xref: /openbmc/linux/arch/ia64/kernel/setup.c (revision 384740dc)
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/paravirt.h>
55 #include <asm/patch.h>
56 #include <asm/pgtable.h>
57 #include <asm/processor.h>
58 #include <asm/sal.h>
59 #include <asm/sections.h>
60 #include <asm/setup.h>
61 #include <asm/smp.h>
62 #include <asm/system.h>
63 #include <asm/tlbflush.h>
64 #include <asm/unistd.h>
65 #include <asm/hpsim.h>
66 
67 #if defined(CONFIG_SMP) && (IA64_CPU_SIZE > PAGE_SIZE)
68 # error "struct cpuinfo_ia64 too big!"
69 #endif
70 
71 #ifdef CONFIG_SMP
72 unsigned long __per_cpu_offset[NR_CPUS];
73 EXPORT_SYMBOL(__per_cpu_offset);
74 #endif
75 
76 DEFINE_PER_CPU(struct cpuinfo_ia64, cpu_info);
77 DEFINE_PER_CPU(unsigned long, local_per_cpu_offset);
78 unsigned long ia64_cycles_per_usec;
79 struct ia64_boot_param *ia64_boot_param;
80 struct screen_info screen_info;
81 unsigned long vga_console_iobase;
82 unsigned long vga_console_membase;
83 
84 static struct resource data_resource = {
85 	.name	= "Kernel data",
86 	.flags	= IORESOURCE_BUSY | IORESOURCE_MEM
87 };
88 
89 static struct resource code_resource = {
90 	.name	= "Kernel code",
91 	.flags	= IORESOURCE_BUSY | IORESOURCE_MEM
92 };
93 
94 static struct resource bss_resource = {
95 	.name	= "Kernel bss",
96 	.flags	= IORESOURCE_BUSY | IORESOURCE_MEM
97 };
98 
99 unsigned long ia64_max_cacheline_size;
100 
101 int dma_get_cache_alignment(void)
102 {
103         return ia64_max_cacheline_size;
104 }
105 EXPORT_SYMBOL(dma_get_cache_alignment);
106 
107 unsigned long ia64_iobase;	/* virtual address for I/O accesses */
108 EXPORT_SYMBOL(ia64_iobase);
109 struct io_space io_space[MAX_IO_SPACES];
110 EXPORT_SYMBOL(io_space);
111 unsigned int num_io_spaces;
112 
113 /*
114  * "flush_icache_range()" needs to know what processor dependent stride size to use
115  * when it makes i-cache(s) coherent with d-caches.
116  */
117 #define	I_CACHE_STRIDE_SHIFT	5	/* Safest way to go: 32 bytes by 32 bytes */
118 unsigned long ia64_i_cache_stride_shift = ~0;
119 
120 /*
121  * The merge_mask variable needs to be set to (max(iommu_page_size(iommu)) - 1).  This
122  * mask specifies a mask of address bits that must be 0 in order for two buffers to be
123  * mergeable by the I/O MMU (i.e., the end address of the first buffer and the start
124  * address of the second buffer must be aligned to (merge_mask+1) in order to be
125  * mergeable).  By default, we assume there is no I/O MMU which can merge physically
126  * discontiguous buffers, so we set the merge_mask to ~0UL, which corresponds to a iommu
127  * page-size of 2^64.
128  */
129 unsigned long ia64_max_iommu_merge_mask = ~0UL;
130 EXPORT_SYMBOL(ia64_max_iommu_merge_mask);
131 
132 /*
133  * We use a special marker for the end of memory and it uses the extra (+1) slot
134  */
135 struct rsvd_region rsvd_region[IA64_MAX_RSVD_REGIONS + 1] __initdata;
136 int num_rsvd_regions __initdata;
137 
138 
139 /*
140  * Filter incoming memory segments based on the primitive map created from the boot
141  * parameters. Segments contained in the map are removed from the memory ranges. A
142  * caller-specified function is called with the memory ranges that remain after filtering.
143  * This routine does not assume the incoming segments are sorted.
144  */
145 int __init
146 filter_rsvd_memory (unsigned long start, unsigned long end, void *arg)
147 {
148 	unsigned long range_start, range_end, prev_start;
149 	void (*func)(unsigned long, unsigned long, int);
150 	int i;
151 
152 #if IGNORE_PFN0
153 	if (start == PAGE_OFFSET) {
154 		printk(KERN_WARNING "warning: skipping physical page 0\n");
155 		start += PAGE_SIZE;
156 		if (start >= end) return 0;
157 	}
158 #endif
159 	/*
160 	 * lowest possible address(walker uses virtual)
161 	 */
162 	prev_start = PAGE_OFFSET;
163 	func = arg;
164 
165 	for (i = 0; i < num_rsvd_regions; ++i) {
166 		range_start = max(start, prev_start);
167 		range_end   = min(end, rsvd_region[i].start);
168 
169 		if (range_start < range_end)
170 			call_pernode_memory(__pa(range_start), range_end - range_start, func);
171 
172 		/* nothing more available in this segment */
173 		if (range_end == end) return 0;
174 
175 		prev_start = rsvd_region[i].end;
176 	}
177 	/* end of memory marker allows full processing inside loop body */
178 	return 0;
179 }
180 
181 /*
182  * Similar to "filter_rsvd_memory()", but the reserved memory ranges
183  * are not filtered out.
184  */
185 int __init
186 filter_memory(unsigned long start, unsigned long end, void *arg)
187 {
188 	void (*func)(unsigned long, unsigned long, int);
189 
190 #if IGNORE_PFN0
191 	if (start == PAGE_OFFSET) {
192 		printk(KERN_WARNING "warning: skipping physical page 0\n");
193 		start += PAGE_SIZE;
194 		if (start >= end)
195 			return 0;
196 	}
197 #endif
198 	func = arg;
199 	if (start < end)
200 		call_pernode_memory(__pa(start), end - start, func);
201 	return 0;
202 }
203 
204 static void __init
205 sort_regions (struct rsvd_region *rsvd_region, int max)
206 {
207 	int j;
208 
209 	/* simple bubble sorting */
210 	while (max--) {
211 		for (j = 0; j < max; ++j) {
212 			if (rsvd_region[j].start > rsvd_region[j+1].start) {
213 				struct rsvd_region tmp;
214 				tmp = rsvd_region[j];
215 				rsvd_region[j] = rsvd_region[j + 1];
216 				rsvd_region[j + 1] = tmp;
217 			}
218 		}
219 	}
220 }
221 
222 /*
223  * Request address space for all standard resources
224  */
225 static int __init register_memory(void)
226 {
227 	code_resource.start = ia64_tpa(_text);
228 	code_resource.end   = ia64_tpa(_etext) - 1;
229 	data_resource.start = ia64_tpa(_etext);
230 	data_resource.end   = ia64_tpa(_edata) - 1;
231 	bss_resource.start  = ia64_tpa(__bss_start);
232 	bss_resource.end    = ia64_tpa(_end) - 1;
233 	efi_initialize_iomem_resources(&code_resource, &data_resource,
234 			&bss_resource);
235 
236 	return 0;
237 }
238 
239 __initcall(register_memory);
240 
241 
242 #ifdef CONFIG_KEXEC
243 
244 /*
245  * This function checks if the reserved crashkernel is allowed on the specific
246  * IA64 machine flavour. Machines without an IO TLB use swiotlb and require
247  * some memory below 4 GB (i.e. in 32 bit area), see the implementation of
248  * lib/swiotlb.c. The hpzx1 architecture has an IO TLB but cannot use that
249  * in kdump case. See the comment in sba_init() in sba_iommu.c.
250  *
251  * So, the only machvec that really supports loading the kdump kernel
252  * over 4 GB is "sn2".
253  */
254 static int __init check_crashkernel_memory(unsigned long pbase, size_t size)
255 {
256 	if (ia64_platform_is("sn2") || ia64_platform_is("uv"))
257 		return 1;
258 	else
259 		return pbase < (1UL << 32);
260 }
261 
262 static void __init setup_crashkernel(unsigned long total, int *n)
263 {
264 	unsigned long long base = 0, size = 0;
265 	int ret;
266 
267 	ret = parse_crashkernel(boot_command_line, total,
268 			&size, &base);
269 	if (ret == 0 && size > 0) {
270 		if (!base) {
271 			sort_regions(rsvd_region, *n);
272 			base = kdump_find_rsvd_region(size,
273 					rsvd_region, *n);
274 		}
275 
276 		if (!check_crashkernel_memory(base, size)) {
277 			pr_warning("crashkernel: There would be kdump memory "
278 				"at %ld GB but this is unusable because it "
279 				"must\nbe below 4 GB. Change the memory "
280 				"configuration of the machine.\n",
281 				(unsigned long)(base >> 30));
282 			return;
283 		}
284 
285 		if (base != ~0UL) {
286 			printk(KERN_INFO "Reserving %ldMB of memory at %ldMB "
287 					"for crashkernel (System RAM: %ldMB)\n",
288 					(unsigned long)(size >> 20),
289 					(unsigned long)(base >> 20),
290 					(unsigned long)(total >> 20));
291 			rsvd_region[*n].start =
292 				(unsigned long)__va(base);
293 			rsvd_region[*n].end =
294 				(unsigned long)__va(base + size);
295 			(*n)++;
296 			crashk_res.start = base;
297 			crashk_res.end = base + size - 1;
298 		}
299 	}
300 	efi_memmap_res.start = ia64_boot_param->efi_memmap;
301 	efi_memmap_res.end = efi_memmap_res.start +
302 		ia64_boot_param->efi_memmap_size;
303 	boot_param_res.start = __pa(ia64_boot_param);
304 	boot_param_res.end = boot_param_res.start +
305 		sizeof(*ia64_boot_param);
306 }
307 #else
308 static inline void __init setup_crashkernel(unsigned long total, int *n)
309 {}
310 #endif
311 
312 /**
313  * reserve_memory - setup reserved memory areas
314  *
315  * Setup the reserved memory areas set aside for the boot parameters,
316  * initrd, etc.  There are currently %IA64_MAX_RSVD_REGIONS defined,
317  * see arch/ia64/include/asm/meminit.h if you need to define more.
318  */
319 void __init
320 reserve_memory (void)
321 {
322 	int n = 0;
323 	unsigned long total_memory;
324 
325 	/*
326 	 * none of the entries in this table overlap
327 	 */
328 	rsvd_region[n].start = (unsigned long) ia64_boot_param;
329 	rsvd_region[n].end   = rsvd_region[n].start + sizeof(*ia64_boot_param);
330 	n++;
331 
332 	rsvd_region[n].start = (unsigned long) __va(ia64_boot_param->efi_memmap);
333 	rsvd_region[n].end   = rsvd_region[n].start + ia64_boot_param->efi_memmap_size;
334 	n++;
335 
336 	rsvd_region[n].start = (unsigned long) __va(ia64_boot_param->command_line);
337 	rsvd_region[n].end   = (rsvd_region[n].start
338 				+ strlen(__va(ia64_boot_param->command_line)) + 1);
339 	n++;
340 
341 	rsvd_region[n].start = (unsigned long) ia64_imva((void *)KERNEL_START);
342 	rsvd_region[n].end   = (unsigned long) ia64_imva(_end);
343 	n++;
344 
345 	n += paravirt_reserve_memory(&rsvd_region[n]);
346 
347 #ifdef CONFIG_BLK_DEV_INITRD
348 	if (ia64_boot_param->initrd_start) {
349 		rsvd_region[n].start = (unsigned long)__va(ia64_boot_param->initrd_start);
350 		rsvd_region[n].end   = rsvd_region[n].start + ia64_boot_param->initrd_size;
351 		n++;
352 	}
353 #endif
354 
355 #ifdef CONFIG_PROC_VMCORE
356 	if (reserve_elfcorehdr(&rsvd_region[n].start,
357 			       &rsvd_region[n].end) == 0)
358 		n++;
359 #endif
360 
361 	total_memory = efi_memmap_init(&rsvd_region[n].start, &rsvd_region[n].end);
362 	n++;
363 
364 	setup_crashkernel(total_memory, &n);
365 
366 	/* end of memory marker */
367 	rsvd_region[n].start = ~0UL;
368 	rsvd_region[n].end   = ~0UL;
369 	n++;
370 
371 	num_rsvd_regions = n;
372 	BUG_ON(IA64_MAX_RSVD_REGIONS + 1 < n);
373 
374 	sort_regions(rsvd_region, num_rsvd_regions);
375 }
376 
377 
378 /**
379  * find_initrd - get initrd parameters from the boot parameter structure
380  *
381  * Grab the initrd start and end from the boot parameter struct given us by
382  * the boot loader.
383  */
384 void __init
385 find_initrd (void)
386 {
387 #ifdef CONFIG_BLK_DEV_INITRD
388 	if (ia64_boot_param->initrd_start) {
389 		initrd_start = (unsigned long)__va(ia64_boot_param->initrd_start);
390 		initrd_end   = initrd_start+ia64_boot_param->initrd_size;
391 
392 		printk(KERN_INFO "Initial ramdisk at: 0x%lx (%lu bytes)\n",
393 		       initrd_start, ia64_boot_param->initrd_size);
394 	}
395 #endif
396 }
397 
398 static void __init
399 io_port_init (void)
400 {
401 	unsigned long phys_iobase;
402 
403 	/*
404 	 * Set `iobase' based on the EFI memory map or, failing that, the
405 	 * value firmware left in ar.k0.
406 	 *
407 	 * Note that in ia32 mode, IN/OUT instructions use ar.k0 to compute
408 	 * the port's virtual address, so ia32_load_state() loads it with a
409 	 * user virtual address.  But in ia64 mode, glibc uses the
410 	 * *physical* address in ar.k0 to mmap the appropriate area from
411 	 * /dev/mem, and the inX()/outX() interfaces use MMIO.  In both
412 	 * cases, user-mode can only use the legacy 0-64K I/O port space.
413 	 *
414 	 * ar.k0 is not involved in kernel I/O port accesses, which can use
415 	 * any of the I/O port spaces and are done via MMIO using the
416 	 * virtual mmio_base from the appropriate io_space[].
417 	 */
418 	phys_iobase = efi_get_iobase();
419 	if (!phys_iobase) {
420 		phys_iobase = ia64_get_kr(IA64_KR_IO_BASE);
421 		printk(KERN_INFO "No I/O port range found in EFI memory map, "
422 			"falling back to AR.KR0 (0x%lx)\n", phys_iobase);
423 	}
424 	ia64_iobase = (unsigned long) ioremap(phys_iobase, 0);
425 	ia64_set_kr(IA64_KR_IO_BASE, __pa(ia64_iobase));
426 
427 	/* setup legacy IO port space */
428 	io_space[0].mmio_base = ia64_iobase;
429 	io_space[0].sparse = 1;
430 	num_io_spaces = 1;
431 }
432 
433 /**
434  * early_console_setup - setup debugging console
435  *
436  * Consoles started here require little enough setup that we can start using
437  * them very early in the boot process, either right after the machine
438  * vector initialization, or even before if the drivers can detect their hw.
439  *
440  * Returns non-zero if a console couldn't be setup.
441  */
442 static inline int __init
443 early_console_setup (char *cmdline)
444 {
445 	int earlycons = 0;
446 
447 #ifdef CONFIG_SERIAL_SGI_L1_CONSOLE
448 	{
449 		extern int sn_serial_console_early_setup(void);
450 		if (!sn_serial_console_early_setup())
451 			earlycons++;
452 	}
453 #endif
454 #ifdef CONFIG_EFI_PCDP
455 	if (!efi_setup_pcdp_console(cmdline))
456 		earlycons++;
457 #endif
458 	if (!simcons_register())
459 		earlycons++;
460 
461 	return (earlycons) ? 0 : -1;
462 }
463 
464 static inline void
465 mark_bsp_online (void)
466 {
467 #ifdef CONFIG_SMP
468 	/* If we register an early console, allow CPU 0 to printk */
469 	cpu_set(smp_processor_id(), cpu_online_map);
470 #endif
471 }
472 
473 static __initdata int nomca;
474 static __init int setup_nomca(char *s)
475 {
476 	nomca = 1;
477 	return 0;
478 }
479 early_param("nomca", setup_nomca);
480 
481 #ifdef CONFIG_PROC_VMCORE
482 /* elfcorehdr= specifies the location of elf core header
483  * stored by the crashed kernel.
484  */
485 static int __init parse_elfcorehdr(char *arg)
486 {
487 	if (!arg)
488 		return -EINVAL;
489 
490         elfcorehdr_addr = memparse(arg, &arg);
491 	return 0;
492 }
493 early_param("elfcorehdr", parse_elfcorehdr);
494 
495 int __init reserve_elfcorehdr(unsigned long *start, unsigned long *end)
496 {
497 	unsigned long length;
498 
499 	/* We get the address using the kernel command line,
500 	 * but the size is extracted from the EFI tables.
501 	 * Both address and size are required for reservation
502 	 * to work properly.
503 	 */
504 
505 	if (elfcorehdr_addr >= ELFCORE_ADDR_MAX)
506 		return -EINVAL;
507 
508 	if ((length = vmcore_find_descriptor_size(elfcorehdr_addr)) == 0) {
509 		elfcorehdr_addr = ELFCORE_ADDR_MAX;
510 		return -EINVAL;
511 	}
512 
513 	*start = (unsigned long)__va(elfcorehdr_addr);
514 	*end = *start + length;
515 	return 0;
516 }
517 
518 #endif /* CONFIG_PROC_VMCORE */
519 
520 void __init
521 setup_arch (char **cmdline_p)
522 {
523 	unw_init();
524 
525 	paravirt_arch_setup_early();
526 
527 	ia64_patch_vtop((u64) __start___vtop_patchlist, (u64) __end___vtop_patchlist);
528 
529 	*cmdline_p = __va(ia64_boot_param->command_line);
530 	strlcpy(boot_command_line, *cmdline_p, COMMAND_LINE_SIZE);
531 
532 	efi_init();
533 	io_port_init();
534 
535 #ifdef CONFIG_IA64_GENERIC
536 	/* machvec needs to be parsed from the command line
537 	 * before parse_early_param() is called to ensure
538 	 * that ia64_mv is initialised before any command line
539 	 * settings may cause console setup to occur
540 	 */
541 	machvec_init_from_cmdline(*cmdline_p);
542 #endif
543 
544 	parse_early_param();
545 
546 	if (early_console_setup(*cmdline_p) == 0)
547 		mark_bsp_online();
548 
549 #ifdef CONFIG_ACPI
550 	/* Initialize the ACPI boot-time table parser */
551 	acpi_table_init();
552 # ifdef CONFIG_ACPI_NUMA
553 	acpi_numa_init();
554 	per_cpu_scan_finalize((cpus_weight(early_cpu_possible_map) == 0 ?
555 		32 : cpus_weight(early_cpu_possible_map)),
556 		additional_cpus > 0 ? additional_cpus : 0);
557 # endif
558 #else
559 # ifdef CONFIG_SMP
560 	smp_build_cpu_map();	/* happens, e.g., with the Ski simulator */
561 # endif
562 #endif /* CONFIG_APCI_BOOT */
563 
564 	find_memory();
565 
566 	/* process SAL system table: */
567 	ia64_sal_init(__va(efi.sal_systab));
568 
569 #ifdef CONFIG_ITANIUM
570 	ia64_patch_rse((u64) __start___rse_patchlist, (u64) __end___rse_patchlist);
571 #else
572 	{
573 		u64 num_phys_stacked;
574 
575 		if (ia64_pal_rse_info(&num_phys_stacked, 0) == 0 && num_phys_stacked > 96)
576 			ia64_patch_rse((u64) __start___rse_patchlist, (u64) __end___rse_patchlist);
577 	}
578 #endif
579 
580 #ifdef CONFIG_SMP
581 	cpu_physical_id(0) = hard_smp_processor_id();
582 #endif
583 
584 	cpu_init();	/* initialize the bootstrap CPU */
585 	mmu_context_init();	/* initialize context_id bitmap */
586 
587 #ifdef CONFIG_ACPI
588 	acpi_boot_init();
589 #endif
590 
591 	paravirt_banner();
592 	paravirt_arch_setup_console(cmdline_p);
593 
594 #ifdef CONFIG_VT
595 	if (!conswitchp) {
596 # if defined(CONFIG_DUMMY_CONSOLE)
597 		conswitchp = &dummy_con;
598 # endif
599 # if defined(CONFIG_VGA_CONSOLE)
600 		/*
601 		 * Non-legacy systems may route legacy VGA MMIO range to system
602 		 * memory.  vga_con probes the MMIO hole, so memory looks like
603 		 * a VGA device to it.  The EFI memory map can tell us if it's
604 		 * memory so we can avoid this problem.
605 		 */
606 		if (efi_mem_type(0xA0000) != EFI_CONVENTIONAL_MEMORY)
607 			conswitchp = &vga_con;
608 # endif
609 	}
610 #endif
611 
612 	/* enable IA-64 Machine Check Abort Handling unless disabled */
613 	if (paravirt_arch_setup_nomca())
614 		nomca = 1;
615 	if (!nomca)
616 		ia64_mca_init();
617 
618 	platform_setup(cmdline_p);
619 #ifndef CONFIG_IA64_HP_SIM
620 	check_sal_cache_flush();
621 #endif
622 	paging_init();
623 }
624 
625 /*
626  * Display cpu info for all CPUs.
627  */
628 static int
629 show_cpuinfo (struct seq_file *m, void *v)
630 {
631 #ifdef CONFIG_SMP
632 #	define lpj	c->loops_per_jiffy
633 #	define cpunum	c->cpu
634 #else
635 #	define lpj	loops_per_jiffy
636 #	define cpunum	0
637 #endif
638 	static struct {
639 		unsigned long mask;
640 		const char *feature_name;
641 	} feature_bits[] = {
642 		{ 1UL << 0, "branchlong" },
643 		{ 1UL << 1, "spontaneous deferral"},
644 		{ 1UL << 2, "16-byte atomic ops" }
645 	};
646 	char features[128], *cp, *sep;
647 	struct cpuinfo_ia64 *c = v;
648 	unsigned long mask;
649 	unsigned long proc_freq;
650 	int i, size;
651 
652 	mask = c->features;
653 
654 	/* build the feature string: */
655 	memcpy(features, "standard", 9);
656 	cp = features;
657 	size = sizeof(features);
658 	sep = "";
659 	for (i = 0; i < ARRAY_SIZE(feature_bits) && size > 1; ++i) {
660 		if (mask & feature_bits[i].mask) {
661 			cp += snprintf(cp, size, "%s%s", sep,
662 				       feature_bits[i].feature_name),
663 			sep = ", ";
664 			mask &= ~feature_bits[i].mask;
665 			size = sizeof(features) - (cp - features);
666 		}
667 	}
668 	if (mask && size > 1) {
669 		/* print unknown features as a hex value */
670 		snprintf(cp, size, "%s0x%lx", sep, mask);
671 	}
672 
673 	proc_freq = cpufreq_quick_get(cpunum);
674 	if (!proc_freq)
675 		proc_freq = c->proc_freq / 1000;
676 
677 	seq_printf(m,
678 		   "processor  : %d\n"
679 		   "vendor     : %s\n"
680 		   "arch       : IA-64\n"
681 		   "family     : %u\n"
682 		   "model      : %u\n"
683 		   "model name : %s\n"
684 		   "revision   : %u\n"
685 		   "archrev    : %u\n"
686 		   "features   : %s\n"
687 		   "cpu number : %lu\n"
688 		   "cpu regs   : %u\n"
689 		   "cpu MHz    : %lu.%03lu\n"
690 		   "itc MHz    : %lu.%06lu\n"
691 		   "BogoMIPS   : %lu.%02lu\n",
692 		   cpunum, c->vendor, c->family, c->model,
693 		   c->model_name, c->revision, c->archrev,
694 		   features, c->ppn, c->number,
695 		   proc_freq / 1000, proc_freq % 1000,
696 		   c->itc_freq / 1000000, c->itc_freq % 1000000,
697 		   lpj*HZ/500000, (lpj*HZ/5000) % 100);
698 #ifdef CONFIG_SMP
699 	seq_printf(m, "siblings   : %u\n", cpus_weight(cpu_core_map[cpunum]));
700 	if (c->socket_id != -1)
701 		seq_printf(m, "physical id: %u\n", c->socket_id);
702 	if (c->threads_per_core > 1 || c->cores_per_socket > 1)
703 		seq_printf(m,
704 			   "core id    : %u\n"
705 			   "thread id  : %u\n",
706 			   c->core_id, c->thread_id);
707 #endif
708 	seq_printf(m,"\n");
709 
710 	return 0;
711 }
712 
713 static void *
714 c_start (struct seq_file *m, loff_t *pos)
715 {
716 #ifdef CONFIG_SMP
717 	while (*pos < NR_CPUS && !cpu_isset(*pos, cpu_online_map))
718 		++*pos;
719 #endif
720 	return *pos < NR_CPUS ? cpu_data(*pos) : NULL;
721 }
722 
723 static void *
724 c_next (struct seq_file *m, void *v, loff_t *pos)
725 {
726 	++*pos;
727 	return c_start(m, pos);
728 }
729 
730 static void
731 c_stop (struct seq_file *m, void *v)
732 {
733 }
734 
735 const struct seq_operations cpuinfo_op = {
736 	.start =	c_start,
737 	.next =		c_next,
738 	.stop =		c_stop,
739 	.show =		show_cpuinfo
740 };
741 
742 #define MAX_BRANDS	8
743 static char brandname[MAX_BRANDS][128];
744 
745 static char * __cpuinit
746 get_model_name(__u8 family, __u8 model)
747 {
748 	static int overflow;
749 	char brand[128];
750 	int i;
751 
752 	memcpy(brand, "Unknown", 8);
753 	if (ia64_pal_get_brand_info(brand)) {
754 		if (family == 0x7)
755 			memcpy(brand, "Merced", 7);
756 		else if (family == 0x1f) switch (model) {
757 			case 0: memcpy(brand, "McKinley", 9); break;
758 			case 1: memcpy(brand, "Madison", 8); break;
759 			case 2: memcpy(brand, "Madison up to 9M cache", 23); break;
760 		}
761 	}
762 	for (i = 0; i < MAX_BRANDS; i++)
763 		if (strcmp(brandname[i], brand) == 0)
764 			return brandname[i];
765 	for (i = 0; i < MAX_BRANDS; i++)
766 		if (brandname[i][0] == '\0')
767 			return strcpy(brandname[i], brand);
768 	if (overflow++ == 0)
769 		printk(KERN_ERR
770 		       "%s: Table overflow. Some processor model information will be missing\n",
771 		       __func__);
772 	return "Unknown";
773 }
774 
775 static void __cpuinit
776 identify_cpu (struct cpuinfo_ia64 *c)
777 {
778 	union {
779 		unsigned long bits[5];
780 		struct {
781 			/* id 0 & 1: */
782 			char vendor[16];
783 
784 			/* id 2 */
785 			u64 ppn;		/* processor serial number */
786 
787 			/* id 3: */
788 			unsigned number		:  8;
789 			unsigned revision	:  8;
790 			unsigned model		:  8;
791 			unsigned family		:  8;
792 			unsigned archrev	:  8;
793 			unsigned reserved	: 24;
794 
795 			/* id 4: */
796 			u64 features;
797 		} field;
798 	} cpuid;
799 	pal_vm_info_1_u_t vm1;
800 	pal_vm_info_2_u_t vm2;
801 	pal_status_t status;
802 	unsigned long impl_va_msb = 50, phys_addr_size = 44;	/* Itanium defaults */
803 	int i;
804 	for (i = 0; i < 5; ++i)
805 		cpuid.bits[i] = ia64_get_cpuid(i);
806 
807 	memcpy(c->vendor, cpuid.field.vendor, 16);
808 #ifdef CONFIG_SMP
809 	c->cpu = smp_processor_id();
810 
811 	/* below default values will be overwritten  by identify_siblings()
812 	 * for Multi-Threading/Multi-Core capable CPUs
813 	 */
814 	c->threads_per_core = c->cores_per_socket = c->num_log = 1;
815 	c->socket_id = -1;
816 
817 	identify_siblings(c);
818 
819 	if (c->threads_per_core > smp_num_siblings)
820 		smp_num_siblings = c->threads_per_core;
821 #endif
822 	c->ppn = cpuid.field.ppn;
823 	c->number = cpuid.field.number;
824 	c->revision = cpuid.field.revision;
825 	c->model = cpuid.field.model;
826 	c->family = cpuid.field.family;
827 	c->archrev = cpuid.field.archrev;
828 	c->features = cpuid.field.features;
829 	c->model_name = get_model_name(c->family, c->model);
830 
831 	status = ia64_pal_vm_summary(&vm1, &vm2);
832 	if (status == PAL_STATUS_SUCCESS) {
833 		impl_va_msb = vm2.pal_vm_info_2_s.impl_va_msb;
834 		phys_addr_size = vm1.pal_vm_info_1_s.phys_add_size;
835 	}
836 	c->unimpl_va_mask = ~((7L<<61) | ((1L << (impl_va_msb + 1)) - 1));
837 	c->unimpl_pa_mask = ~((1L<<63) | ((1L << phys_addr_size) - 1));
838 }
839 
840 void __init
841 setup_per_cpu_areas (void)
842 {
843 	/* start_kernel() requires this... */
844 #ifdef CONFIG_ACPI_HOTPLUG_CPU
845 	prefill_possible_map();
846 #endif
847 }
848 
849 /*
850  * Calculate the max. cache line size.
851  *
852  * In addition, the minimum of the i-cache stride sizes is calculated for
853  * "flush_icache_range()".
854  */
855 static void __cpuinit
856 get_max_cacheline_size (void)
857 {
858 	unsigned long line_size, max = 1;
859 	u64 l, levels, unique_caches;
860         pal_cache_config_info_t cci;
861         s64 status;
862 
863         status = ia64_pal_cache_summary(&levels, &unique_caches);
864         if (status != 0) {
865                 printk(KERN_ERR "%s: ia64_pal_cache_summary() failed (status=%ld)\n",
866                        __func__, status);
867                 max = SMP_CACHE_BYTES;
868 		/* Safest setup for "flush_icache_range()" */
869 		ia64_i_cache_stride_shift = I_CACHE_STRIDE_SHIFT;
870 		goto out;
871         }
872 
873 	for (l = 0; l < levels; ++l) {
874 		status = ia64_pal_cache_config_info(l, /* cache_type (data_or_unified)= */ 2,
875 						    &cci);
876 		if (status != 0) {
877 			printk(KERN_ERR
878 			       "%s: ia64_pal_cache_config_info(l=%lu, 2) failed (status=%ld)\n",
879 			       __func__, l, status);
880 			max = SMP_CACHE_BYTES;
881 			/* The safest setup for "flush_icache_range()" */
882 			cci.pcci_stride = I_CACHE_STRIDE_SHIFT;
883 			cci.pcci_unified = 1;
884 		}
885 		line_size = 1 << cci.pcci_line_size;
886 		if (line_size > max)
887 			max = line_size;
888 		if (!cci.pcci_unified) {
889 			status = ia64_pal_cache_config_info(l,
890 						    /* cache_type (instruction)= */ 1,
891 						    &cci);
892 			if (status != 0) {
893 				printk(KERN_ERR
894 				"%s: ia64_pal_cache_config_info(l=%lu, 1) failed (status=%ld)\n",
895 					__func__, l, status);
896 				/* The safest setup for "flush_icache_range()" */
897 				cci.pcci_stride = I_CACHE_STRIDE_SHIFT;
898 			}
899 		}
900 		if (cci.pcci_stride < ia64_i_cache_stride_shift)
901 			ia64_i_cache_stride_shift = cci.pcci_stride;
902 	}
903   out:
904 	if (max > ia64_max_cacheline_size)
905 		ia64_max_cacheline_size = max;
906 }
907 
908 /*
909  * cpu_init() initializes state that is per-CPU.  This function acts
910  * as a 'CPU state barrier', nothing should get across.
911  */
912 void __cpuinit
913 cpu_init (void)
914 {
915 	extern void __cpuinit ia64_mmu_init (void *);
916 	static unsigned long max_num_phys_stacked = IA64_NUM_PHYS_STACK_REG;
917 	unsigned long num_phys_stacked;
918 	pal_vm_info_2_u_t vmi;
919 	unsigned int max_ctx;
920 	struct cpuinfo_ia64 *cpu_info;
921 	void *cpu_data;
922 
923 	cpu_data = per_cpu_init();
924 #ifdef CONFIG_SMP
925 	/*
926 	 * insert boot cpu into sibling and core mapes
927 	 * (must be done after per_cpu area is setup)
928 	 */
929 	if (smp_processor_id() == 0) {
930 		cpu_set(0, per_cpu(cpu_sibling_map, 0));
931 		cpu_set(0, cpu_core_map[0]);
932 	} else {
933 		/*
934 		 * Set ar.k3 so that assembly code in MCA handler can compute
935 		 * physical addresses of per cpu variables with a simple:
936 		 *   phys = ar.k3 + &per_cpu_var
937 		 * and the alt-dtlb-miss handler can set per-cpu mapping into
938 		 * the TLB when needed. head.S already did this for cpu0.
939 		 */
940 		ia64_set_kr(IA64_KR_PER_CPU_DATA,
941 			    ia64_tpa(cpu_data) - (long) __per_cpu_start);
942 	}
943 #endif
944 
945 	get_max_cacheline_size();
946 
947 	/*
948 	 * We can't pass "local_cpu_data" to identify_cpu() because we haven't called
949 	 * ia64_mmu_init() yet.  And we can't call ia64_mmu_init() first because it
950 	 * depends on the data returned by identify_cpu().  We break the dependency by
951 	 * accessing cpu_data() through the canonical per-CPU address.
952 	 */
953 	cpu_info = cpu_data + ((char *) &__ia64_per_cpu_var(cpu_info) - __per_cpu_start);
954 	identify_cpu(cpu_info);
955 
956 #ifdef CONFIG_MCKINLEY
957 	{
958 #		define FEATURE_SET 16
959 		struct ia64_pal_retval iprv;
960 
961 		if (cpu_info->family == 0x1f) {
962 			PAL_CALL_PHYS(iprv, PAL_PROC_GET_FEATURES, 0, FEATURE_SET, 0);
963 			if ((iprv.status == 0) && (iprv.v0 & 0x80) && (iprv.v2 & 0x80))
964 				PAL_CALL_PHYS(iprv, PAL_PROC_SET_FEATURES,
965 				              (iprv.v1 | 0x80), FEATURE_SET, 0);
966 		}
967 	}
968 #endif
969 
970 	/* Clear the stack memory reserved for pt_regs: */
971 	memset(task_pt_regs(current), 0, sizeof(struct pt_regs));
972 
973 	ia64_set_kr(IA64_KR_FPU_OWNER, 0);
974 
975 	/*
976 	 * Initialize the page-table base register to a global
977 	 * directory with all zeroes.  This ensure that we can handle
978 	 * TLB-misses to user address-space even before we created the
979 	 * first user address-space.  This may happen, e.g., due to
980 	 * aggressive use of lfetch.fault.
981 	 */
982 	ia64_set_kr(IA64_KR_PT_BASE, __pa(ia64_imva(empty_zero_page)));
983 
984 	/*
985 	 * Initialize default control register to defer speculative faults except
986 	 * for those arising from TLB misses, which are not deferred.  The
987 	 * kernel MUST NOT depend on a particular setting of these bits (in other words,
988 	 * the kernel must have recovery code for all speculative accesses).  Turn on
989 	 * dcr.lc as per recommendation by the architecture team.  Most IA-32 apps
990 	 * shouldn't be affected by this (moral: keep your ia32 locks aligned and you'll
991 	 * be fine).
992 	 */
993 	ia64_setreg(_IA64_REG_CR_DCR,  (  IA64_DCR_DP | IA64_DCR_DK | IA64_DCR_DX | IA64_DCR_DR
994 					| IA64_DCR_DA | IA64_DCR_DD | IA64_DCR_LC));
995 	atomic_inc(&init_mm.mm_count);
996 	current->active_mm = &init_mm;
997 	if (current->mm)
998 		BUG();
999 
1000 	ia64_mmu_init(ia64_imva(cpu_data));
1001 	ia64_mca_cpu_init(ia64_imva(cpu_data));
1002 
1003 #ifdef CONFIG_IA32_SUPPORT
1004 	ia32_cpu_init();
1005 #endif
1006 
1007 	/* Clear ITC to eliminate sched_clock() overflows in human time.  */
1008 	ia64_set_itc(0);
1009 
1010 	/* disable all local interrupt sources: */
1011 	ia64_set_itv(1 << 16);
1012 	ia64_set_lrr0(1 << 16);
1013 	ia64_set_lrr1(1 << 16);
1014 	ia64_setreg(_IA64_REG_CR_PMV, 1 << 16);
1015 	ia64_setreg(_IA64_REG_CR_CMCV, 1 << 16);
1016 
1017 	/* clear TPR & XTP to enable all interrupt classes: */
1018 	ia64_setreg(_IA64_REG_CR_TPR, 0);
1019 
1020 	/* Clear any pending interrupts left by SAL/EFI */
1021 	while (ia64_get_ivr() != IA64_SPURIOUS_INT_VECTOR)
1022 		ia64_eoi();
1023 
1024 #ifdef CONFIG_SMP
1025 	normal_xtp();
1026 #endif
1027 
1028 	/* set ia64_ctx.max_rid to the maximum RID that is supported by all CPUs: */
1029 	if (ia64_pal_vm_summary(NULL, &vmi) == 0) {
1030 		max_ctx = (1U << (vmi.pal_vm_info_2_s.rid_size - 3)) - 1;
1031 		setup_ptcg_sem(vmi.pal_vm_info_2_s.max_purges, NPTCG_FROM_PAL);
1032 	} else {
1033 		printk(KERN_WARNING "cpu_init: PAL VM summary failed, assuming 18 RID bits\n");
1034 		max_ctx = (1U << 15) - 1;	/* use architected minimum */
1035 	}
1036 	while (max_ctx < ia64_ctx.max_ctx) {
1037 		unsigned int old = ia64_ctx.max_ctx;
1038 		if (cmpxchg(&ia64_ctx.max_ctx, old, max_ctx) == old)
1039 			break;
1040 	}
1041 
1042 	if (ia64_pal_rse_info(&num_phys_stacked, NULL) != 0) {
1043 		printk(KERN_WARNING "cpu_init: PAL RSE info failed; assuming 96 physical "
1044 		       "stacked regs\n");
1045 		num_phys_stacked = 96;
1046 	}
1047 	/* size of physical stacked register partition plus 8 bytes: */
1048 	if (num_phys_stacked > max_num_phys_stacked) {
1049 		ia64_patch_phys_stack_reg(num_phys_stacked*8 + 8);
1050 		max_num_phys_stacked = num_phys_stacked;
1051 	}
1052 	platform_cpu_init();
1053 	pm_idle = default_idle;
1054 }
1055 
1056 void __init
1057 check_bugs (void)
1058 {
1059 	ia64_patch_mckinley_e9((unsigned long) __start___mckinley_e9_bundles,
1060 			       (unsigned long) __end___mckinley_e9_bundles);
1061 }
1062 
1063 static int __init run_dmi_scan(void)
1064 {
1065 	dmi_scan_machine();
1066 	return 0;
1067 }
1068 core_initcall(run_dmi_scan);
1069