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