From 698f8c2f01ea549d77d7dc3338a12e04c11057b9 Mon Sep 17 00:00:00 2001 From: Daniel Baumann Date: Wed, 17 Apr 2024 14:02:58 +0200 Subject: Adding upstream version 1.64.0+dfsg1. Signed-off-by: Daniel Baumann --- .../rustc_const_eval/src/interpret/terminator.rs | 671 +++++++++++++++++++++ 1 file changed, 671 insertions(+) create mode 100644 compiler/rustc_const_eval/src/interpret/terminator.rs (limited to 'compiler/rustc_const_eval/src/interpret/terminator.rs') diff --git a/compiler/rustc_const_eval/src/interpret/terminator.rs b/compiler/rustc_const_eval/src/interpret/terminator.rs new file mode 100644 index 000000000..d563e35f9 --- /dev/null +++ b/compiler/rustc_const_eval/src/interpret/terminator.rs @@ -0,0 +1,671 @@ +use std::borrow::Cow; + +use rustc_middle::ty::layout::{FnAbiOf, LayoutOf}; +use rustc_middle::ty::Instance; +use rustc_middle::{ + mir, + ty::{self, Ty}, +}; +use rustc_target::abi; +use rustc_target::abi::call::{ArgAbi, ArgAttribute, ArgAttributes, FnAbi, PassMode}; +use rustc_target::spec::abi::Abi; + +use super::{ + FnVal, ImmTy, Immediate, InterpCx, InterpResult, MPlaceTy, Machine, MemoryKind, OpTy, Operand, + PlaceTy, Scalar, StackPopCleanup, StackPopUnwind, +}; + +impl<'mir, 'tcx: 'mir, M: Machine<'mir, 'tcx>> InterpCx<'mir, 'tcx, M> { + pub(super) fn eval_terminator( + &mut self, + terminator: &mir::Terminator<'tcx>, + ) -> InterpResult<'tcx> { + use rustc_middle::mir::TerminatorKind::*; + match terminator.kind { + Return => { + self.pop_stack_frame(/* unwinding */ false)? + } + + Goto { target } => self.go_to_block(target), + + SwitchInt { ref discr, ref targets, switch_ty } => { + let discr = self.read_immediate(&self.eval_operand(discr, None)?)?; + trace!("SwitchInt({:?})", *discr); + assert_eq!(discr.layout.ty, switch_ty); + + // Branch to the `otherwise` case by default, if no match is found. + assert!(!targets.iter().is_empty()); + let mut target_block = targets.otherwise(); + + for (const_int, target) in targets.iter() { + // Compare using MIR BinOp::Eq, to also support pointer values. + // (Avoiding `self.binary_op` as that does some redundant layout computation.) + let res = self + .overflowing_binary_op( + mir::BinOp::Eq, + &discr, + &ImmTy::from_uint(const_int, discr.layout), + )? + .0; + if res.to_bool()? { + target_block = target; + break; + } + } + + self.go_to_block(target_block); + } + + Call { + ref func, + ref args, + destination, + target, + ref cleanup, + from_hir_call: _, + fn_span: _, + } => { + let old_stack = self.frame_idx(); + let old_loc = self.frame().loc; + let func = self.eval_operand(func, None)?; + let args = self.eval_operands(args)?; + + let fn_sig_binder = func.layout.ty.fn_sig(*self.tcx); + let fn_sig = + self.tcx.normalize_erasing_late_bound_regions(self.param_env, fn_sig_binder); + let extra_args = &args[fn_sig.inputs().len()..]; + let extra_args = self.tcx.mk_type_list(extra_args.iter().map(|arg| arg.layout.ty)); + + let (fn_val, fn_abi, with_caller_location) = match *func.layout.ty.kind() { + ty::FnPtr(_sig) => { + let fn_ptr = self.read_pointer(&func)?; + let fn_val = self.get_ptr_fn(fn_ptr)?; + (fn_val, self.fn_abi_of_fn_ptr(fn_sig_binder, extra_args)?, false) + } + ty::FnDef(def_id, substs) => { + let instance = + self.resolve(ty::WithOptConstParam::unknown(def_id), substs)?; + ( + FnVal::Instance(instance), + self.fn_abi_of_instance(instance, extra_args)?, + instance.def.requires_caller_location(*self.tcx), + ) + } + _ => span_bug!( + terminator.source_info.span, + "invalid callee of type {:?}", + func.layout.ty + ), + }; + + let destination = self.eval_place(destination)?; + self.eval_fn_call( + fn_val, + (fn_sig.abi, fn_abi), + &args, + with_caller_location, + &destination, + target, + match (cleanup, fn_abi.can_unwind) { + (Some(cleanup), true) => StackPopUnwind::Cleanup(*cleanup), + (None, true) => StackPopUnwind::Skip, + (_, false) => StackPopUnwind::NotAllowed, + }, + )?; + // Sanity-check that `eval_fn_call` either pushed a new frame or + // did a jump to another block. + if self.frame_idx() == old_stack && self.frame().loc == old_loc { + span_bug!(terminator.source_info.span, "evaluating this call made no progress"); + } + } + + Drop { place, target, unwind } => { + let place = self.eval_place(place)?; + let ty = place.layout.ty; + trace!("TerminatorKind::drop: {:?}, type {}", place, ty); + + let instance = Instance::resolve_drop_in_place(*self.tcx, ty); + self.drop_in_place(&place, instance, target, unwind)?; + } + + Assert { ref cond, expected, ref msg, target, cleanup } => { + let cond_val = + self.read_immediate(&self.eval_operand(cond, None)?)?.to_scalar()?.to_bool()?; + if expected == cond_val { + self.go_to_block(target); + } else { + M::assert_panic(self, msg, cleanup)?; + } + } + + Abort => { + M::abort(self, "the program aborted execution".to_owned())?; + } + + // When we encounter Resume, we've finished unwinding + // cleanup for the current stack frame. We pop it in order + // to continue unwinding the next frame + Resume => { + trace!("unwinding: resuming from cleanup"); + // By definition, a Resume terminator means + // that we're unwinding + self.pop_stack_frame(/* unwinding */ true)?; + return Ok(()); + } + + // It is UB to ever encounter this. + Unreachable => throw_ub!(Unreachable), + + // These should never occur for MIR we actually run. + DropAndReplace { .. } + | FalseEdge { .. } + | FalseUnwind { .. } + | Yield { .. } + | GeneratorDrop => span_bug!( + terminator.source_info.span, + "{:#?} should have been eliminated by MIR pass", + terminator.kind + ), + + // Inline assembly can't be interpreted. + InlineAsm { .. } => throw_unsup_format!("inline assembly is not supported"), + } + + Ok(()) + } + + fn check_argument_compat( + caller_abi: &ArgAbi<'tcx, Ty<'tcx>>, + callee_abi: &ArgAbi<'tcx, Ty<'tcx>>, + ) -> bool { + // Heuristic for type comparison. + let layout_compat = || { + if caller_abi.layout.ty == callee_abi.layout.ty { + // No question + return true; + } + if caller_abi.layout.is_unsized() || callee_abi.layout.is_unsized() { + // No, no, no. We require the types to *exactly* match for unsized arguments. If + // these are somehow unsized "in a different way" (say, `dyn Trait` vs `[i32]`), + // then who knows what happens. + return false; + } + if caller_abi.layout.size != callee_abi.layout.size + || caller_abi.layout.align.abi != callee_abi.layout.align.abi + { + // This cannot go well... + return false; + } + // The rest *should* be okay, but we are extra conservative. + match (caller_abi.layout.abi, callee_abi.layout.abi) { + // Different valid ranges are okay (once we enforce validity, + // that will take care to make it UB to leave the range, just + // like for transmute). + (abi::Abi::Scalar(caller), abi::Abi::Scalar(callee)) => { + caller.primitive() == callee.primitive() + } + ( + abi::Abi::ScalarPair(caller1, caller2), + abi::Abi::ScalarPair(callee1, callee2), + ) => { + caller1.primitive() == callee1.primitive() + && caller2.primitive() == callee2.primitive() + } + // Be conservative + _ => false, + } + }; + // Padding must be fully equal. + let pad_compat = || caller_abi.pad == callee_abi.pad; + // When comparing the PassMode, we have to be smart about comparing the attributes. + let arg_attr_compat = |a1: ArgAttributes, a2: ArgAttributes| { + // There's only one regular attribute that matters for the call ABI: InReg. + // Everything else is things like noalias, dereferencable, nonnull, ... + // (This also applies to pointee_size, pointee_align.) + if a1.regular.contains(ArgAttribute::InReg) != a2.regular.contains(ArgAttribute::InReg) + { + return false; + } + // We also compare the sign extension mode -- this could let the callee make assumptions + // about bits that conceptually were not even passed. + if a1.arg_ext != a2.arg_ext { + return false; + } + return true; + }; + let mode_compat = || match (caller_abi.mode, callee_abi.mode) { + (PassMode::Ignore, PassMode::Ignore) => true, + (PassMode::Direct(a1), PassMode::Direct(a2)) => arg_attr_compat(a1, a2), + (PassMode::Pair(a1, b1), PassMode::Pair(a2, b2)) => { + arg_attr_compat(a1, a2) && arg_attr_compat(b1, b2) + } + (PassMode::Cast(c1), PassMode::Cast(c2)) => c1 == c2, + ( + PassMode::Indirect { attrs: a1, extra_attrs: None, on_stack: s1 }, + PassMode::Indirect { attrs: a2, extra_attrs: None, on_stack: s2 }, + ) => arg_attr_compat(a1, a2) && s1 == s2, + ( + PassMode::Indirect { attrs: a1, extra_attrs: Some(e1), on_stack: s1 }, + PassMode::Indirect { attrs: a2, extra_attrs: Some(e2), on_stack: s2 }, + ) => arg_attr_compat(a1, a2) && arg_attr_compat(e1, e2) && s1 == s2, + _ => false, + }; + + if layout_compat() && pad_compat() && mode_compat() { + return true; + } + trace!( + "check_argument_compat: incompatible ABIs:\ncaller: {:?}\ncallee: {:?}", + caller_abi, + callee_abi + ); + return false; + } + + /// Initialize a single callee argument, checking the types for compatibility. + fn pass_argument<'x, 'y>( + &mut self, + caller_args: &mut impl Iterator< + Item = (&'x OpTy<'tcx, M::Provenance>, &'y ArgAbi<'tcx, Ty<'tcx>>), + >, + callee_abi: &ArgAbi<'tcx, Ty<'tcx>>, + callee_arg: &PlaceTy<'tcx, M::Provenance>, + ) -> InterpResult<'tcx> + where + 'tcx: 'x, + 'tcx: 'y, + { + if matches!(callee_abi.mode, PassMode::Ignore) { + // This one is skipped. + return Ok(()); + } + // Find next caller arg. + let (caller_arg, caller_abi) = caller_args.next().ok_or_else(|| { + err_ub_format!("calling a function with fewer arguments than it requires") + })?; + // Now, check + if !Self::check_argument_compat(caller_abi, callee_abi) { + throw_ub_format!( + "calling a function with argument of type {:?} passing data of type {:?}", + callee_arg.layout.ty, + caller_arg.layout.ty + ) + } + // Special handling for unsized parameters. + if caller_arg.layout.is_unsized() { + // `check_argument_compat` ensures that both have the same type, so we know they will use the metadata the same way. + assert_eq!(caller_arg.layout.ty, callee_arg.layout.ty); + // We have to properly pre-allocate the memory for the callee. + // So let's tear down some wrappers. + // This all has to be in memory, there are no immediate unsized values. + let src = caller_arg.assert_mem_place(); + // The destination cannot be one of these "spread args". + let (dest_frame, dest_local) = callee_arg.assert_local(); + // We are just initializing things, so there can't be anything here yet. + assert!(matches!( + *self.local_to_op(&self.stack()[dest_frame], dest_local, None)?, + Operand::Immediate(Immediate::Uninit) + )); + // Allocate enough memory to hold `src`. + let Some((size, align)) = self.size_and_align_of_mplace(&src)? else { + span_bug!(self.cur_span(), "unsized fn arg with `extern` type tail should not be allowed") + }; + let ptr = self.allocate_ptr(size, align, MemoryKind::Stack)?; + let dest_place = + MPlaceTy::from_aligned_ptr_with_meta(ptr.into(), callee_arg.layout, src.meta); + // Update the local to be that new place. + *M::access_local_mut(self, dest_frame, dest_local)? = Operand::Indirect(*dest_place); + } + // We allow some transmutes here. + // FIXME: Depending on the PassMode, this should reset some padding to uninitialized. (This + // is true for all `copy_op`, but there are a lot of special cases for argument passing + // specifically.) + self.copy_op(&caller_arg, callee_arg, /*allow_transmute*/ true) + } + + /// Call this function -- pushing the stack frame and initializing the arguments. + /// + /// `caller_fn_abi` is used to determine if all the arguments are passed the proper way. + /// However, we also need `caller_abi` to determine if we need to do untupling of arguments. + /// + /// `with_caller_location` indicates whether the caller passed a caller location. Miri + /// implements caller locations without argument passing, but to match `FnAbi` we need to know + /// when those arguments are present. + pub(crate) fn eval_fn_call( + &mut self, + fn_val: FnVal<'tcx, M::ExtraFnVal>, + (caller_abi, caller_fn_abi): (Abi, &FnAbi<'tcx, Ty<'tcx>>), + args: &[OpTy<'tcx, M::Provenance>], + with_caller_location: bool, + destination: &PlaceTy<'tcx, M::Provenance>, + target: Option, + mut unwind: StackPopUnwind, + ) -> InterpResult<'tcx> { + trace!("eval_fn_call: {:#?}", fn_val); + + let instance = match fn_val { + FnVal::Instance(instance) => instance, + FnVal::Other(extra) => { + return M::call_extra_fn( + self, + extra, + caller_abi, + args, + destination, + target, + unwind, + ); + } + }; + + match instance.def { + ty::InstanceDef::Intrinsic(def_id) => { + assert!(self.tcx.is_intrinsic(def_id)); + // caller_fn_abi is not relevant here, we interpret the arguments directly for each intrinsic. + M::call_intrinsic(self, instance, args, destination, target, unwind) + } + ty::InstanceDef::VTableShim(..) + | ty::InstanceDef::ReifyShim(..) + | ty::InstanceDef::ClosureOnceShim { .. } + | ty::InstanceDef::FnPtrShim(..) + | ty::InstanceDef::DropGlue(..) + | ty::InstanceDef::CloneShim(..) + | ty::InstanceDef::Item(_) => { + // We need MIR for this fn + let Some((body, instance)) = + M::find_mir_or_eval_fn(self, instance, caller_abi, args, destination, target, unwind)? else { + return Ok(()); + }; + + // Compute callee information using the `instance` returned by + // `find_mir_or_eval_fn`. + // FIXME: for variadic support, do we have to somehow determine callee's extra_args? + let callee_fn_abi = self.fn_abi_of_instance(instance, ty::List::empty())?; + + if callee_fn_abi.c_variadic || caller_fn_abi.c_variadic { + throw_unsup_format!("calling a c-variadic function is not supported"); + } + + if M::enforce_abi(self) { + if caller_fn_abi.conv != callee_fn_abi.conv { + throw_ub_format!( + "calling a function with calling convention {:?} using calling convention {:?}", + callee_fn_abi.conv, + caller_fn_abi.conv + ) + } + } + + if !matches!(unwind, StackPopUnwind::NotAllowed) && !callee_fn_abi.can_unwind { + // The callee cannot unwind. + unwind = StackPopUnwind::NotAllowed; + } + + self.push_stack_frame( + instance, + body, + destination, + StackPopCleanup::Goto { ret: target, unwind }, + )?; + + // If an error is raised here, pop the frame again to get an accurate backtrace. + // To this end, we wrap it all in a `try` block. + let res: InterpResult<'tcx> = try { + trace!( + "caller ABI: {:?}, args: {:#?}", + caller_abi, + args.iter() + .map(|arg| (arg.layout.ty, format!("{:?}", **arg))) + .collect::>() + ); + trace!( + "spread_arg: {:?}, locals: {:#?}", + body.spread_arg, + body.args_iter() + .map(|local| ( + local, + self.layout_of_local(self.frame(), local, None).unwrap().ty + )) + .collect::>() + ); + + // In principle, we have two iterators: Where the arguments come from, and where + // they go to. + + // For where they come from: If the ABI is RustCall, we untuple the + // last incoming argument. These two iterators do not have the same type, + // so to keep the code paths uniform we accept an allocation + // (for RustCall ABI only). + let caller_args: Cow<'_, [OpTy<'tcx, M::Provenance>]> = + if caller_abi == Abi::RustCall && !args.is_empty() { + // Untuple + let (untuple_arg, args) = args.split_last().unwrap(); + trace!("eval_fn_call: Will pass last argument by untupling"); + Cow::from( + args.iter() + .map(|a| Ok(a.clone())) + .chain( + (0..untuple_arg.layout.fields.count()) + .map(|i| self.operand_field(untuple_arg, i)), + ) + .collect::>>>( + )?, + ) + } else { + // Plain arg passing + Cow::from(args) + }; + // If `with_caller_location` is set we pretend there is an extra argument (that + // we will not pass). + assert_eq!( + caller_args.len() + if with_caller_location { 1 } else { 0 }, + caller_fn_abi.args.len(), + "mismatch between caller ABI and caller arguments", + ); + let mut caller_args = caller_args + .iter() + .zip(caller_fn_abi.args.iter()) + .filter(|arg_and_abi| !matches!(arg_and_abi.1.mode, PassMode::Ignore)); + + // Now we have to spread them out across the callee's locals, + // taking into account the `spread_arg`. If we could write + // this is a single iterator (that handles `spread_arg`), then + // `pass_argument` would be the loop body. It takes care to + // not advance `caller_iter` for ZSTs. + let mut callee_args_abis = callee_fn_abi.args.iter(); + for local in body.args_iter() { + let dest = self.eval_place(mir::Place::from(local))?; + if Some(local) == body.spread_arg { + // Must be a tuple + for i in 0..dest.layout.fields.count() { + let dest = self.place_field(&dest, i)?; + let callee_abi = callee_args_abis.next().unwrap(); + self.pass_argument(&mut caller_args, callee_abi, &dest)?; + } + } else { + // Normal argument + let callee_abi = callee_args_abis.next().unwrap(); + self.pass_argument(&mut caller_args, callee_abi, &dest)?; + } + } + // If the callee needs a caller location, pretend we consume one more argument from the ABI. + if instance.def.requires_caller_location(*self.tcx) { + callee_args_abis.next().unwrap(); + } + // Now we should have no more caller args or callee arg ABIs + assert!( + callee_args_abis.next().is_none(), + "mismatch between callee ABI and callee body arguments" + ); + if caller_args.next().is_some() { + throw_ub_format!("calling a function with more arguments than it expected") + } + // Don't forget to check the return type! + if !Self::check_argument_compat(&caller_fn_abi.ret, &callee_fn_abi.ret) { + throw_ub_format!( + "calling a function with return type {:?} passing \ + return place of type {:?}", + callee_fn_abi.ret.layout.ty, + caller_fn_abi.ret.layout.ty, + ) + } + }; + match res { + Err(err) => { + self.stack_mut().pop(); + Err(err) + } + Ok(()) => Ok(()), + } + } + // cannot use the shim here, because that will only result in infinite recursion + ty::InstanceDef::Virtual(def_id, idx) => { + let mut args = args.to_vec(); + // We have to implement all "object safe receivers". So we have to go search for a + // pointer or `dyn Trait` type, but it could be wrapped in newtypes. So recursively + // unwrap those newtypes until we are there. + let mut receiver = args[0].clone(); + let receiver_place = loop { + match receiver.layout.ty.kind() { + ty::Ref(..) | ty::RawPtr(..) => break self.deref_operand(&receiver)?, + ty::Dynamic(..) => break receiver.assert_mem_place(), // no immediate unsized values + _ => { + // Not there yet, search for the only non-ZST field. + let mut non_zst_field = None; + for i in 0..receiver.layout.fields.count() { + let field = self.operand_field(&receiver, i)?; + if !field.layout.is_zst() { + assert!( + non_zst_field.is_none(), + "multiple non-ZST fields in dyn receiver type {}", + receiver.layout.ty + ); + non_zst_field = Some(field); + } + } + receiver = non_zst_field.unwrap_or_else(|| { + panic!( + "no non-ZST fields in dyn receiver type {}", + receiver.layout.ty + ) + }); + } + } + }; + // Obtain the underlying trait we are working on. + let receiver_tail = self + .tcx + .struct_tail_erasing_lifetimes(receiver_place.layout.ty, self.param_env); + let ty::Dynamic(data, ..) = receiver_tail.kind() else { + span_bug!(self.cur_span(), "dyanmic call on non-`dyn` type {}", receiver_tail) + }; + + // Get the required information from the vtable. + let vptr = receiver_place.meta.unwrap_meta().to_pointer(self)?; + let (dyn_ty, dyn_trait) = self.get_ptr_vtable(vptr)?; + if dyn_trait != data.principal() { + throw_ub_format!( + "`dyn` call on a pointer whose vtable does not match its type" + ); + } + + // Now determine the actual method to call. We can do that in two different ways and + // compare them to ensure everything fits. + let Some(ty::VtblEntry::Method(fn_inst)) = self.get_vtable_entries(vptr)?.get(idx).copied() else { + throw_ub_format!("`dyn` call trying to call something that is not a method") + }; + if cfg!(debug_assertions) { + let tcx = *self.tcx; + + let trait_def_id = tcx.trait_of_item(def_id).unwrap(); + let virtual_trait_ref = + ty::TraitRef::from_method(tcx, trait_def_id, instance.substs); + assert_eq!( + receiver_tail, + virtual_trait_ref.self_ty(), + "mismatch in underlying dyn trait computation within Miri and MIR building", + ); + let existential_trait_ref = + ty::ExistentialTraitRef::erase_self_ty(tcx, virtual_trait_ref); + let concrete_trait_ref = existential_trait_ref.with_self_ty(tcx, dyn_ty); + + let concrete_method = Instance::resolve_for_vtable( + tcx, + self.param_env, + def_id, + instance.substs.rebase_onto(tcx, trait_def_id, concrete_trait_ref.substs), + ) + .unwrap(); + assert_eq!(fn_inst, concrete_method); + } + + // `*mut receiver_place.layout.ty` is almost the layout that we + // want for args[0]: We have to project to field 0 because we want + // a thin pointer. + assert!(receiver_place.layout.is_unsized()); + let receiver_ptr_ty = self.tcx.mk_mut_ptr(receiver_place.layout.ty); + let this_receiver_ptr = self.layout_of(receiver_ptr_ty)?.field(self, 0); + // Adjust receiver argument. + args[0] = OpTy::from(ImmTy::from_immediate( + Scalar::from_maybe_pointer(receiver_place.ptr, self).into(), + this_receiver_ptr, + )); + trace!("Patched receiver operand to {:#?}", args[0]); + // recurse with concrete function + self.eval_fn_call( + FnVal::Instance(fn_inst), + (caller_abi, caller_fn_abi), + &args, + with_caller_location, + destination, + target, + unwind, + ) + } + } + } + + fn drop_in_place( + &mut self, + place: &PlaceTy<'tcx, M::Provenance>, + instance: ty::Instance<'tcx>, + target: mir::BasicBlock, + unwind: Option, + ) -> InterpResult<'tcx> { + trace!("drop_in_place: {:?},\n {:?}, {:?}", *place, place.layout.ty, instance); + // We take the address of the object. This may well be unaligned, which is fine + // for us here. However, unaligned accesses will probably make the actual drop + // implementation fail -- a problem shared by rustc. + let place = self.force_allocation(place)?; + + let (instance, place) = match place.layout.ty.kind() { + ty::Dynamic(..) => { + // Dropping a trait object. Need to find actual drop fn. + let place = self.unpack_dyn_trait(&place)?; + let instance = ty::Instance::resolve_drop_in_place(*self.tcx, place.layout.ty); + (instance, place) + } + _ => (instance, place), + }; + let fn_abi = self.fn_abi_of_instance(instance, ty::List::empty())?; + + let arg = ImmTy::from_immediate( + place.to_ref(self), + self.layout_of(self.tcx.mk_mut_ptr(place.layout.ty))?, + ); + let ret = MPlaceTy::fake_alloc_zst(self.layout_of(self.tcx.types.unit)?); + + self.eval_fn_call( + FnVal::Instance(instance), + (Abi::Rust, fn_abi), + &[arg.into()], + false, + &ret.into(), + Some(target), + match unwind { + Some(cleanup) => StackPopUnwind::Cleanup(cleanup), + None => StackPopUnwind::Skip, + }, + ) + } +} -- cgit v1.2.3