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
27 #include "qemu/osdep.h"
28 #include "qapi/error.h"
29 #include "sysemu/sysemu.h"
30 #include "sysemu/numa.h"
33 #include "hw/fw-path-provider.h"
36 #include "sysemu/device_tree.h"
37 #include "sysemu/block-backend.h"
38 #include "sysemu/cpus.h"
39 #include "sysemu/hw_accel.h"
41 #include "migration/migration.h"
42 #include "mmu-hash64.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"
62 #include "qemu/config-file.h"
63 #include "qemu/error-report.h"
67 #include "hw/compat.h"
68 #include "qemu/cutils.h"
69 #include "hw/ppc/spapr_cpu_core.h"
70 #include "qmp-commands.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
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(sPAPRMachineState *spapr,
99 const char *type, const char *type_ics,
100 const char *type_icp, int nr_servers,
101 int nr_irqs, Error **errp)
103 XICSFabric *xi = XICS_FABRIC(spapr);
104 Error *err = NULL, *local_err = NULL;
106 ICSState *ics = NULL;
109 xics = XICS_COMMON(object_new(type));
110 qdev_set_parent_bus(DEVICE(xics), sysbus_get_default());
111 object_property_set_bool(OBJECT(xics), true, "realized", &err);
116 ics = ICS_SIMPLE(object_new(type_ics));
117 qdev_set_parent_bus(DEVICE(ics), sysbus_get_default());
118 object_property_add_child(OBJECT(spapr), "ics", OBJECT(ics), NULL);
119 object_property_set_int(OBJECT(ics), nr_irqs, "nr-irqs", &err);
120 object_property_add_const_link(OBJECT(ics), "xics", OBJECT(xics), NULL);
121 object_property_set_bool(OBJECT(ics), true, "realized", &local_err);
122 error_propagate(&err, local_err);
127 xics->ss = g_malloc0(nr_servers * sizeof(ICPState));
128 xics->nr_servers = nr_servers;
130 for (i = 0; i < nr_servers; i++) {
131 ICPState *icp = &xics->ss[i];
133 object_initialize(icp, sizeof(*icp), type_icp);
134 object_property_add_child(OBJECT(xics), "icp[*]", OBJECT(icp), NULL);
135 object_property_add_const_link(OBJECT(icp), "xics", OBJECT(xi), NULL);
136 object_property_set_bool(OBJECT(icp), true, "realized", &err);
140 object_unref(OBJECT(icp));
147 error_propagate(errp, err);
149 object_unparent(OBJECT(ics));
151 object_unparent(OBJECT(xics));
155 static XICSState *xics_system_init(MachineState *machine,
156 int nr_servers, int nr_irqs, Error **errp)
158 XICSState *xics = NULL;
163 if (machine_kernel_irqchip_allowed(machine)) {
164 xics = try_create_xics(SPAPR_MACHINE(machine),
165 TYPE_XICS_SPAPR_KVM, TYPE_ICS_KVM,
166 TYPE_KVM_ICP, nr_servers, nr_irqs, &err);
168 if (machine_kernel_irqchip_required(machine) && !xics) {
169 error_reportf_err(err,
170 "kernel_irqchip requested but unavailable: ");
177 xics = try_create_xics(SPAPR_MACHINE(machine),
178 TYPE_XICS_SPAPR, TYPE_ICS_SIMPLE,
179 TYPE_ICP, nr_servers, nr_irqs, errp);
185 static int spapr_fixup_cpu_smt_dt(void *fdt, int offset, PowerPCCPU *cpu,
189 uint32_t servers_prop[smt_threads];
190 uint32_t gservers_prop[smt_threads * 2];
191 int index = ppc_get_vcpu_dt_id(cpu);
193 if (cpu->compat_pvr) {
194 ret = fdt_setprop_cell(fdt, offset, "cpu-version", cpu->compat_pvr);
200 /* Build interrupt servers and gservers properties */
201 for (i = 0; i < smt_threads; i++) {
202 servers_prop[i] = cpu_to_be32(index + i);
203 /* Hack, direct the group queues back to cpu 0 */
204 gservers_prop[i*2] = cpu_to_be32(index + i);
205 gservers_prop[i*2 + 1] = 0;
207 ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-server#s",
208 servers_prop, sizeof(servers_prop));
212 ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-gserver#s",
213 gservers_prop, sizeof(gservers_prop));
218 static int spapr_fixup_cpu_numa_dt(void *fdt, int offset, CPUState *cs)
221 PowerPCCPU *cpu = POWERPC_CPU(cs);
222 int index = ppc_get_vcpu_dt_id(cpu);
223 uint32_t associativity[] = {cpu_to_be32(0x5),
227 cpu_to_be32(cs->numa_node),
230 /* Advertise NUMA via ibm,associativity */
231 if (nb_numa_nodes > 1) {
232 ret = fdt_setprop(fdt, offset, "ibm,associativity", associativity,
233 sizeof(associativity));
239 static int spapr_fixup_cpu_dt(void *fdt, sPAPRMachineState *spapr)
241 int ret = 0, offset, cpus_offset;
244 int smt = kvmppc_smt_threads();
245 uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
248 PowerPCCPU *cpu = POWERPC_CPU(cs);
249 DeviceClass *dc = DEVICE_GET_CLASS(cs);
250 int index = ppc_get_vcpu_dt_id(cpu);
251 int compat_smt = MIN(smp_threads, ppc_compat_max_threads(cpu));
253 if ((index % smt) != 0) {
257 snprintf(cpu_model, 32, "%s@%x", dc->fw_name, index);
259 cpus_offset = fdt_path_offset(fdt, "/cpus");
260 if (cpus_offset < 0) {
261 cpus_offset = fdt_add_subnode(fdt, fdt_path_offset(fdt, "/"),
263 if (cpus_offset < 0) {
267 offset = fdt_subnode_offset(fdt, cpus_offset, cpu_model);
269 offset = fdt_add_subnode(fdt, cpus_offset, cpu_model);
275 ret = fdt_setprop(fdt, offset, "ibm,pft-size",
276 pft_size_prop, sizeof(pft_size_prop));
281 ret = spapr_fixup_cpu_numa_dt(fdt, offset, cs);
286 ret = spapr_fixup_cpu_smt_dt(fdt, offset, cpu, compat_smt);
294 static hwaddr spapr_node0_size(void)
296 MachineState *machine = MACHINE(qdev_get_machine());
300 for (i = 0; i < nb_numa_nodes; ++i) {
301 if (numa_info[i].node_mem) {
302 return MIN(pow2floor(numa_info[i].node_mem),
307 return machine->ram_size;
310 static void add_str(GString *s, const gchar *s1)
312 g_string_append_len(s, s1, strlen(s1) + 1);
315 static int spapr_populate_memory_node(void *fdt, int nodeid, hwaddr start,
318 uint32_t associativity[] = {
319 cpu_to_be32(0x4), /* length */
320 cpu_to_be32(0x0), cpu_to_be32(0x0),
321 cpu_to_be32(0x0), cpu_to_be32(nodeid)
324 uint64_t mem_reg_property[2];
327 mem_reg_property[0] = cpu_to_be64(start);
328 mem_reg_property[1] = cpu_to_be64(size);
330 sprintf(mem_name, "memory@" TARGET_FMT_lx, start);
331 off = fdt_add_subnode(fdt, 0, mem_name);
333 _FDT((fdt_setprop_string(fdt, off, "device_type", "memory")));
334 _FDT((fdt_setprop(fdt, off, "reg", mem_reg_property,
335 sizeof(mem_reg_property))));
336 _FDT((fdt_setprop(fdt, off, "ibm,associativity", associativity,
337 sizeof(associativity))));
341 static int spapr_populate_memory(sPAPRMachineState *spapr, void *fdt)
343 MachineState *machine = MACHINE(spapr);
344 hwaddr mem_start, node_size;
345 int i, nb_nodes = nb_numa_nodes;
346 NodeInfo *nodes = numa_info;
349 /* No NUMA nodes, assume there is just one node with whole RAM */
350 if (!nb_numa_nodes) {
352 ramnode.node_mem = machine->ram_size;
356 for (i = 0, mem_start = 0; i < nb_nodes; ++i) {
357 if (!nodes[i].node_mem) {
360 if (mem_start >= machine->ram_size) {
363 node_size = nodes[i].node_mem;
364 if (node_size > machine->ram_size - mem_start) {
365 node_size = machine->ram_size - mem_start;
369 /* ppc_spapr_init() checks for rma_size <= node0_size already */
370 spapr_populate_memory_node(fdt, i, 0, spapr->rma_size);
371 mem_start += spapr->rma_size;
372 node_size -= spapr->rma_size;
374 for ( ; node_size; ) {
375 hwaddr sizetmp = pow2floor(node_size);
377 /* mem_start != 0 here */
378 if (ctzl(mem_start) < ctzl(sizetmp)) {
379 sizetmp = 1ULL << ctzl(mem_start);
382 spapr_populate_memory_node(fdt, i, mem_start, sizetmp);
383 node_size -= sizetmp;
384 mem_start += sizetmp;
391 /* Populate the "ibm,pa-features" property */
392 static void spapr_populate_pa_features(CPUPPCState *env, void *fdt, int offset)
394 uint8_t pa_features_206[] = { 6, 0,
395 0xf6, 0x1f, 0xc7, 0x00, 0x80, 0xc0 };
396 uint8_t pa_features_207[] = { 24, 0,
397 0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0,
398 0x80, 0x00, 0x00, 0x00, 0x00, 0x00,
399 0x00, 0x00, 0x00, 0x00, 0x80, 0x00,
400 0x80, 0x00, 0x80, 0x00, 0x00, 0x00 };
401 uint8_t *pa_features;
404 switch (env->mmu_model) {
405 case POWERPC_MMU_2_06:
406 case POWERPC_MMU_2_06a:
407 pa_features = pa_features_206;
408 pa_size = sizeof(pa_features_206);
410 case POWERPC_MMU_2_07:
411 case POWERPC_MMU_2_07a:
412 pa_features = pa_features_207;
413 pa_size = sizeof(pa_features_207);
419 if (env->ci_large_pages) {
421 * Note: we keep CI large pages off by default because a 64K capable
422 * guest provisioned with large pages might otherwise try to map a qemu
423 * framebuffer (or other kind of memory mapped PCI BAR) using 64K pages
424 * even if that qemu runs on a 4k host.
425 * We dd this bit back here if we are confident this is not an issue
427 pa_features[3] |= 0x20;
429 if (kvmppc_has_cap_htm() && pa_size > 24) {
430 pa_features[24] |= 0x80; /* Transactional memory support */
433 _FDT((fdt_setprop(fdt, offset, "ibm,pa-features", pa_features, pa_size)));
436 static void spapr_populate_cpu_dt(CPUState *cs, void *fdt, int offset,
437 sPAPRMachineState *spapr)
439 PowerPCCPU *cpu = POWERPC_CPU(cs);
440 CPUPPCState *env = &cpu->env;
441 PowerPCCPUClass *pcc = POWERPC_CPU_GET_CLASS(cs);
442 int index = ppc_get_vcpu_dt_id(cpu);
443 uint32_t segs[] = {cpu_to_be32(28), cpu_to_be32(40),
444 0xffffffff, 0xffffffff};
445 uint32_t tbfreq = kvm_enabled() ? kvmppc_get_tbfreq()
446 : SPAPR_TIMEBASE_FREQ;
447 uint32_t cpufreq = kvm_enabled() ? kvmppc_get_clockfreq() : 1000000000;
448 uint32_t page_sizes_prop[64];
449 size_t page_sizes_prop_size;
450 uint32_t vcpus_per_socket = smp_threads * smp_cores;
451 uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
452 int compat_smt = MIN(smp_threads, ppc_compat_max_threads(cpu));
453 sPAPRDRConnector *drc;
454 sPAPRDRConnectorClass *drck;
457 drc = spapr_dr_connector_by_id(SPAPR_DR_CONNECTOR_TYPE_CPU, index);
459 drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
460 drc_index = drck->get_index(drc);
461 _FDT((fdt_setprop_cell(fdt, offset, "ibm,my-drc-index", drc_index)));
464 _FDT((fdt_setprop_cell(fdt, offset, "reg", index)));
465 _FDT((fdt_setprop_string(fdt, offset, "device_type", "cpu")));
467 _FDT((fdt_setprop_cell(fdt, offset, "cpu-version", env->spr[SPR_PVR])));
468 _FDT((fdt_setprop_cell(fdt, offset, "d-cache-block-size",
469 env->dcache_line_size)));
470 _FDT((fdt_setprop_cell(fdt, offset, "d-cache-line-size",
471 env->dcache_line_size)));
472 _FDT((fdt_setprop_cell(fdt, offset, "i-cache-block-size",
473 env->icache_line_size)));
474 _FDT((fdt_setprop_cell(fdt, offset, "i-cache-line-size",
475 env->icache_line_size)));
477 if (pcc->l1_dcache_size) {
478 _FDT((fdt_setprop_cell(fdt, offset, "d-cache-size",
479 pcc->l1_dcache_size)));
481 error_report("Warning: Unknown L1 dcache size for cpu");
483 if (pcc->l1_icache_size) {
484 _FDT((fdt_setprop_cell(fdt, offset, "i-cache-size",
485 pcc->l1_icache_size)));
487 error_report("Warning: Unknown L1 icache size for cpu");
490 _FDT((fdt_setprop_cell(fdt, offset, "timebase-frequency", tbfreq)));
491 _FDT((fdt_setprop_cell(fdt, offset, "clock-frequency", cpufreq)));
492 _FDT((fdt_setprop_cell(fdt, offset, "slb-size", env->slb_nr)));
493 _FDT((fdt_setprop_cell(fdt, offset, "ibm,slb-size", env->slb_nr)));
494 _FDT((fdt_setprop_string(fdt, offset, "status", "okay")));
495 _FDT((fdt_setprop(fdt, offset, "64-bit", NULL, 0)));
497 if (env->spr_cb[SPR_PURR].oea_read) {
498 _FDT((fdt_setprop(fdt, offset, "ibm,purr", NULL, 0)));
501 if (env->mmu_model & POWERPC_MMU_1TSEG) {
502 _FDT((fdt_setprop(fdt, offset, "ibm,processor-segment-sizes",
503 segs, sizeof(segs))));
506 /* Advertise VMX/VSX (vector extensions) if available
507 * 0 / no property == no vector extensions
508 * 1 == VMX / Altivec available
509 * 2 == VSX available */
510 if (env->insns_flags & PPC_ALTIVEC) {
511 uint32_t vmx = (env->insns_flags2 & PPC2_VSX) ? 2 : 1;
513 _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", vmx)));
516 /* Advertise DFP (Decimal Floating Point) if available
517 * 0 / no property == no DFP
518 * 1 == DFP available */
519 if (env->insns_flags2 & PPC2_DFP) {
520 _FDT((fdt_setprop_cell(fdt, offset, "ibm,dfp", 1)));
523 page_sizes_prop_size = ppc_create_page_sizes_prop(env, page_sizes_prop,
524 sizeof(page_sizes_prop));
525 if (page_sizes_prop_size) {
526 _FDT((fdt_setprop(fdt, offset, "ibm,segment-page-sizes",
527 page_sizes_prop, page_sizes_prop_size)));
530 spapr_populate_pa_features(env, fdt, offset);
532 _FDT((fdt_setprop_cell(fdt, offset, "ibm,chip-id",
533 cs->cpu_index / vcpus_per_socket)));
535 _FDT((fdt_setprop(fdt, offset, "ibm,pft-size",
536 pft_size_prop, sizeof(pft_size_prop))));
538 _FDT(spapr_fixup_cpu_numa_dt(fdt, offset, cs));
540 _FDT(spapr_fixup_cpu_smt_dt(fdt, offset, cpu, compat_smt));
543 static void spapr_populate_cpus_dt_node(void *fdt, sPAPRMachineState *spapr)
548 int smt = kvmppc_smt_threads();
550 cpus_offset = fdt_add_subnode(fdt, 0, "cpus");
552 _FDT((fdt_setprop_cell(fdt, cpus_offset, "#address-cells", 0x1)));
553 _FDT((fdt_setprop_cell(fdt, cpus_offset, "#size-cells", 0x0)));
556 * We walk the CPUs in reverse order to ensure that CPU DT nodes
557 * created by fdt_add_subnode() end up in the right order in FDT
558 * for the guest kernel the enumerate the CPUs correctly.
560 CPU_FOREACH_REVERSE(cs) {
561 PowerPCCPU *cpu = POWERPC_CPU(cs);
562 int index = ppc_get_vcpu_dt_id(cpu);
563 DeviceClass *dc = DEVICE_GET_CLASS(cs);
566 if ((index % smt) != 0) {
570 nodename = g_strdup_printf("%s@%x", dc->fw_name, index);
571 offset = fdt_add_subnode(fdt, cpus_offset, nodename);
574 spapr_populate_cpu_dt(cs, fdt, offset, spapr);
580 * Adds ibm,dynamic-reconfiguration-memory node.
581 * Refer to docs/specs/ppc-spapr-hotplug.txt for the documentation
582 * of this device tree node.
584 static int spapr_populate_drconf_memory(sPAPRMachineState *spapr, void *fdt)
586 MachineState *machine = MACHINE(spapr);
588 uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
589 uint32_t prop_lmb_size[] = {0, cpu_to_be32(lmb_size)};
590 uint32_t hotplug_lmb_start = spapr->hotplug_memory.base / lmb_size;
591 uint32_t nr_lmbs = (spapr->hotplug_memory.base +
592 memory_region_size(&spapr->hotplug_memory.mr)) /
594 uint32_t *int_buf, *cur_index, buf_len;
595 int nr_nodes = nb_numa_nodes ? nb_numa_nodes : 1;
598 * Don't create the node if there is no hotpluggable memory
600 if (machine->ram_size == machine->maxram_size) {
605 * Allocate enough buffer size to fit in ibm,dynamic-memory
606 * or ibm,associativity-lookup-arrays
608 buf_len = MAX(nr_lmbs * SPAPR_DR_LMB_LIST_ENTRY_SIZE + 1, nr_nodes * 4 + 2)
610 cur_index = int_buf = g_malloc0(buf_len);
612 offset = fdt_add_subnode(fdt, 0, "ibm,dynamic-reconfiguration-memory");
614 ret = fdt_setprop(fdt, offset, "ibm,lmb-size", prop_lmb_size,
615 sizeof(prop_lmb_size));
620 ret = fdt_setprop_cell(fdt, offset, "ibm,memory-flags-mask", 0xff);
625 ret = fdt_setprop_cell(fdt, offset, "ibm,memory-preservation-time", 0x0);
630 /* ibm,dynamic-memory */
631 int_buf[0] = cpu_to_be32(nr_lmbs);
633 for (i = 0; i < nr_lmbs; i++) {
634 uint64_t addr = i * lmb_size;
635 uint32_t *dynamic_memory = cur_index;
637 if (i >= hotplug_lmb_start) {
638 sPAPRDRConnector *drc;
639 sPAPRDRConnectorClass *drck;
641 drc = spapr_dr_connector_by_id(SPAPR_DR_CONNECTOR_TYPE_LMB, i);
643 drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
645 dynamic_memory[0] = cpu_to_be32(addr >> 32);
646 dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
647 dynamic_memory[2] = cpu_to_be32(drck->get_index(drc));
648 dynamic_memory[3] = cpu_to_be32(0); /* reserved */
649 dynamic_memory[4] = cpu_to_be32(numa_get_node(addr, NULL));
650 if (memory_region_present(get_system_memory(), addr)) {
651 dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_ASSIGNED);
653 dynamic_memory[5] = cpu_to_be32(0);
657 * LMB information for RMA, boot time RAM and gap b/n RAM and
658 * hotplug memory region -- all these are marked as reserved
659 * and as having no valid DRC.
661 dynamic_memory[0] = cpu_to_be32(addr >> 32);
662 dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
663 dynamic_memory[2] = cpu_to_be32(0);
664 dynamic_memory[3] = cpu_to_be32(0); /* reserved */
665 dynamic_memory[4] = cpu_to_be32(-1);
666 dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_RESERVED |
667 SPAPR_LMB_FLAGS_DRC_INVALID);
670 cur_index += SPAPR_DR_LMB_LIST_ENTRY_SIZE;
672 ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory", int_buf, buf_len);
677 /* ibm,associativity-lookup-arrays */
679 int_buf[0] = cpu_to_be32(nr_nodes);
680 int_buf[1] = cpu_to_be32(4); /* Number of entries per associativity list */
682 for (i = 0; i < nr_nodes; i++) {
683 uint32_t associativity[] = {
689 memcpy(cur_index, associativity, sizeof(associativity));
692 ret = fdt_setprop(fdt, offset, "ibm,associativity-lookup-arrays", int_buf,
693 (cur_index - int_buf) * sizeof(uint32_t));
699 static int spapr_dt_cas_updates(sPAPRMachineState *spapr, void *fdt,
700 sPAPROptionVector *ov5_updates)
702 sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr);
705 /* Generate ibm,dynamic-reconfiguration-memory node if required */
706 if (spapr_ovec_test(ov5_updates, OV5_DRCONF_MEMORY)) {
707 g_assert(smc->dr_lmb_enabled);
708 ret = spapr_populate_drconf_memory(spapr, fdt);
714 offset = fdt_path_offset(fdt, "/chosen");
716 offset = fdt_add_subnode(fdt, 0, "chosen");
721 ret = spapr_ovec_populate_dt(fdt, offset, spapr->ov5_cas,
722 "ibm,architecture-vec-5");
728 int spapr_h_cas_compose_response(sPAPRMachineState *spapr,
729 target_ulong addr, target_ulong size,
730 sPAPROptionVector *ov5_updates)
732 void *fdt, *fdt_skel;
733 sPAPRDeviceTreeUpdateHeader hdr = { .version_id = 1 };
737 /* Create sceleton */
738 fdt_skel = g_malloc0(size);
739 _FDT((fdt_create(fdt_skel, size)));
740 _FDT((fdt_begin_node(fdt_skel, "")));
741 _FDT((fdt_end_node(fdt_skel)));
742 _FDT((fdt_finish(fdt_skel)));
743 fdt = g_malloc0(size);
744 _FDT((fdt_open_into(fdt_skel, fdt, size)));
747 /* Fixup cpu nodes */
748 _FDT((spapr_fixup_cpu_dt(fdt, spapr)));
750 if (spapr_dt_cas_updates(spapr, fdt, ov5_updates)) {
754 /* Pack resulting tree */
755 _FDT((fdt_pack(fdt)));
757 if (fdt_totalsize(fdt) + sizeof(hdr) > size) {
758 trace_spapr_cas_failed(size);
762 cpu_physical_memory_write(addr, &hdr, sizeof(hdr));
763 cpu_physical_memory_write(addr + sizeof(hdr), fdt, fdt_totalsize(fdt));
764 trace_spapr_cas_continue(fdt_totalsize(fdt) + sizeof(hdr));
770 static void spapr_dt_rtas(sPAPRMachineState *spapr, void *fdt)
773 GString *hypertas = g_string_sized_new(256);
774 GString *qemu_hypertas = g_string_sized_new(256);
775 uint32_t refpoints[] = { cpu_to_be32(0x4), cpu_to_be32(0x4) };
776 uint64_t max_hotplug_addr = spapr->hotplug_memory.base +
777 memory_region_size(&spapr->hotplug_memory.mr);
778 uint32_t lrdr_capacity[] = {
779 cpu_to_be32(max_hotplug_addr >> 32),
780 cpu_to_be32(max_hotplug_addr & 0xffffffff),
781 0, cpu_to_be32(SPAPR_MEMORY_BLOCK_SIZE),
782 cpu_to_be32(max_cpus / smp_threads),
785 _FDT(rtas = fdt_add_subnode(fdt, 0, "rtas"));
788 add_str(hypertas, "hcall-pft");
789 add_str(hypertas, "hcall-term");
790 add_str(hypertas, "hcall-dabr");
791 add_str(hypertas, "hcall-interrupt");
792 add_str(hypertas, "hcall-tce");
793 add_str(hypertas, "hcall-vio");
794 add_str(hypertas, "hcall-splpar");
795 add_str(hypertas, "hcall-bulk");
796 add_str(hypertas, "hcall-set-mode");
797 add_str(hypertas, "hcall-sprg0");
798 add_str(hypertas, "hcall-copy");
799 add_str(hypertas, "hcall-debug");
800 add_str(qemu_hypertas, "hcall-memop1");
802 if (!kvm_enabled() || kvmppc_spapr_use_multitce()) {
803 add_str(hypertas, "hcall-multi-tce");
805 _FDT(fdt_setprop(fdt, rtas, "ibm,hypertas-functions",
806 hypertas->str, hypertas->len));
807 g_string_free(hypertas, TRUE);
808 _FDT(fdt_setprop(fdt, rtas, "qemu,hypertas-functions",
809 qemu_hypertas->str, qemu_hypertas->len));
810 g_string_free(qemu_hypertas, TRUE);
812 _FDT(fdt_setprop(fdt, rtas, "ibm,associativity-reference-points",
813 refpoints, sizeof(refpoints)));
815 _FDT(fdt_setprop_cell(fdt, rtas, "rtas-error-log-max",
816 RTAS_ERROR_LOG_MAX));
817 _FDT(fdt_setprop_cell(fdt, rtas, "rtas-event-scan-rate",
818 RTAS_EVENT_SCAN_RATE));
821 _FDT(fdt_setprop(fdt, rtas, "ibm,change-msix-capable", NULL, 0));
825 * According to PAPR, rtas ibm,os-term does not guarantee a return
826 * back to the guest cpu.
828 * While an additional ibm,extended-os-term property indicates
829 * that rtas call return will always occur. Set this property.
831 _FDT(fdt_setprop(fdt, rtas, "ibm,extended-os-term", NULL, 0));
833 _FDT(fdt_setprop(fdt, rtas, "ibm,lrdr-capacity",
834 lrdr_capacity, sizeof(lrdr_capacity)));
836 spapr_dt_rtas_tokens(fdt, rtas);
839 static void spapr_dt_chosen(sPAPRMachineState *spapr, void *fdt)
841 MachineState *machine = MACHINE(spapr);
843 const char *boot_device = machine->boot_order;
844 char *stdout_path = spapr_vio_stdout_path(spapr->vio_bus);
846 char *bootlist = get_boot_devices_list(&cb, true);
848 _FDT(chosen = fdt_add_subnode(fdt, 0, "chosen"));
850 _FDT(fdt_setprop_string(fdt, chosen, "bootargs", machine->kernel_cmdline));
851 _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-start",
852 spapr->initrd_base));
853 _FDT(fdt_setprop_cell(fdt, chosen, "linux,initrd-end",
854 spapr->initrd_base + spapr->initrd_size));
856 if (spapr->kernel_size) {
857 uint64_t kprop[2] = { cpu_to_be64(KERNEL_LOAD_ADDR),
858 cpu_to_be64(spapr->kernel_size) };
860 _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel",
861 &kprop, sizeof(kprop)));
862 if (spapr->kernel_le) {
863 _FDT(fdt_setprop(fdt, chosen, "qemu,boot-kernel-le", NULL, 0));
867 _FDT((fdt_setprop_cell(fdt, chosen, "qemu,boot-menu", boot_menu)));
869 _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-width", graphic_width));
870 _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-height", graphic_height));
871 _FDT(fdt_setprop_cell(fdt, chosen, "qemu,graphic-depth", graphic_depth));
873 if (cb && bootlist) {
876 for (i = 0; i < cb; i++) {
877 if (bootlist[i] == '\n') {
881 _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-list", bootlist));
884 if (boot_device && strlen(boot_device)) {
885 _FDT(fdt_setprop_string(fdt, chosen, "qemu,boot-device", boot_device));
888 if (!spapr->has_graphics && stdout_path) {
889 _FDT(fdt_setprop_string(fdt, chosen, "linux,stdout-path", stdout_path));
896 static void spapr_dt_hypervisor(sPAPRMachineState *spapr, void *fdt)
898 /* The /hypervisor node isn't in PAPR - this is a hack to allow PR
899 * KVM to work under pHyp with some guest co-operation */
901 uint8_t hypercall[16];
903 _FDT(hypervisor = fdt_add_subnode(fdt, 0, "hypervisor"));
904 /* indicate KVM hypercall interface */
905 _FDT(fdt_setprop_string(fdt, hypervisor, "compatible", "linux,kvm"));
906 if (kvmppc_has_cap_fixup_hcalls()) {
908 * Older KVM versions with older guest kernels were broken
909 * with the magic page, don't allow the guest to map it.
911 if (!kvmppc_get_hypercall(first_cpu->env_ptr, hypercall,
912 sizeof(hypercall))) {
913 _FDT(fdt_setprop(fdt, hypervisor, "hcall-instructions",
914 hypercall, sizeof(hypercall)));
919 static void *spapr_build_fdt(sPAPRMachineState *spapr,
923 MachineState *machine = MACHINE(qdev_get_machine());
924 MachineClass *mc = MACHINE_GET_CLASS(machine);
925 sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
931 fdt = g_malloc0(FDT_MAX_SIZE);
932 _FDT((fdt_create_empty_tree(fdt, FDT_MAX_SIZE)));
935 _FDT(fdt_setprop_string(fdt, 0, "device_type", "chrp"));
936 _FDT(fdt_setprop_string(fdt, 0, "model", "IBM pSeries (emulated by qemu)"));
937 _FDT(fdt_setprop_string(fdt, 0, "compatible", "qemu,pseries"));
940 * Add info to guest to indentify which host is it being run on
941 * and what is the uuid of the guest
943 if (kvmppc_get_host_model(&buf)) {
944 _FDT(fdt_setprop_string(fdt, 0, "host-model", buf));
947 if (kvmppc_get_host_serial(&buf)) {
948 _FDT(fdt_setprop_string(fdt, 0, "host-serial", buf));
952 buf = qemu_uuid_unparse_strdup(&qemu_uuid);
954 _FDT(fdt_setprop_string(fdt, 0, "vm,uuid", buf));
956 _FDT(fdt_setprop_string(fdt, 0, "system-id", buf));
960 if (qemu_get_vm_name()) {
961 _FDT(fdt_setprop_string(fdt, 0, "ibm,partition-name",
962 qemu_get_vm_name()));
965 _FDT(fdt_setprop_cell(fdt, 0, "#address-cells", 2));
966 _FDT(fdt_setprop_cell(fdt, 0, "#size-cells", 2));
968 /* /interrupt controller */
969 spapr_dt_xics(spapr->xics, fdt, PHANDLE_XICP);
971 ret = spapr_populate_memory(spapr, fdt);
973 error_report("couldn't setup memory nodes in fdt");
978 spapr_dt_vdevice(spapr->vio_bus, fdt);
980 if (object_resolve_path_type("", TYPE_SPAPR_RNG, NULL)) {
981 ret = spapr_rng_populate_dt(fdt);
983 error_report("could not set up rng device in the fdt");
988 QLIST_FOREACH(phb, &spapr->phbs, list) {
989 ret = spapr_populate_pci_dt(phb, PHANDLE_XICP, fdt);
991 error_report("couldn't setup PCI devices in fdt");
997 spapr_populate_cpus_dt_node(fdt, spapr);
999 if (smc->dr_lmb_enabled) {
1000 _FDT(spapr_drc_populate_dt(fdt, 0, NULL, SPAPR_DR_CONNECTOR_TYPE_LMB));
1003 if (mc->has_hotpluggable_cpus) {
1004 int offset = fdt_path_offset(fdt, "/cpus");
1005 ret = spapr_drc_populate_dt(fdt, offset, NULL,
1006 SPAPR_DR_CONNECTOR_TYPE_CPU);
1008 error_report("Couldn't set up CPU DR device tree properties");
1013 /* /event-sources */
1014 spapr_dt_events(spapr, fdt);
1017 spapr_dt_rtas(spapr, fdt);
1020 spapr_dt_chosen(spapr, fdt);
1023 if (kvm_enabled()) {
1024 spapr_dt_hypervisor(spapr, fdt);
1027 /* Build memory reserve map */
1028 if (spapr->kernel_size) {
1029 _FDT((fdt_add_mem_rsv(fdt, KERNEL_LOAD_ADDR, spapr->kernel_size)));
1031 if (spapr->initrd_size) {
1032 _FDT((fdt_add_mem_rsv(fdt, spapr->initrd_base, spapr->initrd_size)));
1035 /* ibm,client-architecture-support updates */
1036 ret = spapr_dt_cas_updates(spapr, fdt, spapr->ov5_cas);
1038 error_report("couldn't setup CAS properties fdt");
1045 static uint64_t translate_kernel_address(void *opaque, uint64_t addr)
1047 return (addr & 0x0fffffff) + KERNEL_LOAD_ADDR;
1050 static void emulate_spapr_hypercall(PPCVirtualHypervisor *vhyp,
1053 CPUPPCState *env = &cpu->env;
1055 /* The TCG path should also be holding the BQL at this point */
1056 g_assert(qemu_mutex_iothread_locked());
1059 hcall_dprintf("Hypercall made with MSR[PR]=1\n");
1060 env->gpr[3] = H_PRIVILEGE;
1062 env->gpr[3] = spapr_hypercall(cpu, env->gpr[3], &env->gpr[4]);
1066 #define HPTE(_table, _i) (void *)(((uint64_t *)(_table)) + ((_i) * 2))
1067 #define HPTE_VALID(_hpte) (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_VALID)
1068 #define HPTE_DIRTY(_hpte) (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_HPTE_DIRTY)
1069 #define CLEAN_HPTE(_hpte) ((*(uint64_t *)(_hpte)) &= tswap64(~HPTE64_V_HPTE_DIRTY))
1070 #define DIRTY_HPTE(_hpte) ((*(uint64_t *)(_hpte)) |= tswap64(HPTE64_V_HPTE_DIRTY))
1073 * Get the fd to access the kernel htab, re-opening it if necessary
1075 static int get_htab_fd(sPAPRMachineState *spapr)
1077 if (spapr->htab_fd >= 0) {
1078 return spapr->htab_fd;
1081 spapr->htab_fd = kvmppc_get_htab_fd(false);
1082 if (spapr->htab_fd < 0) {
1083 error_report("Unable to open fd for reading hash table from KVM: %s",
1087 return spapr->htab_fd;
1090 static void close_htab_fd(sPAPRMachineState *spapr)
1092 if (spapr->htab_fd >= 0) {
1093 close(spapr->htab_fd);
1095 spapr->htab_fd = -1;
1098 static hwaddr spapr_hpt_mask(PPCVirtualHypervisor *vhyp)
1100 sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1102 return HTAB_SIZE(spapr) / HASH_PTEG_SIZE_64 - 1;
1105 static const ppc_hash_pte64_t *spapr_map_hptes(PPCVirtualHypervisor *vhyp,
1108 sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1109 hwaddr pte_offset = ptex * HASH_PTE_SIZE_64;
1113 * HTAB is controlled by KVM. Fetch into temporary buffer
1115 ppc_hash_pte64_t *hptes = g_malloc(n * HASH_PTE_SIZE_64);
1116 kvmppc_read_hptes(hptes, ptex, n);
1121 * HTAB is controlled by QEMU. Just point to the internally
1124 return (const ppc_hash_pte64_t *)(spapr->htab + pte_offset);
1127 static void spapr_unmap_hptes(PPCVirtualHypervisor *vhyp,
1128 const ppc_hash_pte64_t *hptes,
1131 sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1134 g_free((void *)hptes);
1137 /* Nothing to do for qemu managed HPT */
1140 static void spapr_store_hpte(PPCVirtualHypervisor *vhyp, hwaddr ptex,
1141 uint64_t pte0, uint64_t pte1)
1143 sPAPRMachineState *spapr = SPAPR_MACHINE(vhyp);
1144 hwaddr offset = ptex * HASH_PTE_SIZE_64;
1147 kvmppc_write_hpte(ptex, pte0, pte1);
1149 stq_p(spapr->htab + offset, pte0);
1150 stq_p(spapr->htab + offset + HASH_PTE_SIZE_64 / 2, pte1);
1154 static int spapr_hpt_shift_for_ramsize(uint64_t ramsize)
1158 /* We aim for a hash table of size 1/128 the size of RAM (rounded
1159 * up). The PAPR recommendation is actually 1/64 of RAM size, but
1160 * that's much more than is needed for Linux guests */
1161 shift = ctz64(pow2ceil(ramsize)) - 7;
1162 shift = MAX(shift, 18); /* Minimum architected size */
1163 shift = MIN(shift, 46); /* Maximum architected size */
1167 static void spapr_reallocate_hpt(sPAPRMachineState *spapr, int shift,
1172 /* Clean up any HPT info from a previous boot */
1173 g_free(spapr->htab);
1175 spapr->htab_shift = 0;
1176 close_htab_fd(spapr);
1178 rc = kvmppc_reset_htab(shift);
1180 /* kernel-side HPT needed, but couldn't allocate one */
1181 error_setg_errno(errp, errno,
1182 "Failed to allocate KVM HPT of order %d (try smaller maxmem?)",
1184 /* This is almost certainly fatal, but if the caller really
1185 * wants to carry on with shift == 0, it's welcome to try */
1186 } else if (rc > 0) {
1187 /* kernel-side HPT allocated */
1190 "Requested order %d HPT, but kernel allocated order %ld (try smaller maxmem?)",
1194 spapr->htab_shift = shift;
1197 /* kernel-side HPT not needed, allocate in userspace instead */
1198 size_t size = 1ULL << shift;
1201 spapr->htab = qemu_memalign(size, size);
1203 error_setg_errno(errp, errno,
1204 "Could not allocate HPT of order %d", shift);
1208 memset(spapr->htab, 0, size);
1209 spapr->htab_shift = shift;
1211 for (i = 0; i < size / HASH_PTE_SIZE_64; i++) {
1212 DIRTY_HPTE(HPTE(spapr->htab, i));
1217 static void find_unknown_sysbus_device(SysBusDevice *sbdev, void *opaque)
1219 bool matched = false;
1221 if (object_dynamic_cast(OBJECT(sbdev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
1226 error_report("Device %s is not supported by this machine yet.",
1227 qdev_fw_name(DEVICE(sbdev)));
1232 static void ppc_spapr_reset(void)
1234 MachineState *machine = MACHINE(qdev_get_machine());
1235 sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
1236 PowerPCCPU *first_ppc_cpu;
1237 uint32_t rtas_limit;
1238 hwaddr rtas_addr, fdt_addr;
1242 /* Check for unknown sysbus devices */
1243 foreach_dynamic_sysbus_device(find_unknown_sysbus_device, NULL);
1245 /* Allocate and/or reset the hash page table */
1246 spapr_reallocate_hpt(spapr,
1247 spapr_hpt_shift_for_ramsize(machine->maxram_size),
1250 /* Update the RMA size if necessary */
1251 if (spapr->vrma_adjust) {
1252 spapr->rma_size = kvmppc_rma_size(spapr_node0_size(),
1256 qemu_devices_reset();
1259 * We place the device tree and RTAS just below either the top of the RMA,
1260 * or just below 2GB, whichever is lowere, so that it can be
1261 * processed with 32-bit real mode code if necessary
1263 rtas_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR);
1264 rtas_addr = rtas_limit - RTAS_MAX_SIZE;
1265 fdt_addr = rtas_addr - FDT_MAX_SIZE;
1267 /* if this reset wasn't generated by CAS, we should reset our
1268 * negotiated options and start from scratch */
1269 if (!spapr->cas_reboot) {
1270 spapr_ovec_cleanup(spapr->ov5_cas);
1271 spapr->ov5_cas = spapr_ovec_new();
1274 fdt = spapr_build_fdt(spapr, rtas_addr, spapr->rtas_size);
1276 spapr_load_rtas(spapr, fdt, rtas_addr);
1280 /* Should only fail if we've built a corrupted tree */
1283 if (fdt_totalsize(fdt) > FDT_MAX_SIZE) {
1284 error_report("FDT too big ! 0x%x bytes (max is 0x%x)",
1285 fdt_totalsize(fdt), FDT_MAX_SIZE);
1290 qemu_fdt_dumpdtb(fdt, fdt_totalsize(fdt));
1291 cpu_physical_memory_write(fdt_addr, fdt, fdt_totalsize(fdt));
1294 /* Set up the entry state */
1295 first_ppc_cpu = POWERPC_CPU(first_cpu);
1296 first_ppc_cpu->env.gpr[3] = fdt_addr;
1297 first_ppc_cpu->env.gpr[5] = 0;
1298 first_cpu->halted = 0;
1299 first_ppc_cpu->env.nip = SPAPR_ENTRY_POINT;
1301 spapr->cas_reboot = false;
1304 static void spapr_create_nvram(sPAPRMachineState *spapr)
1306 DeviceState *dev = qdev_create(&spapr->vio_bus->bus, "spapr-nvram");
1307 DriveInfo *dinfo = drive_get(IF_PFLASH, 0, 0);
1310 qdev_prop_set_drive(dev, "drive", blk_by_legacy_dinfo(dinfo),
1314 qdev_init_nofail(dev);
1316 spapr->nvram = (struct sPAPRNVRAM *)dev;
1319 static void spapr_rtc_create(sPAPRMachineState *spapr)
1321 DeviceState *dev = qdev_create(NULL, TYPE_SPAPR_RTC);
1323 qdev_init_nofail(dev);
1326 object_property_add_alias(qdev_get_machine(), "rtc-time",
1327 OBJECT(spapr->rtc), "date", NULL);
1330 /* Returns whether we want to use VGA or not */
1331 static bool spapr_vga_init(PCIBus *pci_bus, Error **errp)
1333 switch (vga_interface_type) {
1340 return pci_vga_init(pci_bus) != NULL;
1343 "Unsupported VGA mode, only -vga std or -vga virtio is supported");
1348 static int spapr_post_load(void *opaque, int version_id)
1350 sPAPRMachineState *spapr = (sPAPRMachineState *)opaque;
1353 /* In earlier versions, there was no separate qdev for the PAPR
1354 * RTC, so the RTC offset was stored directly in sPAPREnvironment.
1355 * So when migrating from those versions, poke the incoming offset
1356 * value into the RTC device */
1357 if (version_id < 3) {
1358 err = spapr_rtc_import_offset(spapr->rtc, spapr->rtc_offset);
1364 static bool version_before_3(void *opaque, int version_id)
1366 return version_id < 3;
1369 static bool spapr_ov5_cas_needed(void *opaque)
1371 sPAPRMachineState *spapr = opaque;
1372 sPAPROptionVector *ov5_mask = spapr_ovec_new();
1373 sPAPROptionVector *ov5_legacy = spapr_ovec_new();
1374 sPAPROptionVector *ov5_removed = spapr_ovec_new();
1377 /* Prior to the introduction of sPAPROptionVector, we had two option
1378 * vectors we dealt with: OV5_FORM1_AFFINITY, and OV5_DRCONF_MEMORY.
1379 * Both of these options encode machine topology into the device-tree
1380 * in such a way that the now-booted OS should still be able to interact
1381 * appropriately with QEMU regardless of what options were actually
1382 * negotiatied on the source side.
1384 * As such, we can avoid migrating the CAS-negotiated options if these
1385 * are the only options available on the current machine/platform.
1386 * Since these are the only options available for pseries-2.7 and
1387 * earlier, this allows us to maintain old->new/new->old migration
1390 * For QEMU 2.8+, there are additional CAS-negotiatable options available
1391 * via default pseries-2.8 machines and explicit command-line parameters.
1392 * Some of these options, like OV5_HP_EVT, *do* require QEMU to be aware
1393 * of the actual CAS-negotiated values to continue working properly. For
1394 * example, availability of memory unplug depends on knowing whether
1395 * OV5_HP_EVT was negotiated via CAS.
1397 * Thus, for any cases where the set of available CAS-negotiatable
1398 * options extends beyond OV5_FORM1_AFFINITY and OV5_DRCONF_MEMORY, we
1399 * include the CAS-negotiated options in the migration stream.
1401 spapr_ovec_set(ov5_mask, OV5_FORM1_AFFINITY);
1402 spapr_ovec_set(ov5_mask, OV5_DRCONF_MEMORY);
1404 /* spapr_ovec_diff returns true if bits were removed. we avoid using
1405 * the mask itself since in the future it's possible "legacy" bits may be
1406 * removed via machine options, which could generate a false positive
1407 * that breaks migration.
1409 spapr_ovec_intersect(ov5_legacy, spapr->ov5, ov5_mask);
1410 cas_needed = spapr_ovec_diff(ov5_removed, spapr->ov5, ov5_legacy);
1412 spapr_ovec_cleanup(ov5_mask);
1413 spapr_ovec_cleanup(ov5_legacy);
1414 spapr_ovec_cleanup(ov5_removed);
1419 static const VMStateDescription vmstate_spapr_ov5_cas = {
1420 .name = "spapr_option_vector_ov5_cas",
1422 .minimum_version_id = 1,
1423 .needed = spapr_ov5_cas_needed,
1424 .fields = (VMStateField[]) {
1425 VMSTATE_STRUCT_POINTER_V(ov5_cas, sPAPRMachineState, 1,
1426 vmstate_spapr_ovec, sPAPROptionVector),
1427 VMSTATE_END_OF_LIST()
1431 static const VMStateDescription vmstate_spapr = {
1434 .minimum_version_id = 1,
1435 .post_load = spapr_post_load,
1436 .fields = (VMStateField[]) {
1437 /* used to be @next_irq */
1438 VMSTATE_UNUSED_BUFFER(version_before_3, 0, 4),
1441 VMSTATE_UINT64_TEST(rtc_offset, sPAPRMachineState, version_before_3),
1443 VMSTATE_PPC_TIMEBASE_V(tb, sPAPRMachineState, 2),
1444 VMSTATE_END_OF_LIST()
1446 .subsections = (const VMStateDescription*[]) {
1447 &vmstate_spapr_ov5_cas,
1452 static int htab_save_setup(QEMUFile *f, void *opaque)
1454 sPAPRMachineState *spapr = opaque;
1456 /* "Iteration" header */
1457 qemu_put_be32(f, spapr->htab_shift);
1460 spapr->htab_save_index = 0;
1461 spapr->htab_first_pass = true;
1463 assert(kvm_enabled());
1470 static void htab_save_first_pass(QEMUFile *f, sPAPRMachineState *spapr,
1473 bool has_timeout = max_ns != -1;
1474 int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
1475 int index = spapr->htab_save_index;
1476 int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
1478 assert(spapr->htab_first_pass);
1483 /* Consume invalid HPTEs */
1484 while ((index < htabslots)
1485 && !HPTE_VALID(HPTE(spapr->htab, index))) {
1487 CLEAN_HPTE(HPTE(spapr->htab, index));
1490 /* Consume valid HPTEs */
1492 while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
1493 && HPTE_VALID(HPTE(spapr->htab, index))) {
1495 CLEAN_HPTE(HPTE(spapr->htab, index));
1498 if (index > chunkstart) {
1499 int n_valid = index - chunkstart;
1501 qemu_put_be32(f, chunkstart);
1502 qemu_put_be16(f, n_valid);
1503 qemu_put_be16(f, 0);
1504 qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
1505 HASH_PTE_SIZE_64 * n_valid);
1508 (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
1512 } while ((index < htabslots) && !qemu_file_rate_limit(f));
1514 if (index >= htabslots) {
1515 assert(index == htabslots);
1517 spapr->htab_first_pass = false;
1519 spapr->htab_save_index = index;
1522 static int htab_save_later_pass(QEMUFile *f, sPAPRMachineState *spapr,
1525 bool final = max_ns < 0;
1526 int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
1527 int examined = 0, sent = 0;
1528 int index = spapr->htab_save_index;
1529 int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
1531 assert(!spapr->htab_first_pass);
1534 int chunkstart, invalidstart;
1536 /* Consume non-dirty HPTEs */
1537 while ((index < htabslots)
1538 && !HPTE_DIRTY(HPTE(spapr->htab, index))) {
1544 /* Consume valid dirty HPTEs */
1545 while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
1546 && HPTE_DIRTY(HPTE(spapr->htab, index))
1547 && HPTE_VALID(HPTE(spapr->htab, index))) {
1548 CLEAN_HPTE(HPTE(spapr->htab, index));
1553 invalidstart = index;
1554 /* Consume invalid dirty HPTEs */
1555 while ((index < htabslots) && (index - invalidstart < USHRT_MAX)
1556 && HPTE_DIRTY(HPTE(spapr->htab, index))
1557 && !HPTE_VALID(HPTE(spapr->htab, index))) {
1558 CLEAN_HPTE(HPTE(spapr->htab, index));
1563 if (index > chunkstart) {
1564 int n_valid = invalidstart - chunkstart;
1565 int n_invalid = index - invalidstart;
1567 qemu_put_be32(f, chunkstart);
1568 qemu_put_be16(f, n_valid);
1569 qemu_put_be16(f, n_invalid);
1570 qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
1571 HASH_PTE_SIZE_64 * n_valid);
1572 sent += index - chunkstart;
1574 if (!final && (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
1579 if (examined >= htabslots) {
1583 if (index >= htabslots) {
1584 assert(index == htabslots);
1587 } while ((examined < htabslots) && (!qemu_file_rate_limit(f) || final));
1589 if (index >= htabslots) {
1590 assert(index == htabslots);
1594 spapr->htab_save_index = index;
1596 return (examined >= htabslots) && (sent == 0) ? 1 : 0;
1599 #define MAX_ITERATION_NS 5000000 /* 5 ms */
1600 #define MAX_KVM_BUF_SIZE 2048
1602 static int htab_save_iterate(QEMUFile *f, void *opaque)
1604 sPAPRMachineState *spapr = opaque;
1608 /* Iteration header */
1609 qemu_put_be32(f, 0);
1612 assert(kvm_enabled());
1614 fd = get_htab_fd(spapr);
1619 rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, MAX_ITERATION_NS);
1623 } else if (spapr->htab_first_pass) {
1624 htab_save_first_pass(f, spapr, MAX_ITERATION_NS);
1626 rc = htab_save_later_pass(f, spapr, MAX_ITERATION_NS);
1630 qemu_put_be32(f, 0);
1631 qemu_put_be16(f, 0);
1632 qemu_put_be16(f, 0);
1637 static int htab_save_complete(QEMUFile *f, void *opaque)
1639 sPAPRMachineState *spapr = opaque;
1642 /* Iteration header */
1643 qemu_put_be32(f, 0);
1648 assert(kvm_enabled());
1650 fd = get_htab_fd(spapr);
1655 rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, -1);
1660 if (spapr->htab_first_pass) {
1661 htab_save_first_pass(f, spapr, -1);
1663 htab_save_later_pass(f, spapr, -1);
1667 qemu_put_be32(f, 0);
1668 qemu_put_be16(f, 0);
1669 qemu_put_be16(f, 0);
1674 static int htab_load(QEMUFile *f, void *opaque, int version_id)
1676 sPAPRMachineState *spapr = opaque;
1677 uint32_t section_hdr;
1680 if (version_id < 1 || version_id > 1) {
1681 error_report("htab_load() bad version");
1685 section_hdr = qemu_get_be32(f);
1688 Error *local_err = NULL;
1690 /* First section gives the htab size */
1691 spapr_reallocate_hpt(spapr, section_hdr, &local_err);
1693 error_report_err(local_err);
1700 assert(kvm_enabled());
1702 fd = kvmppc_get_htab_fd(true);
1704 error_report("Unable to open fd to restore KVM hash table: %s",
1711 uint16_t n_valid, n_invalid;
1713 index = qemu_get_be32(f);
1714 n_valid = qemu_get_be16(f);
1715 n_invalid = qemu_get_be16(f);
1717 if ((index == 0) && (n_valid == 0) && (n_invalid == 0)) {
1722 if ((index + n_valid + n_invalid) >
1723 (HTAB_SIZE(spapr) / HASH_PTE_SIZE_64)) {
1724 /* Bad index in stream */
1726 "htab_load() bad index %d (%hd+%hd entries) in htab stream (htab_shift=%d)",
1727 index, n_valid, n_invalid, spapr->htab_shift);
1733 qemu_get_buffer(f, HPTE(spapr->htab, index),
1734 HASH_PTE_SIZE_64 * n_valid);
1737 memset(HPTE(spapr->htab, index + n_valid), 0,
1738 HASH_PTE_SIZE_64 * n_invalid);
1745 rc = kvmppc_load_htab_chunk(f, fd, index, n_valid, n_invalid);
1760 static void htab_cleanup(void *opaque)
1762 sPAPRMachineState *spapr = opaque;
1764 close_htab_fd(spapr);
1767 static SaveVMHandlers savevm_htab_handlers = {
1768 .save_live_setup = htab_save_setup,
1769 .save_live_iterate = htab_save_iterate,
1770 .save_live_complete_precopy = htab_save_complete,
1771 .cleanup = htab_cleanup,
1772 .load_state = htab_load,
1775 static void spapr_boot_set(void *opaque, const char *boot_device,
1778 MachineState *machine = MACHINE(qdev_get_machine());
1779 machine->boot_order = g_strdup(boot_device);
1783 * Reset routine for LMB DR devices.
1785 * Unlike PCI DR devices, LMB DR devices explicitly register this reset
1786 * routine. Reset for PCI DR devices will be handled by PHB reset routine
1787 * when it walks all its children devices. LMB devices reset occurs
1788 * as part of spapr_ppc_reset().
1790 static void spapr_drc_reset(void *opaque)
1792 sPAPRDRConnector *drc = opaque;
1793 DeviceState *d = DEVICE(drc);
1800 static void spapr_create_lmb_dr_connectors(sPAPRMachineState *spapr)
1802 MachineState *machine = MACHINE(spapr);
1803 uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
1804 uint32_t nr_lmbs = (machine->maxram_size - machine->ram_size)/lmb_size;
1807 for (i = 0; i < nr_lmbs; i++) {
1808 sPAPRDRConnector *drc;
1811 addr = i * lmb_size + spapr->hotplug_memory.base;
1812 drc = spapr_dr_connector_new(OBJECT(spapr), SPAPR_DR_CONNECTOR_TYPE_LMB,
1814 qemu_register_reset(spapr_drc_reset, drc);
1819 * If RAM size, maxmem size and individual node mem sizes aren't aligned
1820 * to SPAPR_MEMORY_BLOCK_SIZE(256MB), then refuse to start the guest
1821 * since we can't support such unaligned sizes with DRCONF_MEMORY.
1823 static void spapr_validate_node_memory(MachineState *machine, Error **errp)
1827 if (machine->ram_size % SPAPR_MEMORY_BLOCK_SIZE) {
1828 error_setg(errp, "Memory size 0x" RAM_ADDR_FMT
1829 " is not aligned to %llu MiB",
1831 SPAPR_MEMORY_BLOCK_SIZE / M_BYTE);
1835 if (machine->maxram_size % SPAPR_MEMORY_BLOCK_SIZE) {
1836 error_setg(errp, "Maximum memory size 0x" RAM_ADDR_FMT
1837 " is not aligned to %llu MiB",
1839 SPAPR_MEMORY_BLOCK_SIZE / M_BYTE);
1843 for (i = 0; i < nb_numa_nodes; i++) {
1844 if (numa_info[i].node_mem % SPAPR_MEMORY_BLOCK_SIZE) {
1846 "Node %d memory size 0x%" PRIx64
1847 " is not aligned to %llu MiB",
1848 i, numa_info[i].node_mem,
1849 SPAPR_MEMORY_BLOCK_SIZE / M_BYTE);
1855 /* find cpu slot in machine->possible_cpus by core_id */
1856 static CPUArchId *spapr_find_cpu_slot(MachineState *ms, uint32_t id, int *idx)
1858 int index = id / smp_threads;
1860 if (index >= ms->possible_cpus->len) {
1866 return &ms->possible_cpus->cpus[index];
1869 static void spapr_init_cpus(sPAPRMachineState *spapr)
1871 MachineState *machine = MACHINE(spapr);
1872 MachineClass *mc = MACHINE_GET_CLASS(machine);
1873 char *type = spapr_get_cpu_core_type(machine->cpu_model);
1874 int smt = kvmppc_smt_threads();
1875 const CPUArchIdList *possible_cpus;
1876 int boot_cores_nr = smp_cpus / smp_threads;
1880 error_report("Unable to find sPAPR CPU Core definition");
1884 possible_cpus = mc->possible_cpu_arch_ids(machine);
1885 if (mc->has_hotpluggable_cpus) {
1886 if (smp_cpus % smp_threads) {
1887 error_report("smp_cpus (%u) must be multiple of threads (%u)",
1888 smp_cpus, smp_threads);
1891 if (max_cpus % smp_threads) {
1892 error_report("max_cpus (%u) must be multiple of threads (%u)",
1893 max_cpus, smp_threads);
1897 if (max_cpus != smp_cpus) {
1898 error_report("This machine version does not support CPU hotplug");
1901 boot_cores_nr = possible_cpus->len;
1904 for (i = 0; i < possible_cpus->len; i++) {
1905 int core_id = i * smp_threads;
1907 if (mc->has_hotpluggable_cpus) {
1908 sPAPRDRConnector *drc =
1909 spapr_dr_connector_new(OBJECT(spapr),
1910 SPAPR_DR_CONNECTOR_TYPE_CPU,
1911 (core_id / smp_threads) * smt);
1913 qemu_register_reset(spapr_drc_reset, drc);
1916 if (i < boot_cores_nr) {
1917 Object *core = object_new(type);
1918 int nr_threads = smp_threads;
1920 /* Handle the partially filled core for older machine types */
1921 if ((i + 1) * smp_threads >= smp_cpus) {
1922 nr_threads = smp_cpus - i * smp_threads;
1925 object_property_set_int(core, nr_threads, "nr-threads",
1927 object_property_set_int(core, core_id, CPU_CORE_PROP_CORE_ID,
1929 object_property_set_bool(core, true, "realized", &error_fatal);
1935 /* pSeries LPAR / sPAPR hardware init */
1936 static void ppc_spapr_init(MachineState *machine)
1938 sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
1939 sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1940 const char *kernel_filename = machine->kernel_filename;
1941 const char *initrd_filename = machine->initrd_filename;
1944 MemoryRegion *sysmem = get_system_memory();
1945 MemoryRegion *ram = g_new(MemoryRegion, 1);
1946 MemoryRegion *rma_region;
1948 hwaddr rma_alloc_size;
1949 hwaddr node0_size = spapr_node0_size();
1950 long load_limit, fw_size;
1952 int smt = kvmppc_smt_threads();
1954 msi_nonbroken = true;
1956 QLIST_INIT(&spapr->phbs);
1958 /* Allocate RMA if necessary */
1959 rma_alloc_size = kvmppc_alloc_rma(&rma);
1961 if (rma_alloc_size == -1) {
1962 error_report("Unable to create RMA");
1966 if (rma_alloc_size && (rma_alloc_size < node0_size)) {
1967 spapr->rma_size = rma_alloc_size;
1969 spapr->rma_size = node0_size;
1971 /* With KVM, we don't actually know whether KVM supports an
1972 * unbounded RMA (PR KVM) or is limited by the hash table size
1973 * (HV KVM using VRMA), so we always assume the latter
1975 * In that case, we also limit the initial allocations for RTAS
1976 * etc... to 256M since we have no way to know what the VRMA size
1977 * is going to be as it depends on the size of the hash table
1978 * isn't determined yet.
1980 if (kvm_enabled()) {
1981 spapr->vrma_adjust = 1;
1982 spapr->rma_size = MIN(spapr->rma_size, 0x10000000);
1985 /* Actually we don't support unbounded RMA anymore since we
1986 * added proper emulation of HV mode. The max we can get is
1987 * 16G which also happens to be what we configure for PAPR
1988 * mode so make sure we don't do anything bigger than that
1990 spapr->rma_size = MIN(spapr->rma_size, 0x400000000ull);
1993 if (spapr->rma_size > node0_size) {
1994 error_report("Numa node 0 has to span the RMA (%#08"HWADDR_PRIx")",
1999 /* Setup a load limit for the ramdisk leaving room for SLOF and FDT */
2000 load_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR) - FW_OVERHEAD;
2002 /* Set up Interrupt Controller before we create the VCPUs */
2003 spapr->xics = xics_system_init(machine,
2004 DIV_ROUND_UP(max_cpus * smt, smp_threads),
2005 XICS_IRQS_SPAPR, &error_fatal);
2007 /* Set up containers for ibm,client-set-architecture negotiated options */
2008 spapr->ov5 = spapr_ovec_new();
2009 spapr->ov5_cas = spapr_ovec_new();
2011 if (smc->dr_lmb_enabled) {
2012 spapr_ovec_set(spapr->ov5, OV5_DRCONF_MEMORY);
2013 spapr_validate_node_memory(machine, &error_fatal);
2016 spapr_ovec_set(spapr->ov5, OV5_FORM1_AFFINITY);
2018 /* advertise support for dedicated HP event source to guests */
2019 if (spapr->use_hotplug_event_source) {
2020 spapr_ovec_set(spapr->ov5, OV5_HP_EVT);
2024 if (machine->cpu_model == NULL) {
2025 machine->cpu_model = kvm_enabled() ? "host" : smc->tcg_default_cpu;
2028 ppc_cpu_parse_features(machine->cpu_model);
2030 spapr_init_cpus(spapr);
2032 if (kvm_enabled()) {
2033 /* Enable H_LOGICAL_CI_* so SLOF can talk to in-kernel devices */
2034 kvmppc_enable_logical_ci_hcalls();
2035 kvmppc_enable_set_mode_hcall();
2037 /* H_CLEAR_MOD/_REF are mandatory in PAPR, but off by default */
2038 kvmppc_enable_clear_ref_mod_hcalls();
2042 memory_region_allocate_system_memory(ram, NULL, "ppc_spapr.ram",
2044 memory_region_add_subregion(sysmem, 0, ram);
2046 if (rma_alloc_size && rma) {
2047 rma_region = g_new(MemoryRegion, 1);
2048 memory_region_init_ram_ptr(rma_region, NULL, "ppc_spapr.rma",
2049 rma_alloc_size, rma);
2050 vmstate_register_ram_global(rma_region);
2051 memory_region_add_subregion(sysmem, 0, rma_region);
2054 /* initialize hotplug memory address space */
2055 if (machine->ram_size < machine->maxram_size) {
2056 ram_addr_t hotplug_mem_size = machine->maxram_size - machine->ram_size;
2058 * Limit the number of hotpluggable memory slots to half the number
2059 * slots that KVM supports, leaving the other half for PCI and other
2060 * devices. However ensure that number of slots doesn't drop below 32.
2062 int max_memslots = kvm_enabled() ? kvm_get_max_memslots() / 2 :
2063 SPAPR_MAX_RAM_SLOTS;
2065 if (max_memslots < SPAPR_MAX_RAM_SLOTS) {
2066 max_memslots = SPAPR_MAX_RAM_SLOTS;
2068 if (machine->ram_slots > max_memslots) {
2069 error_report("Specified number of memory slots %"
2070 PRIu64" exceeds max supported %d",
2071 machine->ram_slots, max_memslots);
2075 spapr->hotplug_memory.base = ROUND_UP(machine->ram_size,
2076 SPAPR_HOTPLUG_MEM_ALIGN);
2077 memory_region_init(&spapr->hotplug_memory.mr, OBJECT(spapr),
2078 "hotplug-memory", hotplug_mem_size);
2079 memory_region_add_subregion(sysmem, spapr->hotplug_memory.base,
2080 &spapr->hotplug_memory.mr);
2083 if (smc->dr_lmb_enabled) {
2084 spapr_create_lmb_dr_connectors(spapr);
2087 filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, "spapr-rtas.bin");
2089 error_report("Could not find LPAR rtas '%s'", "spapr-rtas.bin");
2092 spapr->rtas_size = get_image_size(filename);
2093 if (spapr->rtas_size < 0) {
2094 error_report("Could not get size of LPAR rtas '%s'", filename);
2097 spapr->rtas_blob = g_malloc(spapr->rtas_size);
2098 if (load_image_size(filename, spapr->rtas_blob, spapr->rtas_size) < 0) {
2099 error_report("Could not load LPAR rtas '%s'", filename);
2102 if (spapr->rtas_size > RTAS_MAX_SIZE) {
2103 error_report("RTAS too big ! 0x%zx bytes (max is 0x%x)",
2104 (size_t)spapr->rtas_size, RTAS_MAX_SIZE);
2109 /* Set up RTAS event infrastructure */
2110 spapr_events_init(spapr);
2112 /* Set up the RTC RTAS interfaces */
2113 spapr_rtc_create(spapr);
2115 /* Set up VIO bus */
2116 spapr->vio_bus = spapr_vio_bus_init();
2118 for (i = 0; i < MAX_SERIAL_PORTS; i++) {
2119 if (serial_hds[i]) {
2120 spapr_vty_create(spapr->vio_bus, serial_hds[i]);
2124 /* We always have at least the nvram device on VIO */
2125 spapr_create_nvram(spapr);
2128 spapr_pci_rtas_init();
2130 phb = spapr_create_phb(spapr, 0);
2132 for (i = 0; i < nb_nics; i++) {
2133 NICInfo *nd = &nd_table[i];
2136 nd->model = g_strdup("ibmveth");
2139 if (strcmp(nd->model, "ibmveth") == 0) {
2140 spapr_vlan_create(spapr->vio_bus, nd);
2142 pci_nic_init_nofail(&nd_table[i], phb->bus, nd->model, NULL);
2146 for (i = 0; i <= drive_get_max_bus(IF_SCSI); i++) {
2147 spapr_vscsi_create(spapr->vio_bus);
2151 if (spapr_vga_init(phb->bus, &error_fatal)) {
2152 spapr->has_graphics = true;
2153 machine->usb |= defaults_enabled() && !machine->usb_disabled;
2157 if (smc->use_ohci_by_default) {
2158 pci_create_simple(phb->bus, -1, "pci-ohci");
2160 pci_create_simple(phb->bus, -1, "nec-usb-xhci");
2163 if (spapr->has_graphics) {
2164 USBBus *usb_bus = usb_bus_find(-1);
2166 usb_create_simple(usb_bus, "usb-kbd");
2167 usb_create_simple(usb_bus, "usb-mouse");
2171 if (spapr->rma_size < (MIN_RMA_SLOF << 20)) {
2173 "pSeries SLOF firmware requires >= %ldM guest RMA (Real Mode Area memory)",
2178 if (kernel_filename) {
2179 uint64_t lowaddr = 0;
2181 spapr->kernel_size = load_elf(kernel_filename, translate_kernel_address,
2182 NULL, NULL, &lowaddr, NULL, 1,
2183 PPC_ELF_MACHINE, 0, 0);
2184 if (spapr->kernel_size == ELF_LOAD_WRONG_ENDIAN) {
2185 spapr->kernel_size = load_elf(kernel_filename,
2186 translate_kernel_address, NULL, NULL,
2187 &lowaddr, NULL, 0, PPC_ELF_MACHINE,
2189 spapr->kernel_le = spapr->kernel_size > 0;
2191 if (spapr->kernel_size < 0) {
2192 error_report("error loading %s: %s", kernel_filename,
2193 load_elf_strerror(spapr->kernel_size));
2198 if (initrd_filename) {
2199 /* Try to locate the initrd in the gap between the kernel
2200 * and the firmware. Add a bit of space just in case
2202 spapr->initrd_base = (KERNEL_LOAD_ADDR + spapr->kernel_size
2203 + 0x1ffff) & ~0xffff;
2204 spapr->initrd_size = load_image_targphys(initrd_filename,
2207 - spapr->initrd_base);
2208 if (spapr->initrd_size < 0) {
2209 error_report("could not load initial ram disk '%s'",
2216 if (bios_name == NULL) {
2217 bios_name = FW_FILE_NAME;
2219 filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
2221 error_report("Could not find LPAR firmware '%s'", bios_name);
2224 fw_size = load_image_targphys(filename, 0, FW_MAX_SIZE);
2226 error_report("Could not load LPAR firmware '%s'", filename);
2231 /* FIXME: Should register things through the MachineState's qdev
2232 * interface, this is a legacy from the sPAPREnvironment structure
2233 * which predated MachineState but had a similar function */
2234 vmstate_register(NULL, 0, &vmstate_spapr, spapr);
2235 register_savevm_live(NULL, "spapr/htab", -1, 1,
2236 &savevm_htab_handlers, spapr);
2239 QTAILQ_INIT(&spapr->ccs_list);
2240 qemu_register_reset(spapr_ccs_reset_hook, spapr);
2242 qemu_register_boot_set(spapr_boot_set, spapr);
2244 /* to stop and start vmclock */
2245 if (kvm_enabled()) {
2246 qemu_add_vm_change_state_handler(cpu_ppc_clock_vm_state_change,
2251 static int spapr_kvm_type(const char *vm_type)
2257 if (!strcmp(vm_type, "HV")) {
2261 if (!strcmp(vm_type, "PR")) {
2265 error_report("Unknown kvm-type specified '%s'", vm_type);
2270 * Implementation of an interface to adjust firmware path
2271 * for the bootindex property handling.
2273 static char *spapr_get_fw_dev_path(FWPathProvider *p, BusState *bus,
2276 #define CAST(type, obj, name) \
2277 ((type *)object_dynamic_cast(OBJECT(obj), (name)))
2278 SCSIDevice *d = CAST(SCSIDevice, dev, TYPE_SCSI_DEVICE);
2279 sPAPRPHBState *phb = CAST(sPAPRPHBState, dev, TYPE_SPAPR_PCI_HOST_BRIDGE);
2282 void *spapr = CAST(void, bus->parent, "spapr-vscsi");
2283 VirtIOSCSI *virtio = CAST(VirtIOSCSI, bus->parent, TYPE_VIRTIO_SCSI);
2284 USBDevice *usb = CAST(USBDevice, bus->parent, TYPE_USB_DEVICE);
2288 * Replace "channel@0/disk@0,0" with "disk@8000000000000000":
2289 * We use SRP luns of the form 8000 | (bus << 8) | (id << 5) | lun
2290 * in the top 16 bits of the 64-bit LUN
2292 unsigned id = 0x8000 | (d->id << 8) | d->lun;
2293 return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
2294 (uint64_t)id << 48);
2295 } else if (virtio) {
2297 * We use SRP luns of the form 01000000 | (target << 8) | lun
2298 * in the top 32 bits of the 64-bit LUN
2299 * Note: the quote above is from SLOF and it is wrong,
2300 * the actual binding is:
2301 * swap 0100 or 10 << or 20 << ( target lun-id -- srplun )
2303 unsigned id = 0x1000000 | (d->id << 16) | d->lun;
2304 return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
2305 (uint64_t)id << 32);
2308 * We use SRP luns of the form 01000000 | (usb-port << 16) | lun
2309 * in the top 32 bits of the 64-bit LUN
2311 unsigned usb_port = atoi(usb->port->path);
2312 unsigned id = 0x1000000 | (usb_port << 16) | d->lun;
2313 return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
2314 (uint64_t)id << 32);
2319 * SLOF probes the USB devices, and if it recognizes that the device is a
2320 * storage device, it changes its name to "storage" instead of "usb-host",
2321 * and additionally adds a child node for the SCSI LUN, so the correct
2322 * boot path in SLOF is something like .../storage@1/disk@xxx" instead.
2324 if (strcmp("usb-host", qdev_fw_name(dev)) == 0) {
2325 USBDevice *usbdev = CAST(USBDevice, dev, TYPE_USB_DEVICE);
2326 if (usb_host_dev_is_scsi_storage(usbdev)) {
2327 return g_strdup_printf("storage@%s/disk", usbdev->port->path);
2332 /* Replace "pci" with "pci@800000020000000" */
2333 return g_strdup_printf("pci@%"PRIX64, phb->buid);
2339 static char *spapr_get_kvm_type(Object *obj, Error **errp)
2341 sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2343 return g_strdup(spapr->kvm_type);
2346 static void spapr_set_kvm_type(Object *obj, const char *value, Error **errp)
2348 sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2350 g_free(spapr->kvm_type);
2351 spapr->kvm_type = g_strdup(value);
2354 static bool spapr_get_modern_hotplug_events(Object *obj, Error **errp)
2356 sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2358 return spapr->use_hotplug_event_source;
2361 static void spapr_set_modern_hotplug_events(Object *obj, bool value,
2364 sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2366 spapr->use_hotplug_event_source = value;
2369 static void spapr_machine_initfn(Object *obj)
2371 sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2373 spapr->htab_fd = -1;
2374 spapr->use_hotplug_event_source = true;
2375 object_property_add_str(obj, "kvm-type",
2376 spapr_get_kvm_type, spapr_set_kvm_type, NULL);
2377 object_property_set_description(obj, "kvm-type",
2378 "Specifies the KVM virtualization mode (HV, PR)",
2380 object_property_add_bool(obj, "modern-hotplug-events",
2381 spapr_get_modern_hotplug_events,
2382 spapr_set_modern_hotplug_events,
2384 object_property_set_description(obj, "modern-hotplug-events",
2385 "Use dedicated hotplug event mechanism in"
2386 " place of standard EPOW events when possible"
2387 " (required for memory hot-unplug support)",
2391 static void spapr_machine_finalizefn(Object *obj)
2393 sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2395 g_free(spapr->kvm_type);
2398 void spapr_do_system_reset_on_cpu(CPUState *cs, run_on_cpu_data arg)
2400 cpu_synchronize_state(cs);
2401 ppc_cpu_do_system_reset(cs);
2404 static void spapr_nmi(NMIState *n, int cpu_index, Error **errp)
2409 async_run_on_cpu(cs, spapr_do_system_reset_on_cpu, RUN_ON_CPU_NULL);
2413 static void spapr_add_lmbs(DeviceState *dev, uint64_t addr_start, uint64_t size,
2414 uint32_t node, bool dedicated_hp_event_source,
2417 sPAPRDRConnector *drc;
2418 sPAPRDRConnectorClass *drck;
2419 uint32_t nr_lmbs = size/SPAPR_MEMORY_BLOCK_SIZE;
2420 int i, fdt_offset, fdt_size;
2422 uint64_t addr = addr_start;
2424 for (i = 0; i < nr_lmbs; i++) {
2425 drc = spapr_dr_connector_by_id(SPAPR_DR_CONNECTOR_TYPE_LMB,
2426 addr/SPAPR_MEMORY_BLOCK_SIZE);
2429 fdt = create_device_tree(&fdt_size);
2430 fdt_offset = spapr_populate_memory_node(fdt, node, addr,
2431 SPAPR_MEMORY_BLOCK_SIZE);
2433 drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
2434 drck->attach(drc, dev, fdt, fdt_offset, !dev->hotplugged, errp);
2435 addr += SPAPR_MEMORY_BLOCK_SIZE;
2436 if (!dev->hotplugged) {
2437 /* guests expect coldplugged LMBs to be pre-allocated */
2438 drck->set_allocation_state(drc, SPAPR_DR_ALLOCATION_STATE_USABLE);
2439 drck->set_isolation_state(drc, SPAPR_DR_ISOLATION_STATE_UNISOLATED);
2442 /* send hotplug notification to the
2443 * guest only in case of hotplugged memory
2445 if (dev->hotplugged) {
2446 if (dedicated_hp_event_source) {
2447 drc = spapr_dr_connector_by_id(SPAPR_DR_CONNECTOR_TYPE_LMB,
2448 addr_start / SPAPR_MEMORY_BLOCK_SIZE);
2449 drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
2450 spapr_hotplug_req_add_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
2452 drck->get_index(drc));
2454 spapr_hotplug_req_add_by_count(SPAPR_DR_CONNECTOR_TYPE_LMB,
2460 static void spapr_memory_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
2461 uint32_t node, Error **errp)
2463 Error *local_err = NULL;
2464 sPAPRMachineState *ms = SPAPR_MACHINE(hotplug_dev);
2465 PCDIMMDevice *dimm = PC_DIMM(dev);
2466 PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
2467 MemoryRegion *mr = ddc->get_memory_region(dimm);
2468 uint64_t align = memory_region_get_alignment(mr);
2469 uint64_t size = memory_region_size(mr);
2473 if (size % SPAPR_MEMORY_BLOCK_SIZE) {
2474 error_setg(&local_err, "Hotplugged memory size must be a multiple of "
2475 "%lld MB", SPAPR_MEMORY_BLOCK_SIZE/M_BYTE);
2479 mem_dev = object_property_get_str(OBJECT(dimm), PC_DIMM_MEMDEV_PROP, NULL);
2480 if (mem_dev && !kvmppc_is_mem_backend_page_size_ok(mem_dev)) {
2481 error_setg(&local_err, "Memory backend has bad page size. "
2482 "Use 'memory-backend-file' with correct mem-path.");
2486 pc_dimm_memory_plug(dev, &ms->hotplug_memory, mr, align, &local_err);
2491 addr = object_property_get_int(OBJECT(dimm), PC_DIMM_ADDR_PROP, &local_err);
2493 pc_dimm_memory_unplug(dev, &ms->hotplug_memory, mr);
2497 spapr_add_lmbs(dev, addr, size, node,
2498 spapr_ovec_test(ms->ov5_cas, OV5_HP_EVT),
2502 error_propagate(errp, local_err);
2505 typedef struct sPAPRDIMMState {
2509 static void spapr_lmb_release(DeviceState *dev, void *opaque)
2511 sPAPRDIMMState *ds = (sPAPRDIMMState *)opaque;
2512 HotplugHandler *hotplug_ctrl;
2514 if (--ds->nr_lmbs) {
2521 * Now that all the LMBs have been removed by the guest, call the
2522 * pc-dimm unplug handler to cleanup up the pc-dimm device.
2524 hotplug_ctrl = qdev_get_hotplug_handler(dev);
2525 hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
2528 static void spapr_del_lmbs(DeviceState *dev, uint64_t addr_start, uint64_t size,
2531 sPAPRDRConnector *drc;
2532 sPAPRDRConnectorClass *drck;
2533 uint32_t nr_lmbs = size / SPAPR_MEMORY_BLOCK_SIZE;
2535 sPAPRDIMMState *ds = g_malloc0(sizeof(sPAPRDIMMState));
2536 uint64_t addr = addr_start;
2538 ds->nr_lmbs = nr_lmbs;
2539 for (i = 0; i < nr_lmbs; i++) {
2540 drc = spapr_dr_connector_by_id(SPAPR_DR_CONNECTOR_TYPE_LMB,
2541 addr / SPAPR_MEMORY_BLOCK_SIZE);
2544 drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
2545 drck->detach(drc, dev, spapr_lmb_release, ds, errp);
2546 addr += SPAPR_MEMORY_BLOCK_SIZE;
2549 drc = spapr_dr_connector_by_id(SPAPR_DR_CONNECTOR_TYPE_LMB,
2550 addr_start / SPAPR_MEMORY_BLOCK_SIZE);
2551 drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
2552 spapr_hotplug_req_remove_by_count_indexed(SPAPR_DR_CONNECTOR_TYPE_LMB,
2554 drck->get_index(drc));
2557 static void spapr_memory_unplug(HotplugHandler *hotplug_dev, DeviceState *dev,
2560 sPAPRMachineState *ms = SPAPR_MACHINE(hotplug_dev);
2561 PCDIMMDevice *dimm = PC_DIMM(dev);
2562 PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
2563 MemoryRegion *mr = ddc->get_memory_region(dimm);
2565 pc_dimm_memory_unplug(dev, &ms->hotplug_memory, mr);
2566 object_unparent(OBJECT(dev));
2569 static void spapr_memory_unplug_request(HotplugHandler *hotplug_dev,
2570 DeviceState *dev, Error **errp)
2572 Error *local_err = NULL;
2573 PCDIMMDevice *dimm = PC_DIMM(dev);
2574 PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
2575 MemoryRegion *mr = ddc->get_memory_region(dimm);
2576 uint64_t size = memory_region_size(mr);
2579 addr = object_property_get_int(OBJECT(dimm), PC_DIMM_ADDR_PROP, &local_err);
2584 spapr_del_lmbs(dev, addr, size, &error_abort);
2586 error_propagate(errp, local_err);
2589 void *spapr_populate_hotplug_cpu_dt(CPUState *cs, int *fdt_offset,
2590 sPAPRMachineState *spapr)
2592 PowerPCCPU *cpu = POWERPC_CPU(cs);
2593 DeviceClass *dc = DEVICE_GET_CLASS(cs);
2594 int id = ppc_get_vcpu_dt_id(cpu);
2596 int offset, fdt_size;
2599 fdt = create_device_tree(&fdt_size);
2600 nodename = g_strdup_printf("%s@%x", dc->fw_name, id);
2601 offset = fdt_add_subnode(fdt, 0, nodename);
2603 spapr_populate_cpu_dt(cs, fdt, offset, spapr);
2606 *fdt_offset = offset;
2610 static void spapr_core_unplug(HotplugHandler *hotplug_dev, DeviceState *dev,
2613 MachineState *ms = MACHINE(qdev_get_machine());
2614 CPUCore *cc = CPU_CORE(dev);
2615 CPUArchId *core_slot = spapr_find_cpu_slot(ms, cc->core_id, NULL);
2617 core_slot->cpu = NULL;
2618 object_unparent(OBJECT(dev));
2621 static void spapr_core_release(DeviceState *dev, void *opaque)
2623 HotplugHandler *hotplug_ctrl;
2625 hotplug_ctrl = qdev_get_hotplug_handler(dev);
2626 hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort);
2630 void spapr_core_unplug_request(HotplugHandler *hotplug_dev, DeviceState *dev,
2634 sPAPRDRConnector *drc;
2635 sPAPRDRConnectorClass *drck;
2636 Error *local_err = NULL;
2637 CPUCore *cc = CPU_CORE(dev);
2638 int smt = kvmppc_smt_threads();
2640 if (!spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index)) {
2641 error_setg(errp, "Unable to find CPU core with core-id: %d",
2646 error_setg(errp, "Boot CPU core may not be unplugged");
2650 drc = spapr_dr_connector_by_id(SPAPR_DR_CONNECTOR_TYPE_CPU, index * smt);
2653 drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
2654 drck->detach(drc, dev, spapr_core_release, NULL, &local_err);
2656 error_propagate(errp, local_err);
2660 spapr_hotplug_req_remove_by_index(drc);
2663 static void spapr_core_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
2666 sPAPRMachineState *spapr = SPAPR_MACHINE(OBJECT(hotplug_dev));
2667 MachineClass *mc = MACHINE_GET_CLASS(spapr);
2668 sPAPRCPUCore *core = SPAPR_CPU_CORE(OBJECT(dev));
2669 CPUCore *cc = CPU_CORE(dev);
2670 CPUState *cs = CPU(core->threads);
2671 sPAPRDRConnector *drc;
2672 Error *local_err = NULL;
2675 int smt = kvmppc_smt_threads();
2676 CPUArchId *core_slot;
2679 core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
2681 error_setg(errp, "Unable to find CPU core with core-id: %d",
2685 drc = spapr_dr_connector_by_id(SPAPR_DR_CONNECTOR_TYPE_CPU, index * smt);
2687 g_assert(drc || !mc->has_hotpluggable_cpus);
2690 * Setup CPU DT entries only for hotplugged CPUs. For boot time or
2691 * coldplugged CPUs DT entries are setup in spapr_build_fdt().
2693 if (dev->hotplugged) {
2694 fdt = spapr_populate_hotplug_cpu_dt(cs, &fdt_offset, spapr);
2698 sPAPRDRConnectorClass *drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
2699 drck->attach(drc, dev, fdt, fdt_offset, !dev->hotplugged, &local_err);
2702 error_propagate(errp, local_err);
2707 if (dev->hotplugged) {
2709 * Send hotplug notification interrupt to the guest only in case
2710 * of hotplugged CPUs.
2712 spapr_hotplug_req_add_by_index(drc);
2715 * Set the right DRC states for cold plugged CPU.
2718 sPAPRDRConnectorClass *drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
2719 drck->set_allocation_state(drc, SPAPR_DR_ALLOCATION_STATE_USABLE);
2720 drck->set_isolation_state(drc, SPAPR_DR_ISOLATION_STATE_UNISOLATED);
2723 core_slot->cpu = OBJECT(dev);
2726 static void spapr_core_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
2729 MachineState *machine = MACHINE(OBJECT(hotplug_dev));
2730 MachineClass *mc = MACHINE_GET_CLASS(hotplug_dev);
2731 Error *local_err = NULL;
2732 CPUCore *cc = CPU_CORE(dev);
2733 char *base_core_type = spapr_get_cpu_core_type(machine->cpu_model);
2734 const char *type = object_get_typename(OBJECT(dev));
2735 CPUArchId *core_slot;
2738 if (dev->hotplugged && !mc->has_hotpluggable_cpus) {
2739 error_setg(&local_err, "CPU hotplug not supported for this machine");
2743 if (strcmp(base_core_type, type)) {
2744 error_setg(&local_err, "CPU core type should be %s", base_core_type);
2748 if (cc->core_id % smp_threads) {
2749 error_setg(&local_err, "invalid core id %d", cc->core_id);
2753 core_slot = spapr_find_cpu_slot(MACHINE(hotplug_dev), cc->core_id, &index);
2755 error_setg(&local_err, "core id %d out of range", cc->core_id);
2759 if (core_slot->cpu) {
2760 error_setg(&local_err, "core %d already populated", cc->core_id);
2765 g_free(base_core_type);
2766 error_propagate(errp, local_err);
2769 static void spapr_machine_device_plug(HotplugHandler *hotplug_dev,
2770 DeviceState *dev, Error **errp)
2772 sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(qdev_get_machine());
2774 if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
2777 if (!smc->dr_lmb_enabled) {
2778 error_setg(errp, "Memory hotplug not supported for this machine");
2781 node = object_property_get_int(OBJECT(dev), PC_DIMM_NODE_PROP, errp);
2785 if (node < 0 || node >= MAX_NODES) {
2786 error_setg(errp, "Invaild node %d", node);
2791 * Currently PowerPC kernel doesn't allow hot-adding memory to
2792 * memory-less node, but instead will silently add the memory
2793 * to the first node that has some memory. This causes two
2794 * unexpected behaviours for the user.
2796 * - Memory gets hotplugged to a different node than what the user
2798 * - Since pc-dimm subsystem in QEMU still thinks that memory belongs
2799 * to memory-less node, a reboot will set things accordingly
2800 * and the previously hotplugged memory now ends in the right node.
2801 * This appears as if some memory moved from one node to another.
2803 * So until kernel starts supporting memory hotplug to memory-less
2804 * nodes, just prevent such attempts upfront in QEMU.
2806 if (nb_numa_nodes && !numa_info[node].node_mem) {
2807 error_setg(errp, "Can't hotplug memory to memory-less node %d",
2812 spapr_memory_plug(hotplug_dev, dev, node, errp);
2813 } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
2814 spapr_core_plug(hotplug_dev, dev, errp);
2818 static void spapr_machine_device_unplug(HotplugHandler *hotplug_dev,
2819 DeviceState *dev, Error **errp)
2821 sPAPRMachineState *sms = SPAPR_MACHINE(qdev_get_machine());
2822 MachineClass *mc = MACHINE_GET_CLASS(qdev_get_machine());
2824 if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
2825 if (spapr_ovec_test(sms->ov5_cas, OV5_HP_EVT)) {
2826 spapr_memory_unplug(hotplug_dev, dev, errp);
2828 error_setg(errp, "Memory hot unplug not supported for this guest");
2830 } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
2831 if (!mc->has_hotpluggable_cpus) {
2832 error_setg(errp, "CPU hot unplug not supported on this machine");
2835 spapr_core_unplug(hotplug_dev, dev, errp);
2839 static void spapr_machine_device_unplug_request(HotplugHandler *hotplug_dev,
2840 DeviceState *dev, Error **errp)
2842 sPAPRMachineState *sms = SPAPR_MACHINE(qdev_get_machine());
2843 MachineClass *mc = MACHINE_GET_CLASS(qdev_get_machine());
2845 if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
2846 if (spapr_ovec_test(sms->ov5_cas, OV5_HP_EVT)) {
2847 spapr_memory_unplug_request(hotplug_dev, dev, errp);
2849 /* NOTE: this means there is a window after guest reset, prior to
2850 * CAS negotiation, where unplug requests will fail due to the
2851 * capability not being detected yet. This is a bit different than
2852 * the case with PCI unplug, where the events will be queued and
2853 * eventually handled by the guest after boot
2855 error_setg(errp, "Memory hot unplug not supported for this guest");
2857 } else if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
2858 if (!mc->has_hotpluggable_cpus) {
2859 error_setg(errp, "CPU hot unplug not supported on this machine");
2862 spapr_core_unplug_request(hotplug_dev, dev, errp);
2866 static void spapr_machine_device_pre_plug(HotplugHandler *hotplug_dev,
2867 DeviceState *dev, Error **errp)
2869 if (object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
2870 spapr_core_pre_plug(hotplug_dev, dev, errp);
2874 static HotplugHandler *spapr_get_hotplug_handler(MachineState *machine,
2877 if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM) ||
2878 object_dynamic_cast(OBJECT(dev), TYPE_SPAPR_CPU_CORE)) {
2879 return HOTPLUG_HANDLER(machine);
2884 static unsigned spapr_cpu_index_to_socket_id(unsigned cpu_index)
2886 /* Allocate to NUMA nodes on a "socket" basis (not that concept of
2887 * socket means much for the paravirtualized PAPR platform) */
2888 return cpu_index / smp_threads / smp_cores;
2891 static const CPUArchIdList *spapr_possible_cpu_arch_ids(MachineState *machine)
2894 int spapr_max_cores = max_cpus / smp_threads;
2895 MachineClass *mc = MACHINE_GET_CLASS(machine);
2897 if (!mc->has_hotpluggable_cpus) {
2898 spapr_max_cores = QEMU_ALIGN_UP(smp_cpus, smp_threads) / smp_threads;
2900 if (machine->possible_cpus) {
2901 assert(machine->possible_cpus->len == spapr_max_cores);
2902 return machine->possible_cpus;
2905 machine->possible_cpus = g_malloc0(sizeof(CPUArchIdList) +
2906 sizeof(CPUArchId) * spapr_max_cores);
2907 machine->possible_cpus->len = spapr_max_cores;
2908 for (i = 0; i < machine->possible_cpus->len; i++) {
2909 int core_id = i * smp_threads;
2911 machine->possible_cpus->cpus[i].vcpus_count = smp_threads;
2912 machine->possible_cpus->cpus[i].arch_id = core_id;
2913 machine->possible_cpus->cpus[i].props.has_core_id = true;
2914 machine->possible_cpus->cpus[i].props.core_id = core_id;
2915 /* TODO: add 'has_node/node' here to describe
2916 to which node core belongs */
2918 return machine->possible_cpus;
2921 static void spapr_phb_placement(sPAPRMachineState *spapr, uint32_t index,
2922 uint64_t *buid, hwaddr *pio,
2923 hwaddr *mmio32, hwaddr *mmio64,
2924 unsigned n_dma, uint32_t *liobns, Error **errp)
2927 * New-style PHB window placement.
2929 * Goals: Gives large (1TiB), naturally aligned 64-bit MMIO window
2930 * for each PHB, in addition to 2GiB 32-bit MMIO and 64kiB PIO
2933 * Some guest kernels can't work with MMIO windows above 1<<46
2934 * (64TiB), so we place up to 31 PHBs in the area 32TiB..64TiB
2936 * 32TiB..(33TiB+1984kiB) contains the 64kiB PIO windows for each
2937 * PHB stacked together. (32TiB+2GiB)..(32TiB+64GiB) contains the
2938 * 2GiB 32-bit MMIO windows for each PHB. Then 33..64TiB has the
2939 * 1TiB 64-bit MMIO windows for each PHB.
2941 const uint64_t base_buid = 0x800000020000000ULL;
2942 #define SPAPR_MAX_PHBS ((SPAPR_PCI_LIMIT - SPAPR_PCI_BASE) / \
2943 SPAPR_PCI_MEM64_WIN_SIZE - 1)
2946 /* Sanity check natural alignments */
2947 QEMU_BUILD_BUG_ON((SPAPR_PCI_BASE % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
2948 QEMU_BUILD_BUG_ON((SPAPR_PCI_LIMIT % SPAPR_PCI_MEM64_WIN_SIZE) != 0);
2949 QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM64_WIN_SIZE % SPAPR_PCI_MEM32_WIN_SIZE) != 0);
2950 QEMU_BUILD_BUG_ON((SPAPR_PCI_MEM32_WIN_SIZE % SPAPR_PCI_IO_WIN_SIZE) != 0);
2951 /* Sanity check bounds */
2952 QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_IO_WIN_SIZE) >
2953 SPAPR_PCI_MEM32_WIN_SIZE);
2954 QEMU_BUILD_BUG_ON((SPAPR_MAX_PHBS * SPAPR_PCI_MEM32_WIN_SIZE) >
2955 SPAPR_PCI_MEM64_WIN_SIZE);
2957 if (index >= SPAPR_MAX_PHBS) {
2958 error_setg(errp, "\"index\" for PAPR PHB is too large (max %llu)",
2959 SPAPR_MAX_PHBS - 1);
2963 *buid = base_buid + index;
2964 for (i = 0; i < n_dma; ++i) {
2965 liobns[i] = SPAPR_PCI_LIOBN(index, i);
2968 *pio = SPAPR_PCI_BASE + index * SPAPR_PCI_IO_WIN_SIZE;
2969 *mmio32 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM32_WIN_SIZE;
2970 *mmio64 = SPAPR_PCI_BASE + (index + 1) * SPAPR_PCI_MEM64_WIN_SIZE;
2973 static ICSState *spapr_ics_get(XICSFabric *dev, int irq)
2975 sPAPRMachineState *spapr = SPAPR_MACHINE(dev);
2977 return ics_valid_irq(spapr->ics, irq) ? spapr->ics : NULL;
2980 static void spapr_ics_resend(XICSFabric *dev)
2982 sPAPRMachineState *spapr = SPAPR_MACHINE(dev);
2984 ics_resend(spapr->ics);
2987 static void spapr_machine_class_init(ObjectClass *oc, void *data)
2989 MachineClass *mc = MACHINE_CLASS(oc);
2990 sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(oc);
2991 FWPathProviderClass *fwc = FW_PATH_PROVIDER_CLASS(oc);
2992 NMIClass *nc = NMI_CLASS(oc);
2993 HotplugHandlerClass *hc = HOTPLUG_HANDLER_CLASS(oc);
2994 PPCVirtualHypervisorClass *vhc = PPC_VIRTUAL_HYPERVISOR_CLASS(oc);
2995 XICSFabricClass *xic = XICS_FABRIC_CLASS(oc);
2997 mc->desc = "pSeries Logical Partition (PAPR compliant)";
3000 * We set up the default / latest behaviour here. The class_init
3001 * functions for the specific versioned machine types can override
3002 * these details for backwards compatibility
3004 mc->init = ppc_spapr_init;
3005 mc->reset = ppc_spapr_reset;
3006 mc->block_default_type = IF_SCSI;
3007 mc->max_cpus = 1024;
3008 mc->no_parallel = 1;
3009 mc->default_boot_order = "";
3010 mc->default_ram_size = 512 * M_BYTE;
3011 mc->kvm_type = spapr_kvm_type;
3012 mc->has_dynamic_sysbus = true;
3013 mc->pci_allow_0_address = true;
3014 mc->get_hotplug_handler = spapr_get_hotplug_handler;
3015 hc->pre_plug = spapr_machine_device_pre_plug;
3016 hc->plug = spapr_machine_device_plug;
3017 hc->unplug = spapr_machine_device_unplug;
3018 mc->cpu_index_to_socket_id = spapr_cpu_index_to_socket_id;
3019 mc->possible_cpu_arch_ids = spapr_possible_cpu_arch_ids;
3020 hc->unplug_request = spapr_machine_device_unplug_request;
3022 smc->dr_lmb_enabled = true;
3023 smc->tcg_default_cpu = "POWER8";
3024 mc->has_hotpluggable_cpus = true;
3025 fwc->get_dev_path = spapr_get_fw_dev_path;
3026 nc->nmi_monitor_handler = spapr_nmi;
3027 smc->phb_placement = spapr_phb_placement;
3028 vhc->hypercall = emulate_spapr_hypercall;
3029 vhc->hpt_mask = spapr_hpt_mask;
3030 vhc->map_hptes = spapr_map_hptes;
3031 vhc->unmap_hptes = spapr_unmap_hptes;
3032 vhc->store_hpte = spapr_store_hpte;
3033 xic->ics_get = spapr_ics_get;
3034 xic->ics_resend = spapr_ics_resend;
3037 static const TypeInfo spapr_machine_info = {
3038 .name = TYPE_SPAPR_MACHINE,
3039 .parent = TYPE_MACHINE,
3041 .instance_size = sizeof(sPAPRMachineState),
3042 .instance_init = spapr_machine_initfn,
3043 .instance_finalize = spapr_machine_finalizefn,
3044 .class_size = sizeof(sPAPRMachineClass),
3045 .class_init = spapr_machine_class_init,
3046 .interfaces = (InterfaceInfo[]) {
3047 { TYPE_FW_PATH_PROVIDER },
3049 { TYPE_HOTPLUG_HANDLER },
3050 { TYPE_PPC_VIRTUAL_HYPERVISOR },
3051 { TYPE_XICS_FABRIC },
3056 #define DEFINE_SPAPR_MACHINE(suffix, verstr, latest) \
3057 static void spapr_machine_##suffix##_class_init(ObjectClass *oc, \
3060 MachineClass *mc = MACHINE_CLASS(oc); \
3061 spapr_machine_##suffix##_class_options(mc); \
3063 mc->alias = "pseries"; \
3064 mc->is_default = 1; \
3067 static void spapr_machine_##suffix##_instance_init(Object *obj) \
3069 MachineState *machine = MACHINE(obj); \
3070 spapr_machine_##suffix##_instance_options(machine); \
3072 static const TypeInfo spapr_machine_##suffix##_info = { \
3073 .name = MACHINE_TYPE_NAME("pseries-" verstr), \
3074 .parent = TYPE_SPAPR_MACHINE, \
3075 .class_init = spapr_machine_##suffix##_class_init, \
3076 .instance_init = spapr_machine_##suffix##_instance_init, \
3078 static void spapr_machine_register_##suffix(void) \
3080 type_register(&spapr_machine_##suffix##_info); \
3082 type_init(spapr_machine_register_##suffix)
3087 static void spapr_machine_2_9_instance_options(MachineState *machine)
3091 static void spapr_machine_2_9_class_options(MachineClass *mc)
3093 /* Defaults for the latest behaviour inherited from the base class */
3096 DEFINE_SPAPR_MACHINE(2_9, "2.9", true);
3101 #define SPAPR_COMPAT_2_8 \
3104 static void spapr_machine_2_8_instance_options(MachineState *machine)
3106 spapr_machine_2_9_instance_options(machine);
3109 static void spapr_machine_2_8_class_options(MachineClass *mc)
3111 spapr_machine_2_9_class_options(mc);
3112 SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_8);
3115 DEFINE_SPAPR_MACHINE(2_8, "2.8", false);
3120 #define SPAPR_COMPAT_2_7 \
3123 .driver = TYPE_SPAPR_PCI_HOST_BRIDGE, \
3124 .property = "mem_win_size", \
3125 .value = stringify(SPAPR_PCI_2_7_MMIO_WIN_SIZE),\
3128 .driver = TYPE_SPAPR_PCI_HOST_BRIDGE, \
3129 .property = "mem64_win_size", \
3133 .driver = TYPE_POWERPC_CPU, \
3134 .property = "pre-2.8-migration", \
3138 .driver = TYPE_SPAPR_PCI_HOST_BRIDGE, \
3139 .property = "pre-2.8-migration", \
3143 static void phb_placement_2_7(sPAPRMachineState *spapr, uint32_t index,
3144 uint64_t *buid, hwaddr *pio,
3145 hwaddr *mmio32, hwaddr *mmio64,
3146 unsigned n_dma, uint32_t *liobns, Error **errp)
3148 /* Legacy PHB placement for pseries-2.7 and earlier machine types */
3149 const uint64_t base_buid = 0x800000020000000ULL;
3150 const hwaddr phb_spacing = 0x1000000000ULL; /* 64 GiB */
3151 const hwaddr mmio_offset = 0xa0000000; /* 2 GiB + 512 MiB */
3152 const hwaddr pio_offset = 0x80000000; /* 2 GiB */
3153 const uint32_t max_index = 255;
3154 const hwaddr phb0_alignment = 0x10000000000ULL; /* 1 TiB */
3156 uint64_t ram_top = MACHINE(spapr)->ram_size;
3157 hwaddr phb0_base, phb_base;
3160 /* Do we have hotpluggable memory? */
3161 if (MACHINE(spapr)->maxram_size > ram_top) {
3162 /* Can't just use maxram_size, because there may be an
3163 * alignment gap between normal and hotpluggable memory
3165 ram_top = spapr->hotplug_memory.base +
3166 memory_region_size(&spapr->hotplug_memory.mr);
3169 phb0_base = QEMU_ALIGN_UP(ram_top, phb0_alignment);
3171 if (index > max_index) {
3172 error_setg(errp, "\"index\" for PAPR PHB is too large (max %u)",
3177 *buid = base_buid + index;
3178 for (i = 0; i < n_dma; ++i) {
3179 liobns[i] = SPAPR_PCI_LIOBN(index, i);
3182 phb_base = phb0_base + index * phb_spacing;
3183 *pio = phb_base + pio_offset;
3184 *mmio32 = phb_base + mmio_offset;
3186 * We don't set the 64-bit MMIO window, relying on the PHB's
3187 * fallback behaviour of automatically splitting a large "32-bit"
3188 * window into contiguous 32-bit and 64-bit windows
3192 static void spapr_machine_2_7_instance_options(MachineState *machine)
3194 sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
3196 spapr_machine_2_8_instance_options(machine);
3197 spapr->use_hotplug_event_source = false;
3200 static void spapr_machine_2_7_class_options(MachineClass *mc)
3202 sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3204 spapr_machine_2_8_class_options(mc);
3205 smc->tcg_default_cpu = "POWER7";
3206 SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_7);
3207 smc->phb_placement = phb_placement_2_7;
3210 DEFINE_SPAPR_MACHINE(2_7, "2.7", false);
3215 #define SPAPR_COMPAT_2_6 \
3218 .driver = TYPE_SPAPR_PCI_HOST_BRIDGE,\
3220 .value = stringify(off),\
3223 static void spapr_machine_2_6_instance_options(MachineState *machine)
3225 spapr_machine_2_7_instance_options(machine);
3228 static void spapr_machine_2_6_class_options(MachineClass *mc)
3230 spapr_machine_2_7_class_options(mc);
3231 mc->has_hotpluggable_cpus = false;
3232 SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_6);
3235 DEFINE_SPAPR_MACHINE(2_6, "2.6", false);
3240 #define SPAPR_COMPAT_2_5 \
3243 .driver = "spapr-vlan", \
3244 .property = "use-rx-buffer-pools", \
3248 static void spapr_machine_2_5_instance_options(MachineState *machine)
3250 spapr_machine_2_6_instance_options(machine);
3253 static void spapr_machine_2_5_class_options(MachineClass *mc)
3255 sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3257 spapr_machine_2_6_class_options(mc);
3258 smc->use_ohci_by_default = true;
3259 SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_5);
3262 DEFINE_SPAPR_MACHINE(2_5, "2.5", false);
3267 #define SPAPR_COMPAT_2_4 \
3270 static void spapr_machine_2_4_instance_options(MachineState *machine)
3272 spapr_machine_2_5_instance_options(machine);
3275 static void spapr_machine_2_4_class_options(MachineClass *mc)
3277 sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
3279 spapr_machine_2_5_class_options(mc);
3280 smc->dr_lmb_enabled = false;
3281 SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_4);
3284 DEFINE_SPAPR_MACHINE(2_4, "2.4", false);
3289 #define SPAPR_COMPAT_2_3 \
3292 .driver = "spapr-pci-host-bridge",\
3293 .property = "dynamic-reconfiguration",\
3297 static void spapr_machine_2_3_instance_options(MachineState *machine)
3299 spapr_machine_2_4_instance_options(machine);
3300 savevm_skip_section_footers();
3301 global_state_set_optional();
3302 savevm_skip_configuration();
3305 static void spapr_machine_2_3_class_options(MachineClass *mc)
3307 spapr_machine_2_4_class_options(mc);
3308 SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_3);
3310 DEFINE_SPAPR_MACHINE(2_3, "2.3", false);
3316 #define SPAPR_COMPAT_2_2 \
3319 .driver = TYPE_SPAPR_PCI_HOST_BRIDGE,\
3320 .property = "mem_win_size",\
3321 .value = "0x20000000",\
3324 static void spapr_machine_2_2_instance_options(MachineState *machine)
3326 spapr_machine_2_3_instance_options(machine);
3327 machine->suppress_vmdesc = true;
3330 static void spapr_machine_2_2_class_options(MachineClass *mc)
3332 spapr_machine_2_3_class_options(mc);
3333 SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_2);
3335 DEFINE_SPAPR_MACHINE(2_2, "2.2", false);
3340 #define SPAPR_COMPAT_2_1 \
3343 static void spapr_machine_2_1_instance_options(MachineState *machine)
3345 spapr_machine_2_2_instance_options(machine);
3348 static void spapr_machine_2_1_class_options(MachineClass *mc)
3350 spapr_machine_2_2_class_options(mc);
3351 SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_1);
3353 DEFINE_SPAPR_MACHINE(2_1, "2.1", false);
3355 static void spapr_machine_register_types(void)
3357 type_register_static(&spapr_machine_info);
3360 type_init(spapr_machine_register_types)