[lld][ELF][test] Expand testing of symbols in mergeable sections

[jhenderson]

Summary

Whilst reviewing some internal testing, I noticed a couple of holes in coverage of mergeable sections containing symbols. This patch addresses these holes:

1. Show that mid-piece symbols have their values updated properly when pieces are merged.
2. Show the behaviour of symbols in mergeable pieces when --gc-sections is enabled.

Also make a test more robust to address variations.

[MaskRay]

lld/test/ELF/merge-gc-sym.s

1

Thanks for adding coverage. There are gc-sections-merge*.s. Can you check whether these tests can be reorganized?

[grimar]

This change LG. merge-gc-sym.s name is consistent with merge-sym.s.

I'd do any futher reorganising/cleanup of gc-sections-merge*.s separatelly.

[jhenderson]

lld/test/ELF/merge-gc-sym.s

1

The gc-sections-merge*.s tests all are testing how difference reference variations cause mergeable pieces to be discarded. The three tests essentially translate into using R_X86_64_64 with no addend, R_X86_64_64 with an addend that could be confused with referencing the next piece, and R_X86_64_PC32, where the implicit addend could again cause the same confusion. These are all useful behaviours to test, and are independent of the impact on symbols in mergeable pieces. I think they belong as separate tests consequently (although perhaps renaming them/adding comments to explain things more explicitly might be worthwhile?).

What sort of reorganization did you have in mind?

[jhenderson]

• Rename merge-gc-sym.s to merge-sym-gc.s since it is a parallel test to merge-sym.s.
• Fixed merge-sym.s. The bar and foo symbols seem to get reordered. Their order doesn't matter, so to make the test robust, I have switched to use llvm-readelf (for single-line symbol descriptions) and CHECK-DAG.
• Whilst updating merge-sym.s, I also took the chance to remove the hard-coded address values, making the test more robust.

[jhenderson]

Upload correct diff.

[grimar]

Latest version LGTM too.

lld/test/ELF/merge-gc-sym.s

+# REQUIRES: x86
+## Show how symbols in GCed mergeable pieces behave.
+
+# RUN: llvm-mc -filetype=obj -triple=x86_64-pc-linux %s -o %t.o
+# RUN: ld.lld --gc-sections %t.o -o %t.elf
+# RUN: llvm-readelf %t.elf --sections --syms | FileCheck %s
+
+.section .rodata.merge,"aM",@progbits,4
+a1: ## Unreferenced. In first fragment, kept by a2 reference.
+    .short 1
+a2: ## Referenced.
+    .short 1
+b1: ## Unreferenced. Discarded as second fragment is unreferenced.
+    .short 1
+b2: ## Unreferenced. Discarded as second fragment is unreferenced.
+    .short 1
+c1: ## Referenced.
+    .short 1
+c2: ## Unreferenced. In third fragment, kept by c1 reference.
+    .short 1
+
+.data
+.global _start
+_start:
+    .quad a2
+    .quad c1
+
+# CHECK:      .rodata PROGBITS [[# %x, ADDR:]]
+
+# CHECK:      Symbol table '.symtab' contains 6 entries:
+# CHECK-NEXT:   Num:    Value          {{.*}} Ndx Name
+# CHECK-NEXT:     0:                   {{.*}} UND{{ *$}}
+# CHECK-NEXT:     1: {{0*}}[[#ADDR]]   {{.*}}     a1
+# CHECK-NEXT:     2: {{0*}}[[#ADDR+2]] {{.*}}     a2
+# CHECK-NEXT:     3: {{0*}}[[#ADDR]]   {{.*}}     c1
+# CHECK-NEXT:     4: {{0*}}[[#ADDR+2]] {{.*}}     c2
+# CHECK-NEXT:     5:                   {{.*}}     _start

lld/test/ELF/merge-sym.s

 // REQUIRES: x86
 // RUN: llvm-mc -filetype=obj -triple=x86_64-pc-linux %s -o %t.o
 // RUN: ld.lld %t.o -o %t.so -shared
 // RUN: llvm-readobj --symbols -S %t.so | FileCheck %s
 
         .section        .rodata.cst4,"aM",@progbits,4
         .short 0
 foo:
         .short 42
 
+        .short 0
+bar:
+        .short 42
 
 // CHECK:      Name: .rodata
 // CHECK-NEXT: Type: SHT_PROGBITS
 // CHECK-NEXT: Flags [
 // CHECK-NEXT:   SHF_ALLOC
 // CHECK-NEXT:   SHF_MERGE
 // CHECK-NEXT: ]
 // CHECK-NEXT: Address: 0x20D
 
 // CHECK:      Name: foo
 // CHECK-NEXT: Value: 0x20F
+// CHECK:      Name: bar
+// CHECK-NEXT: Value: 0x20F