The first four patches are fixes for XSA-332. The avoid WARN splats and a performance issue with interdomain events.
Patches 5 and 6 are some additions to event handling in order to add some per pv-device statistics to sysfs and the ability to have a per backend device spurious event delay control.
Patches 7 and 8 are minor fixes I had lying around.
Juergen Gross (8): xen/events: reset affinity of 2-level event when tearing it down xen/events: don't unmask an event channel when an eoi is pending xen/events: avoid handling the same event on two cpus at the same time xen/netback: fix spurious event detection for common event case xen/events: link interdomain events to associated xenbus device xen/events: add per-xenbus device event statistics and settings xen/evtchn: use smp barriers for user event ring xen/evtchn: use READ/WRITE_ONCE() for accessing ring indices
.../ABI/testing/sysfs-devices-xenbus | 41 ++++ drivers/block/xen-blkback/xenbus.c | 2 +- drivers/net/xen-netback/interface.c | 24 ++- drivers/xen/events/events_2l.c | 22 +- drivers/xen/events/events_base.c | 199 +++++++++++++----- drivers/xen/events/events_fifo.c | 7 - drivers/xen/events/events_internal.h | 14 +- drivers/xen/evtchn.c | 29 ++- drivers/xen/pvcalls-back.c | 4 +- drivers/xen/xen-pciback/xenbus.c | 2 +- drivers/xen/xen-scsiback.c | 2 +- drivers/xen/xenbus/xenbus_probe.c | 66 ++++++ include/xen/events.h | 7 +- include/xen/xenbus.h | 7 + 14 files changed, 327 insertions(+), 99 deletions(-) create mode 100644 Documentation/ABI/testing/sysfs-devices-xenbus
When creating a new event channel with 2-level events the affinity needs to be reset initially in order to avoid using an old affinity from earlier usage of the event channel port. So when tearing an event channel down reset all affinity bits.
The same applies to the affinity when onlining a vcpu: all old affinity settings for this vcpu must be reset. As percpu events get initialized before the percpu event channel hook is called, resetting of the affinities happens after offlining a vcpu (this is working, as initial percpu memory is zeroed out).
Cc: stable@vger.kernel.org Reported-by: Julien Grall julien@xen.org Signed-off-by: Juergen Gross jgross@suse.com Reviewed-by: Julien Grall jgrall@amazon.com --- V2: - reset affinity when tearing down the event (Julien Grall) --- drivers/xen/events/events_2l.c | 15 +++++++++++++++ drivers/xen/events/events_base.c | 1 + drivers/xen/events/events_internal.h | 8 ++++++++ 3 files changed, 24 insertions(+)
diff --git a/drivers/xen/events/events_2l.c b/drivers/xen/events/events_2l.c index da87f3a1e351..a7f413c5c190 100644 --- a/drivers/xen/events/events_2l.c +++ b/drivers/xen/events/events_2l.c @@ -47,6 +47,11 @@ static unsigned evtchn_2l_max_channels(void) return EVTCHN_2L_NR_CHANNELS; }
+static void evtchn_2l_remove(evtchn_port_t evtchn, unsigned int cpu) +{ + clear_bit(evtchn, BM(per_cpu(cpu_evtchn_mask, cpu))); +} + static void evtchn_2l_bind_to_cpu(evtchn_port_t evtchn, unsigned int cpu, unsigned int old_cpu) { @@ -355,9 +360,18 @@ static void evtchn_2l_resume(void) EVTCHN_2L_NR_CHANNELS/BITS_PER_EVTCHN_WORD); }
+static int evtchn_2l_percpu_deinit(unsigned int cpu) +{ + memset(per_cpu(cpu_evtchn_mask, cpu), 0, sizeof(xen_ulong_t) * + EVTCHN_2L_NR_CHANNELS/BITS_PER_EVTCHN_WORD); + + return 0; +} + static const struct evtchn_ops evtchn_ops_2l = { .max_channels = evtchn_2l_max_channels, .nr_channels = evtchn_2l_max_channels, + .remove = evtchn_2l_remove, .bind_to_cpu = evtchn_2l_bind_to_cpu, .clear_pending = evtchn_2l_clear_pending, .set_pending = evtchn_2l_set_pending, @@ -367,6 +381,7 @@ static const struct evtchn_ops evtchn_ops_2l = { .unmask = evtchn_2l_unmask, .handle_events = evtchn_2l_handle_events, .resume = evtchn_2l_resume, + .percpu_deinit = evtchn_2l_percpu_deinit, };
void __init xen_evtchn_2l_init(void) diff --git a/drivers/xen/events/events_base.c b/drivers/xen/events/events_base.c index e850f79351cb..6c539db81f8f 100644 --- a/drivers/xen/events/events_base.c +++ b/drivers/xen/events/events_base.c @@ -368,6 +368,7 @@ static int xen_irq_info_pirq_setup(unsigned irq, static void xen_irq_info_cleanup(struct irq_info *info) { set_evtchn_to_irq(info->evtchn, -1); + xen_evtchn_port_remove(info->evtchn, info->cpu); info->evtchn = 0; channels_on_cpu_dec(info); } diff --git a/drivers/xen/events/events_internal.h b/drivers/xen/events/events_internal.h index 0a97c0549db7..18a4090d0709 100644 --- a/drivers/xen/events/events_internal.h +++ b/drivers/xen/events/events_internal.h @@ -14,6 +14,7 @@ struct evtchn_ops { unsigned (*nr_channels)(void);
int (*setup)(evtchn_port_t port); + void (*remove)(evtchn_port_t port, unsigned int cpu); void (*bind_to_cpu)(evtchn_port_t evtchn, unsigned int cpu, unsigned int old_cpu);
@@ -54,6 +55,13 @@ static inline int xen_evtchn_port_setup(evtchn_port_t evtchn) return 0; }
+static inline void xen_evtchn_port_remove(evtchn_port_t evtchn, + unsigned int cpu) +{ + if (evtchn_ops->remove) + evtchn_ops->remove(evtchn, cpu); +} + static inline void xen_evtchn_port_bind_to_cpu(evtchn_port_t evtchn, unsigned int cpu, unsigned int old_cpu)
An event channel should be kept masked when an eoi is pending for it. When being migrated to another cpu it might be unmasked, though.
In order to avoid this keep three different flags for each event channel to be able to distinguish "normal" masking/unmasking from eoi related masking/unmasking and temporary masking. The event channel should only be able to generate an interrupt if all flags are cleared.
Cc: stable@vger.kernel.org Fixes: 54c9de89895e0a36047 ("xen/events: add a new late EOI evtchn framework") Reported-by: Julien Grall julien@xen.org Signed-off-by: Juergen Gross jgross@suse.com --- V2: - introduce a lock around masking/unmasking - merge patch 3 into this one (Jan Beulich) --- drivers/xen/events/events_2l.c | 7 -- drivers/xen/events/events_base.c | 102 +++++++++++++++++++++------ drivers/xen/events/events_fifo.c | 7 -- drivers/xen/events/events_internal.h | 6 -- 4 files changed, 81 insertions(+), 41 deletions(-)
diff --git a/drivers/xen/events/events_2l.c b/drivers/xen/events/events_2l.c index a7f413c5c190..b8f2f971c2f0 100644 --- a/drivers/xen/events/events_2l.c +++ b/drivers/xen/events/events_2l.c @@ -77,12 +77,6 @@ static bool evtchn_2l_is_pending(evtchn_port_t port) return sync_test_bit(port, BM(&s->evtchn_pending[0])); }
-static bool evtchn_2l_test_and_set_mask(evtchn_port_t port) -{ - struct shared_info *s = HYPERVISOR_shared_info; - return sync_test_and_set_bit(port, BM(&s->evtchn_mask[0])); -} - static void evtchn_2l_mask(evtchn_port_t port) { struct shared_info *s = HYPERVISOR_shared_info; @@ -376,7 +370,6 @@ static const struct evtchn_ops evtchn_ops_2l = { .clear_pending = evtchn_2l_clear_pending, .set_pending = evtchn_2l_set_pending, .is_pending = evtchn_2l_is_pending, - .test_and_set_mask = evtchn_2l_test_and_set_mask, .mask = evtchn_2l_mask, .unmask = evtchn_2l_unmask, .handle_events = evtchn_2l_handle_events, diff --git a/drivers/xen/events/events_base.c b/drivers/xen/events/events_base.c index 6c539db81f8f..e157e7506830 100644 --- a/drivers/xen/events/events_base.c +++ b/drivers/xen/events/events_base.c @@ -97,13 +97,18 @@ struct irq_info { short refcnt; u8 spurious_cnt; u8 is_accounted; - enum xen_irq_type type; /* type */ + short type; /* type: IRQT_* */ + u8 mask_reason; /* Why is event channel masked */ +#define EVT_MASK_REASON_EXPLICIT 0x01 +#define EVT_MASK_REASON_TEMPORARY 0x02 +#define EVT_MASK_REASON_EOI_PENDING 0x04 unsigned irq; evtchn_port_t evtchn; /* event channel */ unsigned short cpu; /* cpu bound */ unsigned short eoi_cpu; /* EOI must happen on this cpu-1 */ unsigned int irq_epoch; /* If eoi_cpu valid: irq_epoch of event */ u64 eoi_time; /* Time in jiffies when to EOI. */ + spinlock_t lock;
union { unsigned short virq; @@ -152,6 +157,7 @@ static DEFINE_RWLOCK(evtchn_rwlock); * evtchn_rwlock * IRQ-desc lock * percpu eoi_list_lock + * irq_info->lock */
static LIST_HEAD(xen_irq_list_head); @@ -302,6 +308,8 @@ static int xen_irq_info_common_setup(struct irq_info *info, info->irq = irq; info->evtchn = evtchn; info->cpu = cpu; + info->mask_reason = EVT_MASK_REASON_EXPLICIT; + spin_lock_init(&info->lock);
ret = set_evtchn_to_irq(evtchn, irq); if (ret < 0) @@ -450,6 +458,34 @@ unsigned int cpu_from_evtchn(evtchn_port_t evtchn) return ret; }
+static void do_mask(struct irq_info *info, u8 reason) +{ + unsigned long flags; + + spin_lock_irqsave(&info->lock, flags); + + if (!info->mask_reason) + mask_evtchn(info->evtchn); + + info->mask_reason |= reason; + + spin_unlock_irqrestore(&info->lock, flags); +} + +static void do_unmask(struct irq_info *info, u8 reason) +{ + unsigned long flags; + + spin_lock_irqsave(&info->lock, flags); + + info->mask_reason &= ~reason; + + if (!info->mask_reason) + unmask_evtchn(info->evtchn); + + spin_unlock_irqrestore(&info->lock, flags); +} + #ifdef CONFIG_X86 static bool pirq_check_eoi_map(unsigned irq) { @@ -586,7 +622,7 @@ static void xen_irq_lateeoi_locked(struct irq_info *info, bool spurious) }
info->eoi_time = 0; - unmask_evtchn(evtchn); + do_unmask(info, EVT_MASK_REASON_EOI_PENDING); }
static void xen_irq_lateeoi_worker(struct work_struct *work) @@ -831,7 +867,8 @@ static unsigned int __startup_pirq(unsigned int irq) goto err;
out: - unmask_evtchn(evtchn); + do_unmask(info, EVT_MASK_REASON_EXPLICIT); + eoi_pirq(irq_get_irq_data(irq));
return 0; @@ -858,7 +895,7 @@ static void shutdown_pirq(struct irq_data *data) if (!VALID_EVTCHN(evtchn)) return;
- mask_evtchn(evtchn); + do_mask(info, EVT_MASK_REASON_EXPLICIT); xen_evtchn_close(evtchn); xen_irq_info_cleanup(info); } @@ -1691,10 +1728,10 @@ void rebind_evtchn_irq(evtchn_port_t evtchn, int irq) }
/* Rebind an evtchn so that it gets delivered to a specific cpu */ -static int xen_rebind_evtchn_to_cpu(evtchn_port_t evtchn, unsigned int tcpu) +static int xen_rebind_evtchn_to_cpu(struct irq_info *info, unsigned int tcpu) { struct evtchn_bind_vcpu bind_vcpu; - int masked; + evtchn_port_t evtchn = info ? info->evtchn : 0;
if (!VALID_EVTCHN(evtchn)) return -1; @@ -1710,7 +1747,7 @@ static int xen_rebind_evtchn_to_cpu(evtchn_port_t evtchn, unsigned int tcpu) * Mask the event while changing the VCPU binding to prevent * it being delivered on an unexpected VCPU. */ - masked = test_and_set_mask(evtchn); + do_mask(info, EVT_MASK_REASON_TEMPORARY);
/* * If this fails, it usually just indicates that we're dealing with a @@ -1720,8 +1757,7 @@ static int xen_rebind_evtchn_to_cpu(evtchn_port_t evtchn, unsigned int tcpu) if (HYPERVISOR_event_channel_op(EVTCHNOP_bind_vcpu, &bind_vcpu) >= 0) bind_evtchn_to_cpu(evtchn, tcpu, false);
- if (!masked) - unmask_evtchn(evtchn); + do_unmask(info, EVT_MASK_REASON_TEMPORARY);
return 0; } @@ -1760,7 +1796,7 @@ static int set_affinity_irq(struct irq_data *data, const struct cpumask *dest, unsigned int tcpu = select_target_cpu(dest); int ret;
- ret = xen_rebind_evtchn_to_cpu(evtchn_from_irq(data->irq), tcpu); + ret = xen_rebind_evtchn_to_cpu(info_for_irq(data->irq), tcpu); if (!ret) irq_data_update_effective_affinity(data, cpumask_of(tcpu));
@@ -1769,18 +1805,20 @@ static int set_affinity_irq(struct irq_data *data, const struct cpumask *dest,
static void enable_dynirq(struct irq_data *data) { - evtchn_port_t evtchn = evtchn_from_irq(data->irq); + struct irq_info *info = info_for_irq(data->irq); + evtchn_port_t evtchn = info ? info->evtchn : 0;
if (VALID_EVTCHN(evtchn)) - unmask_evtchn(evtchn); + do_unmask(info, EVT_MASK_REASON_EXPLICIT); }
static void disable_dynirq(struct irq_data *data) { - evtchn_port_t evtchn = evtchn_from_irq(data->irq); + struct irq_info *info = info_for_irq(data->irq); + evtchn_port_t evtchn = info ? info->evtchn : 0;
if (VALID_EVTCHN(evtchn)) - mask_evtchn(evtchn); + do_mask(info, EVT_MASK_REASON_EXPLICIT); }
static void ack_dynirq(struct irq_data *data) @@ -1799,18 +1837,40 @@ static void mask_ack_dynirq(struct irq_data *data) ack_dynirq(data); }
+static void lateeoi_ack_dynirq(struct irq_data *data) +{ + struct irq_info *info = info_for_irq(data->irq); + evtchn_port_t evtchn = info ? info->evtchn : 0; + + if (VALID_EVTCHN(evtchn)) { + do_mask(info, EVT_MASK_REASON_EOI_PENDING); + clear_evtchn(evtchn); + } +} + +static void lateeoi_mask_ack_dynirq(struct irq_data *data) +{ + struct irq_info *info = info_for_irq(data->irq); + evtchn_port_t evtchn = info ? info->evtchn : 0; + + if (VALID_EVTCHN(evtchn)) { + do_mask(info, EVT_MASK_REASON_EXPLICIT | + EVT_MASK_REASON_EOI_PENDING); + clear_evtchn(evtchn); + } +} + static int retrigger_dynirq(struct irq_data *data) { - evtchn_port_t evtchn = evtchn_from_irq(data->irq); - int masked; + struct irq_info *info = info_for_irq(data->irq); + evtchn_port_t evtchn = info ? info->evtchn : 0;
if (!VALID_EVTCHN(evtchn)) return 0;
- masked = test_and_set_mask(evtchn); + do_mask(info, EVT_MASK_REASON_TEMPORARY); set_evtchn(evtchn); - if (!masked) - unmask_evtchn(evtchn); + do_unmask(info, EVT_MASK_REASON_TEMPORARY);
return 1; } @@ -2024,8 +2084,8 @@ static struct irq_chip xen_lateeoi_chip __read_mostly = { .irq_mask = disable_dynirq, .irq_unmask = enable_dynirq,
- .irq_ack = mask_ack_dynirq, - .irq_mask_ack = mask_ack_dynirq, + .irq_ack = lateeoi_ack_dynirq, + .irq_mask_ack = lateeoi_mask_ack_dynirq,
.irq_set_affinity = set_affinity_irq, .irq_retrigger = retrigger_dynirq, diff --git a/drivers/xen/events/events_fifo.c b/drivers/xen/events/events_fifo.c index b234f1766810..ad9fe51d3fb3 100644 --- a/drivers/xen/events/events_fifo.c +++ b/drivers/xen/events/events_fifo.c @@ -209,12 +209,6 @@ static bool evtchn_fifo_is_pending(evtchn_port_t port) return sync_test_bit(EVTCHN_FIFO_BIT(PENDING, word), BM(word)); }
-static bool evtchn_fifo_test_and_set_mask(evtchn_port_t port) -{ - event_word_t *word = event_word_from_port(port); - return sync_test_and_set_bit(EVTCHN_FIFO_BIT(MASKED, word), BM(word)); -} - static void evtchn_fifo_mask(evtchn_port_t port) { event_word_t *word = event_word_from_port(port); @@ -423,7 +417,6 @@ static const struct evtchn_ops evtchn_ops_fifo = { .clear_pending = evtchn_fifo_clear_pending, .set_pending = evtchn_fifo_set_pending, .is_pending = evtchn_fifo_is_pending, - .test_and_set_mask = evtchn_fifo_test_and_set_mask, .mask = evtchn_fifo_mask, .unmask = evtchn_fifo_unmask, .handle_events = evtchn_fifo_handle_events, diff --git a/drivers/xen/events/events_internal.h b/drivers/xen/events/events_internal.h index 18a4090d0709..4d3398eff9cd 100644 --- a/drivers/xen/events/events_internal.h +++ b/drivers/xen/events/events_internal.h @@ -21,7 +21,6 @@ struct evtchn_ops { void (*clear_pending)(evtchn_port_t port); void (*set_pending)(evtchn_port_t port); bool (*is_pending)(evtchn_port_t port); - bool (*test_and_set_mask)(evtchn_port_t port); void (*mask)(evtchn_port_t port); void (*unmask)(evtchn_port_t port);
@@ -84,11 +83,6 @@ static inline bool test_evtchn(evtchn_port_t port) return evtchn_ops->is_pending(port); }
-static inline bool test_and_set_mask(evtchn_port_t port) -{ - return evtchn_ops->test_and_set_mask(port); -} - static inline void mask_evtchn(evtchn_port_t port) { return evtchn_ops->mask(port);
Hi Juergen,
On 19/02/2021 15:40, Juergen Gross wrote:
An event channel should be kept masked when an eoi is pending for it. When being migrated to another cpu it might be unmasked, though.
In order to avoid this keep three different flags for each event channel to be able to distinguish "normal" masking/unmasking from eoi related masking/unmasking and temporary masking. The event channel should only be able to generate an interrupt if all flags are cleared.
Cc: stable@vger.kernel.org Fixes: 54c9de89895e0a36047 ("xen/events: add a new late EOI evtchn framework") Reported-by: Julien Grall julien@xen.org Signed-off-by: Juergen Gross jgross@suse.com
Reviewed-by: Julien Grall jgrall@amazon.com
Cheers,
On 2021-02-19 15:40, Juergen Gross wrote:
An event channel should be kept masked when an eoi is pending for it. When being migrated to another cpu it might be unmasked, though.
In order to avoid this keep three different flags for each event channel to be able to distinguish "normal" masking/unmasking from eoi related masking/unmasking and temporary masking. The event channel should only be able to generate an interrupt if all flags are cleared.
Cc: stable@vger.kernel.org Fixes: 54c9de89895e0a36047 ("xen/events: add a new late EOI evtchn framework") Reported-by: Julien Grall julien@xen.org Signed-off-by: Juergen Gross jgross@suse.com
I tested this patch series backported to a 4.19 kernel and found that when doing a reboot loop of Windows with PV drivers, occasionally it will end up in a state with some event channels pending and masked in dom0 which breaks networking in the guest.
The issue seems to have been introduced with this patch, though at first glance it appears correct. I haven't yet looked into why it is happening. Have you seen anything like this with this patch?
Thanks, Ross
On 23.02.21 10:26, Ross Lagerwall wrote:
On 2021-02-19 15:40, Juergen Gross wrote:
An event channel should be kept masked when an eoi is pending for it. When being migrated to another cpu it might be unmasked, though.
In order to avoid this keep three different flags for each event channel to be able to distinguish "normal" masking/unmasking from eoi related masking/unmasking and temporary masking. The event channel should only be able to generate an interrupt if all flags are cleared.
Cc: stable@vger.kernel.org Fixes: 54c9de89895e0a36047 ("xen/events: add a new late EOI evtchn framework") Reported-by: Julien Grall julien@xen.org Signed-off-by: Juergen Gross jgross@suse.com
I tested this patch series backported to a 4.19 kernel and found that when doing a reboot loop of Windows with PV drivers, occasionally it will end up in a state with some event channels pending and masked in dom0 which breaks networking in the guest.
The issue seems to have been introduced with this patch, though at first glance it appears correct. I haven't yet looked into why it is happening. Have you seen anything like this with this patch?
Sorry it took so long, but now I was able to look into this issue.
I have managed to reproduce it with a pv Linux guest. I'm now adding some debug code to understand what is happening there.
Juergen
On 23.02.21 10:26, Ross Lagerwall wrote:
On 2021-02-19 15:40, Juergen Gross wrote:
An event channel should be kept masked when an eoi is pending for it. When being migrated to another cpu it might be unmasked, though.
In order to avoid this keep three different flags for each event channel to be able to distinguish "normal" masking/unmasking from eoi related masking/unmasking and temporary masking. The event channel should only be able to generate an interrupt if all flags are cleared.
Cc: stable@vger.kernel.org Fixes: 54c9de89895e0a36047 ("xen/events: add a new late EOI evtchn framework") Reported-by: Julien Grall julien@xen.org Signed-off-by: Juergen Gross jgross@suse.com
I tested this patch series backported to a 4.19 kernel and found that when doing a reboot loop of Windows with PV drivers, occasionally it will end up in a state with some event channels pending and masked in dom0 which breaks networking in the guest.
The issue seems to have been introduced with this patch, though at first glance it appears correct. I haven't yet looked into why it is happening. Have you seen anything like this with this patch?
I have found the issue. lateeoi_mask_ack_dynirq() must not set the "eoi" mask reason flag, as this callback will be called when the handler will not be called later, so there will never be a call of xen_irq_lateeoi() to unmask the event channel again.
Juergen
When changing the cpu affinity of an event it can happen today that (with some unlucky timing) the same event will be handled on the old and the new cpu at the same time.
Avoid that by adding an "event active" flag to the per-event data and call the handler only if this flag isn't set.
Cc: stable@vger.kernel.org Reported-by: Julien Grall julien@xen.org Signed-off-by: Juergen Gross jgross@suse.com --- V2: - new patch V3: - use common helper for end of handler action (Julien Grall) - move setting is_active to 0 for lateeoi (Boris Ostrovsky) --- drivers/xen/events/events_base.c | 32 +++++++++++++++++++++----------- 1 file changed, 21 insertions(+), 11 deletions(-)
diff --git a/drivers/xen/events/events_base.c b/drivers/xen/events/events_base.c index e157e7506830..9d7ba7623510 100644 --- a/drivers/xen/events/events_base.c +++ b/drivers/xen/events/events_base.c @@ -102,6 +102,7 @@ struct irq_info { #define EVT_MASK_REASON_EXPLICIT 0x01 #define EVT_MASK_REASON_TEMPORARY 0x02 #define EVT_MASK_REASON_EOI_PENDING 0x04 + u8 is_active; /* Is event just being handled? */ unsigned irq; evtchn_port_t evtchn; /* event channel */ unsigned short cpu; /* cpu bound */ @@ -791,6 +792,12 @@ static void xen_evtchn_close(evtchn_port_t port) BUG(); }
+static void event_handler_exit(struct irq_info *info) +{ + smp_store_release(&info->is_active, 0); + clear_evtchn(info->evtchn); +} + static void pirq_query_unmask(int irq) { struct physdev_irq_status_query irq_status; @@ -809,14 +816,15 @@ static void pirq_query_unmask(int irq)
static void eoi_pirq(struct irq_data *data) { - evtchn_port_t evtchn = evtchn_from_irq(data->irq); + struct irq_info *info = info_for_irq(data->irq); + evtchn_port_t evtchn = info ? info->evtchn : 0; struct physdev_eoi eoi = { .irq = pirq_from_irq(data->irq) }; int rc = 0;
if (!VALID_EVTCHN(evtchn)) return;
- clear_evtchn(evtchn); + event_handler_exit(info);
if (pirq_needs_eoi(data->irq)) { rc = HYPERVISOR_physdev_op(PHYSDEVOP_eoi, &eoi); @@ -1640,6 +1648,8 @@ void handle_irq_for_port(evtchn_port_t port, struct evtchn_loop_ctrl *ctrl) }
info = info_for_irq(irq); + if (xchg_acquire(&info->is_active, 1)) + return;
if (ctrl->defer_eoi) { info->eoi_cpu = smp_processor_id(); @@ -1823,12 +1833,11 @@ static void disable_dynirq(struct irq_data *data)
static void ack_dynirq(struct irq_data *data) { - evtchn_port_t evtchn = evtchn_from_irq(data->irq); - - if (!VALID_EVTCHN(evtchn)) - return; + struct irq_info *info = info_for_irq(data->irq); + evtchn_port_t evtchn = info ? info->evtchn : 0;
- clear_evtchn(evtchn); + if (VALID_EVTCHN(evtchn)) + event_handler_exit(info); }
static void mask_ack_dynirq(struct irq_data *data) @@ -1844,7 +1853,7 @@ static void lateeoi_ack_dynirq(struct irq_data *data)
if (VALID_EVTCHN(evtchn)) { do_mask(info, EVT_MASK_REASON_EOI_PENDING); - clear_evtchn(evtchn); + event_handler_exit(info); } }
@@ -1856,7 +1865,7 @@ static void lateeoi_mask_ack_dynirq(struct irq_data *data) if (VALID_EVTCHN(evtchn)) { do_mask(info, EVT_MASK_REASON_EXPLICIT | EVT_MASK_REASON_EOI_PENDING); - clear_evtchn(evtchn); + event_handler_exit(info); } }
@@ -1969,10 +1978,11 @@ static void restore_cpu_ipis(unsigned int cpu) /* Clear an irq's pending state, in preparation for polling on it */ void xen_clear_irq_pending(int irq) { - evtchn_port_t evtchn = evtchn_from_irq(irq); + struct irq_info *info = info_for_irq(irq); + evtchn_port_t evtchn = info ? info->evtchn : 0;
if (VALID_EVTCHN(evtchn)) - clear_evtchn(evtchn); + event_handler_exit(info); } EXPORT_SYMBOL(xen_clear_irq_pending); void xen_set_irq_pending(int irq)
Hi Juergen,
On 19/02/2021 15:40, Juergen Gross wrote:
When changing the cpu affinity of an event it can happen today that (with some unlucky timing) the same event will be handled on the old and the new cpu at the same time.
Avoid that by adding an "event active" flag to the per-event data and call the handler only if this flag isn't set.
Cc: stable@vger.kernel.org Reported-by: Julien Grall julien@xen.org Signed-off-by: Juergen Gross jgross@suse.com
Reviewed-by: Julien Grall jgrall@amazon.com
Cheers,
In case of a common event for rx and tx queue the event should be regarded to be spurious if no rx and no tx requests are pending.
Unfortunately the condition for testing that is wrong causing to decide a event being spurious if no rx OR no tx requests are pending.
Fix that plus using local variables for rx/tx pending indicators in order to split function calls and if condition.
Cc: stable@vger.kernel.org Fixes: 23025393dbeb3b ("xen/netback: use lateeoi irq binding") Signed-off-by: Juergen Gross jgross@suse.com Reviewed-by: Jan Beulich jbeulich@suse.com Reviewed-by: Paul Durrant paul@xen.org Reviewed-by: Wei Liu wl@xen.org --- V2: - new patch, fixing FreeBSD performance issue --- drivers/net/xen-netback/interface.c | 8 +++++--- 1 file changed, 5 insertions(+), 3 deletions(-)
diff --git a/drivers/net/xen-netback/interface.c b/drivers/net/xen-netback/interface.c index acb786d8b1d8..e02a4fbb74de 100644 --- a/drivers/net/xen-netback/interface.c +++ b/drivers/net/xen-netback/interface.c @@ -162,13 +162,15 @@ irqreturn_t xenvif_interrupt(int irq, void *dev_id) { struct xenvif_queue *queue = dev_id; int old; + bool has_rx, has_tx;
old = atomic_fetch_or(NETBK_COMMON_EOI, &queue->eoi_pending); WARN(old, "Interrupt while EOI pending\n");
- /* Use bitwise or as we need to call both functions. */ - if ((!xenvif_handle_tx_interrupt(queue) | - !xenvif_handle_rx_interrupt(queue))) { + has_tx = xenvif_handle_tx_interrupt(queue); + has_rx = xenvif_handle_rx_interrupt(queue); + + if (!has_rx && !has_tx) { atomic_andnot(NETBK_COMMON_EOI, &queue->eoi_pending); xen_irq_lateeoi(irq, XEN_EOI_FLAG_SPURIOUS); }
On 2/19/21 10:40 AM, Juergen Gross wrote:
The first four patches are fixes for XSA-332. The avoid WARN splats and a performance issue with interdomain events.
Patches 5 and 6 are some additions to event handling in order to add some per pv-device statistics to sysfs and the ability to have a per backend device spurious event delay control.
Patches 7 and 8 are minor fixes I had lying around.
Juergen Gross (8): xen/events: reset affinity of 2-level event when tearing it down xen/events: don't unmask an event channel when an eoi is pending xen/events: avoid handling the same event on two cpus at the same time xen/netback: fix spurious event detection for common event case xen/events: link interdomain events to associated xenbus device xen/events: add per-xenbus device event statistics and settings xen/evtchn: use smp barriers for user event ring xen/evtchn: use READ/WRITE_ONCE() for accessing ring indices
I am going to pick up the last 3 patches since Ross appears to be having some issues with #2 (and 4 and 5 went in via netdev tree)
-boris
linux-stable-mirror@lists.linaro.org
-
Boris Ostrovsky -
Juergen Gross -
Julien Grall -
Jürgen Groß
-
Ross Lagerwall