[PATCH 2/3] Add swiotlb offset preserving mapping when dma_dma_parameters->page_offset_mask is non zero.

[Konrad Rzeszutek Wilk]

On Wed, Jan 27, 2021 at 04:38:28PM -0800, Jianxiong Gao wrote:
> For devices that need to preserve address offset on mapping through
> swiotlb, this patch adds offset preserving based on page_offset_mask
> and keeps the offset if the mask is non zero. This is needed for
> device drivers like NVMe.

<scratches his head>

Didn't you send this patch like a month ago and someone pointed
out that the right fix would be in the NVMe driver?

Is there an issue with fixing the NVMe driver?

> 
> Signed-off-by: Jianxiong Gao <jxgao at google.com>
> ---
>  kernel/dma/swiotlb.c | 25 ++++++++++++++++++++++---
>  1 file changed, 22 insertions(+), 3 deletions(-)
> 
> diff --git a/kernel/dma/swiotlb.c b/kernel/dma/swiotlb.c
> index 7c42df6e6100..4cab35f2c9bc 100644
> --- a/kernel/dma/swiotlb.c
> +++ b/kernel/dma/swiotlb.c
> @@ -468,7 +468,7 @@ phys_addr_t swiotlb_tbl_map_single(struct device *hwdev, phys_addr_t orig_addr,
>  	dma_addr_t tbl_dma_addr = phys_to_dma_unencrypted(hwdev, io_tlb_start);
>  	unsigned long flags;
>  	phys_addr_t tlb_addr;
> -	unsigned int nslots, stride, index, wrap;
> +	unsigned int nslots, stride, index, wrap, page_offset_mask, page_offset;
>  	int i;
>  	unsigned long mask;
>  	unsigned long offset_slots;
> @@ -500,12 +500,16 @@ phys_addr_t swiotlb_tbl_map_single(struct device *hwdev, phys_addr_t orig_addr,
>  		    ? ALIGN(mask + 1, 1 << IO_TLB_SHIFT) >> IO_TLB_SHIFT
>  		    : 1UL << (BITS_PER_LONG - IO_TLB_SHIFT);
>  
> +	page_offset_mask = dma_get_page_offset_mask(hwdev);
> +	page_offset = orig_addr & page_offset_mask;
> +	alloc_size += page_offset;
> +
>  	/*
>  	 * For mappings greater than or equal to a page, we limit the stride
>  	 * (and hence alignment) to a page size.
>  	 */
>  	nslots = ALIGN(alloc_size, 1 << IO_TLB_SHIFT) >> IO_TLB_SHIFT;
> -	if (alloc_size >= PAGE_SIZE)
> +	if ((alloc_size >= PAGE_SIZE) || (page_offset_mask > (1 << IO_TLB_SHIFT)))
>  		stride = (1 << (PAGE_SHIFT - IO_TLB_SHIFT));
>  	else
>  		stride = 1;
> @@ -583,6 +587,11 @@ phys_addr_t swiotlb_tbl_map_single(struct device *hwdev, phys_addr_t orig_addr,
>  	 */
>  	for (i = 0; i < nslots; i++)
>  		io_tlb_orig_addr[index+i] = orig_addr + (i << IO_TLB_SHIFT);
> +	/*
> +	 * When keeping the offset of the original data, we need to advance
> +	 * the tlb_addr by the offset of orig_addr.
> +	 */
> +	tlb_addr += page_offset;
>  	if (!(attrs & DMA_ATTR_SKIP_CPU_SYNC) &&
>  	    (dir == DMA_TO_DEVICE || dir == DMA_BIDIRECTIONAL))
>  		swiotlb_bounce(orig_addr, tlb_addr, mapping_size, DMA_TO_DEVICE);
> @@ -598,7 +607,9 @@ void swiotlb_tbl_unmap_single(struct device *hwdev, phys_addr_t tlb_addr,
>  			      enum dma_data_direction dir, unsigned long attrs)
>  {
>  	unsigned long flags;
> -	int i, count, nslots = ALIGN(alloc_size, 1 << IO_TLB_SHIFT) >> IO_TLB_SHIFT;
> +	unsigned int num_page_offset_slabs, page_offset_mask = dma_get_page_offset_mask(hwdev);
> +	int i, count;
> +	int nslots = ALIGN(alloc_size + tlb_addr & page_offset_mask, 1 << IO_TLB_SHIFT) >> IO_TLB_SHIFT;
>  	int index = (tlb_addr - io_tlb_start) >> IO_TLB_SHIFT;
>  	phys_addr_t orig_addr = io_tlb_orig_addr[index];
>  
> @@ -610,6 +621,14 @@ void swiotlb_tbl_unmap_single(struct device *hwdev, phys_addr_t tlb_addr,
>  	    ((dir == DMA_FROM_DEVICE) || (dir == DMA_BIDIRECTIONAL)))
>  		swiotlb_bounce(orig_addr, tlb_addr, mapping_size, DMA_FROM_DEVICE);
>  
> +	/*
> +	 * When dma_get_page_offset_mask is used, we may have padded more slabs
> +	 * when padding exceeds one slab. We need to move index back to the
> +	 * beginning of the padding.
> +	 */
> +	num_page_offset_slabs =  (tlb_addr & page_offset_mask) / (1 << IO_TLB_SHIFT);
> +	index -= num_page_offset_slabs;
> +
>  	/*
>  	 * Return the buffer to the free list by setting the corresponding
>  	 * entries to indicate the number of contiguous entries available.
> -- 
> 2.27.0
>