summaryrefslogtreecommitdiff
path: root/usr/src/uts/i86pc/os/hma_fpu.c
blob: 14cfa8baed22bbea542e20574794f94920231727 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
/*
 * This file and its contents are supplied under the terms of the
 * Common Development and Distribution License ("CDDL"), version 1.0.
 * You may only use this file in accordance with the terms of version
 * 1.0 of the CDDL.
 *
 * A full copy of the text of the CDDL should have accompanied this
 * source.  A copy of the CDDL is also available via the Internet at
 * http://www.illumos.org/license/CDDL.
 */

/*
 * Copyright (c) 2018, Joyent, Inc.
 */

/*
 * This implements the hypervisor multiplexor FPU API. Its purpose is to make it
 * easy to switch between the host and guest hypervisor while hiding all the
 * details about CR0.TS and how to save the host's state as required.
 */

#include <sys/pcb.h>
#include <sys/kmem.h>
#include <sys/debug.h>
#include <sys/cmn_err.h>
#include <sys/ddi.h>
#include <sys/sunddi.h>
#include <sys/hma.h>
#include <sys/x86_archext.h>
#include <sys/archsystm.h>

struct hma_fpu {
	fpu_ctx_t	hf_guest_fpu;
	kthread_t	*hf_curthread;
	boolean_t	hf_inguest;
};

int
hma_fpu_init(hma_fpu_t *fpu)
{
	struct xsave_state *xs;

	ASSERT0(fpu->hf_inguest);

	switch (fp_save_mech) {
	case FP_FXSAVE:
		bcopy(&sse_initial, fpu->hf_guest_fpu.fpu_regs.kfpu_u.kfpu_fx,
		    sizeof (struct fxsave_state));
		fpu->hf_guest_fpu.fpu_xsave_mask = 0;
		break;
	case FP_XSAVE:
		/*
		 * Zero everything in the xsave case as we may have data in
		 * the structure that's not part of the initial value (which
		 * only really deals with a small portion of the xsave state).
		 */
		xs = fpu->hf_guest_fpu.fpu_regs.kfpu_u.kfpu_xs;
		bzero(xs, cpuid_get_xsave_size());
		bcopy(&avx_initial, xs, sizeof (*xs));
		xs->xs_xstate_bv = XFEATURE_LEGACY_FP | XFEATURE_SSE;
		fpu->hf_guest_fpu.fpu_xsave_mask = XFEATURE_FP_ALL;
		break;
	default:
		panic("Invalid fp_save_mech");
	}

	fpu->hf_guest_fpu.fpu_flags = FPU_EN | FPU_VALID;

	return (0);
}

void
hma_fpu_free(hma_fpu_t *fpu)
{
	if (fpu == NULL)
		return;

	ASSERT3P(fpu->hf_guest_fpu.fpu_regs.kfpu_u.kfpu_generic, !=, NULL);
	kmem_cache_free(fpsave_cachep,
	    fpu->hf_guest_fpu.fpu_regs.kfpu_u.kfpu_generic);
	kmem_free(fpu, sizeof (*fpu));
}

hma_fpu_t *
hma_fpu_alloc(int kmflag)
{
	hma_fpu_t *fpu;

	fpu = kmem_zalloc(sizeof (hma_fpu_t), kmflag);
	if (fpu == NULL)
		return (NULL);

	fpu->hf_guest_fpu.fpu_regs.kfpu_u.kfpu_generic =
	    kmem_cache_alloc(fpsave_cachep, kmflag);
	if (fpu->hf_guest_fpu.fpu_regs.kfpu_u.kfpu_generic == NULL) {
		kmem_free(fpu, sizeof (hma_fpu_t));
		return (NULL);
	}
	fpu->hf_inguest = B_FALSE;

	/*
	 * Make sure the entire structure is zero.
	 */
	switch (fp_save_mech) {
	case FP_FXSAVE:
		bzero(fpu->hf_guest_fpu.fpu_regs.kfpu_u.kfpu_generic,
		    sizeof (struct fxsave_state));
		break;
	case FP_XSAVE:
		bzero(fpu->hf_guest_fpu.fpu_regs.kfpu_u.kfpu_generic,
		    cpuid_get_xsave_size());
		break;
	default:
		panic("Invalid fp_save_mech");
	}

	return (fpu);
}

void
hma_fpu_start_guest(hma_fpu_t *fpu)
{
	/*
	 * Note, we don't check / assert whether or not t_prempt is true because
	 * there are contexts where this is safe to call (from a context op)
	 * where t_preempt may not be set.
	 */
	ASSERT3S(fpu->hf_inguest, ==, B_FALSE);
	ASSERT3P(fpu->hf_curthread, ==, NULL);
	ASSERT3P(curthread->t_lwp, !=, NULL);
	ASSERT3U(fpu->hf_guest_fpu.fpu_flags & FPU_EN, !=, 0);
	ASSERT3U(fpu->hf_guest_fpu.fpu_flags & FPU_VALID, !=, 0);

	fpu->hf_inguest = B_TRUE;
	fpu->hf_curthread = curthread;


	fp_save(&curthread->t_lwp->lwp_pcb.pcb_fpu);
	fp_restore(&fpu->hf_guest_fpu);
	fpu->hf_guest_fpu.fpu_flags &= ~FPU_VALID;
}

void
hma_fpu_stop_guest(hma_fpu_t *fpu)
{
	ASSERT3S(fpu->hf_inguest, ==, B_TRUE);
	ASSERT3P(fpu->hf_curthread, ==, curthread);
	ASSERT3U(fpu->hf_guest_fpu.fpu_flags & FPU_EN, !=, 0);
	ASSERT3U(fpu->hf_guest_fpu.fpu_flags & FPU_VALID, ==, 0);

	/*
	 * Note, we can't use fp_save because it assumes that we're saving to
	 * the thread's PCB and not somewhere else. Because this is a different
	 * FPU context, we instead have to do this ourselves.
	 */
	switch (fp_save_mech) {
	case FP_FXSAVE:
		fpxsave(fpu->hf_guest_fpu.fpu_regs.kfpu_u.kfpu_fx);
		break;
	case FP_XSAVE:
		xsavep(fpu->hf_guest_fpu.fpu_regs.kfpu_u.kfpu_xs,
		    fpu->hf_guest_fpu.fpu_xsave_mask);
		break;
	default:
		panic("Invalid fp_save_mech");
		/*NOTREACHED*/
	}
	fpu->hf_guest_fpu.fpu_flags |= FPU_VALID;

	fp_restore(&curthread->t_lwp->lwp_pcb.pcb_fpu);

	fpu->hf_inguest = B_FALSE;
	fpu->hf_curthread = NULL;
}

void
hma_fpu_get_fxsave_state(const hma_fpu_t *fpu, struct fxsave_state *fx)
{
	const struct fxsave_state *guest;

	ASSERT3S(fpu->hf_inguest, ==, B_FALSE);

	guest = fpu->hf_guest_fpu.fpu_regs.kfpu_u.kfpu_fx;
	bcopy(guest, fx, sizeof (*fx));
}

int
hma_fpu_set_fxsave_state(hma_fpu_t *fpu, const struct fxsave_state *fx)
{
	struct fxsave_state *gfx;
	struct xsave_state *gxs;

	ASSERT3S(fpu->hf_inguest, ==, B_FALSE);

	/*
	 * If reserved bits are set in fx_mxcsr, then we will take a #GP when
	 * we restore them. Reject this outright.
	 *
	 * We do not need to check if we are dealing with state that has pending
	 * exceptions. This was only the case with the original FPU save and
	 * restore mechanisms (fsave/frstor). When using fxsave/fxrstor and
	 * xsave/xrstor they will be deferred to the user using the FPU, which
	 * is what we'd want here (they'd be used in guest context).
	 */
	if ((fx->fx_mxcsr & ~sse_mxcsr_mask) != 0)
		return (EINVAL);

	switch (fp_save_mech) {
	case FP_FXSAVE:
		gfx = fpu->hf_guest_fpu.fpu_regs.kfpu_u.kfpu_fx;
		bcopy(fx, gfx, sizeof (*fx));
		break;
	case FP_XSAVE:
		gxs = fpu->hf_guest_fpu.fpu_regs.kfpu_u.kfpu_xs;
		bzero(gxs, cpuid_get_xsave_size());
		bcopy(fx, &gxs->xs_fxsave, sizeof (*fx));
		gxs->xs_xstate_bv = XFEATURE_LEGACY_FP | XFEATURE_SSE;
		break;
	default:
		panic("Invalid fp_save_mech");
		/* NOTREACHED */
	}

	return (0);
}