x86_64: implement high/low split

6 files changed, 96 insertions, 72 deletions

diff --git a/.devcontainer/x86-64/devcontainer.json b/.devcontainer/x86-64/devcontainer.json
index c23e655..3e3b934 100644
--- a/.devcontainer/x86-64/devcontainer.json
+++ b/.devcontainer/x86-64/devcontainer.json
@@ -1,6 +1,6 @@
 {
     "name": "TeachOS on x86-64",
-    "image": "registry.gitlab.ost.ch:45023/teachos/devcontainers/x86-64:15.1.0-2",
+    "image": "registry.gitlab.ost.ch:45023/teachos/devcontainers/x86-64:15.1.0-3",
     "customizations": {
         "vscode": {
             "extensions": [
diff --git a/CMakeLists.txt b/CMakeLists.txt
index 821640c..7c7cc22 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -48,6 +48,7 @@ target_link_libraries("kernel" PRIVATE
 
 target_link_options("kernel" PRIVATE
     "-no-pie"
+    "-nostdlib"
 )
 
 target_disassemble("kernel")
diff --git a/arch/CMakeLists.txt b/arch/CMakeLists.txt
index eded57e..661de26 100644
--- a/arch/CMakeLists.txt
+++ b/arch/CMakeLists.txt
@@ -16,6 +16,10 @@ target_include_directories("arch-any" INTERFACE
 
 target_link_libraries("arch-any" INTERFACE
     "libs::kstd"
+
+    "c"
+    "gcc"
+    "stdc++"
 )
 
 add_subdirectory("${CMAKE_SYSTEM_PROCESSOR}")
diff --git a/arch/x86_64/scripts/kernel.ld b/arch/x86_64/scripts/kernel.ld
index 8af242f..6ba8b80 100644
--- a/arch/x86_64/scripts/kernel.ld
+++ b/arch/x86_64/scripts/kernel.ld
@@ -4,6 +4,7 @@ ENTRY(_start)
  * Virtual and linear start addresses of the TeachOS kernel
  *****************************************************************************/
 TEACHOS_PMA = 1M;
+TEACHOS_VMA = 0xFFFFFFFF80000000;
 
 PHDRS {
   boot_rodata PT_LOAD FLAGS(4);
@@ -25,12 +26,13 @@ SECTIONS
   . = TEACHOS_PMA;
 
   /***************************************************************************
-   * We want to be able to be able to access all memory (linear and virtual)
+   * We want to be able to be able to access all memory (physical and virtual)
    * during bootstrapping and operation. To achieve this, we define some
    * symbols at the beginning.
    ***************************************************************************/
   _start_physical = .;
-  _start_virtual = .;
+  _start_virtual = . + TEACHOS_VMA;
+
 
   /***************************************************************************
    * The bootstrapping infratructure goes first. We first place the read-only
@@ -63,7 +65,10 @@ SECTIONS
    * Now it is time to load the 64-bit kernel code. We
    * make sure to align the loaded data onto a page boundary.
    ***************************************************************************/
-  .init ALIGN(4K) : AT(ADDR (.init))
+  . = ALIGN(4K);
+  . += TEACHOS_VMA;
+
+  .init ALIGN(4K) : AT(ADDR(.init) - TEACHOS_VMA)
   {
     /* 
      * Make sure that the crt code is wrapped around the compiler generated
@@ -74,7 +79,7 @@ SECTIONS
     KEEP(*crtn.s.o*(.init))
   } :text
 
-  .fini ALIGN(4K) : AT(ADDR (.fini))
+  .fini ALIGN(4K) : AT(ADDR (.fini) - TEACHOS_VMA)
   {
     /* 
      * Make sure that the crt code is wrapped around the compiler generated
@@ -85,29 +90,29 @@ SECTIONS
     KEEP(*crtn.s.o*(.fini))
   }
 
-  .stl_text ALIGN(4K) : AT(ADDR (.stl_text))
+  .stl_text ALIGN(4K) : AT(ADDR (.stl_text) - TEACHOS_VMA)
   {
     *(.stl_text .stl_text*)
     KEEP(*libstdc++.a:*(.text .text.*))
   }
 
-  .text ALIGN(4K) : AT(ADDR (.text))
+  .text ALIGN(4K) : AT(ADDR (.text) - TEACHOS_VMA)
   {
     *(.text .text.*)
   }
 
-  .user_text ALIGN(4K) : AT(ADDR (.user_text))
+  .user_text ALIGN(4K) : AT(ADDR (.user_text) - TEACHOS_VMA)
   {
     *(.user_text .user_text.*)
   }
 
-  .rodata ALIGN(4K) : AT (ADDR (.rodata))
+  .rodata ALIGN(4K) : AT (ADDR (.rodata) - TEACHOS_VMA)
   {
     *(.rodata)
     *(.rodata.*)
   } :rodata
 
-  .ctors ALIGN(4K) : AT (ADDR (.ctors))
+  .ctors ALIGN(4K) : AT (ADDR (.ctors) - TEACHOS_VMA)
   {
     KEEP(*crtbegin.o(.ctors))
     KEEP(*(EXCLUDE_FILE (*crtend.o) .ctors))
@@ -115,7 +120,7 @@ SECTIONS
     KEEP(*crtend.o(.ctors))
   } :data
 
-  .dtors ALIGN(4K) : AT (ADDR (.dtors))
+  .dtors ALIGN(4K) : AT (ADDR (.dtors) - TEACHOS_VMA)
   {
     KEEP(*crtbegin.o(.dtors))
     KEEP(*(EXCLUDE_FILE (*crtend.o) .dtors))
@@ -123,18 +128,18 @@ SECTIONS
     KEEP(*crtend.o(.dtors))
   }
 
-  .bss ALIGN(4K) : AT (ADDR (.bss))
+  .bss ALIGN(4K) : AT (ADDR (.bss) - TEACHOS_VMA)
   {
     *(COMMON)
     *(.bss*)
   }
 
-  .data ALIGN(4K) : AT (ADDR (.data))
+  .data ALIGN(4K) : AT (ADDR (.data) - TEACHOS_VMA)
   {
     *(.data*)
   }
 
-  .user_data ALIGN(4K) : AT (ADDR (.user_data))
+  .user_data ALIGN(4K) : AT (ADDR (.user_data) - TEACHOS_VMA)
   {
     *(.user_data .user_data.*)
   }
@@ -143,8 +148,8 @@ SECTIONS
    * In accordance with the symbol definitions at the start, we generate some
    * symbols to mark the end of our loaded image.
    ***************************************************************************/
-  _end_virtual = ADDR(.bss) + SIZEOF(.bss);
-  _end_physical = _end_virtual;
+  _end_virtual = .;
+  _end_physical = _end_virtual - TEACHOS_VMA;
 
   /DISCARD/ : { *(.comment) }
 }
diff --git a/arch/x86_64/src/boot/boot32.S b/arch/x86_64/src/boot/boot32.S
index 7e6c2ae..3039f38 100644
--- a/arch/x86_64/src/boot/boot32.S
+++ b/arch/x86_64/src/boot/boot32.S
@@ -13,9 +13,13 @@ multiboot_information_pointer: .skip 8
 
 .align 4096
 
-page_map_level_4: .skip 512 * 8
-page_map_level_3: .skip 512 * 8
-page_map_level_2: .skip 512 * 8
+page_maps_start:
+page_map_level_4:      .skip 512 * 8
+page_map_level_3_high: .skip 512 * 8
+page_map_level_3_low:  .skip 512 * 8
+page_map_level_2:      .skip 512 * 8
+page_maps_end = .
+page_maps_size = page_maps_end - page_maps_start
 
 /**
  * @brief Storage for the bootstrap stack.
@@ -40,9 +44,9 @@ global_descriptor_table:
 global_descriptor_table_null = . - global_descriptor_table
 .quad 0
 global_descriptor_table_code = . - global_descriptor_table
-.quad GDT_ACCESSED | GDT_READ_WRITE | GDT_EXECUTABLE | GDT_DESCRIPTOR_TYPE | GDT_PRESENT | GDT_LONG_MODE
+.quad GDT_READ_WRITE | GDT_EXECUTABLE | GDT_DESCRIPTOR_TYPE | GDT_PRESENT | GDT_LONG_MODE | (1 << 55)
 global_descriptor_table_data = . - global_descriptor_table
-.quad GDT_ACCESSED | GDT_READ_WRITE | GDT_DESCRIPTOR_TYPE | GDT_PRESENT
+.quad GDT_READ_WRITE | GDT_DESCRIPTOR_TYPE | GDT_PRESENT | (1 << 54) | (1 << 55)
 global_descriptor_table_end:
 
 message_prefix_panic:                    .string  "Panic: "
@@ -125,7 +129,10 @@ _start:
     call _assert_cpuid_instruction_is_supported
     call _assert_cpu_supports_long_mode
 
+    push $HUGE_PAGES_TO_MAP
     call _prepare_page_maps
+    add $4, %esp
+
     call _enable_paging
     call _enable_sse
     call _reload_gdt
@@ -302,51 +309,74 @@ _assert_cpu_supports_long_mode:
 /**
  * @brief Prepare a recursive page map hierarchy
  *
- * We map all physical memory we were loaded in plus one additional page. The
- * mapping is done in terms of huge pages (2 MiB per page) to save on required
- * page map entries.
- *
+ * @param ebp+8 The number of huge pages to map
  * @return void
  */
 _prepare_page_maps:
     pie_function_start
-    push %ebx
 
-    /* Map the P4 table recursively */
-    lea (page_map_level_4 - 0b)(%esi), %eax
-    mov %eax, %ebx
-    or $0b11, %ebx
-    mov %ebx, (511 * 8)(%eax)
-
-    /* Add an entry to the PML4, pointing to the PML3 */
-    lea (page_map_level_3 - 0b)(%esi), %ebx
-    or $0b11, %ebx
-    mov %ebx, (((0x0000000000100000 >> 39) & 0x1ff) * 8)(%eax)
-
-    /* Add an entry to the PML3, pointing to the PML2 */
-    lea (page_map_level_3 - 0b)(%esi), %eax
-    lea (page_map_level_2 - 0b)(%esi), %ebx
-    or $0b11, %ebx
-    mov %ebx, (((0x0000000000100000 >> 30) & 0x1ff) * 8)(%eax)
-
-    /* Add entries for huge pages to the PML2 */
     push %edi
-    lea (page_map_level_2 - 0b)(%esi), %ebx
-    mov $HUGE_PAGES_TO_MAP, %edi
-    xor %ecx, %ecx
+
+    call _clear_page_map_memory
+
+    lea (page_map_level_4 - 0b)(%esi), %edi
+    mov %edi, %eax
+    or $0b11, %eax
+    mov %eax, (510 * 8)(%edi)
+
+    lea (page_map_level_3_low - 0b)(%esi), %eax
+    or $0b11, %eax
+    mov %eax, (%edi)
+
+    lea (page_map_level_3_high - 0b)(%esi), %eax
+    or $0b11, %eax
+    mov %eax, (511 * 8)(%edi)
+
+    lea (page_map_level_3_low - 0b)(%esi), %edi
+    lea (page_map_level_2 - 0b)(%esi), %eax
+    or $0b11, %eax
+    mov %eax, (%edi)
+
+    lea (page_map_level_3_high - 0b)(%esi), %edi
+    lea (page_map_level_2 - 0b)(%esi), %eax
+    or $0b11, %eax
+    mov %eax, (510 * 8)(%edi)
+
+    lea (page_map_level_2 - 0b)(%esi), %edi
+    mov 8(%ebp), %ecx
 
 1:
+    dec %ecx
     mov $(1 << 21), %eax
     mul %ecx
     or $0b10000011, %eax
-    mov %eax, (%ebx, %ecx, 8)
+    mov %eax, (%edi, %ecx, 8)
 
-    inc %ecx
-    cmp %edi, %ecx
-    jne 1b
+    test %ecx, %ecx
+    jnz 1b
 
     pop %edi
-    pop %ebx
+
+    pie_function_end
+
+/**
+ * @brief Clear all page map memory by filling it with 0s.
+ *
+ * @return void
+ */
+_clear_page_map_memory:
+    pie_function_start
+
+    push %edi
+
+    xor %eax, %eax
+    mov $page_maps_size, %ecx
+    shr $2, %ecx
+    lea (page_maps_start - 0b)(%esi), %edi
+    rep stosl
+
+    pop %edi
+
     pie_function_end
 
 /**
diff --git a/cmake/Platforms/x86_64.cmake b/cmake/Platforms/x86_64.cmake
index c84d9ba..8db6142 100644
--- a/cmake/Platforms/x86_64.cmake
+++ b/cmake/Platforms/x86_64.cmake
@@ -1,17 +1,5 @@
 include_guard(GLOBAL)
 
-macro(find_compiler_file NAME OUTPUT_VARIABLE)
-  execute_process(COMMAND
-    "${CMAKE_CXX_COMPILER}"
-    "-mno-red-zone"
-    "-print-file-name=${NAME}"
-    OUTPUT_STRIP_TRAILING_WHITESPACE
-    OUTPUT_VARIABLE "${OUTPUT_VARIABLE}"
-    ERROR_QUIET
-  )
-  mark_as_advanced("${OUTPUT_VARIABLE}")
-endmacro()
-
 set(PLATFORM_TARGET "x86_64-pc-elf")
 
 set(CMAKE_TRY_COMPILE_TARGET_TYPE "STATIC_LIBRARY")
@@ -27,7 +15,8 @@ find_program(CMAKE_CXX_COMPILER "${CMAKE_CXX_COMPILER_TARGET}-g++" REQUIRED)
 
 set(CMAKE_CXX_FLAGS_INIT
 "-mno-red-zone \
--mcmodel=large \
+-mcmodel=kernel \
+-fno-pie \
 -fno-rtti \
 -fno-exceptions \
 -ffunction-sections \
@@ -46,19 +35,14 @@ set(CMAKE_CXX_FLAGS_RELEASE "-O3 -DNDEBUG -ggdb3")
 set(CMAKE_ASM_FLAGS_RELWITHDEBINFO "-O2 -DNDEBUG -ggdb3")
 set(CMAKE_CXX_FLAGS_RELWITHDEBINFO "-O2 -DNDEBUG -ggdb3")
 
-set(CMAKE_EXE_LINKER_FLAGS_INIT "-nostartfiles -Wl,--gc-sections")
-
-find_compiler_file("crtbegin.o" CRT_BEGIN_PATH)
-find_compiler_file("crtend.o" CRT_END_PATH)
+set(CMAKE_EXE_LINKER_FLAGS_INIT "-Wl,--gc-sections")
 
 set(CMAKE_CXX_LINK_EXECUTABLE
 "<CMAKE_CXX_COMPILER> \
 <FLAGS> \
 <CMAKE_CXX_LINK_FLAGS> \
 <LINK_FLAGS> \
-${CRT_BEGIN_PATH} \
 <OBJECTS> \
 -o <TARGET> \
-<LINK_LIBRARIES> \
-${CRT_END_PATH}"
+<LINK_LIBRARIES>"
 )