xref: /qemu/hw/ppc/spapr.c (revision 3054b0ca4bdd83e0780bd76805ca17aa733031c8)
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 "qemu/cutils.h"
74 #include "hw/ppc/spapr_cpu_core.h"
75 #include "hw/mem/memory-device.h"
76 
77 #include <libfdt.h>
78 
79 /* SLOF memory layout:
80  *
81  * SLOF raw image loaded at 0, copies its romfs right below the flat
82  * device-tree, then position SLOF itself 31M below that
83  *
84  * So we set FW_OVERHEAD to 40MB which should account for all of that
85  * and more
86  *
87  * We load our kernel at 4M, leaving space for SLOF initial image
88  */
89 #define FDT_MAX_SIZE            0x100000
90 #define RTAS_MAX_SIZE           0x10000
91 #define RTAS_MAX_ADDR           0x80000000 /* RTAS must stay below that */
92 #define FW_MAX_SIZE             0x400000
93 #define FW_FILE_NAME            "slof.bin"
94 #define FW_OVERHEAD             0x2800000
95 #define KERNEL_LOAD_ADDR        FW_MAX_SIZE
96 
97 #define MIN_RMA_SLOF            128UL
98 
99 #define PHANDLE_INTC            0x00001111
100 
101 /* These two functions implement the VCPU id numbering: one to compute them
102  * all and one to identify thread 0 of a VCORE. Any change to the first one
103  * is likely to have an impact on the second one, so let's keep them close.
104  */
105 static int spapr_vcpu_id(sPAPRMachineState *spapr, int cpu_index)
106 {
107     assert(spapr->vsmt);
108     return
109         (cpu_index / smp_threads) * spapr->vsmt + cpu_index % smp_threads;
110 }
111 static bool spapr_is_thread0_in_vcore(sPAPRMachineState *spapr,
112                                       PowerPCCPU *cpu)
113 {
114     assert(spapr->vsmt);
115     return spapr_get_vcpu_id(cpu) % spapr->vsmt == 0;
116 }
117 
118 static bool pre_2_10_vmstate_dummy_icp_needed(void *opaque)
119 {
120     /* Dummy entries correspond to unused ICPState objects in older QEMUs,
121      * and newer QEMUs don't even have them. In both cases, we don't want
122      * to send anything on the wire.
123      */
124     return false;
125 }
126 
127 static const VMStateDescription pre_2_10_vmstate_dummy_icp = {
128     .name = "icp/server",
129     .version_id = 1,
130     .minimum_version_id = 1,
131     .needed = pre_2_10_vmstate_dummy_icp_needed,
132     .fields = (VMStateField[]) {
133         VMSTATE_UNUSED(4), /* uint32_t xirr */
134         VMSTATE_UNUSED(1), /* uint8_t pending_priority */
135         VMSTATE_UNUSED(1), /* uint8_t mfrr */
136         VMSTATE_END_OF_LIST()
137     },
138 };
139 
140 static void pre_2_10_vmstate_register_dummy_icp(int i)
141 {
142     vmstate_register(NULL, i, &pre_2_10_vmstate_dummy_icp,
143                      (void *)(uintptr_t) i);
144 }
145 
146 static void pre_2_10_vmstate_unregister_dummy_icp(int i)
147 {
148     vmstate_unregister(NULL, &pre_2_10_vmstate_dummy_icp,
149                        (void *)(uintptr_t) i);
150 }
151 
152 int spapr_max_server_number(sPAPRMachineState *spapr)
153 {
154     assert(spapr->vsmt);
155     return DIV_ROUND_UP(max_cpus * spapr->vsmt, smp_threads);
156 }
157 
158 static int spapr_fixup_cpu_smt_dt(void *fdt, int offset, PowerPCCPU *cpu,
159                                   int smt_threads)
160 {
161     int i, ret = 0;
162     uint32_t servers_prop[smt_threads];
163     uint32_t gservers_prop[smt_threads * 2];
164     int index = spapr_get_vcpu_id(cpu);
165 
166     if (cpu->compat_pvr) {
167         ret = fdt_setprop_cell(fdt, offset, "cpu-version", cpu->compat_pvr);
168         if (ret < 0) {
169             return ret;
170         }
171     }
172 
173     /* Build interrupt servers and gservers properties */
174     for (i = 0; i < smt_threads; i++) {
175         servers_prop[i] = cpu_to_be32(index + i);
176         /* Hack, direct the group queues back to cpu 0 */
177         gservers_prop[i*2] = cpu_to_be32(index + i);
178         gservers_prop[i*2 + 1] = 0;
179     }
180     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-server#s",
181                       servers_prop, sizeof(servers_prop));
182     if (ret < 0) {
183         return ret;
184     }
185     ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-gserver#s",
186                       gservers_prop, sizeof(gservers_prop));
187 
188     return ret;
189 }
190 
191 static int spapr_fixup_cpu_numa_dt(void *fdt, int offset, PowerPCCPU *cpu)
192 {
193     int index = spapr_get_vcpu_id(cpu);
194     uint32_t associativity[] = {cpu_to_be32(0x5),
195                                 cpu_to_be32(0x0),
196                                 cpu_to_be32(0x0),
197                                 cpu_to_be32(0x0),
198                                 cpu_to_be32(cpu->node_id),
199                                 cpu_to_be32(index)};
200 
201     /* Advertise NUMA via ibm,associativity */
202     return fdt_setprop(fdt, offset, "ibm,associativity", associativity,
203                           sizeof(associativity));
204 }
205 
206 /* Populate the "ibm,pa-features" property */
207 static void spapr_populate_pa_features(sPAPRMachineState *spapr,
208                                        PowerPCCPU *cpu,
209                                        void *fdt, int offset,
210                                        bool legacy_guest)
211 {
212     uint8_t pa_features_206[] = { 6, 0,
213         0xf6, 0x1f, 0xc7, 0x00, 0x80, 0xc0 };
214     uint8_t pa_features_207[] = { 24, 0,
215         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0,
216         0x80, 0x00, 0x00, 0x00, 0x00, 0x00,
217         0x00, 0x00, 0x00, 0x00, 0x80, 0x00,
218         0x80, 0x00, 0x80, 0x00, 0x00, 0x00 };
219     uint8_t pa_features_300[] = { 66, 0,
220         /* 0: MMU|FPU|SLB|RUN|DABR|NX, 1: fri[nzpm]|DABRX|SPRG3|SLB0|PP110 */
221         /* 2: VPM|DS205|PPR|DS202|DS206, 3: LSD|URG, SSO, 5: LE|CFAR|EB|LSQ */
222         0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0, /* 0 - 5 */
223         /* 6: DS207 */
224         0x80, 0x00, 0x00, 0x00, 0x00, 0x00, /* 6 - 11 */
225         /* 16: Vector */
226         0x00, 0x00, 0x00, 0x00, 0x80, 0x00, /* 12 - 17 */
227         /* 18: Vec. Scalar, 20: Vec. XOR, 22: HTM */
228         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 18 - 23 */
229         /* 24: Ext. Dec, 26: 64 bit ftrs, 28: PM ftrs */
230         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 24 - 29 */
231         /* 30: MMR, 32: LE atomic, 34: EBB + ext EBB */
232         0x80, 0x00, 0x80, 0x00, 0xC0, 0x00, /* 30 - 35 */
233         /* 36: SPR SO, 38: Copy/Paste, 40: Radix MMU */
234         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 36 - 41 */
235         /* 42: PM, 44: PC RA, 46: SC vec'd */
236         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 42 - 47 */
237         /* 48: SIMD, 50: QP BFP, 52: String */
238         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 48 - 53 */
239         /* 54: DecFP, 56: DecI, 58: SHA */
240         0x80, 0x00, 0x80, 0x00, 0x80, 0x00, /* 54 - 59 */
241         /* 60: NM atomic, 62: RNG */
242         0x80, 0x00, 0x80, 0x00, 0x00, 0x00, /* 60 - 65 */
243     };
244     uint8_t *pa_features = NULL;
245     size_t pa_size;
246 
247     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_06, 0, cpu->compat_pvr)) {
248         pa_features = pa_features_206;
249         pa_size = sizeof(pa_features_206);
250     }
251     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_2_07, 0, cpu->compat_pvr)) {
252         pa_features = pa_features_207;
253         pa_size = sizeof(pa_features_207);
254     }
255     if (ppc_check_compat(cpu, CPU_POWERPC_LOGICAL_3_00, 0, cpu->compat_pvr)) {
256         pa_features = pa_features_300;
257         pa_size = sizeof(pa_features_300);
258     }
259     if (!pa_features) {
260         return;
261     }
262 
263     if (ppc_hash64_has(cpu, PPC_HASH64_CI_LARGEPAGE)) {
264         /*
265          * Note: we keep CI large pages off by default because a 64K capable
266          * guest provisioned with large pages might otherwise try to map a qemu
267          * framebuffer (or other kind of memory mapped PCI BAR) using 64K pages
268          * even if that qemu runs on a 4k host.
269          * We dd this bit back here if we are confident this is not an issue
270          */
271         pa_features[3] |= 0x20;
272     }
273     if ((spapr_get_cap(spapr, SPAPR_CAP_HTM) != 0) && pa_size > 24) {
274         pa_features[24] |= 0x80;    /* Transactional memory support */
275     }
276     if (legacy_guest && pa_size > 40) {
277         /* Workaround for broken kernels that attempt (guest) radix
278          * mode when they can't handle it, if they see the radix bit set
279          * in pa-features. So hide it from them. */
280         pa_features[40 + 2] &= ~0x80; /* Radix MMU */
281     }
282 
283     _FDT((fdt_setprop(fdt, offset, "ibm,pa-features", pa_features, pa_size)));
284 }
285 
286 static int spapr_fixup_cpu_dt(void *fdt, sPAPRMachineState *spapr)
287 {
288     int ret = 0, offset, cpus_offset;
289     CPUState *cs;
290     char cpu_model[32];
291     uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
292 
293     CPU_FOREACH(cs) {
294         PowerPCCPU *cpu = POWERPC_CPU(cs);
295         DeviceClass *dc = DEVICE_GET_CLASS(cs);
296         int index = spapr_get_vcpu_id(cpu);
297         int compat_smt = MIN(smp_threads, ppc_compat_max_vthreads(cpu));
298 
299         if (!spapr_is_thread0_in_vcore(spapr, cpu)) {
300             continue;
301         }
302 
303         snprintf(cpu_model, 32, "%s@%x", dc->fw_name, index);
304 
305         cpus_offset = fdt_path_offset(fdt, "/cpus");
306         if (cpus_offset < 0) {
307             cpus_offset = fdt_add_subnode(fdt, 0, "cpus");
308             if (cpus_offset < 0) {
309                 return cpus_offset;
310             }
311         }
312         offset = fdt_subnode_offset(fdt, cpus_offset, cpu_model);
313         if (offset < 0) {
314             offset = fdt_add_subnode(fdt, cpus_offset, cpu_model);
315             if (offset < 0) {
316                 return offset;
317             }
318         }
319 
320         ret = fdt_setprop(fdt, offset, "ibm,pft-size",
321                           pft_size_prop, sizeof(pft_size_prop));
322         if (ret < 0) {
323             return ret;
324         }
325 
326         if (nb_numa_nodes > 1) {
327             ret = spapr_fixup_cpu_numa_dt(fdt, offset, cpu);
328             if (ret < 0) {
329                 return ret;
330             }
331         }
332 
333         ret = spapr_fixup_cpu_smt_dt(fdt, offset, cpu, compat_smt);
334         if (ret < 0) {
335             return ret;
336         }
337 
338         spapr_populate_pa_features(spapr, cpu, fdt, offset,
339                                    spapr->cas_legacy_guest_workaround);
340     }
341     return ret;
342 }
343 
344 static hwaddr spapr_node0_size(MachineState *machine)
345 {
346     if (nb_numa_nodes) {
347         int i;
348         for (i = 0; i < nb_numa_nodes; ++i) {
349             if (numa_info[i].node_mem) {
350                 return MIN(pow2floor(numa_info[i].node_mem),
351                            machine->ram_size);
352             }
353         }
354     }
355     return machine->ram_size;
356 }
357 
358 static void add_str(GString *s, const gchar *s1)
359 {
360     g_string_append_len(s, s1, strlen(s1) + 1);
361 }
362 
363 static int spapr_populate_memory_node(void *fdt, int nodeid, hwaddr start,
364                                        hwaddr size)
365 {
366     uint32_t associativity[] = {
367         cpu_to_be32(0x4), /* length */
368         cpu_to_be32(0x0), cpu_to_be32(0x0),
369         cpu_to_be32(0x0), cpu_to_be32(nodeid)
370     };
371     char mem_name[32];
372     uint64_t mem_reg_property[2];
373     int off;
374 
375     mem_reg_property[0] = cpu_to_be64(start);
376     mem_reg_property[1] = cpu_to_be64(size);
377 
378     sprintf(mem_name, "memory@" TARGET_FMT_lx, start);
379     off = fdt_add_subnode(fdt, 0, mem_name);
380     _FDT(off);
381     _FDT((fdt_setprop_string(fdt, off, "device_type", "memory")));
382     _FDT((fdt_setprop(fdt, off, "reg", mem_reg_property,
383                       sizeof(mem_reg_property))));
384     _FDT((fdt_setprop(fdt, off, "ibm,associativity", associativity,
385                       sizeof(associativity))));
386     return off;
387 }
388 
389 static int spapr_populate_memory(sPAPRMachineState *spapr, void *fdt)
390 {
391     MachineState *machine = MACHINE(spapr);
392     hwaddr mem_start, node_size;
393     int i, nb_nodes = nb_numa_nodes;
394     NodeInfo *nodes = numa_info;
395     NodeInfo ramnode;
396 
397     /* No NUMA nodes, assume there is just one node with whole RAM */
398     if (!nb_numa_nodes) {
399         nb_nodes = 1;
400         ramnode.node_mem = machine->ram_size;
401         nodes = &ramnode;
402     }
403 
404     for (i = 0, mem_start = 0; i < nb_nodes; ++i) {
405         if (!nodes[i].node_mem) {
406             continue;
407         }
408         if (mem_start >= machine->ram_size) {
409             node_size = 0;
410         } else {
411             node_size = nodes[i].node_mem;
412             if (node_size > machine->ram_size - mem_start) {
413                 node_size = machine->ram_size - mem_start;
414             }
415         }
416         if (!mem_start) {
417             /* spapr_machine_init() checks for rma_size <= node0_size
418              * already */
419             spapr_populate_memory_node(fdt, i, 0, spapr->rma_size);
420             mem_start += spapr->rma_size;
421             node_size -= spapr->rma_size;
422         }
423         for ( ; node_size; ) {
424             hwaddr sizetmp = pow2floor(node_size);
425 
426             /* mem_start != 0 here */
427             if (ctzl(mem_start) < ctzl(sizetmp)) {
428                 sizetmp = 1ULL << ctzl(mem_start);
429             }
430 
431             spapr_populate_memory_node(fdt, i, mem_start, sizetmp);
432             node_size -= sizetmp;
433             mem_start += sizetmp;
434         }
435     }
436 
437     return 0;
438 }
439 
440 static void spapr_populate_cpu_dt(CPUState *cs, void *fdt, int offset,
441                                   sPAPRMachineState *spapr)
442 {
443     PowerPCCPU *cpu = POWERPC_CPU(cs);
444     CPUPPCState *env = &cpu->env;
445     PowerPCCPUClass *pcc = POWERPC_CPU_GET_CLASS(cs);
446     int index = spapr_get_vcpu_id(cpu);
447     uint32_t segs[] = {cpu_to_be32(28), cpu_to_be32(40),
448                        0xffffffff, 0xffffffff};
449     uint32_t tbfreq = kvm_enabled() ? kvmppc_get_tbfreq()
450         : SPAPR_TIMEBASE_FREQ;
451     uint32_t cpufreq = kvm_enabled() ? kvmppc_get_clockfreq() : 1000000000;
452     uint32_t page_sizes_prop[64];
453     size_t page_sizes_prop_size;
454     uint32_t vcpus_per_socket = smp_threads * smp_cores;
455     uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
456     int compat_smt = MIN(smp_threads, ppc_compat_max_vthreads(cpu));
457     sPAPRDRConnector *drc;
458     int drc_index;
459     uint32_t radix_AP_encodings[PPC_PAGE_SIZES_MAX_SZ];
460     int i;
461 
462     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU, index);
463     if (drc) {
464         drc_index = spapr_drc_index(drc);
465         _FDT((fdt_setprop_cell(fdt, offset, "ibm,my-drc-index", drc_index)));
466     }
467 
468     _FDT((fdt_setprop_cell(fdt, offset, "reg", index)));
469     _FDT((fdt_setprop_string(fdt, offset, "device_type", "cpu")));
470 
471     _FDT((fdt_setprop_cell(fdt, offset, "cpu-version", env->spr[SPR_PVR])));
472     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-block-size",
473                            env->dcache_line_size)));
474     _FDT((fdt_setprop_cell(fdt, offset, "d-cache-line-size",
475                            env->dcache_line_size)));
476     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-block-size",
477                            env->icache_line_size)));
478     _FDT((fdt_setprop_cell(fdt, offset, "i-cache-line-size",
479                            env->icache_line_size)));
480 
481     if (pcc->l1_dcache_size) {
482         _FDT((fdt_setprop_cell(fdt, offset, "d-cache-size",
483                                pcc->l1_dcache_size)));
484     } else {
485         warn_report("Unknown L1 dcache size for cpu");
486     }
487     if (pcc->l1_icache_size) {
488         _FDT((fdt_setprop_cell(fdt, offset, "i-cache-size",
489                                pcc->l1_icache_size)));
490     } else {
491         warn_report("Unknown L1 icache size for cpu");
492     }
493 
494     _FDT((fdt_setprop_cell(fdt, offset, "timebase-frequency", tbfreq)));
495     _FDT((fdt_setprop_cell(fdt, offset, "clock-frequency", cpufreq)));
496     _FDT((fdt_setprop_cell(fdt, offset, "slb-size", cpu->hash64_opts->slb_size)));
497     _FDT((fdt_setprop_cell(fdt, offset, "ibm,slb-size", cpu->hash64_opts->slb_size)));
498     _FDT((fdt_setprop_string(fdt, offset, "status", "okay")));
499     _FDT((fdt_setprop(fdt, offset, "64-bit", NULL, 0)));
500 
501     if (env->spr_cb[SPR_PURR].oea_read) {
502         _FDT((fdt_setprop(fdt, offset, "ibm,purr", NULL, 0)));
503     }
504 
505     if (ppc_hash64_has(cpu, PPC_HASH64_1TSEG)) {
506         _FDT((fdt_setprop(fdt, offset, "ibm,processor-segment-sizes",
507                           segs, sizeof(segs))));
508     }
509 
510     /* Advertise VSX (vector extensions) if available
511      *   1               == VMX / Altivec available
512      *   2               == VSX available
513      *
514      * Only CPUs for which we create core types in spapr_cpu_core.c
515      * are possible, and all of those have VMX */
516     if (spapr_get_cap(spapr, SPAPR_CAP_VSX) != 0) {
517         _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 2)));
518     } else {
519         _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", 1)));
520     }
521 
522     /* Advertise DFP (Decimal Floating Point) if available
523      *   0 / no property == no DFP
524      *   1               == DFP available */
525     if (spapr_get_cap(spapr, SPAPR_CAP_DFP) != 0) {
526         _FDT((fdt_setprop_cell(fdt, offset, "ibm,dfp", 1)));
527     }
528 
529     page_sizes_prop_size = ppc_create_page_sizes_prop(cpu, page_sizes_prop,
530                                                       sizeof(page_sizes_prop));
531     if (page_sizes_prop_size) {
532         _FDT((fdt_setprop(fdt, offset, "ibm,segment-page-sizes",
533                           page_sizes_prop, page_sizes_prop_size)));
534     }
535 
536     spapr_populate_pa_features(spapr, cpu, fdt, offset, false);
537 
538     _FDT((fdt_setprop_cell(fdt, offset, "ibm,chip-id",
539                            cs->cpu_index / vcpus_per_socket)));
540 
541     _FDT((fdt_setprop(fdt, offset, "ibm,pft-size",
542                       pft_size_prop, sizeof(pft_size_prop))));
543 
544     if (nb_numa_nodes > 1) {
545         _FDT(spapr_fixup_cpu_numa_dt(fdt, offset, cpu));
546     }
547 
548     _FDT(spapr_fixup_cpu_smt_dt(fdt, offset, cpu, compat_smt));
549 
550     if (pcc->radix_page_info) {
551         for (i = 0; i < pcc->radix_page_info->count; i++) {
552             radix_AP_encodings[i] =
553                 cpu_to_be32(pcc->radix_page_info->entries[i]);
554         }
555         _FDT((fdt_setprop(fdt, offset, "ibm,processor-radix-AP-encodings",
556                           radix_AP_encodings,
557                           pcc->radix_page_info->count *
558                           sizeof(radix_AP_encodings[0]))));
559     }
560 }
561 
562 static void spapr_populate_cpus_dt_node(void *fdt, sPAPRMachineState *spapr)
563 {
564     CPUState **rev;
565     CPUState *cs;
566     int n_cpus;
567     int cpus_offset;
568     char *nodename;
569     int i;
570 
571     cpus_offset = fdt_add_subnode(fdt, 0, "cpus");
572     _FDT(cpus_offset);
573     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#address-cells", 0x1)));
574     _FDT((fdt_setprop_cell(fdt, cpus_offset, "#size-cells", 0x0)));
575 
576     /*
577      * We walk the CPUs in reverse order to ensure that CPU DT nodes
578      * created by fdt_add_subnode() end up in the right order in FDT
579      * for the guest kernel the enumerate the CPUs correctly.
580      *
581      * The CPU list cannot be traversed in reverse order, so we need
582      * to do extra work.
583      */
584     n_cpus = 0;
585     rev = NULL;
586     CPU_FOREACH(cs) {
587         rev = g_renew(CPUState *, rev, n_cpus + 1);
588         rev[n_cpus++] = cs;
589     }
590 
591     for (i = n_cpus - 1; i >= 0; i--) {
592         CPUState *cs = rev[i];
593         PowerPCCPU *cpu = POWERPC_CPU(cs);
594         int index = spapr_get_vcpu_id(cpu);
595         DeviceClass *dc = DEVICE_GET_CLASS(cs);
596         int offset;
597 
598         if (!spapr_is_thread0_in_vcore(spapr, cpu)) {
599             continue;
600         }
601 
602         nodename = g_strdup_printf("%s@%x", dc->fw_name, index);
603         offset = fdt_add_subnode(fdt, cpus_offset, nodename);
604         g_free(nodename);
605         _FDT(offset);
606         spapr_populate_cpu_dt(cs, fdt, offset, spapr);
607     }
608 
609     g_free(rev);
610 }
611 
612 static int spapr_rng_populate_dt(void *fdt)
613 {
614     int node;
615     int ret;
616 
617     node = qemu_fdt_add_subnode(fdt, "/ibm,platform-facilities");
618     if (node <= 0) {
619         return -1;
620     }
621     ret = fdt_setprop_string(fdt, node, "device_type",
622                              "ibm,platform-facilities");
623     ret |= fdt_setprop_cell(fdt, node, "#address-cells", 0x1);
624     ret |= fdt_setprop_cell(fdt, node, "#size-cells", 0x0);
625 
626     node = fdt_add_subnode(fdt, node, "ibm,random-v1");
627     if (node <= 0) {
628         return -1;
629     }
630     ret |= fdt_setprop_string(fdt, node, "compatible", "ibm,random");
631 
632     return ret ? -1 : 0;
633 }
634 
635 static uint32_t spapr_pc_dimm_node(MemoryDeviceInfoList *list, ram_addr_t addr)
636 {
637     MemoryDeviceInfoList *info;
638 
639     for (info = list; info; info = info->next) {
640         MemoryDeviceInfo *value = info->value;
641 
642         if (value && value->type == MEMORY_DEVICE_INFO_KIND_DIMM) {
643             PCDIMMDeviceInfo *pcdimm_info = value->u.dimm.data;
644 
645             if (addr >= pcdimm_info->addr &&
646                 addr < (pcdimm_info->addr + pcdimm_info->size)) {
647                 return pcdimm_info->node;
648             }
649         }
650     }
651 
652     return -1;
653 }
654 
655 struct sPAPRDrconfCellV2 {
656      uint32_t seq_lmbs;
657      uint64_t base_addr;
658      uint32_t drc_index;
659      uint32_t aa_index;
660      uint32_t flags;
661 } QEMU_PACKED;
662 
663 typedef struct DrconfCellQueue {
664     struct sPAPRDrconfCellV2 cell;
665     QSIMPLEQ_ENTRY(DrconfCellQueue) entry;
666 } DrconfCellQueue;
667 
668 static DrconfCellQueue *
669 spapr_get_drconf_cell(uint32_t seq_lmbs, uint64_t base_addr,
670                       uint32_t drc_index, uint32_t aa_index,
671                       uint32_t flags)
672 {
673     DrconfCellQueue *elem;
674 
675     elem = g_malloc0(sizeof(*elem));
676     elem->cell.seq_lmbs = cpu_to_be32(seq_lmbs);
677     elem->cell.base_addr = cpu_to_be64(base_addr);
678     elem->cell.drc_index = cpu_to_be32(drc_index);
679     elem->cell.aa_index = cpu_to_be32(aa_index);
680     elem->cell.flags = cpu_to_be32(flags);
681 
682     return elem;
683 }
684 
685 /* ibm,dynamic-memory-v2 */
686 static int spapr_populate_drmem_v2(sPAPRMachineState *spapr, void *fdt,
687                                    int offset, MemoryDeviceInfoList *dimms)
688 {
689     MachineState *machine = MACHINE(spapr);
690     uint8_t *int_buf, *cur_index;
691     int ret;
692     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
693     uint64_t addr, cur_addr, size;
694     uint32_t nr_boot_lmbs = (machine->device_memory->base / lmb_size);
695     uint64_t mem_end = machine->device_memory->base +
696                        memory_region_size(&machine->device_memory->mr);
697     uint32_t node, buf_len, nr_entries = 0;
698     sPAPRDRConnector *drc;
699     DrconfCellQueue *elem, *next;
700     MemoryDeviceInfoList *info;
701     QSIMPLEQ_HEAD(, DrconfCellQueue) drconf_queue
702         = QSIMPLEQ_HEAD_INITIALIZER(drconf_queue);
703 
704     /* Entry to cover RAM and the gap area */
705     elem = spapr_get_drconf_cell(nr_boot_lmbs, 0, 0, -1,
706                                  SPAPR_LMB_FLAGS_RESERVED |
707                                  SPAPR_LMB_FLAGS_DRC_INVALID);
708     QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
709     nr_entries++;
710 
711     cur_addr = machine->device_memory->base;
712     for (info = dimms; info; info = info->next) {
713         PCDIMMDeviceInfo *di = info->value->u.dimm.data;
714 
715         addr = di->addr;
716         size = di->size;
717         node = di->node;
718 
719         /* Entry for hot-pluggable area */
720         if (cur_addr < addr) {
721             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
722             g_assert(drc);
723             elem = spapr_get_drconf_cell((addr - cur_addr) / lmb_size,
724                                          cur_addr, spapr_drc_index(drc), -1, 0);
725             QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
726             nr_entries++;
727         }
728 
729         /* Entry for DIMM */
730         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, addr / lmb_size);
731         g_assert(drc);
732         elem = spapr_get_drconf_cell(size / lmb_size, addr,
733                                      spapr_drc_index(drc), node,
734                                      SPAPR_LMB_FLAGS_ASSIGNED);
735         QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
736         nr_entries++;
737         cur_addr = addr + size;
738     }
739 
740     /* Entry for remaining hotpluggable area */
741     if (cur_addr < mem_end) {
742         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, cur_addr / lmb_size);
743         g_assert(drc);
744         elem = spapr_get_drconf_cell((mem_end - cur_addr) / lmb_size,
745                                      cur_addr, spapr_drc_index(drc), -1, 0);
746         QSIMPLEQ_INSERT_TAIL(&drconf_queue, elem, entry);
747         nr_entries++;
748     }
749 
750     buf_len = nr_entries * sizeof(struct sPAPRDrconfCellV2) + sizeof(uint32_t);
751     int_buf = cur_index = g_malloc0(buf_len);
752     *(uint32_t *)int_buf = cpu_to_be32(nr_entries);
753     cur_index += sizeof(nr_entries);
754 
755     QSIMPLEQ_FOREACH_SAFE(elem, &drconf_queue, entry, next) {
756         memcpy(cur_index, &elem->cell, sizeof(elem->cell));
757         cur_index += sizeof(elem->cell);
758         QSIMPLEQ_REMOVE(&drconf_queue, elem, DrconfCellQueue, entry);
759         g_free(elem);
760     }
761 
762     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory-v2", int_buf, buf_len);
763     g_free(int_buf);
764     if (ret < 0) {
765         return -1;
766     }
767     return 0;
768 }
769 
770 /* ibm,dynamic-memory */
771 static int spapr_populate_drmem_v1(sPAPRMachineState *spapr, void *fdt,
772                                    int offset, MemoryDeviceInfoList *dimms)
773 {
774     MachineState *machine = MACHINE(spapr);
775     int i, ret;
776     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
777     uint32_t device_lmb_start = machine->device_memory->base / lmb_size;
778     uint32_t nr_lmbs = (machine->device_memory->base +
779                        memory_region_size(&machine->device_memory->mr)) /
780                        lmb_size;
781     uint32_t *int_buf, *cur_index, buf_len;
782 
783     /*
784      * Allocate enough buffer size to fit in ibm,dynamic-memory
785      */
786     buf_len = (nr_lmbs * SPAPR_DR_LMB_LIST_ENTRY_SIZE + 1) * sizeof(uint32_t);
787     cur_index = int_buf = g_malloc0(buf_len);
788     int_buf[0] = cpu_to_be32(nr_lmbs);
789     cur_index++;
790     for (i = 0; i < nr_lmbs; i++) {
791         uint64_t addr = i * lmb_size;
792         uint32_t *dynamic_memory = cur_index;
793 
794         if (i >= device_lmb_start) {
795             sPAPRDRConnector *drc;
796 
797             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB, i);
798             g_assert(drc);
799 
800             dynamic_memory[0] = cpu_to_be32(addr >> 32);
801             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
802             dynamic_memory[2] = cpu_to_be32(spapr_drc_index(drc));
803             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
804             dynamic_memory[4] = cpu_to_be32(spapr_pc_dimm_node(dimms, addr));
805             if (memory_region_present(get_system_memory(), addr)) {
806                 dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_ASSIGNED);
807             } else {
808                 dynamic_memory[5] = cpu_to_be32(0);
809             }
810         } else {
811             /*
812              * LMB information for RMA, boot time RAM and gap b/n RAM and
813              * device memory region -- all these are marked as reserved
814              * and as having no valid DRC.
815              */
816             dynamic_memory[0] = cpu_to_be32(addr >> 32);
817             dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
818             dynamic_memory[2] = cpu_to_be32(0);
819             dynamic_memory[3] = cpu_to_be32(0); /* reserved */
820             dynamic_memory[4] = cpu_to_be32(-1);
821             dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_RESERVED |
822                                             SPAPR_LMB_FLAGS_DRC_INVALID);
823         }
824 
825         cur_index += SPAPR_DR_LMB_LIST_ENTRY_SIZE;
826     }
827     ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory", int_buf, buf_len);
828     g_free(int_buf);
829     if (ret < 0) {
830         return -1;
831     }
832     return 0;
833 }
834 
835 /*
836  * Adds ibm,dynamic-reconfiguration-memory node.
837  * Refer to docs/specs/ppc-spapr-hotplug.txt for the documentation
838  * of this device tree node.
839  */
840 static int spapr_populate_drconf_memory(sPAPRMachineState *spapr, void *fdt)
841 {
842     MachineState *machine = MACHINE(spapr);
843     int ret, i, offset;
844     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
845     uint32_t prop_lmb_size[] = {0, cpu_to_be32(lmb_size)};
846     uint32_t *int_buf, *cur_index, buf_len;
847     int nr_nodes = nb_numa_nodes ? nb_numa_nodes : 1;
848     MemoryDeviceInfoList *dimms = NULL;
849 
850     /*
851      * Don't create the node if there is no device memory
852      */
853     if (machine->ram_size == machine->maxram_size) {
854         return 0;
855     }
856 
857     offset = fdt_add_subnode(fdt, 0, "ibm,dynamic-reconfiguration-memory");
858 
859     ret = fdt_setprop(fdt, offset, "ibm,lmb-size", prop_lmb_size,
860                     sizeof(prop_lmb_size));
861     if (ret < 0) {
862         return ret;
863     }
864 
865     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-flags-mask", 0xff);
866     if (ret < 0) {
867         return ret;
868     }
869 
870     ret = fdt_setprop_cell(fdt, offset, "ibm,memory-preservation-time", 0x0);
871     if (ret < 0) {
872         return ret;
873     }
874 
875     /* ibm,dynamic-memory or ibm,dynamic-memory-v2 */
876     dimms = qmp_memory_device_list();
877     if (spapr_ovec_test(spapr->ov5_cas, OV5_DRMEM_V2)) {
878         ret = spapr_populate_drmem_v2(spapr, fdt, offset, dimms);
879     } else {
880         ret = spapr_populate_drmem_v1(spapr, fdt, offset, dimms);
881     }
882     qapi_free_MemoryDeviceInfoList(dimms);
883 
884     if (ret < 0) {
885         return ret;
886     }
887 
888     /* ibm,associativity-lookup-arrays */
889     buf_len = (nr_nodes * 4 + 2) * sizeof(uint32_t);
890     cur_index = int_buf = g_malloc0(buf_len);
891     int_buf[0] = cpu_to_be32(nr_nodes);
892     int_buf[1] = cpu_to_be32(4); /* Number of entries per associativity list */
893     cur_index += 2;
894     for (i = 0; i < nr_nodes; i++) {
895         uint32_t associativity[] = {
896             cpu_to_be32(0x0),
897             cpu_to_be32(0x0),
898             cpu_to_be32(0x0),
899             cpu_to_be32(i)
900         };
901         memcpy(cur_index, associativity, sizeof(associativity));
902         cur_index += 4;
903     }
904     ret = fdt_setprop(fdt, offset, "ibm,associativity-lookup-arrays", int_buf,
905             (cur_index - int_buf) * sizeof(uint32_t));
906     g_free(int_buf);
907 
908     return ret;
909 }
910 
911 static int spapr_dt_cas_updates(sPAPRMachineState *spapr, void *fdt,
912                                 sPAPROptionVector *ov5_updates)
913 {
914     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
915     int ret = 0, offset;
916 
917     /* Generate ibm,dynamic-reconfiguration-memory node if required */
918     if (spapr_ovec_test(ov5_updates, OV5_DRCONF_MEMORY)) {
919         g_assert(smc->dr_lmb_enabled);
920         ret = spapr_populate_drconf_memory(spapr, fdt);
921         if (ret) {
922             goto out;
923         }
924     }
925 
926     offset = fdt_path_offset(fdt, "/chosen");
927     if (offset < 0) {
928         offset = fdt_add_subnode(fdt, 0, "chosen");
929         if (offset < 0) {
930             return offset;
931         }
932     }
933     ret = spapr_ovec_populate_dt(fdt, offset, spapr->ov5_cas,
934                                  "ibm,architecture-vec-5");
935 
936 out:
937     return ret;
938 }
939 
940 static bool spapr_hotplugged_dev_before_cas(void)
941 {
942     Object *drc_container, *obj;
943     ObjectProperty *prop;
944     ObjectPropertyIterator iter;
945 
946     drc_container = container_get(object_get_root(), "/dr-connector");
947     object_property_iter_init(&iter, drc_container);
948     while ((prop = object_property_iter_next(&iter))) {
949         if (!strstart(prop->type, "link<", NULL)) {
950             continue;
951         }
952         obj = object_property_get_link(drc_container, prop->name, NULL);
953         if (spapr_drc_needed(obj)) {
954             return true;
955         }
956     }
957     return false;
958 }
959 
960 int spapr_h_cas_compose_response(sPAPRMachineState *spapr,
961                                  target_ulong addr, target_ulong size,
962                                  sPAPROptionVector *ov5_updates)
963 {
964     void *fdt, *fdt_skel;
965     sPAPRDeviceTreeUpdateHeader hdr = { .version_id = 1 };
966 
967     if (spapr_hotplugged_dev_before_cas()) {
968         return 1;
969     }
970 
971     if (size < sizeof(hdr) || size > FW_MAX_SIZE) {
972         error_report("SLOF provided an unexpected CAS buffer size "
973                      TARGET_FMT_lu " (min: %zu, max: %u)",
974                      size, sizeof(hdr), FW_MAX_SIZE);
975         exit(EXIT_FAILURE);
976     }
977 
978     size -= sizeof(hdr);
979 
980     /* Create skeleton */
981     fdt_skel = g_malloc0(size);
982     _FDT((fdt_create(fdt_skel, size)));
983     _FDT((fdt_finish_reservemap(fdt_skel)));
984     _FDT((fdt_begin_node(fdt_skel, "")));
985     _FDT((fdt_end_node(fdt_skel)));
986     _FDT((fdt_finish(fdt_skel)));
987     fdt = g_malloc0(size);
988     _FDT((fdt_open_into(fdt_skel, fdt, size)));
989     g_free(fdt_skel);
990 
991     /* Fixup cpu nodes */
992     _FDT((spapr_fixup_cpu_dt(fdt, spapr)));
993 
994     if (spapr_dt_cas_updates(spapr, fdt, ov5_updates)) {
995         return -1;
996     }
997 
998     /* Pack resulting tree */
999     _FDT((fdt_pack(fdt)));
1000 
1001     if (fdt_totalsize(fdt) + sizeof(hdr) > size) {
1002         trace_spapr_cas_failed(size);
1003         return -1;
1004     }
1005 
1006     cpu_physical_memory_write(addr, &hdr, sizeof(hdr));
1007     cpu_physical_memory_write(addr + sizeof(hdr), fdt, fdt_totalsize(fdt));
1008     trace_spapr_cas_continue(fdt_totalsize(fdt) + sizeof(hdr));
1009     g_free(fdt);
1010 
1011     return 0;
1012 }
1013 
1014 static void spapr_dt_rtas(sPAPRMachineState *spapr, void *fdt)
1015 {
1016     int rtas;
1017     GString *hypertas = g_string_sized_new(256);
1018     GString *qemu_hypertas = g_string_sized_new(256);
1019     uint32_t refpoints[] = { cpu_to_be32(0x4), cpu_to_be32(0x4) };
1020     uint64_t max_device_addr = MACHINE(spapr)->device_memory->base +
1021         memory_region_size(&MACHINE(spapr)->device_memory->mr);
1022     uint32_t lrdr_capacity[] = {
1023         cpu_to_be32(max_device_addr >> 32),
1024         cpu_to_be32(max_device_addr & 0xffffffff),
1025         0, cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE),
1026         cpu_to_be32(max_cpus / smp_threads),
1027     };
1028     uint32_t maxdomains[] = {
1029         cpu_to_be32(4),
1030         cpu_to_be32(0),
1031         cpu_to_be32(0),
1032         cpu_to_be32(0),
1033         cpu_to_be32(nb_numa_nodes ? nb_numa_nodes : 1),
1034     };
1035 
1036     _FDT(rtas = fdt_add_subnode(fdt, 0, "rtas"));
1037 
1038     /* hypertas */
1039     add_str(hypertas, "hcall-pft");
1040     add_str(hypertas, "hcall-term");
1041     add_str(hypertas, "hcall-dabr");
1042     add_str(hypertas, "hcall-interrupt");
1043     add_str(hypertas, "hcall-tce");
1044     add_str(hypertas, "hcall-vio");
1045     add_str(hypertas, "hcall-splpar");
1046     add_str(hypertas, "hcall-bulk");
1047     add_str(hypertas, "hcall-set-mode");
1048     add_str(hypertas, "hcall-sprg0");
1049     add_str(hypertas, "hcall-copy");
1050     add_str(hypertas, "hcall-debug");
1051     add_str(hypertas, "hcall-vphn");
1052     add_str(qemu_hypertas, "hcall-memop1");
1053 
1054     if (!kvm_enabled() || kvmppc_spapr_use_multitce()) {
1055         add_str(hypertas, "hcall-multi-tce");
1056     }
1057 
1058     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
1059         add_str(hypertas, "hcall-hpt-resize");
1060     }
1061 
1062     _FDT(fdt_setprop(fdt, rtas, "ibm,hypertas-functions",
1063                      hypertas->str, hypertas->len));
1064     g_string_free(hypertas, TRUE);
1065     _FDT(fdt_setprop(fdt, rtas, "qemu,hypertas-functions",
1066                      qemu_hypertas->str, qemu_hypertas->len));
1067     g_string_free(qemu_hypertas, TRUE);
1068 
1069     _FDT(fdt_setprop(fdt, rtas, "ibm,associativity-reference-points",
1070                      refpoints, sizeof(refpoints)));
1071 
1072     _FDT(fdt_setprop(fdt, rtas, "ibm,max-associativity-domains",
1073                      maxdomains, sizeof(maxdomains)));
1074 
1075     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-error-log-max",
1076                           RTAS_ERROR_LOG_MAX));
1077     _FDT(fdt_setprop_cell(fdt, rtas, "rtas-event-scan-rate",
1078                           RTAS_EVENT_SCAN_RATE));
1079 
1080     g_assert(msi_nonbroken);
1081     _FDT(fdt_setprop(fdt, rtas, "ibm,change-msix-capable", NULL, 0));
1082 
1083     /*
1084      * According to PAPR, rtas ibm,os-term does not guarantee a return
1085      * back to the guest cpu.
1086      *
1087      * While an additional ibm,extended-os-term property indicates
1088      * that rtas call return will always occur. Set this property.
1089      */
1090     _FDT(fdt_setprop(fdt, rtas, "ibm,extended-os-term", NULL, 0));
1091 
1092     _FDT(fdt_setprop(fdt, rtas, "ibm,lrdr-capacity",
1093                      lrdr_capacity, sizeof(lrdr_capacity)));
1094 
1095     spapr_dt_rtas_tokens(fdt, rtas);
1096 }
1097 
1098 /*
1099  * Prepare ibm,arch-vec-5-platform-support, which indicates the MMU
1100  * and the XIVE features that the guest may request and thus the valid
1101  * values for bytes 23..26 of option vector 5:
1102  */
1103 static void spapr_dt_ov5_platform_support(sPAPRMachineState *spapr, void *fdt,
1104                                           int chosen)
1105 {
1106     PowerPCCPU *first_ppc_cpu = POWERPC_CPU(first_cpu);
1107 
1108     char val[2 * 4] = {
1109         23, spapr->irq->ov5, /* Xive mode. */
1110         24, 0x00, /* Hash/Radix, filled in below. */
1111         25, 0x00, /* Hash options: Segment Tables == no, GTSE == no. */
1112         26, 0x40, /* Radix options: GTSE == yes. */
1113     };
1114 
1115     if (!ppc_check_compat(first_ppc_cpu, CPU_POWERPC_LOGICAL_3_00, 0,
1116                           first_ppc_cpu->compat_pvr)) {
1117         /*
1118          * If we're in a pre POWER9 compat mode then the guest should
1119          * do hash and use the legacy interrupt mode
1120          */
1121         val[1] = 0x00; /* XICS */
1122         val[3] = 0x00; /* Hash */
1123     } else if (kvm_enabled()) {
1124         if (kvmppc_has_cap_mmu_radix() && kvmppc_has_cap_mmu_hash_v3()) {
1125             val[3] = 0x80; /* OV5_MMU_BOTH */
1126         } else if (kvmppc_has_cap_mmu_radix()) {
1127             val[3] = 0x40; /* OV5_MMU_RADIX_300 */
1128         } else {
1129             val[3] = 0x00; /* Hash */
1130         }
1131     } else {
1132         /* V3 MMU supports both hash and radix in tcg (with dynamic switching) */
1133         val[3] = 0xC0;
1134     }
1135     _FDT(fdt_setprop(fdt, chosen, "ibm,arch-vec-5-platform-support",
1136                      val, sizeof(val)));
1137 }
1138 
1139 static void spapr_dt_chosen(sPAPRMachineState *spapr, void *fdt)
1140 {
1141     MachineState *machine = MACHINE(spapr);
1142     int chosen;
1143     const char *boot_device = machine->boot_order;
1144     char *stdout_path = spapr_vio_stdout_path(spapr->vio_bus);
1145     size_t cb = 0;
1146     char *bootlist = get_boot_devices_list(&cb);
1147 
1148     _FDT(chosen = fdt_add_subnode(fdt, 0, "chosen"));
1149 
1150     _FDT(fdt_setprop_string(fdt, chosen, "bootargs", machine->kernel_cmdline));
1151     _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-start",
1152                           spapr->initrd_base));
1153     _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-end",
1154                           spapr->initrd_base + spapr->initrd_size));
1155 
1156     if (spapr->kernel_size) {
1157         uint64_t kprop[2] = { cpu_to_be64(KERNEL_LOAD_ADDR),
1158                               cpu_to_be64(spapr->kernel_size) };
1159 
1160         _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel",
1161                          &kprop, sizeof(kprop)));
1162         if (spapr->kernel_le) {
1163             _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel-le", NULL, 0));
1164         }
1165     }
1166     if (boot_menu) {
1167         _FDT((fdt_setprop_cell(fdt, chosen, "qemu,boot-menu", boot_menu)));
1168     }
1169     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-width", graphic_width));
1170     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-height", graphic_height));
1171     _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-depth", graphic_depth));
1172 
1173     if (cb && bootlist) {
1174         int i;
1175 
1176         for (i = 0; i < cb; i++) {
1177             if (bootlist[i] == '\n') {
1178                 bootlist[i] = ' ';
1179             }
1180         }
1181         _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-list", bootlist));
1182     }
1183 
1184     if (boot_device && strlen(boot_device)) {
1185         _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-device", boot_device));
1186     }
1187 
1188     if (!spapr->has_graphics && stdout_path) {
1189         /*
1190          * "linux,stdout-path" and "stdout" properties are deprecated by linux
1191          * kernel. New platforms should only use the "stdout-path" property. Set
1192          * the new property and continue using older property to remain
1193          * compatible with the existing firmware.
1194          */
1195         _FDT(fdt_setprop_string(fdt, chosen, "linux,stdout-path", stdout_path));
1196         _FDT(fdt_setprop_string(fdt, chosen, "stdout-path", stdout_path));
1197     }
1198 
1199     spapr_dt_ov5_platform_support(spapr, fdt, chosen);
1200 
1201     g_free(stdout_path);
1202     g_free(bootlist);
1203 }
1204 
1205 static void spapr_dt_hypervisor(sPAPRMachineState *spapr, void *fdt)
1206 {
1207     /* The /hypervisor node isn't in PAPR - this is a hack to allow PR
1208      * KVM to work under pHyp with some guest co-operation */
1209     int hypervisor;
1210     uint8_t hypercall[16];
1211 
1212     _FDT(hypervisor = fdt_add_subnode(fdt, 0, "hypervisor"));
1213     /* indicate KVM hypercall interface */
1214     _FDT(fdt_setprop_string(fdt, hypervisor, "compatible", "linux,kvm"));
1215     if (kvmppc_has_cap_fixup_hcalls()) {
1216         /*
1217          * Older KVM versions with older guest kernels were broken
1218          * with the magic page, don't allow the guest to map it.
1219          */
1220         if (!kvmppc_get_hypercall(first_cpu->env_ptr, hypercall,
1221                                   sizeof(hypercall))) {
1222             _FDT(fdt_setprop(fdt, hypervisor, "hcall-instructions",
1223                              hypercall, sizeof(hypercall)));
1224         }
1225     }
1226 }
1227 
1228 static void *spapr_build_fdt(sPAPRMachineState *spapr)
1229 {
1230     MachineState *machine = MACHINE(spapr);
1231     MachineClass *mc = MACHINE_GET_CLASS(machine);
1232     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1233     int ret;
1234     void *fdt;
1235     sPAPRPHBState *phb;
1236     char *buf;
1237 
1238     fdt = g_malloc0(FDT_MAX_SIZE);
1239     _FDT((fdt_create_empty_tree(fdt, FDT_MAX_SIZE)));
1240 
1241     /* Root node */
1242     _FDT(fdt_setprop_string(fdt, 0, "device_type", "chrp"));
1243     _FDT(fdt_setprop_string(fdt, 0, "model", "IBM pSeries (emulated by qemu)"));
1244     _FDT(fdt_setprop_string(fdt, 0, "compatible", "qemu,pseries"));
1245 
1246     /*
1247      * Add info to guest to indentify which host is it being run on
1248      * and what is the uuid of the guest
1249      */
1250     if (spapr->host_model && !g_str_equal(spapr->host_model, "none")) {
1251         if (g_str_equal(spapr->host_model, "passthrough")) {
1252             /* -M host-model=passthrough */
1253             if (kvmppc_get_host_model(&buf)) {
1254                 _FDT(fdt_setprop_string(fdt, 0, "host-model", buf));
1255                 g_free(buf);
1256             }
1257         } else {
1258             /* -M host-model=<user-string> */
1259             _FDT(fdt_setprop_string(fdt, 0, "host-model", spapr->host_model));
1260         }
1261     }
1262 
1263     if (spapr->host_serial && !g_str_equal(spapr->host_serial, "none")) {
1264         if (g_str_equal(spapr->host_serial, "passthrough")) {
1265             /* -M host-serial=passthrough */
1266             if (kvmppc_get_host_serial(&buf)) {
1267                 _FDT(fdt_setprop_string(fdt, 0, "host-serial", buf));
1268                 g_free(buf);
1269             }
1270         } else {
1271             /* -M host-serial=<user-string> */
1272             _FDT(fdt_setprop_string(fdt, 0, "host-serial", spapr->host_serial));
1273         }
1274     }
1275 
1276     buf = qemu_uuid_unparse_strdup(&qemu_uuid);
1277 
1278     _FDT(fdt_setprop_string(fdt, 0, "vm,uuid", buf));
1279     if (qemu_uuid_set) {
1280         _FDT(fdt_setprop_string(fdt, 0, "system-id", buf));
1281     }
1282     g_free(buf);
1283 
1284     if (qemu_get_vm_name()) {
1285         _FDT(fdt_setprop_string(fdt, 0, "ibm,partition-name",
1286                                 qemu_get_vm_name()));
1287     }
1288 
1289     _FDT(fdt_setprop_cell(fdt, 0, "#address-cells", 2));
1290     _FDT(fdt_setprop_cell(fdt, 0, "#size-cells", 2));
1291 
1292     /* /interrupt controller */
1293     spapr->irq->dt_populate(spapr, spapr_max_server_number(spapr), fdt,
1294                           PHANDLE_INTC);
1295 
1296     ret = spapr_populate_memory(spapr, fdt);
1297     if (ret < 0) {
1298         error_report("couldn't setup memory nodes in fdt");
1299         exit(1);
1300     }
1301 
1302     /* /vdevice */
1303     spapr_dt_vdevice(spapr->vio_bus, fdt);
1304 
1305     if (object_resolve_path_type("", TYPE_SPAPR_RNG, NULL)) {
1306         ret = spapr_rng_populate_dt(fdt);
1307         if (ret < 0) {
1308             error_report("could not set up rng device in the fdt");
1309             exit(1);
1310         }
1311     }
1312 
1313     QLIST_FOREACH(phb, &spapr->phbs, list) {
1314         ret = spapr_populate_pci_dt(phb, PHANDLE_INTC, fdt,
1315                                     spapr->irq->nr_msis);
1316         if (ret < 0) {
1317             error_report("couldn't setup PCI devices in fdt");
1318             exit(1);
1319         }
1320     }
1321 
1322     /* cpus */
1323     spapr_populate_cpus_dt_node(fdt, spapr);
1324 
1325     if (smc->dr_lmb_enabled) {
1326         _FDT(spapr_drc_populate_dt(fdt, 0, NULL, SPAPR_DR_CONNECTOR_TYPE_LMB));
1327     }
1328 
1329     if (mc->has_hotpluggable_cpus) {
1330         int offset = fdt_path_offset(fdt, "/cpus");
1331         ret = spapr_drc_populate_dt(fdt, offset, NULL,
1332                                     SPAPR_DR_CONNECTOR_TYPE_CPU);
1333         if (ret < 0) {
1334             error_report("Couldn't set up CPU DR device tree properties");
1335             exit(1);
1336         }
1337     }
1338 
1339     /* /event-sources */
1340     spapr_dt_events(spapr, fdt);
1341 
1342     /* /rtas */
1343     spapr_dt_rtas(spapr, fdt);
1344 
1345     /* /chosen */
1346     spapr_dt_chosen(spapr, fdt);
1347 
1348     /* /hypervisor */
1349     if (kvm_enabled()) {
1350         spapr_dt_hypervisor(spapr, fdt);
1351     }
1352 
1353     /* Build memory reserve map */
1354     if (spapr->kernel_size) {
1355         _FDT((fdt_add_mem_rsv(fdt, KERNEL_LOAD_ADDR, spapr->kernel_size)));
1356     }
1357     if (spapr->initrd_size) {
1358         _FDT((fdt_add_mem_rsv(fdt, spapr->initrd_base, spapr->initrd_size)));
1359     }
1360 
1361     /* ibm,client-architecture-support updates */
1362     ret = spapr_dt_cas_updates(spapr, fdt, spapr->ov5_cas);
1363     if (ret < 0) {
1364         error_report("couldn't setup CAS properties fdt");
1365         exit(1);
1366     }
1367 
1368     return fdt;
1369 }
1370 
1371 static uint64_t translate_kernel_address(void *opaque, uint64_t addr)
1372 {
1373     return (addr & 0x0fffffff) + KERNEL_LOAD_ADDR;
1374 }
1375 
1376 static void emulate_spapr_hypercall(PPCVirtualHypervisor *vhyp,
1377                                     PowerPCCPU *cpu)
1378 {
1379     CPUPPCState *env = &cpu->env;
1380 
1381     /* The TCG path should also be holding the BQL at this point */
1382     g_assert(qemu_mutex_iothread_locked());
1383 
1384     if (msr_pr) {
1385         hcall_dprintf("Hypercall made with MSR[PR]=1\n");
1386         env->gpr[3] = H_PRIVILEGE;
1387     } else {
1388         env->gpr[3] = spapr_hypercall(cpu, env->gpr[3], &env->gpr[4]);
1389     }
1390 }
1391 
1392 struct LPCRSyncState {
1393     target_ulong value;
1394     target_ulong mask;
1395 };
1396 
1397 static void do_lpcr_sync(CPUState *cs, run_on_cpu_data arg)
1398 {
1399     struct LPCRSyncState *s = arg.host_ptr;
1400     PowerPCCPU *cpu = POWERPC_CPU(cs);
1401     CPUPPCState *env = &cpu->env;
1402     target_ulong lpcr;
1403 
1404     cpu_synchronize_state(cs);
1405     lpcr = env->spr[SPR_LPCR];
1406     lpcr &= ~s->mask;
1407     lpcr |= s->value;
1408     ppc_store_lpcr(cpu, lpcr);
1409 }
1410 
1411 void spapr_set_all_lpcrs(target_ulong value, target_ulong mask)
1412 {
1413     CPUState *cs;
1414     struct LPCRSyncState s = {
1415         .value = value,
1416         .mask = mask
1417     };
1418     CPU_FOREACH(cs) {
1419         run_on_cpu(cs, do_lpcr_sync, RUN_ON_CPU_HOST_PTR(&s));
1420     }
1421 }
1422 
1423 static uint64_t spapr_get_patbe(PPCVirtualHypervisor *vhyp)
1424 {
1425     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1426 
1427     return spapr->patb_entry;
1428 }
1429 
1430 #define HPTE(_table, _i)   (void *)(((uint64_t *)(_table)) + ((_i) * 2))
1431 #define HPTE_VALID(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_VALID)
1432 #define HPTE_DIRTY(_hpte)  (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_HPTE_DIRTY)
1433 #define CLEAN_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) &= tswap64(~HPTE64_V_HPTE_DIRTY))
1434 #define DIRTY_HPTE(_hpte)  ((*(uint64_t *)(_hpte)) |= tswap64(HPTE64_V_HPTE_DIRTY))
1435 
1436 /*
1437  * Get the fd to access the kernel htab, re-opening it if necessary
1438  */
1439 static int get_htab_fd(sPAPRMachineState *spapr)
1440 {
1441     Error *local_err = NULL;
1442 
1443     if (spapr->htab_fd >= 0) {
1444         return spapr->htab_fd;
1445     }
1446 
1447     spapr->htab_fd = kvmppc_get_htab_fd(false, 0, &local_err);
1448     if (spapr->htab_fd < 0) {
1449         error_report_err(local_err);
1450     }
1451 
1452     return spapr->htab_fd;
1453 }
1454 
1455 void close_htab_fd(sPAPRMachineState *spapr)
1456 {
1457     if (spapr->htab_fd >= 0) {
1458         close(spapr->htab_fd);
1459     }
1460     spapr->htab_fd = -1;
1461 }
1462 
1463 static hwaddr spapr_hpt_mask(PPCVirtualHypervisor *vhyp)
1464 {
1465     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1466 
1467     return HTAB_SIZE(spapr) / HASH_PTEG_SIZE_64 - 1;
1468 }
1469 
1470 static target_ulong spapr_encode_hpt_for_kvm_pr(PPCVirtualHypervisor *vhyp)
1471 {
1472     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1473 
1474     assert(kvm_enabled());
1475 
1476     if (!spapr->htab) {
1477         return 0;
1478     }
1479 
1480     return (target_ulong)(uintptr_t)spapr->htab | (spapr->htab_shift - 18);
1481 }
1482 
1483 static const ppc_hash_pte64_t *spapr_map_hptes(PPCVirtualHypervisor *vhyp,
1484                                                 hwaddr ptex, int n)
1485 {
1486     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1487     hwaddr pte_offset = ptex * HASH_PTE_SIZE_64;
1488 
1489     if (!spapr->htab) {
1490         /*
1491          * HTAB is controlled by KVM. Fetch into temporary buffer
1492          */
1493         ppc_hash_pte64_t *hptes = g_malloc(n * HASH_PTE_SIZE_64);
1494         kvmppc_read_hptes(hptes, ptex, n);
1495         return hptes;
1496     }
1497 
1498     /*
1499      * HTAB is controlled by QEMU. Just point to the internally
1500      * accessible PTEG.
1501      */
1502     return (const ppc_hash_pte64_t *)(spapr->htab + pte_offset);
1503 }
1504 
1505 static void spapr_unmap_hptes(PPCVirtualHypervisor *vhyp,
1506                               const ppc_hash_pte64_t *hptes,
1507                               hwaddr ptex, int n)
1508 {
1509     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1510 
1511     if (!spapr->htab) {
1512         g_free((void *)hptes);
1513     }
1514 
1515     /* Nothing to do for qemu managed HPT */
1516 }
1517 
1518 static void spapr_store_hpte(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1519                              uint64_t pte0, uint64_t pte1)
1520 {
1521     sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1522     hwaddr offset = ptex * HASH_PTE_SIZE_64;
1523 
1524     if (!spapr->htab) {
1525         kvmppc_write_hpte(ptex, pte0, pte1);
1526     } else {
1527         if (pte0 & HPTE64_V_VALID) {
1528             stq_p(spapr->htab + offset + HASH_PTE_SIZE_64 / 2, pte1);
1529             /*
1530              * When setting valid, we write PTE1 first. This ensures
1531              * proper synchronization with the reading code in
1532              * ppc_hash64_pteg_search()
1533              */
1534             smp_wmb();
1535             stq_p(spapr->htab + offset, pte0);
1536         } else {
1537             stq_p(spapr->htab + offset, pte0);
1538             /*
1539              * When clearing it we set PTE0 first. This ensures proper
1540              * synchronization with the reading code in
1541              * ppc_hash64_pteg_search()
1542              */
1543             smp_wmb();
1544             stq_p(spapr->htab + offset + HASH_PTE_SIZE_64 / 2, pte1);
1545         }
1546     }
1547 }
1548 
1549 int spapr_hpt_shift_for_ramsize(uint64_t ramsize)
1550 {
1551     int shift;
1552 
1553     /* We aim for a hash table of size 1/128 the size of RAM (rounded
1554      * up).  The PAPR recommendation is actually 1/64 of RAM size, but
1555      * that's much more than is needed for Linux guests */
1556     shift = ctz64(pow2ceil(ramsize)) - 7;
1557     shift = MAX(shift, 18); /* Minimum architected size */
1558     shift = MIN(shift, 46); /* Maximum architected size */
1559     return shift;
1560 }
1561 
1562 void spapr_free_hpt(sPAPRMachineState *spapr)
1563 {
1564     g_free(spapr->htab);
1565     spapr->htab = NULL;
1566     spapr->htab_shift = 0;
1567     close_htab_fd(spapr);
1568 }
1569 
1570 void spapr_reallocate_hpt(sPAPRMachineState *spapr, int shift,
1571                           Error **errp)
1572 {
1573     long rc;
1574 
1575     /* Clean up any HPT info from a previous boot */
1576     spapr_free_hpt(spapr);
1577 
1578     rc = kvmppc_reset_htab(shift);
1579     if (rc < 0) {
1580         /* kernel-side HPT needed, but couldn't allocate one */
1581         error_setg_errno(errp, errno,
1582                          "Failed to allocate KVM HPT of order %d (try smaller maxmem?)",
1583                          shift);
1584         /* This is almost certainly fatal, but if the caller really
1585          * wants to carry on with shift == 0, it's welcome to try */
1586     } else if (rc > 0) {
1587         /* kernel-side HPT allocated */
1588         if (rc != shift) {
1589             error_setg(errp,
1590                        "Requested order %d HPT, but kernel allocated order %ld (try smaller maxmem?)",
1591                        shift, rc);
1592         }
1593 
1594         spapr->htab_shift = shift;
1595         spapr->htab = NULL;
1596     } else {
1597         /* kernel-side HPT not needed, allocate in userspace instead */
1598         size_t size = 1ULL << shift;
1599         int i;
1600 
1601         spapr->htab = qemu_memalign(size, size);
1602         if (!spapr->htab) {
1603             error_setg_errno(errp, errno,
1604                              "Could not allocate HPT of order %d", shift);
1605             return;
1606         }
1607 
1608         memset(spapr->htab, 0, size);
1609         spapr->htab_shift = shift;
1610 
1611         for (i = 0; i < size / HASH_PTE_SIZE_64; i++) {
1612             DIRTY_HPTE(HPTE(spapr->htab, i));
1613         }
1614     }
1615     /* We're setting up a hash table, so that means we're not radix */
1616     spapr_set_all_lpcrs(0, LPCR_HR | LPCR_UPRT);
1617 }
1618 
1619 void spapr_setup_hpt_and_vrma(sPAPRMachineState *spapr)
1620 {
1621     int hpt_shift;
1622 
1623     if ((spapr->resize_hpt == SPAPR_RESIZE_HPT_DISABLED)
1624         || (spapr->cas_reboot
1625             && !spapr_ovec_test(spapr->ov5_cas, OV5_HPT_RESIZE))) {
1626         hpt_shift = spapr_hpt_shift_for_ramsize(MACHINE(spapr)->maxram_size);
1627     } else {
1628         uint64_t current_ram_size;
1629 
1630         current_ram_size = MACHINE(spapr)->ram_size + get_plugged_memory_size();
1631         hpt_shift = spapr_hpt_shift_for_ramsize(current_ram_size);
1632     }
1633     spapr_reallocate_hpt(spapr, hpt_shift, &error_fatal);
1634 
1635     if (spapr->vrma_adjust) {
1636         spapr->rma_size = kvmppc_rma_size(spapr_node0_size(MACHINE(spapr)),
1637                                           spapr->htab_shift);
1638     }
1639 }
1640 
1641 static int spapr_reset_drcs(Object *child, void *opaque)
1642 {
1643     sPAPRDRConnector *drc =
1644         (sPAPRDRConnector *) object_dynamic_cast(child,
1645                                                  TYPE_SPAPR_DR_CONNECTOR);
1646 
1647     if (drc) {
1648         spapr_drc_reset(drc);
1649     }
1650 
1651     return 0;
1652 }
1653 
1654 static void spapr_machine_reset(void)
1655 {
1656     MachineState *machine = MACHINE(qdev_get_machine());
1657     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
1658     PowerPCCPU *first_ppc_cpu;
1659     uint32_t rtas_limit;
1660     hwaddr rtas_addr, fdt_addr;
1661     void *fdt;
1662     int rc;
1663 
1664     spapr_caps_apply(spapr);
1665 
1666     first_ppc_cpu = POWERPC_CPU(first_cpu);
1667     if (kvm_enabled() && kvmppc_has_cap_mmu_radix() &&
1668         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
1669                               spapr->max_compat_pvr)) {
1670         /* If using KVM with radix mode available, VCPUs can be started
1671          * without a HPT because KVM will start them in radix mode.
1672          * Set the GR bit in PATB so that we know there is no HPT. */
1673         spapr->patb_entry = PATBE1_GR;
1674         spapr_set_all_lpcrs(LPCR_HR | LPCR_UPRT, LPCR_HR | LPCR_UPRT);
1675     } else {
1676         spapr_setup_hpt_and_vrma(spapr);
1677     }
1678 
1679     /* if this reset wasn't generated by CAS, we should reset our
1680      * negotiated options and start from scratch */
1681     if (!spapr->cas_reboot) {
1682         spapr_ovec_cleanup(spapr->ov5_cas);
1683         spapr->ov5_cas = spapr_ovec_new();
1684 
1685         ppc_set_compat(first_ppc_cpu, spapr->max_compat_pvr, &error_fatal);
1686     }
1687 
1688     if (!SPAPR_MACHINE_GET_CLASS(spapr)->legacy_irq_allocation) {
1689         spapr_irq_msi_reset(spapr);
1690     }
1691 
1692     qemu_devices_reset();
1693 
1694     /*
1695      * This is fixing some of the default configuration of the XIVE
1696      * devices. To be called after the reset of the machine devices.
1697      */
1698     spapr_irq_reset(spapr, &error_fatal);
1699 
1700     /* DRC reset may cause a device to be unplugged. This will cause troubles
1701      * if this device is used by another device (eg, a running vhost backend
1702      * will crash QEMU if the DIMM holding the vring goes away). To avoid such
1703      * situations, we reset DRCs after all devices have been reset.
1704      */
1705     object_child_foreach_recursive(object_get_root(), spapr_reset_drcs, NULL);
1706 
1707     spapr_clear_pending_events(spapr);
1708 
1709     /*
1710      * We place the device tree and RTAS just below either the top of the RMA,
1711      * or just below 2GB, whichever is lower, so that it can be
1712      * processed with 32-bit real mode code if necessary
1713      */
1714     rtas_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR);
1715     rtas_addr = rtas_limit - RTAS_MAX_SIZE;
1716     fdt_addr = rtas_addr - FDT_MAX_SIZE;
1717 
1718     fdt = spapr_build_fdt(spapr);
1719 
1720     spapr_load_rtas(spapr, fdt, rtas_addr);
1721 
1722     rc = fdt_pack(fdt);
1723 
1724     /* Should only fail if we've built a corrupted tree */
1725     assert(rc == 0);
1726 
1727     if (fdt_totalsize(fdt) > FDT_MAX_SIZE) {
1728         error_report("FDT too big ! 0x%x bytes (max is 0x%x)",
1729                      fdt_totalsize(fdt), FDT_MAX_SIZE);
1730         exit(1);
1731     }
1732 
1733     /* Load the fdt */
1734     qemu_fdt_dumpdtb(fdt, fdt_totalsize(fdt));
1735     cpu_physical_memory_write(fdt_addr, fdt, fdt_totalsize(fdt));
1736     g_free(spapr->fdt_blob);
1737     spapr->fdt_size = fdt_totalsize(fdt);
1738     spapr->fdt_initial_size = spapr->fdt_size;
1739     spapr->fdt_blob = fdt;
1740 
1741     /* Set up the entry state */
1742     spapr_cpu_set_entry_state(first_ppc_cpu, SPAPR_ENTRY_POINT, fdt_addr);
1743     first_ppc_cpu->env.gpr[5] = 0;
1744 
1745     spapr->cas_reboot = false;
1746 }
1747 
1748 static void spapr_create_nvram(sPAPRMachineState *spapr)
1749 {
1750     DeviceState *dev = qdev_create(&spapr->vio_bus->bus, "spapr-nvram");
1751     DriveInfo *dinfo = drive_get(IF_PFLASH, 0, 0);
1752 
1753     if (dinfo) {
1754         qdev_prop_set_drive(dev, "drive", blk_by_legacy_dinfo(dinfo),
1755                             &error_fatal);
1756     }
1757 
1758     qdev_init_nofail(dev);
1759 
1760     spapr->nvram = (struct sPAPRNVRAM *)dev;
1761 }
1762 
1763 static void spapr_rtc_create(sPAPRMachineState *spapr)
1764 {
1765     object_initialize(&spapr->rtc, sizeof(spapr->rtc), TYPE_SPAPR_RTC);
1766     object_property_add_child(OBJECT(spapr), "rtc", OBJECT(&spapr->rtc),
1767                               &error_fatal);
1768     object_property_set_bool(OBJECT(&spapr->rtc), true, "realized",
1769                               &error_fatal);
1770     object_property_add_alias(OBJECT(spapr), "rtc-time", OBJECT(&spapr->rtc),
1771                               "date", &error_fatal);
1772 }
1773 
1774 /* Returns whether we want to use VGA or not */
1775 static bool spapr_vga_init(PCIBus *pci_bus, Error **errp)
1776 {
1777     switch (vga_interface_type) {
1778     case VGA_NONE:
1779         return false;
1780     case VGA_DEVICE:
1781         return true;
1782     case VGA_STD:
1783     case VGA_VIRTIO:
1784     case VGA_CIRRUS:
1785         return pci_vga_init(pci_bus) != NULL;
1786     default:
1787         error_setg(errp,
1788                    "Unsupported VGA mode, only -vga std or -vga virtio is supported");
1789         return false;
1790     }
1791 }
1792 
1793 static int spapr_pre_load(void *opaque)
1794 {
1795     int rc;
1796 
1797     rc = spapr_caps_pre_load(opaque);
1798     if (rc) {
1799         return rc;
1800     }
1801 
1802     return 0;
1803 }
1804 
1805 static int spapr_post_load(void *opaque, int version_id)
1806 {
1807     sPAPRMachineState *spapr = (sPAPRMachineState *)opaque;
1808     int err = 0;
1809 
1810     err = spapr_caps_post_migration(spapr);
1811     if (err) {
1812         return err;
1813     }
1814 
1815     /*
1816      * In earlier versions, there was no separate qdev for the PAPR
1817      * RTC, so the RTC offset was stored directly in sPAPREnvironment.
1818      * So when migrating from those versions, poke the incoming offset
1819      * value into the RTC device
1820      */
1821     if (version_id < 3) {
1822         err = spapr_rtc_import_offset(&spapr->rtc, spapr->rtc_offset);
1823         if (err) {
1824             return err;
1825         }
1826     }
1827 
1828     if (kvm_enabled() && spapr->patb_entry) {
1829         PowerPCCPU *cpu = POWERPC_CPU(first_cpu);
1830         bool radix = !!(spapr->patb_entry & PATBE1_GR);
1831         bool gtse = !!(cpu->env.spr[SPR_LPCR] & LPCR_GTSE);
1832 
1833         /*
1834          * Update LPCR:HR and UPRT as they may not be set properly in
1835          * the stream
1836          */
1837         spapr_set_all_lpcrs(radix ? (LPCR_HR | LPCR_UPRT) : 0,
1838                             LPCR_HR | LPCR_UPRT);
1839 
1840         err = kvmppc_configure_v3_mmu(cpu, radix, gtse, spapr->patb_entry);
1841         if (err) {
1842             error_report("Process table config unsupported by the host");
1843             return -EINVAL;
1844         }
1845     }
1846 
1847     err = spapr_irq_post_load(spapr, version_id);
1848     if (err) {
1849         return err;
1850     }
1851 
1852     return err;
1853 }
1854 
1855 static int spapr_pre_save(void *opaque)
1856 {
1857     int rc;
1858 
1859     rc = spapr_caps_pre_save(opaque);
1860     if (rc) {
1861         return rc;
1862     }
1863 
1864     return 0;
1865 }
1866 
1867 static bool version_before_3(void *opaque, int version_id)
1868 {
1869     return version_id < 3;
1870 }
1871 
1872 static bool spapr_pending_events_needed(void *opaque)
1873 {
1874     sPAPRMachineState *spapr = (sPAPRMachineState *)opaque;
1875     return !QTAILQ_EMPTY(&spapr->pending_events);
1876 }
1877 
1878 static const VMStateDescription vmstate_spapr_event_entry = {
1879     .name = "spapr_event_log_entry",
1880     .version_id = 1,
1881     .minimum_version_id = 1,
1882     .fields = (VMStateField[]) {
1883         VMSTATE_UINT32(summary, sPAPREventLogEntry),
1884         VMSTATE_UINT32(extended_length, sPAPREventLogEntry),
1885         VMSTATE_VBUFFER_ALLOC_UINT32(extended_log, sPAPREventLogEntry, 0,
1886                                      NULL, extended_length),
1887         VMSTATE_END_OF_LIST()
1888     },
1889 };
1890 
1891 static const VMStateDescription vmstate_spapr_pending_events = {
1892     .name = "spapr_pending_events",
1893     .version_id = 1,
1894     .minimum_version_id = 1,
1895     .needed = spapr_pending_events_needed,
1896     .fields = (VMStateField[]) {
1897         VMSTATE_QTAILQ_V(pending_events, sPAPRMachineState, 1,
1898                          vmstate_spapr_event_entry, sPAPREventLogEntry, next),
1899         VMSTATE_END_OF_LIST()
1900     },
1901 };
1902 
1903 static bool spapr_ov5_cas_needed(void *opaque)
1904 {
1905     sPAPRMachineState *spapr = opaque;
1906     sPAPROptionVector *ov5_mask = spapr_ovec_new();
1907     sPAPROptionVector *ov5_legacy = spapr_ovec_new();
1908     sPAPROptionVector *ov5_removed = spapr_ovec_new();
1909     bool cas_needed;
1910 
1911     /* Prior to the introduction of sPAPROptionVector, we had two option
1912      * vectors we dealt with: OV5_FORM1_AFFINITY, and OV5_DRCONF_MEMORY.
1913      * Both of these options encode machine topology into the device-tree
1914      * in such a way that the now-booted OS should still be able to interact
1915      * appropriately with QEMU regardless of what options were actually
1916      * negotiatied on the source side.
1917      *
1918      * As such, we can avoid migrating the CAS-negotiated options if these
1919      * are the only options available on the current machine/platform.
1920      * Since these are the only options available for pseries-2.7 and
1921      * earlier, this allows us to maintain old->new/new->old migration
1922      * compatibility.
1923      *
1924      * For QEMU 2.8+, there are additional CAS-negotiatable options available
1925      * via default pseries-2.8 machines and explicit command-line parameters.
1926      * Some of these options, like OV5_HP_EVT, *do* require QEMU to be aware
1927      * of the actual CAS-negotiated values to continue working properly. For
1928      * example, availability of memory unplug depends on knowing whether
1929      * OV5_HP_EVT was negotiated via CAS.
1930      *
1931      * Thus, for any cases where the set of available CAS-negotiatable
1932      * options extends beyond OV5_FORM1_AFFINITY and OV5_DRCONF_MEMORY, we
1933      * include the CAS-negotiated options in the migration stream, unless
1934      * if they affect boot time behaviour only.
1935      */
1936     spapr_ovec_set(ov5_mask, OV5_FORM1_AFFINITY);
1937     spapr_ovec_set(ov5_mask, OV5_DRCONF_MEMORY);
1938     spapr_ovec_set(ov5_mask, OV5_DRMEM_V2);
1939 
1940     /* spapr_ovec_diff returns true if bits were removed. we avoid using
1941      * the mask itself since in the future it's possible "legacy" bits may be
1942      * removed via machine options, which could generate a false positive
1943      * that breaks migration.
1944      */
1945     spapr_ovec_intersect(ov5_legacy, spapr->ov5, ov5_mask);
1946     cas_needed = spapr_ovec_diff(ov5_removed, spapr->ov5, ov5_legacy);
1947 
1948     spapr_ovec_cleanup(ov5_mask);
1949     spapr_ovec_cleanup(ov5_legacy);
1950     spapr_ovec_cleanup(ov5_removed);
1951 
1952     return cas_needed;
1953 }
1954 
1955 static const VMStateDescription vmstate_spapr_ov5_cas = {
1956     .name = "spapr_option_vector_ov5_cas",
1957     .version_id = 1,
1958     .minimum_version_id = 1,
1959     .needed = spapr_ov5_cas_needed,
1960     .fields = (VMStateField[]) {
1961         VMSTATE_STRUCT_POINTER_V(ov5_cas, sPAPRMachineState, 1,
1962                                  vmstate_spapr_ovec, sPAPROptionVector),
1963         VMSTATE_END_OF_LIST()
1964     },
1965 };
1966 
1967 static bool spapr_patb_entry_needed(void *opaque)
1968 {
1969     sPAPRMachineState *spapr = opaque;
1970 
1971     return !!spapr->patb_entry;
1972 }
1973 
1974 static const VMStateDescription vmstate_spapr_patb_entry = {
1975     .name = "spapr_patb_entry",
1976     .version_id = 1,
1977     .minimum_version_id = 1,
1978     .needed = spapr_patb_entry_needed,
1979     .fields = (VMStateField[]) {
1980         VMSTATE_UINT64(patb_entry, sPAPRMachineState),
1981         VMSTATE_END_OF_LIST()
1982     },
1983 };
1984 
1985 static bool spapr_irq_map_needed(void *opaque)
1986 {
1987     sPAPRMachineState *spapr = opaque;
1988 
1989     return spapr->irq_map && !bitmap_empty(spapr->irq_map, spapr->irq_map_nr);
1990 }
1991 
1992 static const VMStateDescription vmstate_spapr_irq_map = {
1993     .name = "spapr_irq_map",
1994     .version_id = 1,
1995     .minimum_version_id = 1,
1996     .needed = spapr_irq_map_needed,
1997     .fields = (VMStateField[]) {
1998         VMSTATE_BITMAP(irq_map, sPAPRMachineState, 0, irq_map_nr),
1999         VMSTATE_END_OF_LIST()
2000     },
2001 };
2002 
2003 static bool spapr_dtb_needed(void *opaque)
2004 {
2005     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(opaque);
2006 
2007     return smc->update_dt_enabled;
2008 }
2009 
2010 static int spapr_dtb_pre_load(void *opaque)
2011 {
2012     sPAPRMachineState *spapr = (sPAPRMachineState *)opaque;
2013 
2014     g_free(spapr->fdt_blob);
2015     spapr->fdt_blob = NULL;
2016     spapr->fdt_size = 0;
2017 
2018     return 0;
2019 }
2020 
2021 static const VMStateDescription vmstate_spapr_dtb = {
2022     .name = "spapr_dtb",
2023     .version_id = 1,
2024     .minimum_version_id = 1,
2025     .needed = spapr_dtb_needed,
2026     .pre_load = spapr_dtb_pre_load,
2027     .fields = (VMStateField[]) {
2028         VMSTATE_UINT32(fdt_initial_size, sPAPRMachineState),
2029         VMSTATE_UINT32(fdt_size, sPAPRMachineState),
2030         VMSTATE_VBUFFER_ALLOC_UINT32(fdt_blob, sPAPRMachineState, 0, NULL,
2031                                      fdt_size),
2032         VMSTATE_END_OF_LIST()
2033     },
2034 };
2035 
2036 static const VMStateDescription vmstate_spapr = {
2037     .name = "spapr",
2038     .version_id = 3,
2039     .minimum_version_id = 1,
2040     .pre_load = spapr_pre_load,
2041     .post_load = spapr_post_load,
2042     .pre_save = spapr_pre_save,
2043     .fields = (VMStateField[]) {
2044         /* used to be @next_irq */
2045         VMSTATE_UNUSED_BUFFER(version_before_3, 0, 4),
2046 
2047         /* RTC offset */
2048         VMSTATE_UINT64_TEST(rtc_offset, sPAPRMachineState, version_before_3),
2049 
2050         VMSTATE_PPC_TIMEBASE_V(tb, sPAPRMachineState, 2),
2051         VMSTATE_END_OF_LIST()
2052     },
2053     .subsections = (const VMStateDescription*[]) {
2054         &vmstate_spapr_ov5_cas,
2055         &vmstate_spapr_patb_entry,
2056         &vmstate_spapr_pending_events,
2057         &vmstate_spapr_cap_htm,
2058         &vmstate_spapr_cap_vsx,
2059         &vmstate_spapr_cap_dfp,
2060         &vmstate_spapr_cap_cfpc,
2061         &vmstate_spapr_cap_sbbc,
2062         &vmstate_spapr_cap_ibs,
2063         &vmstate_spapr_irq_map,
2064         &vmstate_spapr_cap_nested_kvm_hv,
2065         &vmstate_spapr_dtb,
2066         NULL
2067     }
2068 };
2069 
2070 static int htab_save_setup(QEMUFile *f, void *opaque)
2071 {
2072     sPAPRMachineState *spapr = opaque;
2073 
2074     /* "Iteration" header */
2075     if (!spapr->htab_shift) {
2076         qemu_put_be32(f, -1);
2077     } else {
2078         qemu_put_be32(f, spapr->htab_shift);
2079     }
2080 
2081     if (spapr->htab) {
2082         spapr->htab_save_index = 0;
2083         spapr->htab_first_pass = true;
2084     } else {
2085         if (spapr->htab_shift) {
2086             assert(kvm_enabled());
2087         }
2088     }
2089 
2090 
2091     return 0;
2092 }
2093 
2094 static void htab_save_chunk(QEMUFile *f, sPAPRMachineState *spapr,
2095                             int chunkstart, int n_valid, int n_invalid)
2096 {
2097     qemu_put_be32(f, chunkstart);
2098     qemu_put_be16(f, n_valid);
2099     qemu_put_be16(f, n_invalid);
2100     qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
2101                     HASH_PTE_SIZE_64 * n_valid);
2102 }
2103 
2104 static void htab_save_end_marker(QEMUFile *f)
2105 {
2106     qemu_put_be32(f, 0);
2107     qemu_put_be16(f, 0);
2108     qemu_put_be16(f, 0);
2109 }
2110 
2111 static void htab_save_first_pass(QEMUFile *f, sPAPRMachineState *spapr,
2112                                  int64_t max_ns)
2113 {
2114     bool has_timeout = max_ns != -1;
2115     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2116     int index = spapr->htab_save_index;
2117     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2118 
2119     assert(spapr->htab_first_pass);
2120 
2121     do {
2122         int chunkstart;
2123 
2124         /* Consume invalid HPTEs */
2125         while ((index < htabslots)
2126                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2127             CLEAN_HPTE(HPTE(spapr->htab, index));
2128             index++;
2129         }
2130 
2131         /* Consume valid HPTEs */
2132         chunkstart = index;
2133         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2134                && HPTE_VALID(HPTE(spapr->htab, index))) {
2135             CLEAN_HPTE(HPTE(spapr->htab, index));
2136             index++;
2137         }
2138 
2139         if (index > chunkstart) {
2140             int n_valid = index - chunkstart;
2141 
2142             htab_save_chunk(f, spapr, chunkstart, n_valid, 0);
2143 
2144             if (has_timeout &&
2145                 (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2146                 break;
2147             }
2148         }
2149     } while ((index < htabslots) && !qemu_file_rate_limit(f));
2150 
2151     if (index >= htabslots) {
2152         assert(index == htabslots);
2153         index = 0;
2154         spapr->htab_first_pass = false;
2155     }
2156     spapr->htab_save_index = index;
2157 }
2158 
2159 static int htab_save_later_pass(QEMUFile *f, sPAPRMachineState *spapr,
2160                                 int64_t max_ns)
2161 {
2162     bool final = max_ns < 0;
2163     int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
2164     int examined = 0, sent = 0;
2165     int index = spapr->htab_save_index;
2166     int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
2167 
2168     assert(!spapr->htab_first_pass);
2169 
2170     do {
2171         int chunkstart, invalidstart;
2172 
2173         /* Consume non-dirty HPTEs */
2174         while ((index < htabslots)
2175                && !HPTE_DIRTY(HPTE(spapr->htab, index))) {
2176             index++;
2177             examined++;
2178         }
2179 
2180         chunkstart = index;
2181         /* Consume valid dirty HPTEs */
2182         while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
2183                && HPTE_DIRTY(HPTE(spapr->htab, index))
2184                && HPTE_VALID(HPTE(spapr->htab, index))) {
2185             CLEAN_HPTE(HPTE(spapr->htab, index));
2186             index++;
2187             examined++;
2188         }
2189 
2190         invalidstart = index;
2191         /* Consume invalid dirty HPTEs */
2192         while ((index < htabslots) && (index - invalidstart < USHRT_MAX)
2193                && HPTE_DIRTY(HPTE(spapr->htab, index))
2194                && !HPTE_VALID(HPTE(spapr->htab, index))) {
2195             CLEAN_HPTE(HPTE(spapr->htab, index));
2196             index++;
2197             examined++;
2198         }
2199 
2200         if (index > chunkstart) {
2201             int n_valid = invalidstart - chunkstart;
2202             int n_invalid = index - invalidstart;
2203 
2204             htab_save_chunk(f, spapr, chunkstart, n_valid, n_invalid);
2205             sent += index - chunkstart;
2206 
2207             if (!final && (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
2208                 break;
2209             }
2210         }
2211 
2212         if (examined >= htabslots) {
2213             break;
2214         }
2215 
2216         if (index >= htabslots) {
2217             assert(index == htabslots);
2218             index = 0;
2219         }
2220     } while ((examined < htabslots) && (!qemu_file_rate_limit(f) || final));
2221 
2222     if (index >= htabslots) {
2223         assert(index == htabslots);
2224         index = 0;
2225     }
2226 
2227     spapr->htab_save_index = index;
2228 
2229     return (examined >= htabslots) && (sent == 0) ? 1 : 0;
2230 }
2231 
2232 #define MAX_ITERATION_NS    5000000 /* 5 ms */
2233 #define MAX_KVM_BUF_SIZE    2048
2234 
2235 static int htab_save_iterate(QEMUFile *f, void *opaque)
2236 {
2237     sPAPRMachineState *spapr = opaque;
2238     int fd;
2239     int rc = 0;
2240 
2241     /* Iteration header */
2242     if (!spapr->htab_shift) {
2243         qemu_put_be32(f, -1);
2244         return 1;
2245     } else {
2246         qemu_put_be32(f, 0);
2247     }
2248 
2249     if (!spapr->htab) {
2250         assert(kvm_enabled());
2251 
2252         fd = get_htab_fd(spapr);
2253         if (fd < 0) {
2254             return fd;
2255         }
2256 
2257         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, MAX_ITERATION_NS);
2258         if (rc < 0) {
2259             return rc;
2260         }
2261     } else  if (spapr->htab_first_pass) {
2262         htab_save_first_pass(f, spapr, MAX_ITERATION_NS);
2263     } else {
2264         rc = htab_save_later_pass(f, spapr, MAX_ITERATION_NS);
2265     }
2266 
2267     htab_save_end_marker(f);
2268 
2269     return rc;
2270 }
2271 
2272 static int htab_save_complete(QEMUFile *f, void *opaque)
2273 {
2274     sPAPRMachineState *spapr = opaque;
2275     int fd;
2276 
2277     /* Iteration header */
2278     if (!spapr->htab_shift) {
2279         qemu_put_be32(f, -1);
2280         return 0;
2281     } else {
2282         qemu_put_be32(f, 0);
2283     }
2284 
2285     if (!spapr->htab) {
2286         int rc;
2287 
2288         assert(kvm_enabled());
2289 
2290         fd = get_htab_fd(spapr);
2291         if (fd < 0) {
2292             return fd;
2293         }
2294 
2295         rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, -1);
2296         if (rc < 0) {
2297             return rc;
2298         }
2299     } else {
2300         if (spapr->htab_first_pass) {
2301             htab_save_first_pass(f, spapr, -1);
2302         }
2303         htab_save_later_pass(f, spapr, -1);
2304     }
2305 
2306     /* End marker */
2307     htab_save_end_marker(f);
2308 
2309     return 0;
2310 }
2311 
2312 static int htab_load(QEMUFile *f, void *opaque, int version_id)
2313 {
2314     sPAPRMachineState *spapr = opaque;
2315     uint32_t section_hdr;
2316     int fd = -1;
2317     Error *local_err = NULL;
2318 
2319     if (version_id < 1 || version_id > 1) {
2320         error_report("htab_load() bad version");
2321         return -EINVAL;
2322     }
2323 
2324     section_hdr = qemu_get_be32(f);
2325 
2326     if (section_hdr == -1) {
2327         spapr_free_hpt(spapr);
2328         return 0;
2329     }
2330 
2331     if (section_hdr) {
2332         /* First section gives the htab size */
2333         spapr_reallocate_hpt(spapr, section_hdr, &local_err);
2334         if (local_err) {
2335             error_report_err(local_err);
2336             return -EINVAL;
2337         }
2338         return 0;
2339     }
2340 
2341     if (!spapr->htab) {
2342         assert(kvm_enabled());
2343 
2344         fd = kvmppc_get_htab_fd(true, 0, &local_err);
2345         if (fd < 0) {
2346             error_report_err(local_err);
2347             return fd;
2348         }
2349     }
2350 
2351     while (true) {
2352         uint32_t index;
2353         uint16_t n_valid, n_invalid;
2354 
2355         index = qemu_get_be32(f);
2356         n_valid = qemu_get_be16(f);
2357         n_invalid = qemu_get_be16(f);
2358 
2359         if ((index == 0) && (n_valid == 0) && (n_invalid == 0)) {
2360             /* End of Stream */
2361             break;
2362         }
2363 
2364         if ((index + n_valid + n_invalid) >
2365             (HTAB_SIZE(spapr) / HASH_PTE_SIZE_64)) {
2366             /* Bad index in stream */
2367             error_report(
2368                 "htab_load() bad index %d (%hd+%hd entries) in htab stream (htab_shift=%d)",
2369                 index, n_valid, n_invalid, spapr->htab_shift);
2370             return -EINVAL;
2371         }
2372 
2373         if (spapr->htab) {
2374             if (n_valid) {
2375                 qemu_get_buffer(f, HPTE(spapr->htab, index),
2376                                 HASH_PTE_SIZE_64 * n_valid);
2377             }
2378             if (n_invalid) {
2379                 memset(HPTE(spapr->htab, index + n_valid), 0,
2380                        HASH_PTE_SIZE_64 * n_invalid);
2381             }
2382         } else {
2383             int rc;
2384 
2385             assert(fd >= 0);
2386 
2387             rc = kvmppc_load_htab_chunk(f, fd, index, n_valid, n_invalid);
2388             if (rc < 0) {
2389                 return rc;
2390             }
2391         }
2392     }
2393 
2394     if (!spapr->htab) {
2395         assert(fd >= 0);
2396         close(fd);
2397     }
2398 
2399     return 0;
2400 }
2401 
2402 static void htab_save_cleanup(void *opaque)
2403 {
2404     sPAPRMachineState *spapr = opaque;
2405 
2406     close_htab_fd(spapr);
2407 }
2408 
2409 static SaveVMHandlers savevm_htab_handlers = {
2410     .save_setup = htab_save_setup,
2411     .save_live_iterate = htab_save_iterate,
2412     .save_live_complete_precopy = htab_save_complete,
2413     .save_cleanup = htab_save_cleanup,
2414     .load_state = htab_load,
2415 };
2416 
2417 static void spapr_boot_set(void *opaque, const char *boot_device,
2418                            Error **errp)
2419 {
2420     MachineState *machine = MACHINE(opaque);
2421     machine->boot_order = g_strdup(boot_device);
2422 }
2423 
2424 static void spapr_create_lmb_dr_connectors(sPAPRMachineState *spapr)
2425 {
2426     MachineState *machine = MACHINE(spapr);
2427     uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
2428     uint32_t nr_lmbs = (machine->maxram_size - machine->ram_size)/lmb_size;
2429     int i;
2430 
2431     for (i = 0; i < nr_lmbs; i++) {
2432         uint64_t addr;
2433 
2434         addr = i * lmb_size + machine->device_memory->base;
2435         spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_LMB,
2436                                addr / lmb_size);
2437     }
2438 }
2439 
2440 /*
2441  * If RAM size, maxmem size and individual node mem sizes aren't aligned
2442  * to SPAPR_MEMORY_BLOCK_SIZE(256MB), then refuse to start the guest
2443  * since we can't support such unaligned sizes with DRCONF_MEMORY.
2444  */
2445 static void spapr_validate_node_memory(MachineState *machine, Error **errp)
2446 {
2447     int i;
2448 
2449     if (machine->ram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2450         error_setg(errp, "Memory size 0x" RAM_ADDR_FMT
2451                    " is not aligned to %" PRIu64 " MiB",
2452                    machine->ram_size,
2453                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2454         return;
2455     }
2456 
2457     if (machine->maxram_size % SPAPR_MEMORY_BLOCK_SIZE) {
2458         error_setg(errp, "Maximum memory size 0x" RAM_ADDR_FMT
2459                    " is not aligned to %" PRIu64 " MiB",
2460                    machine->ram_size,
2461                    SPAPR_MEMORY_BLOCK_SIZE / MiB);
2462         return;
2463     }
2464 
2465     for (i = 0; i < nb_numa_nodes; i++) {
2466         if (numa_info[i].node_mem % SPAPR_MEMORY_BLOCK_SIZE) {
2467             error_setg(errp,
2468                        "Node %d memory size 0x%" PRIx64
2469                        " is not aligned to %" PRIu64 " MiB",
2470                        i, numa_info[i].node_mem,
2471                        SPAPR_MEMORY_BLOCK_SIZE / MiB);
2472             return;
2473         }
2474     }
2475 }
2476 
2477 /* find cpu slot in machine->possible_cpus by core_id */
2478 static CPUArchId *spapr_find_cpu_slot(MachineState *ms, uint32_t id, int *idx)
2479 {
2480     int index = id / smp_threads;
2481 
2482     if (index >= ms->possible_cpus->len) {
2483         return NULL;
2484     }
2485     if (idx) {
2486         *idx = index;
2487     }
2488     return &ms->possible_cpus->cpus[index];
2489 }
2490 
2491 static void spapr_set_vsmt_mode(sPAPRMachineState *spapr, Error **errp)
2492 {
2493     Error *local_err = NULL;
2494     bool vsmt_user = !!spapr->vsmt;
2495     int kvm_smt = kvmppc_smt_threads();
2496     int ret;
2497 
2498     if (!kvm_enabled() && (smp_threads > 1)) {
2499         error_setg(&local_err, "TCG cannot support more than 1 thread/core "
2500                      "on a pseries machine");
2501         goto out;
2502     }
2503     if (!is_power_of_2(smp_threads)) {
2504         error_setg(&local_err, "Cannot support %d threads/core on a pseries "
2505                      "machine because it must be a power of 2", smp_threads);
2506         goto out;
2507     }
2508 
2509     /* Detemine the VSMT mode to use: */
2510     if (vsmt_user) {
2511         if (spapr->vsmt < smp_threads) {
2512             error_setg(&local_err, "Cannot support VSMT mode %d"
2513                          " because it must be >= threads/core (%d)",
2514                          spapr->vsmt, smp_threads);
2515             goto out;
2516         }
2517         /* In this case, spapr->vsmt has been set by the command line */
2518     } else {
2519         /*
2520          * Default VSMT value is tricky, because we need it to be as
2521          * consistent as possible (for migration), but this requires
2522          * changing it for at least some existing cases.  We pick 8 as
2523          * the value that we'd get with KVM on POWER8, the
2524          * overwhelmingly common case in production systems.
2525          */
2526         spapr->vsmt = MAX(8, smp_threads);
2527     }
2528 
2529     /* KVM: If necessary, set the SMT mode: */
2530     if (kvm_enabled() && (spapr->vsmt != kvm_smt)) {
2531         ret = kvmppc_set_smt_threads(spapr->vsmt);
2532         if (ret) {
2533             /* Looks like KVM isn't able to change VSMT mode */
2534             error_setg(&local_err,
2535                        "Failed to set KVM's VSMT mode to %d (errno %d)",
2536                        spapr->vsmt, ret);
2537             /* We can live with that if the default one is big enough
2538              * for the number of threads, and a submultiple of the one
2539              * we want.  In this case we'll waste some vcpu ids, but
2540              * behaviour will be correct */
2541             if ((kvm_smt >= smp_threads) && ((spapr->vsmt % kvm_smt) == 0)) {
2542                 warn_report_err(local_err);
2543                 local_err = NULL;
2544                 goto out;
2545             } else {
2546                 if (!vsmt_user) {
2547                     error_append_hint(&local_err,
2548                                       "On PPC, a VM with %d threads/core"
2549                                       " on a host with %d threads/core"
2550                                       " requires the use of VSMT mode %d.\n",
2551                                       smp_threads, kvm_smt, spapr->vsmt);
2552                 }
2553                 kvmppc_hint_smt_possible(&local_err);
2554                 goto out;
2555             }
2556         }
2557     }
2558     /* else TCG: nothing to do currently */
2559 out:
2560     error_propagate(errp, local_err);
2561 }
2562 
2563 static void spapr_init_cpus(sPAPRMachineState *spapr)
2564 {
2565     MachineState *machine = MACHINE(spapr);
2566     MachineClass *mc = MACHINE_GET_CLASS(machine);
2567     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2568     const char *type = spapr_get_cpu_core_type(machine->cpu_type);
2569     const CPUArchIdList *possible_cpus;
2570     int boot_cores_nr = smp_cpus / smp_threads;
2571     int i;
2572 
2573     possible_cpus = mc->possible_cpu_arch_ids(machine);
2574     if (mc->has_hotpluggable_cpus) {
2575         if (smp_cpus % smp_threads) {
2576             error_report("smp_cpus (%u) must be multiple of threads (%u)",
2577                          smp_cpus, smp_threads);
2578             exit(1);
2579         }
2580         if (max_cpus % smp_threads) {
2581             error_report("max_cpus (%u) must be multiple of threads (%u)",
2582                          max_cpus, smp_threads);
2583             exit(1);
2584         }
2585     } else {
2586         if (max_cpus != smp_cpus) {
2587             error_report("This machine version does not support CPU hotplug");
2588             exit(1);
2589         }
2590         boot_cores_nr = possible_cpus->len;
2591     }
2592 
2593     if (smc->pre_2_10_has_unused_icps) {
2594         int i;
2595 
2596         for (i = 0; i < spapr_max_server_number(spapr); i++) {
2597             /* Dummy entries get deregistered when real ICPState objects
2598              * are registered during CPU core hotplug.
2599              */
2600             pre_2_10_vmstate_register_dummy_icp(i);
2601         }
2602     }
2603 
2604     for (i = 0; i < possible_cpus->len; i++) {
2605         int core_id = i * smp_threads;
2606 
2607         if (mc->has_hotpluggable_cpus) {
2608             spapr_dr_connector_new(OBJECT(spapr), TYPE_SPAPR_DRC_CPU,
2609                                    spapr_vcpu_id(spapr, core_id));
2610         }
2611 
2612         if (i < boot_cores_nr) {
2613             Object *core  = object_new(type);
2614             int nr_threads = smp_threads;
2615 
2616             /* Handle the partially filled core for older machine types */
2617             if ((i + 1) * smp_threads >= smp_cpus) {
2618                 nr_threads = smp_cpus - i * smp_threads;
2619             }
2620 
2621             object_property_set_int(core, nr_threads, "nr-threads",
2622                                     &error_fatal);
2623             object_property_set_int(core, core_id, CPU_CORE_PROP_CORE_ID,
2624                                     &error_fatal);
2625             object_property_set_bool(core, true, "realized", &error_fatal);
2626 
2627             object_unref(core);
2628         }
2629     }
2630 }
2631 
2632 static PCIHostState *spapr_create_default_phb(void)
2633 {
2634     DeviceState *dev;
2635 
2636     dev = qdev_create(NULL, TYPE_SPAPR_PCI_HOST_BRIDGE);
2637     qdev_prop_set_uint32(dev, "index", 0);
2638     qdev_init_nofail(dev);
2639 
2640     return PCI_HOST_BRIDGE(dev);
2641 }
2642 
2643 /* pSeries LPAR / sPAPR hardware init */
2644 static void spapr_machine_init(MachineState *machine)
2645 {
2646     sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
2647     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
2648     const char *kernel_filename = machine->kernel_filename;
2649     const char *initrd_filename = machine->initrd_filename;
2650     PCIHostState *phb;
2651     int i;
2652     MemoryRegion *sysmem = get_system_memory();
2653     MemoryRegion *ram = g_new(MemoryRegion, 1);
2654     hwaddr node0_size = spapr_node0_size(machine);
2655     long load_limit, fw_size;
2656     char *filename;
2657     Error *resize_hpt_err = NULL;
2658 
2659     msi_nonbroken = true;
2660 
2661     QLIST_INIT(&spapr->phbs);
2662     QTAILQ_INIT(&spapr->pending_dimm_unplugs);
2663 
2664     /* Determine capabilities to run with */
2665     spapr_caps_init(spapr);
2666 
2667     kvmppc_check_papr_resize_hpt(&resize_hpt_err);
2668     if (spapr->resize_hpt == SPAPR_RESIZE_HPT_DEFAULT) {
2669         /*
2670          * If the user explicitly requested a mode we should either
2671          * supply it, or fail completely (which we do below).  But if
2672          * it's not set explicitly, we reset our mode to something
2673          * that works
2674          */
2675         if (resize_hpt_err) {
2676             spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
2677             error_free(resize_hpt_err);
2678             resize_hpt_err = NULL;
2679         } else {
2680             spapr->resize_hpt = smc->resize_hpt_default;
2681         }
2682     }
2683 
2684     assert(spapr->resize_hpt != SPAPR_RESIZE_HPT_DEFAULT);
2685 
2686     if ((spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) && resize_hpt_err) {
2687         /*
2688          * User requested HPT resize, but this host can't supply it.  Bail out
2689          */
2690         error_report_err(resize_hpt_err);
2691         exit(1);
2692     }
2693 
2694     spapr->rma_size = node0_size;
2695 
2696     /* With KVM, we don't actually know whether KVM supports an
2697      * unbounded RMA (PR KVM) or is limited by the hash table size
2698      * (HV KVM using VRMA), so we always assume the latter
2699      *
2700      * In that case, we also limit the initial allocations for RTAS
2701      * etc... to 256M since we have no way to know what the VRMA size
2702      * is going to be as it depends on the size of the hash table
2703      * which isn't determined yet.
2704      */
2705     if (kvm_enabled()) {
2706         spapr->vrma_adjust = 1;
2707         spapr->rma_size = MIN(spapr->rma_size, 0x10000000);
2708     }
2709 
2710     /* Actually we don't support unbounded RMA anymore since we added
2711      * proper emulation of HV mode. The max we can get is 16G which
2712      * also happens to be what we configure for PAPR mode so make sure
2713      * we don't do anything bigger than that
2714      */
2715     spapr->rma_size = MIN(spapr->rma_size, 0x400000000ull);
2716 
2717     if (spapr->rma_size > node0_size) {
2718         error_report("Numa node 0 has to span the RMA (%#08"HWADDR_PRIx")",
2719                      spapr->rma_size);
2720         exit(1);
2721     }
2722 
2723     /* Setup a load limit for the ramdisk leaving room for SLOF and FDT */
2724     load_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR) - FW_OVERHEAD;
2725 
2726     /*
2727      * VSMT must be set in order to be able to compute VCPU ids, ie to
2728      * call spapr_max_server_number() or spapr_vcpu_id().
2729      */
2730     spapr_set_vsmt_mode(spapr, &error_fatal);
2731 
2732     /* Set up Interrupt Controller before we create the VCPUs */
2733     spapr_irq_init(spapr, &error_fatal);
2734 
2735     /* Set up containers for ibm,client-architecture-support negotiated options
2736      */
2737     spapr->ov5 = spapr_ovec_new();
2738     spapr->ov5_cas = spapr_ovec_new();
2739 
2740     if (smc->dr_lmb_enabled) {
2741         spapr_ovec_set(spapr->ov5, OV5_DRCONF_MEMORY);
2742         spapr_validate_node_memory(machine, &error_fatal);
2743     }
2744 
2745     spapr_ovec_set(spapr->ov5, OV5_FORM1_AFFINITY);
2746 
2747     /* advertise support for dedicated HP event source to guests */
2748     if (spapr->use_hotplug_event_source) {
2749         spapr_ovec_set(spapr->ov5, OV5_HP_EVT);
2750     }
2751 
2752     /* advertise support for HPT resizing */
2753     if (spapr->resize_hpt != SPAPR_RESIZE_HPT_DISABLED) {
2754         spapr_ovec_set(spapr->ov5, OV5_HPT_RESIZE);
2755     }
2756 
2757     /* advertise support for ibm,dyamic-memory-v2 */
2758     spapr_ovec_set(spapr->ov5, OV5_DRMEM_V2);
2759 
2760     /* advertise XIVE on POWER9 machines */
2761     if (spapr->irq->ov5 & (SPAPR_OV5_XIVE_EXPLOIT | SPAPR_OV5_XIVE_BOTH)) {
2762         if (ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00,
2763                                   0, spapr->max_compat_pvr)) {
2764             spapr_ovec_set(spapr->ov5, OV5_XIVE_EXPLOIT);
2765         } else if (spapr->irq->ov5 & SPAPR_OV5_XIVE_EXPLOIT) {
2766             error_report("XIVE-only machines require a POWER9 CPU");
2767             exit(1);
2768         }
2769     }
2770 
2771     /* init CPUs */
2772     spapr_init_cpus(spapr);
2773 
2774     if ((!kvm_enabled() || kvmppc_has_cap_mmu_radix()) &&
2775         ppc_type_check_compat(machine->cpu_type, CPU_POWERPC_LOGICAL_3_00, 0,
2776                               spapr->max_compat_pvr)) {
2777         /* KVM and TCG always allow GTSE with radix... */
2778         spapr_ovec_set(spapr->ov5, OV5_MMU_RADIX_GTSE);
2779     }
2780     /* ... but not with hash (currently). */
2781 
2782     if (kvm_enabled()) {
2783         /* Enable H_LOGICAL_CI_* so SLOF can talk to in-kernel devices */
2784         kvmppc_enable_logical_ci_hcalls();
2785         kvmppc_enable_set_mode_hcall();
2786 
2787         /* H_CLEAR_MOD/_REF are mandatory in PAPR, but off by default */
2788         kvmppc_enable_clear_ref_mod_hcalls();
2789     }
2790 
2791     /* allocate RAM */
2792     memory_region_allocate_system_memory(ram, NULL, "ppc_spapr.ram",
2793                                          machine->ram_size);
2794     memory_region_add_subregion(sysmem, 0, ram);
2795 
2796     /* always allocate the device memory information */
2797     machine->device_memory = g_malloc0(sizeof(*machine->device_memory));
2798 
2799     /* initialize hotplug memory address space */
2800     if (machine->ram_size < machine->maxram_size) {
2801         ram_addr_t device_mem_size = machine->maxram_size - machine->ram_size;
2802         /*
2803          * Limit the number of hotpluggable memory slots to half the number
2804          * slots that KVM supports, leaving the other half for PCI and other
2805          * devices. However ensure that number of slots doesn't drop below 32.
2806          */
2807         int max_memslots = kvm_enabled() ? kvm_get_max_memslots() / 2 :
2808                            SPAPR_MAX_RAM_SLOTS;
2809 
2810         if (max_memslots < SPAPR_MAX_RAM_SLOTS) {
2811             max_memslots = SPAPR_MAX_RAM_SLOTS;
2812         }
2813         if (machine->ram_slots > max_memslots) {
2814             error_report("Specified number of memory slots %"
2815                          PRIu64" exceeds max supported %d",
2816                          machine->ram_slots, max_memslots);
2817             exit(1);
2818         }
2819 
2820         machine->device_memory->base = ROUND_UP(machine->ram_size,
2821                                                 SPAPR_DEVICE_MEM_ALIGN);
2822         memory_region_init(&machine->device_memory->mr, OBJECT(spapr),
2823                            "device-memory", device_mem_size);
2824         memory_region_add_subregion(sysmem, machine->device_memory->base,
2825                                     &machine->device_memory->mr);
2826     }
2827 
2828     if (smc->dr_lmb_enabled) {
2829         spapr_create_lmb_dr_connectors(spapr);
2830     }
2831 
2832     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, "spapr-rtas.bin");
2833     if (!filename) {
2834         error_report("Could not find LPAR rtas '%s'", "spapr-rtas.bin");
2835         exit(1);
2836     }
2837     spapr->rtas_size = get_image_size(filename);
2838     if (spapr->rtas_size < 0) {
2839         error_report("Could not get size of LPAR rtas '%s'", filename);
2840         exit(1);
2841     }
2842     spapr->rtas_blob = g_malloc(spapr->rtas_size);
2843     if (load_image_size(filename, spapr->rtas_blob, spapr->rtas_size) < 0) {
2844         error_report("Could not load LPAR rtas '%s'", filename);
2845         exit(1);
2846     }
2847     if (spapr->rtas_size > RTAS_MAX_SIZE) {
2848         error_report("RTAS too big ! 0x%zx bytes (max is 0x%x)",
2849                      (size_t)spapr->rtas_size, RTAS_MAX_SIZE);
2850         exit(1);
2851     }
2852     g_free(filename);
2853 
2854     /* Set up RTAS event infrastructure */
2855     spapr_events_init(spapr);
2856 
2857     /* Set up the RTC RTAS interfaces */
2858     spapr_rtc_create(spapr);
2859 
2860     /* Set up VIO bus */
2861     spapr->vio_bus = spapr_vio_bus_init();
2862 
2863     for (i = 0; i < serial_max_hds(); i++) {
2864         if (serial_hd(i)) {
2865             spapr_vty_create(spapr->vio_bus, serial_hd(i));
2866         }
2867     }
2868 
2869     /* We always have at least the nvram device on VIO */
2870     spapr_create_nvram(spapr);
2871 
2872     /* Set up PCI */
2873     spapr_pci_rtas_init();
2874 
2875     phb = spapr_create_default_phb();
2876 
2877     for (i = 0; i < nb_nics; i++) {
2878         NICInfo *nd = &nd_table[i];
2879 
2880         if (!nd->model) {
2881             nd->model = g_strdup("spapr-vlan");
2882         }
2883 
2884         if (g_str_equal(nd->model, "spapr-vlan") ||
2885             g_str_equal(nd->model, "ibmveth")) {
2886             spapr_vlan_create(spapr->vio_bus, nd);
2887         } else {
2888             pci_nic_init_nofail(&nd_table[i], phb->bus, nd->model, NULL);
2889         }
2890     }
2891 
2892     for (i = 0; i <= drive_get_max_bus(IF_SCSI); i++) {
2893         spapr_vscsi_create(spapr->vio_bus);
2894     }
2895 
2896     /* Graphics */
2897     if (spapr_vga_init(phb->bus, &error_fatal)) {
2898         spapr->has_graphics = true;
2899         machine->usb |= defaults_enabled() && !machine->usb_disabled;
2900     }
2901 
2902     if (machine->usb) {
2903         if (smc->use_ohci_by_default) {
2904             pci_create_simple(phb->bus, -1, "pci-ohci");
2905         } else {
2906             pci_create_simple(phb->bus, -1, "nec-usb-xhci");
2907         }
2908 
2909         if (spapr->has_graphics) {
2910             USBBus *usb_bus = usb_bus_find(-1);
2911 
2912             usb_create_simple(usb_bus, "usb-kbd");
2913             usb_create_simple(usb_bus, "usb-mouse");
2914         }
2915     }
2916 
2917     if (spapr->rma_size < (MIN_RMA_SLOF * MiB)) {
2918         error_report(
2919             "pSeries SLOF firmware requires >= %ldM guest RMA (Real Mode Area memory)",
2920             MIN_RMA_SLOF);
2921         exit(1);
2922     }
2923 
2924     if (kernel_filename) {
2925         uint64_t lowaddr = 0;
2926 
2927         spapr->kernel_size = load_elf(kernel_filename, NULL,
2928                                       translate_kernel_address, NULL,
2929                                       NULL, &lowaddr, NULL, 1,
2930                                       PPC_ELF_MACHINE, 0, 0);
2931         if (spapr->kernel_size == ELF_LOAD_WRONG_ENDIAN) {
2932             spapr->kernel_size = load_elf(kernel_filename, NULL,
2933                                           translate_kernel_address, NULL, NULL,
2934                                           &lowaddr, NULL, 0, PPC_ELF_MACHINE,
2935                                           0, 0);
2936             spapr->kernel_le = spapr->kernel_size > 0;
2937         }
2938         if (spapr->kernel_size < 0) {
2939             error_report("error loading %s: %s", kernel_filename,
2940                          load_elf_strerror(spapr->kernel_size));
2941             exit(1);
2942         }
2943 
2944         /* load initrd */
2945         if (initrd_filename) {
2946             /* Try to locate the initrd in the gap between the kernel
2947              * and the firmware. Add a bit of space just in case
2948              */
2949             spapr->initrd_base = (KERNEL_LOAD_ADDR + spapr->kernel_size
2950                                   + 0x1ffff) & ~0xffff;
2951             spapr->initrd_size = load_image_targphys(initrd_filename,
2952                                                      spapr->initrd_base,
2953                                                      load_limit
2954                                                      - spapr->initrd_base);
2955             if (spapr->initrd_size < 0) {
2956                 error_report("could not load initial ram disk '%s'",
2957                              initrd_filename);
2958                 exit(1);
2959             }
2960         }
2961     }
2962 
2963     if (bios_name == NULL) {
2964         bios_name = FW_FILE_NAME;
2965     }
2966     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
2967     if (!filename) {
2968         error_report("Could not find LPAR firmware '%s'", bios_name);
2969         exit(1);
2970     }
2971     fw_size = load_image_targphys(filename, 0, FW_MAX_SIZE);
2972     if (fw_size <= 0) {
2973         error_report("Could not load LPAR firmware '%s'", filename);
2974         exit(1);
2975     }
2976     g_free(filename);
2977 
2978     /* FIXME: Should register things through the MachineState's qdev
2979      * interface, this is a legacy from the sPAPREnvironment structure
2980      * which predated MachineState but had a similar function */
2981     vmstate_register(NULL, 0, &vmstate_spapr, spapr);
2982     register_savevm_live(NULL, "spapr/htab", -1, 1,
2983                          &savevm_htab_handlers, spapr);
2984 
2985     qemu_register_boot_set(spapr_boot_set, spapr);
2986 
2987     if (kvm_enabled()) {
2988         /* to stop and start vmclock */
2989         qemu_add_vm_change_state_handler(cpu_ppc_clock_vm_state_change,
2990                                          &spapr->tb);
2991 
2992         kvmppc_spapr_enable_inkernel_multitce();
2993     }
2994 }
2995 
2996 static int spapr_kvm_type(const char *vm_type)
2997 {
2998     if (!vm_type) {
2999         return 0;
3000     }
3001 
3002     if (!strcmp(vm_type, "HV")) {
3003         return 1;
3004     }
3005 
3006     if (!strcmp(vm_type, "PR")) {
3007         return 2;
3008     }
3009 
3010     error_report("Unknown kvm-type specified '%s'", vm_type);
3011     exit(1);
3012 }
3013 
3014 /*
3015  * Implementation of an interface to adjust firmware path
3016  * for the bootindex property handling.
3017  */
3018 static char *spapr_get_fw_dev_path(FWPathProvider *p, BusState *bus,
3019                                    DeviceState *dev)
3020 {
3021 #define CAST(type, obj, name) \
3022     ((type *)object_dynamic_cast(OBJECT(obj), (name)))
3023     SCSIDevice *d = CAST(SCSIDevice,  dev, TYPE_SCSI_DEVICE);
3024     sPAPRPHBState *phb = CAST(sPAPRPHBState, dev, TYPE_SPAPR_PCI_HOST_BRIDGE);
3025     VHostSCSICommon *vsc = CAST(VHostSCSICommon, dev, TYPE_VHOST_SCSI_COMMON);
3026 
3027     if (d) {
3028         void *spapr = CAST(void, bus->parent, "spapr-vscsi");
3029         VirtIOSCSI *virtio = CAST(VirtIOSCSI, bus->parent, TYPE_VIRTIO_SCSI);
3030         USBDevice *usb = CAST(USBDevice, bus->parent, TYPE_USB_DEVICE);
3031 
3032         if (spapr) {
3033             /*
3034              * Replace "channel@0/disk@0,0" with "disk@8000000000000000":
3035              * In the top 16 bits of the 64-bit LUN, we use SRP luns of the form
3036              * 0x8000 | (target << 8) | (bus << 5) | lun
3037              * (see the "Logical unit addressing format" table in SAM5)
3038              */
3039             unsigned id = 0x8000 | (d->id << 8) | (d->channel << 5) | d->lun;
3040             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3041                                    (uint64_t)id << 48);
3042         } else if (virtio) {
3043             /*
3044              * We use SRP luns of the form 01000000 | (target << 8) | lun
3045              * in the top 32 bits of the 64-bit LUN
3046              * Note: the quote above is from SLOF and it is wrong,
3047              * the actual binding is:
3048              * swap 0100 or 10 << or 20 << ( target lun-id -- srplun )
3049              */
3050             unsigned id = 0x1000000 | (d->id << 16) | d->lun;
3051             if (d->lun >= 256) {
3052                 /* Use the LUN "flat space addressing method" */
3053                 id |= 0x4000;
3054             }
3055             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3056                                    (uint64_t)id << 32);
3057         } else if (usb) {
3058             /*
3059              * We use SRP luns of the form 01000000 | (usb-port << 16) | lun
3060              * in the top 32 bits of the 64-bit LUN
3061              */
3062             unsigned usb_port = atoi(usb->port->path);
3063             unsigned id = 0x1000000 | (usb_port << 16) | d->lun;
3064             return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
3065                                    (uint64_t)id << 32);
3066         }
3067     }
3068 
3069     /*
3070      * SLOF probes the USB devices, and if it recognizes that the device is a
3071      * storage device, it changes its name to "storage" instead of "usb-host",
3072      * and additionally adds a child node for the SCSI LUN, so the correct
3073      * boot path in SLOF is something like .../storage@1/disk@xxx" instead.
3074      */
3075     if (strcmp("usb-host", qdev_fw_name(dev)) == 0) {
3076         USBDevice *usbdev = CAST(USBDevice, dev, TYPE_USB_DEVICE);
3077         if (usb_host_dev_is_scsi_storage(usbdev)) {
3078             return g_strdup_printf("storage@%s/disk", usbdev->port->path);
3079         }
3080     }
3081 
3082     if (phb) {
3083         /* Replace "pci" with "pci@800000020000000" */
3084         return g_strdup_printf("pci@%"PRIX64, phb->buid);
3085     }
3086 
3087     if (vsc) {
3088         /* Same logic as virtio above */
3089         unsigned id = 0x1000000 | (vsc->target << 16) | vsc->lun;
3090         return g_strdup_printf("disk@%"PRIX64, (uint64_t)id << 32);
3091     }
3092 
3093     if (g_str_equal("pci-bridge", qdev_fw_name(dev))) {
3094         /* SLOF uses "pci" instead of "pci-bridge" for PCI bridges */
3095         PCIDevice *pcidev = CAST(PCIDevice, dev, TYPE_PCI_DEVICE);
3096         return g_strdup_printf("pci@%x", PCI_SLOT(pcidev->devfn));
3097     }
3098 
3099     return NULL;
3100 }
3101 
3102 static char *spapr_get_kvm_type(Object *obj, Error **errp)
3103 {
3104     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
3105 
3106     return g_strdup(spapr->kvm_type);
3107 }
3108 
3109 static void spapr_set_kvm_type(Object *obj, const char *value, Error **errp)
3110 {
3111     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
3112 
3113     g_free(spapr->kvm_type);
3114     spapr->kvm_type = g_strdup(value);
3115 }
3116 
3117 static bool spapr_get_modern_hotplug_events(Object *obj, Error **errp)
3118 {
3119     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
3120 
3121     return spapr->use_hotplug_event_source;
3122 }
3123 
3124 static void spapr_set_modern_hotplug_events(Object *obj, bool value,
3125                                             Error **errp)
3126 {
3127     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
3128 
3129     spapr->use_hotplug_event_source = value;
3130 }
3131 
3132 static bool spapr_get_msix_emulation(Object *obj, Error **errp)
3133 {
3134     return true;
3135 }
3136 
3137 static char *spapr_get_resize_hpt(Object *obj, Error **errp)
3138 {
3139     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
3140 
3141     switch (spapr->resize_hpt) {
3142     case SPAPR_RESIZE_HPT_DEFAULT:
3143         return g_strdup("default");
3144     case SPAPR_RESIZE_HPT_DISABLED:
3145         return g_strdup("disabled");
3146     case SPAPR_RESIZE_HPT_ENABLED:
3147         return g_strdup("enabled");
3148     case SPAPR_RESIZE_HPT_REQUIRED:
3149         return g_strdup("required");
3150     }
3151     g_assert_not_reached();
3152 }
3153 
3154 static void spapr_set_resize_hpt(Object *obj, const char *value, Error **errp)
3155 {
3156     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
3157 
3158     if (strcmp(value, "default") == 0) {
3159         spapr->resize_hpt = SPAPR_RESIZE_HPT_DEFAULT;
3160     } else if (strcmp(value, "disabled") == 0) {
3161         spapr->resize_hpt = SPAPR_RESIZE_HPT_DISABLED;
3162     } else if (strcmp(value, "enabled") == 0) {
3163         spapr->resize_hpt = SPAPR_RESIZE_HPT_ENABLED;
3164     } else if (strcmp(value, "required") == 0) {
3165         spapr->resize_hpt = SPAPR_RESIZE_HPT_REQUIRED;
3166     } else {
3167         error_setg(errp, "Bad value for \"resize-hpt\" property");
3168     }
3169 }
3170 
3171 static void spapr_get_vsmt(Object *obj, Visitor *v, const char *name,
3172                                    void *opaque, Error **errp)
3173 {
3174     visit_type_uint32(v, name, (uint32_t *)opaque, errp);
3175 }
3176 
3177 static void spapr_set_vsmt(Object *obj, Visitor *v, const char *name,
3178                                    void *opaque, Error **errp)
3179 {
3180     visit_type_uint32(v, name, (uint32_t *)opaque, errp);
3181 }
3182 
3183 static char *spapr_get_ic_mode(Object *obj, Error **errp)
3184 {
3185     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
3186 
3187     if (spapr->irq == &spapr_irq_xics_legacy) {
3188         return g_strdup("legacy");
3189     } else if (spapr->irq == &spapr_irq_xics) {
3190         return g_strdup("xics");
3191     } else if (spapr->irq == &spapr_irq_xive) {
3192         return g_strdup("xive");
3193     } else if (spapr->irq == &spapr_irq_dual) {
3194         return g_strdup("dual");
3195     }
3196     g_assert_not_reached();
3197 }
3198 
3199 static void spapr_set_ic_mode(Object *obj, const char *value, Error **errp)
3200 {
3201     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
3202 
3203     if (SPAPR_MACHINE_GET_CLASS(spapr)->legacy_irq_allocation) {
3204         error_setg(errp, "This machine only uses the legacy XICS backend, don't pass ic-mode");
3205         return;
3206     }
3207 
3208     /* The legacy IRQ backend can not be set */
3209     if (strcmp(value, "xics") == 0) {
3210         spapr->irq = &spapr_irq_xics;
3211     } else if (strcmp(value, "xive") == 0) {
3212         spapr->irq = &spapr_irq_xive;
3213     } else if (strcmp(value, "dual") == 0) {
3214         spapr->irq = &spapr_irq_dual;
3215     } else {
3216         error_setg(errp, "Bad value for \"ic-mode\" property");
3217     }
3218 }
3219 
3220 static char *spapr_get_host_model(Object *obj, Error **errp)
3221 {
3222     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
3223 
3224     return g_strdup(spapr->host_model);
3225 }
3226 
3227 static void spapr_set_host_model(Object *obj, const char *value, Error **errp)
3228 {
3229     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
3230 
3231     g_free(spapr->host_model);
3232     spapr->host_model = g_strdup(value);
3233 }
3234 
3235 static char *spapr_get_host_serial(Object *obj, Error **errp)
3236 {
3237     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
3238 
3239     return g_strdup(spapr->host_serial);
3240 }
3241 
3242 static void spapr_set_host_serial(Object *obj, const char *value, Error **errp)
3243 {
3244     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
3245 
3246     g_free(spapr->host_serial);
3247     spapr->host_serial = g_strdup(value);
3248 }
3249 
3250 static void spapr_instance_init(Object *obj)
3251 {
3252     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
3253     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
3254 
3255     spapr->htab_fd = -1;
3256     spapr->use_hotplug_event_source = true;
3257     object_property_add_str(obj, "kvm-type",
3258                             spapr_get_kvm_type, spapr_set_kvm_type, NULL);
3259     object_property_set_description(obj, "kvm-type",
3260                                     "Specifies the KVM virtualization mode (HV, PR)",
3261                                     NULL);
3262     object_property_add_bool(obj, "modern-hotplug-events",
3263                             spapr_get_modern_hotplug_events,
3264                             spapr_set_modern_hotplug_events,
3265                             NULL);
3266     object_property_set_description(obj, "modern-hotplug-events",
3267                                     "Use dedicated hotplug event mechanism in"
3268                                     " place of standard EPOW events when possible"
3269                                     " (required for memory hot-unplug support)",
3270                                     NULL);
3271     ppc_compat_add_property(obj, "max-cpu-compat", &spapr->max_compat_pvr,
3272                             "Maximum permitted CPU compatibility mode",
3273                             &error_fatal);
3274 
3275     object_property_add_str(obj, "resize-hpt",
3276                             spapr_get_resize_hpt, spapr_set_resize_hpt, NULL);
3277     object_property_set_description(obj, "resize-hpt",
3278                                     "Resizing of the Hash Page Table (enabled, disabled, required)",
3279                                     NULL);
3280     object_property_add(obj, "vsmt", "uint32", spapr_get_vsmt,
3281                         spapr_set_vsmt, NULL, &spapr->vsmt, &error_abort);
3282     object_property_set_description(obj, "vsmt",
3283                                     "Virtual SMT: KVM behaves as if this were"
3284                                     " the host's SMT mode", &error_abort);
3285     object_property_add_bool(obj, "vfio-no-msix-emulation",
3286                              spapr_get_msix_emulation, NULL, NULL);
3287 
3288     /* The machine class defines the default interrupt controller mode */
3289     spapr->irq = smc->irq;
3290     object_property_add_str(obj, "ic-mode", spapr_get_ic_mode,
3291                             spapr_set_ic_mode, NULL);
3292     object_property_set_description(obj, "ic-mode",
3293                  "Specifies the interrupt controller mode (xics, xive, dual)",
3294                  NULL);
3295 
3296     object_property_add_str(obj, "host-model",
3297         spapr_get_host_model, spapr_set_host_model,
3298         &error_abort);
3299     object_property_set_description(obj, "host-model",
3300         "Set host's model-id to use - none|passthrough|string", &error_abort);
3301     object_property_add_str(obj, "host-serial",
3302         spapr_get_host_serial, spapr_set_host_serial,
3303         &error_abort);
3304     object_property_set_description(obj, "host-serial",
3305         "Set host's system-id to use - none|passthrough|string", &error_abort);
3306 }
3307 
3308 static void spapr_machine_finalizefn(Object *obj)
3309 {
3310     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
3311 
3312     g_free(spapr->kvm_type);
3313 }
3314 
3315 void spapr_do_system_reset_on_cpu(CPUState *cs, run_on_cpu_data arg)
3316 {
3317     cpu_synchronize_state(cs);
3318     ppc_cpu_do_system_reset(cs);
3319 }
3320 
3321 static void spapr_nmi(NMIState *n, int cpu_index, Error **errp)
3322 {
3323     CPUState *cs;
3324 
3325     CPU_FOREACH(cs) {
3326         async_run_on_cpu(cs, spapr_do_system_reset_on_cpu, RUN_ON_CPU_NULL);
3327     }
3328 }
3329 
3330 static void spapr_add_lmbs(DeviceState *dev, uint64_t addr_start, uint64_t size,
3331                            uint32_t node, bool dedicated_hp_event_source,
3332                            Error **errp)
3333 {
3334     sPAPRDRConnector *drc;
3335     uint32_t nr_lmbs = size/SPAPR_MEMORY_BLOCK_SIZE;
3336     int i, fdt_offset, fdt_size;
3337     void *fdt;
3338     uint64_t addr = addr_start;
3339     bool hotplugged = spapr_drc_hotplugged(dev);
3340     Error *local_err = NULL;
3341 
3342     for (i = 0; i < nr_lmbs; i++) {
3343         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3344                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3345         g_assert(drc);
3346 
3347         fdt = create_device_tree(&fdt_size);
3348         fdt_offset = spapr_populate_memory_node(fdt, node, addr,
3349                                                 SPAPR_MEMORY_BLOCK_SIZE);
3350 
3351         spapr_drc_attach(drc, dev, fdt, fdt_offset, &local_err);
3352         if (local_err) {
3353             while (addr > addr_start) {
3354                 addr -= SPAPR_MEMORY_BLOCK_SIZE;
3355                 drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3356                                       addr / SPAPR_MEMORY_BLOCK_SIZE);
3357                 spapr_drc_detach(drc);
3358             }
3359             g_free(fdt);
3360             error_propagate(errp, local_err);
3361             return;
3362         }
3363         if (!hotplugged) {
3364             spapr_drc_reset(drc);
3365         }
3366         addr += SPAPR_MEMORY_BLOCK_SIZE;
3367     }
3368     /* send hotplug notification to the
3369      * guest only in case of hotplugged memory
3370      */
3371     if (hotplugged) {
3372         if (dedicated_hp_event_source) {
3373             drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3374                                   addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3375             spapr_hotplug_req_add_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3376                                                    nr_lmbs,
3377                                                    spapr_drc_index(drc));
3378         } else {
3379             spapr_hotplug_req_add_by_count(SPAPR_DR_CONNECTOR_TYPE_LMB,
3380                                            nr_lmbs);
3381         }
3382     }
3383 }
3384 
3385 static void spapr_memory_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3386                               Error **errp)
3387 {
3388     Error *local_err = NULL;
3389     sPAPRMachineState *ms = SPAPR_MACHINE(hotplug_dev);
3390     PCDIMMDevice *dimm = PC_DIMM(dev);
3391     uint64_t size, addr;
3392     uint32_t node;
3393 
3394     size = memory_device_get_region_size(MEMORY_DEVICE(dev), &error_abort);
3395 
3396     pc_dimm_plug(dimm, MACHINE(ms), &local_err);
3397     if (local_err) {
3398         goto out;
3399     }
3400 
3401     addr = object_property_get_uint(OBJECT(dimm),
3402                                     PC_DIMM_ADDR_PROP, &local_err);
3403     if (local_err) {
3404         goto out_unplug;
3405     }
3406 
3407     node = object_property_get_uint(OBJECT(dev), PC_DIMM_NODE_PROP,
3408                                     &error_abort);
3409     spapr_add_lmbs(dev, addr, size, node,
3410                    spapr_ovec_test(ms->ov5_cas, OV5_HP_EVT),
3411                    &local_err);
3412     if (local_err) {
3413         goto out_unplug;
3414     }
3415 
3416     return;
3417 
3418 out_unplug:
3419     pc_dimm_unplug(dimm, MACHINE(ms));
3420 out:
3421     error_propagate(errp, local_err);
3422 }
3423 
3424 static void spapr_memory_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3425                                   Error **errp)
3426 {
3427     const sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(hotplug_dev);
3428     sPAPRMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3429     PCDIMMDevice *dimm = PC_DIMM(dev);
3430     Error *local_err = NULL;
3431     uint64_t size;
3432     Object *memdev;
3433     hwaddr pagesize;
3434 
3435     if (!smc->dr_lmb_enabled) {
3436         error_setg(errp, "Memory hotplug not supported for this machine");
3437         return;
3438     }
3439 
3440     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &local_err);
3441     if (local_err) {
3442         error_propagate(errp, local_err);
3443         return;
3444     }
3445 
3446     if (size % SPAPR_MEMORY_BLOCK_SIZE) {
3447         error_setg(errp, "Hotplugged memory size must be a multiple of "
3448                       "%" PRIu64 " MB", SPAPR_MEMORY_BLOCK_SIZE / MiB);
3449         return;
3450     }
3451 
3452     memdev = object_property_get_link(OBJECT(dimm), PC_DIMM_MEMDEV_PROP,
3453                                       &error_abort);
3454     pagesize = host_memory_backend_pagesize(MEMORY_BACKEND(memdev));
3455     spapr_check_pagesize(spapr, pagesize, &local_err);
3456     if (local_err) {
3457         error_propagate(errp, local_err);
3458         return;
3459     }
3460 
3461     pc_dimm_pre_plug(dimm, MACHINE(hotplug_dev), NULL, errp);
3462 }
3463 
3464 struct sPAPRDIMMState {
3465     PCDIMMDevice *dimm;
3466     uint32_t nr_lmbs;
3467     QTAILQ_ENTRY(sPAPRDIMMState) next;
3468 };
3469 
3470 static sPAPRDIMMState *spapr_pending_dimm_unplugs_find(sPAPRMachineState *s,
3471                                                        PCDIMMDevice *dimm)
3472 {
3473     sPAPRDIMMState *dimm_state = NULL;
3474 
3475     QTAILQ_FOREACH(dimm_state, &s->pending_dimm_unplugs, next) {
3476         if (dimm_state->dimm == dimm) {
3477             break;
3478         }
3479     }
3480     return dimm_state;
3481 }
3482 
3483 static sPAPRDIMMState *spapr_pending_dimm_unplugs_add(sPAPRMachineState *spapr,
3484                                                       uint32_t nr_lmbs,
3485                                                       PCDIMMDevice *dimm)
3486 {
3487     sPAPRDIMMState *ds = NULL;
3488 
3489     /*
3490      * If this request is for a DIMM whose removal had failed earlier
3491      * (due to guest's refusal to remove the LMBs), we would have this
3492      * dimm already in the pending_dimm_unplugs list. In that
3493      * case don't add again.
3494      */
3495     ds = spapr_pending_dimm_unplugs_find(spapr, dimm);
3496     if (!ds) {
3497         ds = g_malloc0(sizeof(sPAPRDIMMState));
3498         ds->nr_lmbs = nr_lmbs;
3499         ds->dimm = dimm;
3500         QTAILQ_INSERT_HEAD(&spapr->pending_dimm_unplugs, ds, next);
3501     }
3502     return ds;
3503 }
3504 
3505 static void spapr_pending_dimm_unplugs_remove(sPAPRMachineState *spapr,
3506                                               sPAPRDIMMState *dimm_state)
3507 {
3508     QTAILQ_REMOVE(&spapr->pending_dimm_unplugs, dimm_state, next);
3509     g_free(dimm_state);
3510 }
3511 
3512 static sPAPRDIMMState *spapr_recover_pending_dimm_state(sPAPRMachineState *ms,
3513                                                         PCDIMMDevice *dimm)
3514 {
3515     sPAPRDRConnector *drc;
3516     uint64_t size = memory_device_get_region_size(MEMORY_DEVICE(dimm),
3517                                                   &error_abort);
3518     uint32_t nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3519     uint32_t avail_lmbs = 0;
3520     uint64_t addr_start, addr;
3521     int i;
3522 
3523     addr_start = object_property_get_int(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3524                                          &error_abort);
3525 
3526     addr = addr_start;
3527     for (i = 0; i < nr_lmbs; i++) {
3528         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3529                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3530         g_assert(drc);
3531         if (drc->dev) {
3532             avail_lmbs++;
3533         }
3534         addr += SPAPR_MEMORY_BLOCK_SIZE;
3535     }
3536 
3537     return spapr_pending_dimm_unplugs_add(ms, avail_lmbs, dimm);
3538 }
3539 
3540 /* Callback to be called during DRC release. */
3541 void spapr_lmb_release(DeviceState *dev)
3542 {
3543     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3544     sPAPRMachineState *spapr = SPAPR_MACHINE(hotplug_ctrl);
3545     sPAPRDIMMState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3546 
3547     /* This information will get lost if a migration occurs
3548      * during the unplug process. In this case recover it. */
3549     if (ds == NULL) {
3550         ds = spapr_recover_pending_dimm_state(spapr, PC_DIMM(dev));
3551         g_assert(ds);
3552         /* The DRC being examined by the caller at least must be counted */
3553         g_assert(ds->nr_lmbs);
3554     }
3555 
3556     if (--ds->nr_lmbs) {
3557         return;
3558     }
3559 
3560     /*
3561      * Now that all the LMBs have been removed by the guest, call the
3562      * unplug handler chain. This can never fail.
3563      */
3564     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3565 }
3566 
3567 static void spapr_memory_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3568 {
3569     sPAPRMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3570     sPAPRDIMMState *ds = spapr_pending_dimm_unplugs_find(spapr, PC_DIMM(dev));
3571 
3572     pc_dimm_unplug(PC_DIMM(dev), MACHINE(hotplug_dev));
3573     object_unparent(OBJECT(dev));
3574     spapr_pending_dimm_unplugs_remove(spapr, ds);
3575 }
3576 
3577 static void spapr_memory_unplug_request(HotplugHandler *hotplug_dev,
3578                                         DeviceState *dev, Error **errp)
3579 {
3580     sPAPRMachineState *spapr = SPAPR_MACHINE(hotplug_dev);
3581     Error *local_err = NULL;
3582     PCDIMMDevice *dimm = PC_DIMM(dev);
3583     uint32_t nr_lmbs;
3584     uint64_t size, addr_start, addr;
3585     int i;
3586     sPAPRDRConnector *drc;
3587 
3588     size = memory_device_get_region_size(MEMORY_DEVICE(dimm), &error_abort);
3589     nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
3590 
3591     addr_start = object_property_get_uint(OBJECT(dimm), PC_DIMM_ADDR_PROP,
3592                                          &local_err);
3593     if (local_err) {
3594         goto out;
3595     }
3596 
3597     /*
3598      * An existing pending dimm state for this DIMM means that there is an
3599      * unplug operation in progress, waiting for the spapr_lmb_release
3600      * callback to complete the job (BQL can't cover that far). In this case,
3601      * bail out to avoid detaching DRCs that were already released.
3602      */
3603     if (spapr_pending_dimm_unplugs_find(spapr, dimm)) {
3604         error_setg(&local_err,
3605                    "Memory unplug already in progress for device %s",
3606                    dev->id);
3607         goto out;
3608     }
3609 
3610     spapr_pending_dimm_unplugs_add(spapr, nr_lmbs, dimm);
3611 
3612     addr = addr_start;
3613     for (i = 0; i < nr_lmbs; i++) {
3614         drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3615                               addr / SPAPR_MEMORY_BLOCK_SIZE);
3616         g_assert(drc);
3617 
3618         spapr_drc_detach(drc);
3619         addr += SPAPR_MEMORY_BLOCK_SIZE;
3620     }
3621 
3622     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_LMB,
3623                           addr_start / SPAPR_MEMORY_BLOCK_SIZE);
3624     spapr_hotplug_req_remove_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
3625                                               nr_lmbs, spapr_drc_index(drc));
3626 out:
3627     error_propagate(errp, local_err);
3628 }
3629 
3630 static void *spapr_populate_hotplug_cpu_dt(CPUState *cs, int *fdt_offset,
3631                                            sPAPRMachineState *spapr)
3632 {
3633     PowerPCCPU *cpu = POWERPC_CPU(cs);
3634     DeviceClass *dc = DEVICE_GET_CLASS(cs);
3635     int id = spapr_get_vcpu_id(cpu);
3636     void *fdt;
3637     int offset, fdt_size;
3638     char *nodename;
3639 
3640     fdt = create_device_tree(&fdt_size);
3641     nodename = g_strdup_printf("%s@%x", dc->fw_name, id);
3642     offset = fdt_add_subnode(fdt, 0, nodename);
3643 
3644     spapr_populate_cpu_dt(cs, fdt, offset, spapr);
3645     g_free(nodename);
3646 
3647     *fdt_offset = offset;
3648     return fdt;
3649 }
3650 
3651 /* Callback to be called during DRC release. */
3652 void spapr_core_release(DeviceState *dev)
3653 {
3654     HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev);
3655 
3656     /* Call the unplug handler chain. This can never fail. */
3657     hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
3658 }
3659 
3660 static void spapr_core_unplug(HotplugHandler *hotplug_dev, DeviceState *dev)
3661 {
3662     MachineState *ms = MACHINE(hotplug_dev);
3663     sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(ms);
3664     CPUCore *cc = CPU_CORE(dev);
3665     CPUArchId *core_slot = spapr_find_cpu_slot(ms, cc->core_id, NULL);
3666 
3667     if (smc->pre_2_10_has_unused_icps) {
3668         sPAPRCPUCore *sc = SPAPR_CPU_CORE(OBJECT(dev));
3669         int i;
3670 
3671         for (i = 0; i < cc->nr_threads; i++) {
3672             CPUState *cs = CPU(sc->threads[i]);
3673 
3674             pre_2_10_vmstate_register_dummy_icp(cs->cpu_index);
3675         }
3676     }
3677 
3678     assert(core_slot);
3679     core_slot->cpu = NULL;
3680     object_unparent(OBJECT(dev));
3681 }
3682 
3683 static
3684 void spapr_core_unplug_request(HotplugHandler *hotplug_dev, DeviceState *dev,
3685                                Error **errp)
3686 {
3687     sPAPRMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3688     int index;
3689     sPAPRDRConnector *drc;
3690     CPUCore *cc = CPU_CORE(dev);
3691 
3692     if (!spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index)) {
3693         error_setg(errp, "Unable to find CPU core with core-id: %d",
3694                    cc->core_id);
3695         return;
3696     }
3697     if (index == 0) {
3698         error_setg(errp, "Boot CPU core may not be unplugged");
3699         return;
3700     }
3701 
3702     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3703                           spapr_vcpu_id(spapr, cc->core_id));
3704     g_assert(drc);
3705 
3706     spapr_drc_detach(drc);
3707 
3708     spapr_hotplug_req_remove_by_index(drc);
3709 }
3710 
3711 static void spapr_core_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3712                             Error **errp)
3713 {
3714     sPAPRMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
3715     MachineClass *mc = MACHINE_GET_CLASS(spapr);
3716     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3717     sPAPRCPUCore *core = SPAPR_CPU_CORE(OBJECT(dev));
3718     CPUCore *cc = CPU_CORE(dev);
3719     CPUState *cs = CPU(core->threads[0]);
3720     sPAPRDRConnector *drc;
3721     Error *local_err = NULL;
3722     CPUArchId *core_slot;
3723     int index;
3724     bool hotplugged = spapr_drc_hotplugged(dev);
3725 
3726     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3727     if (!core_slot) {
3728         error_setg(errp, "Unable to find CPU core with core-id: %d",
3729                    cc->core_id);
3730         return;
3731     }
3732     drc = spapr_drc_by_id(TYPE_SPAPR_DRC_CPU,
3733                           spapr_vcpu_id(spapr, cc->core_id));
3734 
3735     g_assert(drc || !mc->has_hotpluggable_cpus);
3736 
3737     if (drc) {
3738         void *fdt;
3739         int fdt_offset;
3740 
3741         fdt = spapr_populate_hotplug_cpu_dt(cs, &fdt_offset, spapr);
3742 
3743         spapr_drc_attach(drc, dev, fdt, fdt_offset, &local_err);
3744         if (local_err) {
3745             g_free(fdt);
3746             error_propagate(errp, local_err);
3747             return;
3748         }
3749 
3750         if (hotplugged) {
3751             /*
3752              * Send hotplug notification interrupt to the guest only
3753              * in case of hotplugged CPUs.
3754              */
3755             spapr_hotplug_req_add_by_index(drc);
3756         } else {
3757             spapr_drc_reset(drc);
3758         }
3759     }
3760 
3761     core_slot->cpu = OBJECT(dev);
3762 
3763     if (smc->pre_2_10_has_unused_icps) {
3764         int i;
3765 
3766         for (i = 0; i < cc->nr_threads; i++) {
3767             cs = CPU(core->threads[i]);
3768             pre_2_10_vmstate_unregister_dummy_icp(cs->cpu_index);
3769         }
3770     }
3771 }
3772 
3773 static void spapr_core_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
3774                                 Error **errp)
3775 {
3776     MachineState *machine = MACHINE(OBJECT(hotplug_dev));
3777     MachineClass *mc = MACHINE_GET_CLASS(hotplug_dev);
3778     Error *local_err = NULL;
3779     CPUCore *cc = CPU_CORE(dev);
3780     const char *base_core_type = spapr_get_cpu_core_type(machine->cpu_type);
3781     const char *type = object_get_typename(OBJECT(dev));
3782     CPUArchId *core_slot;
3783     int index;
3784 
3785     if (dev->hotplugged && !mc->has_hotpluggable_cpus) {
3786         error_setg(&local_err, "CPU hotplug not supported for this machine");
3787         goto out;
3788     }
3789 
3790     if (strcmp(base_core_type, type)) {
3791         error_setg(&local_err, "CPU core type should be %s", base_core_type);
3792         goto out;
3793     }
3794 
3795     if (cc->core_id % smp_threads) {
3796         error_setg(&local_err, "invalid core id %d", cc->core_id);
3797         goto out;
3798     }
3799 
3800     /*
3801      * In general we should have homogeneous threads-per-core, but old
3802      * (pre hotplug support) machine types allow the last core to have
3803      * reduced threads as a compatibility hack for when we allowed
3804      * total vcpus not a multiple of threads-per-core.
3805      */
3806     if (mc->has_hotpluggable_cpus && (cc->nr_threads != smp_threads)) {
3807         error_setg(&local_err, "invalid nr-threads %d, must be %d",
3808                    cc->nr_threads, smp_threads);
3809         goto out;
3810     }
3811 
3812     core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
3813     if (!core_slot) {
3814         error_setg(&local_err, "core id %d out of range", cc->core_id);
3815         goto out;
3816     }
3817 
3818     if (core_slot->cpu) {
3819         error_setg(&local_err, "core %d already populated", cc->core_id);
3820         goto out;
3821     }
3822 
3823     numa_cpu_pre_plug(core_slot, dev, &local_err);
3824 
3825 out:
3826     error_propagate(errp, local_err);
3827 }
3828 
3829 static void spapr_machine_device_plug(HotplugHandler *hotplug_dev,
3830                                       DeviceState *dev, Error **errp)
3831 {
3832     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
3833         spapr_memory_plug(hotplug_dev, dev, errp);
3834     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3835         spapr_core_plug(hotplug_dev, dev, errp);
3836     }
3837 }
3838 
3839 static void spapr_machine_device_unplug(HotplugHandler *hotplug_dev,
3840                                         DeviceState *dev, Error **errp)
3841 {
3842     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
3843         spapr_memory_unplug(hotplug_dev, dev);
3844     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3845         spapr_core_unplug(hotplug_dev, dev);
3846     }
3847 }
3848 
3849 static void spapr_machine_device_unplug_request(HotplugHandler *hotplug_dev,
3850                                                 DeviceState *dev, Error **errp)
3851 {
3852     sPAPRMachineState *sms = SPAPR_MACHINE(OBJECT(hotplug_dev));
3853     MachineClass *mc = MACHINE_GET_CLASS(sms);
3854 
3855     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
3856         if (spapr_ovec_test(sms->ov5_cas, OV5_HP_EVT)) {
3857             spapr_memory_unplug_request(hotplug_dev, dev, errp);
3858         } else {
3859             /* NOTE: this means there is a window after guest reset, prior to
3860              * CAS negotiation, where unplug requests will fail due to the
3861              * capability not being detected yet. This is a bit different than
3862              * the case with PCI unplug, where the events will be queued and
3863              * eventually handled by the guest after boot
3864              */
3865             error_setg(errp, "Memory hot unplug not supported for this guest");
3866         }
3867     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3868         if (!mc->has_hotpluggable_cpus) {
3869             error_setg(errp, "CPU hot unplug not supported on this machine");
3870             return;
3871         }
3872         spapr_core_unplug_request(hotplug_dev, dev, errp);
3873     }
3874 }
3875 
3876 static void spapr_machine_device_pre_plug(HotplugHandler *hotplug_dev,
3877                                           DeviceState *dev, Error **errp)
3878 {
3879     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
3880         spapr_memory_pre_plug(hotplug_dev, dev, errp);
3881     } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3882         spapr_core_pre_plug(hotplug_dev, dev, errp);
3883     }
3884 }
3885 
3886 static HotplugHandler *spapr_get_hotplug_handler(MachineState *machine,
3887                                                  DeviceState *dev)
3888 {
3889     if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM) ||
3890         object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
3891         return HOTPLUG_HANDLER(machine);
3892     }
3893     return NULL;
3894 }
3895 
3896 static CpuInstanceProperties
3897 spapr_cpu_index_to_props(MachineState *machine, unsigned cpu_index)
3898 {
3899     CPUArchId *core_slot;
3900     MachineClass *mc = MACHINE_GET_CLASS(machine);
3901 
3902     /* make sure possible_cpu are intialized */
3903     mc->possible_cpu_arch_ids(machine);
3904     /* get CPU core slot containing thread that matches cpu_index */
3905     core_slot = spapr_find_cpu_slot(machine, cpu_index, NULL);
3906     assert(core_slot);
3907     return core_slot->props;
3908 }
3909 
3910 static int64_t spapr_get_default_cpu_node_id(const MachineState *ms, int idx)
3911 {
3912     return idx / smp_cores % nb_numa_nodes;
3913 }
3914 
3915 static const CPUArchIdList *spapr_possible_cpu_arch_ids(MachineState *machine)
3916 {
3917     int i;
3918     const char *core_type;
3919     int spapr_max_cores = max_cpus / smp_threads;
3920     MachineClass *mc = MACHINE_GET_CLASS(machine);
3921 
3922     if (!mc->has_hotpluggable_cpus) {
3923         spapr_max_cores = QEMU_ALIGN_UP(smp_cpus, smp_threads) / smp_threads;
3924     }
3925     if (machine->possible_cpus) {
3926         assert(machine->possible_cpus->len == spapr_max_cores);
3927         return machine->possible_cpus;
3928     }
3929 
3930     core_type = spapr_get_cpu_core_type(machine->cpu_type);
3931     if (!core_type) {
3932         error_report("Unable to find sPAPR CPU Core definition");
3933         exit(1);
3934     }
3935 
3936     machine->possible_cpus = g_malloc0(sizeof(CPUArchIdList) +
3937                              sizeof(CPUArchId) * spapr_max_cores);
3938     machine->possible_cpus->len = spapr_max_cores;
3939     for (i = 0; i < machine->possible_cpus->len; i++) {
3940         int core_id = i * smp_threads;
3941 
3942         machine->possible_cpus->cpus[i].type = core_type;
3943         machine->possible_cpus->cpus[i].vcpus_count = smp_threads;
3944         machine->possible_cpus->cpus[i].arch_id = core_id;
3945         machine->possible_cpus->cpus[i].props.has_core_id = true;
3946         machine->possible_cpus->cpus[i].props.core_id = core_id;
3947     }
3948     return machine->possible_cpus;
3949 }
3950 
3951 static void spapr_phb_placement(sPAPRMachineState *spapr, uint32_t index,
3952                                 uint64_t *buid, hwaddr *pio,
3953                                 hwaddr *mmio32, hwaddr *mmio64,
3954                                 unsigned n_dma, uint32_t *liobns, Error **errp)
3955 {
3956     /*
3957      * New-style PHB window placement.
3958      *
3959      * Goals: Gives large (1TiB), naturally aligned 64-bit MMIO window
3960      * for each PHB, in addition to 2GiB 32-bit MMIO and 64kiB PIO
3961      * windows.
3962      *
3963      * Some guest kernels can't work with MMIO windows above 1<<46
3964      * (64TiB), so we place up to 31 PHBs in the area 32TiB..64TiB
3965      *
3966      * 32TiB..(33TiB+1984kiB) contains the 64kiB PIO windows for each
3967      * PHB stacked together.  (32TiB+2GiB)..(32TiB+64GiB) contains the
3968      * 2GiB 32-bit MMIO windows for each PHB.  Then 33..64TiB has the
3969      * 1TiB 64-bit MMIO windows for each PHB.
3970      */
3971     const uint64_t base_buid = 0x800000020000000ULL;
3972     int i;
3973 
3974     /* Sanity check natural alignments */
3975     QEMU_BUILD_BUG_ON((SPAPR_PCI_BASE % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
3976     QEMU_BUILD_BUG_ON((SPAPR_PCI_LIMIT % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
3977     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM64_WIN_SIZE % SPAPR_PCI_MEM32_WIN_SIZE) != 0);
3978     QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM32_WIN_SIZE % SPAPR_PCI_IO_WIN_SIZE) != 0);
3979     /* Sanity check bounds */
3980     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_IO_WIN_SIZE) >
3981                       SPAPR_PCI_MEM32_WIN_SIZE);
3982     QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_MEM32_WIN_SIZE) >
3983                       SPAPR_PCI_MEM64_WIN_SIZE);
3984 
3985     if (index >= SPAPR_MAX_PHBS) {
3986         error_setg(errp, "\"index\" for PAPR PHB is too large (max %llu)",
3987                    SPAPR_MAX_PHBS - 1);
3988         return;
3989     }
3990 
3991     *buid = base_buid + index;
3992     for (i = 0; i < n_dma; ++i) {
3993         liobns[i] = SPAPR_PCI_LIOBN(index, i);
3994     }
3995 
3996     *pio = SPAPR_PCI_BASE + index * SPAPR_PCI_IO_WIN_SIZE;
3997     *mmio32 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM32_WIN_SIZE;
3998     *mmio64 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM64_WIN_SIZE;
3999 }
4000 
4001 static ICSState *spapr_ics_get(XICSFabric *dev, int irq)
4002 {
4003     sPAPRMachineState *spapr = SPAPR_MACHINE(dev);
4004 
4005     return ics_valid_irq(spapr->ics, irq) ? spapr->ics : NULL;
4006 }
4007 
4008 static void spapr_ics_resend(XICSFabric *dev)
4009 {
4010     sPAPRMachineState *spapr = SPAPR_MACHINE(dev);
4011 
4012     ics_resend(spapr->ics);
4013 }
4014 
4015 static ICPState *spapr_icp_get(XICSFabric *xi, int vcpu_id)
4016 {
4017     PowerPCCPU *cpu = spapr_find_cpu(vcpu_id);
4018 
4019     return cpu ? spapr_cpu_state(cpu)->icp : NULL;
4020 }
4021 
4022 static void spapr_pic_print_info(InterruptStatsProvider *obj,
4023                                  Monitor *mon)
4024 {
4025     sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
4026 
4027     spapr->irq->print_info(spapr, mon);
4028 }
4029 
4030 int spapr_get_vcpu_id(PowerPCCPU *cpu)
4031 {
4032     return cpu->vcpu_id;
4033 }
4034 
4035 void spapr_set_vcpu_id(PowerPCCPU *cpu, int cpu_index, Error **errp)
4036 {
4037     sPAPRMachineState *spapr = SPAPR_MACHINE(qdev_get_machine());
4038     int vcpu_id;
4039 
4040     vcpu_id = spapr_vcpu_id(spapr, cpu_index);
4041 
4042     if (kvm_enabled() && !kvm_vcpu_id_is_valid(vcpu_id)) {
4043         error_setg(errp, "Can't create CPU with id %d in KVM", vcpu_id);
4044         error_append_hint(errp, "Adjust the number of cpus to %d "
4045                           "or try to raise the number of threads per core\n",
4046                           vcpu_id * smp_threads / spapr->vsmt);
4047         return;
4048     }
4049 
4050     cpu->vcpu_id = vcpu_id;
4051 }
4052 
4053 PowerPCCPU *spapr_find_cpu(int vcpu_id)
4054 {
4055     CPUState *cs;
4056 
4057     CPU_FOREACH(cs) {
4058         PowerPCCPU *cpu = POWERPC_CPU(cs);
4059 
4060         if (spapr_get_vcpu_id(cpu) == vcpu_id) {
4061             return cpu;
4062         }
4063     }
4064 
4065     return NULL;
4066 }
4067 
4068 static void spapr_machine_class_init(ObjectClass *oc, void *data)
4069 {
4070     MachineClass *mc = MACHINE_CLASS(oc);
4071     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(oc);
4072     FWPathProviderClass *fwc = FW_PATH_PROVIDER_CLASS(oc);
4073     NMIClass *nc = NMI_CLASS(oc);
4074     HotplugHandlerClass *hc = HOTPLUG_HANDLER_CLASS(oc);
4075     PPCVirtualHypervisorClass *vhc = PPC_VIRTUAL_HYPERVISOR_CLASS(oc);
4076     XICSFabricClass *xic = XICS_FABRIC_CLASS(oc);
4077     InterruptStatsProviderClass *ispc = INTERRUPT_STATS_PROVIDER_CLASS(oc);
4078 
4079     mc->desc = "pSeries Logical Partition (PAPR compliant)";
4080     mc->ignore_boot_device_suffixes = true;
4081 
4082     /*
4083      * We set up the default / latest behaviour here.  The class_init
4084      * functions for the specific versioned machine types can override
4085      * these details for backwards compatibility
4086      */
4087     mc->init = spapr_machine_init;
4088     mc->reset = spapr_machine_reset;
4089     mc->block_default_type = IF_SCSI;
4090     mc->max_cpus = 1024;
4091     mc->no_parallel = 1;
4092     mc->default_boot_order = "";
4093     mc->default_ram_size = 512 * MiB;
4094     mc->default_display = "std";
4095     mc->kvm_type = spapr_kvm_type;
4096     machine_class_allow_dynamic_sysbus_dev(mc, TYPE_SPAPR_PCI_HOST_BRIDGE);
4097     mc->pci_allow_0_address = true;
4098     assert(!mc->get_hotplug_handler);
4099     mc->get_hotplug_handler = spapr_get_hotplug_handler;
4100     hc->pre_plug = spapr_machine_device_pre_plug;
4101     hc->plug = spapr_machine_device_plug;
4102     mc->cpu_index_to_instance_props = spapr_cpu_index_to_props;
4103     mc->get_default_cpu_node_id = spapr_get_default_cpu_node_id;
4104     mc->possible_cpu_arch_ids = spapr_possible_cpu_arch_ids;
4105     hc->unplug_request = spapr_machine_device_unplug_request;
4106     hc->unplug = spapr_machine_device_unplug;
4107 
4108     smc->dr_lmb_enabled = true;
4109     smc->update_dt_enabled = true;
4110     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power9_v2.0");
4111     mc->has_hotpluggable_cpus = true;
4112     smc->resize_hpt_default = SPAPR_RESIZE_HPT_ENABLED;
4113     fwc->get_dev_path = spapr_get_fw_dev_path;
4114     nc->nmi_monitor_handler = spapr_nmi;
4115     smc->phb_placement = spapr_phb_placement;
4116     vhc->hypercall = emulate_spapr_hypercall;
4117     vhc->hpt_mask = spapr_hpt_mask;
4118     vhc->map_hptes = spapr_map_hptes;
4119     vhc->unmap_hptes = spapr_unmap_hptes;
4120     vhc->store_hpte = spapr_store_hpte;
4121     vhc->get_patbe = spapr_get_patbe;
4122     vhc->encode_hpt_for_kvm_pr = spapr_encode_hpt_for_kvm_pr;
4123     xic->ics_get = spapr_ics_get;
4124     xic->ics_resend = spapr_ics_resend;
4125     xic->icp_get = spapr_icp_get;
4126     ispc->print_info = spapr_pic_print_info;
4127     /* Force NUMA node memory size to be a multiple of
4128      * SPAPR_MEMORY_BLOCK_SIZE (256M) since that's the granularity
4129      * in which LMBs are represented and hot-added
4130      */
4131     mc->numa_mem_align_shift = 28;
4132 
4133     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_OFF;
4134     smc->default_caps.caps[SPAPR_CAP_VSX] = SPAPR_CAP_ON;
4135     smc->default_caps.caps[SPAPR_CAP_DFP] = SPAPR_CAP_ON;
4136     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_BROKEN;
4137     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_BROKEN;
4138     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_BROKEN;
4139     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 16; /* 64kiB */
4140     smc->default_caps.caps[SPAPR_CAP_NESTED_KVM_HV] = SPAPR_CAP_OFF;
4141     spapr_caps_add_properties(smc, &error_abort);
4142     smc->irq = &spapr_irq_xics;
4143 }
4144 
4145 static const TypeInfo spapr_machine_info = {
4146     .name          = TYPE_SPAPR_MACHINE,
4147     .parent        = TYPE_MACHINE,
4148     .abstract      = true,
4149     .instance_size = sizeof(sPAPRMachineState),
4150     .instance_init = spapr_instance_init,
4151     .instance_finalize = spapr_machine_finalizefn,
4152     .class_size    = sizeof(sPAPRMachineClass),
4153     .class_init    = spapr_machine_class_init,
4154     .interfaces = (InterfaceInfo[]) {
4155         { TYPE_FW_PATH_PROVIDER },
4156         { TYPE_NMI },
4157         { TYPE_HOTPLUG_HANDLER },
4158         { TYPE_PPC_VIRTUAL_HYPERVISOR },
4159         { TYPE_XICS_FABRIC },
4160         { TYPE_INTERRUPT_STATS_PROVIDER },
4161         { }
4162     },
4163 };
4164 
4165 #define DEFINE_SPAPR_MACHINE(suffix, verstr, latest)                 \
4166     static void spapr_machine_##suffix##_class_init(ObjectClass *oc, \
4167                                                     void *data)      \
4168     {                                                                \
4169         MachineClass *mc = MACHINE_CLASS(oc);                        \
4170         spapr_machine_##suffix##_class_options(mc);                  \
4171         if (latest) {                                                \
4172             mc->alias = "pseries";                                   \
4173             mc->is_default = 1;                                      \
4174         }                                                            \
4175     }                                                                \
4176     static const TypeInfo spapr_machine_##suffix##_info = {          \
4177         .name = MACHINE_TYPE_NAME("pseries-" verstr),                \
4178         .parent = TYPE_SPAPR_MACHINE,                                \
4179         .class_init = spapr_machine_##suffix##_class_init,           \
4180     };                                                               \
4181     static void spapr_machine_register_##suffix(void)                \
4182     {                                                                \
4183         type_register(&spapr_machine_##suffix##_info);               \
4184     }                                                                \
4185     type_init(spapr_machine_register_##suffix)
4186 
4187 /*
4188  * pseries-4.0
4189  */
4190 static void spapr_machine_4_0_class_options(MachineClass *mc)
4191 {
4192     /* Defaults for the latest behaviour inherited from the base class */
4193 }
4194 
4195 DEFINE_SPAPR_MACHINE(4_0, "4.0", true);
4196 
4197 /*
4198  * pseries-3.1
4199  */
4200 static void spapr_machine_3_1_class_options(MachineClass *mc)
4201 {
4202     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4203     static GlobalProperty compat[] = {
4204         { TYPE_SPAPR_MACHINE, "host-model", "passthrough" },
4205         { TYPE_SPAPR_MACHINE, "host-serial", "passthrough" },
4206     };
4207 
4208     spapr_machine_4_0_class_options(mc);
4209     compat_props_add(mc->compat_props, hw_compat_3_1, hw_compat_3_1_len);
4210     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4211 
4212     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power8_v2.0");
4213     smc->update_dt_enabled = false;
4214 }
4215 
4216 DEFINE_SPAPR_MACHINE(3_1, "3.1", false);
4217 
4218 /*
4219  * pseries-3.0
4220  */
4221 
4222 static void spapr_machine_3_0_class_options(MachineClass *mc)
4223 {
4224     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4225 
4226     spapr_machine_3_1_class_options(mc);
4227     compat_props_add(mc->compat_props, hw_compat_3_0, hw_compat_3_0_len);
4228 
4229     smc->legacy_irq_allocation = true;
4230     smc->irq = &spapr_irq_xics_legacy;
4231 }
4232 
4233 DEFINE_SPAPR_MACHINE(3_0, "3.0", false);
4234 
4235 /*
4236  * pseries-2.12
4237  */
4238 static void spapr_machine_2_12_class_options(MachineClass *mc)
4239 {
4240     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4241     static GlobalProperty compat[] = {
4242         { TYPE_POWERPC_CPU, "pre-3.0-migration", "on" },
4243         { TYPE_SPAPR_CPU_CORE, "pre-3.0-migration", "on" },
4244     };
4245 
4246     spapr_machine_3_0_class_options(mc);
4247     compat_props_add(mc->compat_props, hw_compat_2_12, hw_compat_2_12_len);
4248     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4249 
4250     /* We depend on kvm_enabled() to choose a default value for the
4251      * hpt-max-page-size capability. Of course we can't do it here
4252      * because this is too early and the HW accelerator isn't initialzed
4253      * yet. Postpone this to machine init (see default_caps_with_cpu()).
4254      */
4255     smc->default_caps.caps[SPAPR_CAP_HPT_MAXPAGESIZE] = 0;
4256 }
4257 
4258 DEFINE_SPAPR_MACHINE(2_12, "2.12", false);
4259 
4260 static void spapr_machine_2_12_sxxm_class_options(MachineClass *mc)
4261 {
4262     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4263 
4264     spapr_machine_2_12_class_options(mc);
4265     smc->default_caps.caps[SPAPR_CAP_CFPC] = SPAPR_CAP_WORKAROUND;
4266     smc->default_caps.caps[SPAPR_CAP_SBBC] = SPAPR_CAP_WORKAROUND;
4267     smc->default_caps.caps[SPAPR_CAP_IBS] = SPAPR_CAP_FIXED_CCD;
4268 }
4269 
4270 DEFINE_SPAPR_MACHINE(2_12_sxxm, "2.12-sxxm", false);
4271 
4272 /*
4273  * pseries-2.11
4274  */
4275 
4276 static void spapr_machine_2_11_class_options(MachineClass *mc)
4277 {
4278     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4279 
4280     spapr_machine_2_12_class_options(mc);
4281     smc->default_caps.caps[SPAPR_CAP_HTM] = SPAPR_CAP_ON;
4282     compat_props_add(mc->compat_props, hw_compat_2_11, hw_compat_2_11_len);
4283 }
4284 
4285 DEFINE_SPAPR_MACHINE(2_11, "2.11", false);
4286 
4287 /*
4288  * pseries-2.10
4289  */
4290 
4291 static void spapr_machine_2_10_class_options(MachineClass *mc)
4292 {
4293     spapr_machine_2_11_class_options(mc);
4294     compat_props_add(mc->compat_props, hw_compat_2_10, hw_compat_2_10_len);
4295 }
4296 
4297 DEFINE_SPAPR_MACHINE(2_10, "2.10", false);
4298 
4299 /*
4300  * pseries-2.9
4301  */
4302 
4303 static void spapr_machine_2_9_class_options(MachineClass *mc)
4304 {
4305     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4306     static GlobalProperty compat[] = {
4307         { TYPE_POWERPC_CPU, "pre-2.10-migration", "on" },
4308     };
4309 
4310     spapr_machine_2_10_class_options(mc);
4311     compat_props_add(mc->compat_props, hw_compat_2_9, hw_compat_2_9_len);
4312     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4313     mc->numa_auto_assign_ram = numa_legacy_auto_assign_ram;
4314     smc->pre_2_10_has_unused_icps = true;
4315     smc->resize_hpt_default = SPAPR_RESIZE_HPT_DISABLED;
4316 }
4317 
4318 DEFINE_SPAPR_MACHINE(2_9, "2.9", false);
4319 
4320 /*
4321  * pseries-2.8
4322  */
4323 
4324 static void spapr_machine_2_8_class_options(MachineClass *mc)
4325 {
4326     static GlobalProperty compat[] = {
4327         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pcie-extended-configuration-space", "off" },
4328     };
4329 
4330     spapr_machine_2_9_class_options(mc);
4331     compat_props_add(mc->compat_props, hw_compat_2_8, hw_compat_2_8_len);
4332     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4333     mc->numa_mem_align_shift = 23;
4334 }
4335 
4336 DEFINE_SPAPR_MACHINE(2_8, "2.8", false);
4337 
4338 /*
4339  * pseries-2.7
4340  */
4341 
4342 static void phb_placement_2_7(sPAPRMachineState *spapr, uint32_t index,
4343                               uint64_t *buid, hwaddr *pio,
4344                               hwaddr *mmio32, hwaddr *mmio64,
4345                               unsigned n_dma, uint32_t *liobns, Error **errp)
4346 {
4347     /* Legacy PHB placement for pseries-2.7 and earlier machine types */
4348     const uint64_t base_buid = 0x800000020000000ULL;
4349     const hwaddr phb_spacing = 0x1000000000ULL; /* 64 GiB */
4350     const hwaddr mmio_offset = 0xa0000000; /* 2 GiB + 512 MiB */
4351     const hwaddr pio_offset = 0x80000000; /* 2 GiB */
4352     const uint32_t max_index = 255;
4353     const hwaddr phb0_alignment = 0x10000000000ULL; /* 1 TiB */
4354 
4355     uint64_t ram_top = MACHINE(spapr)->ram_size;
4356     hwaddr phb0_base, phb_base;
4357     int i;
4358 
4359     /* Do we have device memory? */
4360     if (MACHINE(spapr)->maxram_size > ram_top) {
4361         /* Can't just use maxram_size, because there may be an
4362          * alignment gap between normal and device memory regions
4363          */
4364         ram_top = MACHINE(spapr)->device_memory->base +
4365             memory_region_size(&MACHINE(spapr)->device_memory->mr);
4366     }
4367 
4368     phb0_base = QEMU_ALIGN_UP(ram_top, phb0_alignment);
4369 
4370     if (index > max_index) {
4371         error_setg(errp, "\"index\" for PAPR PHB is too large (max %u)",
4372                    max_index);
4373         return;
4374     }
4375 
4376     *buid = base_buid + index;
4377     for (i = 0; i < n_dma; ++i) {
4378         liobns[i] = SPAPR_PCI_LIOBN(index, i);
4379     }
4380 
4381     phb_base = phb0_base + index * phb_spacing;
4382     *pio = phb_base + pio_offset;
4383     *mmio32 = phb_base + mmio_offset;
4384     /*
4385      * We don't set the 64-bit MMIO window, relying on the PHB's
4386      * fallback behaviour of automatically splitting a large "32-bit"
4387      * window into contiguous 32-bit and 64-bit windows
4388      */
4389 }
4390 
4391 static void spapr_machine_2_7_class_options(MachineClass *mc)
4392 {
4393     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4394     static GlobalProperty compat[] = {
4395         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem_win_size", "0xf80000000", },
4396         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem64_win_size", "0", },
4397         { TYPE_POWERPC_CPU, "pre-2.8-migration", "on", },
4398         { TYPE_SPAPR_PCI_HOST_BRIDGE, "pre-2.8-migration", "on", },
4399     };
4400 
4401     spapr_machine_2_8_class_options(mc);
4402     mc->default_cpu_type = POWERPC_CPU_TYPE_NAME("power7_v2.3");
4403     mc->default_machine_opts = "modern-hotplug-events=off";
4404     compat_props_add(mc->compat_props, hw_compat_2_7, hw_compat_2_7_len);
4405     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4406     smc->phb_placement = phb_placement_2_7;
4407 }
4408 
4409 DEFINE_SPAPR_MACHINE(2_7, "2.7", false);
4410 
4411 /*
4412  * pseries-2.6
4413  */
4414 
4415 static void spapr_machine_2_6_class_options(MachineClass *mc)
4416 {
4417     static GlobalProperty compat[] = {
4418         { TYPE_SPAPR_PCI_HOST_BRIDGE, "ddw", "off" },
4419     };
4420 
4421     spapr_machine_2_7_class_options(mc);
4422     mc->has_hotpluggable_cpus = false;
4423     compat_props_add(mc->compat_props, hw_compat_2_6, hw_compat_2_6_len);
4424     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4425 }
4426 
4427 DEFINE_SPAPR_MACHINE(2_6, "2.6", false);
4428 
4429 /*
4430  * pseries-2.5
4431  */
4432 
4433 static void spapr_machine_2_5_class_options(MachineClass *mc)
4434 {
4435     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4436     static GlobalProperty compat[] = {
4437         { "spapr-vlan", "use-rx-buffer-pools", "off" },
4438     };
4439 
4440     spapr_machine_2_6_class_options(mc);
4441     smc->use_ohci_by_default = true;
4442     compat_props_add(mc->compat_props, hw_compat_2_5, hw_compat_2_5_len);
4443     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4444 }
4445 
4446 DEFINE_SPAPR_MACHINE(2_5, "2.5", false);
4447 
4448 /*
4449  * pseries-2.4
4450  */
4451 
4452 static void spapr_machine_2_4_class_options(MachineClass *mc)
4453 {
4454     sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
4455 
4456     spapr_machine_2_5_class_options(mc);
4457     smc->dr_lmb_enabled = false;
4458     compat_props_add(mc->compat_props, hw_compat_2_4, hw_compat_2_4_len);
4459 }
4460 
4461 DEFINE_SPAPR_MACHINE(2_4, "2.4", false);
4462 
4463 /*
4464  * pseries-2.3
4465  */
4466 
4467 static void spapr_machine_2_3_class_options(MachineClass *mc)
4468 {
4469     static GlobalProperty compat[] = {
4470         { "spapr-pci-host-bridge", "dynamic-reconfiguration", "off" },
4471     };
4472     spapr_machine_2_4_class_options(mc);
4473     compat_props_add(mc->compat_props, hw_compat_2_3, hw_compat_2_3_len);
4474     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4475 }
4476 DEFINE_SPAPR_MACHINE(2_3, "2.3", false);
4477 
4478 /*
4479  * pseries-2.2
4480  */
4481 
4482 static void spapr_machine_2_2_class_options(MachineClass *mc)
4483 {
4484     static GlobalProperty compat[] = {
4485         { TYPE_SPAPR_PCI_HOST_BRIDGE, "mem_win_size", "0x20000000" },
4486     };
4487 
4488     spapr_machine_2_3_class_options(mc);
4489     compat_props_add(mc->compat_props, hw_compat_2_2, hw_compat_2_2_len);
4490     compat_props_add(mc->compat_props, compat, G_N_ELEMENTS(compat));
4491     mc->default_machine_opts = "modern-hotplug-events=off,suppress-vmdesc=on";
4492 }
4493 DEFINE_SPAPR_MACHINE(2_2, "2.2", false);
4494 
4495 /*
4496  * pseries-2.1
4497  */
4498 
4499 static void spapr_machine_2_1_class_options(MachineClass *mc)
4500 {
4501     spapr_machine_2_2_class_options(mc);
4502     compat_props_add(mc->compat_props, hw_compat_2_1, hw_compat_2_1_len);
4503 }
4504 DEFINE_SPAPR_MACHINE(2_1, "2.1", false);
4505 
4506 static void spapr_machine_register_types(void)
4507 {
4508     type_register_static(&spapr_machine_info);
4509 }
4510 
4511 type_init(spapr_machine_register_types)
4512