xref: /qemu/hw/intc/xics_kvm.c (revision 4ffb7496881ec361deaf1f51c41a933bde3cbf7b)
1 /*
2  * QEMU PowerPC pSeries Logical Partition (aka sPAPR) hardware System Emulator
3  *
4  * PAPR Virtualized Interrupt System, aka ICS/ICP aka xics, in-kernel emulation
5  *
6  * Copyright (c) 2013 David Gibson, IBM Corporation.
7  *
8  * Permission is hereby granted, free of charge, to any person obtaining a copy
9  * of this software and associated documentation files (the "Software"), to deal
10  * in the Software without restriction, including without limitation the rights
11  * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
12  * copies of the Software, and to permit persons to whom the Software is
13  * furnished to do so, subject to the following conditions:
14  *
15  * The above copyright notice and this permission notice shall be included in
16  * all copies or substantial portions of the Software.
17  *
18  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
19  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
20  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
21  * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
22  * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
23  * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
24  * THE SOFTWARE.
25  *
26  */
27 
28 #include "qemu/osdep.h"
29 #include "qapi/error.h"
30 #include "qemu-common.h"
31 #include "cpu.h"
32 #include "trace.h"
33 #include "sysemu/kvm.h"
34 #include "hw/ppc/spapr.h"
35 #include "hw/ppc/spapr_cpu_core.h"
36 #include "hw/ppc/xics.h"
37 #include "hw/ppc/xics_spapr.h"
38 #include "kvm_ppc.h"
39 #include "qemu/config-file.h"
40 #include "qemu/error-report.h"
41 
42 #include <sys/ioctl.h>
43 
44 static int kernel_xics_fd = -1;
45 
46 typedef struct KVMEnabledICP {
47     unsigned long vcpu_id;
48     QLIST_ENTRY(KVMEnabledICP) node;
49 } KVMEnabledICP;
50 
51 static QLIST_HEAD(, KVMEnabledICP)
52     kvm_enabled_icps = QLIST_HEAD_INITIALIZER(&kvm_enabled_icps);
53 
54 static void kvm_disable_icps(void)
55 {
56     KVMEnabledICP *enabled_icp, *next;
57 
58     QLIST_FOREACH_SAFE(enabled_icp, &kvm_enabled_icps, node, next) {
59         QLIST_REMOVE(enabled_icp, node);
60         g_free(enabled_icp);
61     }
62 }
63 
64 /*
65  * ICP-KVM
66  */
67 void icp_get_kvm_state(ICPState *icp)
68 {
69     uint64_t state;
70     int ret;
71 
72     /* The KVM XICS device is not in use */
73     if (kernel_xics_fd == -1) {
74         return;
75     }
76 
77     /* ICP for this CPU thread is not in use, exiting */
78     if (!icp->cs) {
79         return;
80     }
81 
82     ret = kvm_get_one_reg(icp->cs, KVM_REG_PPC_ICP_STATE, &state);
83     if (ret != 0) {
84         error_report("Unable to retrieve KVM interrupt controller state"
85                 " for CPU %ld: %s", kvm_arch_vcpu_id(icp->cs), strerror(errno));
86         exit(1);
87     }
88 
89     icp->xirr = state >> KVM_REG_PPC_ICP_XISR_SHIFT;
90     icp->mfrr = (state >> KVM_REG_PPC_ICP_MFRR_SHIFT)
91         & KVM_REG_PPC_ICP_MFRR_MASK;
92     icp->pending_priority = (state >> KVM_REG_PPC_ICP_PPRI_SHIFT)
93         & KVM_REG_PPC_ICP_PPRI_MASK;
94 }
95 
96 static void do_icp_synchronize_state(CPUState *cpu, run_on_cpu_data arg)
97 {
98     icp_get_kvm_state(arg.host_ptr);
99 }
100 
101 void icp_synchronize_state(ICPState *icp)
102 {
103     if (icp->cs) {
104         run_on_cpu(icp->cs, do_icp_synchronize_state, RUN_ON_CPU_HOST_PTR(icp));
105     }
106 }
107 
108 int icp_set_kvm_state(ICPState *icp, Error **errp)
109 {
110     uint64_t state;
111     int ret;
112 
113     /* The KVM XICS device is not in use */
114     if (kernel_xics_fd == -1) {
115         return 0;
116     }
117 
118     /* ICP for this CPU thread is not in use, exiting */
119     if (!icp->cs) {
120         return 0;
121     }
122 
123     state = ((uint64_t)icp->xirr << KVM_REG_PPC_ICP_XISR_SHIFT)
124         | ((uint64_t)icp->mfrr << KVM_REG_PPC_ICP_MFRR_SHIFT)
125         | ((uint64_t)icp->pending_priority << KVM_REG_PPC_ICP_PPRI_SHIFT);
126 
127     ret = kvm_set_one_reg(icp->cs, KVM_REG_PPC_ICP_STATE, &state);
128     if (ret < 0) {
129         error_setg_errno(errp, -ret,
130                          "Unable to restore KVM interrupt controller state (0x%"
131                          PRIx64 ") for CPU %ld", state,
132                          kvm_arch_vcpu_id(icp->cs));
133         return ret;
134     }
135 
136     return 0;
137 }
138 
139 void icp_kvm_realize(DeviceState *dev, Error **errp)
140 {
141     ICPState *icp = ICP(dev);
142     CPUState *cs;
143     KVMEnabledICP *enabled_icp;
144     unsigned long vcpu_id;
145     int ret;
146 
147     /* The KVM XICS device is not in use */
148     if (kernel_xics_fd == -1) {
149         return;
150     }
151 
152     cs = icp->cs;
153     vcpu_id = kvm_arch_vcpu_id(cs);
154 
155     /*
156      * If we are reusing a parked vCPU fd corresponding to the CPU
157      * which was hot-removed earlier we don't have to renable
158      * KVM_CAP_IRQ_XICS capability again.
159      */
160     QLIST_FOREACH(enabled_icp, &kvm_enabled_icps, node) {
161         if (enabled_icp->vcpu_id == vcpu_id) {
162             return;
163         }
164     }
165 
166     ret = kvm_vcpu_enable_cap(cs, KVM_CAP_IRQ_XICS, 0, kernel_xics_fd, vcpu_id);
167     if (ret < 0) {
168         error_setg(errp, "Unable to connect CPU%ld to kernel XICS: %s", vcpu_id,
169                    strerror(errno));
170         return;
171     }
172     enabled_icp = g_malloc(sizeof(*enabled_icp));
173     enabled_icp->vcpu_id = vcpu_id;
174     QLIST_INSERT_HEAD(&kvm_enabled_icps, enabled_icp, node);
175 }
176 
177 /*
178  * ICS-KVM
179  */
180 void ics_get_kvm_state(ICSState *ics)
181 {
182     uint64_t state;
183     int i;
184 
185     /* The KVM XICS device is not in use */
186     if (kernel_xics_fd == -1) {
187         return;
188     }
189 
190     for (i = 0; i < ics->nr_irqs; i++) {
191         ICSIRQState *irq = &ics->irqs[i];
192 
193         if (ics_irq_free(ics, i)) {
194             continue;
195         }
196 
197         kvm_device_access(kernel_xics_fd, KVM_DEV_XICS_GRP_SOURCES,
198                           i + ics->offset, &state, false, &error_fatal);
199 
200         irq->server = state & KVM_XICS_DESTINATION_MASK;
201         irq->saved_priority = (state >> KVM_XICS_PRIORITY_SHIFT)
202             & KVM_XICS_PRIORITY_MASK;
203         /*
204          * To be consistent with the software emulation in xics.c, we
205          * split out the masked state + priority that we get from the
206          * kernel into 'current priority' (0xff if masked) and
207          * 'saved priority' (if masked, this is the priority the
208          * interrupt had before it was masked).  Masking and unmasking
209          * are done with the ibm,int-off and ibm,int-on RTAS calls.
210          */
211         if (state & KVM_XICS_MASKED) {
212             irq->priority = 0xff;
213         } else {
214             irq->priority = irq->saved_priority;
215         }
216 
217         irq->status = 0;
218         if (state & KVM_XICS_PENDING) {
219             if (state & KVM_XICS_LEVEL_SENSITIVE) {
220                 irq->status |= XICS_STATUS_ASSERTED;
221             } else {
222                 /*
223                  * A pending edge-triggered interrupt (or MSI)
224                  * must have been rejected previously when we
225                  * first detected it and tried to deliver it,
226                  * so mark it as pending and previously rejected
227                  * for consistency with how xics.c works.
228                  */
229                 irq->status |= XICS_STATUS_MASKED_PENDING
230                     | XICS_STATUS_REJECTED;
231             }
232         }
233         if (state & KVM_XICS_PRESENTED) {
234                 irq->status |= XICS_STATUS_PRESENTED;
235         }
236         if (state & KVM_XICS_QUEUED) {
237                 irq->status |= XICS_STATUS_QUEUED;
238         }
239     }
240 }
241 
242 void ics_synchronize_state(ICSState *ics)
243 {
244     ics_get_kvm_state(ics);
245 }
246 
247 int ics_set_kvm_state_one(ICSState *ics, int srcno, Error **errp)
248 {
249     uint64_t state;
250     ICSIRQState *irq = &ics->irqs[srcno];
251     int ret;
252 
253     /* The KVM XICS device is not in use */
254     if (kernel_xics_fd == -1) {
255         return 0;
256     }
257 
258     state = irq->server;
259     state |= (uint64_t)(irq->saved_priority & KVM_XICS_PRIORITY_MASK)
260         << KVM_XICS_PRIORITY_SHIFT;
261     if (irq->priority != irq->saved_priority) {
262         assert(irq->priority == 0xff);
263     }
264 
265     if (irq->priority == 0xff) {
266         state |= KVM_XICS_MASKED;
267     }
268 
269     if (irq->flags & XICS_FLAGS_IRQ_LSI) {
270         state |= KVM_XICS_LEVEL_SENSITIVE;
271         if (irq->status & XICS_STATUS_ASSERTED) {
272             state |= KVM_XICS_PENDING;
273         }
274     } else {
275         if (irq->status & XICS_STATUS_MASKED_PENDING) {
276             state |= KVM_XICS_PENDING;
277         }
278     }
279     if (irq->status & XICS_STATUS_PRESENTED) {
280         state |= KVM_XICS_PRESENTED;
281     }
282     if (irq->status & XICS_STATUS_QUEUED) {
283         state |= KVM_XICS_QUEUED;
284     }
285 
286     ret = kvm_device_access(kernel_xics_fd, KVM_DEV_XICS_GRP_SOURCES,
287                             srcno + ics->offset, &state, true, errp);
288     if (ret < 0) {
289         return ret;
290     }
291 
292     return 0;
293 }
294 
295 int ics_set_kvm_state(ICSState *ics, Error **errp)
296 {
297     int i;
298 
299     /* The KVM XICS device is not in use */
300     if (kernel_xics_fd == -1) {
301         return 0;
302     }
303 
304     for (i = 0; i < ics->nr_irqs; i++) {
305         Error *local_err = NULL;
306         int ret;
307 
308         if (ics_irq_free(ics, i)) {
309             continue;
310         }
311 
312         ret = ics_set_kvm_state_one(ics, i, &local_err);
313         if (ret < 0) {
314             error_propagate(errp, local_err);
315             return ret;
316         }
317     }
318 
319     return 0;
320 }
321 
322 void ics_kvm_set_irq(ICSState *ics, int srcno, int val)
323 {
324     struct kvm_irq_level args;
325     int rc;
326 
327     /* The KVM XICS device should be in use */
328     assert(kernel_xics_fd != -1);
329 
330     args.irq = srcno + ics->offset;
331     if (ics->irqs[srcno].flags & XICS_FLAGS_IRQ_MSI) {
332         if (!val) {
333             return;
334         }
335         args.level = KVM_INTERRUPT_SET;
336     } else {
337         args.level = val ? KVM_INTERRUPT_SET_LEVEL : KVM_INTERRUPT_UNSET;
338     }
339     rc = kvm_vm_ioctl(kvm_state, KVM_IRQ_LINE, &args);
340     if (rc < 0) {
341         perror("kvm_irq_line");
342     }
343 }
344 
345 int xics_kvm_connect(SpaprInterruptController *intc, uint32_t nr_servers,
346                      Error **errp)
347 {
348     ICSState *ics = ICS_SPAPR(intc);
349     int rc;
350     CPUState *cs;
351     Error *local_err = NULL;
352 
353     /*
354      * The KVM XICS device already in use. This is the case when
355      * rebooting under the XICS-only interrupt mode.
356      */
357     if (kernel_xics_fd != -1) {
358         return 0;
359     }
360 
361     if (!kvm_enabled() || !kvm_check_extension(kvm_state, KVM_CAP_IRQ_XICS)) {
362         error_setg(errp,
363                    "KVM and IRQ_XICS capability must be present for in-kernel XICS");
364         return -1;
365     }
366 
367     rc = kvmppc_define_rtas_kernel_token(RTAS_IBM_SET_XIVE, "ibm,set-xive");
368     if (rc < 0) {
369         error_setg_errno(&local_err, -rc,
370                          "kvmppc_define_rtas_kernel_token: ibm,set-xive");
371         goto fail;
372     }
373 
374     rc = kvmppc_define_rtas_kernel_token(RTAS_IBM_GET_XIVE, "ibm,get-xive");
375     if (rc < 0) {
376         error_setg_errno(&local_err, -rc,
377                          "kvmppc_define_rtas_kernel_token: ibm,get-xive");
378         goto fail;
379     }
380 
381     rc = kvmppc_define_rtas_kernel_token(RTAS_IBM_INT_ON, "ibm,int-on");
382     if (rc < 0) {
383         error_setg_errno(&local_err, -rc,
384                          "kvmppc_define_rtas_kernel_token: ibm,int-on");
385         goto fail;
386     }
387 
388     rc = kvmppc_define_rtas_kernel_token(RTAS_IBM_INT_OFF, "ibm,int-off");
389     if (rc < 0) {
390         error_setg_errno(&local_err, -rc,
391                          "kvmppc_define_rtas_kernel_token: ibm,int-off");
392         goto fail;
393     }
394 
395     /* Create the KVM XICS device */
396     rc = kvm_create_device(kvm_state, KVM_DEV_TYPE_XICS, false);
397     if (rc < 0) {
398         error_setg_errno(&local_err, -rc, "Error on KVM_CREATE_DEVICE for XICS");
399         goto fail;
400     }
401 
402     kernel_xics_fd = rc;
403     kvm_kernel_irqchip = true;
404     kvm_msi_via_irqfd_allowed = true;
405     kvm_gsi_direct_mapping = true;
406 
407     /* Create the presenters */
408     CPU_FOREACH(cs) {
409         PowerPCCPU *cpu = POWERPC_CPU(cs);
410 
411         icp_kvm_realize(DEVICE(spapr_cpu_state(cpu)->icp), &local_err);
412         if (local_err) {
413             goto fail;
414         }
415     }
416 
417     /* Update the KVM sources */
418     ics_set_kvm_state(ics, &local_err);
419     if (local_err) {
420         goto fail;
421     }
422 
423     /* Connect the presenters to the initial VCPUs of the machine */
424     CPU_FOREACH(cs) {
425         PowerPCCPU *cpu = POWERPC_CPU(cs);
426         icp_set_kvm_state(spapr_cpu_state(cpu)->icp, &local_err);
427         if (local_err) {
428             goto fail;
429         }
430     }
431 
432     return 0;
433 
434 fail:
435     error_propagate(errp, local_err);
436     xics_kvm_disconnect(intc);
437     return -1;
438 }
439 
440 void xics_kvm_disconnect(SpaprInterruptController *intc)
441 {
442     /*
443      * Only on P9 using the XICS-on XIVE KVM device:
444      *
445      * When the KVM device fd is closed, the device is destroyed and
446      * removed from the list of devices of the VM. The VCPU presenters
447      * are also detached from the device.
448      */
449     if (kernel_xics_fd != -1) {
450         close(kernel_xics_fd);
451         kernel_xics_fd = -1;
452     }
453 
454     kvmppc_define_rtas_kernel_token(0, "ibm,set-xive");
455     kvmppc_define_rtas_kernel_token(0, "ibm,get-xive");
456     kvmppc_define_rtas_kernel_token(0, "ibm,int-on");
457     kvmppc_define_rtas_kernel_token(0, "ibm,int-off");
458 
459     kvm_kernel_irqchip = false;
460     kvm_msi_via_irqfd_allowed = false;
461     kvm_gsi_direct_mapping = false;
462 
463     /* Clear the presenter from the VCPUs */
464     kvm_disable_icps();
465 }
466 
467 /*
468  * This is a heuristic to detect older KVMs on POWER9 hosts that don't
469  * support destruction of a KVM XICS device while the VM is running.
470  * Required to start a spapr machine with ic-mode=dual,kernel-irqchip=on.
471  */
472 bool xics_kvm_has_broken_disconnect(SpaprMachineState *spapr)
473 {
474     int rc;
475 
476     rc = kvm_create_device(kvm_state, KVM_DEV_TYPE_XICS, false);
477     if (rc < 0) {
478         /*
479          * The error is ignored on purpose. The KVM XICS setup code
480          * will catch it again anyway. The goal here is to see if
481          * close() actually destroys the device or not.
482          */
483         return false;
484     }
485 
486     close(rc);
487 
488     rc = kvm_create_device(kvm_state, KVM_DEV_TYPE_XICS, false);
489     if (rc >= 0) {
490         close(rc);
491         return false;
492     }
493 
494     return errno == EEXIST;
495 }
496