1 // SPDX-License-Identifier: GPL-2.0
2 /* Copyright (C) 2021-2022 Intel Corporation */
3 
4 #undef pr_fmt
5 #define pr_fmt(fmt)     "tdx: " fmt
6 
7 #include <linux/cpufeature.h>
8 #include <asm/coco.h>
9 #include <asm/tdx.h>
10 #include <asm/vmx.h>
11 #include <asm/insn.h>
12 #include <asm/insn-eval.h>
13 #include <asm/pgtable.h>
14 
15 /* TDX module Call Leaf IDs */
16 #define TDX_GET_INFO			1
17 #define TDX_GET_VEINFO			3
18 #define TDX_ACCEPT_PAGE			6
19 
20 /* TDX hypercall Leaf IDs */
21 #define TDVMCALL_MAP_GPA		0x10001
22 
23 /* MMIO direction */
24 #define EPT_READ	0
25 #define EPT_WRITE	1
26 
27 /* Port I/O direction */
28 #define PORT_READ	0
29 #define PORT_WRITE	1
30 
31 /* See Exit Qualification for I/O Instructions in VMX documentation */
32 #define VE_IS_IO_IN(e)		((e) & BIT(3))
33 #define VE_GET_IO_SIZE(e)	(((e) & GENMASK(2, 0)) + 1)
34 #define VE_GET_PORT_NUM(e)	((e) >> 16)
35 #define VE_IS_IO_STRING(e)	((e) & BIT(4))
36 
37 #define ATTR_SEPT_VE_DISABLE	BIT(28)
38 
39 /*
40  * Wrapper for standard use of __tdx_hypercall with no output aside from
41  * return code.
42  */
_tdx_hypercall(u64 fn,u64 r12,u64 r13,u64 r14,u64 r15)43 static inline u64 _tdx_hypercall(u64 fn, u64 r12, u64 r13, u64 r14, u64 r15)
44 {
45 	struct tdx_hypercall_args args = {
46 		.r10 = TDX_HYPERCALL_STANDARD,
47 		.r11 = fn,
48 		.r12 = r12,
49 		.r13 = r13,
50 		.r14 = r14,
51 		.r15 = r15,
52 	};
53 
54 	return __tdx_hypercall(&args, 0);
55 }
56 
57 /* Called from __tdx_hypercall() for unrecoverable failure */
__tdx_hypercall_failed(void)58 void __tdx_hypercall_failed(void)
59 {
60 	panic("TDVMCALL failed. TDX module bug?");
61 }
62 
63 /*
64  * The TDG.VP.VMCALL-Instruction-execution sub-functions are defined
65  * independently from but are currently matched 1:1 with VMX EXIT_REASONs.
66  * Reusing the KVM EXIT_REASON macros makes it easier to connect the host and
67  * guest sides of these calls.
68  */
hcall_func(u64 exit_reason)69 static u64 hcall_func(u64 exit_reason)
70 {
71 	return exit_reason;
72 }
73 
74 #ifdef CONFIG_KVM_GUEST
tdx_kvm_hypercall(unsigned int nr,unsigned long p1,unsigned long p2,unsigned long p3,unsigned long p4)75 long tdx_kvm_hypercall(unsigned int nr, unsigned long p1, unsigned long p2,
76 		       unsigned long p3, unsigned long p4)
77 {
78 	struct tdx_hypercall_args args = {
79 		.r10 = nr,
80 		.r11 = p1,
81 		.r12 = p2,
82 		.r13 = p3,
83 		.r14 = p4,
84 	};
85 
86 	return __tdx_hypercall(&args, 0);
87 }
88 EXPORT_SYMBOL_GPL(tdx_kvm_hypercall);
89 #endif
90 
91 /*
92  * Used for TDX guests to make calls directly to the TD module.  This
93  * should only be used for calls that have no legitimate reason to fail
94  * or where the kernel can not survive the call failing.
95  */
tdx_module_call(u64 fn,u64 rcx,u64 rdx,u64 r8,u64 r9,struct tdx_module_output * out)96 static inline void tdx_module_call(u64 fn, u64 rcx, u64 rdx, u64 r8, u64 r9,
97 				   struct tdx_module_output *out)
98 {
99 	if (__tdx_module_call(fn, rcx, rdx, r8, r9, out))
100 		panic("TDCALL %lld failed (Buggy TDX module!)\n", fn);
101 }
102 
tdx_parse_tdinfo(u64 * cc_mask)103 static void tdx_parse_tdinfo(u64 *cc_mask)
104 {
105 	struct tdx_module_output out;
106 	unsigned int gpa_width;
107 	u64 td_attr;
108 
109 	/*
110 	 * TDINFO TDX module call is used to get the TD execution environment
111 	 * information like GPA width, number of available vcpus, debug mode
112 	 * information, etc. More details about the ABI can be found in TDX
113 	 * Guest-Host-Communication Interface (GHCI), section 2.4.2 TDCALL
114 	 * [TDG.VP.INFO].
115 	 */
116 	tdx_module_call(TDX_GET_INFO, 0, 0, 0, 0, &out);
117 
118 	/*
119 	 * The highest bit of a guest physical address is the "sharing" bit.
120 	 * Set it for shared pages and clear it for private pages.
121 	 *
122 	 * The GPA width that comes out of this call is critical. TDX guests
123 	 * can not meaningfully run without it.
124 	 */
125 	gpa_width = out.rcx & GENMASK(5, 0);
126 	*cc_mask = BIT_ULL(gpa_width - 1);
127 
128 	/*
129 	 * The kernel can not handle #VE's when accessing normal kernel
130 	 * memory.  Ensure that no #VE will be delivered for accesses to
131 	 * TD-private memory.  Only VMM-shared memory (MMIO) will #VE.
132 	 */
133 	td_attr = out.rdx;
134 	if (!(td_attr & ATTR_SEPT_VE_DISABLE))
135 		panic("TD misconfiguration: SEPT_VE_DISABLE attibute must be set.\n");
136 }
137 
138 /*
139  * The TDX module spec states that #VE may be injected for a limited set of
140  * reasons:
141  *
142  *  - Emulation of the architectural #VE injection on EPT violation;
143  *
144  *  - As a result of guest TD execution of a disallowed instruction,
145  *    a disallowed MSR access, or CPUID virtualization;
146  *
147  *  - A notification to the guest TD about anomalous behavior;
148  *
149  * The last one is opt-in and is not used by the kernel.
150  *
151  * The Intel Software Developer's Manual describes cases when instruction
152  * length field can be used in section "Information for VM Exits Due to
153  * Instruction Execution".
154  *
155  * For TDX, it ultimately means GET_VEINFO provides reliable instruction length
156  * information if #VE occurred due to instruction execution, but not for EPT
157  * violations.
158  */
ve_instr_len(struct ve_info * ve)159 static int ve_instr_len(struct ve_info *ve)
160 {
161 	switch (ve->exit_reason) {
162 	case EXIT_REASON_HLT:
163 	case EXIT_REASON_MSR_READ:
164 	case EXIT_REASON_MSR_WRITE:
165 	case EXIT_REASON_CPUID:
166 	case EXIT_REASON_IO_INSTRUCTION:
167 		/* It is safe to use ve->instr_len for #VE due instructions */
168 		return ve->instr_len;
169 	case EXIT_REASON_EPT_VIOLATION:
170 		/*
171 		 * For EPT violations, ve->insn_len is not defined. For those,
172 		 * the kernel must decode instructions manually and should not
173 		 * be using this function.
174 		 */
175 		WARN_ONCE(1, "ve->instr_len is not defined for EPT violations");
176 		return 0;
177 	default:
178 		WARN_ONCE(1, "Unexpected #VE-type: %lld\n", ve->exit_reason);
179 		return ve->instr_len;
180 	}
181 }
182 
__halt(const bool irq_disabled,const bool do_sti)183 static u64 __cpuidle __halt(const bool irq_disabled, const bool do_sti)
184 {
185 	struct tdx_hypercall_args args = {
186 		.r10 = TDX_HYPERCALL_STANDARD,
187 		.r11 = hcall_func(EXIT_REASON_HLT),
188 		.r12 = irq_disabled,
189 	};
190 
191 	/*
192 	 * Emulate HLT operation via hypercall. More info about ABI
193 	 * can be found in TDX Guest-Host-Communication Interface
194 	 * (GHCI), section 3.8 TDG.VP.VMCALL<Instruction.HLT>.
195 	 *
196 	 * The VMM uses the "IRQ disabled" param to understand IRQ
197 	 * enabled status (RFLAGS.IF) of the TD guest and to determine
198 	 * whether or not it should schedule the halted vCPU if an
199 	 * IRQ becomes pending. E.g. if IRQs are disabled, the VMM
200 	 * can keep the vCPU in virtual HLT, even if an IRQ is
201 	 * pending, without hanging/breaking the guest.
202 	 */
203 	return __tdx_hypercall(&args, do_sti ? TDX_HCALL_ISSUE_STI : 0);
204 }
205 
handle_halt(struct ve_info * ve)206 static int handle_halt(struct ve_info *ve)
207 {
208 	/*
209 	 * Since non safe halt is mainly used in CPU offlining
210 	 * and the guest will always stay in the halt state, don't
211 	 * call the STI instruction (set do_sti as false).
212 	 */
213 	const bool irq_disabled = irqs_disabled();
214 	const bool do_sti = false;
215 
216 	if (__halt(irq_disabled, do_sti))
217 		return -EIO;
218 
219 	return ve_instr_len(ve);
220 }
221 
tdx_safe_halt(void)222 void __cpuidle tdx_safe_halt(void)
223 {
224 	 /*
225 	  * For do_sti=true case, __tdx_hypercall() function enables
226 	  * interrupts using the STI instruction before the TDCALL. So
227 	  * set irq_disabled as false.
228 	  */
229 	const bool irq_disabled = false;
230 	const bool do_sti = true;
231 
232 	/*
233 	 * Use WARN_ONCE() to report the failure.
234 	 */
235 	if (__halt(irq_disabled, do_sti))
236 		WARN_ONCE(1, "HLT instruction emulation failed\n");
237 }
238 
read_msr(struct pt_regs * regs,struct ve_info * ve)239 static int read_msr(struct pt_regs *regs, struct ve_info *ve)
240 {
241 	struct tdx_hypercall_args args = {
242 		.r10 = TDX_HYPERCALL_STANDARD,
243 		.r11 = hcall_func(EXIT_REASON_MSR_READ),
244 		.r12 = regs->cx,
245 	};
246 
247 	/*
248 	 * Emulate the MSR read via hypercall. More info about ABI
249 	 * can be found in TDX Guest-Host-Communication Interface
250 	 * (GHCI), section titled "TDG.VP.VMCALL<Instruction.RDMSR>".
251 	 */
252 	if (__tdx_hypercall(&args, TDX_HCALL_HAS_OUTPUT))
253 		return -EIO;
254 
255 	regs->ax = lower_32_bits(args.r11);
256 	regs->dx = upper_32_bits(args.r11);
257 	return ve_instr_len(ve);
258 }
259 
write_msr(struct pt_regs * regs,struct ve_info * ve)260 static int write_msr(struct pt_regs *regs, struct ve_info *ve)
261 {
262 	struct tdx_hypercall_args args = {
263 		.r10 = TDX_HYPERCALL_STANDARD,
264 		.r11 = hcall_func(EXIT_REASON_MSR_WRITE),
265 		.r12 = regs->cx,
266 		.r13 = (u64)regs->dx << 32 | regs->ax,
267 	};
268 
269 	/*
270 	 * Emulate the MSR write via hypercall. More info about ABI
271 	 * can be found in TDX Guest-Host-Communication Interface
272 	 * (GHCI) section titled "TDG.VP.VMCALL<Instruction.WRMSR>".
273 	 */
274 	if (__tdx_hypercall(&args, 0))
275 		return -EIO;
276 
277 	return ve_instr_len(ve);
278 }
279 
handle_cpuid(struct pt_regs * regs,struct ve_info * ve)280 static int handle_cpuid(struct pt_regs *regs, struct ve_info *ve)
281 {
282 	struct tdx_hypercall_args args = {
283 		.r10 = TDX_HYPERCALL_STANDARD,
284 		.r11 = hcall_func(EXIT_REASON_CPUID),
285 		.r12 = regs->ax,
286 		.r13 = regs->cx,
287 	};
288 
289 	/*
290 	 * Only allow VMM to control range reserved for hypervisor
291 	 * communication.
292 	 *
293 	 * Return all-zeros for any CPUID outside the range. It matches CPU
294 	 * behaviour for non-supported leaf.
295 	 */
296 	if (regs->ax < 0x40000000 || regs->ax > 0x4FFFFFFF) {
297 		regs->ax = regs->bx = regs->cx = regs->dx = 0;
298 		return ve_instr_len(ve);
299 	}
300 
301 	/*
302 	 * Emulate the CPUID instruction via a hypercall. More info about
303 	 * ABI can be found in TDX Guest-Host-Communication Interface
304 	 * (GHCI), section titled "VP.VMCALL<Instruction.CPUID>".
305 	 */
306 	if (__tdx_hypercall(&args, TDX_HCALL_HAS_OUTPUT))
307 		return -EIO;
308 
309 	/*
310 	 * As per TDX GHCI CPUID ABI, r12-r15 registers contain contents of
311 	 * EAX, EBX, ECX, EDX registers after the CPUID instruction execution.
312 	 * So copy the register contents back to pt_regs.
313 	 */
314 	regs->ax = args.r12;
315 	regs->bx = args.r13;
316 	regs->cx = args.r14;
317 	regs->dx = args.r15;
318 
319 	return ve_instr_len(ve);
320 }
321 
mmio_read(int size,unsigned long addr,unsigned long * val)322 static bool mmio_read(int size, unsigned long addr, unsigned long *val)
323 {
324 	struct tdx_hypercall_args args = {
325 		.r10 = TDX_HYPERCALL_STANDARD,
326 		.r11 = hcall_func(EXIT_REASON_EPT_VIOLATION),
327 		.r12 = size,
328 		.r13 = EPT_READ,
329 		.r14 = addr,
330 		.r15 = *val,
331 	};
332 
333 	if (__tdx_hypercall(&args, TDX_HCALL_HAS_OUTPUT))
334 		return false;
335 	*val = args.r11;
336 	return true;
337 }
338 
mmio_write(int size,unsigned long addr,unsigned long val)339 static bool mmio_write(int size, unsigned long addr, unsigned long val)
340 {
341 	return !_tdx_hypercall(hcall_func(EXIT_REASON_EPT_VIOLATION), size,
342 			       EPT_WRITE, addr, val);
343 }
344 
handle_mmio(struct pt_regs * regs,struct ve_info * ve)345 static int handle_mmio(struct pt_regs *regs, struct ve_info *ve)
346 {
347 	unsigned long *reg, val, vaddr;
348 	char buffer[MAX_INSN_SIZE];
349 	struct insn insn = {};
350 	enum mmio_type mmio;
351 	int size, extend_size;
352 	u8 extend_val = 0;
353 
354 	/* Only in-kernel MMIO is supported */
355 	if (WARN_ON_ONCE(user_mode(regs)))
356 		return -EFAULT;
357 
358 	if (copy_from_kernel_nofault(buffer, (void *)regs->ip, MAX_INSN_SIZE))
359 		return -EFAULT;
360 
361 	if (insn_decode(&insn, buffer, MAX_INSN_SIZE, INSN_MODE_64))
362 		return -EINVAL;
363 
364 	mmio = insn_decode_mmio(&insn, &size);
365 	if (WARN_ON_ONCE(mmio == MMIO_DECODE_FAILED))
366 		return -EINVAL;
367 
368 	if (mmio != MMIO_WRITE_IMM && mmio != MMIO_MOVS) {
369 		reg = insn_get_modrm_reg_ptr(&insn, regs);
370 		if (!reg)
371 			return -EINVAL;
372 	}
373 
374 	/*
375 	 * Reject EPT violation #VEs that split pages.
376 	 *
377 	 * MMIO accesses are supposed to be naturally aligned and therefore
378 	 * never cross page boundaries. Seeing split page accesses indicates
379 	 * a bug or a load_unaligned_zeropad() that stepped into an MMIO page.
380 	 *
381 	 * load_unaligned_zeropad() will recover using exception fixups.
382 	 */
383 	vaddr = (unsigned long)insn_get_addr_ref(&insn, regs);
384 	if (vaddr / PAGE_SIZE != (vaddr + size - 1) / PAGE_SIZE)
385 		return -EFAULT;
386 
387 	/* Handle writes first */
388 	switch (mmio) {
389 	case MMIO_WRITE:
390 		memcpy(&val, reg, size);
391 		if (!mmio_write(size, ve->gpa, val))
392 			return -EIO;
393 		return insn.length;
394 	case MMIO_WRITE_IMM:
395 		val = insn.immediate.value;
396 		if (!mmio_write(size, ve->gpa, val))
397 			return -EIO;
398 		return insn.length;
399 	case MMIO_READ:
400 	case MMIO_READ_ZERO_EXTEND:
401 	case MMIO_READ_SIGN_EXTEND:
402 		/* Reads are handled below */
403 		break;
404 	case MMIO_MOVS:
405 	case MMIO_DECODE_FAILED:
406 		/*
407 		 * MMIO was accessed with an instruction that could not be
408 		 * decoded or handled properly. It was likely not using io.h
409 		 * helpers or accessed MMIO accidentally.
410 		 */
411 		return -EINVAL;
412 	default:
413 		WARN_ONCE(1, "Unknown insn_decode_mmio() decode value?");
414 		return -EINVAL;
415 	}
416 
417 	/* Handle reads */
418 	if (!mmio_read(size, ve->gpa, &val))
419 		return -EIO;
420 
421 	switch (mmio) {
422 	case MMIO_READ:
423 		/* Zero-extend for 32-bit operation */
424 		extend_size = size == 4 ? sizeof(*reg) : 0;
425 		break;
426 	case MMIO_READ_ZERO_EXTEND:
427 		/* Zero extend based on operand size */
428 		extend_size = insn.opnd_bytes;
429 		break;
430 	case MMIO_READ_SIGN_EXTEND:
431 		/* Sign extend based on operand size */
432 		extend_size = insn.opnd_bytes;
433 		if (size == 1 && val & BIT(7))
434 			extend_val = 0xFF;
435 		else if (size > 1 && val & BIT(15))
436 			extend_val = 0xFF;
437 		break;
438 	default:
439 		/* All other cases has to be covered with the first switch() */
440 		WARN_ON_ONCE(1);
441 		return -EINVAL;
442 	}
443 
444 	if (extend_size)
445 		memset(reg, extend_val, extend_size);
446 	memcpy(reg, &val, size);
447 	return insn.length;
448 }
449 
handle_in(struct pt_regs * regs,int size,int port)450 static bool handle_in(struct pt_regs *regs, int size, int port)
451 {
452 	struct tdx_hypercall_args args = {
453 		.r10 = TDX_HYPERCALL_STANDARD,
454 		.r11 = hcall_func(EXIT_REASON_IO_INSTRUCTION),
455 		.r12 = size,
456 		.r13 = PORT_READ,
457 		.r14 = port,
458 	};
459 	u64 mask = GENMASK(BITS_PER_BYTE * size, 0);
460 	bool success;
461 
462 	/*
463 	 * Emulate the I/O read via hypercall. More info about ABI can be found
464 	 * in TDX Guest-Host-Communication Interface (GHCI) section titled
465 	 * "TDG.VP.VMCALL<Instruction.IO>".
466 	 */
467 	success = !__tdx_hypercall(&args, TDX_HCALL_HAS_OUTPUT);
468 
469 	/* Update part of the register affected by the emulated instruction */
470 	regs->ax &= ~mask;
471 	if (success)
472 		regs->ax |= args.r11 & mask;
473 
474 	return success;
475 }
476 
handle_out(struct pt_regs * regs,int size,int port)477 static bool handle_out(struct pt_regs *regs, int size, int port)
478 {
479 	u64 mask = GENMASK(BITS_PER_BYTE * size, 0);
480 
481 	/*
482 	 * Emulate the I/O write via hypercall. More info about ABI can be found
483 	 * in TDX Guest-Host-Communication Interface (GHCI) section titled
484 	 * "TDG.VP.VMCALL<Instruction.IO>".
485 	 */
486 	return !_tdx_hypercall(hcall_func(EXIT_REASON_IO_INSTRUCTION), size,
487 			       PORT_WRITE, port, regs->ax & mask);
488 }
489 
490 /*
491  * Emulate I/O using hypercall.
492  *
493  * Assumes the IO instruction was using ax, which is enforced
494  * by the standard io.h macros.
495  *
496  * Return True on success or False on failure.
497  */
handle_io(struct pt_regs * regs,struct ve_info * ve)498 static int handle_io(struct pt_regs *regs, struct ve_info *ve)
499 {
500 	u32 exit_qual = ve->exit_qual;
501 	int size, port;
502 	bool in, ret;
503 
504 	if (VE_IS_IO_STRING(exit_qual))
505 		return -EIO;
506 
507 	in   = VE_IS_IO_IN(exit_qual);
508 	size = VE_GET_IO_SIZE(exit_qual);
509 	port = VE_GET_PORT_NUM(exit_qual);
510 
511 
512 	if (in)
513 		ret = handle_in(regs, size, port);
514 	else
515 		ret = handle_out(regs, size, port);
516 	if (!ret)
517 		return -EIO;
518 
519 	return ve_instr_len(ve);
520 }
521 
522 /*
523  * Early #VE exception handler. Only handles a subset of port I/O.
524  * Intended only for earlyprintk. If failed, return false.
525  */
tdx_early_handle_ve(struct pt_regs * regs)526 __init bool tdx_early_handle_ve(struct pt_regs *regs)
527 {
528 	struct ve_info ve;
529 	int insn_len;
530 
531 	tdx_get_ve_info(&ve);
532 
533 	if (ve.exit_reason != EXIT_REASON_IO_INSTRUCTION)
534 		return false;
535 
536 	insn_len = handle_io(regs, &ve);
537 	if (insn_len < 0)
538 		return false;
539 
540 	regs->ip += insn_len;
541 	return true;
542 }
543 
tdx_get_ve_info(struct ve_info * ve)544 void tdx_get_ve_info(struct ve_info *ve)
545 {
546 	struct tdx_module_output out;
547 
548 	/*
549 	 * Called during #VE handling to retrieve the #VE info from the
550 	 * TDX module.
551 	 *
552 	 * This has to be called early in #VE handling.  A "nested" #VE which
553 	 * occurs before this will raise a #DF and is not recoverable.
554 	 *
555 	 * The call retrieves the #VE info from the TDX module, which also
556 	 * clears the "#VE valid" flag. This must be done before anything else
557 	 * because any #VE that occurs while the valid flag is set will lead to
558 	 * #DF.
559 	 *
560 	 * Note, the TDX module treats virtual NMIs as inhibited if the #VE
561 	 * valid flag is set. It means that NMI=>#VE will not result in a #DF.
562 	 */
563 	tdx_module_call(TDX_GET_VEINFO, 0, 0, 0, 0, &out);
564 
565 	/* Transfer the output parameters */
566 	ve->exit_reason = out.rcx;
567 	ve->exit_qual   = out.rdx;
568 	ve->gla         = out.r8;
569 	ve->gpa         = out.r9;
570 	ve->instr_len   = lower_32_bits(out.r10);
571 	ve->instr_info  = upper_32_bits(out.r10);
572 }
573 
574 /*
575  * Handle the user initiated #VE.
576  *
577  * On success, returns the number of bytes RIP should be incremented (>=0)
578  * or -errno on error.
579  */
virt_exception_user(struct pt_regs * regs,struct ve_info * ve)580 static int virt_exception_user(struct pt_regs *regs, struct ve_info *ve)
581 {
582 	switch (ve->exit_reason) {
583 	case EXIT_REASON_CPUID:
584 		return handle_cpuid(regs, ve);
585 	default:
586 		pr_warn("Unexpected #VE: %lld\n", ve->exit_reason);
587 		return -EIO;
588 	}
589 }
590 
591 /*
592  * Handle the kernel #VE.
593  *
594  * On success, returns the number of bytes RIP should be incremented (>=0)
595  * or -errno on error.
596  */
virt_exception_kernel(struct pt_regs * regs,struct ve_info * ve)597 static int virt_exception_kernel(struct pt_regs *regs, struct ve_info *ve)
598 {
599 	switch (ve->exit_reason) {
600 	case EXIT_REASON_HLT:
601 		return handle_halt(ve);
602 	case EXIT_REASON_MSR_READ:
603 		return read_msr(regs, ve);
604 	case EXIT_REASON_MSR_WRITE:
605 		return write_msr(regs, ve);
606 	case EXIT_REASON_CPUID:
607 		return handle_cpuid(regs, ve);
608 	case EXIT_REASON_EPT_VIOLATION:
609 		return handle_mmio(regs, ve);
610 	case EXIT_REASON_IO_INSTRUCTION:
611 		return handle_io(regs, ve);
612 	default:
613 		pr_warn("Unexpected #VE: %lld\n", ve->exit_reason);
614 		return -EIO;
615 	}
616 }
617 
tdx_handle_virt_exception(struct pt_regs * regs,struct ve_info * ve)618 bool tdx_handle_virt_exception(struct pt_regs *regs, struct ve_info *ve)
619 {
620 	int insn_len;
621 
622 	if (user_mode(regs))
623 		insn_len = virt_exception_user(regs, ve);
624 	else
625 		insn_len = virt_exception_kernel(regs, ve);
626 	if (insn_len < 0)
627 		return false;
628 
629 	/* After successful #VE handling, move the IP */
630 	regs->ip += insn_len;
631 
632 	return true;
633 }
634 
tdx_tlb_flush_required(bool private)635 static bool tdx_tlb_flush_required(bool private)
636 {
637 	/*
638 	 * TDX guest is responsible for flushing TLB on private->shared
639 	 * transition. VMM is responsible for flushing on shared->private.
640 	 *
641 	 * The VMM _can't_ flush private addresses as it can't generate PAs
642 	 * with the guest's HKID.  Shared memory isn't subject to integrity
643 	 * checking, i.e. the VMM doesn't need to flush for its own protection.
644 	 *
645 	 * There's no need to flush when converting from shared to private,
646 	 * as flushing is the VMM's responsibility in this case, e.g. it must
647 	 * flush to avoid integrity failures in the face of a buggy or
648 	 * malicious guest.
649 	 */
650 	return !private;
651 }
652 
tdx_cache_flush_required(void)653 static bool tdx_cache_flush_required(void)
654 {
655 	/*
656 	 * AMD SME/SEV can avoid cache flushing if HW enforces cache coherence.
657 	 * TDX doesn't have such capability.
658 	 *
659 	 * Flush cache unconditionally.
660 	 */
661 	return true;
662 }
663 
try_accept_one(phys_addr_t * start,unsigned long len,enum pg_level pg_level)664 static bool try_accept_one(phys_addr_t *start, unsigned long len,
665 			  enum pg_level pg_level)
666 {
667 	unsigned long accept_size = page_level_size(pg_level);
668 	u64 tdcall_rcx;
669 	u8 page_size;
670 
671 	if (!IS_ALIGNED(*start, accept_size))
672 		return false;
673 
674 	if (len < accept_size)
675 		return false;
676 
677 	/*
678 	 * Pass the page physical address to the TDX module to accept the
679 	 * pending, private page.
680 	 *
681 	 * Bits 2:0 of RCX encode page size: 0 - 4K, 1 - 2M, 2 - 1G.
682 	 */
683 	switch (pg_level) {
684 	case PG_LEVEL_4K:
685 		page_size = 0;
686 		break;
687 	case PG_LEVEL_2M:
688 		page_size = 1;
689 		break;
690 	case PG_LEVEL_1G:
691 		page_size = 2;
692 		break;
693 	default:
694 		return false;
695 	}
696 
697 	tdcall_rcx = *start | page_size;
698 	if (__tdx_module_call(TDX_ACCEPT_PAGE, tdcall_rcx, 0, 0, 0, NULL))
699 		return false;
700 
701 	*start += accept_size;
702 	return true;
703 }
704 
705 /*
706  * Inform the VMM of the guest's intent for this physical page: shared with
707  * the VMM or private to the guest.  The VMM is expected to change its mapping
708  * of the page in response.
709  */
tdx_enc_status_changed(unsigned long vaddr,int numpages,bool enc)710 static bool tdx_enc_status_changed(unsigned long vaddr, int numpages, bool enc)
711 {
712 	phys_addr_t start = __pa(vaddr);
713 	phys_addr_t end   = __pa(vaddr + numpages * PAGE_SIZE);
714 
715 	if (!enc) {
716 		/* Set the shared (decrypted) bits: */
717 		start |= cc_mkdec(0);
718 		end   |= cc_mkdec(0);
719 	}
720 
721 	/*
722 	 * Notify the VMM about page mapping conversion. More info about ABI
723 	 * can be found in TDX Guest-Host-Communication Interface (GHCI),
724 	 * section "TDG.VP.VMCALL<MapGPA>"
725 	 */
726 	if (_tdx_hypercall(TDVMCALL_MAP_GPA, start, end - start, 0, 0))
727 		return false;
728 
729 	/* private->shared conversion  requires only MapGPA call */
730 	if (!enc)
731 		return true;
732 
733 	/*
734 	 * For shared->private conversion, accept the page using
735 	 * TDX_ACCEPT_PAGE TDX module call.
736 	 */
737 	while (start < end) {
738 		unsigned long len = end - start;
739 
740 		/*
741 		 * Try larger accepts first. It gives chance to VMM to keep
742 		 * 1G/2M SEPT entries where possible and speeds up process by
743 		 * cutting number of hypercalls (if successful).
744 		 */
745 
746 		if (try_accept_one(&start, len, PG_LEVEL_1G))
747 			continue;
748 
749 		if (try_accept_one(&start, len, PG_LEVEL_2M))
750 			continue;
751 
752 		if (!try_accept_one(&start, len, PG_LEVEL_4K))
753 			return false;
754 	}
755 
756 	return true;
757 }
758 
tdx_early_init(void)759 void __init tdx_early_init(void)
760 {
761 	u64 cc_mask;
762 	u32 eax, sig[3];
763 
764 	cpuid_count(TDX_CPUID_LEAF_ID, 0, &eax, &sig[0], &sig[2],  &sig[1]);
765 
766 	if (memcmp(TDX_IDENT, sig, sizeof(sig)))
767 		return;
768 
769 	setup_force_cpu_cap(X86_FEATURE_TDX_GUEST);
770 
771 	cc_set_vendor(CC_VENDOR_INTEL);
772 	tdx_parse_tdinfo(&cc_mask);
773 	cc_set_mask(cc_mask);
774 
775 	/*
776 	 * All bits above GPA width are reserved and kernel treats shared bit
777 	 * as flag, not as part of physical address.
778 	 *
779 	 * Adjust physical mask to only cover valid GPA bits.
780 	 */
781 	physical_mask &= cc_mask - 1;
782 
783 	x86_platform.guest.enc_cache_flush_required = tdx_cache_flush_required;
784 	x86_platform.guest.enc_tlb_flush_required   = tdx_tlb_flush_required;
785 	x86_platform.guest.enc_status_change_finish = tdx_enc_status_changed;
786 
787 	pr_info("Guest detected\n");
788 }
789