xrt-smi rearchitecture Patch 4 (#8714)

* xrt-smi rearchitecture Signed-off-by: Akshay Tondak <[email protected]> * Fix for windows build failure Signed-off-by: Akshay Tondak <[email protected]> * End of file new line handling Signed-off-by: Akshay Tondak <[email protected]> * Move all configuration generation to a common file Signed-off-by: Akshay Tondak <[email protected]> * Adding missing files Signed-off-by: Akshay Tondak <[email protected]> * Windows build failure fix Signed-off-by: Akshay Tondak <[email protected]> * Windows build failure fix and converting to snake case Signed-off-by: Akshay Tondak <[email protected]> * Adding missing XRT_CORE_COMMON_SOURCE define Signed-off-by: Akshay Tondak <[email protected]> * Some review comments handled and pipeline re-trigger Signed-off-by: Akshay Tondak <[email protected]> * Moving re-usable code to base class from which each shim can derive Signed-off-by: Akshay Tondak <[email protected]> * Fixing edge build error Signed-off-by: Akshay Tondak <[email protected]> * Removing some extra includes Signed-off-by: Akshay Tondak <[email protected]> * Formatting review comment changes Signed-off-by: Akshay Tondak <[email protected]> * Exporting required APIs for MCDM Signed-off-by: Akshay Tondak <[email protected]> * Adding std::move for efficient object creation Signed-off-by: Akshay Tondak <[email protected]> --------- Signed-off-by: Akshay Tondak <[email protected]>
Xilinx · Jan 28, 2025 · 275bbcb · 275bbcb
1 parent 337b125
commit 275bbcb
Show file tree

Hide file tree

Showing 24 changed files with 592 additions and 851 deletions.
diff --git a/src/runtime_src/core/common/CMakeLists.txt b/src/runtime_src/core/common/CMakeLists.txt
@@ -43,6 +43,7 @@ add_library(core_common_library_objects OBJECT
   sysinfo.cpp
   xclbin_parser.cpp
   xclbin_swemu.cpp
+  smi.cpp
   )
 
 target_include_directories(core_common_library_objects

diff --git a/src/runtime_src/core/common/smi.cpp b/src/runtime_src/core/common/smi.cpp
@@ -0,0 +1,242 @@
+// SPDX-License-Identifier: Apache-2.0
+// Copyright (C) 2025 Advanced Micro Devices, Inc. All rights reserved.
+
+#define XRT_CORE_COMMON_SOURCE
+
+// Local - Include Files
+#include "smi.h"
+
+// 3rd Party Library - Include Files
+#include <boost/property_tree/json_parser.hpp>
+#include <boost/property_tree/ptree.hpp>
+
+#include <string>
+#include <vector>
+
+namespace xrt_core::smi {
+
+using boost::property_tree::ptree;
+
+ptree 
+option::
+to_ptree() const 
+{
+  boost::property_tree::ptree pt;
+  pt.put("name", name);
+  pt.put("description", description);
+  pt.put("type", type);
+  pt.put("alias", alias);
+  pt.put("default_value", default_value);
+  pt.put("value_type", value_type);
+  if (!description_array.empty()) {
+    boost::property_tree::ptree description_array_ptree;
+    for (const auto& desc : description_array) {
+      boost::property_tree::ptree desc_node;
+      desc_node.put("name", desc.name);
+      desc_node.put("description", desc.description);
+      desc_node.put("type", desc.type);
+      description_array_ptree.push_back(std::make_pair("", desc_node));
+    }
+    pt.add_child("description_array", description_array_ptree);
+  }
+  return pt;
+}
+
+const tuple_vector& 
+smi_base::
+get_validate_test_desc() const 
+{
+  static const tuple_vector validate_test_desc = {
+    {"aie-reconfig-overhead", "Run end-to-end array reconfiguration overhead through shim DMA", "hidden"},
+    {"all", "All applicable validate tests will be executed (default)", "common"},
+    {"cmd-chain-latency", "Run end-to-end latency test using command chaining", "common"},
+    {"cmd-chain-throughput", "Run end-to-end throughput test using command chaining", "common"},
+    {"df-bw", "Run bandwidth test on data fabric", "common"},
+    {"gemm", "Measure the TOPS value of GEMM operations", "common"},
+    {"latency", "Run end-to-end latency test", "common"},
+    {"quick", "Run a subset of four tests: \n1. latency \n2. throughput \n3. cmd-chain-latency \n4. cmd-chain-throughput", "common"},
+    {"spatial-sharing-overhead", "Run Spatial Sharing Overhead Test", "hidden"},
+    {"tct-all-col", "Measure average TCT processing time for all columns", "common"},
+    {"tct-one-col", "Measure average TCT processing time for one column", "common"},
+    {"temporal-sharing-overhead", "Run Temporal Sharing Overhead Test", "hidden"},
+    {"throughput", "Run end-to-end throughput test", "common"},
+    {"aux-connection", "Check if auxiliary power is connected", "common"},
+    {"dma", "Run dma test", "common"},
+    {"thostmem-bw", "Run 'bandwidth kernel' when host memory is enabled", "common"},
+    {"m2m", "Run M2M test", "common"},
+    {"mem-bw", "Run 'bandwidth kernel' and check the throughput", "common"},
+    {"p2p", "Run P2P test", "common"},
+    {"pcie-link", "Check if PCIE link is active", "common"},
+    {"sc-version","Check if SC firmware is up-to-date", "common"},
+    {"verify", "Run 'Hello World' kernel test", "common"}
+  };
+  return validate_test_desc;
+}
+
+const tuple_vector& 
+smi_base::
+get_examine_report_desc() const 
+{
+  static const tuple_vector examine_report_desc = {
+    {"aie-partitions", "AIE partition information", "common"},
+    {"host", "Host information", "common"},
+    {"platform", "Platforms flashed on the device", "common"},
+    {"telemetry", "Telemetry data for the device", "common"},
+    {"aie", "AIE metadata in xclbin", "common"},
+    {"aiemem", "AIE memory tile information", "common"},
+    {"aieshim", "AIE shim tile status", "common"},
+    {"debug-ip-status", "Status of Debug IPs present in xclbin loaded on device", "common"},
+    {"dynamic-regions", "Information about the xclbin and the compute units", "common"},
+    {"electrical", "Electrical and power sensors present on the device", "common"},
+    {"error", "Asyncronus Error present on the device", "common"},
+    {"firewall", "Firewall status", "common"},
+    {"mailbox", "Mailbox metrics of the device", "common"},
+    {"mechanical", "Mechanical sensors on and surrounding the device", "common"},
+    {"memory", "Memory information present on the device", "common"},
+    {"pcie-info", "Pcie information of the device", "common"},
+    {"qspi-status", "QSPI write protection status", "common"},
+    {"thermal", "Thermal sensors present on the device", "common"}
+  };
+  return examine_report_desc;
+}
+
+std::vector<basic_option> 
+smi_base::
+construct_run_option_description() const 
+{
+  std::vector<basic_option> run_option_descriptions;
+  for (const auto& [name, description, type] : get_validate_test_desc()) {
+    run_option_descriptions.push_back({name, description, type});
+  }
+  return run_option_descriptions;
+}
+
+std::vector<basic_option> 
+smi_base::
+construct_report_option_description() const 
+{
+  std::vector<basic_option> report_option_descriptions;
+  for (const auto& [name, description, type] : get_examine_report_desc()) {
+    report_option_descriptions.push_back({name, description, type});
+  }
+  return report_option_descriptions;
+}
+
+ptree 
+smi_base::
+construct_validate_subcommand() const 
+{
+  ptree subcommand;
+  subcommand.put("name", "validate");
+  subcommand.put("description", "Validates the given device by executing the platform's validate executable.");
+  subcommand.put("type", "common");
+
+  std::vector<option> options = {
+    {"device", "d", "The Bus:Device.Function (e.g., 0000:d8:00.0) device of interest", "common", "", "string"},
+    {"format", "f", "Report output format. Valid values are:\n"
+                    "\tJSON        - Latest JSON schema\n"
+                    "\tJSON-2020.2 - JSON 2020.2 schema", "common", "JSON", "string"},
+    {"output", "o", "Direct the output to the given file", "common", "", "string"},
+    {"help", "h", "Help to use this sub-command", "common", "", "none"},
+    {"run", "r", "Run a subset of the test suite. Valid options are:\n",  "common", "",  "array", construct_run_option_description()},
+    {"path", "p", "Path to the directory containing validate xclbins", "hidden", "", "string"},
+    {"param", "", "Extended parameter for a given test. Format: <test-name>:<key>:<value>", "hidden", "", "string"},
+    {"pmode", "", "Specify which power mode to run the benchmarks in. Note: Some tests might be unavailable for some modes", "hidden", "", "string"}
+  };
+
+  ptree options_ptree;
+  for (const auto& option : options) {
+    options_ptree.push_back(std::make_pair("", option.to_ptree()));
+  }
+
+  subcommand.add_child("options", options_ptree);
+  return subcommand;
+}
+
+ptree 
+smi_base::
+construct_examine_subcommand() const 
+{
+  ptree subcommand;
+  subcommand.put("name", "examine");
+  subcommand.put("type", "common");
+  subcommand.put("description", "This command will 'examine' the state of the system/device and will generate a report of interest in a text or JSON format.");
+
+  std::vector<option> options = {
+    {"device", "d", "The Bus:Device.Function (e.g., 0000:d8:00.0) device of interest", "common", "", "string"},
+    {"format", "f", "Report output format. Valid values are:\n"
+                    "\tJSON        - Latest JSON schema\n"
+                    "\tJSON-2020.2 - JSON 2020.2 schema", "common", "", "string"},
+    {"output", "o", "Direct the output to the given file", "common", "", "string"},
+    {"help", "h", "Help to use this sub-command", "common", "", "none"},
+    {"report", "r", "The type of report to be produced. Reports currently available are:\n", "common", "", "array", construct_report_option_description()},
+    {"element", "e", "Filters individual elements(s) from the report. Format: '/<key>/<key>/...'", "hidden", "", "array"}
+  };
+
+  ptree options_ptree;
+  for (const auto& option : options) {
+    options_ptree.push_back(std::make_pair("", option.to_ptree()));
+  }
+
+  subcommand.add_child("options", options_ptree);
+  return subcommand;
+}
+
+ptree 
+smi_base::
+construct_configure_subcommand() const 
+{
+  ptree subcommand;
+  subcommand.put("name", "configure");
+  subcommand.put("type", "common");
+  subcommand.put("description", "Device and host configuration");
+
+  std::vector<option> options = {
+    {"device", "d", "The Bus:Device.Function (e.g., 0000:d8:00.0) device of interest", "common", "", "string"},
+    {"help", "h", "Help to use this sub-command", "common", "", "none"},
+    {"daemon", "", "Update the device daemon configuration", "common", "", "none"},
+    {"purge", "", "Remove the daemon configuration file", "hidden", "", "string"},
+    {"host", "", "IP or hostname for device peer", "common", "", "string"},
+    {"security", "", "Update the security level for the device", "hidden", "", "string"},
+    {"clk_throttle", "", "Enable/disable the device clock throttling", "hidden", "", "string"},
+    {"ct_threshold_power_override", "", "Update the power threshold in watts", "hidden", "", "string"},
+    {"ct_threshold_temp_override", "", "Update the temperature threshold in celsius", "hidden", "", "string"},
+    {"ct_reset", "", "Reset all throttling options", "hidden", "", "string"},
+    {"showx", "", "Display the device configuration settings", "hidden", "", "string"}
+  };
+
+  ptree options_ptree;
+  for (const auto& option : options) {
+    options_ptree.push_back(std::make_pair("", option.to_ptree()));
+  }
+
+  subcommand.add_child("options", options_ptree);
+  return subcommand;
+}
+
+std::string 
+smi_base::
+get_smi_config() const 
+{
+  ptree config;
+  ptree subcommands;
+
+  subcommands.push_back(std::make_pair("", construct_validate_subcommand()));
+  subcommands.push_back(std::make_pair("", construct_examine_subcommand()));
+  subcommands.push_back(std::make_pair("", construct_configure_subcommand()));
+
+  config.add_child("subcommands", subcommands);
+
+  std::ostringstream oss;
+  boost::property_tree::write_json(oss, config, true); // Pretty print with true
+  return oss.str();
+}
+
+std::string
+get_smi_config()
+{
+  xrt_core::smi::smi_base instance;
+
+  return instance.get_smi_config();
+}
+} // namespace xrt_core::smi
diff --git a/src/runtime_src/core/common/smi.h b/src/runtime_src/core/common/smi.h
@@ -0,0 +1,86 @@
+// SPDX-License-Identifier: Apache-2.0
+// Copyright (C) 2025 Advanced Micro Devices, Inc. All rights reserved.
+
+#pragma once
+// Local include files
+#include "config.h"
+
+// 3rd Party Library - Include Files
+#include <boost/property_tree/ptree.hpp>
+
+#include <string>
+#include <tuple>
+#include <vector>
+
+namespace xrt_core::smi {
+
+using tuple_vector = std::vector<std::tuple<std::string, std::string, std::string>>; 
+
+struct basic_option {
+  std::string name;
+  std::string description;
+  std::string type;
+};
+
+struct option : public basic_option {
+  std::string alias;
+  std::string default_value;
+  std::string value_type;
+  std::vector<basic_option> description_array;
+
+  option(const std::string name, 
+         const std::string alias, 
+         const std::string description,
+         const std::string type, 
+         const std::string default_value, 
+         const std::string value_type, 
+         const std::vector<basic_option>& description_array = {})
+      : basic_option{std::move(name), std::move(description), std::move(type)}, 
+        alias(std::move(alias)), 
+        default_value(std::move(default_value)), 
+        value_type(std::move(value_type)), 
+        description_array(std::move(description_array)) {}
+
+  boost::property_tree::ptree to_ptree() const;
+};
+
+// Each shim's smi class derives from this class
+// and adds its custom functionalities. Currently only validate tests and examine
+// reports differ between each shim but going forward, each shim can define its 
+// custom behavior for xrt-smi as required. This also gives us the flexibility
+// to add device specific xrt-smi behavior.
+class smi_base {
+protected:
+
+  XRT_CORE_COMMON_EXPORT
+  virtual const tuple_vector& 
+  get_validate_test_desc() const;
+
+  XRT_CORE_COMMON_EXPORT
+  virtual const tuple_vector& 
+  get_examine_report_desc() const;
+
+  std::vector<basic_option> 
+  construct_run_option_description() const;
+
+  std::vector<basic_option> 
+  construct_report_option_description() const;
+
+  boost::property_tree::ptree 
+  construct_validate_subcommand() const;
+
+  boost::property_tree::ptree 
+  construct_examine_subcommand() const;
+
+  boost::property_tree::ptree 
+  construct_configure_subcommand() const;
+
+public:
+  XRT_CORE_COMMON_EXPORT
+  std::string get_smi_config() const;
+};
+
+XRT_CORE_COMMON_EXPORT
+std::string get_smi_config();
+
+} // namespace xrt_core::smi
diff --git a/src/runtime_src/core/edge/user/device_linux.cpp b/src/runtime_src/core/edge/user/device_linux.cpp
@@ -5,6 +5,7 @@
 #include "xrt.h"
 #include "zynq_dev.h"
 #include "aie_sys_parser.h"
+#include "smi.h"
 
 #include "core/common/debug_ip.h"
 #include "core/common/query_requests.h"
@@ -678,6 +679,30 @@ struct am_counter
   }
 };
 
+struct xrt_smi_config
+{
+  using result_type = std::any;
+
+  static result_type
+  get(const xrt_core::device* device, key_type key, const std::any& reqType)
+  {
+    if (key != key_type::xrt_smi_config)
+      throw xrt_core::query::no_such_key(key, "Not implemented");
+
+    std::string xrt_smi_config;
+    const auto xrt_smi_config_type = std::any_cast<xrt_core::query::xrt_smi_config::type>(reqType);
+    switch (xrt_smi_config_type) {
+    case xrt_core::query::xrt_smi_config::type::options_config:
+      xrt_smi_config = shim_edge::smi::get_smi_config();
+      break;
+    default:
+      throw xrt_core::query::no_such_key(key, "Not implemented");
+    }
+
+    return xrt_smi_config;
+  }
+};
+
 struct asm_counter
 {
   using result_type = query::asm_counter::result_type;
@@ -1058,6 +1083,7 @@ initialize_query_table()
   emplace_func4_request<query::aim_counter,             aim_counter>();
   emplace_func4_request<query::am_counter,              am_counter>();
   emplace_func4_request<query::asm_counter,             asm_counter>();
+  emplace_func4_request<query::xrt_smi_config,          xrt_smi_config>();
   emplace_func4_request<query::lapc_status,             lapc_status>();
   emplace_func4_request<query::spc_status,              spc_status>();
   emplace_func4_request<query::accel_deadlock_status,   accel_deadlock_status>();