Re: [PATCH v6 05/18] asm-generic/tlb: Provide generic tlb_flush() based on flush_tlb_mm()

[Will Deacon <will.deacon@arm.com>]

On Tue, Feb 19, 2019 at 11:31:53AM +0100, Peter Zijlstra wrote:
> When an architecture does not have (an efficient) flush_tlb_range(),
> but instead always uses full TLB invalidates, the current generic
> tlb_flush() is sub-optimal, for it will generate extra flushes in
> order to keep the range small.
> 
> But if we cannot do range flushes, that is a moot concern. Optionally
> provide this simplified default.
> 
> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
> ---
>  include/asm-generic/tlb.h |   41 ++++++++++++++++++++++++++++++++++++++++-
>  1 file changed, 40 insertions(+), 1 deletion(-)
> 
> --- a/include/asm-generic/tlb.h
> +++ b/include/asm-generic/tlb.h
> @@ -114,7 +114,8 @@
>   *    returns the smallest TLB entry size unmapped in this range.
>   *
>   * If an architecture does not provide tlb_flush() a default implementation
> - * based on flush_tlb_range() will be used.
> + * based on flush_tlb_range() will be used, unless MMU_GATHER_NO_RANGE is
> + * specified, in which case we'll default to flush_tlb_mm().
>   *
>   * Additionally there are a few opt-in features:
>   *
> @@ -140,6 +141,9 @@
>   *  the page-table pages. Required if you use HAVE_RCU_TABLE_FREE and your
>   *  architecture uses the Linux page-tables natively.
>   *
> + *  MMU_GATHER_NO_RANGE
> + *
> + *  Use this if your architecture lacks an efficient flush_tlb_range().
>   */
>  #define HAVE_GENERIC_MMU_GATHER
>  
> @@ -302,12 +306,45 @@ static inline void __tlb_reset_range(str
>  	 */
>  }
>  
> +#ifdef CONFIG_MMU_GATHER_NO_RANGE
> +
> +#if defined(tlb_flush) || defined(tlb_start_vma) || defined(tlb_end_vma)
> +#error MMU_GATHER_NO_RANGE relies on default tlb_flush(), tlb_start_vma() and tlb_end_vma()
> +#endif
> +
> +/*
> + * When an architecture does not have efficient means of range flushing TLBs
> + * there is no point in doing intermediate flushes on tlb_end_vma() to keep the
> + * range small. We equally don't have to worry about page granularity or other
> + * things.
> + *
> + * All we need to do is issue a full flush for any !0 range.
> + */
> +static inline void tlb_flush(struct mmu_gather *tlb)
> +{
> +	if (tlb->end)
> +		flush_tlb_mm(tlb->mm);
> +}

I guess another way we could handle these architectures is by
unconditionally resetting tlb->fullmm to 1, but this works too.

Acked-by: Will Deacon <will.deacon@arm.com>

Will