]>
Commit | Line | Data |
---|---|---|
4daace0d JO |
1 | /* |
2 | * Copyright (c) Microsoft Corporation. | |
3 | * | |
4 | * Author: | |
5 | * Jake Oshins <jakeo@microsoft.com> | |
6 | * | |
7 | * This driver acts as a paravirtual front-end for PCI Express root buses. | |
8 | * When a PCI Express function (either an entire device or an SR-IOV | |
9 | * Virtual Function) is being passed through to the VM, this driver exposes | |
10 | * a new bus to the guest VM. This is modeled as a root PCI bus because | |
11 | * no bridges are being exposed to the VM. In fact, with a "Generation 2" | |
12 | * VM within Hyper-V, there may seem to be no PCI bus at all in the VM | |
13 | * until a device as been exposed using this driver. | |
14 | * | |
15 | * Each root PCI bus has its own PCI domain, which is called "Segment" in | |
16 | * the PCI Firmware Specifications. Thus while each device passed through | |
17 | * to the VM using this front-end will appear at "device 0", the domain will | |
18 | * be unique. Typically, each bus will have one PCI function on it, though | |
19 | * this driver does support more than one. | |
20 | * | |
21 | * In order to map the interrupts from the device through to the guest VM, | |
22 | * this driver also implements an IRQ Domain, which handles interrupts (either | |
23 | * MSI or MSI-X) associated with the functions on the bus. As interrupts are | |
24 | * set up, torn down, or reaffined, this driver communicates with the | |
25 | * underlying hypervisor to adjust the mappings in the I/O MMU so that each | |
26 | * interrupt will be delivered to the correct virtual processor at the right | |
27 | * vector. This driver does not support level-triggered (line-based) | |
28 | * interrupts, and will report that the Interrupt Line register in the | |
29 | * function's configuration space is zero. | |
30 | * | |
31 | * The rest of this driver mostly maps PCI concepts onto underlying Hyper-V | |
32 | * facilities. For instance, the configuration space of a function exposed | |
33 | * by Hyper-V is mapped into a single page of memory space, and the | |
34 | * read and write handlers for config space must be aware of this mechanism. | |
35 | * Similarly, device setup and teardown involves messages sent to and from | |
36 | * the PCI back-end driver in Hyper-V. | |
37 | * | |
38 | * This program is free software; you can redistribute it and/or modify it | |
39 | * under the terms of the GNU General Public License version 2 as published | |
40 | * by the Free Software Foundation. | |
41 | * | |
42 | * This program is distributed in the hope that it will be useful, but | |
43 | * WITHOUT ANY WARRANTY; without even the implied warranty of | |
44 | * MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or | |
45 | * NON INFRINGEMENT. See the GNU General Public License for more | |
46 | * details. | |
47 | * | |
48 | */ | |
49 | ||
50 | #include <linux/kernel.h> | |
51 | #include <linux/module.h> | |
52 | #include <linux/pci.h> | |
53 | #include <linux/semaphore.h> | |
54 | #include <linux/irqdomain.h> | |
55 | #include <asm/irqdomain.h> | |
56 | #include <asm/apic.h> | |
57 | #include <linux/msi.h> | |
58 | #include <linux/hyperv.h> | |
59 | #include <asm/mshyperv.h> | |
60 | ||
61 | /* | |
62 | * Protocol versions. The low word is the minor version, the high word the | |
63 | * major version. | |
64 | */ | |
65 | ||
66 | #define PCI_MAKE_VERSION(major, minor) ((u32)(((major) << 16) | (major))) | |
67 | #define PCI_MAJOR_VERSION(version) ((u32)(version) >> 16) | |
68 | #define PCI_MINOR_VERSION(version) ((u32)(version) & 0xff) | |
69 | ||
70 | enum { | |
71 | PCI_PROTOCOL_VERSION_1_1 = PCI_MAKE_VERSION(1, 1), | |
72 | PCI_PROTOCOL_VERSION_CURRENT = PCI_PROTOCOL_VERSION_1_1 | |
73 | }; | |
74 | ||
75 | #define PCI_CONFIG_MMIO_LENGTH 0x2000 | |
76 | #define CFG_PAGE_OFFSET 0x1000 | |
77 | #define CFG_PAGE_SIZE (PCI_CONFIG_MMIO_LENGTH - CFG_PAGE_OFFSET) | |
78 | ||
79 | #define MAX_SUPPORTED_MSI_MESSAGES 0x400 | |
80 | ||
81 | /* | |
82 | * Message Types | |
83 | */ | |
84 | ||
85 | enum pci_message_type { | |
86 | /* | |
87 | * Version 1.1 | |
88 | */ | |
89 | PCI_MESSAGE_BASE = 0x42490000, | |
90 | PCI_BUS_RELATIONS = PCI_MESSAGE_BASE + 0, | |
91 | PCI_QUERY_BUS_RELATIONS = PCI_MESSAGE_BASE + 1, | |
92 | PCI_POWER_STATE_CHANGE = PCI_MESSAGE_BASE + 4, | |
93 | PCI_QUERY_RESOURCE_REQUIREMENTS = PCI_MESSAGE_BASE + 5, | |
94 | PCI_QUERY_RESOURCE_RESOURCES = PCI_MESSAGE_BASE + 6, | |
95 | PCI_BUS_D0ENTRY = PCI_MESSAGE_BASE + 7, | |
96 | PCI_BUS_D0EXIT = PCI_MESSAGE_BASE + 8, | |
97 | PCI_READ_BLOCK = PCI_MESSAGE_BASE + 9, | |
98 | PCI_WRITE_BLOCK = PCI_MESSAGE_BASE + 0xA, | |
99 | PCI_EJECT = PCI_MESSAGE_BASE + 0xB, | |
100 | PCI_QUERY_STOP = PCI_MESSAGE_BASE + 0xC, | |
101 | PCI_REENABLE = PCI_MESSAGE_BASE + 0xD, | |
102 | PCI_QUERY_STOP_FAILED = PCI_MESSAGE_BASE + 0xE, | |
103 | PCI_EJECTION_COMPLETE = PCI_MESSAGE_BASE + 0xF, | |
104 | PCI_RESOURCES_ASSIGNED = PCI_MESSAGE_BASE + 0x10, | |
105 | PCI_RESOURCES_RELEASED = PCI_MESSAGE_BASE + 0x11, | |
106 | PCI_INVALIDATE_BLOCK = PCI_MESSAGE_BASE + 0x12, | |
107 | PCI_QUERY_PROTOCOL_VERSION = PCI_MESSAGE_BASE + 0x13, | |
108 | PCI_CREATE_INTERRUPT_MESSAGE = PCI_MESSAGE_BASE + 0x14, | |
109 | PCI_DELETE_INTERRUPT_MESSAGE = PCI_MESSAGE_BASE + 0x15, | |
110 | PCI_MESSAGE_MAXIMUM | |
111 | }; | |
112 | ||
113 | /* | |
114 | * Structures defining the virtual PCI Express protocol. | |
115 | */ | |
116 | ||
117 | union pci_version { | |
118 | struct { | |
119 | u16 minor_version; | |
120 | u16 major_version; | |
121 | } parts; | |
122 | u32 version; | |
123 | } __packed; | |
124 | ||
125 | /* | |
126 | * Function numbers are 8-bits wide on Express, as interpreted through ARI, | |
127 | * which is all this driver does. This representation is the one used in | |
128 | * Windows, which is what is expected when sending this back and forth with | |
129 | * the Hyper-V parent partition. | |
130 | */ | |
131 | union win_slot_encoding { | |
132 | struct { | |
60e2e2fb DC |
133 | u32 dev:5; |
134 | u32 func:3; | |
4daace0d JO |
135 | u32 reserved:24; |
136 | } bits; | |
137 | u32 slot; | |
138 | } __packed; | |
139 | ||
140 | /* | |
141 | * Pretty much as defined in the PCI Specifications. | |
142 | */ | |
143 | struct pci_function_description { | |
144 | u16 v_id; /* vendor ID */ | |
145 | u16 d_id; /* device ID */ | |
146 | u8 rev; | |
147 | u8 prog_intf; | |
148 | u8 subclass; | |
149 | u8 base_class; | |
150 | u32 subsystem_id; | |
151 | union win_slot_encoding win_slot; | |
152 | u32 ser; /* serial number */ | |
153 | } __packed; | |
154 | ||
155 | /** | |
156 | * struct hv_msi_desc | |
157 | * @vector: IDT entry | |
158 | * @delivery_mode: As defined in Intel's Programmer's | |
159 | * Reference Manual, Volume 3, Chapter 8. | |
160 | * @vector_count: Number of contiguous entries in the | |
161 | * Interrupt Descriptor Table that are | |
162 | * occupied by this Message-Signaled | |
163 | * Interrupt. For "MSI", as first defined | |
164 | * in PCI 2.2, this can be between 1 and | |
165 | * 32. For "MSI-X," as first defined in PCI | |
166 | * 3.0, this must be 1, as each MSI-X table | |
167 | * entry would have its own descriptor. | |
168 | * @reserved: Empty space | |
169 | * @cpu_mask: All the target virtual processors. | |
170 | */ | |
171 | struct hv_msi_desc { | |
172 | u8 vector; | |
173 | u8 delivery_mode; | |
174 | u16 vector_count; | |
175 | u32 reserved; | |
176 | u64 cpu_mask; | |
177 | } __packed; | |
178 | ||
179 | /** | |
180 | * struct tran_int_desc | |
181 | * @reserved: unused, padding | |
182 | * @vector_count: same as in hv_msi_desc | |
183 | * @data: This is the "data payload" value that is | |
184 | * written by the device when it generates | |
185 | * a message-signaled interrupt, either MSI | |
186 | * or MSI-X. | |
187 | * @address: This is the address to which the data | |
188 | * payload is written on interrupt | |
189 | * generation. | |
190 | */ | |
191 | struct tran_int_desc { | |
192 | u16 reserved; | |
193 | u16 vector_count; | |
194 | u32 data; | |
195 | u64 address; | |
196 | } __packed; | |
197 | ||
198 | /* | |
199 | * A generic message format for virtual PCI. | |
200 | * Specific message formats are defined later in the file. | |
201 | */ | |
202 | ||
203 | struct pci_message { | |
0c6045d8 | 204 | u32 type; |
4daace0d JO |
205 | } __packed; |
206 | ||
207 | struct pci_child_message { | |
0c6045d8 | 208 | struct pci_message message_type; |
4daace0d JO |
209 | union win_slot_encoding wslot; |
210 | } __packed; | |
211 | ||
212 | struct pci_incoming_message { | |
213 | struct vmpacket_descriptor hdr; | |
214 | struct pci_message message_type; | |
215 | } __packed; | |
216 | ||
217 | struct pci_response { | |
218 | struct vmpacket_descriptor hdr; | |
219 | s32 status; /* negative values are failures */ | |
220 | } __packed; | |
221 | ||
222 | struct pci_packet { | |
223 | void (*completion_func)(void *context, struct pci_response *resp, | |
224 | int resp_packet_size); | |
225 | void *compl_ctxt; | |
0c6045d8 DC |
226 | |
227 | struct pci_message message[0]; | |
4daace0d JO |
228 | }; |
229 | ||
230 | /* | |
231 | * Specific message types supporting the PCI protocol. | |
232 | */ | |
233 | ||
234 | /* | |
235 | * Version negotiation message. Sent from the guest to the host. | |
236 | * The guest is free to try different versions until the host | |
237 | * accepts the version. | |
238 | * | |
239 | * pci_version: The protocol version requested. | |
240 | * is_last_attempt: If TRUE, this is the last version guest will request. | |
241 | * reservedz: Reserved field, set to zero. | |
242 | */ | |
243 | ||
244 | struct pci_version_request { | |
245 | struct pci_message message_type; | |
246 | enum pci_message_type protocol_version; | |
247 | } __packed; | |
248 | ||
249 | /* | |
250 | * Bus D0 Entry. This is sent from the guest to the host when the virtual | |
251 | * bus (PCI Express port) is ready for action. | |
252 | */ | |
253 | ||
254 | struct pci_bus_d0_entry { | |
255 | struct pci_message message_type; | |
256 | u32 reserved; | |
257 | u64 mmio_base; | |
258 | } __packed; | |
259 | ||
260 | struct pci_bus_relations { | |
261 | struct pci_incoming_message incoming; | |
262 | u32 device_count; | |
7d0f8eec | 263 | struct pci_function_description func[0]; |
4daace0d JO |
264 | } __packed; |
265 | ||
266 | struct pci_q_res_req_response { | |
267 | struct vmpacket_descriptor hdr; | |
268 | s32 status; /* negative values are failures */ | |
269 | u32 probed_bar[6]; | |
270 | } __packed; | |
271 | ||
272 | struct pci_set_power { | |
273 | struct pci_message message_type; | |
274 | union win_slot_encoding wslot; | |
275 | u32 power_state; /* In Windows terms */ | |
276 | u32 reserved; | |
277 | } __packed; | |
278 | ||
279 | struct pci_set_power_response { | |
280 | struct vmpacket_descriptor hdr; | |
281 | s32 status; /* negative values are failures */ | |
282 | union win_slot_encoding wslot; | |
283 | u32 resultant_state; /* In Windows terms */ | |
284 | u32 reserved; | |
285 | } __packed; | |
286 | ||
287 | struct pci_resources_assigned { | |
288 | struct pci_message message_type; | |
289 | union win_slot_encoding wslot; | |
290 | u8 memory_range[0x14][6]; /* not used here */ | |
291 | u32 msi_descriptors; | |
292 | u32 reserved[4]; | |
293 | } __packed; | |
294 | ||
295 | struct pci_create_interrupt { | |
296 | struct pci_message message_type; | |
297 | union win_slot_encoding wslot; | |
298 | struct hv_msi_desc int_desc; | |
299 | } __packed; | |
300 | ||
301 | struct pci_create_int_response { | |
302 | struct pci_response response; | |
303 | u32 reserved; | |
304 | struct tran_int_desc int_desc; | |
305 | } __packed; | |
306 | ||
307 | struct pci_delete_interrupt { | |
308 | struct pci_message message_type; | |
309 | union win_slot_encoding wslot; | |
310 | struct tran_int_desc int_desc; | |
311 | } __packed; | |
312 | ||
313 | struct pci_dev_incoming { | |
314 | struct pci_incoming_message incoming; | |
315 | union win_slot_encoding wslot; | |
316 | } __packed; | |
317 | ||
318 | struct pci_eject_response { | |
0c6045d8 | 319 | struct pci_message message_type; |
4daace0d JO |
320 | union win_slot_encoding wslot; |
321 | u32 status; | |
322 | } __packed; | |
323 | ||
324 | static int pci_ring_size = (4 * PAGE_SIZE); | |
325 | ||
326 | /* | |
327 | * Definitions or interrupt steering hypercall. | |
328 | */ | |
329 | #define HV_PARTITION_ID_SELF ((u64)-1) | |
330 | #define HVCALL_RETARGET_INTERRUPT 0x7e | |
331 | ||
332 | struct retarget_msi_interrupt { | |
333 | u64 partition_id; /* use "self" */ | |
334 | u64 device_id; | |
335 | u32 source; /* 1 for MSI(-X) */ | |
336 | u32 reserved1; | |
337 | u32 address; | |
338 | u32 data; | |
339 | u64 reserved2; | |
340 | u32 vector; | |
341 | u32 flags; | |
342 | u64 vp_mask; | |
343 | } __packed; | |
344 | ||
345 | /* | |
346 | * Driver specific state. | |
347 | */ | |
348 | ||
349 | enum hv_pcibus_state { | |
350 | hv_pcibus_init = 0, | |
351 | hv_pcibus_probed, | |
352 | hv_pcibus_installed, | |
353 | hv_pcibus_maximum | |
354 | }; | |
355 | ||
356 | struct hv_pcibus_device { | |
357 | struct pci_sysdata sysdata; | |
358 | enum hv_pcibus_state state; | |
359 | atomic_t remove_lock; | |
360 | struct hv_device *hdev; | |
361 | resource_size_t low_mmio_space; | |
362 | resource_size_t high_mmio_space; | |
363 | struct resource *mem_config; | |
364 | struct resource *low_mmio_res; | |
365 | struct resource *high_mmio_res; | |
366 | struct completion *survey_event; | |
367 | struct completion remove_event; | |
368 | struct pci_bus *pci_bus; | |
369 | spinlock_t config_lock; /* Avoid two threads writing index page */ | |
370 | spinlock_t device_list_lock; /* Protect lists below */ | |
371 | void __iomem *cfg_addr; | |
372 | ||
373 | struct semaphore enum_sem; | |
374 | struct list_head resources_for_children; | |
375 | ||
376 | struct list_head children; | |
377 | struct list_head dr_list; | |
4daace0d JO |
378 | |
379 | struct msi_domain_info msi_info; | |
380 | struct msi_controller msi_chip; | |
381 | struct irq_domain *irq_domain; | |
0de8ce3e LL |
382 | struct retarget_msi_interrupt retarget_msi_interrupt_params; |
383 | spinlock_t retarget_msi_interrupt_lock; | |
4daace0d JO |
384 | }; |
385 | ||
386 | /* | |
387 | * Tracks "Device Relations" messages from the host, which must be both | |
388 | * processed in order and deferred so that they don't run in the context | |
389 | * of the incoming packet callback. | |
390 | */ | |
391 | struct hv_dr_work { | |
392 | struct work_struct wrk; | |
393 | struct hv_pcibus_device *bus; | |
394 | }; | |
395 | ||
396 | struct hv_dr_state { | |
397 | struct list_head list_entry; | |
398 | u32 device_count; | |
7d0f8eec | 399 | struct pci_function_description func[0]; |
4daace0d JO |
400 | }; |
401 | ||
402 | enum hv_pcichild_state { | |
403 | hv_pcichild_init = 0, | |
404 | hv_pcichild_requirements, | |
405 | hv_pcichild_resourced, | |
406 | hv_pcichild_ejecting, | |
407 | hv_pcichild_maximum | |
408 | }; | |
409 | ||
410 | enum hv_pcidev_ref_reason { | |
411 | hv_pcidev_ref_invalid = 0, | |
412 | hv_pcidev_ref_initial, | |
413 | hv_pcidev_ref_by_slot, | |
414 | hv_pcidev_ref_packet, | |
415 | hv_pcidev_ref_pnp, | |
416 | hv_pcidev_ref_childlist, | |
417 | hv_pcidev_irqdata, | |
418 | hv_pcidev_ref_max | |
419 | }; | |
420 | ||
421 | struct hv_pci_dev { | |
422 | /* List protected by pci_rescan_remove_lock */ | |
423 | struct list_head list_entry; | |
424 | atomic_t refs; | |
425 | enum hv_pcichild_state state; | |
426 | struct pci_function_description desc; | |
427 | bool reported_missing; | |
428 | struct hv_pcibus_device *hbus; | |
429 | struct work_struct wrk; | |
430 | ||
431 | /* | |
432 | * What would be observed if one wrote 0xFFFFFFFF to a BAR and then | |
433 | * read it back, for each of the BAR offsets within config space. | |
434 | */ | |
435 | u32 probed_bar[6]; | |
436 | }; | |
437 | ||
438 | struct hv_pci_compl { | |
439 | struct completion host_event; | |
440 | s32 completion_status; | |
441 | }; | |
442 | ||
443 | /** | |
444 | * hv_pci_generic_compl() - Invoked for a completion packet | |
445 | * @context: Set up by the sender of the packet. | |
446 | * @resp: The response packet | |
447 | * @resp_packet_size: Size in bytes of the packet | |
448 | * | |
449 | * This function is used to trigger an event and report status | |
450 | * for any message for which the completion packet contains a | |
451 | * status and nothing else. | |
452 | */ | |
a5b45b7b DC |
453 | static void hv_pci_generic_compl(void *context, struct pci_response *resp, |
454 | int resp_packet_size) | |
4daace0d JO |
455 | { |
456 | struct hv_pci_compl *comp_pkt = context; | |
457 | ||
458 | if (resp_packet_size >= offsetofend(struct pci_response, status)) | |
459 | comp_pkt->completion_status = resp->status; | |
a5b45b7b DC |
460 | else |
461 | comp_pkt->completion_status = -1; | |
462 | ||
4daace0d JO |
463 | complete(&comp_pkt->host_event); |
464 | } | |
465 | ||
466 | static struct hv_pci_dev *get_pcichild_wslot(struct hv_pcibus_device *hbus, | |
467 | u32 wslot); | |
468 | static void get_pcichild(struct hv_pci_dev *hv_pcidev, | |
469 | enum hv_pcidev_ref_reason reason); | |
470 | static void put_pcichild(struct hv_pci_dev *hv_pcidev, | |
471 | enum hv_pcidev_ref_reason reason); | |
472 | ||
473 | static void get_hvpcibus(struct hv_pcibus_device *hv_pcibus); | |
474 | static void put_hvpcibus(struct hv_pcibus_device *hv_pcibus); | |
475 | ||
476 | /** | |
477 | * devfn_to_wslot() - Convert from Linux PCI slot to Windows | |
478 | * @devfn: The Linux representation of PCI slot | |
479 | * | |
480 | * Windows uses a slightly different representation of PCI slot. | |
481 | * | |
482 | * Return: The Windows representation | |
483 | */ | |
484 | static u32 devfn_to_wslot(int devfn) | |
485 | { | |
486 | union win_slot_encoding wslot; | |
487 | ||
488 | wslot.slot = 0; | |
60e2e2fb DC |
489 | wslot.bits.dev = PCI_SLOT(devfn); |
490 | wslot.bits.func = PCI_FUNC(devfn); | |
4daace0d JO |
491 | |
492 | return wslot.slot; | |
493 | } | |
494 | ||
495 | /** | |
496 | * wslot_to_devfn() - Convert from Windows PCI slot to Linux | |
497 | * @wslot: The Windows representation of PCI slot | |
498 | * | |
499 | * Windows uses a slightly different representation of PCI slot. | |
500 | * | |
501 | * Return: The Linux representation | |
502 | */ | |
503 | static int wslot_to_devfn(u32 wslot) | |
504 | { | |
505 | union win_slot_encoding slot_no; | |
506 | ||
507 | slot_no.slot = wslot; | |
60e2e2fb | 508 | return PCI_DEVFN(slot_no.bits.dev, slot_no.bits.func); |
4daace0d JO |
509 | } |
510 | ||
511 | /* | |
512 | * PCI Configuration Space for these root PCI buses is implemented as a pair | |
513 | * of pages in memory-mapped I/O space. Writing to the first page chooses | |
514 | * the PCI function being written or read. Once the first page has been | |
515 | * written to, the following page maps in the entire configuration space of | |
516 | * the function. | |
517 | */ | |
518 | ||
519 | /** | |
520 | * _hv_pcifront_read_config() - Internal PCI config read | |
521 | * @hpdev: The PCI driver's representation of the device | |
522 | * @where: Offset within config space | |
523 | * @size: Size of the transfer | |
524 | * @val: Pointer to the buffer receiving the data | |
525 | */ | |
526 | static void _hv_pcifront_read_config(struct hv_pci_dev *hpdev, int where, | |
527 | int size, u32 *val) | |
528 | { | |
529 | unsigned long flags; | |
530 | void __iomem *addr = hpdev->hbus->cfg_addr + CFG_PAGE_OFFSET + where; | |
531 | ||
532 | /* | |
533 | * If the attempt is to read the IDs or the ROM BAR, simulate that. | |
534 | */ | |
535 | if (where + size <= PCI_COMMAND) { | |
536 | memcpy(val, ((u8 *)&hpdev->desc.v_id) + where, size); | |
537 | } else if (where >= PCI_CLASS_REVISION && where + size <= | |
538 | PCI_CACHE_LINE_SIZE) { | |
539 | memcpy(val, ((u8 *)&hpdev->desc.rev) + where - | |
540 | PCI_CLASS_REVISION, size); | |
541 | } else if (where >= PCI_SUBSYSTEM_VENDOR_ID && where + size <= | |
542 | PCI_ROM_ADDRESS) { | |
543 | memcpy(val, (u8 *)&hpdev->desc.subsystem_id + where - | |
544 | PCI_SUBSYSTEM_VENDOR_ID, size); | |
545 | } else if (where >= PCI_ROM_ADDRESS && where + size <= | |
546 | PCI_CAPABILITY_LIST) { | |
547 | /* ROM BARs are unimplemented */ | |
548 | *val = 0; | |
549 | } else if (where >= PCI_INTERRUPT_LINE && where + size <= | |
550 | PCI_INTERRUPT_PIN) { | |
551 | /* | |
552 | * Interrupt Line and Interrupt PIN are hard-wired to zero | |
553 | * because this front-end only supports message-signaled | |
554 | * interrupts. | |
555 | */ | |
556 | *val = 0; | |
557 | } else if (where + size <= CFG_PAGE_SIZE) { | |
558 | spin_lock_irqsave(&hpdev->hbus->config_lock, flags); | |
559 | /* Choose the function to be read. (See comment above) */ | |
560 | writel(hpdev->desc.win_slot.slot, hpdev->hbus->cfg_addr); | |
bdd74440 VK |
561 | /* Make sure the function was chosen before we start reading. */ |
562 | mb(); | |
4daace0d JO |
563 | /* Read from that function's config space. */ |
564 | switch (size) { | |
565 | case 1: | |
566 | *val = readb(addr); | |
567 | break; | |
568 | case 2: | |
569 | *val = readw(addr); | |
570 | break; | |
571 | default: | |
572 | *val = readl(addr); | |
573 | break; | |
574 | } | |
bdd74440 VK |
575 | /* |
576 | * Make sure the write was done before we release the spinlock | |
577 | * allowing consecutive reads/writes. | |
578 | */ | |
579 | mb(); | |
4daace0d JO |
580 | spin_unlock_irqrestore(&hpdev->hbus->config_lock, flags); |
581 | } else { | |
582 | dev_err(&hpdev->hbus->hdev->device, | |
583 | "Attempt to read beyond a function's config space.\n"); | |
584 | } | |
585 | } | |
586 | ||
587 | /** | |
588 | * _hv_pcifront_write_config() - Internal PCI config write | |
589 | * @hpdev: The PCI driver's representation of the device | |
590 | * @where: Offset within config space | |
591 | * @size: Size of the transfer | |
592 | * @val: The data being transferred | |
593 | */ | |
594 | static void _hv_pcifront_write_config(struct hv_pci_dev *hpdev, int where, | |
595 | int size, u32 val) | |
596 | { | |
597 | unsigned long flags; | |
598 | void __iomem *addr = hpdev->hbus->cfg_addr + CFG_PAGE_OFFSET + where; | |
599 | ||
600 | if (where >= PCI_SUBSYSTEM_VENDOR_ID && | |
601 | where + size <= PCI_CAPABILITY_LIST) { | |
602 | /* SSIDs and ROM BARs are read-only */ | |
603 | } else if (where >= PCI_COMMAND && where + size <= CFG_PAGE_SIZE) { | |
604 | spin_lock_irqsave(&hpdev->hbus->config_lock, flags); | |
605 | /* Choose the function to be written. (See comment above) */ | |
606 | writel(hpdev->desc.win_slot.slot, hpdev->hbus->cfg_addr); | |
bdd74440 VK |
607 | /* Make sure the function was chosen before we start writing. */ |
608 | wmb(); | |
4daace0d JO |
609 | /* Write to that function's config space. */ |
610 | switch (size) { | |
611 | case 1: | |
612 | writeb(val, addr); | |
613 | break; | |
614 | case 2: | |
615 | writew(val, addr); | |
616 | break; | |
617 | default: | |
618 | writel(val, addr); | |
619 | break; | |
620 | } | |
bdd74440 VK |
621 | /* |
622 | * Make sure the write was done before we release the spinlock | |
623 | * allowing consecutive reads/writes. | |
624 | */ | |
625 | mb(); | |
4daace0d JO |
626 | spin_unlock_irqrestore(&hpdev->hbus->config_lock, flags); |
627 | } else { | |
628 | dev_err(&hpdev->hbus->hdev->device, | |
629 | "Attempt to write beyond a function's config space.\n"); | |
630 | } | |
631 | } | |
632 | ||
633 | /** | |
634 | * hv_pcifront_read_config() - Read configuration space | |
635 | * @bus: PCI Bus structure | |
636 | * @devfn: Device/function | |
637 | * @where: Offset from base | |
638 | * @size: Byte/word/dword | |
639 | * @val: Value to be read | |
640 | * | |
641 | * Return: PCIBIOS_SUCCESSFUL on success | |
642 | * PCIBIOS_DEVICE_NOT_FOUND on failure | |
643 | */ | |
644 | static int hv_pcifront_read_config(struct pci_bus *bus, unsigned int devfn, | |
645 | int where, int size, u32 *val) | |
646 | { | |
647 | struct hv_pcibus_device *hbus = | |
648 | container_of(bus->sysdata, struct hv_pcibus_device, sysdata); | |
649 | struct hv_pci_dev *hpdev; | |
650 | ||
651 | hpdev = get_pcichild_wslot(hbus, devfn_to_wslot(devfn)); | |
652 | if (!hpdev) | |
653 | return PCIBIOS_DEVICE_NOT_FOUND; | |
654 | ||
655 | _hv_pcifront_read_config(hpdev, where, size, val); | |
656 | ||
657 | put_pcichild(hpdev, hv_pcidev_ref_by_slot); | |
658 | return PCIBIOS_SUCCESSFUL; | |
659 | } | |
660 | ||
661 | /** | |
662 | * hv_pcifront_write_config() - Write configuration space | |
663 | * @bus: PCI Bus structure | |
664 | * @devfn: Device/function | |
665 | * @where: Offset from base | |
666 | * @size: Byte/word/dword | |
667 | * @val: Value to be written to device | |
668 | * | |
669 | * Return: PCIBIOS_SUCCESSFUL on success | |
670 | * PCIBIOS_DEVICE_NOT_FOUND on failure | |
671 | */ | |
672 | static int hv_pcifront_write_config(struct pci_bus *bus, unsigned int devfn, | |
673 | int where, int size, u32 val) | |
674 | { | |
675 | struct hv_pcibus_device *hbus = | |
676 | container_of(bus->sysdata, struct hv_pcibus_device, sysdata); | |
677 | struct hv_pci_dev *hpdev; | |
678 | ||
679 | hpdev = get_pcichild_wslot(hbus, devfn_to_wslot(devfn)); | |
680 | if (!hpdev) | |
681 | return PCIBIOS_DEVICE_NOT_FOUND; | |
682 | ||
683 | _hv_pcifront_write_config(hpdev, where, size, val); | |
684 | ||
685 | put_pcichild(hpdev, hv_pcidev_ref_by_slot); | |
686 | return PCIBIOS_SUCCESSFUL; | |
687 | } | |
688 | ||
689 | /* PCIe operations */ | |
690 | static struct pci_ops hv_pcifront_ops = { | |
691 | .read = hv_pcifront_read_config, | |
692 | .write = hv_pcifront_write_config, | |
693 | }; | |
694 | ||
695 | /* Interrupt management hooks */ | |
696 | static void hv_int_desc_free(struct hv_pci_dev *hpdev, | |
697 | struct tran_int_desc *int_desc) | |
698 | { | |
699 | struct pci_delete_interrupt *int_pkt; | |
700 | struct { | |
701 | struct pci_packet pkt; | |
0c6045d8 | 702 | u8 buffer[sizeof(struct pci_delete_interrupt)]; |
4daace0d JO |
703 | } ctxt; |
704 | ||
705 | memset(&ctxt, 0, sizeof(ctxt)); | |
706 | int_pkt = (struct pci_delete_interrupt *)&ctxt.pkt.message; | |
0c6045d8 | 707 | int_pkt->message_type.type = |
4daace0d JO |
708 | PCI_DELETE_INTERRUPT_MESSAGE; |
709 | int_pkt->wslot.slot = hpdev->desc.win_slot.slot; | |
710 | int_pkt->int_desc = *int_desc; | |
711 | vmbus_sendpacket(hpdev->hbus->hdev->channel, int_pkt, sizeof(*int_pkt), | |
712 | (unsigned long)&ctxt.pkt, VM_PKT_DATA_INBAND, 0); | |
713 | kfree(int_desc); | |
714 | } | |
715 | ||
716 | /** | |
717 | * hv_msi_free() - Free the MSI. | |
718 | * @domain: The interrupt domain pointer | |
719 | * @info: Extra MSI-related context | |
720 | * @irq: Identifies the IRQ. | |
721 | * | |
722 | * The Hyper-V parent partition and hypervisor are tracking the | |
723 | * messages that are in use, keeping the interrupt redirection | |
724 | * table up to date. This callback sends a message that frees | |
725 | * the IRT entry and related tracking nonsense. | |
726 | */ | |
727 | static void hv_msi_free(struct irq_domain *domain, struct msi_domain_info *info, | |
728 | unsigned int irq) | |
729 | { | |
730 | struct hv_pcibus_device *hbus; | |
731 | struct hv_pci_dev *hpdev; | |
732 | struct pci_dev *pdev; | |
733 | struct tran_int_desc *int_desc; | |
734 | struct irq_data *irq_data = irq_domain_get_irq_data(domain, irq); | |
735 | struct msi_desc *msi = irq_data_get_msi_desc(irq_data); | |
736 | ||
737 | pdev = msi_desc_to_pci_dev(msi); | |
738 | hbus = info->data; | |
0c6e617f CA |
739 | int_desc = irq_data_get_irq_chip_data(irq_data); |
740 | if (!int_desc) | |
4daace0d JO |
741 | return; |
742 | ||
0c6e617f CA |
743 | irq_data->chip_data = NULL; |
744 | hpdev = get_pcichild_wslot(hbus, devfn_to_wslot(pdev->devfn)); | |
745 | if (!hpdev) { | |
746 | kfree(int_desc); | |
747 | return; | |
4daace0d JO |
748 | } |
749 | ||
0c6e617f | 750 | hv_int_desc_free(hpdev, int_desc); |
4daace0d JO |
751 | put_pcichild(hpdev, hv_pcidev_ref_by_slot); |
752 | } | |
753 | ||
754 | static int hv_set_affinity(struct irq_data *data, const struct cpumask *dest, | |
755 | bool force) | |
756 | { | |
757 | struct irq_data *parent = data->parent_data; | |
758 | ||
759 | return parent->chip->irq_set_affinity(parent, dest, force); | |
760 | } | |
761 | ||
542ccf45 | 762 | static void hv_irq_mask(struct irq_data *data) |
4daace0d JO |
763 | { |
764 | pci_msi_mask_irq(data); | |
765 | } | |
766 | ||
767 | /** | |
768 | * hv_irq_unmask() - "Unmask" the IRQ by setting its current | |
769 | * affinity. | |
770 | * @data: Describes the IRQ | |
771 | * | |
772 | * Build new a destination for the MSI and make a hypercall to | |
773 | * update the Interrupt Redirection Table. "Device Logical ID" | |
774 | * is built out of this PCI bus's instance GUID and the function | |
775 | * number of the device. | |
776 | */ | |
542ccf45 | 777 | static void hv_irq_unmask(struct irq_data *data) |
4daace0d JO |
778 | { |
779 | struct msi_desc *msi_desc = irq_data_get_msi_desc(data); | |
780 | struct irq_cfg *cfg = irqd_cfg(data); | |
0de8ce3e | 781 | struct retarget_msi_interrupt *params; |
4daace0d JO |
782 | struct hv_pcibus_device *hbus; |
783 | struct cpumask *dest; | |
784 | struct pci_bus *pbus; | |
785 | struct pci_dev *pdev; | |
786 | int cpu; | |
0de8ce3e | 787 | unsigned long flags; |
4daace0d JO |
788 | |
789 | dest = irq_data_get_affinity_mask(data); | |
790 | pdev = msi_desc_to_pci_dev(msi_desc); | |
791 | pbus = pdev->bus; | |
792 | hbus = container_of(pbus->sysdata, struct hv_pcibus_device, sysdata); | |
793 | ||
0de8ce3e LL |
794 | spin_lock_irqsave(&hbus->retarget_msi_interrupt_lock, flags); |
795 | ||
796 | params = &hbus->retarget_msi_interrupt_params; | |
797 | memset(params, 0, sizeof(*params)); | |
798 | params->partition_id = HV_PARTITION_ID_SELF; | |
799 | params->source = 1; /* MSI(-X) */ | |
800 | params->address = msi_desc->msg.address_lo; | |
801 | params->data = msi_desc->msg.data; | |
802 | params->device_id = (hbus->hdev->dev_instance.b[5] << 24) | | |
4daace0d JO |
803 | (hbus->hdev->dev_instance.b[4] << 16) | |
804 | (hbus->hdev->dev_instance.b[7] << 8) | | |
805 | (hbus->hdev->dev_instance.b[6] & 0xf8) | | |
806 | PCI_FUNC(pdev->devfn); | |
0de8ce3e | 807 | params->vector = cfg->vector; |
4daace0d JO |
808 | |
809 | for_each_cpu_and(cpu, dest, cpu_online_mask) | |
0de8ce3e LL |
810 | params->vp_mask |= (1ULL << vmbus_cpu_number_to_vp_number(cpu)); |
811 | ||
812 | hv_do_hypercall(HVCALL_RETARGET_INTERRUPT, params, NULL); | |
4daace0d | 813 | |
0de8ce3e | 814 | spin_unlock_irqrestore(&hbus->retarget_msi_interrupt_lock, flags); |
4daace0d JO |
815 | |
816 | pci_msi_unmask_irq(data); | |
817 | } | |
818 | ||
819 | struct compose_comp_ctxt { | |
820 | struct hv_pci_compl comp_pkt; | |
821 | struct tran_int_desc int_desc; | |
822 | }; | |
823 | ||
824 | static void hv_pci_compose_compl(void *context, struct pci_response *resp, | |
825 | int resp_packet_size) | |
826 | { | |
827 | struct compose_comp_ctxt *comp_pkt = context; | |
828 | struct pci_create_int_response *int_resp = | |
829 | (struct pci_create_int_response *)resp; | |
830 | ||
831 | comp_pkt->comp_pkt.completion_status = resp->status; | |
832 | comp_pkt->int_desc = int_resp->int_desc; | |
833 | complete(&comp_pkt->comp_pkt.host_event); | |
834 | } | |
835 | ||
836 | /** | |
837 | * hv_compose_msi_msg() - Supplies a valid MSI address/data | |
838 | * @data: Everything about this MSI | |
839 | * @msg: Buffer that is filled in by this function | |
840 | * | |
841 | * This function unpacks the IRQ looking for target CPU set, IDT | |
842 | * vector and mode and sends a message to the parent partition | |
843 | * asking for a mapping for that tuple in this partition. The | |
844 | * response supplies a data value and address to which that data | |
845 | * should be written to trigger that interrupt. | |
846 | */ | |
847 | static void hv_compose_msi_msg(struct irq_data *data, struct msi_msg *msg) | |
848 | { | |
849 | struct irq_cfg *cfg = irqd_cfg(data); | |
850 | struct hv_pcibus_device *hbus; | |
851 | struct hv_pci_dev *hpdev; | |
852 | struct pci_bus *pbus; | |
853 | struct pci_dev *pdev; | |
854 | struct pci_create_interrupt *int_pkt; | |
855 | struct compose_comp_ctxt comp; | |
856 | struct tran_int_desc *int_desc; | |
857 | struct cpumask *affinity; | |
858 | struct { | |
859 | struct pci_packet pkt; | |
0c6045d8 | 860 | u8 buffer[sizeof(struct pci_create_interrupt)]; |
4daace0d JO |
861 | } ctxt; |
862 | int cpu; | |
863 | int ret; | |
864 | ||
865 | pdev = msi_desc_to_pci_dev(irq_data_get_msi_desc(data)); | |
866 | pbus = pdev->bus; | |
867 | hbus = container_of(pbus->sysdata, struct hv_pcibus_device, sysdata); | |
868 | hpdev = get_pcichild_wslot(hbus, devfn_to_wslot(pdev->devfn)); | |
869 | if (!hpdev) | |
870 | goto return_null_message; | |
871 | ||
872 | /* Free any previous message that might have already been composed. */ | |
873 | if (data->chip_data) { | |
874 | int_desc = data->chip_data; | |
875 | data->chip_data = NULL; | |
876 | hv_int_desc_free(hpdev, int_desc); | |
877 | } | |
878 | ||
879 | int_desc = kzalloc(sizeof(*int_desc), GFP_KERNEL); | |
880 | if (!int_desc) | |
881 | goto drop_reference; | |
882 | ||
883 | memset(&ctxt, 0, sizeof(ctxt)); | |
884 | init_completion(&comp.comp_pkt.host_event); | |
885 | ctxt.pkt.completion_func = hv_pci_compose_compl; | |
886 | ctxt.pkt.compl_ctxt = ∁ | |
887 | int_pkt = (struct pci_create_interrupt *)&ctxt.pkt.message; | |
0c6045d8 | 888 | int_pkt->message_type.type = PCI_CREATE_INTERRUPT_MESSAGE; |
4daace0d JO |
889 | int_pkt->wslot.slot = hpdev->desc.win_slot.slot; |
890 | int_pkt->int_desc.vector = cfg->vector; | |
891 | int_pkt->int_desc.vector_count = 1; | |
892 | int_pkt->int_desc.delivery_mode = | |
893 | (apic->irq_delivery_mode == dest_LowestPrio) ? 1 : 0; | |
894 | ||
895 | /* | |
896 | * This bit doesn't have to work on machines with more than 64 | |
897 | * processors because Hyper-V only supports 64 in a guest. | |
898 | */ | |
899 | affinity = irq_data_get_affinity_mask(data); | |
900 | for_each_cpu_and(cpu, affinity, cpu_online_mask) { | |
901 | int_pkt->int_desc.cpu_mask |= | |
902 | (1ULL << vmbus_cpu_number_to_vp_number(cpu)); | |
903 | } | |
904 | ||
905 | ret = vmbus_sendpacket(hpdev->hbus->hdev->channel, int_pkt, | |
906 | sizeof(*int_pkt), (unsigned long)&ctxt.pkt, | |
907 | VM_PKT_DATA_INBAND, | |
908 | VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED); | |
665e2245 DC |
909 | if (ret) |
910 | goto free_int_desc; | |
911 | ||
912 | wait_for_completion(&comp.comp_pkt.host_event); | |
4daace0d JO |
913 | |
914 | if (comp.comp_pkt.completion_status < 0) { | |
915 | dev_err(&hbus->hdev->device, | |
916 | "Request for interrupt failed: 0x%x", | |
917 | comp.comp_pkt.completion_status); | |
918 | goto free_int_desc; | |
919 | } | |
920 | ||
921 | /* | |
922 | * Record the assignment so that this can be unwound later. Using | |
923 | * irq_set_chip_data() here would be appropriate, but the lock it takes | |
924 | * is already held. | |
925 | */ | |
926 | *int_desc = comp.int_desc; | |
927 | data->chip_data = int_desc; | |
928 | ||
929 | /* Pass up the result. */ | |
930 | msg->address_hi = comp.int_desc.address >> 32; | |
931 | msg->address_lo = comp.int_desc.address & 0xffffffff; | |
932 | msg->data = comp.int_desc.data; | |
933 | ||
934 | put_pcichild(hpdev, hv_pcidev_ref_by_slot); | |
935 | return; | |
936 | ||
937 | free_int_desc: | |
938 | kfree(int_desc); | |
939 | drop_reference: | |
940 | put_pcichild(hpdev, hv_pcidev_ref_by_slot); | |
941 | return_null_message: | |
942 | msg->address_hi = 0; | |
943 | msg->address_lo = 0; | |
944 | msg->data = 0; | |
945 | } | |
946 | ||
947 | /* HW Interrupt Chip Descriptor */ | |
948 | static struct irq_chip hv_msi_irq_chip = { | |
949 | .name = "Hyper-V PCIe MSI", | |
950 | .irq_compose_msi_msg = hv_compose_msi_msg, | |
951 | .irq_set_affinity = hv_set_affinity, | |
952 | .irq_ack = irq_chip_ack_parent, | |
953 | .irq_mask = hv_irq_mask, | |
954 | .irq_unmask = hv_irq_unmask, | |
955 | }; | |
956 | ||
957 | static irq_hw_number_t hv_msi_domain_ops_get_hwirq(struct msi_domain_info *info, | |
958 | msi_alloc_info_t *arg) | |
959 | { | |
960 | return arg->msi_hwirq; | |
961 | } | |
962 | ||
963 | static struct msi_domain_ops hv_msi_ops = { | |
964 | .get_hwirq = hv_msi_domain_ops_get_hwirq, | |
965 | .msi_prepare = pci_msi_prepare, | |
966 | .set_desc = pci_msi_set_desc, | |
967 | .msi_free = hv_msi_free, | |
968 | }; | |
969 | ||
970 | /** | |
971 | * hv_pcie_init_irq_domain() - Initialize IRQ domain | |
972 | * @hbus: The root PCI bus | |
973 | * | |
974 | * This function creates an IRQ domain which will be used for | |
975 | * interrupts from devices that have been passed through. These | |
976 | * devices only support MSI and MSI-X, not line-based interrupts | |
977 | * or simulations of line-based interrupts through PCIe's | |
978 | * fabric-layer messages. Because interrupts are remapped, we | |
979 | * can support multi-message MSI here. | |
980 | * | |
981 | * Return: '0' on success and error value on failure | |
982 | */ | |
983 | static int hv_pcie_init_irq_domain(struct hv_pcibus_device *hbus) | |
984 | { | |
985 | hbus->msi_info.chip = &hv_msi_irq_chip; | |
986 | hbus->msi_info.ops = &hv_msi_ops; | |
987 | hbus->msi_info.flags = (MSI_FLAG_USE_DEF_DOM_OPS | | |
988 | MSI_FLAG_USE_DEF_CHIP_OPS | MSI_FLAG_MULTI_PCI_MSI | | |
989 | MSI_FLAG_PCI_MSIX); | |
990 | hbus->msi_info.handler = handle_edge_irq; | |
991 | hbus->msi_info.handler_name = "edge"; | |
992 | hbus->msi_info.data = hbus; | |
993 | hbus->irq_domain = pci_msi_create_irq_domain(hbus->sysdata.fwnode, | |
994 | &hbus->msi_info, | |
995 | x86_vector_domain); | |
996 | if (!hbus->irq_domain) { | |
997 | dev_err(&hbus->hdev->device, | |
998 | "Failed to build an MSI IRQ domain\n"); | |
999 | return -ENODEV; | |
1000 | } | |
1001 | ||
1002 | return 0; | |
1003 | } | |
1004 | ||
1005 | /** | |
1006 | * get_bar_size() - Get the address space consumed by a BAR | |
1007 | * @bar_val: Value that a BAR returned after -1 was written | |
1008 | * to it. | |
1009 | * | |
1010 | * This function returns the size of the BAR, rounded up to 1 | |
1011 | * page. It has to be rounded up because the hypervisor's page | |
1012 | * table entry that maps the BAR into the VM can't specify an | |
1013 | * offset within a page. The invariant is that the hypervisor | |
1014 | * must place any BARs of smaller than page length at the | |
1015 | * beginning of a page. | |
1016 | * | |
1017 | * Return: Size in bytes of the consumed MMIO space. | |
1018 | */ | |
1019 | static u64 get_bar_size(u64 bar_val) | |
1020 | { | |
1021 | return round_up((1 + ~(bar_val & PCI_BASE_ADDRESS_MEM_MASK)), | |
1022 | PAGE_SIZE); | |
1023 | } | |
1024 | ||
1025 | /** | |
1026 | * survey_child_resources() - Total all MMIO requirements | |
1027 | * @hbus: Root PCI bus, as understood by this driver | |
1028 | */ | |
1029 | static void survey_child_resources(struct hv_pcibus_device *hbus) | |
1030 | { | |
1031 | struct list_head *iter; | |
1032 | struct hv_pci_dev *hpdev; | |
1033 | resource_size_t bar_size = 0; | |
1034 | unsigned long flags; | |
1035 | struct completion *event; | |
1036 | u64 bar_val; | |
1037 | int i; | |
1038 | ||
1039 | /* If nobody is waiting on the answer, don't compute it. */ | |
1040 | event = xchg(&hbus->survey_event, NULL); | |
1041 | if (!event) | |
1042 | return; | |
1043 | ||
1044 | /* If the answer has already been computed, go with it. */ | |
1045 | if (hbus->low_mmio_space || hbus->high_mmio_space) { | |
1046 | complete(event); | |
1047 | return; | |
1048 | } | |
1049 | ||
1050 | spin_lock_irqsave(&hbus->device_list_lock, flags); | |
1051 | ||
1052 | /* | |
1053 | * Due to an interesting quirk of the PCI spec, all memory regions | |
1054 | * for a child device are a power of 2 in size and aligned in memory, | |
1055 | * so it's sufficient to just add them up without tracking alignment. | |
1056 | */ | |
1057 | list_for_each(iter, &hbus->children) { | |
1058 | hpdev = container_of(iter, struct hv_pci_dev, list_entry); | |
1059 | for (i = 0; i < 6; i++) { | |
1060 | if (hpdev->probed_bar[i] & PCI_BASE_ADDRESS_SPACE_IO) | |
1061 | dev_err(&hbus->hdev->device, | |
1062 | "There's an I/O BAR in this list!\n"); | |
1063 | ||
1064 | if (hpdev->probed_bar[i] != 0) { | |
1065 | /* | |
1066 | * A probed BAR has all the upper bits set that | |
1067 | * can be changed. | |
1068 | */ | |
1069 | ||
1070 | bar_val = hpdev->probed_bar[i]; | |
1071 | if (bar_val & PCI_BASE_ADDRESS_MEM_TYPE_64) | |
1072 | bar_val |= | |
1073 | ((u64)hpdev->probed_bar[++i] << 32); | |
1074 | else | |
1075 | bar_val |= 0xffffffff00000000ULL; | |
1076 | ||
1077 | bar_size = get_bar_size(bar_val); | |
1078 | ||
1079 | if (bar_val & PCI_BASE_ADDRESS_MEM_TYPE_64) | |
1080 | hbus->high_mmio_space += bar_size; | |
1081 | else | |
1082 | hbus->low_mmio_space += bar_size; | |
1083 | } | |
1084 | } | |
1085 | } | |
1086 | ||
1087 | spin_unlock_irqrestore(&hbus->device_list_lock, flags); | |
1088 | complete(event); | |
1089 | } | |
1090 | ||
1091 | /** | |
1092 | * prepopulate_bars() - Fill in BARs with defaults | |
1093 | * @hbus: Root PCI bus, as understood by this driver | |
1094 | * | |
1095 | * The core PCI driver code seems much, much happier if the BARs | |
1096 | * for a device have values upon first scan. So fill them in. | |
1097 | * The algorithm below works down from large sizes to small, | |
1098 | * attempting to pack the assignments optimally. The assumption, | |
1099 | * enforced in other parts of the code, is that the beginning of | |
1100 | * the memory-mapped I/O space will be aligned on the largest | |
1101 | * BAR size. | |
1102 | */ | |
1103 | static void prepopulate_bars(struct hv_pcibus_device *hbus) | |
1104 | { | |
1105 | resource_size_t high_size = 0; | |
1106 | resource_size_t low_size = 0; | |
1107 | resource_size_t high_base = 0; | |
1108 | resource_size_t low_base = 0; | |
1109 | resource_size_t bar_size; | |
1110 | struct hv_pci_dev *hpdev; | |
1111 | struct list_head *iter; | |
1112 | unsigned long flags; | |
1113 | u64 bar_val; | |
1114 | u32 command; | |
1115 | bool high; | |
1116 | int i; | |
1117 | ||
1118 | if (hbus->low_mmio_space) { | |
1119 | low_size = 1ULL << (63 - __builtin_clzll(hbus->low_mmio_space)); | |
1120 | low_base = hbus->low_mmio_res->start; | |
1121 | } | |
1122 | ||
1123 | if (hbus->high_mmio_space) { | |
1124 | high_size = 1ULL << | |
1125 | (63 - __builtin_clzll(hbus->high_mmio_space)); | |
1126 | high_base = hbus->high_mmio_res->start; | |
1127 | } | |
1128 | ||
1129 | spin_lock_irqsave(&hbus->device_list_lock, flags); | |
1130 | ||
1131 | /* Pick addresses for the BARs. */ | |
1132 | do { | |
1133 | list_for_each(iter, &hbus->children) { | |
1134 | hpdev = container_of(iter, struct hv_pci_dev, | |
1135 | list_entry); | |
1136 | for (i = 0; i < 6; i++) { | |
1137 | bar_val = hpdev->probed_bar[i]; | |
1138 | if (bar_val == 0) | |
1139 | continue; | |
1140 | high = bar_val & PCI_BASE_ADDRESS_MEM_TYPE_64; | |
1141 | if (high) { | |
1142 | bar_val |= | |
1143 | ((u64)hpdev->probed_bar[i + 1] | |
1144 | << 32); | |
1145 | } else { | |
1146 | bar_val |= 0xffffffffULL << 32; | |
1147 | } | |
1148 | bar_size = get_bar_size(bar_val); | |
1149 | if (high) { | |
1150 | if (high_size != bar_size) { | |
1151 | i++; | |
1152 | continue; | |
1153 | } | |
1154 | _hv_pcifront_write_config(hpdev, | |
1155 | PCI_BASE_ADDRESS_0 + (4 * i), | |
1156 | 4, | |
1157 | (u32)(high_base & 0xffffff00)); | |
1158 | i++; | |
1159 | _hv_pcifront_write_config(hpdev, | |
1160 | PCI_BASE_ADDRESS_0 + (4 * i), | |
1161 | 4, (u32)(high_base >> 32)); | |
1162 | high_base += bar_size; | |
1163 | } else { | |
1164 | if (low_size != bar_size) | |
1165 | continue; | |
1166 | _hv_pcifront_write_config(hpdev, | |
1167 | PCI_BASE_ADDRESS_0 + (4 * i), | |
1168 | 4, | |
1169 | (u32)(low_base & 0xffffff00)); | |
1170 | low_base += bar_size; | |
1171 | } | |
1172 | } | |
1173 | if (high_size <= 1 && low_size <= 1) { | |
1174 | /* Set the memory enable bit. */ | |
1175 | _hv_pcifront_read_config(hpdev, PCI_COMMAND, 2, | |
1176 | &command); | |
1177 | command |= PCI_COMMAND_MEMORY; | |
1178 | _hv_pcifront_write_config(hpdev, PCI_COMMAND, 2, | |
1179 | command); | |
1180 | break; | |
1181 | } | |
1182 | } | |
1183 | ||
1184 | high_size >>= 1; | |
1185 | low_size >>= 1; | |
1186 | } while (high_size || low_size); | |
1187 | ||
1188 | spin_unlock_irqrestore(&hbus->device_list_lock, flags); | |
1189 | } | |
1190 | ||
1191 | /** | |
1192 | * create_root_hv_pci_bus() - Expose a new root PCI bus | |
1193 | * @hbus: Root PCI bus, as understood by this driver | |
1194 | * | |
1195 | * Return: 0 on success, -errno on failure | |
1196 | */ | |
1197 | static int create_root_hv_pci_bus(struct hv_pcibus_device *hbus) | |
1198 | { | |
1199 | /* Register the device */ | |
1200 | hbus->pci_bus = pci_create_root_bus(&hbus->hdev->device, | |
1201 | 0, /* bus number is always zero */ | |
1202 | &hv_pcifront_ops, | |
1203 | &hbus->sysdata, | |
1204 | &hbus->resources_for_children); | |
1205 | if (!hbus->pci_bus) | |
1206 | return -ENODEV; | |
1207 | ||
1208 | hbus->pci_bus->msi = &hbus->msi_chip; | |
1209 | hbus->pci_bus->msi->dev = &hbus->hdev->device; | |
1210 | ||
1211 | pci_scan_child_bus(hbus->pci_bus); | |
1212 | pci_bus_assign_resources(hbus->pci_bus); | |
1213 | pci_bus_add_devices(hbus->pci_bus); | |
1214 | hbus->state = hv_pcibus_installed; | |
1215 | return 0; | |
1216 | } | |
1217 | ||
1218 | struct q_res_req_compl { | |
1219 | struct completion host_event; | |
1220 | struct hv_pci_dev *hpdev; | |
1221 | }; | |
1222 | ||
1223 | /** | |
1224 | * q_resource_requirements() - Query Resource Requirements | |
1225 | * @context: The completion context. | |
1226 | * @resp: The response that came from the host. | |
1227 | * @resp_packet_size: The size in bytes of resp. | |
1228 | * | |
1229 | * This function is invoked on completion of a Query Resource | |
1230 | * Requirements packet. | |
1231 | */ | |
1232 | static void q_resource_requirements(void *context, struct pci_response *resp, | |
1233 | int resp_packet_size) | |
1234 | { | |
1235 | struct q_res_req_compl *completion = context; | |
1236 | struct pci_q_res_req_response *q_res_req = | |
1237 | (struct pci_q_res_req_response *)resp; | |
1238 | int i; | |
1239 | ||
1240 | if (resp->status < 0) { | |
1241 | dev_err(&completion->hpdev->hbus->hdev->device, | |
1242 | "query resource requirements failed: %x\n", | |
1243 | resp->status); | |
1244 | } else { | |
1245 | for (i = 0; i < 6; i++) { | |
1246 | completion->hpdev->probed_bar[i] = | |
1247 | q_res_req->probed_bar[i]; | |
1248 | } | |
1249 | } | |
1250 | ||
1251 | complete(&completion->host_event); | |
1252 | } | |
1253 | ||
1254 | static void get_pcichild(struct hv_pci_dev *hpdev, | |
1255 | enum hv_pcidev_ref_reason reason) | |
1256 | { | |
1257 | atomic_inc(&hpdev->refs); | |
1258 | } | |
1259 | ||
1260 | static void put_pcichild(struct hv_pci_dev *hpdev, | |
1261 | enum hv_pcidev_ref_reason reason) | |
1262 | { | |
1263 | if (atomic_dec_and_test(&hpdev->refs)) | |
1264 | kfree(hpdev); | |
1265 | } | |
1266 | ||
1267 | /** | |
1268 | * new_pcichild_device() - Create a new child device | |
1269 | * @hbus: The internal struct tracking this root PCI bus. | |
1270 | * @desc: The information supplied so far from the host | |
1271 | * about the device. | |
1272 | * | |
1273 | * This function creates the tracking structure for a new child | |
1274 | * device and kicks off the process of figuring out what it is. | |
1275 | * | |
1276 | * Return: Pointer to the new tracking struct | |
1277 | */ | |
1278 | static struct hv_pci_dev *new_pcichild_device(struct hv_pcibus_device *hbus, | |
1279 | struct pci_function_description *desc) | |
1280 | { | |
1281 | struct hv_pci_dev *hpdev; | |
1282 | struct pci_child_message *res_req; | |
1283 | struct q_res_req_compl comp_pkt; | |
8286e96d DC |
1284 | struct { |
1285 | struct pci_packet init_packet; | |
1286 | u8 buffer[sizeof(struct pci_child_message)]; | |
4daace0d JO |
1287 | } pkt; |
1288 | unsigned long flags; | |
1289 | int ret; | |
1290 | ||
1291 | hpdev = kzalloc(sizeof(*hpdev), GFP_ATOMIC); | |
1292 | if (!hpdev) | |
1293 | return NULL; | |
1294 | ||
1295 | hpdev->hbus = hbus; | |
1296 | ||
1297 | memset(&pkt, 0, sizeof(pkt)); | |
1298 | init_completion(&comp_pkt.host_event); | |
1299 | comp_pkt.hpdev = hpdev; | |
1300 | pkt.init_packet.compl_ctxt = &comp_pkt; | |
1301 | pkt.init_packet.completion_func = q_resource_requirements; | |
1302 | res_req = (struct pci_child_message *)&pkt.init_packet.message; | |
0c6045d8 | 1303 | res_req->message_type.type = PCI_QUERY_RESOURCE_REQUIREMENTS; |
4daace0d JO |
1304 | res_req->wslot.slot = desc->win_slot.slot; |
1305 | ||
1306 | ret = vmbus_sendpacket(hbus->hdev->channel, res_req, | |
1307 | sizeof(struct pci_child_message), | |
1308 | (unsigned long)&pkt.init_packet, | |
1309 | VM_PKT_DATA_INBAND, | |
1310 | VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED); | |
1311 | if (ret) | |
1312 | goto error; | |
1313 | ||
1314 | wait_for_completion(&comp_pkt.host_event); | |
1315 | ||
1316 | hpdev->desc = *desc; | |
1317 | get_pcichild(hpdev, hv_pcidev_ref_initial); | |
1318 | get_pcichild(hpdev, hv_pcidev_ref_childlist); | |
1319 | spin_lock_irqsave(&hbus->device_list_lock, flags); | |
1320 | list_add_tail(&hpdev->list_entry, &hbus->children); | |
1321 | spin_unlock_irqrestore(&hbus->device_list_lock, flags); | |
1322 | return hpdev; | |
1323 | ||
1324 | error: | |
1325 | kfree(hpdev); | |
1326 | return NULL; | |
1327 | } | |
1328 | ||
1329 | /** | |
1330 | * get_pcichild_wslot() - Find device from slot | |
1331 | * @hbus: Root PCI bus, as understood by this driver | |
1332 | * @wslot: Location on the bus | |
1333 | * | |
1334 | * This function looks up a PCI device and returns the internal | |
1335 | * representation of it. It acquires a reference on it, so that | |
1336 | * the device won't be deleted while somebody is using it. The | |
1337 | * caller is responsible for calling put_pcichild() to release | |
1338 | * this reference. | |
1339 | * | |
1340 | * Return: Internal representation of a PCI device | |
1341 | */ | |
1342 | static struct hv_pci_dev *get_pcichild_wslot(struct hv_pcibus_device *hbus, | |
1343 | u32 wslot) | |
1344 | { | |
1345 | unsigned long flags; | |
1346 | struct hv_pci_dev *iter, *hpdev = NULL; | |
1347 | ||
1348 | spin_lock_irqsave(&hbus->device_list_lock, flags); | |
1349 | list_for_each_entry(iter, &hbus->children, list_entry) { | |
1350 | if (iter->desc.win_slot.slot == wslot) { | |
1351 | hpdev = iter; | |
1352 | get_pcichild(hpdev, hv_pcidev_ref_by_slot); | |
1353 | break; | |
1354 | } | |
1355 | } | |
1356 | spin_unlock_irqrestore(&hbus->device_list_lock, flags); | |
1357 | ||
1358 | return hpdev; | |
1359 | } | |
1360 | ||
1361 | /** | |
1362 | * pci_devices_present_work() - Handle new list of child devices | |
1363 | * @work: Work struct embedded in struct hv_dr_work | |
1364 | * | |
1365 | * "Bus Relations" is the Windows term for "children of this | |
1366 | * bus." The terminology is preserved here for people trying to | |
1367 | * debug the interaction between Hyper-V and Linux. This | |
1368 | * function is called when the parent partition reports a list | |
1369 | * of functions that should be observed under this PCI Express | |
1370 | * port (bus). | |
1371 | * | |
1372 | * This function updates the list, and must tolerate being | |
1373 | * called multiple times with the same information. The typical | |
1374 | * number of child devices is one, with very atypical cases | |
1375 | * involving three or four, so the algorithms used here can be | |
1376 | * simple and inefficient. | |
1377 | * | |
1378 | * It must also treat the omission of a previously observed device as | |
1379 | * notification that the device no longer exists. | |
1380 | * | |
1381 | * Note that this function is a work item, and it may not be | |
1382 | * invoked in the order that it was queued. Back to back | |
1383 | * updates of the list of present devices may involve queuing | |
1384 | * multiple work items, and this one may run before ones that | |
1385 | * were sent later. As such, this function only does something | |
1386 | * if is the last one in the queue. | |
1387 | */ | |
1388 | static void pci_devices_present_work(struct work_struct *work) | |
1389 | { | |
1390 | u32 child_no; | |
1391 | bool found; | |
1392 | struct list_head *iter; | |
1393 | struct pci_function_description *new_desc; | |
1394 | struct hv_pci_dev *hpdev; | |
1395 | struct hv_pcibus_device *hbus; | |
1396 | struct list_head removed; | |
1397 | struct hv_dr_work *dr_wrk; | |
1398 | struct hv_dr_state *dr = NULL; | |
1399 | unsigned long flags; | |
1400 | ||
1401 | dr_wrk = container_of(work, struct hv_dr_work, wrk); | |
1402 | hbus = dr_wrk->bus; | |
1403 | kfree(dr_wrk); | |
1404 | ||
1405 | INIT_LIST_HEAD(&removed); | |
1406 | ||
1407 | if (down_interruptible(&hbus->enum_sem)) { | |
1408 | put_hvpcibus(hbus); | |
1409 | return; | |
1410 | } | |
1411 | ||
1412 | /* Pull this off the queue and process it if it was the last one. */ | |
1413 | spin_lock_irqsave(&hbus->device_list_lock, flags); | |
1414 | while (!list_empty(&hbus->dr_list)) { | |
1415 | dr = list_first_entry(&hbus->dr_list, struct hv_dr_state, | |
1416 | list_entry); | |
1417 | list_del(&dr->list_entry); | |
1418 | ||
1419 | /* Throw this away if the list still has stuff in it. */ | |
1420 | if (!list_empty(&hbus->dr_list)) { | |
1421 | kfree(dr); | |
1422 | continue; | |
1423 | } | |
1424 | } | |
1425 | spin_unlock_irqrestore(&hbus->device_list_lock, flags); | |
1426 | ||
1427 | if (!dr) { | |
1428 | up(&hbus->enum_sem); | |
1429 | put_hvpcibus(hbus); | |
1430 | return; | |
1431 | } | |
1432 | ||
1433 | /* First, mark all existing children as reported missing. */ | |
1434 | spin_lock_irqsave(&hbus->device_list_lock, flags); | |
1435 | list_for_each(iter, &hbus->children) { | |
1436 | hpdev = container_of(iter, struct hv_pci_dev, | |
1437 | list_entry); | |
1438 | hpdev->reported_missing = true; | |
1439 | } | |
1440 | spin_unlock_irqrestore(&hbus->device_list_lock, flags); | |
1441 | ||
1442 | /* Next, add back any reported devices. */ | |
1443 | for (child_no = 0; child_no < dr->device_count; child_no++) { | |
1444 | found = false; | |
1445 | new_desc = &dr->func[child_no]; | |
1446 | ||
1447 | spin_lock_irqsave(&hbus->device_list_lock, flags); | |
1448 | list_for_each(iter, &hbus->children) { | |
1449 | hpdev = container_of(iter, struct hv_pci_dev, | |
1450 | list_entry); | |
1451 | if ((hpdev->desc.win_slot.slot == | |
1452 | new_desc->win_slot.slot) && | |
1453 | (hpdev->desc.v_id == new_desc->v_id) && | |
1454 | (hpdev->desc.d_id == new_desc->d_id) && | |
1455 | (hpdev->desc.ser == new_desc->ser)) { | |
1456 | hpdev->reported_missing = false; | |
1457 | found = true; | |
1458 | } | |
1459 | } | |
1460 | spin_unlock_irqrestore(&hbus->device_list_lock, flags); | |
1461 | ||
1462 | if (!found) { | |
1463 | hpdev = new_pcichild_device(hbus, new_desc); | |
1464 | if (!hpdev) | |
1465 | dev_err(&hbus->hdev->device, | |
1466 | "couldn't record a child device.\n"); | |
1467 | } | |
1468 | } | |
1469 | ||
1470 | /* Move missing children to a list on the stack. */ | |
1471 | spin_lock_irqsave(&hbus->device_list_lock, flags); | |
1472 | do { | |
1473 | found = false; | |
1474 | list_for_each(iter, &hbus->children) { | |
1475 | hpdev = container_of(iter, struct hv_pci_dev, | |
1476 | list_entry); | |
1477 | if (hpdev->reported_missing) { | |
1478 | found = true; | |
1479 | put_pcichild(hpdev, hv_pcidev_ref_childlist); | |
4f1cb01a | 1480 | list_move_tail(&hpdev->list_entry, &removed); |
4daace0d JO |
1481 | break; |
1482 | } | |
1483 | } | |
1484 | } while (found); | |
1485 | spin_unlock_irqrestore(&hbus->device_list_lock, flags); | |
1486 | ||
1487 | /* Delete everything that should no longer exist. */ | |
1488 | while (!list_empty(&removed)) { | |
1489 | hpdev = list_first_entry(&removed, struct hv_pci_dev, | |
1490 | list_entry); | |
1491 | list_del(&hpdev->list_entry); | |
1492 | put_pcichild(hpdev, hv_pcidev_ref_initial); | |
1493 | } | |
1494 | ||
1495 | /* Tell the core to rescan bus because there may have been changes. */ | |
1496 | if (hbus->state == hv_pcibus_installed) { | |
1497 | pci_lock_rescan_remove(); | |
1498 | pci_scan_child_bus(hbus->pci_bus); | |
1499 | pci_unlock_rescan_remove(); | |
1500 | } else { | |
1501 | survey_child_resources(hbus); | |
1502 | } | |
1503 | ||
1504 | up(&hbus->enum_sem); | |
1505 | put_hvpcibus(hbus); | |
1506 | kfree(dr); | |
1507 | } | |
1508 | ||
1509 | /** | |
1510 | * hv_pci_devices_present() - Handles list of new children | |
1511 | * @hbus: Root PCI bus, as understood by this driver | |
1512 | * @relations: Packet from host listing children | |
1513 | * | |
1514 | * This function is invoked whenever a new list of devices for | |
1515 | * this bus appears. | |
1516 | */ | |
1517 | static void hv_pci_devices_present(struct hv_pcibus_device *hbus, | |
1518 | struct pci_bus_relations *relations) | |
1519 | { | |
1520 | struct hv_dr_state *dr; | |
1521 | struct hv_dr_work *dr_wrk; | |
1522 | unsigned long flags; | |
1523 | ||
1524 | dr_wrk = kzalloc(sizeof(*dr_wrk), GFP_NOWAIT); | |
1525 | if (!dr_wrk) | |
1526 | return; | |
1527 | ||
1528 | dr = kzalloc(offsetof(struct hv_dr_state, func) + | |
1529 | (sizeof(struct pci_function_description) * | |
1530 | (relations->device_count)), GFP_NOWAIT); | |
1531 | if (!dr) { | |
1532 | kfree(dr_wrk); | |
1533 | return; | |
1534 | } | |
1535 | ||
1536 | INIT_WORK(&dr_wrk->wrk, pci_devices_present_work); | |
1537 | dr_wrk->bus = hbus; | |
1538 | dr->device_count = relations->device_count; | |
1539 | if (dr->device_count != 0) { | |
1540 | memcpy(dr->func, relations->func, | |
1541 | sizeof(struct pci_function_description) * | |
1542 | dr->device_count); | |
1543 | } | |
1544 | ||
1545 | spin_lock_irqsave(&hbus->device_list_lock, flags); | |
1546 | list_add_tail(&dr->list_entry, &hbus->dr_list); | |
1547 | spin_unlock_irqrestore(&hbus->device_list_lock, flags); | |
1548 | ||
1549 | get_hvpcibus(hbus); | |
1550 | schedule_work(&dr_wrk->wrk); | |
1551 | } | |
1552 | ||
1553 | /** | |
1554 | * hv_eject_device_work() - Asynchronously handles ejection | |
1555 | * @work: Work struct embedded in internal device struct | |
1556 | * | |
1557 | * This function handles ejecting a device. Windows will | |
1558 | * attempt to gracefully eject a device, waiting 60 seconds to | |
1559 | * hear back from the guest OS that this completed successfully. | |
1560 | * If this timer expires, the device will be forcibly removed. | |
1561 | */ | |
1562 | static void hv_eject_device_work(struct work_struct *work) | |
1563 | { | |
1564 | struct pci_eject_response *ejct_pkt; | |
1565 | struct hv_pci_dev *hpdev; | |
1566 | struct pci_dev *pdev; | |
1567 | unsigned long flags; | |
1568 | int wslot; | |
1569 | struct { | |
1570 | struct pci_packet pkt; | |
0c6045d8 | 1571 | u8 buffer[sizeof(struct pci_eject_response)]; |
4daace0d JO |
1572 | } ctxt; |
1573 | ||
1574 | hpdev = container_of(work, struct hv_pci_dev, wrk); | |
1575 | ||
1576 | if (hpdev->state != hv_pcichild_ejecting) { | |
1577 | put_pcichild(hpdev, hv_pcidev_ref_pnp); | |
1578 | return; | |
1579 | } | |
1580 | ||
1581 | /* | |
1582 | * Ejection can come before or after the PCI bus has been set up, so | |
1583 | * attempt to find it and tear down the bus state, if it exists. This | |
1584 | * must be done without constructs like pci_domain_nr(hbus->pci_bus) | |
1585 | * because hbus->pci_bus may not exist yet. | |
1586 | */ | |
1587 | wslot = wslot_to_devfn(hpdev->desc.win_slot.slot); | |
1588 | pdev = pci_get_domain_bus_and_slot(hpdev->hbus->sysdata.domain, 0, | |
1589 | wslot); | |
1590 | if (pdev) { | |
1591 | pci_stop_and_remove_bus_device(pdev); | |
1592 | pci_dev_put(pdev); | |
1593 | } | |
1594 | ||
e74d2ebd DC |
1595 | spin_lock_irqsave(&hpdev->hbus->device_list_lock, flags); |
1596 | list_del(&hpdev->list_entry); | |
1597 | spin_unlock_irqrestore(&hpdev->hbus->device_list_lock, flags); | |
1598 | ||
4daace0d JO |
1599 | memset(&ctxt, 0, sizeof(ctxt)); |
1600 | ejct_pkt = (struct pci_eject_response *)&ctxt.pkt.message; | |
0c6045d8 | 1601 | ejct_pkt->message_type.type = PCI_EJECTION_COMPLETE; |
4daace0d JO |
1602 | ejct_pkt->wslot.slot = hpdev->desc.win_slot.slot; |
1603 | vmbus_sendpacket(hpdev->hbus->hdev->channel, ejct_pkt, | |
1604 | sizeof(*ejct_pkt), (unsigned long)&ctxt.pkt, | |
1605 | VM_PKT_DATA_INBAND, 0); | |
1606 | ||
4daace0d JO |
1607 | put_pcichild(hpdev, hv_pcidev_ref_childlist); |
1608 | put_pcichild(hpdev, hv_pcidev_ref_pnp); | |
1609 | put_hvpcibus(hpdev->hbus); | |
1610 | } | |
1611 | ||
1612 | /** | |
1613 | * hv_pci_eject_device() - Handles device ejection | |
1614 | * @hpdev: Internal device tracking struct | |
1615 | * | |
1616 | * This function is invoked when an ejection packet arrives. It | |
1617 | * just schedules work so that we don't re-enter the packet | |
1618 | * delivery code handling the ejection. | |
1619 | */ | |
1620 | static void hv_pci_eject_device(struct hv_pci_dev *hpdev) | |
1621 | { | |
1622 | hpdev->state = hv_pcichild_ejecting; | |
1623 | get_pcichild(hpdev, hv_pcidev_ref_pnp); | |
1624 | INIT_WORK(&hpdev->wrk, hv_eject_device_work); | |
1625 | get_hvpcibus(hpdev->hbus); | |
1626 | schedule_work(&hpdev->wrk); | |
1627 | } | |
1628 | ||
1629 | /** | |
1630 | * hv_pci_onchannelcallback() - Handles incoming packets | |
1631 | * @context: Internal bus tracking struct | |
1632 | * | |
1633 | * This function is invoked whenever the host sends a packet to | |
1634 | * this channel (which is private to this root PCI bus). | |
1635 | */ | |
1636 | static void hv_pci_onchannelcallback(void *context) | |
1637 | { | |
1638 | const int packet_size = 0x100; | |
1639 | int ret; | |
1640 | struct hv_pcibus_device *hbus = context; | |
1641 | u32 bytes_recvd; | |
1642 | u64 req_id; | |
1643 | struct vmpacket_descriptor *desc; | |
1644 | unsigned char *buffer; | |
1645 | int bufferlen = packet_size; | |
1646 | struct pci_packet *comp_packet; | |
1647 | struct pci_response *response; | |
1648 | struct pci_incoming_message *new_message; | |
1649 | struct pci_bus_relations *bus_rel; | |
1650 | struct pci_dev_incoming *dev_message; | |
1651 | struct hv_pci_dev *hpdev; | |
1652 | ||
1653 | buffer = kmalloc(bufferlen, GFP_ATOMIC); | |
1654 | if (!buffer) | |
1655 | return; | |
1656 | ||
1657 | while (1) { | |
1658 | ret = vmbus_recvpacket_raw(hbus->hdev->channel, buffer, | |
1659 | bufferlen, &bytes_recvd, &req_id); | |
1660 | ||
1661 | if (ret == -ENOBUFS) { | |
1662 | kfree(buffer); | |
1663 | /* Handle large packet */ | |
1664 | bufferlen = bytes_recvd; | |
1665 | buffer = kmalloc(bytes_recvd, GFP_ATOMIC); | |
1666 | if (!buffer) | |
1667 | return; | |
1668 | continue; | |
1669 | } | |
1670 | ||
837d741e VK |
1671 | /* Zero length indicates there are no more packets. */ |
1672 | if (ret || !bytes_recvd) | |
1673 | break; | |
1674 | ||
4daace0d JO |
1675 | /* |
1676 | * All incoming packets must be at least as large as a | |
1677 | * response. | |
1678 | */ | |
60fcdac8 | 1679 | if (bytes_recvd <= sizeof(struct pci_response)) |
837d741e | 1680 | continue; |
4daace0d JO |
1681 | desc = (struct vmpacket_descriptor *)buffer; |
1682 | ||
1683 | switch (desc->type) { | |
1684 | case VM_PKT_COMP: | |
1685 | ||
1686 | /* | |
1687 | * The host is trusted, and thus it's safe to interpret | |
1688 | * this transaction ID as a pointer. | |
1689 | */ | |
1690 | comp_packet = (struct pci_packet *)req_id; | |
1691 | response = (struct pci_response *)buffer; | |
1692 | comp_packet->completion_func(comp_packet->compl_ctxt, | |
1693 | response, | |
1694 | bytes_recvd); | |
60fcdac8 | 1695 | break; |
4daace0d JO |
1696 | |
1697 | case VM_PKT_DATA_INBAND: | |
1698 | ||
1699 | new_message = (struct pci_incoming_message *)buffer; | |
0c6045d8 | 1700 | switch (new_message->message_type.type) { |
4daace0d JO |
1701 | case PCI_BUS_RELATIONS: |
1702 | ||
1703 | bus_rel = (struct pci_bus_relations *)buffer; | |
1704 | if (bytes_recvd < | |
1705 | offsetof(struct pci_bus_relations, func) + | |
1706 | (sizeof(struct pci_function_description) * | |
1707 | (bus_rel->device_count))) { | |
1708 | dev_err(&hbus->hdev->device, | |
1709 | "bus relations too small\n"); | |
1710 | break; | |
1711 | } | |
1712 | ||
1713 | hv_pci_devices_present(hbus, bus_rel); | |
1714 | break; | |
1715 | ||
1716 | case PCI_EJECT: | |
1717 | ||
1718 | dev_message = (struct pci_dev_incoming *)buffer; | |
1719 | hpdev = get_pcichild_wslot(hbus, | |
1720 | dev_message->wslot.slot); | |
1721 | if (hpdev) { | |
1722 | hv_pci_eject_device(hpdev); | |
1723 | put_pcichild(hpdev, | |
1724 | hv_pcidev_ref_by_slot); | |
1725 | } | |
1726 | break; | |
1727 | ||
1728 | default: | |
1729 | dev_warn(&hbus->hdev->device, | |
1730 | "Unimplemented protocol message %x\n", | |
0c6045d8 | 1731 | new_message->message_type.type); |
4daace0d JO |
1732 | break; |
1733 | } | |
1734 | break; | |
1735 | ||
1736 | default: | |
1737 | dev_err(&hbus->hdev->device, | |
1738 | "unhandled packet type %d, tid %llx len %d\n", | |
1739 | desc->type, req_id, bytes_recvd); | |
1740 | break; | |
1741 | } | |
4daace0d | 1742 | } |
60fcdac8 VK |
1743 | |
1744 | kfree(buffer); | |
4daace0d JO |
1745 | } |
1746 | ||
1747 | /** | |
1748 | * hv_pci_protocol_negotiation() - Set up protocol | |
1749 | * @hdev: VMBus's tracking struct for this root PCI bus | |
1750 | * | |
1751 | * This driver is intended to support running on Windows 10 | |
1752 | * (server) and later versions. It will not run on earlier | |
1753 | * versions, as they assume that many of the operations which | |
1754 | * Linux needs accomplished with a spinlock held were done via | |
1755 | * asynchronous messaging via VMBus. Windows 10 increases the | |
1756 | * surface area of PCI emulation so that these actions can take | |
1757 | * place by suspending a virtual processor for their duration. | |
1758 | * | |
1759 | * This function negotiates the channel protocol version, | |
1760 | * failing if the host doesn't support the necessary protocol | |
1761 | * level. | |
1762 | */ | |
1763 | static int hv_pci_protocol_negotiation(struct hv_device *hdev) | |
1764 | { | |
1765 | struct pci_version_request *version_req; | |
1766 | struct hv_pci_compl comp_pkt; | |
1767 | struct pci_packet *pkt; | |
1768 | int ret; | |
1769 | ||
1770 | /* | |
1771 | * Initiate the handshake with the host and negotiate | |
1772 | * a version that the host can support. We start with the | |
1773 | * highest version number and go down if the host cannot | |
1774 | * support it. | |
1775 | */ | |
1776 | pkt = kzalloc(sizeof(*pkt) + sizeof(*version_req), GFP_KERNEL); | |
1777 | if (!pkt) | |
1778 | return -ENOMEM; | |
1779 | ||
1780 | init_completion(&comp_pkt.host_event); | |
1781 | pkt->completion_func = hv_pci_generic_compl; | |
1782 | pkt->compl_ctxt = &comp_pkt; | |
1783 | version_req = (struct pci_version_request *)&pkt->message; | |
0c6045d8 | 1784 | version_req->message_type.type = PCI_QUERY_PROTOCOL_VERSION; |
4daace0d JO |
1785 | version_req->protocol_version = PCI_PROTOCOL_VERSION_CURRENT; |
1786 | ||
1787 | ret = vmbus_sendpacket(hdev->channel, version_req, | |
1788 | sizeof(struct pci_version_request), | |
1789 | (unsigned long)pkt, VM_PKT_DATA_INBAND, | |
1790 | VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED); | |
1791 | if (ret) | |
1792 | goto exit; | |
1793 | ||
1794 | wait_for_completion(&comp_pkt.host_event); | |
1795 | ||
1796 | if (comp_pkt.completion_status < 0) { | |
1797 | dev_err(&hdev->device, | |
1798 | "PCI Pass-through VSP failed version request %x\n", | |
1799 | comp_pkt.completion_status); | |
1800 | ret = -EPROTO; | |
1801 | goto exit; | |
1802 | } | |
1803 | ||
1804 | ret = 0; | |
1805 | ||
1806 | exit: | |
1807 | kfree(pkt); | |
1808 | return ret; | |
1809 | } | |
1810 | ||
1811 | /** | |
1812 | * hv_pci_free_bridge_windows() - Release memory regions for the | |
1813 | * bus | |
1814 | * @hbus: Root PCI bus, as understood by this driver | |
1815 | */ | |
1816 | static void hv_pci_free_bridge_windows(struct hv_pcibus_device *hbus) | |
1817 | { | |
1818 | /* | |
1819 | * Set the resources back to the way they looked when they | |
1820 | * were allocated by setting IORESOURCE_BUSY again. | |
1821 | */ | |
1822 | ||
1823 | if (hbus->low_mmio_space && hbus->low_mmio_res) { | |
1824 | hbus->low_mmio_res->flags |= IORESOURCE_BUSY; | |
696ca5e8 JO |
1825 | vmbus_free_mmio(hbus->low_mmio_res->start, |
1826 | resource_size(hbus->low_mmio_res)); | |
4daace0d JO |
1827 | } |
1828 | ||
1829 | if (hbus->high_mmio_space && hbus->high_mmio_res) { | |
1830 | hbus->high_mmio_res->flags |= IORESOURCE_BUSY; | |
696ca5e8 JO |
1831 | vmbus_free_mmio(hbus->high_mmio_res->start, |
1832 | resource_size(hbus->high_mmio_res)); | |
4daace0d JO |
1833 | } |
1834 | } | |
1835 | ||
1836 | /** | |
1837 | * hv_pci_allocate_bridge_windows() - Allocate memory regions | |
1838 | * for the bus | |
1839 | * @hbus: Root PCI bus, as understood by this driver | |
1840 | * | |
1841 | * This function calls vmbus_allocate_mmio(), which is itself a | |
1842 | * bit of a compromise. Ideally, we might change the pnp layer | |
1843 | * in the kernel such that it comprehends either PCI devices | |
1844 | * which are "grandchildren of ACPI," with some intermediate bus | |
1845 | * node (in this case, VMBus) or change it such that it | |
1846 | * understands VMBus. The pnp layer, however, has been declared | |
1847 | * deprecated, and not subject to change. | |
1848 | * | |
1849 | * The workaround, implemented here, is to ask VMBus to allocate | |
1850 | * MMIO space for this bus. VMBus itself knows which ranges are | |
1851 | * appropriate by looking at its own ACPI objects. Then, after | |
1852 | * these ranges are claimed, they're modified to look like they | |
1853 | * would have looked if the ACPI and pnp code had allocated | |
1854 | * bridge windows. These descriptors have to exist in this form | |
1855 | * in order to satisfy the code which will get invoked when the | |
1856 | * endpoint PCI function driver calls request_mem_region() or | |
1857 | * request_mem_region_exclusive(). | |
1858 | * | |
1859 | * Return: 0 on success, -errno on failure | |
1860 | */ | |
1861 | static int hv_pci_allocate_bridge_windows(struct hv_pcibus_device *hbus) | |
1862 | { | |
1863 | resource_size_t align; | |
1864 | int ret; | |
1865 | ||
1866 | if (hbus->low_mmio_space) { | |
1867 | align = 1ULL << (63 - __builtin_clzll(hbus->low_mmio_space)); | |
1868 | ret = vmbus_allocate_mmio(&hbus->low_mmio_res, hbus->hdev, 0, | |
1869 | (u64)(u32)0xffffffff, | |
1870 | hbus->low_mmio_space, | |
1871 | align, false); | |
1872 | if (ret) { | |
1873 | dev_err(&hbus->hdev->device, | |
1874 | "Need %#llx of low MMIO space. Consider reconfiguring the VM.\n", | |
1875 | hbus->low_mmio_space); | |
1876 | return ret; | |
1877 | } | |
1878 | ||
1879 | /* Modify this resource to become a bridge window. */ | |
1880 | hbus->low_mmio_res->flags |= IORESOURCE_WINDOW; | |
1881 | hbus->low_mmio_res->flags &= ~IORESOURCE_BUSY; | |
1882 | pci_add_resource(&hbus->resources_for_children, | |
1883 | hbus->low_mmio_res); | |
1884 | } | |
1885 | ||
1886 | if (hbus->high_mmio_space) { | |
1887 | align = 1ULL << (63 - __builtin_clzll(hbus->high_mmio_space)); | |
1888 | ret = vmbus_allocate_mmio(&hbus->high_mmio_res, hbus->hdev, | |
1889 | 0x100000000, -1, | |
1890 | hbus->high_mmio_space, align, | |
1891 | false); | |
1892 | if (ret) { | |
1893 | dev_err(&hbus->hdev->device, | |
1894 | "Need %#llx of high MMIO space. Consider reconfiguring the VM.\n", | |
1895 | hbus->high_mmio_space); | |
1896 | goto release_low_mmio; | |
1897 | } | |
1898 | ||
1899 | /* Modify this resource to become a bridge window. */ | |
1900 | hbus->high_mmio_res->flags |= IORESOURCE_WINDOW; | |
1901 | hbus->high_mmio_res->flags &= ~IORESOURCE_BUSY; | |
1902 | pci_add_resource(&hbus->resources_for_children, | |
1903 | hbus->high_mmio_res); | |
1904 | } | |
1905 | ||
1906 | return 0; | |
1907 | ||
1908 | release_low_mmio: | |
1909 | if (hbus->low_mmio_res) { | |
696ca5e8 JO |
1910 | vmbus_free_mmio(hbus->low_mmio_res->start, |
1911 | resource_size(hbus->low_mmio_res)); | |
4daace0d JO |
1912 | } |
1913 | ||
1914 | return ret; | |
1915 | } | |
1916 | ||
1917 | /** | |
1918 | * hv_allocate_config_window() - Find MMIO space for PCI Config | |
1919 | * @hbus: Root PCI bus, as understood by this driver | |
1920 | * | |
1921 | * This function claims memory-mapped I/O space for accessing | |
1922 | * configuration space for the functions on this bus. | |
1923 | * | |
1924 | * Return: 0 on success, -errno on failure | |
1925 | */ | |
1926 | static int hv_allocate_config_window(struct hv_pcibus_device *hbus) | |
1927 | { | |
1928 | int ret; | |
1929 | ||
1930 | /* | |
1931 | * Set up a region of MMIO space to use for accessing configuration | |
1932 | * space. | |
1933 | */ | |
1934 | ret = vmbus_allocate_mmio(&hbus->mem_config, hbus->hdev, 0, -1, | |
1935 | PCI_CONFIG_MMIO_LENGTH, 0x1000, false); | |
1936 | if (ret) | |
1937 | return ret; | |
1938 | ||
1939 | /* | |
1940 | * vmbus_allocate_mmio() gets used for allocating both device endpoint | |
1941 | * resource claims (those which cannot be overlapped) and the ranges | |
1942 | * which are valid for the children of this bus, which are intended | |
1943 | * to be overlapped by those children. Set the flag on this claim | |
1944 | * meaning that this region can't be overlapped. | |
1945 | */ | |
1946 | ||
1947 | hbus->mem_config->flags |= IORESOURCE_BUSY; | |
1948 | ||
1949 | return 0; | |
1950 | } | |
1951 | ||
1952 | static void hv_free_config_window(struct hv_pcibus_device *hbus) | |
1953 | { | |
696ca5e8 | 1954 | vmbus_free_mmio(hbus->mem_config->start, PCI_CONFIG_MMIO_LENGTH); |
4daace0d JO |
1955 | } |
1956 | ||
1957 | /** | |
1958 | * hv_pci_enter_d0() - Bring the "bus" into the D0 power state | |
1959 | * @hdev: VMBus's tracking struct for this root PCI bus | |
1960 | * | |
1961 | * Return: 0 on success, -errno on failure | |
1962 | */ | |
1963 | static int hv_pci_enter_d0(struct hv_device *hdev) | |
1964 | { | |
1965 | struct hv_pcibus_device *hbus = hv_get_drvdata(hdev); | |
1966 | struct pci_bus_d0_entry *d0_entry; | |
1967 | struct hv_pci_compl comp_pkt; | |
1968 | struct pci_packet *pkt; | |
1969 | int ret; | |
1970 | ||
1971 | /* | |
1972 | * Tell the host that the bus is ready to use, and moved into the | |
1973 | * powered-on state. This includes telling the host which region | |
1974 | * of memory-mapped I/O space has been chosen for configuration space | |
1975 | * access. | |
1976 | */ | |
1977 | pkt = kzalloc(sizeof(*pkt) + sizeof(*d0_entry), GFP_KERNEL); | |
1978 | if (!pkt) | |
1979 | return -ENOMEM; | |
1980 | ||
1981 | init_completion(&comp_pkt.host_event); | |
1982 | pkt->completion_func = hv_pci_generic_compl; | |
1983 | pkt->compl_ctxt = &comp_pkt; | |
1984 | d0_entry = (struct pci_bus_d0_entry *)&pkt->message; | |
0c6045d8 | 1985 | d0_entry->message_type.type = PCI_BUS_D0ENTRY; |
4daace0d JO |
1986 | d0_entry->mmio_base = hbus->mem_config->start; |
1987 | ||
1988 | ret = vmbus_sendpacket(hdev->channel, d0_entry, sizeof(*d0_entry), | |
1989 | (unsigned long)pkt, VM_PKT_DATA_INBAND, | |
1990 | VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED); | |
1991 | if (ret) | |
1992 | goto exit; | |
1993 | ||
1994 | wait_for_completion(&comp_pkt.host_event); | |
1995 | ||
1996 | if (comp_pkt.completion_status < 0) { | |
1997 | dev_err(&hdev->device, | |
1998 | "PCI Pass-through VSP failed D0 Entry with status %x\n", | |
1999 | comp_pkt.completion_status); | |
2000 | ret = -EPROTO; | |
2001 | goto exit; | |
2002 | } | |
2003 | ||
2004 | ret = 0; | |
2005 | ||
2006 | exit: | |
2007 | kfree(pkt); | |
2008 | return ret; | |
2009 | } | |
2010 | ||
2011 | /** | |
2012 | * hv_pci_query_relations() - Ask host to send list of child | |
2013 | * devices | |
2014 | * @hdev: VMBus's tracking struct for this root PCI bus | |
2015 | * | |
2016 | * Return: 0 on success, -errno on failure | |
2017 | */ | |
2018 | static int hv_pci_query_relations(struct hv_device *hdev) | |
2019 | { | |
2020 | struct hv_pcibus_device *hbus = hv_get_drvdata(hdev); | |
2021 | struct pci_message message; | |
2022 | struct completion comp; | |
2023 | int ret; | |
2024 | ||
2025 | /* Ask the host to send along the list of child devices */ | |
2026 | init_completion(&comp); | |
2027 | if (cmpxchg(&hbus->survey_event, NULL, &comp)) | |
2028 | return -ENOTEMPTY; | |
2029 | ||
2030 | memset(&message, 0, sizeof(message)); | |
0c6045d8 | 2031 | message.type = PCI_QUERY_BUS_RELATIONS; |
4daace0d JO |
2032 | |
2033 | ret = vmbus_sendpacket(hdev->channel, &message, sizeof(message), | |
2034 | 0, VM_PKT_DATA_INBAND, 0); | |
2035 | if (ret) | |
2036 | return ret; | |
2037 | ||
2038 | wait_for_completion(&comp); | |
2039 | return 0; | |
2040 | } | |
2041 | ||
2042 | /** | |
2043 | * hv_send_resources_allocated() - Report local resource choices | |
2044 | * @hdev: VMBus's tracking struct for this root PCI bus | |
2045 | * | |
2046 | * The host OS is expecting to be sent a request as a message | |
2047 | * which contains all the resources that the device will use. | |
2048 | * The response contains those same resources, "translated" | |
2049 | * which is to say, the values which should be used by the | |
2050 | * hardware, when it delivers an interrupt. (MMIO resources are | |
2051 | * used in local terms.) This is nice for Windows, and lines up | |
2052 | * with the FDO/PDO split, which doesn't exist in Linux. Linux | |
2053 | * is deeply expecting to scan an emulated PCI configuration | |
2054 | * space. So this message is sent here only to drive the state | |
2055 | * machine on the host forward. | |
2056 | * | |
2057 | * Return: 0 on success, -errno on failure | |
2058 | */ | |
2059 | static int hv_send_resources_allocated(struct hv_device *hdev) | |
2060 | { | |
2061 | struct hv_pcibus_device *hbus = hv_get_drvdata(hdev); | |
2062 | struct pci_resources_assigned *res_assigned; | |
2063 | struct hv_pci_compl comp_pkt; | |
2064 | struct hv_pci_dev *hpdev; | |
2065 | struct pci_packet *pkt; | |
2066 | u32 wslot; | |
2067 | int ret; | |
2068 | ||
2069 | pkt = kmalloc(sizeof(*pkt) + sizeof(*res_assigned), GFP_KERNEL); | |
2070 | if (!pkt) | |
2071 | return -ENOMEM; | |
2072 | ||
2073 | ret = 0; | |
2074 | ||
2075 | for (wslot = 0; wslot < 256; wslot++) { | |
2076 | hpdev = get_pcichild_wslot(hbus, wslot); | |
2077 | if (!hpdev) | |
2078 | continue; | |
2079 | ||
2080 | memset(pkt, 0, sizeof(*pkt) + sizeof(*res_assigned)); | |
2081 | init_completion(&comp_pkt.host_event); | |
2082 | pkt->completion_func = hv_pci_generic_compl; | |
2083 | pkt->compl_ctxt = &comp_pkt; | |
4daace0d | 2084 | res_assigned = (struct pci_resources_assigned *)&pkt->message; |
0c6045d8 | 2085 | res_assigned->message_type.type = PCI_RESOURCES_ASSIGNED; |
4daace0d JO |
2086 | res_assigned->wslot.slot = hpdev->desc.win_slot.slot; |
2087 | ||
2088 | put_pcichild(hpdev, hv_pcidev_ref_by_slot); | |
2089 | ||
2090 | ret = vmbus_sendpacket( | |
2091 | hdev->channel, &pkt->message, | |
2092 | sizeof(*res_assigned), | |
2093 | (unsigned long)pkt, | |
2094 | VM_PKT_DATA_INBAND, | |
2095 | VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED); | |
2096 | if (ret) | |
2097 | break; | |
2098 | ||
2099 | wait_for_completion(&comp_pkt.host_event); | |
2100 | ||
2101 | if (comp_pkt.completion_status < 0) { | |
2102 | ret = -EPROTO; | |
2103 | dev_err(&hdev->device, | |
2104 | "resource allocated returned 0x%x", | |
2105 | comp_pkt.completion_status); | |
2106 | break; | |
2107 | } | |
2108 | } | |
2109 | ||
2110 | kfree(pkt); | |
2111 | return ret; | |
2112 | } | |
2113 | ||
2114 | /** | |
2115 | * hv_send_resources_released() - Report local resources | |
2116 | * released | |
2117 | * @hdev: VMBus's tracking struct for this root PCI bus | |
2118 | * | |
2119 | * Return: 0 on success, -errno on failure | |
2120 | */ | |
2121 | static int hv_send_resources_released(struct hv_device *hdev) | |
2122 | { | |
2123 | struct hv_pcibus_device *hbus = hv_get_drvdata(hdev); | |
2124 | struct pci_child_message pkt; | |
2125 | struct hv_pci_dev *hpdev; | |
2126 | u32 wslot; | |
2127 | int ret; | |
2128 | ||
2129 | for (wslot = 0; wslot < 256; wslot++) { | |
2130 | hpdev = get_pcichild_wslot(hbus, wslot); | |
2131 | if (!hpdev) | |
2132 | continue; | |
2133 | ||
2134 | memset(&pkt, 0, sizeof(pkt)); | |
0c6045d8 | 2135 | pkt.message_type.type = PCI_RESOURCES_RELEASED; |
4daace0d JO |
2136 | pkt.wslot.slot = hpdev->desc.win_slot.slot; |
2137 | ||
2138 | put_pcichild(hpdev, hv_pcidev_ref_by_slot); | |
2139 | ||
2140 | ret = vmbus_sendpacket(hdev->channel, &pkt, sizeof(pkt), 0, | |
2141 | VM_PKT_DATA_INBAND, 0); | |
2142 | if (ret) | |
2143 | return ret; | |
2144 | } | |
2145 | ||
2146 | return 0; | |
2147 | } | |
2148 | ||
2149 | static void get_hvpcibus(struct hv_pcibus_device *hbus) | |
2150 | { | |
2151 | atomic_inc(&hbus->remove_lock); | |
2152 | } | |
2153 | ||
2154 | static void put_hvpcibus(struct hv_pcibus_device *hbus) | |
2155 | { | |
2156 | if (atomic_dec_and_test(&hbus->remove_lock)) | |
2157 | complete(&hbus->remove_event); | |
2158 | } | |
2159 | ||
2160 | /** | |
2161 | * hv_pci_probe() - New VMBus channel probe, for a root PCI bus | |
2162 | * @hdev: VMBus's tracking struct for this root PCI bus | |
2163 | * @dev_id: Identifies the device itself | |
2164 | * | |
2165 | * Return: 0 on success, -errno on failure | |
2166 | */ | |
2167 | static int hv_pci_probe(struct hv_device *hdev, | |
2168 | const struct hv_vmbus_device_id *dev_id) | |
2169 | { | |
2170 | struct hv_pcibus_device *hbus; | |
2171 | int ret; | |
2172 | ||
2173 | hbus = kzalloc(sizeof(*hbus), GFP_KERNEL); | |
2174 | if (!hbus) | |
2175 | return -ENOMEM; | |
2176 | ||
2177 | /* | |
2178 | * The PCI bus "domain" is what is called "segment" in ACPI and | |
2179 | * other specs. Pull it from the instance ID, to get something | |
2180 | * unique. Bytes 8 and 9 are what is used in Windows guests, so | |
2181 | * do the same thing for consistency. Note that, since this code | |
2182 | * only runs in a Hyper-V VM, Hyper-V can (and does) guarantee | |
2183 | * that (1) the only domain in use for something that looks like | |
2184 | * a physical PCI bus (which is actually emulated by the | |
2185 | * hypervisor) is domain 0 and (2) there will be no overlap | |
2186 | * between domains derived from these instance IDs in the same | |
2187 | * VM. | |
2188 | */ | |
2189 | hbus->sysdata.domain = hdev->dev_instance.b[9] | | |
2190 | hdev->dev_instance.b[8] << 8; | |
2191 | ||
2192 | hbus->hdev = hdev; | |
2193 | atomic_inc(&hbus->remove_lock); | |
2194 | INIT_LIST_HEAD(&hbus->children); | |
2195 | INIT_LIST_HEAD(&hbus->dr_list); | |
2196 | INIT_LIST_HEAD(&hbus->resources_for_children); | |
2197 | spin_lock_init(&hbus->config_lock); | |
2198 | spin_lock_init(&hbus->device_list_lock); | |
0de8ce3e | 2199 | spin_lock_init(&hbus->retarget_msi_interrupt_lock); |
4daace0d JO |
2200 | sema_init(&hbus->enum_sem, 1); |
2201 | init_completion(&hbus->remove_event); | |
2202 | ||
2203 | ret = vmbus_open(hdev->channel, pci_ring_size, pci_ring_size, NULL, 0, | |
2204 | hv_pci_onchannelcallback, hbus); | |
2205 | if (ret) | |
2206 | goto free_bus; | |
2207 | ||
2208 | hv_set_drvdata(hdev, hbus); | |
2209 | ||
2210 | ret = hv_pci_protocol_negotiation(hdev); | |
2211 | if (ret) | |
2212 | goto close; | |
2213 | ||
2214 | ret = hv_allocate_config_window(hbus); | |
2215 | if (ret) | |
2216 | goto close; | |
2217 | ||
2218 | hbus->cfg_addr = ioremap(hbus->mem_config->start, | |
2219 | PCI_CONFIG_MMIO_LENGTH); | |
2220 | if (!hbus->cfg_addr) { | |
2221 | dev_err(&hdev->device, | |
2222 | "Unable to map a virtual address for config space\n"); | |
2223 | ret = -ENOMEM; | |
2224 | goto free_config; | |
2225 | } | |
2226 | ||
2227 | hbus->sysdata.fwnode = irq_domain_alloc_fwnode(hbus); | |
2228 | if (!hbus->sysdata.fwnode) { | |
2229 | ret = -ENOMEM; | |
2230 | goto unmap; | |
2231 | } | |
2232 | ||
2233 | ret = hv_pcie_init_irq_domain(hbus); | |
2234 | if (ret) | |
2235 | goto free_fwnode; | |
2236 | ||
2237 | ret = hv_pci_query_relations(hdev); | |
2238 | if (ret) | |
2239 | goto free_irq_domain; | |
2240 | ||
2241 | ret = hv_pci_enter_d0(hdev); | |
2242 | if (ret) | |
2243 | goto free_irq_domain; | |
2244 | ||
2245 | ret = hv_pci_allocate_bridge_windows(hbus); | |
2246 | if (ret) | |
2247 | goto free_irq_domain; | |
2248 | ||
2249 | ret = hv_send_resources_allocated(hdev); | |
2250 | if (ret) | |
2251 | goto free_windows; | |
2252 | ||
2253 | prepopulate_bars(hbus); | |
2254 | ||
2255 | hbus->state = hv_pcibus_probed; | |
2256 | ||
2257 | ret = create_root_hv_pci_bus(hbus); | |
2258 | if (ret) | |
2259 | goto free_windows; | |
2260 | ||
2261 | return 0; | |
2262 | ||
2263 | free_windows: | |
2264 | hv_pci_free_bridge_windows(hbus); | |
2265 | free_irq_domain: | |
2266 | irq_domain_remove(hbus->irq_domain); | |
2267 | free_fwnode: | |
2268 | irq_domain_free_fwnode(hbus->sysdata.fwnode); | |
2269 | unmap: | |
2270 | iounmap(hbus->cfg_addr); | |
2271 | free_config: | |
2272 | hv_free_config_window(hbus); | |
2273 | close: | |
2274 | vmbus_close(hdev->channel); | |
2275 | free_bus: | |
2276 | kfree(hbus); | |
2277 | return ret; | |
2278 | } | |
2279 | ||
17978524 | 2280 | static void hv_pci_bus_exit(struct hv_device *hdev) |
4daace0d | 2281 | { |
17978524 DC |
2282 | struct hv_pcibus_device *hbus = hv_get_drvdata(hdev); |
2283 | struct { | |
4daace0d | 2284 | struct pci_packet teardown_packet; |
17978524 | 2285 | u8 buffer[sizeof(struct pci_message)]; |
4daace0d JO |
2286 | } pkt; |
2287 | struct pci_bus_relations relations; | |
2288 | struct hv_pci_compl comp_pkt; | |
17978524 | 2289 | int ret; |
4daace0d | 2290 | |
17978524 DC |
2291 | /* |
2292 | * After the host sends the RESCIND_CHANNEL message, it doesn't | |
2293 | * access the per-channel ringbuffer any longer. | |
2294 | */ | |
2295 | if (hdev->channel->rescind) | |
2296 | return; | |
2297 | ||
2298 | /* Delete any children which might still exist. */ | |
2299 | memset(&relations, 0, sizeof(relations)); | |
2300 | hv_pci_devices_present(hbus, &relations); | |
2301 | ||
2302 | ret = hv_send_resources_released(hdev); | |
2303 | if (ret) | |
2304 | dev_err(&hdev->device, | |
2305 | "Couldn't send resources released packet(s)\n"); | |
4daace0d | 2306 | |
4daace0d JO |
2307 | memset(&pkt.teardown_packet, 0, sizeof(pkt.teardown_packet)); |
2308 | init_completion(&comp_pkt.host_event); | |
2309 | pkt.teardown_packet.completion_func = hv_pci_generic_compl; | |
2310 | pkt.teardown_packet.compl_ctxt = &comp_pkt; | |
0c6045d8 | 2311 | pkt.teardown_packet.message[0].type = PCI_BUS_D0EXIT; |
4daace0d JO |
2312 | |
2313 | ret = vmbus_sendpacket(hdev->channel, &pkt.teardown_packet.message, | |
2314 | sizeof(struct pci_message), | |
2315 | (unsigned long)&pkt.teardown_packet, | |
2316 | VM_PKT_DATA_INBAND, | |
2317 | VMBUS_DATA_PACKET_FLAG_COMPLETION_REQUESTED); | |
2318 | if (!ret) | |
2319 | wait_for_completion_timeout(&comp_pkt.host_event, 10 * HZ); | |
17978524 DC |
2320 | } |
2321 | ||
2322 | /** | |
2323 | * hv_pci_remove() - Remove routine for this VMBus channel | |
2324 | * @hdev: VMBus's tracking struct for this root PCI bus | |
2325 | * | |
2326 | * Return: 0 on success, -errno on failure | |
2327 | */ | |
2328 | static int hv_pci_remove(struct hv_device *hdev) | |
2329 | { | |
2330 | struct hv_pcibus_device *hbus; | |
4daace0d | 2331 | |
17978524 | 2332 | hbus = hv_get_drvdata(hdev); |
4daace0d JO |
2333 | if (hbus->state == hv_pcibus_installed) { |
2334 | /* Remove the bus from PCI's point of view. */ | |
2335 | pci_lock_rescan_remove(); | |
2336 | pci_stop_root_bus(hbus->pci_bus); | |
2337 | pci_remove_root_bus(hbus->pci_bus); | |
2338 | pci_unlock_rescan_remove(); | |
2339 | } | |
2340 | ||
17978524 | 2341 | hv_pci_bus_exit(hdev); |
deb22e5c | 2342 | |
4daace0d JO |
2343 | vmbus_close(hdev->channel); |
2344 | ||
4daace0d JO |
2345 | iounmap(hbus->cfg_addr); |
2346 | hv_free_config_window(hbus); | |
2347 | pci_free_resource_list(&hbus->resources_for_children); | |
2348 | hv_pci_free_bridge_windows(hbus); | |
2349 | irq_domain_remove(hbus->irq_domain); | |
2350 | irq_domain_free_fwnode(hbus->sysdata.fwnode); | |
2351 | put_hvpcibus(hbus); | |
2352 | wait_for_completion(&hbus->remove_event); | |
2353 | kfree(hbus); | |
2354 | return 0; | |
2355 | } | |
2356 | ||
2357 | static const struct hv_vmbus_device_id hv_pci_id_table[] = { | |
2358 | /* PCI Pass-through Class ID */ | |
2359 | /* 44C4F61D-4444-4400-9D52-802E27EDE19F */ | |
2360 | { HV_PCIE_GUID, }, | |
2361 | { }, | |
2362 | }; | |
2363 | ||
2364 | MODULE_DEVICE_TABLE(vmbus, hv_pci_id_table); | |
2365 | ||
2366 | static struct hv_driver hv_pci_drv = { | |
2367 | .name = "hv_pci", | |
2368 | .id_table = hv_pci_id_table, | |
2369 | .probe = hv_pci_probe, | |
2370 | .remove = hv_pci_remove, | |
2371 | }; | |
2372 | ||
2373 | static void __exit exit_hv_pci_drv(void) | |
2374 | { | |
2375 | vmbus_driver_unregister(&hv_pci_drv); | |
2376 | } | |
2377 | ||
2378 | static int __init init_hv_pci_drv(void) | |
2379 | { | |
2380 | return vmbus_driver_register(&hv_pci_drv); | |
2381 | } | |
2382 | ||
2383 | module_init(init_hv_pci_drv); | |
2384 | module_exit(exit_hv_pci_drv); | |
2385 | ||
2386 | MODULE_DESCRIPTION("Hyper-V PCI"); | |
2387 | MODULE_LICENSE("GPL v2"); |