spapr: Move spapr_cpu_init() to spapr_cpu_core.c
[qemu.git] / hw / ppc / spapr.c
blobb04a3892ea67f824d7ed3d87ec6ec141e433a36d
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/kvm.h"
40 #include "sysemu/device_tree.h"
41 #include "kvm_ppc.h"
42 #include "migration/migration.h"
43 #include "mmu-hash64.h"
44 #include "qom/cpu.h"
46 #include "hw/boards.h"
47 #include "hw/ppc/ppc.h"
48 #include "hw/loader.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"
70 #include <libfdt.h>
72 /* SLOF memory layout:
74 * SLOF raw image loaded at 0, copies its romfs right below the flat
75 * device-tree, then position SLOF itself 31M below that
77 * So we set FW_OVERHEAD to 40MB which should account for all of that
78 * and more
80 * We load our kernel at 4M, leaving space for SLOF initial image
82 #define FDT_MAX_SIZE 0x100000
83 #define RTAS_MAX_SIZE 0x10000
84 #define RTAS_MAX_ADDR 0x80000000 /* RTAS must stay below that */
85 #define FW_MAX_SIZE 0x400000
86 #define FW_FILE_NAME "slof.bin"
87 #define FW_OVERHEAD 0x2800000
88 #define KERNEL_LOAD_ADDR FW_MAX_SIZE
90 #define MIN_RMA_SLOF 128UL
92 #define PHANDLE_XICP 0x00001111
94 #define HTAB_SIZE(spapr) (1ULL << ((spapr)->htab_shift))
96 static XICSState *try_create_xics(const char *type, int nr_servers,
97 int nr_irqs, Error **errp)
99 Error *err = NULL;
100 DeviceState *dev;
102 dev = qdev_create(NULL, type);
103 qdev_prop_set_uint32(dev, "nr_servers", nr_servers);
104 qdev_prop_set_uint32(dev, "nr_irqs", nr_irqs);
105 object_property_set_bool(OBJECT(dev), true, "realized", &err);
106 if (err) {
107 error_propagate(errp, err);
108 object_unparent(OBJECT(dev));
109 return NULL;
111 return XICS_COMMON(dev);
114 static XICSState *xics_system_init(MachineState *machine,
115 int nr_servers, int nr_irqs, Error **errp)
117 XICSState *icp = NULL;
119 if (kvm_enabled()) {
120 Error *err = NULL;
122 if (machine_kernel_irqchip_allowed(machine)) {
123 icp = try_create_xics(TYPE_KVM_XICS, nr_servers, nr_irqs, &err);
125 if (machine_kernel_irqchip_required(machine) && !icp) {
126 error_reportf_err(err,
127 "kernel_irqchip requested but unavailable: ");
128 } else {
129 error_free(err);
133 if (!icp) {
134 icp = try_create_xics(TYPE_XICS, nr_servers, nr_irqs, errp);
137 return icp;
140 static int spapr_fixup_cpu_smt_dt(void *fdt, int offset, PowerPCCPU *cpu,
141 int smt_threads)
143 int i, ret = 0;
144 uint32_t servers_prop[smt_threads];
145 uint32_t gservers_prop[smt_threads * 2];
146 int index = ppc_get_vcpu_dt_id(cpu);
148 if (cpu->cpu_version) {
149 ret = fdt_setprop_cell(fdt, offset, "cpu-version", cpu->cpu_version);
150 if (ret < 0) {
151 return ret;
155 /* Build interrupt servers and gservers properties */
156 for (i = 0; i < smt_threads; i++) {
157 servers_prop[i] = cpu_to_be32(index + i);
158 /* Hack, direct the group queues back to cpu 0 */
159 gservers_prop[i*2] = cpu_to_be32(index + i);
160 gservers_prop[i*2 + 1] = 0;
162 ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-server#s",
163 servers_prop, sizeof(servers_prop));
164 if (ret < 0) {
165 return ret;
167 ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-gserver#s",
168 gservers_prop, sizeof(gservers_prop));
170 return ret;
173 static int spapr_fixup_cpu_numa_dt(void *fdt, int offset, CPUState *cs)
175 int ret = 0;
176 PowerPCCPU *cpu = POWERPC_CPU(cs);
177 int index = ppc_get_vcpu_dt_id(cpu);
178 uint32_t associativity[] = {cpu_to_be32(0x5),
179 cpu_to_be32(0x0),
180 cpu_to_be32(0x0),
181 cpu_to_be32(0x0),
182 cpu_to_be32(cs->numa_node),
183 cpu_to_be32(index)};
185 /* Advertise NUMA via ibm,associativity */
186 if (nb_numa_nodes > 1) {
187 ret = fdt_setprop(fdt, offset, "ibm,associativity", associativity,
188 sizeof(associativity));
191 return ret;
194 static int spapr_fixup_cpu_dt(void *fdt, sPAPRMachineState *spapr)
196 int ret = 0, offset, cpus_offset;
197 CPUState *cs;
198 char cpu_model[32];
199 int smt = kvmppc_smt_threads();
200 uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
202 CPU_FOREACH(cs) {
203 PowerPCCPU *cpu = POWERPC_CPU(cs);
204 DeviceClass *dc = DEVICE_GET_CLASS(cs);
205 int index = ppc_get_vcpu_dt_id(cpu);
207 if ((index % smt) != 0) {
208 continue;
211 snprintf(cpu_model, 32, "%s@%x", dc->fw_name, index);
213 cpus_offset = fdt_path_offset(fdt, "/cpus");
214 if (cpus_offset < 0) {
215 cpus_offset = fdt_add_subnode(fdt, fdt_path_offset(fdt, "/"),
216 "cpus");
217 if (cpus_offset < 0) {
218 return cpus_offset;
221 offset = fdt_subnode_offset(fdt, cpus_offset, cpu_model);
222 if (offset < 0) {
223 offset = fdt_add_subnode(fdt, cpus_offset, cpu_model);
224 if (offset < 0) {
225 return offset;
229 ret = fdt_setprop(fdt, offset, "ibm,pft-size",
230 pft_size_prop, sizeof(pft_size_prop));
231 if (ret < 0) {
232 return ret;
235 ret = spapr_fixup_cpu_numa_dt(fdt, offset, cs);
236 if (ret < 0) {
237 return ret;
240 ret = spapr_fixup_cpu_smt_dt(fdt, offset, cpu,
241 ppc_get_compat_smt_threads(cpu));
242 if (ret < 0) {
243 return ret;
246 return ret;
250 static size_t create_page_sizes_prop(CPUPPCState *env, uint32_t *prop,
251 size_t maxsize)
253 size_t maxcells = maxsize / sizeof(uint32_t);
254 int i, j, count;
255 uint32_t *p = prop;
257 for (i = 0; i < PPC_PAGE_SIZES_MAX_SZ; i++) {
258 struct ppc_one_seg_page_size *sps = &env->sps.sps[i];
260 if (!sps->page_shift) {
261 break;
263 for (count = 0; count < PPC_PAGE_SIZES_MAX_SZ; count++) {
264 if (sps->enc[count].page_shift == 0) {
265 break;
268 if ((p - prop) >= (maxcells - 3 - count * 2)) {
269 break;
271 *(p++) = cpu_to_be32(sps->page_shift);
272 *(p++) = cpu_to_be32(sps->slb_enc);
273 *(p++) = cpu_to_be32(count);
274 for (j = 0; j < count; j++) {
275 *(p++) = cpu_to_be32(sps->enc[j].page_shift);
276 *(p++) = cpu_to_be32(sps->enc[j].pte_enc);
280 return (p - prop) * sizeof(uint32_t);
283 static hwaddr spapr_node0_size(void)
285 MachineState *machine = MACHINE(qdev_get_machine());
287 if (nb_numa_nodes) {
288 int i;
289 for (i = 0; i < nb_numa_nodes; ++i) {
290 if (numa_info[i].node_mem) {
291 return MIN(pow2floor(numa_info[i].node_mem),
292 machine->ram_size);
296 return machine->ram_size;
299 #define _FDT(exp) \
300 do { \
301 int ret = (exp); \
302 if (ret < 0) { \
303 fprintf(stderr, "qemu: error creating device tree: %s: %s\n", \
304 #exp, fdt_strerror(ret)); \
305 exit(1); \
307 } while (0)
309 static void add_str(GString *s, const gchar *s1)
311 g_string_append_len(s, s1, strlen(s1) + 1);
314 static void *spapr_create_fdt_skel(hwaddr initrd_base,
315 hwaddr initrd_size,
316 hwaddr kernel_size,
317 bool little_endian,
318 const char *kernel_cmdline,
319 uint32_t epow_irq)
321 void *fdt;
322 uint32_t start_prop = cpu_to_be32(initrd_base);
323 uint32_t end_prop = cpu_to_be32(initrd_base + initrd_size);
324 GString *hypertas = g_string_sized_new(256);
325 GString *qemu_hypertas = g_string_sized_new(256);
326 uint32_t refpoints[] = {cpu_to_be32(0x4), cpu_to_be32(0x4)};
327 uint32_t interrupt_server_ranges_prop[] = {0, cpu_to_be32(max_cpus)};
328 unsigned char vec5[] = {0x0, 0x0, 0x0, 0x0, 0x0, 0x80};
329 char *buf;
331 add_str(hypertas, "hcall-pft");
332 add_str(hypertas, "hcall-term");
333 add_str(hypertas, "hcall-dabr");
334 add_str(hypertas, "hcall-interrupt");
335 add_str(hypertas, "hcall-tce");
336 add_str(hypertas, "hcall-vio");
337 add_str(hypertas, "hcall-splpar");
338 add_str(hypertas, "hcall-bulk");
339 add_str(hypertas, "hcall-set-mode");
340 add_str(qemu_hypertas, "hcall-memop1");
342 fdt = g_malloc0(FDT_MAX_SIZE);
343 _FDT((fdt_create(fdt, FDT_MAX_SIZE)));
345 if (kernel_size) {
346 _FDT((fdt_add_reservemap_entry(fdt, KERNEL_LOAD_ADDR, kernel_size)));
348 if (initrd_size) {
349 _FDT((fdt_add_reservemap_entry(fdt, initrd_base, initrd_size)));
351 _FDT((fdt_finish_reservemap(fdt)));
353 /* Root node */
354 _FDT((fdt_begin_node(fdt, "")));
355 _FDT((fdt_property_string(fdt, "device_type", "chrp")));
356 _FDT((fdt_property_string(fdt, "model", "IBM pSeries (emulated by qemu)")));
357 _FDT((fdt_property_string(fdt, "compatible", "qemu,pseries")));
360 * Add info to guest to indentify which host is it being run on
361 * and what is the uuid of the guest
363 if (kvmppc_get_host_model(&buf)) {
364 _FDT((fdt_property_string(fdt, "host-model", buf)));
365 g_free(buf);
367 if (kvmppc_get_host_serial(&buf)) {
368 _FDT((fdt_property_string(fdt, "host-serial", buf)));
369 g_free(buf);
372 buf = g_strdup_printf(UUID_FMT, qemu_uuid[0], qemu_uuid[1],
373 qemu_uuid[2], qemu_uuid[3], qemu_uuid[4],
374 qemu_uuid[5], qemu_uuid[6], qemu_uuid[7],
375 qemu_uuid[8], qemu_uuid[9], qemu_uuid[10],
376 qemu_uuid[11], qemu_uuid[12], qemu_uuid[13],
377 qemu_uuid[14], qemu_uuid[15]);
379 _FDT((fdt_property_string(fdt, "vm,uuid", buf)));
380 if (qemu_uuid_set) {
381 _FDT((fdt_property_string(fdt, "system-id", buf)));
383 g_free(buf);
385 if (qemu_get_vm_name()) {
386 _FDT((fdt_property_string(fdt, "ibm,partition-name",
387 qemu_get_vm_name())));
390 _FDT((fdt_property_cell(fdt, "#address-cells", 0x2)));
391 _FDT((fdt_property_cell(fdt, "#size-cells", 0x2)));
393 /* /chosen */
394 _FDT((fdt_begin_node(fdt, "chosen")));
396 /* Set Form1_affinity */
397 _FDT((fdt_property(fdt, "ibm,architecture-vec-5", vec5, sizeof(vec5))));
399 _FDT((fdt_property_string(fdt, "bootargs", kernel_cmdline)));
400 _FDT((fdt_property(fdt, "linux,initrd-start",
401 &start_prop, sizeof(start_prop))));
402 _FDT((fdt_property(fdt, "linux,initrd-end",
403 &end_prop, sizeof(end_prop))));
404 if (kernel_size) {
405 uint64_t kprop[2] = { cpu_to_be64(KERNEL_LOAD_ADDR),
406 cpu_to_be64(kernel_size) };
408 _FDT((fdt_property(fdt, "qemu,boot-kernel", &kprop, sizeof(kprop))));
409 if (little_endian) {
410 _FDT((fdt_property(fdt, "qemu,boot-kernel-le", NULL, 0)));
413 if (boot_menu) {
414 _FDT((fdt_property_cell(fdt, "qemu,boot-menu", boot_menu)));
416 _FDT((fdt_property_cell(fdt, "qemu,graphic-width", graphic_width)));
417 _FDT((fdt_property_cell(fdt, "qemu,graphic-height", graphic_height)));
418 _FDT((fdt_property_cell(fdt, "qemu,graphic-depth", graphic_depth)));
420 _FDT((fdt_end_node(fdt)));
422 /* RTAS */
423 _FDT((fdt_begin_node(fdt, "rtas")));
425 if (!kvm_enabled() || kvmppc_spapr_use_multitce()) {
426 add_str(hypertas, "hcall-multi-tce");
428 _FDT((fdt_property(fdt, "ibm,hypertas-functions", hypertas->str,
429 hypertas->len)));
430 g_string_free(hypertas, TRUE);
431 _FDT((fdt_property(fdt, "qemu,hypertas-functions", qemu_hypertas->str,
432 qemu_hypertas->len)));
433 g_string_free(qemu_hypertas, TRUE);
435 _FDT((fdt_property(fdt, "ibm,associativity-reference-points",
436 refpoints, sizeof(refpoints))));
438 _FDT((fdt_property_cell(fdt, "rtas-error-log-max", RTAS_ERROR_LOG_MAX)));
439 _FDT((fdt_property_cell(fdt, "rtas-event-scan-rate",
440 RTAS_EVENT_SCAN_RATE)));
442 if (msi_nonbroken) {
443 _FDT((fdt_property(fdt, "ibm,change-msix-capable", NULL, 0)));
447 * According to PAPR, rtas ibm,os-term does not guarantee a return
448 * back to the guest cpu.
450 * While an additional ibm,extended-os-term property indicates that
451 * rtas call return will always occur. Set this property.
453 _FDT((fdt_property(fdt, "ibm,extended-os-term", NULL, 0)));
455 _FDT((fdt_end_node(fdt)));
457 /* interrupt controller */
458 _FDT((fdt_begin_node(fdt, "interrupt-controller")));
460 _FDT((fdt_property_string(fdt, "device_type",
461 "PowerPC-External-Interrupt-Presentation")));
462 _FDT((fdt_property_string(fdt, "compatible", "IBM,ppc-xicp")));
463 _FDT((fdt_property(fdt, "interrupt-controller", NULL, 0)));
464 _FDT((fdt_property(fdt, "ibm,interrupt-server-ranges",
465 interrupt_server_ranges_prop,
466 sizeof(interrupt_server_ranges_prop))));
467 _FDT((fdt_property_cell(fdt, "#interrupt-cells", 2)));
468 _FDT((fdt_property_cell(fdt, "linux,phandle", PHANDLE_XICP)));
469 _FDT((fdt_property_cell(fdt, "phandle", PHANDLE_XICP)));
471 _FDT((fdt_end_node(fdt)));
473 /* vdevice */
474 _FDT((fdt_begin_node(fdt, "vdevice")));
476 _FDT((fdt_property_string(fdt, "device_type", "vdevice")));
477 _FDT((fdt_property_string(fdt, "compatible", "IBM,vdevice")));
478 _FDT((fdt_property_cell(fdt, "#address-cells", 0x1)));
479 _FDT((fdt_property_cell(fdt, "#size-cells", 0x0)));
480 _FDT((fdt_property_cell(fdt, "#interrupt-cells", 0x2)));
481 _FDT((fdt_property(fdt, "interrupt-controller", NULL, 0)));
483 _FDT((fdt_end_node(fdt)));
485 /* event-sources */
486 spapr_events_fdt_skel(fdt, epow_irq);
488 /* /hypervisor node */
489 if (kvm_enabled()) {
490 uint8_t hypercall[16];
492 /* indicate KVM hypercall interface */
493 _FDT((fdt_begin_node(fdt, "hypervisor")));
494 _FDT((fdt_property_string(fdt, "compatible", "linux,kvm")));
495 if (kvmppc_has_cap_fixup_hcalls()) {
497 * Older KVM versions with older guest kernels were broken with the
498 * magic page, don't allow the guest to map it.
500 if (!kvmppc_get_hypercall(first_cpu->env_ptr, hypercall,
501 sizeof(hypercall))) {
502 _FDT((fdt_property(fdt, "hcall-instructions", hypercall,
503 sizeof(hypercall))));
506 _FDT((fdt_end_node(fdt)));
509 _FDT((fdt_end_node(fdt))); /* close root node */
510 _FDT((fdt_finish(fdt)));
512 return fdt;
515 static int spapr_populate_memory_node(void *fdt, int nodeid, hwaddr start,
516 hwaddr size)
518 uint32_t associativity[] = {
519 cpu_to_be32(0x4), /* length */
520 cpu_to_be32(0x0), cpu_to_be32(0x0),
521 cpu_to_be32(0x0), cpu_to_be32(nodeid)
523 char mem_name[32];
524 uint64_t mem_reg_property[2];
525 int off;
527 mem_reg_property[0] = cpu_to_be64(start);
528 mem_reg_property[1] = cpu_to_be64(size);
530 sprintf(mem_name, "memory@" TARGET_FMT_lx, start);
531 off = fdt_add_subnode(fdt, 0, mem_name);
532 _FDT(off);
533 _FDT((fdt_setprop_string(fdt, off, "device_type", "memory")));
534 _FDT((fdt_setprop(fdt, off, "reg", mem_reg_property,
535 sizeof(mem_reg_property))));
536 _FDT((fdt_setprop(fdt, off, "ibm,associativity", associativity,
537 sizeof(associativity))));
538 return off;
541 static int spapr_populate_memory(sPAPRMachineState *spapr, void *fdt)
543 MachineState *machine = MACHINE(spapr);
544 hwaddr mem_start, node_size;
545 int i, nb_nodes = nb_numa_nodes;
546 NodeInfo *nodes = numa_info;
547 NodeInfo ramnode;
549 /* No NUMA nodes, assume there is just one node with whole RAM */
550 if (!nb_numa_nodes) {
551 nb_nodes = 1;
552 ramnode.node_mem = machine->ram_size;
553 nodes = &ramnode;
556 for (i = 0, mem_start = 0; i < nb_nodes; ++i) {
557 if (!nodes[i].node_mem) {
558 continue;
560 if (mem_start >= machine->ram_size) {
561 node_size = 0;
562 } else {
563 node_size = nodes[i].node_mem;
564 if (node_size > machine->ram_size - mem_start) {
565 node_size = machine->ram_size - mem_start;
568 if (!mem_start) {
569 /* ppc_spapr_init() checks for rma_size <= node0_size already */
570 spapr_populate_memory_node(fdt, i, 0, spapr->rma_size);
571 mem_start += spapr->rma_size;
572 node_size -= spapr->rma_size;
574 for ( ; node_size; ) {
575 hwaddr sizetmp = pow2floor(node_size);
577 /* mem_start != 0 here */
578 if (ctzl(mem_start) < ctzl(sizetmp)) {
579 sizetmp = 1ULL << ctzl(mem_start);
582 spapr_populate_memory_node(fdt, i, mem_start, sizetmp);
583 node_size -= sizetmp;
584 mem_start += sizetmp;
588 return 0;
591 static void spapr_populate_cpu_dt(CPUState *cs, void *fdt, int offset,
592 sPAPRMachineState *spapr)
594 PowerPCCPU *cpu = POWERPC_CPU(cs);
595 CPUPPCState *env = &cpu->env;
596 PowerPCCPUClass *pcc = POWERPC_CPU_GET_CLASS(cs);
597 int index = ppc_get_vcpu_dt_id(cpu);
598 uint32_t segs[] = {cpu_to_be32(28), cpu_to_be32(40),
599 0xffffffff, 0xffffffff};
600 uint32_t tbfreq = kvm_enabled() ? kvmppc_get_tbfreq()
601 : SPAPR_TIMEBASE_FREQ;
602 uint32_t cpufreq = kvm_enabled() ? kvmppc_get_clockfreq() : 1000000000;
603 uint32_t page_sizes_prop[64];
604 size_t page_sizes_prop_size;
605 uint32_t vcpus_per_socket = smp_threads * smp_cores;
606 uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
608 /* Note: we keep CI large pages off for now because a 64K capable guest
609 * provisioned with large pages might otherwise try to map a qemu
610 * framebuffer (or other kind of memory mapped PCI BAR) using 64K pages
611 * even if that qemu runs on a 4k host.
613 * We can later add this bit back when we are confident this is not
614 * an issue (!HV KVM or 64K host)
616 uint8_t pa_features_206[] = { 6, 0,
617 0xf6, 0x1f, 0xc7, 0x00, 0x80, 0xc0 };
618 uint8_t pa_features_207[] = { 24, 0,
619 0xf6, 0x1f, 0xc7, 0xc0, 0x80, 0xf0,
620 0x80, 0x00, 0x00, 0x00, 0x00, 0x00,
621 0x00, 0x00, 0x00, 0x00, 0x80, 0x00,
622 0x80, 0x00, 0x80, 0x00, 0x80, 0x00 };
623 uint8_t *pa_features;
624 size_t pa_size;
626 _FDT((fdt_setprop_cell(fdt, offset, "reg", index)));
627 _FDT((fdt_setprop_string(fdt, offset, "device_type", "cpu")));
629 _FDT((fdt_setprop_cell(fdt, offset, "cpu-version", env->spr[SPR_PVR])));
630 _FDT((fdt_setprop_cell(fdt, offset, "d-cache-block-size",
631 env->dcache_line_size)));
632 _FDT((fdt_setprop_cell(fdt, offset, "d-cache-line-size",
633 env->dcache_line_size)));
634 _FDT((fdt_setprop_cell(fdt, offset, "i-cache-block-size",
635 env->icache_line_size)));
636 _FDT((fdt_setprop_cell(fdt, offset, "i-cache-line-size",
637 env->icache_line_size)));
639 if (pcc->l1_dcache_size) {
640 _FDT((fdt_setprop_cell(fdt, offset, "d-cache-size",
641 pcc->l1_dcache_size)));
642 } else {
643 fprintf(stderr, "Warning: Unknown L1 dcache size for cpu\n");
645 if (pcc->l1_icache_size) {
646 _FDT((fdt_setprop_cell(fdt, offset, "i-cache-size",
647 pcc->l1_icache_size)));
648 } else {
649 fprintf(stderr, "Warning: Unknown L1 icache size for cpu\n");
652 _FDT((fdt_setprop_cell(fdt, offset, "timebase-frequency", tbfreq)));
653 _FDT((fdt_setprop_cell(fdt, offset, "clock-frequency", cpufreq)));
654 _FDT((fdt_setprop_cell(fdt, offset, "slb-size", env->slb_nr)));
655 _FDT((fdt_setprop_cell(fdt, offset, "ibm,slb-size", env->slb_nr)));
656 _FDT((fdt_setprop_string(fdt, offset, "status", "okay")));
657 _FDT((fdt_setprop(fdt, offset, "64-bit", NULL, 0)));
659 if (env->spr_cb[SPR_PURR].oea_read) {
660 _FDT((fdt_setprop(fdt, offset, "ibm,purr", NULL, 0)));
663 if (env->mmu_model & POWERPC_MMU_1TSEG) {
664 _FDT((fdt_setprop(fdt, offset, "ibm,processor-segment-sizes",
665 segs, sizeof(segs))));
668 /* Advertise VMX/VSX (vector extensions) if available
669 * 0 / no property == no vector extensions
670 * 1 == VMX / Altivec available
671 * 2 == VSX available */
672 if (env->insns_flags & PPC_ALTIVEC) {
673 uint32_t vmx = (env->insns_flags2 & PPC2_VSX) ? 2 : 1;
675 _FDT((fdt_setprop_cell(fdt, offset, "ibm,vmx", vmx)));
678 /* Advertise DFP (Decimal Floating Point) if available
679 * 0 / no property == no DFP
680 * 1 == DFP available */
681 if (env->insns_flags2 & PPC2_DFP) {
682 _FDT((fdt_setprop_cell(fdt, offset, "ibm,dfp", 1)));
685 page_sizes_prop_size = create_page_sizes_prop(env, page_sizes_prop,
686 sizeof(page_sizes_prop));
687 if (page_sizes_prop_size) {
688 _FDT((fdt_setprop(fdt, offset, "ibm,segment-page-sizes",
689 page_sizes_prop, page_sizes_prop_size)));
692 /* Do the ibm,pa-features property, adjust it for ci-large-pages */
693 if (env->mmu_model == POWERPC_MMU_2_06) {
694 pa_features = pa_features_206;
695 pa_size = sizeof(pa_features_206);
696 } else /* env->mmu_model == POWERPC_MMU_2_07 */ {
697 pa_features = pa_features_207;
698 pa_size = sizeof(pa_features_207);
700 if (env->ci_large_pages) {
701 pa_features[3] |= 0x20;
703 _FDT((fdt_setprop(fdt, offset, "ibm,pa-features", pa_features, pa_size)));
705 _FDT((fdt_setprop_cell(fdt, offset, "ibm,chip-id",
706 cs->cpu_index / vcpus_per_socket)));
708 _FDT((fdt_setprop(fdt, offset, "ibm,pft-size",
709 pft_size_prop, sizeof(pft_size_prop))));
711 _FDT(spapr_fixup_cpu_numa_dt(fdt, offset, cs));
713 _FDT(spapr_fixup_cpu_smt_dt(fdt, offset, cpu,
714 ppc_get_compat_smt_threads(cpu)));
717 static void spapr_populate_cpus_dt_node(void *fdt, sPAPRMachineState *spapr)
719 CPUState *cs;
720 int cpus_offset;
721 char *nodename;
722 int smt = kvmppc_smt_threads();
724 cpus_offset = fdt_add_subnode(fdt, 0, "cpus");
725 _FDT(cpus_offset);
726 _FDT((fdt_setprop_cell(fdt, cpus_offset, "#address-cells", 0x1)));
727 _FDT((fdt_setprop_cell(fdt, cpus_offset, "#size-cells", 0x0)));
730 * We walk the CPUs in reverse order to ensure that CPU DT nodes
731 * created by fdt_add_subnode() end up in the right order in FDT
732 * for the guest kernel the enumerate the CPUs correctly.
734 CPU_FOREACH_REVERSE(cs) {
735 PowerPCCPU *cpu = POWERPC_CPU(cs);
736 int index = ppc_get_vcpu_dt_id(cpu);
737 DeviceClass *dc = DEVICE_GET_CLASS(cs);
738 int offset;
740 if ((index % smt) != 0) {
741 continue;
744 nodename = g_strdup_printf("%s@%x", dc->fw_name, index);
745 offset = fdt_add_subnode(fdt, cpus_offset, nodename);
746 g_free(nodename);
747 _FDT(offset);
748 spapr_populate_cpu_dt(cs, fdt, offset, spapr);
754 * Adds ibm,dynamic-reconfiguration-memory node.
755 * Refer to docs/specs/ppc-spapr-hotplug.txt for the documentation
756 * of this device tree node.
758 static int spapr_populate_drconf_memory(sPAPRMachineState *spapr, void *fdt)
760 MachineState *machine = MACHINE(spapr);
761 int ret, i, offset;
762 uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
763 uint32_t prop_lmb_size[] = {0, cpu_to_be32(lmb_size)};
764 uint32_t hotplug_lmb_start = spapr->hotplug_memory.base / lmb_size;
765 uint32_t nr_lmbs = (spapr->hotplug_memory.base +
766 memory_region_size(&spapr->hotplug_memory.mr)) /
767 lmb_size;
768 uint32_t *int_buf, *cur_index, buf_len;
769 int nr_nodes = nb_numa_nodes ? nb_numa_nodes : 1;
772 * Don't create the node if there is no hotpluggable memory
774 if (machine->ram_size == machine->maxram_size) {
775 return 0;
779 * Allocate enough buffer size to fit in ibm,dynamic-memory
780 * or ibm,associativity-lookup-arrays
782 buf_len = MAX(nr_lmbs * SPAPR_DR_LMB_LIST_ENTRY_SIZE + 1, nr_nodes * 4 + 2)
783 * sizeof(uint32_t);
784 cur_index = int_buf = g_malloc0(buf_len);
786 offset = fdt_add_subnode(fdt, 0, "ibm,dynamic-reconfiguration-memory");
788 ret = fdt_setprop(fdt, offset, "ibm,lmb-size", prop_lmb_size,
789 sizeof(prop_lmb_size));
790 if (ret < 0) {
791 goto out;
794 ret = fdt_setprop_cell(fdt, offset, "ibm,memory-flags-mask", 0xff);
795 if (ret < 0) {
796 goto out;
799 ret = fdt_setprop_cell(fdt, offset, "ibm,memory-preservation-time", 0x0);
800 if (ret < 0) {
801 goto out;
804 /* ibm,dynamic-memory */
805 int_buf[0] = cpu_to_be32(nr_lmbs);
806 cur_index++;
807 for (i = 0; i < nr_lmbs; i++) {
808 uint64_t addr = i * lmb_size;
809 uint32_t *dynamic_memory = cur_index;
811 if (i >= hotplug_lmb_start) {
812 sPAPRDRConnector *drc;
813 sPAPRDRConnectorClass *drck;
815 drc = spapr_dr_connector_by_id(SPAPR_DR_CONNECTOR_TYPE_LMB, i);
816 g_assert(drc);
817 drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
819 dynamic_memory[0] = cpu_to_be32(addr >> 32);
820 dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
821 dynamic_memory[2] = cpu_to_be32(drck->get_index(drc));
822 dynamic_memory[3] = cpu_to_be32(0); /* reserved */
823 dynamic_memory[4] = cpu_to_be32(numa_get_node(addr, NULL));
824 if (memory_region_present(get_system_memory(), addr)) {
825 dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_ASSIGNED);
826 } else {
827 dynamic_memory[5] = cpu_to_be32(0);
829 } else {
831 * LMB information for RMA, boot time RAM and gap b/n RAM and
832 * hotplug memory region -- all these are marked as reserved
833 * and as having no valid DRC.
835 dynamic_memory[0] = cpu_to_be32(addr >> 32);
836 dynamic_memory[1] = cpu_to_be32(addr & 0xffffffff);
837 dynamic_memory[2] = cpu_to_be32(0);
838 dynamic_memory[3] = cpu_to_be32(0); /* reserved */
839 dynamic_memory[4] = cpu_to_be32(-1);
840 dynamic_memory[5] = cpu_to_be32(SPAPR_LMB_FLAGS_RESERVED |
841 SPAPR_LMB_FLAGS_DRC_INVALID);
844 cur_index += SPAPR_DR_LMB_LIST_ENTRY_SIZE;
846 ret = fdt_setprop(fdt, offset, "ibm,dynamic-memory", int_buf, buf_len);
847 if (ret < 0) {
848 goto out;
851 /* ibm,associativity-lookup-arrays */
852 cur_index = int_buf;
853 int_buf[0] = cpu_to_be32(nr_nodes);
854 int_buf[1] = cpu_to_be32(4); /* Number of entries per associativity list */
855 cur_index += 2;
856 for (i = 0; i < nr_nodes; i++) {
857 uint32_t associativity[] = {
858 cpu_to_be32(0x0),
859 cpu_to_be32(0x0),
860 cpu_to_be32(0x0),
861 cpu_to_be32(i)
863 memcpy(cur_index, associativity, sizeof(associativity));
864 cur_index += 4;
866 ret = fdt_setprop(fdt, offset, "ibm,associativity-lookup-arrays", int_buf,
867 (cur_index - int_buf) * sizeof(uint32_t));
868 out:
869 g_free(int_buf);
870 return ret;
873 int spapr_h_cas_compose_response(sPAPRMachineState *spapr,
874 target_ulong addr, target_ulong size,
875 bool cpu_update, bool memory_update)
877 void *fdt, *fdt_skel;
878 sPAPRDeviceTreeUpdateHeader hdr = { .version_id = 1 };
879 sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(qdev_get_machine());
881 size -= sizeof(hdr);
883 /* Create sceleton */
884 fdt_skel = g_malloc0(size);
885 _FDT((fdt_create(fdt_skel, size)));
886 _FDT((fdt_begin_node(fdt_skel, "")));
887 _FDT((fdt_end_node(fdt_skel)));
888 _FDT((fdt_finish(fdt_skel)));
889 fdt = g_malloc0(size);
890 _FDT((fdt_open_into(fdt_skel, fdt, size)));
891 g_free(fdt_skel);
893 /* Fixup cpu nodes */
894 if (cpu_update) {
895 _FDT((spapr_fixup_cpu_dt(fdt, spapr)));
898 /* Generate ibm,dynamic-reconfiguration-memory node if required */
899 if (memory_update && smc->dr_lmb_enabled) {
900 _FDT((spapr_populate_drconf_memory(spapr, fdt)));
903 /* Pack resulting tree */
904 _FDT((fdt_pack(fdt)));
906 if (fdt_totalsize(fdt) + sizeof(hdr) > size) {
907 trace_spapr_cas_failed(size);
908 return -1;
911 cpu_physical_memory_write(addr, &hdr, sizeof(hdr));
912 cpu_physical_memory_write(addr + sizeof(hdr), fdt, fdt_totalsize(fdt));
913 trace_spapr_cas_continue(fdt_totalsize(fdt) + sizeof(hdr));
914 g_free(fdt);
916 return 0;
919 static void spapr_finalize_fdt(sPAPRMachineState *spapr,
920 hwaddr fdt_addr,
921 hwaddr rtas_addr,
922 hwaddr rtas_size)
924 MachineState *machine = MACHINE(qdev_get_machine());
925 sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
926 const char *boot_device = machine->boot_order;
927 int ret, i;
928 size_t cb = 0;
929 char *bootlist;
930 void *fdt;
931 sPAPRPHBState *phb;
933 fdt = g_malloc(FDT_MAX_SIZE);
935 /* open out the base tree into a temp buffer for the final tweaks */
936 _FDT((fdt_open_into(spapr->fdt_skel, fdt, FDT_MAX_SIZE)));
938 ret = spapr_populate_memory(spapr, fdt);
939 if (ret < 0) {
940 fprintf(stderr, "couldn't setup memory nodes in fdt\n");
941 exit(1);
944 ret = spapr_populate_vdevice(spapr->vio_bus, fdt);
945 if (ret < 0) {
946 fprintf(stderr, "couldn't setup vio devices in fdt\n");
947 exit(1);
950 if (object_resolve_path_type("", TYPE_SPAPR_RNG, NULL)) {
951 ret = spapr_rng_populate_dt(fdt);
952 if (ret < 0) {
953 fprintf(stderr, "could not set up rng device in the fdt\n");
954 exit(1);
958 QLIST_FOREACH(phb, &spapr->phbs, list) {
959 ret = spapr_populate_pci_dt(phb, PHANDLE_XICP, fdt);
960 if (ret < 0) {
961 error_report("couldn't setup PCI devices in fdt");
962 exit(1);
966 /* RTAS */
967 ret = spapr_rtas_device_tree_setup(fdt, rtas_addr, rtas_size);
968 if (ret < 0) {
969 fprintf(stderr, "Couldn't set up RTAS device tree properties\n");
972 /* cpus */
973 spapr_populate_cpus_dt_node(fdt, spapr);
975 bootlist = get_boot_devices_list(&cb, true);
976 if (cb && bootlist) {
977 int offset = fdt_path_offset(fdt, "/chosen");
978 if (offset < 0) {
979 exit(1);
981 for (i = 0; i < cb; i++) {
982 if (bootlist[i] == '\n') {
983 bootlist[i] = ' ';
987 ret = fdt_setprop_string(fdt, offset, "qemu,boot-list", bootlist);
990 if (boot_device && strlen(boot_device)) {
991 int offset = fdt_path_offset(fdt, "/chosen");
993 if (offset < 0) {
994 exit(1);
996 fdt_setprop_string(fdt, offset, "qemu,boot-device", boot_device);
999 if (!spapr->has_graphics) {
1000 spapr_populate_chosen_stdout(fdt, spapr->vio_bus);
1003 if (smc->dr_lmb_enabled) {
1004 _FDT(spapr_drc_populate_dt(fdt, 0, NULL, SPAPR_DR_CONNECTOR_TYPE_LMB));
1007 _FDT((fdt_pack(fdt)));
1009 if (fdt_totalsize(fdt) > FDT_MAX_SIZE) {
1010 error_report("FDT too big ! 0x%x bytes (max is 0x%x)",
1011 fdt_totalsize(fdt), FDT_MAX_SIZE);
1012 exit(1);
1015 qemu_fdt_dumpdtb(fdt, fdt_totalsize(fdt));
1016 cpu_physical_memory_write(fdt_addr, fdt, fdt_totalsize(fdt));
1018 g_free(bootlist);
1019 g_free(fdt);
1022 static uint64_t translate_kernel_address(void *opaque, uint64_t addr)
1024 return (addr & 0x0fffffff) + KERNEL_LOAD_ADDR;
1027 static void emulate_spapr_hypercall(PowerPCCPU *cpu)
1029 CPUPPCState *env = &cpu->env;
1031 if (msr_pr) {
1032 hcall_dprintf("Hypercall made with MSR[PR]=1\n");
1033 env->gpr[3] = H_PRIVILEGE;
1034 } else {
1035 env->gpr[3] = spapr_hypercall(cpu, env->gpr[3], &env->gpr[4]);
1039 #define HPTE(_table, _i) (void *)(((uint64_t *)(_table)) + ((_i) * 2))
1040 #define HPTE_VALID(_hpte) (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_VALID)
1041 #define HPTE_DIRTY(_hpte) (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_HPTE_DIRTY)
1042 #define CLEAN_HPTE(_hpte) ((*(uint64_t *)(_hpte)) &= tswap64(~HPTE64_V_HPTE_DIRTY))
1043 #define DIRTY_HPTE(_hpte) ((*(uint64_t *)(_hpte)) |= tswap64(HPTE64_V_HPTE_DIRTY))
1046 * Get the fd to access the kernel htab, re-opening it if necessary
1048 static int get_htab_fd(sPAPRMachineState *spapr)
1050 if (spapr->htab_fd >= 0) {
1051 return spapr->htab_fd;
1054 spapr->htab_fd = kvmppc_get_htab_fd(false);
1055 if (spapr->htab_fd < 0) {
1056 error_report("Unable to open fd for reading hash table from KVM: %s",
1057 strerror(errno));
1060 return spapr->htab_fd;
1063 static void close_htab_fd(sPAPRMachineState *spapr)
1065 if (spapr->htab_fd >= 0) {
1066 close(spapr->htab_fd);
1068 spapr->htab_fd = -1;
1071 static int spapr_hpt_shift_for_ramsize(uint64_t ramsize)
1073 int shift;
1075 /* We aim for a hash table of size 1/128 the size of RAM (rounded
1076 * up). The PAPR recommendation is actually 1/64 of RAM size, but
1077 * that's much more than is needed for Linux guests */
1078 shift = ctz64(pow2ceil(ramsize)) - 7;
1079 shift = MAX(shift, 18); /* Minimum architected size */
1080 shift = MIN(shift, 46); /* Maximum architected size */
1081 return shift;
1084 static void spapr_reallocate_hpt(sPAPRMachineState *spapr, int shift,
1085 Error **errp)
1087 long rc;
1089 /* Clean up any HPT info from a previous boot */
1090 g_free(spapr->htab);
1091 spapr->htab = NULL;
1092 spapr->htab_shift = 0;
1093 close_htab_fd(spapr);
1095 rc = kvmppc_reset_htab(shift);
1096 if (rc < 0) {
1097 /* kernel-side HPT needed, but couldn't allocate one */
1098 error_setg_errno(errp, errno,
1099 "Failed to allocate KVM HPT of order %d (try smaller maxmem?)",
1100 shift);
1101 /* This is almost certainly fatal, but if the caller really
1102 * wants to carry on with shift == 0, it's welcome to try */
1103 } else if (rc > 0) {
1104 /* kernel-side HPT allocated */
1105 if (rc != shift) {
1106 error_setg(errp,
1107 "Requested order %d HPT, but kernel allocated order %ld (try smaller maxmem?)",
1108 shift, rc);
1111 spapr->htab_shift = shift;
1112 spapr->htab = NULL;
1113 } else {
1114 /* kernel-side HPT not needed, allocate in userspace instead */
1115 size_t size = 1ULL << shift;
1116 int i;
1118 spapr->htab = qemu_memalign(size, size);
1119 if (!spapr->htab) {
1120 error_setg_errno(errp, errno,
1121 "Could not allocate HPT of order %d", shift);
1122 return;
1125 memset(spapr->htab, 0, size);
1126 spapr->htab_shift = shift;
1128 for (i = 0; i < size / HASH_PTE_SIZE_64; i++) {
1129 DIRTY_HPTE(HPTE(spapr->htab, i));
1134 static int find_unknown_sysbus_device(SysBusDevice *sbdev, void *opaque)
1136 bool matched = false;
1138 if (object_dynamic_cast(OBJECT(sbdev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
1139 matched = true;
1142 if (!matched) {
1143 error_report("Device %s is not supported by this machine yet.",
1144 qdev_fw_name(DEVICE(sbdev)));
1145 exit(1);
1148 return 0;
1151 static void ppc_spapr_reset(void)
1153 MachineState *machine = MACHINE(qdev_get_machine());
1154 sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
1155 PowerPCCPU *first_ppc_cpu;
1156 uint32_t rtas_limit;
1158 /* Check for unknown sysbus devices */
1159 foreach_dynamic_sysbus_device(find_unknown_sysbus_device, NULL);
1161 /* Allocate and/or reset the hash page table */
1162 spapr_reallocate_hpt(spapr,
1163 spapr_hpt_shift_for_ramsize(machine->maxram_size),
1164 &error_fatal);
1166 /* Update the RMA size if necessary */
1167 if (spapr->vrma_adjust) {
1168 spapr->rma_size = kvmppc_rma_size(spapr_node0_size(),
1169 spapr->htab_shift);
1172 qemu_devices_reset();
1175 * We place the device tree and RTAS just below either the top of the RMA,
1176 * or just below 2GB, whichever is lowere, so that it can be
1177 * processed with 32-bit real mode code if necessary
1179 rtas_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR);
1180 spapr->rtas_addr = rtas_limit - RTAS_MAX_SIZE;
1181 spapr->fdt_addr = spapr->rtas_addr - FDT_MAX_SIZE;
1183 /* Load the fdt */
1184 spapr_finalize_fdt(spapr, spapr->fdt_addr, spapr->rtas_addr,
1185 spapr->rtas_size);
1187 /* Copy RTAS over */
1188 cpu_physical_memory_write(spapr->rtas_addr, spapr->rtas_blob,
1189 spapr->rtas_size);
1191 /* Set up the entry state */
1192 first_ppc_cpu = POWERPC_CPU(first_cpu);
1193 first_ppc_cpu->env.gpr[3] = spapr->fdt_addr;
1194 first_ppc_cpu->env.gpr[5] = 0;
1195 first_cpu->halted = 0;
1196 first_ppc_cpu->env.nip = SPAPR_ENTRY_POINT;
1200 static void spapr_create_nvram(sPAPRMachineState *spapr)
1202 DeviceState *dev = qdev_create(&spapr->vio_bus->bus, "spapr-nvram");
1203 DriveInfo *dinfo = drive_get(IF_PFLASH, 0, 0);
1205 if (dinfo) {
1206 qdev_prop_set_drive(dev, "drive", blk_by_legacy_dinfo(dinfo),
1207 &error_fatal);
1210 qdev_init_nofail(dev);
1212 spapr->nvram = (struct sPAPRNVRAM *)dev;
1215 static void spapr_rtc_create(sPAPRMachineState *spapr)
1217 DeviceState *dev = qdev_create(NULL, TYPE_SPAPR_RTC);
1219 qdev_init_nofail(dev);
1220 spapr->rtc = dev;
1222 object_property_add_alias(qdev_get_machine(), "rtc-time",
1223 OBJECT(spapr->rtc), "date", NULL);
1226 /* Returns whether we want to use VGA or not */
1227 static bool spapr_vga_init(PCIBus *pci_bus, Error **errp)
1229 switch (vga_interface_type) {
1230 case VGA_NONE:
1231 return false;
1232 case VGA_DEVICE:
1233 return true;
1234 case VGA_STD:
1235 case VGA_VIRTIO:
1236 return pci_vga_init(pci_bus) != NULL;
1237 default:
1238 error_setg(errp,
1239 "Unsupported VGA mode, only -vga std or -vga virtio is supported");
1240 return false;
1244 static int spapr_post_load(void *opaque, int version_id)
1246 sPAPRMachineState *spapr = (sPAPRMachineState *)opaque;
1247 int err = 0;
1249 /* In earlier versions, there was no separate qdev for the PAPR
1250 * RTC, so the RTC offset was stored directly in sPAPREnvironment.
1251 * So when migrating from those versions, poke the incoming offset
1252 * value into the RTC device */
1253 if (version_id < 3) {
1254 err = spapr_rtc_import_offset(spapr->rtc, spapr->rtc_offset);
1257 return err;
1260 static bool version_before_3(void *opaque, int version_id)
1262 return version_id < 3;
1265 static const VMStateDescription vmstate_spapr = {
1266 .name = "spapr",
1267 .version_id = 3,
1268 .minimum_version_id = 1,
1269 .post_load = spapr_post_load,
1270 .fields = (VMStateField[]) {
1271 /* used to be @next_irq */
1272 VMSTATE_UNUSED_BUFFER(version_before_3, 0, 4),
1274 /* RTC offset */
1275 VMSTATE_UINT64_TEST(rtc_offset, sPAPRMachineState, version_before_3),
1277 VMSTATE_PPC_TIMEBASE_V(tb, sPAPRMachineState, 2),
1278 VMSTATE_END_OF_LIST()
1282 static int htab_save_setup(QEMUFile *f, void *opaque)
1284 sPAPRMachineState *spapr = opaque;
1286 /* "Iteration" header */
1287 qemu_put_be32(f, spapr->htab_shift);
1289 if (spapr->htab) {
1290 spapr->htab_save_index = 0;
1291 spapr->htab_first_pass = true;
1292 } else {
1293 assert(kvm_enabled());
1297 return 0;
1300 static void htab_save_first_pass(QEMUFile *f, sPAPRMachineState *spapr,
1301 int64_t max_ns)
1303 bool has_timeout = max_ns != -1;
1304 int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
1305 int index = spapr->htab_save_index;
1306 int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
1308 assert(spapr->htab_first_pass);
1310 do {
1311 int chunkstart;
1313 /* Consume invalid HPTEs */
1314 while ((index < htabslots)
1315 && !HPTE_VALID(HPTE(spapr->htab, index))) {
1316 index++;
1317 CLEAN_HPTE(HPTE(spapr->htab, index));
1320 /* Consume valid HPTEs */
1321 chunkstart = index;
1322 while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
1323 && HPTE_VALID(HPTE(spapr->htab, index))) {
1324 index++;
1325 CLEAN_HPTE(HPTE(spapr->htab, index));
1328 if (index > chunkstart) {
1329 int n_valid = index - chunkstart;
1331 qemu_put_be32(f, chunkstart);
1332 qemu_put_be16(f, n_valid);
1333 qemu_put_be16(f, 0);
1334 qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
1335 HASH_PTE_SIZE_64 * n_valid);
1337 if (has_timeout &&
1338 (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
1339 break;
1342 } while ((index < htabslots) && !qemu_file_rate_limit(f));
1344 if (index >= htabslots) {
1345 assert(index == htabslots);
1346 index = 0;
1347 spapr->htab_first_pass = false;
1349 spapr->htab_save_index = index;
1352 static int htab_save_later_pass(QEMUFile *f, sPAPRMachineState *spapr,
1353 int64_t max_ns)
1355 bool final = max_ns < 0;
1356 int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
1357 int examined = 0, sent = 0;
1358 int index = spapr->htab_save_index;
1359 int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
1361 assert(!spapr->htab_first_pass);
1363 do {
1364 int chunkstart, invalidstart;
1366 /* Consume non-dirty HPTEs */
1367 while ((index < htabslots)
1368 && !HPTE_DIRTY(HPTE(spapr->htab, index))) {
1369 index++;
1370 examined++;
1373 chunkstart = index;
1374 /* Consume valid dirty HPTEs */
1375 while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
1376 && HPTE_DIRTY(HPTE(spapr->htab, index))
1377 && HPTE_VALID(HPTE(spapr->htab, index))) {
1378 CLEAN_HPTE(HPTE(spapr->htab, index));
1379 index++;
1380 examined++;
1383 invalidstart = index;
1384 /* Consume invalid dirty HPTEs */
1385 while ((index < htabslots) && (index - invalidstart < USHRT_MAX)
1386 && HPTE_DIRTY(HPTE(spapr->htab, index))
1387 && !HPTE_VALID(HPTE(spapr->htab, index))) {
1388 CLEAN_HPTE(HPTE(spapr->htab, index));
1389 index++;
1390 examined++;
1393 if (index > chunkstart) {
1394 int n_valid = invalidstart - chunkstart;
1395 int n_invalid = index - invalidstart;
1397 qemu_put_be32(f, chunkstart);
1398 qemu_put_be16(f, n_valid);
1399 qemu_put_be16(f, n_invalid);
1400 qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
1401 HASH_PTE_SIZE_64 * n_valid);
1402 sent += index - chunkstart;
1404 if (!final && (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
1405 break;
1409 if (examined >= htabslots) {
1410 break;
1413 if (index >= htabslots) {
1414 assert(index == htabslots);
1415 index = 0;
1417 } while ((examined < htabslots) && (!qemu_file_rate_limit(f) || final));
1419 if (index >= htabslots) {
1420 assert(index == htabslots);
1421 index = 0;
1424 spapr->htab_save_index = index;
1426 return (examined >= htabslots) && (sent == 0) ? 1 : 0;
1429 #define MAX_ITERATION_NS 5000000 /* 5 ms */
1430 #define MAX_KVM_BUF_SIZE 2048
1432 static int htab_save_iterate(QEMUFile *f, void *opaque)
1434 sPAPRMachineState *spapr = opaque;
1435 int fd;
1436 int rc = 0;
1438 /* Iteration header */
1439 qemu_put_be32(f, 0);
1441 if (!spapr->htab) {
1442 assert(kvm_enabled());
1444 fd = get_htab_fd(spapr);
1445 if (fd < 0) {
1446 return fd;
1449 rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, MAX_ITERATION_NS);
1450 if (rc < 0) {
1451 return rc;
1453 } else if (spapr->htab_first_pass) {
1454 htab_save_first_pass(f, spapr, MAX_ITERATION_NS);
1455 } else {
1456 rc = htab_save_later_pass(f, spapr, MAX_ITERATION_NS);
1459 /* End marker */
1460 qemu_put_be32(f, 0);
1461 qemu_put_be16(f, 0);
1462 qemu_put_be16(f, 0);
1464 return rc;
1467 static int htab_save_complete(QEMUFile *f, void *opaque)
1469 sPAPRMachineState *spapr = opaque;
1470 int fd;
1472 /* Iteration header */
1473 qemu_put_be32(f, 0);
1475 if (!spapr->htab) {
1476 int rc;
1478 assert(kvm_enabled());
1480 fd = get_htab_fd(spapr);
1481 if (fd < 0) {
1482 return fd;
1485 rc = kvmppc_save_htab(f, fd, MAX_KVM_BUF_SIZE, -1);
1486 if (rc < 0) {
1487 return rc;
1489 close_htab_fd(spapr);
1490 } else {
1491 if (spapr->htab_first_pass) {
1492 htab_save_first_pass(f, spapr, -1);
1494 htab_save_later_pass(f, spapr, -1);
1497 /* End marker */
1498 qemu_put_be32(f, 0);
1499 qemu_put_be16(f, 0);
1500 qemu_put_be16(f, 0);
1502 return 0;
1505 static int htab_load(QEMUFile *f, void *opaque, int version_id)
1507 sPAPRMachineState *spapr = opaque;
1508 uint32_t section_hdr;
1509 int fd = -1;
1511 if (version_id < 1 || version_id > 1) {
1512 error_report("htab_load() bad version");
1513 return -EINVAL;
1516 section_hdr = qemu_get_be32(f);
1518 if (section_hdr) {
1519 Error *local_err = NULL;
1521 /* First section gives the htab size */
1522 spapr_reallocate_hpt(spapr, section_hdr, &local_err);
1523 if (local_err) {
1524 error_report_err(local_err);
1525 return -EINVAL;
1527 return 0;
1530 if (!spapr->htab) {
1531 assert(kvm_enabled());
1533 fd = kvmppc_get_htab_fd(true);
1534 if (fd < 0) {
1535 error_report("Unable to open fd to restore KVM hash table: %s",
1536 strerror(errno));
1540 while (true) {
1541 uint32_t index;
1542 uint16_t n_valid, n_invalid;
1544 index = qemu_get_be32(f);
1545 n_valid = qemu_get_be16(f);
1546 n_invalid = qemu_get_be16(f);
1548 if ((index == 0) && (n_valid == 0) && (n_invalid == 0)) {
1549 /* End of Stream */
1550 break;
1553 if ((index + n_valid + n_invalid) >
1554 (HTAB_SIZE(spapr) / HASH_PTE_SIZE_64)) {
1555 /* Bad index in stream */
1556 error_report(
1557 "htab_load() bad index %d (%hd+%hd entries) in htab stream (htab_shift=%d)",
1558 index, n_valid, n_invalid, spapr->htab_shift);
1559 return -EINVAL;
1562 if (spapr->htab) {
1563 if (n_valid) {
1564 qemu_get_buffer(f, HPTE(spapr->htab, index),
1565 HASH_PTE_SIZE_64 * n_valid);
1567 if (n_invalid) {
1568 memset(HPTE(spapr->htab, index + n_valid), 0,
1569 HASH_PTE_SIZE_64 * n_invalid);
1571 } else {
1572 int rc;
1574 assert(fd >= 0);
1576 rc = kvmppc_load_htab_chunk(f, fd, index, n_valid, n_invalid);
1577 if (rc < 0) {
1578 return rc;
1583 if (!spapr->htab) {
1584 assert(fd >= 0);
1585 close(fd);
1588 return 0;
1591 static SaveVMHandlers savevm_htab_handlers = {
1592 .save_live_setup = htab_save_setup,
1593 .save_live_iterate = htab_save_iterate,
1594 .save_live_complete_precopy = htab_save_complete,
1595 .load_state = htab_load,
1598 static void spapr_boot_set(void *opaque, const char *boot_device,
1599 Error **errp)
1601 MachineState *machine = MACHINE(qdev_get_machine());
1602 machine->boot_order = g_strdup(boot_device);
1606 * Reset routine for LMB DR devices.
1608 * Unlike PCI DR devices, LMB DR devices explicitly register this reset
1609 * routine. Reset for PCI DR devices will be handled by PHB reset routine
1610 * when it walks all its children devices. LMB devices reset occurs
1611 * as part of spapr_ppc_reset().
1613 static void spapr_drc_reset(void *opaque)
1615 sPAPRDRConnector *drc = opaque;
1616 DeviceState *d = DEVICE(drc);
1618 if (d) {
1619 device_reset(d);
1623 static void spapr_create_lmb_dr_connectors(sPAPRMachineState *spapr)
1625 MachineState *machine = MACHINE(spapr);
1626 uint64_t lmb_size = SPAPR_MEMORY_BLOCK_SIZE;
1627 uint32_t nr_lmbs = (machine->maxram_size - machine->ram_size)/lmb_size;
1628 int i;
1630 for (i = 0; i < nr_lmbs; i++) {
1631 sPAPRDRConnector *drc;
1632 uint64_t addr;
1634 addr = i * lmb_size + spapr->hotplug_memory.base;
1635 drc = spapr_dr_connector_new(OBJECT(spapr), SPAPR_DR_CONNECTOR_TYPE_LMB,
1636 addr/lmb_size);
1637 qemu_register_reset(spapr_drc_reset, drc);
1642 * If RAM size, maxmem size and individual node mem sizes aren't aligned
1643 * to SPAPR_MEMORY_BLOCK_SIZE(256MB), then refuse to start the guest
1644 * since we can't support such unaligned sizes with DRCONF_MEMORY.
1646 static void spapr_validate_node_memory(MachineState *machine, Error **errp)
1648 int i;
1650 if (machine->ram_size % SPAPR_MEMORY_BLOCK_SIZE) {
1651 error_setg(errp, "Memory size 0x" RAM_ADDR_FMT
1652 " is not aligned to %llu MiB",
1653 machine->ram_size,
1654 SPAPR_MEMORY_BLOCK_SIZE / M_BYTE);
1655 return;
1658 if (machine->maxram_size % SPAPR_MEMORY_BLOCK_SIZE) {
1659 error_setg(errp, "Maximum memory size 0x" RAM_ADDR_FMT
1660 " is not aligned to %llu MiB",
1661 machine->ram_size,
1662 SPAPR_MEMORY_BLOCK_SIZE / M_BYTE);
1663 return;
1666 for (i = 0; i < nb_numa_nodes; i++) {
1667 if (numa_info[i].node_mem % SPAPR_MEMORY_BLOCK_SIZE) {
1668 error_setg(errp,
1669 "Node %d memory size 0x%" PRIx64
1670 " is not aligned to %llu MiB",
1671 i, numa_info[i].node_mem,
1672 SPAPR_MEMORY_BLOCK_SIZE / M_BYTE);
1673 return;
1678 /* pSeries LPAR / sPAPR hardware init */
1679 static void ppc_spapr_init(MachineState *machine)
1681 sPAPRMachineState *spapr = SPAPR_MACHINE(machine);
1682 sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(machine);
1683 const char *kernel_filename = machine->kernel_filename;
1684 const char *kernel_cmdline = machine->kernel_cmdline;
1685 const char *initrd_filename = machine->initrd_filename;
1686 PowerPCCPU *cpu;
1687 PCIHostState *phb;
1688 int i;
1689 MemoryRegion *sysmem = get_system_memory();
1690 MemoryRegion *ram = g_new(MemoryRegion, 1);
1691 MemoryRegion *rma_region;
1692 void *rma = NULL;
1693 hwaddr rma_alloc_size;
1694 hwaddr node0_size = spapr_node0_size();
1695 uint32_t initrd_base = 0;
1696 long kernel_size = 0, initrd_size = 0;
1697 long load_limit, fw_size;
1698 bool kernel_le = false;
1699 char *filename;
1701 msi_nonbroken = true;
1703 QLIST_INIT(&spapr->phbs);
1705 cpu_ppc_hypercall = emulate_spapr_hypercall;
1707 /* Allocate RMA if necessary */
1708 rma_alloc_size = kvmppc_alloc_rma(&rma);
1710 if (rma_alloc_size == -1) {
1711 error_report("Unable to create RMA");
1712 exit(1);
1715 if (rma_alloc_size && (rma_alloc_size < node0_size)) {
1716 spapr->rma_size = rma_alloc_size;
1717 } else {
1718 spapr->rma_size = node0_size;
1720 /* With KVM, we don't actually know whether KVM supports an
1721 * unbounded RMA (PR KVM) or is limited by the hash table size
1722 * (HV KVM using VRMA), so we always assume the latter
1724 * In that case, we also limit the initial allocations for RTAS
1725 * etc... to 256M since we have no way to know what the VRMA size
1726 * is going to be as it depends on the size of the hash table
1727 * isn't determined yet.
1729 if (kvm_enabled()) {
1730 spapr->vrma_adjust = 1;
1731 spapr->rma_size = MIN(spapr->rma_size, 0x10000000);
1735 if (spapr->rma_size > node0_size) {
1736 error_report("Numa node 0 has to span the RMA (%#08"HWADDR_PRIx")",
1737 spapr->rma_size);
1738 exit(1);
1741 /* Setup a load limit for the ramdisk leaving room for SLOF and FDT */
1742 load_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR) - FW_OVERHEAD;
1744 /* Set up Interrupt Controller before we create the VCPUs */
1745 spapr->icp = xics_system_init(machine,
1746 DIV_ROUND_UP(max_cpus * kvmppc_smt_threads(),
1747 smp_threads),
1748 XICS_IRQS, &error_fatal);
1750 if (smc->dr_lmb_enabled) {
1751 spapr_validate_node_memory(machine, &error_fatal);
1754 /* init CPUs */
1755 if (machine->cpu_model == NULL) {
1756 machine->cpu_model = kvm_enabled() ? "host" : "POWER7";
1758 for (i = 0; i < smp_cpus; i++) {
1759 cpu = cpu_ppc_init(machine->cpu_model);
1760 if (cpu == NULL) {
1761 error_report("Unable to find PowerPC CPU definition");
1762 exit(1);
1764 spapr_cpu_init(spapr, cpu, &error_fatal);
1767 if (kvm_enabled()) {
1768 /* Enable H_LOGICAL_CI_* so SLOF can talk to in-kernel devices */
1769 kvmppc_enable_logical_ci_hcalls();
1770 kvmppc_enable_set_mode_hcall();
1773 /* allocate RAM */
1774 memory_region_allocate_system_memory(ram, NULL, "ppc_spapr.ram",
1775 machine->ram_size);
1776 memory_region_add_subregion(sysmem, 0, ram);
1778 if (rma_alloc_size && rma) {
1779 rma_region = g_new(MemoryRegion, 1);
1780 memory_region_init_ram_ptr(rma_region, NULL, "ppc_spapr.rma",
1781 rma_alloc_size, rma);
1782 vmstate_register_ram_global(rma_region);
1783 memory_region_add_subregion(sysmem, 0, rma_region);
1786 /* initialize hotplug memory address space */
1787 if (machine->ram_size < machine->maxram_size) {
1788 ram_addr_t hotplug_mem_size = machine->maxram_size - machine->ram_size;
1790 * Limit the number of hotpluggable memory slots to half the number
1791 * slots that KVM supports, leaving the other half for PCI and other
1792 * devices. However ensure that number of slots doesn't drop below 32.
1794 int max_memslots = kvm_enabled() ? kvm_get_max_memslots() / 2 :
1795 SPAPR_MAX_RAM_SLOTS;
1797 if (max_memslots < SPAPR_MAX_RAM_SLOTS) {
1798 max_memslots = SPAPR_MAX_RAM_SLOTS;
1800 if (machine->ram_slots > max_memslots) {
1801 error_report("Specified number of memory slots %"
1802 PRIu64" exceeds max supported %d",
1803 machine->ram_slots, max_memslots);
1804 exit(1);
1807 spapr->hotplug_memory.base = ROUND_UP(machine->ram_size,
1808 SPAPR_HOTPLUG_MEM_ALIGN);
1809 memory_region_init(&spapr->hotplug_memory.mr, OBJECT(spapr),
1810 "hotplug-memory", hotplug_mem_size);
1811 memory_region_add_subregion(sysmem, spapr->hotplug_memory.base,
1812 &spapr->hotplug_memory.mr);
1815 if (smc->dr_lmb_enabled) {
1816 spapr_create_lmb_dr_connectors(spapr);
1819 filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, "spapr-rtas.bin");
1820 if (!filename) {
1821 error_report("Could not find LPAR rtas '%s'", "spapr-rtas.bin");
1822 exit(1);
1824 spapr->rtas_size = get_image_size(filename);
1825 if (spapr->rtas_size < 0) {
1826 error_report("Could not get size of LPAR rtas '%s'", filename);
1827 exit(1);
1829 spapr->rtas_blob = g_malloc(spapr->rtas_size);
1830 if (load_image_size(filename, spapr->rtas_blob, spapr->rtas_size) < 0) {
1831 error_report("Could not load LPAR rtas '%s'", filename);
1832 exit(1);
1834 if (spapr->rtas_size > RTAS_MAX_SIZE) {
1835 error_report("RTAS too big ! 0x%zx bytes (max is 0x%x)",
1836 (size_t)spapr->rtas_size, RTAS_MAX_SIZE);
1837 exit(1);
1839 g_free(filename);
1841 /* Set up EPOW events infrastructure */
1842 spapr_events_init(spapr);
1844 /* Set up the RTC RTAS interfaces */
1845 spapr_rtc_create(spapr);
1847 /* Set up VIO bus */
1848 spapr->vio_bus = spapr_vio_bus_init();
1850 for (i = 0; i < MAX_SERIAL_PORTS; i++) {
1851 if (serial_hds[i]) {
1852 spapr_vty_create(spapr->vio_bus, serial_hds[i]);
1856 /* We always have at least the nvram device on VIO */
1857 spapr_create_nvram(spapr);
1859 /* Set up PCI */
1860 spapr_pci_rtas_init();
1862 phb = spapr_create_phb(spapr, 0);
1864 for (i = 0; i < nb_nics; i++) {
1865 NICInfo *nd = &nd_table[i];
1867 if (!nd->model) {
1868 nd->model = g_strdup("ibmveth");
1871 if (strcmp(nd->model, "ibmveth") == 0) {
1872 spapr_vlan_create(spapr->vio_bus, nd);
1873 } else {
1874 pci_nic_init_nofail(&nd_table[i], phb->bus, nd->model, NULL);
1878 for (i = 0; i <= drive_get_max_bus(IF_SCSI); i++) {
1879 spapr_vscsi_create(spapr->vio_bus);
1882 /* Graphics */
1883 if (spapr_vga_init(phb->bus, &error_fatal)) {
1884 spapr->has_graphics = true;
1885 machine->usb |= defaults_enabled() && !machine->usb_disabled;
1888 if (machine->usb) {
1889 if (smc->use_ohci_by_default) {
1890 pci_create_simple(phb->bus, -1, "pci-ohci");
1891 } else {
1892 pci_create_simple(phb->bus, -1, "nec-usb-xhci");
1895 if (spapr->has_graphics) {
1896 USBBus *usb_bus = usb_bus_find(-1);
1898 usb_create_simple(usb_bus, "usb-kbd");
1899 usb_create_simple(usb_bus, "usb-mouse");
1903 if (spapr->rma_size < (MIN_RMA_SLOF << 20)) {
1904 error_report(
1905 "pSeries SLOF firmware requires >= %ldM guest RMA (Real Mode Area memory)",
1906 MIN_RMA_SLOF);
1907 exit(1);
1910 if (kernel_filename) {
1911 uint64_t lowaddr = 0;
1913 kernel_size = load_elf(kernel_filename, translate_kernel_address, NULL,
1914 NULL, &lowaddr, NULL, 1, PPC_ELF_MACHINE,
1915 0, 0);
1916 if (kernel_size == ELF_LOAD_WRONG_ENDIAN) {
1917 kernel_size = load_elf(kernel_filename,
1918 translate_kernel_address, NULL,
1919 NULL, &lowaddr, NULL, 0, PPC_ELF_MACHINE,
1920 0, 0);
1921 kernel_le = kernel_size > 0;
1923 if (kernel_size < 0) {
1924 error_report("error loading %s: %s",
1925 kernel_filename, load_elf_strerror(kernel_size));
1926 exit(1);
1929 /* load initrd */
1930 if (initrd_filename) {
1931 /* Try to locate the initrd in the gap between the kernel
1932 * and the firmware. Add a bit of space just in case
1934 initrd_base = (KERNEL_LOAD_ADDR + kernel_size + 0x1ffff) & ~0xffff;
1935 initrd_size = load_image_targphys(initrd_filename, initrd_base,
1936 load_limit - initrd_base);
1937 if (initrd_size < 0) {
1938 error_report("could not load initial ram disk '%s'",
1939 initrd_filename);
1940 exit(1);
1942 } else {
1943 initrd_base = 0;
1944 initrd_size = 0;
1948 if (bios_name == NULL) {
1949 bios_name = FW_FILE_NAME;
1951 filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
1952 if (!filename) {
1953 error_report("Could not find LPAR firmware '%s'", bios_name);
1954 exit(1);
1956 fw_size = load_image_targphys(filename, 0, FW_MAX_SIZE);
1957 if (fw_size <= 0) {
1958 error_report("Could not load LPAR firmware '%s'", filename);
1959 exit(1);
1961 g_free(filename);
1963 /* FIXME: Should register things through the MachineState's qdev
1964 * interface, this is a legacy from the sPAPREnvironment structure
1965 * which predated MachineState but had a similar function */
1966 vmstate_register(NULL, 0, &vmstate_spapr, spapr);
1967 register_savevm_live(NULL, "spapr/htab", -1, 1,
1968 &savevm_htab_handlers, spapr);
1970 /* Prepare the device tree */
1971 spapr->fdt_skel = spapr_create_fdt_skel(initrd_base, initrd_size,
1972 kernel_size, kernel_le,
1973 kernel_cmdline,
1974 spapr->check_exception_irq);
1975 assert(spapr->fdt_skel != NULL);
1977 /* used by RTAS */
1978 QTAILQ_INIT(&spapr->ccs_list);
1979 qemu_register_reset(spapr_ccs_reset_hook, spapr);
1981 qemu_register_boot_set(spapr_boot_set, spapr);
1984 static int spapr_kvm_type(const char *vm_type)
1986 if (!vm_type) {
1987 return 0;
1990 if (!strcmp(vm_type, "HV")) {
1991 return 1;
1994 if (!strcmp(vm_type, "PR")) {
1995 return 2;
1998 error_report("Unknown kvm-type specified '%s'", vm_type);
1999 exit(1);
2003 * Implementation of an interface to adjust firmware path
2004 * for the bootindex property handling.
2006 static char *spapr_get_fw_dev_path(FWPathProvider *p, BusState *bus,
2007 DeviceState *dev)
2009 #define CAST(type, obj, name) \
2010 ((type *)object_dynamic_cast(OBJECT(obj), (name)))
2011 SCSIDevice *d = CAST(SCSIDevice, dev, TYPE_SCSI_DEVICE);
2012 sPAPRPHBState *phb = CAST(sPAPRPHBState, dev, TYPE_SPAPR_PCI_HOST_BRIDGE);
2014 if (d) {
2015 void *spapr = CAST(void, bus->parent, "spapr-vscsi");
2016 VirtIOSCSI *virtio = CAST(VirtIOSCSI, bus->parent, TYPE_VIRTIO_SCSI);
2017 USBDevice *usb = CAST(USBDevice, bus->parent, TYPE_USB_DEVICE);
2019 if (spapr) {
2021 * Replace "channel@0/disk@0,0" with "disk@8000000000000000":
2022 * We use SRP luns of the form 8000 | (bus << 8) | (id << 5) | lun
2023 * in the top 16 bits of the 64-bit LUN
2025 unsigned id = 0x8000 | (d->id << 8) | d->lun;
2026 return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
2027 (uint64_t)id << 48);
2028 } else if (virtio) {
2030 * We use SRP luns of the form 01000000 | (target << 8) | lun
2031 * in the top 32 bits of the 64-bit LUN
2032 * Note: the quote above is from SLOF and it is wrong,
2033 * the actual binding is:
2034 * swap 0100 or 10 << or 20 << ( target lun-id -- srplun )
2036 unsigned id = 0x1000000 | (d->id << 16) | d->lun;
2037 return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
2038 (uint64_t)id << 32);
2039 } else if (usb) {
2041 * We use SRP luns of the form 01000000 | (usb-port << 16) | lun
2042 * in the top 32 bits of the 64-bit LUN
2044 unsigned usb_port = atoi(usb->port->path);
2045 unsigned id = 0x1000000 | (usb_port << 16) | d->lun;
2046 return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
2047 (uint64_t)id << 32);
2051 if (phb) {
2052 /* Replace "pci" with "pci@800000020000000" */
2053 return g_strdup_printf("pci@%"PRIX64, phb->buid);
2056 return NULL;
2059 static char *spapr_get_kvm_type(Object *obj, Error **errp)
2061 sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2063 return g_strdup(spapr->kvm_type);
2066 static void spapr_set_kvm_type(Object *obj, const char *value, Error **errp)
2068 sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2070 g_free(spapr->kvm_type);
2071 spapr->kvm_type = g_strdup(value);
2074 static void spapr_machine_initfn(Object *obj)
2076 sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2078 spapr->htab_fd = -1;
2079 object_property_add_str(obj, "kvm-type",
2080 spapr_get_kvm_type, spapr_set_kvm_type, NULL);
2081 object_property_set_description(obj, "kvm-type",
2082 "Specifies the KVM virtualization mode (HV, PR)",
2083 NULL);
2086 static void spapr_machine_finalizefn(Object *obj)
2088 sPAPRMachineState *spapr = SPAPR_MACHINE(obj);
2090 g_free(spapr->kvm_type);
2093 static void ppc_cpu_do_nmi_on_cpu(void *arg)
2095 CPUState *cs = arg;
2097 cpu_synchronize_state(cs);
2098 ppc_cpu_do_system_reset(cs);
2101 static void spapr_nmi(NMIState *n, int cpu_index, Error **errp)
2103 CPUState *cs;
2105 CPU_FOREACH(cs) {
2106 async_run_on_cpu(cs, ppc_cpu_do_nmi_on_cpu, cs);
2110 static void spapr_add_lmbs(DeviceState *dev, uint64_t addr, uint64_t size,
2111 uint32_t node, Error **errp)
2113 sPAPRDRConnector *drc;
2114 sPAPRDRConnectorClass *drck;
2115 uint32_t nr_lmbs = size/SPAPR_MEMORY_BLOCK_SIZE;
2116 int i, fdt_offset, fdt_size;
2117 void *fdt;
2119 for (i = 0; i < nr_lmbs; i++) {
2120 drc = spapr_dr_connector_by_id(SPAPR_DR_CONNECTOR_TYPE_LMB,
2121 addr/SPAPR_MEMORY_BLOCK_SIZE);
2122 g_assert(drc);
2124 fdt = create_device_tree(&fdt_size);
2125 fdt_offset = spapr_populate_memory_node(fdt, node, addr,
2126 SPAPR_MEMORY_BLOCK_SIZE);
2128 drck = SPAPR_DR_CONNECTOR_GET_CLASS(drc);
2129 drck->attach(drc, dev, fdt, fdt_offset, !dev->hotplugged, errp);
2130 addr += SPAPR_MEMORY_BLOCK_SIZE;
2132 /* send hotplug notification to the
2133 * guest only in case of hotplugged memory
2135 if (dev->hotplugged) {
2136 spapr_hotplug_req_add_by_count(SPAPR_DR_CONNECTOR_TYPE_LMB, nr_lmbs);
2140 static void spapr_memory_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
2141 uint32_t node, Error **errp)
2143 Error *local_err = NULL;
2144 sPAPRMachineState *ms = SPAPR_MACHINE(hotplug_dev);
2145 PCDIMMDevice *dimm = PC_DIMM(dev);
2146 PCDIMMDeviceClass *ddc = PC_DIMM_GET_CLASS(dimm);
2147 MemoryRegion *mr = ddc->get_memory_region(dimm);
2148 uint64_t align = memory_region_get_alignment(mr);
2149 uint64_t size = memory_region_size(mr);
2150 uint64_t addr;
2152 if (size % SPAPR_MEMORY_BLOCK_SIZE) {
2153 error_setg(&local_err, "Hotplugged memory size must be a multiple of "
2154 "%lld MB", SPAPR_MEMORY_BLOCK_SIZE/M_BYTE);
2155 goto out;
2158 pc_dimm_memory_plug(dev, &ms->hotplug_memory, mr, align, &local_err);
2159 if (local_err) {
2160 goto out;
2163 addr = object_property_get_int(OBJECT(dimm), PC_DIMM_ADDR_PROP, &local_err);
2164 if (local_err) {
2165 pc_dimm_memory_unplug(dev, &ms->hotplug_memory, mr);
2166 goto out;
2169 spapr_add_lmbs(dev, addr, size, node, &error_abort);
2171 out:
2172 error_propagate(errp, local_err);
2175 static void spapr_machine_device_plug(HotplugHandler *hotplug_dev,
2176 DeviceState *dev, Error **errp)
2178 sPAPRMachineClass *smc = SPAPR_MACHINE_GET_CLASS(qdev_get_machine());
2180 if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
2181 int node;
2183 if (!smc->dr_lmb_enabled) {
2184 error_setg(errp, "Memory hotplug not supported for this machine");
2185 return;
2187 node = object_property_get_int(OBJECT(dev), PC_DIMM_NODE_PROP, errp);
2188 if (*errp) {
2189 return;
2191 if (node < 0 || node >= MAX_NODES) {
2192 error_setg(errp, "Invaild node %d", node);
2193 return;
2197 * Currently PowerPC kernel doesn't allow hot-adding memory to
2198 * memory-less node, but instead will silently add the memory
2199 * to the first node that has some memory. This causes two
2200 * unexpected behaviours for the user.
2202 * - Memory gets hotplugged to a different node than what the user
2203 * specified.
2204 * - Since pc-dimm subsystem in QEMU still thinks that memory belongs
2205 * to memory-less node, a reboot will set things accordingly
2206 * and the previously hotplugged memory now ends in the right node.
2207 * This appears as if some memory moved from one node to another.
2209 * So until kernel starts supporting memory hotplug to memory-less
2210 * nodes, just prevent such attempts upfront in QEMU.
2212 if (nb_numa_nodes && !numa_info[node].node_mem) {
2213 error_setg(errp, "Can't hotplug memory to memory-less node %d",
2214 node);
2215 return;
2218 spapr_memory_plug(hotplug_dev, dev, node, errp);
2222 static void spapr_machine_device_unplug(HotplugHandler *hotplug_dev,
2223 DeviceState *dev, Error **errp)
2225 if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
2226 error_setg(errp, "Memory hot unplug not supported by sPAPR");
2230 static HotplugHandler *spapr_get_hotpug_handler(MachineState *machine,
2231 DeviceState *dev)
2233 if (object_dynamic_cast(OBJECT(dev), TYPE_PC_DIMM)) {
2234 return HOTPLUG_HANDLER(machine);
2236 return NULL;
2239 static unsigned spapr_cpu_index_to_socket_id(unsigned cpu_index)
2241 /* Allocate to NUMA nodes on a "socket" basis (not that concept of
2242 * socket means much for the paravirtualized PAPR platform) */
2243 return cpu_index / smp_threads / smp_cores;
2246 static void spapr_machine_class_init(ObjectClass *oc, void *data)
2248 MachineClass *mc = MACHINE_CLASS(oc);
2249 sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(oc);
2250 FWPathProviderClass *fwc = FW_PATH_PROVIDER_CLASS(oc);
2251 NMIClass *nc = NMI_CLASS(oc);
2252 HotplugHandlerClass *hc = HOTPLUG_HANDLER_CLASS(oc);
2254 mc->desc = "pSeries Logical Partition (PAPR compliant)";
2257 * We set up the default / latest behaviour here. The class_init
2258 * functions for the specific versioned machine types can override
2259 * these details for backwards compatibility
2261 mc->init = ppc_spapr_init;
2262 mc->reset = ppc_spapr_reset;
2263 mc->block_default_type = IF_SCSI;
2264 mc->max_cpus = MAX_CPUMASK_BITS;
2265 mc->no_parallel = 1;
2266 mc->default_boot_order = "";
2267 mc->default_ram_size = 512 * M_BYTE;
2268 mc->kvm_type = spapr_kvm_type;
2269 mc->has_dynamic_sysbus = true;
2270 mc->pci_allow_0_address = true;
2271 mc->get_hotplug_handler = spapr_get_hotpug_handler;
2272 hc->plug = spapr_machine_device_plug;
2273 hc->unplug = spapr_machine_device_unplug;
2274 mc->cpu_index_to_socket_id = spapr_cpu_index_to_socket_id;
2276 smc->dr_lmb_enabled = true;
2277 fwc->get_dev_path = spapr_get_fw_dev_path;
2278 nc->nmi_monitor_handler = spapr_nmi;
2281 static const TypeInfo spapr_machine_info = {
2282 .name = TYPE_SPAPR_MACHINE,
2283 .parent = TYPE_MACHINE,
2284 .abstract = true,
2285 .instance_size = sizeof(sPAPRMachineState),
2286 .instance_init = spapr_machine_initfn,
2287 .instance_finalize = spapr_machine_finalizefn,
2288 .class_size = sizeof(sPAPRMachineClass),
2289 .class_init = spapr_machine_class_init,
2290 .interfaces = (InterfaceInfo[]) {
2291 { TYPE_FW_PATH_PROVIDER },
2292 { TYPE_NMI },
2293 { TYPE_HOTPLUG_HANDLER },
2298 #define DEFINE_SPAPR_MACHINE(suffix, verstr, latest) \
2299 static void spapr_machine_##suffix##_class_init(ObjectClass *oc, \
2300 void *data) \
2302 MachineClass *mc = MACHINE_CLASS(oc); \
2303 spapr_machine_##suffix##_class_options(mc); \
2304 if (latest) { \
2305 mc->alias = "pseries"; \
2306 mc->is_default = 1; \
2309 static void spapr_machine_##suffix##_instance_init(Object *obj) \
2311 MachineState *machine = MACHINE(obj); \
2312 spapr_machine_##suffix##_instance_options(machine); \
2314 static const TypeInfo spapr_machine_##suffix##_info = { \
2315 .name = MACHINE_TYPE_NAME("pseries-" verstr), \
2316 .parent = TYPE_SPAPR_MACHINE, \
2317 .class_init = spapr_machine_##suffix##_class_init, \
2318 .instance_init = spapr_machine_##suffix##_instance_init, \
2319 }; \
2320 static void spapr_machine_register_##suffix(void) \
2322 type_register(&spapr_machine_##suffix##_info); \
2324 type_init(spapr_machine_register_##suffix)
2327 * pseries-2.7
2329 static void spapr_machine_2_7_instance_options(MachineState *machine)
2333 static void spapr_machine_2_7_class_options(MachineClass *mc)
2335 /* Defaults for the latest behaviour inherited from the base class */
2338 DEFINE_SPAPR_MACHINE(2_7, "2.7", true);
2341 * pseries-2.6
2343 #define SPAPR_COMPAT_2_6 \
2344 HW_COMPAT_2_6
2346 static void spapr_machine_2_6_instance_options(MachineState *machine)
2350 static void spapr_machine_2_6_class_options(MachineClass *mc)
2352 spapr_machine_2_7_class_options(mc);
2353 SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_6);
2356 DEFINE_SPAPR_MACHINE(2_6, "2.6", false);
2359 * pseries-2.5
2361 #define SPAPR_COMPAT_2_5 \
2362 HW_COMPAT_2_5 \
2364 .driver = "spapr-vlan", \
2365 .property = "use-rx-buffer-pools", \
2366 .value = "off", \
2369 static void spapr_machine_2_5_instance_options(MachineState *machine)
2373 static void spapr_machine_2_5_class_options(MachineClass *mc)
2375 sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
2377 spapr_machine_2_6_class_options(mc);
2378 smc->use_ohci_by_default = true;
2379 SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_5);
2382 DEFINE_SPAPR_MACHINE(2_5, "2.5", false);
2385 * pseries-2.4
2387 #define SPAPR_COMPAT_2_4 \
2388 HW_COMPAT_2_4
2390 static void spapr_machine_2_4_instance_options(MachineState *machine)
2392 spapr_machine_2_5_instance_options(machine);
2395 static void spapr_machine_2_4_class_options(MachineClass *mc)
2397 sPAPRMachineClass *smc = SPAPR_MACHINE_CLASS(mc);
2399 spapr_machine_2_5_class_options(mc);
2400 smc->dr_lmb_enabled = false;
2401 SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_4);
2404 DEFINE_SPAPR_MACHINE(2_4, "2.4", false);
2407 * pseries-2.3
2409 #define SPAPR_COMPAT_2_3 \
2410 HW_COMPAT_2_3 \
2412 .driver = "spapr-pci-host-bridge",\
2413 .property = "dynamic-reconfiguration",\
2414 .value = "off",\
2417 static void spapr_machine_2_3_instance_options(MachineState *machine)
2419 spapr_machine_2_4_instance_options(machine);
2420 savevm_skip_section_footers();
2421 global_state_set_optional();
2422 savevm_skip_configuration();
2425 static void spapr_machine_2_3_class_options(MachineClass *mc)
2427 spapr_machine_2_4_class_options(mc);
2428 SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_3);
2430 DEFINE_SPAPR_MACHINE(2_3, "2.3", false);
2433 * pseries-2.2
2436 #define SPAPR_COMPAT_2_2 \
2437 HW_COMPAT_2_2 \
2439 .driver = TYPE_SPAPR_PCI_HOST_BRIDGE,\
2440 .property = "mem_win_size",\
2441 .value = "0x20000000",\
2444 static void spapr_machine_2_2_instance_options(MachineState *machine)
2446 spapr_machine_2_3_instance_options(machine);
2447 machine->suppress_vmdesc = true;
2450 static void spapr_machine_2_2_class_options(MachineClass *mc)
2452 spapr_machine_2_3_class_options(mc);
2453 SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_2);
2455 DEFINE_SPAPR_MACHINE(2_2, "2.2", false);
2458 * pseries-2.1
2460 #define SPAPR_COMPAT_2_1 \
2461 HW_COMPAT_2_1
2463 static void spapr_machine_2_1_instance_options(MachineState *machine)
2465 spapr_machine_2_2_instance_options(machine);
2468 static void spapr_machine_2_1_class_options(MachineClass *mc)
2470 spapr_machine_2_2_class_options(mc);
2471 SET_MACHINE_COMPAT(mc, SPAPR_COMPAT_2_1);
2473 DEFINE_SPAPR_MACHINE(2_1, "2.1", false);
2475 static void spapr_machine_register_types(void)
2477 type_register_static(&spapr_machine_info);
2480 type_init(spapr_machine_register_types)