From f7ea46e49e4851ee906ef9aad667f38f2d6a3e4f Mon Sep 17 00:00:00 2001
From: "Justin C. Miller" <justin@devjustinian.com>
Date: Sun, 25 Feb 2024 17:09:04 -0800
Subject: [PATCH] [build] Get release mode optimizations working

Added a release config, and fixed a few spots where optimizations broke things:

- Clang was generating incorrect code for run_ctor_list in libc's init.cpp (it
  ignored a check for the end of the list)
- my rep movsb memcpy implementation used incorrect inline asm constraints, so
  it was returning a pointer to the end of the copied range instead of the start.
  Since this function was just inline asm anyway, I rewrote it in asm by hand in
  a new memutils.s file.
---
 assets/build/config.release.yaml       |  3 +++
 src/kernel/interrupts.cpp              | 19 +++++++++++--------
 src/libraries/j6/j6.module             |  1 +
 src/libraries/j6/memutils.cpp          |  8 --------
 src/libraries/j6/memutils.s            |  7 +++++++
 src/libraries/libc/arch/amd64/init.cpp | 17 ++++++-----------
 6 files changed, 28 insertions(+), 27 deletions(-)
 create mode 100644 assets/build/config.release.yaml
 create mode 100644 src/libraries/j6/memutils.s

diff --git a/assets/build/config.release.yaml b/assets/build/config.release.yaml
new file mode 100644
index 0000000..7574ae4
--- /dev/null
+++ b/assets/build/config.release.yaml
@@ -0,0 +1,3 @@
+ccflags: [
+  "-O3",
+]
\ No newline at end of file
diff --git a/src/kernel/interrupts.cpp b/src/kernel/interrupts.cpp
index 89705a6..fd95a7f 100644
--- a/src/kernel/interrupts.cpp
+++ b/src/kernel/interrupts.cpp
@@ -123,16 +123,19 @@ isr_handler(cpu_state *regs)
             uintptr_t cr2 = 0;
             __asm__ __volatile__ ("mov %%cr2, %0" : "=r"(cr2));
 
-            bool user = cr2 < mem::kernel_offset;
-            vm_space::fault_type ft =
-                static_cast<vm_space::fault_type>(regs->errorcode);
+            // The zero page is always invalid
+            if (cr2 > mem::frame_size) {
+                bool user = cr2 < mem::kernel_offset;
+                vm_space::fault_type ft =
+                    static_cast<vm_space::fault_type>(regs->errorcode);
 
-            vm_space &space = user
-                ? obj::process::current().space()
-                : vm_space::kernel_space();
+                vm_space &space = user
+                    ? obj::process::current().space()
+                    : vm_space::kernel_space();
 
-            if (cr2 && space.handle_fault(cr2, ft))
-                break;
+                if (cr2 && space.handle_fault(cr2, ft))
+                    break;
+            }
 
             util::format({message, sizeof(message)},
                 "Page fault: %016lx%s%s%s%s%s", cr2,
diff --git a/src/libraries/j6/j6.module b/src/libraries/j6/j6.module
index 8c4524a..05bac16 100644
--- a/src/libraries/j6/j6.module
+++ b/src/libraries/j6/j6.module
@@ -8,6 +8,7 @@ j6 = module("j6",
         "condition.cpp",
         "init.cpp",
         "memutils.cpp",
+        "memutils.s",
         "mutex.cpp",
         "protocol_ids.cpp",
         "protocols/service_locator.cpp",
diff --git a/src/libraries/j6/memutils.cpp b/src/libraries/j6/memutils.cpp
index ff4fd1f..8e503c5 100644
--- a/src/libraries/j6/memutils.cpp
+++ b/src/libraries/j6/memutils.cpp
@@ -7,14 +7,6 @@
 using namespace j6;
 using namespace __j6libc;
 
-void *memcpy(void * restrict dst, const void * restrict src, size_t n) {
-    asm volatile ("rep movsb"
-        :
-        : "D"(dst), "S"(src), "c"(n)
-        : "memory");
-    return dst;
-}
-
 static void memmove_dispatch(char *s1, const char *s2, size_t n) {
     if (s1 == s2) return;
 
diff --git a/src/libraries/j6/memutils.s b/src/libraries/j6/memutils.s
new file mode 100644
index 0000000..6e21817
--- /dev/null
+++ b/src/libraries/j6/memutils.s
@@ -0,0 +1,7 @@
+global memcpy: function (memcpy.end - memcpy)
+memcpy:
+    mov rax, rdi
+    mov rcx, rdx
+    rep movsb
+    ret
+memcpy.end:
\ No newline at end of file
diff --git a/src/libraries/libc/arch/amd64/init.cpp b/src/libraries/libc/arch/amd64/init.cpp
index 4103b76..5f57c1a 100644
--- a/src/libraries/libc/arch/amd64/init.cpp
+++ b/src/libraries/libc/arch/amd64/init.cpp
@@ -9,16 +9,11 @@ extern cb __init_array_end;
 namespace {
 
 void
-run_ctor_list(cb *array, cb *end)
+run_ctor_list(cb *p, cb *end)
 {
-    if (!array || !end)
-        return;
-
-    size_t i = 0;
-    while (true) {
-        const cb &ctor = array[i++];
-        if (&ctor == end) return;
-        if (ctor) ctor();
+    while (p && end && p < end) {
+        if (p) (*p)();
+        ++p;
     }
 }
 
@@ -31,8 +26,8 @@ run_global_ctors()
 
 } // namespace
 
-extern "C"
-void __init_libc()
+extern "C" void
+__init_libc()
 {
     run_global_ctors();
 }