Implement volatile loads and stores

src/builder.rs

@@ -654,24 +654,12 @@ impl<'a, 'gcc, 'tcx> BuilderMethods<'a, 'tcx> for Builder<'a, 'gcc, 'tcx> {
 
     fn load(&mut self, _ty: Type<'gcc>, ptr: RValue<'gcc>, _align: Align) -> RValue<'gcc> {
         // TODO(antoyo): use ty.
-        let block = self.llbb();
-        let function = block.get_function();
-        // NOTE: instead of returning the dereference here, we have to assign it to a variable in
-        // the current basic block. Otherwise, it could be used in another basic block, causing a
-        // dereference after a drop, for instance.
-        // TODO(antoyo): handle align of the load instruction.
-        let deref = ptr.dereference(None).to_rvalue();
-        let value_type = deref.get_type();
-        unsafe { RETURN_VALUE_COUNT += 1 };
-        let loaded_value = function.new_local(None, value_type, &format!("loadedValue{}", unsafe { RETURN_VALUE_COUNT }));
-        block.add_assignment(None, loaded_value, deref);
-        loaded_value.to_rvalue()
+        // TODO: handle non-natural alignments
+        self.load_inner(ptr, false)
     }
 
     fn volatile_load(&mut self, _ty: Type<'gcc>, ptr: RValue<'gcc>) -> RValue<'gcc> {
-        // TODO(antoyo): use ty.
-        let ptr = self.context.new_cast(None, ptr, ptr.get_type().make_volatile());
-        ptr.dereference(None).to_rvalue()
+        self.load_inner(ptr, true)
     }
 
     fn atomic_load(&mut self, _ty: Type<'gcc>, ptr: RValue<'gcc>, order: AtomicOrdering, size: Size) -> RValue<'gcc> {
@@ -799,17 +787,32 @@ impl<'a, 'gcc, 'tcx> BuilderMethods<'a, 'tcx> for Builder<'a, 'gcc, 'tcx> {
         self.store_with_flags(val, ptr, align, MemFlags::empty())
     }
 
-    fn store_with_flags(&mut self, val: RValue<'gcc>, ptr: RValue<'gcc>, align: Align, _flags: MemFlags) -> RValue<'gcc> {
+    fn store_with_flags(&mut self, mut val: RValue<'gcc>, ptr: RValue<'gcc>, align: Align, flags: MemFlags) -> RValue<'gcc> {
         let ptr = self.check_store(val, ptr);
+
         let destination = ptr.dereference(None);
         // NOTE: libgccjit does not support specifying the alignment on the assignment, so we cast
         // to type so it gets the proper alignment.
         let destination_type = destination.to_rvalue().get_type().unqualified();
-        let aligned_type = destination_type.get_aligned(align.bytes()).make_pointer();
-        let aligned_destination = self.cx.context.new_bitcast(None, ptr, aligned_type);
-        let aligned_destination = aligned_destination.dereference(None);
-        self.llbb().add_assignment(None, aligned_destination, val);
-        // TODO(antoyo): handle align and flags.
+
+        let align = if flags.contains(MemFlags::UNALIGNED) { 1 } else { align.bytes() };
+        let mut modified_destination_type = destination_type.get_aligned(align);
+        if flags.contains(MemFlags::VOLATILE) {
+            modified_destination_type = modified_destination_type.make_volatile();
+        }
+
+        // FIXME: The type checking in `add_assignment` removes only one
+        // qualifier from each side. So the writes `int → volatile int` and
+        // `int → int __attribute__((aligned(1)))` are considered legal but
+        // `int → volatile int __attribute__((aligned(1)))` is not. This seems
+        // like a bug in libgccjit. The easiest way to work around this is to
+        // bitcast `val` to have the matching qualifiers.
+        val = self.cx.context.new_bitcast(None, val, modified_destination_type);
+
+        let modified_ptr = self.cx.context.new_bitcast(None, ptr, modified_destination_type.make_pointer());
+        let modified_destination = modified_ptr.dereference(None);
+        self.llbb().add_assignment(None, modified_destination, val);
+        // TODO: handle `MemFlags::NONTEMPORAL`.
         // NOTE: dummy value here since it's never used. FIXME(antoyo): API should not return a value here?
         self.cx.context.new_rvalue_zero(self.type_i32())
     }
@@ -1296,6 +1299,33 @@ impl<'a, 'gcc, 'tcx> BuilderMethods<'a, 'tcx> for Builder<'a, 'gcc, 'tcx> {
 }
 
 impl<'a, 'gcc, 'tcx> Builder<'a, 'gcc, 'tcx> {
+    fn load_inner(&mut self, ptr: RValue<'gcc>, is_volatile: bool) -> RValue<'gcc> {
+        let block = self.llbb();
+        let function = block.get_function();
+
+        // NOTE: instead of returning the dereference here, we have to assign it to a variable in
+        // the current basic block. Otherwise, it could be used in another basic block, causing a
+        // dereference after a drop, for instance.
+        let deref = ptr.dereference(None).to_rvalue();
+        let value_type = deref.get_type();
+
+        // If `is_volatile == true`, convert `ptr` to `volatile value_type *`
+        // and try dereference again. (libgccjit doesn't provide a way to get
+        // `value_type` directly, unfortunately.)
+        let deref = if is_volatile {
+            let new_ptr_type = value_type.make_volatile().make_pointer();
+            self.pointercast(ptr, new_ptr_type).dereference(None).to_rvalue()
+        }
+        else {
+            deref
+        };
+
+        unsafe { RETURN_VALUE_COUNT += 1 };
+        let loaded_value = function.new_local(None, value_type, &format!("loadedValue{}", unsafe { RETURN_VALUE_COUNT }));
+        block.add_assignment(None, loaded_value, deref);
+        loaded_value.to_rvalue()
+    }
+
     #[cfg(feature="master")]
     pub fn shuffle_vector(&mut self, v1: RValue<'gcc>, v2: RValue<'gcc>, mask: RValue<'gcc>) -> RValue<'gcc> {
         let struct_type = mask.get_type().is_struct().expect("mask of struct type");

tests/run/volatile.rs

@@ -0,0 +1,119 @@
+// Compiler:
+//
+// Run-time:
+//   status: 0
+
+#![no_std]
+#![feature(core_intrinsics, start)]
+
+#[panic_handler]
+fn panic_handler(_: &core::panic::PanicInfo) -> ! {
+    core::intrinsics::abort();
+}
+
+mod libc {
+    #[link(name = "c")]
+    extern "C" {
+        pub fn puts(s: *const u8) -> i32;
+
+        pub fn sigaction(signum: i32, act: *const sigaction, oldact: *mut sigaction) -> i32;
+        pub fn mmap(addr: *mut (), len: usize, prot: i32, flags: i32, fd: i32, offset: i64) -> *mut ();
+        pub fn mprotect(addr: *mut (), len: usize, prot: i32) -> i32;
+    }
+
+    pub const PROT_READ: i32 = 1;
+    pub const PROT_WRITE: i32 = 2;
+    pub const MAP_PRIVATE: i32 = 0x0002;
+    pub const MAP_ANONYMOUS: i32 = 0x0020;
+    pub const MAP_FAILED: *mut u8 = !0 as *mut u8;
+
+    /// glibc sigaction
+    #[repr(C)]
+    pub struct sigaction {
+        pub sa_sigaction: Option<unsafe extern "C" fn(i32, *mut (), *mut ())>,
+        pub sa_mask: [u32; 32],
+        pub sa_flags: i32,
+        pub sa_restorer: Option<unsafe extern "C" fn()>,
+    }
+
+    pub const SA_SIGINFO: i32 = 0x00000004;
+    pub const SIGSEGV: i32 = 11;
+}
+
+static mut COUNT: u32 = 0;
+static mut STORAGE: *mut u8 = core::ptr::null_mut();
+const PAGE_SIZE: usize = 1 << 15;
+
+#[start]
+fn main(_argc: isize, _argv: *const *const u8) -> isize {
+    unsafe {
+        // Register a segfault handler
+        libc::sigaction(
+            libc::SIGSEGV,
+            &libc::sigaction {
+                sa_sigaction: Some(segv_handler),
+                sa_flags: libc::SA_SIGINFO,
+                ..core::mem::zeroed()
+            },
+            core::ptr::null_mut(),
+        );
+
+        STORAGE = libc::mmap(
+            core::ptr::null_mut(),
+            PAGE_SIZE * 2,
+            0,
+            libc::MAP_PRIVATE | libc::MAP_ANONYMOUS,
+            -1,
+            0,
+        ).cast();
+        if STORAGE == libc::MAP_FAILED {
+            libc::puts(b"error: mmap failed\0".as_ptr());
+            return 1;
+        }
+
+        let p_count = (&mut COUNT) as *mut u32;
+        p_count.write_volatile(0);
+
+        // Trigger segfaults
+        STORAGE.add(0).write_volatile(1);
+        STORAGE.add(PAGE_SIZE).write_volatile(1);
+        STORAGE.add(0).write_volatile(1);
+        STORAGE.add(PAGE_SIZE).write_volatile(1);
+        STORAGE.add(0).write_volatile(1);
+        STORAGE.add(PAGE_SIZE).write_volatile(1);
+
+        // The segfault handler should have been called for every
+        // `write_volatile` in `STORAGE`. If the compiler ignores volatility,
+        // some of these writes will be combined, causing a different number of
+        // segfaults.
+        //
+        // This `p_count` read is done by a volatile read. If the compiler
+        // ignores volatility, the compiler will speculate that `*p_count` is
+        // unchanged and remove this check, failing the test.
+        if p_count.read_volatile() != 6 {
+            libc::puts(b"error: segfault count mismatch\0".as_ptr());
+            return 1;
+        }
+
+        0
+    }
+}
+
+unsafe extern "C" fn segv_handler(_: i32, _: *mut (), _: *mut ()) {
+    let p_count = (&mut COUNT) as *mut u32;
+    p_count.write_volatile(p_count.read_volatile() + 1);
+    let count = p_count.read_volatile();
+
+    // Toggle the protected page so that the handler will be called for
+    // each `write_volatile`
+    libc::mprotect(
+        STORAGE.cast(),
+        PAGE_SIZE,
+        if count % 2 == 1 { libc::PROT_READ | libc::PROT_WRITE } else { 0 },
+    );
+    libc::mprotect(
+        STORAGE.add(PAGE_SIZE).cast(),
+        PAGE_SIZE,
+        if count % 2 == 0 { libc::PROT_READ | libc::PROT_WRITE } else { 0 },
+    );
+}