numam-dpdk/drivers/raw/octeontx2_dma/otx2_dpi_rawdev.c

442 lines
12 KiB
C
Raw Normal View History

/* SPDX-License-Identifier: BSD-3-Clause
* Copyright(C) 2019 Marvell International Ltd.
*/
#include <string.h>
#include <unistd.h>
#include <rte_bus.h>
#include <rte_bus_pci.h>
#include <rte_common.h>
#include <rte_eal.h>
#include <rte_lcore.h>
#include <rte_mempool.h>
#include <rte_pci.h>
#include <rte_rawdev.h>
#include <rte_rawdev_pmd.h>
#include <otx2_common.h>
#include "otx2_dpi_rawdev.h"
static const struct rte_pci_id pci_dma_map[] = {
{
RTE_PCI_DEVICE(PCI_VENDOR_ID_CAVIUM,
PCI_DEVID_OCTEONTX2_DPI_VF)
},
{
.vendor_id = 0,
},
};
/* Enable/Disable DMA queue */
static inline int
dma_engine_enb_dis(struct dpi_vf_s *dpivf, const bool enb)
{
if (enb)
otx2_write64(0x1, dpivf->vf_bar0 + DPI_VDMA_EN);
else
otx2_write64(0x0, dpivf->vf_bar0 + DPI_VDMA_EN);
return DPI_DMA_QUEUE_SUCCESS;
}
/* Free DMA Queue instruction buffers, and send close notification to PF */
static inline int
dma_queue_finish(struct dpi_vf_s *dpivf)
{
uint32_t timeout = 0, sleep = 1;
uint64_t reg = 0ULL;
/* Wait for SADDR to become idle */
reg = otx2_read64(dpivf->vf_bar0 + DPI_VDMA_SADDR);
while (!(reg & BIT_ULL(DPI_VDMA_SADDR_REQ_IDLE))) {
rte_delay_ms(sleep);
timeout++;
if (timeout >= DPI_QFINISH_TIMEOUT) {
otx2_dpi_dbg("Timeout!!! Closing Forcibly");
break;
}
reg = otx2_read64(dpivf->vf_bar0 + DPI_VDMA_SADDR);
}
if (otx2_dpi_queue_close(dpivf) < 0)
return -EACCES;
rte_mempool_put(dpivf->chunk_pool, dpivf->base_ptr);
dpivf->vf_bar0 = (uintptr_t)NULL;
return DPI_DMA_QUEUE_SUCCESS;
}
/* Write an arbitrary number of command words to a command queue */
static __rte_always_inline enum dpi_dma_queue_result_e
dma_queue_write(struct dpi_vf_s *dpi, uint16_t cmd_count, uint64_t *cmds)
{
if ((cmd_count < 1) || (cmd_count > 64))
return DPI_DMA_QUEUE_INVALID_PARAM;
if (cmds == NULL)
return DPI_DMA_QUEUE_INVALID_PARAM;
/* Room available in the current buffer for the command */
if (dpi->index + cmd_count < dpi->pool_size_m1) {
uint64_t *ptr = dpi->base_ptr;
ptr += dpi->index;
dpi->index += cmd_count;
while (cmd_count--)
*ptr++ = *cmds++;
} else {
void *new_buffer;
uint64_t *ptr;
int count;
/* Allocate new command buffer, return if failed */
if (rte_mempool_get(dpi->chunk_pool, &new_buffer) ||
new_buffer == NULL) {
return DPI_DMA_QUEUE_NO_MEMORY;
}
ptr = dpi->base_ptr;
/* Figure out how many command words will fit in this buffer.
* One location will be needed for the next buffer pointer.
**/
count = dpi->pool_size_m1 - dpi->index;
ptr += dpi->index;
cmd_count -= count;
while (count--)
*ptr++ = *cmds++;
/* Chunk next ptr is 2DWORDs, second DWORD is reserved. */
*ptr++ = (uint64_t)new_buffer;
*ptr = 0;
/* The current buffer is full and has a link to the next buffer.
* Time to write the rest of the commands into the new buffer.
**/
dpi->base_ptr = new_buffer;
dpi->index = cmd_count;
ptr = new_buffer;
while (cmd_count--)
*ptr++ = *cmds++;
/* queue index may greater than pool size */
if (dpi->index >= dpi->pool_size_m1) {
if (rte_mempool_get(dpi->chunk_pool, &new_buffer) ||
new_buffer == NULL) {
return DPI_DMA_QUEUE_NO_MEMORY;
}
/* Write next buffer address */
*ptr = (uint64_t)new_buffer;
dpi->base_ptr = new_buffer;
dpi->index = 0;
}
}
return DPI_DMA_QUEUE_SUCCESS;
}
/* Submit a DMA command to the DMA queues. */
static __rte_always_inline int
dma_queue_submit(struct rte_rawdev *dev, uint16_t cmd_count, uint64_t *cmds)
{
struct dpi_vf_s *dpivf = dev->dev_private;
enum dpi_dma_queue_result_e result;
result = dma_queue_write(dpivf, cmd_count, cmds);
rte_wmb();
if (likely(result == DPI_DMA_QUEUE_SUCCESS))
otx2_write64((uint64_t)cmd_count,
dpivf->vf_bar0 + DPI_VDMA_DBELL);
return result;
}
/* Enqueue buffers to DMA queue
* returns number of buffers enqueued successfully
*/
static int
otx2_dpi_rawdev_enqueue_bufs(struct rte_rawdev *dev,
struct rte_rawdev_buf **buffers,
unsigned int count, rte_rawdev_obj_t context)
{
struct dpi_dma_queue_ctx_s *ctx = (struct dpi_dma_queue_ctx_s *)context;
struct dpi_dma_buf_ptr_s *cmd;
uint32_t c = 0;
for (c = 0; c < count; c++) {
uint64_t dpi_cmd[DPI_DMA_CMD_SIZE] = {0};
union dpi_dma_instr_hdr_u *hdr;
uint16_t index = 0, i;
hdr = (union dpi_dma_instr_hdr_u *)&dpi_cmd[0];
cmd = (struct dpi_dma_buf_ptr_s *)buffers[c]->buf_addr;
hdr->s.xtype = ctx->xtype & DPI_XTYPE_MASK;
hdr->s.pt = ctx->pt & DPI_HDR_PT_MASK;
/* Request initiated with byte write completion, but completion
* pointer not provided
*/
if ((hdr->s.pt == DPI_HDR_PT_ZBW_CA ||
hdr->s.pt == DPI_HDR_PT_ZBW_NC) && cmd->comp_ptr == NULL)
return c;
cmd->comp_ptr->cdata = DPI_REQ_CDATA;
hdr->s.ptr = (uint64_t)cmd->comp_ptr;
hdr->s.deallocv = ctx->deallocv;
hdr->s.tt = ctx->tt & DPI_W0_TT_MASK;
hdr->s.grp = ctx->grp & DPI_W0_GRP_MASK;
/* If caller provides completion ring details, then only queue
* completion address for later polling.
*/
if (ctx->c_ring) {
ctx->c_ring->compl_data[ctx->c_ring->tail] =
cmd->comp_ptr;
STRM_INC(ctx->c_ring);
}
if (hdr->s.deallocv)
hdr->s.pvfe = 1;
if (hdr->s.pt == DPI_HDR_PT_WQP)
hdr->s.ptr = hdr->s.ptr | DPI_HDR_PT_WQP_STATUSNC;
index += 4;
hdr->s.fport = 0;
hdr->s.lport = 0;
if (ctx->xtype != DPI_XTYPE_INTERNAL_ONLY)
hdr->s.lport = ctx->pem_id;
/* For inbound case, src pointers are last pointers.
* For all other cases, src pointers are first pointers.
*/
if (ctx->xtype == DPI_XTYPE_INBOUND) {
hdr->s.nfst = cmd->wptr_cnt & DPI_MAX_POINTER;
hdr->s.nlst = cmd->rptr_cnt & DPI_MAX_POINTER;
for (i = 0; i < hdr->s.nfst; i++) {
dpi_cmd[index++] = cmd->wptr[i]->u[0];
dpi_cmd[index++] = cmd->wptr[i]->u[1];
}
for (i = 0; i < hdr->s.nlst; i++) {
dpi_cmd[index++] = cmd->rptr[i]->u[0];
dpi_cmd[index++] = cmd->rptr[i]->u[1];
}
} else {
hdr->s.nfst = cmd->rptr_cnt & DPI_MAX_POINTER;
hdr->s.nlst = cmd->wptr_cnt & DPI_MAX_POINTER;
for (i = 0; i < hdr->s.nfst; i++) {
dpi_cmd[index++] = cmd->rptr[i]->u[0];
dpi_cmd[index++] = cmd->rptr[i]->u[1];
}
for (i = 0; i < hdr->s.nlst; i++) {
dpi_cmd[index++] = cmd->wptr[i]->u[0];
dpi_cmd[index++] = cmd->wptr[i]->u[1];
}
}
if (dma_queue_submit(dev, index, dpi_cmd))
return c;
}
return c;
}
/* Check for command completion, returns number of commands completed */
static int
otx2_dpi_rawdev_dequeue_bufs(struct rte_rawdev *dev __rte_unused,
struct rte_rawdev_buf **buffers,
unsigned int count, rte_rawdev_obj_t context)
{
struct dpi_dma_queue_ctx_s *ctx = (struct dpi_dma_queue_ctx_s *)context;
unsigned int i = 0, headp;
/* No completion ring to poll */
if (ctx->c_ring == NULL)
return 0;
headp = ctx->c_ring->head;
for (i = 0; i < count && (headp != ctx->c_ring->tail); i++) {
struct dpi_dma_req_compl_s *comp_ptr =
ctx->c_ring->compl_data[headp];
if (comp_ptr->cdata)
break;
/* Request Completed */
buffers[i] = (void *)comp_ptr;
headp = (headp + 1) % ctx->c_ring->max_cnt;
}
ctx->c_ring->head = headp;
return i;
}
static int
otx2_dpi_rawdev_start(struct rte_rawdev *dev)
{
dev->started = DPI_QUEUE_START;
return DPI_DMA_QUEUE_SUCCESS;
}
static void
otx2_dpi_rawdev_stop(struct rte_rawdev *dev)
{
dev->started = DPI_QUEUE_STOP;
}
static int
otx2_dpi_rawdev_close(struct rte_rawdev *dev)
{
dma_engine_enb_dis(dev->dev_private, false);
dma_queue_finish(dev->dev_private);
return DPI_DMA_QUEUE_SUCCESS;
}
static int
otx2_dpi_rawdev_reset(struct rte_rawdev *dev)
{
return dev ? DPI_QUEUE_STOP : DPI_QUEUE_START;
}
static int
otx2_dpi_rawdev_configure(const struct rte_rawdev *dev, rte_rawdev_obj_t config,
size_t config_size)
{
struct dpi_rawdev_conf_s *conf = config;
struct dpi_vf_s *dpivf = NULL;
void *buf = NULL;
uintptr_t pool;
uint32_t gaura;
if (conf == NULL || config_size != sizeof(*conf)) {
otx2_dpi_dbg("NULL or invalid configuration");
return -EINVAL;
}
dpivf = (struct dpi_vf_s *)dev->dev_private;
dpivf->chunk_pool = conf->chunk_pool;
if (rte_mempool_get(conf->chunk_pool, &buf) || (buf == NULL)) {
otx2_err("Unable allocate buffer");
return -ENODEV;
}
dpivf->base_ptr = buf;
otx2_write64(0x0, dpivf->vf_bar0 + DPI_VDMA_EN);
dpivf->pool_size_m1 = (DPI_CHUNK_SIZE >> 3) - 2;
pool = (uintptr_t)((struct rte_mempool *)conf->chunk_pool)->pool_id;
gaura = npa_lf_aura_handle_to_aura(pool);
otx2_write64(0, dpivf->vf_bar0 + DPI_VDMA_REQQ_CTL);
otx2_write64(((uint64_t)buf >> 7) << 7,
dpivf->vf_bar0 + DPI_VDMA_SADDR);
if (otx2_dpi_queue_open(dpivf, DPI_CHUNK_SIZE, gaura) < 0) {
otx2_err("Unable to open DPI VF %d", dpivf->vf_id);
rte_mempool_put(conf->chunk_pool, buf);
return -EACCES;
}
dma_engine_enb_dis(dpivf, true);
return DPI_DMA_QUEUE_SUCCESS;
}
static const struct rte_rawdev_ops dpi_rawdev_ops = {
.dev_configure = otx2_dpi_rawdev_configure,
.dev_start = otx2_dpi_rawdev_start,
.dev_stop = otx2_dpi_rawdev_stop,
.dev_close = otx2_dpi_rawdev_close,
.dev_reset = otx2_dpi_rawdev_reset,
.enqueue_bufs = otx2_dpi_rawdev_enqueue_bufs,
.dequeue_bufs = otx2_dpi_rawdev_dequeue_bufs,
.dev_selftest = test_otx2_dma_rawdev,
};
static int
otx2_dpi_rawdev_probe(struct rte_pci_driver *pci_drv __rte_unused,
struct rte_pci_device *pci_dev)
{
char name[RTE_RAWDEV_NAME_MAX_LEN];
struct dpi_vf_s *dpivf = NULL;
struct rte_rawdev *rawdev;
uint16_t vf_id;
/* For secondary processes, the primary has done all the work */
if (rte_eal_process_type() != RTE_PROC_PRIMARY)
return DPI_DMA_QUEUE_SUCCESS;
if (pci_dev->mem_resource[0].addr == NULL) {
otx2_dpi_dbg("Empty bars %p %p", pci_dev->mem_resource[0].addr,
pci_dev->mem_resource[2].addr);
return -ENODEV;
}
memset(name, 0, sizeof(name));
snprintf(name, RTE_RAWDEV_NAME_MAX_LEN, "DPI:%x:%02x.%x",
pci_dev->addr.bus, pci_dev->addr.devid,
pci_dev->addr.function);
/* Allocate device structure */
rawdev = rte_rawdev_pmd_allocate(name, sizeof(struct dpi_vf_s),
rte_socket_id());
if (rawdev == NULL) {
otx2_err("Rawdev allocation failed");
return -EINVAL;
}
rawdev->dev_ops = &dpi_rawdev_ops;
rawdev->device = &pci_dev->device;
rawdev->driver_name = pci_dev->driver->driver.name;
dpivf = rawdev->dev_private;
if (dpivf->state != DPI_QUEUE_STOP) {
otx2_dpi_dbg("Device already started!!!");
return -ENODEV;
}
vf_id = ((pci_dev->addr.devid & 0x1F) << 3) |
(pci_dev->addr.function & 0x7);
vf_id -= 1;
dpivf->dev = pci_dev;
dpivf->state = DPI_QUEUE_START;
dpivf->vf_id = vf_id;
dpivf->vf_bar0 = (uintptr_t)pci_dev->mem_resource[0].addr;
dpivf->vf_bar2 = (uintptr_t)pci_dev->mem_resource[2].addr;
return DPI_DMA_QUEUE_SUCCESS;
}
static int
otx2_dpi_rawdev_remove(struct rte_pci_device *pci_dev)
{
char name[RTE_RAWDEV_NAME_MAX_LEN];
struct rte_rawdev *rawdev;
struct dpi_vf_s *dpivf;
if (pci_dev == NULL) {
otx2_dpi_dbg("Invalid pci_dev of the device!");
return -EINVAL;
}
memset(name, 0, sizeof(name));
snprintf(name, RTE_RAWDEV_NAME_MAX_LEN, "DPI:%x:%02x.%x",
pci_dev->addr.bus, pci_dev->addr.devid,
pci_dev->addr.function);
rawdev = rte_rawdev_pmd_get_named_dev(name);
if (rawdev == NULL) {
otx2_dpi_dbg("Invalid device name (%s)", name);
return -EINVAL;
}
dpivf = (struct dpi_vf_s *)rawdev->dev_private;
dma_engine_enb_dis(dpivf, false);
dma_queue_finish(dpivf);
/* rte_rawdev_close is called by pmd_release */
return rte_rawdev_pmd_release(rawdev);
}
static struct rte_pci_driver rte_dpi_rawdev_pmd = {
.id_table = pci_dma_map,
.drv_flags = RTE_PCI_DRV_NEED_MAPPING | RTE_PCI_DRV_NEED_IOVA_AS_VA,
.probe = otx2_dpi_rawdev_probe,
.remove = otx2_dpi_rawdev_remove,
};
RTE_PMD_REGISTER_PCI(dpi_rawdev_pci_driver, rte_dpi_rawdev_pmd);
RTE_PMD_REGISTER_PCI_TABLE(dpi_rawdev_pci_driver, pci_dma_map);
RTE_PMD_REGISTER_KMOD_DEP(dpi_rawdev_pci_driver, "vfio-pci");