xref: /openbmc/qemu/hw/ppc/spapr.c (revision fd506b4f)
1 /*
2  * QEMU PowerPC pSeries Logical Partition (aka sPAPR) hardware System Emulator
3  *
4  * Copyright (c) 2004-2007 Fabrice Bellard
5  * Copyright (c) 2007 Jocelyn Mayer
6  * Copyright (c) 2010 David Gibson, IBM Corporation.
7  *
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:
14  *
15  * The above copyright notice and this permission notice shall be included in
16  * all copies or substantial portions of the Software.
17  *
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.
25  *
26  */
27 #include "sysemu/sysemu.h"
28 #include "hw/hw.h"
29 #include "elf.h"
30 #include "net/net.h"
31 #include "sysemu/blockdev.h"
32 #include "sysemu/cpus.h"
33 #include "sysemu/kvm.h"
34 #include "kvm_ppc.h"
35 
36 #include "hw/boards.h"
37 #include "hw/ppc/ppc.h"
38 #include "hw/loader.h"
39 
40 #include "hw/ppc/spapr.h"
41 #include "hw/ppc/spapr_vio.h"
42 #include "hw/pci-host/spapr.h"
43 #include "hw/ppc/xics.h"
44 #include "hw/pci/msi.h"
45 
46 #include "sysemu/kvm.h"
47 #include "kvm_ppc.h"
48 #include "hw/pci/pci.h"
49 
50 #include "exec/address-spaces.h"
51 #include "hw/usb.h"
52 #include "qemu/config-file.h"
53 
54 #include <libfdt.h>
55 
56 /* SLOF memory layout:
57  *
58  * SLOF raw image loaded at 0, copies its romfs right below the flat
59  * device-tree, then position SLOF itself 31M below that
60  *
61  * So we set FW_OVERHEAD to 40MB which should account for all of that
62  * and more
63  *
64  * We load our kernel at 4M, leaving space for SLOF initial image
65  */
66 #define FDT_MAX_SIZE            0x10000
67 #define RTAS_MAX_SIZE           0x10000
68 #define FW_MAX_SIZE             0x400000
69 #define FW_FILE_NAME            "slof.bin"
70 #define FW_OVERHEAD             0x2800000
71 #define KERNEL_LOAD_ADDR        FW_MAX_SIZE
72 
73 #define MIN_RMA_SLOF            128UL
74 
75 #define TIMEBASE_FREQ           512000000ULL
76 
77 #define MAX_CPUS                256
78 #define XICS_IRQS               1024
79 
80 #define PHANDLE_XICP            0x00001111
81 
82 #define HTAB_SIZE(spapr)        (1ULL << ((spapr)->htab_shift))
83 
84 sPAPREnvironment *spapr;
85 
86 int spapr_allocate_irq(int hint, bool lsi)
87 {
88     int irq;
89 
90     if (hint) {
91         irq = hint;
92         /* FIXME: we should probably check for collisions somehow */
93     } else {
94         irq = spapr->next_irq++;
95     }
96 
97     /* Configure irq type */
98     if (!xics_get_qirq(spapr->icp, irq)) {
99         return 0;
100     }
101 
102     xics_set_irq_type(spapr->icp, irq, lsi);
103 
104     return irq;
105 }
106 
107 /* Allocate block of consequtive IRQs, returns a number of the first */
108 int spapr_allocate_irq_block(int num, bool lsi)
109 {
110     int first = -1;
111     int i;
112 
113     for (i = 0; i < num; ++i) {
114         int irq;
115 
116         irq = spapr_allocate_irq(0, lsi);
117         if (!irq) {
118             return -1;
119         }
120 
121         if (0 == i) {
122             first = irq;
123         }
124 
125         /* If the above doesn't create a consecutive block then that's
126          * an internal bug */
127         assert(irq == (first + i));
128     }
129 
130     return first;
131 }
132 
133 static int spapr_fixup_cpu_dt(void *fdt, sPAPREnvironment *spapr)
134 {
135     int ret = 0, offset;
136     CPUPPCState *env;
137     CPUState *cpu;
138     char cpu_model[32];
139     int smt = kvmppc_smt_threads();
140     uint32_t pft_size_prop[] = {0, cpu_to_be32(spapr->htab_shift)};
141 
142     assert(spapr->cpu_model);
143 
144     for (env = first_cpu; env != NULL; env = env->next_cpu) {
145         cpu = CPU(ppc_env_get_cpu(env));
146         uint32_t associativity[] = {cpu_to_be32(0x5),
147                                     cpu_to_be32(0x0),
148                                     cpu_to_be32(0x0),
149                                     cpu_to_be32(0x0),
150                                     cpu_to_be32(cpu->numa_node),
151                                     cpu_to_be32(cpu->cpu_index)};
152 
153         if ((cpu->cpu_index % smt) != 0) {
154             continue;
155         }
156 
157         snprintf(cpu_model, 32, "/cpus/%s@%x", spapr->cpu_model,
158                  cpu->cpu_index);
159 
160         offset = fdt_path_offset(fdt, cpu_model);
161         if (offset < 0) {
162             return offset;
163         }
164 
165         if (nb_numa_nodes > 1) {
166             ret = fdt_setprop(fdt, offset, "ibm,associativity", associativity,
167                               sizeof(associativity));
168             if (ret < 0) {
169                 return ret;
170             }
171         }
172 
173         ret = fdt_setprop(fdt, offset, "ibm,pft-size",
174                           pft_size_prop, sizeof(pft_size_prop));
175         if (ret < 0) {
176             return ret;
177         }
178     }
179     return ret;
180 }
181 
182 
183 static size_t create_page_sizes_prop(CPUPPCState *env, uint32_t *prop,
184                                      size_t maxsize)
185 {
186     size_t maxcells = maxsize / sizeof(uint32_t);
187     int i, j, count;
188     uint32_t *p = prop;
189 
190     for (i = 0; i < PPC_PAGE_SIZES_MAX_SZ; i++) {
191         struct ppc_one_seg_page_size *sps = &env->sps.sps[i];
192 
193         if (!sps->page_shift) {
194             break;
195         }
196         for (count = 0; count < PPC_PAGE_SIZES_MAX_SZ; count++) {
197             if (sps->enc[count].page_shift == 0) {
198                 break;
199             }
200         }
201         if ((p - prop) >= (maxcells - 3 - count * 2)) {
202             break;
203         }
204         *(p++) = cpu_to_be32(sps->page_shift);
205         *(p++) = cpu_to_be32(sps->slb_enc);
206         *(p++) = cpu_to_be32(count);
207         for (j = 0; j < count; j++) {
208             *(p++) = cpu_to_be32(sps->enc[j].page_shift);
209             *(p++) = cpu_to_be32(sps->enc[j].pte_enc);
210         }
211     }
212 
213     return (p - prop) * sizeof(uint32_t);
214 }
215 
216 #define _FDT(exp) \
217     do { \
218         int ret = (exp);                                           \
219         if (ret < 0) {                                             \
220             fprintf(stderr, "qemu: error creating device tree: %s: %s\n", \
221                     #exp, fdt_strerror(ret));                      \
222             exit(1);                                               \
223         }                                                          \
224     } while (0)
225 
226 
227 static void *spapr_create_fdt_skel(const char *cpu_model,
228                                    hwaddr initrd_base,
229                                    hwaddr initrd_size,
230                                    hwaddr kernel_size,
231                                    const char *boot_device,
232                                    const char *kernel_cmdline,
233                                    uint32_t epow_irq)
234 {
235     void *fdt;
236     CPUPPCState *env;
237     uint32_t start_prop = cpu_to_be32(initrd_base);
238     uint32_t end_prop = cpu_to_be32(initrd_base + initrd_size);
239     char hypertas_prop[] = "hcall-pft\0hcall-term\0hcall-dabr\0hcall-interrupt"
240         "\0hcall-tce\0hcall-vio\0hcall-splpar\0hcall-bulk";
241     char qemu_hypertas_prop[] = "hcall-memop1";
242     uint32_t refpoints[] = {cpu_to_be32(0x4), cpu_to_be32(0x4)};
243     uint32_t interrupt_server_ranges_prop[] = {0, cpu_to_be32(smp_cpus)};
244     char *modelname;
245     int i, smt = kvmppc_smt_threads();
246     unsigned char vec5[] = {0x0, 0x0, 0x0, 0x0, 0x0, 0x80};
247 
248     fdt = g_malloc0(FDT_MAX_SIZE);
249     _FDT((fdt_create(fdt, FDT_MAX_SIZE)));
250 
251     if (kernel_size) {
252         _FDT((fdt_add_reservemap_entry(fdt, KERNEL_LOAD_ADDR, kernel_size)));
253     }
254     if (initrd_size) {
255         _FDT((fdt_add_reservemap_entry(fdt, initrd_base, initrd_size)));
256     }
257     _FDT((fdt_finish_reservemap(fdt)));
258 
259     /* Root node */
260     _FDT((fdt_begin_node(fdt, "")));
261     _FDT((fdt_property_string(fdt, "device_type", "chrp")));
262     _FDT((fdt_property_string(fdt, "model", "IBM pSeries (emulated by qemu)")));
263     _FDT((fdt_property_string(fdt, "compatible", "qemu,pseries")));
264 
265     _FDT((fdt_property_cell(fdt, "#address-cells", 0x2)));
266     _FDT((fdt_property_cell(fdt, "#size-cells", 0x2)));
267 
268     /* /chosen */
269     _FDT((fdt_begin_node(fdt, "chosen")));
270 
271     /* Set Form1_affinity */
272     _FDT((fdt_property(fdt, "ibm,architecture-vec-5", vec5, sizeof(vec5))));
273 
274     _FDT((fdt_property_string(fdt, "bootargs", kernel_cmdline)));
275     _FDT((fdt_property(fdt, "linux,initrd-start",
276                        &start_prop, sizeof(start_prop))));
277     _FDT((fdt_property(fdt, "linux,initrd-end",
278                        &end_prop, sizeof(end_prop))));
279     if (kernel_size) {
280         uint64_t kprop[2] = { cpu_to_be64(KERNEL_LOAD_ADDR),
281                               cpu_to_be64(kernel_size) };
282 
283         _FDT((fdt_property(fdt, "qemu,boot-kernel", &kprop, sizeof(kprop))));
284     }
285     if (boot_device) {
286         _FDT((fdt_property_string(fdt, "qemu,boot-device", boot_device)));
287     }
288     _FDT((fdt_property_cell(fdt, "qemu,graphic-width", graphic_width)));
289     _FDT((fdt_property_cell(fdt, "qemu,graphic-height", graphic_height)));
290     _FDT((fdt_property_cell(fdt, "qemu,graphic-depth", graphic_depth)));
291 
292     _FDT((fdt_end_node(fdt)));
293 
294     /* cpus */
295     _FDT((fdt_begin_node(fdt, "cpus")));
296 
297     _FDT((fdt_property_cell(fdt, "#address-cells", 0x1)));
298     _FDT((fdt_property_cell(fdt, "#size-cells", 0x0)));
299 
300     modelname = g_strdup(cpu_model);
301 
302     for (i = 0; i < strlen(modelname); i++) {
303         modelname[i] = toupper(modelname[i]);
304     }
305 
306     /* This is needed during FDT finalization */
307     spapr->cpu_model = g_strdup(modelname);
308 
309     for (env = first_cpu; env != NULL; env = env->next_cpu) {
310         CPUState *cpu = CPU(ppc_env_get_cpu(env));
311         PowerPCCPUClass *pcc = POWERPC_CPU_GET_CLASS(cpu);
312         int index = cpu->cpu_index;
313         uint32_t servers_prop[smp_threads];
314         uint32_t gservers_prop[smp_threads * 2];
315         char *nodename;
316         uint32_t segs[] = {cpu_to_be32(28), cpu_to_be32(40),
317                            0xffffffff, 0xffffffff};
318         uint32_t tbfreq = kvm_enabled() ? kvmppc_get_tbfreq() : TIMEBASE_FREQ;
319         uint32_t cpufreq = kvm_enabled() ? kvmppc_get_clockfreq() : 1000000000;
320         uint32_t page_sizes_prop[64];
321         size_t page_sizes_prop_size;
322 
323         if ((index % smt) != 0) {
324             continue;
325         }
326 
327         nodename = g_strdup_printf("%s@%x", modelname, index);
328 
329         _FDT((fdt_begin_node(fdt, nodename)));
330 
331         g_free(nodename);
332 
333         _FDT((fdt_property_cell(fdt, "reg", index)));
334         _FDT((fdt_property_string(fdt, "device_type", "cpu")));
335 
336         _FDT((fdt_property_cell(fdt, "cpu-version", env->spr[SPR_PVR])));
337         _FDT((fdt_property_cell(fdt, "d-cache-block-size",
338                                 env->dcache_line_size)));
339         _FDT((fdt_property_cell(fdt, "d-cache-line-size",
340                                 env->dcache_line_size)));
341         _FDT((fdt_property_cell(fdt, "i-cache-block-size",
342                                 env->icache_line_size)));
343         _FDT((fdt_property_cell(fdt, "i-cache-line-size",
344                                 env->icache_line_size)));
345 
346         if (pcc->l1_dcache_size) {
347             _FDT((fdt_property_cell(fdt, "d-cache-size", pcc->l1_dcache_size)));
348         } else {
349             fprintf(stderr, "Warning: Unknown L1 dcache size for cpu\n");
350         }
351         if (pcc->l1_icache_size) {
352             _FDT((fdt_property_cell(fdt, "i-cache-size", pcc->l1_icache_size)));
353         } else {
354             fprintf(stderr, "Warning: Unknown L1 icache size for cpu\n");
355         }
356 
357         _FDT((fdt_property_cell(fdt, "timebase-frequency", tbfreq)));
358         _FDT((fdt_property_cell(fdt, "clock-frequency", cpufreq)));
359         _FDT((fdt_property_cell(fdt, "ibm,slb-size", env->slb_nr)));
360         _FDT((fdt_property_string(fdt, "status", "okay")));
361         _FDT((fdt_property(fdt, "64-bit", NULL, 0)));
362 
363         /* Build interrupt servers and gservers properties */
364         for (i = 0; i < smp_threads; i++) {
365             servers_prop[i] = cpu_to_be32(index + i);
366             /* Hack, direct the group queues back to cpu 0 */
367             gservers_prop[i*2] = cpu_to_be32(index + i);
368             gservers_prop[i*2 + 1] = 0;
369         }
370         _FDT((fdt_property(fdt, "ibm,ppc-interrupt-server#s",
371                            servers_prop, sizeof(servers_prop))));
372         _FDT((fdt_property(fdt, "ibm,ppc-interrupt-gserver#s",
373                            gservers_prop, sizeof(gservers_prop))));
374 
375         if (env->mmu_model & POWERPC_MMU_1TSEG) {
376             _FDT((fdt_property(fdt, "ibm,processor-segment-sizes",
377                                segs, sizeof(segs))));
378         }
379 
380         /* Advertise VMX/VSX (vector extensions) if available
381          *   0 / no property == no vector extensions
382          *   1               == VMX / Altivec available
383          *   2               == VSX available */
384         if (env->insns_flags & PPC_ALTIVEC) {
385             uint32_t vmx = (env->insns_flags2 & PPC2_VSX) ? 2 : 1;
386 
387             _FDT((fdt_property_cell(fdt, "ibm,vmx", vmx)));
388         }
389 
390         /* Advertise DFP (Decimal Floating Point) if available
391          *   0 / no property == no DFP
392          *   1               == DFP available */
393         if (env->insns_flags2 & PPC2_DFP) {
394             _FDT((fdt_property_cell(fdt, "ibm,dfp", 1)));
395         }
396 
397         page_sizes_prop_size = create_page_sizes_prop(env, page_sizes_prop,
398                                                       sizeof(page_sizes_prop));
399         if (page_sizes_prop_size) {
400             _FDT((fdt_property(fdt, "ibm,segment-page-sizes",
401                                page_sizes_prop, page_sizes_prop_size)));
402         }
403 
404         _FDT((fdt_end_node(fdt)));
405     }
406 
407     g_free(modelname);
408 
409     _FDT((fdt_end_node(fdt)));
410 
411     /* RTAS */
412     _FDT((fdt_begin_node(fdt, "rtas")));
413 
414     _FDT((fdt_property(fdt, "ibm,hypertas-functions", hypertas_prop,
415                        sizeof(hypertas_prop))));
416     _FDT((fdt_property(fdt, "qemu,hypertas-functions", qemu_hypertas_prop,
417                        sizeof(qemu_hypertas_prop))));
418 
419     _FDT((fdt_property(fdt, "ibm,associativity-reference-points",
420         refpoints, sizeof(refpoints))));
421 
422     _FDT((fdt_property_cell(fdt, "rtas-error-log-max", RTAS_ERROR_LOG_MAX)));
423 
424     _FDT((fdt_end_node(fdt)));
425 
426     /* interrupt controller */
427     _FDT((fdt_begin_node(fdt, "interrupt-controller")));
428 
429     _FDT((fdt_property_string(fdt, "device_type",
430                               "PowerPC-External-Interrupt-Presentation")));
431     _FDT((fdt_property_string(fdt, "compatible", "IBM,ppc-xicp")));
432     _FDT((fdt_property(fdt, "interrupt-controller", NULL, 0)));
433     _FDT((fdt_property(fdt, "ibm,interrupt-server-ranges",
434                        interrupt_server_ranges_prop,
435                        sizeof(interrupt_server_ranges_prop))));
436     _FDT((fdt_property_cell(fdt, "#interrupt-cells", 2)));
437     _FDT((fdt_property_cell(fdt, "linux,phandle", PHANDLE_XICP)));
438     _FDT((fdt_property_cell(fdt, "phandle", PHANDLE_XICP)));
439 
440     _FDT((fdt_end_node(fdt)));
441 
442     /* vdevice */
443     _FDT((fdt_begin_node(fdt, "vdevice")));
444 
445     _FDT((fdt_property_string(fdt, "device_type", "vdevice")));
446     _FDT((fdt_property_string(fdt, "compatible", "IBM,vdevice")));
447     _FDT((fdt_property_cell(fdt, "#address-cells", 0x1)));
448     _FDT((fdt_property_cell(fdt, "#size-cells", 0x0)));
449     _FDT((fdt_property_cell(fdt, "#interrupt-cells", 0x2)));
450     _FDT((fdt_property(fdt, "interrupt-controller", NULL, 0)));
451 
452     _FDT((fdt_end_node(fdt)));
453 
454     /* event-sources */
455     spapr_events_fdt_skel(fdt, epow_irq);
456 
457     _FDT((fdt_end_node(fdt))); /* close root node */
458     _FDT((fdt_finish(fdt)));
459 
460     return fdt;
461 }
462 
463 static int spapr_populate_memory(sPAPREnvironment *spapr, void *fdt)
464 {
465     uint32_t associativity[] = {cpu_to_be32(0x4), cpu_to_be32(0x0),
466                                 cpu_to_be32(0x0), cpu_to_be32(0x0),
467                                 cpu_to_be32(0x0)};
468     char mem_name[32];
469     hwaddr node0_size, mem_start;
470     uint64_t mem_reg_property[2];
471     int i, off;
472 
473     /* memory node(s) */
474     node0_size = (nb_numa_nodes > 1) ? node_mem[0] : ram_size;
475     if (spapr->rma_size > node0_size) {
476         spapr->rma_size = node0_size;
477     }
478 
479     /* RMA */
480     mem_reg_property[0] = 0;
481     mem_reg_property[1] = cpu_to_be64(spapr->rma_size);
482     off = fdt_add_subnode(fdt, 0, "memory@0");
483     _FDT(off);
484     _FDT((fdt_setprop_string(fdt, off, "device_type", "memory")));
485     _FDT((fdt_setprop(fdt, off, "reg", mem_reg_property,
486                       sizeof(mem_reg_property))));
487     _FDT((fdt_setprop(fdt, off, "ibm,associativity", associativity,
488                       sizeof(associativity))));
489 
490     /* RAM: Node 0 */
491     if (node0_size > spapr->rma_size) {
492         mem_reg_property[0] = cpu_to_be64(spapr->rma_size);
493         mem_reg_property[1] = cpu_to_be64(node0_size - spapr->rma_size);
494 
495         sprintf(mem_name, "memory@" TARGET_FMT_lx, spapr->rma_size);
496         off = fdt_add_subnode(fdt, 0, mem_name);
497         _FDT(off);
498         _FDT((fdt_setprop_string(fdt, off, "device_type", "memory")));
499         _FDT((fdt_setprop(fdt, off, "reg", mem_reg_property,
500                           sizeof(mem_reg_property))));
501         _FDT((fdt_setprop(fdt, off, "ibm,associativity", associativity,
502                           sizeof(associativity))));
503     }
504 
505     /* RAM: Node 1 and beyond */
506     mem_start = node0_size;
507     for (i = 1; i < nb_numa_nodes; i++) {
508         mem_reg_property[0] = cpu_to_be64(mem_start);
509         mem_reg_property[1] = cpu_to_be64(node_mem[i]);
510         associativity[3] = associativity[4] = cpu_to_be32(i);
511         sprintf(mem_name, "memory@" TARGET_FMT_lx, mem_start);
512         off = fdt_add_subnode(fdt, 0, mem_name);
513         _FDT(off);
514         _FDT((fdt_setprop_string(fdt, off, "device_type", "memory")));
515         _FDT((fdt_setprop(fdt, off, "reg", mem_reg_property,
516                           sizeof(mem_reg_property))));
517         _FDT((fdt_setprop(fdt, off, "ibm,associativity", associativity,
518                           sizeof(associativity))));
519         mem_start += node_mem[i];
520     }
521 
522     return 0;
523 }
524 
525 static void spapr_finalize_fdt(sPAPREnvironment *spapr,
526                                hwaddr fdt_addr,
527                                hwaddr rtas_addr,
528                                hwaddr rtas_size)
529 {
530     int ret;
531     void *fdt;
532     sPAPRPHBState *phb;
533 
534     fdt = g_malloc(FDT_MAX_SIZE);
535 
536     /* open out the base tree into a temp buffer for the final tweaks */
537     _FDT((fdt_open_into(spapr->fdt_skel, fdt, FDT_MAX_SIZE)));
538 
539     ret = spapr_populate_memory(spapr, fdt);
540     if (ret < 0) {
541         fprintf(stderr, "couldn't setup memory nodes in fdt\n");
542         exit(1);
543     }
544 
545     ret = spapr_populate_vdevice(spapr->vio_bus, fdt);
546     if (ret < 0) {
547         fprintf(stderr, "couldn't setup vio devices in fdt\n");
548         exit(1);
549     }
550 
551     QLIST_FOREACH(phb, &spapr->phbs, list) {
552         ret = spapr_populate_pci_dt(phb, PHANDLE_XICP, fdt);
553     }
554 
555     if (ret < 0) {
556         fprintf(stderr, "couldn't setup PCI devices in fdt\n");
557         exit(1);
558     }
559 
560     /* RTAS */
561     ret = spapr_rtas_device_tree_setup(fdt, rtas_addr, rtas_size);
562     if (ret < 0) {
563         fprintf(stderr, "Couldn't set up RTAS device tree properties\n");
564     }
565 
566     /* Advertise NUMA via ibm,associativity */
567     ret = spapr_fixup_cpu_dt(fdt, spapr);
568     if (ret < 0) {
569         fprintf(stderr, "Couldn't finalize CPU device tree properties\n");
570     }
571 
572     if (!spapr->has_graphics) {
573         spapr_populate_chosen_stdout(fdt, spapr->vio_bus);
574     }
575 
576     _FDT((fdt_pack(fdt)));
577 
578     if (fdt_totalsize(fdt) > FDT_MAX_SIZE) {
579         hw_error("FDT too big ! 0x%x bytes (max is 0x%x)\n",
580                  fdt_totalsize(fdt), FDT_MAX_SIZE);
581         exit(1);
582     }
583 
584     cpu_physical_memory_write(fdt_addr, fdt, fdt_totalsize(fdt));
585 
586     g_free(fdt);
587 }
588 
589 static uint64_t translate_kernel_address(void *opaque, uint64_t addr)
590 {
591     return (addr & 0x0fffffff) + KERNEL_LOAD_ADDR;
592 }
593 
594 static void emulate_spapr_hypercall(PowerPCCPU *cpu)
595 {
596     CPUPPCState *env = &cpu->env;
597 
598     if (msr_pr) {
599         hcall_dprintf("Hypercall made with MSR[PR]=1\n");
600         env->gpr[3] = H_PRIVILEGE;
601     } else {
602         env->gpr[3] = spapr_hypercall(cpu, env->gpr[3], &env->gpr[4]);
603     }
604 }
605 
606 static void spapr_reset_htab(sPAPREnvironment *spapr)
607 {
608     long shift;
609 
610     /* allocate hash page table.  For now we always make this 16mb,
611      * later we should probably make it scale to the size of guest
612      * RAM */
613 
614     shift = kvmppc_reset_htab(spapr->htab_shift);
615 
616     if (shift > 0) {
617         /* Kernel handles htab, we don't need to allocate one */
618         spapr->htab_shift = shift;
619     } else {
620         if (!spapr->htab) {
621             /* Allocate an htab if we don't yet have one */
622             spapr->htab = qemu_memalign(HTAB_SIZE(spapr), HTAB_SIZE(spapr));
623         }
624 
625         /* And clear it */
626         memset(spapr->htab, 0, HTAB_SIZE(spapr));
627     }
628 
629     /* Update the RMA size if necessary */
630     if (spapr->vrma_adjust) {
631         spapr->rma_size = kvmppc_rma_size(ram_size, spapr->htab_shift);
632     }
633 }
634 
635 static void ppc_spapr_reset(void)
636 {
637     CPUState *first_cpu_cpu;
638 
639     /* Reset the hash table & recalc the RMA */
640     spapr_reset_htab(spapr);
641 
642     qemu_devices_reset();
643 
644     /* Load the fdt */
645     spapr_finalize_fdt(spapr, spapr->fdt_addr, spapr->rtas_addr,
646                        spapr->rtas_size);
647 
648     /* Set up the entry state */
649     first_cpu_cpu = ENV_GET_CPU(first_cpu);
650     first_cpu->gpr[3] = spapr->fdt_addr;
651     first_cpu->gpr[5] = 0;
652     first_cpu_cpu->halted = 0;
653     first_cpu->nip = spapr->entry_point;
654 
655 }
656 
657 static void spapr_cpu_reset(void *opaque)
658 {
659     PowerPCCPU *cpu = opaque;
660     CPUState *cs = CPU(cpu);
661     CPUPPCState *env = &cpu->env;
662 
663     cpu_reset(cs);
664 
665     /* All CPUs start halted.  CPU0 is unhalted from the machine level
666      * reset code and the rest are explicitly started up by the guest
667      * using an RTAS call */
668     cs->halted = 1;
669 
670     env->spr[SPR_HIOR] = 0;
671 
672     env->external_htab = spapr->htab;
673     env->htab_base = -1;
674     env->htab_mask = HTAB_SIZE(spapr) - 1;
675     env->spr[SPR_SDR1] = (unsigned long)spapr->htab |
676         (spapr->htab_shift - 18);
677 }
678 
679 static void spapr_create_nvram(sPAPREnvironment *spapr)
680 {
681     QemuOpts *machine_opts;
682     DeviceState *dev;
683 
684     dev = qdev_create(&spapr->vio_bus->bus, "spapr-nvram");
685 
686     machine_opts = qemu_opts_find(qemu_find_opts("machine"), 0);
687     if (machine_opts) {
688         const char *drivename;
689 
690         drivename = qemu_opt_get(machine_opts, "nvram");
691         if (drivename) {
692             BlockDriverState *bs;
693 
694             bs = bdrv_find(drivename);
695             if (!bs) {
696                 fprintf(stderr, "No such block device \"%s\" for nvram\n",
697                         drivename);
698                 exit(1);
699             }
700             qdev_prop_set_drive_nofail(dev, "drive", bs);
701         }
702     }
703 
704     qdev_init_nofail(dev);
705 
706     spapr->nvram = (struct sPAPRNVRAM *)dev;
707 }
708 
709 /* Returns whether we want to use VGA or not */
710 static int spapr_vga_init(PCIBus *pci_bus)
711 {
712     switch (vga_interface_type) {
713     case VGA_NONE:
714     case VGA_STD:
715         return pci_vga_init(pci_bus) != NULL;
716     default:
717         fprintf(stderr, "This vga model is not supported,"
718                 "currently it only supports -vga std\n");
719         exit(0);
720         break;
721     }
722 }
723 
724 /* pSeries LPAR / sPAPR hardware init */
725 static void ppc_spapr_init(QEMUMachineInitArgs *args)
726 {
727     ram_addr_t ram_size = args->ram_size;
728     const char *cpu_model = args->cpu_model;
729     const char *kernel_filename = args->kernel_filename;
730     const char *kernel_cmdline = args->kernel_cmdline;
731     const char *initrd_filename = args->initrd_filename;
732     const char *boot_device = args->boot_device;
733     PowerPCCPU *cpu;
734     CPUPPCState *env;
735     PCIHostState *phb;
736     int i;
737     MemoryRegion *sysmem = get_system_memory();
738     MemoryRegion *ram = g_new(MemoryRegion, 1);
739     hwaddr rma_alloc_size;
740     uint32_t initrd_base = 0;
741     long kernel_size = 0, initrd_size = 0;
742     long load_limit, rtas_limit, fw_size;
743     char *filename;
744 
745     msi_supported = true;
746 
747     spapr = g_malloc0(sizeof(*spapr));
748     QLIST_INIT(&spapr->phbs);
749 
750     cpu_ppc_hypercall = emulate_spapr_hypercall;
751 
752     /* Allocate RMA if necessary */
753     rma_alloc_size = kvmppc_alloc_rma("ppc_spapr.rma", sysmem);
754 
755     if (rma_alloc_size == -1) {
756         hw_error("qemu: Unable to create RMA\n");
757         exit(1);
758     }
759 
760     if (rma_alloc_size && (rma_alloc_size < ram_size)) {
761         spapr->rma_size = rma_alloc_size;
762     } else {
763         spapr->rma_size = ram_size;
764 
765         /* With KVM, we don't actually know whether KVM supports an
766          * unbounded RMA (PR KVM) or is limited by the hash table size
767          * (HV KVM using VRMA), so we always assume the latter
768          *
769          * In that case, we also limit the initial allocations for RTAS
770          * etc... to 256M since we have no way to know what the VRMA size
771          * is going to be as it depends on the size of the hash table
772          * isn't determined yet.
773          */
774         if (kvm_enabled()) {
775             spapr->vrma_adjust = 1;
776             spapr->rma_size = MIN(spapr->rma_size, 0x10000000);
777         }
778     }
779 
780     /* We place the device tree and RTAS just below either the top of the RMA,
781      * or just below 2GB, whichever is lowere, so that it can be
782      * processed with 32-bit real mode code if necessary */
783     rtas_limit = MIN(spapr->rma_size, 0x80000000);
784     spapr->rtas_addr = rtas_limit - RTAS_MAX_SIZE;
785     spapr->fdt_addr = spapr->rtas_addr - FDT_MAX_SIZE;
786     load_limit = spapr->fdt_addr - FW_OVERHEAD;
787 
788     /* We aim for a hash table of size 1/128 the size of RAM.  The
789      * normal rule of thumb is 1/64 the size of RAM, but that's much
790      * more than needed for the Linux guests we support. */
791     spapr->htab_shift = 18; /* Minimum architected size */
792     while (spapr->htab_shift <= 46) {
793         if ((1ULL << (spapr->htab_shift + 7)) >= ram_size) {
794             break;
795         }
796         spapr->htab_shift++;
797     }
798 
799     /* Set up Interrupt Controller before we create the VCPUs */
800     spapr->icp = xics_system_init(smp_cpus * kvmppc_smt_threads() / smp_threads,
801                                   XICS_IRQS);
802     spapr->next_irq = XICS_IRQ_BASE;
803 
804     /* init CPUs */
805     if (cpu_model == NULL) {
806         cpu_model = kvm_enabled() ? "host" : "POWER7";
807     }
808     for (i = 0; i < smp_cpus; i++) {
809         cpu = cpu_ppc_init(cpu_model);
810         if (cpu == NULL) {
811             fprintf(stderr, "Unable to find PowerPC CPU definition\n");
812             exit(1);
813         }
814         env = &cpu->env;
815 
816         xics_cpu_setup(spapr->icp, cpu);
817 
818         /* Set time-base frequency to 512 MHz */
819         cpu_ppc_tb_init(env, TIMEBASE_FREQ);
820 
821         /* PAPR always has exception vectors in RAM not ROM. To ensure this,
822          * MSR[IP] should never be set.
823          */
824         env->msr_mask &= ~(1 << 6);
825 
826         /* Tell KVM that we're in PAPR mode */
827         if (kvm_enabled()) {
828             kvmppc_set_papr(cpu);
829         }
830 
831         qemu_register_reset(spapr_cpu_reset, cpu);
832     }
833 
834     /* allocate RAM */
835     spapr->ram_limit = ram_size;
836     if (spapr->ram_limit > rma_alloc_size) {
837         ram_addr_t nonrma_base = rma_alloc_size;
838         ram_addr_t nonrma_size = spapr->ram_limit - rma_alloc_size;
839 
840         memory_region_init_ram(ram, "ppc_spapr.ram", nonrma_size);
841         vmstate_register_ram_global(ram);
842         memory_region_add_subregion(sysmem, nonrma_base, ram);
843     }
844 
845     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, "spapr-rtas.bin");
846     spapr->rtas_size = load_image_targphys(filename, spapr->rtas_addr,
847                                            rtas_limit - spapr->rtas_addr);
848     if (spapr->rtas_size < 0) {
849         hw_error("qemu: could not load LPAR rtas '%s'\n", filename);
850         exit(1);
851     }
852     if (spapr->rtas_size > RTAS_MAX_SIZE) {
853         hw_error("RTAS too big ! 0x%lx bytes (max is 0x%x)\n",
854                  spapr->rtas_size, RTAS_MAX_SIZE);
855         exit(1);
856     }
857     g_free(filename);
858 
859     /* Set up EPOW events infrastructure */
860     spapr_events_init(spapr);
861 
862     /* Set up IOMMU */
863     spapr_iommu_init();
864 
865     /* Set up VIO bus */
866     spapr->vio_bus = spapr_vio_bus_init();
867 
868     for (i = 0; i < MAX_SERIAL_PORTS; i++) {
869         if (serial_hds[i]) {
870             spapr_vty_create(spapr->vio_bus, serial_hds[i]);
871         }
872     }
873 
874     /* We always have at least the nvram device on VIO */
875     spapr_create_nvram(spapr);
876 
877     /* Set up PCI */
878     spapr_pci_rtas_init();
879 
880     phb = spapr_create_phb(spapr, 0);
881 
882     for (i = 0; i < nb_nics; i++) {
883         NICInfo *nd = &nd_table[i];
884 
885         if (!nd->model) {
886             nd->model = g_strdup("ibmveth");
887         }
888 
889         if (strcmp(nd->model, "ibmveth") == 0) {
890             spapr_vlan_create(spapr->vio_bus, nd);
891         } else {
892             pci_nic_init_nofail(&nd_table[i], nd->model, NULL);
893         }
894     }
895 
896     for (i = 0; i <= drive_get_max_bus(IF_SCSI); i++) {
897         spapr_vscsi_create(spapr->vio_bus);
898     }
899 
900     /* Graphics */
901     if (spapr_vga_init(phb->bus)) {
902         spapr->has_graphics = true;
903     }
904 
905     if (usb_enabled(spapr->has_graphics)) {
906         pci_create_simple(phb->bus, -1, "pci-ohci");
907         if (spapr->has_graphics) {
908             usbdevice_create("keyboard");
909             usbdevice_create("mouse");
910         }
911     }
912 
913     if (spapr->rma_size < (MIN_RMA_SLOF << 20)) {
914         fprintf(stderr, "qemu: pSeries SLOF firmware requires >= "
915                 "%ldM guest RMA (Real Mode Area memory)\n", MIN_RMA_SLOF);
916         exit(1);
917     }
918 
919     if (kernel_filename) {
920         uint64_t lowaddr = 0;
921 
922         kernel_size = load_elf(kernel_filename, translate_kernel_address, NULL,
923                                NULL, &lowaddr, NULL, 1, ELF_MACHINE, 0);
924         if (kernel_size < 0) {
925             kernel_size = load_image_targphys(kernel_filename,
926                                               KERNEL_LOAD_ADDR,
927                                               load_limit - KERNEL_LOAD_ADDR);
928         }
929         if (kernel_size < 0) {
930             fprintf(stderr, "qemu: could not load kernel '%s'\n",
931                     kernel_filename);
932             exit(1);
933         }
934 
935         /* load initrd */
936         if (initrd_filename) {
937             /* Try to locate the initrd in the gap between the kernel
938              * and the firmware. Add a bit of space just in case
939              */
940             initrd_base = (KERNEL_LOAD_ADDR + kernel_size + 0x1ffff) & ~0xffff;
941             initrd_size = load_image_targphys(initrd_filename, initrd_base,
942                                               load_limit - initrd_base);
943             if (initrd_size < 0) {
944                 fprintf(stderr, "qemu: could not load initial ram disk '%s'\n",
945                         initrd_filename);
946                 exit(1);
947             }
948         } else {
949             initrd_base = 0;
950             initrd_size = 0;
951         }
952     }
953 
954     filename = qemu_find_file(QEMU_FILE_TYPE_BIOS, FW_FILE_NAME);
955     fw_size = load_image_targphys(filename, 0, FW_MAX_SIZE);
956     if (fw_size < 0) {
957         hw_error("qemu: could not load LPAR rtas '%s'\n", filename);
958         exit(1);
959     }
960     g_free(filename);
961 
962     spapr->entry_point = 0x100;
963 
964     /* Prepare the device tree */
965     spapr->fdt_skel = spapr_create_fdt_skel(cpu_model,
966                                             initrd_base, initrd_size,
967                                             kernel_size,
968                                             boot_device, kernel_cmdline,
969                                             spapr->epow_irq);
970     assert(spapr->fdt_skel != NULL);
971 }
972 
973 static QEMUMachine spapr_machine = {
974     .name = "pseries",
975     .desc = "pSeries Logical Partition (PAPR compliant)",
976     .init = ppc_spapr_init,
977     .reset = ppc_spapr_reset,
978     .block_default_type = IF_SCSI,
979     .max_cpus = MAX_CPUS,
980     .no_parallel = 1,
981     .boot_order = NULL,
982 };
983 
984 static void spapr_machine_init(void)
985 {
986     qemu_register_machine(&spapr_machine);
987 }
988 
989 machine_init(spapr_machine_init);
990