Re: [PATCH 2/2] PCI: Only enable realloc auto when root bus has 64bit mmio

[Joseph Salisbury <joseph.salisbury@canonical.com>]

On 01/10/2014 12:13 PM, Yinghai Lu wrote:
> On Fri, Jan 10, 2014 at 8:19 AM, Joseph Salisbury
> <joseph.salisbury@canonical.com> wrote:
>> On 12/11/2013 02:55 PM, Yinghai Lu wrote:
>>> On Wed, Dec 11, 2013 at 11:19 AM, Joseph Salisbury
>>> <joseph.salisbury@canonical.com> wrote:
>>>> On 12/09/2013 03:10 PM, Yinghai Lu wrote:
>>>>> On Mon, Dec 9, 2013 at 11:42 AM, Bjorn Helgaas <bhelgaas@google.com> wrote:
>>>>>> That doesn't answer my question at all.
>>>>>>
>>>>>> I understand that this change makes it so Joseph doesn't have to use
>>>>>> "pci=realloc=off".  But why should auto-reallocation be limited to
>>>>>> buses that have resources above 4GB?  That doesn't make any sense.
>>>>>>
>>>>>> We should fix the reallocation code so it can deal with this case.  If
>>>>>> there's not enough space for everything, obviously we have to leave
>>>>>> something unassigned.  A ROM BAR is a good candidate for leaving
>>>>>> unassigned, because most of the time we can get along without it.
>>>>> Yes, that is ideal and not that simple.
>>>>> but that would be hard to backport to old kernels.
>>>>>
>>>>> BTW, Joseph, can you try
>>>>> git://git.kernel.org/pub/scm/linux/kernel/git/yinghai/linux-yinghai.git
>>>>> for-pci-3.14
>>>>> with pci=realloc=on
>>>>>
>>>>> on that system?
>>>>>
>>>>> Thanks
>>>>>
>>>>> Yinghai
>>>> I noticed there was some back and forth on this thread.  Do you still
>>>> want me to test this version, Yinghai?
>>> Yes, if that works, we would not need to put the patch in upstream for limiting
>>> realloc auto scope.
>>>
>> Another user has confirmed that at test kernel from your branch[0] does
>> resolve the bug.
> Hi, Joe,
>
> Bjorn does not want to limit auto realloc, so this patch can not make
> to upstream at this point.
>
> so we may have to ask user to append "pci=realloc=off" before we find
> more smart way to realloc
> resource.
>
> Thanks
>
> Yinghai

Hi Yinghai,

A new bug[0] was opened due to enabling PCI_REALLOC_ENABLE_AUTO, which
is similar to the original bug[1] we discussed.

Just wondering if there have been any additional ideas on realloc since
this was last discussed?

Thanks,

Joe


[0] http://pad.lv/1363313
[1] http://pad.lv/1245938