NVDIMM Device and Persistent Memory development
 help / color / mirror / Atom feed
From: "Verma, Vishal L" <vishal.l.verma@intel.com>
To: "Jiang, Dave" <dave.jiang@intel.com>,
	"linux-cxl@vger.kernel.org" <linux-cxl@vger.kernel.org>,
	"nvdimm@lists.linux.dev" <nvdimm@lists.linux.dev>
Cc: "Schofield, Alison" <alison.schofield@intel.com>
Subject: Re: [NDCTL PATCH v8 3/4] ndctl: cxl: add QoS class check for CXL region creation
Date: Fri, 1 Mar 2024 23:27:37 +0000	[thread overview]
Message-ID: <95bdc736c3eca0fb49de789ca72b022d0f08ffd6.camel@intel.com> (raw)
In-Reply-To: <20240301223736.1380778-4-dave.jiang@intel.com>

On Fri, 2024-03-01 at 15:36 -0700, Dave Jiang wrote:
> The CFMWS provides a QTG ID. The kernel driver creates a root decoder that
> represents the CFMWS. A qos_class attribute is exported via sysfs for the root
> decoder.
> 
> One or more qos_class tokens are retrieved via QTG ID _DSM from the ACPI0017
> device for a CXL memory device. The input for the _DSM is the read and write
> latency and bandwidth for the path between the device and the CPU. The
> numbers are constructed by the kernel driver for the _DSM input. When a
> device is probed, QoS class tokens  are retrieved. This is useful for a
> hot-plugged CXL memory device that does not have regions created.
> 
> Add a QoS check during region creation. If --enforce-qos/-Q is set and
> the qos_class mismatches, the region creation will fail.
> 
> Reviewed-by: Alison Schofield <alison.schofield@intel.com>
> Signed-off-by: Dave Jiang <dave.jiang@intel.com>
> ---
> v8:
> - Add qos_class comparison in libcxl instead of user set flag. (Vishal)
> - Move check into validate_decoder(). (Vishal)
> ---
>  Documentation/cxl/cxl-create-region.txt |  6 +++
>  cxl/json.c                              |  6 +++
>  cxl/lib/libcxl.c                        | 29 ++++++++++++++
>  cxl/lib/libcxl.sym                      |  1 +
>  cxl/libcxl.h                            |  1 +
>  cxl/region.c                            | 53 ++++++++++++++++++++++++-
>  6 files changed, 95 insertions(+), 1 deletion(-)
> 
> diff --git a/Documentation/cxl/cxl-create-region.txt b/Documentation/cxl/cxl-create-region.txt
> index f11a412bddfe..b244af60b8a6 100644
> --- a/Documentation/cxl/cxl-create-region.txt
> +++ b/Documentation/cxl/cxl-create-region.txt
> @@ -105,6 +105,12 @@ include::bus-option.txt[]
>  	supplied, the first cross-host bridge (if available), decoder that
>  	supports the largest interleave will be chosen.
>  
> +-Q::
> +--enforce-qos::
> +	Parameter to enforce qos_class mismatch failure. Region create operation
> +	will fail of the qos_class of the root decoder and one of the memdev that
> +	backs the region mismatches.
> +
>  include::human-option.txt[]
>  
>  include::debug-option.txt[]
> diff --git a/cxl/json.c b/cxl/json.c
> index c8bd8c27447a..b9ed62abe0e2 100644
> --- a/cxl/json.c
> +++ b/cxl/json.c
> @@ -1238,6 +1238,12 @@ struct json_object *util_cxl_region_to_json(struct cxl_region *region,
>  		}
>  	}
>  
> +	if (!cxl_region_qos_class_matches(region)) {
> +		jobj = json_object_new_boolean(true);
> +		if (jobj)
> +			json_object_object_add(jregion, "qos_class_mismatches", jobj);

Fairly minor but I think 'mismatched' (i.e. present continuous) for a
json attribute reads weirdly to me. Prefer it to be just
'qos_class_mismatch'

> +	}
> +
>  	json_object_set_userdata(jregion, region, NULL);
>  
>  
> diff --git a/cxl/lib/libcxl.c b/cxl/lib/libcxl.c
> index 6c293f1dfc91..3ca0dca8db52 100644
> --- a/cxl/lib/libcxl.c
> +++ b/cxl/lib/libcxl.c
> @@ -414,6 +414,35 @@ CXL_EXPORT int cxl_region_is_enabled(struct cxl_region *region)
>  	return is_enabled(path);
>  }
>  
> +CXL_EXPORT bool cxl_region_qos_class_matches(struct cxl_region *region)

..and since the primary purpose of this API is to see if there is a
mismatch, I think it would be nice to invert the logic, and call it
cxl_region_qos_class_mismatch()

> +{
> +	struct cxl_decoder *root_decoder = cxl_region_get_decoder(region);
> +	struct cxl_memdev_mapping *mapping;
> +
> +	cxl_mapping_foreach(region, mapping) {
> +		struct cxl_decoder *decoder;
> +		struct cxl_memdev *memdev;
> +
> +		decoder = cxl_mapping_get_decoder(mapping);
> +		if (!decoder)
> +			continue;
> +
> +		memdev = cxl_decoder_get_memdev(decoder);
> +		if (!memdev)
> +			continue;
> +
> +		if (region->mode == CXL_DECODER_MODE_RAM) {
> +			if (root_decoder->qos_class != memdev->ram_qos_class)
> +				return false;
> +		} else if (region->mode == CXL_DECODER_MODE_PMEM) {
> +			if (root_decoder->qos_class != memdev->pmem_qos_class)
> +				return false;
> +		}
> +	}
> +
> +	return true;
> +}
> +
>  CXL_EXPORT int cxl_region_disable(struct cxl_region *region)
>  {
>  	const char *devname = cxl_region_get_devname(region);
> diff --git a/cxl/lib/libcxl.sym b/cxl/lib/libcxl.sym
> index 465c78dc6c70..a203b3ce3976 100644
> --- a/cxl/lib/libcxl.sym
> +++ b/cxl/lib/libcxl.sym
> @@ -285,4 +285,5 @@ global:
>  	cxl_root_decoder_get_qos_class;
>  	cxl_memdev_get_pmem_qos_class;
>  	cxl_memdev_get_ram_qos_class;
> +	cxl_region_qos_class_matches;
>  } LIBCXL_7;
> diff --git a/cxl/libcxl.h b/cxl/libcxl.h
> index a180f01cb05e..bd8e924844ff 100644
> --- a/cxl/libcxl.h
> +++ b/cxl/libcxl.h
> @@ -335,6 +335,7 @@ int cxl_region_clear_target(struct cxl_region *region, int position);
>  int cxl_region_clear_all_targets(struct cxl_region *region);
>  int cxl_region_decode_commit(struct cxl_region *region);
>  int cxl_region_decode_reset(struct cxl_region *region);
> +bool cxl_region_qos_class_matches(struct cxl_region *region);
>  
>  #define cxl_region_foreach(decoder, region)                                    \
>  	for (region = cxl_region_get_first(decoder); region != NULL;           \
> diff --git a/cxl/region.c b/cxl/region.c
> index 3a762db4800e..735b3be7759b 100644
> --- a/cxl/region.c
> +++ b/cxl/region.c
> @@ -32,6 +32,7 @@ static struct region_params {
>  	bool force;
>  	bool human;
>  	bool debug;
> +	bool qos_enforce;

Minor nit: just for parity, since the option is --enforce-qos, it would
be nice to call this enforce_qos (and down below as well).

>  } param = {
>  	.ways = INT_MAX,
>  	.granularity = INT_MAX,
> @@ -49,6 +50,7 @@ struct parsed_params {
>  	const char **argv;
>  	struct cxl_decoder *root_decoder;
>  	enum cxl_decoder_mode mode;
> +	bool qos_enforce;
>  };
>  
>  enum region_actions {
> @@ -81,7 +83,8 @@ OPT_STRING('U', "uuid", &param.uuid, \
>  	   "region uuid", "uuid for the new region (default: autogenerate)"), \
>  OPT_BOOLEAN('m', "memdevs", &param.memdevs, \
>  	    "non-option arguments are memdevs"), \
> -OPT_BOOLEAN('u', "human", &param.human, "use human friendly number formats")
> +OPT_BOOLEAN('u', "human", &param.human, "use human friendly number formats"), \
> +OPT_BOOLEAN('Q', "enforce-qos", &param.qos_enforce, "enforce of qos_class matching")
>  
>  static const struct option create_options[] = {
>  	BASE_OPTIONS(),
> @@ -360,6 +363,8 @@ static int parse_create_options(struct cxl_ctx *ctx, int count,
>  		}
>  	}
>  
> +	p->qos_enforce = param.qos_enforce;
> +
>  	return 0;
>  
>  err:
> @@ -423,10 +428,52 @@ static void collect_minsize(struct cxl_ctx *ctx, struct parsed_params *p)
>  	}
>  }
>  
> +static int create_region_validate_qos_class(struct parsed_params *p)
> +{
> +	int root_qos_class;
> +	int qos_class;
> +	int i;
> +
> +	if (!p->qos_enforce)
> +		return 0;
> +
> +	root_qos_class = cxl_root_decoder_get_qos_class(p->root_decoder);
> +	if (root_qos_class == CXL_QOS_CLASS_NONE)
> +		return 0;
> +
> +	for (i = 0; i < p->ways; i++) {
> +		struct json_object *jobj =
> +			json_object_array_get_idx(p->memdevs, i);
> +		struct cxl_memdev *memdev = json_object_get_userdata(jobj);
> +
> +		if (p->mode == CXL_DECODER_MODE_RAM)
> +			qos_class = cxl_memdev_get_ram_qos_class(memdev);
> +		else
> +			qos_class = cxl_memdev_get_pmem_qos_class(memdev);
> +
> +		/* No qos_class entries. Possibly no kernel support */
> +		if (qos_class == CXL_QOS_CLASS_NONE)
> +			break;
> +
> +		if (qos_class != root_qos_class) {
> +			if (p->qos_enforce) {
> +				log_err(&rl, "%s QoS Class mismatches %s\n",

Probably s/mismatches/mismatch/ here too.

> +					cxl_decoder_get_devname(p->root_decoder),
> +					cxl_memdev_get_devname(memdev));
> +
> +				return -ENXIO;
> +			}
> +		}
> +	}
> +
> +	return 0;
> +}
> +
>  static int validate_decoder(struct cxl_decoder *decoder,
>  			    struct parsed_params *p)
>  {
>  	const char *devname = cxl_decoder_get_devname(decoder);
> +	int rc;
>  
>  	switch(p->mode) {
>  	case CXL_DECODER_MODE_RAM:
> @@ -446,6 +493,10 @@ static int validate_decoder(struct cxl_decoder *decoder,
>  		return -EINVAL;
>  	}
>  
> +	rc = create_region_validate_qos_class(p);
> +	if (rc)
> +		return rc;
> +
>  	/* TODO check if the interleave config is possible under this decoder */
>  
>  	return 0;


  reply	other threads:[~2024-03-01 23:27 UTC|newest]

Thread overview: 6+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2024-03-01 22:36 [NDCTL PATCH v8 4/4] ndctl: Add support of qos_class for CXL CLI Dave Jiang
2024-03-01 22:36 ` [NDCTL PATCH v8 1/4] ndctl: cxl: Add QoS class retrieval for the root decoder Dave Jiang
2024-03-01 22:36 ` [NDCTL PATCH v8 2/4] ndctl: cxl: Add QoS class support for the memory device Dave Jiang
2024-03-01 22:36 ` [NDCTL PATCH v8 3/4] ndctl: cxl: add QoS class check for CXL region creation Dave Jiang
2024-03-01 23:27   ` Verma, Vishal L [this message]
2024-03-01 22:36 ` [NDCTL PATCH v8 4/4] ndctl: add test for qos_class in CXL test suite Dave Jiang

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=95bdc736c3eca0fb49de789ca72b022d0f08ffd6.camel@intel.com \
    --to=vishal.l.verma@intel.com \
    --cc=alison.schofield@intel.com \
    --cc=dave.jiang@intel.com \
    --cc=linux-cxl@vger.kernel.org \
    --cc=nvdimm@lists.linux.dev \
    /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 read-only IMAP folder(s) and NNTP newsgroup(s).