On Fri, 26 Mar 2021 03:05:03 +0900 Masami Hiramatsu <mhiramat@xxxxxxxxxx> wrote: > On Wed, 24 Mar 2021 10:40:58 +0900 > Masami Hiramatsu <mhiramat@xxxxxxxxxx> wrote: > > > On Tue, 23 Mar 2021 23:30:07 +0100 > > Peter Zijlstra <peterz@xxxxxxxxxxxxx> wrote: > > > > > On Mon, Mar 22, 2021 at 03:41:40PM +0900, Masami Hiramatsu wrote: > > > > ".global kretprobe_trampoline\n" > > > > ".type kretprobe_trampoline, @function\n" > > > > "kretprobe_trampoline:\n" > > > > #ifdef CONFIG_X86_64 > > > > > > So what happens if we get an NMI here? That is, after the RET but before > > > the push? Then our IP points into the trampoline but we've not done that > > > push yet. > > > > Not only NMI, but also interrupts can happen. There is no cli/sti here. > > > > Anyway, thanks for pointing! > > I think in UNWIND_HINT_TYPE_REGS and UNWIND_HINT_TYPE_REGS_PARTIAL cases > > ORC unwinder also has to check the state->ip and if it is kretprobe_trampoline, > > it should be recovered. > > What about this? > > Hmm, this seems to intoduce another issue on stacktrace from kprobes. > > <...>-137 [003] d.Z. 17.250714: p_full_proxy_read_5: (full_proxy_read+0x5/0x80) > <...>-137 [003] d.Z. 17.250737: <stack trace> > => kprobe_trace_func+0x1d0/0x2c0 > => kprobe_dispatcher+0x39/0x60 > => aggr_pre_handler+0x4f/0x90 > => kprobe_int3_handler+0x152/0x1a0 > => exc_int3+0x47/0x140 > => asm_exc_int3+0x31/0x40 > => 0 > => 0 > => 0 > => 0 > => 0 > => 0 > => 0 > > Let me check... I confirmed this is not related to this series, but occurs when I build kernels with different configs without cleanup. Once I build kernel with CONFIG_UNWIND_GUESS=y (for testing), and after that, I build kernel again with CONFIG_UNWIND_ORC=y (but without make clean), this happened. In this case, I guess ORC data might be corrupted? When I cleanup and rebuild, the stacktrace seems correct. Thank you, -- Masami Hiramatsu <mhiramat@xxxxxxxxxx>
