From 7322df98f59c052dcca20caa63573e77f0a416d7 Mon Sep 17 00:00:00 2001
From: "Justin C. Miller" <justin@devjustinian.com>
Date: Mon, 29 Apr 2024 01:07:18 -0700
Subject: [PATCH] [ld.so] Call all image global ctors, not just libc

With the move to dynamic executables, crt0's _start was only ever
calling libc's __init_libc, which only ran libc's init_array list. Now
make crt0 itself (which is statically linked into every executable) call
it's own init_array list and have ld.so call every other image's ctor
lists.
---
 src/libraries/libc/arch/amd64/crt/crt0.s | 48 ++++++++++++++++++++++++
 src/libraries/libc/arch/amd64/init.cpp   | 27 -------------
 src/user/ld.so/image.cpp                 | 29 ++++++++++----
 src/user/ld.so/image.h                   |  1 +
 4 files changed, 71 insertions(+), 34 deletions(-)

diff --git a/src/libraries/libc/arch/amd64/crt/crt0.s b/src/libraries/libc/arch/amd64/crt/crt0.s
index 299e8cf..05e5e29 100644
--- a/src/libraries/libc/arch/amd64/crt/crt0.s
+++ b/src/libraries/libc/arch/amd64/crt/crt0.s
@@ -5,6 +5,52 @@ extern exit
 extern __init_libj6
 extern __init_libc
 
+extern __preinit_array_start
+extern __preinit_array_end
+extern __init_array_start
+extern __init_array_end
+
+global __run_ctor_list:function hidden (__run_ctor_list.end - __run_ctor_list)
+__run_ctor_list:
+    push rbp
+    mov rbp, rsp
+
+    push rbx
+    push r12
+    mov rbx, rdi
+    mov r12, rsi
+.start:
+    cmp rbx, r12
+    je .fin
+    mov rax, [rbx]
+    call rax
+    add rbx, 8
+    jmp .start
+.fin:
+
+    pop r12
+    pop rbx
+    pop rbp
+    ret
+.end:
+
+global __run_global_ctors:function hidden (__run_global_ctors.end - __run_global_ctors)
+__run_global_ctors:
+    push rbp
+    mov rbp, rsp
+
+    lea rdi, [rel __preinit_array_start]
+    lea rsi, [rel __preinit_array_end]
+    call __run_ctor_list
+    lea rdi, [rel __init_array_start]
+    lea rsi, [rel __init_array_end]
+    call __run_ctor_list
+
+    pop rbp
+    ret
+.end:
+
+
 ; Put the address of the given symbol in rax
 ; This macro is the same as in util/got.inc,
 ; but crt0 can't have a dep on libutil
@@ -31,6 +77,8 @@ _libc_crt0_start:
     lookup_GOT __init_libc
     call rax
 
+    call __run_global_ctors
+
     mov rdi, 0
     mov rsi, rsp
     mov rdx, 0    ; TODO: actually parse stack for argc, argv, envp
diff --git a/src/libraries/libc/arch/amd64/init.cpp b/src/libraries/libc/arch/amd64/init.cpp
index 5f57c1a..973a4cc 100644
--- a/src/libraries/libc/arch/amd64/init.cpp
+++ b/src/libraries/libc/arch/amd64/init.cpp
@@ -1,33 +1,6 @@
 #include <stddef.h>
 
-using cb = void (*)(void);
-extern cb __preinit_array_start;
-extern cb __preinit_array_end;
-extern cb __init_array_start;
-extern cb __init_array_end;
-
-namespace {
-
-void
-run_ctor_list(cb *p, cb *end)
-{
-    while (p && end && p < end) {
-        if (p) (*p)();
-        ++p;
-    }
-}
-
-void
-run_global_ctors()
-{
-    run_ctor_list(&__preinit_array_start, &__preinit_array_end);
-    run_ctor_list(&__init_array_start, &__init_array_end);
-}
-
-} // namespace
-
 extern "C" void
 __init_libc()
 {
-    run_global_ctors();
 }
diff --git a/src/user/ld.so/image.cpp b/src/user/ld.so/image.cpp
index c60e60e..f52d358 100644
--- a/src/user/ld.so/image.cpp
+++ b/src/user/ld.so/image.cpp
@@ -17,6 +17,8 @@
 extern "C" void _ldso_plt_lookup();
 extern image_list all_images;
 
+// From crt0, which ld.so also links
+extern void __run_ctor_list(uintptr_t start, uintptr_t end);
 
 // Can't use strcmp because it's from another library, and
 // this needs to be used as part of relocation or symbol lookup
@@ -327,15 +329,28 @@ image::parse_rela_table(const util::counted<const rela> &table, image_list &ctx)
 void
 image::relocate(image_list &ctx)
 {
-    if (relocated)
-        return;
+    if (!relocated) {
+        parse_rela_table(dynrel, ctx);
+        parse_rela_table(jmprel, ctx);
 
-    parse_rela_table(dynrel, ctx);
-    parse_rela_table(jmprel, ctx);
+        got[1] = reinterpret_cast<uintptr_t>(this);
+        got[2] = reinterpret_cast<uintptr_t>(&_ldso_plt_lookup);
+        relocated = true;
+    }
 
-    got[1] = reinterpret_cast<uintptr_t>(this);
-    got[2] = reinterpret_cast<uintptr_t>(&_ldso_plt_lookup);
-    relocated = true;
+    if (!ctors) {
+        uintptr_t pre_init_start = lookup("__preinit_array_start");
+        uintptr_t pre_init_end = lookup("__preinit_array_end");
+        if (pre_init_start && pre_init_end)
+            __run_ctor_list(pre_init_start, pre_init_end);
+
+        uintptr_t init_start = lookup("__init_array_start");
+        uintptr_t init_end = lookup("__init_array_end");
+        if (init_start && init_end)
+            __run_ctor_list(init_start, init_end);
+
+        ctors = true;
+    }
 }
 
 image_list::item_type *
diff --git a/src/user/ld.so/image.h b/src/user/ld.so/image.h
index e1d04fb..a98ebc2 100644
--- a/src/user/ld.so/image.h
+++ b/src/user/ld.so/image.h
@@ -28,6 +28,7 @@ struct image
     gnu_hash_table const *gnu_hash = nullptr;
 
     bool relocated = false;
+    bool ctors = false;
 
     /// Look up a string table entry in this image's string table.
     const char * string(unsigned index) const {