Re: [PATCH v6 1/2] x86/kexec: Build identity mapping for EFI systab and ACPI tables

[Junichi Nomura]

On 5/15/19 3:58 PM, Borislav Petkov wrote:
> On Wed, May 15, 2019 at 05:17:19AM +0000, Junichi Nomura wrote:
>> Hi Kairui,
>>
>> On 5/13/19 5:02 PM, Baoquan He wrote:
>>> On 05/13/19 at 09:50am, Borislav Petkov wrote:
>>>> On Mon, May 13, 2019 at 03:32:54PM +0800, Baoquan He wrote:
>>>> So we're going to try it again this cycle and if there's no fallout, it
>>>> will go upstream. If not, it will have to be fixed. The usual thing.
>>>>
>>>> And I don't care if Kairui's patch fixes this one problem - judging by
>>>> the fragility of this whole thing, it should be hammered on one more
>>>> cycle on as many boxes as possible to make sure there's no other SNAFUs.
>>>>
>>>> So go test it on more machines instead. I've pushed it here:
>>>>
>>>> https://git.kernel.org/pub/scm/linux/kernel/git/bp/bp.git/log/?h=next-merge-window
>>>
>>> Pingfan has got a machine to reproduce the kexec breakage issue, and
>>> applying these two patches fix it. He planned to paste the test result.
>>> I will ask him to try this branch if he has time, or I can get his
>>> machine to test.
>>>
>>> Junichi, also have a try on Boris's branch in NEC's test environment?
>>
>> while the patch set works on most of the machines I'm testing around,
>> I found kexec(1) fails to load kernel on a few machines if this patch
>> is applied.  Those machines don't have IORES_DESC_ACPI_TABLES region
>> and have ACPI tables in IORES_DESC_ACPI_NV_STORAGE region instead.
> 
> Why? What kind of machines are those?

I don't know.  They are just general purpose Xeon-based servers
and not some special purpose machines.  So I guess there are other
such machines in the wild.

> Why are the ACPI tables in NV storage?
> 
> Looking at crash_setup_memmap_entries(), it already maps that type so I
> guess this is needed.
> 
> + Rafael and leaving in the rest for reference.
> 
>  
>> So I think map_acpi_tables() should try to map both regions.  I tried
>> following change in addition and it worked.
>>
>> -- 
>> Jun'ichi Nomura, NEC Corporation / NEC Solution Innovators, Ltd.
>>
>>
>> diff --git a/arch/x86/kernel/machine_kexec_64.c b/arch/x86/kernel/machine_kexec_64.c
>> index 3c77bdf..3837c4a 100644
>> --- a/arch/x86/kernel/machine_kexec_64.c
>> +++ b/arch/x86/kernel/machine_kexec_64.c
>> @@ -56,12 +56,22 @@ static int mem_region_callback(struct resource *res, void *arg)
>>  {
>>  	unsigned long flags = IORESOURCE_MEM | IORESOURCE_BUSY;
>>  	struct init_pgtable_data data;
>> +	int ret;
>>  
>>  	data.info = info;
>>  	data.level4p = level4p;
>>  	flags = IORESOURCE_MEM | IORESOURCE_BUSY;
>> -	return walk_iomem_res_desc(IORES_DESC_ACPI_TABLES, flags, 0, -1,
>> -				   &data, mem_region_callback);
>> +	ret = walk_iomem_res_desc(IORES_DESC_ACPI_TABLES, flags, 0, -1,
>> +				  &data, mem_region_callback);
>> +	if (ret && ret != -EINVAL)
>> +		return ret;
>> +
>> +	ret = walk_iomem_res_desc(IORES_DESC_ACPI_NV_STORAGE, flags, 0, -1,
>> +				  &data, mem_region_callback);
>> +	if (ret && ret != -EINVAL)
>> +		return ret;
>> +
>> +	return 0;
>>  }
>>  #else
>>  static int map_acpi_tables(struct x86_mapping_info *info, pgd_t *level4p) { return 0; }

-- 
Jun'ichi Nomura, NEC Corporation / NEC Solution Innovators, Ltd.