[opensuse-kernel] Why do we still carry patches.suse/stack-unwind?
Hi! The DWARF2-EH unwinder was dropped from mainline years ago . Commit d1526e2c (Remove stack unwinder for now) Why does the openSUSE kernel still carry this unwinder? If all problems have been resolved please push it back to mainline. -- Thanks, //richard -- To unsubscribe, e-mail: opensuse-kernel+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse-kernel+owner@opensuse.org
On 11/9/13, 10:01 AM, Richard Weinberger wrote:
Hi!
The DWARF2-EH unwinder was dropped from mainline years ago . Commit d1526e2c (Remove stack unwinder for now)
Why does the openSUSE kernel still carry this unwinder?
We carry it in our kernels because the benefits far outweigh the costs in maintaining it ourselves. The in-kernel unwinder uses some heuristics to determine what is or isn't a kernel address and products stack traces of questionable veracity. The DWARF unwinder products much more accurate traces and it's not too invasive to integrate into our kernels.
If all problems have been resolved please push it back to mainline.
The patches, mostly in the vein of annotations, have been flowing upstream gradually. The reasons for it not being included last time were due to Linus objecting to the unwinder accessing stack data and assuming it was trusted. I'm not sure if those objections have been addressed yet. -Jeff -- Jeff Mahoney SUSE Labs
Am 09.11.2013 22:28, schrieb Jeff Mahoney:
On 11/9/13, 10:01 AM, Richard Weinberger wrote:
Hi!
The DWARF2-EH unwinder was dropped from mainline years ago . Commit d1526e2c (Remove stack unwinder for now)
Why does the openSUSE kernel still carry this unwinder?
We carry it in our kernels because the benefits far outweigh the costs in maintaining it ourselves. The in-kernel unwinder uses some heuristics to determine what is or isn't a kernel address and products stack traces of questionable veracity. The DWARF unwinder products much more accurate traces and it's not too invasive to integrate into our kernels.
In which situation produces the in-kernel unwinder incorrect traces? Thanks, //richard -- To unsubscribe, e-mail: opensuse-kernel+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse-kernel+owner@opensuse.org
On 09.11.13 at 22:28, Jeff Mahoney <jeffm@suse.com> wrote: On 11/9/13, 10:01 AM, Richard Weinberger wrote: If all problems have been resolved please push it back to mainline.
The patches, mostly in the vein of annotations, have been flowing upstream gradually. The reasons for it not being included last time were due to Linus objecting to the unwinder accessing stack data and assuming it was trusted. I'm not sure if those objections have been addressed yet.
The situation didn't change, since - afaik - no-one made another try. The original authors (Andi Kleen and me) were told very clearly back then by Linus that he won't accept another attempt by them to get these changes merged, no matter what (i.e. in particular independent on the fact we didn't have any problem with it in our kernels - any issues I recall we had were the result of broken manually generated unwind annotations [i.e. to assembly code, iow where the compiler wasn't to blame, which was one of Linus' main concerns apart from the question of stack accesses that Jeff mentioned]). So someone else would need to step up and perhaps first thing sort out the political aspects. Jan -- To unsubscribe, e-mail: opensuse-kernel+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse-kernel+owner@opensuse.org
participants (4)
-
Jan Beulich -
Jeff Mahoney -
Richard Weinberger -
Richard Weinberger