[valgrind] [Bug 398883] valgrind incorrectly assumes ABI on PowerPC based on endianness
https://bugs.kde.org/show_bug.cgi?id=398883 --- Comment #10 from A. Wilcox (awilfox) --- Indeed, the only technical issue with this patch is that it is outdated. We have an updated version. I will upload the new version to this bug some time this week after testing it against the latest release and main branch. -- You are receiving this mail because: You are watching all bug changes.
[valgrind] [Bug 398883] valgrind incorrectly assumes ABI on PowerPC based on endianness
https://bugs.kde.org/show_bug.cgi?id=398883 --- Comment #9 from Carl Love --- I looked thru the patch. Nothing jumps out at me as being an issue. But, that doesn't guarantee anything given the assembly and all. I did try to apply the patch to test it but it doesn't apply. So, I would say, update the patch and lets test it on the supported BE and LE systems to make sure the patch doesn't break anything. Test it on the non-standard systems where you want to use it. As long as everything works fine, I wouldn't object to it. -- You are receiving this mail because: You are watching all bug changes.
[valgrind] [Bug 398883] valgrind incorrectly assumes ABI on PowerPC based on endianness
https://bugs.kde.org/show_bug.cgi?id=398883 --- Comment #8 from Sam James --- (In reply to Carl Love from comment #4) > > Given that IBM does not recommend using ELFv1 on Little endian and ELFv2 on > Big endian, that is not to say it can't be done. The proposed patch is not > necessarily invalid but IBM will not support these but will also not stand > in the way of someone else trying it but they are on their own for support. In that case, is anyone aware of any technical issue with the patch? -- You are receiving this mail because: You are watching all bug changes.
[valgrind] [Bug 398883] valgrind incorrectly assumes ABI on PowerPC based on endianness
https://bugs.kde.org/show_bug.cgi?id=398883 Sam James changed: What|Removed |Added CC||s...@gentoo.org -- You are receiving this mail because: You are watching all bug changes.
[valgrind] [Bug 398883] valgrind incorrectly assumes ABI on PowerPC based on endianness
https://bugs.kde.org/show_bug.cgi?id=398883 --- Comment #7 from Brandon Bergren --- Ahh, I think we were using different meanings for the word "supported". I was speaking in terms of "what can the hardware do, is specified in the ABI, and is currently in use in the wild", not "what you can actually get official support for". -- You are receiving this mail because: You are watching all bug changes.
[valgrind] [Bug 398883] valgrind incorrectly assumes ABI on PowerPC based on endianness
https://bugs.kde.org/show_bug.cgi?id=398883 Brandon Bergren changed: What|Removed |Added CC||k...@bdragon.rtk0.net --- Comment #6 from Brandon Bergren --- While I agree that using ELFv1 on LE is a completely unsupported configuration and need not be supported, the OpenPOWER ELF V2 ABI is fully defined and intended for use in both little- and big-endian environments. Due to the improved toolchain support (specifically, lld's dropping of ELFv1 support entirely, and ELFv2 being a lot less "weird" of an ABI meaning it tends to have less esoteric issues and can more easily leverage mainstream optimizations) I see the future of BE on POWER to be ELFv2 as well. And yes, FreeBSD is moving towards ELFv2 at a rapid pace. (I personally am running ELFv2 userlands exclusively in FreeBSD on my POWER9 machines.) -- You are receiving this mail because: You are watching all bug changes.
[valgrind] [Bug 398883] valgrind incorrectly assumes ABI on PowerPC based on endianness
https://bugs.kde.org/show_bug.cgi?id=398883 --- Comment #5 from A. Wilcox (awilfox) --- I noted that it was specifically older versions of OpenSuSE. Specifically, the very first release. I believe it is long deprecated; it still, however, does exist. > Similarly, with the musl libc. From my discussion with the GCC team, they > said that musl has been known to play around with ELFv2 ABI on Big Endian but > they are off on their own in this regard as it isn't supported, recommended > or tested by IBM. I fear "play around" may be a wee dismissive. It's a serious port and it has serious consumers, including Adélie Linux and Void Linux. Additionally, FreeBSD is planning to move to ELFv2 on Big Endian, which means eventually FreeBSD will need to undefine VG_PLAT_USES_PPCTOC as well. :) -- You are receiving this mail because: You are watching all bug changes.
[valgrind] [Bug 398883] valgrind incorrectly assumes ABI on PowerPC based on endianness
https://bugs.kde.org/show_bug.cgi?id=398883 --- Comment #4 from Carl Love --- IBM only supports ELFv1 ABI on Big endian machines and ELFv2 ABI on Little endian machines. I checked a little on the statements about the ELF usages. >From what I was told OpenSuSE uses ELFv2 ABI on little endian. So, I would be curious where to get the info that OpenSuSE is using ELFv1 ABI on Little Endian. Similarly, with the musl libc. From my discussion with the GCC team, they said that musl has been known to play around with ELFv2 ABI on Big Endian but they are off on their own in this regard as it isn't supported, recommended or tested by IBM. The key thing is the OS, library, Valgrind etc. all need to either use ELFv1 ABI or ELFv2 ABI. You can't mix these. Given that IBM does not recommend using ELFv1 on Little endian and ELFv2 on Big endian, that is not to say it can't be done. The proposed patch is not necessarily invalid but IBM will not support these but will also not stand in the way of someone else trying it but they are on their own for support. -- You are receiving this mail because: You are watching all bug changes.
[valgrind] [Bug 398883] valgrind incorrectly assumes ABI on PowerPC based on endianness
https://bugs.kde.org/show_bug.cgi?id=398883 Julian Seward changed: What|Removed |Added CC||c...@us.ibm.com --- Comment #3 from Julian Seward --- Carl, do you have any opinion on this? -- You are receiving this mail because: You are watching all bug changes.
[valgrind] [Bug 398883] valgrind incorrectly assumes ABI on PowerPC based on endianness
https://bugs.kde.org/show_bug.cgi?id=398883 A. Wilcox (awilfox) changed: What|Removed |Added Attachment #115278|application/mbox|text/plain mime type|| Attachment #115278|0 |1 is patch|| -- You are receiving this mail because: You are watching all bug changes.
[valgrind] [Bug 398883] valgrind incorrectly assumes ABI on PowerPC based on endianness
https://bugs.kde.org/show_bug.cgi?id=398883 --- Comment #2 from A. Wilcox (awilfox) --- Created attachment 115278 --> https://bugs.kde.org/attachment.cgi?id=115278=edit Ensure ELFv2 is supported on PPC64 This patch has been tested on PPC64 BE, both ELFv1 (Debian) and ELFv2 (Adélie). I don't have any LE hardware to test on, but there should be no functional change. I've tested memcheck and callgrind on both ELF ABIs on a POWER9 host running big-endian and found no issues. -- You are receiving this mail because: You are watching all bug changes.
[valgrind] [Bug 398883] valgrind incorrectly assumes ABI on PowerPC based on endianness
https://bugs.kde.org/show_bug.cgi?id=398883 --- Comment #1 from A. Wilcox (awilfox) --- Okay, it doesn't have to be so hard. There is already VG_PLAT_USES_PPCTOC; so this change just needs to actually use that define instead of just endianness to determine what ABI to use. -- You are receiving this mail because: You are watching all bug changes.
