Hi Tom,
> From: Tom Rix <t...@redhat.com>
> On 9/30/20 12:45 PM, Chautru, Nicolas wrote:
> > Hi Tom,
> >
> >> From: Tom Rix <t...@redhat.com>
> >> On 9/28/20 5:29 PM, Nicolas Chautru wrote:
> >>> Adding capability and functions to support MSI interrupts, call
> >>> backs and inforing.
> >>>
> >>> Signed-off-by: Nicolas Chautru <nicolas.chau...@intel.com>
> >>> Acked-by: Liu Tianjiao <tianjiao....@intel.com>
> >>> ---
> >>> drivers/baseband/acc100/rte_acc100_pmd.c | 288
> >>> ++++++++++++++++++++++++++++++-
> >>> drivers/baseband/acc100/rte_acc100_pmd.h | 15 ++
> >>> 2 files changed, 300 insertions(+), 3 deletions(-)
> >>>
> >>> diff --git a/drivers/baseband/acc100/rte_acc100_pmd.c
> >>> b/drivers/baseband/acc100/rte_acc100_pmd.c
> >>> index 7d4c3df..b6d9e7c 100644
> >>> --- a/drivers/baseband/acc100/rte_acc100_pmd.c
> >>> +++ b/drivers/baseband/acc100/rte_acc100_pmd.c
> >>> @@ -339,6 +339,213 @@
> >>> free_base_addresses(base_addrs, i); }
> >>>
> >>> +/*
> >>> + * Find queue_id of a device queue based on details from the Info Ring.
> >>> + * If a queue isn't found UINT16_MAX is returned.
> >>> + */
> >>> +static inline uint16_t
> >>> +get_queue_id_from_ring_info(struct rte_bbdev_data *data,
> >>> + const union acc100_info_ring_data ring_data) {
> >>> + uint16_t queue_id;
> >>> +
> >>> + for (queue_id = 0; queue_id < data->num_queues; ++queue_id) {
> >>> + struct acc100_queue *acc100_q =
> >>> + data->queues[queue_id].queue_private;
> >>> + if (acc100_q != NULL && acc100_q->aq_id == ring_data.aq_id
> >> &&
> >>> + acc100_q->qgrp_id == ring_data.qg_id &&
> >>> + acc100_q->vf_id == ring_data.vf_id)
> >>> + return queue_id;
> >> If num_queues is large, this linear search will be slow.
> >>
> >> Consider changing the search algorithm.
> > This is not in the time critical part of the code
> ok
> >
> >
> >>> + }
> >>> +
> >>> + return UINT16_MAX;
> >> the interrupt handlers that use this function do not a great job of
> >> handling this error.
> > if that error actualy happened then there is not much else that can be
> done except reporting the unexpected data.
> ok
> >
> >>> +}
> >>> +
> >>> +/* Checks PF Info Ring to find the interrupt cause and handles it
> >>> +accordingly */ static inline void acc100_check_ir(struct
> >>> +acc100_device *acc100_dev) {
> >>> + volatile union acc100_info_ring_data *ring_data;
> >>> + uint16_t info_ring_head = acc100_dev->info_ring_head;
> >>> + if (acc100_dev->info_ring == NULL)
> >>> + return;
> >>> +
> >>> + ring_data = acc100_dev->info_ring + (acc100_dev->info_ring_head &
> >>> + ACC100_INFO_RING_MASK);
> >>> +
> >>> + while (ring_data->valid) {
> >>> + if ((ring_data->int_nb <
> >> ACC100_PF_INT_DMA_DL_DESC_IRQ) || (
> >>> + ring_data->int_nb >
> >>> + ACC100_PF_INT_DMA_DL5G_DESC_IRQ))
> >>> + rte_bbdev_log(WARNING, "InfoRing: ITR:%d
> >> Info:0x%x",
> >>> + ring_data->int_nb, ring_data-
> >>> detailed_info);
> >>> + /* Initialize Info Ring entry and move forward */
> >>> + ring_data->val = 0;
> >>> + info_ring_head++;
> >>> + ring_data = acc100_dev->info_ring +
> >>> + (info_ring_head &
> >> ACC100_INFO_RING_MASK);
> >> These three statements are common for the ring handling, consider a
> >> macro or inline function.
> > ok
> >
> >>> + }
> >>> +}
> >>> +
> >>> +/* Checks PF Info Ring to find the interrupt cause and handles it
> >>> +accordingly */ static inline void
> >>> +acc100_pf_interrupt_handler(struct
> >>> +rte_bbdev *dev) {
> >>> + struct acc100_device *acc100_dev = dev->data->dev_private;
> >>> + volatile union acc100_info_ring_data *ring_data;
> >>> + struct acc100_deq_intr_details deq_intr_det;
> >>> +
> >>> + ring_data = acc100_dev->info_ring + (acc100_dev->info_ring_head &
> >>> + ACC100_INFO_RING_MASK);
> >>> +
> >>> + while (ring_data->valid) {
> >>> +
> >>> + rte_bbdev_log_debug(
> >>> + "ACC100 PF Interrupt received, Info Ring
> >> data: 0x%x",
> >>> + ring_data->val);
> >>> +
> >>> + switch (ring_data->int_nb) {
> >>> + case ACC100_PF_INT_DMA_DL_DESC_IRQ:
> >>> + case ACC100_PF_INT_DMA_UL_DESC_IRQ:
> >>> + case ACC100_PF_INT_DMA_UL5G_DESC_IRQ:
> >>> + case ACC100_PF_INT_DMA_DL5G_DESC_IRQ:
> >>> + deq_intr_det.queue_id =
> >> get_queue_id_from_ring_info(
> >>> + dev->data, *ring_data);
> >>> + if (deq_intr_det.queue_id == UINT16_MAX) {
> >>> + rte_bbdev_log(ERR,
> >>> + "Couldn't find queue: aq_id:
> >> %u, qg_id: %u, vf_id: %u",
> >>> + ring_data->aq_id,
> >>> + ring_data->qg_id,
> >>> + ring_data->vf_id);
> >>> + return;
> >>> + }
> >>> + rte_bbdev_pmd_callback_process(dev,
> >>> + RTE_BBDEV_EVENT_DEQUEUE,
> >> &deq_intr_det);
> >>> + break;
> >>> + default:
> >>> + rte_bbdev_pmd_callback_process(dev,
> >>> + RTE_BBDEV_EVENT_ERROR, NULL);
> >>> + break;
> >>> + }
> >>> +
> >>> + /* Initialize Info Ring entry and move forward */
> >>> + ring_data->val = 0;
> >>> + ++acc100_dev->info_ring_head;
> >>> + ring_data = acc100_dev->info_ring +
> >>> + (acc100_dev->info_ring_head &
> >>> + ACC100_INFO_RING_MASK);
> >>> + }
> >>> +}
> >>> +
> >>> +/* Checks VF Info Ring to find the interrupt cause and handles it
> >>> +accordingly */ static inline void
> >>> +acc100_vf_interrupt_handler(struct
> >>> +rte_bbdev *dev)
> >> very similar to pf case, consider combining.
> >>> +{
> >>> + struct acc100_device *acc100_dev = dev->data->dev_private;
> >>> + volatile union acc100_info_ring_data *ring_data;
> >>> + struct acc100_deq_intr_details deq_intr_det;
> >>> +
> >>> + ring_data = acc100_dev->info_ring + (acc100_dev->info_ring_head &
> >>> + ACC100_INFO_RING_MASK);
> >>> +
> >>> + while (ring_data->valid) {
> >>> +
> >>> + rte_bbdev_log_debug(
> >>> + "ACC100 VF Interrupt received, Info Ring
> >> data: 0x%x",
> >>> + ring_data->val);
> >>> +
> >>> + switch (ring_data->int_nb) {
> >>> + case ACC100_VF_INT_DMA_DL_DESC_IRQ:
> >>> + case ACC100_VF_INT_DMA_UL_DESC_IRQ:
> >>> + case ACC100_VF_INT_DMA_UL5G_DESC_IRQ:
> >>> + case ACC100_VF_INT_DMA_DL5G_DESC_IRQ:
> >>> + /* VFs are not aware of their vf_id - it's set to 0 in
> >>> + * queue structures.
> >>> + */
> >>> + ring_data->vf_id = 0;
> >>> + deq_intr_det.queue_id =
> >> get_queue_id_from_ring_info(
> >>> + dev->data, *ring_data);
> >>> + if (deq_intr_det.queue_id == UINT16_MAX) {
> >>> + rte_bbdev_log(ERR,
> >>> + "Couldn't find queue: aq_id:
> >> %u, qg_id: %u",
> >>> + ring_data->aq_id,
> >>> + ring_data->qg_id);
> >>> + return;
> >>> + }
> >>> + rte_bbdev_pmd_callback_process(dev,
> >>> + RTE_BBDEV_EVENT_DEQUEUE,
> >> &deq_intr_det);
> >>> + break;
> >>> + default:
> >>> + rte_bbdev_pmd_callback_process(dev,
> >>> + RTE_BBDEV_EVENT_ERROR, NULL);
> >>> + break;
> >>> + }
> >>> +
> >>> + /* Initialize Info Ring entry and move forward */
> >>> + ring_data->valid = 0;
> >>> + ++acc100_dev->info_ring_head;
> >>> + ring_data = acc100_dev->info_ring + (acc100_dev-
> >>> info_ring_head
> >>> + & ACC100_INFO_RING_MASK);
> >>> + }
> >>> +}
> >>> +
> >>> +/* Interrupt handler triggered by ACC100 dev for handling specific
> >>> +interrupt */ static void acc100_dev_interrupt_handler(void *cb_arg) {
> >>> + struct rte_bbdev *dev = cb_arg;
> >>> + struct acc100_device *acc100_dev = dev->data->dev_private;
> >>> +
> >>> + /* Read info ring */
> >>> + if (acc100_dev->pf_device)
> >>> + acc100_pf_interrupt_handler(dev);
> >> combined like ..
> >>
> >> acc100_interrupt_handler(dev, is_pf)
> > unsure it will help readability. Much of the code would still be
> > distinct
> ok
> >
> >>> + else
> >>> + acc100_vf_interrupt_handler(dev); }
> >>> +
> >>> +/* Allocate and setup inforing */
> >>> +static int
> >>> +allocate_inforing(struct rte_bbdev *dev)
> >> consider renaming
> >>
> >> allocate_info_ring
> > ok
> >
> >>> +{
> >>> + struct acc100_device *d = dev->data->dev_private;
> >>> + const struct acc100_registry_addr *reg_addr;
> >>> + rte_iova_t info_ring_phys;
> >>> + uint32_t phys_low, phys_high;
> >>> +
> >>> + if (d->info_ring != NULL)
> >>> + return 0; /* Already configured */
> >>> +
> >>> + /* Choose correct registry addresses for the device type */
> >>> + if (d->pf_device)
> >>> + reg_addr = &pf_reg_addr;
> >>> + else
> >>> + reg_addr = &vf_reg_addr;
> >>> + /* Allocate InfoRing */
> >>> + d->info_ring = rte_zmalloc_socket("Info Ring",
> >>> + ACC100_INFO_RING_NUM_ENTRIES *
> >>> + sizeof(*d->info_ring), RTE_CACHE_LINE_SIZE,
> >>> + dev->data->socket_id);
> >>> + if (d->info_ring == NULL) {
> >>> + rte_bbdev_log(ERR,
> >>> + "Failed to allocate Info Ring for %s:%u",
> >>> + dev->device->driver->name,
> >>> + dev->data->dev_id);
> >> The callers do not check that this fails.
> > arguably the error would be self contained if that did fail. But doesn't
> > hurt
> to add, ok.
> >
> >>> + return -ENOMEM;
> >>> + }
> >>> + info_ring_phys = rte_malloc_virt2iova(d->info_ring);
> >>> +
> >>> + /* Setup Info Ring */
> >>> + phys_high = (uint32_t)(info_ring_phys >> 32);
> >>> + phys_low = (uint32_t)(info_ring_phys);
> >>> + acc100_reg_write(d, reg_addr->info_ring_hi, phys_high);
> >>> + acc100_reg_write(d, reg_addr->info_ring_lo, phys_low);
> >>> + acc100_reg_write(d, reg_addr->info_ring_en,
> >> ACC100_REG_IRQ_EN_ALL);
> >>> + d->info_ring_head = (acc100_reg_read(d, reg_addr->info_ring_ptr) &
> >>> + 0xFFF) / sizeof(union acc100_info_ring_data);
> >>> + return 0;
> >>> +}
> >>> +
> >>> +
> >>> /* Allocate 64MB memory used for all software rings */ static int
> >>> acc100_setup_queues(struct rte_bbdev *dev, uint16_t num_queues, int
> >>> socket_id) @@ -426,6 +633,7 @@
> >>> acc100_reg_write(d, reg_addr->tail_ptrs_dl4g_hi, phys_high);
> >>> acc100_reg_write(d, reg_addr->tail_ptrs_dl4g_lo, phys_low);
> >>>
> >>> + allocate_inforing(dev);
> >> need to check here
> >>> d->harq_layout = rte_zmalloc_socket("HARQ Layout",
> >>> ACC100_HARQ_LAYOUT * sizeof(*d->harq_layout),
> >>> RTE_CACHE_LINE_SIZE, dev->data->socket_id); @@ -
> >> 437,13 +645,53 @@
> >>> return 0;
> >>> }
> >>>
> >>> +static int
> >>> +acc100_intr_enable(struct rte_bbdev *dev) {
> >>> + int ret;
> >>> + struct acc100_device *d = dev->data->dev_private;
> >>> +
> >>> + /* Only MSI are currently supported */
> >>> + if (dev->intr_handle->type == RTE_INTR_HANDLE_VFIO_MSI ||
> >>> + dev->intr_handle->type == RTE_INTR_HANDLE_UIO)
> >> {
> >>> +
> >>> + allocate_inforing(dev);
> >> need to check here
> >>> +
> >>> + ret = rte_intr_enable(dev->intr_handle);
> >>> + if (ret < 0) {
> >>> + rte_bbdev_log(ERR,
> >>> + "Couldn't enable interrupts for
> >> device: %s",
> >>> + dev->data->name);
> >>> + rte_free(d->info_ring);
> >>> + return ret;
> >>> + }
> >>> + ret = rte_intr_callback_register(dev->intr_handle,
> >>> + acc100_dev_interrupt_handler, dev);
> >>> + if (ret < 0) {
> >>> + rte_bbdev_log(ERR,
> >>> + "Couldn't register interrupt callback
> >> for device: %s",
> >>> + dev->data->name);
> >>> + rte_free(d->info_ring);
> >> does intr need to be disabled here ?
> > Well I don't see a lot of consistency with other drivers. Sometimes these
> are not even check for failure.
> > I would rather defer changing through other future patch if required as this
> is same code on other bbdev drivers already used (if changed I would rather
> all changed the same way).
>
> ok.
>
>
> >
> >>> + return ret;
> >>> + }
> >>> +
> >>> + return 0;
> >>> + }
> >>> +
> >>> + rte_bbdev_log(ERR, "ACC100 (%s) supports only VFIO MSI
> >> interrupts",
> >>> + dev->data->name);
> >>> + return -ENOTSUP;
> >>> +}
> >>> +
> >>> /* Free 64MB memory used for software rings */ static int
> >>> acc100_dev_close(struct rte_bbdev *dev) {
> >>> struct acc100_device *d = dev->data->dev_private;
> >>> + acc100_check_ir(d);
> >>> if (d->sw_rings_base != NULL) {
> >>> rte_free(d->tail_ptrs);
> >>> + rte_free(d->info_ring);
> >>> rte_free(d->sw_rings_base);
> >>> d->sw_rings_base = NULL;
> >>> }
> >>> @@ -643,6 +891,7 @@
> >>> RTE_BBDEV_TURBO_CRC_TYPE_24B
> >> |
> >> RTE_BBDEV_TURBO_HALF_ITERATION_EVEN |
> >> RTE_BBDEV_TURBO_EARLY_TERMINATION |
> >>> +
> >> RTE_BBDEV_TURBO_DEC_INTERRUPTS |
> >> RTE_BBDEV_TURBO_NEG_LLR_1_BIT_IN |
> >>> RTE_BBDEV_TURBO_MAP_DEC |
> >>>
> >> RTE_BBDEV_TURBO_DEC_TB_CRC_24B_KEEP | @@ -663,6 +912,7
> @@
> >> RTE_BBDEV_TURBO_CRC_24B_ATTACH |
> >> RTE_BBDEV_TURBO_RV_INDEX_BYPASS |
> >>> RTE_BBDEV_TURBO_RATE_MATCH |
> >>> +
> >> RTE_BBDEV_TURBO_ENC_INTERRUPTS |
> >> RTE_BBDEV_TURBO_ENC_SCATTER_GATHER,
> >>> .num_buffers_src =
> >>>
> >> RTE_BBDEV_TURBO_MAX_CODE_BLOCKS, @@ -676,7 +926,8 @@
> >>> .capability_flags =
> >>> RTE_BBDEV_LDPC_RATE_MATCH |
> >>>
> >> RTE_BBDEV_LDPC_CRC_24B_ATTACH |
> >>> -
> >> RTE_BBDEV_LDPC_INTERLEAVER_BYPASS,
> >>> +
> >> RTE_BBDEV_LDPC_INTERLEAVER_BYPASS |
> >>> +
> >> RTE_BBDEV_LDPC_ENC_INTERRUPTS,
> >>> .num_buffers_src =
> >>>
> >> RTE_BBDEV_LDPC_MAX_CODE_BLOCKS,
> >>> .num_buffers_dst =
> >>> @@ -701,7 +952,8 @@
> >>> RTE_BBDEV_LDPC_DECODE_BYPASS |
> >>> RTE_BBDEV_LDPC_DEC_SCATTER_GATHER |
> >>>
> >> RTE_BBDEV_LDPC_HARQ_6BIT_COMPRESSION |
> >>> - RTE_BBDEV_LDPC_LLR_COMPRESSION,
> >>> + RTE_BBDEV_LDPC_LLR_COMPRESSION |
> >>> + RTE_BBDEV_LDPC_DEC_INTERRUPTS,
> >>> .llr_size = 8,
> >>> .llr_decimals = 1,
> >>> .num_buffers_src =
> >>> @@ -751,14 +1003,39 @@
> >>> #else
> >>> dev_info->harq_buffer_size = 0;
> >>> #endif
> >>> + acc100_check_ir(d);
> >>> +}
> >>> +
> >>> +static int
> >>> +acc100_queue_intr_enable(struct rte_bbdev *dev, uint16_t queue_id)
> {
> >>> + struct acc100_queue *q = dev->data-
> >>> queues[queue_id].queue_private;
> >>> +
> >>> + if (dev->intr_handle->type != RTE_INTR_HANDLE_VFIO_MSI &&
> >>> + dev->intr_handle->type != RTE_INTR_HANDLE_UIO)
> >>> + return -ENOTSUP;
> >>> +
> >>> + q->irq_enable = 1;
> >>> + return 0;
> >>> +}
> >>> +
> >>> +static int
> >>> +acc100_queue_intr_disable(struct rte_bbdev *dev, uint16_t queue_id)
> {
> >>> + struct acc100_queue *q = dev->data-
> >>> queues[queue_id].queue_private;
> >>> + q->irq_enable = 0;
> >> A -ENOTSUP above, should need similar check here.
> > How can this fail when we purely disable?
>
> It is for api consistency.
>
> the enable fails
>
> the disable succeeds
>
> that is not consistent.
>
OK can do. Thanks
> Tom
>
> >
> >>> + return 0;
> >>> }
> >>>
> >>> static const struct rte_bbdev_ops acc100_bbdev_ops = {
> >>> .setup_queues = acc100_setup_queues,
> >>> + .intr_enable = acc100_intr_enable,
> >>> .close = acc100_dev_close,
> >>> .info_get = acc100_dev_info_get,
> >>> .queue_setup = acc100_queue_setup,
> >>> .queue_release = acc100_queue_release,
> >>> + .queue_intr_enable = acc100_queue_intr_enable,
> >>> + .queue_intr_disable = acc100_queue_intr_disable
> >>> };
> >>>
> >>> /* ACC100 PCI PF address map */
> >>> @@ -3018,8 +3295,10 @@
> >>> ? (1 << RTE_BBDEV_DATA_ERROR) : 0);
> >>> op->status |= ((rsp.dma_err) ? (1 << RTE_BBDEV_DRV_ERROR) : 0);
> >>> op->status |= ((rsp.fcw_err) ? (1 << RTE_BBDEV_DRV_ERROR) : 0);
> >>> - if (op->status != 0)
> >>> + if (op->status != 0) {
> >>> q_data->queue_stats.dequeue_err_count++;
> >>> + acc100_check_ir(q->d);
> >>> + }
> >>>
> >>> /* CRC invalid if error exists */
> >>> if (!op->status)
> >>> @@ -3076,6 +3355,9 @@
> >>> op->status |= 1 << RTE_BBDEV_SYNDROME_ERROR;
> >>> op->ldpc_dec.iter_count = (uint8_t) rsp.iter_cnt;
> >>>
> >>> + if (op->status & (1 << RTE_BBDEV_DRV_ERROR))
> >>> + acc100_check_ir(q->d);
> >>> +
> >>> /* Check if this is the last desc in batch (Atomic Queue) */
> >>> if (desc->req.last_desc_in_batch) {
> >>> (*aq_dequeued)++;
> >>> diff --git a/drivers/baseband/acc100/rte_acc100_pmd.h
> >>> b/drivers/baseband/acc100/rte_acc100_pmd.h
> >>> index 78686c1..8980fa5 100644
> >>> --- a/drivers/baseband/acc100/rte_acc100_pmd.h
> >>> +++ b/drivers/baseband/acc100/rte_acc100_pmd.h
> >>> @@ -559,7 +559,14 @@ struct acc100_device {
> >>> /* Virtual address of the info memory routed to the this function
> >> under
> >>> * operation, whether it is PF or VF.
> >>> */
> >>> + union acc100_info_ring_data *info_ring;
> >> Need a comment that this array needs a sentinel ?
> > Can clarify a bit expected HW behaviour
> >
> > Thanks
> >
> >> Tom
> >>
> >>> +
> >>> union acc100_harq_layout_data *harq_layout;
> >>> + /* Virtual Info Ring head */
> >>> + uint16_t info_ring_head;
> >>> + /* Number of bytes available for each queue in device, depending
> >> on
> >>> + * how many queues are enabled with configure()
> >>> + */
> >>> uint32_t sw_ring_size;
> >>> uint32_t ddr_size; /* Size in kB */
> >>> uint32_t *tail_ptrs; /* Base address of response tail pointer
> >>> buffer */ @@ -575,4 +582,12 @@ struct acc100_device {
> >>> bool configured; /**< True if this ACC100 device is configured */
> >>> };
> >>>
> >>> +/**
> >>> + * Structure with details about RTE_BBDEV_EVENT_DEQUEUE event.
> It's
> >>> +passed to
> >>> + * the callback function.
> >>> + */
> >>> +struct acc100_deq_intr_details {
> >>> + uint16_t queue_id;
> >>> +};
> >>> +
> >>> #endif /* _RTE_ACC100_PMD_H_ */
