diff options
author | David Woodhouse <David.Woodhouse@intel.com> | 2009-06-28 13:52:20 -0400 |
---|---|---|
committer | David Woodhouse <David.Woodhouse@intel.com> | 2009-06-29 08:39:17 -0400 |
commit | 310a5ab93cb4ce29367238f682affd9ac352f4d0 (patch) | |
tree | dc5bb8f208778ebe0aee0054b7af516a36e2b543 | |
parent | c5395d5c4a82159889cb650de93b591ea51d8c56 (diff) |
intel-iommu: Performance improvement for dma_pte_clear_range()
It's a bit silly to repeatedly call domain_flush_cache() for each PTE
individually, as we clear it. Instead, batch them up and flush a whole
range at a time. We might as well refrain from recalculating the PTE
address from scratch each time round the loop too.
Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>
-rw-r--r-- | drivers/pci/intel-iommu.c | 31 |
1 files changed, 15 insertions, 16 deletions
diff --git a/drivers/pci/intel-iommu.c b/drivers/pci/intel-iommu.c index c5caf7d63a0f..ba7e37f7111a 100644 --- a/drivers/pci/intel-iommu.c +++ b/drivers/pci/intel-iommu.c | |||
@@ -761,34 +761,33 @@ static struct dma_pte *dma_pfn_level_pte(struct dmar_domain *domain, | |||
761 | return NULL; | 761 | return NULL; |
762 | } | 762 | } |
763 | 763 | ||
764 | /* clear one page's page table */ | ||
765 | static void dma_pte_clear_one(struct dmar_domain *domain, unsigned long pfn) | ||
766 | { | ||
767 | struct dma_pte *pte = NULL; | ||
768 | |||
769 | /* get last level pte */ | ||
770 | pte = dma_pfn_level_pte(domain, pfn, 1); | ||
771 | |||
772 | if (pte) { | ||
773 | dma_clear_pte(pte); | ||
774 | domain_flush_cache(domain, pte, sizeof(*pte)); | ||
775 | } | ||
776 | } | ||
777 | |||
778 | /* clear last level pte, a tlb flush should be followed */ | 764 | /* clear last level pte, a tlb flush should be followed */ |
779 | static void dma_pte_clear_range(struct dmar_domain *domain, | 765 | static void dma_pte_clear_range(struct dmar_domain *domain, |
780 | unsigned long start_pfn, | 766 | unsigned long start_pfn, |
781 | unsigned long last_pfn) | 767 | unsigned long last_pfn) |
782 | { | 768 | { |
783 | int addr_width = agaw_to_width(domain->agaw) - VTD_PAGE_SHIFT; | 769 | int addr_width = agaw_to_width(domain->agaw) - VTD_PAGE_SHIFT; |
770 | struct dma_pte *first_pte, *pte; | ||
784 | 771 | ||
785 | BUG_ON(addr_width < BITS_PER_LONG && start_pfn >> addr_width); | 772 | BUG_ON(addr_width < BITS_PER_LONG && start_pfn >> addr_width); |
786 | BUG_ON(addr_width < BITS_PER_LONG && last_pfn >> addr_width); | 773 | BUG_ON(addr_width < BITS_PER_LONG && last_pfn >> addr_width); |
787 | 774 | ||
788 | /* we don't need lock here; nobody else touches the iova range */ | 775 | /* we don't need lock here; nobody else touches the iova range */ |
789 | while (start_pfn <= last_pfn) { | 776 | while (start_pfn <= last_pfn) { |
790 | dma_pte_clear_one(domain, start_pfn); | 777 | first_pte = pte = dma_pfn_level_pte(domain, start_pfn, 1); |
791 | start_pfn++; | 778 | if (!pte) { |
779 | start_pfn = align_to_level(start_pfn + 1, 2); | ||
780 | continue; | ||
781 | } | ||
782 | while (start_pfn <= last_pfn && | ||
783 | (unsigned long)pte >> VTD_PAGE_SHIFT == | ||
784 | (unsigned long)first_pte >> VTD_PAGE_SHIFT) { | ||
785 | dma_clear_pte(pte); | ||
786 | start_pfn++; | ||
787 | pte++; | ||
788 | } | ||
789 | domain_flush_cache(domain, first_pte, | ||
790 | (void *)pte - (void *)first_pte); | ||
792 | } | 791 | } |
793 | } | 792 | } |
794 | 793 | ||