xref: /openbmc/qemu/hw/ppc/spapr.c (revision 806864d9a8a6d5c4cee2ca9bd00474346143113b)
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/kvm.h"
40 #include "sysemu/device_tree.h"
41 #include "kvm_ppc.h"
42 #include "migration/migration.h"
43 #include "mmu-hash64.h"
44 #include "qom/cpu.h"
45 
46 #include "hw/boards.h"
47 #include "hw/ppc/ppc.h"
48 #include "hw/loader.h"
49 
50 #include "hw/ppc/fdt.h"
51 #include "hw/ppc/spapr.h"
52 #include "hw/ppc/spapr_vio.h"
53 #include "hw/pci-host/spapr.h"
54 #include "hw/ppc/xics.h"
55 #include "hw/pci/msi.h"
56 
57 #include "hw/pci/pci.h"
58 #include "hw/scsi/scsi.h"
59 #include "hw/virtio/virtio-scsi.h"
60 
61 #include "exec/address-spaces.h"
62 #include "hw/usb.h"
63 #include "qemu/config-file.h"
64 #include "qemu/error-report.h"
65 #include "trace.h"
66 #include "hw/nmi.h"
67 
68 #include "hw/compat.h"
69 #include "qemu/cutils.h"
70 #include "hw/ppc/spapr_cpu_core.h"
71 #include "qmp-commands.h"
72 
73 #include <libfdt.h>
74 
75 /* SLOF memory layout:
76  *
77  * SLOF raw image loaded at 0, copies its romfs right below the flat
78  * device-tree, then position SLOF itself 31M below that
79  *
80  * So we set FW_OVERHEAD to 40MB which should account for all of that
81  * and more
82  *
83  * We load our kernel at 4M, leaving space for SLOF initial image
84  */
85 #define FDT_MAX_SIZE            0x100000
86 #define RTAS_MAX_SIZE           0x10000
87 #define RTAS_MAX_ADDR           0x80000000 /* RTAS must stay below that */
88 #define FW_MAX_SIZE             0x400000
89 #define FW_FILE_NAME            "slof.bin"
90 #define FW_OVERHEAD             0x2800000
91 #define KERNEL_LOAD_ADDR        FW_MAX_SIZE
92 
93 #define MIN_RMA_SLOF            128UL
94 
95 #define PHANDLE_XICP            0x00001111
96 
97 #define HTAB_SIZE(spapr)        (1ULL << ((spapr)->htab_shift))
98 
99 static XICSState *try_create_xics(const char *type, int nr_servers,
100                                   int nr_irqs, Error **errp)
101 {
102     Error *err = NULL;
103     DeviceState *dev;
104 
105     dev = qdev_create(NULL, type);
106     qdev_prop_set_uint32(dev, "nr_servers", nr_servers);
107     qdev_prop_set_uint32(dev, "nr_irqs", nr_irqs);
108     object_property_set_bool(OBJECT(dev), true, "realized", &err);
109     if (err) {
110         error_propagate(errp, err);
111         object_unparent(OBJECT(dev));
112         return NULL;
113     }
114     return XICS_COMMON(dev);
115 }
116 
117 static XICSState *xics_system_init(MachineState *machine,
118                                    int nr_servers, int nr_irqs, Error **errp)
119 {
120     XICSState *xics = NULL;
121 
122     if (kvm_enabled()) {
123         Error *err = NULL;
124 
125         if (machine_kernel_irqchip_allowed(machine)) {
126             xics = try_create_xics(TYPE_XICS_SPAPR_KVM, nr_servers, nr_irqs,
127                                    &err);
128         }
129         if (machine_kernel_irqchip_required(machine) && !xics) {
130             error_reportf_err(err,
131                               "kernel_irqchip requested but unavailable: ");
132         } else {
133             error_free(err);
134         }
135     }
136 
137     if (!xics) {
138         xics = try_create_xics(TYPE_XICS_SPAPR, nr_servers, nr_irqs, errp);
139     }
140 
141     return xics;
142 }
143 
144 static int spapr_fixup_cpu_smt_dt(void *fdt, int offset, PowerPCCPU *cpu,
145                                   int smt_threads)
146 {
147     int i, ret = 0;
148     uint32_t servers_prop[smt_threads];
149     uint32_t gservers_prop[smt_threads * 2];
150     int index = ppc_get_vcpu_dt_id(cpu);
151 
152     if (cpu->cpu_version) {
153         ret = fdt_setprop_cell(fdt, offset, "cpu-version", cpu->cpu_version);
154         if (ret < 0) {
155             return ret;
156         }
157     }
158 
159     /* Build interrupt servers and gservers properties */
160     for (i = 0; i < smt_threads; i++) {
161         servers_prop[i] = cpu_to_be32(index + i);
162         /* Hack, direct the group queues back to cpu 0 */
163         gservers_prop[i*2] = cpu_to_be32(index + i);
164         gservers_prop[i*2 + 1] = 0;
165     }
166     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-server#s",
167                       servers_prop, sizeof(servers_prop));
168     if (ret < 0) {
169         return ret;
170     }
171     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-gserver#s",
172                       gservers_prop, sizeof(gservers_prop));
173 
174     return ret;
175 }
176 
177 static int spapr_fixup_cpu_numa_dt(void *fdt, int offset, CPUState *cs)
178 {
179     int ret = 0;
180     PowerPCCPU *cpu = POWERPC_CPU(cs);
181     int index = ppc_get_vcpu_dt_id(cpu);
182     uint32_t associativity[] = {cpu_to_be32(0x5),
183                                 cpu_to_be32(0x0),
184                                 cpu_to_be32(0x0),
185                                 cpu_to_be32(0x0),
186                                 cpu_to_be32(cs->numa_node),
187                                 cpu_to_be32(index)};
188 
189     /* Advertise NUMA via ibm,associativity */
190     if (nb_numa_nodes > 1) {
191         ret = fdt_setprop(fdt, offset, "ibm,associativity", associativity,
192                           sizeof(associativity));
193     }
194 
195     return ret;
196 }
197 
198 static int spapr_fixup_cpu_dt(void *fdt, sPAPRMachineState *spapr)
199 {
200     int ret = 0, offset, cpus_offset;
201     CPUState *cs;
202     char cpu_model[32];
203     int smt = kvmppc_smt_threads();
204     uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
205 
206     CPU_FOREACH(cs) {
207         PowerPCCPU *cpu = POWERPC_CPU(cs);
208         DeviceClass *dc = DEVICE_GET_CLASS(cs);
209         int index = ppc_get_vcpu_dt_id(cpu);
210 
211         if ((index % smt) != 0) {
212             continue;
213         }
214 
215         snprintf(cpu_model, 32, "%s@%x", dc->fw_name, index);
216 
217         cpus_offset = fdt_path_offset(fdt, "/cpus");
218         if (cpus_offset < 0) {
219             cpus_offset = fdt_add_subnode(fdt, fdt_path_offset(fdt, "/"),
220                                           "cpus");
221             if (cpus_offset < 0) {
222                 return cpus_offset;
223             }
224         }
225         offset = fdt_subnode_offset(fdt, cpus_offset, cpu_model);
226         if (offset < 0) {
227             offset = fdt_add_subnode(fdt, cpus_offset, cpu_model);
228             if (offset < 0) {
229                 return offset;
230             }
231         }
232 
233         ret = fdt_setprop(fdt, offset, "ibm,pft-size",
234                           pft_size_prop, sizeof(pft_size_prop));
235         if (ret < 0) {
236             return ret;
237         }
238 
239         ret = spapr_fixup_cpu_numa_dt(fdt, offset, cs);
240         if (ret < 0) {
241             return ret;
242         }
243 
244         ret = spapr_fixup_cpu_smt_dt(fdt, offset, cpu,
245                                      ppc_get_compat_smt_threads(cpu));
246         if (ret < 0) {
247             return ret;
248         }
249     }
250     return ret;
251 }
252 
253 static hwaddr spapr_node0_size(void)
254 {
255     MachineState *machine = MACHINE(qdev_get_machine());
256 
257     if (nb_numa_nodes) {
258         int i;
259         for (i = 0; i < nb_numa_nodes; ++i) {
260             if (numa_info[i].node_mem) {
261                 return MIN(pow2floor(numa_info[i].node_mem),
262                            machine->ram_size);
263             }
264         }
265     }
266     return machine->ram_size;
267 }
268 
269 static void add_str(GString *s, const gchar *s1)
270 {
271     g_string_append_len(s, s1, strlen(s1) + 1);
272 }
273 
274 static int spapr_populate_memory_node(void *fdt, int nodeid, hwaddr start,
275                                        hwaddr size)
276 {
277     uint32_t associativity[] = {
278         cpu_to_be32(0x4), /* length */
279         cpu_to_be32(0x0), cpu_to_be32(0x0),
280         cpu_to_be32(0x0), cpu_to_be32(nodeid)
281     };
282     char mem_name[32];
283     uint64_t mem_reg_property[2];
284     int off;
285 
286     mem_reg_property[0] = cpu_to_be64(start);
287     mem_reg_property[1] = cpu_to_be64(size);
288 
289     sprintf(mem_name, "memory@" TARGET_FMT_lx, start);
290     off = fdt_add_subnode(fdt, 0, mem_name);
291     _FDT(off);
292     _FDT((fdt_setprop_string(fdt, off, "device_type", "memory")));
293     _FDT((fdt_setprop(fdt, off, "reg", mem_reg_property,
294                       sizeof(mem_reg_property))));
295     _FDT((fdt_setprop(fdt, off, "ibm,associativity", associativity,
296                       sizeof(associativity))));
297     return off;
298 }
299 
300 static int spapr_populate_memory(sPAPRMachineState *spapr, void *fdt)
301 {
302     MachineState *machine = MACHINE(spapr);
303     hwaddr mem_start, node_size;
304     int i, nb_nodes = nb_numa_nodes;
305     NodeInfo *nodes = numa_info;
306     NodeInfo ramnode;
307 
308     /* No NUMA nodes, assume there is just one node with whole RAM */
309     if (!nb_numa_nodes) {
310         nb_nodes = 1;
311         ramnode.node_mem = machine->ram_size;
312         nodes = &ramnode;
313     }
314 
315     for (i = 0, mem_start = 0; i < nb_nodes; ++i) {
316         if (!nodes[i].node_mem) {
317             continue;
318         }
319         if (mem_start >= machine->ram_size) {
320             node_size = 0;
321         } else {
322             node_size = nodes[i].node_mem;
323             if (node_size > machine->ram_size - mem_start) {
324                 node_size = machine->ram_size - mem_start;
325             }
326         }
327         if (!mem_start) {
328             /* ppc_spapr_init() checks for rma_size <= node0_size already */
329             spapr_populate_memory_node(fdt, i, 0, spapr->rma_size);
330             mem_start += spapr->rma_size;
331             node_size -= spapr->rma_size;
332         }
333         for ( ; node_size; ) {
334             hwaddr sizetmp = pow2floor(node_size);
335 
336             /* mem_start != 0 here */
337             if (ctzl(mem_start) < ctzl(sizetmp)) {
338                 sizetmp = 1ULL << ctzl(mem_start);
339             }
340 
341             spapr_populate_memory_node(fdt, i, mem_start, sizetmp);
342             node_size -= sizetmp;
343             mem_start += sizetmp;
344         }
345     }
346 
347     return 0;
348 }
349 
350 /* Populate the "ibm,pa-features" property */
351 static void spapr_populate_pa_features(CPUPPCState *env, void *fdt, int offset)
352 {
353     uint8_t pa_features_206[] = { 6, 0,
354         0xf6, 0x1f, 0xc7, 0x00, 0x80, 0xc0 };
355     uint8_t pa_features_207[] = { 24, 0,
356         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0,
357         0x80, 0x00, 0x00, 0x00, 0x00, 0x00,
358         0x00, 0x00, 0x00, 0x00, 0x80, 0x00,
359         0x80, 0x00, 0x80, 0x00, 0x00, 0x00 };
360     uint8_t *pa_features;
361     size_t pa_size;
362 
363     switch (env->mmu_model) {
364     case POWERPC_MMU_2_06:
365     case POWERPC_MMU_2_06a:
366         pa_features = pa_features_206;
367         pa_size = sizeof(pa_features_206);
368         break;
369     case POWERPC_MMU_2_07:
370     case POWERPC_MMU_2_07a:
371         pa_features = pa_features_207;
372         pa_size = sizeof(pa_features_207);
373         break;
374     default:
375         return;
376     }
377 
378     if (env->ci_large_pages) {
379         /*
380          * Note: we keep CI large pages off by default because a 64K capable
381          * guest provisioned with large pages might otherwise try to map a qemu
382          * framebuffer (or other kind of memory mapped PCI BAR) using 64K pages
383          * even if that qemu runs on a 4k host.
384          * We dd this bit back here if we are confident this is not an issue
385          */
386         pa_features[3] |= 0x20;
387     }
388     if (kvmppc_has_cap_htm() && pa_size > 24) {
389         pa_features[24] |= 0x80;    /* Transactional memory support */
390     }
391 
392     _FDT((fdt_setprop(fdt, offset, "ibm,pa-features", pa_features, pa_size)));
393 }
394 
395 static void spapr_populate_cpu_dt(CPUState *cs, void *fdt, int offset,
396                                   sPAPRMachineState *spapr)
397 {
398     PowerPCCPU *cpu = POWERPC_CPU(cs);
399     CPUPPCState *env = &cpu->env;
400     PowerPCCPUClass *pcc = POWERPC_CPU_GET_CLASS(cs);
401     int index = ppc_get_vcpu_dt_id(cpu);
402     uint32_t segs[] = {cpu_to_be32(28), cpu_to_be32(40),
403                        0xffffffff, 0xffffffff};
404     uint32_t tbfreq = kvm_enabled() ? kvmppc_get_tbfreq()
405         : SPAPR_TIMEBASE_FREQ;
406     uint32_t cpufreq = kvm_enabled() ? kvmppc_get_clockfreq() : 1000000000;
407     uint32_t page_sizes_prop[64];
408     size_t page_sizes_prop_size;
409     uint32_t vcpus_per_socket = smp_threads * smp_cores;
410     uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
411     sPAPRDRConnector *drc;
412     sPAPRDRConnectorClass *drck;
413     int drc_index;
414 
415     drc = spapr_dr_connector_by_id(SPAPR_DR_CONNECTOR_TYPE_CPU, index);
416     if (drc) {
417         drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
418         drc_index = drck->get_index(drc);
419         _FDT((fdt_setprop_cell(fdt, offset, "ibm,my-drc-index", drc_index)));
420     }
421 
422     _FDT((fdt_setprop_cell(fdt, offset, "reg", index)));
423     _FDT((fdt_setprop_string(fdt, offset, "device_type", "cpu")));
424 
425     _FDT((fdt_setprop_cell(fdt, offset, "cpu-version", env->spr[SPR_PVR])));
426     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-block-size",
427                            env->dcache_line_size)));
428     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-line-size",
429                            env->dcache_line_size)));
430     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-block-size",
431                            env->icache_line_size)));
432     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-line-size",
433                            env->icache_line_size)));
434 
435     if (pcc->l1_dcache_size) {
436         _FDT((fdt_setprop_cell(fdt, offset, "d-cache-size",
437                                pcc->l1_dcache_size)));
438     } else {
439         error_report("Warning: Unknown L1 dcache size for cpu");
440     }
441     if (pcc->l1_icache_size) {
442         _FDT((fdt_setprop_cell(fdt, offset, "i-cache-size",
443                                pcc->l1_icache_size)));
444     } else {
445         error_report("Warning: Unknown L1 icache size for cpu");
446     }
447 
448     _FDT((fdt_setprop_cell(fdt, offset, "timebase-frequency", tbfreq)));
449     _FDT((fdt_setprop_cell(fdt, offset, "clock-frequency", cpufreq)));
450     _FDT((fdt_setprop_cell(fdt, offset, "slb-size", env->slb_nr)));
451     _FDT((fdt_setprop_cell(fdt, offset, "ibm,slb-size", env->slb_nr)));
452     _FDT((fdt_setprop_string(fdt, offset, "status", "okay")));
453     _FDT((fdt_setprop(fdt, offset, "64-bit", NULL, 0)));
454 
455     if (env->spr_cb[SPR_PURR].oea_read) {
456         _FDT((fdt_setprop(fdt, offset, "ibm,purr", NULL, 0)));
457     }
458 
459     if (env->mmu_model & POWERPC_MMU_1TSEG) {
460         _FDT((fdt_setprop(fdt, offset, "ibm,processor-segment-sizes",
461                           segs, sizeof(segs))));
462     }
463 
464     /* Advertise VMX/VSX (vector extensions) if available
465      *   0 / no property == no vector extensions
466      *   1               == VMX / Altivec available
467      *   2               == VSX available */
468     if (env->insns_flags & PPC_ALTIVEC) {
469         uint32_t vmx = (env->insns_flags2 & PPC2_VSX) ? 2 : 1;
470 
471         _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", vmx)));
472     }
473 
474     /* Advertise DFP (Decimal Floating Point) if available
475      *   0 / no property == no DFP
476      *   1               == DFP available */
477     if (env->insns_flags2 & PPC2_DFP) {
478         _FDT((fdt_setprop_cell(fdt, offset, "ibm,dfp", 1)));
479     }
480 
481     page_sizes_prop_size = ppc_create_page_sizes_prop(env, page_sizes_prop,
482                                                   sizeof(page_sizes_prop));
483     if (page_sizes_prop_size) {
484         _FDT((fdt_setprop(fdt, offset, "ibm,segment-page-sizes",
485                           page_sizes_prop, page_sizes_prop_size)));
486     }
487 
488     spapr_populate_pa_features(env, fdt, offset);
489 
490     _FDT((fdt_setprop_cell(fdt, offset, "ibm,chip-id",
491                            cs->cpu_index / vcpus_per_socket)));
492 
493     _FDT((fdt_setprop(fdt, offset, "ibm,pft-size",
494                       pft_size_prop, sizeof(pft_size_prop))));
495 
496     _FDT(spapr_fixup_cpu_numa_dt(fdt, offset, cs));
497 
498     _FDT(spapr_fixup_cpu_smt_dt(fdt, offset, cpu,
499                                 ppc_get_compat_smt_threads(cpu)));
500 }
501 
502 static void spapr_populate_cpus_dt_node(void *fdt, sPAPRMachineState *spapr)
503 {
504     CPUState *cs;
505     int cpus_offset;
506     char *nodename;
507     int smt = kvmppc_smt_threads();
508 
509     cpus_offset = fdt_add_subnode(fdt, 0, "cpus");
510     _FDT(cpus_offset);
511     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#address-cells", 0x1)));
512     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#size-cells", 0x0)));
513 
514     /*
515      * We walk the CPUs in reverse order to ensure that CPU DT nodes
516      * created by fdt_add_subnode() end up in the right order in FDT
517      * for the guest kernel the enumerate the CPUs correctly.
518      */
519     CPU_FOREACH_REVERSE(cs) {
520         PowerPCCPU *cpu = POWERPC_CPU(cs);
521         int index = ppc_get_vcpu_dt_id(cpu);
522         DeviceClass *dc = DEVICE_GET_CLASS(cs);
523         int offset;
524 
525         if ((index % smt) != 0) {
526             continue;
527         }
528 
529         nodename = g_strdup_printf("%s@%x", dc->fw_name, index);
530         offset = fdt_add_subnode(fdt, cpus_offset, nodename);
531         g_free(nodename);
532         _FDT(offset);
533         spapr_populate_cpu_dt(cs, fdt, offset, spapr);
534     }
535 
536 }
537 
538 /*
539  * Adds ibm,dynamic-reconfiguration-memory node.
540  * Refer to docs/specs/ppc-spapr-hotplug.txt for the documentation
541  * of this device tree node.
542  */
543 static int spapr_populate_drconf_memory(sPAPRMachineState *spapr, void *fdt)
544 {
545     MachineState *machine = MACHINE(spapr);
546     int ret, i, offset;
547     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
548     uint32_t prop_lmb_size[] = {0, cpu_to_be32(lmb_size)};
549     uint32_t hotplug_lmb_start = spapr->hotplug_memory.base / lmb_size;
550     uint32_t nr_lmbs = (spapr->hotplug_memory.base +
551                        memory_region_size(&spapr->hotplug_memory.mr)) /
552                        lmb_size;
553     uint32_t *int_buf, *cur_index, buf_len;
554     int nr_nodes = nb_numa_nodes ? nb_numa_nodes : 1;
555 
556     /*
557      * Don't create the node if there is no hotpluggable memory
558      */
559     if (machine->ram_size == machine->maxram_size) {
560         return 0;
561     }
562 
563     /*
564      * Allocate enough buffer size to fit in ibm,dynamic-memory
565      * or ibm,associativity-lookup-arrays
566      */
567     buf_len = MAX(nr_lmbs * SPAPR_DR_LMB_LIST_ENTRY_SIZE + 1, nr_nodes * 4 + 2)
568               * sizeof(uint32_t);
569     cur_index = int_buf = g_malloc0(buf_len);
570 
571     offset = fdt_add_subnode(fdt, 0, "ibm,dynamic-reconfiguration-memory");
572 
573     ret = fdt_setprop(fdt, offset, "ibm,lmb-size", prop_lmb_size,
574                     sizeof(prop_lmb_size));
575     if (ret < 0) {
576         goto out;
577     }
578 
579     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-flags-mask", 0xff);
580     if (ret < 0) {
581         goto out;
582     }
583 
584     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-preservation-time", 0x0);
585     if (ret < 0) {
586         goto out;
587     }
588 
589     /* ibm,dynamic-memory */
590     int_buf[0] = cpu_to_be32(nr_lmbs);
591     cur_index++;
592     for (i = 0; i < nr_lmbs; i++) {
593         uint64_t addr = i * lmb_size;
594         uint32_t *dynamic_memory = cur_index;
595 
596         if (i >= hotplug_lmb_start) {
597             sPAPRDRConnector *drc;
598             sPAPRDRConnectorClass *drck;
599 
600             drc = spapr_dr_connector_by_id(SPAPR_DR_CONNECTOR_TYPE_LMB, i);
601             g_assert(drc);
602             drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
603 
604             dynamic_memory[0] = cpu_to_be32(addr >> 32);
605             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
606             dynamic_memory[2] = cpu_to_be32(drck->get_index(drc));
607             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
608             dynamic_memory[4] = cpu_to_be32(numa_get_node(addr, NULL));
609             if (memory_region_present(get_system_memory(), addr)) {
610                 dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_ASSIGNED);
611             } else {
612                 dynamic_memory[5] = cpu_to_be32(0);
613             }
614         } else {
615             /*
616              * LMB information for RMA, boot time RAM and gap b/n RAM and
617              * hotplug memory region -- all these are marked as reserved
618              * and as having no valid DRC.
619              */
620             dynamic_memory[0] = cpu_to_be32(addr >> 32);
621             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
622             dynamic_memory[2] = cpu_to_be32(0);
623             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
624             dynamic_memory[4] = cpu_to_be32(-1);
625             dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_RESERVED |
626                                             SPAPR_LMB_FLAGS_DRC_INVALID);
627         }
628 
629         cur_index += SPAPR_DR_LMB_LIST_ENTRY_SIZE;
630     }
631     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory", int_buf, buf_len);
632     if (ret < 0) {
633         goto out;
634     }
635 
636     /* ibm,associativity-lookup-arrays */
637     cur_index = int_buf;
638     int_buf[0] = cpu_to_be32(nr_nodes);
639     int_buf[1] = cpu_to_be32(4); /* Number of entries per associativity list */
640     cur_index += 2;
641     for (i = 0; i < nr_nodes; i++) {
642         uint32_t associativity[] = {
643             cpu_to_be32(0x0),
644             cpu_to_be32(0x0),
645             cpu_to_be32(0x0),
646             cpu_to_be32(i)
647         };
648         memcpy(cur_index, associativity, sizeof(associativity));
649         cur_index += 4;
650     }
651     ret = fdt_setprop(fdt, offset, "ibm,associativity-lookup-arrays", int_buf,
652             (cur_index - int_buf) * sizeof(uint32_t));
653 out:
654     g_free(int_buf);
655     return ret;
656 }
657 
658 static int spapr_dt_cas_updates(sPAPRMachineState *spapr, void *fdt,
659                                 sPAPROptionVector *ov5_updates)
660 {
661     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
662     int ret = 0, offset;
663 
664     /* Generate ibm,dynamic-reconfiguration-memory node if required */
665     if (spapr_ovec_test(ov5_updates, OV5_DRCONF_MEMORY)) {
666         g_assert(smc->dr_lmb_enabled);
667         ret = spapr_populate_drconf_memory(spapr, fdt);
668         if (ret) {
669             goto out;
670         }
671     }
672 
673     offset = fdt_path_offset(fdt, "/chosen");
674     if (offset < 0) {
675         offset = fdt_add_subnode(fdt, 0, "chosen");
676         if (offset < 0) {
677             return offset;
678         }
679     }
680     ret = spapr_ovec_populate_dt(fdt, offset, spapr->ov5_cas,
681                                  "ibm,architecture-vec-5");
682 
683 out:
684     return ret;
685 }
686 
687 int spapr_h_cas_compose_response(sPAPRMachineState *spapr,
688                                  target_ulong addr, target_ulong size,
689                                  bool cpu_update,
690                                  sPAPROptionVector *ov5_updates)
691 {
692     void *fdt, *fdt_skel;
693     sPAPRDeviceTreeUpdateHeader hdr = { .version_id = 1 };
694 
695     size -= sizeof(hdr);
696 
697     /* Create sceleton */
698     fdt_skel = g_malloc0(size);
699     _FDT((fdt_create(fdt_skel, size)));
700     _FDT((fdt_begin_node(fdt_skel, "")));
701     _FDT((fdt_end_node(fdt_skel)));
702     _FDT((fdt_finish(fdt_skel)));
703     fdt = g_malloc0(size);
704     _FDT((fdt_open_into(fdt_skel, fdt, size)));
705     g_free(fdt_skel);
706 
707     /* Fixup cpu nodes */
708     if (cpu_update) {
709         _FDT((spapr_fixup_cpu_dt(fdt, spapr)));
710     }
711 
712     if (spapr_dt_cas_updates(spapr, fdt, ov5_updates)) {
713         return -1;
714     }
715 
716     /* Pack resulting tree */
717     _FDT((fdt_pack(fdt)));
718 
719     if (fdt_totalsize(fdt) + sizeof(hdr) > size) {
720         trace_spapr_cas_failed(size);
721         return -1;
722     }
723 
724     cpu_physical_memory_write(addr, &hdr, sizeof(hdr));
725     cpu_physical_memory_write(addr + sizeof(hdr), fdt, fdt_totalsize(fdt));
726     trace_spapr_cas_continue(fdt_totalsize(fdt) + sizeof(hdr));
727     g_free(fdt);
728 
729     return 0;
730 }
731 
732 static void spapr_dt_rtas(sPAPRMachineState *spapr, void *fdt)
733 {
734     int rtas;
735     GString *hypertas = g_string_sized_new(256);
736     GString *qemu_hypertas = g_string_sized_new(256);
737     uint32_t refpoints[] = { cpu_to_be32(0x4), cpu_to_be32(0x4) };
738     uint64_t max_hotplug_addr = spapr->hotplug_memory.base +
739         memory_region_size(&spapr->hotplug_memory.mr);
740     uint32_t lrdr_capacity[] = {
741         cpu_to_be32(max_hotplug_addr >> 32),
742         cpu_to_be32(max_hotplug_addr & 0xffffffff),
743         0, cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE),
744         cpu_to_be32(max_cpus / smp_threads),
745     };
746 
747     _FDT(rtas = fdt_add_subnode(fdt, 0, "rtas"));
748 
749     /* hypertas */
750     add_str(hypertas, "hcall-pft");
751     add_str(hypertas, "hcall-term");
752     add_str(hypertas, "hcall-dabr");
753     add_str(hypertas, "hcall-interrupt");
754     add_str(hypertas, "hcall-tce");
755     add_str(hypertas, "hcall-vio");
756     add_str(hypertas, "hcall-splpar");
757     add_str(hypertas, "hcall-bulk");
758     add_str(hypertas, "hcall-set-mode");
759     add_str(hypertas, "hcall-sprg0");
760     add_str(hypertas, "hcall-copy");
761     add_str(hypertas, "hcall-debug");
762     add_str(qemu_hypertas, "hcall-memop1");
763 
764     if (!kvm_enabled() || kvmppc_spapr_use_multitce()) {
765         add_str(hypertas, "hcall-multi-tce");
766     }
767     _FDT(fdt_setprop(fdt, rtas, "ibm,hypertas-functions",
768                      hypertas->str, hypertas->len));
769     g_string_free(hypertas, TRUE);
770     _FDT(fdt_setprop(fdt, rtas, "qemu,hypertas-functions",
771                      qemu_hypertas->str, qemu_hypertas->len));
772     g_string_free(qemu_hypertas, TRUE);
773 
774     _FDT(fdt_setprop(fdt, rtas, "ibm,associativity-reference-points",
775                      refpoints, sizeof(refpoints)));
776 
777     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-error-log-max",
778                           RTAS_ERROR_LOG_MAX));
779     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-event-scan-rate",
780                           RTAS_EVENT_SCAN_RATE));
781 
782     if (msi_nonbroken) {
783         _FDT(fdt_setprop(fdt, rtas, "ibm,change-msix-capable", NULL, 0));
784     }
785 
786     /*
787      * According to PAPR, rtas ibm,os-term does not guarantee a return
788      * back to the guest cpu.
789      *
790      * While an additional ibm,extended-os-term property indicates
791      * that rtas call return will always occur. Set this property.
792      */
793     _FDT(fdt_setprop(fdt, rtas, "ibm,extended-os-term", NULL, 0));
794 
795     _FDT(fdt_setprop(fdt, rtas, "ibm,lrdr-capacity",
796                      lrdr_capacity, sizeof(lrdr_capacity)));
797 
798     spapr_dt_rtas_tokens(fdt, rtas);
799 }
800 
801 static void spapr_dt_chosen(sPAPRMachineState *spapr, void *fdt)
802 {
803     MachineState *machine = MACHINE(spapr);
804     int chosen;
805     const char *boot_device = machine->boot_order;
806     char *stdout_path = spapr_vio_stdout_path(spapr->vio_bus);
807     size_t cb = 0;
808     char *bootlist = get_boot_devices_list(&cb, true);
809 
810     _FDT(chosen = fdt_add_subnode(fdt, 0, "chosen"));
811 
812     _FDT(fdt_setprop_string(fdt, chosen, "bootargs", machine->kernel_cmdline));
813     _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-start",
814                           spapr->initrd_base));
815     _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-end",
816                           spapr->initrd_base + spapr->initrd_size));
817 
818     if (spapr->kernel_size) {
819         uint64_t kprop[2] = { cpu_to_be64(KERNEL_LOAD_ADDR),
820                               cpu_to_be64(spapr->kernel_size) };
821 
822         _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel",
823                          &kprop, sizeof(kprop)));
824         if (spapr->kernel_le) {
825             _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel-le", NULL, 0));
826         }
827     }
828     if (boot_menu) {
829         _FDT((fdt_setprop_cell(fdt, chosen, "qemu,boot-menu", boot_menu)));
830     }
831     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-width", graphic_width));
832     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-height", graphic_height));
833     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-depth", graphic_depth));
834 
835     if (cb && bootlist) {
836         int i;
837 
838         for (i = 0; i < cb; i++) {
839             if (bootlist[i] == '\n') {
840                 bootlist[i] = ' ';
841             }
842         }
843         _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-list", bootlist));
844     }
845 
846     if (boot_device && strlen(boot_device)) {
847         _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-device", boot_device));
848     }
849 
850     if (!spapr->has_graphics && stdout_path) {
851         _FDT(fdt_setprop_string(fdt, chosen, "linux,stdout-path", stdout_path));
852     }
853 
854     g_free(stdout_path);
855     g_free(bootlist);
856 }
857 
858 static void spapr_dt_hypervisor(sPAPRMachineState *spapr, void *fdt)
859 {
860     /* The /hypervisor node isn't in PAPR - this is a hack to allow PR
861      * KVM to work under pHyp with some guest co-operation */
862     int hypervisor;
863     uint8_t hypercall[16];
864 
865     _FDT(hypervisor = fdt_add_subnode(fdt, 0, "hypervisor"));
866     /* indicate KVM hypercall interface */
867     _FDT(fdt_setprop_string(fdt, hypervisor, "compatible", "linux,kvm"));
868     if (kvmppc_has_cap_fixup_hcalls()) {
869         /*
870          * Older KVM versions with older guest kernels were broken
871          * with the magic page, don't allow the guest to map it.
872          */
873         if (!kvmppc_get_hypercall(first_cpu->env_ptr, hypercall,
874                                   sizeof(hypercall))) {
875             _FDT(fdt_setprop(fdt, hypervisor, "hcall-instructions",
876                              hypercall, sizeof(hypercall)));
877         }
878     }
879 }
880 
881 static void *spapr_build_fdt(sPAPRMachineState *spapr,
882                              hwaddr rtas_addr,
883                              hwaddr rtas_size)
884 {
885     MachineState *machine = MACHINE(qdev_get_machine());
886     MachineClass *mc = MACHINE_GET_CLASS(machine);
887     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
888     int ret;
889     void *fdt;
890     sPAPRPHBState *phb;
891     char *buf;
892 
893     fdt = g_malloc0(FDT_MAX_SIZE);
894     _FDT((fdt_create_empty_tree(fdt, FDT_MAX_SIZE)));
895 
896     /* Root node */
897     _FDT(fdt_setprop_string(fdt, 0, "device_type", "chrp"));
898     _FDT(fdt_setprop_string(fdt, 0, "model", "IBM pSeries (emulated by qemu)"));
899     _FDT(fdt_setprop_string(fdt, 0, "compatible", "qemu,pseries"));
900 
901     /*
902      * Add info to guest to indentify which host is it being run on
903      * and what is the uuid of the guest
904      */
905     if (kvmppc_get_host_model(&buf)) {
906         _FDT(fdt_setprop_string(fdt, 0, "host-model", buf));
907         g_free(buf);
908     }
909     if (kvmppc_get_host_serial(&buf)) {
910         _FDT(fdt_setprop_string(fdt, 0, "host-serial", buf));
911         g_free(buf);
912     }
913 
914     buf = qemu_uuid_unparse_strdup(&qemu_uuid);
915 
916     _FDT(fdt_setprop_string(fdt, 0, "vm,uuid", buf));
917     if (qemu_uuid_set) {
918         _FDT(fdt_setprop_string(fdt, 0, "system-id", buf));
919     }
920     g_free(buf);
921 
922     if (qemu_get_vm_name()) {
923         _FDT(fdt_setprop_string(fdt, 0, "ibm,partition-name",
924                                 qemu_get_vm_name()));
925     }
926 
927     _FDT(fdt_setprop_cell(fdt, 0, "#address-cells", 2));
928     _FDT(fdt_setprop_cell(fdt, 0, "#size-cells", 2));
929 
930     /* /interrupt controller */
931     spapr_dt_xics(spapr->xics, fdt, PHANDLE_XICP);
932 
933     ret = spapr_populate_memory(spapr, fdt);
934     if (ret < 0) {
935         error_report("couldn't setup memory nodes in fdt");
936         exit(1);
937     }
938 
939     /* /vdevice */
940     spapr_dt_vdevice(spapr->vio_bus, fdt);
941 
942     if (object_resolve_path_type("", TYPE_SPAPR_RNG, NULL)) {
943         ret = spapr_rng_populate_dt(fdt);
944         if (ret < 0) {
945             error_report("could not set up rng device in the fdt");
946             exit(1);
947         }
948     }
949 
950     QLIST_FOREACH(phb, &spapr->phbs, list) {
951         ret = spapr_populate_pci_dt(phb, PHANDLE_XICP, fdt);
952         if (ret < 0) {
953             error_report("couldn't setup PCI devices in fdt");
954             exit(1);
955         }
956     }
957 
958     /* cpus */
959     spapr_populate_cpus_dt_node(fdt, spapr);
960 
961     if (smc->dr_lmb_enabled) {
962         _FDT(spapr_drc_populate_dt(fdt, 0, NULL, SPAPR_DR_CONNECTOR_TYPE_LMB));
963     }
964 
965     if (mc->query_hotpluggable_cpus) {
966         int offset = fdt_path_offset(fdt, "/cpus");
967         ret = spapr_drc_populate_dt(fdt, offset, NULL,
968                                     SPAPR_DR_CONNECTOR_TYPE_CPU);
969         if (ret < 0) {
970             error_report("Couldn't set up CPU DR device tree properties");
971             exit(1);
972         }
973     }
974 
975     /* /event-sources */
976     spapr_dt_events(spapr, fdt);
977 
978     /* /rtas */
979     spapr_dt_rtas(spapr, fdt);
980 
981     /* /chosen */
982     spapr_dt_chosen(spapr, fdt);
983 
984     /* /hypervisor */
985     if (kvm_enabled()) {
986         spapr_dt_hypervisor(spapr, fdt);
987     }
988 
989     /* Build memory reserve map */
990     if (spapr->kernel_size) {
991         _FDT((fdt_add_mem_rsv(fdt, KERNEL_LOAD_ADDR, spapr->kernel_size)));
992     }
993     if (spapr->initrd_size) {
994         _FDT((fdt_add_mem_rsv(fdt, spapr->initrd_base, spapr->initrd_size)));
995     }
996 
997     /* ibm,client-architecture-support updates */
998     ret = spapr_dt_cas_updates(spapr, fdt, spapr->ov5_cas);
999     if (ret < 0) {
1000         error_report("couldn't setup CAS properties fdt");
1001         exit(1);
1002     }
1003 
1004     return fdt;
1005 }
1006 
1007 static uint64_t translate_kernel_address(void *opaque, uint64_t addr)
1008 {
1009     return (addr & 0x0fffffff) + KERNEL_LOAD_ADDR;
1010 }
1011 
1012 static void emulate_spapr_hypercall(PowerPCCPU *cpu)
1013 {
1014     CPUPPCState *env = &cpu->env;
1015 
1016     if (msr_pr) {
1017         hcall_dprintf("Hypercall made with MSR[PR]=1\n");
1018         env->gpr[3] = H_PRIVILEGE;
1019     } else {
1020         env->gpr[3] = spapr_hypercall(cpu, env->gpr[3], &env->gpr[4]);
1021     }
1022 }
1023 
1024 #define HPTE(_table, _i)   (void *)(((uint64_t *)(_table)) + ((_i) * 2))
1025 #define HPTE_VALID(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_VALID)
1026 #define HPTE_DIRTY(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_HPTE_DIRTY)
1027 #define CLEAN_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) &= tswap64(~HPTE64_V_HPTE_DIRTY))
1028 #define DIRTY_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) |= tswap64(HPTE64_V_HPTE_DIRTY))
1029 
1030 /*
1031  * Get the fd to access the kernel htab, re-opening it if necessary
1032  */
1033 static int get_htab_fd(sPAPRMachineState *spapr)
1034 {
1035     if (spapr->htab_fd >= 0) {
1036         return spapr->htab_fd;
1037     }
1038 
1039     spapr->htab_fd = kvmppc_get_htab_fd(false);
1040     if (spapr->htab_fd < 0) {
1041         error_report("Unable to open fd for reading hash table from KVM: %s",
1042                      strerror(errno));
1043     }
1044 
1045     return spapr->htab_fd;
1046 }
1047 
1048 static void close_htab_fd(sPAPRMachineState *spapr)
1049 {
1050     if (spapr->htab_fd >= 0) {
1051         close(spapr->htab_fd);
1052     }
1053     spapr->htab_fd = -1;
1054 }
1055 
1056 static int spapr_hpt_shift_for_ramsize(uint64_t ramsize)
1057 {
1058     int shift;
1059 
1060     /* We aim for a hash table of size 1/128 the size of RAM (rounded
1061      * up).  The PAPR recommendation is actually 1/64 of RAM size, but
1062      * that's much more than is needed for Linux guests */
1063     shift = ctz64(pow2ceil(ramsize)) - 7;
1064     shift = MAX(shift, 18); /* Minimum architected size */
1065     shift = MIN(shift, 46); /* Maximum architected size */
1066     return shift;
1067 }
1068 
1069 static void spapr_reallocate_hpt(sPAPRMachineState *spapr, int shift,
1070                                  Error **errp)
1071 {
1072     long rc;
1073 
1074     /* Clean up any HPT info from a previous boot */
1075     g_free(spapr->htab);
1076     spapr->htab = NULL;
1077     spapr->htab_shift = 0;
1078     close_htab_fd(spapr);
1079 
1080     rc = kvmppc_reset_htab(shift);
1081     if (rc < 0) {
1082         /* kernel-side HPT needed, but couldn't allocate one */
1083         error_setg_errno(errp, errno,
1084                          "Failed to allocate KVM HPT of order %d (try smaller maxmem?)",
1085                          shift);
1086         /* This is almost certainly fatal, but if the caller really
1087          * wants to carry on with shift == 0, it's welcome to try */
1088     } else if (rc > 0) {
1089         /* kernel-side HPT allocated */
1090         if (rc != shift) {
1091             error_setg(errp,
1092                        "Requested order %d HPT, but kernel allocated order %ld (try smaller maxmem?)",
1093                        shift, rc);
1094         }
1095 
1096         spapr->htab_shift = shift;
1097         spapr->htab = NULL;
1098     } else {
1099         /* kernel-side HPT not needed, allocate in userspace instead */
1100         size_t size = 1ULL << shift;
1101         int i;
1102 
1103         spapr->htab = qemu_memalign(size, size);
1104         if (!spapr->htab) {
1105             error_setg_errno(errp, errno,
1106                              "Could not allocate HPT of order %d", shift);
1107             return;
1108         }
1109 
1110         memset(spapr->htab, 0, size);
1111         spapr->htab_shift = shift;
1112 
1113         for (i = 0; i < size / HASH_PTE_SIZE_64; i++) {
1114             DIRTY_HPTE(HPTE(spapr->htab, i));
1115         }
1116     }
1117 }
1118 
1119 static void find_unknown_sysbus_device(SysBusDevice *sbdev, void *opaque)
1120 {
1121     bool matched = false;
1122 
1123     if (object_dynamic_cast(OBJECT(sbdev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
1124         matched = true;
1125     }
1126 
1127     if (!matched) {
1128         error_report("Device %s is not supported by this machine yet.",
1129                      qdev_fw_name(DEVICE(sbdev)));
1130         exit(1);
1131     }
1132 }
1133 
1134 static void ppc_spapr_reset(void)
1135 {
1136     MachineState *machine = MACHINE(qdev_get_machine());
1137     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
1138     PowerPCCPU *first_ppc_cpu;
1139     uint32_t rtas_limit;
1140     hwaddr rtas_addr, fdt_addr;
1141     void *fdt;
1142     int rc;
1143 
1144     /* Check for unknown sysbus devices */
1145     foreach_dynamic_sysbus_device(find_unknown_sysbus_device, NULL);
1146 
1147     /* Allocate and/or reset the hash page table */
1148     spapr_reallocate_hpt(spapr,
1149                          spapr_hpt_shift_for_ramsize(machine->maxram_size),
1150                          &error_fatal);
1151 
1152     /* Update the RMA size if necessary */
1153     if (spapr->vrma_adjust) {
1154         spapr->rma_size = kvmppc_rma_size(spapr_node0_size(),
1155                                           spapr->htab_shift);
1156     }
1157 
1158     qemu_devices_reset();
1159 
1160     /*
1161      * We place the device tree and RTAS just below either the top of the RMA,
1162      * or just below 2GB, whichever is lowere, so that it can be
1163      * processed with 32-bit real mode code if necessary
1164      */
1165     rtas_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR);
1166     rtas_addr = rtas_limit - RTAS_MAX_SIZE;
1167     fdt_addr = rtas_addr - FDT_MAX_SIZE;
1168 
1169     /* if this reset wasn't generated by CAS, we should reset our
1170      * negotiated options and start from scratch */
1171     if (!spapr->cas_reboot) {
1172         spapr_ovec_cleanup(spapr->ov5_cas);
1173         spapr->ov5_cas = spapr_ovec_new();
1174     }
1175 
1176     fdt = spapr_build_fdt(spapr, rtas_addr, spapr->rtas_size);
1177 
1178     spapr_load_rtas(spapr, fdt, rtas_addr);
1179 
1180     rc = fdt_pack(fdt);
1181 
1182     /* Should only fail if we've built a corrupted tree */
1183     assert(rc == 0);
1184 
1185     if (fdt_totalsize(fdt) > FDT_MAX_SIZE) {
1186         error_report("FDT too big ! 0x%x bytes (max is 0x%x)",
1187                      fdt_totalsize(fdt), FDT_MAX_SIZE);
1188         exit(1);
1189     }
1190 
1191     /* Load the fdt */
1192     qemu_fdt_dumpdtb(fdt, fdt_totalsize(fdt));
1193     cpu_physical_memory_write(fdt_addr, fdt, fdt_totalsize(fdt));
1194     g_free(fdt);
1195 
1196     /* Set up the entry state */
1197     first_ppc_cpu = POWERPC_CPU(first_cpu);
1198     first_ppc_cpu->env.gpr[3] = fdt_addr;
1199     first_ppc_cpu->env.gpr[5] = 0;
1200     first_cpu->halted = 0;
1201     first_ppc_cpu->env.nip = SPAPR_ENTRY_POINT;
1202 
1203     spapr->cas_reboot = false;
1204 }
1205 
1206 static void spapr_create_nvram(sPAPRMachineState *spapr)
1207 {
1208     DeviceState *dev = qdev_create(&spapr->vio_bus->bus, "spapr-nvram");
1209     DriveInfo *dinfo = drive_get(IF_PFLASH, 0, 0);
1210 
1211     if (dinfo) {
1212         qdev_prop_set_drive(dev, "drive", blk_by_legacy_dinfo(dinfo),
1213                             &error_fatal);
1214     }
1215 
1216     qdev_init_nofail(dev);
1217 
1218     spapr->nvram = (struct sPAPRNVRAM *)dev;
1219 }
1220 
1221 static void spapr_rtc_create(sPAPRMachineState *spapr)
1222 {
1223     DeviceState *dev = qdev_create(NULL, TYPE_SPAPR_RTC);
1224 
1225     qdev_init_nofail(dev);
1226     spapr->rtc = dev;
1227 
1228     object_property_add_alias(qdev_get_machine(), "rtc-time",
1229                               OBJECT(spapr->rtc), "date", NULL);
1230 }
1231 
1232 /* Returns whether we want to use VGA or not */
1233 static bool spapr_vga_init(PCIBus *pci_bus, Error **errp)
1234 {
1235     switch (vga_interface_type) {
1236     case VGA_NONE:
1237         return false;
1238     case VGA_DEVICE:
1239         return true;
1240     case VGA_STD:
1241     case VGA_VIRTIO:
1242         return pci_vga_init(pci_bus) != NULL;
1243     default:
1244         error_setg(errp,
1245                    "Unsupported VGA mode, only -vga std or -vga virtio is supported");
1246         return false;
1247     }
1248 }
1249 
1250 static int spapr_post_load(void *opaque, int version_id)
1251 {
1252     sPAPRMachineState *spapr = (sPAPRMachineState *)opaque;
1253     int err = 0;
1254 
1255     /* In earlier versions, there was no separate qdev for the PAPR
1256      * RTC, so the RTC offset was stored directly in sPAPREnvironment.
1257      * So when migrating from those versions, poke the incoming offset
1258      * value into the RTC device */
1259     if (version_id < 3) {
1260         err = spapr_rtc_import_offset(spapr->rtc, spapr->rtc_offset);
1261     }
1262 
1263     return err;
1264 }
1265 
1266 static bool version_before_3(void *opaque, int version_id)
1267 {
1268     return version_id < 3;
1269 }
1270 
1271 static const VMStateDescription vmstate_spapr = {
1272     .name = "spapr",
1273     .version_id = 3,
1274     .minimum_version_id = 1,
1275     .post_load = spapr_post_load,
1276     .fields = (VMStateField[]) {
1277         /* used to be @next_irq */
1278         VMSTATE_UNUSED_BUFFER(version_before_3, 0, 4),
1279 
1280         /* RTC offset */
1281         VMSTATE_UINT64_TEST(rtc_offset, sPAPRMachineState, version_before_3),
1282 
1283         VMSTATE_PPC_TIMEBASE_V(tb, sPAPRMachineState, 2),
1284         VMSTATE_END_OF_LIST()
1285     },
1286 };
1287 
1288 static int htab_save_setup(QEMUFile *f, void *opaque)
1289 {
1290     sPAPRMachineState *spapr = opaque;
1291 
1292     /* "Iteration" header */
1293     qemu_put_be32(f, spapr->htab_shift);
1294 
1295     if (spapr->htab) {
1296         spapr->htab_save_index = 0;
1297         spapr->htab_first_pass = true;
1298     } else {
1299         assert(kvm_enabled());
1300     }
1301 
1302 
1303     return 0;
1304 }
1305 
1306 static void htab_save_first_pass(QEMUFile *f, sPAPRMachineState *spapr,
1307                                  int64_t max_ns)
1308 {
1309     bool has_timeout = max_ns != -1;
1310     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
1311     int index = spapr->htab_save_index;
1312     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
1313 
1314     assert(spapr->htab_first_pass);
1315 
1316     do {
1317         int chunkstart;
1318 
1319         /* Consume invalid HPTEs */
1320         while ((index < htabslots)
1321                && !HPTE_VALID(HPTE(spapr->htab, index))) {
1322             index++;
1323             CLEAN_HPTE(HPTE(spapr->htab, index));
1324         }
1325 
1326         /* Consume valid HPTEs */
1327         chunkstart = index;
1328         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
1329                && HPTE_VALID(HPTE(spapr->htab, index))) {
1330             index++;
1331             CLEAN_HPTE(HPTE(spapr->htab, index));
1332         }
1333 
1334         if (index > chunkstart) {
1335             int n_valid = index - chunkstart;
1336 
1337             qemu_put_be32(f, chunkstart);
1338             qemu_put_be16(f, n_valid);
1339             qemu_put_be16(f, 0);
1340             qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
1341                             HASH_PTE_SIZE_64 * n_valid);
1342 
1343             if (has_timeout &&
1344                 (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
1345                 break;
1346             }
1347         }
1348     } while ((index < htabslots) && !qemu_file_rate_limit(f));
1349 
1350     if (index >= htabslots) {
1351         assert(index == htabslots);
1352         index = 0;
1353         spapr->htab_first_pass = false;
1354     }
1355     spapr->htab_save_index = index;
1356 }
1357 
1358 static int htab_save_later_pass(QEMUFile *f, sPAPRMachineState *spapr,
1359                                 int64_t max_ns)
1360 {
1361     bool final = max_ns < 0;
1362     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
1363     int examined = 0, sent = 0;
1364     int index = spapr->htab_save_index;
1365     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
1366 
1367     assert(!spapr->htab_first_pass);
1368 
1369     do {
1370         int chunkstart, invalidstart;
1371 
1372         /* Consume non-dirty HPTEs */
1373         while ((index < htabslots)
1374                && !HPTE_DIRTY(HPTE(spapr->htab, index))) {
1375             index++;
1376             examined++;
1377         }
1378 
1379         chunkstart = index;
1380         /* Consume valid dirty HPTEs */
1381         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
1382                && HPTE_DIRTY(HPTE(spapr->htab, index))
1383                && HPTE_VALID(HPTE(spapr->htab, index))) {
1384             CLEAN_HPTE(HPTE(spapr->htab, index));
1385             index++;
1386             examined++;
1387         }
1388 
1389         invalidstart = index;
1390         /* Consume invalid dirty HPTEs */
1391         while ((index < htabslots) && (index - invalidstart < USHRT_MAX)
1392                && HPTE_DIRTY(HPTE(spapr->htab, index))
1393                && !HPTE_VALID(HPTE(spapr->htab, index))) {
1394             CLEAN_HPTE(HPTE(spapr->htab, index));
1395             index++;
1396             examined++;
1397         }
1398 
1399         if (index > chunkstart) {
1400             int n_valid = invalidstart - chunkstart;
1401             int n_invalid = index - invalidstart;
1402 
1403             qemu_put_be32(f, chunkstart);
1404             qemu_put_be16(f, n_valid);
1405             qemu_put_be16(f, n_invalid);
1406             qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
1407                             HASH_PTE_SIZE_64 * n_valid);
1408             sent += index - chunkstart;
1409 
1410             if (!final && (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
1411                 break;
1412             }
1413         }
1414 
1415         if (examined >= htabslots) {
1416             break;
1417         }
1418 
1419         if (index >= htabslots) {
1420             assert(index == htabslots);
1421             index = 0;
1422         }
1423     } while ((examined < htabslots) && (!qemu_file_rate_limit(f) || final));
1424 
1425     if (index >= htabslots) {
1426         assert(index == htabslots);
1427         index = 0;
1428     }
1429 
1430     spapr->htab_save_index = index;
1431 
1432     return (examined >= htabslots) && (sent == 0) ? 1 : 0;
1433 }
1434 
1435 #define MAX_ITERATION_NS    5000000 /* 5 ms */
1436 #define MAX_KVM_BUF_SIZE    2048
1437 
1438 static int htab_save_iterate(QEMUFile *f, void *opaque)
1439 {
1440     sPAPRMachineState *spapr = opaque;
1441     int fd;
1442     int rc = 0;
1443 
1444     /* Iteration header */
1445     qemu_put_be32(f, 0);
1446 
1447     if (!spapr->htab) {
1448         assert(kvm_enabled());
1449 
1450         fd = get_htab_fd(spapr);
1451         if (fd < 0) {
1452             return fd;
1453         }
1454 
1455         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, MAX_ITERATION_NS);
1456         if (rc < 0) {
1457             return rc;
1458         }
1459     } else  if (spapr->htab_first_pass) {
1460         htab_save_first_pass(f, spapr, MAX_ITERATION_NS);
1461     } else {
1462         rc = htab_save_later_pass(f, spapr, MAX_ITERATION_NS);
1463     }
1464 
1465     /* End marker */
1466     qemu_put_be32(f, 0);
1467     qemu_put_be16(f, 0);
1468     qemu_put_be16(f, 0);
1469 
1470     return rc;
1471 }
1472 
1473 static int htab_save_complete(QEMUFile *f, void *opaque)
1474 {
1475     sPAPRMachineState *spapr = opaque;
1476     int fd;
1477 
1478     /* Iteration header */
1479     qemu_put_be32(f, 0);
1480 
1481     if (!spapr->htab) {
1482         int rc;
1483 
1484         assert(kvm_enabled());
1485 
1486         fd = get_htab_fd(spapr);
1487         if (fd < 0) {
1488             return fd;
1489         }
1490 
1491         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, -1);
1492         if (rc < 0) {
1493             return rc;
1494         }
1495     } else {
1496         if (spapr->htab_first_pass) {
1497             htab_save_first_pass(f, spapr, -1);
1498         }
1499         htab_save_later_pass(f, spapr, -1);
1500     }
1501 
1502     /* End marker */
1503     qemu_put_be32(f, 0);
1504     qemu_put_be16(f, 0);
1505     qemu_put_be16(f, 0);
1506 
1507     return 0;
1508 }
1509 
1510 static int htab_load(QEMUFile *f, void *opaque, int version_id)
1511 {
1512     sPAPRMachineState *spapr = opaque;
1513     uint32_t section_hdr;
1514     int fd = -1;
1515 
1516     if (version_id < 1 || version_id > 1) {
1517         error_report("htab_load() bad version");
1518         return -EINVAL;
1519     }
1520 
1521     section_hdr = qemu_get_be32(f);
1522 
1523     if (section_hdr) {
1524         Error *local_err = NULL;
1525 
1526         /* First section gives the htab size */
1527         spapr_reallocate_hpt(spapr, section_hdr, &local_err);
1528         if (local_err) {
1529             error_report_err(local_err);
1530             return -EINVAL;
1531         }
1532         return 0;
1533     }
1534 
1535     if (!spapr->htab) {
1536         assert(kvm_enabled());
1537 
1538         fd = kvmppc_get_htab_fd(true);
1539         if (fd < 0) {
1540             error_report("Unable to open fd to restore KVM hash table: %s",
1541                          strerror(errno));
1542         }
1543     }
1544 
1545     while (true) {
1546         uint32_t index;
1547         uint16_t n_valid, n_invalid;
1548 
1549         index = qemu_get_be32(f);
1550         n_valid = qemu_get_be16(f);
1551         n_invalid = qemu_get_be16(f);
1552 
1553         if ((index == 0) && (n_valid == 0) && (n_invalid == 0)) {
1554             /* End of Stream */
1555             break;
1556         }
1557 
1558         if ((index + n_valid + n_invalid) >
1559             (HTAB_SIZE(spapr) / HASH_PTE_SIZE_64)) {
1560             /* Bad index in stream */
1561             error_report(
1562                 "htab_load() bad index %d (%hd+%hd entries) in htab stream (htab_shift=%d)",
1563                 index, n_valid, n_invalid, spapr->htab_shift);
1564             return -EINVAL;
1565         }
1566 
1567         if (spapr->htab) {
1568             if (n_valid) {
1569                 qemu_get_buffer(f, HPTE(spapr->htab, index),
1570                                 HASH_PTE_SIZE_64 * n_valid);
1571             }
1572             if (n_invalid) {
1573                 memset(HPTE(spapr->htab, index + n_valid), 0,
1574                        HASH_PTE_SIZE_64 * n_invalid);
1575             }
1576         } else {
1577             int rc;
1578 
1579             assert(fd >= 0);
1580 
1581             rc = kvmppc_load_htab_chunk(f, fd, index, n_valid, n_invalid);
1582             if (rc < 0) {
1583                 return rc;
1584             }
1585         }
1586     }
1587 
1588     if (!spapr->htab) {
1589         assert(fd >= 0);
1590         close(fd);
1591     }
1592 
1593     return 0;
1594 }
1595 
1596 static void htab_cleanup(void *opaque)
1597 {
1598     sPAPRMachineState *spapr = opaque;
1599 
1600     close_htab_fd(spapr);
1601 }
1602 
1603 static SaveVMHandlers savevm_htab_handlers = {
1604     .save_live_setup = htab_save_setup,
1605     .save_live_iterate = htab_save_iterate,
1606     .save_live_complete_precopy = htab_save_complete,
1607     .cleanup = htab_cleanup,
1608     .load_state = htab_load,
1609 };
1610 
1611 static void spapr_boot_set(void *opaque, const char *boot_device,
1612                            Error **errp)
1613 {
1614     MachineState *machine = MACHINE(qdev_get_machine());
1615     machine->boot_order = g_strdup(boot_device);
1616 }
1617 
1618 /*
1619  * Reset routine for LMB DR devices.
1620  *
1621  * Unlike PCI DR devices, LMB DR devices explicitly register this reset
1622  * routine. Reset for PCI DR devices will be handled by PHB reset routine
1623  * when it walks all its children devices. LMB devices reset occurs
1624  * as part of spapr_ppc_reset().
1625  */
1626 static void spapr_drc_reset(void *opaque)
1627 {
1628     sPAPRDRConnector *drc = opaque;
1629     DeviceState *d = DEVICE(drc);
1630 
1631     if (d) {
1632         device_reset(d);
1633     }
1634 }
1635 
1636 static void spapr_create_lmb_dr_connectors(sPAPRMachineState *spapr)
1637 {
1638     MachineState *machine = MACHINE(spapr);
1639     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
1640     uint32_t nr_lmbs = (machine->maxram_size - machine->ram_size)/lmb_size;
1641     int i;
1642 
1643     for (i = 0; i < nr_lmbs; i++) {
1644         sPAPRDRConnector *drc;
1645         uint64_t addr;
1646 
1647         addr = i * lmb_size + spapr->hotplug_memory.base;
1648         drc = spapr_dr_connector_new(OBJECT(spapr), SPAPR_DR_CONNECTOR_TYPE_LMB,
1649                                      addr/lmb_size);
1650         qemu_register_reset(spapr_drc_reset, drc);
1651     }
1652 }
1653 
1654 /*
1655  * If RAM size, maxmem size and individual node mem sizes aren't aligned
1656  * to SPAPR_MEMORY_BLOCK_SIZE(256MB), then refuse to start the guest
1657  * since we can't support such unaligned sizes with DRCONF_MEMORY.
1658  */
1659 static void spapr_validate_node_memory(MachineState *machine, Error **errp)
1660 {
1661     int i;
1662 
1663     if (machine->ram_size % SPAPR_MEMORY_BLOCK_SIZE) {
1664         error_setg(errp, "Memory size 0x" RAM_ADDR_FMT
1665                    " is not aligned to %llu MiB",
1666                    machine->ram_size,
1667                    SPAPR_MEMORY_BLOCK_SIZE / M_BYTE);
1668         return;
1669     }
1670 
1671     if (machine->maxram_size % SPAPR_MEMORY_BLOCK_SIZE) {
1672         error_setg(errp, "Maximum memory size 0x" RAM_ADDR_FMT
1673                    " is not aligned to %llu MiB",
1674                    machine->ram_size,
1675                    SPAPR_MEMORY_BLOCK_SIZE / M_BYTE);
1676         return;
1677     }
1678 
1679     for (i = 0; i < nb_numa_nodes; i++) {
1680         if (numa_info[i].node_mem % SPAPR_MEMORY_BLOCK_SIZE) {
1681             error_setg(errp,
1682                        "Node %d memory size 0x%" PRIx64
1683                        " is not aligned to %llu MiB",
1684                        i, numa_info[i].node_mem,
1685                        SPAPR_MEMORY_BLOCK_SIZE / M_BYTE);
1686             return;
1687         }
1688     }
1689 }
1690 
1691 /* pSeries LPAR / sPAPR hardware init */
1692 static void ppc_spapr_init(MachineState *machine)
1693 {
1694     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
1695     MachineClass *mc = MACHINE_GET_CLASS(machine);
1696     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1697     const char *kernel_filename = machine->kernel_filename;
1698     const char *initrd_filename = machine->initrd_filename;
1699     PCIHostState *phb;
1700     int i;
1701     MemoryRegion *sysmem = get_system_memory();
1702     MemoryRegion *ram = g_new(MemoryRegion, 1);
1703     MemoryRegion *rma_region;
1704     void *rma = NULL;
1705     hwaddr rma_alloc_size;
1706     hwaddr node0_size = spapr_node0_size();
1707     long load_limit, fw_size;
1708     char *filename;
1709     int smt = kvmppc_smt_threads();
1710     int spapr_cores = smp_cpus / smp_threads;
1711     int spapr_max_cores = max_cpus / smp_threads;
1712 
1713     if (mc->query_hotpluggable_cpus) {
1714         if (smp_cpus % smp_threads) {
1715             error_report("smp_cpus (%u) must be multiple of threads (%u)",
1716                          smp_cpus, smp_threads);
1717             exit(1);
1718         }
1719         if (max_cpus % smp_threads) {
1720             error_report("max_cpus (%u) must be multiple of threads (%u)",
1721                          max_cpus, smp_threads);
1722             exit(1);
1723         }
1724     }
1725 
1726     msi_nonbroken = true;
1727 
1728     QLIST_INIT(&spapr->phbs);
1729 
1730     cpu_ppc_hypercall = emulate_spapr_hypercall;
1731 
1732     /* Allocate RMA if necessary */
1733     rma_alloc_size = kvmppc_alloc_rma(&rma);
1734 
1735     if (rma_alloc_size == -1) {
1736         error_report("Unable to create RMA");
1737         exit(1);
1738     }
1739 
1740     if (rma_alloc_size && (rma_alloc_size < node0_size)) {
1741         spapr->rma_size = rma_alloc_size;
1742     } else {
1743         spapr->rma_size = node0_size;
1744 
1745         /* With KVM, we don't actually know whether KVM supports an
1746          * unbounded RMA (PR KVM) or is limited by the hash table size
1747          * (HV KVM using VRMA), so we always assume the latter
1748          *
1749          * In that case, we also limit the initial allocations for RTAS
1750          * etc... to 256M since we have no way to know what the VRMA size
1751          * is going to be as it depends on the size of the hash table
1752          * isn't determined yet.
1753          */
1754         if (kvm_enabled()) {
1755             spapr->vrma_adjust = 1;
1756             spapr->rma_size = MIN(spapr->rma_size, 0x10000000);
1757         }
1758 
1759         /* Actually we don't support unbounded RMA anymore since we
1760          * added proper emulation of HV mode. The max we can get is
1761          * 16G which also happens to be what we configure for PAPR
1762          * mode so make sure we don't do anything bigger than that
1763          */
1764         spapr->rma_size = MIN(spapr->rma_size, 0x400000000ull);
1765     }
1766 
1767     if (spapr->rma_size > node0_size) {
1768         error_report("Numa node 0 has to span the RMA (%#08"HWADDR_PRIx")",
1769                      spapr->rma_size);
1770         exit(1);
1771     }
1772 
1773     /* Setup a load limit for the ramdisk leaving room for SLOF and FDT */
1774     load_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR) - FW_OVERHEAD;
1775 
1776     /* Set up Interrupt Controller before we create the VCPUs */
1777     spapr->xics = xics_system_init(machine,
1778                                    DIV_ROUND_UP(max_cpus * smt, smp_threads),
1779                                    XICS_IRQS_SPAPR, &error_fatal);
1780 
1781     /* Set up containers for ibm,client-set-architecture negotiated options */
1782     spapr->ov5 = spapr_ovec_new();
1783     spapr->ov5_cas = spapr_ovec_new();
1784 
1785     if (smc->dr_lmb_enabled) {
1786         spapr_ovec_set(spapr->ov5, OV5_DRCONF_MEMORY);
1787         spapr_validate_node_memory(machine, &error_fatal);
1788     }
1789 
1790     spapr_ovec_set(spapr->ov5, OV5_FORM1_AFFINITY);
1791 
1792     /* advertise support for dedicated HP event source to guests */
1793     if (spapr->use_hotplug_event_source) {
1794         spapr_ovec_set(spapr->ov5, OV5_HP_EVT);
1795     }
1796 
1797     /* init CPUs */
1798     if (machine->cpu_model == NULL) {
1799         machine->cpu_model = kvm_enabled() ? "host" : smc->tcg_default_cpu;
1800     }
1801 
1802     ppc_cpu_parse_features(machine->cpu_model);
1803 
1804     if (mc->query_hotpluggable_cpus) {
1805         char *type = spapr_get_cpu_core_type(machine->cpu_model);
1806 
1807         if (type == NULL) {
1808             error_report("Unable to find sPAPR CPU Core definition");
1809             exit(1);
1810         }
1811 
1812         spapr->cores = g_new0(Object *, spapr_max_cores);
1813         for (i = 0; i < spapr_max_cores; i++) {
1814             int core_id = i * smp_threads;
1815             sPAPRDRConnector *drc =
1816                 spapr_dr_connector_new(OBJECT(spapr),
1817                                        SPAPR_DR_CONNECTOR_TYPE_CPU,
1818                                        (core_id / smp_threads) * smt);
1819 
1820             qemu_register_reset(spapr_drc_reset, drc);
1821 
1822             if (i < spapr_cores) {
1823                 Object *core  = object_new(type);
1824                 object_property_set_int(core, smp_threads, "nr-threads",
1825                                         &error_fatal);
1826                 object_property_set_int(core, core_id, CPU_CORE_PROP_CORE_ID,
1827                                         &error_fatal);
1828                 object_property_set_bool(core, true, "realized", &error_fatal);
1829             }
1830         }
1831         g_free(type);
1832     } else {
1833         for (i = 0; i < smp_cpus; i++) {
1834             PowerPCCPU *cpu = cpu_ppc_init(machine->cpu_model);
1835             if (cpu == NULL) {
1836                 error_report("Unable to find PowerPC CPU definition");
1837                 exit(1);
1838             }
1839             spapr_cpu_init(spapr, cpu, &error_fatal);
1840        }
1841     }
1842 
1843     if (kvm_enabled()) {
1844         /* Enable H_LOGICAL_CI_* so SLOF can talk to in-kernel devices */
1845         kvmppc_enable_logical_ci_hcalls();
1846         kvmppc_enable_set_mode_hcall();
1847 
1848         /* H_CLEAR_MOD/_REF are mandatory in PAPR, but off by default */
1849         kvmppc_enable_clear_ref_mod_hcalls();
1850     }
1851 
1852     /* allocate RAM */
1853     memory_region_allocate_system_memory(ram, NULL, "ppc_spapr.ram",
1854                                          machine->ram_size);
1855     memory_region_add_subregion(sysmem, 0, ram);
1856 
1857     if (rma_alloc_size && rma) {
1858         rma_region = g_new(MemoryRegion, 1);
1859         memory_region_init_ram_ptr(rma_region, NULL, "ppc_spapr.rma",
1860                                    rma_alloc_size, rma);
1861         vmstate_register_ram_global(rma_region);
1862         memory_region_add_subregion(sysmem, 0, rma_region);
1863     }
1864 
1865     /* initialize hotplug memory address space */
1866     if (machine->ram_size < machine->maxram_size) {
1867         ram_addr_t hotplug_mem_size = machine->maxram_size - machine->ram_size;
1868         /*
1869          * Limit the number of hotpluggable memory slots to half the number
1870          * slots that KVM supports, leaving the other half for PCI and other
1871          * devices. However ensure that number of slots doesn't drop below 32.
1872          */
1873         int max_memslots = kvm_enabled() ? kvm_get_max_memslots() / 2 :
1874                            SPAPR_MAX_RAM_SLOTS;
1875 
1876         if (max_memslots < SPAPR_MAX_RAM_SLOTS) {
1877             max_memslots = SPAPR_MAX_RAM_SLOTS;
1878         }
1879         if (machine->ram_slots > max_memslots) {
1880             error_report("Specified number of memory slots %"
1881                          PRIu64" exceeds max supported %d",
1882                          machine->ram_slots, max_memslots);
1883             exit(1);
1884         }
1885 
1886         spapr->hotplug_memory.base = ROUND_UP(machine->ram_size,
1887                                               SPAPR_HOTPLUG_MEM_ALIGN);
1888         memory_region_init(&spapr->hotplug_memory.mr, OBJECT(spapr),
1889                            "hotplug-memory", hotplug_mem_size);
1890         memory_region_add_subregion(sysmem, spapr->hotplug_memory.base,
1891                                     &spapr->hotplug_memory.mr);
1892     }
1893 
1894     if (smc->dr_lmb_enabled) {
1895         spapr_create_lmb_dr_connectors(spapr);
1896     }
1897 
1898     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, "spapr-rtas.bin");
1899     if (!filename) {
1900         error_report("Could not find LPAR rtas '%s'", "spapr-rtas.bin");
1901         exit(1);
1902     }
1903     spapr->rtas_size = get_image_size(filename);
1904     if (spapr->rtas_size < 0) {
1905         error_report("Could not get size of LPAR rtas '%s'", filename);
1906         exit(1);
1907     }
1908     spapr->rtas_blob = g_malloc(spapr->rtas_size);
1909     if (load_image_size(filename, spapr->rtas_blob, spapr->rtas_size) < 0) {
1910         error_report("Could not load LPAR rtas '%s'", filename);
1911         exit(1);
1912     }
1913     if (spapr->rtas_size > RTAS_MAX_SIZE) {
1914         error_report("RTAS too big ! 0x%zx bytes (max is 0x%x)",
1915                      (size_t)spapr->rtas_size, RTAS_MAX_SIZE);
1916         exit(1);
1917     }
1918     g_free(filename);
1919 
1920     /* Set up RTAS event infrastructure */
1921     spapr_events_init(spapr);
1922 
1923     /* Set up the RTC RTAS interfaces */
1924     spapr_rtc_create(spapr);
1925 
1926     /* Set up VIO bus */
1927     spapr->vio_bus = spapr_vio_bus_init();
1928 
1929     for (i = 0; i < MAX_SERIAL_PORTS; i++) {
1930         if (serial_hds[i]) {
1931             spapr_vty_create(spapr->vio_bus, serial_hds[i]);
1932         }
1933     }
1934 
1935     /* We always have at least the nvram device on VIO */
1936     spapr_create_nvram(spapr);
1937 
1938     /* Set up PCI */
1939     spapr_pci_rtas_init();
1940 
1941     phb = spapr_create_phb(spapr, 0);
1942 
1943     for (i = 0; i < nb_nics; i++) {
1944         NICInfo *nd = &nd_table[i];
1945 
1946         if (!nd->model) {
1947             nd->model = g_strdup("ibmveth");
1948         }
1949 
1950         if (strcmp(nd->model, "ibmveth") == 0) {
1951             spapr_vlan_create(spapr->vio_bus, nd);
1952         } else {
1953             pci_nic_init_nofail(&nd_table[i], phb->bus, nd->model, NULL);
1954         }
1955     }
1956 
1957     for (i = 0; i <= drive_get_max_bus(IF_SCSI); i++) {
1958         spapr_vscsi_create(spapr->vio_bus);
1959     }
1960 
1961     /* Graphics */
1962     if (spapr_vga_init(phb->bus, &error_fatal)) {
1963         spapr->has_graphics = true;
1964         machine->usb |= defaults_enabled() && !machine->usb_disabled;
1965     }
1966 
1967     if (machine->usb) {
1968         if (smc->use_ohci_by_default) {
1969             pci_create_simple(phb->bus, -1, "pci-ohci");
1970         } else {
1971             pci_create_simple(phb->bus, -1, "nec-usb-xhci");
1972         }
1973 
1974         if (spapr->has_graphics) {
1975             USBBus *usb_bus = usb_bus_find(-1);
1976 
1977             usb_create_simple(usb_bus, "usb-kbd");
1978             usb_create_simple(usb_bus, "usb-mouse");
1979         }
1980     }
1981 
1982     if (spapr->rma_size < (MIN_RMA_SLOF << 20)) {
1983         error_report(
1984             "pSeries SLOF firmware requires >= %ldM guest RMA (Real Mode Area memory)",
1985             MIN_RMA_SLOF);
1986         exit(1);
1987     }
1988 
1989     if (kernel_filename) {
1990         uint64_t lowaddr = 0;
1991 
1992         spapr->kernel_size = load_elf(kernel_filename, translate_kernel_address,
1993                                       NULL, NULL, &lowaddr, NULL, 1,
1994                                       PPC_ELF_MACHINE, 0, 0);
1995         if (spapr->kernel_size == ELF_LOAD_WRONG_ENDIAN) {
1996             spapr->kernel_size = load_elf(kernel_filename,
1997                                           translate_kernel_address, NULL, NULL,
1998                                           &lowaddr, NULL, 0, PPC_ELF_MACHINE,
1999                                           0, 0);
2000             spapr->kernel_le = spapr->kernel_size > 0;
2001         }
2002         if (spapr->kernel_size < 0) {
2003             error_report("error loading %s: %s", kernel_filename,
2004                          load_elf_strerror(spapr->kernel_size));
2005             exit(1);
2006         }
2007 
2008         /* load initrd */
2009         if (initrd_filename) {
2010             /* Try to locate the initrd in the gap between the kernel
2011              * and the firmware. Add a bit of space just in case
2012              */
2013             spapr->initrd_base = (KERNEL_LOAD_ADDR + spapr->kernel_size
2014                                   + 0x1ffff) & ~0xffff;
2015             spapr->initrd_size = load_image_targphys(initrd_filename,
2016                                                      spapr->initrd_base,
2017                                                      load_limit
2018                                                      - spapr->initrd_base);
2019             if (spapr->initrd_size < 0) {
2020                 error_report("could not load initial ram disk '%s'",
2021                              initrd_filename);
2022                 exit(1);
2023             }
2024         }
2025     }
2026 
2027     if (bios_name == NULL) {
2028         bios_name = FW_FILE_NAME;
2029     }
2030     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
2031     if (!filename) {
2032         error_report("Could not find LPAR firmware '%s'", bios_name);
2033         exit(1);
2034     }
2035     fw_size = load_image_targphys(filename, 0, FW_MAX_SIZE);
2036     if (fw_size <= 0) {
2037         error_report("Could not load LPAR firmware '%s'", filename);
2038         exit(1);
2039     }
2040     g_free(filename);
2041 
2042     /* FIXME: Should register things through the MachineState's qdev
2043      * interface, this is a legacy from the sPAPREnvironment structure
2044      * which predated MachineState but had a similar function */
2045     vmstate_register(NULL, 0, &vmstate_spapr, spapr);
2046     register_savevm_live(NULL, "spapr/htab", -1, 1,
2047                          &savevm_htab_handlers, spapr);
2048 
2049     /* used by RTAS */
2050     QTAILQ_INIT(&spapr->ccs_list);
2051     qemu_register_reset(spapr_ccs_reset_hook, spapr);
2052 
2053     qemu_register_boot_set(spapr_boot_set, spapr);
2054 }
2055 
2056 static int spapr_kvm_type(const char *vm_type)
2057 {
2058     if (!vm_type) {
2059         return 0;
2060     }
2061 
2062     if (!strcmp(vm_type, "HV")) {
2063         return 1;
2064     }
2065 
2066     if (!strcmp(vm_type, "PR")) {
2067         return 2;
2068     }
2069 
2070     error_report("Unknown kvm-type specified '%s'", vm_type);
2071     exit(1);
2072 }
2073 
2074 /*
2075  * Implementation of an interface to adjust firmware path
2076  * for the bootindex property handling.
2077  */
2078 static char *spapr_get_fw_dev_path(FWPathProvider *p, BusState *bus,
2079                                    DeviceState *dev)
2080 {
2081 #define CAST(type, obj, name) \
2082     ((type *)object_dynamic_cast(OBJECT(obj), (name)))
2083     SCSIDevice *d = CAST(SCSIDevice,  dev, TYPE_SCSI_DEVICE);
2084     sPAPRPHBState *phb = CAST(sPAPRPHBState, dev, TYPE_SPAPR_PCI_HOST_BRIDGE);
2085 
2086     if (d) {
2087         void *spapr = CAST(void, bus->parent, "spapr-vscsi");
2088         VirtIOSCSI *virtio = CAST(VirtIOSCSI, bus->parent, TYPE_VIRTIO_SCSI);
2089         USBDevice *usb = CAST(USBDevice, bus->parent, TYPE_USB_DEVICE);
2090 
2091         if (spapr) {
2092             /*
2093              * Replace "channel@0/disk@0,0" with "disk@8000000000000000":
2094              * We use SRP luns of the form 8000 | (bus << 8) | (id << 5) | lun
2095              * in the top 16 bits of the 64-bit LUN
2096              */
2097             unsigned id = 0x8000 | (d->id << 8) | d->lun;
2098             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
2099                                    (uint64_t)id << 48);
2100         } else if (virtio) {
2101             /*
2102              * We use SRP luns of the form 01000000 | (target << 8) | lun
2103              * in the top 32 bits of the 64-bit LUN
2104              * Note: the quote above is from SLOF and it is wrong,
2105              * the actual binding is:
2106              * swap 0100 or 10 << or 20 << ( target lun-id -- srplun )
2107              */
2108             unsigned id = 0x1000000 | (d->id << 16) | d->lun;
2109             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
2110                                    (uint64_t)id << 32);
2111         } else if (usb) {
2112             /*
2113              * We use SRP luns of the form 01000000 | (usb-port << 16) | lun
2114              * in the top 32 bits of the 64-bit LUN
2115              */
2116             unsigned usb_port = atoi(usb->port->path);
2117             unsigned id = 0x1000000 | (usb_port << 16) | d->lun;
2118             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
2119                                    (uint64_t)id << 32);
2120         }
2121     }
2122 
2123     if (phb) {
2124         /* Replace "pci" with "pci@800000020000000" */
2125         return g_strdup_printf("pci@%"PRIX64, phb->buid);
2126     }
2127 
2128     return NULL;
2129 }
2130 
2131 static char *spapr_get_kvm_type(Object *obj, Error **errp)
2132 {
2133     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2134 
2135     return g_strdup(spapr->kvm_type);
2136 }
2137 
2138 static void spapr_set_kvm_type(Object *obj, const char *value, Error **errp)
2139 {
2140     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2141 
2142     g_free(spapr->kvm_type);
2143     spapr->kvm_type = g_strdup(value);
2144 }
2145 
2146 static bool spapr_get_modern_hotplug_events(Object *obj, Error **errp)
2147 {
2148     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2149 
2150     return spapr->use_hotplug_event_source;
2151 }
2152 
2153 static void spapr_set_modern_hotplug_events(Object *obj, bool value,
2154                                             Error **errp)
2155 {
2156     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2157 
2158     spapr->use_hotplug_event_source = value;
2159 }
2160 
2161 static void spapr_machine_initfn(Object *obj)
2162 {
2163     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2164 
2165     spapr->htab_fd = -1;
2166     spapr->use_hotplug_event_source = true;
2167     object_property_add_str(obj, "kvm-type",
2168                             spapr_get_kvm_type, spapr_set_kvm_type, NULL);
2169     object_property_set_description(obj, "kvm-type",
2170                                     "Specifies the KVM virtualization mode (HV, PR)",
2171                                     NULL);
2172     object_property_add_bool(obj, "modern-hotplug-events",
2173                             spapr_get_modern_hotplug_events,
2174                             spapr_set_modern_hotplug_events,
2175                             NULL);
2176     object_property_set_description(obj, "modern-hotplug-events",
2177                                     "Use dedicated hotplug event mechanism in"
2178                                     " place of standard EPOW events when possible"
2179                                     " (required for memory hot-unplug support)",
2180                                     NULL);
2181 }
2182 
2183 static void spapr_machine_finalizefn(Object *obj)
2184 {
2185     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2186 
2187     g_free(spapr->kvm_type);
2188 }
2189 
2190 static void ppc_cpu_do_nmi_on_cpu(CPUState *cs, void *arg)
2191 {
2192     cpu_synchronize_state(cs);
2193     ppc_cpu_do_system_reset(cs);
2194 }
2195 
2196 static void spapr_nmi(NMIState *n, int cpu_index, Error **errp)
2197 {
2198     CPUState *cs;
2199 
2200     CPU_FOREACH(cs) {
2201         async_run_on_cpu(cs, ppc_cpu_do_nmi_on_cpu, NULL);
2202     }
2203 }
2204 
2205 static void spapr_add_lmbs(DeviceState *dev, uint64_t addr_start, uint64_t size,
2206                            uint32_t node, bool dedicated_hp_event_source,
2207                            Error **errp)
2208 {
2209     sPAPRDRConnector *drc;
2210     sPAPRDRConnectorClass *drck;
2211     uint32_t nr_lmbs = size/SPAPR_MEMORY_BLOCK_SIZE;
2212     int i, fdt_offset, fdt_size;
2213     void *fdt;
2214     uint64_t addr = addr_start;
2215 
2216     for (i = 0; i < nr_lmbs; i++) {
2217         drc = spapr_dr_connector_by_id(SPAPR_DR_CONNECTOR_TYPE_LMB,
2218                 addr/SPAPR_MEMORY_BLOCK_SIZE);
2219         g_assert(drc);
2220 
2221         fdt = create_device_tree(&fdt_size);
2222         fdt_offset = spapr_populate_memory_node(fdt, node, addr,
2223                                                 SPAPR_MEMORY_BLOCK_SIZE);
2224 
2225         drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
2226         drck->attach(drc, dev, fdt, fdt_offset, !dev->hotplugged, errp);
2227         addr += SPAPR_MEMORY_BLOCK_SIZE;
2228     }
2229     /* send hotplug notification to the
2230      * guest only in case of hotplugged memory
2231      */
2232     if (dev->hotplugged) {
2233         if (dedicated_hp_event_source) {
2234             drc = spapr_dr_connector_by_id(SPAPR_DR_CONNECTOR_TYPE_LMB,
2235                     addr_start / SPAPR_MEMORY_BLOCK_SIZE);
2236             drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
2237             spapr_hotplug_req_add_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
2238                                                    nr_lmbs,
2239                                                    drck->get_index(drc));
2240         } else {
2241             spapr_hotplug_req_add_by_count(SPAPR_DR_CONNECTOR_TYPE_LMB,
2242                                            nr_lmbs);
2243         }
2244     }
2245 }
2246 
2247 static void spapr_memory_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
2248                               uint32_t node, Error **errp)
2249 {
2250     Error *local_err = NULL;
2251     sPAPRMachineState *ms = SPAPR_MACHINE(hotplug_dev);
2252     PCDIMMDevice *dimm = PC_DIMM(dev);
2253     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
2254     MemoryRegion *mr = ddc->get_memory_region(dimm);
2255     uint64_t align = memory_region_get_alignment(mr);
2256     uint64_t size = memory_region_size(mr);
2257     uint64_t addr;
2258 
2259     if (size % SPAPR_MEMORY_BLOCK_SIZE) {
2260         error_setg(&local_err, "Hotplugged memory size must be a multiple of "
2261                       "%lld MB", SPAPR_MEMORY_BLOCK_SIZE/M_BYTE);
2262         goto out;
2263     }
2264 
2265     pc_dimm_memory_plug(dev, &ms->hotplug_memory, mr, align, &local_err);
2266     if (local_err) {
2267         goto out;
2268     }
2269 
2270     addr = object_property_get_int(OBJECT(dimm), PC_DIMM_ADDR_PROP, &local_err);
2271     if (local_err) {
2272         pc_dimm_memory_unplug(dev, &ms->hotplug_memory, mr);
2273         goto out;
2274     }
2275 
2276     spapr_add_lmbs(dev, addr, size, node,
2277                    spapr_ovec_test(ms->ov5_cas, OV5_HP_EVT),
2278                    &error_abort);
2279 
2280 out:
2281     error_propagate(errp, local_err);
2282 }
2283 
2284 typedef struct sPAPRDIMMState {
2285     uint32_t nr_lmbs;
2286 } sPAPRDIMMState;
2287 
2288 static void spapr_lmb_release(DeviceState *dev, void *opaque)
2289 {
2290     sPAPRDIMMState *ds = (sPAPRDIMMState *)opaque;
2291     HotplugHandler *hotplug_ctrl;
2292 
2293     if (--ds->nr_lmbs) {
2294         return;
2295     }
2296 
2297     g_free(ds);
2298 
2299     /*
2300      * Now that all the LMBs have been removed by the guest, call the
2301      * pc-dimm unplug handler to cleanup up the pc-dimm device.
2302      */
2303     hotplug_ctrl = qdev_get_hotplug_handler(dev);
2304     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
2305 }
2306 
2307 static void spapr_del_lmbs(DeviceState *dev, uint64_t addr_start, uint64_t size,
2308                            Error **errp)
2309 {
2310     sPAPRDRConnector *drc;
2311     sPAPRDRConnectorClass *drck;
2312     uint32_t nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
2313     int i;
2314     sPAPRDIMMState *ds = g_malloc0(sizeof(sPAPRDIMMState));
2315     uint64_t addr = addr_start;
2316 
2317     ds->nr_lmbs = nr_lmbs;
2318     for (i = 0; i < nr_lmbs; i++) {
2319         drc = spapr_dr_connector_by_id(SPAPR_DR_CONNECTOR_TYPE_LMB,
2320                 addr / SPAPR_MEMORY_BLOCK_SIZE);
2321         g_assert(drc);
2322 
2323         drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
2324         drck->detach(drc, dev, spapr_lmb_release, ds, errp);
2325         addr += SPAPR_MEMORY_BLOCK_SIZE;
2326     }
2327 
2328     drc = spapr_dr_connector_by_id(SPAPR_DR_CONNECTOR_TYPE_LMB,
2329                                    addr_start / SPAPR_MEMORY_BLOCK_SIZE);
2330     drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
2331     spapr_hotplug_req_remove_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
2332                                               nr_lmbs,
2333                                               drck->get_index(drc));
2334 }
2335 
2336 static void spapr_memory_unplug(HotplugHandler *hotplug_dev, DeviceState *dev,
2337                                 Error **errp)
2338 {
2339     sPAPRMachineState *ms = SPAPR_MACHINE(hotplug_dev);
2340     PCDIMMDevice *dimm = PC_DIMM(dev);
2341     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
2342     MemoryRegion *mr = ddc->get_memory_region(dimm);
2343 
2344     pc_dimm_memory_unplug(dev, &ms->hotplug_memory, mr);
2345     object_unparent(OBJECT(dev));
2346 }
2347 
2348 static void spapr_memory_unplug_request(HotplugHandler *hotplug_dev,
2349                                         DeviceState *dev, Error **errp)
2350 {
2351     Error *local_err = NULL;
2352     PCDIMMDevice *dimm = PC_DIMM(dev);
2353     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
2354     MemoryRegion *mr = ddc->get_memory_region(dimm);
2355     uint64_t size = memory_region_size(mr);
2356     uint64_t addr;
2357 
2358     addr = object_property_get_int(OBJECT(dimm), PC_DIMM_ADDR_PROP, &local_err);
2359     if (local_err) {
2360         goto out;
2361     }
2362 
2363     spapr_del_lmbs(dev, addr, size, &error_abort);
2364 out:
2365     error_propagate(errp, local_err);
2366 }
2367 
2368 void *spapr_populate_hotplug_cpu_dt(CPUState *cs, int *fdt_offset,
2369                                     sPAPRMachineState *spapr)
2370 {
2371     PowerPCCPU *cpu = POWERPC_CPU(cs);
2372     DeviceClass *dc = DEVICE_GET_CLASS(cs);
2373     int id = ppc_get_vcpu_dt_id(cpu);
2374     void *fdt;
2375     int offset, fdt_size;
2376     char *nodename;
2377 
2378     fdt = create_device_tree(&fdt_size);
2379     nodename = g_strdup_printf("%s@%x", dc->fw_name, id);
2380     offset = fdt_add_subnode(fdt, 0, nodename);
2381 
2382     spapr_populate_cpu_dt(cs, fdt, offset, spapr);
2383     g_free(nodename);
2384 
2385     *fdt_offset = offset;
2386     return fdt;
2387 }
2388 
2389 static void spapr_machine_device_plug(HotplugHandler *hotplug_dev,
2390                                       DeviceState *dev, Error **errp)
2391 {
2392     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(qdev_get_machine());
2393 
2394     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
2395         int node;
2396 
2397         if (!smc->dr_lmb_enabled) {
2398             error_setg(errp, "Memory hotplug not supported for this machine");
2399             return;
2400         }
2401         node = object_property_get_int(OBJECT(dev), PC_DIMM_NODE_PROP, errp);
2402         if (*errp) {
2403             return;
2404         }
2405         if (node < 0 || node >= MAX_NODES) {
2406             error_setg(errp, "Invaild node %d", node);
2407             return;
2408         }
2409 
2410         /*
2411          * Currently PowerPC kernel doesn't allow hot-adding memory to
2412          * memory-less node, but instead will silently add the memory
2413          * to the first node that has some memory. This causes two
2414          * unexpected behaviours for the user.
2415          *
2416          * - Memory gets hotplugged to a different node than what the user
2417          *   specified.
2418          * - Since pc-dimm subsystem in QEMU still thinks that memory belongs
2419          *   to memory-less node, a reboot will set things accordingly
2420          *   and the previously hotplugged memory now ends in the right node.
2421          *   This appears as if some memory moved from one node to another.
2422          *
2423          * So until kernel starts supporting memory hotplug to memory-less
2424          * nodes, just prevent such attempts upfront in QEMU.
2425          */
2426         if (nb_numa_nodes && !numa_info[node].node_mem) {
2427             error_setg(errp, "Can't hotplug memory to memory-less node %d",
2428                        node);
2429             return;
2430         }
2431 
2432         spapr_memory_plug(hotplug_dev, dev, node, errp);
2433     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
2434         spapr_core_plug(hotplug_dev, dev, errp);
2435     }
2436 }
2437 
2438 static void spapr_machine_device_unplug(HotplugHandler *hotplug_dev,
2439                                       DeviceState *dev, Error **errp)
2440 {
2441     sPAPRMachineState *sms = SPAPR_MACHINE(qdev_get_machine());
2442     MachineClass *mc = MACHINE_GET_CLASS(qdev_get_machine());
2443 
2444     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
2445         if (spapr_ovec_test(sms->ov5_cas, OV5_HP_EVT)) {
2446             spapr_memory_unplug(hotplug_dev, dev, errp);
2447         } else {
2448             error_setg(errp, "Memory hot unplug not supported for this guest");
2449         }
2450     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
2451         if (!mc->query_hotpluggable_cpus) {
2452             error_setg(errp, "CPU hot unplug not supported on this machine");
2453             return;
2454         }
2455         spapr_core_unplug(hotplug_dev, dev, errp);
2456     }
2457 }
2458 
2459 static void spapr_machine_device_unplug_request(HotplugHandler *hotplug_dev,
2460                                                 DeviceState *dev, Error **errp)
2461 {
2462     sPAPRMachineState *sms = SPAPR_MACHINE(qdev_get_machine());
2463     MachineClass *mc = MACHINE_GET_CLASS(qdev_get_machine());
2464 
2465     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
2466         if (spapr_ovec_test(sms->ov5_cas, OV5_HP_EVT)) {
2467             spapr_memory_unplug_request(hotplug_dev, dev, errp);
2468         } else {
2469             /* NOTE: this means there is a window after guest reset, prior to
2470              * CAS negotiation, where unplug requests will fail due to the
2471              * capability not being detected yet. This is a bit different than
2472              * the case with PCI unplug, where the events will be queued and
2473              * eventually handled by the guest after boot
2474              */
2475             error_setg(errp, "Memory hot unplug not supported for this guest");
2476         }
2477     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
2478         if (!mc->query_hotpluggable_cpus) {
2479             error_setg(errp, "CPU hot unplug not supported on this machine");
2480             return;
2481         }
2482         spapr_core_unplug(hotplug_dev, dev, errp);
2483     }
2484 }
2485 
2486 static void spapr_machine_device_pre_plug(HotplugHandler *hotplug_dev,
2487                                           DeviceState *dev, Error **errp)
2488 {
2489     if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
2490         spapr_core_pre_plug(hotplug_dev, dev, errp);
2491     }
2492 }
2493 
2494 static HotplugHandler *spapr_get_hotplug_handler(MachineState *machine,
2495                                                  DeviceState *dev)
2496 {
2497     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM) ||
2498         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
2499         return HOTPLUG_HANDLER(machine);
2500     }
2501     return NULL;
2502 }
2503 
2504 static unsigned spapr_cpu_index_to_socket_id(unsigned cpu_index)
2505 {
2506     /* Allocate to NUMA nodes on a "socket" basis (not that concept of
2507      * socket means much for the paravirtualized PAPR platform) */
2508     return cpu_index / smp_threads / smp_cores;
2509 }
2510 
2511 static HotpluggableCPUList *spapr_query_hotpluggable_cpus(MachineState *machine)
2512 {
2513     int i;
2514     HotpluggableCPUList *head = NULL;
2515     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
2516     int spapr_max_cores = max_cpus / smp_threads;
2517 
2518     for (i = 0; i < spapr_max_cores; i++) {
2519         HotpluggableCPUList *list_item = g_new0(typeof(*list_item), 1);
2520         HotpluggableCPU *cpu_item = g_new0(typeof(*cpu_item), 1);
2521         CpuInstanceProperties *cpu_props = g_new0(typeof(*cpu_props), 1);
2522 
2523         cpu_item->type = spapr_get_cpu_core_type(machine->cpu_model);
2524         cpu_item->vcpus_count = smp_threads;
2525         cpu_props->has_core_id = true;
2526         cpu_props->core_id = i * smp_threads;
2527         /* TODO: add 'has_node/node' here to describe
2528            to which node core belongs */
2529 
2530         cpu_item->props = cpu_props;
2531         if (spapr->cores[i]) {
2532             cpu_item->has_qom_path = true;
2533             cpu_item->qom_path = object_get_canonical_path(spapr->cores[i]);
2534         }
2535         list_item->value = cpu_item;
2536         list_item->next = head;
2537         head = list_item;
2538     }
2539     return head;
2540 }
2541 
2542 static void spapr_phb_placement(sPAPRMachineState *spapr, uint32_t index,
2543                                 uint64_t *buid, hwaddr *pio,
2544                                 hwaddr *mmio32, hwaddr *mmio64,
2545                                 unsigned n_dma, uint32_t *liobns, Error **errp)
2546 {
2547     /*
2548      * New-style PHB window placement.
2549      *
2550      * Goals: Gives large (1TiB), naturally aligned 64-bit MMIO window
2551      * for each PHB, in addition to 2GiB 32-bit MMIO and 64kiB PIO
2552      * windows.
2553      *
2554      * Some guest kernels can't work with MMIO windows above 1<<46
2555      * (64TiB), so we place up to 31 PHBs in the area 32TiB..64TiB
2556      *
2557      * 32TiB..(33TiB+1984kiB) contains the 64kiB PIO windows for each
2558      * PHB stacked together.  (32TiB+2GiB)..(32TiB+64GiB) contains the
2559      * 2GiB 32-bit MMIO windows for each PHB.  Then 33..64TiB has the
2560      * 1TiB 64-bit MMIO windows for each PHB.
2561      */
2562     const uint64_t base_buid = 0x800000020000000ULL;
2563     const int max_phbs =
2564         (SPAPR_PCI_LIMIT - SPAPR_PCI_BASE) / SPAPR_PCI_MEM64_WIN_SIZE - 1;
2565     int i;
2566 
2567     /* Sanity check natural alignments */
2568     QEMU_BUILD_BUG_ON((SPAPR_PCI_BASE % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
2569     QEMU_BUILD_BUG_ON((SPAPR_PCI_LIMIT % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
2570     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM64_WIN_SIZE % SPAPR_PCI_MEM32_WIN_SIZE) != 0);
2571     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM32_WIN_SIZE % SPAPR_PCI_IO_WIN_SIZE) != 0);
2572     /* Sanity check bounds */
2573     QEMU_BUILD_BUG_ON((max_phbs * SPAPR_PCI_IO_WIN_SIZE) > SPAPR_PCI_MEM32_WIN_SIZE);
2574     QEMU_BUILD_BUG_ON((max_phbs * SPAPR_PCI_MEM32_WIN_SIZE) > SPAPR_PCI_MEM64_WIN_SIZE);
2575 
2576     if (index >= max_phbs) {
2577         error_setg(errp, "\"index\" for PAPR PHB is too large (max %u)",
2578                    max_phbs - 1);
2579         return;
2580     }
2581 
2582     *buid = base_buid + index;
2583     for (i = 0; i < n_dma; ++i) {
2584         liobns[i] = SPAPR_PCI_LIOBN(index, i);
2585     }
2586 
2587     *pio = SPAPR_PCI_BASE + index * SPAPR_PCI_IO_WIN_SIZE;
2588     *mmio32 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM32_WIN_SIZE;
2589     *mmio64 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM64_WIN_SIZE;
2590 }
2591 
2592 static void spapr_machine_class_init(ObjectClass *oc, void *data)
2593 {
2594     MachineClass *mc = MACHINE_CLASS(oc);
2595     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(oc);
2596     FWPathProviderClass *fwc = FW_PATH_PROVIDER_CLASS(oc);
2597     NMIClass *nc = NMI_CLASS(oc);
2598     HotplugHandlerClass *hc = HOTPLUG_HANDLER_CLASS(oc);
2599 
2600     mc->desc = "pSeries Logical Partition (PAPR compliant)";
2601 
2602     /*
2603      * We set up the default / latest behaviour here.  The class_init
2604      * functions for the specific versioned machine types can override
2605      * these details for backwards compatibility
2606      */
2607     mc->init = ppc_spapr_init;
2608     mc->reset = ppc_spapr_reset;
2609     mc->block_default_type = IF_SCSI;
2610     mc->max_cpus = 255;
2611     mc->no_parallel = 1;
2612     mc->default_boot_order = "";
2613     mc->default_ram_size = 512 * M_BYTE;
2614     mc->kvm_type = spapr_kvm_type;
2615     mc->has_dynamic_sysbus = true;
2616     mc->pci_allow_0_address = true;
2617     mc->get_hotplug_handler = spapr_get_hotplug_handler;
2618     hc->pre_plug = spapr_machine_device_pre_plug;
2619     hc->plug = spapr_machine_device_plug;
2620     hc->unplug = spapr_machine_device_unplug;
2621     mc->cpu_index_to_socket_id = spapr_cpu_index_to_socket_id;
2622     hc->unplug_request = spapr_machine_device_unplug_request;
2623 
2624     smc->dr_lmb_enabled = true;
2625     smc->tcg_default_cpu = "POWER8";
2626     mc->query_hotpluggable_cpus = spapr_query_hotpluggable_cpus;
2627     fwc->get_dev_path = spapr_get_fw_dev_path;
2628     nc->nmi_monitor_handler = spapr_nmi;
2629     smc->phb_placement = spapr_phb_placement;
2630 }
2631 
2632 static const TypeInfo spapr_machine_info = {
2633     .name          = TYPE_SPAPR_MACHINE,
2634     .parent        = TYPE_MACHINE,
2635     .abstract      = true,
2636     .instance_size = sizeof(sPAPRMachineState),
2637     .instance_init = spapr_machine_initfn,
2638     .instance_finalize = spapr_machine_finalizefn,
2639     .class_size    = sizeof(sPAPRMachineClass),
2640     .class_init    = spapr_machine_class_init,
2641     .interfaces = (InterfaceInfo[]) {
2642         { TYPE_FW_PATH_PROVIDER },
2643         { TYPE_NMI },
2644         { TYPE_HOTPLUG_HANDLER },
2645         { }
2646     },
2647 };
2648 
2649 #define DEFINE_SPAPR_MACHINE(suffix, verstr, latest)                 \
2650     static void spapr_machine_##suffix##_class_init(ObjectClass *oc, \
2651                                                     void *data)      \
2652     {                                                                \
2653         MachineClass *mc = MACHINE_CLASS(oc);                        \
2654         spapr_machine_##suffix##_class_options(mc);                  \
2655         if (latest) {                                                \
2656             mc->alias = "pseries";                                   \
2657             mc->is_default = 1;                                      \
2658         }                                                            \
2659     }                                                                \
2660     static void spapr_machine_##suffix##_instance_init(Object *obj)  \
2661     {                                                                \
2662         MachineState *machine = MACHINE(obj);                        \
2663         spapr_machine_##suffix##_instance_options(machine);          \
2664     }                                                                \
2665     static const TypeInfo spapr_machine_##suffix##_info = {          \
2666         .name = MACHINE_TYPE_NAME("pseries-" verstr),                \
2667         .parent = TYPE_SPAPR_MACHINE,                                \
2668         .class_init = spapr_machine_##suffix##_class_init,           \
2669         .instance_init = spapr_machine_##suffix##_instance_init,     \
2670     };                                                               \
2671     static void spapr_machine_register_##suffix(void)                \
2672     {                                                                \
2673         type_register(&spapr_machine_##suffix##_info);               \
2674     }                                                                \
2675     type_init(spapr_machine_register_##suffix)
2676 
2677 /*
2678  * pseries-2.8
2679  */
2680 static void spapr_machine_2_8_instance_options(MachineState *machine)
2681 {
2682 }
2683 
2684 static void spapr_machine_2_8_class_options(MachineClass *mc)
2685 {
2686     /* Defaults for the latest behaviour inherited from the base class */
2687 }
2688 
2689 DEFINE_SPAPR_MACHINE(2_8, "2.8", true);
2690 
2691 /*
2692  * pseries-2.7
2693  */
2694 #define SPAPR_COMPAT_2_7                            \
2695     HW_COMPAT_2_7                                   \
2696     {                                               \
2697         .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,     \
2698         .property = "mem_win_size",                 \
2699         .value    = stringify(SPAPR_PCI_2_7_MMIO_WIN_SIZE),\
2700     },                                              \
2701     {                                               \
2702         .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,     \
2703         .property = "mem64_win_size",               \
2704         .value    = "0",                            \
2705     },
2706 
2707 static void phb_placement_2_7(sPAPRMachineState *spapr, uint32_t index,
2708                               uint64_t *buid, hwaddr *pio,
2709                               hwaddr *mmio32, hwaddr *mmio64,
2710                               unsigned n_dma, uint32_t *liobns, Error **errp)
2711 {
2712     /* Legacy PHB placement for pseries-2.7 and earlier machine types */
2713     const uint64_t base_buid = 0x800000020000000ULL;
2714     const hwaddr phb_spacing = 0x1000000000ULL; /* 64 GiB */
2715     const hwaddr mmio_offset = 0xa0000000; /* 2 GiB + 512 MiB */
2716     const hwaddr pio_offset = 0x80000000; /* 2 GiB */
2717     const uint32_t max_index = 255;
2718     const hwaddr phb0_alignment = 0x10000000000ULL; /* 1 TiB */
2719 
2720     uint64_t ram_top = MACHINE(spapr)->ram_size;
2721     hwaddr phb0_base, phb_base;
2722     int i;
2723 
2724     /* Do we have hotpluggable memory? */
2725     if (MACHINE(spapr)->maxram_size > ram_top) {
2726         /* Can't just use maxram_size, because there may be an
2727          * alignment gap between normal and hotpluggable memory
2728          * regions */
2729         ram_top = spapr->hotplug_memory.base +
2730             memory_region_size(&spapr->hotplug_memory.mr);
2731     }
2732 
2733     phb0_base = QEMU_ALIGN_UP(ram_top, phb0_alignment);
2734 
2735     if (index > max_index) {
2736         error_setg(errp, "\"index\" for PAPR PHB is too large (max %u)",
2737                    max_index);
2738         return;
2739     }
2740 
2741     *buid = base_buid + index;
2742     for (i = 0; i < n_dma; ++i) {
2743         liobns[i] = SPAPR_PCI_LIOBN(index, i);
2744     }
2745 
2746     phb_base = phb0_base + index * phb_spacing;
2747     *pio = phb_base + pio_offset;
2748     *mmio32 = phb_base + mmio_offset;
2749     /*
2750      * We don't set the 64-bit MMIO window, relying on the PHB's
2751      * fallback behaviour of automatically splitting a large "32-bit"
2752      * window into contiguous 32-bit and 64-bit windows
2753      */
2754 }
2755 
2756 static void spapr_machine_2_7_instance_options(MachineState *machine)
2757 {
2758     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
2759 
2760     spapr_machine_2_8_instance_options(machine);
2761     spapr->use_hotplug_event_source = false;
2762 }
2763 
2764 static void spapr_machine_2_7_class_options(MachineClass *mc)
2765 {
2766     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
2767 
2768     spapr_machine_2_8_class_options(mc);
2769     smc->tcg_default_cpu = "POWER7";
2770     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_7);
2771     smc->phb_placement = phb_placement_2_7;
2772 }
2773 
2774 DEFINE_SPAPR_MACHINE(2_7, "2.7", false);
2775 
2776 /*
2777  * pseries-2.6
2778  */
2779 #define SPAPR_COMPAT_2_6 \
2780     HW_COMPAT_2_6 \
2781     { \
2782         .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,\
2783         .property = "ddw",\
2784         .value    = stringify(off),\
2785     },
2786 
2787 static void spapr_machine_2_6_instance_options(MachineState *machine)
2788 {
2789     spapr_machine_2_7_instance_options(machine);
2790 }
2791 
2792 static void spapr_machine_2_6_class_options(MachineClass *mc)
2793 {
2794     spapr_machine_2_7_class_options(mc);
2795     mc->query_hotpluggable_cpus = NULL;
2796     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_6);
2797 }
2798 
2799 DEFINE_SPAPR_MACHINE(2_6, "2.6", false);
2800 
2801 /*
2802  * pseries-2.5
2803  */
2804 #define SPAPR_COMPAT_2_5 \
2805     HW_COMPAT_2_5 \
2806     { \
2807         .driver   = "spapr-vlan", \
2808         .property = "use-rx-buffer-pools", \
2809         .value    = "off", \
2810     },
2811 
2812 static void spapr_machine_2_5_instance_options(MachineState *machine)
2813 {
2814     spapr_machine_2_6_instance_options(machine);
2815 }
2816 
2817 static void spapr_machine_2_5_class_options(MachineClass *mc)
2818 {
2819     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
2820 
2821     spapr_machine_2_6_class_options(mc);
2822     smc->use_ohci_by_default = true;
2823     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_5);
2824 }
2825 
2826 DEFINE_SPAPR_MACHINE(2_5, "2.5", false);
2827 
2828 /*
2829  * pseries-2.4
2830  */
2831 #define SPAPR_COMPAT_2_4 \
2832         HW_COMPAT_2_4
2833 
2834 static void spapr_machine_2_4_instance_options(MachineState *machine)
2835 {
2836     spapr_machine_2_5_instance_options(machine);
2837 }
2838 
2839 static void spapr_machine_2_4_class_options(MachineClass *mc)
2840 {
2841     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
2842 
2843     spapr_machine_2_5_class_options(mc);
2844     smc->dr_lmb_enabled = false;
2845     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_4);
2846 }
2847 
2848 DEFINE_SPAPR_MACHINE(2_4, "2.4", false);
2849 
2850 /*
2851  * pseries-2.3
2852  */
2853 #define SPAPR_COMPAT_2_3 \
2854         HW_COMPAT_2_3 \
2855         {\
2856             .driver   = "spapr-pci-host-bridge",\
2857             .property = "dynamic-reconfiguration",\
2858             .value    = "off",\
2859         },
2860 
2861 static void spapr_machine_2_3_instance_options(MachineState *machine)
2862 {
2863     spapr_machine_2_4_instance_options(machine);
2864     savevm_skip_section_footers();
2865     global_state_set_optional();
2866     savevm_skip_configuration();
2867 }
2868 
2869 static void spapr_machine_2_3_class_options(MachineClass *mc)
2870 {
2871     spapr_machine_2_4_class_options(mc);
2872     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_3);
2873 }
2874 DEFINE_SPAPR_MACHINE(2_3, "2.3", false);
2875 
2876 /*
2877  * pseries-2.2
2878  */
2879 
2880 #define SPAPR_COMPAT_2_2 \
2881         HW_COMPAT_2_2 \
2882         {\
2883             .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,\
2884             .property = "mem_win_size",\
2885             .value    = "0x20000000",\
2886         },
2887 
2888 static void spapr_machine_2_2_instance_options(MachineState *machine)
2889 {
2890     spapr_machine_2_3_instance_options(machine);
2891     machine->suppress_vmdesc = true;
2892 }
2893 
2894 static void spapr_machine_2_2_class_options(MachineClass *mc)
2895 {
2896     spapr_machine_2_3_class_options(mc);
2897     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_2);
2898 }
2899 DEFINE_SPAPR_MACHINE(2_2, "2.2", false);
2900 
2901 /*
2902  * pseries-2.1
2903  */
2904 #define SPAPR_COMPAT_2_1 \
2905         HW_COMPAT_2_1
2906 
2907 static void spapr_machine_2_1_instance_options(MachineState *machine)
2908 {
2909     spapr_machine_2_2_instance_options(machine);
2910 }
2911 
2912 static void spapr_machine_2_1_class_options(MachineClass *mc)
2913 {
2914     spapr_machine_2_2_class_options(mc);
2915     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_1);
2916 }
2917 DEFINE_SPAPR_MACHINE(2_1, "2.1", false);
2918 
2919 static void spapr_machine_register_types(void)
2920 {
2921     type_register_static(&spapr_machine_info);
2922 }
2923 
2924 type_init(spapr_machine_register_types)
2925