Re: [PATCH 08/12] async_raid6_recov: convert to dmaengine_unmap_data

[Bartlomiej Zolnierkiewicz <b.zolnierkie@samsung.com>]

On Thursday 06 December 2012 10:25:54 Dan Williams wrote:
> Use the generic unmap object to unmap dma buffers.
> 
> Cc: Tomasz Figa <t.figa@samsung.com>
> Cc: Kyungmin Park <kyungmin.park@samsung.com>
> Reported-by: Bartlomiej Zolnierkiewicz <b.zolnierkie@samsung.com>
> Signed-off-by: Dan Williams <djbw@fb.com>
> ---
>  crypto/async_tx/async_raid6_recov.c |   69 ++++++++++++++++++++++++-----------
>  1 file changed, 48 insertions(+), 21 deletions(-)
> 
> diff --git a/crypto/async_tx/async_raid6_recov.c b/crypto/async_tx/async_raid6_recov.c
> index a9f08a6..20aea04 100644
> --- a/crypto/async_tx/async_raid6_recov.c
> +++ b/crypto/async_tx/async_raid6_recov.c
> @@ -26,6 +26,7 @@
>  #include <linux/dma-mapping.h>
>  #include <linux/raid/pq.h>
>  #include <linux/async_tx.h>
> +#include <linux/dmaengine.h>
>  
>  static struct dma_async_tx_descriptor *
>  async_sum_product(struct page *dest, struct page **srcs, unsigned char *coef,
> @@ -34,35 +35,47 @@ async_sum_product(struct page *dest, struct page **srcs, unsigned char *coef,
>  	struct dma_chan *chan = async_tx_find_channel(submit, DMA_PQ,
>  						      &dest, 1, srcs, 2, len);
>  	struct dma_device *dma = chan ? chan->device : NULL;
> +	struct dmaengine_unmap_data *unmap = NULL;
>  	const u8 *amul, *bmul;
>  	u8 ax, bx;
>  	u8 *a, *b, *c;
>  
> -	if (dma) {
> -		dma_addr_t dma_dest[2];
> -		dma_addr_t dma_src[2];
> +	if (dma)
> +		unmap = dmaengine_get_unmap_data(dma->dev, 3, GFP_NOIO);
> +
> +	if (unmap) {
>  		struct device *dev = dma->dev;
> +		dma_addr_t pq[2];
>  		struct dma_async_tx_descriptor *tx;
> -		enum dma_ctrl_flags dma_flags = DMA_PREP_PQ_DISABLE_P;
> +		enum dma_ctrl_flags dma_flags = DMA_COMPL_SKIP_SRC_UNMAP |
> +						DMA_COMPL_SKIP_DEST_UNMAP |
> +						DMA_PREP_PQ_DISABLE_P;
>  
>  		if (submit->flags & ASYNC_TX_FENCE)
>  			dma_flags |= DMA_PREP_FENCE;
> -		dma_dest[1] = dma_map_page(dev, dest, 0, len, DMA_BIDIRECTIONAL);
> -		dma_src[0] = dma_map_page(dev, srcs[0], 0, len, DMA_TO_DEVICE);
> -		dma_src[1] = dma_map_page(dev, srcs[1], 0, len, DMA_TO_DEVICE);
> -		tx = dma->device_prep_dma_pq(chan, dma_dest, dma_src, 2, coef,
> +		unmap->addr[0] = dma_map_page(dev, srcs[0], 0, len, DMA_TO_DEVICE);
> +		unmap->addr[1] = dma_map_page(dev, srcs[1], 0, len, DMA_TO_DEVICE);
> +		unmap->to_cnt = 2;
> +
> +		unmap->addr[2] = dma_map_page(dev, dest, 0, len, DMA_BIDIRECTIONAL);
> +		unmap->bidi_cnt = 1;
> +		/* engine only looks at Q, but expects it to follow P */
> +		pq[1] = unmap->addr[2];
> +
> +		unmap->len = len;
> +		tx = dma->device_prep_dma_pq(chan, pq, unmap->addr, 2, coef,
>  					     len, dma_flags);
>  		if (tx) {
> +			dma_set_unmap(tx, unmap);
>  			async_tx_submit(chan, tx, submit);
> +			dmaengine_unmap_put(unmap);
>  			return tx;
>  		}
>  
>  		/* could not get a descriptor, unmap and fall through to
>  		 * the synchronous path
>  		 */
> -		dma_unmap_page(dev, dma_dest[1], len, DMA_BIDIRECTIONAL);
> -		dma_unmap_page(dev, dma_src[0], len, DMA_TO_DEVICE);
> -		dma_unmap_page(dev, dma_src[1], len, DMA_TO_DEVICE);
> +		dmaengine_unmap_put(unmap);
>  	}
>  
>  	/* run the operation synchronously */
> @@ -89,23 +102,38 @@ async_mult(struct page *dest, struct page *src, u8 coef, size_t len,
>  	struct dma_chan *chan = async_tx_find_channel(submit, DMA_PQ,
>  						      &dest, 1, &src, 1, len);
>  	struct dma_device *dma = chan ? chan->device : NULL;
> +	struct dmaengine_unmap_data *unmap = NULL;
>  	const u8 *qmul; /* Q multiplier table */
>  	u8 *d, *s;
>  
> -	if (dma) {
> -		dma_addr_t dma_dest[2];
> -		dma_addr_t dma_src[1];
> +	if (dma)
> +		unmap = dmaengine_get_unmap_data(dma->dev, 3, GFP_NOIO);
> +
> +	if (unmap) {
>  		struct device *dev = dma->dev;
>  		struct dma_async_tx_descriptor *tx;
> -		enum dma_ctrl_flags dma_flags = DMA_PREP_PQ_DISABLE_P;
> +		enum dma_ctrl_flags dma_flags = DMA_COMPL_SKIP_SRC_UNMAP |
> +						DMA_COMPL_SKIP_DEST_UNMAP |
> +						DMA_PREP_PQ_DISABLE_P;
>  
>  		if (submit->flags & ASYNC_TX_FENCE)
>  			dma_flags |= DMA_PREP_FENCE;
> -		dma_dest[1] = dma_map_page(dev, dest, 0, len, DMA_BIDIRECTIONAL);
> -		dma_src[0] = dma_map_page(dev, src, 0, len, DMA_TO_DEVICE);
> -		tx = dma->device_prep_dma_pq(chan, dma_dest, dma_src, 1, &coef,
> -					     len, dma_flags);
> +		unmap->addr[0] = dma_map_page(dev, src, 0, len, DMA_TO_DEVICE);
> +		unmap->to_cnt++;
> +		unmap->addr[1] = dma_map_page(dev, dest, 0, len, DMA_BIDIRECTIONAL);
> +		unmap->bidi_cnt++;
> +		unmap->len = len;
> +
> +		/* this looks funny, but the engine looks for Q at
> +		 * unmap->addr[1] and ignores unmap->addr[0] as a dest
> +		 * due to DMA_PREP_PQ_DISABLE_P
> +		 */
> +		tx = dma->device_prep_dma_pq(chan, unmap->addr, unmap->addr,
> +					     1, &coef, len, dma_flags);

at least iop-adma.c and ioat/dma_v3.c seem to modify content of
unmap->addr[0] which is probably not what we want and therefore
temporary dma_dest array should still be used here

static struct dma_async_tx_descriptor *
iop_adma_prep_dma_pq(struct dma_chan *chan, dma_addr_t *dst, dma_addr_t *src,
		     unsigned int src_cnt, const unsigned char *scf, size_t len,
		     unsigned long flags)
{
...
		/* even if P is disabled its destination address (bits
		 * [3:0]) must match Q.  It is ok if P points to an
		 * invalid address, it won't be written.
		 */
		if (flags & DMA_PREP_PQ_DISABLE_P)
			dst[0] = dst[1] & 0x7;
...
}

static struct dma_async_tx_descriptor *
ioat3_prep_pq(struct dma_chan *chan, dma_addr_t *dst, dma_addr_t *src,
	      unsigned int src_cnt, const unsigned char *scf, size_t len,
	      unsigned long flags)
{
	/* specify valid address for disabled result */
	if (flags & DMA_PREP_PQ_DISABLE_P)
		dst[0] = dst[1];
	if (flags & DMA_PREP_PQ_DISABLE_Q)
		dst[1] = dst[0];
...
}

> +
>  		if (tx) {
> +			dma_set_unmap(tx, unmap);
> +			dmaengine_unmap_put(unmap);
>  			async_tx_submit(chan, tx, submit);
>  			return tx;
>  		}
> @@ -113,8 +141,7 @@ async_mult(struct page *dest, struct page *src, u8 coef, size_t len,
>  		/* could not get a descriptor, unmap and fall through to
>  		 * the synchronous path
>  		 */
> -		dma_unmap_page(dev, dma_dest[1], len, DMA_BIDIRECTIONAL);
> -		dma_unmap_page(dev, dma_src[0], len, DMA_TO_DEVICE);
> +		dmaengine_unmap_put(unmap);
>  	}
>  
>  	/* no channel available, or failed to allocate a descriptor, so

Best regards,
--
Bartlomiej Zolnierkiewicz
Samsung Poland R&D Center