Ananth, On Sat, 28 Nov 2009 09:09:59 +0100, Ananth N Mavinakayanahalli wrote: > I reran tests on a POWER6 machine with the single-step fix Oleg had > cited on lkml. The diff between the two did not change: ... > -PASS: gdb.base/sigbpt.exp: stepi bp at segv; stepi out of handler > +PASS: gdb.base/sigbpt.exp: stepi bp at segv; stepi out of handler (at > breakpoint)
We do not check if GDB is happy with utrace or not. We check if there are any ptrace changes visible to the userland. Any such change should be either a known kernel ptrace bugfix visible on ptrace-testsuite or it is a change=regression by the utrace code. There is no perfect tool to check that but GDB is heuristically sensitive for such changes. If there is a stable change of testcase output between kernels it must be either a known kernel ptrace bugfix visible on ptrace-testsuite or it is a change=regression by the utrace code. (The third possibility is for example the two kernels were more different than just the utrace change as possibly in CAI Qian's testing - or that some subtle timing has changed producing different results in a stable way despite the result is in fact unstable.) One should first check gdb.log diff at this "stepi out of handler" place and if the reason is not obvious there then to manually investigate/debug it. Thanks, Jan
