[MemCpyOpt] Fix the invalid code modification in Memcpy pass #1

kaiyan96 · 2023-10-07T11:19:23Z

No description provided.

Relocate the GEP modification to a later stage of the function performCallSlotOption, ensuring that the code remains unchanged if the optimization fails.

This test is currently flaky on a local Windows amd64 build. The reason is that it relies on the order of `process.threads` but this order is nondeterministic: If we print lldb's inputs and outputs while running, we can see that the breakpoints are always being set correctly, and always being hit: ```sh runCmd: breakpoint set -f "main.c" -l 2 output: Breakpoint 1: where = a.out`func_inner + 1 at main.c:2:9, address = 0x0000000140001001 runCmd: breakpoint set -f "main.c" -l 7 output: Breakpoint 2: where = a.out`main + 17 at main.c:7:5, address = 0x0000000140001021 runCmd: run output: Process 52328 launched: 'C:\workspace\llvm-project\llvm\build\lldb-test-build.noindex\functionalities\unwind\zeroth_frame\TestZerothFrame.test_dwarf\a.out' (x86_64) Process 52328 stopped * thread #1, stop reason = breakpoint 1.1 frame #0: 0x00007ff68f6b1001 a.out`func_inner at main.c:2:9 1 void func_inner() { -> 2 int a = 1; // Set breakpoint 1 here ^ 3 } 4 5 int main() { 6 func_inner(); 7 return 0; // Set breakpoint 2 here ``` However, sometimes the backtrace printed in this test shows that the process is stopped inside NtWaitForWorkViaWorkerFactory from `ntdll.dll`: ```sh Backtrace at the first breakpoint: frame #0: 0x00007ffecc7b3bf4 ntdll.dll`NtWaitForWorkViaWorkerFactory + 20 frame #1: 0x00007ffecc74585e ntdll.dll`RtlClearThreadWorkOnBehalfTicket + 862 frame llvm#2: 0x00007ffecc3e257d kernel32.dll`BaseThreadInitThunk + 29 frame llvm#3: 0x00007ffecc76af28 ntdll.dll`RtlUserThreadStart + 40 ``` When this happens, the test fails with an assertion error that the stopped thread's zeroth frame's current line number does not match the expected line number. This is because the test is looking at the wrong thread: `process.threads[0]`. If we print the list of threads each time the test is run, we notice that threads are sometimes in a different order, within `process.threads`: ```sh Thread 0: thread llvm#4: tid = 0x9c38, 0x00007ffecc7b3bf4 ntdll.dll`NtWaitForWorkViaWorkerFactory + 20 Thread 1: thread llvm#2: tid = 0xa950, 0x00007ffecc7b3bf4 ntdll.dll`NtWaitForWorkViaWorkerFactory + 20 Thread 2: thread #1: tid = 0xab18, 0x00007ff64bc81001 a.out`func_inner at main.c:2:9, stop reason = breakpoint 1.1 Thread 3: thread llvm#3: tid = 0xc514, 0x00007ffecc7b3bf4 ntdll.dll`NtWaitForWorkViaWorkerFactory + 20 Thread 0: thread llvm#3: tid = 0x018c, 0x00007ffecc7b3bf4 ntdll.dll`NtWaitForWorkViaWorkerFactory + 20 Thread 1: thread #1: tid = 0x85c8, 0x00007ff7130c1001 a.out`func_inner at main.c:2:9, stop reason = breakpoint 1.1 Thread 2: thread llvm#2: tid = 0xf344, 0x00007ffecc7b3bf4 ntdll.dll`NtWaitForWorkViaWorkerFactory + 20 Thread 3: thread llvm#4: tid = 0x6a50, 0x00007ffecc7b3bf4 ntdll.dll`NtWaitForWorkViaWorkerFactory + 20 ``` Use `self.thread()` to consistently select the correct thread, instead. Co-authored-by: kendal <[email protected]>

…izations of function templates to USRGenerator (llvm#98027) Given the following: ``` template<typename T> struct A { void f(int); // #1 template<typename U> void f(U); // llvm#2 template<> void f<int>(int); // llvm#3 }; ``` Clang will generate the same USR for `#1` and `llvm#2`. This patch fixes the issue by including the template arguments of dependent class scope explicit specializations in their USRs.

This patch adds a frame recognizer for Clang's `__builtin_verbose_trap`, which behaves like a `__builtin_trap`, but emits a failure-reason string into debug-info in order for debuggers to display it to a user. The frame recognizer triggers when we encounter a frame with a function name that begins with `__clang_trap_msg`, which is the magic prefix Clang emits into debug-info for verbose traps. Once such frame is encountered we display the frame function name as the `Stop Reason` and display that frame to the user. Example output: ``` (lldb) run warning: a.out was compiled with optimization - stepping may behave oddly; variables may not be available. Process 35942 launched: 'a.out' (arm64) Process 35942 stopped * thread #1, queue = 'com.apple.main-thread', stop reason = Misc.: Function is not implemented frame #1: 0x0000000100003fa4 a.out`main [inlined] Dummy::func(this=<unavailable>) at verbose_trap.cpp:3:5 [opt] 1 struct Dummy { 2 void func() { -> 3 __builtin_verbose_trap("Misc.", "Function is not implemented"); 4 } 5 }; 6 7 int main() { (lldb) bt * thread #1, queue = 'com.apple.main-thread', stop reason = Misc.: Function is not implemented frame #0: 0x0000000100003fa4 a.out`main [inlined] __clang_trap_msg$Misc.$Function is not implemented$ at verbose_trap.cpp:0 [opt] * frame #1: 0x0000000100003fa4 a.out`main [inlined] Dummy::func(this=<unavailable>) at verbose_trap.cpp:3:5 [opt] frame llvm#2: 0x0000000100003fa4 a.out`main at verbose_trap.cpp:8:13 [opt] frame llvm#3: 0x0000000189d518b4 dyld`start + 1988 ```

…linux (llvm#99613) Examples of the output: ARM: ``` # ./a.out AddressSanitizer:DEADLYSIGNAL ================================================================= ==122==ERROR: AddressSanitizer: SEGV on unknown address 0x0000007a (pc 0x76e13ac0 bp 0x7eb7fd00 sp 0x7eb7fcc8 T0) ==122==The signal is caused by a READ memory access. ==122==Hint: address points to the zero page. #0 0x76e13ac0 (/lib/libc.so.6+0x7cac0) #1 0x76dce680 in gsignal (/lib/libc.so.6+0x37680) llvm#2 0x005c2250 (/root/a.out+0x145250) llvm#3 0x76db982c (/lib/libc.so.6+0x2282c) llvm#4 0x76db9918 in __libc_start_main (/lib/libc.so.6+0x22918) ==122==Register values: r0 = 0x00000000 r1 = 0x0000007a r2 = 0x0000000b r3 = 0x76d95020 r4 = 0x0000007a r5 = 0x00000001 r6 = 0x005dcc5c r7 = 0x0000010c r8 = 0x0000000b r9 = 0x76f9ece0 r10 = 0x00000000 r11 = 0x7eb7fd00 r12 = 0x76dce670 sp = 0x7eb7fcc8 lr = 0x76e13ab4 pc = 0x76e13ac0 AddressSanitizer can not provide additional info. SUMMARY: AddressSanitizer: SEGV (/lib/libc.so.6+0x7cac0) ==122==ABORTING ``` AArch64: ``` # ./a.out UndefinedBehaviorSanitizer:DEADLYSIGNAL ==99==ERROR: UndefinedBehaviorSanitizer: SEGV on unknown address 0x000000000063 (pc 0x007fbbbc5860 bp 0x007fcfdcb700 sp 0x007fcfdcb700 T99) ==99==The signal is caused by a UNKNOWN memory access. ==99==Hint: address points to the zero page. #0 0x007fbbbc5860 (/lib64/libc.so.6+0x82860) #1 0x007fbbb81578 (/lib64/libc.so.6+0x3e578) llvm#2 0x00556051152c (/root/a.out+0x3152c) llvm#3 0x007fbbb6e268 (/lib64/libc.so.6+0x2b268) llvm#4 0x007fbbb6e344 (/lib64/libc.so.6+0x2b344) llvm#5 0x0055604e45ec (/root/a.out+0x45ec) ==99==Register values: x0 = 0x0000000000000000 x1 = 0x0000000000000063 x2 = 0x000000000000000b x3 = 0x0000007fbbb41440 x4 = 0x0000007fbbb41580 x5 = 0x3669288942d44cce x6 = 0x0000000000000000 x7 = 0x00000055605110b0 x8 = 0x0000000000000083 x9 = 0x0000000000000000 x10 = 0x0000000000000000 x11 = 0x0000000000000000 x12 = 0x0000007fbbdb3360 x13 = 0x0000000000010000 x14 = 0x0000000000000039 x15 = 0x00000000004113a0 x16 = 0x0000007fbbb81560 x17 = 0x0000005560540138 x18 = 0x000000006474e552 x19 = 0x0000000000000063 x20 = 0x0000000000000001 x21 = 0x000000000000000b x22 = 0x0000005560511510 x23 = 0x0000007fcfdcb918 x24 = 0x0000007fbbdb1b50 x25 = 0x0000000000000000 x26 = 0x0000007fbbdb2000 x27 = 0x000000556053f858 x28 = 0x0000000000000000 fp = 0x0000007fcfdcb700 lr = 0x0000007fbbbc584c sp = 0x0000007fcfdcb700 UndefinedBehaviorSanitizer can not provide additional info. SUMMARY: UndefinedBehaviorSanitizer: SEGV (/lib64/libc.so.6+0x82860) ==99==ABORTING ```

…n window scheduling - Fix by comment #1

``` UBSan-Standalone-sparc :: TestCases/Misc/Linux/diag-stacktrace.cpp ``` `FAIL`s on 32 and 64-bit Linux/sparc64 (and on Solaris/sparcv9, too: the test isn't Linux-specific at all). With `UBSAN_OPTIONS=fast_unwind_on_fatal=1`, the stack trace shows a duplicate innermost frame: ``` compiler-rt/test/ubsan/TestCases/Misc/Linux/diag-stacktrace.cpp:14:31: runtime error: execution reached the end of a value-returning function without returning a value #0 0x7003a708 in f() compiler-rt/test/ubsan/TestCases/Misc/Linux/diag-stacktrace.cpp:14:35 #1 0x7003a708 in f() compiler-rt/test/ubsan/TestCases/Misc/Linux/diag-stacktrace.cpp:14:35 llvm#2 0x7003a714 in g() compiler-rt/test/ubsan/TestCases/Misc/Linux/diag-stacktrace.cpp:17:38 ``` which isn't seen with `fast_unwind_on_fatal=0`. This turns out to be another fallout from fixing `__builtin_return_address`/`__builtin_extract_return_addr` on SPARC. In `sanitizer_stacktrace_sparc.cpp` (`BufferedStackTrace::UnwindFast`) the `pc` arg is the return address, while `pc1` from the stack frame (`fr_savpc`) is the address of the `call` insn, leading to a double entry for the innermost frame in `trace_buffer[]`. This patch fixes this by moving the adjustment before all uses. Tested on `sparc64-unknown-linux-gnu` and `sparcv9-sun-solaris2.11` (with the `ubsan/TestCases/Misc/Linux` tests enabled).

``` UBSan-Standalone-sparc :: TestCases/Misc/Linux/diag-stacktrace.cpp ``` `FAIL`s on 32 and 64-bit Linux/sparc64 (and on Solaris/sparcv9, too: the test isn't Linux-specific at all). With `UBSAN_OPTIONS=fast_unwind_on_fatal=1`, the stack trace shows a duplicate innermost frame: ``` compiler-rt/test/ubsan/TestCases/Misc/Linux/diag-stacktrace.cpp:14:31: runtime error: execution reached the end of a value-returning function without returning a value #0 0x7003a708 in f() compiler-rt/test/ubsan/TestCases/Misc/Linux/diag-stacktrace.cpp:14:35 #1 0x7003a708 in f() compiler-rt/test/ubsan/TestCases/Misc/Linux/diag-stacktrace.cpp:14:35 llvm#2 0x7003a714 in g() compiler-rt/test/ubsan/TestCases/Misc/Linux/diag-stacktrace.cpp:17:38 ``` which isn't seen with `fast_unwind_on_fatal=0`. This turns out to be another fallout from fixing `__builtin_return_address`/`__builtin_extract_return_addr` on SPARC. In `sanitizer_stacktrace_sparc.cpp` (`BufferedStackTrace::UnwindFast`) the `pc` arg is the return address, while `pc1` from the stack frame (`fr_savpc`) is the address of the `call` insn, leading to a double entry for the innermost frame in `trace_buffer[]`. This patch fixes this by moving the adjustment before all uses. Tested on `sparc64-unknown-linux-gnu` and `sparcv9-sun-solaris2.11` (with the `ubsan/TestCases/Misc/Linux` tests enabled). (cherry picked from commit 3368a32)

kaiyan96 force-pushed the bugfix_memcpy_invalid_modification branch 5 times, most recently from 602591c to 40084e5 Compare October 9, 2023 02:33

[MemCpyOpt] Fix the invalid code modification for GEP

ab78cf1

Relocate the GEP modification to a later stage of the function performCallSlotOption, ensuring that the code remains unchanged if the optimization fails.

kaiyan96 force-pushed the bugfix_memcpy_invalid_modification branch from 40084e5 to ab78cf1 Compare October 9, 2023 06:10

kaiyan96 closed this Jul 19, 2024

kaiyan96 added a commit that referenced this pull request Jul 25, 2024

[llvm][CodeGen] Address the issue of multiple resource reservations i…

6212248

…n window scheduling - Fix by comment #1

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

[MemCpyOpt] Fix the invalid code modification in Memcpy pass #1

[MemCpyOpt] Fix the invalid code modification in Memcpy pass #1

kaiyan96 commented Oct 7, 2023

[MemCpyOpt] Fix the invalid code modification in Memcpy pass #1

[MemCpyOpt] Fix the invalid code modification in Memcpy pass #1

Conversation

kaiyan96 commented Oct 7, 2023