Re: [PATCH v3] Add support for "info auxv" on NetBSD

[John Baldwin <jhb@FreeBSD.org>]

On 3/27/20 10:04 AM, Kamil Rytarowski wrote:
> On 27.03.2020 17:31, John Baldwin wrote:
>> On 3/26/20 4:26 PM, Kamil Rytarowski wrote:
>>> Ping?
>>>
>>> On 20.03.2020 18:27, Kamil Rytarowski wrote:
>>>> Register nbsd_auxv_parse() that overloads the default (Linux-style)
>>>> AUXV parsing. On NetBSD the type parameter is defined as int32_t
>>>> for all architectures.
>>
>> I would tone down some of the rhetoric.  FreeBSD uses the default AUXV
>> parsing, and I think Solaris does as well, so describing it as Linux-only
>> isn't very accurate.
>>
> 
> It is phrases as Linux-style, not Linux-only.

I would still perhaps drop the paranthetical part as it reads that way
I think even with -style vs -only.

>> (Similarly, I think the earlier reviews I saw around ptrace() claimed that
>> NetBSD was the only OS to use LWPs with ptrace() in the log messages in
>> effect which isn't really true as both Solaris and FreeBSD use LWPs
>> happily with ptrace(), just using a different convention.)
>>
> 
> The pair of (PID+LWP) for getters and setters of registers (among
> certain other ptrace operations) is NetBSD style ptrace(2) API design.
> This leads to the point that NetBSD is currently the only OS where
> get_ptrace_pid() is not compatible (at leat in the current form).

It might be a bit of how it was said, the actual text:

<quote>
    Unlike most other Operating Systems, NetBSD tracks both pid and lwp.
    The process id on NetBSD is stored always in the pid field of ptid.
</quote>

If by "track" you mean "the kernel keeps track of LWPs as distinct
from processes" (which is how I read "track"), then I think it is
inaccurate.


>> I would also tone this down a bit, and at least reference the correct
>> specification.  The ELF spec doesn't define the layout of auxv_t.  The
>> per-architecture psABI documents do.  Also, just saying that you follow
>> the spec doesn't help.  I would suggest something like:
>>
>> /* NetBSD-specific parser for AUXV entries.  NetBSD always uses an int
>>    to store the type as defined in the SVR4 psABI specifications rather
>>    than long as assumed by the default parser.  */
>>
> 
> This is toned down compared to obsd-tdet.c, that says:
> 
> /* Unlike Linux, OpenBSD actually follows the ELF standard.  */

That may be, but that probably isn't who I would choose as a model to
follow.

> OK, is this patch fine after rephrasing the above texts?

It looks fine to me, but it probably needs an approver such as Tom to ok it.
(I can approve FreeBSD-related things, but not really other bits.)

-- 
John Baldwin