1 // SPDX-License-Identifier: GPL-2.0
3 * Common Ultravisor functions and initialization
5 * Copyright IBM Corp. 2019, 2020
7 #define KMSG_COMPONENT "prot_virt"
8 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
10 #include <linux/kernel.h>
11 #include <linux/types.h>
12 #include <linux/sizes.h>
13 #include <linux/bitmap.h>
14 #include <linux/memblock.h>
15 #include <linux/pagemap.h>
16 #include <linux/swap.h>
17 #include <asm/facility.h>
18 #include <asm/sections.h>
21 /* the bootdata_preserved fields come from ones in arch/s390/boot/uv.c */
22 #ifdef CONFIG_PROTECTED_VIRTUALIZATION_GUEST
23 int __bootdata_preserved(prot_virt_guest);
26 struct uv_info __bootdata_preserved(uv_info);
28 #if IS_ENABLED(CONFIG_KVM)
29 int __bootdata_preserved(prot_virt_host);
30 EXPORT_SYMBOL(prot_virt_host);
31 EXPORT_SYMBOL(uv_info);
33 static int __init uv_init(phys_addr_t stor_base, unsigned long stor_len)
35 struct uv_cb_init uvcb = {
36 .header.cmd = UVC_CMD_INIT_UV,
37 .header.len = sizeof(uvcb),
38 .stor_origin = stor_base,
42 if (uv_call(0, (uint64_t)&uvcb)) {
43 pr_err("Ultravisor init failed with rc: 0x%x rrc: 0%x\n",
44 uvcb.header.rc, uvcb.header.rrc);
50 void __init setup_uv(void)
54 if (!is_prot_virt_host())
57 uv_stor_base = memblock_alloc_try_nid(
58 uv_info.uv_base_stor_len, SZ_1M, SZ_2G,
59 MEMBLOCK_ALLOC_ACCESSIBLE, NUMA_NO_NODE);
61 pr_warn("Failed to reserve %lu bytes for ultravisor base storage\n",
62 uv_info.uv_base_stor_len);
66 if (uv_init(__pa(uv_stor_base), uv_info.uv_base_stor_len)) {
67 memblock_free(uv_stor_base, uv_info.uv_base_stor_len);
71 pr_info("Reserving %luMB as ultravisor base storage\n",
72 uv_info.uv_base_stor_len >> 20);
75 pr_info("Disabling support for protected virtualization");
80 * Requests the Ultravisor to pin the page in the shared state. This will
81 * cause an intercept when the guest attempts to unshare the pinned page.
83 static int uv_pin_shared(unsigned long paddr)
85 struct uv_cb_cfs uvcb = {
86 .header.cmd = UVC_CMD_PIN_PAGE_SHARED,
87 .header.len = sizeof(uvcb),
91 if (uv_call(0, (u64)&uvcb))
97 * Requests the Ultravisor to destroy a guest page and make it
98 * accessible to the host. The destroy clears the page instead of
101 * @paddr: Absolute host address of page to be destroyed
103 static int uv_destroy_page(unsigned long paddr)
105 struct uv_cb_cfs uvcb = {
106 .header.cmd = UVC_CMD_DESTR_SEC_STOR,
107 .header.len = sizeof(uvcb),
111 if (uv_call(0, (u64)&uvcb)) {
113 * Older firmware uses 107/d as an indication of a non secure
114 * page. Let us emulate the newer variant (no-op).
116 if (uvcb.header.rc == 0x107 && uvcb.header.rrc == 0xd)
124 * The caller must already hold a reference to the page
126 int uv_destroy_owned_page(unsigned long paddr)
128 struct page *page = phys_to_page(paddr);
132 rc = uv_destroy_page(paddr);
134 clear_bit(PG_arch_1, &page->flags);
140 * Requests the Ultravisor to encrypt a guest page and make it
141 * accessible to the host for paging (export).
143 * @paddr: Absolute host address of page to be exported
145 int uv_convert_from_secure(unsigned long paddr)
147 struct uv_cb_cfs uvcb = {
148 .header.cmd = UVC_CMD_CONV_FROM_SEC_STOR,
149 .header.len = sizeof(uvcb),
153 if (uv_call(0, (u64)&uvcb))
159 * The caller must already hold a reference to the page
161 int uv_convert_owned_from_secure(unsigned long paddr)
163 struct page *page = phys_to_page(paddr);
167 rc = uv_convert_from_secure(paddr);
169 clear_bit(PG_arch_1, &page->flags);
175 * Calculate the expected ref_count for a page that would otherwise have no
176 * further pins. This was cribbed from similar functions in other places in
177 * the kernel, but with some slight modifications. We know that a secure
178 * page can not be a huge page for example.
180 static int expected_page_refs(struct page *page)
184 res = page_mapcount(page);
185 if (PageSwapCache(page)) {
187 } else if (page_mapping(page)) {
189 if (page_has_private(page))
195 static int make_secure_pte(pte_t *ptep, unsigned long addr,
196 struct page *exp_page, struct uv_cb_header *uvcb)
198 pte_t entry = READ_ONCE(*ptep);
200 int expected, cc = 0;
202 if (!pte_present(entry))
204 if (pte_val(entry) & _PAGE_INVALID)
207 page = pte_page(entry);
208 if (page != exp_page)
210 if (PageWriteback(page))
212 expected = expected_page_refs(page);
213 if (!page_ref_freeze(page, expected))
215 set_bit(PG_arch_1, &page->flags);
217 * If the UVC does not succeed or fail immediately, we don't want to
218 * loop for long, or we might get stall notifications.
219 * On the other hand, this is a complex scenario and we are holding a lot of
220 * locks, so we can't easily sleep and reschedule. We try only once,
221 * and if the UVC returned busy or partial completion, we return
222 * -EAGAIN and we let the callers deal with it.
224 cc = __uv_call(0, (u64)uvcb);
225 page_ref_unfreeze(page, expected);
227 * Return -ENXIO if the page was not mapped, -EINVAL for other errors.
228 * If busy or partially completed, return -EAGAIN.
232 else if (cc == UVC_CC_BUSY || cc == UVC_CC_PARTIAL)
234 return uvcb->rc == 0x10a ? -ENXIO : -EINVAL;
238 * Requests the Ultravisor to make a page accessible to a guest.
239 * If it's brought in the first time, it will be cleared. If
240 * it has been exported before, it will be decrypted and integrity
243 int gmap_make_secure(struct gmap *gmap, unsigned long gaddr, void *uvcb)
245 struct vm_area_struct *vma;
246 bool local_drain = false;
255 mmap_read_lock(gmap->mm);
257 uaddr = __gmap_translate(gmap, gaddr);
258 if (IS_ERR_VALUE(uaddr))
260 vma = vma_lookup(gmap->mm, uaddr);
264 * Secure pages cannot be huge and userspace should not combine both.
265 * In case userspace does it anyway this will result in an -EFAULT for
266 * the unpack. The guest is thus never reaching secure mode. If
267 * userspace is playing dirty tricky with mapping huge pages later
268 * on this will result in a segmentation fault.
270 if (is_vm_hugetlb_page(vma))
274 page = follow_page(vma, uaddr, FOLL_WRITE);
275 if (IS_ERR_OR_NULL(page))
279 ptep = get_locked_pte(gmap->mm, uaddr, &ptelock);
280 rc = make_secure_pte(ptep, uaddr, page, uvcb);
281 pte_unmap_unlock(ptep, ptelock);
284 mmap_read_unlock(gmap->mm);
288 * If we are here because the UVC returned busy or partial
289 * completion, this is just a useless check, but it is safe.
291 wait_on_page_writeback(page);
292 } else if (rc == -EBUSY) {
294 * If we have tried a local drain and the page refcount
295 * still does not match our expected safe value, try with a
296 * system wide drain. This is needed if the pagevecs holding
297 * the page are on a different CPU.
301 /* We give up here, and let the caller try again */
305 * We are here if the page refcount does not match the
306 * expected safe value. The main culprits are usually
307 * pagevecs. With lru_add_drain() we drain the pagevecs
308 * on the local CPU so that hopefully the refcount will
309 * reach the expected safe value.
313 /* And now we try again immediately after draining */
315 } else if (rc == -ENXIO) {
316 if (gmap_fault(gmap, gaddr, FAULT_FLAG_WRITE))
322 EXPORT_SYMBOL_GPL(gmap_make_secure);
324 int gmap_convert_to_secure(struct gmap *gmap, unsigned long gaddr)
326 struct uv_cb_cts uvcb = {
327 .header.cmd = UVC_CMD_CONV_TO_SEC_STOR,
328 .header.len = sizeof(uvcb),
329 .guest_handle = gmap->guest_handle,
333 return gmap_make_secure(gmap, gaddr, &uvcb);
335 EXPORT_SYMBOL_GPL(gmap_convert_to_secure);
338 * To be called with the page locked or with an extra reference! This will
339 * prevent gmap_make_secure from touching the page concurrently. Having 2
340 * parallel make_page_accessible is fine, as the UV calls will become a
341 * no-op if the page is already exported.
343 int arch_make_page_accessible(struct page *page)
347 /* Hugepage cannot be protected, so nothing to do */
352 * PG_arch_1 is used in 3 places:
353 * 1. for kernel page tables during early boot
354 * 2. for storage keys of huge pages and KVM
355 * 3. As an indication that this page might be secure. This can
356 * overindicate, e.g. we set the bit before calling
358 * As secure pages are never huge, all 3 variants can co-exists.
360 if (!test_bit(PG_arch_1, &page->flags))
363 rc = uv_pin_shared(page_to_phys(page));
365 clear_bit(PG_arch_1, &page->flags);
369 rc = uv_convert_from_secure(page_to_phys(page));
371 clear_bit(PG_arch_1, &page->flags);
377 EXPORT_SYMBOL_GPL(arch_make_page_accessible);
381 #if defined(CONFIG_PROTECTED_VIRTUALIZATION_GUEST) || IS_ENABLED(CONFIG_KVM)
382 static ssize_t uv_query_facilities(struct kobject *kobj,
383 struct kobj_attribute *attr, char *page)
385 return scnprintf(page, PAGE_SIZE, "%lx\n%lx\n%lx\n%lx\n",
386 uv_info.inst_calls_list[0],
387 uv_info.inst_calls_list[1],
388 uv_info.inst_calls_list[2],
389 uv_info.inst_calls_list[3]);
392 static struct kobj_attribute uv_query_facilities_attr =
393 __ATTR(facilities, 0444, uv_query_facilities, NULL);
395 static ssize_t uv_query_feature_indications(struct kobject *kobj,
396 struct kobj_attribute *attr, char *buf)
398 return sysfs_emit(buf, "%lx\n", uv_info.uv_feature_indications);
401 static struct kobj_attribute uv_query_feature_indications_attr =
402 __ATTR(feature_indications, 0444, uv_query_feature_indications, NULL);
404 static ssize_t uv_query_max_guest_cpus(struct kobject *kobj,
405 struct kobj_attribute *attr, char *page)
407 return scnprintf(page, PAGE_SIZE, "%d\n",
408 uv_info.max_guest_cpu_id + 1);
411 static struct kobj_attribute uv_query_max_guest_cpus_attr =
412 __ATTR(max_cpus, 0444, uv_query_max_guest_cpus, NULL);
414 static ssize_t uv_query_max_guest_vms(struct kobject *kobj,
415 struct kobj_attribute *attr, char *page)
417 return scnprintf(page, PAGE_SIZE, "%d\n",
418 uv_info.max_num_sec_conf);
421 static struct kobj_attribute uv_query_max_guest_vms_attr =
422 __ATTR(max_guests, 0444, uv_query_max_guest_vms, NULL);
424 static ssize_t uv_query_max_guest_addr(struct kobject *kobj,
425 struct kobj_attribute *attr, char *page)
427 return scnprintf(page, PAGE_SIZE, "%lx\n",
428 uv_info.max_sec_stor_addr);
431 static struct kobj_attribute uv_query_max_guest_addr_attr =
432 __ATTR(max_address, 0444, uv_query_max_guest_addr, NULL);
434 static struct attribute *uv_query_attrs[] = {
435 &uv_query_facilities_attr.attr,
436 &uv_query_feature_indications_attr.attr,
437 &uv_query_max_guest_cpus_attr.attr,
438 &uv_query_max_guest_vms_attr.attr,
439 &uv_query_max_guest_addr_attr.attr,
443 static struct attribute_group uv_query_attr_group = {
444 .attrs = uv_query_attrs,
447 static ssize_t uv_is_prot_virt_guest(struct kobject *kobj,
448 struct kobj_attribute *attr, char *page)
452 #ifdef CONFIG_PROTECTED_VIRTUALIZATION_GUEST
453 val = prot_virt_guest;
455 return scnprintf(page, PAGE_SIZE, "%d\n", val);
458 static ssize_t uv_is_prot_virt_host(struct kobject *kobj,
459 struct kobj_attribute *attr, char *page)
463 #if IS_ENABLED(CONFIG_KVM)
464 val = prot_virt_host;
467 return scnprintf(page, PAGE_SIZE, "%d\n", val);
470 static struct kobj_attribute uv_prot_virt_guest =
471 __ATTR(prot_virt_guest, 0444, uv_is_prot_virt_guest, NULL);
473 static struct kobj_attribute uv_prot_virt_host =
474 __ATTR(prot_virt_host, 0444, uv_is_prot_virt_host, NULL);
476 static const struct attribute *uv_prot_virt_attrs[] = {
477 &uv_prot_virt_guest.attr,
478 &uv_prot_virt_host.attr,
482 static struct kset *uv_query_kset;
483 static struct kobject *uv_kobj;
485 static int __init uv_info_init(void)
489 if (!test_facility(158))
492 uv_kobj = kobject_create_and_add("uv", firmware_kobj);
496 rc = sysfs_create_files(uv_kobj, uv_prot_virt_attrs);
500 uv_query_kset = kset_create_and_add("query", NULL, uv_kobj);
501 if (!uv_query_kset) {
506 rc = sysfs_create_group(&uv_query_kset->kobj, &uv_query_attr_group);
510 kset_unregister(uv_query_kset);
512 sysfs_remove_files(uv_kobj, uv_prot_virt_attrs);
514 kobject_del(uv_kobj);
515 kobject_put(uv_kobj);
518 device_initcall(uv_info_init);