From nobody Thu Feb 22 01:23:08 2024 X-Original-To: dev-commits-src-main@mlmmj.nyi.freebsd.org Received: from mx1.freebsd.org (mx1.freebsd.org [IPv6:2610:1c1:1:606c::19:1]) by mlmmj.nyi.freebsd.org (Postfix) with ESMTP id 4TgFlF3JXhz5Bm5m for ; Thu, 22 Feb 2024 01:23:21 +0000 (UTC) (envelope-from jrtc27@jrtc27.com) Received: from mail-wr1-f45.google.com (mail-wr1-f45.google.com [209.85.221.45]) (using TLSv1.3 with cipher TLS_AES_128_GCM_SHA256 (128/128 bits) key-exchange X25519 server-signature RSA-PSS (4096 bits) server-digest SHA256 client-signature RSA-PSS (2048 bits) client-digest SHA256) (Client CN "smtp.gmail.com", Issuer "GTS CA 1D4" (verified OK)) by mx1.freebsd.org (Postfix) with ESMTPS id 4TgFlF13gGz4j8P for ; Thu, 22 Feb 2024 01:23:21 +0000 (UTC) (envelope-from jrtc27@jrtc27.com) Authentication-Results: mx1.freebsd.org; none Received: by mail-wr1-f45.google.com with SMTP id ffacd0b85a97d-33d118a181fso3853053f8f.1 for ; Wed, 21 Feb 2024 17:23:21 -0800 (PST) X-Google-DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=1e100.net; s=20230601; t=1708564999; x=1709169799; h=to:references:message-id:content-transfer-encoding:cc:date :in-reply-to:from:subject:mime-version:x-gm-message-state:from:to:cc :subject:date:message-id:reply-to; bh=GwpFvLKQQKeZvLnn326m21S5WjmEDf570EsB3M7L2rs=; b=czVroTElDpBd1ni4tWKXyE8cYcLvoFiLcpkaKgmOkp9guTBB8evxK0MUqtIqJ5/exY acJl/xLDHQRTAjJ17jw92fmiLpEsR6Sx8lj934XY3wEtnJ1LINyLFM53mK78P+oIRCIb dbKJazgKwWvOk62JfnADwZ1HzduidOhzhjWf91udvvYFpMR8q8wS4KxN95wqe8kfGT/r W+pI+fw0Rm1xOYxlIzqbYsfwuK2Lt70bOdxeduR3T73TFy6Cu1z8nkOyLQQ17iALYj9B IoyRs8c6U2+mtFYgK6yy8x2h1tZ+Z6V+2S8soZKBysBXNZHruyt5qYmTvyrwG5wYCJts isoA== X-Forwarded-Encrypted: i=1; AJvYcCUPjuu1jVGnrOqv957xyNmSCPH+4iE0/qAofrxkkUwQSogBURGmXDHwQU4cqp0KFY3ukZ8uyLHIq5ZkvUUaP52oksLfqgH7e8q1fVE0hW+4GQ== X-Gm-Message-State: AOJu0YyaG27ADdiXw5JlFfGCHl2f/uc6Vx1XGtzMW1clKuuJqk3LjR6A 7CZODfx6sTChivL2rRJTv+5IVDJPEd8HAOJ5FKBvu6aPnqMpLHmwTPRs7+DCwj8= X-Google-Smtp-Source: AGHT+IEjqUnjc++rPljqaNoTqfxqJlwzbZFXOuz6t++L4U9Jc4bMggxltBcp6VraA4qjklWbopgB6w== X-Received: by 2002:a5d:5f91:0:b0:33d:4c06:8614 with SMTP id dr17-20020a5d5f91000000b0033d4c068614mr8124188wrb.37.1708564999371; Wed, 21 Feb 2024 17:23:19 -0800 (PST) Received: from smtpclient.apple ([131.111.5.246]) by smtp.gmail.com with ESMTPSA id b7-20020adfe647000000b0033b406bc689sm18678602wrn.75.2024.02.21.17.23.18 (version=TLS1_2 cipher=ECDHE-ECDSA-AES128-GCM-SHA256 bits=128/128); Wed, 21 Feb 2024 17:23:18 -0800 (PST) Content-Type: text/plain; charset=utf-8 List-Id: Commit messages for the main branch of the src repository List-Archive: https://lists.freebsd.org/archives/dev-commits-src-main List-Help: List-Post: List-Subscribe: List-Unsubscribe: Sender: owner-dev-commits-src-main@freebsd.org X-BeenThere: dev-commits-src-main@freebsd.org Mime-Version: 1.0 (Mac OS X Mail 16.0 \(3774.200.91.1.1\)) Subject: Re: git: 8271d9b99a3b - main - libsys: remove usage of pthread_once and _once_stub From: Jessica Clarke In-Reply-To: Date: Thu, 22 Feb 2024 01:23:08 +0000 Cc: "src-committers@freebsd.org" , "dev-commits-src-all@freebsd.org" , "dev-commits-src-main@freebsd.org" Content-Transfer-Encoding: quoted-printable Message-Id: References: <202402210029.41L0TOH5000231@gitrepo.freebsd.org> <964A29A2-4C51-4037-8EBE-320008D48AE0@freebsd.org> <4715B319-B7DE-4D06-9F27-00CFE5AF89A7@freebsd.org> To: Konstantin Belousov X-Mailer: Apple Mail (2.3774.200.91.1.1) X-Spamd-Bar: ---- X-Rspamd-Queue-Id: 4TgFlF13gGz4j8P X-Rspamd-Pre-Result: action=no action; module=replies; Message is reply to one we originated X-Spamd-Result: default: False [-4.00 / 15.00]; REPLY(-4.00)[]; ASN(0.00)[asn:15169, ipnet:209.85.128.0/17, country:US] On 22 Feb 2024, at 01:11, Konstantin Belousov = wrote: >=20 > On Wed, Feb 21, 2024 at 05:23:10PM +0000, Jessica Clarke wrote: >> On 21 Feb 2024, at 14:17, Konstantin Belousov = wrote: >>>=20 >>> On Wed, Feb 21, 2024 at 12:51:04AM +0000, Jessica Clarke wrote: >>>> On 21 Feb 2024, at 00:29, Konstantin Belousov = wrote: >>>>>=20 >>>>> The branch main has been updated by kib: >>>>>=20 >>>>> URL: = https://cgit.FreeBSD.org/src/commit/?id=3D8271d9b99a3b98c662ee9a6257a14428= 4b7e1728 >>>>>=20 >>>>> commit 8271d9b99a3b98c662ee9a6257a144284b7e1728 >>>>> Author: Konstantin Belousov >>>>> AuthorDate: 2024-02-20 14:45:29 +0000 >>>>> Commit: Konstantin Belousov >>>>> CommitDate: 2024-02-21 00:26:11 +0000 >>>>>=20 >>>>> libsys: remove usage of pthread_once and _once_stub >>>>>=20 >>>>> that existed in auxv.c, use simple bool gate instead. This leaves = a >>>>> small window if two threads try to call _elf_aux_info(3) = simultaneously. >>>>> The situation is safe because auxv parsing is really idempotent. = The >>>>> parsed data is the same, and we store atomic types (int/long/ptr) = so >>>>> double-init does not matter. >>>>=20 >>>> You still need to load acquire and store release aux_once though, >>>> otherwise you can see aux_once as true yet read the pre-initialised >>>> data. In practice that=E2=80=99s surely very hard to hit, but the = code as >>>> written is now wrong. Also, idempotence should probably be made >>>> unnecessary by using 0/1/2 state for uninitialised/initialising/ >>>> initialised, as it=E2=80=99s still technically UB from a C AM = perspective due >>>> to not being data race free if two threads initialise at the same = time. >>>> Better to just do the correct thing rather than risk things going = wrong. >>>=20 >>> There is too much to handle 'in process' state for loosing thread, I = need >>> the whole libthr machinery. >>=20 >> What do you need libthr for? In pseudo-C: >>=20 >> x =3D load_acquire(&aux_once) >> if (__predict_true(x =3D=3D 2)) >> return; >> if (x =3D=3D 1 || !compare_exchange_strong_acquire(&aux_once, &x, 1)) = { >> while (x !=3D 2) { >> yield(); >> x =3D load_acquire(&aux_once) >> } >> return; >> } >> /* initialise as before */ >> store_release(&aux_once, 2); >>=20 >> I believe that=E2=80=99s all you need. Or compare exchange 0 to 1 as = the >> initial operation; makes the source code shorter at the expense of a >> more expensive fast path: >>=20 >> x =3D 0; >> if (__predict_true(!compare_exchange_strong_acquire(&aux_once, &x, = 1)) { >> while (__predict_false(x !=3D 2)) { >> yield(); >> x =3D load_acquire(&aux_once) >> } >> return; >> } >> /* initialise as before */ >> store_release(&aux_once, 2); >>=20 >> I probably have bugs in the above, but you get the gist. > The bug in the fragment above is with the yield(). If low-priority = thread > enters the '1' (in progress) block, and then is preempted by = high-priority > thread also entering init_auxv(), the process would never make a = progress. >=20 > This is why I said that we need libthr (or umtx), to use real locking = and > move the waiting thread off cpu. In kernel, yield can be used in = similar > situations because we can bump the priority, although it is tricky. Yes, priority inversion is an issue here. But this is (without all the C++ abstraction) how libcxxrt implements __cxa_guard_acquire, so if it=E2=80=99s good enough for C++ constructors for static storage = duration objects declared at local scope, surely it=E2=80=99s also good enough = for aux_once? And if it=E2=80=99s not good enough for aux_once then libcxxrt = should be deemed broken... One could easily adapt the above to use UMTX_OP_WAIT/WAKE though. Jess >>> I added the fences, thanks for noting. >>=20 >> Thanks. >>=20 >> Jess >>=20 >>> WRT being UB from pure C, we already have much more assumptions = about >>> atomicity.