xref: /openbmc/qemu/hw/intc/xics_kvm.c (revision 5cc8767d)
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 "hw/hw.h"
33 #include "trace.h"
34 #include "sysemu/kvm.h"
35 #include "hw/ppc/spapr.h"
36 #include "hw/ppc/spapr_cpu_core.h"
37 #include "hw/ppc/xics.h"
38 #include "hw/ppc/xics_spapr.h"
39 #include "kvm_ppc.h"
40 #include "qemu/config-file.h"
41 #include "qemu/error-report.h"
42 
43 #include <sys/ioctl.h>
44 
45 static int kernel_xics_fd = -1;
46 
47 typedef struct KVMEnabledICP {
48     unsigned long vcpu_id;
49     QLIST_ENTRY(KVMEnabledICP) node;
50 } KVMEnabledICP;
51 
52 static QLIST_HEAD(, KVMEnabledICP)
53     kvm_enabled_icps = QLIST_HEAD_INITIALIZER(&kvm_enabled_icps);
54 
55 static void kvm_disable_icps(void)
56 {
57     KVMEnabledICP *enabled_icp, *next;
58 
59     QLIST_FOREACH_SAFE(enabled_icp, &kvm_enabled_icps, node, next) {
60         QLIST_REMOVE(enabled_icp, node);
61         g_free(enabled_icp);
62     }
63 }
64 
65 /*
66  * ICP-KVM
67  */
68 void icp_get_kvm_state(ICPState *icp)
69 {
70     uint64_t state;
71     int ret;
72 
73     /* The KVM XICS device is not in use */
74     if (kernel_xics_fd == -1) {
75         return;
76     }
77 
78     /* ICP for this CPU thread is not in use, exiting */
79     if (!icp->cs) {
80         return;
81     }
82 
83     ret = kvm_get_one_reg(icp->cs, KVM_REG_PPC_ICP_STATE, &state);
84     if (ret != 0) {
85         error_report("Unable to retrieve KVM interrupt controller state"
86                 " for CPU %ld: %s", kvm_arch_vcpu_id(icp->cs), strerror(errno));
87         exit(1);
88     }
89 
90     icp->xirr = state >> KVM_REG_PPC_ICP_XISR_SHIFT;
91     icp->mfrr = (state >> KVM_REG_PPC_ICP_MFRR_SHIFT)
92         & KVM_REG_PPC_ICP_MFRR_MASK;
93     icp->pending_priority = (state >> KVM_REG_PPC_ICP_PPRI_SHIFT)
94         & KVM_REG_PPC_ICP_PPRI_MASK;
95 }
96 
97 static void do_icp_synchronize_state(CPUState *cpu, run_on_cpu_data arg)
98 {
99     icp_get_kvm_state(arg.host_ptr);
100 }
101 
102 void icp_synchronize_state(ICPState *icp)
103 {
104     if (icp->cs) {
105         run_on_cpu(icp->cs, do_icp_synchronize_state, RUN_ON_CPU_HOST_PTR(icp));
106     }
107 }
108 
109 int icp_set_kvm_state(ICPState *icp, Error **errp)
110 {
111     uint64_t state;
112     int ret;
113 
114     /* The KVM XICS device is not in use */
115     if (kernel_xics_fd == -1) {
116         return 0;
117     }
118 
119     /* ICP for this CPU thread is not in use, exiting */
120     if (!icp->cs) {
121         return 0;
122     }
123 
124     state = ((uint64_t)icp->xirr << KVM_REG_PPC_ICP_XISR_SHIFT)
125         | ((uint64_t)icp->mfrr << KVM_REG_PPC_ICP_MFRR_SHIFT)
126         | ((uint64_t)icp->pending_priority << KVM_REG_PPC_ICP_PPRI_SHIFT);
127 
128     ret = kvm_set_one_reg(icp->cs, KVM_REG_PPC_ICP_STATE, &state);
129     if (ret < 0) {
130         error_setg_errno(errp, -ret,
131                          "Unable to restore KVM interrupt controller state (0x%"
132                          PRIx64 ") for CPU %ld", state,
133                          kvm_arch_vcpu_id(icp->cs));
134         return ret;
135     }
136 
137     return 0;
138 }
139 
140 void icp_kvm_realize(DeviceState *dev, Error **errp)
141 {
142     ICPState *icp = ICP(dev);
143     CPUState *cs;
144     KVMEnabledICP *enabled_icp;
145     unsigned long vcpu_id;
146     int ret;
147 
148     /* The KVM XICS device is not in use */
149     if (kernel_xics_fd == -1) {
150         return;
151     }
152 
153     cs = icp->cs;
154     vcpu_id = kvm_arch_vcpu_id(cs);
155 
156     /*
157      * If we are reusing a parked vCPU fd corresponding to the CPU
158      * which was hot-removed earlier we don't have to renable
159      * KVM_CAP_IRQ_XICS capability again.
160      */
161     QLIST_FOREACH(enabled_icp, &kvm_enabled_icps, node) {
162         if (enabled_icp->vcpu_id == vcpu_id) {
163             return;
164         }
165     }
166 
167     ret = kvm_vcpu_enable_cap(cs, KVM_CAP_IRQ_XICS, 0, kernel_xics_fd, vcpu_id);
168     if (ret < 0) {
169         error_setg(errp, "Unable to connect CPU%ld to kernel XICS: %s", vcpu_id,
170                    strerror(errno));
171         return;
172     }
173     enabled_icp = g_malloc(sizeof(*enabled_icp));
174     enabled_icp->vcpu_id = vcpu_id;
175     QLIST_INSERT_HEAD(&kvm_enabled_icps, enabled_icp, node);
176 }
177 
178 /*
179  * ICS-KVM
180  */
181 void ics_get_kvm_state(ICSState *ics)
182 {
183     uint64_t state;
184     int i;
185 
186     /* The KVM XICS device is not in use */
187     if (kernel_xics_fd == -1) {
188         return;
189     }
190 
191     for (i = 0; i < ics->nr_irqs; i++) {
192         ICSIRQState *irq = &ics->irqs[i];
193 
194         kvm_device_access(kernel_xics_fd, KVM_DEV_XICS_GRP_SOURCES,
195                           i + ics->offset, &state, false, &error_fatal);
196 
197         irq->server = state & KVM_XICS_DESTINATION_MASK;
198         irq->saved_priority = (state >> KVM_XICS_PRIORITY_SHIFT)
199             & KVM_XICS_PRIORITY_MASK;
200         /*
201          * To be consistent with the software emulation in xics.c, we
202          * split out the masked state + priority that we get from the
203          * kernel into 'current priority' (0xff if masked) and
204          * 'saved priority' (if masked, this is the priority the
205          * interrupt had before it was masked).  Masking and unmasking
206          * are done with the ibm,int-off and ibm,int-on RTAS calls.
207          */
208         if (state & KVM_XICS_MASKED) {
209             irq->priority = 0xff;
210         } else {
211             irq->priority = irq->saved_priority;
212         }
213 
214         irq->status = 0;
215         if (state & KVM_XICS_PENDING) {
216             if (state & KVM_XICS_LEVEL_SENSITIVE) {
217                 irq->status |= XICS_STATUS_ASSERTED;
218             } else {
219                 /*
220                  * A pending edge-triggered interrupt (or MSI)
221                  * must have been rejected previously when we
222                  * first detected it and tried to deliver it,
223                  * so mark it as pending and previously rejected
224                  * for consistency with how xics.c works.
225                  */
226                 irq->status |= XICS_STATUS_MASKED_PENDING
227                     | XICS_STATUS_REJECTED;
228             }
229         }
230         if (state & KVM_XICS_PRESENTED) {
231                 irq->status |= XICS_STATUS_PRESENTED;
232         }
233         if (state & KVM_XICS_QUEUED) {
234                 irq->status |= XICS_STATUS_QUEUED;
235         }
236     }
237 }
238 
239 void ics_synchronize_state(ICSState *ics)
240 {
241     ics_get_kvm_state(ics);
242 }
243 
244 int ics_set_kvm_state_one(ICSState *ics, int srcno, Error **errp)
245 {
246     uint64_t state;
247     ICSIRQState *irq = &ics->irqs[srcno];
248     int ret;
249 
250     /* The KVM XICS device is not in use */
251     if (kernel_xics_fd == -1) {
252         return 0;
253     }
254 
255     state = irq->server;
256     state |= (uint64_t)(irq->saved_priority & KVM_XICS_PRIORITY_MASK)
257         << KVM_XICS_PRIORITY_SHIFT;
258     if (irq->priority != irq->saved_priority) {
259         assert(irq->priority == 0xff);
260         state |= KVM_XICS_MASKED;
261     }
262 
263     if (irq->flags & XICS_FLAGS_IRQ_LSI) {
264         state |= KVM_XICS_LEVEL_SENSITIVE;
265         if (irq->status & XICS_STATUS_ASSERTED) {
266             state |= KVM_XICS_PENDING;
267         }
268     } else {
269         if (irq->status & XICS_STATUS_MASKED_PENDING) {
270             state |= KVM_XICS_PENDING;
271         }
272     }
273     if (irq->status & XICS_STATUS_PRESENTED) {
274         state |= KVM_XICS_PRESENTED;
275     }
276     if (irq->status & XICS_STATUS_QUEUED) {
277         state |= KVM_XICS_QUEUED;
278     }
279 
280     ret = kvm_device_access(kernel_xics_fd, KVM_DEV_XICS_GRP_SOURCES,
281                             srcno + ics->offset, &state, true, errp);
282     if (ret < 0) {
283         return ret;
284     }
285 
286     return 0;
287 }
288 
289 int ics_set_kvm_state(ICSState *ics, Error **errp)
290 {
291     int i;
292 
293     /* The KVM XICS device is not in use */
294     if (kernel_xics_fd == -1) {
295         return 0;
296     }
297 
298     for (i = 0; i < ics->nr_irqs; i++) {
299         Error *local_err = NULL;
300         int ret;
301 
302         ret = ics_set_kvm_state_one(ics, i, &local_err);
303         if (ret < 0) {
304             error_propagate(errp, local_err);
305             return ret;
306         }
307     }
308 
309     return 0;
310 }
311 
312 void ics_kvm_set_irq(ICSState *ics, int srcno, int val)
313 {
314     struct kvm_irq_level args;
315     int rc;
316 
317     /* The KVM XICS device should be in use */
318     assert(kernel_xics_fd != -1);
319 
320     args.irq = srcno + ics->offset;
321     if (ics->irqs[srcno].flags & XICS_FLAGS_IRQ_MSI) {
322         if (!val) {
323             return;
324         }
325         args.level = KVM_INTERRUPT_SET;
326     } else {
327         args.level = val ? KVM_INTERRUPT_SET_LEVEL : KVM_INTERRUPT_UNSET;
328     }
329     rc = kvm_vm_ioctl(kvm_state, KVM_IRQ_LINE, &args);
330     if (rc < 0) {
331         perror("kvm_irq_line");
332     }
333 }
334 
335 int xics_kvm_connect(SpaprMachineState *spapr, Error **errp)
336 {
337     int rc;
338     CPUState *cs;
339     Error *local_err = NULL;
340 
341     /*
342      * The KVM XICS device already in use. This is the case when
343      * rebooting under the XICS-only interrupt mode.
344      */
345     if (kernel_xics_fd != -1) {
346         return 0;
347     }
348 
349     if (!kvm_enabled() || !kvm_check_extension(kvm_state, KVM_CAP_IRQ_XICS)) {
350         error_setg(errp,
351                    "KVM and IRQ_XICS capability must be present for in-kernel XICS");
352         return -1;
353     }
354 
355     rc = kvmppc_define_rtas_kernel_token(RTAS_IBM_SET_XIVE, "ibm,set-xive");
356     if (rc < 0) {
357         error_setg_errno(&local_err, -rc,
358                          "kvmppc_define_rtas_kernel_token: ibm,set-xive");
359         goto fail;
360     }
361 
362     rc = kvmppc_define_rtas_kernel_token(RTAS_IBM_GET_XIVE, "ibm,get-xive");
363     if (rc < 0) {
364         error_setg_errno(&local_err, -rc,
365                          "kvmppc_define_rtas_kernel_token: ibm,get-xive");
366         goto fail;
367     }
368 
369     rc = kvmppc_define_rtas_kernel_token(RTAS_IBM_INT_ON, "ibm,int-on");
370     if (rc < 0) {
371         error_setg_errno(&local_err, -rc,
372                          "kvmppc_define_rtas_kernel_token: ibm,int-on");
373         goto fail;
374     }
375 
376     rc = kvmppc_define_rtas_kernel_token(RTAS_IBM_INT_OFF, "ibm,int-off");
377     if (rc < 0) {
378         error_setg_errno(&local_err, -rc,
379                          "kvmppc_define_rtas_kernel_token: ibm,int-off");
380         goto fail;
381     }
382 
383     /* Create the KVM XICS device */
384     rc = kvm_create_device(kvm_state, KVM_DEV_TYPE_XICS, false);
385     if (rc < 0) {
386         error_setg_errno(&local_err, -rc, "Error on KVM_CREATE_DEVICE for XICS");
387         goto fail;
388     }
389 
390     kernel_xics_fd = rc;
391     kvm_kernel_irqchip = true;
392     kvm_msi_via_irqfd_allowed = true;
393     kvm_gsi_direct_mapping = true;
394 
395     /* Create the presenters */
396     CPU_FOREACH(cs) {
397         PowerPCCPU *cpu = POWERPC_CPU(cs);
398 
399         icp_kvm_realize(DEVICE(spapr_cpu_state(cpu)->icp), &local_err);
400         if (local_err) {
401             goto fail;
402         }
403     }
404 
405     /* Update the KVM sources */
406     ics_set_kvm_state(spapr->ics, &local_err);
407     if (local_err) {
408         goto fail;
409     }
410 
411     /* Connect the presenters to the initial VCPUs of the machine */
412     CPU_FOREACH(cs) {
413         PowerPCCPU *cpu = POWERPC_CPU(cs);
414         icp_set_kvm_state(spapr_cpu_state(cpu)->icp, &local_err);
415         if (local_err) {
416             goto fail;
417         }
418     }
419 
420     return 0;
421 
422 fail:
423     error_propagate(errp, local_err);
424     xics_kvm_disconnect(spapr, NULL);
425     return -1;
426 }
427 
428 void xics_kvm_disconnect(SpaprMachineState *spapr, Error **errp)
429 {
430     /* The KVM XICS device is not in use */
431     if (kernel_xics_fd == -1) {
432         return;
433     }
434 
435     if (!kvm_enabled() || !kvm_check_extension(kvm_state, KVM_CAP_IRQ_XICS)) {
436         error_setg(errp,
437                    "KVM and IRQ_XICS capability must be present for KVM XICS device");
438         return;
439     }
440 
441     /*
442      * Only on P9 using the XICS-on XIVE KVM device:
443      *
444      * When the KVM device fd is closed, the device is destroyed and
445      * removed from the list of devices of the VM. The VCPU presenters
446      * are also detached from the device.
447      */
448     if (kernel_xics_fd != -1) {
449         close(kernel_xics_fd);
450         kernel_xics_fd = -1;
451     }
452 
453     kvmppc_define_rtas_kernel_token(0, "ibm,set-xive");
454     kvmppc_define_rtas_kernel_token(0, "ibm,get-xive");
455     kvmppc_define_rtas_kernel_token(0, "ibm,int-on");
456     kvmppc_define_rtas_kernel_token(0, "ibm,int-off");
457 
458     kvm_kernel_irqchip = false;
459     kvm_msi_via_irqfd_allowed = false;
460     kvm_gsi_direct_mapping = false;
461 
462     /* Clear the presenter from the VCPUs */
463     kvm_disable_icps();
464 }
465 
466 /*
467  * This is a heuristic to detect older KVMs on POWER9 hosts that don't
468  * support destruction of a KVM XICS device while the VM is running.
469  * Required to start a spapr machine with ic-mode=dual,kernel-irqchip=on.
470  */
471 bool xics_kvm_has_broken_disconnect(SpaprMachineState *spapr)
472 {
473     int rc;
474 
475     rc = kvm_create_device(kvm_state, KVM_DEV_TYPE_XICS, false);
476     if (rc < 0) {
477         /*
478          * The error is ignored on purpose. The KVM XICS setup code
479          * will catch it again anyway. The goal here is to see if
480          * close() actually destroys the device or not.
481          */
482         return false;
483     }
484 
485     close(rc);
486 
487     rc = kvm_create_device(kvm_state, KVM_DEV_TYPE_XICS, false);
488     if (rc >= 0) {
489         close(rc);
490         return false;
491     }
492 
493     return errno == EEXIST;
494 }
495