openvm: Guest program compilation with stock rust compiler. (#115)

Cargo.lock

@@ -2314,7 +2314,9 @@ dependencies = [
 name = "ere-openvm"
 version = "0.0.11"
 dependencies = [
+ "anyhow",
  "build-utils",
+ "cargo_metadata 0.19.2",
  "openvm-build",
  "openvm-circuit",
  "openvm-continuations",
@@ -2325,6 +2327,7 @@ dependencies = [
  "test-utils",
  "thiserror 2.0.12",
  "toml",
+ "tracing",
  "zkvm-interface",
 ]
 

crates/ere-openvm/Cargo.toml

@@ -6,8 +6,11 @@ rust-version.workspace = true
 license.workspace = true
 
 [dependencies]
+anyhow.workspace = true
 serde = { workspace = true, features = ["derive"] }
 thiserror.workspace = true
+cargo_metadata.workspace = true
+tracing.workspace = true
 toml.workspace = true
 
 # OpenVM dependencies

crates/ere-openvm/src/compile_stock_rust.rs

@@ -0,0 +1,160 @@
+use crate::OpenVMProgram;
+use crate::error::CompileError;
+use cargo_metadata::MetadataCommand;
+use openvm_sdk::config::{AppConfig, DEFAULT_APP_LOG_BLOWUP, DEFAULT_LEAF_LOG_BLOWUP, SdkVmConfig};
+use openvm_stark_sdk::config::FriParameters;
+use std::fs;
+use std::path::Path;
+use std::process::Command;
+use tracing::info;
+
+static CARGO_ENCODED_RUSTFLAGS_SEPARATOR: &str = "\x1f";
+const TARGET_TRIPLE: &str = "riscv32ima-unknown-none-elf";
+// Rust flags according to https://github.com/openvm-org/openvm/blob/v1.4.0-rc.8/crates/toolchain/build/src/lib.rs#L291
+const RUSTFLAGS: &[&str] = &[
+    // Replace atomic ops with nonatomic versions since the guest is single threaded.
+    "-C",
+    "passes=lower-atomic",
+    // Specify where to start loading the program in
+    // memory.  The clang linker understands the same
+    // command line arguments as the GNU linker does; see
+    // https://ftp.gnu.org/old-gnu/Manuals/ld-2.9.1/html_mono/ld.html#SEC3
+    // for details.
+    "-C",
+    "link-arg=-Ttext=0x00200800",
+    // Apparently not having an entry point is only a linker warning(!), so
+    // error out in this case.
+    "-C",
+    "link-arg=--fatal-warnings",
+    "-C",
+    "panic=abort",
+    // https://docs.rs/getrandom/0.3.2/getrandom/index.html#opt-in-backends
+    "--cfg",
+    "getrandom_backend=\"custom\"",
+];
+const CARGO_ARGS: &[&str] = &[
+    "build",
+    "--target",
+    TARGET_TRIPLE,
+    "--release",
+    // For bare metal we have to build core and alloc
+    "-Zbuild-std=core,alloc",
+];
+
+pub fn compile_openvm_program_stock_rust(
+    guest_directory: &Path,
+    toolchain: &String,
+) -> Result<OpenVMProgram, CompileError> {
+    wrap_into_openvm_program(
+        compile_program_stock_rust(guest_directory, toolchain)?,
+        guest_directory,
+    )
+}
+
+fn compile_program_stock_rust(
+    guest_directory: &Path,
+    toolchain: &String,
+) -> Result<Vec<u8>, CompileError> {
+    let metadata = MetadataCommand::new().current_dir(guest_directory).exec()?;
+    let package = metadata
+        .root_package()
+        .ok_or_else(|| CompileError::MissingPackageName {
+            path: guest_directory.to_path_buf(),
+        })?;
+
+    let plus_toolchain = format!("+{}", toolchain);
+    let mut cargo_args = [plus_toolchain.as_str()].to_vec();
+    cargo_args.append(&mut CARGO_ARGS.to_vec());
+
+    let encoded_rust_flags = RUSTFLAGS.to_vec().join(CARGO_ENCODED_RUSTFLAGS_SEPARATOR);
+
+    let target_direcotry = guest_directory
+        .join("target")
+        .join(TARGET_TRIPLE)
+        .join("release");
+
+    // Remove target directory.
+    if target_direcotry.exists() {
+        fs::remove_dir_all(&target_direcotry).unwrap();
+    }
+
+    let result = Command::new("cargo")
+        .current_dir(guest_directory)
+        .env("CARGO_ENCODED_RUSTFLAGS", &encoded_rust_flags)
+        .args(cargo_args)
+        .stdout(std::process::Stdio::inherit())
+        .stderr(std::process::Stdio::inherit())
+        .status()
+        .map_err(|source| CompileError::BuildFailure {
+            source: source.into(),
+            crate_path: guest_directory.to_path_buf(),
+        });
+
+    if result.is_err() {
+        return Err(result.err().unwrap());
+    }
+
+    let elf_path = target_direcotry.join(&package.name);
+
+    fs::read(&elf_path).map_err(|e| CompileError::ReadFile {
+        path: elf_path,
+        source: e,
+    })
+}
+
+fn wrap_into_openvm_program(
+    elf: Vec<u8>,
+    guest_directory: &Path,
+) -> Result<OpenVMProgram, CompileError> {
+    let app_config_path = guest_directory.join("openvm.toml");
+    let app_config = if app_config_path.exists() {
+        let toml = fs::read_to_string(&app_config_path).map_err(|source| {
+            CompileError::ReadConfigFailed {
+                source,
+                path: app_config_path.to_path_buf(),
+            }
+        })?;
+        toml::from_str(&toml).map_err(CompileError::DeserializeConfigFailed)?
+    } else {
+        // The default `AppConfig` copied from https://github.com/openvm-org/openvm/blob/ca36de3/crates/cli/src/default.rs#L31.
+        AppConfig {
+            app_fri_params: FriParameters::standard_with_100_bits_conjectured_security(
+                DEFAULT_APP_LOG_BLOWUP,
+            )
+            .into(),
+            // By default it supports RISCV32IM with IO but no precompiles.
+            app_vm_config: SdkVmConfig::builder()
+                .system(Default::default())
+                .rv32i(Default::default())
+                .rv32m(Default::default())
+                .io(Default::default())
+                .build(),
+            leaf_fri_params: FriParameters::standard_with_100_bits_conjectured_security(
+                DEFAULT_LEAF_LOG_BLOWUP,
+            )
+            .into(),
+            compiler_options: Default::default(),
+        }
+    };
+
+    info!("Openvm program compiled OK - {} bytes", elf.len());
+
+    Ok(OpenVMProgram { elf, app_config })
+}
+
+#[cfg(test)]
+mod tests {
+    use crate::compile_stock_rust::compile_openvm_program_stock_rust;
+    use test_utils::host::testing_guest_directory;
+
+    #[test]
+    fn test_stock_compiler_impl() {
+        let guest_directory = testing_guest_directory("openvm", "stock_nightly_no_std");
+        let result = compile_openvm_program_stock_rust(&guest_directory, &"nightly".to_string());
+        assert!(result.is_ok(), "Openvm guest program compilation failure.");
+        assert!(
+            !result.unwrap().elf.is_empty(),
+            "ELF bytes should not be empty."
+        );
+    }
+}

crates/ere-openvm/src/error.rs

@@ -34,6 +34,12 @@ pub enum OpenVMError {
 pub enum CompileError {
     #[error("Failed to build guest, code: {0}")]
     BuildFailed(i32),
+    #[error("`openvm` build failure for {crate_path} failed: {source}")]
+    BuildFailure {
+        #[source]
+        source: anyhow::Error,
+        crate_path: PathBuf,
+    },
     #[error("Guest building skipped (OPENVM_SKIP_BUILD is set)")]
     BuildSkipped,
     #[error("Missing to find unique elf: {0}")]
@@ -44,6 +50,16 @@ pub enum CompileError {
     ReadConfigFailed { source: io::Error, path: PathBuf },
     #[error("Failed to deserialize OpenVM's config file: {0}")]
     DeserializeConfigFailed(toml::de::Error),
+    #[error("`cargo metadata` failed: {0}")]
+    MetadataCommand(#[from] cargo_metadata::Error),
+    #[error("Could not find `[package].name` in guest Cargo.toml at {path}")]
+    MissingPackageName { path: PathBuf },
+    #[error("Failed to read file at {path}: {source}")]
+    ReadFile {
+        path: PathBuf,
+        #[source]
+        source: std::io::Error,
+    },
 }
 
 #[derive(Debug, Error)]

crates/ere-openvm/src/lib.rs

@@ -1,5 +1,6 @@
 #![cfg_attr(not(test), warn(unused_crate_dependencies))]
 
+use crate::compile_stock_rust::compile_openvm_program_stock_rust;
 use crate::error::{CommonError, CompileError, ExecuteError, OpenVMError, ProveError, VerifyError};
 use openvm_build::GuestOptions;
 use openvm_circuit::arch::instructions::exe::VmExe;
@@ -27,6 +28,8 @@ use zkvm_interface::{
     ProverResourceType, PublicValues, zkVM, zkVMError,
 };
 
+mod compile_stock_rust;
+
 include!(concat!(env!("OUT_DIR"), "/name_and_sdk_version.rs"));
 mod error;
 
@@ -44,58 +47,69 @@ impl Compiler for OPENVM_TARGET {
 
     type Program = OpenVMProgram;
 
-    // Inlining `openvm_sdk::Sdk::build` in order to get raw elf bytes.
     fn compile(&self, guest_directory: &Path) -> Result<Self::Program, Self::Error> {
-        let pkg = openvm_build::get_package(guest_directory);
-        let guest_opts = GuestOptions::default().with_profile("release".to_string());
-        let target_dir = match openvm_build::build_guest_package(&pkg, &guest_opts, None, &None) {
-            Ok(target_dir) => target_dir,
-            Err(Some(code)) => return Err(CompileError::BuildFailed(code))?,
-            Err(None) => return Err(CompileError::BuildSkipped)?,
-        };
-
-        let elf_path = openvm_build::find_unique_executable(guest_directory, target_dir, &None)
-            .map_err(|e| CompileError::UniqueElfNotFound(e.into()))?;
-        let elf = fs::read(&elf_path).map_err(|source| CompileError::ReadElfFailed {
-            source,
-            path: elf_path.to_path_buf(),
-        })?;
-
-        let app_config_path = guest_directory.join("openvm.toml");
-        let app_config = if app_config_path.exists() {
-            let toml = fs::read_to_string(&app_config_path).map_err(|source| {
-                CompileError::ReadConfigFailed {
-                    source,
-                    path: app_config_path.to_path_buf(),
-                }
-            })?;
-            toml::from_str(&toml).map_err(CompileError::DeserializeConfigFailed)?
-        } else {
-            // The default `AppConfig` copied from https://github.com/openvm-org/openvm/blob/ca36de3/crates/cli/src/default.rs#L31.
-            AppConfig {
-                app_fri_params: FriParameters::standard_with_100_bits_conjectured_security(
-                    DEFAULT_APP_LOG_BLOWUP,
-                )
-                .into(),
-                // By default it supports RISCV32IM with IO but no precompiles.
-                app_vm_config: SdkVmConfig::builder()
-                    .system(Default::default())
-                    .rv32i(Default::default())
-                    .rv32m(Default::default())
-                    .io(Default::default())
-                    .build(),
-                leaf_fri_params: FriParameters::standard_with_100_bits_conjectured_security(
-                    DEFAULT_LEAF_LOG_BLOWUP,
-                )
-                .into(),
-                compiler_options: Default::default(),
-            }
-        };
-
-        Ok(OpenVMProgram { elf, app_config })
+        let toolchain = env::var("ERE_GUEST_TOOLCHAIN").unwrap_or_else(|_error| "openvm".into());
+        match toolchain.as_str() {
+            "openvm" => Ok(compile_openvm_program(guest_directory)?),
+            _ => Ok(compile_openvm_program_stock_rust(
+                guest_directory,
+                &toolchain,
+            )?),
+        }
     }
 }
 
+// Inlining `openvm_sdk::Sdk::build` in order to get raw elf bytes.
+fn compile_openvm_program(guest_directory: &Path) -> Result<OpenVMProgram, OpenVMError> {
+    let pkg = openvm_build::get_package(guest_directory);
+    let guest_opts = GuestOptions::default().with_profile("release".to_string());
+    let target_dir = match openvm_build::build_guest_package(&pkg, &guest_opts, None, &None) {
+        Ok(target_dir) => target_dir,
+        Err(Some(code)) => return Err(CompileError::BuildFailed(code))?,
+        Err(None) => return Err(CompileError::BuildSkipped)?,
+    };
+
+    let elf_path = openvm_build::find_unique_executable(guest_directory, target_dir, &None)
+        .map_err(|e| CompileError::UniqueElfNotFound(e.into()))?;
+    let elf = fs::read(&elf_path).map_err(|source| CompileError::ReadElfFailed {
+        source,
+        path: elf_path.to_path_buf(),
+    })?;
+
+    let app_config_path = guest_directory.join("openvm.toml");
+    let app_config = if app_config_path.exists() {
+        let toml = fs::read_to_string(&app_config_path).map_err(|source| {
+            CompileError::ReadConfigFailed {
+                source,
+                path: app_config_path.to_path_buf(),
+            }
+        })?;
+        toml::from_str(&toml).map_err(CompileError::DeserializeConfigFailed)?
+    } else {
+        // The default `AppConfig` copied from https://github.com/openvm-org/openvm/blob/ca36de3/crates/cli/src/default.rs#L31.
+        AppConfig {
+            app_fri_params: FriParameters::standard_with_100_bits_conjectured_security(
+                DEFAULT_APP_LOG_BLOWUP,
+            )
+            .into(),
+            // By default it supports RISCV32IM with IO but no precompiles.
+            app_vm_config: SdkVmConfig::builder()
+                .system(Default::default())
+                .rv32i(Default::default())
+                .rv32m(Default::default())
+                .io(Default::default())
+                .build(),
+            leaf_fri_params: FriParameters::standard_with_100_bits_conjectured_security(
+                DEFAULT_LEAF_LOG_BLOWUP,
+            )
+            .into(),
+            compiler_options: Default::default(),
+        }
+    };
+
+    Ok(OpenVMProgram { elf, app_config })
+}
+
 pub struct EreOpenVM {
     app_config: AppConfig<SdkVmConfig>,
     app_exe: Arc<VmExe<F>>,
@@ -266,6 +280,7 @@ pub fn agg_pk_path() -> PathBuf {
 #[cfg(test)]
 mod tests {
     use super::*;
+    use crate::compile_stock_rust::compile_openvm_program_stock_rust;
     use std::sync::OnceLock;
     use test_utils::host::{
         BasicProgramIo, run_zkvm_execute, run_zkvm_prove, testing_guest_directory,
@@ -297,6 +312,17 @@ mod tests {
         run_zkvm_execute(&zkvm, &io);
     }
 
+    #[test]
+    fn test_execute_nightly() {
+        let guest_directory = testing_guest_directory("openvm", "stock_nightly_no_std");
+        let program =
+            compile_openvm_program_stock_rust(&guest_directory, &"nightly".to_string()).unwrap();
+        let zkvm = EreOpenVM::new(program, ProverResourceType::Cpu).unwrap();
+
+        let result = zkvm.execute(&BasicProgramIo::empty());
+        assert!(result.is_ok(), "Openvm execution failure");
+    }
+
     #[test]
     fn test_execute_invalid_inputs() {
         let program = basic_program();

docker/openvm/Dockerfile

@@ -9,6 +9,9 @@ FROM ${BASE_IMAGE_TAG}
 COPY scripts/sdk_installers/install_openvm_sdk.sh /tmp/install_openvm_sdk.sh
 RUN chmod +x /tmp/install_openvm_sdk.sh
 
+# Add `rust-src` component to enable std build for nightly rust.
+RUN rustup +nightly component add rust-src
+
 # Run the OpenVM SDK installation script.
 # This script installs a specific toolchain
 # and installs cargo-openvm.

tests/openvm/stock_nightly_no_std/Cargo.toml

@@ -0,0 +1,7 @@
+[package]
+name = "addition_no_std"
+edition = "2021"
+
+[dependencies]
+
+[workspace]

tests/openvm/stock_nightly_no_std/src/main.rs

@@ -0,0 +1,25 @@
+#![no_std]
+#![no_main]
+extern crate alloc;
+
+use alloc::vec::Vec;
+use core::sync::atomic::Ordering;
+use core::sync::atomic::AtomicU16;
+mod openvm_rt;
+
+fn main() {
+    let a: AtomicU16 = core::hint::black_box(AtomicU16::new(5));
+    let b: AtomicU16 = core::hint::black_box(AtomicU16::new(7));
+
+    if a.load(Ordering::SeqCst) + b.load(Ordering::SeqCst) != 12 {
+        panic!("Something went wrong!");
+    }
+
+    let mut v: Vec<AtomicU16> = Vec::new();
+    v.push(AtomicU16::new(5));
+    v.push(AtomicU16::new(7));
+
+    if v[0].load(Ordering::SeqCst) + v[1].load(Ordering::SeqCst) != 12 {
+        panic!("Something went wrong!");
+    }
+}

tests/openvm/stock_nightly_no_std/src/openvm_rt.rs

@@ -0,0 +1,136 @@
+use core::alloc::{GlobalAlloc, Layout};
+// Import user `main` function
+use crate::main;
+
+// 1. Init global pointer (GP). It's used to optimize jumps by linker. Linker can change jumping from PC(Program Counter) based to GP based.
+// 2. Init stack pointer to the value STACK_TOP. It's stored in sp register.
+// 3. Call __start function defined below.
+// `__global_pointer$` is set by the linker. Its value depends on linker optimization. https://www.programmersought.com/article/77722901592/
+core::arch::global_asm!(
+    r#"
+.section .text._start;
+.globl _start;
+_start:
+    .option push;
+    .option norelax;
+    la gp, __global_pointer$;
+    .option pop;
+    la sp, {0}
+    lw sp, 0(sp)
+    call __start;
+"#,
+    sym STACK_TOP
+);
+
+static STACK_TOP: u32 = 0x0020_0400;
+
+// 1. Call `main` user function
+// 2. Call system halt environment function. It's defined by sp1 vm.
+#[unsafe(no_mangle)]
+fn __start(_argc: isize, _argv: *const *const u8) -> isize {
+    main();
+
+    terminate();
+
+    unreachable!()
+}
+
+// According to https://github.com/openvm-org/openvm/blob/v1.4.0-rc.8/crates/toolchain/openvm/src/process.rs
+#[inline(always)]
+fn terminate() {
+    unsafe {
+        core::arch::asm!(
+            ".insn 4, 0x000b"
+        )
+    }
+}
+
+// According to https://github.com/openvm-org/openvm/blob/v1.4.0-rc.8/crates/toolchain/openvm/src/process.rs
+#[panic_handler]
+fn panic_impl(_panic_info: &core::panic::PanicInfo) -> ! {
+    unsafe {
+        core::arch::asm!(
+            ".insn 4, 0x10000b"
+        )
+    }
+
+    unreachable!()
+}
+
+/// A simple heap allocator.
+///
+/// Allocates memory from left to right, without any deallocation.
+struct SimpleAlloc;
+
+unsafe impl GlobalAlloc for SimpleAlloc {
+    unsafe fn alloc(&self, layout: Layout) -> *mut u8 {
+        unsafe {
+            sys_alloc_aligned(layout.size(), layout.align())
+        }
+    }
+
+    unsafe fn dealloc(&self, _: *mut u8, _: Layout) {}
+}
+
+#[global_allocator]
+static HEAP: SimpleAlloc = SimpleAlloc;
+
+pub const MAX_MEMORY: usize = 0x78000000;
+static mut HEAP_POS: usize = 0;
+#[allow(clippy::missing_safety_doc)]
+#[unsafe(no_mangle)]
+pub unsafe extern "C" fn sys_alloc_aligned(bytes: usize, align: usize) -> *mut u8 {
+    unsafe extern "C" {
+        // https://lld.llvm.org/ELF/linker_script.html#sections-command
+        // `_end` is the last global variable defined by the linker. Its address is the beginning of heap data.
+        unsafe static _end: u8;
+    }
+
+    // SAFETY: Single threaded, so nothing else can touch this while we're working.
+    let mut heap_pos = unsafe { HEAP_POS };
+
+    if heap_pos == 0 {
+        heap_pos = unsafe { (&_end) as *const u8 as usize };
+    }
+
+    let offset = heap_pos & (align - 1);
+    if offset != 0 {
+        heap_pos += align - offset;
+    }
+
+    let ptr = heap_pos as *mut u8;
+    let (heap_pos, overflowed) = heap_pos.overflowing_add(bytes);
+
+    if overflowed || MAX_MEMORY < heap_pos {
+        panic!("Memory limit exceeded (0x78000000)");
+    }
+
+    unsafe { HEAP_POS = heap_pos };
+    ptr
+}
+
+// Assume single-threaded.
+#[cfg(all(target_arch = "riscv32", target_feature = "a"))]
+#[unsafe(no_mangle)]
+fn _critical_section_1_0_acquire() -> u32
+{
+    return 0;
+}
+
+#[cfg(all(target_arch = "riscv32", target_feature = "a"))]
+#[unsafe(no_mangle)]
+fn _critical_section_1_0_release(_: u32)
+{}
+
+// Assume single-threaded.
+#[cfg(all(target_arch = "riscv64", target_feature = "a"))]
+#[unsafe(no_mangle)]
+fn _critical_section_1_0_acquire() -> u64
+{
+    return 0;
+}
+
+#[cfg(all(target_arch = "riscv64", target_feature = "a"))]
+#[unsafe(no_mangle)]
+fn _critical_section_1_0_release(_: u64)
+{}