xref: /freebsd/usr.sbin/bhyve/amd64/task_switch.c (revision 1631382cf2820245cc72965498ff174bb548dd63)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause
3  *
4  * Copyright (c) 2014 Neel Natu <neel@freebsd.org>
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  *
16  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND
17  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
20  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26  * SUCH DAMAGE.
27  */
28 
29 #include <sys/param.h>
30 #include <sys/_iovec.h>
31 #include <sys/mman.h>
32 
33 #include <x86/psl.h>
34 #include <x86/specialreg.h>
35 #include <machine/vmm.h>
36 #include <machine/vmm_instruction_emul.h>
37 
38 #include <assert.h>
39 #include <errno.h>
40 #include <stdbool.h>
41 #include <stdio.h>
42 #include <stdlib.h>
43 
44 #include <vmmapi.h>
45 
46 #include "bhyverun.h"
47 #include "debug.h"
48 
49 /*
50  * Using 'struct i386tss' is tempting but causes myriad sign extension
51  * issues because all of its fields are defined as signed integers.
52  */
53 struct tss32 {
54 	uint16_t	tss_link;
55 	uint16_t	rsvd1;
56 	uint32_t	tss_esp0;
57 	uint16_t	tss_ss0;
58 	uint16_t	rsvd2;
59 	uint32_t	tss_esp1;
60 	uint16_t	tss_ss1;
61 	uint16_t	rsvd3;
62 	uint32_t	tss_esp2;
63 	uint16_t	tss_ss2;
64 	uint16_t	rsvd4;
65 	uint32_t	tss_cr3;
66 	uint32_t	tss_eip;
67 	uint32_t	tss_eflags;
68 	uint32_t	tss_eax;
69 	uint32_t	tss_ecx;
70 	uint32_t	tss_edx;
71 	uint32_t	tss_ebx;
72 	uint32_t	tss_esp;
73 	uint32_t	tss_ebp;
74 	uint32_t	tss_esi;
75 	uint32_t	tss_edi;
76 	uint16_t	tss_es;
77 	uint16_t	rsvd5;
78 	uint16_t	tss_cs;
79 	uint16_t	rsvd6;
80 	uint16_t	tss_ss;
81 	uint16_t	rsvd7;
82 	uint16_t	tss_ds;
83 	uint16_t	rsvd8;
84 	uint16_t	tss_fs;
85 	uint16_t	rsvd9;
86 	uint16_t	tss_gs;
87 	uint16_t	rsvd10;
88 	uint16_t	tss_ldt;
89 	uint16_t	rsvd11;
90 	uint16_t	tss_trap;
91 	uint16_t	tss_iomap;
92 };
93 static_assert(sizeof(struct tss32) == 104, "compile-time assertion failed");
94 
95 #define	SEL_START(sel)	(((sel) & ~0x7))
96 #define	SEL_LIMIT(sel)	(((sel) | 0x7))
97 #define	TSS_BUSY(type)	(((type) & 0x2) != 0)
98 
99 static uint64_t
100 GETREG(struct vcpu *vcpu, int reg)
101 {
102 	uint64_t val;
103 	int error;
104 
105 	error = vm_get_register(vcpu, reg, &val);
106 	assert(error == 0);
107 	return (val);
108 }
109 
110 static void
111 SETREG(struct vcpu *vcpu, int reg, uint64_t val)
112 {
113 	int error;
114 
115 	error = vm_set_register(vcpu, reg, val);
116 	assert(error == 0);
117 }
118 
119 static struct seg_desc
120 usd_to_seg_desc(struct user_segment_descriptor *usd)
121 {
122 	struct seg_desc seg_desc;
123 
124 	seg_desc.base = (u_int)USD_GETBASE(usd);
125 	if (usd->sd_gran)
126 		seg_desc.limit = (u_int)(USD_GETLIMIT(usd) << 12) | 0xfff;
127 	else
128 		seg_desc.limit = (u_int)USD_GETLIMIT(usd);
129 	seg_desc.access = usd->sd_type | usd->sd_dpl << 5 | usd->sd_p << 7;
130 	seg_desc.access |= usd->sd_xx << 12;
131 	seg_desc.access |= usd->sd_def32 << 14;
132 	seg_desc.access |= usd->sd_gran << 15;
133 
134 	return (seg_desc);
135 }
136 
137 /*
138  * Inject an exception with an error code that is a segment selector.
139  * The format of the error code is described in section 6.13, "Error Code",
140  * Intel SDM volume 3.
141  *
142  * Bit 0 (EXT) denotes whether the exception occurred during delivery
143  * of an external event like an interrupt.
144  *
145  * Bit 1 (IDT) indicates whether the selector points to a gate descriptor
146  * in the IDT.
147  *
148  * Bit 2(GDT/LDT) has the usual interpretation of Table Indicator (TI).
149  */
150 static void
151 sel_exception(struct vcpu *vcpu, int vector, uint16_t sel, int ext)
152 {
153 	/*
154 	 * Bit 2 from the selector is retained as-is in the error code.
155 	 *
156 	 * Bit 1 can be safely cleared because none of the selectors
157 	 * encountered during task switch emulation refer to a task
158 	 * gate in the IDT.
159 	 *
160 	 * Bit 0 is set depending on the value of 'ext'.
161 	 */
162 	sel &= ~0x3;
163 	if (ext)
164 		sel |= 0x1;
165 	vm_inject_fault(vcpu, vector, 1, sel);
166 }
167 
168 /*
169  * Return 0 if the selector 'sel' in within the limits of the GDT/LDT
170  * and non-zero otherwise.
171  */
172 static int
173 desc_table_limit_check(struct vcpu *vcpu, uint16_t sel)
174 {
175 	uint64_t base;
176 	uint32_t limit, access;
177 	int error, reg;
178 
179 	reg = ISLDT(sel) ? VM_REG_GUEST_LDTR : VM_REG_GUEST_GDTR;
180 	error = vm_get_desc(vcpu, reg, &base, &limit, &access);
181 	assert(error == 0);
182 
183 	if (reg == VM_REG_GUEST_LDTR) {
184 		if (SEG_DESC_UNUSABLE(access) || !SEG_DESC_PRESENT(access))
185 			return (-1);
186 	}
187 
188 	if (limit < SEL_LIMIT(sel))
189 		return (-1);
190 	else
191 		return (0);
192 }
193 
194 /*
195  * Read/write the segment descriptor 'desc' into the GDT/LDT slot referenced
196  * by the selector 'sel'.
197  *
198  * Returns 0 on success.
199  * Returns 1 if an exception was injected into the guest.
200  * Returns -1 otherwise.
201  */
202 static int
203 desc_table_rw(struct vcpu *vcpu, struct vm_guest_paging *paging,
204     uint16_t sel, struct user_segment_descriptor *desc, bool doread,
205     int *faultptr)
206 {
207 	struct iovec iov[2];
208 	uint64_t base;
209 	uint32_t limit, access;
210 	int error, reg;
211 
212 	reg = ISLDT(sel) ? VM_REG_GUEST_LDTR : VM_REG_GUEST_GDTR;
213 	error = vm_get_desc(vcpu, reg, &base, &limit, &access);
214 	assert(error == 0);
215 	assert(limit >= SEL_LIMIT(sel));
216 
217 	error = vm_copy_setup(vcpu, paging, base + SEL_START(sel),
218 	    sizeof(*desc), doread ? PROT_READ : PROT_WRITE, iov, nitems(iov),
219 	    faultptr);
220 	if (error || *faultptr)
221 		return (error);
222 
223 	if (doread)
224 		vm_copyin(iov, desc, sizeof(*desc));
225 	else
226 		vm_copyout(desc, iov, sizeof(*desc));
227 	return (0);
228 }
229 
230 static int
231 desc_table_read(struct vcpu *vcpu, struct vm_guest_paging *paging,
232     uint16_t sel, struct user_segment_descriptor *desc, int *faultptr)
233 {
234 	return (desc_table_rw(vcpu, paging, sel, desc, true, faultptr));
235 }
236 
237 static int
238 desc_table_write(struct vcpu *vcpu, struct vm_guest_paging *paging,
239     uint16_t sel, struct user_segment_descriptor *desc, int *faultptr)
240 {
241 	return (desc_table_rw(vcpu, paging, sel, desc, false, faultptr));
242 }
243 
244 /*
245  * Read the TSS descriptor referenced by 'sel' into 'desc'.
246  *
247  * Returns 0 on success.
248  * Returns 1 if an exception was injected into the guest.
249  * Returns -1 otherwise.
250  */
251 static int
252 read_tss_descriptor(struct vcpu *vcpu, struct vm_task_switch *ts,
253     uint16_t sel, struct user_segment_descriptor *desc, int *faultptr)
254 {
255 	struct vm_guest_paging sup_paging;
256 	int error;
257 
258 	assert(!ISLDT(sel));
259 	assert(IDXSEL(sel) != 0);
260 
261 	/* Fetch the new TSS descriptor */
262 	if (desc_table_limit_check(vcpu, sel)) {
263 		if (ts->reason == TSR_IRET)
264 			sel_exception(vcpu, IDT_TS, sel, ts->ext);
265 		else
266 			sel_exception(vcpu, IDT_GP, sel, ts->ext);
267 		return (1);
268 	}
269 
270 	sup_paging = ts->paging;
271 	sup_paging.cpl = 0;		/* implicit supervisor mode */
272 	error = desc_table_read(vcpu, &sup_paging, sel, desc, faultptr);
273 	return (error);
274 }
275 
276 static bool
277 code_desc(int sd_type)
278 {
279 	/* code descriptor */
280 	return ((sd_type & 0x18) == 0x18);
281 }
282 
283 static bool
284 stack_desc(int sd_type)
285 {
286 	/* writable data descriptor */
287 	return ((sd_type & 0x1A) == 0x12);
288 }
289 
290 static bool
291 data_desc(int sd_type)
292 {
293 	/* data descriptor or a readable code descriptor */
294 	return ((sd_type & 0x18) == 0x10 || (sd_type & 0x1A) == 0x1A);
295 }
296 
297 static bool
298 ldt_desc(int sd_type)
299 {
300 
301 	return (sd_type == SDT_SYSLDT);
302 }
303 
304 /*
305  * Validate the descriptor 'seg_desc' associated with 'segment'.
306  */
307 static int
308 validate_seg_desc(struct vcpu *vcpu, struct vm_task_switch *ts,
309     int segment, struct seg_desc *seg_desc, int *faultptr)
310 {
311 	struct vm_guest_paging sup_paging;
312 	struct user_segment_descriptor usd;
313 	int error, idtvec;
314 	int cpl, dpl, rpl;
315 	uint16_t sel, cs;
316 	bool ldtseg, codeseg, stackseg, dataseg, conforming;
317 
318 	ldtseg = codeseg = stackseg = dataseg = false;
319 	switch (segment) {
320 	case VM_REG_GUEST_LDTR:
321 		ldtseg = true;
322 		break;
323 	case VM_REG_GUEST_CS:
324 		codeseg = true;
325 		break;
326 	case VM_REG_GUEST_SS:
327 		stackseg = true;
328 		break;
329 	case VM_REG_GUEST_DS:
330 	case VM_REG_GUEST_ES:
331 	case VM_REG_GUEST_FS:
332 	case VM_REG_GUEST_GS:
333 		dataseg = true;
334 		break;
335 	default:
336 		assert(0);
337 	}
338 
339 	/* Get the segment selector */
340 	sel = GETREG(vcpu, segment);
341 
342 	/* LDT selector must point into the GDT */
343 	if (ldtseg && ISLDT(sel)) {
344 		sel_exception(vcpu, IDT_TS, sel, ts->ext);
345 		return (1);
346 	}
347 
348 	/* Descriptor table limit check */
349 	if (desc_table_limit_check(vcpu, sel)) {
350 		sel_exception(vcpu, IDT_TS, sel, ts->ext);
351 		return (1);
352 	}
353 
354 	/* NULL selector */
355 	if (IDXSEL(sel) == 0) {
356 		/* Code and stack segment selectors cannot be NULL */
357 		if (codeseg || stackseg) {
358 			sel_exception(vcpu, IDT_TS, sel, ts->ext);
359 			return (1);
360 		}
361 		seg_desc->base = 0;
362 		seg_desc->limit = 0;
363 		seg_desc->access = 0x10000;	/* unusable */
364 		return (0);
365 	}
366 
367 	/* Read the descriptor from the GDT/LDT */
368 	sup_paging = ts->paging;
369 	sup_paging.cpl = 0;	/* implicit supervisor mode */
370 	error = desc_table_read(vcpu, &sup_paging, sel, &usd, faultptr);
371 	if (error || *faultptr)
372 		return (error);
373 
374 	/* Verify that the descriptor type is compatible with the segment */
375 	if ((ldtseg && !ldt_desc(usd.sd_type)) ||
376 	    (codeseg && !code_desc(usd.sd_type)) ||
377 	    (dataseg && !data_desc(usd.sd_type)) ||
378 	    (stackseg && !stack_desc(usd.sd_type))) {
379 		sel_exception(vcpu, IDT_TS, sel, ts->ext);
380 		return (1);
381 	}
382 
383 	/* Segment must be marked present */
384 	if (!usd.sd_p) {
385 		if (ldtseg)
386 			idtvec = IDT_TS;
387 		else if (stackseg)
388 			idtvec = IDT_SS;
389 		else
390 			idtvec = IDT_NP;
391 		sel_exception(vcpu, idtvec, sel, ts->ext);
392 		return (1);
393 	}
394 
395 	cs = GETREG(vcpu, VM_REG_GUEST_CS);
396 	cpl = cs & SEL_RPL_MASK;
397 	rpl = sel & SEL_RPL_MASK;
398 	dpl = usd.sd_dpl;
399 
400 	if (stackseg && (rpl != cpl || dpl != cpl)) {
401 		sel_exception(vcpu, IDT_TS, sel, ts->ext);
402 		return (1);
403 	}
404 
405 	if (codeseg) {
406 		conforming = (usd.sd_type & 0x4) ? true : false;
407 		if ((conforming && (cpl < dpl)) ||
408 		    (!conforming && (cpl != dpl))) {
409 			sel_exception(vcpu, IDT_TS, sel, ts->ext);
410 			return (1);
411 		}
412 	}
413 
414 	if (dataseg) {
415 		/*
416 		 * A data segment is always non-conforming except when it's
417 		 * descriptor is a readable, conforming code segment.
418 		 */
419 		if (code_desc(usd.sd_type) && (usd.sd_type & 0x4) != 0)
420 			conforming = true;
421 		else
422 			conforming = false;
423 
424 		if (!conforming && (rpl > dpl || cpl > dpl)) {
425 			sel_exception(vcpu, IDT_TS, sel, ts->ext);
426 			return (1);
427 		}
428 	}
429 	*seg_desc = usd_to_seg_desc(&usd);
430 	return (0);
431 }
432 
433 static void
434 tss32_save(struct vcpu *vcpu, struct vm_task_switch *task_switch,
435     uint32_t eip, struct tss32 *tss, struct iovec *iov)
436 {
437 
438 	/* General purpose registers */
439 	tss->tss_eax = GETREG(vcpu, VM_REG_GUEST_RAX);
440 	tss->tss_ecx = GETREG(vcpu, VM_REG_GUEST_RCX);
441 	tss->tss_edx = GETREG(vcpu, VM_REG_GUEST_RDX);
442 	tss->tss_ebx = GETREG(vcpu, VM_REG_GUEST_RBX);
443 	tss->tss_esp = GETREG(vcpu, VM_REG_GUEST_RSP);
444 	tss->tss_ebp = GETREG(vcpu, VM_REG_GUEST_RBP);
445 	tss->tss_esi = GETREG(vcpu, VM_REG_GUEST_RSI);
446 	tss->tss_edi = GETREG(vcpu, VM_REG_GUEST_RDI);
447 
448 	/* Segment selectors */
449 	tss->tss_es = GETREG(vcpu, VM_REG_GUEST_ES);
450 	tss->tss_cs = GETREG(vcpu, VM_REG_GUEST_CS);
451 	tss->tss_ss = GETREG(vcpu, VM_REG_GUEST_SS);
452 	tss->tss_ds = GETREG(vcpu, VM_REG_GUEST_DS);
453 	tss->tss_fs = GETREG(vcpu, VM_REG_GUEST_FS);
454 	tss->tss_gs = GETREG(vcpu, VM_REG_GUEST_GS);
455 
456 	/* eflags and eip */
457 	tss->tss_eflags = GETREG(vcpu, VM_REG_GUEST_RFLAGS);
458 	if (task_switch->reason == TSR_IRET)
459 		tss->tss_eflags &= ~PSL_NT;
460 	tss->tss_eip = eip;
461 
462 	/* Copy updated old TSS into guest memory */
463 	vm_copyout(tss, iov, sizeof(struct tss32));
464 }
465 
466 static void
467 update_seg_desc(struct vcpu *vcpu, int reg, struct seg_desc *sd)
468 {
469 	int error;
470 
471 	error = vm_set_desc(vcpu, reg, sd->base, sd->limit, sd->access);
472 	assert(error == 0);
473 }
474 
475 /*
476  * Update the vcpu registers to reflect the state of the new task.
477  */
478 static int
479 tss32_restore(struct vmctx *ctx, struct vcpu *vcpu, struct vm_task_switch *ts,
480     uint16_t ot_sel, struct tss32 *tss, struct iovec *iov, int *faultptr)
481 {
482 	struct seg_desc seg_desc, seg_desc2;
483 	uint64_t *pdpte, maxphyaddr, reserved;
484 	uint32_t eflags;
485 	int error, i;
486 	bool nested;
487 
488 	nested = false;
489 	if (ts->reason != TSR_IRET && ts->reason != TSR_JMP) {
490 		tss->tss_link = ot_sel;
491 		nested = true;
492 	}
493 
494 	eflags = tss->tss_eflags;
495 	if (nested)
496 		eflags |= PSL_NT;
497 
498 	/* LDTR */
499 	SETREG(vcpu, VM_REG_GUEST_LDTR, tss->tss_ldt);
500 
501 	/* PBDR */
502 	if (ts->paging.paging_mode != PAGING_MODE_FLAT) {
503 		if (ts->paging.paging_mode == PAGING_MODE_PAE) {
504 			/*
505 			 * XXX Assuming 36-bit MAXPHYADDR.
506 			 */
507 			maxphyaddr = (1UL << 36) - 1;
508 			pdpte = paddr_guest2host(ctx, tss->tss_cr3 & ~0x1f, 32);
509 			for (i = 0; i < 4; i++) {
510 				/* Check reserved bits if the PDPTE is valid */
511 				if (!(pdpte[i] & 0x1))
512 					continue;
513 				/*
514 				 * Bits 2:1, 8:5 and bits above the processor's
515 				 * maximum physical address are reserved.
516 				 */
517 				reserved = ~maxphyaddr | 0x1E6;
518 				if (pdpte[i] & reserved) {
519 					vm_inject_gp(vcpu);
520 					return (1);
521 				}
522 			}
523 			SETREG(vcpu, VM_REG_GUEST_PDPTE0, pdpte[0]);
524 			SETREG(vcpu, VM_REG_GUEST_PDPTE1, pdpte[1]);
525 			SETREG(vcpu, VM_REG_GUEST_PDPTE2, pdpte[2]);
526 			SETREG(vcpu, VM_REG_GUEST_PDPTE3, pdpte[3]);
527 		}
528 		SETREG(vcpu, VM_REG_GUEST_CR3, tss->tss_cr3);
529 		ts->paging.cr3 = tss->tss_cr3;
530 	}
531 
532 	/* eflags and eip */
533 	SETREG(vcpu, VM_REG_GUEST_RFLAGS, eflags);
534 	SETREG(vcpu, VM_REG_GUEST_RIP, tss->tss_eip);
535 
536 	/* General purpose registers */
537 	SETREG(vcpu, VM_REG_GUEST_RAX, tss->tss_eax);
538 	SETREG(vcpu, VM_REG_GUEST_RCX, tss->tss_ecx);
539 	SETREG(vcpu, VM_REG_GUEST_RDX, tss->tss_edx);
540 	SETREG(vcpu, VM_REG_GUEST_RBX, tss->tss_ebx);
541 	SETREG(vcpu, VM_REG_GUEST_RSP, tss->tss_esp);
542 	SETREG(vcpu, VM_REG_GUEST_RBP, tss->tss_ebp);
543 	SETREG(vcpu, VM_REG_GUEST_RSI, tss->tss_esi);
544 	SETREG(vcpu, VM_REG_GUEST_RDI, tss->tss_edi);
545 
546 	/* Segment selectors */
547 	SETREG(vcpu, VM_REG_GUEST_ES, tss->tss_es);
548 	SETREG(vcpu, VM_REG_GUEST_CS, tss->tss_cs);
549 	SETREG(vcpu, VM_REG_GUEST_SS, tss->tss_ss);
550 	SETREG(vcpu, VM_REG_GUEST_DS, tss->tss_ds);
551 	SETREG(vcpu, VM_REG_GUEST_FS, tss->tss_fs);
552 	SETREG(vcpu, VM_REG_GUEST_GS, tss->tss_gs);
553 
554 	/*
555 	 * If this is a nested task then write out the new TSS to update
556 	 * the previous link field.
557 	 */
558 	if (nested)
559 		vm_copyout(tss, iov, sizeof(*tss));
560 
561 	/* Validate segment descriptors */
562 	error = validate_seg_desc(vcpu, ts, VM_REG_GUEST_LDTR, &seg_desc,
563 	    faultptr);
564 	if (error || *faultptr)
565 		return (error);
566 	update_seg_desc(vcpu, VM_REG_GUEST_LDTR, &seg_desc);
567 
568 	/*
569 	 * Section "Checks on Guest Segment Registers", Intel SDM, Vol 3.
570 	 *
571 	 * The SS and CS attribute checks on VM-entry are inter-dependent so
572 	 * we need to make sure that both segments are valid before updating
573 	 * either of them. This ensures that the VMCS state can pass the
574 	 * VM-entry checks so the guest can handle any exception injected
575 	 * during task switch emulation.
576 	 */
577 	error = validate_seg_desc(vcpu, ts, VM_REG_GUEST_CS, &seg_desc,
578 	    faultptr);
579 	if (error || *faultptr)
580 		return (error);
581 
582 	error = validate_seg_desc(vcpu, ts, VM_REG_GUEST_SS, &seg_desc2,
583 	    faultptr);
584 	if (error || *faultptr)
585 		return (error);
586 	update_seg_desc(vcpu, VM_REG_GUEST_CS, &seg_desc);
587 	update_seg_desc(vcpu, VM_REG_GUEST_SS, &seg_desc2);
588 	ts->paging.cpl = tss->tss_cs & SEL_RPL_MASK;
589 
590 	error = validate_seg_desc(vcpu, ts, VM_REG_GUEST_DS, &seg_desc,
591 	    faultptr);
592 	if (error || *faultptr)
593 		return (error);
594 	update_seg_desc(vcpu, VM_REG_GUEST_DS, &seg_desc);
595 
596 	error = validate_seg_desc(vcpu, ts, VM_REG_GUEST_ES, &seg_desc,
597 	    faultptr);
598 	if (error || *faultptr)
599 		return (error);
600 	update_seg_desc(vcpu, VM_REG_GUEST_ES, &seg_desc);
601 
602 	error = validate_seg_desc(vcpu, ts, VM_REG_GUEST_FS, &seg_desc,
603 	    faultptr);
604 	if (error || *faultptr)
605 		return (error);
606 	update_seg_desc(vcpu, VM_REG_GUEST_FS, &seg_desc);
607 
608 	error = validate_seg_desc(vcpu, ts, VM_REG_GUEST_GS, &seg_desc,
609 	    faultptr);
610 	if (error || *faultptr)
611 		return (error);
612 	update_seg_desc(vcpu, VM_REG_GUEST_GS, &seg_desc);
613 
614 	return (0);
615 }
616 
617 /*
618  * Push an error code on the stack of the new task. This is needed if the
619  * task switch was triggered by a hardware exception that causes an error
620  * code to be saved (e.g. #PF).
621  */
622 static int
623 push_errcode(struct vcpu *vcpu, struct vm_guest_paging *paging,
624     int task_type, uint32_t errcode, int *faultptr)
625 {
626 	struct iovec iov[2];
627 	struct seg_desc seg_desc;
628 	int stacksize, bytes, error;
629 	uint64_t gla, cr0, rflags;
630 	uint32_t esp;
631 	uint16_t stacksel;
632 
633 	*faultptr = 0;
634 
635 	cr0 = GETREG(vcpu, VM_REG_GUEST_CR0);
636 	rflags = GETREG(vcpu, VM_REG_GUEST_RFLAGS);
637 	stacksel = GETREG(vcpu, VM_REG_GUEST_SS);
638 
639 	error = vm_get_desc(vcpu, VM_REG_GUEST_SS, &seg_desc.base,
640 	    &seg_desc.limit, &seg_desc.access);
641 	assert(error == 0);
642 
643 	/*
644 	 * Section "Error Code" in the Intel SDM vol 3: the error code is
645 	 * pushed on the stack as a doubleword or word (depending on the
646 	 * default interrupt, trap or task gate size).
647 	 */
648 	if (task_type == SDT_SYS386BSY || task_type == SDT_SYS386TSS)
649 		bytes = 4;
650 	else
651 		bytes = 2;
652 
653 	/*
654 	 * PUSH instruction from Intel SDM vol 2: the 'B' flag in the
655 	 * stack-segment descriptor determines the size of the stack
656 	 * pointer outside of 64-bit mode.
657 	 */
658 	if (SEG_DESC_DEF32(seg_desc.access))
659 		stacksize = 4;
660 	else
661 		stacksize = 2;
662 
663 	esp = GETREG(vcpu, VM_REG_GUEST_RSP);
664 	esp -= bytes;
665 
666 	if (vie_calculate_gla(paging->cpu_mode, VM_REG_GUEST_SS,
667 	    &seg_desc, esp, bytes, stacksize, PROT_WRITE, &gla)) {
668 		sel_exception(vcpu, IDT_SS, stacksel, 1);
669 		*faultptr = 1;
670 		return (0);
671 	}
672 
673 	if (vie_alignment_check(paging->cpl, bytes, cr0, rflags, gla)) {
674 		vm_inject_ac(vcpu, 1);
675 		*faultptr = 1;
676 		return (0);
677 	}
678 
679 	error = vm_copy_setup(vcpu, paging, gla, bytes, PROT_WRITE,
680 	    iov, nitems(iov), faultptr);
681 	if (error || *faultptr)
682 		return (error);
683 
684 	vm_copyout(&errcode, iov, bytes);
685 	SETREG(vcpu, VM_REG_GUEST_RSP, esp);
686 	return (0);
687 }
688 
689 /*
690  * Evaluate return value from helper functions and potentially return to
691  * the VM run loop.
692  */
693 #define	CHKERR(error,fault)						\
694 	do {								\
695 		assert((error == 0) || (error == EFAULT));		\
696 		if (error)						\
697 			return (VMEXIT_ABORT);				\
698 		else if (fault)						\
699 			return (VMEXIT_CONTINUE);			\
700 	} while (0)
701 
702 int vmexit_task_switch(struct vmctx *, struct vcpu *, struct vm_run *);
703 
704 int
705 vmexit_task_switch(struct vmctx *ctx, struct vcpu *vcpu, struct vm_run *vmrun)
706 {
707 	struct seg_desc nt;
708 	struct tss32 oldtss, newtss;
709 	struct vm_task_switch *task_switch;
710 	struct vm_guest_paging *paging, sup_paging;
711 	struct user_segment_descriptor nt_desc, ot_desc;
712 	struct iovec nt_iov[2], ot_iov[2];
713 	struct vm_exit *vmexit;
714 	uint64_t cr0, ot_base;
715 	uint32_t eip, ot_lim, access;
716 	int error, ext, fault, minlimit, nt_type, ot_type;
717 	enum task_switch_reason reason;
718 	uint16_t nt_sel, ot_sel;
719 
720 	vmexit = vmrun->vm_exit;
721 	task_switch = &vmexit->u.task_switch;
722 	nt_sel = task_switch->tsssel;
723 	ext = vmexit->u.task_switch.ext;
724 	reason = vmexit->u.task_switch.reason;
725 	paging = &vmexit->u.task_switch.paging;
726 
727 	assert(paging->cpu_mode == CPU_MODE_PROTECTED);
728 
729 	/*
730 	 * Calculate the instruction pointer to store in the old TSS.
731 	 */
732 	eip = vmexit->rip + vmexit->inst_length;
733 
734 	/*
735 	 * Section 4.6, "Access Rights" in Intel SDM Vol 3.
736 	 * The following page table accesses are implicitly supervisor mode:
737 	 * - accesses to GDT or LDT to load segment descriptors
738 	 * - accesses to the task state segment during task switch
739 	 */
740 	sup_paging = *paging;
741 	sup_paging.cpl = 0;	/* implicit supervisor mode */
742 
743 	/* Fetch the new TSS descriptor */
744 	error = read_tss_descriptor(vcpu, task_switch, nt_sel, &nt_desc,
745 	    &fault);
746 	CHKERR(error, fault);
747 
748 	nt = usd_to_seg_desc(&nt_desc);
749 
750 	/* Verify the type of the new TSS */
751 	nt_type = SEG_DESC_TYPE(nt.access);
752 	if (nt_type != SDT_SYS386BSY && nt_type != SDT_SYS386TSS &&
753 	    nt_type != SDT_SYS286BSY && nt_type != SDT_SYS286TSS) {
754 		sel_exception(vcpu, IDT_TS, nt_sel, ext);
755 		goto done;
756 	}
757 
758 	/* TSS descriptor must have present bit set */
759 	if (!SEG_DESC_PRESENT(nt.access)) {
760 		sel_exception(vcpu, IDT_NP, nt_sel, ext);
761 		goto done;
762 	}
763 
764 	/*
765 	 * TSS must have a minimum length of 104 bytes for a 32-bit TSS and
766 	 * 44 bytes for a 16-bit TSS.
767 	 */
768 	if (nt_type == SDT_SYS386BSY || nt_type == SDT_SYS386TSS)
769 		minlimit = 104 - 1;
770 	else if (nt_type == SDT_SYS286BSY || nt_type == SDT_SYS286TSS)
771 		minlimit = 44 - 1;
772 	else
773 		minlimit = 0;
774 
775 	assert(minlimit > 0);
776 	if (nt.limit < (unsigned int)minlimit) {
777 		sel_exception(vcpu, IDT_TS, nt_sel, ext);
778 		goto done;
779 	}
780 
781 	/* TSS must be busy if task switch is due to IRET */
782 	if (reason == TSR_IRET && !TSS_BUSY(nt_type)) {
783 		sel_exception(vcpu, IDT_TS, nt_sel, ext);
784 		goto done;
785 	}
786 
787 	/*
788 	 * TSS must be available (not busy) if task switch reason is
789 	 * CALL, JMP, exception or interrupt.
790 	 */
791 	if (reason != TSR_IRET && TSS_BUSY(nt_type)) {
792 		sel_exception(vcpu, IDT_GP, nt_sel, ext);
793 		goto done;
794 	}
795 
796 	/* Fetch the new TSS */
797 	error = vm_copy_setup(vcpu, &sup_paging, nt.base, minlimit + 1,
798 	    PROT_READ | PROT_WRITE, nt_iov, nitems(nt_iov), &fault);
799 	CHKERR(error, fault);
800 	vm_copyin(nt_iov, &newtss, minlimit + 1);
801 
802 	/* Get the old TSS selector from the guest's task register */
803 	ot_sel = GETREG(vcpu, VM_REG_GUEST_TR);
804 	if (ISLDT(ot_sel) || IDXSEL(ot_sel) == 0) {
805 		/*
806 		 * This might happen if a task switch was attempted without
807 		 * ever loading the task register with LTR. In this case the
808 		 * TR would contain the values from power-on:
809 		 * (sel = 0, base = 0, limit = 0xffff).
810 		 */
811 		sel_exception(vcpu, IDT_TS, ot_sel, task_switch->ext);
812 		goto done;
813 	}
814 
815 	/* Get the old TSS base and limit from the guest's task register */
816 	error = vm_get_desc(vcpu, VM_REG_GUEST_TR, &ot_base, &ot_lim,
817 	    &access);
818 	assert(error == 0);
819 	assert(!SEG_DESC_UNUSABLE(access) && SEG_DESC_PRESENT(access));
820 	ot_type = SEG_DESC_TYPE(access);
821 	assert(ot_type == SDT_SYS386BSY || ot_type == SDT_SYS286BSY);
822 
823 	/* Fetch the old TSS descriptor */
824 	error = read_tss_descriptor(vcpu, task_switch, ot_sel, &ot_desc,
825 	    &fault);
826 	CHKERR(error, fault);
827 
828 	/* Get the old TSS */
829 	error = vm_copy_setup(vcpu, &sup_paging, ot_base, minlimit + 1,
830 	    PROT_READ | PROT_WRITE, ot_iov, nitems(ot_iov), &fault);
831 	CHKERR(error, fault);
832 	vm_copyin(ot_iov, &oldtss, minlimit + 1);
833 
834 	/*
835 	 * Clear the busy bit in the old TSS descriptor if the task switch
836 	 * due to an IRET or JMP instruction.
837 	 */
838 	if (reason == TSR_IRET || reason == TSR_JMP) {
839 		ot_desc.sd_type &= ~0x2;
840 		error = desc_table_write(vcpu, &sup_paging, ot_sel,
841 		    &ot_desc, &fault);
842 		CHKERR(error, fault);
843 	}
844 
845 	if (nt_type == SDT_SYS286BSY || nt_type == SDT_SYS286TSS) {
846 		EPRINTLN("Task switch to 16-bit TSS not supported");
847 		return (VMEXIT_ABORT);
848 	}
849 
850 	/* Save processor state in old TSS */
851 	tss32_save(vcpu, task_switch, eip, &oldtss, ot_iov);
852 
853 	/*
854 	 * If the task switch was triggered for any reason other than IRET
855 	 * then set the busy bit in the new TSS descriptor.
856 	 */
857 	if (reason != TSR_IRET) {
858 		nt_desc.sd_type |= 0x2;
859 		error = desc_table_write(vcpu, &sup_paging, nt_sel,
860 		    &nt_desc, &fault);
861 		CHKERR(error, fault);
862 	}
863 
864 	/* Update task register to point at the new TSS */
865 	SETREG(vcpu, VM_REG_GUEST_TR, nt_sel);
866 
867 	/* Update the hidden descriptor state of the task register */
868 	nt = usd_to_seg_desc(&nt_desc);
869 	update_seg_desc(vcpu, VM_REG_GUEST_TR, &nt);
870 
871 	/* Set CR0.TS */
872 	cr0 = GETREG(vcpu, VM_REG_GUEST_CR0);
873 	SETREG(vcpu, VM_REG_GUEST_CR0, cr0 | CR0_TS);
874 
875 	/*
876 	 * We are now committed to the task switch. Any exceptions encountered
877 	 * after this point will be handled in the context of the new task and
878 	 * the saved instruction pointer will belong to the new task.
879 	 */
880 	error = vm_set_register(vcpu, VM_REG_GUEST_RIP, newtss.tss_eip);
881 	assert(error == 0);
882 
883 	/* Load processor state from new TSS */
884 	error = tss32_restore(ctx, vcpu, task_switch, ot_sel, &newtss, nt_iov,
885 	    &fault);
886 	CHKERR(error, fault);
887 
888 	/*
889 	 * Section "Interrupt Tasks" in Intel SDM, Vol 3: if an exception
890 	 * caused an error code to be generated, this error code is copied
891 	 * to the stack of the new task.
892 	 */
893 	if (task_switch->errcode_valid) {
894 		assert(task_switch->ext);
895 		assert(task_switch->reason == TSR_IDT_GATE);
896 		error = push_errcode(vcpu, &task_switch->paging, nt_type,
897 		    task_switch->errcode, &fault);
898 		CHKERR(error, fault);
899 	}
900 
901 	/*
902 	 * Treatment of virtual-NMI blocking if NMI is delivered through
903 	 * a task gate.
904 	 *
905 	 * Section "Architectural State Before A VM Exit", Intel SDM, Vol3:
906 	 * If the virtual NMIs VM-execution control is 1, VM entry injects
907 	 * an NMI, and delivery of the NMI causes a task switch that causes
908 	 * a VM exit, virtual-NMI blocking is in effect before the VM exit
909 	 * commences.
910 	 *
911 	 * Thus, virtual-NMI blocking is in effect at the time of the task
912 	 * switch VM exit.
913 	 */
914 
915 	/*
916 	 * Treatment of virtual-NMI unblocking on IRET from NMI handler task.
917 	 *
918 	 * Section "Changes to Instruction Behavior in VMX Non-Root Operation"
919 	 * If "virtual NMIs" control is 1 IRET removes any virtual-NMI blocking.
920 	 * This unblocking of virtual-NMI occurs even if IRET causes a fault.
921 	 *
922 	 * Thus, virtual-NMI blocking is cleared at the time of the task switch
923 	 * VM exit.
924 	 */
925 
926 	/*
927 	 * If the task switch was triggered by an event delivered through
928 	 * the IDT then extinguish the pending event from the vcpu's
929 	 * exitintinfo.
930 	 */
931 	if (task_switch->reason == TSR_IDT_GATE) {
932 		error = vm_set_intinfo(vcpu, 0);
933 		assert(error == 0);
934 	}
935 
936 	/*
937 	 * XXX should inject debug exception if 'T' bit is 1
938 	 */
939 done:
940 	return (VMEXIT_CONTINUE);
941 }
942