[BOLT] Fix stack alignment for runtime lib

Summary: Right now, the SAVE_ALL sequence executed upon entry of both of our runtime libs (hugify and instrumentation) will cause the stack to not be aligned at a 16B boundary because it saves 15 8-byte regs. Change the code sequence to adjust for that. The compiler may generate code that assumes the stack is aligned by using movaps instructions, which will crash. (cherry picked from FBD22744307)
llvm · Jul 27, 2020 · c6799a6 · c6799a6
1 parent ed02946
commit c6799a6
Show file tree

Hide file tree

Showing 4 changed files with 15 additions and 9 deletions.
diff --git a/bolt/runtime/common.h b/bolt/runtime/common.h
@@ -6,6 +6,7 @@
 #include <elf.h>
 #endif
 
+// Save all registers while keeping 16B stack alignment
 #define SAVE_ALL                                                               \
   "push %%rax\n"                                                               \
   "push %%rbx\n"                                                               \
@@ -21,9 +22,12 @@
   "push %%r12\n"                                                               \
   "push %%r13\n"                                                               \
   "push %%r14\n"                                                               \
-  "push %%r15\n"
+  "push %%r15\n"                                                               \
+  "sub $8, %%rsp\n"
 
+// Mirrors SAVE_ALL
 #define RESTORE_ALL                                                            \
+  "add $8, %%rsp\n"                                                            \
   "pop %%r15\n"                                                                \
   "pop %%r14\n"                                                                \
   "pop %%r13\n"                                                                \

diff --git a/bolt/runtime/hugify.cpp b/bolt/runtime/hugify.cpp
@@ -166,7 +166,9 @@ extern "C" void __bolt_hugify_self_impl() {
 
 /// This is hooking ELF's entry, it needs to save all machine state.
 extern "C" __attribute((naked)) void __bolt_hugify_self() {
-  __asm__ __volatile__(SAVE_ALL "call __bolt_hugify_self_impl\n" RESTORE_ALL
-                                "jmp *__bolt_hugify_init_ptr(%%rip)\n" ::
-                                    :);
+  __asm__ __volatile__(SAVE_ALL
+                       "call __bolt_hugify_self_impl\n"
+                       RESTORE_ALL
+                       "jmp *__bolt_hugify_init_ptr(%%rip)\n"
+                       :::);
 }
diff --git a/bolt/runtime/instr.cpp b/bolt/runtime/instr.cpp
@@ -1419,8 +1419,8 @@ extern "C" void instrumentIndirectCall(uint64_t Target, uint64_t IndCallID) {
 extern "C" __attribute((naked)) void __bolt_instr_indirect_call()
 {
   __asm__ __volatile__(SAVE_ALL
-                       "mov 0x88(%%rsp), %%rdi\n"
-                       "mov 0x80(%%rsp), %%rsi\n"
+                       "mov 0x90(%%rsp), %%rdi\n"
+                       "mov 0x88(%%rsp), %%rsi\n"
                        "call instrumentIndirectCall\n"
                        RESTORE_ALL
                        "pop %%rdi\n"
@@ -1433,8 +1433,8 @@ extern "C" __attribute((naked)) void __bolt_instr_indirect_call()
 extern "C" __attribute((naked)) void __bolt_instr_indirect_tailcall()
 {
   __asm__ __volatile__(SAVE_ALL
-                       "mov 0x80(%%rsp), %%rdi\n"
-                       "mov 0x78(%%rsp), %%rsi\n"
+                       "mov 0x88(%%rsp), %%rdi\n"
+                       "mov 0x80(%%rsp), %%rsi\n"
                        "call instrumentIndirectCall\n"
                        RESTORE_ALL
                        "add $16, %%rsp\n"

diff --git a/bolt/test/X86/user-func-reorder.c b/bolt/test/X86/user-func-reorder.c
@@ -30,7 +30,7 @@ REQUIRES: system-linux
 RUN: %host_cc %s -o %t.exe -Wl,-q
 
 RUN: llvm-bolt %t.exe -relocs=1 -lite -reorder-functions=user  \
-RUN:   -function-order=%p/Inputs/user_func_order.txt -o %t
+RUN:   -hugify -function-order=%p/Inputs/user_func_order.txt -o %t
 RUN: nm -ns %t | FileCheck %s -check-prefix=CHECK-NM
 RUN: %t 1 2 3 | FileCheck %s -check-prefix=CHECK-OUTPUT