This bug report was migrated from our old Bugzilla tracker.

These attachments are available in the static archive:

• Patch to fix bug 4891 (0001-kmsdrm-Fix-busy-loop-within-libc-s-dlopen.patch, text/plain, 2019-12-17 23:33:44 +0000, 1337 bytes)

Reported in version: 2.0.10
Reported for operating system, platform: Linux, Other

Comments on the original bug report:

On 2019-12-08 14:41:23 +0000, Paul Cercueil wrote:

Running a basic SDL2 app on a MIPS-based Buildroot system running uClibc or musl (glibc untested) results in SDL_Init() locking up and eating the CPU in the process.

From a quick GDB session on the uClibc-based system, it appears to busy-loop in _dl_find_hash() (uclibc-1.0.32/ldso/ldso/dl-hash.c).

The fact that it happens with both musl and uClibc, and only with SDL2 and no other library or app, makes me think it's a bug of SDL2.

On 2019-12-08 14:54:07 +0000, Ryan C. Gordon wrote:

This function generally means the dynamic loader is resolving symbols. If you observed this under the debugger, can you run the “bt” command and paste the output here? It’s probably going to be somewhere under a dlopen() call.

--ryan.

On 2019-12-08 17:54:12 +0000, Paul Cercueil wrote:

Unfortunately it doesn't go very far:

(gdb) bt

0 0x77e6ab88 in _dl_find_hash () from /lib/ld-uClibc.so.0

1 0x77e6aba4 in _dl_find_hash () from /lib/ld-uClibc.so.0

Backtrace stopped: frame did not save the PC

(and yes, I compiled SDL2 with -ggdb3 and GDB did load the symbols).

On 2019-12-08 18:32:22 +0000, Paul Cercueil wrote:

I added debug to src/loadso/dlopen/SDL_sysloadso.c.
From what I can see, it seems to hang within dlopen(), when libdrm.so.2 is opened for the second time.

See below:

gcw0:~ $ gdb ./testsdl2
[...]
(gdb) r
Starting program: /media/data/local/home/testsdl2
[New LWP 264]
SDL dlopen(libdrm.so.2)... done
SDL dlopen(libgbm.so.1)... done
SDL dlsym(drmModeFreeResources)... done
SDL dlsym(drmModeFreeFB)... done
SDL dlsym(drmModeFreeCrtc)... done
SDL dlsym(drmModeFreeConnector)... done
SDL dlsym(drmModeFreeEncoder)... done
SDL dlsym(drmModeGetResources)... done
SDL dlsym(drmModeAddFB)... done
SDL dlsym(drmModeRmFB)... done
SDL dlsym(drmModeGetFB)... done
SDL dlsym(drmModeGetCrtc)... done
SDL dlsym(drmModeSetCrtc)... done
SDL dlsym(drmModeSetCursor)... done
SDL dlsym(drmModeSetCursor2)... done
SDL dlsym(drmModeMoveCursor)... done
SDL dlsym(drmModeGetEncoder)... done
SDL dlsym(drmModeGetConnector)... done
SDL dlsym(drmHandleEvent)... done
SDL dlsym(drmModePageFlip)... done
SDL dlsym(gbm_device_get_fd)... done
SDL dlsym(gbm_device_is_format_supported)... done
SDL dlsym(gbm_device_destroy)... done
SDL dlsym(gbm_create_device)... done
SDL dlsym(gbm_bo_get_width)... done
SDL dlsym(gbm_bo_get_height)... done
SDL dlsym(gbm_bo_get_stride)... done
SDL dlsym(gbm_bo_get_handle)... done
SDL dlsym(gbm_bo_write)... done
SDL dlsym(gbm_bo_get_device)... done
SDL dlsym(gbm_bo_set_user_data)... done
SDL dlsym(gbm_bo_get_user_data)... done
SDL dlsym(gbm_bo_destroy)... done
SDL dlsym(gbm_bo_create)... done
SDL dlsym(gbm_surface_create)... done
SDL dlsym(gbm_surface_destroy)... done
SDL dlsym(gbm_surface_lock_front_buffer)... done
SDL dlsym(gbm_surface_release_buffer)... done
SDL dlopen(libdrm.so.2)...

On 2019-12-17 22:38:53 +0000, Paul Cercueil wrote:

Small update: this bug does not happen if libdrm is pre-loaded:

SDL_PRELOAD=/usr/lib/libdrm.so.2 ./testsdl2

works fine.

On 2019-12-17 23:33:44 +0000, Paul Cercueil wrote:

Created attachment 4113
Patch to fix bug 4891

On 2019-12-20 05:48:06 +0000, Sam Lantinga wrote:

Patch added, thanks!
https://hg.libsdl.org/SDL/rev/4f5bef55183c