DTRACE build failure (/usr/obj/usr/src/tmp/usr/bin/ld: cannot
deischen at freebsd.org
Wed Aug 2 22:02:49 UTC 2006
On Wed, 2 Aug 2006, Kip Macy wrote:
> I described it previously. Signals won't always be sent to the right
> process, i.e. running a process under GDB the debuggee will sometimes
> get the SIGTRAP instead of GDB. When restarting mysql, mysql-server
> becomes unkillable. Removing KSE (bike_sched et al.) makes the problem
> go away.
That sounds more like a bug with gdb/libthread_db. It also doesn't
sound like a major bug either, like "signals are broken with KSE"
makes it sound.
> It is trivial to hit on sun4v. However, I've seen it (or a problem with
> similar symptoms) on x86 using NetApp's development simulator which
> makes very heavy use of signals. I believe NetApp ended up giving up on
> using FreeBSD for running their simulator as this problem made FreeBSD a
> non-starter for them. I no longer work for NetApp so I don't have access
> to the sources for the simulator any more, so I don't know if bike_sched
> fixes that problem for them or not.
> On Wed, 2 Aug 2006, Daniel Eischen wrote:
>> On Wed, 2 Aug 2006, Kip Macy wrote:
>>> sun4u is a dead architecture. Signals are broken with KSE on a
>>> tightly coupled SMP so sun4v doesn't even have KSE in the tree.
>> On every platform or on sparc64? I am unaware of any signal
>> problems in i386.
>>> Unless someone steps up to fix KSE "will not work" is more correct.
>> Please clarify.
More information about the freebsd-current