[ELF] - Linkerscript: implemented SUBALIGN() command.

[grimar]

Summary

You can force input section alignment within an output section by using SUBALIGN. The
value specified overrides any alignment given by input sections, whether larger or smaller.

SUBALIGN is used in many projects in the wild, for example:

https://github.com/kubo39/tinycr/blob/536a8a0ba290b08659d347a0bdddb966cff44441/script.ld
https://github.com/def-/nim-small-coreutils/blob/3cb5e99ab8435220c82b9c307ce8909841655cc8/tools/x86_64/script.ld
https://github.com/takeneco/uniqos/blob/57e802d54979f1f02942e0260687c417e46c58df/arch/x86_64/boot/fd/bootsect/bootsect.ld
https://github.com/sjrct/Frosk/blob/13ee75a64ae98c652c32795bcc5d17ae459d2cfc/kernel/kernel.ld

and many others.

Patch implements it with a single restriction: access to location counter in expression is forbidden. That should not be a problem
as every script I saw uses just integer value here. And the reasons of that is that we assign offsets before we know VA of section,
so we do not know the location counter value. Solution would be to layout input sections right before assign address to output section I guess,
but as I mentioned before that does not seems reasonable to implement for this command.

[ruiu]

I think that it is better to overwrite each input section's Alignment member variable.

ELF/LinkerScript.h

29–30 ↗

(On Diff #66472)

Hmm, this seems too complicated.

85 ↗

(On Diff #66472)

I'd treat "subalign" as one word, so I'd name SubalignExpr.

[ruiu]

ELF/LinkerScript.cpp

350 ↗

(On Diff #66472)

If you want to detect this error, you need to do this at parse time rather than calculation time. Being said that, I think we should just ignore the error as it complicates the code by introducing DotExpr type.

[grimar]

• Addressed review comments.

[ruiu]

ELF/LinkerScript.cpp

289 ↗

(On Diff #68363)

Do not call SubalignExpr in this loop because it's a constant value.

630 ↗

(On Diff #68363)

SubAlign -> Subalign.

915 ↗

(On Diff #68363)

Use readParenExpr.

[grimar]

• Addressed review comments.

[grimar]

ELF/LinkerScript.cpp

289 ↗

(On Diff #68363)

Yes, I thought about that when wrote that initially. Did it to simplify the code a bit. There should be not many output sections and probably none or just few of them them will use SUBALIGN(), so selected one line shorter way.
Fixed.

915 ↗

(On Diff #68363)

Done.

ELF/LinkerScript.h

85 ↗

(On Diff #66472)

Done.

[ruiu]

ELF/LinkerScript.cpp

287 ↗

(On Diff #68489)

Alignment can never be 0 (in LLD we canonicalize 1 as 0). So how about setting Sublign to 0 if SubalignExpr is absent?

289 ↗

(On Diff #68489)

Then you can do

if (Subalign)
  Sec->Alignment = Subalign;

[grimar]

• Addressed review comments.

[ruiu]

LGTM

lld/trunk/ELF/LinkerScript.h

 struct OutputSectionCommand : BaseCommand {
   OutputSectionCommand(StringRef Name)
       : BaseCommand(OutputSectionKind), Name(Name) {}
   static bool classof(const BaseCommand *C);
   StringRef Name;
   Expr AddrExpr;
   Expr AlignExpr;
   Expr LmaExpr;
+  Expr SubalignExpr;
   std::vector<std::unique_ptr<BaseCommand>> Commands;
   std::vector<StringRef> Phdrs;
   std::vector<uint8_t> Filler;
   ConstraintKind Constraint = ConstraintKind::NoConstraint;
 };
 
 enum SortKind { SortNone, SortByName, SortByAlignment };
 

lld/trunk/ELF/LinkerScript.cpp

if (auto *Cmd = dyn_cast<OutputSectionCommand>(Base1.get())) {
       if (!Head)
         continue;
 
       OutputSectionBase<ELFT> *OutSec;
       bool IsNew;
       std::tie(OutSec, IsNew) = Factory.create(Head, Cmd->Name);
       if (IsNew)
         OutputSections->push_back(OutSec);
-      for (InputSectionBase<ELFT> *Sec : V)
+
+      uint32_t Subalign = Cmd->SubalignExpr ? Cmd->SubalignExpr(0) : 0;
+      for (InputSectionBase<ELFT> *Sec : V) {
+        if (Subalign)
+          Sec->Alignment = Subalign;
         if (!Sec->OutSec)
           OutSec->addSection(Sec);
       }
     }
+  }
 
   // Add orphan sections.
   for (const std::unique_ptr<ObjectFile<ELFT>> &F :
        Symtab<ELFT>::X->getObjectFiles()) {
     for (InputSectionBase<ELFT> *S : F->getSections()) {
       if (isDiscarded(S) || S->OutSec)
         continue;
       OutputSectionBase<ELFT> *OutSec;
if (peek() != ":")
     Cmd->AddrExpr = readExpr();
 
   expect(":");
 
   if (skip("AT"))
     Cmd->LmaExpr = readParenExpr();
   if (skip("ALIGN"))
     Cmd->AlignExpr = readParenExpr();
+  if (skip("SUBALIGN"))
+    Cmd->SubalignExpr = readParenExpr();
 
   // Parse constraints.
   if (skip("ONLY_IF_RO"))
     Cmd->Constraint = ConstraintKind::ReadOnly;
   if (skip("ONLY_IF_RW"))
     Cmd->Constraint = ConstraintKind::ReadWrite;
   expect("{");
 

lld/trunk/test/ELF/linkerscript/linkerscript-subalign.s

+# REQUIRES: x86
+# RUN: llvm-mc -filetype=obj -triple=x86_64-unknown-linux %s -o %t1.o
+
+# RUN: echo "SECTIONS { .aaa : { *(.aaa.*) } }" > %t1.script
+# RUN: ld.lld -o %t1 --script %t1.script %t1.o
+# RUN: llvm-objdump -s %t1 | FileCheck -check-prefix=NOALIGN %s
+# NOALIGN:      Contents of section .aaa:
+# NOALIGN-NEXT:   01000000 00000000 00000000 00000000
+# NOALIGN-NEXT:   00000000 00000000 00000000 00000000
+# NOALIGN-NEXT:   02000000 00000000 00000000 00000000
+# NOALIGN-NEXT:   00000000 00000000 00000000 00000000
+# NOALIGN-NEXT:   03000000 00000000 00000000 00000000
+# NOALIGN-NEXT:   00000000 00000000 00000000 00000000
+# NOALIGN-NEXT:   00000000 00000000 00000000 00000000
+# NOALIGN-NEXT:   00000000 00000000 00000000 00000000
+# NOALIGN-NEXT:   04000000 00000000
+
+# RUN: echo "SECTIONS { .aaa : SUBALIGN(1) { *(.aaa.*) } }" > %t2.script
+# RUN: ld.lld -o %t2 --script %t2.script %t1.o
+# RUN: llvm-objdump -s %t2 | FileCheck -check-prefix=SUBALIGN %s
+# SUBALIGN: Contents of section .aaa:
+# SUBALIGN:   01000000 00000000 02000000 00000000
+# SUBALIGN:   03000000 00000000 04000000 00000000
+
+.global _start
+_start:
+ nop
+
+.section .aaa.1, "a"
+.align 16
+.quad 1
+
+.section .aaa.2, "a"
+.align 32
+.quad 2
+
+.section .aaa.3, "a"
+.align 64
+.quad 3
+
+.section .aaa.4, "a"
+.align 128
+.quad 4