diff options
Diffstat (limited to 'arch/x86/kernel/i387.c')
-rw-r--r-- | arch/x86/kernel/i387.c | 292 |
1 files changed, 28 insertions, 264 deletions
diff --git a/arch/x86/kernel/i387.c b/arch/x86/kernel/i387.c index f250431fb505..675a05012449 100644 --- a/arch/x86/kernel/i387.c +++ b/arch/x86/kernel/i387.c | |||
@@ -19,24 +19,17 @@ | |||
19 | #include <asm/fpu-internal.h> | 19 | #include <asm/fpu-internal.h> |
20 | #include <asm/user.h> | 20 | #include <asm/user.h> |
21 | 21 | ||
22 | #ifdef CONFIG_X86_64 | ||
23 | # include <asm/sigcontext32.h> | ||
24 | # include <asm/user32.h> | ||
25 | #else | ||
26 | # define save_i387_xstate_ia32 save_i387_xstate | ||
27 | # define restore_i387_xstate_ia32 restore_i387_xstate | ||
28 | # define _fpstate_ia32 _fpstate | ||
29 | # define _xstate_ia32 _xstate | ||
30 | # define sig_xstate_ia32_size sig_xstate_size | ||
31 | # define fx_sw_reserved_ia32 fx_sw_reserved | ||
32 | # define user_i387_ia32_struct user_i387_struct | ||
33 | # define user32_fxsr_struct user_fxsr_struct | ||
34 | #endif | ||
35 | |||
36 | /* | 22 | /* |
37 | * Were we in an interrupt that interrupted kernel mode? | 23 | * Were we in an interrupt that interrupted kernel mode? |
38 | * | 24 | * |
39 | * We can do a kernel_fpu_begin/end() pair *ONLY* if that | 25 | * For now, with eagerfpu we will return interrupted kernel FPU |
26 | * state as not-idle. TBD: Ideally we can change the return value | ||
27 | * to something like __thread_has_fpu(current). But we need to | ||
28 | * be careful of doing __thread_clear_has_fpu() before saving | ||
29 | * the FPU etc for supporting nested uses etc. For now, take | ||
30 | * the simple route! | ||
31 | * | ||
32 | * On others, we can do a kernel_fpu_begin/end() pair *ONLY* if that | ||
40 | * pair does nothing at all: the thread must not have fpu (so | 33 | * pair does nothing at all: the thread must not have fpu (so |
41 | * that we don't try to save the FPU state), and TS must | 34 | * that we don't try to save the FPU state), and TS must |
42 | * be set (so that the clts/stts pair does nothing that is | 35 | * be set (so that the clts/stts pair does nothing that is |
@@ -44,6 +37,9 @@ | |||
44 | */ | 37 | */ |
45 | static inline bool interrupted_kernel_fpu_idle(void) | 38 | static inline bool interrupted_kernel_fpu_idle(void) |
46 | { | 39 | { |
40 | if (use_eager_fpu()) | ||
41 | return 0; | ||
42 | |||
47 | return !__thread_has_fpu(current) && | 43 | return !__thread_has_fpu(current) && |
48 | (read_cr0() & X86_CR0_TS); | 44 | (read_cr0() & X86_CR0_TS); |
49 | } | 45 | } |
@@ -77,29 +73,29 @@ bool irq_fpu_usable(void) | |||
77 | } | 73 | } |
78 | EXPORT_SYMBOL(irq_fpu_usable); | 74 | EXPORT_SYMBOL(irq_fpu_usable); |
79 | 75 | ||
80 | void kernel_fpu_begin(void) | 76 | void __kernel_fpu_begin(void) |
81 | { | 77 | { |
82 | struct task_struct *me = current; | 78 | struct task_struct *me = current; |
83 | 79 | ||
84 | WARN_ON_ONCE(!irq_fpu_usable()); | ||
85 | preempt_disable(); | ||
86 | if (__thread_has_fpu(me)) { | 80 | if (__thread_has_fpu(me)) { |
87 | __save_init_fpu(me); | 81 | __save_init_fpu(me); |
88 | __thread_clear_has_fpu(me); | 82 | __thread_clear_has_fpu(me); |
89 | /* We do 'stts()' in kernel_fpu_end() */ | 83 | /* We do 'stts()' in __kernel_fpu_end() */ |
90 | } else { | 84 | } else if (!use_eager_fpu()) { |
91 | this_cpu_write(fpu_owner_task, NULL); | 85 | this_cpu_write(fpu_owner_task, NULL); |
92 | clts(); | 86 | clts(); |
93 | } | 87 | } |
94 | } | 88 | } |
95 | EXPORT_SYMBOL(kernel_fpu_begin); | 89 | EXPORT_SYMBOL(__kernel_fpu_begin); |
96 | 90 | ||
97 | void kernel_fpu_end(void) | 91 | void __kernel_fpu_end(void) |
98 | { | 92 | { |
99 | stts(); | 93 | if (use_eager_fpu()) |
100 | preempt_enable(); | 94 | math_state_restore(); |
95 | else | ||
96 | stts(); | ||
101 | } | 97 | } |
102 | EXPORT_SYMBOL(kernel_fpu_end); | 98 | EXPORT_SYMBOL(__kernel_fpu_end); |
103 | 99 | ||
104 | void unlazy_fpu(struct task_struct *tsk) | 100 | void unlazy_fpu(struct task_struct *tsk) |
105 | { | 101 | { |
@@ -113,23 +109,15 @@ void unlazy_fpu(struct task_struct *tsk) | |||
113 | } | 109 | } |
114 | EXPORT_SYMBOL(unlazy_fpu); | 110 | EXPORT_SYMBOL(unlazy_fpu); |
115 | 111 | ||
116 | #ifdef CONFIG_MATH_EMULATION | 112 | unsigned int mxcsr_feature_mask __read_mostly = 0xffffffffu; |
117 | # define HAVE_HWFP (boot_cpu_data.hard_math) | ||
118 | #else | ||
119 | # define HAVE_HWFP 1 | ||
120 | #endif | ||
121 | |||
122 | static unsigned int mxcsr_feature_mask __read_mostly = 0xffffffffu; | ||
123 | unsigned int xstate_size; | 113 | unsigned int xstate_size; |
124 | EXPORT_SYMBOL_GPL(xstate_size); | 114 | EXPORT_SYMBOL_GPL(xstate_size); |
125 | unsigned int sig_xstate_ia32_size = sizeof(struct _fpstate_ia32); | ||
126 | static struct i387_fxsave_struct fx_scratch __cpuinitdata; | 115 | static struct i387_fxsave_struct fx_scratch __cpuinitdata; |
127 | 116 | ||
128 | static void __cpuinit mxcsr_feature_mask_init(void) | 117 | static void __cpuinit mxcsr_feature_mask_init(void) |
129 | { | 118 | { |
130 | unsigned long mask = 0; | 119 | unsigned long mask = 0; |
131 | 120 | ||
132 | clts(); | ||
133 | if (cpu_has_fxsr) { | 121 | if (cpu_has_fxsr) { |
134 | memset(&fx_scratch, 0, sizeof(struct i387_fxsave_struct)); | 122 | memset(&fx_scratch, 0, sizeof(struct i387_fxsave_struct)); |
135 | asm volatile("fxsave %0" : : "m" (fx_scratch)); | 123 | asm volatile("fxsave %0" : : "m" (fx_scratch)); |
@@ -138,7 +126,6 @@ static void __cpuinit mxcsr_feature_mask_init(void) | |||
138 | mask = 0x0000ffbf; | 126 | mask = 0x0000ffbf; |
139 | } | 127 | } |
140 | mxcsr_feature_mask &= mask; | 128 | mxcsr_feature_mask &= mask; |
141 | stts(); | ||
142 | } | 129 | } |
143 | 130 | ||
144 | static void __cpuinit init_thread_xstate(void) | 131 | static void __cpuinit init_thread_xstate(void) |
@@ -192,9 +179,8 @@ void __cpuinit fpu_init(void) | |||
192 | init_thread_xstate(); | 179 | init_thread_xstate(); |
193 | 180 | ||
194 | mxcsr_feature_mask_init(); | 181 | mxcsr_feature_mask_init(); |
195 | /* clean state in init */ | 182 | xsave_init(); |
196 | current_thread_info()->status = 0; | 183 | eager_fpu_init(); |
197 | clear_used_math(); | ||
198 | } | 184 | } |
199 | 185 | ||
200 | void fpu_finit(struct fpu *fpu) | 186 | void fpu_finit(struct fpu *fpu) |
@@ -205,12 +191,7 @@ void fpu_finit(struct fpu *fpu) | |||
205 | } | 191 | } |
206 | 192 | ||
207 | if (cpu_has_fxsr) { | 193 | if (cpu_has_fxsr) { |
208 | struct i387_fxsave_struct *fx = &fpu->state->fxsave; | 194 | fx_finit(&fpu->state->fxsave); |
209 | |||
210 | memset(fx, 0, xstate_size); | ||
211 | fx->cwd = 0x37f; | ||
212 | if (cpu_has_xmm) | ||
213 | fx->mxcsr = MXCSR_DEFAULT; | ||
214 | } else { | 195 | } else { |
215 | struct i387_fsave_struct *fp = &fpu->state->fsave; | 196 | struct i387_fsave_struct *fp = &fpu->state->fsave; |
216 | memset(fp, 0, xstate_size); | 197 | memset(fp, 0, xstate_size); |
@@ -454,7 +435,7 @@ static inline u32 twd_fxsr_to_i387(struct i387_fxsave_struct *fxsave) | |||
454 | * FXSR floating point environment conversions. | 435 | * FXSR floating point environment conversions. |
455 | */ | 436 | */ |
456 | 437 | ||
457 | static void | 438 | void |
458 | convert_from_fxsr(struct user_i387_ia32_struct *env, struct task_struct *tsk) | 439 | convert_from_fxsr(struct user_i387_ia32_struct *env, struct task_struct *tsk) |
459 | { | 440 | { |
460 | struct i387_fxsave_struct *fxsave = &tsk->thread.fpu.state->fxsave; | 441 | struct i387_fxsave_struct *fxsave = &tsk->thread.fpu.state->fxsave; |
@@ -491,8 +472,8 @@ convert_from_fxsr(struct user_i387_ia32_struct *env, struct task_struct *tsk) | |||
491 | memcpy(&to[i], &from[i], sizeof(to[0])); | 472 | memcpy(&to[i], &from[i], sizeof(to[0])); |
492 | } | 473 | } |
493 | 474 | ||
494 | static void convert_to_fxsr(struct task_struct *tsk, | 475 | void convert_to_fxsr(struct task_struct *tsk, |
495 | const struct user_i387_ia32_struct *env) | 476 | const struct user_i387_ia32_struct *env) |
496 | 477 | ||
497 | { | 478 | { |
498 | struct i387_fxsave_struct *fxsave = &tsk->thread.fpu.state->fxsave; | 479 | struct i387_fxsave_struct *fxsave = &tsk->thread.fpu.state->fxsave; |
@@ -589,223 +570,6 @@ int fpregs_set(struct task_struct *target, const struct user_regset *regset, | |||
589 | } | 570 | } |
590 | 571 | ||
591 | /* | 572 | /* |
592 | * Signal frame handlers. | ||
593 | */ | ||
594 | |||
595 | static inline int save_i387_fsave(struct _fpstate_ia32 __user *buf) | ||
596 | { | ||
597 | struct task_struct *tsk = current; | ||
598 | struct i387_fsave_struct *fp = &tsk->thread.fpu.state->fsave; | ||
599 | |||
600 | fp->status = fp->swd; | ||
601 | if (__copy_to_user(buf, fp, sizeof(struct i387_fsave_struct))) | ||
602 | return -1; | ||
603 | return 1; | ||
604 | } | ||
605 | |||
606 | static int save_i387_fxsave(struct _fpstate_ia32 __user *buf) | ||
607 | { | ||
608 | struct task_struct *tsk = current; | ||
609 | struct i387_fxsave_struct *fx = &tsk->thread.fpu.state->fxsave; | ||
610 | struct user_i387_ia32_struct env; | ||
611 | int err = 0; | ||
612 | |||
613 | convert_from_fxsr(&env, tsk); | ||
614 | if (__copy_to_user(buf, &env, sizeof(env))) | ||
615 | return -1; | ||
616 | |||
617 | err |= __put_user(fx->swd, &buf->status); | ||
618 | err |= __put_user(X86_FXSR_MAGIC, &buf->magic); | ||
619 | if (err) | ||
620 | return -1; | ||
621 | |||
622 | if (__copy_to_user(&buf->_fxsr_env[0], fx, xstate_size)) | ||
623 | return -1; | ||
624 | return 1; | ||
625 | } | ||
626 | |||
627 | static int save_i387_xsave(void __user *buf) | ||
628 | { | ||
629 | struct task_struct *tsk = current; | ||
630 | struct _fpstate_ia32 __user *fx = buf; | ||
631 | int err = 0; | ||
632 | |||
633 | |||
634 | sanitize_i387_state(tsk); | ||
635 | |||
636 | /* | ||
637 | * For legacy compatible, we always set FP/SSE bits in the bit | ||
638 | * vector while saving the state to the user context. | ||
639 | * This will enable us capturing any changes(during sigreturn) to | ||
640 | * the FP/SSE bits by the legacy applications which don't touch | ||
641 | * xstate_bv in the xsave header. | ||
642 | * | ||
643 | * xsave aware applications can change the xstate_bv in the xsave | ||
644 | * header as well as change any contents in the memory layout. | ||
645 | * xrestore as part of sigreturn will capture all the changes. | ||
646 | */ | ||
647 | tsk->thread.fpu.state->xsave.xsave_hdr.xstate_bv |= XSTATE_FPSSE; | ||
648 | |||
649 | if (save_i387_fxsave(fx) < 0) | ||
650 | return -1; | ||
651 | |||
652 | err = __copy_to_user(&fx->sw_reserved, &fx_sw_reserved_ia32, | ||
653 | sizeof(struct _fpx_sw_bytes)); | ||
654 | err |= __put_user(FP_XSTATE_MAGIC2, | ||
655 | (__u32 __user *) (buf + sig_xstate_ia32_size | ||
656 | - FP_XSTATE_MAGIC2_SIZE)); | ||
657 | if (err) | ||
658 | return -1; | ||
659 | |||
660 | return 1; | ||
661 | } | ||
662 | |||
663 | int save_i387_xstate_ia32(void __user *buf) | ||
664 | { | ||
665 | struct _fpstate_ia32 __user *fp = (struct _fpstate_ia32 __user *) buf; | ||
666 | struct task_struct *tsk = current; | ||
667 | |||
668 | if (!used_math()) | ||
669 | return 0; | ||
670 | |||
671 | if (!access_ok(VERIFY_WRITE, buf, sig_xstate_ia32_size)) | ||
672 | return -EACCES; | ||
673 | /* | ||
674 | * This will cause a "finit" to be triggered by the next | ||
675 | * attempted FPU operation by the 'current' process. | ||
676 | */ | ||
677 | clear_used_math(); | ||
678 | |||
679 | if (!HAVE_HWFP) { | ||
680 | return fpregs_soft_get(current, NULL, | ||
681 | 0, sizeof(struct user_i387_ia32_struct), | ||
682 | NULL, fp) ? -1 : 1; | ||
683 | } | ||
684 | |||
685 | unlazy_fpu(tsk); | ||
686 | |||
687 | if (cpu_has_xsave) | ||
688 | return save_i387_xsave(fp); | ||
689 | if (cpu_has_fxsr) | ||
690 | return save_i387_fxsave(fp); | ||
691 | else | ||
692 | return save_i387_fsave(fp); | ||
693 | } | ||
694 | |||
695 | static inline int restore_i387_fsave(struct _fpstate_ia32 __user *buf) | ||
696 | { | ||
697 | struct task_struct *tsk = current; | ||
698 | |||
699 | return __copy_from_user(&tsk->thread.fpu.state->fsave, buf, | ||
700 | sizeof(struct i387_fsave_struct)); | ||
701 | } | ||
702 | |||
703 | static int restore_i387_fxsave(struct _fpstate_ia32 __user *buf, | ||
704 | unsigned int size) | ||
705 | { | ||
706 | struct task_struct *tsk = current; | ||
707 | struct user_i387_ia32_struct env; | ||
708 | int err; | ||
709 | |||
710 | err = __copy_from_user(&tsk->thread.fpu.state->fxsave, &buf->_fxsr_env[0], | ||
711 | size); | ||
712 | /* mxcsr reserved bits must be masked to zero for security reasons */ | ||
713 | tsk->thread.fpu.state->fxsave.mxcsr &= mxcsr_feature_mask; | ||
714 | if (err || __copy_from_user(&env, buf, sizeof(env))) | ||
715 | return 1; | ||
716 | convert_to_fxsr(tsk, &env); | ||
717 | |||
718 | return 0; | ||
719 | } | ||
720 | |||
721 | static int restore_i387_xsave(void __user *buf) | ||
722 | { | ||
723 | struct _fpx_sw_bytes fx_sw_user; | ||
724 | struct _fpstate_ia32 __user *fx_user = | ||
725 | ((struct _fpstate_ia32 __user *) buf); | ||
726 | struct i387_fxsave_struct __user *fx = | ||
727 | (struct i387_fxsave_struct __user *) &fx_user->_fxsr_env[0]; | ||
728 | struct xsave_hdr_struct *xsave_hdr = | ||
729 | ¤t->thread.fpu.state->xsave.xsave_hdr; | ||
730 | u64 mask; | ||
731 | int err; | ||
732 | |||
733 | if (check_for_xstate(fx, buf, &fx_sw_user)) | ||
734 | goto fx_only; | ||
735 | |||
736 | mask = fx_sw_user.xstate_bv; | ||
737 | |||
738 | err = restore_i387_fxsave(buf, fx_sw_user.xstate_size); | ||
739 | |||
740 | xsave_hdr->xstate_bv &= pcntxt_mask; | ||
741 | /* | ||
742 | * These bits must be zero. | ||
743 | */ | ||
744 | xsave_hdr->reserved1[0] = xsave_hdr->reserved1[1] = 0; | ||
745 | |||
746 | /* | ||
747 | * Init the state that is not present in the memory layout | ||
748 | * and enabled by the OS. | ||
749 | */ | ||
750 | mask = ~(pcntxt_mask & ~mask); | ||
751 | xsave_hdr->xstate_bv &= mask; | ||
752 | |||
753 | return err; | ||
754 | fx_only: | ||
755 | /* | ||
756 | * Couldn't find the extended state information in the memory | ||
757 | * layout. Restore the FP/SSE and init the other extended state | ||
758 | * enabled by the OS. | ||
759 | */ | ||
760 | xsave_hdr->xstate_bv = XSTATE_FPSSE; | ||
761 | return restore_i387_fxsave(buf, sizeof(struct i387_fxsave_struct)); | ||
762 | } | ||
763 | |||
764 | int restore_i387_xstate_ia32(void __user *buf) | ||
765 | { | ||
766 | int err; | ||
767 | struct task_struct *tsk = current; | ||
768 | struct _fpstate_ia32 __user *fp = (struct _fpstate_ia32 __user *) buf; | ||
769 | |||
770 | if (HAVE_HWFP) | ||
771 | clear_fpu(tsk); | ||
772 | |||
773 | if (!buf) { | ||
774 | if (used_math()) { | ||
775 | clear_fpu(tsk); | ||
776 | clear_used_math(); | ||
777 | } | ||
778 | |||
779 | return 0; | ||
780 | } else | ||
781 | if (!access_ok(VERIFY_READ, buf, sig_xstate_ia32_size)) | ||
782 | return -EACCES; | ||
783 | |||
784 | if (!used_math()) { | ||
785 | err = init_fpu(tsk); | ||
786 | if (err) | ||
787 | return err; | ||
788 | } | ||
789 | |||
790 | if (HAVE_HWFP) { | ||
791 | if (cpu_has_xsave) | ||
792 | err = restore_i387_xsave(buf); | ||
793 | else if (cpu_has_fxsr) | ||
794 | err = restore_i387_fxsave(fp, sizeof(struct | ||
795 | i387_fxsave_struct)); | ||
796 | else | ||
797 | err = restore_i387_fsave(fp); | ||
798 | } else { | ||
799 | err = fpregs_soft_set(current, NULL, | ||
800 | 0, sizeof(struct user_i387_ia32_struct), | ||
801 | NULL, fp) != 0; | ||
802 | } | ||
803 | set_used_math(); | ||
804 | |||
805 | return err; | ||
806 | } | ||
807 | |||
808 | /* | ||
809 | * FPU state for core dumps. | 573 | * FPU state for core dumps. |
810 | * This is only used for a.out dumps now. | 574 | * This is only used for a.out dumps now. |
811 | * It is declared generically using elf_fpregset_t (which is | 575 | * It is declared generically using elf_fpregset_t (which is |