If we support BBM level 2, we can potentially avoid an intermediate TLB invalidation, as hardware is capable of managing the TLB itself in this situation. Hardware will either silently clear out the offending entry, or will take a TLB Conflict Abort Exception. Note that such aborts should not occur on Arm hardware and indeed were not seen on any of the benchmarked systems. Eliding the invalidation results in a 12% improvement on a microbenchmark which targeted the worst case of contpte_convert(), which represents an 80% reduction in the overhead of contpte_convert(). Note also that this patch is pending review to ensure that it is architecturally valid, and we are working with Arm architects to validate this patch. Signed-off-by: Mikołaj Lenczewski <miko.lenczewski@xxxxxxx> --- arch/arm64/mm/contpte.c | 3 --- 1 file changed, 3 deletions(-) diff --git a/arch/arm64/mm/contpte.c b/arch/arm64/mm/contpte.c index fc927be800ee..009690770415 100644 --- a/arch/arm64/mm/contpte.c +++ b/arch/arm64/mm/contpte.c @@ -72,9 +72,6 @@ static void contpte_convert(struct mm_struct *mm, unsigned long addr, __flush_tlb_range(&vma, start_addr, addr, PAGE_SIZE, true, 3); __set_ptes(mm, start_addr, start_ptep, pte, CONT_PTES); - - if (system_supports_bbml2()) - __flush_tlb_range(&vma, start_addr, addr, PAGE_SIZE, true, 3); } void __contpte_try_fold(struct mm_struct *mm, unsigned long addr, -- 2.45.2
