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"
29 #include "hw/fw-path-provider.h"
32 #include "sysemu/blockdev.h"
33 #include "sysemu/cpus.h"
34 #include "sysemu/kvm.h"
36 #include "mmu-hash64.h"
38 #include "hw/boards.h"
39 #include "hw/ppc/ppc.h"
40 #include "hw/loader.h"
42 #include "hw/ppc/spapr.h"
43 #include "hw/ppc/spapr_vio.h"
44 #include "hw/pci-host/spapr.h"
45 #include "hw/ppc/xics.h"
46 #include "hw/pci/msi.h"
48 #include "hw/pci/pci.h"
49 #include "hw/scsi/scsi.h"
50 #include "hw/virtio/virtio-scsi.h"
52 #include "exec/address-spaces.h"
54 #include "qemu/config-file.h"
55 #include "qemu/error-report.h"
59 /* SLOF memory layout:
61 * SLOF raw image loaded at 0, copies its romfs right below the flat
62 * device-tree, then position SLOF itself 31M below that
64 * So we set FW_OVERHEAD to 40MB which should account for all of that
67 * We load our kernel at 4M, leaving space for SLOF initial image
69 #define FDT_MAX_SIZE 0x40000
70 #define RTAS_MAX_SIZE 0x10000
71 #define FW_MAX_SIZE 0x400000
72 #define FW_FILE_NAME "slof.bin"
73 #define FW_OVERHEAD 0x2800000
74 #define KERNEL_LOAD_ADDR FW_MAX_SIZE
76 #define MIN_RMA_SLOF 128UL
78 #define TIMEBASE_FREQ 512000000ULL
81 #define XICS_IRQS 1024
83 #define PHANDLE_XICP 0x00001111
85 #define HTAB_SIZE(spapr) (1ULL << ((spapr)->htab_shift))
87 #define TYPE_SPAPR_MACHINE "spapr-machine"
89 sPAPREnvironment *spapr;
91 int spapr_allocate_irq(int hint, bool lsi)
97 if (hint >= spapr->next_irq) {
98 spapr->next_irq = hint + 1;
100 /* FIXME: we should probably check for collisions somehow */
102 irq = spapr->next_irq++;
105 /* Configure irq type */
106 if (!xics_get_qirq(spapr->icp, irq)) {
110 xics_set_irq_type(spapr->icp, irq, lsi);
116 * Allocate block of consequtive IRQs, returns a number of the first.
117 * If msi==true, aligns the first IRQ number to num.
119 int spapr_allocate_irq_block(int num, bool lsi, bool msi)
125 * MSIMesage::data is used for storing VIRQ so
126 * it has to be aligned to num to support multiple
127 * MSI vectors. MSI-X is not affected by this.
128 * The hint is used for the first IRQ, the rest should
129 * be allocated continuously.
132 assert((num == 1) || (num == 2) || (num == 4) ||
133 (num == 8) || (num == 16) || (num == 32));
134 hint = (spapr->next_irq + num - 1) & ~(num - 1);
137 for (i = 0; i < num; ++i) {
140 irq = spapr_allocate_irq(hint, lsi);
150 /* If the above doesn't create a consecutive block then that's
152 assert(irq == (first + i));
158 static XICSState *try_create_xics(const char *type, int nr_servers,
163 dev = qdev_create(NULL, type);
164 qdev_prop_set_uint32(dev, "nr_servers", nr_servers);
165 qdev_prop_set_uint32(dev, "nr_irqs", nr_irqs);
166 if (qdev_init(dev) < 0) {
170 return XICS_COMMON(dev);
173 static XICSState *xics_system_init(int nr_servers, int nr_irqs)
175 XICSState *icp = NULL;
178 QemuOpts *machine_opts = qemu_get_machine_opts();
179 bool irqchip_allowed = qemu_opt_get_bool(machine_opts,
180 "kernel_irqchip", true);
181 bool irqchip_required = qemu_opt_get_bool(machine_opts,
182 "kernel_irqchip", false);
183 if (irqchip_allowed) {
184 icp = try_create_xics(TYPE_KVM_XICS, nr_servers, nr_irqs);
187 if (irqchip_required && !icp) {
188 perror("Failed to create in-kernel XICS\n");
194 icp = try_create_xics(TYPE_XICS, nr_servers, nr_irqs);
198 perror("Failed to create XICS\n");
205 static int spapr_fixup_cpu_dt(void *fdt, sPAPREnvironment *spapr)
210 int smt = kvmppc_smt_threads();
211 uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
214 DeviceClass *dc = DEVICE_GET_CLASS(cpu);
215 int index = ppc_get_vcpu_dt_id(POWERPC_CPU(cpu));
216 uint32_t associativity[] = {cpu_to_be32(0x5),
220 cpu_to_be32(cpu->numa_node),
223 if ((index % smt) != 0) {
227 snprintf(cpu_model, 32, "/cpus/%s@%x", dc->fw_name,
230 offset = fdt_path_offset(fdt, cpu_model);
235 if (nb_numa_nodes > 1) {
236 ret = fdt_setprop(fdt, offset, "ibm,associativity", associativity,
237 sizeof(associativity));
243 ret = fdt_setprop(fdt, offset, "ibm,pft-size",
244 pft_size_prop, sizeof(pft_size_prop));
253 static size_t create_page_sizes_prop(CPUPPCState *env, uint32_t *prop,
256 size_t maxcells = maxsize / sizeof(uint32_t);
260 for (i = 0; i < PPC_PAGE_SIZES_MAX_SZ; i++) {
261 struct ppc_one_seg_page_size *sps = &env->sps.sps[i];
263 if (!sps->page_shift) {
266 for (count = 0; count < PPC_PAGE_SIZES_MAX_SZ; count++) {
267 if (sps->enc[count].page_shift == 0) {
271 if ((p - prop) >= (maxcells - 3 - count * 2)) {
274 *(p++) = cpu_to_be32(sps->page_shift);
275 *(p++) = cpu_to_be32(sps->slb_enc);
276 *(p++) = cpu_to_be32(count);
277 for (j = 0; j < count; j++) {
278 *(p++) = cpu_to_be32(sps->enc[j].page_shift);
279 *(p++) = cpu_to_be32(sps->enc[j].pte_enc);
283 return (p - prop) * sizeof(uint32_t);
290 fprintf(stderr, "qemu: error creating device tree: %s: %s\n", \
291 #exp, fdt_strerror(ret)); \
297 static void *spapr_create_fdt_skel(hwaddr initrd_base,
301 const char *boot_device,
302 const char *kernel_cmdline,
307 uint32_t start_prop = cpu_to_be32(initrd_base);
308 uint32_t end_prop = cpu_to_be32(initrd_base + initrd_size);
309 char hypertas_prop[] = "hcall-pft\0hcall-term\0hcall-dabr\0hcall-interrupt"
310 "\0hcall-tce\0hcall-vio\0hcall-splpar\0hcall-bulk\0hcall-set-mode";
311 char qemu_hypertas_prop[] = "hcall-memop1";
312 uint32_t refpoints[] = {cpu_to_be32(0x4), cpu_to_be32(0x4)};
313 uint32_t interrupt_server_ranges_prop[] = {0, cpu_to_be32(smp_cpus)};
314 int i, smt = kvmppc_smt_threads();
315 unsigned char vec5[] = {0x0, 0x0, 0x0, 0x0, 0x0, 0x80};
317 fdt = g_malloc0(FDT_MAX_SIZE);
318 _FDT((fdt_create(fdt, FDT_MAX_SIZE)));
321 _FDT((fdt_add_reservemap_entry(fdt, KERNEL_LOAD_ADDR, kernel_size)));
324 _FDT((fdt_add_reservemap_entry(fdt, initrd_base, initrd_size)));
326 _FDT((fdt_finish_reservemap(fdt)));
329 _FDT((fdt_begin_node(fdt, "")));
330 _FDT((fdt_property_string(fdt, "device_type", "chrp")));
331 _FDT((fdt_property_string(fdt, "model", "IBM pSeries (emulated by qemu)")));
332 _FDT((fdt_property_string(fdt, "compatible", "qemu,pseries")));
334 _FDT((fdt_property_cell(fdt, "#address-cells", 0x2)));
335 _FDT((fdt_property_cell(fdt, "#size-cells", 0x2)));
338 _FDT((fdt_begin_node(fdt, "chosen")));
340 /* Set Form1_affinity */
341 _FDT((fdt_property(fdt, "ibm,architecture-vec-5", vec5, sizeof(vec5))));
343 _FDT((fdt_property_string(fdt, "bootargs", kernel_cmdline)));
344 _FDT((fdt_property(fdt, "linux,initrd-start",
345 &start_prop, sizeof(start_prop))));
346 _FDT((fdt_property(fdt, "linux,initrd-end",
347 &end_prop, sizeof(end_prop))));
349 uint64_t kprop[2] = { cpu_to_be64(KERNEL_LOAD_ADDR),
350 cpu_to_be64(kernel_size) };
352 _FDT((fdt_property(fdt, "qemu,boot-kernel", &kprop, sizeof(kprop))));
354 _FDT((fdt_property(fdt, "qemu,boot-kernel-le", NULL, 0)));
358 _FDT((fdt_property_string(fdt, "qemu,boot-device", boot_device)));
360 _FDT((fdt_property_cell(fdt, "qemu,graphic-width", graphic_width)));
361 _FDT((fdt_property_cell(fdt, "qemu,graphic-height", graphic_height)));
362 _FDT((fdt_property_cell(fdt, "qemu,graphic-depth", graphic_depth)));
364 _FDT((fdt_end_node(fdt)));
367 _FDT((fdt_begin_node(fdt, "cpus")));
369 _FDT((fdt_property_cell(fdt, "#address-cells", 0x1)));
370 _FDT((fdt_property_cell(fdt, "#size-cells", 0x0)));
373 PowerPCCPU *cpu = POWERPC_CPU(cs);
374 CPUPPCState *env = &cpu->env;
375 DeviceClass *dc = DEVICE_GET_CLASS(cs);
376 PowerPCCPUClass *pcc = POWERPC_CPU_GET_CLASS(cs);
377 int index = ppc_get_vcpu_dt_id(cpu);
378 uint32_t servers_prop[smp_threads];
379 uint32_t gservers_prop[smp_threads * 2];
381 uint32_t segs[] = {cpu_to_be32(28), cpu_to_be32(40),
382 0xffffffff, 0xffffffff};
383 uint32_t tbfreq = kvm_enabled() ? kvmppc_get_tbfreq() : TIMEBASE_FREQ;
384 uint32_t cpufreq = kvm_enabled() ? kvmppc_get_clockfreq() : 1000000000;
385 uint32_t page_sizes_prop[64];
386 size_t page_sizes_prop_size;
388 if ((index % smt) != 0) {
392 nodename = g_strdup_printf("%s@%x", dc->fw_name, index);
394 _FDT((fdt_begin_node(fdt, nodename)));
398 _FDT((fdt_property_cell(fdt, "reg", index)));
399 _FDT((fdt_property_string(fdt, "device_type", "cpu")));
401 _FDT((fdt_property_cell(fdt, "cpu-version", env->spr[SPR_PVR])));
402 _FDT((fdt_property_cell(fdt, "d-cache-block-size",
403 env->dcache_line_size)));
404 _FDT((fdt_property_cell(fdt, "d-cache-line-size",
405 env->dcache_line_size)));
406 _FDT((fdt_property_cell(fdt, "i-cache-block-size",
407 env->icache_line_size)));
408 _FDT((fdt_property_cell(fdt, "i-cache-line-size",
409 env->icache_line_size)));
411 if (pcc->l1_dcache_size) {
412 _FDT((fdt_property_cell(fdt, "d-cache-size", pcc->l1_dcache_size)));
414 fprintf(stderr, "Warning: Unknown L1 dcache size for cpu\n");
416 if (pcc->l1_icache_size) {
417 _FDT((fdt_property_cell(fdt, "i-cache-size", pcc->l1_icache_size)));
419 fprintf(stderr, "Warning: Unknown L1 icache size for cpu\n");
422 _FDT((fdt_property_cell(fdt, "timebase-frequency", tbfreq)));
423 _FDT((fdt_property_cell(fdt, "clock-frequency", cpufreq)));
424 _FDT((fdt_property_cell(fdt, "ibm,slb-size", env->slb_nr)));
425 _FDT((fdt_property_string(fdt, "status", "okay")));
426 _FDT((fdt_property(fdt, "64-bit", NULL, 0)));
428 /* Build interrupt servers and gservers properties */
429 for (i = 0; i < smp_threads; i++) {
430 servers_prop[i] = cpu_to_be32(index + i);
431 /* Hack, direct the group queues back to cpu 0 */
432 gservers_prop[i*2] = cpu_to_be32(index + i);
433 gservers_prop[i*2 + 1] = 0;
435 _FDT((fdt_property(fdt, "ibm,ppc-interrupt-server#s",
436 servers_prop, sizeof(servers_prop))));
437 _FDT((fdt_property(fdt, "ibm,ppc-interrupt-gserver#s",
438 gservers_prop, sizeof(gservers_prop))));
440 if (env->spr_cb[SPR_PURR].oea_read) {
441 _FDT((fdt_property(fdt, "ibm,purr", NULL, 0)));
444 if (env->mmu_model & POWERPC_MMU_1TSEG) {
445 _FDT((fdt_property(fdt, "ibm,processor-segment-sizes",
446 segs, sizeof(segs))));
449 /* Advertise VMX/VSX (vector extensions) if available
450 * 0 / no property == no vector extensions
451 * 1 == VMX / Altivec available
452 * 2 == VSX available */
453 if (env->insns_flags & PPC_ALTIVEC) {
454 uint32_t vmx = (env->insns_flags2 & PPC2_VSX) ? 2 : 1;
456 _FDT((fdt_property_cell(fdt, "ibm,vmx", vmx)));
459 /* Advertise DFP (Decimal Floating Point) if available
460 * 0 / no property == no DFP
461 * 1 == DFP available */
462 if (env->insns_flags2 & PPC2_DFP) {
463 _FDT((fdt_property_cell(fdt, "ibm,dfp", 1)));
466 page_sizes_prop_size = create_page_sizes_prop(env, page_sizes_prop,
467 sizeof(page_sizes_prop));
468 if (page_sizes_prop_size) {
469 _FDT((fdt_property(fdt, "ibm,segment-page-sizes",
470 page_sizes_prop, page_sizes_prop_size)));
473 _FDT((fdt_end_node(fdt)));
476 _FDT((fdt_end_node(fdt)));
479 _FDT((fdt_begin_node(fdt, "rtas")));
481 _FDT((fdt_property(fdt, "ibm,hypertas-functions", hypertas_prop,
482 sizeof(hypertas_prop))));
483 _FDT((fdt_property(fdt, "qemu,hypertas-functions", qemu_hypertas_prop,
484 sizeof(qemu_hypertas_prop))));
486 _FDT((fdt_property(fdt, "ibm,associativity-reference-points",
487 refpoints, sizeof(refpoints))));
489 _FDT((fdt_property_cell(fdt, "rtas-error-log-max", RTAS_ERROR_LOG_MAX)));
491 _FDT((fdt_end_node(fdt)));
493 /* interrupt controller */
494 _FDT((fdt_begin_node(fdt, "interrupt-controller")));
496 _FDT((fdt_property_string(fdt, "device_type",
497 "PowerPC-External-Interrupt-Presentation")));
498 _FDT((fdt_property_string(fdt, "compatible", "IBM,ppc-xicp")));
499 _FDT((fdt_property(fdt, "interrupt-controller", NULL, 0)));
500 _FDT((fdt_property(fdt, "ibm,interrupt-server-ranges",
501 interrupt_server_ranges_prop,
502 sizeof(interrupt_server_ranges_prop))));
503 _FDT((fdt_property_cell(fdt, "#interrupt-cells", 2)));
504 _FDT((fdt_property_cell(fdt, "linux,phandle", PHANDLE_XICP)));
505 _FDT((fdt_property_cell(fdt, "phandle", PHANDLE_XICP)));
507 _FDT((fdt_end_node(fdt)));
510 _FDT((fdt_begin_node(fdt, "vdevice")));
512 _FDT((fdt_property_string(fdt, "device_type", "vdevice")));
513 _FDT((fdt_property_string(fdt, "compatible", "IBM,vdevice")));
514 _FDT((fdt_property_cell(fdt, "#address-cells", 0x1)));
515 _FDT((fdt_property_cell(fdt, "#size-cells", 0x0)));
516 _FDT((fdt_property_cell(fdt, "#interrupt-cells", 0x2)));
517 _FDT((fdt_property(fdt, "interrupt-controller", NULL, 0)));
519 _FDT((fdt_end_node(fdt)));
522 spapr_events_fdt_skel(fdt, epow_irq);
524 _FDT((fdt_end_node(fdt))); /* close root node */
525 _FDT((fdt_finish(fdt)));
530 static int spapr_populate_memory(sPAPREnvironment *spapr, void *fdt)
532 uint32_t associativity[] = {cpu_to_be32(0x4), cpu_to_be32(0x0),
533 cpu_to_be32(0x0), cpu_to_be32(0x0),
536 hwaddr node0_size, mem_start, node_size;
537 uint64_t mem_reg_property[2];
541 if (nb_numa_nodes > 1 && node_mem[0] < ram_size) {
542 node0_size = node_mem[0];
544 node0_size = ram_size;
548 mem_reg_property[0] = 0;
549 mem_reg_property[1] = cpu_to_be64(spapr->rma_size);
550 off = fdt_add_subnode(fdt, 0, "memory@0");
552 _FDT((fdt_setprop_string(fdt, off, "device_type", "memory")));
553 _FDT((fdt_setprop(fdt, off, "reg", mem_reg_property,
554 sizeof(mem_reg_property))));
555 _FDT((fdt_setprop(fdt, off, "ibm,associativity", associativity,
556 sizeof(associativity))));
559 if (node0_size > spapr->rma_size) {
560 mem_reg_property[0] = cpu_to_be64(spapr->rma_size);
561 mem_reg_property[1] = cpu_to_be64(node0_size - spapr->rma_size);
563 sprintf(mem_name, "memory@" TARGET_FMT_lx, spapr->rma_size);
564 off = fdt_add_subnode(fdt, 0, mem_name);
566 _FDT((fdt_setprop_string(fdt, off, "device_type", "memory")));
567 _FDT((fdt_setprop(fdt, off, "reg", mem_reg_property,
568 sizeof(mem_reg_property))));
569 _FDT((fdt_setprop(fdt, off, "ibm,associativity", associativity,
570 sizeof(associativity))));
573 /* RAM: Node 1 and beyond */
574 mem_start = node0_size;
575 for (i = 1; i < nb_numa_nodes; i++) {
576 mem_reg_property[0] = cpu_to_be64(mem_start);
577 if (mem_start >= ram_size) {
580 node_size = node_mem[i];
581 if (node_size > ram_size - mem_start) {
582 node_size = ram_size - mem_start;
585 mem_reg_property[1] = cpu_to_be64(node_size);
586 associativity[3] = associativity[4] = cpu_to_be32(i);
587 sprintf(mem_name, "memory@" TARGET_FMT_lx, mem_start);
588 off = fdt_add_subnode(fdt, 0, mem_name);
590 _FDT((fdt_setprop_string(fdt, off, "device_type", "memory")));
591 _FDT((fdt_setprop(fdt, off, "reg", mem_reg_property,
592 sizeof(mem_reg_property))));
593 _FDT((fdt_setprop(fdt, off, "ibm,associativity", associativity,
594 sizeof(associativity))));
595 mem_start += node_size;
601 static void spapr_finalize_fdt(sPAPREnvironment *spapr,
612 fdt = g_malloc(FDT_MAX_SIZE);
614 /* open out the base tree into a temp buffer for the final tweaks */
615 _FDT((fdt_open_into(spapr->fdt_skel, fdt, FDT_MAX_SIZE)));
617 ret = spapr_populate_memory(spapr, fdt);
619 fprintf(stderr, "couldn't setup memory nodes in fdt\n");
623 ret = spapr_populate_vdevice(spapr->vio_bus, fdt);
625 fprintf(stderr, "couldn't setup vio devices in fdt\n");
629 QLIST_FOREACH(phb, &spapr->phbs, list) {
630 ret = spapr_populate_pci_dt(phb, PHANDLE_XICP, fdt);
634 fprintf(stderr, "couldn't setup PCI devices in fdt\n");
639 ret = spapr_rtas_device_tree_setup(fdt, rtas_addr, rtas_size);
641 fprintf(stderr, "Couldn't set up RTAS device tree properties\n");
644 /* Advertise NUMA via ibm,associativity */
645 ret = spapr_fixup_cpu_dt(fdt, spapr);
647 fprintf(stderr, "Couldn't finalize CPU device tree properties\n");
650 bootlist = get_boot_devices_list(&cb, true);
651 if (cb && bootlist) {
652 int offset = fdt_path_offset(fdt, "/chosen");
656 for (i = 0; i < cb; i++) {
657 if (bootlist[i] == '\n') {
662 ret = fdt_setprop_string(fdt, offset, "qemu,boot-list", bootlist);
665 if (!spapr->has_graphics) {
666 spapr_populate_chosen_stdout(fdt, spapr->vio_bus);
669 _FDT((fdt_pack(fdt)));
671 if (fdt_totalsize(fdt) > FDT_MAX_SIZE) {
672 hw_error("FDT too big ! 0x%x bytes (max is 0x%x)\n",
673 fdt_totalsize(fdt), FDT_MAX_SIZE);
677 cpu_physical_memory_write(fdt_addr, fdt, fdt_totalsize(fdt));
682 static uint64_t translate_kernel_address(void *opaque, uint64_t addr)
684 return (addr & 0x0fffffff) + KERNEL_LOAD_ADDR;
687 static void emulate_spapr_hypercall(PowerPCCPU *cpu)
689 CPUPPCState *env = &cpu->env;
692 hcall_dprintf("Hypercall made with MSR[PR]=1\n");
693 env->gpr[3] = H_PRIVILEGE;
695 env->gpr[3] = spapr_hypercall(cpu, env->gpr[3], &env->gpr[4]);
699 static void spapr_reset_htab(sPAPREnvironment *spapr)
703 /* allocate hash page table. For now we always make this 16mb,
704 * later we should probably make it scale to the size of guest
707 shift = kvmppc_reset_htab(spapr->htab_shift);
710 /* Kernel handles htab, we don't need to allocate one */
711 spapr->htab_shift = shift;
712 kvmppc_kern_htab = true;
715 /* Allocate an htab if we don't yet have one */
716 spapr->htab = qemu_memalign(HTAB_SIZE(spapr), HTAB_SIZE(spapr));
720 memset(spapr->htab, 0, HTAB_SIZE(spapr));
723 /* Update the RMA size if necessary */
724 if (spapr->vrma_adjust) {
725 hwaddr node0_size = (nb_numa_nodes > 1) ? node_mem[0] : ram_size;
726 spapr->rma_size = kvmppc_rma_size(node0_size, spapr->htab_shift);
730 static void ppc_spapr_reset(void)
732 PowerPCCPU *first_ppc_cpu;
734 /* Reset the hash table & recalc the RMA */
735 spapr_reset_htab(spapr);
737 qemu_devices_reset();
740 spapr_finalize_fdt(spapr, spapr->fdt_addr, spapr->rtas_addr,
743 /* Set up the entry state */
744 first_ppc_cpu = POWERPC_CPU(first_cpu);
745 first_ppc_cpu->env.gpr[3] = spapr->fdt_addr;
746 first_ppc_cpu->env.gpr[5] = 0;
747 first_cpu->halted = 0;
748 first_ppc_cpu->env.nip = spapr->entry_point;
752 static void spapr_cpu_reset(void *opaque)
754 PowerPCCPU *cpu = opaque;
755 CPUState *cs = CPU(cpu);
756 CPUPPCState *env = &cpu->env;
760 /* All CPUs start halted. CPU0 is unhalted from the machine level
761 * reset code and the rest are explicitly started up by the guest
762 * using an RTAS call */
765 env->spr[SPR_HIOR] = 0;
767 env->external_htab = (uint8_t *)spapr->htab;
768 if (kvm_enabled() && !env->external_htab) {
770 * HV KVM, set external_htab to 1 so our ppc_hash64_load_hpte*
771 * functions do the right thing.
773 env->external_htab = (void *)1;
777 * htab_mask is the mask used to normalize hash value to PTEG index.
778 * htab_shift is log2 of hash table size.
779 * We have 8 hpte per group, and each hpte is 16 bytes.
780 * ie have 128 bytes per hpte entry.
782 env->htab_mask = (1ULL << ((spapr)->htab_shift - 7)) - 1;
783 env->spr[SPR_SDR1] = (target_ulong)(uintptr_t)spapr->htab |
784 (spapr->htab_shift - 18);
787 static void spapr_create_nvram(sPAPREnvironment *spapr)
789 DeviceState *dev = qdev_create(&spapr->vio_bus->bus, "spapr-nvram");
790 DriveInfo *dinfo = drive_get(IF_PFLASH, 0, 0);
793 qdev_prop_set_drive_nofail(dev, "drive", dinfo->bdrv);
796 qdev_init_nofail(dev);
798 spapr->nvram = (struct sPAPRNVRAM *)dev;
801 /* Returns whether we want to use VGA or not */
802 static int spapr_vga_init(PCIBus *pci_bus)
804 switch (vga_interface_type) {
810 return pci_vga_init(pci_bus) != NULL;
812 fprintf(stderr, "This vga model is not supported,"
813 "currently it only supports -vga std\n");
818 static const VMStateDescription vmstate_spapr = {
821 .minimum_version_id = 1,
822 .minimum_version_id_old = 1,
823 .fields = (VMStateField []) {
824 VMSTATE_UINT32(next_irq, sPAPREnvironment),
827 VMSTATE_UINT64(rtc_offset, sPAPREnvironment),
829 VMSTATE_END_OF_LIST()
833 #define HPTE(_table, _i) (void *)(((uint64_t *)(_table)) + ((_i) * 2))
834 #define HPTE_VALID(_hpte) (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_VALID)
835 #define HPTE_DIRTY(_hpte) (tswap64(*((uint64_t *)(_hpte))) & HPTE64_V_HPTE_DIRTY)
836 #define CLEAN_HPTE(_hpte) ((*(uint64_t *)(_hpte)) &= tswap64(~HPTE64_V_HPTE_DIRTY))
838 static int htab_save_setup(QEMUFile *f, void *opaque)
840 sPAPREnvironment *spapr = opaque;
842 /* "Iteration" header */
843 qemu_put_be32(f, spapr->htab_shift);
846 spapr->htab_save_index = 0;
847 spapr->htab_first_pass = true;
849 assert(kvm_enabled());
851 spapr->htab_fd = kvmppc_get_htab_fd(false);
852 if (spapr->htab_fd < 0) {
853 fprintf(stderr, "Unable to open fd for reading hash table from KVM: %s\n",
863 static void htab_save_first_pass(QEMUFile *f, sPAPREnvironment *spapr,
866 int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
867 int index = spapr->htab_save_index;
868 int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
870 assert(spapr->htab_first_pass);
875 /* Consume invalid HPTEs */
876 while ((index < htabslots)
877 && !HPTE_VALID(HPTE(spapr->htab, index))) {
879 CLEAN_HPTE(HPTE(spapr->htab, index));
882 /* Consume valid HPTEs */
884 while ((index < htabslots)
885 && HPTE_VALID(HPTE(spapr->htab, index))) {
887 CLEAN_HPTE(HPTE(spapr->htab, index));
890 if (index > chunkstart) {
891 int n_valid = index - chunkstart;
893 qemu_put_be32(f, chunkstart);
894 qemu_put_be16(f, n_valid);
896 qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
897 HASH_PTE_SIZE_64 * n_valid);
899 if ((qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
903 } while ((index < htabslots) && !qemu_file_rate_limit(f));
905 if (index >= htabslots) {
906 assert(index == htabslots);
908 spapr->htab_first_pass = false;
910 spapr->htab_save_index = index;
913 static int htab_save_later_pass(QEMUFile *f, sPAPREnvironment *spapr,
916 bool final = max_ns < 0;
917 int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
918 int examined = 0, sent = 0;
919 int index = spapr->htab_save_index;
920 int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
922 assert(!spapr->htab_first_pass);
925 int chunkstart, invalidstart;
927 /* Consume non-dirty HPTEs */
928 while ((index < htabslots)
929 && !HPTE_DIRTY(HPTE(spapr->htab, index))) {
935 /* Consume valid dirty HPTEs */
936 while ((index < htabslots)
937 && HPTE_DIRTY(HPTE(spapr->htab, index))
938 && HPTE_VALID(HPTE(spapr->htab, index))) {
939 CLEAN_HPTE(HPTE(spapr->htab, index));
944 invalidstart = index;
945 /* Consume invalid dirty HPTEs */
946 while ((index < htabslots)
947 && HPTE_DIRTY(HPTE(spapr->htab, index))
948 && !HPTE_VALID(HPTE(spapr->htab, index))) {
949 CLEAN_HPTE(HPTE(spapr->htab, index));
954 if (index > chunkstart) {
955 int n_valid = invalidstart - chunkstart;
956 int n_invalid = index - invalidstart;
958 qemu_put_be32(f, chunkstart);
959 qemu_put_be16(f, n_valid);
960 qemu_put_be16(f, n_invalid);
961 qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
962 HASH_PTE_SIZE_64 * n_valid);
963 sent += index - chunkstart;
965 if (!final && (qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
970 if (examined >= htabslots) {
974 if (index >= htabslots) {
975 assert(index == htabslots);
978 } while ((examined < htabslots) && (!qemu_file_rate_limit(f) || final));
980 if (index >= htabslots) {
981 assert(index == htabslots);
985 spapr->htab_save_index = index;
987 return (examined >= htabslots) && (sent == 0) ? 1 : 0;
990 #define MAX_ITERATION_NS 5000000 /* 5 ms */
991 #define MAX_KVM_BUF_SIZE 2048
993 static int htab_save_iterate(QEMUFile *f, void *opaque)
995 sPAPREnvironment *spapr = opaque;
998 /* Iteration header */
1002 assert(kvm_enabled());
1004 rc = kvmppc_save_htab(f, spapr->htab_fd,
1005 MAX_KVM_BUF_SIZE, MAX_ITERATION_NS);
1009 } else if (spapr->htab_first_pass) {
1010 htab_save_first_pass(f, spapr, MAX_ITERATION_NS);
1012 rc = htab_save_later_pass(f, spapr, MAX_ITERATION_NS);
1016 qemu_put_be32(f, 0);
1017 qemu_put_be16(f, 0);
1018 qemu_put_be16(f, 0);
1023 static int htab_save_complete(QEMUFile *f, void *opaque)
1025 sPAPREnvironment *spapr = opaque;
1027 /* Iteration header */
1028 qemu_put_be32(f, 0);
1033 assert(kvm_enabled());
1035 rc = kvmppc_save_htab(f, spapr->htab_fd, MAX_KVM_BUF_SIZE, -1);
1039 close(spapr->htab_fd);
1040 spapr->htab_fd = -1;
1042 htab_save_later_pass(f, spapr, -1);
1046 qemu_put_be32(f, 0);
1047 qemu_put_be16(f, 0);
1048 qemu_put_be16(f, 0);
1053 static int htab_load(QEMUFile *f, void *opaque, int version_id)
1055 sPAPREnvironment *spapr = opaque;
1056 uint32_t section_hdr;
1059 if (version_id < 1 || version_id > 1) {
1060 fprintf(stderr, "htab_load() bad version\n");
1064 section_hdr = qemu_get_be32(f);
1067 /* First section, just the hash shift */
1068 if (spapr->htab_shift != section_hdr) {
1075 assert(kvm_enabled());
1077 fd = kvmppc_get_htab_fd(true);
1079 fprintf(stderr, "Unable to open fd to restore KVM hash table: %s\n",
1086 uint16_t n_valid, n_invalid;
1088 index = qemu_get_be32(f);
1089 n_valid = qemu_get_be16(f);
1090 n_invalid = qemu_get_be16(f);
1092 if ((index == 0) && (n_valid == 0) && (n_invalid == 0)) {
1097 if ((index + n_valid + n_invalid) >
1098 (HTAB_SIZE(spapr) / HASH_PTE_SIZE_64)) {
1099 /* Bad index in stream */
1100 fprintf(stderr, "htab_load() bad index %d (%hd+%hd entries) "
1101 "in htab stream (htab_shift=%d)\n", index, n_valid, n_invalid,
1108 qemu_get_buffer(f, HPTE(spapr->htab, index),
1109 HASH_PTE_SIZE_64 * n_valid);
1112 memset(HPTE(spapr->htab, index + n_valid), 0,
1113 HASH_PTE_SIZE_64 * n_invalid);
1120 rc = kvmppc_load_htab_chunk(f, fd, index, n_valid, n_invalid);
1135 static SaveVMHandlers savevm_htab_handlers = {
1136 .save_live_setup = htab_save_setup,
1137 .save_live_iterate = htab_save_iterate,
1138 .save_live_complete = htab_save_complete,
1139 .load_state = htab_load,
1142 /* pSeries LPAR / sPAPR hardware init */
1143 static void ppc_spapr_init(MachineState *machine)
1145 ram_addr_t ram_size = machine->ram_size;
1146 const char *cpu_model = machine->cpu_model;
1147 const char *kernel_filename = machine->kernel_filename;
1148 const char *kernel_cmdline = machine->kernel_cmdline;
1149 const char *initrd_filename = machine->initrd_filename;
1150 const char *boot_device = machine->boot_order;
1155 MemoryRegion *sysmem = get_system_memory();
1156 MemoryRegion *ram = g_new(MemoryRegion, 1);
1157 hwaddr rma_alloc_size;
1158 hwaddr node0_size = (nb_numa_nodes > 1) ? node_mem[0] : ram_size;
1159 uint32_t initrd_base = 0;
1160 long kernel_size = 0, initrd_size = 0;
1161 long load_limit, rtas_limit, fw_size;
1162 bool kernel_le = false;
1165 msi_supported = true;
1167 spapr = g_malloc0(sizeof(*spapr));
1168 QLIST_INIT(&spapr->phbs);
1170 cpu_ppc_hypercall = emulate_spapr_hypercall;
1172 /* Allocate RMA if necessary */
1173 rma_alloc_size = kvmppc_alloc_rma("ppc_spapr.rma", sysmem);
1175 if (rma_alloc_size == -1) {
1176 hw_error("qemu: Unable to create RMA\n");
1180 if (rma_alloc_size && (rma_alloc_size < node0_size)) {
1181 spapr->rma_size = rma_alloc_size;
1183 spapr->rma_size = node0_size;
1185 /* With KVM, we don't actually know whether KVM supports an
1186 * unbounded RMA (PR KVM) or is limited by the hash table size
1187 * (HV KVM using VRMA), so we always assume the latter
1189 * In that case, we also limit the initial allocations for RTAS
1190 * etc... to 256M since we have no way to know what the VRMA size
1191 * is going to be as it depends on the size of the hash table
1192 * isn't determined yet.
1194 if (kvm_enabled()) {
1195 spapr->vrma_adjust = 1;
1196 spapr->rma_size = MIN(spapr->rma_size, 0x10000000);
1200 if (spapr->rma_size > node0_size) {
1201 fprintf(stderr, "Error: Numa node 0 has to span the RMA (%#08"HWADDR_PRIx")\n",
1206 /* We place the device tree and RTAS just below either the top of the RMA,
1207 * or just below 2GB, whichever is lowere, so that it can be
1208 * processed with 32-bit real mode code if necessary */
1209 rtas_limit = MIN(spapr->rma_size, 0x80000000);
1210 spapr->rtas_addr = rtas_limit - RTAS_MAX_SIZE;
1211 spapr->fdt_addr = spapr->rtas_addr - FDT_MAX_SIZE;
1212 load_limit = spapr->fdt_addr - FW_OVERHEAD;
1214 /* We aim for a hash table of size 1/128 the size of RAM. The
1215 * normal rule of thumb is 1/64 the size of RAM, but that's much
1216 * more than needed for the Linux guests we support. */
1217 spapr->htab_shift = 18; /* Minimum architected size */
1218 while (spapr->htab_shift <= 46) {
1219 if ((1ULL << (spapr->htab_shift + 7)) >= ram_size) {
1222 spapr->htab_shift++;
1225 /* Set up Interrupt Controller before we create the VCPUs */
1226 spapr->icp = xics_system_init(smp_cpus * kvmppc_smt_threads() / smp_threads,
1228 spapr->next_irq = XICS_IRQ_BASE;
1231 if (cpu_model == NULL) {
1232 cpu_model = kvm_enabled() ? "host" : "POWER7";
1234 for (i = 0; i < smp_cpus; i++) {
1235 cpu = cpu_ppc_init(cpu_model);
1237 fprintf(stderr, "Unable to find PowerPC CPU definition\n");
1242 /* Set time-base frequency to 512 MHz */
1243 cpu_ppc_tb_init(env, TIMEBASE_FREQ);
1245 /* PAPR always has exception vectors in RAM not ROM. To ensure this,
1246 * MSR[IP] should never be set.
1248 env->msr_mask &= ~(1 << 6);
1250 /* Tell KVM that we're in PAPR mode */
1251 if (kvm_enabled()) {
1252 kvmppc_set_papr(cpu);
1255 xics_cpu_setup(spapr->icp, cpu);
1257 qemu_register_reset(spapr_cpu_reset, cpu);
1261 spapr->ram_limit = ram_size;
1262 if (spapr->ram_limit > rma_alloc_size) {
1263 ram_addr_t nonrma_base = rma_alloc_size;
1264 ram_addr_t nonrma_size = spapr->ram_limit - rma_alloc_size;
1266 memory_region_init_ram(ram, NULL, "ppc_spapr.ram", nonrma_size);
1267 vmstate_register_ram_global(ram);
1268 memory_region_add_subregion(sysmem, nonrma_base, ram);
1271 filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, "spapr-rtas.bin");
1272 spapr->rtas_size = load_image_targphys(filename, spapr->rtas_addr,
1273 rtas_limit - spapr->rtas_addr);
1274 if (spapr->rtas_size < 0) {
1275 hw_error("qemu: could not load LPAR rtas '%s'\n", filename);
1278 if (spapr->rtas_size > RTAS_MAX_SIZE) {
1279 hw_error("RTAS too big ! 0x%lx bytes (max is 0x%x)\n",
1280 spapr->rtas_size, RTAS_MAX_SIZE);
1285 /* Set up EPOW events infrastructure */
1286 spapr_events_init(spapr);
1288 /* Set up VIO bus */
1289 spapr->vio_bus = spapr_vio_bus_init();
1291 for (i = 0; i < MAX_SERIAL_PORTS; i++) {
1292 if (serial_hds[i]) {
1293 spapr_vty_create(spapr->vio_bus, serial_hds[i]);
1297 /* We always have at least the nvram device on VIO */
1298 spapr_create_nvram(spapr);
1301 spapr_pci_msi_init(spapr, SPAPR_PCI_MSI_WINDOW);
1302 spapr_pci_rtas_init();
1304 phb = spapr_create_phb(spapr, 0);
1306 for (i = 0; i < nb_nics; i++) {
1307 NICInfo *nd = &nd_table[i];
1310 nd->model = g_strdup("ibmveth");
1313 if (strcmp(nd->model, "ibmveth") == 0) {
1314 spapr_vlan_create(spapr->vio_bus, nd);
1316 pci_nic_init_nofail(&nd_table[i], phb->bus, nd->model, NULL);
1320 for (i = 0; i <= drive_get_max_bus(IF_SCSI); i++) {
1321 spapr_vscsi_create(spapr->vio_bus);
1325 if (spapr_vga_init(phb->bus)) {
1326 spapr->has_graphics = true;
1329 if (usb_enabled(spapr->has_graphics)) {
1330 pci_create_simple(phb->bus, -1, "pci-ohci");
1331 if (spapr->has_graphics) {
1332 usbdevice_create("keyboard");
1333 usbdevice_create("mouse");
1337 if (spapr->rma_size < (MIN_RMA_SLOF << 20)) {
1338 fprintf(stderr, "qemu: pSeries SLOF firmware requires >= "
1339 "%ldM guest RMA (Real Mode Area memory)\n", MIN_RMA_SLOF);
1343 if (kernel_filename) {
1344 uint64_t lowaddr = 0;
1346 kernel_size = load_elf(kernel_filename, translate_kernel_address, NULL,
1347 NULL, &lowaddr, NULL, 1, ELF_MACHINE, 0);
1348 if (kernel_size == ELF_LOAD_WRONG_ENDIAN) {
1349 kernel_size = load_elf(kernel_filename,
1350 translate_kernel_address, NULL,
1351 NULL, &lowaddr, NULL, 0, ELF_MACHINE, 0);
1352 kernel_le = kernel_size > 0;
1354 if (kernel_size < 0) {
1355 fprintf(stderr, "qemu: error loading %s: %s\n",
1356 kernel_filename, load_elf_strerror(kernel_size));
1361 if (initrd_filename) {
1362 /* Try to locate the initrd in the gap between the kernel
1363 * and the firmware. Add a bit of space just in case
1365 initrd_base = (KERNEL_LOAD_ADDR + kernel_size + 0x1ffff) & ~0xffff;
1366 initrd_size = load_image_targphys(initrd_filename, initrd_base,
1367 load_limit - initrd_base);
1368 if (initrd_size < 0) {
1369 fprintf(stderr, "qemu: could not load initial ram disk '%s'\n",
1379 if (bios_name == NULL) {
1380 bios_name = FW_FILE_NAME;
1382 filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, bios_name);
1383 fw_size = load_image_targphys(filename, 0, FW_MAX_SIZE);
1385 hw_error("qemu: could not load LPAR rtas '%s'\n", filename);
1390 spapr->entry_point = 0x100;
1392 vmstate_register(NULL, 0, &vmstate_spapr, spapr);
1393 register_savevm_live(NULL, "spapr/htab", -1, 1,
1394 &savevm_htab_handlers, spapr);
1396 /* Prepare the device tree */
1397 spapr->fdt_skel = spapr_create_fdt_skel(initrd_base, initrd_size,
1398 kernel_size, kernel_le,
1399 boot_device, kernel_cmdline,
1401 assert(spapr->fdt_skel != NULL);
1404 static int spapr_kvm_type(const char *vm_type)
1410 if (!strcmp(vm_type, "HV")) {
1414 if (!strcmp(vm_type, "PR")) {
1418 error_report("Unknown kvm-type specified '%s'", vm_type);
1423 * Implementation of an interface to adjust firmware patch
1424 * for the bootindex property handling.
1426 static char *spapr_get_fw_dev_path(FWPathProvider *p, BusState *bus,
1429 #define CAST(type, obj, name) \
1430 ((type *)object_dynamic_cast(OBJECT(obj), (name)))
1431 SCSIDevice *d = CAST(SCSIDevice, dev, TYPE_SCSI_DEVICE);
1432 sPAPRPHBState *phb = CAST(sPAPRPHBState, dev, TYPE_SPAPR_PCI_HOST_BRIDGE);
1435 void *spapr = CAST(void, bus->parent, "spapr-vscsi");
1436 VirtIOSCSI *virtio = CAST(VirtIOSCSI, bus->parent, TYPE_VIRTIO_SCSI);
1437 USBDevice *usb = CAST(USBDevice, bus->parent, TYPE_USB_DEVICE);
1441 * Replace "channel@0/disk@0,0" with "disk@8000000000000000":
1442 * We use SRP luns of the form 8000 | (bus << 8) | (id << 5) | lun
1443 * in the top 16 bits of the 64-bit LUN
1445 unsigned id = 0x8000 | (d->id << 8) | d->lun;
1446 return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
1447 (uint64_t)id << 48);
1448 } else if (virtio) {
1450 * We use SRP luns of the form 01000000 | (target << 8) | lun
1451 * in the top 32 bits of the 64-bit LUN
1452 * Note: the quote above is from SLOF and it is wrong,
1453 * the actual binding is:
1454 * swap 0100 or 10 << or 20 << ( target lun-id -- srplun )
1456 unsigned id = 0x1000000 | (d->id << 16) | d->lun;
1457 return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
1458 (uint64_t)id << 32);
1461 * We use SRP luns of the form 01000000 | (usb-port << 16) | lun
1462 * in the top 32 bits of the 64-bit LUN
1464 unsigned usb_port = atoi(usb->port->path);
1465 unsigned id = 0x1000000 | (usb_port << 16) | d->lun;
1466 return g_strdup_printf("%s@%"PRIX64, qdev_fw_name(dev),
1467 (uint64_t)id << 32);
1472 /* Replace "pci" with "pci@800000020000000" */
1473 return g_strdup_printf("pci@%"PRIX64, phb->buid);
1479 static void spapr_machine_class_init(ObjectClass *oc, void *data)
1481 MachineClass *mc = MACHINE_CLASS(oc);
1482 FWPathProviderClass *fwc = FW_PATH_PROVIDER_CLASS(oc);
1484 mc->name = "pseries";
1485 mc->desc = "pSeries Logical Partition (PAPR compliant)";
1487 mc->init = ppc_spapr_init;
1488 mc->reset = ppc_spapr_reset;
1489 mc->block_default_type = IF_SCSI;
1490 mc->max_cpus = MAX_CPUS;
1491 mc->no_parallel = 1;
1492 mc->default_boot_order = NULL;
1493 mc->kvm_type = spapr_kvm_type;
1495 fwc->get_dev_path = spapr_get_fw_dev_path;
1498 static const TypeInfo spapr_machine_info = {
1499 .name = TYPE_SPAPR_MACHINE,
1500 .parent = TYPE_MACHINE,
1501 .class_init = spapr_machine_class_init,
1502 .interfaces = (InterfaceInfo[]) {
1503 { TYPE_FW_PATH_PROVIDER },
1508 static void spapr_machine_register_types(void)
1510 type_register_static(&spapr_machine_info);
1513 type_init(spapr_machine_register_types)