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 29 /* 30 * The common v7_exit_coherency_flush API could not be used because of the 31 * Erratum 799270 workaround. This macro is the same as the common one (in 32 * arch/arm/include/asm/cacheflush.h) except for the erratum handling. 33 */ 34 #define exynos_v7_exit_coherency_flush(level) \ 35 asm volatile( \ 36 "stmfd sp!, {fp, ip}\n\t"\ 37 "mrc p15, 0, r0, c1, c0, 0 @ get SCTLR\n\t" \ 38 "bic r0, r0, #"__stringify(CR_C)"\n\t" \ 39 "mcr p15, 0, r0, c1, c0, 0 @ set SCTLR\n\t" \ 40 "isb\n\t"\ 41 "bl v7_flush_dcache_"__stringify(level)"\n\t" \ 42 "clrex\n\t"\ 43 "mrc p15, 0, r0, c1, c0, 1 @ get ACTLR\n\t" \ 44 "bic r0, r0, #(1 << 6) @ disable local coherency\n\t" \ 45 /* Dummy Load of a device register to avoid Erratum 799270 */ \ 46 "ldr r4, [%0]\n\t" \ 47 "and r4, r4, #0\n\t" \ 48 "orr r0, r0, r4\n\t" \ 49 "mcr p15, 0, r0, c1, c0, 1 @ set ACTLR\n\t" \ 50 "isb\n\t" \ 51 "dsb\n\t" \ 52 "ldmfd sp!, {fp, ip}" \ 53 : \ 54 : "Ir" (S5P_INFORM0) \ 55 : "r0", "r1", "r2", "r3", "r4", "r5", "r6", "r7", \ 56 "r9", "r10", "lr", "memory") 57 58 /* 59 * We can't use regular spinlocks. In the switcher case, it is possible 60 * for an outbound CPU to call power_down() after its inbound counterpart 61 * is already live using the same logical CPU number which trips lockdep 62 * debugging. 63 */ 64 static arch_spinlock_t exynos_mcpm_lock = __ARCH_SPIN_LOCK_UNLOCKED; 65 static int 66 cpu_use_count[EXYNOS5420_CPUS_PER_CLUSTER][EXYNOS5420_NR_CLUSTERS]; 67 68 #define exynos_cluster_usecnt(cluster) \ 69 (cpu_use_count[0][cluster] + \ 70 cpu_use_count[1][cluster] + \ 71 cpu_use_count[2][cluster] + \ 72 cpu_use_count[3][cluster]) 73 74 #define exynos_cluster_unused(cluster) !exynos_cluster_usecnt(cluster) 75 76 static int exynos_cluster_power_control(unsigned int cluster, int enable) 77 { 78 unsigned int tries = 100; 79 unsigned int val; 80 81 if (enable) { 82 exynos_cluster_power_up(cluster); 83 val = S5P_CORE_LOCAL_PWR_EN; 84 } else { 85 exynos_cluster_power_down(cluster); 86 val = 0; 87 } 88 89 /* Wait until cluster power control is applied */ 90 while (tries--) { 91 if (exynos_cluster_power_state(cluster) == val) 92 return 0; 93 94 cpu_relax(); 95 } 96 pr_debug("timed out waiting for cluster %u to power %s\n", cluster, 97 enable ? "on" : "off"); 98 99 return -ETIMEDOUT; 100 } 101 102 static int exynos_power_up(unsigned int cpu, unsigned int cluster) 103 { 104 unsigned int cpunr = cpu + (cluster * EXYNOS5420_CPUS_PER_CLUSTER); 105 int err = 0; 106 107 pr_debug("%s: cpu %u cluster %u\n", __func__, cpu, cluster); 108 if (cpu >= EXYNOS5420_CPUS_PER_CLUSTER || 109 cluster >= EXYNOS5420_NR_CLUSTERS) 110 return -EINVAL; 111 112 /* 113 * Since this is called with IRQs enabled, and no arch_spin_lock_irq 114 * variant exists, we need to disable IRQs manually here. 115 */ 116 local_irq_disable(); 117 arch_spin_lock(&exynos_mcpm_lock); 118 119 cpu_use_count[cpu][cluster]++; 120 if (cpu_use_count[cpu][cluster] == 1) { 121 bool was_cluster_down = 122 (exynos_cluster_usecnt(cluster) == 1); 123 124 /* 125 * Turn on the cluster (L2/COMMON) and then power on the 126 * cores. 127 */ 128 if (was_cluster_down) 129 err = exynos_cluster_power_control(cluster, 1); 130 131 if (!err) 132 exynos_cpu_power_up(cpunr); 133 else 134 exynos_cluster_power_control(cluster, 0); 135 } else if (cpu_use_count[cpu][cluster] != 2) { 136 /* 137 * The only possible values are: 138 * 0 = CPU down 139 * 1 = CPU (still) up 140 * 2 = CPU requested to be up before it had a chance 141 * to actually make itself down. 142 * Any other value is a bug. 143 */ 144 BUG(); 145 } 146 147 arch_spin_unlock(&exynos_mcpm_lock); 148 local_irq_enable(); 149 150 return err; 151 } 152 153 /* 154 * NOTE: This function requires the stack data to be visible through power down 155 * and can only be executed on processors like A15 and A7 that hit the cache 156 * with the C bit clear in the SCTLR register. 157 */ 158 static void exynos_power_down(void) 159 { 160 unsigned int mpidr, cpu, cluster; 161 bool last_man = false, skip_wfi = false; 162 unsigned int cpunr; 163 164 mpidr = read_cpuid_mpidr(); 165 cpu = MPIDR_AFFINITY_LEVEL(mpidr, 0); 166 cluster = MPIDR_AFFINITY_LEVEL(mpidr, 1); 167 cpunr = cpu + (cluster * EXYNOS5420_CPUS_PER_CLUSTER); 168 169 pr_debug("%s: cpu %u cluster %u\n", __func__, cpu, cluster); 170 BUG_ON(cpu >= EXYNOS5420_CPUS_PER_CLUSTER || 171 cluster >= EXYNOS5420_NR_CLUSTERS); 172 173 __mcpm_cpu_going_down(cpu, cluster); 174 175 arch_spin_lock(&exynos_mcpm_lock); 176 BUG_ON(__mcpm_cluster_state(cluster) != CLUSTER_UP); 177 cpu_use_count[cpu][cluster]--; 178 if (cpu_use_count[cpu][cluster] == 0) { 179 exynos_cpu_power_down(cpunr); 180 181 if (exynos_cluster_unused(cluster)) 182 /* TODO: Turn off the cluster here to save power. */ 183 last_man = true; 184 } else if (cpu_use_count[cpu][cluster] == 1) { 185 /* 186 * A power_up request went ahead of us. 187 * Even if we do not want to shut this CPU down, 188 * the caller expects a certain state as if the WFI 189 * was aborted. So let's continue with cache cleaning. 190 */ 191 skip_wfi = true; 192 } else { 193 BUG(); 194 } 195 196 if (last_man && __mcpm_outbound_enter_critical(cpu, cluster)) { 197 arch_spin_unlock(&exynos_mcpm_lock); 198 199 if (read_cpuid_part() == ARM_CPU_PART_CORTEX_A15) { 200 /* 201 * On the Cortex-A15 we need to disable 202 * L2 prefetching before flushing the cache. 203 */ 204 asm volatile( 205 "mcr p15, 1, %0, c15, c0, 3\n\t" 206 "isb\n\t" 207 "dsb" 208 : : "r" (0x400)); 209 } 210 211 /* Flush all cache levels for this cluster. */ 212 exynos_v7_exit_coherency_flush(all); 213 214 /* 215 * Disable cluster-level coherency by masking 216 * incoming snoops and DVM messages: 217 */ 218 cci_disable_port_by_cpu(mpidr); 219 220 __mcpm_outbound_leave_critical(cluster, CLUSTER_DOWN); 221 } else { 222 arch_spin_unlock(&exynos_mcpm_lock); 223 224 /* Disable and flush the local CPU cache. */ 225 exynos_v7_exit_coherency_flush(louis); 226 } 227 228 __mcpm_cpu_down(cpu, cluster); 229 230 /* Now we are prepared for power-down, do it: */ 231 if (!skip_wfi) 232 wfi(); 233 234 /* Not dead at this point? Let our caller cope. */ 235 } 236 237 static int exynos_wait_for_powerdown(unsigned int cpu, unsigned int cluster) 238 { 239 unsigned int tries = 100; 240 unsigned int cpunr = cpu + (cluster * EXYNOS5420_CPUS_PER_CLUSTER); 241 242 pr_debug("%s: cpu %u cluster %u\n", __func__, cpu, cluster); 243 BUG_ON(cpu >= EXYNOS5420_CPUS_PER_CLUSTER || 244 cluster >= EXYNOS5420_NR_CLUSTERS); 245 246 /* Wait for the core state to be OFF */ 247 while (tries--) { 248 if (ACCESS_ONCE(cpu_use_count[cpu][cluster]) == 0) { 249 if ((exynos_cpu_power_state(cpunr) == 0)) 250 return 0; /* success: the CPU is halted */ 251 } 252 253 /* Otherwise, wait and retry: */ 254 msleep(1); 255 } 256 257 return -ETIMEDOUT; /* timeout */ 258 } 259 260 static const struct mcpm_platform_ops exynos_power_ops = { 261 .power_up = exynos_power_up, 262 .power_down = exynos_power_down, 263 .wait_for_powerdown = exynos_wait_for_powerdown, 264 }; 265 266 static void __init exynos_mcpm_usage_count_init(void) 267 { 268 unsigned int mpidr, cpu, cluster; 269 270 mpidr = read_cpuid_mpidr(); 271 cpu = MPIDR_AFFINITY_LEVEL(mpidr, 0); 272 cluster = MPIDR_AFFINITY_LEVEL(mpidr, 1); 273 274 pr_debug("%s: cpu %u cluster %u\n", __func__, cpu, cluster); 275 BUG_ON(cpu >= EXYNOS5420_CPUS_PER_CLUSTER || 276 cluster >= EXYNOS5420_NR_CLUSTERS); 277 278 cpu_use_count[cpu][cluster] = 1; 279 } 280 281 /* 282 * Enable cluster-level coherency, in preparation for turning on the MMU. 283 */ 284 static void __naked exynos_pm_power_up_setup(unsigned int affinity_level) 285 { 286 asm volatile ("\n" 287 "cmp r0, #1\n" 288 "bxne lr\n" 289 "b cci_enable_port_for_self"); 290 } 291 292 static void __init exynos_cache_off(void) 293 { 294 if (read_cpuid_part() == ARM_CPU_PART_CORTEX_A15) { 295 /* disable L2 prefetching on the Cortex-A15 */ 296 asm volatile( 297 "mcr p15, 1, %0, c15, c0, 3\n\t" 298 "isb\n\t" 299 "dsb" 300 : : "r" (0x400)); 301 } 302 exynos_v7_exit_coherency_flush(all); 303 } 304 305 static const struct of_device_id exynos_dt_mcpm_match[] = { 306 { .compatible = "samsung,exynos5420" }, 307 { .compatible = "samsung,exynos5800" }, 308 {}, 309 }; 310 311 static int __init exynos_mcpm_init(void) 312 { 313 struct device_node *node; 314 void __iomem *ns_sram_base_addr; 315 int ret; 316 317 node = of_find_matching_node(NULL, exynos_dt_mcpm_match); 318 if (!node) 319 return -ENODEV; 320 of_node_put(node); 321 322 if (!cci_probed()) 323 return -ENODEV; 324 325 node = of_find_compatible_node(NULL, NULL, 326 "samsung,exynos4210-sysram-ns"); 327 if (!node) 328 return -ENODEV; 329 330 ns_sram_base_addr = of_iomap(node, 0); 331 of_node_put(node); 332 if (!ns_sram_base_addr) { 333 pr_err("failed to map non-secure iRAM base address\n"); 334 return -ENOMEM; 335 } 336 337 /* 338 * To increase the stability of KFC reset we need to program 339 * the PMU SPARE3 register 340 */ 341 __raw_writel(EXYNOS5420_SWRESET_KFC_SEL, S5P_PMU_SPARE3); 342 343 exynos_mcpm_usage_count_init(); 344 345 ret = mcpm_platform_register(&exynos_power_ops); 346 if (!ret) 347 ret = mcpm_sync_init(exynos_pm_power_up_setup); 348 if (!ret) 349 ret = mcpm_loopback(exynos_cache_off); /* turn on the CCI */ 350 if (ret) { 351 iounmap(ns_sram_base_addr); 352 return ret; 353 } 354 355 mcpm_smp_set_ops(); 356 357 pr_info("Exynos MCPM support installed\n"); 358 359 /* 360 * U-Boot SPL is hardcoded to jump to the start of ns_sram_base_addr 361 * as part of secondary_cpu_start(). Let's redirect it to the 362 * mcpm_entry_point(). 363 */ 364 __raw_writel(0xe59f0000, ns_sram_base_addr); /* ldr r0, [pc, #0] */ 365 __raw_writel(0xe12fff10, ns_sram_base_addr + 4); /* bx r0 */ 366 __raw_writel(virt_to_phys(mcpm_entry_point), ns_sram_base_addr + 8); 367 368 iounmap(ns_sram_base_addr); 369 370 return ret; 371 } 372 373 early_initcall(exynos_mcpm_init); 374