#include <linux/platform_device.h>
#include <linux/interrupt.h>
#include <linux/dma-mapping.h>
+#include <linux/dim.h>
#include <linux/slab.h>
#include "dpio.h"
spinlock_t lock_notifications;
struct list_head notifications;
struct device *dev;
+
+ /* Net DIM */
+ struct dim rx_dim;
+ /* protect against concurrent Net DIM updates */
+ spinlock_t dim_lock;
+ u16 event_ctr;
+ u64 bytes;
+ u64 frames;
};
struct dpaa2_io_store {
* potentially being migrated away.
*/
if (cpu < 0)
- cpu = smp_processor_id();
+ cpu = raw_smp_processor_id();
/* If a specific cpu was requested, pick it up immediately */
return dpio_by_cpu[cpu];
}
EXPORT_SYMBOL_GPL(dpaa2_io_service_select);
+static void dpaa2_io_dim_work(struct work_struct *w)
+{
+ struct dim *dim = container_of(w, struct dim, work);
+ struct dim_cq_moder moder =
+ net_dim_get_rx_moderation(dim->mode, dim->profile_ix);
+ struct dpaa2_io *d = container_of(dim, struct dpaa2_io, rx_dim);
+
+ dpaa2_io_set_irq_coalescing(d, moder.usec);
+ dim->state = DIM_START_MEASURE;
+}
+
/**
* dpaa2_io_create() - create a dpaa2_io object.
* @desc: the dpaa2_io descriptor
struct device *dev)
{
struct dpaa2_io *obj = kmalloc(sizeof(*obj), GFP_KERNEL);
+ u32 qman_256_cycles_per_ns;
if (!obj)
return NULL;
obj->dpio_desc = *desc;
obj->swp_desc.cena_bar = obj->dpio_desc.regs_cena;
obj->swp_desc.cinh_bar = obj->dpio_desc.regs_cinh;
+ obj->swp_desc.qman_clk = obj->dpio_desc.qman_clk;
obj->swp_desc.qman_version = obj->dpio_desc.qman_version;
+
+ /* Compute how many 256 QBMAN cycles fit into one ns. This is because
+ * the interrupt timeout period register needs to be specified in QBMAN
+ * clock cycles in increments of 256.
+ */
+ qman_256_cycles_per_ns = 256000 / (obj->swp_desc.qman_clk / 1000000);
+ obj->swp_desc.qman_256_cycles_per_ns = qman_256_cycles_per_ns;
obj->swp = qbman_swp_init(&obj->swp_desc);
if (!obj->swp) {
INIT_LIST_HEAD(&obj->node);
spin_lock_init(&obj->lock_mgmt_cmd);
spin_lock_init(&obj->lock_notifications);
+ spin_lock_init(&obj->dim_lock);
INIT_LIST_HEAD(&obj->notifications);
/* For now only enable DQRR interrupts */
obj->dev = dev;
+ memset(&obj->rx_dim, 0, sizeof(obj->rx_dim));
+ INIT_WORK(&obj->rx_dim.work, dpaa2_io_dim_work);
+ obj->event_ctr = 0;
+ obj->bytes = 0;
+ obj->frames = 0;
+
return obj;
}
struct qbman_swp *swp;
u32 status;
+ obj->event_ctr++;
+
swp = obj->swp;
status = qbman_swp_interrupt_read_status(swp);
if (!status)
qbman_eq_desc_set_no_orp(&ed, 0);
qbman_eq_desc_set_fq(&ed, fqid);
- return qbman_swp_enqueue_multiple(d->swp, &ed, fd, 0, nb);
+ return qbman_swp_enqueue_multiple(d->swp, &ed, fd, NULL, nb);
}
EXPORT_SYMBOL(dpaa2_io_service_enqueue_multiple_fq);
return 0;
}
EXPORT_SYMBOL_GPL(dpaa2_io_query_bp_count);
+
+/**
+ * dpaa2_io_set_irq_coalescing() - Set new IRQ coalescing values
+ * @d: the given DPIO object
+ * @irq_holdoff: interrupt holdoff (timeout) period in us
+ *
+ * Return 0 for success, or negative error code on error.
+ */
+int dpaa2_io_set_irq_coalescing(struct dpaa2_io *d, u32 irq_holdoff)
+{
+ struct qbman_swp *swp = d->swp;
+
+ return qbman_swp_set_irq_coalescing(swp, swp->dqrr.dqrr_size - 1,
+ irq_holdoff);
+}
+EXPORT_SYMBOL(dpaa2_io_set_irq_coalescing);
+
+/**
+ * dpaa2_io_get_irq_coalescing() - Get the current IRQ coalescing parameters
+ * @d: the given DPIO object
+ * @irq_holdoff: interrupt holdoff (timeout) period in us
+ */
+void dpaa2_io_get_irq_coalescing(struct dpaa2_io *d, u32 *irq_holdoff)
+{
+ struct qbman_swp *swp = d->swp;
+
+ qbman_swp_get_irq_coalescing(swp, NULL, irq_holdoff);
+}
+EXPORT_SYMBOL(dpaa2_io_get_irq_coalescing);
+
+/**
+ * dpaa2_io_set_adaptive_coalescing() - Enable/disable adaptive coalescing
+ * @d: the given DPIO object
+ * @use_adaptive_rx_coalesce: adaptive coalescing state
+ */
+void dpaa2_io_set_adaptive_coalescing(struct dpaa2_io *d,
+ int use_adaptive_rx_coalesce)
+{
+ d->swp->use_adaptive_rx_coalesce = use_adaptive_rx_coalesce;
+}
+EXPORT_SYMBOL(dpaa2_io_set_adaptive_coalescing);
+
+/**
+ * dpaa2_io_get_adaptive_coalescing() - Query adaptive coalescing state
+ * @d: the given DPIO object
+ *
+ * Return 1 when adaptive coalescing is enabled on the DPIO object and 0
+ * otherwise.
+ */
+int dpaa2_io_get_adaptive_coalescing(struct dpaa2_io *d)
+{
+ return d->swp->use_adaptive_rx_coalesce;
+}
+EXPORT_SYMBOL(dpaa2_io_get_adaptive_coalescing);
+
+/**
+ * dpaa2_io_update_net_dim() - Update Net DIM
+ * @d: the given DPIO object
+ * @frames: how many frames have been dequeued by the user since the last call
+ * @bytes: how many bytes have been dequeued by the user since the last call
+ */
+void dpaa2_io_update_net_dim(struct dpaa2_io *d, __u64 frames, __u64 bytes)
+{
+ struct dim_sample dim_sample = {};
+
+ if (!d->swp->use_adaptive_rx_coalesce)
+ return;
+
+ spin_lock(&d->dim_lock);
+
+ d->bytes += bytes;
+ d->frames += frames;
+
+ dim_update_sample(d->event_ctr, d->frames, d->bytes, &dim_sample);
+ net_dim(&d->rx_dim, dim_sample);
+
+ spin_unlock(&d->dim_lock);
+}
+EXPORT_SYMBOL(dpaa2_io_update_net_dim);
#define QBMAN_CINH_SWP_EQCR_AM_RT 0x980
#define QBMAN_CINH_SWP_RCR_AM_RT 0x9c0
#define QBMAN_CINH_SWP_DQPI 0xa00
+#define QBMAN_CINH_SWP_DQRR_ITR 0xa80
#define QBMAN_CINH_SWP_DCAP 0xac0
#define QBMAN_CINH_SWP_SDQCR 0xb00
#define QBMAN_CINH_SWP_EQCR_AM_RT2 0xb40
#define QBMAN_CINH_SWP_IER 0xe40
#define QBMAN_CINH_SWP_ISDR 0xe80
#define QBMAN_CINH_SWP_IIR 0xec0
+#define QBMAN_CINH_SWP_ITPR 0xf40
/* CENA register offsets */
#define QBMAN_CENA_SWP_EQCR(n) (0x000 + ((u32)(n) << 6))
& p->eqcr.pi_ci_mask;
p->eqcr.available = p->eqcr.pi_ring_size;
+ /* Initialize the software portal with a irq timeout period of 0us */
+ qbman_swp_set_irq_coalescing(p, p->dqrr.dqrr_size - 1, 0);
+
return p;
}
p = (s->addr_cena + QBMAN_CENA_SWP_EQCR(eqcr_pi & half_mask));
p[0] = cl[0] | s->eqcr.pi_vb;
if (flags && (flags[i] & QBMAN_ENQUEUE_FLAG_DCA)) {
- struct qbman_eq_desc *d = (struct qbman_eq_desc *)p;
+ struct qbman_eq_desc *eq_desc = (struct qbman_eq_desc *)p;
- d->dca = (1 << QB_ENQUEUE_CMD_DCA_EN_SHIFT) |
+ eq_desc->dca = (1 << QB_ENQUEUE_CMD_DCA_EN_SHIFT) |
((flags[i]) & QBMAN_EQCR_DCA_IDXMASK);
}
eqcr_pi++;
int i, num_enqueued = 0;
unsigned long irq_flags;
- spin_lock(&s->access_spinlock);
- local_irq_save(irq_flags);
+ spin_lock_irqsave(&s->access_spinlock, irq_flags);
half_mask = (s->eqcr.pi_ci_mask>>1);
full_mask = s->eqcr.pi_ci_mask;
s->eqcr.available = qm_cyc_diff(s->eqcr.pi_ring_size,
eqcr_ci, s->eqcr.ci);
if (!s->eqcr.available) {
- local_irq_restore(irq_flags);
- spin_unlock(&s->access_spinlock);
+ spin_unlock_irqrestore(&s->access_spinlock, irq_flags);
return 0;
}
}
p = (s->addr_cena + QBMAN_CENA_SWP_EQCR(eqcr_pi & half_mask));
p[0] = cl[0] | s->eqcr.pi_vb;
if (flags && (flags[i] & QBMAN_ENQUEUE_FLAG_DCA)) {
- struct qbman_eq_desc *d = (struct qbman_eq_desc *)p;
+ struct qbman_eq_desc *eq_desc = (struct qbman_eq_desc *)p;
- d->dca = (1 << QB_ENQUEUE_CMD_DCA_EN_SHIFT) |
+ eq_desc->dca = (1 << QB_ENQUEUE_CMD_DCA_EN_SHIFT) |
((flags[i]) & QBMAN_EQCR_DCA_IDXMASK);
}
eqcr_pi++;
dma_wmb();
qbman_write_register(s, QBMAN_CINH_SWP_EQCR_PI,
(QB_RT_BIT)|(s->eqcr.pi)|s->eqcr.pi_vb);
- local_irq_restore(irq_flags);
- spin_unlock(&s->access_spinlock);
+ spin_unlock_irqrestore(&s->access_spinlock, irq_flags);
return num_enqueued;
}
{
return le32_to_cpu(a->fill);
}
+
+/**
+ * qbman_swp_set_irq_coalescing() - Set new IRQ coalescing values
+ * @p: the software portal object
+ * @irq_threshold: interrupt threshold
+ * @irq_holdoff: interrupt holdoff (timeout) period in us
+ *
+ * Return 0 for success, or negative error code on error.
+ */
+int qbman_swp_set_irq_coalescing(struct qbman_swp *p, u32 irq_threshold,
+ u32 irq_holdoff)
+{
+ u32 itp, max_holdoff;
+
+ /* Convert irq_holdoff value from usecs to 256 QBMAN clock cycles
+ * increments. This depends on the QBMAN internal frequency.
+ */
+ itp = (irq_holdoff * 1000) / p->desc->qman_256_cycles_per_ns;
+ if (itp > 4096) {
+ max_holdoff = (p->desc->qman_256_cycles_per_ns * 4096) / 1000;
+ pr_err("irq_holdoff must be <= %uus\n", max_holdoff);
+ return -EINVAL;
+ }
+
+ if (irq_threshold >= p->dqrr.dqrr_size) {
+ pr_err("irq_threshold must be < %u\n", p->dqrr.dqrr_size - 1);
+ return -EINVAL;
+ }
+
+ p->irq_threshold = irq_threshold;
+ p->irq_holdoff = irq_holdoff;
+
+ qbman_write_register(p, QBMAN_CINH_SWP_DQRR_ITR, irq_threshold);
+ qbman_write_register(p, QBMAN_CINH_SWP_ITPR, itp);
+
+ return 0;
+}
+
+/**
+ * qbman_swp_get_irq_coalescing() - Get the current IRQ coalescing parameters
+ * @p: the software portal object
+ * @irq_threshold: interrupt threshold (an IRQ is generated when there are more
+ * DQRR entries in the portal than the threshold)
+ * @irq_holdoff: interrupt holdoff (timeout) period in us
+ */
+void qbman_swp_get_irq_coalescing(struct qbman_swp *p, u32 *irq_threshold,
+ u32 *irq_holdoff)
+{
+ if (irq_threshold)
+ *irq_threshold = p->irq_threshold;
+ if (irq_holdoff)
+ *irq_holdoff = p->irq_holdoff;
+}