CR-1158005 backport (#7551)

* CR-1158005 Report if a device is in a bad state (#7507) * CR-1158005 Add status field to hardware context. Add device status query to legacy query path Signed-off-by: Daniel Benusovich <dbenusov@xilinx.com> * CR-1158005 Fix compilation error Signed-off-by: Daniel Benusovich <dbenusov@xilinx.com> * CR-1158005 Refactor what is displayed for context status. Refactor variable names Signed-off-by: Daniel Benusovich <dbenusov@xilinx.com> * CR-1158005 Add comments for device status Signed-off-by: Daniel Benusovich <dbenusov@xilinx.com> * CR-1158005 Remove context status from hw context. Add device status to json output. Update display for dynamic regions report Signed-off-by: Daniel Benusovich <dbenusov@xilinx.com> --------- Signed-off-by: Daniel Benusovich <dbenusov@xilinx.com> (cherry picked from commit 3554c14) * CR-1158005 Fix compilation issues Signed-off-by: Daniel Benusovich <dbenusov@xilinx.com> --------- Signed-off-by: Daniel Benusovich <dbenusov@xilinx.com>
Xilinx · May 19, 2023 · 73a310b · 73a310b
1 parent 4698c28
commit 73a310b
Show file tree

Hide file tree

Showing 4 changed files with 47 additions and 2 deletions.
diff --git a/src/runtime_src/core/common/query_requests.h b/src/runtime_src/core/common/query_requests.h
@@ -76,6 +76,7 @@ enum class key_type
   ip_layout_raw,
   clock_freq_topology_raw,
   dma_stream,
+  device_status,
   kds_cu_info,
   kds_mode,
   kds_cu_stat,
@@ -725,6 +726,34 @@ struct kds_mode : request
   get(const device*) const = 0;
 };
 
+/**
+ * Extract the status of the device
+ * This states whether or not a device is stuck due to an xclbin issue
+ */
+struct device_status : request
+{
+  using result_type = uint32_t;
+  static const key_type key = key_type::device_status;
+
+  virtual boost::any
+  get(const device*) const = 0;
+
+  static std::string
+  parse_status(const result_type status)
+  {
+    switch (status) {
+      case 0:
+        return "HEALTHY";
+      case 1:
+        return "HANG";
+      case 2:
+        return "UNKNOWN";
+      default:
+        throw xrt_core::system_error(EINVAL, "Invalid device status: " + status);
+    }
+  }
+};
+
 struct kds_cu_stat : request
 {
   struct data {

diff --git a/src/runtime_src/core/pcie/linux/device_linux.cpp b/src/runtime_src/core/pcie/linux/device_linux.cpp
@@ -545,6 +545,7 @@ initialize_query_table()
   emplace_sysfs_get<query::mig_ecc_ce_ffa>                   ("mig", "ecc_ce_ffa");
   emplace_sysfs_get<query::mig_ecc_ue_ffa>                   ("mig", "ecc_ue_ffa");
   emplace_sysfs_get<query::flash_bar_offset>                 ("flash", "bar_off");
+  emplace_sysfs_get<query::device_status>                      ("", "device_bad_state");
   emplace_sysfs_get<query::is_mfg>                           ("", "mfg");
   emplace_sysfs_get<query::mfg_ver>                          ("", "mfg_ver");
   emplace_sysfs_get<query::is_recovery>                      ("", "recovery");

diff --git a/src/runtime_src/core/tools/common/ReportDynamicRegion.cpp b/src/runtime_src/core/tools/common/ReportDynamicRegion.cpp
@@ -278,7 +278,7 @@ ReportDynamicRegion::getPropertyTree20202( const xrt_core::device * _pDevice,
 }
 
 void 
-ReportDynamicRegion::writeReport( const xrt_core::device* /*_pDevice*/,
+ReportDynamicRegion::writeReport( const xrt_core::device* _pDevice,
                        const boost::property_tree::ptree& _pt, 
                        const std::vector<std::string>& /*_elementsFilter*/,
                        std::ostream & _output) const
@@ -292,6 +292,12 @@ ReportDynamicRegion::writeReport( const xrt_core::device* /*_pDevice*/,
   if(pt_dfx.empty())
     return;
 
+  auto device_status = 2;
+  try {
+    device_status = xrt_core::device_query<xrt_core::query::device_status>(_pDevice);
+  } catch (...) {}
+  _output << boost::format("  Device Status: %s\n") % xrt_core::query::device_status::parse_status(device_status);
+
   for(auto& k_dfx : pt_dfx) {
     const boost::property_tree::ptree& dfx = k_dfx.second;
     _output << "Xclbin UUID" << std::endl;

diff --git a/src/runtime_src/core/tools/common/XBHelpMenus.cpp b/src/runtime_src/core/tools/common/XBHelpMenus.cpp
@@ -697,10 +697,19 @@ XBUtilities::produce_reports( xrt_core::device_collection devices,
       ptDevice.put("interface_type", "pcie");
       ptDevice.put("device_id", xrt_core::query::pcie_bdf::to_string(bdf));
 
+      auto device_status = 2;
+      try {
+        device_status = xrt_core::device_query<xrt_core::query::device_status>(device);
+      } catch (...) {}
+      ptDevice.put("device_status", xrt_core::query::device_status::parse_status(device_status));
+
       bool is_mfg = false;
       try {
         is_mfg = xrt_core::device_query<xrt_core::query::is_mfg>(device);
-      } catch (...) {}
+      } 
+      catch (...) {
+        is_mfg = false;
+      }
 
       //if factory mode
       std::string platform = "<not defined>";