Re: [PATCH 4/4] PCI: fix the io resource alignment calculation in pbus_size_io()

[Wei Yang <weiyang@linux.vnet.ibm.com>]

On Tue, Aug 06, 2013 at 11:58:56AM -0600, Bjorn Helgaas wrote:
>On Tue, Aug 06, 2013 at 11:34:10PM +0800, Wei Yang wrote:
>> On Tue, Aug 06, 2013 at 07:39:29AM -0600, Bjorn Helgaas wrote:
>> >> Personal Conclusion
>> >> ============================================================================
>> >> I think Bjorn's patch works.
>> >> Will test on powernv platform and give the result.
>> >
>> >Great, let me know what happens.
>> 
>> In both case, with/with out your patch, the assignment result is the same.
>> Below is the /proc/ioports file.
>> 
>> 00010000-0080ffff : /io-hub@3efe00000000/pciex@3efe00080000
>>   00010000-00010fff : Legacy IO
>>   00020000-0003ffff : PCI Bus 0000:01
>>     00020000-0003ffff : PCI Bus 0000:02
>>       00020000-0002ffff : PCI Bus 0000:40
>>       00030000-0003ffff : PCI Bus 0000:a0
>> 00820000-0101ffff : /io-hub@3efe00000000/pciex@3efe00090000
>>   00820000-00820fff : Legacy IO
>>   00830000-0083ffff : PCI Bus 0001:01
>> 01030000-0182ffff : /io-hub@3efe00000000/pciex@3efe000a0000
>>   01030000-01030fff : Legacy IO
>>   01040000-0104ffff : PCI Bus 0002:01
>> 01840000-0203ffff : /io-hub@3efe00000000/pciex@3efe000b0000
>>   01840000-01840fff : Legacy IO
>>   01850000-0185ffff : PCI Bus 0003:01
>> 02050000-0284ffff : /io-hub@3efe00000000/pciex@3efe000c0000
>>   02050000-02050fff : Legacy IO
>>   02060000-0206ffff : PCI Bus 0004:01
>> 02860000-0305ffff : /io-hub@3efe00000000/pciex@3efe000d0000
>>   02860000-02860fff : Legacy IO
>>   02870000-0287ffff : PCI Bus 0005:01
>> 
>> The arch_align is 0x10000. When sizing for Bus 0000:01, its child has a I/O
>> bridge window with size 0x20000/alignment 0x10000. And the result
>> [0x20000,0x3ffff] looks like size 0x20000/alignment 0x20000 or size
>> 0x20000/alignment 0x10000. 
>
>Obviously, an address with alignment of 0x20000 is *also* aligned to
>0x10000.
>
>> But I still think the alignment is 0x10000.
>
>Yes, as it should be.  These are all bridge windows, which never have to be
>aligned at more than 1K (if supported), 4K, or the arch alignment,
>whichever is largest.

I add some printk which shows the alignment is 0x10000 instead of 0x20000.

>
>I *think* you're saying that the patch works correctly.

Yes.

>
>Bjorn

-- 
Richard Yang
Help you, Help me