Re: [PATCH v2 0/9] DMA engine cookie handling cleanups

[Russell King - ARM Linux <linux@arm.linux.org.uk>]

On Tue, Mar 13, 2012 at 02:10:36PM +0530, Vinod Koul wrote:
> Please see if the below patch is the right fix for build failures in
> addition to one suggested by Jassi.

I'm not sure that Jassi's solution is correct - and I'm wondering whether
any of the DMA engine drivers do the right thing when transfers are
terminated.  Is it right for the DMA status function to return IN_PROGRESS
for a previously submitted cookie which has been terminated?

I can see two answers to that, both equally valid:

1. It allows you to find out exactly where the DMA engine got to before
   the transfer was terminated, and therefore recover from the termination
   if you wish to.

2. Returning in-progress when a cookie will never be completed is
   misleading, and could be misinterpreted by users of the tx_status
   function, especially if they are waiting for a particular transaction
   to complete.

Maybe we need to introduce a DMA_TERMINATED status?

> -------------------x-------------------------x----------------------
> 
> >From 949ff5b8d46b5e3435d21b2651ce3a2599208d44 Mon Sep 17 00:00:00 2001
> From: Vinod Koul <vinod.koul@linux.intel.com>
> Date: Tue, 13 Mar 2012 11:58:12 +0530
> Subject: [PATCH] dmaengine: fix for cookie changes and merge
> 
> Fixed trivial issues in drivers:
> 	drivers/dma/imx-sdma.c
> 	drivers/dma/intel_mid_dma.c
> 	drivers/dma/ioat/dma_v3.c
> 	drivers/dma/iop-adma.c
> 	drivers/dma/sirf-dma.c
> 	drivers/dma/timb_dma.c
> 
> Signed-off-by: Vinod Koul <vinod.koul@linux.intel.com>
> ---
>  drivers/dma/imx-sdma.c      |    1 +
>  drivers/dma/intel_mid_dma.c |    1 +
>  drivers/dma/ioat/dma_v3.c   |    1 +
>  drivers/dma/iop-adma.c      |    1 +
>  drivers/dma/sirf-dma.c      |    2 ++
>  drivers/dma/timb_dma.c      |    6 +-----
>  6 files changed, 7 insertions(+), 5 deletions(-)
> 
> diff --git a/drivers/dma/imx-sdma.c b/drivers/dma/imx-sdma.c
> index ccfc7c4..81f9d57 100644
> --- a/drivers/dma/imx-sdma.c
> +++ b/drivers/dma/imx-sdma.c
> @@ -1127,6 +1127,7 @@ static void sdma_issue_pending(struct dma_chan *chan)
>  	struct sdma_engine *sdma = sdmac->sdma;
>  
>  	if (sdmac->status == DMA_IN_PROGRESS)
> +		sdma_enable_channel(sdma, sdmac->channel);
>  }
>  
>  #define SDMA_SCRIPT_ADDRS_ARRAY_SIZE_V1	34

This looks like a merge conflict resolution.  I don't see this being
caused by my patches as I haven't touched this function.

> diff --git a/drivers/dma/intel_mid_dma.c b/drivers/dma/intel_mid_dma.c
> index d599d96..2449812 100644
> --- a/drivers/dma/intel_mid_dma.c
> +++ b/drivers/dma/intel_mid_dma.c
> @@ -477,6 +477,7 @@ static enum dma_status intel_mid_dma_tx_status(struct dma_chan *chan,
>  						dma_cookie_t cookie,
>  						struct dma_tx_state *txstate)
>  {
> +	struct intel_mid_dma_chan *midc = to_intel_mid_dma_chan(chan);
>  	enum dma_status ret;
>  
>  	ret = dma_cookie_status(chan, cookie, txstate);

Ditto (my patches don't introduce new this new midc, nor do they remove
that line.)

> diff --git a/drivers/dma/ioat/dma_v3.c b/drivers/dma/ioat/dma_v3.c
> index 145eda2..2c4476c 100644
> --- a/drivers/dma/ioat/dma_v3.c
> +++ b/drivers/dma/ioat/dma_v3.c
> @@ -61,6 +61,7 @@
>  #include <linux/dmaengine.h>
>  #include <linux/dma-mapping.h>
>  #include <linux/prefetch.h>
> +#include "../dmaengine.h"
>  #include "registers.h"
>  #include "hw.h"
>  #include "dma.h"
> diff --git a/drivers/dma/iop-adma.c b/drivers/dma/iop-adma.c
> index 1f3a703..4499f88 100644
> --- a/drivers/dma/iop-adma.c
> +++ b/drivers/dma/iop-adma.c
> @@ -894,6 +894,7 @@ static enum dma_status iop_adma_status(struct dma_chan *chan,
>  					struct dma_tx_state *txstate)
>  {
>  	struct iop_adma_chan *iop_chan = to_iop_adma_chan(chan);
> +	int ret;

This was "enum dma_status ret;" before I accidentally removed it.  It
probably should be again, rather than an int.

>  
>  	ret = dma_cookie_status(chan, cookie, txstate);
>  	if (ret == DMA_SUCCESS)
> diff --git a/drivers/dma/sirf-dma.c b/drivers/dma/sirf-dma.c
> index a2cde85..45ba352 100644
> --- a/drivers/dma/sirf-dma.c
> +++ b/drivers/dma/sirf-dma.c
> @@ -18,6 +18,8 @@
>  #include <linux/of_platform.h>
>  #include <linux/sirfsoc_dma.h>
>  
> +#include "dmaengine.h"
> +
>  #define SIRFSOC_DMA_DESCRIPTORS                 16
>  #define SIRFSOC_DMA_CHANNELS                    16
>  

Hmm, guess that's what happens when old patches are brought forward and
things from the original series are forgotten...

> diff --git a/drivers/dma/timb_dma.c b/drivers/dma/timb_dma.c
> index 7805996..d408c22 100644
> --- a/drivers/dma/timb_dma.c
> +++ b/drivers/dma/timb_dma.c
> @@ -510,17 +510,13 @@ static void td_free_chan_resources(struct dma_chan *chan)
>  static enum dma_status td_tx_status(struct dma_chan *chan, dma_cookie_t cookie,
>  				    struct dma_tx_state *txstate)
>  {
> -	struct timb_dma_chan *td_chan =
> -		container_of(chan, struct timb_dma_chan, chan);
>  	enum dma_status ret;
>  
>  	dev_dbg(chan2dev(chan), "%s: Entry\n", __func__);
>  
>  	ret = dma_cookie_status(chan, cookie, txstate);
>  
> -	dev_dbg(chan2dev(chan),
> -		"%s: exit, ret: %d, last_complete: %d, last_used: %d\n",
> -		__func__, ret, last_complete, last_used);
> +	dev_dbg(chan2dev(chan), "%s: exit, ret: %d\n", 	__func__, ret);
>  
>  	return ret;
>  }
> -- 
> 1.7.0.4
> 
> 
> 
> 
> -- 
> ~Vinod
>