pvh/arch-x86_64: Write start_info to guest memory

Fill the hvm_start_info and related structures as specified
in the PVH boot protocol. Write the data structures to guest
memory at the GPA that will be stored in %rbx when the guest starts.

Signed-off-by: Colin Percival <cperciva@freebsd.org>
Co-authored-by: Alejandro Jimenez <alejandro.j.jimenez@oracle.com>

Author: cperciva

src/arch/src/x86_64/layout.rs

@@ -30,5 +30,13 @@ pub const KVM_TSS_ADDRESS: u64 = 0xfffb_d000;
 /// Address of the hvm_start_info struct used in PVH boot
 pub const PVH_INFO_START: u64 = 0x6000;
 
+/// Starting address of array of modules of hvm_modlist_entry type.
+/// Used to enable initrd support using the PVH boot ABI.
+pub const MODLIST_START: u64 = 0x6040;
+
+/// Address of memory map table used in PVH boot. Can overlap
+/// with the zero page address since they are mutually exclusive.
+pub const MEMMAP_START: u64 = 0x7000;
+
 /// The 'zero page', a.k.a linux kernel bootparams.
 pub const ZERO_PAGE_START: u64 = 0x7000;

src/arch/src/x86_64/mod.rs

@@ -1,3 +1,5 @@
+// Copyright © 2020, Oracle and/or its affiliates.
+//
 // Copyright 2018 Amazon.com, Inc. or its affiliates. All Rights Reserved.
 // SPDX-License-Identifier: Apache-2.0
 //
@@ -17,15 +19,41 @@ pub mod msr;
 pub mod regs;
 
 use linux_loader::configurator::linux::LinuxBootConfigurator;
+use linux_loader::configurator::pvh::PvhBootConfigurator;
 use linux_loader::configurator::{BootConfigurator, BootParams};
 use linux_loader::loader::bootparam::boot_params;
-use vm_memory::{Address, GuestAddress, GuestMemory, GuestMemoryMmap, GuestMemoryRegion};
-
+use linux_loader::loader::elf::start_info::{
+    hvm_memmap_table_entry, hvm_modlist_entry, hvm_start_info,
+};
+use vm_memory::{
+    Address, ByteValued, GuestAddress, GuestMemory, GuestMemoryMmap, GuestMemoryRegion,
+};
+
+use super::BootProtocol;
 use crate::InitrdConfig;
 
 // Value taken from https://elixir.bootlin.com/linux/v5.10.68/source/arch/x86/include/uapi/asm/e820.h#L31
 const E820_RAM: u32 = 1;
 
+// Workaround for the Rust orphan rules that guarantee trait coherence by wrapping the foreign type
+// in a tuple structure. Same approach is used by boot_params and BootParamsWrapper.
+#[derive(Copy, Clone, Default)]
+struct StartInfoWrapper(hvm_start_info);
+
+#[derive(Copy, Clone, Default)]
+struct MemmapTableEntryWrapper(hvm_memmap_table_entry);
+
+#[derive(Copy, Clone, Default)]
+struct ModlistEntryWrapper(hvm_modlist_entry);
+
+// It is safe to initialize the following structures. They are wrappers over the structures
+// defined by the start_info module, all of which are formed by fields of integer values.
+unsafe impl ByteValued for StartInfoWrapper {}
+unsafe impl ByteValued for MemmapTableEntryWrapper {}
+unsafe impl ByteValued for ModlistEntryWrapper {}
+
+const MEMMAP_TYPE_RAM: u32 = 1;
+
 /// Errors thrown while configuring x86_64 system.
 #[derive(Debug, PartialEq, derive_more::From)]
 pub enum Error {
@@ -37,6 +65,12 @@ pub enum Error {
     ZeroPageSetup,
     /// Failed to compute initrd address.
     InitrdAddress,
+    /// Error writing module entry to guest memory.
+    ModlistSetup,
+    /// Error writing memory map table to guest memory.
+    MemmapTableSetup,
+    /// Error writing hvm_start_info to guest memory.
+    StartInfoSetup,
 }
 
 // Where BIOS/VGA magic would live on a real PC.
@@ -97,12 +131,139 @@ pub fn initrd_load_addr(guest_mem: &GuestMemoryMmap, initrd_size: usize) -> supe
 /// * `cmdline_size` - Size of the kernel command line in bytes including the null terminator.
 /// * `initrd` - Information about where the ramdisk image was loaded in the `guest_mem`.
 /// * `num_cpus` - Number of virtual CPUs the guest will have.
+/// * `boot_prot` - Boot protocol that will be used to boot the guest.
 pub fn configure_system(
     guest_mem: &GuestMemoryMmap,
     cmdline_addr: GuestAddress,
     cmdline_size: usize,
     initrd: &Option<InitrdConfig>,
     num_cpus: u8,
+    boot_prot: BootProtocol,
+) -> super::Result<()> {
+    // Note that this puts the mptable at the last 1k of Linux's 640k base RAM
+    mptable::setup_mptable(guest_mem, num_cpus).map_err(Error::MpTableSetup)?;
+
+    match boot_prot {
+        BootProtocol::PvhBoot => {
+            configure_pvh(guest_mem, cmdline_addr, initrd)?;
+        }
+        BootProtocol::LinuxBoot => {
+            configure_64bit_boot(guest_mem, cmdline_addr, cmdline_size, initrd)?;
+        }
+    }
+
+    Ok(())
+}
+
+fn configure_pvh(
+    guest_mem: &GuestMemoryMmap,
+    cmdline_addr: GuestAddress,
+    initrd: &Option<InitrdConfig>,
+) -> super::Result<()> {
+    const XEN_HVM_START_MAGIC_VALUE: u32 = 0x336e_c578;
+    let first_addr_past_32bits = GuestAddress(FIRST_ADDR_PAST_32BITS);
+    let end_32bit_gap_start = GuestAddress(MMIO_MEM_START);
+    let himem_start = GuestAddress(layout::HIMEM_START);
+
+    // Vector to hold modules (currently either empty or holding initrd).
+    let mut modules: Vec<hvm_modlist_entry> = Vec::new();
+    if let Some(initrd_config) = initrd {
+        // The initrd has been written to guest memory already, here we just need to
+        // create the module structure that describes it.
+        modules.push(hvm_modlist_entry {
+            paddr: initrd_config.address.raw_value(),
+            size: initrd_config.size as u64,
+            ..Default::default()
+        });
+    }
+
+    // Vector to hold the memory maps which needs to be written to guest memory
+    // at MEMMAP_START after all of the mappings are recorded.
+    let mut memmap: Vec<hvm_memmap_table_entry> = Vec::new();
+
+    // Create the memory map entries.
+    add_memmap_entry(&mut memmap, 0, EBDA_START, MEMMAP_TYPE_RAM)?;
+    let last_addr = guest_mem.last_addr();
+    if last_addr < end_32bit_gap_start {
+        add_memmap_entry(
+            &mut memmap,
+            himem_start.raw_value() as u64,
+            last_addr.unchecked_offset_from(himem_start) as u64 + 1,
+            MEMMAP_TYPE_RAM,
+        )?;
+    } else {
+        add_memmap_entry(
+            &mut memmap,
+            himem_start.raw_value(),
+            end_32bit_gap_start.unchecked_offset_from(himem_start),
+            MEMMAP_TYPE_RAM,
+        )?;
+
+        if last_addr > first_addr_past_32bits {
+            add_memmap_entry(
+                &mut memmap,
+                first_addr_past_32bits.raw_value(),
+                last_addr.unchecked_offset_from(first_addr_past_32bits) + 1,
+                MEMMAP_TYPE_RAM,
+            )?;
+        }
+    }
+
+    // Construct the hvm_start_info structure and serialize it into
+    // boot_params.  This will be stored at PVH_INFO_START address, and %rbx
+    // will be initialized to contain PVH_INFO_START prior to starting the
+    // guest, as required by the PVH ABI.
+    let mut start_info = hvm_start_info {
+        magic: XEN_HVM_START_MAGIC_VALUE,
+        version: 1,
+        cmdline_paddr: cmdline_addr.raw_value(),
+        memmap_paddr: layout::MEMMAP_START,
+        memmap_entries: memmap.len() as u32,
+        nr_modules: modules.len() as u32,
+        ..Default::default()
+    };
+    if !modules.is_empty() {
+        start_info.modlist_paddr = layout::MODLIST_START;
+    }
+    let mut boot_params =
+        BootParams::new::<hvm_start_info>(&start_info, GuestAddress(layout::PVH_INFO_START));
+
+    // Copy the vector with the memmap table to the MEMMAP_START address
+    // which is already saved in the memmap_paddr field of hvm_start_info struct.
+    boot_params.set_sections::<hvm_memmap_table_entry>(&memmap, GuestAddress(layout::MEMMAP_START));
+
+    // Copy the vector with the modules list to the MODLIST_START address.
+    // Note that we only set the modlist_paddr address if there is a nonzero
+    // number of modules, but serializing an empty list is harmless.
+    boot_params.set_modules::<hvm_modlist_entry>(&modules, GuestAddress(layout::MODLIST_START));
+
+    // Write the hvm_start_info struct to guest memory.
+    PvhBootConfigurator::write_bootparams(&boot_params, guest_mem)
+        .map_err(|_| Error::StartInfoSetup)
+}
+
+fn add_memmap_entry(
+    memmap: &mut Vec<hvm_memmap_table_entry>,
+    addr: u64,
+    size: u64,
+    mem_type: u32,
+) -> super::Result<()> {
+    // Add the table entry to the vector
+    memmap.push(hvm_memmap_table_entry {
+        addr,
+        size,
+        type_: mem_type,
+        reserved: 0,
+    });
+
+    Ok(())
+}
+
+fn configure_64bit_boot(
+    guest_mem: &GuestMemoryMmap,
+    cmdline_addr: GuestAddress,
+    cmdline_size: usize,
+    initrd: &Option<InitrdConfig>,
 ) -> super::Result<()> {
     const KERNEL_BOOT_FLAG_MAGIC: u16 = 0xaa55;
     const KERNEL_HDR_MAGIC: u32 = 0x5372_6448;
@@ -113,9 +274,6 @@ pub fn configure_system(
 
     let himem_start = GuestAddress(layout::HIMEM_START);
 
-    // Note that this puts the mptable at the last 1k of Linux's 640k base RAM
-    mptable::setup_mptable(guest_mem, num_cpus)?;
-
     let mut params = boot_params::default();
 
     params.hdr.type_of_loader = KERNEL_LOADER_OTHER;
@@ -218,7 +376,8 @@ mod tests {
         let gm =
             vm_memory::test_utils::create_anon_guest_memory(&[(GuestAddress(0), 0x10000)], false)
                 .unwrap();
-        let config_err = configure_system(&gm, GuestAddress(0), 0, &None, 1);
+        let config_err =
+            configure_system(&gm, GuestAddress(0), 0, &None, 1, BootProtocol::LinuxBoot);
         assert!(config_err.is_err());
         assert_eq!(
             config_err.unwrap_err(),
@@ -229,19 +388,70 @@ mod tests {
         let mem_size = 128 << 20;
         let arch_mem_regions = arch_memory_regions(mem_size);
         let gm = vm_memory::test_utils::create_anon_guest_memory(&arch_mem_regions, false).unwrap();
-        configure_system(&gm, GuestAddress(0), 0, &None, no_vcpus).unwrap();
+        configure_system(
+            &gm,
+            GuestAddress(0),
+            0,
+            &None,
+            no_vcpus,
+            BootProtocol::LinuxBoot,
+        )
+        .unwrap();
+        configure_system(
+            &gm,
+            GuestAddress(0),
+            0,
+            &None,
+            no_vcpus,
+            BootProtocol::PvhBoot,
+        )
+        .unwrap();
 
         // Now assigning some memory that is equal to the start of the 32bit memory hole.
         let mem_size = 3328 << 20;
         let arch_mem_regions = arch_memory_regions(mem_size);
         let gm = vm_memory::test_utils::create_anon_guest_memory(&arch_mem_regions, false).unwrap();
-        configure_system(&gm, GuestAddress(0), 0, &None, no_vcpus).unwrap();
+        configure_system(
+            &gm,
+            GuestAddress(0),
+            0,
+            &None,
+            no_vcpus,
+            BootProtocol::LinuxBoot,
+        )
+        .unwrap();
+        configure_system(
+            &gm,
+            GuestAddress(0),
+            0,
+            &None,
+            no_vcpus,
+            BootProtocol::PvhBoot,
+        )
+        .unwrap();
 
         // Now assigning some memory that falls after the 32bit memory hole.
         let mem_size = 3330 << 20;
         let arch_mem_regions = arch_memory_regions(mem_size);
         let gm = vm_memory::test_utils::create_anon_guest_memory(&arch_mem_regions, false).unwrap();
-        configure_system(&gm, GuestAddress(0), 0, &None, no_vcpus).unwrap();
+        configure_system(
+            &gm,
+            GuestAddress(0),
+            0,
+            &None,
+            no_vcpus,
+            BootProtocol::LinuxBoot,
+        )
+        .unwrap();
+        configure_system(
+            &gm,
+            GuestAddress(0),
+            0,
+            &None,
+            no_vcpus,
+            BootProtocol::PvhBoot,
+        )
+        .unwrap();
     }
 
     #[test]
@@ -283,4 +493,31 @@ mod tests {
         )
         .is_err());
     }
+
+    #[test]
+    fn test_add_memmap_entry() {
+        const MEMMAP_TYPE_RESERVED: u32 = 2;
+
+        let mut memmap: Vec<hvm_memmap_table_entry> = Vec::new();
+
+        let expected_memmap = vec![
+            hvm_memmap_table_entry {
+                addr: 0x0,
+                size: 0x1000,
+                type_: MEMMAP_TYPE_RAM,
+                ..Default::default()
+            },
+            hvm_memmap_table_entry {
+                addr: 0x10000,
+                size: 0xa000,
+                type_: MEMMAP_TYPE_RESERVED,
+                ..Default::default()
+            },
+        ];
+
+        add_memmap_entry(&mut memmap, 0, 0x1000, MEMMAP_TYPE_RAM).unwrap();
+        add_memmap_entry(&mut memmap, 0x10000, 0xa000, MEMMAP_TYPE_RESERVED).unwrap();
+
+        assert_eq!(format!("{:?}", memmap), format!("{:?}", expected_memmap));
+    }
 }

src/vmm/src/builder.rs

@@ -895,6 +895,7 @@ pub fn configure_system_for_boot(
             boot_cmdline.as_str().len() + 1,
             initrd,
             vcpus.len() as u8,
+            entry_point.protocol,
         )
         .map_err(ConfigureSystem)?;
     }