1 /* 2 * Intel I/OAT DMA Linux driver 3 * Copyright(c) 2004 - 2015 Intel Corporation. 4 * 5 * This program is free software; you can redistribute it and/or modify it 6 * under the terms and conditions of the GNU General Public License, 7 * version 2, as published by the Free Software Foundation. 8 * 9 * This program is distributed in the hope that it will be useful, but WITHOUT 10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or 11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for 12 * more details. 13 * 14 * The full GNU General Public License is included in this distribution in 15 * the file called "COPYING". 16 * 17 */ 18 19 #include <linux/init.h> 20 #include <linux/module.h> 21 #include <linux/slab.h> 22 #include <linux/pci.h> 23 #include <linux/interrupt.h> 24 #include <linux/dmaengine.h> 25 #include <linux/delay.h> 26 #include <linux/dma-mapping.h> 27 #include <linux/workqueue.h> 28 #include <linux/prefetch.h> 29 #include <linux/dca.h> 30 #include <linux/aer.h> 31 #include <linux/sizes.h> 32 #include "dma.h" 33 #include "registers.h" 34 #include "hw.h" 35 36 #include "../dmaengine.h" 37 38 MODULE_VERSION(IOAT_DMA_VERSION); 39 MODULE_LICENSE("Dual BSD/GPL"); 40 MODULE_AUTHOR("Intel Corporation"); 41 42 static struct pci_device_id ioat_pci_tbl[] = { 43 /* I/OAT v3 platforms */ 44 { PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_TBG0) }, 45 { PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_TBG1) }, 46 { PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_TBG2) }, 47 { PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_TBG3) }, 48 { PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_TBG4) }, 49 { PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_TBG5) }, 50 { PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_TBG6) }, 51 { PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_TBG7) }, 52 53 /* I/OAT v3.2 platforms */ 54 { PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_JSF0) }, 55 { PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_JSF1) }, 56 { PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_JSF2) }, 57 { PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_JSF3) }, 58 { PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_JSF4) }, 59 { PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_JSF5) }, 60 { PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_JSF6) }, 61 { PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_JSF7) }, 62 { PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_JSF8) }, 63 { PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_JSF9) }, 64 65 { PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_SNB0) }, 66 { PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_SNB1) }, 67 { PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_SNB2) }, 68 { PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_SNB3) }, 69 { PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_SNB4) }, 70 { PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_SNB5) }, 71 { PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_SNB6) }, 72 { PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_SNB7) }, 73 { PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_SNB8) }, 74 { PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_SNB9) }, 75 76 { PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_IVB0) }, 77 { PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_IVB1) }, 78 { PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_IVB2) }, 79 { PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_IVB3) }, 80 { PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_IVB4) }, 81 { PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_IVB5) }, 82 { PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_IVB6) }, 83 { PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_IVB7) }, 84 { PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_IVB8) }, 85 { PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_IVB9) }, 86 87 { PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_HSW0) }, 88 { PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_HSW1) }, 89 { PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_HSW2) }, 90 { PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_HSW3) }, 91 { PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_HSW4) }, 92 { PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_HSW5) }, 93 { PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_HSW6) }, 94 { PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_HSW7) }, 95 { PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_HSW8) }, 96 { PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_HSW9) }, 97 98 { PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_BDX0) }, 99 { PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_BDX1) }, 100 { PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_BDX2) }, 101 { PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_BDX3) }, 102 { PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_BDX4) }, 103 { PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_BDX5) }, 104 { PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_BDX6) }, 105 { PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_BDX7) }, 106 { PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_BDX8) }, 107 { PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_BDX9) }, 108 109 /* I/OAT v3.3 platforms */ 110 { PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_BWD0) }, 111 { PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_BWD1) }, 112 { PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_BWD2) }, 113 { PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_BWD3) }, 114 115 { PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_BDXDE0) }, 116 { PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_BDXDE1) }, 117 { PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_BDXDE2) }, 118 { PCI_VDEVICE(INTEL, PCI_DEVICE_ID_INTEL_IOAT_BDXDE3) }, 119 120 { 0, } 121 }; 122 MODULE_DEVICE_TABLE(pci, ioat_pci_tbl); 123 124 static int ioat_pci_probe(struct pci_dev *pdev, const struct pci_device_id *id); 125 static void ioat_remove(struct pci_dev *pdev); 126 static void 127 ioat_init_channel(struct ioatdma_device *ioat_dma, 128 struct ioatdma_chan *ioat_chan, int idx); 129 static void ioat_intr_quirk(struct ioatdma_device *ioat_dma); 130 static int ioat_enumerate_channels(struct ioatdma_device *ioat_dma); 131 static int ioat3_dma_self_test(struct ioatdma_device *ioat_dma); 132 133 static int ioat_dca_enabled = 1; 134 module_param(ioat_dca_enabled, int, 0644); 135 MODULE_PARM_DESC(ioat_dca_enabled, "control support of dca service (default: 1)"); 136 int ioat_pending_level = 4; 137 module_param(ioat_pending_level, int, 0644); 138 MODULE_PARM_DESC(ioat_pending_level, 139 "high-water mark for pushing ioat descriptors (default: 4)"); 140 static char ioat_interrupt_style[32] = "msix"; 141 module_param_string(ioat_interrupt_style, ioat_interrupt_style, 142 sizeof(ioat_interrupt_style), 0644); 143 MODULE_PARM_DESC(ioat_interrupt_style, 144 "set ioat interrupt style: msix (default), msi, intx"); 145 146 struct kmem_cache *ioat_cache; 147 struct kmem_cache *ioat_sed_cache; 148 149 static bool is_jf_ioat(struct pci_dev *pdev) 150 { 151 switch (pdev->device) { 152 case PCI_DEVICE_ID_INTEL_IOAT_JSF0: 153 case PCI_DEVICE_ID_INTEL_IOAT_JSF1: 154 case PCI_DEVICE_ID_INTEL_IOAT_JSF2: 155 case PCI_DEVICE_ID_INTEL_IOAT_JSF3: 156 case PCI_DEVICE_ID_INTEL_IOAT_JSF4: 157 case PCI_DEVICE_ID_INTEL_IOAT_JSF5: 158 case PCI_DEVICE_ID_INTEL_IOAT_JSF6: 159 case PCI_DEVICE_ID_INTEL_IOAT_JSF7: 160 case PCI_DEVICE_ID_INTEL_IOAT_JSF8: 161 case PCI_DEVICE_ID_INTEL_IOAT_JSF9: 162 return true; 163 default: 164 return false; 165 } 166 } 167 168 static bool is_snb_ioat(struct pci_dev *pdev) 169 { 170 switch (pdev->device) { 171 case PCI_DEVICE_ID_INTEL_IOAT_SNB0: 172 case PCI_DEVICE_ID_INTEL_IOAT_SNB1: 173 case PCI_DEVICE_ID_INTEL_IOAT_SNB2: 174 case PCI_DEVICE_ID_INTEL_IOAT_SNB3: 175 case PCI_DEVICE_ID_INTEL_IOAT_SNB4: 176 case PCI_DEVICE_ID_INTEL_IOAT_SNB5: 177 case PCI_DEVICE_ID_INTEL_IOAT_SNB6: 178 case PCI_DEVICE_ID_INTEL_IOAT_SNB7: 179 case PCI_DEVICE_ID_INTEL_IOAT_SNB8: 180 case PCI_DEVICE_ID_INTEL_IOAT_SNB9: 181 return true; 182 default: 183 return false; 184 } 185 } 186 187 static bool is_ivb_ioat(struct pci_dev *pdev) 188 { 189 switch (pdev->device) { 190 case PCI_DEVICE_ID_INTEL_IOAT_IVB0: 191 case PCI_DEVICE_ID_INTEL_IOAT_IVB1: 192 case PCI_DEVICE_ID_INTEL_IOAT_IVB2: 193 case PCI_DEVICE_ID_INTEL_IOAT_IVB3: 194 case PCI_DEVICE_ID_INTEL_IOAT_IVB4: 195 case PCI_DEVICE_ID_INTEL_IOAT_IVB5: 196 case PCI_DEVICE_ID_INTEL_IOAT_IVB6: 197 case PCI_DEVICE_ID_INTEL_IOAT_IVB7: 198 case PCI_DEVICE_ID_INTEL_IOAT_IVB8: 199 case PCI_DEVICE_ID_INTEL_IOAT_IVB9: 200 return true; 201 default: 202 return false; 203 } 204 205 } 206 207 static bool is_hsw_ioat(struct pci_dev *pdev) 208 { 209 switch (pdev->device) { 210 case PCI_DEVICE_ID_INTEL_IOAT_HSW0: 211 case PCI_DEVICE_ID_INTEL_IOAT_HSW1: 212 case PCI_DEVICE_ID_INTEL_IOAT_HSW2: 213 case PCI_DEVICE_ID_INTEL_IOAT_HSW3: 214 case PCI_DEVICE_ID_INTEL_IOAT_HSW4: 215 case PCI_DEVICE_ID_INTEL_IOAT_HSW5: 216 case PCI_DEVICE_ID_INTEL_IOAT_HSW6: 217 case PCI_DEVICE_ID_INTEL_IOAT_HSW7: 218 case PCI_DEVICE_ID_INTEL_IOAT_HSW8: 219 case PCI_DEVICE_ID_INTEL_IOAT_HSW9: 220 return true; 221 default: 222 return false; 223 } 224 225 } 226 227 static bool is_bdx_ioat(struct pci_dev *pdev) 228 { 229 switch (pdev->device) { 230 case PCI_DEVICE_ID_INTEL_IOAT_BDX0: 231 case PCI_DEVICE_ID_INTEL_IOAT_BDX1: 232 case PCI_DEVICE_ID_INTEL_IOAT_BDX2: 233 case PCI_DEVICE_ID_INTEL_IOAT_BDX3: 234 case PCI_DEVICE_ID_INTEL_IOAT_BDX4: 235 case PCI_DEVICE_ID_INTEL_IOAT_BDX5: 236 case PCI_DEVICE_ID_INTEL_IOAT_BDX6: 237 case PCI_DEVICE_ID_INTEL_IOAT_BDX7: 238 case PCI_DEVICE_ID_INTEL_IOAT_BDX8: 239 case PCI_DEVICE_ID_INTEL_IOAT_BDX9: 240 return true; 241 default: 242 return false; 243 } 244 } 245 246 static bool is_xeon_cb32(struct pci_dev *pdev) 247 { 248 return is_jf_ioat(pdev) || is_snb_ioat(pdev) || is_ivb_ioat(pdev) || 249 is_hsw_ioat(pdev) || is_bdx_ioat(pdev); 250 } 251 252 bool is_bwd_ioat(struct pci_dev *pdev) 253 { 254 switch (pdev->device) { 255 case PCI_DEVICE_ID_INTEL_IOAT_BWD0: 256 case PCI_DEVICE_ID_INTEL_IOAT_BWD1: 257 case PCI_DEVICE_ID_INTEL_IOAT_BWD2: 258 case PCI_DEVICE_ID_INTEL_IOAT_BWD3: 259 /* even though not Atom, BDX-DE has same DMA silicon */ 260 case PCI_DEVICE_ID_INTEL_IOAT_BDXDE0: 261 case PCI_DEVICE_ID_INTEL_IOAT_BDXDE1: 262 case PCI_DEVICE_ID_INTEL_IOAT_BDXDE2: 263 case PCI_DEVICE_ID_INTEL_IOAT_BDXDE3: 264 return true; 265 default: 266 return false; 267 } 268 } 269 270 static bool is_bwd_noraid(struct pci_dev *pdev) 271 { 272 switch (pdev->device) { 273 case PCI_DEVICE_ID_INTEL_IOAT_BWD2: 274 case PCI_DEVICE_ID_INTEL_IOAT_BWD3: 275 case PCI_DEVICE_ID_INTEL_IOAT_BDXDE0: 276 case PCI_DEVICE_ID_INTEL_IOAT_BDXDE1: 277 case PCI_DEVICE_ID_INTEL_IOAT_BDXDE2: 278 case PCI_DEVICE_ID_INTEL_IOAT_BDXDE3: 279 return true; 280 default: 281 return false; 282 } 283 284 } 285 286 /* 287 * Perform a IOAT transaction to verify the HW works. 288 */ 289 #define IOAT_TEST_SIZE 2000 290 291 static void ioat_dma_test_callback(void *dma_async_param) 292 { 293 struct completion *cmp = dma_async_param; 294 295 complete(cmp); 296 } 297 298 /** 299 * ioat_dma_self_test - Perform a IOAT transaction to verify the HW works. 300 * @ioat_dma: dma device to be tested 301 */ 302 static int ioat_dma_self_test(struct ioatdma_device *ioat_dma) 303 { 304 int i; 305 u8 *src; 306 u8 *dest; 307 struct dma_device *dma = &ioat_dma->dma_dev; 308 struct device *dev = &ioat_dma->pdev->dev; 309 struct dma_chan *dma_chan; 310 struct dma_async_tx_descriptor *tx; 311 dma_addr_t dma_dest, dma_src; 312 dma_cookie_t cookie; 313 int err = 0; 314 struct completion cmp; 315 unsigned long tmo; 316 unsigned long flags; 317 318 src = kzalloc(sizeof(u8) * IOAT_TEST_SIZE, GFP_KERNEL); 319 if (!src) 320 return -ENOMEM; 321 dest = kzalloc(sizeof(u8) * IOAT_TEST_SIZE, GFP_KERNEL); 322 if (!dest) { 323 kfree(src); 324 return -ENOMEM; 325 } 326 327 /* Fill in src buffer */ 328 for (i = 0; i < IOAT_TEST_SIZE; i++) 329 src[i] = (u8)i; 330 331 /* Start copy, using first DMA channel */ 332 dma_chan = container_of(dma->channels.next, struct dma_chan, 333 device_node); 334 if (dma->device_alloc_chan_resources(dma_chan) < 1) { 335 dev_err(dev, "selftest cannot allocate chan resource\n"); 336 err = -ENODEV; 337 goto out; 338 } 339 340 dma_src = dma_map_single(dev, src, IOAT_TEST_SIZE, DMA_TO_DEVICE); 341 if (dma_mapping_error(dev, dma_src)) { 342 dev_err(dev, "mapping src buffer failed\n"); 343 goto free_resources; 344 } 345 dma_dest = dma_map_single(dev, dest, IOAT_TEST_SIZE, DMA_FROM_DEVICE); 346 if (dma_mapping_error(dev, dma_dest)) { 347 dev_err(dev, "mapping dest buffer failed\n"); 348 goto unmap_src; 349 } 350 flags = DMA_PREP_INTERRUPT; 351 tx = ioat_dma->dma_dev.device_prep_dma_memcpy(dma_chan, dma_dest, 352 dma_src, IOAT_TEST_SIZE, 353 flags); 354 if (!tx) { 355 dev_err(dev, "Self-test prep failed, disabling\n"); 356 err = -ENODEV; 357 goto unmap_dma; 358 } 359 360 async_tx_ack(tx); 361 init_completion(&cmp); 362 tx->callback = ioat_dma_test_callback; 363 tx->callback_param = &cmp; 364 cookie = tx->tx_submit(tx); 365 if (cookie < 0) { 366 dev_err(dev, "Self-test setup failed, disabling\n"); 367 err = -ENODEV; 368 goto unmap_dma; 369 } 370 dma->device_issue_pending(dma_chan); 371 372 tmo = wait_for_completion_timeout(&cmp, msecs_to_jiffies(3000)); 373 374 if (tmo == 0 || 375 dma->device_tx_status(dma_chan, cookie, NULL) 376 != DMA_COMPLETE) { 377 dev_err(dev, "Self-test copy timed out, disabling\n"); 378 err = -ENODEV; 379 goto unmap_dma; 380 } 381 if (memcmp(src, dest, IOAT_TEST_SIZE)) { 382 dev_err(dev, "Self-test copy failed compare, disabling\n"); 383 err = -ENODEV; 384 goto free_resources; 385 } 386 387 unmap_dma: 388 dma_unmap_single(dev, dma_dest, IOAT_TEST_SIZE, DMA_FROM_DEVICE); 389 unmap_src: 390 dma_unmap_single(dev, dma_src, IOAT_TEST_SIZE, DMA_TO_DEVICE); 391 free_resources: 392 dma->device_free_chan_resources(dma_chan); 393 out: 394 kfree(src); 395 kfree(dest); 396 return err; 397 } 398 399 /** 400 * ioat_dma_setup_interrupts - setup interrupt handler 401 * @ioat_dma: ioat dma device 402 */ 403 int ioat_dma_setup_interrupts(struct ioatdma_device *ioat_dma) 404 { 405 struct ioatdma_chan *ioat_chan; 406 struct pci_dev *pdev = ioat_dma->pdev; 407 struct device *dev = &pdev->dev; 408 struct msix_entry *msix; 409 int i, j, msixcnt; 410 int err = -EINVAL; 411 u8 intrctrl = 0; 412 413 if (!strcmp(ioat_interrupt_style, "msix")) 414 goto msix; 415 if (!strcmp(ioat_interrupt_style, "msi")) 416 goto msi; 417 if (!strcmp(ioat_interrupt_style, "intx")) 418 goto intx; 419 dev_err(dev, "invalid ioat_interrupt_style %s\n", ioat_interrupt_style); 420 goto err_no_irq; 421 422 msix: 423 /* The number of MSI-X vectors should equal the number of channels */ 424 msixcnt = ioat_dma->dma_dev.chancnt; 425 for (i = 0; i < msixcnt; i++) 426 ioat_dma->msix_entries[i].entry = i; 427 428 err = pci_enable_msix_exact(pdev, ioat_dma->msix_entries, msixcnt); 429 if (err) 430 goto msi; 431 432 for (i = 0; i < msixcnt; i++) { 433 msix = &ioat_dma->msix_entries[i]; 434 ioat_chan = ioat_chan_by_index(ioat_dma, i); 435 err = devm_request_irq(dev, msix->vector, 436 ioat_dma_do_interrupt_msix, 0, 437 "ioat-msix", ioat_chan); 438 if (err) { 439 for (j = 0; j < i; j++) { 440 msix = &ioat_dma->msix_entries[j]; 441 ioat_chan = ioat_chan_by_index(ioat_dma, j); 442 devm_free_irq(dev, msix->vector, ioat_chan); 443 } 444 goto msi; 445 } 446 } 447 intrctrl |= IOAT_INTRCTRL_MSIX_VECTOR_CONTROL; 448 ioat_dma->irq_mode = IOAT_MSIX; 449 goto done; 450 451 msi: 452 err = pci_enable_msi(pdev); 453 if (err) 454 goto intx; 455 456 err = devm_request_irq(dev, pdev->irq, ioat_dma_do_interrupt, 0, 457 "ioat-msi", ioat_dma); 458 if (err) { 459 pci_disable_msi(pdev); 460 goto intx; 461 } 462 ioat_dma->irq_mode = IOAT_MSI; 463 goto done; 464 465 intx: 466 err = devm_request_irq(dev, pdev->irq, ioat_dma_do_interrupt, 467 IRQF_SHARED, "ioat-intx", ioat_dma); 468 if (err) 469 goto err_no_irq; 470 471 ioat_dma->irq_mode = IOAT_INTX; 472 done: 473 if (is_bwd_ioat(pdev)) 474 ioat_intr_quirk(ioat_dma); 475 intrctrl |= IOAT_INTRCTRL_MASTER_INT_EN; 476 writeb(intrctrl, ioat_dma->reg_base + IOAT_INTRCTRL_OFFSET); 477 return 0; 478 479 err_no_irq: 480 /* Disable all interrupt generation */ 481 writeb(0, ioat_dma->reg_base + IOAT_INTRCTRL_OFFSET); 482 ioat_dma->irq_mode = IOAT_NOIRQ; 483 dev_err(dev, "no usable interrupts\n"); 484 return err; 485 } 486 487 static void ioat_disable_interrupts(struct ioatdma_device *ioat_dma) 488 { 489 /* Disable all interrupt generation */ 490 writeb(0, ioat_dma->reg_base + IOAT_INTRCTRL_OFFSET); 491 } 492 493 static int ioat_probe(struct ioatdma_device *ioat_dma) 494 { 495 int err = -ENODEV; 496 struct dma_device *dma = &ioat_dma->dma_dev; 497 struct pci_dev *pdev = ioat_dma->pdev; 498 struct device *dev = &pdev->dev; 499 500 ioat_dma->completion_pool = dma_pool_create("completion_pool", dev, 501 sizeof(u64), 502 SMP_CACHE_BYTES, 503 SMP_CACHE_BYTES); 504 505 if (!ioat_dma->completion_pool) { 506 err = -ENOMEM; 507 goto err_out; 508 } 509 510 ioat_enumerate_channels(ioat_dma); 511 512 dma_cap_set(DMA_MEMCPY, dma->cap_mask); 513 dma->dev = &pdev->dev; 514 515 if (!dma->chancnt) { 516 dev_err(dev, "channel enumeration error\n"); 517 goto err_setup_interrupts; 518 } 519 520 err = ioat_dma_setup_interrupts(ioat_dma); 521 if (err) 522 goto err_setup_interrupts; 523 524 err = ioat3_dma_self_test(ioat_dma); 525 if (err) 526 goto err_self_test; 527 528 return 0; 529 530 err_self_test: 531 ioat_disable_interrupts(ioat_dma); 532 err_setup_interrupts: 533 dma_pool_destroy(ioat_dma->completion_pool); 534 err_out: 535 return err; 536 } 537 538 static int ioat_register(struct ioatdma_device *ioat_dma) 539 { 540 int err = dma_async_device_register(&ioat_dma->dma_dev); 541 542 if (err) { 543 ioat_disable_interrupts(ioat_dma); 544 dma_pool_destroy(ioat_dma->completion_pool); 545 } 546 547 return err; 548 } 549 550 static void ioat_dma_remove(struct ioatdma_device *ioat_dma) 551 { 552 struct dma_device *dma = &ioat_dma->dma_dev; 553 554 ioat_disable_interrupts(ioat_dma); 555 556 ioat_kobject_del(ioat_dma); 557 558 dma_async_device_unregister(dma); 559 560 dma_pool_destroy(ioat_dma->completion_pool); 561 562 INIT_LIST_HEAD(&dma->channels); 563 } 564 565 /** 566 * ioat_enumerate_channels - find and initialize the device's channels 567 * @ioat_dma: the ioat dma device to be enumerated 568 */ 569 static int ioat_enumerate_channels(struct ioatdma_device *ioat_dma) 570 { 571 struct ioatdma_chan *ioat_chan; 572 struct device *dev = &ioat_dma->pdev->dev; 573 struct dma_device *dma = &ioat_dma->dma_dev; 574 u8 xfercap_log; 575 int i; 576 577 INIT_LIST_HEAD(&dma->channels); 578 dma->chancnt = readb(ioat_dma->reg_base + IOAT_CHANCNT_OFFSET); 579 dma->chancnt &= 0x1f; /* bits [4:0] valid */ 580 if (dma->chancnt > ARRAY_SIZE(ioat_dma->idx)) { 581 dev_warn(dev, "(%d) exceeds max supported channels (%zu)\n", 582 dma->chancnt, ARRAY_SIZE(ioat_dma->idx)); 583 dma->chancnt = ARRAY_SIZE(ioat_dma->idx); 584 } 585 xfercap_log = readb(ioat_dma->reg_base + IOAT_XFERCAP_OFFSET); 586 xfercap_log &= 0x1f; /* bits [4:0] valid */ 587 if (xfercap_log == 0) 588 return 0; 589 dev_dbg(dev, "%s: xfercap = %d\n", __func__, 1 << xfercap_log); 590 591 for (i = 0; i < dma->chancnt; i++) { 592 ioat_chan = devm_kzalloc(dev, sizeof(*ioat_chan), GFP_KERNEL); 593 if (!ioat_chan) 594 break; 595 596 ioat_init_channel(ioat_dma, ioat_chan, i); 597 ioat_chan->xfercap_log = xfercap_log; 598 spin_lock_init(&ioat_chan->prep_lock); 599 if (ioat_reset_hw(ioat_chan)) { 600 i = 0; 601 break; 602 } 603 } 604 dma->chancnt = i; 605 return i; 606 } 607 608 /** 609 * ioat_free_chan_resources - release all the descriptors 610 * @chan: the channel to be cleaned 611 */ 612 static void ioat_free_chan_resources(struct dma_chan *c) 613 { 614 struct ioatdma_chan *ioat_chan = to_ioat_chan(c); 615 struct ioatdma_device *ioat_dma = ioat_chan->ioat_dma; 616 struct ioat_ring_ent *desc; 617 const int total_descs = 1 << ioat_chan->alloc_order; 618 int descs; 619 int i; 620 621 /* Before freeing channel resources first check 622 * if they have been previously allocated for this channel. 623 */ 624 if (!ioat_chan->ring) 625 return; 626 627 ioat_stop(ioat_chan); 628 ioat_reset_hw(ioat_chan); 629 630 spin_lock_bh(&ioat_chan->cleanup_lock); 631 spin_lock_bh(&ioat_chan->prep_lock); 632 descs = ioat_ring_space(ioat_chan); 633 dev_dbg(to_dev(ioat_chan), "freeing %d idle descriptors\n", descs); 634 for (i = 0; i < descs; i++) { 635 desc = ioat_get_ring_ent(ioat_chan, ioat_chan->head + i); 636 ioat_free_ring_ent(desc, c); 637 } 638 639 if (descs < total_descs) 640 dev_err(to_dev(ioat_chan), "Freeing %d in use descriptors!\n", 641 total_descs - descs); 642 643 for (i = 0; i < total_descs - descs; i++) { 644 desc = ioat_get_ring_ent(ioat_chan, ioat_chan->tail + i); 645 dump_desc_dbg(ioat_chan, desc); 646 ioat_free_ring_ent(desc, c); 647 } 648 649 for (i = 0; i < ioat_chan->desc_chunks; i++) { 650 dma_free_coherent(to_dev(ioat_chan), SZ_2M, 651 ioat_chan->descs[i].virt, 652 ioat_chan->descs[i].hw); 653 ioat_chan->descs[i].virt = NULL; 654 ioat_chan->descs[i].hw = 0; 655 } 656 ioat_chan->desc_chunks = 0; 657 658 kfree(ioat_chan->ring); 659 ioat_chan->ring = NULL; 660 ioat_chan->alloc_order = 0; 661 dma_pool_free(ioat_dma->completion_pool, ioat_chan->completion, 662 ioat_chan->completion_dma); 663 spin_unlock_bh(&ioat_chan->prep_lock); 664 spin_unlock_bh(&ioat_chan->cleanup_lock); 665 666 ioat_chan->last_completion = 0; 667 ioat_chan->completion_dma = 0; 668 ioat_chan->dmacount = 0; 669 } 670 671 /* ioat_alloc_chan_resources - allocate/initialize ioat descriptor ring 672 * @chan: channel to be initialized 673 */ 674 static int ioat_alloc_chan_resources(struct dma_chan *c) 675 { 676 struct ioatdma_chan *ioat_chan = to_ioat_chan(c); 677 struct ioat_ring_ent **ring; 678 u64 status; 679 int order; 680 int i = 0; 681 u32 chanerr; 682 683 /* have we already been set up? */ 684 if (ioat_chan->ring) 685 return 1 << ioat_chan->alloc_order; 686 687 /* Setup register to interrupt and write completion status on error */ 688 writew(IOAT_CHANCTRL_RUN, ioat_chan->reg_base + IOAT_CHANCTRL_OFFSET); 689 690 /* allocate a completion writeback area */ 691 /* doing 2 32bit writes to mmio since 1 64b write doesn't work */ 692 ioat_chan->completion = 693 dma_pool_zalloc(ioat_chan->ioat_dma->completion_pool, 694 GFP_KERNEL, &ioat_chan->completion_dma); 695 if (!ioat_chan->completion) 696 return -ENOMEM; 697 698 writel(((u64)ioat_chan->completion_dma) & 0x00000000FFFFFFFF, 699 ioat_chan->reg_base + IOAT_CHANCMP_OFFSET_LOW); 700 writel(((u64)ioat_chan->completion_dma) >> 32, 701 ioat_chan->reg_base + IOAT_CHANCMP_OFFSET_HIGH); 702 703 order = IOAT_MAX_ORDER; 704 ring = ioat_alloc_ring(c, order, GFP_KERNEL); 705 if (!ring) 706 return -ENOMEM; 707 708 spin_lock_bh(&ioat_chan->cleanup_lock); 709 spin_lock_bh(&ioat_chan->prep_lock); 710 ioat_chan->ring = ring; 711 ioat_chan->head = 0; 712 ioat_chan->issued = 0; 713 ioat_chan->tail = 0; 714 ioat_chan->alloc_order = order; 715 set_bit(IOAT_RUN, &ioat_chan->state); 716 spin_unlock_bh(&ioat_chan->prep_lock); 717 spin_unlock_bh(&ioat_chan->cleanup_lock); 718 719 ioat_start_null_desc(ioat_chan); 720 721 /* check that we got off the ground */ 722 do { 723 udelay(1); 724 status = ioat_chansts(ioat_chan); 725 } while (i++ < 20 && !is_ioat_active(status) && !is_ioat_idle(status)); 726 727 if (is_ioat_active(status) || is_ioat_idle(status)) 728 return 1 << ioat_chan->alloc_order; 729 730 chanerr = readl(ioat_chan->reg_base + IOAT_CHANERR_OFFSET); 731 732 dev_WARN(to_dev(ioat_chan), 733 "failed to start channel chanerr: %#x\n", chanerr); 734 ioat_free_chan_resources(c); 735 return -EFAULT; 736 } 737 738 /* common channel initialization */ 739 static void 740 ioat_init_channel(struct ioatdma_device *ioat_dma, 741 struct ioatdma_chan *ioat_chan, int idx) 742 { 743 struct dma_device *dma = &ioat_dma->dma_dev; 744 struct dma_chan *c = &ioat_chan->dma_chan; 745 unsigned long data = (unsigned long) c; 746 747 ioat_chan->ioat_dma = ioat_dma; 748 ioat_chan->reg_base = ioat_dma->reg_base + (0x80 * (idx + 1)); 749 spin_lock_init(&ioat_chan->cleanup_lock); 750 ioat_chan->dma_chan.device = dma; 751 dma_cookie_init(&ioat_chan->dma_chan); 752 list_add_tail(&ioat_chan->dma_chan.device_node, &dma->channels); 753 ioat_dma->idx[idx] = ioat_chan; 754 init_timer(&ioat_chan->timer); 755 ioat_chan->timer.function = ioat_timer_event; 756 ioat_chan->timer.data = data; 757 tasklet_init(&ioat_chan->cleanup_task, ioat_cleanup_event, data); 758 } 759 760 #define IOAT_NUM_SRC_TEST 6 /* must be <= 8 */ 761 static int ioat_xor_val_self_test(struct ioatdma_device *ioat_dma) 762 { 763 int i, src_idx; 764 struct page *dest; 765 struct page *xor_srcs[IOAT_NUM_SRC_TEST]; 766 struct page *xor_val_srcs[IOAT_NUM_SRC_TEST + 1]; 767 dma_addr_t dma_srcs[IOAT_NUM_SRC_TEST + 1]; 768 dma_addr_t dest_dma; 769 struct dma_async_tx_descriptor *tx; 770 struct dma_chan *dma_chan; 771 dma_cookie_t cookie; 772 u8 cmp_byte = 0; 773 u32 cmp_word; 774 u32 xor_val_result; 775 int err = 0; 776 struct completion cmp; 777 unsigned long tmo; 778 struct device *dev = &ioat_dma->pdev->dev; 779 struct dma_device *dma = &ioat_dma->dma_dev; 780 u8 op = 0; 781 782 dev_dbg(dev, "%s\n", __func__); 783 784 if (!dma_has_cap(DMA_XOR, dma->cap_mask)) 785 return 0; 786 787 for (src_idx = 0; src_idx < IOAT_NUM_SRC_TEST; src_idx++) { 788 xor_srcs[src_idx] = alloc_page(GFP_KERNEL); 789 if (!xor_srcs[src_idx]) { 790 while (src_idx--) 791 __free_page(xor_srcs[src_idx]); 792 return -ENOMEM; 793 } 794 } 795 796 dest = alloc_page(GFP_KERNEL); 797 if (!dest) { 798 while (src_idx--) 799 __free_page(xor_srcs[src_idx]); 800 return -ENOMEM; 801 } 802 803 /* Fill in src buffers */ 804 for (src_idx = 0; src_idx < IOAT_NUM_SRC_TEST; src_idx++) { 805 u8 *ptr = page_address(xor_srcs[src_idx]); 806 807 for (i = 0; i < PAGE_SIZE; i++) 808 ptr[i] = (1 << src_idx); 809 } 810 811 for (src_idx = 0; src_idx < IOAT_NUM_SRC_TEST; src_idx++) 812 cmp_byte ^= (u8) (1 << src_idx); 813 814 cmp_word = (cmp_byte << 24) | (cmp_byte << 16) | 815 (cmp_byte << 8) | cmp_byte; 816 817 memset(page_address(dest), 0, PAGE_SIZE); 818 819 dma_chan = container_of(dma->channels.next, struct dma_chan, 820 device_node); 821 if (dma->device_alloc_chan_resources(dma_chan) < 1) { 822 err = -ENODEV; 823 goto out; 824 } 825 826 /* test xor */ 827 op = IOAT_OP_XOR; 828 829 dest_dma = dma_map_page(dev, dest, 0, PAGE_SIZE, DMA_FROM_DEVICE); 830 if (dma_mapping_error(dev, dest_dma)) 831 goto dma_unmap; 832 833 for (i = 0; i < IOAT_NUM_SRC_TEST; i++) 834 dma_srcs[i] = DMA_ERROR_CODE; 835 for (i = 0; i < IOAT_NUM_SRC_TEST; i++) { 836 dma_srcs[i] = dma_map_page(dev, xor_srcs[i], 0, PAGE_SIZE, 837 DMA_TO_DEVICE); 838 if (dma_mapping_error(dev, dma_srcs[i])) 839 goto dma_unmap; 840 } 841 tx = dma->device_prep_dma_xor(dma_chan, dest_dma, dma_srcs, 842 IOAT_NUM_SRC_TEST, PAGE_SIZE, 843 DMA_PREP_INTERRUPT); 844 845 if (!tx) { 846 dev_err(dev, "Self-test xor prep failed\n"); 847 err = -ENODEV; 848 goto dma_unmap; 849 } 850 851 async_tx_ack(tx); 852 init_completion(&cmp); 853 tx->callback = ioat_dma_test_callback; 854 tx->callback_param = &cmp; 855 cookie = tx->tx_submit(tx); 856 if (cookie < 0) { 857 dev_err(dev, "Self-test xor setup failed\n"); 858 err = -ENODEV; 859 goto dma_unmap; 860 } 861 dma->device_issue_pending(dma_chan); 862 863 tmo = wait_for_completion_timeout(&cmp, msecs_to_jiffies(3000)); 864 865 if (tmo == 0 || 866 dma->device_tx_status(dma_chan, cookie, NULL) != DMA_COMPLETE) { 867 dev_err(dev, "Self-test xor timed out\n"); 868 err = -ENODEV; 869 goto dma_unmap; 870 } 871 872 for (i = 0; i < IOAT_NUM_SRC_TEST; i++) 873 dma_unmap_page(dev, dma_srcs[i], PAGE_SIZE, DMA_TO_DEVICE); 874 875 dma_sync_single_for_cpu(dev, dest_dma, PAGE_SIZE, DMA_FROM_DEVICE); 876 for (i = 0; i < (PAGE_SIZE / sizeof(u32)); i++) { 877 u32 *ptr = page_address(dest); 878 879 if (ptr[i] != cmp_word) { 880 dev_err(dev, "Self-test xor failed compare\n"); 881 err = -ENODEV; 882 goto free_resources; 883 } 884 } 885 dma_sync_single_for_device(dev, dest_dma, PAGE_SIZE, DMA_FROM_DEVICE); 886 887 dma_unmap_page(dev, dest_dma, PAGE_SIZE, DMA_FROM_DEVICE); 888 889 /* skip validate if the capability is not present */ 890 if (!dma_has_cap(DMA_XOR_VAL, dma_chan->device->cap_mask)) 891 goto free_resources; 892 893 op = IOAT_OP_XOR_VAL; 894 895 /* validate the sources with the destintation page */ 896 for (i = 0; i < IOAT_NUM_SRC_TEST; i++) 897 xor_val_srcs[i] = xor_srcs[i]; 898 xor_val_srcs[i] = dest; 899 900 xor_val_result = 1; 901 902 for (i = 0; i < IOAT_NUM_SRC_TEST + 1; i++) 903 dma_srcs[i] = DMA_ERROR_CODE; 904 for (i = 0; i < IOAT_NUM_SRC_TEST + 1; i++) { 905 dma_srcs[i] = dma_map_page(dev, xor_val_srcs[i], 0, PAGE_SIZE, 906 DMA_TO_DEVICE); 907 if (dma_mapping_error(dev, dma_srcs[i])) 908 goto dma_unmap; 909 } 910 tx = dma->device_prep_dma_xor_val(dma_chan, dma_srcs, 911 IOAT_NUM_SRC_TEST + 1, PAGE_SIZE, 912 &xor_val_result, DMA_PREP_INTERRUPT); 913 if (!tx) { 914 dev_err(dev, "Self-test zero prep failed\n"); 915 err = -ENODEV; 916 goto dma_unmap; 917 } 918 919 async_tx_ack(tx); 920 init_completion(&cmp); 921 tx->callback = ioat_dma_test_callback; 922 tx->callback_param = &cmp; 923 cookie = tx->tx_submit(tx); 924 if (cookie < 0) { 925 dev_err(dev, "Self-test zero setup failed\n"); 926 err = -ENODEV; 927 goto dma_unmap; 928 } 929 dma->device_issue_pending(dma_chan); 930 931 tmo = wait_for_completion_timeout(&cmp, msecs_to_jiffies(3000)); 932 933 if (tmo == 0 || 934 dma->device_tx_status(dma_chan, cookie, NULL) != DMA_COMPLETE) { 935 dev_err(dev, "Self-test validate timed out\n"); 936 err = -ENODEV; 937 goto dma_unmap; 938 } 939 940 for (i = 0; i < IOAT_NUM_SRC_TEST + 1; i++) 941 dma_unmap_page(dev, dma_srcs[i], PAGE_SIZE, DMA_TO_DEVICE); 942 943 if (xor_val_result != 0) { 944 dev_err(dev, "Self-test validate failed compare\n"); 945 err = -ENODEV; 946 goto free_resources; 947 } 948 949 memset(page_address(dest), 0, PAGE_SIZE); 950 951 /* test for non-zero parity sum */ 952 op = IOAT_OP_XOR_VAL; 953 954 xor_val_result = 0; 955 for (i = 0; i < IOAT_NUM_SRC_TEST + 1; i++) 956 dma_srcs[i] = DMA_ERROR_CODE; 957 for (i = 0; i < IOAT_NUM_SRC_TEST + 1; i++) { 958 dma_srcs[i] = dma_map_page(dev, xor_val_srcs[i], 0, PAGE_SIZE, 959 DMA_TO_DEVICE); 960 if (dma_mapping_error(dev, dma_srcs[i])) 961 goto dma_unmap; 962 } 963 tx = dma->device_prep_dma_xor_val(dma_chan, dma_srcs, 964 IOAT_NUM_SRC_TEST + 1, PAGE_SIZE, 965 &xor_val_result, DMA_PREP_INTERRUPT); 966 if (!tx) { 967 dev_err(dev, "Self-test 2nd zero prep failed\n"); 968 err = -ENODEV; 969 goto dma_unmap; 970 } 971 972 async_tx_ack(tx); 973 init_completion(&cmp); 974 tx->callback = ioat_dma_test_callback; 975 tx->callback_param = &cmp; 976 cookie = tx->tx_submit(tx); 977 if (cookie < 0) { 978 dev_err(dev, "Self-test 2nd zero setup failed\n"); 979 err = -ENODEV; 980 goto dma_unmap; 981 } 982 dma->device_issue_pending(dma_chan); 983 984 tmo = wait_for_completion_timeout(&cmp, msecs_to_jiffies(3000)); 985 986 if (tmo == 0 || 987 dma->device_tx_status(dma_chan, cookie, NULL) != DMA_COMPLETE) { 988 dev_err(dev, "Self-test 2nd validate timed out\n"); 989 err = -ENODEV; 990 goto dma_unmap; 991 } 992 993 if (xor_val_result != SUM_CHECK_P_RESULT) { 994 dev_err(dev, "Self-test validate failed compare\n"); 995 err = -ENODEV; 996 goto dma_unmap; 997 } 998 999 for (i = 0; i < IOAT_NUM_SRC_TEST + 1; i++) 1000 dma_unmap_page(dev, dma_srcs[i], PAGE_SIZE, DMA_TO_DEVICE); 1001 1002 goto free_resources; 1003 dma_unmap: 1004 if (op == IOAT_OP_XOR) { 1005 if (dest_dma != DMA_ERROR_CODE) 1006 dma_unmap_page(dev, dest_dma, PAGE_SIZE, 1007 DMA_FROM_DEVICE); 1008 for (i = 0; i < IOAT_NUM_SRC_TEST; i++) 1009 if (dma_srcs[i] != DMA_ERROR_CODE) 1010 dma_unmap_page(dev, dma_srcs[i], PAGE_SIZE, 1011 DMA_TO_DEVICE); 1012 } else if (op == IOAT_OP_XOR_VAL) { 1013 for (i = 0; i < IOAT_NUM_SRC_TEST + 1; i++) 1014 if (dma_srcs[i] != DMA_ERROR_CODE) 1015 dma_unmap_page(dev, dma_srcs[i], PAGE_SIZE, 1016 DMA_TO_DEVICE); 1017 } 1018 free_resources: 1019 dma->device_free_chan_resources(dma_chan); 1020 out: 1021 src_idx = IOAT_NUM_SRC_TEST; 1022 while (src_idx--) 1023 __free_page(xor_srcs[src_idx]); 1024 __free_page(dest); 1025 return err; 1026 } 1027 1028 static int ioat3_dma_self_test(struct ioatdma_device *ioat_dma) 1029 { 1030 int rc; 1031 1032 rc = ioat_dma_self_test(ioat_dma); 1033 if (rc) 1034 return rc; 1035 1036 rc = ioat_xor_val_self_test(ioat_dma); 1037 1038 return rc; 1039 } 1040 1041 static void ioat_intr_quirk(struct ioatdma_device *ioat_dma) 1042 { 1043 struct dma_device *dma; 1044 struct dma_chan *c; 1045 struct ioatdma_chan *ioat_chan; 1046 u32 errmask; 1047 1048 dma = &ioat_dma->dma_dev; 1049 1050 /* 1051 * if we have descriptor write back error status, we mask the 1052 * error interrupts 1053 */ 1054 if (ioat_dma->cap & IOAT_CAP_DWBES) { 1055 list_for_each_entry(c, &dma->channels, device_node) { 1056 ioat_chan = to_ioat_chan(c); 1057 errmask = readl(ioat_chan->reg_base + 1058 IOAT_CHANERR_MASK_OFFSET); 1059 errmask |= IOAT_CHANERR_XOR_P_OR_CRC_ERR | 1060 IOAT_CHANERR_XOR_Q_ERR; 1061 writel(errmask, ioat_chan->reg_base + 1062 IOAT_CHANERR_MASK_OFFSET); 1063 } 1064 } 1065 } 1066 1067 static int ioat3_dma_probe(struct ioatdma_device *ioat_dma, int dca) 1068 { 1069 struct pci_dev *pdev = ioat_dma->pdev; 1070 int dca_en = system_has_dca_enabled(pdev); 1071 struct dma_device *dma; 1072 struct dma_chan *c; 1073 struct ioatdma_chan *ioat_chan; 1074 bool is_raid_device = false; 1075 int err; 1076 u16 val16; 1077 1078 dma = &ioat_dma->dma_dev; 1079 dma->device_prep_dma_memcpy = ioat_dma_prep_memcpy_lock; 1080 dma->device_issue_pending = ioat_issue_pending; 1081 dma->device_alloc_chan_resources = ioat_alloc_chan_resources; 1082 dma->device_free_chan_resources = ioat_free_chan_resources; 1083 1084 dma_cap_set(DMA_INTERRUPT, dma->cap_mask); 1085 dma->device_prep_dma_interrupt = ioat_prep_interrupt_lock; 1086 1087 ioat_dma->cap = readl(ioat_dma->reg_base + IOAT_DMA_CAP_OFFSET); 1088 1089 if (is_xeon_cb32(pdev) || is_bwd_noraid(pdev)) 1090 ioat_dma->cap &= 1091 ~(IOAT_CAP_XOR | IOAT_CAP_PQ | IOAT_CAP_RAID16SS); 1092 1093 /* dca is incompatible with raid operations */ 1094 if (dca_en && (ioat_dma->cap & (IOAT_CAP_XOR|IOAT_CAP_PQ))) 1095 ioat_dma->cap &= ~(IOAT_CAP_XOR|IOAT_CAP_PQ); 1096 1097 if (ioat_dma->cap & IOAT_CAP_XOR) { 1098 is_raid_device = true; 1099 dma->max_xor = 8; 1100 1101 dma_cap_set(DMA_XOR, dma->cap_mask); 1102 dma->device_prep_dma_xor = ioat_prep_xor; 1103 1104 dma_cap_set(DMA_XOR_VAL, dma->cap_mask); 1105 dma->device_prep_dma_xor_val = ioat_prep_xor_val; 1106 } 1107 1108 if (ioat_dma->cap & IOAT_CAP_PQ) { 1109 is_raid_device = true; 1110 1111 dma->device_prep_dma_pq = ioat_prep_pq; 1112 dma->device_prep_dma_pq_val = ioat_prep_pq_val; 1113 dma_cap_set(DMA_PQ, dma->cap_mask); 1114 dma_cap_set(DMA_PQ_VAL, dma->cap_mask); 1115 1116 if (ioat_dma->cap & IOAT_CAP_RAID16SS) 1117 dma_set_maxpq(dma, 16, 0); 1118 else 1119 dma_set_maxpq(dma, 8, 0); 1120 1121 if (!(ioat_dma->cap & IOAT_CAP_XOR)) { 1122 dma->device_prep_dma_xor = ioat_prep_pqxor; 1123 dma->device_prep_dma_xor_val = ioat_prep_pqxor_val; 1124 dma_cap_set(DMA_XOR, dma->cap_mask); 1125 dma_cap_set(DMA_XOR_VAL, dma->cap_mask); 1126 1127 if (ioat_dma->cap & IOAT_CAP_RAID16SS) 1128 dma->max_xor = 16; 1129 else 1130 dma->max_xor = 8; 1131 } 1132 } 1133 1134 dma->device_tx_status = ioat_tx_status; 1135 1136 /* starting with CB3.3 super extended descriptors are supported */ 1137 if (ioat_dma->cap & IOAT_CAP_RAID16SS) { 1138 char pool_name[14]; 1139 int i; 1140 1141 for (i = 0; i < MAX_SED_POOLS; i++) { 1142 snprintf(pool_name, 14, "ioat_hw%d_sed", i); 1143 1144 /* allocate SED DMA pool */ 1145 ioat_dma->sed_hw_pool[i] = dmam_pool_create(pool_name, 1146 &pdev->dev, 1147 SED_SIZE * (i + 1), 64, 0); 1148 if (!ioat_dma->sed_hw_pool[i]) 1149 return -ENOMEM; 1150 1151 } 1152 } 1153 1154 if (!(ioat_dma->cap & (IOAT_CAP_XOR | IOAT_CAP_PQ))) 1155 dma_cap_set(DMA_PRIVATE, dma->cap_mask); 1156 1157 err = ioat_probe(ioat_dma); 1158 if (err) 1159 return err; 1160 1161 list_for_each_entry(c, &dma->channels, device_node) { 1162 ioat_chan = to_ioat_chan(c); 1163 writel(IOAT_DMA_DCA_ANY_CPU, 1164 ioat_chan->reg_base + IOAT_DCACTRL_OFFSET); 1165 } 1166 1167 err = ioat_register(ioat_dma); 1168 if (err) 1169 return err; 1170 1171 ioat_kobject_add(ioat_dma, &ioat_ktype); 1172 1173 if (dca) 1174 ioat_dma->dca = ioat_dca_init(pdev, ioat_dma->reg_base); 1175 1176 /* disable relaxed ordering */ 1177 err = pcie_capability_read_word(pdev, IOAT_DEVCTRL_OFFSET, &val16); 1178 if (err) 1179 return err; 1180 1181 /* clear relaxed ordering enable */ 1182 val16 &= ~IOAT_DEVCTRL_ROE; 1183 err = pcie_capability_write_word(pdev, IOAT_DEVCTRL_OFFSET, val16); 1184 if (err) 1185 return err; 1186 1187 return 0; 1188 } 1189 1190 static void ioat_shutdown(struct pci_dev *pdev) 1191 { 1192 struct ioatdma_device *ioat_dma = pci_get_drvdata(pdev); 1193 struct ioatdma_chan *ioat_chan; 1194 int i; 1195 1196 if (!ioat_dma) 1197 return; 1198 1199 for (i = 0; i < IOAT_MAX_CHANS; i++) { 1200 ioat_chan = ioat_dma->idx[i]; 1201 if (!ioat_chan) 1202 continue; 1203 1204 spin_lock_bh(&ioat_chan->prep_lock); 1205 set_bit(IOAT_CHAN_DOWN, &ioat_chan->state); 1206 del_timer_sync(&ioat_chan->timer); 1207 spin_unlock_bh(&ioat_chan->prep_lock); 1208 /* this should quiesce then reset */ 1209 ioat_reset_hw(ioat_chan); 1210 } 1211 1212 ioat_disable_interrupts(ioat_dma); 1213 } 1214 1215 void ioat_resume(struct ioatdma_device *ioat_dma) 1216 { 1217 struct ioatdma_chan *ioat_chan; 1218 u32 chanerr; 1219 int i; 1220 1221 for (i = 0; i < IOAT_MAX_CHANS; i++) { 1222 ioat_chan = ioat_dma->idx[i]; 1223 if (!ioat_chan) 1224 continue; 1225 1226 spin_lock_bh(&ioat_chan->prep_lock); 1227 clear_bit(IOAT_CHAN_DOWN, &ioat_chan->state); 1228 spin_unlock_bh(&ioat_chan->prep_lock); 1229 1230 chanerr = readl(ioat_chan->reg_base + IOAT_CHANERR_OFFSET); 1231 writel(chanerr, ioat_chan->reg_base + IOAT_CHANERR_OFFSET); 1232 1233 /* no need to reset as shutdown already did that */ 1234 } 1235 } 1236 1237 #define DRV_NAME "ioatdma" 1238 1239 static pci_ers_result_t ioat_pcie_error_detected(struct pci_dev *pdev, 1240 enum pci_channel_state error) 1241 { 1242 dev_dbg(&pdev->dev, "%s: PCIe AER error %d\n", DRV_NAME, error); 1243 1244 /* quiesce and block I/O */ 1245 ioat_shutdown(pdev); 1246 1247 return PCI_ERS_RESULT_NEED_RESET; 1248 } 1249 1250 static pci_ers_result_t ioat_pcie_error_slot_reset(struct pci_dev *pdev) 1251 { 1252 pci_ers_result_t result = PCI_ERS_RESULT_RECOVERED; 1253 int err; 1254 1255 dev_dbg(&pdev->dev, "%s post reset handling\n", DRV_NAME); 1256 1257 if (pci_enable_device_mem(pdev) < 0) { 1258 dev_err(&pdev->dev, 1259 "Failed to enable PCIe device after reset.\n"); 1260 result = PCI_ERS_RESULT_DISCONNECT; 1261 } else { 1262 pci_set_master(pdev); 1263 pci_restore_state(pdev); 1264 pci_save_state(pdev); 1265 pci_wake_from_d3(pdev, false); 1266 } 1267 1268 err = pci_cleanup_aer_uncorrect_error_status(pdev); 1269 if (err) { 1270 dev_err(&pdev->dev, 1271 "AER uncorrect error status clear failed: %#x\n", err); 1272 } 1273 1274 return result; 1275 } 1276 1277 static void ioat_pcie_error_resume(struct pci_dev *pdev) 1278 { 1279 struct ioatdma_device *ioat_dma = pci_get_drvdata(pdev); 1280 1281 dev_dbg(&pdev->dev, "%s: AER handling resuming\n", DRV_NAME); 1282 1283 /* initialize and bring everything back */ 1284 ioat_resume(ioat_dma); 1285 } 1286 1287 static const struct pci_error_handlers ioat_err_handler = { 1288 .error_detected = ioat_pcie_error_detected, 1289 .slot_reset = ioat_pcie_error_slot_reset, 1290 .resume = ioat_pcie_error_resume, 1291 }; 1292 1293 static struct pci_driver ioat_pci_driver = { 1294 .name = DRV_NAME, 1295 .id_table = ioat_pci_tbl, 1296 .probe = ioat_pci_probe, 1297 .remove = ioat_remove, 1298 .shutdown = ioat_shutdown, 1299 .err_handler = &ioat_err_handler, 1300 }; 1301 1302 static struct ioatdma_device * 1303 alloc_ioatdma(struct pci_dev *pdev, void __iomem *iobase) 1304 { 1305 struct device *dev = &pdev->dev; 1306 struct ioatdma_device *d = devm_kzalloc(dev, sizeof(*d), GFP_KERNEL); 1307 1308 if (!d) 1309 return NULL; 1310 d->pdev = pdev; 1311 d->reg_base = iobase; 1312 return d; 1313 } 1314 1315 static int ioat_pci_probe(struct pci_dev *pdev, const struct pci_device_id *id) 1316 { 1317 void __iomem * const *iomap; 1318 struct device *dev = &pdev->dev; 1319 struct ioatdma_device *device; 1320 int err; 1321 1322 err = pcim_enable_device(pdev); 1323 if (err) 1324 return err; 1325 1326 err = pcim_iomap_regions(pdev, 1 << IOAT_MMIO_BAR, DRV_NAME); 1327 if (err) 1328 return err; 1329 iomap = pcim_iomap_table(pdev); 1330 if (!iomap) 1331 return -ENOMEM; 1332 1333 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(64)); 1334 if (err) 1335 err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32)); 1336 if (err) 1337 return err; 1338 1339 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64)); 1340 if (err) 1341 err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32)); 1342 if (err) 1343 return err; 1344 1345 device = alloc_ioatdma(pdev, iomap[IOAT_MMIO_BAR]); 1346 if (!device) 1347 return -ENOMEM; 1348 pci_set_master(pdev); 1349 pci_set_drvdata(pdev, device); 1350 1351 device->version = readb(device->reg_base + IOAT_VER_OFFSET); 1352 if (device->version >= IOAT_VER_3_0) { 1353 err = ioat3_dma_probe(device, ioat_dca_enabled); 1354 1355 if (device->version >= IOAT_VER_3_3) 1356 pci_enable_pcie_error_reporting(pdev); 1357 } else 1358 return -ENODEV; 1359 1360 if (err) { 1361 dev_err(dev, "Intel(R) I/OAT DMA Engine init failed\n"); 1362 pci_disable_pcie_error_reporting(pdev); 1363 return -ENODEV; 1364 } 1365 1366 return 0; 1367 } 1368 1369 static void ioat_remove(struct pci_dev *pdev) 1370 { 1371 struct ioatdma_device *device = pci_get_drvdata(pdev); 1372 1373 if (!device) 1374 return; 1375 1376 dev_err(&pdev->dev, "Removing dma and dca services\n"); 1377 if (device->dca) { 1378 unregister_dca_provider(device->dca, &pdev->dev); 1379 free_dca_provider(device->dca); 1380 device->dca = NULL; 1381 } 1382 1383 pci_disable_pcie_error_reporting(pdev); 1384 ioat_dma_remove(device); 1385 } 1386 1387 static int __init ioat_init_module(void) 1388 { 1389 int err = -ENOMEM; 1390 1391 pr_info("%s: Intel(R) QuickData Technology Driver %s\n", 1392 DRV_NAME, IOAT_DMA_VERSION); 1393 1394 ioat_cache = kmem_cache_create("ioat", sizeof(struct ioat_ring_ent), 1395 0, SLAB_HWCACHE_ALIGN, NULL); 1396 if (!ioat_cache) 1397 return -ENOMEM; 1398 1399 ioat_sed_cache = KMEM_CACHE(ioat_sed_ent, 0); 1400 if (!ioat_sed_cache) 1401 goto err_ioat_cache; 1402 1403 err = pci_register_driver(&ioat_pci_driver); 1404 if (err) 1405 goto err_ioat3_cache; 1406 1407 return 0; 1408 1409 err_ioat3_cache: 1410 kmem_cache_destroy(ioat_sed_cache); 1411 1412 err_ioat_cache: 1413 kmem_cache_destroy(ioat_cache); 1414 1415 return err; 1416 } 1417 module_init(ioat_init_module); 1418 1419 static void __exit ioat_exit_module(void) 1420 { 1421 pci_unregister_driver(&ioat_pci_driver); 1422 kmem_cache_destroy(ioat_cache); 1423 } 1424 module_exit(ioat_exit_module); 1425