AI/rocm-systems
2
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

SQTT memory checks

Browse Source 
[ROCm/rocprofiler commit: ee5bc95019]
This commit is contained in:
Evgeny
2018-04-04 12:14:29 -05:00
parent 3c98242e09
commit f1a5ba7386
4 changed files with 60 additions and 44 deletions
Download Patch File Download Diff File Expand all files Collapse all files
projects/rocprofiler/src/core/context.h
+24 -21
View File
@@ -322,6 +322,7 @@ class Context {
}
struct callback_data_t {
const profile_t* profile;
info_vector_t* info_vector;
size_t index;
char* ptr;
@@ -340,7 +341,7 @@ class Context {
if (!complete) WARN_LOGGING("timeout");
}
for (rocprofiler_feature_t* rinfo : *(tuple.info_vector)) rinfo->data.kind = ROCPROFILER_DATA_KIND_UNINIT;
callback_data_t callback_data{tuple.info_vector, tuple.info_vector->size(), NULL};
callback_data_t callback_data{tuple.profile, tuple.info_vector, tuple.info_vector->size(), NULL};
const hsa_status_t status =
api_->hsa_ven_amd_aqlprofile_iterate_data(tuple.profile, DataCallback, &callback_data);
if (status != HSA_STATUS_SUCCESS) AQL_EXC_RAISING(status, "context iterate data failed");
@@ -404,6 +405,7 @@ class Context {
hsa_ven_amd_aqlprofile_info_data_t* ainfo_data, void* data) {
hsa_status_t status = HSA_STATUS_SUCCESS;
callback_data_t* callback_data = reinterpret_cast<callback_data_t*>(data);
const profile_t* profile = callback_data->profile;
info_vector_t& info_vector = *(callback_data->info_vector);
uint32_t index = callback_data->index;
const uint32_t sample_id = ainfo_data->sample_id;
@@ -425,39 +427,40 @@ class Context {
} else if (ainfo_type == HSA_VEN_AMD_AQLPROFILE_INFO_SQTT_DATA) {
if (rinfo->data.result_bytes.copy) {
if (sample_id == 0) {
if (rinfo->data.result_bytes.size == 0) {
const uint32_t output_buffer_size = SqttProfile::GetSize();
const uint32_t output_buffer_size = profile->output_buffer.size;
const uint32_t output_buffer_size64 = output_buffer_size / sizeof(uint64_t);
rinfo->data.result_bytes.ptr = calloc(output_buffer_size64, sizeof(uint64_t));
void* ptr = calloc(output_buffer_size64, sizeof(uint64_t));
rinfo->data.result_bytes.size = output_buffer_size;
} else if (rinfo->data.result_bytes.size != SqttProfile::GetSize()) {
EXC_RAISING(HSA_STATUS_ERROR, "result bytes copy mode, data array size mismatch(" << rinfo->data.result_bytes.size << ")");
}
rinfo->data.result_bytes.ptr = ptr;
callback_data->ptr = reinterpret_cast<char*>(ptr);
}
char* result_bytes_ptr = reinterpret_cast<char*>(rinfo->data.result_bytes.ptr);
const char* end = result_bytes_ptr + rinfo->data.result_bytes.size;
const char* src = reinterpret_cast<char*>(ainfo_data->sqtt_data.ptr);
const uint32_t size = ainfo_data->sqtt_data.size;
char* ptr = (sample_id == 0) ? result_bytes_ptr : callback_data->ptr;
uint64_t* header = reinterpret_cast<uint64_t*>(ptr);
uint32_t size = ainfo_data->sqtt_data.size;
char* ptr = callback_data->ptr;
uint32_t* header = reinterpret_cast<uint32_t*>(ptr);
char* dest = ptr + sizeof(*header);
if ((dest + size) < end) {
hsa_status_t status = hsa_memory_copy(dest, src, size);
if (status == HSA_STATUS_SUCCESS) {
*header = size;
callback_data->ptr = dest + align_size(size, sizeof(uint64_t));
rinfo->data.result_bytes.instance_count = sample_id + 1;
rinfo->data.kind = ROCPROFILER_DATA_KIND_BYTES;
}
} else {
EXC_RAISING(HSA_STATUS_ERROR, "result bytes copy mode, out of boundary, size = " << rinfo->data.result_bytes.size);
if ((dest + size) >= end) {
if (dest < end) size = end - dest;
else EXC_RAISING(HSA_STATUS_ERROR, "SQTT data out of output buffer");
}
hsa_status_t status = hsa_memory_copy(dest, src, size);
if (status == HSA_STATUS_SUCCESS) {
*header = size;
callback_data->ptr = dest + align_size(size, sizeof(uint32_t));
rinfo->data.result_bytes.instance_count = sample_id + 1;
rinfo->data.kind = ROCPROFILER_DATA_KIND_BYTES;
}
} else {
if (sample_id == 0) {
rinfo->data.result_bytes.ptr = ainfo_data->sqtt_data.ptr;
rinfo->data.result_bytes.ptr = profile->output_buffer.ptr;
rinfo->data.result_bytes.size = profile->output_buffer.size;
rinfo->data.result_bytes.instance_count = UINT32_MAX;
}
rinfo->data.result_bytes.instance_count += 1;
rinfo->data.kind = ROCPROFILER_DATA_KIND_BYTES;
}
projects/rocprofiler/src/core/intercept_queue.cpp
+1 -1
View File
@@ -14,5 +14,5 @@ InterceptQueue::obj_map_t* InterceptQueue::obj_map_ = NULL;
const char* InterceptQueue::kernel_none_ = "";
uint64_t InterceptQueue::timeout_ = UINT64_MAX;
Tracker* InterceptQueue::tracker_ = NULL;
bool InterceptQueue::tracker_on_ = true;
bool InterceptQueue::tracker_on_ = false;
} // namespace rocprofiler
projects/rocprofiler/src/core/profile.h
+4 -3
View File
@@ -113,7 +113,8 @@ class Profile {
// Check the profile buffer sizes
status = api->hsa_ven_amd_aqlprofile_start(&profile_, NULL);
if (status != HSA_STATUS_SUCCESS) AQL_EXC_RAISING(status, "aqlprofile_start(NULL)");
Allocate(rsrc);
status = Allocate(rsrc);
if (status != HSA_STATUS_SUCCESS) AQL_EXC_RAISING(status, "Allocate()");
// Generate start/stop profiling packets
status = api->hsa_ven_amd_aqlprofile_start(&profile_, &start);
if (status != HSA_STATUS_SUCCESS) AQL_EXC_RAISING(status, "aqlprofile_start");
@@ -211,7 +212,7 @@ class PmcProfile : public Profile {
class SqttProfile : public Profile {
public:
static inline void SetSize(const uint32_t& size) { output_buffer_size_ = size; }
static inline uint32_t GetSize() { return output_buffer_size_; }
// static inline uint32_t GetSize() { return output_buffer_size_; }
SqttProfile(const util::AgentInfo* agent_info) : Profile(agent_info) {
profile_.type = HSA_VEN_AMD_AQLPROFILE_EVENT_TYPE_SQTT;
@@ -225,9 +226,9 @@ class SqttProfile : public Profile {
}
hsa_status_t Allocate(util::HsaRsrcFactory* rsrc) {
profile_.output_buffer.size = output_buffer_size_;
profile_.command_buffer.ptr =
rsrc->AllocateSysMemory(agent_info_, profile_.command_buffer.size);
profile_.output_buffer.size = output_buffer_size_;
profile_.output_buffer.ptr =
rsrc->AllocateLocalMemory(agent_info_, profile_.output_buffer.size);
return (profile_.command_buffer.ptr && profile_.output_buffer.ptr) ? HSA_STATUS_SUCCESS
projects/rocprofiler/test/ctrl/tool.cpp
+31 -19
View File
@@ -100,8 +100,10 @@ static inline uint32_t GetPid() { return syscall(__NR_getpid); }
static inline uint32_t GetTid() { return syscall(__NR_gettid); }
// Error handler
void fatal(const char* msg) {
fprintf(stderr, "%s\n", msg);
void fatal(const std::string msg) {
fflush(stdout);
fprintf(stderr, "%s\n\n", msg.c_str());
fflush(stderr);
abort();
}
@@ -206,7 +208,7 @@ context_entry_t* alloc_context_entry() {
abort();
}
const uint32_t index = next_context_count();
const uint32_t index = next_context_count() - 1;
auto ret = context_array->insert({index, context_entry_t{}});
if (ret.second == false) {
fprintf(stderr, "context_array corruption, index repeated %u\n", index);
@@ -277,7 +279,6 @@ hsa_status_t trace_data_cb(hsa_ven_amd_aqlprofile_info_type_t info_type,
if (info_type == HSA_VEN_AMD_AQLPROFILE_INFO_SQTT_DATA) {
fprintf(arg->file, " SE(%u) size(%u)\n", info_data->sample_id, info_data->sqtt_data.size);
dump_sqtt_trace(arg->label, info_data->sample_id, info_data->sqtt_data.ptr, info_data->sqtt_data.size);
} else
status = HSA_STATUS_ERROR;
return status;
@@ -305,17 +306,18 @@ void output_results(FILE* file, const rocprofiler_feature_t* features, const uns
uint64_t size = 0;
const char* ptr = reinterpret_cast<const char*>(p->data.result_bytes.ptr);
const char* end = reinterpret_cast<const char*>(ptr + p->data.result_bytes.size);
for (unsigned i = 0; i < p->data.result_bytes.instance_count; ++i) {
const uint32_t chunk_size = *reinterpret_cast<const uint64_t*>(ptr);
const char* chunk_data = ptr + sizeof(uint64_t);
dump_sqtt_trace(label, i, chunk_data, chunk_size);
const uint32_t chunk_size = *reinterpret_cast<const uint32_t*>(ptr);
const char* chunk_data = ptr + sizeof(uint32_t);
if (chunk_data >= end) fatal("SQTT data is out of the result buffer size");
const uint32_t off = align_size(chunk_size, sizeof(uint64_t));
dump_sqtt_trace(label, i, chunk_data, chunk_size);
const uint32_t off = align_size(chunk_size, sizeof(uint32_t));
ptr = chunk_data + off;
size += chunk_size;
if (chunk_data >= end) fatal("SQTT data ptr is out of the result buffer size");
}
fprintf(file, "size(%lu)\n", size);
if (size > p->data.result_bytes.size) fatal("SQTT data size is out of the result buffer size");
free(p->data.result_bytes.ptr);
const_cast<rocprofiler_feature_t*>(p)->data.result_bytes.size = 0;
} else {
@@ -646,7 +648,7 @@ extern "C" PUBLIC_API void OnLoadTool()
}
std::map<std::string, hsa_ven_amd_aqlprofile_parameter_name_t> parameters_dict;
parameters_dict["COMPUTE_UNIT_TARGET"] =
parameters_dict["TARGET_CU"] =
HSA_VEN_AMD_AQLPROFILE_PARAMETER_NAME_COMPUTE_UNIT_TARGET;
parameters_dict["VM_ID_MASK"] =
HSA_VEN_AMD_AQLPROFILE_PARAMETER_NAME_VM_ID_MASK;
@@ -760,13 +762,21 @@ extern "C" PUBLIC_API void OnLoadTool()
unsigned index = metrics_vec.size();
for (auto* entry : traces_list) {
auto params_list = xml->GetNodes("top.trace.parameters");
if (params_list.size() != 1) {
fprintf(stderr, "ROCProfiler: Single input 'parameters' section is supported\n");
abort();
if (params_list.size() > 1) {
fatal("ROCProfiler: Single input 'parameters' section is supported");
}
const std::string& name = entry->opts["name"];
const bool to_copy_data = (entry->opts["copy"] == "true");
printf(" %s (\n", name.c_str());
std::string name = "";
bool to_copy_data = false;
for (const auto& opt : entry->opts) {
// const std::string& name = entry->opts["name"];
// const bool to_copy_data = (entry->opts["copy"] == "true");
if (opt.first == "name") name = opt.second;
else if (opt.first == "copy") to_copy_data = (opt.second == "true");
else fatal("ROCProfiler: Bad trace property '" + opt.first + "'");
}
if (name == "") fatal("ROCProfiler: Bad trace properties, name is not specified");
printf(" %s (", name.c_str());
features[index] = {};
features[index].kind = ROCPROFILER_FEATURE_KIND_TRACE;
features[index].name = strdup(name.c_str());
@@ -783,7 +793,7 @@ extern "C" PUBLIC_API void OnLoadTool()
abort();
}
const uint32_t value = strtol(v.second.c_str(), NULL, 0);
printf(" %s = 0x%x\n", parameter_name.c_str(), value);
printf("\n %s = 0x%x", parameter_name.c_str(), value);
parameters[p_index] = {};
parameters[p_index].parameter_name = parameters_dict[parameter_name];
parameters[p_index].value = value;
@@ -793,7 +803,9 @@ extern "C" PUBLIC_API void OnLoadTool()
features[index].parameters = parameters;
features[index].parameter_count = parameter_count;
}
printf(" )\n");
if (params_list.empty() == false) printf("\n ");
printf(")\n");
fflush(stdout);
++index;
}
fflush(stdout);