count = 0, s = sg; count < (size >> PAGE_SHIFT); s = sg_next(s)) { > phys_addr_t phys = page_to_phys(sg_page(s)); > unsigned int len = PAGE_ALIGN(s->offset + s->length); > > if (!is_coherent && > !dma_get_attr(DMA_ATTR_SKIP_CPU_SYNC, attrs)) > __dma_page_cpu_to_dev(sg_page(s), s->offset, s->length, > dir); > > prot = __dma_direction_to_prot(dir); > > ret = iommu_map(mapping->domain, iova, phys, len, prot); > if (ret < 0) > goto fail; > count += len >> PAGE_SHIFT; > iova += len; > } > > What it's doing is trying to map each scatterlist entry via an IOMMU. > Unsurprisingly, IOMMUs are page based - you can't map a partial IOMMU > page. > > However, what says that the IOMMU page size is the same as the host CPU > page size - it may not be... so the above code is wrong for a completely > different reason. > > What we _should_ be doing is finding out what the IOMMU minimum page > size is, and using that in conjunction with the sg_phys() of the > scatterlist entry to determine the start and length of the mapping > such that the IOMMU mapping covers the range described by the scatterlist > entry. (iommu_map() takes arguments which must be aligned to the IOMMU > minimum page size.) > > However, what we can also see from the above is that we have other code > here using sg_page() - which is a necessity to be able to perform the > required DMA cache maintanence to ensure that the data is visible to the > DMA device. We need to kmap_atomic() these in order to flush them, and > there's no other way other than struct page to represent a highmem page. > > So, I think your intent to want to remove struct page from the I/O > subsystem is a false hope, unless you want to end up rewriting lots of > architecture code, and you can come up with an alternative method to > represent highmem pages. I think there will always be cases that need to do pfn_to_page() for buffer management. Those configurations will be blocked from seeing cases where a pfn is not struct page backed. So, you can have highmem or dma to pmem, but not both. Christoph, this is why I have Kconfig symbols (DEV_PFN etc) to gate whether an arch/config supports pfn-only i/o.
On Wed, Jun 10, 2015 at 09:57:06AM -0700, Dan Williams wrote: > On Wed, Jun 10, 2015 at 9:31 AM, Russell King - ARM Linux > <linux@arm.linux.org.uk> wrote: > > Why? The aim of the code is not to detect whether the address is aligned > > to a page (if it were, it'd be testing for a zero s->offset, or it would > > be testing for an s->offset being a multiple of the page size. > > > > Let's first understand the code that's being modified (which seems to be > > something which isn't done very much today - people seem to just like > > changing code for the hell of it.) > > > > for (count = 0, s = sg; count < (size >> PAGE_SHIFT); s = sg_next(s)) { > > phys_addr_t phys = page_to_phys(sg_page(s)); > > unsigned int len = PAGE_ALIGN(s->offset + s->length); > > > > if (!is_coherent && > > !dma_get_attr(DMA_ATTR_SKIP_CPU_SYNC, attrs)) > > __dma_page_cpu_to_dev(sg_page(s), s->offset, s->length, > > dir); > > > > prot = __dma_direction_to_prot(dir); > > > > ret = iommu_map(mapping->domain, iova, phys, len, prot); > > if (ret < 0) > > goto fail; > > count += len >> PAGE_SHIFT; > > iova += len; > > } > > > > What it's doing is trying to map each scatterlist entry via an IOMMU. > > Unsurprisingly, IOMMUs are page based - you can't map a partial IOMMU > > page. > > > > However, what says that the IOMMU page size is the same as the host CPU > > page size - it may not be... so the above code is wrong for a completely > > different reason. > > > > What we _should_ be doing is finding out what the IOMMU minimum page > > size is, and using that in conjunction with the sg_phys() of the > > scatterlist entry to determine the start and length of the mapping > > such that the IOMMU mapping covers the range described by the scatterlist > > entry. (iommu_map() takes arguments which must be aligned to the IOMMU > > minimum page size.) > > > > However, what we can also see from the above is that we have other code > > here using sg_page() - which is a necessity to be able to perform the > > required DMA cache maintanence to ensure that the data is visible to the > > DMA device. We need to kmap_atomic() these in order to flush them, and > > there's no other way other than struct page to represent a highmem page. > > > > So, I think your intent to want to remove struct page from the I/O > > subsystem is a false hope, unless you want to end up rewriting lots of > > architecture code, and you can come up with an alternative method to > > represent highmem pages. > > I think there will always be cases that need to do pfn_to_page() for > buffer management. Those configurations will be blocked from seeing > cases where a pfn is not struct page backed. So, you can have highmem > or dma to pmem, but not both. Christoph, this is why I have Kconfig > symbols (DEV_PFN etc) to gate whether an arch/config supports pfn-only > i/o. Hmm, pmem... yea, in the SolidRun community, we've basically decided to stick with my updated Marvell BMM layer rather than switch to pmem. I forget the reasons why, but the decision was made after looking at what pmem was doing... In any case, let's not get bogged down in a peripheral issue. What I'm objecting to is that the patches I've seen seem to be just churn without any net benefit. You can't simply make sg_page() return NULL after this change, because you've done nothing with the remaining sg_page() callers to allow them to gracefully handle that case. What I'd like to see is a much fuller series of patches which show the whole progression towards your end goal rather than a piecemeal approach. Right now, it's not clear that there is any benefit to this round of changes.
On Wed, Jun 10, 2015 at 10:13 AM, Russell King - ARM Linux <linux@arm.linux.org.uk> wrote: > On Wed, Jun 10, 2015 at 09:57:06AM -0700, Dan Williams wrote: >> On Wed, Jun 10, 2015 at 9:31 AM, Russell King - ARM Linux >> <linux@arm.linux.org.uk> wrote: >> > Why? The aim of the code is not to detect whether the address is aligned >> > to a page (if it were, it'd be testing for a zero s->offset, or it would >> > be testing for an s->offset being a multiple of the page size. >> > >> > Let's first understand the code that's being modified (which seems to be >> > something which isn't done very much today - people seem to just like >> > changing code for the hell of it.) >> > >> > for (count = 0, s = sg; count < (size >> PAGE_SHIFT); s = sg_next(s)) { >> > phys_addr_t phys = page_to_phys(sg_page(s)); >> > unsigned int len = PAGE_ALIGN(s->offset + s->length); >> > >> > if (!is_coherent && >> > !dma_get_attr(DMA_ATTR_SKIP_CPU_SYNC, attrs)) >> > __dma_page_cpu_to_dev(sg_page(s), s->offset, s->length, >> > dir); >> > >> > prot = __dma_direction_to_prot(dir); >> > >> > ret = iommu_map(mapping->domain, iova, phys, len, prot); >> > if (ret < 0) >> > goto fail; >> > count += len >> PAGE_SHIFT; >> > iova += len; >> > } >> > >> > What it's doing is trying to map each scatterlist entry via an IOMMU. >> > Unsurprisingly, IOMMUs are page based - you can't map a partial IOMMU >> > page. >> > >> > However, what says that the IOMMU page size is the same as the host CPU >> > page size - it may not be... so the above code is wrong for a completely >> > different reason. >> > >> > What we _should_ be doing is finding out what the IOMMU minimum page >> > size is, and using that in conjunction with the sg_phys() of the >> > scatterlist entry to determine the start and length of the mapping >> > such that the IOMMU mapping covers the range described by the scatterlist >> > entry. (iommu_map() takes arguments which must be aligned to the IOMMU >> > minimum page size.) >> > >> > However, what we can also see from the above is that we have other code >> > here using sg_page() - which is a necessity to be able to perform the >> > required DMA cache maintanence to ensure that the data is visible to the >> > DMA device. We need to kmap_atomic() these in order to flush them, and >> > there's no other way other than struct page to represent a highmem page. >> > >> > So, I think your intent to want to remove struct page from the I/O >> > subsystem is a false hope, unless you want to end up rewriting lots of >> > architecture code, and you can come up with an alternative method to >> > represent highmem pages. >> >> I think there will always be cases that need to do pfn_to_page() for >> buffer management. Those configurations will be blocked from seeing >> cases where a pfn is not struct page backed. So, you can have highmem >> or dma to pmem, but not both. Christoph, this is why I have Kconfig >> symbols (DEV_PFN etc) to gate whether an arch/config supports pfn-only >> i/o. > > Hmm, pmem... yea, in the SolidRun community, we've basically decided to > stick with my updated Marvell BMM layer rather than switch to pmem. I > forget the reasons why, but the decision was made after looking at what > pmem was doing... I'd of course be open to exploring if drivers/nvdimm/ could be made more generally useful. > In any case, let's not get bogged down in a peripheral issue. > > What I'm objecting to is that the patches I've seen seem to be just > churn without any net benefit. > > You can't simply make sg_page() return NULL after this change, because > you've done nothing with the remaining sg_page() callers to allow them > to gracefully handle that case. > > What I'd like to see is a much fuller series of patches which show the > whole progression towards your end goal rather than a piecemeal > approach. Right now, it's not clear that there is any benefit to > this round of changes. > Fair enough. I had them as part of a larger series [1]. Christoph suggested that I break out the pure cleanups separately. I'll add you to the next rev of that series. [1]: https://lists.01.org/pipermail/linux-nvdimm/2015-June/001094.html
On Wed, Jun 10, 2015 at 09:00:31AM -0700, Dan Williams wrote: > > So sg_phys() turns out to be 'page_to_phys(sg_page(s)) + s->offset', > > which makes the above statement to: > > > > page_to_phys(sg_page(s)) + s->offset - s->offset; > > > > The compiler will probably optimize that away, but it still doesn't look > > like an improvement. > > The goal is to eventually stop leaking struct page deep into the i/o > stack. Anything that relies on being able to retrieve a struct page > out of an sg entry needs to be converted. I think we need a new > helper for this case "sg_phys_aligned()?". You still have a reference to a struct page, because sg_phys() calls sg_page() too. If you want to get rid of sg_page() something like sg_pfn() migth be a more workable solution than sg_phys_(page_)aligned. But maybe I am just missing the bigger scope of this, so I agree with Russell that it is better so see a patch series which shows the direction you want to go with this. Joerg
diff --git a/arch/arm/mm/dma-mapping.c b/arch/arm/mm/dma-mapping.c index 7e7583ddd607..9f6ff6671f01 100644 --- a/arch/arm/mm/dma-mapping.c +++ b/arch/arm/mm/dma-mapping.c @@ -1502,7 +1502,7 @@ static int __map_sg_chunk(struct device *dev, struct scatterlist *sg, return -ENOMEM; for (count = 0, s = sg; count < (size >> PAGE_SHIFT); s = sg_next(s)) { - phys_addr_t phys = page_to_phys(sg_page(s)); + phys_addr_t phys = sg_phys(s) - s->offset; unsigned int len = PAGE_ALIGN(s->offset + s->length); if (!is_coherent && diff --git a/arch/microblaze/kernel/dma.c b/arch/microblaze/kernel/dma.c index ed7ba8a11822..dcb3c594d626 100644 --- a/arch/microblaze/kernel/dma.c +++ b/arch/microblaze/kernel/dma.c @@ -61,7 +61,7 @@ static int dma_direct_map_sg(struct device *dev, struct scatterlist *sgl, /* FIXME this part of code is untested */ for_each_sg(sgl, sg, nents, i) { sg->dma_address = sg_phys(sg); - __dma_sync(page_to_phys(sg_page(sg)) + sg->offset, + __dma_sync(sg_phys(sg), sg->length, direction); } diff --git a/drivers/iommu/intel-iommu.c b/drivers/iommu/intel-iommu.c index 68d43beccb7e..9b9ada71e0d3 100644 --- a/drivers/iommu/intel-iommu.c +++ b/drivers/iommu/intel-iommu.c @@ -1998,7 +1998,7 @@ static int __domain_mapping(struct dmar_domain *domain, unsigned long iov_pfn, sg_res = aligned_nrpages(sg->offset, sg->length); sg->dma_address = ((dma_addr_t)iov_pfn << VTD_PAGE_SHIFT) + sg->offset; sg->dma_length = sg->length; - pteval = page_to_phys(sg_page(sg)) | prot; + pteval = (sg_phys(sg) - sg->offset) | prot; phys_pfn = pteval >> VTD_PAGE_SHIFT; } @@ -3302,7 +3302,7 @@ static int intel_nontranslate_map_sg(struct device *hddev, for_each_sg(sglist, sg, nelems, i) { BUG_ON(!sg_page(sg)); - sg->dma_address = page_to_phys(sg_page(sg)) + sg->offset; + sg->dma_address = sg_phys(sg); sg->dma_length = sg->length; } return nelems; diff --git a/drivers/iommu/iommu.c b/drivers/iommu/iommu.c index d4f527e56679..59808fc9110d 100644 --- a/drivers/iommu/iommu.c +++ b/drivers/iommu/iommu.c @@ -1147,7 +1147,7 @@ size_t default_iommu_map_sg(struct iommu_domain *domain, unsigned long iova, min_pagesz = 1 << __ffs(domain->ops->pgsize_bitmap); for_each_sg(sg, s, nents, i) { - phys_addr_t phys = page_to_phys(sg_page(s)) + s->offset; + phys_addr_t phys = sg_phys(s); /* * We are mapping on IOMMU page boundaries, so offset within diff --git a/drivers/staging/android/ion/ion_chunk_heap.c b/drivers/staging/android/ion/ion_chunk_heap.c index 3e6ec2ee6802..b7da5d142aa9 100644 --- a/drivers/staging/android/ion/ion_chunk_heap.c +++ b/drivers/staging/android/ion/ion_chunk_heap.c @@ -81,7 +81,7 @@ static int ion_chunk_heap_allocate(struct ion_heap *heap, err: sg = table->sgl; for (i -= 1; i >= 0; i--) { - gen_pool_free(chunk_heap->pool, page_to_phys(sg_page(sg)), + gen_pool_free(chunk_heap->pool, sg_phys(sg) - sg->offset, sg->length); sg = sg_next(sg); } @@ -109,7 +109,7 @@ static void ion_chunk_heap_free(struct ion_buffer *buffer) DMA_BIDIRECTIONAL); for_each_sg(table->sgl, sg, table->nents, i) { - gen_pool_free(chunk_heap->pool, page_to_phys(sg_page(sg)), + gen_pool_free(chunk_heap->pool, sg_phys(sg) - sg->offset, sg->length); } chunk_heap->allocated -= allocated_size;
Coccinelle cleanup to replace open coded sg to physical address translations. This is in preparation for introducing scatterlists that reference __pfn_t. // sg_phys.cocci: convert usage page_to_phys(sg_page(sg)) to sg_phys(sg) // usage: make coccicheck COCCI=sg_phys.cocci MODE=patch virtual patch virtual report virtual org @@ struct scatterlist *sg; @@ - page_to_phys(sg_page(sg)) + sg->offset + sg_phys(sg) @@ struct scatterlist *sg; @@ - page_to_phys(sg_page(sg)) + sg_phys(sg) - sg->offset Cc: Julia Lawall <Julia.Lawall@lip6.fr> Cc: Russell King <linux@arm.linux.org.uk> Cc: Michal Simek <monstr@monstr.eu> Cc: David Woodhouse <dwmw2@infradead.org> Cc: Joerg Roedel <joro@8bytes.org> Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org> Cc: Christoph Hellwig <hch@lst.de> Signed-off-by: Dan Williams <dan.j.williams@intel.com> --- arch/arm/mm/dma-mapping.c | 2 +- arch/microblaze/kernel/dma.c | 2 +- drivers/iommu/intel-iommu.c | 4 ++-- drivers/iommu/iommu.c | 2 +- drivers/staging/android/ion/ion_chunk_heap.c | 4 ++-- 5 files changed, 7 insertions(+), 7 deletions(-)