xref: /openbmc/linux/arch/ia64/kernel/setup.c (revision 8a3360f0)
1 /*
2  * Architecture-specific setup.
3  *
4  * Copyright (C) 1998-2001, 2003-2004 Hewlett-Packard Co
5  *	David Mosberger-Tang <davidm@hpl.hp.com>
6  *	Stephane Eranian <eranian@hpl.hp.com>
7  * Copyright (C) 2000, 2004 Intel Corp
8  * 	Rohit Seth <rohit.seth@intel.com>
9  * 	Suresh Siddha <suresh.b.siddha@intel.com>
10  * 	Gordon Jin <gordon.jin@intel.com>
11  * Copyright (C) 1999 VA Linux Systems
12  * Copyright (C) 1999 Walt Drummond <drummond@valinux.com>
13  *
14  * 12/26/04 S.Siddha, G.Jin, R.Seth
15  *			Add multi-threading and multi-core detection
16  * 11/12/01 D.Mosberger Convert get_cpuinfo() to seq_file based show_cpuinfo().
17  * 04/04/00 D.Mosberger renamed cpu_initialized to cpu_online_map
18  * 03/31/00 R.Seth	cpu_initialized and current->processor fixes
19  * 02/04/00 D.Mosberger	some more get_cpuinfo fixes...
20  * 02/01/00 R.Seth	fixed get_cpuinfo for SMP
21  * 01/07/99 S.Eranian	added the support for command line argument
22  * 06/24/99 W.Drummond	added boot_cpu_data.
23  * 05/28/05 Z. Menyhart	Dynamic stride size for "flush_icache_range()"
24  */
25 #include <linux/module.h>
26 #include <linux/init.h>
27 
28 #include <linux/acpi.h>
29 #include <linux/bootmem.h>
30 #include <linux/console.h>
31 #include <linux/delay.h>
32 #include <linux/kernel.h>
33 #include <linux/reboot.h>
34 #include <linux/sched.h>
35 #include <linux/seq_file.h>
36 #include <linux/string.h>
37 #include <linux/threads.h>
38 #include <linux/screen_info.h>
39 #include <linux/dmi.h>
40 #include <linux/serial.h>
41 #include <linux/serial_core.h>
42 #include <linux/efi.h>
43 #include <linux/initrd.h>
44 #include <linux/pm.h>
45 #include <linux/cpufreq.h>
46 #include <linux/kexec.h>
47 #include <linux/crash_dump.h>
48 
49 #include <asm/ia32.h>
50 #include <asm/machvec.h>
51 #include <asm/mca.h>
52 #include <asm/meminit.h>
53 #include <asm/page.h>
54 #include <asm/patch.h>
55 #include <asm/pgtable.h>
56 #include <asm/processor.h>
57 #include <asm/sal.h>
58 #include <asm/sections.h>
59 #include <asm/setup.h>
60 #include <asm/smp.h>
61 #include <asm/system.h>
62 #include <asm/tlbflush.h>
63 #include <asm/unistd.h>
64 #include <asm/hpsim.h>
65 
66 #if defined(CONFIG_SMP) && (IA64_CPU_SIZE > PAGE_SIZE)
67 # error "struct cpuinfo_ia64 too big!"
68 #endif
69 
70 #ifdef CONFIG_SMP
71 unsigned long __per_cpu_offset[NR_CPUS];
72 EXPORT_SYMBOL(__per_cpu_offset);
73 #endif
74 
75 DEFINE_PER_CPU(struct cpuinfo_ia64, cpu_info);
76 DEFINE_PER_CPU(unsigned long, local_per_cpu_offset);
77 unsigned long ia64_cycles_per_usec;
78 struct ia64_boot_param *ia64_boot_param;
79 struct screen_info screen_info;
80 unsigned long vga_console_iobase;
81 unsigned long vga_console_membase;
82 
83 static struct resource data_resource = {
84 	.name	= "Kernel data",
85 	.flags	= IORESOURCE_BUSY | IORESOURCE_MEM
86 };
87 
88 static struct resource code_resource = {
89 	.name	= "Kernel code",
90 	.flags	= IORESOURCE_BUSY | IORESOURCE_MEM
91 };
92 
93 static struct resource bss_resource = {
94 	.name	= "Kernel bss",
95 	.flags	= IORESOURCE_BUSY | IORESOURCE_MEM
96 };
97 
98 unsigned long ia64_max_cacheline_size;
99 
100 int dma_get_cache_alignment(void)
101 {
102         return ia64_max_cacheline_size;
103 }
104 EXPORT_SYMBOL(dma_get_cache_alignment);
105 
106 unsigned long ia64_iobase;	/* virtual address for I/O accesses */
107 EXPORT_SYMBOL(ia64_iobase);
108 struct io_space io_space[MAX_IO_SPACES];
109 EXPORT_SYMBOL(io_space);
110 unsigned int num_io_spaces;
111 
112 /*
113  * "flush_icache_range()" needs to know what processor dependent stride size to use
114  * when it makes i-cache(s) coherent with d-caches.
115  */
116 #define	I_CACHE_STRIDE_SHIFT	5	/* Safest way to go: 32 bytes by 32 bytes */
117 unsigned long ia64_i_cache_stride_shift = ~0;
118 
119 /*
120  * The merge_mask variable needs to be set to (max(iommu_page_size(iommu)) - 1).  This
121  * mask specifies a mask of address bits that must be 0 in order for two buffers to be
122  * mergeable by the I/O MMU (i.e., the end address of the first buffer and the start
123  * address of the second buffer must be aligned to (merge_mask+1) in order to be
124  * mergeable).  By default, we assume there is no I/O MMU which can merge physically
125  * discontiguous buffers, so we set the merge_mask to ~0UL, which corresponds to a iommu
126  * page-size of 2^64.
127  */
128 unsigned long ia64_max_iommu_merge_mask = ~0UL;
129 EXPORT_SYMBOL(ia64_max_iommu_merge_mask);
130 
131 /*
132  * We use a special marker for the end of memory and it uses the extra (+1) slot
133  */
134 struct rsvd_region rsvd_region[IA64_MAX_RSVD_REGIONS + 1] __initdata;
135 int num_rsvd_regions __initdata;
136 
137 
138 /*
139  * Filter incoming memory segments based on the primitive map created from the boot
140  * parameters. Segments contained in the map are removed from the memory ranges. A
141  * caller-specified function is called with the memory ranges that remain after filtering.
142  * This routine does not assume the incoming segments are sorted.
143  */
144 int __init
145 filter_rsvd_memory (unsigned long start, unsigned long end, void *arg)
146 {
147 	unsigned long range_start, range_end, prev_start;
148 	void (*func)(unsigned long, unsigned long, int);
149 	int i;
150 
151 #if IGNORE_PFN0
152 	if (start == PAGE_OFFSET) {
153 		printk(KERN_WARNING "warning: skipping physical page 0\n");
154 		start += PAGE_SIZE;
155 		if (start >= end) return 0;
156 	}
157 #endif
158 	/*
159 	 * lowest possible address(walker uses virtual)
160 	 */
161 	prev_start = PAGE_OFFSET;
162 	func = arg;
163 
164 	for (i = 0; i < num_rsvd_regions; ++i) {
165 		range_start = max(start, prev_start);
166 		range_end   = min(end, rsvd_region[i].start);
167 
168 		if (range_start < range_end)
169 			call_pernode_memory(__pa(range_start), range_end - range_start, func);
170 
171 		/* nothing more available in this segment */
172 		if (range_end == end) return 0;
173 
174 		prev_start = rsvd_region[i].end;
175 	}
176 	/* end of memory marker allows full processing inside loop body */
177 	return 0;
178 }
179 
180 /*
181  * Similar to "filter_rsvd_memory()", but the reserved memory ranges
182  * are not filtered out.
183  */
184 int __init
185 filter_memory(unsigned long start, unsigned long end, void *arg)
186 {
187 	void (*func)(unsigned long, unsigned long, int);
188 
189 #if IGNORE_PFN0
190 	if (start == PAGE_OFFSET) {
191 		printk(KERN_WARNING "warning: skipping physical page 0\n");
192 		start += PAGE_SIZE;
193 		if (start >= end)
194 			return 0;
195 	}
196 #endif
197 	func = arg;
198 	if (start < end)
199 		call_pernode_memory(__pa(start), end - start, func);
200 	return 0;
201 }
202 
203 static void __init
204 sort_regions (struct rsvd_region *rsvd_region, int max)
205 {
206 	int j;
207 
208 	/* simple bubble sorting */
209 	while (max--) {
210 		for (j = 0; j < max; ++j) {
211 			if (rsvd_region[j].start > rsvd_region[j+1].start) {
212 				struct rsvd_region tmp;
213 				tmp = rsvd_region[j];
214 				rsvd_region[j] = rsvd_region[j + 1];
215 				rsvd_region[j + 1] = tmp;
216 			}
217 		}
218 	}
219 }
220 
221 /*
222  * Request address space for all standard resources
223  */
224 static int __init register_memory(void)
225 {
226 	code_resource.start = ia64_tpa(_text);
227 	code_resource.end   = ia64_tpa(_etext) - 1;
228 	data_resource.start = ia64_tpa(_etext);
229 	data_resource.end   = ia64_tpa(_edata) - 1;
230 	bss_resource.start  = ia64_tpa(__bss_start);
231 	bss_resource.end    = ia64_tpa(_end) - 1;
232 	efi_initialize_iomem_resources(&code_resource, &data_resource,
233 			&bss_resource);
234 
235 	return 0;
236 }
237 
238 __initcall(register_memory);
239 
240 
241 #ifdef CONFIG_KEXEC
242 
243 /*
244  * This function checks if the reserved crashkernel is allowed on the specific
245  * IA64 machine flavour. Machines without an IO TLB use swiotlb and require
246  * some memory below 4 GB (i.e. in 32 bit area), see the implementation of
247  * lib/swiotlb.c. The hpzx1 architecture has an IO TLB but cannot use that
248  * in kdump case. See the comment in sba_init() in sba_iommu.c.
249  *
250  * So, the only machvec that really supports loading the kdump kernel
251  * over 4 GB is "sn2".
252  */
253 static int __init check_crashkernel_memory(unsigned long pbase, size_t size)
254 {
255 	if (ia64_platform_is("sn2") || ia64_platform_is("uv"))
256 		return 1;
257 	else
258 		return pbase < (1UL << 32);
259 }
260 
261 static void __init setup_crashkernel(unsigned long total, int *n)
262 {
263 	unsigned long long base = 0, size = 0;
264 	int ret;
265 
266 	ret = parse_crashkernel(boot_command_line, total,
267 			&size, &base);
268 	if (ret == 0 && size > 0) {
269 		if (!base) {
270 			sort_regions(rsvd_region, *n);
271 			base = kdump_find_rsvd_region(size,
272 					rsvd_region, *n);
273 		}
274 
275 		if (!check_crashkernel_memory(base, size)) {
276 			pr_warning("crashkernel: There would be kdump memory "
277 				"at %ld GB but this is unusable because it "
278 				"must\nbe below 4 GB. Change the memory "
279 				"configuration of the machine.\n",
280 				(unsigned long)(base >> 30));
281 			return;
282 		}
283 
284 		if (base != ~0UL) {
285 			printk(KERN_INFO "Reserving %ldMB of memory at %ldMB "
286 					"for crashkernel (System RAM: %ldMB)\n",
287 					(unsigned long)(size >> 20),
288 					(unsigned long)(base >> 20),
289 					(unsigned long)(total >> 20));
290 			rsvd_region[*n].start =
291 				(unsigned long)__va(base);
292 			rsvd_region[*n].end =
293 				(unsigned long)__va(base + size);
294 			(*n)++;
295 			crashk_res.start = base;
296 			crashk_res.end = base + size - 1;
297 		}
298 	}
299 	efi_memmap_res.start = ia64_boot_param->efi_memmap;
300 	efi_memmap_res.end = efi_memmap_res.start +
301 		ia64_boot_param->efi_memmap_size;
302 	boot_param_res.start = __pa(ia64_boot_param);
303 	boot_param_res.end = boot_param_res.start +
304 		sizeof(*ia64_boot_param);
305 }
306 #else
307 static inline void __init setup_crashkernel(unsigned long total, int *n)
308 {}
309 #endif
310 
311 /**
312  * reserve_memory - setup reserved memory areas
313  *
314  * Setup the reserved memory areas set aside for the boot parameters,
315  * initrd, etc.  There are currently %IA64_MAX_RSVD_REGIONS defined,
316  * see include/asm-ia64/meminit.h if you need to define more.
317  */
318 void __init
319 reserve_memory (void)
320 {
321 	int n = 0;
322 	unsigned long total_memory;
323 
324 	/*
325 	 * none of the entries in this table overlap
326 	 */
327 	rsvd_region[n].start = (unsigned long) ia64_boot_param;
328 	rsvd_region[n].end   = rsvd_region[n].start + sizeof(*ia64_boot_param);
329 	n++;
330 
331 	rsvd_region[n].start = (unsigned long) __va(ia64_boot_param->efi_memmap);
332 	rsvd_region[n].end   = rsvd_region[n].start + ia64_boot_param->efi_memmap_size;
333 	n++;
334 
335 	rsvd_region[n].start = (unsigned long) __va(ia64_boot_param->command_line);
336 	rsvd_region[n].end   = (rsvd_region[n].start
337 				+ strlen(__va(ia64_boot_param->command_line)) + 1);
338 	n++;
339 
340 	rsvd_region[n].start = (unsigned long) ia64_imva((void *)KERNEL_START);
341 	rsvd_region[n].end   = (unsigned long) ia64_imva(_end);
342 	n++;
343 
344 #ifdef CONFIG_BLK_DEV_INITRD
345 	if (ia64_boot_param->initrd_start) {
346 		rsvd_region[n].start = (unsigned long)__va(ia64_boot_param->initrd_start);
347 		rsvd_region[n].end   = rsvd_region[n].start + ia64_boot_param->initrd_size;
348 		n++;
349 	}
350 #endif
351 
352 #ifdef CONFIG_PROC_VMCORE
353 	if (reserve_elfcorehdr(&rsvd_region[n].start,
354 			       &rsvd_region[n].end) == 0)
355 		n++;
356 #endif
357 
358 	total_memory = efi_memmap_init(&rsvd_region[n].start, &rsvd_region[n].end);
359 	n++;
360 
361 	setup_crashkernel(total_memory, &n);
362 
363 	/* end of memory marker */
364 	rsvd_region[n].start = ~0UL;
365 	rsvd_region[n].end   = ~0UL;
366 	n++;
367 
368 	num_rsvd_regions = n;
369 	BUG_ON(IA64_MAX_RSVD_REGIONS + 1 < n);
370 
371 	sort_regions(rsvd_region, num_rsvd_regions);
372 }
373 
374 
375 /**
376  * find_initrd - get initrd parameters from the boot parameter structure
377  *
378  * Grab the initrd start and end from the boot parameter struct given us by
379  * the boot loader.
380  */
381 void __init
382 find_initrd (void)
383 {
384 #ifdef CONFIG_BLK_DEV_INITRD
385 	if (ia64_boot_param->initrd_start) {
386 		initrd_start = (unsigned long)__va(ia64_boot_param->initrd_start);
387 		initrd_end   = initrd_start+ia64_boot_param->initrd_size;
388 
389 		printk(KERN_INFO "Initial ramdisk at: 0x%lx (%lu bytes)\n",
390 		       initrd_start, ia64_boot_param->initrd_size);
391 	}
392 #endif
393 }
394 
395 static void __init
396 io_port_init (void)
397 {
398 	unsigned long phys_iobase;
399 
400 	/*
401 	 * Set `iobase' based on the EFI memory map or, failing that, the
402 	 * value firmware left in ar.k0.
403 	 *
404 	 * Note that in ia32 mode, IN/OUT instructions use ar.k0 to compute
405 	 * the port's virtual address, so ia32_load_state() loads it with a
406 	 * user virtual address.  But in ia64 mode, glibc uses the
407 	 * *physical* address in ar.k0 to mmap the appropriate area from
408 	 * /dev/mem, and the inX()/outX() interfaces use MMIO.  In both
409 	 * cases, user-mode can only use the legacy 0-64K I/O port space.
410 	 *
411 	 * ar.k0 is not involved in kernel I/O port accesses, which can use
412 	 * any of the I/O port spaces and are done via MMIO using the
413 	 * virtual mmio_base from the appropriate io_space[].
414 	 */
415 	phys_iobase = efi_get_iobase();
416 	if (!phys_iobase) {
417 		phys_iobase = ia64_get_kr(IA64_KR_IO_BASE);
418 		printk(KERN_INFO "No I/O port range found in EFI memory map, "
419 			"falling back to AR.KR0 (0x%lx)\n", phys_iobase);
420 	}
421 	ia64_iobase = (unsigned long) ioremap(phys_iobase, 0);
422 	ia64_set_kr(IA64_KR_IO_BASE, __pa(ia64_iobase));
423 
424 	/* setup legacy IO port space */
425 	io_space[0].mmio_base = ia64_iobase;
426 	io_space[0].sparse = 1;
427 	num_io_spaces = 1;
428 }
429 
430 /**
431  * early_console_setup - setup debugging console
432  *
433  * Consoles started here require little enough setup that we can start using
434  * them very early in the boot process, either right after the machine
435  * vector initialization, or even before if the drivers can detect their hw.
436  *
437  * Returns non-zero if a console couldn't be setup.
438  */
439 static inline int __init
440 early_console_setup (char *cmdline)
441 {
442 	int earlycons = 0;
443 
444 #ifdef CONFIG_SERIAL_SGI_L1_CONSOLE
445 	{
446 		extern int sn_serial_console_early_setup(void);
447 		if (!sn_serial_console_early_setup())
448 			earlycons++;
449 	}
450 #endif
451 #ifdef CONFIG_EFI_PCDP
452 	if (!efi_setup_pcdp_console(cmdline))
453 		earlycons++;
454 #endif
455 	if (!simcons_register())
456 		earlycons++;
457 
458 	return (earlycons) ? 0 : -1;
459 }
460 
461 static inline void
462 mark_bsp_online (void)
463 {
464 #ifdef CONFIG_SMP
465 	/* If we register an early console, allow CPU 0 to printk */
466 	cpu_set(smp_processor_id(), cpu_online_map);
467 #endif
468 }
469 
470 static __initdata int nomca;
471 static __init int setup_nomca(char *s)
472 {
473 	nomca = 1;
474 	return 0;
475 }
476 early_param("nomca", setup_nomca);
477 
478 #ifdef CONFIG_PROC_VMCORE
479 /* elfcorehdr= specifies the location of elf core header
480  * stored by the crashed kernel.
481  */
482 static int __init parse_elfcorehdr(char *arg)
483 {
484 	if (!arg)
485 		return -EINVAL;
486 
487         elfcorehdr_addr = memparse(arg, &arg);
488 	return 0;
489 }
490 early_param("elfcorehdr", parse_elfcorehdr);
491 
492 int __init reserve_elfcorehdr(unsigned long *start, unsigned long *end)
493 {
494 	unsigned long length;
495 
496 	/* We get the address using the kernel command line,
497 	 * but the size is extracted from the EFI tables.
498 	 * Both address and size are required for reservation
499 	 * to work properly.
500 	 */
501 
502 	if (elfcorehdr_addr >= ELFCORE_ADDR_MAX)
503 		return -EINVAL;
504 
505 	if ((length = vmcore_find_descriptor_size(elfcorehdr_addr)) == 0) {
506 		elfcorehdr_addr = ELFCORE_ADDR_MAX;
507 		return -EINVAL;
508 	}
509 
510 	*start = (unsigned long)__va(elfcorehdr_addr);
511 	*end = *start + length;
512 	return 0;
513 }
514 
515 #endif /* CONFIG_PROC_VMCORE */
516 
517 void __init
518 setup_arch (char **cmdline_p)
519 {
520 	unw_init();
521 
522 	ia64_patch_vtop((u64) __start___vtop_patchlist, (u64) __end___vtop_patchlist);
523 
524 	*cmdline_p = __va(ia64_boot_param->command_line);
525 	strlcpy(boot_command_line, *cmdline_p, COMMAND_LINE_SIZE);
526 
527 	efi_init();
528 	io_port_init();
529 
530 #ifdef CONFIG_IA64_GENERIC
531 	/* machvec needs to be parsed from the command line
532 	 * before parse_early_param() is called to ensure
533 	 * that ia64_mv is initialised before any command line
534 	 * settings may cause console setup to occur
535 	 */
536 	machvec_init_from_cmdline(*cmdline_p);
537 #endif
538 
539 	parse_early_param();
540 
541 	if (early_console_setup(*cmdline_p) == 0)
542 		mark_bsp_online();
543 
544 #ifdef CONFIG_ACPI
545 	/* Initialize the ACPI boot-time table parser */
546 	acpi_table_init();
547 # ifdef CONFIG_ACPI_NUMA
548 	acpi_numa_init();
549 	per_cpu_scan_finalize((cpus_weight(early_cpu_possible_map) == 0 ?
550 		32 : cpus_weight(early_cpu_possible_map)), additional_cpus);
551 # endif
552 #else
553 # ifdef CONFIG_SMP
554 	smp_build_cpu_map();	/* happens, e.g., with the Ski simulator */
555 # endif
556 #endif /* CONFIG_APCI_BOOT */
557 
558 	find_memory();
559 
560 	/* process SAL system table: */
561 	ia64_sal_init(__va(efi.sal_systab));
562 
563 #ifdef CONFIG_SMP
564 	cpu_physical_id(0) = hard_smp_processor_id();
565 #endif
566 
567 	cpu_init();	/* initialize the bootstrap CPU */
568 	mmu_context_init();	/* initialize context_id bitmap */
569 
570 	check_sal_cache_flush();
571 
572 #ifdef CONFIG_ACPI
573 	acpi_boot_init();
574 #endif
575 
576 #ifdef CONFIG_VT
577 	if (!conswitchp) {
578 # if defined(CONFIG_DUMMY_CONSOLE)
579 		conswitchp = &dummy_con;
580 # endif
581 # if defined(CONFIG_VGA_CONSOLE)
582 		/*
583 		 * Non-legacy systems may route legacy VGA MMIO range to system
584 		 * memory.  vga_con probes the MMIO hole, so memory looks like
585 		 * a VGA device to it.  The EFI memory map can tell us if it's
586 		 * memory so we can avoid this problem.
587 		 */
588 		if (efi_mem_type(0xA0000) != EFI_CONVENTIONAL_MEMORY)
589 			conswitchp = &vga_con;
590 # endif
591 	}
592 #endif
593 
594 	/* enable IA-64 Machine Check Abort Handling unless disabled */
595 	if (!nomca)
596 		ia64_mca_init();
597 
598 	platform_setup(cmdline_p);
599 	paging_init();
600 }
601 
602 /*
603  * Display cpu info for all CPUs.
604  */
605 static int
606 show_cpuinfo (struct seq_file *m, void *v)
607 {
608 #ifdef CONFIG_SMP
609 #	define lpj	c->loops_per_jiffy
610 #	define cpunum	c->cpu
611 #else
612 #	define lpj	loops_per_jiffy
613 #	define cpunum	0
614 #endif
615 	static struct {
616 		unsigned long mask;
617 		const char *feature_name;
618 	} feature_bits[] = {
619 		{ 1UL << 0, "branchlong" },
620 		{ 1UL << 1, "spontaneous deferral"},
621 		{ 1UL << 2, "16-byte atomic ops" }
622 	};
623 	char features[128], *cp, *sep;
624 	struct cpuinfo_ia64 *c = v;
625 	unsigned long mask;
626 	unsigned long proc_freq;
627 	int i, size;
628 
629 	mask = c->features;
630 
631 	/* build the feature string: */
632 	memcpy(features, "standard", 9);
633 	cp = features;
634 	size = sizeof(features);
635 	sep = "";
636 	for (i = 0; i < ARRAY_SIZE(feature_bits) && size > 1; ++i) {
637 		if (mask & feature_bits[i].mask) {
638 			cp += snprintf(cp, size, "%s%s", sep,
639 				       feature_bits[i].feature_name),
640 			sep = ", ";
641 			mask &= ~feature_bits[i].mask;
642 			size = sizeof(features) - (cp - features);
643 		}
644 	}
645 	if (mask && size > 1) {
646 		/* print unknown features as a hex value */
647 		snprintf(cp, size, "%s0x%lx", sep, mask);
648 	}
649 
650 	proc_freq = cpufreq_quick_get(cpunum);
651 	if (!proc_freq)
652 		proc_freq = c->proc_freq / 1000;
653 
654 	seq_printf(m,
655 		   "processor  : %d\n"
656 		   "vendor     : %s\n"
657 		   "arch       : IA-64\n"
658 		   "family     : %u\n"
659 		   "model      : %u\n"
660 		   "model name : %s\n"
661 		   "revision   : %u\n"
662 		   "archrev    : %u\n"
663 		   "features   : %s\n"
664 		   "cpu number : %lu\n"
665 		   "cpu regs   : %u\n"
666 		   "cpu MHz    : %lu.%03lu\n"
667 		   "itc MHz    : %lu.%06lu\n"
668 		   "BogoMIPS   : %lu.%02lu\n",
669 		   cpunum, c->vendor, c->family, c->model,
670 		   c->model_name, c->revision, c->archrev,
671 		   features, c->ppn, c->number,
672 		   proc_freq / 1000, proc_freq % 1000,
673 		   c->itc_freq / 1000000, c->itc_freq % 1000000,
674 		   lpj*HZ/500000, (lpj*HZ/5000) % 100);
675 #ifdef CONFIG_SMP
676 	seq_printf(m, "siblings   : %u\n", cpus_weight(cpu_core_map[cpunum]));
677 	if (c->socket_id != -1)
678 		seq_printf(m, "physical id: %u\n", c->socket_id);
679 	if (c->threads_per_core > 1 || c->cores_per_socket > 1)
680 		seq_printf(m,
681 			   "core id    : %u\n"
682 			   "thread id  : %u\n",
683 			   c->core_id, c->thread_id);
684 #endif
685 	seq_printf(m,"\n");
686 
687 	return 0;
688 }
689 
690 static void *
691 c_start (struct seq_file *m, loff_t *pos)
692 {
693 #ifdef CONFIG_SMP
694 	while (*pos < NR_CPUS && !cpu_isset(*pos, cpu_online_map))
695 		++*pos;
696 #endif
697 	return *pos < NR_CPUS ? cpu_data(*pos) : NULL;
698 }
699 
700 static void *
701 c_next (struct seq_file *m, void *v, loff_t *pos)
702 {
703 	++*pos;
704 	return c_start(m, pos);
705 }
706 
707 static void
708 c_stop (struct seq_file *m, void *v)
709 {
710 }
711 
712 const struct seq_operations cpuinfo_op = {
713 	.start =	c_start,
714 	.next =		c_next,
715 	.stop =		c_stop,
716 	.show =		show_cpuinfo
717 };
718 
719 #define MAX_BRANDS	8
720 static char brandname[MAX_BRANDS][128];
721 
722 static char * __cpuinit
723 get_model_name(__u8 family, __u8 model)
724 {
725 	static int overflow;
726 	char brand[128];
727 	int i;
728 
729 	memcpy(brand, "Unknown", 8);
730 	if (ia64_pal_get_brand_info(brand)) {
731 		if (family == 0x7)
732 			memcpy(brand, "Merced", 7);
733 		else if (family == 0x1f) switch (model) {
734 			case 0: memcpy(brand, "McKinley", 9); break;
735 			case 1: memcpy(brand, "Madison", 8); break;
736 			case 2: memcpy(brand, "Madison up to 9M cache", 23); break;
737 		}
738 	}
739 	for (i = 0; i < MAX_BRANDS; i++)
740 		if (strcmp(brandname[i], brand) == 0)
741 			return brandname[i];
742 	for (i = 0; i < MAX_BRANDS; i++)
743 		if (brandname[i][0] == '\0')
744 			return strcpy(brandname[i], brand);
745 	if (overflow++ == 0)
746 		printk(KERN_ERR
747 		       "%s: Table overflow. Some processor model information will be missing\n",
748 		       __func__);
749 	return "Unknown";
750 }
751 
752 static void __cpuinit
753 identify_cpu (struct cpuinfo_ia64 *c)
754 {
755 	union {
756 		unsigned long bits[5];
757 		struct {
758 			/* id 0 & 1: */
759 			char vendor[16];
760 
761 			/* id 2 */
762 			u64 ppn;		/* processor serial number */
763 
764 			/* id 3: */
765 			unsigned number		:  8;
766 			unsigned revision	:  8;
767 			unsigned model		:  8;
768 			unsigned family		:  8;
769 			unsigned archrev	:  8;
770 			unsigned reserved	: 24;
771 
772 			/* id 4: */
773 			u64 features;
774 		} field;
775 	} cpuid;
776 	pal_vm_info_1_u_t vm1;
777 	pal_vm_info_2_u_t vm2;
778 	pal_status_t status;
779 	unsigned long impl_va_msb = 50, phys_addr_size = 44;	/* Itanium defaults */
780 	int i;
781 	for (i = 0; i < 5; ++i)
782 		cpuid.bits[i] = ia64_get_cpuid(i);
783 
784 	memcpy(c->vendor, cpuid.field.vendor, 16);
785 #ifdef CONFIG_SMP
786 	c->cpu = smp_processor_id();
787 
788 	/* below default values will be overwritten  by identify_siblings()
789 	 * for Multi-Threading/Multi-Core capable CPUs
790 	 */
791 	c->threads_per_core = c->cores_per_socket = c->num_log = 1;
792 	c->socket_id = -1;
793 
794 	identify_siblings(c);
795 
796 	if (c->threads_per_core > smp_num_siblings)
797 		smp_num_siblings = c->threads_per_core;
798 #endif
799 	c->ppn = cpuid.field.ppn;
800 	c->number = cpuid.field.number;
801 	c->revision = cpuid.field.revision;
802 	c->model = cpuid.field.model;
803 	c->family = cpuid.field.family;
804 	c->archrev = cpuid.field.archrev;
805 	c->features = cpuid.field.features;
806 	c->model_name = get_model_name(c->family, c->model);
807 
808 	status = ia64_pal_vm_summary(&vm1, &vm2);
809 	if (status == PAL_STATUS_SUCCESS) {
810 		impl_va_msb = vm2.pal_vm_info_2_s.impl_va_msb;
811 		phys_addr_size = vm1.pal_vm_info_1_s.phys_add_size;
812 	}
813 	c->unimpl_va_mask = ~((7L<<61) | ((1L << (impl_va_msb + 1)) - 1));
814 	c->unimpl_pa_mask = ~((1L<<63) | ((1L << phys_addr_size) - 1));
815 }
816 
817 void __init
818 setup_per_cpu_areas (void)
819 {
820 	/* start_kernel() requires this... */
821 #ifdef CONFIG_ACPI_HOTPLUG_CPU
822 	prefill_possible_map();
823 #endif
824 }
825 
826 /*
827  * Calculate the max. cache line size.
828  *
829  * In addition, the minimum of the i-cache stride sizes is calculated for
830  * "flush_icache_range()".
831  */
832 static void __cpuinit
833 get_max_cacheline_size (void)
834 {
835 	unsigned long line_size, max = 1;
836 	u64 l, levels, unique_caches;
837         pal_cache_config_info_t cci;
838         s64 status;
839 
840         status = ia64_pal_cache_summary(&levels, &unique_caches);
841         if (status != 0) {
842                 printk(KERN_ERR "%s: ia64_pal_cache_summary() failed (status=%ld)\n",
843                        __func__, status);
844                 max = SMP_CACHE_BYTES;
845 		/* Safest setup for "flush_icache_range()" */
846 		ia64_i_cache_stride_shift = I_CACHE_STRIDE_SHIFT;
847 		goto out;
848         }
849 
850 	for (l = 0; l < levels; ++l) {
851 		status = ia64_pal_cache_config_info(l, /* cache_type (data_or_unified)= */ 2,
852 						    &cci);
853 		if (status != 0) {
854 			printk(KERN_ERR
855 			       "%s: ia64_pal_cache_config_info(l=%lu, 2) failed (status=%ld)\n",
856 			       __func__, l, status);
857 			max = SMP_CACHE_BYTES;
858 			/* The safest setup for "flush_icache_range()" */
859 			cci.pcci_stride = I_CACHE_STRIDE_SHIFT;
860 			cci.pcci_unified = 1;
861 		}
862 		line_size = 1 << cci.pcci_line_size;
863 		if (line_size > max)
864 			max = line_size;
865 		if (!cci.pcci_unified) {
866 			status = ia64_pal_cache_config_info(l,
867 						    /* cache_type (instruction)= */ 1,
868 						    &cci);
869 			if (status != 0) {
870 				printk(KERN_ERR
871 				"%s: ia64_pal_cache_config_info(l=%lu, 1) failed (status=%ld)\n",
872 					__func__, l, status);
873 				/* The safest setup for "flush_icache_range()" */
874 				cci.pcci_stride = I_CACHE_STRIDE_SHIFT;
875 			}
876 		}
877 		if (cci.pcci_stride < ia64_i_cache_stride_shift)
878 			ia64_i_cache_stride_shift = cci.pcci_stride;
879 	}
880   out:
881 	if (max > ia64_max_cacheline_size)
882 		ia64_max_cacheline_size = max;
883 }
884 
885 /*
886  * cpu_init() initializes state that is per-CPU.  This function acts
887  * as a 'CPU state barrier', nothing should get across.
888  */
889 void __cpuinit
890 cpu_init (void)
891 {
892 	extern void __cpuinit ia64_mmu_init (void *);
893 	static unsigned long max_num_phys_stacked = IA64_NUM_PHYS_STACK_REG;
894 	unsigned long num_phys_stacked;
895 	pal_vm_info_2_u_t vmi;
896 	unsigned int max_ctx;
897 	struct cpuinfo_ia64 *cpu_info;
898 	void *cpu_data;
899 
900 	cpu_data = per_cpu_init();
901 #ifdef CONFIG_SMP
902 	/*
903 	 * insert boot cpu into sibling and core mapes
904 	 * (must be done after per_cpu area is setup)
905 	 */
906 	if (smp_processor_id() == 0) {
907 		cpu_set(0, per_cpu(cpu_sibling_map, 0));
908 		cpu_set(0, cpu_core_map[0]);
909 	}
910 #endif
911 
912 	/*
913 	 * We set ar.k3 so that assembly code in MCA handler can compute
914 	 * physical addresses of per cpu variables with a simple:
915 	 *   phys = ar.k3 + &per_cpu_var
916 	 */
917 	ia64_set_kr(IA64_KR_PER_CPU_DATA,
918 		    ia64_tpa(cpu_data) - (long) __per_cpu_start);
919 
920 	get_max_cacheline_size();
921 
922 	/*
923 	 * We can't pass "local_cpu_data" to identify_cpu() because we haven't called
924 	 * ia64_mmu_init() yet.  And we can't call ia64_mmu_init() first because it
925 	 * depends on the data returned by identify_cpu().  We break the dependency by
926 	 * accessing cpu_data() through the canonical per-CPU address.
927 	 */
928 	cpu_info = cpu_data + ((char *) &__ia64_per_cpu_var(cpu_info) - __per_cpu_start);
929 	identify_cpu(cpu_info);
930 
931 #ifdef CONFIG_MCKINLEY
932 	{
933 #		define FEATURE_SET 16
934 		struct ia64_pal_retval iprv;
935 
936 		if (cpu_info->family == 0x1f) {
937 			PAL_CALL_PHYS(iprv, PAL_PROC_GET_FEATURES, 0, FEATURE_SET, 0);
938 			if ((iprv.status == 0) && (iprv.v0 & 0x80) && (iprv.v2 & 0x80))
939 				PAL_CALL_PHYS(iprv, PAL_PROC_SET_FEATURES,
940 				              (iprv.v1 | 0x80), FEATURE_SET, 0);
941 		}
942 	}
943 #endif
944 
945 	/* Clear the stack memory reserved for pt_regs: */
946 	memset(task_pt_regs(current), 0, sizeof(struct pt_regs));
947 
948 	ia64_set_kr(IA64_KR_FPU_OWNER, 0);
949 
950 	/*
951 	 * Initialize the page-table base register to a global
952 	 * directory with all zeroes.  This ensure that we can handle
953 	 * TLB-misses to user address-space even before we created the
954 	 * first user address-space.  This may happen, e.g., due to
955 	 * aggressive use of lfetch.fault.
956 	 */
957 	ia64_set_kr(IA64_KR_PT_BASE, __pa(ia64_imva(empty_zero_page)));
958 
959 	/*
960 	 * Initialize default control register to defer speculative faults except
961 	 * for those arising from TLB misses, which are not deferred.  The
962 	 * kernel MUST NOT depend on a particular setting of these bits (in other words,
963 	 * the kernel must have recovery code for all speculative accesses).  Turn on
964 	 * dcr.lc as per recommendation by the architecture team.  Most IA-32 apps
965 	 * shouldn't be affected by this (moral: keep your ia32 locks aligned and you'll
966 	 * be fine).
967 	 */
968 	ia64_setreg(_IA64_REG_CR_DCR,  (  IA64_DCR_DP | IA64_DCR_DK | IA64_DCR_DX | IA64_DCR_DR
969 					| IA64_DCR_DA | IA64_DCR_DD | IA64_DCR_LC));
970 	atomic_inc(&init_mm.mm_count);
971 	current->active_mm = &init_mm;
972 	if (current->mm)
973 		BUG();
974 
975 	ia64_mmu_init(ia64_imva(cpu_data));
976 	ia64_mca_cpu_init(ia64_imva(cpu_data));
977 
978 #ifdef CONFIG_IA32_SUPPORT
979 	ia32_cpu_init();
980 #endif
981 
982 	/* Clear ITC to eliminate sched_clock() overflows in human time.  */
983 	ia64_set_itc(0);
984 
985 	/* disable all local interrupt sources: */
986 	ia64_set_itv(1 << 16);
987 	ia64_set_lrr0(1 << 16);
988 	ia64_set_lrr1(1 << 16);
989 	ia64_setreg(_IA64_REG_CR_PMV, 1 << 16);
990 	ia64_setreg(_IA64_REG_CR_CMCV, 1 << 16);
991 
992 	/* clear TPR & XTP to enable all interrupt classes: */
993 	ia64_setreg(_IA64_REG_CR_TPR, 0);
994 
995 	/* Clear any pending interrupts left by SAL/EFI */
996 	while (ia64_get_ivr() != IA64_SPURIOUS_INT_VECTOR)
997 		ia64_eoi();
998 
999 #ifdef CONFIG_SMP
1000 	normal_xtp();
1001 #endif
1002 
1003 	/* set ia64_ctx.max_rid to the maximum RID that is supported by all CPUs: */
1004 	if (ia64_pal_vm_summary(NULL, &vmi) == 0) {
1005 		max_ctx = (1U << (vmi.pal_vm_info_2_s.rid_size - 3)) - 1;
1006 		setup_ptcg_sem(vmi.pal_vm_info_2_s.max_purges, NPTCG_FROM_PAL);
1007 	} else {
1008 		printk(KERN_WARNING "cpu_init: PAL VM summary failed, assuming 18 RID bits\n");
1009 		max_ctx = (1U << 15) - 1;	/* use architected minimum */
1010 	}
1011 	while (max_ctx < ia64_ctx.max_ctx) {
1012 		unsigned int old = ia64_ctx.max_ctx;
1013 		if (cmpxchg(&ia64_ctx.max_ctx, old, max_ctx) == old)
1014 			break;
1015 	}
1016 
1017 	if (ia64_pal_rse_info(&num_phys_stacked, NULL) != 0) {
1018 		printk(KERN_WARNING "cpu_init: PAL RSE info failed; assuming 96 physical "
1019 		       "stacked regs\n");
1020 		num_phys_stacked = 96;
1021 	}
1022 	/* size of physical stacked register partition plus 8 bytes: */
1023 	if (num_phys_stacked > max_num_phys_stacked) {
1024 		ia64_patch_phys_stack_reg(num_phys_stacked*8 + 8);
1025 		max_num_phys_stacked = num_phys_stacked;
1026 	}
1027 	platform_cpu_init();
1028 	pm_idle = default_idle;
1029 }
1030 
1031 void __init
1032 check_bugs (void)
1033 {
1034 	ia64_patch_mckinley_e9((unsigned long) __start___mckinley_e9_bundles,
1035 			       (unsigned long) __end___mckinley_e9_bundles);
1036 }
1037 
1038 static int __init run_dmi_scan(void)
1039 {
1040 	dmi_scan_machine();
1041 	return 0;
1042 }
1043 core_initcall(run_dmi_scan);
1044