Khalid Aziz | c6202ca | 2018-02-21 10:15:48 -0700 | [diff] [blame] | 1 | /* adi_64.c: support for ADI (Application Data Integrity) feature on |
| 2 | * sparc m7 and newer processors. This feature is also known as |
| 3 | * SSM (Silicon Secured Memory). |
| 4 | * |
| 5 | * Copyright (C) 2016 Oracle and/or its affiliates. All rights reserved. |
| 6 | * Author: Khalid Aziz (khalid.aziz@oracle.com) |
| 7 | * |
| 8 | * This work is licensed under the terms of the GNU GPL, version 2. |
| 9 | */ |
| 10 | #include <linux/init.h> |
Khalid Aziz | 74a0496 | 2018-02-23 15:46:41 -0700 | [diff] [blame] | 11 | #include <linux/slab.h> |
| 12 | #include <linux/mm_types.h> |
Khalid Aziz | c6202ca | 2018-02-21 10:15:48 -0700 | [diff] [blame] | 13 | #include <asm/mdesc.h> |
| 14 | #include <asm/adi_64.h> |
Khalid Aziz | 74a0496 | 2018-02-23 15:46:41 -0700 | [diff] [blame] | 15 | #include <asm/mmu_64.h> |
| 16 | #include <asm/pgtable_64.h> |
| 17 | |
| 18 | /* Each page of storage for ADI tags can accommodate tags for 128 |
| 19 | * pages. When ADI enabled pages are being swapped out, it would be |
| 20 | * prudent to allocate at least enough tag storage space to accommodate |
| 21 | * SWAPFILE_CLUSTER number of pages. Allocate enough tag storage to |
| 22 | * store tags for four SWAPFILE_CLUSTER pages to reduce need for |
| 23 | * further allocations for same vma. |
| 24 | */ |
| 25 | #define TAG_STORAGE_PAGES 8 |
Khalid Aziz | c6202ca | 2018-02-21 10:15:48 -0700 | [diff] [blame] | 26 | |
| 27 | struct adi_config adi_state; |
Khalid Aziz | 74a0496 | 2018-02-23 15:46:41 -0700 | [diff] [blame] | 28 | EXPORT_SYMBOL(adi_state); |
Khalid Aziz | c6202ca | 2018-02-21 10:15:48 -0700 | [diff] [blame] | 29 | |
| 30 | /* mdesc_adi_init() : Parse machine description provided by the |
| 31 | * hypervisor to detect ADI capabilities |
| 32 | * |
| 33 | * Hypervisor reports ADI capabilities of platform in "hwcap-list" property |
| 34 | * for "cpu" node. If the platform supports ADI, "hwcap-list" property |
| 35 | * contains the keyword "adp". If the platform supports ADI, "platform" |
| 36 | * node will contain "adp-blksz", "adp-nbits" and "ue-on-adp" properties |
| 37 | * to describe the ADI capabilities. |
| 38 | */ |
| 39 | void __init mdesc_adi_init(void) |
| 40 | { |
| 41 | struct mdesc_handle *hp = mdesc_grab(); |
| 42 | const char *prop; |
| 43 | u64 pn, *val; |
| 44 | int len; |
| 45 | |
| 46 | if (!hp) |
| 47 | goto adi_not_found; |
| 48 | |
| 49 | pn = mdesc_node_by_name(hp, MDESC_NODE_NULL, "cpu"); |
| 50 | if (pn == MDESC_NODE_NULL) |
| 51 | goto adi_not_found; |
| 52 | |
| 53 | prop = mdesc_get_property(hp, pn, "hwcap-list", &len); |
| 54 | if (!prop) |
| 55 | goto adi_not_found; |
| 56 | |
| 57 | /* |
| 58 | * Look for "adp" keyword in hwcap-list which would indicate |
| 59 | * ADI support |
| 60 | */ |
| 61 | adi_state.enabled = false; |
| 62 | while (len) { |
| 63 | int plen; |
| 64 | |
| 65 | if (!strcmp(prop, "adp")) { |
| 66 | adi_state.enabled = true; |
| 67 | break; |
| 68 | } |
| 69 | |
| 70 | plen = strlen(prop) + 1; |
| 71 | prop += plen; |
| 72 | len -= plen; |
| 73 | } |
| 74 | |
| 75 | if (!adi_state.enabled) |
| 76 | goto adi_not_found; |
| 77 | |
| 78 | /* Find the ADI properties in "platform" node. If all ADI |
| 79 | * properties are not found, ADI support is incomplete and |
| 80 | * do not enable ADI in the kernel. |
| 81 | */ |
| 82 | pn = mdesc_node_by_name(hp, MDESC_NODE_NULL, "platform"); |
| 83 | if (pn == MDESC_NODE_NULL) |
| 84 | goto adi_not_found; |
| 85 | |
| 86 | val = (u64 *) mdesc_get_property(hp, pn, "adp-blksz", &len); |
| 87 | if (!val) |
| 88 | goto adi_not_found; |
| 89 | adi_state.caps.blksz = *val; |
| 90 | |
| 91 | val = (u64 *) mdesc_get_property(hp, pn, "adp-nbits", &len); |
| 92 | if (!val) |
| 93 | goto adi_not_found; |
| 94 | adi_state.caps.nbits = *val; |
| 95 | |
| 96 | val = (u64 *) mdesc_get_property(hp, pn, "ue-on-adp", &len); |
| 97 | if (!val) |
| 98 | goto adi_not_found; |
| 99 | adi_state.caps.ue_on_adi = *val; |
| 100 | |
Khalid Aziz | 74a0496 | 2018-02-23 15:46:41 -0700 | [diff] [blame] | 101 | /* Some of the code to support swapping ADI tags is written |
| 102 | * assumption that two ADI tags can fit inside one byte. If |
| 103 | * this assumption is broken by a future architecture change, |
| 104 | * that code will have to be revisited. If that were to happen, |
| 105 | * disable ADI support so we do not get unpredictable results |
| 106 | * with programs trying to use ADI and their pages getting |
| 107 | * swapped out |
| 108 | */ |
| 109 | if (adi_state.caps.nbits > 4) { |
| 110 | pr_warn("WARNING: ADI tag size >4 on this platform. Disabling AADI support\n"); |
| 111 | adi_state.enabled = false; |
| 112 | } |
| 113 | |
Khalid Aziz | c6202ca | 2018-02-21 10:15:48 -0700 | [diff] [blame] | 114 | mdesc_release(hp); |
| 115 | return; |
| 116 | |
| 117 | adi_not_found: |
| 118 | adi_state.enabled = false; |
| 119 | adi_state.caps.blksz = 0; |
| 120 | adi_state.caps.nbits = 0; |
| 121 | if (hp) |
| 122 | mdesc_release(hp); |
| 123 | } |
Khalid Aziz | 74a0496 | 2018-02-23 15:46:41 -0700 | [diff] [blame] | 124 | |
| 125 | tag_storage_desc_t *find_tag_store(struct mm_struct *mm, |
| 126 | struct vm_area_struct *vma, |
| 127 | unsigned long addr) |
| 128 | { |
| 129 | tag_storage_desc_t *tag_desc = NULL; |
| 130 | unsigned long i, max_desc, flags; |
| 131 | |
| 132 | /* Check if this vma already has tag storage descriptor |
| 133 | * allocated for it. |
| 134 | */ |
| 135 | max_desc = PAGE_SIZE/sizeof(tag_storage_desc_t); |
| 136 | if (mm->context.tag_store) { |
| 137 | tag_desc = mm->context.tag_store; |
| 138 | spin_lock_irqsave(&mm->context.tag_lock, flags); |
| 139 | for (i = 0; i < max_desc; i++) { |
| 140 | if ((addr >= tag_desc->start) && |
| 141 | ((addr + PAGE_SIZE - 1) <= tag_desc->end)) |
| 142 | break; |
| 143 | tag_desc++; |
| 144 | } |
| 145 | spin_unlock_irqrestore(&mm->context.tag_lock, flags); |
| 146 | |
| 147 | /* If no matching entries were found, this must be a |
| 148 | * freshly allocated page |
| 149 | */ |
| 150 | if (i >= max_desc) |
| 151 | tag_desc = NULL; |
| 152 | } |
| 153 | |
| 154 | return tag_desc; |
| 155 | } |
| 156 | |
| 157 | tag_storage_desc_t *alloc_tag_store(struct mm_struct *mm, |
| 158 | struct vm_area_struct *vma, |
| 159 | unsigned long addr) |
| 160 | { |
| 161 | unsigned char *tags; |
| 162 | unsigned long i, size, max_desc, flags; |
| 163 | tag_storage_desc_t *tag_desc, *open_desc; |
| 164 | unsigned long end_addr, hole_start, hole_end; |
| 165 | |
| 166 | max_desc = PAGE_SIZE/sizeof(tag_storage_desc_t); |
| 167 | open_desc = NULL; |
| 168 | hole_start = 0; |
| 169 | hole_end = ULONG_MAX; |
| 170 | end_addr = addr + PAGE_SIZE - 1; |
| 171 | |
| 172 | /* Check if this vma already has tag storage descriptor |
| 173 | * allocated for it. |
| 174 | */ |
| 175 | spin_lock_irqsave(&mm->context.tag_lock, flags); |
| 176 | if (mm->context.tag_store) { |
| 177 | tag_desc = mm->context.tag_store; |
| 178 | |
| 179 | /* Look for a matching entry for this address. While doing |
| 180 | * that, look for the first open slot as well and find |
| 181 | * the hole in already allocated range where this request |
| 182 | * will fit in. |
| 183 | */ |
| 184 | for (i = 0; i < max_desc; i++) { |
| 185 | if (tag_desc->tag_users == 0) { |
| 186 | if (open_desc == NULL) |
| 187 | open_desc = tag_desc; |
| 188 | } else { |
| 189 | if ((addr >= tag_desc->start) && |
| 190 | (tag_desc->end >= (addr + PAGE_SIZE - 1))) { |
| 191 | tag_desc->tag_users++; |
| 192 | goto out; |
| 193 | } |
| 194 | } |
| 195 | if ((tag_desc->start > end_addr) && |
| 196 | (tag_desc->start < hole_end)) |
| 197 | hole_end = tag_desc->start; |
| 198 | if ((tag_desc->end < addr) && |
| 199 | (tag_desc->end > hole_start)) |
| 200 | hole_start = tag_desc->end; |
| 201 | tag_desc++; |
| 202 | } |
| 203 | |
| 204 | } else { |
| 205 | size = sizeof(tag_storage_desc_t)*max_desc; |
| 206 | mm->context.tag_store = kzalloc(size, GFP_NOWAIT|__GFP_NOWARN); |
| 207 | if (mm->context.tag_store == NULL) { |
| 208 | tag_desc = NULL; |
| 209 | goto out; |
| 210 | } |
| 211 | tag_desc = mm->context.tag_store; |
| 212 | for (i = 0; i < max_desc; i++, tag_desc++) |
| 213 | tag_desc->tag_users = 0; |
| 214 | open_desc = mm->context.tag_store; |
| 215 | i = 0; |
| 216 | } |
| 217 | |
| 218 | /* Check if we ran out of tag storage descriptors */ |
| 219 | if (open_desc == NULL) { |
| 220 | tag_desc = NULL; |
| 221 | goto out; |
| 222 | } |
| 223 | |
| 224 | /* Mark this tag descriptor slot in use and then initialize it */ |
| 225 | tag_desc = open_desc; |
| 226 | tag_desc->tag_users = 1; |
| 227 | |
| 228 | /* Tag storage has not been allocated for this vma and space |
| 229 | * is available in tag storage descriptor. Since this page is |
| 230 | * being swapped out, there is high probability subsequent pages |
| 231 | * in the VMA will be swapped out as well. Allocate pages to |
| 232 | * store tags for as many pages in this vma as possible but not |
| 233 | * more than TAG_STORAGE_PAGES. Each byte in tag space holds |
| 234 | * two ADI tags since each ADI tag is 4 bits. Each ADI tag |
| 235 | * covers adi_blksize() worth of addresses. Check if the hole is |
| 236 | * big enough to accommodate full address range for using |
| 237 | * TAG_STORAGE_PAGES number of tag pages. |
| 238 | */ |
| 239 | size = TAG_STORAGE_PAGES * PAGE_SIZE; |
| 240 | end_addr = addr + (size*2*adi_blksize()) - 1; |
| 241 | /* Check for overflow. If overflow occurs, allocate only one page */ |
| 242 | if (end_addr < addr) { |
| 243 | size = PAGE_SIZE; |
| 244 | end_addr = addr + (size*2*adi_blksize()) - 1; |
| 245 | /* If overflow happens with the minimum tag storage |
| 246 | * allocation as well, adjust ending address for this |
| 247 | * tag storage. |
| 248 | */ |
| 249 | if (end_addr < addr) |
| 250 | end_addr = ULONG_MAX; |
| 251 | } |
| 252 | if (hole_end < end_addr) { |
| 253 | /* Available hole is too small on the upper end of |
| 254 | * address. Can we expand the range towards the lower |
| 255 | * address and maximize use of this slot? |
| 256 | */ |
| 257 | unsigned long tmp_addr; |
| 258 | |
| 259 | end_addr = hole_end - 1; |
| 260 | tmp_addr = end_addr - (size*2*adi_blksize()) + 1; |
| 261 | /* Check for underflow. If underflow occurs, allocate |
| 262 | * only one page for storing ADI tags |
| 263 | */ |
| 264 | if (tmp_addr > addr) { |
| 265 | size = PAGE_SIZE; |
| 266 | tmp_addr = end_addr - (size*2*adi_blksize()) - 1; |
| 267 | /* If underflow happens with the minimum tag storage |
| 268 | * allocation as well, adjust starting address for |
| 269 | * this tag storage. |
| 270 | */ |
| 271 | if (tmp_addr > addr) |
| 272 | tmp_addr = 0; |
| 273 | } |
| 274 | if (tmp_addr < hole_start) { |
| 275 | /* Available hole is restricted on lower address |
| 276 | * end as well |
| 277 | */ |
| 278 | tmp_addr = hole_start + 1; |
| 279 | } |
| 280 | addr = tmp_addr; |
| 281 | size = (end_addr + 1 - addr)/(2*adi_blksize()); |
| 282 | size = (size + (PAGE_SIZE-adi_blksize()))/PAGE_SIZE; |
| 283 | size = size * PAGE_SIZE; |
| 284 | } |
| 285 | tags = kzalloc(size, GFP_NOWAIT|__GFP_NOWARN); |
| 286 | if (tags == NULL) { |
| 287 | tag_desc->tag_users = 0; |
| 288 | tag_desc = NULL; |
| 289 | goto out; |
| 290 | } |
| 291 | tag_desc->start = addr; |
| 292 | tag_desc->tags = tags; |
| 293 | tag_desc->end = end_addr; |
| 294 | |
| 295 | out: |
| 296 | spin_unlock_irqrestore(&mm->context.tag_lock, flags); |
| 297 | return tag_desc; |
| 298 | } |
| 299 | |
| 300 | void del_tag_store(tag_storage_desc_t *tag_desc, struct mm_struct *mm) |
| 301 | { |
| 302 | unsigned long flags; |
| 303 | unsigned char *tags = NULL; |
| 304 | |
| 305 | spin_lock_irqsave(&mm->context.tag_lock, flags); |
| 306 | tag_desc->tag_users--; |
| 307 | if (tag_desc->tag_users == 0) { |
| 308 | tag_desc->start = tag_desc->end = 0; |
| 309 | /* Do not free up the tag storage space allocated |
| 310 | * by the first descriptor. This is persistent |
| 311 | * emergency tag storage space for the task. |
| 312 | */ |
| 313 | if (tag_desc != mm->context.tag_store) { |
| 314 | tags = tag_desc->tags; |
| 315 | tag_desc->tags = NULL; |
| 316 | } |
| 317 | } |
| 318 | spin_unlock_irqrestore(&mm->context.tag_lock, flags); |
| 319 | kfree(tags); |
| 320 | } |
| 321 | |
| 322 | #define tag_start(addr, tag_desc) \ |
| 323 | ((tag_desc)->tags + ((addr - (tag_desc)->start)/(2*adi_blksize()))) |
| 324 | |
| 325 | /* Retrieve any saved ADI tags for the page being swapped back in and |
| 326 | * restore these tags to the newly allocated physical page. |
| 327 | */ |
| 328 | void adi_restore_tags(struct mm_struct *mm, struct vm_area_struct *vma, |
| 329 | unsigned long addr, pte_t pte) |
| 330 | { |
| 331 | unsigned char *tag; |
| 332 | tag_storage_desc_t *tag_desc; |
| 333 | unsigned long paddr, tmp, version1, version2; |
| 334 | |
| 335 | /* Check if the swapped out page has an ADI version |
| 336 | * saved. If yes, restore version tag to the newly |
| 337 | * allocated page. |
| 338 | */ |
| 339 | tag_desc = find_tag_store(mm, vma, addr); |
| 340 | if (tag_desc == NULL) |
| 341 | return; |
| 342 | |
| 343 | tag = tag_start(addr, tag_desc); |
| 344 | paddr = pte_val(pte) & _PAGE_PADDR_4V; |
| 345 | for (tmp = paddr; tmp < (paddr+PAGE_SIZE); tmp += adi_blksize()) { |
| 346 | version1 = (*tag) >> 4; |
| 347 | version2 = (*tag) & 0x0f; |
| 348 | *tag++ = 0; |
| 349 | asm volatile("stxa %0, [%1] %2\n\t" |
| 350 | : |
| 351 | : "r" (version1), "r" (tmp), |
| 352 | "i" (ASI_MCD_REAL)); |
| 353 | tmp += adi_blksize(); |
| 354 | asm volatile("stxa %0, [%1] %2\n\t" |
| 355 | : |
| 356 | : "r" (version2), "r" (tmp), |
| 357 | "i" (ASI_MCD_REAL)); |
| 358 | } |
| 359 | asm volatile("membar #Sync\n\t"); |
| 360 | |
| 361 | /* Check and mark this tag space for release later if |
| 362 | * the swapped in page was the last user of tag space |
| 363 | */ |
| 364 | del_tag_store(tag_desc, mm); |
| 365 | } |
| 366 | |
| 367 | /* A page is about to be swapped out. Save any ADI tags associated with |
| 368 | * this physical page so they can be restored later when the page is swapped |
| 369 | * back in. |
| 370 | */ |
| 371 | int adi_save_tags(struct mm_struct *mm, struct vm_area_struct *vma, |
| 372 | unsigned long addr, pte_t oldpte) |
| 373 | { |
| 374 | unsigned char *tag; |
| 375 | tag_storage_desc_t *tag_desc; |
| 376 | unsigned long version1, version2, paddr, tmp; |
| 377 | |
| 378 | tag_desc = alloc_tag_store(mm, vma, addr); |
| 379 | if (tag_desc == NULL) |
| 380 | return -1; |
| 381 | |
| 382 | tag = tag_start(addr, tag_desc); |
| 383 | paddr = pte_val(oldpte) & _PAGE_PADDR_4V; |
| 384 | for (tmp = paddr; tmp < (paddr+PAGE_SIZE); tmp += adi_blksize()) { |
| 385 | asm volatile("ldxa [%1] %2, %0\n\t" |
| 386 | : "=r" (version1) |
| 387 | : "r" (tmp), "i" (ASI_MCD_REAL)); |
| 388 | tmp += adi_blksize(); |
| 389 | asm volatile("ldxa [%1] %2, %0\n\t" |
| 390 | : "=r" (version2) |
| 391 | : "r" (tmp), "i" (ASI_MCD_REAL)); |
| 392 | *tag = (version1 << 4) | version2; |
| 393 | tag++; |
| 394 | } |
| 395 | |
| 396 | return 0; |
| 397 | } |