Revert "gdbstub: Do not kill target in system emulation mode"
[qemu/qmp-unstable.git] / hw / ppc / spapr.c
blobac261ef050e8ad879077dca3ccc2d357c3347441
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 "sysemu/sysemu.h"
28 #include "sysemu/numa.h"
29 #include "hw/hw.h"
30 #include "hw/fw-path-provider.h"
31 #include "elf.h"
32 #include "net/net.h"
33 #include "sysemu/block-backend.h"
34 #include "sysemu/cpus.h"
35 #include "sysemu/kvm.h"
36 #include "kvm_ppc.h"
37 #include "mmu-hash64.h"
38 #include "qom/cpu.h"
40 #include "hw/boards.h"
41 #include "hw/ppc/ppc.h"
42 #include "hw/loader.h"
44 #include "hw/ppc/spapr.h"
45 #include "hw/ppc/spapr_vio.h"
46 #include "hw/pci-host/spapr.h"
47 #include "hw/ppc/xics.h"
48 #include "hw/pci/msi.h"
50 #include "hw/pci/pci.h"
51 #include "hw/scsi/scsi.h"
52 #include "hw/virtio/virtio-scsi.h"
54 #include "exec/address-spaces.h"
55 #include "hw/usb.h"
56 #include "qemu/config-file.h"
57 #include "qemu/error-report.h"
58 #include "trace.h"
59 #include "hw/nmi.h"
61 #include "hw/compat.h"
63 #include <libfdt.h>
65 /* SLOF memory layout:
67 * SLOF raw image loaded at 0, copies its romfs right below the flat
68 * device-tree, then position SLOF itself 31M below that
70 * So we set FW_OVERHEAD to 40MB which should account for all of that
71 * and more
73 * We load our kernel at 4M, leaving space for SLOF initial image
75 #define FDT_MAX_SIZE 0x40000
76 #define RTAS_MAX_SIZE 0x10000
77 #define RTAS_MAX_ADDR 0x80000000 /* RTAS must stay below that */
78 #define FW_MAX_SIZE 0x400000
79 #define FW_FILE_NAME "slof.bin"
80 #define FW_OVERHEAD 0x2800000
81 #define KERNEL_LOAD_ADDR FW_MAX_SIZE
83 #define MIN_RMA_SLOF 128UL
85 #define TIMEBASE_FREQ 512000000ULL
87 #define MAX_CPUS 255
89 #define PHANDLE_XICP 0x00001111
91 #define HTAB_SIZE(spapr) (1ULL << ((spapr)->htab_shift))
93 typedef struct sPAPRMachineState sPAPRMachineState;
95 #define TYPE_SPAPR_MACHINE "spapr-machine"
96 #define SPAPR_MACHINE(obj) \
97 OBJECT_CHECK(sPAPRMachineState, (obj), TYPE_SPAPR_MACHINE)
99 /**
100 * sPAPRMachineState:
102 struct sPAPRMachineState {
103 /*< private >*/
104 MachineState parent_obj;
106 /*< public >*/
107 char *kvm_type;
110 sPAPREnvironment *spapr;
112 static XICSState *try_create_xics(const char *type, int nr_servers,
113 int nr_irqs, Error **errp)
115 Error *err = NULL;
116 DeviceState *dev;
118 dev = qdev_create(NULL, type);
119 qdev_prop_set_uint32(dev, "nr_servers", nr_servers);
120 qdev_prop_set_uint32(dev, "nr_irqs", nr_irqs);
121 object_property_set_bool(OBJECT(dev), true, "realized", &err);
122 if (err) {
123 error_propagate(errp, err);
124 object_unparent(OBJECT(dev));
125 return NULL;
127 return XICS_COMMON(dev);
130 static XICSState *xics_system_init(MachineState *machine,
131 int nr_servers, int nr_irqs)
133 XICSState *icp = NULL;
135 if (kvm_enabled()) {
136 Error *err = NULL;
138 if (machine_kernel_irqchip_allowed(machine)) {
139 icp = try_create_xics(TYPE_KVM_XICS, nr_servers, nr_irqs, &err);
141 if (machine_kernel_irqchip_required(machine) && !icp) {
142 error_report("kernel_irqchip requested but unavailable: %s",
143 error_get_pretty(err));
147 if (!icp) {
148 icp = try_create_xics(TYPE_XICS, nr_servers, nr_irqs, &error_abort);
151 return icp;
154 static int spapr_fixup_cpu_smt_dt(void *fdt, int offset, PowerPCCPU *cpu,
155 int smt_threads)
157 int i, ret = 0;
158 uint32_t servers_prop[smt_threads];
159 uint32_t gservers_prop[smt_threads * 2];
160 int index = ppc_get_vcpu_dt_id(cpu);
162 if (cpu->cpu_version) {
163 ret = fdt_setprop_cell(fdt, offset, "cpu-version", cpu->cpu_version);
164 if (ret < 0) {
165 return ret;
169 /* Build interrupt servers and gservers properties */
170 for (i = 0; i < smt_threads; i++) {
171 servers_prop[i] = cpu_to_be32(index + i);
172 /* Hack, direct the group queues back to cpu 0 */
173 gservers_prop[i*2] = cpu_to_be32(index + i);
174 gservers_prop[i*2 + 1] = 0;
176 ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-server#s",
177 servers_prop, sizeof(servers_prop));
178 if (ret < 0) {
179 return ret;
181 ret = fdt_setprop(fdt, offset, "ibm,ppc-interrupt-gserver#s",
182 gservers_prop, sizeof(gservers_prop));
184 return ret;
187 static int spapr_fixup_cpu_dt(void *fdt, sPAPREnvironment *spapr)
189 int ret = 0, offset, cpus_offset;
190 CPUState *cs;
191 char cpu_model[32];
192 int smt = kvmppc_smt_threads();
193 uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
195 CPU_FOREACH(cs) {
196 PowerPCCPU *cpu = POWERPC_CPU(cs);
197 DeviceClass *dc = DEVICE_GET_CLASS(cs);
198 int index = ppc_get_vcpu_dt_id(cpu);
199 uint32_t associativity[] = {cpu_to_be32(0x5),
200 cpu_to_be32(0x0),
201 cpu_to_be32(0x0),
202 cpu_to_be32(0x0),
203 cpu_to_be32(cs->numa_node),
204 cpu_to_be32(index)};
206 if ((index % smt) != 0) {
207 continue;
210 snprintf(cpu_model, 32, "%s@%x", dc->fw_name, index);
212 cpus_offset = fdt_path_offset(fdt, "/cpus");
213 if (cpus_offset < 0) {
214 cpus_offset = fdt_add_subnode(fdt, fdt_path_offset(fdt, "/"),
215 "cpus");
216 if (cpus_offset < 0) {
217 return cpus_offset;
220 offset = fdt_subnode_offset(fdt, cpus_offset, cpu_model);
221 if (offset < 0) {
222 offset = fdt_add_subnode(fdt, cpus_offset, cpu_model);
223 if (offset < 0) {
224 return offset;
228 if (nb_numa_nodes > 1) {
229 ret = fdt_setprop(fdt, offset, "ibm,associativity", associativity,
230 sizeof(associativity));
231 if (ret < 0) {
232 return ret;
236 ret = fdt_setprop(fdt, offset, "ibm,pft-size",
237 pft_size_prop, sizeof(pft_size_prop));
238 if (ret < 0) {
239 return ret;
242 ret = spapr_fixup_cpu_smt_dt(fdt, offset, cpu,
243 ppc_get_compat_smt_threads(cpu));
244 if (ret < 0) {
245 return ret;
248 return ret;
252 static size_t create_page_sizes_prop(CPUPPCState *env, uint32_t *prop,
253 size_t maxsize)
255 size_t maxcells = maxsize / sizeof(uint32_t);
256 int i, j, count;
257 uint32_t *p = prop;
259 for (i = 0; i < PPC_PAGE_SIZES_MAX_SZ; i++) {
260 struct ppc_one_seg_page_size *sps = &env->sps.sps[i];
262 if (!sps->page_shift) {
263 break;
265 for (count = 0; count < PPC_PAGE_SIZES_MAX_SZ; count++) {
266 if (sps->enc[count].page_shift == 0) {
267 break;
270 if ((p - prop) >= (maxcells - 3 - count * 2)) {
271 break;
273 *(p++) = cpu_to_be32(sps->page_shift);
274 *(p++) = cpu_to_be32(sps->slb_enc);
275 *(p++) = cpu_to_be32(count);
276 for (j = 0; j < count; j++) {
277 *(p++) = cpu_to_be32(sps->enc[j].page_shift);
278 *(p++) = cpu_to_be32(sps->enc[j].pte_enc);
282 return (p - prop) * sizeof(uint32_t);
285 static hwaddr spapr_node0_size(void)
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), ram_size);
295 return ram_size;
298 #define _FDT(exp) \
299 do { \
300 int ret = (exp); \
301 if (ret < 0) { \
302 fprintf(stderr, "qemu: error creating device tree: %s: %s\n", \
303 #exp, fdt_strerror(ret)); \
304 exit(1); \
306 } while (0)
308 static void add_str(GString *s, const gchar *s1)
310 g_string_append_len(s, s1, strlen(s1) + 1);
313 static void *spapr_create_fdt_skel(hwaddr initrd_base,
314 hwaddr initrd_size,
315 hwaddr kernel_size,
316 bool little_endian,
317 const char *kernel_cmdline,
318 uint32_t epow_irq)
320 void *fdt;
321 CPUState *cs;
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(smp_cpus)};
328 int smt = kvmppc_smt_threads();
329 unsigned char vec5[] = {0x0, 0x0, 0x0, 0x0, 0x0, 0x80};
330 QemuOpts *opts = qemu_opts_find(qemu_find_opts("smp-opts"), NULL);
331 unsigned sockets = opts ? qemu_opt_get_number(opts, "sockets", 0) : 0;
332 uint32_t cpus_per_socket = sockets ? (smp_cpus / sockets) : 1;
333 char *buf;
335 add_str(hypertas, "hcall-pft");
336 add_str(hypertas, "hcall-term");
337 add_str(hypertas, "hcall-dabr");
338 add_str(hypertas, "hcall-interrupt");
339 add_str(hypertas, "hcall-tce");
340 add_str(hypertas, "hcall-vio");
341 add_str(hypertas, "hcall-splpar");
342 add_str(hypertas, "hcall-bulk");
343 add_str(hypertas, "hcall-set-mode");
344 add_str(qemu_hypertas, "hcall-memop1");
346 fdt = g_malloc0(FDT_MAX_SIZE);
347 _FDT((fdt_create(fdt, FDT_MAX_SIZE)));
349 if (kernel_size) {
350 _FDT((fdt_add_reservemap_entry(fdt, KERNEL_LOAD_ADDR, kernel_size)));
352 if (initrd_size) {
353 _FDT((fdt_add_reservemap_entry(fdt, initrd_base, initrd_size)));
355 _FDT((fdt_finish_reservemap(fdt)));
357 /* Root node */
358 _FDT((fdt_begin_node(fdt, "")));
359 _FDT((fdt_property_string(fdt, "device_type", "chrp")));
360 _FDT((fdt_property_string(fdt, "model", "IBM pSeries (emulated by qemu)")));
361 _FDT((fdt_property_string(fdt, "compatible", "qemu,pseries")));
364 * Add info to guest to indentify which host is it being run on
365 * and what is the uuid of the guest
367 if (kvmppc_get_host_model(&buf)) {
368 _FDT((fdt_property_string(fdt, "host-model", buf)));
369 g_free(buf);
371 if (kvmppc_get_host_serial(&buf)) {
372 _FDT((fdt_property_string(fdt, "host-serial", buf)));
373 g_free(buf);
376 buf = g_strdup_printf(UUID_FMT, qemu_uuid[0], qemu_uuid[1],
377 qemu_uuid[2], qemu_uuid[3], qemu_uuid[4],
378 qemu_uuid[5], qemu_uuid[6], qemu_uuid[7],
379 qemu_uuid[8], qemu_uuid[9], qemu_uuid[10],
380 qemu_uuid[11], qemu_uuid[12], qemu_uuid[13],
381 qemu_uuid[14], qemu_uuid[15]);
383 _FDT((fdt_property_string(fdt, "vm,uuid", buf)));
384 g_free(buf);
386 _FDT((fdt_property_cell(fdt, "#address-cells", 0x2)));
387 _FDT((fdt_property_cell(fdt, "#size-cells", 0x2)));
389 /* /chosen */
390 _FDT((fdt_begin_node(fdt, "chosen")));
392 /* Set Form1_affinity */
393 _FDT((fdt_property(fdt, "ibm,architecture-vec-5", vec5, sizeof(vec5))));
395 _FDT((fdt_property_string(fdt, "bootargs", kernel_cmdline)));
396 _FDT((fdt_property(fdt, "linux,initrd-start",
397 &start_prop, sizeof(start_prop))));
398 _FDT((fdt_property(fdt, "linux,initrd-end",
399 &end_prop, sizeof(end_prop))));
400 if (kernel_size) {
401 uint64_t kprop[2] = { cpu_to_be64(KERNEL_LOAD_ADDR),
402 cpu_to_be64(kernel_size) };
404 _FDT((fdt_property(fdt, "qemu,boot-kernel", &kprop, sizeof(kprop))));
405 if (little_endian) {
406 _FDT((fdt_property(fdt, "qemu,boot-kernel-le", NULL, 0)));
409 if (boot_menu) {
410 _FDT((fdt_property_cell(fdt, "qemu,boot-menu", boot_menu)));
412 _FDT((fdt_property_cell(fdt, "qemu,graphic-width", graphic_width)));
413 _FDT((fdt_property_cell(fdt, "qemu,graphic-height", graphic_height)));
414 _FDT((fdt_property_cell(fdt, "qemu,graphic-depth", graphic_depth)));
416 _FDT((fdt_end_node(fdt)));
418 /* cpus */
419 _FDT((fdt_begin_node(fdt, "cpus")));
421 _FDT((fdt_property_cell(fdt, "#address-cells", 0x1)));
422 _FDT((fdt_property_cell(fdt, "#size-cells", 0x0)));
424 CPU_FOREACH(cs) {
425 PowerPCCPU *cpu = POWERPC_CPU(cs);
426 CPUPPCState *env = &cpu->env;
427 DeviceClass *dc = DEVICE_GET_CLASS(cs);
428 PowerPCCPUClass *pcc = POWERPC_CPU_GET_CLASS(cs);
429 int index = ppc_get_vcpu_dt_id(cpu);
430 char *nodename;
431 uint32_t segs[] = {cpu_to_be32(28), cpu_to_be32(40),
432 0xffffffff, 0xffffffff};
433 uint32_t tbfreq = kvm_enabled() ? kvmppc_get_tbfreq() : TIMEBASE_FREQ;
434 uint32_t cpufreq = kvm_enabled() ? kvmppc_get_clockfreq() : 1000000000;
435 uint32_t page_sizes_prop[64];
436 size_t page_sizes_prop_size;
438 if ((index % smt) != 0) {
439 continue;
442 nodename = g_strdup_printf("%s@%x", dc->fw_name, index);
444 _FDT((fdt_begin_node(fdt, nodename)));
446 g_free(nodename);
448 _FDT((fdt_property_cell(fdt, "reg", index)));
449 _FDT((fdt_property_string(fdt, "device_type", "cpu")));
451 _FDT((fdt_property_cell(fdt, "cpu-version", env->spr[SPR_PVR])));
452 _FDT((fdt_property_cell(fdt, "d-cache-block-size",
453 env->dcache_line_size)));
454 _FDT((fdt_property_cell(fdt, "d-cache-line-size",
455 env->dcache_line_size)));
456 _FDT((fdt_property_cell(fdt, "i-cache-block-size",
457 env->icache_line_size)));
458 _FDT((fdt_property_cell(fdt, "i-cache-line-size",
459 env->icache_line_size)));
461 if (pcc->l1_dcache_size) {
462 _FDT((fdt_property_cell(fdt, "d-cache-size", pcc->l1_dcache_size)));
463 } else {
464 fprintf(stderr, "Warning: Unknown L1 dcache size for cpu\n");
466 if (pcc->l1_icache_size) {
467 _FDT((fdt_property_cell(fdt, "i-cache-size", pcc->l1_icache_size)));
468 } else {
469 fprintf(stderr, "Warning: Unknown L1 icache size for cpu\n");
472 _FDT((fdt_property_cell(fdt, "timebase-frequency", tbfreq)));
473 _FDT((fdt_property_cell(fdt, "clock-frequency", cpufreq)));
474 _FDT((fdt_property_cell(fdt, "ibm,slb-size", env->slb_nr)));
475 _FDT((fdt_property_string(fdt, "status", "okay")));
476 _FDT((fdt_property(fdt, "64-bit", NULL, 0)));
478 if (env->spr_cb[SPR_PURR].oea_read) {
479 _FDT((fdt_property(fdt, "ibm,purr", NULL, 0)));
482 if (env->mmu_model & POWERPC_MMU_1TSEG) {
483 _FDT((fdt_property(fdt, "ibm,processor-segment-sizes",
484 segs, sizeof(segs))));
487 /* Advertise VMX/VSX (vector extensions) if available
488 * 0 / no property == no vector extensions
489 * 1 == VMX / Altivec available
490 * 2 == VSX available */
491 if (env->insns_flags & PPC_ALTIVEC) {
492 uint32_t vmx = (env->insns_flags2 & PPC2_VSX) ? 2 : 1;
494 _FDT((fdt_property_cell(fdt, "ibm,vmx", vmx)));
497 /* Advertise DFP (Decimal Floating Point) if available
498 * 0 / no property == no DFP
499 * 1 == DFP available */
500 if (env->insns_flags2 & PPC2_DFP) {
501 _FDT((fdt_property_cell(fdt, "ibm,dfp", 1)));
504 page_sizes_prop_size = create_page_sizes_prop(env, page_sizes_prop,
505 sizeof(page_sizes_prop));
506 if (page_sizes_prop_size) {
507 _FDT((fdt_property(fdt, "ibm,segment-page-sizes",
508 page_sizes_prop, page_sizes_prop_size)));
511 _FDT((fdt_property_cell(fdt, "ibm,chip-id",
512 cs->cpu_index / cpus_per_socket)));
514 _FDT((fdt_end_node(fdt)));
517 _FDT((fdt_end_node(fdt)));
519 /* RTAS */
520 _FDT((fdt_begin_node(fdt, "rtas")));
522 if (!kvm_enabled() || kvmppc_spapr_use_multitce()) {
523 add_str(hypertas, "hcall-multi-tce");
525 _FDT((fdt_property(fdt, "ibm,hypertas-functions", hypertas->str,
526 hypertas->len)));
527 g_string_free(hypertas, TRUE);
528 _FDT((fdt_property(fdt, "qemu,hypertas-functions", qemu_hypertas->str,
529 qemu_hypertas->len)));
530 g_string_free(qemu_hypertas, TRUE);
532 _FDT((fdt_property(fdt, "ibm,associativity-reference-points",
533 refpoints, sizeof(refpoints))));
535 _FDT((fdt_property_cell(fdt, "rtas-error-log-max", RTAS_ERROR_LOG_MAX)));
538 * According to PAPR, rtas ibm,os-term does not guarantee a return
539 * back to the guest cpu.
541 * While an additional ibm,extended-os-term property indicates that
542 * rtas call return will always occur. Set this property.
544 _FDT((fdt_property(fdt, "ibm,extended-os-term", NULL, 0)));
546 _FDT((fdt_end_node(fdt)));
548 /* interrupt controller */
549 _FDT((fdt_begin_node(fdt, "interrupt-controller")));
551 _FDT((fdt_property_string(fdt, "device_type",
552 "PowerPC-External-Interrupt-Presentation")));
553 _FDT((fdt_property_string(fdt, "compatible", "IBM,ppc-xicp")));
554 _FDT((fdt_property(fdt, "interrupt-controller", NULL, 0)));
555 _FDT((fdt_property(fdt, "ibm,interrupt-server-ranges",
556 interrupt_server_ranges_prop,
557 sizeof(interrupt_server_ranges_prop))));
558 _FDT((fdt_property_cell(fdt, "#interrupt-cells", 2)));
559 _FDT((fdt_property_cell(fdt, "linux,phandle", PHANDLE_XICP)));
560 _FDT((fdt_property_cell(fdt, "phandle", PHANDLE_XICP)));
562 _FDT((fdt_end_node(fdt)));
564 /* vdevice */
565 _FDT((fdt_begin_node(fdt, "vdevice")));
567 _FDT((fdt_property_string(fdt, "device_type", "vdevice")));
568 _FDT((fdt_property_string(fdt, "compatible", "IBM,vdevice")));
569 _FDT((fdt_property_cell(fdt, "#address-cells", 0x1)));
570 _FDT((fdt_property_cell(fdt, "#size-cells", 0x0)));
571 _FDT((fdt_property_cell(fdt, "#interrupt-cells", 0x2)));
572 _FDT((fdt_property(fdt, "interrupt-controller", NULL, 0)));
574 _FDT((fdt_end_node(fdt)));
576 /* event-sources */
577 spapr_events_fdt_skel(fdt, epow_irq);
579 /* /hypervisor node */
580 if (kvm_enabled()) {
581 uint8_t hypercall[16];
583 /* indicate KVM hypercall interface */
584 _FDT((fdt_begin_node(fdt, "hypervisor")));
585 _FDT((fdt_property_string(fdt, "compatible", "linux,kvm")));
586 if (kvmppc_has_cap_fixup_hcalls()) {
588 * Older KVM versions with older guest kernels were broken with the
589 * magic page, don't allow the guest to map it.
591 kvmppc_get_hypercall(first_cpu->env_ptr, hypercall,
592 sizeof(hypercall));
593 _FDT((fdt_property(fdt, "hcall-instructions", hypercall,
594 sizeof(hypercall))));
596 _FDT((fdt_end_node(fdt)));
599 _FDT((fdt_end_node(fdt))); /* close root node */
600 _FDT((fdt_finish(fdt)));
602 return fdt;
605 int spapr_h_cas_compose_response(target_ulong addr, target_ulong size)
607 void *fdt, *fdt_skel;
608 sPAPRDeviceTreeUpdateHeader hdr = { .version_id = 1 };
610 size -= sizeof(hdr);
612 /* Create sceleton */
613 fdt_skel = g_malloc0(size);
614 _FDT((fdt_create(fdt_skel, size)));
615 _FDT((fdt_begin_node(fdt_skel, "")));
616 _FDT((fdt_end_node(fdt_skel)));
617 _FDT((fdt_finish(fdt_skel)));
618 fdt = g_malloc0(size);
619 _FDT((fdt_open_into(fdt_skel, fdt, size)));
620 g_free(fdt_skel);
622 /* Fix skeleton up */
623 _FDT((spapr_fixup_cpu_dt(fdt, spapr)));
625 /* Pack resulting tree */
626 _FDT((fdt_pack(fdt)));
628 if (fdt_totalsize(fdt) + sizeof(hdr) > size) {
629 trace_spapr_cas_failed(size);
630 return -1;
633 cpu_physical_memory_write(addr, &hdr, sizeof(hdr));
634 cpu_physical_memory_write(addr + sizeof(hdr), fdt, fdt_totalsize(fdt));
635 trace_spapr_cas_continue(fdt_totalsize(fdt) + sizeof(hdr));
636 g_free(fdt);
638 return 0;
641 static void spapr_populate_memory_node(void *fdt, int nodeid, hwaddr start,
642 hwaddr size)
644 uint32_t associativity[] = {
645 cpu_to_be32(0x4), /* length */
646 cpu_to_be32(0x0), cpu_to_be32(0x0),
647 cpu_to_be32(0x0), cpu_to_be32(nodeid)
649 char mem_name[32];
650 uint64_t mem_reg_property[2];
651 int off;
653 mem_reg_property[0] = cpu_to_be64(start);
654 mem_reg_property[1] = cpu_to_be64(size);
656 sprintf(mem_name, "memory@" TARGET_FMT_lx, start);
657 off = fdt_add_subnode(fdt, 0, mem_name);
658 _FDT(off);
659 _FDT((fdt_setprop_string(fdt, off, "device_type", "memory")));
660 _FDT((fdt_setprop(fdt, off, "reg", mem_reg_property,
661 sizeof(mem_reg_property))));
662 _FDT((fdt_setprop(fdt, off, "ibm,associativity", associativity,
663 sizeof(associativity))));
666 static int spapr_populate_memory(sPAPREnvironment *spapr, void *fdt)
668 hwaddr mem_start, node_size;
669 int i, nb_nodes = nb_numa_nodes;
670 NodeInfo *nodes = numa_info;
671 NodeInfo ramnode;
673 /* No NUMA nodes, assume there is just one node with whole RAM */
674 if (!nb_numa_nodes) {
675 nb_nodes = 1;
676 ramnode.node_mem = ram_size;
677 nodes = &ramnode;
680 for (i = 0, mem_start = 0; i < nb_nodes; ++i) {
681 if (!nodes[i].node_mem) {
682 continue;
684 if (mem_start >= ram_size) {
685 node_size = 0;
686 } else {
687 node_size = nodes[i].node_mem;
688 if (node_size > ram_size - mem_start) {
689 node_size = ram_size - mem_start;
692 if (!mem_start) {
693 /* ppc_spapr_init() checks for rma_size <= node0_size already */
694 spapr_populate_memory_node(fdt, i, 0, spapr->rma_size);
695 mem_start += spapr->rma_size;
696 node_size -= spapr->rma_size;
698 for ( ; node_size; ) {
699 hwaddr sizetmp = pow2floor(node_size);
701 /* mem_start != 0 here */
702 if (ctzl(mem_start) < ctzl(sizetmp)) {
703 sizetmp = 1ULL << ctzl(mem_start);
706 spapr_populate_memory_node(fdt, i, mem_start, sizetmp);
707 node_size -= sizetmp;
708 mem_start += sizetmp;
712 return 0;
715 static void spapr_finalize_fdt(sPAPREnvironment *spapr,
716 hwaddr fdt_addr,
717 hwaddr rtas_addr,
718 hwaddr rtas_size)
720 MachineState *machine = MACHINE(qdev_get_machine());
721 const char *boot_device = machine->boot_order;
722 int ret, i;
723 size_t cb = 0;
724 char *bootlist;
725 void *fdt;
726 sPAPRPHBState *phb;
728 fdt = g_malloc(FDT_MAX_SIZE);
730 /* open out the base tree into a temp buffer for the final tweaks */
731 _FDT((fdt_open_into(spapr->fdt_skel, fdt, FDT_MAX_SIZE)));
733 ret = spapr_populate_memory(spapr, fdt);
734 if (ret < 0) {
735 fprintf(stderr, "couldn't setup memory nodes in fdt\n");
736 exit(1);
739 ret = spapr_populate_vdevice(spapr->vio_bus, fdt);
740 if (ret < 0) {
741 fprintf(stderr, "couldn't setup vio devices in fdt\n");
742 exit(1);
745 QLIST_FOREACH(phb, &spapr->phbs, list) {
746 ret = spapr_populate_pci_dt(phb, PHANDLE_XICP, fdt);
749 if (ret < 0) {
750 fprintf(stderr, "couldn't setup PCI devices in fdt\n");
751 exit(1);
754 /* RTAS */
755 ret = spapr_rtas_device_tree_setup(fdt, rtas_addr, rtas_size);
756 if (ret < 0) {
757 fprintf(stderr, "Couldn't set up RTAS device tree properties\n");
760 /* Advertise NUMA via ibm,associativity */
761 ret = spapr_fixup_cpu_dt(fdt, spapr);
762 if (ret < 0) {
763 fprintf(stderr, "Couldn't finalize CPU device tree properties\n");
766 bootlist = get_boot_devices_list(&cb, true);
767 if (cb && bootlist) {
768 int offset = fdt_path_offset(fdt, "/chosen");
769 if (offset < 0) {
770 exit(1);
772 for (i = 0; i < cb; i++) {
773 if (bootlist[i] == '\n') {
774 bootlist[i] = ' ';
778 ret = fdt_setprop_string(fdt, offset, "qemu,boot-list", bootlist);
781 if (boot_device && strlen(boot_device)) {
782 int offset = fdt_path_offset(fdt, "/chosen");
784 if (offset < 0) {
785 exit(1);
787 fdt_setprop_string(fdt, offset, "qemu,boot-device", boot_device);
790 if (!spapr->has_graphics) {
791 spapr_populate_chosen_stdout(fdt, spapr->vio_bus);
794 _FDT((fdt_pack(fdt)));
796 if (fdt_totalsize(fdt) > FDT_MAX_SIZE) {
797 hw_error("FDT too big ! 0x%x bytes (max is 0x%x)\n",
798 fdt_totalsize(fdt), FDT_MAX_SIZE);
799 exit(1);
802 cpu_physical_memory_write(fdt_addr, fdt, fdt_totalsize(fdt));
804 g_free(bootlist);
805 g_free(fdt);
808 static uint64_t translate_kernel_address(void *opaque, uint64_t addr)
810 return (addr & 0x0fffffff) + KERNEL_LOAD_ADDR;
813 static void emulate_spapr_hypercall(PowerPCCPU *cpu)
815 CPUPPCState *env = &cpu->env;
817 if (msr_pr) {
818 hcall_dprintf("Hypercall made with MSR[PR]=1\n");
819 env->gpr[3] = H_PRIVILEGE;
820 } else {
821 env->gpr[3] = spapr_hypercall(cpu, env->gpr[3], &env->gpr[4]);
825 #define HPTE(_table, _i) (void *)(((uint64_t *)(_table)) + ((_i) * 2))
826 #define HPTE_VALID(_hpte) (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_VALID)
827 #define HPTE_DIRTY(_hpte) (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_HPTE_DIRTY)
828 #define CLEAN_HPTE(_hpte) ((*(uint64_t *)(_hpte)) &= tswap64(~HPTE64_V_HPTE_DIRTY))
829 #define DIRTY_HPTE(_hpte) ((*(uint64_t *)(_hpte)) |= tswap64(HPTE64_V_HPTE_DIRTY))
831 static void spapr_reset_htab(sPAPREnvironment *spapr)
833 long shift;
834 int index;
836 /* allocate hash page table. For now we always make this 16mb,
837 * later we should probably make it scale to the size of guest
838 * RAM */
840 shift = kvmppc_reset_htab(spapr->htab_shift);
842 if (shift > 0) {
843 /* Kernel handles htab, we don't need to allocate one */
844 spapr->htab_shift = shift;
845 kvmppc_kern_htab = true;
847 /* Tell readers to update their file descriptor */
848 if (spapr->htab_fd >= 0) {
849 spapr->htab_fd_stale = true;
851 } else {
852 if (!spapr->htab) {
853 /* Allocate an htab if we don't yet have one */
854 spapr->htab = qemu_memalign(HTAB_SIZE(spapr), HTAB_SIZE(spapr));
857 /* And clear it */
858 memset(spapr->htab, 0, HTAB_SIZE(spapr));
860 for (index = 0; index < HTAB_SIZE(spapr) / HASH_PTE_SIZE_64; index++) {
861 DIRTY_HPTE(HPTE(spapr->htab, index));
865 /* Update the RMA size if necessary */
866 if (spapr->vrma_adjust) {
867 spapr->rma_size = kvmppc_rma_size(spapr_node0_size(),
868 spapr->htab_shift);
872 static int find_unknown_sysbus_device(SysBusDevice *sbdev, void *opaque)
874 bool matched = false;
876 if (object_dynamic_cast(OBJECT(sbdev), TYPE_SPAPR_PCI_HOST_BRIDGE)) {
877 matched = true;
880 if (!matched) {
881 error_report("Device %s is not supported by this machine yet.",
882 qdev_fw_name(DEVICE(sbdev)));
883 exit(1);
886 return 0;
890 * A guest reset will cause spapr->htab_fd to become stale if being used.
891 * Reopen the file descriptor to make sure the whole HTAB is properly read.
893 static int spapr_check_htab_fd(sPAPREnvironment *spapr)
895 int rc = 0;
897 if (spapr->htab_fd_stale) {
898 close(spapr->htab_fd);
899 spapr->htab_fd = kvmppc_get_htab_fd(false);
900 if (spapr->htab_fd < 0) {
901 error_report("Unable to open fd for reading hash table from KVM: "
902 "%s", strerror(errno));
903 rc = -1;
905 spapr->htab_fd_stale = false;
908 return rc;
911 static void ppc_spapr_reset(void)
913 PowerPCCPU *first_ppc_cpu;
914 uint32_t rtas_limit;
916 /* Check for unknown sysbus devices */
917 foreach_dynamic_sysbus_device(find_unknown_sysbus_device, NULL);
919 /* Reset the hash table & recalc the RMA */
920 spapr_reset_htab(spapr);
922 qemu_devices_reset();
925 * We place the device tree and RTAS just below either the top of the RMA,
926 * or just below 2GB, whichever is lowere, so that it can be
927 * processed with 32-bit real mode code if necessary
929 rtas_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR);
930 spapr->rtas_addr = rtas_limit - RTAS_MAX_SIZE;
931 spapr->fdt_addr = spapr->rtas_addr - FDT_MAX_SIZE;
933 /* Load the fdt */
934 spapr_finalize_fdt(spapr, spapr->fdt_addr, spapr->rtas_addr,
935 spapr->rtas_size);
937 /* Copy RTAS over */
938 cpu_physical_memory_write(spapr->rtas_addr, spapr->rtas_blob,
939 spapr->rtas_size);
941 /* Set up the entry state */
942 first_ppc_cpu = POWERPC_CPU(first_cpu);
943 first_ppc_cpu->env.gpr[3] = spapr->fdt_addr;
944 first_ppc_cpu->env.gpr[5] = 0;
945 first_cpu->halted = 0;
946 first_ppc_cpu->env.nip = spapr->entry_point;
950 static void spapr_cpu_reset(void *opaque)
952 PowerPCCPU *cpu = opaque;
953 CPUState *cs = CPU(cpu);
954 CPUPPCState *env = &cpu->env;
956 cpu_reset(cs);
958 /* All CPUs start halted. CPU0 is unhalted from the machine level
959 * reset code and the rest are explicitly started up by the guest
960 * using an RTAS call */
961 cs->halted = 1;
963 env->spr[SPR_HIOR] = 0;
965 env->external_htab = (uint8_t *)spapr->htab;
966 if (kvm_enabled() && !env->external_htab) {
968 * HV KVM, set external_htab to 1 so our ppc_hash64_load_hpte*
969 * functions do the right thing.
971 env->external_htab = (void *)1;
973 env->htab_base = -1;
975 * htab_mask is the mask used to normalize hash value to PTEG index.
976 * htab_shift is log2 of hash table size.
977 * We have 8 hpte per group, and each hpte is 16 bytes.
978 * ie have 128 bytes per hpte entry.
980 env->htab_mask = (1ULL << ((spapr)->htab_shift - 7)) - 1;
981 env->spr[SPR_SDR1] = (target_ulong)(uintptr_t)spapr->htab |
982 (spapr->htab_shift - 18);
985 static void spapr_create_nvram(sPAPREnvironment *spapr)
987 DeviceState *dev = qdev_create(&spapr->vio_bus->bus, "spapr-nvram");
988 DriveInfo *dinfo = drive_get(IF_PFLASH, 0, 0);
990 if (dinfo) {
991 qdev_prop_set_drive_nofail(dev, "drive", blk_by_legacy_dinfo(dinfo));
994 qdev_init_nofail(dev);
996 spapr->nvram = (struct sPAPRNVRAM *)dev;
999 static void spapr_rtc_create(sPAPREnvironment *spapr)
1001 DeviceState *dev = qdev_create(NULL, TYPE_SPAPR_RTC);
1003 qdev_init_nofail(dev);
1004 spapr->rtc = dev;
1006 object_property_add_alias(qdev_get_machine(), "rtc-time",
1007 OBJECT(spapr->rtc), "date", NULL);
1010 /* Returns whether we want to use VGA or not */
1011 static int spapr_vga_init(PCIBus *pci_bus)
1013 switch (vga_interface_type) {
1014 case VGA_NONE:
1015 return false;
1016 case VGA_DEVICE:
1017 return true;
1018 case VGA_STD:
1019 return pci_vga_init(pci_bus) != NULL;
1020 default:
1021 fprintf(stderr, "This vga model is not supported,"
1022 "currently it only supports -vga std\n");
1023 exit(0);
1027 static int spapr_post_load(void *opaque, int version_id)
1029 sPAPREnvironment *spapr = (sPAPREnvironment *)opaque;
1030 int err = 0;
1032 /* In earlier versions, there was no separate qdev for the PAPR
1033 * RTC, so the RTC offset was stored directly in sPAPREnvironment.
1034 * So when migrating from those versions, poke the incoming offset
1035 * value into the RTC device */
1036 if (version_id < 3) {
1037 err = spapr_rtc_import_offset(spapr->rtc, spapr->rtc_offset);
1040 return err;
1043 static bool version_before_3(void *opaque, int version_id)
1045 return version_id < 3;
1048 static const VMStateDescription vmstate_spapr = {
1049 .name = "spapr",
1050 .version_id = 3,
1051 .minimum_version_id = 1,
1052 .post_load = spapr_post_load,
1053 .fields = (VMStateField[]) {
1054 /* used to be @next_irq */
1055 VMSTATE_UNUSED_BUFFER(version_before_3, 0, 4),
1057 /* RTC offset */
1058 VMSTATE_UINT64_TEST(rtc_offset, sPAPREnvironment, version_before_3),
1060 VMSTATE_PPC_TIMEBASE_V(tb, sPAPREnvironment, 2),
1061 VMSTATE_END_OF_LIST()
1065 static int htab_save_setup(QEMUFile *f, void *opaque)
1067 sPAPREnvironment *spapr = opaque;
1069 /* "Iteration" header */
1070 qemu_put_be32(f, spapr->htab_shift);
1072 if (spapr->htab) {
1073 spapr->htab_save_index = 0;
1074 spapr->htab_first_pass = true;
1075 } else {
1076 assert(kvm_enabled());
1078 spapr->htab_fd = kvmppc_get_htab_fd(false);
1079 spapr->htab_fd_stale = false;
1080 if (spapr->htab_fd < 0) {
1081 fprintf(stderr, "Unable to open fd for reading hash table from KVM: %s\n",
1082 strerror(errno));
1083 return -1;
1088 return 0;
1091 static void htab_save_first_pass(QEMUFile *f, sPAPREnvironment *spapr,
1092 int64_t max_ns)
1094 int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
1095 int index = spapr->htab_save_index;
1096 int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
1098 assert(spapr->htab_first_pass);
1100 do {
1101 int chunkstart;
1103 /* Consume invalid HPTEs */
1104 while ((index < htabslots)
1105 && !HPTE_VALID(HPTE(spapr->htab, index))) {
1106 index++;
1107 CLEAN_HPTE(HPTE(spapr->htab, index));
1110 /* Consume valid HPTEs */
1111 chunkstart = index;
1112 while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
1113 && HPTE_VALID(HPTE(spapr->htab, index))) {
1114 index++;
1115 CLEAN_HPTE(HPTE(spapr->htab, index));
1118 if (index > chunkstart) {
1119 int n_valid = index - chunkstart;
1121 qemu_put_be32(f, chunkstart);
1122 qemu_put_be16(f, n_valid);
1123 qemu_put_be16(f, 0);
1124 qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
1125 HASH_PTE_SIZE_64 * n_valid);
1127 if ((qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
1128 break;
1131 } while ((index < htabslots) && !qemu_file_rate_limit(f));
1133 if (index >= htabslots) {
1134 assert(index == htabslots);
1135 index = 0;
1136 spapr->htab_first_pass = false;
1138 spapr->htab_save_index = index;
1141 static int htab_save_later_pass(QEMUFile *f, sPAPREnvironment *spapr,
1142 int64_t max_ns)
1144 bool final = max_ns < 0;
1145 int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
1146 int examined = 0, sent = 0;
1147 int index = spapr->htab_save_index;
1148 int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
1150 assert(!spapr->htab_first_pass);
1152 do {
1153 int chunkstart, invalidstart;
1155 /* Consume non-dirty HPTEs */
1156 while ((index < htabslots)
1157 && !HPTE_DIRTY(HPTE(spapr->htab, index))) {
1158 index++;
1159 examined++;
1162 chunkstart = index;
1163 /* Consume valid dirty HPTEs */
1164 while ((index < htabslots) && (index - chunkstart < USHRT_MAX)
1165 && HPTE_DIRTY(HPTE(spapr->htab, index))
1166 && HPTE_VALID(HPTE(spapr->htab, index))) {
1167 CLEAN_HPTE(HPTE(spapr->htab, index));
1168 index++;
1169 examined++;
1172 invalidstart = index;
1173 /* Consume invalid dirty HPTEs */
1174 while ((index < htabslots) && (index - invalidstart < USHRT_MAX)
1175 && HPTE_DIRTY(HPTE(spapr->htab, index))
1176 && !HPTE_VALID(HPTE(spapr->htab, index))) {
1177 CLEAN_HPTE(HPTE(spapr->htab, index));
1178 index++;
1179 examined++;
1182 if (index > chunkstart) {
1183 int n_valid = invalidstart - chunkstart;
1184 int n_invalid = index - invalidstart;
1186 qemu_put_be32(f, chunkstart);
1187 qemu_put_be16(f, n_valid);
1188 qemu_put_be16(f, n_invalid);
1189 qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
1190 HASH_PTE_SIZE_64 * n_valid);
1191 sent += index - chunkstart;
1193 if (!final && (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
1194 break;
1198 if (examined >= htabslots) {
1199 break;
1202 if (index >= htabslots) {
1203 assert(index == htabslots);
1204 index = 0;
1206 } while ((examined < htabslots) && (!qemu_file_rate_limit(f) || final));
1208 if (index >= htabslots) {
1209 assert(index == htabslots);
1210 index = 0;
1213 spapr->htab_save_index = index;
1215 return (examined >= htabslots) && (sent == 0) ? 1 : 0;
1218 #define MAX_ITERATION_NS 5000000 /* 5 ms */
1219 #define MAX_KVM_BUF_SIZE 2048
1221 static int htab_save_iterate(QEMUFile *f, void *opaque)
1223 sPAPREnvironment *spapr = opaque;
1224 int rc = 0;
1226 /* Iteration header */
1227 qemu_put_be32(f, 0);
1229 if (!spapr->htab) {
1230 assert(kvm_enabled());
1232 rc = spapr_check_htab_fd(spapr);
1233 if (rc < 0) {
1234 return rc;
1237 rc = kvmppc_save_htab(f, spapr->htab_fd,
1238 MAX_KVM_BUF_SIZE, MAX_ITERATION_NS);
1239 if (rc < 0) {
1240 return rc;
1242 } else if (spapr->htab_first_pass) {
1243 htab_save_first_pass(f, spapr, MAX_ITERATION_NS);
1244 } else {
1245 rc = htab_save_later_pass(f, spapr, MAX_ITERATION_NS);
1248 /* End marker */
1249 qemu_put_be32(f, 0);
1250 qemu_put_be16(f, 0);
1251 qemu_put_be16(f, 0);
1253 return rc;
1256 static int htab_save_complete(QEMUFile *f, void *opaque)
1258 sPAPREnvironment *spapr = opaque;
1260 /* Iteration header */
1261 qemu_put_be32(f, 0);
1263 if (!spapr->htab) {
1264 int rc;
1266 assert(kvm_enabled());
1268 rc = spapr_check_htab_fd(spapr);
1269 if (rc < 0) {
1270 return rc;
1273 rc = kvmppc_save_htab(f, spapr->htab_fd, MAX_KVM_BUF_SIZE, -1);
1274 if (rc < 0) {
1275 return rc;
1277 close(spapr->htab_fd);
1278 spapr->htab_fd = -1;
1279 } else {
1280 htab_save_later_pass(f, spapr, -1);
1283 /* End marker */
1284 qemu_put_be32(f, 0);
1285 qemu_put_be16(f, 0);
1286 qemu_put_be16(f, 0);
1288 return 0;
1291 static int htab_load(QEMUFile *f, void *opaque, int version_id)
1293 sPAPREnvironment *spapr = opaque;
1294 uint32_t section_hdr;
1295 int fd = -1;
1297 if (version_id < 1 || version_id > 1) {
1298 fprintf(stderr, "htab_load() bad version\n");
1299 return -EINVAL;
1302 section_hdr = qemu_get_be32(f);
1304 if (section_hdr) {
1305 /* First section, just the hash shift */
1306 if (spapr->htab_shift != section_hdr) {
1307 return -EINVAL;
1309 return 0;
1312 if (!spapr->htab) {
1313 assert(kvm_enabled());
1315 fd = kvmppc_get_htab_fd(true);
1316 if (fd < 0) {
1317 fprintf(stderr, "Unable to open fd to restore KVM hash table: %s\n",
1318 strerror(errno));
1322 while (true) {
1323 uint32_t index;
1324 uint16_t n_valid, n_invalid;
1326 index = qemu_get_be32(f);
1327 n_valid = qemu_get_be16(f);
1328 n_invalid = qemu_get_be16(f);
1330 if ((index == 0) && (n_valid == 0) && (n_invalid == 0)) {
1331 /* End of Stream */
1332 break;
1335 if ((index + n_valid + n_invalid) >
1336 (HTAB_SIZE(spapr) / HASH_PTE_SIZE_64)) {
1337 /* Bad index in stream */
1338 fprintf(stderr, "htab_load() bad index %d (%hd+%hd entries) "
1339 "in htab stream (htab_shift=%d)\n", index, n_valid, n_invalid,
1340 spapr->htab_shift);
1341 return -EINVAL;
1344 if (spapr->htab) {
1345 if (n_valid) {
1346 qemu_get_buffer(f, HPTE(spapr->htab, index),
1347 HASH_PTE_SIZE_64 * n_valid);
1349 if (n_invalid) {
1350 memset(HPTE(spapr->htab, index + n_valid), 0,
1351 HASH_PTE_SIZE_64 * n_invalid);
1353 } else {
1354 int rc;
1356 assert(fd >= 0);
1358 rc = kvmppc_load_htab_chunk(f, fd, index, n_valid, n_invalid);
1359 if (rc < 0) {
1360 return rc;
1365 if (!spapr->htab) {
1366 assert(fd >= 0);
1367 close(fd);
1370 return 0;
1373 static SaveVMHandlers savevm_htab_handlers = {
1374 .save_live_setup = htab_save_setup,
1375 .save_live_iterate = htab_save_iterate,
1376 .save_live_complete = htab_save_complete,
1377 .load_state = htab_load,
1380 static void spapr_boot_set(void *opaque, const char *boot_device,
1381 Error **errp)
1383 MachineState *machine = MACHINE(qdev_get_machine());
1384 machine->boot_order = g_strdup(boot_device);
1387 /* pSeries LPAR / sPAPR hardware init */
1388 static void ppc_spapr_init(MachineState *machine)
1390 ram_addr_t ram_size = machine->ram_size;
1391 const char *cpu_model = machine->cpu_model;
1392 const char *kernel_filename = machine->kernel_filename;
1393 const char *kernel_cmdline = machine->kernel_cmdline;
1394 const char *initrd_filename = machine->initrd_filename;
1395 PowerPCCPU *cpu;
1396 CPUPPCState *env;
1397 PCIHostState *phb;
1398 int i;
1399 MemoryRegion *sysmem = get_system_memory();
1400 MemoryRegion *ram = g_new(MemoryRegion, 1);
1401 MemoryRegion *rma_region;
1402 void *rma = NULL;
1403 hwaddr rma_alloc_size;
1404 hwaddr node0_size = spapr_node0_size();
1405 uint32_t initrd_base = 0;
1406 long kernel_size = 0, initrd_size = 0;
1407 long load_limit, fw_size;
1408 bool kernel_le = false;
1409 char *filename;
1411 msi_supported = true;
1413 spapr = g_malloc0(sizeof(*spapr));
1414 QLIST_INIT(&spapr->phbs);
1416 cpu_ppc_hypercall = emulate_spapr_hypercall;
1418 /* Allocate RMA if necessary */
1419 rma_alloc_size = kvmppc_alloc_rma(&rma);
1421 if (rma_alloc_size == -1) {
1422 hw_error("qemu: Unable to create RMA\n");
1423 exit(1);
1426 if (rma_alloc_size && (rma_alloc_size < node0_size)) {
1427 spapr->rma_size = rma_alloc_size;
1428 } else {
1429 spapr->rma_size = node0_size;
1431 /* With KVM, we don't actually know whether KVM supports an
1432 * unbounded RMA (PR KVM) or is limited by the hash table size
1433 * (HV KVM using VRMA), so we always assume the latter
1435 * In that case, we also limit the initial allocations for RTAS
1436 * etc... to 256M since we have no way to know what the VRMA size
1437 * is going to be as it depends on the size of the hash table
1438 * isn't determined yet.
1440 if (kvm_enabled()) {
1441 spapr->vrma_adjust = 1;
1442 spapr->rma_size = MIN(spapr->rma_size, 0x10000000);
1446 if (spapr->rma_size > node0_size) {
1447 fprintf(stderr, "Error: Numa node 0 has to span the RMA (%#08"HWADDR_PRIx")\n",
1448 spapr->rma_size);
1449 exit(1);
1452 /* Setup a load limit for the ramdisk leaving room for SLOF and FDT */
1453 load_limit = MIN(spapr->rma_size, RTAS_MAX_ADDR) - FW_OVERHEAD;
1455 /* We aim for a hash table of size 1/128 the size of RAM. The
1456 * normal rule of thumb is 1/64 the size of RAM, but that's much
1457 * more than needed for the Linux guests we support. */
1458 spapr->htab_shift = 18; /* Minimum architected size */
1459 while (spapr->htab_shift <= 46) {
1460 if ((1ULL << (spapr->htab_shift + 7)) >= ram_size) {
1461 break;
1463 spapr->htab_shift++;
1466 /* Set up Interrupt Controller before we create the VCPUs */
1467 spapr->icp = xics_system_init(machine,
1468 smp_cpus * kvmppc_smt_threads() / smp_threads,
1469 XICS_IRQS);
1471 /* init CPUs */
1472 if (cpu_model == NULL) {
1473 cpu_model = kvm_enabled() ? "host" : "POWER7";
1475 for (i = 0; i < smp_cpus; i++) {
1476 cpu = cpu_ppc_init(cpu_model);
1477 if (cpu == NULL) {
1478 fprintf(stderr, "Unable to find PowerPC CPU definition\n");
1479 exit(1);
1481 env = &cpu->env;
1483 /* Set time-base frequency to 512 MHz */
1484 cpu_ppc_tb_init(env, TIMEBASE_FREQ);
1486 /* PAPR always has exception vectors in RAM not ROM. To ensure this,
1487 * MSR[IP] should never be set.
1489 env->msr_mask &= ~(1 << 6);
1491 /* Tell KVM that we're in PAPR mode */
1492 if (kvm_enabled()) {
1493 kvmppc_set_papr(cpu);
1496 if (cpu->max_compat) {
1497 if (ppc_set_compat(cpu, cpu->max_compat) < 0) {
1498 exit(1);
1502 xics_cpu_setup(spapr->icp, cpu);
1504 qemu_register_reset(spapr_cpu_reset, cpu);
1507 /* allocate RAM */
1508 spapr->ram_limit = ram_size;
1509 memory_region_allocate_system_memory(ram, NULL, "ppc_spapr.ram",
1510 spapr->ram_limit);
1511 memory_region_add_subregion(sysmem, 0, ram);
1513 if (rma_alloc_size && rma) {
1514 rma_region = g_new(MemoryRegion, 1);
1515 memory_region_init_ram_ptr(rma_region, NULL, "ppc_spapr.rma",
1516 rma_alloc_size, rma);
1517 vmstate_register_ram_global(rma_region);
1518 memory_region_add_subregion(sysmem, 0, rma_region);
1521 filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, "spapr-rtas.bin");
1522 if (!filename) {
1523 hw_error("Could not find LPAR rtas '%s'\n", "spapr-rtas.bin");
1524 exit(1);
1526 spapr->rtas_size = get_image_size(filename);
1527 spapr->rtas_blob = g_malloc(spapr->rtas_size);
1528 if (load_image_size(filename, spapr->rtas_blob, spapr->rtas_size) < 0) {
1529 hw_error("qemu: could not load LPAR rtas '%s'\n", filename);
1530 exit(1);
1532 if (spapr->rtas_size > RTAS_MAX_SIZE) {
1533 hw_error("RTAS too big ! 0x%zx bytes (max is 0x%x)\n",
1534 (size_t)spapr->rtas_size, RTAS_MAX_SIZE);
1535 exit(1);
1537 g_free(filename);
1539 /* Set up EPOW events infrastructure */
1540 spapr_events_init(spapr);
1542 /* Set up the RTC RTAS interfaces */
1543 spapr_rtc_create(spapr);
1545 /* Set up VIO bus */
1546 spapr->vio_bus = spapr_vio_bus_init();
1548 for (i = 0; i < MAX_SERIAL_PORTS; i++) {
1549 if (serial_hds[i]) {
1550 spapr_vty_create(spapr->vio_bus, serial_hds[i]);
1554 /* We always have at least the nvram device on VIO */
1555 spapr_create_nvram(spapr);
1557 /* Set up PCI */
1558 spapr_pci_rtas_init();
1560 phb = spapr_create_phb(spapr, 0);
1562 for (i = 0; i < nb_nics; i++) {
1563 NICInfo *nd = &nd_table[i];
1565 if (!nd->model) {
1566 nd->model = g_strdup("ibmveth");
1569 if (strcmp(nd->model, "ibmveth") == 0) {
1570 spapr_vlan_create(spapr->vio_bus, nd);
1571 } else {
1572 pci_nic_init_nofail(&nd_table[i], phb->bus, nd->model, NULL);
1576 for (i = 0; i <= drive_get_max_bus(IF_SCSI); i++) {
1577 spapr_vscsi_create(spapr->vio_bus);
1580 /* Graphics */
1581 if (spapr_vga_init(phb->bus)) {
1582 spapr->has_graphics = true;
1583 machine->usb |= defaults_enabled() && !machine->usb_disabled;
1586 if (machine->usb) {
1587 pci_create_simple(phb->bus, -1, "pci-ohci");
1589 if (spapr->has_graphics) {
1590 USBBus *usb_bus = usb_bus_find(-1);
1592 usb_create_simple(usb_bus, "usb-kbd");
1593 usb_create_simple(usb_bus, "usb-mouse");
1597 if (spapr->rma_size < (MIN_RMA_SLOF << 20)) {
1598 fprintf(stderr, "qemu: pSeries SLOF firmware requires >= "
1599 "%ldM guest RMA (Real Mode Area memory)\n", MIN_RMA_SLOF);
1600 exit(1);
1603 if (kernel_filename) {
1604 uint64_t lowaddr = 0;
1606 kernel_size = load_elf(kernel_filename, translate_kernel_address, NULL,
1607 NULL, &lowaddr, NULL, 1, ELF_MACHINE, 0);
1608 if (kernel_size == ELF_LOAD_WRONG_ENDIAN) {
1609 kernel_size = load_elf(kernel_filename,
1610 translate_kernel_address, NULL,
1611 NULL, &lowaddr, NULL, 0, ELF_MACHINE, 0);
1612 kernel_le = kernel_size > 0;
1614 if (kernel_size < 0) {
1615 fprintf(stderr, "qemu: error loading %s: %s\n",
1616 kernel_filename, load_elf_strerror(kernel_size));
1617 exit(1);
1620 /* load initrd */
1621 if (initrd_filename) {
1622 /* Try to locate the initrd in the gap between the kernel
1623 * and the firmware. Add a bit of space just in case
1625 initrd_base = (KERNEL_LOAD_ADDR + kernel_size + 0x1ffff) & ~0xffff;
1626 initrd_size = load_image_targphys(initrd_filename, initrd_base,
1627 load_limit - initrd_base);
1628 if (initrd_size < 0) {
1629 fprintf(stderr, "qemu: could not load initial ram disk '%s'\n",
1630 initrd_filename);
1631 exit(1);
1633 } else {
1634 initrd_base = 0;
1635 initrd_size = 0;
1639 if (bios_name == NULL) {
1640 bios_name = FW_FILE_NAME;
1642 filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
1643 if (!filename) {
1644 hw_error("Could not find LPAR rtas '%s'\n", bios_name);
1645 exit(1);
1647 fw_size = load_image_targphys(filename, 0, FW_MAX_SIZE);
1648 if (fw_size < 0) {
1649 hw_error("qemu: could not load LPAR rtas '%s'\n", filename);
1650 exit(1);
1652 g_free(filename);
1654 spapr->entry_point = 0x100;
1656 vmstate_register(NULL, 0, &vmstate_spapr, spapr);
1657 register_savevm_live(NULL, "spapr/htab", -1, 1,
1658 &savevm_htab_handlers, spapr);
1660 /* Prepare the device tree */
1661 spapr->fdt_skel = spapr_create_fdt_skel(initrd_base, initrd_size,
1662 kernel_size, kernel_le,
1663 kernel_cmdline, spapr->epow_irq);
1664 assert(spapr->fdt_skel != NULL);
1666 qemu_register_boot_set(spapr_boot_set, spapr);
1669 static int spapr_kvm_type(const char *vm_type)
1671 if (!vm_type) {
1672 return 0;
1675 if (!strcmp(vm_type, "HV")) {
1676 return 1;
1679 if (!strcmp(vm_type, "PR")) {
1680 return 2;
1683 error_report("Unknown kvm-type specified '%s'", vm_type);
1684 exit(1);
1688 * Implementation of an interface to adjust firmware path
1689 * for the bootindex property handling.
1691 static char *spapr_get_fw_dev_path(FWPathProvider *p, BusState *bus,
1692 DeviceState *dev)
1694 #define CAST(type, obj, name) \
1695 ((type *)object_dynamic_cast(OBJECT(obj), (name)))
1696 SCSIDevice *d = CAST(SCSIDevice, dev, TYPE_SCSI_DEVICE);
1697 sPAPRPHBState *phb = CAST(sPAPRPHBState, dev, TYPE_SPAPR_PCI_HOST_BRIDGE);
1699 if (d) {
1700 void *spapr = CAST(void, bus->parent, "spapr-vscsi");
1701 VirtIOSCSI *virtio = CAST(VirtIOSCSI, bus->parent, TYPE_VIRTIO_SCSI);
1702 USBDevice *usb = CAST(USBDevice, bus->parent, TYPE_USB_DEVICE);
1704 if (spapr) {
1706 * Replace "channel@0/disk@0,0" with "disk@8000000000000000":
1707 * We use SRP luns of the form 8000 | (bus << 8) | (id << 5) | lun
1708 * in the top 16 bits of the 64-bit LUN
1710 unsigned id = 0x8000 | (d->id << 8) | d->lun;
1711 return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
1712 (uint64_t)id << 48);
1713 } else if (virtio) {
1715 * We use SRP luns of the form 01000000 | (target << 8) | lun
1716 * in the top 32 bits of the 64-bit LUN
1717 * Note: the quote above is from SLOF and it is wrong,
1718 * the actual binding is:
1719 * swap 0100 or 10 << or 20 << ( target lun-id -- srplun )
1721 unsigned id = 0x1000000 | (d->id << 16) | d->lun;
1722 return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
1723 (uint64_t)id << 32);
1724 } else if (usb) {
1726 * We use SRP luns of the form 01000000 | (usb-port << 16) | lun
1727 * in the top 32 bits of the 64-bit LUN
1729 unsigned usb_port = atoi(usb->port->path);
1730 unsigned id = 0x1000000 | (usb_port << 16) | d->lun;
1731 return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
1732 (uint64_t)id << 32);
1736 if (phb) {
1737 /* Replace "pci" with "pci@800000020000000" */
1738 return g_strdup_printf("pci@%"PRIX64, phb->buid);
1741 return NULL;
1744 static char *spapr_get_kvm_type(Object *obj, Error **errp)
1746 sPAPRMachineState *sm = SPAPR_MACHINE(obj);
1748 return g_strdup(sm->kvm_type);
1751 static void spapr_set_kvm_type(Object *obj, const char *value, Error **errp)
1753 sPAPRMachineState *sm = SPAPR_MACHINE(obj);
1755 g_free(sm->kvm_type);
1756 sm->kvm_type = g_strdup(value);
1759 static void spapr_machine_initfn(Object *obj)
1761 object_property_add_str(obj, "kvm-type",
1762 spapr_get_kvm_type, spapr_set_kvm_type, NULL);
1763 object_property_set_description(obj, "kvm-type",
1764 "Specifies the KVM virtualization mode (HV, PR)",
1765 NULL);
1768 static void ppc_cpu_do_nmi_on_cpu(void *arg)
1770 CPUState *cs = arg;
1772 cpu_synchronize_state(cs);
1773 ppc_cpu_do_system_reset(cs);
1776 static void spapr_nmi(NMIState *n, int cpu_index, Error **errp)
1778 CPUState *cs;
1780 CPU_FOREACH(cs) {
1781 async_run_on_cpu(cs, ppc_cpu_do_nmi_on_cpu, cs);
1785 static void spapr_machine_class_init(ObjectClass *oc, void *data)
1787 MachineClass *mc = MACHINE_CLASS(oc);
1788 FWPathProviderClass *fwc = FW_PATH_PROVIDER_CLASS(oc);
1789 NMIClass *nc = NMI_CLASS(oc);
1791 mc->init = ppc_spapr_init;
1792 mc->reset = ppc_spapr_reset;
1793 mc->block_default_type = IF_SCSI;
1794 mc->max_cpus = MAX_CPUS;
1795 mc->no_parallel = 1;
1796 mc->default_boot_order = "";
1797 mc->kvm_type = spapr_kvm_type;
1798 mc->has_dynamic_sysbus = true;
1800 fwc->get_dev_path = spapr_get_fw_dev_path;
1801 nc->nmi_monitor_handler = spapr_nmi;
1804 static const TypeInfo spapr_machine_info = {
1805 .name = TYPE_SPAPR_MACHINE,
1806 .parent = TYPE_MACHINE,
1807 .abstract = true,
1808 .instance_size = sizeof(sPAPRMachineState),
1809 .instance_init = spapr_machine_initfn,
1810 .class_init = spapr_machine_class_init,
1811 .interfaces = (InterfaceInfo[]) {
1812 { TYPE_FW_PATH_PROVIDER },
1813 { TYPE_NMI },
1818 #define SPAPR_COMPAT_2_2 \
1820 .driver = TYPE_SPAPR_PCI_HOST_BRIDGE,\
1821 .property = "mem_win_size",\
1822 .value = "0x20000000",\
1825 #define SPAPR_COMPAT_2_1 \
1826 SPAPR_COMPAT_2_2
1828 static void spapr_compat_2_3(Object *obj)
1832 static void spapr_compat_2_2(Object *obj)
1834 spapr_compat_2_3(obj);
1837 static void spapr_compat_2_1(Object *obj)
1839 spapr_compat_2_2(obj);
1842 static void spapr_machine_2_3_instance_init(Object *obj)
1844 spapr_compat_2_3(obj);
1845 spapr_machine_initfn(obj);
1848 static void spapr_machine_2_2_instance_init(Object *obj)
1850 spapr_compat_2_2(obj);
1851 spapr_machine_initfn(obj);
1854 static void spapr_machine_2_1_instance_init(Object *obj)
1856 spapr_compat_2_1(obj);
1857 spapr_machine_initfn(obj);
1860 static void spapr_machine_2_1_class_init(ObjectClass *oc, void *data)
1862 MachineClass *mc = MACHINE_CLASS(oc);
1863 static GlobalProperty compat_props[] = {
1864 HW_COMPAT_2_1,
1865 SPAPR_COMPAT_2_1,
1866 { /* end of list */ }
1869 mc->name = "pseries-2.1";
1870 mc->desc = "pSeries Logical Partition (PAPR compliant) v2.1";
1871 mc->compat_props = compat_props;
1874 static const TypeInfo spapr_machine_2_1_info = {
1875 .name = TYPE_SPAPR_MACHINE "2.1",
1876 .parent = TYPE_SPAPR_MACHINE,
1877 .class_init = spapr_machine_2_1_class_init,
1878 .instance_init = spapr_machine_2_1_instance_init,
1881 static void spapr_machine_2_2_class_init(ObjectClass *oc, void *data)
1883 static GlobalProperty compat_props[] = {
1884 SPAPR_COMPAT_2_2,
1885 { /* end of list */ }
1887 MachineClass *mc = MACHINE_CLASS(oc);
1889 mc->name = "pseries-2.2";
1890 mc->desc = "pSeries Logical Partition (PAPR compliant) v2.2";
1891 mc->compat_props = compat_props;
1894 static const TypeInfo spapr_machine_2_2_info = {
1895 .name = TYPE_SPAPR_MACHINE "2.2",
1896 .parent = TYPE_SPAPR_MACHINE,
1897 .class_init = spapr_machine_2_2_class_init,
1898 .instance_init = spapr_machine_2_2_instance_init,
1901 static void spapr_machine_2_3_class_init(ObjectClass *oc, void *data)
1903 MachineClass *mc = MACHINE_CLASS(oc);
1905 mc->name = "pseries-2.3";
1906 mc->desc = "pSeries Logical Partition (PAPR compliant) v2.3";
1909 static const TypeInfo spapr_machine_2_3_info = {
1910 .name = TYPE_SPAPR_MACHINE "2.3",
1911 .parent = TYPE_SPAPR_MACHINE,
1912 .class_init = spapr_machine_2_3_class_init,
1913 .instance_init = spapr_machine_2_3_instance_init,
1916 static void spapr_machine_2_4_class_init(ObjectClass *oc, void *data)
1918 MachineClass *mc = MACHINE_CLASS(oc);
1920 mc->name = "pseries-2.4";
1921 mc->desc = "pSeries Logical Partition (PAPR compliant) v2.4";
1922 mc->alias = "pseries";
1923 mc->is_default = 1;
1926 static const TypeInfo spapr_machine_2_4_info = {
1927 .name = TYPE_SPAPR_MACHINE "2.4",
1928 .parent = TYPE_SPAPR_MACHINE,
1929 .class_init = spapr_machine_2_4_class_init,
1932 static void spapr_machine_register_types(void)
1934 type_register_static(&spapr_machine_info);
1935 type_register_static(&spapr_machine_2_1_info);
1936 type_register_static(&spapr_machine_2_2_info);
1937 type_register_static(&spapr_machine_2_3_info);
1938 type_register_static(&spapr_machine_2_4_info);
1941 type_init(spapr_machine_register_types)