]>
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; | |
e01153c7 | 48 | void __iomem *mmio_base; |
1f166439 | 49 | |
1f166439 | 50 | bool exclusive_vectors; |
1f166439 GZ |
51 | |
52 | struct tasklet_struct datagram_tasklet; | |
53 | struct tasklet_struct bm_tasklet; | |
54 | ||
55 | void *data_buffer; | |
56 | void *notification_bitmap; | |
6d6dfb4f | 57 | dma_addr_t notification_base; |
1f166439 GZ |
58 | }; |
59 | ||
f2db7361 VD |
60 | static bool use_ppn64; |
61 | ||
62 | bool vmci_use_ppn64(void) | |
63 | { | |
64 | return use_ppn64; | |
65 | } | |
66 | ||
1f166439 | 67 | /* vmci_dev singleton device and supporting data*/ |
6d6dfb4f | 68 | struct pci_dev *vmci_pdev; |
1f166439 GZ |
69 | static struct vmci_guest_device *vmci_dev_g; |
70 | static DEFINE_SPINLOCK(vmci_dev_spinlock); | |
71 | ||
72 | static atomic_t vmci_num_guest_devices = ATOMIC_INIT(0); | |
73 | ||
74 | bool vmci_guest_code_active(void) | |
75 | { | |
76 | return atomic_read(&vmci_num_guest_devices) != 0; | |
77 | } | |
78 | ||
79 | u32 vmci_get_vm_context_id(void) | |
80 | { | |
81 | if (vm_context_id == VMCI_INVALID_ID) { | |
1f166439 GZ |
82 | struct vmci_datagram get_cid_msg; |
83 | get_cid_msg.dst = | |
84 | vmci_make_handle(VMCI_HYPERVISOR_CONTEXT_ID, | |
85 | VMCI_GET_CONTEXT_ID); | |
86 | get_cid_msg.src = VMCI_ANON_SRC_HANDLE; | |
87 | get_cid_msg.payload_size = 0; | |
5a19b789 | 88 | vm_context_id = vmci_send_datagram(&get_cid_msg); |
1f166439 GZ |
89 | } |
90 | return vm_context_id; | |
91 | } | |
92 | ||
e01153c7 JH |
93 | static unsigned int vmci_read_reg(struct vmci_guest_device *dev, u32 reg) |
94 | { | |
95 | if (dev->mmio_base != NULL) | |
96 | return readl(dev->mmio_base + reg); | |
97 | return ioread32(dev->iobase + reg); | |
98 | } | |
99 | ||
100 | static void vmci_write_reg(struct vmci_guest_device *dev, u32 val, u32 reg) | |
101 | { | |
102 | if (dev->mmio_base != NULL) | |
103 | writel(val, dev->mmio_base + reg); | |
104 | else | |
105 | iowrite32(val, dev->iobase + reg); | |
106 | } | |
107 | ||
1f166439 GZ |
108 | /* |
109 | * VM to hypervisor call mechanism. We use the standard VMware naming | |
110 | * convention since shared code is calling this function as well. | |
111 | */ | |
112 | int vmci_send_datagram(struct vmci_datagram *dg) | |
113 | { | |
114 | unsigned long flags; | |
115 | int result; | |
116 | ||
117 | /* Check args. */ | |
118 | if (dg == NULL) | |
119 | return VMCI_ERROR_INVALID_ARGS; | |
120 | ||
121 | /* | |
122 | * Need to acquire spinlock on the device because the datagram | |
123 | * data may be spread over multiple pages and the monitor may | |
124 | * interleave device user rpc calls from multiple | |
125 | * VCPUs. Acquiring the spinlock precludes that | |
126 | * possibility. Disabling interrupts to avoid incoming | |
127 | * datagrams during a "rep out" and possibly landing up in | |
128 | * this function. | |
129 | */ | |
130 | spin_lock_irqsave(&vmci_dev_spinlock, flags); | |
131 | ||
132 | if (vmci_dev_g) { | |
133 | iowrite8_rep(vmci_dev_g->iobase + VMCI_DATA_OUT_ADDR, | |
134 | dg, VMCI_DG_SIZE(dg)); | |
e01153c7 | 135 | result = vmci_read_reg(vmci_dev_g, VMCI_RESULT_LOW_ADDR); |
1f166439 GZ |
136 | } else { |
137 | result = VMCI_ERROR_UNAVAILABLE; | |
138 | } | |
139 | ||
140 | spin_unlock_irqrestore(&vmci_dev_spinlock, flags); | |
141 | ||
142 | return result; | |
143 | } | |
144 | EXPORT_SYMBOL_GPL(vmci_send_datagram); | |
145 | ||
146 | /* | |
147 | * Gets called with the new context id if updated or resumed. | |
148 | * Context id. | |
149 | */ | |
150 | static void vmci_guest_cid_update(u32 sub_id, | |
151 | const struct vmci_event_data *event_data, | |
152 | void *client_data) | |
153 | { | |
154 | const struct vmci_event_payld_ctx *ev_payload = | |
155 | vmci_event_data_const_payload(event_data); | |
156 | ||
157 | if (sub_id != ctx_update_sub_id) { | |
158 | pr_devel("Invalid subscriber (ID=0x%x)\n", sub_id); | |
159 | return; | |
160 | } | |
161 | ||
162 | if (!event_data || ev_payload->context_id == VMCI_INVALID_ID) { | |
163 | pr_devel("Invalid event data\n"); | |
164 | return; | |
165 | } | |
166 | ||
167 | pr_devel("Updating context from (ID=0x%x) to (ID=0x%x) on event (type=%d)\n", | |
168 | vm_context_id, ev_payload->context_id, event_data->event); | |
169 | ||
170 | vm_context_id = ev_payload->context_id; | |
171 | } | |
172 | ||
173 | /* | |
174 | * Verify that the host supports the hypercalls we need. If it does not, | |
175 | * try to find fallback hypercalls and use those instead. Returns | |
176 | * true if required hypercalls (or fallback hypercalls) are | |
177 | * supported by the host, false otherwise. | |
178 | */ | |
782f2445 | 179 | static int vmci_check_host_caps(struct pci_dev *pdev) |
1f166439 GZ |
180 | { |
181 | bool result; | |
182 | struct vmci_resource_query_msg *msg; | |
183 | u32 msg_size = sizeof(struct vmci_resource_query_hdr) + | |
184 | VMCI_UTIL_NUM_RESOURCES * sizeof(u32); | |
185 | struct vmci_datagram *check_msg; | |
186 | ||
b2192cfe | 187 | check_msg = kzalloc(msg_size, GFP_KERNEL); |
1f166439 GZ |
188 | if (!check_msg) { |
189 | dev_err(&pdev->dev, "%s: Insufficient memory\n", __func__); | |
782f2445 | 190 | return -ENOMEM; |
1f166439 GZ |
191 | } |
192 | ||
193 | check_msg->dst = vmci_make_handle(VMCI_HYPERVISOR_CONTEXT_ID, | |
194 | VMCI_RESOURCES_QUERY); | |
195 | check_msg->src = VMCI_ANON_SRC_HANDLE; | |
196 | check_msg->payload_size = msg_size - VMCI_DG_HEADERSIZE; | |
197 | msg = (struct vmci_resource_query_msg *)VMCI_DG_PAYLOAD(check_msg); | |
198 | ||
199 | msg->num_resources = VMCI_UTIL_NUM_RESOURCES; | |
200 | msg->resources[0] = VMCI_GET_CONTEXT_ID; | |
201 | ||
202 | /* Checks that hyper calls are supported */ | |
203 | result = vmci_send_datagram(check_msg) == 0x01; | |
204 | kfree(check_msg); | |
205 | ||
206 | dev_dbg(&pdev->dev, "%s: Host capability check: %s\n", | |
207 | __func__, result ? "PASSED" : "FAILED"); | |
208 | ||
209 | /* We need the vector. There are no fallbacks. */ | |
782f2445 | 210 | return result ? 0 : -ENXIO; |
1f166439 GZ |
211 | } |
212 | ||
213 | /* | |
214 | * Reads datagrams from the data in port and dispatches them. We | |
215 | * always start reading datagrams into only the first page of the | |
216 | * datagram buffer. If the datagrams don't fit into one page, we | |
217 | * use the maximum datagram buffer size for the remainder of the | |
218 | * invocation. This is a simple heuristic for not penalizing | |
219 | * small datagrams. | |
220 | * | |
221 | * This function assumes that it has exclusive access to the data | |
222 | * in port for the duration of the call. | |
223 | */ | |
224 | static void vmci_dispatch_dgs(unsigned long data) | |
225 | { | |
226 | struct vmci_guest_device *vmci_dev = (struct vmci_guest_device *)data; | |
227 | u8 *dg_in_buffer = vmci_dev->data_buffer; | |
228 | struct vmci_datagram *dg; | |
229 | size_t dg_in_buffer_size = VMCI_MAX_DG_SIZE; | |
230 | size_t current_dg_in_buffer_size = PAGE_SIZE; | |
231 | size_t remaining_bytes; | |
232 | ||
233 | BUILD_BUG_ON(VMCI_MAX_DG_SIZE < PAGE_SIZE); | |
234 | ||
235 | ioread8_rep(vmci_dev->iobase + VMCI_DATA_IN_ADDR, | |
236 | vmci_dev->data_buffer, current_dg_in_buffer_size); | |
237 | dg = (struct vmci_datagram *)dg_in_buffer; | |
238 | remaining_bytes = current_dg_in_buffer_size; | |
239 | ||
240 | while (dg->dst.resource != VMCI_INVALID_ID || | |
241 | remaining_bytes > PAGE_SIZE) { | |
242 | unsigned dg_in_size; | |
243 | ||
244 | /* | |
245 | * When the input buffer spans multiple pages, a datagram can | |
246 | * start on any page boundary in the buffer. | |
247 | */ | |
248 | if (dg->dst.resource == VMCI_INVALID_ID) { | |
249 | dg = (struct vmci_datagram *)roundup( | |
250 | (uintptr_t)dg + 1, PAGE_SIZE); | |
251 | remaining_bytes = | |
252 | (size_t)(dg_in_buffer + | |
253 | current_dg_in_buffer_size - | |
254 | (u8 *)dg); | |
255 | continue; | |
256 | } | |
257 | ||
258 | dg_in_size = VMCI_DG_SIZE_ALIGNED(dg); | |
259 | ||
260 | if (dg_in_size <= dg_in_buffer_size) { | |
261 | int result; | |
262 | ||
263 | /* | |
264 | * If the remaining bytes in the datagram | |
265 | * buffer doesn't contain the complete | |
266 | * datagram, we first make sure we have enough | |
267 | * room for it and then we read the reminder | |
268 | * of the datagram and possibly any following | |
269 | * datagrams. | |
270 | */ | |
271 | if (dg_in_size > remaining_bytes) { | |
272 | if (remaining_bytes != | |
273 | current_dg_in_buffer_size) { | |
274 | ||
275 | /* | |
276 | * We move the partial | |
277 | * datagram to the front and | |
278 | * read the reminder of the | |
279 | * datagram and possibly | |
280 | * following calls into the | |
281 | * following bytes. | |
282 | */ | |
283 | memmove(dg_in_buffer, dg_in_buffer + | |
284 | current_dg_in_buffer_size - | |
285 | remaining_bytes, | |
286 | remaining_bytes); | |
287 | dg = (struct vmci_datagram *) | |
288 | dg_in_buffer; | |
289 | } | |
290 | ||
291 | if (current_dg_in_buffer_size != | |
292 | dg_in_buffer_size) | |
293 | current_dg_in_buffer_size = | |
294 | dg_in_buffer_size; | |
295 | ||
296 | ioread8_rep(vmci_dev->iobase + | |
297 | VMCI_DATA_IN_ADDR, | |
298 | vmci_dev->data_buffer + | |
299 | remaining_bytes, | |
300 | current_dg_in_buffer_size - | |
301 | remaining_bytes); | |
302 | } | |
303 | ||
304 | /* | |
305 | * We special case event datagrams from the | |
306 | * hypervisor. | |
307 | */ | |
308 | if (dg->src.context == VMCI_HYPERVISOR_CONTEXT_ID && | |
309 | dg->dst.resource == VMCI_EVENT_HANDLER) { | |
310 | result = vmci_event_dispatch(dg); | |
311 | } else { | |
312 | result = vmci_datagram_invoke_guest_handler(dg); | |
313 | } | |
314 | if (result < VMCI_SUCCESS) | |
315 | dev_dbg(vmci_dev->dev, | |
316 | "Datagram with resource (ID=0x%x) failed (err=%d)\n", | |
317 | dg->dst.resource, result); | |
318 | ||
319 | /* On to the next datagram. */ | |
320 | dg = (struct vmci_datagram *)((u8 *)dg + | |
321 | dg_in_size); | |
322 | } else { | |
323 | size_t bytes_to_skip; | |
324 | ||
325 | /* | |
326 | * Datagram doesn't fit in datagram buffer of maximal | |
327 | * size. We drop it. | |
328 | */ | |
329 | dev_dbg(vmci_dev->dev, | |
330 | "Failed to receive datagram (size=%u bytes)\n", | |
331 | dg_in_size); | |
332 | ||
333 | bytes_to_skip = dg_in_size - remaining_bytes; | |
334 | if (current_dg_in_buffer_size != dg_in_buffer_size) | |
335 | current_dg_in_buffer_size = dg_in_buffer_size; | |
336 | ||
337 | for (;;) { | |
338 | ioread8_rep(vmci_dev->iobase + | |
339 | VMCI_DATA_IN_ADDR, | |
340 | vmci_dev->data_buffer, | |
341 | current_dg_in_buffer_size); | |
342 | if (bytes_to_skip <= current_dg_in_buffer_size) | |
343 | break; | |
344 | ||
345 | bytes_to_skip -= current_dg_in_buffer_size; | |
346 | } | |
347 | dg = (struct vmci_datagram *)(dg_in_buffer + | |
348 | bytes_to_skip); | |
349 | } | |
350 | ||
351 | remaining_bytes = | |
352 | (size_t) (dg_in_buffer + current_dg_in_buffer_size - | |
353 | (u8 *)dg); | |
354 | ||
355 | if (remaining_bytes < VMCI_DG_HEADERSIZE) { | |
356 | /* Get the next batch of datagrams. */ | |
357 | ||
358 | ioread8_rep(vmci_dev->iobase + VMCI_DATA_IN_ADDR, | |
359 | vmci_dev->data_buffer, | |
360 | current_dg_in_buffer_size); | |
361 | dg = (struct vmci_datagram *)dg_in_buffer; | |
362 | remaining_bytes = current_dg_in_buffer_size; | |
363 | } | |
364 | } | |
365 | } | |
366 | ||
367 | /* | |
368 | * Scans the notification bitmap for raised flags, clears them | |
369 | * and handles the notifications. | |
370 | */ | |
371 | static void vmci_process_bitmap(unsigned long data) | |
372 | { | |
373 | struct vmci_guest_device *dev = (struct vmci_guest_device *)data; | |
374 | ||
375 | if (!dev->notification_bitmap) { | |
376 | dev_dbg(dev->dev, "No bitmap present in %s\n", __func__); | |
377 | return; | |
378 | } | |
379 | ||
380 | vmci_dbell_scan_notification_entries(dev->notification_bitmap); | |
381 | } | |
382 | ||
1f166439 GZ |
383 | /* |
384 | * Interrupt handler for legacy or MSI interrupt, or for first MSI-X | |
385 | * interrupt (vector VMCI_INTR_DATAGRAM). | |
386 | */ | |
387 | static irqreturn_t vmci_interrupt(int irq, void *_dev) | |
388 | { | |
389 | struct vmci_guest_device *dev = _dev; | |
390 | ||
391 | /* | |
392 | * If we are using MSI-X with exclusive vectors then we simply schedule | |
393 | * the datagram tasklet, since we know the interrupt was meant for us. | |
394 | * Otherwise we must read the ICR to determine what to do. | |
395 | */ | |
396 | ||
3bb434cd | 397 | if (dev->exclusive_vectors) { |
1f166439 GZ |
398 | tasklet_schedule(&dev->datagram_tasklet); |
399 | } else { | |
400 | unsigned int icr; | |
401 | ||
402 | /* Acknowledge interrupt and determine what needs doing. */ | |
e01153c7 | 403 | icr = vmci_read_reg(dev, VMCI_ICR_ADDR); |
1f166439 GZ |
404 | if (icr == 0 || icr == ~0) |
405 | return IRQ_NONE; | |
406 | ||
407 | if (icr & VMCI_ICR_DATAGRAM) { | |
408 | tasklet_schedule(&dev->datagram_tasklet); | |
409 | icr &= ~VMCI_ICR_DATAGRAM; | |
410 | } | |
411 | ||
412 | if (icr & VMCI_ICR_NOTIFICATION) { | |
413 | tasklet_schedule(&dev->bm_tasklet); | |
414 | icr &= ~VMCI_ICR_NOTIFICATION; | |
415 | } | |
416 | ||
417 | if (icr != 0) | |
418 | dev_warn(dev->dev, | |
419 | "Ignoring unknown interrupt cause (%d)\n", | |
420 | icr); | |
421 | } | |
422 | ||
423 | return IRQ_HANDLED; | |
424 | } | |
425 | ||
426 | /* | |
427 | * Interrupt handler for MSI-X interrupt vector VMCI_INTR_NOTIFICATION, | |
428 | * which is for the notification bitmap. Will only get called if we are | |
429 | * using MSI-X with exclusive vectors. | |
430 | */ | |
431 | static irqreturn_t vmci_interrupt_bm(int irq, void *_dev) | |
432 | { | |
433 | struct vmci_guest_device *dev = _dev; | |
434 | ||
435 | /* For MSI-X we can just assume it was meant for us. */ | |
436 | tasklet_schedule(&dev->bm_tasklet); | |
437 | ||
438 | return IRQ_HANDLED; | |
439 | } | |
440 | ||
441 | /* | |
442 | * Most of the initialization at module load time is done here. | |
443 | */ | |
444 | static int vmci_guest_probe_device(struct pci_dev *pdev, | |
445 | const struct pci_device_id *id) | |
446 | { | |
447 | struct vmci_guest_device *vmci_dev; | |
e01153c7 JH |
448 | void __iomem *iobase = NULL; |
449 | void __iomem *mmio_base = NULL; | |
1f166439 | 450 | unsigned int capabilities; |
f2db7361 | 451 | unsigned int caps_in_use; |
1f166439 GZ |
452 | unsigned long cmd; |
453 | int vmci_err; | |
454 | int error; | |
455 | ||
456 | dev_dbg(&pdev->dev, "Probing for vmci/PCI guest device\n"); | |
457 | ||
458 | error = pcim_enable_device(pdev); | |
459 | if (error) { | |
460 | dev_err(&pdev->dev, | |
461 | "Failed to enable VMCI device: %d\n", error); | |
462 | return error; | |
463 | } | |
464 | ||
e01153c7 JH |
465 | /* |
466 | * The VMCI device with mmio access to registers requests 256KB | |
467 | * for BAR1. If present, driver will use new VMCI device | |
468 | * functionality for register access and datagram send/recv. | |
469 | */ | |
1f166439 | 470 | |
e01153c7 JH |
471 | if (pci_resource_len(pdev, 1) == VMCI_WITH_MMIO_ACCESS_BAR_SIZE) { |
472 | dev_info(&pdev->dev, "MMIO register access is available\n"); | |
473 | mmio_base = pci_iomap_range(pdev, 1, VMCI_MMIO_ACCESS_OFFSET, | |
474 | VMCI_MMIO_ACCESS_SIZE); | |
475 | /* If the map fails, we fall back to IOIO access. */ | |
476 | if (!mmio_base) | |
477 | dev_warn(&pdev->dev, "Failed to map MMIO register access\n"); | |
478 | } | |
1f166439 | 479 | |
e01153c7 JH |
480 | if (!mmio_base) { |
481 | error = pcim_iomap_regions(pdev, BIT(0), KBUILD_MODNAME); | |
482 | if (error) { | |
483 | dev_err(&pdev->dev, "Failed to reserve/map IO regions\n"); | |
484 | return error; | |
485 | } | |
486 | iobase = pcim_iomap_table(pdev)[0]; | |
487 | } | |
1f166439 GZ |
488 | |
489 | vmci_dev = devm_kzalloc(&pdev->dev, sizeof(*vmci_dev), GFP_KERNEL); | |
490 | if (!vmci_dev) { | |
491 | dev_err(&pdev->dev, | |
492 | "Can't allocate memory for VMCI device\n"); | |
493 | return -ENOMEM; | |
494 | } | |
495 | ||
496 | vmci_dev->dev = &pdev->dev; | |
1f166439 GZ |
497 | vmci_dev->exclusive_vectors = false; |
498 | vmci_dev->iobase = iobase; | |
e01153c7 | 499 | vmci_dev->mmio_base = mmio_base; |
1f166439 GZ |
500 | |
501 | tasklet_init(&vmci_dev->datagram_tasklet, | |
502 | vmci_dispatch_dgs, (unsigned long)vmci_dev); | |
503 | tasklet_init(&vmci_dev->bm_tasklet, | |
504 | vmci_process_bitmap, (unsigned long)vmci_dev); | |
505 | ||
506 | vmci_dev->data_buffer = vmalloc(VMCI_MAX_DG_SIZE); | |
507 | if (!vmci_dev->data_buffer) { | |
508 | dev_err(&pdev->dev, | |
509 | "Can't allocate memory for datagram buffer\n"); | |
510 | return -ENOMEM; | |
511 | } | |
512 | ||
513 | pci_set_master(pdev); /* To enable queue_pair functionality. */ | |
514 | ||
515 | /* | |
516 | * Verify that the VMCI Device supports the capabilities that | |
517 | * we need. If the device is missing capabilities that we would | |
518 | * like to use, check for fallback capabilities and use those | |
519 | * instead (so we can run a new VM on old hosts). Fail the load if | |
520 | * a required capability is missing and there is no fallback. | |
521 | * | |
522 | * Right now, we need datagrams. There are no fallbacks. | |
523 | */ | |
e01153c7 | 524 | capabilities = vmci_read_reg(vmci_dev, VMCI_CAPS_ADDR); |
1f166439 GZ |
525 | if (!(capabilities & VMCI_CAPS_DATAGRAM)) { |
526 | dev_err(&pdev->dev, "Device does not support datagrams\n"); | |
527 | error = -ENXIO; | |
528 | goto err_free_data_buffer; | |
529 | } | |
f2db7361 VD |
530 | caps_in_use = VMCI_CAPS_DATAGRAM; |
531 | ||
532 | /* | |
533 | * Use 64-bit PPNs if the device supports. | |
534 | * | |
535 | * There is no check for the return value of dma_set_mask_and_coherent | |
536 | * since this driver can handle the default mask values if | |
537 | * dma_set_mask_and_coherent fails. | |
538 | */ | |
539 | if (capabilities & VMCI_CAPS_PPN64) { | |
540 | dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64)); | |
541 | use_ppn64 = true; | |
542 | caps_in_use |= VMCI_CAPS_PPN64; | |
543 | } else { | |
544 | dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(44)); | |
545 | use_ppn64 = false; | |
546 | } | |
1f166439 GZ |
547 | |
548 | /* | |
549 | * If the hardware supports notifications, we will use that as | |
550 | * well. | |
551 | */ | |
552 | if (capabilities & VMCI_CAPS_NOTIFICATIONS) { | |
6d6dfb4f AK |
553 | vmci_dev->notification_bitmap = dma_alloc_coherent( |
554 | &pdev->dev, PAGE_SIZE, &vmci_dev->notification_base, | |
555 | GFP_KERNEL); | |
1f166439 GZ |
556 | if (!vmci_dev->notification_bitmap) { |
557 | dev_warn(&pdev->dev, | |
558 | "Unable to allocate notification bitmap\n"); | |
559 | } else { | |
560 | memset(vmci_dev->notification_bitmap, 0, PAGE_SIZE); | |
f2db7361 | 561 | caps_in_use |= VMCI_CAPS_NOTIFICATIONS; |
1f166439 GZ |
562 | } |
563 | } | |
564 | ||
2d5484c1 JH |
565 | if (mmio_base != NULL) { |
566 | if (capabilities & VMCI_CAPS_DMA_DATAGRAM) { | |
567 | caps_in_use |= VMCI_CAPS_DMA_DATAGRAM; | |
568 | } else { | |
569 | dev_err(&pdev->dev, | |
570 | "Missing capability: VMCI_CAPS_DMA_DATAGRAM\n"); | |
571 | error = -ENXIO; | |
572 | goto err_free_data_buffer; | |
573 | } | |
574 | } | |
575 | ||
f2db7361 | 576 | dev_info(&pdev->dev, "Using capabilities 0x%x\n", caps_in_use); |
1f166439 GZ |
577 | |
578 | /* Let the host know which capabilities we intend to use. */ | |
e01153c7 | 579 | vmci_write_reg(vmci_dev, caps_in_use, VMCI_CAPS_ADDR); |
1f166439 | 580 | |
9b17e6ab JH |
581 | /* Let the device know the size for pages passed down. */ |
582 | if (caps_in_use & VMCI_CAPS_DMA_DATAGRAM) | |
583 | vmci_write_reg(vmci_dev, PAGE_SHIFT, VMCI_GUEST_PAGE_SHIFT); | |
584 | ||
1f166439 GZ |
585 | /* Set up global device so that we can start sending datagrams */ |
586 | spin_lock_irq(&vmci_dev_spinlock); | |
587 | vmci_dev_g = vmci_dev; | |
6d6dfb4f | 588 | vmci_pdev = pdev; |
1f166439 GZ |
589 | spin_unlock_irq(&vmci_dev_spinlock); |
590 | ||
591 | /* | |
592 | * Register notification bitmap with device if that capability is | |
593 | * used. | |
594 | */ | |
f2db7361 | 595 | if (caps_in_use & VMCI_CAPS_NOTIFICATIONS) { |
6d6dfb4f AK |
596 | unsigned long bitmap_ppn = |
597 | vmci_dev->notification_base >> PAGE_SHIFT; | |
1f166439 GZ |
598 | if (!vmci_dbell_register_notification_bitmap(bitmap_ppn)) { |
599 | dev_warn(&pdev->dev, | |
f2db7361 VD |
600 | "VMCI device unable to register notification bitmap with PPN 0x%lx\n", |
601 | bitmap_ppn); | |
782f2445 | 602 | error = -ENXIO; |
1f166439 GZ |
603 | goto err_remove_vmci_dev_g; |
604 | } | |
605 | } | |
606 | ||
607 | /* Check host capabilities. */ | |
782f2445 DT |
608 | error = vmci_check_host_caps(pdev); |
609 | if (error) | |
1f166439 GZ |
610 | goto err_remove_bitmap; |
611 | ||
612 | /* Enable device. */ | |
613 | ||
614 | /* | |
615 | * We subscribe to the VMCI_EVENT_CTX_ID_UPDATE here so we can | |
616 | * update the internal context id when needed. | |
617 | */ | |
618 | vmci_err = vmci_event_subscribe(VMCI_EVENT_CTX_ID_UPDATE, | |
619 | vmci_guest_cid_update, NULL, | |
620 | &ctx_update_sub_id); | |
621 | if (vmci_err < VMCI_SUCCESS) | |
622 | dev_warn(&pdev->dev, | |
623 | "Failed to subscribe to event (type=%d): %d\n", | |
624 | VMCI_EVENT_CTX_ID_UPDATE, vmci_err); | |
625 | ||
626 | /* | |
627 | * Enable interrupts. Try MSI-X first, then MSI, and then fallback on | |
628 | * legacy interrupts. | |
629 | */ | |
3bb434cd CH |
630 | error = pci_alloc_irq_vectors(pdev, VMCI_MAX_INTRS, VMCI_MAX_INTRS, |
631 | PCI_IRQ_MSIX); | |
c3423563 | 632 | if (error < 0) { |
3bb434cd CH |
633 | error = pci_alloc_irq_vectors(pdev, 1, 1, |
634 | PCI_IRQ_MSIX | PCI_IRQ_MSI | PCI_IRQ_LEGACY); | |
c3423563 | 635 | if (error < 0) |
3bb434cd | 636 | goto err_remove_bitmap; |
1f166439 | 637 | } else { |
3bb434cd | 638 | vmci_dev->exclusive_vectors = true; |
1f166439 GZ |
639 | } |
640 | ||
641 | /* | |
642 | * Request IRQ for legacy or MSI interrupts, or for first | |
643 | * MSI-X vector. | |
644 | */ | |
3bb434cd CH |
645 | error = request_irq(pci_irq_vector(pdev, 0), vmci_interrupt, |
646 | IRQF_SHARED, KBUILD_MODNAME, vmci_dev); | |
1f166439 GZ |
647 | if (error) { |
648 | dev_err(&pdev->dev, "Irq %u in use: %d\n", | |
3bb434cd | 649 | pci_irq_vector(pdev, 0), error); |
1f166439 GZ |
650 | goto err_disable_msi; |
651 | } | |
652 | ||
653 | /* | |
654 | * For MSI-X with exclusive vectors we need to request an | |
655 | * interrupt for each vector so that we get a separate | |
656 | * interrupt handler routine. This allows us to distinguish | |
657 | * between the vectors. | |
658 | */ | |
659 | if (vmci_dev->exclusive_vectors) { | |
3bb434cd | 660 | error = request_irq(pci_irq_vector(pdev, 1), |
1f166439 GZ |
661 | vmci_interrupt_bm, 0, KBUILD_MODNAME, |
662 | vmci_dev); | |
663 | if (error) { | |
664 | dev_err(&pdev->dev, | |
665 | "Failed to allocate irq %u: %d\n", | |
3bb434cd | 666 | pci_irq_vector(pdev, 1), error); |
1f166439 GZ |
667 | goto err_free_irq; |
668 | } | |
669 | } | |
670 | ||
671 | dev_dbg(&pdev->dev, "Registered device\n"); | |
672 | ||
673 | atomic_inc(&vmci_num_guest_devices); | |
674 | ||
675 | /* Enable specific interrupt bits. */ | |
676 | cmd = VMCI_IMR_DATAGRAM; | |
f2db7361 | 677 | if (caps_in_use & VMCI_CAPS_NOTIFICATIONS) |
1f166439 | 678 | cmd |= VMCI_IMR_NOTIFICATION; |
e01153c7 | 679 | vmci_write_reg(vmci_dev, cmd, VMCI_IMR_ADDR); |
1f166439 GZ |
680 | |
681 | /* Enable interrupts. */ | |
e01153c7 | 682 | vmci_write_reg(vmci_dev, VMCI_CONTROL_INT_ENABLE, VMCI_CONTROL_ADDR); |
1f166439 GZ |
683 | |
684 | pci_set_drvdata(pdev, vmci_dev); | |
b1bba80a SG |
685 | |
686 | vmci_call_vsock_callback(false); | |
1f166439 GZ |
687 | return 0; |
688 | ||
689 | err_free_irq: | |
3bb434cd | 690 | free_irq(pci_irq_vector(pdev, 0), vmci_dev); |
1f166439 GZ |
691 | tasklet_kill(&vmci_dev->datagram_tasklet); |
692 | tasklet_kill(&vmci_dev->bm_tasklet); | |
693 | ||
694 | err_disable_msi: | |
3bb434cd | 695 | pci_free_irq_vectors(pdev); |
1f166439 GZ |
696 | |
697 | vmci_err = vmci_event_unsubscribe(ctx_update_sub_id); | |
698 | if (vmci_err < VMCI_SUCCESS) | |
699 | dev_warn(&pdev->dev, | |
700 | "Failed to unsubscribe from event (type=%d) with subscriber (ID=0x%x): %d\n", | |
701 | VMCI_EVENT_CTX_ID_UPDATE, ctx_update_sub_id, vmci_err); | |
702 | ||
703 | err_remove_bitmap: | |
704 | if (vmci_dev->notification_bitmap) { | |
e01153c7 | 705 | vmci_write_reg(vmci_dev, VMCI_CONTROL_RESET, VMCI_CONTROL_ADDR); |
6d6dfb4f AK |
706 | dma_free_coherent(&pdev->dev, PAGE_SIZE, |
707 | vmci_dev->notification_bitmap, | |
708 | vmci_dev->notification_base); | |
1f166439 GZ |
709 | } |
710 | ||
711 | err_remove_vmci_dev_g: | |
712 | spin_lock_irq(&vmci_dev_spinlock); | |
6d6dfb4f | 713 | vmci_pdev = NULL; |
1f166439 GZ |
714 | vmci_dev_g = NULL; |
715 | spin_unlock_irq(&vmci_dev_spinlock); | |
716 | ||
717 | err_free_data_buffer: | |
718 | vfree(vmci_dev->data_buffer); | |
719 | ||
720 | /* The rest are managed resources and will be freed by PCI core */ | |
721 | return error; | |
722 | } | |
723 | ||
724 | static void vmci_guest_remove_device(struct pci_dev *pdev) | |
725 | { | |
726 | struct vmci_guest_device *vmci_dev = pci_get_drvdata(pdev); | |
727 | int vmci_err; | |
728 | ||
729 | dev_dbg(&pdev->dev, "Removing device\n"); | |
730 | ||
731 | atomic_dec(&vmci_num_guest_devices); | |
732 | ||
733 | vmci_qp_guest_endpoints_exit(); | |
734 | ||
735 | vmci_err = vmci_event_unsubscribe(ctx_update_sub_id); | |
736 | if (vmci_err < VMCI_SUCCESS) | |
737 | dev_warn(&pdev->dev, | |
738 | "Failed to unsubscribe from event (type=%d) with subscriber (ID=0x%x): %d\n", | |
739 | VMCI_EVENT_CTX_ID_UPDATE, ctx_update_sub_id, vmci_err); | |
740 | ||
741 | spin_lock_irq(&vmci_dev_spinlock); | |
742 | vmci_dev_g = NULL; | |
6d6dfb4f | 743 | vmci_pdev = NULL; |
1f166439 GZ |
744 | spin_unlock_irq(&vmci_dev_spinlock); |
745 | ||
746 | dev_dbg(&pdev->dev, "Resetting vmci device\n"); | |
e01153c7 | 747 | vmci_write_reg(vmci_dev, VMCI_CONTROL_RESET, VMCI_CONTROL_ADDR); |
1f166439 GZ |
748 | |
749 | /* | |
750 | * Free IRQ and then disable MSI/MSI-X as appropriate. For | |
751 | * MSI-X, we might have multiple vectors, each with their own | |
752 | * IRQ, which we must free too. | |
753 | */ | |
3bb434cd CH |
754 | if (vmci_dev->exclusive_vectors) |
755 | free_irq(pci_irq_vector(pdev, 1), vmci_dev); | |
756 | free_irq(pci_irq_vector(pdev, 0), vmci_dev); | |
757 | pci_free_irq_vectors(pdev); | |
1f166439 GZ |
758 | |
759 | tasklet_kill(&vmci_dev->datagram_tasklet); | |
760 | tasklet_kill(&vmci_dev->bm_tasklet); | |
761 | ||
762 | if (vmci_dev->notification_bitmap) { | |
763 | /* | |
764 | * The device reset above cleared the bitmap state of the | |
765 | * device, so we can safely free it here. | |
766 | */ | |
767 | ||
6d6dfb4f AK |
768 | dma_free_coherent(&pdev->dev, PAGE_SIZE, |
769 | vmci_dev->notification_bitmap, | |
770 | vmci_dev->notification_base); | |
1f166439 GZ |
771 | } |
772 | ||
773 | vfree(vmci_dev->data_buffer); | |
774 | ||
775 | /* The rest are managed resources and will be freed by PCI core */ | |
776 | } | |
777 | ||
32182cd3 | 778 | static const struct pci_device_id vmci_ids[] = { |
1f166439 GZ |
779 | { PCI_DEVICE(PCI_VENDOR_ID_VMWARE, PCI_DEVICE_ID_VMWARE_VMCI), }, |
780 | { 0 }, | |
781 | }; | |
782 | MODULE_DEVICE_TABLE(pci, vmci_ids); | |
783 | ||
784 | static struct pci_driver vmci_guest_driver = { | |
785 | .name = KBUILD_MODNAME, | |
786 | .id_table = vmci_ids, | |
787 | .probe = vmci_guest_probe_device, | |
788 | .remove = vmci_guest_remove_device, | |
789 | }; | |
790 | ||
791 | int __init vmci_guest_init(void) | |
792 | { | |
793 | return pci_register_driver(&vmci_guest_driver); | |
794 | } | |
795 | ||
796 | void __exit vmci_guest_exit(void) | |
797 | { | |
798 | pci_unregister_driver(&vmci_guest_driver); | |
799 | } |