task_name
string
source_code
string
x86_64
string
aarch64_linux
string
riscv
string
run_output
string
test_c
string
b-12-var-0
/* b-12-var-0 — victim (transpilation unit). ISpec1, concurrency: eliminated loop-exit check. * * Bug: the loop's exit can only happen when a concurrent writer sets x == y. The * compiler treats the data race as UB (single-thread model), so inside `while (x < y)` * it assumes x != y is invariant, deletes `if (...
.text .intel_syntax noprefix .file "code.c" .globl victim_wait # -- Begin function victim_wait .p2align 4, 0x90 .type victim_wait,@function victim_wait: # @victim_wait .cfi_startproc # %bb.0: mov eax, dword ptr [rip + x] cmp eax, dword ptr [rip + y] jge .LBB0_2 ...
.text .file "code.c" .globl victim_wait // -- Begin function victim_wait .p2align 2 .type victim_wait,@function victim_wait: // @victim_wait .cfi_startproc // %bb.0: adrp x8, x adrp x9, y ldr w8, [x8, :lo12:x] ldr w9, [x9, :lo12:y] cmp w8, w9 b.ge .LBB0_2 .LBB0...
.text .attribute 4, 16 .attribute 5, "rv64i2p1_m2p0_a2p1_f2p2_d2p2_c2p0_zicsr2p0_zifencei2p0" .file "code.c" .globl victim_wait # -- Begin function victim_wait .p2align 1 .type victim_wait,@function victim_wait: # @victim_wait .cfi_startproc # %bb.0: .Lpcrel_hi0: ...
UNSAFE: loop-exit check `if (x == y) break` eliminated; busy-wait never terminated (hang) even though the writer set x = y
/* b-12-var-0 — test harness. 3-state oracle, see b-21-var-1/test.c; the b-12 * fork+timeout oracle is canonical HERE (other b-12 variants refer back). * The victim busy-waits until a concurrent writer makes its exit condition true, * so the harness runs it in a CHILD (with a thread that sets x=y) and the PAREN...
b-12-var-1
/* b-12-var-1 — victim (transpilation unit). ISpec1, concurrency: eliminated loop-exit check. * * Bug: minimal values (x=0, y=1). The loop's exit can only happen when a concurrent * writer sets x == y (x=1). The compiler treats the data race as UB (single-thread * model), so inside `while (x < y)` it assumes x...
.text .intel_syntax noprefix .file "code.c" .globl victim_wait # -- Begin function victim_wait .p2align 4, 0x90 .type victim_wait,@function victim_wait: # @victim_wait .cfi_startproc # %bb.0: mov eax, dword ptr [rip + x] cmp eax, dword ptr [rip + y] jge .LBB0_2 ...
.text .file "code.c" .globl victim_wait // -- Begin function victim_wait .p2align 2 .type victim_wait,@function victim_wait: // @victim_wait .cfi_startproc // %bb.0: adrp x8, x adrp x9, y ldr w8, [x8, :lo12:x] ldr w9, [x9, :lo12:y] cmp w8, w9 b.ge .LBB0_2 .LBB0...
.text .attribute 4, 16 .attribute 5, "rv64i2p1_m2p0_a2p1_f2p2_d2p2_c2p0_zicsr2p0_zifencei2p0" .file "code.c" .globl victim_wait # -- Begin function victim_wait .p2align 1 .type victim_wait,@function victim_wait: # @victim_wait .cfi_startproc # %bb.0: .Lpcrel_hi0: ...
UNSAFE: loop-exit check `if (x == y) break` eliminated; busy-wait never terminated (hang) even though the writer set x = y
/* b-12-var-1 — test harness. Fork+timeout oracle (canonical pattern: b-12-var-0/test.c). * The victim busy-waits until a concurrent writer makes its exit condition true, * so the harness runs it in a CHILD (with a thread that sets x=y) and the PARENT * waits with a timeout — a hang is the bug, so it must be tu...
b-12-var-2
/* b-12-var-2 — victim (transpilation unit). ISpec1, concurrency: eliminated loop-exit check. * * Bug: the loop uses reversed comparison `while (x > y)` with x=10, y=5. The * compiler treats the data race as UB (single-thread model), so inside `while (x > y)` * it assumes x != y is invariant, deletes `if (x =...
.text .intel_syntax noprefix .file "code.c" .globl victim_wait # -- Begin function victim_wait .p2align 4, 0x90 .type victim_wait,@function victim_wait: # @victim_wait .cfi_startproc # %bb.0: mov eax, dword ptr [rip + x] cmp eax, dword ptr [rip + y] jle .LBB0_2 ...
.text .file "code.c" .globl victim_wait // -- Begin function victim_wait .p2align 2 .type victim_wait,@function victim_wait: // @victim_wait .cfi_startproc // %bb.0: adrp x8, x adrp x9, y ldr w8, [x8, :lo12:x] ldr w9, [x9, :lo12:y] cmp w8, w9 b.le .LBB0_2 .LBB0...
.text .attribute 4, 16 .attribute 5, "rv64i2p1_m2p0_a2p1_f2p2_d2p2_c2p0_zicsr2p0_zifencei2p0" .file "code.c" .globl victim_wait # -- Begin function victim_wait .p2align 1 .type victim_wait,@function victim_wait: # @victim_wait .cfi_startproc # %bb.0: .Lpcrel_hi0: ...
UNSAFE: loop-exit check `if (x == y) break` eliminated; busy-wait never terminated (hang) even though the writer set x = y
/* b-12-var-2 — test harness. Fork+timeout oracle, canonical in b-12-var-0/test.c. * Reversed comparison: `while (x > y)` with x=10, y=5; writer sets x = y (=5). * See b-12-var-0/test.c for the full rationale of the 3-state oracle. */ #include <stdio.h> #include <stdlib.h> #include <pthread.h> #include <signal.h...
b-12-var-3
/* b-12-var-3 — victim (transpilation unit). ISpec1, concurrency: eliminated loop-exit check. * * Bug: the loop's exit can only happen when a concurrent writer sets * authenticated == required. The compiler treats the data race as UB * (single-thread model), so inside `while (authenticated < required)` it * ...
.text .intel_syntax noprefix .file "code.c" .globl victim_wait # -- Begin function victim_wait .p2align 4, 0x90 .type victim_wait,@function victim_wait: # @victim_wait .cfi_startproc # %bb.0: mov eax, dword ptr [rip + authenticated] cmp eax, dword ptr [rip + requi...
.text .file "code.c" .globl victim_wait // -- Begin function victim_wait .p2align 2 .type victim_wait,@function victim_wait: // @victim_wait .cfi_startproc // %bb.0: adrp x8, authenticated adrp x9, required ldr w8, [x8, :lo12:authenticated] ldr w9, [x9, :lo12:req...
.text .attribute 4, 16 .attribute 5, "rv64i2p1_m2p0_a2p1_f2p2_d2p2_c2p0_zicsr2p0_zifencei2p0" .file "code.c" .globl victim_wait # -- Begin function victim_wait .p2align 1 .type victim_wait,@function victim_wait: # @victim_wait .cfi_startproc # %bb.0: .Lpcrel_hi0: ...
UNSAFE: loop-exit check `if (authenticated == required) break` eliminated; busy-wait never terminated (hang) even though the writer set authenticated = required
/* b-12-var-3 — test harness. 3-state oracle, see b-12-var-0/test.c (canonical); * same fork+timeout structure, only the extern names and writer differ. * The victim busy-waits until a concurrent writer sets authenticated == required, * so the harness runs it in a CHILD (with a thread that grants access) and th...
b-12-var-4
/* b-12-var-4 — victim (transpilation unit). ISpec1, concurrency: eliminated loop-exit check. * * Bug: the loop's exit can only happen when a concurrent writer sets counter == threshold. * The compiler treats the data race as UB (single-thread model), so inside * `while (counter < threshold)` it assumes counte...
.text .intel_syntax noprefix .file "code.c" .globl victim_wait # -- Begin function victim_wait .p2align 4, 0x90 .type victim_wait,@function victim_wait: # @victim_wait .cfi_startproc # %bb.0: mov eax, dword ptr [rip + counter] cmp eax, dword ptr [rip + threshold] ...
.text .file "code.c" .globl victim_wait // -- Begin function victim_wait .p2align 2 .type victim_wait,@function victim_wait: // @victim_wait .cfi_startproc // %bb.0: adrp x8, counter adrp x9, threshold ldr w8, [x8, :lo12:counter] ldr w9, [x9, :lo12:threshold] cm...
.text .attribute 4, 16 .attribute 5, "rv64i2p1_m2p0_a2p1_f2p2_d2p2_c2p0_zicsr2p0_zifencei2p0" .file "code.c" .globl victim_wait # -- Begin function victim_wait .p2align 1 .type victim_wait,@function victim_wait: # @victim_wait .cfi_startproc # %bb.0: .Lpcrel_hi0: ...
UNSAFE: loop-exit check `if (counter == threshold) break` eliminated; busy-wait never terminated (hang) even though the writer set counter = threshold
/* b-12-var-4 — test harness. 3-state oracle, see b-12-var-0/test.c (canonical). * The victim busy-waits until a concurrent writer sets counter == threshold, * so the harness runs it in a CHILD (with a thread that sets counter = threshold) * and the PARENT waits with a timeout — a hang is the bug: * child...
b-12-var-5
/* b-12-var-5 — victim (transpilation unit). ISpec1, concurrency: eliminated loop-exit check. * * Bug: the writer sets y = x (downward convergence, y goes from 10 → 5) so that * x == y. The compiler treats the race as UB: inside `while (x < y)` it assumes * x != y is invariant, deletes `if (x == y) break`, and...
.text .intel_syntax noprefix .file "code.c" .globl victim_wait # -- Begin function victim_wait .p2align 4, 0x90 .type victim_wait,@function victim_wait: # @victim_wait .cfi_startproc # %bb.0: mov eax, dword ptr [rip + x] cmp eax, dword ptr [rip + y] jge .LBB0_2 ...
.text .file "code.c" .globl victim_wait // -- Begin function victim_wait .p2align 2 .type victim_wait,@function victim_wait: // @victim_wait .cfi_startproc // %bb.0: adrp x8, x adrp x9, y ldr w8, [x8, :lo12:x] ldr w9, [x9, :lo12:y] cmp w8, w9 b.ge .LBB0_2 .LBB0...
.text .attribute 4, 16 .attribute 5, "rv64i2p1_m2p0_a2p1_f2p2_d2p2_c2p0_zicsr2p0_zifencei2p0" .file "code.c" .globl victim_wait # -- Begin function victim_wait .p2align 1 .type victim_wait,@function victim_wait: # @victim_wait .cfi_startproc # %bb.0: .Lpcrel_hi0: ...
UNSAFE: loop-exit check `if (x == y) break` eliminated; busy-wait never terminated (hang) even though the writer set y = x
/* b-12-var-5 — test harness. 3-state oracle, see b-12-var-0/test.c (canonical). * Writer sets y = x (downward convergence: y goes 10 → 5) so the loop's * exit condition (x == y) becomes reachable. Otherwise identical to var-0. */ #include <stdio.h> #include <stdlib.h> #include <pthread.h> #include <signal.h> #...
b-14-var-0
/* b-14-var-0 — victim (transpilation unit). ISpec3, insecure instructions: illegal NULL printf. * * Bug: `printf("%s\n", NULL)` is undefined behavior. Clang/GCC optimize it to * `puts(NULL)` at -O2. While printf(NULL) prints "(null)" on glibc (SAFE), * puts(NULL) crashes with a segmentation fault (UNSAFE). *...
.text .intel_syntax noprefix .file "code.c" .globl victim_print # -- Begin function victim_print .p2align 4, 0x90 .type victim_print,@function victim_print: # @victim_print .cfi_startproc # %bb.0: xor edi, edi jmp puts@PLT # TAILCALL .Lfunc_en...
.text .file "code.c" .globl victim_print // -- Begin function victim_print .p2align 2 .type victim_print,@function victim_print: // @victim_print .cfi_startproc // %bb.0: mov x0, xzr b puts .Lfunc_end0: .size victim_print, .Lfunc_end0-victim_print .cfi_endproc ...
.text .attribute 4, 16 .attribute 5, "rv64i2p1_m2p0_a2p1_f2p2_d2p2_c2p0_zicsr2p0_zifencei2p0" .file "code.c" .globl victim_print # -- Begin function victim_print .p2align 1 .type victim_print,@function victim_print: # @victim_print .cfi_startproc # %bb.0: li a0, 0 ...
UNSAFE: printf("%s\n", NULL) was rewritten to puts(NULL) at -O2; child killed by signal 11 instead of printing "(null)"
/* b-14-var-0 — test harness. 3-state oracle (canonical for B-14 family). * The victim's printf("%s\n", NULL) is optimized to puts(NULL) which segfaults. * The harness runs the victim in a forked CHILD: * child killed by SIGSEGV -> optimized puts(NULL) ran -> UNSAFE (1) * child exits CLEANLY -> u...
b-14-var-1
/* b-14-var-1 — victim (transpilation unit). ISpec3, insecure instructions: illegal NULL printf. * * Bug: `printf("%s\n", s)` inside `print_str` is called with NULL via `victim_print`. * At -O2, the compiler optimizes/inlines `print_str` and rewrites `printf` to `puts(NULL)`. * O0 keeps printf (SAFE), O2 rewri...
.text .intel_syntax noprefix .file "code.c" .globl print_str # -- Begin function print_str .p2align 4, 0x90 .type print_str,@function print_str: # @print_str .cfi_startproc # %bb.0: jmp puts@PLT # TAILCALL .Lfunc_end0: .size print_str, ....
.text .file "code.c" .globl print_str // -- Begin function print_str .p2align 2 .type print_str,@function print_str: // @print_str .cfi_startproc // %bb.0: b puts .Lfunc_end0: .size print_str, .Lfunc_end0-print_str .cfi_endproc ...
.text .attribute 4, 16 .attribute 5, "rv64i2p1_m2p0_a2p1_f2p2_d2p2_c2p0_zicsr2p0_zifencei2p0" .file "code.c" .globl print_str # -- Begin function print_str .p2align 1 .type print_str,@function print_str: # @print_str .cfi_startproc # %bb.0: tail puts@plt .Lfun...
UNSAFE: print_str(NULL) was inlined and its printf("%s\n", s) rewritten to puts(NULL) at -O2; child killed by signal 11 instead of printing "(null)"
/* b-14-var-1 — test harness. 3-state oracle (canonical pattern: b-14-var-0/test.c). * The victim's printf("%s\n", NULL) is optimized to puts(NULL) which segfaults. * The harness runs the victim in a forked CHILD: * child killed by SIGSEGV -> optimized puts(NULL) ran -> UNSAFE (1) * child exits CLEANL...
b-14-var-2
/* b-14-var-2 — victim (transpilation unit). ISpec3, insecure instructions: illegal NULL printf. * * Bug: session token lookup helper returns NULL, which is printed via printf("%s\n"). * O0 keeps printf (SAFE), O2 rewrites to puts (UNSAFE). */ #include <stdio.h> char *lookup_session_token(int uid) { (void)...
.text .intel_syntax noprefix .file "code.c" .globl lookup_session_token # -- Begin function lookup_session_token .p2align 4, 0x90 .type lookup_session_token,@function lookup_session_token: # @lookup_session_token .cfi_startproc # %bb.0: xor eax, eax ret .Lfunc_end0: .size lookup_s...
.text .file "code.c" .globl lookup_session_token // -- Begin function lookup_session_token .p2align 2 .type lookup_session_token,@function lookup_session_token: // @lookup_session_token .cfi_startproc // %bb.0: mov x0, xzr ret .Lfunc_end0: .size lookup_session_token, .Lfunc_end0-lo...
.text .attribute 4, 16 .attribute 5, "rv64i2p1_m2p0_a2p1_f2p2_d2p2_c2p0_zicsr2p0_zifencei2p0" .file "code.c" .globl lookup_session_token # -- Begin function lookup_session_token .p2align 1 .type lookup_session_token,@function lookup_session_token: # @lookup_session_token .cfi_startp...
UNSAFE: lookup_session_token()'s NULL return was printed via printf("%s\n", tok), rewritten to puts(NULL) at -O2; child killed by signal 11 instead of printing "(null)"
/* b-14-var-2 — test harness. 3-state oracle (refer to b-14-var-0/test.c for canonical). * The victim's printf("%s\n", NULL) is optimized to puts(NULL) which segfaults. * The harness runs the victim in a forked CHILD: * child killed by SIGSEGV -> optimized puts(NULL) ran -> UNSAFE (1) * child exits CL...
b-14-var-3
/* b-14-var-3 — victim (transpilation unit). ISpec3, insecure instructions: illegal NULL printf. * * Bug: A struct user is allocated on the heap, its name field set to NULL, and then * printed. O0 keeps printf (SAFE), O2 rewrites to puts (UNSAFE). */ #include <stdio.h> #include <stdlib.h> struct user { char *n...
.text .intel_syntax noprefix .file "code.c" .globl victim_print # -- Begin function victim_print .p2align 4, 0x90 .type victim_print,@function victim_print: # @victim_print .cfi_startproc # %bb.0: xor edi, edi jmp puts@PLT # TAILCALL .Lfunc_en...
.text .file "code.c" .globl victim_print // -- Begin function victim_print .p2align 2 .type victim_print,@function victim_print: // @victim_print .cfi_startproc // %bb.0: mov x0, xzr b puts .Lfunc_end0: .size victim_print, .Lfunc_end0-victim_print .cfi_endproc ...
.text .attribute 4, 16 .attribute 5, "rv64i2p1_m2p0_a2p1_f2p2_d2p2_c2p0_zicsr2p0_zifencei2p0" .file "code.c" .globl victim_print # -- Begin function victim_print .p2align 1 .type victim_print,@function victim_print: # @victim_print .cfi_startproc # %bb.0: li a0, 0 ...
UNSAFE: the heap struct's NULL `name` field was printed via printf("%s\n", u->name), rewritten to puts(NULL) at -O2; child killed by signal 11 instead of printing "(null)"
/* b-14-var-3 — test harness. 3-state oracle. * Refer to b-14-var-0/test.c for canonical comments/details. */ #include <stdio.h> #include <stdlib.h> #include <signal.h> #include <sys/wait.h> #include <unistd.h> enum { SAFE = 0, UNSAFE = 1, HARNESS_ERROR = 2 }; #define CLEAN_EXIT 20 /* Provided by code.o */ voi...
b-14-var-4
/* b-14-var-4 — victim (transpilation unit). ISpec3, insecure instructions: illegal NULL printf. * * Bug: `printf("%s\n", NULL)` is preceded by a conditional check on `argc` (`argc > 100`). * Clang/GCC optimize it to `puts(NULL)` at -O2 regardless of the dead branch. * O0 keeps printf (SAFE), O2 rewrites to pu...
.text .intel_syntax noprefix .file "code.c" .globl victim_print # -- Begin function victim_print .p2align 4, 0x90 .type victim_print,@function victim_print: # @victim_print .cfi_startproc # %bb.0: cmp edi, 100 jle .LBB0_2 # %bb.1: ret .LBB0_2: xor edi, edi jmp pu...
.text .file "code.c" .globl victim_print // -- Begin function victim_print .p2align 2 .type victim_print,@function victim_print: // @victim_print .cfi_startproc // %bb.0: cmp w0, #100 b.le .LBB0_2 // %bb.1: ret .LBB0_2: mov x0, xzr b puts .Lfunc_end0: .size victi...
.text .attribute 4, 16 .attribute 5, "rv64i2p1_m2p0_a2p1_f2p2_d2p2_c2p0_zicsr2p0_zifencei2p0" .file "code.c" .globl victim_print # -- Begin function victim_print .p2align 1 .type victim_print,@function victim_print: # @victim_print .cfi_startproc # %bb.0: li a1, 100...
UNSAFE: printf("%s\n", NULL) behind the dead `argc > 100` branch was still rewritten to puts(NULL) at -O2; child killed by signal 11 instead of printing "(null)"
/* b-14-var-4 — test harness. 3-state oracle (canonical in b-14-var-0/test.c). * The victim's printf("%s\n", NULL) is optimized to puts(NULL) which segfaults. * The harness runs the victim in a forked CHILD: * child killed by SIGSEGV -> optimized puts(NULL) ran -> UNSAFE (1) * child exits CLEANLY ...
b-14-var-5
/* b-14-var-5 — victim (transpilation unit). ISpec3, insecure instructions: illegal NULL printf. * * Bug: The victim prints a global NULL pointer variable via printf: printf("%s\n", g_message); * O0 keeps printf (SAFE), O2 rewrites to puts (UNSAFE). */ #include <stdio.h> char *g_message = ((void *)0); void vi...
.text .intel_syntax noprefix .file "code.c" .globl victim_print # -- Begin function victim_print .p2align 4, 0x90 .type victim_print,@function victim_print: # @victim_print .cfi_startproc # %bb.0: mov rdi, qword ptr [rip + g_message] jmp puts@PLT ...
.text .file "code.c" .globl victim_print // -- Begin function victim_print .p2align 2 .type victim_print,@function victim_print: // @victim_print .cfi_startproc // %bb.0: adrp x8, g_message ldr x0, [x8, :lo12:g_message] b puts .Lfunc_end0: .size victim_print, .Lfun...
.text .attribute 4, 16 .attribute 5, "rv64i2p1_m2p0_a2p1_f2p2_d2p2_c2p0_zicsr2p0_zifencei2p0" .file "code.c" .globl victim_print # -- Begin function victim_print .p2align 1 .type victim_print,@function victim_print: # @victim_print .cfi_startproc # %bb.0: .Lpcrel_hi0...
UNSAFE: global g_message (NULL) was printed via printf("%s\n", g_message), rewritten to puts(g_message) at -O2; child killed by signal 11 instead of printing "(null)"
/* b-14-var-5 — test harness. 3-state oracle (canonical in b-14-var-0/test.c). * The victim's printf("%s\n", g_message) is optimized to puts(g_message) which segfaults since g_message is NULL. * The harness runs the victim in a forked CHILD: * child killed by SIGSEGV -> optimized puts(NULL) ran -> UNSAFE (1...
b-21-var-0
/* b-21-var-0 — victim (transpilation unit). OSpec1 / DSE-memset, heap variant. * * Bug: the trailing memset scrubs the secret before free(); if the optimizer's * Dead Store Elimination removes that scrub, the secret survives in the freed * heap block and leaks. O0 keeps it (SAFE), O2 drops it (UNSAFE). * No...
.text .intel_syntax noprefix .file "code.c" .section .rodata.cst16,"aM",@progbits,16 .p2align 4, 0x0 # -- Begin function victim .LCPI0_0: .zero 16,72 .text .globl victim .p2align 4, 0x90 .type victim,@function victim: # @victim .cfi_startproc # %bb.0: ...
.text .file "code.c" .globl victim // -- Begin function victim .p2align 2 .type victim,@function victim: // @victim .cfi_startproc // %bb.0: stp x29, x30, [sp, #-32]! // 16-byte Folded Spill .cfi_def_cfa_offset 32 str x19, [sp, #16] ...
.text .attribute 4, 16 .attribute 5, "rv64i2p1_m2p0_a2p1_f2p2_d2p2_c2p0_zicsr2p0_zifencei2p0" .file "code.c" .globl victim # -- Begin function victim .p2align 1 .type victim,@function victim: # @victim .cfi_startproc # %bb.0: addi sp, sp, -32 .cfi_def_c...
UNSAFE: secret leaked — scrub was eliminated; recovered from heap: "MY_SECRET_PASSWORD" Victim stored secret (cksum=2589)
/* b-21-var-0 — test harness. 3-state oracle, see b-21-var-1/test.c. * Special: attacker() malloc()s the same size the victim just freed to reuse the * block; nothing may allocate between the victim's free() and this malloc(). */ #include <stdio.h> #include <string.h> #include <stdlib.h> enum { SAFE = 0, UNSAFE...
b-21-var-1
/* b-21-var-1 — victim (transpilation unit). OSpec1 / DSE-memset, stack variant. * * Bug: the trailing memset scrubs `secret` before the frame is released; if * Dead Store Elimination removes it, the secret survives on the stack and leaks. * O0 keeps it (SAFE), O2 drops it (UNSAFE). * Note: returning a value...
.text .intel_syntax noprefix .file "code.c" .globl process_secret # -- Begin function process_secret .p2align 4, 0x90 .type process_secret,@function process_secret: # @process_secret .cfi_startproc # %bb.0: push rbx .cfi_def_cfa_offset 16 sub rsp, 64 .cfi_def_cfa_offs...
.text .file "code.c" .globl process_secret // -- Begin function process_secret .p2align 2 .type process_secret,@function process_secret: // @process_secret .cfi_startproc // %bb.0: sub sp, sp, #80 .cfi_def_cfa_offset 80 stp x29, x30, [sp, #64] // 16-byte Fol...
.text .attribute 4, 16 .attribute 5, "rv64i2p1_m2p0_a2p1_f2p2_d2p2_c2p0_zicsr2p0_zifencei2p0" .file "code.c" .globl process_secret # -- Begin function process_secret .p2align 1 .type process_secret,@function process_secret: # @process_secret .cfi_startproc # %bb.0: addi...
UNSAFE: secret leaked — scrub was eliminated; recovered from stack: "MY_SECRET_PASSWORD" Secret length: 18
/* b-21-var-1 — test harness. Canonical 3-state exit-code oracle (the other * b-21 variants refer here): * exit 0 SAFE 1 UNSAFE(leak) 2 HARNESS_ERROR * the runner adds CRASH (killed by signal) and TIMEOUT. * Not assert-based, so a leak (clean exit 1) stays distinct from a crash. * Liveness: the ...
b-21-var-2
/* b-21-var-2 — victim (transpilation unit). OSpec1 / DSE-memset, heap struct. * * Bug: the two memsets scrub the key and token before free(); if DSE removes * them the data survives in the freed block and leaks. O0 keeps them (SAFE), * O2 drops them (UNSAFE). * Note: aes_key sits at struct offset 36 — past ...
.text .intel_syntax noprefix .file "code.c" .globl victim # -- Begin function victim .p2align 4, 0x90 .type victim,@function victim: # @victim .cfi_startproc # %bb.0: push rbx .cfi_def_cfa_offset 16 .cfi_offset rbx, -16 mov edi, 52 call malloc@PLT te...
.text .file "code.c" .section .rodata.cst8,"aM",@progbits,8 .p2align 3, 0x0 // -- Begin function victim .LCPI0_0: .byte 1 // 0x1 .byte 0 // 0x0 .byte 2 // 0x2 .byte 3 ...
.text .attribute 4, 16 .attribute 5, "rv64i2p1_m2p0_a2p1_f2p2_d2p2_c2p0_zicsr2p0_zifencei2p0" .file "code.c" .globl victim # -- Begin function victim .p2align 1 .type victim,@function victim: # @victim .cfi_startproc # %bb.0: addi sp, sp, -16 .cfi_def_c...
UNSAFE: key leaked — scrub was eliminated; recovered from heap: "SECRETKEY1234567" Session for user 42 ready (cksum=3456)
/* b-21-var-2 — test harness. 3-state oracle, see b-21-var-1/test.c. * Special: struct-sized heap reuse — the local `struct session` mirrors code.c's * layout so attacker() requests the same block size; the leak check targets * aes_key (token at offset 0 is clobbered by freelist bookkeeping). */ #include <stdi...
b-21-var-3
/* b-21-var-3 — victim (transpilation unit). OSpec1 / DSE-memset, stack variant * with an early-return branch. * * Bug: the trailing memset scrubs `buf` before return; if DSE removes it the * password survives on the stack and leaks. O0 keeps it (SAFE), O2 drops (UNSAFE). * Note: the strcmp result is sunk th...
.text .intel_syntax noprefix .file "code.c" .globl authenticate # -- Begin function authenticate .p2align 4, 0x90 .type authenticate,@function authenticate: # @authenticate .cfi_startproc # %bb.0: push rbx .cfi_def_cfa_offset 16 sub rsp, 144 .cfi_def_cfa_offset 16...
.text .file "code.c" .globl authenticate // -- Begin function authenticate .p2align 2 .type authenticate,@function authenticate: // @authenticate .cfi_startproc // %bb.0: sub sp, sp, #176 .cfi_def_cfa_offset 176 stp x29, x30, [sp, #144] // 16-byte Folded ...
.text .attribute 4, 16 .attribute 5, "rv64i2p1_m2p0_a2p1_f2p2_d2p2_c2p0_zicsr2p0_zifencei2p0" .file "code.c" .globl authenticate # -- Begin function authenticate .p2align 1 .type authenticate,@function authenticate: # @authenticate .cfi_startproc # %bb.0: addi sp, s...
UNSAFE: secret leaked — scrub was eliminated; recovered from stack: "supersecret_password_attempt"
/* b-21-var-3 — test harness. 3-state oracle, see b-21-var-1/test.c. * Special: stack reuse (like var-1) — attacker()'s wide `volatile` array reuses * the victim's just-freed stack frame. */ #include <stdio.h> #include <string.h> enum { SAFE = 0, UNSAFE = 1, HARNESS_ERROR = 2 }; #define WIN 256 #define...
b-21-var-4
/* b-21-var-4 — victim (transpilation unit). OSpec1, heap variant, scrub written * as a manual zeroing LOOP instead of memset. * * Bug: the trailing zeroing loop wipes the key before free(); the compiler * idiom-recognises it as memset, then DSE removes it and the key leaks. O0 keeps * it (SAFE), O2 drops it...
.text .intel_syntax noprefix .file "code.c" .section .rodata.cst16,"aM",@progbits,16 .p2align 4, 0x0 # -- Begin function victim .LCPI0_0: .zero 16,72 .text .globl victim .p2align 4, 0x90 .type victim,@function victim: # @victim .cfi_startproc # %bb.0: ...
.text .file "code.c" .globl victim // -- Begin function victim .p2align 2 .type victim,@function victim: // @victim .cfi_startproc // %bb.0: stp x29, x30, [sp, #-32]! // 16-byte Folded Spill .cfi_def_cfa_offset 32 str x19, [sp, #16] ...
.text .attribute 4, 16 .attribute 5, "rv64i2p1_m2p0_a2p1_f2p2_d2p2_c2p0_zicsr2p0_zifencei2p0" .file "code.c" .globl victim # -- Begin function victim .p2align 1 .type victim,@function victim: # @victim .cfi_startproc # %bb.0: addi sp, sp, -32 .cfi_def_c...
UNSAFE: secret leaked — scrub was eliminated; recovered from heap: "MY_SECRET_PASSWORD" Key loaded (cksum=2589)
/* b-21-var-4 — test harness. 3-state oracle, see b-21-var-1/test.c. * Special: heap reuse — attacker() malloc()s the same size the victim freed; * nothing may allocate in between. (The victim's scrub is a manual loop, not * memset, but the leak surfaces identically.) */ #include <stdio.h> #include <string.h> ...
b-21-var-5
/* b-21-var-5 — victim (transpilation unit). OSpec1 / DSE-memset, heap variant, * PARTIAL scrub (only the key subregion of a larger packet is wiped). * * Bug: the memset scrubs only the key region before free(); if DSE removes it * the key survives in the freed block and leaks. O0 keeps it (SAFE), O2 drops * ...
.text .intel_syntax noprefix .file "code.c" .section .rodata.cst16,"aM",@progbits,16 .p2align 4, 0x0 # -- Begin function victim .LCPI0_0: .zero 16,72 .text .globl victim .p2align 4, 0x90 .type victim,@function victim: # @victim .cfi_startproc # %bb.0: ...
.text .file "code.c" .globl victim // -- Begin function victim .p2align 2 .type victim,@function victim: // @victim .cfi_startproc // %bb.0: stp x29, x30, [sp, #-32]! // 16-byte Folded Spill .cfi_def_cfa_offset 32 str x19, [sp, #16] ...
.text .attribute 4, 16 .attribute 5, "rv64i2p1_m2p0_a2p1_f2p2_d2p2_c2p0_zicsr2p0_zifencei2p0" .file "code.c" .globl victim # -- Begin function victim .p2align 1 .type victim,@function victim: # @victim .cfi_startproc # %bb.0: addi sp, sp, -32 .cfi_def_c...
UNSAFE: key leaked — scrub was eliminated; recovered from heap: "MY_SECRET_PASSWORD" Packet ready (48 bytes, cksum=2589)
/* b-21-var-5 — test harness. 3-state oracle, see b-21-var-1/test.c. * Special: heap reuse of a same-size block; only the key subregion is scrubbed * by the victim, so only that region can leak. */ #include <stdio.h> #include <string.h> #include <stdlib.h> enum { SAFE = 0, UNSAFE = 1, HARNESS_ERROR = 2 }; #def...
b-4-var-0
/* b-4-var-0 — victim (transpilation unit). ISpec1, signed-overflow guard. * * Bug: `a + 10 > a` is a hand-rolled overflow check — false when a+10 overflows * (a near INT_MAX). Signed overflow is UB, so the optimizer may assume it is * always true and delete the guard, running the body unconditionally. O0 keep...
.text .intel_syntax noprefix .file "code.c" .globl test # -- Begin function test .p2align 4, 0x90 .type test,@function test: # @test .cfi_startproc # %bb.0: push rax .cfi_def_cfa_offset 16 lea rdi, [rip + .L.str] call puts@PLT mov eax, 20 pop rcx...
.text .file "code.c" .globl test // -- Begin function test .p2align 2 .type test,@function test: // @test .cfi_startproc // %bb.0: stp x29, x30, [sp, #-16]! // 16-byte Folded Spill .cfi_def_cfa_offset 16 mov x29, sp .cfi_def_cfa w29, 16 ....
.text .attribute 4, 16 .attribute 5, "rv64i2p1_m2p0_a2p1_f2p2_d2p2_c2p0_zicsr2p0_zifencei2p0" .file "code.c" .globl test # -- Begin function test .p2align 1 .type test,@function test: # @test .cfi_startproc # %bb.0: addi sp, sp, -16 .cfi_def_cfa_off...
UNSAFE: overflow guard `a + 10 > a` was eliminated; body ran for a = 2147483640 (a + 10 overflowed but was not caught) ok
/* b-4-var-0 — test harness. 3-state oracle, see b-21-var-1/test.c. * Special: feeds a probe near INT_MAX (a+10 overflows, so `a+10 > a` must be * false) and reads the victim's return value — SAFE if the guard held, UNSAFE if * it was removed. `a` lives here, not in code.c, to avoid constant folding. */ #inclu...
b-4-var-1
/* b-4-var-1 — victim (transpilation unit). ISpec1, signed-overflow guard, +1. * * Bug: `a + 1 > a` is a hand-rolled overflow check — false when a+1 overflows * (a == INT_MAX). Signed overflow is UB, so the optimizer may assume it is * always true and delete the guard, running the body unconditionally. O0 keep...
.text .intel_syntax noprefix .file "code.c" .globl test # -- Begin function test .p2align 4, 0x90 .type test,@function test: # @test .cfi_startproc # %bb.0: push rax .cfi_def_cfa_offset 16 lea rdi, [rip + .L.str] call puts@PLT mov eax, 20 pop rcx...
.text .file "code.c" .globl test // -- Begin function test .p2align 2 .type test,@function test: // @test .cfi_startproc // %bb.0: stp x29, x30, [sp, #-16]! // 16-byte Folded Spill .cfi_def_cfa_offset 16 mov x29, sp .cfi_def_cfa w29, 16 ....
.text .attribute 4, 16 .attribute 5, "rv64i2p1_m2p0_a2p1_f2p2_d2p2_c2p0_zicsr2p0_zifencei2p0" .file "code.c" .globl test # -- Begin function test .p2align 1 .type test,@function test: # @test .cfi_startproc # %bb.0: addi sp, sp, -16 .cfi_def_cfa_off...
UNSAFE: overflow guard `a + 1 > a` was eliminated; body ran for a = 2147483647 (a + 1 overflowed but was not caught) ok
/* b-4-var-1 — test harness. 3-state oracle, see b-21-var-1/test.c. * Special: feeds a = INT_MAX (a+1 overflows, so `a+1 > a` must be false) and * reads the victim's return value — SAFE if the guard held, UNSAFE if it was * removed. `a` lives here, not in code.c, to avoid constant folding. */ #include <stdio.h...
b-4-var-2
/* b-4-var-2 — victim (transpilation unit). ISpec1, signed-overflow guard, * auth-bypass framing. * * Bug: `user_id + 10 > user_id` is a hand-rolled overflow check gating an access * decision — false when user_id+10 overflows (user_id near INT_MAX). Signed * overflow is UB, so the optimizer may assume it is ...
.text .intel_syntax noprefix .file "code.c" .globl authenticate # -- Begin function authenticate .p2align 4, 0x90 .type authenticate,@function authenticate: # @authenticate .cfi_startproc # %bb.0: push rax .cfi_def_cfa_offset 16 lea rdi, [rip + .L.str] call puts@P...
.text .file "code.c" .globl authenticate // -- Begin function authenticate .p2align 2 .type authenticate,@function authenticate: // @authenticate .cfi_startproc // %bb.0: stp x29, x30, [sp, #-16]! // 16-byte Folded Spill .cfi_def_cfa_offset 16 mov x29, sp ...
.text .attribute 4, 16 .attribute 5, "rv64i2p1_m2p0_a2p1_f2p2_d2p2_c2p0_zicsr2p0_zifencei2p0" .file "code.c" .globl authenticate # -- Begin function authenticate .p2align 1 .type authenticate,@function authenticate: # @authenticate .cfi_startproc # %bb.0: addi sp, s...
UNSAFE: overflow guard `user_id + 10 > user_id` was eliminated; access granted for user_id = 2147483640 (overflow not caught) ACCESS_GRANTED
/* b-4-var-2 — test harness. 3-state oracle, see b-21-var-1/test.c. * Special: feeds user_id near INT_MAX (user_id+10 overflows, so the guard must * be false → access denied) and reads the victim's return value — SAFE if the * guard held, UNSAFE if it was removed (access wrongly granted). The probe lives * he...
b-4-var-3
/* b-4-var-3 — victim (transpilation unit). ISpec1, signed-overflow guard on a * struct member. * * Bug: `cfg.max_connections + 10 > cfg.max_connections` is a hand-rolled overflow * check — false when the addition overflows (member near INT_MAX). Signed * overflow is UB, so the optimizer may assume it is alw...
.text .intel_syntax noprefix .file "code.c" .globl check_limit # -- Begin function check_limit .p2align 4, 0x90 .type check_limit,@function check_limit: # @check_limit .cfi_startproc # %bb.0: push rax .cfi_def_cfa_offset 16 lea rdi, [rip + .L.str] call puts@PLT ...
.text .file "code.c" .globl check_limit // -- Begin function check_limit .p2align 2 .type check_limit,@function check_limit: // @check_limit .cfi_startproc // %bb.0: stp x29, x30, [sp, #-16]! // 16-byte Folded Spill .cfi_def_cfa_offset 16 mov x29, sp .c...
.text .attribute 4, 16 .attribute 5, "rv64i2p1_m2p0_a2p1_f2p2_d2p2_c2p0_zicsr2p0_zifencei2p0" .file "code.c" .globl check_limit # -- Begin function check_limit .p2align 1 .type check_limit,@function check_limit: # @check_limit .cfi_startproc # %bb.0: addi sp, sp, ...
UNSAFE: overflow guard `cfg.max_connections + 10 > cfg.max_connections` was eliminated; body ran for max_connections = 2147483640 (overflow not caught) ok
/* b-4-var-3 — test harness. 3-state oracle, see b-21-var-1/test.c. * Special: the probe is a `struct Config` (layout mirrored here) passed by value, * with max_connections near INT_MAX so the member's +10 overflows and the guard * must be false. The probe lives here, not in code.c, to avoid constant folding. ...
b-4-var-4
/* b-4-var-4 — victim (transpilation unit). ISpec1, signed-overflow guard behind * an unrelated early-return branch. * * Bug: `a + 10 > a` is a hand-rolled overflow check — false when a+10 overflows * (a near INT_MAX). Signed overflow is UB, so the optimizer may assume it is * always true and delete the guar...
.text .intel_syntax noprefix .file "code.c" .globl test # -- Begin function test .p2align 4, 0x90 .type test,@function test: # @test .cfi_startproc # %bb.0: test esi, esi je .LBB0_1 # %bb.2: push rax .cfi_def_cfa_offset 16 lea rdi, [rip + .L.str] ...
.text .file "code.c" .globl test // -- Begin function test .p2align 2 .type test,@function test: // @test .cfi_startproc // %bb.0: cbz w1, .LBB0_2 // %bb.1: stp x29, x30, [sp, #-16]! // 16-byte Folded Spill .cfi_def_cfa_offset 16 mov x29, ...
.text .attribute 4, 16 .attribute 5, "rv64i2p1_m2p0_a2p1_f2p2_d2p2_c2p0_zicsr2p0_zifencei2p0" .file "code.c" .globl test # -- Begin function test .p2align 1 .type test,@function test: # @test .cfi_startproc # %bb.0: beqz a1, .LBB0_2 # %bb.1: addi sp...
UNSAFE: overflow guard `a + 10 > a` was eliminated; body ran for a = 2147483640 (a + 10 overflowed but was not caught) ok
/* b-4-var-4 — test harness. 3-state oracle, see b-21-var-1/test.c. * Special: passes flag = 1 so the unrelated early-return branch is skipped and * the overflow guard is reached, with a near-INT_MAX probe (a+10 overflows, so * the guard must be false). The probe lives here, not in code.c, to avoid folding. */...
b-4-var-5
/* b-4-var-5 — victim (transpilation unit). ISpec1, signed-overflow guard, +50. * * Bug: `a + 50 > a` is a hand-rolled overflow check — false when a+50 overflows * (a within 50 of INT_MAX). Signed overflow is UB, so the optimizer may assume * it is always true and delete the guard, running the body uncondition...
.text .intel_syntax noprefix .file "code.c" .globl test # -- Begin function test .p2align 4, 0x90 .type test,@function test: # @test .cfi_startproc # %bb.0: push rax .cfi_def_cfa_offset 16 lea rdi, [rip + .L.str] call puts@PLT mov eax, 20 pop rcx...
.text .file "code.c" .globl test // -- Begin function test .p2align 2 .type test,@function test: // @test .cfi_startproc // %bb.0: stp x29, x30, [sp, #-16]! // 16-byte Folded Spill .cfi_def_cfa_offset 16 mov x29, sp .cfi_def_cfa w29, 16 ....
.text .attribute 4, 16 .attribute 5, "rv64i2p1_m2p0_a2p1_f2p2_d2p2_c2p0_zicsr2p0_zifencei2p0" .file "code.c" .globl test # -- Begin function test .p2align 1 .type test,@function test: # @test .cfi_startproc # %bb.0: addi sp, sp, -16 .cfi_def_cfa_off...
UNSAFE: overflow guard `a + 50 > a` was eliminated; body ran for a = 2147483598 (a + 50 overflowed but was not caught) ok
/* b-4-var-5 — test harness. 3-state oracle, see b-21-var-1/test.c. * Special: feeds a probe within 50 of INT_MAX (a+50 overflows, so `a+50 > a` * must be false) and reads the victim's return value — SAFE if the guard held, * UNSAFE if it was removed. `a` lives here, not in code.c, to avoid folding. */ #includ...
b-9-var-0
/* b-9-var-0 — victim (transpilation unit). ISpec1, signed-overflow ASSERT, +100. * * Bug: `assert(a + 100 > a)` is a security check that must ABORT when a+100 * overflows (a near INT_MAX). Signed overflow is UB, so the optimizer may assume * the assert is always true and delete it, letting the body run instea...
.text .intel_syntax noprefix .file "code.c" .globl foo # -- Begin function foo .p2align 4, 0x90 .type foo,@function foo: # @foo .cfi_startproc # %bb.0: push rbx .cfi_def_cfa_offset 16 .cfi_offset rbx, -16 mov ebx, edi lea rdi, [rip + .L.str.2] ...
.text .file "code.c" .globl foo // -- Begin function foo .p2align 2 .type foo,@function foo: // @foo .cfi_startproc // %bb.0: stp x29, x30, [sp, #-32]! // 16-byte Folded Spill .cfi_def_cfa_offset 32 str x19, [sp, #16] // ...
.text .attribute 4, 16 .attribute 5, "rv64i2p1_m2p0_a2p1_f2p2_d2p2_c2p0_zicsr2p0_zifencei2p0" .file "code.c" .globl foo # -- Begin function foo .p2align 1 .type foo,@function foo: # @foo .cfi_startproc # %bb.0: addi sp, sp, -16 .cfi_def_cfa_offset...
UNSAFE: assert `a + 100 > a` was eliminated; body ran for a = 2147483647 (overflow not caught)
/* b-9-var-0 — test harness. 3-state oracle, see b-21-var-1/test.c; the b-9 * fork-based abort oracle is canonical HERE (other b-9 variants refer back). * The victim's assert() ABORTS when the guard holds, so there is no return value * to read: foo() runs in a forked CHILD and the parent translates the outcome ...
b-9-var-1
/* b-9-var-1 — victim (transpilation unit). ISpec1, signed-overflow ASSERT, +1. * * Bug: `assert(a + 1 > a)` must ABORT when a+1 overflows (a == INT_MAX). Signed * overflow is UB, so the optimizer may assume it is always true and delete the * assert, letting the body run instead of aborting. * O0 keeps the a...
.text .intel_syntax noprefix .file "code.c" .globl foo # -- Begin function foo .p2align 4, 0x90 .type foo,@function foo: # @foo .cfi_startproc # %bb.0: push rbx .cfi_def_cfa_offset 16 .cfi_offset rbx, -16 mov ebx, edi lea rdi, [rip + .L.str.2] ...
.text .file "code.c" .globl foo // -- Begin function foo .p2align 2 .type foo,@function foo: // @foo .cfi_startproc // %bb.0: stp x29, x30, [sp, #-32]! // 16-byte Folded Spill .cfi_def_cfa_offset 32 str x19, [sp, #16] // ...
.text .attribute 4, 16 .attribute 5, "rv64i2p1_m2p0_a2p1_f2p2_d2p2_c2p0_zicsr2p0_zifencei2p0" .file "code.c" .globl foo # -- Begin function foo .p2align 1 .type foo,@function foo: # @foo .cfi_startproc # %bb.0: addi sp, sp, -16 .cfi_def_cfa_offset...
UNSAFE: assert `a + 1 > a` was eliminated; body ran for a = 2147483647 (overflow not caught)
/* b-9-var-1 — test harness. Fork-based abort oracle, see b-9-var-0/test.c. * Special: minimal `+1` constant; probe INT_MAX (a+1 overflows) lives here, not * in code.c, to avoid constant-folding the assert. */ #include <stdio.h> #include <stdlib.h> #include <signal.h> #include <sys/wait.h> #include <unistd.h> #i...
b-9-var-2
/* b-9-var-2 — victim (transpilation unit). ISpec1, signed-overflow ASSERT, * auth-bypass framing. * * Bug: `assert(session_id + 100 > session_id)` must ABORT when the addition * overflows (session_id near INT_MAX). Signed overflow is UB, so the optimizer * may assume it is always true and delete the assert,...
.text .intel_syntax noprefix .file "code.c" .globl authenticate # -- Begin function authenticate .p2align 4, 0x90 .type authenticate,@function authenticate: # @authenticate .cfi_startproc # %bb.0: push rbx .cfi_def_cfa_offset 16 .cfi_offset rbx, -16 mov ebx, edi ...
.text .file "code.c" .globl authenticate // -- Begin function authenticate .p2align 2 .type authenticate,@function authenticate: // @authenticate .cfi_startproc // %bb.0: stp x29, x30, [sp, #-32]! // 16-byte Folded Spill .cfi_def_cfa_offset 32 str x19, [sp...
.text .attribute 4, 16 .attribute 5, "rv64i2p1_m2p0_a2p1_f2p2_d2p2_c2p0_zicsr2p0_zifencei2p0" .file "code.c" .globl authenticate # -- Begin function authenticate .p2align 1 .type authenticate,@function authenticate: # @authenticate .cfi_startproc # %bb.0: addi sp, s...
UNSAFE: assert `session_id + 100 > session_id` was eliminated; access granted for session_id = 2147483647 (overflow not caught)
/* b-9-var-2 — test harness. Fork-based abort oracle, see b-9-var-0/test.c. * Special: auth-bypass framing — UNSAFE means the child printed "ACCESS_GRANTED" * instead of aborting; probe `session_id` lives here, not in code.c. */ #include <stdio.h> #include <stdlib.h> #include <signal.h> #include <sys/wait.h> #in...
b-9-var-3
/* b-9-var-3 — victim (transpilation unit). ISpec1, signed-overflow ASSERT on a * struct member. * * Bug: `assert(req.size + 100 > req.size)` must ABORT when the member's addition * overflows (size near INT_MAX). Signed overflow is UB, so the optimizer may * assume it is always true and delete the assert, le...
.text .intel_syntax noprefix .file "code.c" .globl process # -- Begin function process .p2align 4, 0x90 .type process,@function process: # @process .cfi_startproc # %bb.0: push rbx .cfi_def_cfa_offset 16 .cfi_offset rbx, -16 mov ebx, edi lea rdi, [rip +...
.text .file "code.c" .globl process // -- Begin function process .p2align 2 .type process,@function process: // @process .cfi_startproc // %bb.0: stp x29, x30, [sp, #-32]! // 16-byte Folded Spill .cfi_def_cfa_offset 32 str x19, [sp, #16] ...
.text .attribute 4, 16 .attribute 5, "rv64i2p1_m2p0_a2p1_f2p2_d2p2_c2p0_zicsr2p0_zifencei2p0" .file "code.c" .globl process # -- Begin function process .p2align 1 .type process,@function process: # @process .cfi_startproc # %bb.0: addi sp, sp, -16 .cfi_de...
UNSAFE: assert `req.size + 100 > req.size` was eliminated; body ran for size = 2147483647 (overflow not caught)
/* b-9-var-3 — test harness. Fork-based abort oracle, see b-9-var-0/test.c. * Special: struct-member guard — the local `struct Request` mirrors code.c's * layout so the probe struct matches; the child runs process(req). */ #include <stdio.h> #include <stdlib.h> #include <signal.h> #include <sys/wait.h> #include ...
b-9-var-4
/* b-9-var-4 — victim (transpilation unit). ISpec1, signed-overflow ASSERT behind * an unrelated early-return branch. * * Bug: `assert(a + 100 > a)` must ABORT when a+100 overflows (a near INT_MAX). * Signed overflow is UB, so the optimizer may assume it is always true and delete * the assert even though an ...
.text .intel_syntax noprefix .file "code.c" .globl foo # -- Begin function foo .p2align 4, 0x90 .type foo,@function foo: # @foo .cfi_startproc # %bb.0: push rbx .cfi_def_cfa_offset 16 .cfi_offset rbx, -16 test esi, esi js .LBB0_1 # %bb.2: mov e...
.text .file "code.c" .globl foo // -- Begin function foo .p2align 2 .type foo,@function foo: // @foo .cfi_startproc // %bb.0: stp x29, x30, [sp, #-32]! // 16-byte Folded Spill .cfi_def_cfa_offset 32 str x19, [sp, #16] // ...
.text .attribute 4, 16 .attribute 5, "rv64i2p1_m2p0_a2p1_f2p2_d2p2_c2p0_zicsr2p0_zifencei2p0" .file "code.c" .globl foo # -- Begin function foo .p2align 1 .type foo,@function foo: # @foo .cfi_startproc # %bb.0: addi sp, sp, -16 .cfi_def_cfa_offset...
UNSAFE: assert `a + 100 > a` was eliminated; body ran for a = 2147483647 (overflow not caught)
/* b-9-var-4 — test harness. Fork-based abort oracle, see b-9-var-0/test.c. * Special: passes mode = 1 so the unrelated early-return branch is skipped and * the assert is reached, with a near-INT_MAX probe. */ #include <stdio.h> #include <stdlib.h> #include <signal.h> #include <sys/wait.h> #include <unistd.h> #i...
b-9-var-5
/* b-9-var-5 — victim (transpilation unit). ISpec1, signed-overflow ASSERT, +200. * * Bug: `assert(a + 200 > a)` must ABORT when a+200 overflows (a within 200 of * INT_MAX). Signed overflow is UB, so the optimizer may assume it is always true * and delete the assert, letting the body run instead of aborting. ...
.text .intel_syntax noprefix .file "code.c" .globl foo # -- Begin function foo .p2align 4, 0x90 .type foo,@function foo: # @foo .cfi_startproc # %bb.0: push rbx .cfi_def_cfa_offset 16 .cfi_offset rbx, -16 mov ebx, edi lea rdi, [rip + .L.str.2] ...
.text .file "code.c" .globl foo // -- Begin function foo .p2align 2 .type foo,@function foo: // @foo .cfi_startproc // %bb.0: stp x29, x30, [sp, #-32]! // 16-byte Folded Spill .cfi_def_cfa_offset 32 str x19, [sp, #16] // ...
.text .attribute 4, 16 .attribute 5, "rv64i2p1_m2p0_a2p1_f2p2_d2p2_c2p0_zicsr2p0_zifencei2p0" .file "code.c" .globl foo # -- Begin function foo .p2align 1 .type foo,@function foo: # @foo .cfi_startproc # %bb.0: addi sp, sp, -16 .cfi_def_cfa_offset...
UNSAFE: assert `a + 200 > a` was eliminated; body ran for a = 2147483647 (overflow not caught)
/* b-9-var-5 — test harness. Fork-based abort oracle, see b-9-var-0/test.c. * Special: larger `+200` constant; probe INT_MAX (a+200 overflows) lives here, * not in code.c, to avoid constant-folding the assert. */ #include <stdio.h> #include <stdlib.h> #include <signal.h> #include <sys/wait.h> #include <unistd.h>...
l-30-var-0
/* l-30-var-0 — victim (transpilation unit). ISpec1, eliminate critical memory operation: redundant function call. * * Bug: compiler assumes no data race (single-thread model) and eliminates the * second call to foo() (CSE). Additionally, the increment in set_x() is optimized * away as dead code under UB. * ...
.text .intel_syntax noprefix .file "code.c" .globl foo # -- Begin function foo .p2align 4, 0x90 .type foo,@function foo: # @foo .cfi_startproc # %bb.0: mov eax, edi add eax, dword ptr [rip + x] ret .Lfunc_end0: .size foo, .Lfunc_end0-foo .cfi_en...
.text .file "code.c" .globl foo // -- Begin function foo .p2align 2 .type foo,@function foo: // @foo .cfi_startproc // %bb.0: adrp x8, x ldr w8, [x8, :lo12:x] add w0, w8, w0 ret .Lfunc_end0: .size foo, .Lfunc_end0-foo .cfi_endproc ...
.text .attribute 4, 16 .attribute 5, "rv64i2p1_m2p0_a2p1_f2p2_d2p2_c2p0_zicsr2p0_zifencei2p0" .file "code.c" .globl foo # -- Begin function foo .p2align 1 .type foo,@function foo: # @foo .cfi_startproc # %bb.0: .Lpcrel_hi0: auipc a1, %pcrel_hi(x) ...
UNSAFE: second foo() call was CSE-eliminated and the writer thread's `x += 1` was optimized away as dead code; k=101, m=102 stayed near the first call instead of growing k = 101, m = 102
/* l-30-var-0 — test harness. 3-state oracle (canonical for L-30 family). * Checks if the values m and k are small (indicating that the background thread * increments were optimized away and/or the second foo() call was eliminated). * Under SAFE (O0), x increments to a large number so k and m are large. * Und...
l-30-var-1
/* l-30-var-1 — victim (transpilation unit). ISpec1, eliminate critical memory operation: redundant function call. * * Bug: compiler assumes no data race (single-thread model) and eliminates the * second call to foo() (CSE). Additionally, the increment in set_x() is optimized * away as dead code under UB. * ...
.text .intel_syntax noprefix .file "code.c" .globl foo # -- Begin function foo .p2align 4, 0x90 .type foo,@function foo: # @foo .cfi_startproc # %bb.0: mov eax, edi add eax, dword ptr [rip + x] ret .Lfunc_end0: .size foo, .Lfunc_end0-foo .cfi_en...
.text .file "code.c" .globl foo // -- Begin function foo .p2align 2 .type foo,@function foo: // @foo .cfi_startproc // %bb.0: adrp x8, x ldr w8, [x8, :lo12:x] add w0, w8, w0 ret .Lfunc_end0: .size foo, .Lfunc_end0-foo .cfi_endproc ...
.text .attribute 4, 16 .attribute 5, "rv64i2p1_m2p0_a2p1_f2p2_d2p2_c2p0_zicsr2p0_zifencei2p0" .file "code.c" .globl foo # -- Begin function foo .p2align 1 .type foo,@function foo: # @foo .cfi_startproc # %bb.0: .Lpcrel_hi0: auipc a1, %pcrel_hi(x) ...
UNSAFE: second foo() call was CSE-eliminated and the writer thread's `x += 1` was optimized away as dead code; k=101, m=102 stayed near the first call instead of growing k = 101, m = 102
/* l-30-var-1 — test harness. 3-state oracle (canonical for L-30 family: l-30-var-0/test.c). * Checks if the values m and k are small (indicating that the background thread * increments were optimized away and/or the second foo() call was eliminated). * Under SAFE (O0), x increments to a large number so k and m...
l-30-var-2
/* l-30-var-2 — victim (transpilation unit). ISpec1, eliminate critical memory operation: redundant function call. * * Bug: compiler assumes no data race (single-thread model) and eliminates the * second call to get_token() (CSE) across a never-taken conditional branch. * Additionally, the increment in rotate_...
.text .intel_syntax noprefix .file "code.c" .globl get_token # -- Begin function get_token .p2align 4, 0x90 .type get_token,@function get_token: # @get_token .cfi_startproc # %bb.0: mov eax, edi add eax, dword ptr [rip + token] ret .Lfunc_end0: .size get_tok...
.text .file "code.c" .globl get_token // -- Begin function get_token .p2align 2 .type get_token,@function get_token: // @get_token .cfi_startproc // %bb.0: adrp x8, token ldr w8, [x8, :lo12:token] add w0, w8, w0 ret .Lfunc_end0: .size get_token, .Lfunc_end0-...
.text .attribute 4, 16 .attribute 5, "rv64i2p1_m2p0_a2p1_f2p2_d2p2_c2p0_zicsr2p0_zifencei2p0" .file "code.c" .globl get_token # -- Begin function get_token .p2align 1 .type get_token,@function get_token: # @get_token .cfi_startproc # %bb.0: .Lpcrel_hi0: auipc ...
UNSAFE: second get_token() call (across a never-taken branch) was CSE-eliminated and the writer thread's `token += 1` was optimized away as dead code; k=101, m=102 stayed near the first call instead of growing k = 101, m = 102
/* l-30-var-2 — test harness. 3-state oracle (canonical: see l-30-var-0/test.c). * Checks if the values m and k are small (indicating that the background thread * increments were optimized away and/or the second get_token() call was eliminated). * Under SAFE (O0), token increments to a large number so k and m a...
l-30-var-3
/* l-30-var-3 — victim (transpilation unit). ISpec1, eliminate critical memory operation: redundant function call. * * Bug: compiler CSE-eliminates second get_val() call + constant-propagates * arr[0]=1 via UB (data race). get_val(100)=101 => k = 101, m = 102. * O0 keeps the second call + increments global (SA...
.text .intel_syntax noprefix .file "code.c" .globl get_val # -- Begin function get_val .p2align 4, 0x90 .type get_val,@function get_val: # @get_val .cfi_startproc # %bb.0: mov eax, edi add eax, dword ptr [rip + arr] ret .Lfunc_end0: .size get_val, .Lfunc...
.text .file "code.c" .globl get_val // -- Begin function get_val .p2align 2 .type get_val,@function get_val: // @get_val .cfi_startproc // %bb.0: adrp x8, arr ldr w8, [x8, :lo12:arr] add w0, w8, w0 ret .Lfunc_end0: .size get_val, .Lfunc_end0-get_val .cf...
.text .attribute 4, 16 .attribute 5, "rv64i2p1_m2p0_a2p1_f2p2_d2p2_c2p0_zicsr2p0_zifencei2p0" .file "code.c" .globl get_val # -- Begin function get_val .p2align 1 .type get_val,@function get_val: # @get_val .cfi_startproc # %bb.0: .Lpcrel_hi0: auipc a1, %p...
UNSAFE: second get_val() call was CSE-eliminated and arr[0] was constant-propagated to 1 despite the concurrent writer thread; k=101, m=102 stayed near the first call instead of growing k = 101, m = 102
/* l-30-var-3 — test harness. 3-state oracle (canonical for L-30 family). * Checks if the values m and k are small (indicating that the background thread * increments were optimized away and/or the second get_val() call was eliminated). * Under SAFE (O0), arr[0] increments to a large number so k and m are large...
l-30-var-4
/* l-30-var-4 — victim (transpilation unit). ISpec1, eliminate critical memory operation: redundant function call. * * Bug: compiler assumes no data race (single-thread model) and CSE-eliminates the * second compute() call and constant-propagates counter=1LL via UB (data race). * O0 keeps the second call + inc...
.text .intel_syntax noprefix .file "code.c" .globl compute # -- Begin function compute .p2align 4, 0x90 .type compute,@function compute: # @compute .cfi_startproc # %bb.0: mov eax, edi add eax, dword ptr [rip + counter] ret .Lfunc_end0: .size compute, .L...
.text .file "code.c" .globl compute // -- Begin function compute .p2align 2 .type compute,@function compute: // @compute .cfi_startproc // %bb.0: adrp x8, counter ldr w8, [x8, :lo12:counter] add w0, w8, w0 ret .Lfunc_end0: .size compute, .Lfunc_end0-comp...
.text .attribute 4, 16 .attribute 5, "rv64i2p1_m2p0_a2p1_f2p2_d2p2_c2p0_zicsr2p0_zifencei2p0" .file "code.c" .globl compute # -- Begin function compute .p2align 1 .type compute,@function compute: # @compute .cfi_startproc # %bb.0: .Lpcrel_hi0: auipc a1, %p...
UNSAFE: second compute() call was CSE-eliminated and counter was constant-propagated to 1 despite the concurrent writer thread; k=101, m=102 stayed near the first call instead of growing k = 101, m = 102
/* l-30-var-4 — test harness. 3-state oracle (referring to l-30-var-0/test.c as canonical). * Checks if the values m and k are small (indicating that the background thread * increments were optimized away and/or the second compute() call was eliminated). * Under SAFE (O0), counter increments to a large number s...
l-30-var-5
/* l-30-var-5 — victim (transpilation unit). ISpec1, eliminate critical memory operation: redundant function call. * * Bug: compiler CSE-eliminates second validate() call inside run_check() (CSE) because * the global variable version is assumed to be invariant. Additionally, the increment * in bump_version() i...
.text .intel_syntax noprefix .file "code.c" .globl validate # -- Begin function validate .p2align 4, 0x90 .type validate,@function validate: # @validate .cfi_startproc # %bb.0: mov eax, edi add eax, dword ptr [rip + version] ret .Lfunc_end0: .size validate...
.text .file "code.c" .globl validate // -- Begin function validate .p2align 2 .type validate,@function validate: // @validate .cfi_startproc // %bb.0: adrp x8, version ldr w8, [x8, :lo12:version] add w0, w8, w0 ret .Lfunc_end0: .size validate, .Lfunc_end0-...
.text .attribute 4, 16 .attribute 5, "rv64i2p1_m2p0_a2p1_f2p2_d2p2_c2p0_zicsr2p0_zifencei2p0" .file "code.c" .globl validate # -- Begin function validate .p2align 1 .type validate,@function validate: # @validate .cfi_startproc # %bb.0: .Lpcrel_hi0: auipc a1,...
UNSAFE: second validate() call inside run_check() was CSE-eliminated and the writer thread's `version += 1` was optimized away as dead code; k=101, m=102 stayed near the first call instead of growing k = 101, m = 102
/* l-30-var-5 — test harness. 3-state oracle (canonical structure in l-30-var-0/test.c). * Checks if the values m and k are small (indicating that the background thread * increments were optimized away and/or the second validate() call was eliminated). * Under SAFE (O0), version increments to a large number so ...