On 12/02/18 18:41, Roger Pau Monné wrote:
> On Mon, Feb 12, 2018 at 03:04:21PM +0000, Andrew Cooper wrote:
>> On 12/02/18 14:39, Wei Liu wrote:
>>> On Mon, Feb 12, 2018 at 11:23:05AM +0000, Andrew Cooper wrote:
>>>>  .macro ALTERNATIVE oldinstr, newinstr, feature
>>>>  .L\@_orig_s:
>>>>      \oldinstr
>>>>  .L\@_orig_e:
>>>> +     .skip (-((repl_len(1) - orig_len) > 0) * (repl_len(1) - orig_len)), 
>>>> 0x90
>>> Seeing the negation at the beginning, I suppose this should also be a
>>> gas specific macro?
>> The build failures are because clang's integrated assembler can't cope
>> with non-absolute references with .skip, but we already know about this
>> and have code identical to this in tree.  (I temporarily removed it in
>> patch 4).
> Newer clang (6) supports .skip with labels, but doesn't support the
> (-(... And it's having some issues with the rest of the expression,
> will have to check more closely tomorrow.
> I wonder, what's Linux doing in this regard? It seems like clang/llvm
> is quite committed to support building Linux, so it might be good to
> follow suit in this case.

This is basically the same as what Linux does.  Linux unconditionally
uses -no-integrated-as.


Xen-devel mailing list

Reply via email to