ocp-nvme: add get-error-injection command

Documentation/nvme-ocp-get-error-injection.txt

@@ -0,0 +1,57 @@
+nvme-ocp-get-error-injection(1)
+===============================
+
+NAME
+----
+nvme-ocp-get-error-injection - Return set of error injection
+
+SYNOPSIS
+--------
+[verse]
+'nvme ocp get-error-injection' <device> [--no-uuid | -n]
+			[--sel=<select> | -s <select>]
+
+DESCRIPTION
+-----------
+Return set of error injection.
+
+The <device> parameter is mandatory NVMe character device (ex: /dev/nvme0).
+
+This will only work on OCP compliant devices supporting this feature.
+Results for any other device are undefined.
+
+On success it returns 0, error code otherwise.
+
+OPTIONS
+-------
+-n::
+--no-uuid::
+	Do not try to automatically detect UUID index for this command (required
+	for old OCP 1.0 support)
+
+-s <select>::
+--sel=<select>::
+	Select (SEL): This field specifies which value of the attributes
+	to return in the provided data:
++
+[]
+|==================
+|Select|Description
+|0|Current
+|1|Default
+|2|Saved
+|3|Supported capabilities
+|4-7|Reserved
+|==================
+
+EXAMPLES
+--------
+* Has the program issue a get-error-injection to retrieve the 0xC0 get features.
++
+------------
+# nvme ocp get-error-injection /dev/nvme0
+------------
+
+NVME
+----
+Part of the nvme-user suite.

completions/_nvme

@@ -315,6 +315,18 @@ _nvme () {
 				_arguments '*:: :->subcmds'
 				_describe -t commands "nvme ocp tcg-configuration-log options" _ocp_tcg_configuration_log
 				;;
+			(get-error-injection)
+				local _get_error_injection
+				_get_error_injection=(
+				/dev/nvme':supply a device to use (required)'
+				--sel=':0-3: current/default/saved/supported/changed:'
+				-s':alias for --sel'
+				--no-uuid':Skip UUID index search'
+				-n':alias for --no-uuid'
+				)
+				_arguments '*:: :->subcmds'
+				_describe -t commands "nvme ocp get-error-injection options" _get_error_injection
+				;;
 			(*)
 				_files
 				;;
@@ -2502,6 +2514,7 @@ _nvme () {
 			telemetry-string-log':Retrieve Telemetry string Log Page'
 			set-telemetry-profile':Set Telemetry Profile'
 			tcg-configuration-log':tcg configuration log'
+			get-error-injection':get error injection'
 			)
 			_arguments '*:: :->subcmds'
 			_describe -t commands "nvme ocp options" _ocp

completions/bash-nvme-completion.sh

@@ -1478,6 +1478,9 @@ plugin_ocp_opts () {
 		"tcg-configuration-log")
 		opts+=" --output-file= -o"
 			;;
+		"get-error-injection")
+		opts+=" --sel= -s --no-uuid -n"
+			;;
 		"help")
 		opts+=$NO_OPTS
 			;;
@@ -1555,7 +1558,8 @@ _nvme_subcmds () {
 			vs-fw-activate-history device-capability-log \
 			set-dssd-power-state-feature get-dssd-power-state-feature \
 			telemetry-string-log set-telemetry-profile \
-			set-dssd-async-event-config get-dssd-async-event-config"
+			set-dssd-async-event-config get-dssd-async-event-config \
+			get-error-injection"
 	)
 
 	# Associative array mapping plugins to corresponding option completions

plugins/ocp/ocp-nvme.c

@@ -23,6 +23,7 @@
 #include "linux/types.h"
 #include "util/types.h"
 #include "nvme-print.h"
+#include "nvme-wrap.h"
 
 #include "ocp-smart-extended-log.h"
 #include "ocp-clear-features.h"
@@ -113,6 +114,82 @@ struct __packed feature_latency_monitor {
 	__u8  reserved[4083];
 };
 
+struct erri_entry {
+	union {
+		__u8 flags;
+		struct {
+			__u8 enable:1;
+			__u8 single:1;
+			__u8 rsvd2:6;
+		};
+	};
+	__u8 rsvd1;
+	__le16 type;
+	union {
+		__u8 specific[28];
+		struct {
+			__le16 nrtdp;
+			__u8 rsvd4[26];
+		};
+	};
+};
+
+#define ERRI_ENTRIES_MAX 127
+
+enum erri_type {
+	ERRI_TYPE_CPU_CTRL_HANG = 1,
+	ERRI_TYPE_NAND_HANG,
+	ERRI_TYPE_PLP_DEFECT,
+	ERRI_TYPE_LOGICAL_FIRMWARE_ERROR,
+	ERRI_TYPE_DRAM_CORRUPT_CRIT,
+	ERRI_TYPE_DRAM_CORRUPT_NON_CRIT,
+	ERRI_TYPE_NAND_CORRUPT,
+	ERRI_TYPE_SRAM_CORRUPT,
+	ERRI_TYPE_HW_MALFUNCTION,
+	ERRI_TYPE_NO_MORE_NAND_SPARES,
+	ERRI_TYPE_INCOMPLETE_SHUTDOWN,
+};
+
+const char *erri_type_to_string(__le16 type)
+{
+	switch (type) {
+	case ERRI_TYPE_CPU_CTRL_HANG:
+		return "CPU/controller hang";
+	case ERRI_TYPE_NAND_HANG:
+		return "NAND hang";
+	case ERRI_TYPE_PLP_DEFECT:
+		return "PLP defect";
+	case ERRI_TYPE_LOGICAL_FIRMWARE_ERROR:
+		return "logical firmware error";
+	case ERRI_TYPE_DRAM_CORRUPT_CRIT:
+		return "DRAM corruption critical path";
+	case ERRI_TYPE_DRAM_CORRUPT_NON_CRIT:
+		return "DRAM corruption non-critical path";
+	case ERRI_TYPE_NAND_CORRUPT:
+		return "NAND corruption";
+	case ERRI_TYPE_SRAM_CORRUPT:
+		return "SRAM corruption";
+	case ERRI_TYPE_HW_MALFUNCTION:
+		return "HW malfunction";
+	case ERRI_TYPE_NO_MORE_NAND_SPARES:
+		return "no more NAND spares available";
+	case ERRI_TYPE_INCOMPLETE_SHUTDOWN:
+		return "incomplete shutdown";
+	default:
+		break;
+	}
+
+	return "unknown";
+}
+
+struct erri_get_cq_entry {
+	__u32 nume:7;
+	__u32 rsvd7:25;
+};
+
+static const char *sel = "[0-3]: current/default/saved/supported";
+static const char *no_uuid = "Skip UUID index search (UUID index not required for OCP 1.0)";
+
 static int ocp_print_C3_log_normal(struct nvme_dev *dev,
 				   struct ssd_latency_monitor_log *log_data)
 {
@@ -717,7 +794,6 @@ static int eol_plp_failure_mode_set(struct nvme_dev *dev, const __u32 nsid,
 		}
 	}
 
-
 	struct nvme_set_features_args args = {
 		.args_size = sizeof(args),
 		.fd = dev_fd(dev),
@@ -3699,3 +3775,82 @@ static int fw_activation_history_log(int argc, char **argv, struct command *cmd,
 {
 	return ocp_fw_activation_history_log(argc, argv, cmd, plugin);
 }
+
+static int error_injection_get(struct nvme_dev *dev, const __u8 sel, bool uuid)
+{
+	struct erri_get_cq_entry cq_entry;
+	int err;
+	int i;
+	const __u8 fid = 0xc0;
+
+	_cleanup_free_ struct erri_entry *entry = NULL;
+
+	struct nvme_get_features_args args = {
+		.result = (__u32 *)&cq_entry,
+		.data = entry,
+		.args_size = sizeof(args),
+		.fd = dev_fd(dev),
+		.timeout = NVME_DEFAULT_IOCTL_TIMEOUT,
+		.sel = sel,
+		.data_len = sizeof(*entry) * ERRI_ENTRIES_MAX,
+		.fid = fid,
+	};
+
+	if (uuid) {
+		/* OCP 2.0 requires UUID index support */
+		err = ocp_get_uuid_index(dev, &args.uuidx);
+		if (err || !args.uuidx) {
+			nvme_show_error("ERROR: No OCP UUID index found");
+			return err;
+		}
+	}
+
+	entry = nvme_alloc(args.data_len);
+	if (!entry) {
+		nvme_show_error("malloc: %s", strerror(errno));
+		return -errno;
+	}
+
+	err = nvme_cli_get_features(dev, &args);
+	if (!err) {
+		nvme_show_result("Number of Error Injecttions (feature: %#0*x): %#0*x (%s: %d)",
+				 fid ? 4 : 2, fid, cq_entry.nume ? 10 : 8, cq_entry.nume,
+				 nvme_select_to_string(sel), cq_entry.nume);
+		if (sel == NVME_GET_FEATURES_SEL_SUPPORTED)
+			nvme_show_select_result(fid, *args.result);
+		for (i = 0; i < cq_entry.nume; i++) {
+			printf("Entry: %d, Flags: %x (%s%s), Type: %x (%s), NRTDP: %d\n", i,
+			       entry->flags, entry->enable ? "Enabled" : "Disabled",
+			       entry->single ? ", Single instance" : "", entry->type,
+			       erri_type_to_string(entry->type), entry->nrtdp);
+		}
+	} else {
+		nvme_show_error("Could not get feature: %#0*x.", fid ? 4 : 2, fid);
+	}
+
+	return err;
+}
+
+static int get_error_injection(int argc, char **argv, struct command *cmd, struct plugin *plugin)
+{
+	const char *desc = "Return set of error injection";
+	int err;
+	struct config {
+		__u8 sel;
+	};
+	struct config cfg = { 0 };
+
+	_cleanup_nvme_dev_ struct nvme_dev *dev = NULL;
+
+	OPT_ARGS(opts) = {
+		OPT_BYTE("sel", 's', &cfg.sel, sel),
+		OPT_FLAG("no-uuid", 'n', NULL, no_uuid),
+		OPT_END()
+	};
+
+	err = parse_and_open(&dev, argc, argv, desc, opts);
+	if (err)
+		return err;
+
+	return error_injection_get(dev, cfg.sel, !argconfig_parse_seen(opts, "no-uuid"));
+}

plugins/ocp/ocp-nvme.h

@@ -36,6 +36,7 @@ PLUGIN(NAME("ocp", "OCP cloud SSD extensions", OCP_PLUGIN_VERSION),
 		ENTRY("set-dssd-async-event-config", "Set DSSD Async Event Config", set_dssd_async_event_config)
 		ENTRY("get-dssd-async-event-config", "Get DSSD Async Event Config", get_dssd_async_event_config)
 		ENTRY("tcg-configuration-log", "Retrieve TCG Configuration Log Page", ocp_tcg_configuration_log)
+		ENTRY("get-error-injection", "Return set of error injection", get_error_injection)
 	)
 );