Game: Stack overflow crash after intro on v3.0.0 #1772

Open
opened 2026-11-19 18:27:49 +00:00 by BeanBot · 10 comments
Member

Reported on Discord

Longshot version

v3.0.0 (source 87cb36aa)

Channel

Release

What happened

Initially, the game crashed after the intro. The first crash record showed an unhandled stack-overflow exception (0xc00000fd) on the PreActorModelCompose thread. Reporter clarified that this occurs after loading a save at the beginning of the game after the intro, then moving, and says it happens every time. They tried adjusting settings. Screenshots show the title screen with Continue selected and graphics settings (Handheld profile, 1920x1080 output, game's own upscaler, 50% render resolution, frame generation off, V-Sync on).

A later in-game crash report describes a crash while in KokiriForest_Child, after 178 seconds. It reports 0xc0000005 on OptimizeChangeBanc, GPU driver failure while compiling a graphics pipeline, and VK_ERROR_DEVICE_LOST. After the reporter said the NVIDIA driver had apparently removed itself, they reported that stable v3.0 was running again and continued testing.

Most recent report: gameplay ran smoothly until leaving the Gohma boss arena, then the game crashed on a white screen. The attached in-game crash report (2026-11-20 13:10 UTC) says it had run 2337 seconds, in KokiriForest_Child, and crashed with 0xc0000005 (write) in nvoglv64.dll+0xb15f1e. It identifies NVIDIA GeForce RTX 2080 Ti, NVIDIA 621.24, AMD Ryzen 7 3800X, 24 GB RAM, Windows 10 build 19045. The log shows severe GPU memory pressure and allocation failures, including VRAM full, sparse mirrors unable to allocate blocks, texture levels left out, and commits at the system commit limit. It records 14 FPS over the final 5.4 seconds. This driver-crash signature is different from the original stack overflow; whether the original crash still reproduces is unknown.

Graphics card and driver version

NVIDIA GeForce RTX 2080 Ti, NVIDIA 621.24 (from the latest report)

CPU, RAM and operating system

AMD Ryzen 7 3800X 8-Core Processor (16 threads), 24 GB, Windows 10 (build 19045)

Changes from the defaults

Latest report's options: quality preset Custom, output 1920x1080, docked profile, game's own upscaler, render resolution 70%, frame generation off, V-Sync on, fullscreen on, HDR off; no switches set. Screenshots show a separate earlier settings state (Handheld, 50% render resolution, V-Sync on). The reporter also previously mentioned 30 FPS docked at 70% render resolution.

Logs and screenshots

Original attachments: report.txt, longshot_log.txt, longshot_log.1.txt, first crash record, and screenshots. Subsequent in-game crash reports include the later NVIDIA driver crash. Latest attachment message.txt contains the 2026-11-20 report, crash record, and log excerpts; screenshots show the Ocarina of Time title screen with Continue selected and graphics settings. The latest report's log excerpt includes VRAM and system commit pressure and the crash in nvoglv64.dll.

Still needed

Clarify whether the original stack-overflow crash after loading the early save and moving still occurs, and whether the driver was reinstalled/updated (the current report still identifies driver 621.24). Confirm whether the latest crash happened specifically as the reporter left Gohma's arena, and whether this is reproducible. Exact actions/settings immediately before it and a full longshot_log.txt from this latest run would help distinguish the game trigger from the NVIDIA driver/memory failure.

Reported on [Discord](https://discord.com/channels/1569849200047427645/1573039630746652733/1573039630746652733) ### Longshot version v3.0.0 (source 87cb36aa) ### Channel Release ### What happened Initially, the game crashed after the intro. The first crash record showed an unhandled stack-overflow exception (`0xc00000fd`) on the `PreActorModelCompose` thread. Reporter clarified that this occurs after loading a save at the beginning of the game after the intro, then moving, and says it happens every time. They tried adjusting settings. Screenshots show the title screen with Continue selected and graphics settings (Handheld profile, 1920x1080 output, game's own upscaler, 50% render resolution, frame generation off, V-Sync on). A later in-game crash report describes a crash while in KokiriForest_Child, after 178 seconds. It reports `0xc0000005` on `OptimizeChangeBanc`, GPU driver failure while compiling a graphics pipeline, and `VK_ERROR_DEVICE_LOST`. After the reporter said the NVIDIA driver had apparently removed itself, they reported that stable v3.0 was running again and continued testing. Most recent report: gameplay ran smoothly until leaving the Gohma boss arena, then the game crashed on a white screen. The attached in-game crash report (2026-11-20 13:10 UTC) says it had run 2337 seconds, in KokiriForest_Child, and crashed with `0xc0000005` (write) in `nvoglv64.dll+0xb15f1e`. It identifies NVIDIA GeForce RTX 2080 Ti, NVIDIA 621.24, AMD Ryzen 7 3800X, 24 GB RAM, Windows 10 build 19045. The log shows severe GPU memory pressure and allocation failures, including `VRAM full`, sparse mirrors unable to allocate blocks, texture levels left out, and commits at the system commit limit. It records 14 FPS over the final 5.4 seconds. This driver-crash signature is different from the original stack overflow; whether the original crash still reproduces is unknown. ### Graphics card and driver version NVIDIA GeForce RTX 2080 Ti, NVIDIA 621.24 (from the latest report) ### CPU, RAM and operating system AMD Ryzen 7 3800X 8-Core Processor (16 threads), 24 GB, Windows 10 (build 19045) ### Changes from the defaults Latest report's options: quality preset Custom, output 1920x1080, docked profile, game's own upscaler, render resolution 70%, frame generation off, V-Sync on, fullscreen on, HDR off; no switches set. Screenshots show a separate earlier settings state (Handheld, 50% render resolution, V-Sync on). The reporter also previously mentioned 30 FPS docked at 70% render resolution. ### Logs and screenshots Original attachments: `report.txt`, `longshot_log.txt`, `longshot_log.1.txt`, first crash record, and screenshots. Subsequent in-game crash reports include the later NVIDIA driver crash. Latest attachment `message.txt` contains the 2026-11-20 report, crash record, and log excerpts; screenshots show the Ocarina of Time title screen with Continue selected and graphics settings. The latest report's log excerpt includes VRAM and system commit pressure and the crash in `nvoglv64.dll`. ### Still needed Clarify whether the original stack-overflow crash after loading the early save and moving still occurs, and whether the driver was reinstalled/updated (the current report still identifies driver 621.24). Confirm whether the latest crash happened specifically as the reporter left Gohma's arena, and whether this is reproducible. Exact actions/settings immediately before it and a full longshot_log.txt from this latest run would help distinguish the game trigger from the NVIDIA driver/memory failure.
Author
Member
Files from the Discord report: - [report.txt](https://git.magicbean.dev/attachments/51ddc10c-9380-4a1b-b939-e7f3eefe4987) - [longshot_log.txt](https://git.magicbean.dev/attachments/12b9b4a3-b002-454a-9c2c-febc594da1ab) - [longshot_log.1.txt](https://git.magicbean.dev/attachments/4fc61ab9-4978-47e8-8a1f-aa4a8b314319) - [picture.png](https://git.magicbean.dev/attachments/c5f1183f-8a75-4461-a2f6-3169ad3ea5af) - [crash-20261119T182137Z.txt](https://git.magicbean.dev/attachments/30a69ea0-4112-4007-9d32-ab49b49ddda9)
Author
Member

Updated with more from Discord: Added the reproducible steps and frequency reported after filing, plus screenshots/settings; clarified the attached bug-report logs are from a distinct run.

Files:

Updated with more from [Discord](https://discord.com/channels/1569849200047427645/1573039630746652733/1573042862017482783): Added the reproducible steps and frequency reported after filing, plus screenshots/settings; clarified the attached bug-report logs are from a distinct run. Files: - [image.png](https://git.magicbean.dev/attachments/47bbacf7-5ef5-49bc-8fb0-e0832b0a3f3e) - [image.png](https://git.magicbean.dev/attachments/19ad299a-98e8-4608-86b4-5bb0ae468c75)
Author
Member

Updated with more from Discord: Added the subsequent in-game crash report and distinguished its NVIDIA Vulkan device-lost crash from the original stack overflow; noted follow-up details still needed.

Files:

Updated with more from [Discord](https://discord.com/channels/1569849200047427645/1573039630746652733/1573043906722013318): Added the subsequent in-game crash report and distinguished its NVIDIA Vulkan device-lost crash from the original stack overflow; noted follow-up details still needed. Files: - [message.txt](https://git.magicbean.dev/attachments/4471b848-b92b-4f0e-bc91-f246580c1158)
Author
Member

From Discord:

Reporter says the NVIDIA driver appears to have uninstalled itself and they will reinstall it and test again on stable v3.0. No result from that test yet.

From [Discord](https://discord.com/channels/1569849200047427645/1573039630746652733/1573039630746652733): Reporter says the NVIDIA driver appears to have uninstalled itself and they will reinstall it and test again on stable v3.0. No result from that test yet.
Author
Member

Updated with more from Discord: Added the reporter's latest update: stable v3.0 now runs following the NVIDIA-driver issue, with 30 FPS docked at 70% render resolution; whether the earlier crashes recur remains unknown.

Updated with more from [Discord](https://discord.com/channels/1569849200047427645/1573039630746652733/1573050433545441384): Added the reporter's latest update: stable v3.0 now runs following the NVIDIA-driver issue, with 30 FPS docked at 70% render resolution; whether the earlier crashes recur remains unknown.
Author
Member

From Discord:

New Discord update from the reporter: after playing smoothly, Longshot v3.0.0 crashed on a white screen while leaving Queen Gohma's arena. Attached crash report identifies Windows 10 build 19045, RTX 2080 Ti, NVIDIA 621.24, Ryzen 7 3800X, 24 GB RAM; crash at 2,337 s in KokiriForest_Child, exception 0xc0000005 (write) in nvoglv64.dll+0xb15f1e. Log shows severe GPU-memory pressure immediately before the crash: repeated image allocation failures, VRAM full, sparse mirrors at ~881 MB/2048 MB, 2,013 MB of mirrors returned, and commit-limit warnings. This likely represents the later NVIDIA driver/device-loss crash already documented here, with further progression/location context; screenshots in this latest post show v3.0.0 settings. Attachments: Discord message.txt [1573324672588587008]. Still unknown whether the early-save stack overflow reproduced after the driver was reinstalled/Longshot resumed running, and the exact actions leaving the arena.

From [Discord](https://discord.com/channels/1569849200047427645/1573039630746652733/1573039630746652733): New Discord update from the reporter: after playing smoothly, Longshot v3.0.0 crashed on a white screen while leaving Queen Gohma's arena. Attached crash report identifies Windows 10 build 19045, RTX 2080 Ti, NVIDIA 621.24, Ryzen 7 3800X, 24 GB RAM; crash at 2,337 s in KokiriForest_Child, exception 0xc0000005 (write) in nvoglv64.dll+0xb15f1e. Log shows severe GPU-memory pressure immediately before the crash: repeated image allocation failures, VRAM full, sparse mirrors at ~881 MB/2048 MB, 2,013 MB of mirrors returned, and commit-limit warnings. This likely represents the later NVIDIA driver/device-loss crash already documented here, with further progression/location context; screenshots in this latest post show v3.0.0 settings. Attachments: Discord message.txt [1573324672588587008]. Still unknown whether the early-save stack overflow reproduced after the driver was reinstalled/Longshot resumed running, and the exact actions leaving the arena.
Author
Member

Updated with more from Discord: Added the latest white-screen crash after leaving Gohma’s arena, with crash report evidence of an NVIDIA driver fault and GPU-memory allocation pressure.

Files:

Updated with more from [Discord](https://discord.com/channels/1569849200047427645/1573039630746652733/1573324672588587008): Added the latest white-screen crash after leaving Gohma’s arena, with crash report evidence of an NVIDIA driver fault and GPU-memory allocation pressure. Files: - [message.txt](https://git.magicbean.dev/attachments/1ae4d619-30ae-45ae-9f94-9802e1bc212c) - [message.txt](https://git.magicbean.dev/attachments/32234610-5f66-4530-ae8f-a06ed5ee5975)
Contributor

Claiming the first crash here, the stack overflow (0xc00000fd on PreActorModelCompose). I'm testing on Windows 11 with an RTX 3080 (NVIDIA 621.14). The later nvoglv64.dll crash under VRAM pressure is not part of this claim.

What the crash run shows (longshot_log.1.txt, run_id 6e14ecec): it wasn't a stack that ran out. Over the last seconds the PC was at Windows' commit limit. There are mem: commit of 0x10000 bytes at the commit limit made after 50-150 ms lines (8 of them), vk: image ... not made lines, and the renderer leaving texture levels out, all with 24 GB of RAM. Guest threads get a 64 MB stack that is only reserved (host/hos_thread.c, STACK_SIZE_PARAM_IS_A_RESERVATION), so every new stack page needs commit. When Windows can't commit the next one at the commit limit, it raises 0xc00000fd, a "stack overflow", with nearly all of the 64 MB unused. Guest memory already waits out the commit limit (commit_on_touch and commit_wait in host/mkw_glue.c), but a thread's own stack has nothing that does.

Plan: when a thread overflows with most of its reservation unused, wait for commit as guest memory does, set the guard page again and go on. If commit doesn't come back, end with the out-of-memory notice and a crash record that says so, not "stack overflow". I'll write a game-free test first that fails on main: a thread in a Job object with a commit limit that grows its stack while memory is given back. Then the fix, then a run in the game at a forced commit limit.

Prior work: none found (searched "stack overflow", c00000fd, PreActorModelCompose, "thread stack", "stack size"; only the macOS renderer stacks of #420 and #849, a different cause).

Posted by an AI agent (Claude Code) for d4nkn3ss.

Claiming the first crash here, the stack overflow (`0xc00000fd` on PreActorModelCompose). I'm testing on Windows 11 with an RTX 3080 (NVIDIA 621.14). The later `nvoglv64.dll` crash under VRAM pressure is not part of this claim. What the crash run shows (`longshot_log.1.txt`, run_id 6e14ecec): it wasn't a stack that ran out. Over the last seconds the PC was at Windows' commit limit. There are `mem: commit of 0x10000 bytes at the commit limit made after 50-150 ms` lines (8 of them), `vk: image ... not made` lines, and the renderer leaving texture levels out, all with 24 GB of RAM. Guest threads get a 64 MB stack that is only reserved (`host/hos_thread.c`, `STACK_SIZE_PARAM_IS_A_RESERVATION`), so every new stack page needs commit. When Windows can't commit the next one at the commit limit, it raises `0xc00000fd`, a "stack overflow", with nearly all of the 64 MB unused. Guest memory already waits out the commit limit (`commit_on_touch` and `commit_wait` in `host/mkw_glue.c`), but a thread's own stack has nothing that does. Plan: when a thread overflows with most of its reservation unused, wait for commit as guest memory does, set the guard page again and go on. If commit doesn't come back, end with the out-of-memory notice and a crash record that says so, not "stack overflow". I'll write a game-free test first that fails on main: a thread in a Job object with a commit limit that grows its stack while memory is given back. Then the fix, then a run in the game at a forced commit limit. Prior work: none found (searched "stack overflow", `c00000fd`, `PreActorModelCompose`, "thread stack", "stack size"; only the macOS renderer stacks of #420 and #849, a different cause). *Posted by an AI agent (Claude Code) for d4nkn3ss.*
Contributor

#1885 has the fix for the stack overflow: at the commit limit, a thread whose next stack page is refused waits for commit (up to 5 s) and goes on. If commit never comes back, the out-of-memory notice shows first.

Under a forced 14 GB commit limit, main crashed with 0xc00000fd in 2 of 3 starts and the PR in none of 4. At that limit the PR's runs still ended elsewhere, though, twice in NVIDIA's driver (nvoglv64.dll and nvwgf2umx.dll, writes near address 0). That matches the shape of this issue's later nvoglv64.dll crash: the driver's own allocation failing when the PC is out of commit. So for the reporter, the lasting fix is more page file. With 24 GB of RAM, letting Windows manage the page file size, as the out-of-memory notice says, should stop both crashes.

Posted by an AI agent (Claude Code) for d4nkn3ss.

https://git.magicbean.dev/MagicBeanDev/Longshot/pulls/1885 has the fix for the stack overflow: at the commit limit, a thread whose next stack page is refused waits for commit (up to 5 s) and goes on. If commit never comes back, the out-of-memory notice shows first. Under a forced 14 GB commit limit, main crashed with `0xc00000fd` in 2 of 3 starts and the PR in none of 4. At that limit the PR's runs still ended elsewhere, though, twice in NVIDIA's driver (`nvoglv64.dll` and `nvwgf2umx.dll`, writes near address 0). That matches the shape of this issue's later `nvoglv64.dll` crash: the driver's own allocation failing when the PC is out of commit. So for the reporter, the lasting fix is more page file. With 24 GB of RAM, letting Windows manage the page file size, as the out-of-memory notice says, should stop both crashes. *Posted by an AI agent (Claude Code) for d4nkn3ss.*
Author
Member

From Discord:

The reporter's attached longshot_log.txt and longshot_log.1.txt provide a stack-overflow crash record (0xc00000fd) while entering Hyrule Field on v3.0.0, source 87cb36aa. This is a distinct location/scenario from the original stack-overflow report; please correlate cautiously. The uploaded longshot_log.txt contains 6,716 lines; crash-record thread is Resource Control. Their crash record is attached in the Discord post.

From [Discord](https://discord.com/channels/1569849200047427645/1573448048154124409/1573448048154124409): The reporter's attached `longshot_log.txt` and `longshot_log.1.txt` provide a stack-overflow crash record (`0xc00000fd`) while entering Hyrule Field on v3.0.0, source 87cb36aa. This is a distinct location/scenario from the original stack-overflow report; please correlate cautiously. The uploaded `longshot_log.txt` contains 6,716 lines; crash-record thread is Resource Control. Their crash record is attached in the Discord post.
Sign in to join this conversation.
No milestone
No project
No assignees
2 participants
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set

Reference
MagicBeanDev/Longshot#1772
No description provided.