Skip to contentSkip to conversation

iommu/vt-d: boost the mapping process

PATCH Superseded Longpeng(Mike) 2 participants 5 messages Oct 5, 2021 – Oct 7, 2021

Subsystems (2)

v2 of 3 Superseded Latest is v3 →

lore
Share & tools
lore
#0 lore raw

v2 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.

Changes v1 -> v2:
- Fix compile warning on i386 [Baolu]

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
#1 lore raw

PATCH v2 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
index 05a65eb..a590b00 100644
@@ -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
maintainer
#2 lore raw

Re: [PATCH v2 1/2] iommu/vt-d: convert the return type of first_pte_in_page to bool

Quoted 14 lines · Longpeng(Mike)

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 +0 -0
index 05a65eb..a590b00 100644

include/linux/intel-iommu.h static inline bool dma_pte_superpage(struct dma_pte *pte) · line 714

Quoted 8 lines
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);
  }

Probably,

	return IS_ALIGNED((unsigned long)pte, VTD_PAGE_SIZE);

looks neater?

Best regards,
baolu
_______________________________________________
iommu mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/iommu

#3 lore raw

RE: [PATCH v2 1/2] iommu/vt-d: convert the return type of first_pte_in_page to bool

Quoted 39 lines

-----Original Message-----
From: Lu Baolu [mailto:[email protected]]
Sent: Thursday, October 7, 2021 2:18 PM
To: Longpeng (Mike, Cloud Infrastructure Service Product Dept.) <[email protected]>; [email protected]; [email protected]; [email protected]
Cc: [email protected]; [email protected]; [email protected]; Gonglei (Arei) <[email protected]>
Subject: Re: [PATCH v2 1/2] iommu/vt-d: convert the return type of first_pte_in_page to bool

Quoted 14 lines · Longpeng(Mike)

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 +0 -0
index 05a65eb..a590b00 100644

include/linux/intel-iommu.h static inline bool dma_pte_superpage(struct dma_pte *pte) · line 714

Quoted 8 lines
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);
  }

Probably,

Quoted 4 lines
	return IS_ALIGNED((unsigned long)pte, VTD_PAGE_SIZE);

looks neater?

Looks better! I'll include this optimization in v3.

Quoted 2 lines

Best regards,
baolu

#4 lore raw

PATCH v2 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 this 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
index d75f59a..87cbf34 100644
@@ -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
index a590b00..623b407 100644
@@ -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 *)ALIGN((unsigned long)pte, VTD_PAGE_SIZE) - 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 · 6 lines
1.8.3.1

_______________________________________________
iommu mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/iommu