Commit e66b269333

Kendall Condon <goon.pri.low@gmail.com>
2025-05-01 22:52:56
greatly improve capabilities of the fuzzer
This PR significantly improves the capabilities of the fuzzer. The changes made to the fuzzer to accomplish this feat mostly include tracking memory reads from .rodata to determine fresh inputs, new mutations (especially the ones that insert const values from .rodata reads and __sanitizer_conv_const_cmp), and minimizing found inputs. Additionally, the runs per second has greatly been increased due to generating smaller inputs and avoiding clearing the 8-bit pc counters. An additional feature added is that the length of the input file is now stored and the old input file is rerun upon start. Other changes made to the fuzzer include more logical initialization, using one shared file `in` for inputs, creating corpus files with proper sizes, and using hexadecimal-numbered corpus files for simplicity. Furthermore, I added several new fuzz tests to gauge the fuzzer's efficiency. I also tried to add a test for zstandard decompression, which it crashed within 60,000 runs (less than a second.) Bug fixes include: * Fixed a race conditions when multiple fuzzer processes needed to use the same coverage file. * Web interface stats now update even when unique runs is not changing. * Fixed tokenizer.testPropertiesUpheld to allow stray carriage returns since they are valid whitespace.
1 parent a0ec4e2
Changed files (7)
lib
src
codegen
test
standalone
lib/compiler/test_runner.zig
@@ -4,6 +4,7 @@ const builtin = @import("builtin");
 const std = @import("std");
 const testing = std.testing;
 const assert = std.debug.assert;
+const fuzz_abi = std.Build.abi.fuzz;
 
 pub const std_options: std.Options = .{
     .logFn = log,
@@ -57,7 +58,7 @@ pub fn main() void {
     fba.reset();
     if (builtin.fuzz) {
         const cache_dir = opt_cache_dir orelse @panic("missing --cache-dir=[path] argument");
-        fuzzer_init(FuzzerSlice.fromSlice(cache_dir));
+        fuzz_abi.fuzzer_init(.fromSlice(cache_dir));
     }
 
     if (listen) {
@@ -78,7 +79,7 @@ fn mainServer() !void {
     });
 
     if (builtin.fuzz) {
-        const coverage_id = fuzzer_coverage_id();
+        const coverage_id = fuzz_abi.fuzzer_coverage_id();
         try server.serveU64Message(.coverage_id, coverage_id);
     }
 
@@ -152,14 +153,19 @@ fn mainServer() !void {
                 });
             },
             .start_fuzzing => {
+                // This ensures that this code won't be analyzed and hence reference fuzzer symbols
+                // since they are not present.
                 if (!builtin.fuzz) unreachable;
+
                 const index = try server.receiveBody_u32();
                 const test_fn = builtin.test_functions[index];
                 const entry_addr = @intFromPtr(test_fn.func);
+
                 try server.serveU64Message(.fuzz_start_addr, entry_addr);
                 defer if (testing.allocator_instance.deinit() == .leak) std.process.exit(1);
                 is_fuzz_test = false;
-                fuzzer_set_name(test_fn.name.ptr, test_fn.name.len);
+                fuzz_test_index = index;
+
                 test_fn.func() catch |err| switch (err) {
                     error.SkipZigTest => return,
                     else => {
@@ -184,6 +190,8 @@ fn mainServer() !void {
 
 fn mainTerminal() void {
     @disableInstrumentation();
+    if (builtin.fuzz) @panic("fuzz test requires server");
+
     const test_fn_list = builtin.test_functions;
     var ok_count: usize = 0;
     var skip_count: usize = 0;
@@ -333,28 +341,8 @@ pub fn mainSimple() anyerror!void {
     if (failed != 0) std.process.exit(1);
 }
 
-const FuzzerSlice = extern struct {
-    ptr: [*]const u8,
-    len: usize,
-
-    /// Inline to avoid fuzzer instrumentation.
-    inline fn toSlice(s: FuzzerSlice) []const u8 {
-        return s.ptr[0..s.len];
-    }
-
-    /// Inline to avoid fuzzer instrumentation.
-    inline fn fromSlice(s: []const u8) FuzzerSlice {
-        return .{ .ptr = s.ptr, .len = s.len };
-    }
-};
-
 var is_fuzz_test: bool = undefined;
-
-extern fn fuzzer_set_name(name_ptr: [*]const u8, name_len: usize) void;
-extern fn fuzzer_init(cache_dir: FuzzerSlice) void;
-extern fn fuzzer_init_corpus_elem(input_ptr: [*]const u8, input_len: usize) void;
-extern fn fuzzer_start(testOne: *const fn ([*]const u8, usize) callconv(.c) void) void;
-extern fn fuzzer_coverage_id() u64;
+var fuzz_test_index: u32 = undefined;
 
 pub fn fuzz(
     context: anytype,
@@ -385,12 +373,12 @@ pub fn fuzz(
     const global = struct {
         var ctx: @TypeOf(context) = undefined;
 
-        fn fuzzer_one(input_ptr: [*]const u8, input_len: usize) callconv(.c) void {
+        fn test_one(input: fuzz_abi.Slice) callconv(.c) void {
             @disableInstrumentation();
             testing.allocator_instance = .{};
             defer if (testing.allocator_instance.deinit() == .leak) std.process.exit(1);
             log_err_count = 0;
-            testOne(ctx, input_ptr[0..input_len]) catch |err| switch (err) {
+            testOne(ctx, input.toSlice()) catch |err| switch (err) {
                 error.SkipZigTest => return,
                 else => {
                     std.debug.lockStdErr();
@@ -411,10 +399,11 @@ pub fn fuzz(
         testing.allocator_instance = .{};
         defer testing.allocator_instance = prev_allocator_state;
 
-        for (options.corpus) |elem| fuzzer_init_corpus_elem(elem.ptr, elem.len);
-
         global.ctx = context;
-        fuzzer_start(&global.fuzzer_one);
+        fuzz_abi.fuzzer_init_test(&global.test_one, .fromSlice(builtin.test_functions[fuzz_test_index].name));
+        for (options.corpus) |elem|
+            fuzz_abi.fuzzer_new_input(.fromSlice(elem));
+        fuzz_abi.fuzzer_main();
         return;
     }
 
lib/std/Build/abi.zig
@@ -138,6 +138,26 @@ pub const Rebuild = extern struct {
 
 /// ABI bits specifically relating to the fuzzer interface.
 pub const fuzz = struct {
+    pub const TestOne = *const fn (Slice) callconv(.c) void;
+    pub extern fn fuzzer_init(cache_dir_path: Slice) void;
+    pub extern fn fuzzer_coverage_id() u64;
+    pub extern fn fuzzer_init_test(test_one: TestOne, unit_test_name: Slice) void;
+    pub extern fn fuzzer_new_input(bytes: Slice) void;
+    pub extern fn fuzzer_main() void;
+
+    pub const Slice = extern struct {
+        ptr: [*]const u8,
+        len: usize,
+
+        pub fn toSlice(s: Slice) []const u8 {
+            return s.ptr[0..s.len];
+        }
+
+        pub fn fromSlice(s: []const u8) Slice {
+            return .{ .ptr = s.ptr, .len = s.len };
+        }
+    };
+
     /// libfuzzer uses this and its usize is the one that counts. To match the ABI,
     /// make the ints be the size of the target used with libfuzzer.
     ///
lib/std/Build/Fuzz.zig
@@ -252,9 +252,8 @@ pub fn sendUpdate(
     const seen_pcs = cov_header.seenBits();
     const n_runs = @atomicLoad(usize, &cov_header.n_runs, .monotonic);
     const unique_runs = @atomicLoad(usize, &cov_header.unique_runs, .monotonic);
-    if (prev.unique_runs != unique_runs) {
-        // There has been an update.
-        if (prev.unique_runs == 0) {
+    {
+        if (unique_runs != 0 and prev.unique_runs == 0) {
             // We need to send initial context.
             const header: abi.SourceIndexHeader = .{
                 .directories_len = @intCast(coverage_map.coverage.directories.entries.len),
lib/fuzzer.zig
@@ -1,532 +1,1465 @@
 const builtin = @import("builtin");
 const std = @import("std");
-const Allocator = std.mem.Allocator;
+const mem = std.mem;
+const math = std.math;
+const Allocator = mem.Allocator;
 const assert = std.debug.assert;
-const fatal = std.process.fatal;
-const SeenPcsHeader = std.Build.abi.fuzz.SeenPcsHeader;
+const panic = std.debug.panic;
+const abi = std.Build.abi.fuzz;
+const native_endian = builtin.cpu.arch.endian();
 
 pub const std_options = std.Options{
     .logFn = logOverride,
 };
 
-var log_file_buffer: [256]u8 = undefined;
-var log_file_writer: ?std.fs.File.Writer = null;
-
 fn logOverride(
     comptime level: std.log.Level,
     comptime scope: @Type(.enum_literal),
     comptime format: []const u8,
     args: anytype,
 ) void {
-    const fw = if (log_file_writer) |*f| f else f: {
-        const f = fuzzer.cache_dir.createFile("tmp/libfuzzer.log", .{}) catch
-            @panic("failed to open fuzzer log file");
-        log_file_writer = f.writer(&log_file_buffer);
-        break :f &log_file_writer.?;
-    };
+    const f = log_f orelse
+        panic("attempt to use log before initialization, message:\n" ++ format, args);
+    f.lock(.exclusive) catch |e| panic("failed to lock logging file: {t}", .{e});
+    defer f.unlock();
+
+    var buf: [256]u8 = undefined;
+    var fw = f.writer(&buf);
+    const end = f.getEndPos() catch |e| panic("failed to get fuzzer log file end: {t}", .{e});
+    fw.seekTo(end) catch |e| panic("failed to seek to fuzzer log file end: {t}", .{e});
+
     const prefix1 = comptime level.asText();
     const prefix2 = if (scope == .default) ": " else "(" ++ @tagName(scope) ++ "): ";
-    fw.interface.print(prefix1 ++ prefix2 ++ format ++ "\n", args) catch
-        @panic("failed to write to fuzzer log");
-    fw.interface.flush() catch @panic("failed to flush fuzzer log");
+    fw.interface.print(
+        "[{s}] " ++ prefix1 ++ prefix2 ++ format ++ "\n",
+        .{current_test_name orelse "setup"} ++ args,
+    ) catch panic("failed to write to fuzzer log: {t}", .{fw.err.?});
+    fw.interface.flush() catch panic("failed to write to fuzzer log: {t}", .{fw.err.?});
 }
 
-/// Helps determine run uniqueness in the face of recursion.
-export threadlocal var __sancov_lowest_stack: usize = 0;
+var debug_allocator: std.heap.DebugAllocator(.{}) = .init;
+const gpa = switch (builtin.mode) {
+    .Debug => debug_allocator.allocator(),
+    .ReleaseFast, .ReleaseSmall, .ReleaseSafe => std.heap.smp_allocator,
+};
 
-export fn __sanitizer_cov_trace_const_cmp1(arg1: u8, arg2: u8) void {
-    handleCmp(@returnAddress(), arg1, arg2);
-}
+/// Part of `exec`, however seperate to allow it to be set before `exec` is.
+var log_f: ?std.fs.File = null;
+var exec: Executable = .preinit;
+var inst: Instrumentation = .preinit;
+var fuzzer: Fuzzer = undefined;
+var current_test_name: ?[]const u8 = null;
 
-export fn __sanitizer_cov_trace_cmp1(arg1: u8, arg2: u8) void {
-    handleCmp(@returnAddress(), arg1, arg2);
+fn bitsetUsizes(elems: usize) usize {
+    return math.divCeil(usize, elems, @bitSizeOf(usize)) catch unreachable;
 }
 
-export fn __sanitizer_cov_trace_const_cmp2(arg1: u16, arg2: u16) void {
-    handleCmp(@returnAddress(), arg1, arg2);
-}
+const Executable = struct {
+    /// Tracks the hit count for each pc as updated by the process's instrumentation.
+    pc_counters: []u8,
+    /// Read-only memory section containing compiled-in constants found from parsing the executable
+    rodata_addr: usize,
+    rodata_size: usize,
+
+    cache_f: std.fs.Dir,
+    /// Shared copy of all pcs that have been hit stored in a memory-mapped file that can viewed
+    /// while the fuzzer is running.
+    shared_seen_pcs: MemoryMappedList,
+    /// Hash of pcs used to uniquely identify the shared coverage file
+    pc_digest: u64,
+
+    /// A minimal state for this struct which instrumentation can function on.
+    /// Used before this structure is initialized to avoid illegal behavior
+    /// from instrumentation functions being called and using undefined values.
+    pub const preinit: Executable = .{
+        .rodata_addr = 0,
+        .rodata_size = 0,
+        .pc_counters = undefined, // instrumentation works off the __sancov_cntrs section
+        .cache_f = undefined,
+        .shared_seen_pcs = undefined,
+        .pc_digest = undefined,
+    };
 
-export fn __sanitizer_cov_trace_cmp2(arg1: u16, arg2: u16) void {
-    handleCmp(@returnAddress(), arg1, arg2);
-}
+    /// Even on error, this initializes rodata_addr and rodata_size to valid values
+    fn initRodata(self: *Executable) !void {
+        errdefer {
+            self.rodata_addr = 0;
+            self.rodata_size = 0;
+        }
 
-export fn __sanitizer_cov_trace_const_cmp4(arg1: u32, arg2: u32) void {
-    handleCmp(@returnAddress(), arg1, arg2);
-}
+        const exec_path = std.fs.selfExePathAlloc(gpa) catch |e|
+            if (e == error.OutOfMemory) @panic("OOM") else return e;
+        defer gpa.free(exec_path);
+        const exec_file = try std.fs.cwd().openFile(exec_path, .{});
+        defer exec_file.close();
+
+        switch (builtin.object_format) {
+            .elf => {
+                // We use two reader instances since the data they respectively read are
+                // not next to each other in the file.
+                //
+                // Multiple instances is safe since Elf.SectionHeaderIterator always calls
+                // seekTo (which we also use to arbitrarily set the index) and we always
+                // call seekTo to arbitrarily read from the string table.
+                var r_buf: [4096]u8 = undefined;
+                var r = exec_file.reader(&r_buf);
+                var str_r_buf: [4096]u8 = undefined;
+                var str_r = exec_file.reader(&str_r_buf);
+
+                const ehdr: std.elf.Header = try .read(&r.interface);
+                if (ehdr.shstrndx == 0) return error.NoElfStringTable;
+                var shdr_it = ehdr.iterateSectionHeaders(&r);
+
+                shdr_it.index = ehdr.shstrndx;
+                const str_tab_shdr = try shdr_it.next() orelse return error.InvalidElfSection;
+                const str_tab_off = str_tab_shdr.sh_offset;
+
+                shdr_it.index = 0;
+                while (try shdr_it.next()) |shdr| {
+                    const flags: packed struct {
+                        write: bool,
+                        alloc: bool,
+                        execinstr: bool,
+                    } = @bitCast(@as(u3, @truncate(shdr.sh_flags)));
+                    if (shdr.sh_addr == 0 or shdr.sh_size == 0 or flags != @TypeOf(flags){
+                        .alloc = true,
+                        .write = false,
+                        .execinstr = false,
+                    }) continue;
+
+                    const rodata_name = ".rodata\x00";
+                    try str_r.seekTo(try math.add(u64, str_tab_off, shdr.sh_name));
+                    const section_name = str_r.interface.take(rodata_name.len) catch return r.err.?;
+                    if (!std.mem.eql(u8, section_name, rodata_name))
+                        continue;
+
+                    const addr = math.cast(usize, shdr.sh_addr) orelse return error.Overflow;
+                    const size = math.cast(usize, shdr.sh_size) orelse return error.Overflow;
+                    _ = try math.add(usize, addr, size); // make sure there is no overflow
+                    self.rodata_addr = addr;
+                    self.rodata_size = size;
+                    return;
+                }
+                return error.NoRodataSection;
+            },
+            else => return error.UnsupportedObjectFormat,
+        }
+    }
 
-export fn __sanitizer_cov_trace_cmp4(arg1: u32, arg2: u32) void {
-    handleCmp(@returnAddress(), arg1, arg2);
-}
+    fn getCoverageFile(cache_dir: std.fs.Dir, pcs: []const usize, pc_digest: u64) MemoryMappedList {
+        const pc_bitset_usizes = bitsetUsizes(pcs.len);
+        const coverage_file_name = std.fmt.hex(pc_digest);
+        comptime assert(abi.SeenPcsHeader.trailing[0] == .pc_bits_usize);
+        comptime assert(abi.SeenPcsHeader.trailing[1] == .pc_addr);
 
-export fn __sanitizer_cov_trace_const_cmp8(arg1: u64, arg2: u64) void {
-    handleCmp(@returnAddress(), arg1, arg2);
-}
+        var v = cache_dir.makeOpenPath("v", .{}) catch |e|
+            panic("failed to create directory 'v': {t}", .{e});
+        defer v.close();
+        const coverage_file, const populate = if (v.createFile(&coverage_file_name, .{
+            .read = true,
+            // If we create the file, we want to block other processes while we populate it
+            .lock = .exclusive,
+            .exclusive = true,
+        })) |f|
+            .{ f, true }
+        else |e| switch (e) {
+            error.PathAlreadyExists => .{ v.openFile(&coverage_file_name, .{
+                .mode = .read_write,
+                .lock = .shared,
+            }) catch |e2| panic(
+                "failed to open existing coverage file '{s}': {t}",
+                .{ &coverage_file_name, e2 },
+            ), false },
+            else => panic("failed to create coverage file '{s}': {t}", .{ &coverage_file_name, e }),
+        };
 
-export fn __sanitizer_cov_trace_cmp8(arg1: u64, arg2: u64) void {
-    handleCmp(@returnAddress(), arg1, arg2);
-}
+        const coverage_file_len = @sizeOf(abi.SeenPcsHeader) +
+            pc_bitset_usizes * @sizeOf(usize) +
+            pcs.len * @sizeOf(usize);
+        if (populate) {
+            defer coverage_file.lock(.shared) catch |e| panic(
+                "failed to demote lock for coverage file '{s}': {t}",
+                .{ &coverage_file_name, e },
+            );
+            var map = MemoryMappedList.create(coverage_file, 0, coverage_file_len) catch |e| panic(
+                "failed to init memory map for coverage file '{s}': {t}",
+                .{ &coverage_file_name, e },
+            );
+            map.appendSliceAssumeCapacity(mem.asBytes(&abi.SeenPcsHeader{
+                .n_runs = 0,
+                .unique_runs = 0,
+                .pcs_len = pcs.len,
+            }));
+            map.appendNTimesAssumeCapacity(0, pc_bitset_usizes * @sizeOf(usize));
+            map.appendSliceAssumeCapacity(mem.sliceAsBytes(pcs));
+            return map;
+        } else {
+            const size = coverage_file.getEndPos() catch |e| panic(
+                "failed to stat coverage file '{s}': {t}",
+                .{ &coverage_file_name, e },
+            );
+            if (size != coverage_file_len) panic(
+                "incompatible existing coverage file '{s}' (differing lengths: {} != {})",
+                .{ &coverage_file_name, size, coverage_file_len },
+            );
+
+            const map = MemoryMappedList.init(
+                coverage_file,
+                coverage_file_len,
+                coverage_file_len,
+            ) catch |e| panic(
+                "failed to init memory map for coverage file '{s}': {t}",
+                .{ &coverage_file_name, e },
+            );
+
+            const seen_pcs_header: *const abi.SeenPcsHeader = @ptrCast(@volatileCast(map.items));
+            if (seen_pcs_header.pcs_len != pcs.len) panic(
+                "incompatible existing coverage file '{s}' (differing pcs length: {} != {})",
+                .{ &coverage_file_name, seen_pcs_header.pcs_len, pcs.len },
+            );
+            if (mem.indexOfDiff(usize, seen_pcs_header.pcAddrs(), pcs)) |i| panic(
+                "incompatible existing coverage file '{s}' (differing pc at index {d}: {x} != {x})",
+                .{ &coverage_file_name, i, seen_pcs_header.pcAddrs()[i], pcs[i] },
+            );
+
+            return map;
+        }
+    }
 
-export fn __sanitizer_cov_trace_switch(val: u64, cases_ptr: [*]u64) void {
-    const pc = @returnAddress();
-    const len = cases_ptr[0];
-    const val_size_in_bits = cases_ptr[1];
-    const cases = cases_ptr[2..][0..len];
-    fuzzer.traceValue(pc ^ val);
-    _ = val_size_in_bits;
-    _ = cases;
-    //std.log.debug("0x{x}: switch on value {d} ({d} bits) with {d} cases", .{
-    //    pc, val, val_size_in_bits, cases.len,
-    //});
-}
+    pub fn init(cache_dir_path: []const u8) Executable {
+        var self: Executable = undefined;
 
-export fn __sanitizer_cov_trace_pc_indir(callee: usize) void {
-    // Not valuable because we already have pc tracing via 8bit counters.
-    _ = callee;
-    //const pc = @returnAddress();
-    //fuzzer.traceValue(pc ^ callee);
-    //std.log.debug("0x{x}: indirect call to 0x{x}", .{ pc, callee });
-}
-export fn __sanitizer_cov_8bit_counters_init(start: usize, end: usize) void {
-    // clang will emit a call to this function when compiling with code coverage instrumentation.
-    // however fuzzer_init() does not need this information, since it directly reads from the symbol table.
-    _ = start;
-    _ = end;
-}
-export fn __sanitizer_cov_pcs_init(start: usize, end: usize) void {
-    // clang will emit a call to this function when compiling with code coverage instrumentation.
-    // however fuzzer_init() does not need this information, since it directly reads from the symbol table.
-    _ = start;
-    _ = end;
-}
+        const cache_dir = std.fs.cwd().makeOpenPath(cache_dir_path, .{}) catch |e| panic(
+            "failed to open directory '{s}': {t}",
+            .{ cache_dir_path, e },
+        );
+        log_f = cache_dir.createFile("tmp/libfuzzer.log", .{ .truncate = false }) catch |e|
+            panic("failed to create file 'tmp/libfuzzer.log': {t}", .{e});
+        self.cache_f = cache_dir.makeOpenPath("f", .{}) catch |e|
+            panic("failed to open directory 'f': {t}", .{e});
+
+        // Linkers are expected to automatically add symbols prefixed with these for the start and
+        // end of sections whose names are valid C identifiers.
+        const ofmt = builtin.object_format;
+        const section_start_prefix, const section_end_prefix = switch (ofmt) {
+            .elf => .{ "__start_", "__stop_" },
+            .macho => .{ "\x01section$start$__DATA$", "\x01section$end$__DATA$" },
+            else => @compileError("unsupported fuzzing object format '" ++ @tagName(ofmt) ++ "'"),
+        };
 
-fn handleCmp(pc: usize, arg1: u64, arg2: u64) void {
-    fuzzer.traceValue(pc ^ arg1 ^ arg2);
-    //std.log.debug("0x{x}: comparison of {d} and {d}", .{ pc, arg1, arg2 });
-}
+        self.pc_counters = blk: {
+            const pc_counters_start_name = section_start_prefix ++ "__sancov_cntrs";
+            const pc_counters_start = @extern([*]u8, .{
+                .name = pc_counters_start_name,
+                .linkage = .weak,
+            }) orelse panic("missing {s} symbol", .{pc_counters_start_name});
 
-const Fuzzer = struct {
-    rng: std.Random.DefaultPrng,
-    pcs: []const usize,
-    pc_counters: []u8,
-    n_runs: usize,
-    traced_comparisons: std.AutoArrayHashMapUnmanaged(usize, void),
-    /// Tracks which PCs have been seen across all runs that do not crash the fuzzer process.
-    /// Stored in a memory-mapped file so that it can be shared with other
-    /// processes and viewed while the fuzzer is running.
-    seen_pcs: MemoryMappedList,
-    cache_dir: std.fs.Dir,
-    /// Identifies the file name that will be used to store coverage
-    /// information, available to other processes.
-    coverage_id: u64,
-    unit_test_name: []const u8,
-
-    /// The index corresponds to the file name within the f/ subdirectory.
-    /// The string is the input.
-    /// This data is read-only; it caches what is on the filesystem.
-    corpus: std.ArrayListUnmanaged(Input),
-    corpus_directory: std.Build.Cache.Directory,
+            const pc_counters_end_name = section_end_prefix ++ "__sancov_cntrs";
+            const pc_counters_end = @extern([*]u8, .{
+                .name = pc_counters_end_name,
+                .linkage = .weak,
+            }) orelse panic("missing {s} symbol", .{pc_counters_end_name});
 
-    /// The next input that will be given to the testOne function. When the
-    /// current process crashes, this memory-mapped file is used to recover the
-    /// input.
-    ///
-    /// The file size corresponds to the capacity. The length is not stored
-    /// and that is the next thing to work on!
-    input: MemoryMappedList,
+            break :blk pc_counters_start[0 .. pc_counters_end - pc_counters_start];
+        };
 
-    const Input = struct {
-        bytes: []u8,
-        last_traced_comparison: usize,
-    };
+        const pcs = blk: {
+            const pcs_start_name = section_start_prefix ++ "__sancov_pcs1";
+            const pcs_start = @extern([*]usize, .{
+                .name = pcs_start_name,
+                .linkage = .weak,
+            }) orelse panic("missing {s} symbol", .{pcs_start_name});
 
-    const Slice = extern struct {
-        ptr: [*]const u8,
-        len: usize,
+            const pcs_end_name = section_end_prefix ++ "__sancov_pcs1";
+            const pcs_end = @extern([*]usize, .{
+                .name = pcs_end_name,
+                .linkage = .weak,
+            }) orelse panic("missing {s} symbol", .{pcs_end_name});
 
-        fn toZig(s: Slice) []const u8 {
-            return s.ptr[0..s.len];
-        }
+            break :blk pcs_start[0 .. pcs_end - pcs_start];
+        };
 
-        fn fromZig(s: []const u8) Slice {
-            return .{
-                .ptr = s.ptr,
-                .len = s.len,
-            };
+        if (self.pc_counters.len != pcs.len) panic(
+            "pc counters length and pcs length do not match ({} != {})",
+            .{ self.pc_counters.len, pcs.len },
+        );
+
+        self.initRodata() catch |e| if (e != error.UnsupportedObjectFormat) std.log.err(
+            \\failed to enumerate read-only memory: {t}
+            \\efficiency will be severly reduced
+        , .{e});
+
+        self.pc_digest = std.hash.Wyhash.hash(0, mem.sliceAsBytes(pcs));
+        self.shared_seen_pcs = getCoverageFile(cache_dir, pcs, self.pc_digest);
+
+        return self;
+    }
+
+    pub fn pcBitsetIterator(self: Executable) PcBitsetIterator {
+        return .{ .pc_counters = self.pc_counters };
+    }
+
+    /// Iterates over pc_counters returning a bitset for if each of them have been hit
+    pub const PcBitsetIterator = struct {
+        index: usize = 0,
+        pc_counters: []u8,
+
+        pub fn next(self: *PcBitsetIterator) usize {
+            const rest = self.pc_counters[self.index..];
+            if (rest.len >= @bitSizeOf(usize)) {
+                defer self.index += @bitSizeOf(usize);
+                const V = @Vector(@bitSizeOf(usize), u8);
+                return @as(usize, @bitCast(@as(V, @splat(0)) != rest[0..@bitSizeOf(usize)].*));
+            } else if (rest.len != 0) {
+                defer self.index += rest.len;
+                var res: usize = 0;
+                for (0.., rest) |bit_index, byte| {
+                    res |= @shlExact(@as(usize, @intFromBool(byte != 0)), @intCast(bit_index));
+                }
+                return res;
+            } else unreachable;
         }
     };
+};
 
-    fn init(f: *Fuzzer, cache_dir: std.fs.Dir, pc_counters: []u8, pcs: []const usize) !void {
-        f.cache_dir = cache_dir;
-        f.pc_counters = pc_counters;
-        f.pcs = pcs;
-
-        // Choose a file name for the coverage based on a hash of the PCs that will be stored within.
-        const pc_digest = std.hash.Wyhash.hash(0, std.mem.sliceAsBytes(pcs));
-        f.coverage_id = pc_digest;
-        const hex_digest = std.fmt.hex(pc_digest);
-        const coverage_file_path = "v/" ++ hex_digest;
-
-        // Layout of this file:
-        // - Header
-        // - list of PC addresses (usize elements)
-        // - list of hit flag, 1 bit per address (stored in u8 elements)
-        const coverage_file = createFileBail(cache_dir, coverage_file_path, .{
-            .read = true,
-            .truncate = false,
-        });
-        const n_bitset_elems = (pcs.len + @bitSizeOf(usize) - 1) / @bitSizeOf(usize);
-        comptime assert(SeenPcsHeader.trailing[0] == .pc_bits_usize);
-        comptime assert(SeenPcsHeader.trailing[1] == .pc_addr);
-        const bytes_len = @sizeOf(SeenPcsHeader) +
-            n_bitset_elems * @sizeOf(usize) +
-            pcs.len * @sizeOf(usize);
-        const existing_len = coverage_file.getEndPos() catch |err| {
-            fatal("unable to check len of coverage file: {s}", .{@errorName(err)});
+/// Data gathered from instrumentation functions
+/// Seperate from Executable since its state is resetable and changes
+/// Seperate from Fuzzer since it may be needed before fuzzing starts
+const Instrumentation = struct {
+    /// Bitset of seen pcs across all runs excluding fresh pcs.
+    /// This is seperate then shared_seen_pcs because multiple fuzzing processes are likely using
+    /// it which causes contention and unrelated pcs to our campaign being set.
+    seen_pcs: []usize,
+    /// Bitset of seen rodata bytes read across all runs
+    seen_rodata_loads: []usize,
+
+    /// Bitset of run's read bytes that weren't present in seen_loads
+    /// Elements are always zero if !any_new_data_loads
+    new_rodata_loads: []usize,
+    any_new_rodata_loads: bool,
+
+    /// Stores a fresh input's new pcs
+    fresh_pcs: []usize,
+    /// Stores a fresh input's new reads
+    /// Elements are always zero if !any_fresh_rodata_loads
+    fresh_rodata_loads: []usize,
+    any_fresh_rodata_loads: bool,
+
+    /// Pcs which __sanitizer_cov_trace_switch and __sanitizer_cov_trace_const_cmpx
+    /// have been called from and have had their already been added to const_x_vals
+    const_pcs: std.AutoArrayHashMapUnmanaged(usize, void) = .empty,
+    /// Values that have been constant operands in comparisons, switch cases, or memory reads
+    /// There may be duplicates in this array if they came from different addresses, which is
+    /// fine as they are likely more important and hence more likely to be selected.
+    const_vals2: std.ArrayListUnmanaged(u16) = .empty,
+    const_vals4: std.ArrayListUnmanaged(u32) = .empty,
+    const_vals8: std.ArrayListUnmanaged(u64) = .empty,
+    const_vals16: std.ArrayListUnmanaged(u128) = .empty,
+
+    /// A minimal state for this struct which instrumentation can function on.
+    /// Used before this structure is initialized to avoid illegal behavior
+    /// from instrumentation functions being called and using undefined values.
+    pub const preinit: Instrumentation = .{
+        .seen_pcs = undefined, // currently only updated by `Fuzzer`
+        .seen_rodata_loads = undefined,
+        .new_rodata_loads = undefined,
+        .any_new_rodata_loads = undefined,
+        .fresh_pcs = undefined,
+        .fresh_rodata_loads = undefined,
+        .any_fresh_rodata_loads = undefined,
+    };
+
+    pub fn depreinit(self: *Instrumentation) void {
+        self.const_vals2.deinit(gpa);
+        self.const_vals4.deinit(gpa);
+        self.const_vals8.deinit(gpa);
+        self.const_vals16.deinit(gpa);
+        self.* = undefined;
+    }
+
+    pub fn init() Instrumentation {
+        const pc_bitset_usizes = bitsetUsizes(exec.pc_counters.len);
+        const rodata_bitset_usizes = bitsetUsizes(exec.rodata_size);
+        const alloc_usizes = pc_bitset_usizes * 2 + rodata_bitset_usizes * 3;
+        const buf = gpa.alloc(u8, alloc_usizes * @sizeOf(usize)) catch @panic("OOM");
+        var fba_ctx: std.heap.FixedBufferAllocator = .init(buf);
+        const fba = fba_ctx.allocator();
+
+        var self: Instrumentation = .{
+            .seen_pcs = fba.alloc(usize, pc_bitset_usizes) catch unreachable,
+            .seen_rodata_loads = fba.alloc(usize, rodata_bitset_usizes) catch unreachable,
+            .new_rodata_loads = fba.alloc(usize, rodata_bitset_usizes) catch unreachable,
+            .any_new_rodata_loads = undefined,
+            .fresh_pcs = fba.alloc(usize, pc_bitset_usizes) catch unreachable,
+            .fresh_rodata_loads = fba.alloc(usize, rodata_bitset_usizes) catch unreachable,
+            .any_fresh_rodata_loads = undefined,
         };
-        if (existing_len == 0) {
-            coverage_file.setEndPos(bytes_len) catch |err| {
-                fatal("unable to set len of coverage file: {s}", .{@errorName(err)});
-            };
-        } else if (existing_len != bytes_len) {
-            fatal("incompatible existing coverage file (differing lengths)", .{});
+        self.reset();
+        return self;
+    }
+
+    pub fn reset(self: *Instrumentation) void {
+        @memset(self.seen_pcs, 0);
+        @memset(self.seen_rodata_loads, 0);
+        @memset(self.new_rodata_loads, 0);
+        self.any_new_rodata_loads = false;
+        @memset(self.fresh_pcs, 0);
+        @memset(self.fresh_rodata_loads, 0);
+        self.any_fresh_rodata_loads = false;
+        self.const_pcs.clearRetainingCapacity();
+        self.const_vals2.clearRetainingCapacity();
+        self.const_vals4.clearRetainingCapacity();
+        self.const_vals8.clearRetainingCapacity();
+        self.const_vals16.clearRetainingCapacity();
+    }
+
+    /// If false is returned, then the pc is marked as seen
+    pub fn constPcSeen(self: *Instrumentation, pc: usize) bool {
+        return (self.const_pcs.getOrPut(gpa, pc) catch @panic("OOM")).found_existing;
+    }
+
+    pub fn clearNewRodataLoads(self: *Instrumentation) void {
+        if (self.any_new_rodata_loads) {
+            @memset(self.new_rodata_loads, 0);
+            self.any_new_rodata_loads = false;
         }
-        f.seen_pcs = MemoryMappedList.init(coverage_file, existing_len, bytes_len) catch |err| {
-            fatal("unable to init coverage memory map: {s}", .{@errorName(err)});
-        };
-        if (existing_len != 0) {
-            const existing_pcs_bytes = f.seen_pcs.items[@sizeOf(SeenPcsHeader) + @sizeOf(usize) * n_bitset_elems ..][0 .. pcs.len * @sizeOf(usize)];
-            const existing_pcs = std.mem.bytesAsSlice(usize, existing_pcs_bytes);
-            for (existing_pcs, pcs, 0..) |old, new, i| {
-                if (old != new) {
-                    fatal("incompatible existing coverage file (differing PC at index {d}: {x} != {x})", .{
-                        i, old, new,
-                    });
-                }
-            }
-        } else {
-            const header: SeenPcsHeader = .{
-                .n_runs = 0,
-                .unique_runs = 0,
-                .pcs_len = pcs.len,
-            };
-            f.seen_pcs.appendSliceAssumeCapacity(std.mem.asBytes(&header));
-            f.seen_pcs.appendNTimesAssumeCapacity(0, n_bitset_elems * @sizeOf(usize));
-            f.seen_pcs.appendSliceAssumeCapacity(std.mem.sliceAsBytes(pcs));
+    }
+
+    pub fn isFresh(self: *Instrumentation) bool {
+        if (self.any_new_rodata_loads) return true;
+
+        var hit_pcs = exec.pcBitsetIterator();
+        for (self.seen_pcs) |seen_pcs| {
+            if (hit_pcs.next() & ~seen_pcs != 0) return true;
         }
+
+        return false;
     }
 
-    fn initNextInput(f: *Fuzzer) void {
-        while (true) {
-            const i = f.corpus.items.len;
-            var buf: [30]u8 = undefined;
-            const input_sub_path = std.fmt.bufPrint(&buf, "{d}", .{i}) catch unreachable;
-            const input = f.corpus_directory.handle.readFileAlloc(input_sub_path, gpa, .limited(1 << 31)) catch |err| switch (err) {
-                error.FileNotFound => {
-                    // Make this one the next input.
-                    const input_file = f.corpus_directory.handle.createFile(input_sub_path, .{
-                        .exclusive = true,
-                        .truncate = false,
-                        .read = true,
-                    }) catch |e| switch (e) {
-                        error.PathAlreadyExists => continue,
-                        else => fatal("unable to create '{f}{d}: {s}", .{ f.corpus_directory, i, @errorName(err) }),
-                    };
-                    errdefer input_file.close();
-                    // Initialize the mmap for the current input.
-                    f.input = MemoryMappedList.create(input_file, 0, std.heap.page_size_max) catch |e| {
-                        fatal("unable to init memory map for input at '{f}{d}': {s}", .{
-                            f.corpus_directory, i, @errorName(e),
-                        });
-                    };
-                    break;
-                },
-                else => fatal("unable to read '{f}{d}': {s}", .{ f.corpus_directory, i, @errorName(err) }),
-            };
-            errdefer gpa.free(input);
-            f.corpus.append(gpa, .{
-                .bytes = input,
-                .last_traced_comparison = 0,
-            }) catch |err| oom(err);
+    /// Updates fresh_pcs and fresh_rodata_loads
+    /// any_new_rodata_loads and elements of new_rodata_loads are unspecified
+    /// afterwards, but still valid.
+    pub fn setFresh(self: *Instrumentation) void {
+        var hit_pcs = exec.pcBitsetIterator();
+        for (self.seen_pcs, self.fresh_pcs) |seen_pcs, *fresh_pcs| {
+            fresh_pcs.* = hit_pcs.next() & ~seen_pcs;
         }
+
+        mem.swap([]usize, &self.fresh_rodata_loads, &self.new_rodata_loads);
+        mem.swap(bool, &self.any_fresh_rodata_loads, &self.any_new_rodata_loads);
     }
 
-    fn addCorpusElem(f: *Fuzzer, input: []const u8) !void {
-        try f.corpus.append(gpa, .{
-            .bytes = try gpa.dupe(u8, input),
-            .last_traced_comparison = 0,
-        });
+    /// Returns if exec.pc_counters and new_rodata_loads are the same or a superset of fresh_pcs and
+    /// fresh_rodata_loads respectively.
+    pub fn atleastFresh(self: *Instrumentation) bool {
+        var hit_pcs = exec.pcBitsetIterator();
+        for (self.fresh_pcs) |fresh_pcs| {
+            if (fresh_pcs & hit_pcs.next() != fresh_pcs) return false;
+        }
+
+        if (self.any_fresh_rodata_loads) {
+            if (!self.any_new_rodata_loads) return false;
+            for (self.new_rodata_loads, self.fresh_rodata_loads) |n, f| {
+                if (n & f != f) return false;
+            }
+        }
+
+        return true;
     }
 
-    fn start(f: *Fuzzer) !void {
-        const rng = fuzzer.rng.random();
+    /// Updates based off fresh_pcs and fresh_rodata_loads
+    fn updateSeen(self: *Instrumentation) void {
+        comptime assert(abi.SeenPcsHeader.trailing[0] == .pc_bits_usize);
+        const shared_seen_pcs: [*]volatile usize = @ptrCast(
+            exec.shared_seen_pcs.items[@sizeOf(abi.SeenPcsHeader)..].ptr,
+        );
 
-        // Grab the corpus which is namespaced based on `unit_test_name`.
-        {
-            if (f.unit_test_name.len == 0) fatal("test runner never set unit test name", .{});
-            const sub_path = try std.fmt.allocPrint(gpa, "f/{s}", .{f.unit_test_name});
-            f.corpus_directory = .{
-                .handle = f.cache_dir.makeOpenPath(sub_path, .{}) catch |err|
-                    fatal("unable to open corpus directory 'f/{s}': {t}", .{ sub_path, err }),
-                .path = sub_path,
-            };
-            initNextInput(f);
+        for (self.seen_pcs, shared_seen_pcs, self.fresh_pcs) |*seen, *shared_seen, fresh| {
+            seen.* |= fresh;
+            if (fresh != 0)
+                _ = @atomicRmw(usize, shared_seen, .Or, fresh, .monotonic);
         }
 
-        assert(f.n_runs == 0);
-
-        // If the corpus is empty, synthesize one input.
-        if (f.corpus.items.len == 0) {
-            const len = rng.uintLessThanBiased(usize, 200);
-            const slice = try gpa.alloc(u8, len);
-            rng.bytes(slice);
-            f.input.appendSliceAssumeCapacity(slice);
-            try f.corpus.append(gpa, .{
-                .bytes = slice,
-                .last_traced_comparison = 0,
-            });
-            runOne(f, 0);
+        if (self.any_fresh_rodata_loads) {
+            for (self.seen_rodata_loads, self.fresh_rodata_loads) |*s, f|
+                s.* |= f;
         }
+    }
+};
+
+const Fuzzer = struct {
+    arena_ctx: std.heap.ArenaAllocator = .init(gpa),
+    rng: std.Random.DefaultPrng = .init(0),
+    test_one: abi.TestOne,
+    /// The next input that will be given to the testOne function. When the
+    /// current process crashes, this memory-mapped file is used to recover the
+    /// input.
+    input: MemoryMappedList,
+
+    /// Minimized past inputs leading to new pcs or rodata reads. These are randomly mutated in
+    /// round-robin fashion
+    /// Element zero is always an empty input. It is gauraunteed no other elements are empty.
+    corpus: std.ArrayListUnmanaged([]const u8),
+    corpus_pos: usize,
+    /// List of past mutations that have led to new inputs. This way, the mutations that are the
+    /// most effective are the most likely to be selected again. Starts with one of each mutation.
+    mutations: std.ArrayListUnmanaged(Mutation) = .empty,
+
+    /// Filesystem directory containing found inputs for future runs
+    corpus_dir: std.fs.Dir,
+    corpus_dir_idx: usize = 0,
+
+    pub fn init(test_one: abi.TestOne, unit_test_name: []const u8) Fuzzer {
+        var self: Fuzzer = .{
+            .test_one = test_one,
+            .input = undefined,
+            .corpus = .empty,
+            .corpus_pos = 0,
+            .mutations = .empty,
+            .corpus_dir = undefined,
+        };
+        const arena = self.arena_ctx.allocator();
+
+        self.corpus_dir = exec.cache_f.makeOpenPath(unit_test_name, .{}) catch |e|
+            panic("failed to open directory '{s}': {t}", .{ unit_test_name, e });
+        self.input = in: {
+            const f = self.corpus_dir.createFile("in", .{
+                .read = true,
+                .truncate = false,
+                // In case any other fuzz tests are running under the same test name,
+                // the input file is exclusively locked to ensures only one proceeds.
+                .lock = .exclusive,
+                .lock_nonblocking = true,
+            }) catch |e| switch (e) {
+                error.WouldBlock => @panic("input file 'in' is in use by another fuzzing process"),
+                else => panic("failed to create input file 'in': {t}", .{e}),
+            };
+            const size = f.getEndPos() catch |e| panic("failed to stat input file 'in': {t}", .{e});
+            const map = (if (size < std.heap.page_size_max)
+                MemoryMappedList.create(f, 8, std.heap.page_size_max)
+            else
+                MemoryMappedList.init(f, size, size)) catch |e|
+                panic("failed to memory map input file 'in': {t}", .{e});
+
+            // Perform a dry-run of the stored input if there was one in case it might reproduce a
+            // crash.
+            const old_in_len = mem.littleToNative(usize, mem.bytesAsValue(usize, map.items[0..8]).*);
+            if (size >= 8 and old_in_len != 0 and map.items.len - 8 < old_in_len) {
+                test_one(.fromSlice(@volatileCast(map.items[8..][0..old_in_len])));
+            }
+
+            break :in map;
+        };
+        inst.reset();
+
+        self.mutations.appendSlice(gpa, std.meta.tags(Mutation)) catch @panic("OOM");
+        // Ensure there is never an empty corpus. Additionally, an empty input usually leads to
+        // new inputs.
+        self.addInput(&.{});
 
         while (true) {
-            const chosen_index = rng.uintLessThanBiased(usize, f.corpus.items.len);
-            const modification = rng.enumValue(Mutation);
-            f.mutateAndRunOne(chosen_index, modification);
+            var name_buf: [@sizeOf(usize) * 2]u8 = undefined;
+            const bytes = self.corpus_dir.readFileAlloc(
+                std.fmt.bufPrint(&name_buf, "{x}", .{self.corpus_dir_idx}) catch unreachable,
+                arena,
+                .unlimited,
+            ) catch |e| switch (e) {
+                error.FileNotFound => break,
+                else => panic("failed to read corpus file '{x}': {t}", .{ self.corpus_dir_idx, e }),
+            };
+            // No corpus file of length zero will ever be created
+            if (bytes.len == 0)
+                panic("corrupt corpus file '{x}' (len of zero)", .{self.corpus_dir_idx});
+            self.addInput(bytes);
+            self.corpus_dir_idx += 1;
         }
+
+        return self;
     }
 
-    /// `x` represents a possible branch. It is the PC address of the possible
-    /// branch site, hashed together with the value(s) used that determine to
-    /// where it branches.
-    fn traceValue(f: *Fuzzer, x: usize) void {
-        errdefer |err| oom(err);
-        try f.traced_comparisons.put(gpa, x, {});
+    pub fn deinit(self: *Fuzzer) void {
+        self.input.deinit();
+        self.corpus.deinit(gpa);
+        self.mutations.deinit(gpa);
+        self.corpus_dir.close();
+        self.arena_ctx.deinit();
+        self.* = undefined;
     }
 
-    const Mutation = enum {
-        remove_byte,
-        modify_byte,
-        add_byte,
-    };
+    pub fn addInput(self: *Fuzzer, bytes: []const u8) void {
+        self.corpus.append(gpa, bytes) catch @panic("OOM");
+        self.input.clearRetainingCapacity();
+        self.input.ensureTotalCapacity(8 + bytes.len) catch |e|
+            panic("could not resize shared input file: {t}", .{e});
+        self.input.items.len = 8;
+        self.input.appendSliceAssumeCapacity(bytes);
+        self.run();
+        inst.setFresh();
+        inst.updateSeen();
+        inst.clearNewRodataLoads();
+    }
 
-    fn mutateAndRunOne(f: *Fuzzer, corpus_index: usize, mutation: Mutation) void {
-        const rng = fuzzer.rng.random();
-        f.input.clearRetainingCapacity();
-        const old_input = f.corpus.items[corpus_index].bytes;
-        f.input.ensureTotalCapacity(old_input.len + 1) catch @panic("mmap file resize failed");
-        switch (mutation) {
-            .remove_byte => {
-                const omitted_index = rng.uintLessThanBiased(usize, old_input.len);
-                f.input.appendSliceAssumeCapacity(old_input[0..omitted_index]);
-                f.input.appendSliceAssumeCapacity(old_input[omitted_index + 1 ..]);
-            },
-            .modify_byte => {
-                const modified_index = rng.uintLessThanBiased(usize, old_input.len);
-                f.input.appendSliceAssumeCapacity(old_input);
-                f.input.items[modified_index] = rng.int(u8);
-            },
-            .add_byte => {
-                const modified_index = rng.uintLessThanBiased(usize, old_input.len);
-                f.input.appendSliceAssumeCapacity(old_input[0..modified_index]);
-                f.input.appendAssumeCapacity(rng.int(u8));
-                f.input.appendSliceAssumeCapacity(old_input[modified_index..]);
-            },
+    /// Assumes fresh_pcs and fresh_rodata_loads correspond to the input
+    fn minimizeInput(self: *Fuzzer) void {
+        // The minimization technique is kept relatively simple, we sequentially try to remove each
+        // byte and check that the new pcs and memory loads are still hit.
+        var i = self.input.items.len;
+        while (i != 8) {
+            i -= 1;
+            const old = self.input.orderedRemove(i);
+
+            @memset(exec.pc_counters, 0);
+            inst.clearNewRodataLoads();
+            self.run();
+
+            if (!inst.atleastFresh()) {
+                self.input.insertAssumeCapacity(i, old);
+            } else {
+                // This removal may have led to new pcs or memory loads being hit, so we need to
+                // update them to avoid duplicates.
+                inst.setFresh();
+            }
         }
-        runOne(f, corpus_index);
     }
 
-    fn runOne(f: *Fuzzer, corpus_index: usize) void {
-        const header: *volatile SeenPcsHeader = @ptrCast(f.seen_pcs.items[0..@sizeOf(SeenPcsHeader)]);
-
-        f.traced_comparisons.clearRetainingCapacity();
-        @memset(f.pc_counters, 0);
-        __sancov_lowest_stack = std.math.maxInt(usize);
+    fn run(self: *Fuzzer) void {
+        // We don't need to clear pc_counters here; all we care about is new hits and not already
+        // seen hits. Ideally, we wouldn't even have these counters and do something similiar to
+        // what we do for tracking memory (i.e. a __sanitizer_cov function that updates a flag on a
+        // new hit.)
+        assert(!inst.any_new_rodata_loads);
 
-        fuzzer_one(@volatileCast(f.input.items.ptr), f.input.items.len);
+        mem.bytesAsValue(usize, self.input.items[0..8]).* =
+            mem.nativeToLittle(usize, self.input.items.len - 8);
+        self.test_one(.fromSlice(@volatileCast(self.input.items[8..])));
 
-        f.n_runs += 1;
+        const header = mem.bytesAsValue(
+            abi.SeenPcsHeader,
+            exec.shared_seen_pcs.items[0..@sizeOf(abi.SeenPcsHeader)],
+        );
         _ = @atomicRmw(usize, &header.n_runs, .Add, 1, .monotonic);
+    }
 
-        // Track code coverage from all runs.
-        comptime assert(SeenPcsHeader.trailing[0] == .pc_bits_usize);
-        const header_end_ptr: [*]volatile usize = @ptrCast(f.seen_pcs.items[@sizeOf(SeenPcsHeader)..]);
-        const remainder = f.pcs.len % @bitSizeOf(usize);
-        const aligned_len = f.pcs.len - remainder;
-        const seen_pcs = header_end_ptr[0..aligned_len];
-        const pc_counters = std.mem.bytesAsSlice([@bitSizeOf(usize)]u8, f.pc_counters[0..aligned_len]);
-        const V = @Vector(@bitSizeOf(usize), u8);
-        const zero_v: V = @splat(0);
-        var fresh = false;
-        var superset = true;
-
-        for (header_end_ptr[0..pc_counters.len], pc_counters) |*elem, *array| {
-            const v: V = array.*;
-            const mask: usize = @bitCast(v != zero_v);
-            const prev = @atomicRmw(usize, elem, .Or, mask, .monotonic);
-            fresh = fresh or (prev | mask) != prev;
-            superset = superset and (prev | mask) != mask;
-        }
-        if (remainder > 0) {
-            const i = pc_counters.len;
-            const elem = &seen_pcs[i];
-            var mask: usize = 0;
-            for (f.pc_counters[i * @bitSizeOf(usize) ..][0..remainder], 0..) |byte, bit_index| {
-                mask |= @as(usize, @intFromBool(byte != 0)) << @intCast(bit_index);
+    pub fn cycle(self: *Fuzzer) void {
+        const input = self.corpus.items[self.corpus_pos];
+        self.corpus_pos += 1;
+        if (self.corpus_pos == self.corpus.items.len)
+            self.corpus_pos = 0;
+
+        const rng = self.rng.random();
+        while (true) {
+            const m = self.mutations.items[rng.uintLessThanBiased(usize, self.mutations.items.len)];
+            if (!m.mutate(
+                rng,
+                input,
+                &self.input,
+                self.corpus.items,
+                inst.const_vals2.items,
+                inst.const_vals4.items,
+                inst.const_vals8.items,
+                inst.const_vals16.items,
+            )) continue;
+
+            self.run();
+            if (inst.isFresh()) {
+                @branchHint(.unlikely);
+
+                const header = mem.bytesAsValue(
+                    abi.SeenPcsHeader,
+                    exec.shared_seen_pcs.items[0..@sizeOf(abi.SeenPcsHeader)],
+                );
+                _ = @atomicRmw(usize, &header.unique_runs, .Add, 1, .monotonic);
+
+                inst.setFresh();
+                self.minimizeInput();
+                inst.updateSeen();
+                inst.clearNewRodataLoads();
+
+                // An empty-input has always been tried, so if an empty input is fresh then the
+                // test has to be non-deterministic. This has to be checked as duplicate empty
+                // entries are not allowed.
+                if (self.input.items.len - 8 == 0) {
+                    std.log.warn("non-deterministic test (empty input produces different hits)", .{});
+                    _ = @atomicRmw(usize, &header.unique_runs, .Sub, 1, .monotonic);
+                    return;
+                }
+
+                const arena = self.arena_ctx.allocator();
+                const bytes = arena.dupe(u8, @volatileCast(self.input.items[8..])) catch @panic("OOM");
+
+                self.corpus.append(gpa, bytes) catch @panic("OOM");
+                self.mutations.appendNTimes(gpa, m, 6) catch @panic("OOM");
+
+                // Write new corpus to cache
+                var name_buf: [@sizeOf(usize) * 2]u8 = undefined;
+                self.corpus_dir.writeFile(.{
+                    .sub_path = std.fmt.bufPrint(
+                        &name_buf,
+                        "{x}",
+                        .{self.corpus_dir_idx},
+                    ) catch unreachable,
+                    .data = bytes,
+                }) catch |e| panic(
+                    "failed to write corpus file '{x}': {t}",
+                    .{ self.corpus_dir_idx, e },
+                );
+                self.corpus_dir_idx += 1;
             }
-            const prev = @atomicRmw(usize, elem, .Or, mask, .monotonic);
-            fresh = fresh or (prev | mask) != prev;
-            superset = superset and (prev | mask) != mask;
-        }
 
-        // First check if this is a better version of an already existing
-        // input, replacing that input.
-        if (superset or f.traced_comparisons.entries.len >= f.corpus.items[corpus_index].last_traced_comparison) {
-            const new_input = gpa.realloc(f.corpus.items[corpus_index].bytes, f.input.items.len) catch |err| oom(err);
-            f.corpus.items[corpus_index] = .{
-                .bytes = new_input,
-                .last_traced_comparison = f.traced_comparisons.count(),
-            };
-            @memcpy(new_input, @volatileCast(f.input.items));
-            _ = @atomicRmw(usize, &header.unique_runs, .Add, 1, .monotonic);
-            return;
+            break;
         }
+    }
+};
 
-        if (!fresh) return;
+/// Instrumentation must not be triggered before this function is called
+export fn fuzzer_init(cache_dir_path: abi.Slice) void {
+    inst.depreinit();
+    exec = .init(cache_dir_path.toSlice());
+    inst = .init();
+}
 
-        // Input is already committed to the file system, we just need to open a new file
-        // for the next input.
-        // Pre-add it to the corpus list so that it does not get redundantly picked up.
-        f.corpus.append(gpa, .{
-            .bytes = gpa.dupe(u8, @volatileCast(f.input.items)) catch |err| oom(err),
-            .last_traced_comparison = f.traced_comparisons.entries.len,
-        }) catch |err| oom(err);
-        f.input.deinit();
-        initNextInput(f);
+/// Invalid until `fuzzer_init` is called.
+export fn fuzzer_coverage_id() u64 {
+    return exec.pc_digest;
+}
 
-        // TODO: also mark input as "hot" so it gets prioritized for checking mutations above others.
+/// fuzzer_init must be called beforehand
+export fn fuzzer_init_test(test_one: abi.TestOne, unit_test_name: abi.Slice) void {
+    current_test_name = unit_test_name.toSlice();
+    fuzzer = .init(test_one, unit_test_name.toSlice());
+}
 
-        _ = @atomicRmw(usize, &header.unique_runs, .Add, 1, .monotonic);
+/// fuzzer_init_test must be called beforehand
+/// The callee owns the memory of bytes and must not free it until the fuzzer is finished.
+export fn fuzzer_new_input(bytes: abi.Slice) void {
+    // An entry of length zero is always added and duplicates of it are not allowed.
+    if (bytes.len != 0)
+        fuzzer.addInput(bytes.toSlice());
+}
+
+/// fuzzer_init_test must be called first
+export fn fuzzer_main() void {
+    while (true) {
+        fuzzer.cycle();
     }
-};
+}
 
-fn createFileBail(dir: std.fs.Dir, sub_path: []const u8, flags: std.fs.File.CreateFlags) std.fs.File {
-    return dir.createFile(sub_path, flags) catch |err| switch (err) {
-        error.FileNotFound => {
-            const dir_name = std.fs.path.dirname(sub_path).?;
-            dir.makePath(dir_name) catch |e| {
-                fatal("unable to make path '{s}': {s}", .{ dir_name, @errorName(e) });
-            };
-            return dir.createFile(sub_path, flags) catch |e| {
-                fatal("unable to create file '{s}': {s}", .{ sub_path, @errorName(e) });
-            };
-        },
-        else => fatal("unable to create file '{s}': {s}", .{ sub_path, @errorName(err) }),
-    };
+/// Helps determine run uniqueness in the face of recursion.
+/// Currently not used by the fuzzer.
+export threadlocal var __sancov_lowest_stack: usize = 0;
+
+/// Inline since the return address of the callee is required
+inline fn genericConstCmp(T: anytype, val: T, comptime const_vals_field: []const u8) void {
+    if (!inst.constPcSeen(@returnAddress())) {
+        @branchHint(.unlikely);
+        @field(inst, const_vals_field).append(gpa, val) catch @panic("OOM");
+    }
+}
+
+export fn __sanitizer_cov_trace_const_cmp1(const_arg: u8, arg: u8) void {
+    _ = const_arg;
+    _ = arg;
+}
+
+export fn __sanitizer_cov_trace_const_cmp2(const_arg: u16, arg: u16) void {
+    _ = arg;
+    genericConstCmp(u16, const_arg, "const_vals2");
+}
+
+export fn __sanitizer_cov_trace_const_cmp4(const_arg: u32, arg: u32) void {
+    _ = arg;
+    genericConstCmp(u32, const_arg, "const_vals4");
+}
+
+export fn __sanitizer_cov_trace_const_cmp8(const_arg: u64, arg: u64) void {
+    _ = arg;
+    genericConstCmp(u64, const_arg, "const_vals8");
 }
 
-fn oom(err: anytype) noreturn {
-    switch (err) {
-        error.OutOfMemory => @panic("out of memory"),
+export fn __sanitizer_cov_trace_switch(val: u64, cases: [*]const u64) void {
+    _ = val;
+    if (!inst.constPcSeen(@returnAddress())) {
+        @branchHint(.unlikely);
+        const case_bits = cases[1];
+        const cases_slice = cases[2..][0..cases[0]];
+        switch (case_bits) {
+            // 8-bit cases are ignored because they are likely to be randomly generated
+            0...8 => {},
+            9...16 => for (cases_slice) |c|
+                inst.const_vals2.append(gpa, @truncate(c)) catch @panic("OOM"),
+            17...32 => for (cases_slice) |c|
+                inst.const_vals4.append(gpa, @truncate(c)) catch @panic("OOM"),
+            33...64 => for (cases_slice) |c|
+                inst.const_vals8.append(gpa, @truncate(c)) catch @panic("OOM"),
+            else => {}, // Should be impossible
+        }
     }
 }
 
-var debug_allocator: std.heap.GeneralPurposeAllocator(.{}) = .init;
+fn genericLoad(T: anytype, ptr: *align(1) const T, comptime opt_const_vals_field: ?[]const u8) void {
+    const addr = @intFromPtr(ptr);
+    const off = addr -% exec.rodata_addr;
+    if (off >= exec.rodata_size) {
+        @branchHint(.likely);
+        return;
+    }
 
-const gpa = switch (builtin.mode) {
-    .Debug => debug_allocator.allocator(),
-    .ReleaseFast, .ReleaseSmall, .ReleaseSafe => std.heap.smp_allocator,
-};
+    const i = off / @bitSizeOf(usize);
+    // Bits are intentionally truncated since the pointer will almost always be aligned
+    const hit = (@as(usize, (1 << @sizeOf(T)) - 1)) << @intCast(off % @bitSizeOf(usize));
+    const new = hit & ~inst.seen_rodata_loads[i];
+    if (new == 0) {
+        @branchHint(.likely);
+        return;
+    }
 
-var fuzzer: Fuzzer = .{
-    .rng = std.Random.DefaultPrng.init(0),
-    .input = undefined,
-    .pcs = undefined,
-    .pc_counters = undefined,
-    .n_runs = 0,
-    .cache_dir = undefined,
-    .seen_pcs = undefined,
-    .coverage_id = undefined,
-    .unit_test_name = &.{},
-    .corpus = .empty,
-    .corpus_directory = undefined,
-    .traced_comparisons = .empty,
-};
+    inst.new_rodata_loads[i] |= new;
+    inst.any_new_rodata_loads = true;
 
-/// Invalid until `fuzzer_init` is called.
-export fn fuzzer_coverage_id() u64 {
-    return fuzzer.coverage_id;
+    if (opt_const_vals_field) |const_vals_field| {
+        // This may have already been hit and this run is just being used for evaluating the
+        // input, in which case we do not want to readd the same value.
+        if (inst.any_fresh_rodata_loads) {
+            @branchHint(.unlikely);
+            if (new & ~inst.fresh_rodata_loads[i] == 0)
+                return;
+        }
+        @field(inst, const_vals_field).append(gpa, ptr.*) catch @panic("OOM");
+    }
 }
 
-var fuzzer_one: *const fn (input_ptr: [*]const u8, input_len: usize) callconv(.c) void = undefined;
+export fn __sanitizer_cov_load1(ptr: *align(1) const u8) void {
+    genericLoad(u8, ptr, null);
+}
 
-export fn fuzzer_start(testOne: @TypeOf(fuzzer_one)) void {
-    fuzzer_one = testOne;
-    fuzzer.start() catch |err| oom(err);
+export fn __sanitizer_cov_load2(ptr: *align(1) const u16) void {
+    genericLoad(u16, ptr, "const_vals2");
 }
 
-export fn fuzzer_set_name(name_ptr: [*]const u8, name_len: usize) void {
-    fuzzer.unit_test_name = name_ptr[0..name_len];
+export fn __sanitizer_cov_load4(ptr: *align(1) const u32) void {
+    genericLoad(u32, ptr, "const_vals4");
 }
 
-export fn fuzzer_init(cache_dir_struct: Fuzzer.Slice) void {
-    // Linkers are expected to automatically add `__start_<section>` and
-    // `__stop_<section>` symbols when section names are valid C identifiers.
-
-    const ofmt = builtin.object_format;
-
-    const start_symbol_prefix: []const u8 = if (ofmt == .macho)
-        "\x01section$start$__DATA$__"
-    else
-        "__start___";
-    const end_symbol_prefix: []const u8 = if (ofmt == .macho)
-        "\x01section$end$__DATA$__"
-    else
-        "__stop___";
-
-    const pc_counters_start_name = start_symbol_prefix ++ "sancov_cntrs";
-    const pc_counters_start = @extern([*]u8, .{
-        .name = pc_counters_start_name,
-        .linkage = .weak,
-    }) orelse fatal("missing {s} symbol", .{pc_counters_start_name});
-
-    const pc_counters_end_name = end_symbol_prefix ++ "sancov_cntrs";
-    const pc_counters_end = @extern([*]u8, .{
-        .name = pc_counters_end_name,
-        .linkage = .weak,
-    }) orelse fatal("missing {s} symbol", .{pc_counters_end_name});
-
-    const pc_counters = pc_counters_start[0 .. pc_counters_end - pc_counters_start];
-
-    const pcs_start_name = start_symbol_prefix ++ "sancov_pcs1";
-    const pcs_start = @extern([*]usize, .{
-        .name = pcs_start_name,
-        .linkage = .weak,
-    }) orelse fatal("missing {s} symbol", .{pcs_start_name});
-
-    const pcs_end_name = end_symbol_prefix ++ "sancov_pcs1";
-    const pcs_end = @extern([*]usize, .{
-        .name = pcs_end_name,
-        .linkage = .weak,
-    }) orelse fatal("missing {s} symbol", .{pcs_end_name});
-
-    const pcs = pcs_start[0 .. pcs_end - pcs_start];
-
-    const cache_dir_path = cache_dir_struct.toZig();
-    const cache_dir = if (cache_dir_path.len == 0)
-        std.fs.cwd()
-    else
-        std.fs.cwd().makeOpenPath(cache_dir_path, .{ .iterate = true }) catch |err| {
-            fatal("unable to open fuzz directory '{s}': {s}", .{ cache_dir_path, @errorName(err) });
-        };
+export fn __sanitizer_cov_load8(ptr: *align(1) const u64) void {
+    genericLoad(u64, ptr, "const_vals8");
+}
+
+export fn __sanitizer_cov_load16(ptr: *align(1) const u128) void {
+    genericLoad(u128, ptr, "const_vals16");
+}
+
+export fn __sanitizer_cov_trace_cmp1(arg1: u8, arg2: u8) void {
+    _ = arg1;
+    _ = arg2;
+}
+
+export fn __sanitizer_cov_trace_cmp2(arg1: u16, arg2: u16) void {
+    _ = arg1;
+    _ = arg2;
+}
+
+export fn __sanitizer_cov_trace_cmp4(arg1: u32, arg2: u32) void {
+    _ = arg1;
+    _ = arg2;
+}
+
+export fn __sanitizer_cov_trace_cmp8(arg1: u64, arg2: u64) void {
+    _ = arg1;
+    _ = arg2;
+}
+
+export fn __sanitizer_cov_trace_pc_indir(callee: usize) void {
+    // Not valuable because we already have pc tracing via 8bit counters.
+    _ = callee;
+}
+export fn __sanitizer_cov_8bit_counters_init(start: usize, end: usize) void {
+    // clang will emit a call to this function when compiling with code coverage instrumentation.
+    // however, fuzzer_init() does not need this information since it directly reads from the
+    // symbol table.
+    _ = start;
+    _ = end;
+}
+export fn __sanitizer_cov_pcs_init(start: usize, end: usize) void {
+    // clang will emit a call to this function when compiling with code coverage instrumentation.
+    // however, fuzzer_init() does not need this information since it directly reads from the
+    // symbol table.
+    _ = start;
+    _ = end;
+}
 
-    fuzzer.init(cache_dir, pc_counters, pcs) catch |err|
-        fatal("unable to init fuzzer: {s}", .{@errorName(err)});
+/// Copy all of source into dest at position 0.
+/// If the slices overlap, dest.ptr must be <= src.ptr.
+fn volatileCopyForwards(comptime T: type, dest: []volatile T, source: []const volatile T) void {
+    for (dest, source) |*d, s| d.* = s;
 }
 
-export fn fuzzer_init_corpus_elem(input_ptr: [*]const u8, input_len: usize) void {
-    fuzzer.addCorpusElem(input_ptr[0..input_len]) catch |err|
-        fatal("failed to add corpus element: {s}", .{@errorName(err)});
+/// Copy all of source into dest at position 0.
+/// If the slices overlap, dest.ptr must be >= src.ptr.
+fn volatileCopyBackwards(comptime T: type, dest: []volatile T, source: []const volatile T) void {
+    var i = source.len;
+    while (i > 0) {
+        i -= 1;
+        dest[i] = source[i];
+    }
 }
 
+const Mutation = enum {
+    /// Applies .insert_*_span, .push_*_span
+    /// For wtf-8, this limits code units, not code points
+    const max_insert_len = 12;
+    /// Applies to .insert_large_*_span and .push_large_*_span
+    /// 4096 is used as it is a common sector size
+    const max_large_insert_len = 4096;
+    /// Applies to .delete_span and .pop_span
+    const max_delete_len = 16;
+    /// Applies to .set_*span, .move_span, .set_existing_span
+    const max_set_len = 12;
+    const max_replicate_len = 64;
+    const AddValue = i6;
+    const SmallValue = i10;
+
+    delete_byte,
+    delete_span,
+    /// Removes the last byte from the input
+    pop_byte,
+    pop_span,
+    /// Inserts a group of bytes which is already in the input and removes the original copy.
+    move_span,
+    /// Replaces a group of bytes in the input with another group of bytes in the input
+    set_existing_span,
+    insert_existing_span,
+    push_existing_span,
+    set_rng_byte,
+    set_rng_span,
+    insert_rng_byte,
+    insert_rng_span,
+    /// Adds a byte to the end of the input
+    push_rng_byte,
+    push_rng_span,
+    set_zero_byte,
+    set_zero_span,
+    insert_zero_byte,
+    insert_zero_span,
+    push_zero_byte,
+    push_zero_span,
+    /// Inserts a lot of zeros to the end of the input
+    /// This is intended to work with fuzz tests that require data in (large) blocks
+    push_large_zero_span,
+    /// Inserts a group of ascii printable character
+    insert_print_span,
+    /// Inserts a group of character from a...z, A...Z, 0...9, _, and ' '
+    insert_common_span,
+    /// Inserts a group of ascii digits possibly preceded by a `-`
+    insert_integer,
+    /// Code units are evenly distributed between one to four
+    insert_wtf8_char,
+    insert_wtf8_span,
+    /// Inserts a group of bytes from another input
+    insert_splice_span,
+    // utf16 is not yet included since insertion of random bytes should adaquetly check
+    // BMP character, surrogate handling, and occasionally chacters outside of the BMP.
+    set_print_span,
+    set_common_span,
+    set_splice_span,
+    /// Similar to set_splice_span, but the bytes are copied to the same index instead of a random
+    replicate_splice_span,
+    push_print_span,
+    push_common_span,
+    push_integer,
+    push_wtf8_char,
+    push_wtf8_span,
+    push_splice_span,
+    /// Clears a random amount of high bits of a byte
+    truncate_8,
+    truncate_16le,
+    truncate_16be,
+    truncate_32le,
+    truncate_32be,
+    truncate_64le,
+    truncate_64be,
+    /// Flips a random bit
+    xor_1,
+    /// Swaps up to three bits of a byte biased to less bits
+    xor_few_8,
+    /// Swaps up to six bits of a 16-bit value biased to less bits
+    xor_few_16,
+    /// Swaps up to nine bits of a 32-bit value biased to less bits
+    xor_few_32,
+    /// Swaps up to twelve bits of 64-bit value biased to less bits
+    xor_few_64,
+    /// Adds to a byte a value of type AddValue
+    add_8,
+    add_16le,
+    add_16be,
+    add_32le,
+    add_32be,
+    add_64le,
+    add_64be,
+    /// Sets a 16-bit little-endian value to a value of type SmallValue
+    set_small_16le,
+    set_small_16be,
+    set_small_32le,
+    set_small_32be,
+    set_small_64le,
+    set_small_64be,
+    insert_small_16le,
+    insert_small_16be,
+    insert_small_32le,
+    insert_small_32be,
+    insert_small_64le,
+    insert_small_64be,
+    push_small_16le,
+    push_small_16be,
+    push_small_32le,
+    push_small_32be,
+    push_small_64le,
+    push_small_64be,
+    set_const_16,
+    set_const_32,
+    set_const_64,
+    set_const_128,
+    insert_const_16,
+    insert_const_32,
+    insert_const_64,
+    insert_const_128,
+    push_const_16,
+    push_const_32,
+    push_const_64,
+    push_const_128,
+    /// Sets a byte with up to three bits set biased to less bits
+    set_few_8,
+    /// Sets a 16-bit value with up to six bits set biased to less bits
+    set_few_16,
+    /// Sets a 32-bit value with up to nine bits set biased to less bits
+    set_few_32,
+    /// Sets a 64-bit value with up to twelve bits set biased to less bits
+    set_few_64,
+    insert_few_8,
+    insert_few_16,
+    insert_few_32,
+    insert_few_64,
+    push_few_8,
+    push_few_16,
+    push_few_32,
+    push_few_64,
+    /// Randomizes a random contigous group of bits in a byte
+    packed_set_rng_8,
+    packed_set_rng_16le,
+    packed_set_rng_16be,
+    packed_set_rng_32le,
+    packed_set_rng_32be,
+    packed_set_rng_64le,
+    packed_set_rng_64be,
+
+    fn fewValue(rng: std.Random, T: type, comptime bits: u16) T {
+        var result: T = 0;
+        var remaining_bits = rng.intRangeAtMostBiased(u16, 1, bits);
+        while (remaining_bits > 0) {
+            result |= @shlExact(@as(T, 1), rng.int(math.Log2Int(T)));
+            remaining_bits -= 1;
+        }
+        return result;
+    }
+
+    /// Returns if the mutation was applicable to the input
+    pub fn mutate(
+        mutation: Mutation,
+        rng: std.Random,
+        in: []const u8,
+        out: *MemoryMappedList,
+        corpus: []const []const u8,
+        const_vals2: []const u16,
+        const_vals4: []const u32,
+        const_vals8: []const u64,
+        const_vals16: []const u128,
+    ) bool {
+        out.clearRetainingCapacity();
+        const new_capacity = 8 + in.len + @max(
+            16, // builtin 128 value
+            Mutation.max_insert_len,
+            Mutation.max_large_insert_len,
+        );
+        out.ensureTotalCapacity(new_capacity) catch |e|
+            panic("could not resize shared input file: {t}", .{e});
+        out.items.len = 8; // Length field
+
+        const applied = switch (mutation) {
+            inline else => |m| m.comptimeMutate(
+                rng,
+                in,
+                out,
+                corpus,
+                const_vals2,
+                const_vals4,
+                const_vals8,
+                const_vals16,
+            ),
+        };
+        if (!applied)
+            assert(out.items.len == 8)
+        else
+            assert(out.items.len <= new_capacity);
+        return applied;
+    }
+
+    /// Assumes out has already been cleared
+    fn comptimeMutate(
+        comptime mutation: Mutation,
+        rng: std.Random,
+        in: []const u8,
+        out: *MemoryMappedList,
+        corpus: []const []const u8,
+        const_vals2: []const u16,
+        const_vals4: []const u32,
+        const_vals8: []const u64,
+        const_vals16: []const u128,
+    ) bool {
+        const Class = enum { new, remove, rmw, move_span, replicate_splice_span };
+        const class: Class, const class_ctx = switch (mutation) {
+            // zig fmt: off
+            .move_span => .{ .move_span, null },
+            .replicate_splice_span => .{ .replicate_splice_span, null },
+
+            .delete_byte => .{ .remove, .{ .delete, 1 } },
+            .delete_span => .{ .remove, .{ .delete, max_delete_len } },
+
+            .pop_byte => .{ .remove, .{ .pop, 1 } },
+            .pop_span => .{ .remove, .{ .pop, max_delete_len } },
+
+            .set_rng_byte         => .{ .new, .{ .set   ,  1, .rng     , .one              } },
+            .set_zero_byte        => .{ .new, .{ .set   ,  1, .zero    , .one              } },
+            .set_rng_span         => .{ .new, .{ .set   ,  1, .rng     , .many             } },
+            .set_zero_span        => .{ .new, .{ .set   ,  1, .zero    , .many             } },
+            .set_common_span      => .{ .new, .{ .set   ,  1, .common  , .many             } },
+            .set_print_span       => .{ .new, .{ .set   ,  1, .print   , .many             } },
+            .set_existing_span    => .{ .new, .{ .set   ,  2, .existing, .many             } },
+            .set_splice_span      => .{ .new, .{ .set   ,  1, .splice  , .many             } },
+            .set_const_16         => .{ .new, .{ .set   ,  2, .@"const", const_vals2       } },
+            .set_const_32         => .{ .new, .{ .set   ,  4, .@"const", const_vals4       } },
+            .set_const_64         => .{ .new, .{ .set   ,  8, .@"const", const_vals8       } },
+            .set_const_128        => .{ .new, .{ .set   , 16, .@"const", const_vals16      } },
+            .set_small_16le       => .{ .new, .{ .set   ,  2, .small   , .{ i16, .little } } },
+            .set_small_32le       => .{ .new, .{ .set   ,  4, .small   , .{ i32, .little } } },
+            .set_small_64le       => .{ .new, .{ .set   ,  8, .small   , .{ i64, .little } } },
+            .set_small_16be       => .{ .new, .{ .set   ,  2, .small   , .{ i16, .big    } } },
+            .set_small_32be       => .{ .new, .{ .set   ,  4, .small   , .{ i32, .big    } } },
+            .set_small_64be       => .{ .new, .{ .set   ,  8, .small   , .{ i64, .big    } } },
+            .set_few_8            => .{ .new, .{ .set   ,  1, .few     , .{ u8 , 3  }      } },
+            .set_few_16           => .{ .new, .{ .set   ,  2, .few     , .{ u16, 6  }      } },
+            .set_few_32           => .{ .new, .{ .set   ,  4, .few     , .{ u32, 9  }      } },
+            .set_few_64           => .{ .new, .{ .set   ,  8, .few     , .{ u64, 12 }      } },
+
+            .insert_rng_byte      => .{ .new, .{ .insert,  0, .rng     , .one              } },
+            .insert_zero_byte     => .{ .new, .{ .insert,  0, .zero    , .one              } },
+            .insert_rng_span      => .{ .new, .{ .insert,  0, .rng     , .many             } },
+            .insert_zero_span     => .{ .new, .{ .insert,  0, .zero    , .many             } },
+            .insert_print_span    => .{ .new, .{ .insert,  0, .print   , .many             } },
+            .insert_common_span   => .{ .new, .{ .insert,  0, .common  , .many             } },
+            .insert_integer       => .{ .new, .{ .insert,  0, .integer , .many             } },
+            .insert_wtf8_char     => .{ .new, .{ .insert,  0, .wtf8    , .one              } },
+            .insert_wtf8_span     => .{ .new, .{ .insert,  0, .wtf8    , .many             } },
+            .insert_existing_span => .{ .new, .{ .insert,  1, .existing, .many             } },
+            .insert_splice_span   => .{ .new, .{ .insert,  0, .splice  , .many             } },
+            .insert_const_16      => .{ .new, .{ .insert,  0, .@"const", const_vals2       } },
+            .insert_const_32      => .{ .new, .{ .insert,  0, .@"const", const_vals4       } },
+            .insert_const_64      => .{ .new, .{ .insert,  0, .@"const", const_vals8       } },
+            .insert_const_128     => .{ .new, .{ .insert,  0, .@"const", const_vals16      } },
+            .insert_small_16le    => .{ .new, .{ .insert,  0, .small   , .{ i16, .little } } },
+            .insert_small_32le    => .{ .new, .{ .insert,  0, .small   , .{ i32, .little } } },
+            .insert_small_64le    => .{ .new, .{ .insert,  0, .small   , .{ i64, .little } } },
+            .insert_small_16be    => .{ .new, .{ .insert,  0, .small   , .{ i16, .big    } } },
+            .insert_small_32be    => .{ .new, .{ .insert,  0, .small   , .{ i32, .big    } } },
+            .insert_small_64be    => .{ .new, .{ .insert,  0, .small   , .{ i64, .big    } } },
+            .insert_few_8         => .{ .new, .{ .insert,  0, .few     , .{ u8 , 3  }      } },
+            .insert_few_16        => .{ .new, .{ .insert,  0, .few     , .{ u16, 6  }      } },
+            .insert_few_32        => .{ .new, .{ .insert,  0, .few     , .{ u32, 9  }      } },
+            .insert_few_64        => .{ .new, .{ .insert,  0, .few     , .{ u64, 12 }      } },
+
+            .push_rng_byte        => .{ .new, .{ .push  ,  0, .rng     , .one              } },
+            .push_zero_byte       => .{ .new, .{ .push  ,  0, .zero    , .one              } },
+            .push_rng_span        => .{ .new, .{ .push  ,  0, .rng     , .many             } },
+            .push_zero_span       => .{ .new, .{ .push  ,  0, .zero    , .many             } },
+            .push_print_span      => .{ .new, .{ .push  ,  0, .print   , .many             } },
+            .push_common_span     => .{ .new, .{ .push  ,  0, .common  , .many             } },
+            .push_integer         => .{ .new, .{ .push  ,  0, .integer , .many             } },
+            .push_large_zero_span => .{ .new, .{ .push  ,  0, .zero    , .large            } },
+            .push_wtf8_char       => .{ .new, .{ .push  ,  0, .wtf8    , .one              } },
+            .push_wtf8_span       => .{ .new, .{ .push  ,  0, .wtf8    , .many             } },
+            .push_existing_span   => .{ .new, .{ .push  ,  1, .existing, .many             } },
+            .push_splice_span     => .{ .new, .{ .push  ,  0, .splice  , .many             } },
+            .push_const_16        => .{ .new, .{ .push  ,  0, .@"const", const_vals2       } },
+            .push_const_32        => .{ .new, .{ .push  ,  0, .@"const", const_vals4       } },
+            .push_const_64        => .{ .new, .{ .push  ,  0, .@"const", const_vals8       } },
+            .push_const_128       => .{ .new, .{ .push  ,  0, .@"const", const_vals16      } },
+            .push_small_16le      => .{ .new, .{ .push  ,  0, .small   , .{ i16, .little } } },
+            .push_small_32le      => .{ .new, .{ .push  ,  0, .small   , .{ i32, .little } } },
+            .push_small_64le      => .{ .new, .{ .push  ,  0, .small   , .{ i64, .little } } },
+            .push_small_16be      => .{ .new, .{ .push  ,  0, .small   , .{ i16, .big    } } },
+            .push_small_32be      => .{ .new, .{ .push  ,  0, .small   , .{ i32, .big    } } },
+            .push_small_64be      => .{ .new, .{ .push  ,  0, .small   , .{ i64, .big    } } },
+            .push_few_8           => .{ .new, .{ .push  ,  0, .few     , .{ u8 , 3  }      } },
+            .push_few_16          => .{ .new, .{ .push  ,  0, .few     , .{ u16, 6  }      } },
+            .push_few_32          => .{ .new, .{ .push  ,  0, .few     , .{ u32, 9  }      } },
+            .push_few_64          => .{ .new, .{ .push  ,  0, .few     , .{ u64, 12 }      } },
+
+            .xor_1               => .{ .rmw, .{ .xor       , u8 , native_endian, 1  } },
+            .xor_few_8           => .{ .rmw, .{ .xor       , u8 , native_endian, 3  } },
+            .xor_few_16          => .{ .rmw, .{ .xor       , u16, native_endian, 6  } },
+            .xor_few_32          => .{ .rmw, .{ .xor       , u32, native_endian, 9  } },
+            .xor_few_64          => .{ .rmw, .{ .xor       , u64, native_endian, 12 } },
+
+            .truncate_8          => .{ .rmw, .{ .truncate  , u8 , native_endian, {} } },
+            .truncate_16le       => .{ .rmw, .{ .truncate  , u16, .little      , {} } },
+            .truncate_32le       => .{ .rmw, .{ .truncate  , u32, .little      , {} } },
+            .truncate_64le       => .{ .rmw, .{ .truncate  , u64, .little      , {} } },
+            .truncate_16be       => .{ .rmw, .{ .truncate  , u16, .big         , {} } },
+            .truncate_32be       => .{ .rmw, .{ .truncate  , u32, .big         , {} } },
+            .truncate_64be       => .{ .rmw, .{ .truncate  , u64, .big         , {} } },
+
+            .add_8               => .{ .rmw, .{ .add       , i8 , native_endian, {} } },
+            .add_16le            => .{ .rmw, .{ .add       , i16, .little      , {} } },
+            .add_32le            => .{ .rmw, .{ .add       , i32, .little      , {} } },
+            .add_64le            => .{ .rmw, .{ .add       , i64, .little      , {} } },
+            .add_16be            => .{ .rmw, .{ .add       , i16, .big         , {} } },
+            .add_32be            => .{ .rmw, .{ .add       , i32, .big         , {} } },
+            .add_64be            => .{ .rmw, .{ .add       , i64, .big         , {} } },
+
+            .packed_set_rng_8    => .{ .rmw, .{ .packed_rng, u8 , native_endian, {} } },
+            .packed_set_rng_16le => .{ .rmw, .{ .packed_rng, u16, .little      , {} } },
+            .packed_set_rng_32le => .{ .rmw, .{ .packed_rng, u32, .little      , {} } },
+            .packed_set_rng_64le => .{ .rmw, .{ .packed_rng, u64, .little      , {} } },
+            .packed_set_rng_16be => .{ .rmw, .{ .packed_rng, u16, .big         , {} } },
+            .packed_set_rng_32be => .{ .rmw, .{ .packed_rng, u32, .big         , {} } },
+            .packed_set_rng_64be => .{ .rmw, .{ .packed_rng, u64, .big         , {} } },
+            // zig fmt: on
+        };
+
+        switch (class) {
+            .new => {
+                const op: enum {
+                    set,
+                    insert,
+                    push,
+
+                    pub fn maxLen(comptime op: @This(), in_len: usize) usize {
+                        return switch (op) {
+                            .set => @min(in_len, max_set_len),
+                            .insert, .push => max_insert_len,
+                        };
+                    }
+                }, const min_in_len, const data: enum {
+                    rng,
+                    zero,
+                    common,
+                    print,
+                    integer,
+                    wtf8,
+                    existing,
+                    splice,
+                    @"const",
+                    small,
+                    few,
+                }, const data_ctx = class_ctx;
+                const Size = enum { one, many, large };
+                if (in.len < min_in_len) return false;
+                if (data == .@"const" and data_ctx.len == 0) return false;
+
+                const splice_i = if (data == .splice) blk: {
+                    // Element zero always holds an empty input, so we do not select it
+                    if (corpus.len == 1) return false;
+                    break :blk rng.intRangeLessThanBiased(usize, 1, corpus.len);
+                } else undefined;
+
+                // Only needs to be followed for set
+                const len = switch (data) {
+                    else => switch (@as(Size, data_ctx)) {
+                        .one => 1,
+                        .many => rng.intRangeAtMostBiased(usize, 1, op.maxLen(in.len)),
+                        .large => rng.intRangeAtMostBiased(usize, 1, max_large_insert_len),
+                    },
+                    .wtf8 => undefined, // varies by size of each code unit
+                    .splice => rng.intRangeAtMostBiased(usize, 1, @min(
+                        corpus[splice_i].len,
+                        op.maxLen(in.len),
+                    )),
+                    .existing => rng.intRangeAtMostBiased(usize, 1, @min(
+                        in.len,
+                        op.maxLen(in.len),
+                    )),
+                    .@"const" => @sizeOf(@typeInfo(@TypeOf(data_ctx)).pointer.child),
+                    .small, .few => @sizeOf(data_ctx[0]),
+                };
+
+                const i = switch (op) {
+                    .set => rng.uintAtMostBiased(usize, in.len - len),
+                    .insert => rng.uintAtMostBiased(usize, in.len),
+                    .push => in.len,
+                };
+
+                out.appendSliceAssumeCapacity(in[0..i]);
+                switch (data) {
+                    .rng => {
+                        var bytes: [@max(max_insert_len, max_set_len)]u8 = undefined;
+                        rng.bytes(bytes[0..len]);
+                        out.appendSliceAssumeCapacity(bytes[0..len]);
+                    },
+                    .zero => out.appendNTimesAssumeCapacity(0, len),
+                    .common => for (out.addManyAsSliceAssumeCapacity(len)) |*c| {
+                        c.* = switch (rng.int(u6)) {
+                            0 => ' ',
+                            1...10 => |x| '0' + (@as(u8, x) - 1),
+                            11...36 => |x| 'A' + (@as(u8, x) - 11),
+                            37 => '_',
+                            38...63 => |x| 'a' + (@as(u8, x) - 38),
+                        };
+                    },
+                    .print => for (out.addManyAsSliceAssumeCapacity(len)) |*c| {
+                        c.* = rng.intRangeAtMostBiased(u8, 0x20, 0x7E);
+                    },
+                    .integer => {
+                        const negative = len != 0 and rng.boolean();
+                        if (negative) {
+                            out.appendAssumeCapacity('-');
+                        }
+
+                        for (out.addManyAsSliceAssumeCapacity(len - @intFromBool(negative))) |*c| {
+                            c.* = rng.intRangeAtMostBiased(u8, '0', '9');
+                        }
+                    },
+                    .wtf8 => {
+                        comptime assert(op != .set);
+                        var codepoints: usize = if (data_ctx == .one)
+                            1
+                        else
+                            rng.intRangeAtMostBiased(usize, 1, Mutation.max_insert_len / 4);
+
+                        while (true) {
+                            const units1 = rng.int(u2);
+                            const value = switch (units1) {
+                                0 => rng.int(u7),
+                                1 => rng.intRangeAtMostBiased(u11, 0x000080, 0x0007FF),
+                                2 => rng.intRangeAtMostBiased(u16, 0x000800, 0x00FFFF),
+                                3 => rng.intRangeAtMostBiased(u21, 0x010000, 0x10FFFF),
+                            };
+                            const units = @as(u3, units1) + 1;
+
+                            var buf: [4]u8 = undefined;
+                            assert(std.unicode.wtf8Encode(value, &buf) catch unreachable == units);
+                            out.appendSliceAssumeCapacity(buf[0..units]);
+
+                            codepoints -= 1;
+                            if (codepoints == 0) break;
+                        }
+                    },
+                    .existing => {
+                        const j = rng.uintAtMostBiased(usize, in.len - len);
+                        out.appendSliceAssumeCapacity(in[j..][0..len]);
+                    },
+                    .splice => {
+                        const j = rng.uintAtMostBiased(usize, corpus[splice_i].len - len);
+                        out.appendSliceAssumeCapacity(corpus[splice_i][j..][0..len]);
+                    },
+                    .@"const" => out.appendSliceAssumeCapacity(mem.asBytes(
+                        &data_ctx[rng.uintLessThanBiased(usize, data_ctx.len)],
+                    )),
+                    .small => out.appendSliceAssumeCapacity(mem.asBytes(
+                        &mem.nativeTo(data_ctx[0], rng.int(SmallValue), data_ctx[1]),
+                    )),
+                    .few => out.appendSliceAssumeCapacity(mem.asBytes(
+                        &fewValue(rng, data_ctx[0], data_ctx[1]),
+                    )),
+                }
+                switch (op) {
+                    .set => out.appendSliceAssumeCapacity(in[i + len ..]),
+                    .insert => out.appendSliceAssumeCapacity(in[i..]),
+                    .push => {},
+                }
+            },
+            .remove => {
+                if (in.len == 0) return false;
+                const Op = enum { delete, pop };
+                const op: Op, const max_len = class_ctx;
+                // LessThan is used so we don't delete the entire span (which is unproductive since
+                // an empty input has always been tried)
+                const len = if (max_len == 1) 1 else rng.uintLessThanBiased(
+                    usize,
+                    @min(max_len + 1, in.len),
+                );
+                switch (op) {
+                    .delete => {
+                        const i = rng.uintAtMostBiased(usize, in.len - len);
+                        out.appendSliceAssumeCapacity(in[0..i]);
+                        out.appendSliceAssumeCapacity(in[i + len ..]);
+                    },
+                    .pop => out.appendSliceAssumeCapacity(in[0 .. in.len - len]),
+                }
+            },
+            .rmw => {
+                const Op = enum { xor, truncate, add, packed_rng };
+                const op: Op, const T, const endian, const xor_bits = class_ctx;
+                if (in.len < @sizeOf(T)) return false;
+                const Log2T = math.Log2Int(T);
+
+                const idx = rng.uintAtMostBiased(usize, in.len - @sizeOf(T));
+                const old = mem.readInt(T, in[idx..][0..@sizeOf(T)], endian);
+                const new = switch (op) {
+                    .xor => old ^ fewValue(rng, T, xor_bits),
+                    .truncate => old & (@as(T, math.maxInt(T)) >> rng.int(Log2T)),
+                    .add => old +% addend: {
+                        const val = rng.int(Mutation.AddValue);
+                        break :addend if (val == 0) 1 else val;
+                    },
+                    .packed_rng => blk: {
+                        const bits = rng.int(math.Log2Int(T)) +| 1;
+                        break :blk old ^ (rng.int(T) >> bits << rng.uintAtMostBiased(Log2T, bits));
+                    },
+                };
+                out.appendSliceAssumeCapacity(in);
+                mem.bytesAsValue(T, out.items[8..][idx..][0..@sizeOf(T)]).* =
+                    mem.nativeTo(T, new, endian);
+            },
+            .move_span => {
+                if (in.len < 2) return false;
+                // One less since moving whole output will never change anything
+                const len = rng.intRangeAtMostBiased(usize, 1, @min(
+                    in.len - 1,
+                    Mutation.max_set_len,
+                ));
+
+                const src = rng.uintAtMostBiased(usize, in.len - len);
+                // This indexes into the final input
+                const dst = blk: {
+                    const res = rng.uintAtMostBiased(usize, in.len - len - 1);
+                    break :blk res + @intFromBool(res >= src);
+                };
+
+                if (src < dst) {
+                    out.appendSliceAssumeCapacity(in[0..src]);
+                    out.appendSliceAssumeCapacity(in[src + len .. dst + len]);
+                    out.appendSliceAssumeCapacity(in[src..][0..len]);
+                    out.appendSliceAssumeCapacity(in[dst + len ..]);
+                } else {
+                    out.appendSliceAssumeCapacity(in[0..dst]);
+                    out.appendSliceAssumeCapacity(in[src..][0..len]);
+                    out.appendSliceAssumeCapacity(in[dst..src]);
+                    out.appendSliceAssumeCapacity(in[src + len ..]);
+                }
+            },
+            .replicate_splice_span => {
+                if (in.len == 0) return false;
+                if (corpus.len == 1) return false;
+                const from = corpus[rng.intRangeLessThanBiased(usize, 1, corpus.len)];
+                const len = rng.uintLessThanBiased(usize, @min(in.len, from.len, max_replicate_len));
+                const i = rng.uintAtMostBiased(usize, @min(in.len, from.len) - len);
+                out.appendSliceAssumeCapacity(in[0..i]);
+                out.appendSliceAssumeCapacity(from[i..][0..len]);
+                out.appendSliceAssumeCapacity(in[i + len ..]);
+            },
+        }
+        return true;
+    }
+};
+
 /// Like `std.ArrayListUnmanaged(u8)` but backed by memory mapping.
 pub const MemoryMappedList = struct {
     /// Contents of the list.
@@ -654,8 +1587,23 @@ pub const MemoryMappedList = struct {
     fn growCapacity(current: usize, minimum: usize) usize {
         var new = current;
         while (true) {
-            new = std.mem.alignForward(usize, new + new / 2, std.heap.page_size_max);
+            new = mem.alignForward(usize, new + new / 2, std.heap.page_size_max);
             if (new >= minimum) return new;
         }
     }
+
+    pub fn insertAssumeCapacity(l: *MemoryMappedList, i: usize, item: u8) void {
+        assert(l.items.len + 1 <= l.capacity);
+        l.items.len += 1;
+        volatileCopyBackwards(u8, l.items[i + 1 ..], l.items[i .. l.items.len - 1]);
+        l.items[i] = item;
+    }
+
+    pub fn orderedRemove(l: *MemoryMappedList, i: usize) u8 {
+        assert(l.items.len + 1 <= l.capacity);
+        const old = l.items[i];
+        volatileCopyForwards(u8, l.items[i .. l.items.len - 1], l.items[i + 1 ..]);
+        l.items.len -= 1;
+        return old;
+    }
 };
src/codegen/llvm.zig
@@ -1115,7 +1115,7 @@ pub const Object = struct {
                 .NoPrune = false,
                 // Workaround for https://github.com/llvm/llvm-project/pull/106464
                 .StackDepth = true,
-                .TraceLoads = false,
+                .TraceLoads = options.fuzz,
                 .TraceStores = false,
                 .CollectControlFlow = false,
             },
test/standalone/libfuzzer/build.zig
@@ -16,6 +16,7 @@ pub fn build(b: *std.Build) void {
             .optimize = optimize,
             .fuzz = true,
         }),
+        .use_llvm = true, // #23423
     });
 
     b.installArtifact(exe);
test/standalone/libfuzzer/main.zig
@@ -1,29 +1,43 @@
 const std = @import("std");
+const abi = std.Build.abi.fuzz;
+const native_endian = @import("builtin").cpu.arch.endian();
 
-const FuzzerSlice = extern struct {
-    ptr: [*]const u8,
-    len: usize,
+fn testOne(in: abi.Slice) callconv(.c) void {
+    std.debug.assertReadable(in.toSlice());
+}
 
-    fn fromSlice(s: []const u8) FuzzerSlice {
-        return .{ .ptr = s.ptr, .len = s.len };
-    }
-};
+pub fn main() !void {
+    var debug_gpa_ctx: std.heap.DebugAllocator(.{}) = .init;
+    defer _ = debug_gpa_ctx.deinit();
+    const gpa = debug_gpa_ctx.allocator();
 
-extern fn fuzzer_set_name(name_ptr: [*]const u8, name_len: usize) void;
-extern fn fuzzer_init(cache_dir: FuzzerSlice) void;
-extern fn fuzzer_init_corpus_elem(input_ptr: [*]const u8, input_len: usize) void;
-extern fn fuzzer_coverage_id() u64;
+    var args = try std.process.argsWithAllocator(gpa);
+    defer args.deinit();
+    _ = args.skip(); // executable name
 
-pub fn main() !void {
-    var gpa: std.heap.GeneralPurposeAllocator(.{}) = .init;
-    defer _ = gpa.deinit();
-    const args = try std.process.argsAlloc(gpa.allocator());
-    defer std.process.argsFree(gpa.allocator(), args);
+    const cache_dir_path = args.next() orelse @panic("expected cache directory path argument");
+    var cache_dir = try std.fs.cwd().openDir(cache_dir_path, .{});
+    defer cache_dir.close();
+
+    abi.fuzzer_init(.fromSlice(cache_dir_path));
+    abi.fuzzer_init_test(testOne, .fromSlice("test"));
+    abi.fuzzer_new_input(.fromSlice(""));
+    abi.fuzzer_new_input(.fromSlice("hello"));
+
+    const pc_digest = abi.fuzzer_coverage_id();
+    const coverage_file_path = "v/" ++ std.fmt.hex(pc_digest);
+    const coverage_file = try cache_dir.openFile(coverage_file_path, .{});
+    defer coverage_file.close();
 
-    const cache_dir = args[1];
+    var read_buf: [@sizeOf(abi.SeenPcsHeader)]u8 = undefined;
+    var r = coverage_file.reader(&read_buf);
+    const pcs_header = r.interface.takeStruct(abi.SeenPcsHeader, native_endian) catch return r.err.?;
 
-    fuzzer_init(FuzzerSlice.fromSlice(cache_dir));
-    fuzzer_init_corpus_elem("hello".ptr, "hello".len);
-    fuzzer_set_name("test".ptr, "test".len);
-    _ = fuzzer_coverage_id();
+    if (pcs_header.pcs_len == 0)
+        return error.ZeroPcs;
+    const expected_len = @sizeOf(abi.SeenPcsHeader) +
+        try std.math.divCeil(usize, pcs_header.pcs_len, @bitSizeOf(usize)) * @sizeOf(usize) +
+        pcs_header.pcs_len * @sizeOf(usize);
+    if (try coverage_file.getEndPos() != expected_len)
+        return error.WrongEnd;
 }