Loading...
Note: File does not exist in v3.1.
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 <linux/export.h>
9#include <linux/io.h>
10#include <asm/coco.h>
11#include <asm/tdx.h>
12#include <asm/vmx.h>
13#include <asm/insn.h>
14#include <asm/insn-eval.h>
15#include <asm/pgtable.h>
16
17/* TDX module Call Leaf IDs */
18#define TDX_GET_INFO 1
19#define TDX_GET_VEINFO 3
20#define TDX_GET_REPORT 4
21#define TDX_ACCEPT_PAGE 6
22
23/* TDX hypercall Leaf IDs */
24#define TDVMCALL_MAP_GPA 0x10001
25
26/* MMIO direction */
27#define EPT_READ 0
28#define EPT_WRITE 1
29
30/* Port I/O direction */
31#define PORT_READ 0
32#define PORT_WRITE 1
33
34/* See Exit Qualification for I/O Instructions in VMX documentation */
35#define VE_IS_IO_IN(e) ((e) & BIT(3))
36#define VE_GET_IO_SIZE(e) (((e) & GENMASK(2, 0)) + 1)
37#define VE_GET_PORT_NUM(e) ((e) >> 16)
38#define VE_IS_IO_STRING(e) ((e) & BIT(4))
39
40#define ATTR_SEPT_VE_DISABLE BIT(28)
41
42/* TDX Module call error codes */
43#define TDCALL_RETURN_CODE(a) ((a) >> 32)
44#define TDCALL_INVALID_OPERAND 0xc0000100
45
46#define TDREPORT_SUBTYPE_0 0
47
48/*
49 * Wrapper for standard use of __tdx_hypercall with no output aside from
50 * return code.
51 */
52static inline u64 _tdx_hypercall(u64 fn, u64 r12, u64 r13, u64 r14, u64 r15)
53{
54 struct tdx_hypercall_args args = {
55 .r10 = TDX_HYPERCALL_STANDARD,
56 .r11 = fn,
57 .r12 = r12,
58 .r13 = r13,
59 .r14 = r14,
60 .r15 = r15,
61 };
62
63 return __tdx_hypercall(&args, 0);
64}
65
66/* Called from __tdx_hypercall() for unrecoverable failure */
67void __tdx_hypercall_failed(void)
68{
69 panic("TDVMCALL failed. TDX module bug?");
70}
71
72/*
73 * The TDG.VP.VMCALL-Instruction-execution sub-functions are defined
74 * independently from but are currently matched 1:1 with VMX EXIT_REASONs.
75 * Reusing the KVM EXIT_REASON macros makes it easier to connect the host and
76 * guest sides of these calls.
77 */
78static u64 hcall_func(u64 exit_reason)
79{
80 return exit_reason;
81}
82
83#ifdef CONFIG_KVM_GUEST
84long tdx_kvm_hypercall(unsigned int nr, unsigned long p1, unsigned long p2,
85 unsigned long p3, unsigned long p4)
86{
87 struct tdx_hypercall_args args = {
88 .r10 = nr,
89 .r11 = p1,
90 .r12 = p2,
91 .r13 = p3,
92 .r14 = p4,
93 };
94
95 return __tdx_hypercall(&args, 0);
96}
97EXPORT_SYMBOL_GPL(tdx_kvm_hypercall);
98#endif
99
100/*
101 * Used for TDX guests to make calls directly to the TD module. This
102 * should only be used for calls that have no legitimate reason to fail
103 * or where the kernel can not survive the call failing.
104 */
105static inline void tdx_module_call(u64 fn, u64 rcx, u64 rdx, u64 r8, u64 r9,
106 struct tdx_module_output *out)
107{
108 if (__tdx_module_call(fn, rcx, rdx, r8, r9, out))
109 panic("TDCALL %lld failed (Buggy TDX module!)\n", fn);
110}
111
112/**
113 * tdx_mcall_get_report0() - Wrapper to get TDREPORT0 (a.k.a. TDREPORT
114 * subtype 0) using TDG.MR.REPORT TDCALL.
115 * @reportdata: Address of the input buffer which contains user-defined
116 * REPORTDATA to be included into TDREPORT.
117 * @tdreport: Address of the output buffer to store TDREPORT.
118 *
119 * Refer to section titled "TDG.MR.REPORT leaf" in the TDX Module
120 * v1.0 specification for more information on TDG.MR.REPORT TDCALL.
121 * It is used in the TDX guest driver module to get the TDREPORT0.
122 *
123 * Return 0 on success, -EINVAL for invalid operands, or -EIO on
124 * other TDCALL failures.
125 */
126int tdx_mcall_get_report0(u8 *reportdata, u8 *tdreport)
127{
128 u64 ret;
129
130 ret = __tdx_module_call(TDX_GET_REPORT, virt_to_phys(tdreport),
131 virt_to_phys(reportdata), TDREPORT_SUBTYPE_0,
132 0, NULL);
133 if (ret) {
134 if (TDCALL_RETURN_CODE(ret) == TDCALL_INVALID_OPERAND)
135 return -EINVAL;
136 return -EIO;
137 }
138
139 return 0;
140}
141EXPORT_SYMBOL_GPL(tdx_mcall_get_report0);
142
143static void tdx_parse_tdinfo(u64 *cc_mask)
144{
145 struct tdx_module_output out;
146 unsigned int gpa_width;
147 u64 td_attr;
148
149 /*
150 * TDINFO TDX module call is used to get the TD execution environment
151 * information like GPA width, number of available vcpus, debug mode
152 * information, etc. More details about the ABI can be found in TDX
153 * Guest-Host-Communication Interface (GHCI), section 2.4.2 TDCALL
154 * [TDG.VP.INFO].
155 */
156 tdx_module_call(TDX_GET_INFO, 0, 0, 0, 0, &out);
157
158 /*
159 * The highest bit of a guest physical address is the "sharing" bit.
160 * Set it for shared pages and clear it for private pages.
161 *
162 * The GPA width that comes out of this call is critical. TDX guests
163 * can not meaningfully run without it.
164 */
165 gpa_width = out.rcx & GENMASK(5, 0);
166 *cc_mask = BIT_ULL(gpa_width - 1);
167
168 /*
169 * The kernel can not handle #VE's when accessing normal kernel
170 * memory. Ensure that no #VE will be delivered for accesses to
171 * TD-private memory. Only VMM-shared memory (MMIO) will #VE.
172 */
173 td_attr = out.rdx;
174 if (!(td_attr & ATTR_SEPT_VE_DISABLE))
175 panic("TD misconfiguration: SEPT_VE_DISABLE attibute must be set.\n");
176}
177
178/*
179 * The TDX module spec states that #VE may be injected for a limited set of
180 * reasons:
181 *
182 * - Emulation of the architectural #VE injection on EPT violation;
183 *
184 * - As a result of guest TD execution of a disallowed instruction,
185 * a disallowed MSR access, or CPUID virtualization;
186 *
187 * - A notification to the guest TD about anomalous behavior;
188 *
189 * The last one is opt-in and is not used by the kernel.
190 *
191 * The Intel Software Developer's Manual describes cases when instruction
192 * length field can be used in section "Information for VM Exits Due to
193 * Instruction Execution".
194 *
195 * For TDX, it ultimately means GET_VEINFO provides reliable instruction length
196 * information if #VE occurred due to instruction execution, but not for EPT
197 * violations.
198 */
199static int ve_instr_len(struct ve_info *ve)
200{
201 switch (ve->exit_reason) {
202 case EXIT_REASON_HLT:
203 case EXIT_REASON_MSR_READ:
204 case EXIT_REASON_MSR_WRITE:
205 case EXIT_REASON_CPUID:
206 case EXIT_REASON_IO_INSTRUCTION:
207 /* It is safe to use ve->instr_len for #VE due instructions */
208 return ve->instr_len;
209 case EXIT_REASON_EPT_VIOLATION:
210 /*
211 * For EPT violations, ve->insn_len is not defined. For those,
212 * the kernel must decode instructions manually and should not
213 * be using this function.
214 */
215 WARN_ONCE(1, "ve->instr_len is not defined for EPT violations");
216 return 0;
217 default:
218 WARN_ONCE(1, "Unexpected #VE-type: %lld\n", ve->exit_reason);
219 return ve->instr_len;
220 }
221}
222
223static u64 __cpuidle __halt(const bool irq_disabled, const bool do_sti)
224{
225 struct tdx_hypercall_args args = {
226 .r10 = TDX_HYPERCALL_STANDARD,
227 .r11 = hcall_func(EXIT_REASON_HLT),
228 .r12 = irq_disabled,
229 };
230
231 /*
232 * Emulate HLT operation via hypercall. More info about ABI
233 * can be found in TDX Guest-Host-Communication Interface
234 * (GHCI), section 3.8 TDG.VP.VMCALL<Instruction.HLT>.
235 *
236 * The VMM uses the "IRQ disabled" param to understand IRQ
237 * enabled status (RFLAGS.IF) of the TD guest and to determine
238 * whether or not it should schedule the halted vCPU if an
239 * IRQ becomes pending. E.g. if IRQs are disabled, the VMM
240 * can keep the vCPU in virtual HLT, even if an IRQ is
241 * pending, without hanging/breaking the guest.
242 */
243 return __tdx_hypercall(&args, do_sti ? TDX_HCALL_ISSUE_STI : 0);
244}
245
246static int handle_halt(struct ve_info *ve)
247{
248 /*
249 * Since non safe halt is mainly used in CPU offlining
250 * and the guest will always stay in the halt state, don't
251 * call the STI instruction (set do_sti as false).
252 */
253 const bool irq_disabled = irqs_disabled();
254 const bool do_sti = false;
255
256 if (__halt(irq_disabled, do_sti))
257 return -EIO;
258
259 return ve_instr_len(ve);
260}
261
262void __cpuidle tdx_safe_halt(void)
263{
264 /*
265 * For do_sti=true case, __tdx_hypercall() function enables
266 * interrupts using the STI instruction before the TDCALL. So
267 * set irq_disabled as false.
268 */
269 const bool irq_disabled = false;
270 const bool do_sti = true;
271
272 /*
273 * Use WARN_ONCE() to report the failure.
274 */
275 if (__halt(irq_disabled, do_sti))
276 WARN_ONCE(1, "HLT instruction emulation failed\n");
277}
278
279static int read_msr(struct pt_regs *regs, struct ve_info *ve)
280{
281 struct tdx_hypercall_args args = {
282 .r10 = TDX_HYPERCALL_STANDARD,
283 .r11 = hcall_func(EXIT_REASON_MSR_READ),
284 .r12 = regs->cx,
285 };
286
287 /*
288 * Emulate the MSR read via hypercall. More info about ABI
289 * can be found in TDX Guest-Host-Communication Interface
290 * (GHCI), section titled "TDG.VP.VMCALL<Instruction.RDMSR>".
291 */
292 if (__tdx_hypercall(&args, TDX_HCALL_HAS_OUTPUT))
293 return -EIO;
294
295 regs->ax = lower_32_bits(args.r11);
296 regs->dx = upper_32_bits(args.r11);
297 return ve_instr_len(ve);
298}
299
300static int write_msr(struct pt_regs *regs, struct ve_info *ve)
301{
302 struct tdx_hypercall_args args = {
303 .r10 = TDX_HYPERCALL_STANDARD,
304 .r11 = hcall_func(EXIT_REASON_MSR_WRITE),
305 .r12 = regs->cx,
306 .r13 = (u64)regs->dx << 32 | regs->ax,
307 };
308
309 /*
310 * Emulate the MSR write via hypercall. More info about ABI
311 * can be found in TDX Guest-Host-Communication Interface
312 * (GHCI) section titled "TDG.VP.VMCALL<Instruction.WRMSR>".
313 */
314 if (__tdx_hypercall(&args, 0))
315 return -EIO;
316
317 return ve_instr_len(ve);
318}
319
320static int handle_cpuid(struct pt_regs *regs, struct ve_info *ve)
321{
322 struct tdx_hypercall_args args = {
323 .r10 = TDX_HYPERCALL_STANDARD,
324 .r11 = hcall_func(EXIT_REASON_CPUID),
325 .r12 = regs->ax,
326 .r13 = regs->cx,
327 };
328
329 /*
330 * Only allow VMM to control range reserved for hypervisor
331 * communication.
332 *
333 * Return all-zeros for any CPUID outside the range. It matches CPU
334 * behaviour for non-supported leaf.
335 */
336 if (regs->ax < 0x40000000 || regs->ax > 0x4FFFFFFF) {
337 regs->ax = regs->bx = regs->cx = regs->dx = 0;
338 return ve_instr_len(ve);
339 }
340
341 /*
342 * Emulate the CPUID instruction via a hypercall. More info about
343 * ABI can be found in TDX Guest-Host-Communication Interface
344 * (GHCI), section titled "VP.VMCALL<Instruction.CPUID>".
345 */
346 if (__tdx_hypercall(&args, TDX_HCALL_HAS_OUTPUT))
347 return -EIO;
348
349 /*
350 * As per TDX GHCI CPUID ABI, r12-r15 registers contain contents of
351 * EAX, EBX, ECX, EDX registers after the CPUID instruction execution.
352 * So copy the register contents back to pt_regs.
353 */
354 regs->ax = args.r12;
355 regs->bx = args.r13;
356 regs->cx = args.r14;
357 regs->dx = args.r15;
358
359 return ve_instr_len(ve);
360}
361
362static bool mmio_read(int size, unsigned long addr, unsigned long *val)
363{
364 struct tdx_hypercall_args args = {
365 .r10 = TDX_HYPERCALL_STANDARD,
366 .r11 = hcall_func(EXIT_REASON_EPT_VIOLATION),
367 .r12 = size,
368 .r13 = EPT_READ,
369 .r14 = addr,
370 .r15 = *val,
371 };
372
373 if (__tdx_hypercall(&args, TDX_HCALL_HAS_OUTPUT))
374 return false;
375 *val = args.r11;
376 return true;
377}
378
379static bool mmio_write(int size, unsigned long addr, unsigned long val)
380{
381 return !_tdx_hypercall(hcall_func(EXIT_REASON_EPT_VIOLATION), size,
382 EPT_WRITE, addr, val);
383}
384
385static int handle_mmio(struct pt_regs *regs, struct ve_info *ve)
386{
387 unsigned long *reg, val, vaddr;
388 char buffer[MAX_INSN_SIZE];
389 enum insn_mmio_type mmio;
390 struct insn insn = {};
391 int size, extend_size;
392 u8 extend_val = 0;
393
394 /* Only in-kernel MMIO is supported */
395 if (WARN_ON_ONCE(user_mode(regs)))
396 return -EFAULT;
397
398 if (copy_from_kernel_nofault(buffer, (void *)regs->ip, MAX_INSN_SIZE))
399 return -EFAULT;
400
401 if (insn_decode(&insn, buffer, MAX_INSN_SIZE, INSN_MODE_64))
402 return -EINVAL;
403
404 mmio = insn_decode_mmio(&insn, &size);
405 if (WARN_ON_ONCE(mmio == INSN_MMIO_DECODE_FAILED))
406 return -EINVAL;
407
408 if (mmio != INSN_MMIO_WRITE_IMM && mmio != INSN_MMIO_MOVS) {
409 reg = insn_get_modrm_reg_ptr(&insn, regs);
410 if (!reg)
411 return -EINVAL;
412 }
413
414 /*
415 * Reject EPT violation #VEs that split pages.
416 *
417 * MMIO accesses are supposed to be naturally aligned and therefore
418 * never cross page boundaries. Seeing split page accesses indicates
419 * a bug or a load_unaligned_zeropad() that stepped into an MMIO page.
420 *
421 * load_unaligned_zeropad() will recover using exception fixups.
422 */
423 vaddr = (unsigned long)insn_get_addr_ref(&insn, regs);
424 if (vaddr / PAGE_SIZE != (vaddr + size - 1) / PAGE_SIZE)
425 return -EFAULT;
426
427 /* Handle writes first */
428 switch (mmio) {
429 case INSN_MMIO_WRITE:
430 memcpy(&val, reg, size);
431 if (!mmio_write(size, ve->gpa, val))
432 return -EIO;
433 return insn.length;
434 case INSN_MMIO_WRITE_IMM:
435 val = insn.immediate.value;
436 if (!mmio_write(size, ve->gpa, val))
437 return -EIO;
438 return insn.length;
439 case INSN_MMIO_READ:
440 case INSN_MMIO_READ_ZERO_EXTEND:
441 case INSN_MMIO_READ_SIGN_EXTEND:
442 /* Reads are handled below */
443 break;
444 case INSN_MMIO_MOVS:
445 case INSN_MMIO_DECODE_FAILED:
446 /*
447 * MMIO was accessed with an instruction that could not be
448 * decoded or handled properly. It was likely not using io.h
449 * helpers or accessed MMIO accidentally.
450 */
451 return -EINVAL;
452 default:
453 WARN_ONCE(1, "Unknown insn_decode_mmio() decode value?");
454 return -EINVAL;
455 }
456
457 /* Handle reads */
458 if (!mmio_read(size, ve->gpa, &val))
459 return -EIO;
460
461 switch (mmio) {
462 case INSN_MMIO_READ:
463 /* Zero-extend for 32-bit operation */
464 extend_size = size == 4 ? sizeof(*reg) : 0;
465 break;
466 case INSN_MMIO_READ_ZERO_EXTEND:
467 /* Zero extend based on operand size */
468 extend_size = insn.opnd_bytes;
469 break;
470 case INSN_MMIO_READ_SIGN_EXTEND:
471 /* Sign extend based on operand size */
472 extend_size = insn.opnd_bytes;
473 if (size == 1 && val & BIT(7))
474 extend_val = 0xFF;
475 else if (size > 1 && val & BIT(15))
476 extend_val = 0xFF;
477 break;
478 default:
479 /* All other cases has to be covered with the first switch() */
480 WARN_ON_ONCE(1);
481 return -EINVAL;
482 }
483
484 if (extend_size)
485 memset(reg, extend_val, extend_size);
486 memcpy(reg, &val, size);
487 return insn.length;
488}
489
490static bool handle_in(struct pt_regs *regs, int size, int port)
491{
492 struct tdx_hypercall_args args = {
493 .r10 = TDX_HYPERCALL_STANDARD,
494 .r11 = hcall_func(EXIT_REASON_IO_INSTRUCTION),
495 .r12 = size,
496 .r13 = PORT_READ,
497 .r14 = port,
498 };
499 u64 mask = GENMASK(BITS_PER_BYTE * size, 0);
500 bool success;
501
502 /*
503 * Emulate the I/O read via hypercall. More info about ABI can be found
504 * in TDX Guest-Host-Communication Interface (GHCI) section titled
505 * "TDG.VP.VMCALL<Instruction.IO>".
506 */
507 success = !__tdx_hypercall(&args, TDX_HCALL_HAS_OUTPUT);
508
509 /* Update part of the register affected by the emulated instruction */
510 regs->ax &= ~mask;
511 if (success)
512 regs->ax |= args.r11 & mask;
513
514 return success;
515}
516
517static bool handle_out(struct pt_regs *regs, int size, int port)
518{
519 u64 mask = GENMASK(BITS_PER_BYTE * size, 0);
520
521 /*
522 * Emulate the I/O write via hypercall. More info about ABI can be found
523 * in TDX Guest-Host-Communication Interface (GHCI) section titled
524 * "TDG.VP.VMCALL<Instruction.IO>".
525 */
526 return !_tdx_hypercall(hcall_func(EXIT_REASON_IO_INSTRUCTION), size,
527 PORT_WRITE, port, regs->ax & mask);
528}
529
530/*
531 * Emulate I/O using hypercall.
532 *
533 * Assumes the IO instruction was using ax, which is enforced
534 * by the standard io.h macros.
535 *
536 * Return True on success or False on failure.
537 */
538static int handle_io(struct pt_regs *regs, struct ve_info *ve)
539{
540 u32 exit_qual = ve->exit_qual;
541 int size, port;
542 bool in, ret;
543
544 if (VE_IS_IO_STRING(exit_qual))
545 return -EIO;
546
547 in = VE_IS_IO_IN(exit_qual);
548 size = VE_GET_IO_SIZE(exit_qual);
549 port = VE_GET_PORT_NUM(exit_qual);
550
551
552 if (in)
553 ret = handle_in(regs, size, port);
554 else
555 ret = handle_out(regs, size, port);
556 if (!ret)
557 return -EIO;
558
559 return ve_instr_len(ve);
560}
561
562/*
563 * Early #VE exception handler. Only handles a subset of port I/O.
564 * Intended only for earlyprintk. If failed, return false.
565 */
566__init bool tdx_early_handle_ve(struct pt_regs *regs)
567{
568 struct ve_info ve;
569 int insn_len;
570
571 tdx_get_ve_info(&ve);
572
573 if (ve.exit_reason != EXIT_REASON_IO_INSTRUCTION)
574 return false;
575
576 insn_len = handle_io(regs, &ve);
577 if (insn_len < 0)
578 return false;
579
580 regs->ip += insn_len;
581 return true;
582}
583
584void tdx_get_ve_info(struct ve_info *ve)
585{
586 struct tdx_module_output out;
587
588 /*
589 * Called during #VE handling to retrieve the #VE info from the
590 * TDX module.
591 *
592 * This has to be called early in #VE handling. A "nested" #VE which
593 * occurs before this will raise a #DF and is not recoverable.
594 *
595 * The call retrieves the #VE info from the TDX module, which also
596 * clears the "#VE valid" flag. This must be done before anything else
597 * because any #VE that occurs while the valid flag is set will lead to
598 * #DF.
599 *
600 * Note, the TDX module treats virtual NMIs as inhibited if the #VE
601 * valid flag is set. It means that NMI=>#VE will not result in a #DF.
602 */
603 tdx_module_call(TDX_GET_VEINFO, 0, 0, 0, 0, &out);
604
605 /* Transfer the output parameters */
606 ve->exit_reason = out.rcx;
607 ve->exit_qual = out.rdx;
608 ve->gla = out.r8;
609 ve->gpa = out.r9;
610 ve->instr_len = lower_32_bits(out.r10);
611 ve->instr_info = upper_32_bits(out.r10);
612}
613
614/*
615 * Handle the user initiated #VE.
616 *
617 * On success, returns the number of bytes RIP should be incremented (>=0)
618 * or -errno on error.
619 */
620static int virt_exception_user(struct pt_regs *regs, struct ve_info *ve)
621{
622 switch (ve->exit_reason) {
623 case EXIT_REASON_CPUID:
624 return handle_cpuid(regs, ve);
625 default:
626 pr_warn("Unexpected #VE: %lld\n", ve->exit_reason);
627 return -EIO;
628 }
629}
630
631/*
632 * Handle the kernel #VE.
633 *
634 * On success, returns the number of bytes RIP should be incremented (>=0)
635 * or -errno on error.
636 */
637static int virt_exception_kernel(struct pt_regs *regs, struct ve_info *ve)
638{
639 switch (ve->exit_reason) {
640 case EXIT_REASON_HLT:
641 return handle_halt(ve);
642 case EXIT_REASON_MSR_READ:
643 return read_msr(regs, ve);
644 case EXIT_REASON_MSR_WRITE:
645 return write_msr(regs, ve);
646 case EXIT_REASON_CPUID:
647 return handle_cpuid(regs, ve);
648 case EXIT_REASON_EPT_VIOLATION:
649 return handle_mmio(regs, ve);
650 case EXIT_REASON_IO_INSTRUCTION:
651 return handle_io(regs, ve);
652 default:
653 pr_warn("Unexpected #VE: %lld\n", ve->exit_reason);
654 return -EIO;
655 }
656}
657
658bool tdx_handle_virt_exception(struct pt_regs *regs, struct ve_info *ve)
659{
660 int insn_len;
661
662 if (user_mode(regs))
663 insn_len = virt_exception_user(regs, ve);
664 else
665 insn_len = virt_exception_kernel(regs, ve);
666 if (insn_len < 0)
667 return false;
668
669 /* After successful #VE handling, move the IP */
670 regs->ip += insn_len;
671
672 return true;
673}
674
675static bool tdx_tlb_flush_required(bool private)
676{
677 /*
678 * TDX guest is responsible for flushing TLB on private->shared
679 * transition. VMM is responsible for flushing on shared->private.
680 *
681 * The VMM _can't_ flush private addresses as it can't generate PAs
682 * with the guest's HKID. Shared memory isn't subject to integrity
683 * checking, i.e. the VMM doesn't need to flush for its own protection.
684 *
685 * There's no need to flush when converting from shared to private,
686 * as flushing is the VMM's responsibility in this case, e.g. it must
687 * flush to avoid integrity failures in the face of a buggy or
688 * malicious guest.
689 */
690 return !private;
691}
692
693static bool tdx_cache_flush_required(void)
694{
695 /*
696 * AMD SME/SEV can avoid cache flushing if HW enforces cache coherence.
697 * TDX doesn't have such capability.
698 *
699 * Flush cache unconditionally.
700 */
701 return true;
702}
703
704static bool try_accept_one(phys_addr_t *start, unsigned long len,
705 enum pg_level pg_level)
706{
707 unsigned long accept_size = page_level_size(pg_level);
708 u64 tdcall_rcx;
709 u8 page_size;
710
711 if (!IS_ALIGNED(*start, accept_size))
712 return false;
713
714 if (len < accept_size)
715 return false;
716
717 /*
718 * Pass the page physical address to the TDX module to accept the
719 * pending, private page.
720 *
721 * Bits 2:0 of RCX encode page size: 0 - 4K, 1 - 2M, 2 - 1G.
722 */
723 switch (pg_level) {
724 case PG_LEVEL_4K:
725 page_size = 0;
726 break;
727 case PG_LEVEL_2M:
728 page_size = 1;
729 break;
730 case PG_LEVEL_1G:
731 page_size = 2;
732 break;
733 default:
734 return false;
735 }
736
737 tdcall_rcx = *start | page_size;
738 if (__tdx_module_call(TDX_ACCEPT_PAGE, tdcall_rcx, 0, 0, 0, NULL))
739 return false;
740
741 *start += accept_size;
742 return true;
743}
744
745/*
746 * Inform the VMM of the guest's intent for this physical page: shared with
747 * the VMM or private to the guest. The VMM is expected to change its mapping
748 * of the page in response.
749 */
750static bool tdx_enc_status_changed(unsigned long vaddr, int numpages, bool enc)
751{
752 phys_addr_t start = __pa(vaddr);
753 phys_addr_t end = __pa(vaddr + numpages * PAGE_SIZE);
754
755 if (!enc) {
756 /* Set the shared (decrypted) bits: */
757 start |= cc_mkdec(0);
758 end |= cc_mkdec(0);
759 }
760
761 /*
762 * Notify the VMM about page mapping conversion. More info about ABI
763 * can be found in TDX Guest-Host-Communication Interface (GHCI),
764 * section "TDG.VP.VMCALL<MapGPA>"
765 */
766 if (_tdx_hypercall(TDVMCALL_MAP_GPA, start, end - start, 0, 0))
767 return false;
768
769 /* private->shared conversion requires only MapGPA call */
770 if (!enc)
771 return true;
772
773 /*
774 * For shared->private conversion, accept the page using
775 * TDX_ACCEPT_PAGE TDX module call.
776 */
777 while (start < end) {
778 unsigned long len = end - start;
779
780 /*
781 * Try larger accepts first. It gives chance to VMM to keep
782 * 1G/2M SEPT entries where possible and speeds up process by
783 * cutting number of hypercalls (if successful).
784 */
785
786 if (try_accept_one(&start, len, PG_LEVEL_1G))
787 continue;
788
789 if (try_accept_one(&start, len, PG_LEVEL_2M))
790 continue;
791
792 if (!try_accept_one(&start, len, PG_LEVEL_4K))
793 return false;
794 }
795
796 return true;
797}
798
799void __init tdx_early_init(void)
800{
801 u64 cc_mask;
802 u32 eax, sig[3];
803
804 cpuid_count(TDX_CPUID_LEAF_ID, 0, &eax, &sig[0], &sig[2], &sig[1]);
805
806 if (memcmp(TDX_IDENT, sig, sizeof(sig)))
807 return;
808
809 setup_force_cpu_cap(X86_FEATURE_TDX_GUEST);
810
811 cc_set_vendor(CC_VENDOR_INTEL);
812 tdx_parse_tdinfo(&cc_mask);
813 cc_set_mask(cc_mask);
814
815 /*
816 * All bits above GPA width are reserved and kernel treats shared bit
817 * as flag, not as part of physical address.
818 *
819 * Adjust physical mask to only cover valid GPA bits.
820 */
821 physical_mask &= cc_mask - 1;
822
823 x86_platform.guest.enc_cache_flush_required = tdx_cache_flush_required;
824 x86_platform.guest.enc_tlb_flush_required = tdx_tlb_flush_required;
825 x86_platform.guest.enc_status_change_finish = tdx_enc_status_changed;
826
827 pr_info("Guest detected\n");
828}