[PATCH RFC] PCI: imx6: remove outbound io/mem ATU region mapping

[Richard Zhu]

Hi Marek:


> -----Original Message-----
> From: Marek Vasut [mailto:marex at denx.de]
> Sent: Friday, November 29, 2013 10:21 AM
> To: Jingoo Han
> Cc: 'Pratyush Anand'; 'Mohit KUMAR DCG'; 'Tim Harvey'; 'Kishon Vijay Abraham
> I'; linux-pci at vger.kernel.org; linux-arm-kernel at lists.infradead.org; 'Bjorn
> Helgaas'; 'Frank Li'; Zhu Richard-R65037; 'Sascha Hauer'; 'Sean Cross'; 'Shawn
> Guo'; 'Siva Reddy Kallam'; 'Srikanth T Shivanand'; 'Troy Kisky'; 'Yinghai Lu'
> Subject: Re: [PATCH RFC] PCI: imx6: remove outbound io/mem ATU region mapping
> 
> Dear Jingoo Han,
> 
> > On Wednesday, November 27, 2013 5:37 PM, Pratyush Anand wrote:
> > > On Wed, Nov 27, 2013 at 04:24:47PM +0800, Marek Vasut wrote:
> > > > > On Wed, Nov 27, 2013 at 04:46:09AM +0800, Marek Vasut wrote:
> > > > > > Dear Pratyush Anand,
> > > > > >
> > > > > > > Hi,
> > > > > > >
> > > > > > > On Wed, Oct 23, 2013 at 12:55:43PM +0800, Tim Harvey wrote:
> > > > > > > > The IMX6 iATU is used for address translation between the
> > > > > > > > AXI bus address space and PCI address space.  This is used
> > > > > > > > for
> > > > > > > > type0 and type1 config cycles but is not necessary for
> > > > > > > > outbound io/mem regions.
> > > > > > > >
> > > > > > > > This patch removes the calls that inappropriately
> > > > > > > > re-configures the ATU viewport for outbound memory and IO
> > > > > > > > after config cycles and removes them altogether as they are not
> necessary.
> > > > > > > >
> > > > > > > > This resolves issues with PCI devices behind switches and
> > > > > > > > has been tested with a Gige device behind a PLX PEX860x switch.
> > > > > > > > More testing is needed for other configurations.
> > > > > > >
> > > > > > > It seems to me that in your controller you have only one
> > > > > > > view port. Also mem_base and mem_bus_addr is same. Thats why
> > > > > > > it works in your case.
> > > > > >
> > > > > > MX6 has 4 In/4 Out viewports AFAICT.
> > > > >
> > > > > Then if I do not miss anything, MX6 should work even without
> > > > > this patch.
> > > >
> > > > Yes, yet, it does not. On the other hand, this defect is only
> > > > problematic if you use the devices behind a switch. Do you use a
> > > > PCIe switch on your platform please?
> > >
> > > I had tested with device under bridge.
> > > Mohit is in process of validating SPEAr patches with latest kernel.
> > > Mohit, see if we can arrange a switch and test, once SPEAr platform
> > > is ready with latest kernel.
> >
> > I agree with Pratyush Anand's opinion.
> > One of our engineers had tested PCI cards under switch.
> > As far as I know, there was no issue about this.
> > However, currently, I don't have any switches. So, I Cannot test this
> > on Exynos platform.
> 
> OK, I will wait for Richard to come up with confirmation if possible. Looks
> like he's out of the office, so it might take a bit.
> 
[Richard] One Pericom PI7C9X2G303EL pcie switch, and two pcie ep deivces(one is intel e1000e nic,
 the other is one xhci device) are tested on imx6q sabresd board.
Without removing outbound io/mem regions view map during the cfg0/1 read/write cycle, both of these devices
can't work well at my side.
Works well after remove them during the cfg0/1 read/write cycles.

> > Mohit,
> > If you test a switch on SPEAr platform, it will be very helpful. Thank
> > you.
> >
> > Best regards,
> > Jingoo Han
> 
> Best regards,
> Marek Vasut

Best Regards
Richard Zhu