A few questions on imx-sdma

[Sascha Hauer]

On Wed, Jan 19, 2011 at 03:43:00AM +0800, Shawn Guo wrote:
> Hi Sascha,
> 
> I got a few questions when studying imx-sdma.  Can you please help me
> understand?
> 
> * In function sdma_prep_slave_sg, we always get "sdmac->flags = 0;"
> at line 889.  What's the code blow at line 934 for?
> 
> 	if (sdmac->flags & IMX_DMA_SG_LOOP) {
> 		param |= BD_INTR;
> 		if (i + 1 == sg_len)
> 			param |= BD_WRAP;
> 	}

A leftover from the days we did not have the prep_cyclic callback and
implemented sound dma as a looped sg chain. The above can be removed.

> 
> * We have sdmac->status set to DMA_ERROR in err_out of function 
> sdma_prep_dma_cyclic.  Would it be good for us to do the same for
> function sdma_prep_slave_sg.  Otherwise, sdmac->status stays at
> DMA_IN_PROGRESS, which will make the function return immediately
> next time it gets called?

yes

> 
> * It seems the only use of sdmac->status is to check if the channel
> is DMA_IN_PROGRESS when "prep" functions get called.  Is it a worth
> to check DMA_ERROR or DMA_SUCCESS for every single buffer descriptor
> (BD) in interrupt handler?
> 
> * In handler mxc_sdma_handle_channel_normal, callback function gets
> called and last_completed get assigned the cookie no matter
> sdmac->status is DMA_SUCCESS or DMA_ERROR.  How would dma client know
> if this tx succeeded or failed?

Looking at it we should probably just return sdmac->status in
sdma_tx_status.

> 
> * Related to question above, dma client can call function
> dma_async_is_tx_complete to know if the tx succeeded or failed by
> manipulating cookie.  But is there any way for client to know the
> failure in loop (cyclic) case?  In this case, client gets no way to
> query tx status, however any period (BD) in the cyclic could fail.
> Currently, we always do callback in cyclic handler whether status is
> DMA_ERROR or DMA_SUCCESS, and the client (i.e. imx audio driver)
> always assumes the callback means a successful period.  But what if
> one period in the cyclic fails?

I guess the driver should call dma_async_is_tx_complete, assuming this
function is fixed in the sdma driver.

> 
> * Regarding to cyclic case again, is DMA_TERMINATE_ALL
> (down to function sdma_disable_channel) the only way for client to
> stop the tx?  I do not know the exact behavior of sdma when STATSTOP
> bit gets set.  But if it's in the middle of a BD transfer process,
> will this bit setting possibly cause data corruption?

I don't know the exact behaviour either. I only know that the FSL kernel
does this aswell (without polling whether the channel is actually
disabled or something).

Sascha

-- 
Pengutronix e.K.                           |                             |
Industrial Linux Solutions                 | http://www.pengutronix.de/  |
Peiner Str. 6-8, 31137 Hildesheim, Germany | Phone: +49-5121-206917-0    |
Amtsgericht Hildesheim, HRA 2686           | Fax:   +49-5121-206917-5555 |