on 05/01/2014 06:21 Tim Chase said the following: > Hello, > > When working on the problem that resulted in commit > https://github.com/zfsonlinux/zfs/commit/5d862cb0d9a4b6dcc97a88fa0d5a7a717566e5ab > in > ZoL, I got the impression this problem would affect the rest of the OpenZFS > community. I'm posting this note to ask others to check whether their > implementation should get this patch.
Tim, thanks a lot for sharing! In FreeBSD we have the following commit: https://github.com/freebsd/freebsd/commit/a7cf665d I believe that both changes try to fix the same problem and are approximately equivalent. I think that the FreeBSD change was supposed to eventually end up in illumos, but due to the cumbersome RTI process (from a point of view of someone using and hacking on a different OS, of course) I never got around to it. And apparently nobody else was sufficiently interested. > On another note, and one that possibly impacts SA management as well as other > things, I'm curious whether anyone in the OpenZFS community has experienced > problems as various OS distros have adopted gcc 4.8. I realize that FreeBSD > uses a much older version and I've got no idea about Illumos. I'd like to > refer > the community to > https://github.com/zfsonlinux/zfs/commit/5d862cb0d9a4b6dcc97a88fa0d5a7a717566e5ab > which This is the same URL as for the SA commit. You probably wanted to paste a different one here. > is a rather absurd hack for ZoL to deal with problems caused by gcc 4.8's > "aggressive loop optimization". Briefly, the compiler will over-optimize > loops > in some cases where a typical array index iterator is used. The troublemaker > in > ZFS is the dn_blkptr[] array. See the referenced commit for the particular > bug > it caused. I'm pointing this out as a potential issue as gcc 4.8 becomes more > widespread (and, possibly applicable to clang/llvm, too). -- Andriy Gapon _______________________________________________ developer mailing list [email protected] http://lists.open-zfs.org/mailman/listinfo/developer
