1 /* 2 * Copyright 2016-17 IBM Corp. 3 * 4 * This program is free software; you can redistribute it and/or 5 * modify it under the terms of the GNU General Public License 6 * as published by the Free Software Foundation; either version 7 * 2 of the License, or (at your option) any later version. 8 */ 9 10 #define pr_fmt(fmt) "vas: " fmt 11 12 #include <linux/types.h> 13 #include <linux/mutex.h> 14 #include <linux/slab.h> 15 #include <linux/io.h> 16 #include <linux/log2.h> 17 #include <linux/rcupdate.h> 18 #include <linux/cred.h> 19 20 #include "vas.h" 21 #include "copy-paste.h" 22 23 /* 24 * Compute the paste address region for the window @window using the 25 * ->paste_base_addr and ->paste_win_id_shift we got from device tree. 26 */ 27 static void compute_paste_address(struct vas_window *window, u64 *addr, int *len) 28 { 29 int winid; 30 u64 base, shift; 31 32 base = window->vinst->paste_base_addr; 33 shift = window->vinst->paste_win_id_shift; 34 winid = window->winid; 35 36 *addr = base + (winid << shift); 37 if (len) 38 *len = PAGE_SIZE; 39 40 pr_debug("Txwin #%d: Paste addr 0x%llx\n", winid, *addr); 41 } 42 43 u64 vas_win_paste_addr(struct vas_window *win) 44 { 45 u64 addr; 46 47 compute_paste_address(win, &addr, NULL); 48 49 return addr; 50 } 51 EXPORT_SYMBOL(vas_win_paste_addr); 52 53 static inline void get_hvwc_mmio_bar(struct vas_window *window, 54 u64 *start, int *len) 55 { 56 u64 pbaddr; 57 58 pbaddr = window->vinst->hvwc_bar_start; 59 *start = pbaddr + window->winid * VAS_HVWC_SIZE; 60 *len = VAS_HVWC_SIZE; 61 } 62 63 static inline void get_uwc_mmio_bar(struct vas_window *window, 64 u64 *start, int *len) 65 { 66 u64 pbaddr; 67 68 pbaddr = window->vinst->uwc_bar_start; 69 *start = pbaddr + window->winid * VAS_UWC_SIZE; 70 *len = VAS_UWC_SIZE; 71 } 72 73 /* 74 * Map the paste bus address of the given send window into kernel address 75 * space. Unlike MMIO regions (map_mmio_region() below), paste region must 76 * be mapped cache-able and is only applicable to send windows. 77 */ 78 static void *map_paste_region(struct vas_window *txwin) 79 { 80 int len; 81 void *map; 82 char *name; 83 u64 start; 84 85 name = kasprintf(GFP_KERNEL, "window-v%d-w%d", txwin->vinst->vas_id, 86 txwin->winid); 87 if (!name) 88 goto free_name; 89 90 txwin->paste_addr_name = name; 91 compute_paste_address(txwin, &start, &len); 92 93 if (!request_mem_region(start, len, name)) { 94 pr_devel("%s(): request_mem_region(0x%llx, %d) failed\n", 95 __func__, start, len); 96 goto free_name; 97 } 98 99 map = ioremap_cache(start, len); 100 if (!map) { 101 pr_devel("%s(): ioremap_cache(0x%llx, %d) failed\n", __func__, 102 start, len); 103 goto free_name; 104 } 105 106 pr_devel("Mapped paste addr 0x%llx to kaddr 0x%p\n", start, map); 107 return map; 108 109 free_name: 110 kfree(name); 111 return ERR_PTR(-ENOMEM); 112 } 113 114 static void *map_mmio_region(char *name, u64 start, int len) 115 { 116 void *map; 117 118 if (!request_mem_region(start, len, name)) { 119 pr_devel("%s(): request_mem_region(0x%llx, %d) failed\n", 120 __func__, start, len); 121 return NULL; 122 } 123 124 map = ioremap(start, len); 125 if (!map) { 126 pr_devel("%s(): ioremap(0x%llx, %d) failed\n", __func__, start, 127 len); 128 return NULL; 129 } 130 131 return map; 132 } 133 134 static void unmap_region(void *addr, u64 start, int len) 135 { 136 iounmap(addr); 137 release_mem_region((phys_addr_t)start, len); 138 } 139 140 /* 141 * Unmap the paste address region for a window. 142 */ 143 static void unmap_paste_region(struct vas_window *window) 144 { 145 int len; 146 u64 busaddr_start; 147 148 if (window->paste_kaddr) { 149 compute_paste_address(window, &busaddr_start, &len); 150 unmap_region(window->paste_kaddr, busaddr_start, len); 151 window->paste_kaddr = NULL; 152 kfree(window->paste_addr_name); 153 window->paste_addr_name = NULL; 154 } 155 } 156 157 /* 158 * Unmap the MMIO regions for a window. Hold the vas_mutex so we don't 159 * unmap when the window's debugfs dir is in use. This serializes close 160 * of a window even on another VAS instance but since its not a critical 161 * path, just minimize the time we hold the mutex for now. We can add 162 * a per-instance mutex later if necessary. 163 */ 164 static void unmap_winctx_mmio_bars(struct vas_window *window) 165 { 166 int len; 167 void *uwc_map; 168 void *hvwc_map; 169 u64 busaddr_start; 170 171 mutex_lock(&vas_mutex); 172 173 hvwc_map = window->hvwc_map; 174 window->hvwc_map = NULL; 175 176 uwc_map = window->uwc_map; 177 window->uwc_map = NULL; 178 179 mutex_unlock(&vas_mutex); 180 181 if (hvwc_map) { 182 get_hvwc_mmio_bar(window, &busaddr_start, &len); 183 unmap_region(hvwc_map, busaddr_start, len); 184 } 185 186 if (uwc_map) { 187 get_uwc_mmio_bar(window, &busaddr_start, &len); 188 unmap_region(uwc_map, busaddr_start, len); 189 } 190 } 191 192 /* 193 * Find the Hypervisor Window Context (HVWC) MMIO Base Address Region and the 194 * OS/User Window Context (UWC) MMIO Base Address Region for the given window. 195 * Map these bus addresses and save the mapped kernel addresses in @window. 196 */ 197 int map_winctx_mmio_bars(struct vas_window *window) 198 { 199 int len; 200 u64 start; 201 202 get_hvwc_mmio_bar(window, &start, &len); 203 window->hvwc_map = map_mmio_region("HVWCM_Window", start, len); 204 205 get_uwc_mmio_bar(window, &start, &len); 206 window->uwc_map = map_mmio_region("UWCM_Window", start, len); 207 208 if (!window->hvwc_map || !window->uwc_map) { 209 unmap_winctx_mmio_bars(window); 210 return -1; 211 } 212 213 return 0; 214 } 215 216 /* 217 * Reset all valid registers in the HV and OS/User Window Contexts for 218 * the window identified by @window. 219 * 220 * NOTE: We cannot really use a for loop to reset window context. Not all 221 * offsets in a window context are valid registers and the valid 222 * registers are not sequential. And, we can only write to offsets 223 * with valid registers. 224 */ 225 void reset_window_regs(struct vas_window *window) 226 { 227 write_hvwc_reg(window, VREG(LPID), 0ULL); 228 write_hvwc_reg(window, VREG(PID), 0ULL); 229 write_hvwc_reg(window, VREG(XLATE_MSR), 0ULL); 230 write_hvwc_reg(window, VREG(XLATE_LPCR), 0ULL); 231 write_hvwc_reg(window, VREG(XLATE_CTL), 0ULL); 232 write_hvwc_reg(window, VREG(AMR), 0ULL); 233 write_hvwc_reg(window, VREG(SEIDR), 0ULL); 234 write_hvwc_reg(window, VREG(FAULT_TX_WIN), 0ULL); 235 write_hvwc_reg(window, VREG(OSU_INTR_SRC_RA), 0ULL); 236 write_hvwc_reg(window, VREG(HV_INTR_SRC_RA), 0ULL); 237 write_hvwc_reg(window, VREG(PSWID), 0ULL); 238 write_hvwc_reg(window, VREG(LFIFO_BAR), 0ULL); 239 write_hvwc_reg(window, VREG(LDATA_STAMP_CTL), 0ULL); 240 write_hvwc_reg(window, VREG(LDMA_CACHE_CTL), 0ULL); 241 write_hvwc_reg(window, VREG(LRFIFO_PUSH), 0ULL); 242 write_hvwc_reg(window, VREG(CURR_MSG_COUNT), 0ULL); 243 write_hvwc_reg(window, VREG(LNOTIFY_AFTER_COUNT), 0ULL); 244 write_hvwc_reg(window, VREG(LRX_WCRED), 0ULL); 245 write_hvwc_reg(window, VREG(LRX_WCRED_ADDER), 0ULL); 246 write_hvwc_reg(window, VREG(TX_WCRED), 0ULL); 247 write_hvwc_reg(window, VREG(TX_WCRED_ADDER), 0ULL); 248 write_hvwc_reg(window, VREG(LFIFO_SIZE), 0ULL); 249 write_hvwc_reg(window, VREG(WINCTL), 0ULL); 250 write_hvwc_reg(window, VREG(WIN_STATUS), 0ULL); 251 write_hvwc_reg(window, VREG(WIN_CTX_CACHING_CTL), 0ULL); 252 write_hvwc_reg(window, VREG(TX_RSVD_BUF_COUNT), 0ULL); 253 write_hvwc_reg(window, VREG(LRFIFO_WIN_PTR), 0ULL); 254 write_hvwc_reg(window, VREG(LNOTIFY_CTL), 0ULL); 255 write_hvwc_reg(window, VREG(LNOTIFY_PID), 0ULL); 256 write_hvwc_reg(window, VREG(LNOTIFY_LPID), 0ULL); 257 write_hvwc_reg(window, VREG(LNOTIFY_TID), 0ULL); 258 write_hvwc_reg(window, VREG(LNOTIFY_SCOPE), 0ULL); 259 write_hvwc_reg(window, VREG(NX_UTIL_ADDER), 0ULL); 260 261 /* Skip read-only registers: NX_UTIL and NX_UTIL_SE */ 262 263 /* 264 * The send and receive window credit adder registers are also 265 * accessible from HVWC and have been initialized above. We don't 266 * need to initialize from the OS/User Window Context, so skip 267 * following calls: 268 * 269 * write_uwc_reg(window, VREG(TX_WCRED_ADDER), 0ULL); 270 * write_uwc_reg(window, VREG(LRX_WCRED_ADDER), 0ULL); 271 */ 272 } 273 274 /* 275 * Initialize window context registers related to Address Translation. 276 * These registers are common to send/receive windows although they 277 * differ for user/kernel windows. As we resolve the TODOs we may 278 * want to add fields to vas_winctx and move the initialization to 279 * init_vas_winctx_regs(). 280 */ 281 static void init_xlate_regs(struct vas_window *window, bool user_win) 282 { 283 u64 lpcr, val; 284 285 /* 286 * MSR_TA, MSR_US are false for both kernel and user. 287 * MSR_DR and MSR_PR are false for kernel. 288 */ 289 val = 0ULL; 290 val = SET_FIELD(VAS_XLATE_MSR_HV, val, 1); 291 val = SET_FIELD(VAS_XLATE_MSR_SF, val, 1); 292 if (user_win) { 293 val = SET_FIELD(VAS_XLATE_MSR_DR, val, 1); 294 val = SET_FIELD(VAS_XLATE_MSR_PR, val, 1); 295 } 296 write_hvwc_reg(window, VREG(XLATE_MSR), val); 297 298 lpcr = mfspr(SPRN_LPCR); 299 val = 0ULL; 300 /* 301 * NOTE: From Section 5.7.8.1 Segment Lookaside Buffer of the 302 * Power ISA, v3.0B, Page size encoding is 0 = 4KB, 5 = 64KB. 303 * 304 * NOTE: From Section 1.3.1, Address Translation Context of the 305 * Nest MMU Workbook, LPCR_SC should be 0 for Power9. 306 */ 307 val = SET_FIELD(VAS_XLATE_LPCR_PAGE_SIZE, val, 5); 308 val = SET_FIELD(VAS_XLATE_LPCR_ISL, val, lpcr & LPCR_ISL); 309 val = SET_FIELD(VAS_XLATE_LPCR_TC, val, lpcr & LPCR_TC); 310 val = SET_FIELD(VAS_XLATE_LPCR_SC, val, 0); 311 write_hvwc_reg(window, VREG(XLATE_LPCR), val); 312 313 /* 314 * Section 1.3.1 (Address translation Context) of NMMU workbook. 315 * 0b00 Hashed Page Table mode 316 * 0b01 Reserved 317 * 0b10 Radix on HPT 318 * 0b11 Radix on Radix 319 */ 320 val = 0ULL; 321 val = SET_FIELD(VAS_XLATE_MODE, val, radix_enabled() ? 3 : 2); 322 write_hvwc_reg(window, VREG(XLATE_CTL), val); 323 324 /* 325 * TODO: Can we mfspr(AMR) even for user windows? 326 */ 327 val = 0ULL; 328 val = SET_FIELD(VAS_AMR, val, mfspr(SPRN_AMR)); 329 write_hvwc_reg(window, VREG(AMR), val); 330 331 val = 0ULL; 332 val = SET_FIELD(VAS_SEIDR, val, 0); 333 write_hvwc_reg(window, VREG(SEIDR), val); 334 } 335 336 /* 337 * Initialize Reserved Send Buffer Count for the send window. It involves 338 * writing to the register, reading it back to confirm that the hardware 339 * has enough buffers to reserve. See section 1.3.1.2.1 of VAS workbook. 340 * 341 * Since we can only make a best-effort attempt to fulfill the request, 342 * we don't return any errors if we cannot. 343 * 344 * TODO: Reserved (aka dedicated) send buffers are not supported yet. 345 */ 346 static void init_rsvd_tx_buf_count(struct vas_window *txwin, 347 struct vas_winctx *winctx) 348 { 349 write_hvwc_reg(txwin, VREG(TX_RSVD_BUF_COUNT), 0ULL); 350 } 351 352 /* 353 * init_winctx_regs() 354 * Initialize window context registers for a receive window. 355 * Except for caching control and marking window open, the registers 356 * are initialized in the order listed in Section 3.1.4 (Window Context 357 * Cache Register Details) of the VAS workbook although they don't need 358 * to be. 359 * 360 * Design note: For NX receive windows, NX allocates the FIFO buffer in OPAL 361 * (so that it can get a large contiguous area) and passes that buffer 362 * to kernel via device tree. We now write that buffer address to the 363 * FIFO BAR. Would it make sense to do this all in OPAL? i.e have OPAL 364 * write the per-chip RX FIFO addresses to the windows during boot-up 365 * as a one-time task? That could work for NX but what about other 366 * receivers? Let the receivers tell us the rx-fifo buffers for now. 367 */ 368 int init_winctx_regs(struct vas_window *window, struct vas_winctx *winctx) 369 { 370 u64 val; 371 int fifo_size; 372 373 reset_window_regs(window); 374 375 val = 0ULL; 376 val = SET_FIELD(VAS_LPID, val, winctx->lpid); 377 write_hvwc_reg(window, VREG(LPID), val); 378 379 val = 0ULL; 380 val = SET_FIELD(VAS_PID_ID, val, winctx->pidr); 381 write_hvwc_reg(window, VREG(PID), val); 382 383 init_xlate_regs(window, winctx->user_win); 384 385 val = 0ULL; 386 val = SET_FIELD(VAS_FAULT_TX_WIN, val, 0); 387 write_hvwc_reg(window, VREG(FAULT_TX_WIN), val); 388 389 /* In PowerNV, interrupts go to HV. */ 390 write_hvwc_reg(window, VREG(OSU_INTR_SRC_RA), 0ULL); 391 392 val = 0ULL; 393 val = SET_FIELD(VAS_HV_INTR_SRC_RA, val, winctx->irq_port); 394 write_hvwc_reg(window, VREG(HV_INTR_SRC_RA), val); 395 396 val = 0ULL; 397 val = SET_FIELD(VAS_PSWID_EA_HANDLE, val, winctx->pswid); 398 write_hvwc_reg(window, VREG(PSWID), val); 399 400 write_hvwc_reg(window, VREG(SPARE1), 0ULL); 401 write_hvwc_reg(window, VREG(SPARE2), 0ULL); 402 write_hvwc_reg(window, VREG(SPARE3), 0ULL); 403 404 /* 405 * NOTE: VAS expects the FIFO address to be copied into the LFIFO_BAR 406 * register as is - do NOT shift the address into VAS_LFIFO_BAR 407 * bit fields! Ok to set the page migration select fields - 408 * VAS ignores the lower 10+ bits in the address anyway, because 409 * the minimum FIFO size is 1K? 410 * 411 * See also: Design note in function header. 412 */ 413 val = __pa(winctx->rx_fifo); 414 val = SET_FIELD(VAS_PAGE_MIGRATION_SELECT, val, 0); 415 write_hvwc_reg(window, VREG(LFIFO_BAR), val); 416 417 val = 0ULL; 418 val = SET_FIELD(VAS_LDATA_STAMP, val, winctx->data_stamp); 419 write_hvwc_reg(window, VREG(LDATA_STAMP_CTL), val); 420 421 val = 0ULL; 422 val = SET_FIELD(VAS_LDMA_TYPE, val, winctx->dma_type); 423 val = SET_FIELD(VAS_LDMA_FIFO_DISABLE, val, winctx->fifo_disable); 424 write_hvwc_reg(window, VREG(LDMA_CACHE_CTL), val); 425 426 write_hvwc_reg(window, VREG(LRFIFO_PUSH), 0ULL); 427 write_hvwc_reg(window, VREG(CURR_MSG_COUNT), 0ULL); 428 write_hvwc_reg(window, VREG(LNOTIFY_AFTER_COUNT), 0ULL); 429 430 val = 0ULL; 431 val = SET_FIELD(VAS_LRX_WCRED, val, winctx->wcreds_max); 432 write_hvwc_reg(window, VREG(LRX_WCRED), val); 433 434 val = 0ULL; 435 val = SET_FIELD(VAS_TX_WCRED, val, winctx->wcreds_max); 436 write_hvwc_reg(window, VREG(TX_WCRED), val); 437 438 write_hvwc_reg(window, VREG(LRX_WCRED_ADDER), 0ULL); 439 write_hvwc_reg(window, VREG(TX_WCRED_ADDER), 0ULL); 440 441 fifo_size = winctx->rx_fifo_size / 1024; 442 443 val = 0ULL; 444 val = SET_FIELD(VAS_LFIFO_SIZE, val, ilog2(fifo_size)); 445 write_hvwc_reg(window, VREG(LFIFO_SIZE), val); 446 447 /* Update window control and caching control registers last so 448 * we mark the window open only after fully initializing it and 449 * pushing context to cache. 450 */ 451 452 write_hvwc_reg(window, VREG(WIN_STATUS), 0ULL); 453 454 init_rsvd_tx_buf_count(window, winctx); 455 456 /* for a send window, point to the matching receive window */ 457 val = 0ULL; 458 val = SET_FIELD(VAS_LRX_WIN_ID, val, winctx->rx_win_id); 459 write_hvwc_reg(window, VREG(LRFIFO_WIN_PTR), val); 460 461 write_hvwc_reg(window, VREG(SPARE4), 0ULL); 462 463 val = 0ULL; 464 val = SET_FIELD(VAS_NOTIFY_DISABLE, val, winctx->notify_disable); 465 val = SET_FIELD(VAS_INTR_DISABLE, val, winctx->intr_disable); 466 val = SET_FIELD(VAS_NOTIFY_EARLY, val, winctx->notify_early); 467 val = SET_FIELD(VAS_NOTIFY_OSU_INTR, val, winctx->notify_os_intr_reg); 468 write_hvwc_reg(window, VREG(LNOTIFY_CTL), val); 469 470 val = 0ULL; 471 val = SET_FIELD(VAS_LNOTIFY_PID, val, winctx->lnotify_pid); 472 write_hvwc_reg(window, VREG(LNOTIFY_PID), val); 473 474 val = 0ULL; 475 val = SET_FIELD(VAS_LNOTIFY_LPID, val, winctx->lnotify_lpid); 476 write_hvwc_reg(window, VREG(LNOTIFY_LPID), val); 477 478 val = 0ULL; 479 val = SET_FIELD(VAS_LNOTIFY_TID, val, winctx->lnotify_tid); 480 write_hvwc_reg(window, VREG(LNOTIFY_TID), val); 481 482 val = 0ULL; 483 val = SET_FIELD(VAS_LNOTIFY_MIN_SCOPE, val, winctx->min_scope); 484 val = SET_FIELD(VAS_LNOTIFY_MAX_SCOPE, val, winctx->max_scope); 485 write_hvwc_reg(window, VREG(LNOTIFY_SCOPE), val); 486 487 /* Skip read-only registers NX_UTIL and NX_UTIL_SE */ 488 489 write_hvwc_reg(window, VREG(SPARE5), 0ULL); 490 write_hvwc_reg(window, VREG(NX_UTIL_ADDER), 0ULL); 491 write_hvwc_reg(window, VREG(SPARE6), 0ULL); 492 493 /* Finally, push window context to memory and... */ 494 val = 0ULL; 495 val = SET_FIELD(VAS_PUSH_TO_MEM, val, 1); 496 write_hvwc_reg(window, VREG(WIN_CTX_CACHING_CTL), val); 497 498 /* ... mark the window open for business */ 499 val = 0ULL; 500 val = SET_FIELD(VAS_WINCTL_REJ_NO_CREDIT, val, winctx->rej_no_credit); 501 val = SET_FIELD(VAS_WINCTL_PIN, val, winctx->pin_win); 502 val = SET_FIELD(VAS_WINCTL_TX_WCRED_MODE, val, winctx->tx_wcred_mode); 503 val = SET_FIELD(VAS_WINCTL_RX_WCRED_MODE, val, winctx->rx_wcred_mode); 504 val = SET_FIELD(VAS_WINCTL_TX_WORD_MODE, val, winctx->tx_word_mode); 505 val = SET_FIELD(VAS_WINCTL_RX_WORD_MODE, val, winctx->rx_word_mode); 506 val = SET_FIELD(VAS_WINCTL_FAULT_WIN, val, winctx->fault_win); 507 val = SET_FIELD(VAS_WINCTL_NX_WIN, val, winctx->nx_win); 508 val = SET_FIELD(VAS_WINCTL_OPEN, val, 1); 509 write_hvwc_reg(window, VREG(WINCTL), val); 510 511 return 0; 512 } 513 514 static DEFINE_SPINLOCK(vas_ida_lock); 515 516 static void vas_release_window_id(struct ida *ida, int winid) 517 { 518 spin_lock(&vas_ida_lock); 519 ida_remove(ida, winid); 520 spin_unlock(&vas_ida_lock); 521 } 522 523 static int vas_assign_window_id(struct ida *ida) 524 { 525 int rc, winid; 526 527 do { 528 rc = ida_pre_get(ida, GFP_KERNEL); 529 if (!rc) 530 return -EAGAIN; 531 532 spin_lock(&vas_ida_lock); 533 rc = ida_get_new(ida, &winid); 534 spin_unlock(&vas_ida_lock); 535 } while (rc == -EAGAIN); 536 537 if (rc) 538 return rc; 539 540 if (winid > VAS_WINDOWS_PER_CHIP) { 541 pr_err("Too many (%d) open windows\n", winid); 542 vas_release_window_id(ida, winid); 543 return -EAGAIN; 544 } 545 546 return winid; 547 } 548 549 static void vas_window_free(struct vas_window *window) 550 { 551 int winid = window->winid; 552 struct vas_instance *vinst = window->vinst; 553 554 unmap_winctx_mmio_bars(window); 555 556 vas_window_free_dbgdir(window); 557 558 kfree(window); 559 560 vas_release_window_id(&vinst->ida, winid); 561 } 562 563 static struct vas_window *vas_window_alloc(struct vas_instance *vinst) 564 { 565 int winid; 566 struct vas_window *window; 567 568 winid = vas_assign_window_id(&vinst->ida); 569 if (winid < 0) 570 return ERR_PTR(winid); 571 572 window = kzalloc(sizeof(*window), GFP_KERNEL); 573 if (!window) 574 goto out_free; 575 576 window->vinst = vinst; 577 window->winid = winid; 578 579 if (map_winctx_mmio_bars(window)) 580 goto out_free; 581 582 vas_window_init_dbgdir(window); 583 584 return window; 585 586 out_free: 587 kfree(window); 588 vas_release_window_id(&vinst->ida, winid); 589 return ERR_PTR(-ENOMEM); 590 } 591 592 static void put_rx_win(struct vas_window *rxwin) 593 { 594 /* Better not be a send window! */ 595 WARN_ON_ONCE(rxwin->tx_win); 596 597 atomic_dec(&rxwin->num_txwins); 598 } 599 600 /* 601 * Get the VAS receive window associated with NX engine identified 602 * by @cop and if applicable, @pswid. 603 * 604 * See also function header of set_vinst_win(). 605 */ 606 static struct vas_window *get_vinst_rxwin(struct vas_instance *vinst, 607 enum vas_cop_type cop, u32 pswid) 608 { 609 struct vas_window *rxwin; 610 611 mutex_lock(&vinst->mutex); 612 613 if (cop == VAS_COP_TYPE_842 || cop == VAS_COP_TYPE_842_HIPRI) 614 rxwin = vinst->rxwin[cop] ?: ERR_PTR(-EINVAL); 615 else 616 rxwin = ERR_PTR(-EINVAL); 617 618 if (!IS_ERR(rxwin)) 619 atomic_inc(&rxwin->num_txwins); 620 621 mutex_unlock(&vinst->mutex); 622 623 return rxwin; 624 } 625 626 /* 627 * We have two tables of windows in a VAS instance. The first one, 628 * ->windows[], contains all the windows in the instance and allows 629 * looking up a window by its id. It is used to look up send windows 630 * during fault handling and receive windows when pairing user space 631 * send/receive windows. 632 * 633 * The second table, ->rxwin[], contains receive windows that are 634 * associated with NX engines. This table has VAS_COP_TYPE_MAX 635 * entries and is used to look up a receive window by its 636 * coprocessor type. 637 * 638 * Here, we save @window in the ->windows[] table. If it is a receive 639 * window, we also save the window in the ->rxwin[] table. 640 */ 641 static void set_vinst_win(struct vas_instance *vinst, 642 struct vas_window *window) 643 { 644 int id = window->winid; 645 646 mutex_lock(&vinst->mutex); 647 648 /* 649 * There should only be one receive window for a coprocessor type 650 * unless its a user (FTW) window. 651 */ 652 if (!window->user_win && !window->tx_win) { 653 WARN_ON_ONCE(vinst->rxwin[window->cop]); 654 vinst->rxwin[window->cop] = window; 655 } 656 657 WARN_ON_ONCE(vinst->windows[id] != NULL); 658 vinst->windows[id] = window; 659 660 mutex_unlock(&vinst->mutex); 661 } 662 663 /* 664 * Clear this window from the table(s) of windows for this VAS instance. 665 * See also function header of set_vinst_win(). 666 */ 667 static void clear_vinst_win(struct vas_window *window) 668 { 669 int id = window->winid; 670 struct vas_instance *vinst = window->vinst; 671 672 mutex_lock(&vinst->mutex); 673 674 if (!window->user_win && !window->tx_win) { 675 WARN_ON_ONCE(!vinst->rxwin[window->cop]); 676 vinst->rxwin[window->cop] = NULL; 677 } 678 679 WARN_ON_ONCE(vinst->windows[id] != window); 680 vinst->windows[id] = NULL; 681 682 mutex_unlock(&vinst->mutex); 683 } 684 685 static void init_winctx_for_rxwin(struct vas_window *rxwin, 686 struct vas_rx_win_attr *rxattr, 687 struct vas_winctx *winctx) 688 { 689 /* 690 * We first zero (memset()) all fields and only set non-zero fields. 691 * Following fields are 0/false but maybe deserve a comment: 692 * 693 * ->notify_os_intr_reg In powerNV, send intrs to HV 694 * ->notify_disable False for NX windows 695 * ->intr_disable False for Fault Windows 696 * ->xtra_write False for NX windows 697 * ->notify_early NA for NX windows 698 * ->rsvd_txbuf_count NA for Rx windows 699 * ->lpid, ->pid, ->tid NA for Rx windows 700 */ 701 702 memset(winctx, 0, sizeof(struct vas_winctx)); 703 704 winctx->rx_fifo = rxattr->rx_fifo; 705 winctx->rx_fifo_size = rxattr->rx_fifo_size; 706 winctx->wcreds_max = rxwin->wcreds_max; 707 winctx->pin_win = rxattr->pin_win; 708 709 winctx->nx_win = rxattr->nx_win; 710 winctx->fault_win = rxattr->fault_win; 711 winctx->user_win = rxattr->user_win; 712 winctx->rej_no_credit = rxattr->rej_no_credit; 713 winctx->rx_word_mode = rxattr->rx_win_ord_mode; 714 winctx->tx_word_mode = rxattr->tx_win_ord_mode; 715 winctx->rx_wcred_mode = rxattr->rx_wcred_mode; 716 winctx->tx_wcred_mode = rxattr->tx_wcred_mode; 717 winctx->notify_early = rxattr->notify_early; 718 719 if (winctx->nx_win) { 720 winctx->data_stamp = true; 721 winctx->intr_disable = true; 722 winctx->pin_win = true; 723 724 WARN_ON_ONCE(winctx->fault_win); 725 WARN_ON_ONCE(!winctx->rx_word_mode); 726 WARN_ON_ONCE(!winctx->tx_word_mode); 727 WARN_ON_ONCE(winctx->notify_after_count); 728 } else if (winctx->fault_win) { 729 winctx->notify_disable = true; 730 } else if (winctx->user_win) { 731 /* 732 * Section 1.8.1 Low Latency Core-Core Wake up of 733 * the VAS workbook: 734 * 735 * - disable credit checks ([tr]x_wcred_mode = false) 736 * - disable FIFO writes 737 * - enable ASB_Notify, disable interrupt 738 */ 739 winctx->fifo_disable = true; 740 winctx->intr_disable = true; 741 winctx->rx_fifo = NULL; 742 } 743 744 winctx->lnotify_lpid = rxattr->lnotify_lpid; 745 winctx->lnotify_pid = rxattr->lnotify_pid; 746 winctx->lnotify_tid = rxattr->lnotify_tid; 747 winctx->pswid = rxattr->pswid; 748 winctx->dma_type = VAS_DMA_TYPE_INJECT; 749 winctx->tc_mode = rxattr->tc_mode; 750 751 winctx->min_scope = VAS_SCOPE_LOCAL; 752 winctx->max_scope = VAS_SCOPE_VECTORED_GROUP; 753 } 754 755 static bool rx_win_args_valid(enum vas_cop_type cop, 756 struct vas_rx_win_attr *attr) 757 { 758 pr_debug("Rxattr: fault %d, notify %d, intr %d, early %d, fifo %d\n", 759 attr->fault_win, attr->notify_disable, 760 attr->intr_disable, attr->notify_early, 761 attr->rx_fifo_size); 762 763 if (cop >= VAS_COP_TYPE_MAX) 764 return false; 765 766 if (cop != VAS_COP_TYPE_FTW && 767 attr->rx_fifo_size < VAS_RX_FIFO_SIZE_MIN) 768 return false; 769 770 if (attr->rx_fifo_size > VAS_RX_FIFO_SIZE_MAX) 771 return false; 772 773 if (attr->wcreds_max > VAS_RX_WCREDS_MAX) 774 return false; 775 776 if (attr->nx_win) { 777 /* cannot be fault or user window if it is nx */ 778 if (attr->fault_win || attr->user_win) 779 return false; 780 /* 781 * Section 3.1.4.32: NX Windows must not disable notification, 782 * and must not enable interrupts or early notification. 783 */ 784 if (attr->notify_disable || !attr->intr_disable || 785 attr->notify_early) 786 return false; 787 } else if (attr->fault_win) { 788 /* cannot be both fault and user window */ 789 if (attr->user_win) 790 return false; 791 792 /* 793 * Section 3.1.4.32: Fault windows must disable notification 794 * but not interrupts. 795 */ 796 if (!attr->notify_disable || attr->intr_disable) 797 return false; 798 799 } else if (attr->user_win) { 800 /* 801 * User receive windows are only for fast-thread-wakeup 802 * (FTW). They don't need a FIFO and must disable interrupts 803 */ 804 if (attr->rx_fifo || attr->rx_fifo_size || !attr->intr_disable) 805 return false; 806 } else { 807 /* Rx window must be one of NX or Fault or User window. */ 808 return false; 809 } 810 811 return true; 812 } 813 814 void vas_init_rx_win_attr(struct vas_rx_win_attr *rxattr, enum vas_cop_type cop) 815 { 816 memset(rxattr, 0, sizeof(*rxattr)); 817 818 if (cop == VAS_COP_TYPE_842 || cop == VAS_COP_TYPE_842_HIPRI) { 819 rxattr->pin_win = true; 820 rxattr->nx_win = true; 821 rxattr->fault_win = false; 822 rxattr->intr_disable = true; 823 rxattr->rx_wcred_mode = true; 824 rxattr->tx_wcred_mode = true; 825 rxattr->rx_win_ord_mode = true; 826 rxattr->tx_win_ord_mode = true; 827 } else if (cop == VAS_COP_TYPE_FAULT) { 828 rxattr->pin_win = true; 829 rxattr->fault_win = true; 830 rxattr->notify_disable = true; 831 rxattr->rx_wcred_mode = true; 832 rxattr->tx_wcred_mode = true; 833 rxattr->rx_win_ord_mode = true; 834 rxattr->tx_win_ord_mode = true; 835 } else if (cop == VAS_COP_TYPE_FTW) { 836 rxattr->user_win = true; 837 rxattr->intr_disable = true; 838 839 /* 840 * As noted in the VAS Workbook we disable credit checks. 841 * If we enable credit checks in the future, we must also 842 * implement a mechanism to return the user credits or new 843 * paste operations will fail. 844 */ 845 } 846 } 847 EXPORT_SYMBOL_GPL(vas_init_rx_win_attr); 848 849 struct vas_window *vas_rx_win_open(int vasid, enum vas_cop_type cop, 850 struct vas_rx_win_attr *rxattr) 851 { 852 struct vas_window *rxwin; 853 struct vas_winctx winctx; 854 struct vas_instance *vinst; 855 856 if (!rx_win_args_valid(cop, rxattr)) 857 return ERR_PTR(-EINVAL); 858 859 vinst = find_vas_instance(vasid); 860 if (!vinst) { 861 pr_devel("vasid %d not found!\n", vasid); 862 return ERR_PTR(-EINVAL); 863 } 864 pr_devel("Found instance %d\n", vasid); 865 866 rxwin = vas_window_alloc(vinst); 867 if (IS_ERR(rxwin)) { 868 pr_devel("Unable to allocate memory for Rx window\n"); 869 return rxwin; 870 } 871 872 rxwin->tx_win = false; 873 rxwin->nx_win = rxattr->nx_win; 874 rxwin->user_win = rxattr->user_win; 875 rxwin->cop = cop; 876 rxwin->wcreds_max = rxattr->wcreds_max ?: VAS_WCREDS_DEFAULT; 877 if (rxattr->user_win) 878 rxwin->pid = task_pid_vnr(current); 879 880 init_winctx_for_rxwin(rxwin, rxattr, &winctx); 881 init_winctx_regs(rxwin, &winctx); 882 883 set_vinst_win(vinst, rxwin); 884 885 return rxwin; 886 } 887 EXPORT_SYMBOL_GPL(vas_rx_win_open); 888 889 void vas_init_tx_win_attr(struct vas_tx_win_attr *txattr, enum vas_cop_type cop) 890 { 891 memset(txattr, 0, sizeof(*txattr)); 892 893 if (cop == VAS_COP_TYPE_842 || cop == VAS_COP_TYPE_842_HIPRI) { 894 txattr->rej_no_credit = false; 895 txattr->rx_wcred_mode = true; 896 txattr->tx_wcred_mode = true; 897 txattr->rx_win_ord_mode = true; 898 txattr->tx_win_ord_mode = true; 899 } else if (cop == VAS_COP_TYPE_FTW) { 900 txattr->user_win = true; 901 } 902 } 903 EXPORT_SYMBOL_GPL(vas_init_tx_win_attr); 904 905 static void init_winctx_for_txwin(struct vas_window *txwin, 906 struct vas_tx_win_attr *txattr, 907 struct vas_winctx *winctx) 908 { 909 /* 910 * We first zero all fields and only set non-zero ones. Following 911 * are some fields set to 0/false for the stated reason: 912 * 913 * ->notify_os_intr_reg In powernv, send intrs to HV 914 * ->rsvd_txbuf_count Not supported yet. 915 * ->notify_disable False for NX windows 916 * ->xtra_write False for NX windows 917 * ->notify_early NA for NX windows 918 * ->lnotify_lpid NA for Tx windows 919 * ->lnotify_pid NA for Tx windows 920 * ->lnotify_tid NA for Tx windows 921 * ->tx_win_cred_mode Ignore for now for NX windows 922 * ->rx_win_cred_mode Ignore for now for NX windows 923 */ 924 memset(winctx, 0, sizeof(struct vas_winctx)); 925 926 winctx->wcreds_max = txwin->wcreds_max; 927 928 winctx->user_win = txattr->user_win; 929 winctx->nx_win = txwin->rxwin->nx_win; 930 winctx->pin_win = txattr->pin_win; 931 winctx->rej_no_credit = txattr->rej_no_credit; 932 winctx->rsvd_txbuf_enable = txattr->rsvd_txbuf_enable; 933 934 winctx->rx_wcred_mode = txattr->rx_wcred_mode; 935 winctx->tx_wcred_mode = txattr->tx_wcred_mode; 936 winctx->rx_word_mode = txattr->rx_win_ord_mode; 937 winctx->tx_word_mode = txattr->tx_win_ord_mode; 938 winctx->rsvd_txbuf_count = txattr->rsvd_txbuf_count; 939 940 if (winctx->nx_win) { 941 winctx->data_stamp = true; 942 winctx->intr_disable = true; 943 } 944 945 winctx->lpid = txattr->lpid; 946 winctx->pidr = txattr->pidr; 947 winctx->rx_win_id = txwin->rxwin->winid; 948 949 winctx->dma_type = VAS_DMA_TYPE_INJECT; 950 winctx->tc_mode = txattr->tc_mode; 951 winctx->min_scope = VAS_SCOPE_LOCAL; 952 winctx->max_scope = VAS_SCOPE_VECTORED_GROUP; 953 954 winctx->pswid = 0; 955 } 956 957 static bool tx_win_args_valid(enum vas_cop_type cop, 958 struct vas_tx_win_attr *attr) 959 { 960 if (attr->tc_mode != VAS_THRESH_DISABLED) 961 return false; 962 963 if (cop > VAS_COP_TYPE_MAX) 964 return false; 965 966 if (attr->wcreds_max > VAS_TX_WCREDS_MAX) 967 return false; 968 969 if (attr->user_win && 970 (cop != VAS_COP_TYPE_FTW || attr->rsvd_txbuf_count)) 971 return false; 972 973 return true; 974 } 975 976 struct vas_window *vas_tx_win_open(int vasid, enum vas_cop_type cop, 977 struct vas_tx_win_attr *attr) 978 { 979 int rc; 980 struct vas_window *txwin; 981 struct vas_window *rxwin; 982 struct vas_winctx winctx; 983 struct vas_instance *vinst; 984 985 if (!tx_win_args_valid(cop, attr)) 986 return ERR_PTR(-EINVAL); 987 988 vinst = find_vas_instance(vasid); 989 if (!vinst) { 990 pr_devel("vasid %d not found!\n", vasid); 991 return ERR_PTR(-EINVAL); 992 } 993 994 rxwin = get_vinst_rxwin(vinst, cop, attr->pswid); 995 if (IS_ERR(rxwin)) { 996 pr_devel("No RxWin for vasid %d, cop %d\n", vasid, cop); 997 return rxwin; 998 } 999 1000 txwin = vas_window_alloc(vinst); 1001 if (IS_ERR(txwin)) { 1002 rc = PTR_ERR(txwin); 1003 goto put_rxwin; 1004 } 1005 1006 txwin->cop = cop; 1007 txwin->tx_win = 1; 1008 txwin->rxwin = rxwin; 1009 txwin->nx_win = txwin->rxwin->nx_win; 1010 txwin->pid = attr->pid; 1011 txwin->user_win = attr->user_win; 1012 txwin->wcreds_max = attr->wcreds_max ?: VAS_WCREDS_DEFAULT; 1013 1014 init_winctx_for_txwin(txwin, attr, &winctx); 1015 1016 init_winctx_regs(txwin, &winctx); 1017 1018 /* 1019 * If its a kernel send window, map the window address into the 1020 * kernel's address space. For user windows, user must issue an 1021 * mmap() to map the window into their address space. 1022 * 1023 * NOTE: If kernel ever resubmits a user CRB after handling a page 1024 * fault, we will need to map this into kernel as well. 1025 */ 1026 if (!txwin->user_win) { 1027 txwin->paste_kaddr = map_paste_region(txwin); 1028 if (IS_ERR(txwin->paste_kaddr)) { 1029 rc = PTR_ERR(txwin->paste_kaddr); 1030 goto free_window; 1031 } 1032 } 1033 1034 set_vinst_win(vinst, txwin); 1035 1036 return txwin; 1037 1038 free_window: 1039 vas_window_free(txwin); 1040 1041 put_rxwin: 1042 put_rx_win(rxwin); 1043 return ERR_PTR(rc); 1044 1045 } 1046 EXPORT_SYMBOL_GPL(vas_tx_win_open); 1047 1048 int vas_copy_crb(void *crb, int offset) 1049 { 1050 return vas_copy(crb, offset); 1051 } 1052 EXPORT_SYMBOL_GPL(vas_copy_crb); 1053 1054 #define RMA_LSMP_REPORT_ENABLE PPC_BIT(53) 1055 int vas_paste_crb(struct vas_window *txwin, int offset, bool re) 1056 { 1057 int rc; 1058 void *addr; 1059 uint64_t val; 1060 1061 /* 1062 * Only NX windows are supported for now and hardware assumes 1063 * report-enable flag is set for NX windows. Ensure software 1064 * complies too. 1065 */ 1066 WARN_ON_ONCE(txwin->nx_win && !re); 1067 1068 addr = txwin->paste_kaddr; 1069 if (re) { 1070 /* 1071 * Set the REPORT_ENABLE bit (equivalent to writing 1072 * to 1K offset of the paste address) 1073 */ 1074 val = SET_FIELD(RMA_LSMP_REPORT_ENABLE, 0ULL, 1); 1075 addr += val; 1076 } 1077 1078 /* 1079 * Map the raw CR value from vas_paste() to an error code (there 1080 * is just pass or fail for now though). 1081 */ 1082 rc = vas_paste(addr, offset); 1083 if (rc == 2) 1084 rc = 0; 1085 else 1086 rc = -EINVAL; 1087 1088 pr_debug("Txwin #%d: Msg count %llu\n", txwin->winid, 1089 read_hvwc_reg(txwin, VREG(LRFIFO_PUSH))); 1090 1091 return rc; 1092 } 1093 EXPORT_SYMBOL_GPL(vas_paste_crb); 1094 1095 /* 1096 * If credit checking is enabled for this window, poll for the return 1097 * of window credits (i.e for NX engines to process any outstanding CRBs). 1098 * Since NX-842 waits for the CRBs to be processed before closing the 1099 * window, we should not have to wait for too long. 1100 * 1101 * TODO: We retry in 10ms intervals now. We could/should probably peek at 1102 * the VAS_LRFIFO_PUSH_OFFSET register to get an estimate of pending 1103 * CRBs on the FIFO and compute the delay dynamically on each retry. 1104 * But that is not really needed until we support NX-GZIP access from 1105 * user space. (NX-842 driver waits for CSB and Fast thread-wakeup 1106 * doesn't use credit checking). 1107 */ 1108 static void poll_window_credits(struct vas_window *window) 1109 { 1110 u64 val; 1111 int creds, mode; 1112 1113 val = read_hvwc_reg(window, VREG(WINCTL)); 1114 if (window->tx_win) 1115 mode = GET_FIELD(VAS_WINCTL_TX_WCRED_MODE, val); 1116 else 1117 mode = GET_FIELD(VAS_WINCTL_RX_WCRED_MODE, val); 1118 1119 if (!mode) 1120 return; 1121 retry: 1122 if (window->tx_win) { 1123 val = read_hvwc_reg(window, VREG(TX_WCRED)); 1124 creds = GET_FIELD(VAS_TX_WCRED, val); 1125 } else { 1126 val = read_hvwc_reg(window, VREG(LRX_WCRED)); 1127 creds = GET_FIELD(VAS_LRX_WCRED, val); 1128 } 1129 1130 if (creds < window->wcreds_max) { 1131 val = 0; 1132 set_current_state(TASK_UNINTERRUPTIBLE); 1133 schedule_timeout(msecs_to_jiffies(10)); 1134 goto retry; 1135 } 1136 } 1137 1138 /* 1139 * Wait for the window to go to "not-busy" state. It should only take a 1140 * short time to queue a CRB, so window should not be busy for too long. 1141 * Trying 5ms intervals. 1142 */ 1143 static void poll_window_busy_state(struct vas_window *window) 1144 { 1145 int busy; 1146 u64 val; 1147 1148 retry: 1149 val = read_hvwc_reg(window, VREG(WIN_STATUS)); 1150 busy = GET_FIELD(VAS_WIN_BUSY, val); 1151 if (busy) { 1152 val = 0; 1153 set_current_state(TASK_UNINTERRUPTIBLE); 1154 schedule_timeout(msecs_to_jiffies(5)); 1155 goto retry; 1156 } 1157 } 1158 1159 /* 1160 * Have the hardware cast a window out of cache and wait for it to 1161 * be completed. 1162 * 1163 * NOTE: It can take a relatively long time to cast the window context 1164 * out of the cache. It is not strictly necessary to cast out if: 1165 * 1166 * - we clear the "Pin Window" bit (so hardware is free to evict) 1167 * 1168 * - we re-initialize the window context when it is reassigned. 1169 * 1170 * We do the former in vas_win_close() and latter in vas_win_open(). 1171 * So, ignoring the cast-out for now. We can add it as needed. If 1172 * casting out becomes necessary we should consider offloading the 1173 * job to a worker thread, so the window close can proceed quickly. 1174 */ 1175 static void poll_window_castout(struct vas_window *window) 1176 { 1177 /* stub for now */ 1178 } 1179 1180 /* 1181 * Unpin and close a window so no new requests are accepted and the 1182 * hardware can evict this window from cache if necessary. 1183 */ 1184 static void unpin_close_window(struct vas_window *window) 1185 { 1186 u64 val; 1187 1188 val = read_hvwc_reg(window, VREG(WINCTL)); 1189 val = SET_FIELD(VAS_WINCTL_PIN, val, 0); 1190 val = SET_FIELD(VAS_WINCTL_OPEN, val, 0); 1191 write_hvwc_reg(window, VREG(WINCTL), val); 1192 } 1193 1194 /* 1195 * Close a window. 1196 * 1197 * See Section 1.12.1 of VAS workbook v1.05 for details on closing window: 1198 * - Disable new paste operations (unmap paste address) 1199 * - Poll for the "Window Busy" bit to be cleared 1200 * - Clear the Open/Enable bit for the Window. 1201 * - Poll for return of window Credits (implies FIFO empty for Rx win?) 1202 * - Unpin and cast window context out of cache 1203 * 1204 * Besides the hardware, kernel has some bookkeeping of course. 1205 */ 1206 int vas_win_close(struct vas_window *window) 1207 { 1208 if (!window) 1209 return 0; 1210 1211 if (!window->tx_win && atomic_read(&window->num_txwins) != 0) { 1212 pr_devel("Attempting to close an active Rx window!\n"); 1213 WARN_ON_ONCE(1); 1214 return -EBUSY; 1215 } 1216 1217 unmap_paste_region(window); 1218 1219 clear_vinst_win(window); 1220 1221 poll_window_busy_state(window); 1222 1223 unpin_close_window(window); 1224 1225 poll_window_credits(window); 1226 1227 poll_window_castout(window); 1228 1229 /* if send window, drop reference to matching receive window */ 1230 if (window->tx_win) 1231 put_rx_win(window->rxwin); 1232 1233 vas_window_free(window); 1234 1235 return 0; 1236 } 1237 EXPORT_SYMBOL_GPL(vas_win_close); 1238