iommu/vt-d: boost the mapping process
Subsystems (2)
cover iommu/vt-d: boost the mapping process
Hi guys,
We found that the __domain_mapping() would take too long when the memory region is too large, we try to make it faster in this patchset. The performance number can be found in PATCH 2, please review when you free, thanks.
Longpeng(Mike) (2):
iommu/vt-d: convert the return type of first_pte_in_page to bool iommu/vt-d: avoid duplicated removing in __domain_mapping
drivers/iommu/intel/iommu.c | 12 +++++++----- include/linux/intel-iommu.h | 8 +++++++- 2 files changed, 14 insertions(+), 6 deletions(-)
Signature · 1 line
1.8.3.1
PATCH 1/2 iommu/vt-d: convert the return type of first_pte_in_page to bool
first_pte_in_page() returns boolean value, so let's convert its return type to bool.
--- include/linux/intel-iommu.h | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
include/linux/intel-iommu.h +1 -1
@@ -708,7 +708,7 @@ static inline bool dma_pte_superpage(struct dma_pte *pte)return (pte->val &DMA_PTE_LARGE_PAGE ); } -static inline int first_pte_in_page (struct dma_pte *pte) +static inline bool first_pte_in_page (struct dma_pte *pte) {return !((unsigned long )pte & ~VTD_PAGE_MASK ); }
Signature · 1 line
1.8.3.1
PATCH 2/2 iommu/vt-d: avoid duplicated removing in __domain_mapping
In reply to Longpeng(Mike)__domain_mapping() always removes the pages in the range from 'iov_pfn' to 'end_pfn', but the 'end_pfn' is always the last pfn of the range that the caller wants to map.
This would introduce too many duplicated removing and leads the map operation take too long, for example:
Map iova=0x100000,nr_pages=0x7d61800
iov_pfn: 0x100000, end_pfn: 0x7e617ff
iov_pfn: 0x140000, end_pfn: 0x7e617ff
iov_pfn: 0x180000, end_pfn: 0x7e617ff
iov_pfn: 0x1c0000, end_pfn: 0x7e617ff
iov_pfn: 0x200000, end_pfn: 0x7e617ff
...it takes about 50ms in total.
We can reduce the cost by recalculate the 'end_pfn' and limit it to the boundary of the end of the pte page.
Map iova=0x100000,nr_pages=0x7d61800
iov_pfn: 0x100000, end_pfn: 0x13ffff
iov_pfn: 0x140000, end_pfn: 0x17ffff
iov_pfn: 0x180000, end_pfn: 0x1bffff
iov_pfn: 0x1c0000, end_pfn: 0x1fffff
iov_pfn: 0x200000, end_pfn: 0x23ffff
...it only need 9ms now.
--- drivers/iommu/intel/iommu.c | 12 +++++++----- include/linux/intel-iommu.h | 6 ++++++ 2 files changed, 13 insertions(+), 5 deletions(-)
drivers/iommu/intel/iommu.c +7 -5
@@ -2354,12 +2354,18 @@ static void switch_to_super_page(struct dmar_domain *domain,return -ENOMEM ; first_pte = pte; + lvl_pages =lvl_to_nr_pages (largepage_lvl); +BUG_ON (nr_pages < lvl_pages); +/* It is large page*/ if (largepage_lvl >1 ) {unsigned long end_pfn; +unsigned long pages_to_remove; pteval |=DMA_PTE_LARGE_PAGE ; - end_pfn = ((iov_pfn + nr_pages) &level_mask (largepage_lvl)) -1 ; + pages_to_remove = min_t(unsigned long, nr_pages, +nr_pte_to_next_page (pte) * lvl_pages); + end_pfn = iov_pfn + pages_to_remove -1 ;switch_to_super_page (domain, iov_pfn, end_pfn, largepage_lvl); }else { pteval &= ~(uint64_t )DMA_PTE_LARGE_PAGE ;
@@ -2381,10 +2387,6 @@ static void switch_to_super_page(struct dmar_domain *domain,WARN_ON (1 ); } - lvl_pages =lvl_to_nr_pages (largepage_lvl); - -BUG_ON (nr_pages < lvl_pages); - nr_pages -= lvl_pages; iov_pfn += lvl_pages; phys_pfn += lvl_pages;
include/linux/intel-iommu.h +6 -0
@@ -713,6 +713,12 @@ static inline bool first_pte_in_page(struct dma_pte *pte)return !((unsigned long )pte & ~VTD_PAGE_MASK ); } +static inline int nr_pte_to_next_page (struct dma_pte *pte) +{ +return first_pte_in_page (pte) ?BIT_ULL (VTD_STRIDE_SHIFT ) : + (struct dma_pte *)VTD_PAGE_ALIGN ((unsigned long )pte) - pte; +} +extern struct dmar_drhd_unit *dmar_find_matched_drhd_unit (struct pci_dev *dev);extern int dmar_find_matched_atsr_unit (struct pci_dev *dev);
Signature · 1 line
1.8.3.1
Lu Baolu
maintainerRe: [PATCH v1 2/2] iommu/vt-d: avoid duplicated removing in __domain_mapping
Hi Longpeng,
Quoted 28 lines · Longpeng(Mike)
__domain_mapping()always removes the pages in the range from 'iov_pfn' to 'end_pfn', but the 'end_pfn' is always the last pfn of the range that the caller wants to map.This would introduce too many duplicated removing and leads the map operation take too long, for example:
Map iova=0x100000,nr_pages=0x7d61800 iov_pfn: 0x100000, end_pfn: 0x7e617ff iov_pfn: 0x140000, end_pfn: 0x7e617ff iov_pfn: 0x180000, end_pfn: 0x7e617ff iov_pfn: 0x1c0000, end_pfn: 0x7e617ff iov_pfn: 0x200000, end_pfn: 0x7e617ff ... it takes about 50ms in total.We can reduce the cost by recalculate the '
end_pfn' and limit it to the boundary of the end of the pte page.Map iova=0x100000,nr_pages=0x7d61800 iov_pfn: 0x100000, end_pfn: 0x13ffff iov_pfn: 0x140000, end_pfn: 0x17ffff iov_pfn: 0x180000, end_pfn: 0x1bffff iov_pfn: 0x1c0000, end_pfn: 0x1fffff iov_pfn: 0x200000, end_pfn: 0x23ffff ... it only need 9ms now.
Quoted 1 line
Signed-off-by Longpeng(Mike) <[email protected]>
The 0day robot reports below compiling warning when building a 32-bit kernel (make W=1 ARCH=i386):
All errors (new ones prefixed by >>):
In file included from drivers/gpu/drm/i915/i915_drv.h:43,
fromdrivers/gpu/drm/i915/display/intel_display_types.h:47,
from drivers/gpu/drm/i915/display/intel_dsi.h:30,
from <command-line>: include/linux/intel-iommu.h: In function 'nr_pte_to_next_page':
Quoted 1 line
Quoted 1 line
include/linux/intel-iommu.h:719:3: error: cast to pointer from integer of different size [-Werror=int-to-pointer-cast]
719 | (struct dma_pte *)VTD_PAGE_ALIGN((unsigned long)pte) - pte;
| ^cc1: all warnings being treated as errors
vim +719 include/linux/intel-iommu.h
715
716 static inline int nr_pte_to_next_page(struct dma_pte *pte)
717 {
718 return first_pte_in_page(pte) ? BIT_ULL(VTD_STRIDE_SHIFT) :
Quoted 1 line
719 (struct
dma_pte*)VTD_PAGE_ALIGN((unsigned long)pte) - pte;
720 }
721
Can you please take a look at this?
Best regards,
baolu
Quoted 48 lines
--- drivers/iommu/intel/iommu.c | 12 +++++++----- include/linux/intel-iommu.h | 6 ++++++ 2 files changed, 13 insertions(+), 5 deletions(-)drivers/iommu/intel/iommu.c +7
-5@@ -2354,12 +2354,18 @@ static void switch_to_super_page(struct dmar_domain *domain,return -ENOMEM ; first_pte = pte; + lvl_pages =lvl_to_nr_pages (largepage_lvl); +BUG_ON (nr_pages < lvl_pages); +/* It is large page*/ if (largepage_lvl >1 ) {unsigned long end_pfn; +unsigned long pages_to_remove; pteval |=DMA_PTE_LARGE_PAGE ; - end_pfn = ((iov_pfn + nr_pages) &level_mask (largepage_lvl)) -1 ; + pages_to_remove = min_t(unsigned long, nr_pages, +nr_pte_to_next_page (pte) * lvl_pages); + end_pfn = iov_pfn + pages_to_remove -1 ;switch_to_super_page (domain, iov_pfn, end_pfn, largepage_lvl); }else { pteval &= ~(uint64_t )DMA_PTE_LARGE_PAGE ;@@ -2381,10 +2387,6 @@ static void switch_to_super_page(struct dmar_domain *domain,WARN_ON (1 ); } - lvl_pages =lvl_to_nr_pages (largepage_lvl); - -BUG_ON (nr_pages < lvl_pages); - nr_pages -= lvl_pages; iov_pfn += lvl_pages; phys_pfn += lvl_pages;include/linux/intel-iommu.h +0
-0@@ -713,6 +713,12 @@ static inline bool first_pte_in_page(struct dma_pte *pte)return !((unsigned long )pte & ~VTD_PAGE_MASK ); }
include/linux/intel-iommu.h static inline bool first_pte_in_page(struct dma_pte *pte) · line 724
Quoted 10 lines
include/linux/intel-iommu.h +6
-0+static inline int nr_pte_to_next_page (struct dma_pte *pte) +{ +return first_pte_in_page (pte) ?BIT_ULL (VTD_STRIDE_SHIFT ) : + (struct dma_pte *)VTD_PAGE_ALIGN ((unsigned long )pte) - pte; +} +extern struct dmar_drhd_unit *dmar_find_matched_drhd_unit (struct pci_dev *dev);extern int dmar_find_matched_atsr_unit (struct pci_dev *dev);
_______________________________________________
iommu mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/iommu