Re: [PATCH v8 20/22] counter: Implement events_queue_size sysfs attribute
From: William Breathitt Gray
Date: Thu Feb 18 2021 - 07:03:50 EST
On Sun, Feb 14, 2021 at 06:11:46PM +0000, Jonathan Cameron wrote:
> On Fri, 12 Feb 2021 21:13:44 +0900
> William Breathitt Gray <vilhelm.gray@xxxxxxxxx> wrote:
>
> > The events_queue_size sysfs attribute provides a way for users to
> > dynamically configure the Counter events queue size for the Counter
> > character device interface. The size is in number of struct
> > counter_event data structures. The number of elements will be rounded-up
> > to a power of 2 due to a requirement of the kfifo_alloc function called
> > during reallocation of the queue.
> >
> > Cc: Oleksij Rempel <o.rempel@xxxxxxxxxxxxxx>
> > Signed-off-by: William Breathitt Gray <vilhelm.gray@xxxxxxxxx>
> > ---
> > Documentation/ABI/testing/sysfs-bus-counter | 8 +++++++
> > drivers/counter/counter-chrdev.c | 23 +++++++++++++++++++
> > drivers/counter/counter-chrdev.h | 2 ++
> > drivers/counter/counter-sysfs.c | 25 +++++++++++++++++++++
> > 4 files changed, 58 insertions(+)
> >
> > diff --git a/Documentation/ABI/testing/sysfs-bus-counter b/Documentation/ABI/testing/sysfs-bus-counter
> > index 847e96f19d19..f6cb2a8b08a7 100644
> > --- a/Documentation/ABI/testing/sysfs-bus-counter
> > +++ b/Documentation/ABI/testing/sysfs-bus-counter
> > @@ -212,6 +212,14 @@ Description:
> > both edges:
> > Any state transition.
> >
> > +What: /sys/bus/counter/devices/counterX/events_queue_size
> > +KernelVersion: 5.13
> > +Contact: linux-iio@xxxxxxxxxxxxxxx
> > +Description:
> > + Size of the Counter events queue in number of struct
> > + counter_event data structures. The number of elements will be
> > + rounded-up to a power of 2.
> > +
> > What: /sys/bus/counter/devices/counterX/name
> > KernelVersion: 5.2
> > Contact: linux-iio@xxxxxxxxxxxxxxx
> > diff --git a/drivers/counter/counter-chrdev.c b/drivers/counter/counter-chrdev.c
> > index 16f02df7f73d..53eea894e13f 100644
> > --- a/drivers/counter/counter-chrdev.c
> > +++ b/drivers/counter/counter-chrdev.c
> > @@ -375,6 +375,29 @@ void counter_chrdev_remove(struct counter_device *const counter)
> > cdev_del(&counter->chrdev);
> > }
> >
> > +int counter_chrdev_realloc_queue(struct counter_device *const counter,
> > + size_t queue_size)
> > +{
> > + int err;
> > + DECLARE_KFIFO_PTR(events, struct counter_event);
> > + unsigned long flags;
> > +
> > + /* Allocate new events queue */
> > + err = kfifo_alloc(&events, queue_size, GFP_ATOMIC);
>
> Is there any potential for losing events?
We take the events_list_lock down below so we're safe against missing an
event, but past events currently unread in the queue will be lost.
Shortening the size of the queue is inherently a destructive process if
we have more events in the current queue than can fit in the new queue.
Because we a liable to lose some events in such a case, I think it's
best to keep the behavior of this reallocation consistent and have it
provide a fresh empty queue every time, as opposed to sometimes dropping
events and sometimes not.
I also suspect an actual user would be setting the size of their queue
to the required amount before they begin watching events, rather than
adjusting it sporadically during a live operation.
> > + if (err)
> > + return err;
> > +
> > + raw_spin_lock_irqsave(&counter->events_list_lock, flags);
> > +
> > + /* Swap in new events queue */
> > + kfifo_free(&counter->events);
> > + counter->events.kfifo = events.kfifo;
> > +
> > + raw_spin_unlock_irqrestore(&counter->events_list_lock, flags);
> > +
> > + return 0;
> > +}
> > +
> > static int counter_get_data(struct counter_device *const counter,
> > const struct counter_comp_node *const comp_node,
> > u64 *const value)
> > diff --git a/drivers/counter/counter-chrdev.h b/drivers/counter/counter-chrdev.h
> > index cf5a318fe540..ff7fb0191852 100644
> > --- a/drivers/counter/counter-chrdev.h
> > +++ b/drivers/counter/counter-chrdev.h
> > @@ -12,5 +12,7 @@
> > int counter_chrdev_add(struct counter_device *const counter,
> > const dev_t counter_devt);
> > void counter_chrdev_remove(struct counter_device *const counter);
> > +int counter_chrdev_realloc_queue(struct counter_device *const counter,
> > + size_t queue_size);
> >
> > #endif /* _COUNTER_CHRDEV_H_ */
> > diff --git a/drivers/counter/counter-sysfs.c b/drivers/counter/counter-sysfs.c
> > index 0cb3dba950bc..9abc821a3871 100644
> > --- a/drivers/counter/counter-sysfs.c
> > +++ b/drivers/counter/counter-sysfs.c
> > @@ -13,6 +13,7 @@
> > #include <linux/sysfs.h>
> > #include <linux/types.h>
> >
> > +#include "counter-chrdev.h"
> > #include "counter-sysfs.h"
> >
> > /**
> > @@ -737,12 +738,30 @@ static int counter_num_counts_read(struct counter_device *counter, u8 *val)
> > return 0;
> > }
> >
> > +static int counter_events_queue_size_read(struct counter_device *counter,
> > + u64 *val)
> > +{
> > + *val = counter->events.kfifo.mask + 1;
>
> kfifo_size() rather than open coding it?
Ack.
William Breathitt Gray
> > + return 0;
> > +}
> > +
> > +static int counter_events_queue_size_write(struct counter_device *counter,
> > + u64 val)
> > +{
> > + return counter_chrdev_realloc_queue(counter, val);
> > +}
> > +
> > static struct counter_comp counter_num_signals_comp =
> > COUNTER_COMP_DEVICE_U8("num_signals", counter_num_signals_read, NULL);
> >
> > static struct counter_comp counter_num_counts_comp =
> > COUNTER_COMP_DEVICE_U8("num_counts", counter_num_counts_read, NULL);
> >
> > +static struct counter_comp counter_events_queue_size_comp =
> > + COUNTER_COMP_DEVICE_U64("events_queue_size",
> > + counter_events_queue_size_read,
> > + counter_events_queue_size_write);
> > +
> > static int counter_sysfs_attr_add(struct counter_device *const counter,
> > struct counter_attribute_group *group)
> > {
> > @@ -781,6 +800,12 @@ static int counter_sysfs_attr_add(struct counter_device *const counter,
> > if (err < 0)
> > return err;
> >
> > + /* Create num_counts attribute */
> > + err = counter_attr_create(dev, group, &counter_events_queue_size_comp,
> > + scope, NULL);
> > + if (err < 0)
> > + return err;
> > +
> > /* Create an attribute for each extension */
> > for (i = 0; i < counter->num_ext; i++) {
> > ext = counter->ext + i;
>
Attachment:
signature.asc
Description: PGP signature