Return-Path: X-Spam-Checker-Version: SpamAssassin 3.4.0 (2014-02-07) on aws-us-west-2-korg-lkml-1.web.codeaurora.org Received: from phobos.denx.de (phobos.denx.de [85.214.62.61]) (using TLSv1.2 with cipher ECDHE-RSA-AES128-GCM-SHA256 (128/128 bits)) (No client certificate requested) by smtp.lore.kernel.org (Postfix) with ESMTPS id C0D0DC433F5 for ; Fri, 27 May 2022 14:03:32 +0000 (UTC) Received: from h2850616.stratoserver.net (localhost [IPv6:::1]) by phobos.denx.de (Postfix) with ESMTP id 6137B84347; Fri, 27 May 2022 16:03:25 +0200 (CEST) Authentication-Results: phobos.denx.de; dmarc=pass (p=none dis=none) header.from=gmail.com Authentication-Results: phobos.denx.de; spf=pass smtp.mailfrom=u-boot-bounces@lists.denx.de Authentication-Results: phobos.denx.de; dkim=pass (2048-bit key; unprotected) header.d=gmail.com header.i=@gmail.com header.b="UIlkTfru"; dkim-atps=neutral Received: by phobos.denx.de (Postfix, from userid 109) id F3FA484248; Fri, 27 May 2022 16:03:17 +0200 (CEST) Received: from mail-qt1-x829.google.com (mail-qt1-x829.google.com [IPv6:2607:f8b0:4864:20::829]) (using TLSv1.3 with cipher TLS_AES_128_GCM_SHA256 (128/128 bits)) (No client certificate requested) by phobos.denx.de (Postfix) with ESMTPS id 5FE5C83E73 for ; Fri, 27 May 2022 16:03:14 +0200 (CEST) Authentication-Results: phobos.denx.de; dmarc=pass (p=none dis=none) header.from=gmail.com Authentication-Results: phobos.denx.de; spf=pass smtp.mailfrom=seanga2@gmail.com Received: by mail-qt1-x829.google.com with SMTP id w17so5288944qtk.13 for ; Fri, 27 May 2022 07:03:14 -0700 (PDT) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=gmail.com; s=20210112; h=from:to:cc:subject:date:message-id:mime-version :content-transfer-encoding; bh=I5SGjKGIwY1mKpa2IhD5TYwaiHzM0XYE4sncT3P3+9U=; b=UIlkTfru+vR2s1WJWXQrb3Qdzv6AJbeqPYTMH+yx3nnHsWiwNQ8ku0+JUOu55poky5 7wdCQ+C+4iX5DAYx+SLI8bIeh9zfEdea96uZdLyxTTP5ry18RlWhipNVdxSYxhDSQtuf 4fXMKIOQicHVXx/fqOw1P4B60/A/4wk1q4eQM7ZcKgIlSyzNjOnz3I/uBM7hWWw6WCwf uoHlrzQetqsMwsFyI1fIns6xi0cUQ4guF02pidaq/fCaYJdJRtlpxjpmklZCkZ1n8nFA JCvbJClhQ5ZG7iJaSC+DHB8LIlaqZgFChNUyqmR4q83IAPtqmf7ppZ+GnRwiRnfh3kXD y/Cw== X-Google-DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=1e100.net; s=20210112; h=x-gm-message-state:from:to:cc:subject:date:message-id:mime-version :content-transfer-encoding; bh=I5SGjKGIwY1mKpa2IhD5TYwaiHzM0XYE4sncT3P3+9U=; b=j66eqrTM/U2zRvGf3igHp0XCSdGvqcrYcDLgsjBSHbeukMcU4vElPSd6cqOHMelG90 6kcznwHb+xyitOvV691Yi+4uj/QKlvGyJ8EM6h9jJGjzydkYOn0i1HSjN2BqXF11vRY7 C1fQUkVaToSNZPJDgzMX2w07cpkoB721E+X7YaD9QQNdakHI0t91i9H/foyNPhIsAQCd iVxDi7RKWs2BxCIjidMj2Y5loinAdD3tR15jZPzVlwTXKJphhY6qxMfbN4gXfK+NTRYK QQlfUPGq0CVoEQ9/S3iTm/UhJp+V5ePhY44JCN65BOHXyrD/4WpPbo3v7SmRnJpuT9vc CogA== X-Gm-Message-State: AOAM532PniXKf5WYmxcnl+oCxU78Zu4NyOUpFKFMwD52JHmjvnp6hjos 4m19I9QffCEGZUVGXLneP9w0oj0HS9M= X-Google-Smtp-Source: ABdhPJwviWGDhOrIAOMuDYD+YhyUSPgHK/V5Y9fzzDvMGstnhlYbV+seKJaCsYjuPT1u3fu975dwgQ== X-Received: by 2002:a05:622a:1105:b0:2fb:8bbf:fa0a with SMTP id e5-20020a05622a110500b002fb8bbffa0amr9532007qty.452.1653660192900; Fri, 27 May 2022 07:03:12 -0700 (PDT) Received: from godwin.fios-router.home (pool-173-73-95-180.washdc.fios.verizon.net. [173.73.95.180]) by smtp.gmail.com with ESMTPSA id w138-20020a376290000000b0069fd9de088esm2814124qkb.93.2022.05.27.07.03.12 (version=TLS1_3 cipher=TLS_AES_256_GCM_SHA384 bits=256/256); Fri, 27 May 2022 07:03:12 -0700 (PDT) From: Sean Anderson To: u-boot@lists.denx.de Cc: Heinrich Schuchardt , Tom Rini , Andrew Scull , Simon Glass , Sean Anderson Subject: [PATCH v3 1/2] doc: sandbox: Add additional valgrind documentation Date: Fri, 27 May 2022 10:02:59 -0400 Message-Id: <20220527140300.682989-1-seanga2@gmail.com> X-Mailer: git-send-email 2.35.1 MIME-Version: 1.0 Content-Transfer-Encoding: 8bit X-BeenThere: u-boot@lists.denx.de X-Mailman-Version: 2.1.39 Precedence: list List-Id: U-Boot discussion List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Errors-To: u-boot-bounces@lists.denx.de Sender: "U-Boot" X-Virus-Scanned: clamav-milter 0.103.5 at phobos.denx.de X-Virus-Status: Clean This documents some additional options which can be used with valgrind, as well as directions for future work. It also fixes up inline literals to actually be inline literals (and not italics). The content of this documentation is primarily adapted from [1]. [1] https://lore.kernel.org/u-boot/57cb4b49-fa30-1194-9ac3-faa53e8033bd@gmail.com/ Signed-off-by: Sean Anderson --- Changes in v3: - Fix some spelling errors - Reword ut all docs for clarity Changes in v2: - Reworked most documentation to address Heinrich's feedback doc/arch/sandbox.rst | 90 ++++++++++++++++++++++++++++++++++++++++---- 1 file changed, 82 insertions(+), 8 deletions(-) diff --git a/doc/arch/sandbox.rst b/doc/arch/sandbox.rst index e1119492b4..79bcef4f5e 100644 --- a/doc/arch/sandbox.rst +++ b/doc/arch/sandbox.rst @@ -479,18 +479,92 @@ It is possible to run U-Boot under valgrind to check memory allocations:: valgrind ./u-boot -For more detailed results, enable `CONFIG_VALGRIND`. There are many false -positives due to `malloc` itself. Suppress these with:: +However, this does not give very useful results. The sandbox allocates a memory +pool via mmap(). U-Boot's internal malloc() and free() work on this memory pool. +Custom allocators and deallocators are invisible to valgrind by default. To +expose U-Boot's malloc() and free() to valgrind, enable ``CONFIG_VALGRIND``. +Enabling this option will inject placeholder assembler code which valgrind +interprets. This is used to annotate sections of memory as safe or unsafe, and +to inform valgrind about malloc()s and free()s. There are currently no standard +placeholder assembly sequences for RISC-V, so this option cannot be enabled on +that architecture. + +Malloc's bookkeeping information is marked as unsafe by default. However, this +will generate many false positives when malloc itself accesses this information. +These warnings can be suppressed with:: valgrind --suppressions=scripts/u-boot.supp ./u-boot -If you are running sandbox SPL or TPL, then valgrind will not by default -notice when U-Boot jumps from TPL to SPL, or from SPL to U-Boot proper. To -fix this, use `--trace-children=yes`. To show who alloc'd some troublesome -memory, use `--track-origins=yes`. To uncover possible errors, try running all -unit tests with:: +Additionally, you may experience false positives if U-Boot is using a smaller +pointer size than your host architecture. This is because the pointers used by +U-Boot will only contain 32 bits of addressing information. When interpreted as +64-bit pointers, valgrind will think that they are not initialized properly. To +fix this, enable ``CONFIG_SANDBOX64`` (such as via ``sandbox64_defconfig``) +when running on a 64-bit host. - valgrind --track-origins=yes --suppressions=scripts/u-boot.supp ./u-boot -Tc 'ut all' +Additional options +^^^^^^^^^^^^^^^^^^ + +The following options are useful in addition to the above examples: + +* ``--trace-childen=yes`` will tell valgrind to keep tracking subprocesses, such + as when U-Boot jumps from TPL to SPL, or from SPL to U-Boot proper. +* ``--track-origins=yes`` will (for a small overhead) tell valgrind to keep + track of who allocated some troublesome memory. +* ``--error-limit=no`` will enable printing more than 1000 errors in a + single session. +* ``--vgdb=yes --vgdb-error=0`` will let you use gdb to attach like:: + + gdb -ex "target remote | vgdb" u-boot + + This is very helpful for inspecting the program state when there is + an error. +* Passing ``-Tc 'ut all'`` to U-Boot will run unit tests automatically. Note + that not all unit tests will succeed in the default configuration. +* Passing ``-t cooked`` to U-Boot will keep the console in a sane state if you + terminate it early (instead of having to run tset). + +Future work +^^^^^^^^^^^ + +The biggest limitation to the current approach is that supressions don't +"un-taint" uninitialized memory accesses. Currently, dlmalloc's bookkeeping +information is marked as a "red zone." This means that all reads to that zone +are marked as illegal by valgrind. This is fine for regular code, but dlmalloc +really does need to access this area, so we suppress its violations. However, if +dlmalloc then passes a result calculated from a "tainted" access, that result is +still tainted. So the first accessor will raise a warning. This means that every +construct like + +.. code-block:: + + foo = malloc(sizeof(*foo)); + if (!foo) + return -ENOMEM; + +will raise a warning when we check the result of malloc. Whoops. + +There are at least four possible ways to address this: + +* Don't mark dlmalloc bookkeeping information as a red zone. This is the + simplest solution, but reduces the power of valgrind immensely, since we can + no longer determine that (e.g.) access past the end of an array is undefined. +* Implement red zones properly. This would involve growing every allocation by a + fixed amount (16 bytes or so) and then using that extra space for a real red + zone that neither regular code nor dlmalloc needs to access. Unfortunately, + this would probably some fairly intensive surgery to dlmalloc to add/remove + the offset appropriately. +* Mark bookkeeping information as valid before we use it in dlmalloc, and then + mark it invalid before returning. This would be the most correct, but it would + be very tricky to implement since there are so many code paths to mark. I + think it would be the most effort out of the three options here. +* Use the host malloc and free instead of U-Boot's custom allocator. This will + eliminate the need to annotate dlmalloc. However, using a different allocator + for sandbox will mean that bugs in dlmalloc will only be tested when running + on read (or emulated) hardware. + +Until one of the above options are implemented, it will remain difficult +to sift through the massive amount of spurious warnings. Testing ------- -- 2.35.1