]>
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 | 15 | #include <linux/module.h> |
1af5c8ae | 16 | #include <linux/processor.h> |
1f166439 | 17 | #include <linux/sched.h> |
ea8a83a4 | 18 | #include <linux/slab.h> |
1f166439 GZ |
19 | #include <linux/init.h> |
20 | #include <linux/pci.h> | |
21 | #include <linux/smp.h> | |
22 | #include <linux/io.h> | |
ea8a83a4 | 23 | #include <linux/vmalloc.h> |
1f166439 GZ |
24 | |
25 | #include "vmci_datagram.h" | |
26 | #include "vmci_doorbell.h" | |
27 | #include "vmci_context.h" | |
28 | #include "vmci_driver.h" | |
29 | #include "vmci_event.h" | |
30 | ||
1f166439 GZ |
31 | #define PCI_DEVICE_ID_VMWARE_VMCI 0x0740 |
32 | ||
33 | #define VMCI_UTIL_NUM_RESOURCES 1 | |
34 | ||
8389c29a JH |
35 | /* |
36 | * Datagram buffers for DMA send/receive must accommodate at least | |
37 | * a maximum sized datagram and the header. | |
38 | */ | |
39 | #define VMCI_DMA_DG_BUFFER_SIZE (VMCI_MAX_DG_SIZE + PAGE_SIZE) | |
40 | ||
1f166439 GZ |
41 | static bool vmci_disable_msi; |
42 | module_param_named(disable_msi, vmci_disable_msi, bool, 0); | |
43 | MODULE_PARM_DESC(disable_msi, "Disable MSI use in driver - (default=0)"); | |
44 | ||
45 | static bool vmci_disable_msix; | |
46 | module_param_named(disable_msix, vmci_disable_msix, bool, 0); | |
47 | MODULE_PARM_DESC(disable_msix, "Disable MSI-X use in driver - (default=0)"); | |
48 | ||
49 | static u32 ctx_update_sub_id = VMCI_INVALID_ID; | |
50 | static u32 vm_context_id = VMCI_INVALID_ID; | |
51 | ||
52 | struct vmci_guest_device { | |
53 | struct device *dev; /* PCI device we are attached to */ | |
54 | void __iomem *iobase; | |
e01153c7 | 55 | void __iomem *mmio_base; |
1f166439 | 56 | |
1f166439 | 57 | bool exclusive_vectors; |
1f166439 GZ |
58 | |
59 | struct tasklet_struct datagram_tasklet; | |
60 | struct tasklet_struct bm_tasklet; | |
61 | ||
62 | void *data_buffer; | |
8389c29a JH |
63 | dma_addr_t data_buffer_base; |
64 | void *tx_buffer; | |
65 | dma_addr_t tx_buffer_base; | |
1f166439 | 66 | void *notification_bitmap; |
6d6dfb4f | 67 | dma_addr_t notification_base; |
1f166439 GZ |
68 | }; |
69 | ||
f2db7361 VD |
70 | static bool use_ppn64; |
71 | ||
72 | bool vmci_use_ppn64(void) | |
73 | { | |
74 | return use_ppn64; | |
75 | } | |
76 | ||
1f166439 | 77 | /* vmci_dev singleton device and supporting data*/ |
6d6dfb4f | 78 | struct pci_dev *vmci_pdev; |
1f166439 GZ |
79 | static struct vmci_guest_device *vmci_dev_g; |
80 | static DEFINE_SPINLOCK(vmci_dev_spinlock); | |
81 | ||
82 | static atomic_t vmci_num_guest_devices = ATOMIC_INIT(0); | |
83 | ||
84 | bool vmci_guest_code_active(void) | |
85 | { | |
86 | return atomic_read(&vmci_num_guest_devices) != 0; | |
87 | } | |
88 | ||
89 | u32 vmci_get_vm_context_id(void) | |
90 | { | |
91 | if (vm_context_id == VMCI_INVALID_ID) { | |
1f166439 GZ |
92 | struct vmci_datagram get_cid_msg; |
93 | get_cid_msg.dst = | |
94 | vmci_make_handle(VMCI_HYPERVISOR_CONTEXT_ID, | |
95 | VMCI_GET_CONTEXT_ID); | |
96 | get_cid_msg.src = VMCI_ANON_SRC_HANDLE; | |
97 | get_cid_msg.payload_size = 0; | |
5a19b789 | 98 | vm_context_id = vmci_send_datagram(&get_cid_msg); |
1f166439 GZ |
99 | } |
100 | return vm_context_id; | |
101 | } | |
102 | ||
e01153c7 JH |
103 | static unsigned int vmci_read_reg(struct vmci_guest_device *dev, u32 reg) |
104 | { | |
105 | if (dev->mmio_base != NULL) | |
106 | return readl(dev->mmio_base + reg); | |
107 | return ioread32(dev->iobase + reg); | |
108 | } | |
109 | ||
110 | static void vmci_write_reg(struct vmci_guest_device *dev, u32 val, u32 reg) | |
111 | { | |
112 | if (dev->mmio_base != NULL) | |
113 | writel(val, dev->mmio_base + reg); | |
114 | else | |
115 | iowrite32(val, dev->iobase + reg); | |
116 | } | |
117 | ||
1af5c8ae JH |
118 | static int vmci_write_data(struct vmci_guest_device *dev, |
119 | struct vmci_datagram *dg) | |
120 | { | |
121 | int result; | |
122 | ||
123 | if (dev->mmio_base != NULL) { | |
124 | struct vmci_data_in_out_header *buffer_header = dev->tx_buffer; | |
125 | u8 *dg_out_buffer = (u8 *)(buffer_header + 1); | |
126 | ||
127 | if (VMCI_DG_SIZE(dg) > VMCI_MAX_DG_SIZE) | |
128 | return VMCI_ERROR_INVALID_ARGS; | |
129 | ||
130 | /* | |
131 | * Initialize send buffer with outgoing datagram | |
132 | * and set up header for inline data. Device will | |
133 | * not access buffer asynchronously - only after | |
134 | * the write to VMCI_DATA_OUT_LOW_ADDR. | |
135 | */ | |
136 | memcpy(dg_out_buffer, dg, VMCI_DG_SIZE(dg)); | |
137 | buffer_header->opcode = 0; | |
138 | buffer_header->size = VMCI_DG_SIZE(dg); | |
139 | buffer_header->busy = 1; | |
140 | ||
141 | vmci_write_reg(dev, lower_32_bits(dev->tx_buffer_base), | |
142 | VMCI_DATA_OUT_LOW_ADDR); | |
143 | ||
144 | /* Caller holds a spinlock, so cannot block. */ | |
145 | spin_until_cond(buffer_header->busy == 0); | |
146 | ||
147 | result = vmci_read_reg(vmci_dev_g, VMCI_RESULT_LOW_ADDR); | |
148 | if (result == VMCI_SUCCESS) | |
149 | result = (int)buffer_header->result; | |
150 | } else { | |
151 | iowrite8_rep(dev->iobase + VMCI_DATA_OUT_ADDR, | |
152 | dg, VMCI_DG_SIZE(dg)); | |
153 | result = vmci_read_reg(vmci_dev_g, VMCI_RESULT_LOW_ADDR); | |
154 | } | |
155 | ||
156 | return result; | |
157 | } | |
158 | ||
1f166439 GZ |
159 | /* |
160 | * VM to hypervisor call mechanism. We use the standard VMware naming | |
161 | * convention since shared code is calling this function as well. | |
162 | */ | |
163 | int vmci_send_datagram(struct vmci_datagram *dg) | |
164 | { | |
165 | unsigned long flags; | |
166 | int result; | |
167 | ||
168 | /* Check args. */ | |
169 | if (dg == NULL) | |
170 | return VMCI_ERROR_INVALID_ARGS; | |
171 | ||
172 | /* | |
173 | * Need to acquire spinlock on the device because the datagram | |
174 | * data may be spread over multiple pages and the monitor may | |
175 | * interleave device user rpc calls from multiple | |
176 | * VCPUs. Acquiring the spinlock precludes that | |
177 | * possibility. Disabling interrupts to avoid incoming | |
178 | * datagrams during a "rep out" and possibly landing up in | |
179 | * this function. | |
180 | */ | |
181 | spin_lock_irqsave(&vmci_dev_spinlock, flags); | |
182 | ||
183 | if (vmci_dev_g) { | |
1af5c8ae | 184 | vmci_write_data(vmci_dev_g, dg); |
e01153c7 | 185 | result = vmci_read_reg(vmci_dev_g, VMCI_RESULT_LOW_ADDR); |
1f166439 GZ |
186 | } else { |
187 | result = VMCI_ERROR_UNAVAILABLE; | |
188 | } | |
189 | ||
190 | spin_unlock_irqrestore(&vmci_dev_spinlock, flags); | |
191 | ||
192 | return result; | |
193 | } | |
194 | EXPORT_SYMBOL_GPL(vmci_send_datagram); | |
195 | ||
196 | /* | |
197 | * Gets called with the new context id if updated or resumed. | |
198 | * Context id. | |
199 | */ | |
200 | static void vmci_guest_cid_update(u32 sub_id, | |
201 | const struct vmci_event_data *event_data, | |
202 | void *client_data) | |
203 | { | |
204 | const struct vmci_event_payld_ctx *ev_payload = | |
205 | vmci_event_data_const_payload(event_data); | |
206 | ||
207 | if (sub_id != ctx_update_sub_id) { | |
208 | pr_devel("Invalid subscriber (ID=0x%x)\n", sub_id); | |
209 | return; | |
210 | } | |
211 | ||
212 | if (!event_data || ev_payload->context_id == VMCI_INVALID_ID) { | |
213 | pr_devel("Invalid event data\n"); | |
214 | return; | |
215 | } | |
216 | ||
217 | pr_devel("Updating context from (ID=0x%x) to (ID=0x%x) on event (type=%d)\n", | |
218 | vm_context_id, ev_payload->context_id, event_data->event); | |
219 | ||
220 | vm_context_id = ev_payload->context_id; | |
221 | } | |
222 | ||
223 | /* | |
224 | * Verify that the host supports the hypercalls we need. If it does not, | |
225 | * try to find fallback hypercalls and use those instead. Returns | |
226 | * true if required hypercalls (or fallback hypercalls) are | |
227 | * supported by the host, false otherwise. | |
228 | */ | |
782f2445 | 229 | static int vmci_check_host_caps(struct pci_dev *pdev) |
1f166439 GZ |
230 | { |
231 | bool result; | |
232 | struct vmci_resource_query_msg *msg; | |
233 | u32 msg_size = sizeof(struct vmci_resource_query_hdr) + | |
234 | VMCI_UTIL_NUM_RESOURCES * sizeof(u32); | |
235 | struct vmci_datagram *check_msg; | |
236 | ||
b2192cfe | 237 | check_msg = kzalloc(msg_size, GFP_KERNEL); |
1f166439 GZ |
238 | if (!check_msg) { |
239 | dev_err(&pdev->dev, "%s: Insufficient memory\n", __func__); | |
782f2445 | 240 | return -ENOMEM; |
1f166439 GZ |
241 | } |
242 | ||
243 | check_msg->dst = vmci_make_handle(VMCI_HYPERVISOR_CONTEXT_ID, | |
244 | VMCI_RESOURCES_QUERY); | |
245 | check_msg->src = VMCI_ANON_SRC_HANDLE; | |
246 | check_msg->payload_size = msg_size - VMCI_DG_HEADERSIZE; | |
247 | msg = (struct vmci_resource_query_msg *)VMCI_DG_PAYLOAD(check_msg); | |
248 | ||
249 | msg->num_resources = VMCI_UTIL_NUM_RESOURCES; | |
250 | msg->resources[0] = VMCI_GET_CONTEXT_ID; | |
251 | ||
252 | /* Checks that hyper calls are supported */ | |
253 | result = vmci_send_datagram(check_msg) == 0x01; | |
254 | kfree(check_msg); | |
255 | ||
256 | dev_dbg(&pdev->dev, "%s: Host capability check: %s\n", | |
257 | __func__, result ? "PASSED" : "FAILED"); | |
258 | ||
259 | /* We need the vector. There are no fallbacks. */ | |
782f2445 | 260 | return result ? 0 : -ENXIO; |
1f166439 GZ |
261 | } |
262 | ||
263 | /* | |
264 | * Reads datagrams from the data in port and dispatches them. We | |
265 | * always start reading datagrams into only the first page of the | |
266 | * datagram buffer. If the datagrams don't fit into one page, we | |
267 | * use the maximum datagram buffer size for the remainder of the | |
268 | * invocation. This is a simple heuristic for not penalizing | |
269 | * small datagrams. | |
270 | * | |
271 | * This function assumes that it has exclusive access to the data | |
272 | * in port for the duration of the call. | |
273 | */ | |
274 | static void vmci_dispatch_dgs(unsigned long data) | |
275 | { | |
276 | struct vmci_guest_device *vmci_dev = (struct vmci_guest_device *)data; | |
277 | u8 *dg_in_buffer = vmci_dev->data_buffer; | |
278 | struct vmci_datagram *dg; | |
279 | size_t dg_in_buffer_size = VMCI_MAX_DG_SIZE; | |
280 | size_t current_dg_in_buffer_size = PAGE_SIZE; | |
281 | size_t remaining_bytes; | |
282 | ||
283 | BUILD_BUG_ON(VMCI_MAX_DG_SIZE < PAGE_SIZE); | |
284 | ||
285 | ioread8_rep(vmci_dev->iobase + VMCI_DATA_IN_ADDR, | |
286 | vmci_dev->data_buffer, current_dg_in_buffer_size); | |
287 | dg = (struct vmci_datagram *)dg_in_buffer; | |
288 | remaining_bytes = current_dg_in_buffer_size; | |
289 | ||
290 | while (dg->dst.resource != VMCI_INVALID_ID || | |
291 | remaining_bytes > PAGE_SIZE) { | |
292 | unsigned dg_in_size; | |
293 | ||
294 | /* | |
295 | * When the input buffer spans multiple pages, a datagram can | |
296 | * start on any page boundary in the buffer. | |
297 | */ | |
298 | if (dg->dst.resource == VMCI_INVALID_ID) { | |
299 | dg = (struct vmci_datagram *)roundup( | |
300 | (uintptr_t)dg + 1, PAGE_SIZE); | |
301 | remaining_bytes = | |
302 | (size_t)(dg_in_buffer + | |
303 | current_dg_in_buffer_size - | |
304 | (u8 *)dg); | |
305 | continue; | |
306 | } | |
307 | ||
308 | dg_in_size = VMCI_DG_SIZE_ALIGNED(dg); | |
309 | ||
310 | if (dg_in_size <= dg_in_buffer_size) { | |
311 | int result; | |
312 | ||
313 | /* | |
314 | * If the remaining bytes in the datagram | |
315 | * buffer doesn't contain the complete | |
316 | * datagram, we first make sure we have enough | |
317 | * room for it and then we read the reminder | |
318 | * of the datagram and possibly any following | |
319 | * datagrams. | |
320 | */ | |
321 | if (dg_in_size > remaining_bytes) { | |
322 | if (remaining_bytes != | |
323 | current_dg_in_buffer_size) { | |
324 | ||
325 | /* | |
326 | * We move the partial | |
327 | * datagram to the front and | |
328 | * read the reminder of the | |
329 | * datagram and possibly | |
330 | * following calls into the | |
331 | * following bytes. | |
332 | */ | |
333 | memmove(dg_in_buffer, dg_in_buffer + | |
334 | current_dg_in_buffer_size - | |
335 | remaining_bytes, | |
336 | remaining_bytes); | |
337 | dg = (struct vmci_datagram *) | |
338 | dg_in_buffer; | |
339 | } | |
340 | ||
341 | if (current_dg_in_buffer_size != | |
342 | dg_in_buffer_size) | |
343 | current_dg_in_buffer_size = | |
344 | dg_in_buffer_size; | |
345 | ||
346 | ioread8_rep(vmci_dev->iobase + | |
347 | VMCI_DATA_IN_ADDR, | |
348 | vmci_dev->data_buffer + | |
349 | remaining_bytes, | |
350 | current_dg_in_buffer_size - | |
351 | remaining_bytes); | |
352 | } | |
353 | ||
354 | /* | |
355 | * We special case event datagrams from the | |
356 | * hypervisor. | |
357 | */ | |
358 | if (dg->src.context == VMCI_HYPERVISOR_CONTEXT_ID && | |
359 | dg->dst.resource == VMCI_EVENT_HANDLER) { | |
360 | result = vmci_event_dispatch(dg); | |
361 | } else { | |
362 | result = vmci_datagram_invoke_guest_handler(dg); | |
363 | } | |
364 | if (result < VMCI_SUCCESS) | |
365 | dev_dbg(vmci_dev->dev, | |
366 | "Datagram with resource (ID=0x%x) failed (err=%d)\n", | |
367 | dg->dst.resource, result); | |
368 | ||
369 | /* On to the next datagram. */ | |
370 | dg = (struct vmci_datagram *)((u8 *)dg + | |
371 | dg_in_size); | |
372 | } else { | |
373 | size_t bytes_to_skip; | |
374 | ||
375 | /* | |
376 | * Datagram doesn't fit in datagram buffer of maximal | |
377 | * size. We drop it. | |
378 | */ | |
379 | dev_dbg(vmci_dev->dev, | |
380 | "Failed to receive datagram (size=%u bytes)\n", | |
381 | dg_in_size); | |
382 | ||
383 | bytes_to_skip = dg_in_size - remaining_bytes; | |
384 | if (current_dg_in_buffer_size != dg_in_buffer_size) | |
385 | current_dg_in_buffer_size = dg_in_buffer_size; | |
386 | ||
387 | for (;;) { | |
388 | ioread8_rep(vmci_dev->iobase + | |
389 | VMCI_DATA_IN_ADDR, | |
390 | vmci_dev->data_buffer, | |
391 | current_dg_in_buffer_size); | |
392 | if (bytes_to_skip <= current_dg_in_buffer_size) | |
393 | break; | |
394 | ||
395 | bytes_to_skip -= current_dg_in_buffer_size; | |
396 | } | |
397 | dg = (struct vmci_datagram *)(dg_in_buffer + | |
398 | bytes_to_skip); | |
399 | } | |
400 | ||
401 | remaining_bytes = | |
402 | (size_t) (dg_in_buffer + current_dg_in_buffer_size - | |
403 | (u8 *)dg); | |
404 | ||
405 | if (remaining_bytes < VMCI_DG_HEADERSIZE) { | |
406 | /* Get the next batch of datagrams. */ | |
407 | ||
408 | ioread8_rep(vmci_dev->iobase + VMCI_DATA_IN_ADDR, | |
409 | vmci_dev->data_buffer, | |
410 | current_dg_in_buffer_size); | |
411 | dg = (struct vmci_datagram *)dg_in_buffer; | |
412 | remaining_bytes = current_dg_in_buffer_size; | |
413 | } | |
414 | } | |
415 | } | |
416 | ||
417 | /* | |
418 | * Scans the notification bitmap for raised flags, clears them | |
419 | * and handles the notifications. | |
420 | */ | |
421 | static void vmci_process_bitmap(unsigned long data) | |
422 | { | |
423 | struct vmci_guest_device *dev = (struct vmci_guest_device *)data; | |
424 | ||
425 | if (!dev->notification_bitmap) { | |
426 | dev_dbg(dev->dev, "No bitmap present in %s\n", __func__); | |
427 | return; | |
428 | } | |
429 | ||
430 | vmci_dbell_scan_notification_entries(dev->notification_bitmap); | |
431 | } | |
432 | ||
1f166439 GZ |
433 | /* |
434 | * Interrupt handler for legacy or MSI interrupt, or for first MSI-X | |
435 | * interrupt (vector VMCI_INTR_DATAGRAM). | |
436 | */ | |
437 | static irqreturn_t vmci_interrupt(int irq, void *_dev) | |
438 | { | |
439 | struct vmci_guest_device *dev = _dev; | |
440 | ||
441 | /* | |
442 | * If we are using MSI-X with exclusive vectors then we simply schedule | |
443 | * the datagram tasklet, since we know the interrupt was meant for us. | |
444 | * Otherwise we must read the ICR to determine what to do. | |
445 | */ | |
446 | ||
3bb434cd | 447 | if (dev->exclusive_vectors) { |
1f166439 GZ |
448 | tasklet_schedule(&dev->datagram_tasklet); |
449 | } else { | |
450 | unsigned int icr; | |
451 | ||
452 | /* Acknowledge interrupt and determine what needs doing. */ | |
e01153c7 | 453 | icr = vmci_read_reg(dev, VMCI_ICR_ADDR); |
1f166439 GZ |
454 | if (icr == 0 || icr == ~0) |
455 | return IRQ_NONE; | |
456 | ||
457 | if (icr & VMCI_ICR_DATAGRAM) { | |
458 | tasklet_schedule(&dev->datagram_tasklet); | |
459 | icr &= ~VMCI_ICR_DATAGRAM; | |
460 | } | |
461 | ||
462 | if (icr & VMCI_ICR_NOTIFICATION) { | |
463 | tasklet_schedule(&dev->bm_tasklet); | |
464 | icr &= ~VMCI_ICR_NOTIFICATION; | |
465 | } | |
466 | ||
3eed5372 JH |
467 | if (icr & VMCI_ICR_DMA_DATAGRAM) |
468 | icr &= ~VMCI_ICR_DMA_DATAGRAM; | |
469 | ||
1f166439 GZ |
470 | if (icr != 0) |
471 | dev_warn(dev->dev, | |
472 | "Ignoring unknown interrupt cause (%d)\n", | |
473 | icr); | |
474 | } | |
475 | ||
476 | return IRQ_HANDLED; | |
477 | } | |
478 | ||
479 | /* | |
480 | * Interrupt handler for MSI-X interrupt vector VMCI_INTR_NOTIFICATION, | |
481 | * which is for the notification bitmap. Will only get called if we are | |
482 | * using MSI-X with exclusive vectors. | |
483 | */ | |
484 | static irqreturn_t vmci_interrupt_bm(int irq, void *_dev) | |
485 | { | |
486 | struct vmci_guest_device *dev = _dev; | |
487 | ||
488 | /* For MSI-X we can just assume it was meant for us. */ | |
489 | tasklet_schedule(&dev->bm_tasklet); | |
490 | ||
491 | return IRQ_HANDLED; | |
492 | } | |
493 | ||
3eed5372 JH |
494 | /* |
495 | * Interrupt handler for MSI-X interrupt vector VMCI_INTR_DMA_DATAGRAM, | |
496 | * which is for the completion of a DMA datagram send or receive operation. | |
497 | * Will only get called if we are using MSI-X with exclusive vectors. | |
498 | */ | |
499 | static irqreturn_t vmci_interrupt_dma_datagram(int irq, void *_dev) | |
500 | { | |
501 | return IRQ_HANDLED; | |
502 | } | |
503 | ||
8389c29a JH |
504 | static void vmci_free_dg_buffers(struct vmci_guest_device *vmci_dev) |
505 | { | |
506 | if (vmci_dev->mmio_base != NULL) { | |
507 | if (vmci_dev->tx_buffer != NULL) | |
508 | dma_free_coherent(vmci_dev->dev, | |
509 | VMCI_DMA_DG_BUFFER_SIZE, | |
510 | vmci_dev->tx_buffer, | |
511 | vmci_dev->tx_buffer_base); | |
512 | if (vmci_dev->data_buffer != NULL) | |
513 | dma_free_coherent(vmci_dev->dev, | |
514 | VMCI_DMA_DG_BUFFER_SIZE, | |
515 | vmci_dev->data_buffer, | |
516 | vmci_dev->data_buffer_base); | |
517 | } else { | |
518 | vfree(vmci_dev->data_buffer); | |
519 | } | |
520 | } | |
521 | ||
1f166439 GZ |
522 | /* |
523 | * Most of the initialization at module load time is done here. | |
524 | */ | |
525 | static int vmci_guest_probe_device(struct pci_dev *pdev, | |
526 | const struct pci_device_id *id) | |
527 | { | |
528 | struct vmci_guest_device *vmci_dev; | |
e01153c7 JH |
529 | void __iomem *iobase = NULL; |
530 | void __iomem *mmio_base = NULL; | |
3eed5372 | 531 | unsigned int num_irq_vectors; |
1f166439 | 532 | unsigned int capabilities; |
f2db7361 | 533 | unsigned int caps_in_use; |
1f166439 GZ |
534 | unsigned long cmd; |
535 | int vmci_err; | |
536 | int error; | |
537 | ||
538 | dev_dbg(&pdev->dev, "Probing for vmci/PCI guest device\n"); | |
539 | ||
540 | error = pcim_enable_device(pdev); | |
541 | if (error) { | |
542 | dev_err(&pdev->dev, | |
543 | "Failed to enable VMCI device: %d\n", error); | |
544 | return error; | |
545 | } | |
546 | ||
e01153c7 JH |
547 | /* |
548 | * The VMCI device with mmio access to registers requests 256KB | |
549 | * for BAR1. If present, driver will use new VMCI device | |
550 | * functionality for register access and datagram send/recv. | |
551 | */ | |
1f166439 | 552 | |
e01153c7 JH |
553 | if (pci_resource_len(pdev, 1) == VMCI_WITH_MMIO_ACCESS_BAR_SIZE) { |
554 | dev_info(&pdev->dev, "MMIO register access is available\n"); | |
555 | mmio_base = pci_iomap_range(pdev, 1, VMCI_MMIO_ACCESS_OFFSET, | |
556 | VMCI_MMIO_ACCESS_SIZE); | |
557 | /* If the map fails, we fall back to IOIO access. */ | |
558 | if (!mmio_base) | |
559 | dev_warn(&pdev->dev, "Failed to map MMIO register access\n"); | |
560 | } | |
1f166439 | 561 | |
e01153c7 JH |
562 | if (!mmio_base) { |
563 | error = pcim_iomap_regions(pdev, BIT(0), KBUILD_MODNAME); | |
564 | if (error) { | |
565 | dev_err(&pdev->dev, "Failed to reserve/map IO regions\n"); | |
566 | return error; | |
567 | } | |
568 | iobase = pcim_iomap_table(pdev)[0]; | |
569 | } | |
1f166439 GZ |
570 | |
571 | vmci_dev = devm_kzalloc(&pdev->dev, sizeof(*vmci_dev), GFP_KERNEL); | |
572 | if (!vmci_dev) { | |
573 | dev_err(&pdev->dev, | |
574 | "Can't allocate memory for VMCI device\n"); | |
575 | return -ENOMEM; | |
576 | } | |
577 | ||
578 | vmci_dev->dev = &pdev->dev; | |
1f166439 GZ |
579 | vmci_dev->exclusive_vectors = false; |
580 | vmci_dev->iobase = iobase; | |
e01153c7 | 581 | vmci_dev->mmio_base = mmio_base; |
1f166439 GZ |
582 | |
583 | tasklet_init(&vmci_dev->datagram_tasklet, | |
584 | vmci_dispatch_dgs, (unsigned long)vmci_dev); | |
585 | tasklet_init(&vmci_dev->bm_tasklet, | |
586 | vmci_process_bitmap, (unsigned long)vmci_dev); | |
587 | ||
8389c29a JH |
588 | if (mmio_base != NULL) { |
589 | vmci_dev->tx_buffer = dma_alloc_coherent(&pdev->dev, VMCI_DMA_DG_BUFFER_SIZE, | |
590 | &vmci_dev->tx_buffer_base, | |
591 | GFP_KERNEL); | |
592 | if (!vmci_dev->tx_buffer) { | |
593 | dev_err(&pdev->dev, | |
594 | "Can't allocate memory for datagram tx buffer\n"); | |
595 | return -ENOMEM; | |
596 | } | |
597 | ||
598 | vmci_dev->data_buffer = dma_alloc_coherent(&pdev->dev, VMCI_DMA_DG_BUFFER_SIZE, | |
599 | &vmci_dev->data_buffer_base, | |
600 | GFP_KERNEL); | |
601 | } else { | |
602 | vmci_dev->data_buffer = vmalloc(VMCI_MAX_DG_SIZE); | |
603 | } | |
1f166439 GZ |
604 | if (!vmci_dev->data_buffer) { |
605 | dev_err(&pdev->dev, | |
606 | "Can't allocate memory for datagram buffer\n"); | |
8389c29a JH |
607 | error = -ENOMEM; |
608 | goto err_free_data_buffers; | |
1f166439 GZ |
609 | } |
610 | ||
611 | pci_set_master(pdev); /* To enable queue_pair functionality. */ | |
612 | ||
613 | /* | |
614 | * Verify that the VMCI Device supports the capabilities that | |
615 | * we need. If the device is missing capabilities that we would | |
616 | * like to use, check for fallback capabilities and use those | |
617 | * instead (so we can run a new VM on old hosts). Fail the load if | |
618 | * a required capability is missing and there is no fallback. | |
619 | * | |
620 | * Right now, we need datagrams. There are no fallbacks. | |
621 | */ | |
e01153c7 | 622 | capabilities = vmci_read_reg(vmci_dev, VMCI_CAPS_ADDR); |
1f166439 GZ |
623 | if (!(capabilities & VMCI_CAPS_DATAGRAM)) { |
624 | dev_err(&pdev->dev, "Device does not support datagrams\n"); | |
625 | error = -ENXIO; | |
8389c29a | 626 | goto err_free_data_buffers; |
1f166439 | 627 | } |
f2db7361 VD |
628 | caps_in_use = VMCI_CAPS_DATAGRAM; |
629 | ||
630 | /* | |
631 | * Use 64-bit PPNs if the device supports. | |
632 | * | |
633 | * There is no check for the return value of dma_set_mask_and_coherent | |
634 | * since this driver can handle the default mask values if | |
635 | * dma_set_mask_and_coherent fails. | |
636 | */ | |
637 | if (capabilities & VMCI_CAPS_PPN64) { | |
638 | dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64)); | |
639 | use_ppn64 = true; | |
640 | caps_in_use |= VMCI_CAPS_PPN64; | |
641 | } else { | |
642 | dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(44)); | |
643 | use_ppn64 = false; | |
644 | } | |
1f166439 GZ |
645 | |
646 | /* | |
647 | * If the hardware supports notifications, we will use that as | |
648 | * well. | |
649 | */ | |
650 | if (capabilities & VMCI_CAPS_NOTIFICATIONS) { | |
6d6dfb4f AK |
651 | vmci_dev->notification_bitmap = dma_alloc_coherent( |
652 | &pdev->dev, PAGE_SIZE, &vmci_dev->notification_base, | |
653 | GFP_KERNEL); | |
1f166439 GZ |
654 | if (!vmci_dev->notification_bitmap) { |
655 | dev_warn(&pdev->dev, | |
656 | "Unable to allocate notification bitmap\n"); | |
657 | } else { | |
658 | memset(vmci_dev->notification_bitmap, 0, PAGE_SIZE); | |
f2db7361 | 659 | caps_in_use |= VMCI_CAPS_NOTIFICATIONS; |
1f166439 GZ |
660 | } |
661 | } | |
662 | ||
2d5484c1 JH |
663 | if (mmio_base != NULL) { |
664 | if (capabilities & VMCI_CAPS_DMA_DATAGRAM) { | |
665 | caps_in_use |= VMCI_CAPS_DMA_DATAGRAM; | |
666 | } else { | |
667 | dev_err(&pdev->dev, | |
668 | "Missing capability: VMCI_CAPS_DMA_DATAGRAM\n"); | |
669 | error = -ENXIO; | |
8389c29a | 670 | goto err_free_data_buffers; |
2d5484c1 JH |
671 | } |
672 | } | |
673 | ||
f2db7361 | 674 | dev_info(&pdev->dev, "Using capabilities 0x%x\n", caps_in_use); |
1f166439 GZ |
675 | |
676 | /* Let the host know which capabilities we intend to use. */ | |
e01153c7 | 677 | vmci_write_reg(vmci_dev, caps_in_use, VMCI_CAPS_ADDR); |
1f166439 | 678 | |
8389c29a JH |
679 | if (caps_in_use & VMCI_CAPS_DMA_DATAGRAM) { |
680 | /* Let the device know the size for pages passed down. */ | |
9b17e6ab JH |
681 | vmci_write_reg(vmci_dev, PAGE_SHIFT, VMCI_GUEST_PAGE_SHIFT); |
682 | ||
8389c29a JH |
683 | /* Configure the high order parts of the data in/out buffers. */ |
684 | vmci_write_reg(vmci_dev, upper_32_bits(vmci_dev->data_buffer_base), | |
685 | VMCI_DATA_IN_HIGH_ADDR); | |
686 | vmci_write_reg(vmci_dev, upper_32_bits(vmci_dev->tx_buffer_base), | |
687 | VMCI_DATA_OUT_HIGH_ADDR); | |
688 | } | |
689 | ||
1f166439 GZ |
690 | /* Set up global device so that we can start sending datagrams */ |
691 | spin_lock_irq(&vmci_dev_spinlock); | |
692 | vmci_dev_g = vmci_dev; | |
6d6dfb4f | 693 | vmci_pdev = pdev; |
1f166439 GZ |
694 | spin_unlock_irq(&vmci_dev_spinlock); |
695 | ||
696 | /* | |
697 | * Register notification bitmap with device if that capability is | |
698 | * used. | |
699 | */ | |
f2db7361 | 700 | if (caps_in_use & VMCI_CAPS_NOTIFICATIONS) { |
6d6dfb4f AK |
701 | unsigned long bitmap_ppn = |
702 | vmci_dev->notification_base >> PAGE_SHIFT; | |
1f166439 GZ |
703 | if (!vmci_dbell_register_notification_bitmap(bitmap_ppn)) { |
704 | dev_warn(&pdev->dev, | |
f2db7361 VD |
705 | "VMCI device unable to register notification bitmap with PPN 0x%lx\n", |
706 | bitmap_ppn); | |
782f2445 | 707 | error = -ENXIO; |
1f166439 GZ |
708 | goto err_remove_vmci_dev_g; |
709 | } | |
710 | } | |
711 | ||
712 | /* Check host capabilities. */ | |
782f2445 DT |
713 | error = vmci_check_host_caps(pdev); |
714 | if (error) | |
1f166439 GZ |
715 | goto err_remove_bitmap; |
716 | ||
717 | /* Enable device. */ | |
718 | ||
719 | /* | |
720 | * We subscribe to the VMCI_EVENT_CTX_ID_UPDATE here so we can | |
721 | * update the internal context id when needed. | |
722 | */ | |
723 | vmci_err = vmci_event_subscribe(VMCI_EVENT_CTX_ID_UPDATE, | |
724 | vmci_guest_cid_update, NULL, | |
725 | &ctx_update_sub_id); | |
726 | if (vmci_err < VMCI_SUCCESS) | |
727 | dev_warn(&pdev->dev, | |
728 | "Failed to subscribe to event (type=%d): %d\n", | |
729 | VMCI_EVENT_CTX_ID_UPDATE, vmci_err); | |
730 | ||
731 | /* | |
732 | * Enable interrupts. Try MSI-X first, then MSI, and then fallback on | |
733 | * legacy interrupts. | |
734 | */ | |
3eed5372 JH |
735 | if (vmci_dev->mmio_base != NULL) |
736 | num_irq_vectors = VMCI_MAX_INTRS; | |
737 | else | |
738 | num_irq_vectors = VMCI_MAX_INTRS_NOTIFICATION; | |
739 | error = pci_alloc_irq_vectors(pdev, num_irq_vectors, num_irq_vectors, | |
740 | PCI_IRQ_MSIX); | |
c3423563 | 741 | if (error < 0) { |
3bb434cd CH |
742 | error = pci_alloc_irq_vectors(pdev, 1, 1, |
743 | PCI_IRQ_MSIX | PCI_IRQ_MSI | PCI_IRQ_LEGACY); | |
c3423563 | 744 | if (error < 0) |
3bb434cd | 745 | goto err_remove_bitmap; |
1f166439 | 746 | } else { |
3bb434cd | 747 | vmci_dev->exclusive_vectors = true; |
1f166439 GZ |
748 | } |
749 | ||
750 | /* | |
751 | * Request IRQ for legacy or MSI interrupts, or for first | |
752 | * MSI-X vector. | |
753 | */ | |
3bb434cd CH |
754 | error = request_irq(pci_irq_vector(pdev, 0), vmci_interrupt, |
755 | IRQF_SHARED, KBUILD_MODNAME, vmci_dev); | |
1f166439 GZ |
756 | if (error) { |
757 | dev_err(&pdev->dev, "Irq %u in use: %d\n", | |
3bb434cd | 758 | pci_irq_vector(pdev, 0), error); |
1f166439 GZ |
759 | goto err_disable_msi; |
760 | } | |
761 | ||
762 | /* | |
763 | * For MSI-X with exclusive vectors we need to request an | |
764 | * interrupt for each vector so that we get a separate | |
765 | * interrupt handler routine. This allows us to distinguish | |
766 | * between the vectors. | |
767 | */ | |
768 | if (vmci_dev->exclusive_vectors) { | |
3bb434cd | 769 | error = request_irq(pci_irq_vector(pdev, 1), |
1f166439 GZ |
770 | vmci_interrupt_bm, 0, KBUILD_MODNAME, |
771 | vmci_dev); | |
772 | if (error) { | |
773 | dev_err(&pdev->dev, | |
774 | "Failed to allocate irq %u: %d\n", | |
3bb434cd | 775 | pci_irq_vector(pdev, 1), error); |
1f166439 GZ |
776 | goto err_free_irq; |
777 | } | |
3eed5372 JH |
778 | if (caps_in_use & VMCI_CAPS_DMA_DATAGRAM) { |
779 | error = request_irq(pci_irq_vector(pdev, 2), | |
780 | vmci_interrupt_dma_datagram, | |
781 | 0, KBUILD_MODNAME, vmci_dev); | |
782 | if (error) { | |
783 | dev_err(&pdev->dev, | |
784 | "Failed to allocate irq %u: %d\n", | |
785 | pci_irq_vector(pdev, 2), error); | |
786 | goto err_free_bm_irq; | |
787 | } | |
788 | } | |
1f166439 GZ |
789 | } |
790 | ||
791 | dev_dbg(&pdev->dev, "Registered device\n"); | |
792 | ||
793 | atomic_inc(&vmci_num_guest_devices); | |
794 | ||
795 | /* Enable specific interrupt bits. */ | |
796 | cmd = VMCI_IMR_DATAGRAM; | |
f2db7361 | 797 | if (caps_in_use & VMCI_CAPS_NOTIFICATIONS) |
1f166439 | 798 | cmd |= VMCI_IMR_NOTIFICATION; |
3eed5372 JH |
799 | if (caps_in_use & VMCI_CAPS_DMA_DATAGRAM) |
800 | cmd |= VMCI_IMR_DMA_DATAGRAM; | |
e01153c7 | 801 | vmci_write_reg(vmci_dev, cmd, VMCI_IMR_ADDR); |
1f166439 GZ |
802 | |
803 | /* Enable interrupts. */ | |
e01153c7 | 804 | vmci_write_reg(vmci_dev, VMCI_CONTROL_INT_ENABLE, VMCI_CONTROL_ADDR); |
1f166439 GZ |
805 | |
806 | pci_set_drvdata(pdev, vmci_dev); | |
b1bba80a SG |
807 | |
808 | vmci_call_vsock_callback(false); | |
1f166439 GZ |
809 | return 0; |
810 | ||
3eed5372 JH |
811 | err_free_bm_irq: |
812 | free_irq(pci_irq_vector(pdev, 1), vmci_dev); | |
1f166439 | 813 | err_free_irq: |
3bb434cd | 814 | free_irq(pci_irq_vector(pdev, 0), vmci_dev); |
1f166439 GZ |
815 | tasklet_kill(&vmci_dev->datagram_tasklet); |
816 | tasklet_kill(&vmci_dev->bm_tasklet); | |
817 | ||
818 | err_disable_msi: | |
3bb434cd | 819 | pci_free_irq_vectors(pdev); |
1f166439 GZ |
820 | |
821 | vmci_err = vmci_event_unsubscribe(ctx_update_sub_id); | |
822 | if (vmci_err < VMCI_SUCCESS) | |
823 | dev_warn(&pdev->dev, | |
824 | "Failed to unsubscribe from event (type=%d) with subscriber (ID=0x%x): %d\n", | |
825 | VMCI_EVENT_CTX_ID_UPDATE, ctx_update_sub_id, vmci_err); | |
826 | ||
827 | err_remove_bitmap: | |
828 | if (vmci_dev->notification_bitmap) { | |
e01153c7 | 829 | vmci_write_reg(vmci_dev, VMCI_CONTROL_RESET, VMCI_CONTROL_ADDR); |
6d6dfb4f AK |
830 | dma_free_coherent(&pdev->dev, PAGE_SIZE, |
831 | vmci_dev->notification_bitmap, | |
832 | vmci_dev->notification_base); | |
1f166439 GZ |
833 | } |
834 | ||
835 | err_remove_vmci_dev_g: | |
836 | spin_lock_irq(&vmci_dev_spinlock); | |
6d6dfb4f | 837 | vmci_pdev = NULL; |
1f166439 GZ |
838 | vmci_dev_g = NULL; |
839 | spin_unlock_irq(&vmci_dev_spinlock); | |
840 | ||
8389c29a JH |
841 | err_free_data_buffers: |
842 | vmci_free_dg_buffers(vmci_dev); | |
1f166439 GZ |
843 | |
844 | /* The rest are managed resources and will be freed by PCI core */ | |
845 | return error; | |
846 | } | |
847 | ||
848 | static void vmci_guest_remove_device(struct pci_dev *pdev) | |
849 | { | |
850 | struct vmci_guest_device *vmci_dev = pci_get_drvdata(pdev); | |
851 | int vmci_err; | |
852 | ||
853 | dev_dbg(&pdev->dev, "Removing device\n"); | |
854 | ||
855 | atomic_dec(&vmci_num_guest_devices); | |
856 | ||
857 | vmci_qp_guest_endpoints_exit(); | |
858 | ||
859 | vmci_err = vmci_event_unsubscribe(ctx_update_sub_id); | |
860 | if (vmci_err < VMCI_SUCCESS) | |
861 | dev_warn(&pdev->dev, | |
862 | "Failed to unsubscribe from event (type=%d) with subscriber (ID=0x%x): %d\n", | |
863 | VMCI_EVENT_CTX_ID_UPDATE, ctx_update_sub_id, vmci_err); | |
864 | ||
865 | spin_lock_irq(&vmci_dev_spinlock); | |
866 | vmci_dev_g = NULL; | |
6d6dfb4f | 867 | vmci_pdev = NULL; |
1f166439 GZ |
868 | spin_unlock_irq(&vmci_dev_spinlock); |
869 | ||
870 | dev_dbg(&pdev->dev, "Resetting vmci device\n"); | |
e01153c7 | 871 | vmci_write_reg(vmci_dev, VMCI_CONTROL_RESET, VMCI_CONTROL_ADDR); |
1f166439 GZ |
872 | |
873 | /* | |
874 | * Free IRQ and then disable MSI/MSI-X as appropriate. For | |
875 | * MSI-X, we might have multiple vectors, each with their own | |
876 | * IRQ, which we must free too. | |
877 | */ | |
3eed5372 | 878 | if (vmci_dev->exclusive_vectors) { |
3bb434cd | 879 | free_irq(pci_irq_vector(pdev, 1), vmci_dev); |
3eed5372 JH |
880 | if (vmci_dev->mmio_base != NULL) |
881 | free_irq(pci_irq_vector(pdev, 2), vmci_dev); | |
882 | } | |
3bb434cd CH |
883 | free_irq(pci_irq_vector(pdev, 0), vmci_dev); |
884 | pci_free_irq_vectors(pdev); | |
1f166439 GZ |
885 | |
886 | tasklet_kill(&vmci_dev->datagram_tasklet); | |
887 | tasklet_kill(&vmci_dev->bm_tasklet); | |
888 | ||
889 | if (vmci_dev->notification_bitmap) { | |
890 | /* | |
891 | * The device reset above cleared the bitmap state of the | |
892 | * device, so we can safely free it here. | |
893 | */ | |
894 | ||
6d6dfb4f AK |
895 | dma_free_coherent(&pdev->dev, PAGE_SIZE, |
896 | vmci_dev->notification_bitmap, | |
897 | vmci_dev->notification_base); | |
1f166439 GZ |
898 | } |
899 | ||
8389c29a JH |
900 | vmci_free_dg_buffers(vmci_dev); |
901 | ||
902 | if (vmci_dev->mmio_base != NULL) | |
903 | pci_iounmap(pdev, vmci_dev->mmio_base); | |
1f166439 GZ |
904 | |
905 | /* The rest are managed resources and will be freed by PCI core */ | |
906 | } | |
907 | ||
32182cd3 | 908 | static const struct pci_device_id vmci_ids[] = { |
1f166439 GZ |
909 | { PCI_DEVICE(PCI_VENDOR_ID_VMWARE, PCI_DEVICE_ID_VMWARE_VMCI), }, |
910 | { 0 }, | |
911 | }; | |
912 | MODULE_DEVICE_TABLE(pci, vmci_ids); | |
913 | ||
914 | static struct pci_driver vmci_guest_driver = { | |
915 | .name = KBUILD_MODNAME, | |
916 | .id_table = vmci_ids, | |
917 | .probe = vmci_guest_probe_device, | |
918 | .remove = vmci_guest_remove_device, | |
919 | }; | |
920 | ||
921 | int __init vmci_guest_init(void) | |
922 | { | |
923 | return pci_register_driver(&vmci_guest_driver); | |
924 | } | |
925 | ||
926 | void __exit vmci_guest_exit(void) | |
927 | { | |
928 | pci_unregister_driver(&vmci_guest_driver); | |
929 | } |