xref: /openbmc/linux/arch/arm64/kernel/topology.c (revision 9f99d983)
1 /*
2  * arch/arm64/kernel/topology.c
3  *
4  * Copyright (C) 2011,2013,2014 Linaro Limited.
5  *
6  * Based on the arm32 version written by Vincent Guittot in turn based on
7  * arch/sh/kernel/topology.c
8  *
9  * This file is subject to the terms and conditions of the GNU General Public
10  * License.  See the file "COPYING" in the main directory of this archive
11  * for more details.
12  */
13 
14 #include <linux/acpi.h>
15 #include <linux/arch_topology.h>
16 #include <linux/cacheinfo.h>
17 #include <linux/cpufreq.h>
18 #include <linux/init.h>
19 #include <linux/percpu.h>
20 
21 #include <asm/cpu.h>
22 #include <asm/cputype.h>
23 #include <asm/topology.h>
24 
25 void store_cpu_topology(unsigned int cpuid)
26 {
27 	struct cpu_topology *cpuid_topo = &cpu_topology[cpuid];
28 	u64 mpidr;
29 
30 	if (cpuid_topo->package_id != -1)
31 		goto topology_populated;
32 
33 	mpidr = read_cpuid_mpidr();
34 
35 	/* Uniprocessor systems can rely on default topology values */
36 	if (mpidr & MPIDR_UP_BITMASK)
37 		return;
38 
39 	/*
40 	 * This would be the place to create cpu topology based on MPIDR.
41 	 *
42 	 * However, it cannot be trusted to depict the actual topology; some
43 	 * pieces of the architecture enforce an artificial cap on Aff0 values
44 	 * (e.g. GICv3's ICC_SGI1R_EL1 limits it to 15), leading to an
45 	 * artificial cycling of Aff1, Aff2 and Aff3 values. IOW, these end up
46 	 * having absolutely no relationship to the actual underlying system
47 	 * topology, and cannot be reasonably used as core / package ID.
48 	 *
49 	 * If the MT bit is set, Aff0 *could* be used to define a thread ID, but
50 	 * we still wouldn't be able to obtain a sane core ID. This means we
51 	 * need to entirely ignore MPIDR for any topology deduction.
52 	 */
53 	cpuid_topo->thread_id  = -1;
54 	cpuid_topo->core_id    = cpuid;
55 	cpuid_topo->package_id = cpu_to_node(cpuid);
56 
57 	pr_debug("CPU%u: cluster %d core %d thread %d mpidr %#016llx\n",
58 		 cpuid, cpuid_topo->package_id, cpuid_topo->core_id,
59 		 cpuid_topo->thread_id, mpidr);
60 
61 topology_populated:
62 	update_siblings_masks(cpuid);
63 }
64 
65 #ifdef CONFIG_ACPI
66 static bool __init acpi_cpu_is_threaded(int cpu)
67 {
68 	int is_threaded = acpi_pptt_cpu_is_thread(cpu);
69 
70 	/*
71 	 * if the PPTT doesn't have thread information, assume a homogeneous
72 	 * machine and return the current CPU's thread state.
73 	 */
74 	if (is_threaded < 0)
75 		is_threaded = read_cpuid_mpidr() & MPIDR_MT_BITMASK;
76 
77 	return !!is_threaded;
78 }
79 
80 /*
81  * Propagate the topology information of the processor_topology_node tree to the
82  * cpu_topology array.
83  */
84 int __init parse_acpi_topology(void)
85 {
86 	int cpu, topology_id;
87 
88 	if (acpi_disabled)
89 		return 0;
90 
91 	for_each_possible_cpu(cpu) {
92 		int i, cache_id;
93 
94 		topology_id = find_acpi_cpu_topology(cpu, 0);
95 		if (topology_id < 0)
96 			return topology_id;
97 
98 		if (acpi_cpu_is_threaded(cpu)) {
99 			cpu_topology[cpu].thread_id = topology_id;
100 			topology_id = find_acpi_cpu_topology(cpu, 1);
101 			cpu_topology[cpu].core_id   = topology_id;
102 		} else {
103 			cpu_topology[cpu].thread_id  = -1;
104 			cpu_topology[cpu].core_id    = topology_id;
105 		}
106 		topology_id = find_acpi_cpu_topology_cluster(cpu);
107 		cpu_topology[cpu].cluster_id = topology_id;
108 		topology_id = find_acpi_cpu_topology_package(cpu);
109 		cpu_topology[cpu].package_id = topology_id;
110 
111 		i = acpi_find_last_cache_level(cpu);
112 
113 		if (i > 0) {
114 			/*
115 			 * this is the only part of cpu_topology that has
116 			 * a direct relationship with the cache topology
117 			 */
118 			cache_id = find_acpi_cpu_cache_topology(cpu, i);
119 			if (cache_id > 0)
120 				cpu_topology[cpu].llc_id = cache_id;
121 		}
122 	}
123 
124 	return 0;
125 }
126 #endif
127 
128 #ifdef CONFIG_ARM64_AMU_EXTN
129 #define read_corecnt()	read_sysreg_s(SYS_AMEVCNTR0_CORE_EL0)
130 #define read_constcnt()	read_sysreg_s(SYS_AMEVCNTR0_CONST_EL0)
131 #else
132 #define read_corecnt()	(0UL)
133 #define read_constcnt()	(0UL)
134 #endif
135 
136 #undef pr_fmt
137 #define pr_fmt(fmt) "AMU: " fmt
138 
139 static DEFINE_PER_CPU_READ_MOSTLY(unsigned long, arch_max_freq_scale);
140 static DEFINE_PER_CPU(u64, arch_const_cycles_prev);
141 static DEFINE_PER_CPU(u64, arch_core_cycles_prev);
142 static cpumask_var_t amu_fie_cpus;
143 
144 void update_freq_counters_refs(void)
145 {
146 	this_cpu_write(arch_core_cycles_prev, read_corecnt());
147 	this_cpu_write(arch_const_cycles_prev, read_constcnt());
148 }
149 
150 static inline bool freq_counters_valid(int cpu)
151 {
152 	if ((cpu >= nr_cpu_ids) || !cpumask_test_cpu(cpu, cpu_present_mask))
153 		return false;
154 
155 	if (!cpu_has_amu_feat(cpu)) {
156 		pr_debug("CPU%d: counters are not supported.\n", cpu);
157 		return false;
158 	}
159 
160 	if (unlikely(!per_cpu(arch_const_cycles_prev, cpu) ||
161 		     !per_cpu(arch_core_cycles_prev, cpu))) {
162 		pr_debug("CPU%d: cycle counters are not enabled.\n", cpu);
163 		return false;
164 	}
165 
166 	return true;
167 }
168 
169 static int freq_inv_set_max_ratio(int cpu, u64 max_rate, u64 ref_rate)
170 {
171 	u64 ratio;
172 
173 	if (unlikely(!max_rate || !ref_rate)) {
174 		pr_debug("CPU%d: invalid maximum or reference frequency.\n",
175 			 cpu);
176 		return -EINVAL;
177 	}
178 
179 	/*
180 	 * Pre-compute the fixed ratio between the frequency of the constant
181 	 * reference counter and the maximum frequency of the CPU.
182 	 *
183 	 *			    ref_rate
184 	 * arch_max_freq_scale =   ---------- * SCHED_CAPACITY_SCALE²
185 	 *			    max_rate
186 	 *
187 	 * We use a factor of 2 * SCHED_CAPACITY_SHIFT -> SCHED_CAPACITY_SCALE²
188 	 * in order to ensure a good resolution for arch_max_freq_scale for
189 	 * very low reference frequencies (down to the KHz range which should
190 	 * be unlikely).
191 	 */
192 	ratio = ref_rate << (2 * SCHED_CAPACITY_SHIFT);
193 	ratio = div64_u64(ratio, max_rate);
194 	if (!ratio) {
195 		WARN_ONCE(1, "Reference frequency too low.\n");
196 		return -EINVAL;
197 	}
198 
199 	per_cpu(arch_max_freq_scale, cpu) = (unsigned long)ratio;
200 
201 	return 0;
202 }
203 
204 static void amu_scale_freq_tick(void)
205 {
206 	u64 prev_core_cnt, prev_const_cnt;
207 	u64 core_cnt, const_cnt, scale;
208 
209 	prev_const_cnt = this_cpu_read(arch_const_cycles_prev);
210 	prev_core_cnt = this_cpu_read(arch_core_cycles_prev);
211 
212 	update_freq_counters_refs();
213 
214 	const_cnt = this_cpu_read(arch_const_cycles_prev);
215 	core_cnt = this_cpu_read(arch_core_cycles_prev);
216 
217 	if (unlikely(core_cnt <= prev_core_cnt ||
218 		     const_cnt <= prev_const_cnt))
219 		return;
220 
221 	/*
222 	 *	    /\core    arch_max_freq_scale
223 	 * scale =  ------- * --------------------
224 	 *	    /\const   SCHED_CAPACITY_SCALE
225 	 *
226 	 * See validate_cpu_freq_invariance_counters() for details on
227 	 * arch_max_freq_scale and the use of SCHED_CAPACITY_SHIFT.
228 	 */
229 	scale = core_cnt - prev_core_cnt;
230 	scale *= this_cpu_read(arch_max_freq_scale);
231 	scale = div64_u64(scale >> SCHED_CAPACITY_SHIFT,
232 			  const_cnt - prev_const_cnt);
233 
234 	scale = min_t(unsigned long, scale, SCHED_CAPACITY_SCALE);
235 	this_cpu_write(arch_freq_scale, (unsigned long)scale);
236 }
237 
238 static struct scale_freq_data amu_sfd = {
239 	.source = SCALE_FREQ_SOURCE_ARCH,
240 	.set_freq_scale = amu_scale_freq_tick,
241 };
242 
243 static void amu_fie_setup(const struct cpumask *cpus)
244 {
245 	int cpu;
246 
247 	/* We are already set since the last insmod of cpufreq driver */
248 	if (unlikely(cpumask_subset(cpus, amu_fie_cpus)))
249 		return;
250 
251 	for_each_cpu(cpu, cpus) {
252 		if (!freq_counters_valid(cpu) ||
253 		    freq_inv_set_max_ratio(cpu,
254 					   cpufreq_get_hw_max_freq(cpu) * 1000,
255 					   arch_timer_get_rate()))
256 			return;
257 	}
258 
259 	cpumask_or(amu_fie_cpus, amu_fie_cpus, cpus);
260 
261 	topology_set_scale_freq_source(&amu_sfd, amu_fie_cpus);
262 
263 	pr_debug("CPUs[%*pbl]: counters will be used for FIE.",
264 		 cpumask_pr_args(cpus));
265 }
266 
267 static int init_amu_fie_callback(struct notifier_block *nb, unsigned long val,
268 				 void *data)
269 {
270 	struct cpufreq_policy *policy = data;
271 
272 	if (val == CPUFREQ_CREATE_POLICY)
273 		amu_fie_setup(policy->related_cpus);
274 
275 	/*
276 	 * We don't need to handle CPUFREQ_REMOVE_POLICY event as the AMU
277 	 * counters don't have any dependency on cpufreq driver once we have
278 	 * initialized AMU support and enabled invariance. The AMU counters will
279 	 * keep on working just fine in the absence of the cpufreq driver, and
280 	 * for the CPUs for which there are no counters available, the last set
281 	 * value of arch_freq_scale will remain valid as that is the frequency
282 	 * those CPUs are running at.
283 	 */
284 
285 	return 0;
286 }
287 
288 static struct notifier_block init_amu_fie_notifier = {
289 	.notifier_call = init_amu_fie_callback,
290 };
291 
292 static int __init init_amu_fie(void)
293 {
294 	int ret;
295 
296 	if (!zalloc_cpumask_var(&amu_fie_cpus, GFP_KERNEL))
297 		return -ENOMEM;
298 
299 	ret = cpufreq_register_notifier(&init_amu_fie_notifier,
300 					CPUFREQ_POLICY_NOTIFIER);
301 	if (ret)
302 		free_cpumask_var(amu_fie_cpus);
303 
304 	return ret;
305 }
306 core_initcall(init_amu_fie);
307 
308 #ifdef CONFIG_ACPI_CPPC_LIB
309 #include <acpi/cppc_acpi.h>
310 
311 static void cpu_read_corecnt(void *val)
312 {
313 	*(u64 *)val = read_corecnt();
314 }
315 
316 static void cpu_read_constcnt(void *val)
317 {
318 	*(u64 *)val = read_constcnt();
319 }
320 
321 static inline
322 int counters_read_on_cpu(int cpu, smp_call_func_t func, u64 *val)
323 {
324 	/*
325 	 * Abort call on counterless CPU or when interrupts are
326 	 * disabled - can lead to deadlock in smp sync call.
327 	 */
328 	if (!cpu_has_amu_feat(cpu))
329 		return -EOPNOTSUPP;
330 
331 	if (WARN_ON_ONCE(irqs_disabled()))
332 		return -EPERM;
333 
334 	smp_call_function_single(cpu, func, val, 1);
335 
336 	return 0;
337 }
338 
339 /*
340  * Refer to drivers/acpi/cppc_acpi.c for the description of the functions
341  * below.
342  */
343 bool cpc_ffh_supported(void)
344 {
345 	return freq_counters_valid(get_cpu_with_amu_feat());
346 }
347 
348 int cpc_read_ffh(int cpu, struct cpc_reg *reg, u64 *val)
349 {
350 	int ret = -EOPNOTSUPP;
351 
352 	switch ((u64)reg->address) {
353 	case 0x0:
354 		ret = counters_read_on_cpu(cpu, cpu_read_corecnt, val);
355 		break;
356 	case 0x1:
357 		ret = counters_read_on_cpu(cpu, cpu_read_constcnt, val);
358 		break;
359 	}
360 
361 	if (!ret) {
362 		*val &= GENMASK_ULL(reg->bit_offset + reg->bit_width - 1,
363 				    reg->bit_offset);
364 		*val >>= reg->bit_offset;
365 	}
366 
367 	return ret;
368 }
369 
370 int cpc_write_ffh(int cpunum, struct cpc_reg *reg, u64 val)
371 {
372 	return -EOPNOTSUPP;
373 }
374 #endif /* CONFIG_ACPI_CPPC_LIB */
375