Nicolas Pitre | 1c33be5 | 2012-04-12 02:56:10 -0400 | [diff] [blame^] | 1 | /* |
| 2 | * arch/arm/common/bL_switcher.c -- big.LITTLE cluster switcher core driver |
| 3 | * |
| 4 | * Created by: Nicolas Pitre, March 2012 |
| 5 | * Copyright: (C) 2012-2013 Linaro Limited |
| 6 | * |
| 7 | * This program is free software; you can redistribute it and/or modify |
| 8 | * it under the terms of the GNU General Public License version 2 as |
| 9 | * published by the Free Software Foundation. |
| 10 | */ |
| 11 | |
| 12 | #include <linux/init.h> |
| 13 | #include <linux/kernel.h> |
| 14 | #include <linux/module.h> |
| 15 | #include <linux/sched.h> |
| 16 | #include <linux/interrupt.h> |
| 17 | #include <linux/cpu_pm.h> |
| 18 | #include <linux/workqueue.h> |
| 19 | #include <linux/mm.h> |
| 20 | #include <linux/string.h> |
| 21 | #include <linux/irqchip/arm-gic.h> |
| 22 | |
| 23 | #include <asm/smp_plat.h> |
| 24 | #include <asm/suspend.h> |
| 25 | #include <asm/mcpm.h> |
| 26 | #include <asm/bL_switcher.h> |
| 27 | |
| 28 | |
| 29 | /* |
| 30 | * Use our own MPIDR accessors as the generic ones in asm/cputype.h have |
| 31 | * __attribute_const__ and we don't want the compiler to assume any |
| 32 | * constness here as the value _does_ change along some code paths. |
| 33 | */ |
| 34 | |
| 35 | static int read_mpidr(void) |
| 36 | { |
| 37 | unsigned int id; |
| 38 | asm volatile ("mrc p15, 0, %0, c0, c0, 5" : "=r" (id)); |
| 39 | return id & MPIDR_HWID_BITMASK; |
| 40 | } |
| 41 | |
| 42 | /* |
| 43 | * bL switcher core code. |
| 44 | */ |
| 45 | |
| 46 | static void bL_do_switch(void *_unused) |
| 47 | { |
| 48 | unsigned mpidr, cpuid, clusterid, ob_cluster, ib_cluster; |
| 49 | |
| 50 | /* |
| 51 | * We now have a piece of stack borrowed from the init task's. |
| 52 | * Let's also switch to init_mm right away to match it. |
| 53 | */ |
| 54 | cpu_switch_mm(init_mm.pgd, &init_mm); |
| 55 | |
| 56 | pr_debug("%s\n", __func__); |
| 57 | |
| 58 | mpidr = read_mpidr(); |
| 59 | cpuid = MPIDR_AFFINITY_LEVEL(mpidr, 0); |
| 60 | clusterid = MPIDR_AFFINITY_LEVEL(mpidr, 1); |
| 61 | ob_cluster = clusterid; |
| 62 | ib_cluster = clusterid ^ 1; |
| 63 | |
| 64 | /* |
| 65 | * Our state has been saved at this point. Let's release our |
| 66 | * inbound CPU. |
| 67 | */ |
| 68 | mcpm_set_entry_vector(cpuid, ib_cluster, cpu_resume); |
| 69 | sev(); |
| 70 | |
| 71 | /* |
| 72 | * From this point, we must assume that our counterpart CPU might |
| 73 | * have taken over in its parallel world already, as if execution |
| 74 | * just returned from cpu_suspend(). It is therefore important to |
| 75 | * be very careful not to make any change the other guy is not |
| 76 | * expecting. This is why we need stack isolation. |
| 77 | * |
| 78 | * Fancy under cover tasks could be performed here. For now |
| 79 | * we have none. |
| 80 | */ |
| 81 | |
| 82 | /* Let's put ourself down. */ |
| 83 | mcpm_cpu_power_down(); |
| 84 | |
| 85 | /* should never get here */ |
| 86 | BUG(); |
| 87 | } |
| 88 | |
| 89 | /* |
| 90 | * Stack isolation. To ensure 'current' remains valid, we just borrow |
| 91 | * a slice of the init/idle task which should be fairly lightly used. |
| 92 | * The borrowed area starts just above the thread_info structure located |
| 93 | * at the very bottom of the stack, aligned to a cache line. |
| 94 | */ |
| 95 | #define STACK_SIZE 256 |
| 96 | extern void call_with_stack(void (*fn)(void *), void *arg, void *sp); |
| 97 | static int bL_switchpoint(unsigned long _arg) |
| 98 | { |
| 99 | unsigned int mpidr = read_mpidr(); |
| 100 | unsigned int cpuid = MPIDR_AFFINITY_LEVEL(mpidr, 0); |
| 101 | unsigned int clusterid = MPIDR_AFFINITY_LEVEL(mpidr, 1); |
| 102 | unsigned int cpu_index = cpuid + clusterid * MAX_CPUS_PER_CLUSTER; |
| 103 | void *stack = &init_thread_info + 1; |
| 104 | stack = PTR_ALIGN(stack, L1_CACHE_BYTES); |
| 105 | stack += cpu_index * STACK_SIZE + STACK_SIZE; |
| 106 | call_with_stack(bL_do_switch, (void *)_arg, stack); |
| 107 | BUG(); |
| 108 | } |
| 109 | |
| 110 | /* |
| 111 | * Generic switcher interface |
| 112 | */ |
| 113 | |
| 114 | /* |
| 115 | * bL_switch_to - Switch to a specific cluster for the current CPU |
| 116 | * @new_cluster_id: the ID of the cluster to switch to. |
| 117 | * |
| 118 | * This function must be called on the CPU to be switched. |
| 119 | * Returns 0 on success, else a negative status code. |
| 120 | */ |
| 121 | static int bL_switch_to(unsigned int new_cluster_id) |
| 122 | { |
| 123 | unsigned int mpidr, cpuid, clusterid, ob_cluster, ib_cluster, this_cpu; |
| 124 | int ret; |
| 125 | |
| 126 | mpidr = read_mpidr(); |
| 127 | cpuid = MPIDR_AFFINITY_LEVEL(mpidr, 0); |
| 128 | clusterid = MPIDR_AFFINITY_LEVEL(mpidr, 1); |
| 129 | ob_cluster = clusterid; |
| 130 | ib_cluster = clusterid ^ 1; |
| 131 | |
| 132 | if (new_cluster_id == clusterid) |
| 133 | return 0; |
| 134 | |
| 135 | pr_debug("before switch: CPU %d in cluster %d\n", cpuid, clusterid); |
| 136 | |
| 137 | /* Close the gate for our entry vectors */ |
| 138 | mcpm_set_entry_vector(cpuid, ob_cluster, NULL); |
| 139 | mcpm_set_entry_vector(cpuid, ib_cluster, NULL); |
| 140 | |
| 141 | /* |
| 142 | * Let's wake up the inbound CPU now in case it requires some delay |
| 143 | * to come online, but leave it gated in our entry vector code. |
| 144 | */ |
| 145 | ret = mcpm_cpu_power_up(cpuid, ib_cluster); |
| 146 | if (ret) { |
| 147 | pr_err("%s: mcpm_cpu_power_up() returned %d\n", __func__, ret); |
| 148 | return ret; |
| 149 | } |
| 150 | |
| 151 | /* |
| 152 | * From this point we are entering the switch critical zone |
| 153 | * and can't take any interrupts anymore. |
| 154 | */ |
| 155 | local_irq_disable(); |
| 156 | local_fiq_disable(); |
| 157 | |
| 158 | this_cpu = smp_processor_id(); |
| 159 | |
| 160 | /* redirect GIC's SGIs to our counterpart */ |
| 161 | gic_migrate_target(cpuid + ib_cluster*4); |
| 162 | |
| 163 | /* |
| 164 | * Raise a SGI on the inbound CPU to make sure it doesn't stall |
| 165 | * in a possible WFI, such as in mcpm_power_down(). |
| 166 | */ |
| 167 | arch_send_wakeup_ipi_mask(cpumask_of(this_cpu)); |
| 168 | |
| 169 | ret = cpu_pm_enter(); |
| 170 | |
| 171 | /* we can not tolerate errors at this point */ |
| 172 | if (ret) |
| 173 | panic("%s: cpu_pm_enter() returned %d\n", __func__, ret); |
| 174 | |
| 175 | /* Flip the cluster in the CPU logical map for this CPU. */ |
| 176 | cpu_logical_map(this_cpu) ^= (1 << 8); |
| 177 | |
| 178 | /* Let's do the actual CPU switch. */ |
| 179 | ret = cpu_suspend(0, bL_switchpoint); |
| 180 | if (ret > 0) |
| 181 | panic("%s: cpu_suspend() returned %d\n", __func__, ret); |
| 182 | |
| 183 | /* We are executing on the inbound CPU at this point */ |
| 184 | mpidr = read_mpidr(); |
| 185 | cpuid = MPIDR_AFFINITY_LEVEL(mpidr, 0); |
| 186 | clusterid = MPIDR_AFFINITY_LEVEL(mpidr, 1); |
| 187 | pr_debug("after switch: CPU %d in cluster %d\n", cpuid, clusterid); |
| 188 | BUG_ON(clusterid != ib_cluster); |
| 189 | |
| 190 | mcpm_cpu_powered_up(); |
| 191 | |
| 192 | ret = cpu_pm_exit(); |
| 193 | |
| 194 | local_fiq_enable(); |
| 195 | local_irq_enable(); |
| 196 | |
| 197 | if (ret) |
| 198 | pr_err("%s exiting with error %d\n", __func__, ret); |
| 199 | return ret; |
| 200 | } |
| 201 | |
| 202 | struct switch_args { |
| 203 | unsigned int cluster; |
| 204 | struct work_struct work; |
| 205 | }; |
| 206 | |
| 207 | static void __bL_switch_to(struct work_struct *work) |
| 208 | { |
| 209 | struct switch_args *args = container_of(work, struct switch_args, work); |
| 210 | bL_switch_to(args->cluster); |
| 211 | } |
| 212 | |
| 213 | /* |
| 214 | * bL_switch_request - Switch to a specific cluster for the given CPU |
| 215 | * |
| 216 | * @cpu: the CPU to switch |
| 217 | * @new_cluster_id: the ID of the cluster to switch to. |
| 218 | * |
| 219 | * This function causes a cluster switch on the given CPU. If the given |
| 220 | * CPU is the same as the calling CPU then the switch happens right away. |
| 221 | * Otherwise the request is put on a work queue to be scheduled on the |
| 222 | * remote CPU. |
| 223 | */ |
| 224 | void bL_switch_request(unsigned int cpu, unsigned int new_cluster_id) |
| 225 | { |
| 226 | unsigned int this_cpu = get_cpu(); |
| 227 | struct switch_args args; |
| 228 | |
| 229 | if (cpu == this_cpu) { |
| 230 | bL_switch_to(new_cluster_id); |
| 231 | put_cpu(); |
| 232 | return; |
| 233 | } |
| 234 | put_cpu(); |
| 235 | |
| 236 | args.cluster = new_cluster_id; |
| 237 | INIT_WORK_ONSTACK(&args.work, __bL_switch_to); |
| 238 | schedule_work_on(cpu, &args.work); |
| 239 | flush_work(&args.work); |
| 240 | } |
| 241 | EXPORT_SYMBOL_GPL(bL_switch_request); |