nv-tegra-mirror/linux-nv-oot
1
0
Fork 0
mirror of git://nv-tegra.nvidia.com/linux-nv-oot.git synced 2025-12-24 02:01:36 +03:00
Code Issues Packages Projects Releases Wiki Activity

eventlib: unify NVDLA/PVA fence types and add fence recording

Browse Source 
Bug 2170736

Change-Id: If4fdeda140bf2474a08beb2a0d7c3fc1737a1a66
Signed-off-by: Dmitry Antipov <dantipov@nvidia.com>
Reviewed-on: https://git-master.nvidia.com/r/1750906
Reviewed-by: Saleh Dindar <sdindar@nvidia.com>
GVS: Gerrit_Virtual_Submit
Reviewed-by: Colin Tracey <ctracey@nvidia.com>
Reviewed-by: Prashant Gaikwad <pgaikwad@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
This commit is contained in:
Dmitry Antipov
2018-08-14 14:17:05 +03:00
committed by Laxman Dewangan
parent e44a671c9b
commit c9b3772b5e
3 changed files with 80 additions and 69 deletions
Download Patch File Download Diff File Expand all files Collapse all files

7
drivers/video/tegra/host/nvdla/nvdla.h
View File

@@ -23,6 +23,7 @@
#include <linux/completion.h>
#include <linux/mutex.h>
#include <uapi/linux/nvdev_fence.h>
#include <linux/nvhost_nvdla_ioctl.h>
#include "nvhost_buffer.h"
@@ -226,7 +227,7 @@ struct nvdla_device {
struct nvdla_emu_task {
struct nvhost_queue *queue;
struct nvhost_syncpt *sp;
struct nvdla_fence postfences[MAX_NUM_NVDLA_POSTFENCES];
struct nvdev_fence postfences[MAX_NUM_NVDLA_POSTFENCES];
u32 num_postfences;
u32 fence;
u32 fence_counter;
@@ -254,8 +255,8 @@ struct nvdla_task {
struct nvhost_queue *queue;
struct nvhost_buffers *buffers;
struct nvhost_syncpt *sp;
struct nvdla_fence prefences[MAX_NUM_NVDLA_PREFENCES];
struct nvdla_fence postfences[MAX_NUM_NVDLA_POSTFENCES];
struct nvdev_fence prefences[MAX_NUM_NVDLA_PREFENCES];
struct nvdev_fence postfences[MAX_NUM_NVDLA_POSTFENCES];
struct nvdla_status_notify in_task_status[MAX_NUM_NVDLA_IN_TASK_STATUS];
struct nvdla_status_notify out_task_status[MAX_NUM_NVDLA_OUT_TASK_STATUS];
struct nvdla_mem_handle memory_handles[NVDLA_MAX_BUFFERS_PER_TASK];

51
drivers/video/tegra/host/nvdla/nvdla_ioctl.c
View File

@@ -34,6 +34,7 @@
#include "nvdla/nvdla.h"
#include "nvdla/nvdla_debug.h"
#include <uapi/linux/nvdev_fence.h>
#include <linux/nvhost_nvdla_ioctl.h>
#include "dla_os_interface.h"
@@ -126,11 +127,11 @@ static int nvdla_get_q_status(struct nvdla_private *priv, void *args)
{
struct nvdla_get_q_status_args *queue_arg =
(struct nvdla_get_q_status_args *)args;
struct nvdla_fence __user *usr_fence =
(struct nvdla_fence __user *)(uintptr_t)queue_arg->fence;
struct nvdev_fence __user *usr_fence =
(struct nvdev_fence __user *)(uintptr_t)queue_arg->fence;
struct platform_device *pdev = priv->pdev;
struct nvhost_queue *queue = priv->queue;
struct nvdla_fence fence = {0};
struct nvdev_fence fence = {0};
int err = 0;
nvdla_dbg_fn(pdev, "");
@@ -146,7 +147,7 @@ static int nvdla_get_q_status(struct nvdla_private *priv, void *args)
queue->syncpt_id);
nvdla_dbg_info(pdev, "syncpt_id[%u] val[%u]\n", fence.syncpoint_index, fence.syncpoint_value);
if (copy_to_user(usr_fence, &fence, sizeof(struct nvdla_fence))) {
if (copy_to_user(usr_fence, &fence, sizeof(struct nvdev_fence))) {
err = -EFAULT;
nvdla_dbg_err(pdev, "failed to send fence");
goto fail_to_send_fence;
@@ -347,7 +348,7 @@ static int nvdla_get_actions(struct nvdla_ioctl_submit_task *user_task,
/* get pre fences */
if (copy_from_user(task->prefences,
(void __user *)user_task->prefences,
(task->num_prefences * sizeof(struct nvdla_fence)))) {
(task->num_prefences * sizeof(struct nvdev_fence)))) {
err = -EFAULT;
nvdla_dbg_err(pdev, "failed to copy prefences");
goto fail;
@@ -366,7 +367,7 @@ static int nvdla_get_actions(struct nvdla_ioctl_submit_task *user_task,
/* get post fences */
if (copy_from_user(task->postfences,
(void __user *)user_task->postfences,
(task->num_postfences * sizeof(struct nvdla_fence)))) {
(task->num_postfences * sizeof(struct nvdev_fence)))) {
err = -EFAULT;
nvdla_dbg_err(pdev, "failed to copy postfences");
goto fail;
@@ -395,14 +396,14 @@ static int nvdla_send_emu_postfences(struct nvdla_emu_task *task,
struct platform_device *dla_pdev = task->queue->pool->pdev;
struct platform_device *host_pdev =
to_platform_device(dla_pdev->dev.parent);
struct nvdla_fence __user *postfences =
(struct nvdla_fence __user *)(uintptr_t)user_task->postfences;
struct nvdev_fence __user *postfences =
(struct nvdev_fence __user *)(uintptr_t)user_task->postfences;
char fence_name[32];
nvdla_dbg_fn(dla_pdev, "sending post fences");
for (i = 0; i < task->num_postfences; i++) {
if (task->postfences[i].type == NVDLA_FENCE_TYPE_SYNC_FD) {
if (task->postfences[i].type == NVDEV_FENCE_TYPE_SYNC_FD) {
struct nvhost_ctrl_sync_fence_info info;
info.id = task->postfences[i].syncpoint_index;
@@ -432,7 +433,7 @@ static int nvdla_send_emu_postfences(struct nvdla_emu_task *task,
nvdla_dbg_fn(dla_pdev, "copy postfences to user");
/* send post fences */
if (copy_to_user(postfences, task->postfences,
(task->num_postfences * sizeof(struct nvdla_fence)))) {
(task->num_postfences * sizeof(struct nvdev_fence)))) {
err = -EFAULT;
nvdla_dbg_err(dla_pdev, "failed to send postfences");
goto fail;
@@ -450,14 +451,14 @@ static int nvdla_update_postfences(struct nvdla_task *task,
struct platform_device *dla_pdev = task->queue->pool->pdev;
struct platform_device *host_pdev =
to_platform_device(dla_pdev->dev.parent);
struct nvdla_fence __user *postfences =
(struct nvdla_fence __user *)(uintptr_t)user_task->postfences;
struct nvdev_fence __user *postfences =
(struct nvdev_fence __user *)(uintptr_t)user_task->postfences;
char fence_name[32];
nvdla_dbg_fn(dla_pdev, "copy post fences for user");
for (i = 0; i < task->num_postfences; i++) {
if (task->postfences[i].type == NVDLA_FENCE_TYPE_SYNC_FD) {
if (task->postfences[i].type == NVDEV_FENCE_TYPE_SYNC_FD) {
struct nvhost_ctrl_sync_fence_info info;
info.id = task->postfences[i].syncpoint_index;
@@ -487,7 +488,7 @@ static int nvdla_update_postfences(struct nvdla_task *task,
nvdla_dbg_fn(dla_pdev, "copy postfences to user");
/* copy post fences */
if (copy_to_user(postfences, task->postfences,
(task->num_postfences * sizeof(struct nvdla_fence)))) {
(task->num_postfences * sizeof(struct nvdev_fence)))) {
err = -EFAULT;
nvdla_dbg_err(dla_pdev, "failed to copy postfences");
goto fail;
@@ -618,28 +619,28 @@ static void nvdla_dump_task(struct nvdla_task *task)
for (i = 0; i < task->num_prefences; i++) {
nvdla_dbg_info(pdev, "prefence[%d]: type[%u] syncpt_index[%u]"
" syncpt_val[%u] sync_fd[%u] sem_handle[%u]"
" sem_offset[%u] sem_val[%u]",
" syncpt_val[%u] sync_fd[%u] semaphore_handle[%u]"
" semaphore_offset[%u] semaphore_value[%u]",
i, task->prefences[i].type,
task->prefences[i].syncpoint_index,
task->prefences[i].syncpoint_value,
task->prefences[i].sync_fd,
task->prefences[i].sem_handle,
task->prefences[i].sem_offset,
task->prefences[i].sem_val);
task->prefences[i].semaphore_handle,
task->prefences[i].semaphore_offset,
task->prefences[i].semaphore_value);
}
for (i = 0; i < task->num_postfences; i++) {
nvdla_dbg_info(pdev, "postfence[%d]: type[%u] syncpt_index[%u]"
" syncpt_val[%u] sync_fd[%u] sem_handle[%u]"
" sem_offset[%u] sem_val[%u]",
" syncpt_val[%u] sync_fd[%u] semaphore_handle[%u]"
" semaphore_offset[%u] semaphore_value[%u]",
i, task->postfences[i].type,
task->postfences[i].syncpoint_index,
task->postfences[i].syncpoint_value,
task->postfences[i].sync_fd,
task->postfences[i].sem_handle,
task->postfences[i].sem_offset,
task->postfences[i].sem_val);
task->postfences[i].semaphore_handle,
task->postfences[i].semaphore_offset,
task->postfences[i].semaphore_value);
}
for (i = 0; i < task->num_in_task_status; i++) {
@@ -719,7 +720,7 @@ static int nvdla_emu_task_submit(struct nvdla_private *priv, void *arg)
/* get post fences */
if (copy_from_user(task.postfences,
(void __user *)local_tasks[i].postfences,
(task.num_postfences * sizeof(struct nvdla_fence)))) {
(task.num_postfences * sizeof(struct nvdev_fence)))) {
err = -EFAULT;
nvdla_dbg_err(pdev, "failed to copy postfences");
goto exit;

91
drivers/video/tegra/host/nvdla/nvdla_queue.c
View File

@@ -177,9 +177,9 @@ static int nvdla_unmap_task_memory(struct nvdla_task *task)
/* unpin prefences memory */
for (ii = 0; ii < task->num_prefences; ii++) {
if ((task->prefences[ii].type == NVDLA_FENCE_TYPE_SEMAPHORE ||
task->prefences[ii].type == NVDLA_FENCE_TYPE_TS_SEMAPHORE) &&
task->prefences[ii].sem_handle) {
if ((task->prefences[ii].type == NVDEV_FENCE_TYPE_SEMAPHORE ||
task->prefences[ii].type == NVDEV_FENCE_TYPE_SEMAPHORE_TS) &&
task->prefences[ii].semaphore_handle) {
nvhost_buffer_submit_unpin(task->buffers,
&task->prefences_sem_dmabuf[ii], 1);
dma_buf_put(task->prefences_sem_dmabuf[ii]);
@@ -199,9 +199,9 @@ static int nvdla_unmap_task_memory(struct nvdla_task *task)
/* unpin postfences memory */
for (ii = 0; ii < task->num_postfences; ii++) {
if ((task->postfences[ii].type == NVDLA_FENCE_TYPE_SEMAPHORE ||
task->postfences[ii].type == NVDLA_FENCE_TYPE_TS_SEMAPHORE) &&
task->postfences[ii].sem_handle) {
if ((task->postfences[ii].type == NVDEV_FENCE_TYPE_SEMAPHORE ||
task->postfences[ii].type == NVDEV_FENCE_TYPE_SEMAPHORE_TS) &&
task->postfences[ii].semaphore_handle) {
nvhost_buffer_submit_unpin(task->buffers,
&task->postfences_sem_dmabuf[ii], 1);
dma_buf_put(task->postfences_sem_dmabuf[ii]);
@@ -293,7 +293,7 @@ static inline size_t nvdla_profile_status_offset(struct nvdla_task *task)
static void nvdla_queue_update(void *priv, int nr_completed)
{
int task_complete;
int i, task_complete;
struct nvdla_task *task, *safe;
struct nvhost_queue *queue = priv;
struct platform_device *pdev = queue->pool->pdev;
@@ -330,6 +330,11 @@ static void nvdla_queue_update(void *priv, int nr_completed)
timestamp_start,
timestamp_end);
for (i = 0; i < task->num_postfences; i++)
nvhost_eventlib_log_fence(pdev,
NVDEV_FENCE_KIND_POST,
task->postfences + i, timestamp_end);
nvdla_task_free_locked(task);
}
}
@@ -602,13 +607,13 @@ static int nvdla_fill_postactions(struct nvdla_task *task)
/* reset fence counter */
task->fence_counter = 0;
/* fill all postactions */
for (i = 0; i < task->num_postfences; i++) {
/* update action */
switch (task->postfences[i].type) {
case NVDLA_FENCE_TYPE_SYNCPT:
case NVDLA_FENCE_TYPE_SYNC_FD: {
case NVDEV_FENCE_TYPE_SYNCPT:
case NVDEV_FENCE_TYPE_SYNC_FD: {
dma_addr_t syncpt_addr;
u32 gos_id, gos_offset;
@@ -641,22 +646,22 @@ static int nvdla_fill_postactions(struct nvdla_task *task)
i, queue->syncpt_id, &syncpt_addr);
break;
}
case NVDLA_FENCE_TYPE_TS_SEMAPHORE: {
case NVDEV_FENCE_TYPE_SEMAPHORE_TS: {
dma_addr_t dma_addr;
size_t dma_size;
nvdla_dbg_info(pdev, "POSTTS i:%d semh:%u semo:%u v:%d",
i,
task->postfences[i].sem_handle,
task->postfences[i].sem_offset,
task->postfences[i].sem_val);
task->postfences[i].semaphore_handle,
task->postfences[i].semaphore_offset,
task->postfences[i].semaphore_value);
/* TS SEMAPHORE just has extra memory bytes allocated
* to store TS as compared default semaphore.
* override action/opecode type here.
*/
task->postfences_sem_dmabuf[i] =
dma_buf_get(task->postfences[i].sem_handle);
dma_buf_get(task->postfences[i].semaphore_handle);
if (IS_ERR_OR_NULL(task->postfences_sem_dmabuf[i])) {
task->postfences_sem_dmabuf[i] = NULL;
nvdla_dbg_err(pdev, "fail to get buf");
@@ -671,22 +676,22 @@ static int nvdla_fill_postactions(struct nvdla_task *task)
}
next = add_fence_action(next, POSTACTION_TS_SEM,
dma_addr + task->postfences[i].sem_offset,
task->postfences[i].sem_val);
dma_addr + task->postfences[i].semaphore_offset,
task->postfences[i].semaphore_value);
break;
}
case NVDLA_FENCE_TYPE_SEMAPHORE: {
case NVDEV_FENCE_TYPE_SEMAPHORE: {
dma_addr_t dma_addr;
size_t dma_size;
nvdla_dbg_info(pdev, "POST i:%d semh:%u semo:%u v:%d",
i,
task->postfences[i].sem_handle,
task->postfences[i].sem_offset,
task->postfences[i].sem_val);
task->postfences[i].semaphore_handle,
task->postfences[i].semaphore_offset,
task->postfences[i].semaphore_value);
task->postfences_sem_dmabuf[i] =
dma_buf_get(task->postfences[i].sem_handle);
dma_buf_get(task->postfences[i].semaphore_handle);
if (IS_ERR_OR_NULL(task->postfences_sem_dmabuf[i])) {
task->postfences_sem_dmabuf[i] = NULL;
nvdla_dbg_err(pdev, "fail to get buf");
@@ -701,8 +706,8 @@ static int nvdla_fill_postactions(struct nvdla_task *task)
}
next = add_fence_action(next, POSTACTION_SEM,
dma_addr + task->postfences[i].sem_offset,
task->postfences[i].sem_val);
dma_addr + task->postfences[i].semaphore_offset,
task->postfences[i].semaphore_value);
break;
}
default:
@@ -731,6 +736,7 @@ static int nvdla_fill_preactions(struct nvdla_task *task)
struct platform_device *pdev = queue->pool->pdev;
struct nvhost_master *host = nvhost_get_host(pdev);
struct nvhost_syncpt *sp = &host->syncpt;
u64 timestamp = arch_counter_get_cntvct();
struct dla_action_list *preactionl;
uint16_t preactionlist_of;
u8 *next, *start;
@@ -746,8 +752,11 @@ static int nvdla_fill_preactions(struct nvdla_task *task)
/* fill all preactions */
for (i = 0; i < task->num_prefences; i++) {
nvhost_eventlib_log_fence(pdev, NVDEV_FENCE_KIND_PRE,
task->prefences + i, timestamp);
switch (task->prefences[i].type) {
case NVDLA_FENCE_TYPE_SYNC_FD: {
case NVDEV_FENCE_TYPE_SYNC_FD: {
struct sync_fence *f;
struct sync_pt *pt;
u32 id, thresh, j;
@@ -798,7 +807,7 @@ static int nvdla_fill_preactions(struct nvdla_task *task)
}
break;
}
case NVDLA_FENCE_TYPE_SYNCPT: {
case NVDEV_FENCE_TYPE_SYNCPT: {
u32 gos_id, gos_offset;
nvdla_dbg_info(pdev, "i[%d] id[%d] val[%d]",
@@ -835,19 +844,19 @@ static int nvdla_fill_preactions(struct nvdla_task *task)
}
break;
}
case NVDLA_FENCE_TYPE_SEMAPHORE:
case NVDLA_FENCE_TYPE_TS_SEMAPHORE: {
case NVDEV_FENCE_TYPE_SEMAPHORE:
case NVDEV_FENCE_TYPE_SEMAPHORE_TS: {
dma_addr_t dma_addr;
size_t dma_size;
nvdla_dbg_info(pdev, "i[%d] semh[%u] semo[%u] val[%d]",
i,
task->prefences[i].sem_handle,
task->prefences[i].sem_offset,
task->prefences[i].sem_val);
task->prefences[i].semaphore_handle,
task->prefences[i].semaphore_offset,
task->prefences[i].semaphore_value);
task->prefences_sem_dmabuf[i] =
dma_buf_get(task->prefences[i].sem_handle);
dma_buf_get(task->prefences[i].semaphore_handle);
if (IS_ERR_OR_NULL(task->prefences_sem_dmabuf[i])) {
task->prefences_sem_dmabuf[i] = NULL;
nvdla_dbg_err(pdev, "fail to get buf");
@@ -862,8 +871,8 @@ static int nvdla_fill_preactions(struct nvdla_task *task)
}
next = add_fence_action(next, PREACTION_SEM_GE,
dma_addr + task->prefences[i].sem_offset,
task->prefences[i].sem_val);
dma_addr + task->prefences[i].semaphore_offset,
task->prefences[i].semaphore_value);
break;
}
default:
@@ -1013,7 +1022,7 @@ static int nvdla_send_cmd_channel(struct platform_device *pdev,
/* Pick up fences... */
for (i = 0; i < task->num_prefences; i++) {
/* ..and ensure that we have only syncpoints present */
if (task->prefences[i].type != NVDLA_FENCE_TYPE_SYNCPT) {
if (task->prefences[i].type != NVDEV_FENCE_TYPE_SYNCPT) {
nvdla_dbg_err(pdev, "syncpt only supported");
return -EINVAL;
}
@@ -1077,8 +1086,8 @@ int nvdla_emulator_submit(struct nvhost_queue *queue, struct nvdla_emu_task *tas
/* update action */
switch (task->postfences[i].type) {
case NVDLA_FENCE_TYPE_SYNCPT:
case NVDLA_FENCE_TYPE_SYNC_FD: {
case NVDEV_FENCE_TYPE_SYNCPT:
case NVDEV_FENCE_TYPE_SYNC_FD: {
task->fence_counter = task->fence_counter + 1;
break;
}
@@ -1100,8 +1109,8 @@ int nvdla_emulator_submit(struct nvhost_queue *queue, struct nvdla_emu_task *tas
/* Update postfences for all */
counter = task->fence_counter - 1;
for (i = 0; i < task->num_postfences; i++) {
if ((task->postfences[i].type == NVDLA_FENCE_TYPE_SYNCPT) ||
(task->postfences[i].type == NVDLA_FENCE_TYPE_SYNC_FD)) {
if ((task->postfences[i].type == NVDEV_FENCE_TYPE_SYNCPT) ||
(task->postfences[i].type == NVDEV_FENCE_TYPE_SYNC_FD)) {
task->postfences[i].syncpoint_index =
queue->syncpt_id;
task->postfences[i].syncpoint_value =
@@ -1139,8 +1148,8 @@ int nvdla_get_postfences(struct nvhost_queue *queue, void *in_task)
/* Update postfences for all */
counter = task->fence_counter - 1;
for (i = 0; i < task->num_postfences; i++) {
if ((task->postfences[i].type == NVDLA_FENCE_TYPE_SYNCPT) ||
(task->postfences[i].type == NVDLA_FENCE_TYPE_SYNC_FD)) {
if ((task->postfences[i].type == NVDEV_FENCE_TYPE_SYNCPT) ||
(task->postfences[i].type == NVDEV_FENCE_TYPE_SYNC_FD)) {
task->postfences[i].syncpoint_index =
queue->syncpt_id;
task->postfences[i].syncpoint_value =