Re: [PATCH] ARM: edma: special case slot limit workaround

From: Peter Ujfalusi
Date: Thu Oct 15 2015 - 07:06:03 EST


On 10/15/2015 01:48 PM, John Ogness wrote:
> Currently drivers are limited to 19 slots for cyclic transfers.
> However, if the DMA burst size is the same as the period size,
> the period size can be changed to the full buffer size and
> intermediate interrupts activated. Since intermediate interrupts
> will trigger for each burst and the burst size is the same as
> the period size, the driver will get interrupts each period as
> expected. This has the benefit of allowing the functionality of
> many more slots, but only uses 2 slots.
>
> This workaround is only active if more than 19 slots are needed
> and the burst size matches the period size.

To: Vinod
CC: l-o, Sekhar at least

>
> Signed-off-by: John Ogness <john.ogness@xxxxxxxxxxxxx>
> ---
> drivers/dma/edma.c | 25 ++++++++++++++++++++++---
> 1 file changed, 22 insertions(+), 3 deletions(-)
>
> diff --git a/drivers/dma/edma.c b/drivers/dma/edma.c
> index 493c774..16bd193 100644
> --- a/drivers/dma/edma.c
> +++ b/drivers/dma/edma.c
> @@ -585,6 +585,7 @@ static struct dma_async_tx_descriptor *edma_prep_dma_cyclic(
> struct edma_desc *edesc;
> dma_addr_t src_addr, dst_addr;
> enum dma_slave_buswidth dev_width;
> + bool use_intermediate = false;
> u32 burst;
> int i, ret, nslots;
>
> @@ -626,8 +627,21 @@ static struct dma_async_tx_descriptor *edma_prep_dma_cyclic(
> * but the synchronization is difficult to achieve with Cyclic and
> * cannot be guaranteed, so we error out early.
> */
> - if (nslots > MAX_NR_SG)
> - return NULL;
> + if (nslots > MAX_NR_SG) {
> + /*
> + * If the burst and period sizes are the same, we can put
> + * the full buffer into a single period and activate
> + * intermediate interrupts. This will produce interrupts
> + * after each burst, which is also after each desired period.
> + */
> + if (burst == period_len) {
> + period_len = buf_len;
> + nslots = 2;
> + use_intermediate = true;
> + } else {
> + return NULL;
> + }
> + }
>
> edesc = kzalloc(sizeof(*edesc) + nslots *
> sizeof(edesc->pset[0]), GFP_ATOMIC);
> @@ -706,8 +720,13 @@ static struct dma_async_tx_descriptor *edma_prep_dma_cyclic(
> /*
> * Enable period interrupt only if it is requested
> */
> - if (tx_flags & DMA_PREP_INTERRUPT)
> + if (tx_flags & DMA_PREP_INTERRUPT) {
> edesc->pset[i].param.opt |= TCINTEN;
> +
> + /* Also enable intermediate interrupts if necessary */
> + if (use_intermediate)
> + edesc->pset[i].param.opt |= ITCINTEN;
> + }
> }

The workaround looks fine, but please rebase it on the latest edma code from
linux-next.

--
Péter
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/