[U-Boot] [PATCH] pci: Avoid assigning PCI resources that are below 0x1000

[Stefan Roese]

Hi Bin,

On 05.06.19 16:10, Bin Meng wrote:
> Hi Stefan,
> 
> On Wed, Jun 5, 2019 at 9:53 PM Stefan Roese <sr at denx.de> wrote:
>>
>> On 05.06.19 15:07, Bin Meng wrote:
>>> commit b7598a43f2b4 ("[PATCH] Avoid assigning PCI resources from
>>> zero address") only moved the bus lower address to 0x1000 if the
>>> given bus start address is zero. The comment said 0x1000 is a
>>> reasonable starting value, hence we'd better apply the same
>>> adjustment when the given bus start address is below 0x1000.
>>>
>>> Signed-off-by: Bin Meng <bmeng.cn at gmail.com>
>>> ---
>>>
>>>    drivers/pci/pci_auto_common.c | 7 +++++--
>>>    1 file changed, 5 insertions(+), 2 deletions(-)
>>>
>>> diff --git a/drivers/pci/pci_auto_common.c b/drivers/pci/pci_auto_common.c
>>> index 1837873..3ff42f2 100644
>>> --- a/drivers/pci/pci_auto_common.c
>>> +++ b/drivers/pci/pci_auto_common.c
>>> @@ -21,9 +21,12 @@ void pciauto_region_init(struct pci_region *res)
>>>        /*
>>>         * Avoid allocating PCI resources from address 0 -- this is illegal
>>>         * according to PCI 2.1 and moreover, this is known to cause Linux IDE
>>> -      * drivers to fail. Use a reasonable starting value of 0x1000 instead.
>>> +      * drivers to fail. Use a reasonable starting value of 0x1000 instead
>>> +      * if the bus start address is below 0x1000.
>>>         */
>>> -     res->bus_lower = res->bus_start ? res->bus_start : 0x1000;
>>> +     if (res->bus_start < 0x1000)
>>> +             res->bus_start = 0x1000;
>>> +     res->bus_lower = res->bus_start;
>>>    }
>>>
>>>    void pciauto_region_align(struct pci_region *res, pci_size_t size)
>>>
>>
>> Reviewed-by: Stefan Roese <sr at denx.de>
>>
> 
> Thanks for the review. I just re-considered it a little bit, and
> thought that we should not change res->bus_start here, as the original
> logic did not change too. I will send a v2.

Even better. I was a bit reluctant when I saw this change, but trusted
you to have thought it through. ;)

BTW: Where did you spot a problem with the current implementation?
What issue did hit you exactly?

Thanks,
Stefan