xref: /openbmc/linux/arch/x86/kvm/svm/sev.c (revision 6c71a0574249f5e5a45fe055ab5f837023d5eeca)
1  // SPDX-License-Identifier: GPL-2.0-only
2  /*
3   * Kernel-based Virtual Machine driver for Linux
4   *
5   * AMD SVM-SEV support
6   *
7   * Copyright 2010 Red Hat, Inc. and/or its affiliates.
8   */
9  #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
10  
11  #include <linux/kvm_types.h>
12  #include <linux/kvm_host.h>
13  #include <linux/kernel.h>
14  #include <linux/highmem.h>
15  #include <linux/psp.h>
16  #include <linux/psp-sev.h>
17  #include <linux/pagemap.h>
18  #include <linux/swap.h>
19  #include <linux/misc_cgroup.h>
20  #include <linux/processor.h>
21  #include <linux/trace_events.h>
22  
23  #include <asm/pkru.h>
24  #include <asm/trapnr.h>
25  #include <asm/fpu/xcr.h>
26  #include <asm/debugreg.h>
27  
28  #include "mmu.h"
29  #include "x86.h"
30  #include "svm.h"
31  #include "svm_ops.h"
32  #include "cpuid.h"
33  #include "trace.h"
34  
35  #ifndef CONFIG_KVM_AMD_SEV
36  /*
37   * When this config is not defined, SEV feature is not supported and APIs in
38   * this file are not used but this file still gets compiled into the KVM AMD
39   * module.
40   *
41   * We will not have MISC_CG_RES_SEV and MISC_CG_RES_SEV_ES entries in the enum
42   * misc_res_type {} defined in linux/misc_cgroup.h.
43   *
44   * Below macros allow compilation to succeed.
45   */
46  #define MISC_CG_RES_SEV MISC_CG_RES_TYPES
47  #define MISC_CG_RES_SEV_ES MISC_CG_RES_TYPES
48  #endif
49  
50  #ifdef CONFIG_KVM_AMD_SEV
51  /* enable/disable SEV support */
52  static bool sev_enabled = true;
53  module_param_named(sev, sev_enabled, bool, 0444);
54  
55  /* enable/disable SEV-ES support */
56  static bool sev_es_enabled = true;
57  module_param_named(sev_es, sev_es_enabled, bool, 0444);
58  
59  /* enable/disable SEV-ES DebugSwap support */
60  static bool sev_es_debug_swap_enabled = false;
61  module_param_named(debug_swap, sev_es_debug_swap_enabled, bool, 0444);
62  #else
63  #define sev_enabled false
64  #define sev_es_enabled false
65  #define sev_es_debug_swap_enabled false
66  #endif /* CONFIG_KVM_AMD_SEV */
67  
68  static u8 sev_enc_bit;
69  static DECLARE_RWSEM(sev_deactivate_lock);
70  static DEFINE_MUTEX(sev_bitmap_lock);
71  unsigned int max_sev_asid;
72  static unsigned int min_sev_asid;
73  static unsigned long sev_me_mask;
74  static unsigned int nr_asids;
75  static unsigned long *sev_asid_bitmap;
76  static unsigned long *sev_reclaim_asid_bitmap;
77  
78  struct enc_region {
79  	struct list_head list;
80  	unsigned long npages;
81  	struct page **pages;
82  	unsigned long uaddr;
83  	unsigned long size;
84  };
85  
86  /* Called with the sev_bitmap_lock held, or on shutdown  */
sev_flush_asids(unsigned int min_asid,unsigned int max_asid)87  static int sev_flush_asids(unsigned int min_asid, unsigned int max_asid)
88  {
89  	int ret, error = 0;
90  	unsigned int asid;
91  
92  	/* Check if there are any ASIDs to reclaim before performing a flush */
93  	asid = find_next_bit(sev_reclaim_asid_bitmap, nr_asids, min_asid);
94  	if (asid > max_asid)
95  		return -EBUSY;
96  
97  	/*
98  	 * DEACTIVATE will clear the WBINVD indicator causing DF_FLUSH to fail,
99  	 * so it must be guarded.
100  	 */
101  	down_write(&sev_deactivate_lock);
102  
103  	wbinvd_on_all_cpus();
104  	ret = sev_guest_df_flush(&error);
105  
106  	up_write(&sev_deactivate_lock);
107  
108  	if (ret)
109  		pr_err("SEV: DF_FLUSH failed, ret=%d, error=%#x\n", ret, error);
110  
111  	return ret;
112  }
113  
is_mirroring_enc_context(struct kvm * kvm)114  static inline bool is_mirroring_enc_context(struct kvm *kvm)
115  {
116  	return !!to_kvm_svm(kvm)->sev_info.enc_context_owner;
117  }
118  
119  /* Must be called with the sev_bitmap_lock held */
__sev_recycle_asids(unsigned int min_asid,unsigned int max_asid)120  static bool __sev_recycle_asids(unsigned int min_asid, unsigned int max_asid)
121  {
122  	if (sev_flush_asids(min_asid, max_asid))
123  		return false;
124  
125  	/* The flush process will flush all reclaimable SEV and SEV-ES ASIDs */
126  	bitmap_xor(sev_asid_bitmap, sev_asid_bitmap, sev_reclaim_asid_bitmap,
127  		   nr_asids);
128  	bitmap_zero(sev_reclaim_asid_bitmap, nr_asids);
129  
130  	return true;
131  }
132  
sev_misc_cg_try_charge(struct kvm_sev_info * sev)133  static int sev_misc_cg_try_charge(struct kvm_sev_info *sev)
134  {
135  	enum misc_res_type type = sev->es_active ? MISC_CG_RES_SEV_ES : MISC_CG_RES_SEV;
136  	return misc_cg_try_charge(type, sev->misc_cg, 1);
137  }
138  
sev_misc_cg_uncharge(struct kvm_sev_info * sev)139  static void sev_misc_cg_uncharge(struct kvm_sev_info *sev)
140  {
141  	enum misc_res_type type = sev->es_active ? MISC_CG_RES_SEV_ES : MISC_CG_RES_SEV;
142  	misc_cg_uncharge(type, sev->misc_cg, 1);
143  }
144  
sev_asid_new(struct kvm_sev_info * sev)145  static int sev_asid_new(struct kvm_sev_info *sev)
146  {
147  	/*
148  	 * SEV-enabled guests must use asid from min_sev_asid to max_sev_asid.
149  	 * SEV-ES-enabled guest can use from 1 to min_sev_asid - 1.
150  	 * Note: min ASID can end up larger than the max if basic SEV support is
151  	 * effectively disabled by disallowing use of ASIDs for SEV guests.
152  	 */
153  	unsigned int min_asid = sev->es_active ? 1 : min_sev_asid;
154  	unsigned int max_asid = sev->es_active ? min_sev_asid - 1 : max_sev_asid;
155  	unsigned int asid;
156  	bool retry = true;
157  	int ret;
158  
159  	if (min_asid > max_asid)
160  		return -ENOTTY;
161  
162  	WARN_ON(sev->misc_cg);
163  	sev->misc_cg = get_current_misc_cg();
164  	ret = sev_misc_cg_try_charge(sev);
165  	if (ret) {
166  		put_misc_cg(sev->misc_cg);
167  		sev->misc_cg = NULL;
168  		return ret;
169  	}
170  
171  	mutex_lock(&sev_bitmap_lock);
172  
173  again:
174  	asid = find_next_zero_bit(sev_asid_bitmap, max_asid + 1, min_asid);
175  	if (asid > max_asid) {
176  		if (retry && __sev_recycle_asids(min_asid, max_asid)) {
177  			retry = false;
178  			goto again;
179  		}
180  		mutex_unlock(&sev_bitmap_lock);
181  		ret = -EBUSY;
182  		goto e_uncharge;
183  	}
184  
185  	__set_bit(asid, sev_asid_bitmap);
186  
187  	mutex_unlock(&sev_bitmap_lock);
188  
189  	return asid;
190  e_uncharge:
191  	sev_misc_cg_uncharge(sev);
192  	put_misc_cg(sev->misc_cg);
193  	sev->misc_cg = NULL;
194  	return ret;
195  }
196  
sev_get_asid(struct kvm * kvm)197  static unsigned int sev_get_asid(struct kvm *kvm)
198  {
199  	struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
200  
201  	return sev->asid;
202  }
203  
sev_asid_free(struct kvm_sev_info * sev)204  static void sev_asid_free(struct kvm_sev_info *sev)
205  {
206  	struct svm_cpu_data *sd;
207  	int cpu;
208  
209  	mutex_lock(&sev_bitmap_lock);
210  
211  	__set_bit(sev->asid, sev_reclaim_asid_bitmap);
212  
213  	for_each_possible_cpu(cpu) {
214  		sd = per_cpu_ptr(&svm_data, cpu);
215  		sd->sev_vmcbs[sev->asid] = NULL;
216  	}
217  
218  	mutex_unlock(&sev_bitmap_lock);
219  
220  	sev_misc_cg_uncharge(sev);
221  	put_misc_cg(sev->misc_cg);
222  	sev->misc_cg = NULL;
223  }
224  
sev_decommission(unsigned int handle)225  static void sev_decommission(unsigned int handle)
226  {
227  	struct sev_data_decommission decommission;
228  
229  	if (!handle)
230  		return;
231  
232  	decommission.handle = handle;
233  	sev_guest_decommission(&decommission, NULL);
234  }
235  
sev_unbind_asid(struct kvm * kvm,unsigned int handle)236  static void sev_unbind_asid(struct kvm *kvm, unsigned int handle)
237  {
238  	struct sev_data_deactivate deactivate;
239  
240  	if (!handle)
241  		return;
242  
243  	deactivate.handle = handle;
244  
245  	/* Guard DEACTIVATE against WBINVD/DF_FLUSH used in ASID recycling */
246  	down_read(&sev_deactivate_lock);
247  	sev_guest_deactivate(&deactivate, NULL);
248  	up_read(&sev_deactivate_lock);
249  
250  	sev_decommission(handle);
251  }
252  
sev_guest_init(struct kvm * kvm,struct kvm_sev_cmd * argp)253  static int sev_guest_init(struct kvm *kvm, struct kvm_sev_cmd *argp)
254  {
255  	struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
256  	int asid, ret;
257  
258  	if (kvm->created_vcpus)
259  		return -EINVAL;
260  
261  	ret = -EBUSY;
262  	if (unlikely(sev->active))
263  		return ret;
264  
265  	sev->active = true;
266  	sev->es_active = argp->id == KVM_SEV_ES_INIT;
267  	asid = sev_asid_new(sev);
268  	if (asid < 0)
269  		goto e_no_asid;
270  	sev->asid = asid;
271  
272  	ret = sev_platform_init(&argp->error);
273  	if (ret)
274  		goto e_free;
275  
276  	INIT_LIST_HEAD(&sev->regions_list);
277  	INIT_LIST_HEAD(&sev->mirror_vms);
278  
279  	kvm_set_apicv_inhibit(kvm, APICV_INHIBIT_REASON_SEV);
280  
281  	return 0;
282  
283  e_free:
284  	sev_asid_free(sev);
285  	sev->asid = 0;
286  e_no_asid:
287  	sev->es_active = false;
288  	sev->active = false;
289  	return ret;
290  }
291  
sev_bind_asid(struct kvm * kvm,unsigned int handle,int * error)292  static int sev_bind_asid(struct kvm *kvm, unsigned int handle, int *error)
293  {
294  	unsigned int asid = sev_get_asid(kvm);
295  	struct sev_data_activate activate;
296  	int ret;
297  
298  	/* activate ASID on the given handle */
299  	activate.handle = handle;
300  	activate.asid   = asid;
301  	ret = sev_guest_activate(&activate, error);
302  
303  	return ret;
304  }
305  
__sev_issue_cmd(int fd,int id,void * data,int * error)306  static int __sev_issue_cmd(int fd, int id, void *data, int *error)
307  {
308  	struct fd f;
309  	int ret;
310  
311  	f = fdget(fd);
312  	if (!f.file)
313  		return -EBADF;
314  
315  	ret = sev_issue_cmd_external_user(f.file, id, data, error);
316  
317  	fdput(f);
318  	return ret;
319  }
320  
sev_issue_cmd(struct kvm * kvm,int id,void * data,int * error)321  static int sev_issue_cmd(struct kvm *kvm, int id, void *data, int *error)
322  {
323  	struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
324  
325  	return __sev_issue_cmd(sev->fd, id, data, error);
326  }
327  
sev_launch_start(struct kvm * kvm,struct kvm_sev_cmd * argp)328  static int sev_launch_start(struct kvm *kvm, struct kvm_sev_cmd *argp)
329  {
330  	struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
331  	struct sev_data_launch_start start;
332  	struct kvm_sev_launch_start params;
333  	void *dh_blob, *session_blob;
334  	int *error = &argp->error;
335  	int ret;
336  
337  	if (!sev_guest(kvm))
338  		return -ENOTTY;
339  
340  	if (copy_from_user(&params, (void __user *)(uintptr_t)argp->data, sizeof(params)))
341  		return -EFAULT;
342  
343  	memset(&start, 0, sizeof(start));
344  
345  	dh_blob = NULL;
346  	if (params.dh_uaddr) {
347  		dh_blob = psp_copy_user_blob(params.dh_uaddr, params.dh_len);
348  		if (IS_ERR(dh_blob))
349  			return PTR_ERR(dh_blob);
350  
351  		start.dh_cert_address = __sme_set(__pa(dh_blob));
352  		start.dh_cert_len = params.dh_len;
353  	}
354  
355  	session_blob = NULL;
356  	if (params.session_uaddr) {
357  		session_blob = psp_copy_user_blob(params.session_uaddr, params.session_len);
358  		if (IS_ERR(session_blob)) {
359  			ret = PTR_ERR(session_blob);
360  			goto e_free_dh;
361  		}
362  
363  		start.session_address = __sme_set(__pa(session_blob));
364  		start.session_len = params.session_len;
365  	}
366  
367  	start.handle = params.handle;
368  	start.policy = params.policy;
369  
370  	/* create memory encryption context */
371  	ret = __sev_issue_cmd(argp->sev_fd, SEV_CMD_LAUNCH_START, &start, error);
372  	if (ret)
373  		goto e_free_session;
374  
375  	/* Bind ASID to this guest */
376  	ret = sev_bind_asid(kvm, start.handle, error);
377  	if (ret) {
378  		sev_decommission(start.handle);
379  		goto e_free_session;
380  	}
381  
382  	/* return handle to userspace */
383  	params.handle = start.handle;
384  	if (copy_to_user((void __user *)(uintptr_t)argp->data, &params, sizeof(params))) {
385  		sev_unbind_asid(kvm, start.handle);
386  		ret = -EFAULT;
387  		goto e_free_session;
388  	}
389  
390  	sev->handle = start.handle;
391  	sev->fd = argp->sev_fd;
392  
393  e_free_session:
394  	kfree(session_blob);
395  e_free_dh:
396  	kfree(dh_blob);
397  	return ret;
398  }
399  
sev_pin_memory(struct kvm * kvm,unsigned long uaddr,unsigned long ulen,unsigned long * n,int write)400  static struct page **sev_pin_memory(struct kvm *kvm, unsigned long uaddr,
401  				    unsigned long ulen, unsigned long *n,
402  				    int write)
403  {
404  	struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
405  	unsigned long npages, size;
406  	int npinned;
407  	unsigned long locked, lock_limit;
408  	struct page **pages;
409  	unsigned long first, last;
410  	int ret;
411  
412  	lockdep_assert_held(&kvm->lock);
413  
414  	if (ulen == 0 || uaddr + ulen < uaddr)
415  		return ERR_PTR(-EINVAL);
416  
417  	/* Calculate number of pages. */
418  	first = (uaddr & PAGE_MASK) >> PAGE_SHIFT;
419  	last = ((uaddr + ulen - 1) & PAGE_MASK) >> PAGE_SHIFT;
420  	npages = (last - first + 1);
421  
422  	locked = sev->pages_locked + npages;
423  	lock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
424  	if (locked > lock_limit && !capable(CAP_IPC_LOCK)) {
425  		pr_err("SEV: %lu locked pages exceed the lock limit of %lu.\n", locked, lock_limit);
426  		return ERR_PTR(-ENOMEM);
427  	}
428  
429  	if (WARN_ON_ONCE(npages > INT_MAX))
430  		return ERR_PTR(-EINVAL);
431  
432  	/* Avoid using vmalloc for smaller buffers. */
433  	size = npages * sizeof(struct page *);
434  	if (size > PAGE_SIZE)
435  		pages = __vmalloc(size, GFP_KERNEL_ACCOUNT | __GFP_ZERO);
436  	else
437  		pages = kmalloc(size, GFP_KERNEL_ACCOUNT);
438  
439  	if (!pages)
440  		return ERR_PTR(-ENOMEM);
441  
442  	/* Pin the user virtual address. */
443  	npinned = pin_user_pages_fast(uaddr, npages, write ? FOLL_WRITE : 0, pages);
444  	if (npinned != npages) {
445  		pr_err("SEV: Failure locking %lu pages.\n", npages);
446  		ret = -ENOMEM;
447  		goto err;
448  	}
449  
450  	*n = npages;
451  	sev->pages_locked = locked;
452  
453  	return pages;
454  
455  err:
456  	if (npinned > 0)
457  		unpin_user_pages(pages, npinned);
458  
459  	kvfree(pages);
460  	return ERR_PTR(ret);
461  }
462  
sev_unpin_memory(struct kvm * kvm,struct page ** pages,unsigned long npages)463  static void sev_unpin_memory(struct kvm *kvm, struct page **pages,
464  			     unsigned long npages)
465  {
466  	struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
467  
468  	unpin_user_pages(pages, npages);
469  	kvfree(pages);
470  	sev->pages_locked -= npages;
471  }
472  
sev_clflush_pages(struct page * pages[],unsigned long npages)473  static void sev_clflush_pages(struct page *pages[], unsigned long npages)
474  {
475  	uint8_t *page_virtual;
476  	unsigned long i;
477  
478  	if (this_cpu_has(X86_FEATURE_SME_COHERENT) || npages == 0 ||
479  	    pages == NULL)
480  		return;
481  
482  	for (i = 0; i < npages; i++) {
483  		page_virtual = kmap_local_page(pages[i]);
484  		clflush_cache_range(page_virtual, PAGE_SIZE);
485  		kunmap_local(page_virtual);
486  		cond_resched();
487  	}
488  }
489  
get_num_contig_pages(unsigned long idx,struct page ** inpages,unsigned long npages)490  static unsigned long get_num_contig_pages(unsigned long idx,
491  				struct page **inpages, unsigned long npages)
492  {
493  	unsigned long paddr, next_paddr;
494  	unsigned long i = idx + 1, pages = 1;
495  
496  	/* find the number of contiguous pages starting from idx */
497  	paddr = __sme_page_pa(inpages[idx]);
498  	while (i < npages) {
499  		next_paddr = __sme_page_pa(inpages[i++]);
500  		if ((paddr + PAGE_SIZE) == next_paddr) {
501  			pages++;
502  			paddr = next_paddr;
503  			continue;
504  		}
505  		break;
506  	}
507  
508  	return pages;
509  }
510  
sev_launch_update_data(struct kvm * kvm,struct kvm_sev_cmd * argp)511  static int sev_launch_update_data(struct kvm *kvm, struct kvm_sev_cmd *argp)
512  {
513  	unsigned long vaddr, vaddr_end, next_vaddr, npages, pages, size, i;
514  	struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
515  	struct kvm_sev_launch_update_data params;
516  	struct sev_data_launch_update_data data;
517  	struct page **inpages;
518  	int ret;
519  
520  	if (!sev_guest(kvm))
521  		return -ENOTTY;
522  
523  	if (copy_from_user(&params, (void __user *)(uintptr_t)argp->data, sizeof(params)))
524  		return -EFAULT;
525  
526  	vaddr = params.uaddr;
527  	size = params.len;
528  	vaddr_end = vaddr + size;
529  
530  	/* Lock the user memory. */
531  	inpages = sev_pin_memory(kvm, vaddr, size, &npages, 1);
532  	if (IS_ERR(inpages))
533  		return PTR_ERR(inpages);
534  
535  	/*
536  	 * Flush (on non-coherent CPUs) before LAUNCH_UPDATE encrypts pages in
537  	 * place; the cache may contain the data that was written unencrypted.
538  	 */
539  	sev_clflush_pages(inpages, npages);
540  
541  	data.reserved = 0;
542  	data.handle = sev->handle;
543  
544  	for (i = 0; vaddr < vaddr_end; vaddr = next_vaddr, i += pages) {
545  		int offset, len;
546  
547  		/*
548  		 * If the user buffer is not page-aligned, calculate the offset
549  		 * within the page.
550  		 */
551  		offset = vaddr & (PAGE_SIZE - 1);
552  
553  		/* Calculate the number of pages that can be encrypted in one go. */
554  		pages = get_num_contig_pages(i, inpages, npages);
555  
556  		len = min_t(size_t, ((pages * PAGE_SIZE) - offset), size);
557  
558  		data.len = len;
559  		data.address = __sme_page_pa(inpages[i]) + offset;
560  		ret = sev_issue_cmd(kvm, SEV_CMD_LAUNCH_UPDATE_DATA, &data, &argp->error);
561  		if (ret)
562  			goto e_unpin;
563  
564  		size -= len;
565  		next_vaddr = vaddr + len;
566  	}
567  
568  e_unpin:
569  	/* content of memory is updated, mark pages dirty */
570  	for (i = 0; i < npages; i++) {
571  		set_page_dirty_lock(inpages[i]);
572  		mark_page_accessed(inpages[i]);
573  	}
574  	/* unlock the user pages */
575  	sev_unpin_memory(kvm, inpages, npages);
576  	return ret;
577  }
578  
sev_es_sync_vmsa(struct vcpu_svm * svm)579  static int sev_es_sync_vmsa(struct vcpu_svm *svm)
580  {
581  	struct sev_es_save_area *save = svm->sev_es.vmsa;
582  
583  	/* Check some debug related fields before encrypting the VMSA */
584  	if (svm->vcpu.guest_debug || (svm->vmcb->save.dr7 & ~DR7_FIXED_1))
585  		return -EINVAL;
586  
587  	/*
588  	 * SEV-ES will use a VMSA that is pointed to by the VMCB, not
589  	 * the traditional VMSA that is part of the VMCB. Copy the
590  	 * traditional VMSA as it has been built so far (in prep
591  	 * for LAUNCH_UPDATE_VMSA) to be the initial SEV-ES state.
592  	 */
593  	memcpy(save, &svm->vmcb->save, sizeof(svm->vmcb->save));
594  
595  	/* Sync registgers */
596  	save->rax = svm->vcpu.arch.regs[VCPU_REGS_RAX];
597  	save->rbx = svm->vcpu.arch.regs[VCPU_REGS_RBX];
598  	save->rcx = svm->vcpu.arch.regs[VCPU_REGS_RCX];
599  	save->rdx = svm->vcpu.arch.regs[VCPU_REGS_RDX];
600  	save->rsp = svm->vcpu.arch.regs[VCPU_REGS_RSP];
601  	save->rbp = svm->vcpu.arch.regs[VCPU_REGS_RBP];
602  	save->rsi = svm->vcpu.arch.regs[VCPU_REGS_RSI];
603  	save->rdi = svm->vcpu.arch.regs[VCPU_REGS_RDI];
604  #ifdef CONFIG_X86_64
605  	save->r8  = svm->vcpu.arch.regs[VCPU_REGS_R8];
606  	save->r9  = svm->vcpu.arch.regs[VCPU_REGS_R9];
607  	save->r10 = svm->vcpu.arch.regs[VCPU_REGS_R10];
608  	save->r11 = svm->vcpu.arch.regs[VCPU_REGS_R11];
609  	save->r12 = svm->vcpu.arch.regs[VCPU_REGS_R12];
610  	save->r13 = svm->vcpu.arch.regs[VCPU_REGS_R13];
611  	save->r14 = svm->vcpu.arch.regs[VCPU_REGS_R14];
612  	save->r15 = svm->vcpu.arch.regs[VCPU_REGS_R15];
613  #endif
614  	save->rip = svm->vcpu.arch.regs[VCPU_REGS_RIP];
615  
616  	/* Sync some non-GPR registers before encrypting */
617  	save->xcr0 = svm->vcpu.arch.xcr0;
618  	save->pkru = svm->vcpu.arch.pkru;
619  	save->xss  = svm->vcpu.arch.ia32_xss;
620  	save->dr6  = svm->vcpu.arch.dr6;
621  
622  	if (sev_es_debug_swap_enabled) {
623  		save->sev_features |= SVM_SEV_FEAT_DEBUG_SWAP;
624  		pr_warn_once("Enabling DebugSwap with KVM_SEV_ES_INIT. "
625  			     "This will not work starting with Linux 6.10\n");
626  	}
627  
628  	pr_debug("Virtual Machine Save Area (VMSA):\n");
629  	print_hex_dump_debug("", DUMP_PREFIX_NONE, 16, 1, save, sizeof(*save), false);
630  
631  	return 0;
632  }
633  
__sev_launch_update_vmsa(struct kvm * kvm,struct kvm_vcpu * vcpu,int * error)634  static int __sev_launch_update_vmsa(struct kvm *kvm, struct kvm_vcpu *vcpu,
635  				    int *error)
636  {
637  	struct sev_data_launch_update_vmsa vmsa;
638  	struct vcpu_svm *svm = to_svm(vcpu);
639  	int ret;
640  
641  	if (vcpu->guest_debug) {
642  		pr_warn_once("KVM_SET_GUEST_DEBUG for SEV-ES guest is not supported");
643  		return -EINVAL;
644  	}
645  
646  	/* Perform some pre-encryption checks against the VMSA */
647  	ret = sev_es_sync_vmsa(svm);
648  	if (ret)
649  		return ret;
650  
651  	/*
652  	 * The LAUNCH_UPDATE_VMSA command will perform in-place encryption of
653  	 * the VMSA memory content (i.e it will write the same memory region
654  	 * with the guest's key), so invalidate it first.
655  	 */
656  	clflush_cache_range(svm->sev_es.vmsa, PAGE_SIZE);
657  
658  	vmsa.reserved = 0;
659  	vmsa.handle = to_kvm_svm(kvm)->sev_info.handle;
660  	vmsa.address = __sme_pa(svm->sev_es.vmsa);
661  	vmsa.len = PAGE_SIZE;
662  	ret = sev_issue_cmd(kvm, SEV_CMD_LAUNCH_UPDATE_VMSA, &vmsa, error);
663  	if (ret)
664  	  return ret;
665  
666  	vcpu->arch.guest_state_protected = true;
667  
668  	/*
669  	 * SEV-ES guest mandates LBR Virtualization to be _always_ ON. Enable it
670  	 * only after setting guest_state_protected because KVM_SET_MSRS allows
671  	 * dynamic toggling of LBRV (for performance reason) on write access to
672  	 * MSR_IA32_DEBUGCTLMSR when guest_state_protected is not set.
673  	 */
674  	svm_enable_lbrv(vcpu);
675  	return 0;
676  }
677  
sev_launch_update_vmsa(struct kvm * kvm,struct kvm_sev_cmd * argp)678  static int sev_launch_update_vmsa(struct kvm *kvm, struct kvm_sev_cmd *argp)
679  {
680  	struct kvm_vcpu *vcpu;
681  	unsigned long i;
682  	int ret;
683  
684  	if (!sev_es_guest(kvm))
685  		return -ENOTTY;
686  
687  	kvm_for_each_vcpu(i, vcpu, kvm) {
688  		ret = mutex_lock_killable(&vcpu->mutex);
689  		if (ret)
690  			return ret;
691  
692  		ret = __sev_launch_update_vmsa(kvm, vcpu, &argp->error);
693  
694  		mutex_unlock(&vcpu->mutex);
695  		if (ret)
696  			return ret;
697  	}
698  
699  	return 0;
700  }
701  
sev_launch_measure(struct kvm * kvm,struct kvm_sev_cmd * argp)702  static int sev_launch_measure(struct kvm *kvm, struct kvm_sev_cmd *argp)
703  {
704  	void __user *measure = (void __user *)(uintptr_t)argp->data;
705  	struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
706  	struct sev_data_launch_measure data;
707  	struct kvm_sev_launch_measure params;
708  	void __user *p = NULL;
709  	void *blob = NULL;
710  	int ret;
711  
712  	if (!sev_guest(kvm))
713  		return -ENOTTY;
714  
715  	if (copy_from_user(&params, measure, sizeof(params)))
716  		return -EFAULT;
717  
718  	memset(&data, 0, sizeof(data));
719  
720  	/* User wants to query the blob length */
721  	if (!params.len)
722  		goto cmd;
723  
724  	p = (void __user *)(uintptr_t)params.uaddr;
725  	if (p) {
726  		if (params.len > SEV_FW_BLOB_MAX_SIZE)
727  			return -EINVAL;
728  
729  		blob = kzalloc(params.len, GFP_KERNEL_ACCOUNT);
730  		if (!blob)
731  			return -ENOMEM;
732  
733  		data.address = __psp_pa(blob);
734  		data.len = params.len;
735  	}
736  
737  cmd:
738  	data.handle = sev->handle;
739  	ret = sev_issue_cmd(kvm, SEV_CMD_LAUNCH_MEASURE, &data, &argp->error);
740  
741  	/*
742  	 * If we query the session length, FW responded with expected data.
743  	 */
744  	if (!params.len)
745  		goto done;
746  
747  	if (ret)
748  		goto e_free_blob;
749  
750  	if (blob) {
751  		if (copy_to_user(p, blob, params.len))
752  			ret = -EFAULT;
753  	}
754  
755  done:
756  	params.len = data.len;
757  	if (copy_to_user(measure, &params, sizeof(params)))
758  		ret = -EFAULT;
759  e_free_blob:
760  	kfree(blob);
761  	return ret;
762  }
763  
sev_launch_finish(struct kvm * kvm,struct kvm_sev_cmd * argp)764  static int sev_launch_finish(struct kvm *kvm, struct kvm_sev_cmd *argp)
765  {
766  	struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
767  	struct sev_data_launch_finish data;
768  
769  	if (!sev_guest(kvm))
770  		return -ENOTTY;
771  
772  	data.handle = sev->handle;
773  	return sev_issue_cmd(kvm, SEV_CMD_LAUNCH_FINISH, &data, &argp->error);
774  }
775  
sev_guest_status(struct kvm * kvm,struct kvm_sev_cmd * argp)776  static int sev_guest_status(struct kvm *kvm, struct kvm_sev_cmd *argp)
777  {
778  	struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
779  	struct kvm_sev_guest_status params;
780  	struct sev_data_guest_status data;
781  	int ret;
782  
783  	if (!sev_guest(kvm))
784  		return -ENOTTY;
785  
786  	memset(&data, 0, sizeof(data));
787  
788  	data.handle = sev->handle;
789  	ret = sev_issue_cmd(kvm, SEV_CMD_GUEST_STATUS, &data, &argp->error);
790  	if (ret)
791  		return ret;
792  
793  	params.policy = data.policy;
794  	params.state = data.state;
795  	params.handle = data.handle;
796  
797  	if (copy_to_user((void __user *)(uintptr_t)argp->data, &params, sizeof(params)))
798  		ret = -EFAULT;
799  
800  	return ret;
801  }
802  
__sev_issue_dbg_cmd(struct kvm * kvm,unsigned long src,unsigned long dst,int size,int * error,bool enc)803  static int __sev_issue_dbg_cmd(struct kvm *kvm, unsigned long src,
804  			       unsigned long dst, int size,
805  			       int *error, bool enc)
806  {
807  	struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
808  	struct sev_data_dbg data;
809  
810  	data.reserved = 0;
811  	data.handle = sev->handle;
812  	data.dst_addr = dst;
813  	data.src_addr = src;
814  	data.len = size;
815  
816  	return sev_issue_cmd(kvm,
817  			     enc ? SEV_CMD_DBG_ENCRYPT : SEV_CMD_DBG_DECRYPT,
818  			     &data, error);
819  }
820  
__sev_dbg_decrypt(struct kvm * kvm,unsigned long src_paddr,unsigned long dst_paddr,int sz,int * err)821  static int __sev_dbg_decrypt(struct kvm *kvm, unsigned long src_paddr,
822  			     unsigned long dst_paddr, int sz, int *err)
823  {
824  	int offset;
825  
826  	/*
827  	 * Its safe to read more than we are asked, caller should ensure that
828  	 * destination has enough space.
829  	 */
830  	offset = src_paddr & 15;
831  	src_paddr = round_down(src_paddr, 16);
832  	sz = round_up(sz + offset, 16);
833  
834  	return __sev_issue_dbg_cmd(kvm, src_paddr, dst_paddr, sz, err, false);
835  }
836  
__sev_dbg_decrypt_user(struct kvm * kvm,unsigned long paddr,void __user * dst_uaddr,unsigned long dst_paddr,int size,int * err)837  static int __sev_dbg_decrypt_user(struct kvm *kvm, unsigned long paddr,
838  				  void __user *dst_uaddr,
839  				  unsigned long dst_paddr,
840  				  int size, int *err)
841  {
842  	struct page *tpage = NULL;
843  	int ret, offset;
844  
845  	/* if inputs are not 16-byte then use intermediate buffer */
846  	if (!IS_ALIGNED(dst_paddr, 16) ||
847  	    !IS_ALIGNED(paddr,     16) ||
848  	    !IS_ALIGNED(size,      16)) {
849  		tpage = (void *)alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
850  		if (!tpage)
851  			return -ENOMEM;
852  
853  		dst_paddr = __sme_page_pa(tpage);
854  	}
855  
856  	ret = __sev_dbg_decrypt(kvm, paddr, dst_paddr, size, err);
857  	if (ret)
858  		goto e_free;
859  
860  	if (tpage) {
861  		offset = paddr & 15;
862  		if (copy_to_user(dst_uaddr, page_address(tpage) + offset, size))
863  			ret = -EFAULT;
864  	}
865  
866  e_free:
867  	if (tpage)
868  		__free_page(tpage);
869  
870  	return ret;
871  }
872  
__sev_dbg_encrypt_user(struct kvm * kvm,unsigned long paddr,void __user * vaddr,unsigned long dst_paddr,void __user * dst_vaddr,int size,int * error)873  static int __sev_dbg_encrypt_user(struct kvm *kvm, unsigned long paddr,
874  				  void __user *vaddr,
875  				  unsigned long dst_paddr,
876  				  void __user *dst_vaddr,
877  				  int size, int *error)
878  {
879  	struct page *src_tpage = NULL;
880  	struct page *dst_tpage = NULL;
881  	int ret, len = size;
882  
883  	/* If source buffer is not aligned then use an intermediate buffer */
884  	if (!IS_ALIGNED((unsigned long)vaddr, 16)) {
885  		src_tpage = alloc_page(GFP_KERNEL_ACCOUNT);
886  		if (!src_tpage)
887  			return -ENOMEM;
888  
889  		if (copy_from_user(page_address(src_tpage), vaddr, size)) {
890  			__free_page(src_tpage);
891  			return -EFAULT;
892  		}
893  
894  		paddr = __sme_page_pa(src_tpage);
895  	}
896  
897  	/*
898  	 *  If destination buffer or length is not aligned then do read-modify-write:
899  	 *   - decrypt destination in an intermediate buffer
900  	 *   - copy the source buffer in an intermediate buffer
901  	 *   - use the intermediate buffer as source buffer
902  	 */
903  	if (!IS_ALIGNED((unsigned long)dst_vaddr, 16) || !IS_ALIGNED(size, 16)) {
904  		int dst_offset;
905  
906  		dst_tpage = alloc_page(GFP_KERNEL_ACCOUNT);
907  		if (!dst_tpage) {
908  			ret = -ENOMEM;
909  			goto e_free;
910  		}
911  
912  		ret = __sev_dbg_decrypt(kvm, dst_paddr,
913  					__sme_page_pa(dst_tpage), size, error);
914  		if (ret)
915  			goto e_free;
916  
917  		/*
918  		 *  If source is kernel buffer then use memcpy() otherwise
919  		 *  copy_from_user().
920  		 */
921  		dst_offset = dst_paddr & 15;
922  
923  		if (src_tpage)
924  			memcpy(page_address(dst_tpage) + dst_offset,
925  			       page_address(src_tpage), size);
926  		else {
927  			if (copy_from_user(page_address(dst_tpage) + dst_offset,
928  					   vaddr, size)) {
929  				ret = -EFAULT;
930  				goto e_free;
931  			}
932  		}
933  
934  		paddr = __sme_page_pa(dst_tpage);
935  		dst_paddr = round_down(dst_paddr, 16);
936  		len = round_up(size, 16);
937  	}
938  
939  	ret = __sev_issue_dbg_cmd(kvm, paddr, dst_paddr, len, error, true);
940  
941  e_free:
942  	if (src_tpage)
943  		__free_page(src_tpage);
944  	if (dst_tpage)
945  		__free_page(dst_tpage);
946  	return ret;
947  }
948  
sev_dbg_crypt(struct kvm * kvm,struct kvm_sev_cmd * argp,bool dec)949  static int sev_dbg_crypt(struct kvm *kvm, struct kvm_sev_cmd *argp, bool dec)
950  {
951  	unsigned long vaddr, vaddr_end, next_vaddr;
952  	unsigned long dst_vaddr;
953  	struct page **src_p, **dst_p;
954  	struct kvm_sev_dbg debug;
955  	unsigned long n;
956  	unsigned int size;
957  	int ret;
958  
959  	if (!sev_guest(kvm))
960  		return -ENOTTY;
961  
962  	if (copy_from_user(&debug, (void __user *)(uintptr_t)argp->data, sizeof(debug)))
963  		return -EFAULT;
964  
965  	if (!debug.len || debug.src_uaddr + debug.len < debug.src_uaddr)
966  		return -EINVAL;
967  	if (!debug.dst_uaddr)
968  		return -EINVAL;
969  
970  	vaddr = debug.src_uaddr;
971  	size = debug.len;
972  	vaddr_end = vaddr + size;
973  	dst_vaddr = debug.dst_uaddr;
974  
975  	for (; vaddr < vaddr_end; vaddr = next_vaddr) {
976  		int len, s_off, d_off;
977  
978  		/* lock userspace source and destination page */
979  		src_p = sev_pin_memory(kvm, vaddr & PAGE_MASK, PAGE_SIZE, &n, 0);
980  		if (IS_ERR(src_p))
981  			return PTR_ERR(src_p);
982  
983  		dst_p = sev_pin_memory(kvm, dst_vaddr & PAGE_MASK, PAGE_SIZE, &n, 1);
984  		if (IS_ERR(dst_p)) {
985  			sev_unpin_memory(kvm, src_p, n);
986  			return PTR_ERR(dst_p);
987  		}
988  
989  		/*
990  		 * Flush (on non-coherent CPUs) before DBG_{DE,EN}CRYPT read or modify
991  		 * the pages; flush the destination too so that future accesses do not
992  		 * see stale data.
993  		 */
994  		sev_clflush_pages(src_p, 1);
995  		sev_clflush_pages(dst_p, 1);
996  
997  		/*
998  		 * Since user buffer may not be page aligned, calculate the
999  		 * offset within the page.
1000  		 */
1001  		s_off = vaddr & ~PAGE_MASK;
1002  		d_off = dst_vaddr & ~PAGE_MASK;
1003  		len = min_t(size_t, (PAGE_SIZE - s_off), size);
1004  
1005  		if (dec)
1006  			ret = __sev_dbg_decrypt_user(kvm,
1007  						     __sme_page_pa(src_p[0]) + s_off,
1008  						     (void __user *)dst_vaddr,
1009  						     __sme_page_pa(dst_p[0]) + d_off,
1010  						     len, &argp->error);
1011  		else
1012  			ret = __sev_dbg_encrypt_user(kvm,
1013  						     __sme_page_pa(src_p[0]) + s_off,
1014  						     (void __user *)vaddr,
1015  						     __sme_page_pa(dst_p[0]) + d_off,
1016  						     (void __user *)dst_vaddr,
1017  						     len, &argp->error);
1018  
1019  		sev_unpin_memory(kvm, src_p, n);
1020  		sev_unpin_memory(kvm, dst_p, n);
1021  
1022  		if (ret)
1023  			goto err;
1024  
1025  		next_vaddr = vaddr + len;
1026  		dst_vaddr = dst_vaddr + len;
1027  		size -= len;
1028  	}
1029  err:
1030  	return ret;
1031  }
1032  
sev_launch_secret(struct kvm * kvm,struct kvm_sev_cmd * argp)1033  static int sev_launch_secret(struct kvm *kvm, struct kvm_sev_cmd *argp)
1034  {
1035  	struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1036  	struct sev_data_launch_secret data;
1037  	struct kvm_sev_launch_secret params;
1038  	struct page **pages;
1039  	void *blob, *hdr;
1040  	unsigned long n, i;
1041  	int ret, offset;
1042  
1043  	if (!sev_guest(kvm))
1044  		return -ENOTTY;
1045  
1046  	if (copy_from_user(&params, (void __user *)(uintptr_t)argp->data, sizeof(params)))
1047  		return -EFAULT;
1048  
1049  	pages = sev_pin_memory(kvm, params.guest_uaddr, params.guest_len, &n, 1);
1050  	if (IS_ERR(pages))
1051  		return PTR_ERR(pages);
1052  
1053  	/*
1054  	 * Flush (on non-coherent CPUs) before LAUNCH_SECRET encrypts pages in
1055  	 * place; the cache may contain the data that was written unencrypted.
1056  	 */
1057  	sev_clflush_pages(pages, n);
1058  
1059  	/*
1060  	 * The secret must be copied into contiguous memory region, lets verify
1061  	 * that userspace memory pages are contiguous before we issue command.
1062  	 */
1063  	if (get_num_contig_pages(0, pages, n) != n) {
1064  		ret = -EINVAL;
1065  		goto e_unpin_memory;
1066  	}
1067  
1068  	memset(&data, 0, sizeof(data));
1069  
1070  	offset = params.guest_uaddr & (PAGE_SIZE - 1);
1071  	data.guest_address = __sme_page_pa(pages[0]) + offset;
1072  	data.guest_len = params.guest_len;
1073  
1074  	blob = psp_copy_user_blob(params.trans_uaddr, params.trans_len);
1075  	if (IS_ERR(blob)) {
1076  		ret = PTR_ERR(blob);
1077  		goto e_unpin_memory;
1078  	}
1079  
1080  	data.trans_address = __psp_pa(blob);
1081  	data.trans_len = params.trans_len;
1082  
1083  	hdr = psp_copy_user_blob(params.hdr_uaddr, params.hdr_len);
1084  	if (IS_ERR(hdr)) {
1085  		ret = PTR_ERR(hdr);
1086  		goto e_free_blob;
1087  	}
1088  	data.hdr_address = __psp_pa(hdr);
1089  	data.hdr_len = params.hdr_len;
1090  
1091  	data.handle = sev->handle;
1092  	ret = sev_issue_cmd(kvm, SEV_CMD_LAUNCH_UPDATE_SECRET, &data, &argp->error);
1093  
1094  	kfree(hdr);
1095  
1096  e_free_blob:
1097  	kfree(blob);
1098  e_unpin_memory:
1099  	/* content of memory is updated, mark pages dirty */
1100  	for (i = 0; i < n; i++) {
1101  		set_page_dirty_lock(pages[i]);
1102  		mark_page_accessed(pages[i]);
1103  	}
1104  	sev_unpin_memory(kvm, pages, n);
1105  	return ret;
1106  }
1107  
sev_get_attestation_report(struct kvm * kvm,struct kvm_sev_cmd * argp)1108  static int sev_get_attestation_report(struct kvm *kvm, struct kvm_sev_cmd *argp)
1109  {
1110  	void __user *report = (void __user *)(uintptr_t)argp->data;
1111  	struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1112  	struct sev_data_attestation_report data;
1113  	struct kvm_sev_attestation_report params;
1114  	void __user *p;
1115  	void *blob = NULL;
1116  	int ret;
1117  
1118  	if (!sev_guest(kvm))
1119  		return -ENOTTY;
1120  
1121  	if (copy_from_user(&params, (void __user *)(uintptr_t)argp->data, sizeof(params)))
1122  		return -EFAULT;
1123  
1124  	memset(&data, 0, sizeof(data));
1125  
1126  	/* User wants to query the blob length */
1127  	if (!params.len)
1128  		goto cmd;
1129  
1130  	p = (void __user *)(uintptr_t)params.uaddr;
1131  	if (p) {
1132  		if (params.len > SEV_FW_BLOB_MAX_SIZE)
1133  			return -EINVAL;
1134  
1135  		blob = kzalloc(params.len, GFP_KERNEL_ACCOUNT);
1136  		if (!blob)
1137  			return -ENOMEM;
1138  
1139  		data.address = __psp_pa(blob);
1140  		data.len = params.len;
1141  		memcpy(data.mnonce, params.mnonce, sizeof(params.mnonce));
1142  	}
1143  cmd:
1144  	data.handle = sev->handle;
1145  	ret = sev_issue_cmd(kvm, SEV_CMD_ATTESTATION_REPORT, &data, &argp->error);
1146  	/*
1147  	 * If we query the session length, FW responded with expected data.
1148  	 */
1149  	if (!params.len)
1150  		goto done;
1151  
1152  	if (ret)
1153  		goto e_free_blob;
1154  
1155  	if (blob) {
1156  		if (copy_to_user(p, blob, params.len))
1157  			ret = -EFAULT;
1158  	}
1159  
1160  done:
1161  	params.len = data.len;
1162  	if (copy_to_user(report, &params, sizeof(params)))
1163  		ret = -EFAULT;
1164  e_free_blob:
1165  	kfree(blob);
1166  	return ret;
1167  }
1168  
1169  /* Userspace wants to query session length. */
1170  static int
__sev_send_start_query_session_length(struct kvm * kvm,struct kvm_sev_cmd * argp,struct kvm_sev_send_start * params)1171  __sev_send_start_query_session_length(struct kvm *kvm, struct kvm_sev_cmd *argp,
1172  				      struct kvm_sev_send_start *params)
1173  {
1174  	struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1175  	struct sev_data_send_start data;
1176  	int ret;
1177  
1178  	memset(&data, 0, sizeof(data));
1179  	data.handle = sev->handle;
1180  	ret = sev_issue_cmd(kvm, SEV_CMD_SEND_START, &data, &argp->error);
1181  
1182  	params->session_len = data.session_len;
1183  	if (copy_to_user((void __user *)(uintptr_t)argp->data, params,
1184  				sizeof(struct kvm_sev_send_start)))
1185  		ret = -EFAULT;
1186  
1187  	return ret;
1188  }
1189  
sev_send_start(struct kvm * kvm,struct kvm_sev_cmd * argp)1190  static int sev_send_start(struct kvm *kvm, struct kvm_sev_cmd *argp)
1191  {
1192  	struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1193  	struct sev_data_send_start data;
1194  	struct kvm_sev_send_start params;
1195  	void *amd_certs, *session_data;
1196  	void *pdh_cert, *plat_certs;
1197  	int ret;
1198  
1199  	if (!sev_guest(kvm))
1200  		return -ENOTTY;
1201  
1202  	if (copy_from_user(&params, (void __user *)(uintptr_t)argp->data,
1203  				sizeof(struct kvm_sev_send_start)))
1204  		return -EFAULT;
1205  
1206  	/* if session_len is zero, userspace wants to query the session length */
1207  	if (!params.session_len)
1208  		return __sev_send_start_query_session_length(kvm, argp,
1209  				&params);
1210  
1211  	/* some sanity checks */
1212  	if (!params.pdh_cert_uaddr || !params.pdh_cert_len ||
1213  	    !params.session_uaddr || params.session_len > SEV_FW_BLOB_MAX_SIZE)
1214  		return -EINVAL;
1215  
1216  	/* allocate the memory to hold the session data blob */
1217  	session_data = kzalloc(params.session_len, GFP_KERNEL_ACCOUNT);
1218  	if (!session_data)
1219  		return -ENOMEM;
1220  
1221  	/* copy the certificate blobs from userspace */
1222  	pdh_cert = psp_copy_user_blob(params.pdh_cert_uaddr,
1223  				params.pdh_cert_len);
1224  	if (IS_ERR(pdh_cert)) {
1225  		ret = PTR_ERR(pdh_cert);
1226  		goto e_free_session;
1227  	}
1228  
1229  	plat_certs = psp_copy_user_blob(params.plat_certs_uaddr,
1230  				params.plat_certs_len);
1231  	if (IS_ERR(plat_certs)) {
1232  		ret = PTR_ERR(plat_certs);
1233  		goto e_free_pdh;
1234  	}
1235  
1236  	amd_certs = psp_copy_user_blob(params.amd_certs_uaddr,
1237  				params.amd_certs_len);
1238  	if (IS_ERR(amd_certs)) {
1239  		ret = PTR_ERR(amd_certs);
1240  		goto e_free_plat_cert;
1241  	}
1242  
1243  	/* populate the FW SEND_START field with system physical address */
1244  	memset(&data, 0, sizeof(data));
1245  	data.pdh_cert_address = __psp_pa(pdh_cert);
1246  	data.pdh_cert_len = params.pdh_cert_len;
1247  	data.plat_certs_address = __psp_pa(plat_certs);
1248  	data.plat_certs_len = params.plat_certs_len;
1249  	data.amd_certs_address = __psp_pa(amd_certs);
1250  	data.amd_certs_len = params.amd_certs_len;
1251  	data.session_address = __psp_pa(session_data);
1252  	data.session_len = params.session_len;
1253  	data.handle = sev->handle;
1254  
1255  	ret = sev_issue_cmd(kvm, SEV_CMD_SEND_START, &data, &argp->error);
1256  
1257  	if (!ret && copy_to_user((void __user *)(uintptr_t)params.session_uaddr,
1258  			session_data, params.session_len)) {
1259  		ret = -EFAULT;
1260  		goto e_free_amd_cert;
1261  	}
1262  
1263  	params.policy = data.policy;
1264  	params.session_len = data.session_len;
1265  	if (copy_to_user((void __user *)(uintptr_t)argp->data, &params,
1266  				sizeof(struct kvm_sev_send_start)))
1267  		ret = -EFAULT;
1268  
1269  e_free_amd_cert:
1270  	kfree(amd_certs);
1271  e_free_plat_cert:
1272  	kfree(plat_certs);
1273  e_free_pdh:
1274  	kfree(pdh_cert);
1275  e_free_session:
1276  	kfree(session_data);
1277  	return ret;
1278  }
1279  
1280  /* Userspace wants to query either header or trans length. */
1281  static int
__sev_send_update_data_query_lengths(struct kvm * kvm,struct kvm_sev_cmd * argp,struct kvm_sev_send_update_data * params)1282  __sev_send_update_data_query_lengths(struct kvm *kvm, struct kvm_sev_cmd *argp,
1283  				     struct kvm_sev_send_update_data *params)
1284  {
1285  	struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1286  	struct sev_data_send_update_data data;
1287  	int ret;
1288  
1289  	memset(&data, 0, sizeof(data));
1290  	data.handle = sev->handle;
1291  	ret = sev_issue_cmd(kvm, SEV_CMD_SEND_UPDATE_DATA, &data, &argp->error);
1292  
1293  	params->hdr_len = data.hdr_len;
1294  	params->trans_len = data.trans_len;
1295  
1296  	if (copy_to_user((void __user *)(uintptr_t)argp->data, params,
1297  			 sizeof(struct kvm_sev_send_update_data)))
1298  		ret = -EFAULT;
1299  
1300  	return ret;
1301  }
1302  
sev_send_update_data(struct kvm * kvm,struct kvm_sev_cmd * argp)1303  static int sev_send_update_data(struct kvm *kvm, struct kvm_sev_cmd *argp)
1304  {
1305  	struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1306  	struct sev_data_send_update_data data;
1307  	struct kvm_sev_send_update_data params;
1308  	void *hdr, *trans_data;
1309  	struct page **guest_page;
1310  	unsigned long n;
1311  	int ret, offset;
1312  
1313  	if (!sev_guest(kvm))
1314  		return -ENOTTY;
1315  
1316  	if (copy_from_user(&params, (void __user *)(uintptr_t)argp->data,
1317  			sizeof(struct kvm_sev_send_update_data)))
1318  		return -EFAULT;
1319  
1320  	/* userspace wants to query either header or trans length */
1321  	if (!params.trans_len || !params.hdr_len)
1322  		return __sev_send_update_data_query_lengths(kvm, argp, &params);
1323  
1324  	if (!params.trans_uaddr || !params.guest_uaddr ||
1325  	    !params.guest_len || !params.hdr_uaddr)
1326  		return -EINVAL;
1327  
1328  	/* Check if we are crossing the page boundary */
1329  	offset = params.guest_uaddr & (PAGE_SIZE - 1);
1330  	if (params.guest_len > PAGE_SIZE || (params.guest_len + offset) > PAGE_SIZE)
1331  		return -EINVAL;
1332  
1333  	/* Pin guest memory */
1334  	guest_page = sev_pin_memory(kvm, params.guest_uaddr & PAGE_MASK,
1335  				    PAGE_SIZE, &n, 0);
1336  	if (IS_ERR(guest_page))
1337  		return PTR_ERR(guest_page);
1338  
1339  	/* allocate memory for header and transport buffer */
1340  	ret = -ENOMEM;
1341  	hdr = kzalloc(params.hdr_len, GFP_KERNEL_ACCOUNT);
1342  	if (!hdr)
1343  		goto e_unpin;
1344  
1345  	trans_data = kzalloc(params.trans_len, GFP_KERNEL_ACCOUNT);
1346  	if (!trans_data)
1347  		goto e_free_hdr;
1348  
1349  	memset(&data, 0, sizeof(data));
1350  	data.hdr_address = __psp_pa(hdr);
1351  	data.hdr_len = params.hdr_len;
1352  	data.trans_address = __psp_pa(trans_data);
1353  	data.trans_len = params.trans_len;
1354  
1355  	/* The SEND_UPDATE_DATA command requires C-bit to be always set. */
1356  	data.guest_address = (page_to_pfn(guest_page[0]) << PAGE_SHIFT) + offset;
1357  	data.guest_address |= sev_me_mask;
1358  	data.guest_len = params.guest_len;
1359  	data.handle = sev->handle;
1360  
1361  	ret = sev_issue_cmd(kvm, SEV_CMD_SEND_UPDATE_DATA, &data, &argp->error);
1362  
1363  	if (ret)
1364  		goto e_free_trans_data;
1365  
1366  	/* copy transport buffer to user space */
1367  	if (copy_to_user((void __user *)(uintptr_t)params.trans_uaddr,
1368  			 trans_data, params.trans_len)) {
1369  		ret = -EFAULT;
1370  		goto e_free_trans_data;
1371  	}
1372  
1373  	/* Copy packet header to userspace. */
1374  	if (copy_to_user((void __user *)(uintptr_t)params.hdr_uaddr, hdr,
1375  			 params.hdr_len))
1376  		ret = -EFAULT;
1377  
1378  e_free_trans_data:
1379  	kfree(trans_data);
1380  e_free_hdr:
1381  	kfree(hdr);
1382  e_unpin:
1383  	sev_unpin_memory(kvm, guest_page, n);
1384  
1385  	return ret;
1386  }
1387  
sev_send_finish(struct kvm * kvm,struct kvm_sev_cmd * argp)1388  static int sev_send_finish(struct kvm *kvm, struct kvm_sev_cmd *argp)
1389  {
1390  	struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1391  	struct sev_data_send_finish data;
1392  
1393  	if (!sev_guest(kvm))
1394  		return -ENOTTY;
1395  
1396  	data.handle = sev->handle;
1397  	return sev_issue_cmd(kvm, SEV_CMD_SEND_FINISH, &data, &argp->error);
1398  }
1399  
sev_send_cancel(struct kvm * kvm,struct kvm_sev_cmd * argp)1400  static int sev_send_cancel(struct kvm *kvm, struct kvm_sev_cmd *argp)
1401  {
1402  	struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1403  	struct sev_data_send_cancel data;
1404  
1405  	if (!sev_guest(kvm))
1406  		return -ENOTTY;
1407  
1408  	data.handle = sev->handle;
1409  	return sev_issue_cmd(kvm, SEV_CMD_SEND_CANCEL, &data, &argp->error);
1410  }
1411  
sev_receive_start(struct kvm * kvm,struct kvm_sev_cmd * argp)1412  static int sev_receive_start(struct kvm *kvm, struct kvm_sev_cmd *argp)
1413  {
1414  	struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1415  	struct sev_data_receive_start start;
1416  	struct kvm_sev_receive_start params;
1417  	int *error = &argp->error;
1418  	void *session_data;
1419  	void *pdh_data;
1420  	int ret;
1421  
1422  	if (!sev_guest(kvm))
1423  		return -ENOTTY;
1424  
1425  	/* Get parameter from the userspace */
1426  	if (copy_from_user(&params, (void __user *)(uintptr_t)argp->data,
1427  			sizeof(struct kvm_sev_receive_start)))
1428  		return -EFAULT;
1429  
1430  	/* some sanity checks */
1431  	if (!params.pdh_uaddr || !params.pdh_len ||
1432  	    !params.session_uaddr || !params.session_len)
1433  		return -EINVAL;
1434  
1435  	pdh_data = psp_copy_user_blob(params.pdh_uaddr, params.pdh_len);
1436  	if (IS_ERR(pdh_data))
1437  		return PTR_ERR(pdh_data);
1438  
1439  	session_data = psp_copy_user_blob(params.session_uaddr,
1440  			params.session_len);
1441  	if (IS_ERR(session_data)) {
1442  		ret = PTR_ERR(session_data);
1443  		goto e_free_pdh;
1444  	}
1445  
1446  	memset(&start, 0, sizeof(start));
1447  	start.handle = params.handle;
1448  	start.policy = params.policy;
1449  	start.pdh_cert_address = __psp_pa(pdh_data);
1450  	start.pdh_cert_len = params.pdh_len;
1451  	start.session_address = __psp_pa(session_data);
1452  	start.session_len = params.session_len;
1453  
1454  	/* create memory encryption context */
1455  	ret = __sev_issue_cmd(argp->sev_fd, SEV_CMD_RECEIVE_START, &start,
1456  				error);
1457  	if (ret)
1458  		goto e_free_session;
1459  
1460  	/* Bind ASID to this guest */
1461  	ret = sev_bind_asid(kvm, start.handle, error);
1462  	if (ret) {
1463  		sev_decommission(start.handle);
1464  		goto e_free_session;
1465  	}
1466  
1467  	params.handle = start.handle;
1468  	if (copy_to_user((void __user *)(uintptr_t)argp->data,
1469  			 &params, sizeof(struct kvm_sev_receive_start))) {
1470  		ret = -EFAULT;
1471  		sev_unbind_asid(kvm, start.handle);
1472  		goto e_free_session;
1473  	}
1474  
1475      	sev->handle = start.handle;
1476  	sev->fd = argp->sev_fd;
1477  
1478  e_free_session:
1479  	kfree(session_data);
1480  e_free_pdh:
1481  	kfree(pdh_data);
1482  
1483  	return ret;
1484  }
1485  
sev_receive_update_data(struct kvm * kvm,struct kvm_sev_cmd * argp)1486  static int sev_receive_update_data(struct kvm *kvm, struct kvm_sev_cmd *argp)
1487  {
1488  	struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1489  	struct kvm_sev_receive_update_data params;
1490  	struct sev_data_receive_update_data data;
1491  	void *hdr = NULL, *trans = NULL;
1492  	struct page **guest_page;
1493  	unsigned long n;
1494  	int ret, offset;
1495  
1496  	if (!sev_guest(kvm))
1497  		return -EINVAL;
1498  
1499  	if (copy_from_user(&params, (void __user *)(uintptr_t)argp->data,
1500  			sizeof(struct kvm_sev_receive_update_data)))
1501  		return -EFAULT;
1502  
1503  	if (!params.hdr_uaddr || !params.hdr_len ||
1504  	    !params.guest_uaddr || !params.guest_len ||
1505  	    !params.trans_uaddr || !params.trans_len)
1506  		return -EINVAL;
1507  
1508  	/* Check if we are crossing the page boundary */
1509  	offset = params.guest_uaddr & (PAGE_SIZE - 1);
1510  	if (params.guest_len > PAGE_SIZE || (params.guest_len + offset) > PAGE_SIZE)
1511  		return -EINVAL;
1512  
1513  	hdr = psp_copy_user_blob(params.hdr_uaddr, params.hdr_len);
1514  	if (IS_ERR(hdr))
1515  		return PTR_ERR(hdr);
1516  
1517  	trans = psp_copy_user_blob(params.trans_uaddr, params.trans_len);
1518  	if (IS_ERR(trans)) {
1519  		ret = PTR_ERR(trans);
1520  		goto e_free_hdr;
1521  	}
1522  
1523  	memset(&data, 0, sizeof(data));
1524  	data.hdr_address = __psp_pa(hdr);
1525  	data.hdr_len = params.hdr_len;
1526  	data.trans_address = __psp_pa(trans);
1527  	data.trans_len = params.trans_len;
1528  
1529  	/* Pin guest memory */
1530  	guest_page = sev_pin_memory(kvm, params.guest_uaddr & PAGE_MASK,
1531  				    PAGE_SIZE, &n, 1);
1532  	if (IS_ERR(guest_page)) {
1533  		ret = PTR_ERR(guest_page);
1534  		goto e_free_trans;
1535  	}
1536  
1537  	/*
1538  	 * Flush (on non-coherent CPUs) before RECEIVE_UPDATE_DATA, the PSP
1539  	 * encrypts the written data with the guest's key, and the cache may
1540  	 * contain dirty, unencrypted data.
1541  	 */
1542  	sev_clflush_pages(guest_page, n);
1543  
1544  	/* The RECEIVE_UPDATE_DATA command requires C-bit to be always set. */
1545  	data.guest_address = (page_to_pfn(guest_page[0]) << PAGE_SHIFT) + offset;
1546  	data.guest_address |= sev_me_mask;
1547  	data.guest_len = params.guest_len;
1548  	data.handle = sev->handle;
1549  
1550  	ret = sev_issue_cmd(kvm, SEV_CMD_RECEIVE_UPDATE_DATA, &data,
1551  				&argp->error);
1552  
1553  	sev_unpin_memory(kvm, guest_page, n);
1554  
1555  e_free_trans:
1556  	kfree(trans);
1557  e_free_hdr:
1558  	kfree(hdr);
1559  
1560  	return ret;
1561  }
1562  
sev_receive_finish(struct kvm * kvm,struct kvm_sev_cmd * argp)1563  static int sev_receive_finish(struct kvm *kvm, struct kvm_sev_cmd *argp)
1564  {
1565  	struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1566  	struct sev_data_receive_finish data;
1567  
1568  	if (!sev_guest(kvm))
1569  		return -ENOTTY;
1570  
1571  	data.handle = sev->handle;
1572  	return sev_issue_cmd(kvm, SEV_CMD_RECEIVE_FINISH, &data, &argp->error);
1573  }
1574  
is_cmd_allowed_from_mirror(u32 cmd_id)1575  static bool is_cmd_allowed_from_mirror(u32 cmd_id)
1576  {
1577  	/*
1578  	 * Allow mirrors VM to call KVM_SEV_LAUNCH_UPDATE_VMSA to enable SEV-ES
1579  	 * active mirror VMs. Also allow the debugging and status commands.
1580  	 */
1581  	if (cmd_id == KVM_SEV_LAUNCH_UPDATE_VMSA ||
1582  	    cmd_id == KVM_SEV_GUEST_STATUS || cmd_id == KVM_SEV_DBG_DECRYPT ||
1583  	    cmd_id == KVM_SEV_DBG_ENCRYPT)
1584  		return true;
1585  
1586  	return false;
1587  }
1588  
sev_lock_two_vms(struct kvm * dst_kvm,struct kvm * src_kvm)1589  static int sev_lock_two_vms(struct kvm *dst_kvm, struct kvm *src_kvm)
1590  {
1591  	struct kvm_sev_info *dst_sev = &to_kvm_svm(dst_kvm)->sev_info;
1592  	struct kvm_sev_info *src_sev = &to_kvm_svm(src_kvm)->sev_info;
1593  	int r = -EBUSY;
1594  
1595  	if (dst_kvm == src_kvm)
1596  		return -EINVAL;
1597  
1598  	/*
1599  	 * Bail if these VMs are already involved in a migration to avoid
1600  	 * deadlock between two VMs trying to migrate to/from each other.
1601  	 */
1602  	if (atomic_cmpxchg_acquire(&dst_sev->migration_in_progress, 0, 1))
1603  		return -EBUSY;
1604  
1605  	if (atomic_cmpxchg_acquire(&src_sev->migration_in_progress, 0, 1))
1606  		goto release_dst;
1607  
1608  	r = -EINTR;
1609  	if (mutex_lock_killable(&dst_kvm->lock))
1610  		goto release_src;
1611  	if (mutex_lock_killable_nested(&src_kvm->lock, SINGLE_DEPTH_NESTING))
1612  		goto unlock_dst;
1613  	return 0;
1614  
1615  unlock_dst:
1616  	mutex_unlock(&dst_kvm->lock);
1617  release_src:
1618  	atomic_set_release(&src_sev->migration_in_progress, 0);
1619  release_dst:
1620  	atomic_set_release(&dst_sev->migration_in_progress, 0);
1621  	return r;
1622  }
1623  
sev_unlock_two_vms(struct kvm * dst_kvm,struct kvm * src_kvm)1624  static void sev_unlock_two_vms(struct kvm *dst_kvm, struct kvm *src_kvm)
1625  {
1626  	struct kvm_sev_info *dst_sev = &to_kvm_svm(dst_kvm)->sev_info;
1627  	struct kvm_sev_info *src_sev = &to_kvm_svm(src_kvm)->sev_info;
1628  
1629  	mutex_unlock(&dst_kvm->lock);
1630  	mutex_unlock(&src_kvm->lock);
1631  	atomic_set_release(&dst_sev->migration_in_progress, 0);
1632  	atomic_set_release(&src_sev->migration_in_progress, 0);
1633  }
1634  
1635  /* vCPU mutex subclasses.  */
1636  enum sev_migration_role {
1637  	SEV_MIGRATION_SOURCE = 0,
1638  	SEV_MIGRATION_TARGET,
1639  	SEV_NR_MIGRATION_ROLES,
1640  };
1641  
sev_lock_vcpus_for_migration(struct kvm * kvm,enum sev_migration_role role)1642  static int sev_lock_vcpus_for_migration(struct kvm *kvm,
1643  					enum sev_migration_role role)
1644  {
1645  	struct kvm_vcpu *vcpu;
1646  	unsigned long i, j;
1647  
1648  	kvm_for_each_vcpu(i, vcpu, kvm) {
1649  		if (mutex_lock_killable_nested(&vcpu->mutex, role))
1650  			goto out_unlock;
1651  
1652  #ifdef CONFIG_PROVE_LOCKING
1653  		if (!i)
1654  			/*
1655  			 * Reset the role to one that avoids colliding with
1656  			 * the role used for the first vcpu mutex.
1657  			 */
1658  			role = SEV_NR_MIGRATION_ROLES;
1659  		else
1660  			mutex_release(&vcpu->mutex.dep_map, _THIS_IP_);
1661  #endif
1662  	}
1663  
1664  	return 0;
1665  
1666  out_unlock:
1667  
1668  	kvm_for_each_vcpu(j, vcpu, kvm) {
1669  		if (i == j)
1670  			break;
1671  
1672  #ifdef CONFIG_PROVE_LOCKING
1673  		if (j)
1674  			mutex_acquire(&vcpu->mutex.dep_map, role, 0, _THIS_IP_);
1675  #endif
1676  
1677  		mutex_unlock(&vcpu->mutex);
1678  	}
1679  	return -EINTR;
1680  }
1681  
sev_unlock_vcpus_for_migration(struct kvm * kvm)1682  static void sev_unlock_vcpus_for_migration(struct kvm *kvm)
1683  {
1684  	struct kvm_vcpu *vcpu;
1685  	unsigned long i;
1686  	bool first = true;
1687  
1688  	kvm_for_each_vcpu(i, vcpu, kvm) {
1689  		if (first)
1690  			first = false;
1691  		else
1692  			mutex_acquire(&vcpu->mutex.dep_map,
1693  				      SEV_NR_MIGRATION_ROLES, 0, _THIS_IP_);
1694  
1695  		mutex_unlock(&vcpu->mutex);
1696  	}
1697  }
1698  
sev_migrate_from(struct kvm * dst_kvm,struct kvm * src_kvm)1699  static void sev_migrate_from(struct kvm *dst_kvm, struct kvm *src_kvm)
1700  {
1701  	struct kvm_sev_info *dst = &to_kvm_svm(dst_kvm)->sev_info;
1702  	struct kvm_sev_info *src = &to_kvm_svm(src_kvm)->sev_info;
1703  	struct kvm_vcpu *dst_vcpu, *src_vcpu;
1704  	struct vcpu_svm *dst_svm, *src_svm;
1705  	struct kvm_sev_info *mirror;
1706  	unsigned long i;
1707  
1708  	dst->active = true;
1709  	dst->asid = src->asid;
1710  	dst->handle = src->handle;
1711  	dst->pages_locked = src->pages_locked;
1712  	dst->enc_context_owner = src->enc_context_owner;
1713  	dst->es_active = src->es_active;
1714  
1715  	src->asid = 0;
1716  	src->active = false;
1717  	src->handle = 0;
1718  	src->pages_locked = 0;
1719  	src->enc_context_owner = NULL;
1720  	src->es_active = false;
1721  
1722  	list_cut_before(&dst->regions_list, &src->regions_list, &src->regions_list);
1723  
1724  	/*
1725  	 * If this VM has mirrors, "transfer" each mirror's refcount of the
1726  	 * source to the destination (this KVM).  The caller holds a reference
1727  	 * to the source, so there's no danger of use-after-free.
1728  	 */
1729  	list_cut_before(&dst->mirror_vms, &src->mirror_vms, &src->mirror_vms);
1730  	list_for_each_entry(mirror, &dst->mirror_vms, mirror_entry) {
1731  		kvm_get_kvm(dst_kvm);
1732  		kvm_put_kvm(src_kvm);
1733  		mirror->enc_context_owner = dst_kvm;
1734  	}
1735  
1736  	/*
1737  	 * If this VM is a mirror, remove the old mirror from the owners list
1738  	 * and add the new mirror to the list.
1739  	 */
1740  	if (is_mirroring_enc_context(dst_kvm)) {
1741  		struct kvm_sev_info *owner_sev_info =
1742  			&to_kvm_svm(dst->enc_context_owner)->sev_info;
1743  
1744  		list_del(&src->mirror_entry);
1745  		list_add_tail(&dst->mirror_entry, &owner_sev_info->mirror_vms);
1746  	}
1747  
1748  	kvm_for_each_vcpu(i, dst_vcpu, dst_kvm) {
1749  		dst_svm = to_svm(dst_vcpu);
1750  
1751  		sev_init_vmcb(dst_svm);
1752  
1753  		if (!dst->es_active)
1754  			continue;
1755  
1756  		/*
1757  		 * Note, the source is not required to have the same number of
1758  		 * vCPUs as the destination when migrating a vanilla SEV VM.
1759  		 */
1760  		src_vcpu = kvm_get_vcpu(src_kvm, i);
1761  		src_svm = to_svm(src_vcpu);
1762  
1763  		/*
1764  		 * Transfer VMSA and GHCB state to the destination.  Nullify and
1765  		 * clear source fields as appropriate, the state now belongs to
1766  		 * the destination.
1767  		 */
1768  		memcpy(&dst_svm->sev_es, &src_svm->sev_es, sizeof(src_svm->sev_es));
1769  		dst_svm->vmcb->control.ghcb_gpa = src_svm->vmcb->control.ghcb_gpa;
1770  		dst_svm->vmcb->control.vmsa_pa = src_svm->vmcb->control.vmsa_pa;
1771  		dst_vcpu->arch.guest_state_protected = true;
1772  
1773  		memset(&src_svm->sev_es, 0, sizeof(src_svm->sev_es));
1774  		src_svm->vmcb->control.ghcb_gpa = INVALID_PAGE;
1775  		src_svm->vmcb->control.vmsa_pa = INVALID_PAGE;
1776  		src_vcpu->arch.guest_state_protected = false;
1777  	}
1778  }
1779  
sev_check_source_vcpus(struct kvm * dst,struct kvm * src)1780  static int sev_check_source_vcpus(struct kvm *dst, struct kvm *src)
1781  {
1782  	struct kvm_vcpu *src_vcpu;
1783  	unsigned long i;
1784  
1785  	if (!sev_es_guest(src))
1786  		return 0;
1787  
1788  	if (atomic_read(&src->online_vcpus) != atomic_read(&dst->online_vcpus))
1789  		return -EINVAL;
1790  
1791  	kvm_for_each_vcpu(i, src_vcpu, src) {
1792  		if (!src_vcpu->arch.guest_state_protected)
1793  			return -EINVAL;
1794  	}
1795  
1796  	return 0;
1797  }
1798  
sev_vm_move_enc_context_from(struct kvm * kvm,unsigned int source_fd)1799  int sev_vm_move_enc_context_from(struct kvm *kvm, unsigned int source_fd)
1800  {
1801  	struct kvm_sev_info *dst_sev = &to_kvm_svm(kvm)->sev_info;
1802  	struct kvm_sev_info *src_sev, *cg_cleanup_sev;
1803  	struct fd f = fdget(source_fd);
1804  	struct kvm *source_kvm;
1805  	bool charged = false;
1806  	int ret;
1807  
1808  	if (!f.file)
1809  		return -EBADF;
1810  
1811  	if (!file_is_kvm(f.file)) {
1812  		ret = -EBADF;
1813  		goto out_fput;
1814  	}
1815  
1816  	source_kvm = f.file->private_data;
1817  	ret = sev_lock_two_vms(kvm, source_kvm);
1818  	if (ret)
1819  		goto out_fput;
1820  
1821  	if (sev_guest(kvm) || !sev_guest(source_kvm)) {
1822  		ret = -EINVAL;
1823  		goto out_unlock;
1824  	}
1825  
1826  	src_sev = &to_kvm_svm(source_kvm)->sev_info;
1827  
1828  	dst_sev->misc_cg = get_current_misc_cg();
1829  	cg_cleanup_sev = dst_sev;
1830  	if (dst_sev->misc_cg != src_sev->misc_cg) {
1831  		ret = sev_misc_cg_try_charge(dst_sev);
1832  		if (ret)
1833  			goto out_dst_cgroup;
1834  		charged = true;
1835  	}
1836  
1837  	ret = sev_lock_vcpus_for_migration(kvm, SEV_MIGRATION_SOURCE);
1838  	if (ret)
1839  		goto out_dst_cgroup;
1840  	ret = sev_lock_vcpus_for_migration(source_kvm, SEV_MIGRATION_TARGET);
1841  	if (ret)
1842  		goto out_dst_vcpu;
1843  
1844  	ret = sev_check_source_vcpus(kvm, source_kvm);
1845  	if (ret)
1846  		goto out_source_vcpu;
1847  
1848  	sev_migrate_from(kvm, source_kvm);
1849  	kvm_vm_dead(source_kvm);
1850  	cg_cleanup_sev = src_sev;
1851  	ret = 0;
1852  
1853  out_source_vcpu:
1854  	sev_unlock_vcpus_for_migration(source_kvm);
1855  out_dst_vcpu:
1856  	sev_unlock_vcpus_for_migration(kvm);
1857  out_dst_cgroup:
1858  	/* Operates on the source on success, on the destination on failure.  */
1859  	if (charged)
1860  		sev_misc_cg_uncharge(cg_cleanup_sev);
1861  	put_misc_cg(cg_cleanup_sev->misc_cg);
1862  	cg_cleanup_sev->misc_cg = NULL;
1863  out_unlock:
1864  	sev_unlock_two_vms(kvm, source_kvm);
1865  out_fput:
1866  	fdput(f);
1867  	return ret;
1868  }
1869  
sev_mem_enc_ioctl(struct kvm * kvm,void __user * argp)1870  int sev_mem_enc_ioctl(struct kvm *kvm, void __user *argp)
1871  {
1872  	struct kvm_sev_cmd sev_cmd;
1873  	int r;
1874  
1875  	if (!sev_enabled)
1876  		return -ENOTTY;
1877  
1878  	if (!argp)
1879  		return 0;
1880  
1881  	if (copy_from_user(&sev_cmd, argp, sizeof(struct kvm_sev_cmd)))
1882  		return -EFAULT;
1883  
1884  	mutex_lock(&kvm->lock);
1885  
1886  	/* Only the enc_context_owner handles some memory enc operations. */
1887  	if (is_mirroring_enc_context(kvm) &&
1888  	    !is_cmd_allowed_from_mirror(sev_cmd.id)) {
1889  		r = -EINVAL;
1890  		goto out;
1891  	}
1892  
1893  	switch (sev_cmd.id) {
1894  	case KVM_SEV_ES_INIT:
1895  		if (!sev_es_enabled) {
1896  			r = -ENOTTY;
1897  			goto out;
1898  		}
1899  		fallthrough;
1900  	case KVM_SEV_INIT:
1901  		r = sev_guest_init(kvm, &sev_cmd);
1902  		break;
1903  	case KVM_SEV_LAUNCH_START:
1904  		r = sev_launch_start(kvm, &sev_cmd);
1905  		break;
1906  	case KVM_SEV_LAUNCH_UPDATE_DATA:
1907  		r = sev_launch_update_data(kvm, &sev_cmd);
1908  		break;
1909  	case KVM_SEV_LAUNCH_UPDATE_VMSA:
1910  		r = sev_launch_update_vmsa(kvm, &sev_cmd);
1911  		break;
1912  	case KVM_SEV_LAUNCH_MEASURE:
1913  		r = sev_launch_measure(kvm, &sev_cmd);
1914  		break;
1915  	case KVM_SEV_LAUNCH_FINISH:
1916  		r = sev_launch_finish(kvm, &sev_cmd);
1917  		break;
1918  	case KVM_SEV_GUEST_STATUS:
1919  		r = sev_guest_status(kvm, &sev_cmd);
1920  		break;
1921  	case KVM_SEV_DBG_DECRYPT:
1922  		r = sev_dbg_crypt(kvm, &sev_cmd, true);
1923  		break;
1924  	case KVM_SEV_DBG_ENCRYPT:
1925  		r = sev_dbg_crypt(kvm, &sev_cmd, false);
1926  		break;
1927  	case KVM_SEV_LAUNCH_SECRET:
1928  		r = sev_launch_secret(kvm, &sev_cmd);
1929  		break;
1930  	case KVM_SEV_GET_ATTESTATION_REPORT:
1931  		r = sev_get_attestation_report(kvm, &sev_cmd);
1932  		break;
1933  	case KVM_SEV_SEND_START:
1934  		r = sev_send_start(kvm, &sev_cmd);
1935  		break;
1936  	case KVM_SEV_SEND_UPDATE_DATA:
1937  		r = sev_send_update_data(kvm, &sev_cmd);
1938  		break;
1939  	case KVM_SEV_SEND_FINISH:
1940  		r = sev_send_finish(kvm, &sev_cmd);
1941  		break;
1942  	case KVM_SEV_SEND_CANCEL:
1943  		r = sev_send_cancel(kvm, &sev_cmd);
1944  		break;
1945  	case KVM_SEV_RECEIVE_START:
1946  		r = sev_receive_start(kvm, &sev_cmd);
1947  		break;
1948  	case KVM_SEV_RECEIVE_UPDATE_DATA:
1949  		r = sev_receive_update_data(kvm, &sev_cmd);
1950  		break;
1951  	case KVM_SEV_RECEIVE_FINISH:
1952  		r = sev_receive_finish(kvm, &sev_cmd);
1953  		break;
1954  	default:
1955  		r = -EINVAL;
1956  		goto out;
1957  	}
1958  
1959  	if (copy_to_user(argp, &sev_cmd, sizeof(struct kvm_sev_cmd)))
1960  		r = -EFAULT;
1961  
1962  out:
1963  	mutex_unlock(&kvm->lock);
1964  	return r;
1965  }
1966  
sev_mem_enc_register_region(struct kvm * kvm,struct kvm_enc_region * range)1967  int sev_mem_enc_register_region(struct kvm *kvm,
1968  				struct kvm_enc_region *range)
1969  {
1970  	struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1971  	struct enc_region *region;
1972  	int ret = 0;
1973  
1974  	if (!sev_guest(kvm))
1975  		return -ENOTTY;
1976  
1977  	/* If kvm is mirroring encryption context it isn't responsible for it */
1978  	if (is_mirroring_enc_context(kvm))
1979  		return -EINVAL;
1980  
1981  	if (range->addr > ULONG_MAX || range->size > ULONG_MAX)
1982  		return -EINVAL;
1983  
1984  	region = kzalloc(sizeof(*region), GFP_KERNEL_ACCOUNT);
1985  	if (!region)
1986  		return -ENOMEM;
1987  
1988  	mutex_lock(&kvm->lock);
1989  	region->pages = sev_pin_memory(kvm, range->addr, range->size, &region->npages, 1);
1990  	if (IS_ERR(region->pages)) {
1991  		ret = PTR_ERR(region->pages);
1992  		mutex_unlock(&kvm->lock);
1993  		goto e_free;
1994  	}
1995  
1996  	/*
1997  	 * The guest may change the memory encryption attribute from C=0 -> C=1
1998  	 * or vice versa for this memory range. Lets make sure caches are
1999  	 * flushed to ensure that guest data gets written into memory with
2000  	 * correct C-bit.  Note, this must be done before dropping kvm->lock,
2001  	 * as region and its array of pages can be freed by a different task
2002  	 * once kvm->lock is released.
2003  	 */
2004  	sev_clflush_pages(region->pages, region->npages);
2005  
2006  	region->uaddr = range->addr;
2007  	region->size = range->size;
2008  
2009  	list_add_tail(&region->list, &sev->regions_list);
2010  	mutex_unlock(&kvm->lock);
2011  
2012  	return ret;
2013  
2014  e_free:
2015  	kfree(region);
2016  	return ret;
2017  }
2018  
2019  static struct enc_region *
find_enc_region(struct kvm * kvm,struct kvm_enc_region * range)2020  find_enc_region(struct kvm *kvm, struct kvm_enc_region *range)
2021  {
2022  	struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
2023  	struct list_head *head = &sev->regions_list;
2024  	struct enc_region *i;
2025  
2026  	list_for_each_entry(i, head, list) {
2027  		if (i->uaddr == range->addr &&
2028  		    i->size == range->size)
2029  			return i;
2030  	}
2031  
2032  	return NULL;
2033  }
2034  
__unregister_enc_region_locked(struct kvm * kvm,struct enc_region * region)2035  static void __unregister_enc_region_locked(struct kvm *kvm,
2036  					   struct enc_region *region)
2037  {
2038  	sev_unpin_memory(kvm, region->pages, region->npages);
2039  	list_del(&region->list);
2040  	kfree(region);
2041  }
2042  
sev_mem_enc_unregister_region(struct kvm * kvm,struct kvm_enc_region * range)2043  int sev_mem_enc_unregister_region(struct kvm *kvm,
2044  				  struct kvm_enc_region *range)
2045  {
2046  	struct enc_region *region;
2047  	int ret;
2048  
2049  	/* If kvm is mirroring encryption context it isn't responsible for it */
2050  	if (is_mirroring_enc_context(kvm))
2051  		return -EINVAL;
2052  
2053  	mutex_lock(&kvm->lock);
2054  
2055  	if (!sev_guest(kvm)) {
2056  		ret = -ENOTTY;
2057  		goto failed;
2058  	}
2059  
2060  	region = find_enc_region(kvm, range);
2061  	if (!region) {
2062  		ret = -EINVAL;
2063  		goto failed;
2064  	}
2065  
2066  	/*
2067  	 * Ensure that all guest tagged cache entries are flushed before
2068  	 * releasing the pages back to the system for use. CLFLUSH will
2069  	 * not do this, so issue a WBINVD.
2070  	 */
2071  	wbinvd_on_all_cpus();
2072  
2073  	__unregister_enc_region_locked(kvm, region);
2074  
2075  	mutex_unlock(&kvm->lock);
2076  	return 0;
2077  
2078  failed:
2079  	mutex_unlock(&kvm->lock);
2080  	return ret;
2081  }
2082  
sev_vm_copy_enc_context_from(struct kvm * kvm,unsigned int source_fd)2083  int sev_vm_copy_enc_context_from(struct kvm *kvm, unsigned int source_fd)
2084  {
2085  	struct fd f = fdget(source_fd);
2086  	struct kvm *source_kvm;
2087  	struct kvm_sev_info *source_sev, *mirror_sev;
2088  	int ret;
2089  
2090  	if (!f.file)
2091  		return -EBADF;
2092  
2093  	if (!file_is_kvm(f.file)) {
2094  		ret = -EBADF;
2095  		goto e_source_fput;
2096  	}
2097  
2098  	source_kvm = f.file->private_data;
2099  	ret = sev_lock_two_vms(kvm, source_kvm);
2100  	if (ret)
2101  		goto e_source_fput;
2102  
2103  	/*
2104  	 * Mirrors of mirrors should work, but let's not get silly.  Also
2105  	 * disallow out-of-band SEV/SEV-ES init if the target is already an
2106  	 * SEV guest, or if vCPUs have been created.  KVM relies on vCPUs being
2107  	 * created after SEV/SEV-ES initialization, e.g. to init intercepts.
2108  	 */
2109  	if (sev_guest(kvm) || !sev_guest(source_kvm) ||
2110  	    is_mirroring_enc_context(source_kvm) || kvm->created_vcpus) {
2111  		ret = -EINVAL;
2112  		goto e_unlock;
2113  	}
2114  
2115  	/*
2116  	 * The mirror kvm holds an enc_context_owner ref so its asid can't
2117  	 * disappear until we're done with it
2118  	 */
2119  	source_sev = &to_kvm_svm(source_kvm)->sev_info;
2120  	kvm_get_kvm(source_kvm);
2121  	mirror_sev = &to_kvm_svm(kvm)->sev_info;
2122  	list_add_tail(&mirror_sev->mirror_entry, &source_sev->mirror_vms);
2123  
2124  	/* Set enc_context_owner and copy its encryption context over */
2125  	mirror_sev->enc_context_owner = source_kvm;
2126  	mirror_sev->active = true;
2127  	mirror_sev->asid = source_sev->asid;
2128  	mirror_sev->fd = source_sev->fd;
2129  	mirror_sev->es_active = source_sev->es_active;
2130  	mirror_sev->handle = source_sev->handle;
2131  	INIT_LIST_HEAD(&mirror_sev->regions_list);
2132  	INIT_LIST_HEAD(&mirror_sev->mirror_vms);
2133  	ret = 0;
2134  
2135  	/*
2136  	 * Do not copy ap_jump_table. Since the mirror does not share the same
2137  	 * KVM contexts as the original, and they may have different
2138  	 * memory-views.
2139  	 */
2140  
2141  e_unlock:
2142  	sev_unlock_two_vms(kvm, source_kvm);
2143  e_source_fput:
2144  	fdput(f);
2145  	return ret;
2146  }
2147  
sev_vm_destroy(struct kvm * kvm)2148  void sev_vm_destroy(struct kvm *kvm)
2149  {
2150  	struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
2151  	struct list_head *head = &sev->regions_list;
2152  	struct list_head *pos, *q;
2153  
2154  	if (!sev_guest(kvm))
2155  		return;
2156  
2157  	WARN_ON(!list_empty(&sev->mirror_vms));
2158  
2159  	/* If this is a mirror_kvm release the enc_context_owner and skip sev cleanup */
2160  	if (is_mirroring_enc_context(kvm)) {
2161  		struct kvm *owner_kvm = sev->enc_context_owner;
2162  
2163  		mutex_lock(&owner_kvm->lock);
2164  		list_del(&sev->mirror_entry);
2165  		mutex_unlock(&owner_kvm->lock);
2166  		kvm_put_kvm(owner_kvm);
2167  		return;
2168  	}
2169  
2170  	/*
2171  	 * Ensure that all guest tagged cache entries are flushed before
2172  	 * releasing the pages back to the system for use. CLFLUSH will
2173  	 * not do this, so issue a WBINVD.
2174  	 */
2175  	wbinvd_on_all_cpus();
2176  
2177  	/*
2178  	 * if userspace was terminated before unregistering the memory regions
2179  	 * then lets unpin all the registered memory.
2180  	 */
2181  	if (!list_empty(head)) {
2182  		list_for_each_safe(pos, q, head) {
2183  			__unregister_enc_region_locked(kvm,
2184  				list_entry(pos, struct enc_region, list));
2185  			cond_resched();
2186  		}
2187  	}
2188  
2189  	sev_unbind_asid(kvm, sev->handle);
2190  	sev_asid_free(sev);
2191  }
2192  
sev_set_cpu_caps(void)2193  void __init sev_set_cpu_caps(void)
2194  {
2195  	if (!sev_enabled)
2196  		kvm_cpu_cap_clear(X86_FEATURE_SEV);
2197  	if (!sev_es_enabled)
2198  		kvm_cpu_cap_clear(X86_FEATURE_SEV_ES);
2199  }
2200  
sev_hardware_setup(void)2201  void __init sev_hardware_setup(void)
2202  {
2203  #ifdef CONFIG_KVM_AMD_SEV
2204  	unsigned int eax, ebx, ecx, edx, sev_asid_count, sev_es_asid_count;
2205  	bool sev_es_supported = false;
2206  	bool sev_supported = false;
2207  
2208  	if (!sev_enabled || !npt_enabled || !nrips)
2209  		goto out;
2210  
2211  	/*
2212  	 * SEV must obviously be supported in hardware.  Sanity check that the
2213  	 * CPU supports decode assists, which is mandatory for SEV guests to
2214  	 * support instruction emulation.
2215  	 */
2216  	if (!boot_cpu_has(X86_FEATURE_SEV) ||
2217  	    WARN_ON_ONCE(!boot_cpu_has(X86_FEATURE_DECODEASSISTS)))
2218  		goto out;
2219  
2220  	/* Retrieve SEV CPUID information */
2221  	cpuid(0x8000001f, &eax, &ebx, &ecx, &edx);
2222  
2223  	/* Set encryption bit location for SEV-ES guests */
2224  	sev_enc_bit = ebx & 0x3f;
2225  
2226  	/* Maximum number of encrypted guests supported simultaneously */
2227  	max_sev_asid = ecx;
2228  	if (!max_sev_asid)
2229  		goto out;
2230  
2231  	/* Minimum ASID value that should be used for SEV guest */
2232  	min_sev_asid = edx;
2233  	sev_me_mask = 1UL << (ebx & 0x3f);
2234  
2235  	/*
2236  	 * Initialize SEV ASID bitmaps. Allocate space for ASID 0 in the bitmap,
2237  	 * even though it's never used, so that the bitmap is indexed by the
2238  	 * actual ASID.
2239  	 */
2240  	nr_asids = max_sev_asid + 1;
2241  	sev_asid_bitmap = bitmap_zalloc(nr_asids, GFP_KERNEL);
2242  	if (!sev_asid_bitmap)
2243  		goto out;
2244  
2245  	sev_reclaim_asid_bitmap = bitmap_zalloc(nr_asids, GFP_KERNEL);
2246  	if (!sev_reclaim_asid_bitmap) {
2247  		bitmap_free(sev_asid_bitmap);
2248  		sev_asid_bitmap = NULL;
2249  		goto out;
2250  	}
2251  
2252  	if (min_sev_asid <= max_sev_asid) {
2253  		sev_asid_count = max_sev_asid - min_sev_asid + 1;
2254  		WARN_ON_ONCE(misc_cg_set_capacity(MISC_CG_RES_SEV, sev_asid_count));
2255  	}
2256  	sev_supported = true;
2257  
2258  	/* SEV-ES support requested? */
2259  	if (!sev_es_enabled)
2260  		goto out;
2261  
2262  	/*
2263  	 * SEV-ES requires MMIO caching as KVM doesn't have access to the guest
2264  	 * instruction stream, i.e. can't emulate in response to a #NPF and
2265  	 * instead relies on #NPF(RSVD) being reflected into the guest as #VC
2266  	 * (the guest can then do a #VMGEXIT to request MMIO emulation).
2267  	 */
2268  	if (!enable_mmio_caching)
2269  		goto out;
2270  
2271  	/* Does the CPU support SEV-ES? */
2272  	if (!boot_cpu_has(X86_FEATURE_SEV_ES))
2273  		goto out;
2274  
2275  	if (!lbrv) {
2276  		WARN_ONCE(!boot_cpu_has(X86_FEATURE_LBRV),
2277  			  "LBRV must be present for SEV-ES support");
2278  		goto out;
2279  	}
2280  
2281  	/* Has the system been allocated ASIDs for SEV-ES? */
2282  	if (min_sev_asid == 1)
2283  		goto out;
2284  
2285  	sev_es_asid_count = min_sev_asid - 1;
2286  	WARN_ON_ONCE(misc_cg_set_capacity(MISC_CG_RES_SEV_ES, sev_es_asid_count));
2287  	sev_es_supported = true;
2288  
2289  out:
2290  	if (boot_cpu_has(X86_FEATURE_SEV))
2291  		pr_info("SEV %s (ASIDs %u - %u)\n",
2292  			sev_supported ? min_sev_asid <= max_sev_asid ? "enabled" :
2293  								       "unusable" :
2294  								       "disabled",
2295  			min_sev_asid, max_sev_asid);
2296  	if (boot_cpu_has(X86_FEATURE_SEV_ES))
2297  		pr_info("SEV-ES %s (ASIDs %u - %u)\n",
2298  			sev_es_supported ? "enabled" : "disabled",
2299  			min_sev_asid > 1 ? 1 : 0, min_sev_asid - 1);
2300  
2301  	sev_enabled = sev_supported;
2302  	sev_es_enabled = sev_es_supported;
2303  	if (!sev_es_enabled || !cpu_feature_enabled(X86_FEATURE_DEBUG_SWAP) ||
2304  	    !cpu_feature_enabled(X86_FEATURE_NO_NESTED_DATA_BP))
2305  		sev_es_debug_swap_enabled = false;
2306  #endif
2307  }
2308  
sev_hardware_unsetup(void)2309  void sev_hardware_unsetup(void)
2310  {
2311  	if (!sev_enabled)
2312  		return;
2313  
2314  	/* No need to take sev_bitmap_lock, all VMs have been destroyed. */
2315  	sev_flush_asids(1, max_sev_asid);
2316  
2317  	bitmap_free(sev_asid_bitmap);
2318  	bitmap_free(sev_reclaim_asid_bitmap);
2319  
2320  	misc_cg_set_capacity(MISC_CG_RES_SEV, 0);
2321  	misc_cg_set_capacity(MISC_CG_RES_SEV_ES, 0);
2322  }
2323  
sev_cpu_init(struct svm_cpu_data * sd)2324  int sev_cpu_init(struct svm_cpu_data *sd)
2325  {
2326  	if (!sev_enabled)
2327  		return 0;
2328  
2329  	sd->sev_vmcbs = kcalloc(nr_asids, sizeof(void *), GFP_KERNEL);
2330  	if (!sd->sev_vmcbs)
2331  		return -ENOMEM;
2332  
2333  	return 0;
2334  }
2335  
2336  /*
2337   * Pages used by hardware to hold guest encrypted state must be flushed before
2338   * returning them to the system.
2339   */
sev_flush_encrypted_page(struct kvm_vcpu * vcpu,void * va)2340  static void sev_flush_encrypted_page(struct kvm_vcpu *vcpu, void *va)
2341  {
2342  	unsigned int asid = sev_get_asid(vcpu->kvm);
2343  
2344  	/*
2345  	 * Note!  The address must be a kernel address, as regular page walk
2346  	 * checks are performed by VM_PAGE_FLUSH, i.e. operating on a user
2347  	 * address is non-deterministic and unsafe.  This function deliberately
2348  	 * takes a pointer to deter passing in a user address.
2349  	 */
2350  	unsigned long addr = (unsigned long)va;
2351  
2352  	/*
2353  	 * If CPU enforced cache coherency for encrypted mappings of the
2354  	 * same physical page is supported, use CLFLUSHOPT instead. NOTE: cache
2355  	 * flush is still needed in order to work properly with DMA devices.
2356  	 */
2357  	if (boot_cpu_has(X86_FEATURE_SME_COHERENT)) {
2358  		clflush_cache_range(va, PAGE_SIZE);
2359  		return;
2360  	}
2361  
2362  	/*
2363  	 * VM Page Flush takes a host virtual address and a guest ASID.  Fall
2364  	 * back to WBINVD if this faults so as not to make any problems worse
2365  	 * by leaving stale encrypted data in the cache.
2366  	 */
2367  	if (WARN_ON_ONCE(wrmsrl_safe(MSR_AMD64_VM_PAGE_FLUSH, addr | asid)))
2368  		goto do_wbinvd;
2369  
2370  	return;
2371  
2372  do_wbinvd:
2373  	wbinvd_on_all_cpus();
2374  }
2375  
sev_guest_memory_reclaimed(struct kvm * kvm)2376  void sev_guest_memory_reclaimed(struct kvm *kvm)
2377  {
2378  	if (!sev_guest(kvm))
2379  		return;
2380  
2381  	wbinvd_on_all_cpus();
2382  }
2383  
sev_free_vcpu(struct kvm_vcpu * vcpu)2384  void sev_free_vcpu(struct kvm_vcpu *vcpu)
2385  {
2386  	struct vcpu_svm *svm;
2387  
2388  	if (!sev_es_guest(vcpu->kvm))
2389  		return;
2390  
2391  	svm = to_svm(vcpu);
2392  
2393  	if (vcpu->arch.guest_state_protected)
2394  		sev_flush_encrypted_page(vcpu, svm->sev_es.vmsa);
2395  
2396  	__free_page(virt_to_page(svm->sev_es.vmsa));
2397  
2398  	if (svm->sev_es.ghcb_sa_free)
2399  		kvfree(svm->sev_es.ghcb_sa);
2400  }
2401  
dump_ghcb(struct vcpu_svm * svm)2402  static void dump_ghcb(struct vcpu_svm *svm)
2403  {
2404  	struct ghcb *ghcb = svm->sev_es.ghcb;
2405  	unsigned int nbits;
2406  
2407  	/* Re-use the dump_invalid_vmcb module parameter */
2408  	if (!dump_invalid_vmcb) {
2409  		pr_warn_ratelimited("set kvm_amd.dump_invalid_vmcb=1 to dump internal KVM state.\n");
2410  		return;
2411  	}
2412  
2413  	nbits = sizeof(ghcb->save.valid_bitmap) * 8;
2414  
2415  	pr_err("GHCB (GPA=%016llx):\n", svm->vmcb->control.ghcb_gpa);
2416  	pr_err("%-20s%016llx is_valid: %u\n", "sw_exit_code",
2417  	       ghcb->save.sw_exit_code, ghcb_sw_exit_code_is_valid(ghcb));
2418  	pr_err("%-20s%016llx is_valid: %u\n", "sw_exit_info_1",
2419  	       ghcb->save.sw_exit_info_1, ghcb_sw_exit_info_1_is_valid(ghcb));
2420  	pr_err("%-20s%016llx is_valid: %u\n", "sw_exit_info_2",
2421  	       ghcb->save.sw_exit_info_2, ghcb_sw_exit_info_2_is_valid(ghcb));
2422  	pr_err("%-20s%016llx is_valid: %u\n", "sw_scratch",
2423  	       ghcb->save.sw_scratch, ghcb_sw_scratch_is_valid(ghcb));
2424  	pr_err("%-20s%*pb\n", "valid_bitmap", nbits, ghcb->save.valid_bitmap);
2425  }
2426  
sev_es_sync_to_ghcb(struct vcpu_svm * svm)2427  static void sev_es_sync_to_ghcb(struct vcpu_svm *svm)
2428  {
2429  	struct kvm_vcpu *vcpu = &svm->vcpu;
2430  	struct ghcb *ghcb = svm->sev_es.ghcb;
2431  
2432  	/*
2433  	 * The GHCB protocol so far allows for the following data
2434  	 * to be returned:
2435  	 *   GPRs RAX, RBX, RCX, RDX
2436  	 *
2437  	 * Copy their values, even if they may not have been written during the
2438  	 * VM-Exit.  It's the guest's responsibility to not consume random data.
2439  	 */
2440  	ghcb_set_rax(ghcb, vcpu->arch.regs[VCPU_REGS_RAX]);
2441  	ghcb_set_rbx(ghcb, vcpu->arch.regs[VCPU_REGS_RBX]);
2442  	ghcb_set_rcx(ghcb, vcpu->arch.regs[VCPU_REGS_RCX]);
2443  	ghcb_set_rdx(ghcb, vcpu->arch.regs[VCPU_REGS_RDX]);
2444  }
2445  
sev_es_sync_from_ghcb(struct vcpu_svm * svm)2446  static void sev_es_sync_from_ghcb(struct vcpu_svm *svm)
2447  {
2448  	struct vmcb_control_area *control = &svm->vmcb->control;
2449  	struct kvm_vcpu *vcpu = &svm->vcpu;
2450  	struct ghcb *ghcb = svm->sev_es.ghcb;
2451  	u64 exit_code;
2452  
2453  	/*
2454  	 * The GHCB protocol so far allows for the following data
2455  	 * to be supplied:
2456  	 *   GPRs RAX, RBX, RCX, RDX
2457  	 *   XCR0
2458  	 *   CPL
2459  	 *
2460  	 * VMMCALL allows the guest to provide extra registers. KVM also
2461  	 * expects RSI for hypercalls, so include that, too.
2462  	 *
2463  	 * Copy their values to the appropriate location if supplied.
2464  	 */
2465  	memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
2466  
2467  	BUILD_BUG_ON(sizeof(svm->sev_es.valid_bitmap) != sizeof(ghcb->save.valid_bitmap));
2468  	memcpy(&svm->sev_es.valid_bitmap, &ghcb->save.valid_bitmap, sizeof(ghcb->save.valid_bitmap));
2469  
2470  	vcpu->arch.regs[VCPU_REGS_RAX] = kvm_ghcb_get_rax_if_valid(svm, ghcb);
2471  	vcpu->arch.regs[VCPU_REGS_RBX] = kvm_ghcb_get_rbx_if_valid(svm, ghcb);
2472  	vcpu->arch.regs[VCPU_REGS_RCX] = kvm_ghcb_get_rcx_if_valid(svm, ghcb);
2473  	vcpu->arch.regs[VCPU_REGS_RDX] = kvm_ghcb_get_rdx_if_valid(svm, ghcb);
2474  	vcpu->arch.regs[VCPU_REGS_RSI] = kvm_ghcb_get_rsi_if_valid(svm, ghcb);
2475  
2476  	svm->vmcb->save.cpl = kvm_ghcb_get_cpl_if_valid(svm, ghcb);
2477  
2478  	if (kvm_ghcb_xcr0_is_valid(svm)) {
2479  		vcpu->arch.xcr0 = ghcb_get_xcr0(ghcb);
2480  		kvm_update_cpuid_runtime(vcpu);
2481  	}
2482  
2483  	/* Copy the GHCB exit information into the VMCB fields */
2484  	exit_code = ghcb_get_sw_exit_code(ghcb);
2485  	control->exit_code = lower_32_bits(exit_code);
2486  	control->exit_code_hi = upper_32_bits(exit_code);
2487  	control->exit_info_1 = ghcb_get_sw_exit_info_1(ghcb);
2488  	control->exit_info_2 = ghcb_get_sw_exit_info_2(ghcb);
2489  	svm->sev_es.sw_scratch = kvm_ghcb_get_sw_scratch_if_valid(svm, ghcb);
2490  
2491  	/* Clear the valid entries fields */
2492  	memset(ghcb->save.valid_bitmap, 0, sizeof(ghcb->save.valid_bitmap));
2493  }
2494  
kvm_ghcb_get_sw_exit_code(struct vmcb_control_area * control)2495  static u64 kvm_ghcb_get_sw_exit_code(struct vmcb_control_area *control)
2496  {
2497  	return (((u64)control->exit_code_hi) << 32) | control->exit_code;
2498  }
2499  
sev_es_validate_vmgexit(struct vcpu_svm * svm)2500  static int sev_es_validate_vmgexit(struct vcpu_svm *svm)
2501  {
2502  	struct vmcb_control_area *control = &svm->vmcb->control;
2503  	struct kvm_vcpu *vcpu = &svm->vcpu;
2504  	u64 exit_code;
2505  	u64 reason;
2506  
2507  	/*
2508  	 * Retrieve the exit code now even though it may not be marked valid
2509  	 * as it could help with debugging.
2510  	 */
2511  	exit_code = kvm_ghcb_get_sw_exit_code(control);
2512  
2513  	/* Only GHCB Usage code 0 is supported */
2514  	if (svm->sev_es.ghcb->ghcb_usage) {
2515  		reason = GHCB_ERR_INVALID_USAGE;
2516  		goto vmgexit_err;
2517  	}
2518  
2519  	reason = GHCB_ERR_MISSING_INPUT;
2520  
2521  	if (!kvm_ghcb_sw_exit_code_is_valid(svm) ||
2522  	    !kvm_ghcb_sw_exit_info_1_is_valid(svm) ||
2523  	    !kvm_ghcb_sw_exit_info_2_is_valid(svm))
2524  		goto vmgexit_err;
2525  
2526  	switch (exit_code) {
2527  	case SVM_EXIT_READ_DR7:
2528  		break;
2529  	case SVM_EXIT_WRITE_DR7:
2530  		if (!kvm_ghcb_rax_is_valid(svm))
2531  			goto vmgexit_err;
2532  		break;
2533  	case SVM_EXIT_RDTSC:
2534  		break;
2535  	case SVM_EXIT_RDPMC:
2536  		if (!kvm_ghcb_rcx_is_valid(svm))
2537  			goto vmgexit_err;
2538  		break;
2539  	case SVM_EXIT_CPUID:
2540  		if (!kvm_ghcb_rax_is_valid(svm) ||
2541  		    !kvm_ghcb_rcx_is_valid(svm))
2542  			goto vmgexit_err;
2543  		if (vcpu->arch.regs[VCPU_REGS_RAX] == 0xd)
2544  			if (!kvm_ghcb_xcr0_is_valid(svm))
2545  				goto vmgexit_err;
2546  		break;
2547  	case SVM_EXIT_INVD:
2548  		break;
2549  	case SVM_EXIT_IOIO:
2550  		if (control->exit_info_1 & SVM_IOIO_STR_MASK) {
2551  			if (!kvm_ghcb_sw_scratch_is_valid(svm))
2552  				goto vmgexit_err;
2553  		} else {
2554  			if (!(control->exit_info_1 & SVM_IOIO_TYPE_MASK))
2555  				if (!kvm_ghcb_rax_is_valid(svm))
2556  					goto vmgexit_err;
2557  		}
2558  		break;
2559  	case SVM_EXIT_MSR:
2560  		if (!kvm_ghcb_rcx_is_valid(svm))
2561  			goto vmgexit_err;
2562  		if (control->exit_info_1) {
2563  			if (!kvm_ghcb_rax_is_valid(svm) ||
2564  			    !kvm_ghcb_rdx_is_valid(svm))
2565  				goto vmgexit_err;
2566  		}
2567  		break;
2568  	case SVM_EXIT_VMMCALL:
2569  		if (!kvm_ghcb_rax_is_valid(svm) ||
2570  		    !kvm_ghcb_cpl_is_valid(svm))
2571  			goto vmgexit_err;
2572  		break;
2573  	case SVM_EXIT_RDTSCP:
2574  		break;
2575  	case SVM_EXIT_WBINVD:
2576  		break;
2577  	case SVM_EXIT_MONITOR:
2578  		if (!kvm_ghcb_rax_is_valid(svm) ||
2579  		    !kvm_ghcb_rcx_is_valid(svm) ||
2580  		    !kvm_ghcb_rdx_is_valid(svm))
2581  			goto vmgexit_err;
2582  		break;
2583  	case SVM_EXIT_MWAIT:
2584  		if (!kvm_ghcb_rax_is_valid(svm) ||
2585  		    !kvm_ghcb_rcx_is_valid(svm))
2586  			goto vmgexit_err;
2587  		break;
2588  	case SVM_VMGEXIT_MMIO_READ:
2589  	case SVM_VMGEXIT_MMIO_WRITE:
2590  		if (!kvm_ghcb_sw_scratch_is_valid(svm))
2591  			goto vmgexit_err;
2592  		break;
2593  	case SVM_VMGEXIT_NMI_COMPLETE:
2594  	case SVM_VMGEXIT_AP_HLT_LOOP:
2595  	case SVM_VMGEXIT_AP_JUMP_TABLE:
2596  	case SVM_VMGEXIT_UNSUPPORTED_EVENT:
2597  		break;
2598  	default:
2599  		reason = GHCB_ERR_INVALID_EVENT;
2600  		goto vmgexit_err;
2601  	}
2602  
2603  	return 0;
2604  
2605  vmgexit_err:
2606  	if (reason == GHCB_ERR_INVALID_USAGE) {
2607  		vcpu_unimpl(vcpu, "vmgexit: ghcb usage %#x is not valid\n",
2608  			    svm->sev_es.ghcb->ghcb_usage);
2609  	} else if (reason == GHCB_ERR_INVALID_EVENT) {
2610  		vcpu_unimpl(vcpu, "vmgexit: exit code %#llx is not valid\n",
2611  			    exit_code);
2612  	} else {
2613  		vcpu_unimpl(vcpu, "vmgexit: exit code %#llx input is not valid\n",
2614  			    exit_code);
2615  		dump_ghcb(svm);
2616  	}
2617  
2618  	ghcb_set_sw_exit_info_1(svm->sev_es.ghcb, 2);
2619  	ghcb_set_sw_exit_info_2(svm->sev_es.ghcb, reason);
2620  
2621  	/* Resume the guest to "return" the error code. */
2622  	return 1;
2623  }
2624  
sev_es_unmap_ghcb(struct vcpu_svm * svm)2625  void sev_es_unmap_ghcb(struct vcpu_svm *svm)
2626  {
2627  	if (!svm->sev_es.ghcb)
2628  		return;
2629  
2630  	if (svm->sev_es.ghcb_sa_free) {
2631  		/*
2632  		 * The scratch area lives outside the GHCB, so there is a
2633  		 * buffer that, depending on the operation performed, may
2634  		 * need to be synced, then freed.
2635  		 */
2636  		if (svm->sev_es.ghcb_sa_sync) {
2637  			kvm_write_guest(svm->vcpu.kvm,
2638  					svm->sev_es.sw_scratch,
2639  					svm->sev_es.ghcb_sa,
2640  					svm->sev_es.ghcb_sa_len);
2641  			svm->sev_es.ghcb_sa_sync = false;
2642  		}
2643  
2644  		kvfree(svm->sev_es.ghcb_sa);
2645  		svm->sev_es.ghcb_sa = NULL;
2646  		svm->sev_es.ghcb_sa_free = false;
2647  	}
2648  
2649  	trace_kvm_vmgexit_exit(svm->vcpu.vcpu_id, svm->sev_es.ghcb);
2650  
2651  	sev_es_sync_to_ghcb(svm);
2652  
2653  	kvm_vcpu_unmap(&svm->vcpu, &svm->sev_es.ghcb_map, true);
2654  	svm->sev_es.ghcb = NULL;
2655  }
2656  
pre_sev_run(struct vcpu_svm * svm,int cpu)2657  void pre_sev_run(struct vcpu_svm *svm, int cpu)
2658  {
2659  	struct svm_cpu_data *sd = per_cpu_ptr(&svm_data, cpu);
2660  	unsigned int asid = sev_get_asid(svm->vcpu.kvm);
2661  
2662  	/* Assign the asid allocated with this SEV guest */
2663  	svm->asid = asid;
2664  
2665  	/*
2666  	 * Flush guest TLB:
2667  	 *
2668  	 * 1) when different VMCB for the same ASID is to be run on the same host CPU.
2669  	 * 2) or this VMCB was executed on different host CPU in previous VMRUNs.
2670  	 */
2671  	if (sd->sev_vmcbs[asid] == svm->vmcb &&
2672  	    svm->vcpu.arch.last_vmentry_cpu == cpu)
2673  		return;
2674  
2675  	sd->sev_vmcbs[asid] = svm->vmcb;
2676  	svm->vmcb->control.tlb_ctl = TLB_CONTROL_FLUSH_ASID;
2677  	vmcb_mark_dirty(svm->vmcb, VMCB_ASID);
2678  }
2679  
2680  #define GHCB_SCRATCH_AREA_LIMIT		(16ULL * PAGE_SIZE)
setup_vmgexit_scratch(struct vcpu_svm * svm,bool sync,u64 len)2681  static int setup_vmgexit_scratch(struct vcpu_svm *svm, bool sync, u64 len)
2682  {
2683  	struct vmcb_control_area *control = &svm->vmcb->control;
2684  	u64 ghcb_scratch_beg, ghcb_scratch_end;
2685  	u64 scratch_gpa_beg, scratch_gpa_end;
2686  	void *scratch_va;
2687  
2688  	scratch_gpa_beg = svm->sev_es.sw_scratch;
2689  	if (!scratch_gpa_beg) {
2690  		pr_err("vmgexit: scratch gpa not provided\n");
2691  		goto e_scratch;
2692  	}
2693  
2694  	scratch_gpa_end = scratch_gpa_beg + len;
2695  	if (scratch_gpa_end < scratch_gpa_beg) {
2696  		pr_err("vmgexit: scratch length (%#llx) not valid for scratch address (%#llx)\n",
2697  		       len, scratch_gpa_beg);
2698  		goto e_scratch;
2699  	}
2700  
2701  	if ((scratch_gpa_beg & PAGE_MASK) == control->ghcb_gpa) {
2702  		/* Scratch area begins within GHCB */
2703  		ghcb_scratch_beg = control->ghcb_gpa +
2704  				   offsetof(struct ghcb, shared_buffer);
2705  		ghcb_scratch_end = control->ghcb_gpa +
2706  				   offsetof(struct ghcb, reserved_0xff0);
2707  
2708  		/*
2709  		 * If the scratch area begins within the GHCB, it must be
2710  		 * completely contained in the GHCB shared buffer area.
2711  		 */
2712  		if (scratch_gpa_beg < ghcb_scratch_beg ||
2713  		    scratch_gpa_end > ghcb_scratch_end) {
2714  			pr_err("vmgexit: scratch area is outside of GHCB shared buffer area (%#llx - %#llx)\n",
2715  			       scratch_gpa_beg, scratch_gpa_end);
2716  			goto e_scratch;
2717  		}
2718  
2719  		scratch_va = (void *)svm->sev_es.ghcb;
2720  		scratch_va += (scratch_gpa_beg - control->ghcb_gpa);
2721  	} else {
2722  		/*
2723  		 * The guest memory must be read into a kernel buffer, so
2724  		 * limit the size
2725  		 */
2726  		if (len > GHCB_SCRATCH_AREA_LIMIT) {
2727  			pr_err("vmgexit: scratch area exceeds KVM limits (%#llx requested, %#llx limit)\n",
2728  			       len, GHCB_SCRATCH_AREA_LIMIT);
2729  			goto e_scratch;
2730  		}
2731  		scratch_va = kvzalloc(len, GFP_KERNEL_ACCOUNT);
2732  		if (!scratch_va)
2733  			return -ENOMEM;
2734  
2735  		if (kvm_read_guest(svm->vcpu.kvm, scratch_gpa_beg, scratch_va, len)) {
2736  			/* Unable to copy scratch area from guest */
2737  			pr_err("vmgexit: kvm_read_guest for scratch area failed\n");
2738  
2739  			kvfree(scratch_va);
2740  			return -EFAULT;
2741  		}
2742  
2743  		/*
2744  		 * The scratch area is outside the GHCB. The operation will
2745  		 * dictate whether the buffer needs to be synced before running
2746  		 * the vCPU next time (i.e. a read was requested so the data
2747  		 * must be written back to the guest memory).
2748  		 */
2749  		svm->sev_es.ghcb_sa_sync = sync;
2750  		svm->sev_es.ghcb_sa_free = true;
2751  	}
2752  
2753  	svm->sev_es.ghcb_sa = scratch_va;
2754  	svm->sev_es.ghcb_sa_len = len;
2755  
2756  	return 0;
2757  
2758  e_scratch:
2759  	ghcb_set_sw_exit_info_1(svm->sev_es.ghcb, 2);
2760  	ghcb_set_sw_exit_info_2(svm->sev_es.ghcb, GHCB_ERR_INVALID_SCRATCH_AREA);
2761  
2762  	return 1;
2763  }
2764  
set_ghcb_msr_bits(struct vcpu_svm * svm,u64 value,u64 mask,unsigned int pos)2765  static void set_ghcb_msr_bits(struct vcpu_svm *svm, u64 value, u64 mask,
2766  			      unsigned int pos)
2767  {
2768  	svm->vmcb->control.ghcb_gpa &= ~(mask << pos);
2769  	svm->vmcb->control.ghcb_gpa |= (value & mask) << pos;
2770  }
2771  
get_ghcb_msr_bits(struct vcpu_svm * svm,u64 mask,unsigned int pos)2772  static u64 get_ghcb_msr_bits(struct vcpu_svm *svm, u64 mask, unsigned int pos)
2773  {
2774  	return (svm->vmcb->control.ghcb_gpa >> pos) & mask;
2775  }
2776  
set_ghcb_msr(struct vcpu_svm * svm,u64 value)2777  static void set_ghcb_msr(struct vcpu_svm *svm, u64 value)
2778  {
2779  	svm->vmcb->control.ghcb_gpa = value;
2780  }
2781  
sev_handle_vmgexit_msr_protocol(struct vcpu_svm * svm)2782  static int sev_handle_vmgexit_msr_protocol(struct vcpu_svm *svm)
2783  {
2784  	struct vmcb_control_area *control = &svm->vmcb->control;
2785  	struct kvm_vcpu *vcpu = &svm->vcpu;
2786  	u64 ghcb_info;
2787  	int ret = 1;
2788  
2789  	ghcb_info = control->ghcb_gpa & GHCB_MSR_INFO_MASK;
2790  
2791  	trace_kvm_vmgexit_msr_protocol_enter(svm->vcpu.vcpu_id,
2792  					     control->ghcb_gpa);
2793  
2794  	switch (ghcb_info) {
2795  	case GHCB_MSR_SEV_INFO_REQ:
2796  		set_ghcb_msr(svm, GHCB_MSR_SEV_INFO(GHCB_VERSION_MAX,
2797  						    GHCB_VERSION_MIN,
2798  						    sev_enc_bit));
2799  		break;
2800  	case GHCB_MSR_CPUID_REQ: {
2801  		u64 cpuid_fn, cpuid_reg, cpuid_value;
2802  
2803  		cpuid_fn = get_ghcb_msr_bits(svm,
2804  					     GHCB_MSR_CPUID_FUNC_MASK,
2805  					     GHCB_MSR_CPUID_FUNC_POS);
2806  
2807  		/* Initialize the registers needed by the CPUID intercept */
2808  		vcpu->arch.regs[VCPU_REGS_RAX] = cpuid_fn;
2809  		vcpu->arch.regs[VCPU_REGS_RCX] = 0;
2810  
2811  		ret = svm_invoke_exit_handler(vcpu, SVM_EXIT_CPUID);
2812  		if (!ret) {
2813  			/* Error, keep GHCB MSR value as-is */
2814  			break;
2815  		}
2816  
2817  		cpuid_reg = get_ghcb_msr_bits(svm,
2818  					      GHCB_MSR_CPUID_REG_MASK,
2819  					      GHCB_MSR_CPUID_REG_POS);
2820  		if (cpuid_reg == 0)
2821  			cpuid_value = vcpu->arch.regs[VCPU_REGS_RAX];
2822  		else if (cpuid_reg == 1)
2823  			cpuid_value = vcpu->arch.regs[VCPU_REGS_RBX];
2824  		else if (cpuid_reg == 2)
2825  			cpuid_value = vcpu->arch.regs[VCPU_REGS_RCX];
2826  		else
2827  			cpuid_value = vcpu->arch.regs[VCPU_REGS_RDX];
2828  
2829  		set_ghcb_msr_bits(svm, cpuid_value,
2830  				  GHCB_MSR_CPUID_VALUE_MASK,
2831  				  GHCB_MSR_CPUID_VALUE_POS);
2832  
2833  		set_ghcb_msr_bits(svm, GHCB_MSR_CPUID_RESP,
2834  				  GHCB_MSR_INFO_MASK,
2835  				  GHCB_MSR_INFO_POS);
2836  		break;
2837  	}
2838  	case GHCB_MSR_TERM_REQ: {
2839  		u64 reason_set, reason_code;
2840  
2841  		reason_set = get_ghcb_msr_bits(svm,
2842  					       GHCB_MSR_TERM_REASON_SET_MASK,
2843  					       GHCB_MSR_TERM_REASON_SET_POS);
2844  		reason_code = get_ghcb_msr_bits(svm,
2845  						GHCB_MSR_TERM_REASON_MASK,
2846  						GHCB_MSR_TERM_REASON_POS);
2847  		pr_info("SEV-ES guest requested termination: %#llx:%#llx\n",
2848  			reason_set, reason_code);
2849  
2850  		vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
2851  		vcpu->run->system_event.type = KVM_SYSTEM_EVENT_SEV_TERM;
2852  		vcpu->run->system_event.ndata = 1;
2853  		vcpu->run->system_event.data[0] = control->ghcb_gpa;
2854  
2855  		return 0;
2856  	}
2857  	default:
2858  		/* Error, keep GHCB MSR value as-is */
2859  		break;
2860  	}
2861  
2862  	trace_kvm_vmgexit_msr_protocol_exit(svm->vcpu.vcpu_id,
2863  					    control->ghcb_gpa, ret);
2864  
2865  	return ret;
2866  }
2867  
sev_handle_vmgexit(struct kvm_vcpu * vcpu)2868  int sev_handle_vmgexit(struct kvm_vcpu *vcpu)
2869  {
2870  	struct vcpu_svm *svm = to_svm(vcpu);
2871  	struct vmcb_control_area *control = &svm->vmcb->control;
2872  	u64 ghcb_gpa, exit_code;
2873  	int ret;
2874  
2875  	/* Validate the GHCB */
2876  	ghcb_gpa = control->ghcb_gpa;
2877  	if (ghcb_gpa & GHCB_MSR_INFO_MASK)
2878  		return sev_handle_vmgexit_msr_protocol(svm);
2879  
2880  	if (!ghcb_gpa) {
2881  		vcpu_unimpl(vcpu, "vmgexit: GHCB gpa is not set\n");
2882  
2883  		/* Without a GHCB, just return right back to the guest */
2884  		return 1;
2885  	}
2886  
2887  	if (kvm_vcpu_map(vcpu, ghcb_gpa >> PAGE_SHIFT, &svm->sev_es.ghcb_map)) {
2888  		/* Unable to map GHCB from guest */
2889  		vcpu_unimpl(vcpu, "vmgexit: error mapping GHCB [%#llx] from guest\n",
2890  			    ghcb_gpa);
2891  
2892  		/* Without a GHCB, just return right back to the guest */
2893  		return 1;
2894  	}
2895  
2896  	svm->sev_es.ghcb = svm->sev_es.ghcb_map.hva;
2897  
2898  	trace_kvm_vmgexit_enter(vcpu->vcpu_id, svm->sev_es.ghcb);
2899  
2900  	sev_es_sync_from_ghcb(svm);
2901  	ret = sev_es_validate_vmgexit(svm);
2902  	if (ret)
2903  		return ret;
2904  
2905  	ghcb_set_sw_exit_info_1(svm->sev_es.ghcb, 0);
2906  	ghcb_set_sw_exit_info_2(svm->sev_es.ghcb, 0);
2907  
2908  	exit_code = kvm_ghcb_get_sw_exit_code(control);
2909  	switch (exit_code) {
2910  	case SVM_VMGEXIT_MMIO_READ:
2911  		ret = setup_vmgexit_scratch(svm, true, control->exit_info_2);
2912  		if (ret)
2913  			break;
2914  
2915  		ret = kvm_sev_es_mmio_read(vcpu,
2916  					   control->exit_info_1,
2917  					   control->exit_info_2,
2918  					   svm->sev_es.ghcb_sa);
2919  		break;
2920  	case SVM_VMGEXIT_MMIO_WRITE:
2921  		ret = setup_vmgexit_scratch(svm, false, control->exit_info_2);
2922  		if (ret)
2923  			break;
2924  
2925  		ret = kvm_sev_es_mmio_write(vcpu,
2926  					    control->exit_info_1,
2927  					    control->exit_info_2,
2928  					    svm->sev_es.ghcb_sa);
2929  		break;
2930  	case SVM_VMGEXIT_NMI_COMPLETE:
2931  		++vcpu->stat.nmi_window_exits;
2932  		svm->nmi_masked = false;
2933  		kvm_make_request(KVM_REQ_EVENT, vcpu);
2934  		ret = 1;
2935  		break;
2936  	case SVM_VMGEXIT_AP_HLT_LOOP:
2937  		ret = kvm_emulate_ap_reset_hold(vcpu);
2938  		break;
2939  	case SVM_VMGEXIT_AP_JUMP_TABLE: {
2940  		struct kvm_sev_info *sev = &to_kvm_svm(vcpu->kvm)->sev_info;
2941  
2942  		switch (control->exit_info_1) {
2943  		case 0:
2944  			/* Set AP jump table address */
2945  			sev->ap_jump_table = control->exit_info_2;
2946  			break;
2947  		case 1:
2948  			/* Get AP jump table address */
2949  			ghcb_set_sw_exit_info_2(svm->sev_es.ghcb, sev->ap_jump_table);
2950  			break;
2951  		default:
2952  			pr_err("svm: vmgexit: unsupported AP jump table request - exit_info_1=%#llx\n",
2953  			       control->exit_info_1);
2954  			ghcb_set_sw_exit_info_1(svm->sev_es.ghcb, 2);
2955  			ghcb_set_sw_exit_info_2(svm->sev_es.ghcb, GHCB_ERR_INVALID_INPUT);
2956  		}
2957  
2958  		ret = 1;
2959  		break;
2960  	}
2961  	case SVM_VMGEXIT_UNSUPPORTED_EVENT:
2962  		vcpu_unimpl(vcpu,
2963  			    "vmgexit: unsupported event - exit_info_1=%#llx, exit_info_2=%#llx\n",
2964  			    control->exit_info_1, control->exit_info_2);
2965  		ret = -EINVAL;
2966  		break;
2967  	default:
2968  		ret = svm_invoke_exit_handler(vcpu, exit_code);
2969  	}
2970  
2971  	return ret;
2972  }
2973  
sev_es_string_io(struct vcpu_svm * svm,int size,unsigned int port,int in)2974  int sev_es_string_io(struct vcpu_svm *svm, int size, unsigned int port, int in)
2975  {
2976  	int count;
2977  	int bytes;
2978  	int r;
2979  
2980  	if (svm->vmcb->control.exit_info_2 > INT_MAX)
2981  		return -EINVAL;
2982  
2983  	count = svm->vmcb->control.exit_info_2;
2984  	if (unlikely(check_mul_overflow(count, size, &bytes)))
2985  		return -EINVAL;
2986  
2987  	r = setup_vmgexit_scratch(svm, in, bytes);
2988  	if (r)
2989  		return r;
2990  
2991  	return kvm_sev_es_string_io(&svm->vcpu, size, port, svm->sev_es.ghcb_sa,
2992  				    count, in);
2993  }
2994  
sev_es_vcpu_after_set_cpuid(struct vcpu_svm * svm)2995  static void sev_es_vcpu_after_set_cpuid(struct vcpu_svm *svm)
2996  {
2997  	struct kvm_vcpu *vcpu = &svm->vcpu;
2998  
2999  	if (boot_cpu_has(X86_FEATURE_V_TSC_AUX)) {
3000  		bool v_tsc_aux = guest_cpuid_has(vcpu, X86_FEATURE_RDTSCP) ||
3001  				 guest_cpuid_has(vcpu, X86_FEATURE_RDPID);
3002  
3003  		set_msr_interception(vcpu, svm->msrpm, MSR_TSC_AUX, v_tsc_aux, v_tsc_aux);
3004  	}
3005  
3006  	/*
3007  	 * For SEV-ES, accesses to MSR_IA32_XSS should not be intercepted if
3008  	 * the host/guest supports its use.
3009  	 *
3010  	 * guest_can_use() checks a number of requirements on the host/guest to
3011  	 * ensure that MSR_IA32_XSS is available, but it might report true even
3012  	 * if X86_FEATURE_XSAVES isn't configured in the guest to ensure host
3013  	 * MSR_IA32_XSS is always properly restored. For SEV-ES, it is better
3014  	 * to further check that the guest CPUID actually supports
3015  	 * X86_FEATURE_XSAVES so that accesses to MSR_IA32_XSS by misbehaved
3016  	 * guests will still get intercepted and caught in the normal
3017  	 * kvm_emulate_rdmsr()/kvm_emulated_wrmsr() paths.
3018  	 */
3019  	if (guest_can_use(vcpu, X86_FEATURE_XSAVES) &&
3020  	    guest_cpuid_has(vcpu, X86_FEATURE_XSAVES))
3021  		set_msr_interception(vcpu, svm->msrpm, MSR_IA32_XSS, 1, 1);
3022  	else
3023  		set_msr_interception(vcpu, svm->msrpm, MSR_IA32_XSS, 0, 0);
3024  }
3025  
sev_vcpu_after_set_cpuid(struct vcpu_svm * svm)3026  void sev_vcpu_after_set_cpuid(struct vcpu_svm *svm)
3027  {
3028  	struct kvm_vcpu *vcpu = &svm->vcpu;
3029  	struct kvm_cpuid_entry2 *best;
3030  
3031  	/* For sev guests, the memory encryption bit is not reserved in CR3.  */
3032  	best = kvm_find_cpuid_entry(vcpu, 0x8000001F);
3033  	if (best)
3034  		vcpu->arch.reserved_gpa_bits &= ~(1UL << (best->ebx & 0x3f));
3035  
3036  	if (sev_es_guest(svm->vcpu.kvm))
3037  		sev_es_vcpu_after_set_cpuid(svm);
3038  }
3039  
sev_es_init_vmcb(struct vcpu_svm * svm)3040  static void sev_es_init_vmcb(struct vcpu_svm *svm)
3041  {
3042  	struct vmcb *vmcb = svm->vmcb01.ptr;
3043  	struct kvm_vcpu *vcpu = &svm->vcpu;
3044  
3045  	svm->vmcb->control.nested_ctl |= SVM_NESTED_CTL_SEV_ES_ENABLE;
3046  
3047  	/*
3048  	 * An SEV-ES guest requires a VMSA area that is a separate from the
3049  	 * VMCB page. Do not include the encryption mask on the VMSA physical
3050  	 * address since hardware will access it using the guest key.  Note,
3051  	 * the VMSA will be NULL if this vCPU is the destination for intrahost
3052  	 * migration, and will be copied later.
3053  	 */
3054  	if (svm->sev_es.vmsa)
3055  		svm->vmcb->control.vmsa_pa = __pa(svm->sev_es.vmsa);
3056  
3057  	/* Can't intercept CR register access, HV can't modify CR registers */
3058  	svm_clr_intercept(svm, INTERCEPT_CR0_READ);
3059  	svm_clr_intercept(svm, INTERCEPT_CR4_READ);
3060  	svm_clr_intercept(svm, INTERCEPT_CR8_READ);
3061  	svm_clr_intercept(svm, INTERCEPT_CR0_WRITE);
3062  	svm_clr_intercept(svm, INTERCEPT_CR4_WRITE);
3063  	svm_clr_intercept(svm, INTERCEPT_CR8_WRITE);
3064  
3065  	svm_clr_intercept(svm, INTERCEPT_SELECTIVE_CR0);
3066  
3067  	/* Track EFER/CR register changes */
3068  	svm_set_intercept(svm, TRAP_EFER_WRITE);
3069  	svm_set_intercept(svm, TRAP_CR0_WRITE);
3070  	svm_set_intercept(svm, TRAP_CR4_WRITE);
3071  	svm_set_intercept(svm, TRAP_CR8_WRITE);
3072  
3073  	vmcb->control.intercepts[INTERCEPT_DR] = 0;
3074  	if (!sev_es_debug_swap_enabled) {
3075  		vmcb_set_intercept(&vmcb->control, INTERCEPT_DR7_READ);
3076  		vmcb_set_intercept(&vmcb->control, INTERCEPT_DR7_WRITE);
3077  		recalc_intercepts(svm);
3078  	} else {
3079  		/*
3080  		 * Disable #DB intercept iff DebugSwap is enabled.  KVM doesn't
3081  		 * allow debugging SEV-ES guests, and enables DebugSwap iff
3082  		 * NO_NESTED_DATA_BP is supported, so there's no reason to
3083  		 * intercept #DB when DebugSwap is enabled.  For simplicity
3084  		 * with respect to guest debug, intercept #DB for other VMs
3085  		 * even if NO_NESTED_DATA_BP is supported, i.e. even if the
3086  		 * guest can't DoS the CPU with infinite #DB vectoring.
3087  		 */
3088  		clr_exception_intercept(svm, DB_VECTOR);
3089  	}
3090  
3091  	/* Can't intercept XSETBV, HV can't modify XCR0 directly */
3092  	svm_clr_intercept(svm, INTERCEPT_XSETBV);
3093  
3094  	/* Clear intercepts on selected MSRs */
3095  	set_msr_interception(vcpu, svm->msrpm, MSR_EFER, 1, 1);
3096  	set_msr_interception(vcpu, svm->msrpm, MSR_IA32_CR_PAT, 1, 1);
3097  }
3098  
sev_init_vmcb(struct vcpu_svm * svm)3099  void sev_init_vmcb(struct vcpu_svm *svm)
3100  {
3101  	svm->vmcb->control.nested_ctl |= SVM_NESTED_CTL_SEV_ENABLE;
3102  	clr_exception_intercept(svm, UD_VECTOR);
3103  
3104  	/*
3105  	 * Don't intercept #GP for SEV guests, e.g. for the VMware backdoor, as
3106  	 * KVM can't decrypt guest memory to decode the faulting instruction.
3107  	 */
3108  	clr_exception_intercept(svm, GP_VECTOR);
3109  
3110  	if (sev_es_guest(svm->vcpu.kvm))
3111  		sev_es_init_vmcb(svm);
3112  }
3113  
sev_es_vcpu_reset(struct vcpu_svm * svm)3114  void sev_es_vcpu_reset(struct vcpu_svm *svm)
3115  {
3116  	/*
3117  	 * Set the GHCB MSR value as per the GHCB specification when emulating
3118  	 * vCPU RESET for an SEV-ES guest.
3119  	 */
3120  	set_ghcb_msr(svm, GHCB_MSR_SEV_INFO(GHCB_VERSION_MAX,
3121  					    GHCB_VERSION_MIN,
3122  					    sev_enc_bit));
3123  }
3124  
sev_es_prepare_switch_to_guest(struct sev_es_save_area * hostsa)3125  void sev_es_prepare_switch_to_guest(struct sev_es_save_area *hostsa)
3126  {
3127  	/*
3128  	 * All host state for SEV-ES guests is categorized into three swap types
3129  	 * based on how it is handled by hardware during a world switch:
3130  	 *
3131  	 * A: VMRUN:   Host state saved in host save area
3132  	 *    VMEXIT:  Host state loaded from host save area
3133  	 *
3134  	 * B: VMRUN:   Host state _NOT_ saved in host save area
3135  	 *    VMEXIT:  Host state loaded from host save area
3136  	 *
3137  	 * C: VMRUN:   Host state _NOT_ saved in host save area
3138  	 *    VMEXIT:  Host state initialized to default(reset) values
3139  	 *
3140  	 * Manually save type-B state, i.e. state that is loaded by VMEXIT but
3141  	 * isn't saved by VMRUN, that isn't already saved by VMSAVE (performed
3142  	 * by common SVM code).
3143  	 */
3144  	hostsa->xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
3145  	hostsa->pkru = read_pkru();
3146  	hostsa->xss = host_xss;
3147  
3148  	/*
3149  	 * If DebugSwap is enabled, debug registers are loaded but NOT saved by
3150  	 * the CPU (Type-B). If DebugSwap is disabled/unsupported, the CPU both
3151  	 * saves and loads debug registers (Type-A).
3152  	 */
3153  	if (sev_es_debug_swap_enabled) {
3154  		hostsa->dr0 = native_get_debugreg(0);
3155  		hostsa->dr1 = native_get_debugreg(1);
3156  		hostsa->dr2 = native_get_debugreg(2);
3157  		hostsa->dr3 = native_get_debugreg(3);
3158  		hostsa->dr0_addr_mask = amd_get_dr_addr_mask(0);
3159  		hostsa->dr1_addr_mask = amd_get_dr_addr_mask(1);
3160  		hostsa->dr2_addr_mask = amd_get_dr_addr_mask(2);
3161  		hostsa->dr3_addr_mask = amd_get_dr_addr_mask(3);
3162  	}
3163  }
3164  
sev_vcpu_deliver_sipi_vector(struct kvm_vcpu * vcpu,u8 vector)3165  void sev_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
3166  {
3167  	struct vcpu_svm *svm = to_svm(vcpu);
3168  
3169  	/* First SIPI: Use the values as initially set by the VMM */
3170  	if (!svm->sev_es.received_first_sipi) {
3171  		svm->sev_es.received_first_sipi = true;
3172  		return;
3173  	}
3174  
3175  	/*
3176  	 * Subsequent SIPI: Return from an AP Reset Hold VMGEXIT, where
3177  	 * the guest will set the CS and RIP. Set SW_EXIT_INFO_2 to a
3178  	 * non-zero value.
3179  	 */
3180  	if (!svm->sev_es.ghcb)
3181  		return;
3182  
3183  	ghcb_set_sw_exit_info_2(svm->sev_es.ghcb, 1);
3184  }
3185