[lldb-dap] persistent assembly breakpoints (#148061)

Resolves #141955

- Adds data to breakpoints `Source` object, in order for assembly
breakpoints, which rely on a temporary `sourceReference` value, to be
able to resolve in future sessions like normal path+line breakpoints
- Adds optional `instructions_offset` parameter to `BreakpointResolver`

Author: eronnen

lldb/include/lldb/API/SBTarget.h

@@ -658,6 +658,14 @@ class LLDB_API SBTarget {
       lldb::LanguageType symbol_language,
       const SBFileSpecList &module_list, const SBFileSpecList &comp_unit_list);
 
+  lldb::SBBreakpoint BreakpointCreateByName(
+      const char *symbol_name,
+      uint32_t
+          name_type_mask, // Logical OR one or more FunctionNameType enum bits
+      lldb::LanguageType symbol_language, lldb::addr_t offset,
+      bool offset_is_insn_count, const SBFileSpecList &module_list,
+      const SBFileSpecList &comp_unit_list);
+
 #ifdef SWIG
   lldb::SBBreakpoint BreakpointCreateByNames(
       const char **symbol_name, uint32_t num_names,

lldb/include/lldb/Breakpoint/BreakpointResolver.h

@@ -45,9 +45,9 @@ class BreakpointResolver : public Searcher {
   ///   The breakpoint that owns this resolver.
   /// \param[in] resolverType
   ///   The concrete breakpoint resolver type for this breakpoint.
-  BreakpointResolver(const lldb::BreakpointSP &bkpt,
-                     unsigned char resolverType,
-                     lldb::addr_t offset = 0);
+  BreakpointResolver(const lldb::BreakpointSP &bkpt, unsigned char resolverType,
+                     lldb::addr_t offset = 0,
+                     bool offset_is_insn_count = false);
 
   /// The Destructor is virtual, all significant breakpoint resolvers derive
   /// from this class.
@@ -76,6 +76,7 @@ class BreakpointResolver : public Searcher {
   void SetOffset(lldb::addr_t offset);
 
   lldb::addr_t GetOffset() const { return m_offset; }
+  lldb::addr_t GetOffsetIsInsnCount() const { return m_offset_is_insn_count; }
 
   /// In response to this method the resolver scans all the modules in the
   /// breakpoint's target, and adds any new locations it finds.
@@ -220,6 +221,8 @@ class BreakpointResolver : public Searcher {
   lldb::BreakpointWP m_breakpoint; // This is the breakpoint we add locations to.
   lldb::addr_t m_offset;    // A random offset the user asked us to add to any
                             // breakpoints we set.
+  bool m_offset_is_insn_count; // Use the offset as an instruction count
+                               // instead of an address offset.
 
   // Subclass identifier (for llvm isa/dyn_cast)
   const unsigned char SubclassID;

lldb/include/lldb/Breakpoint/BreakpointResolverName.h

@@ -27,7 +27,7 @@ class BreakpointResolverName : public BreakpointResolver {
                          lldb::FunctionNameType name_type_mask,
                          lldb::LanguageType language,
                          Breakpoint::MatchType type, lldb::addr_t offset,
-                         bool skip_prologue);
+                         bool offset_is_insn_count, bool skip_prologue);
 
   // This one takes an array of names.  It is always MatchType = Exact.
   BreakpointResolverName(const lldb::BreakpointSP &bkpt, const char *names[],

lldb/include/lldb/Core/Disassembler.h

@@ -291,6 +291,8 @@ class InstructionList {
 
   size_t GetSize() const;
 
+  size_t GetTotalByteSize() const;
+
   uint32_t GetMaxOpcocdeByteSize() const;
 
   lldb::InstructionSP GetInstructionAtIndex(size_t idx) const;

lldb/include/lldb/Target/Target.h

@@ -18,6 +18,7 @@
 #include "lldb/Breakpoint/BreakpointList.h"
 #include "lldb/Breakpoint/BreakpointName.h"
 #include "lldb/Breakpoint/WatchpointList.h"
+#include "lldb/Core/Address.h"
 #include "lldb/Core/Architecture.h"
 #include "lldb/Core/Disassembler.h"
 #include "lldb/Core/ModuleList.h"
@@ -723,7 +724,7 @@ class Target : public std::enable_shared_from_this<Target>,
   lldb::BreakpointSP CreateBreakpoint(lldb::addr_t load_addr, bool internal,
                                       bool request_hardware);
 
-  // Use this to create a breakpoint from a load address and a module file spec
+  // Use this to create a breakpoint from a file address and a module file spec
   lldb::BreakpointSP CreateAddressInModuleBreakpoint(lldb::addr_t file_addr,
                                                      bool internal,
                                                      const FileSpec &file_spec,
@@ -752,8 +753,8 @@ class Target : public std::enable_shared_from_this<Target>,
       const FileSpecList *containingModules,
       const FileSpecList *containingSourceFiles, const char *func_name,
       lldb::FunctionNameType func_name_type_mask, lldb::LanguageType language,
-      lldb::addr_t offset, LazyBool skip_prologue, bool internal,
-      bool request_hardware);
+      lldb::addr_t offset, bool offset_is_insn_count, LazyBool skip_prologue,
+      bool internal, bool request_hardware);
 
   lldb::BreakpointSP
   CreateExceptionBreakpoint(enum lldb::LanguageType language, bool catch_bp,
@@ -1334,6 +1335,10 @@ class Target : public std::enable_shared_from_this<Target>,
                                const lldb_private::RegisterFlags &flags,
                                uint32_t byte_size);
 
+  llvm::Expected<lldb::DisassemblerSP>
+  ReadInstructions(const Address &start_addr, uint32_t count,
+                   const char *flavor_string = nullptr);
+
   // Target Stop Hooks
   class StopHook : public UserID {
   public:

lldb/packages/Python/lldbsuite/test/tools/lldb-dap/dap_server.py

@@ -107,24 +107,33 @@ def dump_dap_log(log_file):
 
 class Source(object):
     def __init__(
-        self, path: Optional[str] = None, source_reference: Optional[int] = None
+        self,
+        path: Optional[str] = None,
+        source_reference: Optional[int] = None,
+        raw_dict: Optional[dict[str, Any]] = None,
     ):
         self._name = None
         self._path = None
         self._source_reference = None
+        self._raw_dict = None
 
         if path is not None:
             self._name = os.path.basename(path)
             self._path = path
         elif source_reference is not None:
             self._source_reference = source_reference
+        elif raw_dict is not None:
+            self._raw_dict = raw_dict
         else:
             raise ValueError("Either path or source_reference must be provided")
 
     def __str__(self):
         return f"Source(name={self.name}, path={self.path}), source_reference={self.source_reference})"
 
     def as_dict(self):
+        if self._raw_dict is not None:
+            return self._raw_dict
+
         source_dict = {}
         if self._name is not None:
             source_dict["name"] = self._name
@@ -135,6 +144,19 @@ def as_dict(self):
         return source_dict
 
 
+class Breakpoint(object):
+    def __init__(self, obj):
+        self._breakpoint = obj
+
+    def is_verified(self):
+        """Check if the breakpoint is verified."""
+        return self._breakpoint.get("verified", False)
+
+    def source(self):
+        """Get the source of the breakpoint."""
+        return self._breakpoint.get("source", {})
+
+
 class NotSupportedError(KeyError):
     """Raised if a feature is not supported due to its capabilities."""
 
@@ -170,7 +192,7 @@ def __init__(
         self.initialized = False
         self.frame_scopes = {}
         self.init_commands = init_commands
-        self.resolved_breakpoints = {}
+        self.resolved_breakpoints: dict[str, Breakpoint] = {}
 
     @classmethod
     def encode_content(cls, s: str) -> bytes:
@@ -326,8 +348,8 @@ def _process_continued(self, all_threads_continued: bool):
     def _update_verified_breakpoints(self, breakpoints: list[Event]):
         for breakpoint in breakpoints:
             if "id" in breakpoint:
-                self.resolved_breakpoints[str(breakpoint["id"])] = breakpoint.get(
-                    "verified", False
+                self.resolved_breakpoints[str(breakpoint["id"])] = Breakpoint(
+                    breakpoint
                 )
 
     def send_packet(self, command_dict: Request, set_sequence=True):
@@ -484,7 +506,14 @@ def wait_for_breakpoints_to_be_verified(
             if breakpoint_event is None:
                 break
 
-        return [id for id in breakpoint_ids if id not in self.resolved_breakpoints]
+        return [
+            id
+            for id in breakpoint_ids
+            if (
+                id not in self.resolved_breakpoints
+                or not self.resolved_breakpoints[id].is_verified()
+            )
+        ]
 
     def wait_for_exited(self, timeout: Optional[float] = None):
         event_dict = self.wait_for_event("exited", timeout=timeout)

lldb/packages/Python/lldbsuite/test/tools/lldb-dap/lldbdap_testcase.py

@@ -59,24 +59,22 @@ def set_source_breakpoints(
         Each object in data is 1:1 mapping with the entry in lines.
         It contains optional location/hitCondition/logMessage parameters.
         """
-        response = self.dap_server.request_setBreakpoints(
-            Source(source_path), lines, data
+        return self.set_source_breakpoints_from_source(
+            Source(path=source_path), lines, data, wait_for_resolve
         )
-        if response is None or not response["success"]:
-            return []
-        breakpoints = response["body"]["breakpoints"]
-        breakpoint_ids = []
-        for breakpoint in breakpoints:
-            breakpoint_ids.append("%i" % (breakpoint["id"]))
-        if wait_for_resolve:
-            self.wait_for_breakpoints_to_resolve(breakpoint_ids)
-        return breakpoint_ids
 
     def set_source_breakpoints_assembly(
         self, source_reference, lines, data=None, wait_for_resolve=True
+    ):
+        return self.set_source_breakpoints_from_source(
+            Source(source_reference=source_reference), lines, data, wait_for_resolve
+        )
+
+    def set_source_breakpoints_from_source(
+        self, source: Source, lines, data=None, wait_for_resolve=True
     ):
         response = self.dap_server.request_setBreakpoints(
-            Source(source_reference=source_reference),
+            source,
             lines,
             data,
         )

lldb/source/API/SBTarget.cpp

@@ -766,16 +766,19 @@ SBBreakpoint SBTarget::BreakpointCreateByName(const char *symbol_name,
     const bool hardware = false;
     const LazyBool skip_prologue = eLazyBoolCalculate;
     const lldb::addr_t offset = 0;
+    const bool offset_is_insn_count = false;
     if (module_name && module_name[0]) {
       FileSpecList module_spec_list;
       module_spec_list.Append(FileSpec(module_name));
       sb_bp = target_sp->CreateBreakpoint(
           &module_spec_list, nullptr, symbol_name, eFunctionNameTypeAuto,
-          eLanguageTypeUnknown, offset, skip_prologue, internal, hardware);
+          eLanguageTypeUnknown, offset, offset_is_insn_count, skip_prologue,
+          internal, hardware);
     } else {
       sb_bp = target_sp->CreateBreakpoint(
           nullptr, nullptr, symbol_name, eFunctionNameTypeAuto,
-          eLanguageTypeUnknown, offset, skip_prologue, internal, hardware);
+          eLanguageTypeUnknown, offset, offset_is_insn_count, skip_prologue,
+          internal, hardware);
     }
   }
 
@@ -811,6 +814,17 @@ lldb::SBBreakpoint SBTarget::BreakpointCreateByName(
     const SBFileSpecList &comp_unit_list) {
   LLDB_INSTRUMENT_VA(this, symbol_name, name_type_mask, symbol_language,
                      module_list, comp_unit_list);
+  return BreakpointCreateByName(symbol_name, name_type_mask, symbol_language, 0,
+                                false, module_list, comp_unit_list);
+}
+
+lldb::SBBreakpoint SBTarget::BreakpointCreateByName(
+    const char *symbol_name, uint32_t name_type_mask,
+    LanguageType symbol_language, lldb::addr_t offset,
+    bool offset_is_insn_count, const SBFileSpecList &module_list,
+    const SBFileSpecList &comp_unit_list) {
+  LLDB_INSTRUMENT_VA(this, symbol_name, name_type_mask, symbol_language, offset,
+                     offset_is_insn_count, module_list, comp_unit_list);
 
   SBBreakpoint sb_bp;
   if (TargetSP target_sp = GetSP();
@@ -821,7 +835,8 @@ lldb::SBBreakpoint SBTarget::BreakpointCreateByName(
     std::lock_guard<std::recursive_mutex> guard(target_sp->GetAPIMutex());
     FunctionNameType mask = static_cast<FunctionNameType>(name_type_mask);
     sb_bp = target_sp->CreateBreakpoint(module_list.get(), comp_unit_list.get(),
-                                        symbol_name, mask, symbol_language, 0,
+                                        symbol_name, mask, symbol_language,
+                                        offset, offset_is_insn_count,
                                         skip_prologue, internal, hardware);
   }
 
@@ -1955,29 +1970,10 @@ lldb::SBInstructionList SBTarget::ReadInstructions(lldb::SBAddress base_addr,
 
   if (TargetSP target_sp = GetSP()) {
     if (Address *addr_ptr = base_addr.get()) {
-      DataBufferHeap data(
-          target_sp->GetArchitecture().GetMaximumOpcodeByteSize() * count, 0);
-      bool force_live_memory = true;
-      lldb_private::Status error;
-      lldb::addr_t load_addr = LLDB_INVALID_ADDRESS;
-      const size_t bytes_read =
-          target_sp->ReadMemory(*addr_ptr, data.GetBytes(), data.GetByteSize(),
-                                error, force_live_memory, &load_addr);
-
-      const bool data_from_file = load_addr == LLDB_INVALID_ADDRESS;
-      if (!flavor_string || flavor_string[0] == '\0') {
-        // FIXME - we don't have the mechanism in place to do per-architecture
-        // settings.  But since we know that for now we only support flavors on
-        // x86 & x86_64,
-        const llvm::Triple::ArchType arch =
-            target_sp->GetArchitecture().GetTriple().getArch();
-        if (arch == llvm::Triple::x86 || arch == llvm::Triple::x86_64)
-          flavor_string = target_sp->GetDisassemblyFlavor();
+      if (llvm::Expected<DisassemblerSP> disassembler =
+              target_sp->ReadInstructions(*addr_ptr, count, flavor_string)) {
+        sb_instructions.SetDisassembler(*disassembler);
       }
-      sb_instructions.SetDisassembler(Disassembler::DisassembleBytes(
-          target_sp->GetArchitecture(), nullptr, flavor_string,
-          target_sp->GetDisassemblyCPU(), target_sp->GetDisassemblyFeatures(),
-          *addr_ptr, data.GetBytes(), bytes_read, count, data_from_file));
     }
   }
 

lldb/source/Breakpoint/BreakpointResolver.cpp

@@ -42,9 +42,9 @@ const char *BreakpointResolver::g_ty_to_name[] = {"FileAndLine", "Address",
 
 const char *BreakpointResolver::g_option_names[static_cast<uint32_t>(
     BreakpointResolver::OptionNames::LastOptionName)] = {
-    "AddressOffset", "Exact",     "FileName",     "Inlines",     "Language",
-    "LineNumber",    "Column",    "ModuleName",   "NameMask",    "Offset",
-    "PythonClass",   "Regex",     "ScriptArgs",   "SectionName", "SearchDepth",
+    "AddressOffset", "Exact",      "FileName",   "Inlines",     "Language",
+    "LineNumber",    "Column",     "ModuleName", "NameMask",    "Offset",
+    "PythonClass",   "Regex",      "ScriptArgs", "SectionName", "SearchDepth",
     "SkipPrologue",  "SymbolNames"};
 
 const char *BreakpointResolver::ResolverTyToName(enum ResolverTy type) {
@@ -65,8 +65,10 @@ BreakpointResolver::NameToResolverTy(llvm::StringRef name) {
 
 BreakpointResolver::BreakpointResolver(const BreakpointSP &bkpt,
                                        const unsigned char resolverTy,
-                                       lldb::addr_t offset)
-    : m_breakpoint(bkpt), m_offset(offset), SubclassID(resolverTy) {}
+                                       lldb::addr_t offset,
+                                       bool offset_is_insn_count)
+    : m_breakpoint(bkpt), m_offset(offset),
+      m_offset_is_insn_count(offset_is_insn_count), SubclassID(resolverTy) {}
 
 BreakpointResolver::~BreakpointResolver() = default;
 
@@ -364,7 +366,32 @@ void BreakpointResolver::AddLocation(SearchFilter &filter,
 
 BreakpointLocationSP BreakpointResolver::AddLocation(Address loc_addr,
                                                      bool *new_location) {
-  loc_addr.Slide(m_offset);
+  if (m_offset_is_insn_count) {
+    Target &target = GetBreakpoint()->GetTarget();
+    llvm::Expected<DisassemblerSP> expected_instructions =
+        target.ReadInstructions(loc_addr, m_offset);
+    if (!expected_instructions) {
+      LLDB_LOG_ERROR(GetLog(LLDBLog::Breakpoints),
+                     expected_instructions.takeError(),
+                     "error: Unable to read instructions at address 0x{0:x}",
+                     loc_addr.GetLoadAddress(&target));
+      return BreakpointLocationSP();
+    }
+
+    const DisassemblerSP instructions = *expected_instructions;
+    if (!instructions ||
+        instructions->GetInstructionList().GetSize() != m_offset) {
+      LLDB_LOG(GetLog(LLDBLog::Breakpoints),
+               "error: Unable to read {0} instructions at address 0x{1:x}",
+               m_offset, loc_addr.GetLoadAddress(&target));
+      return BreakpointLocationSP();
+    }
+
+    loc_addr.Slide(instructions->GetInstructionList().GetTotalByteSize());
+  } else {
+    loc_addr.Slide(m_offset);
+  }
+
   return GetBreakpoint()->AddLocation(loc_addr, new_location);
 }
 

lldb/source/Breakpoint/BreakpointResolverAddress.cpp

@@ -133,6 +133,11 @@ Searcher::CallbackReturn BreakpointResolverAddress::SearchCallback(
           Address tmp_address;
           if (module_sp->ResolveFileAddress(m_addr.GetOffset(), tmp_address))
             m_addr = tmp_address;
+          else
+            return Searcher::eCallbackReturnStop;
+        } else {
+          // If we didn't find the module, then we can't resolve the address.
+          return Searcher::eCallbackReturnStop;
         }
       }