lkml.org 
[lkml]   [2015]   [Nov]   [4]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
SubjectRe: [PATCH V2 3/3] dma: add Qualcomm Technologies HIDMA channel driver
From
On Wed, Nov 4, 2015 at 2:07 AM, Sinan Kaya <okaya@codeaurora.org> wrote:
> On 11/3/2015 5:10 AM, Andy Shevchenko wrote:
>> On Mon, Nov 2, 2015 at 8:07 AM, Sinan Kaya <okaya@codeaurora.org> wrote:

>>> +static void hidma_issue_pending(struct dma_chan *dmach)
>>> +{
>>
>>
>> Wrong. It should actually start the transfer. tx_submit() just puts
>> the descriptor to a queue.
>>
> Depends on the design.
>
> I started from the Freescale driver (mpc512x_dma.c). It follows the same
> model.
>
> I'll just drop the same comment into this code too.
>
>
> /*
> * We are posting descriptors to the hardware as soon as
> * they are ready, so this function does nothing.
> */

So, the Freescale driver was written before change went effective. I
guess in 2011 DMA Engine drivers should use issue pending.
Please, refactor since this behaviour is expected.

>>> +/*
>>> + * Submit descriptor to hardware.
>>> + * Lock the PM for each descriptor we are sending.
>>> + */
>>> +static dma_cookie_t hidma_tx_submit(struct dma_async_tx_descriptor *txd)
>>> +{
>>> + struct hidma_chan *mchan = to_hidma_chan(txd->chan);
>>> + struct hidma_dev *dmadev = mchan->dmadev;
>>> + struct hidma_desc *mdesc;
>>> + unsigned long irqflags;
>>> + dma_cookie_t cookie;
>>> +
>>> + if (!hidma_ll_isenabled(dmadev->lldev))
>>> + return -ENODEV;
>>> +
>>> + pm_runtime_get_sync(dmadev->ddev.dev);
>>
>>
>> No point to do it here. It should be done on the function that
>> actually starts the transfer (see issue pending).
>>
> comment above

See above as well.

>>> +static int hidma_probe(struct platform_device *pdev)
>>> +{
>>> + struct hidma_dev *dmadev;
>>> + int rc = 0;
>>> + struct resource *trca_resource;
>>> + struct resource *evca_resource;
>>> + int chirq;
>>> + int current_channel_index = atomic_read(&channel_ref_count);
>>> +

>>> + /* Set DMA mask to 64 bits. */
>>> + rc = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64));
>>> + if (rc) {
>>> + dev_warn(&pdev->dev, "unable to set coherent mask to
>>> 64");
>>> + rc = dma_set_mask_and_coherent(&pdev->dev,
>>> DMA_BIT_MASK(32));
>>> + }
>>> + if (rc)
>>> + goto dmafree;

Maybe move these two lines inside previous condition?

>>> +
>>> + dmadev->lldev = hidma_ll_init(dmadev->ddev.dev,
>>> + dmadev->nr_descriptors, dmadev->dev_trca,
>>> + dmadev->dev_evca, dmadev->evridx);
>>> + if (!dmadev->lldev) {
>>> + rc = -EPROBE_DEFER;
>>> + goto dmafree;
>>> + }
>>> +
>>> + rc = devm_request_irq(&pdev->dev, chirq, hidma_chirq_handler, 0,
>>> + "qcom-hidma", &dmadev->lldev);
>>
>>
>> Better to use request_irq().
>>
>
> why? I thought we favored managed functions over standalone functions in
> simplify the exit path.

IRQ is slightly different in workflow. In most cases, unfortunately,
there is no achievement by devm_ variant.
At least I know two for now. One of them is DMA Engine slave drivers,
though I didn't notice if you are using tasklet's here.
Otherwise it's okay.

--
With Best Regards,
Andy Shevchenko


\
 
 \ /
  Last update: 2015-11-04 19:01    [W:0.102 / U:0.472 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site