Thanks @nickdesaulniers for bringing this to the finish line!

• Added comment why allowing cp10/cp11 make sense
• Use check prefixes in coprocessors.s in a sensible mannor

In D59733#2106433, @simon_tatham wrote:

for ARMv7 we only disallowed this pair, and this change removes that restriction. So now, all of them are allowed...

Yes, for Arm v7.

What I'm getting at is: should we turn the whole of isValidCoprocessorNumber into warnings of this kind, on the same principle (that it's OK to write any CDP instruction if it's in the context of code that will dynamically decide at run time which instructions to actually execute)? If not, what makes this v7 constraint different from the v8 ones?

@simon_tatham not sure what you exactly mean, for ARMv7 we only disallowed this pair, and this change removes that restriction. So now, all of them are allowed...

In D59733#1441197, @peter.smith wrote:

From what I can see the mrc in venum_read_pmresr() is not a VFP/Neon instruction. It hails from arch/arm/kernel/perf_event_v7.c and is specific to Qualcomm's Scorpion and Krait CPUs. My understanding is that CP15 is used for the PMU in v7, it is possible that the use of p10 there is a mistake but I'm not knowledgeable enough about Scorpion and Krait to know whether this is the case.

Thanks for the hint! Did pretty much all of that, but used git am instead of arc, thought it does not really make a difference. Will use arc next time!

Committed on master: https://github.com/llvm/llvm-project/commit/f87563661d6661fd4843beb8f391acd077b08dbe

@nickdesaulniers I actually tired to get commit access myself but haven't heard back from Chris yet. So yes would be good if somebody could commit this already.

@efriedma, @jcai19, thanks for feedback, updated the patch accordingly. This now assembles and disassembles to the same encoding as binutils does (#0).

@efriedma I tried the InstAlias approach at first, but it did not work for reasons I don't exactly recall.

This indeed looks much nicer. I was somehow under the assumption that this was not an option since we moved away from using separate instructions for {ld,st}r{,b}t:
https://github.com/llvm/llvm-project/commit/fb3950ec6312dfa4317d8cbf83a1db4aae7428ce

Thanks for the review. I do not have commit access, so I would be glad if somebody could merge that for me.

• add .w alias for dmb/dsb/isb with argument

In D74927#1892934, @psmith wrote:

[..]
Apologies for all the details, there may be a simpler way of doing this.

In D74927#1891751, @MaskRay wrote:

[...]
On the Linux kernel side, it'd be nice to make a change anyway if that moves us away from an edge case.

• Check if Symbol has previously been defined
• Remove unnecessary cast
• Mark function as override

In my test case using Linux compiled for Arm using allyesconfig and ld.lld with these changes applied linking fails as follows:

With that building CONFIG_ARCH_MULTI_V6=y seems to work. Also LLVM emits the correct CPU:

I seem to have a related issue: I am using -march=armv6k and -no-integrated-as, this generates the following output: