nickdesaulniers

// Case 1A: Aggregate Call Return Temporaries (CodeGenFunction::EmitAnyExprToTemp, #68747  
// https://github.com/llvm/llvm-project/issues/68747)
struct a {
  long b[8];
};
struct a f(void);
void g(struct a);
void h(void) {
  g(f());
  g(f());

nickdesaulniers

Hunting Down Private Dirty Memory in Linux Shared Libraries

Shared libraries (.so files) on Linux are designed to save memory by sharing physical memory pages across processes via the kernel's Page Cache. But in practice, shared libraries often consume substantially more physical RAM than expected due to Copy-On-Write (COW) dirtying.

When a process writes to any page in a shared object—or when the dynamic linker (ld.so) writes relocated pointers at startup—that 4 KB page is duplicated in physical RAM as Private_Dirty memory. If a library has 5 MB of dirty pages and is loaded by 100 processes, that library silently costs 500 MB of unshareable RAM.

Here is a practical, command-line-driven playbook for identifying private dirty hotspots in shared objects and resolving them in source code.

---

nickdesaulniers

. /tmpfs/src/git/ndk32/llvm-project/lldb/source/Plugins/Process/Linux/NativeProcessLinux.h
.. /tmpfs/src/git/ndk32/sysroots/ndk/x86/usr/include/c++/v1/csignal
... /tmpfs/src/git/ndk32/sysroots/ndk/x86/usr/include/c++/v1/__config
.... /tmpfs/src/git/ndk32/sysroots/ndk/x86/usr/include/c++/v1/__config_site
.... /tmpfs/src/git/ndk32/sysroots/ndk/x86/usr/include/c++/v1/__configuration/abi.h
..... /tmpfs/src/git/ndk32/sysroots/ndk/x86/usr/include/c++/v1/__configuration/compiler.h
..... /tmpfs/src/git/ndk32/sysroots/ndk/x86/usr/include/c++/v1/__configuration/platform.h
...... /tmpfs/src/git/ndk32/sysroots/ndk/x86/usr/include/features.h
....... /tmpfs/src/git/ndk32/sysroots/ndk/x86/usr/include/sys/cdefs.h
........ /tmpfs/src/git/ndk32/sysroots/ndk/x86/usr/include/android/versioning.h

nickdesaulniers

$ llvm-nm compiler-rt/lib/linux/libclang_rt.builtins-x86_64.a | grep T | cut -d ' ' -f 3
__absvdi2
__absvsi2
__absvti2
__adddf3
__addsf3
__addvdi3
__addvsi3
__addvti3
__ashldi3

nickdesaulniers

(gdb) bt
#0  __GI_raise (sig=sig@entry=6) at ../sysdeps/unix/sysv/linux/raise.c:50
#1  0x0000fffff7ba0a60 in __GI_abort () at abort.c:79
#2  0x000000000040a8c8 in scudo::die ()
    at /home/libc-buildbot/libc-aarch64-ubuntu/libc-aarch64-ubuntu-fullbuild-dbg/llvm-project/compiler-rt/lib/scudo/standalone/linux.cpp:48
#3  0x000000000040b248 in scudo::reportRawError (
    Message=0xffffffffed78 "Scudo ERROR: internal map failure (error desc=Success)\n")
    at /home/libc-buildbot/libc-aarch64-ubuntu/libc-aarch64-ubuntu-fullbuild-dbg/llvm-project/compiler-rt/lib/scudo/standalone/report.cpp:58
#4  0x000000000040b9a4 in scudo::reportMapError (SizeIfOOM=0)
    at /home/libc-buildbot/libc-aarch64-ubuntu/libc-aarch64-ubuntu-fullbuild-dbg/llvm-project/compiler-rt/lib/scudo/standalone/report_linux.cpp:33

nickdesaulniers

commit b0fb5270bc85dd049ddb5f4308e04136256d6783
Author: Nick Desaulniers <ndesaulniers@google.com>
Date:   Thu Oct 3 08:18:10 2024 -0700

    WIP fpbits

diff --git a/libc/src/__support/FPUtil/FPBits.h b/libc/src/__support/FPUtil/FPBits.h
index 5d1f633bb56e..69307653a8e8 100644
--- a/libc/src/__support/FPUtil/FPBits.h
+++ b/libc/src/__support/FPUtil/FPBits.h

nickdesaulniers

diff --git a/tools/testing/selftests/kselftest_harness.h b/tools/testing/selftests/kselftest_harness.h
index b634969cbb6f..da4c91793cdc 100644
--- a/tools/testing/selftests/kselftest_harness.h
+++ b/tools/testing/selftests/kselftest_harness.h
@@ -450,14 +450,18 @@ static inline pid_t clone3_vfork(void)
                                _metadata->exit_code = KSFT_FAIL; \
                        } \
                } \
+               _Bool expected = false; \
+               _Bool desired = true; \

nickdesaulniers

// -Wno-asm-operand-widths
// void input_pressure(int x) {
//     // asm ("# %0"::"m"(x):"r0","r1","r2","r3","r12","lr","r4","r5","r6","r7","r8","r9","r10","r11");
//   asm (""::"m"(x));
// }

struct foo {
  long a, b, c, d, e;
};
int zzz (struct foo my_foo) {

nickdesaulniers

bloaty -n 0 -s file after.a -- before.a
    FILE SIZE        VM SIZE    
 --------------  -------------- 
  +234% +2.05Ki  +212%    +816    .rodata.cst16
  +1.8% +1.48Ki  [ = ]       0    .symtab
  +0.8%    +920  [ = ]       0    .strtab
   +31%    +864   +33%    +870    .text._ZN22__llvm_libc_18_0_0_git8internal27hexadecimal_string_to_floatIeEENS_14StrToNumResultINS0_13ExpandedFloatIT_EEEEPKccNS0_14RoundDirectionE
   +39%    +736   +44%    +760    .text._ZN22__llvm_libc_18_0_0_git8internal20decimal_exp_to_floatIeEENS0_18FloatConvertReturnIT_EENS0_13ExpandedFloatIS3_EEPKcbNS0_14RoundDirectionE
  [NEW]    +480  [NEW]    +411    .text._ZN22__llvm_libc_18_0_0_gitL31construct_thread_name_file_pathERNS_3cpp12StringStreamEi
  [NEW]    +288  [NEW]    +223    .text._ZN22__llvm_libc_18_0_0_git11printf_core11FloatWriter18write_middle_blockEj

nickdesaulniers

struct foo {
    long long x;
};

struct foo foo_factory (void);
struct foo div (struct foo, struct foo);

long long bar (void) {
    return div(div(foo_factory(), foo_factory()), div(foo_factory(), foo_factory())).x;
}