Commit 98229aa3 authored by Thomas Gleixner's avatar Thomas Gleixner

x86/irq: Plug vector cleanup race

We still can end up with a stale vector due to the following:

CPU0                          CPU1                      CPU2
lock_vector()
data->move_in_progress=0
sendIPI()                       
unlock_vector()
                              set_affinity()
                              assign_irq_vector()
                              lock_vector()             handle_IPI
                              move_in_progress = 1      lock_vector()
                              unlock_vector()
                                                        move_in_progress == 1

So we need to serialize the vector assignment against a pending cleanup. The
solution is rather simple now. We not only check for the move_in_progress flag
in assign_irq_vector(), we also check whether there is still a cleanup pending
in the old_domain cpumask. If so, we return -EBUSY to the caller and let him
deal with it. Though we have to be careful in the cpu unplug case. If the
cleanout has not yet completed then the following setaffinity() call would
return -EBUSY. Add code which prevents this.

Full context is here: http://lkml.kernel.org/r/5653B688.4050809@stratus.comReported-and-tested-by: default avatarJoe Lawrence <joe.lawrence@stratus.com>
Signed-off-by: default avatarThomas Gleixner <tglx@linutronix.de>
Tested-by: default avatarBorislav Petkov <bp@alien8.de>
Cc: Jiang Liu <jiang.liu@linux.intel.com>
Cc: Jeremiah Mahler <jmmahler@gmail.com>
Cc: andy.shevchenko@gmail.com
Cc: Guenter Roeck <linux@roeck-us.net>
Cc: stable@vger.kernel.org #4.3+
Link: http://lkml.kernel.org/r/20151231160107.207265407@linutronix.deSigned-off-by: default avatarThomas Gleixner <tglx@linutronix.de>
parent 90a2282e
...@@ -120,7 +120,12 @@ static int __assign_irq_vector(int irq, struct apic_chip_data *d, ...@@ -120,7 +120,12 @@ static int __assign_irq_vector(int irq, struct apic_chip_data *d,
static int current_offset = VECTOR_OFFSET_START % 16; static int current_offset = VECTOR_OFFSET_START % 16;
int cpu, vector; int cpu, vector;
if (d->move_in_progress) /*
* If there is still a move in progress or the previous move has not
* been cleaned up completely, tell the caller to come back later.
*/
if (d->move_in_progress ||
cpumask_intersects(d->old_domain, cpu_online_mask))
return -EBUSY; return -EBUSY;
/* Only try and allocate irqs on cpus that are present */ /* Only try and allocate irqs on cpus that are present */
...@@ -259,7 +264,12 @@ static void clear_irq_vector(int irq, struct apic_chip_data *data) ...@@ -259,7 +264,12 @@ static void clear_irq_vector(int irq, struct apic_chip_data *data)
data->cfg.vector = 0; data->cfg.vector = 0;
cpumask_clear(data->domain); cpumask_clear(data->domain);
if (likely(!data->move_in_progress)) /*
* If move is in progress or the old_domain mask is not empty,
* i.e. the cleanup IPI has not been processed yet, we need to remove
* the old references to desc from all cpus vector tables.
*/
if (!data->move_in_progress && cpumask_empty(data->old_domain))
return; return;
desc = irq_to_desc(irq); desc = irq_to_desc(irq);
...@@ -579,12 +589,25 @@ asmlinkage __visible void smp_irq_move_cleanup_interrupt(void) ...@@ -579,12 +589,25 @@ asmlinkage __visible void smp_irq_move_cleanup_interrupt(void)
goto unlock; goto unlock;
/* /*
* Check if the irq migration is in progress. If so, we * Nothing to cleanup if irq migration is in progress
* haven't received the cleanup request yet for this irq. * or this cpu is not set in the cleanup mask.
*/ */
if (data->move_in_progress) if (data->move_in_progress ||
!cpumask_test_cpu(me, data->old_domain))
goto unlock; goto unlock;
/*
* We have two cases to handle here:
* 1) vector is unchanged but the target mask got reduced
* 2) vector and the target mask has changed
*
* #1 is obvious, but in #2 we have two vectors with the same
* irq descriptor: the old and the new vector. So we need to
* make sure that we only cleanup the old vector. The new
* vector has the current @vector number in the config and
* this cpu is part of the target mask. We better leave that
* one alone.
*/
if (vector == data->cfg.vector && if (vector == data->cfg.vector &&
cpumask_test_cpu(me, data->domain)) cpumask_test_cpu(me, data->domain))
goto unlock; goto unlock;
...@@ -602,6 +625,7 @@ asmlinkage __visible void smp_irq_move_cleanup_interrupt(void) ...@@ -602,6 +625,7 @@ asmlinkage __visible void smp_irq_move_cleanup_interrupt(void)
goto unlock; goto unlock;
} }
__this_cpu_write(vector_irq[vector], VECTOR_UNUSED); __this_cpu_write(vector_irq[vector], VECTOR_UNUSED);
cpumask_clear_cpu(me, data->old_domain);
unlock: unlock:
raw_spin_unlock(&desc->lock); raw_spin_unlock(&desc->lock);
} }
...@@ -645,13 +669,32 @@ void irq_force_complete_move(struct irq_desc *desc) ...@@ -645,13 +669,32 @@ void irq_force_complete_move(struct irq_desc *desc)
__irq_complete_move(cfg, cfg->vector); __irq_complete_move(cfg, cfg->vector);
/* /*
* Remove this cpu from the cleanup mask. The IPI might have been sent * This is tricky. If the cleanup of @data->old_domain has not been
* just before the cpu was removed from the offline mask, but has not * done yet, then the following setaffinity call will fail with
* been processed because the CPU has interrupts disabled and is on * -EBUSY. This can leave the interrupt in a stale state.
* the way out. *
* The cleanup cannot make progress because we hold @desc->lock. So in
* case @data->old_domain is not yet cleaned up, we need to drop the
* lock and acquire it again. @desc cannot go away, because the
* hotplug code holds the sparse irq lock.
*/ */
raw_spin_lock(&vector_lock); raw_spin_lock(&vector_lock);
cpumask_clear_cpu(smp_processor_id(), data->old_domain); /* Clean out all offline cpus (including ourself) first. */
cpumask_and(data->old_domain, data->old_domain, cpu_online_mask);
while (!cpumask_empty(data->old_domain)) {
raw_spin_unlock(&vector_lock);
raw_spin_unlock(&desc->lock);
cpu_relax();
raw_spin_lock(&desc->lock);
/*
* Reevaluate apic_chip_data. It might have been cleared after
* we dropped @desc->lock.
*/
data = apic_chip_data(irqdata);
if (!data)
return;
raw_spin_lock(&vector_lock);
}
raw_spin_unlock(&vector_lock); raw_spin_unlock(&vector_lock);
} }
#endif #endif
......
Markdown is supported
0%
or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment