Re: [PATCH v9 1/3] dmaengine: ptdma: Initial driver for the AMD PTDMA

[Greg KH]

On Wed, Jun 16, 2021 at 12:27:32PM +0530, Sanjay R Mehta wrote:
> 
> 
> On 6/16/2021 11:46 AM, Greg KH wrote:
> > [CAUTION: External Email]
> > 
> > On Wed, Jun 16, 2021 at 10:24:52AM +0530, Sanjay R Mehta wrote:
> >>
> >>
> >> On 6/16/2021 9:45 AM, Vinod Koul wrote:
> >>> [CAUTION: External Email]
> >>>
> >>> On 15-06-21, 16:50, Sanjay R Mehta wrote:
> >>>
> >>>>>> +static struct pt_device *pt_alloc_struct(struct device *dev)
> >>>>>> +{
> >>>>>> +     struct pt_device *pt;
> >>>>>> +
> >>>>>> +     pt = devm_kzalloc(dev, sizeof(*pt), GFP_KERNEL);
> >>>>>> +
> >>>>>> +     if (!pt)
> >>>>>> +             return NULL;
> >>>>>> +     pt->dev = dev;
> >>>>>> +     pt->ord = atomic_inc_return(&pt_ordinal);
> >>>>>
> >>>>> What is the use of this number?
> >>>>>
> >>>>
> >>>> There are eight similar instances of this DMA engine on AMD SOC.
> >>>> It is to differentiate each of these instances.
> >>>
> >>> Are they individual device objects?
> >>>
> >>
> >> Yes, they are individual device objects.
> > 
> > Then what is "ord" for?  Why are you using an atomic variable for this?
> > What does this field do?  Why doesn't the normal way of naming a device
> > come into play here instead?
> > 
> 
> Hi Greg,
> 
> The value of "ord" is incremented for each device instance and then it
> is used to store different name for each device as shown in below snippet.
> 
> 	pt->ord = atomic_inc_return(&pt_ordinal);
> 	snprintf(pt->name, MAX_PT_NAME_LEN, "pt-%u", pt->ord);

Why not use an idr structure like this like all other drivers do?  That
way when devices are removed the numbers are properly reused as well.

And why do you need to save the value?

thanks,

greg k-h