8074 need to add FMA event for SSD wearout

Reviewed by: Yuri Pankov <yuri.pankov@nexenta.com>
Reviewed by: Robert Mustacchi <rm@joyent.com>
Reviewed by: Ken Mays <maybird1776@yahoo.com>
Approved by: Dan McDonald <danmcd@joyent.com>

4 files changed, 156 insertions, 15 deletions

diff --git a/usr/src/lib/fm/libdiskstatus/common/ds_impl.h b/usr/src/lib/fm/libdiskstatus/common/ds_impl.h
index 34f8b15d75..8b8132791d 100644
--- a/usr/src/lib/fm/libdiskstatus/common/ds_impl.h
+++ b/usr/src/lib/fm/libdiskstatus/common/ds_impl.h
@@ -21,13 +21,12 @@
 /*
  * Copyright 2007 Sun Microsystems, Inc.  All rights reserved.
  * Use is subject to license terms.
+ * Copyright 2016 Nexenta Systems, Inc.  All rights reserved.
  */
 
 #ifndef	_DS_IMPL_H
 #define	_DS_IMPL_H
 
-#pragma ident	"%Z%%M%	%I%	%E% SMI"
-
 #include <dlfcn.h>
 #include <libnvpair.h>
 #include <sys/types.h>
@@ -53,6 +52,7 @@ struct disk_status {
 	nvlist_t		*ds_overtemp;	/* overtemp */
 	nvlist_t		*ds_predfail;	/* predict fail */
 	nvlist_t		*ds_testfail;	/* self test fail */
+	nvlist_t		*ds_ssmwearout;	/* SSM wearout fail */
 	int			ds_error;	/* last error */
 	nvlist_t		*ds_state;	/* protocol state */
 };
@@ -60,6 +60,7 @@ struct disk_status {
 #define	DS_FAULT_OVERTEMP	0x1
 #define	DS_FAULT_PREDFAIL	0x2
 #define	DS_FAULT_TESTFAIL	0x4
+#define	DS_FAULT_SSMWEAROUT	0x8
 
 extern void dprintf(const char *, ...);
 extern void ddump(const char *, const void *, size_t);
diff --git a/usr/src/lib/fm/libdiskstatus/common/ds_scsi.c b/usr/src/lib/fm/libdiskstatus/common/ds_scsi.c
index 0b80f4d4c2..081f441dd9 100644
--- a/usr/src/lib/fm/libdiskstatus/common/ds_scsi.c
+++ b/usr/src/lib/fm/libdiskstatus/common/ds_scsi.c
@@ -21,10 +21,9 @@
 /*
  * Copyright 2007 Sun Microsystems, Inc.  All rights reserved.
  * Use is subject to license terms.
+ * Copyright 2016 Nexenta Systems, Inc.  All rights reserved.
  */
 
-#pragma ident	"%Z%%M%	%I%	%E% SMI"
-
 #include <assert.h>
 #include <errno.h>
 #include <libdiskstatus.h>
@@ -77,6 +76,8 @@ static int logpage_temp_verify(ds_scsi_info_t *,
     scsi_log_parameter_header_t *, int, nvlist_t *);
 static int logpage_selftest_verify(ds_scsi_info_t *,
     scsi_log_parameter_header_t *, int, nvlist_t *);
+static int logpage_ssm_verify(ds_scsi_info_t *,
+    scsi_log_parameter_header_t *, int, nvlist_t *);
 
 static int logpage_ie_analyze(ds_scsi_info_t *,
     scsi_log_parameter_header_t *, int);
@@ -84,6 +85,8 @@ static int logpage_temp_analyze(ds_scsi_info_t *,
     scsi_log_parameter_header_t *, int);
 static int logpage_selftest_analyze(ds_scsi_info_t *,
     scsi_log_parameter_header_t *, int);
+static int logpage_ssm_analyze(ds_scsi_info_t *,
+    scsi_log_parameter_header_t *, int);
 
 static struct logpage_validation_entry log_validation[] = {
 	{ LOGPAGE_IE,		LOGPAGE_SUPP_IE,
@@ -94,7 +97,10 @@ static struct logpage_validation_entry log_validation[] = {
 	    logpage_temp_verify, logpage_temp_analyze },
 	{ LOGPAGE_SELFTEST,	LOGPAGE_SUPP_SELFTEST,
 	    "self-test",
-	    logpage_selftest_verify, logpage_selftest_analyze }
+	    logpage_selftest_verify, logpage_selftest_analyze },
+	{ LOGPAGE_SSM,		LOGPAGE_SUPP_SSM,
+	    FM_EREPORT_SCSI_SSMWEAROUT,
+	    logpage_ssm_verify, logpage_ssm_analyze }
 };
 
 #define	NLOG_VALIDATION	(sizeof (log_validation) / sizeof (log_validation[0]))
@@ -757,6 +763,51 @@ logpage_selftest_verify(ds_scsi_info_t *sip,
 }
 
 /*
+ * Verify the contents of the Solid State Media (SSM) log page.
+ * As of SBC3r36 SSM log page contains one log parameter:
+ * "Percentage Used Endurance Indicator" which is mandatory.
+ * For the verification phase, we sanity check this parameter
+ * by making sure it's present and it's length is set to 0x04.
+ */
+static int
+logpage_ssm_verify(ds_scsi_info_t *sip,
+    scsi_log_parameter_header_t *lphp, int log_length, nvlist_t *nvl)
+{
+	ushort_t param_code;
+	int i, plen = 0;
+
+	for (i = 0; i < log_length; i += plen) {
+		lphp = (scsi_log_parameter_header_t *)((char *)lphp + plen);
+		param_code = BE_16(lphp->lph_param);
+
+		switch (param_code) {
+		case LOGPARAM_PRCNT_USED:
+			if (nvlist_add_boolean_value(nvl,
+			    FM_EREPORT_SCSI_SSMWEAROUT, B_TRUE) != 0)
+				return (scsi_set_errno(sip, EDS_NOMEM));
+			if (lphp->lph_length != LOGPARAM_PRCNT_USED_PARAM_LEN) {
+				if (nvlist_add_uint8(nvl,
+				    "invalid-length", lphp->lph_length) != 0)
+					return (scsi_set_errno(sip, EDS_NOMEM));
+
+				dprintf("solid state media logpage bad len\n");
+				break;
+			}
+
+			/* verification succeded */
+			return (0);
+		}
+
+		plen = lphp->lph_length +
+		    sizeof (scsi_log_parameter_header_t);
+	}
+
+	/* verification failed */
+	sip->si_supp_log &= ~LOGPAGE_SUPP_SSM;
+	return (0);
+}
+
+/*
  * Load the current IE mode pages
  */
 static int
@@ -1145,6 +1196,65 @@ logpage_selftest_analyze(ds_scsi_info_t *sip, scsi_log_parameter_header_t *lphp,
 }
 
 /*
+ * Analyze the contents of the Solid State Media (SSM) log page's
+ * "Percentage Used Endurance Indicator" log parameter.
+ * We generate a fault if the percentage used is equal to or over
+ * PRCNT_USED_FAULT_THRSH
+ */
+static int
+logpage_ssm_analyze(ds_scsi_info_t *sip, scsi_log_parameter_header_t *lphp,
+    int log_length)
+{
+	uint16_t param_code;
+	scsi_ssm_log_param_t *ssm;
+	nvlist_t *nvl;
+	int i, plen = 0;
+
+	assert(sip->si_dsp->ds_overtemp == NULL);
+	if (nvlist_alloc(&sip->si_dsp->ds_overtemp, NV_UNIQUE_NAME, 0) != 0)
+		return (scsi_set_errno(sip, EDS_NOMEM));
+	nvl = sip->si_dsp->ds_overtemp;
+
+	for (i = 0; i < log_length; i += plen) {
+		lphp = (scsi_log_parameter_header_t *)((uint8_t *)lphp + plen);
+		param_code = BE_16(lphp->lph_param);
+		ssm = (scsi_ssm_log_param_t *)lphp;
+
+		switch (param_code) {
+		case LOGPARAM_PRCNT_USED:
+			if (lphp->lph_length != LOGPARAM_PRCNT_USED_PARAM_LEN)
+				break;
+
+			if ((nvlist_add_uint8(nvl,
+			    FM_EREPORT_PAYLOAD_SCSI_CURSSMWEAROUT,
+			    ssm->ssm_prcnt_used) != 0) ||
+			    (nvlist_add_uint8(nvl,
+			    FM_EREPORT_PAYLOAD_SCSI_THRSHSSMWEAROUT,
+			    PRCNT_USED_FAULT_THRSH) != 0))
+				return (scsi_set_errno(sip, EDS_NOMEM));
+
+			if (ssm->ssm_prcnt_used >= PRCNT_USED_FAULT_THRSH)
+				sip->si_dsp->ds_faults |= DS_FAULT_SSMWEAROUT;
+
+			return (0);
+		}
+
+		plen = lphp->lph_length +
+		    sizeof (scsi_log_parameter_header_t);
+	}
+
+	/*
+	 * If we got this far we didn't see LOGPARAM_PRCNT_USED
+	 * which is strange since we verified that it's there
+	 */
+	dprintf("solid state media logpage analyze failed\n");
+#if DEBUG
+	abort();
+#endif
+	return (scsi_set_errno(sip, EDS_NOT_SUPPORTED));
+}
+
+/*
  * Analyze the IE mode sense page explicitly.  This is only needed if the IE log
  * page is not supported.
  */
diff --git a/usr/src/lib/fm/libdiskstatus/common/ds_scsi.h b/usr/src/lib/fm/libdiskstatus/common/ds_scsi.h
index 6d2648f06b..db4ee6d591 100644
--- a/usr/src/lib/fm/libdiskstatus/common/ds_scsi.h
+++ b/usr/src/lib/fm/libdiskstatus/common/ds_scsi.h
@@ -21,13 +21,12 @@
 /*
  * Copyright 2007 Sun Microsystems, Inc.  All rights reserved.
  * Use is subject to license terms.
+ * Copyright 2016 Nexenta Systems, Inc.  All rights reserved.
  */
 
 #ifndef	_DS_SCSI_H
 #define	_DS_SCSI_H
 
-#pragma ident	"%Z%%M%	%I%	%E% SMI"
-
 #include <sys/types.h>
 #include <sys/byteorder.h>
 #include <sys/scsi/scsi.h>
@@ -50,12 +49,14 @@ extern "C" {
 typedef struct scsi_log_header {
 #if defined(_BIT_FIELDS_LTOH)
 	uint8_t		lh_code : 6,
-			__reserved : 2;
+			lh_spf : 1,
+			lh_ds : 1;
 #else
-	uint8_t		__reserved : 2,
+	uint8_t		lh_ds : 1,
+			lh_spf : 1,
 			lh_code : 6;
 #endif
-	uint8_t		__reserved2;
+	uint8_t		lh_subpage;
 	uint16_t	lh_length;
 } scsi_log_header_t;
 
@@ -158,6 +159,20 @@ typedef struct scsi_selftest_log_param {
 	    sizeof (scsi_log_parameter_header_t))
 
 /*
+ * Described in SBC3
+ */
+typedef struct scsi_ssm_log_param {
+	scsi_log_parameter_header_t ssm_hdr;
+	uint16_t		__reserved2;
+	uint8_t			__reserved1;
+	uchar_t			ssm_prcnt_used;
+} scsi_ssm_log_param_t;
+
+#define	LOGPARAM_PRCNT_USED		0x0001
+#define	LOGPARAM_PRCNT_USED_PARAM_LEN	0x04
+#define	PRCNT_USED_FAULT_THRSH		90
+
+/*
  * Mode sense/select page header information
  */
 typedef struct scsi_ms_header {
@@ -278,6 +293,8 @@ typedef struct scsi_ie_page {
 #define	LOGPAGE_TEMP		0x0d
 #define	LOGPAGE_SELFTEST	0x10
 #define	LOGPAGE_IE		0x2f
+/* Solid State Media log page code */
+#define	LOGPAGE_SSM		0x11
 
 /* ASC constants */
 #define	ASC_INVALID_OPCODE				0x20
@@ -307,6 +324,7 @@ typedef struct scsi_ie_page {
 #define	LOGPAGE_SUPP_IE			0x1
 #define	LOGPAGE_SUPP_TEMP		0x2
 #define	LOGPAGE_SUPP_SELFTEST		0x4
+#define	LOGPAGE_SUPP_SSM		0x8
 
 #define	MSG_BUFLEN	256
 
diff --git a/usr/src/lib/fm/libdiskstatus/common/libdiskstatus.c b/usr/src/lib/fm/libdiskstatus/common/libdiskstatus.c
index e2cf5ad7a2..18468289a7 100644
--- a/usr/src/lib/fm/libdiskstatus/common/libdiskstatus.c
+++ b/usr/src/lib/fm/libdiskstatus/common/libdiskstatus.c
@@ -21,10 +21,9 @@
 /*
  * Copyright 2007 Sun Microsystems, Inc.  All rights reserved.
  * Use is subject to license terms.
+ * Copyright 2016 Nexenta Systems, Inc.  All rights reserved.
  */
 
-#pragma ident	"%Z%%M%	%I%	%E% SMI"
-
 /*
  * Disk status library
  *
@@ -33,9 +32,10 @@
  * SCSI (and therefore SATA) disks are currently supported.  The library is
  * capable of detecting the following status conditions:
  *
- * 	- Predictive failure
- * 	- Overtemp
- * 	- Self-test failure
+ *	- Predictive failure
+ *	- Overtemp
+ *	- Self-test failure
+ *	- Solid State Media wearout
  */
 
 #include <assert.h>
@@ -131,6 +131,7 @@ disk_status_close(disk_status_t *dsp)
 	nvlist_free(dsp->ds_predfail);
 	nvlist_free(dsp->ds_overtemp);
 	nvlist_free(dsp->ds_testfail);
+	nvlist_free(dsp->ds_ssmwearout);
 	if (dsp->ds_data)
 		dsp->ds_transport->dt_close(dsp->ds_data);
 	(void) close(dsp->ds_fd);
@@ -172,6 +173,8 @@ disk_status_get(disk_status_t *dsp)
 	nvlist_free(dsp->ds_testfail);
 	nvlist_free(dsp->ds_predfail);
 	nvlist_free(dsp->ds_overtemp);
+	nvlist_free(dsp->ds_ssmwearout);
+	dsp->ds_ssmwearout = NULL;
 	dsp->ds_testfail = dsp->ds_overtemp = dsp->ds_predfail = NULL;
 	dsp->ds_faults = 0;
 
@@ -223,6 +226,15 @@ disk_status_get(disk_status_t *dsp)
 			goto nverror;
 	}
 
+	if (dsp->ds_ssmwearout != NULL) {
+		if ((err = nvlist_add_boolean_value(faults,
+		    FM_EREPORT_SCSI_SSMWEAROUT,
+		    (dsp->ds_faults & DS_FAULT_SSMWEAROUT) != 0)) != 0 ||
+		    (err = nvlist_add_nvlist(nvl, FM_EREPORT_SCSI_SSMWEAROUT,
+		    dsp->ds_ssmwearout)) != 0)
+			goto nverror;
+	}
+
 	if ((err = nvlist_add_nvlist(nvl, "faults", faults)) != 0)
 		goto nverror;