[Xen-devel] [PATCH v3] x86/cpu: Sync any remaining RCU callbacks before CPU up/down

* [Xen-devel] [PATCH v3] x86/cpu: Sync any remaining RCU callbacks before CPU up/down
@ 2020-03-04 15:33 Igor Druzhinin
  2020-03-06  9:43 ` Jan Beulich
  0 siblings, 1 reply; 4+ messages in thread
From: Igor Druzhinin @ 2020-03-04 15:33 UTC (permalink / raw)
  To: xen-devel
  Cc: Igor Druzhinin, sstabellini, julien, wl, andrew.cooper3,
	ian.jackson, george.dunlap, jbeulich, roger.pau

During CPU down operation RCU callbacks are scheduled to finish
off some actions later as soon as CPU is fully dead (the same applies
to CPU up operation in case error path is taken). If in the same grace
period another CPU up operation is performed on the same CPU, RCU callback
will be called later on a CPU in a potentially wrong (already up again
instead of still being down) state leading to eventual state inconsistency
and/or crash.

In order to avoid it - flush RCU callbacks explicitly before starting the
next CPU up/down operation.

Reviewed-by: Juergen Gross <jgross@suse.com>
Signed-off-by: Igor Druzhinin <igor.druzhinin@citrix.com>
---
This got discovered trying to resume PV shim with multiple vCPUs on AMD
machine (where park_offline_cpus == 0). RCU callback responsible for
freeing percpu area on CPU offline got finally called after CPU went
online again as the guest performed regular vCPU offline/online operations
on resume.

Note: this patch requires RCU series v3 from Juergen to be applied -
https://lists.xenproject.org/archives/html/xen-devel/2020-03/msg00200.html

v2: changed rcu_barrier() position, updated description
v3: moved rcu_barrier() to common cpu_up/cpu_down code to cover more cases
---
 xen/arch/x86/acpi/power.c | 1 -
 xen/arch/x86/sysctl.c     | 8 --------
 xen/common/cpu.c          | 2 ++
 3 files changed, 2 insertions(+), 9 deletions(-)

diff --git a/xen/arch/x86/acpi/power.c b/xen/arch/x86/acpi/power.c
index b5df00b..847c273 100644
--- a/xen/arch/x86/acpi/power.c
+++ b/xen/arch/x86/acpi/power.c
@@ -305,7 +305,6 @@ static int enter_state(u32 state)
     cpufreq_add_cpu(0);
 
  enable_cpu:
-    rcu_barrier();
     mtrr_aps_sync_begin();
     enable_nonboot_cpus();
     mtrr_aps_sync_end();
diff --git a/xen/arch/x86/sysctl.c b/xen/arch/x86/sysctl.c
index 59a3840..b4e86a8 100644
--- a/xen/arch/x86/sysctl.c
+++ b/xen/arch/x86/sysctl.c
@@ -85,11 +85,7 @@ long cpu_up_helper(void *data)
     int ret = cpu_up(cpu);
 
     if ( ret == -EBUSY )
-    {
-        /* On EBUSY, flush RCU work and have one more go. */
-        rcu_barrier();
         ret = cpu_up(cpu);
-    }
 
     if ( !ret && !opt_smt &&
          cpu_data[cpu].compute_unit_id == INVALID_CUID &&
@@ -110,11 +106,7 @@ long cpu_down_helper(void *data)
     int cpu = (unsigned long)data;
     int ret = cpu_down(cpu);
     if ( ret == -EBUSY )
-    {
-        /* On EBUSY, flush RCU work and have one more go. */
-        rcu_barrier();
         ret = cpu_down(cpu);
-    }
     return ret;
 }
 
diff --git a/xen/common/cpu.c b/xen/common/cpu.c
index 31953f3..1f976db 100644
--- a/xen/common/cpu.c
+++ b/xen/common/cpu.c
@@ -4,6 +4,7 @@
 #include <xen/init.h>
 #include <xen/sched.h>
 #include <xen/stop_machine.h>
+#include <xen/rcupdate.h>
 
 unsigned int __read_mostly nr_cpu_ids = NR_CPUS;
 #ifndef nr_cpumask_bits
@@ -53,6 +54,7 @@ void put_cpu_maps(void)
 
 void cpu_hotplug_begin(void)
 {
+    rcu_barrier();
     write_lock(&cpu_add_remove_lock);
 }
 
-- 
2.7.4


_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xenproject.org
https://lists.xenproject.org/mailman/listinfo/xen-devel

^ permalink raw reply related	[flat|nested] 4+ messages in thread