xref: /openbmc/qemu/hw/ppc/spapr.c (revision a7ff1212e99ff072dfb8db62d5e6d8ce9f4b486c)
1 /*
2  * QEMU PowerPC pSeries Logical Partition (aka sPAPR) hardware System Emulator
3  *
4  * Copyright (c) 2004-2007 Fabrice Bellard
5  * Copyright (c) 2007 Jocelyn Mayer
6  * Copyright (c) 2010 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 #include "qemu/osdep.h"
28 #include "qapi/error.h"
29 #include "sysemu/sysemu.h"
30 #include "sysemu/numa.h"
31 #include "hw/hw.h"
32 #include "qemu/log.h"
33 #include "hw/fw-path-provider.h"
34 #include "elf.h"
35 #include "net/net.h"
36 #include "sysemu/device_tree.h"
37 #include "sysemu/block-backend.h"
38 #include "sysemu/cpus.h"
39 #include "sysemu/hw_accel.h"
40 #include "kvm_ppc.h"
41 #include "migration/migration.h"
42 #include "mmu-hash64.h"
43 #include "qom/cpu.h"
44 
45 #include "hw/boards.h"
46 #include "hw/ppc/ppc.h"
47 #include "hw/loader.h"
48 
49 #include "hw/ppc/fdt.h"
50 #include "hw/ppc/spapr.h"
51 #include "hw/ppc/spapr_vio.h"
52 #include "hw/pci-host/spapr.h"
53 #include "hw/ppc/xics.h"
54 #include "hw/pci/msi.h"
55 
56 #include "hw/pci/pci.h"
57 #include "hw/scsi/scsi.h"
58 #include "hw/virtio/virtio-scsi.h"
59 
60 #include "exec/address-spaces.h"
61 #include "hw/usb.h"
62 #include "qemu/config-file.h"
63 #include "qemu/error-report.h"
64 #include "trace.h"
65 #include "hw/nmi.h"
66 
67 #include "hw/compat.h"
68 #include "qemu/cutils.h"
69 #include "hw/ppc/spapr_cpu_core.h"
70 #include "qmp-commands.h"
71 
72 #include <libfdt.h>
73 
74 /* SLOF memory layout:
75  *
76  * SLOF raw image loaded at 0, copies its romfs right below the flat
77  * device-tree, then position SLOF itself 31M below that
78  *
79  * So we set FW_OVERHEAD to 40MB which should account for all of that
80  * and more
81  *
82  * We load our kernel at 4M, leaving space for SLOF initial image
83  */
84 #define FDT_MAX_SIZE            0x100000
85 #define RTAS_MAX_SIZE           0x10000
86 #define RTAS_MAX_ADDR           0x80000000 /* RTAS must stay below that */
87 #define FW_MAX_SIZE             0x400000
88 #define FW_FILE_NAME            "slof.bin"
89 #define FW_OVERHEAD             0x2800000
90 #define KERNEL_LOAD_ADDR        FW_MAX_SIZE
91 
92 #define MIN_RMA_SLOF            128UL
93 
94 #define PHANDLE_XICP            0x00001111
95 
96 #define HTAB_SIZE(spapr)        (1ULL << ((spapr)->htab_shift))
97 
98 static int try_create_xics(sPAPRMachineState *spapr, const char *type_ics,
99                            const char *type_icp, int nr_servers,
100                            int nr_irqs, Error **errp)
101 {
102     XICSFabric *xi = XICS_FABRIC(spapr);
103     Error *err = NULL, *local_err = NULL;
104     ICSState *ics = NULL;
105     int i;
106 
107     ics = ICS_SIMPLE(object_new(type_ics));
108     qdev_set_parent_bus(DEVICE(ics), sysbus_get_default());
109     object_property_add_child(OBJECT(spapr), "ics", OBJECT(ics), NULL);
110     object_property_set_int(OBJECT(ics), nr_irqs, "nr-irqs", &err);
111     object_property_add_const_link(OBJECT(ics), "xics", OBJECT(xi), NULL);
112     object_property_set_bool(OBJECT(ics), true, "realized", &local_err);
113     error_propagate(&err, local_err);
114     if (err) {
115         goto error;
116     }
117 
118     spapr->icps = g_malloc0(nr_servers * sizeof(ICPState));
119     spapr->nr_servers = nr_servers;
120 
121     for (i = 0; i < nr_servers; i++) {
122         ICPState *icp = &spapr->icps[i];
123 
124         object_initialize(icp, sizeof(*icp), type_icp);
125         qdev_set_parent_bus(DEVICE(icp), sysbus_get_default());
126         object_property_add_child(OBJECT(spapr), "icp[*]", OBJECT(icp), NULL);
127         object_property_add_const_link(OBJECT(icp), "xics", OBJECT(xi), NULL);
128         object_property_set_bool(OBJECT(icp), true, "realized", &err);
129         if (err) {
130             goto error;
131         }
132         object_unref(OBJECT(icp));
133     }
134 
135     spapr->ics = ics;
136     return 0;
137 
138 error:
139     error_propagate(errp, err);
140     if (ics) {
141         object_unparent(OBJECT(ics));
142     }
143     return -1;
144 }
145 
146 static int xics_system_init(MachineState *machine,
147                             int nr_servers, int nr_irqs, Error **errp)
148 {
149     int rc = -1;
150 
151     if (kvm_enabled()) {
152         Error *err = NULL;
153 
154         if (machine_kernel_irqchip_allowed(machine) &&
155             !xics_kvm_init(SPAPR_MACHINE(machine), errp)) {
156             rc = try_create_xics(SPAPR_MACHINE(machine), TYPE_ICS_KVM,
157                                  TYPE_KVM_ICP, nr_servers, nr_irqs, &err);
158         }
159         if (machine_kernel_irqchip_required(machine) && rc < 0) {
160             error_reportf_err(err,
161                               "kernel_irqchip requested but unavailable: ");
162         } else {
163             error_free(err);
164         }
165     }
166 
167     if (rc < 0) {
168         xics_spapr_init(SPAPR_MACHINE(machine), errp);
169         rc = try_create_xics(SPAPR_MACHINE(machine), TYPE_ICS_SIMPLE,
170                                TYPE_ICP, nr_servers, nr_irqs, errp);
171     }
172 
173     return rc;
174 }
175 
176 static int spapr_fixup_cpu_smt_dt(void *fdt, int offset, PowerPCCPU *cpu,
177                                   int smt_threads)
178 {
179     int i, ret = 0;
180     uint32_t servers_prop[smt_threads];
181     uint32_t gservers_prop[smt_threads * 2];
182     int index = ppc_get_vcpu_dt_id(cpu);
183 
184     if (cpu->compat_pvr) {
185         ret = fdt_setprop_cell(fdt, offset, "cpu-version", cpu->compat_pvr);
186         if (ret < 0) {
187             return ret;
188         }
189     }
190 
191     /* Build interrupt servers and gservers properties */
192     for (i = 0; i < smt_threads; i++) {
193         servers_prop[i] = cpu_to_be32(index + i);
194         /* Hack, direct the group queues back to cpu 0 */
195         gservers_prop[i*2] = cpu_to_be32(index + i);
196         gservers_prop[i*2 + 1] = 0;
197     }
198     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-server#s",
199                       servers_prop, sizeof(servers_prop));
200     if (ret < 0) {
201         return ret;
202     }
203     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-gserver#s",
204                       gservers_prop, sizeof(gservers_prop));
205 
206     return ret;
207 }
208 
209 static int spapr_fixup_cpu_numa_dt(void *fdt, int offset, CPUState *cs)
210 {
211     int ret = 0;
212     PowerPCCPU *cpu = POWERPC_CPU(cs);
213     int index = ppc_get_vcpu_dt_id(cpu);
214     uint32_t associativity[] = {cpu_to_be32(0x5),
215                                 cpu_to_be32(0x0),
216                                 cpu_to_be32(0x0),
217                                 cpu_to_be32(0x0),
218                                 cpu_to_be32(cs->numa_node),
219                                 cpu_to_be32(index)};
220 
221     /* Advertise NUMA via ibm,associativity */
222     if (nb_numa_nodes > 1) {
223         ret = fdt_setprop(fdt, offset, "ibm,associativity", associativity,
224                           sizeof(associativity));
225     }
226 
227     return ret;
228 }
229 
230 static int spapr_fixup_cpu_dt(void *fdt, sPAPRMachineState *spapr)
231 {
232     int ret = 0, offset, cpus_offset;
233     CPUState *cs;
234     char cpu_model[32];
235     int smt = kvmppc_smt_threads();
236     uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
237 
238     CPU_FOREACH(cs) {
239         PowerPCCPU *cpu = POWERPC_CPU(cs);
240         DeviceClass *dc = DEVICE_GET_CLASS(cs);
241         int index = ppc_get_vcpu_dt_id(cpu);
242         int compat_smt = MIN(smp_threads, ppc_compat_max_threads(cpu));
243 
244         if ((index % smt) != 0) {
245             continue;
246         }
247 
248         snprintf(cpu_model, 32, "%s@%x", dc->fw_name, index);
249 
250         cpus_offset = fdt_path_offset(fdt, "/cpus");
251         if (cpus_offset < 0) {
252             cpus_offset = fdt_add_subnode(fdt, fdt_path_offset(fdt, "/"),
253                                           "cpus");
254             if (cpus_offset < 0) {
255                 return cpus_offset;
256             }
257         }
258         offset = fdt_subnode_offset(fdt, cpus_offset, cpu_model);
259         if (offset < 0) {
260             offset = fdt_add_subnode(fdt, cpus_offset, cpu_model);
261             if (offset < 0) {
262                 return offset;
263             }
264         }
265 
266         ret = fdt_setprop(fdt, offset, "ibm,pft-size",
267                           pft_size_prop, sizeof(pft_size_prop));
268         if (ret < 0) {
269             return ret;
270         }
271 
272         ret = spapr_fixup_cpu_numa_dt(fdt, offset, cs);
273         if (ret < 0) {
274             return ret;
275         }
276 
277         ret = spapr_fixup_cpu_smt_dt(fdt, offset, cpu, compat_smt);
278         if (ret < 0) {
279             return ret;
280         }
281     }
282     return ret;
283 }
284 
285 static hwaddr spapr_node0_size(void)
286 {
287     MachineState *machine = MACHINE(qdev_get_machine());
288 
289     if (nb_numa_nodes) {
290         int i;
291         for (i = 0; i < nb_numa_nodes; ++i) {
292             if (numa_info[i].node_mem) {
293                 return MIN(pow2floor(numa_info[i].node_mem),
294                            machine->ram_size);
295             }
296         }
297     }
298     return machine->ram_size;
299 }
300 
301 static void add_str(GString *s, const gchar *s1)
302 {
303     g_string_append_len(s, s1, strlen(s1) + 1);
304 }
305 
306 static int spapr_populate_memory_node(void *fdt, int nodeid, hwaddr start,
307                                        hwaddr size)
308 {
309     uint32_t associativity[] = {
310         cpu_to_be32(0x4), /* length */
311         cpu_to_be32(0x0), cpu_to_be32(0x0),
312         cpu_to_be32(0x0), cpu_to_be32(nodeid)
313     };
314     char mem_name[32];
315     uint64_t mem_reg_property[2];
316     int off;
317 
318     mem_reg_property[0] = cpu_to_be64(start);
319     mem_reg_property[1] = cpu_to_be64(size);
320 
321     sprintf(mem_name, "memory@" TARGET_FMT_lx, start);
322     off = fdt_add_subnode(fdt, 0, mem_name);
323     _FDT(off);
324     _FDT((fdt_setprop_string(fdt, off, "device_type", "memory")));
325     _FDT((fdt_setprop(fdt, off, "reg", mem_reg_property,
326                       sizeof(mem_reg_property))));
327     _FDT((fdt_setprop(fdt, off, "ibm,associativity", associativity,
328                       sizeof(associativity))));
329     return off;
330 }
331 
332 static int spapr_populate_memory(sPAPRMachineState *spapr, void *fdt)
333 {
334     MachineState *machine = MACHINE(spapr);
335     hwaddr mem_start, node_size;
336     int i, nb_nodes = nb_numa_nodes;
337     NodeInfo *nodes = numa_info;
338     NodeInfo ramnode;
339 
340     /* No NUMA nodes, assume there is just one node with whole RAM */
341     if (!nb_numa_nodes) {
342         nb_nodes = 1;
343         ramnode.node_mem = machine->ram_size;
344         nodes = &ramnode;
345     }
346 
347     for (i = 0, mem_start = 0; i < nb_nodes; ++i) {
348         if (!nodes[i].node_mem) {
349             continue;
350         }
351         if (mem_start >= machine->ram_size) {
352             node_size = 0;
353         } else {
354             node_size = nodes[i].node_mem;
355             if (node_size > machine->ram_size - mem_start) {
356                 node_size = machine->ram_size - mem_start;
357             }
358         }
359         if (!mem_start) {
360             /* ppc_spapr_init() checks for rma_size <= node0_size already */
361             spapr_populate_memory_node(fdt, i, 0, spapr->rma_size);
362             mem_start += spapr->rma_size;
363             node_size -= spapr->rma_size;
364         }
365         for ( ; node_size; ) {
366             hwaddr sizetmp = pow2floor(node_size);
367 
368             /* mem_start != 0 here */
369             if (ctzl(mem_start) < ctzl(sizetmp)) {
370                 sizetmp = 1ULL << ctzl(mem_start);
371             }
372 
373             spapr_populate_memory_node(fdt, i, mem_start, sizetmp);
374             node_size -= sizetmp;
375             mem_start += sizetmp;
376         }
377     }
378 
379     return 0;
380 }
381 
382 /* Populate the "ibm,pa-features" property */
383 static void spapr_populate_pa_features(CPUPPCState *env, void *fdt, int offset)
384 {
385     uint8_t pa_features_206[] = { 6, 0,
386         0xf6, 0x1f, 0xc7, 0x00, 0x80, 0xc0 };
387     uint8_t pa_features_207[] = { 24, 0,
388         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0,
389         0x80, 0x00, 0x00, 0x00, 0x00, 0x00,
390         0x00, 0x00, 0x00, 0x00, 0x80, 0x00,
391         0x80, 0x00, 0x80, 0x00, 0x00, 0x00 };
392     uint8_t *pa_features;
393     size_t pa_size;
394 
395     switch (env->mmu_model) {
396     case POWERPC_MMU_2_06:
397     case POWERPC_MMU_2_06a:
398         pa_features = pa_features_206;
399         pa_size = sizeof(pa_features_206);
400         break;
401     case POWERPC_MMU_2_07:
402     case POWERPC_MMU_2_07a:
403         pa_features = pa_features_207;
404         pa_size = sizeof(pa_features_207);
405         break;
406     default:
407         return;
408     }
409 
410     if (env->ci_large_pages) {
411         /*
412          * Note: we keep CI large pages off by default because a 64K capable
413          * guest provisioned with large pages might otherwise try to map a qemu
414          * framebuffer (or other kind of memory mapped PCI BAR) using 64K pages
415          * even if that qemu runs on a 4k host.
416          * We dd this bit back here if we are confident this is not an issue
417          */
418         pa_features[3] |= 0x20;
419     }
420     if (kvmppc_has_cap_htm() && pa_size > 24) {
421         pa_features[24] |= 0x80;    /* Transactional memory support */
422     }
423 
424     _FDT((fdt_setprop(fdt, offset, "ibm,pa-features", pa_features, pa_size)));
425 }
426 
427 static void spapr_populate_cpu_dt(CPUState *cs, void *fdt, int offset,
428                                   sPAPRMachineState *spapr)
429 {
430     PowerPCCPU *cpu = POWERPC_CPU(cs);
431     CPUPPCState *env = &cpu->env;
432     PowerPCCPUClass *pcc = POWERPC_CPU_GET_CLASS(cs);
433     int index = ppc_get_vcpu_dt_id(cpu);
434     uint32_t segs[] = {cpu_to_be32(28), cpu_to_be32(40),
435                        0xffffffff, 0xffffffff};
436     uint32_t tbfreq = kvm_enabled() ? kvmppc_get_tbfreq()
437         : SPAPR_TIMEBASE_FREQ;
438     uint32_t cpufreq = kvm_enabled() ? kvmppc_get_clockfreq() : 1000000000;
439     uint32_t page_sizes_prop[64];
440     size_t page_sizes_prop_size;
441     uint32_t vcpus_per_socket = smp_threads * smp_cores;
442     uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
443     int compat_smt = MIN(smp_threads, ppc_compat_max_threads(cpu));
444     sPAPRDRConnector *drc;
445     sPAPRDRConnectorClass *drck;
446     int drc_index;
447 
448     drc = spapr_dr_connector_by_id(SPAPR_DR_CONNECTOR_TYPE_CPU, index);
449     if (drc) {
450         drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
451         drc_index = drck->get_index(drc);
452         _FDT((fdt_setprop_cell(fdt, offset, "ibm,my-drc-index", drc_index)));
453     }
454 
455     _FDT((fdt_setprop_cell(fdt, offset, "reg", index)));
456     _FDT((fdt_setprop_string(fdt, offset, "device_type", "cpu")));
457 
458     _FDT((fdt_setprop_cell(fdt, offset, "cpu-version", env->spr[SPR_PVR])));
459     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-block-size",
460                            env->dcache_line_size)));
461     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-line-size",
462                            env->dcache_line_size)));
463     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-block-size",
464                            env->icache_line_size)));
465     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-line-size",
466                            env->icache_line_size)));
467 
468     if (pcc->l1_dcache_size) {
469         _FDT((fdt_setprop_cell(fdt, offset, "d-cache-size",
470                                pcc->l1_dcache_size)));
471     } else {
472         error_report("Warning: Unknown L1 dcache size for cpu");
473     }
474     if (pcc->l1_icache_size) {
475         _FDT((fdt_setprop_cell(fdt, offset, "i-cache-size",
476                                pcc->l1_icache_size)));
477     } else {
478         error_report("Warning: Unknown L1 icache size for cpu");
479     }
480 
481     _FDT((fdt_setprop_cell(fdt, offset, "timebase-frequency", tbfreq)));
482     _FDT((fdt_setprop_cell(fdt, offset, "clock-frequency", cpufreq)));
483     _FDT((fdt_setprop_cell(fdt, offset, "slb-size", env->slb_nr)));
484     _FDT((fdt_setprop_cell(fdt, offset, "ibm,slb-size", env->slb_nr)));
485     _FDT((fdt_setprop_string(fdt, offset, "status", "okay")));
486     _FDT((fdt_setprop(fdt, offset, "64-bit", NULL, 0)));
487 
488     if (env->spr_cb[SPR_PURR].oea_read) {
489         _FDT((fdt_setprop(fdt, offset, "ibm,purr", NULL, 0)));
490     }
491 
492     if (env->mmu_model & POWERPC_MMU_1TSEG) {
493         _FDT((fdt_setprop(fdt, offset, "ibm,processor-segment-sizes",
494                           segs, sizeof(segs))));
495     }
496 
497     /* Advertise VMX/VSX (vector extensions) if available
498      *   0 / no property == no vector extensions
499      *   1               == VMX / Altivec available
500      *   2               == VSX available */
501     if (env->insns_flags & PPC_ALTIVEC) {
502         uint32_t vmx = (env->insns_flags2 & PPC2_VSX) ? 2 : 1;
503 
504         _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", vmx)));
505     }
506 
507     /* Advertise DFP (Decimal Floating Point) if available
508      *   0 / no property == no DFP
509      *   1               == DFP available */
510     if (env->insns_flags2 & PPC2_DFP) {
511         _FDT((fdt_setprop_cell(fdt, offset, "ibm,dfp", 1)));
512     }
513 
514     page_sizes_prop_size = ppc_create_page_sizes_prop(env, page_sizes_prop,
515                                                   sizeof(page_sizes_prop));
516     if (page_sizes_prop_size) {
517         _FDT((fdt_setprop(fdt, offset, "ibm,segment-page-sizes",
518                           page_sizes_prop, page_sizes_prop_size)));
519     }
520 
521     spapr_populate_pa_features(env, fdt, offset);
522 
523     _FDT((fdt_setprop_cell(fdt, offset, "ibm,chip-id",
524                            cs->cpu_index / vcpus_per_socket)));
525 
526     _FDT((fdt_setprop(fdt, offset, "ibm,pft-size",
527                       pft_size_prop, sizeof(pft_size_prop))));
528 
529     _FDT(spapr_fixup_cpu_numa_dt(fdt, offset, cs));
530 
531     _FDT(spapr_fixup_cpu_smt_dt(fdt, offset, cpu, compat_smt));
532 }
533 
534 static void spapr_populate_cpus_dt_node(void *fdt, sPAPRMachineState *spapr)
535 {
536     CPUState *cs;
537     int cpus_offset;
538     char *nodename;
539     int smt = kvmppc_smt_threads();
540 
541     cpus_offset = fdt_add_subnode(fdt, 0, "cpus");
542     _FDT(cpus_offset);
543     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#address-cells", 0x1)));
544     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#size-cells", 0x0)));
545 
546     /*
547      * We walk the CPUs in reverse order to ensure that CPU DT nodes
548      * created by fdt_add_subnode() end up in the right order in FDT
549      * for the guest kernel the enumerate the CPUs correctly.
550      */
551     CPU_FOREACH_REVERSE(cs) {
552         PowerPCCPU *cpu = POWERPC_CPU(cs);
553         int index = ppc_get_vcpu_dt_id(cpu);
554         DeviceClass *dc = DEVICE_GET_CLASS(cs);
555         int offset;
556 
557         if ((index % smt) != 0) {
558             continue;
559         }
560 
561         nodename = g_strdup_printf("%s@%x", dc->fw_name, index);
562         offset = fdt_add_subnode(fdt, cpus_offset, nodename);
563         g_free(nodename);
564         _FDT(offset);
565         spapr_populate_cpu_dt(cs, fdt, offset, spapr);
566     }
567 
568 }
569 
570 /*
571  * Adds ibm,dynamic-reconfiguration-memory node.
572  * Refer to docs/specs/ppc-spapr-hotplug.txt for the documentation
573  * of this device tree node.
574  */
575 static int spapr_populate_drconf_memory(sPAPRMachineState *spapr, void *fdt)
576 {
577     MachineState *machine = MACHINE(spapr);
578     int ret, i, offset;
579     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
580     uint32_t prop_lmb_size[] = {0, cpu_to_be32(lmb_size)};
581     uint32_t hotplug_lmb_start = spapr->hotplug_memory.base / lmb_size;
582     uint32_t nr_lmbs = (spapr->hotplug_memory.base +
583                        memory_region_size(&spapr->hotplug_memory.mr)) /
584                        lmb_size;
585     uint32_t *int_buf, *cur_index, buf_len;
586     int nr_nodes = nb_numa_nodes ? nb_numa_nodes : 1;
587 
588     /*
589      * Don't create the node if there is no hotpluggable memory
590      */
591     if (machine->ram_size == machine->maxram_size) {
592         return 0;
593     }
594 
595     /*
596      * Allocate enough buffer size to fit in ibm,dynamic-memory
597      * or ibm,associativity-lookup-arrays
598      */
599     buf_len = MAX(nr_lmbs * SPAPR_DR_LMB_LIST_ENTRY_SIZE + 1, nr_nodes * 4 + 2)
600               * sizeof(uint32_t);
601     cur_index = int_buf = g_malloc0(buf_len);
602 
603     offset = fdt_add_subnode(fdt, 0, "ibm,dynamic-reconfiguration-memory");
604 
605     ret = fdt_setprop(fdt, offset, "ibm,lmb-size", prop_lmb_size,
606                     sizeof(prop_lmb_size));
607     if (ret < 0) {
608         goto out;
609     }
610 
611     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-flags-mask", 0xff);
612     if (ret < 0) {
613         goto out;
614     }
615 
616     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-preservation-time", 0x0);
617     if (ret < 0) {
618         goto out;
619     }
620 
621     /* ibm,dynamic-memory */
622     int_buf[0] = cpu_to_be32(nr_lmbs);
623     cur_index++;
624     for (i = 0; i < nr_lmbs; i++) {
625         uint64_t addr = i * lmb_size;
626         uint32_t *dynamic_memory = cur_index;
627 
628         if (i >= hotplug_lmb_start) {
629             sPAPRDRConnector *drc;
630             sPAPRDRConnectorClass *drck;
631 
632             drc = spapr_dr_connector_by_id(SPAPR_DR_CONNECTOR_TYPE_LMB, i);
633             g_assert(drc);
634             drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
635 
636             dynamic_memory[0] = cpu_to_be32(addr >> 32);
637             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
638             dynamic_memory[2] = cpu_to_be32(drck->get_index(drc));
639             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
640             dynamic_memory[4] = cpu_to_be32(numa_get_node(addr, NULL));
641             if (memory_region_present(get_system_memory(), addr)) {
642                 dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_ASSIGNED);
643             } else {
644                 dynamic_memory[5] = cpu_to_be32(0);
645             }
646         } else {
647             /*
648              * LMB information for RMA, boot time RAM and gap b/n RAM and
649              * hotplug memory region -- all these are marked as reserved
650              * and as having no valid DRC.
651              */
652             dynamic_memory[0] = cpu_to_be32(addr >> 32);
653             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
654             dynamic_memory[2] = cpu_to_be32(0);
655             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
656             dynamic_memory[4] = cpu_to_be32(-1);
657             dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_RESERVED |
658                                             SPAPR_LMB_FLAGS_DRC_INVALID);
659         }
660 
661         cur_index += SPAPR_DR_LMB_LIST_ENTRY_SIZE;
662     }
663     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory", int_buf, buf_len);
664     if (ret < 0) {
665         goto out;
666     }
667 
668     /* ibm,associativity-lookup-arrays */
669     cur_index = int_buf;
670     int_buf[0] = cpu_to_be32(nr_nodes);
671     int_buf[1] = cpu_to_be32(4); /* Number of entries per associativity list */
672     cur_index += 2;
673     for (i = 0; i < nr_nodes; i++) {
674         uint32_t associativity[] = {
675             cpu_to_be32(0x0),
676             cpu_to_be32(0x0),
677             cpu_to_be32(0x0),
678             cpu_to_be32(i)
679         };
680         memcpy(cur_index, associativity, sizeof(associativity));
681         cur_index += 4;
682     }
683     ret = fdt_setprop(fdt, offset, "ibm,associativity-lookup-arrays", int_buf,
684             (cur_index - int_buf) * sizeof(uint32_t));
685 out:
686     g_free(int_buf);
687     return ret;
688 }
689 
690 static int spapr_dt_cas_updates(sPAPRMachineState *spapr, void *fdt,
691                                 sPAPROptionVector *ov5_updates)
692 {
693     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
694     int ret = 0, offset;
695 
696     /* Generate ibm,dynamic-reconfiguration-memory node if required */
697     if (spapr_ovec_test(ov5_updates, OV5_DRCONF_MEMORY)) {
698         g_assert(smc->dr_lmb_enabled);
699         ret = spapr_populate_drconf_memory(spapr, fdt);
700         if (ret) {
701             goto out;
702         }
703     }
704 
705     offset = fdt_path_offset(fdt, "/chosen");
706     if (offset < 0) {
707         offset = fdt_add_subnode(fdt, 0, "chosen");
708         if (offset < 0) {
709             return offset;
710         }
711     }
712     ret = spapr_ovec_populate_dt(fdt, offset, spapr->ov5_cas,
713                                  "ibm,architecture-vec-5");
714 
715 out:
716     return ret;
717 }
718 
719 int spapr_h_cas_compose_response(sPAPRMachineState *spapr,
720                                  target_ulong addr, target_ulong size,
721                                  sPAPROptionVector *ov5_updates)
722 {
723     void *fdt, *fdt_skel;
724     sPAPRDeviceTreeUpdateHeader hdr = { .version_id = 1 };
725 
726     size -= sizeof(hdr);
727 
728     /* Create sceleton */
729     fdt_skel = g_malloc0(size);
730     _FDT((fdt_create(fdt_skel, size)));
731     _FDT((fdt_begin_node(fdt_skel, "")));
732     _FDT((fdt_end_node(fdt_skel)));
733     _FDT((fdt_finish(fdt_skel)));
734     fdt = g_malloc0(size);
735     _FDT((fdt_open_into(fdt_skel, fdt, size)));
736     g_free(fdt_skel);
737 
738     /* Fixup cpu nodes */
739     _FDT((spapr_fixup_cpu_dt(fdt, spapr)));
740 
741     if (spapr_dt_cas_updates(spapr, fdt, ov5_updates)) {
742         return -1;
743     }
744 
745     /* Pack resulting tree */
746     _FDT((fdt_pack(fdt)));
747 
748     if (fdt_totalsize(fdt) + sizeof(hdr) > size) {
749         trace_spapr_cas_failed(size);
750         return -1;
751     }
752 
753     cpu_physical_memory_write(addr, &hdr, sizeof(hdr));
754     cpu_physical_memory_write(addr + sizeof(hdr), fdt, fdt_totalsize(fdt));
755     trace_spapr_cas_continue(fdt_totalsize(fdt) + sizeof(hdr));
756     g_free(fdt);
757 
758     return 0;
759 }
760 
761 static void spapr_dt_rtas(sPAPRMachineState *spapr, void *fdt)
762 {
763     int rtas;
764     GString *hypertas = g_string_sized_new(256);
765     GString *qemu_hypertas = g_string_sized_new(256);
766     uint32_t refpoints[] = { cpu_to_be32(0x4), cpu_to_be32(0x4) };
767     uint64_t max_hotplug_addr = spapr->hotplug_memory.base +
768         memory_region_size(&spapr->hotplug_memory.mr);
769     uint32_t lrdr_capacity[] = {
770         cpu_to_be32(max_hotplug_addr >> 32),
771         cpu_to_be32(max_hotplug_addr & 0xffffffff),
772         0, cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE),
773         cpu_to_be32(max_cpus / smp_threads),
774     };
775 
776     _FDT(rtas = fdt_add_subnode(fdt, 0, "rtas"));
777 
778     /* hypertas */
779     add_str(hypertas, "hcall-pft");
780     add_str(hypertas, "hcall-term");
781     add_str(hypertas, "hcall-dabr");
782     add_str(hypertas, "hcall-interrupt");
783     add_str(hypertas, "hcall-tce");
784     add_str(hypertas, "hcall-vio");
785     add_str(hypertas, "hcall-splpar");
786     add_str(hypertas, "hcall-bulk");
787     add_str(hypertas, "hcall-set-mode");
788     add_str(hypertas, "hcall-sprg0");
789     add_str(hypertas, "hcall-copy");
790     add_str(hypertas, "hcall-debug");
791     add_str(qemu_hypertas, "hcall-memop1");
792 
793     if (!kvm_enabled() || kvmppc_spapr_use_multitce()) {
794         add_str(hypertas, "hcall-multi-tce");
795     }
796     _FDT(fdt_setprop(fdt, rtas, "ibm,hypertas-functions",
797                      hypertas->str, hypertas->len));
798     g_string_free(hypertas, TRUE);
799     _FDT(fdt_setprop(fdt, rtas, "qemu,hypertas-functions",
800                      qemu_hypertas->str, qemu_hypertas->len));
801     g_string_free(qemu_hypertas, TRUE);
802 
803     _FDT(fdt_setprop(fdt, rtas, "ibm,associativity-reference-points",
804                      refpoints, sizeof(refpoints)));
805 
806     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-error-log-max",
807                           RTAS_ERROR_LOG_MAX));
808     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-event-scan-rate",
809                           RTAS_EVENT_SCAN_RATE));
810 
811     if (msi_nonbroken) {
812         _FDT(fdt_setprop(fdt, rtas, "ibm,change-msix-capable", NULL, 0));
813     }
814 
815     /*
816      * According to PAPR, rtas ibm,os-term does not guarantee a return
817      * back to the guest cpu.
818      *
819      * While an additional ibm,extended-os-term property indicates
820      * that rtas call return will always occur. Set this property.
821      */
822     _FDT(fdt_setprop(fdt, rtas, "ibm,extended-os-term", NULL, 0));
823 
824     _FDT(fdt_setprop(fdt, rtas, "ibm,lrdr-capacity",
825                      lrdr_capacity, sizeof(lrdr_capacity)));
826 
827     spapr_dt_rtas_tokens(fdt, rtas);
828 }
829 
830 static void spapr_dt_chosen(sPAPRMachineState *spapr, void *fdt)
831 {
832     MachineState *machine = MACHINE(spapr);
833     int chosen;
834     const char *boot_device = machine->boot_order;
835     char *stdout_path = spapr_vio_stdout_path(spapr->vio_bus);
836     size_t cb = 0;
837     char *bootlist = get_boot_devices_list(&cb, true);
838 
839     _FDT(chosen = fdt_add_subnode(fdt, 0, "chosen"));
840 
841     _FDT(fdt_setprop_string(fdt, chosen, "bootargs", machine->kernel_cmdline));
842     _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-start",
843                           spapr->initrd_base));
844     _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-end",
845                           spapr->initrd_base + spapr->initrd_size));
846 
847     if (spapr->kernel_size) {
848         uint64_t kprop[2] = { cpu_to_be64(KERNEL_LOAD_ADDR),
849                               cpu_to_be64(spapr->kernel_size) };
850 
851         _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel",
852                          &kprop, sizeof(kprop)));
853         if (spapr->kernel_le) {
854             _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel-le", NULL, 0));
855         }
856     }
857     if (boot_menu) {
858         _FDT((fdt_setprop_cell(fdt, chosen, "qemu,boot-menu", boot_menu)));
859     }
860     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-width", graphic_width));
861     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-height", graphic_height));
862     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-depth", graphic_depth));
863 
864     if (cb && bootlist) {
865         int i;
866 
867         for (i = 0; i < cb; i++) {
868             if (bootlist[i] == '\n') {
869                 bootlist[i] = ' ';
870             }
871         }
872         _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-list", bootlist));
873     }
874 
875     if (boot_device && strlen(boot_device)) {
876         _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-device", boot_device));
877     }
878 
879     if (!spapr->has_graphics && stdout_path) {
880         _FDT(fdt_setprop_string(fdt, chosen, "linux,stdout-path", stdout_path));
881     }
882 
883     g_free(stdout_path);
884     g_free(bootlist);
885 }
886 
887 static void spapr_dt_hypervisor(sPAPRMachineState *spapr, void *fdt)
888 {
889     /* The /hypervisor node isn't in PAPR - this is a hack to allow PR
890      * KVM to work under pHyp with some guest co-operation */
891     int hypervisor;
892     uint8_t hypercall[16];
893 
894     _FDT(hypervisor = fdt_add_subnode(fdt, 0, "hypervisor"));
895     /* indicate KVM hypercall interface */
896     _FDT(fdt_setprop_string(fdt, hypervisor, "compatible", "linux,kvm"));
897     if (kvmppc_has_cap_fixup_hcalls()) {
898         /*
899          * Older KVM versions with older guest kernels were broken
900          * with the magic page, don't allow the guest to map it.
901          */
902         if (!kvmppc_get_hypercall(first_cpu->env_ptr, hypercall,
903                                   sizeof(hypercall))) {
904             _FDT(fdt_setprop(fdt, hypervisor, "hcall-instructions",
905                              hypercall, sizeof(hypercall)));
906         }
907     }
908 }
909 
910 static void *spapr_build_fdt(sPAPRMachineState *spapr,
911                              hwaddr rtas_addr,
912                              hwaddr rtas_size)
913 {
914     MachineState *machine = MACHINE(qdev_get_machine());
915     MachineClass *mc = MACHINE_GET_CLASS(machine);
916     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
917     int ret;
918     void *fdt;
919     sPAPRPHBState *phb;
920     char *buf;
921 
922     fdt = g_malloc0(FDT_MAX_SIZE);
923     _FDT((fdt_create_empty_tree(fdt, FDT_MAX_SIZE)));
924 
925     /* Root node */
926     _FDT(fdt_setprop_string(fdt, 0, "device_type", "chrp"));
927     _FDT(fdt_setprop_string(fdt, 0, "model", "IBM pSeries (emulated by qemu)"));
928     _FDT(fdt_setprop_string(fdt, 0, "compatible", "qemu,pseries"));
929 
930     /*
931      * Add info to guest to indentify which host is it being run on
932      * and what is the uuid of the guest
933      */
934     if (kvmppc_get_host_model(&buf)) {
935         _FDT(fdt_setprop_string(fdt, 0, "host-model", buf));
936         g_free(buf);
937     }
938     if (kvmppc_get_host_serial(&buf)) {
939         _FDT(fdt_setprop_string(fdt, 0, "host-serial", buf));
940         g_free(buf);
941     }
942 
943     buf = qemu_uuid_unparse_strdup(&qemu_uuid);
944 
945     _FDT(fdt_setprop_string(fdt, 0, "vm,uuid", buf));
946     if (qemu_uuid_set) {
947         _FDT(fdt_setprop_string(fdt, 0, "system-id", buf));
948     }
949     g_free(buf);
950 
951     if (qemu_get_vm_name()) {
952         _FDT(fdt_setprop_string(fdt, 0, "ibm,partition-name",
953                                 qemu_get_vm_name()));
954     }
955 
956     _FDT(fdt_setprop_cell(fdt, 0, "#address-cells", 2));
957     _FDT(fdt_setprop_cell(fdt, 0, "#size-cells", 2));
958 
959     /* /interrupt controller */
960     spapr_dt_xics(spapr->nr_servers, fdt, PHANDLE_XICP);
961 
962     ret = spapr_populate_memory(spapr, fdt);
963     if (ret < 0) {
964         error_report("couldn't setup memory nodes in fdt");
965         exit(1);
966     }
967 
968     /* /vdevice */
969     spapr_dt_vdevice(spapr->vio_bus, fdt);
970 
971     if (object_resolve_path_type("", TYPE_SPAPR_RNG, NULL)) {
972         ret = spapr_rng_populate_dt(fdt);
973         if (ret < 0) {
974             error_report("could not set up rng device in the fdt");
975             exit(1);
976         }
977     }
978 
979     QLIST_FOREACH(phb, &spapr->phbs, list) {
980         ret = spapr_populate_pci_dt(phb, PHANDLE_XICP, fdt);
981         if (ret < 0) {
982             error_report("couldn't setup PCI devices in fdt");
983             exit(1);
984         }
985     }
986 
987     /* cpus */
988     spapr_populate_cpus_dt_node(fdt, spapr);
989 
990     if (smc->dr_lmb_enabled) {
991         _FDT(spapr_drc_populate_dt(fdt, 0, NULL, SPAPR_DR_CONNECTOR_TYPE_LMB));
992     }
993 
994     if (mc->has_hotpluggable_cpus) {
995         int offset = fdt_path_offset(fdt, "/cpus");
996         ret = spapr_drc_populate_dt(fdt, offset, NULL,
997                                     SPAPR_DR_CONNECTOR_TYPE_CPU);
998         if (ret < 0) {
999             error_report("Couldn't set up CPU DR device tree properties");
1000             exit(1);
1001         }
1002     }
1003 
1004     /* /event-sources */
1005     spapr_dt_events(spapr, fdt);
1006 
1007     /* /rtas */
1008     spapr_dt_rtas(spapr, fdt);
1009 
1010     /* /chosen */
1011     spapr_dt_chosen(spapr, fdt);
1012 
1013     /* /hypervisor */
1014     if (kvm_enabled()) {
1015         spapr_dt_hypervisor(spapr, fdt);
1016     }
1017 
1018     /* Build memory reserve map */
1019     if (spapr->kernel_size) {
1020         _FDT((fdt_add_mem_rsv(fdt, KERNEL_LOAD_ADDR, spapr->kernel_size)));
1021     }
1022     if (spapr->initrd_size) {
1023         _FDT((fdt_add_mem_rsv(fdt, spapr->initrd_base, spapr->initrd_size)));
1024     }
1025 
1026     /* ibm,client-architecture-support updates */
1027     ret = spapr_dt_cas_updates(spapr, fdt, spapr->ov5_cas);
1028     if (ret < 0) {
1029         error_report("couldn't setup CAS properties fdt");
1030         exit(1);
1031     }
1032 
1033     return fdt;
1034 }
1035 
1036 static uint64_t translate_kernel_address(void *opaque, uint64_t addr)
1037 {
1038     return (addr & 0x0fffffff) + KERNEL_LOAD_ADDR;
1039 }
1040 
1041 static void emulate_spapr_hypercall(PPCVirtualHypervisor *vhyp,
1042                                     PowerPCCPU *cpu)
1043 {
1044     CPUPPCState *env = &cpu->env;
1045 
1046     /* The TCG path should also be holding the BQL at this point */
1047     g_assert(qemu_mutex_iothread_locked());
1048 
1049     if (msr_pr) {
1050         hcall_dprintf("Hypercall made with MSR[PR]=1\n");
1051         env->gpr[3] = H_PRIVILEGE;
1052     } else {
1053         env->gpr[3] = spapr_hypercall(cpu, env->gpr[3], &env->gpr[4]);
1054     }
1055 }
1056 
1057 #define HPTE(_table, _i)   (void *)(((uint64_t *)(_table)) + ((_i) * 2))
1058 #define HPTE_VALID(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_VALID)
1059 #define HPTE_DIRTY(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_HPTE_DIRTY)
1060 #define CLEAN_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) &= tswap64(~HPTE64_V_HPTE_DIRTY))
1061 #define DIRTY_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) |= tswap64(HPTE64_V_HPTE_DIRTY))
1062 
1063 /*
1064  * Get the fd to access the kernel htab, re-opening it if necessary
1065  */
1066 static int get_htab_fd(sPAPRMachineState *spapr)
1067 {
1068     if (spapr->htab_fd >= 0) {
1069         return spapr->htab_fd;
1070     }
1071 
1072     spapr->htab_fd = kvmppc_get_htab_fd(false);
1073     if (spapr->htab_fd < 0) {
1074         error_report("Unable to open fd for reading hash table from KVM: %s",
1075                      strerror(errno));
1076     }
1077 
1078     return spapr->htab_fd;
1079 }
1080 
1081 static void close_htab_fd(sPAPRMachineState *spapr)
1082 {
1083     if (spapr->htab_fd >= 0) {
1084         close(spapr->htab_fd);
1085     }
1086     spapr->htab_fd = -1;
1087 }
1088 
1089 static hwaddr spapr_hpt_mask(PPCVirtualHypervisor *vhyp)
1090 {
1091     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1092 
1093     return HTAB_SIZE(spapr) / HASH_PTEG_SIZE_64 - 1;
1094 }
1095 
1096 static const ppc_hash_pte64_t *spapr_map_hptes(PPCVirtualHypervisor *vhyp,
1097                                                 hwaddr ptex, int n)
1098 {
1099     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1100     hwaddr pte_offset = ptex * HASH_PTE_SIZE_64;
1101 
1102     if (!spapr->htab) {
1103         /*
1104          * HTAB is controlled by KVM. Fetch into temporary buffer
1105          */
1106         ppc_hash_pte64_t *hptes = g_malloc(n * HASH_PTE_SIZE_64);
1107         kvmppc_read_hptes(hptes, ptex, n);
1108         return hptes;
1109     }
1110 
1111     /*
1112      * HTAB is controlled by QEMU. Just point to the internally
1113      * accessible PTEG.
1114      */
1115     return (const ppc_hash_pte64_t *)(spapr->htab + pte_offset);
1116 }
1117 
1118 static void spapr_unmap_hptes(PPCVirtualHypervisor *vhyp,
1119                               const ppc_hash_pte64_t *hptes,
1120                               hwaddr ptex, int n)
1121 {
1122     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1123 
1124     if (!spapr->htab) {
1125         g_free((void *)hptes);
1126     }
1127 
1128     /* Nothing to do for qemu managed HPT */
1129 }
1130 
1131 static void spapr_store_hpte(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1132                              uint64_t pte0, uint64_t pte1)
1133 {
1134     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1135     hwaddr offset = ptex * HASH_PTE_SIZE_64;
1136 
1137     if (!spapr->htab) {
1138         kvmppc_write_hpte(ptex, pte0, pte1);
1139     } else {
1140         stq_p(spapr->htab + offset, pte0);
1141         stq_p(spapr->htab + offset + HASH_PTE_SIZE_64 / 2, pte1);
1142     }
1143 }
1144 
1145 static int spapr_hpt_shift_for_ramsize(uint64_t ramsize)
1146 {
1147     int shift;
1148 
1149     /* We aim for a hash table of size 1/128 the size of RAM (rounded
1150      * up).  The PAPR recommendation is actually 1/64 of RAM size, but
1151      * that's much more than is needed for Linux guests */
1152     shift = ctz64(pow2ceil(ramsize)) - 7;
1153     shift = MAX(shift, 18); /* Minimum architected size */
1154     shift = MIN(shift, 46); /* Maximum architected size */
1155     return shift;
1156 }
1157 
1158 static void spapr_reallocate_hpt(sPAPRMachineState *spapr, int shift,
1159                                  Error **errp)
1160 {
1161     long rc;
1162 
1163     /* Clean up any HPT info from a previous boot */
1164     g_free(spapr->htab);
1165     spapr->htab = NULL;
1166     spapr->htab_shift = 0;
1167     close_htab_fd(spapr);
1168 
1169     rc = kvmppc_reset_htab(shift);
1170     if (rc < 0) {
1171         /* kernel-side HPT needed, but couldn't allocate one */
1172         error_setg_errno(errp, errno,
1173                          "Failed to allocate KVM HPT of order %d (try smaller maxmem?)",
1174                          shift);
1175         /* This is almost certainly fatal, but if the caller really
1176          * wants to carry on with shift == 0, it's welcome to try */
1177     } else if (rc > 0) {
1178         /* kernel-side HPT allocated */
1179         if (rc != shift) {
1180             error_setg(errp,
1181                        "Requested order %d HPT, but kernel allocated order %ld (try smaller maxmem?)",
1182                        shift, rc);
1183         }
1184 
1185         spapr->htab_shift = shift;
1186         spapr->htab = NULL;
1187     } else {
1188         /* kernel-side HPT not needed, allocate in userspace instead */
1189         size_t size = 1ULL << shift;
1190         int i;
1191 
1192         spapr->htab = qemu_memalign(size, size);
1193         if (!spapr->htab) {
1194             error_setg_errno(errp, errno,
1195                              "Could not allocate HPT of order %d", shift);
1196             return;
1197         }
1198 
1199         memset(spapr->htab, 0, size);
1200         spapr->htab_shift = shift;
1201 
1202         for (i = 0; i < size / HASH_PTE_SIZE_64; i++) {
1203             DIRTY_HPTE(HPTE(spapr->htab, i));
1204         }
1205     }
1206 }
1207 
1208 static void find_unknown_sysbus_device(SysBusDevice *sbdev, void *opaque)
1209 {
1210     bool matched = false;
1211 
1212     if (object_dynamic_cast(OBJECT(sbdev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
1213         matched = true;
1214     }
1215 
1216     if (!matched) {
1217         error_report("Device %s is not supported by this machine yet.",
1218                      qdev_fw_name(DEVICE(sbdev)));
1219         exit(1);
1220     }
1221 }
1222 
1223 static void ppc_spapr_reset(void)
1224 {
1225     MachineState *machine = MACHINE(qdev_get_machine());
1226     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
1227     PowerPCCPU *first_ppc_cpu;
1228     uint32_t rtas_limit;
1229     hwaddr rtas_addr, fdt_addr;
1230     void *fdt;
1231     int rc;
1232 
1233     /* Check for unknown sysbus devices */
1234     foreach_dynamic_sysbus_device(find_unknown_sysbus_device, NULL);
1235 
1236     /* Allocate and/or reset the hash page table */
1237     spapr_reallocate_hpt(spapr,
1238                          spapr_hpt_shift_for_ramsize(machine->maxram_size),
1239                          &error_fatal);
1240 
1241     /* Update the RMA size if necessary */
1242     if (spapr->vrma_adjust) {
1243         spapr->rma_size = kvmppc_rma_size(spapr_node0_size(),
1244                                           spapr->htab_shift);
1245     }
1246 
1247     qemu_devices_reset();
1248 
1249     /*
1250      * We place the device tree and RTAS just below either the top of the RMA,
1251      * or just below 2GB, whichever is lowere, so that it can be
1252      * processed with 32-bit real mode code if necessary
1253      */
1254     rtas_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR);
1255     rtas_addr = rtas_limit - RTAS_MAX_SIZE;
1256     fdt_addr = rtas_addr - FDT_MAX_SIZE;
1257 
1258     /* if this reset wasn't generated by CAS, we should reset our
1259      * negotiated options and start from scratch */
1260     if (!spapr->cas_reboot) {
1261         spapr_ovec_cleanup(spapr->ov5_cas);
1262         spapr->ov5_cas = spapr_ovec_new();
1263     }
1264 
1265     fdt = spapr_build_fdt(spapr, rtas_addr, spapr->rtas_size);
1266 
1267     spapr_load_rtas(spapr, fdt, rtas_addr);
1268 
1269     rc = fdt_pack(fdt);
1270 
1271     /* Should only fail if we've built a corrupted tree */
1272     assert(rc == 0);
1273 
1274     if (fdt_totalsize(fdt) > FDT_MAX_SIZE) {
1275         error_report("FDT too big ! 0x%x bytes (max is 0x%x)",
1276                      fdt_totalsize(fdt), FDT_MAX_SIZE);
1277         exit(1);
1278     }
1279 
1280     /* Load the fdt */
1281     qemu_fdt_dumpdtb(fdt, fdt_totalsize(fdt));
1282     cpu_physical_memory_write(fdt_addr, fdt, fdt_totalsize(fdt));
1283     g_free(fdt);
1284 
1285     /* Set up the entry state */
1286     first_ppc_cpu = POWERPC_CPU(first_cpu);
1287     first_ppc_cpu->env.gpr[3] = fdt_addr;
1288     first_ppc_cpu->env.gpr[5] = 0;
1289     first_cpu->halted = 0;
1290     first_ppc_cpu->env.nip = SPAPR_ENTRY_POINT;
1291 
1292     spapr->cas_reboot = false;
1293 }
1294 
1295 static void spapr_create_nvram(sPAPRMachineState *spapr)
1296 {
1297     DeviceState *dev = qdev_create(&spapr->vio_bus->bus, "spapr-nvram");
1298     DriveInfo *dinfo = drive_get(IF_PFLASH, 0, 0);
1299 
1300     if (dinfo) {
1301         qdev_prop_set_drive(dev, "drive", blk_by_legacy_dinfo(dinfo),
1302                             &error_fatal);
1303     }
1304 
1305     qdev_init_nofail(dev);
1306 
1307     spapr->nvram = (struct sPAPRNVRAM *)dev;
1308 }
1309 
1310 static void spapr_rtc_create(sPAPRMachineState *spapr)
1311 {
1312     DeviceState *dev = qdev_create(NULL, TYPE_SPAPR_RTC);
1313 
1314     qdev_init_nofail(dev);
1315     spapr->rtc = dev;
1316 
1317     object_property_add_alias(qdev_get_machine(), "rtc-time",
1318                               OBJECT(spapr->rtc), "date", NULL);
1319 }
1320 
1321 /* Returns whether we want to use VGA or not */
1322 static bool spapr_vga_init(PCIBus *pci_bus, Error **errp)
1323 {
1324     switch (vga_interface_type) {
1325     case VGA_NONE:
1326         return false;
1327     case VGA_DEVICE:
1328         return true;
1329     case VGA_STD:
1330     case VGA_VIRTIO:
1331         return pci_vga_init(pci_bus) != NULL;
1332     default:
1333         error_setg(errp,
1334                    "Unsupported VGA mode, only -vga std or -vga virtio is supported");
1335         return false;
1336     }
1337 }
1338 
1339 static int spapr_post_load(void *opaque, int version_id)
1340 {
1341     sPAPRMachineState *spapr = (sPAPRMachineState *)opaque;
1342     int err = 0;
1343 
1344     if (!object_dynamic_cast(OBJECT(spapr->ics), TYPE_ICS_KVM)) {
1345         int i;
1346         for (i = 0; i < spapr->nr_servers; i++) {
1347             icp_resend(&spapr->icps[i]);
1348         }
1349     }
1350 
1351     /* In earlier versions, there was no separate qdev for the PAPR
1352      * RTC, so the RTC offset was stored directly in sPAPREnvironment.
1353      * So when migrating from those versions, poke the incoming offset
1354      * value into the RTC device */
1355     if (version_id < 3) {
1356         err = spapr_rtc_import_offset(spapr->rtc, spapr->rtc_offset);
1357     }
1358 
1359     return err;
1360 }
1361 
1362 static bool version_before_3(void *opaque, int version_id)
1363 {
1364     return version_id < 3;
1365 }
1366 
1367 static bool spapr_ov5_cas_needed(void *opaque)
1368 {
1369     sPAPRMachineState *spapr = opaque;
1370     sPAPROptionVector *ov5_mask = spapr_ovec_new();
1371     sPAPROptionVector *ov5_legacy = spapr_ovec_new();
1372     sPAPROptionVector *ov5_removed = spapr_ovec_new();
1373     bool cas_needed;
1374 
1375     /* Prior to the introduction of sPAPROptionVector, we had two option
1376      * vectors we dealt with: OV5_FORM1_AFFINITY, and OV5_DRCONF_MEMORY.
1377      * Both of these options encode machine topology into the device-tree
1378      * in such a way that the now-booted OS should still be able to interact
1379      * appropriately with QEMU regardless of what options were actually
1380      * negotiatied on the source side.
1381      *
1382      * As such, we can avoid migrating the CAS-negotiated options if these
1383      * are the only options available on the current machine/platform.
1384      * Since these are the only options available for pseries-2.7 and
1385      * earlier, this allows us to maintain old->new/new->old migration
1386      * compatibility.
1387      *
1388      * For QEMU 2.8+, there are additional CAS-negotiatable options available
1389      * via default pseries-2.8 machines and explicit command-line parameters.
1390      * Some of these options, like OV5_HP_EVT, *do* require QEMU to be aware
1391      * of the actual CAS-negotiated values to continue working properly. For
1392      * example, availability of memory unplug depends on knowing whether
1393      * OV5_HP_EVT was negotiated via CAS.
1394      *
1395      * Thus, for any cases where the set of available CAS-negotiatable
1396      * options extends beyond OV5_FORM1_AFFINITY and OV5_DRCONF_MEMORY, we
1397      * include the CAS-negotiated options in the migration stream.
1398      */
1399     spapr_ovec_set(ov5_mask, OV5_FORM1_AFFINITY);
1400     spapr_ovec_set(ov5_mask, OV5_DRCONF_MEMORY);
1401 
1402     /* spapr_ovec_diff returns true if bits were removed. we avoid using
1403      * the mask itself since in the future it's possible "legacy" bits may be
1404      * removed via machine options, which could generate a false positive
1405      * that breaks migration.
1406      */
1407     spapr_ovec_intersect(ov5_legacy, spapr->ov5, ov5_mask);
1408     cas_needed = spapr_ovec_diff(ov5_removed, spapr->ov5, ov5_legacy);
1409 
1410     spapr_ovec_cleanup(ov5_mask);
1411     spapr_ovec_cleanup(ov5_legacy);
1412     spapr_ovec_cleanup(ov5_removed);
1413 
1414     return cas_needed;
1415 }
1416 
1417 static const VMStateDescription vmstate_spapr_ov5_cas = {
1418     .name = "spapr_option_vector_ov5_cas",
1419     .version_id = 1,
1420     .minimum_version_id = 1,
1421     .needed = spapr_ov5_cas_needed,
1422     .fields = (VMStateField[]) {
1423         VMSTATE_STRUCT_POINTER_V(ov5_cas, sPAPRMachineState, 1,
1424                                  vmstate_spapr_ovec, sPAPROptionVector),
1425         VMSTATE_END_OF_LIST()
1426     },
1427 };
1428 
1429 static const VMStateDescription vmstate_spapr = {
1430     .name = "spapr",
1431     .version_id = 3,
1432     .minimum_version_id = 1,
1433     .post_load = spapr_post_load,
1434     .fields = (VMStateField[]) {
1435         /* used to be @next_irq */
1436         VMSTATE_UNUSED_BUFFER(version_before_3, 0, 4),
1437 
1438         /* RTC offset */
1439         VMSTATE_UINT64_TEST(rtc_offset, sPAPRMachineState, version_before_3),
1440 
1441         VMSTATE_PPC_TIMEBASE_V(tb, sPAPRMachineState, 2),
1442         VMSTATE_END_OF_LIST()
1443     },
1444     .subsections = (const VMStateDescription*[]) {
1445         &vmstate_spapr_ov5_cas,
1446         NULL
1447     }
1448 };
1449 
1450 static int htab_save_setup(QEMUFile *f, void *opaque)
1451 {
1452     sPAPRMachineState *spapr = opaque;
1453 
1454     /* "Iteration" header */
1455     qemu_put_be32(f, spapr->htab_shift);
1456 
1457     if (spapr->htab) {
1458         spapr->htab_save_index = 0;
1459         spapr->htab_first_pass = true;
1460     } else {
1461         assert(kvm_enabled());
1462     }
1463 
1464 
1465     return 0;
1466 }
1467 
1468 static void htab_save_first_pass(QEMUFile *f, sPAPRMachineState *spapr,
1469                                  int64_t max_ns)
1470 {
1471     bool has_timeout = max_ns != -1;
1472     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
1473     int index = spapr->htab_save_index;
1474     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
1475 
1476     assert(spapr->htab_first_pass);
1477 
1478     do {
1479         int chunkstart;
1480 
1481         /* Consume invalid HPTEs */
1482         while ((index < htabslots)
1483                && !HPTE_VALID(HPTE(spapr->htab, index))) {
1484             index++;
1485             CLEAN_HPTE(HPTE(spapr->htab, index));
1486         }
1487 
1488         /* Consume valid HPTEs */
1489         chunkstart = index;
1490         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
1491                && HPTE_VALID(HPTE(spapr->htab, index))) {
1492             index++;
1493             CLEAN_HPTE(HPTE(spapr->htab, index));
1494         }
1495 
1496         if (index > chunkstart) {
1497             int n_valid = index - chunkstart;
1498 
1499             qemu_put_be32(f, chunkstart);
1500             qemu_put_be16(f, n_valid);
1501             qemu_put_be16(f, 0);
1502             qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
1503                             HASH_PTE_SIZE_64 * n_valid);
1504 
1505             if (has_timeout &&
1506                 (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
1507                 break;
1508             }
1509         }
1510     } while ((index < htabslots) && !qemu_file_rate_limit(f));
1511 
1512     if (index >= htabslots) {
1513         assert(index == htabslots);
1514         index = 0;
1515         spapr->htab_first_pass = false;
1516     }
1517     spapr->htab_save_index = index;
1518 }
1519 
1520 static int htab_save_later_pass(QEMUFile *f, sPAPRMachineState *spapr,
1521                                 int64_t max_ns)
1522 {
1523     bool final = max_ns < 0;
1524     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
1525     int examined = 0, sent = 0;
1526     int index = spapr->htab_save_index;
1527     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
1528 
1529     assert(!spapr->htab_first_pass);
1530 
1531     do {
1532         int chunkstart, invalidstart;
1533 
1534         /* Consume non-dirty HPTEs */
1535         while ((index < htabslots)
1536                && !HPTE_DIRTY(HPTE(spapr->htab, index))) {
1537             index++;
1538             examined++;
1539         }
1540 
1541         chunkstart = index;
1542         /* Consume valid dirty HPTEs */
1543         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
1544                && HPTE_DIRTY(HPTE(spapr->htab, index))
1545                && HPTE_VALID(HPTE(spapr->htab, index))) {
1546             CLEAN_HPTE(HPTE(spapr->htab, index));
1547             index++;
1548             examined++;
1549         }
1550 
1551         invalidstart = index;
1552         /* Consume invalid dirty HPTEs */
1553         while ((index < htabslots) && (index - invalidstart < USHRT_MAX)
1554                && HPTE_DIRTY(HPTE(spapr->htab, index))
1555                && !HPTE_VALID(HPTE(spapr->htab, index))) {
1556             CLEAN_HPTE(HPTE(spapr->htab, index));
1557             index++;
1558             examined++;
1559         }
1560 
1561         if (index > chunkstart) {
1562             int n_valid = invalidstart - chunkstart;
1563             int n_invalid = index - invalidstart;
1564 
1565             qemu_put_be32(f, chunkstart);
1566             qemu_put_be16(f, n_valid);
1567             qemu_put_be16(f, n_invalid);
1568             qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
1569                             HASH_PTE_SIZE_64 * n_valid);
1570             sent += index - chunkstart;
1571 
1572             if (!final && (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
1573                 break;
1574             }
1575         }
1576 
1577         if (examined >= htabslots) {
1578             break;
1579         }
1580 
1581         if (index >= htabslots) {
1582             assert(index == htabslots);
1583             index = 0;
1584         }
1585     } while ((examined < htabslots) && (!qemu_file_rate_limit(f) || final));
1586 
1587     if (index >= htabslots) {
1588         assert(index == htabslots);
1589         index = 0;
1590     }
1591 
1592     spapr->htab_save_index = index;
1593 
1594     return (examined >= htabslots) && (sent == 0) ? 1 : 0;
1595 }
1596 
1597 #define MAX_ITERATION_NS    5000000 /* 5 ms */
1598 #define MAX_KVM_BUF_SIZE    2048
1599 
1600 static int htab_save_iterate(QEMUFile *f, void *opaque)
1601 {
1602     sPAPRMachineState *spapr = opaque;
1603     int fd;
1604     int rc = 0;
1605 
1606     /* Iteration header */
1607     qemu_put_be32(f, 0);
1608 
1609     if (!spapr->htab) {
1610         assert(kvm_enabled());
1611 
1612         fd = get_htab_fd(spapr);
1613         if (fd < 0) {
1614             return fd;
1615         }
1616 
1617         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, MAX_ITERATION_NS);
1618         if (rc < 0) {
1619             return rc;
1620         }
1621     } else  if (spapr->htab_first_pass) {
1622         htab_save_first_pass(f, spapr, MAX_ITERATION_NS);
1623     } else {
1624         rc = htab_save_later_pass(f, spapr, MAX_ITERATION_NS);
1625     }
1626 
1627     /* End marker */
1628     qemu_put_be32(f, 0);
1629     qemu_put_be16(f, 0);
1630     qemu_put_be16(f, 0);
1631 
1632     return rc;
1633 }
1634 
1635 static int htab_save_complete(QEMUFile *f, void *opaque)
1636 {
1637     sPAPRMachineState *spapr = opaque;
1638     int fd;
1639 
1640     /* Iteration header */
1641     qemu_put_be32(f, 0);
1642 
1643     if (!spapr->htab) {
1644         int rc;
1645 
1646         assert(kvm_enabled());
1647 
1648         fd = get_htab_fd(spapr);
1649         if (fd < 0) {
1650             return fd;
1651         }
1652 
1653         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, -1);
1654         if (rc < 0) {
1655             return rc;
1656         }
1657     } else {
1658         if (spapr->htab_first_pass) {
1659             htab_save_first_pass(f, spapr, -1);
1660         }
1661         htab_save_later_pass(f, spapr, -1);
1662     }
1663 
1664     /* End marker */
1665     qemu_put_be32(f, 0);
1666     qemu_put_be16(f, 0);
1667     qemu_put_be16(f, 0);
1668 
1669     return 0;
1670 }
1671 
1672 static int htab_load(QEMUFile *f, void *opaque, int version_id)
1673 {
1674     sPAPRMachineState *spapr = opaque;
1675     uint32_t section_hdr;
1676     int fd = -1;
1677 
1678     if (version_id < 1 || version_id > 1) {
1679         error_report("htab_load() bad version");
1680         return -EINVAL;
1681     }
1682 
1683     section_hdr = qemu_get_be32(f);
1684 
1685     if (section_hdr) {
1686         Error *local_err = NULL;
1687 
1688         /* First section gives the htab size */
1689         spapr_reallocate_hpt(spapr, section_hdr, &local_err);
1690         if (local_err) {
1691             error_report_err(local_err);
1692             return -EINVAL;
1693         }
1694         return 0;
1695     }
1696 
1697     if (!spapr->htab) {
1698         assert(kvm_enabled());
1699 
1700         fd = kvmppc_get_htab_fd(true);
1701         if (fd < 0) {
1702             error_report("Unable to open fd to restore KVM hash table: %s",
1703                          strerror(errno));
1704         }
1705     }
1706 
1707     while (true) {
1708         uint32_t index;
1709         uint16_t n_valid, n_invalid;
1710 
1711         index = qemu_get_be32(f);
1712         n_valid = qemu_get_be16(f);
1713         n_invalid = qemu_get_be16(f);
1714 
1715         if ((index == 0) && (n_valid == 0) && (n_invalid == 0)) {
1716             /* End of Stream */
1717             break;
1718         }
1719 
1720         if ((index + n_valid + n_invalid) >
1721             (HTAB_SIZE(spapr) / HASH_PTE_SIZE_64)) {
1722             /* Bad index in stream */
1723             error_report(
1724                 "htab_load() bad index %d (%hd+%hd entries) in htab stream (htab_shift=%d)",
1725                 index, n_valid, n_invalid, spapr->htab_shift);
1726             return -EINVAL;
1727         }
1728 
1729         if (spapr->htab) {
1730             if (n_valid) {
1731                 qemu_get_buffer(f, HPTE(spapr->htab, index),
1732                                 HASH_PTE_SIZE_64 * n_valid);
1733             }
1734             if (n_invalid) {
1735                 memset(HPTE(spapr->htab, index + n_valid), 0,
1736                        HASH_PTE_SIZE_64 * n_invalid);
1737             }
1738         } else {
1739             int rc;
1740 
1741             assert(fd >= 0);
1742 
1743             rc = kvmppc_load_htab_chunk(f, fd, index, n_valid, n_invalid);
1744             if (rc < 0) {
1745                 return rc;
1746             }
1747         }
1748     }
1749 
1750     if (!spapr->htab) {
1751         assert(fd >= 0);
1752         close(fd);
1753     }
1754 
1755     return 0;
1756 }
1757 
1758 static void htab_cleanup(void *opaque)
1759 {
1760     sPAPRMachineState *spapr = opaque;
1761 
1762     close_htab_fd(spapr);
1763 }
1764 
1765 static SaveVMHandlers savevm_htab_handlers = {
1766     .save_live_setup = htab_save_setup,
1767     .save_live_iterate = htab_save_iterate,
1768     .save_live_complete_precopy = htab_save_complete,
1769     .cleanup = htab_cleanup,
1770     .load_state = htab_load,
1771 };
1772 
1773 static void spapr_boot_set(void *opaque, const char *boot_device,
1774                            Error **errp)
1775 {
1776     MachineState *machine = MACHINE(qdev_get_machine());
1777     machine->boot_order = g_strdup(boot_device);
1778 }
1779 
1780 /*
1781  * Reset routine for LMB DR devices.
1782  *
1783  * Unlike PCI DR devices, LMB DR devices explicitly register this reset
1784  * routine. Reset for PCI DR devices will be handled by PHB reset routine
1785  * when it walks all its children devices. LMB devices reset occurs
1786  * as part of spapr_ppc_reset().
1787  */
1788 static void spapr_drc_reset(void *opaque)
1789 {
1790     sPAPRDRConnector *drc = opaque;
1791     DeviceState *d = DEVICE(drc);
1792 
1793     if (d) {
1794         device_reset(d);
1795     }
1796 }
1797 
1798 static void spapr_create_lmb_dr_connectors(sPAPRMachineState *spapr)
1799 {
1800     MachineState *machine = MACHINE(spapr);
1801     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
1802     uint32_t nr_lmbs = (machine->maxram_size - machine->ram_size)/lmb_size;
1803     int i;
1804 
1805     for (i = 0; i < nr_lmbs; i++) {
1806         sPAPRDRConnector *drc;
1807         uint64_t addr;
1808 
1809         addr = i * lmb_size + spapr->hotplug_memory.base;
1810         drc = spapr_dr_connector_new(OBJECT(spapr), SPAPR_DR_CONNECTOR_TYPE_LMB,
1811                                      addr/lmb_size);
1812         qemu_register_reset(spapr_drc_reset, drc);
1813     }
1814 }
1815 
1816 /*
1817  * If RAM size, maxmem size and individual node mem sizes aren't aligned
1818  * to SPAPR_MEMORY_BLOCK_SIZE(256MB), then refuse to start the guest
1819  * since we can't support such unaligned sizes with DRCONF_MEMORY.
1820  */
1821 static void spapr_validate_node_memory(MachineState *machine, Error **errp)
1822 {
1823     int i;
1824 
1825     if (machine->ram_size % SPAPR_MEMORY_BLOCK_SIZE) {
1826         error_setg(errp, "Memory size 0x" RAM_ADDR_FMT
1827                    " is not aligned to %llu MiB",
1828                    machine->ram_size,
1829                    SPAPR_MEMORY_BLOCK_SIZE / M_BYTE);
1830         return;
1831     }
1832 
1833     if (machine->maxram_size % SPAPR_MEMORY_BLOCK_SIZE) {
1834         error_setg(errp, "Maximum memory size 0x" RAM_ADDR_FMT
1835                    " is not aligned to %llu MiB",
1836                    machine->ram_size,
1837                    SPAPR_MEMORY_BLOCK_SIZE / M_BYTE);
1838         return;
1839     }
1840 
1841     for (i = 0; i < nb_numa_nodes; i++) {
1842         if (numa_info[i].node_mem % SPAPR_MEMORY_BLOCK_SIZE) {
1843             error_setg(errp,
1844                        "Node %d memory size 0x%" PRIx64
1845                        " is not aligned to %llu MiB",
1846                        i, numa_info[i].node_mem,
1847                        SPAPR_MEMORY_BLOCK_SIZE / M_BYTE);
1848             return;
1849         }
1850     }
1851 }
1852 
1853 /* find cpu slot in machine->possible_cpus by core_id */
1854 static CPUArchId *spapr_find_cpu_slot(MachineState *ms, uint32_t id, int *idx)
1855 {
1856     int index = id / smp_threads;
1857 
1858     if (index >= ms->possible_cpus->len) {
1859         return NULL;
1860     }
1861     if (idx) {
1862         *idx = index;
1863     }
1864     return &ms->possible_cpus->cpus[index];
1865 }
1866 
1867 static void spapr_init_cpus(sPAPRMachineState *spapr)
1868 {
1869     MachineState *machine = MACHINE(spapr);
1870     MachineClass *mc = MACHINE_GET_CLASS(machine);
1871     char *type = spapr_get_cpu_core_type(machine->cpu_model);
1872     int smt = kvmppc_smt_threads();
1873     const CPUArchIdList *possible_cpus;
1874     int boot_cores_nr = smp_cpus / smp_threads;
1875     int i;
1876 
1877     if (!type) {
1878         error_report("Unable to find sPAPR CPU Core definition");
1879         exit(1);
1880     }
1881 
1882     possible_cpus = mc->possible_cpu_arch_ids(machine);
1883     if (mc->has_hotpluggable_cpus) {
1884         if (smp_cpus % smp_threads) {
1885             error_report("smp_cpus (%u) must be multiple of threads (%u)",
1886                          smp_cpus, smp_threads);
1887             exit(1);
1888         }
1889         if (max_cpus % smp_threads) {
1890             error_report("max_cpus (%u) must be multiple of threads (%u)",
1891                          max_cpus, smp_threads);
1892             exit(1);
1893         }
1894     } else {
1895         if (max_cpus != smp_cpus) {
1896             error_report("This machine version does not support CPU hotplug");
1897             exit(1);
1898         }
1899         boot_cores_nr = possible_cpus->len;
1900     }
1901 
1902     for (i = 0; i < possible_cpus->len; i++) {
1903         int core_id = i * smp_threads;
1904 
1905         if (mc->has_hotpluggable_cpus) {
1906             sPAPRDRConnector *drc =
1907                 spapr_dr_connector_new(OBJECT(spapr),
1908                                        SPAPR_DR_CONNECTOR_TYPE_CPU,
1909                                        (core_id / smp_threads) * smt);
1910 
1911             qemu_register_reset(spapr_drc_reset, drc);
1912         }
1913 
1914         if (i < boot_cores_nr) {
1915             Object *core  = object_new(type);
1916             int nr_threads = smp_threads;
1917 
1918             /* Handle the partially filled core for older machine types */
1919             if ((i + 1) * smp_threads >= smp_cpus) {
1920                 nr_threads = smp_cpus - i * smp_threads;
1921             }
1922 
1923             object_property_set_int(core, nr_threads, "nr-threads",
1924                                     &error_fatal);
1925             object_property_set_int(core, core_id, CPU_CORE_PROP_CORE_ID,
1926                                     &error_fatal);
1927             object_property_set_bool(core, true, "realized", &error_fatal);
1928         }
1929     }
1930     g_free(type);
1931 }
1932 
1933 /* pSeries LPAR / sPAPR hardware init */
1934 static void ppc_spapr_init(MachineState *machine)
1935 {
1936     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
1937     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1938     const char *kernel_filename = machine->kernel_filename;
1939     const char *initrd_filename = machine->initrd_filename;
1940     PCIHostState *phb;
1941     int i;
1942     MemoryRegion *sysmem = get_system_memory();
1943     MemoryRegion *ram = g_new(MemoryRegion, 1);
1944     MemoryRegion *rma_region;
1945     void *rma = NULL;
1946     hwaddr rma_alloc_size;
1947     hwaddr node0_size = spapr_node0_size();
1948     long load_limit, fw_size;
1949     char *filename;
1950     int smt = kvmppc_smt_threads();
1951 
1952     msi_nonbroken = true;
1953 
1954     QLIST_INIT(&spapr->phbs);
1955 
1956     /* Allocate RMA if necessary */
1957     rma_alloc_size = kvmppc_alloc_rma(&rma);
1958 
1959     if (rma_alloc_size == -1) {
1960         error_report("Unable to create RMA");
1961         exit(1);
1962     }
1963 
1964     if (rma_alloc_size && (rma_alloc_size < node0_size)) {
1965         spapr->rma_size = rma_alloc_size;
1966     } else {
1967         spapr->rma_size = node0_size;
1968 
1969         /* With KVM, we don't actually know whether KVM supports an
1970          * unbounded RMA (PR KVM) or is limited by the hash table size
1971          * (HV KVM using VRMA), so we always assume the latter
1972          *
1973          * In that case, we also limit the initial allocations for RTAS
1974          * etc... to 256M since we have no way to know what the VRMA size
1975          * is going to be as it depends on the size of the hash table
1976          * isn't determined yet.
1977          */
1978         if (kvm_enabled()) {
1979             spapr->vrma_adjust = 1;
1980             spapr->rma_size = MIN(spapr->rma_size, 0x10000000);
1981         }
1982 
1983         /* Actually we don't support unbounded RMA anymore since we
1984          * added proper emulation of HV mode. The max we can get is
1985          * 16G which also happens to be what we configure for PAPR
1986          * mode so make sure we don't do anything bigger than that
1987          */
1988         spapr->rma_size = MIN(spapr->rma_size, 0x400000000ull);
1989     }
1990 
1991     if (spapr->rma_size > node0_size) {
1992         error_report("Numa node 0 has to span the RMA (%#08"HWADDR_PRIx")",
1993                      spapr->rma_size);
1994         exit(1);
1995     }
1996 
1997     /* Setup a load limit for the ramdisk leaving room for SLOF and FDT */
1998     load_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR) - FW_OVERHEAD;
1999 
2000     /* Set up Interrupt Controller before we create the VCPUs */
2001     xics_system_init(machine, DIV_ROUND_UP(max_cpus * smt, smp_threads),
2002                      XICS_IRQS_SPAPR, &error_fatal);
2003 
2004     /* Set up containers for ibm,client-set-architecture negotiated options */
2005     spapr->ov5 = spapr_ovec_new();
2006     spapr->ov5_cas = spapr_ovec_new();
2007 
2008     if (smc->dr_lmb_enabled) {
2009         spapr_ovec_set(spapr->ov5, OV5_DRCONF_MEMORY);
2010         spapr_validate_node_memory(machine, &error_fatal);
2011     }
2012 
2013     spapr_ovec_set(spapr->ov5, OV5_FORM1_AFFINITY);
2014 
2015     /* advertise support for dedicated HP event source to guests */
2016     if (spapr->use_hotplug_event_source) {
2017         spapr_ovec_set(spapr->ov5, OV5_HP_EVT);
2018     }
2019 
2020     /* init CPUs */
2021     if (machine->cpu_model == NULL) {
2022         machine->cpu_model = kvm_enabled() ? "host" : smc->tcg_default_cpu;
2023     }
2024 
2025     ppc_cpu_parse_features(machine->cpu_model);
2026 
2027     spapr_init_cpus(spapr);
2028 
2029     if (kvm_enabled()) {
2030         /* Enable H_LOGICAL_CI_* so SLOF can talk to in-kernel devices */
2031         kvmppc_enable_logical_ci_hcalls();
2032         kvmppc_enable_set_mode_hcall();
2033 
2034         /* H_CLEAR_MOD/_REF are mandatory in PAPR, but off by default */
2035         kvmppc_enable_clear_ref_mod_hcalls();
2036     }
2037 
2038     /* allocate RAM */
2039     memory_region_allocate_system_memory(ram, NULL, "ppc_spapr.ram",
2040                                          machine->ram_size);
2041     memory_region_add_subregion(sysmem, 0, ram);
2042 
2043     if (rma_alloc_size && rma) {
2044         rma_region = g_new(MemoryRegion, 1);
2045         memory_region_init_ram_ptr(rma_region, NULL, "ppc_spapr.rma",
2046                                    rma_alloc_size, rma);
2047         vmstate_register_ram_global(rma_region);
2048         memory_region_add_subregion(sysmem, 0, rma_region);
2049     }
2050 
2051     /* initialize hotplug memory address space */
2052     if (machine->ram_size < machine->maxram_size) {
2053         ram_addr_t hotplug_mem_size = machine->maxram_size - machine->ram_size;
2054         /*
2055          * Limit the number of hotpluggable memory slots to half the number
2056          * slots that KVM supports, leaving the other half for PCI and other
2057          * devices. However ensure that number of slots doesn't drop below 32.
2058          */
2059         int max_memslots = kvm_enabled() ? kvm_get_max_memslots() / 2 :
2060                            SPAPR_MAX_RAM_SLOTS;
2061 
2062         if (max_memslots < SPAPR_MAX_RAM_SLOTS) {
2063             max_memslots = SPAPR_MAX_RAM_SLOTS;
2064         }
2065         if (machine->ram_slots > max_memslots) {
2066             error_report("Specified number of memory slots %"
2067                          PRIu64" exceeds max supported %d",
2068                          machine->ram_slots, max_memslots);
2069             exit(1);
2070         }
2071 
2072         spapr->hotplug_memory.base = ROUND_UP(machine->ram_size,
2073                                               SPAPR_HOTPLUG_MEM_ALIGN);
2074         memory_region_init(&spapr->hotplug_memory.mr, OBJECT(spapr),
2075                            "hotplug-memory", hotplug_mem_size);
2076         memory_region_add_subregion(sysmem, spapr->hotplug_memory.base,
2077                                     &spapr->hotplug_memory.mr);
2078     }
2079 
2080     if (smc->dr_lmb_enabled) {
2081         spapr_create_lmb_dr_connectors(spapr);
2082     }
2083 
2084     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, "spapr-rtas.bin");
2085     if (!filename) {
2086         error_report("Could not find LPAR rtas '%s'", "spapr-rtas.bin");
2087         exit(1);
2088     }
2089     spapr->rtas_size = get_image_size(filename);
2090     if (spapr->rtas_size < 0) {
2091         error_report("Could not get size of LPAR rtas '%s'", filename);
2092         exit(1);
2093     }
2094     spapr->rtas_blob = g_malloc(spapr->rtas_size);
2095     if (load_image_size(filename, spapr->rtas_blob, spapr->rtas_size) < 0) {
2096         error_report("Could not load LPAR rtas '%s'", filename);
2097         exit(1);
2098     }
2099     if (spapr->rtas_size > RTAS_MAX_SIZE) {
2100         error_report("RTAS too big ! 0x%zx bytes (max is 0x%x)",
2101                      (size_t)spapr->rtas_size, RTAS_MAX_SIZE);
2102         exit(1);
2103     }
2104     g_free(filename);
2105 
2106     /* Set up RTAS event infrastructure */
2107     spapr_events_init(spapr);
2108 
2109     /* Set up the RTC RTAS interfaces */
2110     spapr_rtc_create(spapr);
2111 
2112     /* Set up VIO bus */
2113     spapr->vio_bus = spapr_vio_bus_init();
2114 
2115     for (i = 0; i < MAX_SERIAL_PORTS; i++) {
2116         if (serial_hds[i]) {
2117             spapr_vty_create(spapr->vio_bus, serial_hds[i]);
2118         }
2119     }
2120 
2121     /* We always have at least the nvram device on VIO */
2122     spapr_create_nvram(spapr);
2123 
2124     /* Set up PCI */
2125     spapr_pci_rtas_init();
2126 
2127     phb = spapr_create_phb(spapr, 0);
2128 
2129     for (i = 0; i < nb_nics; i++) {
2130         NICInfo *nd = &nd_table[i];
2131 
2132         if (!nd->model) {
2133             nd->model = g_strdup("ibmveth");
2134         }
2135 
2136         if (strcmp(nd->model, "ibmveth") == 0) {
2137             spapr_vlan_create(spapr->vio_bus, nd);
2138         } else {
2139             pci_nic_init_nofail(&nd_table[i], phb->bus, nd->model, NULL);
2140         }
2141     }
2142 
2143     for (i = 0; i <= drive_get_max_bus(IF_SCSI); i++) {
2144         spapr_vscsi_create(spapr->vio_bus);
2145     }
2146 
2147     /* Graphics */
2148     if (spapr_vga_init(phb->bus, &error_fatal)) {
2149         spapr->has_graphics = true;
2150         machine->usb |= defaults_enabled() && !machine->usb_disabled;
2151     }
2152 
2153     if (machine->usb) {
2154         if (smc->use_ohci_by_default) {
2155             pci_create_simple(phb->bus, -1, "pci-ohci");
2156         } else {
2157             pci_create_simple(phb->bus, -1, "nec-usb-xhci");
2158         }
2159 
2160         if (spapr->has_graphics) {
2161             USBBus *usb_bus = usb_bus_find(-1);
2162 
2163             usb_create_simple(usb_bus, "usb-kbd");
2164             usb_create_simple(usb_bus, "usb-mouse");
2165         }
2166     }
2167 
2168     if (spapr->rma_size < (MIN_RMA_SLOF << 20)) {
2169         error_report(
2170             "pSeries SLOF firmware requires >= %ldM guest RMA (Real Mode Area memory)",
2171             MIN_RMA_SLOF);
2172         exit(1);
2173     }
2174 
2175     if (kernel_filename) {
2176         uint64_t lowaddr = 0;
2177 
2178         spapr->kernel_size = load_elf(kernel_filename, translate_kernel_address,
2179                                       NULL, NULL, &lowaddr, NULL, 1,
2180                                       PPC_ELF_MACHINE, 0, 0);
2181         if (spapr->kernel_size == ELF_LOAD_WRONG_ENDIAN) {
2182             spapr->kernel_size = load_elf(kernel_filename,
2183                                           translate_kernel_address, NULL, NULL,
2184                                           &lowaddr, NULL, 0, PPC_ELF_MACHINE,
2185                                           0, 0);
2186             spapr->kernel_le = spapr->kernel_size > 0;
2187         }
2188         if (spapr->kernel_size < 0) {
2189             error_report("error loading %s: %s", kernel_filename,
2190                          load_elf_strerror(spapr->kernel_size));
2191             exit(1);
2192         }
2193 
2194         /* load initrd */
2195         if (initrd_filename) {
2196             /* Try to locate the initrd in the gap between the kernel
2197              * and the firmware. Add a bit of space just in case
2198              */
2199             spapr->initrd_base = (KERNEL_LOAD_ADDR + spapr->kernel_size
2200                                   + 0x1ffff) & ~0xffff;
2201             spapr->initrd_size = load_image_targphys(initrd_filename,
2202                                                      spapr->initrd_base,
2203                                                      load_limit
2204                                                      - spapr->initrd_base);
2205             if (spapr->initrd_size < 0) {
2206                 error_report("could not load initial ram disk '%s'",
2207                              initrd_filename);
2208                 exit(1);
2209             }
2210         }
2211     }
2212 
2213     if (bios_name == NULL) {
2214         bios_name = FW_FILE_NAME;
2215     }
2216     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
2217     if (!filename) {
2218         error_report("Could not find LPAR firmware '%s'", bios_name);
2219         exit(1);
2220     }
2221     fw_size = load_image_targphys(filename, 0, FW_MAX_SIZE);
2222     if (fw_size <= 0) {
2223         error_report("Could not load LPAR firmware '%s'", filename);
2224         exit(1);
2225     }
2226     g_free(filename);
2227 
2228     /* FIXME: Should register things through the MachineState's qdev
2229      * interface, this is a legacy from the sPAPREnvironment structure
2230      * which predated MachineState but had a similar function */
2231     vmstate_register(NULL, 0, &vmstate_spapr, spapr);
2232     register_savevm_live(NULL, "spapr/htab", -1, 1,
2233                          &savevm_htab_handlers, spapr);
2234 
2235     /* used by RTAS */
2236     QTAILQ_INIT(&spapr->ccs_list);
2237     qemu_register_reset(spapr_ccs_reset_hook, spapr);
2238 
2239     qemu_register_boot_set(spapr_boot_set, spapr);
2240 
2241     /* to stop and start vmclock */
2242     if (kvm_enabled()) {
2243         qemu_add_vm_change_state_handler(cpu_ppc_clock_vm_state_change,
2244                                          &spapr->tb);
2245     }
2246 }
2247 
2248 static int spapr_kvm_type(const char *vm_type)
2249 {
2250     if (!vm_type) {
2251         return 0;
2252     }
2253 
2254     if (!strcmp(vm_type, "HV")) {
2255         return 1;
2256     }
2257 
2258     if (!strcmp(vm_type, "PR")) {
2259         return 2;
2260     }
2261 
2262     error_report("Unknown kvm-type specified '%s'", vm_type);
2263     exit(1);
2264 }
2265 
2266 /*
2267  * Implementation of an interface to adjust firmware path
2268  * for the bootindex property handling.
2269  */
2270 static char *spapr_get_fw_dev_path(FWPathProvider *p, BusState *bus,
2271                                    DeviceState *dev)
2272 {
2273 #define CAST(type, obj, name) \
2274     ((type *)object_dynamic_cast(OBJECT(obj), (name)))
2275     SCSIDevice *d = CAST(SCSIDevice,  dev, TYPE_SCSI_DEVICE);
2276     sPAPRPHBState *phb = CAST(sPAPRPHBState, dev, TYPE_SPAPR_PCI_HOST_BRIDGE);
2277 
2278     if (d) {
2279         void *spapr = CAST(void, bus->parent, "spapr-vscsi");
2280         VirtIOSCSI *virtio = CAST(VirtIOSCSI, bus->parent, TYPE_VIRTIO_SCSI);
2281         USBDevice *usb = CAST(USBDevice, bus->parent, TYPE_USB_DEVICE);
2282 
2283         if (spapr) {
2284             /*
2285              * Replace "channel@0/disk@0,0" with "disk@8000000000000000":
2286              * We use SRP luns of the form 8000 | (bus << 8) | (id << 5) | lun
2287              * in the top 16 bits of the 64-bit LUN
2288              */
2289             unsigned id = 0x8000 | (d->id << 8) | d->lun;
2290             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
2291                                    (uint64_t)id << 48);
2292         } else if (virtio) {
2293             /*
2294              * We use SRP luns of the form 01000000 | (target << 8) | lun
2295              * in the top 32 bits of the 64-bit LUN
2296              * Note: the quote above is from SLOF and it is wrong,
2297              * the actual binding is:
2298              * swap 0100 or 10 << or 20 << ( target lun-id -- srplun )
2299              */
2300             unsigned id = 0x1000000 | (d->id << 16) | d->lun;
2301             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
2302                                    (uint64_t)id << 32);
2303         } else if (usb) {
2304             /*
2305              * We use SRP luns of the form 01000000 | (usb-port << 16) | lun
2306              * in the top 32 bits of the 64-bit LUN
2307              */
2308             unsigned usb_port = atoi(usb->port->path);
2309             unsigned id = 0x1000000 | (usb_port << 16) | d->lun;
2310             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
2311                                    (uint64_t)id << 32);
2312         }
2313     }
2314 
2315     /*
2316      * SLOF probes the USB devices, and if it recognizes that the device is a
2317      * storage device, it changes its name to "storage" instead of "usb-host",
2318      * and additionally adds a child node for the SCSI LUN, so the correct
2319      * boot path in SLOF is something like .../storage@1/disk@xxx" instead.
2320      */
2321     if (strcmp("usb-host", qdev_fw_name(dev)) == 0) {
2322         USBDevice *usbdev = CAST(USBDevice, dev, TYPE_USB_DEVICE);
2323         if (usb_host_dev_is_scsi_storage(usbdev)) {
2324             return g_strdup_printf("storage@%s/disk", usbdev->port->path);
2325         }
2326     }
2327 
2328     if (phb) {
2329         /* Replace "pci" with "pci@800000020000000" */
2330         return g_strdup_printf("pci@%"PRIX64, phb->buid);
2331     }
2332 
2333     return NULL;
2334 }
2335 
2336 static char *spapr_get_kvm_type(Object *obj, Error **errp)
2337 {
2338     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2339 
2340     return g_strdup(spapr->kvm_type);
2341 }
2342 
2343 static void spapr_set_kvm_type(Object *obj, const char *value, Error **errp)
2344 {
2345     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2346 
2347     g_free(spapr->kvm_type);
2348     spapr->kvm_type = g_strdup(value);
2349 }
2350 
2351 static bool spapr_get_modern_hotplug_events(Object *obj, Error **errp)
2352 {
2353     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2354 
2355     return spapr->use_hotplug_event_source;
2356 }
2357 
2358 static void spapr_set_modern_hotplug_events(Object *obj, bool value,
2359                                             Error **errp)
2360 {
2361     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2362 
2363     spapr->use_hotplug_event_source = value;
2364 }
2365 
2366 static void spapr_machine_initfn(Object *obj)
2367 {
2368     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2369 
2370     spapr->htab_fd = -1;
2371     spapr->use_hotplug_event_source = true;
2372     object_property_add_str(obj, "kvm-type",
2373                             spapr_get_kvm_type, spapr_set_kvm_type, NULL);
2374     object_property_set_description(obj, "kvm-type",
2375                                     "Specifies the KVM virtualization mode (HV, PR)",
2376                                     NULL);
2377     object_property_add_bool(obj, "modern-hotplug-events",
2378                             spapr_get_modern_hotplug_events,
2379                             spapr_set_modern_hotplug_events,
2380                             NULL);
2381     object_property_set_description(obj, "modern-hotplug-events",
2382                                     "Use dedicated hotplug event mechanism in"
2383                                     " place of standard EPOW events when possible"
2384                                     " (required for memory hot-unplug support)",
2385                                     NULL);
2386 }
2387 
2388 static void spapr_machine_finalizefn(Object *obj)
2389 {
2390     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2391 
2392     g_free(spapr->kvm_type);
2393 }
2394 
2395 void spapr_do_system_reset_on_cpu(CPUState *cs, run_on_cpu_data arg)
2396 {
2397     cpu_synchronize_state(cs);
2398     ppc_cpu_do_system_reset(cs);
2399 }
2400 
2401 static void spapr_nmi(NMIState *n, int cpu_index, Error **errp)
2402 {
2403     CPUState *cs;
2404 
2405     CPU_FOREACH(cs) {
2406         async_run_on_cpu(cs, spapr_do_system_reset_on_cpu, RUN_ON_CPU_NULL);
2407     }
2408 }
2409 
2410 static void spapr_add_lmbs(DeviceState *dev, uint64_t addr_start, uint64_t size,
2411                            uint32_t node, bool dedicated_hp_event_source,
2412                            Error **errp)
2413 {
2414     sPAPRDRConnector *drc;
2415     sPAPRDRConnectorClass *drck;
2416     uint32_t nr_lmbs = size/SPAPR_MEMORY_BLOCK_SIZE;
2417     int i, fdt_offset, fdt_size;
2418     void *fdt;
2419     uint64_t addr = addr_start;
2420 
2421     for (i = 0; i < nr_lmbs; i++) {
2422         drc = spapr_dr_connector_by_id(SPAPR_DR_CONNECTOR_TYPE_LMB,
2423                 addr/SPAPR_MEMORY_BLOCK_SIZE);
2424         g_assert(drc);
2425 
2426         fdt = create_device_tree(&fdt_size);
2427         fdt_offset = spapr_populate_memory_node(fdt, node, addr,
2428                                                 SPAPR_MEMORY_BLOCK_SIZE);
2429 
2430         drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
2431         drck->attach(drc, dev, fdt, fdt_offset, !dev->hotplugged, errp);
2432         addr += SPAPR_MEMORY_BLOCK_SIZE;
2433         if (!dev->hotplugged) {
2434             /* guests expect coldplugged LMBs to be pre-allocated */
2435             drck->set_allocation_state(drc, SPAPR_DR_ALLOCATION_STATE_USABLE);
2436             drck->set_isolation_state(drc, SPAPR_DR_ISOLATION_STATE_UNISOLATED);
2437         }
2438     }
2439     /* send hotplug notification to the
2440      * guest only in case of hotplugged memory
2441      */
2442     if (dev->hotplugged) {
2443         if (dedicated_hp_event_source) {
2444             drc = spapr_dr_connector_by_id(SPAPR_DR_CONNECTOR_TYPE_LMB,
2445                     addr_start / SPAPR_MEMORY_BLOCK_SIZE);
2446             drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
2447             spapr_hotplug_req_add_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
2448                                                    nr_lmbs,
2449                                                    drck->get_index(drc));
2450         } else {
2451             spapr_hotplug_req_add_by_count(SPAPR_DR_CONNECTOR_TYPE_LMB,
2452                                            nr_lmbs);
2453         }
2454     }
2455 }
2456 
2457 static void spapr_memory_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
2458                               uint32_t node, Error **errp)
2459 {
2460     Error *local_err = NULL;
2461     sPAPRMachineState *ms = SPAPR_MACHINE(hotplug_dev);
2462     PCDIMMDevice *dimm = PC_DIMM(dev);
2463     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
2464     MemoryRegion *mr = ddc->get_memory_region(dimm);
2465     uint64_t align = memory_region_get_alignment(mr);
2466     uint64_t size = memory_region_size(mr);
2467     uint64_t addr;
2468     char *mem_dev;
2469 
2470     if (size % SPAPR_MEMORY_BLOCK_SIZE) {
2471         error_setg(&local_err, "Hotplugged memory size must be a multiple of "
2472                       "%lld MB", SPAPR_MEMORY_BLOCK_SIZE/M_BYTE);
2473         goto out;
2474     }
2475 
2476     mem_dev = object_property_get_str(OBJECT(dimm), PC_DIMM_MEMDEV_PROP, NULL);
2477     if (mem_dev && !kvmppc_is_mem_backend_page_size_ok(mem_dev)) {
2478         error_setg(&local_err, "Memory backend has bad page size. "
2479                    "Use 'memory-backend-file' with correct mem-path.");
2480         goto out;
2481     }
2482 
2483     pc_dimm_memory_plug(dev, &ms->hotplug_memory, mr, align, &local_err);
2484     if (local_err) {
2485         goto out;
2486     }
2487 
2488     addr = object_property_get_int(OBJECT(dimm), PC_DIMM_ADDR_PROP, &local_err);
2489     if (local_err) {
2490         pc_dimm_memory_unplug(dev, &ms->hotplug_memory, mr);
2491         goto out;
2492     }
2493 
2494     spapr_add_lmbs(dev, addr, size, node,
2495                    spapr_ovec_test(ms->ov5_cas, OV5_HP_EVT),
2496                    &error_abort);
2497 
2498 out:
2499     error_propagate(errp, local_err);
2500 }
2501 
2502 typedef struct sPAPRDIMMState {
2503     uint32_t nr_lmbs;
2504 } sPAPRDIMMState;
2505 
2506 static void spapr_lmb_release(DeviceState *dev, void *opaque)
2507 {
2508     sPAPRDIMMState *ds = (sPAPRDIMMState *)opaque;
2509     HotplugHandler *hotplug_ctrl;
2510 
2511     if (--ds->nr_lmbs) {
2512         return;
2513     }
2514 
2515     g_free(ds);
2516 
2517     /*
2518      * Now that all the LMBs have been removed by the guest, call the
2519      * pc-dimm unplug handler to cleanup up the pc-dimm device.
2520      */
2521     hotplug_ctrl = qdev_get_hotplug_handler(dev);
2522     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
2523 }
2524 
2525 static void spapr_del_lmbs(DeviceState *dev, uint64_t addr_start, uint64_t size,
2526                            Error **errp)
2527 {
2528     sPAPRDRConnector *drc;
2529     sPAPRDRConnectorClass *drck;
2530     uint32_t nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
2531     int i;
2532     sPAPRDIMMState *ds = g_malloc0(sizeof(sPAPRDIMMState));
2533     uint64_t addr = addr_start;
2534 
2535     ds->nr_lmbs = nr_lmbs;
2536     for (i = 0; i < nr_lmbs; i++) {
2537         drc = spapr_dr_connector_by_id(SPAPR_DR_CONNECTOR_TYPE_LMB,
2538                 addr / SPAPR_MEMORY_BLOCK_SIZE);
2539         g_assert(drc);
2540 
2541         drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
2542         drck->detach(drc, dev, spapr_lmb_release, ds, errp);
2543         addr += SPAPR_MEMORY_BLOCK_SIZE;
2544     }
2545 
2546     drc = spapr_dr_connector_by_id(SPAPR_DR_CONNECTOR_TYPE_LMB,
2547                                    addr_start / SPAPR_MEMORY_BLOCK_SIZE);
2548     drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
2549     spapr_hotplug_req_remove_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
2550                                               nr_lmbs,
2551                                               drck->get_index(drc));
2552 }
2553 
2554 static void spapr_memory_unplug(HotplugHandler *hotplug_dev, DeviceState *dev,
2555                                 Error **errp)
2556 {
2557     sPAPRMachineState *ms = SPAPR_MACHINE(hotplug_dev);
2558     PCDIMMDevice *dimm = PC_DIMM(dev);
2559     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
2560     MemoryRegion *mr = ddc->get_memory_region(dimm);
2561 
2562     pc_dimm_memory_unplug(dev, &ms->hotplug_memory, mr);
2563     object_unparent(OBJECT(dev));
2564 }
2565 
2566 static void spapr_memory_unplug_request(HotplugHandler *hotplug_dev,
2567                                         DeviceState *dev, Error **errp)
2568 {
2569     Error *local_err = NULL;
2570     PCDIMMDevice *dimm = PC_DIMM(dev);
2571     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
2572     MemoryRegion *mr = ddc->get_memory_region(dimm);
2573     uint64_t size = memory_region_size(mr);
2574     uint64_t addr;
2575 
2576     addr = object_property_get_int(OBJECT(dimm), PC_DIMM_ADDR_PROP, &local_err);
2577     if (local_err) {
2578         goto out;
2579     }
2580 
2581     spapr_del_lmbs(dev, addr, size, &error_abort);
2582 out:
2583     error_propagate(errp, local_err);
2584 }
2585 
2586 void *spapr_populate_hotplug_cpu_dt(CPUState *cs, int *fdt_offset,
2587                                     sPAPRMachineState *spapr)
2588 {
2589     PowerPCCPU *cpu = POWERPC_CPU(cs);
2590     DeviceClass *dc = DEVICE_GET_CLASS(cs);
2591     int id = ppc_get_vcpu_dt_id(cpu);
2592     void *fdt;
2593     int offset, fdt_size;
2594     char *nodename;
2595 
2596     fdt = create_device_tree(&fdt_size);
2597     nodename = g_strdup_printf("%s@%x", dc->fw_name, id);
2598     offset = fdt_add_subnode(fdt, 0, nodename);
2599 
2600     spapr_populate_cpu_dt(cs, fdt, offset, spapr);
2601     g_free(nodename);
2602 
2603     *fdt_offset = offset;
2604     return fdt;
2605 }
2606 
2607 static void spapr_core_unplug(HotplugHandler *hotplug_dev, DeviceState *dev,
2608                               Error **errp)
2609 {
2610     MachineState *ms = MACHINE(qdev_get_machine());
2611     CPUCore *cc = CPU_CORE(dev);
2612     CPUArchId *core_slot = spapr_find_cpu_slot(ms, cc->core_id, NULL);
2613 
2614     core_slot->cpu = NULL;
2615     object_unparent(OBJECT(dev));
2616 }
2617 
2618 static void spapr_core_release(DeviceState *dev, void *opaque)
2619 {
2620     HotplugHandler *hotplug_ctrl;
2621 
2622     hotplug_ctrl = qdev_get_hotplug_handler(dev);
2623     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
2624 }
2625 
2626 static
2627 void spapr_core_unplug_request(HotplugHandler *hotplug_dev, DeviceState *dev,
2628                                Error **errp)
2629 {
2630     int index;
2631     sPAPRDRConnector *drc;
2632     sPAPRDRConnectorClass *drck;
2633     Error *local_err = NULL;
2634     CPUCore *cc = CPU_CORE(dev);
2635     int smt = kvmppc_smt_threads();
2636 
2637     if (!spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index)) {
2638         error_setg(errp, "Unable to find CPU core with core-id: %d",
2639                    cc->core_id);
2640         return;
2641     }
2642     if (index == 0) {
2643         error_setg(errp, "Boot CPU core may not be unplugged");
2644         return;
2645     }
2646 
2647     drc = spapr_dr_connector_by_id(SPAPR_DR_CONNECTOR_TYPE_CPU, index * smt);
2648     g_assert(drc);
2649 
2650     drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
2651     drck->detach(drc, dev, spapr_core_release, NULL, &local_err);
2652     if (local_err) {
2653         error_propagate(errp, local_err);
2654         return;
2655     }
2656 
2657     spapr_hotplug_req_remove_by_index(drc);
2658 }
2659 
2660 static void spapr_core_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
2661                             Error **errp)
2662 {
2663     sPAPRMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
2664     MachineClass *mc = MACHINE_GET_CLASS(spapr);
2665     sPAPRCPUCore *core = SPAPR_CPU_CORE(OBJECT(dev));
2666     CPUCore *cc = CPU_CORE(dev);
2667     CPUState *cs = CPU(core->threads);
2668     sPAPRDRConnector *drc;
2669     Error *local_err = NULL;
2670     void *fdt = NULL;
2671     int fdt_offset = 0;
2672     int smt = kvmppc_smt_threads();
2673     CPUArchId *core_slot;
2674     int index;
2675 
2676     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
2677     if (!core_slot) {
2678         error_setg(errp, "Unable to find CPU core with core-id: %d",
2679                    cc->core_id);
2680         return;
2681     }
2682     drc = spapr_dr_connector_by_id(SPAPR_DR_CONNECTOR_TYPE_CPU, index * smt);
2683 
2684     g_assert(drc || !mc->has_hotpluggable_cpus);
2685 
2686     /*
2687      * Setup CPU DT entries only for hotplugged CPUs. For boot time or
2688      * coldplugged CPUs DT entries are setup in spapr_build_fdt().
2689      */
2690     if (dev->hotplugged) {
2691         fdt = spapr_populate_hotplug_cpu_dt(cs, &fdt_offset, spapr);
2692     }
2693 
2694     if (drc) {
2695         sPAPRDRConnectorClass *drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
2696         drck->attach(drc, dev, fdt, fdt_offset, !dev->hotplugged, &local_err);
2697         if (local_err) {
2698             g_free(fdt);
2699             error_propagate(errp, local_err);
2700             return;
2701         }
2702     }
2703 
2704     if (dev->hotplugged) {
2705         /*
2706          * Send hotplug notification interrupt to the guest only in case
2707          * of hotplugged CPUs.
2708          */
2709         spapr_hotplug_req_add_by_index(drc);
2710     } else {
2711         /*
2712          * Set the right DRC states for cold plugged CPU.
2713          */
2714         if (drc) {
2715             sPAPRDRConnectorClass *drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
2716             drck->set_allocation_state(drc, SPAPR_DR_ALLOCATION_STATE_USABLE);
2717             drck->set_isolation_state(drc, SPAPR_DR_ISOLATION_STATE_UNISOLATED);
2718         }
2719     }
2720     core_slot->cpu = OBJECT(dev);
2721 }
2722 
2723 static void spapr_core_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
2724                                 Error **errp)
2725 {
2726     MachineState *machine = MACHINE(OBJECT(hotplug_dev));
2727     MachineClass *mc = MACHINE_GET_CLASS(hotplug_dev);
2728     Error *local_err = NULL;
2729     CPUCore *cc = CPU_CORE(dev);
2730     char *base_core_type = spapr_get_cpu_core_type(machine->cpu_model);
2731     const char *type = object_get_typename(OBJECT(dev));
2732     CPUArchId *core_slot;
2733     int index;
2734 
2735     if (dev->hotplugged && !mc->has_hotpluggable_cpus) {
2736         error_setg(&local_err, "CPU hotplug not supported for this machine");
2737         goto out;
2738     }
2739 
2740     if (strcmp(base_core_type, type)) {
2741         error_setg(&local_err, "CPU core type should be %s", base_core_type);
2742         goto out;
2743     }
2744 
2745     if (cc->core_id % smp_threads) {
2746         error_setg(&local_err, "invalid core id %d", cc->core_id);
2747         goto out;
2748     }
2749 
2750     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
2751     if (!core_slot) {
2752         error_setg(&local_err, "core id %d out of range", cc->core_id);
2753         goto out;
2754     }
2755 
2756     if (core_slot->cpu) {
2757         error_setg(&local_err, "core %d already populated", cc->core_id);
2758         goto out;
2759     }
2760 
2761 out:
2762     g_free(base_core_type);
2763     error_propagate(errp, local_err);
2764 }
2765 
2766 static void spapr_machine_device_plug(HotplugHandler *hotplug_dev,
2767                                       DeviceState *dev, Error **errp)
2768 {
2769     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(qdev_get_machine());
2770 
2771     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
2772         int node;
2773 
2774         if (!smc->dr_lmb_enabled) {
2775             error_setg(errp, "Memory hotplug not supported for this machine");
2776             return;
2777         }
2778         node = object_property_get_int(OBJECT(dev), PC_DIMM_NODE_PROP, errp);
2779         if (*errp) {
2780             return;
2781         }
2782         if (node < 0 || node >= MAX_NODES) {
2783             error_setg(errp, "Invaild node %d", node);
2784             return;
2785         }
2786 
2787         /*
2788          * Currently PowerPC kernel doesn't allow hot-adding memory to
2789          * memory-less node, but instead will silently add the memory
2790          * to the first node that has some memory. This causes two
2791          * unexpected behaviours for the user.
2792          *
2793          * - Memory gets hotplugged to a different node than what the user
2794          *   specified.
2795          * - Since pc-dimm subsystem in QEMU still thinks that memory belongs
2796          *   to memory-less node, a reboot will set things accordingly
2797          *   and the previously hotplugged memory now ends in the right node.
2798          *   This appears as if some memory moved from one node to another.
2799          *
2800          * So until kernel starts supporting memory hotplug to memory-less
2801          * nodes, just prevent such attempts upfront in QEMU.
2802          */
2803         if (nb_numa_nodes && !numa_info[node].node_mem) {
2804             error_setg(errp, "Can't hotplug memory to memory-less node %d",
2805                        node);
2806             return;
2807         }
2808 
2809         spapr_memory_plug(hotplug_dev, dev, node, errp);
2810     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
2811         spapr_core_plug(hotplug_dev, dev, errp);
2812     }
2813 }
2814 
2815 static void spapr_machine_device_unplug(HotplugHandler *hotplug_dev,
2816                                       DeviceState *dev, Error **errp)
2817 {
2818     sPAPRMachineState *sms = SPAPR_MACHINE(qdev_get_machine());
2819     MachineClass *mc = MACHINE_GET_CLASS(qdev_get_machine());
2820 
2821     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
2822         if (spapr_ovec_test(sms->ov5_cas, OV5_HP_EVT)) {
2823             spapr_memory_unplug(hotplug_dev, dev, errp);
2824         } else {
2825             error_setg(errp, "Memory hot unplug not supported for this guest");
2826         }
2827     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
2828         if (!mc->has_hotpluggable_cpus) {
2829             error_setg(errp, "CPU hot unplug not supported on this machine");
2830             return;
2831         }
2832         spapr_core_unplug(hotplug_dev, dev, errp);
2833     }
2834 }
2835 
2836 static void spapr_machine_device_unplug_request(HotplugHandler *hotplug_dev,
2837                                                 DeviceState *dev, Error **errp)
2838 {
2839     sPAPRMachineState *sms = SPAPR_MACHINE(qdev_get_machine());
2840     MachineClass *mc = MACHINE_GET_CLASS(qdev_get_machine());
2841 
2842     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
2843         if (spapr_ovec_test(sms->ov5_cas, OV5_HP_EVT)) {
2844             spapr_memory_unplug_request(hotplug_dev, dev, errp);
2845         } else {
2846             /* NOTE: this means there is a window after guest reset, prior to
2847              * CAS negotiation, where unplug requests will fail due to the
2848              * capability not being detected yet. This is a bit different than
2849              * the case with PCI unplug, where the events will be queued and
2850              * eventually handled by the guest after boot
2851              */
2852             error_setg(errp, "Memory hot unplug not supported for this guest");
2853         }
2854     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
2855         if (!mc->has_hotpluggable_cpus) {
2856             error_setg(errp, "CPU hot unplug not supported on this machine");
2857             return;
2858         }
2859         spapr_core_unplug_request(hotplug_dev, dev, errp);
2860     }
2861 }
2862 
2863 static void spapr_machine_device_pre_plug(HotplugHandler *hotplug_dev,
2864                                           DeviceState *dev, Error **errp)
2865 {
2866     if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
2867         spapr_core_pre_plug(hotplug_dev, dev, errp);
2868     }
2869 }
2870 
2871 static HotplugHandler *spapr_get_hotplug_handler(MachineState *machine,
2872                                                  DeviceState *dev)
2873 {
2874     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM) ||
2875         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
2876         return HOTPLUG_HANDLER(machine);
2877     }
2878     return NULL;
2879 }
2880 
2881 static unsigned spapr_cpu_index_to_socket_id(unsigned cpu_index)
2882 {
2883     /* Allocate to NUMA nodes on a "socket" basis (not that concept of
2884      * socket means much for the paravirtualized PAPR platform) */
2885     return cpu_index / smp_threads / smp_cores;
2886 }
2887 
2888 static const CPUArchIdList *spapr_possible_cpu_arch_ids(MachineState *machine)
2889 {
2890     int i;
2891     int spapr_max_cores = max_cpus / smp_threads;
2892     MachineClass *mc = MACHINE_GET_CLASS(machine);
2893 
2894     if (!mc->has_hotpluggable_cpus) {
2895         spapr_max_cores = QEMU_ALIGN_UP(smp_cpus, smp_threads) / smp_threads;
2896     }
2897     if (machine->possible_cpus) {
2898         assert(machine->possible_cpus->len == spapr_max_cores);
2899         return machine->possible_cpus;
2900     }
2901 
2902     machine->possible_cpus = g_malloc0(sizeof(CPUArchIdList) +
2903                              sizeof(CPUArchId) * spapr_max_cores);
2904     machine->possible_cpus->len = spapr_max_cores;
2905     for (i = 0; i < machine->possible_cpus->len; i++) {
2906         int core_id = i * smp_threads;
2907 
2908         machine->possible_cpus->cpus[i].vcpus_count = smp_threads;
2909         machine->possible_cpus->cpus[i].arch_id = core_id;
2910         machine->possible_cpus->cpus[i].props.has_core_id = true;
2911         machine->possible_cpus->cpus[i].props.core_id = core_id;
2912         /* TODO: add 'has_node/node' here to describe
2913            to which node core belongs */
2914     }
2915     return machine->possible_cpus;
2916 }
2917 
2918 static void spapr_phb_placement(sPAPRMachineState *spapr, uint32_t index,
2919                                 uint64_t *buid, hwaddr *pio,
2920                                 hwaddr *mmio32, hwaddr *mmio64,
2921                                 unsigned n_dma, uint32_t *liobns, Error **errp)
2922 {
2923     /*
2924      * New-style PHB window placement.
2925      *
2926      * Goals: Gives large (1TiB), naturally aligned 64-bit MMIO window
2927      * for each PHB, in addition to 2GiB 32-bit MMIO and 64kiB PIO
2928      * windows.
2929      *
2930      * Some guest kernels can't work with MMIO windows above 1<<46
2931      * (64TiB), so we place up to 31 PHBs in the area 32TiB..64TiB
2932      *
2933      * 32TiB..(33TiB+1984kiB) contains the 64kiB PIO windows for each
2934      * PHB stacked together.  (32TiB+2GiB)..(32TiB+64GiB) contains the
2935      * 2GiB 32-bit MMIO windows for each PHB.  Then 33..64TiB has the
2936      * 1TiB 64-bit MMIO windows for each PHB.
2937      */
2938     const uint64_t base_buid = 0x800000020000000ULL;
2939 #define SPAPR_MAX_PHBS ((SPAPR_PCI_LIMIT - SPAPR_PCI_BASE) / \
2940                         SPAPR_PCI_MEM64_WIN_SIZE - 1)
2941     int i;
2942 
2943     /* Sanity check natural alignments */
2944     QEMU_BUILD_BUG_ON((SPAPR_PCI_BASE % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
2945     QEMU_BUILD_BUG_ON((SPAPR_PCI_LIMIT % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
2946     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM64_WIN_SIZE % SPAPR_PCI_MEM32_WIN_SIZE) != 0);
2947     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM32_WIN_SIZE % SPAPR_PCI_IO_WIN_SIZE) != 0);
2948     /* Sanity check bounds */
2949     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_IO_WIN_SIZE) >
2950                       SPAPR_PCI_MEM32_WIN_SIZE);
2951     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_MEM32_WIN_SIZE) >
2952                       SPAPR_PCI_MEM64_WIN_SIZE);
2953 
2954     if (index >= SPAPR_MAX_PHBS) {
2955         error_setg(errp, "\"index\" for PAPR PHB is too large (max %llu)",
2956                    SPAPR_MAX_PHBS - 1);
2957         return;
2958     }
2959 
2960     *buid = base_buid + index;
2961     for (i = 0; i < n_dma; ++i) {
2962         liobns[i] = SPAPR_PCI_LIOBN(index, i);
2963     }
2964 
2965     *pio = SPAPR_PCI_BASE + index * SPAPR_PCI_IO_WIN_SIZE;
2966     *mmio32 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM32_WIN_SIZE;
2967     *mmio64 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM64_WIN_SIZE;
2968 }
2969 
2970 static ICSState *spapr_ics_get(XICSFabric *dev, int irq)
2971 {
2972     sPAPRMachineState *spapr = SPAPR_MACHINE(dev);
2973 
2974     return ics_valid_irq(spapr->ics, irq) ? spapr->ics : NULL;
2975 }
2976 
2977 static void spapr_ics_resend(XICSFabric *dev)
2978 {
2979     sPAPRMachineState *spapr = SPAPR_MACHINE(dev);
2980 
2981     ics_resend(spapr->ics);
2982 }
2983 
2984 static ICPState *spapr_icp_get(XICSFabric *xi, int server)
2985 {
2986     sPAPRMachineState *spapr = SPAPR_MACHINE(xi);
2987 
2988     return (server < spapr->nr_servers) ? &spapr->icps[server] : NULL;
2989 }
2990 
2991 static void spapr_machine_class_init(ObjectClass *oc, void *data)
2992 {
2993     MachineClass *mc = MACHINE_CLASS(oc);
2994     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(oc);
2995     FWPathProviderClass *fwc = FW_PATH_PROVIDER_CLASS(oc);
2996     NMIClass *nc = NMI_CLASS(oc);
2997     HotplugHandlerClass *hc = HOTPLUG_HANDLER_CLASS(oc);
2998     PPCVirtualHypervisorClass *vhc = PPC_VIRTUAL_HYPERVISOR_CLASS(oc);
2999     XICSFabricClass *xic = XICS_FABRIC_CLASS(oc);
3000 
3001     mc->desc = "pSeries Logical Partition (PAPR compliant)";
3002 
3003     /*
3004      * We set up the default / latest behaviour here.  The class_init
3005      * functions for the specific versioned machine types can override
3006      * these details for backwards compatibility
3007      */
3008     mc->init = ppc_spapr_init;
3009     mc->reset = ppc_spapr_reset;
3010     mc->block_default_type = IF_SCSI;
3011     mc->max_cpus = 1024;
3012     mc->no_parallel = 1;
3013     mc->default_boot_order = "";
3014     mc->default_ram_size = 512 * M_BYTE;
3015     mc->kvm_type = spapr_kvm_type;
3016     mc->has_dynamic_sysbus = true;
3017     mc->pci_allow_0_address = true;
3018     mc->get_hotplug_handler = spapr_get_hotplug_handler;
3019     hc->pre_plug = spapr_machine_device_pre_plug;
3020     hc->plug = spapr_machine_device_plug;
3021     hc->unplug = spapr_machine_device_unplug;
3022     mc->cpu_index_to_socket_id = spapr_cpu_index_to_socket_id;
3023     mc->possible_cpu_arch_ids = spapr_possible_cpu_arch_ids;
3024     hc->unplug_request = spapr_machine_device_unplug_request;
3025 
3026     smc->dr_lmb_enabled = true;
3027     smc->tcg_default_cpu = "POWER8";
3028     mc->has_hotpluggable_cpus = true;
3029     fwc->get_dev_path = spapr_get_fw_dev_path;
3030     nc->nmi_monitor_handler = spapr_nmi;
3031     smc->phb_placement = spapr_phb_placement;
3032     vhc->hypercall = emulate_spapr_hypercall;
3033     vhc->hpt_mask = spapr_hpt_mask;
3034     vhc->map_hptes = spapr_map_hptes;
3035     vhc->unmap_hptes = spapr_unmap_hptes;
3036     vhc->store_hpte = spapr_store_hpte;
3037     xic->ics_get = spapr_ics_get;
3038     xic->ics_resend = spapr_ics_resend;
3039     xic->icp_get = spapr_icp_get;
3040 }
3041 
3042 static const TypeInfo spapr_machine_info = {
3043     .name          = TYPE_SPAPR_MACHINE,
3044     .parent        = TYPE_MACHINE,
3045     .abstract      = true,
3046     .instance_size = sizeof(sPAPRMachineState),
3047     .instance_init = spapr_machine_initfn,
3048     .instance_finalize = spapr_machine_finalizefn,
3049     .class_size    = sizeof(sPAPRMachineClass),
3050     .class_init    = spapr_machine_class_init,
3051     .interfaces = (InterfaceInfo[]) {
3052         { TYPE_FW_PATH_PROVIDER },
3053         { TYPE_NMI },
3054         { TYPE_HOTPLUG_HANDLER },
3055         { TYPE_PPC_VIRTUAL_HYPERVISOR },
3056         { TYPE_XICS_FABRIC },
3057         { }
3058     },
3059 };
3060 
3061 #define DEFINE_SPAPR_MACHINE(suffix, verstr, latest)                 \
3062     static void spapr_machine_##suffix##_class_init(ObjectClass *oc, \
3063                                                     void *data)      \
3064     {                                                                \
3065         MachineClass *mc = MACHINE_CLASS(oc);                        \
3066         spapr_machine_##suffix##_class_options(mc);                  \
3067         if (latest) {                                                \
3068             mc->alias = "pseries";                                   \
3069             mc->is_default = 1;                                      \
3070         }                                                            \
3071     }                                                                \
3072     static void spapr_machine_##suffix##_instance_init(Object *obj)  \
3073     {                                                                \
3074         MachineState *machine = MACHINE(obj);                        \
3075         spapr_machine_##suffix##_instance_options(machine);          \
3076     }                                                                \
3077     static const TypeInfo spapr_machine_##suffix##_info = {          \
3078         .name = MACHINE_TYPE_NAME("pseries-" verstr),                \
3079         .parent = TYPE_SPAPR_MACHINE,                                \
3080         .class_init = spapr_machine_##suffix##_class_init,           \
3081         .instance_init = spapr_machine_##suffix##_instance_init,     \
3082     };                                                               \
3083     static void spapr_machine_register_##suffix(void)                \
3084     {                                                                \
3085         type_register(&spapr_machine_##suffix##_info);               \
3086     }                                                                \
3087     type_init(spapr_machine_register_##suffix)
3088 
3089 /*
3090  * pseries-2.9
3091  */
3092 static void spapr_machine_2_9_instance_options(MachineState *machine)
3093 {
3094 }
3095 
3096 static void spapr_machine_2_9_class_options(MachineClass *mc)
3097 {
3098     /* Defaults for the latest behaviour inherited from the base class */
3099 }
3100 
3101 DEFINE_SPAPR_MACHINE(2_9, "2.9", true);
3102 
3103 /*
3104  * pseries-2.8
3105  */
3106 #define SPAPR_COMPAT_2_8                            \
3107     HW_COMPAT_2_8
3108 
3109 static void spapr_machine_2_8_instance_options(MachineState *machine)
3110 {
3111     spapr_machine_2_9_instance_options(machine);
3112 }
3113 
3114 static void spapr_machine_2_8_class_options(MachineClass *mc)
3115 {
3116     spapr_machine_2_9_class_options(mc);
3117     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_8);
3118 }
3119 
3120 DEFINE_SPAPR_MACHINE(2_8, "2.8", false);
3121 
3122 /*
3123  * pseries-2.7
3124  */
3125 #define SPAPR_COMPAT_2_7                            \
3126     HW_COMPAT_2_7                                   \
3127     {                                               \
3128         .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,     \
3129         .property = "mem_win_size",                 \
3130         .value    = stringify(SPAPR_PCI_2_7_MMIO_WIN_SIZE),\
3131     },                                              \
3132     {                                               \
3133         .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,     \
3134         .property = "mem64_win_size",               \
3135         .value    = "0",                            \
3136     },                                              \
3137     {                                               \
3138         .driver = TYPE_POWERPC_CPU,                 \
3139         .property = "pre-2.8-migration",            \
3140         .value    = "on",                           \
3141     },                                              \
3142     {                                               \
3143         .driver = TYPE_SPAPR_PCI_HOST_BRIDGE,       \
3144         .property = "pre-2.8-migration",            \
3145         .value    = "on",                           \
3146     },
3147 
3148 static void phb_placement_2_7(sPAPRMachineState *spapr, uint32_t index,
3149                               uint64_t *buid, hwaddr *pio,
3150                               hwaddr *mmio32, hwaddr *mmio64,
3151                               unsigned n_dma, uint32_t *liobns, Error **errp)
3152 {
3153     /* Legacy PHB placement for pseries-2.7 and earlier machine types */
3154     const uint64_t base_buid = 0x800000020000000ULL;
3155     const hwaddr phb_spacing = 0x1000000000ULL; /* 64 GiB */
3156     const hwaddr mmio_offset = 0xa0000000; /* 2 GiB + 512 MiB */
3157     const hwaddr pio_offset = 0x80000000; /* 2 GiB */
3158     const uint32_t max_index = 255;
3159     const hwaddr phb0_alignment = 0x10000000000ULL; /* 1 TiB */
3160 
3161     uint64_t ram_top = MACHINE(spapr)->ram_size;
3162     hwaddr phb0_base, phb_base;
3163     int i;
3164 
3165     /* Do we have hotpluggable memory? */
3166     if (MACHINE(spapr)->maxram_size > ram_top) {
3167         /* Can't just use maxram_size, because there may be an
3168          * alignment gap between normal and hotpluggable memory
3169          * regions */
3170         ram_top = spapr->hotplug_memory.base +
3171             memory_region_size(&spapr->hotplug_memory.mr);
3172     }
3173 
3174     phb0_base = QEMU_ALIGN_UP(ram_top, phb0_alignment);
3175 
3176     if (index > max_index) {
3177         error_setg(errp, "\"index\" for PAPR PHB is too large (max %u)",
3178                    max_index);
3179         return;
3180     }
3181 
3182     *buid = base_buid + index;
3183     for (i = 0; i < n_dma; ++i) {
3184         liobns[i] = SPAPR_PCI_LIOBN(index, i);
3185     }
3186 
3187     phb_base = phb0_base + index * phb_spacing;
3188     *pio = phb_base + pio_offset;
3189     *mmio32 = phb_base + mmio_offset;
3190     /*
3191      * We don't set the 64-bit MMIO window, relying on the PHB's
3192      * fallback behaviour of automatically splitting a large "32-bit"
3193      * window into contiguous 32-bit and 64-bit windows
3194      */
3195 }
3196 
3197 static void spapr_machine_2_7_instance_options(MachineState *machine)
3198 {
3199     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
3200 
3201     spapr_machine_2_8_instance_options(machine);
3202     spapr->use_hotplug_event_source = false;
3203 }
3204 
3205 static void spapr_machine_2_7_class_options(MachineClass *mc)
3206 {
3207     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3208 
3209     spapr_machine_2_8_class_options(mc);
3210     smc->tcg_default_cpu = "POWER7";
3211     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_7);
3212     smc->phb_placement = phb_placement_2_7;
3213 }
3214 
3215 DEFINE_SPAPR_MACHINE(2_7, "2.7", false);
3216 
3217 /*
3218  * pseries-2.6
3219  */
3220 #define SPAPR_COMPAT_2_6 \
3221     HW_COMPAT_2_6 \
3222     { \
3223         .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,\
3224         .property = "ddw",\
3225         .value    = stringify(off),\
3226     },
3227 
3228 static void spapr_machine_2_6_instance_options(MachineState *machine)
3229 {
3230     spapr_machine_2_7_instance_options(machine);
3231 }
3232 
3233 static void spapr_machine_2_6_class_options(MachineClass *mc)
3234 {
3235     spapr_machine_2_7_class_options(mc);
3236     mc->has_hotpluggable_cpus = false;
3237     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_6);
3238 }
3239 
3240 DEFINE_SPAPR_MACHINE(2_6, "2.6", false);
3241 
3242 /*
3243  * pseries-2.5
3244  */
3245 #define SPAPR_COMPAT_2_5 \
3246     HW_COMPAT_2_5 \
3247     { \
3248         .driver   = "spapr-vlan", \
3249         .property = "use-rx-buffer-pools", \
3250         .value    = "off", \
3251     },
3252 
3253 static void spapr_machine_2_5_instance_options(MachineState *machine)
3254 {
3255     spapr_machine_2_6_instance_options(machine);
3256 }
3257 
3258 static void spapr_machine_2_5_class_options(MachineClass *mc)
3259 {
3260     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3261 
3262     spapr_machine_2_6_class_options(mc);
3263     smc->use_ohci_by_default = true;
3264     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_5);
3265 }
3266 
3267 DEFINE_SPAPR_MACHINE(2_5, "2.5", false);
3268 
3269 /*
3270  * pseries-2.4
3271  */
3272 #define SPAPR_COMPAT_2_4 \
3273         HW_COMPAT_2_4
3274 
3275 static void spapr_machine_2_4_instance_options(MachineState *machine)
3276 {
3277     spapr_machine_2_5_instance_options(machine);
3278 }
3279 
3280 static void spapr_machine_2_4_class_options(MachineClass *mc)
3281 {
3282     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3283 
3284     spapr_machine_2_5_class_options(mc);
3285     smc->dr_lmb_enabled = false;
3286     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_4);
3287 }
3288 
3289 DEFINE_SPAPR_MACHINE(2_4, "2.4", false);
3290 
3291 /*
3292  * pseries-2.3
3293  */
3294 #define SPAPR_COMPAT_2_3 \
3295         HW_COMPAT_2_3 \
3296         {\
3297             .driver   = "spapr-pci-host-bridge",\
3298             .property = "dynamic-reconfiguration",\
3299             .value    = "off",\
3300         },
3301 
3302 static void spapr_machine_2_3_instance_options(MachineState *machine)
3303 {
3304     spapr_machine_2_4_instance_options(machine);
3305     savevm_skip_section_footers();
3306     global_state_set_optional();
3307     savevm_skip_configuration();
3308 }
3309 
3310 static void spapr_machine_2_3_class_options(MachineClass *mc)
3311 {
3312     spapr_machine_2_4_class_options(mc);
3313     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_3);
3314 }
3315 DEFINE_SPAPR_MACHINE(2_3, "2.3", false);
3316 
3317 /*
3318  * pseries-2.2
3319  */
3320 
3321 #define SPAPR_COMPAT_2_2 \
3322         HW_COMPAT_2_2 \
3323         {\
3324             .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,\
3325             .property = "mem_win_size",\
3326             .value    = "0x20000000",\
3327         },
3328 
3329 static void spapr_machine_2_2_instance_options(MachineState *machine)
3330 {
3331     spapr_machine_2_3_instance_options(machine);
3332     machine->suppress_vmdesc = true;
3333 }
3334 
3335 static void spapr_machine_2_2_class_options(MachineClass *mc)
3336 {
3337     spapr_machine_2_3_class_options(mc);
3338     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_2);
3339 }
3340 DEFINE_SPAPR_MACHINE(2_2, "2.2", false);
3341 
3342 /*
3343  * pseries-2.1
3344  */
3345 #define SPAPR_COMPAT_2_1 \
3346         HW_COMPAT_2_1
3347 
3348 static void spapr_machine_2_1_instance_options(MachineState *machine)
3349 {
3350     spapr_machine_2_2_instance_options(machine);
3351 }
3352 
3353 static void spapr_machine_2_1_class_options(MachineClass *mc)
3354 {
3355     spapr_machine_2_2_class_options(mc);
3356     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_1);
3357 }
3358 DEFINE_SPAPR_MACHINE(2_1, "2.1", false);
3359 
3360 static void spapr_machine_register_types(void)
3361 {
3362     type_register_static(&spapr_machine_info);
3363 }
3364 
3365 type_init(spapr_machine_register_types)
3366