On Tue, 2022-01-25 at 02:28 +0530, Vaibhav Jain wrote:
> Presently after performing an inject-smart the nvdimm flags reported are
> out of date as shown below where no 'smart_notify' or 'flush_fail' flags
> were reported even though they are set after injecting the smart error:
>
> $ sudo inject-smart -fU nmem0
> [
> {
> "dev":"nmem0",
> "health":{
> "health_state":"fatal",
> "shutdown_state":"dirty",
> "shutdown_count":0
> }
> }
> ]
> $ sudo cat /sys/class/nd/ndctl0/device/nmem0/papr/flags
> flush_fail smart_notify
>
> This happens because nvdimm flags are only parsed once during its probe
> and not refreshed even after a inject-smart operation makes them out of
> date. To fix this the patch forces an update of nvdimm flags via newly
> introduced ndctl_refresh_dimm_flags() thats called successfully submitting
> a 'struct ndctl_cmd' in ndctl_cmd_submit(). This ensures that correct
> nvdimm flags are reported after an interaction with the kernel module which
> may trigger a change nvdimm-flags. With this implemented correct nvdimm
> flags are reported after a inject-smart operation:
>
> $ sudo ndctl inject-smart -fU nmem0
> [
> {
> "dev":"nmem0",
> "flag_failed_flush":true,
> "flag_smart_event":true,
> "health":{
> "health_state":"fatal",
> "shutdown_state":"dirty",
> "shutdown_count":0
> }
> }
> ]
>
> The patch refactors populate_dimm_attributes() to move the nvdimm flags
> parsing code to the newly introduced ndctl_refresh_dimm_flags()
> export. Since reading nvdimm flags requires constructing path using
> 'bus_prefix' which is only available during add_dimm(), the patch
> introduces a new member 'struct ndctl_dimm.bus_prefix' to cache its
> value. During ndctl_refresh_dimm_flags() the cached bus_prefix is used to
> read the contents of the nvdimm flag file and pass it on to the appropriate
> flag parsing function. Finally ndctl_refresh_dimm_flags() is invoked at the
> end of ndctl_cmd_submit() if nd-command submission succeeds.
I think instead of making it ndctl_cmd_submit()'s responsibility to
update any stale state, we should simply NDCTL_EXPORT the refresh
function (also rename it to conform to other ndctl_dimm functions -
ndctl_dimm_refresh_flags()). Then if any operation expects to change
state, it can call the refresh API directly.
In this case, ndctl/inject-smart.c would call
ndctl_dimm_refresh_flags() if its command submission succeeded.
Also some minor comments below.
>
> Signed-off-by: Vaibhav Jain <[email protected]>
> Signed-off-by: Shivaprasad G Bhat <[email protected]>
> ---
> Changelog:
>
> Since v3-Resend:
> 164009789816.744139.2870779016511283907.st...@lep8c.aus.stglabs.ibm.com
> * Rebased this on top of latest ndctl-pending tree that includes changes to
> switch to meson build system.
> ---
> ndctl/lib/libndctl.c | 52 ++++++++++++++++++++++++++++++++------------
> ndctl/lib/private.h | 1 +
> ndctl/libndctl.h | 1 +
> 3 files changed, 40 insertions(+), 14 deletions(-)
>
> diff --git a/ndctl/lib/libndctl.c b/ndctl/lib/libndctl.c
> index 5979a92c113c..abff4ececa27 100644
> --- a/ndctl/lib/libndctl.c
> +++ b/ndctl/lib/libndctl.c
> @@ -608,6 +608,7 @@ static void free_dimm(struct ndctl_dimm *dimm)
> free(dimm->unique_id);
> free(dimm->dimm_buf);
> free(dimm->dimm_path);
> + free(dimm->bus_prefix);
> if (dimm->module)
> kmod_module_unref(dimm->module);
> if (dimm->health_eventfd > -1)
> @@ -1670,14 +1671,34 @@ static int ndctl_bind(struct ndctl_ctx *ctx, struct
> kmod_module *module,
> static int ndctl_unbind(struct ndctl_ctx *ctx, const char *devpath);
> static struct kmod_module *to_module(struct ndctl_ctx *ctx, const char
> *alias);
>
> +void ndctl_refresh_dimm_flags(struct ndctl_dimm *dimm)
> +{
> + struct ndctl_ctx *ctx = dimm->bus->ctx;
> + char *path = dimm->dimm_buf;
> + char buf[SYSFS_ATTR_SIZE];
> +
> + /* Construct path to dimm flags sysfs file */
> + sprintf(path, "%s/%s/flags", dimm->dimm_path, dimm->bus_prefix);
> +
> + if (sysfs_read_attr(ctx, path, buf) < 0)
> + return;
> +
> + /* Reset the flags */
> + dimm->flags.flags = 0;
> + if (ndctl_bus_has_nfit(dimm->bus))
> + parse_nfit_mem_flags(dimm, buf);
> + else if (ndctl_bus_is_papr_scm(dimm->bus))
> + parse_papr_flags(dimm, buf);
> +}
> +
> static int populate_dimm_attributes(struct ndctl_dimm *dimm,
> - const char *dimm_base,
> - const char *bus_prefix)
> + const char *dimm_base)
> {
> int i, rc = -1;
> char buf[SYSFS_ATTR_SIZE];
> struct ndctl_ctx *ctx = dimm->bus->ctx;
> char *path = calloc(1, strlen(dimm_base) + 100);
> + const char *bus_prefix = dimm->bus_prefix;
>
> if (!path)
> return -ENOMEM;
> @@ -1761,16 +1782,10 @@ static int populate_dimm_attributes(struct ndctl_dimm
> *dimm,
> }
>
> sprintf(path, "%s/%s/flags", dimm_base, bus_prefix);
> - if (sysfs_read_attr(ctx, path, buf) == 0) {
> - if (ndctl_bus_has_nfit(dimm->bus))
> - parse_nfit_mem_flags(dimm, buf);
> - else if (ndctl_bus_is_papr_scm(dimm->bus)) {
> - dimm->cmd_family = NVDIMM_FAMILY_PAPR;
> - parse_papr_flags(dimm, buf);
> - }
> - }
> -
> dimm->health_eventfd = open(path, O_RDONLY|O_CLOEXEC);
> +
> + ndctl_refresh_dimm_flags(dimm);
> +
> rc = 0;
> err_read:
>
> @@ -1826,8 +1841,9 @@ static int add_papr_dimm(struct ndctl_dimm *dimm, const
> char *dimm_base)
>
> rc = 0;
> } else if (strcmp(buf, "nvdimm_test") == 0) {
> + dimm->cmd_family = NVDIMM_FAMILY_PAPR;
> /* probe via common populate_dimm_attributes() */
> - rc = populate_dimm_attributes(dimm, dimm_base, "papr");
> + rc = populate_dimm_attributes(dimm, dimm_base);
> }
> out:
> free(path);
> @@ -1924,9 +1940,13 @@ static void *add_dimm(void *parent, int id, const char
> *dimm_base)
> dimm->formats = formats;
> /* Check if the given dimm supports nfit */
> if (ndctl_bus_has_nfit(bus)) {
> - rc = populate_dimm_attributes(dimm, dimm_base, "nfit");
> + dimm->bus_prefix = strdup("nfit");
> + rc = dimm->bus_prefix ?
> + populate_dimm_attributes(dimm, dimm_base) : -ENOMEM
> } else if (ndctl_bus_has_of_node(bus)) {
> - rc = add_papr_dimm(dimm, dimm_base);
> + dimm->bus_prefix = strdup("papr");
> + rc = dimm->bus_prefix ?
> + add_papr_dimm(dimm, dimm_base) : -ENOMEM;
For both of the above, it would be a bit more readable to just return
ENOMEM directly after strdup() if it fails, and then carry on with
add_<foo>_dimm().
dimm->bus_prefix = strdup("papr");
if (!dimm->bus_prefix)
return -ENOMEM;
rc = add_papr_dimm(dimm, dimm_base);
...
> }
>
> if (rc == -ENODEV) {
> @@ -3506,6 +3526,10 @@ NDCTL_EXPORT int ndctl_cmd_submit(struct ndctl_cmd
> *cmd)
> rc = -ENXIO;
> }
> close(fd);
> +
> + /* update dimm-flags if command submitted successfully */
> + if (!rc && cmd->dimm)
> + ndctl_refresh_dimm_flags(cmd->dimm);
> out:
> cmd->status = rc;
> return rc;
> diff --git a/ndctl/lib/private.h b/ndctl/lib/private.h
> index 4d8622978790..e5c56295556d 100644
> --- a/ndctl/lib/private.h
> +++ b/ndctl/lib/private.h
> @@ -75,6 +75,7 @@ struct ndctl_dimm {
> char *unique_id;
> char *dimm_path;
> char *dimm_buf;
> + char *bus_prefix;
> int health_eventfd;
> int buf_len;
> int id;
> diff --git a/ndctl/libndctl.h b/ndctl/libndctl.h
> index 4d5cdbf6f619..b1bafd6d9788 100644
> --- a/ndctl/libndctl.h
> +++ b/ndctl/libndctl.h
> @@ -223,6 +223,7 @@ int ndctl_dimm_is_active(struct ndctl_dimm *dimm);
> int ndctl_dimm_is_enabled(struct ndctl_dimm *dimm);
> int ndctl_dimm_disable(struct ndctl_dimm *dimm);
> int ndctl_dimm_enable(struct ndctl_dimm *dimm);
> +void ndctl_refresh_dimm_flags(struct ndctl_dimm *dimm);
>
> struct ndctl_cmd;
> struct ndctl_cmd *ndctl_bus_cmd_new_ars_cap(struct ndctl_bus *bus,
