dma-mapping: update default segment_boundary_mask
Subsystems (8)
cover dma-mapping: update default segment_boundary_mask
These two patches are to update default segment_boundary_mask.
PATCH-1 fixes overflow issues in callers of dma_get_seg_boundary. Previous version was a series: https://lkml.org/lkml/2020/8/31/1026
Then PATCH-2 sets default segment_boundary_mask to ULONG_MAX.
Nicolin Chen (2):
dma-mapping: introduce dma_get_seg_boundary_nr_pages()
dma-mapping: set default segment_boundary_mask to ULONG_MAX
arch/alpha/kernel/pci_iommu.c | 7 +------ arch/ia64/hp/common/sba_iommu.c | 3 +-- arch/powerpc/kernel/iommu.c | 9 ++------- arch/s390/pci/pci_dma.c | 6 ++---- arch/sparc/kernel/iommu-common.c | 10 +++------- arch/sparc/kernel/iommu.c | 3 +-- arch/sparc/kernel/pci_sun4v.c | 3 +-- arch/x86/kernel/amd_gart_64.c | 3 +-- drivers/parisc/ccio-dma.c | 3 +-- drivers/parisc/sba_iommu.c | 3 +-- include/linux/dma-mapping.h | 21 ++++++++++++++++++++- 11 files changed, 34 insertions(+), 37 deletions(-)
Signature · 1 line
2.17.1
PATCH 1/2 dma-mapping: introduce dma_get_seg_boundary_nr_pages()
We found that callers of dma_get_seg_boundary mostly do an ALIGN with page mask and then do a page shift to get number of pages:
ALIGN(boundary + 1, 1 << shift) >> shift
However, the boundary might be as large as ULONG_MAX, which means that a device has no specific boundary limit. So either "+ 1" or passing it to ALIGN() would potentially overflow.
According to kernel defines:
#define ALIGN_MASK(x, mask) (((x) + (mask)) & ~(mask))
#define ALIGN(x, a) ALIGN_MASK(x, (typeof(x))(a) - 1)We can simplify the logic here into a helper function doing:
ALIGN(boundary + 1, 1 << shift) >> shift
= ALIGN_MASK(b + 1, (1 << s) - 1) >> s
= {[b + 1 + (1 << s) - 1] & ~[(1 << s) - 1]} >> s = [b + 1 + (1 << s) - 1] >> s
= [b + (1 << s)] >> s
= (b >> s) + 1
This patch introduces and applies dma_get_seg_boundary_nr_pages() as an overflow-free helper for the dma_get_seg_boundary() callers to get numbers of pages. It also takes care of the NULL dev case for non-DMA API callers.
--- arch/alpha/kernel/pci_iommu.c | 7 +------ arch/ia64/hp/common/sba_iommu.c | 3 +-- arch/powerpc/kernel/iommu.c | 9 ++------- arch/s390/pci/pci_dma.c | 6 ++---- arch/sparc/kernel/iommu-common.c | 10 +++------- arch/sparc/kernel/iommu.c | 3 +-- arch/sparc/kernel/pci_sun4v.c | 3 +-- arch/x86/kernel/amd_gart_64.c | 3 +-- drivers/parisc/ccio-dma.c | 3 +-- drivers/parisc/sba_iommu.c | 3 +-- include/linux/dma-mapping.h | 19 +++++++++++++++++++ 11 files changed, 33 insertions(+), 36 deletions(-)
arch/alpha/kernel/pci_iommu.c +1 -6
@@ -141,12 +141,7 @@ iommu_arena_find_pages(struct device *dev, struct pci_iommu_arena *arena,unsigned long boundary_size; base = arena->dma_base >>PAGE_SHIFT ; -if (dev) { - boundary_size =dma_get_seg_boundary (dev) +1 ; - boundary_size >>=PAGE_SHIFT ; - }else { - boundary_size =1UL << (32 -PAGE_SHIFT ); - } + boundary_size =dma_get_seg_boundary_nr_pages (dev,PAGE_SHIFT );/* Search forward for the first mask-aligned sequence of N free ptes */ ptes = arena->ptes;
arch/ia64/hp/common/sba_iommu.c +1 -2
@@ -485,8 +485,7 @@ sba_search_bitmap(struct ioc *ioc, struct device *dev,ASSERT (((unsigned long ) ioc->res_hint & (sizeof (unsigned long ) -1UL )) ==0 );ASSERT (res_ptr < res_end); - boundary_size = (unsigned long long )dma_get_seg_boundary (dev) +1 ; - boundary_size =ALIGN (boundary_size,1ULL << iovp_shift) >> iovp_shift; + boundary_size =dma_get_seg_boundary_nr_pages (dev, iovp_shift);BUG_ON (ioc->ibase & ~iovp_mask); shift = ioc->ibase >> iovp_shift;
arch/powerpc/kernel/iommu.c +2 -7
@@ -236,15 +236,10 @@ static unsigned long iommu_range_alloc(struct device *dev, } } -if (dev) - boundary_size =ALIGN (dma_get_seg_boundary (dev) +1 , -1 << tbl->it_page_shift); -else - boundary_size =ALIGN (1UL <<32 ,1 << tbl->it_page_shift); -/* 4GB boundary for iseries_hv_alloc and iseries_hv_map */ + boundary_size =dma_get_seg_boundary_nr_pages (dev, tbl->it_page_shift); n =iommu_area_alloc (tbl->it_map, limit, start, npages, tbl->it_offset, - boundary_size >> tbl->it_page_shift, align_mask); + boundary_size, align_mask);if (n ==-1 ) {if (likely (pass ==0 )) {/* First try the pool from the start */
arch/s390/pci/pci_dma.c +2 -4
@@ -261,13 +261,11 @@ static unsigned long __dma_alloc_iommu(struct device *dev, unsigned long start, int size) {struct zpci_dev *zdev =to_zpci (to_pci_dev (dev)); -unsigned long boundary_size; - boundary_size =ALIGN (dma_get_seg_boundary (dev) +1 , -PAGE_SIZE ) >>PAGE_SHIFT ;return iommu_area_alloc (zdev->iommu_bitmap, zdev->iommu_pages, start, size, zdev->start_dma >>PAGE_SHIFT , - boundary_size,0 ); +dma_get_seg_boundary_nr_pages (dev,PAGE_SHIFT ), +0 ); }static dma_addr_t dma_alloc_address (struct device *dev,int size)
arch/sparc/kernel/iommu-common.c +3 -7
@@ -166,13 +166,6 @@ unsigned long iommu_tbl_range_alloc(struct device *dev, } } -if (dev) - boundary_size =ALIGN (dma_get_seg_boundary (dev) +1 , -1 << iommu->table_shift); -else - boundary_size =ALIGN (1ULL <<32 ,1 << iommu->table_shift); - - boundary_size = boundary_size >> iommu->table_shift;/* * if the skip_span_boundary_check had been set during init, we set * things up so that iommu_is_span_boundary() merely checks if the
@@ -181,6 +174,9 @@ unsigned long iommu_tbl_range_alloc(struct device *dev,if ((iommu->flags &IOMMU_NO_SPAN_BOUND ) !=0 ) { shift =0 ; boundary_size = iommu->poolsize * iommu->nr_pools; + }else { + boundary_size =dma_get_seg_boundary_nr_pages (dev, + iommu->table_shift); } n =iommu_area_alloc (iommu->map, limit, start, npages, shift, boundary_size, align_mask);
arch/sparc/kernel/iommu.c +1 -2
@@ -472,8 +472,7 @@ static int dma_4u_map_sg(struct device *dev, struct scatterlist *sglist, outs->dma_length =0 ; max_seg_size =dma_get_max_seg_size (dev); - seg_boundary_size =ALIGN (dma_get_seg_boundary (dev) +1 , -IO_PAGE_SIZE ) >>IO_PAGE_SHIFT ; + seg_boundary_size =dma_get_seg_boundary_nr_pages (dev,IO_PAGE_SHIFT ); base_shift = iommu->tbl.table_map_base >>IO_PAGE_SHIFT ;for_each_sg (sglist, s, nelems, i) {unsigned long paddr, npages, entry, out_entry =0 , slen;
arch/sparc/kernel/pci_sun4v.c +1 -2
@@ -508,8 +508,7 @@ static int dma_4v_map_sg(struct device *dev, struct scatterlist *sglist,iommu_batch_start (dev, prot, ~0UL ); max_seg_size =dma_get_max_seg_size (dev); - seg_boundary_size =ALIGN (dma_get_seg_boundary (dev) +1 , -IO_PAGE_SIZE ) >>IO_PAGE_SHIFT ; + seg_boundary_size =dma_get_seg_boundary_nr_pages (dev,IO_PAGE_SHIFT ); mask = *dev->dma_mask;if (!iommu_use_atu (iommu, mask))
arch/x86/kernel/amd_gart_64.c +1 -2
@@ -96,8 +96,7 @@ static unsigned long alloc_iommu(struct device *dev, int size, base_index =ALIGN (iommu_bus_base &dma_get_seg_boundary (dev),PAGE_SIZE ) >>PAGE_SHIFT ; - boundary_size =ALIGN ((u64 )dma_get_seg_boundary (dev) +1 , -PAGE_SIZE ) >>PAGE_SHIFT ; + boundary_size =dma_get_seg_boundary_nr_pages (dev,PAGE_SHIFT );spin_lock_irqsave (&iommu_bitmap_lock, flags); offset = iommu_area_alloc(iommu_gart_bitmap, iommu_pages, next_bit,
drivers/parisc/ccio-dma.c +1 -2
@@ -356,8 +356,7 @@ ccio_alloc_range(struct ioc *ioc, struct device *dev, size_t size)** ggg sacrifices another 710 to the computer gods. */ - boundary_size =ALIGN ((unsigned long long )dma_get_seg_boundary (dev) +1 , -1ULL <<IOVP_SHIFT ) >>IOVP_SHIFT ; + boundary_size =dma_get_seg_boundary_nr_pages (dev,IOVP_SHIFT );if (pages_needed <=8 ) {/*
drivers/parisc/sba_iommu.c +1 -2
@@ -342,8 +342,7 @@ sba_search_bitmap(struct ioc *ioc, struct device *dev,unsigned long shift;int ret; - boundary_size =ALIGN ((unsigned long long )dma_get_seg_boundary (dev) +1 , -1ULL <<IOVP_SHIFT ) >>IOVP_SHIFT ; + boundary_size =dma_get_seg_boundary_nr_pages (dev,IOVP_SHIFT );#if defined(ZX1_SUPPORT )BUG_ON (ioc->ibase & ~IOVP_MASK );
include/linux/dma-mapping.h +19 -0
@@ -632,6 +632,25 @@ static inline unsigned long dma_get_seg_boundary(struct device *dev)return DMA_BIT_MASK (32 ); } +/** +* dma_get_seg_boundary_nr_pages - return the segment boundary in "page" units +* @dev: device to guery the boundary for +* @page_shift: ilog() of the IOMMU page size +* +* Return the segment boundary in IOMMU page units (which may be different from +* the CPU page size) for the passed in device. +* +* If @dev is NULL a boundary of U32_MAX is assumed, this case is just for +* non-DMA API callers. +*/ +static inline unsigned long dma_get_seg_boundary_nr_pages (struct device *dev, +unsigned int page_shift) +{ +if (!dev) +return (U32_MAX >> page_shift) +1 ; +return (dma_get_seg_boundary (dev) >> page_shift) +1 ; +} +static inline int dma_set_seg_boundary (struct device *dev,unsigned long mask) {if (dev->dma_parms) {
Signature · 1 line
2.17.1
Re: [PATCH 1/2] dma-mapping: introduce dma_get_seg_boundary_nr_pages()
Quoted 9 lines · Nicolin Chen
arch/powerpc/kernel/iommu.c +0
-1@@ -236,15 +236,10 @@ static unsigned long iommu_range_alloc(struct device *dev, } } -if (dev)
arch/powerpc/kernel/iommu.c static unsigned long iommu_range_alloc(struct device *dev, · line 242
Quoted 10 lines
arch/powerpc/kernel/iommu.c +2
-6- boundary_size =ALIGN (dma_get_seg_boundary (dev) +1 , -1 << tbl->it_page_shift); - else - boundary_size =ALIGN (1UL <<32 ,1 << tbl->it_page_shift); -/* 4GB boundary for iseries_hv_alloc and iseries_hv_map */ + boundary_size =dma_get_seg_boundary_nr_pages (dev, tbl->it_page_shift); n =iommu_area_alloc (tbl->it_map, limit, start, npages, tbl->it_offset, - boundary_size >> tbl->it_page_shift, align_mask); + boundary_size, align_mask);
This has changed the units of boundary_size, but it's unused elsewhere in the function so that's OK.
If you need to do a v2 for any other reason, then I'd just drop boundary_size and call dma_get_seg_boundary_nr_pages() directly in the function call.
cheers
Re: [PATCH 1/2] dma-mapping: introduce dma_get_seg_boundary_nr_pages()
In reply to Nicolin ChenQuoted 25 lines · Nicolin Chen
We found that callers of
dma_get_seg_boundarymostly do an ALIGN with page mask and then do a page shift to get number of pages:
ALIGN(boundary + 1, 1 << shift)>> shiftHowever, the boundary might be as large as
ULONG_MAX, which means that a device has no specific boundary limit. So either "+ 1" or passing it toALIGN()would potentially overflow.According to kernel defines:
#define ALIGN_MASK(x, mask) (((x) + (mask)) & ~(mask)) #define ALIGN(x, a) ALIGN_MASK(x, (typeof(x))(a) - 1)We can simplify the logic here into a helper function doing:
ALIGN(boundary + 1, 1 << shift)>> shift
=ALIGN_MASK(b + 1, (1 << s) - 1)>> s
= {[b + 1 + (1 << s) - 1] & ~[(1 << s) - 1]} >> s = [b + 1 + (1 << s) - 1] >> s
= [b + (1 << s)] >> s
= (b >> s) + 1This patch introduces and applies
dma_get_seg_boundary_nr_pages()as an overflow-free helper for thedma_get_seg_boundary()callers to get numbers of pages. It also takes care of the NULL dev case for non-DMA API callers.
...
include/linux/dma-mapping.h static inline unsigned long dma_get_seg_boundary(struct device *dev) · line 651
Quoted 6 lines
include/linux/dma-mapping.h +6
-0+static inline unsigned long dma_get_seg_boundary_nr_pages (struct device *dev, +unsigned int page_shift) +{ +if (!dev) +return (U32_MAX >> page_shift) +1 ; +return (dma_get_seg_boundary (dev) >> page_shift) +1 ;
Can it be better to do something like
unsigned long boundary = dev ? dma_get_seg_boundary(dev) : U32_MAX;
return (boundary >> page_shift) + 1;
?
Quoted 1 line
+}
Signature · 2 lines
With Best Regards, Andy Shevchenko
Re: [PATCH 1/2] dma-mapping: introduce dma_get_seg_boundary_nr_pages()
Quoted 11 lines · Andy Shevchenko
include/linux/dma-mapping.h static inline unsigned long dma_get_seg_boundary(struct device *dev) · line 651
Quoted 4 lines
include/linux/dma-mapping.h +4
-0+{ +if (!dev) +return (U32_MAX >> page_shift) +1 ; +return (dma_get_seg_boundary (dev) >> page_shift) +1 ;Can it be better to do something like
unsigned long boundary = dev ?dma_get_seg_boundary(dev):U32_MAX;return (boundary >> page_shift) + 1;?
I don't really see what that would buy us.
Re: [PATCH 1/2] dma-mapping: introduce dma_get_seg_boundary_nr_pages()
Le 03/09/2020 à 18:12, Christoph Hellwig a écrit :
Quoted 6 lines · Andy Shevchenko
Quoted 5 lines
Quoted 4 lines
include/linux/dma-mapping.h +4
-0+{ +if (!dev) +return (U32_MAX >> page_shift) +1 ; +return (dma_get_seg_boundary (dev) >> page_shift) +1 ;
Quoted 9 lines
Quoted 6 lines
Can it be better to do something like
unsigned long boundary = dev ?dma_get_seg_boundary(dev):U32_MAX;return (boundary >> page_shift) + 1;?
I don't really see what that would buy us.
I guess it would avoid the duplication of >> page_shift) + 1
Christophe
PATCH 2/2 dma-mapping: set default segment_boundary_mask to ULONG_MAX
In reply to Nicolin ChenThe default segment_boundary_mask was set to DMA_BIT_MAKS(32) a decade ago by referencing SCSI/block subsystem, as a 32-bit mask was good enough for most of the devices.
Now more and more drivers set dma_masks above DMA_BIT_MAKS(32) while only a handful of them call dma_set_seg_boundary(). This means that most drivers have a 4GB segmention boundary because DMA API returns a 32-bit default value, though they might not really have such a limit.
The default segment_boundary_mask should mean "no limit" since the device doesn't explicitly set the mask. But a 32-bit mask certainly limits those devices capable of 32+ bits addressing.
So this patch sets default segment_boundary_mask to ULONG_MAX.
--- include/linux/dma-mapping.h | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
include/linux/dma-mapping.h +1 -1
@@ -629,7 +629,7 @@ static inline unsigned long dma_get_seg_boundary(struct device *dev) {if (dev->dma_parms && dev->dma_parms->segment_boundary_mask)return dev->dma_parms->segment_boundary_mask; -return DMA_BIT_MASK (32 ); +return ULONG_MAX ; }/**
Signature · 1 line
2.17.1
Quoted 10 lines · Nicolin Chen
These two patches are to update default
segment_boundary_mask.PATCH-1 fixes overflow issues in callers of
dma_get_seg_boundary. Previous version was a series: https://lkml.org/lkml/2020/8/31/1026Then PATCH-2 sets default
segment_boundary_masktoULONG_MAX.Nicolin Chen (2):
dma-mapping: introducedma_get_seg_boundary_nr_pages()
dma-mapping: set defaultsegment_boundary_masktoULONG_MAX
I gave both of your patches a quick test ride on a couple of dev mainframes, both NVMe, ConnectX and virtio-pci devices all seems to work fine. I already commented on Christoph's mail that I like the helper approach, so as for s390 you can add my
Quoted 14 lines
arch/alpha/kernel/pci_iommu.c | 7 +------ arch/ia64/hp/common/sba_iommu.c | 3 +-- arch/powerpc/kernel/iommu.c | 9 ++------- arch/s390/pci/pci_dma.c | 6 ++---- arch/sparc/kernel/iommu-common.c | 10 +++------- arch/sparc/kernel/iommu.c | 3 +-- arch/sparc/kernel/pci_sun4v.c | 3 +-- arch/x86/kernel/amd_gart_64.c | 3 +-- drivers/parisc/ccio-dma.c | 3 +-- drivers/parisc/sba_iommu.c | 3 +-- include/linux/dma-mapping.h | 21 ++++++++++++++++++++- 11 files changed, 34 insertions(+), 37 deletions(-)
Quoted 12 lines · Niklas Schnelle
Quoted 10 lines · Nicolin Chen
These two patches are to update default
segment_boundary_mask.PATCH-1 fixes overflow issues in callers of
dma_get_seg_boundary. Previous version was a series: https://lkml.org/lkml/2020/8/31/1026Then PATCH-2 sets default
segment_boundary_masktoULONG_MAX.Nicolin Chen (2):
dma-mapping: introducedma_get_seg_boundary_nr_pages()
dma-mapping: set defaultsegment_boundary_masktoULONG_MAX
Quoted 6 lines
I gave both of your patches a quick test ride on a couple of dev mainframes, both NVMe, ConnectX and virtio-pci devices all seems to work fine. I already commented on Christoph's mail that I like the helper approach, so as for s390 you can add my
Acked-by Niklas Schnelle <[email protected]>
Thanks for testing and the ack!
Applied with the recommendation from Michael folded in.