[Crash-utility] [PATCH v3] makedumpfile: fix max_mapnr issue on system has over 44-bit addressing

[Jingbai Ma]

On 10/15/2013 04:22 PM, HATAYAMA Daisuke wrote:
> (2013/10/15 16:55), Jingbai Ma wrote:
>> On 10/15/2013 01:58 PM, HATAYAMA Daisuke wrote:
>>> (2013/10/14 21:16), Jingbai Ma wrote:
>>> <cut>
>>>> @@ -125,7 +126,7 @@ get_max_mapnr(void)
>>>> unsigned long long max_paddr;
>>>>
>>>> if (info->flag_refiltering) {
>>>> - info->max_mapnr = info->dh_memory->max_mapnr;
>>>> + info->max_mapnr = info->kh_memory->max_mapnr_64;
>>>> return TRUE;
>>>> }
>>>>
>>>
>>> Please:
>>>
>>> if (dh.header_version < 6)
>>> info->max_mapnr = info->dh_memmory->max_mapnr;
>>> else
>>> info->max_mapnr = info->kh_memory->max_mapnr_64;
>>
>> If we deal the max_mapnr_64 as below I did, we do not have to check
>> header_version everywhere when we need to the value max_mapnr. I just
>> set it to max_mapnr_64 regardless it's old version or new in the first
>> place we get it. It can simplify the code logic in all code. Or we
>> have to add this version check, it's also error prone.
>>
>
> No, it complicates code logic in the sense that header on the memory and
> header on the disk are not identical.
>
> It seems to me that the case where checking header version is definitely
> necessary is only when assigning either of max_mapnr values to
> info->max_mapnr,
> after which, it's enough to refer to info->max_mapnr.
>

OK, got it. I will send a new patch soon.
Thanks!

>>>
>>>> @@ -783,6 +784,10 @@ get_kdump_compressed_header_info(char *filename)
>>>> ERRMSG("header does not have dump_level member\n");
>>>> return FALSE;
>>>> }
>>>> +
>>>> + if (dh.header_version < 6)
>>>> + kh.max_mapnr_64 = dh.max_mapnr;
>>>> +
>>>
>>> Again, please don't do this. It's confusing if in-memory header data
>>> is not identical to in-disk one.
>>>
>>
>> I have explained the reason why I set the kh.max_mapnr_64 for old
>> version here.
>> If you still think we shouldn't change this value, and should check
>> header_version everywhere when need max_mapnr, I will send a new
>> version to change it.
>>
>>
>
>


-- 
Thanks,
Jingbai Ma