i386: stop guessing the address of the trap frame in ddb backtrace.

[kib]

Summary

Save the address of the trap frame in %ebp on kernel entry. This automatically provides it in struct i386_frame.f_frame to unwinder.

While there, more accurately handle the terminating frames,

[avg]

Looks good to me in general.

sys/i386/i386/db_trace.c

383 ↗

(On Diff #64212)

This comment is a little bit hard to read.
Maybe start with "This can be the case for" or something like that.

384 ↗

(On Diff #64212)

After the change, "--- trap" won't be printed in this case, but probably that's okay?

412 ↗

(On Diff #64212)

The new long condition is a little bit harder to parse than the previous code.
Maybe keep the switch statement but modify the if statement, so that the userland condition falls through to SYSCALL case.

switch (frame_type) {
case TRAP:
case INTERRUPT:
        if ((tf->tf_eflags & PSL_VM) == 0 &&
            (tf->tf_cs & SEL_RPL_MASK) == 0)
                break;
        /* FALLTHROUGH */
case SYSCALL:
        ebp = 0;
        eip = 0;
        break;
}

[kib]

sys/i386/i386/db_trace.c

384 ↗

(On Diff #64212)

But that is the part of the improvement IMO. The kthread does not have trap frame on top of the stack.

I can add a frame type for fork_trampoline() just to print e.g. 'kthread start', but could as well leave it as is. It is really the last frame.

[kib]

Handle avg' suggestions.

Improve comment.
Restore switch(), but use TRAPF_USERMODE().
Print something after last frame of kthread.

[avg]

Looks great!

sys/i386/i386/db_trace.c

384 ↗

(On Diff #64212)

Yes, I agree, that's okay.