Change stack crawling to not be based on unwinding

We always want to crawl the entire stack, and it's
possible to determine the extents of the stack, so just
do a scan over the entire memory range.

Also, change the way the interpreter keeps track of its roots;
we don't really need to associate the roots with a specific
interpreter frame.

This should hopefully clear up the weirdness about libunwind
trying to unwind through the pthreads assembly code, and potentially
also make stack crawling faster.

docs/INSTALLING.md

@@ -66,8 +66,8 @@ git clone git://git.sv.gnu.org/libunwind.git libunwind-trunk
 mkdir libunwind-trunk-install
 cd libunwind-trunk
 git checkout 65ac867416
-# disable shared libraries because we'll be installing this in a place that the loader can't find it:
 autoreconf -i
+# disable shared libraries because we'll be installing this in a place that the loader can't find it:
 ./configure --prefix=$HOME/pyston_deps/libunwind-trunk-install --enable-shared=0
 make -j4
 make install

src/codegen/llvm_interpreter.cpp

@@ -29,6 +29,7 @@
 #include "codegen/irgen/util.h"
 #include "core/common.h"
 #include "core/stats.h"
+#include "core/thread_utils.h"
 #include "core/util.h"
 
 //#undef VERBOSITY
@@ -206,25 +207,23 @@ static void set(SymMap& symbols, const llvm::BasicBlock::iterator& it, Val v) {
     //#define SET(v) symbols.insert(std::make_pair(static_cast<llvm::Value*>(&(*it)), Val(v)))
 }
 
-static std::unordered_map<void*, const SymMap*> interpreter_roots;
 static std::unordered_map<void*, llvm::Instruction*> cur_instruction_map;
 
-void gatherInterpreterRootsForFrame(GCVisitor* visitor, void* frame_ptr) {
-    auto it = interpreter_roots.find(frame_ptr);
-    if (it == interpreter_roots.end()) {
-        printf("%p is not an interpreter frame; they are", frame_ptr);
-        for (const auto& p2 : interpreter_roots) {
-            printf(" %p", p2.first);
-        }
-        printf("\n");
-        abort();
-    }
+typedef std::vector<const SymMap*> root_stack_t;
+threading::PerThreadSet<root_stack_t> root_stack_set;
+threading::PerThread<root_stack_t> thread_local root_stack(&root_stack_set);
 
-    // printf("Gathering roots for frame %p\n", frame_ptr);
-    const SymMap* symbols = it->second;
+void gatherInterpreterRoots(GCVisitor* visitor) {
+    // In theory this lock should be superfluous since we should only call this
+    // inside a sequential section, but lock it anyway:
+    threading::LockedRegion _lock(&root_stack_set.lock);
 
-    for (const auto& p2 : *symbols) {
-        visitor->visitPotential(p2.second.o);
+    for (auto& p : root_stack_set.map) {
+        for (const SymMap* sym_map : *p.second) {
+            for (const auto& p2 : *sym_map) {
+                visitor->visitPotential(p2.second.o);
+            }
+        }
     }
 }
 
@@ -236,8 +235,7 @@ public:
     constexpr UnregisterHelper(void* frame_ptr) : frame_ptr(frame_ptr) {}
 
     ~UnregisterHelper() {
-        assert(interpreter_roots.count(frame_ptr));
-        interpreter_roots.erase(frame_ptr);
+        root_stack.value.pop_back();
 
         assert(cur_instruction_map.count(frame_ptr));
         cur_instruction_map.erase(frame_ptr);
@@ -282,7 +280,7 @@ Box* interpretFunction(llvm::Function* f, int nargs, Box* arg1, Box* arg2, Box*
     SymMap symbols;
 
     void* frame_ptr = __builtin_frame_address(0);
-    interpreter_roots[frame_ptr] = &symbols;
+    root_stack.value.push_back(&symbols);
     UnregisterHelper helper(frame_ptr);
 
     int arg_num = -1;

src/codegen/llvm_interpreter.h

@@ -27,7 +27,7 @@ class LineInfo;
 
 Box* interpretFunction(llvm::Function* f, int nargs, Box* arg1, Box* arg2, Box* arg3, Box** args);
 
-void gatherInterpreterRootsForFrame(GCVisitor* visitor, void* frame_ptr);
+void gatherInterpreterRoots(GCVisitor* visitor);
 const LineInfo* getLineInfoForInterpretedFrame(void* frame_ptr);
 }
 

src/core/common.h

@@ -25,6 +25,8 @@
 #include <unordered_set>
 #include <vector>
 
+#define STACK_GROWS_DOWN 1
+
 #define _STRINGIFY(N) #N
 #define STRINGIFY(N) _STRINGIFY(N)
 

src/core/thread_utils.h

+// Copyright (c) 2014 Dropbox, Inc.
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//    http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#ifndef PYSTON_CORE_THREADUTILS_H
+#define PYSTON_CORE_THREADUTILS_H
+
+#include <pthread.h>
+
+namespace pyston {
+namespace threading {
+
+class LockedRegion {
+private:
+    pthread_mutex_t* mutex;
+
+public:
+    LockedRegion(pthread_mutex_t* mutex) : mutex(mutex) { pthread_mutex_lock(mutex); }
+    ~LockedRegion() { pthread_mutex_unlock(mutex); }
+};
+
+
+
+template <typename T> class PerThreadSet {
+public:
+    pthread_mutex_t lock = PTHREAD_MUTEX_INITIALIZER;
+    std::unordered_map<pthread_t, T*> map;
+};
+
+template <typename T> class PerThread {
+private:
+    PerThreadSet<T>* set;
+    pthread_t self;
+
+public:
+    T value;
+
+    PerThread(PerThreadSet<T>* set) : set(set), self(pthread_self()) {
+        LockedRegion _lock(&set->lock);
+
+        set->map[self] = &value;
+    }
+
+    ~PerThread() {
+        LockedRegion _lock(&set->lock);
+
+        assert(set->map.count(self) == 1);
+        set->map.erase(self);
+    }
+};
+
+} // namespace threading
+} // namespace pyston
+
+#endif

src/core/threading.cpp

@@ -23,6 +23,7 @@
 
 #include "core/common.h"
 #include "core/options.h"
+#include "core/thread_utils.h"
 
 extern "C" int start_thread(void* arg);
 
@@ -39,17 +40,6 @@ int tgkill(int tgid, int tid, int sig) {
     return syscall(SYS_tgkill, tgid, tid, sig);
 }
 
-class LockedRegion {
-private:
-    pthread_mutex_t* mutex;
-
-public:
-    LockedRegion(pthread_mutex_t* mutex) : mutex(mutex) { pthread_mutex_lock(mutex); }
-    ~LockedRegion() { pthread_mutex_unlock(mutex); }
-};
-
-
-
 // Certain thread examination functions won't be valid for a brief
 // period while a thread is starting up.
 // To handle this, track the number of threads in an uninitialized state,
@@ -62,9 +52,18 @@ struct ThreadStartArgs {
 };
 
 static pthread_mutex_t threading_lock = PTHREAD_MUTEX_INITIALIZER;
-static std::unordered_set<pid_t> current_threads;
+struct ThreadInfo {
+    // "bottom" in the sense of a stack, which in a down-growing stack is the highest address:
+    void* stack_bottom;
+    pthread_t pthread_id;
+};
+static std::unordered_map<pid_t, ThreadInfo> current_threads;
 
-static std::atomic<int> signals_waiting(0);
+void* getStackBottom() {
+    return current_threads[gettid()].stack_bottom;
+}
+
+static int signals_waiting(0);
 static std::vector<ThreadState> thread_states;
 std::vector<ThreadState> getAllThreadStates() {
     // TODO need to prevent new threads from starting,
@@ -89,7 +88,8 @@ std::vector<ThreadState> getAllThreadStates() {
 
     pid_t tgid = getpid();
     pid_t mytid = gettid();
-    for (pid_t tid : current_threads) {
+    for (auto& pair : current_threads) {
+        pid_t tid = pair.first;
         if (tid == mytid)
             continue;
         tgkill(tgid, tid, SIGUSR2);
@@ -119,12 +119,19 @@ static void _thread_context_dump(int signum, siginfo_t* info, void* _context) {
         printf("old rip: 0x%lx\n", context->uc_mcontext.gregs[REG_RIP]);
     }
 
-    thread_states.push_back(ThreadState(tid, context));
-    signals_waiting--; // atomic on std::atomic
+#if STACK_GROWS_DOWN
+    void* stack_start = (void*)context->uc_mcontext.gregs[REG_RSP];
+    void* stack_end = current_threads[tid].stack_bottom;
+#else
+    void* stack_start = current_threads[tid].stack_bottom;
+    void* stack_end = (void*)(context->uc_mcontext.gregs[REG_RSP] + sizeof(void*));
+#endif
+    assert(stack_start < stack_end);
+    thread_states.push_back(ThreadState(tid, context, stack_start, stack_end));
+    signals_waiting--;
 }
 
 static void* _thread_start(void* _arg) {
-    pid_t tid = gettid();
     ThreadStartArgs* arg = static_cast<ThreadStartArgs*>(_arg);
     auto start_func = arg->start_func;
     Box* arg1 = arg->arg1;
@@ -135,11 +142,33 @@ static void* _thread_start(void* _arg) {
     {
         LockedRegion _lock(&threading_lock);
 
-        current_threads.insert(tid);
+        pid_t tid = gettid();
+        pthread_t current_thread = pthread_self();
+
+        pthread_attr_t thread_attrs;
+        int code = pthread_getattr_np(current_thread, &thread_attrs);
+        RELEASE_ASSERT(code == 0, "");
+
+        void* stack_start;
+        size_t stack_size;
+        code = pthread_attr_getstack(&thread_attrs, &stack_start, &stack_size);
+        RELEASE_ASSERT(code == 0, "");
+
+        pthread_attr_destroy(&thread_attrs);
+
+        current_threads[tid] = ThreadInfo {
+#if STACK_GROWS_DOWN
+            .stack_bottom = static_cast<char*>(stack_start) + stack_size,
+#else
+            .stack_bottom = stack_start,
+#endif
+            .pthread_id = current_thread,
+        };
+
         num_starting_threads--;
 
         if (VERBOSITY() >= 2)
-            printf("child initialized; tid=%d\n", tid);
+            printf("child initialized; tid=%d\n", gettid());
     }
 
     threading::GLReadRegion _glock;
@@ -149,9 +178,9 @@ static void* _thread_start(void* _arg) {
     {
         LockedRegion _lock(&threading_lock);
 
-        current_threads.erase(tid);
+        current_threads.erase(gettid());
         if (VERBOSITY() >= 2)
-            printf("thread tid=%d exited\n", tid);
+            printf("thread tid=%d exited\n", gettid());
     }
 
     return rtn;
@@ -167,7 +196,7 @@ intptr_t start_thread(void* (*start_func)(Box*, Box*, Box*), Box* arg1, Box* arg
 
     pthread_t thread_id;
     int code = pthread_create(&thread_id, NULL, &_thread_start, args);
-    assert(code == 0);
+    RELEASE_ASSERT(code == 0, "");
     if (VERBOSITY() >= 2)
         printf("pthread thread_id: 0x%lx\n", thread_id);
 
@@ -175,6 +204,50 @@ intptr_t start_thread(void* (*start_func)(Box*, Box*, Box*), Box* arg1, Box* arg
     return thread_id;
 }
 
+// from https://www.sourceware.org/ml/guile/2000-07/msg00214.html
+static void* find_stack() {
+    FILE* input;
+    char* line;
+    char* s;
+    size_t len;
+    char hex[9];
+    void* start;
+    void* end;
+
+    int dummy;
+
+    input = fopen("/proc/self/maps", "r");
+    if (input == NULL)
+        return NULL;
+
+    len = 0;
+    line = NULL;
+    while (getline(&line, &len, input) != -1) {
+        s = strchr(line, '-');
+        if (s == NULL)
+            return NULL;
+        *s++ = '\0';
+
+        start = (void*)strtoul(line, NULL, 16);
+        end = (void*)strtoul(s, NULL, 16);
+
+        if ((void*)&dummy >= start && (void*)&dummy <= end) {
+            free(line);
+            fclose(input);
+
+#if STACK_GROWS_DOWN
+            return end;
+#else
+            return start;
+#endif
+        }
+    }
+
+    free(line);
+    fclose(input);
+    return NULL; /* not found =^P */
+}
+
 intptr_t call_frame_base;
 void registerMainThread() {
     LockedRegion _lock(&threading_lock);
@@ -184,7 +257,9 @@ void registerMainThread() {
     // call_frame_base = (intptr_t)::start_thread;
     call_frame_base = (intptr_t)_thread_start;
 
-    current_threads.insert(gettid());
+    current_threads[gettid()] = ThreadInfo{
+        .stack_bottom = find_stack(), .pthread_id = pthread_self(),
+    };
 
     struct sigaction act;
     act.sa_flags = SA_SIGINFO;

src/core/threading.h

@@ -37,7 +37,10 @@ struct ThreadState {
     pid_t tid; // useful mostly for debugging
     ucontext_t ucontext;
 
-    ThreadState(pid_t tid, ucontext_t* ucontext) : tid(tid) {
+    void* stack_start, *stack_end;
+
+    ThreadState(pid_t tid, ucontext_t* ucontext, void* stack_start, void* stack_end)
+        : tid(tid), stack_start(stack_start), stack_end(stack_end) {
         memcpy(&this->ucontext, ucontext, sizeof(ucontext_t));
         this->ucontext.uc_mcontext.fpregs = &this->ucontext.__fpregs_mem;
     }
@@ -47,6 +50,10 @@ struct ThreadState {
 // as a corollary, this thread is very much not thread safe.
 std::vector<ThreadState> getAllThreadStates();
 
+// Get the stack "bottom" (first pushed data; for stacks that grow down, will
+// be the highest address).
+void* getStackBottom();
+
 #define THREADING_USE_GIL 1
 #define THREADING_USE_GRWL 0
 #define THREADING_SAFE_DATASTRUCTURES THREADING_USE_GRWL

src/gc/root_finder.cpp

@@ -14,14 +14,11 @@
 
 #include "gc/root_finder.h"
 
-#define UNW_LOCAL_ONLY
-#include <libunwind.h>
-
-#include <cstring>
-#include <setjmp.h>
+#include <cassert>
 #include <cstdio>
 #include <cstdlib>
-#include <cassert>
+#include <cstring>
+#include <setjmp.h>
 #include <vector>
 
 #include "codegen/codegen.h"
@@ -43,129 +40,46 @@ namespace gc {
 void collectRoots(void* start, void* end, TraceStack* stack) {
     assert(start <= end);
 
-    void** cur = (void**)start;
-    while (cur < end) {
-        void* p = global_heap.getAllocationFromInteriorPointer(*cur);
-        if (p)
-            stack->push(p);
-        cur++;
-    }
+    TraceStackGCVisitor(stack).visitPotentialRange((void**)start, (void**)end);
 }
 
 
-static void _unwindStack(unw_cursor_t* cursor, TraceStack* stack) {
-    TraceStackGCVisitor visitor(stack);
-
-    unw_word_t ip, sp, bp;
-#ifndef NVALGRIND
-    if (RUNNING_ON_VALGRIND) {
-        memset(&ip, 0, sizeof(ip));
-        memset(&sp, 0, sizeof(sp));
-        memset(&bp, 0, sizeof(bp));
-    }
-#endif
-
-
-    int code;
-    while (true) {
-        int code = unw_step(cursor);
-        // Negative codes are errors, zero means that there isn't a new frame.
-        RELEASE_ASSERT(code >= 0 && "something broke unwinding!", "%d '%s'", code, unw_strerror(code));
-        RELEASE_ASSERT(code != 0, "didn't get to the top of the stack!");
-
-        unw_get_reg(cursor, UNW_REG_IP, &ip);
-        unw_get_reg(cursor, UNW_REG_SP, &sp);
-        unw_get_reg(cursor, UNW_TDEP_BP, &bp);
-
-        void* cur_sp = (void*)sp;
-        void* cur_bp = (void*)bp;
-
-        // std::string name = g.func_addr_registry.getFuncNameAtAddress((void*)ip, true);
-
-        unw_proc_info_t pip;
-        unw_get_proc_info(cursor, &pip);
-
-        // if (VERBOSITY()) printf("ip = 0x%lx (start_ip = 0x%lx), stack = [%p, %p)\n", (long) ip, pip.start_ip, cur_sp,
-        // cur_bp);
-
-        if (pip.start_ip == (uintptr_t)&__libc_start_main) {
-            break;
-        }
-
-        if (pip.start_ip == (intptr_t)interpretFunction) {
-            // TODO Do we still need to crawl the interpreter itself?
-            gatherInterpreterRootsForFrame(&visitor, cur_bp);
-        }
-
-        collectRoots(cur_sp, (char*)cur_bp, stack);
-
-        if (pip.start_ip == threading::call_frame_base) {
-            break;
-        }
-
-        if (cur_bp == NULL) {
-            // TODO I think this indicates an unwind mistake by libunwind?  Not sure.
-            // But if it returns cur_bp=NULL, this is probably just a thread where libunwind
-            // didn't reconstruct the call stack exactly the way we thought.
-            // TODO we probably don't need to do any unwinding here at all; we can just track
-            // the stack min and max for every thread.
-            break;
-        }
-    }
-}
-
 void collectOtherThreadsStacks(TraceStack* stack) {
     std::vector<threading::ThreadState> threads = threading::getAllThreadStates();
 
-    // unw_addr_space_t as = getOtherAddrSpace();
-
     for (threading::ThreadState& tstate : threads) {
-        unw_cursor_t cursor;
-        // int code = unw_init_remote(&cursor, as, &tstate);
-        int code = unw_init_local(&cursor, (ucontext_t*)&tstate.ucontext);
-        assert(code == 0);
-
-        // printf("Collecting thread %d\n", tstate.tid);
-
+        collectRoots(tstate.stack_start, tstate.stack_end, stack);
         collectRoots(&tstate.ucontext, (&tstate.ucontext) + 1, stack);
-
-        _unwindStack(&cursor, stack);
     }
 }
 
 static void collectLocalStack(TraceStack* stack) {
-    unw_cursor_t cursor;
-    unw_context_t uc;
-
     // force callee-save registers onto the stack:
     // Actually, I feel like this is pretty brittle:
     // collectLocalStack itself is allowed to save the callee-save registers
     // on its own stack.
     jmp_buf registers __attribute__((aligned(sizeof(void*))));
 
-#ifndef NVALGRIND
-    if (RUNNING_ON_VALGRIND) {
-        memset(&registers, 0, sizeof(registers));
-        memset(&cursor, 0, sizeof(cursor));
-        memset(&uc, 0, sizeof(uc));
-    }
-#endif
-
     setjmp(registers);
 
     assert(sizeof(registers) % 8 == 0);
     // void* stack_bottom = __builtin_frame_address(0);
-    collectRoots(&registers, &registers + 1, stack);
-
-    unw_getcontext(&uc);
-    unw_init_local(&cursor, &uc);
+    collectRoots(&registers, (&registers) + 1, stack);
 
-    _unwindStack(&cursor, stack);
+    void* stack_bottom = threading::getStackBottom();
+#if STACK_GROWS_DOWN
+    collectRoots(&registers, stack_bottom, stack);
+#else
+    collectRoots(stack_bottom, &registers + 1, stack);
+#endif
 }
 
 void collectStackRoots(TraceStack* stack) {
     collectLocalStack(stack);
     collectOtherThreadsStacks(stack);
+
+    TraceStackGCVisitor visitor(stack);
+    gatherInterpreterRoots(&visitor);
 }
 }
 }

test/unittests/unittests.h

@@ -24,6 +24,12 @@ namespace pyston {
 class PystonTestEnvironment : public testing::Environment {
     void SetUp() override {
         threading::registerMainThread();
+
+        threading::acquireGLRead();
+    }
+
+    void TearDown() override {
+        threading::releaseGLRead();
     }
 };