Re: [PATCH 1/2 v6] x86/kexec_file: add e820 entry in case e820 type string matches to io resource name

[lijiang]

å 2018å11æ15æ 13:58, Dave Young åé:
> On 11/15/18 at 01:44pm, lijiang wrote:
>> å 2018å11æ14æ 19:26, Borislav Petkov åé:
>>> On Wed, Nov 14, 2018 at 03:29:25PM +0800, Lianbo Jiang wrote:
>>>> When load the kernel image and initramfs by kexec_file_load syscall, it can
>>>> not add exact e820 reserved type to kdump kernel e820 table.
>>>>
>>>> Kdump uses walk_iomem_res_desc() to iterate io resources, then adds matched
>>>> desc to e820 table for kdump kernel. But, when convert the e820 type into
>>>> the iores descriptors, several e820 types are converted to 'IORES_DES_NONE'
>>>> in this function e820_type_to_iores_desc(). So the walk_iomem_res_desc()
>>>> will get these unnecessary types(E820_TYPE_RAM/E820_TYPE_UNUSABLE/E820_TYPE
>>>> _KERN) when iterate io resources by the 'IORES_DES_NONE'.
>>>>
>>>> It needs filter out these redundant type(such as E820_TYPE_RAM/E820_TYPE_
>>>> UNUSABLE/E820_TYPE_KERN) in order to add exact e820 reserved type to kdump
>>>> kernel e820 table. Thus it also needs an extra checking in memmap_entry_
>>>> callback() to match the e820 type and resource name.
>>>
>>> Ok, it took me a while to parse what this is trying to say so let's
>>> start from the top:
>>>
>>> * What resource type do you do need in the second kernel?
>>>
>>
>> Thanks for your comment.
>>
>> The e820 reserved ranges need to be passed to the second kernel.
>>
>>> * The most important question: why?
>>>
>>
>> At present, the upstream kernel does not pass the e820 reserved ranges to the
>> second kernel, which might cause two problems:
>>
>> The first one is the MMCONFIG issue, the PCI MMCONFIG(extended mode) requires
>> the reserved region otherwise it falls back to legacy mode, which might lead to
>> the hot-plug device could not be recognized in kdump kernel.
>>
>> Another one is that the e820 reserved ranges do not setup in kdump kernel, which
>> could cause kdump can't work in some machines. To know more information, please
>> refer to the [PATCH 2/2 v6] patch log.
>>
>>
>>> * If it is the reserved resource, why aren't you adding
>>> IORES_DESC_RESERVED or so which to look for instead of this hacky string
>>> comparison?
>>>
>>
>> Adding the new descriptor 'IORES_DESC_RESERVED' is also a good solution. I will
> 
> I was not sure if something else depends on IORES_DESC_NONE and if it is
> easy to split it and add IORES_DESC_RESERVED
> 
> But if you can prove it is safe then it would be a better way.
> 
Thank you, Dave.

These two solutions should be feasible, they can work very well on my machine.

Regards,
Lianbo

> Thanks
> Dave
>