start on paging system; make our own page tables and switch to them

compile_flags.txt

@@ -1,2 +1,5 @@
+-std=c23
 -I
 dependencies/limine
+-I
+kernel/include

kernel/include/paging.h

@@ -0,0 +1,39 @@
+/* Calcite, kernel/include/paging.h
+ * Copyright 2025 Benji Dial
+ *
+ * This program is free software: you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation, either version 3 of the License, or (at
+ * your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful, but
+ * WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
+ * or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
+ * more details.
+ *
+ * You should have received a copy of the GNU General Public License along with
+ * this program. If not, see <https://www.gnu.org/licenses/>.
+ */
+
+#pragma once
+
+#include <stdint.h>
+
+//kernel physical and virtual bases are passed so that we can compute the
+//physical addresses of the kernel's statically allocated paging structures.
+void init_paging(uint64_t kernel_physical_base, void *kernel_virtual_base);
+
+//base and length should be page-aligned
+void mark_physical_memory_free(uint64_t base, uint64_t length);
+
+//maps one page. physical and virtual bases should be page-aligned.
+//virtual address should be within kernel range.
+void map_in_kernel_page_table(
+  uint64_t physical_base, void *virtual_base, int writable, int executable);
+
+//returns a region of contiguous pages in kernel virtual memory where nothing
+//is mapped. length should be page-aligned.
+void *find_free_kernel_region(uint64_t length);
+
+//implemented in paging.asm. the continuation should be noreturn.
+[[noreturn]] void switch_to_kernel_page_tables(void (*continuation)());

kernel/src/entry.c

@@ -16,6 +16,7 @@
  */
 
 #include <limine.h>
+#include <paging.h>
 
 LIMINE_BASE_REVISION(3)
 
@@ -25,24 +26,126 @@ static volatile struct limine_framebuffer_request fb_request = {
   .response = 0
 };
 
+static volatile struct limine_hhdm_request hhdm_request = {
+  .id = LIMINE_HHDM_REQUEST,
+  .revision = 0,
+  .response = 0
+};
+
+static volatile struct limine_kernel_address_request ka_request = {
+  .id = LIMINE_KERNEL_ADDRESS_REQUEST,
+  .revision = 0,
+  .response = 0
+};
+
+static volatile struct limine_memmap_request memmap_request = {
+  .id = LIMINE_MEMMAP_REQUEST,
+  .revision = 0,
+  .response = 0
+};
+
 [[noreturn]] static void die() {
   while (1)
     __asm__ ("hlt");
 }
 
+//these are defined in the linker script. they are page-aligned.
+extern uint8_t __kernel_rx_start;
+extern uint8_t __kernel_rx_end;
+extern uint8_t __kernel_ro_start;
+extern uint8_t __kernel_ro_end;
+extern uint8_t __kernel_rw_start;
+extern uint8_t __kernel_rw_end;
+
+static void map_kernel_region(
+  uint64_t physical_start, void *virtual_start,
+  uint64_t length, int writable, int executable) {
+  for (uint64_t i = 0; i < length; i += 4096)
+    map_in_kernel_page_table(
+      physical_start + i, (uint8_t *)virtual_start + i, writable, executable);
+}
+
+static uint8_t *fb_base;
+static int fb_width;
+static int fb_height;
+static int fb_pitch;
+
+[[noreturn]] static void with_kernel_page_tables();
+
 [[noreturn]] void kernel_entry() {
 
-  if (fb_request.response == 0 || fb_request.response->framebuffer_count == 0)
+  //die if the bootloader hasn't given us something that we need.
+
+  if (fb_request.response == 0 || hhdm_request.response == 0 ||
+      ka_request.response == 0 || memmap_request.response == 0 ||
+      fb_request.response->framebuffer_count == 0)
     die();
 
   struct limine_framebuffer *fb = fb_request.response->framebuffers[0];
+  if (fb->memory_model != LIMINE_FRAMEBUFFER_RGB || fb->bpp != 32 ||
+      fb->red_mask_shift != 16 || fb->red_mask_size != 8 ||
+      fb->green_mask_shift != 8 || fb->green_mask_size != 8 ||
+      fb->blue_mask_shift != 0 || fb->blue_mask_size != 8)
+    die();
 
-  for (uint64_t y = 0; y < fb->height; ++y)
-    for (uint64_t x = 0; x < fb->width; ++x) {
-      uint8_t *pixel = (uint8_t *)fb->address + y * fb->pitch + x * 4;
-      pixel[0] = y * 256 / fb->height;
-      pixel[1] = x * 256 / fb->width;
-      pixel[2] = 0;
+  //set up page tables. we will mark the regions with bootloader structures as
+  //usable, so we need to be careful not to allocate any physical pages until
+  //after we are done using bootloader structures. we map the kernel into our
+  //page tables, and we remap the framebuffer to somewhere in kernel memory.
+  //we do not map the bootloader structures, so we also need to not switch
+  //to the new tables until we are done using them.
+
+  uint64_t kernel_physical_base = ka_request.response->physical_base;
+  uint8_t *kernel_virtual_base = (uint8_t *)ka_request.response->virtual_base;
+  init_paging(kernel_physical_base, kernel_virtual_base);
+
+  struct limine_memmap_response *mm_response = memmap_request.response;
+  for (uint64_t i = 0; i < mm_response->entry_count; ++i) {
+    struct limine_memmap_entry *entry = mm_response->entries[i];
+    if (entry->type == LIMINE_MEMMAP_USABLE ||
+        entry->type == LIMINE_MEMMAP_BOOTLOADER_RECLAIMABLE)
+      //limine guarantees these two types are page-aligned already.
+      mark_physical_memory_free(entry->base, entry->length);
+  }
+
+  map_kernel_region(
+    &__kernel_rx_start - kernel_virtual_base + kernel_physical_base,
+    &__kernel_rx_start, &__kernel_rx_end - &__kernel_rx_start, 0, 1);
+  map_kernel_region(
+    &__kernel_ro_start - kernel_virtual_base + kernel_physical_base,
+    &__kernel_ro_start, &__kernel_ro_end - &__kernel_ro_start, 0, 0);
+  map_kernel_region(
+    &__kernel_rw_start - kernel_virtual_base + kernel_physical_base,
+    &__kernel_rw_start, &__kernel_rw_end - &__kernel_rw_start, 1, 0);
+
+  //we round up to a multiple of a page.
+  uint64_t fb_length = ((fb->height * fb->pitch - 1) / 4096 + 1) * 4096;
+  fb_base = find_free_kernel_region(fb_length);
+  map_kernel_region(
+    (uint64_t)fb->address - hhdm_request.response->offset,
+    fb_base, fb_length, 1, 0);
+
+  //store rest of framebuffer information
+
+  fb_width = fb->width;
+  fb_height = fb->height;
+  fb_pitch = fb->pitch;
+
+  //switch to kernel page tables!
+
+  switch_to_kernel_page_tables(&with_kernel_page_tables);
+
+}
+
+[[noreturn]] static void with_kernel_page_tables() {
+
+  //display our test pattern
+
+  for (int y = 0; y < fb_height; ++y)
+    for (int x = 0; x < fb_width; ++x) {
+      fb_base[y * fb_pitch + x * 4] = y * 256 / fb_height;
+      fb_base[y * fb_pitch + x * 4 + 1] = x * 256 / fb_width;
+      fb_base[y * fb_pitch + x * 4 + 2] = 0;
     }
 
   die();

kernel/src/paging.asm

@@ -0,0 +1,42 @@
+ ; Calcite, kernel/src/paging.asm
+ ; Copyright 2025 Benji Dial
+ ;
+ ; This program is free software: you can redistribute it and/or modify
+ ; it under the terms of the GNU General Public License as published by
+ ; the Free Software Foundation, either version 3 of the License, or (at
+ ; your option) any later version.
+ ;
+ ; This program is distributed in the hope that it will be useful, but
+ ; WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
+ ; or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
+ ; more details.
+ ;
+ ; You should have received a copy of the GNU General Public License along with
+ ; this program. If not, see <https://www.gnu.org/licenses/>.
+
+
+bits 64
+
+;both defined in paging.c
+extern kernel_p4_physical_address
+extern init_stack
+
+init_stack_length equ 16384
+
+section .text
+
+extern switch_to_kernel_page_tables
+switch_to_kernel_page_tables:
+
+  ;switch the page table
+  mov rax, qword [kernel_p4_physical_address]
+  mov cr3, rax
+
+  ;set the stack
+  mov rsp, init_stack + init_stack_length
+
+  ;push 0 before jumping to continuation so that if the continuation
+  ;does return (which it shouldn't), it returns to null instead of
+  ;popping garbage and returning there
+  push qword 0
+  jmp rdi

kernel/src/paging.c

@@ -0,0 +1,147 @@
+/* Calcite, kernel/src/paging.c
+ * Copyright 2025 Benji Dial
+ *
+ * This program is free software: you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation, either version 3 of the License, or (at
+ * your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful, but
+ * WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY
+ * or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
+ * more details.
+ *
+ * You should have received a copy of the GNU General Public License along with
+ * this program. If not, see <https://www.gnu.org/licenses/>.
+ */
+
+#include <paging.h>
+
+#define MAX_PHYSICAL_GB 64ULL
+
+//this is a simple bitmap. physical_map[i / 8] & (1 << (i % 8)) is nonzero if
+//and only if the physical page with base i * 4096 is free.
+static uint8_t physical_map[MAX_PHYSICAL_GB << 15];
+
+//kernel virtual memory is the top gigabyte, i.e. one "page directory" worth
+//of pages. in the future i might make it more reactive, and allow new page
+//directories to be allocated as needed, but for now there is just the one
+//page directory, and all of the page tables in that directory are statically
+//allocated. i do not like the intel/amd names for the paging structures, so
+//i will use these names throughout the rest of this file:
+//  p4 = "page map level 4"
+//  p3 = "page directory pointer table"
+//  p2 = "page directory"
+//  p1 = "page table"
+
+//referenced from paging.asm
+uint64_t kernel_p4_physical_address;
+
+alignas(4096) static uint64_t kernel_p4[512];
+alignas(4096) static uint64_t kernel_p3[512];
+alignas(4096) static uint64_t kernel_p2[512];
+alignas(4096) static uint64_t kernel_p1s[512 * 512];
+
+//referenced from paging.asm. at the end of init, this will either be marked
+//free or reused as an interrupt / syscall stack. if the length of this is
+//changed, it also needs to be changed in paging.asm. there is no guard page,
+//but i don't expect init to need much stack.
+alignas(4096) uint64_t init_stack[16384];
+
+void init_paging(uint64_t kernel_physical_base, void *kernel_virtual_base) {
+
+  uint64_t kernel_load_offset =
+    (uint64_t)kernel_virtual_base - kernel_physical_base;
+
+  kernel_p4_physical_address = (uint64_t)kernel_p4 - kernel_load_offset;
+
+  kernel_p4[511] = ((uint64_t)kernel_p3 - kernel_load_offset) | 0x3;
+  for (int i = 0; i < 511; ++i)
+    kernel_p4[i] = 0;
+
+  kernel_p3[511] = ((uint64_t)kernel_p2 - kernel_load_offset) | 0x3;
+  for (int i = 0; i < 511; ++i)
+    kernel_p3[i] = 0;
+
+  for (int i = 0; i < 512; ++i)
+    kernel_p2[i] =
+      ((uint64_t)kernel_p1s - kernel_load_offset + i * 4096) | 0x03;
+
+  for (int i = 0; i < 512 * 512; ++i)
+    kernel_p1s[i] = 0;
+
+}
+
+void mark_physical_memory_free(uint64_t base, uint64_t length) {
+
+  if (base + length > MAX_PHYSICAL_GB << 30)
+    length = (MAX_PHYSICAL_GB << 30) - base;
+
+  while (length && (base & (0x7 << 12))) {
+    physical_map[base >> 15] |= 1 << ((base >> 12) % 8);
+    base += 4096;
+    length -= 4096;
+  }
+
+  while (length >= 4096 * 8) {
+    physical_map[base >> 15] = 0xff;
+    base += 4096 * 8;
+    length -= 4096 * 8;
+  }
+
+  while (length) {
+    physical_map[base >> 15] |= 1 << ((base >> 12) % 8);
+    base += 4096;
+    length -= 4096;
+  }
+
+}
+
+void map_in_kernel_page_table(
+    uint64_t physical_base, void *virtual_base,
+    int writable, int executable) {
+
+  uint64_t virtual_base_u64 = (uint64_t)virtual_base;
+
+  //should probably die in this case
+  if (virtual_base_u64 < 0xffffffffc0000000)
+    return;
+
+  uint64_t p1s_index = (virtual_base_u64 - 0xffffffffc0000000) >> 12;
+
+  //should probably die in this case too
+  if (kernel_p1s[p1s_index] != 0)
+    return;
+
+  kernel_p1s[p1s_index] =
+    physical_base | (writable ? 0x3 : 0x1) |
+    (executable ? 0 : 0x8000000000000000);
+
+}
+
+void *find_free_kernel_region(uint64_t length) {
+
+  length /= 4096;
+
+  uint64_t next = 0;
+
+  uint64_t run_start = 0;
+  uint64_t run_length = 0;
+
+  while (1) {
+    if (run_length == length)
+      return (void *)(run_start * 4096 + 0xffffffffc0000000);
+    if (next == 512 * 512)
+      //die. TODO: handle this nicer.
+      while (1)
+        __asm__ ("hlt");
+    if (kernel_p1s[next] == 0)
+      ++run_length;
+    else {
+      run_length = 0;
+      run_start = next + 1;
+    }
+    ++next;
+  }
+
+}

makefile

@@ -14,11 +14,15 @@ clean:
 
 # kernel
 
-KERNEL_SOURCES = entry.c
+KERNEL_SOURCES = entry.c paging.asm paging.c
+
+build/kernel/%.asm.o: kernel/src/%.asm
+	@mkdir -p ${@D}
+	nasm -f elf64 $^ -o $@
 
 build/kernel/%.c.o: kernel/src/%.c
 	@mkdir -p ${@D}
-	cc -c ${CC_FLAGS} -I dependencies/limine $^ -o $@
+	cc -c ${CC_FLAGS} -I dependencies/limine -I kernel/include $^ -o $@
 
 build/kernel/kernel.elf: ${KERNEL_SOURCES:%=build/kernel/%.o}
 	ld -T kernel/link.ld $^ -o $@

readme.txt

@@ -17,9 +17,10 @@ Calcite requires some dependencies before it can be built:
  * GNU Binutils (Specifically, "ld" is used to link the kernel.)
  * GNU Make (I don't think I have used any GNU extensions.)
  * GNU xorriso
+ * NASM
 
 On Debian, it is sufficient to run this command:
-  apt install binutils curl gcc make xorriso
+  apt install binutils curl gcc make nasm xorriso
 
 To build Calcite, first run "sh get-dependencies.sh", then run "make".
 This will build a disk image at "build/disk.iso" that can be booted