Sanitizers run on NetBSD/sparc64.

Looks ok for NetBSD.

NetBSD OK!

MAP_NORESERVE is dead on NetBSD as this is the default mode always.

Would it be possible for NetBSD to also use this approach for the libc_pthread_ functions? Or is only a subset exported as libc_pthread_?

Can you follow the NetBSD approach and go for something like TSAN_MAYBE_INTERCEPT_NETBSD_ALIAS instead of patching shared code between all platforms?

Merging is a good idea as NetBSD might duplicate these code chunks too.

In D95696#2532032, @strejda wrote:

Oki, I see, thanks for response. Please give me a few days to analyze (and eventually to implement).

@clayborg please have a look!

These symbols are usually macros + libc calls.

NetBSD uses macros too, but undefines them inside the header for __cplusplus. The proposed patch looks fine. This is a common source of portability issues, the stdio macros.

In D93848#2472902, @MaskRay wrote:

In D93848#2472833, @krytarowski wrote:

Well... technically speaking GLIBC is at least: GNU/Linux, GNU/FreeBSD and HURD.

Hurd and GNU/kFreeBSD are not supported. Even if they do, many defined(__GLIBC__) is inside a #if SANITIZER_LINUX, so they will work automatically.

Well... technically speaking GLIBC is at least: GNU/Linux, GNU/FreeBSD and HURD.

Please include stdlib.h for alloca, unless you target SunOS. alloca.h is unavailable on NetBSD.

In D92667#2434223, @emaste wrote:

No objection here. I'm curious why the two modified tests work on Linux or NetBSD today though?

Ideally we should iterate over the startup process and investigate the state of the r_debug structure. Once it gets initialized, set the software brakpoint to r_brk and keep track of the dynamic loading and unloading of libraries. The tricky part is to detect the right moment to plug into r_brk, early enough in the startup process before calling constructors, loading the libraries etc and late enough to become initialized.

In D92187#2424109, @emaste wrote:

One thing that FreeBSD should do, is to upgrade to the protocol version 1 (stored in r_version), like Linux, NetBSD and OpenBSD.

It looks like Linux has always used r_version=1 (since 1995).

AFAICT we can just add r_ldbase and set version to 1.
See https://reviews.freebsd.org/D27429

In D92187#2423018, @emaste wrote:

I'm curious how gdb handles this, and asked @bsdjhb if he knows off hand.

Is there a brief description of how this works on Linux and/or NetBSD?

I would mark stmmX as an alias to stX and keep stX as the default for all platforms. stmmX could be an alias for everybody for legacy reasons.

I propose to filter all user induced signals at once and check & 0x10000 and SI_USER.

I wonder why NetBSD fails having the same event handling.

In D90757#2378404, @theraven wrote:

Does the new plugin work with processes that are created with pdfork? The last time I tried this, it caused the old plugin to lock up the debugger entirely. Please can you ensure that there are tests that cover pdfork and cap_enter in the child? These are currently quite badly broken.

It's relevant but I encountered issues to pass the review..

Looks OK from the NetBSD point of view.

Host a LLVM buildbot for DragonFly first, otherwise it will be even not ever build-tested.

Please leave builtins intact. ubsan might work, but removing it too is fine.

Please, First remove unused + unusable code for already introduced OS. At most cutting it to ubsan.

This contains unrelated code to ubsan/dragonflybsd. While there can you remove unsupported code for other OSs you introduced? That was never operational?

In D89182#2324267, @labath wrote:

FWIW, PTHREAD_KILL is strongly discouraged on linux. But if the situation is different on freebsd, then fine.

+ @bsdjhb John, could you have a look?

Can we rename the register names in the darwin plugin?

Confirmed to work.

In D87562#2277788, @vitalybuka wrote:

Can you please clang format it as well