On UP1100, emerge fails in below error: Operating system: alpha-whatever-linux2 This system (linux-alpha+bwx-gcc) is not supported. See file INSTALL for details. Reproducible: Always Steps to Reproduce: 1.On BWX platform 2.emerge --ask dev-libs/openssl Actual Results: >>> Emerging (1 of 24) dev-libs/openssl-1.1.1d-r2::gentoo * Fetching files in the background. * To view fetch progress, run in another terminal: * tail -f /var/log/emerge-fetch.log * openssl-1.1.1d.tar.gz BLAKE2B SHA512 size ;-) ... [ ok ] >>> Unpacking source... >>> Unpacking openssl-1.1.1d.tar.gz to /var/tmp/portage/dev-libs/openssl-1.1.1d-r2/work >>> Source unpacked in /var/tmp/portage/dev-libs/openssl-1.1.1d-r2/work >>> Preparing source in /var/tmp/portage/dev-libs/openssl-1.1.1d-r2/work/openssl-1.1.1d ... * Applying openssl-1.1.0j-parallel_install_fix.patch ... [ ok ] * Applying openssl-1.1.1d-fix-zlib.patch ... [ ok ] * Applying openssl-1.1.1d-fix-potential-memleaks-w-BN_to_ASN1_INTEGER.p [ ok ]. Operating system: alpha-whatever-linux2 This system (linux-alpha+bwx-gcc) is not supported. See file INSTALL for details. * ERROR: dev-libs/openssl-1.1.1d-r2::gentoo failed (prepare phase): * I AM NOT SANE * * Call stack: * ebuild.sh, line 125: Called src_prepare * environment, line 2073: Called die * The specific snippet of code: * ./config --test-sanity || die "I AM NOT SANE"; Expected Results: Use system "linux-alpha-gcc" instead of "linux-alpha+bwx-gcc" OpenSSL-1.1.1d does not support system "linux-alpha+bwx-gcc". However command "config" generates "linux-alpha+bwx-gcc" on BWX platforms. BWX platforms is recognized by string "*[678]" in "cpu model" of /proc/cpuinfo. For example, my UP1100's cpu model is "EV67". Valid "system"s should include by results of command "perl Configure LIST", but "linux-alpha+bwx-gcc" is no longer included. All (valid) "system"s listed in Configurations/10-main.conf.
Created attachment 592852 [details] emerge --info
Created attachment 592854 [details] build.log
Created attachment 592856 [details] perl Configure LIST
The same result was obtained on PC164LX. my PC164LX's "cpu model" is "EV56", so ti's a BWX platform.
Created attachment 593270 [details] emerge --info(PC164LX)
Created attachment 593272 [details] build.log(PC164LX)
An Alpha user! I'll check on my UP1500 and see what's going on.
Thank you for your cooperation. I think it is important whether "cpu model" of /proc/cpuinfo matches "*[678]" or not for this to happen. Also, I got the same result on PC164SX. My PC164SX's "cpu model" is "PCA56", so it's a BWX platform.
Created attachment 593342 [details] emerge --info(PC164SX)
Created attachment 593344 [details] build.log(PC164SX)
I think that "system" named "linux-alpha+bwx-gcc" has been abolished in OpenSSL-1.1.1d, so both OpenSSL-1.1.1d's "config" script and "gentoo.config-1.0.2" script provided by Gentoo should be changed so that "linux-alpha+bwx-gcc" is not generated. This issue occurs in "./config --test-sanity" that is executed at the end of emerge's "preparing source" phase. I don't know why this is necessary in this phase, but "./Configure" in the "configuring source" phase is the real configuration for construction. The "system" given as an argument of "./Configure" is the "system" determined by gentoo.config-1.0.2 provided by Gentoo, and is determined by CHOST. If CHOST is "alpha-unknown-linux-gnu" as it is in Gentoo distribution's make.conf, "system" will be "alpha-linux-gcc" and there is no problem, but If CHOST is changed to "alphaev56-unknown-linux" or "alphaev6-unknown-linux-gnu", etc., "system" becomes "alpha-linux+bwx-gcc", which seems to cause problems in the "configuring source" phase. I applied the attached "openssl-1.1.1d-alpha.patch" (ie put it in /etc/portage/patches/dev-libs/openssl-1.1.1d-r2/) and when I emerged on my UP1100 , it was able to install without problems. I also created "gentoo.config-1.0.2-alpha.patch" as a proposed change to gentoo.config-1.0.2. This has not been tried by emerge. Or should we restore the "system" named "linux-alpha+bwx-gcc"? Using this "system" will improve performance, but I don't think RC4_CHAR(RC4_INT) has a wide impact. (And is there any reason to focus on RC4 in the future?) The difference between system "linux-alpha-gcc" and "linux-alpha+bwx-gcc" in "Configure" of OpenSSL-1.0.2t is that "RC4_CHAR" is added to the latter. In "Configurations/10-main.conf" of OpenSSL-1.1.1d, "RC4_CHAR" is disabled in "bn_ops" of system "linux-alpha-gcc", however "RC4_CHAR" is enabled in the inheritance source system "linux-generic64", so I think it is equivalent to system "linux-alpha-gcc" of OpenSSL-1.0.2t. OpenSSL-1.1.1d's "CHANGES" doesn't mention "linux-alpha+bwx-gcc" abolished, so I don't know why upstream has abolished "linux-alpha+bwx-gcc".
Created attachment 593748 [details, diff] openssl-1.1.1d-alpha.patch
Created attachment 593750 [details, diff] gentoo.config-1.0.2-alpha.patch
Not surprisingly, seeing the same issue on my EV6 based DS20.
Thank you for your cooperation. As a note, even if the architecture is EV6 or later, the following "cpu model"s are not recognized as BWX platform by "config" script, these do not match "*[678]".: "EV68AL"(21264B) "EV68CB"(21264C) "EV68CX"(21264D) "EV79"(21364A) "EV69"(what's this?) (from "cpu_names" array in arch/alpha/kernel/setup.c of kernel source)
(In reply to Yoshihiko Iwama from comment #11) > I think that "system" named "linux-alpha+bwx-gcc" has been abolished in > OpenSSL-1.1.1d, so both OpenSSL-1.1.1d's "config" script and > "gentoo.config-1.0.2" script provided by Gentoo should be changed so that > "linux-alpha+bwx-gcc" is not generated. > > This issue occurs in "./config --test-sanity" that is executed at the end of > emerge's "preparing source" phase. I don't know why this is necessary in > this phase, but "./Configure" in the "configuring source" phase is the real > configuration for construction. The "system" given as an argument of > "./Configure" is the "system" determined by gentoo.config-1.0.2 provided by > Gentoo, and is determined by CHOST. If CHOST is "alpha-unknown-linux-gnu" as > it is in Gentoo distribution's make.conf, "system" will be "alpha-linux-gcc" > and there is no problem, but If CHOST is changed to > "alphaev56-unknown-linux" or "alphaev6-unknown-linux-gnu", etc., "system" > becomes "alpha-linux+bwx-gcc", which seems to cause problems in the > "configuring source" phase. > I applied the attached "openssl-1.1.1d-alpha.patch" (ie put it in > /etc/portage/patches/dev-libs/openssl-1.1.1d-r2/) and when I emerged on my > UP1100 , it was able to install without problems. > I also created "gentoo.config-1.0.2-alpha.patch" as a proposed change to > gentoo.config-1.0.2. This has not been tried by emerge. > > Or should we restore the "system" named "linux-alpha+bwx-gcc"? Using this > "system" will improve performance, but I don't think RC4_CHAR(RC4_INT) has a > wide impact. (And is there any reason to focus on RC4 in the future?) The > difference between system "linux-alpha-gcc" and "linux-alpha+bwx-gcc" in > "Configure" of OpenSSL-1.0.2t is that "RC4_CHAR" is added to the latter. > In "Configurations/10-main.conf" of OpenSSL-1.1.1d, "RC4_CHAR" is disabled > in "bn_ops" of system "linux-alpha-gcc", however "RC4_CHAR" is enabled in > the inheritance source system "linux-generic64", so I think it is equivalent > to system "linux-alpha-gcc" of OpenSSL-1.0.2t. > > OpenSSL-1.1.1d's "CHANGES" doesn't mention "linux-alpha+bwx-gcc" abolished, > so I don't know why upstream has abolished "linux-alpha+bwx-gcc". Thank you for the analysis. I think you are correct that linux-alpha+bwx-gcc has been abolished upstream. It looks like it was removed in commit 7ead0c89185c46378e3ed85c0012d083f4b3039b Author: Andy Polyakov <appro@openssl.org> Date: Mon Mar 16 22:33:36 2015 +0100 Configure: fold related configurations more aggressively and clean-up. Reviewed-by: Richard Levitte <levitte@openssl.org> and like you said that configuration defined only the RC4_CHAR in addition to linux-alpha-gcc. Perhaps we should readd linux-alpha+bwx-gcc upstream (it's unclear whether it was removed intentionally, since the string "linux-alpha+bwx" still appears in the 'config' file). I'll try to confirm what you said -- that RC4_CHAR is implied by linux-generic64.
Confirming that the patch from comment #12 allows me to successfully build.
Thank you for your cooperation. Thank you for investigating upstream modifications. I also confirmed that "linux-alpha-gcc" was removed there. The patch in comment #12 for the 'config' script simply enforces "linux-alpha-gcc", so if you re-add "linux-alpha+bwx-gcc", another patch Is required. Because BWX is implemented in all processor models after EV56. If the current condition is "* [678]", as written in comment #15, five models are not enough. Also, for strictness, I think that the settings of "__CNF_CFLAGS" and "__CNF_CXXFLAGS" should be changed. This definition has no meaning when using emerge.
Created attachment 614474 [details, diff] patch Sorry for the delay. This is what I plan to commit, and to send upstream.
The bug has been closed via the following commit(s): https://gitweb.gentoo.org/repo/gentoo.git/commit/?id=e3dec9bf2da04ace6b2e999ff779d117beb65e6e commit e3dec9bf2da04ace6b2e999ff779d117beb65e6e Author: Matt Turner <mattst88@gentoo.org> AuthorDate: 2020-02-25 22:37:15 +0000 Commit: Matt Turner <mattst88@gentoo.org> CommitDate: 2020-02-25 22:40:42 +0000 dev-libs/openssl: Fix the build on alpha Closes: https://bugs.gentoo.org/697840 Signed-off-by: Matt Turner <mattst88@gentoo.org> ...sl-1.1.1d-config-Drop-linux-alpha-gcc-bwx.patch | 42 ++++++++++++++++++++++ dev-libs/openssl/openssl-1.1.1d-r3.ebuild | 1 + 2 files changed, 43 insertions(+)
Thank you very much. I understand that, as a result of benchmarking on EV7, the use of RC4_CHAR is not superior in performance, so linux-alpha+bwx-gcc was not restored and the Configure script was changed. I have confirmed that modified emerge can be successfully run on my UP1100, that's cpu model is 'EV67'. I look forward to long-term support for Linux/Alpha in the upstream as a result of this our action.
