0.7.1 malloc(): unsorted double linked list corrupted

[JupiterRider]

Describe the bug malloc(): unsorted double linked list corrupted

List relevant hardware/software information

• Linux Distribution: Archlinux
• MangoHud version: 0.7.1
• GPU: Intel HD

To Reproduce Steps to reproduce the behavior: Run an application with the --dlsym parameter. For example supertux which links sdl2: mangohud --dlsym supertux2

Additional context When downgrading to 0.7.0 this error doesn't happen.

[flightlessmango]

Does it happen without a config file?

[JupiterRider]

@flightlessmango This happens with and without config file.

[flightlessmango]

Works fine for me

[flightlessmango]

Assuming this has been resolved since 0.7.1 unless more information is provided

[Torston420]

Bug specifically affects apps that require --dlsym to hook, seems intel specific bisected to commit 9411963ad907738f24a8286c0fee6e7f8eccb284 still present on latest commit I am on Arc A770

[flightlessmango]

@Torston420 It would be great if you could get a gdb backtrace since I can't repro it locally

[Torston420]

#0  0x00007ffff68ab32c in ?? () from /usr/lib/libc.so.6
#1  0x00007ffff685a6c8 in raise () from /usr/lib/libc.so.6
#2  0x00007ffff68424b8 in abort () from /usr/lib/libc.so.6
#3  0x00007ffff6843395 in ?? () from /usr/lib/libc.so.6
#4  0x00007ffff68b52a7 in ?? () from /usr/lib/libc.so.6
#5  0x00007ffff68b87ec in ?? () from /usr/lib/libc.so.6
#6  0x00007ffff68b97ed in malloc () from /usr/lib/libc.so.6
#7  0x00007ffff6893f1a in _IO_file_doallocate () from /usr/lib/libc.so.6
#8  0x00007ffff68a2c89 in _IO_doallocbuf () from /usr/lib/libc.so.6
#9  0x00007ffff68a0a65 in _IO_file_underflow () from /usr/lib/libc.so.6
#10 0x00007ffff68a2d2f in _IO_default_uflow () from /usr/lib/libc.so.6
#11 0x00007ffff6895abb in _IO_getline_info () from /usr/lib/libc.so.6
#12 0x00007ffff6894860 in fgets () from /usr/lib/libc.so.6
#13 0x00007ffff790c5df in Intel::intel_gpu_thread() () from /usr/local/lib/mangohud/libMangoHud_opengl.so
#14 0x00007ffff7c99703 in execute_native_thread_routine ()
   from /usr/local/lib/mangohud/libMangoHud_opengl.so
#15 0x00007ffff68a955a in ?? () from /usr/lib/libc.so.6
#16 0x00007ffff6926a3c in ?? () from /usr/lib/libc.so.6

[flightlessmango]

@Torston420 can you try this patch?

diff --git a/src/intel.cpp b/src/intel.cpp
index d61fa56..b010344 100644
--- a/src/intel.cpp
+++ b/src/intel.cpp
@@ -10,6 +10,9 @@ void Intel::intel_gpu_thread(){
     else
         intel_gpu_top = popen("intel_gpu_top -J -s 500", "r");

+    if (intel_gpu_top == NULL)
+        SPDLOG_INFO("popen failed on '{}'", "intel_gpu_top");
+
     int num_line = 0;
     std::string buf;
     int num_iterations = 0;

[Torston420]

#13 0x00007ffff790c5cf in Intel::intel_gpu_thread() () from /usr/local/lib/mangohud/libMangoHud_opengl.so
#14 0x00007ffff7c99753 in execute_native_thread_routine ()
   from /usr/local/lib/mangohud/libMangoHud_opengl.so

otherwise no change

[flightlessmango]

This patch adds some more verbose output to exit code of intel_gpu_top Can run this and see what it returns please?

index d61fa56..b08e27a 100644
--- a/src/intel.cpp
+++ b/src/intel.cpp
@@ -54,16 +54,21 @@ void Intel::intel_gpu_thread(){
             break;
     }

-    int exitcode = pclose(intel_gpu_top) / 256;
-    if (exitcode > 0){
-        if (exitcode == 127)
+    int status = pclose(intel_gpu_top);
+    if (WIFEXITED(status)) {
+        int exitcode = WEXITSTATUS(status);
+        if (exitcode != 0) {
+            SPDLOG_INFO("intel_gpu_top exited with status {}", exitcode);
+            if (exitcode == 127) {
         SPDLOG_INFO("Failed to open '{}'", "intel_gpu_top");
-
-        if (exitcode == 1)
+            } else if (exitcode == 1) {
         SPDLOG_INFO("Missing permissions for '{}'", "intel_gpu_top");
-
+            }
         SPDLOG_INFO("Disabling gpu_stats");
         HUDElements.params->enabled[OVERLAY_PARAM_ENABLED_gpu_stats] = false;
+        }
+    } else if (WIFSIGNALED(status)) {
+        SPDLOG_INFO("intel_gpu_top killed by signal {}", WTERMSIG(status));
     }
 }

[Torston420]

malloc(): unsorted double linked list corrupted
fish: Job 1, 'mangohud --dlsym openrct2' terminated by signal SIGABRT (Abort)

#13 0x00007ffff7911baf in Intel::intel_gpu_thread() () from /usr/local/lib/mangohud/libMangoHud_opengl.so
#14 0x00007ffff7ca7863 in execute_native_thread_routine ()
   from /usr/local/lib/mangohud/libMangoHud_opengl.so

[Torston420]

openrct2 is the app I'm using for these tests if I use gzdoom w/ OpenGL, I get an app hang instead, killing causes SIGABRT with the same error leaving a backtrace here since its the same offending commit, hope this helps in some way

#0  0x00007ffff5aa5ebe in ??? () at /usr/lib/libc.so.6
#1  0x00007ffff5aab0e3 in ??? () at /usr/lib/libc.so.6
#2  0x00007ffff7ca78d8 in std::thread::join() () at /usr/local/lib/mangohud/libMangoHud_opengl.so
#3  0x00007ffff788bccb in init_gpu_stats(unsigned int&, unsigned int, overlay_params&) ()
    at /usr/local/lib/mangohud/libMangoHud_opengl.so
#4  0x00007ffff785e684 in MangoHud::GL::imgui_create(void*, MangoHud::GL::gl_wsi) ()
    at /usr/local/lib/mangohud/libMangoHud_opengl.so
#5  0x00007ffff786669d in glXMakeCurrent () at /usr/local/lib/mangohud/libMangoHud_opengl.so
#6  0x00007ffff6bd23e1 in ??? () at /usr/lib/libSDL2-2.0.so.0
#7  0x00007ffff6bd5343 in ??? () at /usr/lib/libSDL2-2.0.so.0
#8  0x00007ffff6bac4a0 in ??? () at /usr/lib/libSDL2-2.0.so.0
#9  0x0000555555681e85 in ??? ()
#10 0x0000555555e0e217 in ??? ()
#11 0x0000555555682bae in ??? ()
#12 0x0000555555dea088 in ??? ()
#13 0x0000555555925e8c in ??? ()
#14 0x0000555555926d15 in ??? ()
#15 0x0000555555928481 in ??? ()
#16 0x0000555555651e25 in ??? ()
#17 0x00007ffff5a43cd0 in ??? () at /usr/lib/libc.so.6
#18 0x00007ffff5a43d8a in __libc_start_main () at /usr/lib/libc.so.6
#19 0x000055555567b315 in ??? ()

[JupiterRider]

Hey @flightlessmango , is this issue going to be fixed? I just than tell you that it only happens on a Intel GPU together with SDL2 games.

Thanks!

[flightlessmango]

This commit should resolve it 4cbcec30b8f302fecfe3c5ef95f1d0861a649a5f But this comes at the expense of all metrics except load. We'll just have to wait for intel to expose these metrics in sysfs

[JupiterRider]

Thanks! This commit works.