Merge pull request #21587 from alexrp/hexagon-porting

Some initial `hexagon-linux` port work

lib/std/atomic.zig

@@ -433,18 +433,20 @@ pub const cache_line = switch (builtin.cpu.arch) {
     .powerpc64le,
     => 128,
 
+    // https://github.com/llvm/llvm-project/blob/e379094328e49731a606304f7e3559d4f1fa96f9/clang/lib/Basic/Targets/Hexagon.h#L145-L151
+    .hexagon,
+    => if (std.Target.hexagon.featureSetHas(builtin.target.cpu.features, .v73)) 64 else 32,
+
     // - https://github.com/golang/go/blob/3dd58676054223962cd915bb0934d1f9f489d4d2/src/internal/cpu/cpu_arm.go#L7
     // - https://github.com/golang/go/blob/3dd58676054223962cd915bb0934d1f9f489d4d2/src/internal/cpu/cpu_mips.go#L7
     // - https://github.com/golang/go/blob/3dd58676054223962cd915bb0934d1f9f489d4d2/src/internal/cpu/cpu_mipsle.go#L7
     // - https://github.com/golang/go/blob/3dd58676054223962cd915bb0934d1f9f489d4d2/src/internal/cpu/cpu_mips64x.go#L9
     // - https://github.com/golang/go/blob/3dd58676054223962cd915bb0934d1f9f489d4d2/src/internal/cpu/cpu_riscv64.go#L7
-    // - https://github.com/torvalds/linux/blob/3a7e02c040b130b5545e4b115aada7bacd80a2b6/arch/hexagon/include/asm/cache.h#L13
     // - https://github.com/torvalds/linux/blob/3a7e02c040b130b5545e4b115aada7bacd80a2b6/arch/sparc/include/asm/cache.h#L14
     .arm,
     .armeb,
     .thumb,
     .thumbeb,
-    .hexagon,
     .mips,
     .mipsel,
     .mips64,

lib/std/os/linux.zig

@@ -37,8 +37,9 @@ const syscall_bits = switch (native_arch) {
 const arch_bits = switch (native_arch) {
     .x86 => @import("linux/x86.zig"),
     .x86_64 => @import("linux/x86_64.zig"),
-    .aarch64, .aarch64_be => @import("linux/arm64.zig"),
+    .aarch64, .aarch64_be => @import("linux/aarch64.zig"),
-    .arm, .armeb, .thumb, .thumbeb => @import("linux/arm-eabi.zig"),
+    .arm, .armeb, .thumb, .thumbeb => @import("linux/arm.zig"),
+    .hexagon => @import("linux/hexagon.zig"),
     .riscv32 => @import("linux/riscv32.zig"),
     .riscv64 => @import("linux/riscv64.zig"),
     .sparc64 => @import("linux/sparc64.zig"),
@@ -116,7 +117,7 @@ pub const user_desc = arch_bits.user_desc;
 pub const getcontext = arch_bits.getcontext;
 
 pub const tls = @import("linux/tls.zig");
-pub const pie = @import("linux/start_pie.zig");
+pub const pie = @import("linux/pie.zig");
 pub const BPF = @import("linux/bpf.zig");
 pub const IOCTL = @import("linux/ioctl.zig");
 pub const SECCOMP = @import("linux/seccomp.zig");
@@ -277,7 +278,7 @@ pub const MAP = switch (native_arch) {
         UNINITIALIZED: bool = false,
         _: u5 = 0,
     },
-    .s390x => packed struct(u32) {
+    .hexagon, .s390x => packed struct(u32) {
         TYPE: MAP_TYPE,
         FIXED: bool = false,
         ANONYMOUS: bool = false,
@@ -441,7 +442,7 @@ pub const O = switch (native_arch) {
         TMPFILE: bool = false,
         _: u9 = 0,
     },
-    .s390x => packed struct(u32) {
+    .hexagon, .s390x => packed struct(u32) {
         ACCMODE: ACCMODE = .RDONLY,
         _2: u4 = 0,
         CREAT: bool = false,

lib/std/os/linux/hexagon.zig

@@ -0,0 +1,254 @@
+const std = @import("../../std.zig");
+const iovec = std.posix.iovec;
+const iovec_const = std.posix.iovec_const;
+const linux = std.os.linux;
+const SYS = linux.SYS;
+const uid_t = std.os.linux.uid_t;
+const gid_t = std.os.linux.gid_t;
+const pid_t = std.os.linux.pid_t;
+const sockaddr = linux.sockaddr;
+const socklen_t = linux.socklen_t;
+const timespec = std.os.linux.timespec;
+
+pub fn syscall0(number: SYS) usize {
+    return asm volatile ("trap0(#1)"
+        : [ret] "={r0}" (-> usize),
+        : [number] "{r6}" (@intFromEnum(number)),
+        : "memory"
+    );
+}
+
+pub fn syscall1(number: SYS, arg1: usize) usize {
+    return asm volatile ("trap0(#1)"
+        : [ret] "={r0}" (-> usize),
+        : [number] "{r6}" (@intFromEnum(number)),
+          [arg1] "{r0}" (arg1),
+        : "memory"
+    );
+}
+
+pub fn syscall2(number: SYS, arg1: usize, arg2: usize) usize {
+    return asm volatile ("trap0(#1)"
+        : [ret] "={r0}" (-> usize),
+        : [number] "{r6}" (@intFromEnum(number)),
+          [arg1] "{r0}" (arg1),
+          [arg2] "{r1}" (arg2),
+        : "memory"
+    );
+}
+
+pub fn syscall3(number: SYS, arg1: usize, arg2: usize, arg3: usize) usize {
+    return asm volatile ("trap0(#1)"
+        : [ret] "={r0}" (-> usize),
+        : [number] "{r6}" (@intFromEnum(number)),
+          [arg1] "{r0}" (arg1),
+          [arg2] "{r1}" (arg2),
+          [arg3] "{r2}" (arg3),
+        : "memory"
+    );
+}
+
+pub fn syscall4(number: SYS, arg1: usize, arg2: usize, arg3: usize, arg4: usize) usize {
+    return asm volatile ("trap0(#1)"
+        : [ret] "={r0}" (-> usize),
+        : [number] "{r6}" (@intFromEnum(number)),
+          [arg1] "{r0}" (arg1),
+          [arg2] "{r1}" (arg2),
+          [arg3] "{r2}" (arg3),
+          [arg4] "{r3}" (arg4),
+        : "memory"
+    );
+}
+
+pub fn syscall5(number: SYS, arg1: usize, arg2: usize, arg3: usize, arg4: usize, arg5: usize) usize {
+    return asm volatile ("trap0(#1)"
+        : [ret] "={r0}" (-> usize),
+        : [number] "{r6}" (@intFromEnum(number)),
+          [arg1] "{r0}" (arg1),
+          [arg2] "{r1}" (arg2),
+          [arg3] "{r2}" (arg3),
+          [arg4] "{r3}" (arg4),
+          [arg5] "{r4}" (arg5),
+        : "memory"
+    );
+}
+
+pub fn syscall6(
+    number: SYS,
+    arg1: usize,
+    arg2: usize,
+    arg3: usize,
+    arg4: usize,
+    arg5: usize,
+    arg6: usize,
+) usize {
+    return asm volatile ("trap0(#1)"
+        : [ret] "={r0}" (-> usize),
+        : [number] "{r6}" (@intFromEnum(number)),
+          [arg1] "{r0}" (arg1),
+          [arg2] "{r1}" (arg2),
+          [arg3] "{r2}" (arg3),
+          [arg4] "{r3}" (arg4),
+          [arg5] "{r4}" (arg5),
+          [arg6] "{r5}" (arg6),
+        : "memory"
+    );
+}
+
+pub fn clone() callconv(.Naked) usize {
+    // __clone(func, stack, flags, arg, ptid, tls, ctid)
+    //         r0,   r1,    r2,    r3,  r4,   r5,  +0
+    //
+    // syscall(SYS_clone, flags, stack, ptid, ctid, tls)
+    //         r6         r0,    r1,    r2,   r3,   r4
+    asm volatile (
+        \\ allocframe(#8)
+        \\
+        \\ r11 = r0
+        \\ r10 = r3
+        \\
+        \\ r6 = #220 // SYS_clone
+        \\ r0 = r2
+        \\ r1 = and(r1, #-8)
+        \\ r2 = r4
+        \\ r3 = memw(r30 + #8)
+        \\ r4 = r5
+        \\ trap0(#1)
+        \\
+        \\ p0 = cmp.eq(r0, #0)
+        \\ if (!p0) dealloc_return
+        \\
+        \\ r0 = r10
+        \\ callr r11
+        \\
+        \\ r6 = #93 // SYS_exit
+        \\ r0 = #0
+        \\ trap0(#1)
+    );
+}
+
+pub const restore = restore_rt;
+
+pub fn restore_rt() callconv(.Naked) noreturn {
+    asm volatile (
+        \\ trap0(#0)
+        :
+        : [number] "{r6}" (@intFromEnum(SYS.rt_sigreturn)),
+        : "memory"
+    );
+}
+
+pub const F = struct {
+    pub const DUPFD = 0;
+    pub const GETFD = 1;
+    pub const SETFD = 2;
+    pub const GETFL = 3;
+    pub const SETFL = 4;
+    pub const GETLK = 5;
+    pub const SETLK = 6;
+    pub const SETLKW = 7;
+    pub const SETOWN = 8;
+    pub const GETOWN = 9;
+    pub const SETSIG = 10;
+    pub const GETSIG = 11;
+
+    pub const RDLCK = 0;
+    pub const WRLCK = 1;
+    pub const UNLCK = 2;
+
+    pub const SETOWN_EX = 15;
+    pub const GETOWN_EX = 16;
+
+    pub const GETOWNER_UIDS = 17;
+};
+
+pub const timeval = extern struct {
+    sec: time_t,
+    usec: i32,
+};
+
+pub const Flock = extern struct {
+    type: i16,
+    whence: i16,
+    start: off_t,
+    len: off_t,
+    pid: pid_t,
+    __unused: [4]u8,
+};
+
+pub const msghdr = extern struct {
+    name: ?*sockaddr,
+    namelen: socklen_t,
+    iov: [*]iovec,
+    iovlen: i32,
+    __pad1: i32 = 0,
+    control: ?*anyopaque,
+    controllen: socklen_t,
+    __pad2: socklen_t = 0,
+    flags: i32,
+};
+
+pub const msghdr_const = extern struct {
+    name: ?*const sockaddr,
+    namelen: socklen_t,
+    iov: [*]const iovec_const,
+    iovlen: i32,
+    __pad1: i32 = 0,
+    control: ?*const anyopaque,
+    controllen: socklen_t,
+    __pad2: socklen_t = 0,
+    flags: i32,
+};
+
+pub const blksize_t = i32;
+pub const nlink_t = u32;
+pub const time_t = i32;
+pub const mode_t = u32;
+pub const off_t = i64;
+pub const ino_t = u64;
+pub const dev_t = u64;
+pub const blkcnt_t = i64;
+
+// The `stat` definition used by the Linux kernel.
+pub const Stat = extern struct {
+    dev: dev_t,
+    ino: ino_t,
+    mode: mode_t,
+    nlink: nlink_t,
+    uid: uid_t,
+    gid: gid_t,
+    rdev: dev_t,
+    __pad: u32,
+    size: off_t,
+    blksize: blksize_t,
+    __pad2: i32,
+    blocks: blkcnt_t,
+    atim: timespec,
+    mtim: timespec,
+    ctim: timespec,
+    __unused: [2]u32,
+
+    pub fn atime(self: @This()) timespec {
+        return self.atim;
+    }
+
+    pub fn mtime(self: @This()) timespec {
+        return self.mtim;
+    }
+
+    pub fn ctime(self: @This()) timespec {
+        return self.ctim;
+    }
+};
+
+pub const Elf_Symndx = u32;
+
+pub const MMAP2_UNIT = 4096;
+
+pub const VDSO = void;
+
+/// TODO
+pub const ucontext_t = void;
+
+/// TODO
+pub const getcontext = {};

lib/std/os/linux/syscalls.zig

@@ -6852,7 +6852,7 @@ pub const Arc = enum(usize) {
     keyctl = 219,
     clone = 220,
     execve = 221,
-    mmap_pgoff = 222,
+    mmap2 = 222,
     fadvise64_64 = 223,
     swapon = 224,
     swapoff = 225,
@@ -7538,7 +7538,7 @@ pub const Hexagon = enum(usize) {
     keyctl = 219,
     clone = 220,
     execve = 221,
-    mmap_pgoff = 222,
+    mmap2 = 222,
     fadvise64_64 = 223,
     swapon = 224,
     swapoff = 225,

lib/std/os/linux/thumb.zig

@@ -141,7 +141,7 @@ pub fn syscall6(
     );
 }
 
-pub const clone = @import("arm-eabi.zig").clone;
+pub const clone = @import("arm.zig").clone;
 
 pub fn restore() callconv(.Naked) noreturn {
     asm volatile (

lib/std/zig/system.zig

@@ -386,6 +386,16 @@ pub fn resolveTargetQuery(query: Target.Query) DetectError!Target {
         query.cpu_features_sub,
     );
 
+    if (cpu_arch == .hexagon) {
+        // Both LLVM and LLD have broken support for the small data area. Yet LLVM has the feature
+        // on by default for all Hexagon CPUs. Clang sort of solves this by defaulting the `-gpsize`
+        // command line parameter for the Hexagon backend to 0, so that no constants get placed in
+        // the SDA. (This of course breaks down if the user passes `-G <n>` to Clang...) We can't do
+        // the `-gpsize` hack because we can have multiple concurrent LLVM emit jobs, and command
+        // line options in LLVM are shared globally. So just force this feature off. Lovely stuff.
+        result.cpu.features.removeFeature(@intFromEnum(Target.hexagon.Feature.small_data));
+    }
+
     // https://github.com/llvm/llvm-project/issues/105978
     if (result.cpu.arch.isArmOrThumb() and result.floatAbi() == .soft) {
         result.cpu.features.removeFeature(@intFromEnum(Target.arm.Feature.vfp2));

src/codegen/llvm.zig

@@ -12444,6 +12444,7 @@ fn backendSupportsF80(target: std.Target) bool {
 /// if it produces miscompilations.
 fn backendSupportsF16(target: std.Target) bool {
     return switch (target.cpu.arch) {
+        .hexagon,
         .powerpc,
         .powerpcle,
         .powerpc64,

src/link/Elf.zig

@@ -189,7 +189,7 @@ pub fn createEmpty(
     };
 
     const page_size: u32 = switch (target.cpu.arch) {
-        .aarch64, .powerpc64le => 0x10000,
+        .aarch64, .hexagon, .powerpc64le => 0x10000,
         .sparc64 => 0x2000,
         else => 0x1000,
     };

tools/generate_linux_syscalls.zig

@@ -22,6 +22,8 @@ const stdlib_renames = std.StaticStringMap([]const u8).initComptime(.{
     // ARM EABI/Thumb.
     .{ "arm_sync_file_range", "sync_file_range" },
     .{ "arm_fadvise64_64", "fadvise64_64" },
+    // ARC and Hexagon.
+    .{ "mmap_pgoff", "mmap2" },
 });
 
 // Only for newer architectures where we use the C preprocessor.