Re: [PATCH V2 2/3] ARM: shmobile: r8a7790: Add PCI USB host clock support

[Valentine <valentine.barshak@cogentembedded.com>]

On 01/28/2014 04:28 AM, Magnus Damm wrote:
> On Fri, Jan 24, 2014 at 8:37 PM, Ben Dooks <ben.dooks@codethink.co.uk> wrote:
>> On 24/01/14 04:47, Magnus Damm wrote:
>>>
>>> On Sat, Jan 18, 2014 at 3:20 AM, Valentine Barshak
>>> <valentine.barshak@cogentembedded.com> wrote:
>>>>
>>>> This adds internal PCI USB host clock support.
>>>>
>>>> Signed-off-by: Valentine Barshak <valentine.barshak@cogentembedded.com>
>>>> ---
>>>>    arch/arm/mach-shmobile/clock-r8a7790.c | 6 +++++-
>>>>    1 file changed, 5 insertions(+), 1 deletion(-)
>>>>
>>>> Changes in V2:
>>>> * capitalized ARM in the subject;
>>>> * rebased on top the latest devel tag.
>>>
>>>
>>> This patch for PCI USB clock support seems fine to me.
>>>
>>> Acked-by: Magnus Damm <damm@opensource.se>
>>>
>>> It is common practice that maintainers reject code to encourage people
>>> to do further development. I don't think that will help us in this
>>> particular case, so I think it is best to merge this as-is but also
>>> request you to spend your future time on DT development.
>>>
>>> So please work on DT reference support together with CCF for PCI USB.
>>
>>
>> I've already posted a series for getting this device treeed.
>
> Thanks for that!
>
>> As a note, does channel 0 work for you? we are getting a lot of
>> failures with the OHCI controller failing to start with what looks
>> like a bus-master failure.
>
> I've only used USB0 as USBHS, and due to lack of cable detection
> hardware support on Lager (and me missing the cable needed for host) I
> plan on simply fixing USB0 as USBHS Function. Anyone wanting to test
> USB PCI Host should in my opinion use USB1 on Lager. USB2 is also fine
> until USB 3.0 support is tied in there.
>
> As both you and I have noted, the USB PCI stuff can't work well as-is
> due to mismatch in physical address space between the PCI hardware and
> the actual memory on Lager. So as we discussed, bounce buffers would
> be needed and perhaps also IOMMU support but my feeling is that may as
> well be for later in case of USB.

As I have said earlier I have not experienced any issues with the current kernel configuration.
Enabling highmem also didn't show any harm. I'm currently testing LPAE and have not been able
to see any issues as well. I've been using memtest and reading/writing USB storage at the same time.

AFAIU, the DMA is allocated in the lowmem for all USB transfers. Even when disk buffers are allocated
in the highmem, the contents is still copied to a lowmem buffer by the USB storage driver.
If anyone tried to allocate DMA buffer in highmem, using DMA bounce buffers would not help
since bouncing of highmem is not supported.

>
> So until I see USB PCI bounce buffer support or similar I will simply
> assume all errors on USB PCI on r8a7790 and r8a7791 are related to
> physical address size mismatch and lacking software support. Do you
> have anything that points in a different direction?

The only case that I have been able to see any errors is when a different memory split
model is used in the kernel. For example, if 2G/2G user/kernel memory model is used
instead of the default 3G/1G one. In this case DMA-able low memory is 2G and the bounce buffer
is needed.

Thanks,
Val.

>
> Does USB1 and/or USB2 work for you as-is today?
>
> Thanks,
>
> / magnus
> --
> To unsubscribe from this list: send the line "unsubscribe linux-sh" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
>