rework on the IRQ hardening of virtio
Subsystems (7)
v3 cover rework on the IRQ hardening of virtio
Hi All:
This is a rework on the IRQ hardening for virtio which is done previously by the following commits are reverted:
9e35276a5344 ("virtio_pci: harden MSI-X interrupts") 080cd7c3ac87 ("virtio-pci: harden INTX interrupts")
The reason is that it depends on the IRQF_NO_AUTOEN which may conflict with the assumption of the affinity managed IRQ that is used by some virtio drivers. And what's more, it is only done for virtio-pci but not other transports.
In this rework, I try to implement a general virtio solution which borrows the idea of the INTX hardening by re-using per virtqueue boolean vq->broken and toggle it in virtio_device_ready() and virtio_reset_device(). Then we can simply reuse the existing checks in the vring_interrupt() and return early if the driver is not ready.
Note that, I only did compile test on ccw and MMIO transport.
Please review.
Changes since v1:
- Use transport specific irq synchronization method when possible
- Drop the module parameter and enable the hardening unconditonally
- Tweak the barrier/ordering facilities used in the code
- Reanme irq_soft_enabled to driver_ready
- Avoid unnecssary IRQ synchornization (e.g during boot)
Changes since V2:
- add ccw and MMIO support
- rename synchronize_vqs() to synchronize_cbs()
- switch to re-use vq->broken instead of introducing new device attributes for the future virtqueue reset support
- remove unnecssary READ_ONCE()/WRITE_ONCE()
- a new patch to remove device triggerable BUG_ON()
- more tweaks on the comments
Jason Wang (8):
virtio: use virtio_reset_device() when possible
virtio: introduce config op to synchronize vring callbacks
virtio-pci: implement synchronize_cbs()
virtio-mmio: implement synchronize_cbs()
virtio-ccw: implement synchronize_cbs()
virtio: allow to unbreak virtqueue
virtio: harden vring IRQ
virtio: use WARN_ON() to warning illegal status value
Stefano Garzarella (1):
virtio: use virtio_device_ready() in virtio_device_restore()
drivers/char/virtio_console.c | 2 +- drivers/crypto/virtio/virtio_crypto_core.c | 2 +- drivers/s390/virtio/virtio_ccw.c | 31 ++++++++++++++++-- drivers/virtio/virtio.c | 24 ++++++++++---- drivers/virtio/virtio_mmio.c | 9 +++++ drivers/virtio/virtio_pci_common.c | 2 +- drivers/virtio/virtio_pci_common.h | 2 ++ drivers/virtio/virtio_pci_legacy.c | 1 + drivers/virtio/virtio_pci_modern.c | 2 ++ drivers/virtio/virtio_ring.c | 15 +++++---- include/linux/virtio.h | 2 +- include/linux/virtio_config.h | 38 +++++++++++++++++++++- 12 files changed, 110 insertions(+), 20 deletions(-)
Signature · 7 lines
2.25.1 _______________________________________________ Virtualization mailing list [email protected] https://lists.linuxfoundation.org/mailman/listinfo/virtualization
PATCH v3 1/9 virtio: use virtio_device_ready() in virtio_device_restore()
From: Stefano Garzarella <[email protected]>
It will allow us to do extension on virtio_device_ready() without duplicating code.
--- drivers/virtio/virtio.c | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-)
drivers/virtio/virtio.c +3 -2
@@ -526,8 +526,9 @@ int virtio_device_restore(struct virtio_device *dev) goto err; } -/* Finally, tell the device we're all set */ -virtio_add_status (dev,VIRTIO_CONFIG_S_DRIVER_OK ); +/* If restore didn't do it, mark device DRIVER_OK ourselves. */ +if (!(dev->config->get_status (dev) &VIRTIO_CONFIG_S_DRIVER_OK )) +virtio_device_ready (dev);virtio_config_enable (dev);
Signature · 6 lines
2.25.1 _______________________________________________ Virtualization mailing list [email protected] https://lists.linuxfoundation.org/mailman/listinfo/virtualization
Re: [PATCH V3 1/9] virtio: use virtio_device_ready() in virtio_device_restore()
Quoted 13 lines · Jason Wang
From: Stefano Garzarella <[email protected]>
It will allow us to do extension on
virtio_device_ready()without duplicating code.Cc Thomas Gleixner <[email protected]>Cc Peter Zijlstra <[email protected]>Cc "Paul E. McKenney" <[email protected]>Cc Marc Zyngier <[email protected]>Cc Halil Pasic <[email protected]>Cc Cornelia Huck <[email protected]>Signed-off-by Stefano Garzarella <[email protected]>Signed-off-by Jason Wang <[email protected]>
Quoted 3 lines
--- drivers/virtio/virtio.c | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-)
_______________________________________________
Virtualization mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/virtualization
PATCH v3 2/9 virtio: use virtio_reset_device() when possible
In reply to Jason WangThis allows us to do common extension without duplicating code.
--- drivers/virtio/virtio.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)
drivers/virtio/virtio.c +2 -2
@@ -430,7 +430,7 @@ int register_virtio_device(struct virtio_device *dev)/* We always start by resetting the device, in case a previous * driver messed it up. This also tests that code path a little. */ - dev->config->reset (dev); +virtio_reset_device (dev);/* Acknowledge that we've seen the device. */ virtio_add_status (dev,VIRTIO_CONFIG_S_ACKNOWLEDGE );
@@ -496,7 +496,7 @@ int virtio_device_restore(struct virtio_device *dev)/* We always start by resetting the device, in case a previous * driver messed it up. */ - dev->config->reset (dev); +virtio_reset_device (dev);/* Acknowledge that we've seen the device. */ virtio_add_status (dev,VIRTIO_CONFIG_S_ACKNOWLEDGE );
Signature · 1 line
2.25.1
Re: [PATCH V3 2/9] virtio: use virtio_reset_device() when possible
Quoted 9 lines · Jason Wang
This allows us to do common extension without duplicating code.
Cc Thomas Gleixner <[email protected]>Cc Peter Zijlstra <[email protected]>Cc "Paul E. McKenney" <[email protected]>Cc Marc Zyngier <[email protected]>Cc Halil Pasic <[email protected]>Cc Cornelia Huck <[email protected]>Signed-off-by Jason Wang <[email protected]>
Quoted 3 lines
--- drivers/virtio/virtio.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)
_______________________________________________
Virtualization mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/virtualization
PATCH v3 3/9 virtio: introduce config op to synchronize vring callbacks
In reply to Jason WangThis patch introduces new virtio config op to vring callbacks. Transport specific method is required to make sure the write before this function is visible to the vring_interrupt() that is called after the return of this function. For the transport that doesn't provide synchronize_vqs(), use synchornize_rcu() which synchronize with IRQ implicitly as a fallback.
--- include/linux/virtio_config.h | 24 ++++++++++++++++++++++++ 1 file changed, 24 insertions(+)
include/linux/virtio_config.h +24 -0
@@ -57,6 +57,10 @@ struct virtio_shm_region {* include a NULL entry for vqs unused by driver * Returns 0 on success or error status * @del_vqs: free virtqueues found by find_vqs(). +* @synchronize_cbs: synchronize with the virtqueue callbacks (optional) +* Make sure the writes commited before this method is visible to +* vring_interrupt() which is called after this method. +* vdev: the virtio_device * @get_features: get the array of feature bits for this device. * vdev: the virtio_device * Returns the first 64 feature bits (all we currently need).
@@ -89,6 +93,7 @@ struct virtio_config_ops { const char * const names[], const bool *ctx, struct irq_affinity *desc);void (*del_vqs)(struct virtio_device *); +void (*synchronize_cbs)(struct virtio_device *);u64 (*get_features)(struct virtio_device *vdev);int (*finalize_features)(struct virtio_device *vdev);const char *(*bus_name)(struct virtio_device *vdev);
@@ -217,6 +222,25 @@ int virtio_find_vqs_ctx(struct virtio_device *vdev, unsigned nvqs, desc); } +/** +* virtio_synchronize_cbs - synchronize with virtqueue callbacks +* @vdev: the device +*/ +static inline +void virtio_synchronize_cbs (struct virtio_device *dev) +{ +if (dev->config->synchronize_cbs) { + dev->config->synchronize_cbs (dev); + }else { +/* +* A best effort fallback to synchronize with +* interrupts, preemption and softirq. See comment +* above synchronize_rcu(). +*/ +synchronize_rcu (); + } +} +/** * virtio_device_ready - enable vq use in probe function * @vdev: the device
Signature · 6 lines
2.25.1 _______________________________________________ Virtualization mailing list [email protected] https://lists.linuxfoundation.org/mailman/listinfo/virtualization
Cornelia Huck
maintainerRe: [PATCH V3 3/9] virtio: introduce config op to synchronize vring callbacks
Quoted 3 lines · Jason Wang
This patch introduces new virtio config op to vring callbacks. Transport specific method is required to make sure the write before this function is visible to the
vring_interrupt()that is
Which kind of writes? I.e., what is the scope?
Quoted 2 lines
called after the return of this function. For the transport that doesn't provide
synchronize_vqs(), usesynchornize_rcu()which
Typo: synchronize_rcu()
Quoted 18 lines
synchronize with IRQ implicitly as a fallback.
Cc Thomas Gleixner <[email protected]>Cc Peter Zijlstra <[email protected]>Cc "Paul E. McKenney" <[email protected]>Cc Marc Zyngier <[email protected]>Cc Halil Pasic <[email protected]>Cc Cornelia Huck <[email protected]>Signed-off-by Jason Wang <[email protected]>--- include/linux/virtio_config.h | 24 ++++++++++++++++++++++++ 1 file changed, 24 insertions(+)include/linux/virtio_config.h +0
-0
include/linux/virtio_config.h struct virtio_shm_region { · line 62
Quoted 6 lines
include/linux/virtio_config.h +3
-0@@ -57,6 +57,10 @@ struct virtio_shm_region {* include a NULL entry for vqs unused by driver * Returns 0 on success or error status * @del_vqs: free virtqueues found by find_vqs(). +* @synchronize_cbs: synchronize with the virtqueue callbacks (optional) +* Make sure the writes commited before this method is visible to +* vring_interrupt() which is called after this method.
Same here, I think the description needs to be a bit more explicit about which writes we care about here.
include/linux/virtio_config.h struct virtio_config_ops { · line 99
Quoted 12 lines
include/linux/virtio_config.h +2
-0+* vdev: the virtio_device * @get_features: get the array of feature bits for this device. * vdev: the virtio_device * Returns the first 64 feature bits (all we currently need). @@ -89,6 +93,7 @@ struct virtio_config_ops { const char * const names[], const bool *ctx, struct irq_affinity *desc);void (*del_vqs)(struct virtio_device *); +void (*synchronize_cbs)(struct virtio_device *);u64 (*get_features)(struct virtio_device *vdev);int (*finalize_features)(struct virtio_device *vdev);const char *(*bus_name)(struct virtio_device *vdev);
_______________________________________________
Virtualization mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/virtualization
Re: [PATCH V3 3/9] virtio: introduce config op to synchronize vring callbacks
Quoted 8 lines · Cornelia Huck
Quoted 3 lines · Jason Wang
This patch introduces new virtio config op to vring callbacks. Transport specific method is required to make sure the write before this function is visible to the
vring_interrupt()that isWhich kind of writes? I.e., what is the scope?
Any writes before synchronize_cbs(). Is something like the following better?
The function guarantees that all memory operations before it are visible to the vring_interrupt() that is called after it.
Quoted 5 lines
Quoted 3 lines
called after the return of this function. For the transport that doesn't provide
synchronize_vqs(), usesynchornize_rcu()whichTypo:
synchronize_rcu()
Will fix it.
Thanks
Quoted 42 lines
Quoted 19 lines
synchronize with IRQ implicitly as a fallback.
Cc Thomas Gleixner <[email protected]>Cc Peter Zijlstra <[email protected]>Cc "Paul E. McKenney" <[email protected]>Cc Marc Zyngier <[email protected]>Cc Halil Pasic <[email protected]>Cc Cornelia Huck <[email protected]>Signed-off-by Jason Wang <[email protected]>--- include/linux/virtio_config.h | 24 ++++++++++++++++++++++++ 1 file changed, 24 insertions(+)include/linux/virtio_config.h +0
-0include/linux/virtio_config.h struct virtio_shm_region { · line 62
Quoted 6 lines
include/linux/virtio_config.h +3
-0@@ -57,6 +57,10 @@ struct virtio_shm_region {* include a NULL entry for vqs unused by driver * Returns 0 on success or error status * @del_vqs: free virtqueues found by find_vqs(). +* @synchronize_cbs: synchronize with the virtqueue callbacks (optional) +* Make sure the writes commited before this method is visible to +* vring_interrupt() which is called after this method. Same here, I think the description needs to be a bit more explicit about which writes we care about here.
Quoted 12 lines
include/linux/virtio_config.h +2
-0+* vdev: the virtio_device * @get_features: get the array of feature bits for this device. * vdev: the virtio_device * Returns the first 64 feature bits (all we currently need). @@ -89,6 +93,7 @@ struct virtio_config_ops { const char * const names[], const bool *ctx, struct irq_affinity *desc);void (*del_vqs)(struct virtio_device *); +void (*synchronize_cbs)(struct virtio_device *);u64 (*get_features)(struct virtio_device *vdev);int (*finalize_features)(struct virtio_device *vdev);const char *(*bus_name)(struct virtio_device *vdev);
Cornelia Huck
maintainerRe: [PATCH V3 3/9] virtio: introduce config op to synchronize vring callbacks
Quoted 8 lines · Jason Wang
Quoted 7 lines · Cornelia Huck
Quoted 3 lines · Jason Wang
This patch introduces new virtio config op to vring callbacks. Transport specific method is required to make sure the write before this function is visible to the
vring_interrupt()that is
Quoted 6 lines
Quoted 1 line
Which kind of writes? I.e., what is the scope?
Any writes before
synchronize_cbs(). Is something like the following better?The function guarantees that all memory operations before it are visible to the
vring_interrupt()that is called after it.
Maybe "all memory operations on the queue"?
_______________________________________________
Virtualization mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/virtualization
Re: [PATCH V3 3/9] virtio: introduce config op to synchronize vring callbacks
Quoted 8 lines · Cornelia Huck
Quoted 6 lines · Jason Wang
Quoted 5 lines · Cornelia Huck
Quoted 3 lines · Jason Wang
This patch introduces new virtio config op to vring callbacks. Transport specific method is required to make sure the write before this function is visible to the
vring_interrupt()that is
Quoted 6 lines
Quoted 5 lines
Quoted 1 line
Which kind of writes? I.e., what is the scope?
Any writes before
synchronize_cbs(). Is something like the following better?The function guarantees that all memory operations before it are visible to the
vring_interrupt()that is called after it.Maybe "all memory operations on the queue"?
That's fine. Will do.
Thanks
Quoted 1 line
_______________________________________________
Virtualization mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/virtualization
PATCH v3 4/9 virtio-pci: implement synchronize_cbs()
In reply to Jason WangWe can simply reuse vp_synchronize_vectors() for .synchronize_cbs().
--- drivers/virtio/virtio_pci_common.h | 2 ++ drivers/virtio/virtio_pci_legacy.c | 1 + drivers/virtio/virtio_pci_modern.c | 2 ++ 3 files changed, 5 insertions(+)
drivers/virtio/virtio_pci_common.h +2 -0
@@ -105,6 +105,8 @@ static struct virtio_pci_device *to_vp_device(struct virtio_device *vdev)void vp_synchronize_vectors (struct virtio_device *vdev);/* the notify function used when creating a virt queue */ bool vp_notify (struct virtqueue *vq); +/* synchronize with callbacks */ +void vp_synchronize_vqs (struct virtio_device *vdev);/* the config->del_vqs() implementation */ void vp_del_vqs (struct virtio_device *vdev);/* the config->find_vqs() implementation */
drivers/virtio/virtio_pci_legacy.c +1 -0
@@ -192,6 +192,7 @@ static const struct virtio_config_ops virtio_pci_config_ops = { .reset = vp_reset, .find_vqs = vp_find_vqs, .del_vqs = vp_del_vqs, + .synchronize_cbs = vp_synchronize_vectors, .get_features = vp_get_features, .finalize_features = vp_finalize_features, .bus_name = vp_bus_name,
drivers/virtio/virtio_pci_modern.c +2 -0
@@ -394,6 +394,7 @@ static const struct virtio_config_ops virtio_pci_config_nodev_ops = { .reset = vp_reset, .find_vqs = vp_modern_find_vqs, .del_vqs = vp_del_vqs, + .synchronize_cbs = vp_synchronize_vectors, .get_features = vp_get_features, .finalize_features = vp_finalize_features, .bus_name = vp_bus_name,
@@ -411,6 +412,7 @@ static const struct virtio_config_ops virtio_pci_config_ops = { .reset = vp_reset, .find_vqs = vp_modern_find_vqs, .del_vqs = vp_del_vqs, + .synchronize_cbs = vp_synchronize_vectors, .get_features = vp_get_features, .finalize_features = vp_finalize_features, .bus_name = vp_bus_name,
Signature · 6 lines
2.25.1 _______________________________________________ Virtualization mailing list [email protected] https://lists.linuxfoundation.org/mailman/listinfo/virtualization
PATCH v3 5/9 virtio-mmio: implement synchronize_cbs()
In reply to Jason WangSimply synchronize the platform irq that is used by us.
--- drivers/virtio/virtio_mmio.c | 9 +++++++++ 1 file changed, 9 insertions(+)
drivers/virtio/virtio_mmio.c +9 -0
@@ -345,6 +345,14 @@ static void vm_del_vqs(struct virtio_device *vdev)free_irq (platform_get_irq (vm_dev->pdev,0 ), vm_dev); } + +static void vm_synchronize_cbs (struct virtio_device *vdev) +{ +struct virtio_mmio_device *vm_dev =to_virtio_mmio_device (vdev); + +synchronize_irq (platform_get_irq (vm_dev->pdev,0 )); +} +static struct virtqueue *vm_setup_vq (struct virtio_device *vdev,unsigned index,void (*callback)(struct virtqueue *vq),const char *name,bool ctx)
@@ -541,6 +549,7 @@ static const struct virtio_config_ops virtio_mmio_config_ops = { .finalize_features = vm_finalize_features, .bus_name = vm_bus_name, .get_shm_region = vm_get_shm_region, + .synchronize_cbs = vm_synchronize_cbs, };
Signature · 6 lines
2.25.1 _______________________________________________ Virtualization mailing list [email protected] https://lists.linuxfoundation.org/mailman/listinfo/virtualization
PATCH v3 6/9 virtio-ccw: implement synchronize_cbs()
In reply to Jason WangThis patch tries to implement the synchronize_cbs() for ccw. For the vring_interrupt() that is called via virtio_airq_handler(), the synchronization is simply done via the airq_info's lock. For the vring_interrupt() that is called via virtio_ccw_int_handler(), a per device spinlock for irq is introduced ans used in the synchronization method.
--- drivers/s390/virtio/virtio_ccw.c | 27 +++++++++++++++++++++++++++ 1 file changed, 27 insertions(+)
drivers/s390/virtio/virtio_ccw.c +27 -0
@@ -62,6 +62,7 @@ struct virtio_ccw_device {unsigned int revision;/* Transport revision */ wait_queue_head_t wait_q;spinlock_t lock; +spinlock_t irq_lock;struct mutex io_lock;/* Serializes I/O requests */ struct list_head virtqueues;bool is_thinint;
@@ -984,6 +985,27 @@ static const char *virtio_ccw_bus_name(struct virtio_device *vdev)return dev_name (&vcdev->cdev->dev); } +static void virtio_ccw_synchronize_cbs (struct virtio_device *vdev) +{ +struct virtio_ccw_device *vcdev =to_vc_device (vdev); +struct airq_info *info = vcdev->airq_info; + +/* +* Synchronize with the vring_interrupt() called by +* virtio_ccw_int_handler(). +*/ +spin_lock (&vcdev->irq_lock); +spin_unlock (&vcdev->irq_lock); + +if (info) { +/* +* Synchronize with the vring_interrupt() with airq indicator +*/ +write_lock (&info->lock); +write_unlock (&info->lock); + } +} +static const struct virtio_config_ops virtio_ccw_config_ops = { .get_features = virtio_ccw_get_features, .finalize_features = virtio_ccw_finalize_features,
@@ -995,6 +1017,7 @@ static const struct virtio_config_ops virtio_ccw_config_ops = { .find_vqs = virtio_ccw_find_vqs, .del_vqs = virtio_ccw_del_vqs, .bus_name = virtio_ccw_bus_name, + .synchronize_cbs = virtio_ccw_synchronize_cbs, };
@@ -1079,6 +1102,7 @@ static void virtio_ccw_int_handler(struct ccw_device *cdev, {__u32 activity = intparm &VIRTIO_CCW_INTPARM_MASK ;struct virtio_ccw_device *vcdev =dev_get_drvdata (&cdev->dev); +unsigned long flags;int i;struct virtqueue *vq;
@@ -1106,6 +1130,7 @@ static void virtio_ccw_int_handler(struct ccw_device *cdev, vcdev->err = -EIO ; }virtio_ccw_check_activity (vcdev, activity); +spin_lock_irqsave (&vcdev->irq_lock, flags);for_each_set_bit (i,indicators (vcdev),sizeof (*indicators (vcdev)) *BITS_PER_BYTE ) {/* The bit clear must happen before the vring kick. */
@@ -1114,6 +1139,7 @@ static void virtio_ccw_int_handler(struct ccw_device *cdev, vq =virtio_ccw_vq_by_ind (vcdev, i);vring_interrupt (0 , vq); } +spin_unlock_irqrestore (&vcdev->irq_lock, flags);if (test_bit (0 ,indicators2 (vcdev))) {virtio_config_changed (&vcdev->vdev);clear_bit (0 ,indicators2 (vcdev));
@@ -1284,6 +1310,7 @@ static int virtio_ccw_online(struct ccw_device *cdev)init_waitqueue_head (&vcdev->wait_q);INIT_LIST_HEAD (&vcdev->virtqueues);spin_lock_init (&vcdev->lock); +spin_lock_init (&vcdev->irq_lock);mutex_init (&vcdev->io_lock);spin_lock_irqsave (get_ccwdev_lock (cdev), flags);
Signature · 6 lines
2.25.1 _______________________________________________ Virtualization mailing list [email protected] https://lists.linuxfoundation.org/mailman/listinfo/virtualization
Michael S. Tsirkin
maintainerRe: [PATCH V3 6/9] virtio-ccw: implement synchronize_cbs()
Quoted 7 lines · Jason Wang
This patch tries to implement the
synchronize_cbs()for ccw. For thevring_interrupt()that is called viavirtio_airq_handler(), the synchronization is simply done via the airq_info's lock. For thevring_interrupt()that is called viavirtio_ccw_int_handler(), a per device spinlock for irq is introduced ans used in the synchronization method.
Quoted 7 lines
Cc Thomas Gleixner <[email protected]>Cc Peter Zijlstra <[email protected]>Cc "Paul E. McKenney" <[email protected]>Cc Marc Zyngier <[email protected]>Cc Halil Pasic <[email protected]>Cc Cornelia Huck <[email protected]>Signed-off-by Jason Wang <[email protected]>
This is the only one that is giving me pause. Halil, Cornelia, should we be concerned about the performance impact here? Any chance it can be tested?
Quoted 86 lines
--- drivers/s390/virtio/virtio_ccw.c | 27 +++++++++++++++++++++++++++ 1 file changed, 27 insertions(+)drivers/s390/virtio/virtio_ccw.c +27
-0@@ -62,6 +62,7 @@ struct virtio_ccw_device {unsigned int revision;/* Transport revision */ wait_queue_head_t wait_q;spinlock_t lock; +spinlock_t irq_lock;struct mutex io_lock;/* Serializes I/O requests */ struct list_head virtqueues;bool is_thinint;@@ -984,6 +985,27 @@ static const char *virtio_ccw_bus_name(struct virtio_device *vdev)return dev_name (&vcdev->cdev->dev); } +static void virtio_ccw_synchronize_cbs (struct virtio_device *vdev) +{ +struct virtio_ccw_device *vcdev =to_vc_device (vdev); +struct airq_info *info = vcdev->airq_info; + +/* +* Synchronize with the vring_interrupt() called by +* virtio_ccw_int_handler(). +*/ +spin_lock (&vcdev->irq_lock); +spin_unlock (&vcdev->irq_lock); + +if (info) { +/* +* Synchronize with the vring_interrupt() with airq indicator +*/ +write_lock (&info->lock); +write_unlock (&info->lock); + } +} +static const struct virtio_config_ops virtio_ccw_config_ops = { .get_features = virtio_ccw_get_features, .finalize_features = virtio_ccw_finalize_features,@@ -995,6 +1017,7 @@ static const struct virtio_config_ops virtio_ccw_config_ops = { .find_vqs = virtio_ccw_find_vqs, .del_vqs = virtio_ccw_del_vqs, .bus_name = virtio_ccw_bus_name, + .synchronize_cbs = virtio_ccw_synchronize_cbs, };@@ -1079,6 +1102,7 @@ static void virtio_ccw_int_handler(struct ccw_device *cdev, {__u32 activity = intparm &VIRTIO_CCW_INTPARM_MASK ;struct virtio_ccw_device *vcdev =dev_get_drvdata (&cdev->dev); +unsigned long flags;int i;struct virtqueue *vq;@@ -1106,6 +1130,7 @@ static void virtio_ccw_int_handler(struct ccw_device *cdev, vcdev->err = -EIO ; }virtio_ccw_check_activity (vcdev, activity); +spin_lock_irqsave (&vcdev->irq_lock, flags);for_each_set_bit (i,indicators (vcdev),sizeof (*indicators (vcdev)) *BITS_PER_BYTE ) {/* The bit clear must happen before the vring kick. */ @@ -1114,6 +1139,7 @@ static void virtio_ccw_int_handler(struct ccw_device *cdev, vq =virtio_ccw_vq_by_ind (vcdev, i);vring_interrupt (0 , vq); } +spin_unlock_irqrestore (&vcdev->irq_lock, flags);if (test_bit (0 ,indicators2 (vcdev))) {virtio_config_changed (&vcdev->vdev);clear_bit (0 ,indicators2 (vcdev));@@ -1284,6 +1310,7 @@ static int virtio_ccw_online(struct ccw_device *cdev)init_waitqueue_head (&vcdev->wait_q);INIT_LIST_HEAD (&vcdev->virtqueues);spin_lock_init (&vcdev->lock); +spin_lock_init (&vcdev->irq_lock);mutex_init (&vcdev->io_lock);spin_lock_irqsave (get_ccwdev_lock (cdev), flags);Signature · 1 line
2.25.1
_______________________________________________
Virtualization mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/virtualization
Cornelia Huck
maintainerRe: [PATCH V3 6/9] virtio-ccw: implement synchronize_cbs()
Quoted 17 lines · Michael S. Tsirkin
Quoted 7 lines · Jason Wang
This patch tries to implement the
synchronize_cbs()for ccw. For thevring_interrupt()that is called viavirtio_airq_handler(), the synchronization is simply done via the airq_info's lock. For thevring_interrupt()that is called viavirtio_ccw_int_handler(), a per device spinlock for irq is introduced ans used in the synchronization method.Quoted 7 lines
Cc Thomas Gleixner <[email protected]>Cc Peter Zijlstra <[email protected]>Cc "Paul E. McKenney" <[email protected]>Cc Marc Zyngier <[email protected]>Cc Halil Pasic <[email protected]>Cc Cornelia Huck <[email protected]>Signed-off-by Jason Wang <[email protected]>
Quoted 3 lines
This is the only one that is giving me pause. Halil, Cornelia, should we be concerned about the performance impact here? Any chance it can be tested?
We can have a bunch of devices using the same airq structure, and the sync cb creates a choke point, same as registering/unregistering. If invoking the sync cb is a rare operation (same as (un)registering), it should not affect interrupt processing for other devices too much, but it really should be rare.
For testing, you would probably want to use a setup with many devices that share the same airq area (you can fit a lot of devices if they have few queues), generate traffic on the queues, and then do something that triggers the callback (adding/removing a new device in a loop?)
I currently don't have such a setup handy; Halil, would you be able to test that?
Quoted 32 lines
Quoted 32 lines
--- drivers/s390/virtio/virtio_ccw.c | 27 +++++++++++++++++++++++++++ 1 file changed, 27 insertions(+)drivers/s390/virtio/virtio_ccw.c +12
-0@@ -62,6 +62,7 @@ struct virtio_ccw_device {unsigned int revision;/* Transport revision */ wait_queue_head_t wait_q;spinlock_t lock; +spinlock_t irq_lock;struct mutex io_lock;/* Serializes I/O requests */ struct list_head virtqueues;bool is_thinint;@@ -984,6 +985,27 @@ static const char *virtio_ccw_bus_name(struct virtio_device *vdev)return dev_name (&vcdev->cdev->dev); } +static void virtio_ccw_synchronize_cbs (struct virtio_device *vdev) +{ +struct virtio_ccw_device *vcdev =to_vc_device (vdev); +struct airq_info *info = vcdev->airq_info; + +/* +* Synchronize with the vring_interrupt() called by +* virtio_ccw_int_handler(). +*/ +spin_lock (&vcdev->irq_lock); +spin_unlock (&vcdev->irq_lock);
Quoted 8 lines
drivers/s390/virtio/virtio_ccw.c static const char *virtio_ccw_bus_name(struct virtio_device *vdev) · line 1006
Quoted 8 lines
drivers/s390/virtio/virtio_ccw.c +8
-0+ +if (info) { +/* +* Synchronize with the vring_interrupt() with airq indicator +*/ +write_lock (&info->lock); +write_unlock (&info->lock); + }
I think we can make this an either/or operation (devices will either use classic interrupts or adapter interrupts)?
Quoted 5 lines
drivers/s390/virtio/virtio_ccw.c static const char *virtio_ccw_bus_name(struct virtio_device *vdev) · line 1011
Quoted 5 lines
drivers/s390/virtio/virtio_ccw.c +2
-0+} +static const struct virtio_config_ops virtio_ccw_config_ops = { .get_features = virtio_ccw_get_features, .finalize_features = virtio_ccw_finalize_features,
_______________________________________________
Virtualization mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/virtualization
Michael S. Tsirkin
maintainerRe: [PATCH V3 6/9] virtio-ccw: implement synchronize_cbs()
Quoted 19 lines · Cornelia Huck
Quoted 17 lines · Michael S. Tsirkin
Quoted 14 lines · Jason Wang
This patch tries to implement the
synchronize_cbs()for ccw. For thevring_interrupt()that is called viavirtio_airq_handler(), the synchronization is simply done via the airq_info's lock. For thevring_interrupt()that is called viavirtio_ccw_int_handler(), a per device spinlock for irq is introduced ans used in the synchronization method.Cc Thomas Gleixner <[email protected]>Cc Peter Zijlstra <[email protected]>Cc "Paul E. McKenney" <[email protected]>Cc Marc Zyngier <[email protected]>Cc Halil Pasic <[email protected]>Cc Cornelia Huck <[email protected]>Signed-off-by Jason Wang <[email protected]>
Quoted 6 lines
Quoted 3 lines
This is the only one that is giving me pause. Halil, Cornelia, should we be concerned about the performance impact here? Any chance it can be tested?
We can have a bunch of devices using the same airq structure, and the sync cb creates a choke point, same as registering/unregistering.
BTW can callbacks for multiple VQs run on multiple CPUs at the moment? this patch serializes them on a spinlock.
Quoted 62 lines
If
invoking the sync cb is a rare operation (same as (un)registering), it should not affect interrupt processing for other devices too much, but it really should be rare.For testing, you would probably want to use a setup with many devices that share the same airq area (you can fit a lot of devices if they have few queues), generate traffic on the queues, and then do something that triggers the callback (adding/removing a new device in a loop?)
I currently don't have such a setup handy; Halil, would you be able to test that?
Quoted 32 lines
Quoted 32 lines
--- drivers/s390/virtio/virtio_ccw.c | 27 +++++++++++++++++++++++++++ 1 file changed, 27 insertions(+)drivers/s390/virtio/virtio_ccw.c +12
-0@@ -62,6 +62,7 @@ struct virtio_ccw_device {unsigned int revision;/* Transport revision */ wait_queue_head_t wait_q;spinlock_t lock; +spinlock_t irq_lock;struct mutex io_lock;/* Serializes I/O requests */ struct list_head virtqueues;bool is_thinint;@@ -984,6 +985,27 @@ static const char *virtio_ccw_bus_name(struct virtio_device *vdev)return dev_name (&vcdev->cdev->dev); } +static void virtio_ccw_synchronize_cbs (struct virtio_device *vdev) +{ +struct virtio_ccw_device *vcdev =to_vc_device (vdev); +struct airq_info *info = vcdev->airq_info; + +/* +* Synchronize with the vring_interrupt() called by +* virtio_ccw_int_handler(). +*/ +spin_lock (&vcdev->irq_lock); +spin_unlock (&vcdev->irq_lock);Quoted 8 lines
drivers/s390/virtio/virtio_ccw.c static const char *virtio_ccw_bus_name(struct virtio_device *vdev) · line 1006
Quoted 8 lines
drivers/s390/virtio/virtio_ccw.c +8
-0+ +if (info) { +/* +* Synchronize with the vring_interrupt() with airq indicator +*/ +write_lock (&info->lock); +write_unlock (&info->lock); + }I think we can make this an either/or operation (devices will either use classic interrupts or adapter interrupts)?
Quoted 5 lines
Quoted 5 lines
drivers/s390/virtio/virtio_ccw.c +2
-0+} +static const struct virtio_config_ops virtio_ccw_config_ops = { .get_features = virtio_ccw_get_features, .finalize_features = virtio_ccw_finalize_features,
Halil Pasic
maintainerRe: [PATCH V3 6/9] virtio-ccw: implement synchronize_cbs()
Quoted 20 lines · On Mon, 25 Apr 2022 09:59:55 -0400 "Michael S. Tsirkin" <[email protected]> wrote:
Quoted 19 lines · Cornelia Huck
Quoted 17 lines · Michael S. Tsirkin
Quoted 14 lines · Jason Wang
This patch tries to implement the
synchronize_cbs()for ccw. For thevring_interrupt()that is called viavirtio_airq_handler(), the synchronization is simply done via the airq_info's lock. For thevring_interrupt()that is called viavirtio_ccw_int_handler(), a per device spinlock for irq is introduced ans used in the synchronization method.Cc Thomas Gleixner <[email protected]>Cc Peter Zijlstra <[email protected]>Cc "Paul E. McKenney" <[email protected]>Cc Marc Zyngier <[email protected]>Cc Halil Pasic <[email protected]>Cc Cornelia Huck <[email protected]>Signed-off-by Jason Wang <[email protected]>
Quoted 8 lines
Quoted 6 lines
Quoted 3 lines
This is the only one that is giving me pause. Halil, Cornelia, should we be concerned about the performance impact here? Any chance it can be tested?
We can have a bunch of devices using the same airq structure, and the sync cb creates a choke point, same as registering/unregistering.
BTW can callbacks for multiple VQs run on multiple CPUs at the moment?
I'm not sure I understand the question.
I do think we can have multiple CPUs that are executing some portion of virtio_ccw_int_handler(). So I guess the answer is yes. Connie what do you think?
On the other hand we could also end up serializing synchronize_cbs() calls for different devices if they happen to use the same airq_info. But this probably was not your question
Quoted 2 lines
this patch serializes them on a spinlock.
Those could then pile up on the newly introduced spinlock.
Regards,
Halil
_______________________________________________
Virtualization mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/virtualization
Michael S. Tsirkin
maintainerRe: [PATCH V3 6/9] virtio-ccw: implement synchronize_cbs()
Quoted 32 lines · Halil Pasic
Quoted 20 lines · On Mon, 25 Apr 2022 09:59:55 -0400 "Michael S. Tsirkin" <[email protected]> wrote:
Quoted 19 lines · Cornelia Huck
Quoted 17 lines · Michael S. Tsirkin
Quoted 14 lines · Jason Wang
This patch tries to implement the
synchronize_cbs()for ccw. For thevring_interrupt()that is called viavirtio_airq_handler(), the synchronization is simply done via the airq_info's lock. For thevring_interrupt()that is called viavirtio_ccw_int_handler(), a per device spinlock for irq is introduced ans used in the synchronization method.Cc Thomas Gleixner <[email protected]>Cc Peter Zijlstra <[email protected]>Cc "Paul E. McKenney" <[email protected]>Cc Marc Zyngier <[email protected]>Cc Halil Pasic <[email protected]>Cc Cornelia Huck <[email protected]>Signed-off-by Jason Wang <[email protected]>Quoted 8 lines
Quoted 6 lines
Quoted 3 lines
This is the only one that is giving me pause. Halil, Cornelia, should we be concerned about the performance impact here? Any chance it can be tested?
We can have a bunch of devices using the same airq structure, and the sync cb creates a choke point, same as registering/unregistering.
BTW can callbacks for multiple VQs run on multiple CPUs at the moment?
Quoted 8 lines
I'm not sure I understand the question.
I do think we can have multiple CPUs that are executing some portion of
virtio_ccw_int_handler(). So I guess the answer is yes. Connie what do you think?On the other hand we could also end up serializing
synchronize_cbs()calls for different devices if they happen to use the sameairq_info. But this probably was not your question
I am less concerned about synchronize_cbs being slow and more about the slowdown in interrupt processing itself.
Quoted 7 lines
Quoted 2 lines
this patch serializes them on a spinlock.
Those could then pile up on the newly introduced spinlock.
Regards,
Halil
Hmm yea ... not good.
Signature · 6 lines
MST _______________________________________________ Virtualization mailing list [email protected] https://lists.linuxfoundation.org/mailman/listinfo/virtualization
Michael S. Tsirkin
maintainerRe: [PATCH V3 6/9] virtio-ccw: implement synchronize_cbs()
Quoted 49 lines · Michael S. Tsirkin
Quoted 32 lines · Halil Pasic
Quoted 20 lines · On Mon, 25 Apr 2022 09:59:55 -0400 "Michael S. Tsirkin" <[email protected]> wrote:
Quoted 19 lines · Cornelia Huck
Quoted 17 lines · Michael S. Tsirkin
Quoted 14 lines · Jason Wang
This patch tries to implement the
synchronize_cbs()for ccw. For thevring_interrupt()that is called viavirtio_airq_handler(), the synchronization is simply done via the airq_info's lock. For thevring_interrupt()that is called viavirtio_ccw_int_handler(), a per device spinlock for irq is introduced ans used in the synchronization method.Cc Thomas Gleixner <[email protected]>Cc Peter Zijlstra <[email protected]>Cc "Paul E. McKenney" <[email protected]>Cc Marc Zyngier <[email protected]>Cc Halil Pasic <[email protected]>Cc Cornelia Huck <[email protected]>Signed-off-by Jason Wang <[email protected]>Quoted 8 lines
Quoted 6 lines
Quoted 3 lines
This is the only one that is giving me pause. Halil, Cornelia, should we be concerned about the performance impact here? Any chance it can be tested?
We can have a bunch of devices using the same airq structure, and the sync cb creates a choke point, same as registering/unregistering.
BTW can callbacks for multiple VQs run on multiple CPUs at the moment?
Quoted 8 lines
I'm not sure I understand the question.
I do think we can have multiple CPUs that are executing some portion of
virtio_ccw_int_handler(). So I guess the answer is yes. Connie what do you think?On the other hand we could also end up serializing
synchronize_cbs()calls for different devices if they happen to use the sameairq_info. But this probably was not your questionI am less concerned about
synchronize_cbsbeing slow and more about the slowdown in interrupt processing itself.Quoted 3 lines
Quoted 2 lines
this patch serializes them on a spinlock.
Quoted 6 lines
Quoted 4 lines
Those could then pile up on the newly introduced spinlock.
Regards,
HalilHmm yea ... not good.
Is there any other way to synchronize with all callbacks?
Quoted 2 lines
Signature · 1 line
MST
Re: [PATCH V3 6/9] virtio-ccw: implement synchronize_cbs()
Quoted 42 lines · Michael S. Tsirkin
Quoted 41 lines · Michael S. Tsirkin
Quoted 28 lines · Halil Pasic
Quoted 19 lines · On Mon, 25 Apr 2022 09:59:55 -0400 "Michael S. Tsirkin" <[email protected]> wrote:
Quoted 18 lines · Cornelia Huck
Quoted 16 lines · Michael S. Tsirkin
Quoted 14 lines · Jason Wang
This patch tries to implement the
synchronize_cbs()for ccw. For thevring_interrupt()that is called viavirtio_airq_handler(), the synchronization is simply done via the airq_info's lock. For thevring_interrupt()that is called viavirtio_ccw_int_handler(), a per device spinlock for irq is introduced ans used in the synchronization method.Cc Thomas Gleixner <[email protected]>Cc Peter Zijlstra <[email protected]>Cc "Paul E. McKenney" <[email protected]>Cc Marc Zyngier <[email protected]>Cc Halil Pasic <[email protected]>Cc Cornelia Huck <[email protected]>Signed-off-by Jason Wang <[email protected]>Quoted 6 lines
Quoted 5 lines
Quoted 3 lines
This is the only one that is giving me pause. Halil, Cornelia, should we be concerned about the performance impact here? Any chance it can be tested?
We can have a bunch of devices using the same airq structure, and the sync cb creates a choke point, same as registering/unregistering.
BTW can callbacks for multiple VQs run on multiple CPUs at the moment?
Quoted 8 lines
I'm not sure I understand the question.
I do think we can have multiple CPUs that are executing some portion of
virtio_ccw_int_handler(). So I guess the answer is yes. Connie what do you think?On the other hand we could also end up serializing
synchronize_cbs()calls for different devices if they happen to use the sameairq_info. But this probably was not your questionI am less concerned about
synchronize_cbsbeing slow and more about the slowdown in interrupt processing itself.
Quoted 8 lines
Quoted 7 lines
Quoted 6 lines
Quoted 2 lines
this patch serializes them on a spinlock.
Those could then pile up on the newly introduced spinlock.
Regards,
HalilHmm yea ... not good.
Is there any other way to synchronize with all callbacks?
Maybe using rwlock as airq handler?
Thanks
Quoted 3 lines
Quoted 3 lines
Signature · 2 lines
-- MST
_______________________________________________
Virtualization mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/virtualization
Michael S. Tsirkin
maintainerRe: [PATCH V3 6/9] virtio-ccw: implement synchronize_cbs()
Quoted 49 lines · Jason Wang
Quoted 47 lines · Michael S. Tsirkin
Quoted 46 lines · Michael S. Tsirkin
Quoted 27 lines · Halil Pasic
Quoted 18 lines · On Mon, 25 Apr 2022 09:59:55 -0400 "Michael S. Tsirkin" <[email protected]> wrote:
Quoted 17 lines · Cornelia Huck
Quoted 16 lines · Michael S. Tsirkin
Quoted 14 lines · Jason Wang
This patch tries to implement the
synchronize_cbs()for ccw. For thevring_interrupt()that is called viavirtio_airq_handler(), the synchronization is simply done via the airq_info's lock. For thevring_interrupt()that is called viavirtio_ccw_int_handler(), a per device spinlock for irq is introduced ans used in the synchronization method.Cc Thomas Gleixner <[email protected]>Cc Peter Zijlstra <[email protected]>Cc "Paul E. McKenney" <[email protected]>Cc Marc Zyngier <[email protected]>Cc Halil Pasic <[email protected]>Cc Cornelia Huck <[email protected]>Signed-off-by Jason Wang <[email protected]>Quoted 6 lines
Quoted 5 lines
Quoted 3 lines
This is the only one that is giving me pause. Halil, Cornelia, should we be concerned about the performance impact here? Any chance it can be tested?
We can have a bunch of devices using the same airq structure, and the sync cb creates a choke point, same as registering/unregistering.
BTW can callbacks for multiple VQs run on multiple CPUs at the moment?
Quoted 8 lines
I'm not sure I understand the question.
I do think we can have multiple CPUs that are executing some portion of
virtio_ccw_int_handler(). So I guess the answer is yes. Connie what do you think?On the other hand we could also end up serializing
synchronize_cbs()calls for different devices if they happen to use the sameairq_info. But this probably was not your questionI am less concerned about
synchronize_cbsbeing slow and more about the slowdown in interrupt processing itself.Quoted 6 lines
Quoted 2 lines
this patch serializes them on a spinlock.
Those could then pile up on the newly introduced spinlock.
Regards,
Halil
Quoted 8 lines
Quoted 2 lines
Quoted 1 line
Hmm yea ... not good.
Is there any other way to synchronize with all callbacks?
Maybe using rwlock as airq handler?
Thanks
rwlock is still a shared cacheline bouncing between CPUs and a bunch of ordering instructions.
Maybe something per-cpu + some IPIs to run things on all CPUs instead?
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
Signature · 2 lines
-- MST
_______________________________________________
Virtualization mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/virtualization
Michael S. Tsirkin
maintainerRe: [PATCH V3 6/9] virtio-ccw: implement synchronize_cbs()
Quoted 54 lines · Michael S. Tsirkin
Quoted 53 lines · Jason Wang
Quoted 49 lines · Michael S. Tsirkin
Quoted 40 lines · Michael S. Tsirkin
Quoted 27 lines · Halil Pasic
Quoted 18 lines · On Mon, 25 Apr 2022 09:59:55 -0400 "Michael S. Tsirkin" <[email protected]> wrote:
Quoted 17 lines · Cornelia Huck
Quoted 16 lines · Michael S. Tsirkin
Quoted 14 lines · Jason Wang
This patch tries to implement the
synchronize_cbs()for ccw. For thevring_interrupt()that is called viavirtio_airq_handler(), the synchronization is simply done via the airq_info's lock. For thevring_interrupt()that is called viavirtio_ccw_int_handler(), a per device spinlock for irq is introduced ans used in the synchronization method.Cc Thomas Gleixner <[email protected]>Cc Peter Zijlstra <[email protected]>Cc "Paul E. McKenney" <[email protected]>Cc Marc Zyngier <[email protected]>Cc Halil Pasic <[email protected]>Cc Cornelia Huck <[email protected]>Signed-off-by Jason Wang <[email protected]>Quoted 6 lines
Quoted 5 lines
Quoted 3 lines
This is the only one that is giving me pause. Halil, Cornelia, should we be concerned about the performance impact here? Any chance it can be tested?
We can have a bunch of devices using the same airq structure, and the sync cb creates a choke point, same as registering/unregistering.
BTW can callbacks for multiple VQs run on multiple CPUs at the moment?
Quoted 8 lines
I'm not sure I understand the question.
I do think we can have multiple CPUs that are executing some portion of
virtio_ccw_int_handler(). So I guess the answer is yes. Connie what do you think?On the other hand we could also end up serializing
synchronize_cbs()calls for different devices if they happen to use the sameairq_info. But this probably was not your questionI am less concerned about
synchronize_cbsbeing slow and more about the slowdown in interrupt processing itself.Quoted 7 lines
Quoted 6 lines
Quoted 2 lines
this patch serializes them on a spinlock.
Those could then pile up on the newly introduced spinlock.
Regards,
HalilHmm yea ... not good.
Is there any other way to synchronize with all callbacks?
Quoted 8 lines
Quoted 4 lines
Maybe using rwlock as airq handler?
Thanks
rwlock is still a shared cacheline bouncing between CPUs and a bunch of ordering instructions.
Maybe something per-cpu + some IPIs to run things on all CPUs instead?
... and I think classic and device interrupts are different enough here ...
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
Signature · 2 lines
-- MST
_______________________________________________
Virtualization mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/virtualization
Re: [PATCH V3 6/9] virtio-ccw: implement synchronize_cbs()
Quoted 56 lines · Michael S. Tsirkin
Quoted 54 lines · Michael S. Tsirkin
Quoted 53 lines · Jason Wang
Quoted 49 lines · Michael S. Tsirkin
Quoted 40 lines · Michael S. Tsirkin
Quoted 27 lines · Halil Pasic
Quoted 18 lines · On Mon, 25 Apr 2022 09:59:55 -0400 "Michael S. Tsirkin" <[email protected]> wrote:
Quoted 17 lines · Cornelia Huck
Quoted 16 lines · Michael S. Tsirkin
Quoted 14 lines · Jason Wang
This patch tries to implement the
synchronize_cbs()for ccw. For thevring_interrupt()that is called viavirtio_airq_handler(), the synchronization is simply done via the airq_info's lock. For thevring_interrupt()that is called viavirtio_ccw_int_handler(), a per device spinlock for irq is introduced ans used in the synchronization method.Cc Thomas Gleixner <[email protected]>Cc Peter Zijlstra <[email protected]>Cc "Paul E. McKenney" <[email protected]>Cc Marc Zyngier <[email protected]>Cc Halil Pasic <[email protected]>Cc Cornelia Huck <[email protected]>Signed-off-by Jason Wang <[email protected]>Quoted 6 lines
Quoted 5 lines
Quoted 3 lines
This is the only one that is giving me pause. Halil, Cornelia, should we be concerned about the performance impact here? Any chance it can be tested?
We can have a bunch of devices using the same airq structure, and the sync cb creates a choke point, same as registering/unregistering.
BTW can callbacks for multiple VQs run on multiple CPUs at the moment?
Quoted 8 lines
I'm not sure I understand the question.
I do think we can have multiple CPUs that are executing some portion of
virtio_ccw_int_handler(). So I guess the answer is yes. Connie what do you think?On the other hand we could also end up serializing
synchronize_cbs()calls for different devices if they happen to use the sameairq_info. But this probably was not your questionI am less concerned about
synchronize_cbsbeing slow and more about the slowdown in interrupt processing itself.Quoted 7 lines
Quoted 6 lines
Quoted 2 lines
this patch serializes them on a spinlock.
Those could then pile up on the newly introduced spinlock.
Regards,
HalilHmm yea ... not good.
Is there any other way to synchronize with all callbacks?
Quoted 7 lines
Quoted 7 lines
Quoted 4 lines
Maybe using rwlock as airq handler?
Thanks
rwlock is still a shared cacheline bouncing between CPUs and a bunch of ordering instructions.
Yes, but it should be faster than spinlocks anyhow.
Quoted 1 line
Quoted 1 line
Maybe something per-cpu + some IPIs to run things on all CPUs instead?
Is this something like a customized version of synchronzie_rcu_expedited()?
Quoted 3 lines
... and I think classic and device interrupts are different enough here ...
Yes.
Thanks
Quoted 5 lines
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
--
MST
_______________________________________________
Virtualization mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/virtualization
Michael S. Tsirkin
maintainerRe: [PATCH V3 6/9] virtio-ccw: implement synchronize_cbs()
Quoted 62 lines · Jason Wang
Quoted 61 lines · Michael S. Tsirkin
Quoted 59 lines · Michael S. Tsirkin
Quoted 57 lines · Jason Wang
Quoted 41 lines · Michael S. Tsirkin
Quoted 40 lines · Michael S. Tsirkin
Quoted 27 lines · Halil Pasic
Quoted 18 lines · On Mon, 25 Apr 2022 09:59:55 -0400 "Michael S. Tsirkin" <[email protected]> wrote:
Quoted 17 lines · Cornelia Huck
Quoted 16 lines · Michael S. Tsirkin
Quoted 14 lines · Jason Wang
This patch tries to implement the
synchronize_cbs()for ccw. For thevring_interrupt()that is called viavirtio_airq_handler(), the synchronization is simply done via the airq_info's lock. For thevring_interrupt()that is called viavirtio_ccw_int_handler(), a per device spinlock for irq is introduced ans used in the synchronization method.Cc Thomas Gleixner <[email protected]>Cc Peter Zijlstra <[email protected]>Cc "Paul E. McKenney" <[email protected]>Cc Marc Zyngier <[email protected]>Cc Halil Pasic <[email protected]>Cc Cornelia Huck <[email protected]>Signed-off-by Jason Wang <[email protected]>Quoted 6 lines
Quoted 5 lines
Quoted 3 lines
This is the only one that is giving me pause. Halil, Cornelia, should we be concerned about the performance impact here? Any chance it can be tested?
We can have a bunch of devices using the same airq structure, and the sync cb creates a choke point, same as registering/unregistering.
BTW can callbacks for multiple VQs run on multiple CPUs at the moment?
Quoted 8 lines
I'm not sure I understand the question.
I do think we can have multiple CPUs that are executing some portion of
virtio_ccw_int_handler(). So I guess the answer is yes. Connie what do you think?On the other hand we could also end up serializing
synchronize_cbs()calls for different devices if they happen to use the sameairq_info. But this probably was not your questionI am less concerned about
synchronize_cbsbeing slow and more about the slowdown in interrupt processing itself.Quoted 8 lines
Quoted 7 lines
Quoted 6 lines
Quoted 2 lines
this patch serializes them on a spinlock.
Those could then pile up on the newly introduced spinlock.
Regards,
HalilHmm yea ... not good.
Is there any other way to synchronize with all callbacks?
Maybe using rwlock as airq handler?
Thanks
Quoted 8 lines
Quoted 2 lines
Quoted 2 lines
rwlock is still a shared cacheline bouncing between CPUs and a bunch of ordering instructions.
Yes, but it should be faster than spinlocks anyhow.
Quoted 1 line
Quoted 1 line
Maybe something per-cpu + some IPIs to run things on all CPUs instead?
Is this something like a customized version of
synchronzie_rcu_expedited()?
With interrupts running in an RCU read size critical section? Quite possibly that is also an option.
This will need a bunch of documentation since this is not a standard use of RCU, and probably get a confirmation from RCU maintainers that whatever assumptions we make are guaranteed to hold down the road.
Quoted 13 lines
Quoted 3 lines
... and I think classic and device interrupts are different enough here ...
Yes.
Thanks
Quoted 5 lines
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
--
MST
Re: [PATCH V3 6/9] virtio-ccw: implement synchronize_cbs()
Quoted 67 lines · Michael S. Tsirkin
Quoted 65 lines · Jason Wang
Quoted 63 lines · Michael S. Tsirkin
Quoted 61 lines · Michael S. Tsirkin
Quoted 49 lines · Jason Wang
Quoted 47 lines · Michael S. Tsirkin
Quoted 46 lines · Michael S. Tsirkin
Quoted 27 lines · Halil Pasic
Quoted 18 lines · On Mon, 25 Apr 2022 09:59:55 -0400 "Michael S. Tsirkin" <[email protected]> wrote:
Quoted 17 lines · Cornelia Huck
Quoted 16 lines · Michael S. Tsirkin
Quoted 14 lines · Jason Wang
This patch tries to implement the
synchronize_cbs()for ccw. For thevring_interrupt()that is called viavirtio_airq_handler(), the synchronization is simply done via the airq_info's lock. For thevring_interrupt()that is called viavirtio_ccw_int_handler(), a per device spinlock for irq is introduced ans used in the synchronization method.Cc Thomas Gleixner <[email protected]>Cc Peter Zijlstra <[email protected]>Cc "Paul E. McKenney" <[email protected]>Cc Marc Zyngier <[email protected]>Cc Halil Pasic <[email protected]>Cc Cornelia Huck <[email protected]>Signed-off-by Jason Wang <[email protected]>Quoted 6 lines
Quoted 5 lines
Quoted 3 lines
This is the only one that is giving me pause. Halil, Cornelia, should we be concerned about the performance impact here? Any chance it can be tested?
We can have a bunch of devices using the same airq structure, and the sync cb creates a choke point, same as registering/unregistering.
BTW can callbacks for multiple VQs run on multiple CPUs at the moment?
Quoted 8 lines
I'm not sure I understand the question.
I do think we can have multiple CPUs that are executing some portion of
virtio_ccw_int_handler(). So I guess the answer is yes. Connie what do you think?On the other hand we could also end up serializing
synchronize_cbs()calls for different devices if they happen to use the sameairq_info. But this probably was not your questionI am less concerned about
synchronize_cbsbeing slow and more about the slowdown in interrupt processing itself.Quoted 6 lines
Quoted 2 lines
this patch serializes them on a spinlock.
Those could then pile up on the newly introduced spinlock.
Regards,
HalilQuoted 8 lines
Quoted 2 lines
Quoted 1 line
Hmm yea ... not good.
Is there any other way to synchronize with all callbacks?
Maybe using rwlock as airq handler?
Thanks
rwlock is still a shared cacheline bouncing between CPUs and a bunch of ordering instructions.
Quoted 7 lines
Quoted 5 lines
Yes, but it should be faster than spinlocks anyhow.
Quoted 1 line
Quoted 1 line
Maybe something per-cpu + some IPIs to run things on all CPUs instead?
Is this something like a customized version of
synchronzie_rcu_expedited()?With interrupts running in an RCU read size critical section?
For vring_interrupt(), yes.
Quoted 3 lines
Quite possibly that is also an option.
This will need a bunch of documentation since this is not a standard use of RCU,
According to Documentation/RCU/requirements.rst, it looks like a legal case:
"
The Linux kernel has interrupts, and RCU read-side critical sections are legal within interrupt handlers and within interrupt-disabled regions of code, as are invocations of call_rcu().
"
And as discussed, synchronize_rcu_expedited() is not friendly to real time workload. I think we can simply
1) protect vring_interrupt() with rcu_read_lock()
2) use synchronize_rcu() in synchronize_cbs for ccw
And if we care about the long delay we can use per device srcu to reduce that?
Thanks
Quoted 18 lines
and probably get a confirmation
from RCU maintainers that whatever assumptions we make are guaranteed to hold down the road.Quoted 13 lines
Quoted 3 lines
... and I think classic and device interrupts are different enough here ...
Yes.
Thanks
Quoted 5 lines
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
--
MST
Michael S. Tsirkin
maintainerRe: [PATCH V3 6/9] virtio-ccw: implement synchronize_cbs()
Quoted 85 lines · Jason Wang
Quoted 67 lines · Michael S. Tsirkin
Quoted 65 lines · Jason Wang
Quoted 63 lines · Michael S. Tsirkin
Quoted 61 lines · Michael S. Tsirkin
Quoted 49 lines · Jason Wang
Quoted 47 lines · Michael S. Tsirkin
Quoted 46 lines · Michael S. Tsirkin
Quoted 27 lines · Halil Pasic
Quoted 18 lines · On Mon, 25 Apr 2022 09:59:55 -0400 "Michael S. Tsirkin" <[email protected]> wrote:
Quoted 17 lines · Cornelia Huck
Quoted 16 lines · Michael S. Tsirkin
Quoted 14 lines · Jason Wang
This patch tries to implement the synchronize_cbs() for ccw. For the
vring_interrupt() that is called via virtio_airq_handler(), the
synchronization is simply done via the airq_info's lock. For the
vring_interrupt() that is called via virtio_ccw_int_handler(), a per
device spinlock for irq is introduced ans used in the synchronization
method.Cc: Thomas Gleixner <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Cc: "Paul E. McKenney" <[email protected]>
Cc: Marc Zyngier <[email protected]>
Cc: Halil Pasic <[email protected]>
Cc: Cornelia Huck <[email protected]>
Signed-off-by: Jason Wang <[email protected]>Quoted 6 lines
Quoted 5 lines
Quoted 3 lines
This is the only one that is giving me pause. Halil, Cornelia, should we be concerned about the performance impact here? Any chance it can be tested?
We can have a bunch of devices using the same airq structure, and the sync cb creates a choke point, same as registering/unregistering.
BTW can callbacks for multiple VQs run on multiple CPUs at the moment?
Quoted 8 lines
I'm not sure I understand the question.
I do think we can have multiple CPUs that are executing some portion of
virtio_ccw_int_handler(). So I guess the answer is yes. Connie what do you think?On the other hand we could also end up serializing
synchronize_cbs()calls for different devices if they happen to use the sameairq_info. But this probably was not your questionI am less concerned about
synchronize_cbsbeing slow and more about the slowdown in interrupt processing itself.Quoted 6 lines
Quoted 2 lines
this patch serializes them on a spinlock.
Those could then pile up on the newly introduced spinlock.
Regards,
HalilQuoted 8 lines
Quoted 2 lines
Quoted 1 line
Hmm yea ... not good.
Is there any other way to synchronize with all callbacks?
Maybe using rwlock as airq handler?
Thanks
rwlock is still a shared cacheline bouncing between CPUs and a bunch of ordering instructions.
Quoted 7 lines
Quoted 5 lines
Yes, but it should be faster than spinlocks anyhow.
Quoted 1 line
Quoted 1 line
Maybe something per-cpu + some IPIs to run things on all CPUs instead?
Is this something like a customized version of
synchronzie_rcu_expedited()?With interrupts running in an RCU read size critical section?
For
vring_interrupt(), yes.Quoted 3 lines
Quite possibly that is also an option.
This will need a bunch of documentation since this is not a standard use of RCU,According to
Documentation/RCU/requirements.rst, it looks like a legal case:
Quoted 8 lines
"
The Linux kernel has interrupts, and RCU read-side critical sections are legal within interrupt handlers and within interrupt-disabled regions of code, as are invocations ofcall_rcu().
"And as discussed,
synchronize_rcu_expedited()is not friendly to real time workload.
I am not sure hotplug removal is relevant for realtime anyway though.
Quoted 8 lines
I think we can simply
1) protect
vring_interrupt()withrcu_read_lock()
2) usesynchronize_rcu()insynchronize_cbsfor ccwAnd if we care about the long delay we can use per device srcu to reduce that?
Thanks
Maybe ...
Quoted 10 lines
Quoted 10 lines
and probably get a confirmation
from RCU maintainers that whatever assumptions we make are guaranteed to hold down the road.Quoted 6 lines
Quoted 3 lines
... and I think classic and device interrupts are different enough here ...
Yes.
Quoted 8 lines
Quoted 8 lines
Quoted 7 lines
Thanks
Quoted 5 lines
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
--
MST
_______________________________________________
Virtualization mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/virtualization
Michael S. Tsirkin
maintainerRe: [PATCH V3 6/9] virtio-ccw: implement synchronize_cbs()
In reply to Jason WangQuoted 83 lines · Jason Wang
Quoted 67 lines · Michael S. Tsirkin
Quoted 65 lines · Jason Wang
Quoted 63 lines · Michael S. Tsirkin
Quoted 61 lines · Michael S. Tsirkin
Quoted 49 lines · Jason Wang
Quoted 47 lines · Michael S. Tsirkin
Quoted 46 lines · Michael S. Tsirkin
Quoted 27 lines · Halil Pasic
Quoted 18 lines · On Mon, 25 Apr 2022 09:59:55 -0400 "Michael S. Tsirkin" <[email protected]> wrote:
Quoted 17 lines · Cornelia Huck
Quoted 16 lines · Michael S. Tsirkin
Quoted 14 lines · Jason Wang
This patch tries to implement the synchronize_cbs() for ccw. For the
vring_interrupt() that is called via virtio_airq_handler(), the
synchronization is simply done via the airq_info's lock. For the
vring_interrupt() that is called via virtio_ccw_int_handler(), a per
device spinlock for irq is introduced ans used in the synchronization
method.Cc: Thomas Gleixner <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Cc: "Paul E. McKenney" <[email protected]>
Cc: Marc Zyngier <[email protected]>
Cc: Halil Pasic <[email protected]>
Cc: Cornelia Huck <[email protected]>
Signed-off-by: Jason Wang <[email protected]>Quoted 6 lines
Quoted 5 lines
Quoted 3 lines
This is the only one that is giving me pause. Halil, Cornelia, should we be concerned about the performance impact here? Any chance it can be tested?
We can have a bunch of devices using the same airq structure, and the sync cb creates a choke point, same as registering/unregistering.
BTW can callbacks for multiple VQs run on multiple CPUs at the moment?
Quoted 8 lines
I'm not sure I understand the question.
I do think we can have multiple CPUs that are executing some portion of
virtio_ccw_int_handler(). So I guess the answer is yes. Connie what do you think?On the other hand we could also end up serializing
synchronize_cbs()calls for different devices if they happen to use the sameairq_info. But this probably was not your questionI am less concerned about
synchronize_cbsbeing slow and more about the slowdown in interrupt processing itself.Quoted 6 lines
Quoted 2 lines
this patch serializes them on a spinlock.
Those could then pile up on the newly introduced spinlock.
Regards,
HalilQuoted 8 lines
Quoted 2 lines
Quoted 1 line
Hmm yea ... not good.
Is there any other way to synchronize with all callbacks?
Maybe using rwlock as airq handler?
Thanks
rwlock is still a shared cacheline bouncing between CPUs and a bunch of ordering instructions.
Quoted 7 lines
Quoted 5 lines
Yes, but it should be faster than spinlocks anyhow.
Quoted 1 line
Quoted 1 line
Maybe something per-cpu + some IPIs to run things on all CPUs instead?
Is this something like a customized version of
synchronzie_rcu_expedited()?With interrupts running in an RCU read size critical section?
For
vring_interrupt(), yes.Quoted 3 lines
Quite possibly that is also an option.
This will need a bunch of documentation since this is not a standard use of RCU,
Quoted 7 lines
According to
Documentation/RCU/requirements.rst, it looks like a legal case:"
The Linux kernel has interrupts, and RCU read-side critical sections are legal within interrupt handlers and within interrupt-disabled regions of code, as are invocations ofcall_rcu().
"
My problem is it is not clear what data is protected by rcu here. Nothing is tagged with __rcu or uses rcu_dereference. We need at least an ack from rcu maintainers that this is a valid use of rcu and not an undocumented side effect.
Quoted 28 lines
And as discussed,
synchronize_rcu_expedited()is not friendly to real time workload. I think we can simply1) protect
vring_interrupt()withrcu_read_lock()
2) usesynchronize_rcu()insynchronize_cbsfor ccwAnd if we care about the long delay we can use per device srcu to reduce that?
Thanks
Quoted 18 lines
and probably get a confirmation
from RCU maintainers that whatever assumptions we make are guaranteed to hold down the road.Quoted 13 lines
Quoted 3 lines
... and I think classic and device interrupts are different enough here ...
Yes.
Thanks
Quoted 5 lines
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
--
MST
_______________________________________________
Virtualization mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/virtualization
Re: [PATCH V3 6/9] virtio-ccw: implement synchronize_cbs()
Quoted 87 lines · Michael S. Tsirkin
Quoted 85 lines · Jason Wang
Quoted 67 lines · Michael S. Tsirkin
Quoted 65 lines · Jason Wang
Quoted 63 lines · Michael S. Tsirkin
Quoted 61 lines · Michael S. Tsirkin
Quoted 49 lines · Jason Wang
Quoted 47 lines · Michael S. Tsirkin
Quoted 46 lines · Michael S. Tsirkin
Quoted 27 lines · Halil Pasic
Quoted 18 lines · On Mon, 25 Apr 2022 09:59:55 -0400 "Michael S. Tsirkin" <[email protected]> wrote:
Quoted 17 lines · Cornelia Huck
Quoted 16 lines · Michael S. Tsirkin
On Mon, Apr 25, 2022 at 10:44:15AM +0800, Jason Wang wrote:
This patch tries to implement the synchronize_cbs() for ccw. For the
vring_interrupt() that is called via virtio_airq_handler(), the
synchronization is simply done via the airq_info's lock. For the
vring_interrupt() that is called via virtio_ccw_int_handler(), a per
device spinlock for irq is introduced ans used in the synchronization
method.Cc: Thomas Gleixner <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Cc: "Paul E. McKenney" <[email protected]>
Cc: Marc Zyngier <[email protected]>
Cc: Halil Pasic <[email protected]>
Cc: Cornelia Huck <[email protected]>
Signed-off-by: Jason Wang <[email protected]>Quoted 6 lines
Quoted 5 lines
Quoted 3 lines
This is the only one that is giving me pause. Halil, Cornelia,
should we be concerned about the performance impact here?
Any chance it can be tested?We can have a bunch of devices using the same airq structure, and the sync cb creates a choke point, same as registering/unregistering.
BTW can callbacks for multiple VQs run on multiple CPUs at the moment?
Quoted 8 lines
I'm not sure I understand the question.
I do think we can have multiple CPUs that are executing some portion of
virtio_ccw_int_handler(). So I guess the answer is yes. Connie what do you think?On the other hand we could also end up serializing
synchronize_cbs()calls for different devices if they happen to use the sameairq_info. But this probably was not your questionI am less concerned about
synchronize_cbsbeing slow and more about the slowdown in interrupt processing itself.Quoted 6 lines
Quoted 2 lines
this patch serializes them on a spinlock.
Those could then pile up on the newly introduced spinlock.
Regards,
HalilQuoted 8 lines
Quoted 2 lines
Quoted 1 line
Hmm yea ... not good.
Is there any other way to synchronize with all callbacks?
Maybe using rwlock as airq handler?
Thanks
rwlock is still a shared cacheline bouncing between CPUs and a bunch of ordering instructions.
Quoted 7 lines
Quoted 5 lines
Yes, but it should be faster than spinlocks anyhow.
Quoted 1 line
Quoted 1 line
Maybe something per-cpu + some IPIs to run things on all CPUs instead?
Is this something like a customized version of
synchronzie_rcu_expedited()?With interrupts running in an RCU read size critical section?
For
vring_interrupt(), yes.Quoted 3 lines
Quite possibly that is also an option.
This will need a bunch of documentation since this is not a standard use of RCU,According to
Documentation/RCU/requirements.rst, it looks like a legal case:
Quoted 8 lines
Quoted 5 lines
"
The Linux kernel has interrupts, and RCU read-side critical sections are legal within interrupt handlers and within interrupt-disabled regions of code, as are invocations ofcall_rcu().
"My problem is it is not clear what data is protected by rcu here. Nothing is tagged with
__rcuor usesrcu_dereference.
It should be the vq->broken here. We can rcuify it.
Quoted 2 lines
We need at least an ack from rcu maintainers that this is a valid use of rcu and not an undocumented side effect.
Yes.
Thanks
Quoted 30 lines
Quoted 29 lines
And as discussed,
synchronize_rcu_expedited()is not friendly to real time workload. I think we can simply1) protect
vring_interrupt()withrcu_read_lock()
2) usesynchronize_rcu()insynchronize_cbsfor ccwAnd if we care about the long delay we can use per device srcu to reduce that?
Thanks
Quoted 18 lines
and probably get a confirmation
from RCU maintainers that whatever assumptions we make are guaranteed to hold down the road.Quoted 13 lines
Quoted 3 lines
... and I think classic and device interrupts are different enough here ...
Yes.
Thanks
Quoted 5 lines
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
--
MST
Michael S. Tsirkin
maintainerRe: [PATCH V3 6/9] virtio-ccw: implement synchronize_cbs()
Quoted 94 lines · Jason Wang
Quoted 93 lines · Michael S. Tsirkin
Quoted 90 lines · Jason Wang
Quoted 67 lines · Michael S. Tsirkin
Quoted 65 lines · Jason Wang
Quoted 63 lines · Michael S. Tsirkin
Quoted 61 lines · Michael S. Tsirkin
Quoted 49 lines · Jason Wang
Quoted 47 lines · Michael S. Tsirkin
Quoted 46 lines · Michael S. Tsirkin
Quoted 27 lines · Halil Pasic
Quoted 18 lines · On Mon, 25 Apr 2022 09:59:55 -0400 "Michael S. Tsirkin" <[email protected]> wrote:
Quoted 17 lines · Cornelia Huck
On Mon, Apr 25 2022, "Michael S. Tsirkin" <[email protected]> wrote:
On Mon, Apr 25, 2022 at 10:44:15AM +0800, Jason Wang wrote:
This patch tries to implement the synchronize_cbs() for ccw. For the
vring_interrupt() that is called via virtio_airq_handler(), the
synchronization is simply done via the airq_info's lock. For the
vring_interrupt() that is called via virtio_ccw_int_handler(), a per
device spinlock for irq is introduced ans used in the synchronization
method.Cc: Thomas Gleixner <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Cc: "Paul E. McKenney" <[email protected]>
Cc: Marc Zyngier <[email protected]>
Cc: Halil Pasic <[email protected]>
Cc: Cornelia Huck <[email protected]>
Signed-off-by: Jason Wang <[email protected]>Quoted 6 lines
Quoted 5 lines
This is the only one that is giving me pause. Halil, Cornelia,
should we be concerned about the performance impact here?
Any chance it can be tested?
We can have a bunch of devices using the same airq structure, and the
sync cb creates a choke point, same as registering/unregistering.BTW can callbacks for multiple VQs run on multiple CPUs at the moment?
Quoted 8 lines
I'm not sure I understand the question.
I do think we can have multiple CPUs that are executing some portion of
virtio_ccw_int_handler(). So I guess the answer is yes. Connie what do you think?On the other hand we could also end up serializing
synchronize_cbs()calls for different devices if they happen to use the sameairq_info. But this probably was not your questionI am less concerned about
synchronize_cbsbeing slow and more about the slowdown in interrupt processing itself.Quoted 6 lines
Quoted 2 lines
this patch serializes them on a spinlock.
Those could then pile up on the newly introduced spinlock.
Regards,
HalilQuoted 8 lines
Quoted 2 lines
Quoted 1 line
Hmm yea ... not good.
Is there any other way to synchronize with all callbacks?
Maybe using rwlock as airq handler?
Thanks
rwlock is still a shared cacheline bouncing between CPUs and a bunch of ordering instructions.
Quoted 7 lines
Quoted 5 lines
Yes, but it should be faster than spinlocks anyhow.
Quoted 1 line
Quoted 1 line
Maybe something per-cpu + some IPIs to run things on all CPUs instead?
Is this something like a customized version of
synchronzie_rcu_expedited()?With interrupts running in an RCU read size critical section?
For
vring_interrupt(), yes.Quoted 3 lines
Quite possibly that is also an option.
This will need a bunch of documentation since this is not a standard use of RCU,According to
Documentation/RCU/requirements.rst, it looks like a legal case:"
The Linux kernel has interrupts, and RCU read-side critical sections are legal within interrupt handlers and within interrupt-disabled regions of code, as are invocations ofcall_rcu().
"
Quoted 4 lines
Quoted 2 lines
My problem is it is not clear what data is protected by rcu here. Nothing is tagged with
__rcuor usesrcu_dereference.It should be the
vq->brokenhere. We can rcuify it.
I don't exactly exactly see how - rcu_dereference works on pointers. Feel free to send a patch.
Quoted 39 lines
Quoted 3 lines
We need at least an ack from rcu maintainers that this is a valid use of rcu and not an undocumented side effect.
Yes.
Thanks
Quoted 30 lines
Quoted 29 lines
And as discussed,
synchronize_rcu_expedited()is not friendly to real time workload. I think we can simply1) protect
vring_interrupt()withrcu_read_lock()
2) usesynchronize_rcu()insynchronize_cbsfor ccwAnd if we care about the long delay we can use per device srcu to reduce that?
Thanks
Quoted 18 lines
and probably get a confirmation
from RCU maintainers that whatever assumptions we make are guaranteed to hold down the road.Quoted 13 lines
Quoted 3 lines
... and I think classic and device interrupts are different enough here ...
Yes.
Thanks
Quoted 5 lines
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
--
MST
_______________________________________________
Virtualization mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/virtualization
Cornelia Huck
maintainerRe: [PATCH V3 6/9] virtio-ccw: implement synchronize_cbs()
In reply to Michael S. TsirkinQuoted 42 lines · Michael S. Tsirkin
Quoted 41 lines · Michael S. Tsirkin
Quoted 40 lines · Jason Wang
Quoted 38 lines · Michael S. Tsirkin
Quoted 37 lines · Michael S. Tsirkin
Quoted 35 lines · Halil Pasic
Quoted 18 lines · On Mon, 25 Apr 2022 09:59:55 -0400 "Michael S. Tsirkin" <[email protected]> wrote:
Quoted 17 lines · Cornelia Huck
Quoted 16 lines · Michael S. Tsirkin
Quoted 14 lines · Jason Wang
This patch tries to implement the
synchronize_cbs()for ccw. For thevring_interrupt()that is called viavirtio_airq_handler(), the synchronization is simply done via the airq_info's lock. For thevring_interrupt()that is called viavirtio_ccw_int_handler(), a per device spinlock for irq is introduced ans used in the synchronization method.Cc Thomas Gleixner <[email protected]>Cc Peter Zijlstra <[email protected]>Cc "Paul E. McKenney" <[email protected]>Cc Marc Zyngier <[email protected]>Cc Halil Pasic <[email protected]>Cc Cornelia Huck <[email protected]>Signed-off-by Jason Wang <[email protected]>Quoted 6 lines
Quoted 5 lines
Quoted 3 lines
This is the only one that is giving me pause. Halil, Cornelia, should we be concerned about the performance impact here? Any chance it can be tested?
We can have a bunch of devices using the same airq structure, and the sync cb creates a choke point, same as registering/unregistering.
BTW can callbacks for multiple VQs run on multiple CPUs at the moment?
I'm not sure I understand the question.
I do think we can have multiple CPUs that are executing some portion of
virtio_ccw_int_handler(). So I guess the answer is yes. Connie what do you think?On the other hand we could also end up serializing
synchronize_cbs()calls for different devices if they happen to use the sameairq_info. But this probably was not your question
Quoted 6 lines
Quoted 6 lines
Quoted 6 lines
Quoted 6 lines
Quoted 6 lines
I am less concerned about
synchronize_cbsbeing slow and more about the slowdown in interrupt processing itself.Quoted 3 lines
Quoted 2 lines
this patch serializes them on a spinlock.
Those could then pile up on the newly introduced spinlock.
How bad would that be in practice? IIUC, we hit on the spinlock when
- doing synchronize_cbs (should be rare)
- processing queue interrupts for devices using per-device indicators (which is the non-preferred path, which I would basically only expect when running on an ancient or non-standard hypervisor)
- configuration change interrupts (should be rare)
- during setup, reset, etc. (should not be a concern)
Quoted 12 lines
Quoted 12 lines
Quoted 11 lines
Quoted 5 lines
Quoted 4 lines
Quoted 3 lines
Regards,
HalilHmm yea ... not good.
Is there any other way to synchronize with all callbacks?
Maybe using rwlock as airq handler?
Thanks
Quoted 6 lines
Quoted 3 lines
rwlock is still a shared cacheline bouncing between CPUs and a bunch of ordering instructions.
Maybe something per-cpu + some IPIs to run things on all CPUs instead?... and I think classic and device interrupts are different enough here ...
You mean classic (per-device) and adapter interrupts, right?
Michael S. Tsirkin
maintainerRe: [PATCH V3 6/9] virtio-ccw: implement synchronize_cbs()
Quoted 49 lines · Cornelia Huck
Quoted 47 lines · Michael S. Tsirkin
Quoted 46 lines · Michael S. Tsirkin
Quoted 45 lines · Jason Wang
Quoted 43 lines · Michael S. Tsirkin
Quoted 42 lines · Michael S. Tsirkin
Quoted 27 lines · Halil Pasic
Quoted 18 lines · On Mon, 25 Apr 2022 09:59:55 -0400 "Michael S. Tsirkin" <[email protected]> wrote:
Quoted 17 lines · Cornelia Huck
Quoted 16 lines · Michael S. Tsirkin
Quoted 14 lines · Jason Wang
This patch tries to implement the
synchronize_cbs()for ccw. For thevring_interrupt()that is called viavirtio_airq_handler(), the synchronization is simply done via the airq_info's lock. For thevring_interrupt()that is called viavirtio_ccw_int_handler(), a per device spinlock for irq is introduced ans used in the synchronization method.Cc Thomas Gleixner <[email protected]>Cc Peter Zijlstra <[email protected]>Cc "Paul E. McKenney" <[email protected]>Cc Marc Zyngier <[email protected]>Cc Halil Pasic <[email protected]>Cc Cornelia Huck <[email protected]>Signed-off-by Jason Wang <[email protected]>Quoted 6 lines
Quoted 5 lines
Quoted 3 lines
This is the only one that is giving me pause. Halil, Cornelia, should we be concerned about the performance impact here? Any chance it can be tested?
We can have a bunch of devices using the same airq structure, and the sync cb creates a choke point, same as registering/unregistering.
BTW can callbacks for multiple VQs run on multiple CPUs at the moment?
Quoted 8 lines
I'm not sure I understand the question.
I do think we can have multiple CPUs that are executing some portion of
virtio_ccw_int_handler(). So I guess the answer is yes. Connie what do you think?On the other hand we could also end up serializing
synchronize_cbs()calls for different devices if they happen to use the sameairq_info. But this probably was not your questionI am less concerned about
synchronize_cbsbeing slow and more about the slowdown in interrupt processing itself.Quoted 2 lines
Quoted 2 lines
this patch serializes them on a spinlock.
Quoted 7 lines
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Quoted 1 line
Those could then pile up on the newly introduced spinlock.
How bad would that be in practice? IIUC, we hit on the spinlock when
- doingsynchronize_cbs(should be rare)
- processing queue interrupts for devices using per-device indicators (which is the non-preferred path, which I would basically only expect when running on an ancient or non-standard hypervisor)
this one is my concern. I am worried serializing everything on a single lock will drastically regress performance here.
Quoted 15 lines
- configuration change interrupts (should be rare)
- during setup, reset, etc. (should not be a concern)Quoted 12 lines
Quoted 12 lines
Quoted 11 lines
Quoted 5 lines
Quoted 4 lines
Quoted 3 lines
Regards,
HalilHmm yea ... not good.
Is there any other way to synchronize with all callbacks?
Maybe using rwlock as airq handler?
Thanks
Quoted 8 lines
Quoted 6 lines
Quoted 3 lines
rwlock is still a shared cacheline bouncing between CPUs and a bunch of ordering instructions.
Maybe something per-cpu + some IPIs to run things on all CPUs instead?... and I think classic and device interrupts are different enough here ...
You mean classic (per-device) and adapter interrupts, right?
_______________________________________________
Virtualization mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/virtualization
Cornelia Huck
maintainerRe: [PATCH V3 6/9] virtio-ccw: implement synchronize_cbs()
Quoted 52 lines · Michael S. Tsirkin
Quoted 51 lines · Cornelia Huck
Quoted 48 lines · Michael S. Tsirkin
Quoted 47 lines · Michael S. Tsirkin
Quoted 46 lines · Jason Wang
Quoted 44 lines · Michael S. Tsirkin
Quoted 43 lines · Michael S. Tsirkin
Quoted 27 lines · Halil Pasic
Quoted 18 lines · On Mon, 25 Apr 2022 09:59:55 -0400 "Michael S. Tsirkin" <[email protected]> wrote:
Quoted 17 lines · Cornelia Huck
Quoted 16 lines · Michael S. Tsirkin
Quoted 14 lines · Jason Wang
This patch tries to implement the
synchronize_cbs()for ccw. For thevring_interrupt()that is called viavirtio_airq_handler(), the synchronization is simply done via the airq_info's lock. For thevring_interrupt()that is called viavirtio_ccw_int_handler(), a per device spinlock for irq is introduced ans used in the synchronization method.Cc Thomas Gleixner <[email protected]>Cc Peter Zijlstra <[email protected]>Cc "Paul E. McKenney" <[email protected]>Cc Marc Zyngier <[email protected]>Cc Halil Pasic <[email protected]>Cc Cornelia Huck <[email protected]>Signed-off-by Jason Wang <[email protected]>Quoted 6 lines
Quoted 5 lines
Quoted 3 lines
This is the only one that is giving me pause. Halil, Cornelia, should we be concerned about the performance impact here? Any chance it can be tested?
We can have a bunch of devices using the same airq structure, and the sync cb creates a choke point, same as registering/unregistering.
BTW can callbacks for multiple VQs run on multiple CPUs at the moment?
Quoted 8 lines
I'm not sure I understand the question.
I do think we can have multiple CPUs that are executing some portion of
virtio_ccw_int_handler(). So I guess the answer is yes. Connie what do you think?On the other hand we could also end up serializing
synchronize_cbs()calls for different devices if they happen to use the sameairq_info. But this probably was not your questionI am less concerned about
synchronize_cbsbeing slow and more about the slowdown in interrupt processing itself.Quoted 3 lines
Quoted 2 lines
this patch serializes them on a spinlock.
Those could then pile up on the newly introduced spinlock.
Quoted 8 lines
Quoted 5 lines
How bad would that be in practice? IIUC, we hit on the spinlock when
- doingsynchronize_cbs(should be rare)
- processing queue interrupts for devices using per-device indicators (which is the non-preferred path, which I would basically only expect when running on an ancient or non-standard hypervisor)this one is my concern. I am worried serializing everything on a single lock will drastically regress performance here.
Yeah, that case could get much worse. OTOH, how likely is it that any setup that runs a recent kernel will actually end up with devices using per-device indicators? Anything running under a QEMU released in the last couple of years is unlikely to not use airqs, I think. Halil, do you think that the classic indicator setup would be more common on any non-QEMU hypervisors?
IOW, how much effort is it worth spending on optimizing this case? We certainly should explore any simple solutions, but I don't think we need to twist ourselves into pretzels to solve it.
Quoted 4 lines
Quoted 4 lines
- configuration change interrupts (should be rare)
- during setup, reset, etc. (should not be a concern)
_______________________________________________
Virtualization mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/virtualization
Halil Pasic
maintainerRe: [PATCH V3 6/9] virtio-ccw: implement synchronize_cbs()
Quoted 63 lines · On Wed, 27 Apr 2022 11:27:03 +0200 Cornelia Huck <[email protected]> wrote:
Quoted 52 lines · Michael S. Tsirkin
Quoted 51 lines · Cornelia Huck
Quoted 48 lines · Michael S. Tsirkin
Quoted 47 lines · Michael S. Tsirkin
Quoted 46 lines · Jason Wang
Quoted 44 lines · Michael S. Tsirkin
Quoted 43 lines · Michael S. Tsirkin
Quoted 27 lines · Halil Pasic
Quoted 18 lines · On Mon, 25 Apr 2022 09:59:55 -0400 "Michael S. Tsirkin" <[email protected]> wrote:
Quoted 17 lines · Cornelia Huck
Quoted 16 lines · Michael S. Tsirkin
Quoted 14 lines · Jason Wang
This patch tries to implement the synchronize_cbs() for ccw. For the
vring_interrupt() that is called via virtio_airq_handler(), the
synchronization is simply done via the airq_info's lock. For the
vring_interrupt() that is called via virtio_ccw_int_handler(), a per
device spinlock for irq is introduced ans used in the synchronization
method.Cc: Thomas Gleixner <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Cc: "Paul E. McKenney" <[email protected]>
Cc: Marc Zyngier <[email protected]>
Cc: Halil Pasic <[email protected]>
Cc: Cornelia Huck <[email protected]>
Signed-off-by: Jason Wang <[email protected]>Quoted 6 lines
Quoted 5 lines
Quoted 3 lines
This is the only one that is giving me pause. Halil, Cornelia, should we be concerned about the performance impact here? Any chance it can be tested?
We can have a bunch of devices using the same airq structure, and the sync cb creates a choke point, same as registering/unregistering.
BTW can callbacks for multiple VQs run on multiple CPUs at the moment?
Quoted 8 lines
I'm not sure I understand the question.
I do think we can have multiple CPUs that are executing some portion of
virtio_ccw_int_handler(). So I guess the answer is yes. Connie what do you think?On the other hand we could also end up serializing
synchronize_cbs()calls for different devices if they happen to use the sameairq_info. But this probably was not your questionI am less concerned about
synchronize_cbsbeing slow and more about the slowdown in interrupt processing itself.Quoted 3 lines
Quoted 2 lines
this patch serializes them on a spinlock.
Those could then pile up on the newly introduced spinlock.
Quoted 8 lines
Quoted 5 lines
How bad would that be in practice? IIUC, we hit on the spinlock when
- doingsynchronize_cbs(should be rare)
- processing queue interrupts for devices using per-device indicators (which is the non-preferred path, which I would basically only expect when running on an ancient or non-standard hypervisor)this one is my concern. I am worried serializing everything on a single lock will drastically regress performance here.
Quoted 7 lines
Yeah, that case could get much worse. OTOH, how likely is it that any setup that runs a recent kernel will actually end up with devices using per-device indicators? Anything running under a QEMU released in the last couple of years is unlikely to not use airqs, I think. Halil, do you think that the classic indicator setup would be more common on any non-QEMU hypervisors?
I really don't know. My opinion is that, two stages indicators are kind of recommended for anybody who cares about notifications performance.
Quoted 4 lines
IOW, how much effort is it worth spending on optimizing this case? We certainly should explore any simple solutions, but I don't think we need to twist ourselves into pretzels to solve it.
Frankly, I would be fine with an rwlock based solution as proposed by Jason. My rationale is: we recommend two stage indicators, and the two stage indicators are already encumbered by an rwlock on the interrupt path. Yes, the coalescence of adapter interrupts is architecturally different, and so it is with GISA (without GISA, I'm not even sure), so this rwlock end up being worse than the one for 2 stage. But my feeling is, that it should be fine. On the other hand, I don't feel comfortable with plain spinlock, and I am curious about a more advanced solution. But my guess is that rwlock + some testing for the legacy indicator case just to double check if there is a heavy regression despite of our expectations to see none should do the trick.
Regards,
Halil
Quoted 5 lines
Quoted 4 lines
Quoted 4 lines
- configuration change interrupts (should be rare)
- during setup, reset, etc. (should not be a concern)
_______________________________________________
Virtualization mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/virtualization
Re: [PATCH V3 6/9] virtio-ccw: implement synchronize_cbs()
Quoted 83 lines · Halil Pasic
Quoted 63 lines · On Wed, 27 Apr 2022 11:27:03 +0200 Cornelia Huck <[email protected]> wrote:
Quoted 52 lines · Michael S. Tsirkin
Quoted 51 lines · Cornelia Huck
Quoted 48 lines · Michael S. Tsirkin
Quoted 47 lines · Michael S. Tsirkin
Quoted 46 lines · Jason Wang
Quoted 44 lines · Michael S. Tsirkin
Quoted 43 lines · Michael S. Tsirkin
Quoted 27 lines · Halil Pasic
Quoted 18 lines · On Mon, 25 Apr 2022 09:59:55 -0400 "Michael S. Tsirkin" <[email protected]> wrote:
Quoted 17 lines · Cornelia Huck
Quoted 16 lines · Michael S. Tsirkin
On Mon, Apr 25, 2022 at 10:44:15AM +0800, Jason Wang wrote:
This patch tries to implement the synchronize_cbs() for ccw. For the
vring_interrupt() that is called via virtio_airq_handler(), the
synchronization is simply done via the airq_info's lock. For the
vring_interrupt() that is called via virtio_ccw_int_handler(), a per
device spinlock for irq is introduced ans used in the synchronization
method.Cc: Thomas Gleixner <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Cc: "Paul E. McKenney" <[email protected]>
Cc: Marc Zyngier <[email protected]>
Cc: Halil Pasic <[email protected]>
Cc: Cornelia Huck <[email protected]>
Signed-off-by: Jason Wang <[email protected]>Quoted 6 lines
Quoted 5 lines
Quoted 3 lines
This is the only one that is giving me pause. Halil, Cornelia,
should we be concerned about the performance impact here?
Any chance it can be tested?We can have a bunch of devices using the same airq structure, and the sync cb creates a choke point, same as registering/unregistering.
BTW can callbacks for multiple VQs run on multiple CPUs at the moment?
Quoted 8 lines
I'm not sure I understand the question.
I do think we can have multiple CPUs that are executing some portion of
virtio_ccw_int_handler(). So I guess the answer is yes. Connie what do you think?On the other hand we could also end up serializing
synchronize_cbs()calls for different devices if they happen to use the sameairq_info. But this probably was not your questionI am less concerned about
synchronize_cbsbeing slow and more about the slowdown in interrupt processing itself.Quoted 3 lines
Quoted 2 lines
this patch serializes them on a spinlock.
Those could then pile up on the newly introduced spinlock.
Quoted 8 lines
Quoted 5 lines
How bad would that be in practice? IIUC, we hit on the spinlock when
- doingsynchronize_cbs(should be rare)
- processing queue interrupts for devices using per-device indicators (which is the non-preferred path, which I would basically only expect when running on an ancient or non-standard hypervisor)this one is my concern. I am worried serializing everything on a single lock will drastically regress performance here.
Quoted 7 lines
Yeah, that case could get much worse. OTOH, how likely is it that any setup that runs a recent kernel will actually end up with devices using per-device indicators? Anything running under a QEMU released in the last couple of years is unlikely to not use airqs, I think. Halil, do you think that the classic indicator setup would be more common on any non-QEMU hypervisors?
I really don't know. My opinion is that, two stages indicators are kind of recommended for anybody who cares about notifications performance.
Quoted 4 lines
IOW, how much effort is it worth spending on optimizing this case? We certainly should explore any simple solutions, but I don't think we need to twist ourselves into pretzels to solve it.
Quoted 8 lines
Frankly, I would be fine with an rwlock based solution as proposed by Jason. My rationale is: we recommend two stage indicators, and the two stage indicators are already encumbered by an rwlock on the interrupt path. Yes, the coalescence of adapter interrupts is architecturally different, and so it is with GISA (without GISA, I'm not even sure), so this rwlock end up being worse than the one for 2 stage. But my feeling is, that it should be fine. On the other hand, I don't feel comfortable with plain spinlock, and I am curious about a more advanced solution.
Yes, I'm trying to use (S)RCU, let's see if it works.
Quoted 3 lines
But my guess is that rwlock + some testing for the legacy indicator case just to double check if there is a heavy regression despite of our expectations to see none should do the trick.
I suggest this, rwlock (for not airq) seems better than spinlock, but at worst case it will cause cache line bouncing. But I wonder if it's noticeable (anyhow it has been used for airq).
Thanks
Quoted 10 lines
Regards,
HalilQuoted 5 lines
Quoted 4 lines
Quoted 4 lines
- configuration change interrupts (should be rare)
- during setup, reset, etc. (should not be a concern)
_______________________________________________
Virtualization mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/virtualization
Michael S. Tsirkin
maintainerRe: [PATCH V3 6/9] virtio-ccw: implement synchronize_cbs()
Quoted 9 lines · Jason Wang
Quoted 3 lines
But my guess is that rwlock + some testing for the legacy indicator case just to double check if there is a heavy regression despite of our expectations to see none should do the trick.
I suggest this, rwlock (for not airq) seems better than spinlock, but at worst case it will cause cache line bouncing. But I wonder if it's noticeable (anyhow it has been used for airq).
Thanks
Which existing rwlock does airq use right now? Can we take it to sync?
Signature · 6 lines
MST _______________________________________________ Virtualization mailing list [email protected] https://lists.linuxfoundation.org/mailman/listinfo/virtualization
Re: [PATCH V3 6/9] virtio-ccw: implement synchronize_cbs()
Quoted 6 lines · Michael S. Tsirkin
Quoted 4 lines · Jason Wang
Quoted 3 lines
But my guess is that rwlock + some testing for the legacy indicator case just to double check if there is a heavy regression despite of our expectations to see none should do the trick.
Quoted 7 lines
Quoted 5 lines
I suggest this, rwlock (for not airq) seems better than spinlock, but at worst case it will cause cache line bouncing. But I wonder if it's noticeable (anyhow it has been used for airq).
Thanks
Which existing rwlock does airq use right now? Can we take it to sync?
It's the rwlock in airq_info, it has already been used in this patch.
write_lock(&info->lock);
write_unlock(&info->lock);But the problem is, it looks to me there could be a case that airq is not used, (virtio_ccw_int_hander()). That's why the patch use a spinlock, it could be optimized with using a rwlock as well.
Thanks
Quoted 4 lines
--
MST
_______________________________________________
Virtualization mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/virtualization
Michael S. Tsirkin
maintainerRe: [PATCH V3 6/9] virtio-ccw: implement synchronize_cbs()
Quoted 17 lines · Jason Wang
Quoted 6 lines · Michael S. Tsirkin
Quoted 4 lines · Jason Wang
Quoted 3 lines
But my guess is that rwlock + some testing for the legacy indicator case just to double check if there is a heavy regression despite of our expectations to see none should do the trick.
Quoted 7 lines
Quoted 5 lines
I suggest this, rwlock (for not airq) seems better than spinlock, but at worst case it will cause cache line bouncing. But I wonder if it's noticeable (anyhow it has been used for airq).
Thanks
Which existing rwlock does airq use right now? Can we take it to sync?
It's the rwlock in
airq_info, it has already been used in this patch.
Quoted 8 lines
write_lock(&info->lock); write_unlock(&info->lock);But the problem is, it looks to me there could be a case that airq is not used, (
virtio_ccw_int_hander()). That's why the patch use a spinlock, it could be optimized with using a rwlock as well.Thanks
Ah, right. So let's take that on the legacy path too and Halil promises to test to make sure performance isn't impacted too badly?
Quoted 4 lines
Quoted 4 lines
--
MST
_______________________________________________
Virtualization mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/virtualization
Re: [PATCH V3 6/9] virtio-ccw: implement synchronize_cbs()
Quoted 22 lines · Michael S. Tsirkin
Quoted 20 lines · Jason Wang
Quoted 6 lines · Michael S. Tsirkin
Quoted 4 lines · Jason Wang
Quoted 3 lines
But my guess is that rwlock + some testing for the legacy indicator case just to double check if there is a heavy regression despite of our expectations to see none should do the trick.
Quoted 7 lines
Quoted 5 lines
I suggest this, rwlock (for not airq) seems better than spinlock, but at worst case it will cause cache line bouncing. But I wonder if it's noticeable (anyhow it has been used for airq).
Thanks
Which existing rwlock does airq use right now? Can we take it to sync?
It's the rwlock in
airq_info, it has already been used in this patch.write_lock(&info->lock); write_unlock(&info->lock);
Quoted 8 lines
Quoted 5 lines
But the problem is, it looks to me there could be a case that airq is not used, (
virtio_ccw_int_hander()). That's why the patch use a spinlock, it could be optimized with using a rwlock as well.Thanks
Ah, right. So let's take that on the legacy path too and Halil promises to test to make sure performance isn't impacted too badly?
I think what you meant is using a dedicated rwlock instead of trying to reuse one of the airq_info locks.
If this is true, it should be fine.
Thanks
Quoted 6 lines
Quoted 5 lines
Quoted 5 lines
--
MST
_______________________________________________
Virtualization mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/virtualization
Michael S. Tsirkin
maintainerRe: [PATCH V3 6/9] virtio-ccw: implement synchronize_cbs()
Quoted 32 lines · Jason Wang
Quoted 22 lines · Michael S. Tsirkin
Quoted 20 lines · Jason Wang
Quoted 6 lines · Michael S. Tsirkin
Quoted 4 lines · Jason Wang
Quoted 3 lines
But my guess is that rwlock + some testing for the legacy indicator case just to double check if there is a heavy regression despite of our expectations to see none should do the trick.
Quoted 7 lines
Quoted 5 lines
I suggest this, rwlock (for not airq) seems better than spinlock, but at worst case it will cause cache line bouncing. But I wonder if it's noticeable (anyhow it has been used for airq).
Thanks
Which existing rwlock does airq use right now? Can we take it to sync?
It's the rwlock in
airq_info, it has already been used in this patch.write_lock(&info->lock); write_unlock(&info->lock);Quoted 8 lines
Quoted 5 lines
But the problem is, it looks to me there could be a case that airq is not used, (
virtio_ccw_int_hander()). That's why the patch use a spinlock, it could be optimized with using a rwlock as well.Thanks
Ah, right. So let's take that on the legacy path too and Halil promises to test to make sure performance isn't impacted too badly?
Quoted 6 lines
I think what you meant is using a dedicated rwlock instead of trying to reuse one of the
airq_infolocks.If this is true, it should be fine.
Thanks
yes
Quoted 6 lines
Quoted 6 lines
Quoted 5 lines
Quoted 5 lines
--
MST
_______________________________________________
Virtualization mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/virtualization
Cornelia Huck
maintainerRe: [PATCH V3 6/9] virtio-ccw: implement synchronize_cbs()
In reply to Jason WangQuoted 29 lines · Jason Wang
Quoted 28 lines · Michael S. Tsirkin
Quoted 25 lines · Jason Wang
Quoted 6 lines · Michael S. Tsirkin
Quoted 4 lines · Jason Wang
Quoted 3 lines
But my guess is that rwlock + some testing for the legacy indicator case just to double check if there is a heavy regression despite of our expectations to see none should do the trick.
Quoted 7 lines
Quoted 5 lines
I suggest this, rwlock (for not airq) seems better than spinlock, but at worst case it will cause cache line bouncing. But I wonder if it's noticeable (anyhow it has been used for airq).
Thanks
Which existing rwlock does airq use right now? Can we take it to sync?
It's the rwlock in
airq_info, it has already been used in this patch.write_lock(&info->lock); write_unlock(&info->lock);But the problem is, it looks to me there could be a case that airq is not used, (
virtio_ccw_int_hander()). That's why the patch use a spinlock, it could be optimized with using a rwlock as well.Thanks
Quoted 7 lines
Quoted 2 lines
Ah, right. So let's take that on the legacy path too and Halil promises to test to make sure performance isn't impacted too badly?
I think what you meant is using a dedicated rwlock instead of trying to reuse one of the
airq_infolocks.If this is true, it should be fine.
FWIW, that approach makes sense to me as well.
_______________________________________________
Virtualization mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/virtualization
Re: [PATCH V3 6/9] virtio-ccw: implement synchronize_cbs()
Quoted 35 lines · Cornelia Huck
Quoted 32 lines · Jason Wang
Quoted 30 lines · Michael S. Tsirkin
Quoted 17 lines · Jason Wang
Quoted 6 lines · Michael S. Tsirkin
Quoted 4 lines · Jason Wang
Quoted 3 lines
But my guess is that rwlock + some testing for the legacy indicator case just to double check if there is a heavy regression despite of our expectations to see none should do the trick.
Quoted 7 lines
Quoted 5 lines
I suggest this, rwlock (for not airq) seems better than spinlock, but at worst case it will cause cache line bouncing. But I wonder if it's noticeable (anyhow it has been used for airq).
Thanks
Which existing rwlock does airq use right now? Can we take it to sync?
It's the rwlock in
airq_info, it has already been used in this patch.Quoted 8 lines
write_lock(&info->lock); write_unlock(&info->lock);But the problem is, it looks to me there could be a case that airq is not used, (
virtio_ccw_int_hander()). That's why the patch use a spinlock, it could be optimized with using a rwlock as well.Thanks
Ah, right. So let's take that on the legacy path too and Halil promises to test to make sure performance isn't impacted too badly?
Quoted 7 lines
Quoted 4 lines
I think what you meant is using a dedicated rwlock instead of trying to reuse one of the
airq_infolocks.If this is true, it should be fine.
FWIW, that approach makes sense to me as well.
Good to know that. Let me post a new version.
Thanks
_______________________________________________
Virtualization mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/virtualization
Michael S. Tsirkin
maintainerRe: [PATCH V3 6/9] virtio-ccw: implement synchronize_cbs()
In reply to Halil PasicQuoted 82 lines · Halil Pasic
Quoted 63 lines · On Wed, 27 Apr 2022 11:27:03 +0200 Cornelia Huck <[email protected]> wrote:
Quoted 52 lines · Michael S. Tsirkin
Quoted 51 lines · Cornelia Huck
Quoted 48 lines · Michael S. Tsirkin
Quoted 47 lines · Michael S. Tsirkin
Quoted 46 lines · Jason Wang
Quoted 44 lines · Michael S. Tsirkin
Quoted 43 lines · Michael S. Tsirkin
Quoted 27 lines · Halil Pasic
Quoted 18 lines · On Mon, 25 Apr 2022 09:59:55 -0400 "Michael S. Tsirkin" <[email protected]> wrote:
Quoted 17 lines · Cornelia Huck
Quoted 16 lines · Michael S. Tsirkin
On Mon, Apr 25, 2022 at 10:44:15AM +0800, Jason Wang wrote:
This patch tries to implement the synchronize_cbs() for ccw. For the
vring_interrupt() that is called via virtio_airq_handler(), the
synchronization is simply done via the airq_info's lock. For the
vring_interrupt() that is called via virtio_ccw_int_handler(), a per
device spinlock for irq is introduced ans used in the synchronization
method.Cc: Thomas Gleixner <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Cc: "Paul E. McKenney" <[email protected]>
Cc: Marc Zyngier <[email protected]>
Cc: Halil Pasic <[email protected]>
Cc: Cornelia Huck <[email protected]>
Signed-off-by: Jason Wang <[email protected]>Quoted 6 lines
Quoted 5 lines
Quoted 3 lines
This is the only one that is giving me pause. Halil, Cornelia,
should we be concerned about the performance impact here?
Any chance it can be tested?We can have a bunch of devices using the same airq structure, and the sync cb creates a choke point, same as registering/unregistering.
BTW can callbacks for multiple VQs run on multiple CPUs at the moment?
Quoted 8 lines
I'm not sure I understand the question.
I do think we can have multiple CPUs that are executing some portion of
virtio_ccw_int_handler(). So I guess the answer is yes. Connie what do you think?On the other hand we could also end up serializing
synchronize_cbs()calls for different devices if they happen to use the sameairq_info. But this probably was not your questionI am less concerned about
synchronize_cbsbeing slow and more about the slowdown in interrupt processing itself.Quoted 3 lines
Quoted 2 lines
this patch serializes them on a spinlock.
Those could then pile up on the newly introduced spinlock.
Quoted 8 lines
Quoted 5 lines
How bad would that be in practice? IIUC, we hit on the spinlock when
- doingsynchronize_cbs(should be rare)
- processing queue interrupts for devices using per-device indicators (which is the non-preferred path, which I would basically only expect when running on an ancient or non-standard hypervisor)this one is my concern. I am worried serializing everything on a single lock will drastically regress performance here.
Quoted 7 lines
Yeah, that case could get much worse. OTOH, how likely is it that any setup that runs a recent kernel will actually end up with devices using per-device indicators? Anything running under a QEMU released in the last couple of years is unlikely to not use airqs, I think. Halil, do you think that the classic indicator setup would be more common on any non-QEMU hypervisors?
I really don't know. My opinion is that, two stages indicators are kind of recommended for anybody who cares about notifications performance.
Quoted 4 lines
IOW, how much effort is it worth spending on optimizing this case? We certainly should explore any simple solutions, but I don't think we need to twist ourselves into pretzels to solve it.
Quoted 4 lines
Frankly, I would be fine with an rwlock based solution as proposed by Jason. My rationale is: we recommend two stage indicators, and the two stage indicators are already encumbered by an rwlock on the interrupt path.
Which lock is it? Can we take that lock to synchronize?
Quoted 11 lines
Yes, the coalescence of adapter interrupts is architecturally different, and so it is with GISA (without GISA, I'm not even sure), so this rwlock end up being worse than the one for 2 stage. But my feeling is, that it should be fine. On the other hand, I don't feel comfortable with plain spinlock, and I am curious about a more advanced solution. But my guess is that rwlock + some testing for the legacy indicator case just to double check if there is a heavy regression despite of our expectations to see none should do the trick.
Regards,
Halil
Makes sense.
Quoted 5 lines
Quoted 5 lines
Quoted 4 lines
Quoted 4 lines
- configuration change interrupts (should be rare)
- during setup, reset, etc. (should not be a concern)
Halil Pasic
maintainerRe: [PATCH V3 6/9] virtio-ccw: implement synchronize_cbs()
In reply to Cornelia HuckQuoted 23 lines · On Mon, 25 Apr 2022 10:54:24 +0200 Cornelia Huck <[email protected]> wrote:
Quoted 17 lines · Michael S. Tsirkin
Quoted 7 lines · Jason Wang
This patch tries to implement the
synchronize_cbs()for ccw. For thevring_interrupt()that is called viavirtio_airq_handler(), the synchronization is simply done via the airq_info's lock. For thevring_interrupt()that is called viavirtio_ccw_int_handler(), a per device spinlock for irq is introduced ans used in the synchronization method.Quoted 7 lines
Cc Thomas Gleixner <[email protected]>Cc Peter Zijlstra <[email protected]>Cc "Paul E. McKenney" <[email protected]>Cc Marc Zyngier <[email protected]>Cc Halil Pasic <[email protected]>Cc Cornelia Huck <[email protected]>Signed-off-by Jason Wang <[email protected]>Quoted 3 lines
This is the only one that is giving me pause. Halil, Cornelia, should we be concerned about the performance impact here? Any chance it can be tested?
Quoted 5 lines
We can have a bunch of devices using the same airq structure, and the sync cb creates a choke point, same as registering/unregistering. If invoking the sync cb is a rare operation (same as (un)registering), it should not affect interrupt processing for other devices too much, but it really should be rare.
With the notable difference that the critical section in sync_cb is basically empty, so it should be less intrusive that register/unregister.
I would also argue, that since after the reset we (re-)discover our virtqueues and (re-)register adapter interrupts, and thus before or as a part of the reset we probably do an unregister to clean up the adapter interrupts and de-allocate the bits in the info, this should not incur any mayor overhead for the airq case, which is the preferred one.
Or am I missing something?
Quoted 8 lines
For testing, you would probably want to use a setup with many devices that share the same airq area (you can fit a lot of devices if they have few queues), generate traffic on the queues, and then do something that triggers the callback (adding/removing a new device in a loop?)
I currently don't have such a setup handy; Halil, would you be able to test that?
Neither do I. I would also have to start from scratch. I guess it would be also sufficient to do a setup with two devices: a nic with many busy queues, and another device that is responsible for generating the resets.
Regards,
Halil
Quoted 36 lines
Quoted 36 lines
Quoted 36 lines
--- drivers/s390/virtio/virtio_ccw.c | 27 +++++++++++++++++++++++++++ 1 file changed, 27 insertions(+)drivers/s390/virtio/virtio_ccw.c +15
-0@@ -62,6 +62,7 @@ struct virtio_ccw_device {unsigned int revision;/* Transport revision */ wait_queue_head_t wait_q;spinlock_t lock; +spinlock_t irq_lock;struct mutex io_lock;/* Serializes I/O requests */ struct list_head virtqueues;bool is_thinint;@@ -984,6 +985,27 @@ static const char *virtio_ccw_bus_name(struct virtio_device *vdev)return dev_name (&vcdev->cdev->dev); } +static void virtio_ccw_synchronize_cbs (struct virtio_device *vdev) +{ +struct virtio_ccw_device *vcdev =to_vc_device (vdev); +struct airq_info *info = vcdev->airq_info; + +/* +* Synchronize with the vring_interrupt() called by +* virtio_ccw_int_handler(). +*/ +spin_lock (&vcdev->irq_lock); +spin_unlock (&vcdev->irq_lock); + +if (info) { +/*
Quoted 8 lines
Quoted 5 lines
drivers/s390/virtio/virtio_ccw.c static const char *virtio_ccw_bus_name(struct virtio_device *vdev) · line 1006
Quoted 5 lines
drivers/s390/virtio/virtio_ccw.c +5
-0+* Synchronize with the vring_interrupt() with airq indicator +*/ +write_lock (&info->lock); +write_unlock (&info->lock); + }I think we can make this an either/or operation (devices will either use classic interrupts or adapter interrupts)?
Right, for virtqueue notifications. I second Connie's motion!
Quoted 7 lines
Quoted 6 lines
drivers/s390/virtio/virtio_ccw.c static const char *virtio_ccw_bus_name(struct virtio_device *vdev) · line 1011
Quoted 6 lines
drivers/s390/virtio/virtio_ccw.c +2
-0+} +static const struct virtio_config_ops virtio_ccw_config_ops = { .get_features = virtio_ccw_get_features, .finalize_features = virtio_ccw_finalize_features,
_______________________________________________
Virtualization mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/virtualization
Re: [PATCH V3 6/9] virtio-ccw: implement synchronize_cbs()
Quoted 21 lines · Halil Pasic
Quoted 13 lines
Quoted 13 lines
Quoted 13 lines
drivers/s390/virtio/virtio_ccw.c +13
-0+static void virtio_ccw_synchronize_cbs (struct virtio_device *vdev) +{ +struct virtio_ccw_device *vcdev =to_vc_device (vdev); +struct airq_info *info = vcdev->airq_info; + +/* +* Synchronize with the vring_interrupt() called by +* virtio_ccw_int_handler(). +*/ +spin_lock (&vcdev->irq_lock); +spin_unlock (&vcdev->irq_lock); + +if (info) {Quoted 8 lines
Quoted 6 lines
drivers/s390/virtio/virtio_ccw.c static const char *virtio_ccw_bus_name(struct virtio_device *vdev) · line 1006
Quoted 6 lines
drivers/s390/virtio/virtio_ccw.c +6
-0+/* +* Synchronize with the vring_interrupt() with airq indicator +*/ +write_lock (&info->lock); +write_unlock (&info->lock); + }I think we can make this an either/or operation (devices will either use classic interrupts or adapter interrupts)?
Quoted 1 line
Right, for virtqueue notifications. I second Connie's motion!
Will do this.
Thanks
Quoted 1 line
A message that is not in the archive (sent off-list or not yet received): [email protected]
Re: [PATCH V3 6/9] virtio-ccw: implement synchronize_cbs()
Quoted 36 lines · Hillf Danton
Quoted 34 lines · On Mon, 25 Apr 2022 10:44:15 +0800 Jason Wang wrote:
This patch tries to implement the
synchronize_cbs()for ccw. For thevring_interrupt()that is called viavirtio_airq_handler(), the synchronization is simply done via the airq_info's lock. For thevring_interrupt()that is called viavirtio_ccw_int_handler(), a per device spinlock for irq is introduced ans used in the synchronization method.Cc Thomas Gleixner <[email protected]>Cc Peter Zijlstra <[email protected]>Cc "Paul E. McKenney" <[email protected]>Cc Marc Zyngier <[email protected]>Cc Halil Pasic <[email protected]>Cc Cornelia Huck <[email protected]>Signed-off-by Jason Wang <[email protected]>--- drivers/s390/virtio/virtio_ccw.c | 27 +++++++++++++++++++++++++++ 1 file changed, 27 insertions(+)drivers/s390/virtio/virtio_ccw.c +1
-0@@ -62,6 +62,7 @@ struct virtio_ccw_device {unsigned int revision;/* Transport revision */ wait_queue_head_t wait_q;spinlock_t lock; +spinlock_t irq_lock;struct mutex io_lock;/* Serializes I/O requests */ struct list_head virtqueues;bool is_thinint;@@ -984,6 +985,27 @@ static const char *virtio_ccw_bus_name(struct virtio_device *vdev)return dev_name (&vcdev->cdev->dev); }
Quoted 14 lines
drivers/s390/virtio/virtio_ccw.c static const char *virtio_ccw_bus_name(struct virtio_device *vdev) · line 998
Quoted 11 lines
drivers/s390/virtio/virtio_ccw.c +11
-0+static void virtio_ccw_synchronize_cbs (struct virtio_device *vdev) +{ +struct virtio_ccw_device *vcdev =to_vc_device (vdev); +struct airq_info *info = vcdev->airq_info; + +/* +* Synchronize with the vring_interrupt() called by +* virtio_ccw_int_handler(). +*/ +spin_lock (&vcdev->irq_lock); +spin_unlock (&vcdev->irq_lock);Given five lock acquirers, is the sync able to ensure the other four parties have complete their lock/unlock operations?
Just to make sure I understand the meaning of "four parties" here. It looks to me the lock is only used in virtio_ccw_int_handler().
Thanks
Quoted 59 lines
Quoted 58 lines
drivers/s390/virtio/virtio_ccw.c +15
-1+ +if (info) { +/* +* Synchronize with the vring_interrupt() with airq indicator +*/ +write_lock (&info->lock); +write_unlock (&info->lock); + } +} +static const struct virtio_config_ops virtio_ccw_config_ops = { .get_features = virtio_ccw_get_features, .finalize_features = virtio_ccw_finalize_features,@@ -995,6 +1017,7 @@ static const struct virtio_config_ops virtio_ccw_config_ops = { .find_vqs = virtio_ccw_find_vqs, .del_vqs = virtio_ccw_del_vqs, .bus_name = virtio_ccw_bus_name, + .synchronize_cbs = virtio_ccw_synchronize_cbs, };@@ -1079,6 +1102,7 @@ static void virtio_ccw_int_handler(struct ccw_device *cdev, {__u32 activity = intparm &VIRTIO_CCW_INTPARM_MASK ;struct virtio_ccw_device *vcdev =dev_get_drvdata (&cdev->dev); +unsigned long flags;int i;struct virtqueue *vq;@@ -1106,6 +1130,7 @@ static void virtio_ccw_int_handler(struct ccw_device *cdev, vcdev->err = -EIO ; }virtio_ccw_check_activity (vcdev, activity); +spin_lock_irqsave (&vcdev->irq_lock, flags);for_each_set_bit (i,indicators (vcdev),sizeof (*indicators (vcdev)) *BITS_PER_BYTE ) {/* The bit clear must happen before the vring kick. */ @@ -1114,6 +1139,7 @@ static void virtio_ccw_int_handler(struct ccw_device *cdev, vq =virtio_ccw_vq_by_ind (vcdev, i);vring_interrupt (0 , vq); } +spin_unlock_irqrestore (&vcdev->irq_lock, flags);if (test_bit (0 ,indicators2 (vcdev))) {virtio_config_changed (&vcdev->vdev);clear_bit (0 ,indicators2 (vcdev));@@ -1284,6 +1310,7 @@ static int virtio_ccw_online(struct ccw_device *cdev)init_waitqueue_head (&vcdev->wait_q);INIT_LIST_HEAD (&vcdev->virtqueues);spin_lock_init (&vcdev->lock); +spin_lock_init (&vcdev->irq_lock);mutex_init (&vcdev->io_lock);spin_lock_irqsave (get_ccwdev_lock (cdev), flags); --2.25.1
PATCH v3 7/9 virtio: allow to unbreak virtqueue
In reply to Jason WangThis patch allows the virtio_break_device() to accept a boolean value then we can unbreak the virtqueue.
--- drivers/char/virtio_console.c | 2 +- drivers/crypto/virtio/virtio_crypto_core.c | 2 +- drivers/s390/virtio/virtio_ccw.c | 4 ++-- drivers/virtio/virtio_pci_common.c | 2 +- drivers/virtio/virtio_ring.c | 4 ++-- include/linux/virtio.h | 2 +- 6 files changed, 8 insertions(+), 8 deletions(-)
drivers/char/virtio_console.c +1 -1
@@ -1958,7 +1958,7 @@ static void virtcons_remove(struct virtio_device *vdev)spin_unlock_irq (&pdrvdata_lock);/* Device is going away, exit any polling for buffers */ -virtio_break_device (vdev); +virtio_break_device (vdev, true);if (use_multiport (portdev))flush_work (&portdev->control_work);else
drivers/crypto/virtio/virtio_crypto_core.c +1 -1
@@ -215,7 +215,7 @@ static int virtcrypto_update_status(struct virtio_crypto *vcrypto)dev_warn (&vcrypto->vdev->dev,"Unknown status bits: 0x%x\n" , status); -virtio_break_device (vcrypto->vdev); +virtio_break_device (vcrypto->vdev, true);return -EPERM ; }
drivers/s390/virtio/virtio_ccw.c +2 -2
@@ -1211,7 +1211,7 @@ static void virtio_ccw_remove(struct ccw_device *cdev)if (vcdev && cdev->online) {if (vcdev->device_lost) -virtio_break_device (&vcdev->vdev); +virtio_break_device (&vcdev->vdev, true);unregister_virtio_device (&vcdev->vdev);spin_lock_irqsave (get_ccwdev_lock (cdev), flags);dev_set_drvdata (&cdev->dev,NULL );
@@ -1228,7 +1228,7 @@ static int virtio_ccw_offline(struct ccw_device *cdev)if (!vcdev)return 0 ;if (vcdev->device_lost) -virtio_break_device (&vcdev->vdev); +virtio_break_device (&vcdev->vdev, true);unregister_virtio_device (&vcdev->vdev);spin_lock_irqsave (get_ccwdev_lock (cdev), flags);dev_set_drvdata (&cdev->dev,NULL );
drivers/virtio/virtio_pci_common.c +1 -1
@@ -583,7 +583,7 @@ static void virtio_pci_remove(struct pci_dev *pci_dev)* layers can abort any ongoing operation. */ if (!pci_device_is_present (pci_dev)) -virtio_break_device (&vp_dev->vdev); +virtio_break_device (&vp_dev->vdev, true);pci_disable_sriov (pci_dev);
drivers/virtio/virtio_ring.c +2 -2
@@ -2382,7 +2382,7 @@ EXPORT_SYMBOL_GPL(virtqueue_is_broken);* This should prevent the device from being used, allowing drivers to * recover. You may need to grab appropriate locks to flush. */ -void virtio_break_device (struct virtio_device *dev) +void virtio_break_device (struct virtio_device *dev,bool broken) {struct virtqueue *_vq;
@@ -2391,7 +2391,7 @@ void virtio_break_device(struct virtio_device *dev)struct vring_virtqueue *vq =to_vvq (_vq);/* Pairs with READ_ONCE() in virtqueue_is_broken(). */ -WRITE_ONCE (vq->broken, true); +WRITE_ONCE (vq->broken, broken); }spin_unlock (&dev->vqs_list_lock); }
include/linux/virtio.h +1 -1
@@ -130,7 +130,7 @@ int register_virtio_device(struct virtio_device *dev);void unregister_virtio_device (struct virtio_device *dev);bool is_virtio_device (struct device *dev); -void virtio_break_device (struct virtio_device *dev); +void virtio_break_device (struct virtio_device *dev,bool broken);void virtio_config_changed (struct virtio_device *dev);#ifdef CONFIG_PM_SLEEP
Signature · 6 lines
2.25.1 _______________________________________________ Virtualization mailing list [email protected] https://lists.linuxfoundation.org/mailman/listinfo/virtualization
Cornelia Huck
maintainerRe: [PATCH V3 7/9] virtio: allow to unbreak virtqueue
Quoted 79 lines · Jason Wang
This patch allows the
virtio_break_device()to accept a boolean value then we can unbreak the virtqueue.Signed-off-by Jason Wang <[email protected]>--- drivers/char/virtio_console.c | 2 +- drivers/crypto/virtio/virtio_crypto_core.c | 2 +- drivers/s390/virtio/virtio_ccw.c | 4 ++-- drivers/virtio/virtio_pci_common.c | 2 +- drivers/virtio/virtio_ring.c | 4 ++-- include/linux/virtio.h | 2 +- 6 files changed, 8 insertions(+), 8 deletions(-)drivers/char/virtio_console.c +1
-1@@ -1958,7 +1958,7 @@ static void virtcons_remove(struct virtio_device *vdev)spin_unlock_irq (&pdrvdata_lock);/* Device is going away, exit any polling for buffers */ -virtio_break_device (vdev); +virtio_break_device (vdev, true);if (use_multiport (portdev))flush_work (&portdev->control_work);else drivers/crypto/virtio/virtio_crypto_core.c +1
-1@@ -215,7 +215,7 @@ static int virtcrypto_update_status(struct virtio_crypto *vcrypto)dev_warn (&vcrypto->vdev->dev,"Unknown status bits: 0x%x\n" , status); -virtio_break_device (vcrypto->vdev); +virtio_break_device (vcrypto->vdev, true);return -EPERM ; }drivers/s390/virtio/virtio_ccw.c +2
-2@@ -1211,7 +1211,7 @@ static void virtio_ccw_remove(struct ccw_device *cdev)if (vcdev && cdev->online) {if (vcdev->device_lost) -virtio_break_device (&vcdev->vdev); +virtio_break_device (&vcdev->vdev, true);unregister_virtio_device (&vcdev->vdev);spin_lock_irqsave (get_ccwdev_lock (cdev), flags);dev_set_drvdata (&cdev->dev,NULL );@@ -1228,7 +1228,7 @@ static int virtio_ccw_offline(struct ccw_device *cdev)if (!vcdev)return 0 ;if (vcdev->device_lost) -virtio_break_device (&vcdev->vdev); +virtio_break_device (&vcdev->vdev, true);unregister_virtio_device (&vcdev->vdev);spin_lock_irqsave (get_ccwdev_lock (cdev), flags);dev_set_drvdata (&cdev->dev,NULL );drivers/virtio/virtio_pci_common.c +1
-1@@ -583,7 +583,7 @@ static void virtio_pci_remove(struct pci_dev *pci_dev)* layers can abort any ongoing operation. */ if (!pci_device_is_present (pci_dev)) -virtio_break_device (&vp_dev->vdev); +virtio_break_device (&vp_dev->vdev, true);pci_disable_sriov (pci_dev);drivers/virtio/virtio_ring.c +0
-0
drivers/virtio/virtio_ring.c EXPORT_SYMBOL_GPL(virtqueue_is_broken); · line 2385
Quoted 5 lines
drivers/virtio/virtio_ring.c +1
-1@@ -2382,7 +2382,7 @@ EXPORT_SYMBOL_GPL(virtqueue_is_broken);* This should prevent the device from being used, allowing drivers to * recover. You may need to grab appropriate locks to flush. */ -void virtio_break_device (struct virtio_device *dev) +void virtio_break_device (struct virtio_device *dev,bool broken)
I think we need to be careful to say when it is safe to unset 'broken'.
The current callers set all queues to broken in case of surprise removal (ccw, pci), removal (console), or the device behaving badly (crypto). There's also code setting individual queues to broken. We do not want to undo any of these, unless the device has gone through a reset in the meanwhile. Maybe add:
"It is only safe to call this function to *remove* the broken flag for a device that is (re)transitioning to becoming usable; calling it that way during normal usage may have unpredictable consequences."
(Not sure how to word this; especially if we consider future usage of queue reset.)
_______________________________________________
Virtualization mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/virtualization
Michael S. Tsirkin
maintainerRe: [PATCH V3 7/9] virtio: allow to unbreak virtqueue
Quoted 95 lines · Cornelia Huck
Quoted 79 lines · Jason Wang
This patch allows the
virtio_break_device()to accept a boolean value then we can unbreak the virtqueue.Signed-off-by Jason Wang <[email protected]>--- drivers/char/virtio_console.c | 2 +- drivers/crypto/virtio/virtio_crypto_core.c | 2 +- drivers/s390/virtio/virtio_ccw.c | 4 ++-- drivers/virtio/virtio_pci_common.c | 2 +- drivers/virtio/virtio_ring.c | 4 ++-- include/linux/virtio.h | 2 +- 6 files changed, 8 insertions(+), 8 deletions(-)drivers/char/virtio_console.c +1
-1@@ -1958,7 +1958,7 @@ static void virtcons_remove(struct virtio_device *vdev)spin_unlock_irq (&pdrvdata_lock);/* Device is going away, exit any polling for buffers */ -virtio_break_device (vdev); +virtio_break_device (vdev, true);if (use_multiport (portdev))flush_work (&portdev->control_work);else drivers/crypto/virtio/virtio_crypto_core.c +1
-1@@ -215,7 +215,7 @@ static int virtcrypto_update_status(struct virtio_crypto *vcrypto)dev_warn (&vcrypto->vdev->dev,"Unknown status bits: 0x%x\n" , status); -virtio_break_device (vcrypto->vdev); +virtio_break_device (vcrypto->vdev, true);return -EPERM ; }drivers/s390/virtio/virtio_ccw.c +2
-2@@ -1211,7 +1211,7 @@ static void virtio_ccw_remove(struct ccw_device *cdev)if (vcdev && cdev->online) {if (vcdev->device_lost) -virtio_break_device (&vcdev->vdev); +virtio_break_device (&vcdev->vdev, true);unregister_virtio_device (&vcdev->vdev);spin_lock_irqsave (get_ccwdev_lock (cdev), flags);dev_set_drvdata (&cdev->dev,NULL );@@ -1228,7 +1228,7 @@ static int virtio_ccw_offline(struct ccw_device *cdev)if (!vcdev)return 0 ;if (vcdev->device_lost) -virtio_break_device (&vcdev->vdev); +virtio_break_device (&vcdev->vdev, true);unregister_virtio_device (&vcdev->vdev);spin_lock_irqsave (get_ccwdev_lock (cdev), flags);dev_set_drvdata (&cdev->dev,NULL );drivers/virtio/virtio_pci_common.c +1
-1@@ -583,7 +583,7 @@ static void virtio_pci_remove(struct pci_dev *pci_dev)* layers can abort any ongoing operation. */ if (!pci_device_is_present (pci_dev)) -virtio_break_device (&vp_dev->vdev); +virtio_break_device (&vp_dev->vdev, true);pci_disable_sriov (pci_dev);drivers/virtio/virtio_ring.c +0
-0drivers/virtio/virtio_ring.c EXPORT_SYMBOL_GPL(virtqueue_is_broken); · line 2385
Quoted 5 lines
drivers/virtio/virtio_ring.c +1
-1@@ -2382,7 +2382,7 @@ EXPORT_SYMBOL_GPL(virtqueue_is_broken);* This should prevent the device from being used, allowing drivers to * recover. You may need to grab appropriate locks to flush. */ -void virtio_break_device (struct virtio_device *dev) +void virtio_break_device (struct virtio_device *dev,bool broken)I think we need to be careful to say when it is safe to unset 'broken'.
The current callers set all queues to broken in case of surprise removal (ccw, pci), removal (console), or the device behaving badly (crypto). There's also code setting individual queues to broken. We do not want to undo any of these, unless the device has gone through a reset in the meanwhile. Maybe add:
Quoted 6 lines
"It is only safe to call this function to *remove* the broken flag for a device that is (re)transitioning to becoming usable; calling it that way during normal usage may have unpredictable consequences."
(Not sure how to word this; especially if we consider future usage of queue reset.)
Right. I would prefer __virtio_unbreak_device or something similar with a bit comment explaining it's only safe to call during probe.
Signature · 6 lines
MST _______________________________________________ Virtualization mailing list [email protected] https://lists.linuxfoundation.org/mailman/listinfo/virtualization
Re: [PATCH V3 7/9] virtio: allow to unbreak virtqueue
Quoted 95 lines · Michael S. Tsirkin
Quoted 94 lines · Cornelia Huck
Quoted 79 lines · Jason Wang
This patch allows the
virtio_break_device()to accept a boolean value then we can unbreak the virtqueue.Signed-off-by Jason Wang <[email protected]>--- drivers/char/virtio_console.c | 2 +- drivers/crypto/virtio/virtio_crypto_core.c | 2 +- drivers/s390/virtio/virtio_ccw.c | 4 ++-- drivers/virtio/virtio_pci_common.c | 2 +- drivers/virtio/virtio_ring.c | 4 ++-- include/linux/virtio.h | 2 +- 6 files changed, 8 insertions(+), 8 deletions(-)drivers/char/virtio_console.c +1
-1@@ -1958,7 +1958,7 @@ static void virtcons_remove(struct virtio_device *vdev)spin_unlock_irq (&pdrvdata_lock);/* Device is going away, exit any polling for buffers */ -virtio_break_device (vdev); +virtio_break_device (vdev, true);if (use_multiport (portdev))flush_work (&portdev->control_work);else drivers/crypto/virtio/virtio_crypto_core.c +1
-1@@ -215,7 +215,7 @@ static int virtcrypto_update_status(struct virtio_crypto *vcrypto)dev_warn (&vcrypto->vdev->dev,"Unknown status bits: 0x%x\n" , status); -virtio_break_device (vcrypto->vdev); +virtio_break_device (vcrypto->vdev, true);return -EPERM ; }drivers/s390/virtio/virtio_ccw.c +2
-2@@ -1211,7 +1211,7 @@ static void virtio_ccw_remove(struct ccw_device *cdev)if (vcdev && cdev->online) {if (vcdev->device_lost) -virtio_break_device (&vcdev->vdev); +virtio_break_device (&vcdev->vdev, true);unregister_virtio_device (&vcdev->vdev);spin_lock_irqsave (get_ccwdev_lock (cdev), flags);dev_set_drvdata (&cdev->dev,NULL );@@ -1228,7 +1228,7 @@ static int virtio_ccw_offline(struct ccw_device *cdev)if (!vcdev)return 0 ;if (vcdev->device_lost) -virtio_break_device (&vcdev->vdev); +virtio_break_device (&vcdev->vdev, true);unregister_virtio_device (&vcdev->vdev);spin_lock_irqsave (get_ccwdev_lock (cdev), flags);dev_set_drvdata (&cdev->dev,NULL );drivers/virtio/virtio_pci_common.c +1
-1@@ -583,7 +583,7 @@ static void virtio_pci_remove(struct pci_dev *pci_dev)* layers can abort any ongoing operation. */ if (!pci_device_is_present (pci_dev)) -virtio_break_device (&vp_dev->vdev); +virtio_break_device (&vp_dev->vdev, true);pci_disable_sriov (pci_dev);drivers/virtio/virtio_ring.c +0
-0drivers/virtio/virtio_ring.c EXPORT_SYMBOL_GPL(virtqueue_is_broken); · line 2385
Quoted 5 lines
drivers/virtio/virtio_ring.c +1
-1@@ -2382,7 +2382,7 @@ EXPORT_SYMBOL_GPL(virtqueue_is_broken);* This should prevent the device from being used, allowing drivers to * recover. You may need to grab appropriate locks to flush. */ -void virtio_break_device (struct virtio_device *dev) +void virtio_break_device (struct virtio_device *dev,bool broken)I think we need to be careful to say when it is safe to unset 'broken'.
The current callers set all queues to broken in case of surprise removal (ccw, pci), removal (console), or the device behaving badly (crypto). There's also code setting individual queues to broken. We do not want to undo any of these, unless the device has gone through a reset in the meanwhile. Maybe add:
Quoted 8 lines
Quoted 6 lines
"It is only safe to call this function to *remove* the broken flag for a device that is (re)transitioning to becoming usable; calling it that way during normal usage may have unpredictable consequences."
(Not sure how to word this; especially if we consider future usage of queue reset.)
Right. I would prefer
__virtio_unbreak_deviceor something similar
Ok, so I think then we don't want it to be exported.
Quoted 1 line
with a bit comment explaining it's only safe to call during probe.
Will do.
Thanks
Quoted 1 line
_______________________________________________
Virtualization mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/virtualization
PATCH v3 8/9 virtio: harden vring IRQ
In reply to Jason WangThis is a rework on the previous IRQ hardening that is done for virtio-pci where several drawbacks were found and were reverted:
1) try to use IRQF_NO_AUTOEN which is not friendly to affinity managed IRQ that is used by some device such as virtio-blk
2) done only for PCI transport
The vq->broken is re-used in this patch for implementing the IRQ hardening. The vq->broken is set to true during both initialization and reset. And the vq->broken is set to false in virtio_device_ready(). Then vring_interrupt can check and return when vq->broken is true. And in this case, switch to return IRQ_NONE to let the interrupt core aware of such invalid interrupt to prevent IRQ storm.
The reason of using a per queue variable instead of a per device one is that we may need it for per queue reset hardening in the future.
Note that the hardening is only done for vring interrupt since the config interrupt hardening is already done in commit 22b7050a024d7 ("virtio: defer config changed notifications"). But the method that is used by config interrupt can't be reused by the vring interrupt handler because it uses spinlock to do the synchronization which is expensive.
--- drivers/virtio/virtio.c | 15 ++++++++++++--- drivers/virtio/virtio_ring.c | 11 +++++++---- include/linux/virtio_config.h | 12 ++++++++++++ 3 files changed, 31 insertions(+), 7 deletions(-)
drivers/virtio/virtio.c +12 -3
@@ -220,6 +220,15 @@ static int virtio_features_ok(struct virtio_device *dev)* */ void virtio_reset_device (struct virtio_device *dev) { +/* +* The below virtio_synchronize_cbs() guarantees that any +* interrupt for this line arriving after +* virtio_synchronize_vqs() has completed is guaranteed to see +* driver_ready == false. +*/ +virtio_break_device (dev, true); +virtio_synchronize_cbs (dev); + dev->config->reset (dev); }EXPORT_SYMBOL_GPL (virtio_reset_device);
@@ -428,6 +437,9 @@ int register_virtio_device(struct virtio_device *dev) dev->config_enabled = false; dev->config_change_pending = false; +INIT_LIST_HEAD (&dev->vqs); +spin_lock_init (&dev->vqs_list_lock); +/* We always start by resetting the device, in case a previous * driver messed it up. This also tests that code path a little. */ virtio_reset_device (dev);
@@ -435,9 +447,6 @@ int register_virtio_device(struct virtio_device *dev)/* Acknowledge that we've seen the device. */ virtio_add_status (dev,VIRTIO_CONFIG_S_ACKNOWLEDGE ); -INIT_LIST_HEAD (&dev->vqs); -spin_lock_init (&dev->vqs_list_lock); -/* * device_add() causes the bus infrastructure to look for a matching * driver.
drivers/virtio/virtio_ring.c +7 -4
@@ -1690,7 +1690,7 @@ static struct virtqueue *vring_create_virtqueue_packed( vq->we_own_ring = true; vq->notify = notify; vq->weak_barriers = weak_barriers; - vq->broken = false; + vq->broken = true; vq->last_used_idx =0 ; vq->event_triggered = false; vq->num_added =0 ;
@@ -2136,8 +2136,11 @@ irqreturn_t vring_interrupt(int irq, void *_vq)return IRQ_NONE ; } -if (unlikely (vq->broken)) -return IRQ_HANDLED ; +if (unlikely (vq->broken)) { +dev_warn_once (&vq->vq.vdev->dev, +"virtio vring IRQ raised before DRIVER_OK" ); +return IRQ_NONE ; + }/* Just a hint for performance: so it's ok that this can be racy! */ if (vq->event)
@@ -2179,7 +2182,7 @@ struct virtqueue *__vring_new_virtqueue(unsigned int index, vq->we_own_ring = false; vq->notify = notify; vq->weak_barriers = weak_barriers; - vq->broken = false; + vq->broken = true; vq->last_used_idx =0 ; vq->event_triggered = false; vq->num_added =0 ;
include/linux/virtio_config.h +12 -0
@@ -255,6 +255,18 @@ void virtio_device_ready(struct virtio_device *dev)unsigned status = dev->config->get_status (dev);BUG_ON (status &VIRTIO_CONFIG_S_DRIVER_OK ); + +/* +* The virtio_synchronize_cbs() makes sure vring_interrupt() +* will see the driver specific setup if it sees vq->broken +* as false. +*/ +virtio_synchronize_cbs (dev); +virtio_break_device (dev, false); +/* +* The transport is expected ensure the visibility of +* vq->broken before setting VIRTIO_CONFIG_S_DRIVER_OK. +*/ dev->config->set_status (dev, status |VIRTIO_CONFIG_S_DRIVER_OK ); }
Signature · 6 lines
2.25.1 _______________________________________________ Virtualization mailing list [email protected] https://lists.linuxfoundation.org/mailman/listinfo/virtualization
PATCH v3 9/9 virtio: use WARN_ON() to warning illegal status value
In reply to Jason WangWe used to use BUG_ON() in virtio_device_ready() to detect illegal status value, this seems sub-optimal since the value is under the control of the device. Switch to use WARN_ON() instead.
--- include/linux/virtio_config.h | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
include/linux/virtio_config.h +1 -1
@@ -254,7 +254,7 @@ void virtio_device_ready(struct virtio_device *dev) {unsigned status = dev->config->get_status (dev); -BUG_ON (status &VIRTIO_CONFIG_S_DRIVER_OK ); +WARN_ON (status &VIRTIO_CONFIG_S_DRIVER_OK );/* * The virtio_synchronize_cbs() makes sure vring_interrupt()
Signature · 6 lines
2.25.1 _______________________________________________ Virtualization mailing list [email protected] https://lists.linuxfoundation.org/mailman/listinfo/virtualization
Quoted 3 lines · Jason Wang
Hi All:
Forgot to add Xuan to the series.
Thanks
Quoted 66 lines
This is a rework on the IRQ hardening for virtio which is done previously by the following commits are reverted:
9e35276a5344("virtio_pci: harden MSI-X interrupts")080cd7c3ac87("virtio-pci: harden INTX interrupts")The reason is that it depends on the
IRQF_NO_AUTOENwhich may conflict with the assumption of the affinity managed IRQ that is used by some virtio drivers. And what's more, it is only done for virtio-pci but not other transports.In this rework, I try to implement a general virtio solution which borrows the idea of the INTX hardening by re-using per virtqueue boolean
vq->brokenand toggle it invirtio_device_ready()andvirtio_reset_device(). Then we can simply reuse the existing checks in thevring_interrupt()and return early if the driver is not ready.Note that, I only did compile test on ccw and MMIO transport.
Please review.
Changes since v1:
- Use transport specific irq synchronization method when possible
- Drop the module parameter and enable the hardening unconditonally
- Tweak the barrier/ordering facilities used in the code
- Reanmeirq_soft_enabledtodriver_ready
- Avoid unnecssary IRQ synchornization (e.g during boot)Changes since V2:
- add ccw and MMIO support
- renamesynchronize_vqs()tosynchronize_cbs()
- switch to re-usevq->brokeninstead of introducing new device attributes for the future virtqueue reset support
- remove unnecssary READ_ONCE()/WRITE_ONCE()
- a new patch to remove device triggerableBUG_ON()
- more tweaks on the commentsJason Wang (8):
virtio: usevirtio_reset_device()when possible
virtio: introduce config op to synchronize vring callbacks
virtio-pci: implementsynchronize_cbs()
virtio-mmio: implementsynchronize_cbs()
virtio-ccw: implementsynchronize_cbs()
virtio: allow to unbreak virtqueue
virtio: harden vring IRQ
virtio: useWARN_ON()to warning illegal status valueStefano Garzarella (1):
virtio: usevirtio_device_ready()invirtio_device_restore()drivers/char/virtio_console.c | 2 +- drivers/crypto/virtio/virtio_crypto_core.c | 2 +- drivers/s390/virtio/virtio_ccw.c | 31 ++++++++++++++++-- drivers/virtio/virtio.c | 24 ++++++++++---- drivers/virtio/virtio_mmio.c | 9 +++++ drivers/virtio/virtio_pci_common.c | 2 +- drivers/virtio/virtio_pci_common.h | 2 ++ drivers/virtio/virtio_pci_legacy.c | 1 + drivers/virtio/virtio_pci_modern.c | 2 ++ drivers/virtio/virtio_ring.c | 15 +++++---- include/linux/virtio.h | 2 +- include/linux/virtio_config.h | 38 +++++++++++++++++++++- 12 files changed, 110 insertions(+), 20 deletions(-)
Quoted 4 lines
--
2.25.1
_______________________________________________
Virtualization mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/virtualization