xref: /openbmc/linux/drivers/soc/dove/pmu.c (revision a06c488d)
1 /*
2  * Marvell Dove PMU support
3  */
4 #include <linux/io.h>
5 #include <linux/irq.h>
6 #include <linux/irqdomain.h>
7 #include <linux/of.h>
8 #include <linux/of_irq.h>
9 #include <linux/of_address.h>
10 #include <linux/platform_device.h>
11 #include <linux/pm_domain.h>
12 #include <linux/reset.h>
13 #include <linux/reset-controller.h>
14 #include <linux/sched.h>
15 #include <linux/slab.h>
16 #include <linux/soc/dove/pmu.h>
17 #include <linux/spinlock.h>
18 
19 #define NR_PMU_IRQS		7
20 
21 #define PMC_SW_RST		0x30
22 #define PMC_IRQ_CAUSE		0x50
23 #define PMC_IRQ_MASK		0x54
24 
25 #define PMU_PWR			0x10
26 #define PMU_ISO			0x58
27 
28 struct pmu_data {
29 	spinlock_t lock;
30 	struct device_node *of_node;
31 	void __iomem *pmc_base;
32 	void __iomem *pmu_base;
33 	struct irq_chip_generic *irq_gc;
34 	struct irq_domain *irq_domain;
35 #ifdef CONFIG_RESET_CONTROLLER
36 	struct reset_controller_dev reset;
37 #endif
38 };
39 
40 /*
41  * The PMU contains a register to reset various subsystems within the
42  * SoC.  Export this as a reset controller.
43  */
44 #ifdef CONFIG_RESET_CONTROLLER
45 #define rcdev_to_pmu(rcdev) container_of(rcdev, struct pmu_data, reset)
46 
47 static int pmu_reset_reset(struct reset_controller_dev *rc, unsigned long id)
48 {
49 	struct pmu_data *pmu = rcdev_to_pmu(rc);
50 	unsigned long flags;
51 	u32 val;
52 
53 	spin_lock_irqsave(&pmu->lock, flags);
54 	val = readl_relaxed(pmu->pmc_base + PMC_SW_RST);
55 	writel_relaxed(val & ~BIT(id), pmu->pmc_base + PMC_SW_RST);
56 	writel_relaxed(val | BIT(id), pmu->pmc_base + PMC_SW_RST);
57 	spin_unlock_irqrestore(&pmu->lock, flags);
58 
59 	return 0;
60 }
61 
62 static int pmu_reset_assert(struct reset_controller_dev *rc, unsigned long id)
63 {
64 	struct pmu_data *pmu = rcdev_to_pmu(rc);
65 	unsigned long flags;
66 	u32 val = ~BIT(id);
67 
68 	spin_lock_irqsave(&pmu->lock, flags);
69 	val &= readl_relaxed(pmu->pmc_base + PMC_SW_RST);
70 	writel_relaxed(val, pmu->pmc_base + PMC_SW_RST);
71 	spin_unlock_irqrestore(&pmu->lock, flags);
72 
73 	return 0;
74 }
75 
76 static int pmu_reset_deassert(struct reset_controller_dev *rc, unsigned long id)
77 {
78 	struct pmu_data *pmu = rcdev_to_pmu(rc);
79 	unsigned long flags;
80 	u32 val = BIT(id);
81 
82 	spin_lock_irqsave(&pmu->lock, flags);
83 	val |= readl_relaxed(pmu->pmc_base + PMC_SW_RST);
84 	writel_relaxed(val, pmu->pmc_base + PMC_SW_RST);
85 	spin_unlock_irqrestore(&pmu->lock, flags);
86 
87 	return 0;
88 }
89 
90 static struct reset_control_ops pmu_reset_ops = {
91 	.reset = pmu_reset_reset,
92 	.assert = pmu_reset_assert,
93 	.deassert = pmu_reset_deassert,
94 };
95 
96 static struct reset_controller_dev pmu_reset __initdata = {
97 	.ops = &pmu_reset_ops,
98 	.owner = THIS_MODULE,
99 	.nr_resets = 32,
100 };
101 
102 static void __init pmu_reset_init(struct pmu_data *pmu)
103 {
104 	int ret;
105 
106 	pmu->reset = pmu_reset;
107 	pmu->reset.of_node = pmu->of_node;
108 
109 	ret = reset_controller_register(&pmu->reset);
110 	if (ret)
111 		pr_err("pmu: %s failed: %d\n", "reset_controller_register", ret);
112 }
113 #else
114 static void __init pmu_reset_init(struct pmu_data *pmu)
115 {
116 }
117 #endif
118 
119 struct pmu_domain {
120 	struct pmu_data *pmu;
121 	u32 pwr_mask;
122 	u32 rst_mask;
123 	u32 iso_mask;
124 	struct generic_pm_domain base;
125 };
126 
127 #define to_pmu_domain(dom) container_of(dom, struct pmu_domain, base)
128 
129 /*
130  * This deals with the "old" Marvell sequence of bringing a power domain
131  * down/up, which is: apply power, release reset, disable isolators.
132  *
133  * Later devices apparantly use a different sequence: power up, disable
134  * isolators, assert repair signal, enable SRMA clock, enable AXI clock,
135  * enable module clock, deassert reset.
136  *
137  * Note: reading the assembly, it seems that the IO accessors have an
138  * unfortunate side-effect - they cause memory already read into registers
139  * for the if () to be re-read for the bit-set or bit-clear operation.
140  * The code is written to avoid this.
141  */
142 static int pmu_domain_power_off(struct generic_pm_domain *domain)
143 {
144 	struct pmu_domain *pmu_dom = to_pmu_domain(domain);
145 	struct pmu_data *pmu = pmu_dom->pmu;
146 	unsigned long flags;
147 	unsigned int val;
148 	void __iomem *pmu_base = pmu->pmu_base;
149 	void __iomem *pmc_base = pmu->pmc_base;
150 
151 	spin_lock_irqsave(&pmu->lock, flags);
152 
153 	/* Enable isolators */
154 	if (pmu_dom->iso_mask) {
155 		val = ~pmu_dom->iso_mask;
156 		val &= readl_relaxed(pmu_base + PMU_ISO);
157 		writel_relaxed(val, pmu_base + PMU_ISO);
158 	}
159 
160 	/* Reset unit */
161 	if (pmu_dom->rst_mask) {
162 		val = ~pmu_dom->rst_mask;
163 		val &= readl_relaxed(pmc_base + PMC_SW_RST);
164 		writel_relaxed(val, pmc_base + PMC_SW_RST);
165 	}
166 
167 	/* Power down */
168 	val = readl_relaxed(pmu_base + PMU_PWR) | pmu_dom->pwr_mask;
169 	writel_relaxed(val, pmu_base + PMU_PWR);
170 
171 	spin_unlock_irqrestore(&pmu->lock, flags);
172 
173 	return 0;
174 }
175 
176 static int pmu_domain_power_on(struct generic_pm_domain *domain)
177 {
178 	struct pmu_domain *pmu_dom = to_pmu_domain(domain);
179 	struct pmu_data *pmu = pmu_dom->pmu;
180 	unsigned long flags;
181 	unsigned int val;
182 	void __iomem *pmu_base = pmu->pmu_base;
183 	void __iomem *pmc_base = pmu->pmc_base;
184 
185 	spin_lock_irqsave(&pmu->lock, flags);
186 
187 	/* Power on */
188 	val = ~pmu_dom->pwr_mask & readl_relaxed(pmu_base + PMU_PWR);
189 	writel_relaxed(val, pmu_base + PMU_PWR);
190 
191 	/* Release reset */
192 	if (pmu_dom->rst_mask) {
193 		val = pmu_dom->rst_mask;
194 		val |= readl_relaxed(pmc_base + PMC_SW_RST);
195 		writel_relaxed(val, pmc_base + PMC_SW_RST);
196 	}
197 
198 	/* Disable isolators */
199 	if (pmu_dom->iso_mask) {
200 		val = pmu_dom->iso_mask;
201 		val |= readl_relaxed(pmu_base + PMU_ISO);
202 		writel_relaxed(val, pmu_base + PMU_ISO);
203 	}
204 
205 	spin_unlock_irqrestore(&pmu->lock, flags);
206 
207 	return 0;
208 }
209 
210 static void __pmu_domain_register(struct pmu_domain *domain,
211 	struct device_node *np)
212 {
213 	unsigned int val = readl_relaxed(domain->pmu->pmu_base + PMU_PWR);
214 
215 	domain->base.power_off = pmu_domain_power_off;
216 	domain->base.power_on = pmu_domain_power_on;
217 
218 	pm_genpd_init(&domain->base, NULL, !(val & domain->pwr_mask));
219 
220 	if (np)
221 		of_genpd_add_provider_simple(np, &domain->base);
222 }
223 
224 /* PMU IRQ controller */
225 static void pmu_irq_handler(struct irq_desc *desc)
226 {
227 	struct pmu_data *pmu = irq_desc_get_handler_data(desc);
228 	struct irq_chip_generic *gc = pmu->irq_gc;
229 	struct irq_domain *domain = pmu->irq_domain;
230 	void __iomem *base = gc->reg_base;
231 	u32 stat = readl_relaxed(base + PMC_IRQ_CAUSE) & gc->mask_cache;
232 	u32 done = ~0;
233 
234 	if (stat == 0) {
235 		handle_bad_irq(desc);
236 		return;
237 	}
238 
239 	while (stat) {
240 		u32 hwirq = fls(stat) - 1;
241 
242 		stat &= ~(1 << hwirq);
243 		done &= ~(1 << hwirq);
244 
245 		generic_handle_irq(irq_find_mapping(domain, hwirq));
246 	}
247 
248 	/*
249 	 * The PMU mask register is not RW0C: it is RW.  This means that
250 	 * the bits take whatever value is written to them; if you write
251 	 * a '1', you will set the interrupt.
252 	 *
253 	 * Unfortunately this means there is NO race free way to clear
254 	 * these interrupts.
255 	 *
256 	 * So, let's structure the code so that the window is as small as
257 	 * possible.
258 	 */
259 	irq_gc_lock(gc);
260 	done &= readl_relaxed(base + PMC_IRQ_CAUSE);
261 	writel_relaxed(done, base + PMC_IRQ_CAUSE);
262 	irq_gc_unlock(gc);
263 }
264 
265 static int __init dove_init_pmu_irq(struct pmu_data *pmu, int irq)
266 {
267 	const char *name = "pmu_irq";
268 	struct irq_chip_generic *gc;
269 	struct irq_domain *domain;
270 	int ret;
271 
272 	/* mask and clear all interrupts */
273 	writel(0, pmu->pmc_base + PMC_IRQ_MASK);
274 	writel(0, pmu->pmc_base + PMC_IRQ_CAUSE);
275 
276 	domain = irq_domain_add_linear(pmu->of_node, NR_PMU_IRQS,
277 				       &irq_generic_chip_ops, NULL);
278 	if (!domain) {
279 		pr_err("%s: unable to add irq domain\n", name);
280 		return -ENOMEM;
281 	}
282 
283 	ret = irq_alloc_domain_generic_chips(domain, NR_PMU_IRQS, 1, name,
284 					     handle_level_irq,
285 					     IRQ_NOREQUEST | IRQ_NOPROBE, 0,
286 					     IRQ_GC_INIT_MASK_CACHE);
287 	if (ret) {
288 		pr_err("%s: unable to alloc irq domain gc: %d\n", name, ret);
289 		irq_domain_remove(domain);
290 		return ret;
291 	}
292 
293 	gc = irq_get_domain_generic_chip(domain, 0);
294 	gc->reg_base = pmu->pmc_base;
295 	gc->chip_types[0].regs.mask = PMC_IRQ_MASK;
296 	gc->chip_types[0].chip.irq_mask = irq_gc_mask_clr_bit;
297 	gc->chip_types[0].chip.irq_unmask = irq_gc_mask_set_bit;
298 
299 	pmu->irq_domain = domain;
300 	pmu->irq_gc = gc;
301 
302 	irq_set_handler_data(irq, pmu);
303 	irq_set_chained_handler(irq, pmu_irq_handler);
304 
305 	return 0;
306 }
307 
308 int __init dove_init_pmu_legacy(const struct dove_pmu_initdata *initdata)
309 {
310 	const struct dove_pmu_domain_initdata *domain_initdata;
311 	struct pmu_data *pmu;
312 	int ret;
313 
314 	pmu = kzalloc(sizeof(*pmu), GFP_KERNEL);
315 	if (!pmu)
316 		return -ENOMEM;
317 
318 	spin_lock_init(&pmu->lock);
319 	pmu->pmc_base = initdata->pmc_base;
320 	pmu->pmu_base = initdata->pmu_base;
321 
322 	pmu_reset_init(pmu);
323 	for (domain_initdata = initdata->domains; domain_initdata->name;
324 	     domain_initdata++) {
325 		struct pmu_domain *domain;
326 
327 		domain = kzalloc(sizeof(*domain), GFP_KERNEL);
328 		if (domain) {
329 			domain->pmu = pmu;
330 			domain->pwr_mask = domain_initdata->pwr_mask;
331 			domain->rst_mask = domain_initdata->rst_mask;
332 			domain->iso_mask = domain_initdata->iso_mask;
333 			domain->base.name = domain_initdata->name;
334 
335 			__pmu_domain_register(domain, NULL);
336 		}
337 	}
338 
339 	ret = dove_init_pmu_irq(pmu, initdata->irq);
340 	if (ret)
341 		pr_err("dove_init_pmu_irq() failed: %d\n", ret);
342 
343 	if (pmu->irq_domain)
344 		irq_domain_associate_many(pmu->irq_domain,
345 					  initdata->irq_domain_start,
346 					  0, NR_PMU_IRQS);
347 
348 	return 0;
349 }
350 
351 /*
352  * pmu: power-manager@d0000 {
353  *	compatible = "marvell,dove-pmu";
354  *	reg = <0xd0000 0x8000> <0xd8000 0x8000>;
355  *	interrupts = <33>;
356  *	interrupt-controller;
357  *	#reset-cells = 1;
358  *	vpu_domain: vpu-domain {
359  *		#power-domain-cells = <0>;
360  *		marvell,pmu_pwr_mask = <0x00000008>;
361  *		marvell,pmu_iso_mask = <0x00000001>;
362  *		resets = <&pmu 16>;
363  *	};
364  *	gpu_domain: gpu-domain {
365  *		#power-domain-cells = <0>;
366  *		marvell,pmu_pwr_mask = <0x00000004>;
367  *		marvell,pmu_iso_mask = <0x00000002>;
368  *		resets = <&pmu 18>;
369  *	};
370  * };
371  */
372 int __init dove_init_pmu(void)
373 {
374 	struct device_node *np_pmu, *domains_node, *np;
375 	struct pmu_data *pmu;
376 	int ret, parent_irq;
377 
378 	/* Lookup the PMU node */
379 	np_pmu = of_find_compatible_node(NULL, NULL, "marvell,dove-pmu");
380 	if (!np_pmu)
381 		return 0;
382 
383 	domains_node = of_get_child_by_name(np_pmu, "domains");
384 	if (!domains_node) {
385 		pr_err("%s: failed to find domains sub-node\n", np_pmu->name);
386 		return 0;
387 	}
388 
389 	pmu = kzalloc(sizeof(*pmu), GFP_KERNEL);
390 	if (!pmu)
391 		return -ENOMEM;
392 
393 	spin_lock_init(&pmu->lock);
394 	pmu->of_node = np_pmu;
395 	pmu->pmc_base = of_iomap(pmu->of_node, 0);
396 	pmu->pmu_base = of_iomap(pmu->of_node, 1);
397 	if (!pmu->pmc_base || !pmu->pmu_base) {
398 		pr_err("%s: failed to map PMU\n", np_pmu->name);
399 		iounmap(pmu->pmu_base);
400 		iounmap(pmu->pmc_base);
401 		kfree(pmu);
402 		return -ENOMEM;
403 	}
404 
405 	pmu_reset_init(pmu);
406 
407 	for_each_available_child_of_node(domains_node, np) {
408 		struct of_phandle_args args;
409 		struct pmu_domain *domain;
410 
411 		domain = kzalloc(sizeof(*domain), GFP_KERNEL);
412 		if (!domain)
413 			break;
414 
415 		domain->pmu = pmu;
416 		domain->base.name = kstrdup(np->name, GFP_KERNEL);
417 		if (!domain->base.name) {
418 			kfree(domain);
419 			break;
420 		}
421 
422 		of_property_read_u32(np, "marvell,pmu_pwr_mask",
423 				     &domain->pwr_mask);
424 		of_property_read_u32(np, "marvell,pmu_iso_mask",
425 				     &domain->iso_mask);
426 
427 		/*
428 		 * We parse the reset controller property directly here
429 		 * to ensure that we can operate when the reset controller
430 		 * support is not configured into the kernel.
431 		 */
432 		ret = of_parse_phandle_with_args(np, "resets", "#reset-cells",
433 						 0, &args);
434 		if (ret == 0) {
435 			if (args.np == pmu->of_node)
436 				domain->rst_mask = BIT(args.args[0]);
437 			of_node_put(args.np);
438 		}
439 
440 		__pmu_domain_register(domain, np);
441 	}
442 
443 	/* Loss of the interrupt controller is not a fatal error. */
444 	parent_irq = irq_of_parse_and_map(pmu->of_node, 0);
445 	if (!parent_irq) {
446 		pr_err("%s: no interrupt specified\n", np_pmu->name);
447 	} else {
448 		ret = dove_init_pmu_irq(pmu, parent_irq);
449 		if (ret)
450 			pr_err("dove_init_pmu_irq() failed: %d\n", ret);
451 	}
452 
453 	return 0;
454 }
455