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 #if IS_ENABLED(CONFIG_KVM)
  27 int prot_virt_host;
  28 EXPORT_SYMBOL(prot_virt_host);
  29 struct uv_info __bootdata_preserved(uv_info);
  30 EXPORT_SYMBOL(uv_info);
  31
  32 static int __init prot_virt_setup(char *val)
  33 {
  34         bool enabled;
  35         int rc;
  36
  37         rc = kstrtobool(val, &enabled);
  38         if (!rc && enabled)
  39                 prot_virt_host = 1;
  40
  41         if (is_prot_virt_guest() && prot_virt_host) {
  42                 prot_virt_host = 0;
  43                 pr_warn("Protected virtualization not available in protected guests.");
  44         }
  45
  46         if (prot_virt_host && !test_facility(158)) {
  47                 prot_virt_host = 0;
  48                 pr_warn("Protected virtualization not supported by the hardware.");
  49         }
  50
  51         return rc;
  52 }
  53 early_param("prot_virt", prot_virt_setup);
  54
  55 static int __init uv_init(unsigned long stor_base, unsigned long stor_len)
  56 {
  57         struct uv_cb_init uvcb = {
  58                 .header.cmd = UVC_CMD_INIT_UV,
  59                 .header.len = sizeof(uvcb),
  60                 .stor_origin = stor_base,
  61                 .stor_len = stor_len,
  62         };
  63
  64         if (uv_call(0, (uint64_t)&uvcb)) {
  65                 pr_err("Ultravisor init failed with rc: 0x%x rrc: 0%x\n",
  66                        uvcb.header.rc, uvcb.header.rrc);
  67                 return -1;
  68         }
  69         return 0;
  70 }
  71
  72 void __init setup_uv(void)
  73 {
  74         unsigned long uv_stor_base;
  75
  76         uv_stor_base = (unsigned long)memblock_alloc_try_nid(
  77                 uv_info.uv_base_stor_len, SZ_1M, SZ_2G,
  78                 MEMBLOCK_ALLOC_ACCESSIBLE, NUMA_NO_NODE);
  79         if (!uv_stor_base) {
  80                 pr_warn("Failed to reserve %lu bytes for ultravisor base storage\n",
  81                         uv_info.uv_base_stor_len);
  82                 goto fail;
  83         }
  84
  85         if (uv_init(uv_stor_base, uv_info.uv_base_stor_len)) {
  86                 memblock_free(uv_stor_base, uv_info.uv_base_stor_len);
  87                 goto fail;
  88         }
  89
  90         pr_info("Reserving %luMB as ultravisor base storage\n",
  91                 uv_info.uv_base_stor_len >> 20);
  92         return;
  93 fail:
  94         pr_info("Disabling support for protected virtualization");
  95         prot_virt_host = 0;
  96 }
  97
  98 void adjust_to_uv_max(unsigned long *vmax)
  99 {
 100         *vmax = min_t(unsigned long, *vmax, uv_info.max_sec_stor_addr);
 101 }
 102
 103 /*
 104  * Requests the Ultravisor to pin the page in the shared state. This will
 105  * cause an intercept when the guest attempts to unshare the pinned page.
 106  */
 107 static int uv_pin_shared(unsigned long paddr)
 108 {
 109         struct uv_cb_cfs uvcb = {
 110                 .header.cmd = UVC_CMD_PIN_PAGE_SHARED,
 111                 .header.len = sizeof(uvcb),
 112                 .paddr = paddr,
 113         };
 114
 115         if (uv_call(0, (u64)&uvcb))
 116                 return -EINVAL;
 117         return 0;
 118 }
 119
 120 /*
 121  * Requests the Ultravisor to encrypt a guest page and make it
 122  * accessible to the host for paging (export).
 123  *
 124  * @paddr: Absolute host address of page to be exported
 125  */
 126 int uv_convert_from_secure(unsigned long paddr)
 127 {
 128         struct uv_cb_cfs uvcb = {
 129                 .header.cmd = UVC_CMD_CONV_FROM_SEC_STOR,
 130                 .header.len = sizeof(uvcb),
 131                 .paddr = paddr
 132         };
 133
 134         if (uv_call(0, (u64)&uvcb))
 135                 return -EINVAL;
 136         return 0;
 137 }
 138
 139 /*
 140  * Calculate the expected ref_count for a page that would otherwise have no
 141  * further pins. This was cribbed from similar functions in other places in
 142  * the kernel, but with some slight modifications. We know that a secure
 143  * page can not be a huge page for example.
 144  */
 145 static int expected_page_refs(struct page *page)
 146 {
 147         int res;
 148
 149         res = page_mapcount(page);
 150         if (PageSwapCache(page)) {
 151                 res++;
 152         } else if (page_mapping(page)) {
 153                 res++;
 154                 if (page_has_private(page))
 155                         res++;
 156         }
 157         return res;
 158 }
 159
 160 static int make_secure_pte(pte_t *ptep, unsigned long addr,
 161                            struct page *exp_page, struct uv_cb_header *uvcb)
 162 {
 163         pte_t entry = READ_ONCE(*ptep);
 164         struct page *page;
 165         int expected, rc = 0;
 166
 167         if (!pte_present(entry))
 168                 return -ENXIO;
 169         if (pte_val(entry) & _PAGE_INVALID)
 170                 return -ENXIO;
 171
 172         page = pte_page(entry);
 173         if (page != exp_page)
 174                 return -ENXIO;
 175         if (PageWriteback(page))
 176                 return -EAGAIN;
 177         expected = expected_page_refs(page);
 178         if (!page_ref_freeze(page, expected))
 179                 return -EBUSY;
 180         set_bit(PG_arch_1, &page->flags);
 181         rc = uv_call(0, (u64)uvcb);
 182         page_ref_unfreeze(page, expected);
 183         /* Return -ENXIO if the page was not mapped, -EINVAL otherwise */
 184         if (rc)
 185                 rc = uvcb->rc == 0x10a ? -ENXIO : -EINVAL;
 186         return rc;
 187 }
 188
 189 /*
 190  * Requests the Ultravisor to make a page accessible to a guest.
 191  * If it's brought in the first time, it will be cleared. If
 192  * it has been exported before, it will be decrypted and integrity
 193  * checked.
 194  */
 195 int gmap_make_secure(struct gmap *gmap, unsigned long gaddr, void *uvcb)
 196 {
 197         struct vm_area_struct *vma;
 198         bool local_drain = false;
 199         spinlock_t *ptelock;
 200         unsigned long uaddr;
 201         struct page *page;
 202         pte_t *ptep;
 203         int rc;
 204
 205 again:
 206         rc = -EFAULT;
 207         down_read(&gmap->mm->mmap_sem);
 208
 209         uaddr = __gmap_translate(gmap, gaddr);
 210         if (IS_ERR_VALUE(uaddr))
 211                 goto out;
 212         vma = find_vma(gmap->mm, uaddr);
 213         if (!vma)
 214                 goto out;
 215         /*
 216          * Secure pages cannot be huge and userspace should not combine both.
 217          * In case userspace does it anyway this will result in an -EFAULT for
 218          * the unpack. The guest is thus never reaching secure mode. If
 219          * userspace is playing dirty tricky with mapping huge pages later
 220          * on this will result in a segmentation fault.
 221          */
 222         if (is_vm_hugetlb_page(vma))
 223                 goto out;
 224
 225         rc = -ENXIO;
 226         page = follow_page(vma, uaddr, FOLL_WRITE);
 227         if (IS_ERR_OR_NULL(page))
 228                 goto out;
 229
 230         lock_page(page);
 231         ptep = get_locked_pte(gmap->mm, uaddr, &ptelock);
 232         rc = make_secure_pte(ptep, uaddr, page, uvcb);
 233         pte_unmap_unlock(ptep, ptelock);
 234         unlock_page(page);
 235 out:
 236         up_read(&gmap->mm->mmap_sem);
 237
 238         if (rc == -EAGAIN) {
 239                 wait_on_page_writeback(page);
 240         } else if (rc == -EBUSY) {
 241                 /*
 242                  * If we have tried a local drain and the page refcount
 243                  * still does not match our expected safe value, try with a
 244                  * system wide drain. This is needed if the pagevecs holding
 245                  * the page are on a different CPU.
 246                  */
 247                 if (local_drain) {
 248                         lru_add_drain_all();
 249                         /* We give up here, and let the caller try again */
 250                         return -EAGAIN;
 251                 }
 252                 /*
 253                  * We are here if the page refcount does not match the
 254                  * expected safe value. The main culprits are usually
 255                  * pagevecs. With lru_add_drain() we drain the pagevecs
 256                  * on the local CPU so that hopefully the refcount will
 257                  * reach the expected safe value.
 258                  */
 259                 lru_add_drain();
 260                 local_drain = true;
 261                 /* And now we try again immediately after draining */
 262                 goto again;
 263         } else if (rc == -ENXIO) {
 264                 if (gmap_fault(gmap, gaddr, FAULT_FLAG_WRITE))
 265                         return -EFAULT;
 266                 return -EAGAIN;
 267         }
 268         return rc;
 269 }
 270 EXPORT_SYMBOL_GPL(gmap_make_secure);
 271
 272 int gmap_convert_to_secure(struct gmap *gmap, unsigned long gaddr)
 273 {
 274         struct uv_cb_cts uvcb = {
 275                 .header.cmd = UVC_CMD_CONV_TO_SEC_STOR,
 276                 .header.len = sizeof(uvcb),
 277                 .guest_handle = gmap->guest_handle,
 278                 .gaddr = gaddr,
 279         };
 280
 281         return gmap_make_secure(gmap, gaddr, &uvcb);
 282 }
 283 EXPORT_SYMBOL_GPL(gmap_convert_to_secure);
 284
 285 /*
 286  * To be called with the page locked or with an extra reference! This will
 287  * prevent gmap_make_secure from touching the page concurrently. Having 2
 288  * parallel make_page_accessible is fine, as the UV calls will become a
 289  * no-op if the page is already exported.
 290  */
 291 int arch_make_page_accessible(struct page *page)
 292 {
 293         int rc = 0;
 294
 295         /* Hugepage cannot be protected, so nothing to do */
 296         if (PageHuge(page))
 297                 return 0;
 298
 299         /*
 300          * PG_arch_1 is used in 3 places:
 301          * 1. for kernel page tables during early boot
 302          * 2. for storage keys of huge pages and KVM
 303          * 3. As an indication that this page might be secure. This can
 304          *    overindicate, e.g. we set the bit before calling
 305          *    convert_to_secure.
 306          * As secure pages are never huge, all 3 variants can co-exists.
 307          */
 308         if (!test_bit(PG_arch_1, &page->flags))
 309                 return 0;
 310
 311         rc = uv_pin_shared(page_to_phys(page));
 312         if (!rc) {
 313                 clear_bit(PG_arch_1, &page->flags);
 314                 return 0;
 315         }
 316
 317         rc = uv_convert_from_secure(page_to_phys(page));
 318         if (!rc) {
 319                 clear_bit(PG_arch_1, &page->flags);
 320                 return 0;
 321         }
 322
 323         return rc;
 324 }
 325 EXPORT_SYMBOL_GPL(arch_make_page_accessible);
 326
 327 #endif
 328
 329 #if defined(CONFIG_PROTECTED_VIRTUALIZATION_GUEST) || IS_ENABLED(CONFIG_KVM)
 330 static ssize_t uv_query_facilities(struct kobject *kobj,
 331                                    struct kobj_attribute *attr, char *page)
 332 {
 333         return snprintf(page, PAGE_SIZE, "%lx\n%lx\n%lx\n%lx\n",
 334                         uv_info.inst_calls_list[0],
 335                         uv_info.inst_calls_list[1],
 336                         uv_info.inst_calls_list[2],
 337                         uv_info.inst_calls_list[3]);
 338 }
 339
 340 static struct kobj_attribute uv_query_facilities_attr =
 341         __ATTR(facilities, 0444, uv_query_facilities, NULL);
 342
 343 static ssize_t uv_query_max_guest_cpus(struct kobject *kobj,
 344                                        struct kobj_attribute *attr, char *page)
 345 {
 346         return snprintf(page, PAGE_SIZE, "%d\n",
 347                         uv_info.max_guest_cpus);
 348 }
 349
 350 static struct kobj_attribute uv_query_max_guest_cpus_attr =
 351         __ATTR(max_cpus, 0444, uv_query_max_guest_cpus, NULL);
 352
 353 static ssize_t uv_query_max_guest_vms(struct kobject *kobj,
 354                                       struct kobj_attribute *attr, char *page)
 355 {
 356         return snprintf(page, PAGE_SIZE, "%d\n",
 357                         uv_info.max_num_sec_conf);
 358 }
 359
 360 static struct kobj_attribute uv_query_max_guest_vms_attr =
 361         __ATTR(max_guests, 0444, uv_query_max_guest_vms, NULL);
 362
 363 static ssize_t uv_query_max_guest_addr(struct kobject *kobj,
 364                                        struct kobj_attribute *attr, char *page)
 365 {
 366         return snprintf(page, PAGE_SIZE, "%lx\n",
 367                         uv_info.max_sec_stor_addr);
 368 }
 369
 370 static struct kobj_attribute uv_query_max_guest_addr_attr =
 371         __ATTR(max_address, 0444, uv_query_max_guest_addr, NULL);
 372
 373 static struct attribute *uv_query_attrs[] = {
 374         &uv_query_facilities_attr.attr,
 375         &uv_query_max_guest_cpus_attr.attr,
 376         &uv_query_max_guest_vms_attr.attr,
 377         &uv_query_max_guest_addr_attr.attr,
 378         NULL,
 379 };
 380
 381 static struct attribute_group uv_query_attr_group = {
 382         .attrs = uv_query_attrs,
 383 };
 384
 385 static struct kset *uv_query_kset;
 386 static struct kobject *uv_kobj;
 387
 388 static int __init uv_info_init(void)
 389 {
 390         int rc = -ENOMEM;
 391
 392         if (!test_facility(158))
 393                 return 0;
 394
 395         uv_kobj = kobject_create_and_add("uv", firmware_kobj);
 396         if (!uv_kobj)
 397                 return -ENOMEM;
 398
 399         uv_query_kset = kset_create_and_add("query", NULL, uv_kobj);
 400         if (!uv_query_kset)
 401                 goto out_kobj;
 402
 403         rc = sysfs_create_group(&uv_query_kset->kobj, &uv_query_attr_group);
 404         if (!rc)
 405                 return 0;
 406
 407         kset_unregister(uv_query_kset);
 408 out_kobj:
 409         kobject_del(uv_kobj);
 410         kobject_put(uv_kobj);
 411         return rc;
 412 }
 413 device_initcall(uv_info_init);
 414 #endif