Merge pull request #20511 from archbirdplus

runtime page size detection
rework GeneralPurposeAllocator to reduce active mapping count
Allocator VTable API update

Author: andrewrk

lib/fuzzer.zig

@@ -480,7 +480,7 @@ pub const MemoryMappedList = struct {
     /// of this ArrayList in accordance with the respective documentation. In
     /// all cases, "invalidated" means that the memory has been passed to this
     /// allocator's resize or free function.
-    items: []align(std.mem.page_size) volatile u8,
+    items: []align(std.heap.page_size_min) volatile u8,
     /// How many bytes this list can hold without allocating additional memory.
     capacity: usize,
 

lib/std/Build/Fuzz/WebServer.zig

@@ -41,7 +41,7 @@ const fuzzer_arch_os_abi = "wasm32-freestanding";
 const fuzzer_cpu_features = "baseline+atomics+bulk_memory+multivalue+mutable_globals+nontrapping_fptoint+reference_types+sign_ext";
 
 const CoverageMap = struct {
-    mapped_memory: []align(std.mem.page_size) const u8,
+    mapped_memory: []align(std.heap.page_size_min) const u8,
     coverage: Coverage,
     source_locations: []Coverage.SourceLocation,
     /// Elements are indexes into `source_locations` pointing to the unit tests that are being fuzz tested.

lib/std/Thread.zig

@@ -769,7 +769,7 @@ const PosixThreadImpl = struct {
         // Use the same set of parameters used by the libc-less impl.
         const stack_size = @max(config.stack_size, 16 * 1024);
         assert(c.pthread_attr_setstacksize(&attr, stack_size) == .SUCCESS);
-        assert(c.pthread_attr_setguardsize(&attr, std.mem.page_size) == .SUCCESS);
+        assert(c.pthread_attr_setguardsize(&attr, std.heap.pageSize()) == .SUCCESS);
 
         var handle: c.pthread_t = undefined;
         switch (c.pthread_create(
@@ -1155,7 +1155,7 @@ const LinuxThreadImpl = struct {
         completion: Completion = Completion.init(.running),
         child_tid: std.atomic.Value(i32) = std.atomic.Value(i32).init(1),
         parent_tid: i32 = undefined,
-        mapped: []align(std.mem.page_size) u8,
+        mapped: []align(std.heap.page_size_min) u8,
 
         /// Calls `munmap(mapped.ptr, mapped.len)` then `exit(1)` without touching the stack (which lives in `mapped.ptr`).
         /// Ported over from musl libc's pthread detached implementation:
@@ -1362,7 +1362,7 @@ const LinuxThreadImpl = struct {
     };
 
     fn spawn(config: SpawnConfig, comptime f: anytype, args: anytype) !Impl {
-        const page_size = std.mem.page_size;
+        const page_size = std.heap.pageSize();
         const Args = @TypeOf(args);
         const Instance = struct {
             fn_args: Args,

lib/std/array_list.zig

@@ -105,21 +105,19 @@ pub fn ArrayListAligned(comptime T: type, comptime alignment: ?u29) type {
             return result;
         }
 
-        /// The caller owns the returned memory. Empties this ArrayList,
-        /// Its capacity is cleared, making deinit() safe but unnecessary to call.
+        /// The caller owns the returned memory. Empties this ArrayList.
+        /// Its capacity is cleared, making `deinit` safe but unnecessary to call.
         pub fn toOwnedSlice(self: *Self) Allocator.Error!Slice {
             const allocator = self.allocator;
 
             const old_memory = self.allocatedSlice();
-            if (allocator.resize(old_memory, self.items.len)) {
-                const result = self.items;
+            if (allocator.remap(old_memory, self.items.len)) |new_items| {
                 self.* = init(allocator);
-                return result;
+                return new_items;
             }
 
             const new_memory = try allocator.alignedAlloc(T, alignment, self.items.len);
             @memcpy(new_memory, self.items);
-            @memset(self.items, undefined);
             self.clearAndFree();
             return new_memory;
         }
@@ -185,8 +183,9 @@ pub fn ArrayListAligned(comptime T: type, comptime alignment: ?u29) type {
             // extra capacity.
             const new_capacity = growCapacity(self.capacity, new_len);
             const old_memory = self.allocatedSlice();
-            if (self.allocator.resize(old_memory, new_capacity)) {
-                self.capacity = new_capacity;
+            if (self.allocator.remap(old_memory, new_capacity)) |new_memory| {
+                self.items.ptr = new_memory.ptr;
+                self.capacity = new_memory.len;
                 return addManyAtAssumeCapacity(self, index, count);
             }
 
@@ -468,8 +467,9 @@ pub fn ArrayListAligned(comptime T: type, comptime alignment: ?u29) type {
             // the allocator implementation would pointlessly copy our
             // extra capacity.
             const old_memory = self.allocatedSlice();
-            if (self.allocator.resize(old_memory, new_capacity)) {
-                self.capacity = new_capacity;
+            if (self.allocator.remap(old_memory, new_capacity)) |new_memory| {
+                self.items.ptr = new_memory.ptr;
+                self.capacity = new_memory.len;
             } else {
                 const new_memory = try self.allocator.alignedAlloc(T, alignment, new_capacity);
                 @memcpy(new_memory[0..self.items.len], self.items);
@@ -707,15 +707,13 @@ pub fn ArrayListAlignedUnmanaged(comptime T: type, comptime alignment: ?u29) typ
         /// Its capacity is cleared, making deinit() safe but unnecessary to call.
         pub fn toOwnedSlice(self: *Self, allocator: Allocator) Allocator.Error!Slice {
             const old_memory = self.allocatedSlice();
-            if (allocator.resize(old_memory, self.items.len)) {
-                const result = self.items;
+            if (allocator.remap(old_memory, self.items.len)) |new_items| {
                 self.* = .empty;
-                return result;
+                return new_items;
             }
 
             const new_memory = try allocator.alignedAlloc(T, alignment, self.items.len);
             @memcpy(new_memory, self.items);
-            @memset(self.items, undefined);
             self.clearAndFree(allocator);
             return new_memory;
         }
@@ -1031,9 +1029,9 @@ pub fn ArrayListAlignedUnmanaged(comptime T: type, comptime alignment: ?u29) typ
             }
 
             const old_memory = self.allocatedSlice();
-            if (allocator.resize(old_memory, new_len)) {
-                self.capacity = new_len;
-                self.items.len = new_len;
+            if (allocator.remap(old_memory, new_len)) |new_items| {
+                self.capacity = new_items.len;
+                self.items = new_items;
                 return;
             }
 
@@ -1099,8 +1097,9 @@ pub fn ArrayListAlignedUnmanaged(comptime T: type, comptime alignment: ?u29) typ
             // the allocator implementation would pointlessly copy our
             // extra capacity.
             const old_memory = self.allocatedSlice();
-            if (allocator.resize(old_memory, new_capacity)) {
-                self.capacity = new_capacity;
+            if (allocator.remap(old_memory, new_capacity)) |new_memory| {
+                self.items.ptr = new_memory.ptr;
+                self.capacity = new_memory.len;
             } else {
                 const new_memory = try allocator.alignedAlloc(T, alignment, new_capacity);
                 @memcpy(new_memory[0..self.items.len], self.items);

lib/std/c.zig

@@ -3,7 +3,7 @@ const builtin = @import("builtin");
 const c = @This();
 const maxInt = std.math.maxInt;
 const assert = std.debug.assert;
-const page_size = std.mem.page_size;
+const page_size = std.heap.page_size_min;
 const native_abi = builtin.abi;
 const native_arch = builtin.cpu.arch;
 const native_os = builtin.os.tag;
@@ -2227,6 +2227,39 @@ pub const SC = switch (native_os) {
     .linux => linux.SC,
     else => void,
 };
+
+pub const _SC = switch (native_os) {
+    .driverkit, .ios, .macos, .tvos, .visionos, .watchos => enum(c_int) {
+        PAGESIZE = 29,
+    },
+    .dragonfly => enum(c_int) {
+        PAGESIZE = 47,
+    },
+    .freebsd => enum(c_int) {
+        PAGESIZE = 47,
+    },
+    .fuchsia => enum(c_int) {
+        PAGESIZE = 30,
+    },
+    .haiku => enum(c_int) {
+        PAGESIZE = 27,
+    },
+    .linux => enum(c_int) {
+        PAGESIZE = 30,
+    },
+    .netbsd => enum(c_int) {
+        PAGESIZE = 28,
+    },
+    .openbsd => enum(c_int) {
+        PAGESIZE = 28,
+    },
+    .solaris, .illumos => enum(c_int) {
+        PAGESIZE = 11,
+        NPROCESSORS_ONLN = 15,
+    },
+    else => void,
+};
+
 pub const SEEK = switch (native_os) {
     .linux => linux.SEEK,
     .emscripten => emscripten.SEEK,
@@ -7834,6 +7867,11 @@ pub const MAP = switch (native_os) {
     else => void,
 };
 
+pub const MREMAP = switch (native_os) {
+    .linux => linux.MREMAP,
+    else => void,
+};
+
 /// Used by libc to communicate failure. Not actually part of the underlying syscall.
 pub const MAP_FAILED: *anyopaque = @ptrFromInt(maxInt(usize));
 
@@ -9232,7 +9270,7 @@ pub extern "c" fn getpwnam(name: [*:0]const u8) ?*passwd;
 pub extern "c" fn getpwuid(uid: uid_t) ?*passwd;
 pub extern "c" fn getrlimit64(resource: rlimit_resource, rlim: *rlimit) c_int;
 pub extern "c" fn lseek64(fd: fd_t, offset: i64, whence: c_int) i64;
-pub extern "c" fn mmap64(addr: ?*align(std.mem.page_size) anyopaque, len: usize, prot: c_uint, flags: c_uint, fd: fd_t, offset: i64) *anyopaque;
+pub extern "c" fn mmap64(addr: ?*align(page_size) anyopaque, len: usize, prot: c_uint, flags: c_uint, fd: fd_t, offset: i64) *anyopaque;
 pub extern "c" fn open64(path: [*:0]const u8, oflag: O, ...) c_int;
 pub extern "c" fn openat64(fd: c_int, path: [*:0]const u8, oflag: O, ...) c_int;
 pub extern "c" fn pread64(fd: fd_t, buf: [*]u8, nbyte: usize, offset: i64) isize;
@@ -9324,13 +9362,13 @@ pub extern "c" fn signalfd(fd: fd_t, mask: *const sigset_t, flags: u32) c_int;
 
 pub extern "c" fn prlimit(pid: pid_t, resource: rlimit_resource, new_limit: *const rlimit, old_limit: *rlimit) c_int;
 pub extern "c" fn mincore(
-    addr: *align(std.mem.page_size) anyopaque,
+    addr: *align(page_size) anyopaque,
     length: usize,
     vec: [*]u8,
 ) c_int;
 
 pub extern "c" fn madvise(
-    addr: *align(std.mem.page_size) anyopaque,
+    addr: *align(page_size) anyopaque,
     length: usize,
     advice: u32,
 ) c_int;
@@ -9428,6 +9466,10 @@ pub const posix_memalign = switch (native_os) {
     .dragonfly, .netbsd, .freebsd, .solaris, .openbsd, .linux, .macos, .ios, .tvos, .watchos, .visionos => private.posix_memalign,
     else => {},
 };
+pub const sysconf = switch (native_os) {
+    .solaris => solaris.sysconf,
+    else => private.sysconf,
+};
 
 pub const sf_hdtr = switch (native_os) {
     .freebsd, .macos, .ios, .tvos, .watchos, .visionos => extern struct {
@@ -9471,6 +9513,7 @@ pub extern "c" fn write(fd: fd_t, buf: [*]const u8, nbyte: usize) isize;
 pub extern "c" fn pwrite(fd: fd_t, buf: [*]const u8, nbyte: usize, offset: off_t) isize;
 pub extern "c" fn mmap(addr: ?*align(page_size) anyopaque, len: usize, prot: c_uint, flags: MAP, fd: fd_t, offset: off_t) *anyopaque;
 pub extern "c" fn munmap(addr: *align(page_size) const anyopaque, len: usize) c_int;
+pub extern "c" fn mremap(addr: ?*align(page_size) const anyopaque, old_len: usize, new_len: usize, flags: MREMAP, ...) *anyopaque;
 pub extern "c" fn mprotect(addr: *align(page_size) anyopaque, len: usize, prot: c_uint) c_int;
 pub extern "c" fn link(oldpath: [*:0]const u8, newpath: [*:0]const u8) c_int;
 pub extern "c" fn linkat(oldfd: fd_t, oldpath: [*:0]const u8, newfd: fd_t, newpath: [*:0]const u8, flags: c_int) c_int;
@@ -9823,7 +9866,6 @@ pub const SCM = solaris.SCM;
 pub const SETCONTEXT = solaris.SETCONTEXT;
 pub const SETUSTACK = solaris.GETUSTACK;
 pub const SFD = solaris.SFD;
-pub const _SC = solaris._SC;
 pub const cmsghdr = solaris.cmsghdr;
 pub const ctid_t = solaris.ctid_t;
 pub const file_obj = solaris.file_obj;
@@ -9840,7 +9882,6 @@ pub const priority = solaris.priority;
 pub const procfs = solaris.procfs;
 pub const projid_t = solaris.projid_t;
 pub const signalfd_siginfo = solaris.signalfd_siginfo;
-pub const sysconf = solaris.sysconf;
 pub const taskid_t = solaris.taskid_t;
 pub const zoneid_t = solaris.zoneid_t;
 
@@ -9997,6 +10038,7 @@ pub const host_t = darwin.host_t;
 pub const ipc_space_t = darwin.ipc_space_t;
 pub const ipc_space_port_t = darwin.ipc_space_port_t;
 pub const kern_return_t = darwin.kern_return_t;
+pub const vm_size_t = darwin.vm_size_t;
 pub const kevent64 = darwin.kevent64;
 pub const kevent64_s = darwin.kevent64_s;
 pub const mach_absolute_time = darwin.mach_absolute_time;
@@ -10168,6 +10210,7 @@ const private = struct {
     extern "c" fn socket(domain: c_uint, sock_type: c_uint, protocol: c_uint) c_int;
     extern "c" fn stat(noalias path: [*:0]const u8, noalias buf: *Stat) c_int;
     extern "c" fn sigaltstack(ss: ?*stack_t, old_ss: ?*stack_t) c_int;
+    extern "c" fn sysconf(sc: c_int) c_long;
 
     extern "c" fn pthread_setname_np(thread: pthread_t, name: [*:0]const u8) c_int;
     extern "c" fn getcontext(ucp: *ucontext_t) c_int;
@@ -10202,7 +10245,7 @@ const private = struct {
     extern "c" fn __getrusage50(who: c_int, usage: *rusage) c_int;
     extern "c" fn __gettimeofday50(noalias tv: ?*timeval, noalias tz: ?*timezone) c_int;
     extern "c" fn __libc_thr_yield() c_int;
-    extern "c" fn __msync13(addr: *align(std.mem.page_size) const anyopaque, len: usize, flags: c_int) c_int;
+    extern "c" fn __msync13(addr: *align(page_size) const anyopaque, len: usize, flags: c_int) c_int;
     extern "c" fn __nanosleep50(rqtp: *const timespec, rmtp: ?*timespec) c_int;
     extern "c" fn __sigaction14(sig: c_int, noalias act: ?*const Sigaction, noalias oact: ?*Sigaction) c_int;
     extern "c" fn __sigfillset14(set: ?*sigset_t) void;

lib/std/c/solaris.zig

@@ -154,10 +154,6 @@ pub const AF_SUN = struct {
     pub const NOPLM = 0x00000004;
 };
 
-pub const _SC = struct {
-    pub const NPROCESSORS_ONLN = 15;
-};
-
 pub const procfs = struct {
     pub const misc_header = extern struct {
         size: u32,

lib/std/crypto/tlcsprng.zig

@@ -42,7 +42,7 @@ var install_atfork_handler = std.once(struct {
     }
 }.do);
 
-threadlocal var wipe_mem: []align(mem.page_size) u8 = &[_]u8{};
+threadlocal var wipe_mem: []align(std.heap.page_size_min) u8 = &[_]u8{};
 
 fn tlsCsprngFill(_: *anyopaque, buffer: []u8) void {
     if (os_has_arc4random) {
@@ -77,15 +77,15 @@ fn tlsCsprngFill(_: *anyopaque, buffer: []u8) void {
         } else {
             // Use a static thread-local buffer.
             const S = struct {
-                threadlocal var buf: Context align(mem.page_size) = .{
+                threadlocal var buf: Context align(std.heap.page_size_min) = .{
                     .init_state = .uninitialized,
                     .rng = undefined,
                 };
             };
             wipe_mem = mem.asBytes(&S.buf);
         }
     }
-    const ctx = @as(*Context, @ptrCast(wipe_mem.ptr));
+    const ctx: *Context = @ptrCast(wipe_mem.ptr);
 
     switch (ctx.init_state) {
         .uninitialized => {
@@ -141,7 +141,7 @@ fn childAtForkHandler() callconv(.c) void {
 }
 
 fn fillWithCsprng(buffer: []u8) void {
-    const ctx = @as(*Context, @ptrCast(wipe_mem.ptr));
+    const ctx: *Context = @ptrCast(wipe_mem.ptr);
     return ctx.rng.fill(buffer);
 }
 
@@ -157,7 +157,7 @@ fn initAndFill(buffer: []u8) void {
     // the `std.options.cryptoRandomSeed` function is provided.
     std.options.cryptoRandomSeed(&seed);
 
-    const ctx = @as(*Context, @ptrCast(wipe_mem.ptr));
+    const ctx: *Context = @ptrCast(wipe_mem.ptr);
     ctx.rng = Rng.init(seed);
     std.crypto.secureZero(u8, &seed);