Re: [PATCH v2 15/20] kexec: fill note buffers by NT_VMCORE_PAD notes in page-size boundary

[Zhang Yanfei <zhangyanfei@cn.fujitsu.com>]

于 2013年03月09日 11:46, HATAYAMA Daisuke 写道:
> From: Yanfei Zhang <zhangyanfei.yes@gmail.com>
> Subject: Re: [PATCH v2 15/20] kexec: fill note buffers by NT_VMCORE_PAD notes in page-size boundary
> Date: Fri, 8 Mar 2013 21:02:50 +0800
> 
>> 2013/3/8 HATAYAMA Daisuke <d.hatayama@jp.fujitsu.com>:
>>> From: Zhang Yanfei <zhangyanfei@cn.fujitsu.com>
>>> Subject: Re: [PATCH v2 15/20] kexec: fill note buffers by NT_VMCORE_PAD notes in page-size boundary
>>> Date: Thu, 7 Mar 2013 18:11:30 +0800
>>>
>>>> 于 2013年03月02日 16:37, HATAYAMA Daisuke 写道:
>>>>> Fill both crash_notes and vmcoreinfo_note buffers by NT_VMCORE_PAD
>>>>> note type to make them satisfy mmap()'s page-size boundary
>>>>> requirement.
>>>>>
>>>>> So far, end of note segments has been marked by zero-filled elf
>>>>> header. Instead, this patch writes NT_VMCORE_PAD note in the end of
>>>>> note segments until the offset on page-size boundary.
>>>>
>>>>
>>>> In the codes below, it seems that you assign name "VMCOREINFO" for
>>>> note type NT_VMCORE_PAD, right? This is kind of wired, i think. This
>>>> name has been used for NT_VMCORE_DEBUGINFO note already. Why not something
>>>> like "VMCOREPAD" or "PAD"?
>>>>
>>>
>>> It looks you are confusing or don't know name and type. The name is
>>> namespace and in the namespace, there are multiple note types, each of
>>> which has the corresponding data. In other words, data corresponding
>>> to types differ if they belong to differnet name space even if
>>> integers of the types are coincide with.
>>
>> Yes, I knew this. Just as the spec said " a program must recognize both the name
>> and the type to recognize a descriptor.". But I cannot understand what your word
>> "namespace" came from? I think you complicate simple things here.
>>
>> Only with a type, we cannot recognize a descriptor, because "multiple
>> interpretations of
>> a single type value may exist", So we should combine the name and the type
>> together. If both the name and type of two descriptors are the same,
>> we could say we
>> have two same descriptors. If one of them (type or name) are
>> different, we say the
>> two descriptors are different and the two notes have different data.
>>
>> If I am wrong, please correct me.
> 
> ??? I think you're saying here the same thing as my explanation above.
> 
> Although the term ''name space'' never occurs in ELF, it seems to me
> standard to represent the same values as different ones by combining
> additional elements as names to them.
> 
> Well, formally, it is represented as simply tuples or vector
> space. For example, support set S and S' and define new set S x S' by
> 
>   S x S' := { (s, s') | s in S, s' in S' }
> 
> and equality of the S x S' are defined as usual:
> 
>   (s1, s1') == (s2, s2') iff s1 == s2 and s1' == s2'.
> 
> In ELF, S is names and S' is types. There's no other formal meaning
> there.
> 
>>>
>>> The "VMCOREINFO" name represents information exported from
>>> /proc/vmcore that is used in kdump framework. In this sense,
>>> NT_VMCORE_PAD that is specific for /proc/vmcore and kdump framework,
>>> should belong to the "VMCOREINFO" name.
>>
>> I cannot understand the name explanation totally. Does the name really
>> have this meaning? Is there any authentic document? I was always thinking we
>> could feel free to name a name by ourselves!
> 
> Of course, it's optional for you to decide how to name notes within
> the mechanism. But it's important to treat naming for ease of managing
> note types. In addition to the above formal definition, it's important
> to consider what name gives us. It's readability, telling us that note
> types that belong to unique name are treated in common in the sense of
> the name. This is apart from the formal definition above.
> 
> It's certainly possible to distinguish notes by giving names only and
> not giving types. For example, imagine there are new 27 notes and they
> have different names but have 0 as type.
> 
> name          type
> "SOME_NOTE_A" 0
> "SOME_NOTE_B" 0
> ...
> "SOME_NOTE_Z" 0
> 
> Also, for example,
> 
> name        type
> "SOME_NOTE" 0     => NT_SOME_NOTE_A
> "SOME_NOTE" 1     => NT_SOME_NOTE_B
> ...
> "SOME_NOTE" 26    => NT_SOME_NOTE_Z
> 
> For the former case, it *looks to me* that space of time is not used
> effectively and it *looks to me* that space of name is not consumed
> efficiently.
> 
> After all, it amounts to individual preference about naming. I cannot
> say anything more.
> 

I see. I know what you mean now. I was just surprised and puzzled about your
"namespace" concept. 

Other than the name of NT_VMCORE_PAD, no complaints about the code.

Reviewed-by: Zhang Yanfei <zhangyanfei@cn.fujitsu.com>