xref: /openbmc/qemu/hw/ppc/spapr.c (revision 0c2adc175398c37cc40b0164849aada9ad9c56c4)
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 "qapi/visitor.h"
30 #include "sysemu/sysemu.h"
31 #include "sysemu/numa.h"
32 #include "hw/hw.h"
33 #include "qemu/log.h"
34 #include "hw/fw-path-provider.h"
35 #include "elf.h"
36 #include "net/net.h"
37 #include "sysemu/device_tree.h"
38 #include "sysemu/cpus.h"
39 #include "sysemu/hw_accel.h"
40 #include "kvm_ppc.h"
41 #include "migration/misc.h"
42 #include "migration/global_state.h"
43 #include "migration/register.h"
44 #include "mmu-hash64.h"
45 #include "mmu-book3s-v3.h"
46 #include "cpu-models.h"
47 #include "qom/cpu.h"
48 
49 #include "hw/boards.h"
50 #include "hw/ppc/ppc.h"
51 #include "hw/loader.h"
52 
53 #include "hw/ppc/fdt.h"
54 #include "hw/ppc/spapr.h"
55 #include "hw/ppc/spapr_vio.h"
56 #include "hw/pci-host/spapr.h"
57 #include "hw/pci/msi.h"
58 
59 #include "hw/pci/pci.h"
60 #include "hw/scsi/scsi.h"
61 #include "hw/virtio/virtio-scsi.h"
62 #include "hw/virtio/vhost-scsi-common.h"
63 
64 #include "exec/address-spaces.h"
65 #include "exec/ram_addr.h"
66 #include "hw/usb.h"
67 #include "qemu/config-file.h"
68 #include "qemu/error-report.h"
69 #include "trace.h"
70 #include "hw/nmi.h"
71 #include "hw/intc/intc.h"
72 
73 #include "hw/compat.h"
74 #include "qemu/cutils.h"
75 #include "hw/ppc/spapr_cpu_core.h"
76 #include "hw/mem/memory-device.h"
77 
78 #include <libfdt.h>
79 
80 /* SLOF memory layout:
81  *
82  * SLOF raw image loaded at 0, copies its romfs right below the flat
83  * device-tree, then position SLOF itself 31M below that
84  *
85  * So we set FW_OVERHEAD to 40MB which should account for all of that
86  * and more
87  *
88  * We load our kernel at 4M, leaving space for SLOF initial image
89  */
90 #define FDT_MAX_SIZE            0x100000
91 #define RTAS_MAX_SIZE           0x10000
92 #define RTAS_MAX_ADDR           0x80000000 /* RTAS must stay below that */
93 #define FW_MAX_SIZE             0x400000
94 #define FW_FILE_NAME            "slof.bin"
95 #define FW_OVERHEAD             0x2800000
96 #define KERNEL_LOAD_ADDR        FW_MAX_SIZE
97 
98 #define MIN_RMA_SLOF            128UL
99 
100 #define PHANDLE_XICP            0x00001111
101 
102 /* These two functions implement the VCPU id numbering: one to compute them
103  * all and one to identify thread 0 of a VCORE. Any change to the first one
104  * is likely to have an impact on the second one, so let's keep them close.
105  */
106 static int spapr_vcpu_id(sPAPRMachineState *spapr, int cpu_index)
107 {
108     assert(spapr->vsmt);
109     return
110         (cpu_index / smp_threads) * spapr->vsmt + cpu_index % smp_threads;
111 }
112 static bool spapr_is_thread0_in_vcore(sPAPRMachineState *spapr,
113                                       PowerPCCPU *cpu)
114 {
115     assert(spapr->vsmt);
116     return spapr_get_vcpu_id(cpu) % spapr->vsmt == 0;
117 }
118 
119 static bool pre_2_10_vmstate_dummy_icp_needed(void *opaque)
120 {
121     /* Dummy entries correspond to unused ICPState objects in older QEMUs,
122      * and newer QEMUs don't even have them. In both cases, we don't want
123      * to send anything on the wire.
124      */
125     return false;
126 }
127 
128 static const VMStateDescription pre_2_10_vmstate_dummy_icp = {
129     .name = "icp/server",
130     .version_id = 1,
131     .minimum_version_id = 1,
132     .needed = pre_2_10_vmstate_dummy_icp_needed,
133     .fields = (VMStateField[]) {
134         VMSTATE_UNUSED(4), /* uint32_t xirr */
135         VMSTATE_UNUSED(1), /* uint8_t pending_priority */
136         VMSTATE_UNUSED(1), /* uint8_t mfrr */
137         VMSTATE_END_OF_LIST()
138     },
139 };
140 
141 static void pre_2_10_vmstate_register_dummy_icp(int i)
142 {
143     vmstate_register(NULL, i, &pre_2_10_vmstate_dummy_icp,
144                      (void *)(uintptr_t) i);
145 }
146 
147 static void pre_2_10_vmstate_unregister_dummy_icp(int i)
148 {
149     vmstate_unregister(NULL, &pre_2_10_vmstate_dummy_icp,
150                        (void *)(uintptr_t) i);
151 }
152 
153 static int xics_max_server_number(sPAPRMachineState *spapr)
154 {
155     assert(spapr->vsmt);
156     return DIV_ROUND_UP(max_cpus * spapr->vsmt, smp_threads);
157 }
158 
159 static int spapr_fixup_cpu_smt_dt(void *fdt, int offset, PowerPCCPU *cpu,
160                                   int smt_threads)
161 {
162     int i, ret = 0;
163     uint32_t servers_prop[smt_threads];
164     uint32_t gservers_prop[smt_threads * 2];
165     int index = spapr_get_vcpu_id(cpu);
166 
167     if (cpu->compat_pvr) {
168         ret = fdt_setprop_cell(fdt, offset, "cpu-version", cpu->compat_pvr);
169         if (ret < 0) {
170             return ret;
171         }
172     }
173 
174     /* Build interrupt servers and gservers properties */
175     for (i = 0; i < smt_threads; i++) {
176         servers_prop[i] = cpu_to_be32(index + i);
177         /* Hack, direct the group queues back to cpu 0 */
178         gservers_prop[i*2] = cpu_to_be32(index + i);
179         gservers_prop[i*2 + 1] = 0;
180     }
181     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-server#s",
182                       servers_prop, sizeof(servers_prop));
183     if (ret < 0) {
184         return ret;
185     }
186     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-gserver#s",
187                       gservers_prop, sizeof(gservers_prop));
188 
189     return ret;
190 }
191 
192 static int spapr_fixup_cpu_numa_dt(void *fdt, int offset, PowerPCCPU *cpu)
193 {
194     int index = spapr_get_vcpu_id(cpu);
195     uint32_t associativity[] = {cpu_to_be32(0x5),
196                                 cpu_to_be32(0x0),
197                                 cpu_to_be32(0x0),
198                                 cpu_to_be32(0x0),
199                                 cpu_to_be32(cpu->node_id),
200                                 cpu_to_be32(index)};
201 
202     /* Advertise NUMA via ibm,associativity */
203     return fdt_setprop(fdt, offset, "ibm,associativity", associativity,
204                           sizeof(associativity));
205 }
206 
207 /* Populate the "ibm,pa-features" property */
208 static void spapr_populate_pa_features(sPAPRMachineState *spapr,
209                                        PowerPCCPU *cpu,
210                                        void *fdt, int offset,
211                                        bool legacy_guest)
212 {
213     uint8_t pa_features_206[] = { 6, 0,
214         0xf6, 0x1f, 0xc7, 0x00, 0x80, 0xc0 };
215     uint8_t pa_features_207[] = { 24, 0,
216         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0,
217         0x80, 0x00, 0x00, 0x00, 0x00, 0x00,
218         0x00, 0x00, 0x00, 0x00, 0x80, 0x00,
219         0x80, 0x00, 0x80, 0x00, 0x00, 0x00 };
220     uint8_t pa_features_300[] = { 66, 0,
221         /* 0: MMU|FPU|SLB|RUN|DABR|NX, 1: fri[nzpm]|DABRX|SPRG3|SLB0|PP110 */
222         /* 2: VPM|DS205|PPR|DS202|DS206, 3: LSD|URG, SSO, 5: LE|CFAR|EB|LSQ */
223         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0, /* 0 - 5 */
224         /* 6: DS207 */
225         0x80, 0x00, 0x00, 0x00, 0x00, 0x00, /* 6 - 11 */
226         /* 16: Vector */
227         0x00, 0x00, 0x00, 0x00, 0x80, 0x00, /* 12 - 17 */
228         /* 18: Vec. Scalar, 20: Vec. XOR, 22: HTM */
229         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 18 - 23 */
230         /* 24: Ext. Dec, 26: 64 bit ftrs, 28: PM ftrs */
231         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 24 - 29 */
232         /* 30: MMR, 32: LE atomic, 34: EBB + ext EBB */
233         0x80, 0x00, 0x80, 0x00, 0xC0, 0x00, /* 30 - 35 */
234         /* 36: SPR SO, 38: Copy/Paste, 40: Radix MMU */
235         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 36 - 41 */
236         /* 42: PM, 44: PC RA, 46: SC vec'd */
237         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 42 - 47 */
238         /* 48: SIMD, 50: QP BFP, 52: String */
239         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 48 - 53 */
240         /* 54: DecFP, 56: DecI, 58: SHA */
241         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 54 - 59 */
242         /* 60: NM atomic, 62: RNG */
243         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 60 - 65 */
244     };
245     uint8_t *pa_features = NULL;
246     size_t pa_size;
247 
248     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_06, 0, cpu->compat_pvr)) {
249         pa_features = pa_features_206;
250         pa_size = sizeof(pa_features_206);
251     }
252     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_07, 0, cpu->compat_pvr)) {
253         pa_features = pa_features_207;
254         pa_size = sizeof(pa_features_207);
255     }
256     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_3_00, 0, cpu->compat_pvr)) {
257         pa_features = pa_features_300;
258         pa_size = sizeof(pa_features_300);
259     }
260     if (!pa_features) {
261         return;
262     }
263 
264     if (ppc_hash64_has(cpu, PPC_HASH64_CI_LARGEPAGE)) {
265         /*
266          * Note: we keep CI large pages off by default because a 64K capable
267          * guest provisioned with large pages might otherwise try to map a qemu
268          * framebuffer (or other kind of memory mapped PCI BAR) using 64K pages
269          * even if that qemu runs on a 4k host.
270          * We dd this bit back here if we are confident this is not an issue
271          */
272         pa_features[3] |= 0x20;
273     }
274     if ((spapr_get_cap(spapr, SPAPR_CAP_HTM) != 0) && pa_size > 24) {
275         pa_features[24] |= 0x80;    /* Transactional memory support */
276     }
277     if (legacy_guest && pa_size > 40) {
278         /* Workaround for broken kernels that attempt (guest) radix
279          * mode when they can't handle it, if they see the radix bit set
280          * in pa-features. So hide it from them. */
281         pa_features[40 + 2] &= ~0x80; /* Radix MMU */
282     }
283 
284     _FDT((fdt_setprop(fdt, offset, "ibm,pa-features", pa_features, pa_size)));
285 }
286 
287 static int spapr_fixup_cpu_dt(void *fdt, sPAPRMachineState *spapr)
288 {
289     int ret = 0, offset, cpus_offset;
290     CPUState *cs;
291     char cpu_model[32];
292     uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
293 
294     CPU_FOREACH(cs) {
295         PowerPCCPU *cpu = POWERPC_CPU(cs);
296         DeviceClass *dc = DEVICE_GET_CLASS(cs);
297         int index = spapr_get_vcpu_id(cpu);
298         int compat_smt = MIN(smp_threads, ppc_compat_max_vthreads(cpu));
299 
300         if (!spapr_is_thread0_in_vcore(spapr, cpu)) {
301             continue;
302         }
303 
304         snprintf(cpu_model, 32, "%s@%x", dc->fw_name, index);
305 
306         cpus_offset = fdt_path_offset(fdt, "/cpus");
307         if (cpus_offset < 0) {
308             cpus_offset = fdt_add_subnode(fdt, 0, "cpus");
309             if (cpus_offset < 0) {
310                 return cpus_offset;
311             }
312         }
313         offset = fdt_subnode_offset(fdt, cpus_offset, cpu_model);
314         if (offset < 0) {
315             offset = fdt_add_subnode(fdt, cpus_offset, cpu_model);
316             if (offset < 0) {
317                 return offset;
318             }
319         }
320 
321         ret = fdt_setprop(fdt, offset, "ibm,pft-size",
322                           pft_size_prop, sizeof(pft_size_prop));
323         if (ret < 0) {
324             return ret;
325         }
326 
327         if (nb_numa_nodes > 1) {
328             ret = spapr_fixup_cpu_numa_dt(fdt, offset, cpu);
329             if (ret < 0) {
330                 return ret;
331             }
332         }
333 
334         ret = spapr_fixup_cpu_smt_dt(fdt, offset, cpu, compat_smt);
335         if (ret < 0) {
336             return ret;
337         }
338 
339         spapr_populate_pa_features(spapr, cpu, fdt, offset,
340                                    spapr->cas_legacy_guest_workaround);
341     }
342     return ret;
343 }
344 
345 static hwaddr spapr_node0_size(MachineState *machine)
346 {
347     if (nb_numa_nodes) {
348         int i;
349         for (i = 0; i < nb_numa_nodes; ++i) {
350             if (numa_info[i].node_mem) {
351                 return MIN(pow2floor(numa_info[i].node_mem),
352                            machine->ram_size);
353             }
354         }
355     }
356     return machine->ram_size;
357 }
358 
359 static void add_str(GString *s, const gchar *s1)
360 {
361     g_string_append_len(s, s1, strlen(s1) + 1);
362 }
363 
364 static int spapr_populate_memory_node(void *fdt, int nodeid, hwaddr start,
365                                        hwaddr size)
366 {
367     uint32_t associativity[] = {
368         cpu_to_be32(0x4), /* length */
369         cpu_to_be32(0x0), cpu_to_be32(0x0),
370         cpu_to_be32(0x0), cpu_to_be32(nodeid)
371     };
372     char mem_name[32];
373     uint64_t mem_reg_property[2];
374     int off;
375 
376     mem_reg_property[0] = cpu_to_be64(start);
377     mem_reg_property[1] = cpu_to_be64(size);
378 
379     sprintf(mem_name, "memory@" TARGET_FMT_lx, start);
380     off = fdt_add_subnode(fdt, 0, mem_name);
381     _FDT(off);
382     _FDT((fdt_setprop_string(fdt, off, "device_type", "memory")));
383     _FDT((fdt_setprop(fdt, off, "reg", mem_reg_property,
384                       sizeof(mem_reg_property))));
385     _FDT((fdt_setprop(fdt, off, "ibm,associativity", associativity,
386                       sizeof(associativity))));
387     return off;
388 }
389 
390 static int spapr_populate_memory(sPAPRMachineState *spapr, void *fdt)
391 {
392     MachineState *machine = MACHINE(spapr);
393     hwaddr mem_start, node_size;
394     int i, nb_nodes = nb_numa_nodes;
395     NodeInfo *nodes = numa_info;
396     NodeInfo ramnode;
397 
398     /* No NUMA nodes, assume there is just one node with whole RAM */
399     if (!nb_numa_nodes) {
400         nb_nodes = 1;
401         ramnode.node_mem = machine->ram_size;
402         nodes = &ramnode;
403     }
404 
405     for (i = 0, mem_start = 0; i < nb_nodes; ++i) {
406         if (!nodes[i].node_mem) {
407             continue;
408         }
409         if (mem_start >= machine->ram_size) {
410             node_size = 0;
411         } else {
412             node_size = nodes[i].node_mem;
413             if (node_size > machine->ram_size - mem_start) {
414                 node_size = machine->ram_size - mem_start;
415             }
416         }
417         if (!mem_start) {
418             /* spapr_machine_init() checks for rma_size <= node0_size
419              * already */
420             spapr_populate_memory_node(fdt, i, 0, spapr->rma_size);
421             mem_start += spapr->rma_size;
422             node_size -= spapr->rma_size;
423         }
424         for ( ; node_size; ) {
425             hwaddr sizetmp = pow2floor(node_size);
426 
427             /* mem_start != 0 here */
428             if (ctzl(mem_start) < ctzl(sizetmp)) {
429                 sizetmp = 1ULL << ctzl(mem_start);
430             }
431 
432             spapr_populate_memory_node(fdt, i, mem_start, sizetmp);
433             node_size -= sizetmp;
434             mem_start += sizetmp;
435         }
436     }
437 
438     return 0;
439 }
440 
441 static void spapr_populate_cpu_dt(CPUState *cs, void *fdt, int offset,
442                                   sPAPRMachineState *spapr)
443 {
444     PowerPCCPU *cpu = POWERPC_CPU(cs);
445     CPUPPCState *env = &cpu->env;
446     PowerPCCPUClass *pcc = POWERPC_CPU_GET_CLASS(cs);
447     int index = spapr_get_vcpu_id(cpu);
448     uint32_t segs[] = {cpu_to_be32(28), cpu_to_be32(40),
449                        0xffffffff, 0xffffffff};
450     uint32_t tbfreq = kvm_enabled() ? kvmppc_get_tbfreq()
451         : SPAPR_TIMEBASE_FREQ;
452     uint32_t cpufreq = kvm_enabled() ? kvmppc_get_clockfreq() : 1000000000;
453     uint32_t page_sizes_prop[64];
454     size_t page_sizes_prop_size;
455     uint32_t vcpus_per_socket = smp_threads * smp_cores;
456     uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
457     int compat_smt = MIN(smp_threads, ppc_compat_max_vthreads(cpu));
458     sPAPRDRConnector *drc;
459     int drc_index;
460     uint32_t radix_AP_encodings[PPC_PAGE_SIZES_MAX_SZ];
461     int i;
462 
463     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU, index);
464     if (drc) {
465         drc_index = spapr_drc_index(drc);
466         _FDT((fdt_setprop_cell(fdt, offset, "ibm,my-drc-index", drc_index)));
467     }
468 
469     _FDT((fdt_setprop_cell(fdt, offset, "reg", index)));
470     _FDT((fdt_setprop_string(fdt, offset, "device_type", "cpu")));
471 
472     _FDT((fdt_setprop_cell(fdt, offset, "cpu-version", env->spr[SPR_PVR])));
473     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-block-size",
474                            env->dcache_line_size)));
475     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-line-size",
476                            env->dcache_line_size)));
477     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-block-size",
478                            env->icache_line_size)));
479     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-line-size",
480                            env->icache_line_size)));
481 
482     if (pcc->l1_dcache_size) {
483         _FDT((fdt_setprop_cell(fdt, offset, "d-cache-size",
484                                pcc->l1_dcache_size)));
485     } else {
486         warn_report("Unknown L1 dcache size for cpu");
487     }
488     if (pcc->l1_icache_size) {
489         _FDT((fdt_setprop_cell(fdt, offset, "i-cache-size",
490                                pcc->l1_icache_size)));
491     } else {
492         warn_report("Unknown L1 icache size for cpu");
493     }
494 
495     _FDT((fdt_setprop_cell(fdt, offset, "timebase-frequency", tbfreq)));
496     _FDT((fdt_setprop_cell(fdt, offset, "clock-frequency", cpufreq)));
497     _FDT((fdt_setprop_cell(fdt, offset, "slb-size", cpu->hash64_opts->slb_size)));
498     _FDT((fdt_setprop_cell(fdt, offset, "ibm,slb-size", cpu->hash64_opts->slb_size)));
499     _FDT((fdt_setprop_string(fdt, offset, "status", "okay")));
500     _FDT((fdt_setprop(fdt, offset, "64-bit", NULL, 0)));
501 
502     if (env->spr_cb[SPR_PURR].oea_read) {
503         _FDT((fdt_setprop(fdt, offset, "ibm,purr", NULL, 0)));
504     }
505 
506     if (ppc_hash64_has(cpu, PPC_HASH64_1TSEG)) {
507         _FDT((fdt_setprop(fdt, offset, "ibm,processor-segment-sizes",
508                           segs, sizeof(segs))));
509     }
510 
511     /* Advertise VSX (vector extensions) if available
512      *   1               == VMX / Altivec available
513      *   2               == VSX available
514      *
515      * Only CPUs for which we create core types in spapr_cpu_core.c
516      * are possible, and all of those have VMX */
517     if (spapr_get_cap(spapr, SPAPR_CAP_VSX) != 0) {
518         _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 2)));
519     } else {
520         _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 1)));
521     }
522 
523     /* Advertise DFP (Decimal Floating Point) if available
524      *   0 / no property == no DFP
525      *   1               == DFP available */
526     if (spapr_get_cap(spapr, SPAPR_CAP_DFP) != 0) {
527         _FDT((fdt_setprop_cell(fdt, offset, "ibm,dfp", 1)));
528     }
529 
530     page_sizes_prop_size = ppc_create_page_sizes_prop(cpu, page_sizes_prop,
531                                                       sizeof(page_sizes_prop));
532     if (page_sizes_prop_size) {
533         _FDT((fdt_setprop(fdt, offset, "ibm,segment-page-sizes",
534                           page_sizes_prop, page_sizes_prop_size)));
535     }
536 
537     spapr_populate_pa_features(spapr, cpu, fdt, offset, false);
538 
539     _FDT((fdt_setprop_cell(fdt, offset, "ibm,chip-id",
540                            cs->cpu_index / vcpus_per_socket)));
541 
542     _FDT((fdt_setprop(fdt, offset, "ibm,pft-size",
543                       pft_size_prop, sizeof(pft_size_prop))));
544 
545     if (nb_numa_nodes > 1) {
546         _FDT(spapr_fixup_cpu_numa_dt(fdt, offset, cpu));
547     }
548 
549     _FDT(spapr_fixup_cpu_smt_dt(fdt, offset, cpu, compat_smt));
550 
551     if (pcc->radix_page_info) {
552         for (i = 0; i < pcc->radix_page_info->count; i++) {
553             radix_AP_encodings[i] =
554                 cpu_to_be32(pcc->radix_page_info->entries[i]);
555         }
556         _FDT((fdt_setprop(fdt, offset, "ibm,processor-radix-AP-encodings",
557                           radix_AP_encodings,
558                           pcc->radix_page_info->count *
559                           sizeof(radix_AP_encodings[0]))));
560     }
561 }
562 
563 static void spapr_populate_cpus_dt_node(void *fdt, sPAPRMachineState *spapr)
564 {
565     CPUState **rev;
566     CPUState *cs;
567     int n_cpus;
568     int cpus_offset;
569     char *nodename;
570     int i;
571 
572     cpus_offset = fdt_add_subnode(fdt, 0, "cpus");
573     _FDT(cpus_offset);
574     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#address-cells", 0x1)));
575     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#size-cells", 0x0)));
576 
577     /*
578      * We walk the CPUs in reverse order to ensure that CPU DT nodes
579      * created by fdt_add_subnode() end up in the right order in FDT
580      * for the guest kernel the enumerate the CPUs correctly.
581      *
582      * The CPU list cannot be traversed in reverse order, so we need
583      * to do extra work.
584      */
585     n_cpus = 0;
586     rev = NULL;
587     CPU_FOREACH(cs) {
588         rev = g_renew(CPUState *, rev, n_cpus + 1);
589         rev[n_cpus++] = cs;
590     }
591 
592     for (i = n_cpus - 1; i >= 0; i--) {
593         CPUState *cs = rev[i];
594         PowerPCCPU *cpu = POWERPC_CPU(cs);
595         int index = spapr_get_vcpu_id(cpu);
596         DeviceClass *dc = DEVICE_GET_CLASS(cs);
597         int offset;
598 
599         if (!spapr_is_thread0_in_vcore(spapr, cpu)) {
600             continue;
601         }
602 
603         nodename = g_strdup_printf("%s@%x", dc->fw_name, index);
604         offset = fdt_add_subnode(fdt, cpus_offset, nodename);
605         g_free(nodename);
606         _FDT(offset);
607         spapr_populate_cpu_dt(cs, fdt, offset, spapr);
608     }
609 
610     g_free(rev);
611 }
612 
613 static int spapr_rng_populate_dt(void *fdt)
614 {
615     int node;
616     int ret;
617 
618     node = qemu_fdt_add_subnode(fdt, "/ibm,platform-facilities");
619     if (node <= 0) {
620         return -1;
621     }
622     ret = fdt_setprop_string(fdt, node, "device_type",
623                              "ibm,platform-facilities");
624     ret |= fdt_setprop_cell(fdt, node, "#address-cells", 0x1);
625     ret |= fdt_setprop_cell(fdt, node, "#size-cells", 0x0);
626 
627     node = fdt_add_subnode(fdt, node, "ibm,random-v1");
628     if (node <= 0) {
629         return -1;
630     }
631     ret |= fdt_setprop_string(fdt, node, "compatible", "ibm,random");
632 
633     return ret ? -1 : 0;
634 }
635 
636 static uint32_t spapr_pc_dimm_node(MemoryDeviceInfoList *list, ram_addr_t addr)
637 {
638     MemoryDeviceInfoList *info;
639 
640     for (info = list; info; info = info->next) {
641         MemoryDeviceInfo *value = info->value;
642 
643         if (value && value->type == MEMORY_DEVICE_INFO_KIND_DIMM) {
644             PCDIMMDeviceInfo *pcdimm_info = value->u.dimm.data;
645 
646             if (addr >= pcdimm_info->addr &&
647                 addr < (pcdimm_info->addr + pcdimm_info->size)) {
648                 return pcdimm_info->node;
649             }
650         }
651     }
652 
653     return -1;
654 }
655 
656 struct sPAPRDrconfCellV2 {
657      uint32_t seq_lmbs;
658      uint64_t base_addr;
659      uint32_t drc_index;
660      uint32_t aa_index;
661      uint32_t flags;
662 } QEMU_PACKED;
663 
664 typedef struct DrconfCellQueue {
665     struct sPAPRDrconfCellV2 cell;
666     QSIMPLEQ_ENTRY(DrconfCellQueue) entry;
667 } DrconfCellQueue;
668 
669 static DrconfCellQueue *
670 spapr_get_drconf_cell(uint32_t seq_lmbs, uint64_t base_addr,
671                       uint32_t drc_index, uint32_t aa_index,
672                       uint32_t flags)
673 {
674     DrconfCellQueue *elem;
675 
676     elem = g_malloc0(sizeof(*elem));
677     elem->cell.seq_lmbs = cpu_to_be32(seq_lmbs);
678     elem->cell.base_addr = cpu_to_be64(base_addr);
679     elem->cell.drc_index = cpu_to_be32(drc_index);
680     elem->cell.aa_index = cpu_to_be32(aa_index);
681     elem->cell.flags = cpu_to_be32(flags);
682 
683     return elem;
684 }
685 
686 /* ibm,dynamic-memory-v2 */
687 static int spapr_populate_drmem_v2(sPAPRMachineState *spapr, void *fdt,
688                                    int offset, MemoryDeviceInfoList *dimms)
689 {
690     MachineState *machine = MACHINE(spapr);
691     uint8_t *int_buf, *cur_index, buf_len;
692     int ret;
693     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
694     uint64_t addr, cur_addr, size;
695     uint32_t nr_boot_lmbs = (machine->device_memory->base / lmb_size);
696     uint64_t mem_end = machine->device_memory->base +
697                        memory_region_size(&machine->device_memory->mr);
698     uint32_t node, nr_entries = 0;
699     sPAPRDRConnector *drc;
700     DrconfCellQueue *elem, *next;
701     MemoryDeviceInfoList *info;
702     QSIMPLEQ_HEAD(, DrconfCellQueue) drconf_queue
703         = QSIMPLEQ_HEAD_INITIALIZER(drconf_queue);
704 
705     /* Entry to cover RAM and the gap area */
706     elem = spapr_get_drconf_cell(nr_boot_lmbs, 0, 0, -1,
707                                  SPAPR_LMB_FLAGS_RESERVED |
708                                  SPAPR_LMB_FLAGS_DRC_INVALID);
709     QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
710     nr_entries++;
711 
712     cur_addr = machine->device_memory->base;
713     for (info = dimms; info; info = info->next) {
714         PCDIMMDeviceInfo *di = info->value->u.dimm.data;
715 
716         addr = di->addr;
717         size = di->size;
718         node = di->node;
719 
720         /* Entry for hot-pluggable area */
721         if (cur_addr < addr) {
722             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
723             g_assert(drc);
724             elem = spapr_get_drconf_cell((addr - cur_addr) / lmb_size,
725                                          cur_addr, spapr_drc_index(drc), -1, 0);
726             QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
727             nr_entries++;
728         }
729 
730         /* Entry for DIMM */
731         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, addr / lmb_size);
732         g_assert(drc);
733         elem = spapr_get_drconf_cell(size / lmb_size, addr,
734                                      spapr_drc_index(drc), node,
735                                      SPAPR_LMB_FLAGS_ASSIGNED);
736         QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
737         nr_entries++;
738         cur_addr = addr + size;
739     }
740 
741     /* Entry for remaining hotpluggable area */
742     if (cur_addr < mem_end) {
743         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
744         g_assert(drc);
745         elem = spapr_get_drconf_cell((mem_end - cur_addr) / lmb_size,
746                                      cur_addr, spapr_drc_index(drc), -1, 0);
747         QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
748         nr_entries++;
749     }
750 
751     buf_len = nr_entries * sizeof(struct sPAPRDrconfCellV2) + sizeof(uint32_t);
752     int_buf = cur_index = g_malloc0(buf_len);
753     *(uint32_t *)int_buf = cpu_to_be32(nr_entries);
754     cur_index += sizeof(nr_entries);
755 
756     QSIMPLEQ_FOREACH_SAFE(elem, &drconf_queue, entry, next) {
757         memcpy(cur_index, &elem->cell, sizeof(elem->cell));
758         cur_index += sizeof(elem->cell);
759         QSIMPLEQ_REMOVE(&drconf_queue, elem, DrconfCellQueue, entry);
760         g_free(elem);
761     }
762 
763     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory-v2", int_buf, buf_len);
764     g_free(int_buf);
765     if (ret < 0) {
766         return -1;
767     }
768     return 0;
769 }
770 
771 /* ibm,dynamic-memory */
772 static int spapr_populate_drmem_v1(sPAPRMachineState *spapr, void *fdt,
773                                    int offset, MemoryDeviceInfoList *dimms)
774 {
775     MachineState *machine = MACHINE(spapr);
776     int i, ret;
777     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
778     uint32_t device_lmb_start = machine->device_memory->base / lmb_size;
779     uint32_t nr_lmbs = (machine->device_memory->base +
780                        memory_region_size(&machine->device_memory->mr)) /
781                        lmb_size;
782     uint32_t *int_buf, *cur_index, buf_len;
783 
784     /*
785      * Allocate enough buffer size to fit in ibm,dynamic-memory
786      */
787     buf_len = (nr_lmbs * SPAPR_DR_LMB_LIST_ENTRY_SIZE + 1) * sizeof(uint32_t);
788     cur_index = int_buf = g_malloc0(buf_len);
789     int_buf[0] = cpu_to_be32(nr_lmbs);
790     cur_index++;
791     for (i = 0; i < nr_lmbs; i++) {
792         uint64_t addr = i * lmb_size;
793         uint32_t *dynamic_memory = cur_index;
794 
795         if (i >= device_lmb_start) {
796             sPAPRDRConnector *drc;
797 
798             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, i);
799             g_assert(drc);
800 
801             dynamic_memory[0] = cpu_to_be32(addr >> 32);
802             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
803             dynamic_memory[2] = cpu_to_be32(spapr_drc_index(drc));
804             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
805             dynamic_memory[4] = cpu_to_be32(spapr_pc_dimm_node(dimms, addr));
806             if (memory_region_present(get_system_memory(), addr)) {
807                 dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_ASSIGNED);
808             } else {
809                 dynamic_memory[5] = cpu_to_be32(0);
810             }
811         } else {
812             /*
813              * LMB information for RMA, boot time RAM and gap b/n RAM and
814              * device memory region -- all these are marked as reserved
815              * and as having no valid DRC.
816              */
817             dynamic_memory[0] = cpu_to_be32(addr >> 32);
818             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
819             dynamic_memory[2] = cpu_to_be32(0);
820             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
821             dynamic_memory[4] = cpu_to_be32(-1);
822             dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_RESERVED |
823                                             SPAPR_LMB_FLAGS_DRC_INVALID);
824         }
825 
826         cur_index += SPAPR_DR_LMB_LIST_ENTRY_SIZE;
827     }
828     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory", int_buf, buf_len);
829     g_free(int_buf);
830     if (ret < 0) {
831         return -1;
832     }
833     return 0;
834 }
835 
836 /*
837  * Adds ibm,dynamic-reconfiguration-memory node.
838  * Refer to docs/specs/ppc-spapr-hotplug.txt for the documentation
839  * of this device tree node.
840  */
841 static int spapr_populate_drconf_memory(sPAPRMachineState *spapr, void *fdt)
842 {
843     MachineState *machine = MACHINE(spapr);
844     int ret, i, offset;
845     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
846     uint32_t prop_lmb_size[] = {0, cpu_to_be32(lmb_size)};
847     uint32_t *int_buf, *cur_index, buf_len;
848     int nr_nodes = nb_numa_nodes ? nb_numa_nodes : 1;
849     MemoryDeviceInfoList *dimms = NULL;
850 
851     /*
852      * Don't create the node if there is no device memory
853      */
854     if (machine->ram_size == machine->maxram_size) {
855         return 0;
856     }
857 
858     offset = fdt_add_subnode(fdt, 0, "ibm,dynamic-reconfiguration-memory");
859 
860     ret = fdt_setprop(fdt, offset, "ibm,lmb-size", prop_lmb_size,
861                     sizeof(prop_lmb_size));
862     if (ret < 0) {
863         return ret;
864     }
865 
866     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-flags-mask", 0xff);
867     if (ret < 0) {
868         return ret;
869     }
870 
871     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-preservation-time", 0x0);
872     if (ret < 0) {
873         return ret;
874     }
875 
876     /* ibm,dynamic-memory or ibm,dynamic-memory-v2 */
877     dimms = qmp_memory_device_list();
878     if (spapr_ovec_test(spapr->ov5_cas, OV5_DRMEM_V2)) {
879         ret = spapr_populate_drmem_v2(spapr, fdt, offset, dimms);
880     } else {
881         ret = spapr_populate_drmem_v1(spapr, fdt, offset, dimms);
882     }
883     qapi_free_MemoryDeviceInfoList(dimms);
884 
885     if (ret < 0) {
886         return ret;
887     }
888 
889     /* ibm,associativity-lookup-arrays */
890     buf_len = (nr_nodes * 4 + 2) * sizeof(uint32_t);
891     cur_index = int_buf = g_malloc0(buf_len);
892 
893     cur_index = int_buf;
894     int_buf[0] = cpu_to_be32(nr_nodes);
895     int_buf[1] = cpu_to_be32(4); /* Number of entries per associativity list */
896     cur_index += 2;
897     for (i = 0; i < nr_nodes; i++) {
898         uint32_t associativity[] = {
899             cpu_to_be32(0x0),
900             cpu_to_be32(0x0),
901             cpu_to_be32(0x0),
902             cpu_to_be32(i)
903         };
904         memcpy(cur_index, associativity, sizeof(associativity));
905         cur_index += 4;
906     }
907     ret = fdt_setprop(fdt, offset, "ibm,associativity-lookup-arrays", int_buf,
908             (cur_index - int_buf) * sizeof(uint32_t));
909     g_free(int_buf);
910 
911     return ret;
912 }
913 
914 static int spapr_dt_cas_updates(sPAPRMachineState *spapr, void *fdt,
915                                 sPAPROptionVector *ov5_updates)
916 {
917     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
918     int ret = 0, offset;
919 
920     /* Generate ibm,dynamic-reconfiguration-memory node if required */
921     if (spapr_ovec_test(ov5_updates, OV5_DRCONF_MEMORY)) {
922         g_assert(smc->dr_lmb_enabled);
923         ret = spapr_populate_drconf_memory(spapr, fdt);
924         if (ret) {
925             goto out;
926         }
927     }
928 
929     offset = fdt_path_offset(fdt, "/chosen");
930     if (offset < 0) {
931         offset = fdt_add_subnode(fdt, 0, "chosen");
932         if (offset < 0) {
933             return offset;
934         }
935     }
936     ret = spapr_ovec_populate_dt(fdt, offset, spapr->ov5_cas,
937                                  "ibm,architecture-vec-5");
938 
939 out:
940     return ret;
941 }
942 
943 static bool spapr_hotplugged_dev_before_cas(void)
944 {
945     Object *drc_container, *obj;
946     ObjectProperty *prop;
947     ObjectPropertyIterator iter;
948 
949     drc_container = container_get(object_get_root(), "/dr-connector");
950     object_property_iter_init(&iter, drc_container);
951     while ((prop = object_property_iter_next(&iter))) {
952         if (!strstart(prop->type, "link<", NULL)) {
953             continue;
954         }
955         obj = object_property_get_link(drc_container, prop->name, NULL);
956         if (spapr_drc_needed(obj)) {
957             return true;
958         }
959     }
960     return false;
961 }
962 
963 int spapr_h_cas_compose_response(sPAPRMachineState *spapr,
964                                  target_ulong addr, target_ulong size,
965                                  sPAPROptionVector *ov5_updates)
966 {
967     void *fdt, *fdt_skel;
968     sPAPRDeviceTreeUpdateHeader hdr = { .version_id = 1 };
969 
970     if (spapr_hotplugged_dev_before_cas()) {
971         return 1;
972     }
973 
974     if (size < sizeof(hdr) || size > FW_MAX_SIZE) {
975         error_report("SLOF provided an unexpected CAS buffer size "
976                      TARGET_FMT_lu " (min: %zu, max: %u)",
977                      size, sizeof(hdr), FW_MAX_SIZE);
978         exit(EXIT_FAILURE);
979     }
980 
981     size -= sizeof(hdr);
982 
983     /* Create skeleton */
984     fdt_skel = g_malloc0(size);
985     _FDT((fdt_create(fdt_skel, size)));
986     _FDT((fdt_finish_reservemap(fdt_skel)));
987     _FDT((fdt_begin_node(fdt_skel, "")));
988     _FDT((fdt_end_node(fdt_skel)));
989     _FDT((fdt_finish(fdt_skel)));
990     fdt = g_malloc0(size);
991     _FDT((fdt_open_into(fdt_skel, fdt, size)));
992     g_free(fdt_skel);
993 
994     /* Fixup cpu nodes */
995     _FDT((spapr_fixup_cpu_dt(fdt, spapr)));
996 
997     if (spapr_dt_cas_updates(spapr, fdt, ov5_updates)) {
998         return -1;
999     }
1000 
1001     /* Pack resulting tree */
1002     _FDT((fdt_pack(fdt)));
1003 
1004     if (fdt_totalsize(fdt) + sizeof(hdr) > size) {
1005         trace_spapr_cas_failed(size);
1006         return -1;
1007     }
1008 
1009     cpu_physical_memory_write(addr, &hdr, sizeof(hdr));
1010     cpu_physical_memory_write(addr + sizeof(hdr), fdt, fdt_totalsize(fdt));
1011     trace_spapr_cas_continue(fdt_totalsize(fdt) + sizeof(hdr));
1012     g_free(fdt);
1013 
1014     return 0;
1015 }
1016 
1017 static void spapr_dt_rtas(sPAPRMachineState *spapr, void *fdt)
1018 {
1019     int rtas;
1020     GString *hypertas = g_string_sized_new(256);
1021     GString *qemu_hypertas = g_string_sized_new(256);
1022     uint32_t refpoints[] = { cpu_to_be32(0x4), cpu_to_be32(0x4) };
1023     uint64_t max_device_addr = MACHINE(spapr)->device_memory->base +
1024         memory_region_size(&MACHINE(spapr)->device_memory->mr);
1025     uint32_t lrdr_capacity[] = {
1026         cpu_to_be32(max_device_addr >> 32),
1027         cpu_to_be32(max_device_addr & 0xffffffff),
1028         0, cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE),
1029         cpu_to_be32(max_cpus / smp_threads),
1030     };
1031     uint32_t maxdomains[] = {
1032         cpu_to_be32(4),
1033         cpu_to_be32(0),
1034         cpu_to_be32(0),
1035         cpu_to_be32(0),
1036         cpu_to_be32(nb_numa_nodes ? nb_numa_nodes - 1 : 0),
1037     };
1038 
1039     _FDT(rtas = fdt_add_subnode(fdt, 0, "rtas"));
1040 
1041     /* hypertas */
1042     add_str(hypertas, "hcall-pft");
1043     add_str(hypertas, "hcall-term");
1044     add_str(hypertas, "hcall-dabr");
1045     add_str(hypertas, "hcall-interrupt");
1046     add_str(hypertas, "hcall-tce");
1047     add_str(hypertas, "hcall-vio");
1048     add_str(hypertas, "hcall-splpar");
1049     add_str(hypertas, "hcall-bulk");
1050     add_str(hypertas, "hcall-set-mode");
1051     add_str(hypertas, "hcall-sprg0");
1052     add_str(hypertas, "hcall-copy");
1053     add_str(hypertas, "hcall-debug");
1054     add_str(qemu_hypertas, "hcall-memop1");
1055 
1056     if (!kvm_enabled() || kvmppc_spapr_use_multitce()) {
1057         add_str(hypertas, "hcall-multi-tce");
1058     }
1059 
1060     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
1061         add_str(hypertas, "hcall-hpt-resize");
1062     }
1063 
1064     _FDT(fdt_setprop(fdt, rtas, "ibm,hypertas-functions",
1065                      hypertas->str, hypertas->len));
1066     g_string_free(hypertas, TRUE);
1067     _FDT(fdt_setprop(fdt, rtas, "qemu,hypertas-functions",
1068                      qemu_hypertas->str, qemu_hypertas->len));
1069     g_string_free(qemu_hypertas, TRUE);
1070 
1071     _FDT(fdt_setprop(fdt, rtas, "ibm,associativity-reference-points",
1072                      refpoints, sizeof(refpoints)));
1073 
1074     _FDT(fdt_setprop(fdt, rtas, "ibm,max-associativity-domains",
1075                      maxdomains, sizeof(maxdomains)));
1076 
1077     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-error-log-max",
1078                           RTAS_ERROR_LOG_MAX));
1079     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-event-scan-rate",
1080                           RTAS_EVENT_SCAN_RATE));
1081 
1082     g_assert(msi_nonbroken);
1083     _FDT(fdt_setprop(fdt, rtas, "ibm,change-msix-capable", NULL, 0));
1084 
1085     /*
1086      * According to PAPR, rtas ibm,os-term does not guarantee a return
1087      * back to the guest cpu.
1088      *
1089      * While an additional ibm,extended-os-term property indicates
1090      * that rtas call return will always occur. Set this property.
1091      */
1092     _FDT(fdt_setprop(fdt, rtas, "ibm,extended-os-term", NULL, 0));
1093 
1094     _FDT(fdt_setprop(fdt, rtas, "ibm,lrdr-capacity",
1095                      lrdr_capacity, sizeof(lrdr_capacity)));
1096 
1097     spapr_dt_rtas_tokens(fdt, rtas);
1098 }
1099 
1100 /* Prepare ibm,arch-vec-5-platform-support, which indicates the MMU features
1101  * that the guest may request and thus the valid values for bytes 24..26 of
1102  * option vector 5: */
1103 static void spapr_dt_ov5_platform_support(void *fdt, int chosen)
1104 {
1105     PowerPCCPU *first_ppc_cpu = POWERPC_CPU(first_cpu);
1106 
1107     char val[2 * 4] = {
1108         23, 0x00, /* Xive mode, filled in below. */
1109         24, 0x00, /* Hash/Radix, filled in below. */
1110         25, 0x00, /* Hash options: Segment Tables == no, GTSE == no. */
1111         26, 0x40, /* Radix options: GTSE == yes. */
1112     };
1113 
1114     if (!ppc_check_compat(first_ppc_cpu, CPU_POWERPC_LOGICAL_3_00, 0,
1115                           first_ppc_cpu->compat_pvr)) {
1116         /* If we're in a pre POWER9 compat mode then the guest should do hash */
1117         val[3] = 0x00; /* Hash */
1118     } else if (kvm_enabled()) {
1119         if (kvmppc_has_cap_mmu_radix() && kvmppc_has_cap_mmu_hash_v3()) {
1120             val[3] = 0x80; /* OV5_MMU_BOTH */
1121         } else if (kvmppc_has_cap_mmu_radix()) {
1122             val[3] = 0x40; /* OV5_MMU_RADIX_300 */
1123         } else {
1124             val[3] = 0x00; /* Hash */
1125         }
1126     } else {
1127         /* V3 MMU supports both hash and radix in tcg (with dynamic switching) */
1128         val[3] = 0xC0;
1129     }
1130     _FDT(fdt_setprop(fdt, chosen, "ibm,arch-vec-5-platform-support",
1131                      val, sizeof(val)));
1132 }
1133 
1134 static void spapr_dt_chosen(sPAPRMachineState *spapr, void *fdt)
1135 {
1136     MachineState *machine = MACHINE(spapr);
1137     int chosen;
1138     const char *boot_device = machine->boot_order;
1139     char *stdout_path = spapr_vio_stdout_path(spapr->vio_bus);
1140     size_t cb = 0;
1141     char *bootlist = get_boot_devices_list(&cb);
1142 
1143     _FDT(chosen = fdt_add_subnode(fdt, 0, "chosen"));
1144 
1145     _FDT(fdt_setprop_string(fdt, chosen, "bootargs", machine->kernel_cmdline));
1146     _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-start",
1147                           spapr->initrd_base));
1148     _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-end",
1149                           spapr->initrd_base + spapr->initrd_size));
1150 
1151     if (spapr->kernel_size) {
1152         uint64_t kprop[2] = { cpu_to_be64(KERNEL_LOAD_ADDR),
1153                               cpu_to_be64(spapr->kernel_size) };
1154 
1155         _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel",
1156                          &kprop, sizeof(kprop)));
1157         if (spapr->kernel_le) {
1158             _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel-le", NULL, 0));
1159         }
1160     }
1161     if (boot_menu) {
1162         _FDT((fdt_setprop_cell(fdt, chosen, "qemu,boot-menu", boot_menu)));
1163     }
1164     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-width", graphic_width));
1165     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-height", graphic_height));
1166     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-depth", graphic_depth));
1167 
1168     if (cb && bootlist) {
1169         int i;
1170 
1171         for (i = 0; i < cb; i++) {
1172             if (bootlist[i] == '\n') {
1173                 bootlist[i] = ' ';
1174             }
1175         }
1176         _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-list", bootlist));
1177     }
1178 
1179     if (boot_device && strlen(boot_device)) {
1180         _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-device", boot_device));
1181     }
1182 
1183     if (!spapr->has_graphics && stdout_path) {
1184         /*
1185          * "linux,stdout-path" and "stdout" properties are deprecated by linux
1186          * kernel. New platforms should only use the "stdout-path" property. Set
1187          * the new property and continue using older property to remain
1188          * compatible with the existing firmware.
1189          */
1190         _FDT(fdt_setprop_string(fdt, chosen, "linux,stdout-path", stdout_path));
1191         _FDT(fdt_setprop_string(fdt, chosen, "stdout-path", stdout_path));
1192     }
1193 
1194     spapr_dt_ov5_platform_support(fdt, chosen);
1195 
1196     g_free(stdout_path);
1197     g_free(bootlist);
1198 }
1199 
1200 static void spapr_dt_hypervisor(sPAPRMachineState *spapr, void *fdt)
1201 {
1202     /* The /hypervisor node isn't in PAPR - this is a hack to allow PR
1203      * KVM to work under pHyp with some guest co-operation */
1204     int hypervisor;
1205     uint8_t hypercall[16];
1206 
1207     _FDT(hypervisor = fdt_add_subnode(fdt, 0, "hypervisor"));
1208     /* indicate KVM hypercall interface */
1209     _FDT(fdt_setprop_string(fdt, hypervisor, "compatible", "linux,kvm"));
1210     if (kvmppc_has_cap_fixup_hcalls()) {
1211         /*
1212          * Older KVM versions with older guest kernels were broken
1213          * with the magic page, don't allow the guest to map it.
1214          */
1215         if (!kvmppc_get_hypercall(first_cpu->env_ptr, hypercall,
1216                                   sizeof(hypercall))) {
1217             _FDT(fdt_setprop(fdt, hypervisor, "hcall-instructions",
1218                              hypercall, sizeof(hypercall)));
1219         }
1220     }
1221 }
1222 
1223 static void *spapr_build_fdt(sPAPRMachineState *spapr,
1224                              hwaddr rtas_addr,
1225                              hwaddr rtas_size)
1226 {
1227     MachineState *machine = MACHINE(spapr);
1228     MachineClass *mc = MACHINE_GET_CLASS(machine);
1229     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1230     int ret;
1231     void *fdt;
1232     sPAPRPHBState *phb;
1233     char *buf;
1234 
1235     fdt = g_malloc0(FDT_MAX_SIZE);
1236     _FDT((fdt_create_empty_tree(fdt, FDT_MAX_SIZE)));
1237 
1238     /* Root node */
1239     _FDT(fdt_setprop_string(fdt, 0, "device_type", "chrp"));
1240     _FDT(fdt_setprop_string(fdt, 0, "model", "IBM pSeries (emulated by qemu)"));
1241     _FDT(fdt_setprop_string(fdt, 0, "compatible", "qemu,pseries"));
1242 
1243     /*
1244      * Add info to guest to indentify which host is it being run on
1245      * and what is the uuid of the guest
1246      */
1247     if (kvmppc_get_host_model(&buf)) {
1248         _FDT(fdt_setprop_string(fdt, 0, "host-model", buf));
1249         g_free(buf);
1250     }
1251     if (kvmppc_get_host_serial(&buf)) {
1252         _FDT(fdt_setprop_string(fdt, 0, "host-serial", buf));
1253         g_free(buf);
1254     }
1255 
1256     buf = qemu_uuid_unparse_strdup(&qemu_uuid);
1257 
1258     _FDT(fdt_setprop_string(fdt, 0, "vm,uuid", buf));
1259     if (qemu_uuid_set) {
1260         _FDT(fdt_setprop_string(fdt, 0, "system-id", buf));
1261     }
1262     g_free(buf);
1263 
1264     if (qemu_get_vm_name()) {
1265         _FDT(fdt_setprop_string(fdt, 0, "ibm,partition-name",
1266                                 qemu_get_vm_name()));
1267     }
1268 
1269     _FDT(fdt_setprop_cell(fdt, 0, "#address-cells", 2));
1270     _FDT(fdt_setprop_cell(fdt, 0, "#size-cells", 2));
1271 
1272     /* /interrupt controller */
1273     spapr_dt_xics(xics_max_server_number(spapr), fdt, PHANDLE_XICP);
1274 
1275     ret = spapr_populate_memory(spapr, fdt);
1276     if (ret < 0) {
1277         error_report("couldn't setup memory nodes in fdt");
1278         exit(1);
1279     }
1280 
1281     /* /vdevice */
1282     spapr_dt_vdevice(spapr->vio_bus, fdt);
1283 
1284     if (object_resolve_path_type("", TYPE_SPAPR_RNG, NULL)) {
1285         ret = spapr_rng_populate_dt(fdt);
1286         if (ret < 0) {
1287             error_report("could not set up rng device in the fdt");
1288             exit(1);
1289         }
1290     }
1291 
1292     QLIST_FOREACH(phb, &spapr->phbs, list) {
1293         ret = spapr_populate_pci_dt(phb, PHANDLE_XICP, fdt, smc->irq->nr_msis);
1294         if (ret < 0) {
1295             error_report("couldn't setup PCI devices in fdt");
1296             exit(1);
1297         }
1298     }
1299 
1300     /* cpus */
1301     spapr_populate_cpus_dt_node(fdt, spapr);
1302 
1303     if (smc->dr_lmb_enabled) {
1304         _FDT(spapr_drc_populate_dt(fdt, 0, NULL, SPAPR_DR_CONNECTOR_TYPE_LMB));
1305     }
1306 
1307     if (mc->has_hotpluggable_cpus) {
1308         int offset = fdt_path_offset(fdt, "/cpus");
1309         ret = spapr_drc_populate_dt(fdt, offset, NULL,
1310                                     SPAPR_DR_CONNECTOR_TYPE_CPU);
1311         if (ret < 0) {
1312             error_report("Couldn't set up CPU DR device tree properties");
1313             exit(1);
1314         }
1315     }
1316 
1317     /* /event-sources */
1318     spapr_dt_events(spapr, fdt);
1319 
1320     /* /rtas */
1321     spapr_dt_rtas(spapr, fdt);
1322 
1323     /* /chosen */
1324     spapr_dt_chosen(spapr, fdt);
1325 
1326     /* /hypervisor */
1327     if (kvm_enabled()) {
1328         spapr_dt_hypervisor(spapr, fdt);
1329     }
1330 
1331     /* Build memory reserve map */
1332     if (spapr->kernel_size) {
1333         _FDT((fdt_add_mem_rsv(fdt, KERNEL_LOAD_ADDR, spapr->kernel_size)));
1334     }
1335     if (spapr->initrd_size) {
1336         _FDT((fdt_add_mem_rsv(fdt, spapr->initrd_base, spapr->initrd_size)));
1337     }
1338 
1339     /* ibm,client-architecture-support updates */
1340     ret = spapr_dt_cas_updates(spapr, fdt, spapr->ov5_cas);
1341     if (ret < 0) {
1342         error_report("couldn't setup CAS properties fdt");
1343         exit(1);
1344     }
1345 
1346     return fdt;
1347 }
1348 
1349 static uint64_t translate_kernel_address(void *opaque, uint64_t addr)
1350 {
1351     return (addr & 0x0fffffff) + KERNEL_LOAD_ADDR;
1352 }
1353 
1354 static void emulate_spapr_hypercall(PPCVirtualHypervisor *vhyp,
1355                                     PowerPCCPU *cpu)
1356 {
1357     CPUPPCState *env = &cpu->env;
1358 
1359     /* The TCG path should also be holding the BQL at this point */
1360     g_assert(qemu_mutex_iothread_locked());
1361 
1362     if (msr_pr) {
1363         hcall_dprintf("Hypercall made with MSR[PR]=1\n");
1364         env->gpr[3] = H_PRIVILEGE;
1365     } else {
1366         env->gpr[3] = spapr_hypercall(cpu, env->gpr[3], &env->gpr[4]);
1367     }
1368 }
1369 
1370 static uint64_t spapr_get_patbe(PPCVirtualHypervisor *vhyp)
1371 {
1372     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1373 
1374     return spapr->patb_entry;
1375 }
1376 
1377 #define HPTE(_table, _i)   (void *)(((uint64_t *)(_table)) + ((_i) * 2))
1378 #define HPTE_VALID(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_VALID)
1379 #define HPTE_DIRTY(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_HPTE_DIRTY)
1380 #define CLEAN_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) &= tswap64(~HPTE64_V_HPTE_DIRTY))
1381 #define DIRTY_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) |= tswap64(HPTE64_V_HPTE_DIRTY))
1382 
1383 /*
1384  * Get the fd to access the kernel htab, re-opening it if necessary
1385  */
1386 static int get_htab_fd(sPAPRMachineState *spapr)
1387 {
1388     Error *local_err = NULL;
1389 
1390     if (spapr->htab_fd >= 0) {
1391         return spapr->htab_fd;
1392     }
1393 
1394     spapr->htab_fd = kvmppc_get_htab_fd(false, 0, &local_err);
1395     if (spapr->htab_fd < 0) {
1396         error_report_err(local_err);
1397     }
1398 
1399     return spapr->htab_fd;
1400 }
1401 
1402 void close_htab_fd(sPAPRMachineState *spapr)
1403 {
1404     if (spapr->htab_fd >= 0) {
1405         close(spapr->htab_fd);
1406     }
1407     spapr->htab_fd = -1;
1408 }
1409 
1410 static hwaddr spapr_hpt_mask(PPCVirtualHypervisor *vhyp)
1411 {
1412     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1413 
1414     return HTAB_SIZE(spapr) / HASH_PTEG_SIZE_64 - 1;
1415 }
1416 
1417 static target_ulong spapr_encode_hpt_for_kvm_pr(PPCVirtualHypervisor *vhyp)
1418 {
1419     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1420 
1421     assert(kvm_enabled());
1422 
1423     if (!spapr->htab) {
1424         return 0;
1425     }
1426 
1427     return (target_ulong)(uintptr_t)spapr->htab | (spapr->htab_shift - 18);
1428 }
1429 
1430 static const ppc_hash_pte64_t *spapr_map_hptes(PPCVirtualHypervisor *vhyp,
1431                                                 hwaddr ptex, int n)
1432 {
1433     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1434     hwaddr pte_offset = ptex * HASH_PTE_SIZE_64;
1435 
1436     if (!spapr->htab) {
1437         /*
1438          * HTAB is controlled by KVM. Fetch into temporary buffer
1439          */
1440         ppc_hash_pte64_t *hptes = g_malloc(n * HASH_PTE_SIZE_64);
1441         kvmppc_read_hptes(hptes, ptex, n);
1442         return hptes;
1443     }
1444 
1445     /*
1446      * HTAB is controlled by QEMU. Just point to the internally
1447      * accessible PTEG.
1448      */
1449     return (const ppc_hash_pte64_t *)(spapr->htab + pte_offset);
1450 }
1451 
1452 static void spapr_unmap_hptes(PPCVirtualHypervisor *vhyp,
1453                               const ppc_hash_pte64_t *hptes,
1454                               hwaddr ptex, int n)
1455 {
1456     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1457 
1458     if (!spapr->htab) {
1459         g_free((void *)hptes);
1460     }
1461 
1462     /* Nothing to do for qemu managed HPT */
1463 }
1464 
1465 static void spapr_store_hpte(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1466                              uint64_t pte0, uint64_t pte1)
1467 {
1468     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1469     hwaddr offset = ptex * HASH_PTE_SIZE_64;
1470 
1471     if (!spapr->htab) {
1472         kvmppc_write_hpte(ptex, pte0, pte1);
1473     } else {
1474         stq_p(spapr->htab + offset, pte0);
1475         stq_p(spapr->htab + offset + HASH_PTE_SIZE_64 / 2, pte1);
1476     }
1477 }
1478 
1479 int spapr_hpt_shift_for_ramsize(uint64_t ramsize)
1480 {
1481     int shift;
1482 
1483     /* We aim for a hash table of size 1/128 the size of RAM (rounded
1484      * up).  The PAPR recommendation is actually 1/64 of RAM size, but
1485      * that's much more than is needed for Linux guests */
1486     shift = ctz64(pow2ceil(ramsize)) - 7;
1487     shift = MAX(shift, 18); /* Minimum architected size */
1488     shift = MIN(shift, 46); /* Maximum architected size */
1489     return shift;
1490 }
1491 
1492 void spapr_free_hpt(sPAPRMachineState *spapr)
1493 {
1494     g_free(spapr->htab);
1495     spapr->htab = NULL;
1496     spapr->htab_shift = 0;
1497     close_htab_fd(spapr);
1498 }
1499 
1500 void spapr_reallocate_hpt(sPAPRMachineState *spapr, int shift,
1501                           Error **errp)
1502 {
1503     long rc;
1504 
1505     /* Clean up any HPT info from a previous boot */
1506     spapr_free_hpt(spapr);
1507 
1508     rc = kvmppc_reset_htab(shift);
1509     if (rc < 0) {
1510         /* kernel-side HPT needed, but couldn't allocate one */
1511         error_setg_errno(errp, errno,
1512                          "Failed to allocate KVM HPT of order %d (try smaller maxmem?)",
1513                          shift);
1514         /* This is almost certainly fatal, but if the caller really
1515          * wants to carry on with shift == 0, it's welcome to try */
1516     } else if (rc > 0) {
1517         /* kernel-side HPT allocated */
1518         if (rc != shift) {
1519             error_setg(errp,
1520                        "Requested order %d HPT, but kernel allocated order %ld (try smaller maxmem?)",
1521                        shift, rc);
1522         }
1523 
1524         spapr->htab_shift = shift;
1525         spapr->htab = NULL;
1526     } else {
1527         /* kernel-side HPT not needed, allocate in userspace instead */
1528         size_t size = 1ULL << shift;
1529         int i;
1530 
1531         spapr->htab = qemu_memalign(size, size);
1532         if (!spapr->htab) {
1533             error_setg_errno(errp, errno,
1534                              "Could not allocate HPT of order %d", shift);
1535             return;
1536         }
1537 
1538         memset(spapr->htab, 0, size);
1539         spapr->htab_shift = shift;
1540 
1541         for (i = 0; i < size / HASH_PTE_SIZE_64; i++) {
1542             DIRTY_HPTE(HPTE(spapr->htab, i));
1543         }
1544     }
1545     /* We're setting up a hash table, so that means we're not radix */
1546     spapr->patb_entry = 0;
1547 }
1548 
1549 void spapr_setup_hpt_and_vrma(sPAPRMachineState *spapr)
1550 {
1551     int hpt_shift;
1552 
1553     if ((spapr->resize_hpt == SPAPR_RESIZE_HPT_DISABLED)
1554         || (spapr->cas_reboot
1555             && !spapr_ovec_test(spapr->ov5_cas, OV5_HPT_RESIZE))) {
1556         hpt_shift = spapr_hpt_shift_for_ramsize(MACHINE(spapr)->maxram_size);
1557     } else {
1558         uint64_t current_ram_size;
1559 
1560         current_ram_size = MACHINE(spapr)->ram_size + get_plugged_memory_size();
1561         hpt_shift = spapr_hpt_shift_for_ramsize(current_ram_size);
1562     }
1563     spapr_reallocate_hpt(spapr, hpt_shift, &error_fatal);
1564 
1565     if (spapr->vrma_adjust) {
1566         spapr->rma_size = kvmppc_rma_size(spapr_node0_size(MACHINE(spapr)),
1567                                           spapr->htab_shift);
1568     }
1569 }
1570 
1571 static int spapr_reset_drcs(Object *child, void *opaque)
1572 {
1573     sPAPRDRConnector *drc =
1574         (sPAPRDRConnector *) object_dynamic_cast(child,
1575                                                  TYPE_SPAPR_DR_CONNECTOR);
1576 
1577     if (drc) {
1578         spapr_drc_reset(drc);
1579     }
1580 
1581     return 0;
1582 }
1583 
1584 static void spapr_machine_reset(void)
1585 {
1586     MachineState *machine = MACHINE(qdev_get_machine());
1587     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
1588     PowerPCCPU *first_ppc_cpu;
1589     uint32_t rtas_limit;
1590     hwaddr rtas_addr, fdt_addr;
1591     void *fdt;
1592     int rc;
1593 
1594     spapr_caps_apply(spapr);
1595 
1596     first_ppc_cpu = POWERPC_CPU(first_cpu);
1597     if (kvm_enabled() && kvmppc_has_cap_mmu_radix() &&
1598         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
1599                               spapr->max_compat_pvr)) {
1600         /* If using KVM with radix mode available, VCPUs can be started
1601          * without a HPT because KVM will start them in radix mode.
1602          * Set the GR bit in PATB so that we know there is no HPT. */
1603         spapr->patb_entry = PATBE1_GR;
1604     } else {
1605         spapr_setup_hpt_and_vrma(spapr);
1606     }
1607 
1608     /* if this reset wasn't generated by CAS, we should reset our
1609      * negotiated options and start from scratch */
1610     if (!spapr->cas_reboot) {
1611         spapr_ovec_cleanup(spapr->ov5_cas);
1612         spapr->ov5_cas = spapr_ovec_new();
1613 
1614         ppc_set_compat(first_ppc_cpu, spapr->max_compat_pvr, &error_fatal);
1615     }
1616 
1617     if (!SPAPR_MACHINE_GET_CLASS(spapr)->legacy_irq_allocation) {
1618         spapr_irq_msi_reset(spapr);
1619     }
1620 
1621     qemu_devices_reset();
1622 
1623     /* DRC reset may cause a device to be unplugged. This will cause troubles
1624      * if this device is used by another device (eg, a running vhost backend
1625      * will crash QEMU if the DIMM holding the vring goes away). To avoid such
1626      * situations, we reset DRCs after all devices have been reset.
1627      */
1628     object_child_foreach_recursive(object_get_root(), spapr_reset_drcs, NULL);
1629 
1630     spapr_clear_pending_events(spapr);
1631 
1632     /*
1633      * We place the device tree and RTAS just below either the top of the RMA,
1634      * or just below 2GB, whichever is lowere, so that it can be
1635      * processed with 32-bit real mode code if necessary
1636      */
1637     rtas_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR);
1638     rtas_addr = rtas_limit - RTAS_MAX_SIZE;
1639     fdt_addr = rtas_addr - FDT_MAX_SIZE;
1640 
1641     fdt = spapr_build_fdt(spapr, rtas_addr, spapr->rtas_size);
1642 
1643     spapr_load_rtas(spapr, fdt, rtas_addr);
1644 
1645     rc = fdt_pack(fdt);
1646 
1647     /* Should only fail if we've built a corrupted tree */
1648     assert(rc == 0);
1649 
1650     if (fdt_totalsize(fdt) > FDT_MAX_SIZE) {
1651         error_report("FDT too big ! 0x%x bytes (max is 0x%x)",
1652                      fdt_totalsize(fdt), FDT_MAX_SIZE);
1653         exit(1);
1654     }
1655 
1656     /* Load the fdt */
1657     qemu_fdt_dumpdtb(fdt, fdt_totalsize(fdt));
1658     cpu_physical_memory_write(fdt_addr, fdt, fdt_totalsize(fdt));
1659     g_free(fdt);
1660 
1661     /* Set up the entry state */
1662     spapr_cpu_set_entry_state(first_ppc_cpu, SPAPR_ENTRY_POINT, fdt_addr);
1663     first_ppc_cpu->env.gpr[5] = 0;
1664 
1665     spapr->cas_reboot = false;
1666 }
1667 
1668 static void spapr_create_nvram(sPAPRMachineState *spapr)
1669 {
1670     DeviceState *dev = qdev_create(&spapr->vio_bus->bus, "spapr-nvram");
1671     DriveInfo *dinfo = drive_get(IF_PFLASH, 0, 0);
1672 
1673     if (dinfo) {
1674         qdev_prop_set_drive(dev, "drive", blk_by_legacy_dinfo(dinfo),
1675                             &error_fatal);
1676     }
1677 
1678     qdev_init_nofail(dev);
1679 
1680     spapr->nvram = (struct sPAPRNVRAM *)dev;
1681 }
1682 
1683 static void spapr_rtc_create(sPAPRMachineState *spapr)
1684 {
1685     object_initialize(&spapr->rtc, sizeof(spapr->rtc), TYPE_SPAPR_RTC);
1686     object_property_add_child(OBJECT(spapr), "rtc", OBJECT(&spapr->rtc),
1687                               &error_fatal);
1688     object_property_set_bool(OBJECT(&spapr->rtc), true, "realized",
1689                               &error_fatal);
1690     object_property_add_alias(OBJECT(spapr), "rtc-time", OBJECT(&spapr->rtc),
1691                               "date", &error_fatal);
1692 }
1693 
1694 /* Returns whether we want to use VGA or not */
1695 static bool spapr_vga_init(PCIBus *pci_bus, Error **errp)
1696 {
1697     switch (vga_interface_type) {
1698     case VGA_NONE:
1699         return false;
1700     case VGA_DEVICE:
1701         return true;
1702     case VGA_STD:
1703     case VGA_VIRTIO:
1704         return pci_vga_init(pci_bus) != NULL;
1705     default:
1706         error_setg(errp,
1707                    "Unsupported VGA mode, only -vga std or -vga virtio is supported");
1708         return false;
1709     }
1710 }
1711 
1712 static int spapr_pre_load(void *opaque)
1713 {
1714     int rc;
1715 
1716     rc = spapr_caps_pre_load(opaque);
1717     if (rc) {
1718         return rc;
1719     }
1720 
1721     return 0;
1722 }
1723 
1724 static int spapr_post_load(void *opaque, int version_id)
1725 {
1726     sPAPRMachineState *spapr = (sPAPRMachineState *)opaque;
1727     int err = 0;
1728 
1729     err = spapr_caps_post_migration(spapr);
1730     if (err) {
1731         return err;
1732     }
1733 
1734     if (!object_dynamic_cast(OBJECT(spapr->ics), TYPE_ICS_KVM)) {
1735         CPUState *cs;
1736         CPU_FOREACH(cs) {
1737             PowerPCCPU *cpu = POWERPC_CPU(cs);
1738             icp_resend(ICP(cpu->intc));
1739         }
1740     }
1741 
1742     /* In earlier versions, there was no separate qdev for the PAPR
1743      * RTC, so the RTC offset was stored directly in sPAPREnvironment.
1744      * So when migrating from those versions, poke the incoming offset
1745      * value into the RTC device */
1746     if (version_id < 3) {
1747         err = spapr_rtc_import_offset(&spapr->rtc, spapr->rtc_offset);
1748     }
1749 
1750     if (kvm_enabled() && spapr->patb_entry) {
1751         PowerPCCPU *cpu = POWERPC_CPU(first_cpu);
1752         bool radix = !!(spapr->patb_entry & PATBE1_GR);
1753         bool gtse = !!(cpu->env.spr[SPR_LPCR] & LPCR_GTSE);
1754 
1755         err = kvmppc_configure_v3_mmu(cpu, radix, gtse, spapr->patb_entry);
1756         if (err) {
1757             error_report("Process table config unsupported by the host");
1758             return -EINVAL;
1759         }
1760     }
1761 
1762     return err;
1763 }
1764 
1765 static int spapr_pre_save(void *opaque)
1766 {
1767     int rc;
1768 
1769     rc = spapr_caps_pre_save(opaque);
1770     if (rc) {
1771         return rc;
1772     }
1773 
1774     return 0;
1775 }
1776 
1777 static bool version_before_3(void *opaque, int version_id)
1778 {
1779     return version_id < 3;
1780 }
1781 
1782 static bool spapr_pending_events_needed(void *opaque)
1783 {
1784     sPAPRMachineState *spapr = (sPAPRMachineState *)opaque;
1785     return !QTAILQ_EMPTY(&spapr->pending_events);
1786 }
1787 
1788 static const VMStateDescription vmstate_spapr_event_entry = {
1789     .name = "spapr_event_log_entry",
1790     .version_id = 1,
1791     .minimum_version_id = 1,
1792     .fields = (VMStateField[]) {
1793         VMSTATE_UINT32(summary, sPAPREventLogEntry),
1794         VMSTATE_UINT32(extended_length, sPAPREventLogEntry),
1795         VMSTATE_VBUFFER_ALLOC_UINT32(extended_log, sPAPREventLogEntry, 0,
1796                                      NULL, extended_length),
1797         VMSTATE_END_OF_LIST()
1798     },
1799 };
1800 
1801 static const VMStateDescription vmstate_spapr_pending_events = {
1802     .name = "spapr_pending_events",
1803     .version_id = 1,
1804     .minimum_version_id = 1,
1805     .needed = spapr_pending_events_needed,
1806     .fields = (VMStateField[]) {
1807         VMSTATE_QTAILQ_V(pending_events, sPAPRMachineState, 1,
1808                          vmstate_spapr_event_entry, sPAPREventLogEntry, next),
1809         VMSTATE_END_OF_LIST()
1810     },
1811 };
1812 
1813 static bool spapr_ov5_cas_needed(void *opaque)
1814 {
1815     sPAPRMachineState *spapr = opaque;
1816     sPAPROptionVector *ov5_mask = spapr_ovec_new();
1817     sPAPROptionVector *ov5_legacy = spapr_ovec_new();
1818     sPAPROptionVector *ov5_removed = spapr_ovec_new();
1819     bool cas_needed;
1820 
1821     /* Prior to the introduction of sPAPROptionVector, we had two option
1822      * vectors we dealt with: OV5_FORM1_AFFINITY, and OV5_DRCONF_MEMORY.
1823      * Both of these options encode machine topology into the device-tree
1824      * in such a way that the now-booted OS should still be able to interact
1825      * appropriately with QEMU regardless of what options were actually
1826      * negotiatied on the source side.
1827      *
1828      * As such, we can avoid migrating the CAS-negotiated options if these
1829      * are the only options available on the current machine/platform.
1830      * Since these are the only options available for pseries-2.7 and
1831      * earlier, this allows us to maintain old->new/new->old migration
1832      * compatibility.
1833      *
1834      * For QEMU 2.8+, there are additional CAS-negotiatable options available
1835      * via default pseries-2.8 machines and explicit command-line parameters.
1836      * Some of these options, like OV5_HP_EVT, *do* require QEMU to be aware
1837      * of the actual CAS-negotiated values to continue working properly. For
1838      * example, availability of memory unplug depends on knowing whether
1839      * OV5_HP_EVT was negotiated via CAS.
1840      *
1841      * Thus, for any cases where the set of available CAS-negotiatable
1842      * options extends beyond OV5_FORM1_AFFINITY and OV5_DRCONF_MEMORY, we
1843      * include the CAS-negotiated options in the migration stream, unless
1844      * if they affect boot time behaviour only.
1845      */
1846     spapr_ovec_set(ov5_mask, OV5_FORM1_AFFINITY);
1847     spapr_ovec_set(ov5_mask, OV5_DRCONF_MEMORY);
1848     spapr_ovec_set(ov5_mask, OV5_DRMEM_V2);
1849 
1850     /* spapr_ovec_diff returns true if bits were removed. we avoid using
1851      * the mask itself since in the future it's possible "legacy" bits may be
1852      * removed via machine options, which could generate a false positive
1853      * that breaks migration.
1854      */
1855     spapr_ovec_intersect(ov5_legacy, spapr->ov5, ov5_mask);
1856     cas_needed = spapr_ovec_diff(ov5_removed, spapr->ov5, ov5_legacy);
1857 
1858     spapr_ovec_cleanup(ov5_mask);
1859     spapr_ovec_cleanup(ov5_legacy);
1860     spapr_ovec_cleanup(ov5_removed);
1861 
1862     return cas_needed;
1863 }
1864 
1865 static const VMStateDescription vmstate_spapr_ov5_cas = {
1866     .name = "spapr_option_vector_ov5_cas",
1867     .version_id = 1,
1868     .minimum_version_id = 1,
1869     .needed = spapr_ov5_cas_needed,
1870     .fields = (VMStateField[]) {
1871         VMSTATE_STRUCT_POINTER_V(ov5_cas, sPAPRMachineState, 1,
1872                                  vmstate_spapr_ovec, sPAPROptionVector),
1873         VMSTATE_END_OF_LIST()
1874     },
1875 };
1876 
1877 static bool spapr_patb_entry_needed(void *opaque)
1878 {
1879     sPAPRMachineState *spapr = opaque;
1880 
1881     return !!spapr->patb_entry;
1882 }
1883 
1884 static const VMStateDescription vmstate_spapr_patb_entry = {
1885     .name = "spapr_patb_entry",
1886     .version_id = 1,
1887     .minimum_version_id = 1,
1888     .needed = spapr_patb_entry_needed,
1889     .fields = (VMStateField[]) {
1890         VMSTATE_UINT64(patb_entry, sPAPRMachineState),
1891         VMSTATE_END_OF_LIST()
1892     },
1893 };
1894 
1895 static bool spapr_irq_map_needed(void *opaque)
1896 {
1897     sPAPRMachineState *spapr = opaque;
1898 
1899     return spapr->irq_map && !bitmap_empty(spapr->irq_map, spapr->irq_map_nr);
1900 }
1901 
1902 static const VMStateDescription vmstate_spapr_irq_map = {
1903     .name = "spapr_irq_map",
1904     .version_id = 1,
1905     .minimum_version_id = 1,
1906     .needed = spapr_irq_map_needed,
1907     .fields = (VMStateField[]) {
1908         VMSTATE_BITMAP(irq_map, sPAPRMachineState, 0, irq_map_nr),
1909         VMSTATE_END_OF_LIST()
1910     },
1911 };
1912 
1913 static const VMStateDescription vmstate_spapr = {
1914     .name = "spapr",
1915     .version_id = 3,
1916     .minimum_version_id = 1,
1917     .pre_load = spapr_pre_load,
1918     .post_load = spapr_post_load,
1919     .pre_save = spapr_pre_save,
1920     .fields = (VMStateField[]) {
1921         /* used to be @next_irq */
1922         VMSTATE_UNUSED_BUFFER(version_before_3, 0, 4),
1923 
1924         /* RTC offset */
1925         VMSTATE_UINT64_TEST(rtc_offset, sPAPRMachineState, version_before_3),
1926 
1927         VMSTATE_PPC_TIMEBASE_V(tb, sPAPRMachineState, 2),
1928         VMSTATE_END_OF_LIST()
1929     },
1930     .subsections = (const VMStateDescription*[]) {
1931         &vmstate_spapr_ov5_cas,
1932         &vmstate_spapr_patb_entry,
1933         &vmstate_spapr_pending_events,
1934         &vmstate_spapr_cap_htm,
1935         &vmstate_spapr_cap_vsx,
1936         &vmstate_spapr_cap_dfp,
1937         &vmstate_spapr_cap_cfpc,
1938         &vmstate_spapr_cap_sbbc,
1939         &vmstate_spapr_cap_ibs,
1940         &vmstate_spapr_irq_map,
1941         NULL
1942     }
1943 };
1944 
1945 static int htab_save_setup(QEMUFile *f, void *opaque)
1946 {
1947     sPAPRMachineState *spapr = opaque;
1948 
1949     /* "Iteration" header */
1950     if (!spapr->htab_shift) {
1951         qemu_put_be32(f, -1);
1952     } else {
1953         qemu_put_be32(f, spapr->htab_shift);
1954     }
1955 
1956     if (spapr->htab) {
1957         spapr->htab_save_index = 0;
1958         spapr->htab_first_pass = true;
1959     } else {
1960         if (spapr->htab_shift) {
1961             assert(kvm_enabled());
1962         }
1963     }
1964 
1965 
1966     return 0;
1967 }
1968 
1969 static void htab_save_chunk(QEMUFile *f, sPAPRMachineState *spapr,
1970                             int chunkstart, int n_valid, int n_invalid)
1971 {
1972     qemu_put_be32(f, chunkstart);
1973     qemu_put_be16(f, n_valid);
1974     qemu_put_be16(f, n_invalid);
1975     qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
1976                     HASH_PTE_SIZE_64 * n_valid);
1977 }
1978 
1979 static void htab_save_end_marker(QEMUFile *f)
1980 {
1981     qemu_put_be32(f, 0);
1982     qemu_put_be16(f, 0);
1983     qemu_put_be16(f, 0);
1984 }
1985 
1986 static void htab_save_first_pass(QEMUFile *f, sPAPRMachineState *spapr,
1987                                  int64_t max_ns)
1988 {
1989     bool has_timeout = max_ns != -1;
1990     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
1991     int index = spapr->htab_save_index;
1992     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
1993 
1994     assert(spapr->htab_first_pass);
1995 
1996     do {
1997         int chunkstart;
1998 
1999         /* Consume invalid HPTEs */
2000         while ((index < htabslots)
2001                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2002             CLEAN_HPTE(HPTE(spapr->htab, index));
2003             index++;
2004         }
2005 
2006         /* Consume valid HPTEs */
2007         chunkstart = index;
2008         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2009                && HPTE_VALID(HPTE(spapr->htab, index))) {
2010             CLEAN_HPTE(HPTE(spapr->htab, index));
2011             index++;
2012         }
2013 
2014         if (index > chunkstart) {
2015             int n_valid = index - chunkstart;
2016 
2017             htab_save_chunk(f, spapr, chunkstart, n_valid, 0);
2018 
2019             if (has_timeout &&
2020                 (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2021                 break;
2022             }
2023         }
2024     } while ((index < htabslots) && !qemu_file_rate_limit(f));
2025 
2026     if (index >= htabslots) {
2027         assert(index == htabslots);
2028         index = 0;
2029         spapr->htab_first_pass = false;
2030     }
2031     spapr->htab_save_index = index;
2032 }
2033 
2034 static int htab_save_later_pass(QEMUFile *f, sPAPRMachineState *spapr,
2035                                 int64_t max_ns)
2036 {
2037     bool final = max_ns < 0;
2038     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2039     int examined = 0, sent = 0;
2040     int index = spapr->htab_save_index;
2041     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2042 
2043     assert(!spapr->htab_first_pass);
2044 
2045     do {
2046         int chunkstart, invalidstart;
2047 
2048         /* Consume non-dirty HPTEs */
2049         while ((index < htabslots)
2050                && !HPTE_DIRTY(HPTE(spapr->htab, index))) {
2051             index++;
2052             examined++;
2053         }
2054 
2055         chunkstart = index;
2056         /* Consume valid dirty HPTEs */
2057         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2058                && HPTE_DIRTY(HPTE(spapr->htab, index))
2059                && HPTE_VALID(HPTE(spapr->htab, index))) {
2060             CLEAN_HPTE(HPTE(spapr->htab, index));
2061             index++;
2062             examined++;
2063         }
2064 
2065         invalidstart = index;
2066         /* Consume invalid dirty HPTEs */
2067         while ((index < htabslots) && (index - invalidstart < USHRT_MAX)
2068                && HPTE_DIRTY(HPTE(spapr->htab, index))
2069                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2070             CLEAN_HPTE(HPTE(spapr->htab, index));
2071             index++;
2072             examined++;
2073         }
2074 
2075         if (index > chunkstart) {
2076             int n_valid = invalidstart - chunkstart;
2077             int n_invalid = index - invalidstart;
2078 
2079             htab_save_chunk(f, spapr, chunkstart, n_valid, n_invalid);
2080             sent += index - chunkstart;
2081 
2082             if (!final && (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2083                 break;
2084             }
2085         }
2086 
2087         if (examined >= htabslots) {
2088             break;
2089         }
2090 
2091         if (index >= htabslots) {
2092             assert(index == htabslots);
2093             index = 0;
2094         }
2095     } while ((examined < htabslots) && (!qemu_file_rate_limit(f) || final));
2096 
2097     if (index >= htabslots) {
2098         assert(index == htabslots);
2099         index = 0;
2100     }
2101 
2102     spapr->htab_save_index = index;
2103 
2104     return (examined >= htabslots) && (sent == 0) ? 1 : 0;
2105 }
2106 
2107 #define MAX_ITERATION_NS    5000000 /* 5 ms */
2108 #define MAX_KVM_BUF_SIZE    2048
2109 
2110 static int htab_save_iterate(QEMUFile *f, void *opaque)
2111 {
2112     sPAPRMachineState *spapr = opaque;
2113     int fd;
2114     int rc = 0;
2115 
2116     /* Iteration header */
2117     if (!spapr->htab_shift) {
2118         qemu_put_be32(f, -1);
2119         return 1;
2120     } else {
2121         qemu_put_be32(f, 0);
2122     }
2123 
2124     if (!spapr->htab) {
2125         assert(kvm_enabled());
2126 
2127         fd = get_htab_fd(spapr);
2128         if (fd < 0) {
2129             return fd;
2130         }
2131 
2132         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, MAX_ITERATION_NS);
2133         if (rc < 0) {
2134             return rc;
2135         }
2136     } else  if (spapr->htab_first_pass) {
2137         htab_save_first_pass(f, spapr, MAX_ITERATION_NS);
2138     } else {
2139         rc = htab_save_later_pass(f, spapr, MAX_ITERATION_NS);
2140     }
2141 
2142     htab_save_end_marker(f);
2143 
2144     return rc;
2145 }
2146 
2147 static int htab_save_complete(QEMUFile *f, void *opaque)
2148 {
2149     sPAPRMachineState *spapr = opaque;
2150     int fd;
2151 
2152     /* Iteration header */
2153     if (!spapr->htab_shift) {
2154         qemu_put_be32(f, -1);
2155         return 0;
2156     } else {
2157         qemu_put_be32(f, 0);
2158     }
2159 
2160     if (!spapr->htab) {
2161         int rc;
2162 
2163         assert(kvm_enabled());
2164 
2165         fd = get_htab_fd(spapr);
2166         if (fd < 0) {
2167             return fd;
2168         }
2169 
2170         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, -1);
2171         if (rc < 0) {
2172             return rc;
2173         }
2174     } else {
2175         if (spapr->htab_first_pass) {
2176             htab_save_first_pass(f, spapr, -1);
2177         }
2178         htab_save_later_pass(f, spapr, -1);
2179     }
2180 
2181     /* End marker */
2182     htab_save_end_marker(f);
2183 
2184     return 0;
2185 }
2186 
2187 static int htab_load(QEMUFile *f, void *opaque, int version_id)
2188 {
2189     sPAPRMachineState *spapr = opaque;
2190     uint32_t section_hdr;
2191     int fd = -1;
2192     Error *local_err = NULL;
2193 
2194     if (version_id < 1 || version_id > 1) {
2195         error_report("htab_load() bad version");
2196         return -EINVAL;
2197     }
2198 
2199     section_hdr = qemu_get_be32(f);
2200 
2201     if (section_hdr == -1) {
2202         spapr_free_hpt(spapr);
2203         return 0;
2204     }
2205 
2206     if (section_hdr) {
2207         /* First section gives the htab size */
2208         spapr_reallocate_hpt(spapr, section_hdr, &local_err);
2209         if (local_err) {
2210             error_report_err(local_err);
2211             return -EINVAL;
2212         }
2213         return 0;
2214     }
2215 
2216     if (!spapr->htab) {
2217         assert(kvm_enabled());
2218 
2219         fd = kvmppc_get_htab_fd(true, 0, &local_err);
2220         if (fd < 0) {
2221             error_report_err(local_err);
2222             return fd;
2223         }
2224     }
2225 
2226     while (true) {
2227         uint32_t index;
2228         uint16_t n_valid, n_invalid;
2229 
2230         index = qemu_get_be32(f);
2231         n_valid = qemu_get_be16(f);
2232         n_invalid = qemu_get_be16(f);
2233 
2234         if ((index == 0) && (n_valid == 0) && (n_invalid == 0)) {
2235             /* End of Stream */
2236             break;
2237         }
2238 
2239         if ((index + n_valid + n_invalid) >
2240             (HTAB_SIZE(spapr) / HASH_PTE_SIZE_64)) {
2241             /* Bad index in stream */
2242             error_report(
2243                 "htab_load() bad index %d (%hd+%hd entries) in htab stream (htab_shift=%d)",
2244                 index, n_valid, n_invalid, spapr->htab_shift);
2245             return -EINVAL;
2246         }
2247 
2248         if (spapr->htab) {
2249             if (n_valid) {
2250                 qemu_get_buffer(f, HPTE(spapr->htab, index),
2251                                 HASH_PTE_SIZE_64 * n_valid);
2252             }
2253             if (n_invalid) {
2254                 memset(HPTE(spapr->htab, index + n_valid), 0,
2255                        HASH_PTE_SIZE_64 * n_invalid);
2256             }
2257         } else {
2258             int rc;
2259 
2260             assert(fd >= 0);
2261 
2262             rc = kvmppc_load_htab_chunk(f, fd, index, n_valid, n_invalid);
2263             if (rc < 0) {
2264                 return rc;
2265             }
2266         }
2267     }
2268 
2269     if (!spapr->htab) {
2270         assert(fd >= 0);
2271         close(fd);
2272     }
2273 
2274     return 0;
2275 }
2276 
2277 static void htab_save_cleanup(void *opaque)
2278 {
2279     sPAPRMachineState *spapr = opaque;
2280 
2281     close_htab_fd(spapr);
2282 }
2283 
2284 static SaveVMHandlers savevm_htab_handlers = {
2285     .save_setup = htab_save_setup,
2286     .save_live_iterate = htab_save_iterate,
2287     .save_live_complete_precopy = htab_save_complete,
2288     .save_cleanup = htab_save_cleanup,
2289     .load_state = htab_load,
2290 };
2291 
2292 static void spapr_boot_set(void *opaque, const char *boot_device,
2293                            Error **errp)
2294 {
2295     MachineState *machine = MACHINE(opaque);
2296     machine->boot_order = g_strdup(boot_device);
2297 }
2298 
2299 static void spapr_create_lmb_dr_connectors(sPAPRMachineState *spapr)
2300 {
2301     MachineState *machine = MACHINE(spapr);
2302     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
2303     uint32_t nr_lmbs = (machine->maxram_size - machine->ram_size)/lmb_size;
2304     int i;
2305 
2306     for (i = 0; i < nr_lmbs; i++) {
2307         uint64_t addr;
2308 
2309         addr = i * lmb_size + machine->device_memory->base;
2310         spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_LMB,
2311                                addr / lmb_size);
2312     }
2313 }
2314 
2315 /*
2316  * If RAM size, maxmem size and individual node mem sizes aren't aligned
2317  * to SPAPR_MEMORY_BLOCK_SIZE(256MB), then refuse to start the guest
2318  * since we can't support such unaligned sizes with DRCONF_MEMORY.
2319  */
2320 static void spapr_validate_node_memory(MachineState *machine, Error **errp)
2321 {
2322     int i;
2323 
2324     if (machine->ram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2325         error_setg(errp, "Memory size 0x" RAM_ADDR_FMT
2326                    " is not aligned to %" PRIu64 " MiB",
2327                    machine->ram_size,
2328                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2329         return;
2330     }
2331 
2332     if (machine->maxram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2333         error_setg(errp, "Maximum memory size 0x" RAM_ADDR_FMT
2334                    " is not aligned to %" PRIu64 " MiB",
2335                    machine->ram_size,
2336                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2337         return;
2338     }
2339 
2340     for (i = 0; i < nb_numa_nodes; i++) {
2341         if (numa_info[i].node_mem % SPAPR_MEMORY_BLOCK_SIZE) {
2342             error_setg(errp,
2343                        "Node %d memory size 0x%" PRIx64
2344                        " is not aligned to %" PRIu64 " MiB",
2345                        i, numa_info[i].node_mem,
2346                        SPAPR_MEMORY_BLOCK_SIZE / MiB);
2347             return;
2348         }
2349     }
2350 }
2351 
2352 /* find cpu slot in machine->possible_cpus by core_id */
2353 static CPUArchId *spapr_find_cpu_slot(MachineState *ms, uint32_t id, int *idx)
2354 {
2355     int index = id / smp_threads;
2356 
2357     if (index >= ms->possible_cpus->len) {
2358         return NULL;
2359     }
2360     if (idx) {
2361         *idx = index;
2362     }
2363     return &ms->possible_cpus->cpus[index];
2364 }
2365 
2366 static void spapr_set_vsmt_mode(sPAPRMachineState *spapr, Error **errp)
2367 {
2368     Error *local_err = NULL;
2369     bool vsmt_user = !!spapr->vsmt;
2370     int kvm_smt = kvmppc_smt_threads();
2371     int ret;
2372 
2373     if (!kvm_enabled() && (smp_threads > 1)) {
2374         error_setg(&local_err, "TCG cannot support more than 1 thread/core "
2375                      "on a pseries machine");
2376         goto out;
2377     }
2378     if (!is_power_of_2(smp_threads)) {
2379         error_setg(&local_err, "Cannot support %d threads/core on a pseries "
2380                      "machine because it must be a power of 2", smp_threads);
2381         goto out;
2382     }
2383 
2384     /* Detemine the VSMT mode to use: */
2385     if (vsmt_user) {
2386         if (spapr->vsmt < smp_threads) {
2387             error_setg(&local_err, "Cannot support VSMT mode %d"
2388                          " because it must be >= threads/core (%d)",
2389                          spapr->vsmt, smp_threads);
2390             goto out;
2391         }
2392         /* In this case, spapr->vsmt has been set by the command line */
2393     } else {
2394         /*
2395          * Default VSMT value is tricky, because we need it to be as
2396          * consistent as possible (for migration), but this requires
2397          * changing it for at least some existing cases.  We pick 8 as
2398          * the value that we'd get with KVM on POWER8, the
2399          * overwhelmingly common case in production systems.
2400          */
2401         spapr->vsmt = MAX(8, smp_threads);
2402     }
2403 
2404     /* KVM: If necessary, set the SMT mode: */
2405     if (kvm_enabled() && (spapr->vsmt != kvm_smt)) {
2406         ret = kvmppc_set_smt_threads(spapr->vsmt);
2407         if (ret) {
2408             /* Looks like KVM isn't able to change VSMT mode */
2409             error_setg(&local_err,
2410                        "Failed to set KVM's VSMT mode to %d (errno %d)",
2411                        spapr->vsmt, ret);
2412             /* We can live with that if the default one is big enough
2413              * for the number of threads, and a submultiple of the one
2414              * we want.  In this case we'll waste some vcpu ids, but
2415              * behaviour will be correct */
2416             if ((kvm_smt >= smp_threads) && ((spapr->vsmt % kvm_smt) == 0)) {
2417                 warn_report_err(local_err);
2418                 local_err = NULL;
2419                 goto out;
2420             } else {
2421                 if (!vsmt_user) {
2422                     error_append_hint(&local_err,
2423                                       "On PPC, a VM with %d threads/core"
2424                                       " on a host with %d threads/core"
2425                                       " requires the use of VSMT mode %d.\n",
2426                                       smp_threads, kvm_smt, spapr->vsmt);
2427                 }
2428                 kvmppc_hint_smt_possible(&local_err);
2429                 goto out;
2430             }
2431         }
2432     }
2433     /* else TCG: nothing to do currently */
2434 out:
2435     error_propagate(errp, local_err);
2436 }
2437 
2438 static void spapr_init_cpus(sPAPRMachineState *spapr)
2439 {
2440     MachineState *machine = MACHINE(spapr);
2441     MachineClass *mc = MACHINE_GET_CLASS(machine);
2442     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2443     const char *type = spapr_get_cpu_core_type(machine->cpu_type);
2444     const CPUArchIdList *possible_cpus;
2445     int boot_cores_nr = smp_cpus / smp_threads;
2446     int i;
2447 
2448     possible_cpus = mc->possible_cpu_arch_ids(machine);
2449     if (mc->has_hotpluggable_cpus) {
2450         if (smp_cpus % smp_threads) {
2451             error_report("smp_cpus (%u) must be multiple of threads (%u)",
2452                          smp_cpus, smp_threads);
2453             exit(1);
2454         }
2455         if (max_cpus % smp_threads) {
2456             error_report("max_cpus (%u) must be multiple of threads (%u)",
2457                          max_cpus, smp_threads);
2458             exit(1);
2459         }
2460     } else {
2461         if (max_cpus != smp_cpus) {
2462             error_report("This machine version does not support CPU hotplug");
2463             exit(1);
2464         }
2465         boot_cores_nr = possible_cpus->len;
2466     }
2467 
2468     /* VSMT must be set in order to be able to compute VCPU ids, ie to
2469      * call xics_max_server_number() or spapr_vcpu_id().
2470      */
2471     spapr_set_vsmt_mode(spapr, &error_fatal);
2472 
2473     if (smc->pre_2_10_has_unused_icps) {
2474         int i;
2475 
2476         for (i = 0; i < xics_max_server_number(spapr); i++) {
2477             /* Dummy entries get deregistered when real ICPState objects
2478              * are registered during CPU core hotplug.
2479              */
2480             pre_2_10_vmstate_register_dummy_icp(i);
2481         }
2482     }
2483 
2484     for (i = 0; i < possible_cpus->len; i++) {
2485         int core_id = i * smp_threads;
2486 
2487         if (mc->has_hotpluggable_cpus) {
2488             spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_CPU,
2489                                    spapr_vcpu_id(spapr, core_id));
2490         }
2491 
2492         if (i < boot_cores_nr) {
2493             Object *core  = object_new(type);
2494             int nr_threads = smp_threads;
2495 
2496             /* Handle the partially filled core for older machine types */
2497             if ((i + 1) * smp_threads >= smp_cpus) {
2498                 nr_threads = smp_cpus - i * smp_threads;
2499             }
2500 
2501             object_property_set_int(core, nr_threads, "nr-threads",
2502                                     &error_fatal);
2503             object_property_set_int(core, core_id, CPU_CORE_PROP_CORE_ID,
2504                                     &error_fatal);
2505             object_property_set_bool(core, true, "realized", &error_fatal);
2506 
2507             object_unref(core);
2508         }
2509     }
2510 }
2511 
2512 /* pSeries LPAR / sPAPR hardware init */
2513 static void spapr_machine_init(MachineState *machine)
2514 {
2515     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
2516     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2517     const char *kernel_filename = machine->kernel_filename;
2518     const char *initrd_filename = machine->initrd_filename;
2519     PCIHostState *phb;
2520     int i;
2521     MemoryRegion *sysmem = get_system_memory();
2522     MemoryRegion *ram = g_new(MemoryRegion, 1);
2523     hwaddr node0_size = spapr_node0_size(machine);
2524     long load_limit, fw_size;
2525     char *filename;
2526     Error *resize_hpt_err = NULL;
2527 
2528     msi_nonbroken = true;
2529 
2530     QLIST_INIT(&spapr->phbs);
2531     QTAILQ_INIT(&spapr->pending_dimm_unplugs);
2532 
2533     /* Determine capabilities to run with */
2534     spapr_caps_init(spapr);
2535 
2536     kvmppc_check_papr_resize_hpt(&resize_hpt_err);
2537     if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DEFAULT) {
2538         /*
2539          * If the user explicitly requested a mode we should either
2540          * supply it, or fail completely (which we do below).  But if
2541          * it's not set explicitly, we reset our mode to something
2542          * that works
2543          */
2544         if (resize_hpt_err) {
2545             spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
2546             error_free(resize_hpt_err);
2547             resize_hpt_err = NULL;
2548         } else {
2549             spapr->resize_hpt = smc->resize_hpt_default;
2550         }
2551     }
2552 
2553     assert(spapr->resize_hpt != SPAPR_RESIZE_HPT_DEFAULT);
2554 
2555     if ((spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) && resize_hpt_err) {
2556         /*
2557          * User requested HPT resize, but this host can't supply it.  Bail out
2558          */
2559         error_report_err(resize_hpt_err);
2560         exit(1);
2561     }
2562 
2563     spapr->rma_size = node0_size;
2564 
2565     /* With KVM, we don't actually know whether KVM supports an
2566      * unbounded RMA (PR KVM) or is limited by the hash table size
2567      * (HV KVM using VRMA), so we always assume the latter
2568      *
2569      * In that case, we also limit the initial allocations for RTAS
2570      * etc... to 256M since we have no way to know what the VRMA size
2571      * is going to be as it depends on the size of the hash table
2572      * which isn't determined yet.
2573      */
2574     if (kvm_enabled()) {
2575         spapr->vrma_adjust = 1;
2576         spapr->rma_size = MIN(spapr->rma_size, 0x10000000);
2577     }
2578 
2579     /* Actually we don't support unbounded RMA anymore since we added
2580      * proper emulation of HV mode. The max we can get is 16G which
2581      * also happens to be what we configure for PAPR mode so make sure
2582      * we don't do anything bigger than that
2583      */
2584     spapr->rma_size = MIN(spapr->rma_size, 0x400000000ull);
2585 
2586     if (spapr->rma_size > node0_size) {
2587         error_report("Numa node 0 has to span the RMA (%#08"HWADDR_PRIx")",
2588                      spapr->rma_size);
2589         exit(1);
2590     }
2591 
2592     /* Setup a load limit for the ramdisk leaving room for SLOF and FDT */
2593     load_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR) - FW_OVERHEAD;
2594 
2595     /* Set up Interrupt Controller before we create the VCPUs */
2596     smc->irq->init(spapr, &error_fatal);
2597 
2598     /* Set up containers for ibm,client-architecture-support negotiated options
2599      */
2600     spapr->ov5 = spapr_ovec_new();
2601     spapr->ov5_cas = spapr_ovec_new();
2602 
2603     if (smc->dr_lmb_enabled) {
2604         spapr_ovec_set(spapr->ov5, OV5_DRCONF_MEMORY);
2605         spapr_validate_node_memory(machine, &error_fatal);
2606     }
2607 
2608     spapr_ovec_set(spapr->ov5, OV5_FORM1_AFFINITY);
2609 
2610     /* advertise support for dedicated HP event source to guests */
2611     if (spapr->use_hotplug_event_source) {
2612         spapr_ovec_set(spapr->ov5, OV5_HP_EVT);
2613     }
2614 
2615     /* advertise support for HPT resizing */
2616     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
2617         spapr_ovec_set(spapr->ov5, OV5_HPT_RESIZE);
2618     }
2619 
2620     /* advertise support for ibm,dyamic-memory-v2 */
2621     spapr_ovec_set(spapr->ov5, OV5_DRMEM_V2);
2622 
2623     /* init CPUs */
2624     spapr_init_cpus(spapr);
2625 
2626     if ((!kvm_enabled() || kvmppc_has_cap_mmu_radix()) &&
2627         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
2628                               spapr->max_compat_pvr)) {
2629         /* KVM and TCG always allow GTSE with radix... */
2630         spapr_ovec_set(spapr->ov5, OV5_MMU_RADIX_GTSE);
2631     }
2632     /* ... but not with hash (currently). */
2633 
2634     if (kvm_enabled()) {
2635         /* Enable H_LOGICAL_CI_* so SLOF can talk to in-kernel devices */
2636         kvmppc_enable_logical_ci_hcalls();
2637         kvmppc_enable_set_mode_hcall();
2638 
2639         /* H_CLEAR_MOD/_REF are mandatory in PAPR, but off by default */
2640         kvmppc_enable_clear_ref_mod_hcalls();
2641     }
2642 
2643     /* allocate RAM */
2644     memory_region_allocate_system_memory(ram, NULL, "ppc_spapr.ram",
2645                                          machine->ram_size);
2646     memory_region_add_subregion(sysmem, 0, ram);
2647 
2648     /* always allocate the device memory information */
2649     machine->device_memory = g_malloc0(sizeof(*machine->device_memory));
2650 
2651     /* initialize hotplug memory address space */
2652     if (machine->ram_size < machine->maxram_size) {
2653         ram_addr_t device_mem_size = machine->maxram_size - machine->ram_size;
2654         /*
2655          * Limit the number of hotpluggable memory slots to half the number
2656          * slots that KVM supports, leaving the other half for PCI and other
2657          * devices. However ensure that number of slots doesn't drop below 32.
2658          */
2659         int max_memslots = kvm_enabled() ? kvm_get_max_memslots() / 2 :
2660                            SPAPR_MAX_RAM_SLOTS;
2661 
2662         if (max_memslots < SPAPR_MAX_RAM_SLOTS) {
2663             max_memslots = SPAPR_MAX_RAM_SLOTS;
2664         }
2665         if (machine->ram_slots > max_memslots) {
2666             error_report("Specified number of memory slots %"
2667                          PRIu64" exceeds max supported %d",
2668                          machine->ram_slots, max_memslots);
2669             exit(1);
2670         }
2671 
2672         machine->device_memory->base = ROUND_UP(machine->ram_size,
2673                                                 SPAPR_DEVICE_MEM_ALIGN);
2674         memory_region_init(&machine->device_memory->mr, OBJECT(spapr),
2675                            "device-memory", device_mem_size);
2676         memory_region_add_subregion(sysmem, machine->device_memory->base,
2677                                     &machine->device_memory->mr);
2678     }
2679 
2680     if (smc->dr_lmb_enabled) {
2681         spapr_create_lmb_dr_connectors(spapr);
2682     }
2683 
2684     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, "spapr-rtas.bin");
2685     if (!filename) {
2686         error_report("Could not find LPAR rtas '%s'", "spapr-rtas.bin");
2687         exit(1);
2688     }
2689     spapr->rtas_size = get_image_size(filename);
2690     if (spapr->rtas_size < 0) {
2691         error_report("Could not get size of LPAR rtas '%s'", filename);
2692         exit(1);
2693     }
2694     spapr->rtas_blob = g_malloc(spapr->rtas_size);
2695     if (load_image_size(filename, spapr->rtas_blob, spapr->rtas_size) < 0) {
2696         error_report("Could not load LPAR rtas '%s'", filename);
2697         exit(1);
2698     }
2699     if (spapr->rtas_size > RTAS_MAX_SIZE) {
2700         error_report("RTAS too big ! 0x%zx bytes (max is 0x%x)",
2701                      (size_t)spapr->rtas_size, RTAS_MAX_SIZE);
2702         exit(1);
2703     }
2704     g_free(filename);
2705 
2706     /* Set up RTAS event infrastructure */
2707     spapr_events_init(spapr);
2708 
2709     /* Set up the RTC RTAS interfaces */
2710     spapr_rtc_create(spapr);
2711 
2712     /* Set up VIO bus */
2713     spapr->vio_bus = spapr_vio_bus_init();
2714 
2715     for (i = 0; i < serial_max_hds(); i++) {
2716         if (serial_hd(i)) {
2717             spapr_vty_create(spapr->vio_bus, serial_hd(i));
2718         }
2719     }
2720 
2721     /* We always have at least the nvram device on VIO */
2722     spapr_create_nvram(spapr);
2723 
2724     /* Set up PCI */
2725     spapr_pci_rtas_init();
2726 
2727     phb = spapr_create_phb(spapr, 0);
2728 
2729     for (i = 0; i < nb_nics; i++) {
2730         NICInfo *nd = &nd_table[i];
2731 
2732         if (!nd->model) {
2733             nd->model = g_strdup("spapr-vlan");
2734         }
2735 
2736         if (g_str_equal(nd->model, "spapr-vlan") ||
2737             g_str_equal(nd->model, "ibmveth")) {
2738             spapr_vlan_create(spapr->vio_bus, nd);
2739         } else {
2740             pci_nic_init_nofail(&nd_table[i], phb->bus, nd->model, NULL);
2741         }
2742     }
2743 
2744     for (i = 0; i <= drive_get_max_bus(IF_SCSI); i++) {
2745         spapr_vscsi_create(spapr->vio_bus);
2746     }
2747 
2748     /* Graphics */
2749     if (spapr_vga_init(phb->bus, &error_fatal)) {
2750         spapr->has_graphics = true;
2751         machine->usb |= defaults_enabled() && !machine->usb_disabled;
2752     }
2753 
2754     if (machine->usb) {
2755         if (smc->use_ohci_by_default) {
2756             pci_create_simple(phb->bus, -1, "pci-ohci");
2757         } else {
2758             pci_create_simple(phb->bus, -1, "nec-usb-xhci");
2759         }
2760 
2761         if (spapr->has_graphics) {
2762             USBBus *usb_bus = usb_bus_find(-1);
2763 
2764             usb_create_simple(usb_bus, "usb-kbd");
2765             usb_create_simple(usb_bus, "usb-mouse");
2766         }
2767     }
2768 
2769     if (spapr->rma_size < (MIN_RMA_SLOF * MiB)) {
2770         error_report(
2771             "pSeries SLOF firmware requires >= %ldM guest RMA (Real Mode Area memory)",
2772             MIN_RMA_SLOF);
2773         exit(1);
2774     }
2775 
2776     if (kernel_filename) {
2777         uint64_t lowaddr = 0;
2778 
2779         spapr->kernel_size = load_elf(kernel_filename, translate_kernel_address,
2780                                       NULL, NULL, &lowaddr, NULL, 1,
2781                                       PPC_ELF_MACHINE, 0, 0);
2782         if (spapr->kernel_size == ELF_LOAD_WRONG_ENDIAN) {
2783             spapr->kernel_size = load_elf(kernel_filename,
2784                                           translate_kernel_address, NULL, NULL,
2785                                           &lowaddr, NULL, 0, PPC_ELF_MACHINE,
2786                                           0, 0);
2787             spapr->kernel_le = spapr->kernel_size > 0;
2788         }
2789         if (spapr->kernel_size < 0) {
2790             error_report("error loading %s: %s", kernel_filename,
2791                          load_elf_strerror(spapr->kernel_size));
2792             exit(1);
2793         }
2794 
2795         /* load initrd */
2796         if (initrd_filename) {
2797             /* Try to locate the initrd in the gap between the kernel
2798              * and the firmware. Add a bit of space just in case
2799              */
2800             spapr->initrd_base = (KERNEL_LOAD_ADDR + spapr->kernel_size
2801                                   + 0x1ffff) & ~0xffff;
2802             spapr->initrd_size = load_image_targphys(initrd_filename,
2803                                                      spapr->initrd_base,
2804                                                      load_limit
2805                                                      - spapr->initrd_base);
2806             if (spapr->initrd_size < 0) {
2807                 error_report("could not load initial ram disk '%s'",
2808                              initrd_filename);
2809                 exit(1);
2810             }
2811         }
2812     }
2813 
2814     if (bios_name == NULL) {
2815         bios_name = FW_FILE_NAME;
2816     }
2817     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
2818     if (!filename) {
2819         error_report("Could not find LPAR firmware '%s'", bios_name);
2820         exit(1);
2821     }
2822     fw_size = load_image_targphys(filename, 0, FW_MAX_SIZE);
2823     if (fw_size <= 0) {
2824         error_report("Could not load LPAR firmware '%s'", filename);
2825         exit(1);
2826     }
2827     g_free(filename);
2828 
2829     /* FIXME: Should register things through the MachineState's qdev
2830      * interface, this is a legacy from the sPAPREnvironment structure
2831      * which predated MachineState but had a similar function */
2832     vmstate_register(NULL, 0, &vmstate_spapr, spapr);
2833     register_savevm_live(NULL, "spapr/htab", -1, 1,
2834                          &savevm_htab_handlers, spapr);
2835 
2836     qemu_register_boot_set(spapr_boot_set, spapr);
2837 
2838     if (kvm_enabled()) {
2839         /* to stop and start vmclock */
2840         qemu_add_vm_change_state_handler(cpu_ppc_clock_vm_state_change,
2841                                          &spapr->tb);
2842 
2843         kvmppc_spapr_enable_inkernel_multitce();
2844     }
2845 }
2846 
2847 static int spapr_kvm_type(const char *vm_type)
2848 {
2849     if (!vm_type) {
2850         return 0;
2851     }
2852 
2853     if (!strcmp(vm_type, "HV")) {
2854         return 1;
2855     }
2856 
2857     if (!strcmp(vm_type, "PR")) {
2858         return 2;
2859     }
2860 
2861     error_report("Unknown kvm-type specified '%s'", vm_type);
2862     exit(1);
2863 }
2864 
2865 /*
2866  * Implementation of an interface to adjust firmware path
2867  * for the bootindex property handling.
2868  */
2869 static char *spapr_get_fw_dev_path(FWPathProvider *p, BusState *bus,
2870                                    DeviceState *dev)
2871 {
2872 #define CAST(type, obj, name) \
2873     ((type *)object_dynamic_cast(OBJECT(obj), (name)))
2874     SCSIDevice *d = CAST(SCSIDevice,  dev, TYPE_SCSI_DEVICE);
2875     sPAPRPHBState *phb = CAST(sPAPRPHBState, dev, TYPE_SPAPR_PCI_HOST_BRIDGE);
2876     VHostSCSICommon *vsc = CAST(VHostSCSICommon, dev, TYPE_VHOST_SCSI_COMMON);
2877 
2878     if (d) {
2879         void *spapr = CAST(void, bus->parent, "spapr-vscsi");
2880         VirtIOSCSI *virtio = CAST(VirtIOSCSI, bus->parent, TYPE_VIRTIO_SCSI);
2881         USBDevice *usb = CAST(USBDevice, bus->parent, TYPE_USB_DEVICE);
2882 
2883         if (spapr) {
2884             /*
2885              * Replace "channel@0/disk@0,0" with "disk@8000000000000000":
2886              * We use SRP luns of the form 8000 | (bus << 8) | (id << 5) | lun
2887              * in the top 16 bits of the 64-bit LUN
2888              */
2889             unsigned id = 0x8000 | (d->id << 8) | d->lun;
2890             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
2891                                    (uint64_t)id << 48);
2892         } else if (virtio) {
2893             /*
2894              * We use SRP luns of the form 01000000 | (target << 8) | lun
2895              * in the top 32 bits of the 64-bit LUN
2896              * Note: the quote above is from SLOF and it is wrong,
2897              * the actual binding is:
2898              * swap 0100 or 10 << or 20 << ( target lun-id -- srplun )
2899              */
2900             unsigned id = 0x1000000 | (d->id << 16) | d->lun;
2901             if (d->lun >= 256) {
2902                 /* Use the LUN "flat space addressing method" */
2903                 id |= 0x4000;
2904             }
2905             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
2906                                    (uint64_t)id << 32);
2907         } else if (usb) {
2908             /*
2909              * We use SRP luns of the form 01000000 | (usb-port << 16) | lun
2910              * in the top 32 bits of the 64-bit LUN
2911              */
2912             unsigned usb_port = atoi(usb->port->path);
2913             unsigned id = 0x1000000 | (usb_port << 16) | d->lun;
2914             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
2915                                    (uint64_t)id << 32);
2916         }
2917     }
2918 
2919     /*
2920      * SLOF probes the USB devices, and if it recognizes that the device is a
2921      * storage device, it changes its name to "storage" instead of "usb-host",
2922      * and additionally adds a child node for the SCSI LUN, so the correct
2923      * boot path in SLOF is something like .../storage@1/disk@xxx" instead.
2924      */
2925     if (strcmp("usb-host", qdev_fw_name(dev)) == 0) {
2926         USBDevice *usbdev = CAST(USBDevice, dev, TYPE_USB_DEVICE);
2927         if (usb_host_dev_is_scsi_storage(usbdev)) {
2928             return g_strdup_printf("storage@%s/disk", usbdev->port->path);
2929         }
2930     }
2931 
2932     if (phb) {
2933         /* Replace "pci" with "pci@800000020000000" */
2934         return g_strdup_printf("pci@%"PRIX64, phb->buid);
2935     }
2936 
2937     if (vsc) {
2938         /* Same logic as virtio above */
2939         unsigned id = 0x1000000 | (vsc->target << 16) | vsc->lun;
2940         return g_strdup_printf("disk@%"PRIX64, (uint64_t)id << 32);
2941     }
2942 
2943     if (g_str_equal("pci-bridge", qdev_fw_name(dev))) {
2944         /* SLOF uses "pci" instead of "pci-bridge" for PCI bridges */
2945         PCIDevice *pcidev = CAST(PCIDevice, dev, TYPE_PCI_DEVICE);
2946         return g_strdup_printf("pci@%x", PCI_SLOT(pcidev->devfn));
2947     }
2948 
2949     return NULL;
2950 }
2951 
2952 static char *spapr_get_kvm_type(Object *obj, Error **errp)
2953 {
2954     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2955 
2956     return g_strdup(spapr->kvm_type);
2957 }
2958 
2959 static void spapr_set_kvm_type(Object *obj, const char *value, Error **errp)
2960 {
2961     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2962 
2963     g_free(spapr->kvm_type);
2964     spapr->kvm_type = g_strdup(value);
2965 }
2966 
2967 static bool spapr_get_modern_hotplug_events(Object *obj, Error **errp)
2968 {
2969     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2970 
2971     return spapr->use_hotplug_event_source;
2972 }
2973 
2974 static void spapr_set_modern_hotplug_events(Object *obj, bool value,
2975                                             Error **errp)
2976 {
2977     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2978 
2979     spapr->use_hotplug_event_source = value;
2980 }
2981 
2982 static bool spapr_get_msix_emulation(Object *obj, Error **errp)
2983 {
2984     return true;
2985 }
2986 
2987 static char *spapr_get_resize_hpt(Object *obj, Error **errp)
2988 {
2989     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2990 
2991     switch (spapr->resize_hpt) {
2992     case SPAPR_RESIZE_HPT_DEFAULT:
2993         return g_strdup("default");
2994     case SPAPR_RESIZE_HPT_DISABLED:
2995         return g_strdup("disabled");
2996     case SPAPR_RESIZE_HPT_ENABLED:
2997         return g_strdup("enabled");
2998     case SPAPR_RESIZE_HPT_REQUIRED:
2999         return g_strdup("required");
3000     }
3001     g_assert_not_reached();
3002 }
3003 
3004 static void spapr_set_resize_hpt(Object *obj, const char *value, Error **errp)
3005 {
3006     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
3007 
3008     if (strcmp(value, "default") == 0) {
3009         spapr->resize_hpt = SPAPR_RESIZE_HPT_DEFAULT;
3010     } else if (strcmp(value, "disabled") == 0) {
3011         spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
3012     } else if (strcmp(value, "enabled") == 0) {
3013         spapr->resize_hpt = SPAPR_RESIZE_HPT_ENABLED;
3014     } else if (strcmp(value, "required") == 0) {
3015         spapr->resize_hpt = SPAPR_RESIZE_HPT_REQUIRED;
3016     } else {
3017         error_setg(errp, "Bad value for \"resize-hpt\" property");
3018     }
3019 }
3020 
3021 static void spapr_get_vsmt(Object *obj, Visitor *v, const char *name,
3022                                    void *opaque, Error **errp)
3023 {
3024     visit_type_uint32(v, name, (uint32_t *)opaque, errp);
3025 }
3026 
3027 static void spapr_set_vsmt(Object *obj, Visitor *v, const char *name,
3028                                    void *opaque, Error **errp)
3029 {
3030     visit_type_uint32(v, name, (uint32_t *)opaque, errp);
3031 }
3032 
3033 static void spapr_instance_init(Object *obj)
3034 {
3035     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
3036 
3037     spapr->htab_fd = -1;
3038     spapr->use_hotplug_event_source = true;
3039     object_property_add_str(obj, "kvm-type",
3040                             spapr_get_kvm_type, spapr_set_kvm_type, NULL);
3041     object_property_set_description(obj, "kvm-type",
3042                                     "Specifies the KVM virtualization mode (HV, PR)",
3043                                     NULL);
3044     object_property_add_bool(obj, "modern-hotplug-events",
3045                             spapr_get_modern_hotplug_events,
3046                             spapr_set_modern_hotplug_events,
3047                             NULL);
3048     object_property_set_description(obj, "modern-hotplug-events",
3049                                     "Use dedicated hotplug event mechanism in"
3050                                     " place of standard EPOW events when possible"
3051                                     " (required for memory hot-unplug support)",
3052                                     NULL);
3053     ppc_compat_add_property(obj, "max-cpu-compat", &spapr->max_compat_pvr,
3054                             "Maximum permitted CPU compatibility mode",
3055                             &error_fatal);
3056 
3057     object_property_add_str(obj, "resize-hpt",
3058                             spapr_get_resize_hpt, spapr_set_resize_hpt, NULL);
3059     object_property_set_description(obj, "resize-hpt",
3060                                     "Resizing of the Hash Page Table (enabled, disabled, required)",
3061                                     NULL);
3062     object_property_add(obj, "vsmt", "uint32", spapr_get_vsmt,
3063                         spapr_set_vsmt, NULL, &spapr->vsmt, &error_abort);
3064     object_property_set_description(obj, "vsmt",
3065                                     "Virtual SMT: KVM behaves as if this were"
3066                                     " the host's SMT mode", &error_abort);
3067     object_property_add_bool(obj, "vfio-no-msix-emulation",
3068                              spapr_get_msix_emulation, NULL, NULL);
3069 }
3070 
3071 static void spapr_machine_finalizefn(Object *obj)
3072 {
3073     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
3074 
3075     g_free(spapr->kvm_type);
3076 }
3077 
3078 void spapr_do_system_reset_on_cpu(CPUState *cs, run_on_cpu_data arg)
3079 {
3080     cpu_synchronize_state(cs);
3081     ppc_cpu_do_system_reset(cs);
3082 }
3083 
3084 static void spapr_nmi(NMIState *n, int cpu_index, Error **errp)
3085 {
3086     CPUState *cs;
3087 
3088     CPU_FOREACH(cs) {
3089         async_run_on_cpu(cs, spapr_do_system_reset_on_cpu, RUN_ON_CPU_NULL);
3090     }
3091 }
3092 
3093 static void spapr_add_lmbs(DeviceState *dev, uint64_t addr_start, uint64_t size,
3094                            uint32_t node, bool dedicated_hp_event_source,
3095                            Error **errp)
3096 {
3097     sPAPRDRConnector *drc;
3098     uint32_t nr_lmbs = size/SPAPR_MEMORY_BLOCK_SIZE;
3099     int i, fdt_offset, fdt_size;
3100     void *fdt;
3101     uint64_t addr = addr_start;
3102     bool hotplugged = spapr_drc_hotplugged(dev);
3103     Error *local_err = NULL;
3104 
3105     for (i = 0; i < nr_lmbs; i++) {
3106         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3107                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3108         g_assert(drc);
3109 
3110         fdt = create_device_tree(&fdt_size);
3111         fdt_offset = spapr_populate_memory_node(fdt, node, addr,
3112                                                 SPAPR_MEMORY_BLOCK_SIZE);
3113 
3114         spapr_drc_attach(drc, dev, fdt, fdt_offset, &local_err);
3115         if (local_err) {
3116             while (addr > addr_start) {
3117                 addr -= SPAPR_MEMORY_BLOCK_SIZE;
3118                 drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3119                                       addr / SPAPR_MEMORY_BLOCK_SIZE);
3120                 spapr_drc_detach(drc);
3121             }
3122             g_free(fdt);
3123             error_propagate(errp, local_err);
3124             return;
3125         }
3126         if (!hotplugged) {
3127             spapr_drc_reset(drc);
3128         }
3129         addr += SPAPR_MEMORY_BLOCK_SIZE;
3130     }
3131     /* send hotplug notification to the
3132      * guest only in case of hotplugged memory
3133      */
3134     if (hotplugged) {
3135         if (dedicated_hp_event_source) {
3136             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3137                                   addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3138             spapr_hotplug_req_add_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3139                                                    nr_lmbs,
3140                                                    spapr_drc_index(drc));
3141         } else {
3142             spapr_hotplug_req_add_by_count(SPAPR_DR_CONNECTOR_TYPE_LMB,
3143                                            nr_lmbs);
3144         }
3145     }
3146 }
3147 
3148 static void spapr_memory_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3149                               Error **errp)
3150 {
3151     Error *local_err = NULL;
3152     sPAPRMachineState *ms = SPAPR_MACHINE(hotplug_dev);
3153     PCDIMMDevice *dimm = PC_DIMM(dev);
3154     uint64_t size, addr;
3155     uint32_t node;
3156 
3157     size = memory_device_get_region_size(MEMORY_DEVICE(dev), &error_abort);
3158 
3159     pc_dimm_plug(dimm, MACHINE(ms), &local_err);
3160     if (local_err) {
3161         goto out;
3162     }
3163 
3164     addr = object_property_get_uint(OBJECT(dimm),
3165                                     PC_DIMM_ADDR_PROP, &local_err);
3166     if (local_err) {
3167         goto out_unplug;
3168     }
3169 
3170     node = object_property_get_uint(OBJECT(dev), PC_DIMM_NODE_PROP,
3171                                     &error_abort);
3172     spapr_add_lmbs(dev, addr, size, node,
3173                    spapr_ovec_test(ms->ov5_cas, OV5_HP_EVT),
3174                    &local_err);
3175     if (local_err) {
3176         goto out_unplug;
3177     }
3178 
3179     return;
3180 
3181 out_unplug:
3182     pc_dimm_unplug(dimm, MACHINE(ms));
3183 out:
3184     error_propagate(errp, local_err);
3185 }
3186 
3187 static void spapr_memory_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3188                                   Error **errp)
3189 {
3190     const sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(hotplug_dev);
3191     sPAPRMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3192     PCDIMMDevice *dimm = PC_DIMM(dev);
3193     Error *local_err = NULL;
3194     uint64_t size;
3195     Object *memdev;
3196     hwaddr pagesize;
3197 
3198     if (!smc->dr_lmb_enabled) {
3199         error_setg(errp, "Memory hotplug not supported for this machine");
3200         return;
3201     }
3202 
3203     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &local_err);
3204     if (local_err) {
3205         error_propagate(errp, local_err);
3206         return;
3207     }
3208 
3209     if (size % SPAPR_MEMORY_BLOCK_SIZE) {
3210         error_setg(errp, "Hotplugged memory size must be a multiple of "
3211                       "%" PRIu64 " MB", SPAPR_MEMORY_BLOCK_SIZE / MiB);
3212         return;
3213     }
3214 
3215     memdev = object_property_get_link(OBJECT(dimm), PC_DIMM_MEMDEV_PROP,
3216                                       &error_abort);
3217     pagesize = host_memory_backend_pagesize(MEMORY_BACKEND(memdev));
3218     spapr_check_pagesize(spapr, pagesize, &local_err);
3219     if (local_err) {
3220         error_propagate(errp, local_err);
3221         return;
3222     }
3223 
3224     pc_dimm_pre_plug(dimm, MACHINE(hotplug_dev), NULL, errp);
3225 }
3226 
3227 struct sPAPRDIMMState {
3228     PCDIMMDevice *dimm;
3229     uint32_t nr_lmbs;
3230     QTAILQ_ENTRY(sPAPRDIMMState) next;
3231 };
3232 
3233 static sPAPRDIMMState *spapr_pending_dimm_unplugs_find(sPAPRMachineState *s,
3234                                                        PCDIMMDevice *dimm)
3235 {
3236     sPAPRDIMMState *dimm_state = NULL;
3237 
3238     QTAILQ_FOREACH(dimm_state, &s->pending_dimm_unplugs, next) {
3239         if (dimm_state->dimm == dimm) {
3240             break;
3241         }
3242     }
3243     return dimm_state;
3244 }
3245 
3246 static sPAPRDIMMState *spapr_pending_dimm_unplugs_add(sPAPRMachineState *spapr,
3247                                                       uint32_t nr_lmbs,
3248                                                       PCDIMMDevice *dimm)
3249 {
3250     sPAPRDIMMState *ds = NULL;
3251 
3252     /*
3253      * If this request is for a DIMM whose removal had failed earlier
3254      * (due to guest's refusal to remove the LMBs), we would have this
3255      * dimm already in the pending_dimm_unplugs list. In that
3256      * case don't add again.
3257      */
3258     ds = spapr_pending_dimm_unplugs_find(spapr, dimm);
3259     if (!ds) {
3260         ds = g_malloc0(sizeof(sPAPRDIMMState));
3261         ds->nr_lmbs = nr_lmbs;
3262         ds->dimm = dimm;
3263         QTAILQ_INSERT_HEAD(&spapr->pending_dimm_unplugs, ds, next);
3264     }
3265     return ds;
3266 }
3267 
3268 static void spapr_pending_dimm_unplugs_remove(sPAPRMachineState *spapr,
3269                                               sPAPRDIMMState *dimm_state)
3270 {
3271     QTAILQ_REMOVE(&spapr->pending_dimm_unplugs, dimm_state, next);
3272     g_free(dimm_state);
3273 }
3274 
3275 static sPAPRDIMMState *spapr_recover_pending_dimm_state(sPAPRMachineState *ms,
3276                                                         PCDIMMDevice *dimm)
3277 {
3278     sPAPRDRConnector *drc;
3279     uint64_t size = memory_device_get_region_size(MEMORY_DEVICE(dimm),
3280                                                   &error_abort);
3281     uint32_t nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3282     uint32_t avail_lmbs = 0;
3283     uint64_t addr_start, addr;
3284     int i;
3285 
3286     addr_start = object_property_get_int(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3287                                          &error_abort);
3288 
3289     addr = addr_start;
3290     for (i = 0; i < nr_lmbs; i++) {
3291         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3292                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3293         g_assert(drc);
3294         if (drc->dev) {
3295             avail_lmbs++;
3296         }
3297         addr += SPAPR_MEMORY_BLOCK_SIZE;
3298     }
3299 
3300     return spapr_pending_dimm_unplugs_add(ms, avail_lmbs, dimm);
3301 }
3302 
3303 /* Callback to be called during DRC release. */
3304 void spapr_lmb_release(DeviceState *dev)
3305 {
3306     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3307     sPAPRMachineState *spapr = SPAPR_MACHINE(hotplug_ctrl);
3308     sPAPRDIMMState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3309 
3310     /* This information will get lost if a migration occurs
3311      * during the unplug process. In this case recover it. */
3312     if (ds == NULL) {
3313         ds = spapr_recover_pending_dimm_state(spapr, PC_DIMM(dev));
3314         g_assert(ds);
3315         /* The DRC being examined by the caller at least must be counted */
3316         g_assert(ds->nr_lmbs);
3317     }
3318 
3319     if (--ds->nr_lmbs) {
3320         return;
3321     }
3322 
3323     /*
3324      * Now that all the LMBs have been removed by the guest, call the
3325      * unplug handler chain. This can never fail.
3326      */
3327     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3328 }
3329 
3330 static void spapr_memory_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3331 {
3332     sPAPRMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3333     sPAPRDIMMState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3334 
3335     pc_dimm_unplug(PC_DIMM(dev), MACHINE(hotplug_dev));
3336     object_unparent(OBJECT(dev));
3337     spapr_pending_dimm_unplugs_remove(spapr, ds);
3338 }
3339 
3340 static void spapr_memory_unplug_request(HotplugHandler *hotplug_dev,
3341                                         DeviceState *dev, Error **errp)
3342 {
3343     sPAPRMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3344     Error *local_err = NULL;
3345     PCDIMMDevice *dimm = PC_DIMM(dev);
3346     uint32_t nr_lmbs;
3347     uint64_t size, addr_start, addr;
3348     int i;
3349     sPAPRDRConnector *drc;
3350 
3351     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &error_abort);
3352     nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3353 
3354     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3355                                          &local_err);
3356     if (local_err) {
3357         goto out;
3358     }
3359 
3360     /*
3361      * An existing pending dimm state for this DIMM means that there is an
3362      * unplug operation in progress, waiting for the spapr_lmb_release
3363      * callback to complete the job (BQL can't cover that far). In this case,
3364      * bail out to avoid detaching DRCs that were already released.
3365      */
3366     if (spapr_pending_dimm_unplugs_find(spapr, dimm)) {
3367         error_setg(&local_err,
3368                    "Memory unplug already in progress for device %s",
3369                    dev->id);
3370         goto out;
3371     }
3372 
3373     spapr_pending_dimm_unplugs_add(spapr, nr_lmbs, dimm);
3374 
3375     addr = addr_start;
3376     for (i = 0; i < nr_lmbs; i++) {
3377         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3378                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3379         g_assert(drc);
3380 
3381         spapr_drc_detach(drc);
3382         addr += SPAPR_MEMORY_BLOCK_SIZE;
3383     }
3384 
3385     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3386                           addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3387     spapr_hotplug_req_remove_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3388                                               nr_lmbs, spapr_drc_index(drc));
3389 out:
3390     error_propagate(errp, local_err);
3391 }
3392 
3393 static void *spapr_populate_hotplug_cpu_dt(CPUState *cs, int *fdt_offset,
3394                                            sPAPRMachineState *spapr)
3395 {
3396     PowerPCCPU *cpu = POWERPC_CPU(cs);
3397     DeviceClass *dc = DEVICE_GET_CLASS(cs);
3398     int id = spapr_get_vcpu_id(cpu);
3399     void *fdt;
3400     int offset, fdt_size;
3401     char *nodename;
3402 
3403     fdt = create_device_tree(&fdt_size);
3404     nodename = g_strdup_printf("%s@%x", dc->fw_name, id);
3405     offset = fdt_add_subnode(fdt, 0, nodename);
3406 
3407     spapr_populate_cpu_dt(cs, fdt, offset, spapr);
3408     g_free(nodename);
3409 
3410     *fdt_offset = offset;
3411     return fdt;
3412 }
3413 
3414 /* Callback to be called during DRC release. */
3415 void spapr_core_release(DeviceState *dev)
3416 {
3417     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3418 
3419     /* Call the unplug handler chain. This can never fail. */
3420     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3421 }
3422 
3423 static void spapr_core_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3424 {
3425     MachineState *ms = MACHINE(hotplug_dev);
3426     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(ms);
3427     CPUCore *cc = CPU_CORE(dev);
3428     CPUArchId *core_slot = spapr_find_cpu_slot(ms, cc->core_id, NULL);
3429 
3430     if (smc->pre_2_10_has_unused_icps) {
3431         sPAPRCPUCore *sc = SPAPR_CPU_CORE(OBJECT(dev));
3432         int i;
3433 
3434         for (i = 0; i < cc->nr_threads; i++) {
3435             CPUState *cs = CPU(sc->threads[i]);
3436 
3437             pre_2_10_vmstate_register_dummy_icp(cs->cpu_index);
3438         }
3439     }
3440 
3441     assert(core_slot);
3442     core_slot->cpu = NULL;
3443     object_unparent(OBJECT(dev));
3444 }
3445 
3446 static
3447 void spapr_core_unplug_request(HotplugHandler *hotplug_dev, DeviceState *dev,
3448                                Error **errp)
3449 {
3450     sPAPRMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3451     int index;
3452     sPAPRDRConnector *drc;
3453     CPUCore *cc = CPU_CORE(dev);
3454 
3455     if (!spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index)) {
3456         error_setg(errp, "Unable to find CPU core with core-id: %d",
3457                    cc->core_id);
3458         return;
3459     }
3460     if (index == 0) {
3461         error_setg(errp, "Boot CPU core may not be unplugged");
3462         return;
3463     }
3464 
3465     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3466                           spapr_vcpu_id(spapr, cc->core_id));
3467     g_assert(drc);
3468 
3469     spapr_drc_detach(drc);
3470 
3471     spapr_hotplug_req_remove_by_index(drc);
3472 }
3473 
3474 static void spapr_core_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3475                             Error **errp)
3476 {
3477     sPAPRMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3478     MachineClass *mc = MACHINE_GET_CLASS(spapr);
3479     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3480     sPAPRCPUCore *core = SPAPR_CPU_CORE(OBJECT(dev));
3481     CPUCore *cc = CPU_CORE(dev);
3482     CPUState *cs = CPU(core->threads[0]);
3483     sPAPRDRConnector *drc;
3484     Error *local_err = NULL;
3485     CPUArchId *core_slot;
3486     int index;
3487     bool hotplugged = spapr_drc_hotplugged(dev);
3488 
3489     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3490     if (!core_slot) {
3491         error_setg(errp, "Unable to find CPU core with core-id: %d",
3492                    cc->core_id);
3493         return;
3494     }
3495     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3496                           spapr_vcpu_id(spapr, cc->core_id));
3497 
3498     g_assert(drc || !mc->has_hotpluggable_cpus);
3499 
3500     if (drc) {
3501         void *fdt;
3502         int fdt_offset;
3503 
3504         fdt = spapr_populate_hotplug_cpu_dt(cs, &fdt_offset, spapr);
3505 
3506         spapr_drc_attach(drc, dev, fdt, fdt_offset, &local_err);
3507         if (local_err) {
3508             g_free(fdt);
3509             error_propagate(errp, local_err);
3510             return;
3511         }
3512 
3513         if (hotplugged) {
3514             /*
3515              * Send hotplug notification interrupt to the guest only
3516              * in case of hotplugged CPUs.
3517              */
3518             spapr_hotplug_req_add_by_index(drc);
3519         } else {
3520             spapr_drc_reset(drc);
3521         }
3522     }
3523 
3524     core_slot->cpu = OBJECT(dev);
3525 
3526     if (smc->pre_2_10_has_unused_icps) {
3527         int i;
3528 
3529         for (i = 0; i < cc->nr_threads; i++) {
3530             cs = CPU(core->threads[i]);
3531             pre_2_10_vmstate_unregister_dummy_icp(cs->cpu_index);
3532         }
3533     }
3534 }
3535 
3536 static void spapr_core_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3537                                 Error **errp)
3538 {
3539     MachineState *machine = MACHINE(OBJECT(hotplug_dev));
3540     MachineClass *mc = MACHINE_GET_CLASS(hotplug_dev);
3541     Error *local_err = NULL;
3542     CPUCore *cc = CPU_CORE(dev);
3543     const char *base_core_type = spapr_get_cpu_core_type(machine->cpu_type);
3544     const char *type = object_get_typename(OBJECT(dev));
3545     CPUArchId *core_slot;
3546     int index;
3547 
3548     if (dev->hotplugged && !mc->has_hotpluggable_cpus) {
3549         error_setg(&local_err, "CPU hotplug not supported for this machine");
3550         goto out;
3551     }
3552 
3553     if (strcmp(base_core_type, type)) {
3554         error_setg(&local_err, "CPU core type should be %s", base_core_type);
3555         goto out;
3556     }
3557 
3558     if (cc->core_id % smp_threads) {
3559         error_setg(&local_err, "invalid core id %d", cc->core_id);
3560         goto out;
3561     }
3562 
3563     /*
3564      * In general we should have homogeneous threads-per-core, but old
3565      * (pre hotplug support) machine types allow the last core to have
3566      * reduced threads as a compatibility hack for when we allowed
3567      * total vcpus not a multiple of threads-per-core.
3568      */
3569     if (mc->has_hotpluggable_cpus && (cc->nr_threads != smp_threads)) {
3570         error_setg(&local_err, "invalid nr-threads %d, must be %d",
3571                    cc->nr_threads, smp_threads);
3572         goto out;
3573     }
3574 
3575     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3576     if (!core_slot) {
3577         error_setg(&local_err, "core id %d out of range", cc->core_id);
3578         goto out;
3579     }
3580 
3581     if (core_slot->cpu) {
3582         error_setg(&local_err, "core %d already populated", cc->core_id);
3583         goto out;
3584     }
3585 
3586     numa_cpu_pre_plug(core_slot, dev, &local_err);
3587 
3588 out:
3589     error_propagate(errp, local_err);
3590 }
3591 
3592 static void spapr_machine_device_plug(HotplugHandler *hotplug_dev,
3593                                       DeviceState *dev, Error **errp)
3594 {
3595     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
3596         spapr_memory_plug(hotplug_dev, dev, errp);
3597     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3598         spapr_core_plug(hotplug_dev, dev, errp);
3599     }
3600 }
3601 
3602 static void spapr_machine_device_unplug(HotplugHandler *hotplug_dev,
3603                                         DeviceState *dev, Error **errp)
3604 {
3605     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
3606         spapr_memory_unplug(hotplug_dev, dev);
3607     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3608         spapr_core_unplug(hotplug_dev, dev);
3609     }
3610 }
3611 
3612 static void spapr_machine_device_unplug_request(HotplugHandler *hotplug_dev,
3613                                                 DeviceState *dev, Error **errp)
3614 {
3615     sPAPRMachineState *sms = SPAPR_MACHINE(OBJECT(hotplug_dev));
3616     MachineClass *mc = MACHINE_GET_CLASS(sms);
3617 
3618     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
3619         if (spapr_ovec_test(sms->ov5_cas, OV5_HP_EVT)) {
3620             spapr_memory_unplug_request(hotplug_dev, dev, errp);
3621         } else {
3622             /* NOTE: this means there is a window after guest reset, prior to
3623              * CAS negotiation, where unplug requests will fail due to the
3624              * capability not being detected yet. This is a bit different than
3625              * the case with PCI unplug, where the events will be queued and
3626              * eventually handled by the guest after boot
3627              */
3628             error_setg(errp, "Memory hot unplug not supported for this guest");
3629         }
3630     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3631         if (!mc->has_hotpluggable_cpus) {
3632             error_setg(errp, "CPU hot unplug not supported on this machine");
3633             return;
3634         }
3635         spapr_core_unplug_request(hotplug_dev, dev, errp);
3636     }
3637 }
3638 
3639 static void spapr_machine_device_pre_plug(HotplugHandler *hotplug_dev,
3640                                           DeviceState *dev, Error **errp)
3641 {
3642     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
3643         spapr_memory_pre_plug(hotplug_dev, dev, errp);
3644     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3645         spapr_core_pre_plug(hotplug_dev, dev, errp);
3646     }
3647 }
3648 
3649 static HotplugHandler *spapr_get_hotplug_handler(MachineState *machine,
3650                                                  DeviceState *dev)
3651 {
3652     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM) ||
3653         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3654         return HOTPLUG_HANDLER(machine);
3655     }
3656     return NULL;
3657 }
3658 
3659 static CpuInstanceProperties
3660 spapr_cpu_index_to_props(MachineState *machine, unsigned cpu_index)
3661 {
3662     CPUArchId *core_slot;
3663     MachineClass *mc = MACHINE_GET_CLASS(machine);
3664 
3665     /* make sure possible_cpu are intialized */
3666     mc->possible_cpu_arch_ids(machine);
3667     /* get CPU core slot containing thread that matches cpu_index */
3668     core_slot = spapr_find_cpu_slot(machine, cpu_index, NULL);
3669     assert(core_slot);
3670     return core_slot->props;
3671 }
3672 
3673 static int64_t spapr_get_default_cpu_node_id(const MachineState *ms, int idx)
3674 {
3675     return idx / smp_cores % nb_numa_nodes;
3676 }
3677 
3678 static const CPUArchIdList *spapr_possible_cpu_arch_ids(MachineState *machine)
3679 {
3680     int i;
3681     const char *core_type;
3682     int spapr_max_cores = max_cpus / smp_threads;
3683     MachineClass *mc = MACHINE_GET_CLASS(machine);
3684 
3685     if (!mc->has_hotpluggable_cpus) {
3686         spapr_max_cores = QEMU_ALIGN_UP(smp_cpus, smp_threads) / smp_threads;
3687     }
3688     if (machine->possible_cpus) {
3689         assert(machine->possible_cpus->len == spapr_max_cores);
3690         return machine->possible_cpus;
3691     }
3692 
3693     core_type = spapr_get_cpu_core_type(machine->cpu_type);
3694     if (!core_type) {
3695         error_report("Unable to find sPAPR CPU Core definition");
3696         exit(1);
3697     }
3698 
3699     machine->possible_cpus = g_malloc0(sizeof(CPUArchIdList) +
3700                              sizeof(CPUArchId) * spapr_max_cores);
3701     machine->possible_cpus->len = spapr_max_cores;
3702     for (i = 0; i < machine->possible_cpus->len; i++) {
3703         int core_id = i * smp_threads;
3704 
3705         machine->possible_cpus->cpus[i].type = core_type;
3706         machine->possible_cpus->cpus[i].vcpus_count = smp_threads;
3707         machine->possible_cpus->cpus[i].arch_id = core_id;
3708         machine->possible_cpus->cpus[i].props.has_core_id = true;
3709         machine->possible_cpus->cpus[i].props.core_id = core_id;
3710     }
3711     return machine->possible_cpus;
3712 }
3713 
3714 static void spapr_phb_placement(sPAPRMachineState *spapr, uint32_t index,
3715                                 uint64_t *buid, hwaddr *pio,
3716                                 hwaddr *mmio32, hwaddr *mmio64,
3717                                 unsigned n_dma, uint32_t *liobns, Error **errp)
3718 {
3719     /*
3720      * New-style PHB window placement.
3721      *
3722      * Goals: Gives large (1TiB), naturally aligned 64-bit MMIO window
3723      * for each PHB, in addition to 2GiB 32-bit MMIO and 64kiB PIO
3724      * windows.
3725      *
3726      * Some guest kernels can't work with MMIO windows above 1<<46
3727      * (64TiB), so we place up to 31 PHBs in the area 32TiB..64TiB
3728      *
3729      * 32TiB..(33TiB+1984kiB) contains the 64kiB PIO windows for each
3730      * PHB stacked together.  (32TiB+2GiB)..(32TiB+64GiB) contains the
3731      * 2GiB 32-bit MMIO windows for each PHB.  Then 33..64TiB has the
3732      * 1TiB 64-bit MMIO windows for each PHB.
3733      */
3734     const uint64_t base_buid = 0x800000020000000ULL;
3735 #define SPAPR_MAX_PHBS ((SPAPR_PCI_LIMIT - SPAPR_PCI_BASE) / \
3736                         SPAPR_PCI_MEM64_WIN_SIZE - 1)
3737     int i;
3738 
3739     /* Sanity check natural alignments */
3740     QEMU_BUILD_BUG_ON((SPAPR_PCI_BASE % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
3741     QEMU_BUILD_BUG_ON((SPAPR_PCI_LIMIT % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
3742     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM64_WIN_SIZE % SPAPR_PCI_MEM32_WIN_SIZE) != 0);
3743     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM32_WIN_SIZE % SPAPR_PCI_IO_WIN_SIZE) != 0);
3744     /* Sanity check bounds */
3745     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_IO_WIN_SIZE) >
3746                       SPAPR_PCI_MEM32_WIN_SIZE);
3747     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_MEM32_WIN_SIZE) >
3748                       SPAPR_PCI_MEM64_WIN_SIZE);
3749 
3750     if (index >= SPAPR_MAX_PHBS) {
3751         error_setg(errp, "\"index\" for PAPR PHB is too large (max %llu)",
3752                    SPAPR_MAX_PHBS - 1);
3753         return;
3754     }
3755 
3756     *buid = base_buid + index;
3757     for (i = 0; i < n_dma; ++i) {
3758         liobns[i] = SPAPR_PCI_LIOBN(index, i);
3759     }
3760 
3761     *pio = SPAPR_PCI_BASE + index * SPAPR_PCI_IO_WIN_SIZE;
3762     *mmio32 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM32_WIN_SIZE;
3763     *mmio64 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM64_WIN_SIZE;
3764 }
3765 
3766 static ICSState *spapr_ics_get(XICSFabric *dev, int irq)
3767 {
3768     sPAPRMachineState *spapr = SPAPR_MACHINE(dev);
3769 
3770     return ics_valid_irq(spapr->ics, irq) ? spapr->ics : NULL;
3771 }
3772 
3773 static void spapr_ics_resend(XICSFabric *dev)
3774 {
3775     sPAPRMachineState *spapr = SPAPR_MACHINE(dev);
3776 
3777     ics_resend(spapr->ics);
3778 }
3779 
3780 static ICPState *spapr_icp_get(XICSFabric *xi, int vcpu_id)
3781 {
3782     PowerPCCPU *cpu = spapr_find_cpu(vcpu_id);
3783 
3784     return cpu ? ICP(cpu->intc) : NULL;
3785 }
3786 
3787 static void spapr_pic_print_info(InterruptStatsProvider *obj,
3788                                  Monitor *mon)
3789 {
3790     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
3791     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
3792 
3793     smc->irq->print_info(spapr, mon);
3794 }
3795 
3796 int spapr_get_vcpu_id(PowerPCCPU *cpu)
3797 {
3798     return cpu->vcpu_id;
3799 }
3800 
3801 void spapr_set_vcpu_id(PowerPCCPU *cpu, int cpu_index, Error **errp)
3802 {
3803     sPAPRMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
3804     int vcpu_id;
3805 
3806     vcpu_id = spapr_vcpu_id(spapr, cpu_index);
3807 
3808     if (kvm_enabled() && !kvm_vcpu_id_is_valid(vcpu_id)) {
3809         error_setg(errp, "Can't create CPU with id %d in KVM", vcpu_id);
3810         error_append_hint(errp, "Adjust the number of cpus to %d "
3811                           "or try to raise the number of threads per core\n",
3812                           vcpu_id * smp_threads / spapr->vsmt);
3813         return;
3814     }
3815 
3816     cpu->vcpu_id = vcpu_id;
3817 }
3818 
3819 PowerPCCPU *spapr_find_cpu(int vcpu_id)
3820 {
3821     CPUState *cs;
3822 
3823     CPU_FOREACH(cs) {
3824         PowerPCCPU *cpu = POWERPC_CPU(cs);
3825 
3826         if (spapr_get_vcpu_id(cpu) == vcpu_id) {
3827             return cpu;
3828         }
3829     }
3830 
3831     return NULL;
3832 }
3833 
3834 static void spapr_machine_class_init(ObjectClass *oc, void *data)
3835 {
3836     MachineClass *mc = MACHINE_CLASS(oc);
3837     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(oc);
3838     FWPathProviderClass *fwc = FW_PATH_PROVIDER_CLASS(oc);
3839     NMIClass *nc = NMI_CLASS(oc);
3840     HotplugHandlerClass *hc = HOTPLUG_HANDLER_CLASS(oc);
3841     PPCVirtualHypervisorClass *vhc = PPC_VIRTUAL_HYPERVISOR_CLASS(oc);
3842     XICSFabricClass *xic = XICS_FABRIC_CLASS(oc);
3843     InterruptStatsProviderClass *ispc = INTERRUPT_STATS_PROVIDER_CLASS(oc);
3844 
3845     mc->desc = "pSeries Logical Partition (PAPR compliant)";
3846     mc->ignore_boot_device_suffixes = true;
3847 
3848     /*
3849      * We set up the default / latest behaviour here.  The class_init
3850      * functions for the specific versioned machine types can override
3851      * these details for backwards compatibility
3852      */
3853     mc->init = spapr_machine_init;
3854     mc->reset = spapr_machine_reset;
3855     mc->block_default_type = IF_SCSI;
3856     mc->max_cpus = 1024;
3857     mc->no_parallel = 1;
3858     mc->default_boot_order = "";
3859     mc->default_ram_size = 512 * MiB;
3860     mc->default_display = "std";
3861     mc->kvm_type = spapr_kvm_type;
3862     machine_class_allow_dynamic_sysbus_dev(mc, TYPE_SPAPR_PCI_HOST_BRIDGE);
3863     mc->pci_allow_0_address = true;
3864     assert(!mc->get_hotplug_handler);
3865     mc->get_hotplug_handler = spapr_get_hotplug_handler;
3866     hc->pre_plug = spapr_machine_device_pre_plug;
3867     hc->plug = spapr_machine_device_plug;
3868     mc->cpu_index_to_instance_props = spapr_cpu_index_to_props;
3869     mc->get_default_cpu_node_id = spapr_get_default_cpu_node_id;
3870     mc->possible_cpu_arch_ids = spapr_possible_cpu_arch_ids;
3871     hc->unplug_request = spapr_machine_device_unplug_request;
3872     hc->unplug = spapr_machine_device_unplug;
3873 
3874     smc->dr_lmb_enabled = true;
3875     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power8_v2.0");
3876     mc->has_hotpluggable_cpus = true;
3877     smc->resize_hpt_default = SPAPR_RESIZE_HPT_ENABLED;
3878     fwc->get_dev_path = spapr_get_fw_dev_path;
3879     nc->nmi_monitor_handler = spapr_nmi;
3880     smc->phb_placement = spapr_phb_placement;
3881     vhc->hypercall = emulate_spapr_hypercall;
3882     vhc->hpt_mask = spapr_hpt_mask;
3883     vhc->map_hptes = spapr_map_hptes;
3884     vhc->unmap_hptes = spapr_unmap_hptes;
3885     vhc->store_hpte = spapr_store_hpte;
3886     vhc->get_patbe = spapr_get_patbe;
3887     vhc->encode_hpt_for_kvm_pr = spapr_encode_hpt_for_kvm_pr;
3888     xic->ics_get = spapr_ics_get;
3889     xic->ics_resend = spapr_ics_resend;
3890     xic->icp_get = spapr_icp_get;
3891     ispc->print_info = spapr_pic_print_info;
3892     /* Force NUMA node memory size to be a multiple of
3893      * SPAPR_MEMORY_BLOCK_SIZE (256M) since that's the granularity
3894      * in which LMBs are represented and hot-added
3895      */
3896     mc->numa_mem_align_shift = 28;
3897 
3898     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_OFF;
3899     smc->default_caps.caps[SPAPR_CAP_VSX] = SPAPR_CAP_ON;
3900     smc->default_caps.caps[SPAPR_CAP_DFP] = SPAPR_CAP_ON;
3901     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_BROKEN;
3902     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_BROKEN;
3903     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_BROKEN;
3904     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 16; /* 64kiB */
3905     spapr_caps_add_properties(smc, &error_abort);
3906     smc->irq = &spapr_irq_xics;
3907 }
3908 
3909 static const TypeInfo spapr_machine_info = {
3910     .name          = TYPE_SPAPR_MACHINE,
3911     .parent        = TYPE_MACHINE,
3912     .abstract      = true,
3913     .instance_size = sizeof(sPAPRMachineState),
3914     .instance_init = spapr_instance_init,
3915     .instance_finalize = spapr_machine_finalizefn,
3916     .class_size    = sizeof(sPAPRMachineClass),
3917     .class_init    = spapr_machine_class_init,
3918     .interfaces = (InterfaceInfo[]) {
3919         { TYPE_FW_PATH_PROVIDER },
3920         { TYPE_NMI },
3921         { TYPE_HOTPLUG_HANDLER },
3922         { TYPE_PPC_VIRTUAL_HYPERVISOR },
3923         { TYPE_XICS_FABRIC },
3924         { TYPE_INTERRUPT_STATS_PROVIDER },
3925         { }
3926     },
3927 };
3928 
3929 #define DEFINE_SPAPR_MACHINE(suffix, verstr, latest)                 \
3930     static void spapr_machine_##suffix##_class_init(ObjectClass *oc, \
3931                                                     void *data)      \
3932     {                                                                \
3933         MachineClass *mc = MACHINE_CLASS(oc);                        \
3934         spapr_machine_##suffix##_class_options(mc);                  \
3935         if (latest) {                                                \
3936             mc->alias = "pseries";                                   \
3937             mc->is_default = 1;                                      \
3938         }                                                            \
3939     }                                                                \
3940     static void spapr_machine_##suffix##_instance_init(Object *obj)  \
3941     {                                                                \
3942         MachineState *machine = MACHINE(obj);                        \
3943         spapr_machine_##suffix##_instance_options(machine);          \
3944     }                                                                \
3945     static const TypeInfo spapr_machine_##suffix##_info = {          \
3946         .name = MACHINE_TYPE_NAME("pseries-" verstr),                \
3947         .parent = TYPE_SPAPR_MACHINE,                                \
3948         .class_init = spapr_machine_##suffix##_class_init,           \
3949         .instance_init = spapr_machine_##suffix##_instance_init,     \
3950     };                                                               \
3951     static void spapr_machine_register_##suffix(void)                \
3952     {                                                                \
3953         type_register(&spapr_machine_##suffix##_info);               \
3954     }                                                                \
3955     type_init(spapr_machine_register_##suffix)
3956 
3957  /*
3958  * pseries-3.1
3959  */
3960 static void spapr_machine_3_1_instance_options(MachineState *machine)
3961 {
3962 }
3963 
3964 static void spapr_machine_3_1_class_options(MachineClass *mc)
3965 {
3966     /* Defaults for the latest behaviour inherited from the base class */
3967 }
3968 
3969 DEFINE_SPAPR_MACHINE(3_1, "3.1", true);
3970 
3971 /*
3972  * pseries-3.0
3973  */
3974 #define SPAPR_COMPAT_3_0                                              \
3975     HW_COMPAT_3_0
3976 
3977 static void spapr_machine_3_0_instance_options(MachineState *machine)
3978 {
3979     spapr_machine_3_1_instance_options(machine);
3980 }
3981 
3982 static void spapr_machine_3_0_class_options(MachineClass *mc)
3983 {
3984     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3985 
3986     spapr_machine_3_1_class_options(mc);
3987     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_3_0);
3988 
3989     smc->legacy_irq_allocation = true;
3990     smc->irq = &spapr_irq_xics_legacy;
3991 }
3992 
3993 DEFINE_SPAPR_MACHINE(3_0, "3.0", false);
3994 
3995 /*
3996  * pseries-2.12
3997  */
3998 #define SPAPR_COMPAT_2_12                                              \
3999     HW_COMPAT_2_12                                                     \
4000     {                                                                  \
4001         .driver = TYPE_POWERPC_CPU,                                    \
4002         .property = "pre-3.0-migration",                               \
4003         .value    = "on",                                              \
4004     },                                                                 \
4005     {                                                                  \
4006         .driver = TYPE_SPAPR_CPU_CORE,                                 \
4007         .property = "pre-3.0-migration",                               \
4008         .value    = "on",                                              \
4009     },
4010 
4011 static void spapr_machine_2_12_instance_options(MachineState *machine)
4012 {
4013     spapr_machine_3_0_instance_options(machine);
4014 }
4015 
4016 static void spapr_machine_2_12_class_options(MachineClass *mc)
4017 {
4018     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4019 
4020     spapr_machine_3_0_class_options(mc);
4021     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_12);
4022 
4023     /* We depend on kvm_enabled() to choose a default value for the
4024      * hpt-max-page-size capability. Of course we can't do it here
4025      * because this is too early and the HW accelerator isn't initialzed
4026      * yet. Postpone this to machine init (see default_caps_with_cpu()).
4027      */
4028     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 0;
4029 }
4030 
4031 DEFINE_SPAPR_MACHINE(2_12, "2.12", false);
4032 
4033 static void spapr_machine_2_12_sxxm_instance_options(MachineState *machine)
4034 {
4035     spapr_machine_2_12_instance_options(machine);
4036 }
4037 
4038 static void spapr_machine_2_12_sxxm_class_options(MachineClass *mc)
4039 {
4040     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4041 
4042     spapr_machine_2_12_class_options(mc);
4043     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND;
4044     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND;
4045     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_FIXED_CCD;
4046 }
4047 
4048 DEFINE_SPAPR_MACHINE(2_12_sxxm, "2.12-sxxm", false);
4049 
4050 /*
4051  * pseries-2.11
4052  */
4053 #define SPAPR_COMPAT_2_11                                              \
4054     HW_COMPAT_2_11
4055 
4056 static void spapr_machine_2_11_instance_options(MachineState *machine)
4057 {
4058     spapr_machine_2_12_instance_options(machine);
4059 }
4060 
4061 static void spapr_machine_2_11_class_options(MachineClass *mc)
4062 {
4063     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4064 
4065     spapr_machine_2_12_class_options(mc);
4066     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_ON;
4067     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_11);
4068 }
4069 
4070 DEFINE_SPAPR_MACHINE(2_11, "2.11", false);
4071 
4072 /*
4073  * pseries-2.10
4074  */
4075 #define SPAPR_COMPAT_2_10                                              \
4076     HW_COMPAT_2_10
4077 
4078 static void spapr_machine_2_10_instance_options(MachineState *machine)
4079 {
4080     spapr_machine_2_11_instance_options(machine);
4081 }
4082 
4083 static void spapr_machine_2_10_class_options(MachineClass *mc)
4084 {
4085     spapr_machine_2_11_class_options(mc);
4086     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_10);
4087 }
4088 
4089 DEFINE_SPAPR_MACHINE(2_10, "2.10", false);
4090 
4091 /*
4092  * pseries-2.9
4093  */
4094 #define SPAPR_COMPAT_2_9                                               \
4095     HW_COMPAT_2_9                                                      \
4096     {                                                                  \
4097         .driver = TYPE_POWERPC_CPU,                                    \
4098         .property = "pre-2.10-migration",                              \
4099         .value    = "on",                                              \
4100     },                                                                 \
4101 
4102 static void spapr_machine_2_9_instance_options(MachineState *machine)
4103 {
4104     spapr_machine_2_10_instance_options(machine);
4105 }
4106 
4107 static void spapr_machine_2_9_class_options(MachineClass *mc)
4108 {
4109     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4110 
4111     spapr_machine_2_10_class_options(mc);
4112     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_9);
4113     mc->numa_auto_assign_ram = numa_legacy_auto_assign_ram;
4114     smc->pre_2_10_has_unused_icps = true;
4115     smc->resize_hpt_default = SPAPR_RESIZE_HPT_DISABLED;
4116 }
4117 
4118 DEFINE_SPAPR_MACHINE(2_9, "2.9", false);
4119 
4120 /*
4121  * pseries-2.8
4122  */
4123 #define SPAPR_COMPAT_2_8                                        \
4124     HW_COMPAT_2_8                                               \
4125     {                                                           \
4126         .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,                 \
4127         .property = "pcie-extended-configuration-space",        \
4128         .value    = "off",                                      \
4129     },
4130 
4131 static void spapr_machine_2_8_instance_options(MachineState *machine)
4132 {
4133     spapr_machine_2_9_instance_options(machine);
4134 }
4135 
4136 static void spapr_machine_2_8_class_options(MachineClass *mc)
4137 {
4138     spapr_machine_2_9_class_options(mc);
4139     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_8);
4140     mc->numa_mem_align_shift = 23;
4141 }
4142 
4143 DEFINE_SPAPR_MACHINE(2_8, "2.8", false);
4144 
4145 /*
4146  * pseries-2.7
4147  */
4148 #define SPAPR_COMPAT_2_7                            \
4149     HW_COMPAT_2_7                                   \
4150     {                                               \
4151         .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,     \
4152         .property = "mem_win_size",                 \
4153         .value    = stringify(SPAPR_PCI_2_7_MMIO_WIN_SIZE),\
4154     },                                              \
4155     {                                               \
4156         .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,     \
4157         .property = "mem64_win_size",               \
4158         .value    = "0",                            \
4159     },                                              \
4160     {                                               \
4161         .driver = TYPE_POWERPC_CPU,                 \
4162         .property = "pre-2.8-migration",            \
4163         .value    = "on",                           \
4164     },                                              \
4165     {                                               \
4166         .driver = TYPE_SPAPR_PCI_HOST_BRIDGE,       \
4167         .property = "pre-2.8-migration",            \
4168         .value    = "on",                           \
4169     },
4170 
4171 static void phb_placement_2_7(sPAPRMachineState *spapr, uint32_t index,
4172                               uint64_t *buid, hwaddr *pio,
4173                               hwaddr *mmio32, hwaddr *mmio64,
4174                               unsigned n_dma, uint32_t *liobns, Error **errp)
4175 {
4176     /* Legacy PHB placement for pseries-2.7 and earlier machine types */
4177     const uint64_t base_buid = 0x800000020000000ULL;
4178     const hwaddr phb_spacing = 0x1000000000ULL; /* 64 GiB */
4179     const hwaddr mmio_offset = 0xa0000000; /* 2 GiB + 512 MiB */
4180     const hwaddr pio_offset = 0x80000000; /* 2 GiB */
4181     const uint32_t max_index = 255;
4182     const hwaddr phb0_alignment = 0x10000000000ULL; /* 1 TiB */
4183 
4184     uint64_t ram_top = MACHINE(spapr)->ram_size;
4185     hwaddr phb0_base, phb_base;
4186     int i;
4187 
4188     /* Do we have device memory? */
4189     if (MACHINE(spapr)->maxram_size > ram_top) {
4190         /* Can't just use maxram_size, because there may be an
4191          * alignment gap between normal and device memory regions
4192          */
4193         ram_top = MACHINE(spapr)->device_memory->base +
4194             memory_region_size(&MACHINE(spapr)->device_memory->mr);
4195     }
4196 
4197     phb0_base = QEMU_ALIGN_UP(ram_top, phb0_alignment);
4198 
4199     if (index > max_index) {
4200         error_setg(errp, "\"index\" for PAPR PHB is too large (max %u)",
4201                    max_index);
4202         return;
4203     }
4204 
4205     *buid = base_buid + index;
4206     for (i = 0; i < n_dma; ++i) {
4207         liobns[i] = SPAPR_PCI_LIOBN(index, i);
4208     }
4209 
4210     phb_base = phb0_base + index * phb_spacing;
4211     *pio = phb_base + pio_offset;
4212     *mmio32 = phb_base + mmio_offset;
4213     /*
4214      * We don't set the 64-bit MMIO window, relying on the PHB's
4215      * fallback behaviour of automatically splitting a large "32-bit"
4216      * window into contiguous 32-bit and 64-bit windows
4217      */
4218 }
4219 
4220 static void spapr_machine_2_7_instance_options(MachineState *machine)
4221 {
4222     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
4223 
4224     spapr_machine_2_8_instance_options(machine);
4225     spapr->use_hotplug_event_source = false;
4226 }
4227 
4228 static void spapr_machine_2_7_class_options(MachineClass *mc)
4229 {
4230     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4231 
4232     spapr_machine_2_8_class_options(mc);
4233     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power7_v2.3");
4234     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_7);
4235     smc->phb_placement = phb_placement_2_7;
4236 }
4237 
4238 DEFINE_SPAPR_MACHINE(2_7, "2.7", false);
4239 
4240 /*
4241  * pseries-2.6
4242  */
4243 #define SPAPR_COMPAT_2_6 \
4244     HW_COMPAT_2_6 \
4245     { \
4246         .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,\
4247         .property = "ddw",\
4248         .value    = stringify(off),\
4249     },
4250 
4251 static void spapr_machine_2_6_instance_options(MachineState *machine)
4252 {
4253     spapr_machine_2_7_instance_options(machine);
4254 }
4255 
4256 static void spapr_machine_2_6_class_options(MachineClass *mc)
4257 {
4258     spapr_machine_2_7_class_options(mc);
4259     mc->has_hotpluggable_cpus = false;
4260     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_6);
4261 }
4262 
4263 DEFINE_SPAPR_MACHINE(2_6, "2.6", false);
4264 
4265 /*
4266  * pseries-2.5
4267  */
4268 #define SPAPR_COMPAT_2_5 \
4269     HW_COMPAT_2_5 \
4270     { \
4271         .driver   = "spapr-vlan", \
4272         .property = "use-rx-buffer-pools", \
4273         .value    = "off", \
4274     },
4275 
4276 static void spapr_machine_2_5_instance_options(MachineState *machine)
4277 {
4278     spapr_machine_2_6_instance_options(machine);
4279 }
4280 
4281 static void spapr_machine_2_5_class_options(MachineClass *mc)
4282 {
4283     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4284 
4285     spapr_machine_2_6_class_options(mc);
4286     smc->use_ohci_by_default = true;
4287     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_5);
4288 }
4289 
4290 DEFINE_SPAPR_MACHINE(2_5, "2.5", false);
4291 
4292 /*
4293  * pseries-2.4
4294  */
4295 #define SPAPR_COMPAT_2_4 \
4296         HW_COMPAT_2_4
4297 
4298 static void spapr_machine_2_4_instance_options(MachineState *machine)
4299 {
4300     spapr_machine_2_5_instance_options(machine);
4301 }
4302 
4303 static void spapr_machine_2_4_class_options(MachineClass *mc)
4304 {
4305     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4306 
4307     spapr_machine_2_5_class_options(mc);
4308     smc->dr_lmb_enabled = false;
4309     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_4);
4310 }
4311 
4312 DEFINE_SPAPR_MACHINE(2_4, "2.4", false);
4313 
4314 /*
4315  * pseries-2.3
4316  */
4317 #define SPAPR_COMPAT_2_3 \
4318         HW_COMPAT_2_3 \
4319         {\
4320             .driver   = "spapr-pci-host-bridge",\
4321             .property = "dynamic-reconfiguration",\
4322             .value    = "off",\
4323         },
4324 
4325 static void spapr_machine_2_3_instance_options(MachineState *machine)
4326 {
4327     spapr_machine_2_4_instance_options(machine);
4328 }
4329 
4330 static void spapr_machine_2_3_class_options(MachineClass *mc)
4331 {
4332     spapr_machine_2_4_class_options(mc);
4333     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_3);
4334 }
4335 DEFINE_SPAPR_MACHINE(2_3, "2.3", false);
4336 
4337 /*
4338  * pseries-2.2
4339  */
4340 
4341 #define SPAPR_COMPAT_2_2 \
4342         HW_COMPAT_2_2 \
4343         {\
4344             .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,\
4345             .property = "mem_win_size",\
4346             .value    = "0x20000000",\
4347         },
4348 
4349 static void spapr_machine_2_2_instance_options(MachineState *machine)
4350 {
4351     spapr_machine_2_3_instance_options(machine);
4352     machine->suppress_vmdesc = true;
4353 }
4354 
4355 static void spapr_machine_2_2_class_options(MachineClass *mc)
4356 {
4357     spapr_machine_2_3_class_options(mc);
4358     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_2);
4359 }
4360 DEFINE_SPAPR_MACHINE(2_2, "2.2", false);
4361 
4362 /*
4363  * pseries-2.1
4364  */
4365 #define SPAPR_COMPAT_2_1 \
4366         HW_COMPAT_2_1
4367 
4368 static void spapr_machine_2_1_instance_options(MachineState *machine)
4369 {
4370     spapr_machine_2_2_instance_options(machine);
4371 }
4372 
4373 static void spapr_machine_2_1_class_options(MachineClass *mc)
4374 {
4375     spapr_machine_2_2_class_options(mc);
4376     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_1);
4377 }
4378 DEFINE_SPAPR_MACHINE(2_1, "2.1", false);
4379 
4380 static void spapr_machine_register_types(void)
4381 {
4382     type_register_static(&spapr_machine_info);
4383 }
4384 
4385 type_init(spapr_machine_register_types)
4386