x86/debug: Fix DR_STEP vs ptrace_get_debugreg(6)

Commit d53d9bc0 ("x86/debug: Change thread.debugreg6 to
thread.virtual_dr6") changed the semantics of the variable from random
collection of bits, to exactly only those bits that ptrace() needs.

Unfortunately this lost DR_STEP for PTRACE_{BLOCK,SINGLE}STEP.

Furthermore, it turns out that userspace expects DR_STEP to be
unconditionally available, even for manual TF usage outside of
PTRACE_{BLOCK,SINGLE}_STEP.

Fixes: d53d9bc0 ("x86/debug: Change thread.debugreg6 to thread.virtual_dr6")
Reported-by: Kyle Huey <me@kylehuey.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Tested-by: Kyle Huey <me@kylehuey.com> 
Link: https://lore.kernel.org/r/20201027183330.GM2628@hirez.programming.kicks-ass.net

arch/x86/kernel/traps.c

@@ -937,10 +937,13 @@ static __always_inline void exc_debug_user(struct pt_regs *regs,
 	instrumentation_begin();
 
 	/*
-	 * Clear the virtual DR6 value, ptrace() routines will set bits here
-	 * for things it wants signals for.
+	 * Start the virtual/ptrace DR6 value with just the DR_STEP mask
+	 * of the real DR6. ptrace_triggered() will set the DR_TRAPn bits.
+	 *
+	 * Userspace expects DR_STEP to be visible in ptrace_get_debugreg(6)
+	 * even if it is not the result of PTRACE_SINGLESTEP.
 	 */
-	current->thread.virtual_dr6 = 0;
+	current->thread.virtual_dr6 = (dr6 & DR_STEP);
 
 	/*
 	 * The SDM says "The processor clears the BTF flag when it