| 1 | // SPDX-License-Identifier: GPL-2.0-only |
| 2 | /* |
| 3 | * Copyright (C) 2005 Intel Corporation |
| 4 | * Venkatesh Pallipadi <venkatesh.pallipadi@intel.com> |
| 5 | * - Added _PDC for SMP C-states on Intel CPUs |
| 6 | */ |
| 7 | |
| 8 | #include <linux/kernel.h> |
| 9 | #include <linux/export.h> |
| 10 | #include <linux/init.h> |
| 11 | #include <linux/acpi.h> |
| 12 | #include <linux/cpu.h> |
| 13 | #include <linux/sched.h> |
| 14 | |
| 15 | #include <acpi/processor.h> |
| 16 | #include <asm/cpu_device_id.h> |
| 17 | #include <asm/cpuid/api.h> |
| 18 | #include <asm/mwait.h> |
| 19 | #include <asm/special_insns.h> |
| 20 | #include <asm/smp.h> |
| 21 | |
| 22 | /* |
| 23 | * Initialize bm_flags based on the CPU cache properties |
| 24 | * On SMP it depends on cache configuration |
| 25 | * - When cache is not shared among all CPUs, we flush cache |
| 26 | * before entering C3. |
| 27 | * - When cache is shared among all CPUs, we use bm_check |
| 28 | * mechanism as in UP case |
| 29 | * |
| 30 | * This routine is called only after all the CPUs are online |
| 31 | */ |
| 32 | void acpi_processor_power_init_bm_check(struct acpi_processor_flags *flags, |
| 33 | unsigned int cpu) |
| 34 | { |
| 35 | struct cpuinfo_x86 *c = &cpu_data(cpu); |
| 36 | |
| 37 | flags->bm_check = 0; |
| 38 | if (num_online_cpus() == 1) |
| 39 | flags->bm_check = 1; |
| 40 | else if (c->x86_vendor == X86_VENDOR_INTEL) { |
| 41 | /* |
| 42 | * Today all MP CPUs that support C3 share cache. |
| 43 | * And caches should not be flushed by software while |
| 44 | * entering C3 type state. |
| 45 | */ |
| 46 | flags->bm_check = 1; |
| 47 | } |
| 48 | |
| 49 | /* |
| 50 | * On all recent Intel platforms, ARB_DISABLE is a nop. |
| 51 | * So, set bm_control to zero to indicate that ARB_DISABLE |
| 52 | * is not required while entering C3 type state. |
| 53 | */ |
| 54 | if (c->x86_vendor == X86_VENDOR_INTEL && |
| 55 | (c->x86 > 15 || (c->x86_vfm >= INTEL_CORE2_MEROM && c->x86_vfm <= INTEL_FAM6_LAST))) |
| 56 | flags->bm_control = 0; |
| 57 | |
| 58 | if (c->x86_vendor == X86_VENDOR_CENTAUR) { |
| 59 | if (c->x86 > 6 || (c->x86 == 6 && c->x86_model == 0x0f && |
| 60 | c->x86_stepping >= 0x0e)) { |
| 61 | /* |
| 62 | * For all recent Centaur CPUs, the ucode will make sure that each |
| 63 | * core can keep cache coherence with each other while entering C3 |
| 64 | * type state. So, set bm_check to 1 to indicate that the kernel |
| 65 | * doesn't need to execute a cache flush operation (WBINVD) when |
| 66 | * entering C3 type state. |
| 67 | */ |
| 68 | flags->bm_check = 1; |
| 69 | /* |
| 70 | * For all recent Centaur platforms, ARB_DISABLE is a nop. |
| 71 | * Set bm_control to zero to indicate that ARB_DISABLE is |
| 72 | * not required while entering C3 type state. |
| 73 | */ |
| 74 | flags->bm_control = 0; |
| 75 | } |
| 76 | } |
| 77 | |
| 78 | if (c->x86_vendor == X86_VENDOR_ZHAOXIN) { |
| 79 | /* |
| 80 | * All Zhaoxin CPUs that support C3 share cache. |
| 81 | * And caches should not be flushed by software while |
| 82 | * entering C3 type state. |
| 83 | */ |
| 84 | flags->bm_check = 1; |
| 85 | /* |
| 86 | * On all recent Zhaoxin platforms, ARB_DISABLE is a nop. |
| 87 | * So, set bm_control to zero to indicate that ARB_DISABLE |
| 88 | * is not required while entering C3 type state. |
| 89 | */ |
| 90 | flags->bm_control = 0; |
| 91 | } |
| 92 | if (cpu_feature_enabled(X86_FEATURE_ZEN)) { |
| 93 | /* |
| 94 | * For all AMD Zen or newer CPUs that support C3, caches |
| 95 | * should not be flushed by software while entering C3 |
| 96 | * type state. Set bm->check to 1 so that kernel doesn't |
| 97 | * need to execute cache flush operation. |
| 98 | */ |
| 99 | flags->bm_check = 1; |
| 100 | /* |
| 101 | * In current AMD C state implementation ARB_DIS is no longer |
| 102 | * used. So set bm_control to zero to indicate ARB_DIS is not |
| 103 | * required while entering C3 type state. |
| 104 | */ |
| 105 | flags->bm_control = 0; |
| 106 | } |
| 107 | } |
| 108 | EXPORT_SYMBOL(acpi_processor_power_init_bm_check); |
| 109 | |
| 110 | /* The code below handles cstate entry with monitor-mwait pair on Intel*/ |
| 111 | |
| 112 | struct cstate_entry { |
| 113 | struct { |
| 114 | unsigned int eax; |
| 115 | unsigned int ecx; |
| 116 | } states[ACPI_PROCESSOR_MAX_POWER]; |
| 117 | }; |
| 118 | static struct cstate_entry __percpu *cpu_cstate_entry; /* per CPU ptr */ |
| 119 | |
| 120 | static short mwait_supported[ACPI_PROCESSOR_MAX_POWER]; |
| 121 | |
| 122 | #define NATIVE_CSTATE_BEYOND_HALT (2) |
| 123 | |
| 124 | static long acpi_processor_ffh_cstate_probe_cpu(void *_cx) |
| 125 | { |
| 126 | struct acpi_processor_cx *cx = _cx; |
| 127 | long retval; |
| 128 | unsigned int eax, ebx, ecx, edx; |
| 129 | unsigned int edx_part; |
| 130 | unsigned int cstate_type; /* C-state type and not ACPI C-state type */ |
| 131 | unsigned int num_cstate_subtype; |
| 132 | |
| 133 | cpuid(CPUID_LEAF_MWAIT, eax: &eax, ebx: &ebx, ecx: &ecx, edx: &edx); |
| 134 | |
| 135 | /* Check whether this particular cx_type (in CST) is supported or not */ |
| 136 | cstate_type = (((cx->address >> MWAIT_SUBSTATE_SIZE) & |
| 137 | MWAIT_CSTATE_MASK) + 1) & MWAIT_CSTATE_MASK; |
| 138 | edx_part = edx >> (cstate_type * MWAIT_SUBSTATE_SIZE); |
| 139 | num_cstate_subtype = edx_part & MWAIT_SUBSTATE_MASK; |
| 140 | |
| 141 | retval = 0; |
| 142 | /* If the HW does not support any sub-states in this C-state */ |
| 143 | if (num_cstate_subtype == 0) { |
| 144 | pr_warn(FW_BUG "ACPI MWAIT C-state 0x%x not supported by HW (0x%x)\n" , |
| 145 | cx->address, edx_part); |
| 146 | retval = -1; |
| 147 | goto out; |
| 148 | } |
| 149 | |
| 150 | /* mwait ecx extensions INTERRUPT_BREAK should be supported for C2/C3 */ |
| 151 | if (!(ecx & CPUID5_ECX_EXTENSIONS_SUPPORTED) || |
| 152 | !(ecx & CPUID5_ECX_INTERRUPT_BREAK)) { |
| 153 | retval = -1; |
| 154 | goto out; |
| 155 | } |
| 156 | |
| 157 | if (!mwait_supported[cstate_type]) { |
| 158 | mwait_supported[cstate_type] = 1; |
| 159 | printk(KERN_DEBUG |
| 160 | "Monitor-Mwait will be used to enter C-%d state\n" , |
| 161 | cx->type); |
| 162 | } |
| 163 | snprintf(buf: cx->desc, |
| 164 | ACPI_CX_DESC_LEN, fmt: "ACPI FFH MWAIT 0x%x" , |
| 165 | cx->address); |
| 166 | out: |
| 167 | return retval; |
| 168 | } |
| 169 | |
| 170 | int acpi_processor_ffh_cstate_probe(unsigned int cpu, |
| 171 | struct acpi_processor_cx *cx, struct acpi_power_register *reg) |
| 172 | { |
| 173 | struct cstate_entry *percpu_entry; |
| 174 | struct cpuinfo_x86 *c = &cpu_data(cpu); |
| 175 | long retval; |
| 176 | |
| 177 | if (!cpu_cstate_entry || c->cpuid_level < CPUID_LEAF_MWAIT) |
| 178 | return -1; |
| 179 | |
| 180 | if (reg->bit_offset != NATIVE_CSTATE_BEYOND_HALT) |
| 181 | return -1; |
| 182 | |
| 183 | percpu_entry = per_cpu_ptr(cpu_cstate_entry, cpu); |
| 184 | percpu_entry->states[cx->index].eax = 0; |
| 185 | percpu_entry->states[cx->index].ecx = 0; |
| 186 | |
| 187 | /* Make sure we are running on right CPU */ |
| 188 | |
| 189 | retval = call_on_cpu(cpu, fn: acpi_processor_ffh_cstate_probe_cpu, arg: cx, |
| 190 | direct: false); |
| 191 | if (retval == 0) { |
| 192 | /* Use the hint in CST */ |
| 193 | percpu_entry->states[cx->index].eax = cx->address; |
| 194 | percpu_entry->states[cx->index].ecx = MWAIT_ECX_INTERRUPT_BREAK; |
| 195 | } |
| 196 | |
| 197 | /* |
| 198 | * For _CST FFH on Intel, if GAS.access_size bit 1 is cleared, |
| 199 | * then we should skip checking BM_STS for this C-state. |
| 200 | * ref: "Intel Processor Vendor-Specific ACPI Interface Specification" |
| 201 | */ |
| 202 | if ((c->x86_vendor == X86_VENDOR_INTEL) && !(reg->access_size & 0x2)) |
| 203 | cx->bm_sts_skip = 1; |
| 204 | |
| 205 | return retval; |
| 206 | } |
| 207 | EXPORT_SYMBOL_GPL(acpi_processor_ffh_cstate_probe); |
| 208 | |
| 209 | void __noreturn acpi_processor_ffh_play_dead(struct acpi_processor_cx *cx) |
| 210 | { |
| 211 | unsigned int cpu = smp_processor_id(); |
| 212 | struct cstate_entry *percpu_entry; |
| 213 | |
| 214 | percpu_entry = per_cpu_ptr(cpu_cstate_entry, cpu); |
| 215 | mwait_play_dead(eax_hint: percpu_entry->states[cx->index].eax); |
| 216 | } |
| 217 | EXPORT_SYMBOL_GPL(acpi_processor_ffh_play_dead); |
| 218 | |
| 219 | void __cpuidle acpi_processor_ffh_cstate_enter(struct acpi_processor_cx *cx) |
| 220 | { |
| 221 | unsigned int cpu = smp_processor_id(); |
| 222 | struct cstate_entry *percpu_entry; |
| 223 | |
| 224 | percpu_entry = per_cpu_ptr(cpu_cstate_entry, cpu); |
| 225 | mwait_idle_with_hints(eax: percpu_entry->states[cx->index].eax, |
| 226 | ecx: percpu_entry->states[cx->index].ecx); |
| 227 | } |
| 228 | EXPORT_SYMBOL_GPL(acpi_processor_ffh_cstate_enter); |
| 229 | |
| 230 | static int __init ffh_cstate_init(void) |
| 231 | { |
| 232 | struct cpuinfo_x86 *c = &boot_cpu_data; |
| 233 | |
| 234 | if (c->x86_vendor != X86_VENDOR_INTEL && |
| 235 | c->x86_vendor != X86_VENDOR_AMD && |
| 236 | c->x86_vendor != X86_VENDOR_HYGON) |
| 237 | return -1; |
| 238 | |
| 239 | cpu_cstate_entry = alloc_percpu(struct cstate_entry); |
| 240 | return 0; |
| 241 | } |
| 242 | |
| 243 | static void __exit ffh_cstate_exit(void) |
| 244 | { |
| 245 | free_percpu(pdata: cpu_cstate_entry); |
| 246 | cpu_cstate_entry = NULL; |
| 247 | } |
| 248 | |
| 249 | arch_initcall(ffh_cstate_init); |
| 250 | __exitcall(ffh_cstate_exit); |
| 251 | |