2 * xsave/xrstor support.
4 * Author: Suresh Siddha <suresh.b.siddha@intel.com>
6 #include <linux/compat.h>
8 #include <asm/fpu/api.h>
9 #include <asm/fpu/internal.h>
10 #include <asm/sigframe.h>
11 #include <asm/tlbflush.h>
15 * Mask of xstate features supported by the CPU and the kernel:
20 * Represents init state for the supported extended state.
22 struct xsave_struct init_xstate_ctx
;
24 static struct _fpx_sw_bytes fx_sw_reserved
, fx_sw_reserved_ia32
;
25 static unsigned int xstate_offsets
[XFEATURES_NR_MAX
], xstate_sizes
[XFEATURES_NR_MAX
];
26 static unsigned int xstate_comp_offsets
[sizeof(xfeatures_mask
)*8];
28 /* The number of supported xfeatures in xfeatures_mask: */
29 static unsigned int xfeatures_nr
;
32 * When executing XSAVEOPT (optimized XSAVE), if a processor implementation
33 * detects that an FPU state component is still (or is again) in its
34 * initialized state, it may clear the corresponding bit in the header.xfeatures
35 * field, and can skip the writeout of registers to the corresponding memory layout.
37 * This means that when the bit is zero, the state component might still contain
38 * some previous - non-initialized register state.
40 * Before writing xstate information to user-space we sanitize those components,
41 * to always ensure that the memory layout of a feature will be in the init state
42 * if the corresponding header bit is zero. This is to ensure that user-space doesn't
43 * see some stale state in the memory layout during signal handling, debugging etc.
45 void __sanitize_i387_state(struct task_struct
*tsk
)
47 struct i387_fxsave_struct
*fx
= &tsk
->thread
.fpu
.state
.fxsave
;
54 xfeatures
= tsk
->thread
.fpu
.state
.xsave
.header
.xfeatures
;
57 * None of the feature bits are in init state. So nothing else
58 * to do for us, as the memory layout is up to date.
60 if ((xfeatures
& xfeatures_mask
) == xfeatures_mask
)
66 if (!(xfeatures
& XSTATE_FP
)) {
73 memset(&fx
->st_space
[0], 0, 128);
77 * SSE is in init state
79 if (!(xfeatures
& XSTATE_SSE
))
80 memset(&fx
->xmm_space
[0], 0, 256);
83 * First two features are FPU and SSE, which above we handled
84 * in a special way already:
87 xfeatures
= (xfeatures_mask
& ~xfeatures
) >> 2;
90 * Update all the remaining memory layouts according to their
91 * standard xstate layout, if their header bit is in the init
95 if (xfeatures
& 0x1) {
96 int offset
= xstate_offsets
[feature_bit
];
97 int size
= xstate_sizes
[feature_bit
];
99 memcpy((void *)fx
+ offset
,
100 (void *)&init_xstate_ctx
+ offset
,
110 * Check for the presence of extended state information in the
111 * user fpstate pointer in the sigcontext.
113 static inline int check_for_xstate(struct i387_fxsave_struct __user
*buf
,
114 void __user
*fpstate
,
115 struct _fpx_sw_bytes
*fx_sw
)
117 int min_xstate_size
= sizeof(struct i387_fxsave_struct
) +
118 sizeof(struct xstate_header
);
121 if (__copy_from_user(fx_sw
, &buf
->sw_reserved
[0], sizeof(*fx_sw
)))
124 /* Check for the first magic field and other error scenarios. */
125 if (fx_sw
->magic1
!= FP_XSTATE_MAGIC1
||
126 fx_sw
->xstate_size
< min_xstate_size
||
127 fx_sw
->xstate_size
> xstate_size
||
128 fx_sw
->xstate_size
> fx_sw
->extended_size
)
132 * Check for the presence of second magic word at the end of memory
133 * layout. This detects the case where the user just copied the legacy
134 * fpstate layout with out copying the extended state information
135 * in the memory layout.
137 if (__get_user(magic2
, (__u32 __user
*)(fpstate
+ fx_sw
->xstate_size
))
138 || magic2
!= FP_XSTATE_MAGIC2
)
145 * Signal frame handlers.
147 static inline int save_fsave_header(struct task_struct
*tsk
, void __user
*buf
)
150 struct xsave_struct
*xsave
= &tsk
->thread
.fpu
.state
.xsave
;
151 struct user_i387_ia32_struct env
;
152 struct _fpstate_ia32 __user
*fp
= buf
;
154 convert_from_fxsr(&env
, tsk
);
156 if (__copy_to_user(buf
, &env
, sizeof(env
)) ||
157 __put_user(xsave
->i387
.swd
, &fp
->status
) ||
158 __put_user(X86_FXSR_MAGIC
, &fp
->magic
))
161 struct i387_fsave_struct __user
*fp
= buf
;
163 if (__get_user(swd
, &fp
->swd
) || __put_user(swd
, &fp
->status
))
170 static inline int save_xstate_epilog(void __user
*buf
, int ia32_frame
)
172 struct xsave_struct __user
*x
= buf
;
173 struct _fpx_sw_bytes
*sw_bytes
;
177 /* Setup the bytes not touched by the [f]xsave and reserved for SW. */
178 sw_bytes
= ia32_frame
? &fx_sw_reserved_ia32
: &fx_sw_reserved
;
179 err
= __copy_to_user(&x
->i387
.sw_reserved
, sw_bytes
, sizeof(*sw_bytes
));
184 err
|= __put_user(FP_XSTATE_MAGIC2
, (__u32
*)(buf
+ xstate_size
));
187 * Read the xfeatures which we copied (directly from the cpu or
188 * from the state in task struct) to the user buffers.
190 err
|= __get_user(xfeatures
, (__u32
*)&x
->header
.xfeatures
);
193 * For legacy compatible, we always set FP/SSE bits in the bit
194 * vector while saving the state to the user context. This will
195 * enable us capturing any changes(during sigreturn) to
196 * the FP/SSE bits by the legacy applications which don't touch
197 * xfeatures in the xsave header.
199 * xsave aware apps can change the xfeatures in the xsave
200 * header as well as change any contents in the memory layout.
201 * xrestore as part of sigreturn will capture all the changes.
203 xfeatures
|= XSTATE_FPSSE
;
205 err
|= __put_user(xfeatures
, (__u32
*)&x
->header
.xfeatures
);
210 static inline int save_user_xstate(struct xsave_struct __user
*buf
)
215 err
= xsave_user(buf
);
217 err
= fxsave_user((struct i387_fxsave_struct __user
*) buf
);
219 err
= fsave_user((struct i387_fsave_struct __user
*) buf
);
221 if (unlikely(err
) && __clear_user(buf
, xstate_size
))
227 * Save the fpu, extended register state to the user signal frame.
229 * 'buf_fx' is the 64-byte aligned pointer at which the [f|fx|x]save
231 * 'buf' points to the 'buf_fx' or to the fsave header followed by 'buf_fx'.
233 * buf == buf_fx for 64-bit frames and 32-bit fsave frame.
234 * buf != buf_fx for 32-bit frames with fxstate.
236 * If the fpu, extended register state is live, save the state directly
237 * to the user frame pointed by the aligned pointer 'buf_fx'. Otherwise,
238 * copy the thread's fpu state to the user frame starting at 'buf_fx'.
240 * If this is a 32-bit frame with fxstate, put a fsave header before
241 * the aligned state at 'buf_fx'.
243 * For [f]xsave state, update the SW reserved fields in the [f]xsave frame
244 * indicating the absence/presence of the extended state to the user.
246 int save_xstate_sig(void __user
*buf
, void __user
*buf_fx
, int size
)
248 struct xsave_struct
*xsave
= ¤t
->thread
.fpu
.state
.xsave
;
249 struct task_struct
*tsk
= current
;
250 int ia32_fxstate
= (buf
!= buf_fx
);
252 ia32_fxstate
&= (config_enabled(CONFIG_X86_32
) ||
253 config_enabled(CONFIG_IA32_EMULATION
));
255 if (!access_ok(VERIFY_WRITE
, buf
, size
))
258 if (!static_cpu_has(X86_FEATURE_FPU
))
259 return fpregs_soft_get(current
, NULL
, 0,
260 sizeof(struct user_i387_ia32_struct
), NULL
,
261 (struct _fpstate_ia32 __user
*) buf
) ? -1 : 1;
263 if (user_has_fpu()) {
264 /* Save the live register state to the user directly. */
265 if (save_user_xstate(buf_fx
))
267 /* Update the thread's fxstate to save the fsave header. */
269 fpu_fxsave(&tsk
->thread
.fpu
);
271 sanitize_i387_state(tsk
);
272 if (__copy_to_user(buf_fx
, xsave
, xstate_size
))
276 /* Save the fsave header for the 32-bit frames. */
277 if ((ia32_fxstate
|| !use_fxsr()) && save_fsave_header(tsk
, buf
))
280 if (use_fxsr() && save_xstate_epilog(buf_fx
, ia32_fxstate
))
287 sanitize_restored_xstate(struct task_struct
*tsk
,
288 struct user_i387_ia32_struct
*ia32_env
,
289 u64 xfeatures
, int fx_only
)
291 struct xsave_struct
*xsave
= &tsk
->thread
.fpu
.state
.xsave
;
292 struct xstate_header
*header
= &xsave
->header
;
295 /* These bits must be zero. */
296 memset(header
->reserved
, 0, 48);
299 * Init the state that is not present in the memory
300 * layout and not enabled by the OS.
303 header
->xfeatures
= XSTATE_FPSSE
;
305 header
->xfeatures
&= (xfeatures_mask
& xfeatures
);
310 * mscsr reserved bits must be masked to zero for security
313 xsave
->i387
.mxcsr
&= mxcsr_feature_mask
;
315 convert_to_fxsr(tsk
, ia32_env
);
320 * Restore the extended state if present. Otherwise, restore the FP/SSE state.
322 static inline int restore_user_xstate(void __user
*buf
, u64 xbv
, int fx_only
)
325 if ((unsigned long)buf
% 64 || fx_only
) {
326 u64 init_bv
= xfeatures_mask
& ~XSTATE_FPSSE
;
327 xrstor_state(&init_xstate_ctx
, init_bv
);
328 return fxrstor_user(buf
);
330 u64 init_bv
= xfeatures_mask
& ~xbv
;
331 if (unlikely(init_bv
))
332 xrstor_state(&init_xstate_ctx
, init_bv
);
333 return xrestore_user(buf
, xbv
);
335 } else if (use_fxsr()) {
336 return fxrstor_user(buf
);
338 return frstor_user(buf
);
341 int __restore_xstate_sig(void __user
*buf
, void __user
*buf_fx
, int size
)
343 int ia32_fxstate
= (buf
!= buf_fx
);
344 struct task_struct
*tsk
= current
;
345 struct fpu
*fpu
= &tsk
->thread
.fpu
;
346 int state_size
= xstate_size
;
350 ia32_fxstate
&= (config_enabled(CONFIG_X86_32
) ||
351 config_enabled(CONFIG_IA32_EMULATION
));
354 fpu_reset_state(fpu
);
358 if (!access_ok(VERIFY_READ
, buf
, size
))
361 fpu__activate_curr(fpu
);
363 if (!static_cpu_has(X86_FEATURE_FPU
))
364 return fpregs_soft_set(current
, NULL
,
365 0, sizeof(struct user_i387_ia32_struct
),
369 struct _fpx_sw_bytes fx_sw_user
;
370 if (unlikely(check_for_xstate(buf_fx
, buf_fx
, &fx_sw_user
))) {
372 * Couldn't find the extended state information in the
373 * memory layout. Restore just the FP/SSE and init all
374 * the other extended state.
376 state_size
= sizeof(struct i387_fxsave_struct
);
379 state_size
= fx_sw_user
.xstate_size
;
380 xfeatures
= fx_sw_user
.xfeatures
;
386 * For 32-bit frames with fxstate, copy the user state to the
387 * thread's fpu state, reconstruct fxstate from the fsave
388 * header. Sanitize the copied state etc.
390 struct fpu
*fpu
= &tsk
->thread
.fpu
;
391 struct user_i387_ia32_struct env
;
395 * Drop the current fpu which clears fpu->fpstate_active. This ensures
396 * that any context-switch during the copy of the new state,
397 * avoids the intermediate state from getting restored/saved.
398 * Thus avoiding the new restored state from getting corrupted.
399 * We will be ready to restore/save the state only after
400 * fpu->fpstate_active is again set.
404 if (__copy_from_user(&fpu
->state
.xsave
, buf_fx
, state_size
) ||
405 __copy_from_user(&env
, buf
, sizeof(env
))) {
409 sanitize_restored_xstate(tsk
, &env
, xfeatures
, fx_only
);
412 fpu
->fpstate_active
= 1;
413 if (use_eager_fpu()) {
422 * For 64-bit frames and 32-bit fsave frames, restore the user
423 * state to the registers directly (with exceptions handled).
426 if (restore_user_xstate(buf_fx
, xfeatures
, fx_only
)) {
427 fpu_reset_state(fpu
);
436 * Prepare the SW reserved portion of the fxsave memory layout, indicating
437 * the presence of the extended state information in the memory layout
438 * pointed by the fpstate pointer in the sigcontext.
439 * This will be saved when ever the FP and extended state context is
440 * saved on the user stack during the signal handler delivery to the user.
442 static void prepare_fx_sw_frame(void)
444 int fsave_header_size
= sizeof(struct i387_fsave_struct
);
445 int size
= xstate_size
+ FP_XSTATE_MAGIC2_SIZE
;
447 if (config_enabled(CONFIG_X86_32
))
448 size
+= fsave_header_size
;
450 fx_sw_reserved
.magic1
= FP_XSTATE_MAGIC1
;
451 fx_sw_reserved
.extended_size
= size
;
452 fx_sw_reserved
.xfeatures
= xfeatures_mask
;
453 fx_sw_reserved
.xstate_size
= xstate_size
;
455 if (config_enabled(CONFIG_IA32_EMULATION
)) {
456 fx_sw_reserved_ia32
= fx_sw_reserved
;
457 fx_sw_reserved_ia32
.extended_size
+= fsave_header_size
;
462 * Enable the extended processor state save/restore feature.
463 * Called once per CPU onlining.
465 void fpu__init_cpu_xstate(void)
467 if (!cpu_has_xsave
|| !xfeatures_mask
)
470 cr4_set_bits(X86_CR4_OSXSAVE
);
471 xsetbv(XCR_XFEATURE_ENABLED_MASK
, xfeatures_mask
);
475 * Record the offsets and sizes of different state managed by the xsave
478 static void __init
setup_xstate_features(void)
480 int eax
, ebx
, ecx
, edx
, leaf
= 0x2;
482 xfeatures_nr
= fls64(xfeatures_mask
);
485 cpuid_count(XSTATE_CPUID
, leaf
, &eax
, &ebx
, &ecx
, &edx
);
490 xstate_offsets
[leaf
] = ebx
;
491 xstate_sizes
[leaf
] = eax
;
497 static void print_xstate_feature(u64 xstate_mask
, const char *desc
)
499 if (xfeatures_mask
& xstate_mask
) {
500 int xstate_feature
= fls64(xstate_mask
)-1;
502 pr_info("x86/fpu: Supporting XSAVE feature %2d: '%s'\n", xstate_feature
, desc
);
507 * Print out all the supported xstate features:
509 static void print_xstate_features(void)
511 print_xstate_feature(XSTATE_FP
, "x87 floating point registers");
512 print_xstate_feature(XSTATE_SSE
, "SSE registers");
513 print_xstate_feature(XSTATE_YMM
, "AVX registers");
514 print_xstate_feature(XSTATE_BNDREGS
, "MPX bounds registers");
515 print_xstate_feature(XSTATE_BNDCSR
, "MPX CSR");
516 print_xstate_feature(XSTATE_OPMASK
, "AVX-512 opmask");
517 print_xstate_feature(XSTATE_ZMM_Hi256
, "AVX-512 Hi256");
518 print_xstate_feature(XSTATE_Hi16_ZMM
, "AVX-512 ZMM_Hi256");
522 * This function sets up offsets and sizes of all extended states in
523 * xsave area. This supports both standard format and compacted format
524 * of the xsave aread.
529 void setup_xstate_comp(void)
531 unsigned int xstate_comp_sizes
[sizeof(xfeatures_mask
)*8];
535 * The FP xstates and SSE xstates are legacy states. They are always
536 * in the fixed offsets in the xsave area in either compacted form
539 xstate_comp_offsets
[0] = 0;
540 xstate_comp_offsets
[1] = offsetof(struct i387_fxsave_struct
, xmm_space
);
542 if (!cpu_has_xsaves
) {
543 for (i
= 2; i
< xfeatures_nr
; i
++) {
544 if (test_bit(i
, (unsigned long *)&xfeatures_mask
)) {
545 xstate_comp_offsets
[i
] = xstate_offsets
[i
];
546 xstate_comp_sizes
[i
] = xstate_sizes
[i
];
552 xstate_comp_offsets
[2] = FXSAVE_SIZE
+ XSAVE_HDR_SIZE
;
554 for (i
= 2; i
< xfeatures_nr
; i
++) {
555 if (test_bit(i
, (unsigned long *)&xfeatures_mask
))
556 xstate_comp_sizes
[i
] = xstate_sizes
[i
];
558 xstate_comp_sizes
[i
] = 0;
561 xstate_comp_offsets
[i
] = xstate_comp_offsets
[i
-1]
562 + xstate_comp_sizes
[i
-1];
568 * setup the xstate image representing the init state
570 static void setup_init_fpu_buf(void)
572 static int on_boot_cpu
= 1;
581 setup_xstate_features();
582 print_xstate_features();
584 if (cpu_has_xsaves
) {
585 init_xstate_ctx
.header
.xcomp_bv
= (u64
)1 << 63 | xfeatures_mask
;
586 init_xstate_ctx
.header
.xfeatures
= xfeatures_mask
;
590 * Init all the features state with header_bv being 0x0
592 xrstor_state_booting(&init_xstate_ctx
, -1);
595 * Dump the init state again. This is to identify the init state
596 * of any feature which is not represented by all zero's.
598 xsave_state_booting(&init_xstate_ctx
);
602 * Calculate total size of enabled xstates in XCR0/xfeatures_mask.
604 static void __init
init_xstate_size(void)
606 unsigned int eax
, ebx
, ecx
, edx
;
609 if (!cpu_has_xsaves
) {
610 cpuid_count(XSTATE_CPUID
, 0, &eax
, &ebx
, &ecx
, &edx
);
615 xstate_size
= FXSAVE_SIZE
+ XSAVE_HDR_SIZE
;
616 for (i
= 2; i
< 64; i
++) {
617 if (test_bit(i
, (unsigned long *)&xfeatures_mask
)) {
618 cpuid_count(XSTATE_CPUID
, i
, &eax
, &ebx
, &ecx
, &edx
);
625 * Enable and initialize the xsave feature.
626 * Called once per system bootup.
628 * ( Not marked __init because of false positive section warnings. )
630 void fpu__init_system_xstate(void)
632 unsigned int eax
, ebx
, ecx
, edx
;
633 static bool on_boot_cpu
= 1;
639 if (!cpu_has_xsave
) {
640 pr_info("x86/fpu: Legacy x87 FPU detected.\n");
644 if (boot_cpu_data
.cpuid_level
< XSTATE_CPUID
) {
645 WARN(1, "x86/fpu: XSTATE_CPUID missing!\n");
649 cpuid_count(XSTATE_CPUID
, 0, &eax
, &ebx
, &ecx
, &edx
);
650 xfeatures_mask
= eax
+ ((u64
)edx
<< 32);
652 if ((xfeatures_mask
& XSTATE_FPSSE
) != XSTATE_FPSSE
) {
653 pr_err("x86/fpu: FP/SSE not present amongst the CPU's xstate features: 0x%llx.\n", xfeatures_mask
);
658 * Support only the state known to OS.
660 xfeatures_mask
= xfeatures_mask
& XCNTXT_MASK
;
662 /* Enable xstate instructions to be able to continue with initialization: */
663 fpu__init_cpu_xstate();
666 * Recompute the context size for enabled features
670 update_regset_xstate_info(xstate_size
, xfeatures_mask
);
671 prepare_fx_sw_frame();
672 setup_init_fpu_buf();
674 pr_info("x86/fpu: Enabled xstate features 0x%llx, context size is 0x%x bytes, using '%s' format.\n",
677 cpu_has_xsaves
? "compacted" : "standard");
681 * Restore minimal FPU state after suspend:
683 void fpu__resume_cpu(void)
686 * Restore XCR0 on xsave capable CPUs:
689 xsetbv(XCR_XFEATURE_ENABLED_MASK
, xfeatures_mask
);
693 * Given the xsave area and a state inside, this function returns the
694 * address of the state.
696 * This is the API that is called to get xstate address in either
697 * standard format or compacted format of xsave area.
700 * xsave: base address of the xsave area;
701 * xstate: state which is defined in xsave.h (e.g. XSTATE_FP, XSTATE_SSE,
704 * address of the state in the xsave area.
706 void *get_xsave_addr(struct xsave_struct
*xsave
, int xstate
)
708 int feature
= fls64(xstate
) - 1;
709 if (!test_bit(feature
, (unsigned long *)&xfeatures_mask
))
712 return (void *)xsave
+ xstate_comp_offsets
[feature
];
714 EXPORT_SYMBOL_GPL(get_xsave_addr
);