Commit | Line | Data |
---|---|---|
65f06713 TK |
1 | // SPDX-License-Identifier: GPL-2.0+ |
2 | /* | |
3 | * Adjunct processor matrix VFIO device driver callbacks. | |
4 | * | |
5 | * Copyright IBM Corp. 2018 | |
6 | * | |
7 | * Author(s): Tony Krowiak <akrowiak@linux.ibm.com> | |
8 | * Halil Pasic <pasic@linux.ibm.com> | |
9 | * Pierre Morel <pmorel@linux.ibm.com> | |
10 | */ | |
11 | #include <linux/string.h> | |
12 | #include <linux/vfio.h> | |
13 | #include <linux/device.h> | |
14 | #include <linux/list.h> | |
15 | #include <linux/ctype.h> | |
258287c9 TK |
16 | #include <linux/bitops.h> |
17 | #include <linux/kvm_host.h> | |
18 | #include <linux/module.h> | |
68f554b7 | 19 | #include <linux/uuid.h> |
258287c9 | 20 | #include <asm/kvm.h> |
65f06713 TK |
21 | #include <asm/zcrypt.h> |
22 | ||
23 | #include "vfio_ap_private.h" | |
68f554b7 | 24 | #include "vfio_ap_debug.h" |
65f06713 TK |
25 | |
26 | #define VFIO_AP_MDEV_TYPE_HWVIRT "passthrough" | |
27 | #define VFIO_AP_MDEV_NAME_HWVIRT "VFIO AP Passthrough Device" | |
28 | ||
260f3ea1 TK |
29 | #define AP_QUEUE_ASSIGNED "assigned" |
30 | #define AP_QUEUE_UNASSIGNED "unassigned" | |
31 | #define AP_QUEUE_IN_USE "in use" | |
32 | ||
7cb7636a TK |
33 | #define MAX_RESET_CHECK_WAIT 200 /* Sleep max 200ms for reset check */ |
34 | #define AP_RESET_INTERVAL 20 /* Reset sleep interval (20ms) */ | |
35 | ||
70aeefe5 | 36 | static int vfio_ap_mdev_reset_queues(struct ap_queue_table *qtable); |
6c12a638 | 37 | static struct vfio_ap_queue *vfio_ap_find_queue(int apqn); |
eb0feefd | 38 | static const struct vfio_device_ops vfio_ap_matrix_dev_ops; |
3ba41768 | 39 | static int vfio_ap_mdev_reset_queue(struct vfio_ap_queue *q); |
ec89b55e | 40 | |
b84eb8e0 TK |
41 | /** |
42 | * get_update_locks_for_kvm: Acquire the locks required to dynamically update a | |
43 | * KVM guest's APCB in the proper order. | |
44 | * | |
45 | * @kvm: a pointer to a struct kvm object containing the KVM guest's APCB. | |
46 | * | |
47 | * The proper locking order is: | |
48 | * 1. matrix_dev->guests_lock: required to use the KVM pointer to update a KVM | |
49 | * guest's APCB. | |
50 | * 2. kvm->lock: required to update a guest's APCB | |
51 | * 3. matrix_dev->mdevs_lock: required to access data stored in a matrix_mdev | |
52 | * | |
53 | * Note: If @kvm is NULL, the KVM lock will not be taken. | |
54 | */ | |
55 | static inline void get_update_locks_for_kvm(struct kvm *kvm) | |
ec89b55e | 56 | { |
b84eb8e0 TK |
57 | mutex_lock(&matrix_dev->guests_lock); |
58 | if (kvm) | |
59 | mutex_lock(&kvm->lock); | |
60 | mutex_lock(&matrix_dev->mdevs_lock); | |
61 | } | |
ec89b55e | 62 | |
b84eb8e0 TK |
63 | /** |
64 | * release_update_locks_for_kvm: Release the locks used to dynamically update a | |
65 | * KVM guest's APCB in the proper order. | |
66 | * | |
67 | * @kvm: a pointer to a struct kvm object containing the KVM guest's APCB. | |
68 | * | |
69 | * The proper unlocking order is: | |
70 | * 1. matrix_dev->mdevs_lock | |
71 | * 2. kvm->lock | |
72 | * 3. matrix_dev->guests_lock | |
73 | * | |
74 | * Note: If @kvm is NULL, the KVM lock will not be released. | |
75 | */ | |
76 | static inline void release_update_locks_for_kvm(struct kvm *kvm) | |
77 | { | |
78 | mutex_unlock(&matrix_dev->mdevs_lock); | |
79 | if (kvm) | |
80 | mutex_unlock(&kvm->lock); | |
81 | mutex_unlock(&matrix_dev->guests_lock); | |
ec89b55e PM |
82 | } |
83 | ||
84 | /** | |
8ee13ad9 TK |
85 | * get_update_locks_for_mdev: Acquire the locks required to dynamically update a |
86 | * KVM guest's APCB in the proper order. | |
87 | * | |
88 | * @matrix_mdev: a pointer to a struct ap_matrix_mdev object containing the AP | |
89 | * configuration data to use to update a KVM guest's APCB. | |
ec89b55e | 90 | * |
8ee13ad9 TK |
91 | * The proper locking order is: |
92 | * 1. matrix_dev->guests_lock: required to use the KVM pointer to update a KVM | |
93 | * guest's APCB. | |
94 | * 2. matrix_mdev->kvm->lock: required to update a guest's APCB | |
95 | * 3. matrix_dev->mdevs_lock: required to access data stored in a matrix_mdev | |
ec89b55e | 96 | * |
8ee13ad9 TK |
97 | * Note: If @matrix_mdev is NULL or is not attached to a KVM guest, the KVM |
98 | * lock will not be taken. | |
ec89b55e | 99 | */ |
8ee13ad9 TK |
100 | static inline void get_update_locks_for_mdev(struct ap_matrix_mdev *matrix_mdev) |
101 | { | |
102 | mutex_lock(&matrix_dev->guests_lock); | |
103 | if (matrix_mdev && matrix_mdev->kvm) | |
104 | mutex_lock(&matrix_mdev->kvm->lock); | |
105 | mutex_lock(&matrix_dev->mdevs_lock); | |
106 | } | |
107 | ||
108 | /** | |
109 | * release_update_locks_for_mdev: Release the locks used to dynamically update a | |
110 | * KVM guest's APCB in the proper order. | |
111 | * | |
112 | * @matrix_mdev: a pointer to a struct ap_matrix_mdev object containing the AP | |
113 | * configuration data to use to update a KVM guest's APCB. | |
114 | * | |
115 | * The proper unlocking order is: | |
116 | * 1. matrix_dev->mdevs_lock | |
117 | * 2. matrix_mdev->kvm->lock | |
118 | * 3. matrix_dev->guests_lock | |
119 | * | |
120 | * Note: If @matrix_mdev is NULL or is not attached to a KVM guest, the KVM | |
121 | * lock will not be released. | |
122 | */ | |
123 | static inline void release_update_locks_for_mdev(struct ap_matrix_mdev *matrix_mdev) | |
124 | { | |
125 | mutex_unlock(&matrix_dev->mdevs_lock); | |
126 | if (matrix_mdev && matrix_mdev->kvm) | |
127 | mutex_unlock(&matrix_mdev->kvm->lock); | |
128 | mutex_unlock(&matrix_dev->guests_lock); | |
129 | } | |
130 | ||
2c1ee898 TK |
131 | /** |
132 | * get_update_locks_by_apqn: Find the mdev to which an APQN is assigned and | |
133 | * acquire the locks required to update the APCB of | |
134 | * the KVM guest to which the mdev is attached. | |
135 | * | |
136 | * @apqn: the APQN of a queue device. | |
137 | * | |
138 | * The proper locking order is: | |
139 | * 1. matrix_dev->guests_lock: required to use the KVM pointer to update a KVM | |
140 | * guest's APCB. | |
141 | * 2. matrix_mdev->kvm->lock: required to update a guest's APCB | |
142 | * 3. matrix_dev->mdevs_lock: required to access data stored in a matrix_mdev | |
143 | * | |
144 | * Note: If @apqn is not assigned to a matrix_mdev, the matrix_mdev->kvm->lock | |
145 | * will not be taken. | |
146 | * | |
147 | * Return: the ap_matrix_mdev object to which @apqn is assigned or NULL if @apqn | |
148 | * is not assigned to an ap_matrix_mdev. | |
149 | */ | |
150 | static struct ap_matrix_mdev *get_update_locks_by_apqn(int apqn) | |
151 | { | |
152 | struct ap_matrix_mdev *matrix_mdev; | |
153 | ||
154 | mutex_lock(&matrix_dev->guests_lock); | |
155 | ||
156 | list_for_each_entry(matrix_mdev, &matrix_dev->mdev_list, node) { | |
157 | if (test_bit_inv(AP_QID_CARD(apqn), matrix_mdev->matrix.apm) && | |
158 | test_bit_inv(AP_QID_QUEUE(apqn), matrix_mdev->matrix.aqm)) { | |
159 | if (matrix_mdev->kvm) | |
160 | mutex_lock(&matrix_mdev->kvm->lock); | |
161 | ||
162 | mutex_lock(&matrix_dev->mdevs_lock); | |
163 | ||
164 | return matrix_mdev; | |
165 | } | |
166 | } | |
167 | ||
168 | mutex_lock(&matrix_dev->mdevs_lock); | |
169 | ||
170 | return NULL; | |
171 | } | |
172 | ||
173 | /** | |
174 | * get_update_locks_for_queue: get the locks required to update the APCB of the | |
175 | * KVM guest to which the matrix mdev linked to a | |
176 | * vfio_ap_queue object is attached. | |
177 | * | |
178 | * @q: a pointer to a vfio_ap_queue object. | |
179 | * | |
180 | * The proper locking order is: | |
181 | * 1. q->matrix_dev->guests_lock: required to use the KVM pointer to update a | |
182 | * KVM guest's APCB. | |
183 | * 2. q->matrix_mdev->kvm->lock: required to update a guest's APCB | |
184 | * 3. matrix_dev->mdevs_lock: required to access data stored in matrix_mdev | |
185 | * | |
186 | * Note: if @queue is not linked to an ap_matrix_mdev object, the KVM lock | |
187 | * will not be taken. | |
188 | */ | |
189 | static inline void get_update_locks_for_queue(struct vfio_ap_queue *q) | |
190 | { | |
191 | mutex_lock(&matrix_dev->guests_lock); | |
192 | if (q->matrix_mdev && q->matrix_mdev->kvm) | |
193 | mutex_lock(&q->matrix_mdev->kvm->lock); | |
194 | mutex_lock(&matrix_dev->mdevs_lock); | |
195 | } | |
196 | ||
ec89b55e | 197 | /** |
11cb2419 TK |
198 | * vfio_ap_mdev_get_queue - retrieve a queue with a specific APQN from a |
199 | * hash table of queues assigned to a matrix mdev | |
200 | * @matrix_mdev: the matrix mdev | |
201 | * @apqn: The APQN of a queue device | |
ec89b55e | 202 | * |
11cb2419 TK |
203 | * Return: the pointer to the vfio_ap_queue struct representing the queue or |
204 | * NULL if the queue is not assigned to @matrix_mdev | |
ec89b55e | 205 | */ |
11cb2419 | 206 | static struct vfio_ap_queue *vfio_ap_mdev_get_queue( |
ec89b55e PM |
207 | struct ap_matrix_mdev *matrix_mdev, |
208 | int apqn) | |
209 | { | |
210 | struct vfio_ap_queue *q; | |
ec89b55e | 211 | |
11cb2419 TK |
212 | hash_for_each_possible(matrix_mdev->qtable.queues, q, mdev_qnode, |
213 | apqn) { | |
214 | if (q && q->apqn == apqn) | |
215 | return q; | |
216 | } | |
ec89b55e | 217 | |
11cb2419 | 218 | return NULL; |
ec89b55e PM |
219 | } |
220 | ||
221 | /** | |
0c1abe7c | 222 | * vfio_ap_wait_for_irqclear - clears the IR bit or gives up after 5 tries |
ec89b55e PM |
223 | * @apqn: The AP Queue number |
224 | * | |
225 | * Checks the IRQ bit for the status of this APQN using ap_tapq. | |
226 | * Returns if the ap_tapq function succeeded and the bit is clear. | |
227 | * Returns if ap_tapq function failed with invalid, deconfigured or | |
228 | * checkstopped AP. | |
229 | * Otherwise retries up to 5 times after waiting 20ms. | |
ec89b55e PM |
230 | */ |
231 | static void vfio_ap_wait_for_irqclear(int apqn) | |
232 | { | |
233 | struct ap_queue_status status; | |
234 | int retry = 5; | |
235 | ||
236 | do { | |
237 | status = ap_tapq(apqn, NULL); | |
238 | switch (status.response_code) { | |
239 | case AP_RESPONSE_NORMAL: | |
240 | case AP_RESPONSE_RESET_IN_PROGRESS: | |
241 | if (!status.irq_enabled) | |
242 | return; | |
0696178e | 243 | fallthrough; |
ec89b55e PM |
244 | case AP_RESPONSE_BUSY: |
245 | msleep(20); | |
246 | break; | |
247 | case AP_RESPONSE_Q_NOT_AVAIL: | |
248 | case AP_RESPONSE_DECONFIGURED: | |
249 | case AP_RESPONSE_CHECKSTOPPED: | |
250 | default: | |
251 | WARN_ONCE(1, "%s: tapq rc %02x: %04x\n", __func__, | |
252 | status.response_code, apqn); | |
253 | return; | |
254 | } | |
255 | } while (--retry); | |
256 | ||
257 | WARN_ONCE(1, "%s: tapq rc %02x: %04x could not clear IR bit\n", | |
258 | __func__, status.response_code, apqn); | |
259 | } | |
260 | ||
261 | /** | |
0c1abe7c | 262 | * vfio_ap_free_aqic_resources - free vfio_ap_queue resources |
ec89b55e PM |
263 | * @q: The vfio_ap_queue |
264 | * | |
265 | * Unregisters the ISC in the GIB when the saved ISC not invalid. | |
0c1abe7c | 266 | * Unpins the guest's page holding the NIB when it exists. |
3fad3a26 | 267 | * Resets the saved_iova and saved_isc to invalid values. |
ec89b55e PM |
268 | */ |
269 | static void vfio_ap_free_aqic_resources(struct vfio_ap_queue *q) | |
270 | { | |
6c12a638 TK |
271 | if (!q) |
272 | return; | |
273 | if (q->saved_isc != VFIO_AP_ISC_INVALID && | |
274 | !WARN_ON(!(q->matrix_mdev && q->matrix_mdev->kvm))) { | |
ec89b55e | 275 | kvm_s390_gisc_unregister(q->matrix_mdev->kvm, q->saved_isc); |
6c12a638 TK |
276 | q->saved_isc = VFIO_AP_ISC_INVALID; |
277 | } | |
3fad3a26 NC |
278 | if (q->saved_iova && !WARN_ON(!q->matrix_mdev)) { |
279 | vfio_unpin_pages(&q->matrix_mdev->vdev, q->saved_iova, 1); | |
280 | q->saved_iova = 0; | |
6c12a638 | 281 | } |
ec89b55e PM |
282 | } |
283 | ||
284 | /** | |
0c1abe7c | 285 | * vfio_ap_irq_disable - disables and clears an ap_queue interrupt |
ec89b55e PM |
286 | * @q: The vfio_ap_queue |
287 | * | |
288 | * Uses ap_aqic to disable the interruption and in case of success, reset | |
289 | * in progress or IRQ disable command already proceeded: calls | |
290 | * vfio_ap_wait_for_irqclear() to check for the IRQ bit to be clear | |
291 | * and calls vfio_ap_free_aqic_resources() to free the resources associated | |
292 | * with the AP interrupt handling. | |
293 | * | |
294 | * In the case the AP is busy, or a reset is in progress, | |
295 | * retries after 20ms, up to 5 times. | |
296 | * | |
297 | * Returns if ap_aqic function failed with invalid, deconfigured or | |
298 | * checkstopped AP. | |
0c1abe7c RD |
299 | * |
300 | * Return: &struct ap_queue_status | |
ec89b55e | 301 | */ |
6c12a638 | 302 | static struct ap_queue_status vfio_ap_irq_disable(struct vfio_ap_queue *q) |
ec89b55e | 303 | { |
ebf95e88 | 304 | union ap_qirq_ctrl aqic_gisa = { .value = 0 }; |
ec89b55e PM |
305 | struct ap_queue_status status; |
306 | int retries = 5; | |
307 | ||
308 | do { | |
10e19d49 | 309 | status = ap_aqic(q->apqn, aqic_gisa, 0); |
ec89b55e PM |
310 | switch (status.response_code) { |
311 | case AP_RESPONSE_OTHERWISE_CHANGED: | |
312 | case AP_RESPONSE_NORMAL: | |
313 | vfio_ap_wait_for_irqclear(q->apqn); | |
314 | goto end_free; | |
315 | case AP_RESPONSE_RESET_IN_PROGRESS: | |
316 | case AP_RESPONSE_BUSY: | |
317 | msleep(20); | |
318 | break; | |
319 | case AP_RESPONSE_Q_NOT_AVAIL: | |
320 | case AP_RESPONSE_DECONFIGURED: | |
321 | case AP_RESPONSE_CHECKSTOPPED: | |
322 | case AP_RESPONSE_INVALID_ADDRESS: | |
323 | default: | |
324 | /* All cases in default means AP not operational */ | |
325 | WARN_ONCE(1, "%s: ap_aqic status %d\n", __func__, | |
326 | status.response_code); | |
327 | goto end_free; | |
328 | } | |
329 | } while (retries--); | |
330 | ||
331 | WARN_ONCE(1, "%s: ap_aqic status %d\n", __func__, | |
332 | status.response_code); | |
333 | end_free: | |
334 | vfio_ap_free_aqic_resources(q); | |
335 | return status; | |
336 | } | |
337 | ||
783f0a3c TK |
338 | /** |
339 | * vfio_ap_validate_nib - validate a notification indicator byte (nib) address. | |
340 | * | |
341 | * @vcpu: the object representing the vcpu executing the PQAP(AQIC) instruction. | |
342 | * @nib: the location for storing the nib address. | |
783f0a3c TK |
343 | * |
344 | * When the PQAP(AQIC) instruction is executed, general register 2 contains the | |
345 | * address of the notification indicator byte (nib) used for IRQ notification. | |
3fad3a26 | 346 | * This function parses and validates the nib from gr2. |
783f0a3c TK |
347 | * |
348 | * Return: returns zero if the nib address is a valid; otherwise, returns | |
349 | * -EINVAL. | |
350 | */ | |
3fad3a26 | 351 | static int vfio_ap_validate_nib(struct kvm_vcpu *vcpu, dma_addr_t *nib) |
783f0a3c TK |
352 | { |
353 | *nib = vcpu->run->s.regs.gprs[2]; | |
783f0a3c | 354 | |
a64a6d23 HP |
355 | if (!*nib) |
356 | return -EINVAL; | |
3fad3a26 | 357 | if (kvm_is_error_hva(gfn_to_hva(vcpu->kvm, *nib >> PAGE_SHIFT))) |
783f0a3c TK |
358 | return -EINVAL; |
359 | ||
360 | return 0; | |
361 | } | |
362 | ||
ec89b55e | 363 | /** |
0c1abe7c | 364 | * vfio_ap_irq_enable - Enable Interruption for a APQN |
ec89b55e | 365 | * |
ec89b55e | 366 | * @q: the vfio_ap_queue holding AQIC parameters |
5ef4f710 | 367 | * @isc: the guest ISC to register with the GIB interface |
783f0a3c TK |
368 | * @vcpu: the vcpu object containing the registers specifying the parameters |
369 | * passed to the PQAP(AQIC) instruction. | |
ec89b55e PM |
370 | * |
371 | * Pin the NIB saved in *q | |
372 | * Register the guest ISC to GIB interface and retrieve the | |
373 | * host ISC to issue the host side PQAP/AQIC | |
374 | * | |
375 | * Response.status may be set to AP_RESPONSE_INVALID_ADDRESS in case the | |
376 | * vfio_pin_pages failed. | |
377 | * | |
378 | * Otherwise return the ap_queue_status returned by the ap_aqic(), | |
379 | * all retry handling will be done by the guest. | |
0c1abe7c RD |
380 | * |
381 | * Return: &struct ap_queue_status | |
ec89b55e PM |
382 | */ |
383 | static struct ap_queue_status vfio_ap_irq_enable(struct vfio_ap_queue *q, | |
384 | int isc, | |
783f0a3c | 385 | struct kvm_vcpu *vcpu) |
ec89b55e | 386 | { |
ebf95e88 | 387 | union ap_qirq_ctrl aqic_gisa = { .value = 0 }; |
ec89b55e PM |
388 | struct ap_queue_status status = {}; |
389 | struct kvm_s390_gisa *gisa; | |
34a255e6 | 390 | struct page *h_page; |
783f0a3c | 391 | int nisc; |
ec89b55e | 392 | struct kvm *kvm; |
10e19d49 | 393 | phys_addr_t h_nib; |
3fad3a26 | 394 | dma_addr_t nib; |
ec89b55e PM |
395 | int ret; |
396 | ||
783f0a3c | 397 | /* Verify that the notification indicator byte address is valid */ |
3fad3a26 NC |
398 | if (vfio_ap_validate_nib(vcpu, &nib)) { |
399 | VFIO_AP_DBF_WARN("%s: invalid NIB address: nib=%pad, apqn=%#04x\n", | |
400 | __func__, &nib, q->apqn); | |
783f0a3c TK |
401 | |
402 | status.response_code = AP_RESPONSE_INVALID_ADDRESS; | |
403 | return status; | |
404 | } | |
405 | ||
3fad3a26 | 406 | ret = vfio_pin_pages(&q->matrix_mdev->vdev, nib, 1, |
34a255e6 | 407 | IOMMU_READ | IOMMU_WRITE, &h_page); |
ec89b55e PM |
408 | switch (ret) { |
409 | case 1: | |
410 | break; | |
411 | default: | |
783f0a3c | 412 | VFIO_AP_DBF_WARN("%s: vfio_pin_pages failed: rc=%d," |
3fad3a26 NC |
413 | "nib=%pad, apqn=%#04x\n", |
414 | __func__, ret, &nib, q->apqn); | |
783f0a3c | 415 | |
ec89b55e PM |
416 | status.response_code = AP_RESPONSE_INVALID_ADDRESS; |
417 | return status; | |
418 | } | |
419 | ||
420 | kvm = q->matrix_mdev->kvm; | |
421 | gisa = kvm->arch.gisa_int.origin; | |
422 | ||
34a255e6 | 423 | h_nib = page_to_phys(h_page) | (nib & ~PAGE_MASK); |
ec89b55e | 424 | aqic_gisa.gisc = isc; |
783f0a3c TK |
425 | |
426 | nisc = kvm_s390_gisc_register(kvm, isc); | |
427 | if (nisc < 0) { | |
428 | VFIO_AP_DBF_WARN("%s: gisc registration failed: nisc=%d, isc=%d, apqn=%#04x\n", | |
429 | __func__, nisc, isc, q->apqn); | |
430 | ||
431 | status.response_code = AP_RESPONSE_INVALID_GISA; | |
432 | return status; | |
433 | } | |
434 | ||
435 | aqic_gisa.isc = nisc; | |
ec89b55e | 436 | aqic_gisa.ir = 1; |
dbec2800 | 437 | aqic_gisa.gisa = virt_to_phys(gisa) >> 4; |
ec89b55e | 438 | |
10e19d49 | 439 | status = ap_aqic(q->apqn, aqic_gisa, h_nib); |
ec89b55e PM |
440 | switch (status.response_code) { |
441 | case AP_RESPONSE_NORMAL: | |
442 | /* See if we did clear older IRQ configuration */ | |
443 | vfio_ap_free_aqic_resources(q); | |
3fad3a26 | 444 | q->saved_iova = nib; |
ec89b55e PM |
445 | q->saved_isc = isc; |
446 | break; | |
447 | case AP_RESPONSE_OTHERWISE_CHANGED: | |
448 | /* We could not modify IRQ setings: clear new configuration */ | |
3fad3a26 | 449 | vfio_unpin_pages(&q->matrix_mdev->vdev, nib, 1); |
ec89b55e PM |
450 | kvm_s390_gisc_unregister(kvm, isc); |
451 | break; | |
452 | default: | |
453 | pr_warn("%s: apqn %04x: response: %02x\n", __func__, q->apqn, | |
454 | status.response_code); | |
455 | vfio_ap_irq_disable(q); | |
456 | break; | |
457 | } | |
458 | ||
783f0a3c TK |
459 | if (status.response_code != AP_RESPONSE_NORMAL) { |
460 | VFIO_AP_DBF_WARN("%s: PQAP(AQIC) failed with status=%#02x: " | |
461 | "zone=%#x, ir=%#x, gisc=%#x, f=%#x," | |
462 | "gisa=%#x, isc=%#x, apqn=%#04x\n", | |
463 | __func__, status.response_code, | |
464 | aqic_gisa.zone, aqic_gisa.ir, aqic_gisa.gisc, | |
465 | aqic_gisa.gf, aqic_gisa.gisa, aqic_gisa.isc, | |
466 | q->apqn); | |
467 | } | |
468 | ||
ec89b55e PM |
469 | return status; |
470 | } | |
471 | ||
68f554b7 TK |
472 | /** |
473 | * vfio_ap_le_guid_to_be_uuid - convert a little endian guid array into an array | |
474 | * of big endian elements that can be passed by | |
475 | * value to an s390dbf sprintf event function to | |
476 | * format a UUID string. | |
477 | * | |
478 | * @guid: the object containing the little endian guid | |
479 | * @uuid: a six-element array of long values that can be passed by value as | |
480 | * arguments for a formatting string specifying a UUID. | |
481 | * | |
482 | * The S390 Debug Feature (s390dbf) allows the use of "%s" in the sprintf | |
483 | * event functions if the memory for the passed string is available as long as | |
484 | * the debug feature exists. Since a mediated device can be removed at any | |
485 | * time, it's name can not be used because %s passes the reference to the string | |
486 | * in memory and the reference will go stale once the device is removed . | |
487 | * | |
488 | * The s390dbf string formatting function allows a maximum of 9 arguments for a | |
489 | * message to be displayed in the 'sprintf' view. In order to use the bytes | |
490 | * comprising the mediated device's UUID to display the mediated device name, | |
491 | * they will have to be converted into an array whose elements can be passed by | |
492 | * value to sprintf. For example: | |
493 | * | |
494 | * guid array: { 83, 78, 17, 62, bb, f1, f0, 47, 91, 4d, 32, a2, 2e, 3a, 88, 04 } | |
495 | * mdev name: 62177883-f1bb-47f0-914d-32a22e3a8804 | |
496 | * array returned: { 62177883, f1bb, 47f0, 914d, 32a2, 2e3a8804 } | |
497 | * formatting string: "%08lx-%04lx-%04lx-%04lx-%02lx%04lx" | |
498 | */ | |
499 | static void vfio_ap_le_guid_to_be_uuid(guid_t *guid, unsigned long *uuid) | |
500 | { | |
501 | /* | |
502 | * The input guid is ordered in little endian, so it needs to be | |
503 | * reordered for displaying a UUID as a string. This specifies the | |
504 | * guid indices in proper order. | |
505 | */ | |
506 | uuid[0] = le32_to_cpup((__le32 *)guid); | |
507 | uuid[1] = le16_to_cpup((__le16 *)&guid->b[4]); | |
508 | uuid[2] = le16_to_cpup((__le16 *)&guid->b[6]); | |
509 | uuid[3] = *((__u16 *)&guid->b[8]); | |
510 | uuid[4] = *((__u16 *)&guid->b[10]); | |
511 | uuid[5] = *((__u32 *)&guid->b[12]); | |
512 | } | |
513 | ||
ec89b55e | 514 | /** |
0c1abe7c | 515 | * handle_pqap - PQAP instruction callback |
ec89b55e PM |
516 | * |
517 | * @vcpu: The vcpu on which we received the PQAP instruction | |
518 | * | |
519 | * Get the general register contents to initialize internal variables. | |
520 | * REG[0]: APQN | |
521 | * REG[1]: IR and ISC | |
522 | * REG[2]: NIB | |
523 | * | |
524 | * Response.status may be set to following Response Code: | |
525 | * - AP_RESPONSE_Q_NOT_AVAIL: if the queue is not available | |
526 | * - AP_RESPONSE_DECONFIGURED: if the queue is not configured | |
527 | * - AP_RESPONSE_NORMAL (0) : in case of successs | |
528 | * Check vfio_ap_setirq() and vfio_ap_clrirq() for other possible RC. | |
529 | * We take the matrix_dev lock to ensure serialization on queues and | |
530 | * mediated device access. | |
531 | * | |
0c1abe7c RD |
532 | * Return: 0 if we could handle the request inside KVM. |
533 | * Otherwise, returns -EOPNOTSUPP to let QEMU handle the fault. | |
ec89b55e PM |
534 | */ |
535 | static int handle_pqap(struct kvm_vcpu *vcpu) | |
536 | { | |
537 | uint64_t status; | |
538 | uint16_t apqn; | |
68f554b7 | 539 | unsigned long uuid[6]; |
ec89b55e PM |
540 | struct vfio_ap_queue *q; |
541 | struct ap_queue_status qstatus = { | |
542 | .response_code = AP_RESPONSE_Q_NOT_AVAIL, }; | |
543 | struct ap_matrix_mdev *matrix_mdev; | |
544 | ||
68f554b7 TK |
545 | apqn = vcpu->run->s.regs.gprs[0] & 0xffff; |
546 | ||
ec89b55e | 547 | /* If we do not use the AIV facility just go to userland */ |
68f554b7 TK |
548 | if (!(vcpu->arch.sie_block->eca & ECA_AIV)) { |
549 | VFIO_AP_DBF_WARN("%s: AIV facility not installed: apqn=0x%04x, eca=0x%04x\n", | |
550 | __func__, apqn, vcpu->arch.sie_block->eca); | |
551 | ||
ec89b55e | 552 | return -EOPNOTSUPP; |
68f554b7 | 553 | } |
ec89b55e | 554 | |
d0786556 TK |
555 | mutex_lock(&matrix_dev->mdevs_lock); |
556 | ||
68f554b7 TK |
557 | if (!vcpu->kvm->arch.crypto.pqap_hook) { |
558 | VFIO_AP_DBF_WARN("%s: PQAP(AQIC) hook not registered with the vfio_ap driver: apqn=0x%04x\n", | |
559 | __func__, apqn); | |
d0786556 | 560 | |
ec89b55e | 561 | goto out_unlock; |
68f554b7 TK |
562 | } |
563 | ||
ec89b55e PM |
564 | matrix_mdev = container_of(vcpu->kvm->arch.crypto.pqap_hook, |
565 | struct ap_matrix_mdev, pqap_hook); | |
566 | ||
0cc00c8d | 567 | /* If the there is no guest using the mdev, there is nothing to do */ |
68f554b7 TK |
568 | if (!matrix_mdev->kvm) { |
569 | vfio_ap_le_guid_to_be_uuid(&matrix_mdev->mdev->uuid, uuid); | |
570 | VFIO_AP_DBF_WARN("%s: mdev %08lx-%04lx-%04lx-%04lx-%04lx%08lx not in use: apqn=0x%04x\n", | |
571 | __func__, uuid[0], uuid[1], uuid[2], | |
572 | uuid[3], uuid[4], uuid[5], apqn); | |
0cc00c8d | 573 | goto out_unlock; |
68f554b7 | 574 | } |
0cc00c8d | 575 | |
11cb2419 | 576 | q = vfio_ap_mdev_get_queue(matrix_mdev, apqn); |
68f554b7 TK |
577 | if (!q) { |
578 | VFIO_AP_DBF_WARN("%s: Queue %02x.%04x not bound to the vfio_ap driver\n", | |
579 | __func__, AP_QID_CARD(apqn), | |
580 | AP_QID_QUEUE(apqn)); | |
ec89b55e | 581 | goto out_unlock; |
68f554b7 | 582 | } |
ec89b55e PM |
583 | |
584 | status = vcpu->run->s.regs.gprs[1]; | |
585 | ||
586 | /* If IR bit(16) is set we enable the interrupt */ | |
587 | if ((status >> (63 - 16)) & 0x01) | |
783f0a3c | 588 | qstatus = vfio_ap_irq_enable(q, status & 0x07, vcpu); |
ec89b55e PM |
589 | else |
590 | qstatus = vfio_ap_irq_disable(q); | |
591 | ||
592 | out_unlock: | |
593 | memcpy(&vcpu->run->s.regs.gprs[1], &qstatus, sizeof(qstatus)); | |
594 | vcpu->run->s.regs.gprs[1] >>= 32; | |
d0786556 | 595 | mutex_unlock(&matrix_dev->mdevs_lock); |
ec89b55e PM |
596 | return 0; |
597 | } | |
598 | ||
65f06713 TK |
599 | static void vfio_ap_matrix_init(struct ap_config_info *info, |
600 | struct ap_matrix *matrix) | |
601 | { | |
f6047040 HF |
602 | matrix->apm_max = info->apxa ? info->na : 63; |
603 | matrix->aqm_max = info->apxa ? info->nd : 15; | |
604 | matrix->adm_max = info->apxa ? info->nd : 15; | |
65f06713 TK |
605 | } |
606 | ||
51dc562a | 607 | static void vfio_ap_mdev_update_guest_apcb(struct ap_matrix_mdev *matrix_mdev) |
48cae940 | 608 | { |
51dc562a TK |
609 | if (matrix_mdev->kvm) |
610 | kvm_arch_crypto_set_masks(matrix_mdev->kvm, | |
611 | matrix_mdev->shadow_apcb.apm, | |
612 | matrix_mdev->shadow_apcb.aqm, | |
613 | matrix_mdev->shadow_apcb.adm); | |
614 | } | |
615 | ||
616 | static bool vfio_ap_mdev_filter_cdoms(struct ap_matrix_mdev *matrix_mdev) | |
617 | { | |
618 | DECLARE_BITMAP(prev_shadow_adm, AP_DOMAINS); | |
619 | ||
620 | bitmap_copy(prev_shadow_adm, matrix_mdev->shadow_apcb.adm, AP_DOMAINS); | |
48cae940 TK |
621 | bitmap_and(matrix_mdev->shadow_apcb.adm, matrix_mdev->matrix.adm, |
622 | (unsigned long *)matrix_dev->info.adm, AP_DOMAINS); | |
51dc562a TK |
623 | |
624 | return !bitmap_equal(prev_shadow_adm, matrix_mdev->shadow_apcb.adm, | |
625 | AP_DOMAINS); | |
48cae940 TK |
626 | } |
627 | ||
628 | /* | |
629 | * vfio_ap_mdev_filter_matrix - filter the APQNs assigned to the matrix mdev | |
630 | * to ensure no queue devices are passed through to | |
631 | * the guest that are not bound to the vfio_ap | |
632 | * device driver. | |
633 | * | |
634 | * @matrix_mdev: the matrix mdev whose matrix is to be filtered. | |
635 | * | |
636 | * Note: If an APQN referencing a queue device that is not bound to the vfio_ap | |
637 | * driver, its APID will be filtered from the guest's APCB. The matrix | |
638 | * structure precludes filtering an individual APQN, so its APID will be | |
639 | * filtered. | |
51dc562a TK |
640 | * |
641 | * Return: a boolean value indicating whether the KVM guest's APCB was changed | |
642 | * by the filtering or not. | |
48cae940 | 643 | */ |
51dc562a | 644 | static bool vfio_ap_mdev_filter_matrix(unsigned long *apm, unsigned long *aqm, |
48cae940 TK |
645 | struct ap_matrix_mdev *matrix_mdev) |
646 | { | |
48cae940 | 647 | unsigned long apid, apqi, apqn; |
51dc562a TK |
648 | DECLARE_BITMAP(prev_shadow_apm, AP_DEVICES); |
649 | DECLARE_BITMAP(prev_shadow_aqm, AP_DOMAINS); | |
70aeefe5 | 650 | struct vfio_ap_queue *q; |
48cae940 | 651 | |
51dc562a TK |
652 | bitmap_copy(prev_shadow_apm, matrix_mdev->shadow_apcb.apm, AP_DEVICES); |
653 | bitmap_copy(prev_shadow_aqm, matrix_mdev->shadow_apcb.aqm, AP_DOMAINS); | |
48cae940 TK |
654 | vfio_ap_matrix_init(&matrix_dev->info, &matrix_mdev->shadow_apcb); |
655 | ||
656 | /* | |
657 | * Copy the adapters, domains and control domains to the shadow_apcb | |
658 | * from the matrix mdev, but only those that are assigned to the host's | |
659 | * AP configuration. | |
660 | */ | |
661 | bitmap_and(matrix_mdev->shadow_apcb.apm, matrix_mdev->matrix.apm, | |
662 | (unsigned long *)matrix_dev->info.apm, AP_DEVICES); | |
663 | bitmap_and(matrix_mdev->shadow_apcb.aqm, matrix_mdev->matrix.aqm, | |
664 | (unsigned long *)matrix_dev->info.aqm, AP_DOMAINS); | |
665 | ||
666 | for_each_set_bit_inv(apid, apm, AP_DEVICES) { | |
667 | for_each_set_bit_inv(apqi, aqm, AP_DOMAINS) { | |
668 | /* | |
669 | * If the APQN is not bound to the vfio_ap device | |
670 | * driver, then we can't assign it to the guest's | |
671 | * AP configuration. The AP architecture won't | |
672 | * allow filtering of a single APQN, so let's filter | |
673 | * the APID since an adapter represents a physical | |
674 | * hardware device. | |
675 | */ | |
676 | apqn = AP_MKQID(apid, apqi); | |
70aeefe5 TK |
677 | q = vfio_ap_mdev_get_queue(matrix_mdev, apqn); |
678 | if (!q || q->reset_rc) { | |
48cae940 TK |
679 | clear_bit_inv(apid, |
680 | matrix_mdev->shadow_apcb.apm); | |
681 | break; | |
682 | } | |
683 | } | |
684 | } | |
51dc562a TK |
685 | |
686 | return !bitmap_equal(prev_shadow_apm, matrix_mdev->shadow_apcb.apm, | |
687 | AP_DEVICES) || | |
688 | !bitmap_equal(prev_shadow_aqm, matrix_mdev->shadow_apcb.aqm, | |
689 | AP_DOMAINS); | |
48cae940 TK |
690 | } |
691 | ||
7cb5a82e | 692 | static int vfio_ap_mdev_init_dev(struct vfio_device *vdev) |
65f06713 | 693 | { |
7cb5a82e YL |
694 | struct ap_matrix_mdev *matrix_mdev = |
695 | container_of(vdev, struct ap_matrix_mdev, vdev); | |
65f06713 | 696 | |
7cb5a82e | 697 | matrix_mdev->mdev = to_mdev_device(vdev->dev); |
65f06713 | 698 | vfio_ap_matrix_init(&matrix_dev->info, &matrix_mdev->matrix); |
1e753732 | 699 | matrix_mdev->pqap_hook = handle_pqap; |
49b0109f | 700 | vfio_ap_matrix_init(&matrix_dev->info, &matrix_mdev->shadow_apcb); |
11cb2419 | 701 | hash_init(matrix_mdev->qtable.queues); |
65f06713 | 702 | |
7cb5a82e YL |
703 | return 0; |
704 | } | |
705 | ||
706 | static int vfio_ap_mdev_probe(struct mdev_device *mdev) | |
707 | { | |
708 | struct ap_matrix_mdev *matrix_mdev; | |
709 | int ret; | |
710 | ||
711 | matrix_mdev = vfio_alloc_device(ap_matrix_mdev, vdev, &mdev->dev, | |
712 | &vfio_ap_matrix_dev_ops); | |
713 | if (IS_ERR(matrix_mdev)) | |
714 | return PTR_ERR(matrix_mdev); | |
715 | ||
c68ea0d0 | 716 | ret = vfio_register_emulated_iommu_dev(&matrix_mdev->vdev); |
eb0feefd | 717 | if (ret) |
7cb5a82e | 718 | goto err_put_vdev; |
bf48961f | 719 | matrix_mdev->req_trigger = NULL; |
eb0feefd | 720 | dev_set_drvdata(&mdev->dev, matrix_mdev); |
2c1ee898 TK |
721 | mutex_lock(&matrix_dev->mdevs_lock); |
722 | list_add(&matrix_mdev->node, &matrix_dev->mdev_list); | |
723 | mutex_unlock(&matrix_dev->mdevs_lock); | |
65f06713 | 724 | return 0; |
eb0feefd | 725 | |
7cb5a82e YL |
726 | err_put_vdev: |
727 | vfio_put_device(&matrix_mdev->vdev); | |
eb0feefd | 728 | return ret; |
65f06713 TK |
729 | } |
730 | ||
11cb2419 TK |
731 | static void vfio_ap_mdev_link_queue(struct ap_matrix_mdev *matrix_mdev, |
732 | struct vfio_ap_queue *q) | |
733 | { | |
734 | if (q) { | |
735 | q->matrix_mdev = matrix_mdev; | |
736 | hash_add(matrix_mdev->qtable.queues, &q->mdev_qnode, q->apqn); | |
737 | } | |
738 | } | |
739 | ||
740 | static void vfio_ap_mdev_link_apqn(struct ap_matrix_mdev *matrix_mdev, int apqn) | |
741 | { | |
742 | struct vfio_ap_queue *q; | |
743 | ||
744 | q = vfio_ap_find_queue(apqn); | |
745 | vfio_ap_mdev_link_queue(matrix_mdev, q); | |
746 | } | |
747 | ||
748 | static void vfio_ap_unlink_queue_fr_mdev(struct vfio_ap_queue *q) | |
749 | { | |
750 | hash_del(&q->mdev_qnode); | |
751 | } | |
752 | ||
753 | static void vfio_ap_unlink_mdev_fr_queue(struct vfio_ap_queue *q) | |
754 | { | |
755 | q->matrix_mdev = NULL; | |
756 | } | |
757 | ||
11cb2419 TK |
758 | static void vfio_ap_mdev_unlink_fr_queues(struct ap_matrix_mdev *matrix_mdev) |
759 | { | |
760 | struct vfio_ap_queue *q; | |
761 | unsigned long apid, apqi; | |
762 | ||
763 | for_each_set_bit_inv(apid, matrix_mdev->matrix.apm, AP_DEVICES) { | |
764 | for_each_set_bit_inv(apqi, matrix_mdev->matrix.aqm, | |
765 | AP_DOMAINS) { | |
766 | q = vfio_ap_mdev_get_queue(matrix_mdev, | |
767 | AP_MKQID(apid, apqi)); | |
768 | if (q) | |
769 | q->matrix_mdev = NULL; | |
770 | } | |
771 | } | |
772 | } | |
773 | ||
eb0feefd | 774 | static void vfio_ap_mdev_remove(struct mdev_device *mdev) |
65f06713 | 775 | { |
eb0feefd JG |
776 | struct ap_matrix_mdev *matrix_mdev = dev_get_drvdata(&mdev->dev); |
777 | ||
778 | vfio_unregister_group_dev(&matrix_mdev->vdev); | |
65f06713 | 779 | |
2c1ee898 | 780 | mutex_lock(&matrix_dev->guests_lock); |
d0786556 | 781 | mutex_lock(&matrix_dev->mdevs_lock); |
70aeefe5 | 782 | vfio_ap_mdev_reset_queues(&matrix_mdev->qtable); |
11cb2419 | 783 | vfio_ap_mdev_unlink_fr_queues(matrix_mdev); |
65f06713 | 784 | list_del(&matrix_mdev->node); |
d0786556 | 785 | mutex_unlock(&matrix_dev->mdevs_lock); |
2c1ee898 | 786 | mutex_unlock(&matrix_dev->guests_lock); |
7cb5a82e | 787 | vfio_put_device(&matrix_mdev->vdev); |
65f06713 TK |
788 | } |
789 | ||
e2126a73 TK |
790 | #define MDEV_SHARING_ERR "Userspace may not re-assign queue %02lx.%04lx " \ |
791 | "already assigned to %s" | |
96d152bd | 792 | |
e2126a73 TK |
793 | static void vfio_ap_mdev_log_sharing_err(struct ap_matrix_mdev *matrix_mdev, |
794 | unsigned long *apm, | |
795 | unsigned long *aqm) | |
96d152bd | 796 | { |
e2126a73 TK |
797 | unsigned long apid, apqi; |
798 | const struct device *dev = mdev_dev(matrix_mdev->mdev); | |
799 | const char *mdev_name = dev_name(dev); | |
96d152bd | 800 | |
e2126a73 TK |
801 | for_each_set_bit_inv(apid, apm, AP_DEVICES) |
802 | for_each_set_bit_inv(apqi, aqm, AP_DOMAINS) | |
803 | dev_warn(dev, MDEV_SHARING_ERR, apid, apqi, mdev_name); | |
96d152bd TK |
804 | } |
805 | ||
806 | /** | |
e2126a73 | 807 | * vfio_ap_mdev_verify_no_sharing - verify APQNs are not shared by matrix mdevs |
0c1abe7c | 808 | * |
e2126a73 TK |
809 | * @mdev_apm: mask indicating the APIDs of the APQNs to be verified |
810 | * @mdev_aqm: mask indicating the APQIs of the APQNs to be verified | |
96d152bd | 811 | * |
e2126a73 TK |
812 | * Verifies that each APQN derived from the Cartesian product of a bitmap of |
813 | * AP adapter IDs and AP queue indexes is not configured for any matrix | |
96d152bd TK |
814 | * mediated device. AP queue sharing is not allowed. |
815 | * | |
e2126a73 | 816 | * Return: 0 if the APQNs are not shared; otherwise return -EADDRINUSE. |
96d152bd | 817 | */ |
e2126a73 TK |
818 | static int vfio_ap_mdev_verify_no_sharing(unsigned long *mdev_apm, |
819 | unsigned long *mdev_aqm) | |
96d152bd | 820 | { |
e2126a73 | 821 | struct ap_matrix_mdev *matrix_mdev; |
96d152bd TK |
822 | DECLARE_BITMAP(apm, AP_DEVICES); |
823 | DECLARE_BITMAP(aqm, AP_DOMAINS); | |
824 | ||
e2126a73 TK |
825 | list_for_each_entry(matrix_mdev, &matrix_dev->mdev_list, node) { |
826 | /* | |
827 | * If the input apm and aqm are fields of the matrix_mdev | |
828 | * object, then move on to the next matrix_mdev. | |
829 | */ | |
830 | if (mdev_apm == matrix_mdev->matrix.apm && | |
831 | mdev_aqm == matrix_mdev->matrix.aqm) | |
96d152bd TK |
832 | continue; |
833 | ||
834 | memset(apm, 0, sizeof(apm)); | |
835 | memset(aqm, 0, sizeof(aqm)); | |
836 | ||
837 | /* | |
838 | * We work on full longs, as we can only exclude the leftover | |
839 | * bits in non-inverse order. The leftover is all zeros. | |
840 | */ | |
e2126a73 TK |
841 | if (!bitmap_and(apm, mdev_apm, matrix_mdev->matrix.apm, |
842 | AP_DEVICES)) | |
96d152bd TK |
843 | continue; |
844 | ||
e2126a73 TK |
845 | if (!bitmap_and(aqm, mdev_aqm, matrix_mdev->matrix.aqm, |
846 | AP_DOMAINS)) | |
96d152bd TK |
847 | continue; |
848 | ||
e2126a73 TK |
849 | vfio_ap_mdev_log_sharing_err(matrix_mdev, apm, aqm); |
850 | ||
96d152bd TK |
851 | return -EADDRINUSE; |
852 | } | |
853 | ||
854 | return 0; | |
855 | } | |
856 | ||
3f85d1df TK |
857 | /** |
858 | * vfio_ap_mdev_validate_masks - verify that the APQNs assigned to the mdev are | |
859 | * not reserved for the default zcrypt driver and | |
860 | * are not assigned to another mdev. | |
861 | * | |
862 | * @matrix_mdev: the mdev to which the APQNs being validated are assigned. | |
863 | * | |
864 | * Return: One of the following values: | |
865 | * o the error returned from the ap_apqn_in_matrix_owned_by_def_drv() function, | |
866 | * most likely -EBUSY indicating the ap_perms_mutex lock is already held. | |
867 | * o EADDRNOTAVAIL if an APQN assigned to @matrix_mdev is reserved for the | |
868 | * zcrypt default driver. | |
869 | * o EADDRINUSE if an APQN assigned to @matrix_mdev is assigned to another mdev | |
870 | * o A zero indicating validation succeeded. | |
871 | */ | |
e2126a73 TK |
872 | static int vfio_ap_mdev_validate_masks(struct ap_matrix_mdev *matrix_mdev) |
873 | { | |
874 | if (ap_apqn_in_matrix_owned_by_def_drv(matrix_mdev->matrix.apm, | |
875 | matrix_mdev->matrix.aqm)) | |
876 | return -EADDRNOTAVAIL; | |
877 | ||
878 | return vfio_ap_mdev_verify_no_sharing(matrix_mdev->matrix.apm, | |
879 | matrix_mdev->matrix.aqm); | |
880 | } | |
881 | ||
11cb2419 TK |
882 | static void vfio_ap_mdev_link_adapter(struct ap_matrix_mdev *matrix_mdev, |
883 | unsigned long apid) | |
884 | { | |
885 | unsigned long apqi; | |
886 | ||
887 | for_each_set_bit_inv(apqi, matrix_mdev->matrix.aqm, AP_DOMAINS) | |
888 | vfio_ap_mdev_link_apqn(matrix_mdev, | |
889 | AP_MKQID(apid, apqi)); | |
890 | } | |
891 | ||
96d152bd | 892 | /** |
0c1abe7c RD |
893 | * assign_adapter_store - parses the APID from @buf and sets the |
894 | * corresponding bit in the mediated matrix device's APM | |
96d152bd TK |
895 | * |
896 | * @dev: the matrix device | |
897 | * @attr: the mediated matrix device's assign_adapter attribute | |
898 | * @buf: a buffer containing the AP adapter number (APID) to | |
899 | * be assigned | |
900 | * @count: the number of bytes in @buf | |
901 | * | |
0c1abe7c | 902 | * Return: the number of bytes processed if the APID is valid; otherwise, |
96d152bd TK |
903 | * returns one of the following errors: |
904 | * | |
905 | * 1. -EINVAL | |
906 | * The APID is not a valid number | |
907 | * | |
908 | * 2. -ENODEV | |
909 | * The APID exceeds the maximum value configured for the system | |
910 | * | |
911 | * 3. -EADDRNOTAVAIL | |
912 | * An APQN derived from the cross product of the APID being assigned | |
913 | * and the APQIs previously assigned is not bound to the vfio_ap device | |
914 | * driver; or, if no APQIs have yet been assigned, the APID is not | |
915 | * contained in an APQN bound to the vfio_ap device driver. | |
916 | * | |
917 | * 4. -EADDRINUSE | |
918 | * An APQN derived from the cross product of the APID being assigned | |
919 | * and the APQIs previously assigned is being used by another mediated | |
920 | * matrix device | |
3f85d1df TK |
921 | * |
922 | * 5. -EAGAIN | |
923 | * A lock required to validate the mdev's AP configuration could not | |
924 | * be obtained. | |
96d152bd TK |
925 | */ |
926 | static ssize_t assign_adapter_store(struct device *dev, | |
927 | struct device_attribute *attr, | |
928 | const char *buf, size_t count) | |
929 | { | |
930 | int ret; | |
931 | unsigned long apid; | |
48cae940 | 932 | DECLARE_BITMAP(apm_delta, AP_DEVICES); |
eb0feefd | 933 | struct ap_matrix_mdev *matrix_mdev = dev_get_drvdata(dev); |
96d152bd | 934 | |
3f85d1df | 935 | mutex_lock(&ap_perms_mutex); |
8ee13ad9 | 936 | get_update_locks_for_mdev(matrix_mdev); |
258287c9 | 937 | |
96d152bd TK |
938 | ret = kstrtoul(buf, 0, &apid); |
939 | if (ret) | |
0cc00c8d | 940 | goto done; |
96d152bd | 941 | |
0cc00c8d TK |
942 | if (apid > matrix_mdev->matrix.apm_max) { |
943 | ret = -ENODEV; | |
944 | goto done; | |
945 | } | |
96d152bd | 946 | |
1918f2b2 TK |
947 | if (test_bit_inv(apid, matrix_mdev->matrix.apm)) { |
948 | ret = count; | |
949 | goto done; | |
950 | } | |
951 | ||
96d152bd TK |
952 | set_bit_inv(apid, matrix_mdev->matrix.apm); |
953 | ||
e2126a73 TK |
954 | ret = vfio_ap_mdev_validate_masks(matrix_mdev); |
955 | if (ret) { | |
956 | clear_bit_inv(apid, matrix_mdev->matrix.apm); | |
96d152bd | 957 | goto done; |
e2126a73 | 958 | } |
96d152bd | 959 | |
11cb2419 | 960 | vfio_ap_mdev_link_adapter(matrix_mdev, apid); |
48cae940 TK |
961 | memset(apm_delta, 0, sizeof(apm_delta)); |
962 | set_bit_inv(apid, apm_delta); | |
96d152bd | 963 | |
51dc562a TK |
964 | if (vfio_ap_mdev_filter_matrix(apm_delta, |
965 | matrix_mdev->matrix.aqm, matrix_mdev)) | |
966 | vfio_ap_mdev_update_guest_apcb(matrix_mdev); | |
96d152bd TK |
967 | |
968 | ret = count; | |
96d152bd | 969 | done: |
8ee13ad9 | 970 | release_update_locks_for_mdev(matrix_mdev); |
3f85d1df | 971 | mutex_unlock(&ap_perms_mutex); |
96d152bd TK |
972 | |
973 | return ret; | |
974 | } | |
975 | static DEVICE_ATTR_WO(assign_adapter); | |
976 | ||
70aeefe5 TK |
977 | static struct vfio_ap_queue |
978 | *vfio_ap_unlink_apqn_fr_mdev(struct ap_matrix_mdev *matrix_mdev, | |
979 | unsigned long apid, unsigned long apqi) | |
980 | { | |
981 | struct vfio_ap_queue *q = NULL; | |
982 | ||
983 | q = vfio_ap_mdev_get_queue(matrix_mdev, AP_MKQID(apid, apqi)); | |
984 | /* If the queue is assigned to the matrix mdev, unlink it. */ | |
985 | if (q) | |
986 | vfio_ap_unlink_queue_fr_mdev(q); | |
987 | ||
988 | return q; | |
989 | } | |
990 | ||
991 | /** | |
992 | * vfio_ap_mdev_unlink_adapter - unlink all queues associated with unassigned | |
993 | * adapter from the matrix mdev to which the | |
994 | * adapter was assigned. | |
995 | * @matrix_mdev: the matrix mediated device to which the adapter was assigned. | |
996 | * @apid: the APID of the unassigned adapter. | |
997 | * @qtable: table for storing queues associated with unassigned adapter. | |
998 | */ | |
11cb2419 | 999 | static void vfio_ap_mdev_unlink_adapter(struct ap_matrix_mdev *matrix_mdev, |
70aeefe5 TK |
1000 | unsigned long apid, |
1001 | struct ap_queue_table *qtable) | |
11cb2419 TK |
1002 | { |
1003 | unsigned long apqi; | |
1004 | struct vfio_ap_queue *q; | |
1005 | ||
1006 | for_each_set_bit_inv(apqi, matrix_mdev->matrix.aqm, AP_DOMAINS) { | |
70aeefe5 | 1007 | q = vfio_ap_unlink_apqn_fr_mdev(matrix_mdev, apid, apqi); |
11cb2419 | 1008 | |
70aeefe5 TK |
1009 | if (q && qtable) { |
1010 | if (test_bit_inv(apid, matrix_mdev->shadow_apcb.apm) && | |
1011 | test_bit_inv(apqi, matrix_mdev->shadow_apcb.aqm)) | |
1012 | hash_add(qtable->queues, &q->mdev_qnode, | |
1013 | q->apqn); | |
1014 | } | |
11cb2419 TK |
1015 | } |
1016 | } | |
1017 | ||
70aeefe5 TK |
1018 | static void vfio_ap_mdev_hot_unplug_adapter(struct ap_matrix_mdev *matrix_mdev, |
1019 | unsigned long apid) | |
1020 | { | |
1021 | int loop_cursor; | |
1022 | struct vfio_ap_queue *q; | |
1023 | struct ap_queue_table *qtable = kzalloc(sizeof(*qtable), GFP_KERNEL); | |
1024 | ||
1025 | hash_init(qtable->queues); | |
1026 | vfio_ap_mdev_unlink_adapter(matrix_mdev, apid, qtable); | |
1027 | ||
1028 | if (test_bit_inv(apid, matrix_mdev->shadow_apcb.apm)) { | |
1029 | clear_bit_inv(apid, matrix_mdev->shadow_apcb.apm); | |
1030 | vfio_ap_mdev_update_guest_apcb(matrix_mdev); | |
1031 | } | |
1032 | ||
1033 | vfio_ap_mdev_reset_queues(qtable); | |
1034 | ||
1035 | hash_for_each(qtable->queues, loop_cursor, q, mdev_qnode) { | |
1036 | vfio_ap_unlink_mdev_fr_queue(q); | |
1037 | hash_del(&q->mdev_qnode); | |
1038 | } | |
1039 | ||
1040 | kfree(qtable); | |
1041 | } | |
1042 | ||
96d152bd | 1043 | /** |
0c1abe7c RD |
1044 | * unassign_adapter_store - parses the APID from @buf and clears the |
1045 | * corresponding bit in the mediated matrix device's APM | |
96d152bd TK |
1046 | * |
1047 | * @dev: the matrix device | |
1048 | * @attr: the mediated matrix device's unassign_adapter attribute | |
1049 | * @buf: a buffer containing the adapter number (APID) to be unassigned | |
1050 | * @count: the number of bytes in @buf | |
1051 | * | |
0c1abe7c | 1052 | * Return: the number of bytes processed if the APID is valid; otherwise, |
96d152bd TK |
1053 | * returns one of the following errors: |
1054 | * -EINVAL if the APID is not a number | |
1055 | * -ENODEV if the APID it exceeds the maximum value configured for the | |
1056 | * system | |
1057 | */ | |
1058 | static ssize_t unassign_adapter_store(struct device *dev, | |
1059 | struct device_attribute *attr, | |
1060 | const char *buf, size_t count) | |
1061 | { | |
1062 | int ret; | |
1063 | unsigned long apid; | |
eb0feefd | 1064 | struct ap_matrix_mdev *matrix_mdev = dev_get_drvdata(dev); |
96d152bd | 1065 | |
8ee13ad9 | 1066 | get_update_locks_for_mdev(matrix_mdev); |
258287c9 | 1067 | |
96d152bd TK |
1068 | ret = kstrtoul(buf, 0, &apid); |
1069 | if (ret) | |
0cc00c8d | 1070 | goto done; |
96d152bd | 1071 | |
0cc00c8d TK |
1072 | if (apid > matrix_mdev->matrix.apm_max) { |
1073 | ret = -ENODEV; | |
1074 | goto done; | |
1075 | } | |
96d152bd | 1076 | |
1918f2b2 TK |
1077 | if (!test_bit_inv(apid, matrix_mdev->matrix.apm)) { |
1078 | ret = count; | |
1079 | goto done; | |
1080 | } | |
1081 | ||
96d152bd | 1082 | clear_bit_inv((unsigned long)apid, matrix_mdev->matrix.apm); |
70aeefe5 | 1083 | vfio_ap_mdev_hot_unplug_adapter(matrix_mdev, apid); |
0cc00c8d TK |
1084 | ret = count; |
1085 | done: | |
8ee13ad9 | 1086 | release_update_locks_for_mdev(matrix_mdev); |
0cc00c8d | 1087 | return ret; |
96d152bd | 1088 | } |
46623ab3 | 1089 | static DEVICE_ATTR_WO(unassign_adapter); |
96d152bd | 1090 | |
11cb2419 TK |
1091 | static void vfio_ap_mdev_link_domain(struct ap_matrix_mdev *matrix_mdev, |
1092 | unsigned long apqi) | |
3211da0c | 1093 | { |
3211da0c | 1094 | unsigned long apid; |
3211da0c | 1095 | |
11cb2419 TK |
1096 | for_each_set_bit_inv(apid, matrix_mdev->matrix.apm, AP_DEVICES) |
1097 | vfio_ap_mdev_link_apqn(matrix_mdev, | |
1098 | AP_MKQID(apid, apqi)); | |
3211da0c TK |
1099 | } |
1100 | ||
1101 | /** | |
0c1abe7c RD |
1102 | * assign_domain_store - parses the APQI from @buf and sets the |
1103 | * corresponding bit in the mediated matrix device's AQM | |
1104 | * | |
3211da0c TK |
1105 | * @dev: the matrix device |
1106 | * @attr: the mediated matrix device's assign_domain attribute | |
1107 | * @buf: a buffer containing the AP queue index (APQI) of the domain to | |
1108 | * be assigned | |
1109 | * @count: the number of bytes in @buf | |
1110 | * | |
0c1abe7c | 1111 | * Return: the number of bytes processed if the APQI is valid; otherwise returns |
3211da0c TK |
1112 | * one of the following errors: |
1113 | * | |
1114 | * 1. -EINVAL | |
1115 | * The APQI is not a valid number | |
1116 | * | |
1117 | * 2. -ENODEV | |
1118 | * The APQI exceeds the maximum value configured for the system | |
1119 | * | |
1120 | * 3. -EADDRNOTAVAIL | |
1121 | * An APQN derived from the cross product of the APQI being assigned | |
1122 | * and the APIDs previously assigned is not bound to the vfio_ap device | |
1123 | * driver; or, if no APIDs have yet been assigned, the APQI is not | |
1124 | * contained in an APQN bound to the vfio_ap device driver. | |
1125 | * | |
1126 | * 4. -EADDRINUSE | |
1127 | * An APQN derived from the cross product of the APQI being assigned | |
1128 | * and the APIDs previously assigned is being used by another mediated | |
1129 | * matrix device | |
3f85d1df TK |
1130 | * |
1131 | * 5. -EAGAIN | |
1132 | * The lock required to validate the mdev's AP configuration could not | |
1133 | * be obtained. | |
3211da0c TK |
1134 | */ |
1135 | static ssize_t assign_domain_store(struct device *dev, | |
1136 | struct device_attribute *attr, | |
1137 | const char *buf, size_t count) | |
1138 | { | |
1139 | int ret; | |
1140 | unsigned long apqi; | |
48cae940 | 1141 | DECLARE_BITMAP(aqm_delta, AP_DOMAINS); |
eb0feefd | 1142 | struct ap_matrix_mdev *matrix_mdev = dev_get_drvdata(dev); |
3211da0c | 1143 | |
3f85d1df | 1144 | mutex_lock(&ap_perms_mutex); |
8ee13ad9 | 1145 | get_update_locks_for_mdev(matrix_mdev); |
258287c9 | 1146 | |
3211da0c TK |
1147 | ret = kstrtoul(buf, 0, &apqi); |
1148 | if (ret) | |
0cc00c8d | 1149 | goto done; |
51dc562a TK |
1150 | |
1151 | if (apqi > matrix_mdev->matrix.aqm_max) { | |
0cc00c8d TK |
1152 | ret = -ENODEV; |
1153 | goto done; | |
1154 | } | |
3211da0c | 1155 | |
1918f2b2 TK |
1156 | if (test_bit_inv(apqi, matrix_mdev->matrix.aqm)) { |
1157 | ret = count; | |
1158 | goto done; | |
1159 | } | |
1160 | ||
3211da0c TK |
1161 | set_bit_inv(apqi, matrix_mdev->matrix.aqm); |
1162 | ||
e2126a73 TK |
1163 | ret = vfio_ap_mdev_validate_masks(matrix_mdev); |
1164 | if (ret) { | |
1165 | clear_bit_inv(apqi, matrix_mdev->matrix.aqm); | |
3211da0c | 1166 | goto done; |
e2126a73 | 1167 | } |
3211da0c | 1168 | |
11cb2419 | 1169 | vfio_ap_mdev_link_domain(matrix_mdev, apqi); |
48cae940 TK |
1170 | memset(aqm_delta, 0, sizeof(aqm_delta)); |
1171 | set_bit_inv(apqi, aqm_delta); | |
3211da0c | 1172 | |
51dc562a TK |
1173 | if (vfio_ap_mdev_filter_matrix(matrix_mdev->matrix.apm, aqm_delta, |
1174 | matrix_mdev)) | |
1175 | vfio_ap_mdev_update_guest_apcb(matrix_mdev); | |
3211da0c TK |
1176 | |
1177 | ret = count; | |
3211da0c | 1178 | done: |
8ee13ad9 | 1179 | release_update_locks_for_mdev(matrix_mdev); |
3f85d1df | 1180 | mutex_unlock(&ap_perms_mutex); |
3211da0c TK |
1181 | |
1182 | return ret; | |
1183 | } | |
46623ab3 | 1184 | static DEVICE_ATTR_WO(assign_domain); |
3211da0c | 1185 | |
11cb2419 | 1186 | static void vfio_ap_mdev_unlink_domain(struct ap_matrix_mdev *matrix_mdev, |
70aeefe5 TK |
1187 | unsigned long apqi, |
1188 | struct ap_queue_table *qtable) | |
11cb2419 TK |
1189 | { |
1190 | unsigned long apid; | |
1191 | struct vfio_ap_queue *q; | |
1192 | ||
1193 | for_each_set_bit_inv(apid, matrix_mdev->matrix.apm, AP_DEVICES) { | |
70aeefe5 TK |
1194 | q = vfio_ap_unlink_apqn_fr_mdev(matrix_mdev, apid, apqi); |
1195 | ||
1196 | if (q && qtable) { | |
1197 | if (test_bit_inv(apid, matrix_mdev->shadow_apcb.apm) && | |
1198 | test_bit_inv(apqi, matrix_mdev->shadow_apcb.aqm)) | |
1199 | hash_add(qtable->queues, &q->mdev_qnode, | |
1200 | q->apqn); | |
1201 | } | |
1202 | } | |
1203 | } | |
1204 | ||
1205 | static void vfio_ap_mdev_hot_unplug_domain(struct ap_matrix_mdev *matrix_mdev, | |
1206 | unsigned long apqi) | |
1207 | { | |
1208 | int loop_cursor; | |
1209 | struct vfio_ap_queue *q; | |
1210 | struct ap_queue_table *qtable = kzalloc(sizeof(*qtable), GFP_KERNEL); | |
1211 | ||
1212 | hash_init(qtable->queues); | |
1213 | vfio_ap_mdev_unlink_domain(matrix_mdev, apqi, qtable); | |
11cb2419 | 1214 | |
70aeefe5 TK |
1215 | if (test_bit_inv(apqi, matrix_mdev->shadow_apcb.aqm)) { |
1216 | clear_bit_inv(apqi, matrix_mdev->shadow_apcb.aqm); | |
1217 | vfio_ap_mdev_update_guest_apcb(matrix_mdev); | |
11cb2419 | 1218 | } |
70aeefe5 TK |
1219 | |
1220 | vfio_ap_mdev_reset_queues(qtable); | |
1221 | ||
1222 | hash_for_each(qtable->queues, loop_cursor, q, mdev_qnode) { | |
1223 | vfio_ap_unlink_mdev_fr_queue(q); | |
1224 | hash_del(&q->mdev_qnode); | |
1225 | } | |
1226 | ||
1227 | kfree(qtable); | |
11cb2419 | 1228 | } |
3211da0c TK |
1229 | |
1230 | /** | |
0c1abe7c RD |
1231 | * unassign_domain_store - parses the APQI from @buf and clears the |
1232 | * corresponding bit in the mediated matrix device's AQM | |
3211da0c TK |
1233 | * |
1234 | * @dev: the matrix device | |
1235 | * @attr: the mediated matrix device's unassign_domain attribute | |
1236 | * @buf: a buffer containing the AP queue index (APQI) of the domain to | |
1237 | * be unassigned | |
1238 | * @count: the number of bytes in @buf | |
1239 | * | |
0c1abe7c | 1240 | * Return: the number of bytes processed if the APQI is valid; otherwise, |
3211da0c TK |
1241 | * returns one of the following errors: |
1242 | * -EINVAL if the APQI is not a number | |
1243 | * -ENODEV if the APQI exceeds the maximum value configured for the system | |
1244 | */ | |
1245 | static ssize_t unassign_domain_store(struct device *dev, | |
1246 | struct device_attribute *attr, | |
1247 | const char *buf, size_t count) | |
1248 | { | |
1249 | int ret; | |
1250 | unsigned long apqi; | |
eb0feefd | 1251 | struct ap_matrix_mdev *matrix_mdev = dev_get_drvdata(dev); |
3211da0c | 1252 | |
8ee13ad9 | 1253 | get_update_locks_for_mdev(matrix_mdev); |
258287c9 | 1254 | |
3211da0c TK |
1255 | ret = kstrtoul(buf, 0, &apqi); |
1256 | if (ret) | |
0cc00c8d | 1257 | goto done; |
3211da0c | 1258 | |
0cc00c8d TK |
1259 | if (apqi > matrix_mdev->matrix.aqm_max) { |
1260 | ret = -ENODEV; | |
1261 | goto done; | |
1262 | } | |
3211da0c | 1263 | |
1918f2b2 TK |
1264 | if (!test_bit_inv(apqi, matrix_mdev->matrix.aqm)) { |
1265 | ret = count; | |
1266 | goto done; | |
1267 | } | |
1268 | ||
3211da0c | 1269 | clear_bit_inv((unsigned long)apqi, matrix_mdev->matrix.aqm); |
70aeefe5 | 1270 | vfio_ap_mdev_hot_unplug_domain(matrix_mdev, apqi); |
0cc00c8d | 1271 | ret = count; |
3211da0c | 1272 | |
0cc00c8d | 1273 | done: |
8ee13ad9 | 1274 | release_update_locks_for_mdev(matrix_mdev); |
0cc00c8d | 1275 | return ret; |
3211da0c | 1276 | } |
46623ab3 | 1277 | static DEVICE_ATTR_WO(unassign_domain); |
3211da0c | 1278 | |
3b1eab7f | 1279 | /** |
0c1abe7c RD |
1280 | * assign_control_domain_store - parses the domain ID from @buf and sets |
1281 | * the corresponding bit in the mediated matrix device's ADM | |
1282 | * | |
3b1eab7f TK |
1283 | * @dev: the matrix device |
1284 | * @attr: the mediated matrix device's assign_control_domain attribute | |
1285 | * @buf: a buffer containing the domain ID to be assigned | |
1286 | * @count: the number of bytes in @buf | |
1287 | * | |
0c1abe7c | 1288 | * Return: the number of bytes processed if the domain ID is valid; otherwise, |
3b1eab7f TK |
1289 | * returns one of the following errors: |
1290 | * -EINVAL if the ID is not a number | |
1291 | * -ENODEV if the ID exceeds the maximum value configured for the system | |
1292 | */ | |
1293 | static ssize_t assign_control_domain_store(struct device *dev, | |
1294 | struct device_attribute *attr, | |
1295 | const char *buf, size_t count) | |
1296 | { | |
1297 | int ret; | |
1298 | unsigned long id; | |
eb0feefd | 1299 | struct ap_matrix_mdev *matrix_mdev = dev_get_drvdata(dev); |
3b1eab7f | 1300 | |
8ee13ad9 | 1301 | get_update_locks_for_mdev(matrix_mdev); |
258287c9 | 1302 | |
3b1eab7f TK |
1303 | ret = kstrtoul(buf, 0, &id); |
1304 | if (ret) | |
0cc00c8d | 1305 | goto done; |
3b1eab7f | 1306 | |
0cc00c8d TK |
1307 | if (id > matrix_mdev->matrix.adm_max) { |
1308 | ret = -ENODEV; | |
1309 | goto done; | |
1310 | } | |
3b1eab7f | 1311 | |
1918f2b2 TK |
1312 | if (test_bit_inv(id, matrix_mdev->matrix.adm)) { |
1313 | ret = count; | |
1314 | goto done; | |
1315 | } | |
1316 | ||
3b1eab7f TK |
1317 | /* Set the bit in the ADM (bitmask) corresponding to the AP control |
1318 | * domain number (id). The bits in the mask, from most significant to | |
1319 | * least significant, correspond to IDs 0 up to the one less than the | |
1320 | * number of control domains that can be assigned. | |
1321 | */ | |
3b1eab7f | 1322 | set_bit_inv(id, matrix_mdev->matrix.adm); |
51dc562a TK |
1323 | if (vfio_ap_mdev_filter_cdoms(matrix_mdev)) |
1324 | vfio_ap_mdev_update_guest_apcb(matrix_mdev); | |
1325 | ||
0cc00c8d TK |
1326 | ret = count; |
1327 | done: | |
8ee13ad9 | 1328 | release_update_locks_for_mdev(matrix_mdev); |
0cc00c8d | 1329 | return ret; |
3b1eab7f | 1330 | } |
46623ab3 | 1331 | static DEVICE_ATTR_WO(assign_control_domain); |
3b1eab7f TK |
1332 | |
1333 | /** | |
0c1abe7c RD |
1334 | * unassign_control_domain_store - parses the domain ID from @buf and |
1335 | * clears the corresponding bit in the mediated matrix device's ADM | |
3b1eab7f TK |
1336 | * |
1337 | * @dev: the matrix device | |
1338 | * @attr: the mediated matrix device's unassign_control_domain attribute | |
1339 | * @buf: a buffer containing the domain ID to be unassigned | |
1340 | * @count: the number of bytes in @buf | |
1341 | * | |
0c1abe7c | 1342 | * Return: the number of bytes processed if the domain ID is valid; otherwise, |
3b1eab7f TK |
1343 | * returns one of the following errors: |
1344 | * -EINVAL if the ID is not a number | |
1345 | * -ENODEV if the ID exceeds the maximum value configured for the system | |
1346 | */ | |
1347 | static ssize_t unassign_control_domain_store(struct device *dev, | |
1348 | struct device_attribute *attr, | |
1349 | const char *buf, size_t count) | |
1350 | { | |
1351 | int ret; | |
1352 | unsigned long domid; | |
eb0feefd | 1353 | struct ap_matrix_mdev *matrix_mdev = dev_get_drvdata(dev); |
3b1eab7f | 1354 | |
8ee13ad9 | 1355 | get_update_locks_for_mdev(matrix_mdev); |
258287c9 | 1356 | |
3b1eab7f TK |
1357 | ret = kstrtoul(buf, 0, &domid); |
1358 | if (ret) | |
0cc00c8d | 1359 | goto done; |
51dc562a TK |
1360 | |
1361 | if (domid > matrix_mdev->matrix.adm_max) { | |
0cc00c8d TK |
1362 | ret = -ENODEV; |
1363 | goto done; | |
1364 | } | |
3b1eab7f | 1365 | |
1918f2b2 TK |
1366 | if (!test_bit_inv(domid, matrix_mdev->matrix.adm)) { |
1367 | ret = count; | |
1368 | goto done; | |
1369 | } | |
1370 | ||
3b1eab7f | 1371 | clear_bit_inv(domid, matrix_mdev->matrix.adm); |
48cae940 | 1372 | |
51dc562a | 1373 | if (test_bit_inv(domid, matrix_mdev->shadow_apcb.adm)) { |
48cae940 | 1374 | clear_bit_inv(domid, matrix_mdev->shadow_apcb.adm); |
51dc562a TK |
1375 | vfio_ap_mdev_update_guest_apcb(matrix_mdev); |
1376 | } | |
48cae940 | 1377 | |
0cc00c8d TK |
1378 | ret = count; |
1379 | done: | |
8ee13ad9 | 1380 | release_update_locks_for_mdev(matrix_mdev); |
0cc00c8d | 1381 | return ret; |
3b1eab7f | 1382 | } |
46623ab3 | 1383 | static DEVICE_ATTR_WO(unassign_control_domain); |
3b1eab7f TK |
1384 | |
1385 | static ssize_t control_domains_show(struct device *dev, | |
1386 | struct device_attribute *dev_attr, | |
1387 | char *buf) | |
1388 | { | |
1389 | unsigned long id; | |
1390 | int nchars = 0; | |
1391 | int n; | |
1392 | char *bufpos = buf; | |
eb0feefd | 1393 | struct ap_matrix_mdev *matrix_mdev = dev_get_drvdata(dev); |
3b1eab7f TK |
1394 | unsigned long max_domid = matrix_mdev->matrix.adm_max; |
1395 | ||
d0786556 | 1396 | mutex_lock(&matrix_dev->mdevs_lock); |
3b1eab7f TK |
1397 | for_each_set_bit_inv(id, matrix_mdev->matrix.adm, max_domid + 1) { |
1398 | n = sprintf(bufpos, "%04lx\n", id); | |
1399 | bufpos += n; | |
1400 | nchars += n; | |
1401 | } | |
d0786556 | 1402 | mutex_unlock(&matrix_dev->mdevs_lock); |
3b1eab7f TK |
1403 | |
1404 | return nchars; | |
1405 | } | |
46623ab3 | 1406 | static DEVICE_ATTR_RO(control_domains); |
3b1eab7f | 1407 | |
f7f795c5 | 1408 | static ssize_t vfio_ap_mdev_matrix_show(struct ap_matrix *matrix, char *buf) |
81b2b4b7 | 1409 | { |
81b2b4b7 TK |
1410 | char *bufpos = buf; |
1411 | unsigned long apid; | |
1412 | unsigned long apqi; | |
1413 | unsigned long apid1; | |
1414 | unsigned long apqi1; | |
f7f795c5 TK |
1415 | unsigned long napm_bits = matrix->apm_max + 1; |
1416 | unsigned long naqm_bits = matrix->aqm_max + 1; | |
81b2b4b7 TK |
1417 | int nchars = 0; |
1418 | int n; | |
1419 | ||
f7f795c5 TK |
1420 | apid1 = find_first_bit_inv(matrix->apm, napm_bits); |
1421 | apqi1 = find_first_bit_inv(matrix->aqm, naqm_bits); | |
81b2b4b7 TK |
1422 | |
1423 | if ((apid1 < napm_bits) && (apqi1 < naqm_bits)) { | |
f7f795c5 TK |
1424 | for_each_set_bit_inv(apid, matrix->apm, napm_bits) { |
1425 | for_each_set_bit_inv(apqi, matrix->aqm, | |
81b2b4b7 TK |
1426 | naqm_bits) { |
1427 | n = sprintf(bufpos, "%02lx.%04lx\n", apid, | |
1428 | apqi); | |
1429 | bufpos += n; | |
1430 | nchars += n; | |
1431 | } | |
1432 | } | |
1433 | } else if (apid1 < napm_bits) { | |
f7f795c5 | 1434 | for_each_set_bit_inv(apid, matrix->apm, napm_bits) { |
81b2b4b7 TK |
1435 | n = sprintf(bufpos, "%02lx.\n", apid); |
1436 | bufpos += n; | |
1437 | nchars += n; | |
1438 | } | |
1439 | } else if (apqi1 < naqm_bits) { | |
f7f795c5 | 1440 | for_each_set_bit_inv(apqi, matrix->aqm, naqm_bits) { |
81b2b4b7 TK |
1441 | n = sprintf(bufpos, ".%04lx\n", apqi); |
1442 | bufpos += n; | |
1443 | nchars += n; | |
1444 | } | |
1445 | } | |
1446 | ||
f7f795c5 TK |
1447 | return nchars; |
1448 | } | |
1449 | ||
1450 | static ssize_t matrix_show(struct device *dev, struct device_attribute *attr, | |
1451 | char *buf) | |
1452 | { | |
1453 | ssize_t nchars; | |
1454 | struct ap_matrix_mdev *matrix_mdev = dev_get_drvdata(dev); | |
1455 | ||
1456 | mutex_lock(&matrix_dev->mdevs_lock); | |
1457 | nchars = vfio_ap_mdev_matrix_show(&matrix_mdev->matrix, buf); | |
d0786556 | 1458 | mutex_unlock(&matrix_dev->mdevs_lock); |
81b2b4b7 TK |
1459 | |
1460 | return nchars; | |
1461 | } | |
46623ab3 | 1462 | static DEVICE_ATTR_RO(matrix); |
81b2b4b7 | 1463 | |
f7f795c5 TK |
1464 | static ssize_t guest_matrix_show(struct device *dev, |
1465 | struct device_attribute *attr, char *buf) | |
1466 | { | |
1467 | ssize_t nchars; | |
1468 | struct ap_matrix_mdev *matrix_mdev = dev_get_drvdata(dev); | |
1469 | ||
1470 | mutex_lock(&matrix_dev->mdevs_lock); | |
1471 | nchars = vfio_ap_mdev_matrix_show(&matrix_mdev->shadow_apcb, buf); | |
1472 | mutex_unlock(&matrix_dev->mdevs_lock); | |
1473 | ||
1474 | return nchars; | |
1475 | } | |
1476 | static DEVICE_ATTR_RO(guest_matrix); | |
1477 | ||
96d152bd TK |
1478 | static struct attribute *vfio_ap_mdev_attrs[] = { |
1479 | &dev_attr_assign_adapter.attr, | |
1480 | &dev_attr_unassign_adapter.attr, | |
3211da0c TK |
1481 | &dev_attr_assign_domain.attr, |
1482 | &dev_attr_unassign_domain.attr, | |
3b1eab7f TK |
1483 | &dev_attr_assign_control_domain.attr, |
1484 | &dev_attr_unassign_control_domain.attr, | |
1485 | &dev_attr_control_domains.attr, | |
81b2b4b7 | 1486 | &dev_attr_matrix.attr, |
f7f795c5 | 1487 | &dev_attr_guest_matrix.attr, |
3211da0c | 1488 | NULL, |
96d152bd TK |
1489 | }; |
1490 | ||
1491 | static struct attribute_group vfio_ap_mdev_attr_group = { | |
1492 | .attrs = vfio_ap_mdev_attrs | |
1493 | }; | |
1494 | ||
1495 | static const struct attribute_group *vfio_ap_mdev_attr_groups[] = { | |
1496 | &vfio_ap_mdev_attr_group, | |
1497 | NULL | |
1498 | }; | |
1499 | ||
258287c9 | 1500 | /** |
0c1abe7c RD |
1501 | * vfio_ap_mdev_set_kvm - sets all data for @matrix_mdev that are needed |
1502 | * to manage AP resources for the guest whose state is represented by @kvm | |
258287c9 TK |
1503 | * |
1504 | * @matrix_mdev: a mediated matrix device | |
1505 | * @kvm: reference to KVM instance | |
1506 | * | |
0c1abe7c | 1507 | * Return: 0 if no other mediated matrix device has a reference to @kvm; |
258287c9 TK |
1508 | * otherwise, returns an -EPERM. |
1509 | */ | |
1510 | static int vfio_ap_mdev_set_kvm(struct ap_matrix_mdev *matrix_mdev, | |
1511 | struct kvm *kvm) | |
1512 | { | |
1513 | struct ap_matrix_mdev *m; | |
1514 | ||
0cc00c8d | 1515 | if (kvm->arch.crypto.crycbd) { |
86956e70 TK |
1516 | down_write(&kvm->arch.crypto.pqap_hook_rwsem); |
1517 | kvm->arch.crypto.pqap_hook = &matrix_mdev->pqap_hook; | |
1518 | up_write(&kvm->arch.crypto.pqap_hook_rwsem); | |
1519 | ||
b84eb8e0 | 1520 | get_update_locks_for_kvm(kvm); |
86956e70 | 1521 | |
0cc00c8d | 1522 | list_for_each_entry(m, &matrix_dev->mdev_list, node) { |
86956e70 | 1523 | if (m != matrix_mdev && m->kvm == kvm) { |
b84eb8e0 | 1524 | release_update_locks_for_kvm(kvm); |
0cc00c8d | 1525 | return -EPERM; |
86956e70 | 1526 | } |
0cc00c8d | 1527 | } |
258287c9 | 1528 | |
0cc00c8d | 1529 | kvm_get_kvm(kvm); |
1e753732 | 1530 | matrix_mdev->kvm = kvm; |
51dc562a | 1531 | vfio_ap_mdev_update_guest_apcb(matrix_mdev); |
1e753732 | 1532 | |
b84eb8e0 | 1533 | release_update_locks_for_kvm(kvm); |
0cc00c8d | 1534 | } |
258287c9 TK |
1535 | |
1536 | return 0; | |
1537 | } | |
1538 | ||
2a54e347 MR |
1539 | static void unmap_iova(struct ap_matrix_mdev *matrix_mdev, u64 iova, u64 length) |
1540 | { | |
1541 | struct ap_queue_table *qtable = &matrix_mdev->qtable; | |
1542 | struct vfio_ap_queue *q; | |
1543 | int loop_cursor; | |
1544 | ||
1545 | hash_for_each(qtable->queues, loop_cursor, q, mdev_qnode) { | |
1546 | if (q->saved_iova >= iova && q->saved_iova < iova + length) | |
1547 | vfio_ap_irq_disable(q); | |
1548 | } | |
1549 | } | |
1550 | ||
ce4b4657 JG |
1551 | static void vfio_ap_mdev_dma_unmap(struct vfio_device *vdev, u64 iova, |
1552 | u64 length) | |
62e358ce | 1553 | { |
ce4b4657 JG |
1554 | struct ap_matrix_mdev *matrix_mdev = |
1555 | container_of(vdev, struct ap_matrix_mdev, vdev); | |
62e358ce | 1556 | |
2a54e347 MR |
1557 | mutex_lock(&matrix_dev->mdevs_lock); |
1558 | ||
1559 | unmap_iova(matrix_mdev, iova, length); | |
1560 | ||
1561 | mutex_unlock(&matrix_dev->mdevs_lock); | |
62e358ce PM |
1562 | } |
1563 | ||
0cc00c8d | 1564 | /** |
0c1abe7c RD |
1565 | * vfio_ap_mdev_unset_kvm - performs clean-up of resources no longer needed |
1566 | * by @matrix_mdev. | |
0cc00c8d TK |
1567 | * |
1568 | * @matrix_mdev: a matrix mediated device | |
0cc00c8d | 1569 | */ |
71078220 | 1570 | static void vfio_ap_mdev_unset_kvm(struct ap_matrix_mdev *matrix_mdev) |
f21916ec | 1571 | { |
71078220 TK |
1572 | struct kvm *kvm = matrix_mdev->kvm; |
1573 | ||
86956e70 TK |
1574 | if (kvm && kvm->arch.crypto.crycbd) { |
1575 | down_write(&kvm->arch.crypto.pqap_hook_rwsem); | |
1576 | kvm->arch.crypto.pqap_hook = NULL; | |
1577 | up_write(&kvm->arch.crypto.pqap_hook_rwsem); | |
0cc00c8d | 1578 | |
b84eb8e0 | 1579 | get_update_locks_for_kvm(kvm); |
86956e70 TK |
1580 | |
1581 | kvm_arch_crypto_clear_masks(kvm); | |
70aeefe5 | 1582 | vfio_ap_mdev_reset_queues(&matrix_mdev->qtable); |
86956e70 | 1583 | kvm_put_kvm(kvm); |
0cc00c8d | 1584 | matrix_mdev->kvm = NULL; |
86956e70 | 1585 | |
b84eb8e0 | 1586 | release_update_locks_for_kvm(kvm); |
0cc00c8d | 1587 | } |
f21916ec TK |
1588 | } |
1589 | ||
6c12a638 | 1590 | static struct vfio_ap_queue *vfio_ap_find_queue(int apqn) |
ec89b55e | 1591 | { |
034921cd | 1592 | struct ap_queue *queue; |
6c12a638 | 1593 | struct vfio_ap_queue *q = NULL; |
ec89b55e | 1594 | |
034921cd TK |
1595 | queue = ap_get_qdev(apqn); |
1596 | if (!queue) | |
1597 | return NULL; | |
1598 | ||
1599 | if (queue->ap_dev.device.driver == &matrix_dev->vfio_ap_drv->driver) | |
1600 | q = dev_get_drvdata(&queue->ap_dev.device); | |
1601 | ||
1602 | put_device(&queue->ap_dev.device); | |
6c12a638 TK |
1603 | |
1604 | return q; | |
ec89b55e PM |
1605 | } |
1606 | ||
0daf9878 TK |
1607 | static int apq_status_check(int apqn, struct ap_queue_status *status) |
1608 | { | |
1609 | switch (status->response_code) { | |
1610 | case AP_RESPONSE_NORMAL: | |
1611 | case AP_RESPONSE_RESET_IN_PROGRESS: | |
1612 | if (status->queue_empty && !status->irq_enabled) | |
1613 | return 0; | |
1614 | return -EBUSY; | |
1615 | case AP_RESPONSE_DECONFIGURED: | |
1616 | /* | |
1617 | * If the AP queue is deconfigured, any subsequent AP command | |
1618 | * targeting the queue will fail with the same response code. On the | |
1619 | * other hand, when an AP adapter is deconfigured, the associated | |
1620 | * queues are reset, so let's return a value indicating the reset | |
1621 | * for which we're waiting completed successfully. | |
1622 | */ | |
1623 | return 0; | |
1624 | default: | |
1625 | WARN(true, | |
1626 | "failed to verify reset of queue %02x.%04x: TAPQ rc=%u\n", | |
1627 | AP_QID_CARD(apqn), AP_QID_QUEUE(apqn), | |
1628 | status->response_code); | |
1629 | return -EIO; | |
1630 | } | |
1631 | } | |
1632 | ||
62414d90 TK |
1633 | static int apq_reset_check(struct vfio_ap_queue *q) |
1634 | { | |
7cb7636a TK |
1635 | int ret; |
1636 | int iters = MAX_RESET_CHECK_WAIT / AP_RESET_INTERVAL; | |
62414d90 TK |
1637 | struct ap_queue_status status; |
1638 | ||
7cb7636a TK |
1639 | for (; iters > 0; iters--) { |
1640 | msleep(AP_RESET_INTERVAL); | |
62414d90 | 1641 | status = ap_tapq(q->apqn, NULL); |
0daf9878 TK |
1642 | ret = apq_status_check(q->apqn, &status); |
1643 | if (ret != -EBUSY) | |
1644 | return ret; | |
62414d90 TK |
1645 | } |
1646 | WARN_ONCE(iters <= 0, | |
1647 | "timeout verifying reset of queue %02x.%04x (%u, %u, %u)", | |
1648 | AP_QID_CARD(q->apqn), AP_QID_QUEUE(q->apqn), | |
1649 | status.queue_empty, status.irq_enabled, status.response_code); | |
0daf9878 | 1650 | return ret; |
62414d90 TK |
1651 | } |
1652 | ||
3ba41768 | 1653 | static int vfio_ap_mdev_reset_queue(struct vfio_ap_queue *q) |
46a7263d TK |
1654 | { |
1655 | struct ap_queue_status status; | |
6c12a638 | 1656 | int ret; |
46a7263d | 1657 | |
6c12a638 TK |
1658 | if (!q) |
1659 | return 0; | |
6c12a638 | 1660 | retry_zapq: |
4bdf3c39 | 1661 | status = ap_zapq(q->apqn, 0); |
70aeefe5 | 1662 | q->reset_rc = status.response_code; |
6c12a638 TK |
1663 | switch (status.response_code) { |
1664 | case AP_RESPONSE_NORMAL: | |
1665 | ret = 0; | |
5a42b348 TK |
1666 | /* if the reset has not completed, wait for it to take effect */ |
1667 | if (!status.queue_empty || status.irq_enabled) | |
1668 | ret = apq_reset_check(q); | |
6c12a638 TK |
1669 | break; |
1670 | case AP_RESPONSE_RESET_IN_PROGRESS: | |
3ba41768 TK |
1671 | /* |
1672 | * There is a reset issued by another process in progress. Let's wait | |
1673 | * for that to complete. Since we have no idea whether it was a RAPQ or | |
1674 | * ZAPQ, then if it completes successfully, let's issue the ZAPQ. | |
1675 | */ | |
1676 | ret = apq_reset_check(q); | |
1677 | if (ret) | |
1678 | break; | |
1679 | goto retry_zapq; | |
6c12a638 | 1680 | case AP_RESPONSE_DECONFIGURED: |
51d4d987 TK |
1681 | /* |
1682 | * When an AP adapter is deconfigured, the associated | |
1683 | * queues are reset, so let's return a value indicating the reset | |
1684 | * completed successfully. | |
1685 | */ | |
1686 | ret = 0; | |
1687 | break; | |
6c12a638 | 1688 | default: |
70aeefe5 TK |
1689 | WARN(true, |
1690 | "PQAP/ZAPQ for %02x.%04x failed with invalid rc=%u\n", | |
1691 | AP_QID_CARD(q->apqn), AP_QID_QUEUE(q->apqn), | |
6c12a638 TK |
1692 | status.response_code); |
1693 | return -EIO; | |
1694 | } | |
1695 | ||
6c12a638 TK |
1696 | vfio_ap_free_aqic_resources(q); |
1697 | ||
1698 | return ret; | |
46a7263d TK |
1699 | } |
1700 | ||
70aeefe5 | 1701 | static int vfio_ap_mdev_reset_queues(struct ap_queue_table *qtable) |
46a7263d | 1702 | { |
11cb2419 | 1703 | int ret, loop_cursor, rc = 0; |
6c12a638 | 1704 | struct vfio_ap_queue *q; |
46a7263d | 1705 | |
70aeefe5 | 1706 | hash_for_each(qtable->queues, loop_cursor, q, mdev_qnode) { |
3ba41768 | 1707 | ret = vfio_ap_mdev_reset_queue(q); |
11cb2419 TK |
1708 | /* |
1709 | * Regardless whether a queue turns out to be busy, or | |
1710 | * is not operational, we need to continue resetting | |
1711 | * the remaining queues. | |
1712 | */ | |
1713 | if (ret) | |
1714 | rc = ret; | |
46a7263d TK |
1715 | } |
1716 | ||
1717 | return rc; | |
1718 | } | |
1719 | ||
eb0feefd | 1720 | static int vfio_ap_mdev_open_device(struct vfio_device *vdev) |
258287c9 | 1721 | { |
eb0feefd JG |
1722 | struct ap_matrix_mdev *matrix_mdev = |
1723 | container_of(vdev, struct ap_matrix_mdev, vdev); | |
258287c9 | 1724 | |
421cfe65 MR |
1725 | if (!vdev->kvm) |
1726 | return -EINVAL; | |
258287c9 | 1727 | |
ce4b4657 | 1728 | return vfio_ap_mdev_set_kvm(matrix_mdev, vdev->kvm); |
258287c9 TK |
1729 | } |
1730 | ||
eb0feefd | 1731 | static void vfio_ap_mdev_close_device(struct vfio_device *vdev) |
258287c9 | 1732 | { |
eb0feefd JG |
1733 | struct ap_matrix_mdev *matrix_mdev = |
1734 | container_of(vdev, struct ap_matrix_mdev, vdev); | |
258287c9 | 1735 | |
71078220 | 1736 | vfio_ap_mdev_unset_kvm(matrix_mdev); |
258287c9 TK |
1737 | } |
1738 | ||
e06670c5 TK |
1739 | static int vfio_ap_mdev_get_device_info(unsigned long arg) |
1740 | { | |
1741 | unsigned long minsz; | |
1742 | struct vfio_device_info info; | |
1743 | ||
1744 | minsz = offsetofend(struct vfio_device_info, num_irqs); | |
1745 | ||
1746 | if (copy_from_user(&info, (void __user *)arg, minsz)) | |
1747 | return -EFAULT; | |
1748 | ||
1749 | if (info.argsz < minsz) | |
1750 | return -EINVAL; | |
1751 | ||
cd8a377e | 1752 | info.flags = VFIO_DEVICE_FLAGS_AP | VFIO_DEVICE_FLAGS_RESET; |
e06670c5 | 1753 | info.num_regions = 0; |
6afc7700 TK |
1754 | info.num_irqs = VFIO_AP_NUM_IRQS; |
1755 | ||
1756 | return copy_to_user((void __user *)arg, &info, minsz) ? -EFAULT : 0; | |
1757 | } | |
1758 | ||
1759 | static ssize_t vfio_ap_get_irq_info(unsigned long arg) | |
1760 | { | |
1761 | unsigned long minsz; | |
1762 | struct vfio_irq_info info; | |
1763 | ||
1764 | minsz = offsetofend(struct vfio_irq_info, count); | |
1765 | ||
1766 | if (copy_from_user(&info, (void __user *)arg, minsz)) | |
1767 | return -EFAULT; | |
1768 | ||
1769 | if (info.argsz < minsz || info.index >= VFIO_AP_NUM_IRQS) | |
1770 | return -EINVAL; | |
1771 | ||
1772 | switch (info.index) { | |
1773 | case VFIO_AP_REQ_IRQ_INDEX: | |
1774 | info.count = 1; | |
1775 | info.flags = VFIO_IRQ_INFO_EVENTFD; | |
1776 | break; | |
1777 | default: | |
1778 | return -EINVAL; | |
1779 | } | |
e06670c5 | 1780 | |
942df4be | 1781 | return copy_to_user((void __user *)arg, &info, minsz) ? -EFAULT : 0; |
e06670c5 TK |
1782 | } |
1783 | ||
bf48961f TK |
1784 | static int vfio_ap_irq_set_init(struct vfio_irq_set *irq_set, unsigned long arg) |
1785 | { | |
1786 | int ret; | |
1787 | size_t data_size; | |
1788 | unsigned long minsz; | |
1789 | ||
1790 | minsz = offsetofend(struct vfio_irq_set, count); | |
1791 | ||
1792 | if (copy_from_user(irq_set, (void __user *)arg, minsz)) | |
1793 | return -EFAULT; | |
1794 | ||
1795 | ret = vfio_set_irqs_validate_and_prepare(irq_set, 1, VFIO_AP_NUM_IRQS, | |
1796 | &data_size); | |
1797 | if (ret) | |
1798 | return ret; | |
1799 | ||
1800 | if (!(irq_set->flags & VFIO_IRQ_SET_ACTION_TRIGGER)) | |
1801 | return -EINVAL; | |
1802 | ||
1803 | return 0; | |
1804 | } | |
1805 | ||
1806 | static int vfio_ap_set_request_irq(struct ap_matrix_mdev *matrix_mdev, | |
1807 | unsigned long arg) | |
1808 | { | |
1809 | s32 fd; | |
1810 | void __user *data; | |
1811 | unsigned long minsz; | |
1812 | struct eventfd_ctx *req_trigger; | |
1813 | ||
1814 | minsz = offsetofend(struct vfio_irq_set, count); | |
1815 | data = (void __user *)(arg + minsz); | |
1816 | ||
1817 | if (get_user(fd, (s32 __user *)data)) | |
1818 | return -EFAULT; | |
1819 | ||
1820 | if (fd == -1) { | |
1821 | if (matrix_mdev->req_trigger) | |
1822 | eventfd_ctx_put(matrix_mdev->req_trigger); | |
1823 | matrix_mdev->req_trigger = NULL; | |
1824 | } else if (fd >= 0) { | |
1825 | req_trigger = eventfd_ctx_fdget(fd); | |
1826 | if (IS_ERR(req_trigger)) | |
1827 | return PTR_ERR(req_trigger); | |
1828 | ||
1829 | if (matrix_mdev->req_trigger) | |
1830 | eventfd_ctx_put(matrix_mdev->req_trigger); | |
1831 | ||
1832 | matrix_mdev->req_trigger = req_trigger; | |
1833 | } else { | |
1834 | return -EINVAL; | |
1835 | } | |
1836 | ||
1837 | return 0; | |
1838 | } | |
1839 | ||
1840 | static int vfio_ap_set_irqs(struct ap_matrix_mdev *matrix_mdev, | |
1841 | unsigned long arg) | |
1842 | { | |
1843 | int ret; | |
1844 | struct vfio_irq_set irq_set; | |
1845 | ||
1846 | ret = vfio_ap_irq_set_init(&irq_set, arg); | |
1847 | if (ret) | |
1848 | return ret; | |
1849 | ||
1850 | switch (irq_set.flags & VFIO_IRQ_SET_DATA_TYPE_MASK) { | |
1851 | case VFIO_IRQ_SET_DATA_EVENTFD: | |
1852 | switch (irq_set.index) { | |
1853 | case VFIO_AP_REQ_IRQ_INDEX: | |
1854 | return vfio_ap_set_request_irq(matrix_mdev, arg); | |
1855 | default: | |
1856 | return -EINVAL; | |
1857 | } | |
1858 | default: | |
1859 | return -EINVAL; | |
1860 | } | |
1861 | } | |
1862 | ||
eb0feefd | 1863 | static ssize_t vfio_ap_mdev_ioctl(struct vfio_device *vdev, |
e06670c5 TK |
1864 | unsigned int cmd, unsigned long arg) |
1865 | { | |
eb0feefd JG |
1866 | struct ap_matrix_mdev *matrix_mdev = |
1867 | container_of(vdev, struct ap_matrix_mdev, vdev); | |
e06670c5 TK |
1868 | int ret; |
1869 | ||
d0786556 | 1870 | mutex_lock(&matrix_dev->mdevs_lock); |
e06670c5 TK |
1871 | switch (cmd) { |
1872 | case VFIO_DEVICE_GET_INFO: | |
1873 | ret = vfio_ap_mdev_get_device_info(arg); | |
1874 | break; | |
cd8a377e | 1875 | case VFIO_DEVICE_RESET: |
70aeefe5 | 1876 | ret = vfio_ap_mdev_reset_queues(&matrix_mdev->qtable); |
cd8a377e | 1877 | break; |
6afc7700 TK |
1878 | case VFIO_DEVICE_GET_IRQ_INFO: |
1879 | ret = vfio_ap_get_irq_info(arg); | |
1880 | break; | |
bf48961f TK |
1881 | case VFIO_DEVICE_SET_IRQS: |
1882 | ret = vfio_ap_set_irqs(matrix_mdev, arg); | |
1883 | break; | |
e06670c5 TK |
1884 | default: |
1885 | ret = -EOPNOTSUPP; | |
1886 | break; | |
1887 | } | |
d0786556 | 1888 | mutex_unlock(&matrix_dev->mdevs_lock); |
e06670c5 TK |
1889 | |
1890 | return ret; | |
1891 | } | |
1892 | ||
260f3ea1 TK |
1893 | static struct ap_matrix_mdev *vfio_ap_mdev_for_queue(struct vfio_ap_queue *q) |
1894 | { | |
1895 | struct ap_matrix_mdev *matrix_mdev; | |
1896 | unsigned long apid = AP_QID_CARD(q->apqn); | |
1897 | unsigned long apqi = AP_QID_QUEUE(q->apqn); | |
1898 | ||
1899 | list_for_each_entry(matrix_mdev, &matrix_dev->mdev_list, node) { | |
1900 | if (test_bit_inv(apid, matrix_mdev->matrix.apm) && | |
1901 | test_bit_inv(apqi, matrix_mdev->matrix.aqm)) | |
1902 | return matrix_mdev; | |
1903 | } | |
1904 | ||
1905 | return NULL; | |
1906 | } | |
1907 | ||
1908 | static ssize_t status_show(struct device *dev, | |
1909 | struct device_attribute *attr, | |
1910 | char *buf) | |
1911 | { | |
1912 | ssize_t nchars = 0; | |
1913 | struct vfio_ap_queue *q; | |
1914 | struct ap_matrix_mdev *matrix_mdev; | |
1915 | struct ap_device *apdev = to_ap_dev(dev); | |
1916 | ||
d0786556 | 1917 | mutex_lock(&matrix_dev->mdevs_lock); |
260f3ea1 TK |
1918 | q = dev_get_drvdata(&apdev->device); |
1919 | matrix_mdev = vfio_ap_mdev_for_queue(q); | |
1920 | ||
1921 | if (matrix_mdev) { | |
1922 | if (matrix_mdev->kvm) | |
1923 | nchars = scnprintf(buf, PAGE_SIZE, "%s\n", | |
1924 | AP_QUEUE_IN_USE); | |
1925 | else | |
1926 | nchars = scnprintf(buf, PAGE_SIZE, "%s\n", | |
1927 | AP_QUEUE_ASSIGNED); | |
1928 | } else { | |
1929 | nchars = scnprintf(buf, PAGE_SIZE, "%s\n", | |
1930 | AP_QUEUE_UNASSIGNED); | |
1931 | } | |
1932 | ||
d0786556 | 1933 | mutex_unlock(&matrix_dev->mdevs_lock); |
260f3ea1 TK |
1934 | |
1935 | return nchars; | |
1936 | } | |
1937 | ||
1938 | static DEVICE_ATTR_RO(status); | |
1939 | ||
1940 | static struct attribute *vfio_queue_attrs[] = { | |
1941 | &dev_attr_status.attr, | |
1942 | NULL, | |
1943 | }; | |
1944 | ||
1945 | static const struct attribute_group vfio_queue_attr_group = { | |
1946 | .attrs = vfio_queue_attrs, | |
1947 | }; | |
1948 | ||
eb0feefd | 1949 | static const struct vfio_device_ops vfio_ap_matrix_dev_ops = { |
7cb5a82e | 1950 | .init = vfio_ap_mdev_init_dev, |
eb0feefd JG |
1951 | .open_device = vfio_ap_mdev_open_device, |
1952 | .close_device = vfio_ap_mdev_close_device, | |
1953 | .ioctl = vfio_ap_mdev_ioctl, | |
ce4b4657 | 1954 | .dma_unmap = vfio_ap_mdev_dma_unmap, |
4741f2e9 JG |
1955 | .bind_iommufd = vfio_iommufd_emulated_bind, |
1956 | .unbind_iommufd = vfio_iommufd_emulated_unbind, | |
1957 | .attach_ioas = vfio_iommufd_emulated_attach_ioas, | |
eb0feefd JG |
1958 | }; |
1959 | ||
1960 | static struct mdev_driver vfio_ap_matrix_driver = { | |
290aac5d | 1961 | .device_api = VFIO_DEVICE_API_AP_STRING, |
9c799c22 | 1962 | .max_instances = MAX_ZDEV_ENTRIES_EXT, |
eb0feefd JG |
1963 | .driver = { |
1964 | .name = "vfio_ap_mdev", | |
1965 | .owner = THIS_MODULE, | |
1966 | .mod_name = KBUILD_MODNAME, | |
1967 | .dev_groups = vfio_ap_mdev_attr_groups, | |
1968 | }, | |
1969 | .probe = vfio_ap_mdev_probe, | |
1970 | .remove = vfio_ap_mdev_remove, | |
65f06713 TK |
1971 | }; |
1972 | ||
1973 | int vfio_ap_mdev_register(void) | |
1974 | { | |
eb0feefd JG |
1975 | int ret; |
1976 | ||
eb0feefd JG |
1977 | ret = mdev_register_driver(&vfio_ap_matrix_driver); |
1978 | if (ret) | |
1979 | return ret; | |
1980 | ||
da44c340 | 1981 | matrix_dev->mdev_type.sysfs_name = VFIO_AP_MDEV_TYPE_HWVIRT; |
0bc79069 | 1982 | matrix_dev->mdev_type.pretty_name = VFIO_AP_MDEV_NAME_HWVIRT; |
da44c340 | 1983 | matrix_dev->mdev_types[0] = &matrix_dev->mdev_type; |
89345d51 | 1984 | ret = mdev_register_parent(&matrix_dev->parent, &matrix_dev->device, |
da44c340 CH |
1985 | &vfio_ap_matrix_driver, |
1986 | matrix_dev->mdev_types, 1); | |
eb0feefd JG |
1987 | if (ret) |
1988 | goto err_driver; | |
1989 | return 0; | |
1990 | ||
1991 | err_driver: | |
1992 | mdev_unregister_driver(&vfio_ap_matrix_driver); | |
1993 | return ret; | |
65f06713 TK |
1994 | } |
1995 | ||
1996 | void vfio_ap_mdev_unregister(void) | |
1997 | { | |
89345d51 | 1998 | mdev_unregister_parent(&matrix_dev->parent); |
eb0feefd | 1999 | mdev_unregister_driver(&vfio_ap_matrix_driver); |
65f06713 | 2000 | } |
260f3ea1 TK |
2001 | |
2002 | int vfio_ap_mdev_probe_queue(struct ap_device *apdev) | |
2003 | { | |
2004 | int ret; | |
2005 | struct vfio_ap_queue *q; | |
2c1ee898 | 2006 | struct ap_matrix_mdev *matrix_mdev; |
260f3ea1 TK |
2007 | |
2008 | ret = sysfs_create_group(&apdev->device.kobj, &vfio_queue_attr_group); | |
2009 | if (ret) | |
2010 | return ret; | |
2011 | ||
2012 | q = kzalloc(sizeof(*q), GFP_KERNEL); | |
08866d34 CJ |
2013 | if (!q) { |
2014 | ret = -ENOMEM; | |
2015 | goto err_remove_group; | |
2016 | } | |
260f3ea1 | 2017 | |
260f3ea1 TK |
2018 | q->apqn = to_ap_queue(&apdev->device)->qid; |
2019 | q->saved_isc = VFIO_AP_ISC_INVALID; | |
2c1ee898 TK |
2020 | matrix_mdev = get_update_locks_by_apqn(q->apqn); |
2021 | ||
2022 | if (matrix_mdev) { | |
2023 | vfio_ap_mdev_link_queue(matrix_mdev, q); | |
09d31ff7 TK |
2024 | |
2025 | if (vfio_ap_mdev_filter_matrix(matrix_mdev->matrix.apm, | |
2026 | matrix_mdev->matrix.aqm, | |
2027 | matrix_mdev)) | |
2028 | vfio_ap_mdev_update_guest_apcb(matrix_mdev); | |
48cae940 | 2029 | } |
260f3ea1 | 2030 | dev_set_drvdata(&apdev->device, q); |
2c1ee898 | 2031 | release_update_locks_for_mdev(matrix_mdev); |
260f3ea1 TK |
2032 | |
2033 | return 0; | |
08866d34 CJ |
2034 | |
2035 | err_remove_group: | |
2036 | sysfs_remove_group(&apdev->device.kobj, &vfio_queue_attr_group); | |
2037 | return ret; | |
260f3ea1 TK |
2038 | } |
2039 | ||
2040 | void vfio_ap_mdev_remove_queue(struct ap_device *apdev) | |
2041 | { | |
09d31ff7 | 2042 | unsigned long apid, apqi; |
260f3ea1 | 2043 | struct vfio_ap_queue *q; |
2c1ee898 | 2044 | struct ap_matrix_mdev *matrix_mdev; |
260f3ea1 | 2045 | |
260f3ea1 TK |
2046 | sysfs_remove_group(&apdev->device.kobj, &vfio_queue_attr_group); |
2047 | q = dev_get_drvdata(&apdev->device); | |
2c1ee898 TK |
2048 | get_update_locks_for_queue(q); |
2049 | matrix_mdev = q->matrix_mdev; | |
11cb2419 | 2050 | |
2c1ee898 | 2051 | if (matrix_mdev) { |
11cb2419 TK |
2052 | vfio_ap_unlink_queue_fr_mdev(q); |
2053 | ||
48cae940 | 2054 | apid = AP_QID_CARD(q->apqn); |
09d31ff7 TK |
2055 | apqi = AP_QID_QUEUE(q->apqn); |
2056 | ||
2057 | /* | |
2058 | * If the queue is assigned to the guest's APCB, then remove | |
2059 | * the adapter's APID from the APCB and hot it into the guest. | |
2060 | */ | |
2061 | if (test_bit_inv(apid, matrix_mdev->shadow_apcb.apm) && | |
2062 | test_bit_inv(apqi, matrix_mdev->shadow_apcb.aqm)) { | |
2063 | clear_bit_inv(apid, matrix_mdev->shadow_apcb.apm); | |
2064 | vfio_ap_mdev_update_guest_apcb(matrix_mdev); | |
2065 | } | |
48cae940 TK |
2066 | } |
2067 | ||
3ba41768 | 2068 | vfio_ap_mdev_reset_queue(q); |
260f3ea1 TK |
2069 | dev_set_drvdata(&apdev->device, NULL); |
2070 | kfree(q); | |
2c1ee898 | 2071 | release_update_locks_for_mdev(matrix_mdev); |
260f3ea1 | 2072 | } |
3f85d1df TK |
2073 | |
2074 | /** | |
2075 | * vfio_ap_mdev_resource_in_use: check whether any of a set of APQNs is | |
2076 | * assigned to a mediated device under the control | |
2077 | * of the vfio_ap device driver. | |
2078 | * | |
2079 | * @apm: a bitmap specifying a set of APIDs comprising the APQNs to check. | |
2080 | * @aqm: a bitmap specifying a set of APQIs comprising the APQNs to check. | |
2081 | * | |
2082 | * Return: | |
2083 | * * -EADDRINUSE if one or more of the APQNs specified via @apm/@aqm are | |
2084 | * assigned to a mediated device under the control of the vfio_ap | |
2085 | * device driver. | |
2086 | * * Otherwise, return 0. | |
2087 | */ | |
2088 | int vfio_ap_mdev_resource_in_use(unsigned long *apm, unsigned long *aqm) | |
2089 | { | |
2090 | int ret; | |
2091 | ||
2092 | mutex_lock(&matrix_dev->guests_lock); | |
2093 | mutex_lock(&matrix_dev->mdevs_lock); | |
2094 | ret = vfio_ap_mdev_verify_no_sharing(apm, aqm); | |
2095 | mutex_unlock(&matrix_dev->mdevs_lock); | |
2096 | mutex_unlock(&matrix_dev->guests_lock); | |
2097 | ||
2098 | return ret; | |
2099 | } | |
eeb386ae TK |
2100 | |
2101 | /** | |
2102 | * vfio_ap_mdev_hot_unplug_cfg - hot unplug the adapters, domains and control | |
2103 | * domains that have been removed from the host's | |
2104 | * AP configuration from a guest. | |
2105 | * | |
2106 | * @matrix_mdev: an ap_matrix_mdev object attached to a KVM guest. | |
2107 | * @aprem: the adapters that have been removed from the host's AP configuration | |
2108 | * @aqrem: the domains that have been removed from the host's AP configuration | |
2109 | * @cdrem: the control domains that have been removed from the host's AP | |
2110 | * configuration. | |
2111 | */ | |
2112 | static void vfio_ap_mdev_hot_unplug_cfg(struct ap_matrix_mdev *matrix_mdev, | |
2113 | unsigned long *aprem, | |
2114 | unsigned long *aqrem, | |
2115 | unsigned long *cdrem) | |
2116 | { | |
2117 | int do_hotplug = 0; | |
2118 | ||
2119 | if (!bitmap_empty(aprem, AP_DEVICES)) { | |
2120 | do_hotplug |= bitmap_andnot(matrix_mdev->shadow_apcb.apm, | |
2121 | matrix_mdev->shadow_apcb.apm, | |
2122 | aprem, AP_DEVICES); | |
2123 | } | |
2124 | ||
2125 | if (!bitmap_empty(aqrem, AP_DOMAINS)) { | |
2126 | do_hotplug |= bitmap_andnot(matrix_mdev->shadow_apcb.aqm, | |
2127 | matrix_mdev->shadow_apcb.aqm, | |
2128 | aqrem, AP_DEVICES); | |
2129 | } | |
2130 | ||
2131 | if (!bitmap_empty(cdrem, AP_DOMAINS)) | |
2132 | do_hotplug |= bitmap_andnot(matrix_mdev->shadow_apcb.adm, | |
2133 | matrix_mdev->shadow_apcb.adm, | |
2134 | cdrem, AP_DOMAINS); | |
2135 | ||
2136 | if (do_hotplug) | |
2137 | vfio_ap_mdev_update_guest_apcb(matrix_mdev); | |
2138 | } | |
2139 | ||
2140 | /** | |
2141 | * vfio_ap_mdev_cfg_remove - determines which guests are using the adapters, | |
2142 | * domains and control domains that have been removed | |
2143 | * from the host AP configuration and unplugs them | |
2144 | * from those guests. | |
2145 | * | |
2146 | * @ap_remove: bitmap specifying which adapters have been removed from the host | |
2147 | * config. | |
2148 | * @aq_remove: bitmap specifying which domains have been removed from the host | |
2149 | * config. | |
2150 | * @cd_remove: bitmap specifying which control domains have been removed from | |
2151 | * the host config. | |
2152 | */ | |
2153 | static void vfio_ap_mdev_cfg_remove(unsigned long *ap_remove, | |
2154 | unsigned long *aq_remove, | |
2155 | unsigned long *cd_remove) | |
2156 | { | |
2157 | struct ap_matrix_mdev *matrix_mdev; | |
2158 | DECLARE_BITMAP(aprem, AP_DEVICES); | |
2159 | DECLARE_BITMAP(aqrem, AP_DOMAINS); | |
2160 | DECLARE_BITMAP(cdrem, AP_DOMAINS); | |
2161 | int do_remove = 0; | |
2162 | ||
2163 | list_for_each_entry(matrix_mdev, &matrix_dev->mdev_list, node) { | |
2164 | mutex_lock(&matrix_mdev->kvm->lock); | |
2165 | mutex_lock(&matrix_dev->mdevs_lock); | |
2166 | ||
2167 | do_remove |= bitmap_and(aprem, ap_remove, | |
2168 | matrix_mdev->matrix.apm, | |
2169 | AP_DEVICES); | |
2170 | do_remove |= bitmap_and(aqrem, aq_remove, | |
2171 | matrix_mdev->matrix.aqm, | |
2172 | AP_DOMAINS); | |
2173 | do_remove |= bitmap_andnot(cdrem, cd_remove, | |
2174 | matrix_mdev->matrix.adm, | |
2175 | AP_DOMAINS); | |
2176 | ||
2177 | if (do_remove) | |
2178 | vfio_ap_mdev_hot_unplug_cfg(matrix_mdev, aprem, aqrem, | |
2179 | cdrem); | |
2180 | ||
2181 | mutex_unlock(&matrix_dev->mdevs_lock); | |
2182 | mutex_unlock(&matrix_mdev->kvm->lock); | |
2183 | } | |
2184 | } | |
2185 | ||
2186 | /** | |
2187 | * vfio_ap_mdev_on_cfg_remove - responds to the removal of adapters, domains and | |
2188 | * control domains from the host AP configuration | |
2189 | * by unplugging them from the guests that are | |
2190 | * using them. | |
2191 | * @cur_config_info: the current host AP configuration information | |
2192 | * @prev_config_info: the previous host AP configuration information | |
2193 | */ | |
2194 | static void vfio_ap_mdev_on_cfg_remove(struct ap_config_info *cur_config_info, | |
2195 | struct ap_config_info *prev_config_info) | |
2196 | { | |
2197 | int do_remove; | |
2198 | DECLARE_BITMAP(aprem, AP_DEVICES); | |
2199 | DECLARE_BITMAP(aqrem, AP_DOMAINS); | |
2200 | DECLARE_BITMAP(cdrem, AP_DOMAINS); | |
2201 | ||
2202 | do_remove = bitmap_andnot(aprem, | |
2203 | (unsigned long *)prev_config_info->apm, | |
2204 | (unsigned long *)cur_config_info->apm, | |
2205 | AP_DEVICES); | |
2206 | do_remove |= bitmap_andnot(aqrem, | |
2207 | (unsigned long *)prev_config_info->aqm, | |
2208 | (unsigned long *)cur_config_info->aqm, | |
2209 | AP_DEVICES); | |
2210 | do_remove |= bitmap_andnot(cdrem, | |
2211 | (unsigned long *)prev_config_info->adm, | |
2212 | (unsigned long *)cur_config_info->adm, | |
2213 | AP_DEVICES); | |
2214 | ||
2215 | if (do_remove) | |
2216 | vfio_ap_mdev_cfg_remove(aprem, aqrem, cdrem); | |
2217 | } | |
2218 | ||
2219 | /** | |
2220 | * vfio_ap_filter_apid_by_qtype: filter APIDs from an AP mask for adapters that | |
2221 | * are older than AP type 10 (CEX4). | |
2222 | * @apm: a bitmap of the APIDs to examine | |
2223 | * @aqm: a bitmap of the APQIs of the queues to query for the AP type. | |
2224 | */ | |
2225 | static void vfio_ap_filter_apid_by_qtype(unsigned long *apm, unsigned long *aqm) | |
2226 | { | |
2227 | bool apid_cleared; | |
2228 | struct ap_queue_status status; | |
211c06d8 HF |
2229 | unsigned long apid, apqi; |
2230 | struct ap_tapq_gr2 info; | |
eeb386ae TK |
2231 | |
2232 | for_each_set_bit_inv(apid, apm, AP_DEVICES) { | |
2233 | apid_cleared = false; | |
2234 | ||
2235 | for_each_set_bit_inv(apqi, aqm, AP_DOMAINS) { | |
2236 | status = ap_test_queue(AP_MKQID(apid, apqi), 1, &info); | |
2237 | switch (status.response_code) { | |
2238 | /* | |
2239 | * According to the architecture in each case | |
2240 | * below, the queue's info should be filled. | |
2241 | */ | |
2242 | case AP_RESPONSE_NORMAL: | |
2243 | case AP_RESPONSE_RESET_IN_PROGRESS: | |
2244 | case AP_RESPONSE_DECONFIGURED: | |
2245 | case AP_RESPONSE_CHECKSTOPPED: | |
2246 | case AP_RESPONSE_BUSY: | |
eeb386ae TK |
2247 | /* |
2248 | * The vfio_ap device driver only | |
2249 | * supports CEX4 and newer adapters, so | |
2250 | * remove the APID if the adapter is | |
2251 | * older than a CEX4. | |
2252 | */ | |
211c06d8 | 2253 | if (info.at < AP_DEVICE_TYPE_CEX4) { |
eeb386ae TK |
2254 | clear_bit_inv(apid, apm); |
2255 | apid_cleared = true; | |
2256 | } | |
2257 | ||
2258 | break; | |
2259 | ||
2260 | default: | |
2261 | /* | |
2262 | * If we don't know the adapter type, | |
2263 | * clear its APID since it can't be | |
2264 | * determined whether the vfio_ap | |
2265 | * device driver supports it. | |
2266 | */ | |
2267 | clear_bit_inv(apid, apm); | |
2268 | apid_cleared = true; | |
2269 | break; | |
2270 | } | |
2271 | ||
2272 | /* | |
2273 | * If we've already cleared the APID from the apm, there | |
2274 | * is no need to continue examining the remainin AP | |
2275 | * queues to determine the type of the adapter. | |
2276 | */ | |
2277 | if (apid_cleared) | |
2278 | continue; | |
2279 | } | |
2280 | } | |
2281 | } | |
2282 | ||
2283 | /** | |
2284 | * vfio_ap_mdev_cfg_add - store bitmaps specifying the adapters, domains and | |
2285 | * control domains that have been added to the host's | |
2286 | * AP configuration for each matrix mdev to which they | |
2287 | * are assigned. | |
2288 | * | |
2289 | * @apm_add: a bitmap specifying the adapters that have been added to the AP | |
2290 | * configuration. | |
2291 | * @aqm_add: a bitmap specifying the domains that have been added to the AP | |
2292 | * configuration. | |
2293 | * @adm_add: a bitmap specifying the control domains that have been added to the | |
2294 | * AP configuration. | |
2295 | */ | |
2296 | static void vfio_ap_mdev_cfg_add(unsigned long *apm_add, unsigned long *aqm_add, | |
2297 | unsigned long *adm_add) | |
2298 | { | |
2299 | struct ap_matrix_mdev *matrix_mdev; | |
2300 | ||
2301 | if (list_empty(&matrix_dev->mdev_list)) | |
2302 | return; | |
2303 | ||
2304 | vfio_ap_filter_apid_by_qtype(apm_add, aqm_add); | |
2305 | ||
2306 | list_for_each_entry(matrix_mdev, &matrix_dev->mdev_list, node) { | |
2307 | bitmap_and(matrix_mdev->apm_add, | |
2308 | matrix_mdev->matrix.apm, apm_add, AP_DEVICES); | |
2309 | bitmap_and(matrix_mdev->aqm_add, | |
2310 | matrix_mdev->matrix.aqm, aqm_add, AP_DOMAINS); | |
2311 | bitmap_and(matrix_mdev->adm_add, | |
2312 | matrix_mdev->matrix.adm, adm_add, AP_DEVICES); | |
2313 | } | |
2314 | } | |
2315 | ||
2316 | /** | |
2317 | * vfio_ap_mdev_on_cfg_add - responds to the addition of adapters, domains and | |
2318 | * control domains to the host AP configuration | |
2319 | * by updating the bitmaps that specify what adapters, | |
2320 | * domains and control domains have been added so they | |
2321 | * can be hot plugged into the guest when the AP bus | |
2322 | * scan completes (see vfio_ap_on_scan_complete | |
2323 | * function). | |
2324 | * @cur_config_info: the current AP configuration information | |
2325 | * @prev_config_info: the previous AP configuration information | |
2326 | */ | |
2327 | static void vfio_ap_mdev_on_cfg_add(struct ap_config_info *cur_config_info, | |
2328 | struct ap_config_info *prev_config_info) | |
2329 | { | |
2330 | bool do_add; | |
2331 | DECLARE_BITMAP(apm_add, AP_DEVICES); | |
2332 | DECLARE_BITMAP(aqm_add, AP_DOMAINS); | |
2333 | DECLARE_BITMAP(adm_add, AP_DOMAINS); | |
2334 | ||
2335 | do_add = bitmap_andnot(apm_add, | |
2336 | (unsigned long *)cur_config_info->apm, | |
2337 | (unsigned long *)prev_config_info->apm, | |
2338 | AP_DEVICES); | |
2339 | do_add |= bitmap_andnot(aqm_add, | |
2340 | (unsigned long *)cur_config_info->aqm, | |
2341 | (unsigned long *)prev_config_info->aqm, | |
2342 | AP_DOMAINS); | |
2343 | do_add |= bitmap_andnot(adm_add, | |
2344 | (unsigned long *)cur_config_info->adm, | |
2345 | (unsigned long *)prev_config_info->adm, | |
2346 | AP_DOMAINS); | |
2347 | ||
2348 | if (do_add) | |
2349 | vfio_ap_mdev_cfg_add(apm_add, aqm_add, adm_add); | |
2350 | } | |
2351 | ||
2352 | /** | |
2353 | * vfio_ap_on_cfg_changed - handles notification of changes to the host AP | |
2354 | * configuration. | |
2355 | * | |
2356 | * @cur_cfg_info: the current host AP configuration | |
2357 | * @prev_cfg_info: the previous host AP configuration | |
2358 | */ | |
2359 | void vfio_ap_on_cfg_changed(struct ap_config_info *cur_cfg_info, | |
2360 | struct ap_config_info *prev_cfg_info) | |
2361 | { | |
2362 | if (!cur_cfg_info || !prev_cfg_info) | |
2363 | return; | |
2364 | ||
2365 | mutex_lock(&matrix_dev->guests_lock); | |
2366 | ||
2367 | vfio_ap_mdev_on_cfg_remove(cur_cfg_info, prev_cfg_info); | |
2368 | vfio_ap_mdev_on_cfg_add(cur_cfg_info, prev_cfg_info); | |
2369 | memcpy(&matrix_dev->info, cur_cfg_info, sizeof(*cur_cfg_info)); | |
2370 | ||
2371 | mutex_unlock(&matrix_dev->guests_lock); | |
2372 | } | |
2373 | ||
2374 | static void vfio_ap_mdev_hot_plug_cfg(struct ap_matrix_mdev *matrix_mdev) | |
2375 | { | |
2376 | bool do_hotplug = false; | |
2377 | int filter_domains = 0; | |
2378 | int filter_adapters = 0; | |
2379 | DECLARE_BITMAP(apm, AP_DEVICES); | |
2380 | DECLARE_BITMAP(aqm, AP_DOMAINS); | |
2381 | ||
2382 | mutex_lock(&matrix_mdev->kvm->lock); | |
2383 | mutex_lock(&matrix_dev->mdevs_lock); | |
2384 | ||
2385 | filter_adapters = bitmap_and(apm, matrix_mdev->matrix.apm, | |
2386 | matrix_mdev->apm_add, AP_DEVICES); | |
2387 | filter_domains = bitmap_and(aqm, matrix_mdev->matrix.aqm, | |
2388 | matrix_mdev->aqm_add, AP_DOMAINS); | |
2389 | ||
2390 | if (filter_adapters && filter_domains) | |
2391 | do_hotplug |= vfio_ap_mdev_filter_matrix(apm, aqm, matrix_mdev); | |
2392 | else if (filter_adapters) | |
2393 | do_hotplug |= | |
2394 | vfio_ap_mdev_filter_matrix(apm, | |
2395 | matrix_mdev->shadow_apcb.aqm, | |
2396 | matrix_mdev); | |
2397 | else | |
2398 | do_hotplug |= | |
2399 | vfio_ap_mdev_filter_matrix(matrix_mdev->shadow_apcb.apm, | |
2400 | aqm, matrix_mdev); | |
2401 | ||
2402 | if (bitmap_intersects(matrix_mdev->matrix.adm, matrix_mdev->adm_add, | |
2403 | AP_DOMAINS)) | |
2404 | do_hotplug |= vfio_ap_mdev_filter_cdoms(matrix_mdev); | |
2405 | ||
2406 | if (do_hotplug) | |
2407 | vfio_ap_mdev_update_guest_apcb(matrix_mdev); | |
2408 | ||
2409 | mutex_unlock(&matrix_dev->mdevs_lock); | |
2410 | mutex_unlock(&matrix_mdev->kvm->lock); | |
2411 | } | |
2412 | ||
2413 | void vfio_ap_on_scan_complete(struct ap_config_info *new_config_info, | |
2414 | struct ap_config_info *old_config_info) | |
2415 | { | |
2416 | struct ap_matrix_mdev *matrix_mdev; | |
2417 | ||
2418 | mutex_lock(&matrix_dev->guests_lock); | |
2419 | ||
2420 | list_for_each_entry(matrix_mdev, &matrix_dev->mdev_list, node) { | |
2421 | if (bitmap_empty(matrix_mdev->apm_add, AP_DEVICES) && | |
2422 | bitmap_empty(matrix_mdev->aqm_add, AP_DOMAINS) && | |
2423 | bitmap_empty(matrix_mdev->adm_add, AP_DOMAINS)) | |
2424 | continue; | |
2425 | ||
2426 | vfio_ap_mdev_hot_plug_cfg(matrix_mdev); | |
2427 | bitmap_clear(matrix_mdev->apm_add, 0, AP_DEVICES); | |
2428 | bitmap_clear(matrix_mdev->aqm_add, 0, AP_DOMAINS); | |
2429 | bitmap_clear(matrix_mdev->adm_add, 0, AP_DOMAINS); | |
2430 | } | |
2431 | ||
2432 | mutex_unlock(&matrix_dev->guests_lock); | |
2433 | } |