Re: [RFC][PATCH 4/7] x86/debug: Move historical SYSENTER junk into exc_debug_kernel()

[peterz]

On Thu, Aug 20, 2020 at 09:43:15AM -0700, Andy Lutomirski wrote:
> I’ve lost track of how many bugs QEMU and KVM have in this space.
> Let’s keep it as a warning, but a bug. But let’s get rid of the
> totally bogus TIF_SINGLESTEP manipulation.

OK, I've shuffled the series around to fix that ordering problem in
patch 4 and added the below patch at the end.

Although I'm not entirely sure it actually leaks a #DB or just wrecks
the state.. *shrug*.

---
Subject: x86/debug: Remove the historical junk
From: Peter Zijlstra <peterz@xxxxxxxxxxxxx>
Date: Thu Aug 20 18:28:37 CEST 2020


Suggested-by: Brian Gerst <brgerst@xxxxxxxxx>
Signed-off-by: Peter Zijlstra (Intel) <peterz@xxxxxxxxxxxxx>
---
 arch/x86/kernel/traps.c |   24 ++++++++++--------------
 1 file changed, 10 insertions(+), 14 deletions(-)

--- a/arch/x86/kernel/traps.c
+++ b/arch/x86/kernel/traps.c
@@ -839,22 +839,18 @@ static __always_inline void exc_debug_ke
 		goto out;
 
 	/*
-	 * Reload dr6, the notifier might have changed it.
+	 * The kernel doesn't use TF single-step outside of:
+	 *
+	 *  - Kprobes, consumed through kprobe_debug_handler()
+	 *  - KGDB, consumed through notify_debug()
+	 *
+	 * So if we get here with DR_STEP set, something is wonky.
+	 *
+	 * A known way to trigger this is through QEMU's GDB stub,
+	 * which leaks #DB into the guest and causes IST recursion.
 	 */
-	dr6 = current->thread.debugreg6;
-
-	if (WARN_ON_ONCE(dr6 & DR_STEP)) {
-		/*
-		 * Historical junk that used to handle SYSENTER single-stepping.
-		 * This should be unreachable now.  If we survive for a while
-		 * without anyone hitting this warning, we'll turn this into
-		 * an oops.
-		 */
-		dr6 &= ~DR_STEP;
-		set_thread_flag(TIF_SINGLESTEP);
+	if (WARN_ON_ONCE(current->thread.debugreg6 & DR_STEP))
 		regs->flags &= ~X86_EFLAGS_TF;
-	}
-
 out:
 	instrumentation_end();
 	idtentry_exit_nmi(regs, irq_state);