1 // SPDX-License-Identifier: GPL-2.0-or-later
9 #define pr_fmt(fmt) "hv-24x7: " fmt
24 #include "hv-24x7.h"
25 #include "hv-24x7-catalog.h"
26 #include "hv-common.h"
36 static bool domain_is_valid(unsigned domain)  in domain_is_valid()  argument
38 	switch (domain) {  in domain_is_valid()
39 #define DOMAIN(n, v, x, c)		\  in domain_is_valid()  macro
42 #include "hv-24x7-domains.h"  in domain_is_valid()
43 #undef DOMAIN  in domain_is_valid()
50 static bool is_physical_domain(unsigned domain)  in is_physical_domain()  argument
52 	switch (domain) {  in is_physical_domain()
53 #define DOMAIN(n, v, x, c)		\  in is_physical_domain()  macro
56 #include "hv-24x7-domains.h"  in is_physical_domain()
57 #undef DOMAIN  in is_physical_domain()
78  * chip details through the get-system-parameter rtas call.
94 	call_status = rtas_call(rtas_token("ibm,get-system-parameter"), 3, 1,  in read_24x7_sys_info()
101 		pr_err("Error calling get-system-parameter %d\n",  in read_24x7_sys_info()
123 static bool domain_needs_aggregation(unsigned int domain)  in domain_needs_aggregation()  argument
126 			(domain == HV_PERF_DOMAIN_PHYS_CORE ||  in domain_needs_aggregation()
127 			 (domain >= HV_PERF_DOMAIN_VCPU_HOME_CORE &&  in domain_needs_aggregation()
128 			  domain <= HV_PERF_DOMAIN_VCPU_REMOTE_NODE));  in domain_needs_aggregation()
131 static const char *domain_name(unsigned domain)  in domain_name()  argument
133 	if (!domain_is_valid(domain))  in domain_name()
136 	switch (domain) {  in domain_name()
138 	case HV_PERF_DOMAIN_PHYS_CORE:		return "Physical Core";  in domain_name()
139 	case HV_PERF_DOMAIN_VCPU_HOME_CORE:	return "VCPU Home Core";  in domain_name()
145 	WARN_ON_ONCE(domain);  in domain_name()
149 static bool catalog_entry_domain_is_valid(unsigned domain)  in catalog_entry_domain_is_valid()  argument
153 		return is_physical_domain(domain);  in catalog_entry_domain_is_valid()
155 		return domain_is_valid(domain);  in catalog_entry_domain_is_valid()
160  * - Think of the hcall as an interface to a 4d array of counters:
161  *   - x = domains
162  *   - y = indexes in the domain (core, chip, vcpu, node, etc)
163  *   - z = offset into the counter space
164  *   - w = lpars (guest vms, "logical partitions")
165  * - A single request is: x,y,y_last,z,z_last,w,w_last
166  *   - this means we can retrieve a rectangle of counters in y,z for a single x.
168  * - Things to consider (ignoring w):
169  *   - input  cost_per_request = 16
170  *   - output cost_per_result(ys,zs)  = 8 + 8 * ys + ys * zs
171  *   - limited number of requests per hcall (must fit into 4K bytes)
172  *     - 4k = 16 [buffer header] - 16 [request size] * request_count
173  *     - 255 requests per hcall
174  *   - sometimes it will be more efficient to read extra data and discard
179  *  perf stat -e 'hv_24x7/domain=2,offset=8,vcpu=0,lpar=0xffffffff/'
182 /* u3 0-6, one of HV_24X7_PERF_DOMAIN */
183 EVENT_DEFINE_RANGE_FORMAT(domain, config, 0, 3);
185 EVENT_DEFINE_RANGE_FORMAT(core, config, 16, 31);
249 	return (H24x7_DATA_BUFFER_SIZE - sizeof(struct hv_24x7_request_buffer))  in max_num_requests()
255 	*len = be16_to_cpu(ev->event_name_len) - 2;  in event_name()
256 	return (char *)ev->remainder;  in event_name()
261 	unsigned nl = be16_to_cpu(ev->event_name_len);  in event_desc()
262 	__be16 *desc_len = (__be16 *)(ev->remainder + nl - 2);  in event_desc()
264 	*len = be16_to_cpu(*desc_len) - 2;  in event_desc()
265 	return (char *)ev->remainder + nl;  in event_desc()
270 	unsigned nl = be16_to_cpu(ev->event_name_len);  in event_long_desc()
271 	__be16 *desc_len_ = (__be16 *)(ev->remainder + nl - 2);  in event_long_desc()
273 	__be16 *long_desc_len = (__be16 *)(ev->remainder + nl + desc_len - 2);  in event_long_desc()
275 	*len = be16_to_cpu(*long_desc_len) - 2;  in event_long_desc()
276 	return (char *)ev->remainder + nl + desc_len;  in event_long_desc()
289  *  - padding for desc, name, and long/detailed desc is required to be '\0'
300 	unsigned nl = be16_to_cpu(ev->event_name_len);  in event_end()
313 	dl_ = (__be16 *)(ev->remainder + nl - 2);  in event_end()
328 	ldl_ = (__be16 *)(ev->remainder + nl + dl - 2);  in event_end()
367  * data for this sysfs entry based on the event's domain.
369  * Events belonging to the Chip domain can only be monitored in that domain.
370  * i.e the domain for these events is a fixed/knwon value.
372  * Events belonging to the Core domain can be monitored either in the physical
373  * core or in one of the virtual CPU domains. So the domain value for these
375  * the Core events with 'domain=?' so the perf-tool can error check required
378  * NOTE: For the Core domain events, rather than making domain a required
380  *	 override the domain to one of the VCPU domains.
384  *	 If we set domain=2 (PHYS_CHIP) and allow user to override this field
389  *		perf stat -e hv_24x7/HPM_PCYC,offset=0x20/
393  *	By not assigning a default value to the domain for the Core events,
396  *		- Specifying values for parameters with "=?" is required.
398  *		- Specifying (i.e overriding) values for other parameters
401 static char *event_fmt(struct hv_24x7_event_data *event, unsigned domain)  in event_fmt()  argument
408 	switch (domain) {  in event_fmt()
410 		snprintf(buf, sizeof(buf), "%d", domain);  in event_fmt()
418 		sindex = "core";  in event_fmt()
427 			"domain=%s,offset=0x%x,%s=?,lpar=%s",  in event_fmt()
429 			be16_to_cpu(event->event_counter_offs) +  in event_fmt()
430 				be16_to_cpu(event->event_group_record_offs),  in event_fmt()
448 	return sprintf(buf, "%s\n", (char *)d->var);  in device_show_string()
482 	sysfs_attr_init(&attr->attr.attr);  in device_str_attr_create_()
484 	attr->var = str;  in device_str_attr_create_()
485 	attr->attr.attr.name = name;  in device_str_attr_create_()
486 	attr->attr.attr.mode = 0444;  in device_str_attr_create_()
487 	attr->attr.show = device_show_string;  in device_str_attr_create_()
489 	return &attr->attr.attr;  in device_str_attr_create_()
534 				       unsigned domain,  in event_to_attr()  argument
541 	if (!domain_is_valid(domain)) {  in event_to_attr()
542 		pr_warn("catalog event %u has invalid domain %u\n",  in event_to_attr()
543 				ix, domain);  in event_to_attr()
547 	val = event_fmt(event, domain);  in event_to_attr()
605 	*attrs = event_to_attr(ix, event, event->domain, nonce);  in event_data_to_attrs()
607 		return -1;  in event_data_to_attrs()
618 	unsigned domain;  member
626 		return -1;  in memord()
641 		return -1;  in ev_uniq_ord()
646 			  unsigned domain)  in event_uniq_add()  argument
648 	struct rb_node **new = &(root->rb_node), *parent = NULL;  in event_uniq_add()
657 		result = ev_uniq_ord(name, nl, domain, it->name, it->nl,  in event_uniq_add()
658 					it->domain);  in event_uniq_add()
662 			new = &((*new)->rb_left);  in event_uniq_add()
664 			new = &((*new)->rb_right);  in event_uniq_add()
666 			it->ct++;  in event_uniq_add()
668 						name, it->ct);  in event_uniq_add()
669 			return it->ct;  in event_uniq_add()
675 		return -ENOMEM;  in event_uniq_add()
681 		.domain = domain,  in event_uniq_add()
685 	rb_link_node(&data->node, parent, new);  in event_uniq_add()
686 	rb_insert_color(&data->node, root);  in event_uniq_add()
688 	/* data->ct */  in event_uniq_add()
710  * Otherwise, return -1 (and print as appropriate).
722 		return -1;  in catalog_event_len_validate()
726 				event_data_bytes - offset);  in catalog_event_len_validate()
727 		return -1;  in catalog_event_len_validate()
733 		return -1;  in catalog_event_len_validate()
736 	ev_len = be16_to_cpu(event->length);  in catalog_event_len_validate()
747 		return -1;  in catalog_event_len_validate()
755 		return -1;  in catalog_event_len_validate()
761 		return -1;  in catalog_event_len_validate()
798 		ret = -ENOMEM;  in create_events_from_catalog()
804 		ret = -EIO;  in create_events_from_catalog()
808 	catalog_version_num = be64_to_cpu(page_0->version);  in create_events_from_catalog()
809 	catalog_page_len = be32_to_cpu(page_0->length);  in create_events_from_catalog()
813 		ret = -EIO;  in create_events_from_catalog()
819 	event_entry_count = be16_to_cpu(page_0->event_entry_count);  in create_events_from_catalog()
820 	event_data_offs   = be16_to_cpu(page_0->event_data_offs);  in create_events_from_catalog()
821 	event_data_len    = be16_to_cpu(page_0->event_data_len);  in create_events_from_catalog()
829 			|| (MAX_4K - event_data_offs < event_data_len)) {  in create_events_from_catalog()
832 		ret = -EIO;  in create_events_from_catalog()
837 		pr_err("event data %zu-%zu does not fit inside catalog 0-%zu\n",  in create_events_from_catalog()
841 		ret = -EIO;  in create_events_from_catalog()
845 	if (SIZE_MAX - 1 < event_entry_count) {  in create_events_from_catalog()
847 		ret = -EIO;  in create_events_from_catalog()
860 		ret = -ENOMEM;  in create_events_from_catalog()
880 			ret = -EIO;  in create_events_from_catalog()
892 		size_t offset = (void *)event - (void *)event_data;  in create_events_from_catalog()
909 		if (event->event_group_record_len == 0) {  in create_events_from_catalog()
916 		if (!catalog_entry_domain_is_valid(event->domain)) {  in create_events_from_catalog()
917 			pr_info("event %zu (%.*s) has invalid domain %d\n",  in create_events_from_catalog()
918 					event_idx, nl, name, event->domain);  in create_events_from_catalog()
933 		ret = -ENOMEM;  in create_events_from_catalog()
940 		ret = -ENOMEM;  in create_events_from_catalog()
947 		ret = -ENOMEM;  in create_events_from_catalog()
955 			event_idx++, ev_len = be16_to_cpu(event->length),  in create_events_from_catalog()
964 		if (event->event_group_record_len == 0)  in create_events_from_catalog()
966 		if (!catalog_entry_domain_is_valid(event->domain))  in create_events_from_catalog()
973 		nonce = event_uniq_add(&ev_uniq, name, nl, event->domain);  in create_events_from_catalog()
1038 		return -ENOMEM;  in catalog_read()
1042 		ret = -EIO;  in catalog_read()
1046 	catalog_version_num = be64_to_cpu(page_0->version);  in catalog_read()
1047 	catalog_page_len = be32_to_cpu(page_0->length);  in catalog_read()
1060 			ret = -EIO;  in catalog_read()
1065 	copy_len = 4096 - offset_in_page;  in catalog_read()
1117 		return -ENOMEM;					\
1120 		ret = -EIO;					\
1131 		(unsigned long long)be64_to_cpu(page_0->version));
1133 		(unsigned long long)be32_to_cpu(page_0->length) * 4096);
1191 	request_buffer->interface_version = interface_version;  in init_24x7_request()
1192 	/* memset above set request_buffer->num_requests to 0 */  in init_24x7_request()
1216 		req = request_buffer->requests;  in make_24x7_request()
1218 				      req->performance_domain, req->data_offset,  in make_24x7_request()
1219 				      req->starting_ix, req->starting_lpar_ix,  in make_24x7_request()
1220 				      ret, ret, result_buffer->detailed_rc,  in make_24x7_request()
1221 				      result_buffer->failing_request_ix);  in make_24x7_request()
1222 		return -EIO;  in make_24x7_request()
1243 	if (request_buffer->num_requests >=  in add_event_to_24x7_request()
1244 	    max_num_requests(request_buffer->interface_version)) {  in add_event_to_24x7_request()
1246 				request_buffer->num_requests);  in add_event_to_24x7_request()
1247 		return -EINVAL;  in add_event_to_24x7_request()
1261 	req_size = H24x7_REQUEST_SIZE(request_buffer->interface_version);  in add_event_to_24x7_request()
1263 	i = request_buffer->num_requests++;  in add_event_to_24x7_request()
1264 	req = (void *) request_buffer->requests + i * req_size;  in add_event_to_24x7_request()
1266 	req->performance_domain = event_get_domain(event);  in add_event_to_24x7_request()
1267 	req->data_size = cpu_to_be16(8);  in add_event_to_24x7_request()
1268 	req->data_offset = cpu_to_be32(event_get_offset(event));  in add_event_to_24x7_request()
1269 	req->starting_lpar_ix = cpu_to_be16(event_get_lpar(event));  in add_event_to_24x7_request()
1270 	req->max_num_lpars = cpu_to_be16(1);  in add_event_to_24x7_request()
1271 	req->starting_ix = cpu_to_be16(idx);  in add_event_to_24x7_request()
1272 	req->max_ix = cpu_to_be16(1);  in add_event_to_24x7_request()
1274 	if (request_buffer->interface_version > 1) {  in add_event_to_24x7_request()
1275 		if (domain_needs_aggregation(req->performance_domain))  in add_event_to_24x7_request()
1276 			req->max_num_thread_groups = -1;  in add_event_to_24x7_request()
1277 		else if (req->performance_domain != HV_PERF_DOMAIN_PHYS_CHIP) {  in add_event_to_24x7_request()
1278 			req->starting_thread_group_ix = idx % 2;  in add_event_to_24x7_request()
1279 			req->max_num_thread_groups = 1;  in add_event_to_24x7_request()
1287  * get_count_from_result - get event count from all result elements in result
1303 	u16 num_elements = be16_to_cpu(res->num_elements_returned);  in get_count_from_result()
1304 	u16 data_size = be16_to_cpu(res->result_element_data_size);  in get_count_from_result()
1315 			 res->result_ix);  in get_count_from_result()
1318 			*next = (struct hv_24x7_result *) res->elements;  in get_count_from_result()
1320 		return -ENODATA;  in get_count_from_result()
1331 		       res->result_ix, num_elements);  in get_count_from_result()
1333 		return -EIO;  in get_count_from_result()
1338 			 res->result_ix, data_size);  in get_count_from_result()
1340 		return -ENOTSUPP;  in get_count_from_result()
1343 	if (resb->interface_version == 1)  in get_count_from_result()
1351 	for (i = count = 0, element_data = res->elements + data_offset;  in get_count_from_result()
1360 		*next = element_data - data_offset;  in get_count_from_result()
1389 				    result_buffer->results, count, NULL);  in single_24x7_request()
1401 	unsigned domain;  in h_24x7_event_init()  local
1406 	if (event->attr.type != event->pmu->type)  in h_24x7_event_init()
1407 		return -ENOENT;  in h_24x7_event_init()
1414 				event->attr.config,  in h_24x7_event_init()
1416 				event->attr.config1,  in h_24x7_event_init()
1418 				event->attr.config2,  in h_24x7_event_init()
1420 		return -EINVAL;  in h_24x7_event_init()
1425 		return -EOPNOTSUPP;  in h_24x7_event_init()
1430 		return -EINVAL;  in h_24x7_event_init()
1433 	domain = event_get_domain(event);  in h_24x7_event_init()
1434 	if (domain >= HV_PERF_DOMAIN_MAX) {  in h_24x7_event_init()
1435 		pr_devel("invalid domain %d\n", domain);  in h_24x7_event_init()
1436 		return -EINVAL;  in h_24x7_event_init()
1442 		return -EIO;  in h_24x7_event_init()
1446 	if (!caps.collect_privileged && (is_physical_domain(domain) ||  in h_24x7_event_init()
1449 				is_physical_domain(domain),  in h_24x7_event_init()
1451 		return -EACCES;  in h_24x7_event_init()
1457 		return -EIO;  in h_24x7_event_init()
1459 	(void)local64_xchg(&event->hw.prev_count, ct);  in h_24x7_event_init()
1479 	prev = local64_xchg(&event->hw.prev_count, now);  in update_event_count()
1480 	local64_add(now - prev, &event->count);  in update_event_count()
1514 			 * so ->commit_txn() can quickly find/update count.  in h_24x7_event_read()
1516 			i = request_buffer->num_requests - 1;  in h_24x7_event_read()
1519 			h24x7hw->events[i] = event;  in h_24x7_event_read()
1533 		local64_set(&event->hw.prev_count, h_24x7_get_value(event));  in h_24x7_event_start()
1590  * and dont need/support ADD transactions. Clear ->txn_flags but otherwise
1627 	for (i = 0, res = result_buffer->results;  in h_24x7_event_commit_txn()
1628 	     i < result_buffer->num_results; i++, res = next_res) {  in h_24x7_event_commit_txn()
1629 		struct perf_event *event = h24x7hw->events[res->result_ix];  in h_24x7_event_commit_txn()
1699 		return -1;  in ppc_hv_24x7_cpu_offline()
1725 		return -ENODEV;  in hv_24x7_init()
1726 	} else if (!cur_cpu_spec->oprofile_cpu_type)  in hv_24x7_init()
1727 		return -ENODEV;  in hv_24x7_init()
1730 	if (!strcmp(cur_cpu_spec->oprofile_cpu_type, "ppc64/power8"))  in hv_24x7_init()
1744 		return -ENODEV;  in hv_24x7_init()
1747 	hv_page_cache = kmem_cache_create("hv-page-4096", 4096, 4096, 0, NULL);  in hv_24x7_init()
1749 		return -ENOMEM;  in hv_24x7_init()
1766 	r = perf_pmu_register(&h_24x7_pmu, h_24x7_pmu.name, -1);  in hv_24x7_init()