1 // SPDX-License-Identifier: GPL-2.0+
3 * Adjunct processor matrix VFIO device driver callbacks.
5 * Copyright IBM Corp. 2018
7 * Author(s): Tony Krowiak <akrowiak@linux.ibm.com>
8 * Halil Pasic <pasic@linux.ibm.com>
9 * Pierre Morel <pmorel@linux.ibm.com>
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>
16 #include <linux/bitops.h>
17 #include <linux/kvm_host.h>
18 #include <linux/module.h>
19 #include <linux/uuid.h>
21 #include <asm/zcrypt.h>
23 #include "vfio_ap_private.h"
24 #include "vfio_ap_debug.h"
26 #define VFIO_AP_MDEV_TYPE_HWVIRT "passthrough"
27 #define VFIO_AP_MDEV_NAME_HWVIRT "VFIO AP Passthrough Device"
29 #define AP_QUEUE_ASSIGNED "assigned"
30 #define AP_QUEUE_UNASSIGNED "unassigned"
31 #define AP_QUEUE_IN_USE "in use"
33 #define AP_RESET_INTERVAL 20 /* Reset sleep interval (20ms) */
35 static int vfio_ap_mdev_reset_queues(struct ap_matrix_mdev *matrix_mdev);
36 static int vfio_ap_mdev_reset_qlist(struct list_head *qlist);
37 static struct vfio_ap_queue *vfio_ap_find_queue(int apqn);
38 static const struct vfio_device_ops vfio_ap_matrix_dev_ops;
39 static void vfio_ap_mdev_reset_queue(struct vfio_ap_queue *q);
42 * get_update_locks_for_kvm: Acquire the locks required to dynamically update a
43 * KVM guest's APCB in the proper order.
45 * @kvm: a pointer to a struct kvm object containing the KVM guest's APCB.
47 * The proper locking order is:
48 * 1. matrix_dev->guests_lock: required to use the KVM pointer to update a KVM
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
53 * Note: If @kvm is NULL, the KVM lock will not be taken.
55 static inline void get_update_locks_for_kvm(struct kvm *kvm)
57 mutex_lock(&matrix_dev->guests_lock);
59 mutex_lock(&kvm->lock);
60 mutex_lock(&matrix_dev->mdevs_lock);
64 * release_update_locks_for_kvm: Release the locks used to dynamically update a
65 * KVM guest's APCB in the proper order.
67 * @kvm: a pointer to a struct kvm object containing the KVM guest's APCB.
69 * The proper unlocking order is:
70 * 1. matrix_dev->mdevs_lock
72 * 3. matrix_dev->guests_lock
74 * Note: If @kvm is NULL, the KVM lock will not be released.
76 static inline void release_update_locks_for_kvm(struct kvm *kvm)
78 mutex_unlock(&matrix_dev->mdevs_lock);
80 mutex_unlock(&kvm->lock);
81 mutex_unlock(&matrix_dev->guests_lock);
85 * get_update_locks_for_mdev: Acquire the locks required to dynamically update a
86 * KVM guest's APCB in the proper order.
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.
91 * The proper locking order is:
92 * 1. matrix_dev->guests_lock: required to use the KVM pointer to update a KVM
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
97 * Note: If @matrix_mdev is NULL or is not attached to a KVM guest, the KVM
98 * lock will not be taken.
100 static inline void get_update_locks_for_mdev(struct ap_matrix_mdev *matrix_mdev)
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);
109 * release_update_locks_for_mdev: Release the locks used to dynamically update a
110 * KVM guest's APCB in the proper order.
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.
115 * The proper unlocking order is:
116 * 1. matrix_dev->mdevs_lock
117 * 2. matrix_mdev->kvm->lock
118 * 3. matrix_dev->guests_lock
120 * Note: If @matrix_mdev is NULL or is not attached to a KVM guest, the KVM
121 * lock will not be released.
123 static inline void release_update_locks_for_mdev(struct ap_matrix_mdev *matrix_mdev)
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);
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.
136 * @apqn: the APQN of a queue device.
138 * The proper locking order is:
139 * 1. matrix_dev->guests_lock: required to use the KVM pointer to update a KVM
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
144 * Note: If @apqn is not assigned to a matrix_mdev, the matrix_mdev->kvm->lock
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.
150 static struct ap_matrix_mdev *get_update_locks_by_apqn(int apqn)
152 struct ap_matrix_mdev *matrix_mdev;
154 mutex_lock(&matrix_dev->guests_lock);
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);
162 mutex_lock(&matrix_dev->mdevs_lock);
168 mutex_lock(&matrix_dev->mdevs_lock);
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.
178 * @q: a pointer to a vfio_ap_queue object.
180 * The proper locking order is:
181 * 1. q->matrix_dev->guests_lock: required to use the KVM pointer to update a
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
186 * Note: if @queue is not linked to an ap_matrix_mdev object, the KVM lock
189 static inline void get_update_locks_for_queue(struct vfio_ap_queue *q)
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);
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
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
206 static struct vfio_ap_queue *vfio_ap_mdev_get_queue(
207 struct ap_matrix_mdev *matrix_mdev,
210 struct vfio_ap_queue *q;
212 hash_for_each_possible(matrix_mdev->qtable.queues, q, mdev_qnode,
214 if (q && q->apqn == apqn)
222 * vfio_ap_wait_for_irqclear - clears the IR bit or gives up after 5 tries
223 * @apqn: The AP Queue number
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
229 * Otherwise retries up to 5 times after waiting 20ms.
231 static void vfio_ap_wait_for_irqclear(int apqn)
233 struct ap_queue_status status;
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)
244 case AP_RESPONSE_BUSY:
247 case AP_RESPONSE_Q_NOT_AVAIL:
248 case AP_RESPONSE_DECONFIGURED:
249 case AP_RESPONSE_CHECKSTOPPED:
251 WARN_ONCE(1, "%s: tapq rc %02x: %04x\n", __func__,
252 status.response_code, apqn);
257 WARN_ONCE(1, "%s: tapq rc %02x: %04x could not clear IR bit\n",
258 __func__, status.response_code, apqn);
262 * vfio_ap_free_aqic_resources - free vfio_ap_queue resources
263 * @q: The vfio_ap_queue
265 * Unregisters the ISC in the GIB when the saved ISC not invalid.
266 * Unpins the guest's page holding the NIB when it exists.
267 * Resets the saved_iova and saved_isc to invalid values.
269 static void vfio_ap_free_aqic_resources(struct vfio_ap_queue *q)
273 if (q->saved_isc != VFIO_AP_ISC_INVALID &&
274 !WARN_ON(!(q->matrix_mdev && q->matrix_mdev->kvm))) {
275 kvm_s390_gisc_unregister(q->matrix_mdev->kvm, q->saved_isc);
276 q->saved_isc = VFIO_AP_ISC_INVALID;
278 if (q->saved_iova && !WARN_ON(!q->matrix_mdev)) {
279 vfio_unpin_pages(&q->matrix_mdev->vdev, q->saved_iova, 1);
285 * vfio_ap_irq_disable - disables and clears an ap_queue interrupt
286 * @q: The vfio_ap_queue
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.
294 * In the case the AP is busy, or a reset is in progress,
295 * retries after 20ms, up to 5 times.
297 * Returns if ap_aqic function failed with invalid, deconfigured or
300 * Return: &struct ap_queue_status
302 static struct ap_queue_status vfio_ap_irq_disable(struct vfio_ap_queue *q)
304 union ap_qirq_ctrl aqic_gisa = { .value = 0 };
305 struct ap_queue_status status;
309 status = ap_aqic(q->apqn, aqic_gisa, 0);
310 switch (status.response_code) {
311 case AP_RESPONSE_OTHERWISE_CHANGED:
312 case AP_RESPONSE_NORMAL:
313 vfio_ap_wait_for_irqclear(q->apqn);
315 case AP_RESPONSE_RESET_IN_PROGRESS:
316 case AP_RESPONSE_BUSY:
319 case AP_RESPONSE_Q_NOT_AVAIL:
320 case AP_RESPONSE_DECONFIGURED:
321 case AP_RESPONSE_CHECKSTOPPED:
322 case AP_RESPONSE_INVALID_ADDRESS:
324 /* All cases in default means AP not operational */
325 WARN_ONCE(1, "%s: ap_aqic status %d\n", __func__,
326 status.response_code);
331 WARN_ONCE(1, "%s: ap_aqic status %d\n", __func__,
332 status.response_code);
334 vfio_ap_free_aqic_resources(q);
339 * vfio_ap_validate_nib - validate a notification indicator byte (nib) address.
341 * @vcpu: the object representing the vcpu executing the PQAP(AQIC) instruction.
342 * @nib: the location for storing the nib address.
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.
346 * This function parses and validates the nib from gr2.
348 * Return: returns zero if the nib address is a valid; otherwise, returns
351 static int vfio_ap_validate_nib(struct kvm_vcpu *vcpu, dma_addr_t *nib)
353 *nib = vcpu->run->s.regs.gprs[2];
357 if (kvm_is_error_hva(gfn_to_hva(vcpu->kvm, *nib >> PAGE_SHIFT)))
363 static int ensure_nib_shared(unsigned long addr, struct gmap *gmap)
368 * The nib has to be located in shared storage since guest and
369 * host access it. vfio_pin_pages() will do a pin shared and
370 * if that fails (possibly because it's not a shared page) it
371 * calls export. We try to do a second pin shared here so that
372 * the UV gives us an error code if we try to pin a non-shared
375 * If the page is already pinned shared the UV will return a success.
377 ret = uv_pin_shared(addr);
379 /* vfio_pin_pages() likely exported the page so let's re-import */
380 gmap_convert_to_secure(gmap, addr);
386 * vfio_ap_irq_enable - Enable Interruption for a APQN
388 * @q: the vfio_ap_queue holding AQIC parameters
389 * @isc: the guest ISC to register with the GIB interface
390 * @vcpu: the vcpu object containing the registers specifying the parameters
391 * passed to the PQAP(AQIC) instruction.
393 * Pin the NIB saved in *q
394 * Register the guest ISC to GIB interface and retrieve the
395 * host ISC to issue the host side PQAP/AQIC
397 * status.response_code may be set to AP_RESPONSE_INVALID_ADDRESS in case the
398 * vfio_pin_pages or kvm_s390_gisc_register failed.
400 * Otherwise return the ap_queue_status returned by the ap_aqic(),
401 * all retry handling will be done by the guest.
403 * Return: &struct ap_queue_status
405 static struct ap_queue_status vfio_ap_irq_enable(struct vfio_ap_queue *q,
407 struct kvm_vcpu *vcpu)
409 union ap_qirq_ctrl aqic_gisa = { .value = 0 };
410 struct ap_queue_status status = {};
411 struct kvm_s390_gisa *gisa;
419 /* Verify that the notification indicator byte address is valid */
420 if (vfio_ap_validate_nib(vcpu, &nib)) {
421 VFIO_AP_DBF_WARN("%s: invalid NIB address: nib=%pad, apqn=%#04x\n",
422 __func__, &nib, q->apqn);
424 status.response_code = AP_RESPONSE_INVALID_ADDRESS;
428 ret = vfio_pin_pages(&q->matrix_mdev->vdev, nib, 1,
429 IOMMU_READ | IOMMU_WRITE, &h_page);
434 VFIO_AP_DBF_WARN("%s: vfio_pin_pages failed: rc=%d,"
435 "nib=%pad, apqn=%#04x\n",
436 __func__, ret, &nib, q->apqn);
438 status.response_code = AP_RESPONSE_INVALID_ADDRESS;
442 kvm = q->matrix_mdev->kvm;
443 gisa = kvm->arch.gisa_int.origin;
445 h_nib = page_to_phys(h_page) | (nib & ~PAGE_MASK);
446 aqic_gisa.gisc = isc;
448 /* NIB in non-shared storage is a rc 6 for PV guests */
449 if (kvm_s390_pv_cpu_is_protected(vcpu) &&
450 ensure_nib_shared(h_nib & PAGE_MASK, kvm->arch.gmap)) {
451 vfio_unpin_pages(&q->matrix_mdev->vdev, nib, 1);
452 status.response_code = AP_RESPONSE_INVALID_ADDRESS;
456 nisc = kvm_s390_gisc_register(kvm, isc);
458 VFIO_AP_DBF_WARN("%s: gisc registration failed: nisc=%d, isc=%d, apqn=%#04x\n",
459 __func__, nisc, isc, q->apqn);
461 vfio_unpin_pages(&q->matrix_mdev->vdev, nib, 1);
462 status.response_code = AP_RESPONSE_INVALID_ADDRESS;
466 aqic_gisa.isc = nisc;
468 aqic_gisa.gisa = virt_to_phys(gisa) >> 4;
470 status = ap_aqic(q->apqn, aqic_gisa, h_nib);
471 switch (status.response_code) {
472 case AP_RESPONSE_NORMAL:
473 /* See if we did clear older IRQ configuration */
474 vfio_ap_free_aqic_resources(q);
478 case AP_RESPONSE_OTHERWISE_CHANGED:
479 /* We could not modify IRQ settings: clear new configuration */
480 ret = kvm_s390_gisc_unregister(kvm, isc);
482 VFIO_AP_DBF_WARN("%s: kvm_s390_gisc_unregister: rc=%d isc=%d, apqn=%#04x\n",
483 __func__, ret, isc, q->apqn);
484 vfio_unpin_pages(&q->matrix_mdev->vdev, nib, 1);
487 pr_warn("%s: apqn %04x: response: %02x\n", __func__, q->apqn,
488 status.response_code);
489 vfio_ap_irq_disable(q);
493 if (status.response_code != AP_RESPONSE_NORMAL) {
494 VFIO_AP_DBF_WARN("%s: PQAP(AQIC) failed with status=%#02x: "
495 "zone=%#x, ir=%#x, gisc=%#x, f=%#x,"
496 "gisa=%#x, isc=%#x, apqn=%#04x\n",
497 __func__, status.response_code,
498 aqic_gisa.zone, aqic_gisa.ir, aqic_gisa.gisc,
499 aqic_gisa.gf, aqic_gisa.gisa, aqic_gisa.isc,
507 * vfio_ap_le_guid_to_be_uuid - convert a little endian guid array into an array
508 * of big endian elements that can be passed by
509 * value to an s390dbf sprintf event function to
510 * format a UUID string.
512 * @guid: the object containing the little endian guid
513 * @uuid: a six-element array of long values that can be passed by value as
514 * arguments for a formatting string specifying a UUID.
516 * The S390 Debug Feature (s390dbf) allows the use of "%s" in the sprintf
517 * event functions if the memory for the passed string is available as long as
518 * the debug feature exists. Since a mediated device can be removed at any
519 * time, it's name can not be used because %s passes the reference to the string
520 * in memory and the reference will go stale once the device is removed .
522 * The s390dbf string formatting function allows a maximum of 9 arguments for a
523 * message to be displayed in the 'sprintf' view. In order to use the bytes
524 * comprising the mediated device's UUID to display the mediated device name,
525 * they will have to be converted into an array whose elements can be passed by
526 * value to sprintf. For example:
528 * guid array: { 83, 78, 17, 62, bb, f1, f0, 47, 91, 4d, 32, a2, 2e, 3a, 88, 04 }
529 * mdev name: 62177883-f1bb-47f0-914d-32a22e3a8804
530 * array returned: { 62177883, f1bb, 47f0, 914d, 32a2, 2e3a8804 }
531 * formatting string: "%08lx-%04lx-%04lx-%04lx-%02lx%04lx"
533 static void vfio_ap_le_guid_to_be_uuid(guid_t *guid, unsigned long *uuid)
536 * The input guid is ordered in little endian, so it needs to be
537 * reordered for displaying a UUID as a string. This specifies the
538 * guid indices in proper order.
540 uuid[0] = le32_to_cpup((__le32 *)guid);
541 uuid[1] = le16_to_cpup((__le16 *)&guid->b[4]);
542 uuid[2] = le16_to_cpup((__le16 *)&guid->b[6]);
543 uuid[3] = *((__u16 *)&guid->b[8]);
544 uuid[4] = *((__u16 *)&guid->b[10]);
545 uuid[5] = *((__u32 *)&guid->b[12]);
549 * handle_pqap - PQAP instruction callback
551 * @vcpu: The vcpu on which we received the PQAP instruction
553 * Get the general register contents to initialize internal variables.
558 * Response.status may be set to following Response Code:
559 * - AP_RESPONSE_Q_NOT_AVAIL: if the queue is not available
560 * - AP_RESPONSE_DECONFIGURED: if the queue is not configured
561 * - AP_RESPONSE_NORMAL (0) : in case of success
562 * Check vfio_ap_setirq() and vfio_ap_clrirq() for other possible RC.
563 * We take the matrix_dev lock to ensure serialization on queues and
564 * mediated device access.
566 * Return: 0 if we could handle the request inside KVM.
567 * Otherwise, returns -EOPNOTSUPP to let QEMU handle the fault.
569 static int handle_pqap(struct kvm_vcpu *vcpu)
573 unsigned long uuid[6];
574 struct vfio_ap_queue *q;
575 struct ap_queue_status qstatus = {
576 .response_code = AP_RESPONSE_Q_NOT_AVAIL, };
577 struct ap_matrix_mdev *matrix_mdev;
579 apqn = vcpu->run->s.regs.gprs[0] & 0xffff;
581 /* If we do not use the AIV facility just go to userland */
582 if (!(vcpu->arch.sie_block->eca & ECA_AIV)) {
583 VFIO_AP_DBF_WARN("%s: AIV facility not installed: apqn=0x%04x, eca=0x%04x\n",
584 __func__, apqn, vcpu->arch.sie_block->eca);
589 mutex_lock(&matrix_dev->mdevs_lock);
591 if (!vcpu->kvm->arch.crypto.pqap_hook) {
592 VFIO_AP_DBF_WARN("%s: PQAP(AQIC) hook not registered with the vfio_ap driver: apqn=0x%04x\n",
598 matrix_mdev = container_of(vcpu->kvm->arch.crypto.pqap_hook,
599 struct ap_matrix_mdev, pqap_hook);
601 /* If the there is no guest using the mdev, there is nothing to do */
602 if (!matrix_mdev->kvm) {
603 vfio_ap_le_guid_to_be_uuid(&matrix_mdev->mdev->uuid, uuid);
604 VFIO_AP_DBF_WARN("%s: mdev %08lx-%04lx-%04lx-%04lx-%04lx%08lx not in use: apqn=0x%04x\n",
605 __func__, uuid[0], uuid[1], uuid[2],
606 uuid[3], uuid[4], uuid[5], apqn);
610 q = vfio_ap_mdev_get_queue(matrix_mdev, apqn);
612 VFIO_AP_DBF_WARN("%s: Queue %02x.%04x not bound to the vfio_ap driver\n",
613 __func__, AP_QID_CARD(apqn),
618 status = vcpu->run->s.regs.gprs[1];
620 /* If IR bit(16) is set we enable the interrupt */
621 if ((status >> (63 - 16)) & 0x01)
622 qstatus = vfio_ap_irq_enable(q, status & 0x07, vcpu);
624 qstatus = vfio_ap_irq_disable(q);
627 memcpy(&vcpu->run->s.regs.gprs[1], &qstatus, sizeof(qstatus));
628 vcpu->run->s.regs.gprs[1] >>= 32;
629 mutex_unlock(&matrix_dev->mdevs_lock);
633 static void vfio_ap_matrix_init(struct ap_config_info *info,
634 struct ap_matrix *matrix)
636 matrix->apm_max = info->apxa ? info->na : 63;
637 matrix->aqm_max = info->apxa ? info->nd : 15;
638 matrix->adm_max = info->apxa ? info->nd : 15;
641 static void vfio_ap_mdev_update_guest_apcb(struct ap_matrix_mdev *matrix_mdev)
643 if (matrix_mdev->kvm)
644 kvm_arch_crypto_set_masks(matrix_mdev->kvm,
645 matrix_mdev->shadow_apcb.apm,
646 matrix_mdev->shadow_apcb.aqm,
647 matrix_mdev->shadow_apcb.adm);
650 static bool vfio_ap_mdev_filter_cdoms(struct ap_matrix_mdev *matrix_mdev)
652 DECLARE_BITMAP(prev_shadow_adm, AP_DOMAINS);
654 bitmap_copy(prev_shadow_adm, matrix_mdev->shadow_apcb.adm, AP_DOMAINS);
655 bitmap_and(matrix_mdev->shadow_apcb.adm, matrix_mdev->matrix.adm,
656 (unsigned long *)matrix_dev->info.adm, AP_DOMAINS);
658 return !bitmap_equal(prev_shadow_adm, matrix_mdev->shadow_apcb.adm,
662 static bool _queue_passable(struct vfio_ap_queue *q)
667 switch (q->reset_status.response_code) {
668 case AP_RESPONSE_NORMAL:
669 case AP_RESPONSE_DECONFIGURED:
670 case AP_RESPONSE_CHECKSTOPPED:
678 * vfio_ap_mdev_filter_matrix - filter the APQNs assigned to the matrix mdev
679 * to ensure no queue devices are passed through to
680 * the guest that are not bound to the vfio_ap
683 * @matrix_mdev: the matrix mdev whose matrix is to be filtered.
684 * @apm_filtered: a 256-bit bitmap for storing the APIDs filtered from the
685 * guest's AP configuration that are still in the host's AP
688 * Note: If an APQN referencing a queue device that is not bound to the vfio_ap
689 * driver, its APID will be filtered from the guest's APCB. The matrix
690 * structure precludes filtering an individual APQN, so its APID will be
691 * filtered. Consequently, all queues associated with the adapter that
692 * are in the host's AP configuration must be reset. If queues are
693 * subsequently made available again to the guest, they should re-appear
696 * Return: a boolean value indicating whether the KVM guest's APCB was changed
697 * by the filtering or not.
699 static bool vfio_ap_mdev_filter_matrix(struct ap_matrix_mdev *matrix_mdev,
700 unsigned long *apm_filtered)
702 unsigned long apid, apqi, apqn;
703 DECLARE_BITMAP(prev_shadow_apm, AP_DEVICES);
704 DECLARE_BITMAP(prev_shadow_aqm, AP_DOMAINS);
706 bitmap_copy(prev_shadow_apm, matrix_mdev->shadow_apcb.apm, AP_DEVICES);
707 bitmap_copy(prev_shadow_aqm, matrix_mdev->shadow_apcb.aqm, AP_DOMAINS);
708 vfio_ap_matrix_init(&matrix_dev->info, &matrix_mdev->shadow_apcb);
709 bitmap_clear(apm_filtered, 0, AP_DEVICES);
712 * Copy the adapters, domains and control domains to the shadow_apcb
713 * from the matrix mdev, but only those that are assigned to the host's
716 bitmap_and(matrix_mdev->shadow_apcb.apm, matrix_mdev->matrix.apm,
717 (unsigned long *)matrix_dev->info.apm, AP_DEVICES);
718 bitmap_and(matrix_mdev->shadow_apcb.aqm, matrix_mdev->matrix.aqm,
719 (unsigned long *)matrix_dev->info.aqm, AP_DOMAINS);
721 for_each_set_bit_inv(apid, matrix_mdev->shadow_apcb.apm, AP_DEVICES) {
722 for_each_set_bit_inv(apqi, matrix_mdev->shadow_apcb.aqm,
725 * If the APQN is not bound to the vfio_ap device
726 * driver, then we can't assign it to the guest's
727 * AP configuration. The AP architecture won't
728 * allow filtering of a single APQN, so let's filter
729 * the APID since an adapter represents a physical
732 apqn = AP_MKQID(apid, apqi);
733 if (!_queue_passable(vfio_ap_mdev_get_queue(matrix_mdev, apqn))) {
734 clear_bit_inv(apid, matrix_mdev->shadow_apcb.apm);
737 * If the adapter was previously plugged into
738 * the guest, let's let the caller know that
739 * the APID was filtered.
741 if (test_bit_inv(apid, prev_shadow_apm))
742 set_bit_inv(apid, apm_filtered);
749 return !bitmap_equal(prev_shadow_apm, matrix_mdev->shadow_apcb.apm,
751 !bitmap_equal(prev_shadow_aqm, matrix_mdev->shadow_apcb.aqm,
755 static int vfio_ap_mdev_init_dev(struct vfio_device *vdev)
757 struct ap_matrix_mdev *matrix_mdev =
758 container_of(vdev, struct ap_matrix_mdev, vdev);
760 matrix_mdev->mdev = to_mdev_device(vdev->dev);
761 vfio_ap_matrix_init(&matrix_dev->info, &matrix_mdev->matrix);
762 matrix_mdev->pqap_hook = handle_pqap;
763 vfio_ap_matrix_init(&matrix_dev->info, &matrix_mdev->shadow_apcb);
764 hash_init(matrix_mdev->qtable.queues);
769 static int vfio_ap_mdev_probe(struct mdev_device *mdev)
771 struct ap_matrix_mdev *matrix_mdev;
774 matrix_mdev = vfio_alloc_device(ap_matrix_mdev, vdev, &mdev->dev,
775 &vfio_ap_matrix_dev_ops);
776 if (IS_ERR(matrix_mdev))
777 return PTR_ERR(matrix_mdev);
779 ret = vfio_register_emulated_iommu_dev(&matrix_mdev->vdev);
782 matrix_mdev->req_trigger = NULL;
783 dev_set_drvdata(&mdev->dev, matrix_mdev);
784 mutex_lock(&matrix_dev->mdevs_lock);
785 list_add(&matrix_mdev->node, &matrix_dev->mdev_list);
786 mutex_unlock(&matrix_dev->mdevs_lock);
790 vfio_put_device(&matrix_mdev->vdev);
794 static void vfio_ap_mdev_link_queue(struct ap_matrix_mdev *matrix_mdev,
795 struct vfio_ap_queue *q)
798 q->matrix_mdev = matrix_mdev;
799 hash_add(matrix_mdev->qtable.queues, &q->mdev_qnode, q->apqn);
803 static void vfio_ap_mdev_link_apqn(struct ap_matrix_mdev *matrix_mdev, int apqn)
805 struct vfio_ap_queue *q;
807 q = vfio_ap_find_queue(apqn);
808 vfio_ap_mdev_link_queue(matrix_mdev, q);
811 static void vfio_ap_unlink_queue_fr_mdev(struct vfio_ap_queue *q)
813 hash_del(&q->mdev_qnode);
816 static void vfio_ap_unlink_mdev_fr_queue(struct vfio_ap_queue *q)
818 q->matrix_mdev = NULL;
821 static void vfio_ap_mdev_unlink_fr_queues(struct ap_matrix_mdev *matrix_mdev)
823 struct vfio_ap_queue *q;
824 unsigned long apid, apqi;
826 for_each_set_bit_inv(apid, matrix_mdev->matrix.apm, AP_DEVICES) {
827 for_each_set_bit_inv(apqi, matrix_mdev->matrix.aqm,
829 q = vfio_ap_mdev_get_queue(matrix_mdev,
830 AP_MKQID(apid, apqi));
832 q->matrix_mdev = NULL;
837 static void vfio_ap_mdev_remove(struct mdev_device *mdev)
839 struct ap_matrix_mdev *matrix_mdev = dev_get_drvdata(&mdev->dev);
841 vfio_unregister_group_dev(&matrix_mdev->vdev);
843 mutex_lock(&matrix_dev->guests_lock);
844 mutex_lock(&matrix_dev->mdevs_lock);
845 vfio_ap_mdev_reset_queues(matrix_mdev);
846 vfio_ap_mdev_unlink_fr_queues(matrix_mdev);
847 list_del(&matrix_mdev->node);
848 mutex_unlock(&matrix_dev->mdevs_lock);
849 mutex_unlock(&matrix_dev->guests_lock);
850 vfio_put_device(&matrix_mdev->vdev);
853 #define MDEV_SHARING_ERR "Userspace may not re-assign queue %02lx.%04lx " \
854 "already assigned to %s"
856 static void vfio_ap_mdev_log_sharing_err(struct ap_matrix_mdev *matrix_mdev,
860 unsigned long apid, apqi;
861 const struct device *dev = mdev_dev(matrix_mdev->mdev);
862 const char *mdev_name = dev_name(dev);
864 for_each_set_bit_inv(apid, apm, AP_DEVICES)
865 for_each_set_bit_inv(apqi, aqm, AP_DOMAINS)
866 dev_warn(dev, MDEV_SHARING_ERR, apid, apqi, mdev_name);
870 * vfio_ap_mdev_verify_no_sharing - verify APQNs are not shared by matrix mdevs
872 * @mdev_apm: mask indicating the APIDs of the APQNs to be verified
873 * @mdev_aqm: mask indicating the APQIs of the APQNs to be verified
875 * Verifies that each APQN derived from the Cartesian product of a bitmap of
876 * AP adapter IDs and AP queue indexes is not configured for any matrix
877 * mediated device. AP queue sharing is not allowed.
879 * Return: 0 if the APQNs are not shared; otherwise return -EADDRINUSE.
881 static int vfio_ap_mdev_verify_no_sharing(unsigned long *mdev_apm,
882 unsigned long *mdev_aqm)
884 struct ap_matrix_mdev *matrix_mdev;
885 DECLARE_BITMAP(apm, AP_DEVICES);
886 DECLARE_BITMAP(aqm, AP_DOMAINS);
888 list_for_each_entry(matrix_mdev, &matrix_dev->mdev_list, node) {
890 * If the input apm and aqm are fields of the matrix_mdev
891 * object, then move on to the next matrix_mdev.
893 if (mdev_apm == matrix_mdev->matrix.apm &&
894 mdev_aqm == matrix_mdev->matrix.aqm)
897 memset(apm, 0, sizeof(apm));
898 memset(aqm, 0, sizeof(aqm));
901 * We work on full longs, as we can only exclude the leftover
902 * bits in non-inverse order. The leftover is all zeros.
904 if (!bitmap_and(apm, mdev_apm, matrix_mdev->matrix.apm,
908 if (!bitmap_and(aqm, mdev_aqm, matrix_mdev->matrix.aqm,
912 vfio_ap_mdev_log_sharing_err(matrix_mdev, apm, aqm);
921 * vfio_ap_mdev_validate_masks - verify that the APQNs assigned to the mdev are
922 * not reserved for the default zcrypt driver and
923 * are not assigned to another mdev.
925 * @matrix_mdev: the mdev to which the APQNs being validated are assigned.
927 * Return: One of the following values:
928 * o the error returned from the ap_apqn_in_matrix_owned_by_def_drv() function,
929 * most likely -EBUSY indicating the ap_perms_mutex lock is already held.
930 * o EADDRNOTAVAIL if an APQN assigned to @matrix_mdev is reserved for the
931 * zcrypt default driver.
932 * o EADDRINUSE if an APQN assigned to @matrix_mdev is assigned to another mdev
933 * o A zero indicating validation succeeded.
935 static int vfio_ap_mdev_validate_masks(struct ap_matrix_mdev *matrix_mdev)
937 if (ap_apqn_in_matrix_owned_by_def_drv(matrix_mdev->matrix.apm,
938 matrix_mdev->matrix.aqm))
939 return -EADDRNOTAVAIL;
941 return vfio_ap_mdev_verify_no_sharing(matrix_mdev->matrix.apm,
942 matrix_mdev->matrix.aqm);
945 static void vfio_ap_mdev_link_adapter(struct ap_matrix_mdev *matrix_mdev,
950 for_each_set_bit_inv(apqi, matrix_mdev->matrix.aqm, AP_DOMAINS)
951 vfio_ap_mdev_link_apqn(matrix_mdev,
952 AP_MKQID(apid, apqi));
955 static void collect_queues_to_reset(struct ap_matrix_mdev *matrix_mdev,
957 struct list_head *qlist)
959 struct vfio_ap_queue *q;
962 for_each_set_bit_inv(apqi, matrix_mdev->shadow_apcb.aqm, AP_DOMAINS) {
963 q = vfio_ap_mdev_get_queue(matrix_mdev, AP_MKQID(apid, apqi));
965 list_add_tail(&q->reset_qnode, qlist);
969 static void reset_queues_for_apid(struct ap_matrix_mdev *matrix_mdev,
972 struct list_head qlist;
974 INIT_LIST_HEAD(&qlist);
975 collect_queues_to_reset(matrix_mdev, apid, &qlist);
976 vfio_ap_mdev_reset_qlist(&qlist);
979 static int reset_queues_for_apids(struct ap_matrix_mdev *matrix_mdev,
980 unsigned long *apm_reset)
982 struct list_head qlist;
985 if (bitmap_empty(apm_reset, AP_DEVICES))
988 INIT_LIST_HEAD(&qlist);
990 for_each_set_bit_inv(apid, apm_reset, AP_DEVICES)
991 collect_queues_to_reset(matrix_mdev, apid, &qlist);
993 return vfio_ap_mdev_reset_qlist(&qlist);
997 * assign_adapter_store - parses the APID from @buf and sets the
998 * corresponding bit in the mediated matrix device's APM
1000 * @dev: the matrix device
1001 * @attr: the mediated matrix device's assign_adapter attribute
1002 * @buf: a buffer containing the AP adapter number (APID) to
1004 * @count: the number of bytes in @buf
1006 * Return: the number of bytes processed if the APID is valid; otherwise,
1007 * returns one of the following errors:
1010 * The APID is not a valid number
1013 * The APID exceeds the maximum value configured for the system
1016 * An APQN derived from the cross product of the APID being assigned
1017 * and the APQIs previously assigned is not bound to the vfio_ap device
1018 * driver; or, if no APQIs have yet been assigned, the APID is not
1019 * contained in an APQN bound to the vfio_ap device driver.
1022 * An APQN derived from the cross product of the APID being assigned
1023 * and the APQIs previously assigned is being used by another mediated
1027 * A lock required to validate the mdev's AP configuration could not
1030 static ssize_t assign_adapter_store(struct device *dev,
1031 struct device_attribute *attr,
1032 const char *buf, size_t count)
1036 DECLARE_BITMAP(apm_filtered, AP_DEVICES);
1037 struct ap_matrix_mdev *matrix_mdev = dev_get_drvdata(dev);
1039 mutex_lock(&ap_perms_mutex);
1040 get_update_locks_for_mdev(matrix_mdev);
1042 ret = kstrtoul(buf, 0, &apid);
1046 if (apid > matrix_mdev->matrix.apm_max) {
1051 if (test_bit_inv(apid, matrix_mdev->matrix.apm)) {
1056 set_bit_inv(apid, matrix_mdev->matrix.apm);
1058 ret = vfio_ap_mdev_validate_masks(matrix_mdev);
1060 clear_bit_inv(apid, matrix_mdev->matrix.apm);
1064 vfio_ap_mdev_link_adapter(matrix_mdev, apid);
1066 if (vfio_ap_mdev_filter_matrix(matrix_mdev, apm_filtered)) {
1067 vfio_ap_mdev_update_guest_apcb(matrix_mdev);
1068 reset_queues_for_apids(matrix_mdev, apm_filtered);
1073 release_update_locks_for_mdev(matrix_mdev);
1074 mutex_unlock(&ap_perms_mutex);
1078 static DEVICE_ATTR_WO(assign_adapter);
1080 static struct vfio_ap_queue
1081 *vfio_ap_unlink_apqn_fr_mdev(struct ap_matrix_mdev *matrix_mdev,
1082 unsigned long apid, unsigned long apqi)
1084 struct vfio_ap_queue *q = NULL;
1086 q = vfio_ap_mdev_get_queue(matrix_mdev, AP_MKQID(apid, apqi));
1087 /* If the queue is assigned to the matrix mdev, unlink it. */
1089 vfio_ap_unlink_queue_fr_mdev(q);
1095 * vfio_ap_mdev_unlink_adapter - unlink all queues associated with unassigned
1096 * adapter from the matrix mdev to which the
1097 * adapter was assigned.
1098 * @matrix_mdev: the matrix mediated device to which the adapter was assigned.
1099 * @apid: the APID of the unassigned adapter.
1100 * @qlist: list for storing queues associated with unassigned adapter that
1103 static void vfio_ap_mdev_unlink_adapter(struct ap_matrix_mdev *matrix_mdev,
1105 struct list_head *qlist)
1108 struct vfio_ap_queue *q;
1110 for_each_set_bit_inv(apqi, matrix_mdev->matrix.aqm, AP_DOMAINS) {
1111 q = vfio_ap_unlink_apqn_fr_mdev(matrix_mdev, apid, apqi);
1114 if (test_bit_inv(apid, matrix_mdev->shadow_apcb.apm) &&
1115 test_bit_inv(apqi, matrix_mdev->shadow_apcb.aqm))
1116 list_add_tail(&q->reset_qnode, qlist);
1121 static void vfio_ap_mdev_hot_unplug_adapter(struct ap_matrix_mdev *matrix_mdev,
1124 struct vfio_ap_queue *q, *tmpq;
1125 struct list_head qlist;
1127 INIT_LIST_HEAD(&qlist);
1128 vfio_ap_mdev_unlink_adapter(matrix_mdev, apid, &qlist);
1130 if (test_bit_inv(apid, matrix_mdev->shadow_apcb.apm)) {
1131 clear_bit_inv(apid, matrix_mdev->shadow_apcb.apm);
1132 vfio_ap_mdev_update_guest_apcb(matrix_mdev);
1135 vfio_ap_mdev_reset_qlist(&qlist);
1137 list_for_each_entry_safe(q, tmpq, &qlist, reset_qnode) {
1138 vfio_ap_unlink_mdev_fr_queue(q);
1139 list_del(&q->reset_qnode);
1144 * unassign_adapter_store - parses the APID from @buf and clears the
1145 * corresponding bit in the mediated matrix device's APM
1147 * @dev: the matrix device
1148 * @attr: the mediated matrix device's unassign_adapter attribute
1149 * @buf: a buffer containing the adapter number (APID) to be unassigned
1150 * @count: the number of bytes in @buf
1152 * Return: the number of bytes processed if the APID is valid; otherwise,
1153 * returns one of the following errors:
1154 * -EINVAL if the APID is not a number
1155 * -ENODEV if the APID it exceeds the maximum value configured for the
1158 static ssize_t unassign_adapter_store(struct device *dev,
1159 struct device_attribute *attr,
1160 const char *buf, size_t count)
1164 struct ap_matrix_mdev *matrix_mdev = dev_get_drvdata(dev);
1166 get_update_locks_for_mdev(matrix_mdev);
1168 ret = kstrtoul(buf, 0, &apid);
1172 if (apid > matrix_mdev->matrix.apm_max) {
1177 if (!test_bit_inv(apid, matrix_mdev->matrix.apm)) {
1182 clear_bit_inv((unsigned long)apid, matrix_mdev->matrix.apm);
1183 vfio_ap_mdev_hot_unplug_adapter(matrix_mdev, apid);
1186 release_update_locks_for_mdev(matrix_mdev);
1189 static DEVICE_ATTR_WO(unassign_adapter);
1191 static void vfio_ap_mdev_link_domain(struct ap_matrix_mdev *matrix_mdev,
1196 for_each_set_bit_inv(apid, matrix_mdev->matrix.apm, AP_DEVICES)
1197 vfio_ap_mdev_link_apqn(matrix_mdev,
1198 AP_MKQID(apid, apqi));
1202 * assign_domain_store - parses the APQI from @buf and sets the
1203 * corresponding bit in the mediated matrix device's AQM
1205 * @dev: the matrix device
1206 * @attr: the mediated matrix device's assign_domain attribute
1207 * @buf: a buffer containing the AP queue index (APQI) of the domain to
1209 * @count: the number of bytes in @buf
1211 * Return: the number of bytes processed if the APQI is valid; otherwise returns
1212 * one of the following errors:
1215 * The APQI is not a valid number
1218 * The APQI exceeds the maximum value configured for the system
1221 * An APQN derived from the cross product of the APQI being assigned
1222 * and the APIDs previously assigned is not bound to the vfio_ap device
1223 * driver; or, if no APIDs have yet been assigned, the APQI is not
1224 * contained in an APQN bound to the vfio_ap device driver.
1227 * An APQN derived from the cross product of the APQI being assigned
1228 * and the APIDs previously assigned is being used by another mediated
1232 * The lock required to validate the mdev's AP configuration could not
1235 static ssize_t assign_domain_store(struct device *dev,
1236 struct device_attribute *attr,
1237 const char *buf, size_t count)
1241 DECLARE_BITMAP(apm_filtered, AP_DEVICES);
1242 struct ap_matrix_mdev *matrix_mdev = dev_get_drvdata(dev);
1244 mutex_lock(&ap_perms_mutex);
1245 get_update_locks_for_mdev(matrix_mdev);
1247 ret = kstrtoul(buf, 0, &apqi);
1251 if (apqi > matrix_mdev->matrix.aqm_max) {
1256 if (test_bit_inv(apqi, matrix_mdev->matrix.aqm)) {
1261 set_bit_inv(apqi, matrix_mdev->matrix.aqm);
1263 ret = vfio_ap_mdev_validate_masks(matrix_mdev);
1265 clear_bit_inv(apqi, matrix_mdev->matrix.aqm);
1269 vfio_ap_mdev_link_domain(matrix_mdev, apqi);
1271 if (vfio_ap_mdev_filter_matrix(matrix_mdev, apm_filtered)) {
1272 vfio_ap_mdev_update_guest_apcb(matrix_mdev);
1273 reset_queues_for_apids(matrix_mdev, apm_filtered);
1278 release_update_locks_for_mdev(matrix_mdev);
1279 mutex_unlock(&ap_perms_mutex);
1283 static DEVICE_ATTR_WO(assign_domain);
1285 static void vfio_ap_mdev_unlink_domain(struct ap_matrix_mdev *matrix_mdev,
1287 struct list_head *qlist)
1290 struct vfio_ap_queue *q;
1292 for_each_set_bit_inv(apid, matrix_mdev->matrix.apm, AP_DEVICES) {
1293 q = vfio_ap_unlink_apqn_fr_mdev(matrix_mdev, apid, apqi);
1296 if (test_bit_inv(apid, matrix_mdev->shadow_apcb.apm) &&
1297 test_bit_inv(apqi, matrix_mdev->shadow_apcb.aqm))
1298 list_add_tail(&q->reset_qnode, qlist);
1303 static void vfio_ap_mdev_hot_unplug_domain(struct ap_matrix_mdev *matrix_mdev,
1306 struct vfio_ap_queue *q, *tmpq;
1307 struct list_head qlist;
1309 INIT_LIST_HEAD(&qlist);
1310 vfio_ap_mdev_unlink_domain(matrix_mdev, apqi, &qlist);
1312 if (test_bit_inv(apqi, matrix_mdev->shadow_apcb.aqm)) {
1313 clear_bit_inv(apqi, matrix_mdev->shadow_apcb.aqm);
1314 vfio_ap_mdev_update_guest_apcb(matrix_mdev);
1317 vfio_ap_mdev_reset_qlist(&qlist);
1319 list_for_each_entry_safe(q, tmpq, &qlist, reset_qnode) {
1320 vfio_ap_unlink_mdev_fr_queue(q);
1321 list_del(&q->reset_qnode);
1326 * unassign_domain_store - parses the APQI from @buf and clears the
1327 * corresponding bit in the mediated matrix device's AQM
1329 * @dev: the matrix device
1330 * @attr: the mediated matrix device's unassign_domain attribute
1331 * @buf: a buffer containing the AP queue index (APQI) of the domain to
1333 * @count: the number of bytes in @buf
1335 * Return: the number of bytes processed if the APQI is valid; otherwise,
1336 * returns one of the following errors:
1337 * -EINVAL if the APQI is not a number
1338 * -ENODEV if the APQI exceeds the maximum value configured for the system
1340 static ssize_t unassign_domain_store(struct device *dev,
1341 struct device_attribute *attr,
1342 const char *buf, size_t count)
1346 struct ap_matrix_mdev *matrix_mdev = dev_get_drvdata(dev);
1348 get_update_locks_for_mdev(matrix_mdev);
1350 ret = kstrtoul(buf, 0, &apqi);
1354 if (apqi > matrix_mdev->matrix.aqm_max) {
1359 if (!test_bit_inv(apqi, matrix_mdev->matrix.aqm)) {
1364 clear_bit_inv((unsigned long)apqi, matrix_mdev->matrix.aqm);
1365 vfio_ap_mdev_hot_unplug_domain(matrix_mdev, apqi);
1369 release_update_locks_for_mdev(matrix_mdev);
1372 static DEVICE_ATTR_WO(unassign_domain);
1375 * assign_control_domain_store - parses the domain ID from @buf and sets
1376 * the corresponding bit in the mediated matrix device's ADM
1378 * @dev: the matrix device
1379 * @attr: the mediated matrix device's assign_control_domain attribute
1380 * @buf: a buffer containing the domain ID to be assigned
1381 * @count: the number of bytes in @buf
1383 * Return: the number of bytes processed if the domain ID is valid; otherwise,
1384 * returns one of the following errors:
1385 * -EINVAL if the ID is not a number
1386 * -ENODEV if the ID exceeds the maximum value configured for the system
1388 static ssize_t assign_control_domain_store(struct device *dev,
1389 struct device_attribute *attr,
1390 const char *buf, size_t count)
1394 struct ap_matrix_mdev *matrix_mdev = dev_get_drvdata(dev);
1396 get_update_locks_for_mdev(matrix_mdev);
1398 ret = kstrtoul(buf, 0, &id);
1402 if (id > matrix_mdev->matrix.adm_max) {
1407 if (test_bit_inv(id, matrix_mdev->matrix.adm)) {
1412 /* Set the bit in the ADM (bitmask) corresponding to the AP control
1413 * domain number (id). The bits in the mask, from most significant to
1414 * least significant, correspond to IDs 0 up to the one less than the
1415 * number of control domains that can be assigned.
1417 set_bit_inv(id, matrix_mdev->matrix.adm);
1418 if (vfio_ap_mdev_filter_cdoms(matrix_mdev))
1419 vfio_ap_mdev_update_guest_apcb(matrix_mdev);
1423 release_update_locks_for_mdev(matrix_mdev);
1426 static DEVICE_ATTR_WO(assign_control_domain);
1429 * unassign_control_domain_store - parses the domain ID from @buf and
1430 * clears the corresponding bit in the mediated matrix device's ADM
1432 * @dev: the matrix device
1433 * @attr: the mediated matrix device's unassign_control_domain attribute
1434 * @buf: a buffer containing the domain ID to be unassigned
1435 * @count: the number of bytes in @buf
1437 * Return: the number of bytes processed if the domain ID is valid; otherwise,
1438 * returns one of the following errors:
1439 * -EINVAL if the ID is not a number
1440 * -ENODEV if the ID exceeds the maximum value configured for the system
1442 static ssize_t unassign_control_domain_store(struct device *dev,
1443 struct device_attribute *attr,
1444 const char *buf, size_t count)
1447 unsigned long domid;
1448 struct ap_matrix_mdev *matrix_mdev = dev_get_drvdata(dev);
1450 get_update_locks_for_mdev(matrix_mdev);
1452 ret = kstrtoul(buf, 0, &domid);
1456 if (domid > matrix_mdev->matrix.adm_max) {
1461 if (!test_bit_inv(domid, matrix_mdev->matrix.adm)) {
1466 clear_bit_inv(domid, matrix_mdev->matrix.adm);
1468 if (test_bit_inv(domid, matrix_mdev->shadow_apcb.adm)) {
1469 clear_bit_inv(domid, matrix_mdev->shadow_apcb.adm);
1470 vfio_ap_mdev_update_guest_apcb(matrix_mdev);
1475 release_update_locks_for_mdev(matrix_mdev);
1478 static DEVICE_ATTR_WO(unassign_control_domain);
1480 static ssize_t control_domains_show(struct device *dev,
1481 struct device_attribute *dev_attr,
1488 struct ap_matrix_mdev *matrix_mdev = dev_get_drvdata(dev);
1489 unsigned long max_domid = matrix_mdev->matrix.adm_max;
1491 mutex_lock(&matrix_dev->mdevs_lock);
1492 for_each_set_bit_inv(id, matrix_mdev->matrix.adm, max_domid + 1) {
1493 n = sprintf(bufpos, "%04lx\n", id);
1497 mutex_unlock(&matrix_dev->mdevs_lock);
1501 static DEVICE_ATTR_RO(control_domains);
1503 static ssize_t vfio_ap_mdev_matrix_show(struct ap_matrix *matrix, char *buf)
1508 unsigned long apid1;
1509 unsigned long apqi1;
1510 unsigned long napm_bits = matrix->apm_max + 1;
1511 unsigned long naqm_bits = matrix->aqm_max + 1;
1515 apid1 = find_first_bit_inv(matrix->apm, napm_bits);
1516 apqi1 = find_first_bit_inv(matrix->aqm, naqm_bits);
1518 if ((apid1 < napm_bits) && (apqi1 < naqm_bits)) {
1519 for_each_set_bit_inv(apid, matrix->apm, napm_bits) {
1520 for_each_set_bit_inv(apqi, matrix->aqm,
1522 n = sprintf(bufpos, "%02lx.%04lx\n", apid,
1528 } else if (apid1 < napm_bits) {
1529 for_each_set_bit_inv(apid, matrix->apm, napm_bits) {
1530 n = sprintf(bufpos, "%02lx.\n", apid);
1534 } else if (apqi1 < naqm_bits) {
1535 for_each_set_bit_inv(apqi, matrix->aqm, naqm_bits) {
1536 n = sprintf(bufpos, ".%04lx\n", apqi);
1545 static ssize_t matrix_show(struct device *dev, struct device_attribute *attr,
1549 struct ap_matrix_mdev *matrix_mdev = dev_get_drvdata(dev);
1551 mutex_lock(&matrix_dev->mdevs_lock);
1552 nchars = vfio_ap_mdev_matrix_show(&matrix_mdev->matrix, buf);
1553 mutex_unlock(&matrix_dev->mdevs_lock);
1557 static DEVICE_ATTR_RO(matrix);
1559 static ssize_t guest_matrix_show(struct device *dev,
1560 struct device_attribute *attr, char *buf)
1563 struct ap_matrix_mdev *matrix_mdev = dev_get_drvdata(dev);
1565 mutex_lock(&matrix_dev->mdevs_lock);
1566 nchars = vfio_ap_mdev_matrix_show(&matrix_mdev->shadow_apcb, buf);
1567 mutex_unlock(&matrix_dev->mdevs_lock);
1571 static DEVICE_ATTR_RO(guest_matrix);
1573 static struct attribute *vfio_ap_mdev_attrs[] = {
1574 &dev_attr_assign_adapter.attr,
1575 &dev_attr_unassign_adapter.attr,
1576 &dev_attr_assign_domain.attr,
1577 &dev_attr_unassign_domain.attr,
1578 &dev_attr_assign_control_domain.attr,
1579 &dev_attr_unassign_control_domain.attr,
1580 &dev_attr_control_domains.attr,
1581 &dev_attr_matrix.attr,
1582 &dev_attr_guest_matrix.attr,
1586 static struct attribute_group vfio_ap_mdev_attr_group = {
1587 .attrs = vfio_ap_mdev_attrs
1590 static const struct attribute_group *vfio_ap_mdev_attr_groups[] = {
1591 &vfio_ap_mdev_attr_group,
1596 * vfio_ap_mdev_set_kvm - sets all data for @matrix_mdev that are needed
1597 * to manage AP resources for the guest whose state is represented by @kvm
1599 * @matrix_mdev: a mediated matrix device
1600 * @kvm: reference to KVM instance
1602 * Return: 0 if no other mediated matrix device has a reference to @kvm;
1603 * otherwise, returns an -EPERM.
1605 static int vfio_ap_mdev_set_kvm(struct ap_matrix_mdev *matrix_mdev,
1608 struct ap_matrix_mdev *m;
1610 if (kvm->arch.crypto.crycbd) {
1611 down_write(&kvm->arch.crypto.pqap_hook_rwsem);
1612 kvm->arch.crypto.pqap_hook = &matrix_mdev->pqap_hook;
1613 up_write(&kvm->arch.crypto.pqap_hook_rwsem);
1615 get_update_locks_for_kvm(kvm);
1617 list_for_each_entry(m, &matrix_dev->mdev_list, node) {
1618 if (m != matrix_mdev && m->kvm == kvm) {
1619 release_update_locks_for_kvm(kvm);
1625 matrix_mdev->kvm = kvm;
1626 vfio_ap_mdev_update_guest_apcb(matrix_mdev);
1628 release_update_locks_for_kvm(kvm);
1634 static void unmap_iova(struct ap_matrix_mdev *matrix_mdev, u64 iova, u64 length)
1636 struct ap_queue_table *qtable = &matrix_mdev->qtable;
1637 struct vfio_ap_queue *q;
1640 hash_for_each(qtable->queues, loop_cursor, q, mdev_qnode) {
1641 if (q->saved_iova >= iova && q->saved_iova < iova + length)
1642 vfio_ap_irq_disable(q);
1646 static void vfio_ap_mdev_dma_unmap(struct vfio_device *vdev, u64 iova,
1649 struct ap_matrix_mdev *matrix_mdev =
1650 container_of(vdev, struct ap_matrix_mdev, vdev);
1652 mutex_lock(&matrix_dev->mdevs_lock);
1654 unmap_iova(matrix_mdev, iova, length);
1656 mutex_unlock(&matrix_dev->mdevs_lock);
1660 * vfio_ap_mdev_unset_kvm - performs clean-up of resources no longer needed
1663 * @matrix_mdev: a matrix mediated device
1665 static void vfio_ap_mdev_unset_kvm(struct ap_matrix_mdev *matrix_mdev)
1667 struct kvm *kvm = matrix_mdev->kvm;
1669 if (kvm && kvm->arch.crypto.crycbd) {
1670 down_write(&kvm->arch.crypto.pqap_hook_rwsem);
1671 kvm->arch.crypto.pqap_hook = NULL;
1672 up_write(&kvm->arch.crypto.pqap_hook_rwsem);
1674 get_update_locks_for_kvm(kvm);
1676 kvm_arch_crypto_clear_masks(kvm);
1677 vfio_ap_mdev_reset_queues(matrix_mdev);
1679 matrix_mdev->kvm = NULL;
1681 release_update_locks_for_kvm(kvm);
1685 static struct vfio_ap_queue *vfio_ap_find_queue(int apqn)
1687 struct ap_queue *queue;
1688 struct vfio_ap_queue *q = NULL;
1690 queue = ap_get_qdev(apqn);
1694 if (queue->ap_dev.device.driver == &matrix_dev->vfio_ap_drv->driver)
1695 q = dev_get_drvdata(&queue->ap_dev.device);
1697 put_device(&queue->ap_dev.device);
1702 static int apq_status_check(int apqn, struct ap_queue_status *status)
1704 switch (status->response_code) {
1705 case AP_RESPONSE_NORMAL:
1706 case AP_RESPONSE_DECONFIGURED:
1707 case AP_RESPONSE_CHECKSTOPPED:
1709 case AP_RESPONSE_RESET_IN_PROGRESS:
1710 case AP_RESPONSE_BUSY:
1712 case AP_RESPONSE_ASSOC_SECRET_NOT_UNIQUE:
1713 case AP_RESPONSE_ASSOC_FAILED:
1715 * These asynchronous response codes indicate a PQAP(AAPQ)
1716 * instruction to associate a secret with the guest failed. All
1717 * subsequent AP instructions will end with the asynchronous
1718 * response code until the AP queue is reset; so, let's return
1719 * a value indicating a reset needs to be performed again.
1724 "failed to verify reset of queue %02x.%04x: TAPQ rc=%u\n",
1725 AP_QID_CARD(apqn), AP_QID_QUEUE(apqn),
1726 status->response_code);
1731 #define WAIT_MSG "Waited %dms for reset of queue %02x.%04x (%u, %u, %u)"
1733 static void apq_reset_check(struct work_struct *reset_work)
1735 int ret = -EBUSY, elapsed = 0;
1736 struct ap_queue_status status;
1737 struct vfio_ap_queue *q;
1739 q = container_of(reset_work, struct vfio_ap_queue, reset_work);
1740 memcpy(&status, &q->reset_status, sizeof(status));
1742 msleep(AP_RESET_INTERVAL);
1743 elapsed += AP_RESET_INTERVAL;
1744 status = ap_tapq(q->apqn, NULL);
1745 ret = apq_status_check(q->apqn, &status);
1748 if (ret == -EBUSY) {
1749 pr_notice_ratelimited(WAIT_MSG, elapsed,
1750 AP_QID_CARD(q->apqn),
1751 AP_QID_QUEUE(q->apqn),
1752 status.response_code,
1754 status.irq_enabled);
1756 if (q->reset_status.response_code == AP_RESPONSE_RESET_IN_PROGRESS ||
1757 q->reset_status.response_code == AP_RESPONSE_BUSY ||
1758 q->reset_status.response_code == AP_RESPONSE_STATE_CHANGE_IN_PROGRESS ||
1760 status = ap_zapq(q->apqn, 0);
1761 memcpy(&q->reset_status, &status, sizeof(status));
1764 if (q->saved_isc != VFIO_AP_ISC_INVALID)
1765 vfio_ap_free_aqic_resources(q);
1771 static void vfio_ap_mdev_reset_queue(struct vfio_ap_queue *q)
1773 struct ap_queue_status status;
1777 status = ap_zapq(q->apqn, 0);
1778 memcpy(&q->reset_status, &status, sizeof(status));
1779 switch (status.response_code) {
1780 case AP_RESPONSE_NORMAL:
1781 case AP_RESPONSE_RESET_IN_PROGRESS:
1782 case AP_RESPONSE_BUSY:
1783 case AP_RESPONSE_STATE_CHANGE_IN_PROGRESS:
1785 * Let's verify whether the ZAPQ completed successfully on a work queue.
1787 queue_work(system_long_wq, &q->reset_work);
1789 case AP_RESPONSE_DECONFIGURED:
1790 case AP_RESPONSE_CHECKSTOPPED:
1791 vfio_ap_free_aqic_resources(q);
1795 "PQAP/ZAPQ for %02x.%04x failed with invalid rc=%u\n",
1796 AP_QID_CARD(q->apqn), AP_QID_QUEUE(q->apqn),
1797 status.response_code);
1801 static int vfio_ap_mdev_reset_queues(struct ap_matrix_mdev *matrix_mdev)
1803 int ret = 0, loop_cursor;
1804 struct vfio_ap_queue *q;
1806 hash_for_each(matrix_mdev->qtable.queues, loop_cursor, q, mdev_qnode)
1807 vfio_ap_mdev_reset_queue(q);
1809 hash_for_each(matrix_mdev->qtable.queues, loop_cursor, q, mdev_qnode) {
1810 flush_work(&q->reset_work);
1812 if (q->reset_status.response_code)
1819 static int vfio_ap_mdev_reset_qlist(struct list_head *qlist)
1822 struct vfio_ap_queue *q;
1824 list_for_each_entry(q, qlist, reset_qnode)
1825 vfio_ap_mdev_reset_queue(q);
1827 list_for_each_entry(q, qlist, reset_qnode) {
1828 flush_work(&q->reset_work);
1830 if (q->reset_status.response_code)
1837 static int vfio_ap_mdev_open_device(struct vfio_device *vdev)
1839 struct ap_matrix_mdev *matrix_mdev =
1840 container_of(vdev, struct ap_matrix_mdev, vdev);
1845 return vfio_ap_mdev_set_kvm(matrix_mdev, vdev->kvm);
1848 static void vfio_ap_mdev_close_device(struct vfio_device *vdev)
1850 struct ap_matrix_mdev *matrix_mdev =
1851 container_of(vdev, struct ap_matrix_mdev, vdev);
1853 vfio_ap_mdev_unset_kvm(matrix_mdev);
1856 static void vfio_ap_mdev_request(struct vfio_device *vdev, unsigned int count)
1858 struct device *dev = vdev->dev;
1859 struct ap_matrix_mdev *matrix_mdev;
1861 matrix_mdev = container_of(vdev, struct ap_matrix_mdev, vdev);
1863 if (matrix_mdev->req_trigger) {
1865 dev_notice_ratelimited(dev,
1866 "Relaying device request to user (#%u)\n",
1869 eventfd_signal(matrix_mdev->req_trigger);
1870 } else if (count == 0) {
1872 "No device request registered, blocked until released by user\n");
1876 static int vfio_ap_mdev_get_device_info(unsigned long arg)
1878 unsigned long minsz;
1879 struct vfio_device_info info;
1881 minsz = offsetofend(struct vfio_device_info, num_irqs);
1883 if (copy_from_user(&info, (void __user *)arg, minsz))
1886 if (info.argsz < minsz)
1889 info.flags = VFIO_DEVICE_FLAGS_AP | VFIO_DEVICE_FLAGS_RESET;
1890 info.num_regions = 0;
1891 info.num_irqs = VFIO_AP_NUM_IRQS;
1893 return copy_to_user((void __user *)arg, &info, minsz) ? -EFAULT : 0;
1896 static ssize_t vfio_ap_get_irq_info(unsigned long arg)
1898 unsigned long minsz;
1899 struct vfio_irq_info info;
1901 minsz = offsetofend(struct vfio_irq_info, count);
1903 if (copy_from_user(&info, (void __user *)arg, minsz))
1906 if (info.argsz < minsz || info.index >= VFIO_AP_NUM_IRQS)
1909 switch (info.index) {
1910 case VFIO_AP_REQ_IRQ_INDEX:
1912 info.flags = VFIO_IRQ_INFO_EVENTFD;
1918 return copy_to_user((void __user *)arg, &info, minsz) ? -EFAULT : 0;
1921 static int vfio_ap_irq_set_init(struct vfio_irq_set *irq_set, unsigned long arg)
1925 unsigned long minsz;
1927 minsz = offsetofend(struct vfio_irq_set, count);
1929 if (copy_from_user(irq_set, (void __user *)arg, minsz))
1932 ret = vfio_set_irqs_validate_and_prepare(irq_set, 1, VFIO_AP_NUM_IRQS,
1937 if (!(irq_set->flags & VFIO_IRQ_SET_ACTION_TRIGGER))
1943 static int vfio_ap_set_request_irq(struct ap_matrix_mdev *matrix_mdev,
1948 unsigned long minsz;
1949 struct eventfd_ctx *req_trigger;
1951 minsz = offsetofend(struct vfio_irq_set, count);
1952 data = (void __user *)(arg + minsz);
1954 if (get_user(fd, (s32 __user *)data))
1958 if (matrix_mdev->req_trigger)
1959 eventfd_ctx_put(matrix_mdev->req_trigger);
1960 matrix_mdev->req_trigger = NULL;
1961 } else if (fd >= 0) {
1962 req_trigger = eventfd_ctx_fdget(fd);
1963 if (IS_ERR(req_trigger))
1964 return PTR_ERR(req_trigger);
1966 if (matrix_mdev->req_trigger)
1967 eventfd_ctx_put(matrix_mdev->req_trigger);
1969 matrix_mdev->req_trigger = req_trigger;
1977 static int vfio_ap_set_irqs(struct ap_matrix_mdev *matrix_mdev,
1981 struct vfio_irq_set irq_set;
1983 ret = vfio_ap_irq_set_init(&irq_set, arg);
1987 switch (irq_set.flags & VFIO_IRQ_SET_DATA_TYPE_MASK) {
1988 case VFIO_IRQ_SET_DATA_EVENTFD:
1989 switch (irq_set.index) {
1990 case VFIO_AP_REQ_IRQ_INDEX:
1991 return vfio_ap_set_request_irq(matrix_mdev, arg);
2000 static ssize_t vfio_ap_mdev_ioctl(struct vfio_device *vdev,
2001 unsigned int cmd, unsigned long arg)
2003 struct ap_matrix_mdev *matrix_mdev =
2004 container_of(vdev, struct ap_matrix_mdev, vdev);
2007 mutex_lock(&matrix_dev->mdevs_lock);
2009 case VFIO_DEVICE_GET_INFO:
2010 ret = vfio_ap_mdev_get_device_info(arg);
2012 case VFIO_DEVICE_RESET:
2013 ret = vfio_ap_mdev_reset_queues(matrix_mdev);
2015 case VFIO_DEVICE_GET_IRQ_INFO:
2016 ret = vfio_ap_get_irq_info(arg);
2018 case VFIO_DEVICE_SET_IRQS:
2019 ret = vfio_ap_set_irqs(matrix_mdev, arg);
2025 mutex_unlock(&matrix_dev->mdevs_lock);
2030 static struct ap_matrix_mdev *vfio_ap_mdev_for_queue(struct vfio_ap_queue *q)
2032 struct ap_matrix_mdev *matrix_mdev;
2033 unsigned long apid = AP_QID_CARD(q->apqn);
2034 unsigned long apqi = AP_QID_QUEUE(q->apqn);
2036 list_for_each_entry(matrix_mdev, &matrix_dev->mdev_list, node) {
2037 if (test_bit_inv(apid, matrix_mdev->matrix.apm) &&
2038 test_bit_inv(apqi, matrix_mdev->matrix.aqm))
2045 static ssize_t status_show(struct device *dev,
2046 struct device_attribute *attr,
2050 struct vfio_ap_queue *q;
2051 unsigned long apid, apqi;
2052 struct ap_matrix_mdev *matrix_mdev;
2053 struct ap_device *apdev = to_ap_dev(dev);
2055 mutex_lock(&matrix_dev->mdevs_lock);
2056 q = dev_get_drvdata(&apdev->device);
2057 matrix_mdev = vfio_ap_mdev_for_queue(q);
2059 /* If the queue is assigned to the matrix mediated device, then
2060 * determine whether it is passed through to a guest; otherwise,
2061 * indicate that it is unassigned.
2064 apid = AP_QID_CARD(q->apqn);
2065 apqi = AP_QID_QUEUE(q->apqn);
2067 * If the queue is passed through to the guest, then indicate
2068 * that it is in use; otherwise, indicate that it is
2069 * merely assigned to a matrix mediated device.
2071 if (matrix_mdev->kvm &&
2072 test_bit_inv(apid, matrix_mdev->shadow_apcb.apm) &&
2073 test_bit_inv(apqi, matrix_mdev->shadow_apcb.aqm))
2074 nchars = scnprintf(buf, PAGE_SIZE, "%s\n",
2077 nchars = scnprintf(buf, PAGE_SIZE, "%s\n",
2080 nchars = scnprintf(buf, PAGE_SIZE, "%s\n",
2081 AP_QUEUE_UNASSIGNED);
2084 mutex_unlock(&matrix_dev->mdevs_lock);
2089 static DEVICE_ATTR_RO(status);
2091 static struct attribute *vfio_queue_attrs[] = {
2092 &dev_attr_status.attr,
2096 static const struct attribute_group vfio_queue_attr_group = {
2097 .attrs = vfio_queue_attrs,
2100 static const struct vfio_device_ops vfio_ap_matrix_dev_ops = {
2101 .init = vfio_ap_mdev_init_dev,
2102 .open_device = vfio_ap_mdev_open_device,
2103 .close_device = vfio_ap_mdev_close_device,
2104 .ioctl = vfio_ap_mdev_ioctl,
2105 .dma_unmap = vfio_ap_mdev_dma_unmap,
2106 .bind_iommufd = vfio_iommufd_emulated_bind,
2107 .unbind_iommufd = vfio_iommufd_emulated_unbind,
2108 .attach_ioas = vfio_iommufd_emulated_attach_ioas,
2109 .detach_ioas = vfio_iommufd_emulated_detach_ioas,
2110 .request = vfio_ap_mdev_request
2113 static struct mdev_driver vfio_ap_matrix_driver = {
2114 .device_api = VFIO_DEVICE_API_AP_STRING,
2115 .max_instances = MAX_ZDEV_ENTRIES_EXT,
2117 .name = "vfio_ap_mdev",
2118 .owner = THIS_MODULE,
2119 .mod_name = KBUILD_MODNAME,
2120 .dev_groups = vfio_ap_mdev_attr_groups,
2122 .probe = vfio_ap_mdev_probe,
2123 .remove = vfio_ap_mdev_remove,
2126 int vfio_ap_mdev_register(void)
2130 ret = mdev_register_driver(&vfio_ap_matrix_driver);
2134 matrix_dev->mdev_type.sysfs_name = VFIO_AP_MDEV_TYPE_HWVIRT;
2135 matrix_dev->mdev_type.pretty_name = VFIO_AP_MDEV_NAME_HWVIRT;
2136 matrix_dev->mdev_types[0] = &matrix_dev->mdev_type;
2137 ret = mdev_register_parent(&matrix_dev->parent, &matrix_dev->device,
2138 &vfio_ap_matrix_driver,
2139 matrix_dev->mdev_types, 1);
2145 mdev_unregister_driver(&vfio_ap_matrix_driver);
2149 void vfio_ap_mdev_unregister(void)
2151 mdev_unregister_parent(&matrix_dev->parent);
2152 mdev_unregister_driver(&vfio_ap_matrix_driver);
2155 int vfio_ap_mdev_probe_queue(struct ap_device *apdev)
2158 struct vfio_ap_queue *q;
2159 DECLARE_BITMAP(apm_filtered, AP_DEVICES);
2160 struct ap_matrix_mdev *matrix_mdev;
2162 ret = sysfs_create_group(&apdev->device.kobj, &vfio_queue_attr_group);
2166 q = kzalloc(sizeof(*q), GFP_KERNEL);
2169 goto err_remove_group;
2172 q->apqn = to_ap_queue(&apdev->device)->qid;
2173 q->saved_isc = VFIO_AP_ISC_INVALID;
2174 memset(&q->reset_status, 0, sizeof(q->reset_status));
2175 INIT_WORK(&q->reset_work, apq_reset_check);
2176 matrix_mdev = get_update_locks_by_apqn(q->apqn);
2179 vfio_ap_mdev_link_queue(matrix_mdev, q);
2182 * If we're in the process of handling the adding of adapters or
2183 * domains to the host's AP configuration, then let the
2184 * vfio_ap device driver's on_scan_complete callback filter the
2185 * matrix and update the guest's AP configuration after all of
2186 * the new queue devices are probed.
2188 if (!bitmap_empty(matrix_mdev->apm_add, AP_DEVICES) ||
2189 !bitmap_empty(matrix_mdev->aqm_add, AP_DOMAINS))
2192 if (vfio_ap_mdev_filter_matrix(matrix_mdev, apm_filtered)) {
2193 vfio_ap_mdev_update_guest_apcb(matrix_mdev);
2194 reset_queues_for_apids(matrix_mdev, apm_filtered);
2199 dev_set_drvdata(&apdev->device, q);
2200 release_update_locks_for_mdev(matrix_mdev);
2205 sysfs_remove_group(&apdev->device.kobj, &vfio_queue_attr_group);
2209 void vfio_ap_mdev_remove_queue(struct ap_device *apdev)
2211 unsigned long apid, apqi;
2212 struct vfio_ap_queue *q;
2213 struct ap_matrix_mdev *matrix_mdev;
2215 sysfs_remove_group(&apdev->device.kobj, &vfio_queue_attr_group);
2216 q = dev_get_drvdata(&apdev->device);
2217 get_update_locks_for_queue(q);
2218 matrix_mdev = q->matrix_mdev;
2219 apid = AP_QID_CARD(q->apqn);
2220 apqi = AP_QID_QUEUE(q->apqn);
2223 /* If the queue is assigned to the guest's AP configuration */
2224 if (test_bit_inv(apid, matrix_mdev->shadow_apcb.apm) &&
2225 test_bit_inv(apqi, matrix_mdev->shadow_apcb.aqm)) {
2227 * Since the queues are defined via a matrix of adapters
2228 * and domains, it is not possible to hot unplug a
2229 * single queue; so, let's unplug the adapter.
2231 clear_bit_inv(apid, matrix_mdev->shadow_apcb.apm);
2232 vfio_ap_mdev_update_guest_apcb(matrix_mdev);
2233 reset_queues_for_apid(matrix_mdev, apid);
2239 * If the queue is not in the host's AP configuration, then resetting
2240 * it will fail with response code 01, (APQN not valid); so, let's make
2241 * sure it is in the host's config.
2243 if (test_bit_inv(apid, (unsigned long *)matrix_dev->info.apm) &&
2244 test_bit_inv(apqi, (unsigned long *)matrix_dev->info.aqm)) {
2245 vfio_ap_mdev_reset_queue(q);
2246 flush_work(&q->reset_work);
2251 vfio_ap_unlink_queue_fr_mdev(q);
2253 dev_set_drvdata(&apdev->device, NULL);
2255 release_update_locks_for_mdev(matrix_mdev);
2259 * vfio_ap_mdev_resource_in_use: check whether any of a set of APQNs is
2260 * assigned to a mediated device under the control
2261 * of the vfio_ap device driver.
2263 * @apm: a bitmap specifying a set of APIDs comprising the APQNs to check.
2264 * @aqm: a bitmap specifying a set of APQIs comprising the APQNs to check.
2267 * * -EADDRINUSE if one or more of the APQNs specified via @apm/@aqm are
2268 * assigned to a mediated device under the control of the vfio_ap
2270 * * Otherwise, return 0.
2272 int vfio_ap_mdev_resource_in_use(unsigned long *apm, unsigned long *aqm)
2276 mutex_lock(&matrix_dev->guests_lock);
2277 mutex_lock(&matrix_dev->mdevs_lock);
2278 ret = vfio_ap_mdev_verify_no_sharing(apm, aqm);
2279 mutex_unlock(&matrix_dev->mdevs_lock);
2280 mutex_unlock(&matrix_dev->guests_lock);
2286 * vfio_ap_mdev_hot_unplug_cfg - hot unplug the adapters, domains and control
2287 * domains that have been removed from the host's
2288 * AP configuration from a guest.
2290 * @matrix_mdev: an ap_matrix_mdev object attached to a KVM guest.
2291 * @aprem: the adapters that have been removed from the host's AP configuration
2292 * @aqrem: the domains that have been removed from the host's AP configuration
2293 * @cdrem: the control domains that have been removed from the host's AP
2296 static void vfio_ap_mdev_hot_unplug_cfg(struct ap_matrix_mdev *matrix_mdev,
2297 unsigned long *aprem,
2298 unsigned long *aqrem,
2299 unsigned long *cdrem)
2303 if (!bitmap_empty(aprem, AP_DEVICES)) {
2304 do_hotplug |= bitmap_andnot(matrix_mdev->shadow_apcb.apm,
2305 matrix_mdev->shadow_apcb.apm,
2309 if (!bitmap_empty(aqrem, AP_DOMAINS)) {
2310 do_hotplug |= bitmap_andnot(matrix_mdev->shadow_apcb.aqm,
2311 matrix_mdev->shadow_apcb.aqm,
2315 if (!bitmap_empty(cdrem, AP_DOMAINS))
2316 do_hotplug |= bitmap_andnot(matrix_mdev->shadow_apcb.adm,
2317 matrix_mdev->shadow_apcb.adm,
2321 vfio_ap_mdev_update_guest_apcb(matrix_mdev);
2325 * vfio_ap_mdev_cfg_remove - determines which guests are using the adapters,
2326 * domains and control domains that have been removed
2327 * from the host AP configuration and unplugs them
2328 * from those guests.
2330 * @ap_remove: bitmap specifying which adapters have been removed from the host
2332 * @aq_remove: bitmap specifying which domains have been removed from the host
2334 * @cd_remove: bitmap specifying which control domains have been removed from
2337 static void vfio_ap_mdev_cfg_remove(unsigned long *ap_remove,
2338 unsigned long *aq_remove,
2339 unsigned long *cd_remove)
2341 struct ap_matrix_mdev *matrix_mdev;
2342 DECLARE_BITMAP(aprem, AP_DEVICES);
2343 DECLARE_BITMAP(aqrem, AP_DOMAINS);
2344 DECLARE_BITMAP(cdrem, AP_DOMAINS);
2347 list_for_each_entry(matrix_mdev, &matrix_dev->mdev_list, node) {
2348 mutex_lock(&matrix_mdev->kvm->lock);
2349 mutex_lock(&matrix_dev->mdevs_lock);
2351 do_remove |= bitmap_and(aprem, ap_remove,
2352 matrix_mdev->matrix.apm,
2354 do_remove |= bitmap_and(aqrem, aq_remove,
2355 matrix_mdev->matrix.aqm,
2357 do_remove |= bitmap_andnot(cdrem, cd_remove,
2358 matrix_mdev->matrix.adm,
2362 vfio_ap_mdev_hot_unplug_cfg(matrix_mdev, aprem, aqrem,
2365 mutex_unlock(&matrix_dev->mdevs_lock);
2366 mutex_unlock(&matrix_mdev->kvm->lock);
2371 * vfio_ap_mdev_on_cfg_remove - responds to the removal of adapters, domains and
2372 * control domains from the host AP configuration
2373 * by unplugging them from the guests that are
2375 * @cur_config_info: the current host AP configuration information
2376 * @prev_config_info: the previous host AP configuration information
2378 static void vfio_ap_mdev_on_cfg_remove(struct ap_config_info *cur_config_info,
2379 struct ap_config_info *prev_config_info)
2382 DECLARE_BITMAP(aprem, AP_DEVICES);
2383 DECLARE_BITMAP(aqrem, AP_DOMAINS);
2384 DECLARE_BITMAP(cdrem, AP_DOMAINS);
2386 do_remove = bitmap_andnot(aprem,
2387 (unsigned long *)prev_config_info->apm,
2388 (unsigned long *)cur_config_info->apm,
2390 do_remove |= bitmap_andnot(aqrem,
2391 (unsigned long *)prev_config_info->aqm,
2392 (unsigned long *)cur_config_info->aqm,
2394 do_remove |= bitmap_andnot(cdrem,
2395 (unsigned long *)prev_config_info->adm,
2396 (unsigned long *)cur_config_info->adm,
2400 vfio_ap_mdev_cfg_remove(aprem, aqrem, cdrem);
2404 * vfio_ap_filter_apid_by_qtype: filter APIDs from an AP mask for adapters that
2405 * are older than AP type 10 (CEX4).
2406 * @apm: a bitmap of the APIDs to examine
2407 * @aqm: a bitmap of the APQIs of the queues to query for the AP type.
2409 static void vfio_ap_filter_apid_by_qtype(unsigned long *apm, unsigned long *aqm)
2412 struct ap_queue_status status;
2413 unsigned long apid, apqi;
2414 struct ap_tapq_hwinfo info;
2416 for_each_set_bit_inv(apid, apm, AP_DEVICES) {
2417 apid_cleared = false;
2419 for_each_set_bit_inv(apqi, aqm, AP_DOMAINS) {
2420 status = ap_test_queue(AP_MKQID(apid, apqi), 1, &info);
2421 switch (status.response_code) {
2423 * According to the architecture in each case
2424 * below, the queue's info should be filled.
2426 case AP_RESPONSE_NORMAL:
2427 case AP_RESPONSE_RESET_IN_PROGRESS:
2428 case AP_RESPONSE_DECONFIGURED:
2429 case AP_RESPONSE_CHECKSTOPPED:
2430 case AP_RESPONSE_BUSY:
2432 * The vfio_ap device driver only
2433 * supports CEX4 and newer adapters, so
2434 * remove the APID if the adapter is
2435 * older than a CEX4.
2437 if (info.at < AP_DEVICE_TYPE_CEX4) {
2438 clear_bit_inv(apid, apm);
2439 apid_cleared = true;
2446 * If we don't know the adapter type,
2447 * clear its APID since it can't be
2448 * determined whether the vfio_ap
2449 * device driver supports it.
2451 clear_bit_inv(apid, apm);
2452 apid_cleared = true;
2457 * If we've already cleared the APID from the apm, there
2458 * is no need to continue examining the remainin AP
2459 * queues to determine the type of the adapter.
2468 * vfio_ap_mdev_cfg_add - store bitmaps specifying the adapters, domains and
2469 * control domains that have been added to the host's
2470 * AP configuration for each matrix mdev to which they
2473 * @apm_add: a bitmap specifying the adapters that have been added to the AP
2475 * @aqm_add: a bitmap specifying the domains that have been added to the AP
2477 * @adm_add: a bitmap specifying the control domains that have been added to the
2480 static void vfio_ap_mdev_cfg_add(unsigned long *apm_add, unsigned long *aqm_add,
2481 unsigned long *adm_add)
2483 struct ap_matrix_mdev *matrix_mdev;
2485 if (list_empty(&matrix_dev->mdev_list))
2488 vfio_ap_filter_apid_by_qtype(apm_add, aqm_add);
2490 list_for_each_entry(matrix_mdev, &matrix_dev->mdev_list, node) {
2491 bitmap_and(matrix_mdev->apm_add,
2492 matrix_mdev->matrix.apm, apm_add, AP_DEVICES);
2493 bitmap_and(matrix_mdev->aqm_add,
2494 matrix_mdev->matrix.aqm, aqm_add, AP_DOMAINS);
2495 bitmap_and(matrix_mdev->adm_add,
2496 matrix_mdev->matrix.adm, adm_add, AP_DEVICES);
2501 * vfio_ap_mdev_on_cfg_add - responds to the addition of adapters, domains and
2502 * control domains to the host AP configuration
2503 * by updating the bitmaps that specify what adapters,
2504 * domains and control domains have been added so they
2505 * can be hot plugged into the guest when the AP bus
2506 * scan completes (see vfio_ap_on_scan_complete
2508 * @cur_config_info: the current AP configuration information
2509 * @prev_config_info: the previous AP configuration information
2511 static void vfio_ap_mdev_on_cfg_add(struct ap_config_info *cur_config_info,
2512 struct ap_config_info *prev_config_info)
2515 DECLARE_BITMAP(apm_add, AP_DEVICES);
2516 DECLARE_BITMAP(aqm_add, AP_DOMAINS);
2517 DECLARE_BITMAP(adm_add, AP_DOMAINS);
2519 do_add = bitmap_andnot(apm_add,
2520 (unsigned long *)cur_config_info->apm,
2521 (unsigned long *)prev_config_info->apm,
2523 do_add |= bitmap_andnot(aqm_add,
2524 (unsigned long *)cur_config_info->aqm,
2525 (unsigned long *)prev_config_info->aqm,
2527 do_add |= bitmap_andnot(adm_add,
2528 (unsigned long *)cur_config_info->adm,
2529 (unsigned long *)prev_config_info->adm,
2533 vfio_ap_mdev_cfg_add(apm_add, aqm_add, adm_add);
2537 * vfio_ap_on_cfg_changed - handles notification of changes to the host AP
2540 * @cur_cfg_info: the current host AP configuration
2541 * @prev_cfg_info: the previous host AP configuration
2543 void vfio_ap_on_cfg_changed(struct ap_config_info *cur_cfg_info,
2544 struct ap_config_info *prev_cfg_info)
2546 if (!cur_cfg_info || !prev_cfg_info)
2549 mutex_lock(&matrix_dev->guests_lock);
2551 vfio_ap_mdev_on_cfg_remove(cur_cfg_info, prev_cfg_info);
2552 vfio_ap_mdev_on_cfg_add(cur_cfg_info, prev_cfg_info);
2553 memcpy(&matrix_dev->info, cur_cfg_info, sizeof(*cur_cfg_info));
2555 mutex_unlock(&matrix_dev->guests_lock);
2558 static void vfio_ap_mdev_hot_plug_cfg(struct ap_matrix_mdev *matrix_mdev)
2560 DECLARE_BITMAP(apm_filtered, AP_DEVICES);
2561 bool filter_domains, filter_adapters, filter_cdoms, do_hotplug = false;
2563 mutex_lock(&matrix_mdev->kvm->lock);
2564 mutex_lock(&matrix_dev->mdevs_lock);
2566 filter_adapters = bitmap_intersects(matrix_mdev->matrix.apm,
2567 matrix_mdev->apm_add, AP_DEVICES);
2568 filter_domains = bitmap_intersects(matrix_mdev->matrix.aqm,
2569 matrix_mdev->aqm_add, AP_DOMAINS);
2570 filter_cdoms = bitmap_intersects(matrix_mdev->matrix.adm,
2571 matrix_mdev->adm_add, AP_DOMAINS);
2573 if (filter_adapters || filter_domains)
2574 do_hotplug = vfio_ap_mdev_filter_matrix(matrix_mdev, apm_filtered);
2577 do_hotplug |= vfio_ap_mdev_filter_cdoms(matrix_mdev);
2580 vfio_ap_mdev_update_guest_apcb(matrix_mdev);
2582 reset_queues_for_apids(matrix_mdev, apm_filtered);
2584 mutex_unlock(&matrix_dev->mdevs_lock);
2585 mutex_unlock(&matrix_mdev->kvm->lock);
2588 void vfio_ap_on_scan_complete(struct ap_config_info *new_config_info,
2589 struct ap_config_info *old_config_info)
2591 struct ap_matrix_mdev *matrix_mdev;
2593 mutex_lock(&matrix_dev->guests_lock);
2595 list_for_each_entry(matrix_mdev, &matrix_dev->mdev_list, node) {
2596 if (bitmap_empty(matrix_mdev->apm_add, AP_DEVICES) &&
2597 bitmap_empty(matrix_mdev->aqm_add, AP_DOMAINS) &&
2598 bitmap_empty(matrix_mdev->adm_add, AP_DOMAINS))
2601 vfio_ap_mdev_hot_plug_cfg(matrix_mdev);
2602 bitmap_clear(matrix_mdev->apm_add, 0, AP_DEVICES);
2603 bitmap_clear(matrix_mdev->aqm_add, 0, AP_DOMAINS);
2604 bitmap_clear(matrix_mdev->adm_add, 0, AP_DOMAINS);
2607 mutex_unlock(&matrix_dev->guests_lock);