Re: [PATCH v2 RESEND 1/2] ARM: spectre-v2: Fix potential missing mitigations

[Xie Yuanbin <xieyuanbin1@huawei.com>]

On Tue, 28 Oct 2025 17:20:05 +0100, Sebastian Andrzej Siewior wrote:
> If I apply both patches (of yours) then it sends a
> signal with disabled interrupts which breaks my PREEMPT_RT case.

I am not familiar with PREEMPT_RT yet and do not know that signals cannot
be sent with disabled interrupts and PREEMPT_RT=y.
I apologize for this.

On Tue, 28 Oct 2025 19:20:52 +0100, Sebastian Andrzej Siewior wrote:
> !LPAE does do_bad_area() -> __do_user_fault() and does not trigger the
> warning in harden_branch_predictor() because the interrupts are off.
> On PREEMPT_RT this leads to an error due to accessing spinlock_t from
> force_sig_fault() with disabled interrupts.

This seems to be a more serious bug, and may require another patch to
fix it. Not only !LPAE is affected, but LAPE=y is also affected:
do_translation_fault() -> do_bad_area() -> __do_user_fault()
This code path seems very easy to trigger.

> I guess the requirement is to invoke harden_branch_predictor() on the
> same CPU that triggered the page_fault, right? Couldn't we then move
> harden_branch_predictor() a little bit earlier, invoke it in the >=
> TASK_SIZE case and then enable interrupts if they were enabled?
>
> That would make me happy ;)

This seems to only fix the warning in harden_branch_predictor, but cannot
fix the issue of sending signals with disabled interrupts mentioned above.

What about adding:

diff --git a/arch/arm/mm/fault.c b/arch/arm/mm/fault.c
index 09dde89a88ed..b9c9c80db109 100644
--- a/arch/arm/mm/fault.c
+++ b/arch/arm/mm/fault.c
@@ -182,6 +182,12 @@ __do_user_fault(unsigned long addr, unsigned int fsr, unsigned int sig,
 		int code, struct pt_regs *regs)
 {
 	struct task_struct *tsk = current;
+	const bool save_irqs_disabled = irqs_disabled();
+
+	if (save_irqs_disabled) {
+		preempt_disable();
+		local_irq_enable();
+	}

 	if (addr > TASK_SIZE)
 		harden_branch_predictor();
@@ -207,6 +213,11 @@ __do_user_fault(unsigned long addr, unsigned int fsr, unsigned int sig,
 	tsk->thread.error_code = fsr;
 	tsk->thread.trap_no = 14;
 	force_sig_fault(sig, code, (void __user *)addr);
+
+	if (save_irqs_disabled) {
+		local_irq_disable();
+		preempt_enable_no_resched();
+	}
 }

and the modification of patch 1 is still retained.

Xie Yuanbin