1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Copyright (c) 2018 Chen-Yu Tsai
4 *
5 * Chen-Yu Tsai <wens@csie.org>
6 *
7 * arch/arm/mach-sunxi/mc_smp.c
8 *
9 * Based on Allwinner code, arch/arm/mach-exynos/mcpm-exynos.c, and
10 * arch/arm/mach-hisi/platmcpm.c
11 * Cluster cache enable trampoline code adapted from MCPM framework
12 */
13
14#include <linux/arm-cci.h>
15#include <linux/cpu_pm.h>
16#include <linux/delay.h>
17#include <linux/io.h>
18#include <linux/iopoll.h>
19#include <linux/irqchip/arm-gic.h>
20#include <linux/of.h>
21#include <linux/of_address.h>
22#include <linux/of_device.h>
23#include <linux/smp.h>
24
25#include <asm/cacheflush.h>
26#include <asm/cp15.h>
27#include <asm/cputype.h>
28#include <asm/idmap.h>
29#include <asm/smp_plat.h>
30#include <asm/suspend.h>
31
32#define SUNXI_CPUS_PER_CLUSTER		4
33#define SUNXI_NR_CLUSTERS		2
34
35#define POLL_USEC	100
36#define TIMEOUT_USEC	100000
37
38#define CPUCFG_CX_CTRL_REG0(c)		(0x10 * (c))
39#define CPUCFG_CX_CTRL_REG0_L1_RST_DISABLE(n)	BIT(n)
40#define CPUCFG_CX_CTRL_REG0_L1_RST_DISABLE_ALL	0xf
41#define CPUCFG_CX_CTRL_REG0_L2_RST_DISABLE_A7	BIT(4)
42#define CPUCFG_CX_CTRL_REG0_L2_RST_DISABLE_A15	BIT(0)
43#define CPUCFG_CX_CTRL_REG1(c)		(0x10 * (c) + 0x4)
44#define CPUCFG_CX_CTRL_REG1_ACINACTM	BIT(0)
45#define CPUCFG_CX_STATUS(c)		(0x30 + 0x4 * (c))
46#define CPUCFG_CX_STATUS_STANDBYWFI(n)	BIT(16 + (n))
47#define CPUCFG_CX_STATUS_STANDBYWFIL2	BIT(0)
48#define CPUCFG_CX_RST_CTRL(c)		(0x80 + 0x4 * (c))
49#define CPUCFG_CX_RST_CTRL_DBG_SOC_RST	BIT(24)
50#define CPUCFG_CX_RST_CTRL_ETM_RST(n)	BIT(20 + (n))
51#define CPUCFG_CX_RST_CTRL_ETM_RST_ALL	(0xf << 20)
52#define CPUCFG_CX_RST_CTRL_DBG_RST(n)	BIT(16 + (n))
53#define CPUCFG_CX_RST_CTRL_DBG_RST_ALL	(0xf << 16)
54#define CPUCFG_CX_RST_CTRL_H_RST	BIT(12)
55#define CPUCFG_CX_RST_CTRL_L2_RST	BIT(8)
56#define CPUCFG_CX_RST_CTRL_CX_RST(n)	BIT(4 + (n))
57#define CPUCFG_CX_RST_CTRL_CORE_RST(n)	BIT(n)
58#define CPUCFG_CX_RST_CTRL_CORE_RST_ALL	(0xf << 0)
59
60#define PRCM_CPU_PO_RST_CTRL(c)		(0x4 + 0x4 * (c))
61#define PRCM_CPU_PO_RST_CTRL_CORE(n)	BIT(n)
62#define PRCM_CPU_PO_RST_CTRL_CORE_ALL	0xf
63#define PRCM_PWROFF_GATING_REG(c)	(0x100 + 0x4 * (c))
64/* The power off register for clusters are different from a80 and a83t */
65#define PRCM_PWROFF_GATING_REG_CLUSTER_SUN8I	BIT(0)
66#define PRCM_PWROFF_GATING_REG_CLUSTER_SUN9I	BIT(4)
67#define PRCM_PWROFF_GATING_REG_CORE(n)	BIT(n)
68#define PRCM_PWR_SWITCH_REG(c, cpu)	(0x140 + 0x10 * (c) + 0x4 * (cpu))
69#define PRCM_CPU_SOFT_ENTRY_REG		0x164
70
71/* R_CPUCFG registers, specific to sun8i-a83t */
72#define R_CPUCFG_CLUSTER_PO_RST_CTRL(c)	(0x30 + (c) * 0x4)
73#define R_CPUCFG_CLUSTER_PO_RST_CTRL_CORE(n)	BIT(n)
74#define R_CPUCFG_CPU_SOFT_ENTRY_REG		0x01a4
75
76#define CPU0_SUPPORT_HOTPLUG_MAGIC0	0xFA50392F
77#define CPU0_SUPPORT_HOTPLUG_MAGIC1	0x790DCA3A
78
79static void __iomem *cpucfg_base;
80static void __iomem *prcm_base;
81static void __iomem *sram_b_smp_base;
82static void __iomem *r_cpucfg_base;
83
84extern void sunxi_mc_smp_secondary_startup(void);
85extern void sunxi_mc_smp_resume(void);
86static bool is_a83t;
87
88static bool sunxi_core_is_cortex_a15(unsigned int core, unsigned int cluster)
89{
90	struct device_node *node;
91	int cpu = cluster * SUNXI_CPUS_PER_CLUSTER + core;
92	bool is_compatible;
93
94	node = of_cpu_device_node_get(cpu);
95
96	/* In case of_cpu_device_node_get fails */
97	if (!node)
98		node = of_get_cpu_node(cpu, NULL);
99
100	if (!node) {
101		/*
102		 * There's no point in returning an error, since we
103		 * would be mid way in a core or cluster power sequence.
104		 */
105		pr_err("%s: Couldn't get CPU cluster %u core %u device node\n",
106		       __func__, cluster, core);
107
108		return false;
109	}
110
111	is_compatible = of_device_is_compatible(node, "arm,cortex-a15");
112	of_node_put(node);
113	return is_compatible;
114}
115
116static int sunxi_cpu_power_switch_set(unsigned int cpu, unsigned int cluster,
117				      bool enable)
118{
119	u32 reg;
120
121	/* control sequence from Allwinner A80 user manual v1.2 PRCM section */
122	reg = readl(prcm_base + PRCM_PWR_SWITCH_REG(cluster, cpu));
123	if (enable) {
124		if (reg == 0x00) {
125			pr_debug("power clamp for cluster %u cpu %u already open\n",
126				 cluster, cpu);
127			return 0;
128		}
129
130		writel(0xff, prcm_base + PRCM_PWR_SWITCH_REG(cluster, cpu));
131		udelay(10);
132		writel(0xfe, prcm_base + PRCM_PWR_SWITCH_REG(cluster, cpu));
133		udelay(10);
134		writel(0xf8, prcm_base + PRCM_PWR_SWITCH_REG(cluster, cpu));
135		udelay(10);
136		writel(0xf0, prcm_base + PRCM_PWR_SWITCH_REG(cluster, cpu));
137		udelay(10);
138		writel(0x00, prcm_base + PRCM_PWR_SWITCH_REG(cluster, cpu));
139		udelay(10);
140	} else {
141		writel(0xff, prcm_base + PRCM_PWR_SWITCH_REG(cluster, cpu));
142		udelay(10);
143	}
144
145	return 0;
146}
147
148static void sunxi_cpu0_hotplug_support_set(bool enable)
149{
150	if (enable) {
151		writel(CPU0_SUPPORT_HOTPLUG_MAGIC0, sram_b_smp_base);
152		writel(CPU0_SUPPORT_HOTPLUG_MAGIC1, sram_b_smp_base + 0x4);
153	} else {
154		writel(0x0, sram_b_smp_base);
155		writel(0x0, sram_b_smp_base + 0x4);
156	}
157}
158
159static int sunxi_cpu_powerup(unsigned int cpu, unsigned int cluster)
160{
161	u32 reg;
162
163	pr_debug("%s: cluster %u cpu %u\n", __func__, cluster, cpu);
164	if (cpu >= SUNXI_CPUS_PER_CLUSTER || cluster >= SUNXI_NR_CLUSTERS)
165		return -EINVAL;
166
167	/* Set hotplug support magic flags for cpu0 */
168	if (cluster == 0 && cpu == 0)
169		sunxi_cpu0_hotplug_support_set(true);
170
171	/* assert processor power-on reset */
172	reg = readl(prcm_base + PRCM_CPU_PO_RST_CTRL(cluster));
173	reg &= ~PRCM_CPU_PO_RST_CTRL_CORE(cpu);
174	writel(reg, prcm_base + PRCM_CPU_PO_RST_CTRL(cluster));
175
176	if (is_a83t) {
177		/* assert cpu power-on reset */
178		reg  = readl(r_cpucfg_base +
179			     R_CPUCFG_CLUSTER_PO_RST_CTRL(cluster));
180		reg &= ~(R_CPUCFG_CLUSTER_PO_RST_CTRL_CORE(cpu));
181		writel(reg, r_cpucfg_base +
182		       R_CPUCFG_CLUSTER_PO_RST_CTRL(cluster));
183		udelay(10);
184	}
185
186	/* Cortex-A7: hold L1 reset disable signal low */
187	if (!sunxi_core_is_cortex_a15(cpu, cluster)) {
188		reg = readl(cpucfg_base + CPUCFG_CX_CTRL_REG0(cluster));
189		reg &= ~CPUCFG_CX_CTRL_REG0_L1_RST_DISABLE(cpu);
190		writel(reg, cpucfg_base + CPUCFG_CX_CTRL_REG0(cluster));
191	}
192
193	/* assert processor related resets */
194	reg = readl(cpucfg_base + CPUCFG_CX_RST_CTRL(cluster));
195	reg &= ~CPUCFG_CX_RST_CTRL_DBG_RST(cpu);
196
197	/*
198	 * Allwinner code also asserts resets for NEON on A15. According
199	 * to ARM manuals, asserting power-on reset is sufficient.
200	 */
201	if (!sunxi_core_is_cortex_a15(cpu, cluster))
202		reg &= ~CPUCFG_CX_RST_CTRL_ETM_RST(cpu);
203
204	writel(reg, cpucfg_base + CPUCFG_CX_RST_CTRL(cluster));
205
206	/* open power switch */
207	sunxi_cpu_power_switch_set(cpu, cluster, true);
208
209	/* Handle A83T bit swap */
210	if (is_a83t) {
211		if (cpu == 0)
212			cpu = 4;
213	}
214
215	/* clear processor power gate */
216	reg = readl(prcm_base + PRCM_PWROFF_GATING_REG(cluster));
217	reg &= ~PRCM_PWROFF_GATING_REG_CORE(cpu);
218	writel(reg, prcm_base + PRCM_PWROFF_GATING_REG(cluster));
219	udelay(20);
220
221	/* Handle A83T bit swap */
222	if (is_a83t) {
223		if (cpu == 4)
224			cpu = 0;
225	}
226
227	/* de-assert processor power-on reset */
228	reg = readl(prcm_base + PRCM_CPU_PO_RST_CTRL(cluster));
229	reg |= PRCM_CPU_PO_RST_CTRL_CORE(cpu);
230	writel(reg, prcm_base + PRCM_CPU_PO_RST_CTRL(cluster));
231
232	if (is_a83t) {
233		reg  = readl(r_cpucfg_base +
234			     R_CPUCFG_CLUSTER_PO_RST_CTRL(cluster));
235		reg |= R_CPUCFG_CLUSTER_PO_RST_CTRL_CORE(cpu);
236		writel(reg, r_cpucfg_base +
237		       R_CPUCFG_CLUSTER_PO_RST_CTRL(cluster));
238		udelay(10);
239	}
240
241	/* de-assert all processor resets */
242	reg = readl(cpucfg_base + CPUCFG_CX_RST_CTRL(cluster));
243	reg |= CPUCFG_CX_RST_CTRL_DBG_RST(cpu);
244	reg |= CPUCFG_CX_RST_CTRL_CORE_RST(cpu);
245	if (!sunxi_core_is_cortex_a15(cpu, cluster))
246		reg |= CPUCFG_CX_RST_CTRL_ETM_RST(cpu);
247	else
248		reg |= CPUCFG_CX_RST_CTRL_CX_RST(cpu); /* NEON */
249	writel(reg, cpucfg_base + CPUCFG_CX_RST_CTRL(cluster));
250
251	return 0;
252}
253
254static int sunxi_cluster_powerup(unsigned int cluster)
255{
256	u32 reg;
257
258	pr_debug("%s: cluster %u\n", __func__, cluster);
259	if (cluster >= SUNXI_NR_CLUSTERS)
260		return -EINVAL;
261
262	/* For A83T, assert cluster cores resets */
263	if (is_a83t) {
264		reg = readl(cpucfg_base + CPUCFG_CX_RST_CTRL(cluster));
265		reg &= ~CPUCFG_CX_RST_CTRL_CORE_RST_ALL;   /* Core Reset    */
266		writel(reg, cpucfg_base + CPUCFG_CX_RST_CTRL(cluster));
267		udelay(10);
268	}
269
270	/* assert ACINACTM */
271	reg = readl(cpucfg_base + CPUCFG_CX_CTRL_REG1(cluster));
272	reg |= CPUCFG_CX_CTRL_REG1_ACINACTM;
273	writel(reg, cpucfg_base + CPUCFG_CX_CTRL_REG1(cluster));
274
275	/* assert cluster processor power-on resets */
276	reg = readl(prcm_base + PRCM_CPU_PO_RST_CTRL(cluster));
277	reg &= ~PRCM_CPU_PO_RST_CTRL_CORE_ALL;
278	writel(reg, prcm_base + PRCM_CPU_PO_RST_CTRL(cluster));
279
280	/* assert cluster cores resets */
281	if (is_a83t) {
282		reg  = readl(r_cpucfg_base +
283			     R_CPUCFG_CLUSTER_PO_RST_CTRL(cluster));
284		reg &= ~CPUCFG_CX_RST_CTRL_CORE_RST_ALL;
285		writel(reg, r_cpucfg_base +
286		       R_CPUCFG_CLUSTER_PO_RST_CTRL(cluster));
287		udelay(10);
288	}
289
290	/* assert cluster resets */
291	reg = readl(cpucfg_base + CPUCFG_CX_RST_CTRL(cluster));
292	reg &= ~CPUCFG_CX_RST_CTRL_DBG_SOC_RST;
293	reg &= ~CPUCFG_CX_RST_CTRL_DBG_RST_ALL;
294	reg &= ~CPUCFG_CX_RST_CTRL_H_RST;
295	reg &= ~CPUCFG_CX_RST_CTRL_L2_RST;
296
297	/*
298	 * Allwinner code also asserts resets for NEON on A15. According
299	 * to ARM manuals, asserting power-on reset is sufficient.
300	 */
301	if (!sunxi_core_is_cortex_a15(0, cluster))
302		reg &= ~CPUCFG_CX_RST_CTRL_ETM_RST_ALL;
303
304	writel(reg, cpucfg_base + CPUCFG_CX_RST_CTRL(cluster));
305
306	/* hold L1/L2 reset disable signals low */
307	reg = readl(cpucfg_base + CPUCFG_CX_CTRL_REG0(cluster));
308	if (sunxi_core_is_cortex_a15(0, cluster)) {
309		/* Cortex-A15: hold L2RSTDISABLE low */
310		reg &= ~CPUCFG_CX_CTRL_REG0_L2_RST_DISABLE_A15;
311	} else {
312		/* Cortex-A7: hold L1RSTDISABLE and L2RSTDISABLE low */
313		reg &= ~CPUCFG_CX_CTRL_REG0_L1_RST_DISABLE_ALL;
314		reg &= ~CPUCFG_CX_CTRL_REG0_L2_RST_DISABLE_A7;
315	}
316	writel(reg, cpucfg_base + CPUCFG_CX_CTRL_REG0(cluster));
317
318	/* clear cluster power gate */
319	reg = readl(prcm_base + PRCM_PWROFF_GATING_REG(cluster));
320	if (is_a83t)
321		reg &= ~PRCM_PWROFF_GATING_REG_CLUSTER_SUN8I;
322	else
323		reg &= ~PRCM_PWROFF_GATING_REG_CLUSTER_SUN9I;
324	writel(reg, prcm_base + PRCM_PWROFF_GATING_REG(cluster));
325	udelay(20);
326
327	/* de-assert cluster resets */
328	reg = readl(cpucfg_base + CPUCFG_CX_RST_CTRL(cluster));
329	reg |= CPUCFG_CX_RST_CTRL_DBG_SOC_RST;
330	reg |= CPUCFG_CX_RST_CTRL_H_RST;
331	reg |= CPUCFG_CX_RST_CTRL_L2_RST;
332	writel(reg, cpucfg_base + CPUCFG_CX_RST_CTRL(cluster));
333
334	/* de-assert ACINACTM */
335	reg = readl(cpucfg_base + CPUCFG_CX_CTRL_REG1(cluster));
336	reg &= ~CPUCFG_CX_CTRL_REG1_ACINACTM;
337	writel(reg, cpucfg_base + CPUCFG_CX_CTRL_REG1(cluster));
338
339	return 0;
340}
341
342/*
343 * This bit is shared between the initial nocache_trampoline call to
344 * enable CCI-400 and proper cluster cache disable before power down.
345 */
346static void sunxi_cluster_cache_disable_without_axi(void)
347{
348	if (read_cpuid_part() == ARM_CPU_PART_CORTEX_A15) {
349		/*
350		 * On the Cortex-A15 we need to disable
351		 * L2 prefetching before flushing the cache.
352		 */
353		asm volatile(
354		"mcr	p15, 1, %0, c15, c0, 3\n"
355		"isb\n"
356		"dsb"
357		: : "r" (0x400));
358	}
359
360	/* Flush all cache levels for this cluster. */
361	v7_exit_coherency_flush(all);
362
363	/*
364	 * Disable cluster-level coherency by masking
365	 * incoming snoops and DVM messages:
366	 */
367	cci_disable_port_by_cpu(read_cpuid_mpidr());
368}
369
370static int sunxi_mc_smp_cpu_table[SUNXI_NR_CLUSTERS][SUNXI_CPUS_PER_CLUSTER];
371int sunxi_mc_smp_first_comer;
372
373static DEFINE_SPINLOCK(boot_lock);
374
375static bool sunxi_mc_smp_cluster_is_down(unsigned int cluster)
376{
377	int i;
378
379	for (i = 0; i < SUNXI_CPUS_PER_CLUSTER; i++)
380		if (sunxi_mc_smp_cpu_table[cluster][i])
381			return false;
382	return true;
383}
384
385static void sunxi_mc_smp_secondary_init(unsigned int cpu)
386{
387	/* Clear hotplug support magic flags for cpu0 */
388	if (cpu == 0)
389		sunxi_cpu0_hotplug_support_set(false);
390}
391
392static int sunxi_mc_smp_boot_secondary(unsigned int l_cpu, struct task_struct *idle)
393{
394	unsigned int mpidr, cpu, cluster;
395
396	mpidr = cpu_logical_map(l_cpu);
397	cpu = MPIDR_AFFINITY_LEVEL(mpidr, 0);
398	cluster = MPIDR_AFFINITY_LEVEL(mpidr, 1);
399
400	if (!cpucfg_base)
401		return -ENODEV;
402	if (cluster >= SUNXI_NR_CLUSTERS || cpu >= SUNXI_CPUS_PER_CLUSTER)
403		return -EINVAL;
404
405	spin_lock_irq(&boot_lock);
406
407	if (sunxi_mc_smp_cpu_table[cluster][cpu])
408		goto out;
409
410	if (sunxi_mc_smp_cluster_is_down(cluster)) {
411		sunxi_mc_smp_first_comer = true;
412		sunxi_cluster_powerup(cluster);
413	} else {
414		sunxi_mc_smp_first_comer = false;
415	}
416
417	/* This is read by incoming CPUs with their cache and MMU disabled */
418	sync_cache_w(&sunxi_mc_smp_first_comer);
419	sunxi_cpu_powerup(cpu, cluster);
420
421out:
422	sunxi_mc_smp_cpu_table[cluster][cpu]++;
423	spin_unlock_irq(&boot_lock);
424
425	return 0;
426}
427
428#ifdef CONFIG_HOTPLUG_CPU
429static void sunxi_cluster_cache_disable(void)
430{
431	unsigned int cluster = MPIDR_AFFINITY_LEVEL(read_cpuid_mpidr(), 1);
432	u32 reg;
433
434	pr_debug("%s: cluster %u\n", __func__, cluster);
435
436	sunxi_cluster_cache_disable_without_axi();
437
438	/* last man standing, assert ACINACTM */
439	reg = readl(cpucfg_base + CPUCFG_CX_CTRL_REG1(cluster));
440	reg |= CPUCFG_CX_CTRL_REG1_ACINACTM;
441	writel(reg, cpucfg_base + CPUCFG_CX_CTRL_REG1(cluster));
442}
443
444static void sunxi_mc_smp_cpu_die(unsigned int l_cpu)
445{
446	unsigned int mpidr, cpu, cluster;
447	bool last_man;
448
449	mpidr = cpu_logical_map(l_cpu);
450	cpu = MPIDR_AFFINITY_LEVEL(mpidr, 0);
451	cluster = MPIDR_AFFINITY_LEVEL(mpidr, 1);
452	pr_debug("%s: cluster %u cpu %u\n", __func__, cluster, cpu);
453
454	spin_lock(&boot_lock);
455	sunxi_mc_smp_cpu_table[cluster][cpu]--;
456	if (sunxi_mc_smp_cpu_table[cluster][cpu] == 1) {
457		/* A power_up request went ahead of us. */
458		pr_debug("%s: aborting due to a power up request\n",
459			 __func__);
460		spin_unlock(&boot_lock);
461		return;
462	} else if (sunxi_mc_smp_cpu_table[cluster][cpu] > 1) {
463		pr_err("Cluster %d CPU%d boots multiple times\n",
464		       cluster, cpu);
465		BUG();
466	}
467
468	last_man = sunxi_mc_smp_cluster_is_down(cluster);
469	spin_unlock(&boot_lock);
470
471	gic_cpu_if_down(0);
472	if (last_man)
473		sunxi_cluster_cache_disable();
474	else
475		v7_exit_coherency_flush(louis);
476
477	for (;;)
478		wfi();
479}
480
481static int sunxi_cpu_powerdown(unsigned int cpu, unsigned int cluster)
482{
483	u32 reg;
484	int gating_bit = cpu;
485
486	pr_debug("%s: cluster %u cpu %u\n", __func__, cluster, cpu);
487	if (cpu >= SUNXI_CPUS_PER_CLUSTER || cluster >= SUNXI_NR_CLUSTERS)
488		return -EINVAL;
489
490	if (is_a83t && cpu == 0)
491		gating_bit = 4;
492
493	/* gate processor power */
494	reg = readl(prcm_base + PRCM_PWROFF_GATING_REG(cluster));
495	reg |= PRCM_PWROFF_GATING_REG_CORE(gating_bit);
496	writel(reg, prcm_base + PRCM_PWROFF_GATING_REG(cluster));
497	udelay(20);
498
499	/* close power switch */
500	sunxi_cpu_power_switch_set(cpu, cluster, false);
501
502	return 0;
503}
504
505static int sunxi_cluster_powerdown(unsigned int cluster)
506{
507	u32 reg;
508
509	pr_debug("%s: cluster %u\n", __func__, cluster);
510	if (cluster >= SUNXI_NR_CLUSTERS)
511		return -EINVAL;
512
513	/* assert cluster resets or system will hang */
514	pr_debug("%s: assert cluster reset\n", __func__);
515	reg = readl(cpucfg_base + CPUCFG_CX_RST_CTRL(cluster));
516	reg &= ~CPUCFG_CX_RST_CTRL_DBG_SOC_RST;
517	reg &= ~CPUCFG_CX_RST_CTRL_H_RST;
518	reg &= ~CPUCFG_CX_RST_CTRL_L2_RST;
519	writel(reg, cpucfg_base + CPUCFG_CX_RST_CTRL(cluster));
520
521	/* gate cluster power */
522	pr_debug("%s: gate cluster power\n", __func__);
523	reg = readl(prcm_base + PRCM_PWROFF_GATING_REG(cluster));
524	if (is_a83t)
525		reg |= PRCM_PWROFF_GATING_REG_CLUSTER_SUN8I;
526	else
527		reg |= PRCM_PWROFF_GATING_REG_CLUSTER_SUN9I;
528	writel(reg, prcm_base + PRCM_PWROFF_GATING_REG(cluster));
529	udelay(20);
530
531	return 0;
532}
533
534static int sunxi_mc_smp_cpu_kill(unsigned int l_cpu)
535{
536	unsigned int mpidr, cpu, cluster;
537	unsigned int tries, count;
538	int ret = 0;
539	u32 reg;
540
541	mpidr = cpu_logical_map(l_cpu);
542	cpu = MPIDR_AFFINITY_LEVEL(mpidr, 0);
543	cluster = MPIDR_AFFINITY_LEVEL(mpidr, 1);
544
545	/* This should never happen */
546	if (WARN_ON(cluster >= SUNXI_NR_CLUSTERS ||
547		    cpu >= SUNXI_CPUS_PER_CLUSTER))
548		return 0;
549
550	/* wait for CPU core to die and enter WFI */
551	count = TIMEOUT_USEC / POLL_USEC;
552	spin_lock_irq(&boot_lock);
553	for (tries = 0; tries < count; tries++) {
554		spin_unlock_irq(&boot_lock);
555		usleep_range(POLL_USEC / 2, POLL_USEC);
556		spin_lock_irq(&boot_lock);
557
558		/*
559		 * If the user turns off a bunch of cores at the same
560		 * time, the kernel might call cpu_kill before some of
561		 * them are ready. This is because boot_lock serializes
562		 * both cpu_die and cpu_kill callbacks. Either one could
563		 * run first. We should wait for cpu_die to complete.
564		 */
565		if (sunxi_mc_smp_cpu_table[cluster][cpu])
566			continue;
567
568		reg = readl(cpucfg_base + CPUCFG_CX_STATUS(cluster));
569		if (reg & CPUCFG_CX_STATUS_STANDBYWFI(cpu))
570			break;
571	}
572
573	if (tries >= count) {
574		ret = ETIMEDOUT;
575		goto out;
576	}
577
578	/* power down CPU core */
579	sunxi_cpu_powerdown(cpu, cluster);
580
581	if (!sunxi_mc_smp_cluster_is_down(cluster))
582		goto out;
583
584	/* wait for cluster L2 WFI */
585	ret = readl_poll_timeout(cpucfg_base + CPUCFG_CX_STATUS(cluster), reg,
586				 reg & CPUCFG_CX_STATUS_STANDBYWFIL2,
587				 POLL_USEC, TIMEOUT_USEC);
588	if (ret) {
589		/*
590		 * Ignore timeout on the cluster. Leaving the cluster on
591		 * will not affect system execution, just use a bit more
592		 * power. But returning an error here will only confuse
593		 * the user as the CPU has already been shutdown.
594		 */
595		ret = 0;
596		goto out;
597	}
598
599	/* Power down cluster */
600	sunxi_cluster_powerdown(cluster);
601
602out:
603	spin_unlock_irq(&boot_lock);
604	pr_debug("%s: cluster %u cpu %u powerdown: %d\n",
605		 __func__, cluster, cpu, ret);
606	return !ret;
607}
608
609static bool sunxi_mc_smp_cpu_can_disable(unsigned int cpu)
610{
611	/* CPU0 hotplug not handled for sun8i-a83t */
612	if (is_a83t)
613		if (cpu == 0)
614			return false;
615	return true;
616}
617#endif
618
619static const struct smp_operations sunxi_mc_smp_smp_ops __initconst = {
620	.smp_secondary_init	= sunxi_mc_smp_secondary_init,
621	.smp_boot_secondary	= sunxi_mc_smp_boot_secondary,
622#ifdef CONFIG_HOTPLUG_CPU
623	.cpu_die		= sunxi_mc_smp_cpu_die,
624	.cpu_kill		= sunxi_mc_smp_cpu_kill,
625	.cpu_can_disable	= sunxi_mc_smp_cpu_can_disable,
626#endif
627};
628
629static bool __init sunxi_mc_smp_cpu_table_init(void)
630{
631	unsigned int mpidr, cpu, cluster;
632
633	mpidr = read_cpuid_mpidr();
634	cpu = MPIDR_AFFINITY_LEVEL(mpidr, 0);
635	cluster = MPIDR_AFFINITY_LEVEL(mpidr, 1);
636
637	if (cluster >= SUNXI_NR_CLUSTERS || cpu >= SUNXI_CPUS_PER_CLUSTER) {
638		pr_err("%s: boot CPU is out of bounds!\n", __func__);
639		return false;
640	}
641	sunxi_mc_smp_cpu_table[cluster][cpu] = 1;
642	return true;
643}
644
645/*
646 * Adapted from arch/arm/common/mc_smp_entry.c
647 *
648 * We need the trampoline code to enable CCI-400 on the first cluster
649 */
650typedef typeof(cpu_reset) phys_reset_t;
651
652static int __init nocache_trampoline(unsigned long __unused)
653{
654	phys_reset_t phys_reset;
655
656	setup_mm_for_reboot();
657	sunxi_cluster_cache_disable_without_axi();
658
659	phys_reset = (phys_reset_t)(unsigned long)__pa_symbol(cpu_reset);
660	phys_reset(__pa_symbol(sunxi_mc_smp_resume), false);
661	BUG();
662}
663
664static int __init sunxi_mc_smp_loopback(void)
665{
666	int ret;
667
668	/*
669	 * We're going to soft-restart the current CPU through the
670	 * low-level MCPM code by leveraging the suspend/resume
671	 * infrastructure. Let's play it safe by using cpu_pm_enter()
672	 * in case the CPU init code path resets the VFP or similar.
673	 */
674	sunxi_mc_smp_first_comer = true;
675	local_irq_disable();
676	local_fiq_disable();
677	ret = cpu_pm_enter();
678	if (!ret) {
679		ret = cpu_suspend(0, nocache_trampoline);
680		cpu_pm_exit();
681	}
682	local_fiq_enable();
683	local_irq_enable();
684	sunxi_mc_smp_first_comer = false;
685
686	return ret;
687}
688
689/*
690 * This holds any device nodes that we requested resources for,
691 * so that we may easily release resources in the error path.
692 */
693struct sunxi_mc_smp_nodes {
694	struct device_node *prcm_node;
695	struct device_node *cpucfg_node;
696	struct device_node *sram_node;
697	struct device_node *r_cpucfg_node;
698};
699
700/* This structure holds SoC-specific bits tied to an enable-method string. */
701struct sunxi_mc_smp_data {
702	const char *enable_method;
703	int (*get_smp_nodes)(struct sunxi_mc_smp_nodes *nodes);
704	bool is_a83t;
705};
706
707static void __init sunxi_mc_smp_put_nodes(struct sunxi_mc_smp_nodes *nodes)
708{
709	of_node_put(nodes->prcm_node);
710	of_node_put(nodes->cpucfg_node);
711	of_node_put(nodes->sram_node);
712	of_node_put(nodes->r_cpucfg_node);
713	memset(nodes, 0, sizeof(*nodes));
714}
715
716static int __init sun9i_a80_get_smp_nodes(struct sunxi_mc_smp_nodes *nodes)
717{
718	nodes->prcm_node = of_find_compatible_node(NULL, NULL,
719						   "allwinner,sun9i-a80-prcm");
720	if (!nodes->prcm_node) {
721		pr_err("%s: PRCM not available\n", __func__);
722		return -ENODEV;
723	}
724
725	nodes->cpucfg_node = of_find_compatible_node(NULL, NULL,
726						     "allwinner,sun9i-a80-cpucfg");
727	if (!nodes->cpucfg_node) {
728		pr_err("%s: CPUCFG not available\n", __func__);
729		return -ENODEV;
730	}
731
732	nodes->sram_node = of_find_compatible_node(NULL, NULL,
733						   "allwinner,sun9i-a80-smp-sram");
734	if (!nodes->sram_node) {
735		pr_err("%s: Secure SRAM not available\n", __func__);
736		return -ENODEV;
737	}
738
739	return 0;
740}
741
742static int __init sun8i_a83t_get_smp_nodes(struct sunxi_mc_smp_nodes *nodes)
743{
744	nodes->prcm_node = of_find_compatible_node(NULL, NULL,
745						   "allwinner,sun8i-a83t-r-ccu");
746	if (!nodes->prcm_node) {
747		pr_err("%s: PRCM not available\n", __func__);
748		return -ENODEV;
749	}
750
751	nodes->cpucfg_node = of_find_compatible_node(NULL, NULL,
752						     "allwinner,sun8i-a83t-cpucfg");
753	if (!nodes->cpucfg_node) {
754		pr_err("%s: CPUCFG not available\n", __func__);
755		return -ENODEV;
756	}
757
758	nodes->r_cpucfg_node = of_find_compatible_node(NULL, NULL,
759						       "allwinner,sun8i-a83t-r-cpucfg");
760	if (!nodes->r_cpucfg_node) {
761		pr_err("%s: RCPUCFG not available\n", __func__);
762		return -ENODEV;
763	}
764
765	return 0;
766}
767
768static const struct sunxi_mc_smp_data sunxi_mc_smp_data[] __initconst = {
769	{
770		.enable_method	= "allwinner,sun9i-a80-smp",
771		.get_smp_nodes	= sun9i_a80_get_smp_nodes,
772	},
773	{
774		.enable_method	= "allwinner,sun8i-a83t-smp",
775		.get_smp_nodes	= sun8i_a83t_get_smp_nodes,
776		.is_a83t	= true,
777	},
778};
779
780static int __init sunxi_mc_smp_init(void)
781{
782	struct sunxi_mc_smp_nodes nodes = { 0 };
783	struct device_node *node;
784	struct resource res;
785	void __iomem *addr;
786	int i, ret;
787
788	/*
789	 * Don't bother checking the "cpus" node, as an enable-method
790	 * property in that node is undocumented.
791	 */
792	node = of_cpu_device_node_get(0);
793	if (!node)
794		return -ENODEV;
795
796	/*
797	 * We can't actually use the enable-method magic in the kernel.
798	 * Our loopback / trampoline code uses the CPU suspend framework,
799	 * which requires the identity mapping be available. It would not
800	 * yet be available if we used the .init_cpus or .prepare_cpus
801	 * callbacks in smp_operations, which we would use if we were to
802	 * use CPU_METHOD_OF_DECLARE
803	 */
804	for (i = 0; i < ARRAY_SIZE(sunxi_mc_smp_data); i++) {
805		ret = of_property_match_string(node, "enable-method",
806					       sunxi_mc_smp_data[i].enable_method);
807		if (ret >= 0)
808			break;
809	}
810
811	of_node_put(node);
812	if (ret < 0)
813		return -ENODEV;
814
815	is_a83t = sunxi_mc_smp_data[i].is_a83t;
816
817	if (!sunxi_mc_smp_cpu_table_init())
818		return -EINVAL;
819
820	if (!cci_probed()) {
821		pr_err("%s: CCI-400 not available\n", __func__);
822		return -ENODEV;
823	}
824
825	/* Get needed device tree nodes */
826	ret = sunxi_mc_smp_data[i].get_smp_nodes(&nodes);
827	if (ret)
828		goto err_put_nodes;
829
830	/*
831	 * Unfortunately we can not request the I/O region for the PRCM.
832	 * It is shared with the PRCM clock.
833	 */
834	prcm_base = of_iomap(nodes.prcm_node, 0);
835	if (!prcm_base) {
836		pr_err("%s: failed to map PRCM registers\n", __func__);
837		ret = -ENOMEM;
838		goto err_put_nodes;
839	}
840
841	cpucfg_base = of_io_request_and_map(nodes.cpucfg_node, 0,
842					    "sunxi-mc-smp");
843	if (IS_ERR(cpucfg_base)) {
844		ret = PTR_ERR(cpucfg_base);
845		pr_err("%s: failed to map CPUCFG registers: %d\n",
846		       __func__, ret);
847		goto err_unmap_prcm;
848	}
849
850	if (is_a83t) {
851		r_cpucfg_base = of_io_request_and_map(nodes.r_cpucfg_node,
852						      0, "sunxi-mc-smp");
853		if (IS_ERR(r_cpucfg_base)) {
854			ret = PTR_ERR(r_cpucfg_base);
855			pr_err("%s: failed to map R-CPUCFG registers\n",
856			       __func__);
857			goto err_unmap_release_cpucfg;
858		}
859	} else {
860		sram_b_smp_base = of_io_request_and_map(nodes.sram_node, 0,
861							"sunxi-mc-smp");
862		if (IS_ERR(sram_b_smp_base)) {
863			ret = PTR_ERR(sram_b_smp_base);
864			pr_err("%s: failed to map secure SRAM\n", __func__);
865			goto err_unmap_release_cpucfg;
866		}
867	}
868
869	/* Configure CCI-400 for boot cluster */
870	ret = sunxi_mc_smp_loopback();
871	if (ret) {
872		pr_err("%s: failed to configure boot cluster: %d\n",
873		       __func__, ret);
874		goto err_unmap_release_sram_rcpucfg;
875	}
876
877	/* We don't need the device nodes anymore */
878	sunxi_mc_smp_put_nodes(&nodes);
879
880	/* Set the hardware entry point address */
881	if (is_a83t)
882		addr = r_cpucfg_base + R_CPUCFG_CPU_SOFT_ENTRY_REG;
883	else
884		addr = prcm_base + PRCM_CPU_SOFT_ENTRY_REG;
885	writel(__pa_symbol(sunxi_mc_smp_secondary_startup), addr);
886
887	/* Actually enable multi cluster SMP */
888	smp_set_ops(&sunxi_mc_smp_smp_ops);
889
890	pr_info("sunxi multi cluster SMP support installed\n");
891
892	return 0;
893
894err_unmap_release_sram_rcpucfg:
895	if (is_a83t) {
896		iounmap(r_cpucfg_base);
897		of_address_to_resource(nodes.r_cpucfg_node, 0, &res);
898	} else {
899		iounmap(sram_b_smp_base);
900		of_address_to_resource(nodes.sram_node, 0, &res);
901	}
902	release_mem_region(res.start, resource_size(&res));
903err_unmap_release_cpucfg:
904	iounmap(cpucfg_base);
905	of_address_to_resource(nodes.cpucfg_node, 0, &res);
906	release_mem_region(res.start, resource_size(&res));
907err_unmap_prcm:
908	iounmap(prcm_base);
909err_put_nodes:
910	sunxi_mc_smp_put_nodes(&nodes);
911	return ret;
912}
913
914early_initcall(sunxi_mc_smp_init);
915