hw_breakpoint: Fix cpu check in task_bp_pinned(cpu)
authorOleg Nesterov <oleg@redhat.com>
Thu, 20 Jun 2013 15:50:06 +0000 (17:50 +0200)
committerIngo Molnar <mingo@kernel.org>
Thu, 20 Jun 2013 15:57:00 +0000 (17:57 +0200)
commit8b4d801b2b123b6c09742f861fe44a8527b84d47
tree12f820394b663a494bc70d63307fad8589a497ef
parent003002e04ed38618fc37b92ba128f5ca79d39f4f
hw_breakpoint: Fix cpu check in task_bp_pinned(cpu)

trinity fuzzer triggered WARN_ONCE("Can't find any breakpoint
slot") in arch_install_hw_breakpoint() but the problem is not
arch-specific.

The problem is, task_bp_pinned(cpu) checks "cpu == iter->cpu"
but this doesn't account the "all cpus" events with iter->cpu <
0.

This means that, say, register_user_hw_breakpoint(tsk) can
happily create the arbitrary number > HBP_NUM of breakpoints
which can not be activated. toggle_bp_task_slot() is equally
wrong by the same reason and nr_task_bp_pinned[] can have
negative entries.

Simple test:

# perl -e 'sleep 1 while 1' &
# perf record -e mem:0x10,mem:0x10,mem:0x10,mem:0x10,mem:0x10 -p `pidof perl`

Before this patch this triggers the same problem/WARN_ON(),
after the patch it correctly fails with -ENOSPC.

Reported-by: Vince Weaver <vincent.weaver@maine.edu>
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: <stable@vger.kernel.org>
Link: http://lkml.kernel.org/r/20130620155006.GA6324@redhat.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
kernel/events/hw_breakpoint.c