[llvm] [BOLT] Add --custom-allocation-vma flag (PR #136385)

[via llvm-commits]

llvmbot wrote:


<!--LLVM PR SUMMARY COMMENT-->

@llvm/pr-subscribers-bolt

Author: Rafael Auler (rafaelauler)

<details>
<summary>Changes</summary>

Add an advanced-user flag so we are able to rewrite binaries when we fail
to identify a suitable location to put new code. User then can supply a custom
location via --custom-allocation-vma. This happens more obviously if the binary
has segments mapped to very high addresses.

---
Full diff: https://github.com/llvm/llvm-project/pull/136385.diff


2 Files Affected:

- (modified) bolt/lib/Rewrite/RewriteInstance.cpp (+26) 
- (added) bolt/test/X86/high-segments.s (+46) 


``````````diff
diff --git a/bolt/lib/Rewrite/RewriteInstance.cpp b/bolt/lib/Rewrite/RewriteInstance.cpp
index 37dcfa868c211..518c8150d6728 100644
--- a/bolt/lib/Rewrite/RewriteInstance.cpp
+++ b/bolt/lib/Rewrite/RewriteInstance.cpp
@@ -237,6 +237,13 @@ UseGnuStack("use-gnu-stack",
   cl::ZeroOrMore,
   cl::cat(BoltCategory));
 
+static cl::opt<uint64_t>
+CustomAllocationVMA("custom-allocation-vma",
+           cl::desc("use a custom address at which new code will be put, "
+                    "bypassing BOLT's logic to detect where to put code"),
+  cl::ZeroOrMore,
+  cl::cat(BoltCategory));
+
 static cl::opt<bool>
 SequentialDisassembly("sequential-disassembly",
   cl::desc("performs disassembly sequentially"),
@@ -592,6 +599,25 @@ Error RewriteInstance::discoverStorage() {
 
   FirstNonAllocatableOffset = NextAvailableOffset;
 
+  if (opts::CustomAllocationVMA) {
+    // If user specified a custom address where we should start writing new
+    // data, honor that.
+    NextAvailableAddress = opts::CustomAllocationVMA;
+    // Sanity check the user-supplied address and emit warnings if something
+    // seems off.
+    for (const ELF64LE::Phdr &Phdr : PHs) {
+      switch (Phdr.p_type) {
+        case ELF::PT_LOAD:
+          if (NextAvailableAddress >= Phdr.p_vaddr &&
+              NextAvailableAddress < Phdr.p_vaddr + Phdr.p_memsz) {
+            BC->errs() << "BOLT-WARNING: user-supplied allocation vma 0x"
+                       << Twine::utohexstr(NextAvailableAddress)
+                       << " conflicts with ELF segment at 0x"
+                       << Twine::utohexstr(Phdr.p_vaddr) << "\n";
+          }
+      }
+    }
+  }
   NextAvailableAddress = alignTo(NextAvailableAddress, BC->PageAlign);
   NextAvailableOffset = alignTo(NextAvailableOffset, BC->PageAlign);
 
diff --git a/bolt/test/X86/high-segments.s b/bolt/test/X86/high-segments.s
new file mode 100644
index 0000000000000..dfddf18004c2a
--- /dev/null
+++ b/bolt/test/X86/high-segments.s
@@ -0,0 +1,46 @@
+// Check that we are able to rewrite binaries when we fail to identify a
+// suitable location to put new code and user supplies a custom one via
+// --custom-allocation-vma. This happens more obviously if the binary has
+// segments mapped to very high addresses.
+
+// In this example, my.reserved.section is mapped to a segment to be loaded
+// at address 0x10000000000, while regular text should be at 0x200000. We
+// pick a vma in the middle at 0x700000 to carve space for BOLT to put data,
+// since BOLT's usual route of allocating after the last segment will put
+// code far away and that will blow up relocations from main.
+
+// RUN: split-file %s %t
+// RUN: llvm-mc -filetype=obj -triple x86_64-unknown-unknown %t/main.s -o %t.o
+// RUN: %clang %cflags -no-pie %t.o -o %t.exe -Wl,-T %t/main.ls
+// RUN: llvm-bolt %t.exe -o %t.bolt --custom-allocation-vma=0x700000
+
+//--- main.s
+  .type            reserved_space, at object
+  .section        .my.reserved.section,"awx", at nobits
+  .globl           reserved_space
+  .p2align         4, 0x0
+reserved_space:
+  .zero  0x80000000
+  .size   reserved_space, 0x80000000
+
+	.text
+  .globl main
+  .globl _start
+  .type main, %function
+_start:
+main:
+	.cfi_startproc
+  nop
+  nop
+  nop
+  retq
+	.cfi_endproc
+.size main, .-main
+
+//--- main.ls
+SECTIONS
+{
+    .my.reserved.section 1<<40 : {
+      *(.my.reserved.section);
+    }
+} INSERT BEFORE .comment;

``````````

</details>


https://github.com/llvm/llvm-project/pull/136385