1 /*
2  * Copyright (c) 2014 Samsung Electronics Co., Ltd.
3  *		http://www.samsung.com
4  *
5  * arch/arm/mach-exynos/mcpm-exynos.c
6  *
7  * Based on arch/arm/mach-vexpress/dcscb.c
8  *
9  * This program is free software; you can redistribute it and/or modify
10  * it under the terms of the GNU General Public License version 2 as
11  * published by the Free Software Foundation.
12  */
13 
14 #include <linux/arm-cci.h>
15 #include <linux/delay.h>
16 #include <linux/io.h>
17 #include <linux/of_address.h>
18 
19 #include <asm/cputype.h>
20 #include <asm/cp15.h>
21 #include <asm/mcpm.h>
22 
23 #include "regs-pmu.h"
24 #include "common.h"
25 
26 #define EXYNOS5420_CPUS_PER_CLUSTER	4
27 #define EXYNOS5420_NR_CLUSTERS		2
28 #define MCPM_BOOT_ADDR_OFFSET		0x1c
29 
30 /*
31  * The common v7_exit_coherency_flush API could not be used because of the
32  * Erratum 799270 workaround. This macro is the same as the common one (in
33  * arch/arm/include/asm/cacheflush.h) except for the erratum handling.
34  */
35 #define exynos_v7_exit_coherency_flush(level) \
36 	asm volatile( \
37 	"stmfd	sp!, {fp, ip}\n\t"\
38 	"mrc	p15, 0, r0, c1, c0, 0	@ get SCTLR\n\t" \
39 	"bic	r0, r0, #"__stringify(CR_C)"\n\t" \
40 	"mcr	p15, 0, r0, c1, c0, 0	@ set SCTLR\n\t" \
41 	"isb\n\t"\
42 	"bl	v7_flush_dcache_"__stringify(level)"\n\t" \
43 	"clrex\n\t"\
44 	"mrc	p15, 0, r0, c1, c0, 1	@ get ACTLR\n\t" \
45 	"bic	r0, r0, #(1 << 6)	@ disable local coherency\n\t" \
46 	/* Dummy Load of a device register to avoid Erratum 799270 */ \
47 	"ldr	r4, [%0]\n\t" \
48 	"and	r4, r4, #0\n\t" \
49 	"orr	r0, r0, r4\n\t" \
50 	"mcr	p15, 0, r0, c1, c0, 1	@ set ACTLR\n\t" \
51 	"isb\n\t" \
52 	"dsb\n\t" \
53 	"ldmfd	sp!, {fp, ip}" \
54 	: \
55 	: "Ir" (S5P_INFORM0) \
56 	: "r0", "r1", "r2", "r3", "r4", "r5", "r6", "r7", \
57 	  "r9", "r10", "lr", "memory")
58 
59 /*
60  * We can't use regular spinlocks. In the switcher case, it is possible
61  * for an outbound CPU to call power_down() after its inbound counterpart
62  * is already live using the same logical CPU number which trips lockdep
63  * debugging.
64  */
65 static arch_spinlock_t exynos_mcpm_lock = __ARCH_SPIN_LOCK_UNLOCKED;
66 static int
67 cpu_use_count[EXYNOS5420_CPUS_PER_CLUSTER][EXYNOS5420_NR_CLUSTERS];
68 
69 #define exynos_cluster_usecnt(cluster) \
70 	(cpu_use_count[0][cluster] +   \
71 	 cpu_use_count[1][cluster] +   \
72 	 cpu_use_count[2][cluster] +   \
73 	 cpu_use_count[3][cluster])
74 
75 #define exynos_cluster_unused(cluster) !exynos_cluster_usecnt(cluster)
76 
77 static int exynos_cluster_power_control(unsigned int cluster, int enable)
78 {
79 	unsigned int tries = 100;
80 	unsigned int val;
81 
82 	if (enable) {
83 		exynos_cluster_power_up(cluster);
84 		val = S5P_CORE_LOCAL_PWR_EN;
85 	} else {
86 		exynos_cluster_power_down(cluster);
87 		val = 0;
88 	}
89 
90 	/* Wait until cluster power control is applied */
91 	while (tries--) {
92 		if (exynos_cluster_power_state(cluster) == val)
93 			return 0;
94 
95 		cpu_relax();
96 	}
97 	pr_debug("timed out waiting for cluster %u to power %s\n", cluster,
98 		enable ? "on" : "off");
99 
100 	return -ETIMEDOUT;
101 }
102 
103 static int exynos_power_up(unsigned int cpu, unsigned int cluster)
104 {
105 	unsigned int cpunr = cpu + (cluster * EXYNOS5420_CPUS_PER_CLUSTER);
106 	int err = 0;
107 
108 	pr_debug("%s: cpu %u cluster %u\n", __func__, cpu, cluster);
109 	if (cpu >= EXYNOS5420_CPUS_PER_CLUSTER ||
110 		cluster >= EXYNOS5420_NR_CLUSTERS)
111 		return -EINVAL;
112 
113 	/*
114 	 * Since this is called with IRQs enabled, and no arch_spin_lock_irq
115 	 * variant exists, we need to disable IRQs manually here.
116 	 */
117 	local_irq_disable();
118 	arch_spin_lock(&exynos_mcpm_lock);
119 
120 	cpu_use_count[cpu][cluster]++;
121 	if (cpu_use_count[cpu][cluster] == 1) {
122 		bool was_cluster_down =
123 			(exynos_cluster_usecnt(cluster) == 1);
124 
125 		/*
126 		 * Turn on the cluster (L2/COMMON) and then power on the
127 		 * cores.
128 		 */
129 		if (was_cluster_down)
130 			err = exynos_cluster_power_control(cluster, 1);
131 
132 		if (!err)
133 			exynos_cpu_power_up(cpunr);
134 		else
135 			exynos_cluster_power_control(cluster, 0);
136 	} else if (cpu_use_count[cpu][cluster] != 2) {
137 		/*
138 		 * The only possible values are:
139 		 * 0 = CPU down
140 		 * 1 = CPU (still) up
141 		 * 2 = CPU requested to be up before it had a chance
142 		 *     to actually make itself down.
143 		 * Any other value is a bug.
144 		 */
145 		BUG();
146 	}
147 
148 	arch_spin_unlock(&exynos_mcpm_lock);
149 	local_irq_enable();
150 
151 	return err;
152 }
153 
154 /*
155  * NOTE: This function requires the stack data to be visible through power down
156  * and can only be executed on processors like A15 and A7 that hit the cache
157  * with the C bit clear in the SCTLR register.
158  */
159 static void exynos_power_down(void)
160 {
161 	unsigned int mpidr, cpu, cluster;
162 	bool last_man = false, skip_wfi = false;
163 	unsigned int cpunr;
164 
165 	mpidr = read_cpuid_mpidr();
166 	cpu = MPIDR_AFFINITY_LEVEL(mpidr, 0);
167 	cluster = MPIDR_AFFINITY_LEVEL(mpidr, 1);
168 	cpunr =  cpu + (cluster * EXYNOS5420_CPUS_PER_CLUSTER);
169 
170 	pr_debug("%s: cpu %u cluster %u\n", __func__, cpu, cluster);
171 	BUG_ON(cpu >= EXYNOS5420_CPUS_PER_CLUSTER ||
172 			cluster >= EXYNOS5420_NR_CLUSTERS);
173 
174 	__mcpm_cpu_going_down(cpu, cluster);
175 
176 	arch_spin_lock(&exynos_mcpm_lock);
177 	BUG_ON(__mcpm_cluster_state(cluster) != CLUSTER_UP);
178 	cpu_use_count[cpu][cluster]--;
179 	if (cpu_use_count[cpu][cluster] == 0) {
180 		exynos_cpu_power_down(cpunr);
181 
182 		if (exynos_cluster_unused(cluster))
183 			/* TODO: Turn off the cluster here to save power. */
184 			last_man = true;
185 	} else if (cpu_use_count[cpu][cluster] == 1) {
186 		/*
187 		 * A power_up request went ahead of us.
188 		 * Even if we do not want to shut this CPU down,
189 		 * the caller expects a certain state as if the WFI
190 		 * was aborted.  So let's continue with cache cleaning.
191 		 */
192 		skip_wfi = true;
193 	} else {
194 		BUG();
195 	}
196 
197 	if (last_man && __mcpm_outbound_enter_critical(cpu, cluster)) {
198 		arch_spin_unlock(&exynos_mcpm_lock);
199 
200 		if (read_cpuid_part_number() == ARM_CPU_PART_CORTEX_A15) {
201 			/*
202 			 * On the Cortex-A15 we need to disable
203 			 * L2 prefetching before flushing the cache.
204 			 */
205 			asm volatile(
206 			"mcr	p15, 1, %0, c15, c0, 3\n\t"
207 			"isb\n\t"
208 			"dsb"
209 			: : "r" (0x400));
210 		}
211 
212 		/* Flush all cache levels for this cluster. */
213 		exynos_v7_exit_coherency_flush(all);
214 
215 		/*
216 		 * Disable cluster-level coherency by masking
217 		 * incoming snoops and DVM messages:
218 		 */
219 		cci_disable_port_by_cpu(mpidr);
220 
221 		__mcpm_outbound_leave_critical(cluster, CLUSTER_DOWN);
222 	} else {
223 		arch_spin_unlock(&exynos_mcpm_lock);
224 
225 		/* Disable and flush the local CPU cache. */
226 		exynos_v7_exit_coherency_flush(louis);
227 	}
228 
229 	__mcpm_cpu_down(cpu, cluster);
230 
231 	/* Now we are prepared for power-down, do it: */
232 	if (!skip_wfi)
233 		wfi();
234 
235 	/* Not dead at this point?  Let our caller cope. */
236 }
237 
238 static int exynos_wait_for_powerdown(unsigned int cpu, unsigned int cluster)
239 {
240 	unsigned int tries = 100;
241 	unsigned int cpunr = cpu + (cluster * EXYNOS5420_CPUS_PER_CLUSTER);
242 
243 	pr_debug("%s: cpu %u cluster %u\n", __func__, cpu, cluster);
244 	BUG_ON(cpu >= EXYNOS5420_CPUS_PER_CLUSTER ||
245 			cluster >= EXYNOS5420_NR_CLUSTERS);
246 
247 	/* Wait for the core state to be OFF */
248 	while (tries--) {
249 		if (ACCESS_ONCE(cpu_use_count[cpu][cluster]) == 0) {
250 			if ((exynos_cpu_power_state(cpunr) == 0))
251 				return 0; /* success: the CPU is halted */
252 		}
253 
254 		/* Otherwise, wait and retry: */
255 		msleep(1);
256 	}
257 
258 	return -ETIMEDOUT; /* timeout */
259 }
260 
261 static const struct mcpm_platform_ops exynos_power_ops = {
262 	.power_up		= exynos_power_up,
263 	.power_down		= exynos_power_down,
264 	.wait_for_powerdown	= exynos_wait_for_powerdown,
265 };
266 
267 static void __init exynos_mcpm_usage_count_init(void)
268 {
269 	unsigned int mpidr, cpu, cluster;
270 
271 	mpidr = read_cpuid_mpidr();
272 	cpu = MPIDR_AFFINITY_LEVEL(mpidr, 0);
273 	cluster = MPIDR_AFFINITY_LEVEL(mpidr, 1);
274 
275 	pr_debug("%s: cpu %u cluster %u\n", __func__, cpu, cluster);
276 	BUG_ON(cpu >= EXYNOS5420_CPUS_PER_CLUSTER  ||
277 			cluster >= EXYNOS5420_NR_CLUSTERS);
278 
279 	cpu_use_count[cpu][cluster] = 1;
280 }
281 
282 /*
283  * Enable cluster-level coherency, in preparation for turning on the MMU.
284  */
285 static void __naked exynos_pm_power_up_setup(unsigned int affinity_level)
286 {
287 	asm volatile ("\n"
288 	"cmp	r0, #1\n"
289 	"bxne	lr\n"
290 	"b	cci_enable_port_for_self");
291 }
292 
293 static const struct of_device_id exynos_dt_mcpm_match[] = {
294 	{ .compatible = "samsung,exynos5420" },
295 	{ .compatible = "samsung,exynos5800" },
296 	{},
297 };
298 
299 static int __init exynos_mcpm_init(void)
300 {
301 	struct device_node *node;
302 	void __iomem *ns_sram_base_addr;
303 	int ret;
304 
305 	node = of_find_matching_node(NULL, exynos_dt_mcpm_match);
306 	if (!node)
307 		return -ENODEV;
308 	of_node_put(node);
309 
310 	if (!cci_probed())
311 		return -ENODEV;
312 
313 	node = of_find_compatible_node(NULL, NULL,
314 			"samsung,exynos4210-sysram-ns");
315 	if (!node)
316 		return -ENODEV;
317 
318 	ns_sram_base_addr = of_iomap(node, 0);
319 	of_node_put(node);
320 	if (!ns_sram_base_addr) {
321 		pr_err("failed to map non-secure iRAM base address\n");
322 		return -ENOMEM;
323 	}
324 
325 	/*
326 	 * To increase the stability of KFC reset we need to program
327 	 * the PMU SPARE3 register
328 	 */
329 	__raw_writel(EXYNOS5420_SWRESET_KFC_SEL, S5P_PMU_SPARE3);
330 
331 	exynos_mcpm_usage_count_init();
332 
333 	ret = mcpm_platform_register(&exynos_power_ops);
334 	if (!ret)
335 		ret = mcpm_sync_init(exynos_pm_power_up_setup);
336 	if (ret) {
337 		iounmap(ns_sram_base_addr);
338 		return ret;
339 	}
340 
341 	mcpm_smp_set_ops();
342 
343 	pr_info("Exynos MCPM support installed\n");
344 
345 	/*
346 	 * Future entries into the kernel can now go
347 	 * through the cluster entry vectors.
348 	 */
349 	__raw_writel(virt_to_phys(mcpm_entry_point),
350 			ns_sram_base_addr + MCPM_BOOT_ADDR_OFFSET);
351 
352 	iounmap(ns_sram_base_addr);
353 
354 	return ret;
355 }
356 
357 early_initcall(exynos_mcpm_init);
358