Re: [Xen-devel] [PATCH v2 22/23] x86: make Xen early boot code relocatable

[Jan Beulich]

>>> On 31.08.15 at 21:49, <address@hidden> wrote:
> On Fri, Aug 28, 2015 at 08:16:05AM -0600, Jan Beulich wrote:
>> >>> On 28.08.15 at 15:42, <address@hidden> wrote:
>> > Now that said - do you have suggestions on how to make this work
>> > with GRUB in the picture?
>>
>> ... I don't think I'm the one to make suggestions on how to make
>> things work with grub in the picture when I continue to be of the
>> opinion that it shouldn't have been brought into the picture in the
>> first place.
> 
> Could you be more specific what is wrong with this patch or at least last
> hunk which you reviewed? What is real technical reason that it could not
> be accepted? If idea is wrong in general please tell me where and why.
> Otherwise I am not able to work out other better solution.

The patch may not be wrong technically (and I never said so), it is
just that the way you carry things out is too intrusive for my taste.
Since Andrew is happy with the change in principle, I think I wouldn't
veto it going in with his R-b.

> By the way, once I have put 3 (IIRC) proposals for this problem on the 
> table.
> Even we discussed this issue in Shanghai. You and Andrew approved more or
> less this one. So, I am a bit disappointed that you withdraw your approval
> (yes, partial but still the approval) at this stage with just vague 
> explanations.

I've never approved of anything here. At the hackathon we've only
hashed out possible options.

>> But for the purely technical (patch) aspect: Anything (e.g.
>> macroization such that at least some sym_phys() uses can remain
>> untouched) allowing to limit the impact of said patch on the source
>> code (thus helping review and perhaps also long term
>> maintainability) would be a step towards talking me into
>> withdrawing my objection.
> 
> Ditto. This is too vague. So, I will be very grateful if you review this
> patch until the end or at least tell me what (if you add why it will be
> nice) exactly should be fixed.

How is this to vague? I gave a possible direction (macroization) as
well as the criteria (less impact on existing code; to be slightly more
precise I'd specifically like to see most of the open coded %fs: uses
gone).

Again - if Andrew thinks this is the right thing to do, I'll defer to him
for reviewing these final few patches of the series, since as said
before to me this is a workaround for a misguided design, and as
such I'm not willing to accept as intrusive a patch as this one is. (And
no, I don't really buy his argument of Xen boot code needing to be
relocatable anyway - the legacy boot path doesn't really need to
fear there not being free memory right above 1Mb. At least I have
seen no proof of there being any system [supporting legacy boot]
where Xen can't boot that way. And even if there was, it would
still need to be determined whether this is legitimately so, or again
due to poorly written firmware.)

Jan