| 1 | // SPDX-License-Identifier: GPL-2.0-only | 
|---|
| 2 | /* | 
|---|
| 3 | * x86 FPU boot time init code: | 
|---|
| 4 | */ | 
|---|
| 5 | #include <asm/fpu/api.h> | 
|---|
| 6 | #include <asm/tlbflush.h> | 
|---|
| 7 | #include <asm/setup.h> | 
|---|
| 8 |  | 
|---|
| 9 | #include <linux/sched.h> | 
|---|
| 10 | #include <linux/sched/task.h> | 
|---|
| 11 | #include <linux/init.h> | 
|---|
| 12 |  | 
|---|
| 13 | #include "internal.h" | 
|---|
| 14 | #include "legacy.h" | 
|---|
| 15 | #include "xstate.h" | 
|---|
| 16 |  | 
|---|
| 17 | /* | 
|---|
| 18 | * Initialize the registers found in all CPUs, CR0 and CR4: | 
|---|
| 19 | */ | 
|---|
| 20 | static void fpu__init_cpu_generic(void) | 
|---|
| 21 | { | 
|---|
| 22 | unsigned long cr0; | 
|---|
| 23 | unsigned long cr4_mask = 0; | 
|---|
| 24 |  | 
|---|
| 25 | if (boot_cpu_has(X86_FEATURE_FXSR)) | 
|---|
| 26 | cr4_mask |= X86_CR4_OSFXSR; | 
|---|
| 27 | if (boot_cpu_has(X86_FEATURE_XMM)) | 
|---|
| 28 | cr4_mask |= X86_CR4_OSXMMEXCPT; | 
|---|
| 29 | if (cr4_mask) | 
|---|
| 30 | cr4_set_bits(mask: cr4_mask); | 
|---|
| 31 |  | 
|---|
| 32 | cr0 = read_cr0(); | 
|---|
| 33 | cr0 &= ~(X86_CR0_TS|X86_CR0_EM); /* clear TS and EM */ | 
|---|
| 34 | if (!boot_cpu_has(X86_FEATURE_FPU)) | 
|---|
| 35 | cr0 |= X86_CR0_EM; | 
|---|
| 36 | write_cr0(x: cr0); | 
|---|
| 37 |  | 
|---|
| 38 | /* Flush out any pending x87 state: */ | 
|---|
| 39 | #ifdef CONFIG_MATH_EMULATION | 
|---|
| 40 | if (!boot_cpu_has(X86_FEATURE_FPU)) | 
|---|
| 41 | ; | 
|---|
| 42 | else | 
|---|
| 43 | #endif | 
|---|
| 44 | asm volatile ( "fninit"); | 
|---|
| 45 | } | 
|---|
| 46 |  | 
|---|
| 47 | /* | 
|---|
| 48 | * Enable all supported FPU features. Called when a CPU is brought online: | 
|---|
| 49 | */ | 
|---|
| 50 | void fpu__init_cpu(void) | 
|---|
| 51 | { | 
|---|
| 52 | fpu__init_cpu_generic(); | 
|---|
| 53 | fpu__init_cpu_xstate(); | 
|---|
| 54 |  | 
|---|
| 55 | /* Start allowing kernel-mode FPU: */ | 
|---|
| 56 | this_cpu_write(kernel_fpu_allowed, true); | 
|---|
| 57 | } | 
|---|
| 58 |  | 
|---|
| 59 | static bool __init fpu__probe_without_cpuid(void) | 
|---|
| 60 | { | 
|---|
| 61 | unsigned long cr0; | 
|---|
| 62 | u16 fsw, fcw; | 
|---|
| 63 |  | 
|---|
| 64 | fsw = fcw = 0xffff; | 
|---|
| 65 |  | 
|---|
| 66 | cr0 = read_cr0(); | 
|---|
| 67 | cr0 &= ~(X86_CR0_TS | X86_CR0_EM); | 
|---|
| 68 | write_cr0(x: cr0); | 
|---|
| 69 |  | 
|---|
| 70 | asm volatile( "fninit ; fnstsw %0 ; fnstcw %1": "+m"(fsw), "+m"(fcw)); | 
|---|
| 71 |  | 
|---|
| 72 | pr_info( "x86/fpu: Probing for FPU: FSW=0x%04hx FCW=0x%04hx\n", fsw, fcw); | 
|---|
| 73 |  | 
|---|
| 74 | return fsw == 0 && (fcw & 0x103f) == 0x003f; | 
|---|
| 75 | } | 
|---|
| 76 |  | 
|---|
| 77 | static void __init fpu__init_system_early_generic(void) | 
|---|
| 78 | { | 
|---|
| 79 | set_thread_flag(TIF_NEED_FPU_LOAD); | 
|---|
| 80 |  | 
|---|
| 81 | if (!boot_cpu_has(X86_FEATURE_CPUID) && | 
|---|
| 82 | !test_bit(X86_FEATURE_FPU, (unsigned long *)cpu_caps_cleared)) { | 
|---|
| 83 | if (fpu__probe_without_cpuid()) | 
|---|
| 84 | setup_force_cpu_cap(X86_FEATURE_FPU); | 
|---|
| 85 | else | 
|---|
| 86 | setup_clear_cpu_cap(X86_FEATURE_FPU); | 
|---|
| 87 | } | 
|---|
| 88 |  | 
|---|
| 89 | #ifndef CONFIG_MATH_EMULATION | 
|---|
| 90 | if (!test_cpu_cap(&boot_cpu_data, X86_FEATURE_FPU)) { | 
|---|
| 91 | pr_emerg( "x86/fpu: Giving up, no FPU found and no math emulation present\n"); | 
|---|
| 92 | for (;;) | 
|---|
| 93 | asm volatile( "hlt"); | 
|---|
| 94 | } | 
|---|
| 95 | #endif | 
|---|
| 96 | } | 
|---|
| 97 |  | 
|---|
| 98 | /* | 
|---|
| 99 | * Boot time FPU feature detection code: | 
|---|
| 100 | */ | 
|---|
| 101 | unsigned int mxcsr_feature_mask __ro_after_init = 0xffffffffu; | 
|---|
| 102 |  | 
|---|
| 103 | static void __init fpu__init_system_mxcsr(void) | 
|---|
| 104 | { | 
|---|
| 105 | unsigned int mask = 0; | 
|---|
| 106 |  | 
|---|
| 107 | if (boot_cpu_has(X86_FEATURE_FXSR)) { | 
|---|
| 108 | /* Static because GCC does not get 16-byte stack alignment right: */ | 
|---|
| 109 | static struct fxregs_state fxregs __initdata; | 
|---|
| 110 |  | 
|---|
| 111 | asm volatile( "fxsave %0": "+m"(fxregs)); | 
|---|
| 112 |  | 
|---|
| 113 | mask = fxregs.mxcsr_mask; | 
|---|
| 114 |  | 
|---|
| 115 | /* | 
|---|
| 116 | * If zero then use the default features mask, | 
|---|
| 117 | * which has all features set, except the | 
|---|
| 118 | * denormals-are-zero feature bit: | 
|---|
| 119 | */ | 
|---|
| 120 | if (mask == 0) | 
|---|
| 121 | mask = 0x0000ffbf; | 
|---|
| 122 | } | 
|---|
| 123 | mxcsr_feature_mask &= mask; | 
|---|
| 124 | } | 
|---|
| 125 |  | 
|---|
| 126 | /* | 
|---|
| 127 | * Once per bootup FPU initialization sequences that will run on most x86 CPUs: | 
|---|
| 128 | */ | 
|---|
| 129 | static void __init fpu__init_system_generic(void) | 
|---|
| 130 | { | 
|---|
| 131 | /* | 
|---|
| 132 | * Set up the legacy init FPU context. Will be updated when the | 
|---|
| 133 | * CPU supports XSAVE[S]. | 
|---|
| 134 | */ | 
|---|
| 135 | fpstate_init_user(fpstate: &init_fpstate); | 
|---|
| 136 |  | 
|---|
| 137 | fpu__init_system_mxcsr(); | 
|---|
| 138 | } | 
|---|
| 139 |  | 
|---|
| 140 | /* | 
|---|
| 141 | * Enforce that 'MEMBER' is the last field of 'TYPE'. | 
|---|
| 142 | * | 
|---|
| 143 | * Align the computed size with alignment of the TYPE, | 
|---|
| 144 | * because that's how C aligns structs. | 
|---|
| 145 | */ | 
|---|
| 146 | #define CHECK_MEMBER_AT_END_OF(TYPE, MEMBER) \ | 
|---|
| 147 | BUILD_BUG_ON(sizeof(TYPE) !=         \ | 
|---|
| 148 | ALIGN(offsetofend(TYPE, MEMBER), _Alignof(TYPE))) | 
|---|
| 149 |  | 
|---|
| 150 | /* | 
|---|
| 151 | * We append the 'struct fpu' to the task_struct: | 
|---|
| 152 | */ | 
|---|
| 153 | static void __init fpu__init_task_struct_size(void) | 
|---|
| 154 | { | 
|---|
| 155 | int task_size = sizeof(struct task_struct); | 
|---|
| 156 |  | 
|---|
| 157 | task_size += sizeof(struct fpu); | 
|---|
| 158 |  | 
|---|
| 159 | /* | 
|---|
| 160 | * Subtract off the static size of the register state. | 
|---|
| 161 | * It potentially has a bunch of padding. | 
|---|
| 162 | */ | 
|---|
| 163 | task_size -= sizeof(union fpregs_state); | 
|---|
| 164 |  | 
|---|
| 165 | /* | 
|---|
| 166 | * Add back the dynamically-calculated register state | 
|---|
| 167 | * size. | 
|---|
| 168 | */ | 
|---|
| 169 | task_size += fpu_kernel_cfg.default_size; | 
|---|
| 170 |  | 
|---|
| 171 | /* | 
|---|
| 172 | * We dynamically size 'struct fpu', so we require that | 
|---|
| 173 | * 'state' be at the end of 'it: | 
|---|
| 174 | */ | 
|---|
| 175 | CHECK_MEMBER_AT_END_OF(struct fpu, __fpstate); | 
|---|
| 176 |  | 
|---|
| 177 | arch_task_struct_size = task_size; | 
|---|
| 178 | } | 
|---|
| 179 |  | 
|---|
| 180 | /* | 
|---|
| 181 | * Set up the user and kernel xstate sizes based on the legacy FPU context size. | 
|---|
| 182 | * | 
|---|
| 183 | * We set this up first, and later it will be overwritten by | 
|---|
| 184 | * fpu__init_system_xstate() if the CPU knows about xstates. | 
|---|
| 185 | */ | 
|---|
| 186 | static void __init fpu__init_system_xstate_size_legacy(void) | 
|---|
| 187 | { | 
|---|
| 188 | unsigned int size; | 
|---|
| 189 |  | 
|---|
| 190 | /* | 
|---|
| 191 | * Note that the size configuration might be overwritten later | 
|---|
| 192 | * during fpu__init_system_xstate(). | 
|---|
| 193 | */ | 
|---|
| 194 | if (!cpu_feature_enabled(X86_FEATURE_FPU)) { | 
|---|
| 195 | size = sizeof(struct swregs_state); | 
|---|
| 196 | } else if (cpu_feature_enabled(X86_FEATURE_FXSR)) { | 
|---|
| 197 | size = sizeof(struct fxregs_state); | 
|---|
| 198 | fpu_user_cfg.legacy_features = XFEATURE_MASK_FPSSE; | 
|---|
| 199 | } else { | 
|---|
| 200 | size = sizeof(struct fregs_state); | 
|---|
| 201 | fpu_user_cfg.legacy_features = XFEATURE_MASK_FP; | 
|---|
| 202 | } | 
|---|
| 203 |  | 
|---|
| 204 | fpu_kernel_cfg.max_size = size; | 
|---|
| 205 | fpu_kernel_cfg.default_size = size; | 
|---|
| 206 | fpu_user_cfg.max_size = size; | 
|---|
| 207 | fpu_user_cfg.default_size = size; | 
|---|
| 208 | guest_default_cfg.size = size; | 
|---|
| 209 | } | 
|---|
| 210 |  | 
|---|
| 211 | /* | 
|---|
| 212 | * Called on the boot CPU once per system bootup, to set up the initial | 
|---|
| 213 | * FPU state that is later cloned into all processes: | 
|---|
| 214 | */ | 
|---|
| 215 | void __init fpu__init_system(void) | 
|---|
| 216 | { | 
|---|
| 217 | fpu__init_system_early_generic(); | 
|---|
| 218 |  | 
|---|
| 219 | /* | 
|---|
| 220 | * The FPU has to be operational for some of the | 
|---|
| 221 | * later FPU init activities: | 
|---|
| 222 | */ | 
|---|
| 223 | fpu__init_cpu(); | 
|---|
| 224 |  | 
|---|
| 225 | fpu__init_system_generic(); | 
|---|
| 226 | fpu__init_system_xstate_size_legacy(); | 
|---|
| 227 | fpu__init_system_xstate(legacy_size: fpu_kernel_cfg.max_size); | 
|---|
| 228 | fpu__init_task_struct_size(); | 
|---|
| 229 | } | 
|---|
| 230 |  | 
|---|