Re: BUG: debug_exception_enter() disables preemption and may call sleeping functions on aarch64 with RT

[Luis Claudio R. Goncalves]

On Tue, Feb 11, 2025 at 11:34:26AM -0300, Luis Claudio R. Goncalves wrote:
> On Mon, Feb 10, 2025 at 12:49:45PM +0000, Mark Rutland wrote:
> > On Fri, Feb 07, 2025 at 11:22:57AM -0300, Luis Claudio R. Goncalves wrote:
...
> > I don't have an immediate suggestion; I'll need to go think about this
> > for a bit. Unfortunatealy, there are several nested cans of worms here.
> > :/
> > 
> > In theory, we can go split out the EL0 "debug exceptions" into separate
> > handlers, and wouldn't generally need to disable preemption for things
> > like BRK or single-step.
> 
> If this is an acceptable workaround, until we have the real solution,
> I can work on that :)
> 
> Luis

I tested the prototype below and it survived 6h of ssdd and the ptrace LTP
tests running simultaneously, in a tight loop. Would something along these
lines be an acceptable workaround?


diff --git a/arch/arm64/mm/fault.c b/arch/arm64/mm/fault.c
index 8b281cf308b30..eb3b54710024f 100644
--- a/arch/arm64/mm/fault.c
+++ b/arch/arm64/mm/fault.c
@@ -933,18 +933,20 @@ void __init hook_debug_fault_code(int nr,
  * accidentally schedule in exception context and it will force a warning
  * if we somehow manage to schedule by accident.
  */
-static void debug_exception_enter(struct pt_regs *regs)
+static void debug_exception_enter(struct pt_regs *regs, int touch_preemption)
 {
-	preempt_disable();
+	if (touch_preemption)
+		preempt_disable();
 
 	/* This code is a bit fragile.  Test it. */
 	RCU_LOCKDEP_WARN(!rcu_is_watching(), "exception_enter didn't work");
 }
 NOKPROBE_SYMBOL(debug_exception_enter);
 
-static void debug_exception_exit(struct pt_regs *regs)
+static void debug_exception_exit(struct pt_regs *regs, int touch_preemption)
 {
-	preempt_enable_no_resched();
+	if (touch_preemption)
+		preempt_enable_no_resched();
 }
 NOKPROBE_SYMBOL(debug_exception_exit);
 
@@ -953,8 +955,14 @@ void do_debug_exception(unsigned long addr_if_watchpoint, unsigned long esr,
 {
 	const struct fault_info *inf = esr_to_debug_fault_info(esr);
 	unsigned long pc = instruction_pointer(regs);
+	unsigned long req = ESR_ELx_EC(esr);
+	int touch_preemption;
 
-	debug_exception_enter(regs);
+	touch_preemption = !(IS_ENABLED(CONFIG_PREEMPT_RT) &&
+		(req == ESR_ELx_EC_SOFTSTP_LOW || req == ESR_ELx_EC_BRK64
+		 || req == ESR_ELx_EC_BKPT32 || req == ESR_ELx_EC_SOFTSTP_CUR));
+
+	debug_exception_enter(regs, touch_preemption);
 
 	if (user_mode(regs) && !is_ttbr0_addr(pc))
 		arm64_apply_bp_hardening();
@@ -963,7 +971,7 @@ void do_debug_exception(unsigned long addr_if_watchpoint, unsigned long esr,
 		arm64_notify_die(inf->name, regs, inf->sig, inf->code, pc, esr);
 	}
 
-	debug_exception_exit(regs);
+	debug_exception_exit(regs, touch_preemption);
 }
 NOKPROBE_SYMBOL(do_debug_exception);