On Tue, 23 May 2023 at 14:44, Tobias Nygren <tnn%netbsd.org@localhost> wrote: > > On Wed, 10 May 2023 10:18:01 +0100 > Chavdar Ivanov <ci4ic4%gmail.com@localhost> wrote: > > > Any recent ideas about this? Frankly, this version is next to unusable > > until this is resolved; perhaps downgrade firefox to the last known > > working version until a solution is found? > > USE_GLX_TLS should maybe be backed out in xsrc until someone has > time to investigate it. I unfortunately do not have time to look into > this in the near future. It might be worthwhile to try to merge this > fix (please try and report back if it helps): > > https://github.com/NetBSD/pkgsrc/commit/0e33e24256c28aed530eb25e07d1c2f23bdf1e12 It looks like most of that commit was already in xsrc (as of netbsd-10), but there were two extra parts for glxcurrent.c. One was a cosmetic adjustment for defining __THREAD_INITIAL_EXEC void *__glX_tls_Context The other made some changes inside a GLX_USE_TLS and changed it to USE_ELF_TLS - but neither seem to be actually defined in-tree? Having said that, I applied them to a netbsd-10 tree and rebooted into it, and my firefox has been running for fifteen minutes without the LD_PRELOAD=/usr/X11R7/lib/libEGL.so workaround previously needed to stop a crash within a few seconds In case anyone else would like to try to reproduce, the patch is below, and a netbsd-10 build from the last couple of days with the patch applies is syncing to https://sync.absd.org/amd64/10-with-glxcurrent-patch/ (the README.md will be the last thing to sync, so if it its present, the sync is done :) (I'm parking my proposed addition of LD_PRELOAD=/usr/X11R7/lib/libEGL.so to the firefox startup pending someone confirming if they either see the same thing with the patch, or that my system is Just In A Strange State) Thanks David
Attachment:
diff
Description: Binary data