George Zhang | 1f16643 | 2013-01-08 15:55:32 -0800 | [diff] [blame] | 1 | /* |
| 2 | * VMware VMCI Driver |
| 3 | * |
| 4 | * Copyright (C) 2012 VMware, Inc. All rights reserved. |
| 5 | * |
| 6 | * This program is free software; you can redistribute it and/or modify it |
| 7 | * under the terms of the GNU General Public License as published by the |
| 8 | * Free Software Foundation version 2 and no later version. |
| 9 | * |
| 10 | * This program is distributed in the hope that it will be useful, but |
| 11 | * WITHOUT ANY WARRANTY; without even the implied warranty of MERCHANTABILITY |
| 12 | * or FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License |
| 13 | * for more details. |
| 14 | */ |
| 15 | |
| 16 | #include <linux/vmw_vmci_defs.h> |
| 17 | #include <linux/vmw_vmci_api.h> |
| 18 | #include <linux/moduleparam.h> |
| 19 | #include <linux/interrupt.h> |
| 20 | #include <linux/highmem.h> |
| 21 | #include <linux/kernel.h> |
Dmitry Torokhov | ea8a83a | 2013-01-10 15:41:43 -0800 | [diff] [blame] | 22 | #include <linux/mm.h> |
George Zhang | 1f16643 | 2013-01-08 15:55:32 -0800 | [diff] [blame] | 23 | #include <linux/module.h> |
| 24 | #include <linux/sched.h> |
Dmitry Torokhov | ea8a83a | 2013-01-10 15:41:43 -0800 | [diff] [blame] | 25 | #include <linux/slab.h> |
George Zhang | 1f16643 | 2013-01-08 15:55:32 -0800 | [diff] [blame] | 26 | #include <linux/init.h> |
| 27 | #include <linux/pci.h> |
| 28 | #include <linux/smp.h> |
| 29 | #include <linux/io.h> |
Dmitry Torokhov | ea8a83a | 2013-01-10 15:41:43 -0800 | [diff] [blame] | 30 | #include <linux/vmalloc.h> |
George Zhang | 1f16643 | 2013-01-08 15:55:32 -0800 | [diff] [blame] | 31 | |
| 32 | #include "vmci_datagram.h" |
| 33 | #include "vmci_doorbell.h" |
| 34 | #include "vmci_context.h" |
| 35 | #include "vmci_driver.h" |
| 36 | #include "vmci_event.h" |
| 37 | |
| 38 | #define PCI_VENDOR_ID_VMWARE 0x15AD |
| 39 | #define PCI_DEVICE_ID_VMWARE_VMCI 0x0740 |
| 40 | |
| 41 | #define VMCI_UTIL_NUM_RESOURCES 1 |
| 42 | |
| 43 | static bool vmci_disable_msi; |
| 44 | module_param_named(disable_msi, vmci_disable_msi, bool, 0); |
| 45 | MODULE_PARM_DESC(disable_msi, "Disable MSI use in driver - (default=0)"); |
| 46 | |
| 47 | static bool vmci_disable_msix; |
| 48 | module_param_named(disable_msix, vmci_disable_msix, bool, 0); |
| 49 | MODULE_PARM_DESC(disable_msix, "Disable MSI-X use in driver - (default=0)"); |
| 50 | |
| 51 | static u32 ctx_update_sub_id = VMCI_INVALID_ID; |
| 52 | static u32 vm_context_id = VMCI_INVALID_ID; |
| 53 | |
| 54 | struct vmci_guest_device { |
| 55 | struct device *dev; /* PCI device we are attached to */ |
| 56 | void __iomem *iobase; |
| 57 | |
| 58 | unsigned int irq; |
| 59 | unsigned int intr_type; |
| 60 | bool exclusive_vectors; |
| 61 | struct msix_entry msix_entries[VMCI_MAX_INTRS]; |
| 62 | |
| 63 | struct tasklet_struct datagram_tasklet; |
| 64 | struct tasklet_struct bm_tasklet; |
| 65 | |
| 66 | void *data_buffer; |
| 67 | void *notification_bitmap; |
Andy King | 6d6dfb4 | 2013-08-23 09:22:14 -0700 | [diff] [blame] | 68 | dma_addr_t notification_base; |
George Zhang | 1f16643 | 2013-01-08 15:55:32 -0800 | [diff] [blame] | 69 | }; |
| 70 | |
| 71 | /* vmci_dev singleton device and supporting data*/ |
Andy King | 6d6dfb4 | 2013-08-23 09:22:14 -0700 | [diff] [blame] | 72 | struct pci_dev *vmci_pdev; |
George Zhang | 1f16643 | 2013-01-08 15:55:32 -0800 | [diff] [blame] | 73 | static struct vmci_guest_device *vmci_dev_g; |
| 74 | static DEFINE_SPINLOCK(vmci_dev_spinlock); |
| 75 | |
| 76 | static atomic_t vmci_num_guest_devices = ATOMIC_INIT(0); |
| 77 | |
| 78 | bool vmci_guest_code_active(void) |
| 79 | { |
| 80 | return atomic_read(&vmci_num_guest_devices) != 0; |
| 81 | } |
| 82 | |
| 83 | u32 vmci_get_vm_context_id(void) |
| 84 | { |
| 85 | if (vm_context_id == VMCI_INVALID_ID) { |
George Zhang | 1f16643 | 2013-01-08 15:55:32 -0800 | [diff] [blame] | 86 | struct vmci_datagram get_cid_msg; |
| 87 | get_cid_msg.dst = |
| 88 | vmci_make_handle(VMCI_HYPERVISOR_CONTEXT_ID, |
| 89 | VMCI_GET_CONTEXT_ID); |
| 90 | get_cid_msg.src = VMCI_ANON_SRC_HANDLE; |
| 91 | get_cid_msg.payload_size = 0; |
Andy King | 5a19b78 | 2013-01-22 09:15:04 -0800 | [diff] [blame] | 92 | vm_context_id = vmci_send_datagram(&get_cid_msg); |
George Zhang | 1f16643 | 2013-01-08 15:55:32 -0800 | [diff] [blame] | 93 | } |
| 94 | return vm_context_id; |
| 95 | } |
| 96 | |
| 97 | /* |
| 98 | * VM to hypervisor call mechanism. We use the standard VMware naming |
| 99 | * convention since shared code is calling this function as well. |
| 100 | */ |
| 101 | int vmci_send_datagram(struct vmci_datagram *dg) |
| 102 | { |
| 103 | unsigned long flags; |
| 104 | int result; |
| 105 | |
| 106 | /* Check args. */ |
| 107 | if (dg == NULL) |
| 108 | return VMCI_ERROR_INVALID_ARGS; |
| 109 | |
| 110 | /* |
| 111 | * Need to acquire spinlock on the device because the datagram |
| 112 | * data may be spread over multiple pages and the monitor may |
| 113 | * interleave device user rpc calls from multiple |
| 114 | * VCPUs. Acquiring the spinlock precludes that |
| 115 | * possibility. Disabling interrupts to avoid incoming |
| 116 | * datagrams during a "rep out" and possibly landing up in |
| 117 | * this function. |
| 118 | */ |
| 119 | spin_lock_irqsave(&vmci_dev_spinlock, flags); |
| 120 | |
| 121 | if (vmci_dev_g) { |
| 122 | iowrite8_rep(vmci_dev_g->iobase + VMCI_DATA_OUT_ADDR, |
| 123 | dg, VMCI_DG_SIZE(dg)); |
| 124 | result = ioread32(vmci_dev_g->iobase + VMCI_RESULT_LOW_ADDR); |
| 125 | } else { |
| 126 | result = VMCI_ERROR_UNAVAILABLE; |
| 127 | } |
| 128 | |
| 129 | spin_unlock_irqrestore(&vmci_dev_spinlock, flags); |
| 130 | |
| 131 | return result; |
| 132 | } |
| 133 | EXPORT_SYMBOL_GPL(vmci_send_datagram); |
| 134 | |
| 135 | /* |
| 136 | * Gets called with the new context id if updated or resumed. |
| 137 | * Context id. |
| 138 | */ |
| 139 | static void vmci_guest_cid_update(u32 sub_id, |
| 140 | const struct vmci_event_data *event_data, |
| 141 | void *client_data) |
| 142 | { |
| 143 | const struct vmci_event_payld_ctx *ev_payload = |
| 144 | vmci_event_data_const_payload(event_data); |
| 145 | |
| 146 | if (sub_id != ctx_update_sub_id) { |
| 147 | pr_devel("Invalid subscriber (ID=0x%x)\n", sub_id); |
| 148 | return; |
| 149 | } |
| 150 | |
| 151 | if (!event_data || ev_payload->context_id == VMCI_INVALID_ID) { |
| 152 | pr_devel("Invalid event data\n"); |
| 153 | return; |
| 154 | } |
| 155 | |
| 156 | pr_devel("Updating context from (ID=0x%x) to (ID=0x%x) on event (type=%d)\n", |
| 157 | vm_context_id, ev_payload->context_id, event_data->event); |
| 158 | |
| 159 | vm_context_id = ev_payload->context_id; |
| 160 | } |
| 161 | |
| 162 | /* |
| 163 | * Verify that the host supports the hypercalls we need. If it does not, |
| 164 | * try to find fallback hypercalls and use those instead. Returns |
| 165 | * true if required hypercalls (or fallback hypercalls) are |
| 166 | * supported by the host, false otherwise. |
| 167 | */ |
| 168 | static bool vmci_check_host_caps(struct pci_dev *pdev) |
| 169 | { |
| 170 | bool result; |
| 171 | struct vmci_resource_query_msg *msg; |
| 172 | u32 msg_size = sizeof(struct vmci_resource_query_hdr) + |
| 173 | VMCI_UTIL_NUM_RESOURCES * sizeof(u32); |
| 174 | struct vmci_datagram *check_msg; |
| 175 | |
| 176 | check_msg = kmalloc(msg_size, GFP_KERNEL); |
| 177 | if (!check_msg) { |
| 178 | dev_err(&pdev->dev, "%s: Insufficient memory\n", __func__); |
| 179 | return false; |
| 180 | } |
| 181 | |
| 182 | check_msg->dst = vmci_make_handle(VMCI_HYPERVISOR_CONTEXT_ID, |
| 183 | VMCI_RESOURCES_QUERY); |
| 184 | check_msg->src = VMCI_ANON_SRC_HANDLE; |
| 185 | check_msg->payload_size = msg_size - VMCI_DG_HEADERSIZE; |
| 186 | msg = (struct vmci_resource_query_msg *)VMCI_DG_PAYLOAD(check_msg); |
| 187 | |
| 188 | msg->num_resources = VMCI_UTIL_NUM_RESOURCES; |
| 189 | msg->resources[0] = VMCI_GET_CONTEXT_ID; |
| 190 | |
| 191 | /* Checks that hyper calls are supported */ |
| 192 | result = vmci_send_datagram(check_msg) == 0x01; |
| 193 | kfree(check_msg); |
| 194 | |
| 195 | dev_dbg(&pdev->dev, "%s: Host capability check: %s\n", |
| 196 | __func__, result ? "PASSED" : "FAILED"); |
| 197 | |
| 198 | /* We need the vector. There are no fallbacks. */ |
| 199 | return result; |
| 200 | } |
| 201 | |
| 202 | /* |
| 203 | * Reads datagrams from the data in port and dispatches them. We |
| 204 | * always start reading datagrams into only the first page of the |
| 205 | * datagram buffer. If the datagrams don't fit into one page, we |
| 206 | * use the maximum datagram buffer size for the remainder of the |
| 207 | * invocation. This is a simple heuristic for not penalizing |
| 208 | * small datagrams. |
| 209 | * |
| 210 | * This function assumes that it has exclusive access to the data |
| 211 | * in port for the duration of the call. |
| 212 | */ |
| 213 | static void vmci_dispatch_dgs(unsigned long data) |
| 214 | { |
| 215 | struct vmci_guest_device *vmci_dev = (struct vmci_guest_device *)data; |
| 216 | u8 *dg_in_buffer = vmci_dev->data_buffer; |
| 217 | struct vmci_datagram *dg; |
| 218 | size_t dg_in_buffer_size = VMCI_MAX_DG_SIZE; |
| 219 | size_t current_dg_in_buffer_size = PAGE_SIZE; |
| 220 | size_t remaining_bytes; |
| 221 | |
| 222 | BUILD_BUG_ON(VMCI_MAX_DG_SIZE < PAGE_SIZE); |
| 223 | |
| 224 | ioread8_rep(vmci_dev->iobase + VMCI_DATA_IN_ADDR, |
| 225 | vmci_dev->data_buffer, current_dg_in_buffer_size); |
| 226 | dg = (struct vmci_datagram *)dg_in_buffer; |
| 227 | remaining_bytes = current_dg_in_buffer_size; |
| 228 | |
| 229 | while (dg->dst.resource != VMCI_INVALID_ID || |
| 230 | remaining_bytes > PAGE_SIZE) { |
| 231 | unsigned dg_in_size; |
| 232 | |
| 233 | /* |
| 234 | * When the input buffer spans multiple pages, a datagram can |
| 235 | * start on any page boundary in the buffer. |
| 236 | */ |
| 237 | if (dg->dst.resource == VMCI_INVALID_ID) { |
| 238 | dg = (struct vmci_datagram *)roundup( |
| 239 | (uintptr_t)dg + 1, PAGE_SIZE); |
| 240 | remaining_bytes = |
| 241 | (size_t)(dg_in_buffer + |
| 242 | current_dg_in_buffer_size - |
| 243 | (u8 *)dg); |
| 244 | continue; |
| 245 | } |
| 246 | |
| 247 | dg_in_size = VMCI_DG_SIZE_ALIGNED(dg); |
| 248 | |
| 249 | if (dg_in_size <= dg_in_buffer_size) { |
| 250 | int result; |
| 251 | |
| 252 | /* |
| 253 | * If the remaining bytes in the datagram |
| 254 | * buffer doesn't contain the complete |
| 255 | * datagram, we first make sure we have enough |
| 256 | * room for it and then we read the reminder |
| 257 | * of the datagram and possibly any following |
| 258 | * datagrams. |
| 259 | */ |
| 260 | if (dg_in_size > remaining_bytes) { |
| 261 | if (remaining_bytes != |
| 262 | current_dg_in_buffer_size) { |
| 263 | |
| 264 | /* |
| 265 | * We move the partial |
| 266 | * datagram to the front and |
| 267 | * read the reminder of the |
| 268 | * datagram and possibly |
| 269 | * following calls into the |
| 270 | * following bytes. |
| 271 | */ |
| 272 | memmove(dg_in_buffer, dg_in_buffer + |
| 273 | current_dg_in_buffer_size - |
| 274 | remaining_bytes, |
| 275 | remaining_bytes); |
| 276 | dg = (struct vmci_datagram *) |
| 277 | dg_in_buffer; |
| 278 | } |
| 279 | |
| 280 | if (current_dg_in_buffer_size != |
| 281 | dg_in_buffer_size) |
| 282 | current_dg_in_buffer_size = |
| 283 | dg_in_buffer_size; |
| 284 | |
| 285 | ioread8_rep(vmci_dev->iobase + |
| 286 | VMCI_DATA_IN_ADDR, |
| 287 | vmci_dev->data_buffer + |
| 288 | remaining_bytes, |
| 289 | current_dg_in_buffer_size - |
| 290 | remaining_bytes); |
| 291 | } |
| 292 | |
| 293 | /* |
| 294 | * We special case event datagrams from the |
| 295 | * hypervisor. |
| 296 | */ |
| 297 | if (dg->src.context == VMCI_HYPERVISOR_CONTEXT_ID && |
| 298 | dg->dst.resource == VMCI_EVENT_HANDLER) { |
| 299 | result = vmci_event_dispatch(dg); |
| 300 | } else { |
| 301 | result = vmci_datagram_invoke_guest_handler(dg); |
| 302 | } |
| 303 | if (result < VMCI_SUCCESS) |
| 304 | dev_dbg(vmci_dev->dev, |
| 305 | "Datagram with resource (ID=0x%x) failed (err=%d)\n", |
| 306 | dg->dst.resource, result); |
| 307 | |
| 308 | /* On to the next datagram. */ |
| 309 | dg = (struct vmci_datagram *)((u8 *)dg + |
| 310 | dg_in_size); |
| 311 | } else { |
| 312 | size_t bytes_to_skip; |
| 313 | |
| 314 | /* |
| 315 | * Datagram doesn't fit in datagram buffer of maximal |
| 316 | * size. We drop it. |
| 317 | */ |
| 318 | dev_dbg(vmci_dev->dev, |
| 319 | "Failed to receive datagram (size=%u bytes)\n", |
| 320 | dg_in_size); |
| 321 | |
| 322 | bytes_to_skip = dg_in_size - remaining_bytes; |
| 323 | if (current_dg_in_buffer_size != dg_in_buffer_size) |
| 324 | current_dg_in_buffer_size = dg_in_buffer_size; |
| 325 | |
| 326 | for (;;) { |
| 327 | ioread8_rep(vmci_dev->iobase + |
| 328 | VMCI_DATA_IN_ADDR, |
| 329 | vmci_dev->data_buffer, |
| 330 | current_dg_in_buffer_size); |
| 331 | if (bytes_to_skip <= current_dg_in_buffer_size) |
| 332 | break; |
| 333 | |
| 334 | bytes_to_skip -= current_dg_in_buffer_size; |
| 335 | } |
| 336 | dg = (struct vmci_datagram *)(dg_in_buffer + |
| 337 | bytes_to_skip); |
| 338 | } |
| 339 | |
| 340 | remaining_bytes = |
| 341 | (size_t) (dg_in_buffer + current_dg_in_buffer_size - |
| 342 | (u8 *)dg); |
| 343 | |
| 344 | if (remaining_bytes < VMCI_DG_HEADERSIZE) { |
| 345 | /* Get the next batch of datagrams. */ |
| 346 | |
| 347 | ioread8_rep(vmci_dev->iobase + VMCI_DATA_IN_ADDR, |
| 348 | vmci_dev->data_buffer, |
| 349 | current_dg_in_buffer_size); |
| 350 | dg = (struct vmci_datagram *)dg_in_buffer; |
| 351 | remaining_bytes = current_dg_in_buffer_size; |
| 352 | } |
| 353 | } |
| 354 | } |
| 355 | |
| 356 | /* |
| 357 | * Scans the notification bitmap for raised flags, clears them |
| 358 | * and handles the notifications. |
| 359 | */ |
| 360 | static void vmci_process_bitmap(unsigned long data) |
| 361 | { |
| 362 | struct vmci_guest_device *dev = (struct vmci_guest_device *)data; |
| 363 | |
| 364 | if (!dev->notification_bitmap) { |
| 365 | dev_dbg(dev->dev, "No bitmap present in %s\n", __func__); |
| 366 | return; |
| 367 | } |
| 368 | |
| 369 | vmci_dbell_scan_notification_entries(dev->notification_bitmap); |
| 370 | } |
| 371 | |
| 372 | /* |
| 373 | * Enable MSI-X. Try exclusive vectors first, then shared vectors. |
| 374 | */ |
| 375 | static int vmci_enable_msix(struct pci_dev *pdev, |
| 376 | struct vmci_guest_device *vmci_dev) |
| 377 | { |
| 378 | int i; |
| 379 | int result; |
| 380 | |
| 381 | for (i = 0; i < VMCI_MAX_INTRS; ++i) { |
| 382 | vmci_dev->msix_entries[i].entry = i; |
| 383 | vmci_dev->msix_entries[i].vector = i; |
| 384 | } |
| 385 | |
| 386 | result = pci_enable_msix(pdev, vmci_dev->msix_entries, VMCI_MAX_INTRS); |
| 387 | if (result == 0) |
| 388 | vmci_dev->exclusive_vectors = true; |
| 389 | else if (result > 0) |
| 390 | result = pci_enable_msix(pdev, vmci_dev->msix_entries, 1); |
| 391 | |
| 392 | return result; |
| 393 | } |
| 394 | |
| 395 | /* |
| 396 | * Interrupt handler for legacy or MSI interrupt, or for first MSI-X |
| 397 | * interrupt (vector VMCI_INTR_DATAGRAM). |
| 398 | */ |
| 399 | static irqreturn_t vmci_interrupt(int irq, void *_dev) |
| 400 | { |
| 401 | struct vmci_guest_device *dev = _dev; |
| 402 | |
| 403 | /* |
| 404 | * If we are using MSI-X with exclusive vectors then we simply schedule |
| 405 | * the datagram tasklet, since we know the interrupt was meant for us. |
| 406 | * Otherwise we must read the ICR to determine what to do. |
| 407 | */ |
| 408 | |
| 409 | if (dev->intr_type == VMCI_INTR_TYPE_MSIX && dev->exclusive_vectors) { |
| 410 | tasklet_schedule(&dev->datagram_tasklet); |
| 411 | } else { |
| 412 | unsigned int icr; |
| 413 | |
| 414 | /* Acknowledge interrupt and determine what needs doing. */ |
| 415 | icr = ioread32(dev->iobase + VMCI_ICR_ADDR); |
| 416 | if (icr == 0 || icr == ~0) |
| 417 | return IRQ_NONE; |
| 418 | |
| 419 | if (icr & VMCI_ICR_DATAGRAM) { |
| 420 | tasklet_schedule(&dev->datagram_tasklet); |
| 421 | icr &= ~VMCI_ICR_DATAGRAM; |
| 422 | } |
| 423 | |
| 424 | if (icr & VMCI_ICR_NOTIFICATION) { |
| 425 | tasklet_schedule(&dev->bm_tasklet); |
| 426 | icr &= ~VMCI_ICR_NOTIFICATION; |
| 427 | } |
| 428 | |
| 429 | if (icr != 0) |
| 430 | dev_warn(dev->dev, |
| 431 | "Ignoring unknown interrupt cause (%d)\n", |
| 432 | icr); |
| 433 | } |
| 434 | |
| 435 | return IRQ_HANDLED; |
| 436 | } |
| 437 | |
| 438 | /* |
| 439 | * Interrupt handler for MSI-X interrupt vector VMCI_INTR_NOTIFICATION, |
| 440 | * which is for the notification bitmap. Will only get called if we are |
| 441 | * using MSI-X with exclusive vectors. |
| 442 | */ |
| 443 | static irqreturn_t vmci_interrupt_bm(int irq, void *_dev) |
| 444 | { |
| 445 | struct vmci_guest_device *dev = _dev; |
| 446 | |
| 447 | /* For MSI-X we can just assume it was meant for us. */ |
| 448 | tasklet_schedule(&dev->bm_tasklet); |
| 449 | |
| 450 | return IRQ_HANDLED; |
| 451 | } |
| 452 | |
| 453 | /* |
| 454 | * Most of the initialization at module load time is done here. |
| 455 | */ |
| 456 | static int vmci_guest_probe_device(struct pci_dev *pdev, |
| 457 | const struct pci_device_id *id) |
| 458 | { |
| 459 | struct vmci_guest_device *vmci_dev; |
| 460 | void __iomem *iobase; |
| 461 | unsigned int capabilities; |
| 462 | unsigned long cmd; |
| 463 | int vmci_err; |
| 464 | int error; |
| 465 | |
| 466 | dev_dbg(&pdev->dev, "Probing for vmci/PCI guest device\n"); |
| 467 | |
| 468 | error = pcim_enable_device(pdev); |
| 469 | if (error) { |
| 470 | dev_err(&pdev->dev, |
| 471 | "Failed to enable VMCI device: %d\n", error); |
| 472 | return error; |
| 473 | } |
| 474 | |
| 475 | error = pcim_iomap_regions(pdev, 1 << 0, KBUILD_MODNAME); |
| 476 | if (error) { |
| 477 | dev_err(&pdev->dev, "Failed to reserve/map IO regions\n"); |
| 478 | return error; |
| 479 | } |
| 480 | |
| 481 | iobase = pcim_iomap_table(pdev)[0]; |
| 482 | |
| 483 | dev_info(&pdev->dev, "Found VMCI PCI device at %#lx, irq %u\n", |
| 484 | (unsigned long)iobase, pdev->irq); |
| 485 | |
| 486 | vmci_dev = devm_kzalloc(&pdev->dev, sizeof(*vmci_dev), GFP_KERNEL); |
| 487 | if (!vmci_dev) { |
| 488 | dev_err(&pdev->dev, |
| 489 | "Can't allocate memory for VMCI device\n"); |
| 490 | return -ENOMEM; |
| 491 | } |
| 492 | |
| 493 | vmci_dev->dev = &pdev->dev; |
| 494 | vmci_dev->intr_type = VMCI_INTR_TYPE_INTX; |
| 495 | vmci_dev->exclusive_vectors = false; |
| 496 | vmci_dev->iobase = iobase; |
| 497 | |
| 498 | tasklet_init(&vmci_dev->datagram_tasklet, |
| 499 | vmci_dispatch_dgs, (unsigned long)vmci_dev); |
| 500 | tasklet_init(&vmci_dev->bm_tasklet, |
| 501 | vmci_process_bitmap, (unsigned long)vmci_dev); |
| 502 | |
| 503 | vmci_dev->data_buffer = vmalloc(VMCI_MAX_DG_SIZE); |
| 504 | if (!vmci_dev->data_buffer) { |
| 505 | dev_err(&pdev->dev, |
| 506 | "Can't allocate memory for datagram buffer\n"); |
| 507 | return -ENOMEM; |
| 508 | } |
| 509 | |
| 510 | pci_set_master(pdev); /* To enable queue_pair functionality. */ |
| 511 | |
| 512 | /* |
| 513 | * Verify that the VMCI Device supports the capabilities that |
| 514 | * we need. If the device is missing capabilities that we would |
| 515 | * like to use, check for fallback capabilities and use those |
| 516 | * instead (so we can run a new VM on old hosts). Fail the load if |
| 517 | * a required capability is missing and there is no fallback. |
| 518 | * |
| 519 | * Right now, we need datagrams. There are no fallbacks. |
| 520 | */ |
| 521 | capabilities = ioread32(vmci_dev->iobase + VMCI_CAPS_ADDR); |
| 522 | if (!(capabilities & VMCI_CAPS_DATAGRAM)) { |
| 523 | dev_err(&pdev->dev, "Device does not support datagrams\n"); |
| 524 | error = -ENXIO; |
| 525 | goto err_free_data_buffer; |
| 526 | } |
| 527 | |
| 528 | /* |
| 529 | * If the hardware supports notifications, we will use that as |
| 530 | * well. |
| 531 | */ |
| 532 | if (capabilities & VMCI_CAPS_NOTIFICATIONS) { |
Andy King | 6d6dfb4 | 2013-08-23 09:22:14 -0700 | [diff] [blame] | 533 | vmci_dev->notification_bitmap = dma_alloc_coherent( |
| 534 | &pdev->dev, PAGE_SIZE, &vmci_dev->notification_base, |
| 535 | GFP_KERNEL); |
George Zhang | 1f16643 | 2013-01-08 15:55:32 -0800 | [diff] [blame] | 536 | if (!vmci_dev->notification_bitmap) { |
| 537 | dev_warn(&pdev->dev, |
| 538 | "Unable to allocate notification bitmap\n"); |
| 539 | } else { |
| 540 | memset(vmci_dev->notification_bitmap, 0, PAGE_SIZE); |
| 541 | capabilities |= VMCI_CAPS_NOTIFICATIONS; |
| 542 | } |
| 543 | } |
| 544 | |
| 545 | dev_info(&pdev->dev, "Using capabilities 0x%x\n", capabilities); |
| 546 | |
| 547 | /* Let the host know which capabilities we intend to use. */ |
| 548 | iowrite32(capabilities, vmci_dev->iobase + VMCI_CAPS_ADDR); |
| 549 | |
| 550 | /* Set up global device so that we can start sending datagrams */ |
| 551 | spin_lock_irq(&vmci_dev_spinlock); |
| 552 | vmci_dev_g = vmci_dev; |
Andy King | 6d6dfb4 | 2013-08-23 09:22:14 -0700 | [diff] [blame] | 553 | vmci_pdev = pdev; |
George Zhang | 1f16643 | 2013-01-08 15:55:32 -0800 | [diff] [blame] | 554 | spin_unlock_irq(&vmci_dev_spinlock); |
| 555 | |
| 556 | /* |
| 557 | * Register notification bitmap with device if that capability is |
| 558 | * used. |
| 559 | */ |
| 560 | if (capabilities & VMCI_CAPS_NOTIFICATIONS) { |
Andy King | 6d6dfb4 | 2013-08-23 09:22:14 -0700 | [diff] [blame] | 561 | unsigned long bitmap_ppn = |
| 562 | vmci_dev->notification_base >> PAGE_SHIFT; |
George Zhang | 1f16643 | 2013-01-08 15:55:32 -0800 | [diff] [blame] | 563 | if (!vmci_dbell_register_notification_bitmap(bitmap_ppn)) { |
| 564 | dev_warn(&pdev->dev, |
| 565 | "VMCI device unable to register notification bitmap with PPN 0x%x\n", |
| 566 | (u32) bitmap_ppn); |
| 567 | goto err_remove_vmci_dev_g; |
| 568 | } |
| 569 | } |
| 570 | |
| 571 | /* Check host capabilities. */ |
| 572 | if (!vmci_check_host_caps(pdev)) |
| 573 | goto err_remove_bitmap; |
| 574 | |
| 575 | /* Enable device. */ |
| 576 | |
| 577 | /* |
| 578 | * We subscribe to the VMCI_EVENT_CTX_ID_UPDATE here so we can |
| 579 | * update the internal context id when needed. |
| 580 | */ |
| 581 | vmci_err = vmci_event_subscribe(VMCI_EVENT_CTX_ID_UPDATE, |
| 582 | vmci_guest_cid_update, NULL, |
| 583 | &ctx_update_sub_id); |
| 584 | if (vmci_err < VMCI_SUCCESS) |
| 585 | dev_warn(&pdev->dev, |
| 586 | "Failed to subscribe to event (type=%d): %d\n", |
| 587 | VMCI_EVENT_CTX_ID_UPDATE, vmci_err); |
| 588 | |
| 589 | /* |
| 590 | * Enable interrupts. Try MSI-X first, then MSI, and then fallback on |
| 591 | * legacy interrupts. |
| 592 | */ |
| 593 | if (!vmci_disable_msix && !vmci_enable_msix(pdev, vmci_dev)) { |
| 594 | vmci_dev->intr_type = VMCI_INTR_TYPE_MSIX; |
| 595 | vmci_dev->irq = vmci_dev->msix_entries[0].vector; |
| 596 | } else if (!vmci_disable_msi && !pci_enable_msi(pdev)) { |
| 597 | vmci_dev->intr_type = VMCI_INTR_TYPE_MSI; |
| 598 | vmci_dev->irq = pdev->irq; |
| 599 | } else { |
| 600 | vmci_dev->intr_type = VMCI_INTR_TYPE_INTX; |
| 601 | vmci_dev->irq = pdev->irq; |
| 602 | } |
| 603 | |
| 604 | /* |
| 605 | * Request IRQ for legacy or MSI interrupts, or for first |
| 606 | * MSI-X vector. |
| 607 | */ |
| 608 | error = request_irq(vmci_dev->irq, vmci_interrupt, IRQF_SHARED, |
| 609 | KBUILD_MODNAME, vmci_dev); |
| 610 | if (error) { |
| 611 | dev_err(&pdev->dev, "Irq %u in use: %d\n", |
| 612 | vmci_dev->irq, error); |
| 613 | goto err_disable_msi; |
| 614 | } |
| 615 | |
| 616 | /* |
| 617 | * For MSI-X with exclusive vectors we need to request an |
| 618 | * interrupt for each vector so that we get a separate |
| 619 | * interrupt handler routine. This allows us to distinguish |
| 620 | * between the vectors. |
| 621 | */ |
| 622 | if (vmci_dev->exclusive_vectors) { |
| 623 | error = request_irq(vmci_dev->msix_entries[1].vector, |
| 624 | vmci_interrupt_bm, 0, KBUILD_MODNAME, |
| 625 | vmci_dev); |
| 626 | if (error) { |
| 627 | dev_err(&pdev->dev, |
| 628 | "Failed to allocate irq %u: %d\n", |
| 629 | vmci_dev->msix_entries[1].vector, error); |
| 630 | goto err_free_irq; |
| 631 | } |
| 632 | } |
| 633 | |
| 634 | dev_dbg(&pdev->dev, "Registered device\n"); |
| 635 | |
| 636 | atomic_inc(&vmci_num_guest_devices); |
| 637 | |
| 638 | /* Enable specific interrupt bits. */ |
| 639 | cmd = VMCI_IMR_DATAGRAM; |
| 640 | if (capabilities & VMCI_CAPS_NOTIFICATIONS) |
| 641 | cmd |= VMCI_IMR_NOTIFICATION; |
| 642 | iowrite32(cmd, vmci_dev->iobase + VMCI_IMR_ADDR); |
| 643 | |
| 644 | /* Enable interrupts. */ |
| 645 | iowrite32(VMCI_CONTROL_INT_ENABLE, |
| 646 | vmci_dev->iobase + VMCI_CONTROL_ADDR); |
| 647 | |
| 648 | pci_set_drvdata(pdev, vmci_dev); |
| 649 | return 0; |
| 650 | |
| 651 | err_free_irq: |
Wei Yongjun | 9089e3b | 2013-09-06 14:39:28 +0800 | [diff] [blame] | 652 | free_irq(vmci_dev->irq, vmci_dev); |
George Zhang | 1f16643 | 2013-01-08 15:55:32 -0800 | [diff] [blame] | 653 | tasklet_kill(&vmci_dev->datagram_tasklet); |
| 654 | tasklet_kill(&vmci_dev->bm_tasklet); |
| 655 | |
| 656 | err_disable_msi: |
| 657 | if (vmci_dev->intr_type == VMCI_INTR_TYPE_MSIX) |
| 658 | pci_disable_msix(pdev); |
| 659 | else if (vmci_dev->intr_type == VMCI_INTR_TYPE_MSI) |
| 660 | pci_disable_msi(pdev); |
| 661 | |
| 662 | vmci_err = vmci_event_unsubscribe(ctx_update_sub_id); |
| 663 | if (vmci_err < VMCI_SUCCESS) |
| 664 | dev_warn(&pdev->dev, |
| 665 | "Failed to unsubscribe from event (type=%d) with subscriber (ID=0x%x): %d\n", |
| 666 | VMCI_EVENT_CTX_ID_UPDATE, ctx_update_sub_id, vmci_err); |
| 667 | |
| 668 | err_remove_bitmap: |
| 669 | if (vmci_dev->notification_bitmap) { |
| 670 | iowrite32(VMCI_CONTROL_RESET, |
| 671 | vmci_dev->iobase + VMCI_CONTROL_ADDR); |
Andy King | 6d6dfb4 | 2013-08-23 09:22:14 -0700 | [diff] [blame] | 672 | dma_free_coherent(&pdev->dev, PAGE_SIZE, |
| 673 | vmci_dev->notification_bitmap, |
| 674 | vmci_dev->notification_base); |
George Zhang | 1f16643 | 2013-01-08 15:55:32 -0800 | [diff] [blame] | 675 | } |
| 676 | |
| 677 | err_remove_vmci_dev_g: |
| 678 | spin_lock_irq(&vmci_dev_spinlock); |
Andy King | 6d6dfb4 | 2013-08-23 09:22:14 -0700 | [diff] [blame] | 679 | vmci_pdev = NULL; |
George Zhang | 1f16643 | 2013-01-08 15:55:32 -0800 | [diff] [blame] | 680 | vmci_dev_g = NULL; |
| 681 | spin_unlock_irq(&vmci_dev_spinlock); |
| 682 | |
| 683 | err_free_data_buffer: |
| 684 | vfree(vmci_dev->data_buffer); |
| 685 | |
| 686 | /* The rest are managed resources and will be freed by PCI core */ |
| 687 | return error; |
| 688 | } |
| 689 | |
| 690 | static void vmci_guest_remove_device(struct pci_dev *pdev) |
| 691 | { |
| 692 | struct vmci_guest_device *vmci_dev = pci_get_drvdata(pdev); |
| 693 | int vmci_err; |
| 694 | |
| 695 | dev_dbg(&pdev->dev, "Removing device\n"); |
| 696 | |
| 697 | atomic_dec(&vmci_num_guest_devices); |
| 698 | |
| 699 | vmci_qp_guest_endpoints_exit(); |
| 700 | |
| 701 | vmci_err = vmci_event_unsubscribe(ctx_update_sub_id); |
| 702 | if (vmci_err < VMCI_SUCCESS) |
| 703 | dev_warn(&pdev->dev, |
| 704 | "Failed to unsubscribe from event (type=%d) with subscriber (ID=0x%x): %d\n", |
| 705 | VMCI_EVENT_CTX_ID_UPDATE, ctx_update_sub_id, vmci_err); |
| 706 | |
| 707 | spin_lock_irq(&vmci_dev_spinlock); |
| 708 | vmci_dev_g = NULL; |
Andy King | 6d6dfb4 | 2013-08-23 09:22:14 -0700 | [diff] [blame] | 709 | vmci_pdev = NULL; |
George Zhang | 1f16643 | 2013-01-08 15:55:32 -0800 | [diff] [blame] | 710 | spin_unlock_irq(&vmci_dev_spinlock); |
| 711 | |
| 712 | dev_dbg(&pdev->dev, "Resetting vmci device\n"); |
| 713 | iowrite32(VMCI_CONTROL_RESET, vmci_dev->iobase + VMCI_CONTROL_ADDR); |
| 714 | |
| 715 | /* |
| 716 | * Free IRQ and then disable MSI/MSI-X as appropriate. For |
| 717 | * MSI-X, we might have multiple vectors, each with their own |
| 718 | * IRQ, which we must free too. |
| 719 | */ |
| 720 | free_irq(vmci_dev->irq, vmci_dev); |
| 721 | if (vmci_dev->intr_type == VMCI_INTR_TYPE_MSIX) { |
| 722 | if (vmci_dev->exclusive_vectors) |
| 723 | free_irq(vmci_dev->msix_entries[1].vector, vmci_dev); |
| 724 | pci_disable_msix(pdev); |
| 725 | } else if (vmci_dev->intr_type == VMCI_INTR_TYPE_MSI) { |
| 726 | pci_disable_msi(pdev); |
| 727 | } |
| 728 | |
| 729 | tasklet_kill(&vmci_dev->datagram_tasklet); |
| 730 | tasklet_kill(&vmci_dev->bm_tasklet); |
| 731 | |
| 732 | if (vmci_dev->notification_bitmap) { |
| 733 | /* |
| 734 | * The device reset above cleared the bitmap state of the |
| 735 | * device, so we can safely free it here. |
| 736 | */ |
| 737 | |
Andy King | 6d6dfb4 | 2013-08-23 09:22:14 -0700 | [diff] [blame] | 738 | dma_free_coherent(&pdev->dev, PAGE_SIZE, |
| 739 | vmci_dev->notification_bitmap, |
| 740 | vmci_dev->notification_base); |
George Zhang | 1f16643 | 2013-01-08 15:55:32 -0800 | [diff] [blame] | 741 | } |
| 742 | |
| 743 | vfree(vmci_dev->data_buffer); |
| 744 | |
| 745 | /* The rest are managed resources and will be freed by PCI core */ |
| 746 | } |
| 747 | |
| 748 | static DEFINE_PCI_DEVICE_TABLE(vmci_ids) = { |
| 749 | { PCI_DEVICE(PCI_VENDOR_ID_VMWARE, PCI_DEVICE_ID_VMWARE_VMCI), }, |
| 750 | { 0 }, |
| 751 | }; |
| 752 | MODULE_DEVICE_TABLE(pci, vmci_ids); |
| 753 | |
| 754 | static struct pci_driver vmci_guest_driver = { |
| 755 | .name = KBUILD_MODNAME, |
| 756 | .id_table = vmci_ids, |
| 757 | .probe = vmci_guest_probe_device, |
| 758 | .remove = vmci_guest_remove_device, |
| 759 | }; |
| 760 | |
| 761 | int __init vmci_guest_init(void) |
| 762 | { |
| 763 | return pci_register_driver(&vmci_guest_driver); |
| 764 | } |
| 765 | |
| 766 | void __exit vmci_guest_exit(void) |
| 767 | { |
| 768 | pci_unregister_driver(&vmci_guest_driver); |
| 769 | } |