Re: [PATCH 2/3] dmaeninge: xilinx_dma: Fix bug in multiple frame stores scenario in vdma

From: Jose Abreu
Date: Fri Dec 16 2016 - 05:14:02 EST


Hi Kedar,


On 15-12-2016 19:09, Appana Durga Kedareswara Rao wrote:
> Hi Jose Miguel Abreu,
>
> Thanks for the review....
>
>>> - last = segment;
>>> + for (j = 0; j < chan->num_frms; ) {
>>> + list_for_each_entry(segment, &desc->segments, node)
>> {
>>> + if (chan->ext_addr)
>>> + vdma_desc_write_64(chan,
>>> +
>> XILINX_VDMA_REG_START_ADDRESS_64(i++),
>>> + segment->hw.buf_addr,
>>> + segment->hw.buf_addr_msb);
>>> + else
>>> + vdma_desc_write(chan,
>>> +
>> XILINX_VDMA_REG_START_ADDRESS(i++),
>>> + segment->hw.buf_addr);
>>> +
>>> + last = segment;
>> Hmm, is it possible to submit more than one segment? If so, then i and j will get
>> out of sync.
> If h/w is configured for more than 1 frame buffer and user submits more than one frame buffer
> We can submit more than one frame/ segment to hw right??

I'm not sure. When I used VDMA driver I always submitted only one
segment and multiple descriptors. But the problem is, for example:

If you have:
descriptor1 (2 segments)
descriptor2 (2 segments)

And you have 3 frame buffers in the HW.

Then:
1st frame buffer will have: descriptor1 -> segment1
2nd frame buffer will have: descriptor1 -> segment2
3rd frame buffer will have: descriptor2 -> segment1
but, 4th frame buffer will have: descriptor2 -> segment2 <----
INVALID because there is only 3 frame buffers

So, maybe a check inside the loop "list_for_each_entry(segment,
&desc->segments, node)" could be a nice to have.

>
>>> + }
>>> + list_del(&desc->node);
>>> + list_add_tail(&desc->node, &chan->active_list);
>>> + j++;
>> But if i is non zero and pending_list has more than num_frms then i will not
>> wrap-around as it should and will write to invalid framebuffer location, right?
> Yep will fix in v2...
>
> If (if (list_empty(&chan->pending_list)) || (i == chan->num_frms)
> break;
>
> Above condition is sufficient right???

Looks ok.

>
>>> + if (list_empty(&chan->pending_list))
>>> + break;
>>> + desc = list_first_entry(&chan->pending_list,
>>> + struct
>> xilinx_dma_tx_descriptor,
>>> + node);
>>> }
>>>
>>> if (!last)
>>> @@ -1114,14 +1124,13 @@ static void xilinx_vdma_start_transfer(struct
>> xilinx_dma_chan *chan)
>>> vdma_desc_write(chan, XILINX_DMA_REG_FRMDLY_STRIDE,
>>> last->hw.stride);
>>> vdma_desc_write(chan, XILINX_DMA_REG_VSIZE, last-
>>> hw.vsize);
>> Maybe a check that all framebuffers contain valid addresses should be done
>> before programming vsize so that VDMA does not try to write to invalid
>> addresses.
> Do we really need to check for valid address???
> I didn't get you what to do you mean by invalid address could you please explain???
> In the driver we are reading form the pending_list which will be updated by pep_interleaved_dma
> Call so we are under assumption that user sends the proper address right???

What I mean by valid address is to check that i variable has
already been incremented by num_frms at least once since a VDMA
reset. This way you know that you have programmed all the
addresses of the frame buffers with an address and they are non-zero.

Best regards,
Jose Miguel Abreu

>
>>> +
>>> + chan->desc_submitcount += j;
>>> + chan->desc_pendingcount -= j;
>>> }
>>>
>>> chan->idle = false;
>>> if (!chan->has_sg) {
>>> - list_del(&desc->node);
>>> - list_add_tail(&desc->node, &chan->active_list);
>>> - chan->desc_submitcount++;
>>> - chan->desc_pendingcount--;
>>> if (chan->desc_submitcount == chan->num_frms)
>>> chan->desc_submitcount = 0;
>> "desc_submitcount >= chan->num_frms would be safer here.
> Sure will fix in v2...
>
> Regards,
> Kedar.
>
>>> } else {
>> Best regards,
>> Jose Miguel Abreu
>> --
>> To unsubscribe from this list: send the line "unsubscribe dmaengine" in the body
>> of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at
>> http://vger.kernel.org/majordomo-info.html