[U-Boot] [PATCH 3/5] x86: queensbay: Support pre-relocation dm pci

[Simon Glass]

Hi Bin,

On 5 August 2015 at 20:48, Bin Meng <bmeng.cn at gmail.com> wrote:
> Hi Simon,
>
> On Wed, Aug 5, 2015 at 10:39 PM, Simon Glass <sjg at chromium.org> wrote:
>> Hi Bin,
>>
>> On 2 August 2015 at 19:52, Bin Meng <bmeng.cn at gmail.com> wrote:
>>> Hi Simon,
>>>
>>> On Mon, Aug 3, 2015 at 6:31 AM, Simon Glass <sjg at chromium.org> wrote:
>>>> Hi Bin,
>>>>
>>>> On 27 July 2015 at 01:33, Bin Meng <bmeng.cn at gmail.com> wrote:
>>>>> Increase CONFIG_MALLOC_F_LEN so that dm pci does not fail with -ENOMEM
>>>>> before relocation. This makes pci uart work again on Intel Crown Bay.
>>>>>
>>>>> Signed-off-by: Bin Meng <bmeng.cn at gmail.com>
>>>>> ---
>>>>>
>>>>>  arch/x86/cpu/queensbay/Kconfig |  3 +++
>>>>>  arch/x86/cpu/queensbay/tnc.c   | 13 +++++++++++++
>>>>>  arch/x86/dts/crownbay.dts      |  2 --
>>>>>  3 files changed, 16 insertions(+), 2 deletions(-)
>>>>>
>>>>> diff --git a/arch/x86/cpu/queensbay/Kconfig b/arch/x86/cpu/queensbay/Kconfig
>>>>> index fbf85f2..8eb619c 100644
>>>>> --- a/arch/x86/cpu/queensbay/Kconfig
>>>>> +++ b/arch/x86/cpu/queensbay/Kconfig
>>>>> @@ -42,4 +42,7 @@ config CPU_ADDR_BITS
>>>>>         int
>>>>>         default 32
>>>>>
>>>>> +config SYS_MALLOC_F_LEN
>>>>> +       default 0x8000
>>>>
>>>> 32KB? Wow that's a huge amount. How much does it actually use? Perhaps
>>>> we should change PCI so that it respects the u-boot,dm-pre-reloc
>>>> property?
>>>>
>>>
>>> I just add a zero to the original value :) I did not measure the
>>> actual memory consumption since after fsp_init() we have plenty of
>>> memory to be used as the stack. For queensbay platform, it has lots of
>>> pci devices each of which will need allocate some memory for dm pci to
>>> use.
>>
>> I think it is worth checking this and using a value closer to what you
>> need. board_init_f.c has a debug output for it I think.
>>
>
> OK, I will have a check.
>
>> Also I wonder if we can avoid creating PCI devices for everything
>> pre-relocation, when we only need serial?
>>
>
> Yes, looks that we only need serial. However x86 is a specific case
> since pci bus is the fundamental bus that is used to set up other
> devices. After conversion to dm pci, if we implement it correctly, any
> call to a pci api will trigger the enumeration and it is probably a
> side effect of using dm?  I even think we can hardcoded pci uart in
> the pre-relocation phase without using dm at all, but that is not a
> portable way.

Yes that's right. Of course for the UART we can do a special case at
least for the debug UART, since that should be available early (before
DM).

How about this:

- early stage, before board_init_f(): use special x86 PCI functions, enable UART
- pre-reloc: set up only a few devices which are needed before relocation
- post-reloc: full scan of PCI

Regards,
Simon