delve/pkg/proc/evalop/ops.go

252 lines
6.5 KiB
Go
Raw Normal View History

proc: use stack machine to evaluate expressions (#3508) * proc: use stack machine to evaluate expressions This commit splits expression evaluation into two parts. The first part (in pkg/proc/evalop/evalcompile.go) "compiles" as ast.Expr into a list of instructions (defined in pkg/proc/evalop/ops.go) for a stack machine (defined by `proc.(*evalStack)`). The second part is a stack machine (implemented by `proc.(*EvalScope).eval` and `proc.(*EvalScope).evalOne`) that has two modes of operation: in the main mode it executes inteructions from the list (by calling `evalOne`), in the second mode it executes the call injection protocol by calling `funcCallStep` repeatedly until it either the protocol finishes, needs more input from the stack machine (to set call arguments) or fails. This approach has several benefits: - it is now possible to remove the goroutine we use to evaluate expression and the channel used to communicate with the Continue loop. - every time we resume the target to execute the call injection protocol we need to update several local variables to match the changed state of the target, this is now done at the top level of the evaluation loop instead of being hidden inside a recurisive evaluator - using runtime.Pin to pin addresses returned by an injected call would allow us to use a more natural evaluation order for function calls, which would solve some bugs #3310, allow users to inspect values returned by a call injection #1599 and allow implementing some other features #1465. Doing this with the recursive evaluator, while keeping backwards compatibility with versions of Go that do not have runtime.Pin is very hard. However after this change we can simply conditionally change how compileFunctionCall works and add some opcodes. * review round 1 * review round 2
2023-10-17 18:21:59 +00:00
package evalop
import (
"go/ast"
"go/constant"
"github.com/go-delve/delve/pkg/dwarf/godwarf"
)
// Op is a stack machine opcode
type Op interface {
depthCheck() (npop, npush int)
}
// PushCurg pushes the current goroutine on the stack.
type PushCurg struct {
}
func (*PushCurg) depthCheck() (npop, npush int) { return 0, 1 }
// PushFrameoff pushes the frame offset for the current frame on the stack.
type PushFrameoff struct {
}
func (*PushFrameoff) depthCheck() (npop, npush int) { return 0, 1 }
// PushThreadID pushes the ID of the current thread on the stack.
type PushThreadID struct {
}
func (*PushThreadID) depthCheck() (npop, npush int) { return 0, 1 }
// PushConst pushes a constant on the stack.
type PushConst struct {
Value constant.Value
}
func (*PushConst) depthCheck() (npop, npush int) { return 0, 1 }
// PushLocal pushes the local variable with the given name on the stack.
type PushLocal struct {
Name string
Frame int64
proc: use stack machine to evaluate expressions (#3508) * proc: use stack machine to evaluate expressions This commit splits expression evaluation into two parts. The first part (in pkg/proc/evalop/evalcompile.go) "compiles" as ast.Expr into a list of instructions (defined in pkg/proc/evalop/ops.go) for a stack machine (defined by `proc.(*evalStack)`). The second part is a stack machine (implemented by `proc.(*EvalScope).eval` and `proc.(*EvalScope).evalOne`) that has two modes of operation: in the main mode it executes inteructions from the list (by calling `evalOne`), in the second mode it executes the call injection protocol by calling `funcCallStep` repeatedly until it either the protocol finishes, needs more input from the stack machine (to set call arguments) or fails. This approach has several benefits: - it is now possible to remove the goroutine we use to evaluate expression and the channel used to communicate with the Continue loop. - every time we resume the target to execute the call injection protocol we need to update several local variables to match the changed state of the target, this is now done at the top level of the evaluation loop instead of being hidden inside a recurisive evaluator - using runtime.Pin to pin addresses returned by an injected call would allow us to use a more natural evaluation order for function calls, which would solve some bugs #3310, allow users to inspect values returned by a call injection #1599 and allow implementing some other features #1465. Doing this with the recursive evaluator, while keeping backwards compatibility with versions of Go that do not have runtime.Pin is very hard. However after this change we can simply conditionally change how compileFunctionCall works and add some opcodes. * review round 1 * review round 2
2023-10-17 18:21:59 +00:00
}
func (*PushLocal) depthCheck() (npop, npush int) { return 0, 1 }
// PushNil pushes an untyped nil on the stack.
type PushNil struct {
}
func (*PushNil) depthCheck() (npop, npush int) { return 0, 1 }
// PushRegister pushes the CPU register Regnum on the stack.
type PushRegister struct {
Regnum int
Regname string
}
func (*PushRegister) depthCheck() (npop, npush int) { return 0, 1 }
// PushPackageVar pushes a package variable on the stack.
type PushPackageVar struct {
PkgName, Name string // if PkgName == "" use current function's package
}
func (*PushPackageVar) depthCheck() (npop, npush int) { return 0, 1 }
// Select replaces the topmost stack variable v with v.Name.
type Select struct {
Name string
}
func (*Select) depthCheck() (npop, npush int) { return 1, 1 }
// TypeAssert replaces the topmost stack variable v with v.(DwarfType).
type TypeAssert struct {
DwarfType godwarf.Type
Node *ast.TypeAssertExpr
}
func (*TypeAssert) depthCheck() (npop, npush int) { return 1, 1 }
// PointerDeref replaces the topmost stack variable v with *v.
type PointerDeref struct {
Node *ast.StarExpr
}
func (*PointerDeref) depthCheck() (npop, npush int) { return 1, 1 }
// Unary applies the given unary operator to the topmost stack variable.
type Unary struct {
Node *ast.UnaryExpr
}
func (*Unary) depthCheck() (npop, npush int) { return 1, 1 }
// AddrOf replaces the topmost stack variable v with &v.
type AddrOf struct {
Node *ast.UnaryExpr
}
func (*AddrOf) depthCheck() (npop, npush int) { return 1, 1 }
// TypeCast replaces the topmost stack variable v with (DwarfType)(v).
type TypeCast struct {
DwarfType godwarf.Type
Node *ast.CallExpr
}
func (*TypeCast) depthCheck() (npop, npush int) { return 1, 1 }
// Reslice implements a reslice operation.
// If HasHigh is set it pops three variables, low, high and v, and pushes
// v[low:high].
// Otherwise it pops two variables, low and v, and pushes v[low:].
type Reslice struct {
HasHigh bool
Node *ast.SliceExpr
}
func (op *Reslice) depthCheck() (npop, npush int) {
if op.HasHigh {
return 3, 1
} else {
return 2, 1
}
}
// Index pops two variables, idx and v, and pushes v[idx].
type Index struct {
Node *ast.IndexExpr
}
func (*Index) depthCheck() (npop, npush int) { return 2, 1 }
2023-10-19 18:04:31 +00:00
// Jump looks at the topmost stack variable and if it satisfies the
proc: use stack machine to evaluate expressions (#3508) * proc: use stack machine to evaluate expressions This commit splits expression evaluation into two parts. The first part (in pkg/proc/evalop/evalcompile.go) "compiles" as ast.Expr into a list of instructions (defined in pkg/proc/evalop/ops.go) for a stack machine (defined by `proc.(*evalStack)`). The second part is a stack machine (implemented by `proc.(*EvalScope).eval` and `proc.(*EvalScope).evalOne`) that has two modes of operation: in the main mode it executes inteructions from the list (by calling `evalOne`), in the second mode it executes the call injection protocol by calling `funcCallStep` repeatedly until it either the protocol finishes, needs more input from the stack machine (to set call arguments) or fails. This approach has several benefits: - it is now possible to remove the goroutine we use to evaluate expression and the channel used to communicate with the Continue loop. - every time we resume the target to execute the call injection protocol we need to update several local variables to match the changed state of the target, this is now done at the top level of the evaluation loop instead of being hidden inside a recurisive evaluator - using runtime.Pin to pin addresses returned by an injected call would allow us to use a more natural evaluation order for function calls, which would solve some bugs #3310, allow users to inspect values returned by a call injection #1599 and allow implementing some other features #1465. Doing this with the recursive evaluator, while keeping backwards compatibility with versions of Go that do not have runtime.Pin is very hard. However after this change we can simply conditionally change how compileFunctionCall works and add some opcodes. * review round 1 * review round 2
2023-10-17 18:21:59 +00:00
// condition specified by When it jumps to the stack machine instruction at
// Target+1.
// If Pop is set the topmost stack variable is also popped.
type Jump struct {
When JumpCond
Pop bool
Target int
Node ast.Expr
}
func (jmpif *Jump) depthCheck() (npop, npush int) {
if jmpif.Pop {
return 1, 0
}
return 0, 0
}
// JumpCond specifies a condition for the Jump instruction.
type JumpCond uint8
const (
JumpIfFalse JumpCond = iota
JumpIfTrue
)
// Binary pops two variables from the stack, applies the specified binary
// operator to them and pushes the result back on the stack.
type Binary struct {
Node *ast.BinaryExpr
}
func (*Binary) depthCheck() (npop, npush int) { return 2, 1 }
// BoolToConst pops the topmost variable from the stack, which must be a
// boolean variable, and converts it to a constant.
type BoolToConst struct {
}
func (*BoolToConst) depthCheck() (npop, npush int) { return 1, 1 }
// Pop removes the topmost variable from the stack.
type Pop struct {
}
func (*Pop) depthCheck() (npop, npush int) { return 1, 0 }
// BuiltinCall pops len(Args) argument from the stack, calls the specified
// builtin on them and pushes the result back on the stack.
type BuiltinCall struct {
Name string
Args []ast.Expr
}
func (bc *BuiltinCall) depthCheck() (npop, npush int) {
return len(bc.Args), 1
}
// CallInjectionStart starts call injection by calling
// runtime.debugCallVn.
type CallInjectionStart struct {
id int // identifier for all the call injection instructions that belong to the same sequence, this only exists to make reading listings easier
HasFunc bool // target function already pushed on the stack
Node *ast.CallExpr
}
func (*CallInjectionStart) depthCheck() (npop, npush int) { return 0, 1 }
// CallInjectionSetTarget starts the call injection, after
// runtime.debugCallVn set up the stack for us, by copying the entry point
// of the function, setting the closure register and copying the receiver.
type CallInjectionSetTarget struct {
id int
}
func (*CallInjectionSetTarget) depthCheck() (npop, npush int) { return 1, 0 }
// CallInjectionCopyArg copies one argument for call injection.
type CallInjectionCopyArg struct {
id int
ArgNum int
ArgExpr ast.Expr
}
func (*CallInjectionCopyArg) depthCheck() (npop, npush int) { return 1, 0 }
// CallInjectionComplete resumes target execution so that the injected call can run.
type CallInjectionComplete struct {
id int
}
func (*CallInjectionComplete) depthCheck() (npop, npush int) { return 0, 1 }
// CallInjectionAllocString uses the call injection protocol to allocate the
// value of a string literal somewhere on the target's memory so that it can
// be assigned to a variable (or passed to a function).
// There are three phases to CallInjectionAllocString, distinguished by the
// Phase field. They must always appear in sequence in the program:
//
// CallInjectionAllocString{Phase: 0}
// CallInjectionAllocString{Phase: 1}
// CallInjectionAllocString{Phase: 2}
type CallInjectionAllocString struct {
Phase int
}
func (op *CallInjectionAllocString) depthCheck() (npop, npush int) { return 1, 1 }
// SetValue pops to variables from the stack, lhv and rhv, and sets lhv to
// rhv.
type SetValue struct {
lhe, Rhe ast.Expr
}
func (*SetValue) depthCheck() (npop, npush int) { return 2, 0 }