Expand SPARC FP-based unwinding support to more OSs

[alexrp]

[alexrp]

#25569 gets us most of the way there.

However, there's still a bug in DWARF unwinding in ReleaseSafe. Apply this diff:

diff --git a/lib/std/debug.zig b/lib/std/debug.zig
index 545a7b8a19..a2893ae659 100644
--- a/lib/std/debug.zig
+++ b/lib/std/debug.zig
@@ -888,8 +888,6 @@ const StackIterator = union(enum) {
 .powerpcle,
 .powerpc64,
 .powerpc64le,
- .sparc,
- .sparc64,
 => .ideal,
 // https://developer.apple.com/documentation/xcode/writing-arm64-code-for-apple-platforms#Respect-the-purpose-of-specific-CPU-registers
 .aarch64 => if (builtin.target.os.tag.isDarwin()) .safe else .unsafe,

With this program:

const std = @import("std");
pub fn main() void {
 foo();
 bar(true);
}

noinline fn foo() void {
 std.debug.dumpCurrentStackTrace(.{});
}

noinline fn bar(x: bool) void {
 baz(x);
}

noinline fn baz(x: bool) void {
 // _ = x; @panic("oh no");
 if (x) @as(*allowzero volatile u32, @ptrFromInt(0)).* = 42;
}

Run:

zig4 build-exe crash.zig -target sparc64-linux -OReleaseSafe && qemu-sparc64 ./crash

Result:

/home/alexrp/Source/scratch/crash.zig:3:8: 0x1107a34 in main (crash)
 foo();
 ^
/home/alexrp/Source/scratch/crash.zig:3:8: 0x1107a34 in main (crash)
 foo();
 ^
/home/alexrp/Source/scratch/crash.zig:3:8: 0x1107a34 in main (crash)
 foo();
 ^
/home/alexrp/Source/scratch/crash.zig:3:8: 0x1107a34 in main (crash)
 foo();

...snip...

/home/alexrp/Source/scratch/crash.zig:3:8: 0x1107a34 in main (crash)
 foo();
 ^
Stopping trace after 10001 frames (large frame count may indicate broken debug info)
Segmentation fault at address 0x0
/home/alexrp/Source/scratch/crash.zig:17:41: 0x1107f59 in baz (crash)
 if (x) @as(*allowzero volatile u32, @ptrFromInt(0)).* = 42;
 ^
/home/alexrp/Source/ziglang/zig/lib/std/start.zig:618:22: 0x1107694 in callMain (crash)
 root.main();
 ^
fish: Job 1, 'qemu-sparc64 ./crash' terminated by signal SIGABRT (Abort)

I'm not sure what's up with this or why it doesn't happen in Debug. But I also can't be bothered to investigate further right now since FP unwinding is superior on SPARC anyway.

[alexrp]

Also just noting that I won't be implementing SPARC unwinding support for any OS other than Linux just based on headers; unlike other architectures, the signal context ABI is really unusual and tricky on SPARC, so it's not really practical to do without real hardware or a reliable userspace emulator.

The OSs with missing support are mainly OpenBSD, NetBSD, and Solaris. Haiku too, technically, but I suspect that port isn't anywhere near functional.

[alexrp]

After talking to @mlugg, we decided that we don't care about DWARF unwinding on SPARC since FP unwinding just seems better in all contexts.

So this issue now tracks adding FP unwinding support for SPARC on OpenBSD and NetBSD.