Re: [PATCH 02/11] PCI: Try to allocate mem64 above 4G at first

[Bjorn Helgaas]

On Wed, May 30, 2012 at 1:40 AM, Steven Newbury <steve@xxxxxxxxxxxxxxx> wrote:
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
>
> On 30/05/12 00:27, Bjorn Helgaas wrote:
>> On Tue, May 29, 2012 at 2:40 PM, Yinghai Lu <yinghai@xxxxxxxxxx>
>> wrote:
>>> On Tue, May 29, 2012 at 12:23 PM, Bjorn Helgaas
>>> <bhelgaas@xxxxxxxxxx> wrote:
>>>> On Tue, May 29, 2012 at 12:17 PM, Yinghai Lu
>>>> <yinghai@xxxxxxxxxx> wrote:
>>>>> On Tue, May 29, 2012 at 10:57 AM, H. Peter Anvin
>>>>> <hpa@xxxxxxxxx> wrote:
>>>>>> On 05/29/2012 10:55 AM, Yinghai Lu wrote:
>
> *SNIP*
>
>>>
>>>
>>> ok, please check the version, that put back PCI_MAX_RESOURCE_32
>>> for io ports.
>>>
>>> also RFC to limit for 16 bit ioport handling.  only help other
>>> arches that does support 32bit ioports but have bridges only
>>> support 16bit io ports.
>>
>> I don't understand this one at all.  It looks like you mashed
>> together at least two changes: (1) prefer I/O space above 64K if
>> available, and (2) mark secondary bus resources with
>> IORESOURCE_IO_32 when the P2P bridge I/O window address decode type
>> is PCI_IO_RANGE_TYPE_32 and use that to limit allocations.
>>
>> I don't see the justification for (1).  What problem does this
>> solve?
>>
> I can potentially see this helping with hotplug, where a new device
> has only 16 bit io ports, but if there's no space remaining under 64k
> allocation would fail.  Preferring above 64k means preserving that
> limited resource.  This is exactly equivalent to my original
> motivation for preferring 64 bit PCI mem in order to preserve 32 bit
> address space for hotplug devices without 64 bit BARs.

You're right, the spec does allow the upper 16 bits of I/O BARs to be
hardwired to zero (PCI spec rev 3.0, p. 226), so this part does make
some sense.  I don't think it applies to x86, since I don't think
there's a way to generate an I/O access to anything above 64K, but it
could help other arches.

I'm inclined to be conservative and wait until we find a problem where
a patch like this would help.
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/