On Wed, 2015-07-22 at 18:08 +0800, Zhiqiang Hou wrote:

From: Hou Zhiqiang <redacted>

You CCed this to 
b21284@freescale.com.  Who is that?  It would be nice to use "friendly" 
e-mail addresses, but at least include the name along with the e-mail address.

I suggest CCing the people who added these device trees.

Impact board list:
P1020MBG-PC. P1022DS, P2020RDB
All above boards have its PCIE memory range less than 0xbfff_ffff,

If you mean that the physical address of the memory region is <= 0xbfff_ffff, 
I don't see the relevance.

but in dts its boundary value was 0xe0000000. Both of them was maped
to the same boundary 0xe0000000 which was Overlapped and crossed.

By "boundary" do you mean the PCIe bus address?  Why is it a problem for 
these independent PCIe root complexes to have the same PCIe bus addresses?

 Cpu will access the illicit memery addr and detect error then lead to
cpu stall.  So update dts for these boards.

What is illicit about it?

Why isn't the problem seen in the 36-bit device trees, which do the same 
thing?

-Scott