@@ -38,7 +38,8 @@ skeldma_info_get(const struct rte_dma_dev *dev, struct rte_dma_info *dev_info,
dev_info->dev_capa = RTE_DMA_CAPA_MEM_TO_MEM |
RTE_DMA_CAPA_SVA |
RTE_DMA_CAPA_OPS_COPY |
- RTE_DMA_CAPA_OPS_COPY_SG;
+ RTE_DMA_CAPA_OPS_COPY_SG |
+ RTE_DMA_CAPA_OPS_FILL;
dev_info->max_vchans = 1;
dev_info->max_desc = SKELDMA_MAX_DESC;
dev_info->min_desc = SKELDMA_MIN_DESC;
@@ -100,8 +101,19 @@ do_copy_sg(struct skeldma_desc *desc)
}
}
+static inline void
+do_fill(struct skeldma_desc *desc)
+{
+ uint8_t *fills = (uint8_t *)&desc->fill.pattern;
+ uint8_t *dst = (uint8_t *)desc->fill.dst;
+ uint32_t i;
+
+ for (i = 0; i < desc->fill.len; i++)
+ dst[i] = fills[i % 8];
+}
+
static uint32_t
-cpucopy_thread(void *param)
+cpuwork_thread(void *param)
{
#define SLEEP_THRESHOLD 10000
#define SLEEP_US_VAL 10
@@ -127,6 +139,8 @@ cpucopy_thread(void *param)
rte_memcpy(desc->copy.dst, desc->copy.src, desc->copy.len);
else if (desc->op == SKELDMA_OP_COPY_SG)
do_copy_sg(desc);
+ else if (desc->op == SKELDMA_OP_FILL)
+ do_fill(desc);
__atomic_fetch_add(&hw->completed_count, 1, __ATOMIC_RELEASE);
(void)rte_ring_enqueue(hw->desc_completed, (void *)desc);
@@ -162,7 +176,7 @@ skeldma_start(struct rte_dma_dev *dev)
* 1) fflush pending/running/completed ring to empty ring.
* 2) init ring idx to zero.
* 3) init running statistics.
- * 4) mark cpucopy task exit_flag to false.
+ * 4) mark cpuwork task exit_flag to false.
*/
fflush_ring(hw, hw->desc_pending);
fflush_ring(hw, hw->desc_running);
@@ -178,9 +192,9 @@ skeldma_start(struct rte_dma_dev *dev)
snprintf(name, sizeof(name), "dma-skel%d", dev->data->dev_id);
ret = rte_thread_create_internal_control(&hw->thread, name,
- cpucopy_thread, dev);
+ cpuwork_thread, dev);
if (ret) {
- SKELDMA_LOG(ERR, "Start cpucopy thread fail!");
+ SKELDMA_LOG(ERR, "Start cpuwork thread fail!");
return -EINVAL;
}
@@ -462,6 +476,34 @@ skeldma_copy_sg(void *dev_private, uint16_t vchan,
return hw->ridx++;
}
+static int
+skeldma_fill(void *dev_private, uint16_t vchan,
+ uint64_t pattern, rte_iova_t dst,
+ uint32_t length, uint64_t flags)
+{
+ struct skeldma_hw *hw = dev_private;
+ struct skeldma_desc *desc;
+ int ret;
+
+ RTE_SET_USED(vchan);
+
+ ret = rte_ring_dequeue(hw->desc_empty, (void **)&desc);
+ if (ret)
+ return -ENOSPC;
+ desc->op = SKELDMA_OP_FILL;
+ desc->ridx = hw->ridx;
+ desc->fill.dst = (void *)(uintptr_t)dst;
+ desc->fill.len = length;
+ desc->fill.pattern = pattern;
+ if (flags & RTE_DMA_OP_FLAG_SUBMIT)
+ submit(hw, desc);
+ else
+ (void)rte_ring_enqueue(hw->desc_pending, (void *)desc);
+ hw->submitted_count++;
+
+ return hw->ridx++;
+}
+
static int
skeldma_submit(void *dev_private, uint16_t vchan)
{
@@ -573,6 +615,7 @@ skeldma_create(const char *name, struct rte_vdev_device *vdev, int lcore_id)
dev->fp_obj->dev_private = dev->data->dev_private;
dev->fp_obj->copy = skeldma_copy;
dev->fp_obj->copy_sg = skeldma_copy_sg;
+ dev->fp_obj->fill = skeldma_fill;
dev->fp_obj->submit = skeldma_submit;
dev->fp_obj->completed = skeldma_completed;
dev->fp_obj->completed_status = skeldma_completed_status;
@@ -16,6 +16,7 @@
enum skeldma_op {
SKELDMA_OP_COPY,
SKELDMA_OP_COPY_SG,
+ SKELDMA_OP_FILL,
};
struct skeldma_desc {
@@ -34,14 +35,19 @@ struct skeldma_desc {
uint16_t nb_src;
uint16_t nb_dst;
} copy_sg;
+ struct {
+ void *dst;
+ uint32_t len;
+ uint64_t pattern;
+ } fill;
};
};
struct skeldma_hw {
- int lcore_id; /* cpucopy task affinity core */
+ int lcore_id; /* cpuwork task affinity core */
int socket_id;
- rte_thread_t thread; /* cpucopy task thread */
- volatile int exit_flag; /* cpucopy task exit flag */
+ rte_thread_t thread; /* cpuwork task thread */
+ volatile int exit_flag; /* cpuwork task exit flag */
struct skeldma_desc *desc_mem;
@@ -57,7 +63,7 @@ struct skeldma_hw {
* |get completed |------------------| |
* | | |
* | v v
- * ----------- cpucopy thread working -----------
+ * ----------- cpuwork thread working -----------
* |completed|<-------------------------------| running |
* ----------- -----------
*/
@@ -72,7 +78,7 @@ struct skeldma_hw {
uint16_t last_ridx;
uint64_t submitted_count;
- /* Cache delimiter for cpucopy thread's operation data */
+ /* Cache delimiter for cpuwork thread's operation data */
char cache2 __rte_cache_aligned;
volatile uint32_t zero_req_count;
uint64_t completed_count;