[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <20250917163623.00004a3c@huawei.com>
Date: Wed, 17 Sep 2025 16:36:23 +0100
From: Jonathan Cameron <jonathan.cameron@...wei.com>
To: Neeraj Kumar <s.neeraj@...sung.com>
CC: <linux-cxl@...r.kernel.org>, <nvdimm@...ts.linux.dev>,
<linux-kernel@...r.kernel.org>, <gost.dev@...sung.com>,
<a.manzanares@...sung.com>, <vishak.g@...sung.com>,
<neeraj.kernel@...il.com>, <cpgs@...sung.com>
Subject: Re: [PATCH V3 06/20] nvdimm/region_label: Add region label update
support
On Wed, 17 Sep 2025 19:11:02 +0530
Neeraj Kumar <s.neeraj@...sung.com> wrote:
> Modified __pmem_label_update() to update region labels into LSA
>
I'm struggling to follow the use of the union for the two label types
in much of this code. To me if feels like that should only be a thing
at the init_labels() point on the basis I think it's only there that
we need to handle both in the same storage.
For the rest I'd just pay the small price of duplication that will
occur if you just split he functions up.
> Signed-off-by: Neeraj Kumar <s.neeraj@...sung.com>
> ---
> drivers/nvdimm/label.c | 269 ++++++++++++++++++++++++++------
> drivers/nvdimm/label.h | 15 ++
> drivers/nvdimm/namespace_devs.c | 12 ++
> drivers/nvdimm/nd.h | 38 ++++-
> include/linux/libnvdimm.h | 8 +
> 5 files changed, 289 insertions(+), 53 deletions(-)
>
> diff --git a/drivers/nvdimm/label.c b/drivers/nvdimm/label.c
> index 182f8c9a01bf..209c73f6b7e7 100644
> --- a/drivers/nvdimm/label.c
> +++ b/drivers/nvdimm/label.c
> @@ -381,6 +381,16 @@ static void nsl_calculate_checksum(struct nvdimm_drvdata *ndd,
> nsl_set_checksum(ndd, nd_label, sum);
> }
> +static bool is_label_reapable(struct nd_interleave_set *nd_set,
> + struct nd_namespace_pmem *nspm,
> + struct nvdimm_drvdata *ndd,
> + union nd_lsa_label *label,
> + enum label_type ltype,
> + unsigned long *flags)
> +{
> + switch (ltype) {
> + case NS_LABEL_TYPE:
> + if (test_and_clear_bit(ND_LABEL_REAP, flags) ||
> + nsl_uuid_equal(ndd, &label->ns_label, nspm->uuid))
> + return true;
> +
> + break;
> + case RG_LABEL_TYPE:
> + if (region_label_uuid_equal(&label->region_label,
> + &nd_set->uuid))
> + return true;
> +
> + break;
> + }
> +
> + return false;
> +}
> +
> +static int __pmem_label_update(struct nd_region *nd_region,
> + struct nd_mapping *nd_mapping,
> + struct nd_namespace_pmem *nspm,
> + int pos, unsigned long flags,
> + enum label_type ltype)
> +{
> + struct nd_interleave_set *nd_set = nd_region->nd_set;
> + struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
> + struct nd_namespace_index *nsindex;
> + struct nd_label_ent *label_ent;
> + union nd_lsa_label *lsa_label;
> + unsigned long *free;
> + struct device *dev;
> + u32 nslot, slot;
> + size_t offset;
> + int rc;
> +
> + if (!preamble_next(ndd, &nsindex, &free, &nslot))
> + return -ENXIO;
> +
> /* allocate and write the label to the staging (next) index */
> slot = nd_label_alloc_slot(ndd);
> if (slot == UINT_MAX)
> return -ENXIO;
> dev_dbg(ndd->dev, "allocated: %d\n", slot);
>
> - nd_label = to_label(ndd, slot);
> - memset(nd_label, 0, sizeof_namespace_label(ndd));
> - nsl_set_type(ndd, nd_label);
> - nsl_set_uuid(ndd, nd_label, nspm->uuid);
> - nsl_set_name(ndd, nd_label, nspm->alt_name);
> - nsl_set_flags(ndd, nd_label, flags);
> - nsl_set_nlabel(ndd, nd_label, nd_region->ndr_mappings);
> - nsl_set_nrange(ndd, nd_label, 1);
> - nsl_set_position(ndd, nd_label, pos);
> - nsl_set_isetcookie(ndd, nd_label, cookie);
> - nsl_set_rawsize(ndd, nd_label, resource_size(res));
> - nsl_set_lbasize(ndd, nd_label, nspm->lbasize);
> - nsl_set_dpa(ndd, nd_label, res->start);
> - nsl_set_slot(ndd, nd_label, slot);
> - nsl_set_alignment(ndd, nd_label, 0);
> - nsl_set_type_guid(ndd, nd_label, &nd_set->type_guid);
> - nsl_set_region_uuid(ndd, nd_label, NULL);
> - nsl_set_claim_class(ndd, nd_label, ndns->claim_class);
> - nsl_calculate_checksum(ndd, nd_label);
> - nd_dbg_dpa(nd_region, ndd, res, "\n");
> + lsa_label = (union nd_lsa_label *) to_label(ndd, slot);
This cast feels rather dubious.
If the union makes sense in general, then this should be changed
to return the union.
> + memset(lsa_label, 0, sizeof_namespace_label(ndd));
> +
> + switch (ltype) {
> + case NS_LABEL_TYPE:
> + dev = &nspm->nsio.common.dev;
> + rc = namespace_label_update(nd_region, nd_mapping,
> + nspm, pos, flags, &lsa_label->ns_label,
> + nsindex, slot);
> + if (rc)
> + return rc;
> +
> + break;
> + case RG_LABEL_TYPE:
> + dev = &nd_region->dev;
> + region_label_update(nd_region, &lsa_label->region_label,
> + nd_mapping, pos, flags, slot);
> +
> + break;
> + }
>
> /* update label */
> - offset = nd_label_offset(ndd, nd_label);
> - rc = nvdimm_set_config_data(ndd, offset, nd_label,
> + offset = nd_label_offset(ndd, &lsa_label->ns_label);
This doesn't make sense as the type might be either an ns_label or a region_label.
If there is a generic header (I'm guessing there is) then define that as part of the
union with just the shared parts and use that to avoid any implication of what the type
is in calls like this. Also make nd_label_offset() take the union not the specific bit.
> + rc = nvdimm_set_config_data(ndd, offset, lsa_label,
> sizeof_namespace_label(ndd));
> if (rc < 0)
> return rc;
> @@ -955,8 +1054,10 @@ static int __pmem_label_update(struct nd_region *nd_region,
> list_for_each_entry(label_ent, &nd_mapping->labels, list) {
> if (!label_ent->label)
> continue;
> - if (test_and_clear_bit(ND_LABEL_REAP, &label_ent->flags) ||
> - nsl_uuid_equal(ndd, label_ent->label, nspm->uuid))
> +
> + if (is_label_reapable(nd_set, nspm, ndd,
> + (union nd_lsa_label *) label_ent->label,
If we are going with the union that label_ent->label should be a union that
we don't need to cast.
> + ltype, &label_ent->flags))
> reap_victim(nd_mapping, label_ent);
> }
>
> @@ -966,19 +1067,20 @@ static int __pmem_label_update(struct nd_region *nd_region,
> if (rc)
> return rc;
>
> - list_for_each_entry(label_ent, &nd_mapping->labels, list)
> - if (!label_ent->label) {
> - label_ent->label = nd_label;
> - nd_label = NULL;
> - break;
> - }
> - dev_WARN_ONCE(&nspm->nsio.common.dev, nd_label,
> - "failed to track label: %d\n",
> - to_slot(ndd, nd_label));
> - if (nd_label)
> - rc = -ENXIO;
> + list_for_each_entry(label_ent, &nd_mapping->labels, list) {
> + if (label_ent->label)
> + continue;
This flow change is unrelated to the rest here. I'd push it back
to the simpler patch that change the locking. Make sure to call it out there
though. Or just don't do it and keep this patch a little more readable!
>
> - return rc;
> + label_ent->label = &lsa_label->ns_label;
> + lsa_label = NULL;
> + break;
> + }
> + dev_WARN_ONCE(dev, lsa_label, "failed to track label: %d\n",
> + to_slot(ndd, &lsa_label->ns_label));
> + if (lsa_label)
> + return -ENXIO;
> +
> + return 0;
> }
>
> static int init_labels(struct nd_mapping *nd_mapping, int num_labels)
> @@ -1068,6 +1170,21 @@ static int del_labels(struct nd_mapping *nd_mapping, uuid_t *uuid)
> nd_inc_seq(__le32_to_cpu(nsindex->seq)), 0);
> }
>
> +static int find_region_label_count(struct nvdimm_drvdata *ndd,
> + struct nd_mapping *nd_mapping)
> +{
> + struct nd_label_ent *label_ent;
> + int region_label_cnt = 0;
> +
> + guard(mutex)(&nd_mapping->lock);
> + list_for_each_entry(label_ent, &nd_mapping->labels, list)
> + if (is_region_label(ndd,
> + (union nd_lsa_label *) label_ent->label))
As above. If it's a union make label_ent->label that union type.
> + region_label_cnt++;
> +
> + return region_label_cnt;
> +}
> +
> int nd_pmem_namespace_label_update(struct nd_region *nd_region,
> struct nd_namespace_pmem *nspm, resource_size_t size)
> {
> @@ -1076,6 +1193,7 @@ int nd_pmem_namespace_label_update(struct nd_region *nd_region,
> for (i = 0; i < nd_region->ndr_mappings; i++) {
> struct nd_mapping *nd_mapping = &nd_region->mapping[i];
> struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
> + int region_label_cnt = 0;
> struct resource *res;
> int count = 0;
>
> @@ -1091,12 +1209,19 @@ int nd_pmem_namespace_label_update(struct nd_region *nd_region,
> count++;
> WARN_ON_ONCE(!count);
>
> - rc = init_labels(nd_mapping, count);
> + region_label_cnt = find_region_label_count(ndd, nd_mapping);
> + /*
> + * init_labels() scan labels and allocate new label based
> + * on its second parameter (num_labels). Therefore to
> + * allocate new namespace label also include previously
> + * added region label
> + */
> + rc = init_labels(nd_mapping, count + region_label_cnt);
> if (rc < 0)
> return rc;
>
> rc = __pmem_label_update(nd_region, nd_mapping, nspm, i,
> - NSLABEL_FLAG_UPDATING);
> + NSLABEL_FLAG_UPDATING, NS_LABEL_TYPE);
The amount of shared code in __pmem_label_update() across the two types in
the union isn't that high. I'd be tempted to just split it for simplicity.
> if (rc)
> return rc;
> }
> @@ -1108,7 +1233,47 @@ int nd_pmem_namespace_label_update(struct nd_region *nd_region,
> for (i = 0; i < nd_region->ndr_mappings; i++) {
> struct nd_mapping *nd_mapping = &nd_region->mapping[i];
>
> - rc = __pmem_label_update(nd_region, nd_mapping, nspm, i, 0);
> + rc = __pmem_label_update(nd_region, nd_mapping, nspm, i, 0,
> + NS_LABEL_TYPE);
> + if (rc)
> + return rc;
> + }
> +
> + return 0;
> +}
> +
> +int nd_pmem_region_label_update(struct nd_region *nd_region)
> +{
> + int i, rc;
> +
> + for (i = 0; i < nd_region->ndr_mappings; i++) {
> + struct nd_mapping *nd_mapping = &nd_region->mapping[i];
> + struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
> + int region_label_cnt = 0;
> +
> + /* No need to update region label for non cxl format */
> + if (!ndd->cxl)
> + return 0;
> +
> + region_label_cnt = find_region_label_count(ndd, nd_mapping);
> + rc = init_labels(nd_mapping, region_label_cnt + 1);
> + if (rc < 0)
> + return rc;
> +
> + rc = __pmem_label_update(nd_region, nd_mapping, NULL, i,
> + NSLABEL_FLAG_UPDATING, RG_LABEL_TYPE);
> + if (rc)
> + return rc;
> + }
> +
> + /* Clear the UPDATING flag per UEFI 2.7 expectations */
> + for (i = 0; i < nd_region->ndr_mappings; i++) {
> + struct nd_mapping *nd_mapping = &nd_region->mapping[i];
> + struct nvdimm_drvdata *ndd = to_ndd(nd_mapping);
> +
> + WARN_ON_ONCE(!ndd->cxl);
> + rc = __pmem_label_update(nd_region, nd_mapping, NULL, i, 0,
> + RG_LABEL_TYPE);
> if (rc)
> return rc;
> }
> diff --git a/drivers/nvdimm/label.h b/drivers/nvdimm/label.h
> index 0650fb4b9821..284e2a763b49 100644
> --- a/drivers/nvdimm/label.h
> +++ b/drivers/nvdimm/label.h
> @@ -30,6 +30,11 @@ enum {
> ND_NSINDEX_INIT = 0x1,
> };
>
> +enum label_type {
> + RG_LABEL_TYPE,
> + NS_LABEL_TYPE,
> +};
> +
> /**
> * struct nd_namespace_index - label set superblock
> * @sig: NAMESPACE_INDEX\0
> @@ -183,6 +188,15 @@ struct nd_namespace_label {
> };
> };
>
> +/*
> + * LSA 2.1 format introduces region label, which can also reside
> + * into LSA along with only namespace label as per v1.1 and v1.2
> + */
> +union nd_lsa_label {
> + struct nd_namespace_label ns_label;
> + struct cxl_region_label region_label;
> +};
> diff --git a/drivers/nvdimm/namespace_devs.c b/drivers/nvdimm/namespace_devs.c
> index 3271b1c8569a..559f822ef24f 100644
> --- a/drivers/nvdimm/namespace_devs.c
> +++ b/drivers/nvdimm/namespace_devs.c
> @@ -232,6 +232,18 @@ static ssize_t __alt_name_store(struct device *dev, const char *buf,
> return rc;
> }
>
> +int nd_region_label_update(struct nd_region *nd_region)
> +{
> + int rc;
> +
> + nvdimm_bus_lock(&nd_region->dev);
> + rc = nd_pmem_region_label_update(nd_region);
> + nvdimm_bus_unlock(&nd_region->dev);
Looks like it would be worth introducing a
DEFINE_GUARD() for this lock.
Not necessarily in this series however.
> +
> + return rc;
> +}
> +EXPORT_SYMBOL_GPL(nd_region_label_update);
> +
> static int nd_namespace_label_update(struct nd_region *nd_region,
> struct device *dev)
> {
> diff --git a/drivers/nvdimm/nd.h b/drivers/nvdimm/nd.h
> index e362611d82cc..f04c042dcfa9 100644
> --- a/drivers/nvdimm/nd.h
> +++ b/drivers/nvdimm/nd.h
> bool nsl_validate_type_guid(struct nvdimm_drvdata *ndd,
> struct nd_namespace_label *nd_label, guid_t *guid);
> enum nvdimm_claim_class nsl_get_claim_class(struct nvdimm_drvdata *ndd,
> @@ -399,7 +432,10 @@ enum nd_label_flags {
> struct nd_label_ent {
> struct list_head list;
> unsigned long flags;
> - struct nd_namespace_label *label;
> + union {
> + struct nd_namespace_label *label;
> + struct cxl_region_label *region_label;
It is a bit odd to have a union above of the two types in
here but then union the pointers here.
I'm also failing to find where region_label is used in this patch.
> + };
> };
>
> enum nd_mapping_lock_class {
_rc)
Powered by blists - more mailing lists