kata-containers/cloud-hypervisor
1
0
Fork 0
mirror of https://github.com/cloud-hypervisor/cloud-hypervisor.git synced 2026-08-05 02:19:16 +00:00
Code Issues Packages Projects Releases Wiki Activity

arch, vmm: Remove support for LinuxBoot

Browse Source 
By supporting just PVH boot on x86-64 we simplify our boot path
substatially.

Fixes: #2231

Signed-off-by: Rob Bradford <robert.bradford@intel.com>
This commit is contained in:
Rob Bradford
2021-04-29 17:11:32 +01:00
committed by Sebastien Boeuf
parent 1f4a34ffc8
commit da8136e49d
6 changed files with 62 additions and 568 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
arch/src/lib.rs
View File

@@ -98,8 +98,8 @@ pub mod x86_64;
#[cfg(target_arch = "x86_64")]
pub use x86_64::{
arch_memory_regions, configure_system, configure_vcpu, get_host_cpu_phys_bits,
initramfs_load_addr, layout, layout::CMDLINE_MAX_SIZE, layout::CMDLINE_START, regs,
BootProtocol, CpuidPatch, CpuidReg, EntryPoint,
initramfs_load_addr, layout, layout::CMDLINE_MAX_SIZE, layout::CMDLINE_START, regs, CpuidPatch,
CpuidReg, EntryPoint,
};
/// Safe wrapper for `sysconf(_SC_PAGESIZE)`.

305
arch/src/x86_64/mod.rs
View File

@@ -14,7 +14,7 @@ pub mod regs;
use crate::InitramfsConfig;
use crate::RegionType;
use hypervisor::{CpuId, CpuIdEntry, CPUID_FLAG_VALID_INDEX};
use linux_loader::loader::bootparam::{boot_params, setup_header};
use linux_loader::loader::bootparam::boot_params;
use linux_loader::loader::elf::start_info::{
hvm_memmap_table_entry, hvm_modlist_entry, hvm_start_info,
};
@@ -28,21 +28,6 @@ use std::arch::x86_64;
#[cfg(feature = "tdx")]
pub mod tdx;
#[derive(Debug, Copy, Clone)]
pub enum BootProtocol {
LinuxBoot,
PvhBoot,
}
impl ::std::fmt::Display for BootProtocol {
fn fmt(&self, f: &mut ::std::fmt::Formatter) -> ::std::fmt::Result {
match self {
BootProtocol::LinuxBoot => write!(f, "Linux 64-bit boot protocol"),
BootProtocol::PvhBoot => write!(f, "PVH boot protocol"),
}
}
}
#[derive(Debug, Copy, Clone)]
/// Specifies the entry point address where the guest must start
/// executing code, as well as which of the supported boot protocols
@@ -50,10 +35,6 @@ impl ::std::fmt::Display for BootProtocol {
pub struct EntryPoint {
/// Address in guest memory where the guest must start execution
pub entry_addr: GuestAddress,
/// Specifies which boot protocol to use
pub protocol: BootProtocol,
/// This field is used for bzImage to fill zero page
pub setup_header: Option<setup_header>,
}
const E820_RAM: u32 = 1;
@@ -351,17 +332,10 @@ pub fn configure_vcpu(
regs::setup_msrs(fd).map_err(Error::MsrsConfiguration)?;
if let Some(kernel_entry_point) = kernel_entry_point {
// Safe to unwrap because this method is called after the VM is configured
regs::setup_regs(
fd,
kernel_entry_point.entry_addr.raw_value(),
layout::BOOT_STACK_POINTER.raw_value(),
layout::ZERO_PAGE_START.raw_value(),
kernel_entry_point.protocol,
)
.map_err(Error::RegsConfiguration)?;
regs::setup_regs(fd, kernel_entry_point.entry_addr.raw_value())
.map_err(Error::RegsConfiguration)?;
regs::setup_fpu(fd).map_err(Error::FpuConfiguration)?;
regs::setup_sregs(&vm_memory.memory(), fd, kernel_entry_point.protocol)
.map_err(Error::SregsConfiguration)?;
regs::setup_sregs(&vm_memory.memory(), fd).map_err(Error::SregsConfiguration)?;
}
interrupts::set_lint(fd).map_err(|e| Error::LocalIntConfiguration(e.into()))?;
Ok(())
@@ -426,12 +400,9 @@ pub fn arch_memory_regions(size: GuestUsize) -> Vec<(GuestAddress, usize, Region
pub fn configure_system(
guest_mem: &GuestMemoryMmap,
cmdline_addr: GuestAddress,
cmdline_size: usize,
initramfs: &Option<InitramfsConfig>,
_num_cpus: u8,
setup_hdr: Option<setup_header>,
rsdp_addr: Option<GuestAddress>,
boot_prot: BootProtocol,
sgx_epc_region: Option<SgxEpcRegion>,
) -> super::Result<()> {
let size = smbios::setup_smbios(guest_mem).map_err(Error::SmbiosSetup)?;
@@ -448,31 +419,13 @@ pub fn configure_system(
}
}
match boot_prot {
BootProtocol::PvhBoot => {
configure_pvh(
guest_mem,
cmdline_addr,
initramfs,
rsdp_addr,
sgx_epc_region,
)?;
}
BootProtocol::LinuxBoot => {
error!("Using deprecated LinuxBoot protocol: Please configure your kernel with CONFIG_PVH=y and supply the `vmlinux` file to `--kernel`");
configure_64bit_boot(
guest_mem,
cmdline_addr,
cmdline_size,
initramfs,
setup_hdr,
rsdp_addr,
sgx_epc_region,
)?;
}
}
Ok(())
configure_pvh(
guest_mem,
cmdline_addr,
initramfs,
rsdp_addr,
sgx_epc_region,
)
}
fn configure_pvh(
@@ -615,121 +568,6 @@ fn add_memmap_entry(memmap: &mut Vec<hvm_memmap_table_entry>, addr: u64, size: u
});
}
fn configure_64bit_boot(
guest_mem: &GuestMemoryMmap,
cmdline_addr: GuestAddress,
cmdline_size: usize,
initramfs: &Option<InitramfsConfig>,
setup_hdr: Option<setup_header>,
rsdp_addr: Option<GuestAddress>,
sgx_epc_region: Option<SgxEpcRegion>,
) -> super::Result<()> {
const KERNEL_BOOT_FLAG_MAGIC: u16 = 0xaa55;
const KERNEL_HDR_MAGIC: u32 = 0x53726448;
const KERNEL_LOADER_OTHER: u8 = 0xff;
const KERNEL_MIN_ALIGNMENT_BYTES: u32 = 0x1000000; // Must be non-zero.
let mut params: BootParamsWrapper = BootParamsWrapper(boot_params::default());
if let Some(hdr) = setup_hdr {
// We should use the header if the loader provides one (e.g. from a bzImage).
params.0.hdr = hdr;
} else {
params.0.hdr.boot_flag = KERNEL_BOOT_FLAG_MAGIC;
params.0.hdr.header = KERNEL_HDR_MAGIC;
params.0.hdr.kernel_alignment = KERNEL_MIN_ALIGNMENT_BYTES;
};
// Common bootparams settings
if params.0.hdr.type_of_loader == 0 {
params.0.hdr.type_of_loader = KERNEL_LOADER_OTHER;
}
params.0.hdr.cmd_line_ptr = cmdline_addr.raw_value() as u32;
params.0.hdr.cmdline_size = cmdline_size as u32;
if let Some(initramfs_config) = initramfs {
params.0.hdr.ramdisk_image = initramfs_config.address.raw_value() as u32;
params.0.hdr.ramdisk_size = initramfs_config.size as u32;
}
add_e820_entry(&mut params.0, 0, layout::EBDA_START.raw_value(), E820_RAM)?;
let mem_end = guest_mem.last_addr();
if mem_end < layout::MEM_32BIT_RESERVED_START {
add_e820_entry(
&mut params.0,
layout::HIGH_RAM_START.raw_value(),
mem_end.unchecked_offset_from(layout::HIGH_RAM_START) + 1,
E820_RAM,
)?;
} else {
add_e820_entry(
&mut params.0,
layout::HIGH_RAM_START.raw_value(),
layout::MEM_32BIT_RESERVED_START.unchecked_offset_from(layout::HIGH_RAM_START),
E820_RAM,
)?;
if mem_end > layout::RAM_64BIT_START {
add_e820_entry(
&mut params.0,
layout::RAM_64BIT_START.raw_value(),
mem_end.unchecked_offset_from(layout::RAM_64BIT_START) + 1,
E820_RAM,
)?;
}
}
add_e820_entry(
&mut params.0,
layout::PCI_MMCONFIG_START.0,
layout::PCI_MMCONFIG_SIZE,
E820_RESERVED,
)?;
if let Some(sgx_epc_region) = sgx_epc_region {
add_e820_entry(
&mut params.0,
sgx_epc_region.start().raw_value(),
sgx_epc_region.size() as u64,
E820_RESERVED,
)?;
}
if let Some(rsdp_addr) = rsdp_addr {
params.0.acpi_rsdp_addr = rsdp_addr.0;
}
let zero_page_addr = layout::ZERO_PAGE_START;
guest_mem
.checked_offset(zero_page_addr, mem::size_of::<boot_params>())
.ok_or(super::Error::ZeroPagePastRamEnd)?;
guest_mem
.write_obj(params, zero_page_addr)
.map_err(super::Error::ZeroPageSetup)?;
Ok(())
}
/// Add an e820 region to the e820 map.
/// Returns Ok(()) if successful, or an error if there is no space left in the map.
fn add_e820_entry(
params: &mut boot_params,
addr: u64,
size: u64,
mem_type: u32,
) -> Result<(), Error> {
if params.e820_entries >= params.e820_table.len() as u8 {
return Err(Error::E820Configuration);
}
params.e820_table[params.e820_entries as usize].addr = addr;
params.e820_table[params.e820_entries as usize].size = size;
params.e820_table[params.e820_entries as usize].type_ = mem_type;
params.e820_entries += 1;
Ok(())
}
/// Returns the memory address where the initramfs could be loaded.
pub fn initramfs_load_addr(
guest_mem: &GuestMemoryMmap,
@@ -888,7 +726,6 @@ pub fn update_cpuid_sgx(cpuid: &mut CpuId, epc_sections: Vec<SgxEpcSection>) ->
#[cfg(test)]
mod tests {
use super::*;
use linux_loader::loader::bootparam::boot_e820_entry;
#[test]
fn regions_lt_4gb() {
@@ -913,12 +750,9 @@ mod tests {
let config_err = configure_system(
&gm,
GuestAddress(0),
0,
&None,
1,
None,
Some(layout::RSDP_POINTER),
BootProtocol::LinuxBoot,
None,
);
assert!(config_err.is_err());
@@ -932,31 +766,8 @@ mod tests {
.map(|r| (r.0, r.1))
.collect();
let gm = GuestMemoryMmap::from_ranges(&ram_regions).unwrap();
configure_system(
&gm,
GuestAddress(0),
0,
&None,
no_vcpus,
None,
None,
BootProtocol::LinuxBoot,
None,
)
.unwrap();
configure_system(
&gm,
GuestAddress(0),
0,
&None,
no_vcpus,
None,
None,
BootProtocol::PvhBoot,
None,
)
.unwrap();
configure_system(&gm, GuestAddress(0), &None, no_vcpus, None, None).unwrap();
// Now assigning some memory that is equal to the start of the 32bit memory hole.
let mem_size = 3328 << 20;
@@ -967,31 +778,9 @@ mod tests {
.map(|r| (r.0, r.1))
.collect();
let gm = GuestMemoryMmap::from_ranges(&ram_regions).unwrap();
configure_system(
&gm,
GuestAddress(0),
0,
&None,
no_vcpus,
None,
None,
BootProtocol::LinuxBoot,
None,
)
.unwrap();
configure_system(&gm, GuestAddress(0), &None, no_vcpus, None, None).unwrap();
configure_system(
&gm,
GuestAddress(0),
0,
&None,
no_vcpus,
None,
None,
BootProtocol::PvhBoot,
None,
)
.unwrap();
configure_system(&gm, GuestAddress(0), &None, no_vcpus, None, None).unwrap();
// Now assigning some memory that falls after the 32bit memory hole.
let mem_size = 3330 << 20;
@@ -1002,71 +791,9 @@ mod tests {
.map(|r| (r.0, r.1))
.collect();
let gm = GuestMemoryMmap::from_ranges(&ram_regions).unwrap();
configure_system(
&gm,
GuestAddress(0),
0,
&None,
no_vcpus,
None,
None,
BootProtocol::LinuxBoot,
None,
)
.unwrap();
configure_system(&gm, GuestAddress(0), &None, no_vcpus, None, None).unwrap();
configure_system(
&gm,
GuestAddress(0),
0,
&None,
no_vcpus,
None,
None,
BootProtocol::PvhBoot,
None,
)
.unwrap();
}
#[test]
fn test_add_e820_entry() {
let e820_table = [(boot_e820_entry {
addr: 0x1,
size: 4,
type_: 1,
}); 128];
let expected_params = boot_params {
e820_table,
e820_entries: 1,
..Default::default()
};
let mut params: boot_params = Default::default();
add_e820_entry(
&mut params,
e820_table[0].addr,
e820_table[0].size,
e820_table[0].type_,
)
.unwrap();
assert_eq!(
format!("{:?}", params.e820_table[0]),
format!("{:?}", expected_params.e820_table[0])
);
assert_eq!(params.e820_entries, expected_params.e820_entries);
// Exercise the scenario where the field storing the length of the e820 entry table is
// is bigger than the allocated memory.
params.e820_entries = params.e820_table.len() as u8 + 1;
assert!(add_e820_entry(
&mut params,
e820_table[0].addr,
e820_table[0].size,
e820_table[0].type_
)
.is_err());
configure_system(&gm, GuestAddress(0), &None, no_vcpus, None, None).unwrap();
}
#[test]

184
arch/src/x86_64/regs.rs
View File

@@ -9,13 +9,10 @@
use std::sync::Arc;
use std::{mem, result};
use super::BootProtocol;
use hypervisor::arch::x86::gdt::{gdt_entry, segment_from_gdt};
use hypervisor::arch::x86::regs::*;
use hypervisor::x86_64::{FpuState, SpecialRegisters, StandardRegisters};
use layout::{
BOOT_GDT_START, BOOT_IDT_START, PDE_START, PDPTE_START, PML4_START, PML5_START, PVH_INFO_START,
};
use layout::{BOOT_GDT_START, BOOT_IDT_START, PVH_INFO_START};
use vm_memory::{Address, Bytes, GuestMemory, GuestMemoryError, GuestMemoryMmap};
#[derive(Debug)]
@@ -81,32 +78,12 @@ pub fn setup_msrs(vcpu: &Arc<dyn hypervisor::Vcpu>) -> Result<()> {
///
/// * `vcpu` - Structure for the VCPU that holds the VCPU's fd.
/// * `boot_ip` - Starting instruction pointer.
/// * `boot_sp` - Starting stack pointer.
/// * `boot_si` - Must point to zero page address per Linux ABI.
pub fn setup_regs(
vcpu: &Arc<dyn hypervisor::Vcpu>,
boot_ip: u64,
boot_sp: u64,
boot_si: u64,
boot_prot: BootProtocol,
) -> Result<()> {
let regs: StandardRegisters = match boot_prot {
// Configure regs as required by PVH boot protocol.
BootProtocol::PvhBoot => StandardRegisters {
rflags: 0x0000000000000002u64,
rbx: PVH_INFO_START.raw_value(),
rip: boot_ip,
..Default::default()
},
// Configure regs as required by Linux 64-bit boot protocol.
BootProtocol::LinuxBoot => StandardRegisters {
rflags: 0x0000000000000002u64,
rip: boot_ip,
rsp: boot_sp,
rbp: boot_sp,
rsi: boot_si,
..Default::default()
},
pub fn setup_regs(vcpu: &Arc<dyn hypervisor::Vcpu>, boot_ip: u64) -> Result<()> {
let regs = StandardRegisters {
rflags: 0x0000000000000002u64,
rbx: PVH_INFO_START.raw_value(),
rip: boot_ip,
..Default::default()
};
vcpu.set_regs(&regs).map_err(Error::SetBaseRegisters)
}
@@ -117,19 +94,9 @@ pub fn setup_regs(
///
/// * `mem` - The memory that will be passed to the guest.
/// * `vcpu` - Structure for the VCPU that holds the VCPU's fd.
pub fn setup_sregs(
mem: &GuestMemoryMmap,
vcpu: &Arc<dyn hypervisor::Vcpu>,
boot_prot: BootProtocol,
) -> Result<()> {
pub fn setup_sregs(mem: &GuestMemoryMmap, vcpu: &Arc<dyn hypervisor::Vcpu>) -> Result<()> {
let mut sregs: SpecialRegisters = vcpu.get_sregs().map_err(Error::GetStatusRegisters)?;
configure_segments_and_sregs(mem, &mut sregs, boot_prot)?;
if let BootProtocol::LinuxBoot = boot_prot {
setup_page_tables(mem, &mut sregs)?; // TODO(dgreid) - Can this be done once per system instead?
}
configure_segments_and_sregs(mem, &mut sregs)?;
vcpu.set_sregs(&sregs).map_err(Error::SetStatusRegisters)
}
@@ -156,27 +123,15 @@ fn write_idt_value(val: u64, guest_mem: &GuestMemoryMmap) -> Result<()> {
pub fn configure_segments_and_sregs(
mem: &GuestMemoryMmap,
sregs: &mut SpecialRegisters,
boot_prot: BootProtocol,
) -> Result<()> {
let gdt_table: [u64; BOOT_GDT_MAX as usize] = match boot_prot {
BootProtocol::PvhBoot => {
// Configure GDT entries as specified by PVH boot protocol
[
gdt_entry(0, 0, 0), // NULL
gdt_entry(0xc09b, 0, 0xffffffff), // CODE
gdt_entry(0xc093, 0, 0xffffffff), // DATA
gdt_entry(0x008b, 0, 0x67), // TSS
]
}
BootProtocol::LinuxBoot => {
// Configure GDT entries as specified by Linux 64bit boot protocol
[
gdt_entry(0, 0, 0), // NULL
gdt_entry(0xa09b, 0, 0xfffff), // CODE
gdt_entry(0xc093, 0, 0xfffff), // DATA
gdt_entry(0x808b, 0, 0xfffff), // TSS
]
}
let gdt_table: [u64; BOOT_GDT_MAX as usize] = {
// Configure GDT entries as specified by PVH boot protocol
[
gdt_entry(0, 0, 0), // NULL
gdt_entry(0xc09b, 0, 0xffffffff), // CODE
gdt_entry(0xc093, 0, 0xffffffff), // DATA
gdt_entry(0x008b, 0, 0x67), // TSS
]
};
let code_seg = segment_from_gdt(gdt_table[1], 1);
@@ -200,51 +155,8 @@ pub fn configure_segments_and_sregs(
sregs.ss = data_seg;
sregs.tr = tss_seg;
match boot_prot {
BootProtocol::PvhBoot => {
sregs.cr0 = CR0_PE;
sregs.cr4 = 0;
}
BootProtocol::LinuxBoot => {
/* 64-bit protected mode */
sregs.cr0 |= CR0_PE;
sregs.efer |= EFER_LME | EFER_LMA;
}
}
Ok(())
}
pub fn setup_page_tables(mem: &GuestMemoryMmap, sregs: &mut SpecialRegisters) -> Result<()> {
// Puts PML5 or PML4 right after zero page but aligned to 4k.
if unsafe { std::arch::x86_64::__cpuid(7).ecx } & (1 << 16) != 0 {
// Entry covering VA [0..256TB)
mem.write_obj(PML4_START.raw_value() | 0x03, PML5_START)
.map_err(Error::WritePml5Address)?;
sregs.cr3 = PML5_START.raw_value();
sregs.cr4 |= CR4_LA57;
} else {
sregs.cr3 = PML4_START.raw_value();
}
// Entry covering VA [0..512GB)
mem.write_obj(PDPTE_START.raw_value() | 0x03, PML4_START)
.map_err(Error::WritePml4Address)?;
// Entry covering VA [0..1GB)
mem.write_obj(PDE_START.raw_value() | 0x03, PDPTE_START)
.map_err(Error::WritePdpteAddress)?;
// 512 2MB entries together covering VA [0..1GB). Note we are assuming
// CPU supports 2MB pages (/proc/cpuinfo has 'pse'). All modern CPUs do.
for i in 0..512 {
mem.write_obj((i << 21) + 0x83u64, PDE_START.unchecked_add(i * 8))
.map_err(Error::WritePdeAddress)?;
}
sregs.cr4 |= CR4_PAE;
sregs.cr0 |= CR0_PG;
sregs.cr0 = CR0_PE;
sregs.cr4 = 0;
Ok(())
}
@@ -268,36 +180,7 @@ mod tests {
fn segments_and_sregs() {
let mut sregs: SpecialRegisters = Default::default();
let gm = create_guest_mem();
configure_segments_and_sregs(&gm, &mut sregs, BootProtocol::LinuxBoot).unwrap();
assert_eq!(0x0, read_u64(&gm, BOOT_GDT_START));
assert_eq!(
0xaf9b000000ffff,
read_u64(&gm, BOOT_GDT_START.unchecked_add(8))
);
assert_eq!(
0xcf93000000ffff,
read_u64(&gm, BOOT_GDT_START.unchecked_add(16))
);
assert_eq!(
0x8f8b000000ffff,
read_u64(&gm, BOOT_GDT_START.unchecked_add(24))
);
assert_eq!(0x0, read_u64(&gm, BOOT_IDT_START));
assert_eq!(0, sregs.cs.base);
assert_eq!(0xffffffff, sregs.ds.limit);
assert_eq!(0x10, sregs.es.selector);
assert_eq!(1, sregs.fs.present);
assert_eq!(1, sregs.gs.g);
assert_eq!(0, sregs.ss.avl);
assert_eq!(0, sregs.tr.base);
assert_eq!(0xffffffff, sregs.tr.limit);
assert_eq!(0, sregs.tr.avl);
assert_eq!(CR0_PE, sregs.cr0);
assert_eq!(EFER_LME | EFER_LMA, sregs.efer);
configure_segments_and_sregs(&gm, &mut sregs, BootProtocol::PvhBoot).unwrap();
configure_segments_and_sregs(&gm, &mut sregs).unwrap();
assert_eq!(0x0, read_u64(&gm, BOOT_GDT_START));
assert_eq!(
0xcf9b000000ffff,
@@ -327,31 +210,4 @@ mod tests {
assert_eq!(CR0_PE, sregs.cr0);
assert_eq!(0, sregs.cr4);
}
#[test]
fn page_tables() {
let mut sregs: SpecialRegisters = Default::default();
let gm = create_guest_mem();
setup_page_tables(&gm, &mut sregs).unwrap();
if unsafe { std::arch::x86_64::__cpuid(7).ecx } & (1 << 16) != 0 {
assert_eq!(0xa003, read_u64(&gm, PML5_START));
}
assert_eq!(0xb003, read_u64(&gm, PML4_START));
assert_eq!(0xc003, read_u64(&gm, PDPTE_START));
for i in 0..512 {
assert_eq!(
(i << 21) + 0x83u64,
read_u64(&gm, PDE_START.unchecked_add(i * 8))
);
}
if unsafe { std::arch::x86_64::__cpuid(7).ecx } & (1 << 16) != 0 {
assert_eq!(PML5_START.raw_value(), sregs.cr3);
} else {
assert_eq!(PML4_START.raw_value(), sregs.cr3);
}
assert_eq!(CR4_PAE, sregs.cr4);
assert_eq!(CR0_PG, sregs.cr0);
}
}