Closed Bug 1457657 Opened 7 years ago Closed 7 years ago

Something named libnvidia-tls is calling arch_prctl(ARCH_GET_FS)

Categories

(Core :: Security: Process Sandboxing, enhancement, P5)

Product:
Core
Component:
Security: Process Sandboxing
Version:
60 Branch
Platform:
x86_64
Linux
Type:
enhancement

Tracking

()

Status:
RESOLVED WONTFIX

People

(Reporter: jld, Assigned: jld)

Details

Crash Data

NVIDIA has their own thread-local storage library, it seems — and it's using arch_prctl on amd64 to access the vestigial segment base registers.
There's something weird going on here. It took a lot of my local copies of this library; they don't seem to be importing the symbols they'd use for this syscall, or using the syscall instruction directly, or loading the value of ARCH_GET_FS into a register. They do have thread-local data, but glibc also doesn't use ARCH_GET_FS, except to special-case it in its syscall wrappers for the x32 ABI (amd64 with 32-bit pointers). Also, in my copy of libnvidia-tls.so.390.48, offset 0x911 is in the .rela.dyn section, not executable code. *And*, the GPU reported in the telemetry environment is weird: bp-3c08aedf-57c8-4711-a1be-c62eb0180427 gives a different version from the crashing library (384.111 vs. 390.48), while bp-4cf89208-afd2-431f-b8dc-d45da0180424 says it's using Mesa/Gallium. I'm going to deprioritize this until/unless more of these turn up.
Priority: P1 → P5
Summary: libnvidia-tls is calling arch_prctl(ARCH_GET_FS) → Something named libnvidia-tls is calling arch_prctl(ARCH_GET_FS)
Closing because no crash reported since 12 weeks.
Status: NEW → RESOLVED
Closed: 7 years ago
Resolution: --- → WONTFIX
Closing because no crash reported since 12 weeks.
You need to log in before you can comment on or make changes to this bug.