Re: Firefox52 broken on NetBSD9.3 / pkgsrc-2022Q4
David Brownlee <[email protected]> Sat, 18 Mar 2023 22:13:10 +0000
| Newsgroups | gmane.os.netbsd.ports.sparc64 |
|---|---|
| Message-ID | <CAGN_6pb5wMCXHb9JHu7bOvcUTTYcMEnJJz8m=cVjWo8urU7XZA@mail.gmail.com> |
On Sat, 18 Mar 2023 at 22:09, Connor McLaughlan <[email protected]> wrote: > > I found the problem! > > It seems a certain thread of firefox52 is crashing without notice > leading to the error and I finally found the cause after replacing > almost the whole dependency tree it occurred to me that neither the > dependencies nor the firefox pkgversion might be the cause. > Searching through the pkgsrc-2021Q4 release notes i found that > hardening was elevated and more hardening was enabled exactly in this > release where the problems seems to have started. > > Release notes tell: > " - PKGSRC_USE_MKPIE was enabled on some architectures. Packages that > provide static libraries to other packages should be rebuilt with > MKPIE enabled. Some packages may experience runtime errors on > platforms like i386 - usually this is due to unsafe assembly code, > which should be disabled when MKPIE is enabled. > - PKGSRC_USE_SSP was bumped from "yes" to "strong". This is not > expected to have a noticable effect." > > So i searched for how to disable those when experiencing runtime > errors and found this webpage: > https://www.netbsd.org/docs/pkgsrc/hardening.html#hardening.caveats.pie > > I added all workarounds as told to mozilla-common.mk before .include > "../../mk/bsd.prefs.mk": > MKPIE_SUPPORTED= no > FORTIFY_SUPPORTED= no > RELRO_SUPPORTED= no > SSP_SUPPORTED= no > > And after recompiling firefox52 it is working again! > I possibly overshot here..possibly only SSP and/or MKPIE need to be disabled. > > So how to proceed from here? It seems not only firefox52 is affected > on sparc64, but also thunderbird52 and possibly other packages... Excellent catch. We can definitely disable the appropriate options on sparc64 (with a nice comment in the Makefiles for the future curious :) Would you be willing to run another couple of tests to find exactly which need to be disabled? Thanks! David