8000 Crash: the game crashed (signal 11 at guest offset 0x2169edc) · Issue #99 · Wooloomooloo2/bloodborne_mac_port · GitHub
[go: up one dir, main page]

Skip to content

Crash: the game crashed (signal 11 at guest offset 0x2169edc) #99

Closed

Description

@ELLDev

What were you doing when it happened?
had just spawned on hunters dreams

Diagnostics: please drag crash-2026-10-27_23-16-08.zip (the launcher showed it in Finder) onto this issue: it has the full session log and macOS's crash reports.

Bloodborne for Mac 0.5.10: crash report

  • Mac: Mac14,10 (Apple M2 Pro), 16 GB, macOS 26.5.1 (Build 25F80)
  • Settings: output 1920x1080, upscaler off, cap 60 FPS, fixed timing, full screen
  • Session: started 2026-10-27 23:16:08 -03:00, ran 101 min 32 s
  • Exit status: 139
  • Cause: the game crashed (signal 11 at guest offset 0x2169edc) (at 00:57:41.342)
  • Log: game-2026-10-27_23-16-08.log, 2487 lines: 1 critical, 439 errors, 14 warnings
Warnings and errors before the cause (5 most recent)
00:57:37.001 ERROR  GPU [Render.Vulkan] <Error> vk_pipeline_common.cpp:58 Fail: Graphics pipeline fs_0x000000002390c887,vs_0x0000000097f2d384 not built (the Metal pipeline state failed): its draws are skipped
00:57:37.003 ERROR  GPU [Render] <Error> mtl_pso_cache.mm:699 GetRenderPipeline: Metal render pipeline (vs 0x632a64205d4 fs 0xe54146c6792): vertex attribute 0 (location 0) format Undefined has no Metal vertex format
00:57:37.003 ERROR  GPU [Render.Vulkan] <Error> vk_pipeline_common.cpp:58 Fail: Graphics pipeline fs_0x00000000e54449c7,vs_0x0000000063220f67 not built (the Metal pipeline state failed): its draws are skipped
00:57:37.028 ERROR  GPU [Render] <Error> mtl_pso_cache.mm:699 GetRenderPipeline: Metal render pipeline (vs 0x5999706a8b6 fs 0x74afee5712c): vertex attribute 0 (location 0) format Undefined has no Metal vertex format
00:57:37.028 ERROR  GPU [Render.Vulkan] <Error> vk_pipeline_common.cpp:58 Fail: Graphics pipeline fs_0x00000000749e4f9e,vs_0x000000005992ac36 not built (the Metal pipeline state failed): its draws are skipped
game-2026-10-27_23-16-08.log (16 lines around the cause)
00:57:36.988 ERROR  GPU [Render] <Error> mtl_pso_cache.mm:699 GetRenderPipeline: Metal render pipeline (vs 0xb140d7e9ec fs 0x83ecd611598): vertex attribute 0 (location 0) format Undefined has no Metal vertex format
00:57:36.988 ERROR  GPU [Render.Vulkan] <Error> vk_pipeline_common.cpp:58 Fail: Graphics pipeline fs_0x0000000083da8467,vs_0x000000000b0acf50 not built (the Metal pipeline state failed): its draws are skipped
00:57:37.001 ERROR  GPU [Render] <Error> mtl_pso_cache.mm:699 GetRenderPipeline: Metal render pipeline (vs 0x97f431c357c fs 0x2398f683ec6): vertex attribute 0 (location 0) format Undefined has no Metal vertex format
00:57:37.001 ERROR  GPU [Render.Vulkan] <Error> vk_pipeline_common.cpp:58 Fail: Graphics pipeline fs_0x000000002390c887,vs_0x0000000097f2d384 not built (the Metal pipeline state failed): its draws are skipped
00:57:37.003 ERROR  GPU [Render] <Error> mtl_pso_cache.mm:699 GetRenderPipeline: Metal render pipeline (vs 0x632a64205d4 fs 0xe54146c6792): vertex attribute 0 (location 0) format Undefined has no Metal vertex format
00:57:37.003 ERROR  GPU [Render.Vulkan] <Error> vk_pipeline_common.cpp:58 Fail: Graphics pipeline fs_0x00000000e54449c7,vs_0x0000000063220f67 not built (the Metal pipeline state failed): its draws are skipped
00:57:37.028 ERROR  GPU [Render] <Error> mtl_pso_cache.mm:699 GetRenderPipeline: Metal render pipeline (vs 0x5999706a8b6 fs 0x74afee5712c): vertex attribute 0 (location 0) format Undefined has no Metal vertex format
00:57:37.028 ERROR  GPU [Render.Vulkan] <Error> vk_pipeline_common.cpp:58 Fail: Graphics pipeline fs_0x00000000749e4f9e,vs_0x000000005992ac36 not built (the Metal pipeline state failed): its draws are skipped
00:57:41.342 CRIT   Crash: the game crashed (signal 11 at guest offset 0x2169edc)
00:57:41.342 INFO   Guest fault (signal 11) at guest offset 0x2169edc, address 0x0
00:57:41.342 INFO     rax=0 rbx=0 rdi=0 rsi=0x70282adf70 rsp=0x806a5b870
00:57:41.460 INFO     #2 0x806a5a950
00:57:41.460 INFO     #3 0x800e2f157
00:57:41.460 INFO     #4 0x802286260
00:57:41.460 INFO     #5 0x8021b4dca
00:57:41.460 INFO     #6 0x802199bbe

Activity

  1. ELLDev commented on Oct 28, 2026

    @ELLDev
    Author
  2. Wooloomooloo2 commented on Oct 28, 2026

    @Wooloomooloo2
    Owner

    Same crash reported in #101 on 0.5.9 (64 GB M3 Max, Hunter's Dream, about 90 minutes in), so it isn't memory or 0.5.10 related. A fix that skips the missing bind instead of crashing is in review in #100, aimed at 0.5.11.

  3. added a commit that references this issue on Oct 28, 2026
  4. noelmiller commented on Oct 29, 2026

    @noelmiller
    Contributor

    The guard from #100 works, but it only covers one of two dispatchers that get the same missing record, so the crash just moves 1 ms later.

    What I saw. 0.5.10 (master aa89fdd plus online play), 113 minutes of Hunter's Dream co-op, then:

    23:30:47.072 INFO   Render guard: absent shader bind record (#99), context 0x70282a2840, occurrence 1
    23:30:47.073 CRIT   Crash: the game crashed (signal 11 at guest offset 0x216a08c)
    23:30:47.073 INFO   Guest fault (signal 11) at guest offset 0x216a08c, address 0x0
    23:30:47.073 INFO     rax=0 rbx=0x70282a2840 rdi=0 rsi=0x70282adf70 rsp=0x806a5b870
    23:30:47.073 INFO     #3 0x800e2f181
    

    Same context in RBX as the guard line, address zero, RAX/RDI zero: the #99 pattern again, at a new offset.

    Why. 0x216a08c is inside a twin of the guarded dispatcher, at 0x216a070. It has the same structure:

    216a070: push rbp; mov rbp,rsp; push r15; push r14; push rbx; push rax
    216a07a: mov rbx,rdi
    216a07d: lea r14,[rbx+0x10]
    216a081: mov rdi,rdx
    216a084: call 0x26a94c0        ; same identity accessor
    216a089: mov r15,rax
    216a08c: mov eax,[r15]         ; <- fault on a null record
    ...
    216a298: mov rdi,r14; add rsp,8; pop rbx; pop r14; pop r15; pop rbp
    216a2a5: jmp 0x26af700         ; same unsupported-type tail into the finalizer
    

    The caller runs the two back to back for the same record: 0xe2f152 calls 0x2169ec0, then 0xe2f17c calls 0x216a070. Frame #3 above (0xe2f181) is the return address of that second call. Without the guard the game died in the first one, so the second never showed up in the reports. With the guard, an absent record now always lands in the second.

    There are only three callers of 0x26a94c0 that take the record from RDX. The third, at 0x216a770, already does test rdx,rdx; je first, so these two are the whole set.

    Fix. I extended runtime_render_guard.c to hook both the same way:

    • A site table: entry and tail signatures for each dispatcher. The twin's 17 moved bytes also have no relative addressing.
    • All or nothing: any mismatch in either leaves both unpatched.
    • The log line now names the dispatcher that saw the null record.

    tests/test_render_guard.c now runs both dispatchers' original code against a prepared 1.09 ELF:

    PASS: original 1.09 null bind in 0x2169ec0 reproduces the reported guest fault at 0x2169edc
    PASS: original 1.09 null bind in 0x216a070 reproduces the reported guest fault at 0x216a08c
    Render guard: null shader binds (0x2169ec0, 0x216a070) hooked
    PASS: 0x2169ec0: null bind follows existing finish; 40 non-null cases preserve writes, context and return
    PASS: 0x216a070: null bind follows existing finish; 40 non-null cases preserve writes, context and return
    PASS: a null record passes both dispatchers in turn
    

    (The twin also calls a writer at 0x26bb6d0, which the fixture now records.) I also added a short section to the #99 entry in docs/FINDINGS.md. It's in the game now: the launch logs both hooks. It hasn't been through a long co-op session yet, so it isn't proven in play. The same caveats as #100 apply: this contains the null dereference; where the record goes missing is still unknown.

    Happy to open it as a PR against master if that's useful.

    Setup: M5 Max, 36 GB, macOS 27.0.1, 2560x1440, Hunter's Dream co-op.

    Written with Claude Code.

  5. Wooloomooloo2 commented on Oct 31, 2026

    @Wooloomooloo2
    Owner

    Thanks @noelmiller, that's exactly what happened: the same missing record also reached the release step (#113, #112), and after that the game asks for its value too. 0.5.12 guards both (#114 from @evill33t), so the whole sequence for a missing record now skips instead of crashing. Closing. If it happens again on 0.5.12, please reopen with the crash zip.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions

      0