xref: /openbmc/qemu/hw/intc/xics_kvm.c (revision 795c40b8)
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/xics.h"
37 #include "kvm_ppc.h"
38 #include "qemu/config-file.h"
39 #include "qemu/error-report.h"
40 
41 #include <sys/ioctl.h>
42 
43 static int kernel_xics_fd = -1;
44 
45 /*
46  * ICP-KVM
47  */
48 static void icp_get_kvm_state(ICPState *icp)
49 {
50     uint64_t state;
51     struct kvm_one_reg reg = {
52         .id = KVM_REG_PPC_ICP_STATE,
53         .addr = (uintptr_t)&state,
54     };
55     int ret;
56 
57     /* ICP for this CPU thread is not in use, exiting */
58     if (!icp->cs) {
59         return;
60     }
61 
62     ret = kvm_vcpu_ioctl(icp->cs, KVM_GET_ONE_REG, &reg);
63     if (ret != 0) {
64         error_report("Unable to retrieve KVM interrupt controller state"
65                 " for CPU %ld: %s", kvm_arch_vcpu_id(icp->cs), strerror(errno));
66         exit(1);
67     }
68 
69     icp->xirr = state >> KVM_REG_PPC_ICP_XISR_SHIFT;
70     icp->mfrr = (state >> KVM_REG_PPC_ICP_MFRR_SHIFT)
71         & KVM_REG_PPC_ICP_MFRR_MASK;
72     icp->pending_priority = (state >> KVM_REG_PPC_ICP_PPRI_SHIFT)
73         & KVM_REG_PPC_ICP_PPRI_MASK;
74 }
75 
76 static int icp_set_kvm_state(ICPState *icp, int version_id)
77 {
78     uint64_t state;
79     struct kvm_one_reg reg = {
80         .id = KVM_REG_PPC_ICP_STATE,
81         .addr = (uintptr_t)&state,
82     };
83     int ret;
84 
85     /* ICP for this CPU thread is not in use, exiting */
86     if (!icp->cs) {
87         return 0;
88     }
89 
90     state = ((uint64_t)icp->xirr << KVM_REG_PPC_ICP_XISR_SHIFT)
91         | ((uint64_t)icp->mfrr << KVM_REG_PPC_ICP_MFRR_SHIFT)
92         | ((uint64_t)icp->pending_priority << KVM_REG_PPC_ICP_PPRI_SHIFT);
93 
94     ret = kvm_vcpu_ioctl(icp->cs, KVM_SET_ONE_REG, &reg);
95     if (ret != 0) {
96         error_report("Unable to restore KVM interrupt controller state (0x%"
97                 PRIx64 ") for CPU %ld: %s", state, kvm_arch_vcpu_id(icp->cs),
98                 strerror(errno));
99         return ret;
100     }
101 
102     return 0;
103 }
104 
105 static void icp_kvm_reset(void *dev)
106 {
107     ICPState *icp = ICP(dev);
108 
109     icp->xirr = 0;
110     icp->pending_priority = 0xff;
111     icp->mfrr = 0xff;
112 
113     /* Make all outputs as deasserted only if the CPU thread is in use */
114     if (icp->output) {
115         qemu_set_irq(icp->output, 0);
116     }
117 
118     icp_set_kvm_state(icp, 1);
119 }
120 
121 static void icp_kvm_cpu_setup(ICPState *icp, PowerPCCPU *cpu)
122 {
123     CPUState *cs = CPU(cpu);
124     int ret;
125 
126     if (kernel_xics_fd == -1) {
127         abort();
128     }
129 
130     /*
131      * If we are reusing a parked vCPU fd corresponding to the CPU
132      * which was hot-removed earlier we don't have to renable
133      * KVM_CAP_IRQ_XICS capability again.
134      */
135     if (icp->cap_irq_xics_enabled) {
136         return;
137     }
138 
139     ret = kvm_vcpu_enable_cap(cs, KVM_CAP_IRQ_XICS, 0, kernel_xics_fd,
140                               kvm_arch_vcpu_id(cs));
141     if (ret < 0) {
142         error_report("Unable to connect CPU%ld to kernel XICS: %s",
143                      kvm_arch_vcpu_id(cs), strerror(errno));
144         exit(1);
145     }
146     icp->cap_irq_xics_enabled = true;
147 }
148 
149 static void icp_kvm_realize(DeviceState *dev, Error **errp)
150 {
151     qemu_register_reset(icp_kvm_reset, dev);
152 }
153 
154 static void icp_kvm_class_init(ObjectClass *klass, void *data)
155 {
156     DeviceClass *dc = DEVICE_CLASS(klass);
157     ICPStateClass *icpc = ICP_CLASS(klass);
158 
159     dc->realize = icp_kvm_realize;
160     icpc->pre_save = icp_get_kvm_state;
161     icpc->post_load = icp_set_kvm_state;
162     icpc->cpu_setup = icp_kvm_cpu_setup;
163 }
164 
165 static const TypeInfo icp_kvm_info = {
166     .name = TYPE_KVM_ICP,
167     .parent = TYPE_ICP,
168     .instance_size = sizeof(ICPState),
169     .class_init = icp_kvm_class_init,
170     .class_size = sizeof(ICPStateClass),
171 };
172 
173 /*
174  * ICS-KVM
175  */
176 static void ics_get_kvm_state(ICSState *ics)
177 {
178     uint64_t state;
179     struct kvm_device_attr attr = {
180         .flags = 0,
181         .group = KVM_DEV_XICS_GRP_SOURCES,
182         .addr = (uint64_t)(uintptr_t)&state,
183     };
184     int i;
185 
186     for (i = 0; i < ics->nr_irqs; i++) {
187         ICSIRQState *irq = &ics->irqs[i];
188         int ret;
189 
190         attr.attr = i + ics->offset;
191 
192         ret = ioctl(kernel_xics_fd, KVM_GET_DEVICE_ATTR, &attr);
193         if (ret != 0) {
194             error_report("Unable to retrieve KVM interrupt controller state"
195                     " for IRQ %d: %s", i + ics->offset, strerror(errno));
196             exit(1);
197         }
198 
199         irq->server = state & KVM_XICS_DESTINATION_MASK;
200         irq->saved_priority = (state >> KVM_XICS_PRIORITY_SHIFT)
201             & KVM_XICS_PRIORITY_MASK;
202         /*
203          * To be consistent with the software emulation in xics.c, we
204          * split out the masked state + priority that we get from the
205          * kernel into 'current priority' (0xff if masked) and
206          * 'saved priority' (if masked, this is the priority the
207          * interrupt had before it was masked).  Masking and unmasking
208          * are done with the ibm,int-off and ibm,int-on RTAS calls.
209          */
210         if (state & KVM_XICS_MASKED) {
211             irq->priority = 0xff;
212         } else {
213             irq->priority = irq->saved_priority;
214         }
215 
216         irq->status = 0;
217         if (state & KVM_XICS_PENDING) {
218             if (state & KVM_XICS_LEVEL_SENSITIVE) {
219                 irq->status |= XICS_STATUS_ASSERTED;
220             } else {
221                 /*
222                  * A pending edge-triggered interrupt (or MSI)
223                  * must have been rejected previously when we
224                  * first detected it and tried to deliver it,
225                  * so mark it as pending and previously rejected
226                  * for consistency with how xics.c works.
227                  */
228                 irq->status |= XICS_STATUS_MASKED_PENDING
229                     | XICS_STATUS_REJECTED;
230             }
231         }
232         if (state & KVM_XICS_PRESENTED) {
233                 irq->status |= XICS_STATUS_PRESENTED;
234         }
235         if (state & KVM_XICS_QUEUED) {
236                 irq->status |= XICS_STATUS_QUEUED;
237         }
238     }
239 }
240 
241 static int ics_set_kvm_state(ICSState *ics, int version_id)
242 {
243     uint64_t state;
244     struct kvm_device_attr attr = {
245         .flags = 0,
246         .group = KVM_DEV_XICS_GRP_SOURCES,
247         .addr = (uint64_t)(uintptr_t)&state,
248     };
249     int i;
250 
251     for (i = 0; i < ics->nr_irqs; i++) {
252         ICSIRQState *irq = &ics->irqs[i];
253         int ret;
254 
255         attr.attr = i + ics->offset;
256 
257         state = irq->server;
258         state |= (uint64_t)(irq->saved_priority & KVM_XICS_PRIORITY_MASK)
259             << KVM_XICS_PRIORITY_SHIFT;
260         if (irq->priority != irq->saved_priority) {
261             assert(irq->priority == 0xff);
262             state |= KVM_XICS_MASKED;
263         }
264 
265         if (ics->irqs[i].flags & XICS_FLAGS_IRQ_LSI) {
266             state |= KVM_XICS_LEVEL_SENSITIVE;
267             if (irq->status & XICS_STATUS_ASSERTED) {
268                 state |= KVM_XICS_PENDING;
269             }
270         } else {
271             if (irq->status & XICS_STATUS_MASKED_PENDING) {
272                 state |= KVM_XICS_PENDING;
273             }
274         }
275         if (irq->status & XICS_STATUS_PRESENTED) {
276                 state |= KVM_XICS_PRESENTED;
277         }
278         if (irq->status & XICS_STATUS_QUEUED) {
279                 state |= KVM_XICS_QUEUED;
280         }
281 
282         ret = ioctl(kernel_xics_fd, KVM_SET_DEVICE_ATTR, &attr);
283         if (ret != 0) {
284             error_report("Unable to restore KVM interrupt controller state"
285                     " for IRQs %d: %s", i + ics->offset, strerror(errno));
286             return ret;
287         }
288     }
289 
290     return 0;
291 }
292 
293 static void ics_kvm_set_irq(void *opaque, int srcno, int val)
294 {
295     ICSState *ics = opaque;
296     struct kvm_irq_level args;
297     int rc;
298 
299     args.irq = srcno + ics->offset;
300     if (ics->irqs[srcno].flags & XICS_FLAGS_IRQ_MSI) {
301         if (!val) {
302             return;
303         }
304         args.level = KVM_INTERRUPT_SET;
305     } else {
306         args.level = val ? KVM_INTERRUPT_SET_LEVEL : KVM_INTERRUPT_UNSET;
307     }
308     rc = kvm_vm_ioctl(kvm_state, KVM_IRQ_LINE, &args);
309     if (rc < 0) {
310         perror("kvm_irq_line");
311     }
312 }
313 
314 static void ics_kvm_reset(void *dev)
315 {
316     ICSState *ics = ICS_SIMPLE(dev);
317     int i;
318     uint8_t flags[ics->nr_irqs];
319 
320     for (i = 0; i < ics->nr_irqs; i++) {
321         flags[i] = ics->irqs[i].flags;
322     }
323 
324     memset(ics->irqs, 0, sizeof(ICSIRQState) * ics->nr_irqs);
325 
326     for (i = 0; i < ics->nr_irqs; i++) {
327         ics->irqs[i].priority = 0xff;
328         ics->irqs[i].saved_priority = 0xff;
329         ics->irqs[i].flags = flags[i];
330     }
331 
332     ics_set_kvm_state(ics, 1);
333 }
334 
335 static void ics_kvm_realize(DeviceState *dev, Error **errp)
336 {
337     ICSState *ics = ICS_SIMPLE(dev);
338 
339     if (!ics->nr_irqs) {
340         error_setg(errp, "Number of interrupts needs to be greater 0");
341         return;
342     }
343     ics->irqs = g_malloc0(ics->nr_irqs * sizeof(ICSIRQState));
344     ics->qirqs = qemu_allocate_irqs(ics_kvm_set_irq, ics, ics->nr_irqs);
345 
346     qemu_register_reset(ics_kvm_reset, dev);
347 }
348 
349 static void ics_kvm_class_init(ObjectClass *klass, void *data)
350 {
351     ICSStateClass *icsc = ICS_BASE_CLASS(klass);
352 
353     icsc->realize = ics_kvm_realize;
354     icsc->pre_save = ics_get_kvm_state;
355     icsc->post_load = ics_set_kvm_state;
356 }
357 
358 static const TypeInfo ics_kvm_info = {
359     .name = TYPE_ICS_KVM,
360     .parent = TYPE_ICS_SIMPLE,
361     .instance_size = sizeof(ICSState),
362     .class_init = ics_kvm_class_init,
363 };
364 
365 /*
366  * XICS-KVM
367  */
368 
369 static void rtas_dummy(PowerPCCPU *cpu, sPAPRMachineState *spapr,
370                        uint32_t token,
371                        uint32_t nargs, target_ulong args,
372                        uint32_t nret, target_ulong rets)
373 {
374     error_report("pseries: %s must never be called for in-kernel XICS",
375                  __func__);
376 }
377 
378 int xics_kvm_init(sPAPRMachineState *spapr, Error **errp)
379 {
380     int rc;
381     struct kvm_create_device xics_create_device = {
382         .type = KVM_DEV_TYPE_XICS,
383         .flags = 0,
384     };
385 
386     if (!kvm_enabled() || !kvm_check_extension(kvm_state, KVM_CAP_IRQ_XICS)) {
387         error_setg(errp,
388                    "KVM and IRQ_XICS capability must be present for in-kernel XICS");
389         goto fail;
390     }
391 
392     spapr_rtas_register(RTAS_IBM_SET_XIVE, "ibm,set-xive", rtas_dummy);
393     spapr_rtas_register(RTAS_IBM_GET_XIVE, "ibm,get-xive", rtas_dummy);
394     spapr_rtas_register(RTAS_IBM_INT_OFF, "ibm,int-off", rtas_dummy);
395     spapr_rtas_register(RTAS_IBM_INT_ON, "ibm,int-on", rtas_dummy);
396 
397     rc = kvmppc_define_rtas_kernel_token(RTAS_IBM_SET_XIVE, "ibm,set-xive");
398     if (rc < 0) {
399         error_setg(errp, "kvmppc_define_rtas_kernel_token: ibm,set-xive");
400         goto fail;
401     }
402 
403     rc = kvmppc_define_rtas_kernel_token(RTAS_IBM_GET_XIVE, "ibm,get-xive");
404     if (rc < 0) {
405         error_setg(errp, "kvmppc_define_rtas_kernel_token: ibm,get-xive");
406         goto fail;
407     }
408 
409     rc = kvmppc_define_rtas_kernel_token(RTAS_IBM_INT_ON, "ibm,int-on");
410     if (rc < 0) {
411         error_setg(errp, "kvmppc_define_rtas_kernel_token: ibm,int-on");
412         goto fail;
413     }
414 
415     rc = kvmppc_define_rtas_kernel_token(RTAS_IBM_INT_OFF, "ibm,int-off");
416     if (rc < 0) {
417         error_setg(errp, "kvmppc_define_rtas_kernel_token: ibm,int-off");
418         goto fail;
419     }
420 
421     /* Create the kernel ICP */
422     rc = kvm_vm_ioctl(kvm_state, KVM_CREATE_DEVICE, &xics_create_device);
423     if (rc < 0) {
424         error_setg_errno(errp, -rc, "Error on KVM_CREATE_DEVICE for XICS");
425         goto fail;
426     }
427 
428     kernel_xics_fd = xics_create_device.fd;
429 
430     kvm_kernel_irqchip = true;
431     kvm_msi_via_irqfd_allowed = true;
432     kvm_gsi_direct_mapping = true;
433 
434     return rc;
435 
436 fail:
437     kvmppc_define_rtas_kernel_token(0, "ibm,set-xive");
438     kvmppc_define_rtas_kernel_token(0, "ibm,get-xive");
439     kvmppc_define_rtas_kernel_token(0, "ibm,int-on");
440     kvmppc_define_rtas_kernel_token(0, "ibm,int-off");
441     return -1;
442 }
443 
444 static void xics_kvm_register_types(void)
445 {
446     type_register_static(&ics_kvm_info);
447     type_register_static(&icp_kvm_info);
448 }
449 
450 type_init(xics_kvm_register_types)
451