rocm-systems/src/perfctr.c

/*
 * Copyright © 2014 Advanced Micro Devices, Inc.
 *
 * 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 (including
 * the next paragraph) 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 <stdlib.h>
#include <stdio.h>
#include <string.h>
#include <linux/perf_event.h>
#include <sys/syscall.h>
#include "libhsakmt.h"
#include "pmc_table.h"
#include "linux/kfd_ioctl.h"
#include <unistd.h>
#include <sys/ioctl.h>
#include <errno.h>
#include <sys/mman.h>
#include <fcntl.h>
#include <semaphore.h>

#define BITS_PER_BYTE		CHAR_BIT

#define HSA_PERF_MAGIC4CC	0x54415348

enum perf_trace_state {
	PERF_TRACE_STATE__STOPPED = 0,
	PERF_TRACE_STATE__STARTED
};

struct perf_trace_block {
	enum perf_block_id block_id;
	uint32_t num_counters;
	uint64_t *counter_id;
	int *perf_event_fd;
};

struct perf_trace {
	uint32_t magic4cc;
	uint32_t gpu_id;
	enum perf_trace_state state;
	uint32_t num_blocks;
	void *buf;
	uint64_t buf_size;
	struct perf_trace_block blocks[0];
};

enum perf_trace_action {
	PERF_TRACE_ACTION__ACQUIRE = 0,
	PERF_TRACE_ACTION__RELEASE
};

struct perf_shared_table {
	uint32_t magic4cc;
	uint32_t iommu_slots_left;
};

struct perf_counts_values {
	union {
		struct {
			u64 val;
			u64 ena;
			u64 run;
		};
		u64 values[3];
	};
};

static HsaCounterProperties **counter_props;
static unsigned int counter_props_count;
static const char shmem_name[] = "/hsakmt_shared_mem";
static int shmem_fd;
static const char sem_name[] = "hsakmt_semaphore";
static sem_t *sem = SEM_FAILED;
struct perf_shared_table *shared_table;

static ssize_t readn(int fd, void *buf, size_t n)
{
	size_t left = n;
	ssize_t bytes;

	while (left) {
		bytes = read(fd, buf, left);
		if (!bytes) /* reach EOF */
			return (n - left);
		if (bytes < 0) {
			if (errno == EINTR) /* read got interrupted */
				continue;
			else
				return -errno;
		}
		left -= bytes;
		buf = VOID_PTR_ADD(buf, bytes);
	}
	return n;
}

static HSAKMT_STATUS init_shared_region(void)
{
	sem = sem_open(sem_name, O_CREAT, 0666, 1);
	if (sem == SEM_FAILED)
		return HSAKMT_STATUS_ERROR;

	shmem_fd = shm_open(shmem_name, O_CREAT | O_RDWR, 0666);
	if (shmem_fd < 0)
		goto exit_1;

	if (ftruncate(shmem_fd, sizeof(struct perf_shared_table)) < 0)
		goto exit_2;

	shared_table = mmap(NULL, sizeof(*shared_table),
			PROT_READ | PROT_WRITE, MAP_SHARED, shmem_fd, 0);
	if (shared_table == MAP_FAILED)
		goto exit_2;

	return HSAKMT_STATUS_SUCCESS;

exit_2:
	shm_unlink(shmem_name);
	shmem_fd = 0;
exit_1:
	sem_close(sem);
	sem_unlink(sem_name);
	sem = SEM_FAILED;
	return HSAKMT_STATUS_ERROR;
}

static void destroy_shared_region(void)
{
	if (shared_table && shared_table != MAP_FAILED)
		munmap(shared_table, sizeof(*shared_table));

	if (shmem_fd > 0) {
		close(shmem_fd);
		shm_unlink(shmem_name);
	}

	if (sem != SEM_FAILED) {
		sem_close(sem);
		sem_unlink(sem_name);
		sem = SEM_FAILED;
	}
}

static void init_perf_shared_table(void)
{
	sem_wait(sem);

	/* If the magic number exists, the perf shared table has been
	 * initialized by another process and is in use. Don't overwrite it.
	 */
	if (shared_table->magic4cc == HSA_PERF_MAGIC4CC) {
		sem_post(sem);
		return;
	}

	/* write the perf content */
	shared_table->magic4cc = HSA_PERF_MAGIC4CC;
	shared_table->iommu_slots_left =
		pmc_table_get_max_concurrent(PERFCOUNTER_BLOCKID__IOMMUV2);

	sem_post(sem);
}

HSAKMT_STATUS init_counter_props(unsigned int NumNodes)
{
	counter_props = calloc(NumNodes, sizeof(struct HsaCounterProperties *));
	if (!counter_props) {
		pr_warn("Profiling is not available.\n");
		return HSAKMT_STATUS_NO_MEMORY;
	}

	counter_props_count = NumNodes;
	alloc_pmc_blocks();

	if (init_shared_region() != HSAKMT_STATUS_SUCCESS) {
		pr_warn("Profiling of privileged blocks is not available.\n");
		return HSAKMT_STATUS_ERROR;
	}
	init_perf_shared_table();

	return HSAKMT_STATUS_SUCCESS;
}

void destroy_counter_props(void)
{
	unsigned int i;

	destroy_shared_region();

	if (!counter_props)
		return;

	for (i = 0; i < counter_props_count; i++)
		if (counter_props[i]) {
			free(counter_props[i]);
			counter_props[i] = NULL;
		}

	free(counter_props);
	free_pmc_blocks();
}

static int blockid2uuid(enum perf_block_id block_id, HSA_UUID *uuid)
{
	int rc = 0;

	switch (block_id) {
	case PERFCOUNTER_BLOCKID__CB:
		*uuid = HSA_PROFILEBLOCK_AMD_CB;
		break;
	case PERFCOUNTER_BLOCKID__CPF:
		*uuid = HSA_PROFILEBLOCK_AMD_CPF;
		break;
	case PERFCOUNTER_BLOCKID__CPG:
		*uuid = HSA_PROFILEBLOCK_AMD_CPG;
		break;
	case PERFCOUNTER_BLOCKID__DB:
		*uuid = HSA_PROFILEBLOCK_AMD_DB;
		break;
	case PERFCOUNTER_BLOCKID__GDS:
		*uuid = HSA_PROFILEBLOCK_AMD_GDS;
		break;
	case PERFCOUNTER_BLOCKID__GRBM:
		*uuid = HSA_PROFILEBLOCK_AMD_GRBM;
		break;
	case PERFCOUNTER_BLOCKID__GRBMSE:
		*uuid = HSA_PROFILEBLOCK_AMD_GRBMSE;
		break;
	case PERFCOUNTER_BLOCKID__IA:
		*uuid = HSA_PROFILEBLOCK_AMD_IA;
		break;
	case PERFCOUNTER_BLOCKID__MC:
		*uuid = HSA_PROFILEBLOCK_AMD_MC;
		break;
	case PERFCOUNTER_BLOCKID__PASC:
		*uuid = HSA_PROFILEBLOCK_AMD_PASC;
		break;
	case PERFCOUNTER_BLOCKID__PASU:
		*uuid = HSA_PROFILEBLOCK_AMD_PASU;
		break;
	case PERFCOUNTER_BLOCKID__SPI:
		*uuid = HSA_PROFILEBLOCK_AMD_SPI;
		break;
	case PERFCOUNTER_BLOCKID__SRBM:
		*uuid = HSA_PROFILEBLOCK_AMD_SRBM;
		break;
	case PERFCOUNTER_BLOCKID__SQ:
		*uuid = HSA_PROFILEBLOCK_AMD_SQ;
		break;
	case PERFCOUNTER_BLOCKID__SX:
		*uuid = HSA_PROFILEBLOCK_AMD_SX;
		break;
	case PERFCOUNTER_BLOCKID__TA:
		*uuid = HSA_PROFILEBLOCK_AMD_TA;
		break;
	case PERFCOUNTER_BLOCKID__TCA:
		*uuid = HSA_PROFILEBLOCK_AMD_TCA;
		break;
	case PERFCOUNTER_BLOCKID__TCC:
		*uuid = HSA_PROFILEBLOCK_AMD_TCC;
		break;
	case PERFCOUNTER_BLOCKID__TCP:
		*uuid = HSA_PROFILEBLOCK_AMD_TCP;
		break;
	case PERFCOUNTER_BLOCKID__TCS:
		*uuid = HSA_PROFILEBLOCK_AMD_TCS;
		break;
	case PERFCOUNTER_BLOCKID__TD:
		*uuid = HSA_PROFILEBLOCK_AMD_TD;
		break;
	case PERFCOUNTER_BLOCKID__VGT:
		*uuid = HSA_PROFILEBLOCK_AMD_VGT;
		break;
	case PERFCOUNTER_BLOCKID__WD:
		*uuid = HSA_PROFILEBLOCK_AMD_WD;
		break;
	case PERFCOUNTER_BLOCKID__IOMMUV2:
		*uuid = HSA_PROFILEBLOCK_AMD_IOMMUV2;
		break;
	default:
		/* If we reach this point, it's a bug */
		rc = -1;
		break;
	}

	return rc;
}

static HSAuint32 get_block_concurrent_limit(uint32_t node_id,
						HSAuint32 block_id)
{
	uint32_t i;
	HsaCounterBlockProperties *block = &counter_props[node_id]->Blocks[0];

	for (i = 0; i < PERFCOUNTER_BLOCKID__MAX; i++) {
		if (block->Counters[0].BlockIndex == block_id)
			return block->NumConcurrent;
		block = (HsaCounterBlockProperties *)&block->Counters[block->NumCounters];
	}

	return 0;
}

static HSAKMT_STATUS update_block_slots(enum perf_trace_action action,
					uint32_t block_id, uint32_t num_slots)
{
	uint32_t *slots_left;
	HSAKMT_STATUS ret = HSAKMT_STATUS_SUCCESS;

	if (shmem_fd <= 0)
		return HSAKMT_STATUS_UNAVAILABLE;
	if (sem == SEM_FAILED)
		return HSAKMT_STATUS_UNAVAILABLE;

	sem_wait(sem);

	if (block_id == PERFCOUNTER_BLOCKID__IOMMUV2)
		slots_left = &shared_table->iommu_slots_left;
	else {
		ret = HSAKMT_STATUS_UNAVAILABLE;
		goto out;
	}

	switch (action) {
	case PERF_TRACE_ACTION__ACQUIRE:
		if (*slots_left >= num_slots)
			*slots_left -= num_slots;
		else
			ret = HSAKMT_STATUS_UNAVAILABLE;
		break;
	case PERF_TRACE_ACTION__RELEASE:
		if ((*slots_left + num_slots) <=
				pmc_table_get_max_concurrent(block_id))
			*slots_left += num_slots;
		else
			ret = HSAKMT_STATUS_ERROR;
		break;
	default:
		ret = HSAKMT_STATUS_INVALID_PARAMETER;
		break;
	}

out:
	sem_post(sem);

	return ret;
}

static unsigned int get_perf_event_type(enum perf_block_id block_id)
{
	FILE *file = NULL;
	unsigned int type = 0;

	if (block_id == PERFCOUNTER_BLOCKID__IOMMUV2) {
		/* Starting from kernel 4.12, amd_iommu_0 is used */
		file = fopen("/sys/bus/event_source/devices/amd_iommu_0/type",
			 "r");
		if (!file)
			file = fopen(/* kernel 4.11 and older */
				"/sys/bus/event_source/devices/amd_iommu/type",
				"r");
	}

	if (!file)
		return 0;

	if (fscanf(file, "%d", &type) != 1)
		type = 0;
	fclose(file);

	return type;
}

/* close_perf_event_fd - Close all FDs opened for this block.
 * When RT acquires the trace access, RT has no ideas about each
 * individual FD opened for this block. We should treat the whole
 * block as one and close all of them.
 */
static void close_perf_event_fd(struct perf_trace_block *block)
{
	uint32_t i;

	if (!block || !block->perf_event_fd)
		return;

	for (i = 0; i < block->num_counters; i++)
		if (block->perf_event_fd[i] > 0) {
			close(block->perf_event_fd[i]);
			block->perf_event_fd[i] = 0;
		}
}

/* open_perf_event_fd - Open FDs required for this block.
 * If one of them fails, we should close all FDs that have been
 * opened because RT has no ideas about those FDs successfully
 * opened and it won't send anything to close them.
 */
static HSAKMT_STATUS open_perf_event_fd(struct perf_trace_block *block)
{
	struct perf_event_attr attr;
	uint32_t i;
	HSAKMT_STATUS ret = HSAKMT_STATUS_SUCCESS;

	if (!block || !block->perf_event_fd)
		return HSAKMT_STATUS_INVALID_HANDLE;

	if (getuid()) {
		pr_err("Must be root to open perf_event.\n");
		return HSAKMT_STATUS_ERROR;
	}

	memset(&attr, 0, sizeof(struct perf_event_attr));
	attr.type = get_perf_event_type(block->block_id);
	if (!attr.type)
		return HSAKMT_STATUS_ERROR;

	for (i = 0; i < block->num_counters; i++) {
		attr.size = sizeof(struct perf_event_attr);
		attr.config = block->counter_id[i];
		attr.read_format = PERF_FORMAT_TOTAL_TIME_ENABLED |
					PERF_FORMAT_TOTAL_TIME_RUNNING;
		attr.disabled = 1;
		attr.inherit = 1;

		/* We are profiling system wide, not per cpu, so no threads,
		 * no groups -> pid=-1 and group_fd=-1. cpu = 0
		 * flags=PERF_FLAG_FD_NO_GROUP
		 */
		block->perf_event_fd[i] = syscall(__NR_perf_event_open, &attr,
					-1, 0, -1, PERF_FLAG_FD_NO_GROUP);

		if (block->perf_event_fd[i] < 0) {
			ret = HSAKMT_STATUS_ERROR;
			close_perf_event_fd(block);
			break;
		}
	}

	return ret;
}

static HSAKMT_STATUS perf_trace_ioctl(struct perf_trace_block *block,
				      uint32_t cmd)
{
	uint32_t i;

	for (i = 0; i < block->num_counters; i++) {
		if (block->perf_event_fd[i] < 0)
			return HSAKMT_STATUS_UNAVAILABLE;
		if (ioctl(block->perf_event_fd[i], cmd, NULL))
			return HSAKMT_STATUS_ERROR;
	}

	return HSAKMT_STATUS_SUCCESS;
}

static HSAKMT_STATUS query_trace(int fd, uint64_t *buf)
{
	struct perf_counts_values content;

	if (fd < 0)
		return HSAKMT_STATUS_ERROR;
	if (readn(fd, &content, sizeof(content)) != sizeof(content))
		return HSAKMT_STATUS_ERROR;

	*buf = content.val;
	return HSAKMT_STATUS_SUCCESS;
}

HSAKMT_STATUS HSAKMTAPI hsaKmtPmcGetCounterProperties(HSAuint32 NodeId,
						      HsaCounterProperties **CounterProperties)
{
	HSAKMT_STATUS rc = HSAKMT_STATUS_SUCCESS;
	uint32_t gpu_id, i, block_id;
	uint32_t counter_props_size = 0;
	uint32_t total_counters = 0;
	uint32_t total_concurrent = 0;
	struct perf_counter_block block = {0};
	uint32_t total_blocks = 0;
	HsaCounterBlockProperties *block_prop;

	if (!counter_props)
		return HSAKMT_STATUS_NO_MEMORY;

	if (!CounterProperties)
		return HSAKMT_STATUS_INVALID_PARAMETER;

	if (validate_nodeid(NodeId, &gpu_id) != HSAKMT_STATUS_SUCCESS)
		return HSAKMT_STATUS_INVALID_NODE_UNIT;

	if (counter_props[NodeId]) {
		*CounterProperties = counter_props[NodeId];
		return HSAKMT_STATUS_SUCCESS;
	}

	for (i = 0; i < PERFCOUNTER_BLOCKID__MAX; i++) {
		rc = get_block_properties(NodeId, i, &block);
		if (rc != HSAKMT_STATUS_SUCCESS)
			return rc;
		total_concurrent += block.num_of_slots;
		total_counters += block.num_of_counters;
		/* If num_of_slots=0, this block doesn't exist */
		if (block.num_of_slots)
			total_blocks++;
	}

	counter_props_size = sizeof(HsaCounterProperties) +
			sizeof(HsaCounterBlockProperties) * (total_blocks - 1) +
			sizeof(HsaCounter) * (total_counters - total_blocks);

	counter_props[NodeId] = malloc(counter_props_size);
	if (!counter_props[NodeId])
		return HSAKMT_STATUS_NO_MEMORY;

	counter_props[NodeId]->NumBlocks = total_blocks;
	counter_props[NodeId]->NumConcurrent = total_concurrent;

	block_prop = &counter_props[NodeId]->Blocks[0];
	for (block_id = 0; block_id < PERFCOUNTER_BLOCKID__MAX; block_id++) {
		rc = get_block_properties(NodeId, block_id, &block);
		if (rc != HSAKMT_STATUS_SUCCESS) {
			free(counter_props[NodeId]);
			counter_props[NodeId] = NULL;
			return rc;
		}

		if (!block.num_of_slots) /* not a valid block */
			continue;

		blockid2uuid(block_id, &block_prop->BlockId);
		block_prop->NumCounters = block.num_of_counters;
		block_prop->NumConcurrent = block.num_of_slots;
		for (i = 0; i < block.num_of_counters; i++) {
			block_prop->Counters[i].BlockIndex = block_id;
			block_prop->Counters[i].CounterId = block.counter_ids[i];
			block_prop->Counters[i].CounterSizeInBits = block.counter_size_in_bits;
			block_prop->Counters[i].CounterMask = block.counter_mask;
			block_prop->Counters[i].Flags.ui32.Global = 1;
			if (block_id == PERFCOUNTER_BLOCKID__IOMMUV2)
				block_prop->Counters[i].Type = HSA_PROFILE_TYPE_PRIVILEGED_IMMEDIATE;
			else
				block_prop->Counters[i].Type = HSA_PROFILE_TYPE_NONPRIV_IMMEDIATE;
		}

		block_prop = (HsaCounterBlockProperties *)&block_prop->Counters[block_prop->NumCounters];
	}

	*CounterProperties = counter_props[NodeId];

	return HSAKMT_STATUS_SUCCESS;
}

/* Registers a set of (HW) counters to be used for tracing/profiling */
HSAKMT_STATUS HSAKMTAPI hsaKmtPmcRegisterTrace(HSAuint32 NodeId,
					       HSAuint32 NumberOfCounters,
					       HsaCounter *Counters,
					       HsaPmcTraceRoot *TraceRoot)
{
	uint32_t gpu_id, i, j;
	uint64_t min_buf_size = 0;
	struct perf_trace *trace = NULL;
	uint32_t concurrent_limit;
	const uint32_t MAX_COUNTERS = 512;
	uint64_t counter_id[PERFCOUNTER_BLOCKID__MAX][MAX_COUNTERS];
	uint32_t num_counters[PERFCOUNTER_BLOCKID__MAX] = {0};
	uint32_t block, num_blocks = 0, total_counters = 0;
	uint64_t *counter_id_ptr;
	int *fd_ptr;

	pr_debug("[%s] Number of counters %d\n", __func__, NumberOfCounters);

	if (!counter_props)
		return HSAKMT_STATUS_NO_MEMORY;

	if (!Counters || !TraceRoot || NumberOfCounters == 0)
		return HSAKMT_STATUS_INVALID_PARAMETER;

	if (validate_nodeid(NodeId, &gpu_id) != HSAKMT_STATUS_SUCCESS)
		return HSAKMT_STATUS_INVALID_NODE_UNIT;

	if (NumberOfCounters > MAX_COUNTERS) {
		pr_err("MAX_COUNTERS is too small for %d.\n",
			NumberOfCounters);
		return HSAKMT_STATUS_NO_MEMORY;
	}

	/* Calculating the minimum buffer size */
	for (i = 0; i < NumberOfCounters; i++) {
		if (Counters[i].BlockIndex >= PERFCOUNTER_BLOCKID__MAX)
			return HSAKMT_STATUS_INVALID_PARAMETER;
		/* Only privileged counters need to register */
		if (Counters[i].Type > HSA_PROFILE_TYPE_PRIVILEGED_STREAMING)
			continue;
		min_buf_size += Counters[i].CounterSizeInBits/BITS_PER_BYTE;
		/* j: the first blank entry in the block to record counter_id */
		j = num_counters[Counters[i].BlockIndex];
		counter_id[Counters[i].BlockIndex][j] = Counters[i].CounterId;
		num_counters[Counters[i].BlockIndex]++;
		total_counters++;
	}

	/* Verify that the number of counters per block is not larger than the
	 * number of slots.
	 */
	for (i = 0; i < PERFCOUNTER_BLOCKID__MAX; i++) {
		if (!num_counters[i])
			continue;
		concurrent_limit = get_block_concurrent_limit(NodeId, i);
		if (!concurrent_limit) {
			pr_err("Invalid block ID: %d\n", i);
			return HSAKMT_STATUS_INVALID_PARAMETER;
		}
		if (num_counters[i] > concurrent_limit) {
			pr_err("Counters exceed the limit.\n");
			return HSAKMT_STATUS_INVALID_PARAMETER;
		}
		num_blocks++;
	}

	if (!num_blocks)
		return HSAKMT_STATUS_INVALID_PARAMETER;

	/* Now we have sorted blocks/counters information in
	 * num_counters[block_id] and counter_id[block_id][]. Allocate trace
	 * and record the information.
	 */
	trace = (struct perf_trace *)calloc(sizeof(struct perf_trace)
			+ sizeof(struct perf_trace_block) * num_blocks
			+ sizeof(uint64_t) * total_counters
			+ sizeof(int) * total_counters,
			1);
	if (!trace)
		return HSAKMT_STATUS_NO_MEMORY;

	/* Allocated area is partitioned as:
	 * +---------------------------------+ trace
	 * |    perf_trace                   |
	 * |---------------------------------| trace->blocks[0]
	 * | perf_trace_block 0              |
	 * | ....                            |
	 * | perf_trace_block N-1            | trace->blocks[N-1]
	 * |---------------------------------| <-- counter_id_ptr starts here
	 * | block 0's counter IDs(uint64_t) |
	 * | ......                          |
	 * | block N-1's counter IDs         |
	 * |---------------------------------| <-- perf_event_fd starts here
	 * | block 0's perf_event_fds(int)   |
	 * | ......                          |
	 * | block N-1's perf_event_fds      |
	 * +---------------------------------+
	 */
	block = 0;
	counter_id_ptr = (uint64_t *)((char *)
			trace + sizeof(struct perf_trace)
			+ sizeof(struct perf_trace_block) * num_blocks);
	fd_ptr = (int *)(counter_id_ptr + total_counters);
	/* Fill in each block's information to the TraceId */
	for (i = 0; i < PERFCOUNTER_BLOCKID__MAX; i++) {
		if (!num_counters[i]) /* not a block to trace */
			continue;
		/* Following perf_trace + perf_trace_block x N are those
		 * counter_id arrays. Assign the counter_id array belonging to
		 * this block.
		 */
		trace->blocks[block].counter_id = counter_id_ptr;
		/* Fill in counter IDs to the counter_id array. */
		for (j = 0; j < num_counters[i]; j++)
			trace->blocks[block].counter_id[j] = counter_id[i][j];
		trace->blocks[block].perf_event_fd = fd_ptr;
		/* how many counters to trace */
		trace->blocks[block].num_counters = num_counters[i];
		/* block index in "enum perf_block_id" */
		trace->blocks[block].block_id = i;
		block++; /* move to next */
		counter_id_ptr += num_counters[i];
		fd_ptr += num_counters[i];
	}

	trace->magic4cc = HSA_PERF_MAGIC4CC;
	trace->gpu_id = gpu_id;
	trace->state = PERF_TRACE_STATE__STOPPED;
	trace->num_blocks = num_blocks;

	TraceRoot->NumberOfPasses = 1;
	TraceRoot->TraceBufferMinSizeBytes = PAGE_ALIGN_UP(min_buf_size);
	TraceRoot->TraceId = PORT_VPTR_TO_UINT64(trace);

	return HSAKMT_STATUS_SUCCESS;
}

/* Unregisters a set of (HW) counters used for tracing/profiling */

HSAKMT_STATUS HSAKMTAPI hsaKmtPmcUnregisterTrace(HSAuint32 NodeId,
						 HSATraceId TraceId)
{
	uint32_t gpu_id;
	struct perf_trace *trace;

	pr_debug("[%s] Trace ID 0x%lx\n", __func__, TraceId);

	if (TraceId == 0)
		return HSAKMT_STATUS_INVALID_PARAMETER;

	if (validate_nodeid(NodeId, &gpu_id) != HSAKMT_STATUS_SUCCESS)
		return HSAKMT_STATUS_INVALID_NODE_UNIT;

	trace = (struct perf_trace *)PORT_UINT64_TO_VPTR(TraceId);

	if (trace->magic4cc != HSA_PERF_MAGIC4CC)
		return HSAKMT_STATUS_INVALID_HANDLE;

	if (trace->gpu_id != gpu_id)
		return HSAKMT_STATUS_INVALID_NODE_UNIT;

	/* If the trace is in the running state, stop it */
	if (trace->state == PERF_TRACE_STATE__STARTED) {
		HSAKMT_STATUS status = hsaKmtPmcStopTrace(TraceId);

		if (status != HSAKMT_STATUS_SUCCESS)
			return status;
	}

	free(trace);

	return HSAKMT_STATUS_SUCCESS;
}

HSAKMT_STATUS HSAKMTAPI hsaKmtPmcAcquireTraceAccess(HSAuint32 NodeId,
						    HSATraceId TraceId)
{
	struct perf_trace *trace;
	HSAKMT_STATUS ret = HSAKMT_STATUS_SUCCESS;
	uint32_t gpu_id, i;
	int j;

	pr_debug("[%s] Trace ID 0x%lx\n", __func__, TraceId);

	if (TraceId == 0)
		return HSAKMT_STATUS_INVALID_PARAMETER;

	trace = (struct perf_trace *)PORT_UINT64_TO_VPTR(TraceId);

	if (trace->magic4cc != HSA_PERF_MAGIC4CC)
		return HSAKMT_STATUS_INVALID_HANDLE;

	if (validate_nodeid(NodeId, &gpu_id) != HSAKMT_STATUS_SUCCESS)
		return HSAKMT_STATUS_INVALID_NODE_UNIT;

	for (i = 0; i < trace->num_blocks; i++) {
		ret = update_block_slots(PERF_TRACE_ACTION__ACQUIRE,
					trace->blocks[i].block_id,
					trace->blocks[i].num_counters);
		if (ret != HSAKMT_STATUS_SUCCESS)
			goto out;
		ret = open_perf_event_fd(&trace->blocks[i]);
		if (ret != HSAKMT_STATUS_SUCCESS) {
			i++; /* to release slots just reserved */
			goto out;
		}
	}

out:
	if (ret != HSAKMT_STATUS_SUCCESS) {
		for (j = i-1; j >= 0; j--) {
			update_block_slots(PERF_TRACE_ACTION__RELEASE,
					trace->blocks[j].block_id,
					trace->blocks[j].num_counters);
			close_perf_event_fd(&trace->blocks[j]);
		}
	}

	return ret;
}

HSAKMT_STATUS HSAKMTAPI hsaKmtPmcReleaseTraceAccess(HSAuint32 NodeId,
						    HSATraceId TraceId)
{
	struct perf_trace *trace;
	uint32_t i;

	pr_debug("[%s] Trace ID 0x%lx\n", __func__, TraceId);

	if (TraceId == 0)
		return HSAKMT_STATUS_INVALID_PARAMETER;

	trace = (struct perf_trace *)PORT_UINT64_TO_VPTR(TraceId);

	if (trace->magic4cc != HSA_PERF_MAGIC4CC)
		return HSAKMT_STATUS_INVALID_HANDLE;

	for (i = 0; i < trace->num_blocks; i++) {
		update_block_slots(PERF_TRACE_ACTION__RELEASE,
				trace->blocks[i].block_id,
				trace->blocks[i].num_counters);
		close_perf_event_fd(&trace->blocks[i]);
	}

	return HSAKMT_STATUS_SUCCESS;
}


/* Starts tracing operation on a previously established set of performance counters */
HSAKMT_STATUS HSAKMTAPI hsaKmtPmcStartTrace(HSATraceId TraceId,
					    void *TraceBuffer,
					    HSAuint64 TraceBufferSizeBytes)
{
	struct perf_trace *trace =
			(struct perf_trace *)PORT_UINT64_TO_VPTR(TraceId);
	uint32_t i;
	int32_t j;
	HSAKMT_STATUS ret = HSAKMT_STATUS_SUCCESS;

	pr_debug("[%s] Trace ID 0x%lx\n", __func__, TraceId);

	if (TraceId == 0 || !TraceBuffer || TraceBufferSizeBytes == 0)
		return HSAKMT_STATUS_INVALID_PARAMETER;

	if (trace->magic4cc != HSA_PERF_MAGIC4CC)
		return HSAKMT_STATUS_INVALID_HANDLE;

	for (i = 0; i < trace->num_blocks; i++) {
		ret = perf_trace_ioctl(&trace->blocks[i],
					PERF_EVENT_IOC_ENABLE);
		if (ret != HSAKMT_STATUS_SUCCESS)
			break;
	}
	if (ret != HSAKMT_STATUS_SUCCESS) {
		/* Disable enabled blocks before returning the failure. */
		j = (int32_t)i;
		while (--j >= 0)
			perf_trace_ioctl(&trace->blocks[j],
					PERF_EVENT_IOC_DISABLE);
		return ret;
	}

	trace->state = PERF_TRACE_STATE__STARTED;
	trace->buf = TraceBuffer;
	trace->buf_size = TraceBufferSizeBytes;

	return HSAKMT_STATUS_SUCCESS;
}


/*Forces an update of all the counters that a previously started trace operation has registered */

HSAKMT_STATUS HSAKMTAPI hsaKmtPmcQueryTrace(HSATraceId TraceId)
{
	struct perf_trace *trace =
			(struct perf_trace *)PORT_UINT64_TO_VPTR(TraceId);
	uint32_t i, j;
	HSAKMT_STATUS ret = HSAKMT_STATUS_SUCCESS;
	uint64_t *buf;
	uint64_t buf_filled = 0;

	if (TraceId == 0)
		return HSAKMT_STATUS_INVALID_PARAMETER;

	if (trace->magic4cc != HSA_PERF_MAGIC4CC)
		return HSAKMT_STATUS_INVALID_HANDLE;

	buf = (uint64_t *)trace->buf;
	pr_debug("[%s] Trace buffer(%p): ", __func__, buf);
	for (i = 0; i < trace->num_blocks; i++)
		for (j = 0; j < trace->blocks[i].num_counters; j++) {
			buf_filled += sizeof(uint64_t);
			if (buf_filled > trace->buf_size)
				return HSAKMT_STATUS_NO_MEMORY;
			ret = query_trace(trace->blocks[i].perf_event_fd[j],
					buf);
			if (ret != HSAKMT_STATUS_SUCCESS)
				return ret;
			pr_debug("%lu_", *buf);
			buf++;
		}
	pr_debug("\n");

	return HSAKMT_STATUS_SUCCESS;
}


/* Stops tracing operation on a previously established set of performance counters */
HSAKMT_STATUS HSAKMTAPI hsaKmtPmcStopTrace(HSATraceId TraceId)
{
	struct perf_trace *trace =
			(struct perf_trace *)PORT_UINT64_TO_VPTR(TraceId);
	uint32_t i;
	HSAKMT_STATUS ret = HSAKMT_STATUS_SUCCESS;

	pr_debug("[%s] Trace ID 0x%lx\n", __func__, TraceId);

	if (TraceId == 0)
		return HSAKMT_STATUS_INVALID_PARAMETER;

	if (trace->magic4cc != HSA_PERF_MAGIC4CC)
		return HSAKMT_STATUS_INVALID_HANDLE;

	for (i = 0; i < trace->num_blocks; i++) {
		ret = perf_trace_ioctl(&trace->blocks[i],
					PERF_EVENT_IOC_DISABLE);
		if (ret != HSAKMT_STATUS_SUCCESS)
			return ret;
	}

	trace->state = PERF_TRACE_STATE__STOPPED;

	return ret;
}