Files
linux-nvgpu/drivers/gpu/nvgpu/common/nvs/nvs-control-interface-parser.c
Debarshi Dutta 562c4f6ea3 gpu: nvgpu: add infra for manual mode submits in KMD
Added infrastructure for enabling parsing Control-Fifo's
ring buffers(i.e. send/receive).

Initialization of these buffers are handled as part of
nvgpu_nvs_buffer_alloc() call itself.

A follow-up change shall implement the methods defined here
as part of the existing NVS worker thread.

The changes adhered to the design laid out in the header
nvsched/include/nvs/nvs-control-interface.h.

Jira NVGPU-8619

Signed-off-by: Debarshi Dutta <ddutta@nvidia.com>
Change-Id: I2050e6fb681eba80e01cf547ada37a955e58315a
Reviewed-on: https://git-master.nvidia.com/r/c/linux-nvgpu/+/2764518
Tested-by: mobile promotions <svcmobile_promotions@nvidia.com>
Reviewed-by: mobile promotions <svcmobile_promotions@nvidia.com>
2022-10-10 14:07:52 -07:00

322 lines
9.1 KiB
C

/*
* Copyright (c) 2022, NVIDIA CORPORATION. All rights reserved.
*
* Permission is hereby granted, free of charge, to any person obtaining a
* copy of this software and associated documentation files (the "Software"),
* to deal in the Software without restriction, including without limitation
* the rights to use, copy, modify, merge, publish, distribute, sublicense,
* and/or sell copies of the Software, and to permit persons to whom the
* Software is furnished to do so, subject to the following conditions:
*
* The above copyright notice and this permission notice shall be included in
* all copies or substantial portions of the Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
* FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
* DEALINGS IN THE SOFTWARE.
*/
#include <nvs/log.h>
#include <nvgpu/utils.h>
#include <nvgpu/barrier.h>
#include <nvgpu/atomic.h>
#include <nvgpu/static_analysis.h>
#include <nvgpu/string.h>
#include <nvgpu/log.h>
#include <nvgpu/kmem.h>
#include <nvgpu/nvs.h>
#include <nvgpu/nvs-control-interface-parser.h>
static const u32 min_queue_size = 2U;
static u32 nvs_control_atomic_read(void *const address)
{
u32 value;
value = NV_READ_ONCE(*(u32 *)address);
nvgpu_rmb();
return value;
}
static void nvs_control_atomic_write(void *address, u32 value)
{
NV_WRITE_ONCE(*(u32 *)address, value);
nvgpu_wmb();
}
static u64 nvs_control_atomic64_read(void *const address)
{
u64 value;
value = NV_READ_ONCE(*(u64 *)address);
nvgpu_rmb();
return value;
}
static void nvs_control_atomic64_write(void *address, u64 value)
{
NV_WRITE_ONCE(*(u64 *)address, value);
nvgpu_wmb();
}
void nvs_control_fifo_sender_write_message(struct nvs_control_fifo_sender *const sender,
const u32 msg_number, const u32 msg_sequence_tag,
const u64 msg_timestamp_ns)
{
u64 updated_put_revolutions;
struct nvs_domain_message * const write_loc = &sender->fifo[sender->put_index];
nvgpu_mb();
write_loc->type = msg_number;
write_loc->sequence_tag = msg_sequence_tag;
write_loc->timestamp_ns = msg_timestamp_ns;
nvgpu_memcpy((u8 *)&write_loc->payload, (u8 *)sender->internal_buffer,
NVS_DOMAIN_MESSAGE_MAX_PAYLOAD_SIZE);
nvgpu_wmb();
sender->put_index++;
if (sender->put_index == sender->num_queue_entries) {
sender->put_index = 0;
sender->num_revolutions = nvgpu_wrapping_add_u32(sender->num_revolutions, 1U);
}
updated_put_revolutions = hi32_lo32_to_u64(sender->num_revolutions,
sender->put_index);
nvs_control_atomic64_write(&sender->control_interface->put_revolutions,
updated_put_revolutions);
}
void nvs_control_fifo_sender_out_of_space(struct nvs_control_fifo_sender *const sender)
{
sender->num_dropped_messages++;
nvs_control_atomic64_write(&sender->control_interface->num_dropped_messages,
sender->num_dropped_messages);
}
int nvs_control_fifo_sender_can_write(struct nvs_control_fifo_sender *const sender)
{
u64 curr_put_revolution;
u64 occupied_slots;
u64 available_slots;
u64 curr_get_index = nvs_control_atomic_read(&sender->control_interface->get);
/* bound check the get index */
if (curr_get_index >= sender->num_queue_entries) {
nvgpu_err(sender->g, "Get is out of bounds");
return -EINVAL;
} else if (curr_get_index == NVS_DOMAIN_MSG_FIFO_CONTROL_GET_FLOW_CTRL_DISABLED) {
/* space is always available */
return 0;
}
/* bound check the put index */
if (sender->put_index >= sender->num_queue_entries) {
nvgpu_err(sender->g, "Put is out of bounds. Probable memory corruption");
return -EBADF;
}
curr_put_revolution = nvs_control_atomic64_read(
&sender->control_interface->put_revolutions);
if (curr_put_revolution != hi32_lo32_to_u64(sender->num_revolutions,
sender->put_index)) {
nvgpu_err(sender->g, "Put index has changed since our last index");
return -EINVAL;
}
if (sender->put_index >= curr_get_index) {
occupied_slots = sender->put_index - curr_get_index;
} else {
occupied_slots = sender->num_queue_entries - curr_get_index + sender->put_index;
}
available_slots = sender->num_queue_entries - occupied_slots - 1U;
if (available_slots == 0) {
return -EAGAIN;
}
return 0;
}
struct nvs_control_fifo_sender *nvs_control_fifo_sender_initialize(
struct gk20a *g, struct nvs_domain_msg_fifo *const ring_buffer,
u64 buffer_size_bytes)
{
u64 current_put_revolutions;
struct nvs_control_fifo_sender *sender = NULL;
const u32 lower_bound_buffer_size =
nvgpu_safe_add_u32(nvgpu_safe_mult_u32(min_queue_size,
sizeof(struct nvs_domain_message)),
sizeof(struct nvs_domain_msg_fifo_control));
if (ring_buffer == NULL) {
nvgpu_err(g, "ring buffer is NULL");
return NULL;
}
if (buffer_size_bytes < lower_bound_buffer_size) {
nvgpu_err(g, "buffer size must be a minimum of 2 entries");
return NULL;
}
sender = nvgpu_kzalloc(g, sizeof(*sender));
if (sender == NULL) {
return NULL;
}
sender->g = g;
sender->num_queue_entries = (nvgpu_safe_sub_u64(buffer_size_bytes,
sizeof(struct nvs_domain_msg_fifo_control))) / sizeof(
struct nvs_domain_message);
sender->fifo = ring_buffer->messages;
sender->control_interface = &ring_buffer->control;
current_put_revolutions = nvs_control_atomic64_read(
&sender->control_interface->put_revolutions);
sender->put_index = u64_lo32(current_put_revolutions);
sender->num_revolutions = u64_hi32(current_put_revolutions);
sender->num_dropped_messages = nvs_control_atomic64_read(
&sender->control_interface->num_dropped_messages);
if (sender->put_index >= sender->num_queue_entries) {
nvgpu_err(g, "Put Index more than Max Queue size");
nvgpu_kfree(g, sender);
return NULL;
}
return sender;
}
void nvs_control_fifo_read_message(struct nvs_control_fifo_receiver *const receiver)
{
struct nvs_domain_message *const read_loc = &receiver->fifo[receiver->get_index];
nvgpu_rmb();
/* Copy the message from the buffer */
receiver->msg_type = read_loc->type;
receiver->msg_sequence = read_loc->sequence_tag;
receiver->msg_timestamp_ns = read_loc->timestamp_ns;
memset((u8 *)&receiver->internal_buffer, 0, NVS_DOMAIN_MESSAGE_MAX_PAYLOAD_SIZE);
nvgpu_memcpy((u8 *)&receiver->internal_buffer,
(u8 *)&read_loc->payload, NVS_DOMAIN_MESSAGE_MAX_PAYLOAD_SIZE);
nvgpu_mb();
receiver->get_index++;
if (receiver->get_index == receiver->num_queue_entries) {
receiver->get_index = 0;
}
nvs_control_atomic_write(&receiver->control_interface->get, receiver->get_index);
}
int nvs_control_fifo_receiver_can_read(struct nvs_control_fifo_receiver *const receiver)
{
u32 put;
u64 curr_put_revolution;
curr_put_revolution = nvs_control_atomic64_read(
&receiver->control_interface->put_revolutions);
put = u64_lo32(curr_put_revolution);
if (put == receiver->get_index) {
nvs_dbg(receiver->g, "No new message");
return -EAGAIN;
}
return 0;
}
struct nvs_control_fifo_receiver *nvs_control_fifo_receiver_initialize(
struct gk20a *g, struct nvs_domain_msg_fifo *const fifo,
u64 buffer_size_bytes)
{
struct nvs_control_fifo_receiver *receiver = NULL;
u64 num_put_revolutions;
const u32 lower_bound_buffer_size =
nvgpu_safe_add_u32(nvgpu_safe_mult_u32(min_queue_size,
sizeof(struct nvs_domain_message)),
sizeof(struct nvs_domain_msg_fifo_control));
if (fifo == NULL) {
nvgpu_err(g, "ring buffer is NULL");
return NULL;
}
if (buffer_size_bytes < lower_bound_buffer_size) {
nvgpu_err(g, "buffer size must be a minimum of 2 entries");
return NULL;
}
receiver = nvgpu_kzalloc(g, sizeof(*receiver));
if (receiver == NULL) {
return NULL;
}
receiver->g = g;
receiver->num_queue_entries = (nvgpu_safe_sub_u64(buffer_size_bytes,
sizeof(struct nvs_domain_msg_fifo_control))) / sizeof(
struct nvs_domain_message);
receiver->fifo = fifo->messages;
receiver->control_interface = &fifo->control;
num_put_revolutions = nvs_control_atomic64_read(
&receiver->control_interface->put_revolutions);
receiver->get_index = u64_lo32(num_put_revolutions);
if (receiver->get_index >= receiver->num_queue_entries) {
nvgpu_err(g, "Get Index more than Max Queue size");
nvgpu_kfree(g, receiver);
return NULL;
}
nvs_control_fifo_enable_flow_control(receiver->control_interface, receiver->get_index);
return receiver;
}
void nvs_control_fifo_sender_exit(struct gk20a *g,
struct nvs_control_fifo_sender *const sender)
{
nvs_control_fifo_disable_flow_control(sender->control_interface);
nvgpu_kfree(g, sender);
}
void nvs_control_fifo_receiver_exit(struct gk20a *g,
struct nvs_control_fifo_receiver *const receiver)
{
nvs_control_fifo_disable_flow_control(receiver->control_interface);
nvgpu_kfree(g, receiver);
}
void nvs_control_fifo_enable_flow_control(struct nvs_domain_msg_fifo_control *control_interface,
u32 get_index)
{
nvs_control_atomic_write(&control_interface->get, get_index);
}
void nvs_control_fifo_disable_flow_control(struct nvs_domain_msg_fifo_control *control_interface)
{
nvs_control_atomic_write(&control_interface->get,
NVS_DOMAIN_MSG_FIFO_CONTROL_GET_FLOW_CTRL_DISABLED);
}