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_decommission(unsigned int handle)
204 struct sev_data_decommission decommission;
209 decommission.handle = handle;
210 sev_guest_decommission(&decommission, NULL);
213 static void sev_unbind_asid(struct kvm *kvm, unsigned int handle)
215 struct sev_data_deactivate deactivate;
220 deactivate.handle = handle;
222 /* Guard DEACTIVATE against WBINVD/DF_FLUSH used in ASID recycling */
223 down_read(&sev_deactivate_lock);
224 sev_guest_deactivate(&deactivate, NULL);
225 up_read(&sev_deactivate_lock);
227 sev_decommission(handle);
230 static int sev_guest_init(struct kvm *kvm, struct kvm_sev_cmd *argp)
232 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
233 bool es_active = argp->id == KVM_SEV_ES_INIT;
236 if (kvm->created_vcpus)
240 if (unlikely(sev->active))
243 sev->es_active = es_active;
244 asid = sev_asid_new(sev);
249 ret = sev_platform_init(&argp->error);
255 INIT_LIST_HEAD(&sev->regions_list);
263 sev->es_active = false;
267 static int sev_bind_asid(struct kvm *kvm, unsigned int handle, int *error)
269 struct sev_data_activate activate;
270 int asid = sev_get_asid(kvm);
273 /* activate ASID on the given handle */
274 activate.handle = handle;
275 activate.asid = asid;
276 ret = sev_guest_activate(&activate, error);
281 static int __sev_issue_cmd(int fd, int id, void *data, int *error)
290 ret = sev_issue_cmd_external_user(f.file, id, data, error);
296 static int sev_issue_cmd(struct kvm *kvm, int id, void *data, int *error)
298 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
300 return __sev_issue_cmd(sev->fd, id, data, error);
303 static int sev_launch_start(struct kvm *kvm, struct kvm_sev_cmd *argp)
305 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
306 struct sev_data_launch_start start;
307 struct kvm_sev_launch_start params;
308 void *dh_blob, *session_blob;
309 int *error = &argp->error;
315 if (copy_from_user(¶ms, (void __user *)(uintptr_t)argp->data, sizeof(params)))
318 memset(&start, 0, sizeof(start));
321 if (params.dh_uaddr) {
322 dh_blob = psp_copy_user_blob(params.dh_uaddr, params.dh_len);
324 return PTR_ERR(dh_blob);
326 start.dh_cert_address = __sme_set(__pa(dh_blob));
327 start.dh_cert_len = params.dh_len;
331 if (params.session_uaddr) {
332 session_blob = psp_copy_user_blob(params.session_uaddr, params.session_len);
333 if (IS_ERR(session_blob)) {
334 ret = PTR_ERR(session_blob);
338 start.session_address = __sme_set(__pa(session_blob));
339 start.session_len = params.session_len;
342 start.handle = params.handle;
343 start.policy = params.policy;
345 /* create memory encryption context */
346 ret = __sev_issue_cmd(argp->sev_fd, SEV_CMD_LAUNCH_START, &start, error);
350 /* Bind ASID to this guest */
351 ret = sev_bind_asid(kvm, start.handle, error);
353 sev_decommission(start.handle);
357 /* return handle to userspace */
358 params.handle = start.handle;
359 if (copy_to_user((void __user *)(uintptr_t)argp->data, ¶ms, sizeof(params))) {
360 sev_unbind_asid(kvm, start.handle);
365 sev->handle = start.handle;
366 sev->fd = argp->sev_fd;
375 static struct page **sev_pin_memory(struct kvm *kvm, unsigned long uaddr,
376 unsigned long ulen, unsigned long *n,
379 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
380 unsigned long npages, size;
382 unsigned long locked, lock_limit;
384 unsigned long first, last;
387 lockdep_assert_held(&kvm->lock);
389 if (ulen == 0 || uaddr + ulen < uaddr)
390 return ERR_PTR(-EINVAL);
392 /* Calculate number of pages. */
393 first = (uaddr & PAGE_MASK) >> PAGE_SHIFT;
394 last = ((uaddr + ulen - 1) & PAGE_MASK) >> PAGE_SHIFT;
395 npages = (last - first + 1);
397 locked = sev->pages_locked + npages;
398 lock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
399 if (locked > lock_limit && !capable(CAP_IPC_LOCK)) {
400 pr_err("SEV: %lu locked pages exceed the lock limit of %lu.\n", locked, lock_limit);
401 return ERR_PTR(-ENOMEM);
404 if (WARN_ON_ONCE(npages > INT_MAX))
405 return ERR_PTR(-EINVAL);
407 /* Avoid using vmalloc for smaller buffers. */
408 size = npages * sizeof(struct page *);
409 if (size > PAGE_SIZE)
410 pages = __vmalloc(size, GFP_KERNEL_ACCOUNT | __GFP_ZERO);
412 pages = kmalloc(size, GFP_KERNEL_ACCOUNT);
415 return ERR_PTR(-ENOMEM);
417 /* Pin the user virtual address. */
418 npinned = pin_user_pages_fast(uaddr, npages, write ? FOLL_WRITE : 0, pages);
419 if (npinned != npages) {
420 pr_err("SEV: Failure locking %lu pages.\n", npages);
426 sev->pages_locked = locked;
432 unpin_user_pages(pages, npinned);
438 static void sev_unpin_memory(struct kvm *kvm, struct page **pages,
439 unsigned long npages)
441 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
443 unpin_user_pages(pages, npages);
445 sev->pages_locked -= npages;
448 static void sev_clflush_pages(struct page *pages[], unsigned long npages)
450 uint8_t *page_virtual;
453 if (this_cpu_has(X86_FEATURE_SME_COHERENT) || npages == 0 ||
457 for (i = 0; i < npages; i++) {
458 page_virtual = kmap_atomic(pages[i]);
459 clflush_cache_range(page_virtual, PAGE_SIZE);
460 kunmap_atomic(page_virtual);
464 static unsigned long get_num_contig_pages(unsigned long idx,
465 struct page **inpages, unsigned long npages)
467 unsigned long paddr, next_paddr;
468 unsigned long i = idx + 1, pages = 1;
470 /* find the number of contiguous pages starting from idx */
471 paddr = __sme_page_pa(inpages[idx]);
473 next_paddr = __sme_page_pa(inpages[i++]);
474 if ((paddr + PAGE_SIZE) == next_paddr) {
485 static int sev_launch_update_data(struct kvm *kvm, struct kvm_sev_cmd *argp)
487 unsigned long vaddr, vaddr_end, next_vaddr, npages, pages, size, i;
488 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
489 struct kvm_sev_launch_update_data params;
490 struct sev_data_launch_update_data data;
491 struct page **inpages;
497 if (copy_from_user(¶ms, (void __user *)(uintptr_t)argp->data, sizeof(params)))
500 vaddr = params.uaddr;
502 vaddr_end = vaddr + size;
504 /* Lock the user memory. */
505 inpages = sev_pin_memory(kvm, vaddr, size, &npages, 1);
507 return PTR_ERR(inpages);
510 * Flush (on non-coherent CPUs) before LAUNCH_UPDATE encrypts pages in
511 * place; the cache may contain the data that was written unencrypted.
513 sev_clflush_pages(inpages, npages);
516 data.handle = sev->handle;
518 for (i = 0; vaddr < vaddr_end; vaddr = next_vaddr, i += pages) {
522 * If the user buffer is not page-aligned, calculate the offset
525 offset = vaddr & (PAGE_SIZE - 1);
527 /* Calculate the number of pages that can be encrypted in one go. */
528 pages = get_num_contig_pages(i, inpages, npages);
530 len = min_t(size_t, ((pages * PAGE_SIZE) - offset), size);
533 data.address = __sme_page_pa(inpages[i]) + offset;
534 ret = sev_issue_cmd(kvm, SEV_CMD_LAUNCH_UPDATE_DATA, &data, &argp->error);
539 next_vaddr = vaddr + len;
543 /* content of memory is updated, mark pages dirty */
544 for (i = 0; i < npages; i++) {
545 set_page_dirty_lock(inpages[i]);
546 mark_page_accessed(inpages[i]);
548 /* unlock the user pages */
549 sev_unpin_memory(kvm, inpages, npages);
553 static int sev_es_sync_vmsa(struct vcpu_svm *svm)
555 struct vmcb_save_area *save = &svm->vmcb->save;
557 /* Check some debug related fields before encrypting the VMSA */
558 if (svm->vcpu.guest_debug || (save->dr7 & ~DR7_FIXED_1))
561 /* Sync registgers */
562 save->rax = svm->vcpu.arch.regs[VCPU_REGS_RAX];
563 save->rbx = svm->vcpu.arch.regs[VCPU_REGS_RBX];
564 save->rcx = svm->vcpu.arch.regs[VCPU_REGS_RCX];
565 save->rdx = svm->vcpu.arch.regs[VCPU_REGS_RDX];
566 save->rsp = svm->vcpu.arch.regs[VCPU_REGS_RSP];
567 save->rbp = svm->vcpu.arch.regs[VCPU_REGS_RBP];
568 save->rsi = svm->vcpu.arch.regs[VCPU_REGS_RSI];
569 save->rdi = svm->vcpu.arch.regs[VCPU_REGS_RDI];
571 save->r8 = svm->vcpu.arch.regs[VCPU_REGS_R8];
572 save->r9 = svm->vcpu.arch.regs[VCPU_REGS_R9];
573 save->r10 = svm->vcpu.arch.regs[VCPU_REGS_R10];
574 save->r11 = svm->vcpu.arch.regs[VCPU_REGS_R11];
575 save->r12 = svm->vcpu.arch.regs[VCPU_REGS_R12];
576 save->r13 = svm->vcpu.arch.regs[VCPU_REGS_R13];
577 save->r14 = svm->vcpu.arch.regs[VCPU_REGS_R14];
578 save->r15 = svm->vcpu.arch.regs[VCPU_REGS_R15];
580 save->rip = svm->vcpu.arch.regs[VCPU_REGS_RIP];
582 /* Sync some non-GPR registers before encrypting */
583 save->xcr0 = svm->vcpu.arch.xcr0;
584 save->pkru = svm->vcpu.arch.pkru;
585 save->xss = svm->vcpu.arch.ia32_xss;
588 * SEV-ES will use a VMSA that is pointed to by the VMCB, not
589 * the traditional VMSA that is part of the VMCB. Copy the
590 * traditional VMSA as it has been built so far (in prep
591 * for LAUNCH_UPDATE_VMSA) to be the initial SEV-ES state.
593 memcpy(svm->vmsa, save, sizeof(*save));
598 static int sev_launch_update_vmsa(struct kvm *kvm, struct kvm_sev_cmd *argp)
600 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
601 struct sev_data_launch_update_vmsa vmsa;
602 struct kvm_vcpu *vcpu;
605 if (!sev_es_guest(kvm))
610 kvm_for_each_vcpu(i, vcpu, kvm) {
611 struct vcpu_svm *svm = to_svm(vcpu);
613 /* Perform some pre-encryption checks against the VMSA */
614 ret = sev_es_sync_vmsa(svm);
619 * The LAUNCH_UPDATE_VMSA command will perform in-place
620 * encryption of the VMSA memory content (i.e it will write
621 * the same memory region with the guest's key), so invalidate
624 clflush_cache_range(svm->vmsa, PAGE_SIZE);
626 vmsa.handle = sev->handle;
627 vmsa.address = __sme_pa(svm->vmsa);
628 vmsa.len = PAGE_SIZE;
629 ret = sev_issue_cmd(kvm, SEV_CMD_LAUNCH_UPDATE_VMSA, &vmsa,
634 svm->vcpu.arch.guest_state_protected = true;
640 static int sev_launch_measure(struct kvm *kvm, struct kvm_sev_cmd *argp)
642 void __user *measure = (void __user *)(uintptr_t)argp->data;
643 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
644 struct sev_data_launch_measure data;
645 struct kvm_sev_launch_measure params;
646 void __user *p = NULL;
653 if (copy_from_user(¶ms, measure, sizeof(params)))
656 memset(&data, 0, sizeof(data));
658 /* User wants to query the blob length */
662 p = (void __user *)(uintptr_t)params.uaddr;
664 if (params.len > SEV_FW_BLOB_MAX_SIZE)
667 blob = kmalloc(params.len, GFP_KERNEL_ACCOUNT);
671 data.address = __psp_pa(blob);
672 data.len = params.len;
676 data.handle = sev->handle;
677 ret = sev_issue_cmd(kvm, SEV_CMD_LAUNCH_MEASURE, &data, &argp->error);
680 * If we query the session length, FW responded with expected data.
689 if (copy_to_user(p, blob, params.len))
694 params.len = data.len;
695 if (copy_to_user(measure, ¶ms, sizeof(params)))
702 static int sev_launch_finish(struct kvm *kvm, struct kvm_sev_cmd *argp)
704 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
705 struct sev_data_launch_finish data;
710 data.handle = sev->handle;
711 return sev_issue_cmd(kvm, SEV_CMD_LAUNCH_FINISH, &data, &argp->error);
714 static int sev_guest_status(struct kvm *kvm, struct kvm_sev_cmd *argp)
716 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
717 struct kvm_sev_guest_status params;
718 struct sev_data_guest_status data;
724 memset(&data, 0, sizeof(data));
726 data.handle = sev->handle;
727 ret = sev_issue_cmd(kvm, SEV_CMD_GUEST_STATUS, &data, &argp->error);
731 params.policy = data.policy;
732 params.state = data.state;
733 params.handle = data.handle;
735 if (copy_to_user((void __user *)(uintptr_t)argp->data, ¶ms, sizeof(params)))
741 static int __sev_issue_dbg_cmd(struct kvm *kvm, unsigned long src,
742 unsigned long dst, int size,
743 int *error, bool enc)
745 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
746 struct sev_data_dbg data;
749 data.handle = sev->handle;
754 return sev_issue_cmd(kvm,
755 enc ? SEV_CMD_DBG_ENCRYPT : SEV_CMD_DBG_DECRYPT,
759 static int __sev_dbg_decrypt(struct kvm *kvm, unsigned long src_paddr,
760 unsigned long dst_paddr, int sz, int *err)
765 * Its safe to read more than we are asked, caller should ensure that
766 * destination has enough space.
768 offset = src_paddr & 15;
769 src_paddr = round_down(src_paddr, 16);
770 sz = round_up(sz + offset, 16);
772 return __sev_issue_dbg_cmd(kvm, src_paddr, dst_paddr, sz, err, false);
775 static int __sev_dbg_decrypt_user(struct kvm *kvm, unsigned long paddr,
776 void __user *dst_uaddr,
777 unsigned long dst_paddr,
780 struct page *tpage = NULL;
783 /* if inputs are not 16-byte then use intermediate buffer */
784 if (!IS_ALIGNED(dst_paddr, 16) ||
785 !IS_ALIGNED(paddr, 16) ||
786 !IS_ALIGNED(size, 16)) {
787 tpage = (void *)alloc_page(GFP_KERNEL);
791 dst_paddr = __sme_page_pa(tpage);
794 ret = __sev_dbg_decrypt(kvm, paddr, dst_paddr, size, err);
800 if (copy_to_user(dst_uaddr, page_address(tpage) + offset, size))
811 static int __sev_dbg_encrypt_user(struct kvm *kvm, unsigned long paddr,
813 unsigned long dst_paddr,
814 void __user *dst_vaddr,
815 int size, int *error)
817 struct page *src_tpage = NULL;
818 struct page *dst_tpage = NULL;
821 /* If source buffer is not aligned then use an intermediate buffer */
822 if (!IS_ALIGNED((unsigned long)vaddr, 16)) {
823 src_tpage = alloc_page(GFP_KERNEL);
827 if (copy_from_user(page_address(src_tpage), vaddr, size)) {
828 __free_page(src_tpage);
832 paddr = __sme_page_pa(src_tpage);
836 * If destination buffer or length is not aligned then do read-modify-write:
837 * - decrypt destination in an intermediate buffer
838 * - copy the source buffer in an intermediate buffer
839 * - use the intermediate buffer as source buffer
841 if (!IS_ALIGNED((unsigned long)dst_vaddr, 16) || !IS_ALIGNED(size, 16)) {
844 dst_tpage = alloc_page(GFP_KERNEL);
850 ret = __sev_dbg_decrypt(kvm, dst_paddr,
851 __sme_page_pa(dst_tpage), size, error);
856 * If source is kernel buffer then use memcpy() otherwise
859 dst_offset = dst_paddr & 15;
862 memcpy(page_address(dst_tpage) + dst_offset,
863 page_address(src_tpage), size);
865 if (copy_from_user(page_address(dst_tpage) + dst_offset,
872 paddr = __sme_page_pa(dst_tpage);
873 dst_paddr = round_down(dst_paddr, 16);
874 len = round_up(size, 16);
877 ret = __sev_issue_dbg_cmd(kvm, paddr, dst_paddr, len, error, true);
881 __free_page(src_tpage);
883 __free_page(dst_tpage);
887 static int sev_dbg_crypt(struct kvm *kvm, struct kvm_sev_cmd *argp, bool dec)
889 unsigned long vaddr, vaddr_end, next_vaddr;
890 unsigned long dst_vaddr;
891 struct page **src_p, **dst_p;
892 struct kvm_sev_dbg debug;
900 if (copy_from_user(&debug, (void __user *)(uintptr_t)argp->data, sizeof(debug)))
903 if (!debug.len || debug.src_uaddr + debug.len < debug.src_uaddr)
905 if (!debug.dst_uaddr)
908 vaddr = debug.src_uaddr;
910 vaddr_end = vaddr + size;
911 dst_vaddr = debug.dst_uaddr;
913 for (; vaddr < vaddr_end; vaddr = next_vaddr) {
914 int len, s_off, d_off;
916 /* lock userspace source and destination page */
917 src_p = sev_pin_memory(kvm, vaddr & PAGE_MASK, PAGE_SIZE, &n, 0);
919 return PTR_ERR(src_p);
921 dst_p = sev_pin_memory(kvm, dst_vaddr & PAGE_MASK, PAGE_SIZE, &n, 1);
923 sev_unpin_memory(kvm, src_p, n);
924 return PTR_ERR(dst_p);
928 * Flush (on non-coherent CPUs) before DBG_{DE,EN}CRYPT read or modify
929 * the pages; flush the destination too so that future accesses do not
932 sev_clflush_pages(src_p, 1);
933 sev_clflush_pages(dst_p, 1);
936 * Since user buffer may not be page aligned, calculate the
937 * offset within the page.
939 s_off = vaddr & ~PAGE_MASK;
940 d_off = dst_vaddr & ~PAGE_MASK;
941 len = min_t(size_t, (PAGE_SIZE - s_off), size);
944 ret = __sev_dbg_decrypt_user(kvm,
945 __sme_page_pa(src_p[0]) + s_off,
946 (void __user *)dst_vaddr,
947 __sme_page_pa(dst_p[0]) + d_off,
950 ret = __sev_dbg_encrypt_user(kvm,
951 __sme_page_pa(src_p[0]) + s_off,
952 (void __user *)vaddr,
953 __sme_page_pa(dst_p[0]) + d_off,
954 (void __user *)dst_vaddr,
957 sev_unpin_memory(kvm, src_p, n);
958 sev_unpin_memory(kvm, dst_p, n);
963 next_vaddr = vaddr + len;
964 dst_vaddr = dst_vaddr + len;
971 static int sev_launch_secret(struct kvm *kvm, struct kvm_sev_cmd *argp)
973 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
974 struct sev_data_launch_secret data;
975 struct kvm_sev_launch_secret params;
984 if (copy_from_user(¶ms, (void __user *)(uintptr_t)argp->data, sizeof(params)))
987 pages = sev_pin_memory(kvm, params.guest_uaddr, params.guest_len, &n, 1);
989 return PTR_ERR(pages);
992 * Flush (on non-coherent CPUs) before LAUNCH_SECRET encrypts pages in
993 * place; the cache may contain the data that was written unencrypted.
995 sev_clflush_pages(pages, n);
998 * The secret must be copied into contiguous memory region, lets verify
999 * that userspace memory pages are contiguous before we issue command.
1001 if (get_num_contig_pages(0, pages, n) != n) {
1003 goto e_unpin_memory;
1006 memset(&data, 0, sizeof(data));
1008 offset = params.guest_uaddr & (PAGE_SIZE - 1);
1009 data.guest_address = __sme_page_pa(pages[0]) + offset;
1010 data.guest_len = params.guest_len;
1012 blob = psp_copy_user_blob(params.trans_uaddr, params.trans_len);
1014 ret = PTR_ERR(blob);
1015 goto e_unpin_memory;
1018 data.trans_address = __psp_pa(blob);
1019 data.trans_len = params.trans_len;
1021 hdr = psp_copy_user_blob(params.hdr_uaddr, params.hdr_len);
1026 data.hdr_address = __psp_pa(hdr);
1027 data.hdr_len = params.hdr_len;
1029 data.handle = sev->handle;
1030 ret = sev_issue_cmd(kvm, SEV_CMD_LAUNCH_UPDATE_SECRET, &data, &argp->error);
1037 /* content of memory is updated, mark pages dirty */
1038 for (i = 0; i < n; i++) {
1039 set_page_dirty_lock(pages[i]);
1040 mark_page_accessed(pages[i]);
1042 sev_unpin_memory(kvm, pages, n);
1046 static int sev_get_attestation_report(struct kvm *kvm, struct kvm_sev_cmd *argp)
1048 void __user *report = (void __user *)(uintptr_t)argp->data;
1049 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1050 struct sev_data_attestation_report data;
1051 struct kvm_sev_attestation_report params;
1056 if (!sev_guest(kvm))
1059 if (copy_from_user(¶ms, (void __user *)(uintptr_t)argp->data, sizeof(params)))
1062 memset(&data, 0, sizeof(data));
1064 /* User wants to query the blob length */
1068 p = (void __user *)(uintptr_t)params.uaddr;
1070 if (params.len > SEV_FW_BLOB_MAX_SIZE)
1073 blob = kmalloc(params.len, GFP_KERNEL_ACCOUNT);
1077 data.address = __psp_pa(blob);
1078 data.len = params.len;
1079 memcpy(data.mnonce, params.mnonce, sizeof(params.mnonce));
1082 data.handle = sev->handle;
1083 ret = sev_issue_cmd(kvm, SEV_CMD_ATTESTATION_REPORT, &data, &argp->error);
1085 * If we query the session length, FW responded with expected data.
1094 if (copy_to_user(p, blob, params.len))
1099 params.len = data.len;
1100 if (copy_to_user(report, ¶ms, sizeof(params)))
1107 /* Userspace wants to query session length. */
1109 __sev_send_start_query_session_length(struct kvm *kvm, struct kvm_sev_cmd *argp,
1110 struct kvm_sev_send_start *params)
1112 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1113 struct sev_data_send_start data;
1116 memset(&data, 0, sizeof(data));
1117 data.handle = sev->handle;
1118 ret = sev_issue_cmd(kvm, SEV_CMD_SEND_START, &data, &argp->error);
1120 params->session_len = data.session_len;
1121 if (copy_to_user((void __user *)(uintptr_t)argp->data, params,
1122 sizeof(struct kvm_sev_send_start)))
1128 static int sev_send_start(struct kvm *kvm, struct kvm_sev_cmd *argp)
1130 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1131 struct sev_data_send_start data;
1132 struct kvm_sev_send_start params;
1133 void *amd_certs, *session_data;
1134 void *pdh_cert, *plat_certs;
1137 if (!sev_guest(kvm))
1140 if (copy_from_user(¶ms, (void __user *)(uintptr_t)argp->data,
1141 sizeof(struct kvm_sev_send_start)))
1144 /* if session_len is zero, userspace wants to query the session length */
1145 if (!params.session_len)
1146 return __sev_send_start_query_session_length(kvm, argp,
1149 /* some sanity checks */
1150 if (!params.pdh_cert_uaddr || !params.pdh_cert_len ||
1151 !params.session_uaddr || params.session_len > SEV_FW_BLOB_MAX_SIZE)
1154 /* allocate the memory to hold the session data blob */
1155 session_data = kmalloc(params.session_len, GFP_KERNEL_ACCOUNT);
1159 /* copy the certificate blobs from userspace */
1160 pdh_cert = psp_copy_user_blob(params.pdh_cert_uaddr,
1161 params.pdh_cert_len);
1162 if (IS_ERR(pdh_cert)) {
1163 ret = PTR_ERR(pdh_cert);
1164 goto e_free_session;
1167 plat_certs = psp_copy_user_blob(params.plat_certs_uaddr,
1168 params.plat_certs_len);
1169 if (IS_ERR(plat_certs)) {
1170 ret = PTR_ERR(plat_certs);
1174 amd_certs = psp_copy_user_blob(params.amd_certs_uaddr,
1175 params.amd_certs_len);
1176 if (IS_ERR(amd_certs)) {
1177 ret = PTR_ERR(amd_certs);
1178 goto e_free_plat_cert;
1181 /* populate the FW SEND_START field with system physical address */
1182 memset(&data, 0, sizeof(data));
1183 data.pdh_cert_address = __psp_pa(pdh_cert);
1184 data.pdh_cert_len = params.pdh_cert_len;
1185 data.plat_certs_address = __psp_pa(plat_certs);
1186 data.plat_certs_len = params.plat_certs_len;
1187 data.amd_certs_address = __psp_pa(amd_certs);
1188 data.amd_certs_len = params.amd_certs_len;
1189 data.session_address = __psp_pa(session_data);
1190 data.session_len = params.session_len;
1191 data.handle = sev->handle;
1193 ret = sev_issue_cmd(kvm, SEV_CMD_SEND_START, &data, &argp->error);
1195 if (!ret && copy_to_user((void __user *)(uintptr_t)params.session_uaddr,
1196 session_data, params.session_len)) {
1198 goto e_free_amd_cert;
1201 params.policy = data.policy;
1202 params.session_len = data.session_len;
1203 if (copy_to_user((void __user *)(uintptr_t)argp->data, ¶ms,
1204 sizeof(struct kvm_sev_send_start)))
1214 kfree(session_data);
1218 /* Userspace wants to query either header or trans length. */
1220 __sev_send_update_data_query_lengths(struct kvm *kvm, struct kvm_sev_cmd *argp,
1221 struct kvm_sev_send_update_data *params)
1223 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1224 struct sev_data_send_update_data data;
1227 memset(&data, 0, sizeof(data));
1228 data.handle = sev->handle;
1229 ret = sev_issue_cmd(kvm, SEV_CMD_SEND_UPDATE_DATA, &data, &argp->error);
1231 params->hdr_len = data.hdr_len;
1232 params->trans_len = data.trans_len;
1234 if (copy_to_user((void __user *)(uintptr_t)argp->data, params,
1235 sizeof(struct kvm_sev_send_update_data)))
1241 static int sev_send_update_data(struct kvm *kvm, struct kvm_sev_cmd *argp)
1243 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1244 struct sev_data_send_update_data data;
1245 struct kvm_sev_send_update_data params;
1246 void *hdr, *trans_data;
1247 struct page **guest_page;
1251 if (!sev_guest(kvm))
1254 if (copy_from_user(¶ms, (void __user *)(uintptr_t)argp->data,
1255 sizeof(struct kvm_sev_send_update_data)))
1258 /* userspace wants to query either header or trans length */
1259 if (!params.trans_len || !params.hdr_len)
1260 return __sev_send_update_data_query_lengths(kvm, argp, ¶ms);
1262 if (!params.trans_uaddr || !params.guest_uaddr ||
1263 !params.guest_len || !params.hdr_uaddr)
1266 /* Check if we are crossing the page boundary */
1267 offset = params.guest_uaddr & (PAGE_SIZE - 1);
1268 if ((params.guest_len + offset > PAGE_SIZE))
1271 /* Pin guest memory */
1272 guest_page = sev_pin_memory(kvm, params.guest_uaddr & PAGE_MASK,
1277 /* allocate memory for header and transport buffer */
1279 hdr = kmalloc(params.hdr_len, GFP_KERNEL_ACCOUNT);
1283 trans_data = kmalloc(params.trans_len, GFP_KERNEL_ACCOUNT);
1287 memset(&data, 0, sizeof(data));
1288 data.hdr_address = __psp_pa(hdr);
1289 data.hdr_len = params.hdr_len;
1290 data.trans_address = __psp_pa(trans_data);
1291 data.trans_len = params.trans_len;
1293 /* The SEND_UPDATE_DATA command requires C-bit to be always set. */
1294 data.guest_address = (page_to_pfn(guest_page[0]) << PAGE_SHIFT) + offset;
1295 data.guest_address |= sev_me_mask;
1296 data.guest_len = params.guest_len;
1297 data.handle = sev->handle;
1299 ret = sev_issue_cmd(kvm, SEV_CMD_SEND_UPDATE_DATA, &data, &argp->error);
1302 goto e_free_trans_data;
1304 /* copy transport buffer to user space */
1305 if (copy_to_user((void __user *)(uintptr_t)params.trans_uaddr,
1306 trans_data, params.trans_len)) {
1308 goto e_free_trans_data;
1311 /* Copy packet header to userspace. */
1312 ret = copy_to_user((void __user *)(uintptr_t)params.hdr_uaddr, hdr,
1320 sev_unpin_memory(kvm, guest_page, n);
1325 static int sev_send_finish(struct kvm *kvm, struct kvm_sev_cmd *argp)
1327 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1328 struct sev_data_send_finish data;
1330 if (!sev_guest(kvm))
1333 data.handle = sev->handle;
1334 return sev_issue_cmd(kvm, SEV_CMD_SEND_FINISH, &data, &argp->error);
1337 static int sev_send_cancel(struct kvm *kvm, struct kvm_sev_cmd *argp)
1339 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1340 struct sev_data_send_cancel data;
1342 if (!sev_guest(kvm))
1345 data.handle = sev->handle;
1346 return sev_issue_cmd(kvm, SEV_CMD_SEND_CANCEL, &data, &argp->error);
1349 static int sev_receive_start(struct kvm *kvm, struct kvm_sev_cmd *argp)
1351 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1352 struct sev_data_receive_start start;
1353 struct kvm_sev_receive_start params;
1354 int *error = &argp->error;
1359 if (!sev_guest(kvm))
1362 /* Get parameter from the userspace */
1363 if (copy_from_user(¶ms, (void __user *)(uintptr_t)argp->data,
1364 sizeof(struct kvm_sev_receive_start)))
1367 /* some sanity checks */
1368 if (!params.pdh_uaddr || !params.pdh_len ||
1369 !params.session_uaddr || !params.session_len)
1372 pdh_data = psp_copy_user_blob(params.pdh_uaddr, params.pdh_len);
1373 if (IS_ERR(pdh_data))
1374 return PTR_ERR(pdh_data);
1376 session_data = psp_copy_user_blob(params.session_uaddr,
1377 params.session_len);
1378 if (IS_ERR(session_data)) {
1379 ret = PTR_ERR(session_data);
1383 memset(&start, 0, sizeof(start));
1384 start.handle = params.handle;
1385 start.policy = params.policy;
1386 start.pdh_cert_address = __psp_pa(pdh_data);
1387 start.pdh_cert_len = params.pdh_len;
1388 start.session_address = __psp_pa(session_data);
1389 start.session_len = params.session_len;
1391 /* create memory encryption context */
1392 ret = __sev_issue_cmd(argp->sev_fd, SEV_CMD_RECEIVE_START, &start,
1395 goto e_free_session;
1397 /* Bind ASID to this guest */
1398 ret = sev_bind_asid(kvm, start.handle, error);
1400 goto e_free_session;
1402 params.handle = start.handle;
1403 if (copy_to_user((void __user *)(uintptr_t)argp->data,
1404 ¶ms, sizeof(struct kvm_sev_receive_start))) {
1406 sev_unbind_asid(kvm, start.handle);
1407 goto e_free_session;
1410 sev->handle = start.handle;
1411 sev->fd = argp->sev_fd;
1414 kfree(session_data);
1421 static int sev_receive_update_data(struct kvm *kvm, struct kvm_sev_cmd *argp)
1423 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1424 struct kvm_sev_receive_update_data params;
1425 struct sev_data_receive_update_data data;
1426 void *hdr = NULL, *trans = NULL;
1427 struct page **guest_page;
1431 if (!sev_guest(kvm))
1434 if (copy_from_user(¶ms, (void __user *)(uintptr_t)argp->data,
1435 sizeof(struct kvm_sev_receive_update_data)))
1438 if (!params.hdr_uaddr || !params.hdr_len ||
1439 !params.guest_uaddr || !params.guest_len ||
1440 !params.trans_uaddr || !params.trans_len)
1443 /* Check if we are crossing the page boundary */
1444 offset = params.guest_uaddr & (PAGE_SIZE - 1);
1445 if ((params.guest_len + offset > PAGE_SIZE))
1448 hdr = psp_copy_user_blob(params.hdr_uaddr, params.hdr_len);
1450 return PTR_ERR(hdr);
1452 trans = psp_copy_user_blob(params.trans_uaddr, params.trans_len);
1453 if (IS_ERR(trans)) {
1454 ret = PTR_ERR(trans);
1458 memset(&data, 0, sizeof(data));
1459 data.hdr_address = __psp_pa(hdr);
1460 data.hdr_len = params.hdr_len;
1461 data.trans_address = __psp_pa(trans);
1462 data.trans_len = params.trans_len;
1464 /* Pin guest memory */
1466 guest_page = sev_pin_memory(kvm, params.guest_uaddr & PAGE_MASK,
1471 /* The RECEIVE_UPDATE_DATA command requires C-bit to be always set. */
1472 data.guest_address = (page_to_pfn(guest_page[0]) << PAGE_SHIFT) + offset;
1473 data.guest_address |= sev_me_mask;
1474 data.guest_len = params.guest_len;
1475 data.handle = sev->handle;
1477 ret = sev_issue_cmd(kvm, SEV_CMD_RECEIVE_UPDATE_DATA, &data,
1480 sev_unpin_memory(kvm, guest_page, n);
1490 static int sev_receive_finish(struct kvm *kvm, struct kvm_sev_cmd *argp)
1492 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1493 struct sev_data_receive_finish data;
1495 if (!sev_guest(kvm))
1498 data.handle = sev->handle;
1499 return sev_issue_cmd(kvm, SEV_CMD_RECEIVE_FINISH, &data, &argp->error);
1502 int svm_mem_enc_op(struct kvm *kvm, void __user *argp)
1504 struct kvm_sev_cmd sev_cmd;
1513 if (copy_from_user(&sev_cmd, argp, sizeof(struct kvm_sev_cmd)))
1516 mutex_lock(&kvm->lock);
1518 /* enc_context_owner handles all memory enc operations */
1519 if (is_mirroring_enc_context(kvm)) {
1524 switch (sev_cmd.id) {
1525 case KVM_SEV_ES_INIT:
1526 if (!sev_es_enabled) {
1532 r = sev_guest_init(kvm, &sev_cmd);
1534 case KVM_SEV_LAUNCH_START:
1535 r = sev_launch_start(kvm, &sev_cmd);
1537 case KVM_SEV_LAUNCH_UPDATE_DATA:
1538 r = sev_launch_update_data(kvm, &sev_cmd);
1540 case KVM_SEV_LAUNCH_UPDATE_VMSA:
1541 r = sev_launch_update_vmsa(kvm, &sev_cmd);
1543 case KVM_SEV_LAUNCH_MEASURE:
1544 r = sev_launch_measure(kvm, &sev_cmd);
1546 case KVM_SEV_LAUNCH_FINISH:
1547 r = sev_launch_finish(kvm, &sev_cmd);
1549 case KVM_SEV_GUEST_STATUS:
1550 r = sev_guest_status(kvm, &sev_cmd);
1552 case KVM_SEV_DBG_DECRYPT:
1553 r = sev_dbg_crypt(kvm, &sev_cmd, true);
1555 case KVM_SEV_DBG_ENCRYPT:
1556 r = sev_dbg_crypt(kvm, &sev_cmd, false);
1558 case KVM_SEV_LAUNCH_SECRET:
1559 r = sev_launch_secret(kvm, &sev_cmd);
1561 case KVM_SEV_GET_ATTESTATION_REPORT:
1562 r = sev_get_attestation_report(kvm, &sev_cmd);
1564 case KVM_SEV_SEND_START:
1565 r = sev_send_start(kvm, &sev_cmd);
1567 case KVM_SEV_SEND_UPDATE_DATA:
1568 r = sev_send_update_data(kvm, &sev_cmd);
1570 case KVM_SEV_SEND_FINISH:
1571 r = sev_send_finish(kvm, &sev_cmd);
1573 case KVM_SEV_SEND_CANCEL:
1574 r = sev_send_cancel(kvm, &sev_cmd);
1576 case KVM_SEV_RECEIVE_START:
1577 r = sev_receive_start(kvm, &sev_cmd);
1579 case KVM_SEV_RECEIVE_UPDATE_DATA:
1580 r = sev_receive_update_data(kvm, &sev_cmd);
1582 case KVM_SEV_RECEIVE_FINISH:
1583 r = sev_receive_finish(kvm, &sev_cmd);
1590 if (copy_to_user(argp, &sev_cmd, sizeof(struct kvm_sev_cmd)))
1594 mutex_unlock(&kvm->lock);
1598 int svm_register_enc_region(struct kvm *kvm,
1599 struct kvm_enc_region *range)
1601 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1602 struct enc_region *region;
1605 if (!sev_guest(kvm))
1608 /* If kvm is mirroring encryption context it isn't responsible for it */
1609 if (is_mirroring_enc_context(kvm))
1612 if (range->addr > ULONG_MAX || range->size > ULONG_MAX)
1615 region = kzalloc(sizeof(*region), GFP_KERNEL_ACCOUNT);
1619 mutex_lock(&kvm->lock);
1620 region->pages = sev_pin_memory(kvm, range->addr, range->size, ®ion->npages, 1);
1621 if (IS_ERR(region->pages)) {
1622 ret = PTR_ERR(region->pages);
1623 mutex_unlock(&kvm->lock);
1627 region->uaddr = range->addr;
1628 region->size = range->size;
1630 list_add_tail(®ion->list, &sev->regions_list);
1631 mutex_unlock(&kvm->lock);
1634 * The guest may change the memory encryption attribute from C=0 -> C=1
1635 * or vice versa for this memory range. Lets make sure caches are
1636 * flushed to ensure that guest data gets written into memory with
1639 sev_clflush_pages(region->pages, region->npages);
1648 static struct enc_region *
1649 find_enc_region(struct kvm *kvm, struct kvm_enc_region *range)
1651 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1652 struct list_head *head = &sev->regions_list;
1653 struct enc_region *i;
1655 list_for_each_entry(i, head, list) {
1656 if (i->uaddr == range->addr &&
1657 i->size == range->size)
1664 static void __unregister_enc_region_locked(struct kvm *kvm,
1665 struct enc_region *region)
1667 sev_unpin_memory(kvm, region->pages, region->npages);
1668 list_del(®ion->list);
1672 int svm_unregister_enc_region(struct kvm *kvm,
1673 struct kvm_enc_region *range)
1675 struct enc_region *region;
1678 /* If kvm is mirroring encryption context it isn't responsible for it */
1679 if (is_mirroring_enc_context(kvm))
1682 mutex_lock(&kvm->lock);
1684 if (!sev_guest(kvm)) {
1689 region = find_enc_region(kvm, range);
1696 * Ensure that all guest tagged cache entries are flushed before
1697 * releasing the pages back to the system for use. CLFLUSH will
1698 * not do this, so issue a WBINVD.
1700 wbinvd_on_all_cpus();
1702 __unregister_enc_region_locked(kvm, region);
1704 mutex_unlock(&kvm->lock);
1708 mutex_unlock(&kvm->lock);
1712 int svm_vm_copy_asid_from(struct kvm *kvm, unsigned int source_fd)
1714 struct file *source_kvm_file;
1715 struct kvm *source_kvm;
1716 struct kvm_sev_info *mirror_sev;
1720 source_kvm_file = fget(source_fd);
1721 if (!file_is_kvm(source_kvm_file)) {
1726 source_kvm = source_kvm_file->private_data;
1727 mutex_lock(&source_kvm->lock);
1729 if (!sev_guest(source_kvm)) {
1731 goto e_source_unlock;
1734 /* Mirrors of mirrors should work, but let's not get silly */
1735 if (is_mirroring_enc_context(source_kvm) || source_kvm == kvm) {
1737 goto e_source_unlock;
1740 asid = to_kvm_svm(source_kvm)->sev_info.asid;
1743 * The mirror kvm holds an enc_context_owner ref so its asid can't
1744 * disappear until we're done with it
1746 kvm_get_kvm(source_kvm);
1748 fput(source_kvm_file);
1749 mutex_unlock(&source_kvm->lock);
1750 mutex_lock(&kvm->lock);
1752 if (sev_guest(kvm)) {
1754 goto e_mirror_unlock;
1757 /* Set enc_context_owner and copy its encryption context over */
1758 mirror_sev = &to_kvm_svm(kvm)->sev_info;
1759 mirror_sev->enc_context_owner = source_kvm;
1760 mirror_sev->asid = asid;
1761 mirror_sev->active = true;
1763 mutex_unlock(&kvm->lock);
1767 mutex_unlock(&kvm->lock);
1768 kvm_put_kvm(source_kvm);
1771 mutex_unlock(&source_kvm->lock);
1773 if (source_kvm_file)
1774 fput(source_kvm_file);
1778 void sev_vm_destroy(struct kvm *kvm)
1780 struct kvm_sev_info *sev = &to_kvm_svm(kvm)->sev_info;
1781 struct list_head *head = &sev->regions_list;
1782 struct list_head *pos, *q;
1784 if (!sev_guest(kvm))
1787 /* If this is a mirror_kvm release the enc_context_owner and skip sev cleanup */
1788 if (is_mirroring_enc_context(kvm)) {
1789 kvm_put_kvm(sev->enc_context_owner);
1793 mutex_lock(&kvm->lock);
1796 * Ensure that all guest tagged cache entries are flushed before
1797 * releasing the pages back to the system for use. CLFLUSH will
1798 * not do this, so issue a WBINVD.
1800 wbinvd_on_all_cpus();
1803 * if userspace was terminated before unregistering the memory regions
1804 * then lets unpin all the registered memory.
1806 if (!list_empty(head)) {
1807 list_for_each_safe(pos, q, head) {
1808 __unregister_enc_region_locked(kvm,
1809 list_entry(pos, struct enc_region, list));
1814 mutex_unlock(&kvm->lock);
1816 sev_unbind_asid(kvm, sev->handle);
1820 void __init sev_set_cpu_caps(void)
1823 kvm_cpu_cap_clear(X86_FEATURE_SEV);
1824 if (!sev_es_enabled)
1825 kvm_cpu_cap_clear(X86_FEATURE_SEV_ES);
1828 void __init sev_hardware_setup(void)
1830 #ifdef CONFIG_KVM_AMD_SEV
1831 unsigned int eax, ebx, ecx, edx, sev_asid_count, sev_es_asid_count;
1832 bool sev_es_supported = false;
1833 bool sev_supported = false;
1835 if (!sev_enabled || !npt_enabled)
1838 /* Does the CPU support SEV? */
1839 if (!boot_cpu_has(X86_FEATURE_SEV))
1842 /* Retrieve SEV CPUID information */
1843 cpuid(0x8000001f, &eax, &ebx, &ecx, &edx);
1845 /* Set encryption bit location for SEV-ES guests */
1846 sev_enc_bit = ebx & 0x3f;
1848 /* Maximum number of encrypted guests supported simultaneously */
1853 /* Minimum ASID value that should be used for SEV guest */
1855 sev_me_mask = 1UL << (ebx & 0x3f);
1857 /* Initialize SEV ASID bitmaps */
1858 sev_asid_bitmap = bitmap_zalloc(max_sev_asid, GFP_KERNEL);
1859 if (!sev_asid_bitmap)
1862 sev_reclaim_asid_bitmap = bitmap_zalloc(max_sev_asid, GFP_KERNEL);
1863 if (!sev_reclaim_asid_bitmap) {
1864 bitmap_free(sev_asid_bitmap);
1865 sev_asid_bitmap = NULL;
1869 sev_asid_count = max_sev_asid - min_sev_asid + 1;
1870 if (misc_cg_set_capacity(MISC_CG_RES_SEV, sev_asid_count))
1873 pr_info("SEV supported: %u ASIDs\n", sev_asid_count);
1874 sev_supported = true;
1876 /* SEV-ES support requested? */
1877 if (!sev_es_enabled)
1880 /* Does the CPU support SEV-ES? */
1881 if (!boot_cpu_has(X86_FEATURE_SEV_ES))
1884 /* Has the system been allocated ASIDs for SEV-ES? */
1885 if (min_sev_asid == 1)
1888 sev_es_asid_count = min_sev_asid - 1;
1889 if (misc_cg_set_capacity(MISC_CG_RES_SEV_ES, sev_es_asid_count))
1892 pr_info("SEV-ES supported: %u ASIDs\n", sev_es_asid_count);
1893 sev_es_supported = true;
1896 sev_enabled = sev_supported;
1897 sev_es_enabled = sev_es_supported;
1901 void sev_hardware_teardown(void)
1906 /* No need to take sev_bitmap_lock, all VMs have been destroyed. */
1907 sev_flush_asids(0, max_sev_asid);
1909 bitmap_free(sev_asid_bitmap);
1910 bitmap_free(sev_reclaim_asid_bitmap);
1912 misc_cg_set_capacity(MISC_CG_RES_SEV, 0);
1913 misc_cg_set_capacity(MISC_CG_RES_SEV_ES, 0);
1916 int sev_cpu_init(struct svm_cpu_data *sd)
1921 sd->sev_vmcbs = kcalloc(max_sev_asid + 1, sizeof(void *), GFP_KERNEL);
1929 * Pages used by hardware to hold guest encrypted state must be flushed before
1930 * returning them to the system.
1932 static void sev_flush_guest_memory(struct vcpu_svm *svm, void *va,
1936 * If hardware enforced cache coherency for encrypted mappings of the
1937 * same physical page is supported, nothing to do.
1939 if (boot_cpu_has(X86_FEATURE_SME_COHERENT))
1943 * If the VM Page Flush MSR is supported, use it to flush the page
1944 * (using the page virtual address and the guest ASID).
1946 if (boot_cpu_has(X86_FEATURE_VM_PAGE_FLUSH)) {
1947 struct kvm_sev_info *sev;
1948 unsigned long va_start;
1951 /* Align start and stop to page boundaries. */
1952 va_start = (unsigned long)va;
1953 start = (u64)va_start & PAGE_MASK;
1954 stop = PAGE_ALIGN((u64)va_start + len);
1957 sev = &to_kvm_svm(svm->vcpu.kvm)->sev_info;
1959 while (start < stop) {
1960 wrmsrl(MSR_AMD64_VM_PAGE_FLUSH,
1969 WARN(1, "Address overflow, using WBINVD\n");
1973 * Hardware should always have one of the above features,
1974 * but if not, use WBINVD and issue a warning.
1976 WARN_ONCE(1, "Using WBINVD to flush guest memory\n");
1977 wbinvd_on_all_cpus();
1980 void sev_free_vcpu(struct kvm_vcpu *vcpu)
1982 struct vcpu_svm *svm;
1984 if (!sev_es_guest(vcpu->kvm))
1989 if (vcpu->arch.guest_state_protected)
1990 sev_flush_guest_memory(svm, svm->vmsa, PAGE_SIZE);
1991 __free_page(virt_to_page(svm->vmsa));
1993 if (svm->ghcb_sa_free)
1994 kfree(svm->ghcb_sa);
1997 static void dump_ghcb(struct vcpu_svm *svm)
1999 struct ghcb *ghcb = svm->ghcb;
2002 /* Re-use the dump_invalid_vmcb module parameter */
2003 if (!dump_invalid_vmcb) {
2004 pr_warn_ratelimited("set kvm_amd.dump_invalid_vmcb=1 to dump internal KVM state.\n");
2008 nbits = sizeof(ghcb->save.valid_bitmap) * 8;
2010 pr_err("GHCB (GPA=%016llx):\n", svm->vmcb->control.ghcb_gpa);
2011 pr_err("%-20s%016llx is_valid: %u\n", "sw_exit_code",
2012 ghcb->save.sw_exit_code, ghcb_sw_exit_code_is_valid(ghcb));
2013 pr_err("%-20s%016llx is_valid: %u\n", "sw_exit_info_1",
2014 ghcb->save.sw_exit_info_1, ghcb_sw_exit_info_1_is_valid(ghcb));
2015 pr_err("%-20s%016llx is_valid: %u\n", "sw_exit_info_2",
2016 ghcb->save.sw_exit_info_2, ghcb_sw_exit_info_2_is_valid(ghcb));
2017 pr_err("%-20s%016llx is_valid: %u\n", "sw_scratch",
2018 ghcb->save.sw_scratch, ghcb_sw_scratch_is_valid(ghcb));
2019 pr_err("%-20s%*pb\n", "valid_bitmap", nbits, ghcb->save.valid_bitmap);
2022 static void sev_es_sync_to_ghcb(struct vcpu_svm *svm)
2024 struct kvm_vcpu *vcpu = &svm->vcpu;
2025 struct ghcb *ghcb = svm->ghcb;
2028 * The GHCB protocol so far allows for the following data
2030 * GPRs RAX, RBX, RCX, RDX
2032 * Copy their values, even if they may not have been written during the
2033 * VM-Exit. It's the guest's responsibility to not consume random data.
2035 ghcb_set_rax(ghcb, vcpu->arch.regs[VCPU_REGS_RAX]);
2036 ghcb_set_rbx(ghcb, vcpu->arch.regs[VCPU_REGS_RBX]);
2037 ghcb_set_rcx(ghcb, vcpu->arch.regs[VCPU_REGS_RCX]);
2038 ghcb_set_rdx(ghcb, vcpu->arch.regs[VCPU_REGS_RDX]);
2041 static void sev_es_sync_from_ghcb(struct vcpu_svm *svm)
2043 struct vmcb_control_area *control = &svm->vmcb->control;
2044 struct kvm_vcpu *vcpu = &svm->vcpu;
2045 struct ghcb *ghcb = svm->ghcb;
2049 * The GHCB protocol so far allows for the following data
2051 * GPRs RAX, RBX, RCX, RDX
2055 * VMMCALL allows the guest to provide extra registers. KVM also
2056 * expects RSI for hypercalls, so include that, too.
2058 * Copy their values to the appropriate location if supplied.
2060 memset(vcpu->arch.regs, 0, sizeof(vcpu->arch.regs));
2062 vcpu->arch.regs[VCPU_REGS_RAX] = ghcb_get_rax_if_valid(ghcb);
2063 vcpu->arch.regs[VCPU_REGS_RBX] = ghcb_get_rbx_if_valid(ghcb);
2064 vcpu->arch.regs[VCPU_REGS_RCX] = ghcb_get_rcx_if_valid(ghcb);
2065 vcpu->arch.regs[VCPU_REGS_RDX] = ghcb_get_rdx_if_valid(ghcb);
2066 vcpu->arch.regs[VCPU_REGS_RSI] = ghcb_get_rsi_if_valid(ghcb);
2068 svm->vmcb->save.cpl = ghcb_get_cpl_if_valid(ghcb);
2070 if (ghcb_xcr0_is_valid(ghcb)) {
2071 vcpu->arch.xcr0 = ghcb_get_xcr0(ghcb);
2072 kvm_update_cpuid_runtime(vcpu);
2075 /* Copy the GHCB exit information into the VMCB fields */
2076 exit_code = ghcb_get_sw_exit_code(ghcb);
2077 control->exit_code = lower_32_bits(exit_code);
2078 control->exit_code_hi = upper_32_bits(exit_code);
2079 control->exit_info_1 = ghcb_get_sw_exit_info_1(ghcb);
2080 control->exit_info_2 = ghcb_get_sw_exit_info_2(ghcb);
2082 /* Clear the valid entries fields */
2083 memset(ghcb->save.valid_bitmap, 0, sizeof(ghcb->save.valid_bitmap));
2086 static int sev_es_validate_vmgexit(struct vcpu_svm *svm)
2088 struct kvm_vcpu *vcpu;
2094 /* Only GHCB Usage code 0 is supported */
2095 if (ghcb->ghcb_usage)
2099 * Retrieve the exit code now even though is may not be marked valid
2100 * as it could help with debugging.
2102 exit_code = ghcb_get_sw_exit_code(ghcb);
2104 if (!ghcb_sw_exit_code_is_valid(ghcb) ||
2105 !ghcb_sw_exit_info_1_is_valid(ghcb) ||
2106 !ghcb_sw_exit_info_2_is_valid(ghcb))
2109 switch (ghcb_get_sw_exit_code(ghcb)) {
2110 case SVM_EXIT_READ_DR7:
2112 case SVM_EXIT_WRITE_DR7:
2113 if (!ghcb_rax_is_valid(ghcb))
2116 case SVM_EXIT_RDTSC:
2118 case SVM_EXIT_RDPMC:
2119 if (!ghcb_rcx_is_valid(ghcb))
2122 case SVM_EXIT_CPUID:
2123 if (!ghcb_rax_is_valid(ghcb) ||
2124 !ghcb_rcx_is_valid(ghcb))
2126 if (ghcb_get_rax(ghcb) == 0xd)
2127 if (!ghcb_xcr0_is_valid(ghcb))
2133 if (ghcb_get_sw_exit_info_1(ghcb) & SVM_IOIO_STR_MASK) {
2134 if (!ghcb_sw_scratch_is_valid(ghcb))
2137 if (!(ghcb_get_sw_exit_info_1(ghcb) & SVM_IOIO_TYPE_MASK))
2138 if (!ghcb_rax_is_valid(ghcb))
2143 if (!ghcb_rcx_is_valid(ghcb))
2145 if (ghcb_get_sw_exit_info_1(ghcb)) {
2146 if (!ghcb_rax_is_valid(ghcb) ||
2147 !ghcb_rdx_is_valid(ghcb))
2151 case SVM_EXIT_VMMCALL:
2152 if (!ghcb_rax_is_valid(ghcb) ||
2153 !ghcb_cpl_is_valid(ghcb))
2156 case SVM_EXIT_RDTSCP:
2158 case SVM_EXIT_WBINVD:
2160 case SVM_EXIT_MONITOR:
2161 if (!ghcb_rax_is_valid(ghcb) ||
2162 !ghcb_rcx_is_valid(ghcb) ||
2163 !ghcb_rdx_is_valid(ghcb))
2166 case SVM_EXIT_MWAIT:
2167 if (!ghcb_rax_is_valid(ghcb) ||
2168 !ghcb_rcx_is_valid(ghcb))
2171 case SVM_VMGEXIT_MMIO_READ:
2172 case SVM_VMGEXIT_MMIO_WRITE:
2173 if (!ghcb_sw_scratch_is_valid(ghcb))
2176 case SVM_VMGEXIT_NMI_COMPLETE:
2177 case SVM_VMGEXIT_AP_HLT_LOOP:
2178 case SVM_VMGEXIT_AP_JUMP_TABLE:
2179 case SVM_VMGEXIT_UNSUPPORTED_EVENT:
2190 if (ghcb->ghcb_usage) {
2191 vcpu_unimpl(vcpu, "vmgexit: ghcb usage %#x is not valid\n",
2194 vcpu_unimpl(vcpu, "vmgexit: exit reason %#llx is not valid\n",
2199 vcpu->run->exit_reason = KVM_EXIT_INTERNAL_ERROR;
2200 vcpu->run->internal.suberror = KVM_INTERNAL_ERROR_UNEXPECTED_EXIT_REASON;
2201 vcpu->run->internal.ndata = 2;
2202 vcpu->run->internal.data[0] = exit_code;
2203 vcpu->run->internal.data[1] = vcpu->arch.last_vmentry_cpu;
2208 void sev_es_unmap_ghcb(struct vcpu_svm *svm)
2213 if (svm->ghcb_sa_free) {
2215 * The scratch area lives outside the GHCB, so there is a
2216 * buffer that, depending on the operation performed, may
2217 * need to be synced, then freed.
2219 if (svm->ghcb_sa_sync) {
2220 kvm_write_guest(svm->vcpu.kvm,
2221 ghcb_get_sw_scratch(svm->ghcb),
2222 svm->ghcb_sa, svm->ghcb_sa_len);
2223 svm->ghcb_sa_sync = false;
2226 kfree(svm->ghcb_sa);
2227 svm->ghcb_sa = NULL;
2228 svm->ghcb_sa_free = false;
2231 trace_kvm_vmgexit_exit(svm->vcpu.vcpu_id, svm->ghcb);
2233 sev_es_sync_to_ghcb(svm);
2235 kvm_vcpu_unmap(&svm->vcpu, &svm->ghcb_map, true);
2239 void pre_sev_run(struct vcpu_svm *svm, int cpu)
2241 struct svm_cpu_data *sd = per_cpu(svm_data, cpu);
2242 int asid = sev_get_asid(svm->vcpu.kvm);
2244 /* Assign the asid allocated with this SEV guest */
2250 * 1) when different VMCB for the same ASID is to be run on the same host CPU.
2251 * 2) or this VMCB was executed on different host CPU in previous VMRUNs.
2253 if (sd->sev_vmcbs[asid] == svm->vmcb &&
2254 svm->vcpu.arch.last_vmentry_cpu == cpu)
2257 sd->sev_vmcbs[asid] = svm->vmcb;
2258 svm->vmcb->control.tlb_ctl = TLB_CONTROL_FLUSH_ASID;
2259 vmcb_mark_dirty(svm->vmcb, VMCB_ASID);
2262 #define GHCB_SCRATCH_AREA_LIMIT (16ULL * PAGE_SIZE)
2263 static bool setup_vmgexit_scratch(struct vcpu_svm *svm, bool sync, u64 len)
2265 struct vmcb_control_area *control = &svm->vmcb->control;
2266 struct ghcb *ghcb = svm->ghcb;
2267 u64 ghcb_scratch_beg, ghcb_scratch_end;
2268 u64 scratch_gpa_beg, scratch_gpa_end;
2271 scratch_gpa_beg = ghcb_get_sw_scratch(ghcb);
2272 if (!scratch_gpa_beg) {
2273 pr_err("vmgexit: scratch gpa not provided\n");
2277 scratch_gpa_end = scratch_gpa_beg + len;
2278 if (scratch_gpa_end < scratch_gpa_beg) {
2279 pr_err("vmgexit: scratch length (%#llx) not valid for scratch address (%#llx)\n",
2280 len, scratch_gpa_beg);
2284 if ((scratch_gpa_beg & PAGE_MASK) == control->ghcb_gpa) {
2285 /* Scratch area begins within GHCB */
2286 ghcb_scratch_beg = control->ghcb_gpa +
2287 offsetof(struct ghcb, shared_buffer);
2288 ghcb_scratch_end = control->ghcb_gpa +
2289 offsetof(struct ghcb, reserved_1);
2292 * If the scratch area begins within the GHCB, it must be
2293 * completely contained in the GHCB shared buffer area.
2295 if (scratch_gpa_beg < ghcb_scratch_beg ||
2296 scratch_gpa_end > ghcb_scratch_end) {
2297 pr_err("vmgexit: scratch area is outside of GHCB shared buffer area (%#llx - %#llx)\n",
2298 scratch_gpa_beg, scratch_gpa_end);
2302 scratch_va = (void *)svm->ghcb;
2303 scratch_va += (scratch_gpa_beg - control->ghcb_gpa);
2306 * The guest memory must be read into a kernel buffer, so
2309 if (len > GHCB_SCRATCH_AREA_LIMIT) {
2310 pr_err("vmgexit: scratch area exceeds KVM limits (%#llx requested, %#llx limit)\n",
2311 len, GHCB_SCRATCH_AREA_LIMIT);
2314 scratch_va = kzalloc(len, GFP_KERNEL_ACCOUNT);
2318 if (kvm_read_guest(svm->vcpu.kvm, scratch_gpa_beg, scratch_va, len)) {
2319 /* Unable to copy scratch area from guest */
2320 pr_err("vmgexit: kvm_read_guest for scratch area failed\n");
2327 * The scratch area is outside the GHCB. The operation will
2328 * dictate whether the buffer needs to be synced before running
2329 * the vCPU next time (i.e. a read was requested so the data
2330 * must be written back to the guest memory).
2332 svm->ghcb_sa_sync = sync;
2333 svm->ghcb_sa_free = true;
2336 svm->ghcb_sa = scratch_va;
2337 svm->ghcb_sa_len = len;
2342 static void set_ghcb_msr_bits(struct vcpu_svm *svm, u64 value, u64 mask,
2345 svm->vmcb->control.ghcb_gpa &= ~(mask << pos);
2346 svm->vmcb->control.ghcb_gpa |= (value & mask) << pos;
2349 static u64 get_ghcb_msr_bits(struct vcpu_svm *svm, u64 mask, unsigned int pos)
2351 return (svm->vmcb->control.ghcb_gpa >> pos) & mask;
2354 static void set_ghcb_msr(struct vcpu_svm *svm, u64 value)
2356 svm->vmcb->control.ghcb_gpa = value;
2359 static int sev_handle_vmgexit_msr_protocol(struct vcpu_svm *svm)
2361 struct vmcb_control_area *control = &svm->vmcb->control;
2362 struct kvm_vcpu *vcpu = &svm->vcpu;
2366 ghcb_info = control->ghcb_gpa & GHCB_MSR_INFO_MASK;
2368 trace_kvm_vmgexit_msr_protocol_enter(svm->vcpu.vcpu_id,
2371 switch (ghcb_info) {
2372 case GHCB_MSR_SEV_INFO_REQ:
2373 set_ghcb_msr(svm, GHCB_MSR_SEV_INFO(GHCB_VERSION_MAX,
2377 case GHCB_MSR_CPUID_REQ: {
2378 u64 cpuid_fn, cpuid_reg, cpuid_value;
2380 cpuid_fn = get_ghcb_msr_bits(svm,
2381 GHCB_MSR_CPUID_FUNC_MASK,
2382 GHCB_MSR_CPUID_FUNC_POS);
2384 /* Initialize the registers needed by the CPUID intercept */
2385 vcpu->arch.regs[VCPU_REGS_RAX] = cpuid_fn;
2386 vcpu->arch.regs[VCPU_REGS_RCX] = 0;
2388 ret = svm_invoke_exit_handler(vcpu, SVM_EXIT_CPUID);
2394 cpuid_reg = get_ghcb_msr_bits(svm,
2395 GHCB_MSR_CPUID_REG_MASK,
2396 GHCB_MSR_CPUID_REG_POS);
2398 cpuid_value = vcpu->arch.regs[VCPU_REGS_RAX];
2399 else if (cpuid_reg == 1)
2400 cpuid_value = vcpu->arch.regs[VCPU_REGS_RBX];
2401 else if (cpuid_reg == 2)
2402 cpuid_value = vcpu->arch.regs[VCPU_REGS_RCX];
2404 cpuid_value = vcpu->arch.regs[VCPU_REGS_RDX];
2406 set_ghcb_msr_bits(svm, cpuid_value,
2407 GHCB_MSR_CPUID_VALUE_MASK,
2408 GHCB_MSR_CPUID_VALUE_POS);
2410 set_ghcb_msr_bits(svm, GHCB_MSR_CPUID_RESP,
2415 case GHCB_MSR_TERM_REQ: {
2416 u64 reason_set, reason_code;
2418 reason_set = get_ghcb_msr_bits(svm,
2419 GHCB_MSR_TERM_REASON_SET_MASK,
2420 GHCB_MSR_TERM_REASON_SET_POS);
2421 reason_code = get_ghcb_msr_bits(svm,
2422 GHCB_MSR_TERM_REASON_MASK,
2423 GHCB_MSR_TERM_REASON_POS);
2424 pr_info("SEV-ES guest requested termination: %#llx:%#llx\n",
2425 reason_set, reason_code);
2432 trace_kvm_vmgexit_msr_protocol_exit(svm->vcpu.vcpu_id,
2433 control->ghcb_gpa, ret);
2438 int sev_handle_vmgexit(struct kvm_vcpu *vcpu)
2440 struct vcpu_svm *svm = to_svm(vcpu);
2441 struct vmcb_control_area *control = &svm->vmcb->control;
2442 u64 ghcb_gpa, exit_code;
2446 /* Validate the GHCB */
2447 ghcb_gpa = control->ghcb_gpa;
2448 if (ghcb_gpa & GHCB_MSR_INFO_MASK)
2449 return sev_handle_vmgexit_msr_protocol(svm);
2452 vcpu_unimpl(vcpu, "vmgexit: GHCB gpa is not set\n");
2456 if (kvm_vcpu_map(vcpu, ghcb_gpa >> PAGE_SHIFT, &svm->ghcb_map)) {
2457 /* Unable to map GHCB from guest */
2458 vcpu_unimpl(vcpu, "vmgexit: error mapping GHCB [%#llx] from guest\n",
2463 svm->ghcb = svm->ghcb_map.hva;
2464 ghcb = svm->ghcb_map.hva;
2466 trace_kvm_vmgexit_enter(vcpu->vcpu_id, ghcb);
2468 exit_code = ghcb_get_sw_exit_code(ghcb);
2470 ret = sev_es_validate_vmgexit(svm);
2474 sev_es_sync_from_ghcb(svm);
2475 ghcb_set_sw_exit_info_1(ghcb, 0);
2476 ghcb_set_sw_exit_info_2(ghcb, 0);
2479 switch (exit_code) {
2480 case SVM_VMGEXIT_MMIO_READ:
2481 if (!setup_vmgexit_scratch(svm, true, control->exit_info_2))
2484 ret = kvm_sev_es_mmio_read(vcpu,
2485 control->exit_info_1,
2486 control->exit_info_2,
2489 case SVM_VMGEXIT_MMIO_WRITE:
2490 if (!setup_vmgexit_scratch(svm, false, control->exit_info_2))
2493 ret = kvm_sev_es_mmio_write(vcpu,
2494 control->exit_info_1,
2495 control->exit_info_2,
2498 case SVM_VMGEXIT_NMI_COMPLETE:
2499 ret = svm_invoke_exit_handler(vcpu, SVM_EXIT_IRET);
2501 case SVM_VMGEXIT_AP_HLT_LOOP:
2502 ret = kvm_emulate_ap_reset_hold(vcpu);
2504 case SVM_VMGEXIT_AP_JUMP_TABLE: {
2505 struct kvm_sev_info *sev = &to_kvm_svm(vcpu->kvm)->sev_info;
2507 switch (control->exit_info_1) {
2509 /* Set AP jump table address */
2510 sev->ap_jump_table = control->exit_info_2;
2513 /* Get AP jump table address */
2514 ghcb_set_sw_exit_info_2(ghcb, sev->ap_jump_table);
2517 pr_err("svm: vmgexit: unsupported AP jump table request - exit_info_1=%#llx\n",
2518 control->exit_info_1);
2519 ghcb_set_sw_exit_info_1(ghcb, 1);
2520 ghcb_set_sw_exit_info_2(ghcb,
2522 SVM_EVTINJ_TYPE_EXEPT |
2529 case SVM_VMGEXIT_UNSUPPORTED_EVENT:
2531 "vmgexit: unsupported event - exit_info_1=%#llx, exit_info_2=%#llx\n",
2532 control->exit_info_1, control->exit_info_2);
2535 ret = svm_invoke_exit_handler(vcpu, exit_code);
2541 int sev_es_string_io(struct vcpu_svm *svm, int size, unsigned int port, int in)
2543 if (!setup_vmgexit_scratch(svm, in, svm->vmcb->control.exit_info_2))
2546 return kvm_sev_es_string_io(&svm->vcpu, size, port,
2547 svm->ghcb_sa, svm->ghcb_sa_len, in);
2550 void sev_es_init_vmcb(struct vcpu_svm *svm)
2552 struct kvm_vcpu *vcpu = &svm->vcpu;
2554 svm->vmcb->control.nested_ctl |= SVM_NESTED_CTL_SEV_ES_ENABLE;
2555 svm->vmcb->control.virt_ext |= LBR_CTL_ENABLE_MASK;
2558 * An SEV-ES guest requires a VMSA area that is a separate from the
2559 * VMCB page. Do not include the encryption mask on the VMSA physical
2560 * address since hardware will access it using the guest key.
2562 svm->vmcb->control.vmsa_pa = __pa(svm->vmsa);
2564 /* Can't intercept CR register access, HV can't modify CR registers */
2565 svm_clr_intercept(svm, INTERCEPT_CR0_READ);
2566 svm_clr_intercept(svm, INTERCEPT_CR4_READ);
2567 svm_clr_intercept(svm, INTERCEPT_CR8_READ);
2568 svm_clr_intercept(svm, INTERCEPT_CR0_WRITE);
2569 svm_clr_intercept(svm, INTERCEPT_CR4_WRITE);
2570 svm_clr_intercept(svm, INTERCEPT_CR8_WRITE);
2572 svm_clr_intercept(svm, INTERCEPT_SELECTIVE_CR0);
2574 /* Track EFER/CR register changes */
2575 svm_set_intercept(svm, TRAP_EFER_WRITE);
2576 svm_set_intercept(svm, TRAP_CR0_WRITE);
2577 svm_set_intercept(svm, TRAP_CR4_WRITE);
2578 svm_set_intercept(svm, TRAP_CR8_WRITE);
2580 /* No support for enable_vmware_backdoor */
2581 clr_exception_intercept(svm, GP_VECTOR);
2583 /* Can't intercept XSETBV, HV can't modify XCR0 directly */
2584 svm_clr_intercept(svm, INTERCEPT_XSETBV);
2586 /* Clear intercepts on selected MSRs */
2587 set_msr_interception(vcpu, svm->msrpm, MSR_EFER, 1, 1);
2588 set_msr_interception(vcpu, svm->msrpm, MSR_IA32_CR_PAT, 1, 1);
2589 set_msr_interception(vcpu, svm->msrpm, MSR_IA32_LASTBRANCHFROMIP, 1, 1);
2590 set_msr_interception(vcpu, svm->msrpm, MSR_IA32_LASTBRANCHTOIP, 1, 1);
2591 set_msr_interception(vcpu, svm->msrpm, MSR_IA32_LASTINTFROMIP, 1, 1);
2592 set_msr_interception(vcpu, svm->msrpm, MSR_IA32_LASTINTTOIP, 1, 1);
2595 void sev_es_create_vcpu(struct vcpu_svm *svm)
2598 * Set the GHCB MSR value as per the GHCB specification when creating
2599 * a vCPU for an SEV-ES guest.
2601 set_ghcb_msr(svm, GHCB_MSR_SEV_INFO(GHCB_VERSION_MAX,
2606 void sev_es_prepare_guest_switch(struct vcpu_svm *svm, unsigned int cpu)
2608 struct svm_cpu_data *sd = per_cpu(svm_data, cpu);
2609 struct vmcb_save_area *hostsa;
2612 * As an SEV-ES guest, hardware will restore the host state on VMEXIT,
2613 * of which one step is to perform a VMLOAD. Since hardware does not
2614 * perform a VMSAVE on VMRUN, the host savearea must be updated.
2616 vmsave(__sme_page_pa(sd->save_area));
2618 /* XCR0 is restored on VMEXIT, save the current host value */
2619 hostsa = (struct vmcb_save_area *)(page_address(sd->save_area) + 0x400);
2620 hostsa->xcr0 = xgetbv(XCR_XFEATURE_ENABLED_MASK);
2622 /* PKRU is restored on VMEXIT, save the current host value */
2623 hostsa->pkru = read_pkru();
2625 /* MSR_IA32_XSS is restored on VMEXIT, save the currnet host value */
2626 hostsa->xss = host_xss;
2629 void sev_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
2631 struct vcpu_svm *svm = to_svm(vcpu);
2633 /* First SIPI: Use the values as initially set by the VMM */
2634 if (!svm->received_first_sipi) {
2635 svm->received_first_sipi = true;
2640 * Subsequent SIPI: Return from an AP Reset Hold VMGEXIT, where
2641 * the guest will set the CS and RIP. Set SW_EXIT_INFO_2 to a
2647 ghcb_set_sw_exit_info_2(svm->ghcb, 1);