1 // SPDX-License-Identifier: GPL-2.0-only
3 * Kernel-based Virtual Machine driver for Linux
7 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
10 #include <linux/kvm_types.h>
11 #include <linux/kvm_host.h>
12 #include <linux/kernel.h>
13 #include <linux/highmem.h>
14 #include <linux/psp-sev.h>
15 #include <linux/pagemap.h>
16 #include <linux/swap.h>
17 #include <linux/misc_cgroup.h>
18 #include <linux/processor.h>
19 #include <linux/trace_events.h>
22 #include <asm/trapnr.h>
23 #include <asm/fpu/xcr.h>
31 #ifndef CONFIG_KVM_AMD_SEV
33 * When this config is not defined, SEV feature is not supported and APIs in
34 * this file are not used but this file still gets compiled into the KVM AMD
37 * We will not have MISC_CG_RES_SEV and MISC_CG_RES_SEV_ES entries in the enum
38 * misc_res_type {} defined in linux/misc_cgroup.h.
40 * Below macros allow compilation to succeed.
42 #define MISC_CG_RES_SEV MISC_CG_RES_TYPES
43 #define MISC_CG_RES_SEV_ES MISC_CG_RES_TYPES
46 #ifdef CONFIG_KVM_AMD_SEV
47 /* enable/disable SEV support */
48 static bool sev_enabled = true;
49 module_param_named(sev, sev_enabled, bool, 0444);
51 /* enable/disable SEV-ES support */
52 static bool sev_es_enabled = true;
53 module_param_named(sev_es, sev_es_enabled, bool, 0444);
55 #define sev_enabled false
56 #define sev_es_enabled false
57 #endif /* CONFIG_KVM_AMD_SEV */
59 static u8 sev_enc_bit;
60 static DECLARE_RWSEM(sev_deactivate_lock);
61 static DEFINE_MUTEX(sev_bitmap_lock);
62 unsigned int max_sev_asid;
63 static unsigned int min_sev_asid;
64 static unsigned long sev_me_mask;
65 static unsigned int nr_asids;
66 static unsigned long *sev_asid_bitmap;
67 static unsigned long *sev_reclaim_asid_bitmap;
70 struct list_head list;
77 /* Called with the sev_bitmap_lock held, or on shutdown */
78 static int sev_flush_asids(int min_asid, int max_asid)
80 int ret, asid, error = 0;
82 /* Check if there are any ASIDs to reclaim before performing a flush */
83 asid = find_next_bit(sev_reclaim_asid_bitmap, nr_asids, min_asid);
88 * DEACTIVATE will clear the WBINVD indicator causing DF_FLUSH to fail,
89 * so it must be guarded.
91 down_write(&sev_deactivate_lock);
94 ret = sev_guest_df_flush(&error);
96 up_write(&sev_deactivate_lock);
99 pr_err("SEV: DF_FLUSH failed, ret=%d, error=%#x\n", ret, error);
104 static inline bool is_mirroring_enc_context(struct kvm *kvm)
106 return !!to_kvm_svm(kvm)->sev_info.enc_context_owner;
109 /* Must be called with the sev_bitmap_lock held */
110 static bool __sev_recycle_asids(int min_asid, int max_asid)
112 if (sev_flush_asids(min_asid, max_asid))
115 /* The flush process will flush all reclaimable SEV and SEV-ES ASIDs */
116 bitmap_xor(sev_asid_bitmap, sev_asid_bitmap, sev_reclaim_asid_bitmap,
118 bitmap_zero(sev_reclaim_asid_bitmap, nr_asids);
123 static int sev_misc_cg_try_charge(struct kvm_sev_info *sev)
125 enum misc_res_type type = sev->es_active ? MISC_CG_RES_SEV_ES : MISC_CG_RES_SEV;
126 return misc_cg_try_charge(type, sev->misc_cg, 1);
129 static void sev_misc_cg_uncharge(struct kvm_sev_info *sev)
131 enum misc_res_type type = sev->es_active ? MISC_CG_RES_SEV_ES : MISC_CG_RES_SEV;
132 misc_cg_uncharge(type, sev->misc_cg, 1);
135 static int sev_asid_new(struct kvm_sev_info *sev)
137 int asid, min_asid, max_asid, ret;
140 WARN_ON(sev->misc_cg);
141 sev->misc_cg = get_current_misc_cg();
142 ret = sev_misc_cg_try_charge(sev);
144 put_misc_cg(sev->misc_cg);
149 mutex_lock(&sev_bitmap_lock);
152 * SEV-enabled guests must use asid from min_sev_asid to max_sev_asid.
153 * SEV-ES-enabled guest can use from 1 to min_sev_asid - 1.
155 min_asid = sev->es_active ? 1 : min_sev_asid;
156 max_asid = sev->es_active ? min_sev_asid - 1 : max_sev_asid;
158 asid = find_next_zero_bit(sev_asid_bitmap, max_asid + 1, min_asid);
159 if (asid > max_asid) {
160 if (retry && __sev_recycle_asids(min_asid, max_asid)) {
164 mutex_unlock(&sev_bitmap_lock);
169 __set_bit(asid, sev_asid_bitmap);
171 mutex_unlock(&sev_bitmap_lock);
175 sev_misc_cg_uncharge(sev);
176 put_misc_cg(sev->misc_cg);
181 static int sev_get_asid(struct kvm *kvm)
183 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
188 static void sev_asid_free(struct kvm_sev_info *sev)
190 struct svm_cpu_data *sd;
193 mutex_lock(&sev_bitmap_lock);
195 __set_bit(sev->asid, sev_reclaim_asid_bitmap);
197 for_each_possible_cpu(cpu) {
198 sd = per_cpu(svm_data, cpu);
199 sd->sev_vmcbs[sev->asid] = NULL;
202 mutex_unlock(&sev_bitmap_lock);
204 sev_misc_cg_uncharge(sev);
205 put_misc_cg(sev->misc_cg);
209 static void sev_decommission(unsigned int handle)
211 struct sev_data_decommission decommission;
216 decommission.handle = handle;
217 sev_guest_decommission(&decommission, NULL);
220 static void sev_unbind_asid(struct kvm *kvm, unsigned int handle)
222 struct sev_data_deactivate deactivate;
227 deactivate.handle = handle;
229 /* Guard DEACTIVATE against WBINVD/DF_FLUSH used in ASID recycling */
230 down_read(&sev_deactivate_lock);
231 sev_guest_deactivate(&deactivate, NULL);
232 up_read(&sev_deactivate_lock);
234 sev_decommission(handle);
237 static int sev_guest_init(struct kvm *kvm, struct kvm_sev_cmd *argp)
239 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
242 if (kvm->created_vcpus)
246 if (unlikely(sev->active))
250 sev->es_active = argp->id == KVM_SEV_ES_INIT;
251 asid = sev_asid_new(sev);
256 ret = sev_platform_init(&argp->error);
260 INIT_LIST_HEAD(&sev->regions_list);
261 INIT_LIST_HEAD(&sev->mirror_vms);
263 kvm_set_apicv_inhibit(kvm, APICV_INHIBIT_REASON_SEV);
271 sev->es_active = false;
276 static int sev_bind_asid(struct kvm *kvm, unsigned int handle, int *error)
278 struct sev_data_activate activate;
279 int asid = sev_get_asid(kvm);
282 /* activate ASID on the given handle */
283 activate.handle = handle;
284 activate.asid = asid;
285 ret = sev_guest_activate(&activate, error);
290 static int __sev_issue_cmd(int fd, int id, void *data, int *error)
299 ret = sev_issue_cmd_external_user(f.file, id, data, error);
305 static int sev_issue_cmd(struct kvm *kvm, int id, void *data, int *error)
307 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
309 return __sev_issue_cmd(sev->fd, id, data, error);
312 static int sev_launch_start(struct kvm *kvm, struct kvm_sev_cmd *argp)
314 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
315 struct sev_data_launch_start start;
316 struct kvm_sev_launch_start params;
317 void *dh_blob, *session_blob;
318 int *error = &argp->error;
324 if (copy_from_user(¶ms, (void __user *)(uintptr_t)argp->data, sizeof(params)))
327 memset(&start, 0, sizeof(start));
330 if (params.dh_uaddr) {
331 dh_blob = psp_copy_user_blob(params.dh_uaddr, params.dh_len);
333 return PTR_ERR(dh_blob);
335 start.dh_cert_address = __sme_set(__pa(dh_blob));
336 start.dh_cert_len = params.dh_len;
340 if (params.session_uaddr) {
341 session_blob = psp_copy_user_blob(params.session_uaddr, params.session_len);
342 if (IS_ERR(session_blob)) {
343 ret = PTR_ERR(session_blob);
347 start.session_address = __sme_set(__pa(session_blob));
348 start.session_len = params.session_len;
351 start.handle = params.handle;
352 start.policy = params.policy;
354 /* create memory encryption context */
355 ret = __sev_issue_cmd(argp->sev_fd, SEV_CMD_LAUNCH_START, &start, error);
359 /* Bind ASID to this guest */
360 ret = sev_bind_asid(kvm, start.handle, error);
362 sev_decommission(start.handle);
366 /* return handle to userspace */
367 params.handle = start.handle;
368 if (copy_to_user((void __user *)(uintptr_t)argp->data, ¶ms, sizeof(params))) {
369 sev_unbind_asid(kvm, start.handle);
374 sev->handle = start.handle;
375 sev->fd = argp->sev_fd;
384 static struct page **sev_pin_memory(struct kvm *kvm, unsigned long uaddr,
385 unsigned long ulen, unsigned long *n,
388 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
389 unsigned long npages, size;
391 unsigned long locked, lock_limit;
393 unsigned long first, last;
396 lockdep_assert_held(&kvm->lock);
398 if (ulen == 0 || uaddr + ulen < uaddr)
399 return ERR_PTR(-EINVAL);
401 /* Calculate number of pages. */
402 first = (uaddr & PAGE_MASK) >> PAGE_SHIFT;
403 last = ((uaddr + ulen - 1) & PAGE_MASK) >> PAGE_SHIFT;
404 npages = (last - first + 1);
406 locked = sev->pages_locked + npages;
407 lock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
408 if (locked > lock_limit && !capable(CAP_IPC_LOCK)) {
409 pr_err("SEV: %lu locked pages exceed the lock limit of %lu.\n", locked, lock_limit);
410 return ERR_PTR(-ENOMEM);
413 if (WARN_ON_ONCE(npages > INT_MAX))
414 return ERR_PTR(-EINVAL);
416 /* Avoid using vmalloc for smaller buffers. */
417 size = npages * sizeof(struct page *);
418 if (size > PAGE_SIZE)
419 pages = __vmalloc(size, GFP_KERNEL_ACCOUNT | __GFP_ZERO);
421 pages = kmalloc(size, GFP_KERNEL_ACCOUNT);
424 return ERR_PTR(-ENOMEM);
426 /* Pin the user virtual address. */
427 npinned = pin_user_pages_fast(uaddr, npages, write ? FOLL_WRITE : 0, pages);
428 if (npinned != npages) {
429 pr_err("SEV: Failure locking %lu pages.\n", npages);
435 sev->pages_locked = locked;
441 unpin_user_pages(pages, npinned);
447 static void sev_unpin_memory(struct kvm *kvm, struct page **pages,
448 unsigned long npages)
450 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
452 unpin_user_pages(pages, npages);
454 sev->pages_locked -= npages;
457 static void sev_clflush_pages(struct page *pages[], unsigned long npages)
459 uint8_t *page_virtual;
462 if (this_cpu_has(X86_FEATURE_SME_COHERENT) || npages == 0 ||
466 for (i = 0; i < npages; i++) {
467 page_virtual = kmap_atomic(pages[i]);
468 clflush_cache_range(page_virtual, PAGE_SIZE);
469 kunmap_atomic(page_virtual);
474 static unsigned long get_num_contig_pages(unsigned long idx,
475 struct page **inpages, unsigned long npages)
477 unsigned long paddr, next_paddr;
478 unsigned long i = idx + 1, pages = 1;
480 /* find the number of contiguous pages starting from idx */
481 paddr = __sme_page_pa(inpages[idx]);
483 next_paddr = __sme_page_pa(inpages[i++]);
484 if ((paddr + PAGE_SIZE) == next_paddr) {
495 static int sev_launch_update_data(struct kvm *kvm, struct kvm_sev_cmd *argp)
497 unsigned long vaddr, vaddr_end, next_vaddr, npages, pages, size, i;
498 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
499 struct kvm_sev_launch_update_data params;
500 struct sev_data_launch_update_data data;
501 struct page **inpages;
507 if (copy_from_user(¶ms, (void __user *)(uintptr_t)argp->data, sizeof(params)))
510 vaddr = params.uaddr;
512 vaddr_end = vaddr + size;
514 /* Lock the user memory. */
515 inpages = sev_pin_memory(kvm, vaddr, size, &npages, 1);
517 return PTR_ERR(inpages);
520 * Flush (on non-coherent CPUs) before LAUNCH_UPDATE encrypts pages in
521 * place; the cache may contain the data that was written unencrypted.
523 sev_clflush_pages(inpages, npages);
526 data.handle = sev->handle;
528 for (i = 0; vaddr < vaddr_end; vaddr = next_vaddr, i += pages) {
532 * If the user buffer is not page-aligned, calculate the offset
535 offset = vaddr & (PAGE_SIZE - 1);
537 /* Calculate the number of pages that can be encrypted in one go. */
538 pages = get_num_contig_pages(i, inpages, npages);
540 len = min_t(size_t, ((pages * PAGE_SIZE) - offset), size);
543 data.address = __sme_page_pa(inpages[i]) + offset;
544 ret = sev_issue_cmd(kvm, SEV_CMD_LAUNCH_UPDATE_DATA, &data, &argp->error);
549 next_vaddr = vaddr + len;
553 /* content of memory is updated, mark pages dirty */
554 for (i = 0; i < npages; i++) {
555 set_page_dirty_lock(inpages[i]);
556 mark_page_accessed(inpages[i]);
558 /* unlock the user pages */
559 sev_unpin_memory(kvm, inpages, npages);
563 static int sev_es_sync_vmsa(struct vcpu_svm *svm)
565 struct sev_es_save_area *save = svm->sev_es.vmsa;
567 /* Check some debug related fields before encrypting the VMSA */
568 if (svm->vcpu.guest_debug || (svm->vmcb->save.dr7 & ~DR7_FIXED_1))
572 * SEV-ES will use a VMSA that is pointed to by the VMCB, not
573 * the traditional VMSA that is part of the VMCB. Copy the
574 * traditional VMSA as it has been built so far (in prep
575 * for LAUNCH_UPDATE_VMSA) to be the initial SEV-ES state.
577 memcpy(save, &svm->vmcb->save, sizeof(svm->vmcb->save));
579 /* Sync registgers */
580 save->rax = svm->vcpu.arch.regs[VCPU_REGS_RAX];
581 save->rbx = svm->vcpu.arch.regs[VCPU_REGS_RBX];
582 save->rcx = svm->vcpu.arch.regs[VCPU_REGS_RCX];
583 save->rdx = svm->vcpu.arch.regs[VCPU_REGS_RDX];
584 save->rsp = svm->vcpu.arch.regs[VCPU_REGS_RSP];
585 save->rbp = svm->vcpu.arch.regs[VCPU_REGS_RBP];
586 save->rsi = svm->vcpu.arch.regs[VCPU_REGS_RSI];
587 save->rdi = svm->vcpu.arch.regs[VCPU_REGS_RDI];
589 save->r8 = svm->vcpu.arch.regs[VCPU_REGS_R8];
590 save->r9 = svm->vcpu.arch.regs[VCPU_REGS_R9];
591 save->r10 = svm->vcpu.arch.regs[VCPU_REGS_R10];
592 save->r11 = svm->vcpu.arch.regs[VCPU_REGS_R11];
593 save->r12 = svm->vcpu.arch.regs[VCPU_REGS_R12];
594 save->r13 = svm->vcpu.arch.regs[VCPU_REGS_R13];
595 save->r14 = svm->vcpu.arch.regs[VCPU_REGS_R14];
596 save->r15 = svm->vcpu.arch.regs[VCPU_REGS_R15];
598 save->rip = svm->vcpu.arch.regs[VCPU_REGS_RIP];
600 /* Sync some non-GPR registers before encrypting */
601 save->xcr0 = svm->vcpu.arch.xcr0;
602 save->pkru = svm->vcpu.arch.pkru;
603 save->xss = svm->vcpu.arch.ia32_xss;
604 save->dr6 = svm->vcpu.arch.dr6;
609 static int __sev_launch_update_vmsa(struct kvm *kvm, struct kvm_vcpu *vcpu,
612 struct sev_data_launch_update_vmsa vmsa;
613 struct vcpu_svm *svm = to_svm(vcpu);
616 /* Perform some pre-encryption checks against the VMSA */
617 ret = sev_es_sync_vmsa(svm);
622 * The LAUNCH_UPDATE_VMSA command will perform in-place encryption of
623 * the VMSA memory content (i.e it will write the same memory region
624 * with the guest's key), so invalidate it first.
626 clflush_cache_range(svm->sev_es.vmsa, PAGE_SIZE);
629 vmsa.handle = to_kvm_svm(kvm)->sev_info.handle;
630 vmsa.address = __sme_pa(svm->sev_es.vmsa);
631 vmsa.len = PAGE_SIZE;
632 ret = sev_issue_cmd(kvm, SEV_CMD_LAUNCH_UPDATE_VMSA, &vmsa, error);
636 vcpu->arch.guest_state_protected = true;
640 static int sev_launch_update_vmsa(struct kvm *kvm, struct kvm_sev_cmd *argp)
642 struct kvm_vcpu *vcpu;
646 if (!sev_es_guest(kvm))
649 kvm_for_each_vcpu(i, vcpu, kvm) {
650 ret = mutex_lock_killable(&vcpu->mutex);
654 ret = __sev_launch_update_vmsa(kvm, vcpu, &argp->error);
656 mutex_unlock(&vcpu->mutex);
664 static int sev_launch_measure(struct kvm *kvm, struct kvm_sev_cmd *argp)
666 void __user *measure = (void __user *)(uintptr_t)argp->data;
667 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
668 struct sev_data_launch_measure data;
669 struct kvm_sev_launch_measure params;
670 void __user *p = NULL;
677 if (copy_from_user(¶ms, measure, sizeof(params)))
680 memset(&data, 0, sizeof(data));
682 /* User wants to query the blob length */
686 p = (void __user *)(uintptr_t)params.uaddr;
688 if (params.len > SEV_FW_BLOB_MAX_SIZE)
691 blob = kzalloc(params.len, GFP_KERNEL_ACCOUNT);
695 data.address = __psp_pa(blob);
696 data.len = params.len;
700 data.handle = sev->handle;
701 ret = sev_issue_cmd(kvm, SEV_CMD_LAUNCH_MEASURE, &data, &argp->error);
704 * If we query the session length, FW responded with expected data.
713 if (copy_to_user(p, blob, params.len))
718 params.len = data.len;
719 if (copy_to_user(measure, ¶ms, sizeof(params)))
726 static int sev_launch_finish(struct kvm *kvm, struct kvm_sev_cmd *argp)
728 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
729 struct sev_data_launch_finish data;
734 data.handle = sev->handle;
735 return sev_issue_cmd(kvm, SEV_CMD_LAUNCH_FINISH, &data, &argp->error);
738 static int sev_guest_status(struct kvm *kvm, struct kvm_sev_cmd *argp)
740 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
741 struct kvm_sev_guest_status params;
742 struct sev_data_guest_status data;
748 memset(&data, 0, sizeof(data));
750 data.handle = sev->handle;
751 ret = sev_issue_cmd(kvm, SEV_CMD_GUEST_STATUS, &data, &argp->error);
755 params.policy = data.policy;
756 params.state = data.state;
757 params.handle = data.handle;
759 if (copy_to_user((void __user *)(uintptr_t)argp->data, ¶ms, sizeof(params)))
765 static int __sev_issue_dbg_cmd(struct kvm *kvm, unsigned long src,
766 unsigned long dst, int size,
767 int *error, bool enc)
769 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
770 struct sev_data_dbg data;
773 data.handle = sev->handle;
778 return sev_issue_cmd(kvm,
779 enc ? SEV_CMD_DBG_ENCRYPT : SEV_CMD_DBG_DECRYPT,
783 static int __sev_dbg_decrypt(struct kvm *kvm, unsigned long src_paddr,
784 unsigned long dst_paddr, int sz, int *err)
789 * Its safe to read more than we are asked, caller should ensure that
790 * destination has enough space.
792 offset = src_paddr & 15;
793 src_paddr = round_down(src_paddr, 16);
794 sz = round_up(sz + offset, 16);
796 return __sev_issue_dbg_cmd(kvm, src_paddr, dst_paddr, sz, err, false);
799 static int __sev_dbg_decrypt_user(struct kvm *kvm, unsigned long paddr,
800 void __user *dst_uaddr,
801 unsigned long dst_paddr,
804 struct page *tpage = NULL;
807 /* if inputs are not 16-byte then use intermediate buffer */
808 if (!IS_ALIGNED(dst_paddr, 16) ||
809 !IS_ALIGNED(paddr, 16) ||
810 !IS_ALIGNED(size, 16)) {
811 tpage = (void *)alloc_page(GFP_KERNEL | __GFP_ZERO);
815 dst_paddr = __sme_page_pa(tpage);
818 ret = __sev_dbg_decrypt(kvm, paddr, dst_paddr, size, err);
824 if (copy_to_user(dst_uaddr, page_address(tpage) + offset, size))
835 static int __sev_dbg_encrypt_user(struct kvm *kvm, unsigned long paddr,
837 unsigned long dst_paddr,
838 void __user *dst_vaddr,
839 int size, int *error)
841 struct page *src_tpage = NULL;
842 struct page *dst_tpage = NULL;
845 /* If source buffer is not aligned then use an intermediate buffer */
846 if (!IS_ALIGNED((unsigned long)vaddr, 16)) {
847 src_tpage = alloc_page(GFP_KERNEL_ACCOUNT);
851 if (copy_from_user(page_address(src_tpage), vaddr, size)) {
852 __free_page(src_tpage);
856 paddr = __sme_page_pa(src_tpage);
860 * If destination buffer or length is not aligned then do read-modify-write:
861 * - decrypt destination in an intermediate buffer
862 * - copy the source buffer in an intermediate buffer
863 * - use the intermediate buffer as source buffer
865 if (!IS_ALIGNED((unsigned long)dst_vaddr, 16) || !IS_ALIGNED(size, 16)) {
868 dst_tpage = alloc_page(GFP_KERNEL_ACCOUNT);
874 ret = __sev_dbg_decrypt(kvm, dst_paddr,
875 __sme_page_pa(dst_tpage), size, error);
880 * If source is kernel buffer then use memcpy() otherwise
883 dst_offset = dst_paddr & 15;
886 memcpy(page_address(dst_tpage) + dst_offset,
887 page_address(src_tpage), size);
889 if (copy_from_user(page_address(dst_tpage) + dst_offset,
896 paddr = __sme_page_pa(dst_tpage);
897 dst_paddr = round_down(dst_paddr, 16);
898 len = round_up(size, 16);
901 ret = __sev_issue_dbg_cmd(kvm, paddr, dst_paddr, len, error, true);
905 __free_page(src_tpage);
907 __free_page(dst_tpage);
911 static int sev_dbg_crypt(struct kvm *kvm, struct kvm_sev_cmd *argp, bool dec)
913 unsigned long vaddr, vaddr_end, next_vaddr;
914 unsigned long dst_vaddr;
915 struct page **src_p, **dst_p;
916 struct kvm_sev_dbg debug;
924 if (copy_from_user(&debug, (void __user *)(uintptr_t)argp->data, sizeof(debug)))
927 if (!debug.len || debug.src_uaddr + debug.len < debug.src_uaddr)
929 if (!debug.dst_uaddr)
932 vaddr = debug.src_uaddr;
934 vaddr_end = vaddr + size;
935 dst_vaddr = debug.dst_uaddr;
937 for (; vaddr < vaddr_end; vaddr = next_vaddr) {
938 int len, s_off, d_off;
940 /* lock userspace source and destination page */
941 src_p = sev_pin_memory(kvm, vaddr & PAGE_MASK, PAGE_SIZE, &n, 0);
943 return PTR_ERR(src_p);
945 dst_p = sev_pin_memory(kvm, dst_vaddr & PAGE_MASK, PAGE_SIZE, &n, 1);
947 sev_unpin_memory(kvm, src_p, n);
948 return PTR_ERR(dst_p);
952 * Flush (on non-coherent CPUs) before DBG_{DE,EN}CRYPT read or modify
953 * the pages; flush the destination too so that future accesses do not
956 sev_clflush_pages(src_p, 1);
957 sev_clflush_pages(dst_p, 1);
960 * Since user buffer may not be page aligned, calculate the
961 * offset within the page.
963 s_off = vaddr & ~PAGE_MASK;
964 d_off = dst_vaddr & ~PAGE_MASK;
965 len = min_t(size_t, (PAGE_SIZE - s_off), size);
968 ret = __sev_dbg_decrypt_user(kvm,
969 __sme_page_pa(src_p[0]) + s_off,
970 (void __user *)dst_vaddr,
971 __sme_page_pa(dst_p[0]) + d_off,
974 ret = __sev_dbg_encrypt_user(kvm,
975 __sme_page_pa(src_p[0]) + s_off,
976 (void __user *)vaddr,
977 __sme_page_pa(dst_p[0]) + d_off,
978 (void __user *)dst_vaddr,
981 sev_unpin_memory(kvm, src_p, n);
982 sev_unpin_memory(kvm, dst_p, n);
987 next_vaddr = vaddr + len;
988 dst_vaddr = dst_vaddr + len;
995 static int sev_launch_secret(struct kvm *kvm, struct kvm_sev_cmd *argp)
997 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
998 struct sev_data_launch_secret data;
999 struct kvm_sev_launch_secret params;
1000 struct page **pages;
1005 if (!sev_guest(kvm))
1008 if (copy_from_user(¶ms, (void __user *)(uintptr_t)argp->data, sizeof(params)))
1011 pages = sev_pin_memory(kvm, params.guest_uaddr, params.guest_len, &n, 1);
1013 return PTR_ERR(pages);
1016 * Flush (on non-coherent CPUs) before LAUNCH_SECRET encrypts pages in
1017 * place; the cache may contain the data that was written unencrypted.
1019 sev_clflush_pages(pages, n);
1022 * The secret must be copied into contiguous memory region, lets verify
1023 * that userspace memory pages are contiguous before we issue command.
1025 if (get_num_contig_pages(0, pages, n) != n) {
1027 goto e_unpin_memory;
1030 memset(&data, 0, sizeof(data));
1032 offset = params.guest_uaddr & (PAGE_SIZE - 1);
1033 data.guest_address = __sme_page_pa(pages[0]) + offset;
1034 data.guest_len = params.guest_len;
1036 blob = psp_copy_user_blob(params.trans_uaddr, params.trans_len);
1038 ret = PTR_ERR(blob);
1039 goto e_unpin_memory;
1042 data.trans_address = __psp_pa(blob);
1043 data.trans_len = params.trans_len;
1045 hdr = psp_copy_user_blob(params.hdr_uaddr, params.hdr_len);
1050 data.hdr_address = __psp_pa(hdr);
1051 data.hdr_len = params.hdr_len;
1053 data.handle = sev->handle;
1054 ret = sev_issue_cmd(kvm, SEV_CMD_LAUNCH_UPDATE_SECRET, &data, &argp->error);
1061 /* content of memory is updated, mark pages dirty */
1062 for (i = 0; i < n; i++) {
1063 set_page_dirty_lock(pages[i]);
1064 mark_page_accessed(pages[i]);
1066 sev_unpin_memory(kvm, pages, n);
1070 static int sev_get_attestation_report(struct kvm *kvm, struct kvm_sev_cmd *argp)
1072 void __user *report = (void __user *)(uintptr_t)argp->data;
1073 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1074 struct sev_data_attestation_report data;
1075 struct kvm_sev_attestation_report params;
1080 if (!sev_guest(kvm))
1083 if (copy_from_user(¶ms, (void __user *)(uintptr_t)argp->data, sizeof(params)))
1086 memset(&data, 0, sizeof(data));
1088 /* User wants to query the blob length */
1092 p = (void __user *)(uintptr_t)params.uaddr;
1094 if (params.len > SEV_FW_BLOB_MAX_SIZE)
1097 blob = kzalloc(params.len, GFP_KERNEL_ACCOUNT);
1101 data.address = __psp_pa(blob);
1102 data.len = params.len;
1103 memcpy(data.mnonce, params.mnonce, sizeof(params.mnonce));
1106 data.handle = sev->handle;
1107 ret = sev_issue_cmd(kvm, SEV_CMD_ATTESTATION_REPORT, &data, &argp->error);
1109 * If we query the session length, FW responded with expected data.
1118 if (copy_to_user(p, blob, params.len))
1123 params.len = data.len;
1124 if (copy_to_user(report, ¶ms, sizeof(params)))
1131 /* Userspace wants to query session length. */
1133 __sev_send_start_query_session_length(struct kvm *kvm, struct kvm_sev_cmd *argp,
1134 struct kvm_sev_send_start *params)
1136 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1137 struct sev_data_send_start data;
1140 memset(&data, 0, sizeof(data));
1141 data.handle = sev->handle;
1142 ret = sev_issue_cmd(kvm, SEV_CMD_SEND_START, &data, &argp->error);
1144 params->session_len = data.session_len;
1145 if (copy_to_user((void __user *)(uintptr_t)argp->data, params,
1146 sizeof(struct kvm_sev_send_start)))
1152 static int sev_send_start(struct kvm *kvm, struct kvm_sev_cmd *argp)
1154 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1155 struct sev_data_send_start data;
1156 struct kvm_sev_send_start params;
1157 void *amd_certs, *session_data;
1158 void *pdh_cert, *plat_certs;
1161 if (!sev_guest(kvm))
1164 if (copy_from_user(¶ms, (void __user *)(uintptr_t)argp->data,
1165 sizeof(struct kvm_sev_send_start)))
1168 /* if session_len is zero, userspace wants to query the session length */
1169 if (!params.session_len)
1170 return __sev_send_start_query_session_length(kvm, argp,
1173 /* some sanity checks */
1174 if (!params.pdh_cert_uaddr || !params.pdh_cert_len ||
1175 !params.session_uaddr || params.session_len > SEV_FW_BLOB_MAX_SIZE)
1178 /* allocate the memory to hold the session data blob */
1179 session_data = kzalloc(params.session_len, GFP_KERNEL_ACCOUNT);
1183 /* copy the certificate blobs from userspace */
1184 pdh_cert = psp_copy_user_blob(params.pdh_cert_uaddr,
1185 params.pdh_cert_len);
1186 if (IS_ERR(pdh_cert)) {
1187 ret = PTR_ERR(pdh_cert);
1188 goto e_free_session;
1191 plat_certs = psp_copy_user_blob(params.plat_certs_uaddr,
1192 params.plat_certs_len);
1193 if (IS_ERR(plat_certs)) {
1194 ret = PTR_ERR(plat_certs);
1198 amd_certs = psp_copy_user_blob(params.amd_certs_uaddr,
1199 params.amd_certs_len);
1200 if (IS_ERR(amd_certs)) {
1201 ret = PTR_ERR(amd_certs);
1202 goto e_free_plat_cert;
1205 /* populate the FW SEND_START field with system physical address */
1206 memset(&data, 0, sizeof(data));
1207 data.pdh_cert_address = __psp_pa(pdh_cert);
1208 data.pdh_cert_len = params.pdh_cert_len;
1209 data.plat_certs_address = __psp_pa(plat_certs);
1210 data.plat_certs_len = params.plat_certs_len;
1211 data.amd_certs_address = __psp_pa(amd_certs);
1212 data.amd_certs_len = params.amd_certs_len;
1213 data.session_address = __psp_pa(session_data);
1214 data.session_len = params.session_len;
1215 data.handle = sev->handle;
1217 ret = sev_issue_cmd(kvm, SEV_CMD_SEND_START, &data, &argp->error);
1219 if (!ret && copy_to_user((void __user *)(uintptr_t)params.session_uaddr,
1220 session_data, params.session_len)) {
1222 goto e_free_amd_cert;
1225 params.policy = data.policy;
1226 params.session_len = data.session_len;
1227 if (copy_to_user((void __user *)(uintptr_t)argp->data, ¶ms,
1228 sizeof(struct kvm_sev_send_start)))
1238 kfree(session_data);
1242 /* Userspace wants to query either header or trans length. */
1244 __sev_send_update_data_query_lengths(struct kvm *kvm, struct kvm_sev_cmd *argp,
1245 struct kvm_sev_send_update_data *params)
1247 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1248 struct sev_data_send_update_data data;
1251 memset(&data, 0, sizeof(data));
1252 data.handle = sev->handle;
1253 ret = sev_issue_cmd(kvm, SEV_CMD_SEND_UPDATE_DATA, &data, &argp->error);
1255 params->hdr_len = data.hdr_len;
1256 params->trans_len = data.trans_len;
1258 if (copy_to_user((void __user *)(uintptr_t)argp->data, params,
1259 sizeof(struct kvm_sev_send_update_data)))
1265 static int sev_send_update_data(struct kvm *kvm, struct kvm_sev_cmd *argp)
1267 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1268 struct sev_data_send_update_data data;
1269 struct kvm_sev_send_update_data params;
1270 void *hdr, *trans_data;
1271 struct page **guest_page;
1275 if (!sev_guest(kvm))
1278 if (copy_from_user(¶ms, (void __user *)(uintptr_t)argp->data,
1279 sizeof(struct kvm_sev_send_update_data)))
1282 /* userspace wants to query either header or trans length */
1283 if (!params.trans_len || !params.hdr_len)
1284 return __sev_send_update_data_query_lengths(kvm, argp, ¶ms);
1286 if (!params.trans_uaddr || !params.guest_uaddr ||
1287 !params.guest_len || !params.hdr_uaddr)
1290 /* Check if we are crossing the page boundary */
1291 offset = params.guest_uaddr & (PAGE_SIZE - 1);
1292 if ((params.guest_len + offset > PAGE_SIZE))
1295 /* Pin guest memory */
1296 guest_page = sev_pin_memory(kvm, params.guest_uaddr & PAGE_MASK,
1298 if (IS_ERR(guest_page))
1299 return PTR_ERR(guest_page);
1301 /* allocate memory for header and transport buffer */
1303 hdr = kzalloc(params.hdr_len, GFP_KERNEL_ACCOUNT);
1307 trans_data = kzalloc(params.trans_len, GFP_KERNEL_ACCOUNT);
1311 memset(&data, 0, sizeof(data));
1312 data.hdr_address = __psp_pa(hdr);
1313 data.hdr_len = params.hdr_len;
1314 data.trans_address = __psp_pa(trans_data);
1315 data.trans_len = params.trans_len;
1317 /* The SEND_UPDATE_DATA command requires C-bit to be always set. */
1318 data.guest_address = (page_to_pfn(guest_page[0]) << PAGE_SHIFT) + offset;
1319 data.guest_address |= sev_me_mask;
1320 data.guest_len = params.guest_len;
1321 data.handle = sev->handle;
1323 ret = sev_issue_cmd(kvm, SEV_CMD_SEND_UPDATE_DATA, &data, &argp->error);
1326 goto e_free_trans_data;
1328 /* copy transport buffer to user space */
1329 if (copy_to_user((void __user *)(uintptr_t)params.trans_uaddr,
1330 trans_data, params.trans_len)) {
1332 goto e_free_trans_data;
1335 /* Copy packet header to userspace. */
1336 if (copy_to_user((void __user *)(uintptr_t)params.hdr_uaddr, hdr,
1345 sev_unpin_memory(kvm, guest_page, n);
1350 static int sev_send_finish(struct kvm *kvm, struct kvm_sev_cmd *argp)
1352 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1353 struct sev_data_send_finish data;
1355 if (!sev_guest(kvm))
1358 data.handle = sev->handle;
1359 return sev_issue_cmd(kvm, SEV_CMD_SEND_FINISH, &data, &argp->error);
1362 static int sev_send_cancel(struct kvm *kvm, struct kvm_sev_cmd *argp)
1364 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1365 struct sev_data_send_cancel data;
1367 if (!sev_guest(kvm))
1370 data.handle = sev->handle;
1371 return sev_issue_cmd(kvm, SEV_CMD_SEND_CANCEL, &data, &argp->error);
1374 static int sev_receive_start(struct kvm *kvm, struct kvm_sev_cmd *argp)
1376 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1377 struct sev_data_receive_start start;
1378 struct kvm_sev_receive_start params;
1379 int *error = &argp->error;
1384 if (!sev_guest(kvm))
1387 /* Get parameter from the userspace */
1388 if (copy_from_user(¶ms, (void __user *)(uintptr_t)argp->data,
1389 sizeof(struct kvm_sev_receive_start)))
1392 /* some sanity checks */
1393 if (!params.pdh_uaddr || !params.pdh_len ||
1394 !params.session_uaddr || !params.session_len)
1397 pdh_data = psp_copy_user_blob(params.pdh_uaddr, params.pdh_len);
1398 if (IS_ERR(pdh_data))
1399 return PTR_ERR(pdh_data);
1401 session_data = psp_copy_user_blob(params.session_uaddr,
1402 params.session_len);
1403 if (IS_ERR(session_data)) {
1404 ret = PTR_ERR(session_data);
1408 memset(&start, 0, sizeof(start));
1409 start.handle = params.handle;
1410 start.policy = params.policy;
1411 start.pdh_cert_address = __psp_pa(pdh_data);
1412 start.pdh_cert_len = params.pdh_len;
1413 start.session_address = __psp_pa(session_data);
1414 start.session_len = params.session_len;
1416 /* create memory encryption context */
1417 ret = __sev_issue_cmd(argp->sev_fd, SEV_CMD_RECEIVE_START, &start,
1420 goto e_free_session;
1422 /* Bind ASID to this guest */
1423 ret = sev_bind_asid(kvm, start.handle, error);
1425 sev_decommission(start.handle);
1426 goto e_free_session;
1429 params.handle = start.handle;
1430 if (copy_to_user((void __user *)(uintptr_t)argp->data,
1431 ¶ms, sizeof(struct kvm_sev_receive_start))) {
1433 sev_unbind_asid(kvm, start.handle);
1434 goto e_free_session;
1437 sev->handle = start.handle;
1438 sev->fd = argp->sev_fd;
1441 kfree(session_data);
1448 static int sev_receive_update_data(struct kvm *kvm, struct kvm_sev_cmd *argp)
1450 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1451 struct kvm_sev_receive_update_data params;
1452 struct sev_data_receive_update_data data;
1453 void *hdr = NULL, *trans = NULL;
1454 struct page **guest_page;
1458 if (!sev_guest(kvm))
1461 if (copy_from_user(¶ms, (void __user *)(uintptr_t)argp->data,
1462 sizeof(struct kvm_sev_receive_update_data)))
1465 if (!params.hdr_uaddr || !params.hdr_len ||
1466 !params.guest_uaddr || !params.guest_len ||
1467 !params.trans_uaddr || !params.trans_len)
1470 /* Check if we are crossing the page boundary */
1471 offset = params.guest_uaddr & (PAGE_SIZE - 1);
1472 if ((params.guest_len + offset > PAGE_SIZE))
1475 hdr = psp_copy_user_blob(params.hdr_uaddr, params.hdr_len);
1477 return PTR_ERR(hdr);
1479 trans = psp_copy_user_blob(params.trans_uaddr, params.trans_len);
1480 if (IS_ERR(trans)) {
1481 ret = PTR_ERR(trans);
1485 memset(&data, 0, sizeof(data));
1486 data.hdr_address = __psp_pa(hdr);
1487 data.hdr_len = params.hdr_len;
1488 data.trans_address = __psp_pa(trans);
1489 data.trans_len = params.trans_len;
1491 /* Pin guest memory */
1492 guest_page = sev_pin_memory(kvm, params.guest_uaddr & PAGE_MASK,
1494 if (IS_ERR(guest_page)) {
1495 ret = PTR_ERR(guest_page);
1500 * Flush (on non-coherent CPUs) before RECEIVE_UPDATE_DATA, the PSP
1501 * encrypts the written data with the guest's key, and the cache may
1502 * contain dirty, unencrypted data.
1504 sev_clflush_pages(guest_page, n);
1506 /* The RECEIVE_UPDATE_DATA command requires C-bit to be always set. */
1507 data.guest_address = (page_to_pfn(guest_page[0]) << PAGE_SHIFT) + offset;
1508 data.guest_address |= sev_me_mask;
1509 data.guest_len = params.guest_len;
1510 data.handle = sev->handle;
1512 ret = sev_issue_cmd(kvm, SEV_CMD_RECEIVE_UPDATE_DATA, &data,
1515 sev_unpin_memory(kvm, guest_page, n);
1525 static int sev_receive_finish(struct kvm *kvm, struct kvm_sev_cmd *argp)
1527 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1528 struct sev_data_receive_finish data;
1530 if (!sev_guest(kvm))
1533 data.handle = sev->handle;
1534 return sev_issue_cmd(kvm, SEV_CMD_RECEIVE_FINISH, &data, &argp->error);
1537 static bool is_cmd_allowed_from_mirror(u32 cmd_id)
1540 * Allow mirrors VM to call KVM_SEV_LAUNCH_UPDATE_VMSA to enable SEV-ES
1541 * active mirror VMs. Also allow the debugging and status commands.
1543 if (cmd_id == KVM_SEV_LAUNCH_UPDATE_VMSA ||
1544 cmd_id == KVM_SEV_GUEST_STATUS || cmd_id == KVM_SEV_DBG_DECRYPT ||
1545 cmd_id == KVM_SEV_DBG_ENCRYPT)
1551 static int sev_lock_two_vms(struct kvm *dst_kvm, struct kvm *src_kvm)
1553 struct kvm_sev_info *dst_sev = &to_kvm_svm(dst_kvm)->sev_info;
1554 struct kvm_sev_info *src_sev = &to_kvm_svm(src_kvm)->sev_info;
1557 if (dst_kvm == src_kvm)
1561 * Bail if these VMs are already involved in a migration to avoid
1562 * deadlock between two VMs trying to migrate to/from each other.
1564 if (atomic_cmpxchg_acquire(&dst_sev->migration_in_progress, 0, 1))
1567 if (atomic_cmpxchg_acquire(&src_sev->migration_in_progress, 0, 1))
1571 if (mutex_lock_killable(&dst_kvm->lock))
1573 if (mutex_lock_killable_nested(&src_kvm->lock, SINGLE_DEPTH_NESTING))
1578 mutex_unlock(&dst_kvm->lock);
1580 atomic_set_release(&src_sev->migration_in_progress, 0);
1582 atomic_set_release(&dst_sev->migration_in_progress, 0);
1586 static void sev_unlock_two_vms(struct kvm *dst_kvm, struct kvm *src_kvm)
1588 struct kvm_sev_info *dst_sev = &to_kvm_svm(dst_kvm)->sev_info;
1589 struct kvm_sev_info *src_sev = &to_kvm_svm(src_kvm)->sev_info;
1591 mutex_unlock(&dst_kvm->lock);
1592 mutex_unlock(&src_kvm->lock);
1593 atomic_set_release(&dst_sev->migration_in_progress, 0);
1594 atomic_set_release(&src_sev->migration_in_progress, 0);
1597 /* vCPU mutex subclasses. */
1598 enum sev_migration_role {
1599 SEV_MIGRATION_SOURCE = 0,
1600 SEV_MIGRATION_TARGET,
1601 SEV_NR_MIGRATION_ROLES,
1604 static int sev_lock_vcpus_for_migration(struct kvm *kvm,
1605 enum sev_migration_role role)
1607 struct kvm_vcpu *vcpu;
1611 kvm_for_each_vcpu(i, vcpu, kvm) {
1612 if (mutex_lock_killable_nested(&vcpu->mutex, role))
1617 * Reset the role to one that avoids colliding with
1618 * the role used for the first vcpu mutex.
1620 role = SEV_NR_MIGRATION_ROLES;
1623 mutex_release(&vcpu->mutex.dep_map, _THIS_IP_);
1632 kvm_for_each_vcpu(j, vcpu, kvm) {
1639 mutex_acquire(&vcpu->mutex.dep_map, role, 0, _THIS_IP_);
1642 mutex_unlock(&vcpu->mutex);
1647 static void sev_unlock_vcpus_for_migration(struct kvm *kvm)
1649 struct kvm_vcpu *vcpu;
1653 kvm_for_each_vcpu(i, vcpu, kvm) {
1657 mutex_acquire(&vcpu->mutex.dep_map,
1658 SEV_NR_MIGRATION_ROLES, 0, _THIS_IP_);
1660 mutex_unlock(&vcpu->mutex);
1664 static void sev_migrate_from(struct kvm *dst_kvm, struct kvm *src_kvm)
1666 struct kvm_sev_info *dst = &to_kvm_svm(dst_kvm)->sev_info;
1667 struct kvm_sev_info *src = &to_kvm_svm(src_kvm)->sev_info;
1668 struct kvm_vcpu *dst_vcpu, *src_vcpu;
1669 struct vcpu_svm *dst_svm, *src_svm;
1670 struct kvm_sev_info *mirror;
1674 dst->asid = src->asid;
1675 dst->handle = src->handle;
1676 dst->pages_locked = src->pages_locked;
1677 dst->enc_context_owner = src->enc_context_owner;
1678 dst->es_active = src->es_active;
1681 src->active = false;
1683 src->pages_locked = 0;
1684 src->enc_context_owner = NULL;
1685 src->es_active = false;
1687 list_cut_before(&dst->regions_list, &src->regions_list, &src->regions_list);
1690 * If this VM has mirrors, "transfer" each mirror's refcount of the
1691 * source to the destination (this KVM). The caller holds a reference
1692 * to the source, so there's no danger of use-after-free.
1694 list_cut_before(&dst->mirror_vms, &src->mirror_vms, &src->mirror_vms);
1695 list_for_each_entry(mirror, &dst->mirror_vms, mirror_entry) {
1696 kvm_get_kvm(dst_kvm);
1697 kvm_put_kvm(src_kvm);
1698 mirror->enc_context_owner = dst_kvm;
1702 * If this VM is a mirror, remove the old mirror from the owners list
1703 * and add the new mirror to the list.
1705 if (is_mirroring_enc_context(dst_kvm)) {
1706 struct kvm_sev_info *owner_sev_info =
1707 &to_kvm_svm(dst->enc_context_owner)->sev_info;
1709 list_del(&src->mirror_entry);
1710 list_add_tail(&dst->mirror_entry, &owner_sev_info->mirror_vms);
1713 kvm_for_each_vcpu(i, dst_vcpu, dst_kvm) {
1714 dst_svm = to_svm(dst_vcpu);
1716 sev_init_vmcb(dst_svm);
1718 if (!dst->es_active)
1722 * Note, the source is not required to have the same number of
1723 * vCPUs as the destination when migrating a vanilla SEV VM.
1725 src_vcpu = kvm_get_vcpu(dst_kvm, i);
1726 src_svm = to_svm(src_vcpu);
1729 * Transfer VMSA and GHCB state to the destination. Nullify and
1730 * clear source fields as appropriate, the state now belongs to
1733 memcpy(&dst_svm->sev_es, &src_svm->sev_es, sizeof(src_svm->sev_es));
1734 dst_svm->vmcb->control.ghcb_gpa = src_svm->vmcb->control.ghcb_gpa;
1735 dst_svm->vmcb->control.vmsa_pa = src_svm->vmcb->control.vmsa_pa;
1736 dst_vcpu->arch.guest_state_protected = true;
1738 memset(&src_svm->sev_es, 0, sizeof(src_svm->sev_es));
1739 src_svm->vmcb->control.ghcb_gpa = INVALID_PAGE;
1740 src_svm->vmcb->control.vmsa_pa = INVALID_PAGE;
1741 src_vcpu->arch.guest_state_protected = false;
1745 static int sev_check_source_vcpus(struct kvm *dst, struct kvm *src)
1747 struct kvm_vcpu *src_vcpu;
1750 if (!sev_es_guest(src))
1753 if (atomic_read(&src->online_vcpus) != atomic_read(&dst->online_vcpus))
1756 kvm_for_each_vcpu(i, src_vcpu, src) {
1757 if (!src_vcpu->arch.guest_state_protected)
1764 int sev_vm_move_enc_context_from(struct kvm *kvm, unsigned int source_fd)
1766 struct kvm_sev_info *dst_sev = &to_kvm_svm(kvm)->sev_info;
1767 struct kvm_sev_info *src_sev, *cg_cleanup_sev;
1768 struct file *source_kvm_file;
1769 struct kvm *source_kvm;
1770 bool charged = false;
1773 source_kvm_file = fget(source_fd);
1774 if (!file_is_kvm(source_kvm_file)) {
1779 source_kvm = source_kvm_file->private_data;
1780 ret = sev_lock_two_vms(kvm, source_kvm);
1784 if (sev_guest(kvm) || !sev_guest(source_kvm)) {
1789 src_sev = &to_kvm_svm(source_kvm)->sev_info;
1791 dst_sev->misc_cg = get_current_misc_cg();
1792 cg_cleanup_sev = dst_sev;
1793 if (dst_sev->misc_cg != src_sev->misc_cg) {
1794 ret = sev_misc_cg_try_charge(dst_sev);
1796 goto out_dst_cgroup;
1800 ret = sev_lock_vcpus_for_migration(kvm, SEV_MIGRATION_SOURCE);
1802 goto out_dst_cgroup;
1803 ret = sev_lock_vcpus_for_migration(source_kvm, SEV_MIGRATION_TARGET);
1807 ret = sev_check_source_vcpus(kvm, source_kvm);
1809 goto out_source_vcpu;
1811 sev_migrate_from(kvm, source_kvm);
1812 kvm_vm_dead(source_kvm);
1813 cg_cleanup_sev = src_sev;
1817 sev_unlock_vcpus_for_migration(source_kvm);
1819 sev_unlock_vcpus_for_migration(kvm);
1821 /* Operates on the source on success, on the destination on failure. */
1823 sev_misc_cg_uncharge(cg_cleanup_sev);
1824 put_misc_cg(cg_cleanup_sev->misc_cg);
1825 cg_cleanup_sev->misc_cg = NULL;
1827 sev_unlock_two_vms(kvm, source_kvm);
1829 if (source_kvm_file)
1830 fput(source_kvm_file);
1834 int sev_mem_enc_ioctl(struct kvm *kvm, void __user *argp)
1836 struct kvm_sev_cmd sev_cmd;
1845 if (copy_from_user(&sev_cmd, argp, sizeof(struct kvm_sev_cmd)))
1848 mutex_lock(&kvm->lock);
1850 /* Only the enc_context_owner handles some memory enc operations. */
1851 if (is_mirroring_enc_context(kvm) &&
1852 !is_cmd_allowed_from_mirror(sev_cmd.id)) {
1857 switch (sev_cmd.id) {
1858 case KVM_SEV_ES_INIT:
1859 if (!sev_es_enabled) {
1865 r = sev_guest_init(kvm, &sev_cmd);
1867 case KVM_SEV_LAUNCH_START:
1868 r = sev_launch_start(kvm, &sev_cmd);
1870 case KVM_SEV_LAUNCH_UPDATE_DATA:
1871 r = sev_launch_update_data(kvm, &sev_cmd);
1873 case KVM_SEV_LAUNCH_UPDATE_VMSA:
1874 r = sev_launch_update_vmsa(kvm, &sev_cmd);
1876 case KVM_SEV_LAUNCH_MEASURE:
1877 r = sev_launch_measure(kvm, &sev_cmd);
1879 case KVM_SEV_LAUNCH_FINISH:
1880 r = sev_launch_finish(kvm, &sev_cmd);
1882 case KVM_SEV_GUEST_STATUS:
1883 r = sev_guest_status(kvm, &sev_cmd);
1885 case KVM_SEV_DBG_DECRYPT:
1886 r = sev_dbg_crypt(kvm, &sev_cmd, true);
1888 case KVM_SEV_DBG_ENCRYPT:
1889 r = sev_dbg_crypt(kvm, &sev_cmd, false);
1891 case KVM_SEV_LAUNCH_SECRET:
1892 r = sev_launch_secret(kvm, &sev_cmd);
1894 case KVM_SEV_GET_ATTESTATION_REPORT:
1895 r = sev_get_attestation_report(kvm, &sev_cmd);
1897 case KVM_SEV_SEND_START:
1898 r = sev_send_start(kvm, &sev_cmd);
1900 case KVM_SEV_SEND_UPDATE_DATA:
1901 r = sev_send_update_data(kvm, &sev_cmd);
1903 case KVM_SEV_SEND_FINISH:
1904 r = sev_send_finish(kvm, &sev_cmd);
1906 case KVM_SEV_SEND_CANCEL:
1907 r = sev_send_cancel(kvm, &sev_cmd);
1909 case KVM_SEV_RECEIVE_START:
1910 r = sev_receive_start(kvm, &sev_cmd);
1912 case KVM_SEV_RECEIVE_UPDATE_DATA:
1913 r = sev_receive_update_data(kvm, &sev_cmd);
1915 case KVM_SEV_RECEIVE_FINISH:
1916 r = sev_receive_finish(kvm, &sev_cmd);
1923 if (copy_to_user(argp, &sev_cmd, sizeof(struct kvm_sev_cmd)))
1927 mutex_unlock(&kvm->lock);
1931 int sev_mem_enc_register_region(struct kvm *kvm,
1932 struct kvm_enc_region *range)
1934 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1935 struct enc_region *region;
1938 if (!sev_guest(kvm))
1941 /* If kvm is mirroring encryption context it isn't responsible for it */
1942 if (is_mirroring_enc_context(kvm))
1945 if (range->addr > ULONG_MAX || range->size > ULONG_MAX)
1948 region = kzalloc(sizeof(*region), GFP_KERNEL_ACCOUNT);
1952 mutex_lock(&kvm->lock);
1953 region->pages = sev_pin_memory(kvm, range->addr, range->size, ®ion->npages, 1);
1954 if (IS_ERR(region->pages)) {
1955 ret = PTR_ERR(region->pages);
1956 mutex_unlock(&kvm->lock);
1960 region->uaddr = range->addr;
1961 region->size = range->size;
1963 list_add_tail(®ion->list, &sev->regions_list);
1964 mutex_unlock(&kvm->lock);
1967 * The guest may change the memory encryption attribute from C=0 -> C=1
1968 * or vice versa for this memory range. Lets make sure caches are
1969 * flushed to ensure that guest data gets written into memory with
1972 sev_clflush_pages(region->pages, region->npages);
1981 static struct enc_region *
1982 find_enc_region(struct kvm *kvm, struct kvm_enc_region *range)
1984 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1985 struct list_head *head = &sev->regions_list;
1986 struct enc_region *i;
1988 list_for_each_entry(i, head, list) {
1989 if (i->uaddr == range->addr &&
1990 i->size == range->size)
1997 static void __unregister_enc_region_locked(struct kvm *kvm,
1998 struct enc_region *region)
2000 sev_unpin_memory(kvm, region->pages, region->npages);
2001 list_del(®ion->list);
2005 int sev_mem_enc_unregister_region(struct kvm *kvm,
2006 struct kvm_enc_region *range)
2008 struct enc_region *region;
2011 /* If kvm is mirroring encryption context it isn't responsible for it */
2012 if (is_mirroring_enc_context(kvm))
2015 mutex_lock(&kvm->lock);
2017 if (!sev_guest(kvm)) {
2022 region = find_enc_region(kvm, range);
2029 * Ensure that all guest tagged cache entries are flushed before
2030 * releasing the pages back to the system for use. CLFLUSH will
2031 * not do this, so issue a WBINVD.
2033 wbinvd_on_all_cpus();
2035 __unregister_enc_region_locked(kvm, region);
2037 mutex_unlock(&kvm->lock);
2041 mutex_unlock(&kvm->lock);
2045 int sev_vm_copy_enc_context_from(struct kvm *kvm, unsigned int source_fd)
2047 struct file *source_kvm_file;
2048 struct kvm *source_kvm;
2049 struct kvm_sev_info *source_sev, *mirror_sev;
2052 source_kvm_file = fget(source_fd);
2053 if (!file_is_kvm(source_kvm_file)) {
2058 source_kvm = source_kvm_file->private_data;
2059 ret = sev_lock_two_vms(kvm, source_kvm);
2064 * Mirrors of mirrors should work, but let's not get silly. Also
2065 * disallow out-of-band SEV/SEV-ES init if the target is already an
2066 * SEV guest, or if vCPUs have been created. KVM relies on vCPUs being
2067 * created after SEV/SEV-ES initialization, e.g. to init intercepts.
2069 if (sev_guest(kvm) || !sev_guest(source_kvm) ||
2070 is_mirroring_enc_context(source_kvm) || kvm->created_vcpus) {
2076 * The mirror kvm holds an enc_context_owner ref so its asid can't
2077 * disappear until we're done with it
2079 source_sev = &to_kvm_svm(source_kvm)->sev_info;
2080 kvm_get_kvm(source_kvm);
2081 mirror_sev = &to_kvm_svm(kvm)->sev_info;
2082 list_add_tail(&mirror_sev->mirror_entry, &source_sev->mirror_vms);
2084 /* Set enc_context_owner and copy its encryption context over */
2085 mirror_sev->enc_context_owner = source_kvm;
2086 mirror_sev->active = true;
2087 mirror_sev->asid = source_sev->asid;
2088 mirror_sev->fd = source_sev->fd;
2089 mirror_sev->es_active = source_sev->es_active;
2090 mirror_sev->handle = source_sev->handle;
2091 INIT_LIST_HEAD(&mirror_sev->regions_list);
2092 INIT_LIST_HEAD(&mirror_sev->mirror_vms);
2096 * Do not copy ap_jump_table. Since the mirror does not share the same
2097 * KVM contexts as the original, and they may have different
2102 sev_unlock_two_vms(kvm, source_kvm);
2104 if (source_kvm_file)
2105 fput(source_kvm_file);
2109 void sev_vm_destroy(struct kvm *kvm)
2111 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
2112 struct list_head *head = &sev->regions_list;
2113 struct list_head *pos, *q;
2115 if (!sev_guest(kvm))
2118 WARN_ON(!list_empty(&sev->mirror_vms));
2120 /* If this is a mirror_kvm release the enc_context_owner and skip sev cleanup */
2121 if (is_mirroring_enc_context(kvm)) {
2122 struct kvm *owner_kvm = sev->enc_context_owner;
2124 mutex_lock(&owner_kvm->lock);
2125 list_del(&sev->mirror_entry);
2126 mutex_unlock(&owner_kvm->lock);
2127 kvm_put_kvm(owner_kvm);
2132 * Ensure that all guest tagged cache entries are flushed before
2133 * releasing the pages back to the system for use. CLFLUSH will
2134 * not do this, so issue a WBINVD.
2136 wbinvd_on_all_cpus();
2139 * if userspace was terminated before unregistering the memory regions
2140 * then lets unpin all the registered memory.
2142 if (!list_empty(head)) {
2143 list_for_each_safe(pos, q, head) {
2144 __unregister_enc_region_locked(kvm,
2145 list_entry(pos, struct enc_region, list));
2150 sev_unbind_asid(kvm, sev->handle);
2154 void __init sev_set_cpu_caps(void)
2157 kvm_cpu_cap_clear(X86_FEATURE_SEV);
2158 if (!sev_es_enabled)
2159 kvm_cpu_cap_clear(X86_FEATURE_SEV_ES);
2162 void __init sev_hardware_setup(void)
2164 #ifdef CONFIG_KVM_AMD_SEV
2165 unsigned int eax, ebx, ecx, edx, sev_asid_count, sev_es_asid_count;
2166 bool sev_es_supported = false;
2167 bool sev_supported = false;
2169 if (!sev_enabled || !npt_enabled)
2173 * SEV must obviously be supported in hardware. Sanity check that the
2174 * CPU supports decode assists, which is mandatory for SEV guests to
2175 * support instruction emulation.
2177 if (!boot_cpu_has(X86_FEATURE_SEV) ||
2178 WARN_ON_ONCE(!boot_cpu_has(X86_FEATURE_DECODEASSISTS)))
2181 /* Retrieve SEV CPUID information */
2182 cpuid(0x8000001f, &eax, &ebx, &ecx, &edx);
2184 /* Set encryption bit location for SEV-ES guests */
2185 sev_enc_bit = ebx & 0x3f;
2187 /* Maximum number of encrypted guests supported simultaneously */
2192 /* Minimum ASID value that should be used for SEV guest */
2194 sev_me_mask = 1UL << (ebx & 0x3f);
2197 * Initialize SEV ASID bitmaps. Allocate space for ASID 0 in the bitmap,
2198 * even though it's never used, so that the bitmap is indexed by the
2201 nr_asids = max_sev_asid + 1;
2202 sev_asid_bitmap = bitmap_zalloc(nr_asids, GFP_KERNEL);
2203 if (!sev_asid_bitmap)
2206 sev_reclaim_asid_bitmap = bitmap_zalloc(nr_asids, GFP_KERNEL);
2207 if (!sev_reclaim_asid_bitmap) {
2208 bitmap_free(sev_asid_bitmap);
2209 sev_asid_bitmap = NULL;
2213 sev_asid_count = max_sev_asid - min_sev_asid + 1;
2214 if (misc_cg_set_capacity(MISC_CG_RES_SEV, sev_asid_count))
2217 pr_info("SEV supported: %u ASIDs\n", sev_asid_count);
2218 sev_supported = true;
2220 /* SEV-ES support requested? */
2221 if (!sev_es_enabled)
2224 /* Does the CPU support SEV-ES? */
2225 if (!boot_cpu_has(X86_FEATURE_SEV_ES))
2228 /* Has the system been allocated ASIDs for SEV-ES? */
2229 if (min_sev_asid == 1)
2232 sev_es_asid_count = min_sev_asid - 1;
2233 if (misc_cg_set_capacity(MISC_CG_RES_SEV_ES, sev_es_asid_count))
2236 pr_info("SEV-ES supported: %u ASIDs\n", sev_es_asid_count);
2237 sev_es_supported = true;
2240 sev_enabled = sev_supported;
2241 sev_es_enabled = sev_es_supported;
2245 void sev_hardware_unsetup(void)
2250 /* No need to take sev_bitmap_lock, all VMs have been destroyed. */
2251 sev_flush_asids(1, max_sev_asid);
2253 bitmap_free(sev_asid_bitmap);
2254 bitmap_free(sev_reclaim_asid_bitmap);
2256 misc_cg_set_capacity(MISC_CG_RES_SEV, 0);
2257 misc_cg_set_capacity(MISC_CG_RES_SEV_ES, 0);
2260 int sev_cpu_init(struct svm_cpu_data *sd)
2265 sd->sev_vmcbs = kcalloc(nr_asids, sizeof(void *), GFP_KERNEL);
2273 * Pages used by hardware to hold guest encrypted state must be flushed before
2274 * returning them to the system.
2276 static void sev_flush_encrypted_page(struct kvm_vcpu *vcpu, void *va)
2278 int asid = to_kvm_svm(vcpu->kvm)->sev_info.asid;
2281 * Note! The address must be a kernel address, as regular page walk
2282 * checks are performed by VM_PAGE_FLUSH, i.e. operating on a user
2283 * address is non-deterministic and unsafe. This function deliberately
2284 * takes a pointer to deter passing in a user address.
2286 unsigned long addr = (unsigned long)va;
2289 * If CPU enforced cache coherency for encrypted mappings of the
2290 * same physical page is supported, use CLFLUSHOPT instead. NOTE: cache
2291 * flush is still needed in order to work properly with DMA devices.
2293 if (boot_cpu_has(X86_FEATURE_SME_COHERENT)) {
2294 clflush_cache_range(va, PAGE_SIZE);
2299 * VM Page Flush takes a host virtual address and a guest ASID. Fall
2300 * back to WBINVD if this faults so as not to make any problems worse
2301 * by leaving stale encrypted data in the cache.
2303 if (WARN_ON_ONCE(wrmsrl_safe(MSR_AMD64_VM_PAGE_FLUSH, addr | asid)))
2309 wbinvd_on_all_cpus();
2312 void sev_guest_memory_reclaimed(struct kvm *kvm)
2314 if (!sev_guest(kvm))
2317 wbinvd_on_all_cpus();
2320 void sev_free_vcpu(struct kvm_vcpu *vcpu)
2322 struct vcpu_svm *svm;
2324 if (!sev_es_guest(vcpu->kvm))
2329 if (vcpu->arch.guest_state_protected)
2330 sev_flush_encrypted_page(vcpu, svm->sev_es.vmsa);
2332 __free_page(virt_to_page(svm->sev_es.vmsa));
2334 if (svm->sev_es.ghcb_sa_free)
2335 kvfree(svm->sev_es.ghcb_sa);
2338 static void dump_ghcb(struct vcpu_svm *svm)
2340 struct ghcb *ghcb = svm->sev_es.ghcb;
2343 /* Re-use the dump_invalid_vmcb module parameter */
2344 if (!dump_invalid_vmcb) {
2345 pr_warn_ratelimited("set kvm_amd.dump_invalid_vmcb=1 to dump internal KVM state.\n");
2349 nbits = sizeof(ghcb->save.valid_bitmap) * 8;
2351 pr_err("GHCB (GPA=%016llx):\n", svm->vmcb->control.ghcb_gpa);
2352 pr_err("%-20s%016llx is_valid: %u\n", "sw_exit_code",
2353 ghcb->save.sw_exit_code, ghcb_sw_exit_code_is_valid(ghcb));
2354 pr_err("%-20s%016llx is_valid: %u\n", "sw_exit_info_1",
2355 ghcb->save.sw_exit_info_1, ghcb_sw_exit_info_1_is_valid(ghcb));
2356 pr_err("%-20s%016llx is_valid: %u\n", "sw_exit_info_2",
2357 ghcb->save.sw_exit_info_2, ghcb_sw_exit_info_2_is_valid(ghcb));
2358 pr_err("%-20s%016llx is_valid: %u\n", "sw_scratch",
2359 ghcb->save.sw_scratch, ghcb_sw_scratch_is_valid(ghcb));
2360 pr_err("%-20s%*pb\n", "valid_bitmap", nbits, ghcb->save.valid_bitmap);
2363 static void sev_es_sync_to_ghcb(struct vcpu_svm *svm)
2365 struct kvm_vcpu *vcpu = &svm->vcpu;
2366 struct ghcb *ghcb = svm->sev_es.ghcb;
2369 * The GHCB protocol so far allows for the following data
2371 * GPRs RAX, RBX, RCX, RDX
2373 * Copy their values, even if they may not have been written during the
2374 * VM-Exit. It's the guest's responsibility to not consume random data.
2376 ghcb_set_rax(ghcb, vcpu->arch.regs[VCPU_REGS_RAX]);
2377 ghcb_set_rbx(ghcb, vcpu->arch.regs[VCPU_REGS_RBX]);
2378 ghcb_set_rcx(ghcb, vcpu->arch.regs[VCPU_REGS_RCX]);
2379 ghcb_set_rdx(ghcb, vcpu->arch.regs[VCPU_REGS_RDX]);
2382 static void sev_es_sync_from_ghcb(struct vcpu_svm *svm)
2384 struct vmcb_control_area *control = &svm->vmcb->control;
2385 struct kvm_vcpu *vcpu = &svm->vcpu;
2386 struct ghcb *ghcb = svm->sev_es.ghcb;
2390 * The GHCB protocol so far allows for the following data
2392 * GPRs RAX, RBX, RCX, RDX
2396 * VMMCALL allows the guest to provide extra registers. KVM also
2397 * expects RSI for hypercalls, so include that, too.
2399 * Copy their values to the appropriate location if supplied.
2401 memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
2403 vcpu->arch.regs[VCPU_REGS_RAX] = ghcb_get_rax_if_valid(ghcb);
2404 vcpu->arch.regs[VCPU_REGS_RBX] = ghcb_get_rbx_if_valid(ghcb);
2405 vcpu->arch.regs[VCPU_REGS_RCX] = ghcb_get_rcx_if_valid(ghcb);
2406 vcpu->arch.regs[VCPU_REGS_RDX] = ghcb_get_rdx_if_valid(ghcb);
2407 vcpu->arch.regs[VCPU_REGS_RSI] = ghcb_get_rsi_if_valid(ghcb);
2409 svm->vmcb->save.cpl = ghcb_get_cpl_if_valid(ghcb);
2411 if (ghcb_xcr0_is_valid(ghcb)) {
2412 vcpu->arch.xcr0 = ghcb_get_xcr0(ghcb);
2413 kvm_update_cpuid_runtime(vcpu);
2416 /* Copy the GHCB exit information into the VMCB fields */
2417 exit_code = ghcb_get_sw_exit_code(ghcb);
2418 control->exit_code = lower_32_bits(exit_code);
2419 control->exit_code_hi = upper_32_bits(exit_code);
2420 control->exit_info_1 = ghcb_get_sw_exit_info_1(ghcb);
2421 control->exit_info_2 = ghcb_get_sw_exit_info_2(ghcb);
2423 /* Clear the valid entries fields */
2424 memset(ghcb->save.valid_bitmap, 0, sizeof(ghcb->save.valid_bitmap));
2427 static int sev_es_validate_vmgexit(struct vcpu_svm *svm)
2429 struct kvm_vcpu *vcpu;
2434 ghcb = svm->sev_es.ghcb;
2437 * Retrieve the exit code now even though it may not be marked valid
2438 * as it could help with debugging.
2440 exit_code = ghcb_get_sw_exit_code(ghcb);
2442 /* Only GHCB Usage code 0 is supported */
2443 if (ghcb->ghcb_usage) {
2444 reason = GHCB_ERR_INVALID_USAGE;
2448 reason = GHCB_ERR_MISSING_INPUT;
2450 if (!ghcb_sw_exit_code_is_valid(ghcb) ||
2451 !ghcb_sw_exit_info_1_is_valid(ghcb) ||
2452 !ghcb_sw_exit_info_2_is_valid(ghcb))
2455 switch (ghcb_get_sw_exit_code(ghcb)) {
2456 case SVM_EXIT_READ_DR7:
2458 case SVM_EXIT_WRITE_DR7:
2459 if (!ghcb_rax_is_valid(ghcb))
2462 case SVM_EXIT_RDTSC:
2464 case SVM_EXIT_RDPMC:
2465 if (!ghcb_rcx_is_valid(ghcb))
2468 case SVM_EXIT_CPUID:
2469 if (!ghcb_rax_is_valid(ghcb) ||
2470 !ghcb_rcx_is_valid(ghcb))
2472 if (ghcb_get_rax(ghcb) == 0xd)
2473 if (!ghcb_xcr0_is_valid(ghcb))
2479 if (ghcb_get_sw_exit_info_1(ghcb) & SVM_IOIO_STR_MASK) {
2480 if (!ghcb_sw_scratch_is_valid(ghcb))
2483 if (!(ghcb_get_sw_exit_info_1(ghcb) & SVM_IOIO_TYPE_MASK))
2484 if (!ghcb_rax_is_valid(ghcb))
2489 if (!ghcb_rcx_is_valid(ghcb))
2491 if (ghcb_get_sw_exit_info_1(ghcb)) {
2492 if (!ghcb_rax_is_valid(ghcb) ||
2493 !ghcb_rdx_is_valid(ghcb))
2497 case SVM_EXIT_VMMCALL:
2498 if (!ghcb_rax_is_valid(ghcb) ||
2499 !ghcb_cpl_is_valid(ghcb))
2502 case SVM_EXIT_RDTSCP:
2504 case SVM_EXIT_WBINVD:
2506 case SVM_EXIT_MONITOR:
2507 if (!ghcb_rax_is_valid(ghcb) ||
2508 !ghcb_rcx_is_valid(ghcb) ||
2509 !ghcb_rdx_is_valid(ghcb))
2512 case SVM_EXIT_MWAIT:
2513 if (!ghcb_rax_is_valid(ghcb) ||
2514 !ghcb_rcx_is_valid(ghcb))
2517 case SVM_VMGEXIT_MMIO_READ:
2518 case SVM_VMGEXIT_MMIO_WRITE:
2519 if (!ghcb_sw_scratch_is_valid(ghcb))
2522 case SVM_VMGEXIT_NMI_COMPLETE:
2523 case SVM_VMGEXIT_AP_HLT_LOOP:
2524 case SVM_VMGEXIT_AP_JUMP_TABLE:
2525 case SVM_VMGEXIT_UNSUPPORTED_EVENT:
2528 reason = GHCB_ERR_INVALID_EVENT;
2537 if (reason == GHCB_ERR_INVALID_USAGE) {
2538 vcpu_unimpl(vcpu, "vmgexit: ghcb usage %#x is not valid\n",
2540 } else if (reason == GHCB_ERR_INVALID_EVENT) {
2541 vcpu_unimpl(vcpu, "vmgexit: exit code %#llx is not valid\n",
2544 vcpu_unimpl(vcpu, "vmgexit: exit code %#llx input is not valid\n",
2549 /* Clear the valid entries fields */
2550 memset(ghcb->save.valid_bitmap, 0, sizeof(ghcb->save.valid_bitmap));
2552 ghcb_set_sw_exit_info_1(ghcb, 2);
2553 ghcb_set_sw_exit_info_2(ghcb, reason);
2555 /* Resume the guest to "return" the error code. */
2559 void sev_es_unmap_ghcb(struct vcpu_svm *svm)
2561 if (!svm->sev_es.ghcb)
2564 if (svm->sev_es.ghcb_sa_free) {
2566 * The scratch area lives outside the GHCB, so there is a
2567 * buffer that, depending on the operation performed, may
2568 * need to be synced, then freed.
2570 if (svm->sev_es.ghcb_sa_sync) {
2571 kvm_write_guest(svm->vcpu.kvm,
2572 ghcb_get_sw_scratch(svm->sev_es.ghcb),
2573 svm->sev_es.ghcb_sa,
2574 svm->sev_es.ghcb_sa_len);
2575 svm->sev_es.ghcb_sa_sync = false;
2578 kvfree(svm->sev_es.ghcb_sa);
2579 svm->sev_es.ghcb_sa = NULL;
2580 svm->sev_es.ghcb_sa_free = false;
2583 trace_kvm_vmgexit_exit(svm->vcpu.vcpu_id, svm->sev_es.ghcb);
2585 sev_es_sync_to_ghcb(svm);
2587 kvm_vcpu_unmap(&svm->vcpu, &svm->sev_es.ghcb_map, true);
2588 svm->sev_es.ghcb = NULL;
2591 void pre_sev_run(struct vcpu_svm *svm, int cpu)
2593 struct svm_cpu_data *sd = per_cpu(svm_data, cpu);
2594 int asid = sev_get_asid(svm->vcpu.kvm);
2596 /* Assign the asid allocated with this SEV guest */
2602 * 1) when different VMCB for the same ASID is to be run on the same host CPU.
2603 * 2) or this VMCB was executed on different host CPU in previous VMRUNs.
2605 if (sd->sev_vmcbs[asid] == svm->vmcb &&
2606 svm->vcpu.arch.last_vmentry_cpu == cpu)
2609 sd->sev_vmcbs[asid] = svm->vmcb;
2610 svm->vmcb->control.tlb_ctl = TLB_CONTROL_FLUSH_ASID;
2611 vmcb_mark_dirty(svm->vmcb, VMCB_ASID);
2614 #define GHCB_SCRATCH_AREA_LIMIT (16ULL * PAGE_SIZE)
2615 static int setup_vmgexit_scratch(struct vcpu_svm *svm, bool sync, u64 len)
2617 struct vmcb_control_area *control = &svm->vmcb->control;
2618 struct ghcb *ghcb = svm->sev_es.ghcb;
2619 u64 ghcb_scratch_beg, ghcb_scratch_end;
2620 u64 scratch_gpa_beg, scratch_gpa_end;
2623 scratch_gpa_beg = ghcb_get_sw_scratch(ghcb);
2624 if (!scratch_gpa_beg) {
2625 pr_err("vmgexit: scratch gpa not provided\n");
2629 scratch_gpa_end = scratch_gpa_beg + len;
2630 if (scratch_gpa_end < scratch_gpa_beg) {
2631 pr_err("vmgexit: scratch length (%#llx) not valid for scratch address (%#llx)\n",
2632 len, scratch_gpa_beg);
2636 if ((scratch_gpa_beg & PAGE_MASK) == control->ghcb_gpa) {
2637 /* Scratch area begins within GHCB */
2638 ghcb_scratch_beg = control->ghcb_gpa +
2639 offsetof(struct ghcb, shared_buffer);
2640 ghcb_scratch_end = control->ghcb_gpa +
2641 offsetof(struct ghcb, reserved_1);
2644 * If the scratch area begins within the GHCB, it must be
2645 * completely contained in the GHCB shared buffer area.
2647 if (scratch_gpa_beg < ghcb_scratch_beg ||
2648 scratch_gpa_end > ghcb_scratch_end) {
2649 pr_err("vmgexit: scratch area is outside of GHCB shared buffer area (%#llx - %#llx)\n",
2650 scratch_gpa_beg, scratch_gpa_end);
2654 scratch_va = (void *)svm->sev_es.ghcb;
2655 scratch_va += (scratch_gpa_beg - control->ghcb_gpa);
2658 * The guest memory must be read into a kernel buffer, so
2661 if (len > GHCB_SCRATCH_AREA_LIMIT) {
2662 pr_err("vmgexit: scratch area exceeds KVM limits (%#llx requested, %#llx limit)\n",
2663 len, GHCB_SCRATCH_AREA_LIMIT);
2666 scratch_va = kvzalloc(len, GFP_KERNEL_ACCOUNT);
2670 if (kvm_read_guest(svm->vcpu.kvm, scratch_gpa_beg, scratch_va, len)) {
2671 /* Unable to copy scratch area from guest */
2672 pr_err("vmgexit: kvm_read_guest for scratch area failed\n");
2679 * The scratch area is outside the GHCB. The operation will
2680 * dictate whether the buffer needs to be synced before running
2681 * the vCPU next time (i.e. a read was requested so the data
2682 * must be written back to the guest memory).
2684 svm->sev_es.ghcb_sa_sync = sync;
2685 svm->sev_es.ghcb_sa_free = true;
2688 svm->sev_es.ghcb_sa = scratch_va;
2689 svm->sev_es.ghcb_sa_len = len;
2694 ghcb_set_sw_exit_info_1(ghcb, 2);
2695 ghcb_set_sw_exit_info_2(ghcb, GHCB_ERR_INVALID_SCRATCH_AREA);
2700 static void set_ghcb_msr_bits(struct vcpu_svm *svm, u64 value, u64 mask,
2703 svm->vmcb->control.ghcb_gpa &= ~(mask << pos);
2704 svm->vmcb->control.ghcb_gpa |= (value & mask) << pos;
2707 static u64 get_ghcb_msr_bits(struct vcpu_svm *svm, u64 mask, unsigned int pos)
2709 return (svm->vmcb->control.ghcb_gpa >> pos) & mask;
2712 static void set_ghcb_msr(struct vcpu_svm *svm, u64 value)
2714 svm->vmcb->control.ghcb_gpa = value;
2717 static int sev_handle_vmgexit_msr_protocol(struct vcpu_svm *svm)
2719 struct vmcb_control_area *control = &svm->vmcb->control;
2720 struct kvm_vcpu *vcpu = &svm->vcpu;
2724 ghcb_info = control->ghcb_gpa & GHCB_MSR_INFO_MASK;
2726 trace_kvm_vmgexit_msr_protocol_enter(svm->vcpu.vcpu_id,
2729 switch (ghcb_info) {
2730 case GHCB_MSR_SEV_INFO_REQ:
2731 set_ghcb_msr(svm, GHCB_MSR_SEV_INFO(GHCB_VERSION_MAX,
2735 case GHCB_MSR_CPUID_REQ: {
2736 u64 cpuid_fn, cpuid_reg, cpuid_value;
2738 cpuid_fn = get_ghcb_msr_bits(svm,
2739 GHCB_MSR_CPUID_FUNC_MASK,
2740 GHCB_MSR_CPUID_FUNC_POS);
2742 /* Initialize the registers needed by the CPUID intercept */
2743 vcpu->arch.regs[VCPU_REGS_RAX] = cpuid_fn;
2744 vcpu->arch.regs[VCPU_REGS_RCX] = 0;
2746 ret = svm_invoke_exit_handler(vcpu, SVM_EXIT_CPUID);
2748 /* Error, keep GHCB MSR value as-is */
2752 cpuid_reg = get_ghcb_msr_bits(svm,
2753 GHCB_MSR_CPUID_REG_MASK,
2754 GHCB_MSR_CPUID_REG_POS);
2756 cpuid_value = vcpu->arch.regs[VCPU_REGS_RAX];
2757 else if (cpuid_reg == 1)
2758 cpuid_value = vcpu->arch.regs[VCPU_REGS_RBX];
2759 else if (cpuid_reg == 2)
2760 cpuid_value = vcpu->arch.regs[VCPU_REGS_RCX];
2762 cpuid_value = vcpu->arch.regs[VCPU_REGS_RDX];
2764 set_ghcb_msr_bits(svm, cpuid_value,
2765 GHCB_MSR_CPUID_VALUE_MASK,
2766 GHCB_MSR_CPUID_VALUE_POS);
2768 set_ghcb_msr_bits(svm, GHCB_MSR_CPUID_RESP,
2773 case GHCB_MSR_TERM_REQ: {
2774 u64 reason_set, reason_code;
2776 reason_set = get_ghcb_msr_bits(svm,
2777 GHCB_MSR_TERM_REASON_SET_MASK,
2778 GHCB_MSR_TERM_REASON_SET_POS);
2779 reason_code = get_ghcb_msr_bits(svm,
2780 GHCB_MSR_TERM_REASON_MASK,
2781 GHCB_MSR_TERM_REASON_POS);
2782 pr_info("SEV-ES guest requested termination: %#llx:%#llx\n",
2783 reason_set, reason_code);
2785 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
2786 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_SEV_TERM;
2787 vcpu->run->system_event.ndata = 1;
2788 vcpu->run->system_event.data[0] = control->ghcb_gpa;
2793 /* Error, keep GHCB MSR value as-is */
2797 trace_kvm_vmgexit_msr_protocol_exit(svm->vcpu.vcpu_id,
2798 control->ghcb_gpa, ret);
2803 int sev_handle_vmgexit(struct kvm_vcpu *vcpu)
2805 struct vcpu_svm *svm = to_svm(vcpu);
2806 struct vmcb_control_area *control = &svm->vmcb->control;
2807 u64 ghcb_gpa, exit_code;
2811 /* Validate the GHCB */
2812 ghcb_gpa = control->ghcb_gpa;
2813 if (ghcb_gpa & GHCB_MSR_INFO_MASK)
2814 return sev_handle_vmgexit_msr_protocol(svm);
2817 vcpu_unimpl(vcpu, "vmgexit: GHCB gpa is not set\n");
2819 /* Without a GHCB, just return right back to the guest */
2823 if (kvm_vcpu_map(vcpu, ghcb_gpa >> PAGE_SHIFT, &svm->sev_es.ghcb_map)) {
2824 /* Unable to map GHCB from guest */
2825 vcpu_unimpl(vcpu, "vmgexit: error mapping GHCB [%#llx] from guest\n",
2828 /* Without a GHCB, just return right back to the guest */
2832 svm->sev_es.ghcb = svm->sev_es.ghcb_map.hva;
2833 ghcb = svm->sev_es.ghcb_map.hva;
2835 trace_kvm_vmgexit_enter(vcpu->vcpu_id, ghcb);
2837 exit_code = ghcb_get_sw_exit_code(ghcb);
2839 ret = sev_es_validate_vmgexit(svm);
2843 sev_es_sync_from_ghcb(svm);
2844 ghcb_set_sw_exit_info_1(ghcb, 0);
2845 ghcb_set_sw_exit_info_2(ghcb, 0);
2847 switch (exit_code) {
2848 case SVM_VMGEXIT_MMIO_READ:
2849 ret = setup_vmgexit_scratch(svm, true, control->exit_info_2);
2853 ret = kvm_sev_es_mmio_read(vcpu,
2854 control->exit_info_1,
2855 control->exit_info_2,
2856 svm->sev_es.ghcb_sa);
2858 case SVM_VMGEXIT_MMIO_WRITE:
2859 ret = setup_vmgexit_scratch(svm, false, control->exit_info_2);
2863 ret = kvm_sev_es_mmio_write(vcpu,
2864 control->exit_info_1,
2865 control->exit_info_2,
2866 svm->sev_es.ghcb_sa);
2868 case SVM_VMGEXIT_NMI_COMPLETE:
2869 ret = svm_invoke_exit_handler(vcpu, SVM_EXIT_IRET);
2871 case SVM_VMGEXIT_AP_HLT_LOOP:
2872 ret = kvm_emulate_ap_reset_hold(vcpu);
2874 case SVM_VMGEXIT_AP_JUMP_TABLE: {
2875 struct kvm_sev_info *sev = &to_kvm_svm(vcpu->kvm)->sev_info;
2877 switch (control->exit_info_1) {
2879 /* Set AP jump table address */
2880 sev->ap_jump_table = control->exit_info_2;
2883 /* Get AP jump table address */
2884 ghcb_set_sw_exit_info_2(ghcb, sev->ap_jump_table);
2887 pr_err("svm: vmgexit: unsupported AP jump table request - exit_info_1=%#llx\n",
2888 control->exit_info_1);
2889 ghcb_set_sw_exit_info_1(ghcb, 2);
2890 ghcb_set_sw_exit_info_2(ghcb, GHCB_ERR_INVALID_INPUT);
2896 case SVM_VMGEXIT_UNSUPPORTED_EVENT:
2898 "vmgexit: unsupported event - exit_info_1=%#llx, exit_info_2=%#llx\n",
2899 control->exit_info_1, control->exit_info_2);
2903 ret = svm_invoke_exit_handler(vcpu, exit_code);
2909 int sev_es_string_io(struct vcpu_svm *svm, int size, unsigned int port, int in)
2915 if (svm->vmcb->control.exit_info_2 > INT_MAX)
2918 count = svm->vmcb->control.exit_info_2;
2919 if (unlikely(check_mul_overflow(count, size, &bytes)))
2922 r = setup_vmgexit_scratch(svm, in, bytes);
2926 return kvm_sev_es_string_io(&svm->vcpu, size, port, svm->sev_es.ghcb_sa,
2930 static void sev_es_init_vmcb(struct vcpu_svm *svm)
2932 struct kvm_vcpu *vcpu = &svm->vcpu;
2934 svm->vmcb->control.nested_ctl |= SVM_NESTED_CTL_SEV_ES_ENABLE;
2935 svm->vmcb->control.virt_ext |= LBR_CTL_ENABLE_MASK;
2938 * An SEV-ES guest requires a VMSA area that is a separate from the
2939 * VMCB page. Do not include the encryption mask on the VMSA physical
2940 * address since hardware will access it using the guest key.
2942 svm->vmcb->control.vmsa_pa = __pa(svm->sev_es.vmsa);
2944 /* Can't intercept CR register access, HV can't modify CR registers */
2945 svm_clr_intercept(svm, INTERCEPT_CR0_READ);
2946 svm_clr_intercept(svm, INTERCEPT_CR4_READ);
2947 svm_clr_intercept(svm, INTERCEPT_CR8_READ);
2948 svm_clr_intercept(svm, INTERCEPT_CR0_WRITE);
2949 svm_clr_intercept(svm, INTERCEPT_CR4_WRITE);
2950 svm_clr_intercept(svm, INTERCEPT_CR8_WRITE);
2952 svm_clr_intercept(svm, INTERCEPT_SELECTIVE_CR0);
2954 /* Track EFER/CR register changes */
2955 svm_set_intercept(svm, TRAP_EFER_WRITE);
2956 svm_set_intercept(svm, TRAP_CR0_WRITE);
2957 svm_set_intercept(svm, TRAP_CR4_WRITE);
2958 svm_set_intercept(svm, TRAP_CR8_WRITE);
2960 /* No support for enable_vmware_backdoor */
2961 clr_exception_intercept(svm, GP_VECTOR);
2963 /* Can't intercept XSETBV, HV can't modify XCR0 directly */
2964 svm_clr_intercept(svm, INTERCEPT_XSETBV);
2966 /* Clear intercepts on selected MSRs */
2967 set_msr_interception(vcpu, svm->msrpm, MSR_EFER, 1, 1);
2968 set_msr_interception(vcpu, svm->msrpm, MSR_IA32_CR_PAT, 1, 1);
2969 set_msr_interception(vcpu, svm->msrpm, MSR_IA32_LASTBRANCHFROMIP, 1, 1);
2970 set_msr_interception(vcpu, svm->msrpm, MSR_IA32_LASTBRANCHTOIP, 1, 1);
2971 set_msr_interception(vcpu, svm->msrpm, MSR_IA32_LASTINTFROMIP, 1, 1);
2972 set_msr_interception(vcpu, svm->msrpm, MSR_IA32_LASTINTTOIP, 1, 1);
2974 if (boot_cpu_has(X86_FEATURE_V_TSC_AUX) &&
2975 (guest_cpuid_has(&svm->vcpu, X86_FEATURE_RDTSCP) ||
2976 guest_cpuid_has(&svm->vcpu, X86_FEATURE_RDPID))) {
2977 set_msr_interception(vcpu, svm->msrpm, MSR_TSC_AUX, 1, 1);
2978 if (guest_cpuid_has(&svm->vcpu, X86_FEATURE_RDTSCP))
2979 svm_clr_intercept(svm, INTERCEPT_RDTSCP);
2983 void sev_init_vmcb(struct vcpu_svm *svm)
2985 svm->vmcb->control.nested_ctl |= SVM_NESTED_CTL_SEV_ENABLE;
2986 clr_exception_intercept(svm, UD_VECTOR);
2988 if (sev_es_guest(svm->vcpu.kvm))
2989 sev_es_init_vmcb(svm);
2992 void sev_es_vcpu_reset(struct vcpu_svm *svm)
2995 * Set the GHCB MSR value as per the GHCB specification when emulating
2996 * vCPU RESET for an SEV-ES guest.
2998 set_ghcb_msr(svm, GHCB_MSR_SEV_INFO(GHCB_VERSION_MAX,
3003 void sev_es_prepare_switch_to_guest(struct sev_es_save_area *hostsa)
3006 * As an SEV-ES guest, hardware will restore the host state on VMEXIT,
3007 * of which one step is to perform a VMLOAD. KVM performs the
3008 * corresponding VMSAVE in svm_prepare_guest_switch for both
3009 * traditional and SEV-ES guests.
3012 /* XCR0 is restored on VMEXIT, save the current host value */
3013 hostsa->xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
3015 /* PKRU is restored on VMEXIT, save the current host value */
3016 hostsa->pkru = read_pkru();
3018 /* MSR_IA32_XSS is restored on VMEXIT, save the currnet host value */
3019 hostsa->xss = host_xss;
3022 void sev_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
3024 struct vcpu_svm *svm = to_svm(vcpu);
3026 /* First SIPI: Use the values as initially set by the VMM */
3027 if (!svm->sev_es.received_first_sipi) {
3028 svm->sev_es.received_first_sipi = true;
3033 * Subsequent SIPI: Return from an AP Reset Hold VMGEXIT, where
3034 * the guest will set the CS and RIP. Set SW_EXIT_INFO_2 to a
3037 if (!svm->sev_es.ghcb)
3040 ghcb_set_sw_exit_info_2(svm->sev_es.ghcb, 1);