Merge tag 'v2.8.0-rc3'
[qemu/ar7.git] / hw / ppc / spapr.c
bloba642e663d497c27c029a254ed2b1306b1125ccea
1 /*
2 * QEMU PowerPC pSeries Logical Partition (aka sPAPR) hardware System Emulator
4 * Copyright (c) 2004-2007 Fabrice Bellard
5 * Copyright (c) 2007 Jocelyn Mayer
6 * Copyright (c) 2010 David Gibson, IBM Corporation.
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:
15 * The above copyright notice and this permission notice shall be included in
16 * all copies or substantial portions of the Software.
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.
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"
45 #include "hw/boards.h"
46 #include "hw/ppc/ppc.h"
47 #include "hw/loader.h"
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"
56 #include "hw/pci/pci.h"
57 #include "hw/scsi/scsi.h"
58 #include "hw/virtio/virtio-scsi.h"
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"
67 #include "hw/compat.h"
68 #include "qemu/cutils.h"
69 #include "hw/ppc/spapr_cpu_core.h"
70 #include "qmp-commands.h"
72 #include <libfdt.h>
74 /* SLOF memory layout:
76 * SLOF raw image loaded at 0, copies its romfs right below the flat
77 * device-tree, then position SLOF itself 31M below that
79 * So we set FW_OVERHEAD to 40MB which should account for all of that
80 * and more
82 * We load our kernel at 4M, leaving space for SLOF initial image
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
92 #define MIN_RMA_SLOF 128UL
94 #define PHANDLE_XICP 0x00001111
96 #define HTAB_SIZE(spapr) (1ULL << ((spapr)->htab_shift))
98 static XICSState *try_create_xics(const char *type, int nr_servers,
99 int nr_irqs, Error **errp)
101 Error *err = NULL;
102 DeviceState *dev;
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;
113 return XICS_COMMON(dev);
116 static XICSState *xics_system_init(MachineState *machine,
117 int nr_servers, int nr_irqs, Error **errp)
119 XICSState *xics = NULL;
121 if (kvm_enabled()) {
122 Error *err = NULL;
124 if (machine_kernel_irqchip_allowed(machine)) {
125 xics = try_create_xics(TYPE_XICS_SPAPR_KVM, nr_servers, nr_irqs,
126 &err);
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);
136 if (!xics) {
137 xics = try_create_xics(TYPE_XICS_SPAPR, nr_servers, nr_irqs, errp);
140 return xics;
143 static int spapr_fixup_cpu_smt_dt(void *fdt, int offset, PowerPCCPU *cpu,
144 int smt_threads)
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);
151 if (cpu->cpu_version) {
152 ret = fdt_setprop_cell(fdt, offset, "cpu-version", cpu->cpu_version);
153 if (ret < 0) {
154 return ret;
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;
165 ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-server#s",
166 servers_prop, sizeof(servers_prop));
167 if (ret < 0) {
168 return ret;
170 ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-gserver#s",
171 gservers_prop, sizeof(gservers_prop));
173 return ret;
176 static int spapr_fixup_cpu_numa_dt(void *fdt, int offset, CPUState *cs)
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)};
188 /* Advertise NUMA via ibm,associativity */
189 if (nb_numa_nodes > 1) {
190 ret = fdt_setprop(fdt, offset, "ibm,associativity", associativity,
191 sizeof(associativity));
194 return ret;
197 static int spapr_fixup_cpu_dt(void *fdt, sPAPRMachineState *spapr)
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)};
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);
210 if ((index % smt) != 0) {
211 continue;
214 snprintf(cpu_model, 32, "%s@%x", dc->fw_name, index);
216 cpus_offset = fdt_path_offset(fdt, "/cpus");
217 if (cpus_offset < 0) {
218 cpus_offset = fdt_add_subnode(fdt, fdt_path_offset(fdt, "/"),
219 "cpus");
220 if (cpus_offset < 0) {
221 return cpus_offset;
224 offset = fdt_subnode_offset(fdt, cpus_offset, cpu_model);
225 if (offset < 0) {
226 offset = fdt_add_subnode(fdt, cpus_offset, cpu_model);
227 if (offset < 0) {
228 return offset;
232 ret = fdt_setprop(fdt, offset, "ibm,pft-size",
233 pft_size_prop, sizeof(pft_size_prop));
234 if (ret < 0) {
235 return ret;
238 ret = spapr_fixup_cpu_numa_dt(fdt, offset, cs);
239 if (ret < 0) {
240 return ret;
243 ret = spapr_fixup_cpu_smt_dt(fdt, offset, cpu,
244 ppc_get_compat_smt_threads(cpu));
245 if (ret < 0) {
246 return ret;
249 return ret;
252 static hwaddr spapr_node0_size(void)
254 MachineState *machine = MACHINE(qdev_get_machine());
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);
265 return machine->ram_size;
268 static void add_str(GString *s, const gchar *s1)
270 g_string_append_len(s, s1, strlen(s1) + 1);
273 static int spapr_populate_memory_node(void *fdt, int nodeid, hwaddr start,
274 hwaddr size)
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)
281 char mem_name[32];
282 uint64_t mem_reg_property[2];
283 int off;
285 mem_reg_property[0] = cpu_to_be64(start);
286 mem_reg_property[1] = cpu_to_be64(size);
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;
299 static int spapr_populate_memory(sPAPRMachineState *spapr, void *fdt)
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;
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;
314 for (i = 0, mem_start = 0; i < nb_nodes; ++i) {
315 if (!nodes[i].node_mem) {
316 continue;
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;
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;
332 for ( ; node_size; ) {
333 hwaddr sizetmp = pow2floor(node_size);
335 /* mem_start != 0 here */
336 if (ctzl(mem_start) < ctzl(sizetmp)) {
337 sizetmp = 1ULL << ctzl(mem_start);
340 spapr_populate_memory_node(fdt, i, mem_start, sizetmp);
341 node_size -= sizetmp;
342 mem_start += sizetmp;
346 return 0;
349 /* Populate the "ibm,pa-features" property */
350 static void spapr_populate_pa_features(CPUPPCState *env, void *fdt, int offset)
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;
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;
377 if (env->ci_large_pages) {
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
385 pa_features[3] |= 0x20;
387 if (kvmppc_has_cap_htm() && pa_size > 24) {
388 pa_features[24] |= 0x80; /* Transactional memory support */
391 _FDT((fdt_setprop(fdt, offset, "ibm,pa-features", pa_features, pa_size)));
394 static void spapr_populate_cpu_dt(CPUState *cs, void *fdt, int offset,
395 sPAPRMachineState *spapr)
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 sPAPRDRConnector *drc;
411 sPAPRDRConnectorClass *drck;
412 int drc_index;
414 drc = spapr_dr_connector_by_id(SPAPR_DR_CONNECTOR_TYPE_CPU, index);
415 if (drc) {
416 drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
417 drc_index = drck->get_index(drc);
418 _FDT((fdt_setprop_cell(fdt, offset, "ibm,my-drc-index", drc_index)));
421 _FDT((fdt_setprop_cell(fdt, offset, "reg", index)));
422 _FDT((fdt_setprop_string(fdt, offset, "device_type", "cpu")));
424 _FDT((fdt_setprop_cell(fdt, offset, "cpu-version", env->spr[SPR_PVR])));
425 _FDT((fdt_setprop_cell(fdt, offset, "d-cache-block-size",
426 env->dcache_line_size)));
427 _FDT((fdt_setprop_cell(fdt, offset, "d-cache-line-size",
428 env->dcache_line_size)));
429 _FDT((fdt_setprop_cell(fdt, offset, "i-cache-block-size",
430 env->icache_line_size)));
431 _FDT((fdt_setprop_cell(fdt, offset, "i-cache-line-size",
432 env->icache_line_size)));
434 if (pcc->l1_dcache_size) {
435 _FDT((fdt_setprop_cell(fdt, offset, "d-cache-size",
436 pcc->l1_dcache_size)));
437 } else {
438 error_report("Warning: Unknown L1 dcache size for cpu");
440 if (pcc->l1_icache_size) {
441 _FDT((fdt_setprop_cell(fdt, offset, "i-cache-size",
442 pcc->l1_icache_size)));
443 } else {
444 error_report("Warning: Unknown L1 icache size for cpu");
447 _FDT((fdt_setprop_cell(fdt, offset, "timebase-frequency", tbfreq)));
448 _FDT((fdt_setprop_cell(fdt, offset, "clock-frequency", cpufreq)));
449 _FDT((fdt_setprop_cell(fdt, offset, "slb-size", env->slb_nr)));
450 _FDT((fdt_setprop_cell(fdt, offset, "ibm,slb-size", env->slb_nr)));
451 _FDT((fdt_setprop_string(fdt, offset, "status", "okay")));
452 _FDT((fdt_setprop(fdt, offset, "64-bit", NULL, 0)));
454 if (env->spr_cb[SPR_PURR].oea_read) {
455 _FDT((fdt_setprop(fdt, offset, "ibm,purr", NULL, 0)));
458 if (env->mmu_model & POWERPC_MMU_1TSEG) {
459 _FDT((fdt_setprop(fdt, offset, "ibm,processor-segment-sizes",
460 segs, sizeof(segs))));
463 /* Advertise VMX/VSX (vector extensions) if available
464 * 0 / no property == no vector extensions
465 * 1 == VMX / Altivec available
466 * 2 == VSX available */
467 if (env->insns_flags & PPC_ALTIVEC) {
468 uint32_t vmx = (env->insns_flags2 & PPC2_VSX) ? 2 : 1;
470 _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", vmx)));
473 /* Advertise DFP (Decimal Floating Point) if available
474 * 0 / no property == no DFP
475 * 1 == DFP available */
476 if (env->insns_flags2 & PPC2_DFP) {
477 _FDT((fdt_setprop_cell(fdt, offset, "ibm,dfp", 1)));
480 page_sizes_prop_size = ppc_create_page_sizes_prop(env, page_sizes_prop,
481 sizeof(page_sizes_prop));
482 if (page_sizes_prop_size) {
483 _FDT((fdt_setprop(fdt, offset, "ibm,segment-page-sizes",
484 page_sizes_prop, page_sizes_prop_size)));
487 spapr_populate_pa_features(env, fdt, offset);
489 _FDT((fdt_setprop_cell(fdt, offset, "ibm,chip-id",
490 cs->cpu_index / vcpus_per_socket)));
492 _FDT((fdt_setprop(fdt, offset, "ibm,pft-size",
493 pft_size_prop, sizeof(pft_size_prop))));
495 _FDT(spapr_fixup_cpu_numa_dt(fdt, offset, cs));
497 _FDT(spapr_fixup_cpu_smt_dt(fdt, offset, cpu,
498 ppc_get_compat_smt_threads(cpu)));
501 static void spapr_populate_cpus_dt_node(void *fdt, sPAPRMachineState *spapr)
503 CPUState *cs;
504 int cpus_offset;
505 char *nodename;
506 int smt = kvmppc_smt_threads();
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)));
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.
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;
524 if ((index % smt) != 0) {
525 continue;
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);
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.
542 static int spapr_populate_drconf_memory(sPAPRMachineState *spapr, void *fdt)
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;
556 * Don't create the node if there is no hotpluggable memory
558 if (machine->ram_size == machine->maxram_size) {
559 return 0;
563 * Allocate enough buffer size to fit in ibm,dynamic-memory
564 * or ibm,associativity-lookup-arrays
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);
570 offset = fdt_add_subnode(fdt, 0, "ibm,dynamic-reconfiguration-memory");
572 ret = fdt_setprop(fdt, offset, "ibm,lmb-size", prop_lmb_size,
573 sizeof(prop_lmb_size));
574 if (ret < 0) {
575 goto out;
578 ret = fdt_setprop_cell(fdt, offset, "ibm,memory-flags-mask", 0xff);
579 if (ret < 0) {
580 goto out;
583 ret = fdt_setprop_cell(fdt, offset, "ibm,memory-preservation-time", 0x0);
584 if (ret < 0) {
585 goto out;
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;
595 if (i >= hotplug_lmb_start) {
596 sPAPRDRConnector *drc;
597 sPAPRDRConnectorClass *drck;
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);
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);
613 } else {
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.
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);
628 cur_index += SPAPR_DR_LMB_LIST_ENTRY_SIZE;
630 ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory", int_buf, buf_len);
631 if (ret < 0) {
632 goto out;
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)
647 memcpy(cur_index, associativity, sizeof(associativity));
648 cur_index += 4;
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;
657 static int spapr_dt_cas_updates(sPAPRMachineState *spapr, void *fdt,
658 sPAPROptionVector *ov5_updates)
660 sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
661 int ret = 0, offset;
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;
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;
679 ret = spapr_ovec_populate_dt(fdt, offset, spapr->ov5_cas,
680 "ibm,architecture-vec-5");
682 out:
683 return ret;
686 int spapr_h_cas_compose_response(sPAPRMachineState *spapr,
687 target_ulong addr, target_ulong size,
688 bool cpu_update,
689 sPAPROptionVector *ov5_updates)
691 void *fdt, *fdt_skel;
692 sPAPRDeviceTreeUpdateHeader hdr = { .version_id = 1 };
694 size -= sizeof(hdr);
696 /* Create sceleton */
697 fdt_skel = g_malloc0(size);
698 _FDT((fdt_create(fdt_skel, size)));
699 _FDT((fdt_begin_node(fdt_skel, "")));
700 _FDT((fdt_end_node(fdt_skel)));
701 _FDT((fdt_finish(fdt_skel)));
702 fdt = g_malloc0(size);
703 _FDT((fdt_open_into(fdt_skel, fdt, size)));
704 g_free(fdt_skel);
706 /* Fixup cpu nodes */
707 if (cpu_update) {
708 _FDT((spapr_fixup_cpu_dt(fdt, spapr)));
711 if (spapr_dt_cas_updates(spapr, fdt, ov5_updates)) {
712 return -1;
715 /* Pack resulting tree */
716 _FDT((fdt_pack(fdt)));
718 if (fdt_totalsize(fdt) + sizeof(hdr) > size) {
719 trace_spapr_cas_failed(size);
720 return -1;
723 cpu_physical_memory_write(addr, &hdr, sizeof(hdr));
724 cpu_physical_memory_write(addr + sizeof(hdr), fdt, fdt_totalsize(fdt));
725 trace_spapr_cas_continue(fdt_totalsize(fdt) + sizeof(hdr));
726 g_free(fdt);
728 return 0;
731 static void spapr_dt_rtas(sPAPRMachineState *spapr, void *fdt)
733 int rtas;
734 GString *hypertas = g_string_sized_new(256);
735 GString *qemu_hypertas = g_string_sized_new(256);
736 uint32_t refpoints[] = { cpu_to_be32(0x4), cpu_to_be32(0x4) };
737 uint64_t max_hotplug_addr = spapr->hotplug_memory.base +
738 memory_region_size(&spapr->hotplug_memory.mr);
739 uint32_t lrdr_capacity[] = {
740 cpu_to_be32(max_hotplug_addr >> 32),
741 cpu_to_be32(max_hotplug_addr & 0xffffffff),
742 0, cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE),
743 cpu_to_be32(max_cpus / smp_threads),
746 _FDT(rtas = fdt_add_subnode(fdt, 0, "rtas"));
748 /* hypertas */
749 add_str(hypertas, "hcall-pft");
750 add_str(hypertas, "hcall-term");
751 add_str(hypertas, "hcall-dabr");
752 add_str(hypertas, "hcall-interrupt");
753 add_str(hypertas, "hcall-tce");
754 add_str(hypertas, "hcall-vio");
755 add_str(hypertas, "hcall-splpar");
756 add_str(hypertas, "hcall-bulk");
757 add_str(hypertas, "hcall-set-mode");
758 add_str(hypertas, "hcall-sprg0");
759 add_str(hypertas, "hcall-copy");
760 add_str(hypertas, "hcall-debug");
761 add_str(qemu_hypertas, "hcall-memop1");
763 if (!kvm_enabled() || kvmppc_spapr_use_multitce()) {
764 add_str(hypertas, "hcall-multi-tce");
766 _FDT(fdt_setprop(fdt, rtas, "ibm,hypertas-functions",
767 hypertas->str, hypertas->len));
768 g_string_free(hypertas, TRUE);
769 _FDT(fdt_setprop(fdt, rtas, "qemu,hypertas-functions",
770 qemu_hypertas->str, qemu_hypertas->len));
771 g_string_free(qemu_hypertas, TRUE);
773 _FDT(fdt_setprop(fdt, rtas, "ibm,associativity-reference-points",
774 refpoints, sizeof(refpoints)));
776 _FDT(fdt_setprop_cell(fdt, rtas, "rtas-error-log-max",
777 RTAS_ERROR_LOG_MAX));
778 _FDT(fdt_setprop_cell(fdt, rtas, "rtas-event-scan-rate",
779 RTAS_EVENT_SCAN_RATE));
781 if (msi_nonbroken) {
782 _FDT(fdt_setprop(fdt, rtas, "ibm,change-msix-capable", NULL, 0));
786 * According to PAPR, rtas ibm,os-term does not guarantee a return
787 * back to the guest cpu.
789 * While an additional ibm,extended-os-term property indicates
790 * that rtas call return will always occur. Set this property.
792 _FDT(fdt_setprop(fdt, rtas, "ibm,extended-os-term", NULL, 0));
794 _FDT(fdt_setprop(fdt, rtas, "ibm,lrdr-capacity",
795 lrdr_capacity, sizeof(lrdr_capacity)));
797 spapr_dt_rtas_tokens(fdt, rtas);
800 static void spapr_dt_chosen(sPAPRMachineState *spapr, void *fdt)
802 MachineState *machine = MACHINE(spapr);
803 int chosen;
804 const char *boot_device = machine->boot_order;
805 char *stdout_path = spapr_vio_stdout_path(spapr->vio_bus);
806 size_t cb = 0;
807 char *bootlist = get_boot_devices_list(&cb, true);
809 _FDT(chosen = fdt_add_subnode(fdt, 0, "chosen"));
811 _FDT(fdt_setprop_string(fdt, chosen, "bootargs", machine->kernel_cmdline));
812 _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-start",
813 spapr->initrd_base));
814 _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-end",
815 spapr->initrd_base + spapr->initrd_size));
817 if (spapr->kernel_size) {
818 uint64_t kprop[2] = { cpu_to_be64(KERNEL_LOAD_ADDR),
819 cpu_to_be64(spapr->kernel_size) };
821 _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel",
822 &kprop, sizeof(kprop)));
823 if (spapr->kernel_le) {
824 _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel-le", NULL, 0));
827 if (boot_menu) {
828 _FDT((fdt_setprop_cell(fdt, chosen, "qemu,boot-menu", boot_menu)));
830 _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-width", graphic_width));
831 _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-height", graphic_height));
832 _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-depth", graphic_depth));
834 if (cb && bootlist) {
835 int i;
837 for (i = 0; i < cb; i++) {
838 if (bootlist[i] == '\n') {
839 bootlist[i] = ' ';
842 _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-list", bootlist));
845 if (boot_device && strlen(boot_device)) {
846 _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-device", boot_device));
849 if (!spapr->has_graphics && stdout_path) {
850 _FDT(fdt_setprop_string(fdt, chosen, "linux,stdout-path", stdout_path));
853 g_free(stdout_path);
854 g_free(bootlist);
857 static void spapr_dt_hypervisor(sPAPRMachineState *spapr, void *fdt)
859 /* The /hypervisor node isn't in PAPR - this is a hack to allow PR
860 * KVM to work under pHyp with some guest co-operation */
861 int hypervisor;
862 uint8_t hypercall[16];
864 _FDT(hypervisor = fdt_add_subnode(fdt, 0, "hypervisor"));
865 /* indicate KVM hypercall interface */
866 _FDT(fdt_setprop_string(fdt, hypervisor, "compatible", "linux,kvm"));
867 if (kvmppc_has_cap_fixup_hcalls()) {
869 * Older KVM versions with older guest kernels were broken
870 * with the magic page, don't allow the guest to map it.
872 if (!kvmppc_get_hypercall(first_cpu->env_ptr, hypercall,
873 sizeof(hypercall))) {
874 _FDT(fdt_setprop(fdt, hypervisor, "hcall-instructions",
875 hypercall, sizeof(hypercall)));
880 static void *spapr_build_fdt(sPAPRMachineState *spapr,
881 hwaddr rtas_addr,
882 hwaddr rtas_size)
884 MachineState *machine = MACHINE(qdev_get_machine());
885 MachineClass *mc = MACHINE_GET_CLASS(machine);
886 sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
887 int ret;
888 void *fdt;
889 sPAPRPHBState *phb;
890 char *buf;
892 fdt = g_malloc0(FDT_MAX_SIZE);
893 _FDT((fdt_create_empty_tree(fdt, FDT_MAX_SIZE)));
895 /* Root node */
896 _FDT(fdt_setprop_string(fdt, 0, "device_type", "chrp"));
897 _FDT(fdt_setprop_string(fdt, 0, "model", "IBM pSeries (emulated by qemu)"));
898 _FDT(fdt_setprop_string(fdt, 0, "compatible", "qemu,pseries"));
901 * Add info to guest to indentify which host is it being run on
902 * and what is the uuid of the guest
904 if (kvmppc_get_host_model(&buf)) {
905 _FDT(fdt_setprop_string(fdt, 0, "host-model", buf));
906 g_free(buf);
908 if (kvmppc_get_host_serial(&buf)) {
909 _FDT(fdt_setprop_string(fdt, 0, "host-serial", buf));
910 g_free(buf);
913 buf = qemu_uuid_unparse_strdup(&qemu_uuid);
915 _FDT(fdt_setprop_string(fdt, 0, "vm,uuid", buf));
916 if (qemu_uuid_set) {
917 _FDT(fdt_setprop_string(fdt, 0, "system-id", buf));
919 g_free(buf);
921 if (qemu_get_vm_name()) {
922 _FDT(fdt_setprop_string(fdt, 0, "ibm,partition-name",
923 qemu_get_vm_name()));
926 _FDT(fdt_setprop_cell(fdt, 0, "#address-cells", 2));
927 _FDT(fdt_setprop_cell(fdt, 0, "#size-cells", 2));
929 /* /interrupt controller */
930 spapr_dt_xics(spapr->xics, fdt, PHANDLE_XICP);
932 ret = spapr_populate_memory(spapr, fdt);
933 if (ret < 0) {
934 error_report("couldn't setup memory nodes in fdt");
935 exit(1);
938 /* /vdevice */
939 spapr_dt_vdevice(spapr->vio_bus, fdt);
941 if (object_resolve_path_type("", TYPE_SPAPR_RNG, NULL)) {
942 ret = spapr_rng_populate_dt(fdt);
943 if (ret < 0) {
944 error_report("could not set up rng device in the fdt");
945 exit(1);
949 QLIST_FOREACH(phb, &spapr->phbs, list) {
950 ret = spapr_populate_pci_dt(phb, PHANDLE_XICP, fdt);
951 if (ret < 0) {
952 error_report("couldn't setup PCI devices in fdt");
953 exit(1);
957 /* cpus */
958 spapr_populate_cpus_dt_node(fdt, spapr);
960 if (smc->dr_lmb_enabled) {
961 _FDT(spapr_drc_populate_dt(fdt, 0, NULL, SPAPR_DR_CONNECTOR_TYPE_LMB));
964 if (mc->query_hotpluggable_cpus) {
965 int offset = fdt_path_offset(fdt, "/cpus");
966 ret = spapr_drc_populate_dt(fdt, offset, NULL,
967 SPAPR_DR_CONNECTOR_TYPE_CPU);
968 if (ret < 0) {
969 error_report("Couldn't set up CPU DR device tree properties");
970 exit(1);
974 /* /event-sources */
975 spapr_dt_events(spapr, fdt);
977 /* /rtas */
978 spapr_dt_rtas(spapr, fdt);
980 /* /chosen */
981 spapr_dt_chosen(spapr, fdt);
983 /* /hypervisor */
984 if (kvm_enabled()) {
985 spapr_dt_hypervisor(spapr, fdt);
988 /* Build memory reserve map */
989 if (spapr->kernel_size) {
990 _FDT((fdt_add_mem_rsv(fdt, KERNEL_LOAD_ADDR, spapr->kernel_size)));
992 if (spapr->initrd_size) {
993 _FDT((fdt_add_mem_rsv(fdt, spapr->initrd_base, spapr->initrd_size)));
996 /* ibm,client-architecture-support updates */
997 ret = spapr_dt_cas_updates(spapr, fdt, spapr->ov5_cas);
998 if (ret < 0) {
999 error_report("couldn't setup CAS properties fdt");
1000 exit(1);
1003 return fdt;
1006 static uint64_t translate_kernel_address(void *opaque, uint64_t addr)
1008 return (addr & 0x0fffffff) + KERNEL_LOAD_ADDR;
1011 static void emulate_spapr_hypercall(PowerPCCPU *cpu)
1013 CPUPPCState *env = &cpu->env;
1015 if (msr_pr) {
1016 hcall_dprintf("Hypercall made with MSR[PR]=1\n");
1017 env->gpr[3] = H_PRIVILEGE;
1018 } else {
1019 env->gpr[3] = spapr_hypercall(cpu, env->gpr[3], &env->gpr[4]);
1023 #define HPTE(_table, _i) (void *)(((uint64_t *)(_table)) + ((_i) * 2))
1024 #define HPTE_VALID(_hpte) (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_VALID)
1025 #define HPTE_DIRTY(_hpte) (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_HPTE_DIRTY)
1026 #define CLEAN_HPTE(_hpte) ((*(uint64_t *)(_hpte)) &= tswap64(~HPTE64_V_HPTE_DIRTY))
1027 #define DIRTY_HPTE(_hpte) ((*(uint64_t *)(_hpte)) |= tswap64(HPTE64_V_HPTE_DIRTY))
1030 * Get the fd to access the kernel htab, re-opening it if necessary
1032 static int get_htab_fd(sPAPRMachineState *spapr)
1034 if (spapr->htab_fd >= 0) {
1035 return spapr->htab_fd;
1038 spapr->htab_fd = kvmppc_get_htab_fd(false);
1039 if (spapr->htab_fd < 0) {
1040 error_report("Unable to open fd for reading hash table from KVM: %s",
1041 strerror(errno));
1044 return spapr->htab_fd;
1047 static void close_htab_fd(sPAPRMachineState *spapr)
1049 if (spapr->htab_fd >= 0) {
1050 close(spapr->htab_fd);
1052 spapr->htab_fd = -1;
1055 static int spapr_hpt_shift_for_ramsize(uint64_t ramsize)
1057 int shift;
1059 /* We aim for a hash table of size 1/128 the size of RAM (rounded
1060 * up). The PAPR recommendation is actually 1/64 of RAM size, but
1061 * that's much more than is needed for Linux guests */
1062 shift = ctz64(pow2ceil(ramsize)) - 7;
1063 shift = MAX(shift, 18); /* Minimum architected size */
1064 shift = MIN(shift, 46); /* Maximum architected size */
1065 return shift;
1068 static void spapr_reallocate_hpt(sPAPRMachineState *spapr, int shift,
1069 Error **errp)
1071 long rc;
1073 /* Clean up any HPT info from a previous boot */
1074 g_free(spapr->htab);
1075 spapr->htab = NULL;
1076 spapr->htab_shift = 0;
1077 close_htab_fd(spapr);
1079 rc = kvmppc_reset_htab(shift);
1080 if (rc < 0) {
1081 /* kernel-side HPT needed, but couldn't allocate one */
1082 error_setg_errno(errp, errno,
1083 "Failed to allocate KVM HPT of order %d (try smaller maxmem?)",
1084 shift);
1085 /* This is almost certainly fatal, but if the caller really
1086 * wants to carry on with shift == 0, it's welcome to try */
1087 } else if (rc > 0) {
1088 /* kernel-side HPT allocated */
1089 if (rc != shift) {
1090 error_setg(errp,
1091 "Requested order %d HPT, but kernel allocated order %ld (try smaller maxmem?)",
1092 shift, rc);
1095 spapr->htab_shift = shift;
1096 spapr->htab = NULL;
1097 } else {
1098 /* kernel-side HPT not needed, allocate in userspace instead */
1099 size_t size = 1ULL << shift;
1100 int i;
1102 spapr->htab = qemu_memalign(size, size);
1103 if (!spapr->htab) {
1104 error_setg_errno(errp, errno,
1105 "Could not allocate HPT of order %d", shift);
1106 return;
1109 memset(spapr->htab, 0, size);
1110 spapr->htab_shift = shift;
1112 for (i = 0; i < size / HASH_PTE_SIZE_64; i++) {
1113 DIRTY_HPTE(HPTE(spapr->htab, i));
1118 static void find_unknown_sysbus_device(SysBusDevice *sbdev, void *opaque)
1120 bool matched = false;
1122 if (object_dynamic_cast(OBJECT(sbdev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
1123 matched = true;
1126 if (!matched) {
1127 error_report("Device %s is not supported by this machine yet.",
1128 qdev_fw_name(DEVICE(sbdev)));
1129 exit(1);
1133 static void ppc_spapr_reset(void)
1135 MachineState *machine = MACHINE(qdev_get_machine());
1136 sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
1137 PowerPCCPU *first_ppc_cpu;
1138 uint32_t rtas_limit;
1139 hwaddr rtas_addr, fdt_addr;
1140 void *fdt;
1141 int rc;
1143 /* Check for unknown sysbus devices */
1144 foreach_dynamic_sysbus_device(find_unknown_sysbus_device, NULL);
1146 /* Allocate and/or reset the hash page table */
1147 spapr_reallocate_hpt(spapr,
1148 spapr_hpt_shift_for_ramsize(machine->maxram_size),
1149 &error_fatal);
1151 /* Update the RMA size if necessary */
1152 if (spapr->vrma_adjust) {
1153 spapr->rma_size = kvmppc_rma_size(spapr_node0_size(),
1154 spapr->htab_shift);
1157 qemu_devices_reset();
1160 * We place the device tree and RTAS just below either the top of the RMA,
1161 * or just below 2GB, whichever is lowere, so that it can be
1162 * processed with 32-bit real mode code if necessary
1164 rtas_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR);
1165 rtas_addr = rtas_limit - RTAS_MAX_SIZE;
1166 fdt_addr = rtas_addr - FDT_MAX_SIZE;
1168 /* if this reset wasn't generated by CAS, we should reset our
1169 * negotiated options and start from scratch */
1170 if (!spapr->cas_reboot) {
1171 spapr_ovec_cleanup(spapr->ov5_cas);
1172 spapr->ov5_cas = spapr_ovec_new();
1175 fdt = spapr_build_fdt(spapr, rtas_addr, spapr->rtas_size);
1177 spapr_load_rtas(spapr, fdt, rtas_addr);
1179 rc = fdt_pack(fdt);
1181 /* Should only fail if we've built a corrupted tree */
1182 assert(rc == 0);
1184 if (fdt_totalsize(fdt) > FDT_MAX_SIZE) {
1185 error_report("FDT too big ! 0x%x bytes (max is 0x%x)",
1186 fdt_totalsize(fdt), FDT_MAX_SIZE);
1187 exit(1);
1190 /* Load the fdt */
1191 qemu_fdt_dumpdtb(fdt, fdt_totalsize(fdt));
1192 cpu_physical_memory_write(fdt_addr, fdt, fdt_totalsize(fdt));
1193 g_free(fdt);
1195 /* Set up the entry state */
1196 first_ppc_cpu = POWERPC_CPU(first_cpu);
1197 first_ppc_cpu->env.gpr[3] = fdt_addr;
1198 first_ppc_cpu->env.gpr[5] = 0;
1199 first_cpu->halted = 0;
1200 first_ppc_cpu->env.nip = SPAPR_ENTRY_POINT;
1202 spapr->cas_reboot = false;
1205 static void spapr_create_nvram(sPAPRMachineState *spapr)
1207 DeviceState *dev = qdev_create(&spapr->vio_bus->bus, "spapr-nvram");
1208 DriveInfo *dinfo = drive_get(IF_PFLASH, 0, 0);
1210 if (dinfo) {
1211 qdev_prop_set_drive(dev, "drive", blk_by_legacy_dinfo(dinfo),
1212 &error_fatal);
1215 qdev_init_nofail(dev);
1217 spapr->nvram = (struct sPAPRNVRAM *)dev;
1220 static void spapr_rtc_create(sPAPRMachineState *spapr)
1222 DeviceState *dev = qdev_create(NULL, TYPE_SPAPR_RTC);
1224 qdev_init_nofail(dev);
1225 spapr->rtc = dev;
1227 object_property_add_alias(qdev_get_machine(), "rtc-time",
1228 OBJECT(spapr->rtc), "date", NULL);
1231 /* Returns whether we want to use VGA or not */
1232 static bool spapr_vga_init(PCIBus *pci_bus, Error **errp)
1234 switch (vga_interface_type) {
1235 case VGA_NONE:
1236 return false;
1237 case VGA_DEVICE:
1238 return true;
1239 case VGA_STD:
1240 case VGA_VIRTIO:
1241 return pci_vga_init(pci_bus) != NULL;
1242 default:
1243 error_setg(errp,
1244 "Unsupported VGA mode, only -vga std or -vga virtio is supported");
1245 return false;
1249 static int spapr_post_load(void *opaque, int version_id)
1251 sPAPRMachineState *spapr = (sPAPRMachineState *)opaque;
1252 int err = 0;
1254 /* In earlier versions, there was no separate qdev for the PAPR
1255 * RTC, so the RTC offset was stored directly in sPAPREnvironment.
1256 * So when migrating from those versions, poke the incoming offset
1257 * value into the RTC device */
1258 if (version_id < 3) {
1259 err = spapr_rtc_import_offset(spapr->rtc, spapr->rtc_offset);
1262 return err;
1265 static bool version_before_3(void *opaque, int version_id)
1267 return version_id < 3;
1270 static bool spapr_ov5_cas_needed(void *opaque)
1272 sPAPRMachineState *spapr = opaque;
1273 sPAPROptionVector *ov5_mask = spapr_ovec_new();
1274 sPAPROptionVector *ov5_legacy = spapr_ovec_new();
1275 sPAPROptionVector *ov5_removed = spapr_ovec_new();
1276 bool cas_needed;
1278 /* Prior to the introduction of sPAPROptionVector, we had two option
1279 * vectors we dealt with: OV5_FORM1_AFFINITY, and OV5_DRCONF_MEMORY.
1280 * Both of these options encode machine topology into the device-tree
1281 * in such a way that the now-booted OS should still be able to interact
1282 * appropriately with QEMU regardless of what options were actually
1283 * negotiatied on the source side.
1285 * As such, we can avoid migrating the CAS-negotiated options if these
1286 * are the only options available on the current machine/platform.
1287 * Since these are the only options available for pseries-2.7 and
1288 * earlier, this allows us to maintain old->new/new->old migration
1289 * compatibility.
1291 * For QEMU 2.8+, there are additional CAS-negotiatable options available
1292 * via default pseries-2.8 machines and explicit command-line parameters.
1293 * Some of these options, like OV5_HP_EVT, *do* require QEMU to be aware
1294 * of the actual CAS-negotiated values to continue working properly. For
1295 * example, availability of memory unplug depends on knowing whether
1296 * OV5_HP_EVT was negotiated via CAS.
1298 * Thus, for any cases where the set of available CAS-negotiatable
1299 * options extends beyond OV5_FORM1_AFFINITY and OV5_DRCONF_MEMORY, we
1300 * include the CAS-negotiated options in the migration stream.
1302 spapr_ovec_set(ov5_mask, OV5_FORM1_AFFINITY);
1303 spapr_ovec_set(ov5_mask, OV5_DRCONF_MEMORY);
1305 /* spapr_ovec_diff returns true if bits were removed. we avoid using
1306 * the mask itself since in the future it's possible "legacy" bits may be
1307 * removed via machine options, which could generate a false positive
1308 * that breaks migration.
1310 spapr_ovec_intersect(ov5_legacy, spapr->ov5, ov5_mask);
1311 cas_needed = spapr_ovec_diff(ov5_removed, spapr->ov5, ov5_legacy);
1313 spapr_ovec_cleanup(ov5_mask);
1314 spapr_ovec_cleanup(ov5_legacy);
1315 spapr_ovec_cleanup(ov5_removed);
1317 return cas_needed;
1320 static const VMStateDescription vmstate_spapr_ov5_cas = {
1321 .name = "spapr_option_vector_ov5_cas",
1322 .version_id = 1,
1323 .minimum_version_id = 1,
1324 .needed = spapr_ov5_cas_needed,
1325 .fields = (VMStateField[]) {
1326 VMSTATE_STRUCT_POINTER_V(ov5_cas, sPAPRMachineState, 1,
1327 vmstate_spapr_ovec, sPAPROptionVector),
1328 VMSTATE_END_OF_LIST()
1332 static const VMStateDescription vmstate_spapr = {
1333 .name = "spapr",
1334 .version_id = 3,
1335 .minimum_version_id = 1,
1336 .post_load = spapr_post_load,
1337 .fields = (VMStateField[]) {
1338 /* used to be @next_irq */
1339 VMSTATE_UNUSED_BUFFER(version_before_3, 0, 4),
1341 /* RTC offset */
1342 VMSTATE_UINT64_TEST(rtc_offset, sPAPRMachineState, version_before_3),
1344 VMSTATE_PPC_TIMEBASE_V(tb, sPAPRMachineState, 2),
1345 VMSTATE_END_OF_LIST()
1347 .subsections = (const VMStateDescription*[]) {
1348 &vmstate_spapr_ov5_cas,
1349 NULL
1353 static int htab_save_setup(QEMUFile *f, void *opaque)
1355 sPAPRMachineState *spapr = opaque;
1357 /* "Iteration" header */
1358 qemu_put_be32(f, spapr->htab_shift);
1360 if (spapr->htab) {
1361 spapr->htab_save_index = 0;
1362 spapr->htab_first_pass = true;
1363 } else {
1364 assert(kvm_enabled());
1368 return 0;
1371 static void htab_save_first_pass(QEMUFile *f, sPAPRMachineState *spapr,
1372 int64_t max_ns)
1374 bool has_timeout = max_ns != -1;
1375 int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
1376 int index = spapr->htab_save_index;
1377 int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
1379 assert(spapr->htab_first_pass);
1381 do {
1382 int chunkstart;
1384 /* Consume invalid HPTEs */
1385 while ((index < htabslots)
1386 && !HPTE_VALID(HPTE(spapr->htab, index))) {
1387 index++;
1388 CLEAN_HPTE(HPTE(spapr->htab, index));
1391 /* Consume valid HPTEs */
1392 chunkstart = index;
1393 while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
1394 && HPTE_VALID(HPTE(spapr->htab, index))) {
1395 index++;
1396 CLEAN_HPTE(HPTE(spapr->htab, index));
1399 if (index > chunkstart) {
1400 int n_valid = index - chunkstart;
1402 qemu_put_be32(f, chunkstart);
1403 qemu_put_be16(f, n_valid);
1404 qemu_put_be16(f, 0);
1405 qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
1406 HASH_PTE_SIZE_64 * n_valid);
1408 if (has_timeout &&
1409 (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
1410 break;
1413 } while ((index < htabslots) && !qemu_file_rate_limit(f));
1415 if (index >= htabslots) {
1416 assert(index == htabslots);
1417 index = 0;
1418 spapr->htab_first_pass = false;
1420 spapr->htab_save_index = index;
1423 static int htab_save_later_pass(QEMUFile *f, sPAPRMachineState *spapr,
1424 int64_t max_ns)
1426 bool final = max_ns < 0;
1427 int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
1428 int examined = 0, sent = 0;
1429 int index = spapr->htab_save_index;
1430 int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
1432 assert(!spapr->htab_first_pass);
1434 do {
1435 int chunkstart, invalidstart;
1437 /* Consume non-dirty HPTEs */
1438 while ((index < htabslots)
1439 && !HPTE_DIRTY(HPTE(spapr->htab, index))) {
1440 index++;
1441 examined++;
1444 chunkstart = index;
1445 /* Consume valid dirty HPTEs */
1446 while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
1447 && HPTE_DIRTY(HPTE(spapr->htab, index))
1448 && HPTE_VALID(HPTE(spapr->htab, index))) {
1449 CLEAN_HPTE(HPTE(spapr->htab, index));
1450 index++;
1451 examined++;
1454 invalidstart = index;
1455 /* Consume invalid dirty HPTEs */
1456 while ((index < htabslots) && (index - invalidstart < USHRT_MAX)
1457 && HPTE_DIRTY(HPTE(spapr->htab, index))
1458 && !HPTE_VALID(HPTE(spapr->htab, index))) {
1459 CLEAN_HPTE(HPTE(spapr->htab, index));
1460 index++;
1461 examined++;
1464 if (index > chunkstart) {
1465 int n_valid = invalidstart - chunkstart;
1466 int n_invalid = index - invalidstart;
1468 qemu_put_be32(f, chunkstart);
1469 qemu_put_be16(f, n_valid);
1470 qemu_put_be16(f, n_invalid);
1471 qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
1472 HASH_PTE_SIZE_64 * n_valid);
1473 sent += index - chunkstart;
1475 if (!final && (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
1476 break;
1480 if (examined >= htabslots) {
1481 break;
1484 if (index >= htabslots) {
1485 assert(index == htabslots);
1486 index = 0;
1488 } while ((examined < htabslots) && (!qemu_file_rate_limit(f) || final));
1490 if (index >= htabslots) {
1491 assert(index == htabslots);
1492 index = 0;
1495 spapr->htab_save_index = index;
1497 return (examined >= htabslots) && (sent == 0) ? 1 : 0;
1500 #define MAX_ITERATION_NS 5000000 /* 5 ms */
1501 #define MAX_KVM_BUF_SIZE 2048
1503 static int htab_save_iterate(QEMUFile *f, void *opaque)
1505 sPAPRMachineState *spapr = opaque;
1506 int fd;
1507 int rc = 0;
1509 /* Iteration header */
1510 qemu_put_be32(f, 0);
1512 if (!spapr->htab) {
1513 assert(kvm_enabled());
1515 fd = get_htab_fd(spapr);
1516 if (fd < 0) {
1517 return fd;
1520 rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, MAX_ITERATION_NS);
1521 if (rc < 0) {
1522 return rc;
1524 } else if (spapr->htab_first_pass) {
1525 htab_save_first_pass(f, spapr, MAX_ITERATION_NS);
1526 } else {
1527 rc = htab_save_later_pass(f, spapr, MAX_ITERATION_NS);
1530 /* End marker */
1531 qemu_put_be32(f, 0);
1532 qemu_put_be16(f, 0);
1533 qemu_put_be16(f, 0);
1535 return rc;
1538 static int htab_save_complete(QEMUFile *f, void *opaque)
1540 sPAPRMachineState *spapr = opaque;
1541 int fd;
1543 /* Iteration header */
1544 qemu_put_be32(f, 0);
1546 if (!spapr->htab) {
1547 int rc;
1549 assert(kvm_enabled());
1551 fd = get_htab_fd(spapr);
1552 if (fd < 0) {
1553 return fd;
1556 rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, -1);
1557 if (rc < 0) {
1558 return rc;
1560 } else {
1561 if (spapr->htab_first_pass) {
1562 htab_save_first_pass(f, spapr, -1);
1564 htab_save_later_pass(f, spapr, -1);
1567 /* End marker */
1568 qemu_put_be32(f, 0);
1569 qemu_put_be16(f, 0);
1570 qemu_put_be16(f, 0);
1572 return 0;
1575 static int htab_load(QEMUFile *f, void *opaque, int version_id)
1577 sPAPRMachineState *spapr = opaque;
1578 uint32_t section_hdr;
1579 int fd = -1;
1581 if (version_id < 1 || version_id > 1) {
1582 error_report("htab_load() bad version");
1583 return -EINVAL;
1586 section_hdr = qemu_get_be32(f);
1588 if (section_hdr) {
1589 Error *local_err = NULL;
1591 /* First section gives the htab size */
1592 spapr_reallocate_hpt(spapr, section_hdr, &local_err);
1593 if (local_err) {
1594 error_report_err(local_err);
1595 return -EINVAL;
1597 return 0;
1600 if (!spapr->htab) {
1601 assert(kvm_enabled());
1603 fd = kvmppc_get_htab_fd(true);
1604 if (fd < 0) {
1605 error_report("Unable to open fd to restore KVM hash table: %s",
1606 strerror(errno));
1610 while (true) {
1611 uint32_t index;
1612 uint16_t n_valid, n_invalid;
1614 index = qemu_get_be32(f);
1615 n_valid = qemu_get_be16(f);
1616 n_invalid = qemu_get_be16(f);
1618 if ((index == 0) && (n_valid == 0) && (n_invalid == 0)) {
1619 /* End of Stream */
1620 break;
1623 if ((index + n_valid + n_invalid) >
1624 (HTAB_SIZE(spapr) / HASH_PTE_SIZE_64)) {
1625 /* Bad index in stream */
1626 error_report(
1627 "htab_load() bad index %d (%hd+%hd entries) in htab stream (htab_shift=%d)",
1628 index, n_valid, n_invalid, spapr->htab_shift);
1629 return -EINVAL;
1632 if (spapr->htab) {
1633 if (n_valid) {
1634 qemu_get_buffer(f, HPTE(spapr->htab, index),
1635 HASH_PTE_SIZE_64 * n_valid);
1637 if (n_invalid) {
1638 memset(HPTE(spapr->htab, index + n_valid), 0,
1639 HASH_PTE_SIZE_64 * n_invalid);
1641 } else {
1642 int rc;
1644 assert(fd >= 0);
1646 rc = kvmppc_load_htab_chunk(f, fd, index, n_valid, n_invalid);
1647 if (rc < 0) {
1648 return rc;
1653 if (!spapr->htab) {
1654 assert(fd >= 0);
1655 close(fd);
1658 return 0;
1661 static void htab_cleanup(void *opaque)
1663 sPAPRMachineState *spapr = opaque;
1665 close_htab_fd(spapr);
1668 static SaveVMHandlers savevm_htab_handlers = {
1669 .save_live_setup = htab_save_setup,
1670 .save_live_iterate = htab_save_iterate,
1671 .save_live_complete_precopy = htab_save_complete,
1672 .cleanup = htab_cleanup,
1673 .load_state = htab_load,
1676 static void spapr_boot_set(void *opaque, const char *boot_device,
1677 Error **errp)
1679 MachineState *machine = MACHINE(qdev_get_machine());
1680 machine->boot_order = g_strdup(boot_device);
1684 * Reset routine for LMB DR devices.
1686 * Unlike PCI DR devices, LMB DR devices explicitly register this reset
1687 * routine. Reset for PCI DR devices will be handled by PHB reset routine
1688 * when it walks all its children devices. LMB devices reset occurs
1689 * as part of spapr_ppc_reset().
1691 static void spapr_drc_reset(void *opaque)
1693 sPAPRDRConnector *drc = opaque;
1694 DeviceState *d = DEVICE(drc);
1696 if (d) {
1697 device_reset(d);
1701 static void spapr_create_lmb_dr_connectors(sPAPRMachineState *spapr)
1703 MachineState *machine = MACHINE(spapr);
1704 uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
1705 uint32_t nr_lmbs = (machine->maxram_size - machine->ram_size)/lmb_size;
1706 int i;
1708 for (i = 0; i < nr_lmbs; i++) {
1709 sPAPRDRConnector *drc;
1710 uint64_t addr;
1712 addr = i * lmb_size + spapr->hotplug_memory.base;
1713 drc = spapr_dr_connector_new(OBJECT(spapr), SPAPR_DR_CONNECTOR_TYPE_LMB,
1714 addr/lmb_size);
1715 qemu_register_reset(spapr_drc_reset, drc);
1720 * If RAM size, maxmem size and individual node mem sizes aren't aligned
1721 * to SPAPR_MEMORY_BLOCK_SIZE(256MB), then refuse to start the guest
1722 * since we can't support such unaligned sizes with DRCONF_MEMORY.
1724 static void spapr_validate_node_memory(MachineState *machine, Error **errp)
1726 int i;
1728 if (machine->ram_size % SPAPR_MEMORY_BLOCK_SIZE) {
1729 error_setg(errp, "Memory size 0x" RAM_ADDR_FMT
1730 " is not aligned to %llu MiB",
1731 machine->ram_size,
1732 SPAPR_MEMORY_BLOCK_SIZE / M_BYTE);
1733 return;
1736 if (machine->maxram_size % SPAPR_MEMORY_BLOCK_SIZE) {
1737 error_setg(errp, "Maximum memory size 0x" RAM_ADDR_FMT
1738 " is not aligned to %llu MiB",
1739 machine->ram_size,
1740 SPAPR_MEMORY_BLOCK_SIZE / M_BYTE);
1741 return;
1744 for (i = 0; i < nb_numa_nodes; i++) {
1745 if (numa_info[i].node_mem % SPAPR_MEMORY_BLOCK_SIZE) {
1746 error_setg(errp,
1747 "Node %d memory size 0x%" PRIx64
1748 " is not aligned to %llu MiB",
1749 i, numa_info[i].node_mem,
1750 SPAPR_MEMORY_BLOCK_SIZE / M_BYTE);
1751 return;
1756 /* pSeries LPAR / sPAPR hardware init */
1757 static void ppc_spapr_init(MachineState *machine)
1759 sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
1760 MachineClass *mc = MACHINE_GET_CLASS(machine);
1761 sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1762 const char *kernel_filename = machine->kernel_filename;
1763 const char *initrd_filename = machine->initrd_filename;
1764 PCIHostState *phb;
1765 int i;
1766 MemoryRegion *sysmem = get_system_memory();
1767 MemoryRegion *ram = g_new(MemoryRegion, 1);
1768 MemoryRegion *rma_region;
1769 void *rma = NULL;
1770 hwaddr rma_alloc_size;
1771 hwaddr node0_size = spapr_node0_size();
1772 long load_limit, fw_size;
1773 char *filename;
1774 int smt = kvmppc_smt_threads();
1775 int spapr_cores = smp_cpus / smp_threads;
1776 int spapr_max_cores = max_cpus / smp_threads;
1778 if (mc->query_hotpluggable_cpus) {
1779 if (smp_cpus % smp_threads) {
1780 error_report("smp_cpus (%u) must be multiple of threads (%u)",
1781 smp_cpus, smp_threads);
1782 exit(1);
1784 if (max_cpus % smp_threads) {
1785 error_report("max_cpus (%u) must be multiple of threads (%u)",
1786 max_cpus, smp_threads);
1787 exit(1);
1791 msi_nonbroken = true;
1793 QLIST_INIT(&spapr->phbs);
1795 cpu_ppc_hypercall = emulate_spapr_hypercall;
1797 /* Allocate RMA if necessary */
1798 rma_alloc_size = kvmppc_alloc_rma(&rma);
1800 if (rma_alloc_size == -1) {
1801 error_report("Unable to create RMA");
1802 exit(1);
1805 if (rma_alloc_size && (rma_alloc_size < node0_size)) {
1806 spapr->rma_size = rma_alloc_size;
1807 } else {
1808 spapr->rma_size = node0_size;
1810 /* With KVM, we don't actually know whether KVM supports an
1811 * unbounded RMA (PR KVM) or is limited by the hash table size
1812 * (HV KVM using VRMA), so we always assume the latter
1814 * In that case, we also limit the initial allocations for RTAS
1815 * etc... to 256M since we have no way to know what the VRMA size
1816 * is going to be as it depends on the size of the hash table
1817 * isn't determined yet.
1819 if (kvm_enabled()) {
1820 spapr->vrma_adjust = 1;
1821 spapr->rma_size = MIN(spapr->rma_size, 0x10000000);
1824 /* Actually we don't support unbounded RMA anymore since we
1825 * added proper emulation of HV mode. The max we can get is
1826 * 16G which also happens to be what we configure for PAPR
1827 * mode so make sure we don't do anything bigger than that
1829 spapr->rma_size = MIN(spapr->rma_size, 0x400000000ull);
1832 if (spapr->rma_size > node0_size) {
1833 error_report("Numa node 0 has to span the RMA (%#08"HWADDR_PRIx")",
1834 spapr->rma_size);
1835 exit(1);
1838 /* Setup a load limit for the ramdisk leaving room for SLOF and FDT */
1839 load_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR) - FW_OVERHEAD;
1841 /* Set up Interrupt Controller before we create the VCPUs */
1842 spapr->xics = xics_system_init(machine,
1843 DIV_ROUND_UP(max_cpus * smt, smp_threads),
1844 XICS_IRQS_SPAPR, &error_fatal);
1846 /* Set up containers for ibm,client-set-architecture negotiated options */
1847 spapr->ov5 = spapr_ovec_new();
1848 spapr->ov5_cas = spapr_ovec_new();
1850 if (smc->dr_lmb_enabled) {
1851 spapr_ovec_set(spapr->ov5, OV5_DRCONF_MEMORY);
1852 spapr_validate_node_memory(machine, &error_fatal);
1855 spapr_ovec_set(spapr->ov5, OV5_FORM1_AFFINITY);
1857 /* advertise support for dedicated HP event source to guests */
1858 if (spapr->use_hotplug_event_source) {
1859 spapr_ovec_set(spapr->ov5, OV5_HP_EVT);
1862 /* init CPUs */
1863 if (machine->cpu_model == NULL) {
1864 machine->cpu_model = kvm_enabled() ? "host" : smc->tcg_default_cpu;
1867 ppc_cpu_parse_features(machine->cpu_model);
1869 if (mc->query_hotpluggable_cpus) {
1870 char *type = spapr_get_cpu_core_type(machine->cpu_model);
1872 if (type == NULL) {
1873 error_report("Unable to find sPAPR CPU Core definition");
1874 exit(1);
1877 spapr->cores = g_new0(Object *, spapr_max_cores);
1878 for (i = 0; i < spapr_max_cores; i++) {
1879 int core_id = i * smp_threads;
1880 sPAPRDRConnector *drc =
1881 spapr_dr_connector_new(OBJECT(spapr),
1882 SPAPR_DR_CONNECTOR_TYPE_CPU,
1883 (core_id / smp_threads) * smt);
1885 qemu_register_reset(spapr_drc_reset, drc);
1887 if (i < spapr_cores) {
1888 Object *core = object_new(type);
1889 object_property_set_int(core, smp_threads, "nr-threads",
1890 &error_fatal);
1891 object_property_set_int(core, core_id, CPU_CORE_PROP_CORE_ID,
1892 &error_fatal);
1893 object_property_set_bool(core, true, "realized", &error_fatal);
1896 g_free(type);
1897 } else {
1898 for (i = 0; i < smp_cpus; i++) {
1899 PowerPCCPU *cpu = cpu_ppc_init(machine->cpu_model);
1900 if (cpu == NULL) {
1901 error_report("Unable to find PowerPC CPU definition");
1902 exit(1);
1904 spapr_cpu_init(spapr, cpu, &error_fatal);
1908 if (kvm_enabled()) {
1909 /* Enable H_LOGICAL_CI_* so SLOF can talk to in-kernel devices */
1910 kvmppc_enable_logical_ci_hcalls();
1911 kvmppc_enable_set_mode_hcall();
1913 /* H_CLEAR_MOD/_REF are mandatory in PAPR, but off by default */
1914 kvmppc_enable_clear_ref_mod_hcalls();
1917 /* allocate RAM */
1918 memory_region_allocate_system_memory(ram, NULL, "ppc_spapr.ram",
1919 machine->ram_size);
1920 memory_region_add_subregion(sysmem, 0, ram);
1922 if (rma_alloc_size && rma) {
1923 rma_region = g_new(MemoryRegion, 1);
1924 memory_region_init_ram_ptr(rma_region, NULL, "ppc_spapr.rma",
1925 rma_alloc_size, rma);
1926 vmstate_register_ram_global(rma_region);
1927 memory_region_add_subregion(sysmem, 0, rma_region);
1930 /* initialize hotplug memory address space */
1931 if (machine->ram_size < machine->maxram_size) {
1932 ram_addr_t hotplug_mem_size = machine->maxram_size - machine->ram_size;
1934 * Limit the number of hotpluggable memory slots to half the number
1935 * slots that KVM supports, leaving the other half for PCI and other
1936 * devices. However ensure that number of slots doesn't drop below 32.
1938 int max_memslots = kvm_enabled() ? kvm_get_max_memslots() / 2 :
1939 SPAPR_MAX_RAM_SLOTS;
1941 if (max_memslots < SPAPR_MAX_RAM_SLOTS) {
1942 max_memslots = SPAPR_MAX_RAM_SLOTS;
1944 if (machine->ram_slots > max_memslots) {
1945 error_report("Specified number of memory slots %"
1946 PRIu64" exceeds max supported %d",
1947 machine->ram_slots, max_memslots);
1948 exit(1);
1951 spapr->hotplug_memory.base = ROUND_UP(machine->ram_size,
1952 SPAPR_HOTPLUG_MEM_ALIGN);
1953 memory_region_init(&spapr->hotplug_memory.mr, OBJECT(spapr),
1954 "hotplug-memory", hotplug_mem_size);
1955 memory_region_add_subregion(sysmem, spapr->hotplug_memory.base,
1956 &spapr->hotplug_memory.mr);
1959 if (smc->dr_lmb_enabled) {
1960 spapr_create_lmb_dr_connectors(spapr);
1963 filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, "spapr-rtas.bin");
1964 if (!filename) {
1965 error_report("Could not find LPAR rtas '%s'", "spapr-rtas.bin");
1966 exit(1);
1968 spapr->rtas_size = get_image_size(filename);
1969 if (spapr->rtas_size < 0) {
1970 error_report("Could not get size of LPAR rtas '%s'", filename);
1971 exit(1);
1973 spapr->rtas_blob = g_malloc(spapr->rtas_size);
1974 if (load_image_size(filename, spapr->rtas_blob, spapr->rtas_size) < 0) {
1975 error_report("Could not load LPAR rtas '%s'", filename);
1976 exit(1);
1978 if (spapr->rtas_size > RTAS_MAX_SIZE) {
1979 error_report("RTAS too big ! 0x%zx bytes (max is 0x%x)",
1980 (size_t)spapr->rtas_size, RTAS_MAX_SIZE);
1981 exit(1);
1983 g_free(filename);
1985 /* Set up RTAS event infrastructure */
1986 spapr_events_init(spapr);
1988 /* Set up the RTC RTAS interfaces */
1989 spapr_rtc_create(spapr);
1991 /* Set up VIO bus */
1992 spapr->vio_bus = spapr_vio_bus_init();
1994 for (i = 0; i < MAX_SERIAL_PORTS; i++) {
1995 if (serial_hds[i]) {
1996 spapr_vty_create(spapr->vio_bus, serial_hds[i]);
2000 /* We always have at least the nvram device on VIO */
2001 spapr_create_nvram(spapr);
2003 /* Set up PCI */
2004 spapr_pci_rtas_init();
2006 phb = spapr_create_phb(spapr, 0);
2008 for (i = 0; i < nb_nics; i++) {
2009 NICInfo *nd = &nd_table[i];
2011 if (!nd->model) {
2012 nd->model = g_strdup("ibmveth");
2015 if (strcmp(nd->model, "ibmveth") == 0) {
2016 spapr_vlan_create(spapr->vio_bus, nd);
2017 } else {
2018 pci_nic_init_nofail(&nd_table[i], phb->bus, nd->model, NULL);
2022 for (i = 0; i <= drive_get_max_bus(IF_SCSI); i++) {
2023 spapr_vscsi_create(spapr->vio_bus);
2026 /* Graphics */
2027 if (spapr_vga_init(phb->bus, &error_fatal)) {
2028 spapr->has_graphics = true;
2029 machine->usb |= defaults_enabled() && !machine->usb_disabled;
2032 if (machine->usb) {
2033 if (smc->use_ohci_by_default) {
2034 pci_create_simple(phb->bus, -1, "pci-ohci");
2035 } else {
2036 pci_create_simple(phb->bus, -1, "nec-usb-xhci");
2039 if (spapr->has_graphics) {
2040 USBBus *usb_bus = usb_bus_find(-1);
2042 usb_create_simple(usb_bus, "usb-kbd");
2043 usb_create_simple(usb_bus, "usb-mouse");
2047 if (spapr->rma_size < (MIN_RMA_SLOF << 20)) {
2048 error_report(
2049 "pSeries SLOF firmware requires >= %ldM guest RMA (Real Mode Area memory)",
2050 MIN_RMA_SLOF);
2051 exit(1);
2054 if (kernel_filename) {
2055 uint64_t lowaddr = 0;
2057 spapr->kernel_size = load_elf(kernel_filename, translate_kernel_address,
2058 NULL, NULL, &lowaddr, NULL, 1,
2059 PPC_ELF_MACHINE, 0, 0);
2060 if (spapr->kernel_size == ELF_LOAD_WRONG_ENDIAN) {
2061 spapr->kernel_size = load_elf(kernel_filename,
2062 translate_kernel_address, NULL, NULL,
2063 &lowaddr, NULL, 0, PPC_ELF_MACHINE,
2064 0, 0);
2065 spapr->kernel_le = spapr->kernel_size > 0;
2067 if (spapr->kernel_size < 0) {
2068 error_report("error loading %s: %s", kernel_filename,
2069 load_elf_strerror(spapr->kernel_size));
2070 exit(1);
2073 /* load initrd */
2074 if (initrd_filename) {
2075 /* Try to locate the initrd in the gap between the kernel
2076 * and the firmware. Add a bit of space just in case
2078 spapr->initrd_base = (KERNEL_LOAD_ADDR + spapr->kernel_size
2079 + 0x1ffff) & ~0xffff;
2080 spapr->initrd_size = load_image_targphys(initrd_filename,
2081 spapr->initrd_base,
2082 load_limit
2083 - spapr->initrd_base);
2084 if (spapr->initrd_size < 0) {
2085 error_report("could not load initial ram disk '%s'",
2086 initrd_filename);
2087 exit(1);
2092 if (bios_name == NULL) {
2093 bios_name = FW_FILE_NAME;
2095 filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
2096 if (!filename) {
2097 error_report("Could not find LPAR firmware '%s'", bios_name);
2098 exit(1);
2100 fw_size = load_image_targphys(filename, 0, FW_MAX_SIZE);
2101 if (fw_size <= 0) {
2102 error_report("Could not load LPAR firmware '%s'", filename);
2103 exit(1);
2105 g_free(filename);
2107 /* FIXME: Should register things through the MachineState's qdev
2108 * interface, this is a legacy from the sPAPREnvironment structure
2109 * which predated MachineState but had a similar function */
2110 vmstate_register(NULL, 0, &vmstate_spapr, spapr);
2111 register_savevm_live(NULL, "spapr/htab", -1, 1,
2112 &savevm_htab_handlers, spapr);
2114 /* used by RTAS */
2115 QTAILQ_INIT(&spapr->ccs_list);
2116 qemu_register_reset(spapr_ccs_reset_hook, spapr);
2118 qemu_register_boot_set(spapr_boot_set, spapr);
2121 static int spapr_kvm_type(const char *vm_type)
2123 if (!vm_type) {
2124 return 0;
2127 if (!strcmp(vm_type, "HV")) {
2128 return 1;
2131 if (!strcmp(vm_type, "PR")) {
2132 return 2;
2135 error_report("Unknown kvm-type specified '%s'", vm_type);
2136 exit(1);
2140 * Implementation of an interface to adjust firmware path
2141 * for the bootindex property handling.
2143 static char *spapr_get_fw_dev_path(FWPathProvider *p, BusState *bus,
2144 DeviceState *dev)
2146 #define CAST(type, obj, name) \
2147 ((type *)object_dynamic_cast(OBJECT(obj), (name)))
2148 SCSIDevice *d = CAST(SCSIDevice, dev, TYPE_SCSI_DEVICE);
2149 sPAPRPHBState *phb = CAST(sPAPRPHBState, dev, TYPE_SPAPR_PCI_HOST_BRIDGE);
2151 if (d) {
2152 void *spapr = CAST(void, bus->parent, "spapr-vscsi");
2153 VirtIOSCSI *virtio = CAST(VirtIOSCSI, bus->parent, TYPE_VIRTIO_SCSI);
2154 USBDevice *usb = CAST(USBDevice, bus->parent, TYPE_USB_DEVICE);
2156 if (spapr) {
2158 * Replace "channel@0/disk@0,0" with "disk@8000000000000000":
2159 * We use SRP luns of the form 8000 | (bus << 8) | (id << 5) | lun
2160 * in the top 16 bits of the 64-bit LUN
2162 unsigned id = 0x8000 | (d->id << 8) | d->lun;
2163 return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
2164 (uint64_t)id << 48);
2165 } else if (virtio) {
2167 * We use SRP luns of the form 01000000 | (target << 8) | lun
2168 * in the top 32 bits of the 64-bit LUN
2169 * Note: the quote above is from SLOF and it is wrong,
2170 * the actual binding is:
2171 * swap 0100 or 10 << or 20 << ( target lun-id -- srplun )
2173 unsigned id = 0x1000000 | (d->id << 16) | d->lun;
2174 return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
2175 (uint64_t)id << 32);
2176 } else if (usb) {
2178 * We use SRP luns of the form 01000000 | (usb-port << 16) | lun
2179 * in the top 32 bits of the 64-bit LUN
2181 unsigned usb_port = atoi(usb->port->path);
2182 unsigned id = 0x1000000 | (usb_port << 16) | d->lun;
2183 return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
2184 (uint64_t)id << 32);
2188 if (phb) {
2189 /* Replace "pci" with "pci@800000020000000" */
2190 return g_strdup_printf("pci@%"PRIX64, phb->buid);
2193 return NULL;
2196 static char *spapr_get_kvm_type(Object *obj, Error **errp)
2198 sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2200 return g_strdup(spapr->kvm_type);
2203 static void spapr_set_kvm_type(Object *obj, const char *value, Error **errp)
2205 sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2207 g_free(spapr->kvm_type);
2208 spapr->kvm_type = g_strdup(value);
2211 static bool spapr_get_modern_hotplug_events(Object *obj, Error **errp)
2213 sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2215 return spapr->use_hotplug_event_source;
2218 static void spapr_set_modern_hotplug_events(Object *obj, bool value,
2219 Error **errp)
2221 sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2223 spapr->use_hotplug_event_source = value;
2226 static void spapr_machine_initfn(Object *obj)
2228 sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2230 spapr->htab_fd = -1;
2231 spapr->use_hotplug_event_source = true;
2232 object_property_add_str(obj, "kvm-type",
2233 spapr_get_kvm_type, spapr_set_kvm_type, NULL);
2234 object_property_set_description(obj, "kvm-type",
2235 "Specifies the KVM virtualization mode (HV, PR)",
2236 NULL);
2237 object_property_add_bool(obj, "modern-hotplug-events",
2238 spapr_get_modern_hotplug_events,
2239 spapr_set_modern_hotplug_events,
2240 NULL);
2241 object_property_set_description(obj, "modern-hotplug-events",
2242 "Use dedicated hotplug event mechanism in"
2243 " place of standard EPOW events when possible"
2244 " (required for memory hot-unplug support)",
2245 NULL);
2248 static void spapr_machine_finalizefn(Object *obj)
2250 sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2252 g_free(spapr->kvm_type);
2255 static void ppc_cpu_do_nmi_on_cpu(CPUState *cs, run_on_cpu_data arg)
2257 cpu_synchronize_state(cs);
2258 ppc_cpu_do_system_reset(cs);
2261 static void spapr_nmi(NMIState *n, int cpu_index, Error **errp)
2263 CPUState *cs;
2265 CPU_FOREACH(cs) {
2266 async_run_on_cpu(cs, ppc_cpu_do_nmi_on_cpu, RUN_ON_CPU_NULL);
2270 static void spapr_add_lmbs(DeviceState *dev, uint64_t addr_start, uint64_t size,
2271 uint32_t node, bool dedicated_hp_event_source,
2272 Error **errp)
2274 sPAPRDRConnector *drc;
2275 sPAPRDRConnectorClass *drck;
2276 uint32_t nr_lmbs = size/SPAPR_MEMORY_BLOCK_SIZE;
2277 int i, fdt_offset, fdt_size;
2278 void *fdt;
2279 uint64_t addr = addr_start;
2281 for (i = 0; i < nr_lmbs; i++) {
2282 drc = spapr_dr_connector_by_id(SPAPR_DR_CONNECTOR_TYPE_LMB,
2283 addr/SPAPR_MEMORY_BLOCK_SIZE);
2284 g_assert(drc);
2286 fdt = create_device_tree(&fdt_size);
2287 fdt_offset = spapr_populate_memory_node(fdt, node, addr,
2288 SPAPR_MEMORY_BLOCK_SIZE);
2290 drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
2291 drck->attach(drc, dev, fdt, fdt_offset, !dev->hotplugged, errp);
2292 addr += SPAPR_MEMORY_BLOCK_SIZE;
2293 if (!dev->hotplugged) {
2294 /* guests expect coldplugged LMBs to be pre-allocated */
2295 drck->set_allocation_state(drc, SPAPR_DR_ALLOCATION_STATE_USABLE);
2296 drck->set_isolation_state(drc, SPAPR_DR_ISOLATION_STATE_UNISOLATED);
2299 /* send hotplug notification to the
2300 * guest only in case of hotplugged memory
2302 if (dev->hotplugged) {
2303 if (dedicated_hp_event_source) {
2304 drc = spapr_dr_connector_by_id(SPAPR_DR_CONNECTOR_TYPE_LMB,
2305 addr_start / SPAPR_MEMORY_BLOCK_SIZE);
2306 drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
2307 spapr_hotplug_req_add_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
2308 nr_lmbs,
2309 drck->get_index(drc));
2310 } else {
2311 spapr_hotplug_req_add_by_count(SPAPR_DR_CONNECTOR_TYPE_LMB,
2312 nr_lmbs);
2317 static void spapr_memory_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
2318 uint32_t node, Error **errp)
2320 Error *local_err = NULL;
2321 sPAPRMachineState *ms = SPAPR_MACHINE(hotplug_dev);
2322 PCDIMMDevice *dimm = PC_DIMM(dev);
2323 PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
2324 MemoryRegion *mr = ddc->get_memory_region(dimm);
2325 uint64_t align = memory_region_get_alignment(mr);
2326 uint64_t size = memory_region_size(mr);
2327 uint64_t addr;
2329 if (size % SPAPR_MEMORY_BLOCK_SIZE) {
2330 error_setg(&local_err, "Hotplugged memory size must be a multiple of "
2331 "%lld MB", SPAPR_MEMORY_BLOCK_SIZE/M_BYTE);
2332 goto out;
2335 pc_dimm_memory_plug(dev, &ms->hotplug_memory, mr, align, &local_err);
2336 if (local_err) {
2337 goto out;
2340 addr = object_property_get_int(OBJECT(dimm), PC_DIMM_ADDR_PROP, &local_err);
2341 if (local_err) {
2342 pc_dimm_memory_unplug(dev, &ms->hotplug_memory, mr);
2343 goto out;
2346 spapr_add_lmbs(dev, addr, size, node,
2347 spapr_ovec_test(ms->ov5_cas, OV5_HP_EVT),
2348 &error_abort);
2350 out:
2351 error_propagate(errp, local_err);
2354 typedef struct sPAPRDIMMState {
2355 uint32_t nr_lmbs;
2356 } sPAPRDIMMState;
2358 static void spapr_lmb_release(DeviceState *dev, void *opaque)
2360 sPAPRDIMMState *ds = (sPAPRDIMMState *)opaque;
2361 HotplugHandler *hotplug_ctrl;
2363 if (--ds->nr_lmbs) {
2364 return;
2367 g_free(ds);
2370 * Now that all the LMBs have been removed by the guest, call the
2371 * pc-dimm unplug handler to cleanup up the pc-dimm device.
2373 hotplug_ctrl = qdev_get_hotplug_handler(dev);
2374 hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
2377 static void spapr_del_lmbs(DeviceState *dev, uint64_t addr_start, uint64_t size,
2378 Error **errp)
2380 sPAPRDRConnector *drc;
2381 sPAPRDRConnectorClass *drck;
2382 uint32_t nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
2383 int i;
2384 sPAPRDIMMState *ds = g_malloc0(sizeof(sPAPRDIMMState));
2385 uint64_t addr = addr_start;
2387 ds->nr_lmbs = nr_lmbs;
2388 for (i = 0; i < nr_lmbs; i++) {
2389 drc = spapr_dr_connector_by_id(SPAPR_DR_CONNECTOR_TYPE_LMB,
2390 addr / SPAPR_MEMORY_BLOCK_SIZE);
2391 g_assert(drc);
2393 drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
2394 drck->detach(drc, dev, spapr_lmb_release, ds, errp);
2395 addr += SPAPR_MEMORY_BLOCK_SIZE;
2398 drc = spapr_dr_connector_by_id(SPAPR_DR_CONNECTOR_TYPE_LMB,
2399 addr_start / SPAPR_MEMORY_BLOCK_SIZE);
2400 drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
2401 spapr_hotplug_req_remove_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
2402 nr_lmbs,
2403 drck->get_index(drc));
2406 static void spapr_memory_unplug(HotplugHandler *hotplug_dev, DeviceState *dev,
2407 Error **errp)
2409 sPAPRMachineState *ms = SPAPR_MACHINE(hotplug_dev);
2410 PCDIMMDevice *dimm = PC_DIMM(dev);
2411 PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
2412 MemoryRegion *mr = ddc->get_memory_region(dimm);
2414 pc_dimm_memory_unplug(dev, &ms->hotplug_memory, mr);
2415 object_unparent(OBJECT(dev));
2418 static void spapr_memory_unplug_request(HotplugHandler *hotplug_dev,
2419 DeviceState *dev, Error **errp)
2421 Error *local_err = NULL;
2422 PCDIMMDevice *dimm = PC_DIMM(dev);
2423 PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
2424 MemoryRegion *mr = ddc->get_memory_region(dimm);
2425 uint64_t size = memory_region_size(mr);
2426 uint64_t addr;
2428 addr = object_property_get_int(OBJECT(dimm), PC_DIMM_ADDR_PROP, &local_err);
2429 if (local_err) {
2430 goto out;
2433 spapr_del_lmbs(dev, addr, size, &error_abort);
2434 out:
2435 error_propagate(errp, local_err);
2438 void *spapr_populate_hotplug_cpu_dt(CPUState *cs, int *fdt_offset,
2439 sPAPRMachineState *spapr)
2441 PowerPCCPU *cpu = POWERPC_CPU(cs);
2442 DeviceClass *dc = DEVICE_GET_CLASS(cs);
2443 int id = ppc_get_vcpu_dt_id(cpu);
2444 void *fdt;
2445 int offset, fdt_size;
2446 char *nodename;
2448 fdt = create_device_tree(&fdt_size);
2449 nodename = g_strdup_printf("%s@%x", dc->fw_name, id);
2450 offset = fdt_add_subnode(fdt, 0, nodename);
2452 spapr_populate_cpu_dt(cs, fdt, offset, spapr);
2453 g_free(nodename);
2455 *fdt_offset = offset;
2456 return fdt;
2459 static void spapr_machine_device_plug(HotplugHandler *hotplug_dev,
2460 DeviceState *dev, Error **errp)
2462 sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(qdev_get_machine());
2464 if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
2465 int node;
2467 if (!smc->dr_lmb_enabled) {
2468 error_setg(errp, "Memory hotplug not supported for this machine");
2469 return;
2471 node = object_property_get_int(OBJECT(dev), PC_DIMM_NODE_PROP, errp);
2472 if (*errp) {
2473 return;
2475 if (node < 0 || node >= MAX_NODES) {
2476 error_setg(errp, "Invaild node %d", node);
2477 return;
2481 * Currently PowerPC kernel doesn't allow hot-adding memory to
2482 * memory-less node, but instead will silently add the memory
2483 * to the first node that has some memory. This causes two
2484 * unexpected behaviours for the user.
2486 * - Memory gets hotplugged to a different node than what the user
2487 * specified.
2488 * - Since pc-dimm subsystem in QEMU still thinks that memory belongs
2489 * to memory-less node, a reboot will set things accordingly
2490 * and the previously hotplugged memory now ends in the right node.
2491 * This appears as if some memory moved from one node to another.
2493 * So until kernel starts supporting memory hotplug to memory-less
2494 * nodes, just prevent such attempts upfront in QEMU.
2496 if (nb_numa_nodes && !numa_info[node].node_mem) {
2497 error_setg(errp, "Can't hotplug memory to memory-less node %d",
2498 node);
2499 return;
2502 spapr_memory_plug(hotplug_dev, dev, node, errp);
2503 } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
2504 spapr_core_plug(hotplug_dev, dev, errp);
2508 static void spapr_machine_device_unplug(HotplugHandler *hotplug_dev,
2509 DeviceState *dev, Error **errp)
2511 sPAPRMachineState *sms = SPAPR_MACHINE(qdev_get_machine());
2512 MachineClass *mc = MACHINE_GET_CLASS(qdev_get_machine());
2514 if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
2515 if (spapr_ovec_test(sms->ov5_cas, OV5_HP_EVT)) {
2516 spapr_memory_unplug(hotplug_dev, dev, errp);
2517 } else {
2518 error_setg(errp, "Memory hot unplug not supported for this guest");
2520 } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
2521 if (!mc->query_hotpluggable_cpus) {
2522 error_setg(errp, "CPU hot unplug not supported on this machine");
2523 return;
2525 spapr_core_unplug(hotplug_dev, dev, errp);
2529 static void spapr_machine_device_unplug_request(HotplugHandler *hotplug_dev,
2530 DeviceState *dev, Error **errp)
2532 sPAPRMachineState *sms = SPAPR_MACHINE(qdev_get_machine());
2533 MachineClass *mc = MACHINE_GET_CLASS(qdev_get_machine());
2535 if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
2536 if (spapr_ovec_test(sms->ov5_cas, OV5_HP_EVT)) {
2537 spapr_memory_unplug_request(hotplug_dev, dev, errp);
2538 } else {
2539 /* NOTE: this means there is a window after guest reset, prior to
2540 * CAS negotiation, where unplug requests will fail due to the
2541 * capability not being detected yet. This is a bit different than
2542 * the case with PCI unplug, where the events will be queued and
2543 * eventually handled by the guest after boot
2545 error_setg(errp, "Memory hot unplug not supported for this guest");
2547 } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
2548 if (!mc->query_hotpluggable_cpus) {
2549 error_setg(errp, "CPU hot unplug not supported on this machine");
2550 return;
2552 spapr_core_unplug(hotplug_dev, dev, errp);
2556 static void spapr_machine_device_pre_plug(HotplugHandler *hotplug_dev,
2557 DeviceState *dev, Error **errp)
2559 if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
2560 spapr_core_pre_plug(hotplug_dev, dev, errp);
2564 static HotplugHandler *spapr_get_hotplug_handler(MachineState *machine,
2565 DeviceState *dev)
2567 if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM) ||
2568 object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
2569 return HOTPLUG_HANDLER(machine);
2571 return NULL;
2574 static unsigned spapr_cpu_index_to_socket_id(unsigned cpu_index)
2576 /* Allocate to NUMA nodes on a "socket" basis (not that concept of
2577 * socket means much for the paravirtualized PAPR platform) */
2578 return cpu_index / smp_threads / smp_cores;
2581 static HotpluggableCPUList *spapr_query_hotpluggable_cpus(MachineState *machine)
2583 int i;
2584 HotpluggableCPUList *head = NULL;
2585 sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
2586 int spapr_max_cores = max_cpus / smp_threads;
2588 for (i = 0; i < spapr_max_cores; i++) {
2589 HotpluggableCPUList *list_item = g_new0(typeof(*list_item), 1);
2590 HotpluggableCPU *cpu_item = g_new0(typeof(*cpu_item), 1);
2591 CpuInstanceProperties *cpu_props = g_new0(typeof(*cpu_props), 1);
2593 cpu_item->type = spapr_get_cpu_core_type(machine->cpu_model);
2594 cpu_item->vcpus_count = smp_threads;
2595 cpu_props->has_core_id = true;
2596 cpu_props->core_id = i * smp_threads;
2597 /* TODO: add 'has_node/node' here to describe
2598 to which node core belongs */
2600 cpu_item->props = cpu_props;
2601 if (spapr->cores[i]) {
2602 cpu_item->has_qom_path = true;
2603 cpu_item->qom_path = object_get_canonical_path(spapr->cores[i]);
2605 list_item->value = cpu_item;
2606 list_item->next = head;
2607 head = list_item;
2609 return head;
2612 static void spapr_phb_placement(sPAPRMachineState *spapr, uint32_t index,
2613 uint64_t *buid, hwaddr *pio,
2614 hwaddr *mmio32, hwaddr *mmio64,
2615 unsigned n_dma, uint32_t *liobns, Error **errp)
2618 * New-style PHB window placement.
2620 * Goals: Gives large (1TiB), naturally aligned 64-bit MMIO window
2621 * for each PHB, in addition to 2GiB 32-bit MMIO and 64kiB PIO
2622 * windows.
2624 * Some guest kernels can't work with MMIO windows above 1<<46
2625 * (64TiB), so we place up to 31 PHBs in the area 32TiB..64TiB
2627 * 32TiB..(33TiB+1984kiB) contains the 64kiB PIO windows for each
2628 * PHB stacked together. (32TiB+2GiB)..(32TiB+64GiB) contains the
2629 * 2GiB 32-bit MMIO windows for each PHB. Then 33..64TiB has the
2630 * 1TiB 64-bit MMIO windows for each PHB.
2632 const uint64_t base_buid = 0x800000020000000ULL;
2633 const int max_phbs =
2634 (SPAPR_PCI_LIMIT - SPAPR_PCI_BASE) / SPAPR_PCI_MEM64_WIN_SIZE - 1;
2635 int i;
2637 /* Sanity check natural alignments */
2638 QEMU_BUILD_BUG_ON((SPAPR_PCI_BASE % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
2639 QEMU_BUILD_BUG_ON((SPAPR_PCI_LIMIT % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
2640 QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM64_WIN_SIZE % SPAPR_PCI_MEM32_WIN_SIZE) != 0);
2641 QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM32_WIN_SIZE % SPAPR_PCI_IO_WIN_SIZE) != 0);
2642 /* Sanity check bounds */
2643 QEMU_BUILD_BUG_ON((max_phbs * SPAPR_PCI_IO_WIN_SIZE) > SPAPR_PCI_MEM32_WIN_SIZE);
2644 QEMU_BUILD_BUG_ON((max_phbs * SPAPR_PCI_MEM32_WIN_SIZE) > SPAPR_PCI_MEM64_WIN_SIZE);
2646 if (index >= max_phbs) {
2647 error_setg(errp, "\"index\" for PAPR PHB is too large (max %u)",
2648 max_phbs - 1);
2649 return;
2652 *buid = base_buid + index;
2653 for (i = 0; i < n_dma; ++i) {
2654 liobns[i] = SPAPR_PCI_LIOBN(index, i);
2657 *pio = SPAPR_PCI_BASE + index * SPAPR_PCI_IO_WIN_SIZE;
2658 *mmio32 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM32_WIN_SIZE;
2659 *mmio64 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM64_WIN_SIZE;
2662 static void spapr_machine_class_init(ObjectClass *oc, void *data)
2664 MachineClass *mc = MACHINE_CLASS(oc);
2665 sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(oc);
2666 FWPathProviderClass *fwc = FW_PATH_PROVIDER_CLASS(oc);
2667 NMIClass *nc = NMI_CLASS(oc);
2668 HotplugHandlerClass *hc = HOTPLUG_HANDLER_CLASS(oc);
2670 mc->desc = "pSeries Logical Partition (PAPR compliant)";
2673 * We set up the default / latest behaviour here. The class_init
2674 * functions for the specific versioned machine types can override
2675 * these details for backwards compatibility
2677 mc->init = ppc_spapr_init;
2678 mc->reset = ppc_spapr_reset;
2679 mc->block_default_type = IF_SCSI;
2680 mc->max_cpus = 255;
2681 mc->no_parallel = 1;
2682 mc->default_boot_order = "";
2683 mc->default_ram_size = 512 * M_BYTE;
2684 mc->kvm_type = spapr_kvm_type;
2685 mc->has_dynamic_sysbus = true;
2686 mc->pci_allow_0_address = true;
2687 mc->get_hotplug_handler = spapr_get_hotplug_handler;
2688 hc->pre_plug = spapr_machine_device_pre_plug;
2689 hc->plug = spapr_machine_device_plug;
2690 hc->unplug = spapr_machine_device_unplug;
2691 mc->cpu_index_to_socket_id = spapr_cpu_index_to_socket_id;
2692 hc->unplug_request = spapr_machine_device_unplug_request;
2694 smc->dr_lmb_enabled = true;
2695 smc->tcg_default_cpu = "POWER8";
2696 mc->query_hotpluggable_cpus = spapr_query_hotpluggable_cpus;
2697 fwc->get_dev_path = spapr_get_fw_dev_path;
2698 nc->nmi_monitor_handler = spapr_nmi;
2699 smc->phb_placement = spapr_phb_placement;
2702 static const TypeInfo spapr_machine_info = {
2703 .name = TYPE_SPAPR_MACHINE,
2704 .parent = TYPE_MACHINE,
2705 .abstract = true,
2706 .instance_size = sizeof(sPAPRMachineState),
2707 .instance_init = spapr_machine_initfn,
2708 .instance_finalize = spapr_machine_finalizefn,
2709 .class_size = sizeof(sPAPRMachineClass),
2710 .class_init = spapr_machine_class_init,
2711 .interfaces = (InterfaceInfo[]) {
2712 { TYPE_FW_PATH_PROVIDER },
2713 { TYPE_NMI },
2714 { TYPE_HOTPLUG_HANDLER },
2719 #define DEFINE_SPAPR_MACHINE(suffix, verstr, latest) \
2720 static void spapr_machine_##suffix##_class_init(ObjectClass *oc, \
2721 void *data) \
2723 MachineClass *mc = MACHINE_CLASS(oc); \
2724 spapr_machine_##suffix##_class_options(mc); \
2725 if (latest) { \
2726 mc->alias = "pseries"; \
2727 mc->is_default = 1; \
2730 static void spapr_machine_##suffix##_instance_init(Object *obj) \
2732 MachineState *machine = MACHINE(obj); \
2733 spapr_machine_##suffix##_instance_options(machine); \
2735 static const TypeInfo spapr_machine_##suffix##_info = { \
2736 .name = MACHINE_TYPE_NAME("pseries-" verstr), \
2737 .parent = TYPE_SPAPR_MACHINE, \
2738 .class_init = spapr_machine_##suffix##_class_init, \
2739 .instance_init = spapr_machine_##suffix##_instance_init, \
2740 }; \
2741 static void spapr_machine_register_##suffix(void) \
2743 type_register(&spapr_machine_##suffix##_info); \
2745 type_init(spapr_machine_register_##suffix)
2748 * pseries-2.8
2750 static void spapr_machine_2_8_instance_options(MachineState *machine)
2754 static void spapr_machine_2_8_class_options(MachineClass *mc)
2756 /* Defaults for the latest behaviour inherited from the base class */
2759 DEFINE_SPAPR_MACHINE(2_8, "2.8", true);
2762 * pseries-2.7
2764 #define SPAPR_COMPAT_2_7 \
2765 HW_COMPAT_2_7 \
2767 .driver = TYPE_SPAPR_PCI_HOST_BRIDGE, \
2768 .property = "mem_win_size", \
2769 .value = stringify(SPAPR_PCI_2_7_MMIO_WIN_SIZE),\
2770 }, \
2772 .driver = TYPE_SPAPR_PCI_HOST_BRIDGE, \
2773 .property = "mem64_win_size", \
2774 .value = "0", \
2775 }, \
2777 .driver = TYPE_POWERPC_CPU, \
2778 .property = "pre-2.8-migration", \
2779 .value = "on", \
2780 }, \
2782 .driver = TYPE_SPAPR_PCI_HOST_BRIDGE, \
2783 .property = "pre-2.8-migration", \
2784 .value = "on", \
2787 static void phb_placement_2_7(sPAPRMachineState *spapr, uint32_t index,
2788 uint64_t *buid, hwaddr *pio,
2789 hwaddr *mmio32, hwaddr *mmio64,
2790 unsigned n_dma, uint32_t *liobns, Error **errp)
2792 /* Legacy PHB placement for pseries-2.7 and earlier machine types */
2793 const uint64_t base_buid = 0x800000020000000ULL;
2794 const hwaddr phb_spacing = 0x1000000000ULL; /* 64 GiB */
2795 const hwaddr mmio_offset = 0xa0000000; /* 2 GiB + 512 MiB */
2796 const hwaddr pio_offset = 0x80000000; /* 2 GiB */
2797 const uint32_t max_index = 255;
2798 const hwaddr phb0_alignment = 0x10000000000ULL; /* 1 TiB */
2800 uint64_t ram_top = MACHINE(spapr)->ram_size;
2801 hwaddr phb0_base, phb_base;
2802 int i;
2804 /* Do we have hotpluggable memory? */
2805 if (MACHINE(spapr)->maxram_size > ram_top) {
2806 /* Can't just use maxram_size, because there may be an
2807 * alignment gap between normal and hotpluggable memory
2808 * regions */
2809 ram_top = spapr->hotplug_memory.base +
2810 memory_region_size(&spapr->hotplug_memory.mr);
2813 phb0_base = QEMU_ALIGN_UP(ram_top, phb0_alignment);
2815 if (index > max_index) {
2816 error_setg(errp, "\"index\" for PAPR PHB is too large (max %u)",
2817 max_index);
2818 return;
2821 *buid = base_buid + index;
2822 for (i = 0; i < n_dma; ++i) {
2823 liobns[i] = SPAPR_PCI_LIOBN(index, i);
2826 phb_base = phb0_base + index * phb_spacing;
2827 *pio = phb_base + pio_offset;
2828 *mmio32 = phb_base + mmio_offset;
2830 * We don't set the 64-bit MMIO window, relying on the PHB's
2831 * fallback behaviour of automatically splitting a large "32-bit"
2832 * window into contiguous 32-bit and 64-bit windows
2836 static void spapr_machine_2_7_instance_options(MachineState *machine)
2838 sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
2840 spapr_machine_2_8_instance_options(machine);
2841 spapr->use_hotplug_event_source = false;
2844 static void spapr_machine_2_7_class_options(MachineClass *mc)
2846 sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
2848 spapr_machine_2_8_class_options(mc);
2849 smc->tcg_default_cpu = "POWER7";
2850 SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_7);
2851 smc->phb_placement = phb_placement_2_7;
2854 DEFINE_SPAPR_MACHINE(2_7, "2.7", false);
2857 * pseries-2.6
2859 #define SPAPR_COMPAT_2_6 \
2860 HW_COMPAT_2_6 \
2862 .driver = TYPE_SPAPR_PCI_HOST_BRIDGE,\
2863 .property = "ddw",\
2864 .value = stringify(off),\
2867 static void spapr_machine_2_6_instance_options(MachineState *machine)
2869 spapr_machine_2_7_instance_options(machine);
2872 static void spapr_machine_2_6_class_options(MachineClass *mc)
2874 spapr_machine_2_7_class_options(mc);
2875 mc->query_hotpluggable_cpus = NULL;
2876 SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_6);
2879 DEFINE_SPAPR_MACHINE(2_6, "2.6", false);
2882 * pseries-2.5
2884 #define SPAPR_COMPAT_2_5 \
2885 HW_COMPAT_2_5 \
2887 .driver = "spapr-vlan", \
2888 .property = "use-rx-buffer-pools", \
2889 .value = "off", \
2892 static void spapr_machine_2_5_instance_options(MachineState *machine)
2894 spapr_machine_2_6_instance_options(machine);
2897 static void spapr_machine_2_5_class_options(MachineClass *mc)
2899 sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
2901 spapr_machine_2_6_class_options(mc);
2902 smc->use_ohci_by_default = true;
2903 SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_5);
2906 DEFINE_SPAPR_MACHINE(2_5, "2.5", false);
2909 * pseries-2.4
2911 #define SPAPR_COMPAT_2_4 \
2912 HW_COMPAT_2_4
2914 static void spapr_machine_2_4_instance_options(MachineState *machine)
2916 spapr_machine_2_5_instance_options(machine);
2919 static void spapr_machine_2_4_class_options(MachineClass *mc)
2921 sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
2923 spapr_machine_2_5_class_options(mc);
2924 smc->dr_lmb_enabled = false;
2925 SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_4);
2928 DEFINE_SPAPR_MACHINE(2_4, "2.4", false);
2931 * pseries-2.3
2933 #define SPAPR_COMPAT_2_3 \
2934 HW_COMPAT_2_3 \
2936 .driver = "spapr-pci-host-bridge",\
2937 .property = "dynamic-reconfiguration",\
2938 .value = "off",\
2941 static void spapr_machine_2_3_instance_options(MachineState *machine)
2943 spapr_machine_2_4_instance_options(machine);
2944 savevm_skip_section_footers();
2945 global_state_set_optional();
2946 savevm_skip_configuration();
2949 static void spapr_machine_2_3_class_options(MachineClass *mc)
2951 spapr_machine_2_4_class_options(mc);
2952 SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_3);
2954 DEFINE_SPAPR_MACHINE(2_3, "2.3", false);
2957 * pseries-2.2
2960 #define SPAPR_COMPAT_2_2 \
2961 HW_COMPAT_2_2 \
2963 .driver = TYPE_SPAPR_PCI_HOST_BRIDGE,\
2964 .property = "mem_win_size",\
2965 .value = "0x20000000",\
2968 static void spapr_machine_2_2_instance_options(MachineState *machine)
2970 spapr_machine_2_3_instance_options(machine);
2971 machine->suppress_vmdesc = true;
2974 static void spapr_machine_2_2_class_options(MachineClass *mc)
2976 spapr_machine_2_3_class_options(mc);
2977 SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_2);
2979 DEFINE_SPAPR_MACHINE(2_2, "2.2", false);
2982 * pseries-2.1
2984 #define SPAPR_COMPAT_2_1 \
2985 HW_COMPAT_2_1
2987 static void spapr_machine_2_1_instance_options(MachineState *machine)
2989 spapr_machine_2_2_instance_options(machine);
2992 static void spapr_machine_2_1_class_options(MachineClass *mc)
2994 spapr_machine_2_2_class_options(mc);
2995 SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_1);
2997 DEFINE_SPAPR_MACHINE(2_1, "2.1", false);
2999 static void spapr_machine_register_types(void)
3001 type_register_static(&spapr_machine_info);
3004 type_init(spapr_machine_register_types)