allow dynamic configuration of the heap memory (#287)

This PR changes the implementation of the emulated EVM heap memory:
Instead of linking in a C implementation the code is emitted directly
into the contract module. Which allows making it configurable via a
compiler parameter (a follow up PR to this).

---------

Signed-off-by: Cyrill Leutwiler <bigcyrill@hotmail.com>

crates/llvm-context/src/lib.rs

@@ -28,6 +28,7 @@ pub use self::polkavm::context::function::runtime::entry::Entry as PolkaVMEntryF
 pub use self::polkavm::context::function::runtime::revive::Exit as PolkaVMExitFunction;
 pub use self::polkavm::context::function::runtime::revive::WordToPointer as PolkaVMWordToPointerFunction;
 pub use self::polkavm::context::function::runtime::runtime_code::RuntimeCode as PolkaVMRuntimeCodeFunction;
+pub use self::polkavm::context::function::runtime::sbrk::Sbrk as PolkaVMSbrkFunction;
 pub use self::polkavm::context::function::runtime::FUNCTION_DEPLOY_CODE as PolkaVMFunctionDeployCode;
 pub use self::polkavm::context::function::runtime::FUNCTION_ENTRY as PolkaVMFunctionEntry;
 pub use self::polkavm::context::function::runtime::FUNCTION_RUNTIME_CODE as PolkaVMFunctionRuntimeCode;

crates/llvm-context/src/polkavm/const/mod.rs

@@ -9,6 +9,12 @@ pub static XLEN: usize = revive_common::BIT_LENGTH_X32;
 /// The calldata size global variable name.
 pub static GLOBAL_CALLDATA_SIZE: &str = "calldatasize";
 
+/// The heap size global variable name.
+pub static GLOBAL_HEAP_SIZE: &str = "__heap_size";
+
+/// The heap memory global variable name.
+pub static GLOBAL_HEAP_MEMORY: &str = "__heap_memory";
+
 /// The spill buffer global variable name.
 pub static GLOBAL_ADDRESS_SPILL_BUFFER: &str = "address_spill_buffer";
 

crates/llvm-context/src/polkavm/context/function/runtime/entry.rs

@@ -31,6 +31,21 @@ impl Entry {
             context.xlen_type().get_undef(),
         );
 
+        context.set_global(
+            crate::polkavm::GLOBAL_HEAP_SIZE,
+            context.xlen_type(),
+            AddressSpace::Stack,
+            context.xlen_type().const_zero(),
+        );
+
+        let heap_memory_type = context.byte_type().array_type(context.heap_size);
+        context.set_global(
+            crate::polkavm::GLOBAL_HEAP_MEMORY,
+            heap_memory_type,
+            AddressSpace::Stack,
+            heap_memory_type.const_zero(),
+        );
+
         let address_type = context.integer_type(revive_common::BIT_LENGTH_ETH_ADDRESS);
         context.set_global(
             crate::polkavm::GLOBAL_ADDRESS_SPILL_BUFFER,

crates/llvm-context/src/polkavm/context/function/runtime/mod.rs

@@ -5,6 +5,7 @@ pub mod deploy_code;
 pub mod entry;
 pub mod revive;
 pub mod runtime_code;
+pub mod sbrk;
 
 /// The main entry function name.
 pub const FUNCTION_ENTRY: &str = "__entry";

crates/llvm-context/src/polkavm/context/function/runtime/sbrk.rs

@@ -0,0 +1,144 @@
+//! Emulates the linear EVM heap memory via a simulated `sbrk` system call.
+
+use inkwell::values::BasicValue;
+
+use crate::polkavm::context::attribute::Attribute;
+use crate::polkavm::context::runtime::RuntimeFunction;
+use crate::polkavm::context::Context;
+use crate::polkavm::Dependency;
+use crate::polkavm::WriteLLVM;
+
+/// Simulates the `sbrk` system call, reproducing the semantics of the EVM heap memory.
+///
+/// Parameters:
+/// - The `offset` into the emulated EVM heap memory.
+/// - The `size` of the allocation emulated EVM heap memory.
+///
+/// Returns:
+/// - A pointer to the EVM heap memory at given `offset`.
+///
+/// Semantics:
+/// - Traps if the offset is out of bounds.
+/// - Aligns the total heap memory size to the EVM word size.
+/// - Traps if the memory size would be greater than the configured EVM heap memory size.
+/// - Maintains the total memory size (`msize`) in global heap size value.
+pub struct Sbrk;
+
+impl<D> RuntimeFunction<D> for Sbrk
+where
+    D: Dependency + Clone,
+{
+    const NAME: &'static str = "__sbrk_internal";
+
+    const ATTRIBUTES: &'static [Attribute] = &[
+        Attribute::NoFree,
+        Attribute::NoRecurse,
+        Attribute::WillReturn,
+    ];
+
+    fn r#type<'ctx>(context: &Context<'ctx, D>) -> inkwell::types::FunctionType<'ctx> {
+        context.llvm().ptr_type(Default::default()).fn_type(
+            &[context.xlen_type().into(), context.xlen_type().into()],
+            false,
+        )
+    }
+
+    fn emit_body<'ctx>(
+        &self,
+        context: &mut Context<'ctx, D>,
+    ) -> anyhow::Result<Option<inkwell::values::BasicValueEnum<'ctx>>> {
+        let offset = Self::paramater(context, 0).into_int_value();
+        let size = Self::paramater(context, 1).into_int_value();
+
+        let trap_block = context.append_basic_block("trap");
+        let offset_in_bounds_block = context.append_basic_block("offset_in_bounds");
+        let is_offset_out_of_bounds = context.builder().build_int_compare(
+            inkwell::IntPredicate::UGE,
+            offset,
+            context.heap_size(),
+            "offset_out_of_bounds",
+        )?;
+        context.build_conditional_branch(
+            is_offset_out_of_bounds,
+            trap_block,
+            offset_in_bounds_block,
+        )?;
+
+        context.set_basic_block(trap_block);
+        context.build_call(context.intrinsics().trap, &[], "invalid_trap");
+        context.build_unreachable();
+
+        context.set_basic_block(offset_in_bounds_block);
+        let mask = context
+            .xlen_type()
+            .const_int(revive_common::BYTE_LENGTH_WORD as u64 - 1, false);
+        let total_size = context
+            .builder()
+            .build_int_add(offset, size, "total_size")?;
+        let memory_size = context.builder().build_and(
+            context.builder().build_int_add(total_size, mask, "mask")?,
+            context.builder().build_not(mask, "mask_not")?,
+            "memory_size",
+        )?;
+        let size_in_bounds_block = context.append_basic_block("size_in_bounds");
+        let is_size_out_of_bounds = context.builder().build_int_compare(
+            inkwell::IntPredicate::UGT,
+            memory_size,
+            context.heap_size(),
+            "size_out_of_bounds",
+        )?;
+        context.build_conditional_branch(
+            is_size_out_of_bounds,
+            trap_block,
+            size_in_bounds_block,
+        )?;
+
+        context.set_basic_block(size_in_bounds_block);
+        let return_block = context.append_basic_block("return_pointer");
+        let new_size_block = context.append_basic_block("new_size");
+        let is_new_size = context.builder().build_int_compare(
+            inkwell::IntPredicate::UGT,
+            memory_size,
+            context
+                .get_global_value(crate::polkavm::GLOBAL_HEAP_SIZE)?
+                .into_int_value(),
+            "is_new_size",
+        )?;
+        context.build_conditional_branch(is_new_size, new_size_block, return_block)?;
+
+        context.set_basic_block(new_size_block);
+        context.build_store(
+            context.get_global(crate::polkavm::GLOBAL_HEAP_SIZE)?.into(),
+            memory_size,
+        )?;
+        context.build_unconditional_branch(return_block);
+
+        context.set_basic_block(return_block);
+        Ok(Some(
+            context
+                .build_gep(
+                    context
+                        .get_global(crate::polkavm::GLOBAL_HEAP_MEMORY)?
+                        .into(),
+                    &[context.xlen_type().const_zero(), offset],
+                    context.byte_type(),
+                    "allocation_start_pointer",
+                )
+                .value
+                .as_basic_value_enum(),
+        ))
+    }
+}
+
+impl<D> WriteLLVM<D> for Sbrk
+where
+    D: Dependency + Clone,
+{
+    fn declare(&mut self, context: &mut Context<D>) -> anyhow::Result<()> {
+        <Self as RuntimeFunction<_>>::declare(self, context)
+    }
+
+    fn into_llvm(self, context: &mut Context<D>) -> anyhow::Result<()> {
+        <Self as RuntimeFunction<_>>::emit(&self, context)
+    }
+}

crates/llvm-context/src/polkavm/context/mod.rs

@@ -33,6 +33,7 @@ use crate::polkavm::Dependency;
 use crate::target_machine::target::Target;
 use crate::target_machine::TargetMachine;
 use crate::PolkaVMLoadHeapWordFunction;
+use crate::PolkaVMSbrkFunction;
 use crate::PolkaVMStoreHeapWordFunction;
 
 use self::address_space::AddressSpace;
@@ -85,6 +86,8 @@ where
     loop_stack: Vec<Loop<'ctx>>,
     /// The extra LLVM arguments that were used during target initialization.
     llvm_arguments: &'ctx [String],
+    /// The emulated EVM linear heap memory size.
+    heap_size: u32,
 
     /// The project dependency manager. It can be any entity implementing the trait.
     /// The manager is used to get information about contracts and their dependencies during
@@ -254,6 +257,7 @@ where
             current_function: None,
             loop_stack: Vec::with_capacity(Self::LOOP_STACK_INITIAL_CAPACITY),
             llvm_arguments,
+            heap_size: 65536,
 
             dependency_manager,
             include_metadata_hash,
@@ -1084,32 +1088,40 @@ where
         offset: inkwell::values::IntValue<'ctx>,
         size: inkwell::values::IntValue<'ctx>,
     ) -> anyhow::Result<inkwell::values::PointerValue<'ctx>> {
-        Ok(self
-            .builder()
-            .build_call(
-                self.runtime_api_method(revive_runtime_api::polkavm_imports::SBRK),
-                &[offset.into(), size.into()],
-                "call_sbrk",
-            )?
+        let call_site_value = self.builder().build_call(
+            <PolkaVMSbrkFunction as RuntimeFunction<D>>::declaration(self).function_value(),
+            &[offset.into(), size.into()],
+            "alloc_start",
+        )?;
+
+        call_site_value.add_attribute(
+            inkwell::attributes::AttributeLoc::Return,
+            self.llvm
+                .create_enum_attribute(Attribute::NonNull as u32, 0),
+        );
+        call_site_value.add_attribute(
+            inkwell::attributes::AttributeLoc::Return,
+            self.llvm
+                .create_enum_attribute(Attribute::NoUndef as u32, 0),
+        );
+
+        Ok(call_site_value
             .try_as_basic_value()
             .left()
-            .expect("sbrk returns a pointer")
+            .unwrap_or_else(|| {
+                panic!(
+                    "revive runtime function {} should return a value",
+                    <PolkaVMSbrkFunction as RuntimeFunction<D>>::NAME,
+                )
+            })
             .into_pointer_value())
     }
 
     /// Build a call to PolkaVM `msize` for querying the linear memory size.
     pub fn build_msize(&self) -> anyhow::Result<inkwell::values::IntValue<'ctx>> {
-        let memory_size_pointer = self
-            .module()
-            .get_global(revive_runtime_api::polkavm_imports::MEMORY_SIZE)
-            .expect("the memory size symbol should have been declared")
-            .as_pointer_value();
-        let memory_size_value = self.builder().build_load(
-            self.xlen_type(),
-            memory_size_pointer,
-            "memory_size_value",
-        )?;
-        Ok(memory_size_value.into_int_value())
+        Ok(self
+            .get_global_value(crate::polkavm::GLOBAL_HEAP_SIZE)?
+            .into_int_value())
     }
 
     /// Returns a pointer to `offset` into the heap, allocating
@@ -1428,4 +1440,8 @@ where
     pub fn optimizer_settings(&self) -> &OptimizerSettings {
         self.optimizer.settings()
     }
+
+    pub fn heap_size(&self) -> inkwell::values::IntValue<'ctx> {
+        self.xlen_type().const_int(self.heap_size as u64, false)
+    }
 }