[ndctl,04/16] ndctl/dimm: Detect firmware-update vs ARS conflict

Commit Message

Dan Williams July 7, 2020, 2:40 a.m. UTC

Prevent firmware-update while ARS is active unless --force is supplied.
Translate the FW_ERETRY return code as a conflict with a background DIMM
process.

Signed-off-by: Dan Williams <[e-mail]>
---
 Documentation/ndctl/ndctl-update-firmware.txt |    5 ++++
 ndctl/dimm.c                                  |   32 ++++++++++++++++++-------
 2 files changed, 28 insertions(+), 9 deletions(-)

Patch

diff --git a/Documentation/ndctl/ndctl-update-firmware.txt b/Documentation/ndctl/ndctl-update-firmware.txt
index f93da6bf15e7..bcf61abaa989 100644
--- a/Documentation/ndctl/ndctl-update-firmware.txt
+++ b/Documentation/ndctl/ndctl-update-firmware.txt
@@ -47,6 +47,11 @@  include::xable-bus-options.txt[]
 --firmware::
 	firmware file used to perform the update
 
+-i::
+--force::
+	Ignore in-progress Address Range Scrub and try to submit the
+	firmware update.
+
 -v::
 --verbose::
         Emit debug messages for the namespace check process.
diff --git a/ndctl/dimm.c b/ndctl/dimm.c
index 8523ead36eb5..f67f78e6c420 100644
--- a/ndctl/dimm.c
+++ b/ndctl/dimm.c
@@ -18,7 +18,6 @@ 
 #include <unistd.h>
 #include <limits.h>
 #include <syslog.h>
-#include <util/log.h>
 #include <util/size.h>
 #include <uuid/uuid.h>
 #include <util/json.h>
@@ -33,6 +32,11 @@ 
 #include <ndctl/firmware-update.h>
 #include <util/keys.h>
 
+static const char *cmd_name = "dimm";
+static int err_count;
+#define err(fmt, ...) \
+	({ err_count++; error("%s: " fmt, cmd_name, ##__VA_ARGS__); })
+
 struct action_context {
 	struct json_object *jdimms;
 	enum ndctl_namespace_version labelversion;
@@ -832,24 +836,30 @@  static int update_firmware(struct ndctl_dimm *dimm,
 
 static int action_update(struct ndctl_dimm *dimm, struct action_context *actx)
 {
+	struct ndctl_bus *bus = ndctl_dimm_get_bus(dimm);
+	const char *devname = ndctl_dimm_get_devname(dimm);
 	int rc;
 
 	rc = ndctl_dimm_fw_update_supported(dimm);
 	switch (rc) {
 	case -ENOTTY:
-		error("%s: firmware update not supported by ndctl.",
-			ndctl_dimm_get_devname(dimm));
+		err("%s: firmware update not supported by ndctl.", devname);
 		return rc;
 	case -EOPNOTSUPP:
-		error("%s: firmware update not supported by the kernel",
-			ndctl_dimm_get_devname(dimm));
+		err("%s: firmware update not supported by the kernel", devname);
 		return rc;
 	case -EIO:
-		error("%s: firmware update not supported by either platform firmware or the kernel.",
-			ndctl_dimm_get_devname(dimm));
+		err("%s: firmware update not supported by either platform firmware or the kernel.",
+				devname);
 		return rc;
 	}
 
+	if (ndctl_bus_get_scrub_state(bus) > 0 && !param.force) {
+		err("%s: scrub active, retry after 'ndctl wait-scrub'",
+				devname);
+		return -EBUSY;
+	}
+
 	rc = update_verify_input(actx);
 	if (rc < 0)
 		return rc;
@@ -1073,7 +1083,8 @@  OPT_STRING('i', "input", &param.infile, "input-file", \
 
 #define UPDATE_OPTIONS() \
 OPT_STRING('f', "firmware", &param.infile, "firmware-file", \
-	"firmware filename for update")
+	"firmware filename for update"), \
+OPT_BOOLEAN('i', "force", &param.force, "ignore ARS status, try to force update")
 
 #define INIT_OPTIONS() \
 OPT_BOOLEAN('f', "force", &param.force, \
@@ -1386,7 +1397,10 @@  int cmd_enable_dimm(int argc, const char **argv, struct ndctl_ctx *ctx)
 
 int cmd_update_firmware(int argc, const char **argv, struct ndctl_ctx *ctx)
 {
-	int count = dimm_action(argc, argv, ctx, action_update, update_options,
+	int count;
+
+	cmd_name = "update firmware";
+	count = dimm_action(argc, argv, ctx, action_update, update_options,
 			"ndctl update-firmware <nmem0> [<nmem1>..<nmemN>] [<options>]");
 
 	fprintf(stderr, "updated %d nmem%s.\n", count >= 0 ? count : 0,