x86/kprobes: Fix crash when probing CS CALL instructionsgrafted

When using eBPF to probe CS CALL instructions within a function,
a crash can be triggered.

The eBPF tool probes offset 257 of the __hrtimer_run_queues()
function:

<__hrtimer_run_queues+249>:  nopl   0x0(%rax,%rax,1)
<__hrtimer_run_queues+254>:  mov    %r14,%rdi
<__hrtimer_run_queues+257>:  cs call <__x86_indirect_thunk_r12>
<__hrtimer_run_queues+263>:  mov    %eax,%r12d
<__hrtimer_run_queues+266>:  xchg   %ax,%ax
<__hrtimer_run_queues+268>:  mov    %r13,%rdi

Which triggers this crash:

  BUG: unable to handle page fault for address: 00000000000f41c9
  #PF: supervisor write access in kernel mode
  #PF: error_code(0x0002) - not-present page
  PGD 0 P4D 0
  Oops: 0002 [#1] SMP NOPTI
  CPU: 1 PID: 0 Comm: swapper/1 Kdump: loaded Tainted: P
  RIP: 0010:__hrtimer_run_queues+0x106/0x230

Note that __hrtimer_run_queues+0x106 is __hrtimer_run_queues+262, which is
at the 6th byte of the above CS CALL instruction. Since the CS CALL
instruction occupies 6 bytes, the exception occurred in the middle of that
call instruction.

The root cause is that when using eBPF tools to probe in the middle of a
function, a kprobe with INT3 is used as the underlying implementation.

During single-step emulation of the original CALL instruction,
int3_emulate_call() assumes that the probed CALL instruction is 5 bytes
long. However, the actual CS-prefixed CALL instruction occupies 6 bytes,
so it constructs an incorrect exception return address. When the CPU
returns from the kprobe handler, the next instruction to be executed is at
the address of the last byte of that CS CALL instruction. Coincidentally,
starting from that address, the CPU fetches and decodes a completely
different instruction, which ultimately triggers a kernel crash.

Fix the issue by using the actual instruction length obtained from
the instruction decoder when constructing the exception return
address, rather than relying on the hardcoded CALL_INSN_SIZE macro.

[ mingo: Refined the changelog ]

Fixes: 6256e668b7af ("x86/kprobes: Use int3 instead of debug trap for single-step")
Suggested-by: Masami Hiramatsu (Google) <mhiramat@kernel.org>
Signed-off-by: Jinke Han <jinkehan@didiglobal.com>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Reviewed-by: Masami Hiramatsu (Google) <mhiramat@kernel.org>
Acked-by: Yafang Shao <laoar.shao@gmail.com>
Acked-by: Borislav Petkov <bp@alien8.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Link: https://patch.msgid.link/20260908073742.GA10517@didi-ThinkCentre-M920t-N000

6 files changed, 763 insertions, 0 deletions

diff --git a/rust/pin-init/examples/big_struct_in_place.rs b/rust/pin-init/examples/big_struct_in_place.rs
new file mode 100644
index 000000000..c05139927
--- /dev/null
+++ b/rust/pin-init/examples/big_struct_in_place.rs
@@ -0,0 +1,43 @@
+// SPDX-License-Identifier: Apache-2.0 OR MIT
+
+use pin_init::*;
+
+// Struct with size over 1GiB
+#[derive(Debug)]
+#[allow(dead_code)]
+pub struct BigStruct {
+    buf: [u8; 1024 * 1024 * 1024],
+    a: u64,
+    b: u64,
+    c: u64,
+    d: u64,
+    managed_buf: ManagedBuf,
+}
+
+#[derive(Debug)]
+pub struct ManagedBuf {
+    buf: [u8; 1024 * 1024],
+}
+
+impl ManagedBuf {
+    pub fn new() -> impl Init<Self> {
+        init!(ManagedBuf { buf <- init_zeroed() })
+    }
+}
+
+fn main() {
+    #[cfg(any(feature = "std", feature = "alloc"))]
+    {
+        // we want to initialize the struct in-place, otherwise we would get a stackoverflow
+        let buf: Box<BigStruct> = Box::init(init!(BigStruct {
+            buf <- init_zeroed(),
+            a: 7,
+            b: 186,
+            c: 7789,
+            d: 34,
+            managed_buf <- ManagedBuf::new(),
+        }))
+        .unwrap();
+        println!("{}", core::mem::size_of_val(&*buf));
+    }
+}
diff --git a/rust/pin-init/examples/error.rs b/rust/pin-init/examples/error.rs
new file mode 100644
index 000000000..96f095398
--- /dev/null
+++ b/rust/pin-init/examples/error.rs
@@ -0,0 +1,31 @@
+// SPDX-License-Identifier: Apache-2.0 OR MIT
+
+#![cfg_attr(feature = "alloc", feature(allocator_api))]
+
+use core::convert::Infallible;
+
+#[cfg(feature = "alloc")]
+use std::alloc::AllocError;
+
+#[derive(Debug)]
+pub struct Error;
+
+impl From<Infallible> for Error {
+    #[inline]
+    fn from(e: Infallible) -> Self {
+        match e {}
+    }
+}
+
+#[cfg(feature = "alloc")]
+impl From<AllocError> for Error {
+    #[inline]
+    fn from(_: AllocError) -> Self {
+        Self
+    }
+}
+
+#[allow(dead_code)]
+fn main() {
+    let _ = Error;
+}
diff --git a/rust/pin-init/examples/linked_list.rs b/rust/pin-init/examples/linked_list.rs
new file mode 100644
index 000000000..424585fe2
--- /dev/null
+++ b/rust/pin-init/examples/linked_list.rs
@@ -0,0 +1,168 @@
+// SPDX-License-Identifier: Apache-2.0 OR MIT
+
+#![allow(clippy::undocumented_unsafe_blocks)]
+#![cfg_attr(feature = "alloc", feature(allocator_api))]
+
+use core::{
+    cell::Cell,
+    marker::PhantomPinned,
+    pin::Pin,
+    ptr::{self, NonNull},
+};
+
+use pin_init::*;
+
+#[allow(unused_attributes)]
+mod error;
+#[allow(unused_imports)]
+use error::Error;
+
+#[pin_data(PinnedDrop)]
+#[repr(C)]
+#[derive(Debug)]
+pub struct ListHead {
+    next: Link,
+    prev: Link,
+    #[pin]
+    pin: PhantomPinned,
+}
+
+impl ListHead {
+    #[inline]
+    pub fn new() -> impl PinInit<Self> {
+        pin_init!(&this in Self {
+            next: unsafe { Link::new_unchecked(this) },
+            prev: unsafe { Link::new_unchecked(this) },
+            pin: PhantomPinned,
+        })
+    }
+
+    #[inline]
+    #[allow(dead_code)]
+    pub fn insert_next(list: &ListHead) -> impl PinInit<Self> + '_ {
+        pin_init!(&this in Self {
+            prev: list.next.prev().replace(unsafe { Link::new_unchecked(this)}),
+            next: list.next.replace(unsafe { Link::new_unchecked(this)}),
+            pin: PhantomPinned,
+        })
+    }
+
+    #[inline]
+    pub fn insert_prev(list: &ListHead) -> impl PinInit<Self> + '_ {
+        pin_init!(&this in Self {
+            next: list.prev.next().replace(unsafe { Link::new_unchecked(this)}),
+            prev: list.prev.replace(unsafe { Link::new_unchecked(this)}),
+            pin: PhantomPinned,
+        })
+    }
+
+    #[inline]
+    pub fn next(&self) -> Option<NonNull<Self>> {
+        if ptr::eq(self.next.as_ptr(), self) {
+            None
+        } else {
+            Some(unsafe { NonNull::new_unchecked(self.next.as_ptr() as *mut Self) })
+        }
+    }
+
+    #[allow(dead_code)]
+    pub fn size(&self) -> usize {
+        let mut size = 1;
+        let mut cur = self.next.clone();
+        while !ptr::eq(self, cur.cur()) {
+            cur = cur.next().clone();
+            size += 1;
+        }
+        size
+    }
+}
+
+#[pinned_drop]
+impl PinnedDrop for ListHead {
+    //#[inline]
+    fn drop(self: Pin<&mut Self>) {
+        if !ptr::eq(self.next.as_ptr(), &*self) {
+            let next = unsafe { &*self.next.as_ptr() };
+            let prev = unsafe { &*self.prev.as_ptr() };
+            next.prev.set(&self.prev);
+            prev.next.set(&self.next);
+        }
+    }
+}
+
+#[repr(transparent)]
+#[derive(Clone, Debug)]
+struct Link(Cell<NonNull<ListHead>>);
+
+impl Link {
+    /// # Safety
+    ///
+    /// The contents of the pointer should form a consistent circular
+    /// linked list; for example, a "next" link should be pointed back
+    /// by the target `ListHead`'s "prev" link and a "prev" link should be
+    /// pointed back by the target `ListHead`'s "next" link.
+    #[inline]
+    unsafe fn new_unchecked(ptr: NonNull<ListHead>) -> Self {
+        Self(Cell::new(ptr))
+    }
+
+    #[inline]
+    fn next(&self) -> &Link {
+        unsafe { &(*self.0.get().as_ptr()).next }
+    }
+
+    #[inline]
+    #[allow(dead_code)]
+    fn prev(&self) -> &Link {
+        unsafe { &(*self.0.get().as_ptr()).prev }
+    }
+
+    #[allow(dead_code)]
+    fn cur(&self) -> &ListHead {
+        unsafe { &*self.0.get().as_ptr() }
+    }
+
+    #[inline]
+    fn replace(&self, other: Link) -> Link {
+        unsafe { Link::new_unchecked(self.0.replace(other.0.get())) }
+    }
+
+    #[inline]
+    fn as_ptr(&self) -> *const ListHead {
+        self.0.get().as_ptr()
+    }
+
+    #[inline]
+    fn set(&self, val: &Link) {
+        self.0.set(val.0.get());
+    }
+}
+
+#[allow(dead_code)]
+#[cfg(not(any(feature = "std", feature = "alloc")))]
+fn main() {}
+
+#[allow(dead_code)]
+#[cfg_attr(test, test)]
+#[cfg(any(feature = "std", feature = "alloc"))]
+fn main() -> Result<(), Error> {
+    let a = Box::pin_init(ListHead::new())?;
+    stack_pin_init!(let b = ListHead::insert_next(&a));
+    stack_pin_init!(let c = ListHead::insert_next(&a));
+    stack_pin_init!(let d = ListHead::insert_next(&b));
+    let e = Box::pin_init(ListHead::insert_next(&b))?;
+    println!("a ({a:p}): {a:?}");
+    println!("b ({b:p}): {b:?}");
+    println!("c ({c:p}): {c:?}");
+    println!("d ({d:p}): {d:?}");
+    println!("e ({e:p}): {e:?}");
+    let mut inspect = &*a;
+    while let Some(next) = inspect.next() {
+        println!("({inspect:p}): {inspect:?}");
+        inspect = unsafe { &*next.as_ptr() };
+        if core::ptr::eq(inspect, &*a) {
+            break;
+        }
+    }
+    Ok(())
+}
diff --git a/rust/pin-init/examples/mutex.rs b/rust/pin-init/examples/mutex.rs
new file mode 100644
index 000000000..e8d4dbb66
--- /dev/null
+++ b/rust/pin-init/examples/mutex.rs
@@ -0,0 +1,217 @@
+// SPDX-License-Identifier: Apache-2.0 OR MIT
+
+#![allow(clippy::undocumented_unsafe_blocks)]
+#![cfg_attr(feature = "alloc", feature(allocator_api))]
+#![allow(clippy::missing_safety_doc)]
+
+use core::{
+    cell::{Cell, UnsafeCell},
+    marker::PhantomPinned,
+    ops::{Deref, DerefMut},
+    pin::Pin,
+    sync::atomic::{AtomicBool, Ordering},
+};
+#[cfg(feature = "std")]
+use std::{
+    sync::Arc,
+    thread::{self, sleep, Builder, Thread},
+    time::Duration,
+};
+
+use pin_init::*;
+#[allow(unused_attributes)]
+#[path = "./linked_list.rs"]
+pub mod linked_list;
+use linked_list::*;
+
+pub struct SpinLock {
+    inner: AtomicBool,
+}
+
+impl SpinLock {
+    #[inline]
+    pub fn acquire(&self) -> SpinLockGuard<'_> {
+        while self
+            .inner
+            .compare_exchange(false, true, Ordering::Acquire, Ordering::Relaxed)
+            .is_err()
+        {
+            #[cfg(feature = "std")]
+            while self.inner.load(Ordering::Relaxed) {
+                thread::yield_now();
+            }
+        }
+        SpinLockGuard(self)
+    }
+
+    #[inline]
+    #[allow(clippy::new_without_default)]
+    pub const fn new() -> Self {
+        Self {
+            inner: AtomicBool::new(false),
+        }
+    }
+}
+
+pub struct SpinLockGuard<'a>(&'a SpinLock);
+
+impl Drop for SpinLockGuard<'_> {
+    #[inline]
+    fn drop(&mut self) {
+        self.0.inner.store(false, Ordering::Release);
+    }
+}
+
+#[pin_data]
+pub struct CMutex<T> {
+    #[pin]
+    wait_list: ListHead,
+    spin_lock: SpinLock,
+    locked: Cell<bool>,
+    #[pin]
+    data: UnsafeCell<T>,
+}
+
+impl<T> CMutex<T> {
+    #[inline]
+    pub fn new(val: impl PinInit<T>) -> impl PinInit<Self> {
+        pin_init!(CMutex {
+            wait_list <- ListHead::new(),
+            spin_lock: SpinLock::new(),
+            locked: Cell::new(false),
+            data <- UnsafeCell::pin_init(val),
+        })
+    }
+
+    #[inline]
+    pub fn lock(&self) -> Pin<CMutexGuard<'_, T>> {
+        let mut sguard = self.spin_lock.acquire();
+        if self.locked.get() {
+            stack_pin_init!(let _wait_entry = WaitEntry::insert_new(&self.wait_list));
+            // println!("wait list length: {}", self.wait_list.size());
+            while self.locked.get() {
+                drop(sguard);
+                #[cfg(feature = "std")]
+                thread::park();
+                sguard = self.spin_lock.acquire();
+            }
+        }
+        self.locked.set(true);
+        unsafe {
+            Pin::new_unchecked(CMutexGuard {
+                mtx: self,
+                _pin: PhantomPinned,
+            })
+        }
+    }
+
+    #[allow(dead_code)]
+    pub fn get_data_mut(self: Pin<&mut Self>) -> &mut T {
+        // SAFETY: we have an exclusive reference and thus nobody has access to data.
+        unsafe { &mut *self.data.get() }
+    }
+}
+
+unsafe impl<T: Send> Send for CMutex<T> {}
+unsafe impl<T: Send> Sync for CMutex<T> {}
+
+pub struct CMutexGuard<'a, T> {
+    mtx: &'a CMutex<T>,
+    _pin: PhantomPinned,
+}
+
+impl<T> Drop for CMutexGuard<'_, T> {
+    #[inline]
+    fn drop(&mut self) {
+        let sguard = self.mtx.spin_lock.acquire();
+        self.mtx.locked.set(false);
+        if let Some(list_field) = self.mtx.wait_list.next() {
+            let _wait_entry = list_field.as_ptr().cast::<WaitEntry>();
+            #[cfg(feature = "std")]
+            unsafe {
+                (*_wait_entry).thread.unpark()
+            };
+        }
+        drop(sguard);
+    }
+}
+
+impl<T> Deref for CMutexGuard<'_, T> {
+    type Target = T;
+
+    #[inline]
+    fn deref(&self) -> &Self::Target {
+        unsafe { &*self.mtx.data.get() }
+    }
+}
+
+impl<T> DerefMut for CMutexGuard<'_, T> {
+    #[inline]
+    fn deref_mut(&mut self) -> &mut Self::Target {
+        unsafe { &mut *self.mtx.data.get() }
+    }
+}
+
+#[pin_data]
+#[repr(C)]
+struct WaitEntry {
+    #[pin]
+    wait_list: ListHead,
+    #[cfg(feature = "std")]
+    thread: Thread,
+}
+
+impl WaitEntry {
+    #[inline]
+    fn insert_new(list: &ListHead) -> impl PinInit<Self> + '_ {
+        #[cfg(feature = "std")]
+        {
+            pin_init!(Self {
+                thread: thread::current(),
+                wait_list <- ListHead::insert_prev(list),
+            })
+        }
+        #[cfg(not(feature = "std"))]
+        {
+            pin_init!(Self {
+                wait_list <- ListHead::insert_prev(list),
+            })
+        }
+    }
+}
+
+#[cfg_attr(test, test)]
+#[allow(dead_code)]
+fn main() {
+    #[cfg(feature = "std")]
+    {
+        let mtx: Pin<Arc<CMutex<usize>>> = Arc::pin_init(CMutex::new(0)).unwrap();
+        let mut handles = vec![];
+        let thread_count = 20;
+        let workload = if cfg!(miri) { 100 } else { 1_000 };
+        for i in 0..thread_count {
+            let mtx = mtx.clone();
+            handles.push(
+                Builder::new()
+                    .name(format!("worker #{i}"))
+                    .spawn(move || {
+                        for _ in 0..workload {
+                            *mtx.lock() += 1;
+                        }
+                        println!("{i} halfway");
+                        sleep(Duration::from_millis((i as u64) * 10));
+                        for _ in 0..workload {
+                            *mtx.lock() += 1;
+                        }
+                        println!("{i} finished");
+                    })
+                    .expect("should not fail"),
+            );
+        }
+        for h in handles {
+            h.join().expect("thread panicked");
+        }
+        println!("{:?}", *mtx.lock());
+        assert_eq!(*mtx.lock(), workload * thread_count * 2);
+    }
+}
diff --git a/rust/pin-init/examples/pthread_mutex.rs b/rust/pin-init/examples/pthread_mutex.rs
new file mode 100644
index 000000000..00f457e68
--- /dev/null
+++ b/rust/pin-init/examples/pthread_mutex.rs
@@ -0,0 +1,183 @@
+// SPDX-License-Identifier: Apache-2.0 OR MIT
+
+// inspired by <https://github.com/nbdd0121/pin-init/blob/trunk/examples/pthread_mutex.rs>
+#![allow(clippy::undocumented_unsafe_blocks)]
+#![cfg_attr(feature = "alloc", feature(allocator_api))]
+
+#[cfg(not(windows))]
+mod pthread_mtx {
+    #[cfg(feature = "alloc")]
+    use core::alloc::AllocError;
+    use core::{
+        cell::UnsafeCell,
+        marker::PhantomPinned,
+        mem::MaybeUninit,
+        ops::{Deref, DerefMut},
+        pin::Pin,
+    };
+    use pin_init::*;
+    use std::convert::Infallible;
+
+    #[pin_data(PinnedDrop)]
+    pub struct PThreadMutex<T> {
+        #[pin]
+        raw: UnsafeCell<libc::pthread_mutex_t>,
+        data: UnsafeCell<T>,
+        #[pin]
+        pin: PhantomPinned,
+    }
+
+    unsafe impl<T: Send> Send for PThreadMutex<T> {}
+    unsafe impl<T: Send> Sync for PThreadMutex<T> {}
+
+    #[pinned_drop]
+    impl<T> PinnedDrop for PThreadMutex<T> {
+        fn drop(self: Pin<&mut Self>) {
+            unsafe {
+                libc::pthread_mutex_destroy(self.raw.get());
+            }
+        }
+    }
+
+    #[derive(Debug)]
+    pub enum Error {
+        #[allow(dead_code)]
+        IO(std::io::Error),
+        #[allow(dead_code)]
+        Alloc,
+    }
+
+    impl From<Infallible> for Error {
+        fn from(e: Infallible) -> Self {
+            match e {}
+        }
+    }
+
+    #[cfg(feature = "alloc")]
+    impl From<AllocError> for Error {
+        fn from(_: AllocError) -> Self {
+            Self::Alloc
+        }
+    }
+
+    impl<T> PThreadMutex<T> {
+        #[allow(dead_code)]
+        pub fn new(data: T) -> impl PinInit<Self, Error> {
+            fn init_raw() -> impl PinInit<UnsafeCell<libc::pthread_mutex_t>, Error> {
+                let init = |slot: *mut UnsafeCell<libc::pthread_mutex_t>| {
+                    // we can cast, because `UnsafeCell` has the same layout as T.
+                    let slot: *mut libc::pthread_mutex_t = slot.cast();
+                    let mut attr = MaybeUninit::uninit();
+                    let attr = attr.as_mut_ptr();
+                    // SAFETY: ptr is valid
+                    let ret = unsafe { libc::pthread_mutexattr_init(attr) };
+                    if ret != 0 {
+                        return Err(Error::IO(std::io::Error::from_raw_os_error(ret)));
+                    }
+                    // SAFETY: attr is initialized
+                    let ret = unsafe {
+                        libc::pthread_mutexattr_settype(attr, libc::PTHREAD_MUTEX_NORMAL)
+                    };
+                    if ret != 0 {
+                        // SAFETY: attr is initialized
+                        unsafe { libc::pthread_mutexattr_destroy(attr) };
+                        return Err(Error::IO(std::io::Error::from_raw_os_error(ret)));
+                    }
+                    // SAFETY: slot is valid
+                    unsafe { slot.write(libc::PTHREAD_MUTEX_INITIALIZER) };
+                    // SAFETY: attr and slot are valid ptrs and attr is initialized
+                    let ret = unsafe { libc::pthread_mutex_init(slot, attr) };
+                    // SAFETY: attr was initialized
+                    unsafe { libc::pthread_mutexattr_destroy(attr) };
+                    if ret != 0 {
+                        return Err(Error::IO(std::io::Error::from_raw_os_error(ret)));
+                    }
+                    Ok(())
+                };
+                // SAFETY: mutex has been initialized
+                unsafe { pin_init_from_closure(init) }
+            }
+            pin_init!(Self {
+                data: UnsafeCell::new(data),
+                raw <- init_raw(),
+                pin: PhantomPinned,
+            }? Error)
+        }
+
+        #[allow(dead_code)]
+        pub fn lock(&self) -> PThreadMutexGuard<'_, T> {
+            // SAFETY: raw is always initialized
+            unsafe { libc::pthread_mutex_lock(self.raw.get()) };
+            PThreadMutexGuard { mtx: self }
+        }
+    }
+
+    pub struct PThreadMutexGuard<'a, T> {
+        mtx: &'a PThreadMutex<T>,
+    }
+
+    impl<T> Drop for PThreadMutexGuard<'_, T> {
+        fn drop(&mut self) {
+            // SAFETY: raw is always initialized
+            unsafe { libc::pthread_mutex_unlock(self.mtx.raw.get()) };
+        }
+    }
+
+    impl<T> Deref for PThreadMutexGuard<'_, T> {
+        type Target = T;
+
+        fn deref(&self) -> &Self::Target {
+            unsafe { &*self.mtx.data.get() }
+        }
+    }
+
+    impl<T> DerefMut for PThreadMutexGuard<'_, T> {
+        fn deref_mut(&mut self) -> &mut Self::Target {
+            unsafe { &mut *self.mtx.data.get() }
+        }
+    }
+}
+
+#[cfg_attr(test, test)]
+#[cfg_attr(all(test, miri), ignore)]
+fn main() {
+    #[cfg(all(any(feature = "std", feature = "alloc"), not(windows)))]
+    {
+        use core::pin::Pin;
+        use pin_init::*;
+        use pthread_mtx::*;
+        use std::{
+            sync::Arc,
+            thread::{sleep, Builder},
+            time::Duration,
+        };
+        let mtx: Pin<Arc<PThreadMutex<usize>>> = Arc::try_pin_init(PThreadMutex::new(0)).unwrap();
+        let mut handles = vec![];
+        let thread_count = 20;
+        let workload = 1_000_000;
+        for i in 0..thread_count {
+            let mtx = mtx.clone();
+            handles.push(
+                Builder::new()
+                    .name(format!("worker #{i}"))
+                    .spawn(move || {
+                        for _ in 0..workload {
+                            *mtx.lock() += 1;
+                        }
+                        println!("{i} halfway");
+                        sleep(Duration::from_millis((i as u64) * 10));
+                        for _ in 0..workload {
+                            *mtx.lock() += 1;
+                        }
+                        println!("{i} finished");
+                    })
+                    .expect("should not fail"),
+            );
+        }
+        for h in handles {
+            h.join().expect("thread panicked");
+        }
+        println!("{:?}", *mtx.lock());
+        assert_eq!(*mtx.lock(), workload * thread_count * 2);
+    }
+}
diff --git a/rust/pin-init/examples/static_init.rs b/rust/pin-init/examples/static_init.rs
new file mode 100644
index 000000000..8dd52313c
--- /dev/null
+++ b/rust/pin-init/examples/static_init.rs
@@ -0,0 +1,121 @@
+// SPDX-License-Identifier: Apache-2.0 OR MIT
+
+#![allow(clippy::undocumented_unsafe_blocks)]
+#![cfg_attr(feature = "alloc", feature(allocator_api))]
+#![allow(unused_imports)]
+
+use core::{
+    cell::{Cell, UnsafeCell},
+    mem::MaybeUninit,
+    ops,
+    pin::Pin,
+    time::Duration,
+};
+use pin_init::*;
+#[cfg(feature = "std")]
+use std::{
+    sync::Arc,
+    thread::{sleep, Builder},
+};
+
+#[allow(unused_attributes)]
+mod mutex;
+use mutex::*;
+
+pub struct StaticInit<T, I> {
+    cell: UnsafeCell<MaybeUninit<T>>,
+    init: Cell<Option<I>>,
+    lock: SpinLock,
+    present: Cell<bool>,
+}
+
+unsafe impl<T: Sync, I> Sync for StaticInit<T, I> {}
+unsafe impl<T: Send, I> Send for StaticInit<T, I> {}
+
+impl<T, I: PinInit<T>> StaticInit<T, I> {
+    pub const fn new(init: I) -> Self {
+        Self {
+            cell: UnsafeCell::new(MaybeUninit::uninit()),
+            init: Cell::new(Some(init)),
+            lock: SpinLock::new(),
+            present: Cell::new(false),
+        }
+    }
+}
+
+impl<T, I: PinInit<T>> ops::Deref for StaticInit<T, I> {
+    type Target = T;
+    fn deref(&self) -> &Self::Target {
+        if self.present.get() {
+            unsafe { (*self.cell.get()).assume_init_ref() }
+        } else {
+            println!("acquire spinlock on static init");
+            let _guard = self.lock.acquire();
+            println!("rechecking present...");
+            std::thread::sleep(std::time::Duration::from_millis(200));
+            if self.present.get() {
+                return unsafe { (*self.cell.get()).assume_init_ref() };
+            }
+            println!("doing init");
+            let ptr = self.cell.get().cast::<T>();
+            match self.init.take() {
+                Some(f) => unsafe { pin_init::raw_init(ptr, f) },
+                None => unsafe { core::hint::unreachable_unchecked() },
+            }
+            self.present.set(true);
+            unsafe { (*self.cell.get()).assume_init_ref() }
+        }
+    }
+}
+
+pub struct CountInit;
+
+unsafe impl PinInit<CMutex<usize>> for CountInit {
+    unsafe fn __init(self, slot: *mut CMutex<usize>) -> Result<(), core::convert::Infallible> {
+        let init = CMutex::new(0);
+        std::thread::sleep(std::time::Duration::from_millis(1000));
+        unsafe { pin_init::raw_init(slot, init) };
+        Ok(())
+    }
+}
+
+pub static COUNT: StaticInit<CMutex<usize>, CountInit> = StaticInit::new(CountInit);
+
+fn main() {
+    #[cfg(feature = "std")]
+    {
+        let mtx: Pin<Arc<CMutex<usize>>> = Arc::pin_init(CMutex::new(0)).unwrap();
+        let mut handles = vec![];
+        let thread_count = 20;
+        let workload = 1_000;
+        for i in 0..thread_count {
+            let mtx = mtx.clone();
+            handles.push(
+                Builder::new()
+                    .name(format!("worker #{i}"))
+                    .spawn(move || {
+                        for _ in 0..workload {
+                            *COUNT.lock() += 1;
+                            std::thread::sleep(std::time::Duration::from_millis(10));
+                            *mtx.lock() += 1;
+                            std::thread::sleep(std::time::Duration::from_millis(10));
+                            *COUNT.lock() += 1;
+                        }
+                        println!("{i} halfway");
+                        sleep(Duration::from_millis((i as u64) * 10));
+                        for _ in 0..workload {
+                            std::thread::sleep(std::time::Duration::from_millis(10));
+                            *mtx.lock() += 1;
+                        }
+                        println!("{i} finished");
+                    })
+                    .expect("should not fail"),
+            );
+        }
+        for h in handles {
+            h.join().expect("thread panicked");
+        }
+        println!("{:?}, {:?}", *mtx.lock(), *COUNT.lock());
+        assert_eq!(*mtx.lock(), workload * thread_count * 2);
+    }
+}