xref: /openbmc/qemu/hw/intc/xics_kvm.c (revision 56b11587dfc79dc6453d857159801dcc38f24d0c)
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 "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     /* ICP for this CPU thread is not in use, exiting */
73     if (!icp->cs) {
74         return;
75     }
76 
77     ret = kvm_get_one_reg(icp->cs, KVM_REG_PPC_ICP_STATE, &state);
78     if (ret != 0) {
79         error_report("Unable to retrieve KVM interrupt controller state"
80                 " for CPU %ld: %s", kvm_arch_vcpu_id(icp->cs), strerror(errno));
81         exit(1);
82     }
83 
84     icp->xirr = state >> KVM_REG_PPC_ICP_XISR_SHIFT;
85     icp->mfrr = (state >> KVM_REG_PPC_ICP_MFRR_SHIFT)
86         & KVM_REG_PPC_ICP_MFRR_MASK;
87     icp->pending_priority = (state >> KVM_REG_PPC_ICP_PPRI_SHIFT)
88         & KVM_REG_PPC_ICP_PPRI_MASK;
89 }
90 
91 static void do_icp_synchronize_state(CPUState *cpu, run_on_cpu_data arg)
92 {
93     icp_get_kvm_state(arg.host_ptr);
94 }
95 
96 void icp_synchronize_state(ICPState *icp)
97 {
98     if (icp->cs) {
99         run_on_cpu(icp->cs, do_icp_synchronize_state, RUN_ON_CPU_HOST_PTR(icp));
100     }
101 }
102 
103 int icp_set_kvm_state(ICPState *icp)
104 {
105     uint64_t state;
106     int ret;
107 
108     /* ICP for this CPU thread is not in use, exiting */
109     if (!icp->cs) {
110         return 0;
111     }
112 
113     state = ((uint64_t)icp->xirr << KVM_REG_PPC_ICP_XISR_SHIFT)
114         | ((uint64_t)icp->mfrr << KVM_REG_PPC_ICP_MFRR_SHIFT)
115         | ((uint64_t)icp->pending_priority << KVM_REG_PPC_ICP_PPRI_SHIFT);
116 
117     ret = kvm_set_one_reg(icp->cs, KVM_REG_PPC_ICP_STATE, &state);
118     if (ret != 0) {
119         error_report("Unable to restore KVM interrupt controller state (0x%"
120                 PRIx64 ") for CPU %ld: %s", state, kvm_arch_vcpu_id(icp->cs),
121                 strerror(errno));
122         return ret;
123     }
124 
125     return 0;
126 }
127 
128 void icp_kvm_realize(DeviceState *dev, Error **errp)
129 {
130     ICPState *icp = ICP(dev);
131     CPUState *cs;
132     KVMEnabledICP *enabled_icp;
133     unsigned long vcpu_id;
134     int ret;
135 
136     if (kernel_xics_fd == -1) {
137         abort();
138     }
139 
140     cs = icp->cs;
141     vcpu_id = kvm_arch_vcpu_id(cs);
142 
143     /*
144      * If we are reusing a parked vCPU fd corresponding to the CPU
145      * which was hot-removed earlier we don't have to renable
146      * KVM_CAP_IRQ_XICS capability again.
147      */
148     QLIST_FOREACH(enabled_icp, &kvm_enabled_icps, node) {
149         if (enabled_icp->vcpu_id == vcpu_id) {
150             return;
151         }
152     }
153 
154     ret = kvm_vcpu_enable_cap(cs, KVM_CAP_IRQ_XICS, 0, kernel_xics_fd, vcpu_id);
155     if (ret < 0) {
156         error_setg(errp, "Unable to connect CPU%ld to kernel XICS: %s", vcpu_id,
157                    strerror(errno));
158         return;
159     }
160     enabled_icp = g_malloc(sizeof(*enabled_icp));
161     enabled_icp->vcpu_id = vcpu_id;
162     QLIST_INSERT_HEAD(&kvm_enabled_icps, enabled_icp, node);
163 }
164 
165 /*
166  * ICS-KVM
167  */
168 void ics_get_kvm_state(ICSState *ics)
169 {
170     uint64_t state;
171     int i;
172 
173     for (i = 0; i < ics->nr_irqs; i++) {
174         ICSIRQState *irq = &ics->irqs[i];
175 
176         kvm_device_access(kernel_xics_fd, KVM_DEV_XICS_GRP_SOURCES,
177                           i + ics->offset, &state, false, &error_fatal);
178 
179         irq->server = state & KVM_XICS_DESTINATION_MASK;
180         irq->saved_priority = (state >> KVM_XICS_PRIORITY_SHIFT)
181             & KVM_XICS_PRIORITY_MASK;
182         /*
183          * To be consistent with the software emulation in xics.c, we
184          * split out the masked state + priority that we get from the
185          * kernel into 'current priority' (0xff if masked) and
186          * 'saved priority' (if masked, this is the priority the
187          * interrupt had before it was masked).  Masking and unmasking
188          * are done with the ibm,int-off and ibm,int-on RTAS calls.
189          */
190         if (state & KVM_XICS_MASKED) {
191             irq->priority = 0xff;
192         } else {
193             irq->priority = irq->saved_priority;
194         }
195 
196         irq->status = 0;
197         if (state & KVM_XICS_PENDING) {
198             if (state & KVM_XICS_LEVEL_SENSITIVE) {
199                 irq->status |= XICS_STATUS_ASSERTED;
200             } else {
201                 /*
202                  * A pending edge-triggered interrupt (or MSI)
203                  * must have been rejected previously when we
204                  * first detected it and tried to deliver it,
205                  * so mark it as pending and previously rejected
206                  * for consistency with how xics.c works.
207                  */
208                 irq->status |= XICS_STATUS_MASKED_PENDING
209                     | XICS_STATUS_REJECTED;
210             }
211         }
212         if (state & KVM_XICS_PRESENTED) {
213                 irq->status |= XICS_STATUS_PRESENTED;
214         }
215         if (state & KVM_XICS_QUEUED) {
216                 irq->status |= XICS_STATUS_QUEUED;
217         }
218     }
219 }
220 
221 void ics_synchronize_state(ICSState *ics)
222 {
223     ics_get_kvm_state(ics);
224 }
225 
226 int ics_set_kvm_state_one(ICSState *ics, int srcno)
227 {
228     uint64_t state;
229     Error *local_err = NULL;
230     ICSIRQState *irq = &ics->irqs[srcno];
231     int ret;
232 
233     state = irq->server;
234     state |= (uint64_t)(irq->saved_priority & KVM_XICS_PRIORITY_MASK)
235         << KVM_XICS_PRIORITY_SHIFT;
236     if (irq->priority != irq->saved_priority) {
237         assert(irq->priority == 0xff);
238         state |= KVM_XICS_MASKED;
239     }
240 
241     if (irq->flags & XICS_FLAGS_IRQ_LSI) {
242         state |= KVM_XICS_LEVEL_SENSITIVE;
243         if (irq->status & XICS_STATUS_ASSERTED) {
244             state |= KVM_XICS_PENDING;
245         }
246     } else {
247         if (irq->status & XICS_STATUS_MASKED_PENDING) {
248             state |= KVM_XICS_PENDING;
249         }
250     }
251     if (irq->status & XICS_STATUS_PRESENTED) {
252         state |= KVM_XICS_PRESENTED;
253     }
254     if (irq->status & XICS_STATUS_QUEUED) {
255         state |= KVM_XICS_QUEUED;
256     }
257 
258     ret = kvm_device_access(kernel_xics_fd, KVM_DEV_XICS_GRP_SOURCES,
259                             srcno + ics->offset, &state, true, &local_err);
260     if (local_err) {
261         error_report_err(local_err);
262         return ret;
263     }
264 
265     return 0;
266 }
267 
268 int ics_set_kvm_state(ICSState *ics)
269 {
270     int i;
271 
272     for (i = 0; i < ics->nr_irqs; i++) {
273         int ret;
274 
275         ret = ics_set_kvm_state_one(ics, i);
276         if (ret) {
277             return ret;
278         }
279     }
280 
281     return 0;
282 }
283 
284 void ics_kvm_set_irq(ICSState *ics, int srcno, int val)
285 {
286     struct kvm_irq_level args;
287     int rc;
288 
289     args.irq = srcno + ics->offset;
290     if (ics->irqs[srcno].flags & XICS_FLAGS_IRQ_MSI) {
291         if (!val) {
292             return;
293         }
294         args.level = KVM_INTERRUPT_SET;
295     } else {
296         args.level = val ? KVM_INTERRUPT_SET_LEVEL : KVM_INTERRUPT_UNSET;
297     }
298     rc = kvm_vm_ioctl(kvm_state, KVM_IRQ_LINE, &args);
299     if (rc < 0) {
300         perror("kvm_irq_line");
301     }
302 }
303 
304 static void rtas_dummy(PowerPCCPU *cpu, SpaprMachineState *spapr,
305                        uint32_t token,
306                        uint32_t nargs, target_ulong args,
307                        uint32_t nret, target_ulong rets)
308 {
309     error_report("pseries: %s must never be called for in-kernel XICS",
310                  __func__);
311 }
312 
313 int xics_kvm_init(SpaprMachineState *spapr, Error **errp)
314 {
315     int rc;
316 
317     if (!kvm_enabled() || !kvm_check_extension(kvm_state, KVM_CAP_IRQ_XICS)) {
318         error_setg(errp,
319                    "KVM and IRQ_XICS capability must be present for in-kernel XICS");
320         goto fail;
321     }
322 
323     spapr_rtas_register(RTAS_IBM_SET_XIVE, "ibm,set-xive", rtas_dummy);
324     spapr_rtas_register(RTAS_IBM_GET_XIVE, "ibm,get-xive", rtas_dummy);
325     spapr_rtas_register(RTAS_IBM_INT_OFF, "ibm,int-off", rtas_dummy);
326     spapr_rtas_register(RTAS_IBM_INT_ON, "ibm,int-on", rtas_dummy);
327 
328     rc = kvmppc_define_rtas_kernel_token(RTAS_IBM_SET_XIVE, "ibm,set-xive");
329     if (rc < 0) {
330         error_setg(errp, "kvmppc_define_rtas_kernel_token: ibm,set-xive");
331         goto fail;
332     }
333 
334     rc = kvmppc_define_rtas_kernel_token(RTAS_IBM_GET_XIVE, "ibm,get-xive");
335     if (rc < 0) {
336         error_setg(errp, "kvmppc_define_rtas_kernel_token: ibm,get-xive");
337         goto fail;
338     }
339 
340     rc = kvmppc_define_rtas_kernel_token(RTAS_IBM_INT_ON, "ibm,int-on");
341     if (rc < 0) {
342         error_setg(errp, "kvmppc_define_rtas_kernel_token: ibm,int-on");
343         goto fail;
344     }
345 
346     rc = kvmppc_define_rtas_kernel_token(RTAS_IBM_INT_OFF, "ibm,int-off");
347     if (rc < 0) {
348         error_setg(errp, "kvmppc_define_rtas_kernel_token: ibm,int-off");
349         goto fail;
350     }
351 
352     /* Create the KVM XICS device */
353     rc = kvm_create_device(kvm_state, KVM_DEV_TYPE_XICS, false);
354     if (rc < 0) {
355         error_setg_errno(errp, -rc, "Error on KVM_CREATE_DEVICE for XICS");
356         goto fail;
357     }
358 
359     kernel_xics_fd = rc;
360     kvm_kernel_irqchip = true;
361     kvm_msi_via_irqfd_allowed = true;
362     kvm_gsi_direct_mapping = true;
363 
364     return 0;
365 
366 fail:
367     kvmppc_define_rtas_kernel_token(0, "ibm,set-xive");
368     kvmppc_define_rtas_kernel_token(0, "ibm,get-xive");
369     kvmppc_define_rtas_kernel_token(0, "ibm,int-on");
370     kvmppc_define_rtas_kernel_token(0, "ibm,int-off");
371     return -1;
372 }
373 
374 void xics_kvm_disconnect(SpaprMachineState *spapr, Error **errp)
375 {
376     /* The KVM XICS device is not in use */
377     if (kernel_xics_fd == -1) {
378         return;
379     }
380 
381     if (!kvm_enabled() || !kvm_check_extension(kvm_state, KVM_CAP_IRQ_XICS)) {
382         error_setg(errp,
383                    "KVM and IRQ_XICS capability must be present for KVM XICS device");
384         return;
385     }
386 
387     /*
388      * Only on P9 using the XICS-on XIVE KVM device:
389      *
390      * When the KVM device fd is closed, the device is destroyed and
391      * removed from the list of devices of the VM. The VCPU presenters
392      * are also detached from the device.
393      */
394     close(kernel_xics_fd);
395     kernel_xics_fd = -1;
396 
397     spapr_rtas_unregister(RTAS_IBM_SET_XIVE);
398     spapr_rtas_unregister(RTAS_IBM_GET_XIVE);
399     spapr_rtas_unregister(RTAS_IBM_INT_OFF);
400     spapr_rtas_unregister(RTAS_IBM_INT_ON);
401 
402     kvmppc_define_rtas_kernel_token(0, "ibm,set-xive");
403     kvmppc_define_rtas_kernel_token(0, "ibm,get-xive");
404     kvmppc_define_rtas_kernel_token(0, "ibm,int-on");
405     kvmppc_define_rtas_kernel_token(0, "ibm,int-off");
406 
407     kvm_kernel_irqchip = false;
408     kvm_msi_via_irqfd_allowed = false;
409     kvm_gsi_direct_mapping = false;
410 
411     /* Clear the presenter from the VCPUs */
412     kvm_disable_icps();
413 }
414