Commit | Line | Data |
---|---|---|
685a6bf8 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
1f166439 GZ |
2 | /* |
3 | * VMware VMCI Driver | |
4 | * | |
5 | * Copyright (C) 2012 VMware, Inc. All rights reserved. | |
1f166439 GZ |
6 | */ |
7 | ||
8 | #include <linux/vmw_vmci_defs.h> | |
9 | #include <linux/vmw_vmci_api.h> | |
10 | #include <linux/moduleparam.h> | |
11 | #include <linux/interrupt.h> | |
12 | #include <linux/highmem.h> | |
13 | #include <linux/kernel.h> | |
ea8a83a4 | 14 | #include <linux/mm.h> |
1f166439 GZ |
15 | #include <linux/module.h> |
16 | #include <linux/sched.h> | |
ea8a83a4 | 17 | #include <linux/slab.h> |
1f166439 GZ |
18 | #include <linux/init.h> |
19 | #include <linux/pci.h> | |
20 | #include <linux/smp.h> | |
21 | #include <linux/io.h> | |
ea8a83a4 | 22 | #include <linux/vmalloc.h> |
1f166439 GZ |
23 | |
24 | #include "vmci_datagram.h" | |
25 | #include "vmci_doorbell.h" | |
26 | #include "vmci_context.h" | |
27 | #include "vmci_driver.h" | |
28 | #include "vmci_event.h" | |
29 | ||
1f166439 GZ |
30 | #define PCI_DEVICE_ID_VMWARE_VMCI 0x0740 |
31 | ||
32 | #define VMCI_UTIL_NUM_RESOURCES 1 | |
33 | ||
34 | static bool vmci_disable_msi; | |
35 | module_param_named(disable_msi, vmci_disable_msi, bool, 0); | |
36 | MODULE_PARM_DESC(disable_msi, "Disable MSI use in driver - (default=0)"); | |
37 | ||
38 | static bool vmci_disable_msix; | |
39 | module_param_named(disable_msix, vmci_disable_msix, bool, 0); | |
40 | MODULE_PARM_DESC(disable_msix, "Disable MSI-X use in driver - (default=0)"); | |
41 | ||
42 | static u32 ctx_update_sub_id = VMCI_INVALID_ID; | |
43 | static u32 vm_context_id = VMCI_INVALID_ID; | |
44 | ||
45 | struct vmci_guest_device { | |
46 | struct device *dev; /* PCI device we are attached to */ | |
47 | void __iomem *iobase; | |
48 | ||
1f166439 | 49 | bool exclusive_vectors; |
1f166439 GZ |
50 | |
51 | struct tasklet_struct datagram_tasklet; | |
52 | struct tasklet_struct bm_tasklet; | |
53 | ||
54 | void *data_buffer; | |
55 | void *notification_bitmap; | |
6d6dfb4f | 56 | dma_addr_t notification_base; |
1f166439 GZ |
57 | }; |
58 | ||
f2db7361 VD |
59 | static bool use_ppn64; |
60 | ||
61 | bool vmci_use_ppn64(void) | |
62 | { | |
63 | return use_ppn64; | |
64 | } | |
65 | ||
1f166439 | 66 | /* vmci_dev singleton device and supporting data*/ |
6d6dfb4f | 67 | struct pci_dev *vmci_pdev; |
1f166439 GZ |
68 | static struct vmci_guest_device *vmci_dev_g; |
69 | static DEFINE_SPINLOCK(vmci_dev_spinlock); | |
70 | ||
71 | static atomic_t vmci_num_guest_devices = ATOMIC_INIT(0); | |
72 | ||
73 | bool vmci_guest_code_active(void) | |
74 | { | |
75 | return atomic_read(&vmci_num_guest_devices) != 0; | |
76 | } | |
77 | ||
78 | u32 vmci_get_vm_context_id(void) | |
79 | { | |
80 | if (vm_context_id == VMCI_INVALID_ID) { | |
1f166439 GZ |
81 | struct vmci_datagram get_cid_msg; |
82 | get_cid_msg.dst = | |
83 | vmci_make_handle(VMCI_HYPERVISOR_CONTEXT_ID, | |
84 | VMCI_GET_CONTEXT_ID); | |
85 | get_cid_msg.src = VMCI_ANON_SRC_HANDLE; | |
86 | get_cid_msg.payload_size = 0; | |
5a19b789 | 87 | vm_context_id = vmci_send_datagram(&get_cid_msg); |
1f166439 GZ |
88 | } |
89 | return vm_context_id; | |
90 | } | |
91 | ||
92 | /* | |
93 | * VM to hypervisor call mechanism. We use the standard VMware naming | |
94 | * convention since shared code is calling this function as well. | |
95 | */ | |
96 | int vmci_send_datagram(struct vmci_datagram *dg) | |
97 | { | |
98 | unsigned long flags; | |
99 | int result; | |
100 | ||
101 | /* Check args. */ | |
102 | if (dg == NULL) | |
103 | return VMCI_ERROR_INVALID_ARGS; | |
104 | ||
105 | /* | |
106 | * Need to acquire spinlock on the device because the datagram | |
107 | * data may be spread over multiple pages and the monitor may | |
108 | * interleave device user rpc calls from multiple | |
109 | * VCPUs. Acquiring the spinlock precludes that | |
110 | * possibility. Disabling interrupts to avoid incoming | |
111 | * datagrams during a "rep out" and possibly landing up in | |
112 | * this function. | |
113 | */ | |
114 | spin_lock_irqsave(&vmci_dev_spinlock, flags); | |
115 | ||
116 | if (vmci_dev_g) { | |
117 | iowrite8_rep(vmci_dev_g->iobase + VMCI_DATA_OUT_ADDR, | |
118 | dg, VMCI_DG_SIZE(dg)); | |
119 | result = ioread32(vmci_dev_g->iobase + VMCI_RESULT_LOW_ADDR); | |
120 | } else { | |
121 | result = VMCI_ERROR_UNAVAILABLE; | |
122 | } | |
123 | ||
124 | spin_unlock_irqrestore(&vmci_dev_spinlock, flags); | |
125 | ||
126 | return result; | |
127 | } | |
128 | EXPORT_SYMBOL_GPL(vmci_send_datagram); | |
129 | ||
130 | /* | |
131 | * Gets called with the new context id if updated or resumed. | |
132 | * Context id. | |
133 | */ | |
134 | static void vmci_guest_cid_update(u32 sub_id, | |
135 | const struct vmci_event_data *event_data, | |
136 | void *client_data) | |
137 | { | |
138 | const struct vmci_event_payld_ctx *ev_payload = | |
139 | vmci_event_data_const_payload(event_data); | |
140 | ||
141 | if (sub_id != ctx_update_sub_id) { | |
142 | pr_devel("Invalid subscriber (ID=0x%x)\n", sub_id); | |
143 | return; | |
144 | } | |
145 | ||
146 | if (!event_data || ev_payload->context_id == VMCI_INVALID_ID) { | |
147 | pr_devel("Invalid event data\n"); | |
148 | return; | |
149 | } | |
150 | ||
151 | pr_devel("Updating context from (ID=0x%x) to (ID=0x%x) on event (type=%d)\n", | |
152 | vm_context_id, ev_payload->context_id, event_data->event); | |
153 | ||
154 | vm_context_id = ev_payload->context_id; | |
155 | } | |
156 | ||
157 | /* | |
158 | * Verify that the host supports the hypercalls we need. If it does not, | |
159 | * try to find fallback hypercalls and use those instead. Returns | |
160 | * true if required hypercalls (or fallback hypercalls) are | |
161 | * supported by the host, false otherwise. | |
162 | */ | |
782f2445 | 163 | static int vmci_check_host_caps(struct pci_dev *pdev) |
1f166439 GZ |
164 | { |
165 | bool result; | |
166 | struct vmci_resource_query_msg *msg; | |
167 | u32 msg_size = sizeof(struct vmci_resource_query_hdr) + | |
168 | VMCI_UTIL_NUM_RESOURCES * sizeof(u32); | |
169 | struct vmci_datagram *check_msg; | |
170 | ||
171 | check_msg = kmalloc(msg_size, GFP_KERNEL); | |
172 | if (!check_msg) { | |
173 | dev_err(&pdev->dev, "%s: Insufficient memory\n", __func__); | |
782f2445 | 174 | return -ENOMEM; |
1f166439 GZ |
175 | } |
176 | ||
177 | check_msg->dst = vmci_make_handle(VMCI_HYPERVISOR_CONTEXT_ID, | |
178 | VMCI_RESOURCES_QUERY); | |
179 | check_msg->src = VMCI_ANON_SRC_HANDLE; | |
180 | check_msg->payload_size = msg_size - VMCI_DG_HEADERSIZE; | |
181 | msg = (struct vmci_resource_query_msg *)VMCI_DG_PAYLOAD(check_msg); | |
182 | ||
183 | msg->num_resources = VMCI_UTIL_NUM_RESOURCES; | |
184 | msg->resources[0] = VMCI_GET_CONTEXT_ID; | |
185 | ||
186 | /* Checks that hyper calls are supported */ | |
187 | result = vmci_send_datagram(check_msg) == 0x01; | |
188 | kfree(check_msg); | |
189 | ||
190 | dev_dbg(&pdev->dev, "%s: Host capability check: %s\n", | |
191 | __func__, result ? "PASSED" : "FAILED"); | |
192 | ||
193 | /* We need the vector. There are no fallbacks. */ | |
782f2445 | 194 | return result ? 0 : -ENXIO; |
1f166439 GZ |
195 | } |
196 | ||
197 | /* | |
198 | * Reads datagrams from the data in port and dispatches them. We | |
199 | * always start reading datagrams into only the first page of the | |
200 | * datagram buffer. If the datagrams don't fit into one page, we | |
201 | * use the maximum datagram buffer size for the remainder of the | |
202 | * invocation. This is a simple heuristic for not penalizing | |
203 | * small datagrams. | |
204 | * | |
205 | * This function assumes that it has exclusive access to the data | |
206 | * in port for the duration of the call. | |
207 | */ | |
208 | static void vmci_dispatch_dgs(unsigned long data) | |
209 | { | |
210 | struct vmci_guest_device *vmci_dev = (struct vmci_guest_device *)data; | |
211 | u8 *dg_in_buffer = vmci_dev->data_buffer; | |
212 | struct vmci_datagram *dg; | |
213 | size_t dg_in_buffer_size = VMCI_MAX_DG_SIZE; | |
214 | size_t current_dg_in_buffer_size = PAGE_SIZE; | |
215 | size_t remaining_bytes; | |
216 | ||
217 | BUILD_BUG_ON(VMCI_MAX_DG_SIZE < PAGE_SIZE); | |
218 | ||
219 | ioread8_rep(vmci_dev->iobase + VMCI_DATA_IN_ADDR, | |
220 | vmci_dev->data_buffer, current_dg_in_buffer_size); | |
221 | dg = (struct vmci_datagram *)dg_in_buffer; | |
222 | remaining_bytes = current_dg_in_buffer_size; | |
223 | ||
224 | while (dg->dst.resource != VMCI_INVALID_ID || | |
225 | remaining_bytes > PAGE_SIZE) { | |
226 | unsigned dg_in_size; | |
227 | ||
228 | /* | |
229 | * When the input buffer spans multiple pages, a datagram can | |
230 | * start on any page boundary in the buffer. | |
231 | */ | |
232 | if (dg->dst.resource == VMCI_INVALID_ID) { | |
233 | dg = (struct vmci_datagram *)roundup( | |
234 | (uintptr_t)dg + 1, PAGE_SIZE); | |
235 | remaining_bytes = | |
236 | (size_t)(dg_in_buffer + | |
237 | current_dg_in_buffer_size - | |
238 | (u8 *)dg); | |
239 | continue; | |
240 | } | |
241 | ||
242 | dg_in_size = VMCI_DG_SIZE_ALIGNED(dg); | |
243 | ||
244 | if (dg_in_size <= dg_in_buffer_size) { | |
245 | int result; | |
246 | ||
247 | /* | |
248 | * If the remaining bytes in the datagram | |
249 | * buffer doesn't contain the complete | |
250 | * datagram, we first make sure we have enough | |
251 | * room for it and then we read the reminder | |
252 | * of the datagram and possibly any following | |
253 | * datagrams. | |
254 | */ | |
255 | if (dg_in_size > remaining_bytes) { | |
256 | if (remaining_bytes != | |
257 | current_dg_in_buffer_size) { | |
258 | ||
259 | /* | |
260 | * We move the partial | |
261 | * datagram to the front and | |
262 | * read the reminder of the | |
263 | * datagram and possibly | |
264 | * following calls into the | |
265 | * following bytes. | |
266 | */ | |
267 | memmove(dg_in_buffer, dg_in_buffer + | |
268 | current_dg_in_buffer_size - | |
269 | remaining_bytes, | |
270 | remaining_bytes); | |
271 | dg = (struct vmci_datagram *) | |
272 | dg_in_buffer; | |
273 | } | |
274 | ||
275 | if (current_dg_in_buffer_size != | |
276 | dg_in_buffer_size) | |
277 | current_dg_in_buffer_size = | |
278 | dg_in_buffer_size; | |
279 | ||
280 | ioread8_rep(vmci_dev->iobase + | |
281 | VMCI_DATA_IN_ADDR, | |
282 | vmci_dev->data_buffer + | |
283 | remaining_bytes, | |
284 | current_dg_in_buffer_size - | |
285 | remaining_bytes); | |
286 | } | |
287 | ||
288 | /* | |
289 | * We special case event datagrams from the | |
290 | * hypervisor. | |
291 | */ | |
292 | if (dg->src.context == VMCI_HYPERVISOR_CONTEXT_ID && | |
293 | dg->dst.resource == VMCI_EVENT_HANDLER) { | |
294 | result = vmci_event_dispatch(dg); | |
295 | } else { | |
296 | result = vmci_datagram_invoke_guest_handler(dg); | |
297 | } | |
298 | if (result < VMCI_SUCCESS) | |
299 | dev_dbg(vmci_dev->dev, | |
300 | "Datagram with resource (ID=0x%x) failed (err=%d)\n", | |
301 | dg->dst.resource, result); | |
302 | ||
303 | /* On to the next datagram. */ | |
304 | dg = (struct vmci_datagram *)((u8 *)dg + | |
305 | dg_in_size); | |
306 | } else { | |
307 | size_t bytes_to_skip; | |
308 | ||
309 | /* | |
310 | * Datagram doesn't fit in datagram buffer of maximal | |
311 | * size. We drop it. | |
312 | */ | |
313 | dev_dbg(vmci_dev->dev, | |
314 | "Failed to receive datagram (size=%u bytes)\n", | |
315 | dg_in_size); | |
316 | ||
317 | bytes_to_skip = dg_in_size - remaining_bytes; | |
318 | if (current_dg_in_buffer_size != dg_in_buffer_size) | |
319 | current_dg_in_buffer_size = dg_in_buffer_size; | |
320 | ||
321 | for (;;) { | |
322 | ioread8_rep(vmci_dev->iobase + | |
323 | VMCI_DATA_IN_ADDR, | |
324 | vmci_dev->data_buffer, | |
325 | current_dg_in_buffer_size); | |
326 | if (bytes_to_skip <= current_dg_in_buffer_size) | |
327 | break; | |
328 | ||
329 | bytes_to_skip -= current_dg_in_buffer_size; | |
330 | } | |
331 | dg = (struct vmci_datagram *)(dg_in_buffer + | |
332 | bytes_to_skip); | |
333 | } | |
334 | ||
335 | remaining_bytes = | |
336 | (size_t) (dg_in_buffer + current_dg_in_buffer_size - | |
337 | (u8 *)dg); | |
338 | ||
339 | if (remaining_bytes < VMCI_DG_HEADERSIZE) { | |
340 | /* Get the next batch of datagrams. */ | |
341 | ||
342 | ioread8_rep(vmci_dev->iobase + VMCI_DATA_IN_ADDR, | |
343 | vmci_dev->data_buffer, | |
344 | current_dg_in_buffer_size); | |
345 | dg = (struct vmci_datagram *)dg_in_buffer; | |
346 | remaining_bytes = current_dg_in_buffer_size; | |
347 | } | |
348 | } | |
349 | } | |
350 | ||
351 | /* | |
352 | * Scans the notification bitmap for raised flags, clears them | |
353 | * and handles the notifications. | |
354 | */ | |
355 | static void vmci_process_bitmap(unsigned long data) | |
356 | { | |
357 | struct vmci_guest_device *dev = (struct vmci_guest_device *)data; | |
358 | ||
359 | if (!dev->notification_bitmap) { | |
360 | dev_dbg(dev->dev, "No bitmap present in %s\n", __func__); | |
361 | return; | |
362 | } | |
363 | ||
364 | vmci_dbell_scan_notification_entries(dev->notification_bitmap); | |
365 | } | |
366 | ||
1f166439 GZ |
367 | /* |
368 | * Interrupt handler for legacy or MSI interrupt, or for first MSI-X | |
369 | * interrupt (vector VMCI_INTR_DATAGRAM). | |
370 | */ | |
371 | static irqreturn_t vmci_interrupt(int irq, void *_dev) | |
372 | { | |
373 | struct vmci_guest_device *dev = _dev; | |
374 | ||
375 | /* | |
376 | * If we are using MSI-X with exclusive vectors then we simply schedule | |
377 | * the datagram tasklet, since we know the interrupt was meant for us. | |
378 | * Otherwise we must read the ICR to determine what to do. | |
379 | */ | |
380 | ||
3bb434cd | 381 | if (dev->exclusive_vectors) { |
1f166439 GZ |
382 | tasklet_schedule(&dev->datagram_tasklet); |
383 | } else { | |
384 | unsigned int icr; | |
385 | ||
386 | /* Acknowledge interrupt and determine what needs doing. */ | |
387 | icr = ioread32(dev->iobase + VMCI_ICR_ADDR); | |
388 | if (icr == 0 || icr == ~0) | |
389 | return IRQ_NONE; | |
390 | ||
391 | if (icr & VMCI_ICR_DATAGRAM) { | |
392 | tasklet_schedule(&dev->datagram_tasklet); | |
393 | icr &= ~VMCI_ICR_DATAGRAM; | |
394 | } | |
395 | ||
396 | if (icr & VMCI_ICR_NOTIFICATION) { | |
397 | tasklet_schedule(&dev->bm_tasklet); | |
398 | icr &= ~VMCI_ICR_NOTIFICATION; | |
399 | } | |
400 | ||
401 | if (icr != 0) | |
402 | dev_warn(dev->dev, | |
403 | "Ignoring unknown interrupt cause (%d)\n", | |
404 | icr); | |
405 | } | |
406 | ||
407 | return IRQ_HANDLED; | |
408 | } | |
409 | ||
410 | /* | |
411 | * Interrupt handler for MSI-X interrupt vector VMCI_INTR_NOTIFICATION, | |
412 | * which is for the notification bitmap. Will only get called if we are | |
413 | * using MSI-X with exclusive vectors. | |
414 | */ | |
415 | static irqreturn_t vmci_interrupt_bm(int irq, void *_dev) | |
416 | { | |
417 | struct vmci_guest_device *dev = _dev; | |
418 | ||
419 | /* For MSI-X we can just assume it was meant for us. */ | |
420 | tasklet_schedule(&dev->bm_tasklet); | |
421 | ||
422 | return IRQ_HANDLED; | |
423 | } | |
424 | ||
425 | /* | |
426 | * Most of the initialization at module load time is done here. | |
427 | */ | |
428 | static int vmci_guest_probe_device(struct pci_dev *pdev, | |
429 | const struct pci_device_id *id) | |
430 | { | |
431 | struct vmci_guest_device *vmci_dev; | |
432 | void __iomem *iobase; | |
433 | unsigned int capabilities; | |
f2db7361 | 434 | unsigned int caps_in_use; |
1f166439 GZ |
435 | unsigned long cmd; |
436 | int vmci_err; | |
437 | int error; | |
438 | ||
439 | dev_dbg(&pdev->dev, "Probing for vmci/PCI guest device\n"); | |
440 | ||
441 | error = pcim_enable_device(pdev); | |
442 | if (error) { | |
443 | dev_err(&pdev->dev, | |
444 | "Failed to enable VMCI device: %d\n", error); | |
445 | return error; | |
446 | } | |
447 | ||
448 | error = pcim_iomap_regions(pdev, 1 << 0, KBUILD_MODNAME); | |
449 | if (error) { | |
450 | dev_err(&pdev->dev, "Failed to reserve/map IO regions\n"); | |
451 | return error; | |
452 | } | |
453 | ||
454 | iobase = pcim_iomap_table(pdev)[0]; | |
455 | ||
456 | dev_info(&pdev->dev, "Found VMCI PCI device at %#lx, irq %u\n", | |
457 | (unsigned long)iobase, pdev->irq); | |
458 | ||
459 | vmci_dev = devm_kzalloc(&pdev->dev, sizeof(*vmci_dev), GFP_KERNEL); | |
460 | if (!vmci_dev) { | |
461 | dev_err(&pdev->dev, | |
462 | "Can't allocate memory for VMCI device\n"); | |
463 | return -ENOMEM; | |
464 | } | |
465 | ||
466 | vmci_dev->dev = &pdev->dev; | |
1f166439 GZ |
467 | vmci_dev->exclusive_vectors = false; |
468 | vmci_dev->iobase = iobase; | |
469 | ||
470 | tasklet_init(&vmci_dev->datagram_tasklet, | |
471 | vmci_dispatch_dgs, (unsigned long)vmci_dev); | |
472 | tasklet_init(&vmci_dev->bm_tasklet, | |
473 | vmci_process_bitmap, (unsigned long)vmci_dev); | |
474 | ||
475 | vmci_dev->data_buffer = vmalloc(VMCI_MAX_DG_SIZE); | |
476 | if (!vmci_dev->data_buffer) { | |
477 | dev_err(&pdev->dev, | |
478 | "Can't allocate memory for datagram buffer\n"); | |
479 | return -ENOMEM; | |
480 | } | |
481 | ||
482 | pci_set_master(pdev); /* To enable queue_pair functionality. */ | |
483 | ||
484 | /* | |
485 | * Verify that the VMCI Device supports the capabilities that | |
486 | * we need. If the device is missing capabilities that we would | |
487 | * like to use, check for fallback capabilities and use those | |
488 | * instead (so we can run a new VM on old hosts). Fail the load if | |
489 | * a required capability is missing and there is no fallback. | |
490 | * | |
491 | * Right now, we need datagrams. There are no fallbacks. | |
492 | */ | |
493 | capabilities = ioread32(vmci_dev->iobase + VMCI_CAPS_ADDR); | |
494 | if (!(capabilities & VMCI_CAPS_DATAGRAM)) { | |
495 | dev_err(&pdev->dev, "Device does not support datagrams\n"); | |
496 | error = -ENXIO; | |
497 | goto err_free_data_buffer; | |
498 | } | |
f2db7361 VD |
499 | caps_in_use = VMCI_CAPS_DATAGRAM; |
500 | ||
501 | /* | |
502 | * Use 64-bit PPNs if the device supports. | |
503 | * | |
504 | * There is no check for the return value of dma_set_mask_and_coherent | |
505 | * since this driver can handle the default mask values if | |
506 | * dma_set_mask_and_coherent fails. | |
507 | */ | |
508 | if (capabilities & VMCI_CAPS_PPN64) { | |
509 | dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64)); | |
510 | use_ppn64 = true; | |
511 | caps_in_use |= VMCI_CAPS_PPN64; | |
512 | } else { | |
513 | dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(44)); | |
514 | use_ppn64 = false; | |
515 | } | |
1f166439 GZ |
516 | |
517 | /* | |
518 | * If the hardware supports notifications, we will use that as | |
519 | * well. | |
520 | */ | |
521 | if (capabilities & VMCI_CAPS_NOTIFICATIONS) { | |
6d6dfb4f AK |
522 | vmci_dev->notification_bitmap = dma_alloc_coherent( |
523 | &pdev->dev, PAGE_SIZE, &vmci_dev->notification_base, | |
524 | GFP_KERNEL); | |
1f166439 GZ |
525 | if (!vmci_dev->notification_bitmap) { |
526 | dev_warn(&pdev->dev, | |
527 | "Unable to allocate notification bitmap\n"); | |
528 | } else { | |
529 | memset(vmci_dev->notification_bitmap, 0, PAGE_SIZE); | |
f2db7361 | 530 | caps_in_use |= VMCI_CAPS_NOTIFICATIONS; |
1f166439 GZ |
531 | } |
532 | } | |
533 | ||
f2db7361 | 534 | dev_info(&pdev->dev, "Using capabilities 0x%x\n", caps_in_use); |
1f166439 GZ |
535 | |
536 | /* Let the host know which capabilities we intend to use. */ | |
f2db7361 | 537 | iowrite32(caps_in_use, vmci_dev->iobase + VMCI_CAPS_ADDR); |
1f166439 GZ |
538 | |
539 | /* Set up global device so that we can start sending datagrams */ | |
540 | spin_lock_irq(&vmci_dev_spinlock); | |
541 | vmci_dev_g = vmci_dev; | |
6d6dfb4f | 542 | vmci_pdev = pdev; |
1f166439 GZ |
543 | spin_unlock_irq(&vmci_dev_spinlock); |
544 | ||
545 | /* | |
546 | * Register notification bitmap with device if that capability is | |
547 | * used. | |
548 | */ | |
f2db7361 | 549 | if (caps_in_use & VMCI_CAPS_NOTIFICATIONS) { |
6d6dfb4f AK |
550 | unsigned long bitmap_ppn = |
551 | vmci_dev->notification_base >> PAGE_SHIFT; | |
1f166439 GZ |
552 | if (!vmci_dbell_register_notification_bitmap(bitmap_ppn)) { |
553 | dev_warn(&pdev->dev, | |
f2db7361 VD |
554 | "VMCI device unable to register notification bitmap with PPN 0x%lx\n", |
555 | bitmap_ppn); | |
782f2445 | 556 | error = -ENXIO; |
1f166439 GZ |
557 | goto err_remove_vmci_dev_g; |
558 | } | |
559 | } | |
560 | ||
561 | /* Check host capabilities. */ | |
782f2445 DT |
562 | error = vmci_check_host_caps(pdev); |
563 | if (error) | |
1f166439 GZ |
564 | goto err_remove_bitmap; |
565 | ||
566 | /* Enable device. */ | |
567 | ||
568 | /* | |
569 | * We subscribe to the VMCI_EVENT_CTX_ID_UPDATE here so we can | |
570 | * update the internal context id when needed. | |
571 | */ | |
572 | vmci_err = vmci_event_subscribe(VMCI_EVENT_CTX_ID_UPDATE, | |
573 | vmci_guest_cid_update, NULL, | |
574 | &ctx_update_sub_id); | |
575 | if (vmci_err < VMCI_SUCCESS) | |
576 | dev_warn(&pdev->dev, | |
577 | "Failed to subscribe to event (type=%d): %d\n", | |
578 | VMCI_EVENT_CTX_ID_UPDATE, vmci_err); | |
579 | ||
580 | /* | |
581 | * Enable interrupts. Try MSI-X first, then MSI, and then fallback on | |
582 | * legacy interrupts. | |
583 | */ | |
3bb434cd CH |
584 | error = pci_alloc_irq_vectors(pdev, VMCI_MAX_INTRS, VMCI_MAX_INTRS, |
585 | PCI_IRQ_MSIX); | |
c3423563 | 586 | if (error < 0) { |
3bb434cd CH |
587 | error = pci_alloc_irq_vectors(pdev, 1, 1, |
588 | PCI_IRQ_MSIX | PCI_IRQ_MSI | PCI_IRQ_LEGACY); | |
c3423563 | 589 | if (error < 0) |
3bb434cd | 590 | goto err_remove_bitmap; |
1f166439 | 591 | } else { |
3bb434cd | 592 | vmci_dev->exclusive_vectors = true; |
1f166439 GZ |
593 | } |
594 | ||
595 | /* | |
596 | * Request IRQ for legacy or MSI interrupts, or for first | |
597 | * MSI-X vector. | |
598 | */ | |
3bb434cd CH |
599 | error = request_irq(pci_irq_vector(pdev, 0), vmci_interrupt, |
600 | IRQF_SHARED, KBUILD_MODNAME, vmci_dev); | |
1f166439 GZ |
601 | if (error) { |
602 | dev_err(&pdev->dev, "Irq %u in use: %d\n", | |
3bb434cd | 603 | pci_irq_vector(pdev, 0), error); |
1f166439 GZ |
604 | goto err_disable_msi; |
605 | } | |
606 | ||
607 | /* | |
608 | * For MSI-X with exclusive vectors we need to request an | |
609 | * interrupt for each vector so that we get a separate | |
610 | * interrupt handler routine. This allows us to distinguish | |
611 | * between the vectors. | |
612 | */ | |
613 | if (vmci_dev->exclusive_vectors) { | |
3bb434cd | 614 | error = request_irq(pci_irq_vector(pdev, 1), |
1f166439 GZ |
615 | vmci_interrupt_bm, 0, KBUILD_MODNAME, |
616 | vmci_dev); | |
617 | if (error) { | |
618 | dev_err(&pdev->dev, | |
619 | "Failed to allocate irq %u: %d\n", | |
3bb434cd | 620 | pci_irq_vector(pdev, 1), error); |
1f166439 GZ |
621 | goto err_free_irq; |
622 | } | |
623 | } | |
624 | ||
625 | dev_dbg(&pdev->dev, "Registered device\n"); | |
626 | ||
627 | atomic_inc(&vmci_num_guest_devices); | |
628 | ||
629 | /* Enable specific interrupt bits. */ | |
630 | cmd = VMCI_IMR_DATAGRAM; | |
f2db7361 | 631 | if (caps_in_use & VMCI_CAPS_NOTIFICATIONS) |
1f166439 GZ |
632 | cmd |= VMCI_IMR_NOTIFICATION; |
633 | iowrite32(cmd, vmci_dev->iobase + VMCI_IMR_ADDR); | |
634 | ||
635 | /* Enable interrupts. */ | |
636 | iowrite32(VMCI_CONTROL_INT_ENABLE, | |
637 | vmci_dev->iobase + VMCI_CONTROL_ADDR); | |
638 | ||
639 | pci_set_drvdata(pdev, vmci_dev); | |
b1bba80a SG |
640 | |
641 | vmci_call_vsock_callback(false); | |
1f166439 GZ |
642 | return 0; |
643 | ||
644 | err_free_irq: | |
3bb434cd | 645 | free_irq(pci_irq_vector(pdev, 0), vmci_dev); |
1f166439 GZ |
646 | tasklet_kill(&vmci_dev->datagram_tasklet); |
647 | tasklet_kill(&vmci_dev->bm_tasklet); | |
648 | ||
649 | err_disable_msi: | |
3bb434cd | 650 | pci_free_irq_vectors(pdev); |
1f166439 GZ |
651 | |
652 | vmci_err = vmci_event_unsubscribe(ctx_update_sub_id); | |
653 | if (vmci_err < VMCI_SUCCESS) | |
654 | dev_warn(&pdev->dev, | |
655 | "Failed to unsubscribe from event (type=%d) with subscriber (ID=0x%x): %d\n", | |
656 | VMCI_EVENT_CTX_ID_UPDATE, ctx_update_sub_id, vmci_err); | |
657 | ||
658 | err_remove_bitmap: | |
659 | if (vmci_dev->notification_bitmap) { | |
660 | iowrite32(VMCI_CONTROL_RESET, | |
661 | vmci_dev->iobase + VMCI_CONTROL_ADDR); | |
6d6dfb4f AK |
662 | dma_free_coherent(&pdev->dev, PAGE_SIZE, |
663 | vmci_dev->notification_bitmap, | |
664 | vmci_dev->notification_base); | |
1f166439 GZ |
665 | } |
666 | ||
667 | err_remove_vmci_dev_g: | |
668 | spin_lock_irq(&vmci_dev_spinlock); | |
6d6dfb4f | 669 | vmci_pdev = NULL; |
1f166439 GZ |
670 | vmci_dev_g = NULL; |
671 | spin_unlock_irq(&vmci_dev_spinlock); | |
672 | ||
673 | err_free_data_buffer: | |
674 | vfree(vmci_dev->data_buffer); | |
675 | ||
676 | /* The rest are managed resources and will be freed by PCI core */ | |
677 | return error; | |
678 | } | |
679 | ||
680 | static void vmci_guest_remove_device(struct pci_dev *pdev) | |
681 | { | |
682 | struct vmci_guest_device *vmci_dev = pci_get_drvdata(pdev); | |
683 | int vmci_err; | |
684 | ||
685 | dev_dbg(&pdev->dev, "Removing device\n"); | |
686 | ||
687 | atomic_dec(&vmci_num_guest_devices); | |
688 | ||
689 | vmci_qp_guest_endpoints_exit(); | |
690 | ||
691 | vmci_err = vmci_event_unsubscribe(ctx_update_sub_id); | |
692 | if (vmci_err < VMCI_SUCCESS) | |
693 | dev_warn(&pdev->dev, | |
694 | "Failed to unsubscribe from event (type=%d) with subscriber (ID=0x%x): %d\n", | |
695 | VMCI_EVENT_CTX_ID_UPDATE, ctx_update_sub_id, vmci_err); | |
696 | ||
697 | spin_lock_irq(&vmci_dev_spinlock); | |
698 | vmci_dev_g = NULL; | |
6d6dfb4f | 699 | vmci_pdev = NULL; |
1f166439 GZ |
700 | spin_unlock_irq(&vmci_dev_spinlock); |
701 | ||
702 | dev_dbg(&pdev->dev, "Resetting vmci device\n"); | |
703 | iowrite32(VMCI_CONTROL_RESET, vmci_dev->iobase + VMCI_CONTROL_ADDR); | |
704 | ||
705 | /* | |
706 | * Free IRQ and then disable MSI/MSI-X as appropriate. For | |
707 | * MSI-X, we might have multiple vectors, each with their own | |
708 | * IRQ, which we must free too. | |
709 | */ | |
3bb434cd CH |
710 | if (vmci_dev->exclusive_vectors) |
711 | free_irq(pci_irq_vector(pdev, 1), vmci_dev); | |
712 | free_irq(pci_irq_vector(pdev, 0), vmci_dev); | |
713 | pci_free_irq_vectors(pdev); | |
1f166439 GZ |
714 | |
715 | tasklet_kill(&vmci_dev->datagram_tasklet); | |
716 | tasklet_kill(&vmci_dev->bm_tasklet); | |
717 | ||
718 | if (vmci_dev->notification_bitmap) { | |
719 | /* | |
720 | * The device reset above cleared the bitmap state of the | |
721 | * device, so we can safely free it here. | |
722 | */ | |
723 | ||
6d6dfb4f AK |
724 | dma_free_coherent(&pdev->dev, PAGE_SIZE, |
725 | vmci_dev->notification_bitmap, | |
726 | vmci_dev->notification_base); | |
1f166439 GZ |
727 | } |
728 | ||
729 | vfree(vmci_dev->data_buffer); | |
730 | ||
731 | /* The rest are managed resources and will be freed by PCI core */ | |
732 | } | |
733 | ||
32182cd3 | 734 | static const struct pci_device_id vmci_ids[] = { |
1f166439 GZ |
735 | { PCI_DEVICE(PCI_VENDOR_ID_VMWARE, PCI_DEVICE_ID_VMWARE_VMCI), }, |
736 | { 0 }, | |
737 | }; | |
738 | MODULE_DEVICE_TABLE(pci, vmci_ids); | |
739 | ||
740 | static struct pci_driver vmci_guest_driver = { | |
741 | .name = KBUILD_MODNAME, | |
742 | .id_table = vmci_ids, | |
743 | .probe = vmci_guest_probe_device, | |
744 | .remove = vmci_guest_remove_device, | |
745 | }; | |
746 | ||
747 | int __init vmci_guest_init(void) | |
748 | { | |
749 | return pci_register_driver(&vmci_guest_driver); | |
750 | } | |
751 | ||
752 | void __exit vmci_guest_exit(void) | |
753 | { | |
754 | pci_unregister_driver(&vmci_guest_driver); | |
755 | } |