x86/resctrl: Add "*" shorthand to set io_alloc CBM for all domains

Introduce a wildcard domain ID selector "*" for the io_alloc_cbm interface. This allows a user to update the Capacity Bitmask (CBM) across all cache domains in a single operation.

Currently, configuring io_alloc_cbm requires an explicit ID for each domain, which is cumbersome on systems with high core counts and numerous cache clusters. Supporting a wildcard selector simplifies automation and management tasks.

For example, a user can now write "*=0" to the io_alloc_cbm file to program every domain to the hardware-defined minimum CBM. Note that the value provided must still adhere to the constraints defined in the resource's min_cbm_bits.

Signed-off-by Aaron Tomlin <[email protected]>

---
  Documentation/filesystems/resctrl.rst |  8 ++++
  fs/resctrl/ctrlmondata.c              | 60 ++++++++++++++-------------
  2 files changed, 40 insertions(+), 28 deletions(-)

Documentation/filesystems/resctrl.rst +0 -0

index 8c8ce678148a..e4d92c865e44 100644

Documentation/filesystems/resctrl.rst +1 -0

+
  		When CDP is enabled "io_alloc_cbm" associated with the CDP_DATA and CDP_CODE
  		resources may reflect the same values. For example, values read from and
  		written to /sys/fs/resctrl/info/L3DATA/io_alloc_cbm may be reflected by

fs/resctrl/ctrlmondata.c +29 -26

index b2d178d3556e..7b3119f91b9d 100644
@@ -873,41 +873,45 @@ static int resctrl_io_alloc_parse_line(char *line,  struct rdt_resource *r,
  	struct rdt_ctrl_domain *d;
  	char *dom = NULL, *id;
  	unsigned long dom_id;
+	bool update_all;
 
-next:
-	if (!line || line[0] == '\0')
-		return 0;
+	while (line && line[0] != '\0') {
+		update_all = false;
+		dom = strsep(&line, ";");
+		id = strsep(&dom, "=");
 
-	dom = strsep(&line, ";");
-	id = strsep(&dom, "=");
-	if (!dom || kstrtoul(id, 10, &dom_id)) {
-		rdt_last_cmd_puts("Missing '=' or non-numeric domain\n");
-		return -EINVAL;
-	}
+		if (id && !strcmp(id, "*")) {
+			update_all = true;
+		} else if (!dom || kstrtoul(id, 10, &dom_id)) {
+			rdt_last_cmd_puts("Missing '=' or non-numeric domain\n");
+			return -EINVAL;
+		}
 
-	dom = strim(dom);
-	list_for_each_entry(d, &r->ctrl_domains, hdr.list) {
-		if (d->hdr.id == dom_id) {
-			data.buf = dom;
-			data.mode = RDT_MODE_SHAREABLE;
-			data.closid = closid;
-			if (parse_cbm(&data, s, d))
-				return -EINVAL;
-			/*
-			 * Keep io_alloc CLOSID's CBM of CDP_CODE and CDP_DATA
-			 * in sync.
-			 */
-			if (resctrl_arch_get_cdp_enabled(r->rid)) {
-				peer_type = resctrl_peer_type(s->conf_type);
-				memcpy(&d->staged_config[peer_type],
-				       &d->staged_config[s->conf_type],
-				       sizeof(d->staged_config[0]));
+		dom = strim(dom);
+		list_for_each_entry(d, &r->ctrl_domains, hdr.list) {
+			if (update_all || d->hdr.id == dom_id) {
+				data.buf = dom;
+				data.mode = RDT_MODE_SHAREABLE;
+				data.closid = closid;
+				if (parse_cbm(&data, s, d))
+					return -EINVAL;
+				/*
+				 * Keep io_alloc CLOSID's CBM of CDP_CODE and CDP_DATA
+				 * in sync.
+				 */
+				if (resctrl_arch_get_cdp_enabled(r->rid)) {
+					peer_type = resctrl_peer_type(s->conf_type);
+					d->staged_config[peer_type] =
+						d->staged_config[s->conf_type];
+				}
+

fs/resctrl/ctrlmondata.c +0 -0

  ssize_t resctrl_io_alloc_cbm_write(struct kernfs_open_file *of, char *buf,

To be consistent add the output also.

 # cat /sys/fs/resctrl/info/L3/io_alloc_cbm
     0=0;1=0

Tested the patch and looks good to me.

Introduce a wildcard domain ID selector "*" for the io_alloc_cbm interface. This allows a user to update the Capacity Bitmask (CBM) across all cache domains in a single operation.

Currently, configuring io_alloc_cbm requires an explicit ID for each domain, which is cumbersome on systems with high core counts and numerous cache clusters. Supporting a wildcard selector simplifies automation and management tasks.

For example, a user can now write "*=0" to the io_alloc_cbm file to program every domain to the hardware-defined minimum CBM. Note that the value provided must still adhere to the constraints defined in the resource's min_cbm_bits.

Signed-off-by Aaron Tomlin <[email protected]>

---
 Documentation/filesystems/resctrl.rst |  8 ++++
 fs/resctrl/ctrlmondata.c              | 60 ++++++++++++++-------------
 2 files changed, 40 insertions(+), 28 deletions(-)

Documentation/filesystems/resctrl.rst +0 -0

index 8c8ce678148a..e4d92c865e44 100644

Documentation/filesystems/resctrl.rst +5 -0

+
+		Example::
+
+			# echo "*=0" > /sys/fs/resctrl/info/L3/io_alloc_cbm
+
 		When CDP is enabled "io_alloc_cbm" associated with the CDP_DATA and CDP_CODE
 		resources may reflect the same values. For example, values read from and
 		written to /sys/fs/resctrl/info/L3DATA/io_alloc_cbm may be reflected by

fs/resctrl/ctrlmondata.c +0 -0

index b2d178d3556e..7b3119f91b9d 100644

nit: the value is not "required". Compare with, for example, "An ID of '*' configures all domains with provided CBM"

Why is it necessary to change to a while loop? From what I can tell this is not required to support this particular feature and thus makes this patch consist of two logical changes instead of one (see "Separate your changes" in Documentation/process/submitting-patches.rst). Since the focus of this patch is to support "*" the switch to a while loop distracts from the core change due to generating so many line changes.

I have two concerns with a switch to a while loop:

The switch to the while loop has a subtle but significant consequence that changes the behavior of this flow to return "success" when the user provides an invalid domain ID. The caller, resctrl_io_alloc_cbm_write(), expects "success" to mean that the configuration has been staged and can be configured on hardware. Now it may be possible that hardware configuration attempted without a configuration staged?

Apart from the user interface change the switch to the while loop changes the common pattern in resctrl used for parsing similar content from user space. See fs/resctrl/ctrlmondata.c:parse_line(), fs/resctrl/monitor.c:resctrl_parse_mbm_assignment(), fs/resctrl/rdtgroup.c:mon_config_write(). When working with resctrl I prefer to use common patterns. If there is an issue with a particular pattern then all instances should be changed.

Please just change the code needed to support the new feature.

Have you tried this feature with some edge cases? When considering, for example, a user just providing "*" then from this point on "dom" is assumed to point to a valid CBM but is actually NULL. Since this parses user input it is required to be robust against any input.