perf/x86: Implement immediate enforcement of /sys/devices/cpu/rdpmc v…

…alue of 0

When you successfully write 0 to /sys/devices/cpu/rdpmc, the RDPMC
instruction should be disabled unconditionally and immediately (after you
close the SYSFS file) by the documentation.

Instead, in the current implementation the PMU must be reloaded which
happens only eventually some time in the future. Only after that the RDPMC
instruction becomes disabled (on ring 3) on the respective core.

This change makes the treatment of the 0 value as blocking and as
unconditional as the current treatment of the 2 value, only the CR4.PCE
bit is naturally set to false instead of true.

Signed-off-by: Anthony Steinhauser <[email protected]>
Cc: Alexander Shishkin <[email protected]>
Cc: Arnaldo Carvalho de Melo <[email protected]>
Cc: Jiri Olsa <[email protected]>
Cc: Linus Torvalds <[email protected]>
Cc: Mark Rutland <[email protected]>
Cc: Namhyung Kim <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Cc: Stephane Eranian <[email protected]>
Cc: Thomas Gleixner <[email protected]>
Cc: Vince Weaver <[email protected]>
Cc: [email protected]
Link: https://lkml.kernel.org/r/[email protected]
Signed-off-by: Ingo Molnar <[email protected]>

arch/x86/events/core.c

@@ -49,6 +49,7 @@ DEFINE_PER_CPU(struct cpu_hw_events, cpu_hw_events) = {
 	.enabled = 1,
 };
 
+DEFINE_STATIC_KEY_FALSE(rdpmc_never_available_key);
 DEFINE_STATIC_KEY_FALSE(rdpmc_always_available_key);
 
 u64 __read_mostly hw_cache_event_ids
@@ -2181,21 +2182,26 @@ static ssize_t set_attr_rdpmc(struct device *cdev,
 	if (x86_pmu.attr_rdpmc_broken)
 		return -ENOTSUPP;
 
-	if ((val == 2) != (x86_pmu.attr_rdpmc == 2)) {
+	if (val != x86_pmu.attr_rdpmc) {
 		/*
-		 * Changing into or out of always available, aka
-		 * perf-event-bypassing mode.  This path is extremely slow,
+		 * Changing into or out of never available or always available,
+		 * aka perf-event-bypassing mode. This path is extremely slow,
 		 * but only root can trigger it, so it's okay.
 		 */
+		if (val == 0)
+			static_branch_inc(&rdpmc_never_available_key);
+		else if (x86_pmu.attr_rdpmc == 0)
+			static_branch_dec(&rdpmc_never_available_key);
+
 		if (val == 2)
 			static_branch_inc(&rdpmc_always_available_key);
-		else
+		else if (x86_pmu.attr_rdpmc == 2)
 			static_branch_dec(&rdpmc_always_available_key);
+
 		on_each_cpu(refresh_pce, NULL, 1);
+		x86_pmu.attr_rdpmc = val;
 	}
 
-	x86_pmu.attr_rdpmc = val;
-
 	return count;
 }
 

arch/x86/include/asm/mmu_context.h

@@ -26,12 +26,14 @@ static inline void paravirt_activate_mm(struct mm_struct *prev,
 
 #ifdef CONFIG_PERF_EVENTS
 
+DECLARE_STATIC_KEY_FALSE(rdpmc_never_available_key);
 DECLARE_STATIC_KEY_FALSE(rdpmc_always_available_key);
 
 static inline void load_mm_cr4_irqsoff(struct mm_struct *mm)
 {
 	if (static_branch_unlikely(&rdpmc_always_available_key) ||
-	    atomic_read(&mm->context.perf_rdpmc_allowed))
+	    (!static_branch_unlikely(&rdpmc_never_available_key) &&
+	     atomic_read(&mm->context.perf_rdpmc_allowed)))
 		cr4_set_bits_irqsoff(X86_CR4_PCE);
 	else
 		cr4_clear_bits_irqsoff(X86_CR4_PCE);