Re: [PATCH] i2c: ismt: fix wrong device address when unmap the data buffer

[Peter Rosin]

On 2017-06-12 11:38, Andy Shevchenko wrote:
> On Mon, Jun 12, 2017 at 12:28 PM, Peter Rosin <peda@xxxxxxxxxx> wrote:
>> On 2017-06-12 11:11, Andy Shevchenko wrote:
>>> On Mon, Jun 12, 2017 at 6:42 AM, Song liwei <liwei.song@xxxxxxxxxxxxx> wrote:
>>>> From: Liwei Song <liwei.song@xxxxxxxxxxxxx>
> 
>>>> After finished I2C block read/write, when unmap the data buffer,
>>>> a wrong device address was pass to dma_unmap_single(),
> 
>>>> the right
>>>> device address should be "dev" not "&adap->dev", the relation is
>>>> *(&adap->dev) == dev.
>>>
>>> This is confusing. You are telling that there are two copies of struct
>>> device here?
>>
>> Yes, there are two copies.
> 
> No, there is not. See below.

What I meant was that there are the struct device in pci_dev->dev and the
struct device in adap->dev. That seems like two copies of struct device
to me. I didn't mean that they are copies in the sense that they have the
same content, but in the sense that they are both struct device.

I guess we can argue ourselves blue over this point.

> There are two struct devices, 

Hmm, two struct devices, I seem to recall that from somewhere... :-)

>                               one is a real PCI device, which
> represents actual device what *does* DMA.
> This struct should be used according to DMA API.

When you put it like that, it's obvious that the patch is correct. I had
this feeling that little thought had gone into the choice to pick "dev"
over "&adap->dev", that's all.

> Another struct device which is wrongly used is an artificial one that
> represents I2C adapter in terms of Linux kernel.

Cheers,
peda