ia64/xen-unstable

view xen/arch/ia64/xen/xensetup.c @ 14037:9364bea18bc4

[IA64] First step to support big-endian PV domUs

Signed-off-by: Dietmar Hahn <dietmar.hahn@fujitsu-siemens.com>
author awilliam@xenbuild2.aw
date Thu Feb 22 09:42:13 2007 -0700 (2007-02-22)
parents 6c63ff548888
children 215b799fa181
line source
1 /******************************************************************************
2 * xensetup.c
3 * Copyright (c) 2004-2005 Hewlett-Packard Co
4 * Dan Magenheimer <dan.magenheimer@hp.com>
5 */
7 #include <xen/config.h>
8 #include <xen/lib.h>
9 #include <xen/errno.h>
10 //#include <xen/spinlock.h>
11 #include <xen/multiboot.h>
12 #include <xen/sched.h>
13 #include <xen/mm.h>
14 #include <public/version.h>
15 #include <xen/gdbstub.h>
16 #include <xen/version.h>
17 #include <xen/console.h>
18 #include <xen/domain.h>
19 #include <xen/serial.h>
20 #include <xen/trace.h>
21 #include <xen/keyhandler.h>
22 #include <asm/meminit.h>
23 #include <asm/page.h>
24 #include <asm/setup.h>
25 #include <xen/string.h>
26 #include <asm/vmx.h>
27 #include <linux/efi.h>
28 #include <asm/iosapic.h>
29 #include <xen/softirq.h>
30 #include <xen/rcupdate.h>
32 unsigned long xenheap_phys_end, total_pages;
34 char saved_command_line[COMMAND_LINE_SIZE];
35 char dom0_command_line[COMMAND_LINE_SIZE];
37 cpumask_t cpu_present_map;
39 extern unsigned long domain0_ready;
41 int find_max_pfn (unsigned long, unsigned long, void *);
43 /* FIXME: which header these declarations should be there ? */
44 extern long is_platform_hp_ski(void);
45 extern void early_setup_arch(char **);
46 extern void late_setup_arch(char **);
47 extern void hpsim_serial_init(void);
48 extern void alloc_dom0(void);
49 extern void setup_per_cpu_areas(void);
50 extern void mem_init(void);
51 extern void init_IRQ(void);
52 extern void trap_init(void);
53 extern void xen_patch_kernel(void);
55 /* opt_nosmp: If true, secondary processors are ignored. */
56 static int opt_nosmp;
57 boolean_param("nosmp", opt_nosmp);
59 /* maxcpus: maximum number of CPUs to activate. */
60 static unsigned int max_cpus = NR_CPUS;
61 integer_param("maxcpus", max_cpus);
63 /* xencons: if true enable xenconsole input (and irq).
64 Note: you have to disable 8250 serials in domains (to avoid use of the
65 same resource). */
66 static int opt_xencons = 1;
67 integer_param("xencons", opt_xencons);
69 /* Toggle to allow non-legacy xencons UARTs to run in polling mode */
70 static int opt_xencons_poll;
71 boolean_param("xencons_poll", opt_xencons_poll);
73 /*
74 * opt_xenheap_megabytes: Size of Xen heap in megabytes, including:
75 * xen image
76 * bootmap bits
77 * xen heap
78 * Note: To allow xenheap size configurable, the prerequisite is
79 * to configure elilo allowing relocation defaultly. Then since
80 * elilo chooses 256M as alignment when relocating, alignment issue
81 * on IPF can be addressed.
82 */
83 unsigned int opt_xenheap_megabytes = XENHEAP_DEFAULT_MB;
84 unsigned long xenheap_size = XENHEAP_DEFAULT_SIZE;
85 extern long running_on_sim;
86 unsigned long xen_pstart;
87 void *xen_heap_start __read_mostly;
89 static int
90 xen_count_pages(u64 start, u64 end, void *arg)
91 {
92 unsigned long *count = arg;
94 /* FIXME: do we need consider difference between DMA-usable memory and
95 * normal memory? Seems that HV has no requirement to operate DMA which
96 * is owned by Dom0? */
97 *count += (end - start) >> PAGE_SHIFT;
98 return 0;
99 }
101 static void __init do_initcalls(void)
102 {
103 initcall_t *call;
104 for ( call = &__initcall_start; call < &__initcall_end; call++ )
105 (*call)();
106 }
108 /*
109 * IPF loader only supports one commaind line currently, for
110 * both xen and guest kernel. This function provides pre-parse
111 * to mixed command line, to split it into two parts.
112 *
113 * User should split the parameters by "--", with strings after
114 * spliter for guest kernel. Missing "--" means whole line belongs
115 * to guest. Example:
116 * "com2=57600,8n1 console=com2 -- console=ttyS1 console=tty
117 * root=/dev/sda3 ro"
118 */
119 static char null[4] = { 0 };
121 void early_cmdline_parse(char **cmdline_p)
122 {
123 char *guest_cmd;
124 static const char * const split = "--";
126 if (*cmdline_p == NULL) {
127 *cmdline_p = &null[0];
128 saved_command_line[0] = '\0';
129 dom0_command_line[0] = '\0';
130 return;
131 }
133 guest_cmd = strstr(*cmdline_p, split);
134 /* If no spliter, whole line is for guest */
135 if (guest_cmd == NULL) {
136 guest_cmd = *cmdline_p;
137 *cmdline_p = &null[0];
138 } else {
139 *guest_cmd = '\0'; /* Split boot parameters for xen and guest */
140 guest_cmd += strlen(split);
141 while (*guest_cmd == ' ') guest_cmd++;
142 }
144 strlcpy(saved_command_line, *cmdline_p, COMMAND_LINE_SIZE);
145 strlcpy(dom0_command_line, guest_cmd, COMMAND_LINE_SIZE);
146 return;
147 }
149 struct ns16550_defaults ns16550_com1 = {
150 .data_bits = 8,
151 .parity = 'n',
152 .stop_bits = 1
153 };
155 unsigned int ns16550_com1_gsi;
156 unsigned int ns16550_com1_polarity;
157 unsigned int ns16550_com1_trigger;
159 struct ns16550_defaults ns16550_com2 = {
160 .data_bits = 8,
161 .parity = 'n',
162 .stop_bits = 1
163 };
165 /* efi_print: print efi table at boot */
166 static int opt_efi_print;
167 boolean_param("efi_print", opt_efi_print);
169 /* print EFI memory map: */
170 static void
171 efi_print(void)
172 {
173 void *efi_map_start, *efi_map_end;
174 u64 efi_desc_size;
176 efi_memory_desc_t *md;
177 void *p;
178 int i;
180 if (!opt_efi_print)
181 return;
183 efi_map_start = __va(ia64_boot_param->efi_memmap);
184 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
185 efi_desc_size = ia64_boot_param->efi_memdesc_size;
187 for (i = 0, p = efi_map_start; p < efi_map_end; ++i, p += efi_desc_size) {
188 md = p;
189 printk("mem%02u: type=%2u, attr=0x%016lx, range=[0x%016lx-0x%016lx) "
190 "(%luMB)\n", i, md->type, md->attribute, md->phys_addr,
191 md->phys_addr + (md->num_pages << EFI_PAGE_SHIFT),
192 md->num_pages >> (20 - EFI_PAGE_SHIFT));
193 }
194 }
196 /*
197 * These functions are utility functions for getting and
198 * testing memory descriptors for allocating the xenheap area.
199 */
200 static efi_memory_desc_t *
201 efi_get_md (unsigned long phys_addr)
202 {
203 void *efi_map_start, *efi_map_end, *p;
204 efi_memory_desc_t *md;
205 u64 efi_desc_size;
207 efi_map_start = __va(ia64_boot_param->efi_memmap);
208 efi_map_end = efi_map_start + ia64_boot_param->efi_memmap_size;
209 efi_desc_size = ia64_boot_param->efi_memdesc_size;
211 for (p = efi_map_start; p < efi_map_end; p += efi_desc_size) {
212 md = p;
213 if (phys_addr - md->phys_addr < (md->num_pages << EFI_PAGE_SHIFT))
214 return md;
215 }
216 return 0;
217 }
219 static int
220 is_xenheap_usable_memory(efi_memory_desc_t *md)
221 {
222 if (!(md->attribute & EFI_MEMORY_WB))
223 return 0;
225 switch (md->type) {
226 case EFI_LOADER_CODE:
227 case EFI_LOADER_DATA:
228 case EFI_BOOT_SERVICES_CODE:
229 case EFI_BOOT_SERVICES_DATA:
230 case EFI_CONVENTIONAL_MEMORY:
231 return 1;
232 }
233 return 0;
234 }
236 static inline int
237 md_overlaps(efi_memory_desc_t *md, unsigned long phys_addr)
238 {
239 return (phys_addr - md->phys_addr < (md->num_pages << EFI_PAGE_SHIFT));
240 }
242 #define MD_SIZE(md) (md->num_pages << EFI_PAGE_SHIFT)
244 void start_kernel(void)
245 {
246 char *cmdline;
247 unsigned long nr_pages;
248 unsigned long dom0_memory_start, dom0_memory_size;
249 unsigned long dom0_initrd_start, dom0_initrd_size;
250 unsigned long md_end, relo_start, relo_end, relo_size = 0;
251 struct domain *idle_domain;
252 struct vcpu *dom0_vcpu0;
253 efi_memory_desc_t *kern_md, *last_md, *md;
254 #ifdef CONFIG_SMP
255 int i;
256 #endif
258 /* Be sure the struct shared_info size is <= XSI_SIZE. */
259 BUILD_BUG_ON(sizeof(struct shared_info) > XSI_SIZE);
261 running_on_sim = is_platform_hp_ski();
262 /* Kernel may be relocated by EFI loader */
263 xen_pstart = ia64_tpa(KERNEL_START);
265 early_setup_arch(&cmdline);
267 /* We initialise the serial devices very early so we can get debugging. */
268 if (running_on_sim)
269 hpsim_serial_init();
270 else {
271 ns16550_init(0, &ns16550_com1);
272 ns16550_init(1, &ns16550_com2);
273 }
274 serial_init_preirq();
276 init_console();
277 set_printk_prefix("(XEN) ");
279 if (running_on_sim || ia64_boot_param->domain_start == 0 ||
280 ia64_boot_param->domain_size == 0) {
281 /* This is possible only with the old elilo, which does not support
282 a vmm. Fix now, and continue without initrd. */
283 printk ("Your elilo is not Xen-aware. Bootparams fixed\n");
284 ia64_boot_param->domain_start = ia64_boot_param->initrd_start;
285 ia64_boot_param->domain_size = ia64_boot_param->initrd_size;
286 ia64_boot_param->initrd_start = 0;
287 ia64_boot_param->initrd_size = 0;
288 }
290 printk("Xen command line: %s\n", saved_command_line);
291 /* xenheap should be in same TR-covered range with xen image */
292 xenheap_phys_end = xen_pstart + xenheap_size;
293 printk("xen image pstart: 0x%lx, xenheap pend: 0x%lx\n",
294 xen_pstart, xenheap_phys_end);
296 xen_patch_kernel();
298 kern_md = md = efi_get_md(xen_pstart);
299 md_end = __pa(ia64_imva(&_end));
300 relo_start = xenheap_phys_end;
302 /*
303 * Scan through the memory descriptors after the kernel
304 * image to make sure we have enough room for the xenheap
305 * area, pushing out whatever may already be there.
306 */
307 while (relo_start + relo_size >= md_end) {
308 md = efi_get_md(md_end);
310 BUG_ON(!md);
311 BUG_ON(!is_xenheap_usable_memory(md));
313 md_end = md->phys_addr + MD_SIZE(md);
314 /*
315 * The dom0 kernel or initrd could overlap, reserve space
316 * at the end to relocate them later.
317 */
318 if (md->type == EFI_LOADER_DATA) {
319 /* Test for ranges we're not prepared to move */
320 BUG_ON(md_overlaps(md, __pa(ia64_boot_param)) ||
321 md_overlaps(md, ia64_boot_param->efi_memmap) ||
322 md_overlaps(md, ia64_boot_param->command_line));
324 relo_size += MD_SIZE(md);
325 /* If range overlaps the end, push out the relocation start */
326 if (md_end > relo_start)
327 relo_start = md_end;
328 }
329 }
330 last_md = md;
331 relo_end = relo_start + relo_size;
333 md_end = __pa(ia64_imva(&_end));
335 /*
336 * Move any relocated data out into the previously found relocation
337 * area. Any extra memory descriptrs are moved out to the end
338 * and set to zero pages.
339 */
340 for (md = efi_get_md(md_end) ;; md = efi_get_md(md_end)) {
341 md_end = md->phys_addr + MD_SIZE(md);
343 if (md->type == EFI_LOADER_DATA) {
344 unsigned long relo_offset;
346 if (md_overlaps(md, ia64_boot_param->domain_start)) {
347 relo_offset = ia64_boot_param->domain_start - md->phys_addr;
348 printk("Moving Dom0 kernel image: 0x%lx -> 0x%lx (%ld KiB)\n",
349 ia64_boot_param->domain_start, relo_start + relo_offset,
350 ia64_boot_param->domain_size >> 10);
351 ia64_boot_param->domain_start = relo_start + relo_offset;
352 }
353 if (ia64_boot_param->initrd_size &&
354 md_overlaps(md, ia64_boot_param->initrd_start)) {
355 relo_offset = ia64_boot_param->initrd_start - md->phys_addr;
356 printk("Moving Dom0 initrd image: 0x%lx -> 0x%lx (%ld KiB)\n",
357 ia64_boot_param->initrd_start, relo_start + relo_offset,
358 ia64_boot_param->initrd_size >> 10);
359 ia64_boot_param->initrd_start = relo_start + relo_offset;
360 }
361 memcpy(__va(relo_start), __va(md->phys_addr), MD_SIZE(md));
362 relo_start += MD_SIZE(md);
363 }
365 if (md == kern_md)
366 continue;
367 if (md == last_md)
368 break;
370 md->phys_addr = relo_end;
371 md->num_pages = 0;
372 }
374 /* Trim the last entry */
375 md->phys_addr = relo_end;
376 md->num_pages = (md_end - relo_end) >> EFI_PAGE_SHIFT;
378 /*
379 * Expand the new kernel/xenheap (and maybe dom0/initrd) out to
380 * the full size. This range will already be type EFI_LOADER_DATA,
381 * therefore the xenheap area is now protected being allocated for
382 * use by find_memmap_space() in efi.c
383 */
384 kern_md->num_pages = (relo_end - kern_md->phys_addr) >> EFI_PAGE_SHIFT;
386 reserve_memory();
388 /* first find highest page frame number */
389 max_page = 0;
390 efi_memmap_walk(find_max_pfn, &max_page);
391 printk("find_memory: efi_memmap_walk returns max_page=%lx\n",max_page);
392 efi_print();
394 xen_heap_start = memguard_init(ia64_imva(&_end));
395 printk("Before xen_heap_start: %p\n", xen_heap_start);
396 xen_heap_start = __va(init_boot_allocator(__pa(xen_heap_start)));
397 printk("After xen_heap_start: %p\n", xen_heap_start);
399 efi_memmap_walk(filter_rsvd_memory, init_boot_pages);
400 efi_memmap_walk(xen_count_pages, &nr_pages);
402 printk("System RAM: %luMB (%lukB)\n",
403 nr_pages >> (20 - PAGE_SHIFT),
404 nr_pages << (PAGE_SHIFT - 10));
405 total_pages = nr_pages;
407 init_frametable();
409 trap_init();
411 alloc_dom0();
413 end_boot_allocator();
415 init_xenheap_pages(__pa(xen_heap_start), xenheap_phys_end);
416 printk("Xen heap: %luMB (%lukB)\n",
417 (xenheap_phys_end-__pa(xen_heap_start)) >> 20,
418 (xenheap_phys_end-__pa(xen_heap_start)) >> 10);
420 late_setup_arch(&cmdline);
422 scheduler_init();
423 idle_vcpu[0] = (struct vcpu*) ia64_r13;
424 idle_domain = domain_create(IDLE_DOMAIN_ID, 0);
425 if ( (idle_domain == NULL) || (alloc_vcpu(idle_domain, 0, 0) == NULL) )
426 BUG();
428 alloc_dom_xen_and_dom_io();
429 setup_per_cpu_areas();
430 mem_init();
432 local_irq_disable();
433 init_IRQ ();
434 init_xen_time(); /* initialise the time */
435 timer_init();
437 rcu_init();
439 #ifdef CONFIG_XEN_IA64_TLBFLUSH_CLOCK
440 open_softirq(NEW_TLBFLUSH_CLOCK_PERIOD_SOFTIRQ, new_tlbflush_clock_period);
441 #endif
443 #ifdef CONFIG_SMP
444 if ( opt_nosmp )
445 {
446 max_cpus = 0;
447 smp_num_siblings = 1;
448 //boot_cpu_data.x86_num_cores = 1;
449 }
451 /* A vcpu is created for the idle domain on every physical cpu.
452 Limit the number of cpus to the maximum number of vcpus. */
453 if (max_cpus > MAX_VIRT_CPUS)
454 max_cpus = MAX_VIRT_CPUS;
456 smp_prepare_cpus(max_cpus);
458 /* We aren't hotplug-capable yet. */
459 for_each_cpu ( i )
460 cpu_set(i, cpu_present_map);
462 /* Enable IRQ to receive IPI (needed for ITC sync). */
463 local_irq_enable();
465 printk("num_online_cpus=%d, max_cpus=%d\n",num_online_cpus(),max_cpus);
466 for_each_present_cpu ( i )
467 {
468 if ( num_online_cpus() >= max_cpus )
469 break;
470 if ( !cpu_online(i) ) {
471 rcu_online_cpu(i);
472 __cpu_up(i);
473 }
474 }
476 local_irq_disable();
478 printk("Brought up %ld CPUs\n", (long)num_online_cpus());
479 smp_cpus_done(max_cpus);
480 #endif
482 initialise_gdb(); /* could be moved earlier */
484 do_initcalls();
485 sort_main_extable();
487 init_rid_allocator ();
489 local_irq_enable();
491 if (opt_xencons) {
492 initialize_keytable();
493 if (ns16550_com1_gsi) {
494 if (opt_xencons_poll ||
495 iosapic_register_intr(ns16550_com1_gsi,
496 ns16550_com1_polarity,
497 ns16550_com1_trigger) < 0) {
498 ns16550_com1.irq = 0;
499 ns16550_init(0, &ns16550_com1);
500 }
501 }
502 serial_init_postirq();
504 /* Hide the HCDP table from dom0 */
505 efi.hcdp = NULL;
506 }
508 expose_p2m_init();
510 /* Create initial domain 0. */
511 dom0 = domain_create(0, 0);
512 if (dom0 == NULL)
513 panic("Error creating domain 0\n");
514 dom0_vcpu0 = alloc_vcpu(dom0, 0, 0);
515 if (dom0_vcpu0 == NULL || vcpu_late_initialise(dom0_vcpu0) != 0)
516 panic("Cannot allocate dom0 vcpu 0\n");
518 dom0->is_privileged = 1;
520 /*
521 * We're going to setup domain0 using the module(s) that we stashed safely
522 * above our heap. The second module, if present, is an initrd ramdisk.
523 */
524 dom0_memory_start = (unsigned long) __va(ia64_boot_param->domain_start);
525 dom0_memory_size = ia64_boot_param->domain_size;
526 dom0_initrd_start = (unsigned long) __va(ia64_boot_param->initrd_start);
527 dom0_initrd_size = ia64_boot_param->initrd_size;
529 if ( construct_dom0(dom0, dom0_memory_start, dom0_memory_size,
530 dom0_initrd_start,dom0_initrd_size,
531 0) != 0)
532 panic("Could not set up DOM0 guest OS\n");
534 if (!running_on_sim) // slow on ski and pages are pre-initialized to zero
535 scrub_heap_pages();
537 init_trace_bufs();
539 if (opt_xencons) {
540 console_endboot();
541 serial_endboot();
542 }
544 domain0_ready = 1;
546 schedulers_start();
548 domain_unpause_by_systemcontroller(dom0);
550 startup_cpu_idle_loop();
551 }
553 void arch_get_xen_caps(xen_capabilities_info_t *info)
554 {
555 int major = xen_major_version();
556 int minor = xen_minor_version();
557 char s[32];
559 (*info)[0] = '\0';
561 snprintf(s, sizeof(s), "xen-%d.%d-ia64 ", major, minor);
562 safe_strcat(*info, s);
564 snprintf(s, sizeof(s), "xen-%d.%d-ia64be ", major, minor);
565 safe_strcat(*info, s);
567 if (vmx_enabled)
568 {
569 snprintf(s, sizeof(s), "hvm-%d.%d-ia64 ", major, minor);
570 safe_strcat(*info, s);
571 }
572 }