2 * QEMU PowerPC pSeries Logical Partition (aka sPAPR) hardware System Emulator
4 * Copyright (c) 2004-2007 Fabrice Bellard
5 * Copyright (c) 2007 Jocelyn Mayer
6 * Copyright (c) 2010 David Gibson, IBM Corporation.
8 * Permission is hereby granted, free of charge, to any person obtaining a copy
9 * of this software and associated documentation files (the "Software"), to deal
10 * in the Software without restriction, including without limitation the rights
11 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
12 * copies of the Software, and to permit persons to whom the Software is
13 * furnished to do so, subject to the following conditions:
15 * The above copyright notice and this permission notice shall be included in
16 * all copies or substantial portions of the Software.
18 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
19 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
20 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL
21 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
22 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
23 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN
27 #include "sysemu/sysemu.h"
31 #include "sysemu/blockdev.h"
32 #include "sysemu/cpus.h"
33 #include "sysemu/kvm.h"
35 #include "mmu-hash64.h"
37 #include "hw/boards.h"
38 #include "hw/ppc/ppc.h"
39 #include "hw/loader.h"
41 #include "hw/ppc/spapr.h"
42 #include "hw/ppc/spapr_vio.h"
43 #include "hw/pci-host/spapr.h"
44 #include "hw/ppc/xics.h"
45 #include "hw/pci/msi.h"
47 #include "hw/pci/pci.h"
49 #include "exec/address-spaces.h"
51 #include "qemu/config-file.h"
55 /* SLOF memory layout:
57 * SLOF raw image loaded at 0, copies its romfs right below the flat
58 * device-tree, then position SLOF itself 31M below that
60 * So we set FW_OVERHEAD to 40MB which should account for all of that
63 * We load our kernel at 4M, leaving space for SLOF initial image
65 #define FDT_MAX_SIZE 0x40000
66 #define RTAS_MAX_SIZE 0x10000
67 #define FW_MAX_SIZE 0x400000
68 #define FW_FILE_NAME "slof.bin"
69 #define FW_OVERHEAD 0x2800000
70 #define KERNEL_LOAD_ADDR FW_MAX_SIZE
72 #define MIN_RMA_SLOF 128UL
74 #define TIMEBASE_FREQ 512000000ULL
77 #define XICS_IRQS 1024
79 #define PHANDLE_XICP 0x00001111
81 #define HTAB_SIZE(spapr) (1ULL << ((spapr)->htab_shift))
83 sPAPREnvironment *spapr;
85 int spapr_allocate_irq(int hint, bool lsi)
91 if (hint >= spapr->next_irq) {
92 spapr->next_irq = hint + 1;
94 /* FIXME: we should probably check for collisions somehow */
96 irq = spapr->next_irq++;
99 /* Configure irq type */
100 if (!xics_get_qirq(spapr->icp, irq)) {
104 xics_set_irq_type(spapr->icp, irq, lsi);
110 * Allocate block of consequtive IRQs, returns a number of the first.
111 * If msi==true, aligns the first IRQ number to num.
113 int spapr_allocate_irq_block(int num, bool lsi, bool msi)
119 * MSIMesage::data is used for storing VIRQ so
120 * it has to be aligned to num to support multiple
121 * MSI vectors. MSI-X is not affected by this.
122 * The hint is used for the first IRQ, the rest should
123 * be allocated continuously.
126 assert((num == 1) || (num == 2) || (num == 4) ||
127 (num == 8) || (num == 16) || (num == 32));
128 hint = (spapr->next_irq + num - 1) & ~(num - 1);
131 for (i = 0; i < num; ++i) {
134 irq = spapr_allocate_irq(hint, lsi);
144 /* If the above doesn't create a consecutive block then that's
146 assert(irq == (first + i));
152 static XICSState *try_create_xics(const char *type, int nr_servers,
157 dev = qdev_create(NULL, type);
158 qdev_prop_set_uint32(dev, "nr_servers", nr_servers);
159 qdev_prop_set_uint32(dev, "nr_irqs", nr_irqs);
160 if (qdev_init(dev) < 0) {
164 return XICS_COMMON(dev);
167 static XICSState *xics_system_init(int nr_servers, int nr_irqs)
169 XICSState *icp = NULL;
172 QemuOpts *machine_opts = qemu_get_machine_opts();
173 bool irqchip_allowed = qemu_opt_get_bool(machine_opts,
174 "kernel_irqchip", true);
175 bool irqchip_required = qemu_opt_get_bool(machine_opts,
176 "kernel_irqchip", false);
177 if (irqchip_allowed) {
178 icp = try_create_xics(TYPE_KVM_XICS, nr_servers, nr_irqs);
181 if (irqchip_required && !icp) {
182 perror("Failed to create in-kernel XICS\n");
188 icp = try_create_xics(TYPE_XICS, nr_servers, nr_irqs);
192 perror("Failed to create XICS\n");
199 static int spapr_fixup_cpu_dt(void *fdt, sPAPREnvironment *spapr)
204 int smt = kvmppc_smt_threads();
205 uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
208 DeviceClass *dc = DEVICE_GET_CLASS(cpu);
209 uint32_t associativity[] = {cpu_to_be32(0x5),
213 cpu_to_be32(cpu->numa_node),
214 cpu_to_be32(cpu->cpu_index)};
216 if ((cpu->cpu_index % smt) != 0) {
220 snprintf(cpu_model, 32, "/cpus/%s@%x", dc->fw_name,
223 offset = fdt_path_offset(fdt, cpu_model);
228 if (nb_numa_nodes > 1) {
229 ret = fdt_setprop(fdt, offset, "ibm,associativity", associativity,
230 sizeof(associativity));
236 ret = fdt_setprop(fdt, offset, "ibm,pft-size",
237 pft_size_prop, sizeof(pft_size_prop));
246 static size_t create_page_sizes_prop(CPUPPCState *env, uint32_t *prop,
249 size_t maxcells = maxsize / sizeof(uint32_t);
253 for (i = 0; i < PPC_PAGE_SIZES_MAX_SZ; i++) {
254 struct ppc_one_seg_page_size *sps = &env->sps.sps[i];
256 if (!sps->page_shift) {
259 for (count = 0; count < PPC_PAGE_SIZES_MAX_SZ; count++) {
260 if (sps->enc[count].page_shift == 0) {
264 if ((p - prop) >= (maxcells - 3 - count * 2)) {
267 *(p++) = cpu_to_be32(sps->page_shift);
268 *(p++) = cpu_to_be32(sps->slb_enc);
269 *(p++) = cpu_to_be32(count);
270 for (j = 0; j < count; j++) {
271 *(p++) = cpu_to_be32(sps->enc[j].page_shift);
272 *(p++) = cpu_to_be32(sps->enc[j].pte_enc);
276 return (p - prop) * sizeof(uint32_t);
283 fprintf(stderr, "qemu: error creating device tree: %s: %s\n", \
284 #exp, fdt_strerror(ret)); \
290 static void *spapr_create_fdt_skel(hwaddr initrd_base,
294 const char *boot_device,
295 const char *kernel_cmdline,
300 uint32_t start_prop = cpu_to_be32(initrd_base);
301 uint32_t end_prop = cpu_to_be32(initrd_base + initrd_size);
302 char hypertas_prop[] = "hcall-pft\0hcall-term\0hcall-dabr\0hcall-interrupt"
303 "\0hcall-tce\0hcall-vio\0hcall-splpar\0hcall-bulk\0hcall-set-mode";
304 char qemu_hypertas_prop[] = "hcall-memop1";
305 uint32_t refpoints[] = {cpu_to_be32(0x4), cpu_to_be32(0x4)};
306 uint32_t interrupt_server_ranges_prop[] = {0, cpu_to_be32(smp_cpus)};
307 int i, smt = kvmppc_smt_threads();
308 unsigned char vec5[] = {0x0, 0x0, 0x0, 0x0, 0x0, 0x80};
310 fdt = g_malloc0(FDT_MAX_SIZE);
311 _FDT((fdt_create(fdt, FDT_MAX_SIZE)));
314 _FDT((fdt_add_reservemap_entry(fdt, KERNEL_LOAD_ADDR, kernel_size)));
317 _FDT((fdt_add_reservemap_entry(fdt, initrd_base, initrd_size)));
319 _FDT((fdt_finish_reservemap(fdt)));
322 _FDT((fdt_begin_node(fdt, "")));
323 _FDT((fdt_property_string(fdt, "device_type", "chrp")));
324 _FDT((fdt_property_string(fdt, "model", "IBM pSeries (emulated by qemu)")));
325 _FDT((fdt_property_string(fdt, "compatible", "qemu,pseries")));
327 _FDT((fdt_property_cell(fdt, "#address-cells", 0x2)));
328 _FDT((fdt_property_cell(fdt, "#size-cells", 0x2)));
331 _FDT((fdt_begin_node(fdt, "chosen")));
333 /* Set Form1_affinity */
334 _FDT((fdt_property(fdt, "ibm,architecture-vec-5", vec5, sizeof(vec5))));
336 _FDT((fdt_property_string(fdt, "bootargs", kernel_cmdline)));
337 _FDT((fdt_property(fdt, "linux,initrd-start",
338 &start_prop, sizeof(start_prop))));
339 _FDT((fdt_property(fdt, "linux,initrd-end",
340 &end_prop, sizeof(end_prop))));
342 uint64_t kprop[2] = { cpu_to_be64(KERNEL_LOAD_ADDR),
343 cpu_to_be64(kernel_size) };
345 _FDT((fdt_property(fdt, "qemu,boot-kernel", &kprop, sizeof(kprop))));
347 _FDT((fdt_property(fdt, "qemu,boot-kernel-le", NULL, 0)));
351 _FDT((fdt_property_string(fdt, "qemu,boot-device", boot_device)));
353 _FDT((fdt_property_cell(fdt, "qemu,graphic-width", graphic_width)));
354 _FDT((fdt_property_cell(fdt, "qemu,graphic-height", graphic_height)));
355 _FDT((fdt_property_cell(fdt, "qemu,graphic-depth", graphic_depth)));
357 _FDT((fdt_end_node(fdt)));
360 _FDT((fdt_begin_node(fdt, "cpus")));
362 _FDT((fdt_property_cell(fdt, "#address-cells", 0x1)));
363 _FDT((fdt_property_cell(fdt, "#size-cells", 0x0)));
366 PowerPCCPU *cpu = POWERPC_CPU(cs);
367 CPUPPCState *env = &cpu->env;
368 DeviceClass *dc = DEVICE_GET_CLASS(cs);
369 PowerPCCPUClass *pcc = POWERPC_CPU_GET_CLASS(cs);
370 int index = cs->cpu_index;
371 uint32_t servers_prop[smp_threads];
372 uint32_t gservers_prop[smp_threads * 2];
374 uint32_t segs[] = {cpu_to_be32(28), cpu_to_be32(40),
375 0xffffffff, 0xffffffff};
376 uint32_t tbfreq = kvm_enabled() ? kvmppc_get_tbfreq() : TIMEBASE_FREQ;
377 uint32_t cpufreq = kvm_enabled() ? kvmppc_get_clockfreq() : 1000000000;
378 uint32_t page_sizes_prop[64];
379 size_t page_sizes_prop_size;
381 if ((index % smt) != 0) {
385 nodename = g_strdup_printf("%s@%x", dc->fw_name, index);
387 _FDT((fdt_begin_node(fdt, nodename)));
391 _FDT((fdt_property_cell(fdt, "reg", index)));
392 _FDT((fdt_property_string(fdt, "device_type", "cpu")));
394 _FDT((fdt_property_cell(fdt, "cpu-version", env->spr[SPR_PVR])));
395 _FDT((fdt_property_cell(fdt, "d-cache-block-size",
396 env->dcache_line_size)));
397 _FDT((fdt_property_cell(fdt, "d-cache-line-size",
398 env->dcache_line_size)));
399 _FDT((fdt_property_cell(fdt, "i-cache-block-size",
400 env->icache_line_size)));
401 _FDT((fdt_property_cell(fdt, "i-cache-line-size",
402 env->icache_line_size)));
404 if (pcc->l1_dcache_size) {
405 _FDT((fdt_property_cell(fdt, "d-cache-size", pcc->l1_dcache_size)));
407 fprintf(stderr, "Warning: Unknown L1 dcache size for cpu\n");
409 if (pcc->l1_icache_size) {
410 _FDT((fdt_property_cell(fdt, "i-cache-size", pcc->l1_icache_size)));
412 fprintf(stderr, "Warning: Unknown L1 icache size for cpu\n");
415 _FDT((fdt_property_cell(fdt, "timebase-frequency", tbfreq)));
416 _FDT((fdt_property_cell(fdt, "clock-frequency", cpufreq)));
417 _FDT((fdt_property_cell(fdt, "ibm,slb-size", env->slb_nr)));
418 _FDT((fdt_property_string(fdt, "status", "okay")));
419 _FDT((fdt_property(fdt, "64-bit", NULL, 0)));
421 /* Build interrupt servers and gservers properties */
422 for (i = 0; i < smp_threads; i++) {
423 servers_prop[i] = cpu_to_be32(index + i);
424 /* Hack, direct the group queues back to cpu 0 */
425 gservers_prop[i*2] = cpu_to_be32(index + i);
426 gservers_prop[i*2 + 1] = 0;
428 _FDT((fdt_property(fdt, "ibm,ppc-interrupt-server#s",
429 servers_prop, sizeof(servers_prop))));
430 _FDT((fdt_property(fdt, "ibm,ppc-interrupt-gserver#s",
431 gservers_prop, sizeof(gservers_prop))));
433 if (env->spr_cb[SPR_PURR].oea_read) {
434 _FDT((fdt_property(fdt, "ibm,purr", NULL, 0)));
437 if (env->mmu_model & POWERPC_MMU_1TSEG) {
438 _FDT((fdt_property(fdt, "ibm,processor-segment-sizes",
439 segs, sizeof(segs))));
442 /* Advertise VMX/VSX (vector extensions) if available
443 * 0 / no property == no vector extensions
444 * 1 == VMX / Altivec available
445 * 2 == VSX available */
446 if (env->insns_flags & PPC_ALTIVEC) {
447 uint32_t vmx = (env->insns_flags2 & PPC2_VSX) ? 2 : 1;
449 _FDT((fdt_property_cell(fdt, "ibm,vmx", vmx)));
452 /* Advertise DFP (Decimal Floating Point) if available
453 * 0 / no property == no DFP
454 * 1 == DFP available */
455 if (env->insns_flags2 & PPC2_DFP) {
456 _FDT((fdt_property_cell(fdt, "ibm,dfp", 1)));
459 page_sizes_prop_size = create_page_sizes_prop(env, page_sizes_prop,
460 sizeof(page_sizes_prop));
461 if (page_sizes_prop_size) {
462 _FDT((fdt_property(fdt, "ibm,segment-page-sizes",
463 page_sizes_prop, page_sizes_prop_size)));
466 _FDT((fdt_end_node(fdt)));
469 _FDT((fdt_end_node(fdt)));
472 _FDT((fdt_begin_node(fdt, "rtas")));
474 _FDT((fdt_property(fdt, "ibm,hypertas-functions", hypertas_prop,
475 sizeof(hypertas_prop))));
476 _FDT((fdt_property(fdt, "qemu,hypertas-functions", qemu_hypertas_prop,
477 sizeof(qemu_hypertas_prop))));
479 _FDT((fdt_property(fdt, "ibm,associativity-reference-points",
480 refpoints, sizeof(refpoints))));
482 _FDT((fdt_property_cell(fdt, "rtas-error-log-max", RTAS_ERROR_LOG_MAX)));
484 _FDT((fdt_end_node(fdt)));
486 /* interrupt controller */
487 _FDT((fdt_begin_node(fdt, "interrupt-controller")));
489 _FDT((fdt_property_string(fdt, "device_type",
490 "PowerPC-External-Interrupt-Presentation")));
491 _FDT((fdt_property_string(fdt, "compatible", "IBM,ppc-xicp")));
492 _FDT((fdt_property(fdt, "interrupt-controller", NULL, 0)));
493 _FDT((fdt_property(fdt, "ibm,interrupt-server-ranges",
494 interrupt_server_ranges_prop,
495 sizeof(interrupt_server_ranges_prop))));
496 _FDT((fdt_property_cell(fdt, "#interrupt-cells", 2)));
497 _FDT((fdt_property_cell(fdt, "linux,phandle", PHANDLE_XICP)));
498 _FDT((fdt_property_cell(fdt, "phandle", PHANDLE_XICP)));
500 _FDT((fdt_end_node(fdt)));
503 _FDT((fdt_begin_node(fdt, "vdevice")));
505 _FDT((fdt_property_string(fdt, "device_type", "vdevice")));
506 _FDT((fdt_property_string(fdt, "compatible", "IBM,vdevice")));
507 _FDT((fdt_property_cell(fdt, "#address-cells", 0x1)));
508 _FDT((fdt_property_cell(fdt, "#size-cells", 0x0)));
509 _FDT((fdt_property_cell(fdt, "#interrupt-cells", 0x2)));
510 _FDT((fdt_property(fdt, "interrupt-controller", NULL, 0)));
512 _FDT((fdt_end_node(fdt)));
515 spapr_events_fdt_skel(fdt, epow_irq);
517 _FDT((fdt_end_node(fdt))); /* close root node */
518 _FDT((fdt_finish(fdt)));
523 static int spapr_populate_memory(sPAPREnvironment *spapr, void *fdt)
525 uint32_t associativity[] = {cpu_to_be32(0x4), cpu_to_be32(0x0),
526 cpu_to_be32(0x0), cpu_to_be32(0x0),
529 hwaddr node0_size, mem_start, node_size;
530 uint64_t mem_reg_property[2];
534 if (nb_numa_nodes > 1 && node_mem[0] < ram_size) {
535 node0_size = node_mem[0];
537 node0_size = ram_size;
541 mem_reg_property[0] = 0;
542 mem_reg_property[1] = cpu_to_be64(spapr->rma_size);
543 off = fdt_add_subnode(fdt, 0, "memory@0");
545 _FDT((fdt_setprop_string(fdt, off, "device_type", "memory")));
546 _FDT((fdt_setprop(fdt, off, "reg", mem_reg_property,
547 sizeof(mem_reg_property))));
548 _FDT((fdt_setprop(fdt, off, "ibm,associativity", associativity,
549 sizeof(associativity))));
552 if (node0_size > spapr->rma_size) {
553 mem_reg_property[0] = cpu_to_be64(spapr->rma_size);
554 mem_reg_property[1] = cpu_to_be64(node0_size - spapr->rma_size);
556 sprintf(mem_name, "memory@" TARGET_FMT_lx, spapr->rma_size);
557 off = fdt_add_subnode(fdt, 0, mem_name);
559 _FDT((fdt_setprop_string(fdt, off, "device_type", "memory")));
560 _FDT((fdt_setprop(fdt, off, "reg", mem_reg_property,
561 sizeof(mem_reg_property))));
562 _FDT((fdt_setprop(fdt, off, "ibm,associativity", associativity,
563 sizeof(associativity))));
566 /* RAM: Node 1 and beyond */
567 mem_start = node0_size;
568 for (i = 1; i < nb_numa_nodes; i++) {
569 mem_reg_property[0] = cpu_to_be64(mem_start);
570 if (mem_start >= ram_size) {
573 node_size = node_mem[i];
574 if (node_size > ram_size - mem_start) {
575 node_size = ram_size - mem_start;
578 mem_reg_property[1] = cpu_to_be64(node_size);
579 associativity[3] = associativity[4] = cpu_to_be32(i);
580 sprintf(mem_name, "memory@" TARGET_FMT_lx, mem_start);
581 off = fdt_add_subnode(fdt, 0, mem_name);
583 _FDT((fdt_setprop_string(fdt, off, "device_type", "memory")));
584 _FDT((fdt_setprop(fdt, off, "reg", mem_reg_property,
585 sizeof(mem_reg_property))));
586 _FDT((fdt_setprop(fdt, off, "ibm,associativity", associativity,
587 sizeof(associativity))));
588 mem_start += node_size;
594 static void spapr_finalize_fdt(sPAPREnvironment *spapr,
603 fdt = g_malloc(FDT_MAX_SIZE);
605 /* open out the base tree into a temp buffer for the final tweaks */
606 _FDT((fdt_open_into(spapr->fdt_skel, fdt, FDT_MAX_SIZE)));
608 ret = spapr_populate_memory(spapr, fdt);
610 fprintf(stderr, "couldn't setup memory nodes in fdt\n");
614 ret = spapr_populate_vdevice(spapr->vio_bus, fdt);
616 fprintf(stderr, "couldn't setup vio devices in fdt\n");
620 QLIST_FOREACH(phb, &spapr->phbs, list) {
621 ret = spapr_populate_pci_dt(phb, PHANDLE_XICP, fdt);
625 fprintf(stderr, "couldn't setup PCI devices in fdt\n");
630 ret = spapr_rtas_device_tree_setup(fdt, rtas_addr, rtas_size);
632 fprintf(stderr, "Couldn't set up RTAS device tree properties\n");
635 /* Advertise NUMA via ibm,associativity */
636 ret = spapr_fixup_cpu_dt(fdt, spapr);
638 fprintf(stderr, "Couldn't finalize CPU device tree properties\n");
641 if (!spapr->has_graphics) {
642 spapr_populate_chosen_stdout(fdt, spapr->vio_bus);
645 _FDT((fdt_pack(fdt)));
647 if (fdt_totalsize(fdt) > FDT_MAX_SIZE) {
648 hw_error("FDT too big ! 0x%x bytes (max is 0x%x)\n",
649 fdt_totalsize(fdt), FDT_MAX_SIZE);
653 cpu_physical_memory_write(fdt_addr, fdt, fdt_totalsize(fdt));
658 static uint64_t translate_kernel_address(void *opaque, uint64_t addr)
660 return (addr & 0x0fffffff) + KERNEL_LOAD_ADDR;
663 static void emulate_spapr_hypercall(PowerPCCPU *cpu)
665 CPUPPCState *env = &cpu->env;
668 hcall_dprintf("Hypercall made with MSR[PR]=1\n");
669 env->gpr[3] = H_PRIVILEGE;
671 env->gpr[3] = spapr_hypercall(cpu, env->gpr[3], &env->gpr[4]);
675 static void spapr_reset_htab(sPAPREnvironment *spapr)
679 /* allocate hash page table. For now we always make this 16mb,
680 * later we should probably make it scale to the size of guest
683 shift = kvmppc_reset_htab(spapr->htab_shift);
686 /* Kernel handles htab, we don't need to allocate one */
687 spapr->htab_shift = shift;
690 /* Allocate an htab if we don't yet have one */
691 spapr->htab = qemu_memalign(HTAB_SIZE(spapr), HTAB_SIZE(spapr));
695 memset(spapr->htab, 0, HTAB_SIZE(spapr));
698 /* Update the RMA size if necessary */
699 if (spapr->vrma_adjust) {
700 hwaddr node0_size = (nb_numa_nodes > 1) ? node_mem[0] : ram_size;
701 spapr->rma_size = kvmppc_rma_size(node0_size, spapr->htab_shift);
705 static void ppc_spapr_reset(void)
707 PowerPCCPU *first_ppc_cpu;
709 /* Reset the hash table & recalc the RMA */
710 spapr_reset_htab(spapr);
712 qemu_devices_reset();
715 spapr_finalize_fdt(spapr, spapr->fdt_addr, spapr->rtas_addr,
718 /* Set up the entry state */
719 first_ppc_cpu = POWERPC_CPU(first_cpu);
720 first_ppc_cpu->env.gpr[3] = spapr->fdt_addr;
721 first_ppc_cpu->env.gpr[5] = 0;
722 first_cpu->halted = 0;
723 first_ppc_cpu->env.nip = spapr->entry_point;
727 static void spapr_cpu_reset(void *opaque)
729 PowerPCCPU *cpu = opaque;
730 CPUState *cs = CPU(cpu);
731 CPUPPCState *env = &cpu->env;
735 /* All CPUs start halted. CPU0 is unhalted from the machine level
736 * reset code and the rest are explicitly started up by the guest
737 * using an RTAS call */
740 env->spr[SPR_HIOR] = 0;
742 env->external_htab = (uint8_t *)spapr->htab;
744 env->htab_mask = HTAB_SIZE(spapr) - 1;
745 env->spr[SPR_SDR1] = (target_ulong)(uintptr_t)spapr->htab |
746 (spapr->htab_shift - 18);
749 static void spapr_create_nvram(sPAPREnvironment *spapr)
751 DeviceState *dev = qdev_create(&spapr->vio_bus->bus, "spapr-nvram");
752 DriveInfo *dinfo = drive_get(IF_PFLASH, 0, 0);
755 qdev_prop_set_drive_nofail(dev, "drive", dinfo->bdrv);
758 qdev_init_nofail(dev);
760 spapr->nvram = (struct sPAPRNVRAM *)dev;
763 /* Returns whether we want to use VGA or not */
764 static int spapr_vga_init(PCIBus *pci_bus)
766 switch (vga_interface_type) {
769 return pci_vga_init(pci_bus) != NULL;
771 fprintf(stderr, "This vga model is not supported,"
772 "currently it only supports -vga std\n");
778 static const VMStateDescription vmstate_spapr = {
781 .minimum_version_id = 1,
782 .minimum_version_id_old = 1,
783 .fields = (VMStateField []) {
784 VMSTATE_UINT32(next_irq, sPAPREnvironment),
787 VMSTATE_UINT64(rtc_offset, sPAPREnvironment),
789 VMSTATE_END_OF_LIST()
793 #define HPTE(_table, _i) (void *)(((uint64_t *)(_table)) + ((_i) * 2))
794 #define HPTE_VALID(_hpte) (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_VALID)
795 #define HPTE_DIRTY(_hpte) (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_HPTE_DIRTY)
796 #define CLEAN_HPTE(_hpte) ((*(uint64_t *)(_hpte)) &= tswap64(~HPTE64_V_HPTE_DIRTY))
798 static int htab_save_setup(QEMUFile *f, void *opaque)
800 sPAPREnvironment *spapr = opaque;
802 /* "Iteration" header */
803 qemu_put_be32(f, spapr->htab_shift);
806 spapr->htab_save_index = 0;
807 spapr->htab_first_pass = true;
809 assert(kvm_enabled());
811 spapr->htab_fd = kvmppc_get_htab_fd(false);
812 if (spapr->htab_fd < 0) {
813 fprintf(stderr, "Unable to open fd for reading hash table from KVM: %s\n",
823 static void htab_save_first_pass(QEMUFile *f, sPAPREnvironment *spapr,
826 int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
827 int index = spapr->htab_save_index;
828 int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
830 assert(spapr->htab_first_pass);
835 /* Consume invalid HPTEs */
836 while ((index < htabslots)
837 && !HPTE_VALID(HPTE(spapr->htab, index))) {
839 CLEAN_HPTE(HPTE(spapr->htab, index));
842 /* Consume valid HPTEs */
844 while ((index < htabslots)
845 && HPTE_VALID(HPTE(spapr->htab, index))) {
847 CLEAN_HPTE(HPTE(spapr->htab, index));
850 if (index > chunkstart) {
851 int n_valid = index - chunkstart;
853 qemu_put_be32(f, chunkstart);
854 qemu_put_be16(f, n_valid);
856 qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
857 HASH_PTE_SIZE_64 * n_valid);
859 if ((qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
863 } while ((index < htabslots) && !qemu_file_rate_limit(f));
865 if (index >= htabslots) {
866 assert(index == htabslots);
868 spapr->htab_first_pass = false;
870 spapr->htab_save_index = index;
873 static int htab_save_later_pass(QEMUFile *f, sPAPREnvironment *spapr,
876 bool final = max_ns < 0;
877 int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
878 int examined = 0, sent = 0;
879 int index = spapr->htab_save_index;
880 int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
882 assert(!spapr->htab_first_pass);
885 int chunkstart, invalidstart;
887 /* Consume non-dirty HPTEs */
888 while ((index < htabslots)
889 && !HPTE_DIRTY(HPTE(spapr->htab, index))) {
895 /* Consume valid dirty HPTEs */
896 while ((index < htabslots)
897 && HPTE_DIRTY(HPTE(spapr->htab, index))
898 && HPTE_VALID(HPTE(spapr->htab, index))) {
899 CLEAN_HPTE(HPTE(spapr->htab, index));
904 invalidstart = index;
905 /* Consume invalid dirty HPTEs */
906 while ((index < htabslots)
907 && HPTE_DIRTY(HPTE(spapr->htab, index))
908 && !HPTE_VALID(HPTE(spapr->htab, index))) {
909 CLEAN_HPTE(HPTE(spapr->htab, index));
914 if (index > chunkstart) {
915 int n_valid = invalidstart - chunkstart;
916 int n_invalid = index - invalidstart;
918 qemu_put_be32(f, chunkstart);
919 qemu_put_be16(f, n_valid);
920 qemu_put_be16(f, n_invalid);
921 qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
922 HASH_PTE_SIZE_64 * n_valid);
923 sent += index - chunkstart;
925 if (!final && (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
930 if (examined >= htabslots) {
934 if (index >= htabslots) {
935 assert(index == htabslots);
938 } while ((examined < htabslots) && (!qemu_file_rate_limit(f) || final));
940 if (index >= htabslots) {
941 assert(index == htabslots);
945 spapr->htab_save_index = index;
947 return (examined >= htabslots) && (sent == 0) ? 1 : 0;
950 #define MAX_ITERATION_NS 5000000 /* 5 ms */
951 #define MAX_KVM_BUF_SIZE 2048
953 static int htab_save_iterate(QEMUFile *f, void *opaque)
955 sPAPREnvironment *spapr = opaque;
958 /* Iteration header */
962 assert(kvm_enabled());
964 rc = kvmppc_save_htab(f, spapr->htab_fd,
965 MAX_KVM_BUF_SIZE, MAX_ITERATION_NS);
969 } else if (spapr->htab_first_pass) {
970 htab_save_first_pass(f, spapr, MAX_ITERATION_NS);
972 rc = htab_save_later_pass(f, spapr, MAX_ITERATION_NS);
983 static int htab_save_complete(QEMUFile *f, void *opaque)
985 sPAPREnvironment *spapr = opaque;
987 /* Iteration header */
993 assert(kvm_enabled());
995 rc = kvmppc_save_htab(f, spapr->htab_fd, MAX_KVM_BUF_SIZE, -1);
999 close(spapr->htab_fd);
1000 spapr->htab_fd = -1;
1002 htab_save_later_pass(f, spapr, -1);
1006 qemu_put_be32(f, 0);
1007 qemu_put_be16(f, 0);
1008 qemu_put_be16(f, 0);
1013 static int htab_load(QEMUFile *f, void *opaque, int version_id)
1015 sPAPREnvironment *spapr = opaque;
1016 uint32_t section_hdr;
1019 if (version_id < 1 || version_id > 1) {
1020 fprintf(stderr, "htab_load() bad version\n");
1024 section_hdr = qemu_get_be32(f);
1027 /* First section, just the hash shift */
1028 if (spapr->htab_shift != section_hdr) {
1035 assert(kvm_enabled());
1037 fd = kvmppc_get_htab_fd(true);
1039 fprintf(stderr, "Unable to open fd to restore KVM hash table: %s\n",
1046 uint16_t n_valid, n_invalid;
1048 index = qemu_get_be32(f);
1049 n_valid = qemu_get_be16(f);
1050 n_invalid = qemu_get_be16(f);
1052 if ((index == 0) && (n_valid == 0) && (n_invalid == 0)) {
1057 if ((index + n_valid + n_invalid) >
1058 (HTAB_SIZE(spapr) / HASH_PTE_SIZE_64)) {
1059 /* Bad index in stream */
1060 fprintf(stderr, "htab_load() bad index %d (%hd+%hd entries) "
1061 "in htab stream (htab_shift=%d)\n", index, n_valid, n_invalid,
1068 qemu_get_buffer(f, HPTE(spapr->htab, index),
1069 HASH_PTE_SIZE_64 * n_valid);
1072 memset(HPTE(spapr->htab, index + n_valid), 0,
1073 HASH_PTE_SIZE_64 * n_invalid);
1080 rc = kvmppc_load_htab_chunk(f, fd, index, n_valid, n_invalid);
1095 static SaveVMHandlers savevm_htab_handlers = {
1096 .save_live_setup = htab_save_setup,
1097 .save_live_iterate = htab_save_iterate,
1098 .save_live_complete = htab_save_complete,
1099 .load_state = htab_load,
1102 /* pSeries LPAR / sPAPR hardware init */
1103 static void ppc_spapr_init(QEMUMachineInitArgs *args)
1105 ram_addr_t ram_size = args->ram_size;
1106 const char *cpu_model = args->cpu_model;
1107 const char *kernel_filename = args->kernel_filename;
1108 const char *kernel_cmdline = args->kernel_cmdline;
1109 const char *initrd_filename = args->initrd_filename;
1110 const char *boot_device = args->boot_order;
1115 MemoryRegion *sysmem = get_system_memory();
1116 MemoryRegion *ram = g_new(MemoryRegion, 1);
1117 hwaddr rma_alloc_size;
1118 hwaddr node0_size = (nb_numa_nodes > 1) ? node_mem[0] : ram_size;
1119 uint32_t initrd_base = 0;
1120 long kernel_size = 0, initrd_size = 0;
1121 long load_limit, rtas_limit, fw_size;
1122 bool kernel_le = false;
1125 msi_supported = true;
1127 spapr = g_malloc0(sizeof(*spapr));
1128 QLIST_INIT(&spapr->phbs);
1130 cpu_ppc_hypercall = emulate_spapr_hypercall;
1132 /* Allocate RMA if necessary */
1133 rma_alloc_size = kvmppc_alloc_rma("ppc_spapr.rma", sysmem);
1135 if (rma_alloc_size == -1) {
1136 hw_error("qemu: Unable to create RMA\n");
1140 if (rma_alloc_size && (rma_alloc_size < node0_size)) {
1141 spapr->rma_size = rma_alloc_size;
1143 spapr->rma_size = node0_size;
1145 /* With KVM, we don't actually know whether KVM supports an
1146 * unbounded RMA (PR KVM) or is limited by the hash table size
1147 * (HV KVM using VRMA), so we always assume the latter
1149 * In that case, we also limit the initial allocations for RTAS
1150 * etc... to 256M since we have no way to know what the VRMA size
1151 * is going to be as it depends on the size of the hash table
1152 * isn't determined yet.
1154 if (kvm_enabled()) {
1155 spapr->vrma_adjust = 1;
1156 spapr->rma_size = MIN(spapr->rma_size, 0x10000000);
1160 if (spapr->rma_size > node0_size) {
1161 fprintf(stderr, "Error: Numa node 0 has to span the RMA (%#08"HWADDR_PRIx")\n",
1166 /* We place the device tree and RTAS just below either the top of the RMA,
1167 * or just below 2GB, whichever is lowere, so that it can be
1168 * processed with 32-bit real mode code if necessary */
1169 rtas_limit = MIN(spapr->rma_size, 0x80000000);
1170 spapr->rtas_addr = rtas_limit - RTAS_MAX_SIZE;
1171 spapr->fdt_addr = spapr->rtas_addr - FDT_MAX_SIZE;
1172 load_limit = spapr->fdt_addr - FW_OVERHEAD;
1174 /* We aim for a hash table of size 1/128 the size of RAM. The
1175 * normal rule of thumb is 1/64 the size of RAM, but that's much
1176 * more than needed for the Linux guests we support. */
1177 spapr->htab_shift = 18; /* Minimum architected size */
1178 while (spapr->htab_shift <= 46) {
1179 if ((1ULL << (spapr->htab_shift + 7)) >= ram_size) {
1182 spapr->htab_shift++;
1185 /* Set up Interrupt Controller before we create the VCPUs */
1186 spapr->icp = xics_system_init(smp_cpus * kvmppc_smt_threads() / smp_threads,
1188 spapr->next_irq = XICS_IRQ_BASE;
1191 if (cpu_model == NULL) {
1192 cpu_model = kvm_enabled() ? "host" : "POWER7";
1194 for (i = 0; i < smp_cpus; i++) {
1195 cpu = cpu_ppc_init(cpu_model);
1197 fprintf(stderr, "Unable to find PowerPC CPU definition\n");
1202 /* Set time-base frequency to 512 MHz */
1203 cpu_ppc_tb_init(env, TIMEBASE_FREQ);
1205 /* PAPR always has exception vectors in RAM not ROM. To ensure this,
1206 * MSR[IP] should never be set.
1208 env->msr_mask &= ~(1 << 6);
1210 /* Tell KVM that we're in PAPR mode */
1211 if (kvm_enabled()) {
1212 kvmppc_set_papr(cpu);
1215 xics_cpu_setup(spapr->icp, cpu);
1217 qemu_register_reset(spapr_cpu_reset, cpu);
1221 spapr->ram_limit = ram_size;
1222 if (spapr->ram_limit > rma_alloc_size) {
1223 ram_addr_t nonrma_base = rma_alloc_size;
1224 ram_addr_t nonrma_size = spapr->ram_limit - rma_alloc_size;
1226 memory_region_init_ram(ram, NULL, "ppc_spapr.ram", nonrma_size);
1227 vmstate_register_ram_global(ram);
1228 memory_region_add_subregion(sysmem, nonrma_base, ram);
1231 filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, "spapr-rtas.bin");
1232 spapr->rtas_size = load_image_targphys(filename, spapr->rtas_addr,
1233 rtas_limit - spapr->rtas_addr);
1234 if (spapr->rtas_size < 0) {
1235 hw_error("qemu: could not load LPAR rtas '%s'\n", filename);
1238 if (spapr->rtas_size > RTAS_MAX_SIZE) {
1239 hw_error("RTAS too big ! 0x%lx bytes (max is 0x%x)\n",
1240 spapr->rtas_size, RTAS_MAX_SIZE);
1245 /* Set up EPOW events infrastructure */
1246 spapr_events_init(spapr);
1248 /* Set up VIO bus */
1249 spapr->vio_bus = spapr_vio_bus_init();
1251 for (i = 0; i < MAX_SERIAL_PORTS; i++) {
1252 if (serial_hds[i]) {
1253 spapr_vty_create(spapr->vio_bus, serial_hds[i]);
1257 /* We always have at least the nvram device on VIO */
1258 spapr_create_nvram(spapr);
1261 spapr_pci_msi_init(spapr, SPAPR_PCI_MSI_WINDOW);
1262 spapr_pci_rtas_init();
1264 phb = spapr_create_phb(spapr, 0);
1266 for (i = 0; i < nb_nics; i++) {
1267 NICInfo *nd = &nd_table[i];
1270 nd->model = g_strdup("ibmveth");
1273 if (strcmp(nd->model, "ibmveth") == 0) {
1274 spapr_vlan_create(spapr->vio_bus, nd);
1276 pci_nic_init_nofail(&nd_table[i], phb->bus, nd->model, NULL);
1280 for (i = 0; i <= drive_get_max_bus(IF_SCSI); i++) {
1281 spapr_vscsi_create(spapr->vio_bus);
1285 if (spapr_vga_init(phb->bus)) {
1286 spapr->has_graphics = true;
1289 if (usb_enabled(spapr->has_graphics)) {
1290 pci_create_simple(phb->bus, -1, "pci-ohci");
1291 if (spapr->has_graphics) {
1292 usbdevice_create("keyboard");
1293 usbdevice_create("mouse");
1297 if (spapr->rma_size < (MIN_RMA_SLOF << 20)) {
1298 fprintf(stderr, "qemu: pSeries SLOF firmware requires >= "
1299 "%ldM guest RMA (Real Mode Area memory)\n", MIN_RMA_SLOF);
1303 if (kernel_filename) {
1304 uint64_t lowaddr = 0;
1306 kernel_size = load_elf(kernel_filename, translate_kernel_address, NULL,
1307 NULL, &lowaddr, NULL, 1, ELF_MACHINE, 0);
1308 if (kernel_size < 0) {
1309 kernel_size = load_elf(kernel_filename,
1310 translate_kernel_address, NULL,
1311 NULL, &lowaddr, NULL, 0, ELF_MACHINE, 0);
1312 kernel_le = kernel_size > 0;
1314 if (kernel_size < 0) {
1315 kernel_size = load_image_targphys(kernel_filename,
1317 load_limit - KERNEL_LOAD_ADDR);
1319 if (kernel_size < 0) {
1320 fprintf(stderr, "qemu: could not load kernel '%s'\n",
1326 if (initrd_filename) {
1327 /* Try to locate the initrd in the gap between the kernel
1328 * and the firmware. Add a bit of space just in case
1330 initrd_base = (KERNEL_LOAD_ADDR + kernel_size + 0x1ffff) & ~0xffff;
1331 initrd_size = load_image_targphys(initrd_filename, initrd_base,
1332 load_limit - initrd_base);
1333 if (initrd_size < 0) {
1334 fprintf(stderr, "qemu: could not load initial ram disk '%s'\n",
1344 if (bios_name == NULL) {
1345 bios_name = FW_FILE_NAME;
1347 filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
1348 fw_size = load_image_targphys(filename, 0, FW_MAX_SIZE);
1350 hw_error("qemu: could not load LPAR rtas '%s'\n", filename);
1355 spapr->entry_point = 0x100;
1357 vmstate_register(NULL, 0, &vmstate_spapr, spapr);
1358 register_savevm_live(NULL, "spapr/htab", -1, 1,
1359 &savevm_htab_handlers, spapr);
1361 /* Prepare the device tree */
1362 spapr->fdt_skel = spapr_create_fdt_skel(initrd_base, initrd_size,
1363 kernel_size, kernel_le,
1364 boot_device, kernel_cmdline,
1366 assert(spapr->fdt_skel != NULL);
1369 static QEMUMachine spapr_machine = {
1371 .desc = "pSeries Logical Partition (PAPR compliant)",
1373 .init = ppc_spapr_init,
1374 .reset = ppc_spapr_reset,
1375 .block_default_type = IF_SCSI,
1376 .max_cpus = MAX_CPUS,
1378 .default_boot_order = NULL,
1381 static void spapr_machine_init(void)
1383 qemu_register_machine(&spapr_machine);
1386 machine_init(spapr_machine_init);