diff --git a/.github/workflows/ci.yml b/.github/workflows/ci.yml
index fd3b294d..ab242f38 100644
--- a/.github/workflows/ci.yml
+++ b/.github/workflows/ci.yml
@@ -209,7 +209,7 @@ jobs:
           python -m pip install -U pip nox
           cargo xtask test-py
         env:
-          CARGO_TARGET_DIR: ${{ github.workspace }}
+          CARGO_TARGET_DIR: ${{ github.workspace }}/target
 
       - name: Test cross compilation
         if: ${{ matrix.platform.os == 'ubuntu-latest' && matrix.python-version == '3.9' }}
@@ -265,14 +265,10 @@ jobs:
           host=$(rustc -Vv | grep host | sed 's/host: //')
           curl -fsSL https://github.com/taiki-e/cargo-llvm-cov/releases/download/v${CARGO_LLVM_COV_VERSION}/cargo-llvm-cov-"$host".tar.gz | tar xzf - -C ~/.cargo/bin
         env:
-          CARGO_LLVM_COV_VERSION: 0.1.11
+          CARGO_LLVM_COV_VERSION: 0.1.15
+      - run: pip install -U pip nox
       - run: |
-          cargo llvm-cov clean --workspace
-          cargo llvm-cov --package $ALL_PACKAGES --no-report
-          cargo llvm-cov --package $ALL_PACKAGES --no-report --features abi3
-          cargo llvm-cov --package $ALL_PACKAGES --no-report --features full
-          cargo llvm-cov --package $ALL_PACKAGES --no-report --features "abi3 full"
-          cargo llvm-cov --package $ALL_PACKAGES --no-run --lcov --output-path coverage.lcov
+          cargo xtask coverage --output-lcov coverage.lcov
         shell: bash
         env:
           ALL_PACKAGES: pyo3 pyo3-build-config pyo3-macros-backend pyo3-macros
diff --git a/Cargo.toml b/Cargo.toml
index d88464fb..fc69989c 100644
--- a/Cargo.toml
+++ b/Cargo.toml
@@ -128,6 +128,8 @@ harness = false
 members = [
     "pyo3-macros",
     "pyo3-macros-backend",
+    "pytests/pyo3-benchmarks",
+    "pytests/pyo3-pytests",
     "examples",
     "xtask"
 ]
diff --git a/pytests/pyo3-benchmarks/Cargo.toml b/pytests/pyo3-benchmarks/Cargo.toml
index 70798099..f931e19a 100644
--- a/pytests/pyo3-benchmarks/Cargo.toml
+++ b/pytests/pyo3-benchmarks/Cargo.toml
@@ -14,5 +14,3 @@ features = ["extension-module"]
 [lib]
 name = "_pyo3_benchmarks"
 crate-type = ["cdylib"]
-
-[workspace]
\ No newline at end of file
diff --git a/pytests/pyo3-benchmarks/setup.py b/pytests/pyo3-benchmarks/setup.py
index b16f1ed5..f38bebdf 100644
--- a/pytests/pyo3-benchmarks/setup.py
+++ b/pytests/pyo3-benchmarks/setup.py
@@ -1,7 +1,8 @@
+import os
+
 from setuptools import setup
 from setuptools_rust import RustExtension
 
-
 setup(
     name="pyo3-benchmarks",
     version="0.1.0",
@@ -18,7 +19,8 @@ setup(
     rust_extensions=[
         RustExtension(
             "pyo3_benchmarks._pyo3_benchmarks",
-            debug=False,
+            # build debug when measuring coverage, otherwise release
+            debug="CARGO_LLVM_COV_TARGET_DIR" in os.environ,
         ),
     ],
     include_package_data=True,
diff --git a/pytests/pyo3-pytests/Cargo.toml b/pytests/pyo3-pytests/Cargo.toml
index 61f76891..8fdcda88 100644
--- a/pytests/pyo3-pytests/Cargo.toml
+++ b/pytests/pyo3-pytests/Cargo.toml
@@ -25,5 +25,3 @@ classifier=[
     "Operating System :: POSIX",
     "Operating System :: MacOS :: MacOS X",
 ]
-
-[workspace]
\ No newline at end of file
diff --git a/pytests/pyo3-pytests/noxfile.py b/pytests/pyo3-pytests/noxfile.py
index e6d1b300..17a6b80f 100644
--- a/pytests/pyo3-pytests/noxfile.py
+++ b/pytests/pyo3-pytests/noxfile.py
@@ -4,5 +4,6 @@ import nox
 @nox.session
 def python(session):
     session.install("-rrequirements-dev.txt")
-    session.install(".", "--no-build-isolation")
+    session.install("maturin")
+    session.run_always("maturin", "develop")
     session.run("pytest")
diff --git a/pytests/pyo3-pytests/pyproject.toml b/pytests/pyo3-pytests/pyproject.toml
index 9a2f56d9..5f0a84ee 100644
--- a/pytests/pyo3-pytests/pyproject.toml
+++ b/pytests/pyo3-pytests/pyproject.toml
@@ -1,3 +1,4 @@
 [build-system]
-requires = ["maturin>=0.10,<0.11"]
+# FIXME: branch is necessary for coverage to work
+requires = ["maturin @ git+https://github.com/davidhewitt/maturin@profile-arg"]
 build-backend = "maturin"
diff --git a/pytests/pyo3-pytests/requirements-dev.txt b/pytests/pyo3-pytests/requirements-dev.txt
index d70f6391..cc06de69 100644
--- a/pytests/pyo3-pytests/requirements-dev.txt
+++ b/pytests/pyo3-pytests/requirements-dev.txt
@@ -2,4 +2,3 @@ hypothesis>=3.55
 pytest>=3.5.0
 psutil>=5.6
 pip>=21.3
-maturin>=0.12,<0.13
diff --git a/xtask/src/main.rs b/xtask/src/main.rs
index e9fa71a2..2b919815 100644
--- a/xtask/src/main.rs
+++ b/xtask/src/main.rs
@@ -1,19 +1,26 @@
-use anyhow::{Context, Result};
+use anyhow::{ensure, Context, Result};
 use std::{collections::HashMap, process::Command};
 use structopt::StructOpt;
 
 #[derive(StructOpt)]
 enum Subcommand {
     /// Runs `cargo llvm-cov` for the PyO3 codebase.
-    Coverage,
+    Coverage(CoverageOpts),
     /// Runs tests in examples/ and pytests/
     TestPy,
 }
 
+#[derive(StructOpt)]
+struct CoverageOpts {
+    /// Creates an lcov output instead of printing to the terminal.
+    #[structopt(long)]
+    output_lcov: Option<String>,
+}
+
 impl Subcommand {
     fn execute(self) -> Result<()> {
         match self {
-            Subcommand::Coverage => subcommand_coverage(),
+            Subcommand::Coverage(opts) => subcommand_coverage(opts),
             Subcommand::TestPy => run_python_tests(None),
         }
     }
@@ -24,32 +31,84 @@ fn main() -> Result<()> {
 }
 
 /// Runs `cargo llvm-cov` for the PyO3 codebase.
-fn subcommand_coverage() -> Result<()> {
-    run(&mut llvm_cov_command(&["clean", "--workspace"]))?;
-    run(&mut llvm_cov_command(&["--no-report"]))?;
-
-    // FIXME: add various feature combinations using 'full' feature.
-    // run(&mut llvm_cov_command(&["--no-report"]))?;
-
-    // XXX: the following block doesn't work until https://github.com/taiki-e/cargo-llvm-cov/pull/115 is merged
+fn subcommand_coverage(opts: CoverageOpts) -> Result<()> {
     let env = get_coverage_env()?;
-    run_python_tests(&env)?;
-    // (after here works with stable llvm-cov)
 
-    // TODO: add an argument to make it possible to generate lcov report & use this in CI.
-    run(&mut llvm_cov_command(&["--no-run", "--summary-only"]))?;
+    run(llvm_cov_command(&["clean", "--workspace"]).envs(&env))?;
+
+    run(Command::new("cargo")
+        .args(&["test", "--manifest-path", "pyo3-build-config/Cargo.toml"])
+        .envs(&env))?;
+    run(Command::new("cargo")
+        .args(&["test", "--manifest-path", "pyo3-macros-backend/Cargo.toml"])
+        .envs(&env))?;
+    run(Command::new("cargo")
+        .args(&["test", "--manifest-path", "pyo3-macros/Cargo.toml"])
+        .envs(&env))?;
+
+    run(Command::new("cargo").arg("test").envs(&env))?;
+    run(Command::new("cargo")
+        .args(&["test", "--features", "abi3"])
+        .envs(&env))?;
+    run(Command::new("cargo")
+        .args(&["test", "--features", "full"])
+        .envs(&env))?;
+    run(Command::new("cargo")
+        .args(&["test", "--features", "abi3 full"])
+        .envs(&env))?;
+
+    run_python_tests(&env)?;
+
+    match opts.output_lcov {
+        Some(path) => {
+            run(llvm_cov_command(&["--no-run", "--lcov", "--output-path", &path]).envs(&env))?
+        }
+        None => run(llvm_cov_command(&["--no-run", "--summary-only"]).envs(&env))?,
+    }
+
     Ok(())
 }
 
 fn run(command: &mut Command) -> Result<()> {
     println!("running: {}", format_command(command));
-    command.spawn()?.wait()?;
+    let status = command.spawn()?.wait()?;
+    ensure! {
+        status.success(),
+        "process did not run successfully ({exit}): {command}",
+        exit = match status.code() {
+            Some(code) => format!("exit code {}", code),
+            None => "terminated by signal".into(),
+        },
+        command = format_command(command),
+    };
     Ok(())
 }
 
+fn get_output(command: &mut Command) -> Result<std::process::Output> {
+    let output = command.output()?;
+    ensure! {
+        output.status.success(),
+        "process did not run successfully ({exit}): {command}",
+        exit = match output.status.code() {
+            Some(code) => format!("exit code {}", code),
+            None => "terminated by signal".into(),
+        },
+        command = format_command(command),
+    };
+    Ok(output)
+}
+
 fn llvm_cov_command(args: &[&str]) -> Command {
     let mut command = Command::new("cargo");
-    command.args(&["llvm-cov", "--package=pyo3"]).args(args);
+    command
+        .args(&[
+            "llvm-cov",
+            "--package=pyo3",
+            "--package=pyo3-build-config",
+            "--package=pyo3-macros-backend",
+            "--package=pyo3-macros",
+        ])
+        .args(args);
     command
 }
 
@@ -85,11 +144,25 @@ fn get_coverage_env() -> Result<HashMap<String, String>> {
     let output = String::from_utf8(llvm_cov_command(&["show-env"]).output()?.stdout)?;
 
     for line in output.trim().split('\n') {
-        let (key, value) = split_once(line, '=').context("expected '=' in each output line")?;
+        let (key, value) = split_once(line, '=')
+            .context("expected '=' in each line of output from llvm-cov show-env")?;
         env.insert(key.to_owned(), value.trim_matches('"').to_owned());
     }
 
-    env.insert("RUSTUP_TOOLCHAIN".to_owned(), "nightly".to_owned());
+    // Ensure that examples/ and pytests/ all build to the correct target directory to collect
+    // coverage artifacts.
+    env.insert(
+        "CARGO_TARGET_DIR".to_owned(),
+        env.get("CARGO_LLVM_COV_TARGET_DIR").unwrap().to_owned(),
+    );
+
+    // Coverage only works on nightly.
+    let rustc_version =
+        String::from_utf8(get_output(Command::new("rustc").arg("--version"))?.stdout)
+            .context("failed to parse rust version as utf8")?;
+    if !rustc_version.contains("nightly") {
+        env.insert("RUSTUP_TOOLCHAIN".to_owned(), "nightly".to_owned());
+    }
 
     Ok(env)
 }