Greg Kroah-Hartman | b244131 | 2017-11-01 15:07:57 +0100 | [diff] [blame] | 1 | // SPDX-License-Identifier: GPL-2.0 |
Dong Jia Shi | 0a19e61 | 2017-03-17 04:17:32 +0100 | [diff] [blame] | 2 | /* |
| 3 | * channel program interfaces |
| 4 | * |
| 5 | * Copyright IBM Corp. 2017 |
| 6 | * |
| 7 | * Author(s): Dong Jia Shi <bjsdjshi@linux.vnet.ibm.com> |
| 8 | * Xiao Feng Ren <renxiaof@linux.vnet.ibm.com> |
| 9 | */ |
| 10 | |
| 11 | #include <linux/mm.h> |
| 12 | #include <linux/slab.h> |
| 13 | #include <linux/iommu.h> |
| 14 | #include <linux/vfio.h> |
| 15 | #include <asm/idals.h> |
| 16 | |
| 17 | #include "vfio_ccw_cp.h" |
| 18 | |
| 19 | /* |
| 20 | * Max length for ccw chain. |
| 21 | * XXX: Limit to 256, need to check more? |
| 22 | */ |
| 23 | #define CCWCHAIN_LEN_MAX 256 |
| 24 | |
| 25 | struct pfn_array { |
Dong Jia Shi | 80c57f7 | 2018-05-23 04:56:42 +0200 | [diff] [blame] | 26 | /* Starting guest physical I/O address. */ |
Dong Jia Shi | 0a19e61 | 2017-03-17 04:17:32 +0100 | [diff] [blame] | 27 | unsigned long pa_iova; |
Dong Jia Shi | 80c57f7 | 2018-05-23 04:56:42 +0200 | [diff] [blame] | 28 | /* Array that stores PFNs of the pages need to pin. */ |
Dong Jia Shi | 0a19e61 | 2017-03-17 04:17:32 +0100 | [diff] [blame] | 29 | unsigned long *pa_iova_pfn; |
Dong Jia Shi | 80c57f7 | 2018-05-23 04:56:42 +0200 | [diff] [blame] | 30 | /* Array that receives PFNs of the pages pinned. */ |
Dong Jia Shi | 0a19e61 | 2017-03-17 04:17:32 +0100 | [diff] [blame] | 31 | unsigned long *pa_pfn; |
Dong Jia Shi | 5c1cfb1 | 2018-05-23 04:56:43 +0200 | [diff] [blame] | 32 | /* Number of pages pinned from @pa_iova. */ |
Dong Jia Shi | 0a19e61 | 2017-03-17 04:17:32 +0100 | [diff] [blame] | 33 | int pa_nr; |
| 34 | }; |
| 35 | |
| 36 | struct pfn_array_table { |
| 37 | struct pfn_array *pat_pa; |
| 38 | int pat_nr; |
| 39 | }; |
| 40 | |
| 41 | struct ccwchain { |
| 42 | struct list_head next; |
| 43 | struct ccw1 *ch_ccw; |
| 44 | /* Guest physical address of the current chain. */ |
| 45 | u64 ch_iova; |
| 46 | /* Count of the valid ccws in chain. */ |
| 47 | int ch_len; |
| 48 | /* Pinned PAGEs for the original data. */ |
| 49 | struct pfn_array_table *ch_pat; |
| 50 | }; |
| 51 | |
| 52 | /* |
Dong Jia Shi | 5c1cfb1 | 2018-05-23 04:56:43 +0200 | [diff] [blame] | 53 | * pfn_array_alloc_pin() - alloc memory for PFNs, then pin user pages in memory |
Dong Jia Shi | 0a19e61 | 2017-03-17 04:17:32 +0100 | [diff] [blame] | 54 | * @pa: pfn_array on which to perform the operation |
| 55 | * @mdev: the mediated device to perform pin/unpin operations |
Dong Jia Shi | 5c1cfb1 | 2018-05-23 04:56:43 +0200 | [diff] [blame] | 56 | * @iova: target guest physical address |
| 57 | * @len: number of bytes that should be pinned from @iova |
Dong Jia Shi | 0a19e61 | 2017-03-17 04:17:32 +0100 | [diff] [blame] | 58 | * |
Dong Jia Shi | 5c1cfb1 | 2018-05-23 04:56:43 +0200 | [diff] [blame] | 59 | * Attempt to allocate memory for PFNs, and pin user pages in memory. |
Dong Jia Shi | 0a19e61 | 2017-03-17 04:17:32 +0100 | [diff] [blame] | 60 | * |
| 61 | * Usage of pfn_array: |
Dong Jia Shi | 5c1cfb1 | 2018-05-23 04:56:43 +0200 | [diff] [blame] | 62 | * We expect (pa_nr == 0) and (pa_iova_pfn == NULL), any field in |
| 63 | * this structure will be filled in by this function. |
Dong Jia Shi | 0a19e61 | 2017-03-17 04:17:32 +0100 | [diff] [blame] | 64 | * |
| 65 | * Returns: |
| 66 | * Number of pages pinned on success. |
Dong Jia Shi | 5c1cfb1 | 2018-05-23 04:56:43 +0200 | [diff] [blame] | 67 | * If @pa->pa_nr is not 0, or @pa->pa_iova_pfn is not NULL initially, |
| 68 | * returns -EINVAL. |
Dong Jia Shi | 0a19e61 | 2017-03-17 04:17:32 +0100 | [diff] [blame] | 69 | * If no pages were pinned, returns -errno. |
| 70 | */ |
Dong Jia Shi | 0a19e61 | 2017-03-17 04:17:32 +0100 | [diff] [blame] | 71 | static int pfn_array_alloc_pin(struct pfn_array *pa, struct device *mdev, |
| 72 | u64 iova, unsigned int len) |
| 73 | { |
Dong Jia Shi | 5c1cfb1 | 2018-05-23 04:56:43 +0200 | [diff] [blame] | 74 | int i, ret = 0; |
Dong Jia Shi | 0a19e61 | 2017-03-17 04:17:32 +0100 | [diff] [blame] | 75 | |
Dong Jia Shi | 4cebc5d | 2017-10-11 04:38:22 +0200 | [diff] [blame] | 76 | if (!len) |
| 77 | return 0; |
| 78 | |
Dong Jia Shi | 5c1cfb1 | 2018-05-23 04:56:43 +0200 | [diff] [blame] | 79 | if (pa->pa_nr || pa->pa_iova_pfn) |
Dong Jia Shi | 0a19e61 | 2017-03-17 04:17:32 +0100 | [diff] [blame] | 80 | return -EINVAL; |
| 81 | |
| 82 | pa->pa_iova = iova; |
| 83 | |
| 84 | pa->pa_nr = ((iova & ~PAGE_MASK) + len + (PAGE_SIZE - 1)) >> PAGE_SHIFT; |
| 85 | if (!pa->pa_nr) |
| 86 | return -EINVAL; |
| 87 | |
| 88 | pa->pa_iova_pfn = kcalloc(pa->pa_nr, |
| 89 | sizeof(*pa->pa_iova_pfn) + |
| 90 | sizeof(*pa->pa_pfn), |
| 91 | GFP_KERNEL); |
| 92 | if (unlikely(!pa->pa_iova_pfn)) |
| 93 | return -ENOMEM; |
| 94 | pa->pa_pfn = pa->pa_iova_pfn + pa->pa_nr; |
| 95 | |
Dong Jia Shi | 5c1cfb1 | 2018-05-23 04:56:43 +0200 | [diff] [blame] | 96 | pa->pa_iova_pfn[0] = pa->pa_iova >> PAGE_SHIFT; |
| 97 | for (i = 1; i < pa->pa_nr; i++) |
| 98 | pa->pa_iova_pfn[i] = pa->pa_iova_pfn[i - 1] + 1; |
Dong Jia Shi | 0a19e61 | 2017-03-17 04:17:32 +0100 | [diff] [blame] | 99 | |
Dong Jia Shi | 5c1cfb1 | 2018-05-23 04:56:43 +0200 | [diff] [blame] | 100 | ret = vfio_pin_pages(mdev, pa->pa_iova_pfn, pa->pa_nr, |
| 101 | IOMMU_READ | IOMMU_WRITE, pa->pa_pfn); |
| 102 | |
| 103 | if (ret < 0) { |
| 104 | goto err_out; |
| 105 | } else if (ret > 0 && ret != pa->pa_nr) { |
| 106 | vfio_unpin_pages(mdev, pa->pa_iova_pfn, ret); |
Dong Jia Shi | 0a19e61 | 2017-03-17 04:17:32 +0100 | [diff] [blame] | 107 | ret = -EINVAL; |
Dong Jia Shi | 5c1cfb1 | 2018-05-23 04:56:43 +0200 | [diff] [blame] | 108 | goto err_out; |
| 109 | } |
Dong Jia Shi | 0a19e61 | 2017-03-17 04:17:32 +0100 | [diff] [blame] | 110 | |
| 111 | return ret; |
Dong Jia Shi | 5c1cfb1 | 2018-05-23 04:56:43 +0200 | [diff] [blame] | 112 | |
| 113 | err_out: |
| 114 | pa->pa_nr = 0; |
| 115 | kfree(pa->pa_iova_pfn); |
| 116 | pa->pa_iova_pfn = NULL; |
| 117 | |
| 118 | return ret; |
| 119 | } |
| 120 | |
| 121 | /* Unpin the pages before releasing the memory. */ |
| 122 | static void pfn_array_unpin_free(struct pfn_array *pa, struct device *mdev) |
| 123 | { |
| 124 | vfio_unpin_pages(mdev, pa->pa_iova_pfn, pa->pa_nr); |
| 125 | pa->pa_nr = 0; |
| 126 | kfree(pa->pa_iova_pfn); |
Dong Jia Shi | 0a19e61 | 2017-03-17 04:17:32 +0100 | [diff] [blame] | 127 | } |
| 128 | |
| 129 | static int pfn_array_table_init(struct pfn_array_table *pat, int nr) |
| 130 | { |
| 131 | pat->pat_pa = kcalloc(nr, sizeof(*pat->pat_pa), GFP_KERNEL); |
| 132 | if (unlikely(ZERO_OR_NULL_PTR(pat->pat_pa))) { |
| 133 | pat->pat_nr = 0; |
| 134 | return -ENOMEM; |
| 135 | } |
| 136 | |
| 137 | pat->pat_nr = nr; |
| 138 | |
| 139 | return 0; |
| 140 | } |
| 141 | |
| 142 | static void pfn_array_table_unpin_free(struct pfn_array_table *pat, |
| 143 | struct device *mdev) |
| 144 | { |
| 145 | int i; |
| 146 | |
| 147 | for (i = 0; i < pat->pat_nr; i++) |
| 148 | pfn_array_unpin_free(pat->pat_pa + i, mdev); |
| 149 | |
| 150 | if (pat->pat_nr) { |
| 151 | kfree(pat->pat_pa); |
| 152 | pat->pat_pa = NULL; |
| 153 | pat->pat_nr = 0; |
| 154 | } |
| 155 | } |
| 156 | |
| 157 | static bool pfn_array_table_iova_pinned(struct pfn_array_table *pat, |
| 158 | unsigned long iova) |
| 159 | { |
| 160 | struct pfn_array *pa = pat->pat_pa; |
| 161 | unsigned long iova_pfn = iova >> PAGE_SHIFT; |
| 162 | int i, j; |
| 163 | |
| 164 | for (i = 0; i < pat->pat_nr; i++, pa++) |
| 165 | for (j = 0; j < pa->pa_nr; j++) |
| 166 | if (pa->pa_iova_pfn[i] == iova_pfn) |
| 167 | return true; |
| 168 | |
| 169 | return false; |
| 170 | } |
| 171 | /* Create the list idal words for a pfn_array_table. */ |
| 172 | static inline void pfn_array_table_idal_create_words( |
| 173 | struct pfn_array_table *pat, |
| 174 | unsigned long *idaws) |
| 175 | { |
| 176 | struct pfn_array *pa; |
| 177 | int i, j, k; |
| 178 | |
| 179 | /* |
| 180 | * Idal words (execept the first one) rely on the memory being 4k |
| 181 | * aligned. If a user virtual address is 4K aligned, then it's |
| 182 | * corresponding kernel physical address will also be 4K aligned. Thus |
| 183 | * there will be no problem here to simply use the phys to create an |
| 184 | * idaw. |
| 185 | */ |
| 186 | k = 0; |
| 187 | for (i = 0; i < pat->pat_nr; i++) { |
| 188 | pa = pat->pat_pa + i; |
| 189 | for (j = 0; j < pa->pa_nr; j++) { |
| 190 | idaws[k] = pa->pa_pfn[j] << PAGE_SHIFT; |
| 191 | if (k == 0) |
| 192 | idaws[k] += pa->pa_iova & (PAGE_SIZE - 1); |
| 193 | k++; |
| 194 | } |
| 195 | } |
| 196 | } |
| 197 | |
| 198 | |
| 199 | /* |
| 200 | * Within the domain (@mdev), copy @n bytes from a guest physical |
| 201 | * address (@iova) to a host physical address (@to). |
| 202 | */ |
| 203 | static long copy_from_iova(struct device *mdev, |
| 204 | void *to, u64 iova, |
| 205 | unsigned long n) |
| 206 | { |
| 207 | struct pfn_array pa = {0}; |
| 208 | u64 from; |
| 209 | int i, ret; |
| 210 | unsigned long l, m; |
| 211 | |
| 212 | ret = pfn_array_alloc_pin(&pa, mdev, iova, n); |
| 213 | if (ret <= 0) |
| 214 | return ret; |
| 215 | |
| 216 | l = n; |
| 217 | for (i = 0; i < pa.pa_nr; i++) { |
| 218 | from = pa.pa_pfn[i] << PAGE_SHIFT; |
| 219 | m = PAGE_SIZE; |
| 220 | if (i == 0) { |
| 221 | from += iova & (PAGE_SIZE - 1); |
| 222 | m -= iova & (PAGE_SIZE - 1); |
| 223 | } |
| 224 | |
| 225 | m = min(l, m); |
| 226 | memcpy(to + (n - l), (void *)from, m); |
| 227 | |
| 228 | l -= m; |
| 229 | if (l == 0) |
| 230 | break; |
| 231 | } |
| 232 | |
| 233 | pfn_array_unpin_free(&pa, mdev); |
| 234 | |
| 235 | return l; |
| 236 | } |
| 237 | |
| 238 | static long copy_ccw_from_iova(struct channel_program *cp, |
| 239 | struct ccw1 *to, u64 iova, |
| 240 | unsigned long len) |
| 241 | { |
Dong Jia Shi | d686f21 | 2017-03-17 04:17:42 +0100 | [diff] [blame] | 242 | struct ccw0 ccw0; |
| 243 | struct ccw1 *pccw1; |
| 244 | int ret; |
| 245 | int i; |
| 246 | |
| 247 | ret = copy_from_iova(cp->mdev, to, iova, len * sizeof(struct ccw1)); |
| 248 | if (ret) |
| 249 | return ret; |
| 250 | |
| 251 | if (!cp->orb.cmd.fmt) { |
| 252 | pccw1 = to; |
| 253 | for (i = 0; i < len; i++) { |
| 254 | ccw0 = *(struct ccw0 *)pccw1; |
| 255 | if ((pccw1->cmd_code & 0x0f) == CCW_CMD_TIC) { |
| 256 | pccw1->cmd_code = CCW_CMD_TIC; |
| 257 | pccw1->flags = 0; |
| 258 | pccw1->count = 0; |
| 259 | } else { |
| 260 | pccw1->cmd_code = ccw0.cmd_code; |
| 261 | pccw1->flags = ccw0.flags; |
| 262 | pccw1->count = ccw0.count; |
| 263 | } |
| 264 | pccw1->cda = ccw0.cda; |
| 265 | pccw1++; |
| 266 | } |
| 267 | } |
| 268 | |
| 269 | return ret; |
Dong Jia Shi | 0a19e61 | 2017-03-17 04:17:32 +0100 | [diff] [blame] | 270 | } |
| 271 | |
| 272 | /* |
| 273 | * Helpers to operate ccwchain. |
| 274 | */ |
| 275 | #define ccw_is_test(_ccw) (((_ccw)->cmd_code & 0x0F) == 0) |
| 276 | |
| 277 | #define ccw_is_noop(_ccw) ((_ccw)->cmd_code == CCW_CMD_NOOP) |
| 278 | |
| 279 | #define ccw_is_tic(_ccw) ((_ccw)->cmd_code == CCW_CMD_TIC) |
| 280 | |
| 281 | #define ccw_is_idal(_ccw) ((_ccw)->flags & CCW_FLAG_IDA) |
| 282 | |
| 283 | |
| 284 | #define ccw_is_chain(_ccw) ((_ccw)->flags & (CCW_FLAG_CC | CCW_FLAG_DC)) |
| 285 | |
| 286 | static struct ccwchain *ccwchain_alloc(struct channel_program *cp, int len) |
| 287 | { |
| 288 | struct ccwchain *chain; |
| 289 | void *data; |
| 290 | size_t size; |
| 291 | |
| 292 | /* Make ccw address aligned to 8. */ |
| 293 | size = ((sizeof(*chain) + 7L) & -8L) + |
| 294 | sizeof(*chain->ch_ccw) * len + |
| 295 | sizeof(*chain->ch_pat) * len; |
| 296 | chain = kzalloc(size, GFP_DMA | GFP_KERNEL); |
| 297 | if (!chain) |
| 298 | return NULL; |
| 299 | |
| 300 | data = (u8 *)chain + ((sizeof(*chain) + 7L) & -8L); |
| 301 | chain->ch_ccw = (struct ccw1 *)data; |
| 302 | |
| 303 | data = (u8 *)(chain->ch_ccw) + sizeof(*chain->ch_ccw) * len; |
| 304 | chain->ch_pat = (struct pfn_array_table *)data; |
| 305 | |
| 306 | chain->ch_len = len; |
| 307 | |
| 308 | list_add_tail(&chain->next, &cp->ccwchain_list); |
| 309 | |
| 310 | return chain; |
| 311 | } |
| 312 | |
| 313 | static void ccwchain_free(struct ccwchain *chain) |
| 314 | { |
| 315 | list_del(&chain->next); |
| 316 | kfree(chain); |
| 317 | } |
| 318 | |
| 319 | /* Free resource for a ccw that allocated memory for its cda. */ |
| 320 | static void ccwchain_cda_free(struct ccwchain *chain, int idx) |
| 321 | { |
| 322 | struct ccw1 *ccw = chain->ch_ccw + idx; |
| 323 | |
Jason J. Herne | 408358b | 2017-11-07 10:22:32 -0500 | [diff] [blame] | 324 | if (ccw_is_test(ccw) || ccw_is_noop(ccw) || ccw_is_tic(ccw)) |
| 325 | return; |
Dong Jia Shi | 0a19e61 | 2017-03-17 04:17:32 +0100 | [diff] [blame] | 326 | if (!ccw->count) |
| 327 | return; |
| 328 | |
| 329 | kfree((void *)(u64)ccw->cda); |
| 330 | } |
| 331 | |
| 332 | /* Unpin the pages then free the memory resources. */ |
| 333 | static void cp_unpin_free(struct channel_program *cp) |
| 334 | { |
| 335 | struct ccwchain *chain, *temp; |
| 336 | int i; |
| 337 | |
| 338 | list_for_each_entry_safe(chain, temp, &cp->ccwchain_list, next) { |
| 339 | for (i = 0; i < chain->ch_len; i++) { |
| 340 | pfn_array_table_unpin_free(chain->ch_pat + i, |
| 341 | cp->mdev); |
| 342 | ccwchain_cda_free(chain, i); |
| 343 | } |
| 344 | ccwchain_free(chain); |
| 345 | } |
| 346 | } |
| 347 | |
| 348 | /** |
| 349 | * ccwchain_calc_length - calculate the length of the ccw chain. |
| 350 | * @iova: guest physical address of the target ccw chain |
| 351 | * @cp: channel_program on which to perform the operation |
| 352 | * |
| 353 | * This is the chain length not considering any TICs. |
| 354 | * You need to do a new round for each TIC target. |
| 355 | * |
Halil Pasic | fb9e788 | 2018-05-16 19:33:42 +0200 | [diff] [blame] | 356 | * The program is also validated for absence of not yet supported |
| 357 | * indirect data addressing scenarios. |
| 358 | * |
Dong Jia Shi | 0a19e61 | 2017-03-17 04:17:32 +0100 | [diff] [blame] | 359 | * Returns: the length of the ccw chain or -errno. |
| 360 | */ |
| 361 | static int ccwchain_calc_length(u64 iova, struct channel_program *cp) |
| 362 | { |
| 363 | struct ccw1 *ccw, *p; |
| 364 | int cnt; |
| 365 | |
| 366 | /* |
| 367 | * Copy current chain from guest to host kernel. |
| 368 | * Currently the chain length is limited to CCWCHAIN_LEN_MAX (256). |
| 369 | * So copying 2K is enough (safe). |
| 370 | */ |
| 371 | p = ccw = kcalloc(CCWCHAIN_LEN_MAX, sizeof(*ccw), GFP_KERNEL); |
| 372 | if (!ccw) |
| 373 | return -ENOMEM; |
| 374 | |
| 375 | cnt = copy_ccw_from_iova(cp, ccw, iova, CCWCHAIN_LEN_MAX); |
| 376 | if (cnt) { |
| 377 | kfree(ccw); |
| 378 | return cnt; |
| 379 | } |
| 380 | |
| 381 | cnt = 0; |
| 382 | do { |
| 383 | cnt++; |
| 384 | |
Halil Pasic | fb9e788 | 2018-05-16 19:33:42 +0200 | [diff] [blame] | 385 | /* |
| 386 | * As we don't want to fail direct addressing even if the |
| 387 | * orb specified one of the unsupported formats, we defer |
| 388 | * checking for IDAWs in unsupported formats to here. |
| 389 | */ |
| 390 | if ((!cp->orb.cmd.c64 || cp->orb.cmd.i2k) && ccw_is_idal(ccw)) |
| 391 | return -EOPNOTSUPP; |
| 392 | |
Dong Jia Shi | 0a19e61 | 2017-03-17 04:17:32 +0100 | [diff] [blame] | 393 | if ((!ccw_is_chain(ccw)) && (!ccw_is_tic(ccw))) |
| 394 | break; |
| 395 | |
| 396 | ccw++; |
| 397 | } while (cnt < CCWCHAIN_LEN_MAX + 1); |
| 398 | |
| 399 | if (cnt == CCWCHAIN_LEN_MAX + 1) |
| 400 | cnt = -EINVAL; |
| 401 | |
| 402 | kfree(p); |
| 403 | return cnt; |
| 404 | } |
| 405 | |
| 406 | static int tic_target_chain_exists(struct ccw1 *tic, struct channel_program *cp) |
| 407 | { |
| 408 | struct ccwchain *chain; |
| 409 | u32 ccw_head, ccw_tail; |
| 410 | |
| 411 | list_for_each_entry(chain, &cp->ccwchain_list, next) { |
| 412 | ccw_head = chain->ch_iova; |
| 413 | ccw_tail = ccw_head + (chain->ch_len - 1) * sizeof(struct ccw1); |
| 414 | |
| 415 | if ((ccw_head <= tic->cda) && (tic->cda <= ccw_tail)) |
| 416 | return 1; |
| 417 | } |
| 418 | |
| 419 | return 0; |
| 420 | } |
| 421 | |
| 422 | static int ccwchain_loop_tic(struct ccwchain *chain, |
| 423 | struct channel_program *cp); |
| 424 | |
| 425 | static int ccwchain_handle_tic(struct ccw1 *tic, struct channel_program *cp) |
| 426 | { |
| 427 | struct ccwchain *chain; |
| 428 | int len, ret; |
| 429 | |
| 430 | /* May transfer to an existing chain. */ |
| 431 | if (tic_target_chain_exists(tic, cp)) |
| 432 | return 0; |
| 433 | |
| 434 | /* Get chain length. */ |
| 435 | len = ccwchain_calc_length(tic->cda, cp); |
| 436 | if (len < 0) |
| 437 | return len; |
| 438 | |
| 439 | /* Need alloc a new chain for this one. */ |
| 440 | chain = ccwchain_alloc(cp, len); |
| 441 | if (!chain) |
| 442 | return -ENOMEM; |
| 443 | chain->ch_iova = tic->cda; |
| 444 | |
| 445 | /* Copy the new chain from user. */ |
| 446 | ret = copy_ccw_from_iova(cp, chain->ch_ccw, tic->cda, len); |
| 447 | if (ret) { |
| 448 | ccwchain_free(chain); |
| 449 | return ret; |
| 450 | } |
| 451 | |
| 452 | /* Loop for tics on this new chain. */ |
| 453 | return ccwchain_loop_tic(chain, cp); |
| 454 | } |
| 455 | |
| 456 | /* Loop for TICs. */ |
| 457 | static int ccwchain_loop_tic(struct ccwchain *chain, struct channel_program *cp) |
| 458 | { |
| 459 | struct ccw1 *tic; |
| 460 | int i, ret; |
| 461 | |
| 462 | for (i = 0; i < chain->ch_len; i++) { |
| 463 | tic = chain->ch_ccw + i; |
| 464 | |
| 465 | if (!ccw_is_tic(tic)) |
| 466 | continue; |
| 467 | |
| 468 | ret = ccwchain_handle_tic(tic, cp); |
| 469 | if (ret) |
| 470 | return ret; |
| 471 | } |
| 472 | |
| 473 | return 0; |
| 474 | } |
| 475 | |
| 476 | static int ccwchain_fetch_tic(struct ccwchain *chain, |
| 477 | int idx, |
| 478 | struct channel_program *cp) |
| 479 | { |
| 480 | struct ccw1 *ccw = chain->ch_ccw + idx; |
| 481 | struct ccwchain *iter; |
| 482 | u32 ccw_head, ccw_tail; |
| 483 | |
| 484 | list_for_each_entry(iter, &cp->ccwchain_list, next) { |
| 485 | ccw_head = iter->ch_iova; |
| 486 | ccw_tail = ccw_head + (iter->ch_len - 1) * sizeof(struct ccw1); |
| 487 | |
| 488 | if ((ccw_head <= ccw->cda) && (ccw->cda <= ccw_tail)) { |
Jason J. Herne | c389377 | 2017-07-21 03:14:36 +0200 | [diff] [blame] | 489 | ccw->cda = (__u32) (addr_t) (((char *)iter->ch_ccw) + |
Dong Jia Shi | 0a19e61 | 2017-03-17 04:17:32 +0100 | [diff] [blame] | 490 | (ccw->cda - ccw_head)); |
| 491 | return 0; |
| 492 | } |
| 493 | } |
| 494 | |
| 495 | return -EFAULT; |
| 496 | } |
| 497 | |
| 498 | static int ccwchain_fetch_direct(struct ccwchain *chain, |
| 499 | int idx, |
| 500 | struct channel_program *cp) |
| 501 | { |
| 502 | struct ccw1 *ccw; |
| 503 | struct pfn_array_table *pat; |
| 504 | unsigned long *idaws; |
Dong Jia Shi | 6238f92 | 2018-05-23 04:56:44 +0200 | [diff] [blame] | 505 | int ret; |
Dong Jia Shi | 0a19e61 | 2017-03-17 04:17:32 +0100 | [diff] [blame] | 506 | |
| 507 | ccw = chain->ch_ccw + idx; |
| 508 | |
Dong Jia Shi | 4cebc5d | 2017-10-11 04:38:22 +0200 | [diff] [blame] | 509 | if (!ccw->count) { |
| 510 | /* |
| 511 | * We just want the translation result of any direct ccw |
| 512 | * to be an IDA ccw, so let's add the IDA flag for it. |
| 513 | * Although the flag will be ignored by firmware. |
| 514 | */ |
| 515 | ccw->flags |= CCW_FLAG_IDA; |
| 516 | return 0; |
| 517 | } |
| 518 | |
Dong Jia Shi | 0a19e61 | 2017-03-17 04:17:32 +0100 | [diff] [blame] | 519 | /* |
| 520 | * Pin data page(s) in memory. |
| 521 | * The number of pages actually is the count of the idaws which will be |
| 522 | * needed when translating a direct ccw to a idal ccw. |
| 523 | */ |
| 524 | pat = chain->ch_pat + idx; |
Dong Jia Shi | 6238f92 | 2018-05-23 04:56:44 +0200 | [diff] [blame] | 525 | ret = pfn_array_table_init(pat, 1); |
| 526 | if (ret) |
| 527 | goto out_init; |
| 528 | |
| 529 | ret = pfn_array_alloc_pin(pat->pat_pa, cp->mdev, ccw->cda, ccw->count); |
| 530 | if (ret < 0) |
| 531 | goto out_init; |
Dong Jia Shi | 0a19e61 | 2017-03-17 04:17:32 +0100 | [diff] [blame] | 532 | |
| 533 | /* Translate this direct ccw to a idal ccw. */ |
Dong Jia Shi | 6238f92 | 2018-05-23 04:56:44 +0200 | [diff] [blame] | 534 | idaws = kcalloc(ret, sizeof(*idaws), GFP_DMA | GFP_KERNEL); |
Dong Jia Shi | 0a19e61 | 2017-03-17 04:17:32 +0100 | [diff] [blame] | 535 | if (!idaws) { |
Dong Jia Shi | 6238f92 | 2018-05-23 04:56:44 +0200 | [diff] [blame] | 536 | ret = -ENOMEM; |
| 537 | goto out_unpin; |
Dong Jia Shi | 0a19e61 | 2017-03-17 04:17:32 +0100 | [diff] [blame] | 538 | } |
| 539 | ccw->cda = (__u32) virt_to_phys(idaws); |
| 540 | ccw->flags |= CCW_FLAG_IDA; |
| 541 | |
| 542 | pfn_array_table_idal_create_words(pat, idaws); |
| 543 | |
| 544 | return 0; |
Dong Jia Shi | 6238f92 | 2018-05-23 04:56:44 +0200 | [diff] [blame] | 545 | |
| 546 | out_unpin: |
| 547 | pfn_array_table_unpin_free(pat, cp->mdev); |
| 548 | out_init: |
| 549 | ccw->cda = 0; |
| 550 | return ret; |
Dong Jia Shi | 0a19e61 | 2017-03-17 04:17:32 +0100 | [diff] [blame] | 551 | } |
| 552 | |
| 553 | static int ccwchain_fetch_idal(struct ccwchain *chain, |
| 554 | int idx, |
| 555 | struct channel_program *cp) |
| 556 | { |
| 557 | struct ccw1 *ccw; |
| 558 | struct pfn_array_table *pat; |
| 559 | unsigned long *idaws; |
| 560 | u64 idaw_iova; |
| 561 | unsigned int idaw_nr, idaw_len; |
| 562 | int i, ret; |
| 563 | |
| 564 | ccw = chain->ch_ccw + idx; |
| 565 | |
Dong Jia Shi | 4cebc5d | 2017-10-11 04:38:22 +0200 | [diff] [blame] | 566 | if (!ccw->count) |
| 567 | return 0; |
| 568 | |
Dong Jia Shi | 0a19e61 | 2017-03-17 04:17:32 +0100 | [diff] [blame] | 569 | /* Calculate size of idaws. */ |
| 570 | ret = copy_from_iova(cp->mdev, &idaw_iova, ccw->cda, sizeof(idaw_iova)); |
| 571 | if (ret) |
| 572 | return ret; |
| 573 | idaw_nr = idal_nr_words((void *)(idaw_iova), ccw->count); |
| 574 | idaw_len = idaw_nr * sizeof(*idaws); |
| 575 | |
| 576 | /* Pin data page(s) in memory. */ |
| 577 | pat = chain->ch_pat + idx; |
| 578 | ret = pfn_array_table_init(pat, idaw_nr); |
| 579 | if (ret) |
Dong Jia Shi | 6238f92 | 2018-05-23 04:56:44 +0200 | [diff] [blame] | 580 | goto out_init; |
Dong Jia Shi | 0a19e61 | 2017-03-17 04:17:32 +0100 | [diff] [blame] | 581 | |
| 582 | /* Translate idal ccw to use new allocated idaws. */ |
| 583 | idaws = kzalloc(idaw_len, GFP_DMA | GFP_KERNEL); |
| 584 | if (!idaws) { |
| 585 | ret = -ENOMEM; |
| 586 | goto out_unpin; |
| 587 | } |
| 588 | |
| 589 | ret = copy_from_iova(cp->mdev, idaws, ccw->cda, idaw_len); |
| 590 | if (ret) |
| 591 | goto out_free_idaws; |
| 592 | |
| 593 | ccw->cda = virt_to_phys(idaws); |
| 594 | |
| 595 | for (i = 0; i < idaw_nr; i++) { |
| 596 | idaw_iova = *(idaws + i); |
Dong Jia Shi | 0a19e61 | 2017-03-17 04:17:32 +0100 | [diff] [blame] | 597 | |
| 598 | ret = pfn_array_alloc_pin(pat->pat_pa + i, cp->mdev, |
| 599 | idaw_iova, 1); |
| 600 | if (ret < 0) |
| 601 | goto out_free_idaws; |
| 602 | } |
| 603 | |
| 604 | pfn_array_table_idal_create_words(pat, idaws); |
| 605 | |
| 606 | return 0; |
| 607 | |
| 608 | out_free_idaws: |
| 609 | kfree(idaws); |
| 610 | out_unpin: |
| 611 | pfn_array_table_unpin_free(pat, cp->mdev); |
Dong Jia Shi | 6238f92 | 2018-05-23 04:56:44 +0200 | [diff] [blame] | 612 | out_init: |
| 613 | ccw->cda = 0; |
Dong Jia Shi | 0a19e61 | 2017-03-17 04:17:32 +0100 | [diff] [blame] | 614 | return ret; |
| 615 | } |
| 616 | |
| 617 | /* |
| 618 | * Fetch one ccw. |
| 619 | * To reduce memory copy, we'll pin the cda page in memory, |
| 620 | * and to get rid of the cda 2G limitiaion of ccw1, we'll translate |
| 621 | * direct ccws to idal ccws. |
| 622 | */ |
| 623 | static int ccwchain_fetch_one(struct ccwchain *chain, |
| 624 | int idx, |
| 625 | struct channel_program *cp) |
| 626 | { |
| 627 | struct ccw1 *ccw = chain->ch_ccw + idx; |
| 628 | |
| 629 | if (ccw_is_test(ccw) || ccw_is_noop(ccw)) |
| 630 | return 0; |
| 631 | |
| 632 | if (ccw_is_tic(ccw)) |
| 633 | return ccwchain_fetch_tic(chain, idx, cp); |
| 634 | |
| 635 | if (ccw_is_idal(ccw)) |
| 636 | return ccwchain_fetch_idal(chain, idx, cp); |
| 637 | |
| 638 | return ccwchain_fetch_direct(chain, idx, cp); |
| 639 | } |
| 640 | |
| 641 | /** |
| 642 | * cp_init() - allocate ccwchains for a channel program. |
| 643 | * @cp: channel_program on which to perform the operation |
| 644 | * @mdev: the mediated device to perform pin/unpin operations |
| 645 | * @orb: control block for the channel program from the guest |
| 646 | * |
| 647 | * This creates one or more ccwchain(s), and copies the raw data of |
| 648 | * the target channel program from @orb->cmd.iova to the new ccwchain(s). |
| 649 | * |
| 650 | * Limitations: |
| 651 | * 1. Supports only prefetch enabled mode. |
| 652 | * 2. Supports idal(c64) ccw chaining. |
| 653 | * 3. Supports 4k idaw. |
| 654 | * |
| 655 | * Returns: |
| 656 | * %0 on success and a negative error value on failure. |
| 657 | */ |
| 658 | int cp_init(struct channel_program *cp, struct device *mdev, union orb *orb) |
| 659 | { |
| 660 | u64 iova = orb->cmd.cpa; |
| 661 | struct ccwchain *chain; |
| 662 | int len, ret; |
| 663 | |
| 664 | /* |
| 665 | * XXX: |
| 666 | * Only support prefetch enable mode now. |
Dong Jia Shi | 0a19e61 | 2017-03-17 04:17:32 +0100 | [diff] [blame] | 667 | */ |
Halil Pasic | fb9e788 | 2018-05-16 19:33:42 +0200 | [diff] [blame] | 668 | if (!orb->cmd.pfch) |
Dong Jia Shi | 0a19e61 | 2017-03-17 04:17:32 +0100 | [diff] [blame] | 669 | return -EOPNOTSUPP; |
| 670 | |
| 671 | INIT_LIST_HEAD(&cp->ccwchain_list); |
| 672 | memcpy(&cp->orb, orb, sizeof(*orb)); |
| 673 | cp->mdev = mdev; |
| 674 | |
| 675 | /* Get chain length. */ |
| 676 | len = ccwchain_calc_length(iova, cp); |
| 677 | if (len < 0) |
| 678 | return len; |
| 679 | |
| 680 | /* Alloc mem for the head chain. */ |
| 681 | chain = ccwchain_alloc(cp, len); |
| 682 | if (!chain) |
| 683 | return -ENOMEM; |
| 684 | chain->ch_iova = iova; |
| 685 | |
| 686 | /* Copy the head chain from guest. */ |
| 687 | ret = copy_ccw_from_iova(cp, chain->ch_ccw, iova, len); |
| 688 | if (ret) { |
| 689 | ccwchain_free(chain); |
| 690 | return ret; |
| 691 | } |
| 692 | |
| 693 | /* Now loop for its TICs. */ |
| 694 | ret = ccwchain_loop_tic(chain, cp); |
| 695 | if (ret) |
| 696 | cp_unpin_free(cp); |
Halil Pasic | fb9e788 | 2018-05-16 19:33:42 +0200 | [diff] [blame] | 697 | /* It is safe to force: if not set but idals used |
| 698 | * ccwchain_calc_length returns an error. |
| 699 | */ |
| 700 | cp->orb.cmd.c64 = 1; |
Dong Jia Shi | 0a19e61 | 2017-03-17 04:17:32 +0100 | [diff] [blame] | 701 | |
| 702 | return ret; |
| 703 | } |
| 704 | |
| 705 | |
| 706 | /** |
| 707 | * cp_free() - free resources for channel program. |
| 708 | * @cp: channel_program on which to perform the operation |
| 709 | * |
| 710 | * This unpins the memory pages and frees the memory space occupied by |
| 711 | * @cp, which must have been returned by a previous call to cp_init(). |
| 712 | * Otherwise, undefined behavior occurs. |
| 713 | */ |
| 714 | void cp_free(struct channel_program *cp) |
| 715 | { |
| 716 | cp_unpin_free(cp); |
| 717 | } |
| 718 | |
| 719 | /** |
| 720 | * cp_prefetch() - translate a guest physical address channel program to |
| 721 | * a real-device runnable channel program. |
| 722 | * @cp: channel_program on which to perform the operation |
| 723 | * |
| 724 | * This function translates the guest-physical-address channel program |
| 725 | * and stores the result to ccwchain list. @cp must have been |
| 726 | * initialized by a previous call with cp_init(). Otherwise, undefined |
| 727 | * behavior occurs. |
Halil Pasic | d66a735 | 2018-04-24 13:26:56 +0200 | [diff] [blame] | 728 | * For each chain composing the channel program: |
| 729 | * - On entry ch_len holds the count of CCWs to be translated. |
| 730 | * - On exit ch_len is adjusted to the count of successfully translated CCWs. |
| 731 | * This allows cp_free to find in ch_len the count of CCWs to free in a chain. |
Dong Jia Shi | 0a19e61 | 2017-03-17 04:17:32 +0100 | [diff] [blame] | 732 | * |
| 733 | * The S/390 CCW Translation APIS (prefixed by 'cp_') are introduced |
| 734 | * as helpers to do ccw chain translation inside the kernel. Basically |
| 735 | * they accept a channel program issued by a virtual machine, and |
| 736 | * translate the channel program to a real-device runnable channel |
| 737 | * program. |
| 738 | * |
| 739 | * These APIs will copy the ccws into kernel-space buffers, and update |
| 740 | * the guest phsical addresses with their corresponding host physical |
| 741 | * addresses. Then channel I/O device drivers could issue the |
| 742 | * translated channel program to real devices to perform an I/O |
| 743 | * operation. |
| 744 | * |
| 745 | * These interfaces are designed to support translation only for |
| 746 | * channel programs, which are generated and formatted by a |
| 747 | * guest. Thus this will make it possible for things like VFIO to |
| 748 | * leverage the interfaces to passthrough a channel I/O mediated |
| 749 | * device in QEMU. |
| 750 | * |
| 751 | * We support direct ccw chaining by translating them to idal ccws. |
| 752 | * |
| 753 | * Returns: |
| 754 | * %0 on success and a negative error value on failure. |
| 755 | */ |
| 756 | int cp_prefetch(struct channel_program *cp) |
| 757 | { |
| 758 | struct ccwchain *chain; |
| 759 | int len, idx, ret; |
| 760 | |
| 761 | list_for_each_entry(chain, &cp->ccwchain_list, next) { |
| 762 | len = chain->ch_len; |
| 763 | for (idx = 0; idx < len; idx++) { |
| 764 | ret = ccwchain_fetch_one(chain, idx, cp); |
| 765 | if (ret) |
Halil Pasic | d66a735 | 2018-04-24 13:26:56 +0200 | [diff] [blame] | 766 | goto out_err; |
Dong Jia Shi | 0a19e61 | 2017-03-17 04:17:32 +0100 | [diff] [blame] | 767 | } |
| 768 | } |
| 769 | |
| 770 | return 0; |
Halil Pasic | d66a735 | 2018-04-24 13:26:56 +0200 | [diff] [blame] | 771 | out_err: |
| 772 | /* Only cleanup the chain elements that were actually translated. */ |
| 773 | chain->ch_len = idx; |
| 774 | list_for_each_entry_continue(chain, &cp->ccwchain_list, next) { |
| 775 | chain->ch_len = 0; |
| 776 | } |
| 777 | return ret; |
Dong Jia Shi | 0a19e61 | 2017-03-17 04:17:32 +0100 | [diff] [blame] | 778 | } |
| 779 | |
| 780 | /** |
| 781 | * cp_get_orb() - get the orb of the channel program |
| 782 | * @cp: channel_program on which to perform the operation |
| 783 | * @intparm: new intparm for the returned orb |
| 784 | * @lpm: candidate value of the logical-path mask for the returned orb |
| 785 | * |
| 786 | * This function returns the address of the updated orb of the channel |
| 787 | * program. Channel I/O device drivers could use this orb to issue a |
| 788 | * ssch. |
| 789 | */ |
| 790 | union orb *cp_get_orb(struct channel_program *cp, u32 intparm, u8 lpm) |
| 791 | { |
| 792 | union orb *orb; |
| 793 | struct ccwchain *chain; |
| 794 | struct ccw1 *cpa; |
| 795 | |
| 796 | orb = &cp->orb; |
| 797 | |
| 798 | orb->cmd.intparm = intparm; |
| 799 | orb->cmd.fmt = 1; |
| 800 | orb->cmd.key = PAGE_DEFAULT_KEY >> 4; |
| 801 | |
| 802 | if (orb->cmd.lpm == 0) |
| 803 | orb->cmd.lpm = lpm; |
| 804 | |
| 805 | chain = list_first_entry(&cp->ccwchain_list, struct ccwchain, next); |
| 806 | cpa = chain->ch_ccw; |
| 807 | orb->cmd.cpa = (__u32) __pa(cpa); |
| 808 | |
| 809 | return orb; |
| 810 | } |
| 811 | |
| 812 | /** |
| 813 | * cp_update_scsw() - update scsw for a channel program. |
| 814 | * @cp: channel_program on which to perform the operation |
| 815 | * @scsw: I/O results of the channel program and also the target to be |
| 816 | * updated |
| 817 | * |
| 818 | * @scsw contains the I/O results of the channel program that pointed |
| 819 | * to by @cp. However what @scsw->cpa stores is a host physical |
| 820 | * address, which is meaningless for the guest, which is waiting for |
| 821 | * the I/O results. |
| 822 | * |
| 823 | * This function updates @scsw->cpa to its coressponding guest physical |
| 824 | * address. |
| 825 | */ |
| 826 | void cp_update_scsw(struct channel_program *cp, union scsw *scsw) |
| 827 | { |
| 828 | struct ccwchain *chain; |
| 829 | u32 cpa = scsw->cmd.cpa; |
| 830 | u32 ccw_head, ccw_tail; |
| 831 | |
| 832 | /* |
| 833 | * LATER: |
| 834 | * For now, only update the cmd.cpa part. We may need to deal with |
| 835 | * other portions of the schib as well, even if we don't return them |
| 836 | * in the ioctl directly. Path status changes etc. |
| 837 | */ |
| 838 | list_for_each_entry(chain, &cp->ccwchain_list, next) { |
| 839 | ccw_head = (u32)(u64)chain->ch_ccw; |
| 840 | ccw_tail = (u32)(u64)(chain->ch_ccw + chain->ch_len - 1); |
| 841 | |
| 842 | if ((ccw_head <= cpa) && (cpa <= ccw_tail)) { |
| 843 | /* |
| 844 | * (cpa - ccw_head) is the offset value of the host |
| 845 | * physical ccw to its chain head. |
| 846 | * Adding this value to the guest physical ccw chain |
| 847 | * head gets us the guest cpa. |
| 848 | */ |
| 849 | cpa = chain->ch_iova + (cpa - ccw_head); |
| 850 | break; |
| 851 | } |
| 852 | } |
| 853 | |
| 854 | scsw->cmd.cpa = cpa; |
| 855 | } |
| 856 | |
| 857 | /** |
| 858 | * cp_iova_pinned() - check if an iova is pinned for a ccw chain. |
Sebastian Ott | 364e3f9 | 2018-01-29 12:55:29 +0100 | [diff] [blame] | 859 | * @cp: channel_program on which to perform the operation |
Dong Jia Shi | 0a19e61 | 2017-03-17 04:17:32 +0100 | [diff] [blame] | 860 | * @iova: the iova to check |
| 861 | * |
| 862 | * If the @iova is currently pinned for the ccw chain, return true; |
| 863 | * else return false. |
| 864 | */ |
| 865 | bool cp_iova_pinned(struct channel_program *cp, u64 iova) |
| 866 | { |
| 867 | struct ccwchain *chain; |
| 868 | int i; |
| 869 | |
| 870 | list_for_each_entry(chain, &cp->ccwchain_list, next) { |
| 871 | for (i = 0; i < chain->ch_len; i++) |
| 872 | if (pfn_array_table_iova_pinned(chain->ch_pat + i, |
| 873 | iova)) |
| 874 | return true; |
| 875 | } |
| 876 | |
| 877 | return false; |
| 878 | } |