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/fpu/signal.h>
11 #include <asm/sigframe.h>
12 #include <asm/tlbflush.h>
14 static const char *xfeature_names
[] =
16 "x87 floating point registers" ,
19 "MPX bounds registers" ,
24 "unknown xstate feature" ,
28 * Mask of xstate features supported by the CPU and the kernel:
30 u64 xfeatures_mask __read_mostly
;
33 * Represents init state for the supported extended state.
35 struct xsave_struct init_xstate_ctx
;
37 static struct _fpx_sw_bytes fx_sw_reserved
, fx_sw_reserved_ia32
;
38 static unsigned int xstate_offsets
[XFEATURES_NR_MAX
], xstate_sizes
[XFEATURES_NR_MAX
];
39 static unsigned int xstate_comp_offsets
[sizeof(xfeatures_mask
)*8];
41 /* The number of supported xfeatures in xfeatures_mask: */
42 static unsigned int xfeatures_nr
;
45 * Return whether the system supports a given xfeature.
47 * Also return the name of the (most advanced) feature that the caller requested:
49 int cpu_has_xfeatures(u64 xfeatures_needed
, const char **feature_name
)
51 u64 xfeatures_missing
= xfeatures_needed
& ~xfeatures_mask
;
53 if (unlikely(feature_name
)) {
54 long xfeature_idx
, max_idx
;
57 * So we use FLS here to be able to print the most advanced
58 * feature that was requested but is missing. So if a driver
59 * asks about "XSTATE_SSE | XSTATE_YMM" we'll print the
60 * missing AVX feature - this is the most informative message
63 if (xfeatures_missing
)
64 xfeatures_print
= xfeatures_missing
;
66 xfeatures_print
= xfeatures_needed
;
68 xfeature_idx
= fls64(xfeatures_print
)-1;
69 max_idx
= ARRAY_SIZE(xfeature_names
)-1;
70 xfeature_idx
= min(xfeature_idx
, max_idx
);
72 *feature_name
= xfeature_names
[xfeature_idx
];
75 if (xfeatures_missing
)
80 EXPORT_SYMBOL_GPL(cpu_has_xfeatures
);
83 * When executing XSAVEOPT (optimized XSAVE), if a processor implementation
84 * detects that an FPU state component is still (or is again) in its
85 * initialized state, it may clear the corresponding bit in the header.xfeatures
86 * field, and can skip the writeout of registers to the corresponding memory layout.
88 * This means that when the bit is zero, the state component might still contain
89 * some previous - non-initialized register state.
91 * Before writing xstate information to user-space we sanitize those components,
92 * to always ensure that the memory layout of a feature will be in the init state
93 * if the corresponding header bit is zero. This is to ensure that user-space doesn't
94 * see some stale state in the memory layout during signal handling, debugging etc.
96 void fpstate_sanitize_xstate(struct fpu
*fpu
)
98 struct i387_fxsave_struct
*fx
= &fpu
->state
.fxsave
;
105 xfeatures
= fpu
->state
.xsave
.header
.xfeatures
;
108 * None of the feature bits are in init state. So nothing else
109 * to do for us, as the memory layout is up to date.
111 if ((xfeatures
& xfeatures_mask
) == xfeatures_mask
)
115 * FP is in init state
117 if (!(xfeatures
& XSTATE_FP
)) {
124 memset(&fx
->st_space
[0], 0, 128);
128 * SSE is in init state
130 if (!(xfeatures
& XSTATE_SSE
))
131 memset(&fx
->xmm_space
[0], 0, 256);
134 * First two features are FPU and SSE, which above we handled
135 * in a special way already:
138 xfeatures
= (xfeatures_mask
& ~xfeatures
) >> 2;
141 * Update all the remaining memory layouts according to their
142 * standard xstate layout, if their header bit is in the init
146 if (xfeatures
& 0x1) {
147 int offset
= xstate_offsets
[feature_bit
];
148 int size
= xstate_sizes
[feature_bit
];
150 memcpy((void *)fx
+ offset
,
151 (void *)&init_xstate_ctx
+ offset
,
161 * Check for the presence of extended state information in the
162 * user fpstate pointer in the sigcontext.
164 static inline int check_for_xstate(struct i387_fxsave_struct __user
*buf
,
165 void __user
*fpstate
,
166 struct _fpx_sw_bytes
*fx_sw
)
168 int min_xstate_size
= sizeof(struct i387_fxsave_struct
) +
169 sizeof(struct xstate_header
);
172 if (__copy_from_user(fx_sw
, &buf
->sw_reserved
[0], sizeof(*fx_sw
)))
175 /* Check for the first magic field and other error scenarios. */
176 if (fx_sw
->magic1
!= FP_XSTATE_MAGIC1
||
177 fx_sw
->xstate_size
< min_xstate_size
||
178 fx_sw
->xstate_size
> xstate_size
||
179 fx_sw
->xstate_size
> fx_sw
->extended_size
)
183 * Check for the presence of second magic word at the end of memory
184 * layout. This detects the case where the user just copied the legacy
185 * fpstate layout with out copying the extended state information
186 * in the memory layout.
188 if (__get_user(magic2
, (__u32 __user
*)(fpstate
+ fx_sw
->xstate_size
))
189 || magic2
!= FP_XSTATE_MAGIC2
)
196 * Signal frame handlers.
198 static inline int save_fsave_header(struct task_struct
*tsk
, void __user
*buf
)
201 struct xsave_struct
*xsave
= &tsk
->thread
.fpu
.state
.xsave
;
202 struct user_i387_ia32_struct env
;
203 struct _fpstate_ia32 __user
*fp
= buf
;
205 convert_from_fxsr(&env
, tsk
);
207 if (__copy_to_user(buf
, &env
, sizeof(env
)) ||
208 __put_user(xsave
->i387
.swd
, &fp
->status
) ||
209 __put_user(X86_FXSR_MAGIC
, &fp
->magic
))
212 struct i387_fsave_struct __user
*fp
= buf
;
214 if (__get_user(swd
, &fp
->swd
) || __put_user(swd
, &fp
->status
))
221 static inline int save_xstate_epilog(void __user
*buf
, int ia32_frame
)
223 struct xsave_struct __user
*x
= buf
;
224 struct _fpx_sw_bytes
*sw_bytes
;
228 /* Setup the bytes not touched by the [f]xsave and reserved for SW. */
229 sw_bytes
= ia32_frame
? &fx_sw_reserved_ia32
: &fx_sw_reserved
;
230 err
= __copy_to_user(&x
->i387
.sw_reserved
, sw_bytes
, sizeof(*sw_bytes
));
235 err
|= __put_user(FP_XSTATE_MAGIC2
, (__u32
*)(buf
+ xstate_size
));
238 * Read the xfeatures which we copied (directly from the cpu or
239 * from the state in task struct) to the user buffers.
241 err
|= __get_user(xfeatures
, (__u32
*)&x
->header
.xfeatures
);
244 * For legacy compatible, we always set FP/SSE bits in the bit
245 * vector while saving the state to the user context. This will
246 * enable us capturing any changes(during sigreturn) to
247 * the FP/SSE bits by the legacy applications which don't touch
248 * xfeatures in the xsave header.
250 * xsave aware apps can change the xfeatures in the xsave
251 * header as well as change any contents in the memory layout.
252 * xrestore as part of sigreturn will capture all the changes.
254 xfeatures
|= XSTATE_FPSSE
;
256 err
|= __put_user(xfeatures
, (__u32
*)&x
->header
.xfeatures
);
261 static inline int copy_fpregs_to_sigframe(struct xsave_struct __user
*buf
)
266 err
= xsave_user(buf
);
268 err
= fxsave_user((struct i387_fxsave_struct __user
*) buf
);
270 err
= fsave_user((struct i387_fsave_struct __user
*) buf
);
272 if (unlikely(err
) && __clear_user(buf
, xstate_size
))
278 * Save the fpu, extended register state to the user signal frame.
280 * 'buf_fx' is the 64-byte aligned pointer at which the [f|fx|x]save
282 * 'buf' points to the 'buf_fx' or to the fsave header followed by 'buf_fx'.
284 * buf == buf_fx for 64-bit frames and 32-bit fsave frame.
285 * buf != buf_fx for 32-bit frames with fxstate.
287 * If the fpu, extended register state is live, save the state directly
288 * to the user frame pointed by the aligned pointer 'buf_fx'. Otherwise,
289 * copy the thread's fpu state to the user frame starting at 'buf_fx'.
291 * If this is a 32-bit frame with fxstate, put a fsave header before
292 * the aligned state at 'buf_fx'.
294 * For [f]xsave state, update the SW reserved fields in the [f]xsave frame
295 * indicating the absence/presence of the extended state to the user.
297 int copy_fpstate_to_sigframe(void __user
*buf
, void __user
*buf_fx
, int size
)
299 struct xsave_struct
*xsave
= ¤t
->thread
.fpu
.state
.xsave
;
300 struct task_struct
*tsk
= current
;
301 int ia32_fxstate
= (buf
!= buf_fx
);
303 ia32_fxstate
&= (config_enabled(CONFIG_X86_32
) ||
304 config_enabled(CONFIG_IA32_EMULATION
));
306 if (!access_ok(VERIFY_WRITE
, buf
, size
))
309 if (!static_cpu_has(X86_FEATURE_FPU
))
310 return fpregs_soft_get(current
, NULL
, 0,
311 sizeof(struct user_i387_ia32_struct
), NULL
,
312 (struct _fpstate_ia32 __user
*) buf
) ? -1 : 1;
314 if (fpregs_active()) {
315 /* Save the live register state to the user directly. */
316 if (copy_fpregs_to_sigframe(buf_fx
))
318 /* Update the thread's fxstate to save the fsave header. */
320 fpu_fxsave(&tsk
->thread
.fpu
);
322 fpstate_sanitize_xstate(&tsk
->thread
.fpu
);
323 if (__copy_to_user(buf_fx
, xsave
, xstate_size
))
327 /* Save the fsave header for the 32-bit frames. */
328 if ((ia32_fxstate
|| !use_fxsr()) && save_fsave_header(tsk
, buf
))
331 if (use_fxsr() && save_xstate_epilog(buf_fx
, ia32_fxstate
))
338 sanitize_restored_xstate(struct task_struct
*tsk
,
339 struct user_i387_ia32_struct
*ia32_env
,
340 u64 xfeatures
, int fx_only
)
342 struct xsave_struct
*xsave
= &tsk
->thread
.fpu
.state
.xsave
;
343 struct xstate_header
*header
= &xsave
->header
;
346 /* These bits must be zero. */
347 memset(header
->reserved
, 0, 48);
350 * Init the state that is not present in the memory
351 * layout and not enabled by the OS.
354 header
->xfeatures
= XSTATE_FPSSE
;
356 header
->xfeatures
&= (xfeatures_mask
& xfeatures
);
361 * mscsr reserved bits must be masked to zero for security
364 xsave
->i387
.mxcsr
&= mxcsr_feature_mask
;
366 convert_to_fxsr(tsk
, ia32_env
);
371 * Restore the extended state if present. Otherwise, restore the FP/SSE state.
373 static inline int restore_user_xstate(void __user
*buf
, u64 xbv
, int fx_only
)
376 if ((unsigned long)buf
% 64 || fx_only
) {
377 u64 init_bv
= xfeatures_mask
& ~XSTATE_FPSSE
;
378 xrstor_state(&init_xstate_ctx
, init_bv
);
379 return fxrstor_user(buf
);
381 u64 init_bv
= xfeatures_mask
& ~xbv
;
382 if (unlikely(init_bv
))
383 xrstor_state(&init_xstate_ctx
, init_bv
);
384 return xrestore_user(buf
, xbv
);
386 } else if (use_fxsr()) {
387 return fxrstor_user(buf
);
389 return frstor_user(buf
);
392 static int __fpu__restore_sig(void __user
*buf
, void __user
*buf_fx
, int size
)
394 int ia32_fxstate
= (buf
!= buf_fx
);
395 struct task_struct
*tsk
= current
;
396 struct fpu
*fpu
= &tsk
->thread
.fpu
;
397 int state_size
= xstate_size
;
401 ia32_fxstate
&= (config_enabled(CONFIG_X86_32
) ||
402 config_enabled(CONFIG_IA32_EMULATION
));
409 if (!access_ok(VERIFY_READ
, buf
, size
))
412 fpu__activate_curr(fpu
);
414 if (!static_cpu_has(X86_FEATURE_FPU
))
415 return fpregs_soft_set(current
, NULL
,
416 0, sizeof(struct user_i387_ia32_struct
),
420 struct _fpx_sw_bytes fx_sw_user
;
421 if (unlikely(check_for_xstate(buf_fx
, buf_fx
, &fx_sw_user
))) {
423 * Couldn't find the extended state information in the
424 * memory layout. Restore just the FP/SSE and init all
425 * the other extended state.
427 state_size
= sizeof(struct i387_fxsave_struct
);
430 state_size
= fx_sw_user
.xstate_size
;
431 xfeatures
= fx_sw_user
.xfeatures
;
437 * For 32-bit frames with fxstate, copy the user state to the
438 * thread's fpu state, reconstruct fxstate from the fsave
439 * header. Sanitize the copied state etc.
441 struct fpu
*fpu
= &tsk
->thread
.fpu
;
442 struct user_i387_ia32_struct env
;
446 * Drop the current fpu which clears fpu->fpstate_active. This ensures
447 * that any context-switch during the copy of the new state,
448 * avoids the intermediate state from getting restored/saved.
449 * Thus avoiding the new restored state from getting corrupted.
450 * We will be ready to restore/save the state only after
451 * fpu->fpstate_active is again set.
455 if (__copy_from_user(&fpu
->state
.xsave
, buf_fx
, state_size
) ||
456 __copy_from_user(&env
, buf
, sizeof(env
))) {
460 sanitize_restored_xstate(tsk
, &env
, xfeatures
, fx_only
);
463 fpu
->fpstate_active
= 1;
464 if (use_eager_fpu()) {
473 * For 64-bit frames and 32-bit fsave frames, restore the user
474 * state to the registers directly (with exceptions handled).
477 if (restore_user_xstate(buf_fx
, xfeatures
, fx_only
)) {
486 static inline int xstate_sigframe_size(void)
488 return use_xsave() ? xstate_size
+ FP_XSTATE_MAGIC2_SIZE
: xstate_size
;
492 * Restore FPU state from a sigframe:
494 int fpu__restore_sig(void __user
*buf
, int ia32_frame
)
496 void __user
*buf_fx
= buf
;
497 int size
= xstate_sigframe_size();
499 if (ia32_frame
&& use_fxsr()) {
500 buf_fx
= buf
+ sizeof(struct i387_fsave_struct
);
501 size
+= sizeof(struct i387_fsave_struct
);
504 return __fpu__restore_sig(buf
, buf_fx
, size
);
508 fpu__alloc_mathframe(unsigned long sp
, int ia32_frame
,
509 unsigned long *buf_fx
, unsigned long *size
)
511 unsigned long frame_size
= xstate_sigframe_size();
513 *buf_fx
= sp
= round_down(sp
- frame_size
, 64);
514 if (ia32_frame
&& use_fxsr()) {
515 frame_size
+= sizeof(struct i387_fsave_struct
);
516 sp
-= sizeof(struct i387_fsave_struct
);
524 * Prepare the SW reserved portion of the fxsave memory layout, indicating
525 * the presence of the extended state information in the memory layout
526 * pointed by the fpstate pointer in the sigcontext.
527 * This will be saved when ever the FP and extended state context is
528 * saved on the user stack during the signal handler delivery to the user.
530 static void prepare_fx_sw_frame(void)
532 int fsave_header_size
= sizeof(struct i387_fsave_struct
);
533 int size
= xstate_size
+ FP_XSTATE_MAGIC2_SIZE
;
535 if (config_enabled(CONFIG_X86_32
))
536 size
+= fsave_header_size
;
538 fx_sw_reserved
.magic1
= FP_XSTATE_MAGIC1
;
539 fx_sw_reserved
.extended_size
= size
;
540 fx_sw_reserved
.xfeatures
= xfeatures_mask
;
541 fx_sw_reserved
.xstate_size
= xstate_size
;
543 if (config_enabled(CONFIG_IA32_EMULATION
)) {
544 fx_sw_reserved_ia32
= fx_sw_reserved
;
545 fx_sw_reserved_ia32
.extended_size
+= fsave_header_size
;
550 * Enable the extended processor state save/restore feature.
551 * Called once per CPU onlining.
553 void fpu__init_cpu_xstate(void)
555 if (!cpu_has_xsave
|| !xfeatures_mask
)
558 cr4_set_bits(X86_CR4_OSXSAVE
);
559 xsetbv(XCR_XFEATURE_ENABLED_MASK
, xfeatures_mask
);
563 * Record the offsets and sizes of different state managed by the xsave
566 static void __init
setup_xstate_features(void)
568 int eax
, ebx
, ecx
, edx
, leaf
= 0x2;
570 xfeatures_nr
= fls64(xfeatures_mask
);
573 cpuid_count(XSTATE_CPUID
, leaf
, &eax
, &ebx
, &ecx
, &edx
);
578 xstate_offsets
[leaf
] = ebx
;
579 xstate_sizes
[leaf
] = eax
;
585 static void print_xstate_feature(u64 xstate_mask
)
587 const char *feature_name
;
589 if (cpu_has_xfeatures(xstate_mask
, &feature_name
))
590 pr_info("x86/fpu: Supporting XSAVE feature 0x%02Lx: '%s'\n", xstate_mask
, feature_name
);
594 * Print out all the supported xstate features:
596 static void print_xstate_features(void)
598 print_xstate_feature(XSTATE_FP
);
599 print_xstate_feature(XSTATE_SSE
);
600 print_xstate_feature(XSTATE_YMM
);
601 print_xstate_feature(XSTATE_BNDREGS
);
602 print_xstate_feature(XSTATE_BNDCSR
);
603 print_xstate_feature(XSTATE_OPMASK
);
604 print_xstate_feature(XSTATE_ZMM_Hi256
);
605 print_xstate_feature(XSTATE_Hi16_ZMM
);
609 * This function sets up offsets and sizes of all extended states in
610 * xsave area. This supports both standard format and compacted format
611 * of the xsave aread.
616 void setup_xstate_comp(void)
618 unsigned int xstate_comp_sizes
[sizeof(xfeatures_mask
)*8];
622 * The FP xstates and SSE xstates are legacy states. They are always
623 * in the fixed offsets in the xsave area in either compacted form
626 xstate_comp_offsets
[0] = 0;
627 xstate_comp_offsets
[1] = offsetof(struct i387_fxsave_struct
, xmm_space
);
629 if (!cpu_has_xsaves
) {
630 for (i
= 2; i
< xfeatures_nr
; i
++) {
631 if (test_bit(i
, (unsigned long *)&xfeatures_mask
)) {
632 xstate_comp_offsets
[i
] = xstate_offsets
[i
];
633 xstate_comp_sizes
[i
] = xstate_sizes
[i
];
639 xstate_comp_offsets
[2] = FXSAVE_SIZE
+ XSAVE_HDR_SIZE
;
641 for (i
= 2; i
< xfeatures_nr
; i
++) {
642 if (test_bit(i
, (unsigned long *)&xfeatures_mask
))
643 xstate_comp_sizes
[i
] = xstate_sizes
[i
];
645 xstate_comp_sizes
[i
] = 0;
648 xstate_comp_offsets
[i
] = xstate_comp_offsets
[i
-1]
649 + xstate_comp_sizes
[i
-1];
655 * setup the xstate image representing the init state
657 static void setup_init_fpu_buf(void)
659 static int on_boot_cpu
= 1;
668 setup_xstate_features();
669 print_xstate_features();
671 if (cpu_has_xsaves
) {
672 init_xstate_ctx
.header
.xcomp_bv
= (u64
)1 << 63 | xfeatures_mask
;
673 init_xstate_ctx
.header
.xfeatures
= xfeatures_mask
;
677 * Init all the features state with header_bv being 0x0
679 xrstor_state_booting(&init_xstate_ctx
, -1);
682 * Dump the init state again. This is to identify the init state
683 * of any feature which is not represented by all zero's.
685 xsave_state_booting(&init_xstate_ctx
);
689 * Calculate total size of enabled xstates in XCR0/xfeatures_mask.
691 static void __init
init_xstate_size(void)
693 unsigned int eax
, ebx
, ecx
, edx
;
696 if (!cpu_has_xsaves
) {
697 cpuid_count(XSTATE_CPUID
, 0, &eax
, &ebx
, &ecx
, &edx
);
702 xstate_size
= FXSAVE_SIZE
+ XSAVE_HDR_SIZE
;
703 for (i
= 2; i
< 64; i
++) {
704 if (test_bit(i
, (unsigned long *)&xfeatures_mask
)) {
705 cpuid_count(XSTATE_CPUID
, i
, &eax
, &ebx
, &ecx
, &edx
);
712 * Enable and initialize the xsave feature.
713 * Called once per system bootup.
715 * ( Not marked __init because of false positive section warnings. )
717 void fpu__init_system_xstate(void)
719 unsigned int eax
, ebx
, ecx
, edx
;
720 static bool on_boot_cpu
= 1;
726 if (!cpu_has_xsave
) {
727 pr_info("x86/fpu: Legacy x87 FPU detected.\n");
731 if (boot_cpu_data
.cpuid_level
< XSTATE_CPUID
) {
732 WARN(1, "x86/fpu: XSTATE_CPUID missing!\n");
736 cpuid_count(XSTATE_CPUID
, 0, &eax
, &ebx
, &ecx
, &edx
);
737 xfeatures_mask
= eax
+ ((u64
)edx
<< 32);
739 if ((xfeatures_mask
& XSTATE_FPSSE
) != XSTATE_FPSSE
) {
740 pr_err("x86/fpu: FP/SSE not present amongst the CPU's xstate features: 0x%llx.\n", xfeatures_mask
);
745 * Support only the state known to OS.
747 xfeatures_mask
= xfeatures_mask
& XCNTXT_MASK
;
749 /* Enable xstate instructions to be able to continue with initialization: */
750 fpu__init_cpu_xstate();
753 * Recompute the context size for enabled features
757 update_regset_xstate_info(xstate_size
, xfeatures_mask
);
758 prepare_fx_sw_frame();
759 setup_init_fpu_buf();
761 pr_info("x86/fpu: Enabled xstate features 0x%llx, context size is 0x%x bytes, using '%s' format.\n",
764 cpu_has_xsaves
? "compacted" : "standard");
768 * Restore minimal FPU state after suspend:
770 void fpu__resume_cpu(void)
773 * Restore XCR0 on xsave capable CPUs:
776 xsetbv(XCR_XFEATURE_ENABLED_MASK
, xfeatures_mask
);
780 * Given the xsave area and a state inside, this function returns the
781 * address of the state.
783 * This is the API that is called to get xstate address in either
784 * standard format or compacted format of xsave area.
787 * xsave: base address of the xsave area;
788 * xstate: state which is defined in xsave.h (e.g. XSTATE_FP, XSTATE_SSE,
791 * address of the state in the xsave area.
793 void *get_xsave_addr(struct xsave_struct
*xsave
, int xstate
)
795 int feature
= fls64(xstate
) - 1;
796 if (!test_bit(feature
, (unsigned long *)&xfeatures_mask
))
799 return (void *)xsave
+ xstate_comp_offsets
[feature
];
801 EXPORT_SYMBOL_GPL(get_xsave_addr
);