Add informational note to user-signalled SIGSEGV

[topolarity]

If we hit a segfault from a user-triggered SIGSEGV, that usually means that someone has tampered with our signal handlers (as in JuliaInterop/Clang.jl#549)

Prints as:

julia> Threads.@threads for i in 1:1000; zeros(1024, 1024) .+ zeros(1024, 1024); end

[29561] signal 11 (-6): Segmentation fault
signal came from user-land, signal handlers may be mis-configured.
in expression starting at REPL[3]:1

This will not help if the signal handler was dropped on the floor or intercepted by another application without forwarding, but it will at least be useful to detect bad signal forwarding from another library.

[giordano]

Oh, maybe this would help with PyCall and PythonCall, there are often segfaults because of multithreaded GC interacting badly with the python process.

[ararslan]

Suggested change

	jl_safe_printf("signal came from user-land, signal handlers may be mis-configured.\n");
	jl_safe_printf("signal came from userland, signal handlers may be misconfigured.\n");

[ararslan]

This should also work on FreeBSD, I think

[vtjnash]

What does "signal came from userland" mean? Every signal comes from userland in some way (except perhaps SIGPWR, and that is very uncommon).

[vtjnash]

Suggested change

	if ((sig == SIGBUS || sig == SIGSEGV) && (si_code <= 0))
	if ((sig == SIGBUS || sig == SIGSEGV) && (si_code <= 0)) # SI_FROMUSER(&si_code)

But none of these seem really that interesting here to deserve a special comment when our signal handler is installed but gets one of these sent:

/*
 * si_code values
 * Digital reserves positive values for kernel-generated signals.
 */
#define SI_USER         0               /* sent by kill, sigsend, raise */
#define SI_KERNEL       0x80            /* sent by the kernel from somewhere */
#define SI_QUEUE        -1              /* sent by sigqueue */
#define SI_TIMER        -2              /* sent by timer expiration */
#define SI_MESGQ        -3              /* sent by real time mesq state change */
#define SI_ASYNCIO      -4              /* sent by AIO completion */
#define SI_SIGIO        -5              /* sent by queued SIGIO */
#define SI_TKILL        -6              /* sent by tkill system call */
#define SI_DETHREAD     -7              /* sent by execve() killing subsidiary threads */
#define SI_ASYNCNL      -60             /* sent by glibc async name lookup completion */

[topolarity]

The main idea for the note is to detect signal forwarding that's incompatible with our use of siginfo in the segv_handler.

As far as I can tell kill, raise, sigqueue, and tgkill can all "forward" an intercepted SIGSEGV (or trigger an artificial one), but none can preserve the si_code + si_addr that we depend on to identify a safepoint-segfault from a real one.

IIUC the only forwarding that preserves a sufficient amount of the signal metadata is rt_sigqueueinfo(2) (or manually calling / chaining the signal handler, which is what the JVM does)

[gbaraldi]

I think the goal is to distinguish a raise from a MMU triggered segfault