Cargo.lock

@@ -220,7 +220,7 @@ name = "hbxrt"
 version = "0.1.0"
 dependencies = [
  "hbvm",
- "memmap2",
+ "nix",
 ]
 
 [[package]]
@@ -256,15 +256,6 @@ version = "2.6.4"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "f665ee40bc4a3c5590afb1e9677db74a508659dfd71e126420da8274909a0167"
 
-[[package]]
-name = "memmap2"
-version = "0.9.0"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "deaba38d7abf1d4cca21cc89e932e542ba2b9258664d2a9ef0e61512039c9375"
-dependencies = [
- "libc",
-]
-
 [[package]]
 name = "miniz_oxide"
 version = "0.7.1"

c-abi.md

@@ -0,0 +1,29 @@
+# C ABI (proposal)
+
+## C datatypes
+| C Type      | Description              |     Size (B) |
+|:------------|:-------------------------|-------------:|
+| char        | Character / byte         |            8 |
+| short       | Short integer            |           16 |
+| int         | Integer                  |           32 |
+| long        | Long integer             |           64 |
+| long long   | Long long integer        |           64 |
+| T*          | Pointer                  |           64 |
+| float       | Single-precision float   |           32 |
+| double      | Double-precision float   |           64 |
+| long double | Extended-precision float | **Bikeshed** |
+
+## Registers
+| Register | ABI Name | Description    | Saver  |
+|:---------|:---------|:---------------|:-------|
+| `r0`     | —        | Zero register  | N/A    |
+| `r1`     | `ra`     | Return address | Caller |
+| `r2`     | `sp`     | Stack pointer  | Callee |
+| `r3`     | `tp`     | Thread pointer | N/A    |
+
+**TODO:** Parameters
+
+**TODO:** Saved
+
+**TODO:** Temp
+

hbbytecode/instructions.in

@@ -47,7 +47,7 @@
 0x2E, ADDI16,   RRH,  "Addition with immediate (16b)"               ;
 0x2F, ADDI32,   RRW,  "Addition with immediate (32b)"               ;
 0x30, ADDI64,   RRD,  "Addition with immediate (64b)"               ;
-0x31, MULI8,    RRB,  "Multiplication with immediate (8b)"          ;
+0x31, MULI8,    RRW,  "Multiplication with immediate (8b)"          ;
 0x32, MULI16,   RRH,  "Multiplication with immediate (16b)"         ;
 0x33, MULI32,   RRW,  "Multiplication with immediate (32b)"         ;
 0x34, MULI64,   RRD,  "Multiplication with immediate (64b)"         ;
@@ -114,7 +114,7 @@
 0x71, FTI64,    RRB,  "Float 64 to int"                             ;
 0x72, FC32T64,  RR,   "Float 64 to Float 32"                        ;
 0x73, FC64T32,  RRB,  "Float 32 to Float 64"                        ;
-0x74, LRA16,    RRO,  "Load relative immediate (16 bit)"            ;
+0x74, LRA16,    RRP,  "Load relative immediate (16 bit)"            ;
 0x75, LDR16,    RRPH, "Load from relative address (16 bit)"         ;
 0x76, STR16,    RRPH, "Store to relative address (16 bit)"          ;
 0x77, JMP16,    P,    "Relative jump (16 bit)"                      ;

hbvm/src/bmc.rs

@@ -43,19 +43,17 @@ impl BlockCopier {
     /// - Same as for [`Memory::load`] and [`Memory::store`]
     pub unsafe fn poll(&mut self, memory: &mut impl Memory) -> Poll<Result<(), BlkCopyError>> {
         // Safety: Assuming uninit of array of MaybeUninit is sound
-        let mut buf = AlignedBuf(unsafe { MaybeUninit::uninit().assume_init() });
+        let mut buf = AlignedBuf(MaybeUninit::uninit().assume_init());
 
         // We have at least one buffer size to copy
         if self.n_buffers != 0 {
-            if let Err(e) = unsafe {
+            if let Err(e) = act(
-                act(
                 memory,
                 self.src,
                 self.dst,
                 buf.0.as_mut_ptr().cast(),
                 BUF_SIZE,
-                )
+            ) {
-            } {
                 return Poll::Ready(Err(e));
             }
 
@@ -75,15 +73,13 @@ impl BlockCopier {
         }
 
         if self.rem != 0 {
-            if let Err(e) = unsafe {
+            if let Err(e) = act(
-                act(
                 memory,
                 self.src,
                 self.dst,
                 buf.0.as_mut_ptr().cast(),
                 self.rem,
-                )
+            ) {
-            } {
                 return Poll::Ready(Err(e));
             }
         }
@@ -101,7 +97,6 @@ unsafe fn act(
     buf: *mut u8,
     count: usize,
 ) -> Result<(), BlkCopyError> {
-    unsafe {
     // Load to buffer
     memory
         .load(src, buf, count)
@@ -117,7 +112,6 @@ unsafe fn act(
             access_reason: MemoryAccessReason::Store,
             addr,
         })?;
-    }
 
     Ok(())
 }

hbvm/src/float/aarch64.rs

@@ -46,7 +46,7 @@ unsafe fn set_rounding_mode(mode: RoundingMode) {
     }
 
     let fpcr: u64;
-    unsafe { asm!("mrs {}, fpcr", out(reg) fpcr) };
+    asm!("mrs {}, fpcr", out(reg) fpcr);
 
     let fpcr = fpcr & !(0b11 << 22)
         | (match mode {
@@ -56,7 +56,7 @@ unsafe fn set_rounding_mode(mode: RoundingMode) {
             RoundingMode::Down => 0b10,
         }) << 22;
 
-    unsafe { asm!("msr fpcr, {}", in(reg) fpcr) };
+    asm!("msr fpcr, {}", in(reg) fpcr);
 }
 
 #[inline(always)]

hbvm/src/float/x86_64.rs

@@ -56,14 +56,12 @@ fnsdef! {
 ///   [`default_rounding_mode`], you have to rely on inline assembly
 #[inline(always)]
 unsafe fn set_rounding_mode(mode: RoundingMode) {
-    unsafe {
     arin::_MM_SET_ROUNDING_MODE(match mode {
         RoundingMode::NearestEven => return,
         RoundingMode::Truncate => arin::_MM_ROUND_TOWARD_ZERO,
         RoundingMode::Up => arin::_MM_ROUND_UP,
         RoundingMode::Down => arin::_MM_ROUND_DOWN,
     })
-    }
 }
 
 #[inline(always)]

hbvm/src/lib.rs

@@ -12,7 +12,6 @@
 
 #![no_std]
 #![cfg_attr(feature = "nightly", feature(fn_align))]
-#![deny(unsafe_op_in_unsafe_fn)]
 
 #[cfg(feature = "alloc")]
 extern crate alloc;

hbvm/src/mem/softpaging/icache.rs

@@ -48,14 +48,14 @@ impl ICache {
 
         let pbase = self
             .data
-            .or_else(|| unsafe { self.fetch_page(self.base + self.size, root_pt) })?;
+            .or_else(|| self.fetch_page(self.base + self.size, root_pt))?;
 
         // Get address base
         let base = addr.map(|x| x & self.mask);
 
         // Base not matching, fetch anew
         if base != self.base {
-            unsafe { self.fetch_page(base, root_pt) }?;
+            self.fetch_page(base, root_pt)?;
         };
 
         let offset = addr.get() & !self.mask;
@@ -68,27 +68,25 @@ impl ICache {
         let first_copy = requ_size.saturating_sub(rem);
 
         // Copy non-overflowing part
-        unsafe { copy_nonoverlapping(pbase.as_ptr(), ret.as_mut_ptr().cast::<u8>(), first_copy) };
+        copy_nonoverlapping(pbase.as_ptr(), ret.as_mut_ptr().cast::<u8>(), first_copy);
 
         // Copy overflow
         if rem != 0 {
-            let pbase = unsafe { self.fetch_page(self.base + self.size, root_pt) }?;
+            let pbase = self.fetch_page(self.base + self.size, root_pt)?;
 
             // Unlikely, unsupported scenario
             if rem > self.size as _ {
                 return None;
             }
 
-            unsafe {
             copy_nonoverlapping(
                 pbase.as_ptr(),
                 ret.as_mut_ptr().cast::<u8>().add(first_copy),
                 rem,
-                )
+            );
-            };
         }
 
-        Some(unsafe { ret.assume_init() })
+        Some(ret.assume_init())
     }
 
     /// Fetch a page

hbvm/src/mem/softpaging/mapping.rs

@@ -36,11 +36,9 @@ impl<'p, A, const OUT_PROG_EXEC: bool> SoftPagedMem<'p, A, OUT_PROG_EXEC> {
 
         // Walk pagetable levels
         for lvl in (lookup_depth + 1..5).rev() {
-            let entry = unsafe {
+            let entry = (*current_pt)
-                (*current_pt)
                 .table
-                    .get_unchecked_mut(addr_extract_index(target, lvl))
+                .get_unchecked_mut(addr_extract_index(target, lvl));
-            };
 
             let ptr = entry.ptr();
             match entry.permission() {
@@ -48,13 +46,13 @@ impl<'p, A, const OUT_PROG_EXEC: bool> SoftPagedMem<'p, A, OUT_PROG_EXEC> {
                 // No worries! Let's create one (allocates).
                 Permission::Empty => {
                     // Increase children count
-                    unsafe { *current_pt }.childen += 1;
+                    (*current_pt).childen += 1;
 
                     let table = Box::into_raw(Box::new(PtPointedData {
                         pt: PageTable::default(),
                     }));
 
-                    unsafe { core::ptr::write(entry, PtEntry::new(table, Permission::Node)) };
+                    core::ptr::write(entry, PtEntry::new(table, Permission::Node));
                     current_pt = table as _;
                 }
                 // Continue walking
@@ -65,11 +63,9 @@ impl<'p, A, const OUT_PROG_EXEC: bool> SoftPagedMem<'p, A, OUT_PROG_EXEC> {
             }
         }
 
-        let node = unsafe {
+        let node = (*current_pt)
-            (*current_pt)
             .table
-                .get_unchecked_mut(addr_extract_index(target, lookup_depth))
+            .get_unchecked_mut(addr_extract_index(target, lookup_depth));
-        };
 
         // Check if node is not mapped
         if node.permission() != Permission::Empty {
@@ -77,10 +73,8 @@ impl<'p, A, const OUT_PROG_EXEC: bool> SoftPagedMem<'p, A, OUT_PROG_EXEC> {
         }
 
         // Write entry
-        unsafe {
         (*current_pt).childen += 1;
         core::ptr::write(node, PtEntry::new(host.cast(), perm));
-        }
 
         Ok(())
     }

hbvm/src/mem/softpaging/mod.rs

@@ -51,7 +51,7 @@ impl<'p, PfH: HandlePageFault, const OUT_PROG_EXEC: bool> Memory
             target,
             count,
             perm_check::readable,
-            |src, dst, count| unsafe { core::ptr::copy_nonoverlapping(src, dst, count) },
+            |src, dst, count| core::ptr::copy_nonoverlapping(src, dst, count),
         )
         .map_err(LoadError)
     }
@@ -72,7 +72,7 @@ impl<'p, PfH: HandlePageFault, const OUT_PROG_EXEC: bool> Memory
             source.cast_mut(),
             count,
             perm_check::writable,
-            |dst, src, count| unsafe { core::ptr::copy_nonoverlapping(src, dst, count) },
+            |dst, src, count| core::ptr::copy_nonoverlapping(src, dst, count),
         )
         .map_err(StoreError)
     }
@@ -80,14 +80,16 @@ impl<'p, PfH: HandlePageFault, const OUT_PROG_EXEC: bool> Memory
     #[inline(always)]
     unsafe fn prog_read<T>(&mut self, addr: Address) -> T {
         if OUT_PROG_EXEC && addr.truncate_usize() > self.program.len() {
-            return unsafe { self.icache.fetch::<T>(addr, self.root_pt) }
+            return self
+                .icache
+                .fetch::<T>(addr, self.root_pt)
                 .unwrap_or_else(|| unsafe { core::mem::zeroed() });
         }
 
         let addr = addr.truncate_usize();
         self.program
             .get(addr..addr + size_of::<T>())
-            .map(|x| unsafe { x.as_ptr().cast::<T>().read() })
+            .map(|x| x.as_ptr().cast::<T>().read())
             .unwrap_or_else(|| unsafe { core::mem::zeroed() })
     }
 }

hbvm/src/vmrun.rs

@@ -17,8 +17,7 @@ use {
 
 macro_rules! handler {
     ($self:expr, |$ty:ident ($($ident:pat),* $(,)?)| $expr:expr) => {{
-        #[allow(unused_unsafe)]
+        let $ty($($ident),*) = $self.decode::<$ty>();
-        let $ty($($ident),*) = unsafe { $self.decode::<$ty>() };
         #[allow(clippy::no_effect)] let e = $expr;
         $self.bump_pc::<$ty>();
         e
@@ -41,14 +40,14 @@ where
             // Contribution guide:
             // - Zero register shall never be overwitten. It's value has to always be 0.
             //     - Prefer `Self::read_reg` and `Self::write_reg` functions
-            // - Try to use `handler!` macro for decoding and then bumping program counter
+            // - Extract parameters using `param!` macro
             // - Prioritise speed over code size
             //     - Memory is cheap, CPUs not that much
             // - Do not heap allocate at any cost
             //     - Yes, user-provided trap handler may allocate,
             //       but that is not our »fault«.
             // - Unsafe is kinda must, but be sure you have validated everything
-            //     - Your contributions have to pass sanitizers, fuzzer and Miri
+            //     - Your contributions have to pass sanitizers and Miri
             // - Strictly follow the spec
             //     - The spec does not specify how you perform actions, in what order,
             //       just that the observable effects have to be performed in order and
@@ -375,16 +374,13 @@ where
     /// Bump instruction pointer
     #[inline(always)]
     fn bump_pc<T: Copy>(&mut self) {
-        self.pc = self
+        self.pc = self.pc.wrapping_add(core::mem::size_of::<T>());
-            .pc
-            .wrapping_add(core::mem::size_of::<T>())
-            .wrapping_add(1);
     }
 
     /// Decode instruction operands
     #[inline(always)]
     unsafe fn decode<T: Copy>(&mut self) -> T {
-        unsafe { self.memory.prog_read::<T>(self.pc + 1_u64) }
+        self.memory.prog_read::<T>(self.pc + 1_u64)
     }
 
     /// Load
@@ -401,7 +397,6 @@ where
             _ => 0,
         };
 
-        unsafe {
         self.memory.load(
             self.ldst_addr_uber(dst, base, offset, count, n)?,
             self.registers
@@ -409,8 +404,7 @@ where
                 .add(usize::from(dst) + usize::from(n))
                 .cast(),
             usize::from(count).saturating_sub(n.into()),
-            )
+        )?;
-        }?;
 
         Ok(())
     }
@@ -424,13 +418,11 @@ where
         offset: u64,
         count: u16,
     ) -> Result<(), VmRunError> {
-        unsafe {
         self.memory.store(
             self.ldst_addr_uber(dst, base, offset, count, 0)?,
             self.registers.as_ptr().add(usize::from(dst)).cast(),
             count.into(),
-            )
+        )?;
-        }?;
         Ok(())
     }
 
@@ -443,7 +435,7 @@ where
     /// Perform binary operating over two registers
     #[inline(always)]
     unsafe fn binary_op<T: ValueVariant>(&mut self, op: impl Fn(T, T) -> T) {
-        let OpsRRR(tg, a0, a1) = unsafe { self.decode() };
+        let OpsRRR(tg, a0, a1) = self.decode();
         self.write_reg(
             tg,
             op(self.read_reg(a0).cast::<T>(), self.read_reg(a1).cast::<T>()),
@@ -458,7 +450,7 @@ where
         #[repr(packed)]
         struct OpsRRImm<I>(OpsRR, I);
 
-        let OpsRRImm::<T>(OpsRR(tg, reg), imm) = unsafe { self.decode() };
+        let OpsRRImm::<T>(OpsRR(tg, reg), imm) = self.decode();
         self.write_reg(tg, op(self.read_reg(reg).cast::<T>(), imm));
         self.bump_pc::<OpsRRImm<T>>();
     }
@@ -466,7 +458,7 @@ where
     /// Perform binary operation over register and shift immediate
     #[inline(always)]
     unsafe fn binary_op_shift<T: ValueVariant>(&mut self, op: impl Fn(T, u32) -> T) {
-        let OpsRRR(tg, a0, a1) = unsafe { self.decode() };
+        let OpsRRR(tg, a0, a1) = self.decode();
         self.write_reg(
             tg,
             op(
@@ -480,7 +472,7 @@ where
     /// Perform binary operation over register and shift immediate
     #[inline(always)]
     unsafe fn binary_op_ims<T: ValueVariant>(&mut self, op: impl Fn(T, u32) -> T) {
-        let OpsRRB(tg, reg, imm) = unsafe { self.decode() };
+        let OpsRRB(tg, reg, imm) = self.decode();
         self.write_reg(tg, op(self.read_reg(reg).cast::<T>(), imm.into()));
         self.bump_pc::<OpsRRW>();
     }
@@ -539,7 +531,7 @@ where
     /// Jump at `PC + #3` if ordering on `#0 <=> #1` is equal to expected
     #[inline(always)]
     unsafe fn cond_jmp<T: ValueVariant + Ord>(&mut self, expected: Ordering) {
-        let OpsRRP(a0, a1, ja) = unsafe { self.decode() };
+        let OpsRRP(a0, a1, ja) = self.decode();
         if self
             .read_reg(a0)
             .cast::<T>()

hbxrt/Cargo.toml

@@ -6,4 +6,4 @@ default-run = "hbxrt"
 
 [dependencies]
 hbvm.path = "../hbvm"
-memmap2   = "0.9"
+nix       = { version = "0.27", features = ["mman", "signal"] }

hbxrt/src/main.rs

@@ -1,50 +1,65 @@
 //! Holey Bytes Experimental Runtime
-
-#![deny(unsafe_op_in_unsafe_fn)]
-
 mod mem;
 
 use {
     hbvm::{mem::Address, Vm, VmRunOk},
-    memmap2::Mmap,
+    nix::sys::mman::{mmap, MapFlags, ProtFlags},
-    std::{env::args, fs::File, mem::MaybeUninit, process::exit},
+    std::{env::args, fs::File, num::NonZeroUsize, process::exit},
 };
 
 fn main() -> Result<(), Box<dyn std::error::Error>> {
-    eprintln!("== HB×RT (Holey Bytes Experimental Runtime) v0.1 ==");
+    eprintln!("== HB×RT (Holey Bytes Linux Runtime) v0.1 ==");
     eprintln!("[W] Currently supporting only flat images");
 
-    let mut args = args().skip(1);
+    let Some(image_path) = args().nth(1) else {
-    let Some(image_path) = args.next() else {
         eprintln!("[E] Missing image path");
         exit(1);
     };
 
-    let dsls = args.next().as_deref() == Some("-L");
-    if cfg!(not(target_os = "linux")) && dsls {
-        eprintln!("[E] Unsupported platform for Direct Linux syscall mode");
-        exit(1);
-    }
-
-    if dsls {
-        eprintln!("[I] Direct Linux syscall mode activated")
-    }
-
-    // Allocate stack
-    let mut stack = Box::new(MaybeUninit::<[u8; 1024 * 1024 * 2]>::uninit());
-    eprintln!("[I] Stack allocated at {:p}", stack.as_ptr());
-
     // Load program
     eprintln!("[I] Loading image from \"{image_path}\"");
-    let file_handle = File::open(image_path)?;
+    let file = File::open(image_path)?;
-    let mmap = unsafe { Mmap::map(&file_handle) }?;
+    let ptr = unsafe {
+        mmap(
+            None,
+            NonZeroUsize::new(file.metadata()?.len() as usize).ok_or("File is empty")?,
+            ProtFlags::PROT_READ,
+            MapFlags::MAP_PRIVATE,
+            Some(&file),
+            0,
+        )?
+    };
 
-    eprintln!("[I] Image loaded at {:p}", mmap.as_ptr());
+    eprintln!("[I] Image loaded at {ptr:p}");
-
-    let mut vm = unsafe { Vm::<_, 0>::new(mem::HostMemory, Address::new(mmap.as_ptr() as u64)) };
-    vm.write_reg(254, stack.as_mut_ptr() as u64);
 
     // Execute program
+    let mut vm = unsafe { Vm::<_, 0>::new(mem::HostMemory, Address::new(ptr as u64)) };
+
+    // Memory access fault handling
+    unsafe {
+        use nix::sys::signal;
+
+        extern "C" fn action(
+            _: std::ffi::c_int,
+            info: *mut nix::libc::siginfo_t,
+            _: *mut std::ffi::c_void,
+        ) {
+            unsafe {
+                eprintln!("[E] Memory access fault at {:p}", (*info).si_addr());
+                exit(2);
+            }
+        }
+
+        signal::sigaction(
+            signal::Signal::SIGSEGV,
+            &nix::sys::signal::SigAction::new(
+                signal::SigHandler::SigAction(action),
+                signal::SaFlags::SA_NODEFER,
+                nix::sys::signalfd::SigSet::empty(),
+            ),
+        )?;
+    }
+
     let stat = loop {
         match vm.run() {
             Ok(VmRunOk::Breakpoint) => eprintln!(
@@ -52,7 +67,7 @@ fn main() -> Result<(), Box<dyn std::error::Error>> {
                 vm.pc, vm.registers
             ),
             Ok(VmRunOk::Timer) => (),
-            Ok(VmRunOk::Ecall) if dsls => unsafe {
+            Ok(VmRunOk::Ecall) => unsafe {
                 std::arch::asm!(
                     "syscall",
                     inlateout("rax") vm.registers[1].0,
@@ -64,10 +79,6 @@ fn main() -> Result<(), Box<dyn std::error::Error>> {
                     in("r9")  vm.registers[7].0,
                 )
             },
-            Ok(VmRunOk::Ecall) => {
-                eprintln!("[E] General environment calls not supported");
-                exit(1);
-            }
             Ok(VmRunOk::End) => break Ok(()),
             Err(e) => break Err(e),
         }

hbxrt/src/mem.rs

@@ -26,6 +26,6 @@ impl Memory for HostMemory {
 
     #[inline]
     unsafe fn prog_read<T: Copy>(&mut self, addr: Address) -> T {
-        unsafe { core::ptr::read(addr.get() as *const T) }
+        core::ptr::read(addr.get() as *const T)
     }
 }

spec.md

@@ -480,26 +480,18 @@ Program counter stays on the currently executed instruction
 | long long   | Long long integer        | 8          |
 | float       | Single-precision float   | 4          |
 | double      | Double-precision float   | 8          |
-| long double | Extended-precision float | 8          |
+| long double | Extended-precision float | TBD        |
-
-- Bikeshedding note: `long double` is now 8 bytes as
-  the base ISA does not support `f128`. an extension
-  for that should be made.
 
 ## Call convention
-- Registers r1 – r31 are caller saved
+- Registers r1 – r30 are caller saved
-- Registers r32 – r255 are callee saved
+- Registers r31 – r255 are callee saved
 
 | Register | Description         | Saver  |
-|:-----------|:--------------------|:-------|
+|:---------|:--------------------|:-------|
 | r0       | Hard-wired zero     | N/A    |
 | r1 - r2  | Return values       | Caller |
 | r2 - r11 | Function parameters | Caller |
-| r12 - r30  | General purpose     | Caller |
+| r30      | Return address      | Caller |
-| r31        | Return address      | Caller |
-| r32 - r253 | General purpose     | Callee |
-| r254       | Stack pointer       | Callee |
-| r255       | Thread pointer      | N/A    |
 
-- If return value is too big to fit r1, r2 is also used.
+If return value is too big to fit one register, r2 is also used.
-- Values larger than two double-words are passed by reference
+TODO: Stack pointer, Thread pointer, ...