[lld] [lld][ELF] Introduce an option to keep data section prefix. (PR #148985)

[Mingming Liu via llvm-commits]

================
@@ -103,13 +103,48 @@ StringRef LinkerScript::getOutputSectionName(const InputSectionBase *s) const {
     return ".text";
   }
 
-  for (StringRef v : {".data.rel.ro", ".data",       ".rodata",
-                      ".bss.rel.ro",  ".bss",        ".ldata",
-                      ".lrodata",     ".lbss",       ".gcc_except_table",
-                      ".init_array",  ".fini_array", ".tbss",
-                      ".tdata",       ".ARM.exidx",  ".ARM.extab",
-                      ".ctors",       ".dtors",      ".sbss",
-                      ".sdata",       ".srodata"})
+  // When zKeepDataSectionPrefix is true, keep .hot and .unlikely suffixes
+  // in data sections.
+  static constexpr StringRef dataSectionPrefixes[] = {
+      ".data.rel.ro", ".data", ".rodata", ".bss.rel.ro", ".bss",
+  };
+
+  for (auto [index, v] : llvm::enumerate(dataSectionPrefixes)) {
+    StringRef secName = s->name;
+    if (!secName.consume_front(v))
+      continue;
+    if (!secName.empty() && secName[0] != '.') {
+      continue;
+    }
+
+    // The section name starts with 'v', and the remaining string is either
+    // empty or starts with a '.' character. If keep-data-section-prefix is
+    // false, map s to output section with name 'v'.
+    if (!ctx.arg.zKeepDataSectionPrefix)
+      return v;
+    // Preserve .hot or .unlikely suffixes in data sections with
+    // keep-data-section-prefix=true.
+    if (isSectionPrefix(".hot", secName))
+      return s->name.substr(0, v.size() + 4);
+    if (isSectionPrefix(".unlikely", secName))
+      return s->name.substr(0, v.size() + 9);
+    if (index == 2) {
----------------
mingmingl-llvm wrote:

> I think it's ok for linker to map input .bss.hot. to .bss.hot even if compilers don't generate .bss.hot. .

yes, it's ok to map `.bss.hot.` to `.bss.hot`, in which case `if (isSectionPrefix(".hot", secName))` at L127 will return `.bss.hot`, and code at L131 `(if index==2)` won't run.

As discussed in https://github.com/llvm/llvm-project/pull/148985#discussion_r2221133248 and https://github.com/llvm/llvm-project/pull/148985#discussion_r2233382641,  patterns like `.rodata.str1.1.unlikely.` or `.rodata.cst4.hot.` are not mapped before `if (index==2)` because of the strings between section name and hot/unlikely prefix, and `if(index==2)` is added to map them. Meanwhile, we can use `s->name.substr` to return a StringRef by referencing `s->name`.

To get rid of `if (index==2)` specialization, one implementation would be something like [1]. If this version is more readable, I can update the patch to it.

[1]

```
// records the hot prefixes in data array. 
// Note `bss.rel.ro` doesn't have hot prefix.
static constexpr StringRef dataSectionHotPrefixes[] = {
      ".data.rel.ro.hot", ".data.hot", ".rodata.hot", ".bss.rel.ro", ".bss.hot",
  };
// records the unlikely suffixes in data array
// Note `bss.rel.ro` doesn't have unlikely prefix.
static constexpr StringRef dataSectionUnlikelyPrefixes[] = {
      ".data.rel.ro.unlikely", ".data.unlikely", ".rodata.unlikely", ".bss.rel.ro", ".bss.unlikely",
  };

for (auto [index, v] : llvm::enumerate(dataSectionPrefixes)) {
   ... // code before `ctx.arg.zKeepDataSectionPrefix`
   if (!ctx.arg.zKeepDataSectionPrefix)
       return v;
    // We require the trailing dot after hot or unlikely to exist to map them
    // to hot or unlikely variant.
     if (s->name.ends_with(".hot.")
        return dataSectionHotPrefixes[index];
     if (s->name.ends_with(".unlikely.")
        return dataSectionUnlikelyPrefixes[index]; 
}

```



https://github.com/llvm/llvm-project/pull/148985