arch/x86/kernel/fpu/init.c

   1 /*
   2  * x86 FPU boot time init code:
   3  */
   4 #include <asm/fpu/internal.h>
   5 #include <asm/tlbflush.h>
   6
   7 #include <linux/sched.h>
   8
   9 /*
  10  * Initialize the TS bit in CR0 according to the style of context-switches
  11  * we are using:
  12  */
  13 static void fpu__init_cpu_ctx_switch(void)
  14 {
  15         if (!cpu_has_eager_fpu)
  16                 stts();
  17         else
  18                 clts();
  19 }
  20
  21 /*
  22  * Initialize the registers found in all CPUs, CR0 and CR4:
  23  */
  24 static void fpu__init_cpu_generic(void)
  25 {
  26         unsigned long cr0;
  27         unsigned long cr4_mask = 0;
  28
  29         if (cpu_has_fxsr)
  30                 cr4_mask |= X86_CR4_OSFXSR;
  31         if (cpu_has_xmm)
  32                 cr4_mask |= X86_CR4_OSXMMEXCPT;
  33         if (cr4_mask)
  34                 cr4_set_bits(cr4_mask);
  35
  36         cr0 = read_cr0();
  37         cr0 &= ~(X86_CR0_TS|X86_CR0_EM); /* clear TS and EM */
  38         if (!cpu_has_fpu)
  39                 cr0 |= X86_CR0_EM;
  40         write_cr0(cr0);
  41
  42         /* Flush out any pending x87 state: */
  43 #ifdef CONFIG_MATH_EMULATION
  44         if (!cpu_has_fpu)
  45                 fpstate_init_soft(&current->thread.fpu.state.soft);
  46         else
  47 #endif
  48                 asm volatile ("fninit");
  49 }
  50
  51 /*
  52  * Enable all supported FPU features. Called when a CPU is brought online:
  53  */
  54 void fpu__init_cpu(void)
  55 {
  56         fpu__init_cpu_generic();
  57         fpu__init_cpu_xstate();
  58         fpu__init_cpu_ctx_switch();
  59 }
  60
  61 /*
  62  * The earliest FPU detection code.
  63  *
  64  * Set the X86_FEATURE_FPU CPU-capability bit based on
  65  * trying to execute an actual sequence of FPU instructions:
  66  */
  67 static void fpu__init_system_early_generic(struct cpuinfo_x86 *c)
  68 {
  69         unsigned long cr0;
  70         u16 fsw, fcw;
  71
  72         fsw = fcw = 0xffff;
  73
  74         cr0 = read_cr0();
  75         cr0 &= ~(X86_CR0_TS | X86_CR0_EM);
  76         write_cr0(cr0);
  77
  78         asm volatile("fninit ; fnstsw %0 ; fnstcw %1"
  79                      : "+m" (fsw), "+m" (fcw));
  80
  81         if (fsw == 0 && (fcw & 0x103f) == 0x003f)
  82                 set_cpu_cap(c, X86_FEATURE_FPU);
  83         else
  84                 clear_cpu_cap(c, X86_FEATURE_FPU);
  85
  86 #ifndef CONFIG_MATH_EMULATION
  87         if (!cpu_has_fpu) {
  88                 pr_emerg("x86/fpu: Giving up, no FPU found and no math emulation present\n");
  89                 for (;;)
  90                         asm volatile("hlt");
  91         }
  92 #endif
  93 }
  94
  95 /*
  96  * Boot time FPU feature detection code:
  97  */
  98 unsigned int mxcsr_feature_mask __read_mostly = 0xffffffffu;
  99
 100 static void __init fpu__init_system_mxcsr(void)
 101 {
 102         unsigned int mask = 0;
 103
 104         if (cpu_has_fxsr) {
 105                 /* Static because GCC does not get 16-byte stack alignment right: */
 106                 static struct fxregs_state fxregs __initdata;
 107
 108                 asm volatile("fxsave %0" : "+m" (fxregs));
 109
 110                 mask = fxregs.mxcsr_mask;
 111
 112                 /*
 113                  * If zero then use the default features mask,
 114                  * which has all features set, except the
 115                  * denormals-are-zero feature bit:
 116                  */
 117                 if (mask == 0)
 118                         mask = 0x0000ffbf;
 119         }
 120         mxcsr_feature_mask &= mask;
 121 }
 122
 123 /*
 124  * Once per bootup FPU initialization sequences that will run on most x86 CPUs:
 125  */
 126 static void __init fpu__init_system_generic(void)
 127 {
 128         /*
 129          * Set up the legacy init FPU context. (xstate init might overwrite this
 130          * with a more modern format, if the CPU supports it.)
 131          */
 132         fpstate_init_fxstate(&init_fpstate.fxsave);
 133
 134         fpu__init_system_mxcsr();
 135 }
 136
 137 /*
 138  * Size of the FPU context state. All tasks in the system use the
 139  * same context size, regardless of what portion they use.
 140  * This is inherent to the XSAVE architecture which puts all state
 141  * components into a single, continuous memory block:
 142  */
 143 unsigned int xstate_size;
 144 EXPORT_SYMBOL_GPL(xstate_size);
 145
 146 /* Enforce that 'MEMBER' is the last field of 'TYPE': */
 147 #define CHECK_MEMBER_AT_END_OF(TYPE, MEMBER) \
 148         BUILD_BUG_ON(sizeof(TYPE) != offsetofend(TYPE, MEMBER))
 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         /*
 158          * Subtract off the static size of the register state.
 159          * It potentially has a bunch of padding.
 160          */
 161         task_size -= sizeof(((struct task_struct *)0)->thread.fpu.state);
 162
 163         /*
 164          * Add back the dynamically-calculated register state
 165          * size.
 166          */
 167         task_size += xstate_size;
 168
 169         /*
 170          * We dynamically size 'struct fpu', so we require that
 171          * it be at the end of 'thread_struct' and that
 172          * 'thread_struct' be at the end of 'task_struct'.  If
 173          * you hit a compile error here, check the structure to
 174          * see if something got added to the end.
 175          */
 176         CHECK_MEMBER_AT_END_OF(struct fpu, state);
 177         CHECK_MEMBER_AT_END_OF(struct thread_struct, fpu);
 178         CHECK_MEMBER_AT_END_OF(struct task_struct, thread);
 179
 180         arch_task_struct_size = task_size;
 181 }
 182
 183 /*
 184  * Set up the xstate_size based on the legacy FPU context size.
 185  *
 186  * We set this up first, and later it will be overwritten by
 187  * fpu__init_system_xstate() if the CPU knows about xstates.
 188  */
 189 static void __init fpu__init_system_xstate_size_legacy(void)
 190 {
 191         static int on_boot_cpu = 1;
 192
 193         WARN_ON_FPU(!on_boot_cpu);
 194         on_boot_cpu = 0;
 195
 196         /*
 197          * Note that xstate_size might be overwriten later during
 198          * fpu__init_system_xstate().
 199          */
 200
 201         if (!cpu_has_fpu) {
 202                 /*
 203                  * Disable xsave as we do not support it if i387
 204                  * emulation is enabled.
 205                  */
 206                 setup_clear_cpu_cap(X86_FEATURE_XSAVE);
 207                 setup_clear_cpu_cap(X86_FEATURE_XSAVEOPT);
 208                 xstate_size = sizeof(struct swregs_state);
 209         } else {
 210                 if (cpu_has_fxsr)
 211                         xstate_size = sizeof(struct fxregs_state);
 212                 else
 213                         xstate_size = sizeof(struct fregs_state);
 214         }
 215         /*
 216          * Quirk: we don't yet handle the XSAVES* instructions
 217          * correctly, as we don't correctly convert between
 218          * standard and compacted format when interfacing
 219          * with user-space - so disable it for now.
 220          *
 221          * The difference is small: with recent CPUs the
 222          * compacted format is only marginally smaller than
 223          * the standard FPU state format.
 224          *
 225          * ( This is easy to backport while we are fixing
 226          *   XSAVES* support. )
 227          */
 228         setup_clear_cpu_cap(X86_FEATURE_XSAVES);
 229 }
 230
 231 /*
 232  * FPU context switching strategies:
 233  *
 234  * Against popular belief, we don't do lazy FPU saves, due to the
 235  * task migration complications it brings on SMP - we only do
 236  * lazy FPU restores.
 237  *
 238  * 'lazy' is the traditional strategy, which is based on setting
 239  * CR0::TS to 1 during context-switch (instead of doing a full
 240  * restore of the FPU state), which causes the first FPU instruction
 241  * after the context switch (whenever it is executed) to fault - at
 242  * which point we lazily restore the FPU state into FPU registers.
 243  *
 244  * Tasks are of course under no obligation to execute FPU instructions,
 245  * so it can easily happen that another context-switch occurs without
 246  * a single FPU instruction being executed. If we eventually switch
 247  * back to the original task (that still owns the FPU) then we have
 248  * not only saved the restores along the way, but we also have the
 249  * FPU ready to be used for the original task.
 250  *
 251  * 'eager' switching is used on modern CPUs, there we switch the FPU
 252  * state during every context switch, regardless of whether the task
 253  * has used FPU instructions in that time slice or not. This is done
 254  * because modern FPU context saving instructions are able to optimize
 255  * state saving and restoration in hardware: they can detect both
 256  * unused and untouched FPU state and optimize accordingly.
 257  *
 258  * [ Note that even in 'lazy' mode we might optimize context switches
 259  *   to use 'eager' restores, if we detect that a task is using the FPU
 260  *   frequently. See the fpu->counter logic in fpu/internal.h for that. ]
 261  */
 262 static enum { AUTO, ENABLE, DISABLE } eagerfpu = AUTO;
 263
 264 static int __init eager_fpu_setup(char *s)
 265 {
 266         if (!strcmp(s, "on"))
 267                 eagerfpu = ENABLE;
 268         else if (!strcmp(s, "off"))
 269                 eagerfpu = DISABLE;
 270         else if (!strcmp(s, "auto"))
 271                 eagerfpu = AUTO;
 272         return 1;
 273 }
 274 __setup("eagerfpu=", eager_fpu_setup);
 275
 276 /*
 277  * Pick the FPU context switching strategy:
 278  */
 279 static void __init fpu__init_system_ctx_switch(void)
 280 {
 281         static bool on_boot_cpu = 1;
 282
 283         WARN_ON_FPU(!on_boot_cpu);
 284         on_boot_cpu = 0;
 285
 286         WARN_ON_FPU(current->thread.fpu.fpstate_active);
 287         current_thread_info()->status = 0;
 288
 289         /* Auto enable eagerfpu for xsaveopt */
 290         if (cpu_has_xsaveopt && eagerfpu != DISABLE)
 291                 eagerfpu = ENABLE;
 292
 293         if (xfeatures_mask & XFEATURE_MASK_EAGER) {
 294                 if (eagerfpu == DISABLE) {
 295                         pr_err("x86/fpu: eagerfpu switching disabled, disabling the following xstate features: 0x%llx.\n",
 296                                xfeatures_mask & XFEATURE_MASK_EAGER);
 297                         xfeatures_mask &= ~XFEATURE_MASK_EAGER;
 298                 } else {
 299                         eagerfpu = ENABLE;
 300                 }
 301         }
 302
 303         if (eagerfpu == ENABLE)
 304                 setup_force_cpu_cap(X86_FEATURE_EAGER_FPU);
 305
 306         printk(KERN_INFO "x86/fpu: Using '%s' FPU context switches.\n", eagerfpu == ENABLE ? "eager" : "lazy");
 307 }
 308
 309 /*
 310  * Called on the boot CPU once per system bootup, to set up the initial
 311  * FPU state that is later cloned into all processes:
 312  */
 313 void __init fpu__init_system(struct cpuinfo_x86 *c)
 314 {
 315         fpu__init_system_early_generic(c);
 316
 317         /*
 318          * The FPU has to be operational for some of the
 319          * later FPU init activities:
 320          */
 321         fpu__init_cpu();
 322
 323         /*
 324          * But don't leave CR0::TS set yet, as some of the FPU setup
 325          * methods depend on being able to execute FPU instructions
 326          * that will fault on a set TS, such as the FXSAVE in
 327          * fpu__init_system_mxcsr().
 328          */
 329         clts();
 330
 331         fpu__init_system_generic();
 332         fpu__init_system_xstate_size_legacy();
 333         fpu__init_system_xstate();
 334         fpu__init_task_struct_size();
 335
 336         fpu__init_system_ctx_switch();
 337 }
 338
 339 /*
 340  * Boot parameter to turn off FPU support and fall back to math-emu:
 341  */
 342 static int __init no_387(char *s)
 343 {
 344         setup_clear_cpu_cap(X86_FEATURE_FPU);
 345         return 1;
 346 }
 347 __setup("no387", no_387);
 348
 349 /*
 350  * Disable all xstate CPU features:
 351  */
 352 static int __init x86_noxsave_setup(char *s)
 353 {
 354         if (strlen(s))
 355                 return 0;
 356
 357         fpu__xstate_clear_all_cpu_caps();
 358
 359         return 1;
 360 }
 361 __setup("noxsave", x86_noxsave_setup);
 362
 363 /*
 364  * Disable the XSAVEOPT instruction specifically:
 365  */
 366 static int __init x86_noxsaveopt_setup(char *s)
 367 {
 368         setup_clear_cpu_cap(X86_FEATURE_XSAVEOPT);
 369
 370         return 1;
 371 }
 372 __setup("noxsaveopt", x86_noxsaveopt_setup);
 373
 374 /*
 375  * Disable the XSAVES instruction:
 376  */
 377 static int __init x86_noxsaves_setup(char *s)
 378 {
 379         setup_clear_cpu_cap(X86_FEATURE_XSAVES);
 380
 381         return 1;
 382 }
 383 __setup("noxsaves", x86_noxsaves_setup);
 384
 385 /*
 386  * Disable FX save/restore and SSE support:
 387  */
 388 static int __init x86_nofxsr_setup(char *s)
 389 {
 390         setup_clear_cpu_cap(X86_FEATURE_FXSR);
 391         setup_clear_cpu_cap(X86_FEATURE_FXSR_OPT);
 392         setup_clear_cpu_cap(X86_FEATURE_XMM);
 393
 394         return 1;
 395 }
 396 __setup("nofxsr", x86_nofxsr_setup);