arch/s390/kernel/uv.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 /*
   3  * Common Ultravisor functions and initialization
   4  *
   5  * Copyright IBM Corp. 2019, 2020
   6  */
   7 #define KMSG_COMPONENT "prot_virt"
   8 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
   9
  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>
  19 #include <asm/uv.h>
  20
  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);
  24 #endif
  25
  26 struct uv_info __bootdata_preserved(uv_info);
  27
  28 #if IS_ENABLED(CONFIG_KVM)
  29 int __bootdata_preserved(prot_virt_host);
  30 EXPORT_SYMBOL(prot_virt_host);
  31 EXPORT_SYMBOL(uv_info);
  32
  33 static int __init uv_init(unsigned long stor_base, unsigned long stor_len)
  34 {
  35         struct uv_cb_init uvcb = {
  36                 .header.cmd = UVC_CMD_INIT_UV,
  37                 .header.len = sizeof(uvcb),
  38                 .stor_origin = stor_base,
  39                 .stor_len = stor_len,
  40         };
  41
  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);
  45                 return -1;
  46         }
  47         return 0;
  48 }
  49
  50 void __init setup_uv(void)
  51 {
  52         unsigned long uv_stor_base;
  53
  54         /*
  55          * keep these conditions in line with kasan init code has_uv_sec_stor_limit()
  56          */
  57         if (!is_prot_virt_host())
  58                 return;
  59
  60         if (is_prot_virt_guest()) {
  61                 prot_virt_host = 0;
  62                 pr_warn("Protected virtualization not available in protected guests.");
  63                 return;
  64         }
  65
  66         if (!test_facility(158)) {
  67                 prot_virt_host = 0;
  68                 pr_warn("Protected virtualization not supported by the hardware.");
  69                 return;
  70         }
  71
  72         uv_stor_base = (unsigned long)memblock_alloc_try_nid(
  73                 uv_info.uv_base_stor_len, SZ_1M, SZ_2G,
  74                 MEMBLOCK_ALLOC_ACCESSIBLE, NUMA_NO_NODE);
  75         if (!uv_stor_base) {
  76                 pr_warn("Failed to reserve %lu bytes for ultravisor base storage\n",
  77                         uv_info.uv_base_stor_len);
  78                 goto fail;
  79         }
  80
  81         if (uv_init(uv_stor_base, uv_info.uv_base_stor_len)) {
  82                 memblock_free(uv_stor_base, uv_info.uv_base_stor_len);
  83                 goto fail;
  84         }
  85
  86         pr_info("Reserving %luMB as ultravisor base storage\n",
  87                 uv_info.uv_base_stor_len >> 20);
  88         return;
  89 fail:
  90         pr_info("Disabling support for protected virtualization");
  91         prot_virt_host = 0;
  92 }
  93
  94 void adjust_to_uv_max(unsigned long *vmax)
  95 {
  96         if (uv_info.max_sec_stor_addr)
  97                 *vmax = min_t(unsigned long, *vmax, uv_info.max_sec_stor_addr);
  98 }
  99
 100 /*
 101  * Requests the Ultravisor to pin the page in the shared state. This will
 102  * cause an intercept when the guest attempts to unshare the pinned page.
 103  */
 104 static int uv_pin_shared(unsigned long paddr)
 105 {
 106         struct uv_cb_cfs uvcb = {
 107                 .header.cmd = UVC_CMD_PIN_PAGE_SHARED,
 108                 .header.len = sizeof(uvcb),
 109                 .paddr = paddr,
 110         };
 111
 112         if (uv_call(0, (u64)&uvcb))
 113                 return -EINVAL;
 114         return 0;
 115 }
 116
 117 /*
 118  * Requests the Ultravisor to destroy a guest page and make it
 119  * accessible to the host. The destroy clears the page instead of
 120  * exporting.
 121  *
 122  * @paddr: Absolute host address of page to be destroyed
 123  */
 124 int uv_destroy_page(unsigned long paddr)
 125 {
 126         struct uv_cb_cfs uvcb = {
 127                 .header.cmd = UVC_CMD_DESTR_SEC_STOR,
 128                 .header.len = sizeof(uvcb),
 129                 .paddr = paddr
 130         };
 131
 132         if (uv_call(0, (u64)&uvcb)) {
 133                 /*
 134                  * Older firmware uses 107/d as an indication of a non secure
 135                  * page. Let us emulate the newer variant (no-op).
 136                  */
 137                 if (uvcb.header.rc == 0x107 && uvcb.header.rrc == 0xd)
 138                         return 0;
 139                 return -EINVAL;
 140         }
 141         return 0;
 142 }
 143
 144 /*
 145  * Requests the Ultravisor to encrypt a guest page and make it
 146  * accessible to the host for paging (export).
 147  *
 148  * @paddr: Absolute host address of page to be exported
 149  */
 150 int uv_convert_from_secure(unsigned long paddr)
 151 {
 152         struct uv_cb_cfs uvcb = {
 153                 .header.cmd = UVC_CMD_CONV_FROM_SEC_STOR,
 154                 .header.len = sizeof(uvcb),
 155                 .paddr = paddr
 156         };
 157
 158         if (uv_call(0, (u64)&uvcb))
 159                 return -EINVAL;
 160         return 0;
 161 }
 162
 163 /*
 164  * Calculate the expected ref_count for a page that would otherwise have no
 165  * further pins. This was cribbed from similar functions in other places in
 166  * the kernel, but with some slight modifications. We know that a secure
 167  * page can not be a huge page for example.
 168  */
 169 static int expected_page_refs(struct page *page)
 170 {
 171         int res;
 172
 173         res = page_mapcount(page);
 174         if (PageSwapCache(page)) {
 175                 res++;
 176         } else if (page_mapping(page)) {
 177                 res++;
 178                 if (page_has_private(page))
 179                         res++;
 180         }
 181         return res;
 182 }
 183
 184 static int make_secure_pte(pte_t *ptep, unsigned long addr,
 185                            struct page *exp_page, struct uv_cb_header *uvcb)
 186 {
 187         pte_t entry = READ_ONCE(*ptep);
 188         struct page *page;
 189         int expected, rc = 0;
 190
 191         if (!pte_present(entry))
 192                 return -ENXIO;
 193         if (pte_val(entry) & _PAGE_INVALID)
 194                 return -ENXIO;
 195
 196         page = pte_page(entry);
 197         if (page != exp_page)
 198                 return -ENXIO;
 199         if (PageWriteback(page))
 200                 return -EAGAIN;
 201         expected = expected_page_refs(page);
 202         if (!page_ref_freeze(page, expected))
 203                 return -EBUSY;
 204         set_bit(PG_arch_1, &page->flags);
 205         rc = uv_call(0, (u64)uvcb);
 206         page_ref_unfreeze(page, expected);
 207         /* Return -ENXIO if the page was not mapped, -EINVAL otherwise */
 208         if (rc)
 209                 rc = uvcb->rc == 0x10a ? -ENXIO : -EINVAL;
 210         return rc;
 211 }
 212
 213 /*
 214  * Requests the Ultravisor to make a page accessible to a guest.
 215  * If it's brought in the first time, it will be cleared. If
 216  * it has been exported before, it will be decrypted and integrity
 217  * checked.
 218  */
 219 int gmap_make_secure(struct gmap *gmap, unsigned long gaddr, void *uvcb)
 220 {
 221         struct vm_area_struct *vma;
 222         bool local_drain = false;
 223         spinlock_t *ptelock;
 224         unsigned long uaddr;
 225         struct page *page;
 226         pte_t *ptep;
 227         int rc;
 228
 229 again:
 230         rc = -EFAULT;
 231         mmap_read_lock(gmap->mm);
 232
 233         uaddr = __gmap_translate(gmap, gaddr);
 234         if (IS_ERR_VALUE(uaddr))
 235                 goto out;
 236         vma = find_vma(gmap->mm, uaddr);
 237         if (!vma)
 238                 goto out;
 239         /*
 240          * Secure pages cannot be huge and userspace should not combine both.
 241          * In case userspace does it anyway this will result in an -EFAULT for
 242          * the unpack. The guest is thus never reaching secure mode. If
 243          * userspace is playing dirty tricky with mapping huge pages later
 244          * on this will result in a segmentation fault.
 245          */
 246         if (is_vm_hugetlb_page(vma))
 247                 goto out;
 248
 249         rc = -ENXIO;
 250         page = follow_page(vma, uaddr, FOLL_WRITE);
 251         if (IS_ERR_OR_NULL(page))
 252                 goto out;
 253
 254         lock_page(page);
 255         ptep = get_locked_pte(gmap->mm, uaddr, &ptelock);
 256         rc = make_secure_pte(ptep, uaddr, page, uvcb);
 257         pte_unmap_unlock(ptep, ptelock);
 258         unlock_page(page);
 259 out:
 260         mmap_read_unlock(gmap->mm);
 261
 262         if (rc == -EAGAIN) {
 263                 wait_on_page_writeback(page);
 264         } else if (rc == -EBUSY) {
 265                 /*
 266                  * If we have tried a local drain and the page refcount
 267                  * still does not match our expected safe value, try with a
 268                  * system wide drain. This is needed if the pagevecs holding
 269                  * the page are on a different CPU.
 270                  */
 271                 if (local_drain) {
 272                         lru_add_drain_all();
 273                         /* We give up here, and let the caller try again */
 274                         return -EAGAIN;
 275                 }
 276                 /*
 277                  * We are here if the page refcount does not match the
 278                  * expected safe value. The main culprits are usually
 279                  * pagevecs. With lru_add_drain() we drain the pagevecs
 280                  * on the local CPU so that hopefully the refcount will
 281                  * reach the expected safe value.
 282                  */
 283                 lru_add_drain();
 284                 local_drain = true;
 285                 /* And now we try again immediately after draining */
 286                 goto again;
 287         } else if (rc == -ENXIO) {
 288                 if (gmap_fault(gmap, gaddr, FAULT_FLAG_WRITE))
 289                         return -EFAULT;
 290                 return -EAGAIN;
 291         }
 292         return rc;
 293 }
 294 EXPORT_SYMBOL_GPL(gmap_make_secure);
 295
 296 int gmap_convert_to_secure(struct gmap *gmap, unsigned long gaddr)
 297 {
 298         struct uv_cb_cts uvcb = {
 299                 .header.cmd = UVC_CMD_CONV_TO_SEC_STOR,
 300                 .header.len = sizeof(uvcb),
 301                 .guest_handle = gmap->guest_handle,
 302                 .gaddr = gaddr,
 303         };
 304
 305         return gmap_make_secure(gmap, gaddr, &uvcb);
 306 }
 307 EXPORT_SYMBOL_GPL(gmap_convert_to_secure);
 308
 309 /*
 310  * To be called with the page locked or with an extra reference! This will
 311  * prevent gmap_make_secure from touching the page concurrently. Having 2
 312  * parallel make_page_accessible is fine, as the UV calls will become a
 313  * no-op if the page is already exported.
 314  */
 315 int arch_make_page_accessible(struct page *page)
 316 {
 317         int rc = 0;
 318
 319         /* Hugepage cannot be protected, so nothing to do */
 320         if (PageHuge(page))
 321                 return 0;
 322
 323         /*
 324          * PG_arch_1 is used in 3 places:
 325          * 1. for kernel page tables during early boot
 326          * 2. for storage keys of huge pages and KVM
 327          * 3. As an indication that this page might be secure. This can
 328          *    overindicate, e.g. we set the bit before calling
 329          *    convert_to_secure.
 330          * As secure pages are never huge, all 3 variants can co-exists.
 331          */
 332         if (!test_bit(PG_arch_1, &page->flags))
 333                 return 0;
 334
 335         rc = uv_pin_shared(page_to_phys(page));
 336         if (!rc) {
 337                 clear_bit(PG_arch_1, &page->flags);
 338                 return 0;
 339         }
 340
 341         rc = uv_convert_from_secure(page_to_phys(page));
 342         if (!rc) {
 343                 clear_bit(PG_arch_1, &page->flags);
 344                 return 0;
 345         }
 346
 347         return rc;
 348 }
 349 EXPORT_SYMBOL_GPL(arch_make_page_accessible);
 350
 351 #endif
 352
 353 #if defined(CONFIG_PROTECTED_VIRTUALIZATION_GUEST) || IS_ENABLED(CONFIG_KVM)
 354 static ssize_t uv_query_facilities(struct kobject *kobj,
 355                                    struct kobj_attribute *attr, char *page)
 356 {
 357         return scnprintf(page, PAGE_SIZE, "%lx\n%lx\n%lx\n%lx\n",
 358                         uv_info.inst_calls_list[0],
 359                         uv_info.inst_calls_list[1],
 360                         uv_info.inst_calls_list[2],
 361                         uv_info.inst_calls_list[3]);
 362 }
 363
 364 static struct kobj_attribute uv_query_facilities_attr =
 365         __ATTR(facilities, 0444, uv_query_facilities, NULL);
 366
 367 static ssize_t uv_query_max_guest_cpus(struct kobject *kobj,
 368                                        struct kobj_attribute *attr, char *page)
 369 {
 370         return scnprintf(page, PAGE_SIZE, "%d\n",
 371                         uv_info.max_guest_cpus);
 372 }
 373
 374 static struct kobj_attribute uv_query_max_guest_cpus_attr =
 375         __ATTR(max_cpus, 0444, uv_query_max_guest_cpus, NULL);
 376
 377 static ssize_t uv_query_max_guest_vms(struct kobject *kobj,
 378                                       struct kobj_attribute *attr, char *page)
 379 {
 380         return scnprintf(page, PAGE_SIZE, "%d\n",
 381                         uv_info.max_num_sec_conf);
 382 }
 383
 384 static struct kobj_attribute uv_query_max_guest_vms_attr =
 385         __ATTR(max_guests, 0444, uv_query_max_guest_vms, NULL);
 386
 387 static ssize_t uv_query_max_guest_addr(struct kobject *kobj,
 388                                        struct kobj_attribute *attr, char *page)
 389 {
 390         return scnprintf(page, PAGE_SIZE, "%lx\n",
 391                         uv_info.max_sec_stor_addr);
 392 }
 393
 394 static struct kobj_attribute uv_query_max_guest_addr_attr =
 395         __ATTR(max_address, 0444, uv_query_max_guest_addr, NULL);
 396
 397 static struct attribute *uv_query_attrs[] = {
 398         &uv_query_facilities_attr.attr,
 399         &uv_query_max_guest_cpus_attr.attr,
 400         &uv_query_max_guest_vms_attr.attr,
 401         &uv_query_max_guest_addr_attr.attr,
 402         NULL,
 403 };
 404
 405 static struct attribute_group uv_query_attr_group = {
 406         .attrs = uv_query_attrs,
 407 };
 408
 409 static struct kset *uv_query_kset;
 410 static struct kobject *uv_kobj;
 411
 412 static int __init uv_info_init(void)
 413 {
 414         int rc = -ENOMEM;
 415
 416         if (!test_facility(158))
 417                 return 0;
 418
 419         uv_kobj = kobject_create_and_add("uv", firmware_kobj);
 420         if (!uv_kobj)
 421                 return -ENOMEM;
 422
 423         uv_query_kset = kset_create_and_add("query", NULL, uv_kobj);
 424         if (!uv_query_kset)
 425                 goto out_kobj;
 426
 427         rc = sysfs_create_group(&uv_query_kset->kobj, &uv_query_attr_group);
 428         if (!rc)
 429                 return 0;
 430
 431         kset_unregister(uv_query_kset);
 432 out_kobj:
 433         kobject_del(uv_kobj);
 434         kobject_put(uv_kobj);
 435         return rc;
 436 }
 437 device_initcall(uv_info_init);
 438 #endif