On Tue, Jan 05, 2016 at 01:06:15PM +0800, Wangnan (F) wrote:
On 2016/1/5 0:55, Will Deacon wrote:I'm still really uneasy about this change. Pairing up the signal delivery
The problem seems to be that we take the debug exception before thePlease have a look at , which I improve test__bp_signal() to
breakpointed instruction has been executed and call perf_bp_event at
that moment, so when we single-step the faulting instruction we actually
step into the SIGIO handler and end up getting stuck.
Your fix doesn't really address this afaict, in that you don't (can't?)
* A longjmp out of a signal handler
* A watchpoint and a breakpoint that fire on the same instruction
* User-controlled single-step from a signal handler that enables a
* Nested signals
check bullet 2 and 4 you mentioned above. Seems my fix is correct.
with the sigreturn to keep track of the debug state is extremely fragile
and I'm not keen on adding this logic there. I also think we need to
track the address that the breakpoint is originally taken on so that we
can only perform the extra sigreturn work if we're returning to the same
instruction. Furthermore, I wouldn't want to do this for signals other
than those generated directly by a breakpoint.
An alternative would be to postpone the signal delivery until after the
stepping has been taken care of, but that's a change in ABI and I worry
we'll break somebody relying on the current behaviour.
What exactly does x86 do? I couldn't figure it out from the code.