Enable stack traces on segfault in stage1 on linux

[Rocknest]

Proof of concept for linux x86_64 and i386.

Example stack trace in stage1 (bug #2309):

Received SIGSEGV at instruction 0x83b400 (addr=0x0)
Frame address: 0x7fffe0d615d0
Stack address: 0x7fffe0d61320

long-path/handler.zig:64:5: 0x90185a in segv_handler.handler.segvPanic (userland)
    @panic(""); // Segmentation fault
    ^
long-path/ir.cpp:7481:13: 0x83b3ff in ir_gen_node_raw (long-path/ir.cpp)
    switch (node->type) {
            ^
long-path/ir.cpp:7616:29: 0x833a5a in ir_gen_node_extra (long-path/ir.cpp)
    IrInstruction *result = ir_gen_node_raw(irb, node, scope, lval);
                            ^
long-path/ir.cpp:7622:12: 0x846d46 in ir_gen_node (long-path/ir.cpp)
    return ir_gen_node_extra(irb, node, scope, LValNone);
           ^
====== etc =====
long-path/codegen.cpp:8347:9: 0x80d692 in gen_root_source (long-path/codegen.cpp)
long-path/codegen.cpp:9377:13: 0x80c2d9 in codegen_build_and_link (long-path/codegen.cpp)
long-path/main.cpp:1161:17: 0x7d7dc2 in main (long-path/main.cpp)
???:?:?: 0x7f50e3c61b96 in ??? (???)
???:?:?: 0x630e258d4c544154 in ??? (???)

Received SIGSEGV at instruction 0x9014dd (addr=0x0)
Frame address: 0x7fffe0d606e0
Stack address: 0x7fffe0d60640
Aborted (core dumped)

Due to bugs in stdlib there is a second segfault when it tries to unwind beyond the last stack frame.

[andrewrk]

Nice proof of concept. I think this feature would be great to have enabled by default for some targets, such as linux. I do want the ability to opt out, however. I think a good way to do this will be for me to implement #2189 and then #1439. Then the startup code before main() can look at the root source file and see if the programmer wants to opt out, otherwise do this segfault handling to gain stack traces.

There will also be a way to explicitly install the segfault handler. That API could be used for example in the stage1 compiler, where libc is the one that calls main().

[emekoi]

related: #1740 and windows implementation.

[shawnl]

Also for SIGBUS. (unaligned memory faults).

[Rocknest]

@shawnl how do i even trigger SIGBUS? I tried casting u8 pointers to u64 pointers, but it works fine

[shawnl]

@Rocknest I believe using an aligned SIMD load on unaligned memory will do it. (I've mainly dealt with it on old ARM chips)

[andrewrk]

@Rocknest thanks for this, the above commit is inspired from your PR. I was able to avoid the tricks, and the implementation is in the standard library, on by default for all programs, with a way to opt out. The stage1 compiler opts in by manually calling std.debug.attachSegfaultHandler()