On Wed, 1 Jul 2026 06:11:37 +0000 Peter Wang (王信友) <[email protected]> wrote: > However, I am curious: if the HBA is removed, implying that the > storage would become unusable, might the system encounter an > I/O hang or shutdown, potentially preventing its detection? > Perhaps it's a theoretical issue that would not manifest > in a real-world situation?
Note, it doesn't necessarily mean that the device itself was removed. The issue is that a pointer to an allocated descriptor is saved in the ring buffer. Maybe once the device is created it will never go way. But what happens if for some reason the descriptor is freed and reallocated? Now the old descriptor pointer is still in the ring buffer. What in the logic guarantees that the pointer will never be freed? And lets say there is an issue and the hba is freed and you debug this by dumping the trace buffer via ftrace_dump_on_oops. Now the dump itself may crash and you don't have a way to debug what happened. One other point that causes issues here. It makes user space tracing useless. Try tracing this with "trace-cmd record". These events will not be able to be parsed. -- Steve
