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>
20 #include <asm/fpu/internal.h>
22 #include <asm/trapnr.h>
30 #define __ex(x) __kvm_handle_fault_on_reboot(x)
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 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, pos, error = 0;
82 /* Check if there are any ASIDs to reclaim before performing a flush */
83 pos = find_next_bit(sev_reclaim_asid_bitmap, max_asid, 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, max_sev_asid);
123 static int sev_asid_new(struct kvm_sev_info *sev)
125 int pos, min_asid, max_asid, ret;
127 enum misc_res_type type;
129 type = sev->es_active ? MISC_CG_RES_SEV_ES : MISC_CG_RES_SEV;
130 WARN_ON(sev->misc_cg);
131 sev->misc_cg = get_current_misc_cg();
132 ret = misc_cg_try_charge(type, sev->misc_cg, 1);
134 put_misc_cg(sev->misc_cg);
139 mutex_lock(&sev_bitmap_lock);
142 * SEV-enabled guests must use asid from min_sev_asid to max_sev_asid.
143 * SEV-ES-enabled guest can use from 1 to min_sev_asid - 1.
145 min_asid = sev->es_active ? 0 : min_sev_asid - 1;
146 max_asid = sev->es_active ? min_sev_asid - 1 : max_sev_asid;
148 pos = find_next_zero_bit(sev_asid_bitmap, max_sev_asid, min_asid);
149 if (pos >= max_asid) {
150 if (retry && __sev_recycle_asids(min_asid, max_asid)) {
154 mutex_unlock(&sev_bitmap_lock);
159 __set_bit(pos, sev_asid_bitmap);
161 mutex_unlock(&sev_bitmap_lock);
165 misc_cg_uncharge(type, sev->misc_cg, 1);
166 put_misc_cg(sev->misc_cg);
171 static int sev_get_asid(struct kvm *kvm)
173 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
178 static void sev_asid_free(struct kvm_sev_info *sev)
180 struct svm_cpu_data *sd;
182 enum misc_res_type type;
184 mutex_lock(&sev_bitmap_lock);
187 __set_bit(pos, sev_reclaim_asid_bitmap);
189 for_each_possible_cpu(cpu) {
190 sd = per_cpu(svm_data, cpu);
191 sd->sev_vmcbs[pos] = NULL;
194 mutex_unlock(&sev_bitmap_lock);
196 type = sev->es_active ? MISC_CG_RES_SEV_ES : MISC_CG_RES_SEV;
197 misc_cg_uncharge(type, sev->misc_cg, 1);
198 put_misc_cg(sev->misc_cg);
202 static void sev_unbind_asid(struct kvm *kvm, unsigned int handle)
204 struct sev_data_decommission decommission;
205 struct sev_data_deactivate deactivate;
210 deactivate.handle = handle;
212 /* Guard DEACTIVATE against WBINVD/DF_FLUSH used in ASID recycling */
213 down_read(&sev_deactivate_lock);
214 sev_guest_deactivate(&deactivate, NULL);
215 up_read(&sev_deactivate_lock);
217 /* decommission handle */
218 decommission.handle = handle;
219 sev_guest_decommission(&decommission, NULL);
222 static int sev_guest_init(struct kvm *kvm, struct kvm_sev_cmd *argp)
224 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
225 bool es_active = argp->id == KVM_SEV_ES_INIT;
228 if (kvm->created_vcpus)
232 if (unlikely(sev->active))
235 sev->es_active = es_active;
236 asid = sev_asid_new(sev);
241 ret = sev_platform_init(&argp->error);
247 INIT_LIST_HEAD(&sev->regions_list);
255 sev->es_active = false;
259 static int sev_bind_asid(struct kvm *kvm, unsigned int handle, int *error)
261 struct sev_data_activate activate;
262 int asid = sev_get_asid(kvm);
265 /* activate ASID on the given handle */
266 activate.handle = handle;
267 activate.asid = asid;
268 ret = sev_guest_activate(&activate, error);
273 static int __sev_issue_cmd(int fd, int id, void *data, int *error)
282 ret = sev_issue_cmd_external_user(f.file, id, data, error);
288 static int sev_issue_cmd(struct kvm *kvm, int id, void *data, int *error)
290 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
292 return __sev_issue_cmd(sev->fd, id, data, error);
295 static int sev_launch_start(struct kvm *kvm, struct kvm_sev_cmd *argp)
297 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
298 struct sev_data_launch_start start;
299 struct kvm_sev_launch_start params;
300 void *dh_blob, *session_blob;
301 int *error = &argp->error;
307 if (copy_from_user(¶ms, (void __user *)(uintptr_t)argp->data, sizeof(params)))
310 memset(&start, 0, sizeof(start));
313 if (params.dh_uaddr) {
314 dh_blob = psp_copy_user_blob(params.dh_uaddr, params.dh_len);
316 return PTR_ERR(dh_blob);
318 start.dh_cert_address = __sme_set(__pa(dh_blob));
319 start.dh_cert_len = params.dh_len;
323 if (params.session_uaddr) {
324 session_blob = psp_copy_user_blob(params.session_uaddr, params.session_len);
325 if (IS_ERR(session_blob)) {
326 ret = PTR_ERR(session_blob);
330 start.session_address = __sme_set(__pa(session_blob));
331 start.session_len = params.session_len;
334 start.handle = params.handle;
335 start.policy = params.policy;
337 /* create memory encryption context */
338 ret = __sev_issue_cmd(argp->sev_fd, SEV_CMD_LAUNCH_START, &start, error);
342 /* Bind ASID to this guest */
343 ret = sev_bind_asid(kvm, start.handle, error);
347 /* return handle to userspace */
348 params.handle = start.handle;
349 if (copy_to_user((void __user *)(uintptr_t)argp->data, ¶ms, sizeof(params))) {
350 sev_unbind_asid(kvm, start.handle);
355 sev->handle = start.handle;
356 sev->fd = argp->sev_fd;
365 static struct page **sev_pin_memory(struct kvm *kvm, unsigned long uaddr,
366 unsigned long ulen, unsigned long *n,
369 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
370 unsigned long npages, size;
372 unsigned long locked, lock_limit;
374 unsigned long first, last;
377 lockdep_assert_held(&kvm->lock);
379 if (ulen == 0 || uaddr + ulen < uaddr)
380 return ERR_PTR(-EINVAL);
382 /* Calculate number of pages. */
383 first = (uaddr & PAGE_MASK) >> PAGE_SHIFT;
384 last = ((uaddr + ulen - 1) & PAGE_MASK) >> PAGE_SHIFT;
385 npages = (last - first + 1);
387 locked = sev->pages_locked + npages;
388 lock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
389 if (locked > lock_limit && !capable(CAP_IPC_LOCK)) {
390 pr_err("SEV: %lu locked pages exceed the lock limit of %lu.\n", locked, lock_limit);
391 return ERR_PTR(-ENOMEM);
394 if (WARN_ON_ONCE(npages > INT_MAX))
395 return ERR_PTR(-EINVAL);
397 /* Avoid using vmalloc for smaller buffers. */
398 size = npages * sizeof(struct page *);
399 if (size > PAGE_SIZE)
400 pages = __vmalloc(size, GFP_KERNEL_ACCOUNT | __GFP_ZERO);
402 pages = kmalloc(size, GFP_KERNEL_ACCOUNT);
405 return ERR_PTR(-ENOMEM);
407 /* Pin the user virtual address. */
408 npinned = pin_user_pages_fast(uaddr, npages, write ? FOLL_WRITE : 0, pages);
409 if (npinned != npages) {
410 pr_err("SEV: Failure locking %lu pages.\n", npages);
416 sev->pages_locked = locked;
422 unpin_user_pages(pages, npinned);
428 static void sev_unpin_memory(struct kvm *kvm, struct page **pages,
429 unsigned long npages)
431 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
433 unpin_user_pages(pages, npages);
435 sev->pages_locked -= npages;
438 static void sev_clflush_pages(struct page *pages[], unsigned long npages)
440 uint8_t *page_virtual;
443 if (this_cpu_has(X86_FEATURE_SME_COHERENT) || npages == 0 ||
447 for (i = 0; i < npages; i++) {
448 page_virtual = kmap_atomic(pages[i]);
449 clflush_cache_range(page_virtual, PAGE_SIZE);
450 kunmap_atomic(page_virtual);
454 static unsigned long get_num_contig_pages(unsigned long idx,
455 struct page **inpages, unsigned long npages)
457 unsigned long paddr, next_paddr;
458 unsigned long i = idx + 1, pages = 1;
460 /* find the number of contiguous pages starting from idx */
461 paddr = __sme_page_pa(inpages[idx]);
463 next_paddr = __sme_page_pa(inpages[i++]);
464 if ((paddr + PAGE_SIZE) == next_paddr) {
475 static int sev_launch_update_data(struct kvm *kvm, struct kvm_sev_cmd *argp)
477 unsigned long vaddr, vaddr_end, next_vaddr, npages, pages, size, i;
478 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
479 struct kvm_sev_launch_update_data params;
480 struct sev_data_launch_update_data data;
481 struct page **inpages;
487 if (copy_from_user(¶ms, (void __user *)(uintptr_t)argp->data, sizeof(params)))
490 vaddr = params.uaddr;
492 vaddr_end = vaddr + size;
494 /* Lock the user memory. */
495 inpages = sev_pin_memory(kvm, vaddr, size, &npages, 1);
497 return PTR_ERR(inpages);
500 * Flush (on non-coherent CPUs) before LAUNCH_UPDATE encrypts pages in
501 * place; the cache may contain the data that was written unencrypted.
503 sev_clflush_pages(inpages, npages);
506 data.handle = sev->handle;
508 for (i = 0; vaddr < vaddr_end; vaddr = next_vaddr, i += pages) {
512 * If the user buffer is not page-aligned, calculate the offset
515 offset = vaddr & (PAGE_SIZE - 1);
517 /* Calculate the number of pages that can be encrypted in one go. */
518 pages = get_num_contig_pages(i, inpages, npages);
520 len = min_t(size_t, ((pages * PAGE_SIZE) - offset), size);
523 data.address = __sme_page_pa(inpages[i]) + offset;
524 ret = sev_issue_cmd(kvm, SEV_CMD_LAUNCH_UPDATE_DATA, &data, &argp->error);
529 next_vaddr = vaddr + len;
533 /* content of memory is updated, mark pages dirty */
534 for (i = 0; i < npages; i++) {
535 set_page_dirty_lock(inpages[i]);
536 mark_page_accessed(inpages[i]);
538 /* unlock the user pages */
539 sev_unpin_memory(kvm, inpages, npages);
543 static int sev_es_sync_vmsa(struct vcpu_svm *svm)
545 struct vmcb_save_area *save = &svm->vmcb->save;
547 /* Check some debug related fields before encrypting the VMSA */
548 if (svm->vcpu.guest_debug || (save->dr7 & ~DR7_FIXED_1))
551 /* Sync registgers */
552 save->rax = svm->vcpu.arch.regs[VCPU_REGS_RAX];
553 save->rbx = svm->vcpu.arch.regs[VCPU_REGS_RBX];
554 save->rcx = svm->vcpu.arch.regs[VCPU_REGS_RCX];
555 save->rdx = svm->vcpu.arch.regs[VCPU_REGS_RDX];
556 save->rsp = svm->vcpu.arch.regs[VCPU_REGS_RSP];
557 save->rbp = svm->vcpu.arch.regs[VCPU_REGS_RBP];
558 save->rsi = svm->vcpu.arch.regs[VCPU_REGS_RSI];
559 save->rdi = svm->vcpu.arch.regs[VCPU_REGS_RDI];
561 save->r8 = svm->vcpu.arch.regs[VCPU_REGS_R8];
562 save->r9 = svm->vcpu.arch.regs[VCPU_REGS_R9];
563 save->r10 = svm->vcpu.arch.regs[VCPU_REGS_R10];
564 save->r11 = svm->vcpu.arch.regs[VCPU_REGS_R11];
565 save->r12 = svm->vcpu.arch.regs[VCPU_REGS_R12];
566 save->r13 = svm->vcpu.arch.regs[VCPU_REGS_R13];
567 save->r14 = svm->vcpu.arch.regs[VCPU_REGS_R14];
568 save->r15 = svm->vcpu.arch.regs[VCPU_REGS_R15];
570 save->rip = svm->vcpu.arch.regs[VCPU_REGS_RIP];
572 /* Sync some non-GPR registers before encrypting */
573 save->xcr0 = svm->vcpu.arch.xcr0;
574 save->pkru = svm->vcpu.arch.pkru;
575 save->xss = svm->vcpu.arch.ia32_xss;
578 * SEV-ES will use a VMSA that is pointed to by the VMCB, not
579 * the traditional VMSA that is part of the VMCB. Copy the
580 * traditional VMSA as it has been built so far (in prep
581 * for LAUNCH_UPDATE_VMSA) to be the initial SEV-ES state.
583 memcpy(svm->vmsa, save, sizeof(*save));
588 static int sev_launch_update_vmsa(struct kvm *kvm, struct kvm_sev_cmd *argp)
590 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
591 struct sev_data_launch_update_vmsa vmsa;
592 struct kvm_vcpu *vcpu;
595 if (!sev_es_guest(kvm))
600 kvm_for_each_vcpu(i, vcpu, kvm) {
601 struct vcpu_svm *svm = to_svm(vcpu);
603 /* Perform some pre-encryption checks against the VMSA */
604 ret = sev_es_sync_vmsa(svm);
609 * The LAUNCH_UPDATE_VMSA command will perform in-place
610 * encryption of the VMSA memory content (i.e it will write
611 * the same memory region with the guest's key), so invalidate
614 clflush_cache_range(svm->vmsa, PAGE_SIZE);
616 vmsa.handle = sev->handle;
617 vmsa.address = __sme_pa(svm->vmsa);
618 vmsa.len = PAGE_SIZE;
619 ret = sev_issue_cmd(kvm, SEV_CMD_LAUNCH_UPDATE_VMSA, &vmsa,
624 svm->vcpu.arch.guest_state_protected = true;
630 static int sev_launch_measure(struct kvm *kvm, struct kvm_sev_cmd *argp)
632 void __user *measure = (void __user *)(uintptr_t)argp->data;
633 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
634 struct sev_data_launch_measure data;
635 struct kvm_sev_launch_measure params;
636 void __user *p = NULL;
643 if (copy_from_user(¶ms, measure, sizeof(params)))
646 memset(&data, 0, sizeof(data));
648 /* User wants to query the blob length */
652 p = (void __user *)(uintptr_t)params.uaddr;
654 if (params.len > SEV_FW_BLOB_MAX_SIZE)
657 blob = kmalloc(params.len, GFP_KERNEL_ACCOUNT);
661 data.address = __psp_pa(blob);
662 data.len = params.len;
666 data.handle = sev->handle;
667 ret = sev_issue_cmd(kvm, SEV_CMD_LAUNCH_MEASURE, &data, &argp->error);
670 * If we query the session length, FW responded with expected data.
679 if (copy_to_user(p, blob, params.len))
684 params.len = data.len;
685 if (copy_to_user(measure, ¶ms, sizeof(params)))
692 static int sev_launch_finish(struct kvm *kvm, struct kvm_sev_cmd *argp)
694 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
695 struct sev_data_launch_finish data;
700 data.handle = sev->handle;
701 return sev_issue_cmd(kvm, SEV_CMD_LAUNCH_FINISH, &data, &argp->error);
704 static int sev_guest_status(struct kvm *kvm, struct kvm_sev_cmd *argp)
706 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
707 struct kvm_sev_guest_status params;
708 struct sev_data_guest_status data;
714 memset(&data, 0, sizeof(data));
716 data.handle = sev->handle;
717 ret = sev_issue_cmd(kvm, SEV_CMD_GUEST_STATUS, &data, &argp->error);
721 params.policy = data.policy;
722 params.state = data.state;
723 params.handle = data.handle;
725 if (copy_to_user((void __user *)(uintptr_t)argp->data, ¶ms, sizeof(params)))
731 static int __sev_issue_dbg_cmd(struct kvm *kvm, unsigned long src,
732 unsigned long dst, int size,
733 int *error, bool enc)
735 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
736 struct sev_data_dbg data;
739 data.handle = sev->handle;
744 return sev_issue_cmd(kvm,
745 enc ? SEV_CMD_DBG_ENCRYPT : SEV_CMD_DBG_DECRYPT,
749 static int __sev_dbg_decrypt(struct kvm *kvm, unsigned long src_paddr,
750 unsigned long dst_paddr, int sz, int *err)
755 * Its safe to read more than we are asked, caller should ensure that
756 * destination has enough space.
758 offset = src_paddr & 15;
759 src_paddr = round_down(src_paddr, 16);
760 sz = round_up(sz + offset, 16);
762 return __sev_issue_dbg_cmd(kvm, src_paddr, dst_paddr, sz, err, false);
765 static int __sev_dbg_decrypt_user(struct kvm *kvm, unsigned long paddr,
766 void __user *dst_uaddr,
767 unsigned long dst_paddr,
770 struct page *tpage = NULL;
773 /* if inputs are not 16-byte then use intermediate buffer */
774 if (!IS_ALIGNED(dst_paddr, 16) ||
775 !IS_ALIGNED(paddr, 16) ||
776 !IS_ALIGNED(size, 16)) {
777 tpage = (void *)alloc_page(GFP_KERNEL);
781 dst_paddr = __sme_page_pa(tpage);
784 ret = __sev_dbg_decrypt(kvm, paddr, dst_paddr, size, err);
790 if (copy_to_user(dst_uaddr, page_address(tpage) + offset, size))
801 static int __sev_dbg_encrypt_user(struct kvm *kvm, unsigned long paddr,
803 unsigned long dst_paddr,
804 void __user *dst_vaddr,
805 int size, int *error)
807 struct page *src_tpage = NULL;
808 struct page *dst_tpage = NULL;
811 /* If source buffer is not aligned then use an intermediate buffer */
812 if (!IS_ALIGNED((unsigned long)vaddr, 16)) {
813 src_tpage = alloc_page(GFP_KERNEL);
817 if (copy_from_user(page_address(src_tpage), vaddr, size)) {
818 __free_page(src_tpage);
822 paddr = __sme_page_pa(src_tpage);
826 * If destination buffer or length is not aligned then do read-modify-write:
827 * - decrypt destination in an intermediate buffer
828 * - copy the source buffer in an intermediate buffer
829 * - use the intermediate buffer as source buffer
831 if (!IS_ALIGNED((unsigned long)dst_vaddr, 16) || !IS_ALIGNED(size, 16)) {
834 dst_tpage = alloc_page(GFP_KERNEL);
840 ret = __sev_dbg_decrypt(kvm, dst_paddr,
841 __sme_page_pa(dst_tpage), size, error);
846 * If source is kernel buffer then use memcpy() otherwise
849 dst_offset = dst_paddr & 15;
852 memcpy(page_address(dst_tpage) + dst_offset,
853 page_address(src_tpage), size);
855 if (copy_from_user(page_address(dst_tpage) + dst_offset,
862 paddr = __sme_page_pa(dst_tpage);
863 dst_paddr = round_down(dst_paddr, 16);
864 len = round_up(size, 16);
867 ret = __sev_issue_dbg_cmd(kvm, paddr, dst_paddr, len, error, true);
871 __free_page(src_tpage);
873 __free_page(dst_tpage);
877 static int sev_dbg_crypt(struct kvm *kvm, struct kvm_sev_cmd *argp, bool dec)
879 unsigned long vaddr, vaddr_end, next_vaddr;
880 unsigned long dst_vaddr;
881 struct page **src_p, **dst_p;
882 struct kvm_sev_dbg debug;
890 if (copy_from_user(&debug, (void __user *)(uintptr_t)argp->data, sizeof(debug)))
893 if (!debug.len || debug.src_uaddr + debug.len < debug.src_uaddr)
895 if (!debug.dst_uaddr)
898 vaddr = debug.src_uaddr;
900 vaddr_end = vaddr + size;
901 dst_vaddr = debug.dst_uaddr;
903 for (; vaddr < vaddr_end; vaddr = next_vaddr) {
904 int len, s_off, d_off;
906 /* lock userspace source and destination page */
907 src_p = sev_pin_memory(kvm, vaddr & PAGE_MASK, PAGE_SIZE, &n, 0);
909 return PTR_ERR(src_p);
911 dst_p = sev_pin_memory(kvm, dst_vaddr & PAGE_MASK, PAGE_SIZE, &n, 1);
913 sev_unpin_memory(kvm, src_p, n);
914 return PTR_ERR(dst_p);
918 * Flush (on non-coherent CPUs) before DBG_{DE,EN}CRYPT read or modify
919 * the pages; flush the destination too so that future accesses do not
922 sev_clflush_pages(src_p, 1);
923 sev_clflush_pages(dst_p, 1);
926 * Since user buffer may not be page aligned, calculate the
927 * offset within the page.
929 s_off = vaddr & ~PAGE_MASK;
930 d_off = dst_vaddr & ~PAGE_MASK;
931 len = min_t(size_t, (PAGE_SIZE - s_off), size);
934 ret = __sev_dbg_decrypt_user(kvm,
935 __sme_page_pa(src_p[0]) + s_off,
936 (void __user *)dst_vaddr,
937 __sme_page_pa(dst_p[0]) + d_off,
940 ret = __sev_dbg_encrypt_user(kvm,
941 __sme_page_pa(src_p[0]) + s_off,
942 (void __user *)vaddr,
943 __sme_page_pa(dst_p[0]) + d_off,
944 (void __user *)dst_vaddr,
947 sev_unpin_memory(kvm, src_p, n);
948 sev_unpin_memory(kvm, dst_p, n);
953 next_vaddr = vaddr + len;
954 dst_vaddr = dst_vaddr + len;
961 static int sev_launch_secret(struct kvm *kvm, struct kvm_sev_cmd *argp)
963 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
964 struct sev_data_launch_secret data;
965 struct kvm_sev_launch_secret params;
974 if (copy_from_user(¶ms, (void __user *)(uintptr_t)argp->data, sizeof(params)))
977 pages = sev_pin_memory(kvm, params.guest_uaddr, params.guest_len, &n, 1);
979 return PTR_ERR(pages);
982 * Flush (on non-coherent CPUs) before LAUNCH_SECRET encrypts pages in
983 * place; the cache may contain the data that was written unencrypted.
985 sev_clflush_pages(pages, n);
988 * The secret must be copied into contiguous memory region, lets verify
989 * that userspace memory pages are contiguous before we issue command.
991 if (get_num_contig_pages(0, pages, n) != n) {
996 memset(&data, 0, sizeof(data));
998 offset = params.guest_uaddr & (PAGE_SIZE - 1);
999 data.guest_address = __sme_page_pa(pages[0]) + offset;
1000 data.guest_len = params.guest_len;
1002 blob = psp_copy_user_blob(params.trans_uaddr, params.trans_len);
1004 ret = PTR_ERR(blob);
1005 goto e_unpin_memory;
1008 data.trans_address = __psp_pa(blob);
1009 data.trans_len = params.trans_len;
1011 hdr = psp_copy_user_blob(params.hdr_uaddr, params.hdr_len);
1016 data.hdr_address = __psp_pa(hdr);
1017 data.hdr_len = params.hdr_len;
1019 data.handle = sev->handle;
1020 ret = sev_issue_cmd(kvm, SEV_CMD_LAUNCH_UPDATE_SECRET, &data, &argp->error);
1027 /* content of memory is updated, mark pages dirty */
1028 for (i = 0; i < n; i++) {
1029 set_page_dirty_lock(pages[i]);
1030 mark_page_accessed(pages[i]);
1032 sev_unpin_memory(kvm, pages, n);
1036 static int sev_get_attestation_report(struct kvm *kvm, struct kvm_sev_cmd *argp)
1038 void __user *report = (void __user *)(uintptr_t)argp->data;
1039 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1040 struct sev_data_attestation_report data;
1041 struct kvm_sev_attestation_report params;
1046 if (!sev_guest(kvm))
1049 if (copy_from_user(¶ms, (void __user *)(uintptr_t)argp->data, sizeof(params)))
1052 memset(&data, 0, sizeof(data));
1054 /* User wants to query the blob length */
1058 p = (void __user *)(uintptr_t)params.uaddr;
1060 if (params.len > SEV_FW_BLOB_MAX_SIZE)
1063 blob = kmalloc(params.len, GFP_KERNEL_ACCOUNT);
1067 data.address = __psp_pa(blob);
1068 data.len = params.len;
1069 memcpy(data.mnonce, params.mnonce, sizeof(params.mnonce));
1072 data.handle = sev->handle;
1073 ret = sev_issue_cmd(kvm, SEV_CMD_ATTESTATION_REPORT, &data, &argp->error);
1075 * If we query the session length, FW responded with expected data.
1084 if (copy_to_user(p, blob, params.len))
1089 params.len = data.len;
1090 if (copy_to_user(report, ¶ms, sizeof(params)))
1097 /* Userspace wants to query session length. */
1099 __sev_send_start_query_session_length(struct kvm *kvm, struct kvm_sev_cmd *argp,
1100 struct kvm_sev_send_start *params)
1102 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1103 struct sev_data_send_start data;
1106 data.handle = sev->handle;
1107 ret = sev_issue_cmd(kvm, SEV_CMD_SEND_START, &data, &argp->error);
1111 params->session_len = data.session_len;
1112 if (copy_to_user((void __user *)(uintptr_t)argp->data, params,
1113 sizeof(struct kvm_sev_send_start)))
1119 static int sev_send_start(struct kvm *kvm, struct kvm_sev_cmd *argp)
1121 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1122 struct sev_data_send_start data;
1123 struct kvm_sev_send_start params;
1124 void *amd_certs, *session_data;
1125 void *pdh_cert, *plat_certs;
1128 if (!sev_guest(kvm))
1131 if (copy_from_user(¶ms, (void __user *)(uintptr_t)argp->data,
1132 sizeof(struct kvm_sev_send_start)))
1135 /* if session_len is zero, userspace wants to query the session length */
1136 if (!params.session_len)
1137 return __sev_send_start_query_session_length(kvm, argp,
1140 /* some sanity checks */
1141 if (!params.pdh_cert_uaddr || !params.pdh_cert_len ||
1142 !params.session_uaddr || params.session_len > SEV_FW_BLOB_MAX_SIZE)
1145 /* allocate the memory to hold the session data blob */
1146 session_data = kmalloc(params.session_len, GFP_KERNEL_ACCOUNT);
1150 /* copy the certificate blobs from userspace */
1151 pdh_cert = psp_copy_user_blob(params.pdh_cert_uaddr,
1152 params.pdh_cert_len);
1153 if (IS_ERR(pdh_cert)) {
1154 ret = PTR_ERR(pdh_cert);
1155 goto e_free_session;
1158 plat_certs = psp_copy_user_blob(params.plat_certs_uaddr,
1159 params.plat_certs_len);
1160 if (IS_ERR(plat_certs)) {
1161 ret = PTR_ERR(plat_certs);
1165 amd_certs = psp_copy_user_blob(params.amd_certs_uaddr,
1166 params.amd_certs_len);
1167 if (IS_ERR(amd_certs)) {
1168 ret = PTR_ERR(amd_certs);
1169 goto e_free_plat_cert;
1172 /* populate the FW SEND_START field with system physical address */
1173 memset(&data, 0, sizeof(data));
1174 data.pdh_cert_address = __psp_pa(pdh_cert);
1175 data.pdh_cert_len = params.pdh_cert_len;
1176 data.plat_certs_address = __psp_pa(plat_certs);
1177 data.plat_certs_len = params.plat_certs_len;
1178 data.amd_certs_address = __psp_pa(amd_certs);
1179 data.amd_certs_len = params.amd_certs_len;
1180 data.session_address = __psp_pa(session_data);
1181 data.session_len = params.session_len;
1182 data.handle = sev->handle;
1184 ret = sev_issue_cmd(kvm, SEV_CMD_SEND_START, &data, &argp->error);
1186 if (!ret && copy_to_user((void __user *)(uintptr_t)params.session_uaddr,
1187 session_data, params.session_len)) {
1189 goto e_free_amd_cert;
1192 params.policy = data.policy;
1193 params.session_len = data.session_len;
1194 if (copy_to_user((void __user *)(uintptr_t)argp->data, ¶ms,
1195 sizeof(struct kvm_sev_send_start)))
1205 kfree(session_data);
1209 /* Userspace wants to query either header or trans length. */
1211 __sev_send_update_data_query_lengths(struct kvm *kvm, struct kvm_sev_cmd *argp,
1212 struct kvm_sev_send_update_data *params)
1214 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1215 struct sev_data_send_update_data data;
1218 data.handle = sev->handle;
1219 ret = sev_issue_cmd(kvm, SEV_CMD_SEND_UPDATE_DATA, &data, &argp->error);
1223 params->hdr_len = data.hdr_len;
1224 params->trans_len = data.trans_len;
1226 if (copy_to_user((void __user *)(uintptr_t)argp->data, params,
1227 sizeof(struct kvm_sev_send_update_data)))
1233 static int sev_send_update_data(struct kvm *kvm, struct kvm_sev_cmd *argp)
1235 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1236 struct sev_data_send_update_data data;
1237 struct kvm_sev_send_update_data params;
1238 void *hdr, *trans_data;
1239 struct page **guest_page;
1243 if (!sev_guest(kvm))
1246 if (copy_from_user(¶ms, (void __user *)(uintptr_t)argp->data,
1247 sizeof(struct kvm_sev_send_update_data)))
1250 /* userspace wants to query either header or trans length */
1251 if (!params.trans_len || !params.hdr_len)
1252 return __sev_send_update_data_query_lengths(kvm, argp, ¶ms);
1254 if (!params.trans_uaddr || !params.guest_uaddr ||
1255 !params.guest_len || !params.hdr_uaddr)
1258 /* Check if we are crossing the page boundary */
1259 offset = params.guest_uaddr & (PAGE_SIZE - 1);
1260 if ((params.guest_len + offset > PAGE_SIZE))
1263 /* Pin guest memory */
1264 guest_page = sev_pin_memory(kvm, params.guest_uaddr & PAGE_MASK,
1269 /* allocate memory for header and transport buffer */
1271 hdr = kmalloc(params.hdr_len, GFP_KERNEL_ACCOUNT);
1275 trans_data = kmalloc(params.trans_len, GFP_KERNEL_ACCOUNT);
1279 memset(&data, 0, sizeof(data));
1280 data.hdr_address = __psp_pa(hdr);
1281 data.hdr_len = params.hdr_len;
1282 data.trans_address = __psp_pa(trans_data);
1283 data.trans_len = params.trans_len;
1285 /* The SEND_UPDATE_DATA command requires C-bit to be always set. */
1286 data.guest_address = (page_to_pfn(guest_page[0]) << PAGE_SHIFT) + offset;
1287 data.guest_address |= sev_me_mask;
1288 data.guest_len = params.guest_len;
1289 data.handle = sev->handle;
1291 ret = sev_issue_cmd(kvm, SEV_CMD_SEND_UPDATE_DATA, &data, &argp->error);
1294 goto e_free_trans_data;
1296 /* copy transport buffer to user space */
1297 if (copy_to_user((void __user *)(uintptr_t)params.trans_uaddr,
1298 trans_data, params.trans_len)) {
1300 goto e_free_trans_data;
1303 /* Copy packet header to userspace. */
1304 ret = copy_to_user((void __user *)(uintptr_t)params.hdr_uaddr, hdr,
1312 sev_unpin_memory(kvm, guest_page, n);
1317 static int sev_send_finish(struct kvm *kvm, struct kvm_sev_cmd *argp)
1319 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1320 struct sev_data_send_finish data;
1322 if (!sev_guest(kvm))
1325 data.handle = sev->handle;
1326 return sev_issue_cmd(kvm, SEV_CMD_SEND_FINISH, &data, &argp->error);
1329 static int sev_send_cancel(struct kvm *kvm, struct kvm_sev_cmd *argp)
1331 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1332 struct sev_data_send_cancel data;
1334 if (!sev_guest(kvm))
1337 data.handle = sev->handle;
1338 return sev_issue_cmd(kvm, SEV_CMD_SEND_CANCEL, &data, &argp->error);
1341 static int sev_receive_start(struct kvm *kvm, struct kvm_sev_cmd *argp)
1343 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1344 struct sev_data_receive_start start;
1345 struct kvm_sev_receive_start params;
1346 int *error = &argp->error;
1351 if (!sev_guest(kvm))
1354 /* Get parameter from the userspace */
1355 if (copy_from_user(¶ms, (void __user *)(uintptr_t)argp->data,
1356 sizeof(struct kvm_sev_receive_start)))
1359 /* some sanity checks */
1360 if (!params.pdh_uaddr || !params.pdh_len ||
1361 !params.session_uaddr || !params.session_len)
1364 pdh_data = psp_copy_user_blob(params.pdh_uaddr, params.pdh_len);
1365 if (IS_ERR(pdh_data))
1366 return PTR_ERR(pdh_data);
1368 session_data = psp_copy_user_blob(params.session_uaddr,
1369 params.session_len);
1370 if (IS_ERR(session_data)) {
1371 ret = PTR_ERR(session_data);
1375 memset(&start, 0, sizeof(start));
1376 start.handle = params.handle;
1377 start.policy = params.policy;
1378 start.pdh_cert_address = __psp_pa(pdh_data);
1379 start.pdh_cert_len = params.pdh_len;
1380 start.session_address = __psp_pa(session_data);
1381 start.session_len = params.session_len;
1383 /* create memory encryption context */
1384 ret = __sev_issue_cmd(argp->sev_fd, SEV_CMD_RECEIVE_START, &start,
1387 goto e_free_session;
1389 /* Bind ASID to this guest */
1390 ret = sev_bind_asid(kvm, start.handle, error);
1392 goto e_free_session;
1394 params.handle = start.handle;
1395 if (copy_to_user((void __user *)(uintptr_t)argp->data,
1396 ¶ms, sizeof(struct kvm_sev_receive_start))) {
1398 sev_unbind_asid(kvm, start.handle);
1399 goto e_free_session;
1402 sev->handle = start.handle;
1403 sev->fd = argp->sev_fd;
1406 kfree(session_data);
1413 static int sev_receive_update_data(struct kvm *kvm, struct kvm_sev_cmd *argp)
1415 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1416 struct kvm_sev_receive_update_data params;
1417 struct sev_data_receive_update_data data;
1418 void *hdr = NULL, *trans = NULL;
1419 struct page **guest_page;
1423 if (!sev_guest(kvm))
1426 if (copy_from_user(¶ms, (void __user *)(uintptr_t)argp->data,
1427 sizeof(struct kvm_sev_receive_update_data)))
1430 if (!params.hdr_uaddr || !params.hdr_len ||
1431 !params.guest_uaddr || !params.guest_len ||
1432 !params.trans_uaddr || !params.trans_len)
1435 /* Check if we are crossing the page boundary */
1436 offset = params.guest_uaddr & (PAGE_SIZE - 1);
1437 if ((params.guest_len + offset > PAGE_SIZE))
1440 hdr = psp_copy_user_blob(params.hdr_uaddr, params.hdr_len);
1442 return PTR_ERR(hdr);
1444 trans = psp_copy_user_blob(params.trans_uaddr, params.trans_len);
1445 if (IS_ERR(trans)) {
1446 ret = PTR_ERR(trans);
1450 memset(&data, 0, sizeof(data));
1451 data.hdr_address = __psp_pa(hdr);
1452 data.hdr_len = params.hdr_len;
1453 data.trans_address = __psp_pa(trans);
1454 data.trans_len = params.trans_len;
1456 /* Pin guest memory */
1458 guest_page = sev_pin_memory(kvm, params.guest_uaddr & PAGE_MASK,
1463 /* The RECEIVE_UPDATE_DATA command requires C-bit to be always set. */
1464 data.guest_address = (page_to_pfn(guest_page[0]) << PAGE_SHIFT) + offset;
1465 data.guest_address |= sev_me_mask;
1466 data.guest_len = params.guest_len;
1467 data.handle = sev->handle;
1469 ret = sev_issue_cmd(kvm, SEV_CMD_RECEIVE_UPDATE_DATA, &data,
1472 sev_unpin_memory(kvm, guest_page, n);
1482 static int sev_receive_finish(struct kvm *kvm, struct kvm_sev_cmd *argp)
1484 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1485 struct sev_data_receive_finish data;
1487 if (!sev_guest(kvm))
1490 data.handle = sev->handle;
1491 return sev_issue_cmd(kvm, SEV_CMD_RECEIVE_FINISH, &data, &argp->error);
1494 int svm_mem_enc_op(struct kvm *kvm, void __user *argp)
1496 struct kvm_sev_cmd sev_cmd;
1505 if (copy_from_user(&sev_cmd, argp, sizeof(struct kvm_sev_cmd)))
1508 mutex_lock(&kvm->lock);
1510 /* enc_context_owner handles all memory enc operations */
1511 if (is_mirroring_enc_context(kvm)) {
1516 switch (sev_cmd.id) {
1517 case KVM_SEV_ES_INIT:
1518 if (!sev_es_enabled) {
1524 r = sev_guest_init(kvm, &sev_cmd);
1526 case KVM_SEV_LAUNCH_START:
1527 r = sev_launch_start(kvm, &sev_cmd);
1529 case KVM_SEV_LAUNCH_UPDATE_DATA:
1530 r = sev_launch_update_data(kvm, &sev_cmd);
1532 case KVM_SEV_LAUNCH_UPDATE_VMSA:
1533 r = sev_launch_update_vmsa(kvm, &sev_cmd);
1535 case KVM_SEV_LAUNCH_MEASURE:
1536 r = sev_launch_measure(kvm, &sev_cmd);
1538 case KVM_SEV_LAUNCH_FINISH:
1539 r = sev_launch_finish(kvm, &sev_cmd);
1541 case KVM_SEV_GUEST_STATUS:
1542 r = sev_guest_status(kvm, &sev_cmd);
1544 case KVM_SEV_DBG_DECRYPT:
1545 r = sev_dbg_crypt(kvm, &sev_cmd, true);
1547 case KVM_SEV_DBG_ENCRYPT:
1548 r = sev_dbg_crypt(kvm, &sev_cmd, false);
1550 case KVM_SEV_LAUNCH_SECRET:
1551 r = sev_launch_secret(kvm, &sev_cmd);
1553 case KVM_SEV_GET_ATTESTATION_REPORT:
1554 r = sev_get_attestation_report(kvm, &sev_cmd);
1556 case KVM_SEV_SEND_START:
1557 r = sev_send_start(kvm, &sev_cmd);
1559 case KVM_SEV_SEND_UPDATE_DATA:
1560 r = sev_send_update_data(kvm, &sev_cmd);
1562 case KVM_SEV_SEND_FINISH:
1563 r = sev_send_finish(kvm, &sev_cmd);
1565 case KVM_SEV_SEND_CANCEL:
1566 r = sev_send_cancel(kvm, &sev_cmd);
1568 case KVM_SEV_RECEIVE_START:
1569 r = sev_receive_start(kvm, &sev_cmd);
1571 case KVM_SEV_RECEIVE_UPDATE_DATA:
1572 r = sev_receive_update_data(kvm, &sev_cmd);
1574 case KVM_SEV_RECEIVE_FINISH:
1575 r = sev_receive_finish(kvm, &sev_cmd);
1582 if (copy_to_user(argp, &sev_cmd, sizeof(struct kvm_sev_cmd)))
1586 mutex_unlock(&kvm->lock);
1590 int svm_register_enc_region(struct kvm *kvm,
1591 struct kvm_enc_region *range)
1593 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1594 struct enc_region *region;
1597 if (!sev_guest(kvm))
1600 /* If kvm is mirroring encryption context it isn't responsible for it */
1601 if (is_mirroring_enc_context(kvm))
1604 if (range->addr > ULONG_MAX || range->size > ULONG_MAX)
1607 region = kzalloc(sizeof(*region), GFP_KERNEL_ACCOUNT);
1611 mutex_lock(&kvm->lock);
1612 region->pages = sev_pin_memory(kvm, range->addr, range->size, ®ion->npages, 1);
1613 if (IS_ERR(region->pages)) {
1614 ret = PTR_ERR(region->pages);
1615 mutex_unlock(&kvm->lock);
1619 region->uaddr = range->addr;
1620 region->size = range->size;
1622 list_add_tail(®ion->list, &sev->regions_list);
1623 mutex_unlock(&kvm->lock);
1626 * The guest may change the memory encryption attribute from C=0 -> C=1
1627 * or vice versa for this memory range. Lets make sure caches are
1628 * flushed to ensure that guest data gets written into memory with
1631 sev_clflush_pages(region->pages, region->npages);
1640 static struct enc_region *
1641 find_enc_region(struct kvm *kvm, struct kvm_enc_region *range)
1643 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1644 struct list_head *head = &sev->regions_list;
1645 struct enc_region *i;
1647 list_for_each_entry(i, head, list) {
1648 if (i->uaddr == range->addr &&
1649 i->size == range->size)
1656 static void __unregister_enc_region_locked(struct kvm *kvm,
1657 struct enc_region *region)
1659 sev_unpin_memory(kvm, region->pages, region->npages);
1660 list_del(®ion->list);
1664 int svm_unregister_enc_region(struct kvm *kvm,
1665 struct kvm_enc_region *range)
1667 struct enc_region *region;
1670 /* If kvm is mirroring encryption context it isn't responsible for it */
1671 if (is_mirroring_enc_context(kvm))
1674 mutex_lock(&kvm->lock);
1676 if (!sev_guest(kvm)) {
1681 region = find_enc_region(kvm, range);
1688 * Ensure that all guest tagged cache entries are flushed before
1689 * releasing the pages back to the system for use. CLFLUSH will
1690 * not do this, so issue a WBINVD.
1692 wbinvd_on_all_cpus();
1694 __unregister_enc_region_locked(kvm, region);
1696 mutex_unlock(&kvm->lock);
1700 mutex_unlock(&kvm->lock);
1704 int svm_vm_copy_asid_from(struct kvm *kvm, unsigned int source_fd)
1706 struct file *source_kvm_file;
1707 struct kvm *source_kvm;
1708 struct kvm_sev_info *mirror_sev;
1712 source_kvm_file = fget(source_fd);
1713 if (!file_is_kvm(source_kvm_file)) {
1718 source_kvm = source_kvm_file->private_data;
1719 mutex_lock(&source_kvm->lock);
1721 if (!sev_guest(source_kvm)) {
1723 goto e_source_unlock;
1726 /* Mirrors of mirrors should work, but let's not get silly */
1727 if (is_mirroring_enc_context(source_kvm) || source_kvm == kvm) {
1729 goto e_source_unlock;
1732 asid = to_kvm_svm(source_kvm)->sev_info.asid;
1735 * The mirror kvm holds an enc_context_owner ref so its asid can't
1736 * disappear until we're done with it
1738 kvm_get_kvm(source_kvm);
1740 fput(source_kvm_file);
1741 mutex_unlock(&source_kvm->lock);
1742 mutex_lock(&kvm->lock);
1744 if (sev_guest(kvm)) {
1746 goto e_mirror_unlock;
1749 /* Set enc_context_owner and copy its encryption context over */
1750 mirror_sev = &to_kvm_svm(kvm)->sev_info;
1751 mirror_sev->enc_context_owner = source_kvm;
1752 mirror_sev->asid = asid;
1753 mirror_sev->active = true;
1755 mutex_unlock(&kvm->lock);
1759 mutex_unlock(&kvm->lock);
1760 kvm_put_kvm(source_kvm);
1763 mutex_unlock(&source_kvm->lock);
1765 if (source_kvm_file)
1766 fput(source_kvm_file);
1770 void sev_vm_destroy(struct kvm *kvm)
1772 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1773 struct list_head *head = &sev->regions_list;
1774 struct list_head *pos, *q;
1776 if (!sev_guest(kvm))
1779 /* If this is a mirror_kvm release the enc_context_owner and skip sev cleanup */
1780 if (is_mirroring_enc_context(kvm)) {
1781 kvm_put_kvm(sev->enc_context_owner);
1785 mutex_lock(&kvm->lock);
1788 * Ensure that all guest tagged cache entries are flushed before
1789 * releasing the pages back to the system for use. CLFLUSH will
1790 * not do this, so issue a WBINVD.
1792 wbinvd_on_all_cpus();
1795 * if userspace was terminated before unregistering the memory regions
1796 * then lets unpin all the registered memory.
1798 if (!list_empty(head)) {
1799 list_for_each_safe(pos, q, head) {
1800 __unregister_enc_region_locked(kvm,
1801 list_entry(pos, struct enc_region, list));
1806 mutex_unlock(&kvm->lock);
1808 sev_unbind_asid(kvm, sev->handle);
1812 void __init sev_set_cpu_caps(void)
1815 kvm_cpu_cap_clear(X86_FEATURE_SEV);
1816 if (!sev_es_enabled)
1817 kvm_cpu_cap_clear(X86_FEATURE_SEV_ES);
1820 void __init sev_hardware_setup(void)
1822 #ifdef CONFIG_KVM_AMD_SEV
1823 unsigned int eax, ebx, ecx, edx, sev_asid_count, sev_es_asid_count;
1824 bool sev_es_supported = false;
1825 bool sev_supported = false;
1827 if (!sev_enabled || !npt_enabled)
1830 /* Does the CPU support SEV? */
1831 if (!boot_cpu_has(X86_FEATURE_SEV))
1834 /* Retrieve SEV CPUID information */
1835 cpuid(0x8000001f, &eax, &ebx, &ecx, &edx);
1837 /* Set encryption bit location for SEV-ES guests */
1838 sev_enc_bit = ebx & 0x3f;
1840 /* Maximum number of encrypted guests supported simultaneously */
1845 /* Minimum ASID value that should be used for SEV guest */
1847 sev_me_mask = 1UL << (ebx & 0x3f);
1849 /* Initialize SEV ASID bitmaps */
1850 sev_asid_bitmap = bitmap_zalloc(max_sev_asid, GFP_KERNEL);
1851 if (!sev_asid_bitmap)
1854 sev_reclaim_asid_bitmap = bitmap_zalloc(max_sev_asid, GFP_KERNEL);
1855 if (!sev_reclaim_asid_bitmap) {
1856 bitmap_free(sev_asid_bitmap);
1857 sev_asid_bitmap = NULL;
1861 sev_asid_count = max_sev_asid - min_sev_asid + 1;
1862 if (misc_cg_set_capacity(MISC_CG_RES_SEV, sev_asid_count))
1865 pr_info("SEV supported: %u ASIDs\n", sev_asid_count);
1866 sev_supported = true;
1868 /* SEV-ES support requested? */
1869 if (!sev_es_enabled)
1872 /* Does the CPU support SEV-ES? */
1873 if (!boot_cpu_has(X86_FEATURE_SEV_ES))
1876 /* Has the system been allocated ASIDs for SEV-ES? */
1877 if (min_sev_asid == 1)
1880 sev_es_asid_count = min_sev_asid - 1;
1881 if (misc_cg_set_capacity(MISC_CG_RES_SEV_ES, sev_es_asid_count))
1884 pr_info("SEV-ES supported: %u ASIDs\n", sev_es_asid_count);
1885 sev_es_supported = true;
1888 sev_enabled = sev_supported;
1889 sev_es_enabled = sev_es_supported;
1893 void sev_hardware_teardown(void)
1898 /* No need to take sev_bitmap_lock, all VMs have been destroyed. */
1899 sev_flush_asids(0, max_sev_asid);
1901 bitmap_free(sev_asid_bitmap);
1902 bitmap_free(sev_reclaim_asid_bitmap);
1904 misc_cg_set_capacity(MISC_CG_RES_SEV, 0);
1905 misc_cg_set_capacity(MISC_CG_RES_SEV_ES, 0);
1908 int sev_cpu_init(struct svm_cpu_data *sd)
1913 sd->sev_vmcbs = kcalloc(max_sev_asid + 1, sizeof(void *), GFP_KERNEL);
1921 * Pages used by hardware to hold guest encrypted state must be flushed before
1922 * returning them to the system.
1924 static void sev_flush_guest_memory(struct vcpu_svm *svm, void *va,
1928 * If hardware enforced cache coherency for encrypted mappings of the
1929 * same physical page is supported, nothing to do.
1931 if (boot_cpu_has(X86_FEATURE_SME_COHERENT))
1935 * If the VM Page Flush MSR is supported, use it to flush the page
1936 * (using the page virtual address and the guest ASID).
1938 if (boot_cpu_has(X86_FEATURE_VM_PAGE_FLUSH)) {
1939 struct kvm_sev_info *sev;
1940 unsigned long va_start;
1943 /* Align start and stop to page boundaries. */
1944 va_start = (unsigned long)va;
1945 start = (u64)va_start & PAGE_MASK;
1946 stop = PAGE_ALIGN((u64)va_start + len);
1949 sev = &to_kvm_svm(svm->vcpu.kvm)->sev_info;
1951 while (start < stop) {
1952 wrmsrl(MSR_AMD64_VM_PAGE_FLUSH,
1961 WARN(1, "Address overflow, using WBINVD\n");
1965 * Hardware should always have one of the above features,
1966 * but if not, use WBINVD and issue a warning.
1968 WARN_ONCE(1, "Using WBINVD to flush guest memory\n");
1969 wbinvd_on_all_cpus();
1972 void sev_free_vcpu(struct kvm_vcpu *vcpu)
1974 struct vcpu_svm *svm;
1976 if (!sev_es_guest(vcpu->kvm))
1981 if (vcpu->arch.guest_state_protected)
1982 sev_flush_guest_memory(svm, svm->vmsa, PAGE_SIZE);
1983 __free_page(virt_to_page(svm->vmsa));
1985 if (svm->ghcb_sa_free)
1986 kfree(svm->ghcb_sa);
1989 static void dump_ghcb(struct vcpu_svm *svm)
1991 struct ghcb *ghcb = svm->ghcb;
1994 /* Re-use the dump_invalid_vmcb module parameter */
1995 if (!dump_invalid_vmcb) {
1996 pr_warn_ratelimited("set kvm_amd.dump_invalid_vmcb=1 to dump internal KVM state.\n");
2000 nbits = sizeof(ghcb->save.valid_bitmap) * 8;
2002 pr_err("GHCB (GPA=%016llx):\n", svm->vmcb->control.ghcb_gpa);
2003 pr_err("%-20s%016llx is_valid: %u\n", "sw_exit_code",
2004 ghcb->save.sw_exit_code, ghcb_sw_exit_code_is_valid(ghcb));
2005 pr_err("%-20s%016llx is_valid: %u\n", "sw_exit_info_1",
2006 ghcb->save.sw_exit_info_1, ghcb_sw_exit_info_1_is_valid(ghcb));
2007 pr_err("%-20s%016llx is_valid: %u\n", "sw_exit_info_2",
2008 ghcb->save.sw_exit_info_2, ghcb_sw_exit_info_2_is_valid(ghcb));
2009 pr_err("%-20s%016llx is_valid: %u\n", "sw_scratch",
2010 ghcb->save.sw_scratch, ghcb_sw_scratch_is_valid(ghcb));
2011 pr_err("%-20s%*pb\n", "valid_bitmap", nbits, ghcb->save.valid_bitmap);
2014 static void sev_es_sync_to_ghcb(struct vcpu_svm *svm)
2016 struct kvm_vcpu *vcpu = &svm->vcpu;
2017 struct ghcb *ghcb = svm->ghcb;
2020 * The GHCB protocol so far allows for the following data
2022 * GPRs RAX, RBX, RCX, RDX
2024 * Copy their values, even if they may not have been written during the
2025 * VM-Exit. It's the guest's responsibility to not consume random data.
2027 ghcb_set_rax(ghcb, vcpu->arch.regs[VCPU_REGS_RAX]);
2028 ghcb_set_rbx(ghcb, vcpu->arch.regs[VCPU_REGS_RBX]);
2029 ghcb_set_rcx(ghcb, vcpu->arch.regs[VCPU_REGS_RCX]);
2030 ghcb_set_rdx(ghcb, vcpu->arch.regs[VCPU_REGS_RDX]);
2033 static void sev_es_sync_from_ghcb(struct vcpu_svm *svm)
2035 struct vmcb_control_area *control = &svm->vmcb->control;
2036 struct kvm_vcpu *vcpu = &svm->vcpu;
2037 struct ghcb *ghcb = svm->ghcb;
2041 * The GHCB protocol so far allows for the following data
2043 * GPRs RAX, RBX, RCX, RDX
2047 * VMMCALL allows the guest to provide extra registers. KVM also
2048 * expects RSI for hypercalls, so include that, too.
2050 * Copy their values to the appropriate location if supplied.
2052 memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
2054 vcpu->arch.regs[VCPU_REGS_RAX] = ghcb_get_rax_if_valid(ghcb);
2055 vcpu->arch.regs[VCPU_REGS_RBX] = ghcb_get_rbx_if_valid(ghcb);
2056 vcpu->arch.regs[VCPU_REGS_RCX] = ghcb_get_rcx_if_valid(ghcb);
2057 vcpu->arch.regs[VCPU_REGS_RDX] = ghcb_get_rdx_if_valid(ghcb);
2058 vcpu->arch.regs[VCPU_REGS_RSI] = ghcb_get_rsi_if_valid(ghcb);
2060 svm->vmcb->save.cpl = ghcb_get_cpl_if_valid(ghcb);
2062 if (ghcb_xcr0_is_valid(ghcb)) {
2063 vcpu->arch.xcr0 = ghcb_get_xcr0(ghcb);
2064 kvm_update_cpuid_runtime(vcpu);
2067 /* Copy the GHCB exit information into the VMCB fields */
2068 exit_code = ghcb_get_sw_exit_code(ghcb);
2069 control->exit_code = lower_32_bits(exit_code);
2070 control->exit_code_hi = upper_32_bits(exit_code);
2071 control->exit_info_1 = ghcb_get_sw_exit_info_1(ghcb);
2072 control->exit_info_2 = ghcb_get_sw_exit_info_2(ghcb);
2074 /* Clear the valid entries fields */
2075 memset(ghcb->save.valid_bitmap, 0, sizeof(ghcb->save.valid_bitmap));
2078 static int sev_es_validate_vmgexit(struct vcpu_svm *svm)
2080 struct kvm_vcpu *vcpu;
2086 /* Only GHCB Usage code 0 is supported */
2087 if (ghcb->ghcb_usage)
2091 * Retrieve the exit code now even though is may not be marked valid
2092 * as it could help with debugging.
2094 exit_code = ghcb_get_sw_exit_code(ghcb);
2096 if (!ghcb_sw_exit_code_is_valid(ghcb) ||
2097 !ghcb_sw_exit_info_1_is_valid(ghcb) ||
2098 !ghcb_sw_exit_info_2_is_valid(ghcb))
2101 switch (ghcb_get_sw_exit_code(ghcb)) {
2102 case SVM_EXIT_READ_DR7:
2104 case SVM_EXIT_WRITE_DR7:
2105 if (!ghcb_rax_is_valid(ghcb))
2108 case SVM_EXIT_RDTSC:
2110 case SVM_EXIT_RDPMC:
2111 if (!ghcb_rcx_is_valid(ghcb))
2114 case SVM_EXIT_CPUID:
2115 if (!ghcb_rax_is_valid(ghcb) ||
2116 !ghcb_rcx_is_valid(ghcb))
2118 if (ghcb_get_rax(ghcb) == 0xd)
2119 if (!ghcb_xcr0_is_valid(ghcb))
2125 if (ghcb_get_sw_exit_info_1(ghcb) & SVM_IOIO_STR_MASK) {
2126 if (!ghcb_sw_scratch_is_valid(ghcb))
2129 if (!(ghcb_get_sw_exit_info_1(ghcb) & SVM_IOIO_TYPE_MASK))
2130 if (!ghcb_rax_is_valid(ghcb))
2135 if (!ghcb_rcx_is_valid(ghcb))
2137 if (ghcb_get_sw_exit_info_1(ghcb)) {
2138 if (!ghcb_rax_is_valid(ghcb) ||
2139 !ghcb_rdx_is_valid(ghcb))
2143 case SVM_EXIT_VMMCALL:
2144 if (!ghcb_rax_is_valid(ghcb) ||
2145 !ghcb_cpl_is_valid(ghcb))
2148 case SVM_EXIT_RDTSCP:
2150 case SVM_EXIT_WBINVD:
2152 case SVM_EXIT_MONITOR:
2153 if (!ghcb_rax_is_valid(ghcb) ||
2154 !ghcb_rcx_is_valid(ghcb) ||
2155 !ghcb_rdx_is_valid(ghcb))
2158 case SVM_EXIT_MWAIT:
2159 if (!ghcb_rax_is_valid(ghcb) ||
2160 !ghcb_rcx_is_valid(ghcb))
2163 case SVM_VMGEXIT_MMIO_READ:
2164 case SVM_VMGEXIT_MMIO_WRITE:
2165 if (!ghcb_sw_scratch_is_valid(ghcb))
2168 case SVM_VMGEXIT_NMI_COMPLETE:
2169 case SVM_VMGEXIT_AP_HLT_LOOP:
2170 case SVM_VMGEXIT_AP_JUMP_TABLE:
2171 case SVM_VMGEXIT_UNSUPPORTED_EVENT:
2182 if (ghcb->ghcb_usage) {
2183 vcpu_unimpl(vcpu, "vmgexit: ghcb usage %#x is not valid\n",
2186 vcpu_unimpl(vcpu, "vmgexit: exit reason %#llx is not valid\n",
2191 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
2192 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_UNEXPECTED_EXIT_REASON;
2193 vcpu->run->internal.ndata = 2;
2194 vcpu->run->internal.data[0] = exit_code;
2195 vcpu->run->internal.data[1] = vcpu->arch.last_vmentry_cpu;
2200 void sev_es_unmap_ghcb(struct vcpu_svm *svm)
2205 if (svm->ghcb_sa_free) {
2207 * The scratch area lives outside the GHCB, so there is a
2208 * buffer that, depending on the operation performed, may
2209 * need to be synced, then freed.
2211 if (svm->ghcb_sa_sync) {
2212 kvm_write_guest(svm->vcpu.kvm,
2213 ghcb_get_sw_scratch(svm->ghcb),
2214 svm->ghcb_sa, svm->ghcb_sa_len);
2215 svm->ghcb_sa_sync = false;
2218 kfree(svm->ghcb_sa);
2219 svm->ghcb_sa = NULL;
2220 svm->ghcb_sa_free = false;
2223 trace_kvm_vmgexit_exit(svm->vcpu.vcpu_id, svm->ghcb);
2225 sev_es_sync_to_ghcb(svm);
2227 kvm_vcpu_unmap(&svm->vcpu, &svm->ghcb_map, true);
2231 void pre_sev_run(struct vcpu_svm *svm, int cpu)
2233 struct svm_cpu_data *sd = per_cpu(svm_data, cpu);
2234 int asid = sev_get_asid(svm->vcpu.kvm);
2236 /* Assign the asid allocated with this SEV guest */
2242 * 1) when different VMCB for the same ASID is to be run on the same host CPU.
2243 * 2) or this VMCB was executed on different host CPU in previous VMRUNs.
2245 if (sd->sev_vmcbs[asid] == svm->vmcb &&
2246 svm->vcpu.arch.last_vmentry_cpu == cpu)
2249 sd->sev_vmcbs[asid] = svm->vmcb;
2250 svm->vmcb->control.tlb_ctl = TLB_CONTROL_FLUSH_ASID;
2251 vmcb_mark_dirty(svm->vmcb, VMCB_ASID);
2254 #define GHCB_SCRATCH_AREA_LIMIT (16ULL * PAGE_SIZE)
2255 static bool setup_vmgexit_scratch(struct vcpu_svm *svm, bool sync, u64 len)
2257 struct vmcb_control_area *control = &svm->vmcb->control;
2258 struct ghcb *ghcb = svm->ghcb;
2259 u64 ghcb_scratch_beg, ghcb_scratch_end;
2260 u64 scratch_gpa_beg, scratch_gpa_end;
2263 scratch_gpa_beg = ghcb_get_sw_scratch(ghcb);
2264 if (!scratch_gpa_beg) {
2265 pr_err("vmgexit: scratch gpa not provided\n");
2269 scratch_gpa_end = scratch_gpa_beg + len;
2270 if (scratch_gpa_end < scratch_gpa_beg) {
2271 pr_err("vmgexit: scratch length (%#llx) not valid for scratch address (%#llx)\n",
2272 len, scratch_gpa_beg);
2276 if ((scratch_gpa_beg & PAGE_MASK) == control->ghcb_gpa) {
2277 /* Scratch area begins within GHCB */
2278 ghcb_scratch_beg = control->ghcb_gpa +
2279 offsetof(struct ghcb, shared_buffer);
2280 ghcb_scratch_end = control->ghcb_gpa +
2281 offsetof(struct ghcb, reserved_1);
2284 * If the scratch area begins within the GHCB, it must be
2285 * completely contained in the GHCB shared buffer area.
2287 if (scratch_gpa_beg < ghcb_scratch_beg ||
2288 scratch_gpa_end > ghcb_scratch_end) {
2289 pr_err("vmgexit: scratch area is outside of GHCB shared buffer area (%#llx - %#llx)\n",
2290 scratch_gpa_beg, scratch_gpa_end);
2294 scratch_va = (void *)svm->ghcb;
2295 scratch_va += (scratch_gpa_beg - control->ghcb_gpa);
2298 * The guest memory must be read into a kernel buffer, so
2301 if (len > GHCB_SCRATCH_AREA_LIMIT) {
2302 pr_err("vmgexit: scratch area exceeds KVM limits (%#llx requested, %#llx limit)\n",
2303 len, GHCB_SCRATCH_AREA_LIMIT);
2306 scratch_va = kzalloc(len, GFP_KERNEL_ACCOUNT);
2310 if (kvm_read_guest(svm->vcpu.kvm, scratch_gpa_beg, scratch_va, len)) {
2311 /* Unable to copy scratch area from guest */
2312 pr_err("vmgexit: kvm_read_guest for scratch area failed\n");
2319 * The scratch area is outside the GHCB. The operation will
2320 * dictate whether the buffer needs to be synced before running
2321 * the vCPU next time (i.e. a read was requested so the data
2322 * must be written back to the guest memory).
2324 svm->ghcb_sa_sync = sync;
2325 svm->ghcb_sa_free = true;
2328 svm->ghcb_sa = scratch_va;
2329 svm->ghcb_sa_len = len;
2334 static void set_ghcb_msr_bits(struct vcpu_svm *svm, u64 value, u64 mask,
2337 svm->vmcb->control.ghcb_gpa &= ~(mask << pos);
2338 svm->vmcb->control.ghcb_gpa |= (value & mask) << pos;
2341 static u64 get_ghcb_msr_bits(struct vcpu_svm *svm, u64 mask, unsigned int pos)
2343 return (svm->vmcb->control.ghcb_gpa >> pos) & mask;
2346 static void set_ghcb_msr(struct vcpu_svm *svm, u64 value)
2348 svm->vmcb->control.ghcb_gpa = value;
2351 static int sev_handle_vmgexit_msr_protocol(struct vcpu_svm *svm)
2353 struct vmcb_control_area *control = &svm->vmcb->control;
2354 struct kvm_vcpu *vcpu = &svm->vcpu;
2358 ghcb_info = control->ghcb_gpa & GHCB_MSR_INFO_MASK;
2360 trace_kvm_vmgexit_msr_protocol_enter(svm->vcpu.vcpu_id,
2363 switch (ghcb_info) {
2364 case GHCB_MSR_SEV_INFO_REQ:
2365 set_ghcb_msr(svm, GHCB_MSR_SEV_INFO(GHCB_VERSION_MAX,
2369 case GHCB_MSR_CPUID_REQ: {
2370 u64 cpuid_fn, cpuid_reg, cpuid_value;
2372 cpuid_fn = get_ghcb_msr_bits(svm,
2373 GHCB_MSR_CPUID_FUNC_MASK,
2374 GHCB_MSR_CPUID_FUNC_POS);
2376 /* Initialize the registers needed by the CPUID intercept */
2377 vcpu->arch.regs[VCPU_REGS_RAX] = cpuid_fn;
2378 vcpu->arch.regs[VCPU_REGS_RCX] = 0;
2380 ret = svm_invoke_exit_handler(vcpu, SVM_EXIT_CPUID);
2386 cpuid_reg = get_ghcb_msr_bits(svm,
2387 GHCB_MSR_CPUID_REG_MASK,
2388 GHCB_MSR_CPUID_REG_POS);
2390 cpuid_value = vcpu->arch.regs[VCPU_REGS_RAX];
2391 else if (cpuid_reg == 1)
2392 cpuid_value = vcpu->arch.regs[VCPU_REGS_RBX];
2393 else if (cpuid_reg == 2)
2394 cpuid_value = vcpu->arch.regs[VCPU_REGS_RCX];
2396 cpuid_value = vcpu->arch.regs[VCPU_REGS_RDX];
2398 set_ghcb_msr_bits(svm, cpuid_value,
2399 GHCB_MSR_CPUID_VALUE_MASK,
2400 GHCB_MSR_CPUID_VALUE_POS);
2402 set_ghcb_msr_bits(svm, GHCB_MSR_CPUID_RESP,
2407 case GHCB_MSR_TERM_REQ: {
2408 u64 reason_set, reason_code;
2410 reason_set = get_ghcb_msr_bits(svm,
2411 GHCB_MSR_TERM_REASON_SET_MASK,
2412 GHCB_MSR_TERM_REASON_SET_POS);
2413 reason_code = get_ghcb_msr_bits(svm,
2414 GHCB_MSR_TERM_REASON_MASK,
2415 GHCB_MSR_TERM_REASON_POS);
2416 pr_info("SEV-ES guest requested termination: %#llx:%#llx\n",
2417 reason_set, reason_code);
2424 trace_kvm_vmgexit_msr_protocol_exit(svm->vcpu.vcpu_id,
2425 control->ghcb_gpa, ret);
2430 int sev_handle_vmgexit(struct kvm_vcpu *vcpu)
2432 struct vcpu_svm *svm = to_svm(vcpu);
2433 struct vmcb_control_area *control = &svm->vmcb->control;
2434 u64 ghcb_gpa, exit_code;
2438 /* Validate the GHCB */
2439 ghcb_gpa = control->ghcb_gpa;
2440 if (ghcb_gpa & GHCB_MSR_INFO_MASK)
2441 return sev_handle_vmgexit_msr_protocol(svm);
2444 vcpu_unimpl(vcpu, "vmgexit: GHCB gpa is not set\n");
2448 if (kvm_vcpu_map(vcpu, ghcb_gpa >> PAGE_SHIFT, &svm->ghcb_map)) {
2449 /* Unable to map GHCB from guest */
2450 vcpu_unimpl(vcpu, "vmgexit: error mapping GHCB [%#llx] from guest\n",
2455 svm->ghcb = svm->ghcb_map.hva;
2456 ghcb = svm->ghcb_map.hva;
2458 trace_kvm_vmgexit_enter(vcpu->vcpu_id, ghcb);
2460 exit_code = ghcb_get_sw_exit_code(ghcb);
2462 ret = sev_es_validate_vmgexit(svm);
2466 sev_es_sync_from_ghcb(svm);
2467 ghcb_set_sw_exit_info_1(ghcb, 0);
2468 ghcb_set_sw_exit_info_2(ghcb, 0);
2471 switch (exit_code) {
2472 case SVM_VMGEXIT_MMIO_READ:
2473 if (!setup_vmgexit_scratch(svm, true, control->exit_info_2))
2476 ret = kvm_sev_es_mmio_read(vcpu,
2477 control->exit_info_1,
2478 control->exit_info_2,
2481 case SVM_VMGEXIT_MMIO_WRITE:
2482 if (!setup_vmgexit_scratch(svm, false, control->exit_info_2))
2485 ret = kvm_sev_es_mmio_write(vcpu,
2486 control->exit_info_1,
2487 control->exit_info_2,
2490 case SVM_VMGEXIT_NMI_COMPLETE:
2491 ret = svm_invoke_exit_handler(vcpu, SVM_EXIT_IRET);
2493 case SVM_VMGEXIT_AP_HLT_LOOP:
2494 ret = kvm_emulate_ap_reset_hold(vcpu);
2496 case SVM_VMGEXIT_AP_JUMP_TABLE: {
2497 struct kvm_sev_info *sev = &to_kvm_svm(vcpu->kvm)->sev_info;
2499 switch (control->exit_info_1) {
2501 /* Set AP jump table address */
2502 sev->ap_jump_table = control->exit_info_2;
2505 /* Get AP jump table address */
2506 ghcb_set_sw_exit_info_2(ghcb, sev->ap_jump_table);
2509 pr_err("svm: vmgexit: unsupported AP jump table request - exit_info_1=%#llx\n",
2510 control->exit_info_1);
2511 ghcb_set_sw_exit_info_1(ghcb, 1);
2512 ghcb_set_sw_exit_info_2(ghcb,
2514 SVM_EVTINJ_TYPE_EXEPT |
2521 case SVM_VMGEXIT_UNSUPPORTED_EVENT:
2523 "vmgexit: unsupported event - exit_info_1=%#llx, exit_info_2=%#llx\n",
2524 control->exit_info_1, control->exit_info_2);
2527 ret = svm_invoke_exit_handler(vcpu, exit_code);
2533 int sev_es_string_io(struct vcpu_svm *svm, int size, unsigned int port, int in)
2535 if (!setup_vmgexit_scratch(svm, in, svm->vmcb->control.exit_info_2))
2538 return kvm_sev_es_string_io(&svm->vcpu, size, port,
2539 svm->ghcb_sa, svm->ghcb_sa_len, in);
2542 void sev_es_init_vmcb(struct vcpu_svm *svm)
2544 struct kvm_vcpu *vcpu = &svm->vcpu;
2546 svm->vmcb->control.nested_ctl |= SVM_NESTED_CTL_SEV_ES_ENABLE;
2547 svm->vmcb->control.virt_ext |= LBR_CTL_ENABLE_MASK;
2550 * An SEV-ES guest requires a VMSA area that is a separate from the
2551 * VMCB page. Do not include the encryption mask on the VMSA physical
2552 * address since hardware will access it using the guest key.
2554 svm->vmcb->control.vmsa_pa = __pa(svm->vmsa);
2556 /* Can't intercept CR register access, HV can't modify CR registers */
2557 svm_clr_intercept(svm, INTERCEPT_CR0_READ);
2558 svm_clr_intercept(svm, INTERCEPT_CR4_READ);
2559 svm_clr_intercept(svm, INTERCEPT_CR8_READ);
2560 svm_clr_intercept(svm, INTERCEPT_CR0_WRITE);
2561 svm_clr_intercept(svm, INTERCEPT_CR4_WRITE);
2562 svm_clr_intercept(svm, INTERCEPT_CR8_WRITE);
2564 svm_clr_intercept(svm, INTERCEPT_SELECTIVE_CR0);
2566 /* Track EFER/CR register changes */
2567 svm_set_intercept(svm, TRAP_EFER_WRITE);
2568 svm_set_intercept(svm, TRAP_CR0_WRITE);
2569 svm_set_intercept(svm, TRAP_CR4_WRITE);
2570 svm_set_intercept(svm, TRAP_CR8_WRITE);
2572 /* No support for enable_vmware_backdoor */
2573 clr_exception_intercept(svm, GP_VECTOR);
2575 /* Can't intercept XSETBV, HV can't modify XCR0 directly */
2576 svm_clr_intercept(svm, INTERCEPT_XSETBV);
2578 /* Clear intercepts on selected MSRs */
2579 set_msr_interception(vcpu, svm->msrpm, MSR_EFER, 1, 1);
2580 set_msr_interception(vcpu, svm->msrpm, MSR_IA32_CR_PAT, 1, 1);
2581 set_msr_interception(vcpu, svm->msrpm, MSR_IA32_LASTBRANCHFROMIP, 1, 1);
2582 set_msr_interception(vcpu, svm->msrpm, MSR_IA32_LASTBRANCHTOIP, 1, 1);
2583 set_msr_interception(vcpu, svm->msrpm, MSR_IA32_LASTINTFROMIP, 1, 1);
2584 set_msr_interception(vcpu, svm->msrpm, MSR_IA32_LASTINTTOIP, 1, 1);
2587 void sev_es_create_vcpu(struct vcpu_svm *svm)
2590 * Set the GHCB MSR value as per the GHCB specification when creating
2591 * a vCPU for an SEV-ES guest.
2593 set_ghcb_msr(svm, GHCB_MSR_SEV_INFO(GHCB_VERSION_MAX,
2598 void sev_es_prepare_guest_switch(struct vcpu_svm *svm, unsigned int cpu)
2600 struct svm_cpu_data *sd = per_cpu(svm_data, cpu);
2601 struct vmcb_save_area *hostsa;
2604 * As an SEV-ES guest, hardware will restore the host state on VMEXIT,
2605 * of which one step is to perform a VMLOAD. Since hardware does not
2606 * perform a VMSAVE on VMRUN, the host savearea must be updated.
2608 vmsave(__sme_page_pa(sd->save_area));
2610 /* XCR0 is restored on VMEXIT, save the current host value */
2611 hostsa = (struct vmcb_save_area *)(page_address(sd->save_area) + 0x400);
2612 hostsa->xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
2614 /* PKRU is restored on VMEXIT, save the current host value */
2615 hostsa->pkru = read_pkru();
2617 /* MSR_IA32_XSS is restored on VMEXIT, save the currnet host value */
2618 hostsa->xss = host_xss;
2621 void sev_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
2623 struct vcpu_svm *svm = to_svm(vcpu);
2625 /* First SIPI: Use the values as initially set by the VMM */
2626 if (!svm->received_first_sipi) {
2627 svm->received_first_sipi = true;
2632 * Subsequent SIPI: Return from an AP Reset Hold VMGEXIT, where
2633 * the guest will set the CS and RIP. Set SW_EXIT_INFO_2 to a
2639 ghcb_set_sw_exit_info_2(svm->ghcb, 1);