xref: /openbmc/qemu/hw/ppc/spapr.c (revision bf31c56f096ae11367c6ba19a47387d008791996)
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 uint32_t spapr_pc_dimm_node(MemoryDeviceInfoList *list, ram_addr_t addr)
614 {
615     MemoryDeviceInfoList *info;
616 
617     for (info = list; info; info = info->next) {
618         MemoryDeviceInfo *value = info->value;
619 
620         if (value && value->type == MEMORY_DEVICE_INFO_KIND_DIMM) {
621             PCDIMMDeviceInfo *pcdimm_info = value->u.dimm.data;
622 
623             if (addr >= pcdimm_info->addr &&
624                 addr < (pcdimm_info->addr + pcdimm_info->size)) {
625                 return pcdimm_info->node;
626             }
627         }
628     }
629 
630     return -1;
631 }
632 
633 struct sPAPRDrconfCellV2 {
634      uint32_t seq_lmbs;
635      uint64_t base_addr;
636      uint32_t drc_index;
637      uint32_t aa_index;
638      uint32_t flags;
639 } QEMU_PACKED;
640 
641 typedef struct DrconfCellQueue {
642     struct sPAPRDrconfCellV2 cell;
643     QSIMPLEQ_ENTRY(DrconfCellQueue) entry;
644 } DrconfCellQueue;
645 
646 static DrconfCellQueue *
647 spapr_get_drconf_cell(uint32_t seq_lmbs, uint64_t base_addr,
648                       uint32_t drc_index, uint32_t aa_index,
649                       uint32_t flags)
650 {
651     DrconfCellQueue *elem;
652 
653     elem = g_malloc0(sizeof(*elem));
654     elem->cell.seq_lmbs = cpu_to_be32(seq_lmbs);
655     elem->cell.base_addr = cpu_to_be64(base_addr);
656     elem->cell.drc_index = cpu_to_be32(drc_index);
657     elem->cell.aa_index = cpu_to_be32(aa_index);
658     elem->cell.flags = cpu_to_be32(flags);
659 
660     return elem;
661 }
662 
663 /* ibm,dynamic-memory-v2 */
664 static int spapr_populate_drmem_v2(sPAPRMachineState *spapr, void *fdt,
665                                    int offset, MemoryDeviceInfoList *dimms)
666 {
667     MachineState *machine = MACHINE(spapr);
668     uint8_t *int_buf, *cur_index, buf_len;
669     int ret;
670     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
671     uint64_t addr, cur_addr, size;
672     uint32_t nr_boot_lmbs = (machine->device_memory->base / lmb_size);
673     uint64_t mem_end = machine->device_memory->base +
674                        memory_region_size(&machine->device_memory->mr);
675     uint32_t node, nr_entries = 0;
676     sPAPRDRConnector *drc;
677     DrconfCellQueue *elem, *next;
678     MemoryDeviceInfoList *info;
679     QSIMPLEQ_HEAD(, DrconfCellQueue) drconf_queue
680         = QSIMPLEQ_HEAD_INITIALIZER(drconf_queue);
681 
682     /* Entry to cover RAM and the gap area */
683     elem = spapr_get_drconf_cell(nr_boot_lmbs, 0, 0, -1,
684                                  SPAPR_LMB_FLAGS_RESERVED |
685                                  SPAPR_LMB_FLAGS_DRC_INVALID);
686     QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
687     nr_entries++;
688 
689     cur_addr = machine->device_memory->base;
690     for (info = dimms; info; info = info->next) {
691         PCDIMMDeviceInfo *di = info->value->u.dimm.data;
692 
693         addr = di->addr;
694         size = di->size;
695         node = di->node;
696 
697         /* Entry for hot-pluggable area */
698         if (cur_addr < addr) {
699             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
700             g_assert(drc);
701             elem = spapr_get_drconf_cell((addr - cur_addr) / lmb_size,
702                                          cur_addr, spapr_drc_index(drc), -1, 0);
703             QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
704             nr_entries++;
705         }
706 
707         /* Entry for DIMM */
708         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, addr / lmb_size);
709         g_assert(drc);
710         elem = spapr_get_drconf_cell(size / lmb_size, addr,
711                                      spapr_drc_index(drc), node,
712                                      SPAPR_LMB_FLAGS_ASSIGNED);
713         QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
714         nr_entries++;
715         cur_addr = addr + size;
716     }
717 
718     /* Entry for remaining hotpluggable area */
719     if (cur_addr < mem_end) {
720         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
721         g_assert(drc);
722         elem = spapr_get_drconf_cell((mem_end - cur_addr) / lmb_size,
723                                      cur_addr, spapr_drc_index(drc), -1, 0);
724         QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
725         nr_entries++;
726     }
727 
728     buf_len = nr_entries * sizeof(struct sPAPRDrconfCellV2) + sizeof(uint32_t);
729     int_buf = cur_index = g_malloc0(buf_len);
730     *(uint32_t *)int_buf = cpu_to_be32(nr_entries);
731     cur_index += sizeof(nr_entries);
732 
733     QSIMPLEQ_FOREACH_SAFE(elem, &drconf_queue, entry, next) {
734         memcpy(cur_index, &elem->cell, sizeof(elem->cell));
735         cur_index += sizeof(elem->cell);
736         QSIMPLEQ_REMOVE(&drconf_queue, elem, DrconfCellQueue, entry);
737         g_free(elem);
738     }
739 
740     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory-v2", int_buf, buf_len);
741     g_free(int_buf);
742     if (ret < 0) {
743         return -1;
744     }
745     return 0;
746 }
747 
748 /* ibm,dynamic-memory */
749 static int spapr_populate_drmem_v1(sPAPRMachineState *spapr, void *fdt,
750                                    int offset, MemoryDeviceInfoList *dimms)
751 {
752     MachineState *machine = MACHINE(spapr);
753     int i, ret;
754     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
755     uint32_t device_lmb_start = machine->device_memory->base / lmb_size;
756     uint32_t nr_lmbs = (machine->device_memory->base +
757                        memory_region_size(&machine->device_memory->mr)) /
758                        lmb_size;
759     uint32_t *int_buf, *cur_index, buf_len;
760 
761     /*
762      * Allocate enough buffer size to fit in ibm,dynamic-memory
763      */
764     buf_len = (nr_lmbs * SPAPR_DR_LMB_LIST_ENTRY_SIZE + 1) * sizeof(uint32_t);
765     cur_index = int_buf = g_malloc0(buf_len);
766     int_buf[0] = cpu_to_be32(nr_lmbs);
767     cur_index++;
768     for (i = 0; i < nr_lmbs; i++) {
769         uint64_t addr = i * lmb_size;
770         uint32_t *dynamic_memory = cur_index;
771 
772         if (i >= device_lmb_start) {
773             sPAPRDRConnector *drc;
774 
775             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, i);
776             g_assert(drc);
777 
778             dynamic_memory[0] = cpu_to_be32(addr >> 32);
779             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
780             dynamic_memory[2] = cpu_to_be32(spapr_drc_index(drc));
781             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
782             dynamic_memory[4] = cpu_to_be32(spapr_pc_dimm_node(dimms, addr));
783             if (memory_region_present(get_system_memory(), addr)) {
784                 dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_ASSIGNED);
785             } else {
786                 dynamic_memory[5] = cpu_to_be32(0);
787             }
788         } else {
789             /*
790              * LMB information for RMA, boot time RAM and gap b/n RAM and
791              * device memory region -- all these are marked as reserved
792              * and as having no valid DRC.
793              */
794             dynamic_memory[0] = cpu_to_be32(addr >> 32);
795             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
796             dynamic_memory[2] = cpu_to_be32(0);
797             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
798             dynamic_memory[4] = cpu_to_be32(-1);
799             dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_RESERVED |
800                                             SPAPR_LMB_FLAGS_DRC_INVALID);
801         }
802 
803         cur_index += SPAPR_DR_LMB_LIST_ENTRY_SIZE;
804     }
805     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory", int_buf, buf_len);
806     g_free(int_buf);
807     if (ret < 0) {
808         return -1;
809     }
810     return 0;
811 }
812 
813 /*
814  * Adds ibm,dynamic-reconfiguration-memory node.
815  * Refer to docs/specs/ppc-spapr-hotplug.txt for the documentation
816  * of this device tree node.
817  */
818 static int spapr_populate_drconf_memory(sPAPRMachineState *spapr, void *fdt)
819 {
820     MachineState *machine = MACHINE(spapr);
821     int ret, i, offset;
822     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
823     uint32_t prop_lmb_size[] = {0, cpu_to_be32(lmb_size)};
824     uint32_t *int_buf, *cur_index, buf_len;
825     int nr_nodes = nb_numa_nodes ? nb_numa_nodes : 1;
826     MemoryDeviceInfoList *dimms = NULL;
827 
828     /*
829      * Don't create the node if there is no device memory
830      */
831     if (machine->ram_size == machine->maxram_size) {
832         return 0;
833     }
834 
835     offset = fdt_add_subnode(fdt, 0, "ibm,dynamic-reconfiguration-memory");
836 
837     ret = fdt_setprop(fdt, offset, "ibm,lmb-size", prop_lmb_size,
838                     sizeof(prop_lmb_size));
839     if (ret < 0) {
840         return ret;
841     }
842 
843     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-flags-mask", 0xff);
844     if (ret < 0) {
845         return ret;
846     }
847 
848     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-preservation-time", 0x0);
849     if (ret < 0) {
850         return ret;
851     }
852 
853     /* ibm,dynamic-memory or ibm,dynamic-memory-v2 */
854     dimms = qmp_memory_device_list();
855     if (spapr_ovec_test(spapr->ov5_cas, OV5_DRMEM_V2)) {
856         ret = spapr_populate_drmem_v2(spapr, fdt, offset, dimms);
857     } else {
858         ret = spapr_populate_drmem_v1(spapr, fdt, offset, dimms);
859     }
860     qapi_free_MemoryDeviceInfoList(dimms);
861 
862     if (ret < 0) {
863         return ret;
864     }
865 
866     /* ibm,associativity-lookup-arrays */
867     buf_len = (nr_nodes * 4 + 2) * sizeof(uint32_t);
868     cur_index = int_buf = g_malloc0(buf_len);
869 
870     cur_index = int_buf;
871     int_buf[0] = cpu_to_be32(nr_nodes);
872     int_buf[1] = cpu_to_be32(4); /* Number of entries per associativity list */
873     cur_index += 2;
874     for (i = 0; i < nr_nodes; i++) {
875         uint32_t associativity[] = {
876             cpu_to_be32(0x0),
877             cpu_to_be32(0x0),
878             cpu_to_be32(0x0),
879             cpu_to_be32(i)
880         };
881         memcpy(cur_index, associativity, sizeof(associativity));
882         cur_index += 4;
883     }
884     ret = fdt_setprop(fdt, offset, "ibm,associativity-lookup-arrays", int_buf,
885             (cur_index - int_buf) * sizeof(uint32_t));
886     g_free(int_buf);
887 
888     return ret;
889 }
890 
891 static int spapr_dt_cas_updates(sPAPRMachineState *spapr, void *fdt,
892                                 sPAPROptionVector *ov5_updates)
893 {
894     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
895     int ret = 0, offset;
896 
897     /* Generate ibm,dynamic-reconfiguration-memory node if required */
898     if (spapr_ovec_test(ov5_updates, OV5_DRCONF_MEMORY)) {
899         g_assert(smc->dr_lmb_enabled);
900         ret = spapr_populate_drconf_memory(spapr, fdt);
901         if (ret) {
902             goto out;
903         }
904     }
905 
906     offset = fdt_path_offset(fdt, "/chosen");
907     if (offset < 0) {
908         offset = fdt_add_subnode(fdt, 0, "chosen");
909         if (offset < 0) {
910             return offset;
911         }
912     }
913     ret = spapr_ovec_populate_dt(fdt, offset, spapr->ov5_cas,
914                                  "ibm,architecture-vec-5");
915 
916 out:
917     return ret;
918 }
919 
920 static bool spapr_hotplugged_dev_before_cas(void)
921 {
922     Object *drc_container, *obj;
923     ObjectProperty *prop;
924     ObjectPropertyIterator iter;
925 
926     drc_container = container_get(object_get_root(), "/dr-connector");
927     object_property_iter_init(&iter, drc_container);
928     while ((prop = object_property_iter_next(&iter))) {
929         if (!strstart(prop->type, "link<", NULL)) {
930             continue;
931         }
932         obj = object_property_get_link(drc_container, prop->name, NULL);
933         if (spapr_drc_needed(obj)) {
934             return true;
935         }
936     }
937     return false;
938 }
939 
940 int spapr_h_cas_compose_response(sPAPRMachineState *spapr,
941                                  target_ulong addr, target_ulong size,
942                                  sPAPROptionVector *ov5_updates)
943 {
944     void *fdt, *fdt_skel;
945     sPAPRDeviceTreeUpdateHeader hdr = { .version_id = 1 };
946 
947     if (spapr_hotplugged_dev_before_cas()) {
948         return 1;
949     }
950 
951     if (size < sizeof(hdr) || size > FW_MAX_SIZE) {
952         error_report("SLOF provided an unexpected CAS buffer size "
953                      TARGET_FMT_lu " (min: %zu, max: %u)",
954                      size, sizeof(hdr), FW_MAX_SIZE);
955         exit(EXIT_FAILURE);
956     }
957 
958     size -= sizeof(hdr);
959 
960     /* Create skeleton */
961     fdt_skel = g_malloc0(size);
962     _FDT((fdt_create(fdt_skel, size)));
963     _FDT((fdt_finish_reservemap(fdt_skel)));
964     _FDT((fdt_begin_node(fdt_skel, "")));
965     _FDT((fdt_end_node(fdt_skel)));
966     _FDT((fdt_finish(fdt_skel)));
967     fdt = g_malloc0(size);
968     _FDT((fdt_open_into(fdt_skel, fdt, size)));
969     g_free(fdt_skel);
970 
971     /* Fixup cpu nodes */
972     _FDT((spapr_fixup_cpu_dt(fdt, spapr)));
973 
974     if (spapr_dt_cas_updates(spapr, fdt, ov5_updates)) {
975         return -1;
976     }
977 
978     /* Pack resulting tree */
979     _FDT((fdt_pack(fdt)));
980 
981     if (fdt_totalsize(fdt) + sizeof(hdr) > size) {
982         trace_spapr_cas_failed(size);
983         return -1;
984     }
985 
986     cpu_physical_memory_write(addr, &hdr, sizeof(hdr));
987     cpu_physical_memory_write(addr + sizeof(hdr), fdt, fdt_totalsize(fdt));
988     trace_spapr_cas_continue(fdt_totalsize(fdt) + sizeof(hdr));
989     g_free(fdt);
990 
991     return 0;
992 }
993 
994 static void spapr_dt_rtas(sPAPRMachineState *spapr, void *fdt)
995 {
996     int rtas;
997     GString *hypertas = g_string_sized_new(256);
998     GString *qemu_hypertas = g_string_sized_new(256);
999     uint32_t refpoints[] = { cpu_to_be32(0x4), cpu_to_be32(0x4) };
1000     uint64_t max_device_addr = MACHINE(spapr)->device_memory->base +
1001         memory_region_size(&MACHINE(spapr)->device_memory->mr);
1002     uint32_t lrdr_capacity[] = {
1003         cpu_to_be32(max_device_addr >> 32),
1004         cpu_to_be32(max_device_addr & 0xffffffff),
1005         0, cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE),
1006         cpu_to_be32(max_cpus / smp_threads),
1007     };
1008     uint32_t maxdomains[] = {
1009         cpu_to_be32(4),
1010         cpu_to_be32(0),
1011         cpu_to_be32(0),
1012         cpu_to_be32(0),
1013         cpu_to_be32(nb_numa_nodes ? nb_numa_nodes - 1 : 0),
1014     };
1015 
1016     _FDT(rtas = fdt_add_subnode(fdt, 0, "rtas"));
1017 
1018     /* hypertas */
1019     add_str(hypertas, "hcall-pft");
1020     add_str(hypertas, "hcall-term");
1021     add_str(hypertas, "hcall-dabr");
1022     add_str(hypertas, "hcall-interrupt");
1023     add_str(hypertas, "hcall-tce");
1024     add_str(hypertas, "hcall-vio");
1025     add_str(hypertas, "hcall-splpar");
1026     add_str(hypertas, "hcall-bulk");
1027     add_str(hypertas, "hcall-set-mode");
1028     add_str(hypertas, "hcall-sprg0");
1029     add_str(hypertas, "hcall-copy");
1030     add_str(hypertas, "hcall-debug");
1031     add_str(qemu_hypertas, "hcall-memop1");
1032 
1033     if (!kvm_enabled() || kvmppc_spapr_use_multitce()) {
1034         add_str(hypertas, "hcall-multi-tce");
1035     }
1036 
1037     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
1038         add_str(hypertas, "hcall-hpt-resize");
1039     }
1040 
1041     _FDT(fdt_setprop(fdt, rtas, "ibm,hypertas-functions",
1042                      hypertas->str, hypertas->len));
1043     g_string_free(hypertas, TRUE);
1044     _FDT(fdt_setprop(fdt, rtas, "qemu,hypertas-functions",
1045                      qemu_hypertas->str, qemu_hypertas->len));
1046     g_string_free(qemu_hypertas, TRUE);
1047 
1048     _FDT(fdt_setprop(fdt, rtas, "ibm,associativity-reference-points",
1049                      refpoints, sizeof(refpoints)));
1050 
1051     _FDT(fdt_setprop(fdt, rtas, "ibm,max-associativity-domains",
1052                      maxdomains, sizeof(maxdomains)));
1053 
1054     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-error-log-max",
1055                           RTAS_ERROR_LOG_MAX));
1056     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-event-scan-rate",
1057                           RTAS_EVENT_SCAN_RATE));
1058 
1059     g_assert(msi_nonbroken);
1060     _FDT(fdt_setprop(fdt, rtas, "ibm,change-msix-capable", NULL, 0));
1061 
1062     /*
1063      * According to PAPR, rtas ibm,os-term does not guarantee a return
1064      * back to the guest cpu.
1065      *
1066      * While an additional ibm,extended-os-term property indicates
1067      * that rtas call return will always occur. Set this property.
1068      */
1069     _FDT(fdt_setprop(fdt, rtas, "ibm,extended-os-term", NULL, 0));
1070 
1071     _FDT(fdt_setprop(fdt, rtas, "ibm,lrdr-capacity",
1072                      lrdr_capacity, sizeof(lrdr_capacity)));
1073 
1074     spapr_dt_rtas_tokens(fdt, rtas);
1075 }
1076 
1077 /* Prepare ibm,arch-vec-5-platform-support, which indicates the MMU features
1078  * that the guest may request and thus the valid values for bytes 24..26 of
1079  * option vector 5: */
1080 static void spapr_dt_ov5_platform_support(void *fdt, int chosen)
1081 {
1082     PowerPCCPU *first_ppc_cpu = POWERPC_CPU(first_cpu);
1083 
1084     char val[2 * 4] = {
1085         23, 0x00, /* Xive mode, filled in below. */
1086         24, 0x00, /* Hash/Radix, filled in below. */
1087         25, 0x00, /* Hash options: Segment Tables == no, GTSE == no. */
1088         26, 0x40, /* Radix options: GTSE == yes. */
1089     };
1090 
1091     if (!ppc_check_compat(first_ppc_cpu, CPU_POWERPC_LOGICAL_3_00, 0,
1092                           first_ppc_cpu->compat_pvr)) {
1093         /* If we're in a pre POWER9 compat mode then the guest should do hash */
1094         val[3] = 0x00; /* Hash */
1095     } else if (kvm_enabled()) {
1096         if (kvmppc_has_cap_mmu_radix() && kvmppc_has_cap_mmu_hash_v3()) {
1097             val[3] = 0x80; /* OV5_MMU_BOTH */
1098         } else if (kvmppc_has_cap_mmu_radix()) {
1099             val[3] = 0x40; /* OV5_MMU_RADIX_300 */
1100         } else {
1101             val[3] = 0x00; /* Hash */
1102         }
1103     } else {
1104         /* V3 MMU supports both hash and radix in tcg (with dynamic switching) */
1105         val[3] = 0xC0;
1106     }
1107     _FDT(fdt_setprop(fdt, chosen, "ibm,arch-vec-5-platform-support",
1108                      val, sizeof(val)));
1109 }
1110 
1111 static void spapr_dt_chosen(sPAPRMachineState *spapr, void *fdt)
1112 {
1113     MachineState *machine = MACHINE(spapr);
1114     int chosen;
1115     const char *boot_device = machine->boot_order;
1116     char *stdout_path = spapr_vio_stdout_path(spapr->vio_bus);
1117     size_t cb = 0;
1118     char *bootlist = get_boot_devices_list(&cb);
1119 
1120     _FDT(chosen = fdt_add_subnode(fdt, 0, "chosen"));
1121 
1122     _FDT(fdt_setprop_string(fdt, chosen, "bootargs", machine->kernel_cmdline));
1123     _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-start",
1124                           spapr->initrd_base));
1125     _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-end",
1126                           spapr->initrd_base + spapr->initrd_size));
1127 
1128     if (spapr->kernel_size) {
1129         uint64_t kprop[2] = { cpu_to_be64(KERNEL_LOAD_ADDR),
1130                               cpu_to_be64(spapr->kernel_size) };
1131 
1132         _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel",
1133                          &kprop, sizeof(kprop)));
1134         if (spapr->kernel_le) {
1135             _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel-le", NULL, 0));
1136         }
1137     }
1138     if (boot_menu) {
1139         _FDT((fdt_setprop_cell(fdt, chosen, "qemu,boot-menu", boot_menu)));
1140     }
1141     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-width", graphic_width));
1142     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-height", graphic_height));
1143     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-depth", graphic_depth));
1144 
1145     if (cb && bootlist) {
1146         int i;
1147 
1148         for (i = 0; i < cb; i++) {
1149             if (bootlist[i] == '\n') {
1150                 bootlist[i] = ' ';
1151             }
1152         }
1153         _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-list", bootlist));
1154     }
1155 
1156     if (boot_device && strlen(boot_device)) {
1157         _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-device", boot_device));
1158     }
1159 
1160     if (!spapr->has_graphics && stdout_path) {
1161         /*
1162          * "linux,stdout-path" and "stdout" properties are deprecated by linux
1163          * kernel. New platforms should only use the "stdout-path" property. Set
1164          * the new property and continue using older property to remain
1165          * compatible with the existing firmware.
1166          */
1167         _FDT(fdt_setprop_string(fdt, chosen, "linux,stdout-path", stdout_path));
1168         _FDT(fdt_setprop_string(fdt, chosen, "stdout-path", stdout_path));
1169     }
1170 
1171     spapr_dt_ov5_platform_support(fdt, chosen);
1172 
1173     g_free(stdout_path);
1174     g_free(bootlist);
1175 }
1176 
1177 static void spapr_dt_hypervisor(sPAPRMachineState *spapr, void *fdt)
1178 {
1179     /* The /hypervisor node isn't in PAPR - this is a hack to allow PR
1180      * KVM to work under pHyp with some guest co-operation */
1181     int hypervisor;
1182     uint8_t hypercall[16];
1183 
1184     _FDT(hypervisor = fdt_add_subnode(fdt, 0, "hypervisor"));
1185     /* indicate KVM hypercall interface */
1186     _FDT(fdt_setprop_string(fdt, hypervisor, "compatible", "linux,kvm"));
1187     if (kvmppc_has_cap_fixup_hcalls()) {
1188         /*
1189          * Older KVM versions with older guest kernels were broken
1190          * with the magic page, don't allow the guest to map it.
1191          */
1192         if (!kvmppc_get_hypercall(first_cpu->env_ptr, hypercall,
1193                                   sizeof(hypercall))) {
1194             _FDT(fdt_setprop(fdt, hypervisor, "hcall-instructions",
1195                              hypercall, sizeof(hypercall)));
1196         }
1197     }
1198 }
1199 
1200 static void *spapr_build_fdt(sPAPRMachineState *spapr,
1201                              hwaddr rtas_addr,
1202                              hwaddr rtas_size)
1203 {
1204     MachineState *machine = MACHINE(spapr);
1205     MachineClass *mc = MACHINE_GET_CLASS(machine);
1206     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1207     int ret;
1208     void *fdt;
1209     sPAPRPHBState *phb;
1210     char *buf;
1211 
1212     fdt = g_malloc0(FDT_MAX_SIZE);
1213     _FDT((fdt_create_empty_tree(fdt, FDT_MAX_SIZE)));
1214 
1215     /* Root node */
1216     _FDT(fdt_setprop_string(fdt, 0, "device_type", "chrp"));
1217     _FDT(fdt_setprop_string(fdt, 0, "model", "IBM pSeries (emulated by qemu)"));
1218     _FDT(fdt_setprop_string(fdt, 0, "compatible", "qemu,pseries"));
1219 
1220     /*
1221      * Add info to guest to indentify which host is it being run on
1222      * and what is the uuid of the guest
1223      */
1224     if (kvmppc_get_host_model(&buf)) {
1225         _FDT(fdt_setprop_string(fdt, 0, "host-model", buf));
1226         g_free(buf);
1227     }
1228     if (kvmppc_get_host_serial(&buf)) {
1229         _FDT(fdt_setprop_string(fdt, 0, "host-serial", buf));
1230         g_free(buf);
1231     }
1232 
1233     buf = qemu_uuid_unparse_strdup(&qemu_uuid);
1234 
1235     _FDT(fdt_setprop_string(fdt, 0, "vm,uuid", buf));
1236     if (qemu_uuid_set) {
1237         _FDT(fdt_setprop_string(fdt, 0, "system-id", buf));
1238     }
1239     g_free(buf);
1240 
1241     if (qemu_get_vm_name()) {
1242         _FDT(fdt_setprop_string(fdt, 0, "ibm,partition-name",
1243                                 qemu_get_vm_name()));
1244     }
1245 
1246     _FDT(fdt_setprop_cell(fdt, 0, "#address-cells", 2));
1247     _FDT(fdt_setprop_cell(fdt, 0, "#size-cells", 2));
1248 
1249     /* /interrupt controller */
1250     spapr_dt_xics(xics_max_server_number(spapr), fdt, PHANDLE_XICP);
1251 
1252     ret = spapr_populate_memory(spapr, fdt);
1253     if (ret < 0) {
1254         error_report("couldn't setup memory nodes in fdt");
1255         exit(1);
1256     }
1257 
1258     /* /vdevice */
1259     spapr_dt_vdevice(spapr->vio_bus, fdt);
1260 
1261     if (object_resolve_path_type("", TYPE_SPAPR_RNG, NULL)) {
1262         ret = spapr_rng_populate_dt(fdt);
1263         if (ret < 0) {
1264             error_report("could not set up rng device in the fdt");
1265             exit(1);
1266         }
1267     }
1268 
1269     QLIST_FOREACH(phb, &spapr->phbs, list) {
1270         ret = spapr_populate_pci_dt(phb, PHANDLE_XICP, fdt);
1271         if (ret < 0) {
1272             error_report("couldn't setup PCI devices in fdt");
1273             exit(1);
1274         }
1275     }
1276 
1277     /* cpus */
1278     spapr_populate_cpus_dt_node(fdt, spapr);
1279 
1280     if (smc->dr_lmb_enabled) {
1281         _FDT(spapr_drc_populate_dt(fdt, 0, NULL, SPAPR_DR_CONNECTOR_TYPE_LMB));
1282     }
1283 
1284     if (mc->has_hotpluggable_cpus) {
1285         int offset = fdt_path_offset(fdt, "/cpus");
1286         ret = spapr_drc_populate_dt(fdt, offset, NULL,
1287                                     SPAPR_DR_CONNECTOR_TYPE_CPU);
1288         if (ret < 0) {
1289             error_report("Couldn't set up CPU DR device tree properties");
1290             exit(1);
1291         }
1292     }
1293 
1294     /* /event-sources */
1295     spapr_dt_events(spapr, fdt);
1296 
1297     /* /rtas */
1298     spapr_dt_rtas(spapr, fdt);
1299 
1300     /* /chosen */
1301     spapr_dt_chosen(spapr, fdt);
1302 
1303     /* /hypervisor */
1304     if (kvm_enabled()) {
1305         spapr_dt_hypervisor(spapr, fdt);
1306     }
1307 
1308     /* Build memory reserve map */
1309     if (spapr->kernel_size) {
1310         _FDT((fdt_add_mem_rsv(fdt, KERNEL_LOAD_ADDR, spapr->kernel_size)));
1311     }
1312     if (spapr->initrd_size) {
1313         _FDT((fdt_add_mem_rsv(fdt, spapr->initrd_base, spapr->initrd_size)));
1314     }
1315 
1316     /* ibm,client-architecture-support updates */
1317     ret = spapr_dt_cas_updates(spapr, fdt, spapr->ov5_cas);
1318     if (ret < 0) {
1319         error_report("couldn't setup CAS properties fdt");
1320         exit(1);
1321     }
1322 
1323     return fdt;
1324 }
1325 
1326 static uint64_t translate_kernel_address(void *opaque, uint64_t addr)
1327 {
1328     return (addr & 0x0fffffff) + KERNEL_LOAD_ADDR;
1329 }
1330 
1331 static void emulate_spapr_hypercall(PPCVirtualHypervisor *vhyp,
1332                                     PowerPCCPU *cpu)
1333 {
1334     CPUPPCState *env = &cpu->env;
1335 
1336     /* The TCG path should also be holding the BQL at this point */
1337     g_assert(qemu_mutex_iothread_locked());
1338 
1339     if (msr_pr) {
1340         hcall_dprintf("Hypercall made with MSR[PR]=1\n");
1341         env->gpr[3] = H_PRIVILEGE;
1342     } else {
1343         env->gpr[3] = spapr_hypercall(cpu, env->gpr[3], &env->gpr[4]);
1344     }
1345 }
1346 
1347 static uint64_t spapr_get_patbe(PPCVirtualHypervisor *vhyp)
1348 {
1349     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1350 
1351     return spapr->patb_entry;
1352 }
1353 
1354 #define HPTE(_table, _i)   (void *)(((uint64_t *)(_table)) + ((_i) * 2))
1355 #define HPTE_VALID(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_VALID)
1356 #define HPTE_DIRTY(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_HPTE_DIRTY)
1357 #define CLEAN_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) &= tswap64(~HPTE64_V_HPTE_DIRTY))
1358 #define DIRTY_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) |= tswap64(HPTE64_V_HPTE_DIRTY))
1359 
1360 /*
1361  * Get the fd to access the kernel htab, re-opening it if necessary
1362  */
1363 static int get_htab_fd(sPAPRMachineState *spapr)
1364 {
1365     Error *local_err = NULL;
1366 
1367     if (spapr->htab_fd >= 0) {
1368         return spapr->htab_fd;
1369     }
1370 
1371     spapr->htab_fd = kvmppc_get_htab_fd(false, 0, &local_err);
1372     if (spapr->htab_fd < 0) {
1373         error_report_err(local_err);
1374     }
1375 
1376     return spapr->htab_fd;
1377 }
1378 
1379 void close_htab_fd(sPAPRMachineState *spapr)
1380 {
1381     if (spapr->htab_fd >= 0) {
1382         close(spapr->htab_fd);
1383     }
1384     spapr->htab_fd = -1;
1385 }
1386 
1387 static hwaddr spapr_hpt_mask(PPCVirtualHypervisor *vhyp)
1388 {
1389     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1390 
1391     return HTAB_SIZE(spapr) / HASH_PTEG_SIZE_64 - 1;
1392 }
1393 
1394 static target_ulong spapr_encode_hpt_for_kvm_pr(PPCVirtualHypervisor *vhyp)
1395 {
1396     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1397 
1398     assert(kvm_enabled());
1399 
1400     if (!spapr->htab) {
1401         return 0;
1402     }
1403 
1404     return (target_ulong)(uintptr_t)spapr->htab | (spapr->htab_shift - 18);
1405 }
1406 
1407 static const ppc_hash_pte64_t *spapr_map_hptes(PPCVirtualHypervisor *vhyp,
1408                                                 hwaddr ptex, int n)
1409 {
1410     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1411     hwaddr pte_offset = ptex * HASH_PTE_SIZE_64;
1412 
1413     if (!spapr->htab) {
1414         /*
1415          * HTAB is controlled by KVM. Fetch into temporary buffer
1416          */
1417         ppc_hash_pte64_t *hptes = g_malloc(n * HASH_PTE_SIZE_64);
1418         kvmppc_read_hptes(hptes, ptex, n);
1419         return hptes;
1420     }
1421 
1422     /*
1423      * HTAB is controlled by QEMU. Just point to the internally
1424      * accessible PTEG.
1425      */
1426     return (const ppc_hash_pte64_t *)(spapr->htab + pte_offset);
1427 }
1428 
1429 static void spapr_unmap_hptes(PPCVirtualHypervisor *vhyp,
1430                               const ppc_hash_pte64_t *hptes,
1431                               hwaddr ptex, int n)
1432 {
1433     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1434 
1435     if (!spapr->htab) {
1436         g_free((void *)hptes);
1437     }
1438 
1439     /* Nothing to do for qemu managed HPT */
1440 }
1441 
1442 static void spapr_store_hpte(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1443                              uint64_t pte0, uint64_t pte1)
1444 {
1445     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1446     hwaddr offset = ptex * HASH_PTE_SIZE_64;
1447 
1448     if (!spapr->htab) {
1449         kvmppc_write_hpte(ptex, pte0, pte1);
1450     } else {
1451         stq_p(spapr->htab + offset, pte0);
1452         stq_p(spapr->htab + offset + HASH_PTE_SIZE_64 / 2, pte1);
1453     }
1454 }
1455 
1456 int spapr_hpt_shift_for_ramsize(uint64_t ramsize)
1457 {
1458     int shift;
1459 
1460     /* We aim for a hash table of size 1/128 the size of RAM (rounded
1461      * up).  The PAPR recommendation is actually 1/64 of RAM size, but
1462      * that's much more than is needed for Linux guests */
1463     shift = ctz64(pow2ceil(ramsize)) - 7;
1464     shift = MAX(shift, 18); /* Minimum architected size */
1465     shift = MIN(shift, 46); /* Maximum architected size */
1466     return shift;
1467 }
1468 
1469 void spapr_free_hpt(sPAPRMachineState *spapr)
1470 {
1471     g_free(spapr->htab);
1472     spapr->htab = NULL;
1473     spapr->htab_shift = 0;
1474     close_htab_fd(spapr);
1475 }
1476 
1477 void spapr_reallocate_hpt(sPAPRMachineState *spapr, int shift,
1478                           Error **errp)
1479 {
1480     long rc;
1481 
1482     /* Clean up any HPT info from a previous boot */
1483     spapr_free_hpt(spapr);
1484 
1485     rc = kvmppc_reset_htab(shift);
1486     if (rc < 0) {
1487         /* kernel-side HPT needed, but couldn't allocate one */
1488         error_setg_errno(errp, errno,
1489                          "Failed to allocate KVM HPT of order %d (try smaller maxmem?)",
1490                          shift);
1491         /* This is almost certainly fatal, but if the caller really
1492          * wants to carry on with shift == 0, it's welcome to try */
1493     } else if (rc > 0) {
1494         /* kernel-side HPT allocated */
1495         if (rc != shift) {
1496             error_setg(errp,
1497                        "Requested order %d HPT, but kernel allocated order %ld (try smaller maxmem?)",
1498                        shift, rc);
1499         }
1500 
1501         spapr->htab_shift = shift;
1502         spapr->htab = NULL;
1503     } else {
1504         /* kernel-side HPT not needed, allocate in userspace instead */
1505         size_t size = 1ULL << shift;
1506         int i;
1507 
1508         spapr->htab = qemu_memalign(size, size);
1509         if (!spapr->htab) {
1510             error_setg_errno(errp, errno,
1511                              "Could not allocate HPT of order %d", shift);
1512             return;
1513         }
1514 
1515         memset(spapr->htab, 0, size);
1516         spapr->htab_shift = shift;
1517 
1518         for (i = 0; i < size / HASH_PTE_SIZE_64; i++) {
1519             DIRTY_HPTE(HPTE(spapr->htab, i));
1520         }
1521     }
1522     /* We're setting up a hash table, so that means we're not radix */
1523     spapr->patb_entry = 0;
1524 }
1525 
1526 void spapr_setup_hpt_and_vrma(sPAPRMachineState *spapr)
1527 {
1528     int hpt_shift;
1529 
1530     if ((spapr->resize_hpt == SPAPR_RESIZE_HPT_DISABLED)
1531         || (spapr->cas_reboot
1532             && !spapr_ovec_test(spapr->ov5_cas, OV5_HPT_RESIZE))) {
1533         hpt_shift = spapr_hpt_shift_for_ramsize(MACHINE(spapr)->maxram_size);
1534     } else {
1535         uint64_t current_ram_size;
1536 
1537         current_ram_size = MACHINE(spapr)->ram_size + get_plugged_memory_size();
1538         hpt_shift = spapr_hpt_shift_for_ramsize(current_ram_size);
1539     }
1540     spapr_reallocate_hpt(spapr, hpt_shift, &error_fatal);
1541 
1542     if (spapr->vrma_adjust) {
1543         spapr->rma_size = kvmppc_rma_size(spapr_node0_size(MACHINE(spapr)),
1544                                           spapr->htab_shift);
1545     }
1546 }
1547 
1548 static int spapr_reset_drcs(Object *child, void *opaque)
1549 {
1550     sPAPRDRConnector *drc =
1551         (sPAPRDRConnector *) object_dynamic_cast(child,
1552                                                  TYPE_SPAPR_DR_CONNECTOR);
1553 
1554     if (drc) {
1555         spapr_drc_reset(drc);
1556     }
1557 
1558     return 0;
1559 }
1560 
1561 static void spapr_machine_reset(void)
1562 {
1563     MachineState *machine = MACHINE(qdev_get_machine());
1564     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
1565     PowerPCCPU *first_ppc_cpu;
1566     uint32_t rtas_limit;
1567     hwaddr rtas_addr, fdt_addr;
1568     void *fdt;
1569     int rc;
1570 
1571     spapr_caps_apply(spapr);
1572 
1573     first_ppc_cpu = POWERPC_CPU(first_cpu);
1574     if (kvm_enabled() && kvmppc_has_cap_mmu_radix() &&
1575         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
1576                               spapr->max_compat_pvr)) {
1577         /* If using KVM with radix mode available, VCPUs can be started
1578          * without a HPT because KVM will start them in radix mode.
1579          * Set the GR bit in PATB so that we know there is no HPT. */
1580         spapr->patb_entry = PATBE1_GR;
1581     } else {
1582         spapr_setup_hpt_and_vrma(spapr);
1583     }
1584 
1585     /* if this reset wasn't generated by CAS, we should reset our
1586      * negotiated options and start from scratch */
1587     if (!spapr->cas_reboot) {
1588         spapr_ovec_cleanup(spapr->ov5_cas);
1589         spapr->ov5_cas = spapr_ovec_new();
1590 
1591         ppc_set_compat(first_ppc_cpu, spapr->max_compat_pvr, &error_fatal);
1592     }
1593 
1594     if (!SPAPR_MACHINE_GET_CLASS(spapr)->legacy_irq_allocation) {
1595         spapr_irq_msi_reset(spapr);
1596     }
1597 
1598     qemu_devices_reset();
1599 
1600     /* DRC reset may cause a device to be unplugged. This will cause troubles
1601      * if this device is used by another device (eg, a running vhost backend
1602      * will crash QEMU if the DIMM holding the vring goes away). To avoid such
1603      * situations, we reset DRCs after all devices have been reset.
1604      */
1605     object_child_foreach_recursive(object_get_root(), spapr_reset_drcs, NULL);
1606 
1607     spapr_clear_pending_events(spapr);
1608 
1609     /*
1610      * We place the device tree and RTAS just below either the top of the RMA,
1611      * or just below 2GB, whichever is lowere, so that it can be
1612      * processed with 32-bit real mode code if necessary
1613      */
1614     rtas_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR);
1615     rtas_addr = rtas_limit - RTAS_MAX_SIZE;
1616     fdt_addr = rtas_addr - FDT_MAX_SIZE;
1617 
1618     fdt = spapr_build_fdt(spapr, rtas_addr, spapr->rtas_size);
1619 
1620     spapr_load_rtas(spapr, fdt, rtas_addr);
1621 
1622     rc = fdt_pack(fdt);
1623 
1624     /* Should only fail if we've built a corrupted tree */
1625     assert(rc == 0);
1626 
1627     if (fdt_totalsize(fdt) > FDT_MAX_SIZE) {
1628         error_report("FDT too big ! 0x%x bytes (max is 0x%x)",
1629                      fdt_totalsize(fdt), FDT_MAX_SIZE);
1630         exit(1);
1631     }
1632 
1633     /* Load the fdt */
1634     qemu_fdt_dumpdtb(fdt, fdt_totalsize(fdt));
1635     cpu_physical_memory_write(fdt_addr, fdt, fdt_totalsize(fdt));
1636     g_free(fdt);
1637 
1638     /* Set up the entry state */
1639     spapr_cpu_set_entry_state(first_ppc_cpu, SPAPR_ENTRY_POINT, fdt_addr);
1640     first_ppc_cpu->env.gpr[5] = 0;
1641 
1642     spapr->cas_reboot = false;
1643 }
1644 
1645 static void spapr_create_nvram(sPAPRMachineState *spapr)
1646 {
1647     DeviceState *dev = qdev_create(&spapr->vio_bus->bus, "spapr-nvram");
1648     DriveInfo *dinfo = drive_get(IF_PFLASH, 0, 0);
1649 
1650     if (dinfo) {
1651         qdev_prop_set_drive(dev, "drive", blk_by_legacy_dinfo(dinfo),
1652                             &error_fatal);
1653     }
1654 
1655     qdev_init_nofail(dev);
1656 
1657     spapr->nvram = (struct sPAPRNVRAM *)dev;
1658 }
1659 
1660 static void spapr_rtc_create(sPAPRMachineState *spapr)
1661 {
1662     object_initialize(&spapr->rtc, sizeof(spapr->rtc), TYPE_SPAPR_RTC);
1663     object_property_add_child(OBJECT(spapr), "rtc", OBJECT(&spapr->rtc),
1664                               &error_fatal);
1665     object_property_set_bool(OBJECT(&spapr->rtc), true, "realized",
1666                               &error_fatal);
1667     object_property_add_alias(OBJECT(spapr), "rtc-time", OBJECT(&spapr->rtc),
1668                               "date", &error_fatal);
1669 }
1670 
1671 /* Returns whether we want to use VGA or not */
1672 static bool spapr_vga_init(PCIBus *pci_bus, Error **errp)
1673 {
1674     switch (vga_interface_type) {
1675     case VGA_NONE:
1676         return false;
1677     case VGA_DEVICE:
1678         return true;
1679     case VGA_STD:
1680     case VGA_VIRTIO:
1681         return pci_vga_init(pci_bus) != NULL;
1682     default:
1683         error_setg(errp,
1684                    "Unsupported VGA mode, only -vga std or -vga virtio is supported");
1685         return false;
1686     }
1687 }
1688 
1689 static int spapr_pre_load(void *opaque)
1690 {
1691     int rc;
1692 
1693     rc = spapr_caps_pre_load(opaque);
1694     if (rc) {
1695         return rc;
1696     }
1697 
1698     return 0;
1699 }
1700 
1701 static int spapr_post_load(void *opaque, int version_id)
1702 {
1703     sPAPRMachineState *spapr = (sPAPRMachineState *)opaque;
1704     int err = 0;
1705 
1706     err = spapr_caps_post_migration(spapr);
1707     if (err) {
1708         return err;
1709     }
1710 
1711     if (!object_dynamic_cast(OBJECT(spapr->ics), TYPE_ICS_KVM)) {
1712         CPUState *cs;
1713         CPU_FOREACH(cs) {
1714             PowerPCCPU *cpu = POWERPC_CPU(cs);
1715             icp_resend(ICP(cpu->intc));
1716         }
1717     }
1718 
1719     /* In earlier versions, there was no separate qdev for the PAPR
1720      * RTC, so the RTC offset was stored directly in sPAPREnvironment.
1721      * So when migrating from those versions, poke the incoming offset
1722      * value into the RTC device */
1723     if (version_id < 3) {
1724         err = spapr_rtc_import_offset(&spapr->rtc, spapr->rtc_offset);
1725     }
1726 
1727     if (kvm_enabled() && spapr->patb_entry) {
1728         PowerPCCPU *cpu = POWERPC_CPU(first_cpu);
1729         bool radix = !!(spapr->patb_entry & PATBE1_GR);
1730         bool gtse = !!(cpu->env.spr[SPR_LPCR] & LPCR_GTSE);
1731 
1732         err = kvmppc_configure_v3_mmu(cpu, radix, gtse, spapr->patb_entry);
1733         if (err) {
1734             error_report("Process table config unsupported by the host");
1735             return -EINVAL;
1736         }
1737     }
1738 
1739     return err;
1740 }
1741 
1742 static int spapr_pre_save(void *opaque)
1743 {
1744     int rc;
1745 
1746     rc = spapr_caps_pre_save(opaque);
1747     if (rc) {
1748         return rc;
1749     }
1750 
1751     return 0;
1752 }
1753 
1754 static bool version_before_3(void *opaque, int version_id)
1755 {
1756     return version_id < 3;
1757 }
1758 
1759 static bool spapr_pending_events_needed(void *opaque)
1760 {
1761     sPAPRMachineState *spapr = (sPAPRMachineState *)opaque;
1762     return !QTAILQ_EMPTY(&spapr->pending_events);
1763 }
1764 
1765 static const VMStateDescription vmstate_spapr_event_entry = {
1766     .name = "spapr_event_log_entry",
1767     .version_id = 1,
1768     .minimum_version_id = 1,
1769     .fields = (VMStateField[]) {
1770         VMSTATE_UINT32(summary, sPAPREventLogEntry),
1771         VMSTATE_UINT32(extended_length, sPAPREventLogEntry),
1772         VMSTATE_VBUFFER_ALLOC_UINT32(extended_log, sPAPREventLogEntry, 0,
1773                                      NULL, extended_length),
1774         VMSTATE_END_OF_LIST()
1775     },
1776 };
1777 
1778 static const VMStateDescription vmstate_spapr_pending_events = {
1779     .name = "spapr_pending_events",
1780     .version_id = 1,
1781     .minimum_version_id = 1,
1782     .needed = spapr_pending_events_needed,
1783     .fields = (VMStateField[]) {
1784         VMSTATE_QTAILQ_V(pending_events, sPAPRMachineState, 1,
1785                          vmstate_spapr_event_entry, sPAPREventLogEntry, next),
1786         VMSTATE_END_OF_LIST()
1787     },
1788 };
1789 
1790 static bool spapr_ov5_cas_needed(void *opaque)
1791 {
1792     sPAPRMachineState *spapr = opaque;
1793     sPAPROptionVector *ov5_mask = spapr_ovec_new();
1794     sPAPROptionVector *ov5_legacy = spapr_ovec_new();
1795     sPAPROptionVector *ov5_removed = spapr_ovec_new();
1796     bool cas_needed;
1797 
1798     /* Prior to the introduction of sPAPROptionVector, we had two option
1799      * vectors we dealt with: OV5_FORM1_AFFINITY, and OV5_DRCONF_MEMORY.
1800      * Both of these options encode machine topology into the device-tree
1801      * in such a way that the now-booted OS should still be able to interact
1802      * appropriately with QEMU regardless of what options were actually
1803      * negotiatied on the source side.
1804      *
1805      * As such, we can avoid migrating the CAS-negotiated options if these
1806      * are the only options available on the current machine/platform.
1807      * Since these are the only options available for pseries-2.7 and
1808      * earlier, this allows us to maintain old->new/new->old migration
1809      * compatibility.
1810      *
1811      * For QEMU 2.8+, there are additional CAS-negotiatable options available
1812      * via default pseries-2.8 machines and explicit command-line parameters.
1813      * Some of these options, like OV5_HP_EVT, *do* require QEMU to be aware
1814      * of the actual CAS-negotiated values to continue working properly. For
1815      * example, availability of memory unplug depends on knowing whether
1816      * OV5_HP_EVT was negotiated via CAS.
1817      *
1818      * Thus, for any cases where the set of available CAS-negotiatable
1819      * options extends beyond OV5_FORM1_AFFINITY and OV5_DRCONF_MEMORY, we
1820      * include the CAS-negotiated options in the migration stream, unless
1821      * if they affect boot time behaviour only.
1822      */
1823     spapr_ovec_set(ov5_mask, OV5_FORM1_AFFINITY);
1824     spapr_ovec_set(ov5_mask, OV5_DRCONF_MEMORY);
1825     spapr_ovec_set(ov5_mask, OV5_DRMEM_V2);
1826 
1827     /* spapr_ovec_diff returns true if bits were removed. we avoid using
1828      * the mask itself since in the future it's possible "legacy" bits may be
1829      * removed via machine options, which could generate a false positive
1830      * that breaks migration.
1831      */
1832     spapr_ovec_intersect(ov5_legacy, spapr->ov5, ov5_mask);
1833     cas_needed = spapr_ovec_diff(ov5_removed, spapr->ov5, ov5_legacy);
1834 
1835     spapr_ovec_cleanup(ov5_mask);
1836     spapr_ovec_cleanup(ov5_legacy);
1837     spapr_ovec_cleanup(ov5_removed);
1838 
1839     return cas_needed;
1840 }
1841 
1842 static const VMStateDescription vmstate_spapr_ov5_cas = {
1843     .name = "spapr_option_vector_ov5_cas",
1844     .version_id = 1,
1845     .minimum_version_id = 1,
1846     .needed = spapr_ov5_cas_needed,
1847     .fields = (VMStateField[]) {
1848         VMSTATE_STRUCT_POINTER_V(ov5_cas, sPAPRMachineState, 1,
1849                                  vmstate_spapr_ovec, sPAPROptionVector),
1850         VMSTATE_END_OF_LIST()
1851     },
1852 };
1853 
1854 static bool spapr_patb_entry_needed(void *opaque)
1855 {
1856     sPAPRMachineState *spapr = opaque;
1857 
1858     return !!spapr->patb_entry;
1859 }
1860 
1861 static const VMStateDescription vmstate_spapr_patb_entry = {
1862     .name = "spapr_patb_entry",
1863     .version_id = 1,
1864     .minimum_version_id = 1,
1865     .needed = spapr_patb_entry_needed,
1866     .fields = (VMStateField[]) {
1867         VMSTATE_UINT64(patb_entry, sPAPRMachineState),
1868         VMSTATE_END_OF_LIST()
1869     },
1870 };
1871 
1872 static bool spapr_irq_map_needed(void *opaque)
1873 {
1874     sPAPRMachineState *spapr = opaque;
1875 
1876     return spapr->irq_map && !bitmap_empty(spapr->irq_map, spapr->irq_map_nr);
1877 }
1878 
1879 static const VMStateDescription vmstate_spapr_irq_map = {
1880     .name = "spapr_irq_map",
1881     .version_id = 1,
1882     .minimum_version_id = 1,
1883     .needed = spapr_irq_map_needed,
1884     .fields = (VMStateField[]) {
1885         VMSTATE_BITMAP(irq_map, sPAPRMachineState, 0, irq_map_nr),
1886         VMSTATE_END_OF_LIST()
1887     },
1888 };
1889 
1890 static const VMStateDescription vmstate_spapr = {
1891     .name = "spapr",
1892     .version_id = 3,
1893     .minimum_version_id = 1,
1894     .pre_load = spapr_pre_load,
1895     .post_load = spapr_post_load,
1896     .pre_save = spapr_pre_save,
1897     .fields = (VMStateField[]) {
1898         /* used to be @next_irq */
1899         VMSTATE_UNUSED_BUFFER(version_before_3, 0, 4),
1900 
1901         /* RTC offset */
1902         VMSTATE_UINT64_TEST(rtc_offset, sPAPRMachineState, version_before_3),
1903 
1904         VMSTATE_PPC_TIMEBASE_V(tb, sPAPRMachineState, 2),
1905         VMSTATE_END_OF_LIST()
1906     },
1907     .subsections = (const VMStateDescription*[]) {
1908         &vmstate_spapr_ov5_cas,
1909         &vmstate_spapr_patb_entry,
1910         &vmstate_spapr_pending_events,
1911         &vmstate_spapr_cap_htm,
1912         &vmstate_spapr_cap_vsx,
1913         &vmstate_spapr_cap_dfp,
1914         &vmstate_spapr_cap_cfpc,
1915         &vmstate_spapr_cap_sbbc,
1916         &vmstate_spapr_cap_ibs,
1917         &vmstate_spapr_irq_map,
1918         NULL
1919     }
1920 };
1921 
1922 static int htab_save_setup(QEMUFile *f, void *opaque)
1923 {
1924     sPAPRMachineState *spapr = opaque;
1925 
1926     /* "Iteration" header */
1927     if (!spapr->htab_shift) {
1928         qemu_put_be32(f, -1);
1929     } else {
1930         qemu_put_be32(f, spapr->htab_shift);
1931     }
1932 
1933     if (spapr->htab) {
1934         spapr->htab_save_index = 0;
1935         spapr->htab_first_pass = true;
1936     } else {
1937         if (spapr->htab_shift) {
1938             assert(kvm_enabled());
1939         }
1940     }
1941 
1942 
1943     return 0;
1944 }
1945 
1946 static void htab_save_chunk(QEMUFile *f, sPAPRMachineState *spapr,
1947                             int chunkstart, int n_valid, int n_invalid)
1948 {
1949     qemu_put_be32(f, chunkstart);
1950     qemu_put_be16(f, n_valid);
1951     qemu_put_be16(f, n_invalid);
1952     qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
1953                     HASH_PTE_SIZE_64 * n_valid);
1954 }
1955 
1956 static void htab_save_end_marker(QEMUFile *f)
1957 {
1958     qemu_put_be32(f, 0);
1959     qemu_put_be16(f, 0);
1960     qemu_put_be16(f, 0);
1961 }
1962 
1963 static void htab_save_first_pass(QEMUFile *f, sPAPRMachineState *spapr,
1964                                  int64_t max_ns)
1965 {
1966     bool has_timeout = max_ns != -1;
1967     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
1968     int index = spapr->htab_save_index;
1969     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
1970 
1971     assert(spapr->htab_first_pass);
1972 
1973     do {
1974         int chunkstart;
1975 
1976         /* Consume invalid HPTEs */
1977         while ((index < htabslots)
1978                && !HPTE_VALID(HPTE(spapr->htab, index))) {
1979             CLEAN_HPTE(HPTE(spapr->htab, index));
1980             index++;
1981         }
1982 
1983         /* Consume valid HPTEs */
1984         chunkstart = index;
1985         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
1986                && HPTE_VALID(HPTE(spapr->htab, index))) {
1987             CLEAN_HPTE(HPTE(spapr->htab, index));
1988             index++;
1989         }
1990 
1991         if (index > chunkstart) {
1992             int n_valid = index - chunkstart;
1993 
1994             htab_save_chunk(f, spapr, chunkstart, n_valid, 0);
1995 
1996             if (has_timeout &&
1997                 (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
1998                 break;
1999             }
2000         }
2001     } while ((index < htabslots) && !qemu_file_rate_limit(f));
2002 
2003     if (index >= htabslots) {
2004         assert(index == htabslots);
2005         index = 0;
2006         spapr->htab_first_pass = false;
2007     }
2008     spapr->htab_save_index = index;
2009 }
2010 
2011 static int htab_save_later_pass(QEMUFile *f, sPAPRMachineState *spapr,
2012                                 int64_t max_ns)
2013 {
2014     bool final = max_ns < 0;
2015     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2016     int examined = 0, sent = 0;
2017     int index = spapr->htab_save_index;
2018     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2019 
2020     assert(!spapr->htab_first_pass);
2021 
2022     do {
2023         int chunkstart, invalidstart;
2024 
2025         /* Consume non-dirty HPTEs */
2026         while ((index < htabslots)
2027                && !HPTE_DIRTY(HPTE(spapr->htab, index))) {
2028             index++;
2029             examined++;
2030         }
2031 
2032         chunkstart = index;
2033         /* Consume valid dirty HPTEs */
2034         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2035                && HPTE_DIRTY(HPTE(spapr->htab, index))
2036                && HPTE_VALID(HPTE(spapr->htab, index))) {
2037             CLEAN_HPTE(HPTE(spapr->htab, index));
2038             index++;
2039             examined++;
2040         }
2041 
2042         invalidstart = index;
2043         /* Consume invalid dirty HPTEs */
2044         while ((index < htabslots) && (index - invalidstart < USHRT_MAX)
2045                && HPTE_DIRTY(HPTE(spapr->htab, index))
2046                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2047             CLEAN_HPTE(HPTE(spapr->htab, index));
2048             index++;
2049             examined++;
2050         }
2051 
2052         if (index > chunkstart) {
2053             int n_valid = invalidstart - chunkstart;
2054             int n_invalid = index - invalidstart;
2055 
2056             htab_save_chunk(f, spapr, chunkstart, n_valid, n_invalid);
2057             sent += index - chunkstart;
2058 
2059             if (!final && (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2060                 break;
2061             }
2062         }
2063 
2064         if (examined >= htabslots) {
2065             break;
2066         }
2067 
2068         if (index >= htabslots) {
2069             assert(index == htabslots);
2070             index = 0;
2071         }
2072     } while ((examined < htabslots) && (!qemu_file_rate_limit(f) || final));
2073 
2074     if (index >= htabslots) {
2075         assert(index == htabslots);
2076         index = 0;
2077     }
2078 
2079     spapr->htab_save_index = index;
2080 
2081     return (examined >= htabslots) && (sent == 0) ? 1 : 0;
2082 }
2083 
2084 #define MAX_ITERATION_NS    5000000 /* 5 ms */
2085 #define MAX_KVM_BUF_SIZE    2048
2086 
2087 static int htab_save_iterate(QEMUFile *f, void *opaque)
2088 {
2089     sPAPRMachineState *spapr = opaque;
2090     int fd;
2091     int rc = 0;
2092 
2093     /* Iteration header */
2094     if (!spapr->htab_shift) {
2095         qemu_put_be32(f, -1);
2096         return 1;
2097     } else {
2098         qemu_put_be32(f, 0);
2099     }
2100 
2101     if (!spapr->htab) {
2102         assert(kvm_enabled());
2103 
2104         fd = get_htab_fd(spapr);
2105         if (fd < 0) {
2106             return fd;
2107         }
2108 
2109         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, MAX_ITERATION_NS);
2110         if (rc < 0) {
2111             return rc;
2112         }
2113     } else  if (spapr->htab_first_pass) {
2114         htab_save_first_pass(f, spapr, MAX_ITERATION_NS);
2115     } else {
2116         rc = htab_save_later_pass(f, spapr, MAX_ITERATION_NS);
2117     }
2118 
2119     htab_save_end_marker(f);
2120 
2121     return rc;
2122 }
2123 
2124 static int htab_save_complete(QEMUFile *f, void *opaque)
2125 {
2126     sPAPRMachineState *spapr = opaque;
2127     int fd;
2128 
2129     /* Iteration header */
2130     if (!spapr->htab_shift) {
2131         qemu_put_be32(f, -1);
2132         return 0;
2133     } else {
2134         qemu_put_be32(f, 0);
2135     }
2136 
2137     if (!spapr->htab) {
2138         int rc;
2139 
2140         assert(kvm_enabled());
2141 
2142         fd = get_htab_fd(spapr);
2143         if (fd < 0) {
2144             return fd;
2145         }
2146 
2147         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, -1);
2148         if (rc < 0) {
2149             return rc;
2150         }
2151     } else {
2152         if (spapr->htab_first_pass) {
2153             htab_save_first_pass(f, spapr, -1);
2154         }
2155         htab_save_later_pass(f, spapr, -1);
2156     }
2157 
2158     /* End marker */
2159     htab_save_end_marker(f);
2160 
2161     return 0;
2162 }
2163 
2164 static int htab_load(QEMUFile *f, void *opaque, int version_id)
2165 {
2166     sPAPRMachineState *spapr = opaque;
2167     uint32_t section_hdr;
2168     int fd = -1;
2169     Error *local_err = NULL;
2170 
2171     if (version_id < 1 || version_id > 1) {
2172         error_report("htab_load() bad version");
2173         return -EINVAL;
2174     }
2175 
2176     section_hdr = qemu_get_be32(f);
2177 
2178     if (section_hdr == -1) {
2179         spapr_free_hpt(spapr);
2180         return 0;
2181     }
2182 
2183     if (section_hdr) {
2184         /* First section gives the htab size */
2185         spapr_reallocate_hpt(spapr, section_hdr, &local_err);
2186         if (local_err) {
2187             error_report_err(local_err);
2188             return -EINVAL;
2189         }
2190         return 0;
2191     }
2192 
2193     if (!spapr->htab) {
2194         assert(kvm_enabled());
2195 
2196         fd = kvmppc_get_htab_fd(true, 0, &local_err);
2197         if (fd < 0) {
2198             error_report_err(local_err);
2199             return fd;
2200         }
2201     }
2202 
2203     while (true) {
2204         uint32_t index;
2205         uint16_t n_valid, n_invalid;
2206 
2207         index = qemu_get_be32(f);
2208         n_valid = qemu_get_be16(f);
2209         n_invalid = qemu_get_be16(f);
2210 
2211         if ((index == 0) && (n_valid == 0) && (n_invalid == 0)) {
2212             /* End of Stream */
2213             break;
2214         }
2215 
2216         if ((index + n_valid + n_invalid) >
2217             (HTAB_SIZE(spapr) / HASH_PTE_SIZE_64)) {
2218             /* Bad index in stream */
2219             error_report(
2220                 "htab_load() bad index %d (%hd+%hd entries) in htab stream (htab_shift=%d)",
2221                 index, n_valid, n_invalid, spapr->htab_shift);
2222             return -EINVAL;
2223         }
2224 
2225         if (spapr->htab) {
2226             if (n_valid) {
2227                 qemu_get_buffer(f, HPTE(spapr->htab, index),
2228                                 HASH_PTE_SIZE_64 * n_valid);
2229             }
2230             if (n_invalid) {
2231                 memset(HPTE(spapr->htab, index + n_valid), 0,
2232                        HASH_PTE_SIZE_64 * n_invalid);
2233             }
2234         } else {
2235             int rc;
2236 
2237             assert(fd >= 0);
2238 
2239             rc = kvmppc_load_htab_chunk(f, fd, index, n_valid, n_invalid);
2240             if (rc < 0) {
2241                 return rc;
2242             }
2243         }
2244     }
2245 
2246     if (!spapr->htab) {
2247         assert(fd >= 0);
2248         close(fd);
2249     }
2250 
2251     return 0;
2252 }
2253 
2254 static void htab_save_cleanup(void *opaque)
2255 {
2256     sPAPRMachineState *spapr = opaque;
2257 
2258     close_htab_fd(spapr);
2259 }
2260 
2261 static SaveVMHandlers savevm_htab_handlers = {
2262     .save_setup = htab_save_setup,
2263     .save_live_iterate = htab_save_iterate,
2264     .save_live_complete_precopy = htab_save_complete,
2265     .save_cleanup = htab_save_cleanup,
2266     .load_state = htab_load,
2267 };
2268 
2269 static void spapr_boot_set(void *opaque, const char *boot_device,
2270                            Error **errp)
2271 {
2272     MachineState *machine = MACHINE(opaque);
2273     machine->boot_order = g_strdup(boot_device);
2274 }
2275 
2276 static void spapr_create_lmb_dr_connectors(sPAPRMachineState *spapr)
2277 {
2278     MachineState *machine = MACHINE(spapr);
2279     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
2280     uint32_t nr_lmbs = (machine->maxram_size - machine->ram_size)/lmb_size;
2281     int i;
2282 
2283     for (i = 0; i < nr_lmbs; i++) {
2284         uint64_t addr;
2285 
2286         addr = i * lmb_size + machine->device_memory->base;
2287         spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_LMB,
2288                                addr / lmb_size);
2289     }
2290 }
2291 
2292 /*
2293  * If RAM size, maxmem size and individual node mem sizes aren't aligned
2294  * to SPAPR_MEMORY_BLOCK_SIZE(256MB), then refuse to start the guest
2295  * since we can't support such unaligned sizes with DRCONF_MEMORY.
2296  */
2297 static void spapr_validate_node_memory(MachineState *machine, Error **errp)
2298 {
2299     int i;
2300 
2301     if (machine->ram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2302         error_setg(errp, "Memory size 0x" RAM_ADDR_FMT
2303                    " is not aligned to %" PRIu64 " MiB",
2304                    machine->ram_size,
2305                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2306         return;
2307     }
2308 
2309     if (machine->maxram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2310         error_setg(errp, "Maximum memory size 0x" RAM_ADDR_FMT
2311                    " is not aligned to %" PRIu64 " MiB",
2312                    machine->ram_size,
2313                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2314         return;
2315     }
2316 
2317     for (i = 0; i < nb_numa_nodes; i++) {
2318         if (numa_info[i].node_mem % SPAPR_MEMORY_BLOCK_SIZE) {
2319             error_setg(errp,
2320                        "Node %d memory size 0x%" PRIx64
2321                        " is not aligned to %" PRIu64 " MiB",
2322                        i, numa_info[i].node_mem,
2323                        SPAPR_MEMORY_BLOCK_SIZE / MiB);
2324             return;
2325         }
2326     }
2327 }
2328 
2329 /* find cpu slot in machine->possible_cpus by core_id */
2330 static CPUArchId *spapr_find_cpu_slot(MachineState *ms, uint32_t id, int *idx)
2331 {
2332     int index = id / smp_threads;
2333 
2334     if (index >= ms->possible_cpus->len) {
2335         return NULL;
2336     }
2337     if (idx) {
2338         *idx = index;
2339     }
2340     return &ms->possible_cpus->cpus[index];
2341 }
2342 
2343 static void spapr_set_vsmt_mode(sPAPRMachineState *spapr, Error **errp)
2344 {
2345     Error *local_err = NULL;
2346     bool vsmt_user = !!spapr->vsmt;
2347     int kvm_smt = kvmppc_smt_threads();
2348     int ret;
2349 
2350     if (!kvm_enabled() && (smp_threads > 1)) {
2351         error_setg(&local_err, "TCG cannot support more than 1 thread/core "
2352                      "on a pseries machine");
2353         goto out;
2354     }
2355     if (!is_power_of_2(smp_threads)) {
2356         error_setg(&local_err, "Cannot support %d threads/core on a pseries "
2357                      "machine because it must be a power of 2", smp_threads);
2358         goto out;
2359     }
2360 
2361     /* Detemine the VSMT mode to use: */
2362     if (vsmt_user) {
2363         if (spapr->vsmt < smp_threads) {
2364             error_setg(&local_err, "Cannot support VSMT mode %d"
2365                          " because it must be >= threads/core (%d)",
2366                          spapr->vsmt, smp_threads);
2367             goto out;
2368         }
2369         /* In this case, spapr->vsmt has been set by the command line */
2370     } else {
2371         /*
2372          * Default VSMT value is tricky, because we need it to be as
2373          * consistent as possible (for migration), but this requires
2374          * changing it for at least some existing cases.  We pick 8 as
2375          * the value that we'd get with KVM on POWER8, the
2376          * overwhelmingly common case in production systems.
2377          */
2378         spapr->vsmt = MAX(8, smp_threads);
2379     }
2380 
2381     /* KVM: If necessary, set the SMT mode: */
2382     if (kvm_enabled() && (spapr->vsmt != kvm_smt)) {
2383         ret = kvmppc_set_smt_threads(spapr->vsmt);
2384         if (ret) {
2385             /* Looks like KVM isn't able to change VSMT mode */
2386             error_setg(&local_err,
2387                        "Failed to set KVM's VSMT mode to %d (errno %d)",
2388                        spapr->vsmt, ret);
2389             /* We can live with that if the default one is big enough
2390              * for the number of threads, and a submultiple of the one
2391              * we want.  In this case we'll waste some vcpu ids, but
2392              * behaviour will be correct */
2393             if ((kvm_smt >= smp_threads) && ((spapr->vsmt % kvm_smt) == 0)) {
2394                 warn_report_err(local_err);
2395                 local_err = NULL;
2396                 goto out;
2397             } else {
2398                 if (!vsmt_user) {
2399                     error_append_hint(&local_err,
2400                                       "On PPC, a VM with %d threads/core"
2401                                       " on a host with %d threads/core"
2402                                       " requires the use of VSMT mode %d.\n",
2403                                       smp_threads, kvm_smt, spapr->vsmt);
2404                 }
2405                 kvmppc_hint_smt_possible(&local_err);
2406                 goto out;
2407             }
2408         }
2409     }
2410     /* else TCG: nothing to do currently */
2411 out:
2412     error_propagate(errp, local_err);
2413 }
2414 
2415 static void spapr_init_cpus(sPAPRMachineState *spapr)
2416 {
2417     MachineState *machine = MACHINE(spapr);
2418     MachineClass *mc = MACHINE_GET_CLASS(machine);
2419     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2420     const char *type = spapr_get_cpu_core_type(machine->cpu_type);
2421     const CPUArchIdList *possible_cpus;
2422     int boot_cores_nr = smp_cpus / smp_threads;
2423     int i;
2424 
2425     possible_cpus = mc->possible_cpu_arch_ids(machine);
2426     if (mc->has_hotpluggable_cpus) {
2427         if (smp_cpus % smp_threads) {
2428             error_report("smp_cpus (%u) must be multiple of threads (%u)",
2429                          smp_cpus, smp_threads);
2430             exit(1);
2431         }
2432         if (max_cpus % smp_threads) {
2433             error_report("max_cpus (%u) must be multiple of threads (%u)",
2434                          max_cpus, smp_threads);
2435             exit(1);
2436         }
2437     } else {
2438         if (max_cpus != smp_cpus) {
2439             error_report("This machine version does not support CPU hotplug");
2440             exit(1);
2441         }
2442         boot_cores_nr = possible_cpus->len;
2443     }
2444 
2445     /* VSMT must be set in order to be able to compute VCPU ids, ie to
2446      * call xics_max_server_number() or spapr_vcpu_id().
2447      */
2448     spapr_set_vsmt_mode(spapr, &error_fatal);
2449 
2450     if (smc->pre_2_10_has_unused_icps) {
2451         int i;
2452 
2453         for (i = 0; i < xics_max_server_number(spapr); i++) {
2454             /* Dummy entries get deregistered when real ICPState objects
2455              * are registered during CPU core hotplug.
2456              */
2457             pre_2_10_vmstate_register_dummy_icp(i);
2458         }
2459     }
2460 
2461     for (i = 0; i < possible_cpus->len; i++) {
2462         int core_id = i * smp_threads;
2463 
2464         if (mc->has_hotpluggable_cpus) {
2465             spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_CPU,
2466                                    spapr_vcpu_id(spapr, core_id));
2467         }
2468 
2469         if (i < boot_cores_nr) {
2470             Object *core  = object_new(type);
2471             int nr_threads = smp_threads;
2472 
2473             /* Handle the partially filled core for older machine types */
2474             if ((i + 1) * smp_threads >= smp_cpus) {
2475                 nr_threads = smp_cpus - i * smp_threads;
2476             }
2477 
2478             object_property_set_int(core, nr_threads, "nr-threads",
2479                                     &error_fatal);
2480             object_property_set_int(core, core_id, CPU_CORE_PROP_CORE_ID,
2481                                     &error_fatal);
2482             object_property_set_bool(core, true, "realized", &error_fatal);
2483         }
2484     }
2485 }
2486 
2487 /* pSeries LPAR / sPAPR hardware init */
2488 static void spapr_machine_init(MachineState *machine)
2489 {
2490     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
2491     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2492     const char *kernel_filename = machine->kernel_filename;
2493     const char *initrd_filename = machine->initrd_filename;
2494     PCIHostState *phb;
2495     int i;
2496     MemoryRegion *sysmem = get_system_memory();
2497     MemoryRegion *ram = g_new(MemoryRegion, 1);
2498     hwaddr node0_size = spapr_node0_size(machine);
2499     long load_limit, fw_size;
2500     char *filename;
2501     Error *resize_hpt_err = NULL;
2502 
2503     msi_nonbroken = true;
2504 
2505     QLIST_INIT(&spapr->phbs);
2506     QTAILQ_INIT(&spapr->pending_dimm_unplugs);
2507 
2508     /* Determine capabilities to run with */
2509     spapr_caps_init(spapr);
2510 
2511     kvmppc_check_papr_resize_hpt(&resize_hpt_err);
2512     if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DEFAULT) {
2513         /*
2514          * If the user explicitly requested a mode we should either
2515          * supply it, or fail completely (which we do below).  But if
2516          * it's not set explicitly, we reset our mode to something
2517          * that works
2518          */
2519         if (resize_hpt_err) {
2520             spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
2521             error_free(resize_hpt_err);
2522             resize_hpt_err = NULL;
2523         } else {
2524             spapr->resize_hpt = smc->resize_hpt_default;
2525         }
2526     }
2527 
2528     assert(spapr->resize_hpt != SPAPR_RESIZE_HPT_DEFAULT);
2529 
2530     if ((spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) && resize_hpt_err) {
2531         /*
2532          * User requested HPT resize, but this host can't supply it.  Bail out
2533          */
2534         error_report_err(resize_hpt_err);
2535         exit(1);
2536     }
2537 
2538     spapr->rma_size = node0_size;
2539 
2540     /* With KVM, we don't actually know whether KVM supports an
2541      * unbounded RMA (PR KVM) or is limited by the hash table size
2542      * (HV KVM using VRMA), so we always assume the latter
2543      *
2544      * In that case, we also limit the initial allocations for RTAS
2545      * etc... to 256M since we have no way to know what the VRMA size
2546      * is going to be as it depends on the size of the hash table
2547      * which isn't determined yet.
2548      */
2549     if (kvm_enabled()) {
2550         spapr->vrma_adjust = 1;
2551         spapr->rma_size = MIN(spapr->rma_size, 0x10000000);
2552     }
2553 
2554     /* Actually we don't support unbounded RMA anymore since we added
2555      * proper emulation of HV mode. The max we can get is 16G which
2556      * also happens to be what we configure for PAPR mode so make sure
2557      * we don't do anything bigger than that
2558      */
2559     spapr->rma_size = MIN(spapr->rma_size, 0x400000000ull);
2560 
2561     if (spapr->rma_size > node0_size) {
2562         error_report("Numa node 0 has to span the RMA (%#08"HWADDR_PRIx")",
2563                      spapr->rma_size);
2564         exit(1);
2565     }
2566 
2567     /* Setup a load limit for the ramdisk leaving room for SLOF and FDT */
2568     load_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR) - FW_OVERHEAD;
2569 
2570     /* Set up Interrupt Controller before we create the VCPUs */
2571     smc->irq->init(spapr, &error_fatal);
2572 
2573     /* Set up containers for ibm,client-architecture-support negotiated options
2574      */
2575     spapr->ov5 = spapr_ovec_new();
2576     spapr->ov5_cas = spapr_ovec_new();
2577 
2578     if (smc->dr_lmb_enabled) {
2579         spapr_ovec_set(spapr->ov5, OV5_DRCONF_MEMORY);
2580         spapr_validate_node_memory(machine, &error_fatal);
2581     }
2582 
2583     spapr_ovec_set(spapr->ov5, OV5_FORM1_AFFINITY);
2584 
2585     /* advertise support for dedicated HP event source to guests */
2586     if (spapr->use_hotplug_event_source) {
2587         spapr_ovec_set(spapr->ov5, OV5_HP_EVT);
2588     }
2589 
2590     /* advertise support for HPT resizing */
2591     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
2592         spapr_ovec_set(spapr->ov5, OV5_HPT_RESIZE);
2593     }
2594 
2595     /* advertise support for ibm,dyamic-memory-v2 */
2596     spapr_ovec_set(spapr->ov5, OV5_DRMEM_V2);
2597 
2598     /* init CPUs */
2599     spapr_init_cpus(spapr);
2600 
2601     if ((!kvm_enabled() || kvmppc_has_cap_mmu_radix()) &&
2602         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
2603                               spapr->max_compat_pvr)) {
2604         /* KVM and TCG always allow GTSE with radix... */
2605         spapr_ovec_set(spapr->ov5, OV5_MMU_RADIX_GTSE);
2606     }
2607     /* ... but not with hash (currently). */
2608 
2609     if (kvm_enabled()) {
2610         /* Enable H_LOGICAL_CI_* so SLOF can talk to in-kernel devices */
2611         kvmppc_enable_logical_ci_hcalls();
2612         kvmppc_enable_set_mode_hcall();
2613 
2614         /* H_CLEAR_MOD/_REF are mandatory in PAPR, but off by default */
2615         kvmppc_enable_clear_ref_mod_hcalls();
2616     }
2617 
2618     /* allocate RAM */
2619     memory_region_allocate_system_memory(ram, NULL, "ppc_spapr.ram",
2620                                          machine->ram_size);
2621     memory_region_add_subregion(sysmem, 0, ram);
2622 
2623     /* always allocate the device memory information */
2624     machine->device_memory = g_malloc0(sizeof(*machine->device_memory));
2625 
2626     /* initialize hotplug memory address space */
2627     if (machine->ram_size < machine->maxram_size) {
2628         ram_addr_t device_mem_size = machine->maxram_size - machine->ram_size;
2629         /*
2630          * Limit the number of hotpluggable memory slots to half the number
2631          * slots that KVM supports, leaving the other half for PCI and other
2632          * devices. However ensure that number of slots doesn't drop below 32.
2633          */
2634         int max_memslots = kvm_enabled() ? kvm_get_max_memslots() / 2 :
2635                            SPAPR_MAX_RAM_SLOTS;
2636 
2637         if (max_memslots < SPAPR_MAX_RAM_SLOTS) {
2638             max_memslots = SPAPR_MAX_RAM_SLOTS;
2639         }
2640         if (machine->ram_slots > max_memslots) {
2641             error_report("Specified number of memory slots %"
2642                          PRIu64" exceeds max supported %d",
2643                          machine->ram_slots, max_memslots);
2644             exit(1);
2645         }
2646 
2647         machine->device_memory->base = ROUND_UP(machine->ram_size,
2648                                                 SPAPR_DEVICE_MEM_ALIGN);
2649         memory_region_init(&machine->device_memory->mr, OBJECT(spapr),
2650                            "device-memory", device_mem_size);
2651         memory_region_add_subregion(sysmem, machine->device_memory->base,
2652                                     &machine->device_memory->mr);
2653     }
2654 
2655     if (smc->dr_lmb_enabled) {
2656         spapr_create_lmb_dr_connectors(spapr);
2657     }
2658 
2659     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, "spapr-rtas.bin");
2660     if (!filename) {
2661         error_report("Could not find LPAR rtas '%s'", "spapr-rtas.bin");
2662         exit(1);
2663     }
2664     spapr->rtas_size = get_image_size(filename);
2665     if (spapr->rtas_size < 0) {
2666         error_report("Could not get size of LPAR rtas '%s'", filename);
2667         exit(1);
2668     }
2669     spapr->rtas_blob = g_malloc(spapr->rtas_size);
2670     if (load_image_size(filename, spapr->rtas_blob, spapr->rtas_size) < 0) {
2671         error_report("Could not load LPAR rtas '%s'", filename);
2672         exit(1);
2673     }
2674     if (spapr->rtas_size > RTAS_MAX_SIZE) {
2675         error_report("RTAS too big ! 0x%zx bytes (max is 0x%x)",
2676                      (size_t)spapr->rtas_size, RTAS_MAX_SIZE);
2677         exit(1);
2678     }
2679     g_free(filename);
2680 
2681     /* Set up RTAS event infrastructure */
2682     spapr_events_init(spapr);
2683 
2684     /* Set up the RTC RTAS interfaces */
2685     spapr_rtc_create(spapr);
2686 
2687     /* Set up VIO bus */
2688     spapr->vio_bus = spapr_vio_bus_init();
2689 
2690     for (i = 0; i < serial_max_hds(); i++) {
2691         if (serial_hd(i)) {
2692             spapr_vty_create(spapr->vio_bus, serial_hd(i));
2693         }
2694     }
2695 
2696     /* We always have at least the nvram device on VIO */
2697     spapr_create_nvram(spapr);
2698 
2699     /* Set up PCI */
2700     spapr_pci_rtas_init();
2701 
2702     phb = spapr_create_phb(spapr, 0);
2703 
2704     for (i = 0; i < nb_nics; i++) {
2705         NICInfo *nd = &nd_table[i];
2706 
2707         if (!nd->model) {
2708             nd->model = g_strdup("spapr-vlan");
2709         }
2710 
2711         if (g_str_equal(nd->model, "spapr-vlan") ||
2712             g_str_equal(nd->model, "ibmveth")) {
2713             spapr_vlan_create(spapr->vio_bus, nd);
2714         } else {
2715             pci_nic_init_nofail(&nd_table[i], phb->bus, nd->model, NULL);
2716         }
2717     }
2718 
2719     for (i = 0; i <= drive_get_max_bus(IF_SCSI); i++) {
2720         spapr_vscsi_create(spapr->vio_bus);
2721     }
2722 
2723     /* Graphics */
2724     if (spapr_vga_init(phb->bus, &error_fatal)) {
2725         spapr->has_graphics = true;
2726         machine->usb |= defaults_enabled() && !machine->usb_disabled;
2727     }
2728 
2729     if (machine->usb) {
2730         if (smc->use_ohci_by_default) {
2731             pci_create_simple(phb->bus, -1, "pci-ohci");
2732         } else {
2733             pci_create_simple(phb->bus, -1, "nec-usb-xhci");
2734         }
2735 
2736         if (spapr->has_graphics) {
2737             USBBus *usb_bus = usb_bus_find(-1);
2738 
2739             usb_create_simple(usb_bus, "usb-kbd");
2740             usb_create_simple(usb_bus, "usb-mouse");
2741         }
2742     }
2743 
2744     if (spapr->rma_size < (MIN_RMA_SLOF * MiB)) {
2745         error_report(
2746             "pSeries SLOF firmware requires >= %ldM guest RMA (Real Mode Area memory)",
2747             MIN_RMA_SLOF);
2748         exit(1);
2749     }
2750 
2751     if (kernel_filename) {
2752         uint64_t lowaddr = 0;
2753 
2754         spapr->kernel_size = load_elf(kernel_filename, translate_kernel_address,
2755                                       NULL, NULL, &lowaddr, NULL, 1,
2756                                       PPC_ELF_MACHINE, 0, 0);
2757         if (spapr->kernel_size == ELF_LOAD_WRONG_ENDIAN) {
2758             spapr->kernel_size = load_elf(kernel_filename,
2759                                           translate_kernel_address, NULL, NULL,
2760                                           &lowaddr, NULL, 0, PPC_ELF_MACHINE,
2761                                           0, 0);
2762             spapr->kernel_le = spapr->kernel_size > 0;
2763         }
2764         if (spapr->kernel_size < 0) {
2765             error_report("error loading %s: %s", kernel_filename,
2766                          load_elf_strerror(spapr->kernel_size));
2767             exit(1);
2768         }
2769 
2770         /* load initrd */
2771         if (initrd_filename) {
2772             /* Try to locate the initrd in the gap between the kernel
2773              * and the firmware. Add a bit of space just in case
2774              */
2775             spapr->initrd_base = (KERNEL_LOAD_ADDR + spapr->kernel_size
2776                                   + 0x1ffff) & ~0xffff;
2777             spapr->initrd_size = load_image_targphys(initrd_filename,
2778                                                      spapr->initrd_base,
2779                                                      load_limit
2780                                                      - spapr->initrd_base);
2781             if (spapr->initrd_size < 0) {
2782                 error_report("could not load initial ram disk '%s'",
2783                              initrd_filename);
2784                 exit(1);
2785             }
2786         }
2787     }
2788 
2789     if (bios_name == NULL) {
2790         bios_name = FW_FILE_NAME;
2791     }
2792     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
2793     if (!filename) {
2794         error_report("Could not find LPAR firmware '%s'", bios_name);
2795         exit(1);
2796     }
2797     fw_size = load_image_targphys(filename, 0, FW_MAX_SIZE);
2798     if (fw_size <= 0) {
2799         error_report("Could not load LPAR firmware '%s'", filename);
2800         exit(1);
2801     }
2802     g_free(filename);
2803 
2804     /* FIXME: Should register things through the MachineState's qdev
2805      * interface, this is a legacy from the sPAPREnvironment structure
2806      * which predated MachineState but had a similar function */
2807     vmstate_register(NULL, 0, &vmstate_spapr, spapr);
2808     register_savevm_live(NULL, "spapr/htab", -1, 1,
2809                          &savevm_htab_handlers, spapr);
2810 
2811     qemu_register_boot_set(spapr_boot_set, spapr);
2812 
2813     if (kvm_enabled()) {
2814         /* to stop and start vmclock */
2815         qemu_add_vm_change_state_handler(cpu_ppc_clock_vm_state_change,
2816                                          &spapr->tb);
2817 
2818         kvmppc_spapr_enable_inkernel_multitce();
2819     }
2820 }
2821 
2822 static int spapr_kvm_type(const char *vm_type)
2823 {
2824     if (!vm_type) {
2825         return 0;
2826     }
2827 
2828     if (!strcmp(vm_type, "HV")) {
2829         return 1;
2830     }
2831 
2832     if (!strcmp(vm_type, "PR")) {
2833         return 2;
2834     }
2835 
2836     error_report("Unknown kvm-type specified '%s'", vm_type);
2837     exit(1);
2838 }
2839 
2840 /*
2841  * Implementation of an interface to adjust firmware path
2842  * for the bootindex property handling.
2843  */
2844 static char *spapr_get_fw_dev_path(FWPathProvider *p, BusState *bus,
2845                                    DeviceState *dev)
2846 {
2847 #define CAST(type, obj, name) \
2848     ((type *)object_dynamic_cast(OBJECT(obj), (name)))
2849     SCSIDevice *d = CAST(SCSIDevice,  dev, TYPE_SCSI_DEVICE);
2850     sPAPRPHBState *phb = CAST(sPAPRPHBState, dev, TYPE_SPAPR_PCI_HOST_BRIDGE);
2851     VHostSCSICommon *vsc = CAST(VHostSCSICommon, dev, TYPE_VHOST_SCSI_COMMON);
2852 
2853     if (d) {
2854         void *spapr = CAST(void, bus->parent, "spapr-vscsi");
2855         VirtIOSCSI *virtio = CAST(VirtIOSCSI, bus->parent, TYPE_VIRTIO_SCSI);
2856         USBDevice *usb = CAST(USBDevice, bus->parent, TYPE_USB_DEVICE);
2857 
2858         if (spapr) {
2859             /*
2860              * Replace "channel@0/disk@0,0" with "disk@8000000000000000":
2861              * We use SRP luns of the form 8000 | (bus << 8) | (id << 5) | lun
2862              * in the top 16 bits of the 64-bit LUN
2863              */
2864             unsigned id = 0x8000 | (d->id << 8) | d->lun;
2865             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
2866                                    (uint64_t)id << 48);
2867         } else if (virtio) {
2868             /*
2869              * We use SRP luns of the form 01000000 | (target << 8) | lun
2870              * in the top 32 bits of the 64-bit LUN
2871              * Note: the quote above is from SLOF and it is wrong,
2872              * the actual binding is:
2873              * swap 0100 or 10 << or 20 << ( target lun-id -- srplun )
2874              */
2875             unsigned id = 0x1000000 | (d->id << 16) | d->lun;
2876             if (d->lun >= 256) {
2877                 /* Use the LUN "flat space addressing method" */
2878                 id |= 0x4000;
2879             }
2880             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
2881                                    (uint64_t)id << 32);
2882         } else if (usb) {
2883             /*
2884              * We use SRP luns of the form 01000000 | (usb-port << 16) | lun
2885              * in the top 32 bits of the 64-bit LUN
2886              */
2887             unsigned usb_port = atoi(usb->port->path);
2888             unsigned id = 0x1000000 | (usb_port << 16) | d->lun;
2889             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
2890                                    (uint64_t)id << 32);
2891         }
2892     }
2893 
2894     /*
2895      * SLOF probes the USB devices, and if it recognizes that the device is a
2896      * storage device, it changes its name to "storage" instead of "usb-host",
2897      * and additionally adds a child node for the SCSI LUN, so the correct
2898      * boot path in SLOF is something like .../storage@1/disk@xxx" instead.
2899      */
2900     if (strcmp("usb-host", qdev_fw_name(dev)) == 0) {
2901         USBDevice *usbdev = CAST(USBDevice, dev, TYPE_USB_DEVICE);
2902         if (usb_host_dev_is_scsi_storage(usbdev)) {
2903             return g_strdup_printf("storage@%s/disk", usbdev->port->path);
2904         }
2905     }
2906 
2907     if (phb) {
2908         /* Replace "pci" with "pci@800000020000000" */
2909         return g_strdup_printf("pci@%"PRIX64, phb->buid);
2910     }
2911 
2912     if (vsc) {
2913         /* Same logic as virtio above */
2914         unsigned id = 0x1000000 | (vsc->target << 16) | vsc->lun;
2915         return g_strdup_printf("disk@%"PRIX64, (uint64_t)id << 32);
2916     }
2917 
2918     if (g_str_equal("pci-bridge", qdev_fw_name(dev))) {
2919         /* SLOF uses "pci" instead of "pci-bridge" for PCI bridges */
2920         PCIDevice *pcidev = CAST(PCIDevice, dev, TYPE_PCI_DEVICE);
2921         return g_strdup_printf("pci@%x", PCI_SLOT(pcidev->devfn));
2922     }
2923 
2924     return NULL;
2925 }
2926 
2927 static char *spapr_get_kvm_type(Object *obj, Error **errp)
2928 {
2929     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2930 
2931     return g_strdup(spapr->kvm_type);
2932 }
2933 
2934 static void spapr_set_kvm_type(Object *obj, const char *value, Error **errp)
2935 {
2936     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2937 
2938     g_free(spapr->kvm_type);
2939     spapr->kvm_type = g_strdup(value);
2940 }
2941 
2942 static bool spapr_get_modern_hotplug_events(Object *obj, Error **errp)
2943 {
2944     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2945 
2946     return spapr->use_hotplug_event_source;
2947 }
2948 
2949 static void spapr_set_modern_hotplug_events(Object *obj, bool value,
2950                                             Error **errp)
2951 {
2952     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2953 
2954     spapr->use_hotplug_event_source = value;
2955 }
2956 
2957 static bool spapr_get_msix_emulation(Object *obj, Error **errp)
2958 {
2959     return true;
2960 }
2961 
2962 static char *spapr_get_resize_hpt(Object *obj, Error **errp)
2963 {
2964     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2965 
2966     switch (spapr->resize_hpt) {
2967     case SPAPR_RESIZE_HPT_DEFAULT:
2968         return g_strdup("default");
2969     case SPAPR_RESIZE_HPT_DISABLED:
2970         return g_strdup("disabled");
2971     case SPAPR_RESIZE_HPT_ENABLED:
2972         return g_strdup("enabled");
2973     case SPAPR_RESIZE_HPT_REQUIRED:
2974         return g_strdup("required");
2975     }
2976     g_assert_not_reached();
2977 }
2978 
2979 static void spapr_set_resize_hpt(Object *obj, const char *value, Error **errp)
2980 {
2981     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2982 
2983     if (strcmp(value, "default") == 0) {
2984         spapr->resize_hpt = SPAPR_RESIZE_HPT_DEFAULT;
2985     } else if (strcmp(value, "disabled") == 0) {
2986         spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
2987     } else if (strcmp(value, "enabled") == 0) {
2988         spapr->resize_hpt = SPAPR_RESIZE_HPT_ENABLED;
2989     } else if (strcmp(value, "required") == 0) {
2990         spapr->resize_hpt = SPAPR_RESIZE_HPT_REQUIRED;
2991     } else {
2992         error_setg(errp, "Bad value for \"resize-hpt\" property");
2993     }
2994 }
2995 
2996 static void spapr_get_vsmt(Object *obj, Visitor *v, const char *name,
2997                                    void *opaque, Error **errp)
2998 {
2999     visit_type_uint32(v, name, (uint32_t *)opaque, errp);
3000 }
3001 
3002 static void spapr_set_vsmt(Object *obj, Visitor *v, const char *name,
3003                                    void *opaque, Error **errp)
3004 {
3005     visit_type_uint32(v, name, (uint32_t *)opaque, errp);
3006 }
3007 
3008 static void spapr_instance_init(Object *obj)
3009 {
3010     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
3011 
3012     spapr->htab_fd = -1;
3013     spapr->use_hotplug_event_source = true;
3014     object_property_add_str(obj, "kvm-type",
3015                             spapr_get_kvm_type, spapr_set_kvm_type, NULL);
3016     object_property_set_description(obj, "kvm-type",
3017                                     "Specifies the KVM virtualization mode (HV, PR)",
3018                                     NULL);
3019     object_property_add_bool(obj, "modern-hotplug-events",
3020                             spapr_get_modern_hotplug_events,
3021                             spapr_set_modern_hotplug_events,
3022                             NULL);
3023     object_property_set_description(obj, "modern-hotplug-events",
3024                                     "Use dedicated hotplug event mechanism in"
3025                                     " place of standard EPOW events when possible"
3026                                     " (required for memory hot-unplug support)",
3027                                     NULL);
3028     ppc_compat_add_property(obj, "max-cpu-compat", &spapr->max_compat_pvr,
3029                             "Maximum permitted CPU compatibility mode",
3030                             &error_fatal);
3031 
3032     object_property_add_str(obj, "resize-hpt",
3033                             spapr_get_resize_hpt, spapr_set_resize_hpt, NULL);
3034     object_property_set_description(obj, "resize-hpt",
3035                                     "Resizing of the Hash Page Table (enabled, disabled, required)",
3036                                     NULL);
3037     object_property_add(obj, "vsmt", "uint32", spapr_get_vsmt,
3038                         spapr_set_vsmt, NULL, &spapr->vsmt, &error_abort);
3039     object_property_set_description(obj, "vsmt",
3040                                     "Virtual SMT: KVM behaves as if this were"
3041                                     " the host's SMT mode", &error_abort);
3042     object_property_add_bool(obj, "vfio-no-msix-emulation",
3043                              spapr_get_msix_emulation, NULL, NULL);
3044 }
3045 
3046 static void spapr_machine_finalizefn(Object *obj)
3047 {
3048     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
3049 
3050     g_free(spapr->kvm_type);
3051 }
3052 
3053 void spapr_do_system_reset_on_cpu(CPUState *cs, run_on_cpu_data arg)
3054 {
3055     cpu_synchronize_state(cs);
3056     ppc_cpu_do_system_reset(cs);
3057 }
3058 
3059 static void spapr_nmi(NMIState *n, int cpu_index, Error **errp)
3060 {
3061     CPUState *cs;
3062 
3063     CPU_FOREACH(cs) {
3064         async_run_on_cpu(cs, spapr_do_system_reset_on_cpu, RUN_ON_CPU_NULL);
3065     }
3066 }
3067 
3068 static void spapr_add_lmbs(DeviceState *dev, uint64_t addr_start, uint64_t size,
3069                            uint32_t node, bool dedicated_hp_event_source,
3070                            Error **errp)
3071 {
3072     sPAPRDRConnector *drc;
3073     uint32_t nr_lmbs = size/SPAPR_MEMORY_BLOCK_SIZE;
3074     int i, fdt_offset, fdt_size;
3075     void *fdt;
3076     uint64_t addr = addr_start;
3077     bool hotplugged = spapr_drc_hotplugged(dev);
3078     Error *local_err = NULL;
3079 
3080     for (i = 0; i < nr_lmbs; i++) {
3081         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3082                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3083         g_assert(drc);
3084 
3085         fdt = create_device_tree(&fdt_size);
3086         fdt_offset = spapr_populate_memory_node(fdt, node, addr,
3087                                                 SPAPR_MEMORY_BLOCK_SIZE);
3088 
3089         spapr_drc_attach(drc, dev, fdt, fdt_offset, &local_err);
3090         if (local_err) {
3091             while (addr > addr_start) {
3092                 addr -= SPAPR_MEMORY_BLOCK_SIZE;
3093                 drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3094                                       addr / SPAPR_MEMORY_BLOCK_SIZE);
3095                 spapr_drc_detach(drc);
3096             }
3097             g_free(fdt);
3098             error_propagate(errp, local_err);
3099             return;
3100         }
3101         if (!hotplugged) {
3102             spapr_drc_reset(drc);
3103         }
3104         addr += SPAPR_MEMORY_BLOCK_SIZE;
3105     }
3106     /* send hotplug notification to the
3107      * guest only in case of hotplugged memory
3108      */
3109     if (hotplugged) {
3110         if (dedicated_hp_event_source) {
3111             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3112                                   addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3113             spapr_hotplug_req_add_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3114                                                    nr_lmbs,
3115                                                    spapr_drc_index(drc));
3116         } else {
3117             spapr_hotplug_req_add_by_count(SPAPR_DR_CONNECTOR_TYPE_LMB,
3118                                            nr_lmbs);
3119         }
3120     }
3121 }
3122 
3123 static void spapr_memory_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3124                               Error **errp)
3125 {
3126     Error *local_err = NULL;
3127     sPAPRMachineState *ms = SPAPR_MACHINE(hotplug_dev);
3128     PCDIMMDevice *dimm = PC_DIMM(dev);
3129     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
3130     MemoryRegion *mr = ddc->get_memory_region(dimm, &error_abort);
3131     uint64_t size, addr;
3132     uint32_t node;
3133 
3134     size = memory_region_size(mr);
3135 
3136     pc_dimm_plug(dev, MACHINE(ms), &local_err);
3137     if (local_err) {
3138         goto out;
3139     }
3140 
3141     addr = object_property_get_uint(OBJECT(dimm),
3142                                     PC_DIMM_ADDR_PROP, &local_err);
3143     if (local_err) {
3144         goto out_unplug;
3145     }
3146 
3147     node = object_property_get_uint(OBJECT(dev), PC_DIMM_NODE_PROP,
3148                                     &error_abort);
3149     spapr_add_lmbs(dev, addr, size, node,
3150                    spapr_ovec_test(ms->ov5_cas, OV5_HP_EVT),
3151                    &local_err);
3152     if (local_err) {
3153         goto out_unplug;
3154     }
3155 
3156     return;
3157 
3158 out_unplug:
3159     pc_dimm_unplug(dev, MACHINE(ms));
3160 out:
3161     error_propagate(errp, local_err);
3162 }
3163 
3164 static void spapr_memory_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3165                                   Error **errp)
3166 {
3167     const sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(hotplug_dev);
3168     sPAPRMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3169     PCDIMMDevice *dimm = PC_DIMM(dev);
3170     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
3171     Error *local_err = NULL;
3172     MemoryRegion *mr;
3173     uint64_t size;
3174     Object *memdev;
3175     hwaddr pagesize;
3176 
3177     if (!smc->dr_lmb_enabled) {
3178         error_setg(errp, "Memory hotplug not supported for this machine");
3179         return;
3180     }
3181 
3182     mr = ddc->get_memory_region(dimm, errp);
3183     if (!mr) {
3184         return;
3185     }
3186     size = memory_region_size(mr);
3187 
3188     if (size % SPAPR_MEMORY_BLOCK_SIZE) {
3189         error_setg(errp, "Hotplugged memory size must be a multiple of "
3190                       "%" PRIu64 " MB", SPAPR_MEMORY_BLOCK_SIZE / MiB);
3191         return;
3192     }
3193 
3194     memdev = object_property_get_link(OBJECT(dimm), PC_DIMM_MEMDEV_PROP,
3195                                       &error_abort);
3196     pagesize = host_memory_backend_pagesize(MEMORY_BACKEND(memdev));
3197     spapr_check_pagesize(spapr, pagesize, &local_err);
3198     if (local_err) {
3199         error_propagate(errp, local_err);
3200         return;
3201     }
3202 
3203     pc_dimm_pre_plug(dev, MACHINE(hotplug_dev), NULL, errp);
3204 }
3205 
3206 struct sPAPRDIMMState {
3207     PCDIMMDevice *dimm;
3208     uint32_t nr_lmbs;
3209     QTAILQ_ENTRY(sPAPRDIMMState) next;
3210 };
3211 
3212 static sPAPRDIMMState *spapr_pending_dimm_unplugs_find(sPAPRMachineState *s,
3213                                                        PCDIMMDevice *dimm)
3214 {
3215     sPAPRDIMMState *dimm_state = NULL;
3216 
3217     QTAILQ_FOREACH(dimm_state, &s->pending_dimm_unplugs, next) {
3218         if (dimm_state->dimm == dimm) {
3219             break;
3220         }
3221     }
3222     return dimm_state;
3223 }
3224 
3225 static sPAPRDIMMState *spapr_pending_dimm_unplugs_add(sPAPRMachineState *spapr,
3226                                                       uint32_t nr_lmbs,
3227                                                       PCDIMMDevice *dimm)
3228 {
3229     sPAPRDIMMState *ds = NULL;
3230 
3231     /*
3232      * If this request is for a DIMM whose removal had failed earlier
3233      * (due to guest's refusal to remove the LMBs), we would have this
3234      * dimm already in the pending_dimm_unplugs list. In that
3235      * case don't add again.
3236      */
3237     ds = spapr_pending_dimm_unplugs_find(spapr, dimm);
3238     if (!ds) {
3239         ds = g_malloc0(sizeof(sPAPRDIMMState));
3240         ds->nr_lmbs = nr_lmbs;
3241         ds->dimm = dimm;
3242         QTAILQ_INSERT_HEAD(&spapr->pending_dimm_unplugs, ds, next);
3243     }
3244     return ds;
3245 }
3246 
3247 static void spapr_pending_dimm_unplugs_remove(sPAPRMachineState *spapr,
3248                                               sPAPRDIMMState *dimm_state)
3249 {
3250     QTAILQ_REMOVE(&spapr->pending_dimm_unplugs, dimm_state, next);
3251     g_free(dimm_state);
3252 }
3253 
3254 static sPAPRDIMMState *spapr_recover_pending_dimm_state(sPAPRMachineState *ms,
3255                                                         PCDIMMDevice *dimm)
3256 {
3257     sPAPRDRConnector *drc;
3258     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
3259     MemoryRegion *mr = ddc->get_memory_region(dimm, &error_abort);
3260     uint64_t size = memory_region_size(mr);
3261     uint32_t nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3262     uint32_t avail_lmbs = 0;
3263     uint64_t addr_start, addr;
3264     int i;
3265 
3266     addr_start = object_property_get_int(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3267                                          &error_abort);
3268 
3269     addr = addr_start;
3270     for (i = 0; i < nr_lmbs; i++) {
3271         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3272                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3273         g_assert(drc);
3274         if (drc->dev) {
3275             avail_lmbs++;
3276         }
3277         addr += SPAPR_MEMORY_BLOCK_SIZE;
3278     }
3279 
3280     return spapr_pending_dimm_unplugs_add(ms, avail_lmbs, dimm);
3281 }
3282 
3283 /* Callback to be called during DRC release. */
3284 void spapr_lmb_release(DeviceState *dev)
3285 {
3286     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3287     sPAPRMachineState *spapr = SPAPR_MACHINE(hotplug_ctrl);
3288     sPAPRDIMMState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3289 
3290     /* This information will get lost if a migration occurs
3291      * during the unplug process. In this case recover it. */
3292     if (ds == NULL) {
3293         ds = spapr_recover_pending_dimm_state(spapr, PC_DIMM(dev));
3294         g_assert(ds);
3295         /* The DRC being examined by the caller at least must be counted */
3296         g_assert(ds->nr_lmbs);
3297     }
3298 
3299     if (--ds->nr_lmbs) {
3300         return;
3301     }
3302 
3303     /*
3304      * Now that all the LMBs have been removed by the guest, call the
3305      * unplug handler chain. This can never fail.
3306      */
3307     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3308 }
3309 
3310 static void spapr_memory_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3311 {
3312     sPAPRMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3313     sPAPRDIMMState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3314 
3315     pc_dimm_unplug(dev, MACHINE(hotplug_dev));
3316     object_unparent(OBJECT(dev));
3317     spapr_pending_dimm_unplugs_remove(spapr, ds);
3318 }
3319 
3320 static void spapr_memory_unplug_request(HotplugHandler *hotplug_dev,
3321                                         DeviceState *dev, Error **errp)
3322 {
3323     sPAPRMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3324     Error *local_err = NULL;
3325     PCDIMMDevice *dimm = PC_DIMM(dev);
3326     PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
3327     MemoryRegion *mr = ddc->get_memory_region(dimm, &error_abort);
3328     uint32_t nr_lmbs;
3329     uint64_t size, addr_start, addr;
3330     int i;
3331     sPAPRDRConnector *drc;
3332 
3333     size = memory_region_size(mr);
3334     nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3335 
3336     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3337                                          &local_err);
3338     if (local_err) {
3339         goto out;
3340     }
3341 
3342     /*
3343      * An existing pending dimm state for this DIMM means that there is an
3344      * unplug operation in progress, waiting for the spapr_lmb_release
3345      * callback to complete the job (BQL can't cover that far). In this case,
3346      * bail out to avoid detaching DRCs that were already released.
3347      */
3348     if (spapr_pending_dimm_unplugs_find(spapr, dimm)) {
3349         error_setg(&local_err,
3350                    "Memory unplug already in progress for device %s",
3351                    dev->id);
3352         goto out;
3353     }
3354 
3355     spapr_pending_dimm_unplugs_add(spapr, nr_lmbs, dimm);
3356 
3357     addr = addr_start;
3358     for (i = 0; i < nr_lmbs; i++) {
3359         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3360                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3361         g_assert(drc);
3362 
3363         spapr_drc_detach(drc);
3364         addr += SPAPR_MEMORY_BLOCK_SIZE;
3365     }
3366 
3367     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3368                           addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3369     spapr_hotplug_req_remove_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3370                                               nr_lmbs, spapr_drc_index(drc));
3371 out:
3372     error_propagate(errp, local_err);
3373 }
3374 
3375 static void *spapr_populate_hotplug_cpu_dt(CPUState *cs, int *fdt_offset,
3376                                            sPAPRMachineState *spapr)
3377 {
3378     PowerPCCPU *cpu = POWERPC_CPU(cs);
3379     DeviceClass *dc = DEVICE_GET_CLASS(cs);
3380     int id = spapr_get_vcpu_id(cpu);
3381     void *fdt;
3382     int offset, fdt_size;
3383     char *nodename;
3384 
3385     fdt = create_device_tree(&fdt_size);
3386     nodename = g_strdup_printf("%s@%x", dc->fw_name, id);
3387     offset = fdt_add_subnode(fdt, 0, nodename);
3388 
3389     spapr_populate_cpu_dt(cs, fdt, offset, spapr);
3390     g_free(nodename);
3391 
3392     *fdt_offset = offset;
3393     return fdt;
3394 }
3395 
3396 /* Callback to be called during DRC release. */
3397 void spapr_core_release(DeviceState *dev)
3398 {
3399     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3400 
3401     /* Call the unplug handler chain. This can never fail. */
3402     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3403 }
3404 
3405 static void spapr_core_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3406 {
3407     MachineState *ms = MACHINE(hotplug_dev);
3408     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(ms);
3409     CPUCore *cc = CPU_CORE(dev);
3410     CPUArchId *core_slot = spapr_find_cpu_slot(ms, cc->core_id, NULL);
3411 
3412     if (smc->pre_2_10_has_unused_icps) {
3413         sPAPRCPUCore *sc = SPAPR_CPU_CORE(OBJECT(dev));
3414         int i;
3415 
3416         for (i = 0; i < cc->nr_threads; i++) {
3417             CPUState *cs = CPU(sc->threads[i]);
3418 
3419             pre_2_10_vmstate_register_dummy_icp(cs->cpu_index);
3420         }
3421     }
3422 
3423     assert(core_slot);
3424     core_slot->cpu = NULL;
3425     object_unparent(OBJECT(dev));
3426 }
3427 
3428 static
3429 void spapr_core_unplug_request(HotplugHandler *hotplug_dev, DeviceState *dev,
3430                                Error **errp)
3431 {
3432     sPAPRMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3433     int index;
3434     sPAPRDRConnector *drc;
3435     CPUCore *cc = CPU_CORE(dev);
3436 
3437     if (!spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index)) {
3438         error_setg(errp, "Unable to find CPU core with core-id: %d",
3439                    cc->core_id);
3440         return;
3441     }
3442     if (index == 0) {
3443         error_setg(errp, "Boot CPU core may not be unplugged");
3444         return;
3445     }
3446 
3447     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3448                           spapr_vcpu_id(spapr, cc->core_id));
3449     g_assert(drc);
3450 
3451     spapr_drc_detach(drc);
3452 
3453     spapr_hotplug_req_remove_by_index(drc);
3454 }
3455 
3456 static void spapr_core_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3457                             Error **errp)
3458 {
3459     sPAPRMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3460     MachineClass *mc = MACHINE_GET_CLASS(spapr);
3461     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3462     sPAPRCPUCore *core = SPAPR_CPU_CORE(OBJECT(dev));
3463     CPUCore *cc = CPU_CORE(dev);
3464     CPUState *cs = CPU(core->threads[0]);
3465     sPAPRDRConnector *drc;
3466     Error *local_err = NULL;
3467     CPUArchId *core_slot;
3468     int index;
3469     bool hotplugged = spapr_drc_hotplugged(dev);
3470 
3471     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3472     if (!core_slot) {
3473         error_setg(errp, "Unable to find CPU core with core-id: %d",
3474                    cc->core_id);
3475         return;
3476     }
3477     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3478                           spapr_vcpu_id(spapr, cc->core_id));
3479 
3480     g_assert(drc || !mc->has_hotpluggable_cpus);
3481 
3482     if (drc) {
3483         void *fdt;
3484         int fdt_offset;
3485 
3486         fdt = spapr_populate_hotplug_cpu_dt(cs, &fdt_offset, spapr);
3487 
3488         spapr_drc_attach(drc, dev, fdt, fdt_offset, &local_err);
3489         if (local_err) {
3490             g_free(fdt);
3491             error_propagate(errp, local_err);
3492             return;
3493         }
3494 
3495         if (hotplugged) {
3496             /*
3497              * Send hotplug notification interrupt to the guest only
3498              * in case of hotplugged CPUs.
3499              */
3500             spapr_hotplug_req_add_by_index(drc);
3501         } else {
3502             spapr_drc_reset(drc);
3503         }
3504     }
3505 
3506     core_slot->cpu = OBJECT(dev);
3507 
3508     if (smc->pre_2_10_has_unused_icps) {
3509         int i;
3510 
3511         for (i = 0; i < cc->nr_threads; i++) {
3512             cs = CPU(core->threads[i]);
3513             pre_2_10_vmstate_unregister_dummy_icp(cs->cpu_index);
3514         }
3515     }
3516 }
3517 
3518 static void spapr_core_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3519                                 Error **errp)
3520 {
3521     MachineState *machine = MACHINE(OBJECT(hotplug_dev));
3522     MachineClass *mc = MACHINE_GET_CLASS(hotplug_dev);
3523     Error *local_err = NULL;
3524     CPUCore *cc = CPU_CORE(dev);
3525     const char *base_core_type = spapr_get_cpu_core_type(machine->cpu_type);
3526     const char *type = object_get_typename(OBJECT(dev));
3527     CPUArchId *core_slot;
3528     int index;
3529 
3530     if (dev->hotplugged && !mc->has_hotpluggable_cpus) {
3531         error_setg(&local_err, "CPU hotplug not supported for this machine");
3532         goto out;
3533     }
3534 
3535     if (strcmp(base_core_type, type)) {
3536         error_setg(&local_err, "CPU core type should be %s", base_core_type);
3537         goto out;
3538     }
3539 
3540     if (cc->core_id % smp_threads) {
3541         error_setg(&local_err, "invalid core id %d", cc->core_id);
3542         goto out;
3543     }
3544 
3545     /*
3546      * In general we should have homogeneous threads-per-core, but old
3547      * (pre hotplug support) machine types allow the last core to have
3548      * reduced threads as a compatibility hack for when we allowed
3549      * total vcpus not a multiple of threads-per-core.
3550      */
3551     if (mc->has_hotpluggable_cpus && (cc->nr_threads != smp_threads)) {
3552         error_setg(&local_err, "invalid nr-threads %d, must be %d",
3553                    cc->nr_threads, smp_threads);
3554         goto out;
3555     }
3556 
3557     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3558     if (!core_slot) {
3559         error_setg(&local_err, "core id %d out of range", cc->core_id);
3560         goto out;
3561     }
3562 
3563     if (core_slot->cpu) {
3564         error_setg(&local_err, "core %d already populated", cc->core_id);
3565         goto out;
3566     }
3567 
3568     numa_cpu_pre_plug(core_slot, dev, &local_err);
3569 
3570 out:
3571     error_propagate(errp, local_err);
3572 }
3573 
3574 static void spapr_machine_device_plug(HotplugHandler *hotplug_dev,
3575                                       DeviceState *dev, Error **errp)
3576 {
3577     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
3578         spapr_memory_plug(hotplug_dev, dev, errp);
3579     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3580         spapr_core_plug(hotplug_dev, dev, errp);
3581     }
3582 }
3583 
3584 static void spapr_machine_device_unplug(HotplugHandler *hotplug_dev,
3585                                         DeviceState *dev, Error **errp)
3586 {
3587     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
3588         spapr_memory_unplug(hotplug_dev, dev);
3589     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3590         spapr_core_unplug(hotplug_dev, dev);
3591     }
3592 }
3593 
3594 static void spapr_machine_device_unplug_request(HotplugHandler *hotplug_dev,
3595                                                 DeviceState *dev, Error **errp)
3596 {
3597     sPAPRMachineState *sms = SPAPR_MACHINE(OBJECT(hotplug_dev));
3598     MachineClass *mc = MACHINE_GET_CLASS(sms);
3599 
3600     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
3601         if (spapr_ovec_test(sms->ov5_cas, OV5_HP_EVT)) {
3602             spapr_memory_unplug_request(hotplug_dev, dev, errp);
3603         } else {
3604             /* NOTE: this means there is a window after guest reset, prior to
3605              * CAS negotiation, where unplug requests will fail due to the
3606              * capability not being detected yet. This is a bit different than
3607              * the case with PCI unplug, where the events will be queued and
3608              * eventually handled by the guest after boot
3609              */
3610             error_setg(errp, "Memory hot unplug not supported for this guest");
3611         }
3612     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3613         if (!mc->has_hotpluggable_cpus) {
3614             error_setg(errp, "CPU hot unplug not supported on this machine");
3615             return;
3616         }
3617         spapr_core_unplug_request(hotplug_dev, dev, errp);
3618     }
3619 }
3620 
3621 static void spapr_machine_device_pre_plug(HotplugHandler *hotplug_dev,
3622                                           DeviceState *dev, Error **errp)
3623 {
3624     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
3625         spapr_memory_pre_plug(hotplug_dev, dev, errp);
3626     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3627         spapr_core_pre_plug(hotplug_dev, dev, errp);
3628     }
3629 }
3630 
3631 static HotplugHandler *spapr_get_hotplug_handler(MachineState *machine,
3632                                                  DeviceState *dev)
3633 {
3634     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM) ||
3635         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3636         return HOTPLUG_HANDLER(machine);
3637     }
3638     return NULL;
3639 }
3640 
3641 static CpuInstanceProperties
3642 spapr_cpu_index_to_props(MachineState *machine, unsigned cpu_index)
3643 {
3644     CPUArchId *core_slot;
3645     MachineClass *mc = MACHINE_GET_CLASS(machine);
3646 
3647     /* make sure possible_cpu are intialized */
3648     mc->possible_cpu_arch_ids(machine);
3649     /* get CPU core slot containing thread that matches cpu_index */
3650     core_slot = spapr_find_cpu_slot(machine, cpu_index, NULL);
3651     assert(core_slot);
3652     return core_slot->props;
3653 }
3654 
3655 static int64_t spapr_get_default_cpu_node_id(const MachineState *ms, int idx)
3656 {
3657     return idx / smp_cores % nb_numa_nodes;
3658 }
3659 
3660 static const CPUArchIdList *spapr_possible_cpu_arch_ids(MachineState *machine)
3661 {
3662     int i;
3663     const char *core_type;
3664     int spapr_max_cores = max_cpus / smp_threads;
3665     MachineClass *mc = MACHINE_GET_CLASS(machine);
3666 
3667     if (!mc->has_hotpluggable_cpus) {
3668         spapr_max_cores = QEMU_ALIGN_UP(smp_cpus, smp_threads) / smp_threads;
3669     }
3670     if (machine->possible_cpus) {
3671         assert(machine->possible_cpus->len == spapr_max_cores);
3672         return machine->possible_cpus;
3673     }
3674 
3675     core_type = spapr_get_cpu_core_type(machine->cpu_type);
3676     if (!core_type) {
3677         error_report("Unable to find sPAPR CPU Core definition");
3678         exit(1);
3679     }
3680 
3681     machine->possible_cpus = g_malloc0(sizeof(CPUArchIdList) +
3682                              sizeof(CPUArchId) * spapr_max_cores);
3683     machine->possible_cpus->len = spapr_max_cores;
3684     for (i = 0; i < machine->possible_cpus->len; i++) {
3685         int core_id = i * smp_threads;
3686 
3687         machine->possible_cpus->cpus[i].type = core_type;
3688         machine->possible_cpus->cpus[i].vcpus_count = smp_threads;
3689         machine->possible_cpus->cpus[i].arch_id = core_id;
3690         machine->possible_cpus->cpus[i].props.has_core_id = true;
3691         machine->possible_cpus->cpus[i].props.core_id = core_id;
3692     }
3693     return machine->possible_cpus;
3694 }
3695 
3696 static void spapr_phb_placement(sPAPRMachineState *spapr, uint32_t index,
3697                                 uint64_t *buid, hwaddr *pio,
3698                                 hwaddr *mmio32, hwaddr *mmio64,
3699                                 unsigned n_dma, uint32_t *liobns, Error **errp)
3700 {
3701     /*
3702      * New-style PHB window placement.
3703      *
3704      * Goals: Gives large (1TiB), naturally aligned 64-bit MMIO window
3705      * for each PHB, in addition to 2GiB 32-bit MMIO and 64kiB PIO
3706      * windows.
3707      *
3708      * Some guest kernels can't work with MMIO windows above 1<<46
3709      * (64TiB), so we place up to 31 PHBs in the area 32TiB..64TiB
3710      *
3711      * 32TiB..(33TiB+1984kiB) contains the 64kiB PIO windows for each
3712      * PHB stacked together.  (32TiB+2GiB)..(32TiB+64GiB) contains the
3713      * 2GiB 32-bit MMIO windows for each PHB.  Then 33..64TiB has the
3714      * 1TiB 64-bit MMIO windows for each PHB.
3715      */
3716     const uint64_t base_buid = 0x800000020000000ULL;
3717 #define SPAPR_MAX_PHBS ((SPAPR_PCI_LIMIT - SPAPR_PCI_BASE) / \
3718                         SPAPR_PCI_MEM64_WIN_SIZE - 1)
3719     int i;
3720 
3721     /* Sanity check natural alignments */
3722     QEMU_BUILD_BUG_ON((SPAPR_PCI_BASE % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
3723     QEMU_BUILD_BUG_ON((SPAPR_PCI_LIMIT % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
3724     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM64_WIN_SIZE % SPAPR_PCI_MEM32_WIN_SIZE) != 0);
3725     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM32_WIN_SIZE % SPAPR_PCI_IO_WIN_SIZE) != 0);
3726     /* Sanity check bounds */
3727     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_IO_WIN_SIZE) >
3728                       SPAPR_PCI_MEM32_WIN_SIZE);
3729     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_MEM32_WIN_SIZE) >
3730                       SPAPR_PCI_MEM64_WIN_SIZE);
3731 
3732     if (index >= SPAPR_MAX_PHBS) {
3733         error_setg(errp, "\"index\" for PAPR PHB is too large (max %llu)",
3734                    SPAPR_MAX_PHBS - 1);
3735         return;
3736     }
3737 
3738     *buid = base_buid + index;
3739     for (i = 0; i < n_dma; ++i) {
3740         liobns[i] = SPAPR_PCI_LIOBN(index, i);
3741     }
3742 
3743     *pio = SPAPR_PCI_BASE + index * SPAPR_PCI_IO_WIN_SIZE;
3744     *mmio32 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM32_WIN_SIZE;
3745     *mmio64 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM64_WIN_SIZE;
3746 }
3747 
3748 static ICSState *spapr_ics_get(XICSFabric *dev, int irq)
3749 {
3750     sPAPRMachineState *spapr = SPAPR_MACHINE(dev);
3751 
3752     return ics_valid_irq(spapr->ics, irq) ? spapr->ics : NULL;
3753 }
3754 
3755 static void spapr_ics_resend(XICSFabric *dev)
3756 {
3757     sPAPRMachineState *spapr = SPAPR_MACHINE(dev);
3758 
3759     ics_resend(spapr->ics);
3760 }
3761 
3762 static ICPState *spapr_icp_get(XICSFabric *xi, int vcpu_id)
3763 {
3764     PowerPCCPU *cpu = spapr_find_cpu(vcpu_id);
3765 
3766     return cpu ? ICP(cpu->intc) : NULL;
3767 }
3768 
3769 static void spapr_pic_print_info(InterruptStatsProvider *obj,
3770                                  Monitor *mon)
3771 {
3772     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
3773     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
3774 
3775     smc->irq->print_info(spapr, mon);
3776 }
3777 
3778 int spapr_get_vcpu_id(PowerPCCPU *cpu)
3779 {
3780     return cpu->vcpu_id;
3781 }
3782 
3783 void spapr_set_vcpu_id(PowerPCCPU *cpu, int cpu_index, Error **errp)
3784 {
3785     sPAPRMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
3786     int vcpu_id;
3787 
3788     vcpu_id = spapr_vcpu_id(spapr, cpu_index);
3789 
3790     if (kvm_enabled() && !kvm_vcpu_id_is_valid(vcpu_id)) {
3791         error_setg(errp, "Can't create CPU with id %d in KVM", vcpu_id);
3792         error_append_hint(errp, "Adjust the number of cpus to %d "
3793                           "or try to raise the number of threads per core\n",
3794                           vcpu_id * smp_threads / spapr->vsmt);
3795         return;
3796     }
3797 
3798     cpu->vcpu_id = vcpu_id;
3799 }
3800 
3801 PowerPCCPU *spapr_find_cpu(int vcpu_id)
3802 {
3803     CPUState *cs;
3804 
3805     CPU_FOREACH(cs) {
3806         PowerPCCPU *cpu = POWERPC_CPU(cs);
3807 
3808         if (spapr_get_vcpu_id(cpu) == vcpu_id) {
3809             return cpu;
3810         }
3811     }
3812 
3813     return NULL;
3814 }
3815 
3816 static void spapr_machine_class_init(ObjectClass *oc, void *data)
3817 {
3818     MachineClass *mc = MACHINE_CLASS(oc);
3819     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(oc);
3820     FWPathProviderClass *fwc = FW_PATH_PROVIDER_CLASS(oc);
3821     NMIClass *nc = NMI_CLASS(oc);
3822     HotplugHandlerClass *hc = HOTPLUG_HANDLER_CLASS(oc);
3823     PPCVirtualHypervisorClass *vhc = PPC_VIRTUAL_HYPERVISOR_CLASS(oc);
3824     XICSFabricClass *xic = XICS_FABRIC_CLASS(oc);
3825     InterruptStatsProviderClass *ispc = INTERRUPT_STATS_PROVIDER_CLASS(oc);
3826 
3827     mc->desc = "pSeries Logical Partition (PAPR compliant)";
3828     mc->ignore_boot_device_suffixes = true;
3829 
3830     /*
3831      * We set up the default / latest behaviour here.  The class_init
3832      * functions for the specific versioned machine types can override
3833      * these details for backwards compatibility
3834      */
3835     mc->init = spapr_machine_init;
3836     mc->reset = spapr_machine_reset;
3837     mc->block_default_type = IF_SCSI;
3838     mc->max_cpus = 1024;
3839     mc->no_parallel = 1;
3840     mc->default_boot_order = "";
3841     mc->default_ram_size = 512 * MiB;
3842     mc->default_display = "std";
3843     mc->kvm_type = spapr_kvm_type;
3844     machine_class_allow_dynamic_sysbus_dev(mc, TYPE_SPAPR_PCI_HOST_BRIDGE);
3845     mc->pci_allow_0_address = true;
3846     assert(!mc->get_hotplug_handler);
3847     mc->get_hotplug_handler = spapr_get_hotplug_handler;
3848     hc->pre_plug = spapr_machine_device_pre_plug;
3849     hc->plug = spapr_machine_device_plug;
3850     mc->cpu_index_to_instance_props = spapr_cpu_index_to_props;
3851     mc->get_default_cpu_node_id = spapr_get_default_cpu_node_id;
3852     mc->possible_cpu_arch_ids = spapr_possible_cpu_arch_ids;
3853     hc->unplug_request = spapr_machine_device_unplug_request;
3854     hc->unplug = spapr_machine_device_unplug;
3855 
3856     smc->dr_lmb_enabled = true;
3857     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power8_v2.0");
3858     mc->has_hotpluggable_cpus = true;
3859     smc->resize_hpt_default = SPAPR_RESIZE_HPT_ENABLED;
3860     fwc->get_dev_path = spapr_get_fw_dev_path;
3861     nc->nmi_monitor_handler = spapr_nmi;
3862     smc->phb_placement = spapr_phb_placement;
3863     vhc->hypercall = emulate_spapr_hypercall;
3864     vhc->hpt_mask = spapr_hpt_mask;
3865     vhc->map_hptes = spapr_map_hptes;
3866     vhc->unmap_hptes = spapr_unmap_hptes;
3867     vhc->store_hpte = spapr_store_hpte;
3868     vhc->get_patbe = spapr_get_patbe;
3869     vhc->encode_hpt_for_kvm_pr = spapr_encode_hpt_for_kvm_pr;
3870     xic->ics_get = spapr_ics_get;
3871     xic->ics_resend = spapr_ics_resend;
3872     xic->icp_get = spapr_icp_get;
3873     ispc->print_info = spapr_pic_print_info;
3874     /* Force NUMA node memory size to be a multiple of
3875      * SPAPR_MEMORY_BLOCK_SIZE (256M) since that's the granularity
3876      * in which LMBs are represented and hot-added
3877      */
3878     mc->numa_mem_align_shift = 28;
3879 
3880     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_OFF;
3881     smc->default_caps.caps[SPAPR_CAP_VSX] = SPAPR_CAP_ON;
3882     smc->default_caps.caps[SPAPR_CAP_DFP] = SPAPR_CAP_ON;
3883     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_BROKEN;
3884     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_BROKEN;
3885     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_BROKEN;
3886     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 16; /* 64kiB */
3887     spapr_caps_add_properties(smc, &error_abort);
3888     smc->irq = &spapr_irq_xics;
3889 }
3890 
3891 static const TypeInfo spapr_machine_info = {
3892     .name          = TYPE_SPAPR_MACHINE,
3893     .parent        = TYPE_MACHINE,
3894     .abstract      = true,
3895     .instance_size = sizeof(sPAPRMachineState),
3896     .instance_init = spapr_instance_init,
3897     .instance_finalize = spapr_machine_finalizefn,
3898     .class_size    = sizeof(sPAPRMachineClass),
3899     .class_init    = spapr_machine_class_init,
3900     .interfaces = (InterfaceInfo[]) {
3901         { TYPE_FW_PATH_PROVIDER },
3902         { TYPE_NMI },
3903         { TYPE_HOTPLUG_HANDLER },
3904         { TYPE_PPC_VIRTUAL_HYPERVISOR },
3905         { TYPE_XICS_FABRIC },
3906         { TYPE_INTERRUPT_STATS_PROVIDER },
3907         { }
3908     },
3909 };
3910 
3911 #define DEFINE_SPAPR_MACHINE(suffix, verstr, latest)                 \
3912     static void spapr_machine_##suffix##_class_init(ObjectClass *oc, \
3913                                                     void *data)      \
3914     {                                                                \
3915         MachineClass *mc = MACHINE_CLASS(oc);                        \
3916         spapr_machine_##suffix##_class_options(mc);                  \
3917         if (latest) {                                                \
3918             mc->alias = "pseries";                                   \
3919             mc->is_default = 1;                                      \
3920         }                                                            \
3921     }                                                                \
3922     static void spapr_machine_##suffix##_instance_init(Object *obj)  \
3923     {                                                                \
3924         MachineState *machine = MACHINE(obj);                        \
3925         spapr_machine_##suffix##_instance_options(machine);          \
3926     }                                                                \
3927     static const TypeInfo spapr_machine_##suffix##_info = {          \
3928         .name = MACHINE_TYPE_NAME("pseries-" verstr),                \
3929         .parent = TYPE_SPAPR_MACHINE,                                \
3930         .class_init = spapr_machine_##suffix##_class_init,           \
3931         .instance_init = spapr_machine_##suffix##_instance_init,     \
3932     };                                                               \
3933     static void spapr_machine_register_##suffix(void)                \
3934     {                                                                \
3935         type_register(&spapr_machine_##suffix##_info);               \
3936     }                                                                \
3937     type_init(spapr_machine_register_##suffix)
3938 
3939  /*
3940  * pseries-3.1
3941  */
3942 static void spapr_machine_3_1_instance_options(MachineState *machine)
3943 {
3944 }
3945 
3946 static void spapr_machine_3_1_class_options(MachineClass *mc)
3947 {
3948     /* Defaults for the latest behaviour inherited from the base class */
3949 }
3950 
3951 DEFINE_SPAPR_MACHINE(3_1, "3.1", true);
3952 
3953 /*
3954  * pseries-3.0
3955  */
3956 #define SPAPR_COMPAT_3_0                                              \
3957     HW_COMPAT_3_0
3958 
3959 static void spapr_machine_3_0_instance_options(MachineState *machine)
3960 {
3961     spapr_machine_3_1_instance_options(machine);
3962 }
3963 
3964 static void spapr_machine_3_0_class_options(MachineClass *mc)
3965 {
3966     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3967 
3968     spapr_machine_3_1_class_options(mc);
3969     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_3_0);
3970 
3971     smc->legacy_irq_allocation = true;
3972 }
3973 
3974 DEFINE_SPAPR_MACHINE(3_0, "3.0", false);
3975 
3976 /*
3977  * pseries-2.12
3978  */
3979 #define SPAPR_COMPAT_2_12                                              \
3980     HW_COMPAT_2_12                                                     \
3981     {                                                                  \
3982         .driver = TYPE_POWERPC_CPU,                                    \
3983         .property = "pre-3.0-migration",                               \
3984         .value    = "on",                                              \
3985     },                                                                 \
3986     {                                                                  \
3987         .driver = TYPE_SPAPR_CPU_CORE,                                 \
3988         .property = "pre-3.0-migration",                               \
3989         .value    = "on",                                              \
3990     },
3991 
3992 static void spapr_machine_2_12_instance_options(MachineState *machine)
3993 {
3994     spapr_machine_3_0_instance_options(machine);
3995 }
3996 
3997 static void spapr_machine_2_12_class_options(MachineClass *mc)
3998 {
3999     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4000 
4001     spapr_machine_3_0_class_options(mc);
4002     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_12);
4003 
4004     /* We depend on kvm_enabled() to choose a default value for the
4005      * hpt-max-page-size capability. Of course we can't do it here
4006      * because this is too early and the HW accelerator isn't initialzed
4007      * yet. Postpone this to machine init (see default_caps_with_cpu()).
4008      */
4009     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 0;
4010 }
4011 
4012 DEFINE_SPAPR_MACHINE(2_12, "2.12", false);
4013 
4014 static void spapr_machine_2_12_sxxm_instance_options(MachineState *machine)
4015 {
4016     spapr_machine_2_12_instance_options(machine);
4017 }
4018 
4019 static void spapr_machine_2_12_sxxm_class_options(MachineClass *mc)
4020 {
4021     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4022 
4023     spapr_machine_2_12_class_options(mc);
4024     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND;
4025     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND;
4026     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_FIXED_CCD;
4027 }
4028 
4029 DEFINE_SPAPR_MACHINE(2_12_sxxm, "2.12-sxxm", false);
4030 
4031 /*
4032  * pseries-2.11
4033  */
4034 #define SPAPR_COMPAT_2_11                                              \
4035     HW_COMPAT_2_11
4036 
4037 static void spapr_machine_2_11_instance_options(MachineState *machine)
4038 {
4039     spapr_machine_2_12_instance_options(machine);
4040 }
4041 
4042 static void spapr_machine_2_11_class_options(MachineClass *mc)
4043 {
4044     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4045 
4046     spapr_machine_2_12_class_options(mc);
4047     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_ON;
4048     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_11);
4049 }
4050 
4051 DEFINE_SPAPR_MACHINE(2_11, "2.11", false);
4052 
4053 /*
4054  * pseries-2.10
4055  */
4056 #define SPAPR_COMPAT_2_10                                              \
4057     HW_COMPAT_2_10
4058 
4059 static void spapr_machine_2_10_instance_options(MachineState *machine)
4060 {
4061     spapr_machine_2_11_instance_options(machine);
4062 }
4063 
4064 static void spapr_machine_2_10_class_options(MachineClass *mc)
4065 {
4066     spapr_machine_2_11_class_options(mc);
4067     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_10);
4068 }
4069 
4070 DEFINE_SPAPR_MACHINE(2_10, "2.10", false);
4071 
4072 /*
4073  * pseries-2.9
4074  */
4075 #define SPAPR_COMPAT_2_9                                               \
4076     HW_COMPAT_2_9                                                      \
4077     {                                                                  \
4078         .driver = TYPE_POWERPC_CPU,                                    \
4079         .property = "pre-2.10-migration",                              \
4080         .value    = "on",                                              \
4081     },                                                                 \
4082 
4083 static void spapr_machine_2_9_instance_options(MachineState *machine)
4084 {
4085     spapr_machine_2_10_instance_options(machine);
4086 }
4087 
4088 static void spapr_machine_2_9_class_options(MachineClass *mc)
4089 {
4090     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4091 
4092     spapr_machine_2_10_class_options(mc);
4093     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_9);
4094     mc->numa_auto_assign_ram = numa_legacy_auto_assign_ram;
4095     smc->pre_2_10_has_unused_icps = true;
4096     smc->resize_hpt_default = SPAPR_RESIZE_HPT_DISABLED;
4097 }
4098 
4099 DEFINE_SPAPR_MACHINE(2_9, "2.9", false);
4100 
4101 /*
4102  * pseries-2.8
4103  */
4104 #define SPAPR_COMPAT_2_8                                        \
4105     HW_COMPAT_2_8                                               \
4106     {                                                           \
4107         .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,                 \
4108         .property = "pcie-extended-configuration-space",        \
4109         .value    = "off",                                      \
4110     },
4111 
4112 static void spapr_machine_2_8_instance_options(MachineState *machine)
4113 {
4114     spapr_machine_2_9_instance_options(machine);
4115 }
4116 
4117 static void spapr_machine_2_8_class_options(MachineClass *mc)
4118 {
4119     spapr_machine_2_9_class_options(mc);
4120     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_8);
4121     mc->numa_mem_align_shift = 23;
4122 }
4123 
4124 DEFINE_SPAPR_MACHINE(2_8, "2.8", false);
4125 
4126 /*
4127  * pseries-2.7
4128  */
4129 #define SPAPR_COMPAT_2_7                            \
4130     HW_COMPAT_2_7                                   \
4131     {                                               \
4132         .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,     \
4133         .property = "mem_win_size",                 \
4134         .value    = stringify(SPAPR_PCI_2_7_MMIO_WIN_SIZE),\
4135     },                                              \
4136     {                                               \
4137         .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,     \
4138         .property = "mem64_win_size",               \
4139         .value    = "0",                            \
4140     },                                              \
4141     {                                               \
4142         .driver = TYPE_POWERPC_CPU,                 \
4143         .property = "pre-2.8-migration",            \
4144         .value    = "on",                           \
4145     },                                              \
4146     {                                               \
4147         .driver = TYPE_SPAPR_PCI_HOST_BRIDGE,       \
4148         .property = "pre-2.8-migration",            \
4149         .value    = "on",                           \
4150     },
4151 
4152 static void phb_placement_2_7(sPAPRMachineState *spapr, uint32_t index,
4153                               uint64_t *buid, hwaddr *pio,
4154                               hwaddr *mmio32, hwaddr *mmio64,
4155                               unsigned n_dma, uint32_t *liobns, Error **errp)
4156 {
4157     /* Legacy PHB placement for pseries-2.7 and earlier machine types */
4158     const uint64_t base_buid = 0x800000020000000ULL;
4159     const hwaddr phb_spacing = 0x1000000000ULL; /* 64 GiB */
4160     const hwaddr mmio_offset = 0xa0000000; /* 2 GiB + 512 MiB */
4161     const hwaddr pio_offset = 0x80000000; /* 2 GiB */
4162     const uint32_t max_index = 255;
4163     const hwaddr phb0_alignment = 0x10000000000ULL; /* 1 TiB */
4164 
4165     uint64_t ram_top = MACHINE(spapr)->ram_size;
4166     hwaddr phb0_base, phb_base;
4167     int i;
4168 
4169     /* Do we have device memory? */
4170     if (MACHINE(spapr)->maxram_size > ram_top) {
4171         /* Can't just use maxram_size, because there may be an
4172          * alignment gap between normal and device memory regions
4173          */
4174         ram_top = MACHINE(spapr)->device_memory->base +
4175             memory_region_size(&MACHINE(spapr)->device_memory->mr);
4176     }
4177 
4178     phb0_base = QEMU_ALIGN_UP(ram_top, phb0_alignment);
4179 
4180     if (index > max_index) {
4181         error_setg(errp, "\"index\" for PAPR PHB is too large (max %u)",
4182                    max_index);
4183         return;
4184     }
4185 
4186     *buid = base_buid + index;
4187     for (i = 0; i < n_dma; ++i) {
4188         liobns[i] = SPAPR_PCI_LIOBN(index, i);
4189     }
4190 
4191     phb_base = phb0_base + index * phb_spacing;
4192     *pio = phb_base + pio_offset;
4193     *mmio32 = phb_base + mmio_offset;
4194     /*
4195      * We don't set the 64-bit MMIO window, relying on the PHB's
4196      * fallback behaviour of automatically splitting a large "32-bit"
4197      * window into contiguous 32-bit and 64-bit windows
4198      */
4199 }
4200 
4201 static void spapr_machine_2_7_instance_options(MachineState *machine)
4202 {
4203     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
4204 
4205     spapr_machine_2_8_instance_options(machine);
4206     spapr->use_hotplug_event_source = false;
4207 }
4208 
4209 static void spapr_machine_2_7_class_options(MachineClass *mc)
4210 {
4211     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4212 
4213     spapr_machine_2_8_class_options(mc);
4214     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power7_v2.3");
4215     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_7);
4216     smc->phb_placement = phb_placement_2_7;
4217 }
4218 
4219 DEFINE_SPAPR_MACHINE(2_7, "2.7", false);
4220 
4221 /*
4222  * pseries-2.6
4223  */
4224 #define SPAPR_COMPAT_2_6 \
4225     HW_COMPAT_2_6 \
4226     { \
4227         .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,\
4228         .property = "ddw",\
4229         .value    = stringify(off),\
4230     },
4231 
4232 static void spapr_machine_2_6_instance_options(MachineState *machine)
4233 {
4234     spapr_machine_2_7_instance_options(machine);
4235 }
4236 
4237 static void spapr_machine_2_6_class_options(MachineClass *mc)
4238 {
4239     spapr_machine_2_7_class_options(mc);
4240     mc->has_hotpluggable_cpus = false;
4241     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_6);
4242 }
4243 
4244 DEFINE_SPAPR_MACHINE(2_6, "2.6", false);
4245 
4246 /*
4247  * pseries-2.5
4248  */
4249 #define SPAPR_COMPAT_2_5 \
4250     HW_COMPAT_2_5 \
4251     { \
4252         .driver   = "spapr-vlan", \
4253         .property = "use-rx-buffer-pools", \
4254         .value    = "off", \
4255     },
4256 
4257 static void spapr_machine_2_5_instance_options(MachineState *machine)
4258 {
4259     spapr_machine_2_6_instance_options(machine);
4260 }
4261 
4262 static void spapr_machine_2_5_class_options(MachineClass *mc)
4263 {
4264     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4265 
4266     spapr_machine_2_6_class_options(mc);
4267     smc->use_ohci_by_default = true;
4268     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_5);
4269 }
4270 
4271 DEFINE_SPAPR_MACHINE(2_5, "2.5", false);
4272 
4273 /*
4274  * pseries-2.4
4275  */
4276 #define SPAPR_COMPAT_2_4 \
4277         HW_COMPAT_2_4
4278 
4279 static void spapr_machine_2_4_instance_options(MachineState *machine)
4280 {
4281     spapr_machine_2_5_instance_options(machine);
4282 }
4283 
4284 static void spapr_machine_2_4_class_options(MachineClass *mc)
4285 {
4286     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4287 
4288     spapr_machine_2_5_class_options(mc);
4289     smc->dr_lmb_enabled = false;
4290     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_4);
4291 }
4292 
4293 DEFINE_SPAPR_MACHINE(2_4, "2.4", false);
4294 
4295 /*
4296  * pseries-2.3
4297  */
4298 #define SPAPR_COMPAT_2_3 \
4299         HW_COMPAT_2_3 \
4300         {\
4301             .driver   = "spapr-pci-host-bridge",\
4302             .property = "dynamic-reconfiguration",\
4303             .value    = "off",\
4304         },
4305 
4306 static void spapr_machine_2_3_instance_options(MachineState *machine)
4307 {
4308     spapr_machine_2_4_instance_options(machine);
4309 }
4310 
4311 static void spapr_machine_2_3_class_options(MachineClass *mc)
4312 {
4313     spapr_machine_2_4_class_options(mc);
4314     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_3);
4315 }
4316 DEFINE_SPAPR_MACHINE(2_3, "2.3", false);
4317 
4318 /*
4319  * pseries-2.2
4320  */
4321 
4322 #define SPAPR_COMPAT_2_2 \
4323         HW_COMPAT_2_2 \
4324         {\
4325             .driver   = TYPE_SPAPR_PCI_HOST_BRIDGE,\
4326             .property = "mem_win_size",\
4327             .value    = "0x20000000",\
4328         },
4329 
4330 static void spapr_machine_2_2_instance_options(MachineState *machine)
4331 {
4332     spapr_machine_2_3_instance_options(machine);
4333     machine->suppress_vmdesc = true;
4334 }
4335 
4336 static void spapr_machine_2_2_class_options(MachineClass *mc)
4337 {
4338     spapr_machine_2_3_class_options(mc);
4339     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_2);
4340 }
4341 DEFINE_SPAPR_MACHINE(2_2, "2.2", false);
4342 
4343 /*
4344  * pseries-2.1
4345  */
4346 #define SPAPR_COMPAT_2_1 \
4347         HW_COMPAT_2_1
4348 
4349 static void spapr_machine_2_1_instance_options(MachineState *machine)
4350 {
4351     spapr_machine_2_2_instance_options(machine);
4352 }
4353 
4354 static void spapr_machine_2_1_class_options(MachineClass *mc)
4355 {
4356     spapr_machine_2_2_class_options(mc);
4357     SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_1);
4358 }
4359 DEFINE_SPAPR_MACHINE(2_1, "2.1", false);
4360 
4361 static void spapr_machine_register_types(void)
4362 {
4363     type_register_static(&spapr_machine_info);
4364 }
4365 
4366 type_init(spapr_machine_register_types)
4367