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>
32 #ifndef CONFIG_KVM_AMD_SEV
34 * When this config is not defined, SEV feature is not supported and APIs in
35 * this file are not used but this file still gets compiled into the KVM AMD
38 * We will not have MISC_CG_RES_SEV and MISC_CG_RES_SEV_ES entries in the enum
39 * misc_res_type {} defined in linux/misc_cgroup.h.
41 * Below macros allow compilation to succeed.
43 #define MISC_CG_RES_SEV MISC_CG_RES_TYPES
44 #define MISC_CG_RES_SEV_ES MISC_CG_RES_TYPES
47 #ifdef CONFIG_KVM_AMD_SEV
48 /* enable/disable SEV support */
49 static bool sev_enabled = true;
50 module_param_named(sev, sev_enabled, bool, 0444);
52 /* enable/disable SEV-ES support */
53 static bool sev_es_enabled = true;
54 module_param_named(sev_es, sev_es_enabled, bool, 0444);
56 #define sev_enabled false
57 #define sev_es_enabled false
58 #endif /* CONFIG_KVM_AMD_SEV */
60 static u8 sev_enc_bit;
61 static DECLARE_RWSEM(sev_deactivate_lock);
62 static DEFINE_MUTEX(sev_bitmap_lock);
63 unsigned int max_sev_asid;
64 static unsigned int min_sev_asid;
65 static unsigned long sev_me_mask;
66 static unsigned int nr_asids;
67 static unsigned long *sev_asid_bitmap;
68 static unsigned long *sev_reclaim_asid_bitmap;
71 struct list_head list;
78 /* Called with the sev_bitmap_lock held, or on shutdown */
79 static int sev_flush_asids(int min_asid, int max_asid)
81 int ret, asid, error = 0;
83 /* Check if there are any ASIDs to reclaim before performing a flush */
84 asid = find_next_bit(sev_reclaim_asid_bitmap, nr_asids, min_asid);
89 * DEACTIVATE will clear the WBINVD indicator causing DF_FLUSH to fail,
90 * so it must be guarded.
92 down_write(&sev_deactivate_lock);
95 ret = sev_guest_df_flush(&error);
97 up_write(&sev_deactivate_lock);
100 pr_err("SEV: DF_FLUSH failed, ret=%d, error=%#x\n", ret, error);
105 static inline bool is_mirroring_enc_context(struct kvm *kvm)
107 return !!to_kvm_svm(kvm)->sev_info.enc_context_owner;
110 /* Must be called with the sev_bitmap_lock held */
111 static bool __sev_recycle_asids(int min_asid, int max_asid)
113 if (sev_flush_asids(min_asid, max_asid))
116 /* The flush process will flush all reclaimable SEV and SEV-ES ASIDs */
117 bitmap_xor(sev_asid_bitmap, sev_asid_bitmap, sev_reclaim_asid_bitmap,
119 bitmap_zero(sev_reclaim_asid_bitmap, nr_asids);
124 static int sev_misc_cg_try_charge(struct kvm_sev_info *sev)
126 enum misc_res_type type = sev->es_active ? MISC_CG_RES_SEV_ES : MISC_CG_RES_SEV;
127 return misc_cg_try_charge(type, sev->misc_cg, 1);
130 static void sev_misc_cg_uncharge(struct kvm_sev_info *sev)
132 enum misc_res_type type = sev->es_active ? MISC_CG_RES_SEV_ES : MISC_CG_RES_SEV;
133 misc_cg_uncharge(type, sev->misc_cg, 1);
136 static int sev_asid_new(struct kvm_sev_info *sev)
138 int asid, min_asid, max_asid, ret;
141 WARN_ON(sev->misc_cg);
142 sev->misc_cg = get_current_misc_cg();
143 ret = sev_misc_cg_try_charge(sev);
145 put_misc_cg(sev->misc_cg);
150 mutex_lock(&sev_bitmap_lock);
153 * SEV-enabled guests must use asid from min_sev_asid to max_sev_asid.
154 * SEV-ES-enabled guest can use from 1 to min_sev_asid - 1.
156 min_asid = sev->es_active ? 1 : min_sev_asid;
157 max_asid = sev->es_active ? min_sev_asid - 1 : max_sev_asid;
159 asid = find_next_zero_bit(sev_asid_bitmap, max_asid + 1, min_asid);
160 if (asid > max_asid) {
161 if (retry && __sev_recycle_asids(min_asid, max_asid)) {
165 mutex_unlock(&sev_bitmap_lock);
170 __set_bit(asid, sev_asid_bitmap);
172 mutex_unlock(&sev_bitmap_lock);
176 sev_misc_cg_uncharge(sev);
177 put_misc_cg(sev->misc_cg);
182 static int sev_get_asid(struct kvm *kvm)
184 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
189 static void sev_asid_free(struct kvm_sev_info *sev)
191 struct svm_cpu_data *sd;
194 mutex_lock(&sev_bitmap_lock);
196 __set_bit(sev->asid, sev_reclaim_asid_bitmap);
198 for_each_possible_cpu(cpu) {
199 sd = per_cpu_ptr(&svm_data, cpu);
200 sd->sev_vmcbs[sev->asid] = NULL;
203 mutex_unlock(&sev_bitmap_lock);
205 sev_misc_cg_uncharge(sev);
206 put_misc_cg(sev->misc_cg);
210 static void sev_decommission(unsigned int handle)
212 struct sev_data_decommission decommission;
217 decommission.handle = handle;
218 sev_guest_decommission(&decommission, NULL);
221 static void sev_unbind_asid(struct kvm *kvm, unsigned int handle)
223 struct sev_data_deactivate deactivate;
228 deactivate.handle = handle;
230 /* Guard DEACTIVATE against WBINVD/DF_FLUSH used in ASID recycling */
231 down_read(&sev_deactivate_lock);
232 sev_guest_deactivate(&deactivate, NULL);
233 up_read(&sev_deactivate_lock);
235 sev_decommission(handle);
238 static int sev_guest_init(struct kvm *kvm, struct kvm_sev_cmd *argp)
240 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
243 if (kvm->created_vcpus)
247 if (unlikely(sev->active))
251 sev->es_active = argp->id == KVM_SEV_ES_INIT;
252 asid = sev_asid_new(sev);
257 ret = sev_platform_init(&argp->error);
261 INIT_LIST_HEAD(&sev->regions_list);
262 INIT_LIST_HEAD(&sev->mirror_vms);
264 kvm_set_apicv_inhibit(kvm, APICV_INHIBIT_REASON_SEV);
272 sev->es_active = false;
277 static int sev_bind_asid(struct kvm *kvm, unsigned int handle, int *error)
279 struct sev_data_activate activate;
280 int asid = sev_get_asid(kvm);
283 /* activate ASID on the given handle */
284 activate.handle = handle;
285 activate.asid = asid;
286 ret = sev_guest_activate(&activate, error);
291 static int __sev_issue_cmd(int fd, int id, void *data, int *error)
300 ret = sev_issue_cmd_external_user(f.file, id, data, error);
306 static int sev_issue_cmd(struct kvm *kvm, int id, void *data, int *error)
308 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
310 return __sev_issue_cmd(sev->fd, id, data, error);
313 static int sev_launch_start(struct kvm *kvm, struct kvm_sev_cmd *argp)
315 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
316 struct sev_data_launch_start start;
317 struct kvm_sev_launch_start params;
318 void *dh_blob, *session_blob;
319 int *error = &argp->error;
325 if (copy_from_user(¶ms, (void __user *)(uintptr_t)argp->data, sizeof(params)))
328 memset(&start, 0, sizeof(start));
331 if (params.dh_uaddr) {
332 dh_blob = psp_copy_user_blob(params.dh_uaddr, params.dh_len);
334 return PTR_ERR(dh_blob);
336 start.dh_cert_address = __sme_set(__pa(dh_blob));
337 start.dh_cert_len = params.dh_len;
341 if (params.session_uaddr) {
342 session_blob = psp_copy_user_blob(params.session_uaddr, params.session_len);
343 if (IS_ERR(session_blob)) {
344 ret = PTR_ERR(session_blob);
348 start.session_address = __sme_set(__pa(session_blob));
349 start.session_len = params.session_len;
352 start.handle = params.handle;
353 start.policy = params.policy;
355 /* create memory encryption context */
356 ret = __sev_issue_cmd(argp->sev_fd, SEV_CMD_LAUNCH_START, &start, error);
360 /* Bind ASID to this guest */
361 ret = sev_bind_asid(kvm, start.handle, error);
363 sev_decommission(start.handle);
367 /* return handle to userspace */
368 params.handle = start.handle;
369 if (copy_to_user((void __user *)(uintptr_t)argp->data, ¶ms, sizeof(params))) {
370 sev_unbind_asid(kvm, start.handle);
375 sev->handle = start.handle;
376 sev->fd = argp->sev_fd;
385 static struct page **sev_pin_memory(struct kvm *kvm, unsigned long uaddr,
386 unsigned long ulen, unsigned long *n,
389 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
390 unsigned long npages, size;
392 unsigned long locked, lock_limit;
394 unsigned long first, last;
397 lockdep_assert_held(&kvm->lock);
399 if (ulen == 0 || uaddr + ulen < uaddr)
400 return ERR_PTR(-EINVAL);
402 /* Calculate number of pages. */
403 first = (uaddr & PAGE_MASK) >> PAGE_SHIFT;
404 last = ((uaddr + ulen - 1) & PAGE_MASK) >> PAGE_SHIFT;
405 npages = (last - first + 1);
407 locked = sev->pages_locked + npages;
408 lock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
409 if (locked > lock_limit && !capable(CAP_IPC_LOCK)) {
410 pr_err("SEV: %lu locked pages exceed the lock limit of %lu.\n", locked, lock_limit);
411 return ERR_PTR(-ENOMEM);
414 if (WARN_ON_ONCE(npages > INT_MAX))
415 return ERR_PTR(-EINVAL);
417 /* Avoid using vmalloc for smaller buffers. */
418 size = npages * sizeof(struct page *);
419 if (size > PAGE_SIZE)
420 pages = __vmalloc(size, GFP_KERNEL_ACCOUNT | __GFP_ZERO);
422 pages = kmalloc(size, GFP_KERNEL_ACCOUNT);
425 return ERR_PTR(-ENOMEM);
427 /* Pin the user virtual address. */
428 npinned = pin_user_pages_fast(uaddr, npages, write ? FOLL_WRITE : 0, pages);
429 if (npinned != npages) {
430 pr_err("SEV: Failure locking %lu pages.\n", npages);
436 sev->pages_locked = locked;
442 unpin_user_pages(pages, npinned);
448 static void sev_unpin_memory(struct kvm *kvm, struct page **pages,
449 unsigned long npages)
451 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
453 unpin_user_pages(pages, npages);
455 sev->pages_locked -= npages;
458 static void sev_clflush_pages(struct page *pages[], unsigned long npages)
460 uint8_t *page_virtual;
463 if (this_cpu_has(X86_FEATURE_SME_COHERENT) || npages == 0 ||
467 for (i = 0; i < npages; i++) {
468 page_virtual = kmap_local_page(pages[i]);
469 clflush_cache_range(page_virtual, PAGE_SIZE);
470 kunmap_local(page_virtual);
475 static unsigned long get_num_contig_pages(unsigned long idx,
476 struct page **inpages, unsigned long npages)
478 unsigned long paddr, next_paddr;
479 unsigned long i = idx + 1, pages = 1;
481 /* find the number of contiguous pages starting from idx */
482 paddr = __sme_page_pa(inpages[idx]);
484 next_paddr = __sme_page_pa(inpages[i++]);
485 if ((paddr + PAGE_SIZE) == next_paddr) {
496 static int sev_launch_update_data(struct kvm *kvm, struct kvm_sev_cmd *argp)
498 unsigned long vaddr, vaddr_end, next_vaddr, npages, pages, size, i;
499 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
500 struct kvm_sev_launch_update_data params;
501 struct sev_data_launch_update_data data;
502 struct page **inpages;
508 if (copy_from_user(¶ms, (void __user *)(uintptr_t)argp->data, sizeof(params)))
511 vaddr = params.uaddr;
513 vaddr_end = vaddr + size;
515 /* Lock the user memory. */
516 inpages = sev_pin_memory(kvm, vaddr, size, &npages, 1);
518 return PTR_ERR(inpages);
521 * Flush (on non-coherent CPUs) before LAUNCH_UPDATE encrypts pages in
522 * place; the cache may contain the data that was written unencrypted.
524 sev_clflush_pages(inpages, npages);
527 data.handle = sev->handle;
529 for (i = 0; vaddr < vaddr_end; vaddr = next_vaddr, i += pages) {
533 * If the user buffer is not page-aligned, calculate the offset
536 offset = vaddr & (PAGE_SIZE - 1);
538 /* Calculate the number of pages that can be encrypted in one go. */
539 pages = get_num_contig_pages(i, inpages, npages);
541 len = min_t(size_t, ((pages * PAGE_SIZE) - offset), size);
544 data.address = __sme_page_pa(inpages[i]) + offset;
545 ret = sev_issue_cmd(kvm, SEV_CMD_LAUNCH_UPDATE_DATA, &data, &argp->error);
550 next_vaddr = vaddr + len;
554 /* content of memory is updated, mark pages dirty */
555 for (i = 0; i < npages; i++) {
556 set_page_dirty_lock(inpages[i]);
557 mark_page_accessed(inpages[i]);
559 /* unlock the user pages */
560 sev_unpin_memory(kvm, inpages, npages);
564 static int sev_es_sync_vmsa(struct vcpu_svm *svm)
566 struct sev_es_save_area *save = svm->sev_es.vmsa;
568 /* Check some debug related fields before encrypting the VMSA */
569 if (svm->vcpu.guest_debug || (svm->vmcb->save.dr7 & ~DR7_FIXED_1))
573 * SEV-ES will use a VMSA that is pointed to by the VMCB, not
574 * the traditional VMSA that is part of the VMCB. Copy the
575 * traditional VMSA as it has been built so far (in prep
576 * for LAUNCH_UPDATE_VMSA) to be the initial SEV-ES state.
578 memcpy(save, &svm->vmcb->save, sizeof(svm->vmcb->save));
580 /* Sync registgers */
581 save->rax = svm->vcpu.arch.regs[VCPU_REGS_RAX];
582 save->rbx = svm->vcpu.arch.regs[VCPU_REGS_RBX];
583 save->rcx = svm->vcpu.arch.regs[VCPU_REGS_RCX];
584 save->rdx = svm->vcpu.arch.regs[VCPU_REGS_RDX];
585 save->rsp = svm->vcpu.arch.regs[VCPU_REGS_RSP];
586 save->rbp = svm->vcpu.arch.regs[VCPU_REGS_RBP];
587 save->rsi = svm->vcpu.arch.regs[VCPU_REGS_RSI];
588 save->rdi = svm->vcpu.arch.regs[VCPU_REGS_RDI];
590 save->r8 = svm->vcpu.arch.regs[VCPU_REGS_R8];
591 save->r9 = svm->vcpu.arch.regs[VCPU_REGS_R9];
592 save->r10 = svm->vcpu.arch.regs[VCPU_REGS_R10];
593 save->r11 = svm->vcpu.arch.regs[VCPU_REGS_R11];
594 save->r12 = svm->vcpu.arch.regs[VCPU_REGS_R12];
595 save->r13 = svm->vcpu.arch.regs[VCPU_REGS_R13];
596 save->r14 = svm->vcpu.arch.regs[VCPU_REGS_R14];
597 save->r15 = svm->vcpu.arch.regs[VCPU_REGS_R15];
599 save->rip = svm->vcpu.arch.regs[VCPU_REGS_RIP];
601 /* Sync some non-GPR registers before encrypting */
602 save->xcr0 = svm->vcpu.arch.xcr0;
603 save->pkru = svm->vcpu.arch.pkru;
604 save->xss = svm->vcpu.arch.ia32_xss;
605 save->dr6 = svm->vcpu.arch.dr6;
607 pr_debug("Virtual Machine Save Area (VMSA):\n");
608 print_hex_dump_debug("", DUMP_PREFIX_NONE, 16, 1, save, sizeof(*save), false);
613 static int __sev_launch_update_vmsa(struct kvm *kvm, struct kvm_vcpu *vcpu,
616 struct sev_data_launch_update_vmsa vmsa;
617 struct vcpu_svm *svm = to_svm(vcpu);
620 /* Perform some pre-encryption checks against the VMSA */
621 ret = sev_es_sync_vmsa(svm);
626 * The LAUNCH_UPDATE_VMSA command will perform in-place encryption of
627 * the VMSA memory content (i.e it will write the same memory region
628 * with the guest's key), so invalidate it first.
630 clflush_cache_range(svm->sev_es.vmsa, PAGE_SIZE);
633 vmsa.handle = to_kvm_svm(kvm)->sev_info.handle;
634 vmsa.address = __sme_pa(svm->sev_es.vmsa);
635 vmsa.len = PAGE_SIZE;
636 ret = sev_issue_cmd(kvm, SEV_CMD_LAUNCH_UPDATE_VMSA, &vmsa, error);
640 vcpu->arch.guest_state_protected = true;
644 static int sev_launch_update_vmsa(struct kvm *kvm, struct kvm_sev_cmd *argp)
646 struct kvm_vcpu *vcpu;
650 if (!sev_es_guest(kvm))
653 kvm_for_each_vcpu(i, vcpu, kvm) {
654 ret = mutex_lock_killable(&vcpu->mutex);
658 ret = __sev_launch_update_vmsa(kvm, vcpu, &argp->error);
660 mutex_unlock(&vcpu->mutex);
668 static int sev_launch_measure(struct kvm *kvm, struct kvm_sev_cmd *argp)
670 void __user *measure = (void __user *)(uintptr_t)argp->data;
671 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
672 struct sev_data_launch_measure data;
673 struct kvm_sev_launch_measure params;
674 void __user *p = NULL;
681 if (copy_from_user(¶ms, measure, sizeof(params)))
684 memset(&data, 0, sizeof(data));
686 /* User wants to query the blob length */
690 p = (void __user *)(uintptr_t)params.uaddr;
692 if (params.len > SEV_FW_BLOB_MAX_SIZE)
695 blob = kzalloc(params.len, GFP_KERNEL_ACCOUNT);
699 data.address = __psp_pa(blob);
700 data.len = params.len;
704 data.handle = sev->handle;
705 ret = sev_issue_cmd(kvm, SEV_CMD_LAUNCH_MEASURE, &data, &argp->error);
708 * If we query the session length, FW responded with expected data.
717 if (copy_to_user(p, blob, params.len))
722 params.len = data.len;
723 if (copy_to_user(measure, ¶ms, sizeof(params)))
730 static int sev_launch_finish(struct kvm *kvm, struct kvm_sev_cmd *argp)
732 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
733 struct sev_data_launch_finish data;
738 data.handle = sev->handle;
739 return sev_issue_cmd(kvm, SEV_CMD_LAUNCH_FINISH, &data, &argp->error);
742 static int sev_guest_status(struct kvm *kvm, struct kvm_sev_cmd *argp)
744 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
745 struct kvm_sev_guest_status params;
746 struct sev_data_guest_status data;
752 memset(&data, 0, sizeof(data));
754 data.handle = sev->handle;
755 ret = sev_issue_cmd(kvm, SEV_CMD_GUEST_STATUS, &data, &argp->error);
759 params.policy = data.policy;
760 params.state = data.state;
761 params.handle = data.handle;
763 if (copy_to_user((void __user *)(uintptr_t)argp->data, ¶ms, sizeof(params)))
769 static int __sev_issue_dbg_cmd(struct kvm *kvm, unsigned long src,
770 unsigned long dst, int size,
771 int *error, bool enc)
773 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
774 struct sev_data_dbg data;
777 data.handle = sev->handle;
782 return sev_issue_cmd(kvm,
783 enc ? SEV_CMD_DBG_ENCRYPT : SEV_CMD_DBG_DECRYPT,
787 static int __sev_dbg_decrypt(struct kvm *kvm, unsigned long src_paddr,
788 unsigned long dst_paddr, int sz, int *err)
793 * Its safe to read more than we are asked, caller should ensure that
794 * destination has enough space.
796 offset = src_paddr & 15;
797 src_paddr = round_down(src_paddr, 16);
798 sz = round_up(sz + offset, 16);
800 return __sev_issue_dbg_cmd(kvm, src_paddr, dst_paddr, sz, err, false);
803 static int __sev_dbg_decrypt_user(struct kvm *kvm, unsigned long paddr,
804 void __user *dst_uaddr,
805 unsigned long dst_paddr,
808 struct page *tpage = NULL;
811 /* if inputs are not 16-byte then use intermediate buffer */
812 if (!IS_ALIGNED(dst_paddr, 16) ||
813 !IS_ALIGNED(paddr, 16) ||
814 !IS_ALIGNED(size, 16)) {
815 tpage = (void *)alloc_page(GFP_KERNEL | __GFP_ZERO);
819 dst_paddr = __sme_page_pa(tpage);
822 ret = __sev_dbg_decrypt(kvm, paddr, dst_paddr, size, err);
828 if (copy_to_user(dst_uaddr, page_address(tpage) + offset, size))
839 static int __sev_dbg_encrypt_user(struct kvm *kvm, unsigned long paddr,
841 unsigned long dst_paddr,
842 void __user *dst_vaddr,
843 int size, int *error)
845 struct page *src_tpage = NULL;
846 struct page *dst_tpage = NULL;
849 /* If source buffer is not aligned then use an intermediate buffer */
850 if (!IS_ALIGNED((unsigned long)vaddr, 16)) {
851 src_tpage = alloc_page(GFP_KERNEL_ACCOUNT);
855 if (copy_from_user(page_address(src_tpage), vaddr, size)) {
856 __free_page(src_tpage);
860 paddr = __sme_page_pa(src_tpage);
864 * If destination buffer or length is not aligned then do read-modify-write:
865 * - decrypt destination in an intermediate buffer
866 * - copy the source buffer in an intermediate buffer
867 * - use the intermediate buffer as source buffer
869 if (!IS_ALIGNED((unsigned long)dst_vaddr, 16) || !IS_ALIGNED(size, 16)) {
872 dst_tpage = alloc_page(GFP_KERNEL_ACCOUNT);
878 ret = __sev_dbg_decrypt(kvm, dst_paddr,
879 __sme_page_pa(dst_tpage), size, error);
884 * If source is kernel buffer then use memcpy() otherwise
887 dst_offset = dst_paddr & 15;
890 memcpy(page_address(dst_tpage) + dst_offset,
891 page_address(src_tpage), size);
893 if (copy_from_user(page_address(dst_tpage) + dst_offset,
900 paddr = __sme_page_pa(dst_tpage);
901 dst_paddr = round_down(dst_paddr, 16);
902 len = round_up(size, 16);
905 ret = __sev_issue_dbg_cmd(kvm, paddr, dst_paddr, len, error, true);
909 __free_page(src_tpage);
911 __free_page(dst_tpage);
915 static int sev_dbg_crypt(struct kvm *kvm, struct kvm_sev_cmd *argp, bool dec)
917 unsigned long vaddr, vaddr_end, next_vaddr;
918 unsigned long dst_vaddr;
919 struct page **src_p, **dst_p;
920 struct kvm_sev_dbg debug;
928 if (copy_from_user(&debug, (void __user *)(uintptr_t)argp->data, sizeof(debug)))
931 if (!debug.len || debug.src_uaddr + debug.len < debug.src_uaddr)
933 if (!debug.dst_uaddr)
936 vaddr = debug.src_uaddr;
938 vaddr_end = vaddr + size;
939 dst_vaddr = debug.dst_uaddr;
941 for (; vaddr < vaddr_end; vaddr = next_vaddr) {
942 int len, s_off, d_off;
944 /* lock userspace source and destination page */
945 src_p = sev_pin_memory(kvm, vaddr & PAGE_MASK, PAGE_SIZE, &n, 0);
947 return PTR_ERR(src_p);
949 dst_p = sev_pin_memory(kvm, dst_vaddr & PAGE_MASK, PAGE_SIZE, &n, 1);
951 sev_unpin_memory(kvm, src_p, n);
952 return PTR_ERR(dst_p);
956 * Flush (on non-coherent CPUs) before DBG_{DE,EN}CRYPT read or modify
957 * the pages; flush the destination too so that future accesses do not
960 sev_clflush_pages(src_p, 1);
961 sev_clflush_pages(dst_p, 1);
964 * Since user buffer may not be page aligned, calculate the
965 * offset within the page.
967 s_off = vaddr & ~PAGE_MASK;
968 d_off = dst_vaddr & ~PAGE_MASK;
969 len = min_t(size_t, (PAGE_SIZE - s_off), size);
972 ret = __sev_dbg_decrypt_user(kvm,
973 __sme_page_pa(src_p[0]) + s_off,
974 (void __user *)dst_vaddr,
975 __sme_page_pa(dst_p[0]) + d_off,
978 ret = __sev_dbg_encrypt_user(kvm,
979 __sme_page_pa(src_p[0]) + s_off,
980 (void __user *)vaddr,
981 __sme_page_pa(dst_p[0]) + d_off,
982 (void __user *)dst_vaddr,
985 sev_unpin_memory(kvm, src_p, n);
986 sev_unpin_memory(kvm, dst_p, n);
991 next_vaddr = vaddr + len;
992 dst_vaddr = dst_vaddr + len;
999 static int sev_launch_secret(struct kvm *kvm, struct kvm_sev_cmd *argp)
1001 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1002 struct sev_data_launch_secret data;
1003 struct kvm_sev_launch_secret params;
1004 struct page **pages;
1009 if (!sev_guest(kvm))
1012 if (copy_from_user(¶ms, (void __user *)(uintptr_t)argp->data, sizeof(params)))
1015 pages = sev_pin_memory(kvm, params.guest_uaddr, params.guest_len, &n, 1);
1017 return PTR_ERR(pages);
1020 * Flush (on non-coherent CPUs) before LAUNCH_SECRET encrypts pages in
1021 * place; the cache may contain the data that was written unencrypted.
1023 sev_clflush_pages(pages, n);
1026 * The secret must be copied into contiguous memory region, lets verify
1027 * that userspace memory pages are contiguous before we issue command.
1029 if (get_num_contig_pages(0, pages, n) != n) {
1031 goto e_unpin_memory;
1034 memset(&data, 0, sizeof(data));
1036 offset = params.guest_uaddr & (PAGE_SIZE - 1);
1037 data.guest_address = __sme_page_pa(pages[0]) + offset;
1038 data.guest_len = params.guest_len;
1040 blob = psp_copy_user_blob(params.trans_uaddr, params.trans_len);
1042 ret = PTR_ERR(blob);
1043 goto e_unpin_memory;
1046 data.trans_address = __psp_pa(blob);
1047 data.trans_len = params.trans_len;
1049 hdr = psp_copy_user_blob(params.hdr_uaddr, params.hdr_len);
1054 data.hdr_address = __psp_pa(hdr);
1055 data.hdr_len = params.hdr_len;
1057 data.handle = sev->handle;
1058 ret = sev_issue_cmd(kvm, SEV_CMD_LAUNCH_UPDATE_SECRET, &data, &argp->error);
1065 /* content of memory is updated, mark pages dirty */
1066 for (i = 0; i < n; i++) {
1067 set_page_dirty_lock(pages[i]);
1068 mark_page_accessed(pages[i]);
1070 sev_unpin_memory(kvm, pages, n);
1074 static int sev_get_attestation_report(struct kvm *kvm, struct kvm_sev_cmd *argp)
1076 void __user *report = (void __user *)(uintptr_t)argp->data;
1077 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1078 struct sev_data_attestation_report data;
1079 struct kvm_sev_attestation_report params;
1084 if (!sev_guest(kvm))
1087 if (copy_from_user(¶ms, (void __user *)(uintptr_t)argp->data, sizeof(params)))
1090 memset(&data, 0, sizeof(data));
1092 /* User wants to query the blob length */
1096 p = (void __user *)(uintptr_t)params.uaddr;
1098 if (params.len > SEV_FW_BLOB_MAX_SIZE)
1101 blob = kzalloc(params.len, GFP_KERNEL_ACCOUNT);
1105 data.address = __psp_pa(blob);
1106 data.len = params.len;
1107 memcpy(data.mnonce, params.mnonce, sizeof(params.mnonce));
1110 data.handle = sev->handle;
1111 ret = sev_issue_cmd(kvm, SEV_CMD_ATTESTATION_REPORT, &data, &argp->error);
1113 * If we query the session length, FW responded with expected data.
1122 if (copy_to_user(p, blob, params.len))
1127 params.len = data.len;
1128 if (copy_to_user(report, ¶ms, sizeof(params)))
1135 /* Userspace wants to query session length. */
1137 __sev_send_start_query_session_length(struct kvm *kvm, struct kvm_sev_cmd *argp,
1138 struct kvm_sev_send_start *params)
1140 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1141 struct sev_data_send_start data;
1144 memset(&data, 0, sizeof(data));
1145 data.handle = sev->handle;
1146 ret = sev_issue_cmd(kvm, SEV_CMD_SEND_START, &data, &argp->error);
1148 params->session_len = data.session_len;
1149 if (copy_to_user((void __user *)(uintptr_t)argp->data, params,
1150 sizeof(struct kvm_sev_send_start)))
1156 static int sev_send_start(struct kvm *kvm, struct kvm_sev_cmd *argp)
1158 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1159 struct sev_data_send_start data;
1160 struct kvm_sev_send_start params;
1161 void *amd_certs, *session_data;
1162 void *pdh_cert, *plat_certs;
1165 if (!sev_guest(kvm))
1168 if (copy_from_user(¶ms, (void __user *)(uintptr_t)argp->data,
1169 sizeof(struct kvm_sev_send_start)))
1172 /* if session_len is zero, userspace wants to query the session length */
1173 if (!params.session_len)
1174 return __sev_send_start_query_session_length(kvm, argp,
1177 /* some sanity checks */
1178 if (!params.pdh_cert_uaddr || !params.pdh_cert_len ||
1179 !params.session_uaddr || params.session_len > SEV_FW_BLOB_MAX_SIZE)
1182 /* allocate the memory to hold the session data blob */
1183 session_data = kzalloc(params.session_len, GFP_KERNEL_ACCOUNT);
1187 /* copy the certificate blobs from userspace */
1188 pdh_cert = psp_copy_user_blob(params.pdh_cert_uaddr,
1189 params.pdh_cert_len);
1190 if (IS_ERR(pdh_cert)) {
1191 ret = PTR_ERR(pdh_cert);
1192 goto e_free_session;
1195 plat_certs = psp_copy_user_blob(params.plat_certs_uaddr,
1196 params.plat_certs_len);
1197 if (IS_ERR(plat_certs)) {
1198 ret = PTR_ERR(plat_certs);
1202 amd_certs = psp_copy_user_blob(params.amd_certs_uaddr,
1203 params.amd_certs_len);
1204 if (IS_ERR(amd_certs)) {
1205 ret = PTR_ERR(amd_certs);
1206 goto e_free_plat_cert;
1209 /* populate the FW SEND_START field with system physical address */
1210 memset(&data, 0, sizeof(data));
1211 data.pdh_cert_address = __psp_pa(pdh_cert);
1212 data.pdh_cert_len = params.pdh_cert_len;
1213 data.plat_certs_address = __psp_pa(plat_certs);
1214 data.plat_certs_len = params.plat_certs_len;
1215 data.amd_certs_address = __psp_pa(amd_certs);
1216 data.amd_certs_len = params.amd_certs_len;
1217 data.session_address = __psp_pa(session_data);
1218 data.session_len = params.session_len;
1219 data.handle = sev->handle;
1221 ret = sev_issue_cmd(kvm, SEV_CMD_SEND_START, &data, &argp->error);
1223 if (!ret && copy_to_user((void __user *)(uintptr_t)params.session_uaddr,
1224 session_data, params.session_len)) {
1226 goto e_free_amd_cert;
1229 params.policy = data.policy;
1230 params.session_len = data.session_len;
1231 if (copy_to_user((void __user *)(uintptr_t)argp->data, ¶ms,
1232 sizeof(struct kvm_sev_send_start)))
1242 kfree(session_data);
1246 /* Userspace wants to query either header or trans length. */
1248 __sev_send_update_data_query_lengths(struct kvm *kvm, struct kvm_sev_cmd *argp,
1249 struct kvm_sev_send_update_data *params)
1251 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1252 struct sev_data_send_update_data data;
1255 memset(&data, 0, sizeof(data));
1256 data.handle = sev->handle;
1257 ret = sev_issue_cmd(kvm, SEV_CMD_SEND_UPDATE_DATA, &data, &argp->error);
1259 params->hdr_len = data.hdr_len;
1260 params->trans_len = data.trans_len;
1262 if (copy_to_user((void __user *)(uintptr_t)argp->data, params,
1263 sizeof(struct kvm_sev_send_update_data)))
1269 static int sev_send_update_data(struct kvm *kvm, struct kvm_sev_cmd *argp)
1271 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1272 struct sev_data_send_update_data data;
1273 struct kvm_sev_send_update_data params;
1274 void *hdr, *trans_data;
1275 struct page **guest_page;
1279 if (!sev_guest(kvm))
1282 if (copy_from_user(¶ms, (void __user *)(uintptr_t)argp->data,
1283 sizeof(struct kvm_sev_send_update_data)))
1286 /* userspace wants to query either header or trans length */
1287 if (!params.trans_len || !params.hdr_len)
1288 return __sev_send_update_data_query_lengths(kvm, argp, ¶ms);
1290 if (!params.trans_uaddr || !params.guest_uaddr ||
1291 !params.guest_len || !params.hdr_uaddr)
1294 /* Check if we are crossing the page boundary */
1295 offset = params.guest_uaddr & (PAGE_SIZE - 1);
1296 if ((params.guest_len + offset > PAGE_SIZE))
1299 /* Pin guest memory */
1300 guest_page = sev_pin_memory(kvm, params.guest_uaddr & PAGE_MASK,
1302 if (IS_ERR(guest_page))
1303 return PTR_ERR(guest_page);
1305 /* allocate memory for header and transport buffer */
1307 hdr = kzalloc(params.hdr_len, GFP_KERNEL_ACCOUNT);
1311 trans_data = kzalloc(params.trans_len, GFP_KERNEL_ACCOUNT);
1315 memset(&data, 0, sizeof(data));
1316 data.hdr_address = __psp_pa(hdr);
1317 data.hdr_len = params.hdr_len;
1318 data.trans_address = __psp_pa(trans_data);
1319 data.trans_len = params.trans_len;
1321 /* The SEND_UPDATE_DATA command requires C-bit to be always set. */
1322 data.guest_address = (page_to_pfn(guest_page[0]) << PAGE_SHIFT) + offset;
1323 data.guest_address |= sev_me_mask;
1324 data.guest_len = params.guest_len;
1325 data.handle = sev->handle;
1327 ret = sev_issue_cmd(kvm, SEV_CMD_SEND_UPDATE_DATA, &data, &argp->error);
1330 goto e_free_trans_data;
1332 /* copy transport buffer to user space */
1333 if (copy_to_user((void __user *)(uintptr_t)params.trans_uaddr,
1334 trans_data, params.trans_len)) {
1336 goto e_free_trans_data;
1339 /* Copy packet header to userspace. */
1340 if (copy_to_user((void __user *)(uintptr_t)params.hdr_uaddr, hdr,
1349 sev_unpin_memory(kvm, guest_page, n);
1354 static int sev_send_finish(struct kvm *kvm, struct kvm_sev_cmd *argp)
1356 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1357 struct sev_data_send_finish data;
1359 if (!sev_guest(kvm))
1362 data.handle = sev->handle;
1363 return sev_issue_cmd(kvm, SEV_CMD_SEND_FINISH, &data, &argp->error);
1366 static int sev_send_cancel(struct kvm *kvm, struct kvm_sev_cmd *argp)
1368 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1369 struct sev_data_send_cancel data;
1371 if (!sev_guest(kvm))
1374 data.handle = sev->handle;
1375 return sev_issue_cmd(kvm, SEV_CMD_SEND_CANCEL, &data, &argp->error);
1378 static int sev_receive_start(struct kvm *kvm, struct kvm_sev_cmd *argp)
1380 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1381 struct sev_data_receive_start start;
1382 struct kvm_sev_receive_start params;
1383 int *error = &argp->error;
1388 if (!sev_guest(kvm))
1391 /* Get parameter from the userspace */
1392 if (copy_from_user(¶ms, (void __user *)(uintptr_t)argp->data,
1393 sizeof(struct kvm_sev_receive_start)))
1396 /* some sanity checks */
1397 if (!params.pdh_uaddr || !params.pdh_len ||
1398 !params.session_uaddr || !params.session_len)
1401 pdh_data = psp_copy_user_blob(params.pdh_uaddr, params.pdh_len);
1402 if (IS_ERR(pdh_data))
1403 return PTR_ERR(pdh_data);
1405 session_data = psp_copy_user_blob(params.session_uaddr,
1406 params.session_len);
1407 if (IS_ERR(session_data)) {
1408 ret = PTR_ERR(session_data);
1412 memset(&start, 0, sizeof(start));
1413 start.handle = params.handle;
1414 start.policy = params.policy;
1415 start.pdh_cert_address = __psp_pa(pdh_data);
1416 start.pdh_cert_len = params.pdh_len;
1417 start.session_address = __psp_pa(session_data);
1418 start.session_len = params.session_len;
1420 /* create memory encryption context */
1421 ret = __sev_issue_cmd(argp->sev_fd, SEV_CMD_RECEIVE_START, &start,
1424 goto e_free_session;
1426 /* Bind ASID to this guest */
1427 ret = sev_bind_asid(kvm, start.handle, error);
1429 sev_decommission(start.handle);
1430 goto e_free_session;
1433 params.handle = start.handle;
1434 if (copy_to_user((void __user *)(uintptr_t)argp->data,
1435 ¶ms, sizeof(struct kvm_sev_receive_start))) {
1437 sev_unbind_asid(kvm, start.handle);
1438 goto e_free_session;
1441 sev->handle = start.handle;
1442 sev->fd = argp->sev_fd;
1445 kfree(session_data);
1452 static int sev_receive_update_data(struct kvm *kvm, struct kvm_sev_cmd *argp)
1454 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1455 struct kvm_sev_receive_update_data params;
1456 struct sev_data_receive_update_data data;
1457 void *hdr = NULL, *trans = NULL;
1458 struct page **guest_page;
1462 if (!sev_guest(kvm))
1465 if (copy_from_user(¶ms, (void __user *)(uintptr_t)argp->data,
1466 sizeof(struct kvm_sev_receive_update_data)))
1469 if (!params.hdr_uaddr || !params.hdr_len ||
1470 !params.guest_uaddr || !params.guest_len ||
1471 !params.trans_uaddr || !params.trans_len)
1474 /* Check if we are crossing the page boundary */
1475 offset = params.guest_uaddr & (PAGE_SIZE - 1);
1476 if ((params.guest_len + offset > PAGE_SIZE))
1479 hdr = psp_copy_user_blob(params.hdr_uaddr, params.hdr_len);
1481 return PTR_ERR(hdr);
1483 trans = psp_copy_user_blob(params.trans_uaddr, params.trans_len);
1484 if (IS_ERR(trans)) {
1485 ret = PTR_ERR(trans);
1489 memset(&data, 0, sizeof(data));
1490 data.hdr_address = __psp_pa(hdr);
1491 data.hdr_len = params.hdr_len;
1492 data.trans_address = __psp_pa(trans);
1493 data.trans_len = params.trans_len;
1495 /* Pin guest memory */
1496 guest_page = sev_pin_memory(kvm, params.guest_uaddr & PAGE_MASK,
1498 if (IS_ERR(guest_page)) {
1499 ret = PTR_ERR(guest_page);
1504 * Flush (on non-coherent CPUs) before RECEIVE_UPDATE_DATA, the PSP
1505 * encrypts the written data with the guest's key, and the cache may
1506 * contain dirty, unencrypted data.
1508 sev_clflush_pages(guest_page, n);
1510 /* The RECEIVE_UPDATE_DATA command requires C-bit to be always set. */
1511 data.guest_address = (page_to_pfn(guest_page[0]) << PAGE_SHIFT) + offset;
1512 data.guest_address |= sev_me_mask;
1513 data.guest_len = params.guest_len;
1514 data.handle = sev->handle;
1516 ret = sev_issue_cmd(kvm, SEV_CMD_RECEIVE_UPDATE_DATA, &data,
1519 sev_unpin_memory(kvm, guest_page, n);
1529 static int sev_receive_finish(struct kvm *kvm, struct kvm_sev_cmd *argp)
1531 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1532 struct sev_data_receive_finish data;
1534 if (!sev_guest(kvm))
1537 data.handle = sev->handle;
1538 return sev_issue_cmd(kvm, SEV_CMD_RECEIVE_FINISH, &data, &argp->error);
1541 static bool is_cmd_allowed_from_mirror(u32 cmd_id)
1544 * Allow mirrors VM to call KVM_SEV_LAUNCH_UPDATE_VMSA to enable SEV-ES
1545 * active mirror VMs. Also allow the debugging and status commands.
1547 if (cmd_id == KVM_SEV_LAUNCH_UPDATE_VMSA ||
1548 cmd_id == KVM_SEV_GUEST_STATUS || cmd_id == KVM_SEV_DBG_DECRYPT ||
1549 cmd_id == KVM_SEV_DBG_ENCRYPT)
1555 static int sev_lock_two_vms(struct kvm *dst_kvm, struct kvm *src_kvm)
1557 struct kvm_sev_info *dst_sev = &to_kvm_svm(dst_kvm)->sev_info;
1558 struct kvm_sev_info *src_sev = &to_kvm_svm(src_kvm)->sev_info;
1561 if (dst_kvm == src_kvm)
1565 * Bail if these VMs are already involved in a migration to avoid
1566 * deadlock between two VMs trying to migrate to/from each other.
1568 if (atomic_cmpxchg_acquire(&dst_sev->migration_in_progress, 0, 1))
1571 if (atomic_cmpxchg_acquire(&src_sev->migration_in_progress, 0, 1))
1575 if (mutex_lock_killable(&dst_kvm->lock))
1577 if (mutex_lock_killable_nested(&src_kvm->lock, SINGLE_DEPTH_NESTING))
1582 mutex_unlock(&dst_kvm->lock);
1584 atomic_set_release(&src_sev->migration_in_progress, 0);
1586 atomic_set_release(&dst_sev->migration_in_progress, 0);
1590 static void sev_unlock_two_vms(struct kvm *dst_kvm, struct kvm *src_kvm)
1592 struct kvm_sev_info *dst_sev = &to_kvm_svm(dst_kvm)->sev_info;
1593 struct kvm_sev_info *src_sev = &to_kvm_svm(src_kvm)->sev_info;
1595 mutex_unlock(&dst_kvm->lock);
1596 mutex_unlock(&src_kvm->lock);
1597 atomic_set_release(&dst_sev->migration_in_progress, 0);
1598 atomic_set_release(&src_sev->migration_in_progress, 0);
1601 /* vCPU mutex subclasses. */
1602 enum sev_migration_role {
1603 SEV_MIGRATION_SOURCE = 0,
1604 SEV_MIGRATION_TARGET,
1605 SEV_NR_MIGRATION_ROLES,
1608 static int sev_lock_vcpus_for_migration(struct kvm *kvm,
1609 enum sev_migration_role role)
1611 struct kvm_vcpu *vcpu;
1614 kvm_for_each_vcpu(i, vcpu, kvm) {
1615 if (mutex_lock_killable_nested(&vcpu->mutex, role))
1618 #ifdef CONFIG_PROVE_LOCKING
1621 * Reset the role to one that avoids colliding with
1622 * the role used for the first vcpu mutex.
1624 role = SEV_NR_MIGRATION_ROLES;
1626 mutex_release(&vcpu->mutex.dep_map, _THIS_IP_);
1634 kvm_for_each_vcpu(j, vcpu, kvm) {
1638 #ifdef CONFIG_PROVE_LOCKING
1640 mutex_acquire(&vcpu->mutex.dep_map, role, 0, _THIS_IP_);
1643 mutex_unlock(&vcpu->mutex);
1648 static void sev_unlock_vcpus_for_migration(struct kvm *kvm)
1650 struct kvm_vcpu *vcpu;
1654 kvm_for_each_vcpu(i, vcpu, kvm) {
1658 mutex_acquire(&vcpu->mutex.dep_map,
1659 SEV_NR_MIGRATION_ROLES, 0, _THIS_IP_);
1661 mutex_unlock(&vcpu->mutex);
1665 static void sev_migrate_from(struct kvm *dst_kvm, struct kvm *src_kvm)
1667 struct kvm_sev_info *dst = &to_kvm_svm(dst_kvm)->sev_info;
1668 struct kvm_sev_info *src = &to_kvm_svm(src_kvm)->sev_info;
1669 struct kvm_vcpu *dst_vcpu, *src_vcpu;
1670 struct vcpu_svm *dst_svm, *src_svm;
1671 struct kvm_sev_info *mirror;
1675 dst->asid = src->asid;
1676 dst->handle = src->handle;
1677 dst->pages_locked = src->pages_locked;
1678 dst->enc_context_owner = src->enc_context_owner;
1679 dst->es_active = src->es_active;
1682 src->active = false;
1684 src->pages_locked = 0;
1685 src->enc_context_owner = NULL;
1686 src->es_active = false;
1688 list_cut_before(&dst->regions_list, &src->regions_list, &src->regions_list);
1691 * If this VM has mirrors, "transfer" each mirror's refcount of the
1692 * source to the destination (this KVM). The caller holds a reference
1693 * to the source, so there's no danger of use-after-free.
1695 list_cut_before(&dst->mirror_vms, &src->mirror_vms, &src->mirror_vms);
1696 list_for_each_entry(mirror, &dst->mirror_vms, mirror_entry) {
1697 kvm_get_kvm(dst_kvm);
1698 kvm_put_kvm(src_kvm);
1699 mirror->enc_context_owner = dst_kvm;
1703 * If this VM is a mirror, remove the old mirror from the owners list
1704 * and add the new mirror to the list.
1706 if (is_mirroring_enc_context(dst_kvm)) {
1707 struct kvm_sev_info *owner_sev_info =
1708 &to_kvm_svm(dst->enc_context_owner)->sev_info;
1710 list_del(&src->mirror_entry);
1711 list_add_tail(&dst->mirror_entry, &owner_sev_info->mirror_vms);
1714 kvm_for_each_vcpu(i, dst_vcpu, dst_kvm) {
1715 dst_svm = to_svm(dst_vcpu);
1717 sev_init_vmcb(dst_svm);
1719 if (!dst->es_active)
1723 * Note, the source is not required to have the same number of
1724 * vCPUs as the destination when migrating a vanilla SEV VM.
1726 src_vcpu = kvm_get_vcpu(dst_kvm, i);
1727 src_svm = to_svm(src_vcpu);
1730 * Transfer VMSA and GHCB state to the destination. Nullify and
1731 * clear source fields as appropriate, the state now belongs to
1734 memcpy(&dst_svm->sev_es, &src_svm->sev_es, sizeof(src_svm->sev_es));
1735 dst_svm->vmcb->control.ghcb_gpa = src_svm->vmcb->control.ghcb_gpa;
1736 dst_svm->vmcb->control.vmsa_pa = src_svm->vmcb->control.vmsa_pa;
1737 dst_vcpu->arch.guest_state_protected = true;
1739 memset(&src_svm->sev_es, 0, sizeof(src_svm->sev_es));
1740 src_svm->vmcb->control.ghcb_gpa = INVALID_PAGE;
1741 src_svm->vmcb->control.vmsa_pa = INVALID_PAGE;
1742 src_vcpu->arch.guest_state_protected = false;
1746 static int sev_check_source_vcpus(struct kvm *dst, struct kvm *src)
1748 struct kvm_vcpu *src_vcpu;
1751 if (!sev_es_guest(src))
1754 if (atomic_read(&src->online_vcpus) != atomic_read(&dst->online_vcpus))
1757 kvm_for_each_vcpu(i, src_vcpu, src) {
1758 if (!src_vcpu->arch.guest_state_protected)
1765 int sev_vm_move_enc_context_from(struct kvm *kvm, unsigned int source_fd)
1767 struct kvm_sev_info *dst_sev = &to_kvm_svm(kvm)->sev_info;
1768 struct kvm_sev_info *src_sev, *cg_cleanup_sev;
1769 struct file *source_kvm_file;
1770 struct kvm *source_kvm;
1771 bool charged = false;
1774 source_kvm_file = fget(source_fd);
1775 if (!file_is_kvm(source_kvm_file)) {
1780 source_kvm = source_kvm_file->private_data;
1781 ret = sev_lock_two_vms(kvm, source_kvm);
1785 if (sev_guest(kvm) || !sev_guest(source_kvm)) {
1790 src_sev = &to_kvm_svm(source_kvm)->sev_info;
1792 dst_sev->misc_cg = get_current_misc_cg();
1793 cg_cleanup_sev = dst_sev;
1794 if (dst_sev->misc_cg != src_sev->misc_cg) {
1795 ret = sev_misc_cg_try_charge(dst_sev);
1797 goto out_dst_cgroup;
1801 ret = sev_lock_vcpus_for_migration(kvm, SEV_MIGRATION_SOURCE);
1803 goto out_dst_cgroup;
1804 ret = sev_lock_vcpus_for_migration(source_kvm, SEV_MIGRATION_TARGET);
1808 ret = sev_check_source_vcpus(kvm, source_kvm);
1810 goto out_source_vcpu;
1812 sev_migrate_from(kvm, source_kvm);
1813 kvm_vm_dead(source_kvm);
1814 cg_cleanup_sev = src_sev;
1818 sev_unlock_vcpus_for_migration(source_kvm);
1820 sev_unlock_vcpus_for_migration(kvm);
1822 /* Operates on the source on success, on the destination on failure. */
1824 sev_misc_cg_uncharge(cg_cleanup_sev);
1825 put_misc_cg(cg_cleanup_sev->misc_cg);
1826 cg_cleanup_sev->misc_cg = NULL;
1828 sev_unlock_two_vms(kvm, source_kvm);
1830 if (source_kvm_file)
1831 fput(source_kvm_file);
1835 int sev_mem_enc_ioctl(struct kvm *kvm, void __user *argp)
1837 struct kvm_sev_cmd sev_cmd;
1846 if (copy_from_user(&sev_cmd, argp, sizeof(struct kvm_sev_cmd)))
1849 mutex_lock(&kvm->lock);
1851 /* Only the enc_context_owner handles some memory enc operations. */
1852 if (is_mirroring_enc_context(kvm) &&
1853 !is_cmd_allowed_from_mirror(sev_cmd.id)) {
1858 switch (sev_cmd.id) {
1859 case KVM_SEV_ES_INIT:
1860 if (!sev_es_enabled) {
1866 r = sev_guest_init(kvm, &sev_cmd);
1868 case KVM_SEV_LAUNCH_START:
1869 r = sev_launch_start(kvm, &sev_cmd);
1871 case KVM_SEV_LAUNCH_UPDATE_DATA:
1872 r = sev_launch_update_data(kvm, &sev_cmd);
1874 case KVM_SEV_LAUNCH_UPDATE_VMSA:
1875 r = sev_launch_update_vmsa(kvm, &sev_cmd);
1877 case KVM_SEV_LAUNCH_MEASURE:
1878 r = sev_launch_measure(kvm, &sev_cmd);
1880 case KVM_SEV_LAUNCH_FINISH:
1881 r = sev_launch_finish(kvm, &sev_cmd);
1883 case KVM_SEV_GUEST_STATUS:
1884 r = sev_guest_status(kvm, &sev_cmd);
1886 case KVM_SEV_DBG_DECRYPT:
1887 r = sev_dbg_crypt(kvm, &sev_cmd, true);
1889 case KVM_SEV_DBG_ENCRYPT:
1890 r = sev_dbg_crypt(kvm, &sev_cmd, false);
1892 case KVM_SEV_LAUNCH_SECRET:
1893 r = sev_launch_secret(kvm, &sev_cmd);
1895 case KVM_SEV_GET_ATTESTATION_REPORT:
1896 r = sev_get_attestation_report(kvm, &sev_cmd);
1898 case KVM_SEV_SEND_START:
1899 r = sev_send_start(kvm, &sev_cmd);
1901 case KVM_SEV_SEND_UPDATE_DATA:
1902 r = sev_send_update_data(kvm, &sev_cmd);
1904 case KVM_SEV_SEND_FINISH:
1905 r = sev_send_finish(kvm, &sev_cmd);
1907 case KVM_SEV_SEND_CANCEL:
1908 r = sev_send_cancel(kvm, &sev_cmd);
1910 case KVM_SEV_RECEIVE_START:
1911 r = sev_receive_start(kvm, &sev_cmd);
1913 case KVM_SEV_RECEIVE_UPDATE_DATA:
1914 r = sev_receive_update_data(kvm, &sev_cmd);
1916 case KVM_SEV_RECEIVE_FINISH:
1917 r = sev_receive_finish(kvm, &sev_cmd);
1924 if (copy_to_user(argp, &sev_cmd, sizeof(struct kvm_sev_cmd)))
1928 mutex_unlock(&kvm->lock);
1932 int sev_mem_enc_register_region(struct kvm *kvm,
1933 struct kvm_enc_region *range)
1935 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1936 struct enc_region *region;
1939 if (!sev_guest(kvm))
1942 /* If kvm is mirroring encryption context it isn't responsible for it */
1943 if (is_mirroring_enc_context(kvm))
1946 if (range->addr > ULONG_MAX || range->size > ULONG_MAX)
1949 region = kzalloc(sizeof(*region), GFP_KERNEL_ACCOUNT);
1953 mutex_lock(&kvm->lock);
1954 region->pages = sev_pin_memory(kvm, range->addr, range->size, ®ion->npages, 1);
1955 if (IS_ERR(region->pages)) {
1956 ret = PTR_ERR(region->pages);
1957 mutex_unlock(&kvm->lock);
1961 region->uaddr = range->addr;
1962 region->size = range->size;
1964 list_add_tail(®ion->list, &sev->regions_list);
1965 mutex_unlock(&kvm->lock);
1968 * The guest may change the memory encryption attribute from C=0 -> C=1
1969 * or vice versa for this memory range. Lets make sure caches are
1970 * flushed to ensure that guest data gets written into memory with
1973 sev_clflush_pages(region->pages, region->npages);
1982 static struct enc_region *
1983 find_enc_region(struct kvm *kvm, struct kvm_enc_region *range)
1985 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1986 struct list_head *head = &sev->regions_list;
1987 struct enc_region *i;
1989 list_for_each_entry(i, head, list) {
1990 if (i->uaddr == range->addr &&
1991 i->size == range->size)
1998 static void __unregister_enc_region_locked(struct kvm *kvm,
1999 struct enc_region *region)
2001 sev_unpin_memory(kvm, region->pages, region->npages);
2002 list_del(®ion->list);
2006 int sev_mem_enc_unregister_region(struct kvm *kvm,
2007 struct kvm_enc_region *range)
2009 struct enc_region *region;
2012 /* If kvm is mirroring encryption context it isn't responsible for it */
2013 if (is_mirroring_enc_context(kvm))
2016 mutex_lock(&kvm->lock);
2018 if (!sev_guest(kvm)) {
2023 region = find_enc_region(kvm, range);
2030 * Ensure that all guest tagged cache entries are flushed before
2031 * releasing the pages back to the system for use. CLFLUSH will
2032 * not do this, so issue a WBINVD.
2034 wbinvd_on_all_cpus();
2036 __unregister_enc_region_locked(kvm, region);
2038 mutex_unlock(&kvm->lock);
2042 mutex_unlock(&kvm->lock);
2046 int sev_vm_copy_enc_context_from(struct kvm *kvm, unsigned int source_fd)
2048 struct file *source_kvm_file;
2049 struct kvm *source_kvm;
2050 struct kvm_sev_info *source_sev, *mirror_sev;
2053 source_kvm_file = fget(source_fd);
2054 if (!file_is_kvm(source_kvm_file)) {
2059 source_kvm = source_kvm_file->private_data;
2060 ret = sev_lock_two_vms(kvm, source_kvm);
2065 * Mirrors of mirrors should work, but let's not get silly. Also
2066 * disallow out-of-band SEV/SEV-ES init if the target is already an
2067 * SEV guest, or if vCPUs have been created. KVM relies on vCPUs being
2068 * created after SEV/SEV-ES initialization, e.g. to init intercepts.
2070 if (sev_guest(kvm) || !sev_guest(source_kvm) ||
2071 is_mirroring_enc_context(source_kvm) || kvm->created_vcpus) {
2077 * The mirror kvm holds an enc_context_owner ref so its asid can't
2078 * disappear until we're done with it
2080 source_sev = &to_kvm_svm(source_kvm)->sev_info;
2081 kvm_get_kvm(source_kvm);
2082 mirror_sev = &to_kvm_svm(kvm)->sev_info;
2083 list_add_tail(&mirror_sev->mirror_entry, &source_sev->mirror_vms);
2085 /* Set enc_context_owner and copy its encryption context over */
2086 mirror_sev->enc_context_owner = source_kvm;
2087 mirror_sev->active = true;
2088 mirror_sev->asid = source_sev->asid;
2089 mirror_sev->fd = source_sev->fd;
2090 mirror_sev->es_active = source_sev->es_active;
2091 mirror_sev->handle = source_sev->handle;
2092 INIT_LIST_HEAD(&mirror_sev->regions_list);
2093 INIT_LIST_HEAD(&mirror_sev->mirror_vms);
2097 * Do not copy ap_jump_table. Since the mirror does not share the same
2098 * KVM contexts as the original, and they may have different
2103 sev_unlock_two_vms(kvm, source_kvm);
2105 if (source_kvm_file)
2106 fput(source_kvm_file);
2110 void sev_vm_destroy(struct kvm *kvm)
2112 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
2113 struct list_head *head = &sev->regions_list;
2114 struct list_head *pos, *q;
2116 if (!sev_guest(kvm))
2119 WARN_ON(!list_empty(&sev->mirror_vms));
2121 /* If this is a mirror_kvm release the enc_context_owner and skip sev cleanup */
2122 if (is_mirroring_enc_context(kvm)) {
2123 struct kvm *owner_kvm = sev->enc_context_owner;
2125 mutex_lock(&owner_kvm->lock);
2126 list_del(&sev->mirror_entry);
2127 mutex_unlock(&owner_kvm->lock);
2128 kvm_put_kvm(owner_kvm);
2133 * Ensure that all guest tagged cache entries are flushed before
2134 * releasing the pages back to the system for use. CLFLUSH will
2135 * not do this, so issue a WBINVD.
2137 wbinvd_on_all_cpus();
2140 * if userspace was terminated before unregistering the memory regions
2141 * then lets unpin all the registered memory.
2143 if (!list_empty(head)) {
2144 list_for_each_safe(pos, q, head) {
2145 __unregister_enc_region_locked(kvm,
2146 list_entry(pos, struct enc_region, list));
2151 sev_unbind_asid(kvm, sev->handle);
2155 void __init sev_set_cpu_caps(void)
2158 kvm_cpu_cap_clear(X86_FEATURE_SEV);
2159 if (!sev_es_enabled)
2160 kvm_cpu_cap_clear(X86_FEATURE_SEV_ES);
2163 void __init sev_hardware_setup(void)
2165 #ifdef CONFIG_KVM_AMD_SEV
2166 unsigned int eax, ebx, ecx, edx, sev_asid_count, sev_es_asid_count;
2167 bool sev_es_supported = false;
2168 bool sev_supported = false;
2170 if (!sev_enabled || !npt_enabled)
2174 * SEV must obviously be supported in hardware. Sanity check that the
2175 * CPU supports decode assists, which is mandatory for SEV guests to
2176 * support instruction emulation.
2178 if (!boot_cpu_has(X86_FEATURE_SEV) ||
2179 WARN_ON_ONCE(!boot_cpu_has(X86_FEATURE_DECODEASSISTS)))
2182 /* Retrieve SEV CPUID information */
2183 cpuid(0x8000001f, &eax, &ebx, &ecx, &edx);
2185 /* Set encryption bit location for SEV-ES guests */
2186 sev_enc_bit = ebx & 0x3f;
2188 /* Maximum number of encrypted guests supported simultaneously */
2193 /* Minimum ASID value that should be used for SEV guest */
2195 sev_me_mask = 1UL << (ebx & 0x3f);
2198 * Initialize SEV ASID bitmaps. Allocate space for ASID 0 in the bitmap,
2199 * even though it's never used, so that the bitmap is indexed by the
2202 nr_asids = max_sev_asid + 1;
2203 sev_asid_bitmap = bitmap_zalloc(nr_asids, GFP_KERNEL);
2204 if (!sev_asid_bitmap)
2207 sev_reclaim_asid_bitmap = bitmap_zalloc(nr_asids, GFP_KERNEL);
2208 if (!sev_reclaim_asid_bitmap) {
2209 bitmap_free(sev_asid_bitmap);
2210 sev_asid_bitmap = NULL;
2214 sev_asid_count = max_sev_asid - min_sev_asid + 1;
2215 if (misc_cg_set_capacity(MISC_CG_RES_SEV, sev_asid_count))
2218 pr_info("SEV supported: %u ASIDs\n", sev_asid_count);
2219 sev_supported = true;
2221 /* SEV-ES support requested? */
2222 if (!sev_es_enabled)
2226 * SEV-ES requires MMIO caching as KVM doesn't have access to the guest
2227 * instruction stream, i.e. can't emulate in response to a #NPF and
2228 * instead relies on #NPF(RSVD) being reflected into the guest as #VC
2229 * (the guest can then do a #VMGEXIT to request MMIO emulation).
2231 if (!enable_mmio_caching)
2234 /* Does the CPU support SEV-ES? */
2235 if (!boot_cpu_has(X86_FEATURE_SEV_ES))
2238 /* Has the system been allocated ASIDs for SEV-ES? */
2239 if (min_sev_asid == 1)
2242 sev_es_asid_count = min_sev_asid - 1;
2243 if (misc_cg_set_capacity(MISC_CG_RES_SEV_ES, sev_es_asid_count))
2246 pr_info("SEV-ES supported: %u ASIDs\n", sev_es_asid_count);
2247 sev_es_supported = true;
2250 sev_enabled = sev_supported;
2251 sev_es_enabled = sev_es_supported;
2255 void sev_hardware_unsetup(void)
2260 /* No need to take sev_bitmap_lock, all VMs have been destroyed. */
2261 sev_flush_asids(1, max_sev_asid);
2263 bitmap_free(sev_asid_bitmap);
2264 bitmap_free(sev_reclaim_asid_bitmap);
2266 misc_cg_set_capacity(MISC_CG_RES_SEV, 0);
2267 misc_cg_set_capacity(MISC_CG_RES_SEV_ES, 0);
2270 int sev_cpu_init(struct svm_cpu_data *sd)
2275 sd->sev_vmcbs = kcalloc(nr_asids, sizeof(void *), GFP_KERNEL);
2283 * Pages used by hardware to hold guest encrypted state must be flushed before
2284 * returning them to the system.
2286 static void sev_flush_encrypted_page(struct kvm_vcpu *vcpu, void *va)
2288 int asid = to_kvm_svm(vcpu->kvm)->sev_info.asid;
2291 * Note! The address must be a kernel address, as regular page walk
2292 * checks are performed by VM_PAGE_FLUSH, i.e. operating on a user
2293 * address is non-deterministic and unsafe. This function deliberately
2294 * takes a pointer to deter passing in a user address.
2296 unsigned long addr = (unsigned long)va;
2299 * If CPU enforced cache coherency for encrypted mappings of the
2300 * same physical page is supported, use CLFLUSHOPT instead. NOTE: cache
2301 * flush is still needed in order to work properly with DMA devices.
2303 if (boot_cpu_has(X86_FEATURE_SME_COHERENT)) {
2304 clflush_cache_range(va, PAGE_SIZE);
2309 * VM Page Flush takes a host virtual address and a guest ASID. Fall
2310 * back to WBINVD if this faults so as not to make any problems worse
2311 * by leaving stale encrypted data in the cache.
2313 if (WARN_ON_ONCE(wrmsrl_safe(MSR_AMD64_VM_PAGE_FLUSH, addr | asid)))
2319 wbinvd_on_all_cpus();
2322 void sev_guest_memory_reclaimed(struct kvm *kvm)
2324 if (!sev_guest(kvm))
2327 wbinvd_on_all_cpus();
2330 void sev_free_vcpu(struct kvm_vcpu *vcpu)
2332 struct vcpu_svm *svm;
2334 if (!sev_es_guest(vcpu->kvm))
2339 if (vcpu->arch.guest_state_protected)
2340 sev_flush_encrypted_page(vcpu, svm->sev_es.vmsa);
2342 __free_page(virt_to_page(svm->sev_es.vmsa));
2344 if (svm->sev_es.ghcb_sa_free)
2345 kvfree(svm->sev_es.ghcb_sa);
2348 static void dump_ghcb(struct vcpu_svm *svm)
2350 struct ghcb *ghcb = svm->sev_es.ghcb;
2353 /* Re-use the dump_invalid_vmcb module parameter */
2354 if (!dump_invalid_vmcb) {
2355 pr_warn_ratelimited("set kvm_amd.dump_invalid_vmcb=1 to dump internal KVM state.\n");
2359 nbits = sizeof(ghcb->save.valid_bitmap) * 8;
2361 pr_err("GHCB (GPA=%016llx):\n", svm->vmcb->control.ghcb_gpa);
2362 pr_err("%-20s%016llx is_valid: %u\n", "sw_exit_code",
2363 ghcb->save.sw_exit_code, ghcb_sw_exit_code_is_valid(ghcb));
2364 pr_err("%-20s%016llx is_valid: %u\n", "sw_exit_info_1",
2365 ghcb->save.sw_exit_info_1, ghcb_sw_exit_info_1_is_valid(ghcb));
2366 pr_err("%-20s%016llx is_valid: %u\n", "sw_exit_info_2",
2367 ghcb->save.sw_exit_info_2, ghcb_sw_exit_info_2_is_valid(ghcb));
2368 pr_err("%-20s%016llx is_valid: %u\n", "sw_scratch",
2369 ghcb->save.sw_scratch, ghcb_sw_scratch_is_valid(ghcb));
2370 pr_err("%-20s%*pb\n", "valid_bitmap", nbits, ghcb->save.valid_bitmap);
2373 static void sev_es_sync_to_ghcb(struct vcpu_svm *svm)
2375 struct kvm_vcpu *vcpu = &svm->vcpu;
2376 struct ghcb *ghcb = svm->sev_es.ghcb;
2379 * The GHCB protocol so far allows for the following data
2381 * GPRs RAX, RBX, RCX, RDX
2383 * Copy their values, even if they may not have been written during the
2384 * VM-Exit. It's the guest's responsibility to not consume random data.
2386 ghcb_set_rax(ghcb, vcpu->arch.regs[VCPU_REGS_RAX]);
2387 ghcb_set_rbx(ghcb, vcpu->arch.regs[VCPU_REGS_RBX]);
2388 ghcb_set_rcx(ghcb, vcpu->arch.regs[VCPU_REGS_RCX]);
2389 ghcb_set_rdx(ghcb, vcpu->arch.regs[VCPU_REGS_RDX]);
2392 static void sev_es_sync_from_ghcb(struct vcpu_svm *svm)
2394 struct vmcb_control_area *control = &svm->vmcb->control;
2395 struct kvm_vcpu *vcpu = &svm->vcpu;
2396 struct ghcb *ghcb = svm->sev_es.ghcb;
2400 * The GHCB protocol so far allows for the following data
2402 * GPRs RAX, RBX, RCX, RDX
2406 * VMMCALL allows the guest to provide extra registers. KVM also
2407 * expects RSI for hypercalls, so include that, too.
2409 * Copy their values to the appropriate location if supplied.
2411 memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
2413 vcpu->arch.regs[VCPU_REGS_RAX] = ghcb_get_rax_if_valid(ghcb);
2414 vcpu->arch.regs[VCPU_REGS_RBX] = ghcb_get_rbx_if_valid(ghcb);
2415 vcpu->arch.regs[VCPU_REGS_RCX] = ghcb_get_rcx_if_valid(ghcb);
2416 vcpu->arch.regs[VCPU_REGS_RDX] = ghcb_get_rdx_if_valid(ghcb);
2417 vcpu->arch.regs[VCPU_REGS_RSI] = ghcb_get_rsi_if_valid(ghcb);
2419 svm->vmcb->save.cpl = ghcb_get_cpl_if_valid(ghcb);
2421 if (ghcb_xcr0_is_valid(ghcb)) {
2422 vcpu->arch.xcr0 = ghcb_get_xcr0(ghcb);
2423 kvm_update_cpuid_runtime(vcpu);
2426 /* Copy the GHCB exit information into the VMCB fields */
2427 exit_code = ghcb_get_sw_exit_code(ghcb);
2428 control->exit_code = lower_32_bits(exit_code);
2429 control->exit_code_hi = upper_32_bits(exit_code);
2430 control->exit_info_1 = ghcb_get_sw_exit_info_1(ghcb);
2431 control->exit_info_2 = ghcb_get_sw_exit_info_2(ghcb);
2433 /* Clear the valid entries fields */
2434 memset(ghcb->save.valid_bitmap, 0, sizeof(ghcb->save.valid_bitmap));
2437 static int sev_es_validate_vmgexit(struct vcpu_svm *svm)
2439 struct kvm_vcpu *vcpu;
2444 ghcb = svm->sev_es.ghcb;
2447 * Retrieve the exit code now even though it may not be marked valid
2448 * as it could help with debugging.
2450 exit_code = ghcb_get_sw_exit_code(ghcb);
2452 /* Only GHCB Usage code 0 is supported */
2453 if (ghcb->ghcb_usage) {
2454 reason = GHCB_ERR_INVALID_USAGE;
2458 reason = GHCB_ERR_MISSING_INPUT;
2460 if (!ghcb_sw_exit_code_is_valid(ghcb) ||
2461 !ghcb_sw_exit_info_1_is_valid(ghcb) ||
2462 !ghcb_sw_exit_info_2_is_valid(ghcb))
2465 switch (ghcb_get_sw_exit_code(ghcb)) {
2466 case SVM_EXIT_READ_DR7:
2468 case SVM_EXIT_WRITE_DR7:
2469 if (!ghcb_rax_is_valid(ghcb))
2472 case SVM_EXIT_RDTSC:
2474 case SVM_EXIT_RDPMC:
2475 if (!ghcb_rcx_is_valid(ghcb))
2478 case SVM_EXIT_CPUID:
2479 if (!ghcb_rax_is_valid(ghcb) ||
2480 !ghcb_rcx_is_valid(ghcb))
2482 if (ghcb_get_rax(ghcb) == 0xd)
2483 if (!ghcb_xcr0_is_valid(ghcb))
2489 if (ghcb_get_sw_exit_info_1(ghcb) & SVM_IOIO_STR_MASK) {
2490 if (!ghcb_sw_scratch_is_valid(ghcb))
2493 if (!(ghcb_get_sw_exit_info_1(ghcb) & SVM_IOIO_TYPE_MASK))
2494 if (!ghcb_rax_is_valid(ghcb))
2499 if (!ghcb_rcx_is_valid(ghcb))
2501 if (ghcb_get_sw_exit_info_1(ghcb)) {
2502 if (!ghcb_rax_is_valid(ghcb) ||
2503 !ghcb_rdx_is_valid(ghcb))
2507 case SVM_EXIT_VMMCALL:
2508 if (!ghcb_rax_is_valid(ghcb) ||
2509 !ghcb_cpl_is_valid(ghcb))
2512 case SVM_EXIT_RDTSCP:
2514 case SVM_EXIT_WBINVD:
2516 case SVM_EXIT_MONITOR:
2517 if (!ghcb_rax_is_valid(ghcb) ||
2518 !ghcb_rcx_is_valid(ghcb) ||
2519 !ghcb_rdx_is_valid(ghcb))
2522 case SVM_EXIT_MWAIT:
2523 if (!ghcb_rax_is_valid(ghcb) ||
2524 !ghcb_rcx_is_valid(ghcb))
2527 case SVM_VMGEXIT_MMIO_READ:
2528 case SVM_VMGEXIT_MMIO_WRITE:
2529 if (!ghcb_sw_scratch_is_valid(ghcb))
2532 case SVM_VMGEXIT_NMI_COMPLETE:
2533 case SVM_VMGEXIT_AP_HLT_LOOP:
2534 case SVM_VMGEXIT_AP_JUMP_TABLE:
2535 case SVM_VMGEXIT_UNSUPPORTED_EVENT:
2538 reason = GHCB_ERR_INVALID_EVENT;
2547 if (reason == GHCB_ERR_INVALID_USAGE) {
2548 vcpu_unimpl(vcpu, "vmgexit: ghcb usage %#x is not valid\n",
2550 } else if (reason == GHCB_ERR_INVALID_EVENT) {
2551 vcpu_unimpl(vcpu, "vmgexit: exit code %#llx is not valid\n",
2554 vcpu_unimpl(vcpu, "vmgexit: exit code %#llx input is not valid\n",
2559 /* Clear the valid entries fields */
2560 memset(ghcb->save.valid_bitmap, 0, sizeof(ghcb->save.valid_bitmap));
2562 ghcb_set_sw_exit_info_1(ghcb, 2);
2563 ghcb_set_sw_exit_info_2(ghcb, reason);
2565 /* Resume the guest to "return" the error code. */
2569 void sev_es_unmap_ghcb(struct vcpu_svm *svm)
2571 if (!svm->sev_es.ghcb)
2574 if (svm->sev_es.ghcb_sa_free) {
2576 * The scratch area lives outside the GHCB, so there is a
2577 * buffer that, depending on the operation performed, may
2578 * need to be synced, then freed.
2580 if (svm->sev_es.ghcb_sa_sync) {
2581 kvm_write_guest(svm->vcpu.kvm,
2582 ghcb_get_sw_scratch(svm->sev_es.ghcb),
2583 svm->sev_es.ghcb_sa,
2584 svm->sev_es.ghcb_sa_len);
2585 svm->sev_es.ghcb_sa_sync = false;
2588 kvfree(svm->sev_es.ghcb_sa);
2589 svm->sev_es.ghcb_sa = NULL;
2590 svm->sev_es.ghcb_sa_free = false;
2593 trace_kvm_vmgexit_exit(svm->vcpu.vcpu_id, svm->sev_es.ghcb);
2595 sev_es_sync_to_ghcb(svm);
2597 kvm_vcpu_unmap(&svm->vcpu, &svm->sev_es.ghcb_map, true);
2598 svm->sev_es.ghcb = NULL;
2601 void pre_sev_run(struct vcpu_svm *svm, int cpu)
2603 struct svm_cpu_data *sd = per_cpu_ptr(&svm_data, cpu);
2604 int asid = sev_get_asid(svm->vcpu.kvm);
2606 /* Assign the asid allocated with this SEV guest */
2612 * 1) when different VMCB for the same ASID is to be run on the same host CPU.
2613 * 2) or this VMCB was executed on different host CPU in previous VMRUNs.
2615 if (sd->sev_vmcbs[asid] == svm->vmcb &&
2616 svm->vcpu.arch.last_vmentry_cpu == cpu)
2619 sd->sev_vmcbs[asid] = svm->vmcb;
2620 svm->vmcb->control.tlb_ctl = TLB_CONTROL_FLUSH_ASID;
2621 vmcb_mark_dirty(svm->vmcb, VMCB_ASID);
2624 #define GHCB_SCRATCH_AREA_LIMIT (16ULL * PAGE_SIZE)
2625 static int setup_vmgexit_scratch(struct vcpu_svm *svm, bool sync, u64 len)
2627 struct vmcb_control_area *control = &svm->vmcb->control;
2628 struct ghcb *ghcb = svm->sev_es.ghcb;
2629 u64 ghcb_scratch_beg, ghcb_scratch_end;
2630 u64 scratch_gpa_beg, scratch_gpa_end;
2633 scratch_gpa_beg = ghcb_get_sw_scratch(ghcb);
2634 if (!scratch_gpa_beg) {
2635 pr_err("vmgexit: scratch gpa not provided\n");
2639 scratch_gpa_end = scratch_gpa_beg + len;
2640 if (scratch_gpa_end < scratch_gpa_beg) {
2641 pr_err("vmgexit: scratch length (%#llx) not valid for scratch address (%#llx)\n",
2642 len, scratch_gpa_beg);
2646 if ((scratch_gpa_beg & PAGE_MASK) == control->ghcb_gpa) {
2647 /* Scratch area begins within GHCB */
2648 ghcb_scratch_beg = control->ghcb_gpa +
2649 offsetof(struct ghcb, shared_buffer);
2650 ghcb_scratch_end = control->ghcb_gpa +
2651 offsetof(struct ghcb, reserved_0xff0);
2654 * If the scratch area begins within the GHCB, it must be
2655 * completely contained in the GHCB shared buffer area.
2657 if (scratch_gpa_beg < ghcb_scratch_beg ||
2658 scratch_gpa_end > ghcb_scratch_end) {
2659 pr_err("vmgexit: scratch area is outside of GHCB shared buffer area (%#llx - %#llx)\n",
2660 scratch_gpa_beg, scratch_gpa_end);
2664 scratch_va = (void *)svm->sev_es.ghcb;
2665 scratch_va += (scratch_gpa_beg - control->ghcb_gpa);
2668 * The guest memory must be read into a kernel buffer, so
2671 if (len > GHCB_SCRATCH_AREA_LIMIT) {
2672 pr_err("vmgexit: scratch area exceeds KVM limits (%#llx requested, %#llx limit)\n",
2673 len, GHCB_SCRATCH_AREA_LIMIT);
2676 scratch_va = kvzalloc(len, GFP_KERNEL_ACCOUNT);
2680 if (kvm_read_guest(svm->vcpu.kvm, scratch_gpa_beg, scratch_va, len)) {
2681 /* Unable to copy scratch area from guest */
2682 pr_err("vmgexit: kvm_read_guest for scratch area failed\n");
2689 * The scratch area is outside the GHCB. The operation will
2690 * dictate whether the buffer needs to be synced before running
2691 * the vCPU next time (i.e. a read was requested so the data
2692 * must be written back to the guest memory).
2694 svm->sev_es.ghcb_sa_sync = sync;
2695 svm->sev_es.ghcb_sa_free = true;
2698 svm->sev_es.ghcb_sa = scratch_va;
2699 svm->sev_es.ghcb_sa_len = len;
2704 ghcb_set_sw_exit_info_1(ghcb, 2);
2705 ghcb_set_sw_exit_info_2(ghcb, GHCB_ERR_INVALID_SCRATCH_AREA);
2710 static void set_ghcb_msr_bits(struct vcpu_svm *svm, u64 value, u64 mask,
2713 svm->vmcb->control.ghcb_gpa &= ~(mask << pos);
2714 svm->vmcb->control.ghcb_gpa |= (value & mask) << pos;
2717 static u64 get_ghcb_msr_bits(struct vcpu_svm *svm, u64 mask, unsigned int pos)
2719 return (svm->vmcb->control.ghcb_gpa >> pos) & mask;
2722 static void set_ghcb_msr(struct vcpu_svm *svm, u64 value)
2724 svm->vmcb->control.ghcb_gpa = value;
2727 static int sev_handle_vmgexit_msr_protocol(struct vcpu_svm *svm)
2729 struct vmcb_control_area *control = &svm->vmcb->control;
2730 struct kvm_vcpu *vcpu = &svm->vcpu;
2734 ghcb_info = control->ghcb_gpa & GHCB_MSR_INFO_MASK;
2736 trace_kvm_vmgexit_msr_protocol_enter(svm->vcpu.vcpu_id,
2739 switch (ghcb_info) {
2740 case GHCB_MSR_SEV_INFO_REQ:
2741 set_ghcb_msr(svm, GHCB_MSR_SEV_INFO(GHCB_VERSION_MAX,
2745 case GHCB_MSR_CPUID_REQ: {
2746 u64 cpuid_fn, cpuid_reg, cpuid_value;
2748 cpuid_fn = get_ghcb_msr_bits(svm,
2749 GHCB_MSR_CPUID_FUNC_MASK,
2750 GHCB_MSR_CPUID_FUNC_POS);
2752 /* Initialize the registers needed by the CPUID intercept */
2753 vcpu->arch.regs[VCPU_REGS_RAX] = cpuid_fn;
2754 vcpu->arch.regs[VCPU_REGS_RCX] = 0;
2756 ret = svm_invoke_exit_handler(vcpu, SVM_EXIT_CPUID);
2758 /* Error, keep GHCB MSR value as-is */
2762 cpuid_reg = get_ghcb_msr_bits(svm,
2763 GHCB_MSR_CPUID_REG_MASK,
2764 GHCB_MSR_CPUID_REG_POS);
2766 cpuid_value = vcpu->arch.regs[VCPU_REGS_RAX];
2767 else if (cpuid_reg == 1)
2768 cpuid_value = vcpu->arch.regs[VCPU_REGS_RBX];
2769 else if (cpuid_reg == 2)
2770 cpuid_value = vcpu->arch.regs[VCPU_REGS_RCX];
2772 cpuid_value = vcpu->arch.regs[VCPU_REGS_RDX];
2774 set_ghcb_msr_bits(svm, cpuid_value,
2775 GHCB_MSR_CPUID_VALUE_MASK,
2776 GHCB_MSR_CPUID_VALUE_POS);
2778 set_ghcb_msr_bits(svm, GHCB_MSR_CPUID_RESP,
2783 case GHCB_MSR_TERM_REQ: {
2784 u64 reason_set, reason_code;
2786 reason_set = get_ghcb_msr_bits(svm,
2787 GHCB_MSR_TERM_REASON_SET_MASK,
2788 GHCB_MSR_TERM_REASON_SET_POS);
2789 reason_code = get_ghcb_msr_bits(svm,
2790 GHCB_MSR_TERM_REASON_MASK,
2791 GHCB_MSR_TERM_REASON_POS);
2792 pr_info("SEV-ES guest requested termination: %#llx:%#llx\n",
2793 reason_set, reason_code);
2795 vcpu->run->exit_reason = KVM_EXIT_SYSTEM_EVENT;
2796 vcpu->run->system_event.type = KVM_SYSTEM_EVENT_SEV_TERM;
2797 vcpu->run->system_event.ndata = 1;
2798 vcpu->run->system_event.data[0] = control->ghcb_gpa;
2803 /* Error, keep GHCB MSR value as-is */
2807 trace_kvm_vmgexit_msr_protocol_exit(svm->vcpu.vcpu_id,
2808 control->ghcb_gpa, ret);
2813 int sev_handle_vmgexit(struct kvm_vcpu *vcpu)
2815 struct vcpu_svm *svm = to_svm(vcpu);
2816 struct vmcb_control_area *control = &svm->vmcb->control;
2817 u64 ghcb_gpa, exit_code;
2821 /* Validate the GHCB */
2822 ghcb_gpa = control->ghcb_gpa;
2823 if (ghcb_gpa & GHCB_MSR_INFO_MASK)
2824 return sev_handle_vmgexit_msr_protocol(svm);
2827 vcpu_unimpl(vcpu, "vmgexit: GHCB gpa is not set\n");
2829 /* Without a GHCB, just return right back to the guest */
2833 if (kvm_vcpu_map(vcpu, ghcb_gpa >> PAGE_SHIFT, &svm->sev_es.ghcb_map)) {
2834 /* Unable to map GHCB from guest */
2835 vcpu_unimpl(vcpu, "vmgexit: error mapping GHCB [%#llx] from guest\n",
2838 /* Without a GHCB, just return right back to the guest */
2842 svm->sev_es.ghcb = svm->sev_es.ghcb_map.hva;
2843 ghcb = svm->sev_es.ghcb_map.hva;
2845 trace_kvm_vmgexit_enter(vcpu->vcpu_id, ghcb);
2847 exit_code = ghcb_get_sw_exit_code(ghcb);
2849 ret = sev_es_validate_vmgexit(svm);
2853 sev_es_sync_from_ghcb(svm);
2854 ghcb_set_sw_exit_info_1(ghcb, 0);
2855 ghcb_set_sw_exit_info_2(ghcb, 0);
2857 switch (exit_code) {
2858 case SVM_VMGEXIT_MMIO_READ:
2859 ret = setup_vmgexit_scratch(svm, true, control->exit_info_2);
2863 ret = kvm_sev_es_mmio_read(vcpu,
2864 control->exit_info_1,
2865 control->exit_info_2,
2866 svm->sev_es.ghcb_sa);
2868 case SVM_VMGEXIT_MMIO_WRITE:
2869 ret = setup_vmgexit_scratch(svm, false, control->exit_info_2);
2873 ret = kvm_sev_es_mmio_write(vcpu,
2874 control->exit_info_1,
2875 control->exit_info_2,
2876 svm->sev_es.ghcb_sa);
2878 case SVM_VMGEXIT_NMI_COMPLETE:
2879 ret = svm_invoke_exit_handler(vcpu, SVM_EXIT_IRET);
2881 case SVM_VMGEXIT_AP_HLT_LOOP:
2882 ret = kvm_emulate_ap_reset_hold(vcpu);
2884 case SVM_VMGEXIT_AP_JUMP_TABLE: {
2885 struct kvm_sev_info *sev = &to_kvm_svm(vcpu->kvm)->sev_info;
2887 switch (control->exit_info_1) {
2889 /* Set AP jump table address */
2890 sev->ap_jump_table = control->exit_info_2;
2893 /* Get AP jump table address */
2894 ghcb_set_sw_exit_info_2(ghcb, sev->ap_jump_table);
2897 pr_err("svm: vmgexit: unsupported AP jump table request - exit_info_1=%#llx\n",
2898 control->exit_info_1);
2899 ghcb_set_sw_exit_info_1(ghcb, 2);
2900 ghcb_set_sw_exit_info_2(ghcb, GHCB_ERR_INVALID_INPUT);
2906 case SVM_VMGEXIT_UNSUPPORTED_EVENT:
2908 "vmgexit: unsupported event - exit_info_1=%#llx, exit_info_2=%#llx\n",
2909 control->exit_info_1, control->exit_info_2);
2913 ret = svm_invoke_exit_handler(vcpu, exit_code);
2919 int sev_es_string_io(struct vcpu_svm *svm, int size, unsigned int port, int in)
2925 if (svm->vmcb->control.exit_info_2 > INT_MAX)
2928 count = svm->vmcb->control.exit_info_2;
2929 if (unlikely(check_mul_overflow(count, size, &bytes)))
2932 r = setup_vmgexit_scratch(svm, in, bytes);
2936 return kvm_sev_es_string_io(&svm->vcpu, size, port, svm->sev_es.ghcb_sa,
2940 static void sev_es_init_vmcb(struct vcpu_svm *svm)
2942 struct kvm_vcpu *vcpu = &svm->vcpu;
2944 svm->vmcb->control.nested_ctl |= SVM_NESTED_CTL_SEV_ES_ENABLE;
2945 svm->vmcb->control.virt_ext |= LBR_CTL_ENABLE_MASK;
2948 * An SEV-ES guest requires a VMSA area that is a separate from the
2949 * VMCB page. Do not include the encryption mask on the VMSA physical
2950 * address since hardware will access it using the guest key.
2952 svm->vmcb->control.vmsa_pa = __pa(svm->sev_es.vmsa);
2954 /* Can't intercept CR register access, HV can't modify CR registers */
2955 svm_clr_intercept(svm, INTERCEPT_CR0_READ);
2956 svm_clr_intercept(svm, INTERCEPT_CR4_READ);
2957 svm_clr_intercept(svm, INTERCEPT_CR8_READ);
2958 svm_clr_intercept(svm, INTERCEPT_CR0_WRITE);
2959 svm_clr_intercept(svm, INTERCEPT_CR4_WRITE);
2960 svm_clr_intercept(svm, INTERCEPT_CR8_WRITE);
2962 svm_clr_intercept(svm, INTERCEPT_SELECTIVE_CR0);
2964 /* Track EFER/CR register changes */
2965 svm_set_intercept(svm, TRAP_EFER_WRITE);
2966 svm_set_intercept(svm, TRAP_CR0_WRITE);
2967 svm_set_intercept(svm, TRAP_CR4_WRITE);
2968 svm_set_intercept(svm, TRAP_CR8_WRITE);
2970 /* No support for enable_vmware_backdoor */
2971 clr_exception_intercept(svm, GP_VECTOR);
2973 /* Can't intercept XSETBV, HV can't modify XCR0 directly */
2974 svm_clr_intercept(svm, INTERCEPT_XSETBV);
2976 /* Clear intercepts on selected MSRs */
2977 set_msr_interception(vcpu, svm->msrpm, MSR_EFER, 1, 1);
2978 set_msr_interception(vcpu, svm->msrpm, MSR_IA32_CR_PAT, 1, 1);
2979 set_msr_interception(vcpu, svm->msrpm, MSR_IA32_LASTBRANCHFROMIP, 1, 1);
2980 set_msr_interception(vcpu, svm->msrpm, MSR_IA32_LASTBRANCHTOIP, 1, 1);
2981 set_msr_interception(vcpu, svm->msrpm, MSR_IA32_LASTINTFROMIP, 1, 1);
2982 set_msr_interception(vcpu, svm->msrpm, MSR_IA32_LASTINTTOIP, 1, 1);
2984 if (boot_cpu_has(X86_FEATURE_V_TSC_AUX) &&
2985 (guest_cpuid_has(&svm->vcpu, X86_FEATURE_RDTSCP) ||
2986 guest_cpuid_has(&svm->vcpu, X86_FEATURE_RDPID))) {
2987 set_msr_interception(vcpu, svm->msrpm, MSR_TSC_AUX, 1, 1);
2988 if (guest_cpuid_has(&svm->vcpu, X86_FEATURE_RDTSCP))
2989 svm_clr_intercept(svm, INTERCEPT_RDTSCP);
2993 void sev_init_vmcb(struct vcpu_svm *svm)
2995 svm->vmcb->control.nested_ctl |= SVM_NESTED_CTL_SEV_ENABLE;
2996 clr_exception_intercept(svm, UD_VECTOR);
2998 if (sev_es_guest(svm->vcpu.kvm))
2999 sev_es_init_vmcb(svm);
3002 void sev_es_vcpu_reset(struct vcpu_svm *svm)
3005 * Set the GHCB MSR value as per the GHCB specification when emulating
3006 * vCPU RESET for an SEV-ES guest.
3008 set_ghcb_msr(svm, GHCB_MSR_SEV_INFO(GHCB_VERSION_MAX,
3013 void sev_es_prepare_switch_to_guest(struct sev_es_save_area *hostsa)
3016 * As an SEV-ES guest, hardware will restore the host state on VMEXIT,
3017 * of which one step is to perform a VMLOAD. KVM performs the
3018 * corresponding VMSAVE in svm_prepare_guest_switch for both
3019 * traditional and SEV-ES guests.
3022 /* XCR0 is restored on VMEXIT, save the current host value */
3023 hostsa->xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
3025 /* PKRU is restored on VMEXIT, save the current host value */
3026 hostsa->pkru = read_pkru();
3028 /* MSR_IA32_XSS is restored on VMEXIT, save the currnet host value */
3029 hostsa->xss = host_xss;
3032 void sev_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
3034 struct vcpu_svm *svm = to_svm(vcpu);
3036 /* First SIPI: Use the values as initially set by the VMM */
3037 if (!svm->sev_es.received_first_sipi) {
3038 svm->sev_es.received_first_sipi = true;
3043 * Subsequent SIPI: Return from an AP Reset Hold VMGEXIT, where
3044 * the guest will set the CS and RIP. Set SW_EXIT_INFO_2 to a
3047 if (!svm->sev_es.ghcb)
3050 ghcb_set_sw_exit_info_2(svm->sev_es.ghcb, 1);