Re: [PATCH v4 2/4] usb: dwc2: add compatible data for rockchip soc

["Kever.Yang" <kever.yang@rock-chips.com>]

On 08/08/2014 04:52 AM, Doug Anderson wrote:
> Paul,
>
> On Thu, Aug 7, 2014 at 11:26 AM, Paul Zimmerman
> <Paul.Zimmerman@synopsys.com> wrote:
>>> From: Kever Yang [mailto:kever.yang@gmail.com] On Behalf Of Kever Yang
>>> Sent: Thursday, August 07, 2014 2:35 AM
>>>
>>> This patch add compatible data for dwc2 controller found on
>>> rk3066, rk3188 and rk3288 processors from rockchip.
>>>
>>> Signed-off-by: Kever Yang <kever.yang@rock-chips.com>
>>> Acked-by: Paul Zimmerman <paulz@synopsys.com>
>>> ---
>>>
>>> Changes in v4:
>>> - max_transfer_size change to 65536, this should be enough
>>>    for most transfer, the hardware auto-detect will set this
>>>    to 0x7ffff which may make dma_alloc_coherent fail when
>>>    non-dword aligned buf from driver like usbnet happen.
>> Hi Kever,
>>
>> Did you test this change thoroughly? I have vague memories of any
>> value above 65535 causing problems, at least on my hardware. And I
>> see it is set to 65535 in both pci.c and platform.c. I could be
>> wrong, but I thought I should mention it.
> Certainly it is documented in the header file to have a max of 65535:
>
>   * @max_transfer_size:  The maximum transfer size supported, in bytes
>   *                       2047 to 65,535
>   *                      Actual maximum value is autodetected and also
>   *                      the default.
Sorry for didn't check the header file, I'll change it to 65535 and 
resubmit.
>
> ...but looking at the register definition that I see, the size can be
> up to 19 bits.  A 19-bit transfer far exceeds 65535.  Do you remember
> what the error was?  Certainly I can imagine there being errors with
> large calls to dma_alloc_coherent()...
>
> I know that with Kever's change I can do USB Ethernet downloads, so it
> is at least working to some degree.  ...to me it feels like Kever
> should resubmit with 65535 (to match the documentation) and then work
> in the background to figure out what the max_transfer_size really
> ought to be.
You are right.
> -Doug
>
>
>