[U-Boot] [RFC PATCH] driver: pci: Fix regions for local memory

[Simon Glass]

Hi York,

On 16 March 2017 at 21:29, york sun <york.sun at nxp.com> wrote:
> On 03/16/2017 08:14 PM, Simon Glass wrote:
>> Hi York,
>>
>> On 16 March 2017 at 16:47, Simon Glass <sjg at chromium.org> wrote:
>>> On 9 February 2017 at 11:35, York Sun <york.sun at nxp.com> wrote:
>>>> When adding local memory to PCI region, gd->ram_size is correct only
>>>> if the memory is in one continuous block. In case memory is split
>>>> into several banks, each bank should be added separately.
>>>>
>>>> Signed-off-by: York Sun <york.sun at nxp.com>
>>>> CC: Simon Glass <sjg at chromium.org>
>>>> ---
>>>> It was spotted when I was rewriting the code to reserve secure memory
>>>> and forgot to reduce gd->ram_size. PCIe resumes working after fixing
>>>> gd->ram_size. For my case, the memory is split into two banks. So
>>>> base + gd->ram_size is not in memory. I don't know how it worked before.
>>>> This change seems reasonable without digging into PCI code.
>>>>
>>>>  drivers/pci/pci-uclass.c | 18 +++++++++++++++++-
>>>>  1 file changed, 17 insertions(+), 1 deletion(-)
>>>>
>>>
>>> Acked-by: Simon Glass <sjg at chromium.org>
>>
>> Unfortunately this breaks chromebook_link (x86).
>>
>> Do you have any ideas or should I dig into it?
>>
>
> Sorry I have no idea. If you can look into it, that will be great.

The problem is that x86 sets up PCI before relocation, thus before
gd->bd is available.

I think several changes are needed:

1. Something like:

bool use_dram_banks = false;

#ifdef CONFIG_NR_DRAM_BANKS
use_dram_banks = gd->bd != NULL;
#endif
if (use_dram_banks) {
   your new code
} else {
   old code
}

2. DRAM banks can have a size of 0, so check for that and skip if needed.

3. Check you don't overflow the size of the pci controller regions[] array.

Sorry I didn't notice this when reviewing it. It looked fine to me!

Regards,
Simon