From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from proxmox.maurer-it.com (proxmox.maurer-it.com [212.186.127.180]) (using TLSv1.2 with cipher ECDHE-RSA-AES256-GCM-SHA384 (256/256 bits)) (No client certificate requested) by ml01.01.org (Postfix) with ESMTPS id 3BD562095BB68 for ; Wed, 6 Sep 2017 06:32:37 -0700 (PDT) Received: from proxmox.maurer-it.com (localhost [127.0.0.1]) by proxmox.maurer-it.com (Proxmox) with ESMTP id 9129F110E054; Wed, 6 Sep 2017 15:35:25 +0200 (CEST) To: Laszlo Ersek , edk2-devel@lists.01.org Cc: Ard Biesheuvel , "Gao, Liming" , "Zhu, Yonghong" References: <9fbd059f-18d8-a798-da00-951f85c9fd1a@proxmox.com> <554afa66-cc81-6115-ff83-2b0b6f01d455@redhat.com> From: Thomas Lamprecht Message-ID: <697823cf-4e21-137c-eb5c-8b4a2bce14d0@proxmox.com> Date: Wed, 6 Sep 2017 15:35:23 +0200 User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:52.0) Gecko/20100101 Thunderbird/52.3.0 MIME-Version: 1.0 In-Reply-To: <554afa66-cc81-6115-ff83-2b0b6f01d455@redhat.com> Subject: Re: OvmfPkg/IoMmuDxe: unused-const-variable warning fails release build with GCC 6.3 X-BeenThere: edk2-devel@lists.01.org X-Mailman-Version: 2.1.22 Precedence: list List-Id: EDK II Development List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , X-List-Received-Date: Wed, 06 Sep 2017 13:32:37 -0000 Content-Type: text/plain; charset=utf-8; format=flowed Content-Language: en-US Content-Transfer-Encoding: quoted-printable On 09/06/2017 01:15 PM, Laszlo Ersek wrote: > Hello Thomas, >=20 > (CC Ard, Liming, Yonghong) >=20 > On 09/06/17 09:44, Thomas Lamprecht wrote: >> Hi, >> >> commit 2ad6ba80a1bd58382bde6b994070f7c01d2fb48d >> Author: Laszlo Ersek >> Date: Wed Aug 30 14:00:58 2017 +0200 >> >> OvmfPkg/IoMmuDxe: IoMmuMap(): log nicer and more informative DEBU= G msgs >> >> triggers a unused-const-variable warning for mBusMasterOperationName, >> and thus for releases with warnings-as-errors a build error. >> >> I'm using a quite vanilla Debian Stretch machine on amd64/x86_64 >> (running in qemu/KVM) as build host, gcc version is 6.3.0 20170516 >> (Debian 6.3.0-18) >> >> My build procedure looks like: >> >> # make -C BaseTools/ >> # . edksetup.sh >> # OvmfPkg/build.sh -a X64 -b RELEASE -n 4 -t GCC5 >> >> With current master (12cfc9009e7cf1a69ca675110c2cf6e21b152992) checked= >> out. >> >> I suspect that gcc, at least in this version, cannot track the usage >> of the variable in crime in the DEBUG macros, and thus (falsely?) >> detects this warning. >> >> So I'm not quite sure if this is a problem with edk2/Ovmf itself or a >> problem stemming from gcc. >> >> The following patch fixes the problem quite nonchalantly by adding an >> unused attribute, which is highly probably not what is wanted as a >> clean fix - I guess - but it achieves my desired result :) >> >> ----8<---- >> diff --git a/OvmfPkg/IoMmuDxe/AmdSevIoMmu.c >> b/OvmfPkg/IoMmuDxe/AmdSevIoMmu.c >> index bc57de5b57..8c0b8b0931 100644 >> --- a/OvmfPkg/IoMmuDxe/AmdSevIoMmu.c >> +++ b/OvmfPkg/IoMmuDxe/AmdSevIoMmu.c >> @@ -48,7 +48,7 @@ STATIC LIST_ENTRY mRecycledMapInfos =3D >> INITIALIZE_LIST_HEAD_VARIABLE ( >> // ASCII names for EDKII_IOMMU_OPERATION constants, for debug loggin= g. >> // >> STATIC CONST CHAR8 * CONST >> -mBusMasterOperationName[EdkiiIoMmuOperationMaximum] =3D { >> +mBusMasterOperationName[EdkiiIoMmuOperationMaximum] __attribute__ >> ((unused)) =3D { >> "Read", >> "Write", >> "CommonBuffer", >> ---->8---- >> >> It naturally could be that I've got something wrong, wouldn't be the >> first time, but I suspect that it's edk2 in combination with my GCC >> version this time. As I lack in depth knowledge of Ovmf it'd be nice >> if someone could confirm my suspicion. >> >> CC'ing Laszlo as he is the author of the patch which lets this problem= >> trigger and following this mailing list since a bit it seems that he >> surely has the in depth knowledge. Sorry that I wasn't patient enough >> to test other compiler version before posting here. >=20 > Thanks for the report. >=20 > When I (recently) wrote the above commit, I was fully aware that the > variable would become unused if DEBUG macro invocations were compiled > out of the code (i.e., in the RELEASE build target). >=20 > However, that didn't worry me because we had already introduced a > distinction for such variables, between RELEASE and DEBUG builds. As yo= u > can expect, this phenomenon is quite wide-spread; for example a Status > variable may capture a function return value, only to be ASSERT()-ed > upon. In a RELEASE build, the ASSERT() disappears, and we don't want th= e > compiler to yell at us for setting, but never checking, Status. >=20 Ah yes, that makes sense. > ... Please refer to git commit 20d00edf21d2 ("BaseTools/GCC: set > -Wno-unused-but-set-variables only on RELEASE builds", 2016-03-24). >=20 > Looks like gcc-6 has a knob specific to const variables that we should > give the same treatment. Looking at the following pages: >=20 > https://gcc.gnu.org/onlinedocs/gcc-4.9.4/gcc/Warning-Options.html > https://gcc.gnu.org/onlinedocs/gcc-5.4.0/gcc/Warning-Options.html > https://gcc.gnu.org/onlinedocs/gcc-6.4.0/gcc/Warning-Options.html > https://gcc.gnu.org/onlinedocs/gcc-7.2.0/gcc/Warning-Options.html >=20 > it appears that "-Wunused-const-variable" has first appeared in gcc-6. >=20 > Now, that's a problem. In edk2, we currently do not have a specific GCC= 6 > toolchain setting. We have several GCC4x toolchains, and one GCC5 > toolchain. Normally I would have asked you to send a patch, similar to > commit 20d00edf21d2 above, to add "-Wno-unused-const-variable" to the > "oldest" applicable macros in "BaseTools/Conf/tools_def.template", for > example: >=20 > RELEASE_GCC5_IA32_CC_FLAGS > RELEASE_GCC5_X64_CC_FLAGS > RELEASE_GCC5_ARM_CC_FLAGS > RELEASE_GCC5_AARCH64_CC_FLAGS >=20 > But gcc-5 would not recognize this option. I think we might have to > introduce the GCC6 toolchain for this. >=20 After giving the "BaseTools/Conf/tools_def.template" another look, it seems that adding the target would not be the main work here, but testing it on all platforms and verifying that all other flags are still correct and that for all platforms would be, or? I could add a GCC6 one which mainly inherits from GCC5, i.e. in the same way as GCC49 is already the "parent" for GCC5. But I could currently only compile and test OVMF for amd64 and aarch64. I mean from a developer perspective it seems OK to do that and add the target even without full coverage and the adapt the target toolchain - or the platform where appropriate, on a case per case basis, once the new toolchain will see more use there. > Here's what I suggest: please file a TianoCore BZ about this issue, at > . Select "BaseTools" as Package. Feel > free to reference the mailing list thread in the BZ: >=20 > https://lists.01.org/pipermail/edk2-devel/2017-September/014225.html >=20 Done, I got #700 : > Now, addressing that will take a while. Meanwhile, can you please check= > if simply removing the CONST suppresses the issue? (Note, the CONST > should be removed from the "mBusMasterOperationName" variable, *not* > from the pointed-to CHAR8.) >=20 > If that works, can you please submit a patch like this? >=20 No, it does not work, it then does not falls under the "-Wno-unused-but-set-variable" case as its static, if I understand the gc= c Warning Options doc correctly: > -Wunused-but-set-variable: > Warn whenever a local variable is assigned to, ... Whereas: > -Wunused-variable > Warn whenever a local or static variable is unused aside from its= > declaration. This option implies -Wunused-const-variable=3D1 for = C, ... So "no-unused-but-set" allows the local variable case but not the static variable case. >> diff --git a/OvmfPkg/IoMmuDxe/AmdSevIoMmu.c b/OvmfPkg/IoMmuDxe/AmdSevI= oMmu.c >> index bc57de5b572b..45d1c909b5ad 100644 >> --- a/OvmfPkg/IoMmuDxe/AmdSevIoMmu.c >> +++ b/OvmfPkg/IoMmuDxe/AmdSevIoMmu.c >> @@ -47,7 +47,9 @@ STATIC LIST_ENTRY mRecycledMapInfos =3D INITIALIZE_L= IST_HEAD_VARIABLE ( >> // >> // ASCII names for EDKII_IOMMU_OPERATION constants, for debug loggin= g. >> // >> -STATIC CONST CHAR8 * CONST >> +// Not CONST-qualified in order to suppress gcc-6+ warnings. >> +// >> +STATIC CONST CHAR8 * >> mBusMasterOperationName[EdkiiIoMmuOperationMaximum] =3D { >> "Read", >> "Write", >=20 > We could commit this patch for now. Once the BZ you file for BaseTools > is fixed, we can revert this patch. >=20 With your proposed workaround changes my compiler emits: [...] /root/edk2/OvmfPkg/IoMmuDxe/AmdSevIoMmu.c:51:1: error: =E2=80=98mBusMaste= rOperationName=E2=80=99 defined but not used [-Werror=3Dunused-variable] mBusMasterOperationName[EdkiiIoMmuOperationMaximum] =3D { ^~~~~~~~~~~~~~~~~~~~~~~ Another idea, why not also compile out the declaration of the "mBusMasterOperationName" variable when building a RELEASE? Something like: ----8<---- diff --git a/OvmfPkg/IoMmuDxe/AmdSevIoMmu.c b/OvmfPkg/IoMmuDxe/AmdSevIoMm= u.c index bc57de5b57..5af5ce258f 100644 --- a/OvmfPkg/IoMmuDxe/AmdSevIoMmu.c +++ b/OvmfPkg/IoMmuDxe/AmdSevIoMmu.c @@ -47,6 +47,7 @@ STATIC LIST_ENTRY mRecycledMapInfos =3D INITIALIZE_LIST= _HEAD_VARIABLE ( // // ASCII names for EDKII_IOMMU_OPERATION constants, for debug logging. // +#ifdef EFI_DEBUG^M STATIC CONST CHAR8 * CONST mBusMasterOperationName[EdkiiIoMmuOperationMaximum] =3D { "Read", @@ -56,6 +57,7 @@ mBusMasterOperationName[EdkiiIoMmuOperationMaximum] =3D= { "Write64", "CommonBuffer64" }; +#endif^M // // The following structure enables Map() and Unmap() to perform in-plac= e ---->8---- works here, not sure if its up to your coding standards. cheers, Thomas