On 2024/3/4 23:15, Jann Horn wrote: > On Mon, Mar 4, 2024 at 3:02 AM Jiangfeng Xiao <xiaojiangfeng@xxxxxxxxxx> wrote: >> When the last instruction of a noreturn function is a call >> to another function, the return address falls outside >> of the function boundary. This seems to cause kernel >> to interrupt the backtrace. > [...] >> Delete __noreturn from usercopy_abort, > > This sounds like the actual bug is in the backtracing logic? I don't > think removing __noreturn annotations from an individual function is a > good fix, since the same thing can happen with other __noreturn > functions depending on what choices the compiler makes. > . > Yes, you make a point. This may be a bug is in the backtracing logic, but the kernel backtracing always parses symbols using (lr) instead of (lr-4). This may be due to historical reasons or more comprehensive considerations. In addition, modifying the implementation logic of the kernel backtracing has a great impact. Therefore, I do not dare to modify the implementation logic of the kernel backtracing. Not all noreturn functions are ended with calling other functions. Therefore, only a few individual functions may have the same problem. In addition, deleting '__noreturn' from usercopy_abort does not change the internal behavior of usercopy_abort function. Therefore, there is no risk. Deleting '__noreturn' from usercopy_abort is the solution that I can think of with minimal impact and minimum risk. If you will submit a better patch to solve this problem, I would like to learn from you. Thank you.
