[RESEND,V7,4/6] i2c: qup: Add bam dma capabilities

Commit Message

Sricharan Ramabadhran March 21, 2016, 2:01 p.m. UTC

Hi wolfram,

> On Mon, Feb 22, 2016 at 05:38:15PM +0530, Sricharan R wrote:
> > QUP cores can be attached to a BAM module, which acts as a dma engine
> > for the QUP core. When DMA with BAM is enabled, the BAM consumer
> pipe
> > transmitted data is written to the output FIFO and the BAM producer
> > pipe received data is read from the input FIFO.
> >
> > With BAM capabilities, qup-i2c core can transfer more than
> > 256 bytes, without a 'stop' which is not possible otherwise.
> >
> > Signed-off-by: Sricharan R <[e-mail]>
> > Reviewed-by: Andy Gross <[e-mail]>
> > Tested-by: Archit Taneja <[e-mail]>
> > Tested-by: Telkar Nagender <[e-mail]>
> 
> Applied to for-next, thanks!

There is a bug with only this patch which comes out when CONFIG_DEBUG_SG
option gets
enabled.  I have patch like below to fix it, but wanted to ask you if it
would correct to have this in the
original patch or should I sent a separate fix ? Sorry for the mess on this.
Not sure if the pull request is sent
already.

 
@@ -1268,6 +1265,8 @@ static int qup_i2c_xfer_v2(struct i2c_adapter *adap,
 		}
 	}
 
+	idx = 0;
+
 	do {
 		if (msgs[idx].len == 0) {
 			ret = -EINVAL;
@@ -1407,27 +1406,21 @@ static int qup_i2c_probe(struct platform_device
*pdev)
 
 		/* 2 tag bytes for each block + 5 for start, stop tags */
 		size = blocks * 2 + 5;
-		qup->dpool = dma_pool_create("qup_i2c-dma-pool", &pdev->dev,
-					     size, 4, 0);
 
-		qup->start_tag.start = dma_pool_alloc(qup->dpool,
GFP_KERNEL,
-						      &qup->start_tag.addr);
+		qup->start_tag.start = devm_kzalloc(&pdev->dev,
+						    size, GFP_KERNEL);
 		if (!qup->start_tag.start) {
 			ret = -ENOMEM;
 			goto fail_dma;
 		}
 
-		qup->brx.tag.start = dma_pool_alloc(qup->dpool,
-						    GFP_KERNEL,
-						    &qup->brx.tag.addr);
+		qup->brx.tag.start = devm_kzalloc(&pdev->dev, size,
GFP_KERNEL);
 		if (!qup->brx.tag.start) {
 			ret = -ENOMEM;
 			goto fail_dma;
 		}
 
-		qup->btx.tag.start = dma_pool_alloc(qup->dpool,
-						    GFP_KERNEL,
-						    &qup->btx.tag.addr);
+		qup->btx.tag.start = devm_kzalloc(&pdev->dev, size,
GFP_KERNEL);
 		if (!qup->btx.tag.start) {
 			ret = -ENOMEM;
 			goto fail_dma;
@@ -1566,13 +1559,6 @@ static int qup_i2c_remove(struct platform_device
*pdev)
 	struct qup_i2c_dev *qup = platform_get_drvdata(pdev);
 
 	if (qup->is_dma) {
-		dma_pool_free(qup->dpool, qup->start_tag.start,
-			      qup->start_tag.addr);
-		dma_pool_free(qup->dpool, qup->brx.tag.start,
-			      qup->brx.tag.addr);
-		dma_pool_free(qup->dpool, qup->btx.tag.start,
-			      qup->btx.tag.addr);
-		dma_pool_destroy(qup->dpool);
 		dma_release_channel(qup->btx.dma);
 		dma_release_channel(qup->brx.dma);
 	}
--

Regards,
 Sricharan

Comments

Wolfram Sang March 21, 2016, 4:44 p.m. UTC | #1

> enabled.  I have patch like below to fix it, but wanted to ask you if it
> would correct to have this in the
> original patch or should I sent a separate fix ? Sorry for the mess on this.

Once a patch is applied, it is in linux-next, so you need to send
incremental patches to fix things.

Patch

diff --git a/drivers/i2c/busses/i2c-qup.c b/drivers/i2c/busses/i2c-qup.c
index 23eaabb..a1fae0c 100644
--- a/drivers/i2c/busses/i2c-qup.c
+++ b/drivers/i2c/busses/i2c-qup.c
@@ -595,9 +595,6 @@  static int qup_sg_set_buf(struct scatterlist *sg, void
*buf,
 	if (!ret)
 		return -EINVAL;
 
-	if (!map)
-		sg_dma_address(sg) = tg->addr + ((u8 *)buf - tg->start);
-
 	return 0;
 }