Re: [PATCH v3 05/11] vduse: Support automatic irq callback affinity
From: Yongji Xie
Date: Fri Mar 17 2023 - 03:05:15 EST
On Thu, Mar 16, 2023 at 5:03 PM Jason Wang <jasowang@xxxxxxxxxx> wrote:
>
>
> 在 2023/2/28 17:41, Xie Yongji 写道:
> > This brings current interrupt affinity spreading mechanism
> > to vduse device. We will make use of group_cpus_evenly()
> > to create an irq callback affinity mask for each virtqueue of
> > vduse device. Then we will spread IRQs between CPUs in the affinity
> > mask, in a round-robin manner, to run the irq callback.
> >
> > Signed-off-by: Xie Yongji <xieyongji@xxxxxxxxxxxxx>
> > ---
> > drivers/vdpa/vdpa_user/vduse_dev.c | 130 +++++++++++++++++++++++++++--
> > 1 file changed, 123 insertions(+), 7 deletions(-)
> >
> > diff --git a/drivers/vdpa/vdpa_user/vduse_dev.c b/drivers/vdpa/vdpa_user/vduse_dev.c
> > index 98359d87a06f..bde28a8692d5 100644
> > --- a/drivers/vdpa/vdpa_user/vduse_dev.c
> > +++ b/drivers/vdpa/vdpa_user/vduse_dev.c
> > @@ -23,6 +23,8 @@
> > #include <linux/nospec.h>
> > #include <linux/vmalloc.h>
> > #include <linux/sched/mm.h>
> > +#include <linux/interrupt.h>
> > +#include <linux/group_cpus.h>
> > #include <uapi/linux/vduse.h>
> > #include <uapi/linux/vdpa.h>
> > #include <uapi/linux/virtio_config.h>
> > @@ -41,6 +43,8 @@
> > #define VDUSE_IOVA_SIZE (128 * 1024 * 1024)
> > #define VDUSE_MSG_DEFAULT_TIMEOUT 30
> >
> > +#define IRQ_UNBOUND -1
> > +
> > struct vduse_virtqueue {
> > u16 index;
> > u16 num_max;
> > @@ -57,6 +61,8 @@ struct vduse_virtqueue {
> > struct vdpa_callback cb;
> > struct work_struct inject;
> > struct work_struct kick;
> > + int irq_effective_cpu;
> > + struct cpumask irq_affinity;
> > };
> >
> > struct vduse_dev;
> > @@ -128,6 +134,7 @@ static struct class *vduse_class;
> > static struct cdev vduse_ctrl_cdev;
> > static struct cdev vduse_cdev;
> > static struct workqueue_struct *vduse_irq_wq;
> > +static struct workqueue_struct *vduse_irq_bound_wq;
> >
> > static u32 allowed_device_id[] = {
> > VIRTIO_ID_BLOCK,
> > @@ -708,6 +715,82 @@ static u32 vduse_vdpa_get_generation(struct vdpa_device *vdpa)
> > return dev->generation;
> > }
> >
> > +static void default_calc_sets(struct irq_affinity *affd, unsigned int affvecs)
> > +{
> > + affd->nr_sets = 1;
> > + affd->set_size[0] = affvecs;
> > +}
> > +
> > +struct cpumask *
> > +create_affinity_masks(unsigned int nvecs, struct irq_affinity *affd)
> > +{
> > + unsigned int affvecs = 0, curvec, usedvecs, i;
> > + struct cpumask *masks = NULL;
> > +
> > + if (nvecs > affd->pre_vectors + affd->post_vectors)
> > + affvecs = nvecs - affd->pre_vectors - affd->post_vectors;
> > +
> > + if (!affd->calc_sets)
> > + affd->calc_sets = default_calc_sets;
> > +
> > + affd->calc_sets(affd, affvecs);
> > +
> > + if (!affvecs)
> > + return NULL;
> > +
> > + masks = kcalloc(nvecs, sizeof(*masks), GFP_KERNEL);
> > + if (!masks)
> > + return NULL;
> > +
> > + /* Fill out vectors at the beginning that don't need affinity */
> > + for (curvec = 0; curvec < affd->pre_vectors; curvec++)
> > + cpumask_setall(&masks[curvec]);
> > +
> > + for (i = 0, usedvecs = 0; i < affd->nr_sets; i++) {
> > + unsigned int this_vecs = affd->set_size[i];
> > + int j;
> > + struct cpumask *result = group_cpus_evenly(this_vecs);
> > +
> > + if (!result) {
> > + kfree(masks);
> > + return NULL;
> > + }
> > +
> > + for (j = 0; j < this_vecs; j++)
> > + cpumask_copy(&masks[curvec + j], &result[j]);
> > + kfree(result);
> > +
> > + curvec += this_vecs;
> > + usedvecs += this_vecs;
> > + }
> > +
> > + /* Fill out vectors at the end that don't need affinity */
> > + if (usedvecs >= affvecs)
> > + curvec = affd->pre_vectors + affvecs;
> > + else
> > + curvec = affd->pre_vectors + usedvecs;
> > + for (; curvec < nvecs; curvec++)
> > + cpumask_setall(&masks[curvec]);
> > +
> > + return masks;
> > +}
> > +
> > +static void vduse_vdpa_set_irq_affinity(struct vdpa_device *vdpa,
> > + struct irq_affinity *desc)
> > +{
> > + struct vduse_dev *dev = vdpa_to_vduse(vdpa);
> > + struct cpumask *masks;
> > + int i;
> > +
> > + masks = create_affinity_masks(dev->vq_num, desc);
> > + if (!masks)
> > + return;
> > +
> > + for (i = 0; i < dev->vq_num; i++)
> > + cpumask_copy(&dev->vqs[i]->irq_affinity, &masks[i]);
> > + kfree(masks);
> > +}
> > +
> > static int vduse_vdpa_set_map(struct vdpa_device *vdpa,
> > unsigned int asid,
> > struct vhost_iotlb *iotlb)
> > @@ -758,6 +841,7 @@ static const struct vdpa_config_ops vduse_vdpa_config_ops = {
> > .get_config = vduse_vdpa_get_config,
> > .set_config = vduse_vdpa_set_config,
> > .get_generation = vduse_vdpa_get_generation,
> > + .set_irq_affinity = vduse_vdpa_set_irq_affinity,
> > .reset = vduse_vdpa_reset,
> > .set_map = vduse_vdpa_set_map,
> > .free = vduse_vdpa_free,
> > @@ -917,7 +1001,8 @@ static void vduse_vq_irq_inject(struct work_struct *work)
> > }
> >
> > static int vduse_dev_queue_irq_work(struct vduse_dev *dev,
> > - struct work_struct *irq_work)
> > + struct work_struct *irq_work,
> > + int irq_effective_cpu)
> > {
> > int ret = -EINVAL;
> >
> > @@ -926,7 +1011,11 @@ static int vduse_dev_queue_irq_work(struct vduse_dev *dev,
> > goto unlock;
> >
> > ret = 0;
> > - queue_work(vduse_irq_wq, irq_work);
> > + if (irq_effective_cpu == IRQ_UNBOUND)
> > + queue_work(vduse_irq_wq, irq_work);
> > + else
> > + queue_work_on(irq_effective_cpu,
> > + vduse_irq_bound_wq, irq_work);
> > unlock:
> > up_read(&dev->rwsem);
> >
> > @@ -1029,6 +1118,22 @@ static int vduse_dev_reg_umem(struct vduse_dev *dev,
> > return ret;
> > }
> >
> > +static void vduse_vq_update_effective_cpu(struct vduse_virtqueue *vq)
> > +{
> > + int curr_cpu = vq->irq_effective_cpu;
> > +
> > + while (true) {
> > + curr_cpu = cpumask_next(curr_cpu, &vq->irq_affinity);
> > + if (cpu_online(curr_cpu))
> > + break;
> > +
> > + if (curr_cpu >= nr_cpu_ids)
> > + curr_cpu = -1;
>
>
> IRQ_UNBOUND?
>
Will fix it.
>
> > + }
> > +
> > + vq->irq_effective_cpu = curr_cpu;
> > +}
> > +
> > static long vduse_dev_ioctl(struct file *file, unsigned int cmd,
> > unsigned long arg)
> > {
> > @@ -1111,7 +1216,7 @@ static long vduse_dev_ioctl(struct file *file, unsigned int cmd,
> > break;
> > }
> > case VDUSE_DEV_INJECT_CONFIG_IRQ:
> > - ret = vduse_dev_queue_irq_work(dev, &dev->inject);
> > + ret = vduse_dev_queue_irq_work(dev, &dev->inject, IRQ_UNBOUND);
> > break;
> > case VDUSE_VQ_SETUP: {
> > struct vduse_vq_config config;
> > @@ -1198,7 +1303,10 @@ static long vduse_dev_ioctl(struct file *file, unsigned int cmd,
> > break;
> >
> > index = array_index_nospec(index, dev->vq_num);
> > - ret = vduse_dev_queue_irq_work(dev, &dev->vqs[index]->inject);
> > +
> > + vduse_vq_update_effective_cpu(dev->vqs[index]);
> > + ret = vduse_dev_queue_irq_work(dev, &dev->vqs[index]->inject,
> > + dev->vqs[index]->irq_effective_cpu);
> > break;
> > }
> > case VDUSE_IOTLB_REG_UMEM: {
> > @@ -1367,10 +1475,12 @@ static int vduse_dev_init_vqs(struct vduse_dev *dev, u32 vq_align, u32 vq_num)
> > goto err;
> >
> > dev->vqs[i]->index = i;
> > + dev->vqs[i]->irq_effective_cpu = -1;
>
>
> IRQ_UNBOUND?
>
Will fix it.
Thanks,
Yongji