[lld] [lld] Add infrastructure for handling RISCV vendor-specific relocations. (PR #159987)

[Sam Elliott via llvm-commits]

================
@@ -1476,3 +1482,54 @@ void elf::mergeRISCVAttributesSections(Ctx &ctx) {
 }
 
 void elf::setRISCVTargetInfo(Ctx &ctx) { ctx.target.reset(new RISCV(ctx)); }
+
+template <class ELFT, class RelTy>
+void RISCV::scanSectionImpl(InputSectionBase &sec, Relocs<RelTy> rels) {
+  RelocScan rs(ctx, &sec);
+  // Many relocations end up in sec.relocations.
+  sec.relocations.reserve(rels.size());
+
+  StringRef rvVendor;
+  for (auto it = rels.begin(); it != rels.end(); ++it) {
+    RelType type = it->getType(false);
+    uint32_t symIndex = it->getSymbol(false);
+    Symbol &sym = sec.getFile<ELFT>()->getSymbol(symIndex);
+    const uint8_t *loc = sec.content().data() + it->r_offset;
+
+    if (type == R_RISCV_VENDOR) {
+      if (!rvVendor.empty())
+        Err(ctx) << getErrorLoc(ctx, loc)
+                 << "malformed consecutive R_RISCV_VENDOR relocations";
+      rvVendor = sym.getName();
+      continue;
+    } else if (!rvVendor.empty()) {
+      Err(ctx) << getErrorLoc(ctx, loc)
----------------
lenary wrote:

The idea is that, once implemented, we'll override `type` in here to become a vendor-specific value (i.e. a value that can differentiate QUALCOMM's CUSTOM192 from e.g. CherIoT's CUSTOM192)?

And then those will be mapped back to the conventional RelExpr?

This is not a blocking comment, I'm just trying to understand the next step for someone implementing their relocations.

https://github.com/llvm/llvm-project/pull/159987