integration-test: boot multiboot2_chainloader itself also via Multiboot2

Since I don't use QEMU Multiboot1-direct boot anymore, we can use Multiboot2
here as well.

This also fights some problems with Limine which doesn't want to load the
Multiboot1 kernel in an UEFI environment.

Author: phip1611

integration-test/bins/Cargo.lock

@@ -12,7 +12,6 @@ publish = false
 anyhow.workspace = true
 log.workspace = true
 good_memory_allocator.workspace = true
-multiboot = "0.8"
 multiboot2.workspace = true
 multiboot2-header.workspace = true
 util.workspace = true

integration-test/bins/multiboot2_chainloader/Cargo.toml

@@ -11,9 +11,9 @@ PHDRS
 
 SECTIONS {
     /* Chainloader linked at 8M, payload at 16M */
-    .text 8M : AT(8M) ALIGN(4K)
+    .text 12M : AT(12M) ALIGN(4K)
     {
-        KEEP(*(.multiboot_header));
+        KEEP(*(.multiboot2_header));
         *(.text .text.*)
     } : kernel_rx
 

integration-test/bins/multiboot2_chainloader/link.ld

@@ -1,19 +1,58 @@
 use alloc::boxed::Box;
+use alloc::vec::Vec;
 use elf_rs::{ElfFile, ProgramHeaderEntry, ProgramType};
+use log::{debug, info};
 use multiboot2::{
-    BootLoaderNameTag, CommandLineTag, MaybeDynSized, MemoryArea, MemoryAreaType, MemoryMapTag,
-    ModuleTag, SmbiosTag,
+    BootLoaderNameTag, CommandLineTag, EFIMemoryAreaType, MaybeDynSized, MemoryArea,
+    MemoryAreaType, MemoryMapTag, ModuleTag, SmbiosTag,
 };
 
+fn get_free_mmap_areas(
+    mbi: &multiboot2::BootInformation,
+) -> Vec<(u64 /* start */, u64 /* size */)> {
+    match (mbi.memory_map_tag(), mbi.efi_memory_map_tag()) {
+        (Some(mmt), None) => mmt
+            .memory_areas()
+            .iter()
+            .filter(|ma| ma.typ() == MemoryAreaType::Available)
+            .map(|ma| (ma.start_address(), ma.size()))
+            .collect::<alloc::vec::Vec<_>>(),
+        (_, Some(mmt)) => mmt
+            .memory_areas()
+            .filter(|ma| ma.ty == EFIMemoryAreaType::CONVENTIONAL)
+            .map(|ma| (ma.phys_start, ma.page_count * 4096))
+            .collect::<alloc::vec::Vec<_>>(),
+        _ => panic!("No usable memory map"),
+    }
+}
+
+fn assert_load_segment_fits_into_memory(
+    start: u64,
+    size: u64,
+    free_areas: &[(u64 /* start */, u64 /* size */)],
+) {
+    let end = start + size;
+    let range = free_areas
+        .iter()
+        .find(|(a_start, a_size)| start >= *a_start && end <= a_start + a_size);
+    if let Some(range) = range {
+        debug!("Can load load segment (0x{start:x?}, {size:x?}) into free memory area {range:#x?}");
+    } else {
+        panic!("Can't load load segment  (0x{start:x?}, {size:x?}) into any area!");
+    }
+}
+
 /// Loads the first module into memory. Assumes that the module is a ELF file.
 /// The handoff is performed according to the Multiboot2 spec.
-pub fn load_module(mut modules: multiboot::information::ModuleIter) -> ! {
+pub fn load_module(mbi: &multiboot2::BootInformation) -> ! {
+    let mut modules = mbi.module_tags();
+
     // Load the ELF from the Multiboot1 boot module.
     let elf_mod = modules.next().expect("Should have payload");
     let elf_bytes = unsafe {
         core::slice::from_raw_parts(
-            elf_mod.start as *const u64 as *const u8,
-            (elf_mod.end - elf_mod.start) as usize,
+            elf_mod.start_address() as *const u64 as *const u8,
+            elf_mod.module_size() as usize,
         )
     };
     let elf = elf_rs::Elf32::from_bytes(elf_bytes).expect("Should be valid ELF");
@@ -28,10 +67,11 @@ pub fn load_module(mut modules: multiboot::information::ModuleIter) -> ! {
         log::info!("Multiboot2 header:\n{hdr:#?}");
     }
 
-    // Map the load segments into memory (at their corresponding link).
+    // Load the load segments into memory (at their corresponding link address).
     {
-        let elf = elf_rs::Elf32::from_bytes(elf_bytes).expect("Should be valid ELF");
+        let free_areas = get_free_mmap_areas(mbi);
         elf.program_header_iter()
+            .inspect(|ph| assert_load_segment_fits_into_memory(ph.vaddr(), ph.memsz(), &free_areas))
             .filter(|ph| ph.ph_type() == ProgramType::LOAD)
             .for_each(|ph| {
                 map_memory(ph);
@@ -53,9 +93,9 @@ pub fn load_module(mut modules: multiboot::information::ModuleIter) -> ! {
             MemoryAreaType::Reserved,
         )]))
         .add_module(ModuleTag::new(
-            elf_mod.start as u32,
-            elf_mod.end as u32,
-            elf_mod.string.unwrap(),
+            elf_mod.start_address(),
+            elf_mod.end_address(),
+            elf_mod.cmdline().unwrap(),
         ))
         // Test that we can add SmbiosTag multiple times.
         .add_smbios(SmbiosTag::new(1, 1, &[1, 2, 3]))
@@ -66,7 +106,7 @@ pub fn load_module(mut modules: multiboot::information::ModuleIter) -> ! {
 
     log::info!(
         "Handing over to ELF: {}",
-        elf_mod.string.unwrap_or("<unknown>")
+        elf_mod.cmdline().unwrap_or("<unknown>")
     );
 
     // handoff
@@ -84,7 +124,7 @@ pub fn load_module(mut modules: multiboot::information::ModuleIter) -> ! {
 /// address space. The loader assumes that the addresses to not clash with the
 /// loader (or anything else).
 fn map_memory(ph: ProgramHeaderEntry) {
-    log::debug!("Mapping LOAD segment {ph:#?}");
+    debug!("Mapping LOAD segment {ph:#?}");
     let dest_ptr = ph.vaddr() as *mut u8;
     let content = ph.content().expect("Should have content");
     unsafe { core::ptr::copy(content.as_ptr(), dest_ptr, content.len()) };

integration-test/bins/multiboot2_chainloader/src/loader.rs

@@ -3,7 +3,6 @@
 #![feature(error_in_core)]
 
 mod loader;
-mod multiboot;
 
 extern crate alloc;
 
@@ -12,6 +11,7 @@ extern crate util;
 
 use util::init_environment;
 
+core::arch::global_asm!(include_str!("multiboot2_header.S"), options(att_syntax));
 core::arch::global_asm!(include_str!("start.S"), options(att_syntax));
 
 /// Entry into the Rust code from assembly using the x86 SystemV calling
@@ -20,7 +20,21 @@ core::arch::global_asm!(include_str!("start.S"), options(att_syntax));
 fn rust_entry(multiboot_magic: u32, multiboot_hdr: *const u32) -> ! {
     init_environment();
     log::debug!("multiboot_hdr={multiboot_hdr:x?}, multiboot_magic=0x{multiboot_magic:x?}");
-    let mbi = multiboot::get_mbi(multiboot_magic, multiboot_hdr as u32).unwrap();
-    let module_iter = mbi.modules().expect("Should provide modules");
-    loader::load_module(module_iter);
+    assert_eq!(multiboot_magic, multiboot2::MAGIC);
+    let mbi = unsafe { multiboot2::BootInformation::load(multiboot_hdr.cast()) }.unwrap();
+
+    if let Some(mmap) = mbi.efi_memory_map_tag() {
+        log::debug!("efi memory map:",);
+        for desc in mmap.memory_areas() {
+            log::warn!(
+                "  start=0x{:016x?} size={:016x?} type={:?}, attr={:?}",
+                desc.phys_start,
+                desc.page_count * 4096,
+                desc.ty,
+                desc.att
+            );
+        }
+    }
+
+    loader::load_module(&mbi);
 }

integration-test/bins/multiboot2_chainloader/src/main.rs

@@ -0,0 +1,25 @@
+# The assembly code uses the GNU Assembly (GAS) flavor with Intel noprefix
+# syntax.
+
+# Symbol from main.rs
+.EXTERN start
+
+.code32
+.align 8
+.section .multiboot2_header
+
+    mb2_header_start:
+        .long  0xe85250d6                # magic number
+        .long  0                         # architecture 0 (protected mode i386)
+        .long  mb2_header_end - mb2_header_start # header length
+        # checksum
+        .long  0x100000000 - (0xe85250d6 + 0 + (mb2_header_end - mb2_header_start))
+
+        # REQUIRED END TAG
+        .align 8
+        .Lmb2_header_tag_end_start:
+            .word  0       # type  (16bit)
+            .word  0       # flags (16bit)
+            .long  .Lmb2_header_tag_end_end - .Lmb2_header_tag_end_start # size  (32bit)
+        .Lmb2_header_tag_end_end:
+    mb2_header_end:

integration-test/bins/multiboot2_chainloader/src/multiboot.rs

@@ -3,19 +3,6 @@
 
 .code32
 
-.section .multiboot_header, "a", @progbits
-
-/*
- * Multiboot v1 Header.
- * Required so that we can be booted by QEMU via the "-kernel" parameter.
- */
-.align 8
-.global multiboot_header
-multiboot_header:
-    .long   0x1badb002
-    .long   0x0
-    .long  -0x1badb002
-
 .section .text
 
 .global start

integration-test/bins/multiboot2_chainloader/src/multiboot2_header.S

@@ -11,9 +11,9 @@ PHDRS
 
 SECTIONS {
     /* Chainloader linked at 8M, payload at 16M */
-    .text 16M : AT(16M) ALIGN(4K)
+    .text 24M : AT(24M) ALIGN(4K)
     {
-        *(.multiboot2_header)
+        KEEP(*(.multiboot2_header));
         *(.text .text.*)
     } : kernel_rx
 

integration-test/bins/multiboot2_chainloader/src/start.S

@@ -1,10 +1,10 @@
 use good_memory_allocator::SpinLockedAllocator;
 
-#[repr(align(0x4000))]
-struct Align16K<T>(T);
+#[repr(align(0x1000))]
+struct PageAlign<T>(T);
 
-/// 16 KiB naturally aligned backing storage for heap.
-static mut HEAP: Align16K<[u8; 0x4000]> = Align16K([0; 0x4000]);
+/// 16 KiB page-aligned backing storage for heap.
+static mut HEAP: PageAlign<[u8; 0x4000]> = PageAlign([0; 0x4000]);
 
 #[global_allocator]
 static ALLOCATOR: SpinLockedAllocator = SpinLockedAllocator::empty();

integration-test/bins/multiboot2_payload/link.ld

@@ -44,11 +44,12 @@ function fn_build_rust_bins() {
     cargo build --release --verbose
     cd -
 
+    echo "Verifying multiboot2_chainloader ..."
     test -f $BINS_DIR/multiboot2_chainloader
     file --brief $BINS_DIR/multiboot2_chainloader | grep -q "ELF 32-bit LSB executable"
-    # For simplicity, the chainloader itself boots via Multiboot 1. Sufficient.
-    grub-file --is-x86-multiboot $BINS_DIR/multiboot2_chainloader
+    grub-file --is-x86-multiboot2 $BINS_DIR/multiboot2_chainloader
 
+    echo "Verifying multiboot2_payload ..."
     test -f $BINS_DIR/multiboot2_payload
     file --brief $BINS_DIR/multiboot2_payload | grep -q "ELF 32-bit LSB executable"
     grub-file --is-x86-multiboot2 $BINS_DIR/multiboot2_payload

integration-test/bins/util/src/allocator.rs

@@ -4,8 +4,7 @@ VERBOSE=yes
 INTERFACE_BRANDING=integration-test
 
 :integration-test
-# For simplicity reasons, the loader itself boots via Multiboot 1. Sufficient.
-PROTOCOL=multiboot
+PROTOCOL=multiboot2
 KERNEL_PATH=boot:///kernel
 KERNEL_CMDLINE=some kernel cmdline
 MODULE_PATH=boot:///payload