?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
spapr: reuse machine->possible_cpus instead of cores[]
[qemu/kevin.git] / hw / ppc / spapr.c
bloba0aa69ecb2994dd990804c8160ae44b8ce2ddb7e
1 /*
2 * QEMU PowerPC pSeries Logical Partition (aka sPAPR) hardware System Emulator
3 *
4 * Copyright (c) 2004-2007 Fabrice Bellard
5 * Copyright (c) 2007 Jocelyn Mayer
6 * Copyright (c) 2010 David Gibson, IBM Corporation.
7 *
8 * Permission is hereby granted, free of charge, to any person obtaining a copy
9 * of this software and associated documentation files (the "Software"), to deal
10 * in the Software without restriction, including without limitation the rights
11 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
12 * copies of the Software, and to permit persons to whom the Software is
13 * furnished to do so, subject to the following conditions:
14 *
15 * The above copyright notice and this permission notice shall be included in
16 * all copies or substantial portions of the Software.
17 *
18 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
19 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
20 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
21 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
22 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
23 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
24 * THE SOFTWARE.
25 *
26 */
27 #include "qemu/osdep.h"
28 #include "qapi/error.h"
29 #include "sysemu/sysemu.h"
30 #include "sysemu/numa.h"
31 #include "hw/hw.h"
32 #include "qemu/log.h"
33 #include "hw/fw-path-provider.h"
34 #include "elf.h"
35 #include "net/net.h"
36 #include "sysemu/device_tree.h"
37 #include "sysemu/block-backend.h"
38 #include "sysemu/cpus.h"
39 #include "sysemu/hw_accel.h"
40 #include "kvm_ppc.h"
41 #include "migration/migration.h"
42 #include "mmu-hash64.h"
43 #include "qom/cpu.h"
44
45 #include "hw/boards.h"
46 #include "hw/ppc/ppc.h"
47 #include "hw/loader.h"
48
49 #include "hw/ppc/fdt.h"
50 #include "hw/ppc/spapr.h"
51 #include "hw/ppc/spapr_vio.h"
52 #include "hw/pci-host/spapr.h"
53 #include "hw/ppc/xics.h"
54 #include "hw/pci/msi.h"
55
56 #include "hw/pci/pci.h"
57 #include "hw/scsi/scsi.h"
58 #include "hw/virtio/virtio-scsi.h"
59
60 #include "exec/address-spaces.h"
61 #include "hw/usb.h"
62 #include "qemu/config-file.h"
63 #include "qemu/error-report.h"
64 #include "trace.h"
65 #include "hw/nmi.h"
66
67 #include "hw/compat.h"
68 #include "qemu/cutils.h"
69 #include "hw/ppc/spapr_cpu_core.h"
70 #include "qmp-commands.h"
71
72 #include <libfdt.h>
73
74 /* SLOF memory layout:
75 *
76 * SLOF raw image loaded at 0, copies its romfs right below the flat
77 * device-tree, then position SLOF itself 31M below that
78 *
79 * So we set FW_OVERHEAD to 40MB which should account for all of that
80 * and more
81 *
82 * We load our kernel at 4M, leaving space for SLOF initial image
83 */
84 #define FDT_MAX_SIZE 0x100000
85 #define RTAS_MAX_SIZE 0x10000
86 #define RTAS_MAX_ADDR 0x80000000 /* RTAS must stay below that */
87 #define FW_MAX_SIZE 0x400000
88 #define FW_FILE_NAME "slof.bin"
89 #define FW_OVERHEAD 0x2800000
90 #define KERNEL_LOAD_ADDR FW_MAX_SIZE
91
92 #define MIN_RMA_SLOF 128UL
93
94 #define PHANDLE_XICP 0x00001111
95
96 #define HTAB_SIZE(spapr) (1ULL << ((spapr)->htab_shift))
97
98 static XICSState *try_create_xics(const char *type, int nr_servers,
99 int nr_irqs, Error **errp)
100 {
101 Error *err = NULL;
102 DeviceState *dev;
103
104 dev = qdev_create(NULL, type);
105 qdev_prop_set_uint32(dev, "nr_servers", nr_servers);
106 qdev_prop_set_uint32(dev, "nr_irqs", nr_irqs);
107 object_property_set_bool(OBJECT(dev), true, "realized", &err);
108 if (err) {
109 error_propagate(errp, err);
110 object_unparent(OBJECT(dev));
111 return NULL;
112 }
113 return XICS_COMMON(dev);
114 }
115
116 static XICSState *xics_system_init(MachineState *machine,
117 int nr_servers, int nr_irqs, Error **errp)
118 {
119 XICSState *xics = NULL;
120
121 if (kvm_enabled()) {
122 Error *err = NULL;
123
124 if (machine_kernel_irqchip_allowed(machine)) {
125 xics = try_create_xics(TYPE_XICS_SPAPR_KVM, nr_servers, nr_irqs,
126 &err);
127 }
128 if (machine_kernel_irqchip_required(machine) && !xics) {
129 error_reportf_err(err,
130 "kernel_irqchip requested but unavailable: ");
131 } else {
132 error_free(err);
133 }
134 }
135
136 if (!xics) {
137 xics = try_create_xics(TYPE_XICS_SPAPR, nr_servers, nr_irqs, errp);
138 }
139
140 return xics;
141 }
142
143 static int spapr_fixup_cpu_smt_dt(void *fdt, int offset, PowerPCCPU *cpu,
144 int smt_threads)
145 {
146 int i, ret = 0;
147 uint32_t servers_prop[smt_threads];
148 uint32_t gservers_prop[smt_threads * 2];
149 int index = ppc_get_vcpu_dt_id(cpu);
150
151 if (cpu->compat_pvr) {
152 ret = fdt_setprop_cell(fdt, offset, "cpu-version", cpu->compat_pvr);
153 if (ret < 0) {
154 return ret;
155 }
156 }
157
158 /* Build interrupt servers and gservers properties */
159 for (i = 0; i < smt_threads; i++) {
160 servers_prop[i] = cpu_to_be32(index + i);
161 /* Hack, direct the group queues back to cpu 0 */
162 gservers_prop[i*2] = cpu_to_be32(index + i);
163 gservers_prop[i*2 + 1] = 0;
164 }
165 ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-server#s",
166 servers_prop, sizeof(servers_prop));
167 if (ret < 0) {
168 return ret;
169 }
170 ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-gserver#s",
171 gservers_prop, sizeof(gservers_prop));
172
173 return ret;
174 }
175
176 static int spapr_fixup_cpu_numa_dt(void *fdt, int offset, CPUState *cs)
177 {
178 int ret = 0;
179 PowerPCCPU *cpu = POWERPC_CPU(cs);
180 int index = ppc_get_vcpu_dt_id(cpu);
181 uint32_t associativity[] = {cpu_to_be32(0x5),
182 cpu_to_be32(0x0),
183 cpu_to_be32(0x0),
184 cpu_to_be32(0x0),
185 cpu_to_be32(cs->numa_node),
186 cpu_to_be32(index)};
187
188 /* Advertise NUMA via ibm,associativity */
189 if (nb_numa_nodes > 1) {
190 ret = fdt_setprop(fdt, offset, "ibm,associativity", associativity,
191 sizeof(associativity));
192 }
193
194 return ret;
195 }
196
197 static int spapr_fixup_cpu_dt(void *fdt, sPAPRMachineState *spapr)
198 {
199 int ret = 0, offset, cpus_offset;
200 CPUState *cs;
201 char cpu_model[32];
202 int smt = kvmppc_smt_threads();
203 uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
204
205 CPU_FOREACH(cs) {
206 PowerPCCPU *cpu = POWERPC_CPU(cs);
207 DeviceClass *dc = DEVICE_GET_CLASS(cs);
208 int index = ppc_get_vcpu_dt_id(cpu);
209 int compat_smt = MIN(smp_threads, ppc_compat_max_threads(cpu));
210
211 if ((index % smt) != 0) {
212 continue;
213 }
214
215 snprintf(cpu_model, 32, "%s@%x", dc->fw_name, index);
216
217 cpus_offset = fdt_path_offset(fdt, "/cpus");
218 if (cpus_offset < 0) {
219 cpus_offset = fdt_add_subnode(fdt, fdt_path_offset(fdt, "/"),
220 "cpus");
221 if (cpus_offset < 0) {
222 return cpus_offset;
223 }
224 }
225 offset = fdt_subnode_offset(fdt, cpus_offset, cpu_model);
226 if (offset < 0) {
227 offset = fdt_add_subnode(fdt, cpus_offset, cpu_model);
228 if (offset < 0) {
229 return offset;
230 }
231 }
232
233 ret = fdt_setprop(fdt, offset, "ibm,pft-size",
234 pft_size_prop, sizeof(pft_size_prop));
235 if (ret < 0) {
236 return ret;
237 }
238
239 ret = spapr_fixup_cpu_numa_dt(fdt, offset, cs);
240 if (ret < 0) {
241 return ret;
242 }
243
244 ret = spapr_fixup_cpu_smt_dt(fdt, offset, cpu, compat_smt);
245 if (ret < 0) {
246 return ret;
247 }
248 }
249 return ret;
250 }
251
252 static hwaddr spapr_node0_size(void)
253 {
254 MachineState *machine = MACHINE(qdev_get_machine());
255
256 if (nb_numa_nodes) {
257 int i;
258 for (i = 0; i < nb_numa_nodes; ++i) {
259 if (numa_info[i].node_mem) {
260 return MIN(pow2floor(numa_info[i].node_mem),
261 machine->ram_size);
262 }
263 }
264 }
265 return machine->ram_size;
266 }
267
268 static void add_str(GString *s, const gchar *s1)
269 {
270 g_string_append_len(s, s1, strlen(s1) + 1);
271 }
272
273 static int spapr_populate_memory_node(void *fdt, int nodeid, hwaddr start,
274 hwaddr size)
275 {
276 uint32_t associativity[] = {
277 cpu_to_be32(0x4), /* length */
278 cpu_to_be32(0x0), cpu_to_be32(0x0),
279 cpu_to_be32(0x0), cpu_to_be32(nodeid)
280 };
281 char mem_name[32];
282 uint64_t mem_reg_property[2];
283 int off;
284
285 mem_reg_property[0] = cpu_to_be64(start);
286 mem_reg_property[1] = cpu_to_be64(size);
287
288 sprintf(mem_name, "memory@" TARGET_FMT_lx, start);
289 off = fdt_add_subnode(fdt, 0, mem_name);
290 _FDT(off);
291 _FDT((fdt_setprop_string(fdt, off, "device_type", "memory")));
292 _FDT((fdt_setprop(fdt, off, "reg", mem_reg_property,
293 sizeof(mem_reg_property))));
294 _FDT((fdt_setprop(fdt, off, "ibm,associativity", associativity,
295 sizeof(associativity))));
296 return off;
297 }
298
299 static int spapr_populate_memory(sPAPRMachineState *spapr, void *fdt)
300 {
301 MachineState *machine = MACHINE(spapr);
302 hwaddr mem_start, node_size;
303 int i, nb_nodes = nb_numa_nodes;
304 NodeInfo *nodes = numa_info;
305 NodeInfo ramnode;
306
307 /* No NUMA nodes, assume there is just one node with whole RAM */
308 if (!nb_numa_nodes) {
309 nb_nodes = 1;
310 ramnode.node_mem = machine->ram_size;
311 nodes = &ramnode;
312 }
313
314 for (i = 0, mem_start = 0; i < nb_nodes; ++i) {
315 if (!nodes[i].node_mem) {
316 continue;
317 }
318 if (mem_start >= machine->ram_size) {
319 node_size = 0;
320 } else {
321 node_size = nodes[i].node_mem;
322 if (node_size > machine->ram_size - mem_start) {
323 node_size = machine->ram_size - mem_start;
324 }
325 }
326 if (!mem_start) {
327 /* ppc_spapr_init() checks for rma_size <= node0_size already */
328 spapr_populate_memory_node(fdt, i, 0, spapr->rma_size);
329 mem_start += spapr->rma_size;
330 node_size -= spapr->rma_size;
331 }
332 for ( ; node_size; ) {
333 hwaddr sizetmp = pow2floor(node_size);
334
335 /* mem_start != 0 here */
336 if (ctzl(mem_start) < ctzl(sizetmp)) {
337 sizetmp = 1ULL << ctzl(mem_start);
338 }
339
340 spapr_populate_memory_node(fdt, i, mem_start, sizetmp);
341 node_size -= sizetmp;
342 mem_start += sizetmp;
343 }
344 }
345
346 return 0;
347 }
348
349 /* Populate the "ibm,pa-features" property */
350 static void spapr_populate_pa_features(CPUPPCState *env, void *fdt, int offset)
351 {
352 uint8_t pa_features_206[] = { 6, 0,
353 0xf6, 0x1f, 0xc7, 0x00, 0x80, 0xc0 };
354 uint8_t pa_features_207[] = { 24, 0,
355 0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0,
356 0x80, 0x00, 0x00, 0x00, 0x00, 0x00,
357 0x00, 0x00, 0x00, 0x00, 0x80, 0x00,
358 0x80, 0x00, 0x80, 0x00, 0x00, 0x00 };
359 uint8_t *pa_features;
360 size_t pa_size;
361
362 switch (env->mmu_model) {
363 case POWERPC_MMU_2_06:
364 case POWERPC_MMU_2_06a:
365 pa_features = pa_features_206;
366 pa_size = sizeof(pa_features_206);
367 break;
368 case POWERPC_MMU_2_07:
369 case POWERPC_MMU_2_07a:
370 pa_features = pa_features_207;
371 pa_size = sizeof(pa_features_207);
372 break;
373 default:
374 return;
375 }
376
377 if (env->ci_large_pages) {
378 /*
379 * Note: we keep CI large pages off by default because a 64K capable
380 * guest provisioned with large pages might otherwise try to map a qemu
381 * framebuffer (or other kind of memory mapped PCI BAR) using 64K pages
382 * even if that qemu runs on a 4k host.
383 * We dd this bit back here if we are confident this is not an issue
384 */
385 pa_features[3] |= 0x20;
386 }
387 if (kvmppc_has_cap_htm() && pa_size > 24) {
388 pa_features[24] |= 0x80; /* Transactional memory support */
389 }
390
391 _FDT((fdt_setprop(fdt, offset, "ibm,pa-features", pa_features, pa_size)));
392 }
393
394 static void spapr_populate_cpu_dt(CPUState *cs, void *fdt, int offset,
395 sPAPRMachineState *spapr)
396 {
397 PowerPCCPU *cpu = POWERPC_CPU(cs);
398 CPUPPCState *env = &cpu->env;
399 PowerPCCPUClass *pcc = POWERPC_CPU_GET_CLASS(cs);
400 int index = ppc_get_vcpu_dt_id(cpu);
401 uint32_t segs[] = {cpu_to_be32(28), cpu_to_be32(40),
402 0xffffffff, 0xffffffff};
403 uint32_t tbfreq = kvm_enabled() ? kvmppc_get_tbfreq()
404 : SPAPR_TIMEBASE_FREQ;
405 uint32_t cpufreq = kvm_enabled() ? kvmppc_get_clockfreq() : 1000000000;
406 uint32_t page_sizes_prop[64];
407 size_t page_sizes_prop_size;
408 uint32_t vcpus_per_socket = smp_threads * smp_cores;
409 uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
410 int compat_smt = MIN(smp_threads, ppc_compat_max_threads(cpu));
411 sPAPRDRConnector *drc;
412 sPAPRDRConnectorClass *drck;
413 int drc_index;
414
415 drc = spapr_dr_connector_by_id(SPAPR_DR_CONNECTOR_TYPE_CPU, index);
416 if (drc) {
417 drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
418 drc_index = drck->get_index(drc);
419 _FDT((fdt_setprop_cell(fdt, offset, "ibm,my-drc-index", drc_index)));
420 }
421
422 _FDT((fdt_setprop_cell(fdt, offset, "reg", index)));
423 _FDT((fdt_setprop_string(fdt, offset, "device_type", "cpu")));
424
425 _FDT((fdt_setprop_cell(fdt, offset, "cpu-version", env->spr[SPR_PVR])));
426 _FDT((fdt_setprop_cell(fdt, offset, "d-cache-block-size",
427 env->dcache_line_size)));
428 _FDT((fdt_setprop_cell(fdt, offset, "d-cache-line-size",
429 env->dcache_line_size)));
430 _FDT((fdt_setprop_cell(fdt, offset, "i-cache-block-size",
431 env->icache_line_size)));
432 _FDT((fdt_setprop_cell(fdt, offset, "i-cache-line-size",
433 env->icache_line_size)));
434
435 if (pcc->l1_dcache_size) {
436 _FDT((fdt_setprop_cell(fdt, offset, "d-cache-size",
437 pcc->l1_dcache_size)));
438 } else {
439 error_report("Warning: Unknown L1 dcache size for cpu");
440 }
441 if (pcc->l1_icache_size) {
442 _FDT((fdt_setprop_cell(fdt, offset, "i-cache-size",
443 pcc->l1_icache_size)));
444 } else {
445 error_report("Warning: Unknown L1 icache size for cpu");
446 }
447
448 _FDT((fdt_setprop_cell(fdt, offset, "timebase-frequency", tbfreq)));
449 _FDT((fdt_setprop_cell(fdt, offset, "clock-frequency", cpufreq)));
450 _FDT((fdt_setprop_cell(fdt, offset, "slb-size", env->slb_nr)));
451 _FDT((fdt_setprop_cell(fdt, offset, "ibm,slb-size", env->slb_nr)));
452 _FDT((fdt_setprop_string(fdt, offset, "status", "okay")));
453 _FDT((fdt_setprop(fdt, offset, "64-bit", NULL, 0)));
454
455 if (env->spr_cb[SPR_PURR].oea_read) {
456 _FDT((fdt_setprop(fdt, offset, "ibm,purr", NULL, 0)));
457 }
458
459 if (env->mmu_model & POWERPC_MMU_1TSEG) {
460 _FDT((fdt_setprop(fdt, offset, "ibm,processor-segment-sizes",
461 segs, sizeof(segs))));
462 }
463
464 /* Advertise VMX/VSX (vector extensions) if available
465 * 0 / no property == no vector extensions
466 * 1 == VMX / Altivec available
467 * 2 == VSX available */
468 if (env->insns_flags & PPC_ALTIVEC) {
469 uint32_t vmx = (env->insns_flags2 & PPC2_VSX) ? 2 : 1;
470
471 _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", vmx)));
472 }
473
474 /* Advertise DFP (Decimal Floating Point) if available
475 * 0 / no property == no DFP
476 * 1 == DFP available */
477 if (env->insns_flags2 & PPC2_DFP) {
478 _FDT((fdt_setprop_cell(fdt, offset, "ibm,dfp", 1)));
479 }
480
481 page_sizes_prop_size = ppc_create_page_sizes_prop(env, page_sizes_prop,
482 sizeof(page_sizes_prop));
483 if (page_sizes_prop_size) {
484 _FDT((fdt_setprop(fdt, offset, "ibm,segment-page-sizes",
485 page_sizes_prop, page_sizes_prop_size)));
486 }
487
488 spapr_populate_pa_features(env, fdt, offset);
489
490 _FDT((fdt_setprop_cell(fdt, offset, "ibm,chip-id",
491 cs->cpu_index / vcpus_per_socket)));
492
493 _FDT((fdt_setprop(fdt, offset, "ibm,pft-size",
494 pft_size_prop, sizeof(pft_size_prop))));
495
496 _FDT(spapr_fixup_cpu_numa_dt(fdt, offset, cs));
497
498 _FDT(spapr_fixup_cpu_smt_dt(fdt, offset, cpu, compat_smt));
499 }
500
501 static void spapr_populate_cpus_dt_node(void *fdt, sPAPRMachineState *spapr)
502 {
503 CPUState *cs;
504 int cpus_offset;
505 char *nodename;
506 int smt = kvmppc_smt_threads();
507
508 cpus_offset = fdt_add_subnode(fdt, 0, "cpus");
509 _FDT(cpus_offset);
510 _FDT((fdt_setprop_cell(fdt, cpus_offset, "#address-cells", 0x1)));
511 _FDT((fdt_setprop_cell(fdt, cpus_offset, "#size-cells", 0x0)));
512
513 /*
514 * We walk the CPUs in reverse order to ensure that CPU DT nodes
515 * created by fdt_add_subnode() end up in the right order in FDT
516 * for the guest kernel the enumerate the CPUs correctly.
517 */
518 CPU_FOREACH_REVERSE(cs) {
519 PowerPCCPU *cpu = POWERPC_CPU(cs);
520 int index = ppc_get_vcpu_dt_id(cpu);
521 DeviceClass *dc = DEVICE_GET_CLASS(cs);
522 int offset;
523
524 if ((index % smt) != 0) {
525 continue;
526 }
527
528 nodename = g_strdup_printf("%s@%x", dc->fw_name, index);
529 offset = fdt_add_subnode(fdt, cpus_offset, nodename);
530 g_free(nodename);
531 _FDT(offset);
532 spapr_populate_cpu_dt(cs, fdt, offset, spapr);
533 }
534
535 }
536
537 /*
538 * Adds ibm,dynamic-reconfiguration-memory node.
539 * Refer to docs/specs/ppc-spapr-hotplug.txt for the documentation
540 * of this device tree node.
541 */
542 static int spapr_populate_drconf_memory(sPAPRMachineState *spapr, void *fdt)
543 {
544 MachineState *machine = MACHINE(spapr);
545 int ret, i, offset;
546 uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
547 uint32_t prop_lmb_size[] = {0, cpu_to_be32(lmb_size)};
548 uint32_t hotplug_lmb_start = spapr->hotplug_memory.base / lmb_size;
549 uint32_t nr_lmbs = (spapr->hotplug_memory.base +
550 memory_region_size(&spapr->hotplug_memory.mr)) /
551 lmb_size;
552 uint32_t *int_buf, *cur_index, buf_len;
553 int nr_nodes = nb_numa_nodes ? nb_numa_nodes : 1;
554
555 /*
556 * Don't create the node if there is no hotpluggable memory
557 */
558 if (machine->ram_size == machine->maxram_size) {
559 return 0;
560 }
561
562 /*
563 * Allocate enough buffer size to fit in ibm,dynamic-memory
564 * or ibm,associativity-lookup-arrays
565 */
566 buf_len = MAX(nr_lmbs * SPAPR_DR_LMB_LIST_ENTRY_SIZE + 1, nr_nodes * 4 + 2)
567 * sizeof(uint32_t);
568 cur_index = int_buf = g_malloc0(buf_len);
569
570 offset = fdt_add_subnode(fdt, 0, "ibm,dynamic-reconfiguration-memory");
571
572 ret = fdt_setprop(fdt, offset, "ibm,lmb-size", prop_lmb_size,
573 sizeof(prop_lmb_size));
574 if (ret < 0) {
575 goto out;
576 }
577
578 ret = fdt_setprop_cell(fdt, offset, "ibm,memory-flags-mask", 0xff);
579 if (ret < 0) {
580 goto out;
581 }
582
583 ret = fdt_setprop_cell(fdt, offset, "ibm,memory-preservation-time", 0x0);
584 if (ret < 0) {
585 goto out;
586 }
587
588 /* ibm,dynamic-memory */
589 int_buf[0] = cpu_to_be32(nr_lmbs);
590 cur_index++;
591 for (i = 0; i < nr_lmbs; i++) {
592 uint64_t addr = i * lmb_size;
593 uint32_t *dynamic_memory = cur_index;
594
595 if (i >= hotplug_lmb_start) {
596 sPAPRDRConnector *drc;
597 sPAPRDRConnectorClass *drck;
598
599 drc = spapr_dr_connector_by_id(SPAPR_DR_CONNECTOR_TYPE_LMB, i);
600 g_assert(drc);
601 drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
602
603 dynamic_memory[0] = cpu_to_be32(addr >> 32);
604 dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
605 dynamic_memory[2] = cpu_to_be32(drck->get_index(drc));
606 dynamic_memory[3] = cpu_to_be32(0); /* reserved */
607 dynamic_memory[4] = cpu_to_be32(numa_get_node(addr, NULL));
608 if (memory_region_present(get_system_memory(), addr)) {
609 dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_ASSIGNED);
610 } else {
611 dynamic_memory[5] = cpu_to_be32(0);
612 }
613 } else {
614 /*
615 * LMB information for RMA, boot time RAM and gap b/n RAM and
616 * hotplug memory region -- all these are marked as reserved
617 * and as having no valid DRC.
618 */
619 dynamic_memory[0] = cpu_to_be32(addr >> 32);
620 dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
621 dynamic_memory[2] = cpu_to_be32(0);
622 dynamic_memory[3] = cpu_to_be32(0); /* reserved */
623 dynamic_memory[4] = cpu_to_be32(-1);
624 dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_RESERVED |
625 SPAPR_LMB_FLAGS_DRC_INVALID);
626 }
627
628 cur_index += SPAPR_DR_LMB_LIST_ENTRY_SIZE;
629 }
630 ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory", int_buf, buf_len);
631 if (ret < 0) {
632 goto out;
633 }
634
635 /* ibm,associativity-lookup-arrays */
636 cur_index = int_buf;
637 int_buf[0] = cpu_to_be32(nr_nodes);
638 int_buf[1] = cpu_to_be32(4); /* Number of entries per associativity list */
639 cur_index += 2;
640 for (i = 0; i < nr_nodes; i++) {
641 uint32_t associativity[] = {
642 cpu_to_be32(0x0),
643 cpu_to_be32(0x0),
644 cpu_to_be32(0x0),
645 cpu_to_be32(i)
646 };
647 memcpy(cur_index, associativity, sizeof(associativity));
648 cur_index += 4;
649 }
650 ret = fdt_setprop(fdt, offset, "ibm,associativity-lookup-arrays", int_buf,
651 (cur_index - int_buf) * sizeof(uint32_t));
652 out:
653 g_free(int_buf);
654 return ret;
655 }
656
657 static int spapr_dt_cas_updates(sPAPRMachineState *spapr, void *fdt,
658 sPAPROptionVector *ov5_updates)
659 {
660 sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
661 int ret = 0, offset;
662
663 /* Generate ibm,dynamic-reconfiguration-memory node if required */
664 if (spapr_ovec_test(ov5_updates, OV5_DRCONF_MEMORY)) {
665 g_assert(smc->dr_lmb_enabled);
666 ret = spapr_populate_drconf_memory(spapr, fdt);
667 if (ret) {
668 goto out;
669 }
670 }
671
672 offset = fdt_path_offset(fdt, "/chosen");
673 if (offset < 0) {
674 offset = fdt_add_subnode(fdt, 0, "chosen");
675 if (offset < 0) {
676 return offset;
677 }
678 }
679 ret = spapr_ovec_populate_dt(fdt, offset, spapr->ov5_cas,
680 "ibm,architecture-vec-5");
681
682 out:
683 return ret;
684 }
685
686 int spapr_h_cas_compose_response(sPAPRMachineState *spapr,
687 target_ulong addr, target_ulong size,
688 sPAPROptionVector *ov5_updates)
689 {
690 void *fdt, *fdt_skel;
691 sPAPRDeviceTreeUpdateHeader hdr = { .version_id = 1 };
692
693 size -= sizeof(hdr);
694
695 /* Create sceleton */
696 fdt_skel = g_malloc0(size);
697 _FDT((fdt_create(fdt_skel, size)));
698 _FDT((fdt_begin_node(fdt_skel, "")));
699 _FDT((fdt_end_node(fdt_skel)));
700 _FDT((fdt_finish(fdt_skel)));
701 fdt = g_malloc0(size);
702 _FDT((fdt_open_into(fdt_skel, fdt, size)));
703 g_free(fdt_skel);
704
705 /* Fixup cpu nodes */
706 _FDT((spapr_fixup_cpu_dt(fdt, spapr)));
707
708 if (spapr_dt_cas_updates(spapr, fdt, ov5_updates)) {
709 return -1;
710 }
711
712 /* Pack resulting tree */
713 _FDT((fdt_pack(fdt)));
714
715 if (fdt_totalsize(fdt) + sizeof(hdr) > size) {
716 trace_spapr_cas_failed(size);
717 return -1;
718 }
719
720 cpu_physical_memory_write(addr, &hdr, sizeof(hdr));
721 cpu_physical_memory_write(addr + sizeof(hdr), fdt, fdt_totalsize(fdt));
722 trace_spapr_cas_continue(fdt_totalsize(fdt) + sizeof(hdr));
723 g_free(fdt);
724
725 return 0;
726 }
727
728 static void spapr_dt_rtas(sPAPRMachineState *spapr, void *fdt)
729 {
730 int rtas;
731 GString *hypertas = g_string_sized_new(256);
732 GString *qemu_hypertas = g_string_sized_new(256);
733 uint32_t refpoints[] = { cpu_to_be32(0x4), cpu_to_be32(0x4) };
734 uint64_t max_hotplug_addr = spapr->hotplug_memory.base +
735 memory_region_size(&spapr->hotplug_memory.mr);
736 uint32_t lrdr_capacity[] = {
737 cpu_to_be32(max_hotplug_addr >> 32),
738 cpu_to_be32(max_hotplug_addr & 0xffffffff),
739 0, cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE),
740 cpu_to_be32(max_cpus / smp_threads),
741 };
742
743 _FDT(rtas = fdt_add_subnode(fdt, 0, "rtas"));
744
745 /* hypertas */
746 add_str(hypertas, "hcall-pft");
747 add_str(hypertas, "hcall-term");
748 add_str(hypertas, "hcall-dabr");
749 add_str(hypertas, "hcall-interrupt");
750 add_str(hypertas, "hcall-tce");
751 add_str(hypertas, "hcall-vio");
752 add_str(hypertas, "hcall-splpar");
753 add_str(hypertas, "hcall-bulk");
754 add_str(hypertas, "hcall-set-mode");
755 add_str(hypertas, "hcall-sprg0");
756 add_str(hypertas, "hcall-copy");
757 add_str(hypertas, "hcall-debug");
758 add_str(qemu_hypertas, "hcall-memop1");
759
760 if (!kvm_enabled() || kvmppc_spapr_use_multitce()) {
761 add_str(hypertas, "hcall-multi-tce");
762 }
763 _FDT(fdt_setprop(fdt, rtas, "ibm,hypertas-functions",
764 hypertas->str, hypertas->len));
765 g_string_free(hypertas, TRUE);
766 _FDT(fdt_setprop(fdt, rtas, "qemu,hypertas-functions",
767 qemu_hypertas->str, qemu_hypertas->len));
768 g_string_free(qemu_hypertas, TRUE);
769
770 _FDT(fdt_setprop(fdt, rtas, "ibm,associativity-reference-points",
771 refpoints, sizeof(refpoints)));
772
773 _FDT(fdt_setprop_cell(fdt, rtas, "rtas-error-log-max",
774 RTAS_ERROR_LOG_MAX));
775 _FDT(fdt_setprop_cell(fdt, rtas, "rtas-event-scan-rate",
776 RTAS_EVENT_SCAN_RATE));
777
778 if (msi_nonbroken) {
779 _FDT(fdt_setprop(fdt, rtas, "ibm,change-msix-capable", NULL, 0));
780 }
781
782 /*
783 * According to PAPR, rtas ibm,os-term does not guarantee a return
784 * back to the guest cpu.
785 *
786 * While an additional ibm,extended-os-term property indicates
787 * that rtas call return will always occur. Set this property.
788 */
789 _FDT(fdt_setprop(fdt, rtas, "ibm,extended-os-term", NULL, 0));
790
791 _FDT(fdt_setprop(fdt, rtas, "ibm,lrdr-capacity",
792 lrdr_capacity, sizeof(lrdr_capacity)));
793
794 spapr_dt_rtas_tokens(fdt, rtas);
795 }
796
797 static void spapr_dt_chosen(sPAPRMachineState *spapr, void *fdt)
798 {
799 MachineState *machine = MACHINE(spapr);
800 int chosen;
801 const char *boot_device = machine->boot_order;
802 char *stdout_path = spapr_vio_stdout_path(spapr->vio_bus);
803 size_t cb = 0;
804 char *bootlist = get_boot_devices_list(&cb, true);
805
806 _FDT(chosen = fdt_add_subnode(fdt, 0, "chosen"));
807
808 _FDT(fdt_setprop_string(fdt, chosen, "bootargs", machine->kernel_cmdline));
809 _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-start",
810 spapr->initrd_base));
811 _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-end",
812 spapr->initrd_base + spapr->initrd_size));
813
814 if (spapr->kernel_size) {
815 uint64_t kprop[2] = { cpu_to_be64(KERNEL_LOAD_ADDR),
816 cpu_to_be64(spapr->kernel_size) };
817
818 _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel",
819 &kprop, sizeof(kprop)));
820 if (spapr->kernel_le) {
821 _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel-le", NULL, 0));
822 }
823 }
824 if (boot_menu) {
825 _FDT((fdt_setprop_cell(fdt, chosen, "qemu,boot-menu", boot_menu)));
826 }
827 _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-width", graphic_width));
828 _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-height", graphic_height));
829 _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-depth", graphic_depth));
830
831 if (cb && bootlist) {
832 int i;
833
834 for (i = 0; i < cb; i++) {
835 if (bootlist[i] == '\n') {
836 bootlist[i] = ' ';
837 }
838 }
839 _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-list", bootlist));
840 }
841
842 if (boot_device && strlen(boot_device)) {
843 _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-device", boot_device));
844 }
845
846 if (!spapr->has_graphics && stdout_path) {
847 _FDT(fdt_setprop_string(fdt, chosen, "linux,stdout-path", stdout_path));
848 }
849
850 g_free(stdout_path);
851 g_free(bootlist);
852 }
853
854 static void spapr_dt_hypervisor(sPAPRMachineState *spapr, void *fdt)
855 {
856 /* The /hypervisor node isn't in PAPR - this is a hack to allow PR
857 * KVM to work under pHyp with some guest co-operation */
858 int hypervisor;
859 uint8_t hypercall[16];
860
861 _FDT(hypervisor = fdt_add_subnode(fdt, 0, "hypervisor"));
862 /* indicate KVM hypercall interface */
863 _FDT(fdt_setprop_string(fdt, hypervisor, "compatible", "linux,kvm"));
864 if (kvmppc_has_cap_fixup_hcalls()) {
865 /*
866 * Older KVM versions with older guest kernels were broken
867 * with the magic page, don't allow the guest to map it.
868 */
869 if (!kvmppc_get_hypercall(first_cpu->env_ptr, hypercall,
870 sizeof(hypercall))) {
871 _FDT(fdt_setprop(fdt, hypervisor, "hcall-instructions",
872 hypercall, sizeof(hypercall)));
873 }
874 }
875 }
876
877 static void *spapr_build_fdt(sPAPRMachineState *spapr,
878 hwaddr rtas_addr,
879 hwaddr rtas_size)
880 {
881 MachineState *machine = MACHINE(qdev_get_machine());
882 MachineClass *mc = MACHINE_GET_CLASS(machine);
883 sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
884 int ret;
885 void *fdt;
886 sPAPRPHBState *phb;
887 char *buf;
888
889 fdt = g_malloc0(FDT_MAX_SIZE);
890 _FDT((fdt_create_empty_tree(fdt, FDT_MAX_SIZE)));
891
892 /* Root node */
893 _FDT(fdt_setprop_string(fdt, 0, "device_type", "chrp"));
894 _FDT(fdt_setprop_string(fdt, 0, "model", "IBM pSeries (emulated by qemu)"));
895 _FDT(fdt_setprop_string(fdt, 0, "compatible", "qemu,pseries"));
896
897 /*
898 * Add info to guest to indentify which host is it being run on
899 * and what is the uuid of the guest
900 */
901 if (kvmppc_get_host_model(&buf)) {
902 _FDT(fdt_setprop_string(fdt, 0, "host-model", buf));
903 g_free(buf);
904 }
905 if (kvmppc_get_host_serial(&buf)) {
906 _FDT(fdt_setprop_string(fdt, 0, "host-serial", buf));
907 g_free(buf);
908 }
909
910 buf = qemu_uuid_unparse_strdup(&qemu_uuid);
911
912 _FDT(fdt_setprop_string(fdt, 0, "vm,uuid", buf));
913 if (qemu_uuid_set) {
914 _FDT(fdt_setprop_string(fdt, 0, "system-id", buf));
915 }
916 g_free(buf);
917
918 if (qemu_get_vm_name()) {
919 _FDT(fdt_setprop_string(fdt, 0, "ibm,partition-name",
920 qemu_get_vm_name()));
921 }
922
923 _FDT(fdt_setprop_cell(fdt, 0, "#address-cells", 2));
924 _FDT(fdt_setprop_cell(fdt, 0, "#size-cells", 2));
925
926 /* /interrupt controller */
927 spapr_dt_xics(spapr->xics, fdt, PHANDLE_XICP);
928
929 ret = spapr_populate_memory(spapr, fdt);
930 if (ret < 0) {
931 error_report("couldn't setup memory nodes in fdt");
932 exit(1);
933 }
934
935 /* /vdevice */
936 spapr_dt_vdevice(spapr->vio_bus, fdt);
937
938 if (object_resolve_path_type("", TYPE_SPAPR_RNG, NULL)) {
939 ret = spapr_rng_populate_dt(fdt);
940 if (ret < 0) {
941 error_report("could not set up rng device in the fdt");
942 exit(1);
943 }
944 }
945
946 QLIST_FOREACH(phb, &spapr->phbs, list) {
947 ret = spapr_populate_pci_dt(phb, PHANDLE_XICP, fdt);
948 if (ret < 0) {
949 error_report("couldn't setup PCI devices in fdt");
950 exit(1);
951 }
952 }
953
954 /* cpus */
955 spapr_populate_cpus_dt_node(fdt, spapr);
956
957 if (smc->dr_lmb_enabled) {
958 _FDT(spapr_drc_populate_dt(fdt, 0, NULL, SPAPR_DR_CONNECTOR_TYPE_LMB));
959 }
960
961 if (mc->query_hotpluggable_cpus) {
962 int offset = fdt_path_offset(fdt, "/cpus");
963 ret = spapr_drc_populate_dt(fdt, offset, NULL,
964 SPAPR_DR_CONNECTOR_TYPE_CPU);
965 if (ret < 0) {
966 error_report("Couldn't set up CPU DR device tree properties");
967 exit(1);
968 }
969 }
970
971 /* /event-sources */
972 spapr_dt_events(spapr, fdt);
973
974 /* /rtas */
975 spapr_dt_rtas(spapr, fdt);
976
977 /* /chosen */
978 spapr_dt_chosen(spapr, fdt);
979
980 /* /hypervisor */
981 if (kvm_enabled()) {
982 spapr_dt_hypervisor(spapr, fdt);
983 }
984
985 /* Build memory reserve map */
986 if (spapr->kernel_size) {
987 _FDT((fdt_add_mem_rsv(fdt, KERNEL_LOAD_ADDR, spapr->kernel_size)));
988 }
989 if (spapr->initrd_size) {
990 _FDT((fdt_add_mem_rsv(fdt, spapr->initrd_base, spapr->initrd_size)));
991 }
992
993 /* ibm,client-architecture-support updates */
994 ret = spapr_dt_cas_updates(spapr, fdt, spapr->ov5_cas);
995 if (ret < 0) {
996 error_report("couldn't setup CAS properties fdt");
997 exit(1);
998 }
999
1000 return fdt;
1001 }
1002
1003 static uint64_t translate_kernel_address(void *opaque, uint64_t addr)
1004 {
1005 return (addr & 0x0fffffff) + KERNEL_LOAD_ADDR;
1006 }
1007
1008 static void emulate_spapr_hypercall(PPCVirtualHypervisor *vhyp,
1009 PowerPCCPU *cpu)
1010 {
1011 CPUPPCState *env = &cpu->env;
1012
1013 if (msr_pr) {
1014 hcall_dprintf("Hypercall made with MSR[PR]=1\n");
1015 env->gpr[3] = H_PRIVILEGE;
1016 } else {
1017 env->gpr[3] = spapr_hypercall(cpu, env->gpr[3], &env->gpr[4]);
1018 }
1019 }
1020
1021 #define HPTE(_table, _i) (void *)(((uint64_t *)(_table)) + ((_i) * 2))
1022 #define HPTE_VALID(_hpte) (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_VALID)
1023 #define HPTE_DIRTY(_hpte) (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_HPTE_DIRTY)
1024 #define CLEAN_HPTE(_hpte) ((*(uint64_t *)(_hpte)) &= tswap64(~HPTE64_V_HPTE_DIRTY))
1025 #define DIRTY_HPTE(_hpte) ((*(uint64_t *)(_hpte)) |= tswap64(HPTE64_V_HPTE_DIRTY))
1026
1027 /*
1028 * Get the fd to access the kernel htab, re-opening it if necessary
1029 */
1030 static int get_htab_fd(sPAPRMachineState *spapr)
1031 {
1032 if (spapr->htab_fd >= 0) {
1033 return spapr->htab_fd;
1034 }
1035
1036 spapr->htab_fd = kvmppc_get_htab_fd(false);
1037 if (spapr->htab_fd < 0) {
1038 error_report("Unable to open fd for reading hash table from KVM: %s",
1039 strerror(errno));
1040 }
1041
1042 return spapr->htab_fd;
1043 }
1044
1045 static void close_htab_fd(sPAPRMachineState *spapr)
1046 {
1047 if (spapr->htab_fd >= 0) {
1048 close(spapr->htab_fd);
1049 }
1050 spapr->htab_fd = -1;
1051 }
1052
1053 static int spapr_hpt_shift_for_ramsize(uint64_t ramsize)
1054 {
1055 int shift;
1056
1057 /* We aim for a hash table of size 1/128 the size of RAM (rounded
1058 * up). The PAPR recommendation is actually 1/64 of RAM size, but
1059 * that's much more than is needed for Linux guests */
1060 shift = ctz64(pow2ceil(ramsize)) - 7;
1061 shift = MAX(shift, 18); /* Minimum architected size */
1062 shift = MIN(shift, 46); /* Maximum architected size */
1063 return shift;
1064 }
1065
1066 static void spapr_reallocate_hpt(sPAPRMachineState *spapr, int shift,
1067 Error **errp)
1068 {
1069 long rc;
1070
1071 /* Clean up any HPT info from a previous boot */
1072 g_free(spapr->htab);
1073 spapr->htab = NULL;
1074 spapr->htab_shift = 0;
1075 close_htab_fd(spapr);
1076
1077 rc = kvmppc_reset_htab(shift);
1078 if (rc < 0) {
1079 /* kernel-side HPT needed, but couldn't allocate one */
1080 error_setg_errno(errp, errno,
1081 "Failed to allocate KVM HPT of order %d (try smaller maxmem?)",
1082 shift);
1083 /* This is almost certainly fatal, but if the caller really
1084 * wants to carry on with shift == 0, it's welcome to try */
1085 } else if (rc > 0) {
1086 /* kernel-side HPT allocated */
1087 if (rc != shift) {
1088 error_setg(errp,
1089 "Requested order %d HPT, but kernel allocated order %ld (try smaller maxmem?)",
1090 shift, rc);
1091 }
1092
1093 spapr->htab_shift = shift;
1094 spapr->htab = NULL;
1095 } else {
1096 /* kernel-side HPT not needed, allocate in userspace instead */
1097 size_t size = 1ULL << shift;
1098 int i;
1099
1100 spapr->htab = qemu_memalign(size, size);
1101 if (!spapr->htab) {
1102 error_setg_errno(errp, errno,
1103 "Could not allocate HPT of order %d", shift);
1104 return;
1105 }
1106
1107 memset(spapr->htab, 0, size);
1108 spapr->htab_shift = shift;
1109
1110 for (i = 0; i < size / HASH_PTE_SIZE_64; i++) {
1111 DIRTY_HPTE(HPTE(spapr->htab, i));
1112 }
1113 }
1114 }
1115
1116 static void find_unknown_sysbus_device(SysBusDevice *sbdev, void *opaque)
1117 {
1118 bool matched = false;
1119
1120 if (object_dynamic_cast(OBJECT(sbdev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
1121 matched = true;
1122 }
1123
1124 if (!matched) {
1125 error_report("Device %s is not supported by this machine yet.",
1126 qdev_fw_name(DEVICE(sbdev)));
1127 exit(1);
1128 }
1129 }
1130
1131 static void ppc_spapr_reset(void)
1132 {
1133 MachineState *machine = MACHINE(qdev_get_machine());
1134 sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
1135 PowerPCCPU *first_ppc_cpu;
1136 uint32_t rtas_limit;
1137 hwaddr rtas_addr, fdt_addr;
1138 void *fdt;
1139 int rc;
1140
1141 /* Check for unknown sysbus devices */
1142 foreach_dynamic_sysbus_device(find_unknown_sysbus_device, NULL);
1143
1144 /* Allocate and/or reset the hash page table */
1145 spapr_reallocate_hpt(spapr,
1146 spapr_hpt_shift_for_ramsize(machine->maxram_size),
1147 &error_fatal);
1148
1149 /* Update the RMA size if necessary */
1150 if (spapr->vrma_adjust) {
1151 spapr->rma_size = kvmppc_rma_size(spapr_node0_size(),
1152 spapr->htab_shift);
1153 }
1154
1155 qemu_devices_reset();
1156
1157 /*
1158 * We place the device tree and RTAS just below either the top of the RMA,
1159 * or just below 2GB, whichever is lowere, so that it can be
1160 * processed with 32-bit real mode code if necessary
1161 */
1162 rtas_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR);
1163 rtas_addr = rtas_limit - RTAS_MAX_SIZE;
1164 fdt_addr = rtas_addr - FDT_MAX_SIZE;
1165
1166 /* if this reset wasn't generated by CAS, we should reset our
1167 * negotiated options and start from scratch */
1168 if (!spapr->cas_reboot) {
1169 spapr_ovec_cleanup(spapr->ov5_cas);
1170 spapr->ov5_cas = spapr_ovec_new();
1171 }
1172
1173 fdt = spapr_build_fdt(spapr, rtas_addr, spapr->rtas_size);
1174
1175 spapr_load_rtas(spapr, fdt, rtas_addr);
1176
1177 rc = fdt_pack(fdt);
1178
1179 /* Should only fail if we've built a corrupted tree */
1180 assert(rc == 0);
1181
1182 if (fdt_totalsize(fdt) > FDT_MAX_SIZE) {
1183 error_report("FDT too big ! 0x%x bytes (max is 0x%x)",
1184 fdt_totalsize(fdt), FDT_MAX_SIZE);
1185 exit(1);
1186 }
1187
1188 /* Load the fdt */
1189 qemu_fdt_dumpdtb(fdt, fdt_totalsize(fdt));
1190 cpu_physical_memory_write(fdt_addr, fdt, fdt_totalsize(fdt));
1191 g_free(fdt);
1192
1193 /* Set up the entry state */
1194 first_ppc_cpu = POWERPC_CPU(first_cpu);
1195 first_ppc_cpu->env.gpr[3] = fdt_addr;
1196 first_ppc_cpu->env.gpr[5] = 0;
1197 first_cpu->halted = 0;
1198 first_ppc_cpu->env.nip = SPAPR_ENTRY_POINT;
1199
1200 spapr->cas_reboot = false;
1201 }
1202
1203 static void spapr_create_nvram(sPAPRMachineState *spapr)
1204 {
1205 DeviceState *dev = qdev_create(&spapr->vio_bus->bus, "spapr-nvram");
1206 DriveInfo *dinfo = drive_get(IF_PFLASH, 0, 0);
1207
1208 if (dinfo) {
1209 qdev_prop_set_drive(dev, "drive", blk_by_legacy_dinfo(dinfo),
1210 &error_fatal);
1211 }
1212
1213 qdev_init_nofail(dev);
1214
1215 spapr->nvram = (struct sPAPRNVRAM *)dev;
1216 }
1217
1218 static void spapr_rtc_create(sPAPRMachineState *spapr)
1219 {
1220 DeviceState *dev = qdev_create(NULL, TYPE_SPAPR_RTC);
1221
1222 qdev_init_nofail(dev);
1223 spapr->rtc = dev;
1224
1225 object_property_add_alias(qdev_get_machine(), "rtc-time",
1226 OBJECT(spapr->rtc), "date", NULL);
1227 }
1228
1229 /* Returns whether we want to use VGA or not */
1230 static bool spapr_vga_init(PCIBus *pci_bus, Error **errp)
1231 {
1232 switch (vga_interface_type) {
1233 case VGA_NONE:
1234 return false;
1235 case VGA_DEVICE:
1236 return true;
1237 case VGA_STD:
1238 case VGA_VIRTIO:
1239 return pci_vga_init(pci_bus) != NULL;
1240 default:
1241 error_setg(errp,
1242 "Unsupported VGA mode, only -vga std or -vga virtio is supported");
1243 return false;
1244 }
1245 }
1246
1247 static int spapr_post_load(void *opaque, int version_id)
1248 {
1249 sPAPRMachineState *spapr = (sPAPRMachineState *)opaque;
1250 int err = 0;
1251
1252 /* In earlier versions, there was no separate qdev for the PAPR
1253 * RTC, so the RTC offset was stored directly in sPAPREnvironment.
1254 * So when migrating from those versions, poke the incoming offset
1255 * value into the RTC device */
1256 if (version_id < 3) {
1257 err = spapr_rtc_import_offset(spapr->rtc, spapr->rtc_offset);
1258 }
1259
1260 return err;
1261 }
1262
1263 static bool version_before_3(void *opaque, int version_id)
1264 {
1265 return version_id < 3;
1266 }
1267
1268 static bool spapr_ov5_cas_needed(void *opaque)
1269 {
1270 sPAPRMachineState *spapr = opaque;
1271 sPAPROptionVector *ov5_mask = spapr_ovec_new();
1272 sPAPROptionVector *ov5_legacy = spapr_ovec_new();
1273 sPAPROptionVector *ov5_removed = spapr_ovec_new();
1274 bool cas_needed;
1275
1276 /* Prior to the introduction of sPAPROptionVector, we had two option
1277 * vectors we dealt with: OV5_FORM1_AFFINITY, and OV5_DRCONF_MEMORY.
1278 * Both of these options encode machine topology into the device-tree
1279 * in such a way that the now-booted OS should still be able to interact
1280 * appropriately with QEMU regardless of what options were actually
1281 * negotiatied on the source side.
1282 *
1283 * As such, we can avoid migrating the CAS-negotiated options if these
1284 * are the only options available on the current machine/platform.
1285 * Since these are the only options available for pseries-2.7 and
1286 * earlier, this allows us to maintain old->new/new->old migration
1287 * compatibility.
1288 *
1289 * For QEMU 2.8+, there are additional CAS-negotiatable options available
1290 * via default pseries-2.8 machines and explicit command-line parameters.
1291 * Some of these options, like OV5_HP_EVT, *do* require QEMU to be aware
1292 * of the actual CAS-negotiated values to continue working properly. For
1293 * example, availability of memory unplug depends on knowing whether
1294 * OV5_HP_EVT was negotiated via CAS.
1295 *
1296 * Thus, for any cases where the set of available CAS-negotiatable
1297 * options extends beyond OV5_FORM1_AFFINITY and OV5_DRCONF_MEMORY, we
1298 * include the CAS-negotiated options in the migration stream.
1299 */
1300 spapr_ovec_set(ov5_mask, OV5_FORM1_AFFINITY);
1301 spapr_ovec_set(ov5_mask, OV5_DRCONF_MEMORY);
1302
1303 /* spapr_ovec_diff returns true if bits were removed. we avoid using
1304 * the mask itself since in the future it's possible "legacy" bits may be
1305 * removed via machine options, which could generate a false positive
1306 * that breaks migration.
1307 */
1308 spapr_ovec_intersect(ov5_legacy, spapr->ov5, ov5_mask);
1309 cas_needed = spapr_ovec_diff(ov5_removed, spapr->ov5, ov5_legacy);
1310
1311 spapr_ovec_cleanup(ov5_mask);
1312 spapr_ovec_cleanup(ov5_legacy);
1313 spapr_ovec_cleanup(ov5_removed);
1314
1315 return cas_needed;
1316 }
1317
1318 static const VMStateDescription vmstate_spapr_ov5_cas = {
1319 .name = "spapr_option_vector_ov5_cas",
1320 .version_id = 1,
1321 .minimum_version_id = 1,
1322 .needed = spapr_ov5_cas_needed,
1323 .fields = (VMStateField[]) {
1324 VMSTATE_STRUCT_POINTER_V(ov5_cas, sPAPRMachineState, 1,
1325 vmstate_spapr_ovec, sPAPROptionVector),
1326 VMSTATE_END_OF_LIST()
1327 },
1328 };
1329
1330 static const VMStateDescription vmstate_spapr = {
1331 .name = "spapr",
1332 .version_id = 3,
1333 .minimum_version_id = 1,
1334 .post_load = spapr_post_load,
1335 .fields = (VMStateField[]) {
1336 /* used to be @next_irq */
1337 VMSTATE_UNUSED_BUFFER(version_before_3, 0, 4),
1338
1339 /* RTC offset */
1340 VMSTATE_UINT64_TEST(rtc_offset, sPAPRMachineState, version_before_3),
1341
1342 VMSTATE_PPC_TIMEBASE_V(tb, sPAPRMachineState, 2),
1343 VMSTATE_END_OF_LIST()
1344 },
1345 .subsections = (const VMStateDescription*[]) {
1346 &vmstate_spapr_ov5_cas,
1347 NULL
1348 }
1349 };
1350
1351 static int htab_save_setup(QEMUFile *f, void *opaque)
1352 {
1353 sPAPRMachineState *spapr = opaque;
1354
1355 /* "Iteration" header */
1356 qemu_put_be32(f, spapr->htab_shift);
1357
1358 if (spapr->htab) {
1359 spapr->htab_save_index = 0;
1360 spapr->htab_first_pass = true;
1361 } else {
1362 assert(kvm_enabled());
1363 }
1364
1365
1366 return 0;
1367 }
1368
1369 static void htab_save_first_pass(QEMUFile *f, sPAPRMachineState *spapr,
1370 int64_t max_ns)
1371 {
1372 bool has_timeout = max_ns != -1;
1373 int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
1374 int index = spapr->htab_save_index;
1375 int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
1376
1377 assert(spapr->htab_first_pass);
1378
1379 do {
1380 int chunkstart;
1381
1382 /* Consume invalid HPTEs */
1383 while ((index < htabslots)
1384 && !HPTE_VALID(HPTE(spapr->htab, index))) {
1385 index++;
1386 CLEAN_HPTE(HPTE(spapr->htab, index));
1387 }
1388
1389 /* Consume valid HPTEs */
1390 chunkstart = index;
1391 while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
1392 && HPTE_VALID(HPTE(spapr->htab, index))) {
1393 index++;
1394 CLEAN_HPTE(HPTE(spapr->htab, index));
1395 }
1396
1397 if (index > chunkstart) {
1398 int n_valid = index - chunkstart;
1399
1400 qemu_put_be32(f, chunkstart);
1401 qemu_put_be16(f, n_valid);
1402 qemu_put_be16(f, 0);
1403 qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
1404 HASH_PTE_SIZE_64 * n_valid);
1405
1406 if (has_timeout &&
1407 (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
1408 break;
1409 }
1410 }
1411 } while ((index < htabslots) && !qemu_file_rate_limit(f));
1412
1413 if (index >= htabslots) {
1414 assert(index == htabslots);
1415 index = 0;
1416 spapr->htab_first_pass = false;
1417 }
1418 spapr->htab_save_index = index;
1419 }
1420
1421 static int htab_save_later_pass(QEMUFile *f, sPAPRMachineState *spapr,
1422 int64_t max_ns)
1423 {
1424 bool final = max_ns < 0;
1425 int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
1426 int examined = 0, sent = 0;
1427 int index = spapr->htab_save_index;
1428 int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
1429
1430 assert(!spapr->htab_first_pass);
1431
1432 do {
1433 int chunkstart, invalidstart;
1434
1435 /* Consume non-dirty HPTEs */
1436 while ((index < htabslots)
1437 && !HPTE_DIRTY(HPTE(spapr->htab, index))) {
1438 index++;
1439 examined++;
1440 }
1441
1442 chunkstart = index;
1443 /* Consume valid dirty HPTEs */
1444 while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
1445 && HPTE_DIRTY(HPTE(spapr->htab, index))
1446 && HPTE_VALID(HPTE(spapr->htab, index))) {
1447 CLEAN_HPTE(HPTE(spapr->htab, index));
1448 index++;
1449 examined++;
1450 }
1451
1452 invalidstart = index;
1453 /* Consume invalid dirty HPTEs */
1454 while ((index < htabslots) && (index - invalidstart < USHRT_MAX)
1455 && HPTE_DIRTY(HPTE(spapr->htab, index))
1456 && !HPTE_VALID(HPTE(spapr->htab, index))) {
1457 CLEAN_HPTE(HPTE(spapr->htab, index));
1458 index++;
1459 examined++;
1460 }
1461
1462 if (index > chunkstart) {
1463 int n_valid = invalidstart - chunkstart;
1464 int n_invalid = index - invalidstart;
1465
1466 qemu_put_be32(f, chunkstart);
1467 qemu_put_be16(f, n_valid);
1468 qemu_put_be16(f, n_invalid);
1469 qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
1470 HASH_PTE_SIZE_64 * n_valid);
1471 sent += index - chunkstart;
1472
1473 if (!final && (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
1474 break;
1475 }
1476 }
1477
1478 if (examined >= htabslots) {
1479 break;
1480 }
1481
1482 if (index >= htabslots) {
1483 assert(index == htabslots);
1484 index = 0;
1485 }
1486 } while ((examined < htabslots) && (!qemu_file_rate_limit(f) || final));
1487
1488 if (index >= htabslots) {
1489 assert(index == htabslots);
1490 index = 0;
1491 }
1492
1493 spapr->htab_save_index = index;
1494
1495 return (examined >= htabslots) && (sent == 0) ? 1 : 0;
1496 }
1497
1498 #define MAX_ITERATION_NS 5000000 /* 5 ms */
1499 #define MAX_KVM_BUF_SIZE 2048
1500
1501 static int htab_save_iterate(QEMUFile *f, void *opaque)
1502 {
1503 sPAPRMachineState *spapr = opaque;
1504 int fd;
1505 int rc = 0;
1506
1507 /* Iteration header */
1508 qemu_put_be32(f, 0);
1509
1510 if (!spapr->htab) {
1511 assert(kvm_enabled());
1512
1513 fd = get_htab_fd(spapr);
1514 if (fd < 0) {
1515 return fd;
1516 }
1517
1518 rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, MAX_ITERATION_NS);
1519 if (rc < 0) {
1520 return rc;
1521 }
1522 } else if (spapr->htab_first_pass) {
1523 htab_save_first_pass(f, spapr, MAX_ITERATION_NS);
1524 } else {
1525 rc = htab_save_later_pass(f, spapr, MAX_ITERATION_NS);
1526 }
1527
1528 /* End marker */
1529 qemu_put_be32(f, 0);
1530 qemu_put_be16(f, 0);
1531 qemu_put_be16(f, 0);
1532
1533 return rc;
1534 }
1535
1536 static int htab_save_complete(QEMUFile *f, void *opaque)
1537 {
1538 sPAPRMachineState *spapr = opaque;
1539 int fd;
1540
1541 /* Iteration header */
1542 qemu_put_be32(f, 0);
1543
1544 if (!spapr->htab) {
1545 int rc;
1546
1547 assert(kvm_enabled());
1548
1549 fd = get_htab_fd(spapr);
1550 if (fd < 0) {
1551 return fd;
1552 }
1553
1554 rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, -1);
1555 if (rc < 0) {
1556 return rc;
1557 }
1558 } else {
1559 if (spapr->htab_first_pass) {
1560 htab_save_first_pass(f, spapr, -1);
1561 }
1562 htab_save_later_pass(f, spapr, -1);
1563 }
1564
1565 /* End marker */
1566 qemu_put_be32(f, 0);
1567 qemu_put_be16(f, 0);
1568 qemu_put_be16(f, 0);
1569
1570 return 0;
1571 }
1572
1573 static int htab_load(QEMUFile *f, void *opaque, int version_id)
1574 {
1575 sPAPRMachineState *spapr = opaque;
1576 uint32_t section_hdr;
1577 int fd = -1;
1578
1579 if (version_id < 1 || version_id > 1) {
1580 error_report("htab_load() bad version");
1581 return -EINVAL;
1582 }
1583
1584 section_hdr = qemu_get_be32(f);
1585
1586 if (section_hdr) {
1587 Error *local_err = NULL;
1588
1589 /* First section gives the htab size */
1590 spapr_reallocate_hpt(spapr, section_hdr, &local_err);
1591 if (local_err) {
1592 error_report_err(local_err);
1593 return -EINVAL;
1594 }
1595 return 0;
1596 }
1597
1598 if (!spapr->htab) {
1599 assert(kvm_enabled());
1600
1601 fd = kvmppc_get_htab_fd(true);
1602 if (fd < 0) {
1603 error_report("Unable to open fd to restore KVM hash table: %s",
1604 strerror(errno));
1605 }
1606 }
1607
1608 while (true) {
1609 uint32_t index;
1610 uint16_t n_valid, n_invalid;
1611
1612 index = qemu_get_be32(f);
1613 n_valid = qemu_get_be16(f);
1614 n_invalid = qemu_get_be16(f);
1615
1616 if ((index == 0) && (n_valid == 0) && (n_invalid == 0)) {
1617 /* End of Stream */
1618 break;
1619 }
1620
1621 if ((index + n_valid + n_invalid) >
1622 (HTAB_SIZE(spapr) / HASH_PTE_SIZE_64)) {
1623 /* Bad index in stream */
1624 error_report(
1625 "htab_load() bad index %d (%hd+%hd entries) in htab stream (htab_shift=%d)",
1626 index, n_valid, n_invalid, spapr->htab_shift);
1627 return -EINVAL;
1628 }
1629
1630 if (spapr->htab) {
1631 if (n_valid) {
1632 qemu_get_buffer(f, HPTE(spapr->htab, index),
1633 HASH_PTE_SIZE_64 * n_valid);
1634 }
1635 if (n_invalid) {
1636 memset(HPTE(spapr->htab, index + n_valid), 0,
1637 HASH_PTE_SIZE_64 * n_invalid);
1638 }
1639 } else {
1640 int rc;
1641
1642 assert(fd >= 0);
1643
1644 rc = kvmppc_load_htab_chunk(f, fd, index, n_valid, n_invalid);
1645 if (rc < 0) {
1646 return rc;
1647 }
1648 }
1649 }
1650
1651 if (!spapr->htab) {
1652 assert(fd >= 0);
1653 close(fd);
1654 }
1655
1656 return 0;
1657 }
1658
1659 static void htab_cleanup(void *opaque)
1660 {
1661 sPAPRMachineState *spapr = opaque;
1662
1663 close_htab_fd(spapr);
1664 }
1665
1666 static SaveVMHandlers savevm_htab_handlers = {
1667 .save_live_setup = htab_save_setup,
1668 .save_live_iterate = htab_save_iterate,
1669 .save_live_complete_precopy = htab_save_complete,
1670 .cleanup = htab_cleanup,
1671 .load_state = htab_load,
1672 };
1673
1674 static void spapr_boot_set(void *opaque, const char *boot_device,
1675 Error **errp)
1676 {
1677 MachineState *machine = MACHINE(qdev_get_machine());
1678 machine->boot_order = g_strdup(boot_device);
1679 }
1680
1681 /*
1682 * Reset routine for LMB DR devices.
1683 *
1684 * Unlike PCI DR devices, LMB DR devices explicitly register this reset
1685 * routine. Reset for PCI DR devices will be handled by PHB reset routine
1686 * when it walks all its children devices. LMB devices reset occurs
1687 * as part of spapr_ppc_reset().
1688 */
1689 static void spapr_drc_reset(void *opaque)
1690 {
1691 sPAPRDRConnector *drc = opaque;
1692 DeviceState *d = DEVICE(drc);
1693
1694 if (d) {
1695 device_reset(d);
1696 }
1697 }
1698
1699 static void spapr_create_lmb_dr_connectors(sPAPRMachineState *spapr)
1700 {
1701 MachineState *machine = MACHINE(spapr);
1702 uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
1703 uint32_t nr_lmbs = (machine->maxram_size - machine->ram_size)/lmb_size;
1704 int i;
1705
1706 for (i = 0; i < nr_lmbs; i++) {
1707 sPAPRDRConnector *drc;
1708 uint64_t addr;
1709
1710 addr = i * lmb_size + spapr->hotplug_memory.base;
1711 drc = spapr_dr_connector_new(OBJECT(spapr), SPAPR_DR_CONNECTOR_TYPE_LMB,
1712 addr/lmb_size);
1713 qemu_register_reset(spapr_drc_reset, drc);
1714 }
1715 }
1716
1717 /*
1718 * If RAM size, maxmem size and individual node mem sizes aren't aligned
1719 * to SPAPR_MEMORY_BLOCK_SIZE(256MB), then refuse to start the guest
1720 * since we can't support such unaligned sizes with DRCONF_MEMORY.
1721 */
1722 static void spapr_validate_node_memory(MachineState *machine, Error **errp)
1723 {
1724 int i;
1725
1726 if (machine->ram_size % SPAPR_MEMORY_BLOCK_SIZE) {
1727 error_setg(errp, "Memory size 0x" RAM_ADDR_FMT
1728 " is not aligned to %llu MiB",
1729 machine->ram_size,
1730 SPAPR_MEMORY_BLOCK_SIZE / M_BYTE);
1731 return;
1732 }
1733
1734 if (machine->maxram_size % SPAPR_MEMORY_BLOCK_SIZE) {
1735 error_setg(errp, "Maximum memory size 0x" RAM_ADDR_FMT
1736 " is not aligned to %llu MiB",
1737 machine->ram_size,
1738 SPAPR_MEMORY_BLOCK_SIZE / M_BYTE);
1739 return;
1740 }
1741
1742 for (i = 0; i < nb_numa_nodes; i++) {
1743 if (numa_info[i].node_mem % SPAPR_MEMORY_BLOCK_SIZE) {
1744 error_setg(errp,
1745 "Node %d memory size 0x%" PRIx64
1746 " is not aligned to %llu MiB",
1747 i, numa_info[i].node_mem,
1748 SPAPR_MEMORY_BLOCK_SIZE / M_BYTE);
1749 return;
1750 }
1751 }
1752 }
1753
1754 /* find cpu slot in machine->possible_cpus by core_id */
1755 static CPUArchId *spapr_find_cpu_slot(MachineState *ms, uint32_t id, int *idx)
1756 {
1757 int index = id / smp_threads;
1758
1759 if (index >= ms->possible_cpus->len) {
1760 return NULL;
1761 }
1762 if (idx) {
1763 *idx = index;
1764 }
1765 return &ms->possible_cpus->cpus[index];
1766 }
1767
1768 static void spapr_init_cpus(sPAPRMachineState *spapr)
1769 {
1770 MachineState *machine = MACHINE(spapr);
1771 MachineClass *mc = MACHINE_GET_CLASS(machine);
1772 char *type = spapr_get_cpu_core_type(machine->cpu_model);
1773 int smt = kvmppc_smt_threads();
1774 const CPUArchIdList *possible_cpus;
1775 int boot_cores_nr = smp_cpus / smp_threads;
1776 int i;
1777
1778 if (!type) {
1779 error_report("Unable to find sPAPR CPU Core definition");
1780 exit(1);
1781 }
1782
1783 possible_cpus = mc->possible_cpu_arch_ids(machine);
1784 if (mc->query_hotpluggable_cpus) {
1785 if (smp_cpus % smp_threads) {
1786 error_report("smp_cpus (%u) must be multiple of threads (%u)",
1787 smp_cpus, smp_threads);
1788 exit(1);
1789 }
1790 if (max_cpus % smp_threads) {
1791 error_report("max_cpus (%u) must be multiple of threads (%u)",
1792 max_cpus, smp_threads);
1793 exit(1);
1794 }
1795 } else {
1796 if (max_cpus != smp_cpus) {
1797 error_report("This machine version does not support CPU hotplug");
1798 exit(1);
1799 }
1800 boot_cores_nr = possible_cpus->len;
1801 }
1802
1803 for (i = 0; i < possible_cpus->len; i++) {
1804 int core_id = i * smp_threads;
1805
1806 if (mc->query_hotpluggable_cpus) {
1807 sPAPRDRConnector *drc =
1808 spapr_dr_connector_new(OBJECT(spapr),
1809 SPAPR_DR_CONNECTOR_TYPE_CPU,
1810 (core_id / smp_threads) * smt);
1811
1812 qemu_register_reset(spapr_drc_reset, drc);
1813 }
1814
1815 if (i < boot_cores_nr) {
1816 Object *core = object_new(type);
1817 int nr_threads = smp_threads;
1818
1819 /* Handle the partially filled core for older machine types */
1820 if ((i + 1) * smp_threads >= smp_cpus) {
1821 nr_threads = smp_cpus - i * smp_threads;
1822 }
1823
1824 object_property_set_int(core, nr_threads, "nr-threads",
1825 &error_fatal);
1826 object_property_set_int(core, core_id, CPU_CORE_PROP_CORE_ID,
1827 &error_fatal);
1828 object_property_set_bool(core, true, "realized", &error_fatal);
1829 }
1830 }
1831 g_free(type);
1832 }
1833
1834 /* pSeries LPAR / sPAPR hardware init */
1835 static void ppc_spapr_init(MachineState *machine)
1836 {
1837 sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
1838 sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1839 const char *kernel_filename = machine->kernel_filename;
1840 const char *initrd_filename = machine->initrd_filename;
1841 PCIHostState *phb;
1842 int i;
1843 MemoryRegion *sysmem = get_system_memory();
1844 MemoryRegion *ram = g_new(MemoryRegion, 1);
1845 MemoryRegion *rma_region;
1846 void *rma = NULL;
1847 hwaddr rma_alloc_size;
1848 hwaddr node0_size = spapr_node0_size();
1849 long load_limit, fw_size;
1850 char *filename;
1851 int smt = kvmppc_smt_threads();
1852
1853 msi_nonbroken = true;
1854
1855 QLIST_INIT(&spapr->phbs);
1856
1857 /* Allocate RMA if necessary */
1858 rma_alloc_size = kvmppc_alloc_rma(&rma);
1859
1860 if (rma_alloc_size == -1) {
1861 error_report("Unable to create RMA");
1862 exit(1);
1863 }
1864
1865 if (rma_alloc_size && (rma_alloc_size < node0_size)) {
1866 spapr->rma_size = rma_alloc_size;
1867 } else {
1868 spapr->rma_size = node0_size;
1869
1870 /* With KVM, we don't actually know whether KVM supports an
1871 * unbounded RMA (PR KVM) or is limited by the hash table size
1872 * (HV KVM using VRMA), so we always assume the latter
1873 *
1874 * In that case, we also limit the initial allocations for RTAS
1875 * etc... to 256M since we have no way to know what the VRMA size
1876 * is going to be as it depends on the size of the hash table
1877 * isn't determined yet.
1878 */
1879 if (kvm_enabled()) {
1880 spapr->vrma_adjust = 1;
1881 spapr->rma_size = MIN(spapr->rma_size, 0x10000000);
1882 }
1883
1884 /* Actually we don't support unbounded RMA anymore since we
1885 * added proper emulation of HV mode. The max we can get is
1886 * 16G which also happens to be what we configure for PAPR
1887 * mode so make sure we don't do anything bigger than that
1888 */
1889 spapr->rma_size = MIN(spapr->rma_size, 0x400000000ull);
1890 }
1891
1892 if (spapr->rma_size > node0_size) {
1893 error_report("Numa node 0 has to span the RMA (%#08"HWADDR_PRIx")",
1894 spapr->rma_size);
1895 exit(1);
1896 }
1897
1898 /* Setup a load limit for the ramdisk leaving room for SLOF and FDT */
1899 load_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR) - FW_OVERHEAD;
1900
1901 /* Set up Interrupt Controller before we create the VCPUs */
1902 spapr->xics = xics_system_init(machine,
1903 DIV_ROUND_UP(max_cpus * smt, smp_threads),
1904 XICS_IRQS_SPAPR, &error_fatal);
1905
1906 /* Set up containers for ibm,client-set-architecture negotiated options */
1907 spapr->ov5 = spapr_ovec_new();
1908 spapr->ov5_cas = spapr_ovec_new();
1909
1910 if (smc->dr_lmb_enabled) {
1911 spapr_ovec_set(spapr->ov5, OV5_DRCONF_MEMORY);
1912 spapr_validate_node_memory(machine, &error_fatal);
1913 }
1914
1915 spapr_ovec_set(spapr->ov5, OV5_FORM1_AFFINITY);
1916
1917 /* advertise support for dedicated HP event source to guests */
1918 if (spapr->use_hotplug_event_source) {
1919 spapr_ovec_set(spapr->ov5, OV5_HP_EVT);
1920 }
1921
1922 /* init CPUs */
1923 if (machine->cpu_model == NULL) {
1924 machine->cpu_model = kvm_enabled() ? "host" : smc->tcg_default_cpu;
1925 }
1926
1927 ppc_cpu_parse_features(machine->cpu_model);
1928
1929 spapr_init_cpus(spapr);
1930
1931 if (kvm_enabled()) {
1932 /* Enable H_LOGICAL_CI_* so SLOF can talk to in-kernel devices */
1933 kvmppc_enable_logical_ci_hcalls();
1934 kvmppc_enable_set_mode_hcall();
1935
1936 /* H_CLEAR_MOD/_REF are mandatory in PAPR, but off by default */
1937 kvmppc_enable_clear_ref_mod_hcalls();
1938 }
1939
1940 /* allocate RAM */
1941 memory_region_allocate_system_memory(ram, NULL, "ppc_spapr.ram",
1942 machine->ram_size);
1943 memory_region_add_subregion(sysmem, 0, ram);
1944
1945 if (rma_alloc_size && rma) {
1946 rma_region = g_new(MemoryRegion, 1);
1947 memory_region_init_ram_ptr(rma_region, NULL, "ppc_spapr.rma",
1948 rma_alloc_size, rma);
1949 vmstate_register_ram_global(rma_region);
1950 memory_region_add_subregion(sysmem, 0, rma_region);
1951 }
1952
1953 /* initialize hotplug memory address space */
1954 if (machine->ram_size < machine->maxram_size) {
1955 ram_addr_t hotplug_mem_size = machine->maxram_size - machine->ram_size;
1956 /*
1957 * Limit the number of hotpluggable memory slots to half the number
1958 * slots that KVM supports, leaving the other half for PCI and other
1959 * devices. However ensure that number of slots doesn't drop below 32.
1960 */
1961 int max_memslots = kvm_enabled() ? kvm_get_max_memslots() / 2 :
1962 SPAPR_MAX_RAM_SLOTS;
1963
1964 if (max_memslots < SPAPR_MAX_RAM_SLOTS) {
1965 max_memslots = SPAPR_MAX_RAM_SLOTS;
1966 }
1967 if (machine->ram_slots > max_memslots) {
1968 error_report("Specified number of memory slots %"
1969 PRIu64" exceeds max supported %d",
1970 machine->ram_slots, max_memslots);
1971 exit(1);
1972 }
1973
1974 spapr->hotplug_memory.base = ROUND_UP(machine->ram_size,
1975 SPAPR_HOTPLUG_MEM_ALIGN);
1976 memory_region_init(&spapr->hotplug_memory.mr, OBJECT(spapr),
1977 "hotplug-memory", hotplug_mem_size);
1978 memory_region_add_subregion(sysmem, spapr->hotplug_memory.base,
1979 &spapr->hotplug_memory.mr);
1980 }
1981
1982 if (smc->dr_lmb_enabled) {
1983 spapr_create_lmb_dr_connectors(spapr);
1984 }
1985
1986 filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, "spapr-rtas.bin");
1987 if (!filename) {
1988 error_report("Could not find LPAR rtas '%s'", "spapr-rtas.bin");
1989 exit(1);
1990 }
1991 spapr->rtas_size = get_image_size(filename);
1992 if (spapr->rtas_size < 0) {
1993 error_report("Could not get size of LPAR rtas '%s'", filename);
1994 exit(1);
1995 }
1996 spapr->rtas_blob = g_malloc(spapr->rtas_size);
1997 if (load_image_size(filename, spapr->rtas_blob, spapr->rtas_size) < 0) {
1998 error_report("Could not load LPAR rtas '%s'", filename);
1999 exit(1);
2000 }
2001 if (spapr->rtas_size > RTAS_MAX_SIZE) {
2002 error_report("RTAS too big ! 0x%zx bytes (max is 0x%x)",
2003 (size_t)spapr->rtas_size, RTAS_MAX_SIZE);
2004 exit(1);
2005 }
2006 g_free(filename);
2007
2008 /* Set up RTAS event infrastructure */
2009 spapr_events_init(spapr);
2010
2011 /* Set up the RTC RTAS interfaces */
2012 spapr_rtc_create(spapr);
2013
2014 /* Set up VIO bus */
2015 spapr->vio_bus = spapr_vio_bus_init();
2016
2017 for (i = 0; i < MAX_SERIAL_PORTS; i++) {
2018 if (serial_hds[i]) {
2019 spapr_vty_create(spapr->vio_bus, serial_hds[i]);
2020 }
2021 }
2022
2023 /* We always have at least the nvram device on VIO */
2024 spapr_create_nvram(spapr);
2025
2026 /* Set up PCI */
2027 spapr_pci_rtas_init();
2028
2029 phb = spapr_create_phb(spapr, 0);
2030
2031 for (i = 0; i < nb_nics; i++) {
2032 NICInfo *nd = &nd_table[i];
2033
2034 if (!nd->model) {
2035 nd->model = g_strdup("ibmveth");
2036 }
2037
2038 if (strcmp(nd->model, "ibmveth") == 0) {
2039 spapr_vlan_create(spapr->vio_bus, nd);
2040 } else {
2041 pci_nic_init_nofail(&nd_table[i], phb->bus, nd->model, NULL);
2042 }
2043 }
2044
2045 for (i = 0; i <= drive_get_max_bus(IF_SCSI); i++) {
2046 spapr_vscsi_create(spapr->vio_bus);
2047 }
2048
2049 /* Graphics */
2050 if (spapr_vga_init(phb->bus, &error_fatal)) {
2051 spapr->has_graphics = true;
2052 machine->usb |= defaults_enabled() && !machine->usb_disabled;
2053 }
2054
2055 if (machine->usb) {
2056 if (smc->use_ohci_by_default) {
2057 pci_create_simple(phb->bus, -1, "pci-ohci");
2058 } else {
2059 pci_create_simple(phb->bus, -1, "nec-usb-xhci");
2060 }
2061
2062 if (spapr->has_graphics) {
2063 USBBus *usb_bus = usb_bus_find(-1);
2064
2065 usb_create_simple(usb_bus, "usb-kbd");
2066 usb_create_simple(usb_bus, "usb-mouse");
2067 }
2068 }
2069
2070 if (spapr->rma_size < (MIN_RMA_SLOF << 20)) {
2071 error_report(
2072 "pSeries SLOF firmware requires >= %ldM guest RMA (Real Mode Area memory)",
2073 MIN_RMA_SLOF);
2074 exit(1);
2075 }
2076
2077 if (kernel_filename) {
2078 uint64_t lowaddr = 0;
2079
2080 spapr->kernel_size = load_elf(kernel_filename, translate_kernel_address,
2081 NULL, NULL, &lowaddr, NULL, 1,
2082 PPC_ELF_MACHINE, 0, 0);
2083 if (spapr->kernel_size == ELF_LOAD_WRONG_ENDIAN) {
2084 spapr->kernel_size = load_elf(kernel_filename,
2085 translate_kernel_address, NULL, NULL,
2086 &lowaddr, NULL, 0, PPC_ELF_MACHINE,
2087 0, 0);
2088 spapr->kernel_le = spapr->kernel_size > 0;
2089 }
2090 if (spapr->kernel_size < 0) {
2091 error_report("error loading %s: %s", kernel_filename,
2092 load_elf_strerror(spapr->kernel_size));
2093 exit(1);
2094 }
2095
2096 /* load initrd */
2097 if (initrd_filename) {
2098 /* Try to locate the initrd in the gap between the kernel
2099 * and the firmware. Add a bit of space just in case
2100 */
2101 spapr->initrd_base = (KERNEL_LOAD_ADDR + spapr->kernel_size
2102 + 0x1ffff) & ~0xffff;
2103 spapr->initrd_size = load_image_targphys(initrd_filename,
2104 spapr->initrd_base,
2105 load_limit
2106 - spapr->initrd_base);
2107 if (spapr->initrd_size < 0) {
2108 error_report("could not load initial ram disk '%s'",
2109 initrd_filename);
2110 exit(1);
2111 }
2112 }
2113 }
2114
2115 if (bios_name == NULL) {
2116 bios_name = FW_FILE_NAME;
2117 }
2118 filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
2119 if (!filename) {
2120 error_report("Could not find LPAR firmware '%s'", bios_name);
2121 exit(1);
2122 }
2123 fw_size = load_image_targphys(filename, 0, FW_MAX_SIZE);
2124 if (fw_size <= 0) {
2125 error_report("Could not load LPAR firmware '%s'", filename);
2126 exit(1);
2127 }
2128 g_free(filename);
2129
2130 /* FIXME: Should register things through the MachineState's qdev
2131 * interface, this is a legacy from the sPAPREnvironment structure
2132 * which predated MachineState but had a similar function */
2133 vmstate_register(NULL, 0, &vmstate_spapr, spapr);
2134 register_savevm_live(NULL, "spapr/htab", -1, 1,
2135 &savevm_htab_handlers, spapr);
2136
2137 /* used by RTAS */
2138 QTAILQ_INIT(&spapr->ccs_list);
2139 qemu_register_reset(spapr_ccs_reset_hook, spapr);
2140
2141 qemu_register_boot_set(spapr_boot_set, spapr);
2142
2143 /* to stop and start vmclock */
2144 if (kvm_enabled()) {
2145 qemu_add_vm_change_state_handler(cpu_ppc_clock_vm_state_change,
2146 &spapr->tb);
2147 }
2148 }
2149
2150 static int spapr_kvm_type(const char *vm_type)
2151 {
2152 if (!vm_type) {
2153 return 0;
2154 }
2155
2156 if (!strcmp(vm_type, "HV")) {
2157 return 1;
2158 }
2159
2160 if (!strcmp(vm_type, "PR")) {
2161 return 2;
2162 }
2163
2164 error_report("Unknown kvm-type specified '%s'", vm_type);
2165 exit(1);
2166 }
2167
2168 /*
2169 * Implementation of an interface to adjust firmware path
2170 * for the bootindex property handling.
2171 */
2172 static char *spapr_get_fw_dev_path(FWPathProvider *p, BusState *bus,
2173 DeviceState *dev)
2174 {
2175 #define CAST(type, obj, name) \
2176 ((type *)object_dynamic_cast(OBJECT(obj), (name)))
2177 SCSIDevice *d = CAST(SCSIDevice, dev, TYPE_SCSI_DEVICE);
2178 sPAPRPHBState *phb = CAST(sPAPRPHBState, dev, TYPE_SPAPR_PCI_HOST_BRIDGE);
2179
2180 if (d) {
2181 void *spapr = CAST(void, bus->parent, "spapr-vscsi");
2182 VirtIOSCSI *virtio = CAST(VirtIOSCSI, bus->parent, TYPE_VIRTIO_SCSI);
2183 USBDevice *usb = CAST(USBDevice, bus->parent, TYPE_USB_DEVICE);
2184
2185 if (spapr) {
2186 /*
2187 * Replace "channel@0/disk@0,0" with "disk@8000000000000000":
2188 * We use SRP luns of the form 8000 | (bus << 8) | (id << 5) | lun
2189 * in the top 16 bits of the 64-bit LUN
2190 */
2191 unsigned id = 0x8000 | (d->id << 8) | d->lun;
2192 return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
2193 (uint64_t)id << 48);
2194 } else if (virtio) {
2195 /*
2196 * We use SRP luns of the form 01000000 | (target << 8) | lun
2197 * in the top 32 bits of the 64-bit LUN
2198 * Note: the quote above is from SLOF and it is wrong,
2199 * the actual binding is:
2200 * swap 0100 or 10 << or 20 << ( target lun-id -- srplun )
2201 */
2202 unsigned id = 0x1000000 | (d->id << 16) | d->lun;
2203 return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
2204 (uint64_t)id << 32);
2205 } else if (usb) {
2206 /*
2207 * We use SRP luns of the form 01000000 | (usb-port << 16) | lun
2208 * in the top 32 bits of the 64-bit LUN
2209 */
2210 unsigned usb_port = atoi(usb->port->path);
2211 unsigned id = 0x1000000 | (usb_port << 16) | d->lun;
2212 return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
2213 (uint64_t)id << 32);
2214 }
2215 }
2216
2217 /*
2218 * SLOF probes the USB devices, and if it recognizes that the device is a
2219 * storage device, it changes its name to "storage" instead of "usb-host",
2220 * and additionally adds a child node for the SCSI LUN, so the correct
2221 * boot path in SLOF is something like .../storage@1/disk@xxx" instead.
2222 */
2223 if (strcmp("usb-host", qdev_fw_name(dev)) == 0) {
2224 USBDevice *usbdev = CAST(USBDevice, dev, TYPE_USB_DEVICE);
2225 if (usb_host_dev_is_scsi_storage(usbdev)) {
2226 return g_strdup_printf("storage@%s/disk", usbdev->port->path);
2227 }
2228 }
2229
2230 if (phb) {
2231 /* Replace "pci" with "pci@800000020000000" */
2232 return g_strdup_printf("pci@%"PRIX64, phb->buid);
2233 }
2234
2235 return NULL;
2236 }
2237
2238 static char *spapr_get_kvm_type(Object *obj, Error **errp)
2239 {
2240 sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2241
2242 return g_strdup(spapr->kvm_type);
2243 }
2244
2245 static void spapr_set_kvm_type(Object *obj, const char *value, Error **errp)
2246 {
2247 sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2248
2249 g_free(spapr->kvm_type);
2250 spapr->kvm_type = g_strdup(value);
2251 }
2252
2253 static bool spapr_get_modern_hotplug_events(Object *obj, Error **errp)
2254 {
2255 sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2256
2257 return spapr->use_hotplug_event_source;
2258 }
2259
2260 static void spapr_set_modern_hotplug_events(Object *obj, bool value,
2261 Error **errp)
2262 {
2263 sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2264
2265 spapr->use_hotplug_event_source = value;
2266 }
2267
2268 static void spapr_machine_initfn(Object *obj)
2269 {
2270 sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2271
2272 spapr->htab_fd = -1;
2273 spapr->use_hotplug_event_source = true;
2274 object_property_add_str(obj, "kvm-type",
2275 spapr_get_kvm_type, spapr_set_kvm_type, NULL);
2276 object_property_set_description(obj, "kvm-type",
2277 "Specifies the KVM virtualization mode (HV, PR)",
2278 NULL);
2279 object_property_add_bool(obj, "modern-hotplug-events",
2280 spapr_get_modern_hotplug_events,
2281 spapr_set_modern_hotplug_events,
2282 NULL);
2283 object_property_set_description(obj, "modern-hotplug-events",
2284 "Use dedicated hotplug event mechanism in"
2285 " place of standard EPOW events when possible"
2286 " (required for memory hot-unplug support)",
2287 NULL);
2288 }
2289
2290 static void spapr_machine_finalizefn(Object *obj)
2291 {
2292 sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2293
2294 g_free(spapr->kvm_type);
2295 }
2296
2297 void spapr_do_system_reset_on_cpu(CPUState *cs, run_on_cpu_data arg)
2298 {
2299 cpu_synchronize_state(cs);
2300 ppc_cpu_do_system_reset(cs);
2301 }
2302
2303 static void spapr_nmi(NMIState *n, int cpu_index, Error **errp)
2304 {
2305 CPUState *cs;
2306
2307 CPU_FOREACH(cs) {
2308 async_run_on_cpu(cs, spapr_do_system_reset_on_cpu, RUN_ON_CPU_NULL);
2309 }
2310 }
2311
2312 static void spapr_add_lmbs(DeviceState *dev, uint64_t addr_start, uint64_t size,
2313 uint32_t node, bool dedicated_hp_event_source,
2314 Error **errp)
2315 {
2316 sPAPRDRConnector *drc;
2317 sPAPRDRConnectorClass *drck;
2318 uint32_t nr_lmbs = size/SPAPR_MEMORY_BLOCK_SIZE;
2319 int i, fdt_offset, fdt_size;
2320 void *fdt;
2321 uint64_t addr = addr_start;
2322
2323 for (i = 0; i < nr_lmbs; i++) {
2324 drc = spapr_dr_connector_by_id(SPAPR_DR_CONNECTOR_TYPE_LMB,
2325 addr/SPAPR_MEMORY_BLOCK_SIZE);
2326 g_assert(drc);
2327
2328 fdt = create_device_tree(&fdt_size);
2329 fdt_offset = spapr_populate_memory_node(fdt, node, addr,
2330 SPAPR_MEMORY_BLOCK_SIZE);
2331
2332 drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
2333 drck->attach(drc, dev, fdt, fdt_offset, !dev->hotplugged, errp);
2334 addr += SPAPR_MEMORY_BLOCK_SIZE;
2335 if (!dev->hotplugged) {
2336 /* guests expect coldplugged LMBs to be pre-allocated */
2337 drck->set_allocation_state(drc, SPAPR_DR_ALLOCATION_STATE_USABLE);
2338 drck->set_isolation_state(drc, SPAPR_DR_ISOLATION_STATE_UNISOLATED);
2339 }
2340 }
2341 /* send hotplug notification to the
2342 * guest only in case of hotplugged memory
2343 */
2344 if (dev->hotplugged) {
2345 if (dedicated_hp_event_source) {
2346 drc = spapr_dr_connector_by_id(SPAPR_DR_CONNECTOR_TYPE_LMB,
2347 addr_start / SPAPR_MEMORY_BLOCK_SIZE);
2348 drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
2349 spapr_hotplug_req_add_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
2350 nr_lmbs,
2351 drck->get_index(drc));
2352 } else {
2353 spapr_hotplug_req_add_by_count(SPAPR_DR_CONNECTOR_TYPE_LMB,
2354 nr_lmbs);
2355 }
2356 }
2357 }
2358
2359 static void spapr_memory_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
2360 uint32_t node, Error **errp)
2361 {
2362 Error *local_err = NULL;
2363 sPAPRMachineState *ms = SPAPR_MACHINE(hotplug_dev);
2364 PCDIMMDevice *dimm = PC_DIMM(dev);
2365 PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
2366 MemoryRegion *mr = ddc->get_memory_region(dimm);
2367 uint64_t align = memory_region_get_alignment(mr);
2368 uint64_t size = memory_region_size(mr);
2369 uint64_t addr;
2370
2371 if (size % SPAPR_MEMORY_BLOCK_SIZE) {
2372 error_setg(&local_err, "Hotplugged memory size must be a multiple of "
2373 "%lld MB", SPAPR_MEMORY_BLOCK_SIZE/M_BYTE);
2374 goto out;
2375 }
2376
2377 pc_dimm_memory_plug(dev, &ms->hotplug_memory, mr, align, &local_err);
2378 if (local_err) {
2379 goto out;
2380 }
2381
2382 addr = object_property_get_int(OBJECT(dimm), PC_DIMM_ADDR_PROP, &local_err);
2383 if (local_err) {
2384 pc_dimm_memory_unplug(dev, &ms->hotplug_memory, mr);
2385 goto out;
2386 }
2387
2388 spapr_add_lmbs(dev, addr, size, node,
2389 spapr_ovec_test(ms->ov5_cas, OV5_HP_EVT),
2390 &error_abort);
2391
2392 out:
2393 error_propagate(errp, local_err);
2394 }
2395
2396 typedef struct sPAPRDIMMState {
2397 uint32_t nr_lmbs;
2398 } sPAPRDIMMState;
2399
2400 static void spapr_lmb_release(DeviceState *dev, void *opaque)
2401 {
2402 sPAPRDIMMState *ds = (sPAPRDIMMState *)opaque;
2403 HotplugHandler *hotplug_ctrl;
2404
2405 if (--ds->nr_lmbs) {
2406 return;
2407 }
2408
2409 g_free(ds);
2410
2411 /*
2412 * Now that all the LMBs have been removed by the guest, call the
2413 * pc-dimm unplug handler to cleanup up the pc-dimm device.
2414 */
2415 hotplug_ctrl = qdev_get_hotplug_handler(dev);
2416 hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
2417 }
2418
2419 static void spapr_del_lmbs(DeviceState *dev, uint64_t addr_start, uint64_t size,
2420 Error **errp)
2421 {
2422 sPAPRDRConnector *drc;
2423 sPAPRDRConnectorClass *drck;
2424 uint32_t nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
2425 int i;
2426 sPAPRDIMMState *ds = g_malloc0(sizeof(sPAPRDIMMState));
2427 uint64_t addr = addr_start;
2428
2429 ds->nr_lmbs = nr_lmbs;
2430 for (i = 0; i < nr_lmbs; i++) {
2431 drc = spapr_dr_connector_by_id(SPAPR_DR_CONNECTOR_TYPE_LMB,
2432 addr / SPAPR_MEMORY_BLOCK_SIZE);
2433 g_assert(drc);
2434
2435 drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
2436 drck->detach(drc, dev, spapr_lmb_release, ds, errp);
2437 addr += SPAPR_MEMORY_BLOCK_SIZE;
2438 }
2439
2440 drc = spapr_dr_connector_by_id(SPAPR_DR_CONNECTOR_TYPE_LMB,
2441 addr_start / SPAPR_MEMORY_BLOCK_SIZE);
2442 drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
2443 spapr_hotplug_req_remove_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
2444 nr_lmbs,
2445 drck->get_index(drc));
2446 }
2447
2448 static void spapr_memory_unplug(HotplugHandler *hotplug_dev, DeviceState *dev,
2449 Error **errp)
2450 {
2451 sPAPRMachineState *ms = SPAPR_MACHINE(hotplug_dev);
2452 PCDIMMDevice *dimm = PC_DIMM(dev);
2453 PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
2454 MemoryRegion *mr = ddc->get_memory_region(dimm);
2455
2456 pc_dimm_memory_unplug(dev, &ms->hotplug_memory, mr);
2457 object_unparent(OBJECT(dev));
2458 }
2459
2460 static void spapr_memory_unplug_request(HotplugHandler *hotplug_dev,
2461 DeviceState *dev, Error **errp)
2462 {
2463 Error *local_err = NULL;
2464 PCDIMMDevice *dimm = PC_DIMM(dev);
2465 PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
2466 MemoryRegion *mr = ddc->get_memory_region(dimm);
2467 uint64_t size = memory_region_size(mr);
2468 uint64_t addr;
2469
2470 addr = object_property_get_int(OBJECT(dimm), PC_DIMM_ADDR_PROP, &local_err);
2471 if (local_err) {
2472 goto out;
2473 }
2474
2475 spapr_del_lmbs(dev, addr, size, &error_abort);
2476 out:
2477 error_propagate(errp, local_err);
2478 }
2479
2480 void *spapr_populate_hotplug_cpu_dt(CPUState *cs, int *fdt_offset,
2481 sPAPRMachineState *spapr)
2482 {
2483 PowerPCCPU *cpu = POWERPC_CPU(cs);
2484 DeviceClass *dc = DEVICE_GET_CLASS(cs);
2485 int id = ppc_get_vcpu_dt_id(cpu);
2486 void *fdt;
2487 int offset, fdt_size;
2488 char *nodename;
2489
2490 fdt = create_device_tree(&fdt_size);
2491 nodename = g_strdup_printf("%s@%x", dc->fw_name, id);
2492 offset = fdt_add_subnode(fdt, 0, nodename);
2493
2494 spapr_populate_cpu_dt(cs, fdt, offset, spapr);
2495 g_free(nodename);
2496
2497 *fdt_offset = offset;
2498 return fdt;
2499 }
2500
2501 static void spapr_core_unplug(HotplugHandler *hotplug_dev, DeviceState *dev,
2502 Error **errp)
2503 {
2504 MachineState *ms = MACHINE(qdev_get_machine());
2505 CPUCore *cc = CPU_CORE(dev);
2506 CPUArchId *core_slot = spapr_find_cpu_slot(ms, cc->core_id, NULL);
2507
2508 core_slot->cpu = NULL;
2509 object_unparent(OBJECT(dev));
2510 }
2511
2512 static void spapr_core_release(DeviceState *dev, void *opaque)
2513 {
2514 HotplugHandler *hotplug_ctrl;
2515
2516 hotplug_ctrl = qdev_get_hotplug_handler(dev);
2517 hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
2518 }
2519
2520 static
2521 void spapr_core_unplug_request(HotplugHandler *hotplug_dev, DeviceState *dev,
2522 Error **errp)
2523 {
2524 int index;
2525 sPAPRDRConnector *drc;
2526 sPAPRDRConnectorClass *drck;
2527 Error *local_err = NULL;
2528 CPUCore *cc = CPU_CORE(dev);
2529 int smt = kvmppc_smt_threads();
2530
2531 if (!spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index)) {
2532 error_setg(errp, "Unable to find CPU core with core-id: %d",
2533 cc->core_id);
2534 return;
2535 }
2536 if (index == 0) {
2537 error_setg(errp, "Boot CPU core may not be unplugged");
2538 return;
2539 }
2540
2541 drc = spapr_dr_connector_by_id(SPAPR_DR_CONNECTOR_TYPE_CPU, index * smt);
2542 g_assert(drc);
2543
2544 drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
2545 drck->detach(drc, dev, spapr_core_release, NULL, &local_err);
2546 if (local_err) {
2547 error_propagate(errp, local_err);
2548 return;
2549 }
2550
2551 spapr_hotplug_req_remove_by_index(drc);
2552 }
2553
2554 static void spapr_core_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
2555 Error **errp)
2556 {
2557 sPAPRMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
2558 MachineClass *mc = MACHINE_GET_CLASS(spapr);
2559 sPAPRCPUCore *core = SPAPR_CPU_CORE(OBJECT(dev));
2560 CPUCore *cc = CPU_CORE(dev);
2561 CPUState *cs = CPU(core->threads);
2562 sPAPRDRConnector *drc;
2563 Error *local_err = NULL;
2564 void *fdt = NULL;
2565 int fdt_offset = 0;
2566 int smt = kvmppc_smt_threads();
2567 CPUArchId *core_slot;
2568 int index;
2569
2570 core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
2571 if (!core_slot) {
2572 error_setg(errp, "Unable to find CPU core with core-id: %d",
2573 cc->core_id);
2574 return;
2575 }
2576 drc = spapr_dr_connector_by_id(SPAPR_DR_CONNECTOR_TYPE_CPU, index * smt);
2577
2578 g_assert(drc || !mc->query_hotpluggable_cpus);
2579
2580 /*
2581 * Setup CPU DT entries only for hotplugged CPUs. For boot time or
2582 * coldplugged CPUs DT entries are setup in spapr_build_fdt().
2583 */
2584 if (dev->hotplugged) {
2585 fdt = spapr_populate_hotplug_cpu_dt(cs, &fdt_offset, spapr);
2586 }
2587
2588 if (drc) {
2589 sPAPRDRConnectorClass *drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
2590 drck->attach(drc, dev, fdt, fdt_offset, !dev->hotplugged, &local_err);
2591 if (local_err) {
2592 g_free(fdt);
2593 error_propagate(errp, local_err);
2594 return;
2595 }
2596 }
2597
2598 if (dev->hotplugged) {
2599 /*
2600 * Send hotplug notification interrupt to the guest only in case
2601 * of hotplugged CPUs.
2602 */
2603 spapr_hotplug_req_add_by_index(drc);
2604 } else {
2605 /*
2606 * Set the right DRC states for cold plugged CPU.
2607 */
2608 if (drc) {
2609 sPAPRDRConnectorClass *drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
2610 drck->set_allocation_state(drc, SPAPR_DR_ALLOCATION_STATE_USABLE);
2611 drck->set_isolation_state(drc, SPAPR_DR_ISOLATION_STATE_UNISOLATED);
2612 }
2613 }
2614 core_slot->cpu = OBJECT(dev);
2615 }
2616
2617 static void spapr_core_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
2618 Error **errp)
2619 {
2620 MachineState *machine = MACHINE(OBJECT(hotplug_dev));
2621 MachineClass *mc = MACHINE_GET_CLASS(hotplug_dev);
2622 Error *local_err = NULL;
2623 CPUCore *cc = CPU_CORE(dev);
2624 char *base_core_type = spapr_get_cpu_core_type(machine->cpu_model);
2625 const char *type = object_get_typename(OBJECT(dev));
2626 CPUArchId *core_slot;
2627 int index;
2628
2629 if (dev->hotplugged && !mc->query_hotpluggable_cpus) {
2630 error_setg(&local_err, "CPU hotplug not supported for this machine");
2631 goto out;
2632 }
2633
2634 if (strcmp(base_core_type, type)) {
2635 error_setg(&local_err, "CPU core type should be %s", base_core_type);
2636 goto out;
2637 }
2638
2639 if (cc->core_id % smp_threads) {
2640 error_setg(&local_err, "invalid core id %d", cc->core_id);
2641 goto out;
2642 }
2643
2644 core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
2645 if (!core_slot) {
2646 error_setg(&local_err, "core id %d out of range", cc->core_id);
2647 goto out;
2648 }
2649
2650 if (core_slot->cpu) {
2651 error_setg(&local_err, "core %d already populated", cc->core_id);
2652 goto out;
2653 }
2654
2655 out:
2656 g_free(base_core_type);
2657 error_propagate(errp, local_err);
2658 }
2659
2660 static void spapr_machine_device_plug(HotplugHandler *hotplug_dev,
2661 DeviceState *dev, Error **errp)
2662 {
2663 sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(qdev_get_machine());
2664
2665 if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
2666 int node;
2667
2668 if (!smc->dr_lmb_enabled) {
2669 error_setg(errp, "Memory hotplug not supported for this machine");
2670 return;
2671 }
2672 node = object_property_get_int(OBJECT(dev), PC_DIMM_NODE_PROP, errp);
2673 if (*errp) {
2674 return;
2675 }
2676 if (node < 0 || node >= MAX_NODES) {
2677 error_setg(errp, "Invaild node %d", node);
2678 return;
2679 }
2680
2681 /*
2682 * Currently PowerPC kernel doesn't allow hot-adding memory to
2683 * memory-less node, but instead will silently add the memory
2684 * to the first node that has some memory. This causes two
2685 * unexpected behaviours for the user.
2686 *
2687 * - Memory gets hotplugged to a different node than what the user
2688 * specified.
2689 * - Since pc-dimm subsystem in QEMU still thinks that memory belongs
2690 * to memory-less node, a reboot will set things accordingly
2691 * and the previously hotplugged memory now ends in the right node.
2692 * This appears as if some memory moved from one node to another.
2693 *
2694 * So until kernel starts supporting memory hotplug to memory-less
2695 * nodes, just prevent such attempts upfront in QEMU.
2696 */
2697 if (nb_numa_nodes && !numa_info[node].node_mem) {
2698 error_setg(errp, "Can't hotplug memory to memory-less node %d",
2699 node);
2700 return;
2701 }
2702
2703 spapr_memory_plug(hotplug_dev, dev, node, errp);
2704 } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
2705 spapr_core_plug(hotplug_dev, dev, errp);
2706 }
2707 }
2708
2709 static void spapr_machine_device_unplug(HotplugHandler *hotplug_dev,
2710 DeviceState *dev, Error **errp)
2711 {
2712 sPAPRMachineState *sms = SPAPR_MACHINE(qdev_get_machine());
2713 MachineClass *mc = MACHINE_GET_CLASS(qdev_get_machine());
2714
2715 if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
2716 if (spapr_ovec_test(sms->ov5_cas, OV5_HP_EVT)) {
2717 spapr_memory_unplug(hotplug_dev, dev, errp);
2718 } else {
2719 error_setg(errp, "Memory hot unplug not supported for this guest");
2720 }
2721 } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
2722 if (!mc->query_hotpluggable_cpus) {
2723 error_setg(errp, "CPU hot unplug not supported on this machine");
2724 return;
2725 }
2726 spapr_core_unplug(hotplug_dev, dev, errp);
2727 }
2728 }
2729
2730 static void spapr_machine_device_unplug_request(HotplugHandler *hotplug_dev,
2731 DeviceState *dev, Error **errp)
2732 {
2733 sPAPRMachineState *sms = SPAPR_MACHINE(qdev_get_machine());
2734 MachineClass *mc = MACHINE_GET_CLASS(qdev_get_machine());
2735
2736 if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
2737 if (spapr_ovec_test(sms->ov5_cas, OV5_HP_EVT)) {
2738 spapr_memory_unplug_request(hotplug_dev, dev, errp);
2739 } else {
2740 /* NOTE: this means there is a window after guest reset, prior to
2741 * CAS negotiation, where unplug requests will fail due to the
2742 * capability not being detected yet. This is a bit different than
2743 * the case with PCI unplug, where the events will be queued and
2744 * eventually handled by the guest after boot
2745 */
2746 error_setg(errp, "Memory hot unplug not supported for this guest");
2747 }
2748 } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
2749 if (!mc->query_hotpluggable_cpus) {
2750 error_setg(errp, "CPU hot unplug not supported on this machine");
2751 return;
2752 }
2753 spapr_core_unplug_request(hotplug_dev, dev, errp);
2754 }
2755 }
2756
2757 static void spapr_machine_device_pre_plug(HotplugHandler *hotplug_dev,
2758 DeviceState *dev, Error **errp)
2759 {
2760 if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
2761 spapr_core_pre_plug(hotplug_dev, dev, errp);
2762 }
2763 }
2764
2765 static HotplugHandler *spapr_get_hotplug_handler(MachineState *machine,
2766 DeviceState *dev)
2767 {
2768 if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM) ||
2769 object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
2770 return HOTPLUG_HANDLER(machine);
2771 }
2772 return NULL;
2773 }
2774
2775 static unsigned spapr_cpu_index_to_socket_id(unsigned cpu_index)
2776 {
2777 /* Allocate to NUMA nodes on a "socket" basis (not that concept of
2778 * socket means much for the paravirtualized PAPR platform) */
2779 return cpu_index / smp_threads / smp_cores;
2780 }
2781
2782 static const CPUArchIdList *spapr_possible_cpu_arch_ids(MachineState *machine)
2783 {
2784 int i;
2785 int spapr_max_cores = max_cpus / smp_threads;
2786 MachineClass *mc = MACHINE_GET_CLASS(machine);
2787
2788 if (!mc->query_hotpluggable_cpus) {
2789 spapr_max_cores = QEMU_ALIGN_UP(smp_cpus, smp_threads) / smp_threads;
2790 }
2791 if (machine->possible_cpus) {
2792 assert(machine->possible_cpus->len == spapr_max_cores);
2793 return machine->possible_cpus;
2794 }
2795
2796 machine->possible_cpus = g_malloc0(sizeof(CPUArchIdList) +
2797 sizeof(CPUArchId) * spapr_max_cores);
2798 machine->possible_cpus->len = spapr_max_cores;
2799 for (i = 0; i < machine->possible_cpus->len; i++) {
2800 int core_id = i * smp_threads;
2801
2802 machine->possible_cpus->cpus[i].arch_id = core_id;
2803 machine->possible_cpus->cpus[i].props.has_core_id = true;
2804 machine->possible_cpus->cpus[i].props.core_id = core_id;
2805 /* TODO: add 'has_node/node' here to describe
2806 to which node core belongs */
2807 }
2808 return machine->possible_cpus;
2809 }
2810
2811 static HotpluggableCPUList *spapr_query_hotpluggable_cpus(MachineState *machine)
2812 {
2813 int i;
2814 Object *cpu;
2815 HotpluggableCPUList *head = NULL;
2816 const char *cpu_type;
2817
2818 cpu = machine->possible_cpus->cpus[0].cpu;
2819 assert(cpu); /* Boot cpu is always present */
2820 cpu_type = object_get_typename(cpu);
2821 for (i = 0; i < machine->possible_cpus->len; i++) {
2822 HotpluggableCPUList *list_item = g_new0(typeof(*list_item), 1);
2823 HotpluggableCPU *cpu_item = g_new0(typeof(*cpu_item), 1);
2824
2825 cpu_item->type = g_strdup(cpu_type);
2826 cpu_item->vcpus_count = smp_threads; // TODO: ??? generalize
2827 cpu_item->props = g_memdup(&machine->possible_cpus->cpus[i].props,
2828 sizeof(*cpu_item->props));
2829
2830 cpu = machine->possible_cpus->cpus[i].cpu;
2831 if (cpu) {
2832 cpu_item->has_qom_path = true;
2833 cpu_item->qom_path = object_get_canonical_path(cpu);
2834 }
2835 list_item->value = cpu_item;
2836 list_item->next = head;
2837 head = list_item;
2838 }
2839 return head;
2840 }
2841
2842 static void spapr_phb_placement(sPAPRMachineState *spapr, uint32_t index,
2843 uint64_t *buid, hwaddr *pio,
2844 hwaddr *mmio32, hwaddr *mmio64,
2845 unsigned n_dma, uint32_t *liobns, Error **errp)
2846 {
2847 /*
2848 * New-style PHB window placement.
2849 *
2850 * Goals: Gives large (1TiB), naturally aligned 64-bit MMIO window
2851 * for each PHB, in addition to 2GiB 32-bit MMIO and 64kiB PIO
2852 * windows.
2853 *
2854 * Some guest kernels can't work with MMIO windows above 1<<46
2855 * (64TiB), so we place up to 31 PHBs in the area 32TiB..64TiB
2856 *
2857 * 32TiB..(33TiB+1984kiB) contains the 64kiB PIO windows for each
2858 * PHB stacked together. (32TiB+2GiB)..(32TiB+64GiB) contains the
2859 * 2GiB 32-bit MMIO windows for each PHB. Then 33..64TiB has the
2860 * 1TiB 64-bit MMIO windows for each PHB.
2861 */
2862 const uint64_t base_buid = 0x800000020000000ULL;
2863 #define SPAPR_MAX_PHBS ((SPAPR_PCI_LIMIT - SPAPR_PCI_BASE) / \
2864 SPAPR_PCI_MEM64_WIN_SIZE - 1)
2865 int i;
2866
2867 /* Sanity check natural alignments */
2868 QEMU_BUILD_BUG_ON((SPAPR_PCI_BASE % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
2869 QEMU_BUILD_BUG_ON((SPAPR_PCI_LIMIT % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
2870 QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM64_WIN_SIZE % SPAPR_PCI_MEM32_WIN_SIZE) != 0);
2871 QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM32_WIN_SIZE % SPAPR_PCI_IO_WIN_SIZE) != 0);
2872 /* Sanity check bounds */
2873 QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_IO_WIN_SIZE) >
2874 SPAPR_PCI_MEM32_WIN_SIZE);
2875 QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_MEM32_WIN_SIZE) >
2876 SPAPR_PCI_MEM64_WIN_SIZE);
2877
2878 if (index >= SPAPR_MAX_PHBS) {
2879 error_setg(errp, "\"index\" for PAPR PHB is too large (max %llu)",
2880 SPAPR_MAX_PHBS - 1);
2881 return;
2882 }
2883
2884 *buid = base_buid + index;
2885 for (i = 0; i < n_dma; ++i) {
2886 liobns[i] = SPAPR_PCI_LIOBN(index, i);
2887 }
2888
2889 *pio = SPAPR_PCI_BASE + index * SPAPR_PCI_IO_WIN_SIZE;
2890 *mmio32 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM32_WIN_SIZE;
2891 *mmio64 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM64_WIN_SIZE;
2892 }
2893
2894 static void spapr_machine_class_init(ObjectClass *oc, void *data)
2895 {
2896 MachineClass *mc = MACHINE_CLASS(oc);
2897 sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(oc);
2898 FWPathProviderClass *fwc = FW_PATH_PROVIDER_CLASS(oc);
2899 NMIClass *nc = NMI_CLASS(oc);
2900 HotplugHandlerClass *hc = HOTPLUG_HANDLER_CLASS(oc);
2901 PPCVirtualHypervisorClass *vhc = PPC_VIRTUAL_HYPERVISOR_CLASS(oc);
2902
2903 mc->desc = "pSeries Logical Partition (PAPR compliant)";
2904
2905 /*
2906 * We set up the default / latest behaviour here. The class_init
2907 * functions for the specific versioned machine types can override
2908 * these details for backwards compatibility
2909 */
2910 mc->init = ppc_spapr_init;
2911 mc->reset = ppc_spapr_reset;
2912 mc->block_default_type = IF_SCSI;
2913 mc->max_cpus = 255;
2914 mc->no_parallel = 1;
2915 mc->default_boot_order = "";
2916 mc->default_ram_size = 512 * M_BYTE;
2917 mc->kvm_type = spapr_kvm_type;
2918 mc->has_dynamic_sysbus = true;
2919 mc->pci_allow_0_address = true;
2920 mc->get_hotplug_handler = spapr_get_hotplug_handler;
2921 hc->pre_plug = spapr_machine_device_pre_plug;
2922 hc->plug = spapr_machine_device_plug;
2923 hc->unplug = spapr_machine_device_unplug;
2924 mc->cpu_index_to_socket_id = spapr_cpu_index_to_socket_id;
2925 mc->possible_cpu_arch_ids = spapr_possible_cpu_arch_ids;
2926 hc->unplug_request = spapr_machine_device_unplug_request;
2927
2928 smc->dr_lmb_enabled = true;
2929 smc->tcg_default_cpu = "POWER8";
2930 mc->query_hotpluggable_cpus = spapr_query_hotpluggable_cpus;
2931 fwc->get_dev_path = spapr_get_fw_dev_path;
2932 nc->nmi_monitor_handler = spapr_nmi;
2933 smc->phb_placement = spapr_phb_placement;
2934 vhc->hypercall = emulate_spapr_hypercall;
2935 }
2936
2937 static const TypeInfo spapr_machine_info = {
2938 .name = TYPE_SPAPR_MACHINE,
2939 .parent = TYPE_MACHINE,
2940 .abstract = true,
2941 .instance_size = sizeof(sPAPRMachineState),
2942 .instance_init = spapr_machine_initfn,
2943 .instance_finalize = spapr_machine_finalizefn,
2944 .class_size = sizeof(sPAPRMachineClass),
2945 .class_init = spapr_machine_class_init,
2946 .interfaces = (InterfaceInfo[]) {
2947 { TYPE_FW_PATH_PROVIDER },
2948 { TYPE_NMI },
2949 { TYPE_HOTPLUG_HANDLER },
2950 { TYPE_PPC_VIRTUAL_HYPERVISOR },
2951 { }
2952 },
2953 };
2954
2955 #define DEFINE_SPAPR_MACHINE(suffix, verstr, latest) \
2956 static void spapr_machine_##suffix##_class_init(ObjectClass *oc, \
2957 void *data) \
2958 { \
2959 MachineClass *mc = MACHINE_CLASS(oc); \
2960 spapr_machine_##suffix##_class_options(mc); \
2961 if (latest) { \
2962 mc->alias = "pseries"; \
2963 mc->is_default = 1; \
2964 } \
2965 } \
2966 static void spapr_machine_##suffix##_instance_init(Object *obj) \
2967 { \
2968 MachineState *machine = MACHINE(obj); \
2969 spapr_machine_##suffix##_instance_options(machine); \
2970 } \
2971 static const TypeInfo spapr_machine_##suffix##_info = { \
2972 .name = MACHINE_TYPE_NAME("pseries-" verstr), \
2973 .parent = TYPE_SPAPR_MACHINE, \
2974 .class_init = spapr_machine_##suffix##_class_init, \
2975 .instance_init = spapr_machine_##suffix##_instance_init, \
2976 }; \
2977 static void spapr_machine_register_##suffix(void) \
2978 { \
2979 type_register(&spapr_machine_##suffix##_info); \
2980 } \
2981 type_init(spapr_machine_register_##suffix)
2982
2983 /*
2984 * pseries-2.9
2985 */
2986 static void spapr_machine_2_9_instance_options(MachineState *machine)
2987 {
2988 }
2989
2990 static void spapr_machine_2_9_class_options(MachineClass *mc)
2991 {
2992 /* Defaults for the latest behaviour inherited from the base class */
2993 }
2994
2995 DEFINE_SPAPR_MACHINE(2_9, "2.9", true);
2996
2997 /*
2998 * pseries-2.8
2999 */
3000 #define SPAPR_COMPAT_2_8 \
3001 HW_COMPAT_2_8
3002
3003 static void spapr_machine_2_8_instance_options(MachineState *machine)
3004 {
3005 spapr_machine_2_9_instance_options(machine);
3006 }
3007
3008 static void spapr_machine_2_8_class_options(MachineClass *mc)
3009 {
3010 spapr_machine_2_9_class_options(mc);
3011 SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_8);
3012 }
3013
3014 DEFINE_SPAPR_MACHINE(2_8, "2.8", false);
3015
3016 /*
3017 * pseries-2.7
3018 */
3019 #define SPAPR_COMPAT_2_7 \
3020 HW_COMPAT_2_7 \
3021 { \
3022 .driver = TYPE_SPAPR_PCI_HOST_BRIDGE, \
3023 .property = "mem_win_size", \
3024 .value = stringify(SPAPR_PCI_2_7_MMIO_WIN_SIZE),\
3025 }, \
3026 { \
3027 .driver = TYPE_SPAPR_PCI_HOST_BRIDGE, \
3028 .property = "mem64_win_size", \
3029 .value = "0", \
3030 }, \
3031 { \
3032 .driver = TYPE_POWERPC_CPU, \
3033 .property = "pre-2.8-migration", \
3034 .value = "on", \
3035 }, \
3036 { \
3037 .driver = TYPE_SPAPR_PCI_HOST_BRIDGE, \
3038 .property = "pre-2.8-migration", \
3039 .value = "on", \
3040 },
3041
3042 static void phb_placement_2_7(sPAPRMachineState *spapr, uint32_t index,
3043 uint64_t *buid, hwaddr *pio,
3044 hwaddr *mmio32, hwaddr *mmio64,
3045 unsigned n_dma, uint32_t *liobns, Error **errp)
3046 {
3047 /* Legacy PHB placement for pseries-2.7 and earlier machine types */
3048 const uint64_t base_buid = 0x800000020000000ULL;
3049 const hwaddr phb_spacing = 0x1000000000ULL; /* 64 GiB */
3050 const hwaddr mmio_offset = 0xa0000000; /* 2 GiB + 512 MiB */
3051 const hwaddr pio_offset = 0x80000000; /* 2 GiB */
3052 const uint32_t max_index = 255;
3053 const hwaddr phb0_alignment = 0x10000000000ULL; /* 1 TiB */
3054
3055 uint64_t ram_top = MACHINE(spapr)->ram_size;
3056 hwaddr phb0_base, phb_base;
3057 int i;
3058
3059 /* Do we have hotpluggable memory? */
3060 if (MACHINE(spapr)->maxram_size > ram_top) {
3061 /* Can't just use maxram_size, because there may be an
3062 * alignment gap between normal and hotpluggable memory
3063 * regions */
3064 ram_top = spapr->hotplug_memory.base +
3065 memory_region_size(&spapr->hotplug_memory.mr);
3066 }
3067
3068 phb0_base = QEMU_ALIGN_UP(ram_top, phb0_alignment);
3069
3070 if (index > max_index) {
3071 error_setg(errp, "\"index\" for PAPR PHB is too large (max %u)",
3072 max_index);
3073 return;
3074 }
3075
3076 *buid = base_buid + index;
3077 for (i = 0; i < n_dma; ++i) {
3078 liobns[i] = SPAPR_PCI_LIOBN(index, i);
3079 }
3080
3081 phb_base = phb0_base + index * phb_spacing;
3082 *pio = phb_base + pio_offset;
3083 *mmio32 = phb_base + mmio_offset;
3084 /*
3085 * We don't set the 64-bit MMIO window, relying on the PHB's
3086 * fallback behaviour of automatically splitting a large "32-bit"
3087 * window into contiguous 32-bit and 64-bit windows
3088 */
3089 }
3090
3091 static void spapr_machine_2_7_instance_options(MachineState *machine)
3092 {
3093 sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
3094
3095 spapr_machine_2_8_instance_options(machine);
3096 spapr->use_hotplug_event_source = false;
3097 }
3098
3099 static void spapr_machine_2_7_class_options(MachineClass *mc)
3100 {
3101 sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3102
3103 spapr_machine_2_8_class_options(mc);
3104 smc->tcg_default_cpu = "POWER7";
3105 SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_7);
3106 smc->phb_placement = phb_placement_2_7;
3107 }
3108
3109 DEFINE_SPAPR_MACHINE(2_7, "2.7", false);
3110
3111 /*
3112 * pseries-2.6
3113 */
3114 #define SPAPR_COMPAT_2_6 \
3115 HW_COMPAT_2_6 \
3116 { \
3117 .driver = TYPE_SPAPR_PCI_HOST_BRIDGE,\
3118 .property = "ddw",\
3119 .value = stringify(off),\
3120 },
3121
3122 static void spapr_machine_2_6_instance_options(MachineState *machine)
3123 {
3124 spapr_machine_2_7_instance_options(machine);
3125 }
3126
3127 static void spapr_machine_2_6_class_options(MachineClass *mc)
3128 {
3129 spapr_machine_2_7_class_options(mc);
3130 mc->query_hotpluggable_cpus = NULL;
3131 SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_6);
3132 }
3133
3134 DEFINE_SPAPR_MACHINE(2_6, "2.6", false);
3135
3136 /*
3137 * pseries-2.5
3138 */
3139 #define SPAPR_COMPAT_2_5 \
3140 HW_COMPAT_2_5 \
3141 { \
3142 .driver = "spapr-vlan", \
3143 .property = "use-rx-buffer-pools", \
3144 .value = "off", \
3145 },
3146
3147 static void spapr_machine_2_5_instance_options(MachineState *machine)
3148 {
3149 spapr_machine_2_6_instance_options(machine);
3150 }
3151
3152 static void spapr_machine_2_5_class_options(MachineClass *mc)
3153 {
3154 sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3155
3156 spapr_machine_2_6_class_options(mc);
3157 smc->use_ohci_by_default = true;
3158 SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_5);
3159 }
3160
3161 DEFINE_SPAPR_MACHINE(2_5, "2.5", false);
3162
3163 /*
3164 * pseries-2.4
3165 */
3166 #define SPAPR_COMPAT_2_4 \
3167 HW_COMPAT_2_4
3168
3169 static void spapr_machine_2_4_instance_options(MachineState *machine)
3170 {
3171 spapr_machine_2_5_instance_options(machine);
3172 }
3173
3174 static void spapr_machine_2_4_class_options(MachineClass *mc)
3175 {
3176 sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3177
3178 spapr_machine_2_5_class_options(mc);
3179 smc->dr_lmb_enabled = false;
3180 SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_4);
3181 }
3182
3183 DEFINE_SPAPR_MACHINE(2_4, "2.4", false);
3184
3185 /*
3186 * pseries-2.3
3187 */
3188 #define SPAPR_COMPAT_2_3 \
3189 HW_COMPAT_2_3 \
3190 {\
3191 .driver = "spapr-pci-host-bridge",\
3192 .property = "dynamic-reconfiguration",\
3193 .value = "off",\
3194 },
3195
3196 static void spapr_machine_2_3_instance_options(MachineState *machine)
3197 {
3198 spapr_machine_2_4_instance_options(machine);
3199 savevm_skip_section_footers();
3200 global_state_set_optional();
3201 savevm_skip_configuration();
3202 }
3203
3204 static void spapr_machine_2_3_class_options(MachineClass *mc)
3205 {
3206 spapr_machine_2_4_class_options(mc);
3207 SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_3);
3208 }
3209 DEFINE_SPAPR_MACHINE(2_3, "2.3", false);
3210
3211 /*
3212 * pseries-2.2
3213 */
3214
3215 #define SPAPR_COMPAT_2_2 \
3216 HW_COMPAT_2_2 \
3217 {\
3218 .driver = TYPE_SPAPR_PCI_HOST_BRIDGE,\
3219 .property = "mem_win_size",\
3220 .value = "0x20000000",\
3221 },
3222
3223 static void spapr_machine_2_2_instance_options(MachineState *machine)
3224 {
3225 spapr_machine_2_3_instance_options(machine);
3226 machine->suppress_vmdesc = true;
3227 }
3228
3229 static void spapr_machine_2_2_class_options(MachineClass *mc)
3230 {
3231 spapr_machine_2_3_class_options(mc);
3232 SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_2);
3233 }
3234 DEFINE_SPAPR_MACHINE(2_2, "2.2", false);
3235
3236 /*
3237 * pseries-2.1
3238 */
3239 #define SPAPR_COMPAT_2_1 \
3240 HW_COMPAT_2_1
3241
3242 static void spapr_machine_2_1_instance_options(MachineState *machine)
3243 {
3244 spapr_machine_2_2_instance_options(machine);
3245 }
3246
3247 static void spapr_machine_2_1_class_options(MachineClass *mc)
3248 {
3249 spapr_machine_2_2_class_options(mc);
3250 SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_1);
3251 }
3252 DEFINE_SPAPR_MACHINE(2_1, "2.1", false);
3253
3254 static void spapr_machine_register_types(void)
3255 {
3256 type_register_static(&spapr_machine_info);
3257 }
3258
3259 type_init(spapr_machine_register_types)
Random patches