linux-nvdimm.lists.01.org archive mirror
 help / color / mirror / Atom feed
From: Vaibhav Jain <vaibhav@linux.ibm.com>
To: linuxppc-dev@lists.ozlabs.org, linux-nvdimm@lists.01.org
Cc: Vaibhav Jain <vaibhav@linux.ibm.com>,
	"Aneesh Kumar K . V" <aneesh.kumar@linux.ibm.com>,
	Michael Ellerman <mpe@ellerman.id.au>
Subject: [RFC PATCH 3/4] powerpc/papr_scm: Implement support for PAPR_SCM_PDSM_READ_PERF_STATS
Date: Mon, 18 May 2020 16:38:13 +0530	[thread overview]
Message-ID: <20200518110814.145644-4-vaibhav@linux.ibm.com> (raw)
In-Reply-To: <20200518110814.145644-1-vaibhav@linux.ibm.com>

Implement support for pdsm READ_PERF_STATS to be used by libndctl to
fetch all NVDIMM performance statistics. The stats are to be exchanged
via newly introduced 'struct nd_pdsm_get_perf_stats' which is
allocated and sent by libndctl to papr_scm. The struct contains
members 'in_offset' and 'in_length' to provide incremental access to
performance statistics data buffer and workaround 'libnvdimm' limit of
256 bytes evelope size.

The patch introduces new function 'papr_scm_read_perf_stats()' to
service this pdsm and copy the requested chunk of performance stats to
the libndctl provided payload buffer for the given offset and length.

Signed-off-by: Vaibhav Jain <vaibhav@linux.ibm.com>
---
 arch/powerpc/include/uapi/asm/papr_scm_pdsm.h | 35 +++++++
 arch/powerpc/platforms/pseries/papr_scm.c     | 91 +++++++++++++++++++
 2 files changed, 126 insertions(+)

diff --git a/arch/powerpc/include/uapi/asm/papr_scm_pdsm.h b/arch/powerpc/include/uapi/asm/papr_scm_pdsm.h
index 40ec55d06f4c..2db4ffdff285 100644
--- a/arch/powerpc/include/uapi/asm/papr_scm_pdsm.h
+++ b/arch/powerpc/include/uapi/asm/papr_scm_pdsm.h
@@ -115,6 +115,7 @@ enum papr_scm_pdsm {
 	PAPR_SCM_PDSM_MIN = 0x0,
 	PAPR_SCM_PDSM_HEALTH,
 	PAPR_SCM_PDSM_FETCH_PERF_STATS,
+	PAPR_SCM_PDSM_READ_PERF_STATS,
 	PAPR_SCM_PDSM_MAX,
 };
 
@@ -183,4 +184,38 @@ struct nd_pdsm_fetch_perf_stats_v1 {
 #define nd_pdsm_fetch_perf_stats nd_pdsm_fetch_perf_stats_v1
 #define ND_PDSM_FETCH_PERF_STATS_VERSION 1
 
+/*
+ * Holds a single performance stat. papr_scm owns a buffer that holds an array
+ * of all the available stats and their values. Access to the buffer is provided
+ * via PERF_STAT_SIZE and READ_PERF_STATS psdm.
+ * id : id of the performance stat. Usually acsii encode stat name.
+ * val : Non normalized value of the id.
+ */
+
+struct nd_pdsm_perf_stat {
+	__u64 id;
+	__u64 val;
+};
+
+/*
+ * Returns a chunk of performance stats buffer data to libndctl.
+ * This is needed to overcome the 256 byte envelope size limit enforced by
+ * libnvdimm.
+ * in_offset: The starting offset to perf stats data buffer.
+ * in_length: Length of data to be copied to 'stats_data'
+ * stats_data: Holds the chunk of requested perf stats data buffer.
+ *
+ * Note: To prevent races in reading performance stats, in_offset and in_length
+ * should multiple of 16-Bytes. If they are not then papr_scm will return an
+ * -EINVAL error.
+ */
+struct nd_pdsm_read_perf_stats_v1 {
+	__u32 in_offset;
+	__u32 in_length;
+	struct nd_pdsm_perf_stat stats_data[];
+} __packed;
+
+#define nd_pdsm_read_perf_stats nd_pdsm_read_perf_stats_v1
+#define ND_PDSM_READ_PERF_STATS_VERSION 1
+
 #endif /* _UAPI_ASM_POWERPC_PAPR_SCM_PDSM_H_ */
diff --git a/arch/powerpc/platforms/pseries/papr_scm.c b/arch/powerpc/platforms/pseries/papr_scm.c
index f8b37a830aed..06744d7fe727 100644
--- a/arch/powerpc/platforms/pseries/papr_scm.c
+++ b/arch/powerpc/platforms/pseries/papr_scm.c
@@ -525,6 +525,94 @@ static int is_cmd_valid(struct nvdimm *nvdimm, unsigned int cmd, void *buf,
 	return 0;
 }
 
+/*
+ * Read the contents of dimm performance statistics buffer at the given
+ * 'in_offset' and copy 'in_length' number of bytes to the pkg payload.
+ * Both 'in_offset' and 'in_length' are expected to be in multiples of
+ * 16-Bytes to prevent a read/write race that may cause malformed values
+ * top be returned as performance statistics buffer content.
+ */
+static int papr_scm_read_perf_stats(struct papr_scm_priv *p,
+				    struct nd_pdsm_cmd_pkg *pkg)
+{
+	int rc;
+	struct nd_pdsm_read_perf_stats *stats =
+		(struct nd_pdsm_read_perf_stats *)pdsm_cmd_to_payload(pkg);
+	const size_t copysize = sizeof(struct nd_pdsm_read_perf_stats);
+	off_t offset;
+
+	/*
+	 * If the requested payload version is greater than one we know
+	 * about, return the payload version we know about and let
+	 * caller/userspace handle.
+	 */
+	if (pkg->payload_version > ND_PDSM_READ_PERF_STATS_VERSION)
+		pkg->payload_version = ND_PDSM_READ_PERF_STATS_VERSION;
+
+	if (pkg->hdr.nd_size_out < copysize) {
+		dev_dbg(&p->pdev->dev, "Truncated payload (%u). Expected (%lu)",
+			pkg->hdr.nd_size_out, copysize);
+		rc = -ENOSPC;
+		goto out;
+	}
+
+	/* Protect concurrent modifications to papr_scm_priv */
+	rc = mutex_lock_interruptible(&p->health_mutex);
+	if (rc)
+		goto out;
+
+	if (!p->len_stat_buffer) {
+		dev_dbg(&p->pdev->dev, "Perf stats: req for unsupported device");
+		rc = -ENOENT;
+		goto mutex_unlock_out;
+	}
+
+	/* calculate offset skipping the perf_stats buffer header */
+	offset = stats->in_offset + sizeof(*p->perf_stats);
+	/* Cap the copy length to extend of stats buffer */
+	stats->in_length = min(stats->in_length,
+			       (__u32)(p->len_stat_buffer - offset));
+
+	/*
+	 * Ensure that offset and length are valid and multiples of 16 bytes.
+	 * PDSM FETCH_PERF_STATS can interleave in between PDSM READ_PERF_STAT.
+	 * Since this is a read/write race hence malformed performance stats
+	 * buffer contents that may be returned.
+	 * A 16-Byte read alignment constraint forces a read granularity of
+	 * same the size of each performance stat and they are guaranteed to
+	 * remain stable during 'health_mutex' lock context.
+	 */
+	if (offset >= p->len_stat_buffer || (offset % 16) ||
+	    (stats->in_length % 16)) {
+		dev_dbg(&p->pdev->dev,
+			"Perf stats: Invalid offset(0x%lx) or length(0x%x)",
+			offset, stats->in_length);
+		rc = -EINVAL;
+		goto mutex_unlock_out;
+	}
+
+	/* Put the stats buffer data in the payload buffer */
+	memcpy(stats->stats_data,
+	       (void *)p->perf_stats + offset, stats->in_length);
+
+	pkg->hdr.nd_fw_size = stats->in_length;
+
+	dev_dbg(&p->pdev->dev, "Copying payload size=%u version=0x%x\n",
+		stats->in_length, pkg->payload_version);
+
+mutex_unlock_out:
+	mutex_unlock(&p->health_mutex);
+out:
+	/*
+	 * Put the error in out package and return success from function
+	 * so that errors if any are propogated back to userspace.
+	 */
+	pkg->cmd_status = rc;
+	dev_dbg(&p->pdev->dev, "completion code = %d\n", rc);
+
+	return 0;
+}
+
 /* Return the size in bytes for returning all perf stats to libndctl */
 static int papr_scm_fetch_perf_stats(struct papr_scm_priv *p,
 				     struct nd_pdsm_cmd_pkg *pkg)
@@ -664,6 +752,9 @@ static int papr_scm_service_pdsm(struct papr_scm_priv *p,
 	case PAPR_SCM_PDSM_FETCH_PERF_STATS:
 		return papr_scm_fetch_perf_stats(p, call_pkg);
 
+	case PAPR_SCM_PDSM_READ_PERF_STATS:
+		return papr_scm_read_perf_stats(p, call_pkg);
+
 	default:
 		dev_dbg(&p->pdev->dev, "Unsupported PDSM request 0x%llx\n",
 			call_pkg->hdr.nd_command);
-- 
2.26.2
_______________________________________________
Linux-nvdimm mailing list -- linux-nvdimm@lists.01.org
To unsubscribe send an email to linux-nvdimm-leave@lists.01.org

  parent reply	other threads:[~2020-05-18 11:08 UTC|newest]

Thread overview: 6+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2020-05-18 11:08 [RFC PATCH 0/4] powerpc/papr_scm: Add support for reporting NVDIMM performance statistics Vaibhav Jain
2020-05-18 11:08 ` [RFC PATCH 1/4] powerpc/papr_scm: Fetch nvdimm performance stats from PHYP Vaibhav Jain
2020-05-18 11:08 ` [RFC PATCH 2/4] powerpc/papr_scm: Add support for PAPR_SCM_PDSM_FETCH_PERF_STATS Vaibhav Jain
2020-05-18 11:08 ` Vaibhav Jain [this message]
2020-05-18 11:08 ` [RFC PATCH 4/4] powerpc/papr_scm: Add support for PDSM GET_PERF_STAT Vaibhav Jain
2020-10-21 16:52 ` [RFC PATCH 0/4] powerpc/papr_scm: Add support for reporting NVDIMM performance statistics Michal Suchánek

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20200518110814.145644-4-vaibhav@linux.ibm.com \
    --to=vaibhav@linux.ibm.com \
    --cc=aneesh.kumar@linux.ibm.com \
    --cc=linux-nvdimm@lists.01.org \
    --cc=linuxppc-dev@lists.ozlabs.org \
    --cc=mpe@ellerman.id.au \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).