[PATCH 1/3] powerpc/watchpoints: Disable preemption in thread_change_pc()

Benjamin Gray bgray at linux.ibm.com
Tue Aug 29 16:34:55 AEST 2023

thread_change_pc() uses CPU local data, so must be protected from
swapping CPUs while it is reading the breakpoint struct.

The error is more noticeable after 1e60f3564bad ("powerpc/watchpoints:
Track perf single step directly on the breakpoint"), which added an
unconditional __this_cpu_read() call in thread_change_pc(). However the
existing __this_cpu_read() that runs if a breakpoint does need to be
re-inserted has the same issue.

Signed-off-by: Benjamin Gray <bgray at linux.ibm.com>


There's probably a more idiomatic way to express this. We technically
don't need to disable preemption for the entire function: we should only
need to disable preemption within each loop iteration while handling the
pointer we are working with. Each iteration itself is independent.
 arch/powerpc/kernel/hw_breakpoint.c | 7 ++++++-
 1 file changed, 6 insertions(+), 1 deletion(-)

diff --git a/arch/powerpc/kernel/hw_breakpoint.c b/arch/powerpc/kernel/hw_breakpoint.c
index b8513dc3e53a..2854376870cf 100644
--- a/arch/powerpc/kernel/hw_breakpoint.c
+++ b/arch/powerpc/kernel/hw_breakpoint.c
@@ -230,13 +230,15 @@ void thread_change_pc(struct task_struct *tsk, struct pt_regs *regs)
 	struct arch_hw_breakpoint *info;
 	int i;
+	preempt_disable();
 	for (i = 0; i < nr_wp_slots(); i++) {
 		struct perf_event *bp = __this_cpu_read(bp_per_reg[i]);
 		if (unlikely(bp && counter_arch_bp(bp)->perf_single_step))
 			goto reset;
-	return;
+	goto out;
 	regs_set_return_msr(regs, regs->msr & ~MSR_SE);
@@ -245,6 +247,9 @@ void thread_change_pc(struct task_struct *tsk, struct pt_regs *regs)
 		__set_breakpoint(i, info);
 		info->perf_single_step = false;
+	preempt_enable();
 static bool is_larx_stcx_instr(int type)

More information about the Linuxppc-dev mailing list