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);
268 sev->es_active = false;
273 static int sev_bind_asid(struct kvm *kvm, unsigned int handle, int *error)
275 struct sev_data_activate activate;
276 int asid = sev_get_asid(kvm);
279 /* activate ASID on the given handle */
280 activate.handle = handle;
281 activate.asid = asid;
282 ret = sev_guest_activate(&activate, error);
287 static int __sev_issue_cmd(int fd, int id, void *data, int *error)
296 ret = sev_issue_cmd_external_user(f.file, id, data, error);
302 static int sev_issue_cmd(struct kvm *kvm, int id, void *data, int *error)
304 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
306 return __sev_issue_cmd(sev->fd, id, data, error);
309 static int sev_launch_start(struct kvm *kvm, struct kvm_sev_cmd *argp)
311 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
312 struct sev_data_launch_start start;
313 struct kvm_sev_launch_start params;
314 void *dh_blob, *session_blob;
315 int *error = &argp->error;
321 if (copy_from_user(¶ms, (void __user *)(uintptr_t)argp->data, sizeof(params)))
324 memset(&start, 0, sizeof(start));
327 if (params.dh_uaddr) {
328 dh_blob = psp_copy_user_blob(params.dh_uaddr, params.dh_len);
330 return PTR_ERR(dh_blob);
332 start.dh_cert_address = __sme_set(__pa(dh_blob));
333 start.dh_cert_len = params.dh_len;
337 if (params.session_uaddr) {
338 session_blob = psp_copy_user_blob(params.session_uaddr, params.session_len);
339 if (IS_ERR(session_blob)) {
340 ret = PTR_ERR(session_blob);
344 start.session_address = __sme_set(__pa(session_blob));
345 start.session_len = params.session_len;
348 start.handle = params.handle;
349 start.policy = params.policy;
351 /* create memory encryption context */
352 ret = __sev_issue_cmd(argp->sev_fd, SEV_CMD_LAUNCH_START, &start, error);
356 /* Bind ASID to this guest */
357 ret = sev_bind_asid(kvm, start.handle, error);
359 sev_decommission(start.handle);
363 /* return handle to userspace */
364 params.handle = start.handle;
365 if (copy_to_user((void __user *)(uintptr_t)argp->data, ¶ms, sizeof(params))) {
366 sev_unbind_asid(kvm, start.handle);
371 sev->handle = start.handle;
372 sev->fd = argp->sev_fd;
381 static struct page **sev_pin_memory(struct kvm *kvm, unsigned long uaddr,
382 unsigned long ulen, unsigned long *n,
385 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
386 unsigned long npages, size;
388 unsigned long locked, lock_limit;
390 unsigned long first, last;
393 lockdep_assert_held(&kvm->lock);
395 if (ulen == 0 || uaddr + ulen < uaddr)
396 return ERR_PTR(-EINVAL);
398 /* Calculate number of pages. */
399 first = (uaddr & PAGE_MASK) >> PAGE_SHIFT;
400 last = ((uaddr + ulen - 1) & PAGE_MASK) >> PAGE_SHIFT;
401 npages = (last - first + 1);
403 locked = sev->pages_locked + npages;
404 lock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
405 if (locked > lock_limit && !capable(CAP_IPC_LOCK)) {
406 pr_err("SEV: %lu locked pages exceed the lock limit of %lu.\n", locked, lock_limit);
407 return ERR_PTR(-ENOMEM);
410 if (WARN_ON_ONCE(npages > INT_MAX))
411 return ERR_PTR(-EINVAL);
413 /* Avoid using vmalloc for smaller buffers. */
414 size = npages * sizeof(struct page *);
415 if (size > PAGE_SIZE)
416 pages = __vmalloc(size, GFP_KERNEL_ACCOUNT | __GFP_ZERO);
418 pages = kmalloc(size, GFP_KERNEL_ACCOUNT);
421 return ERR_PTR(-ENOMEM);
423 /* Pin the user virtual address. */
424 npinned = pin_user_pages_fast(uaddr, npages, write ? FOLL_WRITE : 0, pages);
425 if (npinned != npages) {
426 pr_err("SEV: Failure locking %lu pages.\n", npages);
432 sev->pages_locked = locked;
438 unpin_user_pages(pages, npinned);
444 static void sev_unpin_memory(struct kvm *kvm, struct page **pages,
445 unsigned long npages)
447 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
449 unpin_user_pages(pages, npages);
451 sev->pages_locked -= npages;
454 static void sev_clflush_pages(struct page *pages[], unsigned long npages)
456 uint8_t *page_virtual;
459 if (this_cpu_has(X86_FEATURE_SME_COHERENT) || npages == 0 ||
463 for (i = 0; i < npages; i++) {
464 page_virtual = kmap_atomic(pages[i]);
465 clflush_cache_range(page_virtual, PAGE_SIZE);
466 kunmap_atomic(page_virtual);
470 static unsigned long get_num_contig_pages(unsigned long idx,
471 struct page **inpages, unsigned long npages)
473 unsigned long paddr, next_paddr;
474 unsigned long i = idx + 1, pages = 1;
476 /* find the number of contiguous pages starting from idx */
477 paddr = __sme_page_pa(inpages[idx]);
479 next_paddr = __sme_page_pa(inpages[i++]);
480 if ((paddr + PAGE_SIZE) == next_paddr) {
491 static int sev_launch_update_data(struct kvm *kvm, struct kvm_sev_cmd *argp)
493 unsigned long vaddr, vaddr_end, next_vaddr, npages, pages, size, i;
494 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
495 struct kvm_sev_launch_update_data params;
496 struct sev_data_launch_update_data data;
497 struct page **inpages;
503 if (copy_from_user(¶ms, (void __user *)(uintptr_t)argp->data, sizeof(params)))
506 vaddr = params.uaddr;
508 vaddr_end = vaddr + size;
510 /* Lock the user memory. */
511 inpages = sev_pin_memory(kvm, vaddr, size, &npages, 1);
513 return PTR_ERR(inpages);
516 * Flush (on non-coherent CPUs) before LAUNCH_UPDATE encrypts pages in
517 * place; the cache may contain the data that was written unencrypted.
519 sev_clflush_pages(inpages, npages);
522 data.handle = sev->handle;
524 for (i = 0; vaddr < vaddr_end; vaddr = next_vaddr, i += pages) {
528 * If the user buffer is not page-aligned, calculate the offset
531 offset = vaddr & (PAGE_SIZE - 1);
533 /* Calculate the number of pages that can be encrypted in one go. */
534 pages = get_num_contig_pages(i, inpages, npages);
536 len = min_t(size_t, ((pages * PAGE_SIZE) - offset), size);
539 data.address = __sme_page_pa(inpages[i]) + offset;
540 ret = sev_issue_cmd(kvm, SEV_CMD_LAUNCH_UPDATE_DATA, &data, &argp->error);
545 next_vaddr = vaddr + len;
549 /* content of memory is updated, mark pages dirty */
550 for (i = 0; i < npages; i++) {
551 set_page_dirty_lock(inpages[i]);
552 mark_page_accessed(inpages[i]);
554 /* unlock the user pages */
555 sev_unpin_memory(kvm, inpages, npages);
559 static int sev_es_sync_vmsa(struct vcpu_svm *svm)
561 struct vmcb_save_area *save = &svm->vmcb->save;
563 /* Check some debug related fields before encrypting the VMSA */
564 if (svm->vcpu.guest_debug || (save->dr7 & ~DR7_FIXED_1))
567 /* Sync registgers */
568 save->rax = svm->vcpu.arch.regs[VCPU_REGS_RAX];
569 save->rbx = svm->vcpu.arch.regs[VCPU_REGS_RBX];
570 save->rcx = svm->vcpu.arch.regs[VCPU_REGS_RCX];
571 save->rdx = svm->vcpu.arch.regs[VCPU_REGS_RDX];
572 save->rsp = svm->vcpu.arch.regs[VCPU_REGS_RSP];
573 save->rbp = svm->vcpu.arch.regs[VCPU_REGS_RBP];
574 save->rsi = svm->vcpu.arch.regs[VCPU_REGS_RSI];
575 save->rdi = svm->vcpu.arch.regs[VCPU_REGS_RDI];
577 save->r8 = svm->vcpu.arch.regs[VCPU_REGS_R8];
578 save->r9 = svm->vcpu.arch.regs[VCPU_REGS_R9];
579 save->r10 = svm->vcpu.arch.regs[VCPU_REGS_R10];
580 save->r11 = svm->vcpu.arch.regs[VCPU_REGS_R11];
581 save->r12 = svm->vcpu.arch.regs[VCPU_REGS_R12];
582 save->r13 = svm->vcpu.arch.regs[VCPU_REGS_R13];
583 save->r14 = svm->vcpu.arch.regs[VCPU_REGS_R14];
584 save->r15 = svm->vcpu.arch.regs[VCPU_REGS_R15];
586 save->rip = svm->vcpu.arch.regs[VCPU_REGS_RIP];
588 /* Sync some non-GPR registers before encrypting */
589 save->xcr0 = svm->vcpu.arch.xcr0;
590 save->pkru = svm->vcpu.arch.pkru;
591 save->xss = svm->vcpu.arch.ia32_xss;
592 save->dr6 = svm->vcpu.arch.dr6;
595 * SEV-ES will use a VMSA that is pointed to by the VMCB, not
596 * the traditional VMSA that is part of the VMCB. Copy the
597 * traditional VMSA as it has been built so far (in prep
598 * for LAUNCH_UPDATE_VMSA) to be the initial SEV-ES state.
600 memcpy(svm->sev_es.vmsa, save, sizeof(*save));
605 static int __sev_launch_update_vmsa(struct kvm *kvm, struct kvm_vcpu *vcpu,
608 struct sev_data_launch_update_vmsa vmsa;
609 struct vcpu_svm *svm = to_svm(vcpu);
612 /* Perform some pre-encryption checks against the VMSA */
613 ret = sev_es_sync_vmsa(svm);
618 * The LAUNCH_UPDATE_VMSA command will perform in-place encryption of
619 * the VMSA memory content (i.e it will write the same memory region
620 * with the guest's key), so invalidate it first.
622 clflush_cache_range(svm->sev_es.vmsa, PAGE_SIZE);
625 vmsa.handle = to_kvm_svm(kvm)->sev_info.handle;
626 vmsa.address = __sme_pa(svm->sev_es.vmsa);
627 vmsa.len = PAGE_SIZE;
628 ret = sev_issue_cmd(kvm, SEV_CMD_LAUNCH_UPDATE_VMSA, &vmsa, error);
632 vcpu->arch.guest_state_protected = true;
636 static int sev_launch_update_vmsa(struct kvm *kvm, struct kvm_sev_cmd *argp)
638 struct kvm_vcpu *vcpu;
642 if (!sev_es_guest(kvm))
645 kvm_for_each_vcpu(i, vcpu, kvm) {
646 ret = mutex_lock_killable(&vcpu->mutex);
650 ret = __sev_launch_update_vmsa(kvm, vcpu, &argp->error);
652 mutex_unlock(&vcpu->mutex);
660 static int sev_launch_measure(struct kvm *kvm, struct kvm_sev_cmd *argp)
662 void __user *measure = (void __user *)(uintptr_t)argp->data;
663 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
664 struct sev_data_launch_measure data;
665 struct kvm_sev_launch_measure params;
666 void __user *p = NULL;
673 if (copy_from_user(¶ms, measure, sizeof(params)))
676 memset(&data, 0, sizeof(data));
678 /* User wants to query the blob length */
682 p = (void __user *)(uintptr_t)params.uaddr;
684 if (params.len > SEV_FW_BLOB_MAX_SIZE)
687 blob = kmalloc(params.len, GFP_KERNEL_ACCOUNT);
691 data.address = __psp_pa(blob);
692 data.len = params.len;
696 data.handle = sev->handle;
697 ret = sev_issue_cmd(kvm, SEV_CMD_LAUNCH_MEASURE, &data, &argp->error);
700 * If we query the session length, FW responded with expected data.
709 if (copy_to_user(p, blob, params.len))
714 params.len = data.len;
715 if (copy_to_user(measure, ¶ms, sizeof(params)))
722 static int sev_launch_finish(struct kvm *kvm, struct kvm_sev_cmd *argp)
724 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
725 struct sev_data_launch_finish data;
730 data.handle = sev->handle;
731 return sev_issue_cmd(kvm, SEV_CMD_LAUNCH_FINISH, &data, &argp->error);
734 static int sev_guest_status(struct kvm *kvm, struct kvm_sev_cmd *argp)
736 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
737 struct kvm_sev_guest_status params;
738 struct sev_data_guest_status data;
744 memset(&data, 0, sizeof(data));
746 data.handle = sev->handle;
747 ret = sev_issue_cmd(kvm, SEV_CMD_GUEST_STATUS, &data, &argp->error);
751 params.policy = data.policy;
752 params.state = data.state;
753 params.handle = data.handle;
755 if (copy_to_user((void __user *)(uintptr_t)argp->data, ¶ms, sizeof(params)))
761 static int __sev_issue_dbg_cmd(struct kvm *kvm, unsigned long src,
762 unsigned long dst, int size,
763 int *error, bool enc)
765 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
766 struct sev_data_dbg data;
769 data.handle = sev->handle;
774 return sev_issue_cmd(kvm,
775 enc ? SEV_CMD_DBG_ENCRYPT : SEV_CMD_DBG_DECRYPT,
779 static int __sev_dbg_decrypt(struct kvm *kvm, unsigned long src_paddr,
780 unsigned long dst_paddr, int sz, int *err)
785 * Its safe to read more than we are asked, caller should ensure that
786 * destination has enough space.
788 offset = src_paddr & 15;
789 src_paddr = round_down(src_paddr, 16);
790 sz = round_up(sz + offset, 16);
792 return __sev_issue_dbg_cmd(kvm, src_paddr, dst_paddr, sz, err, false);
795 static int __sev_dbg_decrypt_user(struct kvm *kvm, unsigned long paddr,
796 void __user *dst_uaddr,
797 unsigned long dst_paddr,
800 struct page *tpage = NULL;
803 /* if inputs are not 16-byte then use intermediate buffer */
804 if (!IS_ALIGNED(dst_paddr, 16) ||
805 !IS_ALIGNED(paddr, 16) ||
806 !IS_ALIGNED(size, 16)) {
807 tpage = (void *)alloc_page(GFP_KERNEL);
811 dst_paddr = __sme_page_pa(tpage);
814 ret = __sev_dbg_decrypt(kvm, paddr, dst_paddr, size, err);
820 if (copy_to_user(dst_uaddr, page_address(tpage) + offset, size))
831 static int __sev_dbg_encrypt_user(struct kvm *kvm, unsigned long paddr,
833 unsigned long dst_paddr,
834 void __user *dst_vaddr,
835 int size, int *error)
837 struct page *src_tpage = NULL;
838 struct page *dst_tpage = NULL;
841 /* If source buffer is not aligned then use an intermediate buffer */
842 if (!IS_ALIGNED((unsigned long)vaddr, 16)) {
843 src_tpage = alloc_page(GFP_KERNEL);
847 if (copy_from_user(page_address(src_tpage), vaddr, size)) {
848 __free_page(src_tpage);
852 paddr = __sme_page_pa(src_tpage);
856 * If destination buffer or length is not aligned then do read-modify-write:
857 * - decrypt destination in an intermediate buffer
858 * - copy the source buffer in an intermediate buffer
859 * - use the intermediate buffer as source buffer
861 if (!IS_ALIGNED((unsigned long)dst_vaddr, 16) || !IS_ALIGNED(size, 16)) {
864 dst_tpage = alloc_page(GFP_KERNEL);
870 ret = __sev_dbg_decrypt(kvm, dst_paddr,
871 __sme_page_pa(dst_tpage), size, error);
876 * If source is kernel buffer then use memcpy() otherwise
879 dst_offset = dst_paddr & 15;
882 memcpy(page_address(dst_tpage) + dst_offset,
883 page_address(src_tpage), size);
885 if (copy_from_user(page_address(dst_tpage) + dst_offset,
892 paddr = __sme_page_pa(dst_tpage);
893 dst_paddr = round_down(dst_paddr, 16);
894 len = round_up(size, 16);
897 ret = __sev_issue_dbg_cmd(kvm, paddr, dst_paddr, len, error, true);
901 __free_page(src_tpage);
903 __free_page(dst_tpage);
907 static int sev_dbg_crypt(struct kvm *kvm, struct kvm_sev_cmd *argp, bool dec)
909 unsigned long vaddr, vaddr_end, next_vaddr;
910 unsigned long dst_vaddr;
911 struct page **src_p, **dst_p;
912 struct kvm_sev_dbg debug;
920 if (copy_from_user(&debug, (void __user *)(uintptr_t)argp->data, sizeof(debug)))
923 if (!debug.len || debug.src_uaddr + debug.len < debug.src_uaddr)
925 if (!debug.dst_uaddr)
928 vaddr = debug.src_uaddr;
930 vaddr_end = vaddr + size;
931 dst_vaddr = debug.dst_uaddr;
933 for (; vaddr < vaddr_end; vaddr = next_vaddr) {
934 int len, s_off, d_off;
936 /* lock userspace source and destination page */
937 src_p = sev_pin_memory(kvm, vaddr & PAGE_MASK, PAGE_SIZE, &n, 0);
939 return PTR_ERR(src_p);
941 dst_p = sev_pin_memory(kvm, dst_vaddr & PAGE_MASK, PAGE_SIZE, &n, 1);
943 sev_unpin_memory(kvm, src_p, n);
944 return PTR_ERR(dst_p);
948 * Flush (on non-coherent CPUs) before DBG_{DE,EN}CRYPT read or modify
949 * the pages; flush the destination too so that future accesses do not
952 sev_clflush_pages(src_p, 1);
953 sev_clflush_pages(dst_p, 1);
956 * Since user buffer may not be page aligned, calculate the
957 * offset within the page.
959 s_off = vaddr & ~PAGE_MASK;
960 d_off = dst_vaddr & ~PAGE_MASK;
961 len = min_t(size_t, (PAGE_SIZE - s_off), size);
964 ret = __sev_dbg_decrypt_user(kvm,
965 __sme_page_pa(src_p[0]) + s_off,
966 (void __user *)dst_vaddr,
967 __sme_page_pa(dst_p[0]) + d_off,
970 ret = __sev_dbg_encrypt_user(kvm,
971 __sme_page_pa(src_p[0]) + s_off,
972 (void __user *)vaddr,
973 __sme_page_pa(dst_p[0]) + d_off,
974 (void __user *)dst_vaddr,
977 sev_unpin_memory(kvm, src_p, n);
978 sev_unpin_memory(kvm, dst_p, n);
983 next_vaddr = vaddr + len;
984 dst_vaddr = dst_vaddr + len;
991 static int sev_launch_secret(struct kvm *kvm, struct kvm_sev_cmd *argp)
993 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
994 struct sev_data_launch_secret data;
995 struct kvm_sev_launch_secret params;
1001 if (!sev_guest(kvm))
1004 if (copy_from_user(¶ms, (void __user *)(uintptr_t)argp->data, sizeof(params)))
1007 pages = sev_pin_memory(kvm, params.guest_uaddr, params.guest_len, &n, 1);
1009 return PTR_ERR(pages);
1012 * Flush (on non-coherent CPUs) before LAUNCH_SECRET encrypts pages in
1013 * place; the cache may contain the data that was written unencrypted.
1015 sev_clflush_pages(pages, n);
1018 * The secret must be copied into contiguous memory region, lets verify
1019 * that userspace memory pages are contiguous before we issue command.
1021 if (get_num_contig_pages(0, pages, n) != n) {
1023 goto e_unpin_memory;
1026 memset(&data, 0, sizeof(data));
1028 offset = params.guest_uaddr & (PAGE_SIZE - 1);
1029 data.guest_address = __sme_page_pa(pages[0]) + offset;
1030 data.guest_len = params.guest_len;
1032 blob = psp_copy_user_blob(params.trans_uaddr, params.trans_len);
1034 ret = PTR_ERR(blob);
1035 goto e_unpin_memory;
1038 data.trans_address = __psp_pa(blob);
1039 data.trans_len = params.trans_len;
1041 hdr = psp_copy_user_blob(params.hdr_uaddr, params.hdr_len);
1046 data.hdr_address = __psp_pa(hdr);
1047 data.hdr_len = params.hdr_len;
1049 data.handle = sev->handle;
1050 ret = sev_issue_cmd(kvm, SEV_CMD_LAUNCH_UPDATE_SECRET, &data, &argp->error);
1057 /* content of memory is updated, mark pages dirty */
1058 for (i = 0; i < n; i++) {
1059 set_page_dirty_lock(pages[i]);
1060 mark_page_accessed(pages[i]);
1062 sev_unpin_memory(kvm, pages, n);
1066 static int sev_get_attestation_report(struct kvm *kvm, struct kvm_sev_cmd *argp)
1068 void __user *report = (void __user *)(uintptr_t)argp->data;
1069 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1070 struct sev_data_attestation_report data;
1071 struct kvm_sev_attestation_report params;
1076 if (!sev_guest(kvm))
1079 if (copy_from_user(¶ms, (void __user *)(uintptr_t)argp->data, sizeof(params)))
1082 memset(&data, 0, sizeof(data));
1084 /* User wants to query the blob length */
1088 p = (void __user *)(uintptr_t)params.uaddr;
1090 if (params.len > SEV_FW_BLOB_MAX_SIZE)
1093 blob = kmalloc(params.len, GFP_KERNEL_ACCOUNT);
1097 data.address = __psp_pa(blob);
1098 data.len = params.len;
1099 memcpy(data.mnonce, params.mnonce, sizeof(params.mnonce));
1102 data.handle = sev->handle;
1103 ret = sev_issue_cmd(kvm, SEV_CMD_ATTESTATION_REPORT, &data, &argp->error);
1105 * If we query the session length, FW responded with expected data.
1114 if (copy_to_user(p, blob, params.len))
1119 params.len = data.len;
1120 if (copy_to_user(report, ¶ms, sizeof(params)))
1127 /* Userspace wants to query session length. */
1129 __sev_send_start_query_session_length(struct kvm *kvm, struct kvm_sev_cmd *argp,
1130 struct kvm_sev_send_start *params)
1132 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1133 struct sev_data_send_start data;
1136 memset(&data, 0, sizeof(data));
1137 data.handle = sev->handle;
1138 ret = sev_issue_cmd(kvm, SEV_CMD_SEND_START, &data, &argp->error);
1140 params->session_len = data.session_len;
1141 if (copy_to_user((void __user *)(uintptr_t)argp->data, params,
1142 sizeof(struct kvm_sev_send_start)))
1148 static int sev_send_start(struct kvm *kvm, struct kvm_sev_cmd *argp)
1150 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1151 struct sev_data_send_start data;
1152 struct kvm_sev_send_start params;
1153 void *amd_certs, *session_data;
1154 void *pdh_cert, *plat_certs;
1157 if (!sev_guest(kvm))
1160 if (copy_from_user(¶ms, (void __user *)(uintptr_t)argp->data,
1161 sizeof(struct kvm_sev_send_start)))
1164 /* if session_len is zero, userspace wants to query the session length */
1165 if (!params.session_len)
1166 return __sev_send_start_query_session_length(kvm, argp,
1169 /* some sanity checks */
1170 if (!params.pdh_cert_uaddr || !params.pdh_cert_len ||
1171 !params.session_uaddr || params.session_len > SEV_FW_BLOB_MAX_SIZE)
1174 /* allocate the memory to hold the session data blob */
1175 session_data = kmalloc(params.session_len, GFP_KERNEL_ACCOUNT);
1179 /* copy the certificate blobs from userspace */
1180 pdh_cert = psp_copy_user_blob(params.pdh_cert_uaddr,
1181 params.pdh_cert_len);
1182 if (IS_ERR(pdh_cert)) {
1183 ret = PTR_ERR(pdh_cert);
1184 goto e_free_session;
1187 plat_certs = psp_copy_user_blob(params.plat_certs_uaddr,
1188 params.plat_certs_len);
1189 if (IS_ERR(plat_certs)) {
1190 ret = PTR_ERR(plat_certs);
1194 amd_certs = psp_copy_user_blob(params.amd_certs_uaddr,
1195 params.amd_certs_len);
1196 if (IS_ERR(amd_certs)) {
1197 ret = PTR_ERR(amd_certs);
1198 goto e_free_plat_cert;
1201 /* populate the FW SEND_START field with system physical address */
1202 memset(&data, 0, sizeof(data));
1203 data.pdh_cert_address = __psp_pa(pdh_cert);
1204 data.pdh_cert_len = params.pdh_cert_len;
1205 data.plat_certs_address = __psp_pa(plat_certs);
1206 data.plat_certs_len = params.plat_certs_len;
1207 data.amd_certs_address = __psp_pa(amd_certs);
1208 data.amd_certs_len = params.amd_certs_len;
1209 data.session_address = __psp_pa(session_data);
1210 data.session_len = params.session_len;
1211 data.handle = sev->handle;
1213 ret = sev_issue_cmd(kvm, SEV_CMD_SEND_START, &data, &argp->error);
1215 if (!ret && copy_to_user((void __user *)(uintptr_t)params.session_uaddr,
1216 session_data, params.session_len)) {
1218 goto e_free_amd_cert;
1221 params.policy = data.policy;
1222 params.session_len = data.session_len;
1223 if (copy_to_user((void __user *)(uintptr_t)argp->data, ¶ms,
1224 sizeof(struct kvm_sev_send_start)))
1234 kfree(session_data);
1238 /* Userspace wants to query either header or trans length. */
1240 __sev_send_update_data_query_lengths(struct kvm *kvm, struct kvm_sev_cmd *argp,
1241 struct kvm_sev_send_update_data *params)
1243 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1244 struct sev_data_send_update_data data;
1247 memset(&data, 0, sizeof(data));
1248 data.handle = sev->handle;
1249 ret = sev_issue_cmd(kvm, SEV_CMD_SEND_UPDATE_DATA, &data, &argp->error);
1251 params->hdr_len = data.hdr_len;
1252 params->trans_len = data.trans_len;
1254 if (copy_to_user((void __user *)(uintptr_t)argp->data, params,
1255 sizeof(struct kvm_sev_send_update_data)))
1261 static int sev_send_update_data(struct kvm *kvm, struct kvm_sev_cmd *argp)
1263 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1264 struct sev_data_send_update_data data;
1265 struct kvm_sev_send_update_data params;
1266 void *hdr, *trans_data;
1267 struct page **guest_page;
1271 if (!sev_guest(kvm))
1274 if (copy_from_user(¶ms, (void __user *)(uintptr_t)argp->data,
1275 sizeof(struct kvm_sev_send_update_data)))
1278 /* userspace wants to query either header or trans length */
1279 if (!params.trans_len || !params.hdr_len)
1280 return __sev_send_update_data_query_lengths(kvm, argp, ¶ms);
1282 if (!params.trans_uaddr || !params.guest_uaddr ||
1283 !params.guest_len || !params.hdr_uaddr)
1286 /* Check if we are crossing the page boundary */
1287 offset = params.guest_uaddr & (PAGE_SIZE - 1);
1288 if ((params.guest_len + offset > PAGE_SIZE))
1291 /* Pin guest memory */
1292 guest_page = sev_pin_memory(kvm, params.guest_uaddr & PAGE_MASK,
1294 if (IS_ERR(guest_page))
1295 return PTR_ERR(guest_page);
1297 /* allocate memory for header and transport buffer */
1299 hdr = kmalloc(params.hdr_len, GFP_KERNEL_ACCOUNT);
1303 trans_data = kmalloc(params.trans_len, GFP_KERNEL_ACCOUNT);
1307 memset(&data, 0, sizeof(data));
1308 data.hdr_address = __psp_pa(hdr);
1309 data.hdr_len = params.hdr_len;
1310 data.trans_address = __psp_pa(trans_data);
1311 data.trans_len = params.trans_len;
1313 /* The SEND_UPDATE_DATA command requires C-bit to be always set. */
1314 data.guest_address = (page_to_pfn(guest_page[0]) << PAGE_SHIFT) + offset;
1315 data.guest_address |= sev_me_mask;
1316 data.guest_len = params.guest_len;
1317 data.handle = sev->handle;
1319 ret = sev_issue_cmd(kvm, SEV_CMD_SEND_UPDATE_DATA, &data, &argp->error);
1322 goto e_free_trans_data;
1324 /* copy transport buffer to user space */
1325 if (copy_to_user((void __user *)(uintptr_t)params.trans_uaddr,
1326 trans_data, params.trans_len)) {
1328 goto e_free_trans_data;
1331 /* Copy packet header to userspace. */
1332 if (copy_to_user((void __user *)(uintptr_t)params.hdr_uaddr, hdr,
1341 sev_unpin_memory(kvm, guest_page, n);
1346 static int sev_send_finish(struct kvm *kvm, struct kvm_sev_cmd *argp)
1348 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1349 struct sev_data_send_finish data;
1351 if (!sev_guest(kvm))
1354 data.handle = sev->handle;
1355 return sev_issue_cmd(kvm, SEV_CMD_SEND_FINISH, &data, &argp->error);
1358 static int sev_send_cancel(struct kvm *kvm, struct kvm_sev_cmd *argp)
1360 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1361 struct sev_data_send_cancel data;
1363 if (!sev_guest(kvm))
1366 data.handle = sev->handle;
1367 return sev_issue_cmd(kvm, SEV_CMD_SEND_CANCEL, &data, &argp->error);
1370 static int sev_receive_start(struct kvm *kvm, struct kvm_sev_cmd *argp)
1372 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1373 struct sev_data_receive_start start;
1374 struct kvm_sev_receive_start params;
1375 int *error = &argp->error;
1380 if (!sev_guest(kvm))
1383 /* Get parameter from the userspace */
1384 if (copy_from_user(¶ms, (void __user *)(uintptr_t)argp->data,
1385 sizeof(struct kvm_sev_receive_start)))
1388 /* some sanity checks */
1389 if (!params.pdh_uaddr || !params.pdh_len ||
1390 !params.session_uaddr || !params.session_len)
1393 pdh_data = psp_copy_user_blob(params.pdh_uaddr, params.pdh_len);
1394 if (IS_ERR(pdh_data))
1395 return PTR_ERR(pdh_data);
1397 session_data = psp_copy_user_blob(params.session_uaddr,
1398 params.session_len);
1399 if (IS_ERR(session_data)) {
1400 ret = PTR_ERR(session_data);
1404 memset(&start, 0, sizeof(start));
1405 start.handle = params.handle;
1406 start.policy = params.policy;
1407 start.pdh_cert_address = __psp_pa(pdh_data);
1408 start.pdh_cert_len = params.pdh_len;
1409 start.session_address = __psp_pa(session_data);
1410 start.session_len = params.session_len;
1412 /* create memory encryption context */
1413 ret = __sev_issue_cmd(argp->sev_fd, SEV_CMD_RECEIVE_START, &start,
1416 goto e_free_session;
1418 /* Bind ASID to this guest */
1419 ret = sev_bind_asid(kvm, start.handle, error);
1421 sev_decommission(start.handle);
1422 goto e_free_session;
1425 params.handle = start.handle;
1426 if (copy_to_user((void __user *)(uintptr_t)argp->data,
1427 ¶ms, sizeof(struct kvm_sev_receive_start))) {
1429 sev_unbind_asid(kvm, start.handle);
1430 goto e_free_session;
1433 sev->handle = start.handle;
1434 sev->fd = argp->sev_fd;
1437 kfree(session_data);
1444 static int sev_receive_update_data(struct kvm *kvm, struct kvm_sev_cmd *argp)
1446 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1447 struct kvm_sev_receive_update_data params;
1448 struct sev_data_receive_update_data data;
1449 void *hdr = NULL, *trans = NULL;
1450 struct page **guest_page;
1454 if (!sev_guest(kvm))
1457 if (copy_from_user(¶ms, (void __user *)(uintptr_t)argp->data,
1458 sizeof(struct kvm_sev_receive_update_data)))
1461 if (!params.hdr_uaddr || !params.hdr_len ||
1462 !params.guest_uaddr || !params.guest_len ||
1463 !params.trans_uaddr || !params.trans_len)
1466 /* Check if we are crossing the page boundary */
1467 offset = params.guest_uaddr & (PAGE_SIZE - 1);
1468 if ((params.guest_len + offset > PAGE_SIZE))
1471 hdr = psp_copy_user_blob(params.hdr_uaddr, params.hdr_len);
1473 return PTR_ERR(hdr);
1475 trans = psp_copy_user_blob(params.trans_uaddr, params.trans_len);
1476 if (IS_ERR(trans)) {
1477 ret = PTR_ERR(trans);
1481 memset(&data, 0, sizeof(data));
1482 data.hdr_address = __psp_pa(hdr);
1483 data.hdr_len = params.hdr_len;
1484 data.trans_address = __psp_pa(trans);
1485 data.trans_len = params.trans_len;
1487 /* Pin guest memory */
1488 guest_page = sev_pin_memory(kvm, params.guest_uaddr & PAGE_MASK,
1490 if (IS_ERR(guest_page)) {
1491 ret = PTR_ERR(guest_page);
1496 * Flush (on non-coherent CPUs) before RECEIVE_UPDATE_DATA, the PSP
1497 * encrypts the written data with the guest's key, and the cache may
1498 * contain dirty, unencrypted data.
1500 sev_clflush_pages(guest_page, n);
1502 /* The RECEIVE_UPDATE_DATA command requires C-bit to be always set. */
1503 data.guest_address = (page_to_pfn(guest_page[0]) << PAGE_SHIFT) + offset;
1504 data.guest_address |= sev_me_mask;
1505 data.guest_len = params.guest_len;
1506 data.handle = sev->handle;
1508 ret = sev_issue_cmd(kvm, SEV_CMD_RECEIVE_UPDATE_DATA, &data,
1511 sev_unpin_memory(kvm, guest_page, n);
1521 static int sev_receive_finish(struct kvm *kvm, struct kvm_sev_cmd *argp)
1523 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1524 struct sev_data_receive_finish data;
1526 if (!sev_guest(kvm))
1529 data.handle = sev->handle;
1530 return sev_issue_cmd(kvm, SEV_CMD_RECEIVE_FINISH, &data, &argp->error);
1533 static bool is_cmd_allowed_from_mirror(u32 cmd_id)
1536 * Allow mirrors VM to call KVM_SEV_LAUNCH_UPDATE_VMSA to enable SEV-ES
1537 * active mirror VMs. Also allow the debugging and status commands.
1539 if (cmd_id == KVM_SEV_LAUNCH_UPDATE_VMSA ||
1540 cmd_id == KVM_SEV_GUEST_STATUS || cmd_id == KVM_SEV_DBG_DECRYPT ||
1541 cmd_id == KVM_SEV_DBG_ENCRYPT)
1547 static int sev_lock_two_vms(struct kvm *dst_kvm, struct kvm *src_kvm)
1549 struct kvm_sev_info *dst_sev = &to_kvm_svm(dst_kvm)->sev_info;
1550 struct kvm_sev_info *src_sev = &to_kvm_svm(src_kvm)->sev_info;
1553 if (dst_kvm == src_kvm)
1557 * Bail if these VMs are already involved in a migration to avoid
1558 * deadlock between two VMs trying to migrate to/from each other.
1560 if (atomic_cmpxchg_acquire(&dst_sev->migration_in_progress, 0, 1))
1563 if (atomic_cmpxchg_acquire(&src_sev->migration_in_progress, 0, 1))
1567 if (mutex_lock_killable(&dst_kvm->lock))
1569 if (mutex_lock_killable_nested(&src_kvm->lock, SINGLE_DEPTH_NESTING))
1574 mutex_unlock(&dst_kvm->lock);
1576 atomic_set_release(&src_sev->migration_in_progress, 0);
1578 atomic_set_release(&dst_sev->migration_in_progress, 0);
1582 static void sev_unlock_two_vms(struct kvm *dst_kvm, struct kvm *src_kvm)
1584 struct kvm_sev_info *dst_sev = &to_kvm_svm(dst_kvm)->sev_info;
1585 struct kvm_sev_info *src_sev = &to_kvm_svm(src_kvm)->sev_info;
1587 mutex_unlock(&dst_kvm->lock);
1588 mutex_unlock(&src_kvm->lock);
1589 atomic_set_release(&dst_sev->migration_in_progress, 0);
1590 atomic_set_release(&src_sev->migration_in_progress, 0);
1594 static int sev_lock_vcpus_for_migration(struct kvm *kvm)
1596 struct kvm_vcpu *vcpu;
1599 kvm_for_each_vcpu(i, vcpu, kvm) {
1600 if (mutex_lock_killable(&vcpu->mutex))
1607 kvm_for_each_vcpu(j, vcpu, kvm) {
1611 mutex_unlock(&vcpu->mutex);
1616 static void sev_unlock_vcpus_for_migration(struct kvm *kvm)
1618 struct kvm_vcpu *vcpu;
1621 kvm_for_each_vcpu(i, vcpu, kvm) {
1622 mutex_unlock(&vcpu->mutex);
1626 static void sev_migrate_from(struct kvm_sev_info *dst,
1627 struct kvm_sev_info *src)
1630 dst->asid = src->asid;
1631 dst->handle = src->handle;
1632 dst->pages_locked = src->pages_locked;
1633 dst->enc_context_owner = src->enc_context_owner;
1636 src->active = false;
1638 src->pages_locked = 0;
1639 src->enc_context_owner = NULL;
1641 list_cut_before(&dst->regions_list, &src->regions_list, &src->regions_list);
1644 static int sev_es_migrate_from(struct kvm *dst, struct kvm *src)
1647 struct kvm_vcpu *dst_vcpu, *src_vcpu;
1648 struct vcpu_svm *dst_svm, *src_svm;
1650 if (atomic_read(&src->online_vcpus) != atomic_read(&dst->online_vcpus))
1653 kvm_for_each_vcpu(i, src_vcpu, src) {
1654 if (!src_vcpu->arch.guest_state_protected)
1658 kvm_for_each_vcpu(i, src_vcpu, src) {
1659 src_svm = to_svm(src_vcpu);
1660 dst_vcpu = kvm_get_vcpu(dst, i);
1661 dst_svm = to_svm(dst_vcpu);
1664 * Transfer VMSA and GHCB state to the destination. Nullify and
1665 * clear source fields as appropriate, the state now belongs to
1668 memcpy(&dst_svm->sev_es, &src_svm->sev_es, sizeof(src_svm->sev_es));
1669 dst_svm->vmcb->control.ghcb_gpa = src_svm->vmcb->control.ghcb_gpa;
1670 dst_svm->vmcb->control.vmsa_pa = src_svm->vmcb->control.vmsa_pa;
1671 dst_vcpu->arch.guest_state_protected = true;
1673 memset(&src_svm->sev_es, 0, sizeof(src_svm->sev_es));
1674 src_svm->vmcb->control.ghcb_gpa = INVALID_PAGE;
1675 src_svm->vmcb->control.vmsa_pa = INVALID_PAGE;
1676 src_vcpu->arch.guest_state_protected = false;
1678 to_kvm_svm(src)->sev_info.es_active = false;
1679 to_kvm_svm(dst)->sev_info.es_active = true;
1684 int svm_vm_migrate_from(struct kvm *kvm, unsigned int source_fd)
1686 struct kvm_sev_info *dst_sev = &to_kvm_svm(kvm)->sev_info;
1687 struct kvm_sev_info *src_sev, *cg_cleanup_sev;
1688 struct file *source_kvm_file;
1689 struct kvm *source_kvm;
1690 bool charged = false;
1693 source_kvm_file = fget(source_fd);
1694 if (!file_is_kvm(source_kvm_file)) {
1699 source_kvm = source_kvm_file->private_data;
1700 ret = sev_lock_two_vms(kvm, source_kvm);
1704 if (sev_guest(kvm) || !sev_guest(source_kvm)) {
1709 src_sev = &to_kvm_svm(source_kvm)->sev_info;
1712 * VMs mirroring src's encryption context rely on it to keep the
1713 * ASID allocated, but below we are clearing src_sev->asid.
1715 if (src_sev->num_mirrored_vms) {
1720 dst_sev->misc_cg = get_current_misc_cg();
1721 cg_cleanup_sev = dst_sev;
1722 if (dst_sev->misc_cg != src_sev->misc_cg) {
1723 ret = sev_misc_cg_try_charge(dst_sev);
1725 goto out_dst_cgroup;
1729 ret = sev_lock_vcpus_for_migration(kvm);
1731 goto out_dst_cgroup;
1732 ret = sev_lock_vcpus_for_migration(source_kvm);
1736 if (sev_es_guest(source_kvm)) {
1737 ret = sev_es_migrate_from(kvm, source_kvm);
1739 goto out_source_vcpu;
1741 sev_migrate_from(dst_sev, src_sev);
1742 kvm_vm_dead(source_kvm);
1743 cg_cleanup_sev = src_sev;
1747 sev_unlock_vcpus_for_migration(source_kvm);
1749 sev_unlock_vcpus_for_migration(kvm);
1751 /* Operates on the source on success, on the destination on failure. */
1753 sev_misc_cg_uncharge(cg_cleanup_sev);
1754 put_misc_cg(cg_cleanup_sev->misc_cg);
1755 cg_cleanup_sev->misc_cg = NULL;
1757 sev_unlock_two_vms(kvm, source_kvm);
1759 if (source_kvm_file)
1760 fput(source_kvm_file);
1764 int svm_mem_enc_op(struct kvm *kvm, void __user *argp)
1766 struct kvm_sev_cmd sev_cmd;
1775 if (copy_from_user(&sev_cmd, argp, sizeof(struct kvm_sev_cmd)))
1778 mutex_lock(&kvm->lock);
1780 /* Only the enc_context_owner handles some memory enc operations. */
1781 if (is_mirroring_enc_context(kvm) &&
1782 !is_cmd_allowed_from_mirror(sev_cmd.id)) {
1787 switch (sev_cmd.id) {
1788 case KVM_SEV_ES_INIT:
1789 if (!sev_es_enabled) {
1795 r = sev_guest_init(kvm, &sev_cmd);
1797 case KVM_SEV_LAUNCH_START:
1798 r = sev_launch_start(kvm, &sev_cmd);
1800 case KVM_SEV_LAUNCH_UPDATE_DATA:
1801 r = sev_launch_update_data(kvm, &sev_cmd);
1803 case KVM_SEV_LAUNCH_UPDATE_VMSA:
1804 r = sev_launch_update_vmsa(kvm, &sev_cmd);
1806 case KVM_SEV_LAUNCH_MEASURE:
1807 r = sev_launch_measure(kvm, &sev_cmd);
1809 case KVM_SEV_LAUNCH_FINISH:
1810 r = sev_launch_finish(kvm, &sev_cmd);
1812 case KVM_SEV_GUEST_STATUS:
1813 r = sev_guest_status(kvm, &sev_cmd);
1815 case KVM_SEV_DBG_DECRYPT:
1816 r = sev_dbg_crypt(kvm, &sev_cmd, true);
1818 case KVM_SEV_DBG_ENCRYPT:
1819 r = sev_dbg_crypt(kvm, &sev_cmd, false);
1821 case KVM_SEV_LAUNCH_SECRET:
1822 r = sev_launch_secret(kvm, &sev_cmd);
1824 case KVM_SEV_GET_ATTESTATION_REPORT:
1825 r = sev_get_attestation_report(kvm, &sev_cmd);
1827 case KVM_SEV_SEND_START:
1828 r = sev_send_start(kvm, &sev_cmd);
1830 case KVM_SEV_SEND_UPDATE_DATA:
1831 r = sev_send_update_data(kvm, &sev_cmd);
1833 case KVM_SEV_SEND_FINISH:
1834 r = sev_send_finish(kvm, &sev_cmd);
1836 case KVM_SEV_SEND_CANCEL:
1837 r = sev_send_cancel(kvm, &sev_cmd);
1839 case KVM_SEV_RECEIVE_START:
1840 r = sev_receive_start(kvm, &sev_cmd);
1842 case KVM_SEV_RECEIVE_UPDATE_DATA:
1843 r = sev_receive_update_data(kvm, &sev_cmd);
1845 case KVM_SEV_RECEIVE_FINISH:
1846 r = sev_receive_finish(kvm, &sev_cmd);
1853 if (copy_to_user(argp, &sev_cmd, sizeof(struct kvm_sev_cmd)))
1857 mutex_unlock(&kvm->lock);
1861 int svm_register_enc_region(struct kvm *kvm,
1862 struct kvm_enc_region *range)
1864 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1865 struct enc_region *region;
1868 if (!sev_guest(kvm))
1871 /* If kvm is mirroring encryption context it isn't responsible for it */
1872 if (is_mirroring_enc_context(kvm))
1875 if (range->addr > ULONG_MAX || range->size > ULONG_MAX)
1878 region = kzalloc(sizeof(*region), GFP_KERNEL_ACCOUNT);
1882 mutex_lock(&kvm->lock);
1883 region->pages = sev_pin_memory(kvm, range->addr, range->size, ®ion->npages, 1);
1884 if (IS_ERR(region->pages)) {
1885 ret = PTR_ERR(region->pages);
1886 mutex_unlock(&kvm->lock);
1890 region->uaddr = range->addr;
1891 region->size = range->size;
1893 list_add_tail(®ion->list, &sev->regions_list);
1894 mutex_unlock(&kvm->lock);
1897 * The guest may change the memory encryption attribute from C=0 -> C=1
1898 * or vice versa for this memory range. Lets make sure caches are
1899 * flushed to ensure that guest data gets written into memory with
1902 sev_clflush_pages(region->pages, region->npages);
1911 static struct enc_region *
1912 find_enc_region(struct kvm *kvm, struct kvm_enc_region *range)
1914 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1915 struct list_head *head = &sev->regions_list;
1916 struct enc_region *i;
1918 list_for_each_entry(i, head, list) {
1919 if (i->uaddr == range->addr &&
1920 i->size == range->size)
1927 static void __unregister_enc_region_locked(struct kvm *kvm,
1928 struct enc_region *region)
1930 sev_unpin_memory(kvm, region->pages, region->npages);
1931 list_del(®ion->list);
1935 int svm_unregister_enc_region(struct kvm *kvm,
1936 struct kvm_enc_region *range)
1938 struct enc_region *region;
1941 /* If kvm is mirroring encryption context it isn't responsible for it */
1942 if (is_mirroring_enc_context(kvm))
1945 mutex_lock(&kvm->lock);
1947 if (!sev_guest(kvm)) {
1952 region = find_enc_region(kvm, range);
1959 * Ensure that all guest tagged cache entries are flushed before
1960 * releasing the pages back to the system for use. CLFLUSH will
1961 * not do this, so issue a WBINVD.
1963 wbinvd_on_all_cpus();
1965 __unregister_enc_region_locked(kvm, region);
1967 mutex_unlock(&kvm->lock);
1971 mutex_unlock(&kvm->lock);
1975 int svm_vm_copy_asid_from(struct kvm *kvm, unsigned int source_fd)
1977 struct file *source_kvm_file;
1978 struct kvm *source_kvm;
1979 struct kvm_sev_info *source_sev, *mirror_sev;
1982 source_kvm_file = fget(source_fd);
1983 if (!file_is_kvm(source_kvm_file)) {
1988 source_kvm = source_kvm_file->private_data;
1989 ret = sev_lock_two_vms(kvm, source_kvm);
1994 * Mirrors of mirrors should work, but let's not get silly. Also
1995 * disallow out-of-band SEV/SEV-ES init if the target is already an
1996 * SEV guest, or if vCPUs have been created. KVM relies on vCPUs being
1997 * created after SEV/SEV-ES initialization, e.g. to init intercepts.
1999 if (sev_guest(kvm) || !sev_guest(source_kvm) ||
2000 is_mirroring_enc_context(source_kvm) || kvm->created_vcpus) {
2006 * The mirror kvm holds an enc_context_owner ref so its asid can't
2007 * disappear until we're done with it
2009 source_sev = &to_kvm_svm(source_kvm)->sev_info;
2010 kvm_get_kvm(source_kvm);
2011 source_sev->num_mirrored_vms++;
2013 /* Set enc_context_owner and copy its encryption context over */
2014 mirror_sev = &to_kvm_svm(kvm)->sev_info;
2015 mirror_sev->enc_context_owner = source_kvm;
2016 mirror_sev->active = true;
2017 mirror_sev->asid = source_sev->asid;
2018 mirror_sev->fd = source_sev->fd;
2019 mirror_sev->es_active = source_sev->es_active;
2020 mirror_sev->handle = source_sev->handle;
2021 INIT_LIST_HEAD(&mirror_sev->regions_list);
2025 * Do not copy ap_jump_table. Since the mirror does not share the same
2026 * KVM contexts as the original, and they may have different
2031 sev_unlock_two_vms(kvm, source_kvm);
2033 if (source_kvm_file)
2034 fput(source_kvm_file);
2038 void sev_vm_destroy(struct kvm *kvm)
2040 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
2041 struct list_head *head = &sev->regions_list;
2042 struct list_head *pos, *q;
2044 WARN_ON(sev->num_mirrored_vms);
2046 if (!sev_guest(kvm))
2049 /* If this is a mirror_kvm release the enc_context_owner and skip sev cleanup */
2050 if (is_mirroring_enc_context(kvm)) {
2051 struct kvm *owner_kvm = sev->enc_context_owner;
2052 struct kvm_sev_info *owner_sev = &to_kvm_svm(owner_kvm)->sev_info;
2054 mutex_lock(&owner_kvm->lock);
2055 if (!WARN_ON(!owner_sev->num_mirrored_vms))
2056 owner_sev->num_mirrored_vms--;
2057 mutex_unlock(&owner_kvm->lock);
2058 kvm_put_kvm(owner_kvm);
2063 * Ensure that all guest tagged cache entries are flushed before
2064 * releasing the pages back to the system for use. CLFLUSH will
2065 * not do this, so issue a WBINVD.
2067 wbinvd_on_all_cpus();
2070 * if userspace was terminated before unregistering the memory regions
2071 * then lets unpin all the registered memory.
2073 if (!list_empty(head)) {
2074 list_for_each_safe(pos, q, head) {
2075 __unregister_enc_region_locked(kvm,
2076 list_entry(pos, struct enc_region, list));
2081 sev_unbind_asid(kvm, sev->handle);
2085 void __init sev_set_cpu_caps(void)
2088 kvm_cpu_cap_clear(X86_FEATURE_SEV);
2089 if (!sev_es_enabled)
2090 kvm_cpu_cap_clear(X86_FEATURE_SEV_ES);
2093 void __init sev_hardware_setup(void)
2095 #ifdef CONFIG_KVM_AMD_SEV
2096 unsigned int eax, ebx, ecx, edx, sev_asid_count, sev_es_asid_count;
2097 bool sev_es_supported = false;
2098 bool sev_supported = false;
2100 if (!sev_enabled || !npt_enabled)
2103 /* Does the CPU support SEV? */
2104 if (!boot_cpu_has(X86_FEATURE_SEV))
2107 /* Retrieve SEV CPUID information */
2108 cpuid(0x8000001f, &eax, &ebx, &ecx, &edx);
2110 /* Set encryption bit location for SEV-ES guests */
2111 sev_enc_bit = ebx & 0x3f;
2113 /* Maximum number of encrypted guests supported simultaneously */
2118 /* Minimum ASID value that should be used for SEV guest */
2120 sev_me_mask = 1UL << (ebx & 0x3f);
2123 * Initialize SEV ASID bitmaps. Allocate space for ASID 0 in the bitmap,
2124 * even though it's never used, so that the bitmap is indexed by the
2127 nr_asids = max_sev_asid + 1;
2128 sev_asid_bitmap = bitmap_zalloc(nr_asids, GFP_KERNEL);
2129 if (!sev_asid_bitmap)
2132 sev_reclaim_asid_bitmap = bitmap_zalloc(nr_asids, GFP_KERNEL);
2133 if (!sev_reclaim_asid_bitmap) {
2134 bitmap_free(sev_asid_bitmap);
2135 sev_asid_bitmap = NULL;
2139 sev_asid_count = max_sev_asid - min_sev_asid + 1;
2140 if (misc_cg_set_capacity(MISC_CG_RES_SEV, sev_asid_count))
2143 pr_info("SEV supported: %u ASIDs\n", sev_asid_count);
2144 sev_supported = true;
2146 /* SEV-ES support requested? */
2147 if (!sev_es_enabled)
2150 /* Does the CPU support SEV-ES? */
2151 if (!boot_cpu_has(X86_FEATURE_SEV_ES))
2154 /* Has the system been allocated ASIDs for SEV-ES? */
2155 if (min_sev_asid == 1)
2158 sev_es_asid_count = min_sev_asid - 1;
2159 if (misc_cg_set_capacity(MISC_CG_RES_SEV_ES, sev_es_asid_count))
2162 pr_info("SEV-ES supported: %u ASIDs\n", sev_es_asid_count);
2163 sev_es_supported = true;
2166 sev_enabled = sev_supported;
2167 sev_es_enabled = sev_es_supported;
2171 void sev_hardware_teardown(void)
2176 /* No need to take sev_bitmap_lock, all VMs have been destroyed. */
2177 sev_flush_asids(1, max_sev_asid);
2179 bitmap_free(sev_asid_bitmap);
2180 bitmap_free(sev_reclaim_asid_bitmap);
2182 misc_cg_set_capacity(MISC_CG_RES_SEV, 0);
2183 misc_cg_set_capacity(MISC_CG_RES_SEV_ES, 0);
2186 int sev_cpu_init(struct svm_cpu_data *sd)
2191 sd->sev_vmcbs = kcalloc(nr_asids, sizeof(void *), GFP_KERNEL);
2199 * Pages used by hardware to hold guest encrypted state must be flushed before
2200 * returning them to the system.
2202 static void sev_flush_guest_memory(struct vcpu_svm *svm, void *va,
2206 * If hardware enforced cache coherency for encrypted mappings of the
2207 * same physical page is supported, nothing to do.
2209 if (boot_cpu_has(X86_FEATURE_SME_COHERENT))
2213 * If the VM Page Flush MSR is supported, use it to flush the page
2214 * (using the page virtual address and the guest ASID).
2216 if (boot_cpu_has(X86_FEATURE_VM_PAGE_FLUSH)) {
2217 struct kvm_sev_info *sev;
2218 unsigned long va_start;
2221 /* Align start and stop to page boundaries. */
2222 va_start = (unsigned long)va;
2223 start = (u64)va_start & PAGE_MASK;
2224 stop = PAGE_ALIGN((u64)va_start + len);
2227 sev = &to_kvm_svm(svm->vcpu.kvm)->sev_info;
2229 while (start < stop) {
2230 wrmsrl(MSR_AMD64_VM_PAGE_FLUSH,
2239 WARN(1, "Address overflow, using WBINVD\n");
2243 * Hardware should always have one of the above features,
2244 * but if not, use WBINVD and issue a warning.
2246 WARN_ONCE(1, "Using WBINVD to flush guest memory\n");
2247 wbinvd_on_all_cpus();
2250 void sev_free_vcpu(struct kvm_vcpu *vcpu)
2252 struct vcpu_svm *svm;
2254 if (!sev_es_guest(vcpu->kvm))
2259 if (vcpu->arch.guest_state_protected)
2260 sev_flush_guest_memory(svm, svm->sev_es.vmsa, PAGE_SIZE);
2261 __free_page(virt_to_page(svm->sev_es.vmsa));
2263 if (svm->sev_es.ghcb_sa_free)
2264 kvfree(svm->sev_es.ghcb_sa);
2267 static void dump_ghcb(struct vcpu_svm *svm)
2269 struct ghcb *ghcb = svm->sev_es.ghcb;
2272 /* Re-use the dump_invalid_vmcb module parameter */
2273 if (!dump_invalid_vmcb) {
2274 pr_warn_ratelimited("set kvm_amd.dump_invalid_vmcb=1 to dump internal KVM state.\n");
2278 nbits = sizeof(ghcb->save.valid_bitmap) * 8;
2280 pr_err("GHCB (GPA=%016llx):\n", svm->vmcb->control.ghcb_gpa);
2281 pr_err("%-20s%016llx is_valid: %u\n", "sw_exit_code",
2282 ghcb->save.sw_exit_code, ghcb_sw_exit_code_is_valid(ghcb));
2283 pr_err("%-20s%016llx is_valid: %u\n", "sw_exit_info_1",
2284 ghcb->save.sw_exit_info_1, ghcb_sw_exit_info_1_is_valid(ghcb));
2285 pr_err("%-20s%016llx is_valid: %u\n", "sw_exit_info_2",
2286 ghcb->save.sw_exit_info_2, ghcb_sw_exit_info_2_is_valid(ghcb));
2287 pr_err("%-20s%016llx is_valid: %u\n", "sw_scratch",
2288 ghcb->save.sw_scratch, ghcb_sw_scratch_is_valid(ghcb));
2289 pr_err("%-20s%*pb\n", "valid_bitmap", nbits, ghcb->save.valid_bitmap);
2292 static void sev_es_sync_to_ghcb(struct vcpu_svm *svm)
2294 struct kvm_vcpu *vcpu = &svm->vcpu;
2295 struct ghcb *ghcb = svm->sev_es.ghcb;
2298 * The GHCB protocol so far allows for the following data
2300 * GPRs RAX, RBX, RCX, RDX
2302 * Copy their values, even if they may not have been written during the
2303 * VM-Exit. It's the guest's responsibility to not consume random data.
2305 ghcb_set_rax(ghcb, vcpu->arch.regs[VCPU_REGS_RAX]);
2306 ghcb_set_rbx(ghcb, vcpu->arch.regs[VCPU_REGS_RBX]);
2307 ghcb_set_rcx(ghcb, vcpu->arch.regs[VCPU_REGS_RCX]);
2308 ghcb_set_rdx(ghcb, vcpu->arch.regs[VCPU_REGS_RDX]);
2311 static void sev_es_sync_from_ghcb(struct vcpu_svm *svm)
2313 struct vmcb_control_area *control = &svm->vmcb->control;
2314 struct kvm_vcpu *vcpu = &svm->vcpu;
2315 struct ghcb *ghcb = svm->sev_es.ghcb;
2319 * The GHCB protocol so far allows for the following data
2321 * GPRs RAX, RBX, RCX, RDX
2325 * VMMCALL allows the guest to provide extra registers. KVM also
2326 * expects RSI for hypercalls, so include that, too.
2328 * Copy their values to the appropriate location if supplied.
2330 memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
2332 vcpu->arch.regs[VCPU_REGS_RAX] = ghcb_get_rax_if_valid(ghcb);
2333 vcpu->arch.regs[VCPU_REGS_RBX] = ghcb_get_rbx_if_valid(ghcb);
2334 vcpu->arch.regs[VCPU_REGS_RCX] = ghcb_get_rcx_if_valid(ghcb);
2335 vcpu->arch.regs[VCPU_REGS_RDX] = ghcb_get_rdx_if_valid(ghcb);
2336 vcpu->arch.regs[VCPU_REGS_RSI] = ghcb_get_rsi_if_valid(ghcb);
2338 svm->vmcb->save.cpl = ghcb_get_cpl_if_valid(ghcb);
2340 if (ghcb_xcr0_is_valid(ghcb)) {
2341 vcpu->arch.xcr0 = ghcb_get_xcr0(ghcb);
2342 kvm_update_cpuid_runtime(vcpu);
2345 /* Copy the GHCB exit information into the VMCB fields */
2346 exit_code = ghcb_get_sw_exit_code(ghcb);
2347 control->exit_code = lower_32_bits(exit_code);
2348 control->exit_code_hi = upper_32_bits(exit_code);
2349 control->exit_info_1 = ghcb_get_sw_exit_info_1(ghcb);
2350 control->exit_info_2 = ghcb_get_sw_exit_info_2(ghcb);
2352 /* Clear the valid entries fields */
2353 memset(ghcb->save.valid_bitmap, 0, sizeof(ghcb->save.valid_bitmap));
2356 static bool sev_es_validate_vmgexit(struct vcpu_svm *svm)
2358 struct kvm_vcpu *vcpu;
2363 ghcb = svm->sev_es.ghcb;
2366 * Retrieve the exit code now even though it may not be marked valid
2367 * as it could help with debugging.
2369 exit_code = ghcb_get_sw_exit_code(ghcb);
2371 /* Only GHCB Usage code 0 is supported */
2372 if (ghcb->ghcb_usage) {
2373 reason = GHCB_ERR_INVALID_USAGE;
2377 reason = GHCB_ERR_MISSING_INPUT;
2379 if (!ghcb_sw_exit_code_is_valid(ghcb) ||
2380 !ghcb_sw_exit_info_1_is_valid(ghcb) ||
2381 !ghcb_sw_exit_info_2_is_valid(ghcb))
2384 switch (ghcb_get_sw_exit_code(ghcb)) {
2385 case SVM_EXIT_READ_DR7:
2387 case SVM_EXIT_WRITE_DR7:
2388 if (!ghcb_rax_is_valid(ghcb))
2391 case SVM_EXIT_RDTSC:
2393 case SVM_EXIT_RDPMC:
2394 if (!ghcb_rcx_is_valid(ghcb))
2397 case SVM_EXIT_CPUID:
2398 if (!ghcb_rax_is_valid(ghcb) ||
2399 !ghcb_rcx_is_valid(ghcb))
2401 if (ghcb_get_rax(ghcb) == 0xd)
2402 if (!ghcb_xcr0_is_valid(ghcb))
2408 if (ghcb_get_sw_exit_info_1(ghcb) & SVM_IOIO_STR_MASK) {
2409 if (!ghcb_sw_scratch_is_valid(ghcb))
2412 if (!(ghcb_get_sw_exit_info_1(ghcb) & SVM_IOIO_TYPE_MASK))
2413 if (!ghcb_rax_is_valid(ghcb))
2418 if (!ghcb_rcx_is_valid(ghcb))
2420 if (ghcb_get_sw_exit_info_1(ghcb)) {
2421 if (!ghcb_rax_is_valid(ghcb) ||
2422 !ghcb_rdx_is_valid(ghcb))
2426 case SVM_EXIT_VMMCALL:
2427 if (!ghcb_rax_is_valid(ghcb) ||
2428 !ghcb_cpl_is_valid(ghcb))
2431 case SVM_EXIT_RDTSCP:
2433 case SVM_EXIT_WBINVD:
2435 case SVM_EXIT_MONITOR:
2436 if (!ghcb_rax_is_valid(ghcb) ||
2437 !ghcb_rcx_is_valid(ghcb) ||
2438 !ghcb_rdx_is_valid(ghcb))
2441 case SVM_EXIT_MWAIT:
2442 if (!ghcb_rax_is_valid(ghcb) ||
2443 !ghcb_rcx_is_valid(ghcb))
2446 case SVM_VMGEXIT_MMIO_READ:
2447 case SVM_VMGEXIT_MMIO_WRITE:
2448 if (!ghcb_sw_scratch_is_valid(ghcb))
2451 case SVM_VMGEXIT_NMI_COMPLETE:
2452 case SVM_VMGEXIT_AP_HLT_LOOP:
2453 case SVM_VMGEXIT_AP_JUMP_TABLE:
2454 case SVM_VMGEXIT_UNSUPPORTED_EVENT:
2457 reason = GHCB_ERR_INVALID_EVENT;
2466 if (reason == GHCB_ERR_INVALID_USAGE) {
2467 vcpu_unimpl(vcpu, "vmgexit: ghcb usage %#x is not valid\n",
2469 } else if (reason == GHCB_ERR_INVALID_EVENT) {
2470 vcpu_unimpl(vcpu, "vmgexit: exit code %#llx is not valid\n",
2473 vcpu_unimpl(vcpu, "vmgexit: exit code %#llx input is not valid\n",
2478 /* Clear the valid entries fields */
2479 memset(ghcb->save.valid_bitmap, 0, sizeof(ghcb->save.valid_bitmap));
2481 ghcb_set_sw_exit_info_1(ghcb, 2);
2482 ghcb_set_sw_exit_info_2(ghcb, reason);
2487 void sev_es_unmap_ghcb(struct vcpu_svm *svm)
2489 if (!svm->sev_es.ghcb)
2492 if (svm->sev_es.ghcb_sa_free) {
2494 * The scratch area lives outside the GHCB, so there is a
2495 * buffer that, depending on the operation performed, may
2496 * need to be synced, then freed.
2498 if (svm->sev_es.ghcb_sa_sync) {
2499 kvm_write_guest(svm->vcpu.kvm,
2500 ghcb_get_sw_scratch(svm->sev_es.ghcb),
2501 svm->sev_es.ghcb_sa,
2502 svm->sev_es.ghcb_sa_len);
2503 svm->sev_es.ghcb_sa_sync = false;
2506 kvfree(svm->sev_es.ghcb_sa);
2507 svm->sev_es.ghcb_sa = NULL;
2508 svm->sev_es.ghcb_sa_free = false;
2511 trace_kvm_vmgexit_exit(svm->vcpu.vcpu_id, svm->sev_es.ghcb);
2513 sev_es_sync_to_ghcb(svm);
2515 kvm_vcpu_unmap(&svm->vcpu, &svm->sev_es.ghcb_map, true);
2516 svm->sev_es.ghcb = NULL;
2519 void pre_sev_run(struct vcpu_svm *svm, int cpu)
2521 struct svm_cpu_data *sd = per_cpu(svm_data, cpu);
2522 int asid = sev_get_asid(svm->vcpu.kvm);
2524 /* Assign the asid allocated with this SEV guest */
2530 * 1) when different VMCB for the same ASID is to be run on the same host CPU.
2531 * 2) or this VMCB was executed on different host CPU in previous VMRUNs.
2533 if (sd->sev_vmcbs[asid] == svm->vmcb &&
2534 svm->vcpu.arch.last_vmentry_cpu == cpu)
2537 sd->sev_vmcbs[asid] = svm->vmcb;
2538 svm->vmcb->control.tlb_ctl = TLB_CONTROL_FLUSH_ASID;
2539 vmcb_mark_dirty(svm->vmcb, VMCB_ASID);
2542 #define GHCB_SCRATCH_AREA_LIMIT (16ULL * PAGE_SIZE)
2543 static bool setup_vmgexit_scratch(struct vcpu_svm *svm, bool sync, u64 len)
2545 struct vmcb_control_area *control = &svm->vmcb->control;
2546 struct ghcb *ghcb = svm->sev_es.ghcb;
2547 u64 ghcb_scratch_beg, ghcb_scratch_end;
2548 u64 scratch_gpa_beg, scratch_gpa_end;
2551 scratch_gpa_beg = ghcb_get_sw_scratch(ghcb);
2552 if (!scratch_gpa_beg) {
2553 pr_err("vmgexit: scratch gpa not provided\n");
2557 scratch_gpa_end = scratch_gpa_beg + len;
2558 if (scratch_gpa_end < scratch_gpa_beg) {
2559 pr_err("vmgexit: scratch length (%#llx) not valid for scratch address (%#llx)\n",
2560 len, scratch_gpa_beg);
2564 if ((scratch_gpa_beg & PAGE_MASK) == control->ghcb_gpa) {
2565 /* Scratch area begins within GHCB */
2566 ghcb_scratch_beg = control->ghcb_gpa +
2567 offsetof(struct ghcb, shared_buffer);
2568 ghcb_scratch_end = control->ghcb_gpa +
2569 offsetof(struct ghcb, reserved_1);
2572 * If the scratch area begins within the GHCB, it must be
2573 * completely contained in the GHCB shared buffer area.
2575 if (scratch_gpa_beg < ghcb_scratch_beg ||
2576 scratch_gpa_end > ghcb_scratch_end) {
2577 pr_err("vmgexit: scratch area is outside of GHCB shared buffer area (%#llx - %#llx)\n",
2578 scratch_gpa_beg, scratch_gpa_end);
2582 scratch_va = (void *)svm->sev_es.ghcb;
2583 scratch_va += (scratch_gpa_beg - control->ghcb_gpa);
2586 * The guest memory must be read into a kernel buffer, so
2589 if (len > GHCB_SCRATCH_AREA_LIMIT) {
2590 pr_err("vmgexit: scratch area exceeds KVM limits (%#llx requested, %#llx limit)\n",
2591 len, GHCB_SCRATCH_AREA_LIMIT);
2594 scratch_va = kvzalloc(len, GFP_KERNEL_ACCOUNT);
2598 if (kvm_read_guest(svm->vcpu.kvm, scratch_gpa_beg, scratch_va, len)) {
2599 /* Unable to copy scratch area from guest */
2600 pr_err("vmgexit: kvm_read_guest for scratch area failed\n");
2607 * The scratch area is outside the GHCB. The operation will
2608 * dictate whether the buffer needs to be synced before running
2609 * the vCPU next time (i.e. a read was requested so the data
2610 * must be written back to the guest memory).
2612 svm->sev_es.ghcb_sa_sync = sync;
2613 svm->sev_es.ghcb_sa_free = true;
2616 svm->sev_es.ghcb_sa = scratch_va;
2617 svm->sev_es.ghcb_sa_len = len;
2622 ghcb_set_sw_exit_info_1(ghcb, 2);
2623 ghcb_set_sw_exit_info_2(ghcb, GHCB_ERR_INVALID_SCRATCH_AREA);
2628 static void set_ghcb_msr_bits(struct vcpu_svm *svm, u64 value, u64 mask,
2631 svm->vmcb->control.ghcb_gpa &= ~(mask << pos);
2632 svm->vmcb->control.ghcb_gpa |= (value & mask) << pos;
2635 static u64 get_ghcb_msr_bits(struct vcpu_svm *svm, u64 mask, unsigned int pos)
2637 return (svm->vmcb->control.ghcb_gpa >> pos) & mask;
2640 static void set_ghcb_msr(struct vcpu_svm *svm, u64 value)
2642 svm->vmcb->control.ghcb_gpa = value;
2645 static int sev_handle_vmgexit_msr_protocol(struct vcpu_svm *svm)
2647 struct vmcb_control_area *control = &svm->vmcb->control;
2648 struct kvm_vcpu *vcpu = &svm->vcpu;
2652 ghcb_info = control->ghcb_gpa & GHCB_MSR_INFO_MASK;
2654 trace_kvm_vmgexit_msr_protocol_enter(svm->vcpu.vcpu_id,
2657 switch (ghcb_info) {
2658 case GHCB_MSR_SEV_INFO_REQ:
2659 set_ghcb_msr(svm, GHCB_MSR_SEV_INFO(GHCB_VERSION_MAX,
2663 case GHCB_MSR_CPUID_REQ: {
2664 u64 cpuid_fn, cpuid_reg, cpuid_value;
2666 cpuid_fn = get_ghcb_msr_bits(svm,
2667 GHCB_MSR_CPUID_FUNC_MASK,
2668 GHCB_MSR_CPUID_FUNC_POS);
2670 /* Initialize the registers needed by the CPUID intercept */
2671 vcpu->arch.regs[VCPU_REGS_RAX] = cpuid_fn;
2672 vcpu->arch.regs[VCPU_REGS_RCX] = 0;
2674 ret = svm_invoke_exit_handler(vcpu, SVM_EXIT_CPUID);
2676 /* Error, keep GHCB MSR value as-is */
2680 cpuid_reg = get_ghcb_msr_bits(svm,
2681 GHCB_MSR_CPUID_REG_MASK,
2682 GHCB_MSR_CPUID_REG_POS);
2684 cpuid_value = vcpu->arch.regs[VCPU_REGS_RAX];
2685 else if (cpuid_reg == 1)
2686 cpuid_value = vcpu->arch.regs[VCPU_REGS_RBX];
2687 else if (cpuid_reg == 2)
2688 cpuid_value = vcpu->arch.regs[VCPU_REGS_RCX];
2690 cpuid_value = vcpu->arch.regs[VCPU_REGS_RDX];
2692 set_ghcb_msr_bits(svm, cpuid_value,
2693 GHCB_MSR_CPUID_VALUE_MASK,
2694 GHCB_MSR_CPUID_VALUE_POS);
2696 set_ghcb_msr_bits(svm, GHCB_MSR_CPUID_RESP,
2701 case GHCB_MSR_TERM_REQ: {
2702 u64 reason_set, reason_code;
2704 reason_set = get_ghcb_msr_bits(svm,
2705 GHCB_MSR_TERM_REASON_SET_MASK,
2706 GHCB_MSR_TERM_REASON_SET_POS);
2707 reason_code = get_ghcb_msr_bits(svm,
2708 GHCB_MSR_TERM_REASON_MASK,
2709 GHCB_MSR_TERM_REASON_POS);
2710 pr_info("SEV-ES guest requested termination: %#llx:%#llx\n",
2711 reason_set, reason_code);
2717 /* Error, keep GHCB MSR value as-is */
2721 trace_kvm_vmgexit_msr_protocol_exit(svm->vcpu.vcpu_id,
2722 control->ghcb_gpa, ret);
2727 int sev_handle_vmgexit(struct kvm_vcpu *vcpu)
2729 struct vcpu_svm *svm = to_svm(vcpu);
2730 struct vmcb_control_area *control = &svm->vmcb->control;
2731 u64 ghcb_gpa, exit_code;
2735 /* Validate the GHCB */
2736 ghcb_gpa = control->ghcb_gpa;
2737 if (ghcb_gpa & GHCB_MSR_INFO_MASK)
2738 return sev_handle_vmgexit_msr_protocol(svm);
2741 vcpu_unimpl(vcpu, "vmgexit: GHCB gpa is not set\n");
2743 /* Without a GHCB, just return right back to the guest */
2747 if (kvm_vcpu_map(vcpu, ghcb_gpa >> PAGE_SHIFT, &svm->sev_es.ghcb_map)) {
2748 /* Unable to map GHCB from guest */
2749 vcpu_unimpl(vcpu, "vmgexit: error mapping GHCB [%#llx] from guest\n",
2752 /* Without a GHCB, just return right back to the guest */
2756 svm->sev_es.ghcb = svm->sev_es.ghcb_map.hva;
2757 ghcb = svm->sev_es.ghcb_map.hva;
2759 trace_kvm_vmgexit_enter(vcpu->vcpu_id, ghcb);
2761 exit_code = ghcb_get_sw_exit_code(ghcb);
2763 if (!sev_es_validate_vmgexit(svm))
2766 sev_es_sync_from_ghcb(svm);
2767 ghcb_set_sw_exit_info_1(ghcb, 0);
2768 ghcb_set_sw_exit_info_2(ghcb, 0);
2771 switch (exit_code) {
2772 case SVM_VMGEXIT_MMIO_READ:
2773 if (!setup_vmgexit_scratch(svm, true, control->exit_info_2))
2776 ret = kvm_sev_es_mmio_read(vcpu,
2777 control->exit_info_1,
2778 control->exit_info_2,
2779 svm->sev_es.ghcb_sa);
2781 case SVM_VMGEXIT_MMIO_WRITE:
2782 if (!setup_vmgexit_scratch(svm, false, control->exit_info_2))
2785 ret = kvm_sev_es_mmio_write(vcpu,
2786 control->exit_info_1,
2787 control->exit_info_2,
2788 svm->sev_es.ghcb_sa);
2790 case SVM_VMGEXIT_NMI_COMPLETE:
2791 ret = svm_invoke_exit_handler(vcpu, SVM_EXIT_IRET);
2793 case SVM_VMGEXIT_AP_HLT_LOOP:
2794 ret = kvm_emulate_ap_reset_hold(vcpu);
2796 case SVM_VMGEXIT_AP_JUMP_TABLE: {
2797 struct kvm_sev_info *sev = &to_kvm_svm(vcpu->kvm)->sev_info;
2799 switch (control->exit_info_1) {
2801 /* Set AP jump table address */
2802 sev->ap_jump_table = control->exit_info_2;
2805 /* Get AP jump table address */
2806 ghcb_set_sw_exit_info_2(ghcb, sev->ap_jump_table);
2809 pr_err("svm: vmgexit: unsupported AP jump table request - exit_info_1=%#llx\n",
2810 control->exit_info_1);
2811 ghcb_set_sw_exit_info_1(ghcb, 2);
2812 ghcb_set_sw_exit_info_2(ghcb, GHCB_ERR_INVALID_INPUT);
2817 case SVM_VMGEXIT_UNSUPPORTED_EVENT:
2819 "vmgexit: unsupported event - exit_info_1=%#llx, exit_info_2=%#llx\n",
2820 control->exit_info_1, control->exit_info_2);
2824 ret = svm_invoke_exit_handler(vcpu, exit_code);
2830 int sev_es_string_io(struct vcpu_svm *svm, int size, unsigned int port, int in)
2835 if (svm->vmcb->control.exit_info_2 > INT_MAX)
2838 count = svm->vmcb->control.exit_info_2;
2839 if (unlikely(check_mul_overflow(count, size, &bytes)))
2842 if (!setup_vmgexit_scratch(svm, in, bytes))
2845 return kvm_sev_es_string_io(&svm->vcpu, size, port, svm->sev_es.ghcb_sa,
2849 void sev_es_init_vmcb(struct vcpu_svm *svm)
2851 struct kvm_vcpu *vcpu = &svm->vcpu;
2853 svm->vmcb->control.nested_ctl |= SVM_NESTED_CTL_SEV_ES_ENABLE;
2854 svm->vmcb->control.virt_ext |= LBR_CTL_ENABLE_MASK;
2857 * An SEV-ES guest requires a VMSA area that is a separate from the
2858 * VMCB page. Do not include the encryption mask on the VMSA physical
2859 * address since hardware will access it using the guest key.
2861 svm->vmcb->control.vmsa_pa = __pa(svm->sev_es.vmsa);
2863 /* Can't intercept CR register access, HV can't modify CR registers */
2864 svm_clr_intercept(svm, INTERCEPT_CR0_READ);
2865 svm_clr_intercept(svm, INTERCEPT_CR4_READ);
2866 svm_clr_intercept(svm, INTERCEPT_CR8_READ);
2867 svm_clr_intercept(svm, INTERCEPT_CR0_WRITE);
2868 svm_clr_intercept(svm, INTERCEPT_CR4_WRITE);
2869 svm_clr_intercept(svm, INTERCEPT_CR8_WRITE);
2871 svm_clr_intercept(svm, INTERCEPT_SELECTIVE_CR0);
2873 /* Track EFER/CR register changes */
2874 svm_set_intercept(svm, TRAP_EFER_WRITE);
2875 svm_set_intercept(svm, TRAP_CR0_WRITE);
2876 svm_set_intercept(svm, TRAP_CR4_WRITE);
2877 svm_set_intercept(svm, TRAP_CR8_WRITE);
2879 /* No support for enable_vmware_backdoor */
2880 clr_exception_intercept(svm, GP_VECTOR);
2882 /* Can't intercept XSETBV, HV can't modify XCR0 directly */
2883 svm_clr_intercept(svm, INTERCEPT_XSETBV);
2885 /* Clear intercepts on selected MSRs */
2886 set_msr_interception(vcpu, svm->msrpm, MSR_EFER, 1, 1);
2887 set_msr_interception(vcpu, svm->msrpm, MSR_IA32_CR_PAT, 1, 1);
2888 set_msr_interception(vcpu, svm->msrpm, MSR_IA32_LASTBRANCHFROMIP, 1, 1);
2889 set_msr_interception(vcpu, svm->msrpm, MSR_IA32_LASTBRANCHTOIP, 1, 1);
2890 set_msr_interception(vcpu, svm->msrpm, MSR_IA32_LASTINTFROMIP, 1, 1);
2891 set_msr_interception(vcpu, svm->msrpm, MSR_IA32_LASTINTTOIP, 1, 1);
2894 void sev_es_vcpu_reset(struct vcpu_svm *svm)
2897 * Set the GHCB MSR value as per the GHCB specification when emulating
2898 * vCPU RESET for an SEV-ES guest.
2900 set_ghcb_msr(svm, GHCB_MSR_SEV_INFO(GHCB_VERSION_MAX,
2905 void sev_es_prepare_guest_switch(struct vcpu_svm *svm, unsigned int cpu)
2907 struct svm_cpu_data *sd = per_cpu(svm_data, cpu);
2908 struct vmcb_save_area *hostsa;
2911 * As an SEV-ES guest, hardware will restore the host state on VMEXIT,
2912 * of which one step is to perform a VMLOAD. Since hardware does not
2913 * perform a VMSAVE on VMRUN, the host savearea must be updated.
2915 vmsave(__sme_page_pa(sd->save_area));
2917 /* XCR0 is restored on VMEXIT, save the current host value */
2918 hostsa = (struct vmcb_save_area *)(page_address(sd->save_area) + 0x400);
2919 hostsa->xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
2921 /* PKRU is restored on VMEXIT, save the current host value */
2922 hostsa->pkru = read_pkru();
2924 /* MSR_IA32_XSS is restored on VMEXIT, save the currnet host value */
2925 hostsa->xss = host_xss;
2928 void sev_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
2930 struct vcpu_svm *svm = to_svm(vcpu);
2932 /* First SIPI: Use the values as initially set by the VMM */
2933 if (!svm->sev_es.received_first_sipi) {
2934 svm->sev_es.received_first_sipi = true;
2939 * Subsequent SIPI: Return from an AP Reset Hold VMGEXIT, where
2940 * the guest will set the CS and RIP. Set SW_EXIT_INFO_2 to a
2943 if (!svm->sev_es.ghcb)
2946 ghcb_set_sw_exit_info_2(svm->sev_es.ghcb, 1);