xref: /openbmc/linux/arch/ia64/kernel/setup.c (revision 08f9530a)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Architecture-specific setup.
4  *
5  * Copyright (C) 1998-2001, 2003-2004 Hewlett-Packard Co
6  *	David Mosberger-Tang <davidm@hpl.hp.com>
7  *	Stephane Eranian <eranian@hpl.hp.com>
8  * Copyright (C) 2000, 2004 Intel Corp
9  * 	Rohit Seth <rohit.seth@intel.com>
10  * 	Suresh Siddha <suresh.b.siddha@intel.com>
11  * 	Gordon Jin <gordon.jin@intel.com>
12  * Copyright (C) 1999 VA Linux Systems
13  * Copyright (C) 1999 Walt Drummond <drummond@valinux.com>
14  *
15  * 12/26/04 S.Siddha, G.Jin, R.Seth
16  *			Add multi-threading and multi-core detection
17  * 11/12/01 D.Mosberger Convert get_cpuinfo() to seq_file based show_cpuinfo().
18  * 04/04/00 D.Mosberger renamed cpu_initialized to cpu_online_map
19  * 03/31/00 R.Seth	cpu_initialized and current->processor fixes
20  * 02/04/00 D.Mosberger	some more get_cpuinfo fixes...
21  * 02/01/00 R.Seth	fixed get_cpuinfo for SMP
22  * 01/07/99 S.Eranian	added the support for command line argument
23  * 06/24/99 W.Drummond	added boot_cpu_data.
24  * 05/28/05 Z. Menyhart	Dynamic stride size for "flush_icache_range()"
25  */
26 #include <linux/module.h>
27 #include <linux/init.h>
28 
29 #include <linux/acpi.h>
30 #include <linux/console.h>
31 #include <linux/delay.h>
32 #include <linux/cpu.h>
33 #include <linux/kernel.h>
34 #include <linux/memblock.h>
35 #include <linux/reboot.h>
36 #include <linux/sched/mm.h>
37 #include <linux/sched/clock.h>
38 #include <linux/sched/task_stack.h>
39 #include <linux/seq_file.h>
40 #include <linux/string.h>
41 #include <linux/threads.h>
42 #include <linux/screen_info.h>
43 #include <linux/dmi.h>
44 #include <linux/serial.h>
45 #include <linux/serial_core.h>
46 #include <linux/efi.h>
47 #include <linux/initrd.h>
48 #include <linux/pm.h>
49 #include <linux/cpufreq.h>
50 #include <linux/kexec.h>
51 #include <linux/crash_dump.h>
52 
53 #include <asm/machvec.h>
54 #include <asm/mca.h>
55 #include <asm/meminit.h>
56 #include <asm/page.h>
57 #include <asm/patch.h>
58 #include <asm/pgtable.h>
59 #include <asm/processor.h>
60 #include <asm/sal.h>
61 #include <asm/sections.h>
62 #include <asm/setup.h>
63 #include <asm/smp.h>
64 #include <asm/tlbflush.h>
65 #include <asm/unistd.h>
66 #include <asm/hpsim.h>
67 
68 #if defined(CONFIG_SMP) && (IA64_CPU_SIZE > PAGE_SIZE)
69 # error "struct cpuinfo_ia64 too big!"
70 #endif
71 
72 #ifdef CONFIG_SMP
73 unsigned long __per_cpu_offset[NR_CPUS];
74 EXPORT_SYMBOL(__per_cpu_offset);
75 #endif
76 
77 DEFINE_PER_CPU(struct cpuinfo_ia64, ia64_cpu_info);
78 EXPORT_SYMBOL(ia64_cpu_info);
79 DEFINE_PER_CPU(unsigned long, local_per_cpu_offset);
80 #ifdef CONFIG_SMP
81 EXPORT_SYMBOL(local_per_cpu_offset);
82 #endif
83 unsigned long ia64_cycles_per_usec;
84 struct ia64_boot_param *ia64_boot_param;
85 struct screen_info screen_info;
86 unsigned long vga_console_iobase;
87 unsigned long vga_console_membase;
88 
89 static struct resource data_resource = {
90 	.name	= "Kernel data",
91 	.flags	= IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
92 };
93 
94 static struct resource code_resource = {
95 	.name	= "Kernel code",
96 	.flags	= IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
97 };
98 
99 static struct resource bss_resource = {
100 	.name	= "Kernel bss",
101 	.flags	= IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
102 };
103 
104 unsigned long ia64_max_cacheline_size;
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  * "clflush_cache_range()" needs to know what processor dependent stride size to
120  * use when it flushes cache lines including both d-cache and i-cache.
121  */
122 /* Safest way to go: 32 bytes by 32 bytes */
123 #define	CACHE_STRIDE_SHIFT	5
124 unsigned long ia64_cache_stride_shift = ~0;
125 
126 /*
127  * We use a special marker for the end of memory and it uses the extra (+1) slot
128  */
129 struct rsvd_region rsvd_region[IA64_MAX_RSVD_REGIONS + 1] __initdata;
130 int num_rsvd_regions __initdata;
131 
132 
133 /*
134  * Filter incoming memory segments based on the primitive map created from the boot
135  * parameters. Segments contained in the map are removed from the memory ranges. A
136  * caller-specified function is called with the memory ranges that remain after filtering.
137  * This routine does not assume the incoming segments are sorted.
138  */
139 int __init
140 filter_rsvd_memory (u64 start, u64 end, void *arg)
141 {
142 	u64 range_start, range_end, prev_start;
143 	void (*func)(unsigned long, unsigned long, int);
144 	int i;
145 
146 #if IGNORE_PFN0
147 	if (start == PAGE_OFFSET) {
148 		printk(KERN_WARNING "warning: skipping physical page 0\n");
149 		start += PAGE_SIZE;
150 		if (start >= end) return 0;
151 	}
152 #endif
153 	/*
154 	 * lowest possible address(walker uses virtual)
155 	 */
156 	prev_start = PAGE_OFFSET;
157 	func = arg;
158 
159 	for (i = 0; i < num_rsvd_regions; ++i) {
160 		range_start = max(start, prev_start);
161 		range_end   = min(end, rsvd_region[i].start);
162 
163 		if (range_start < range_end)
164 			call_pernode_memory(__pa(range_start), range_end - range_start, func);
165 
166 		/* nothing more available in this segment */
167 		if (range_end == end) return 0;
168 
169 		prev_start = rsvd_region[i].end;
170 	}
171 	/* end of memory marker allows full processing inside loop body */
172 	return 0;
173 }
174 
175 /*
176  * Similar to "filter_rsvd_memory()", but the reserved memory ranges
177  * are not filtered out.
178  */
179 int __init
180 filter_memory(u64 start, u64 end, void *arg)
181 {
182 	void (*func)(unsigned long, unsigned long, int);
183 
184 #if IGNORE_PFN0
185 	if (start == PAGE_OFFSET) {
186 		printk(KERN_WARNING "warning: skipping physical page 0\n");
187 		start += PAGE_SIZE;
188 		if (start >= end)
189 			return 0;
190 	}
191 #endif
192 	func = arg;
193 	if (start < end)
194 		call_pernode_memory(__pa(start), end - start, func);
195 	return 0;
196 }
197 
198 static void __init
199 sort_regions (struct rsvd_region *rsvd_region, int max)
200 {
201 	int j;
202 
203 	/* simple bubble sorting */
204 	while (max--) {
205 		for (j = 0; j < max; ++j) {
206 			if (rsvd_region[j].start > rsvd_region[j+1].start) {
207 				struct rsvd_region tmp;
208 				tmp = rsvd_region[j];
209 				rsvd_region[j] = rsvd_region[j + 1];
210 				rsvd_region[j + 1] = tmp;
211 			}
212 		}
213 	}
214 }
215 
216 /* merge overlaps */
217 static int __init
218 merge_regions (struct rsvd_region *rsvd_region, int max)
219 {
220 	int i;
221 	for (i = 1; i < max; ++i) {
222 		if (rsvd_region[i].start >= rsvd_region[i-1].end)
223 			continue;
224 		if (rsvd_region[i].end > rsvd_region[i-1].end)
225 			rsvd_region[i-1].end = rsvd_region[i].end;
226 		--max;
227 		memmove(&rsvd_region[i], &rsvd_region[i+1],
228 			(max - i) * sizeof(struct rsvd_region));
229 	}
230 	return max;
231 }
232 
233 /*
234  * Request address space for all standard resources
235  */
236 static int __init register_memory(void)
237 {
238 	code_resource.start = ia64_tpa(_text);
239 	code_resource.end   = ia64_tpa(_etext) - 1;
240 	data_resource.start = ia64_tpa(_etext);
241 	data_resource.end   = ia64_tpa(_edata) - 1;
242 	bss_resource.start  = ia64_tpa(__bss_start);
243 	bss_resource.end    = ia64_tpa(_end) - 1;
244 	efi_initialize_iomem_resources(&code_resource, &data_resource,
245 			&bss_resource);
246 
247 	return 0;
248 }
249 
250 __initcall(register_memory);
251 
252 
253 #ifdef CONFIG_KEXEC
254 
255 /*
256  * This function checks if the reserved crashkernel is allowed on the specific
257  * IA64 machine flavour. Machines without an IO TLB use swiotlb and require
258  * some memory below 4 GB (i.e. in 32 bit area), see the implementation of
259  * lib/swiotlb.c. The hpzx1 architecture has an IO TLB but cannot use that
260  * in kdump case. See the comment in sba_init() in sba_iommu.c.
261  *
262  * So, the only machvec that really supports loading the kdump kernel
263  * over 4 GB is "sn2".
264  */
265 static int __init check_crashkernel_memory(unsigned long pbase, size_t size)
266 {
267 	if (ia64_platform_is("sn2") || ia64_platform_is("uv"))
268 		return 1;
269 	else
270 		return pbase < (1UL << 32);
271 }
272 
273 static void __init setup_crashkernel(unsigned long total, int *n)
274 {
275 	unsigned long long base = 0, size = 0;
276 	int ret;
277 
278 	ret = parse_crashkernel(boot_command_line, total,
279 			&size, &base);
280 	if (ret == 0 && size > 0) {
281 		if (!base) {
282 			sort_regions(rsvd_region, *n);
283 			*n = merge_regions(rsvd_region, *n);
284 			base = kdump_find_rsvd_region(size,
285 					rsvd_region, *n);
286 		}
287 
288 		if (!check_crashkernel_memory(base, size)) {
289 			pr_warning("crashkernel: There would be kdump memory "
290 				"at %ld GB but this is unusable because it "
291 				"must\nbe below 4 GB. Change the memory "
292 				"configuration of the machine.\n",
293 				(unsigned long)(base >> 30));
294 			return;
295 		}
296 
297 		if (base != ~0UL) {
298 			printk(KERN_INFO "Reserving %ldMB of memory at %ldMB "
299 					"for crashkernel (System RAM: %ldMB)\n",
300 					(unsigned long)(size >> 20),
301 					(unsigned long)(base >> 20),
302 					(unsigned long)(total >> 20));
303 			rsvd_region[*n].start =
304 				(unsigned long)__va(base);
305 			rsvd_region[*n].end =
306 				(unsigned long)__va(base + size);
307 			(*n)++;
308 			crashk_res.start = base;
309 			crashk_res.end = base + size - 1;
310 		}
311 	}
312 	efi_memmap_res.start = ia64_boot_param->efi_memmap;
313 	efi_memmap_res.end = efi_memmap_res.start +
314 		ia64_boot_param->efi_memmap_size;
315 	boot_param_res.start = __pa(ia64_boot_param);
316 	boot_param_res.end = boot_param_res.start +
317 		sizeof(*ia64_boot_param);
318 }
319 #else
320 static inline void __init setup_crashkernel(unsigned long total, int *n)
321 {}
322 #endif
323 
324 /**
325  * reserve_memory - setup reserved memory areas
326  *
327  * Setup the reserved memory areas set aside for the boot parameters,
328  * initrd, etc.  There are currently %IA64_MAX_RSVD_REGIONS defined,
329  * see arch/ia64/include/asm/meminit.h if you need to define more.
330  */
331 void __init
332 reserve_memory (void)
333 {
334 	int n = 0;
335 	unsigned long total_memory;
336 
337 	/*
338 	 * none of the entries in this table overlap
339 	 */
340 	rsvd_region[n].start = (unsigned long) ia64_boot_param;
341 	rsvd_region[n].end   = rsvd_region[n].start + sizeof(*ia64_boot_param);
342 	n++;
343 
344 	rsvd_region[n].start = (unsigned long) __va(ia64_boot_param->efi_memmap);
345 	rsvd_region[n].end   = rsvd_region[n].start + ia64_boot_param->efi_memmap_size;
346 	n++;
347 
348 	rsvd_region[n].start = (unsigned long) __va(ia64_boot_param->command_line);
349 	rsvd_region[n].end   = (rsvd_region[n].start
350 				+ strlen(__va(ia64_boot_param->command_line)) + 1);
351 	n++;
352 
353 	rsvd_region[n].start = (unsigned long) ia64_imva((void *)KERNEL_START);
354 	rsvd_region[n].end   = (unsigned long) ia64_imva(_end);
355 	n++;
356 
357 #ifdef CONFIG_BLK_DEV_INITRD
358 	if (ia64_boot_param->initrd_start) {
359 		rsvd_region[n].start = (unsigned long)__va(ia64_boot_param->initrd_start);
360 		rsvd_region[n].end   = rsvd_region[n].start + ia64_boot_param->initrd_size;
361 		n++;
362 	}
363 #endif
364 
365 #ifdef CONFIG_CRASH_DUMP
366 	if (reserve_elfcorehdr(&rsvd_region[n].start,
367 			       &rsvd_region[n].end) == 0)
368 		n++;
369 #endif
370 
371 	total_memory = efi_memmap_init(&rsvd_region[n].start, &rsvd_region[n].end);
372 	n++;
373 
374 	setup_crashkernel(total_memory, &n);
375 
376 	/* end of memory marker */
377 	rsvd_region[n].start = ~0UL;
378 	rsvd_region[n].end   = ~0UL;
379 	n++;
380 
381 	num_rsvd_regions = n;
382 	BUG_ON(IA64_MAX_RSVD_REGIONS + 1 < n);
383 
384 	sort_regions(rsvd_region, num_rsvd_regions);
385 	num_rsvd_regions = merge_regions(rsvd_region, num_rsvd_regions);
386 
387 	/* reserve all regions except the end of memory marker with memblock */
388 	for (n = 0; n < num_rsvd_regions - 1; n++) {
389 		struct rsvd_region *region = &rsvd_region[n];
390 		phys_addr_t addr = __pa(region->start);
391 		phys_addr_t size = region->end - region->start;
392 
393 		memblock_reserve(addr, size);
394 	}
395 }
396 
397 /**
398  * find_initrd - get initrd parameters from the boot parameter structure
399  *
400  * Grab the initrd start and end from the boot parameter struct given us by
401  * the boot loader.
402  */
403 void __init
404 find_initrd (void)
405 {
406 #ifdef CONFIG_BLK_DEV_INITRD
407 	if (ia64_boot_param->initrd_start) {
408 		initrd_start = (unsigned long)__va(ia64_boot_param->initrd_start);
409 		initrd_end   = initrd_start+ia64_boot_param->initrd_size;
410 
411 		printk(KERN_INFO "Initial ramdisk at: 0x%lx (%llu bytes)\n",
412 		       initrd_start, ia64_boot_param->initrd_size);
413 	}
414 #endif
415 }
416 
417 static void __init
418 io_port_init (void)
419 {
420 	unsigned long phys_iobase;
421 
422 	/*
423 	 * Set `iobase' based on the EFI memory map or, failing that, the
424 	 * value firmware left in ar.k0.
425 	 *
426 	 * Note that in ia32 mode, IN/OUT instructions use ar.k0 to compute
427 	 * the port's virtual address, so ia32_load_state() loads it with a
428 	 * user virtual address.  But in ia64 mode, glibc uses the
429 	 * *physical* address in ar.k0 to mmap the appropriate area from
430 	 * /dev/mem, and the inX()/outX() interfaces use MMIO.  In both
431 	 * cases, user-mode can only use the legacy 0-64K I/O port space.
432 	 *
433 	 * ar.k0 is not involved in kernel I/O port accesses, which can use
434 	 * any of the I/O port spaces and are done via MMIO using the
435 	 * virtual mmio_base from the appropriate io_space[].
436 	 */
437 	phys_iobase = efi_get_iobase();
438 	if (!phys_iobase) {
439 		phys_iobase = ia64_get_kr(IA64_KR_IO_BASE);
440 		printk(KERN_INFO "No I/O port range found in EFI memory map, "
441 			"falling back to AR.KR0 (0x%lx)\n", phys_iobase);
442 	}
443 	ia64_iobase = (unsigned long) ioremap(phys_iobase, 0);
444 	ia64_set_kr(IA64_KR_IO_BASE, __pa(ia64_iobase));
445 
446 	/* setup legacy IO port space */
447 	io_space[0].mmio_base = ia64_iobase;
448 	io_space[0].sparse = 1;
449 	num_io_spaces = 1;
450 }
451 
452 /**
453  * early_console_setup - setup debugging console
454  *
455  * Consoles started here require little enough setup that we can start using
456  * them very early in the boot process, either right after the machine
457  * vector initialization, or even before if the drivers can detect their hw.
458  *
459  * Returns non-zero if a console couldn't be setup.
460  */
461 static inline int __init
462 early_console_setup (char *cmdline)
463 {
464 	int earlycons = 0;
465 
466 #ifdef CONFIG_EFI_PCDP
467 	if (!efi_setup_pcdp_console(cmdline))
468 		earlycons++;
469 #endif
470 	if (!simcons_register())
471 		earlycons++;
472 
473 	return (earlycons) ? 0 : -1;
474 }
475 
476 static inline void
477 mark_bsp_online (void)
478 {
479 #ifdef CONFIG_SMP
480 	/* If we register an early console, allow CPU 0 to printk */
481 	set_cpu_online(smp_processor_id(), true);
482 #endif
483 }
484 
485 static __initdata int nomca;
486 static __init int setup_nomca(char *s)
487 {
488 	nomca = 1;
489 	return 0;
490 }
491 early_param("nomca", setup_nomca);
492 
493 #ifdef CONFIG_CRASH_DUMP
494 int __init reserve_elfcorehdr(u64 *start, u64 *end)
495 {
496 	u64 length;
497 
498 	/* We get the address using the kernel command line,
499 	 * but the size is extracted from the EFI tables.
500 	 * Both address and size are required for reservation
501 	 * to work properly.
502 	 */
503 
504 	if (!is_vmcore_usable())
505 		return -EINVAL;
506 
507 	if ((length = vmcore_find_descriptor_size(elfcorehdr_addr)) == 0) {
508 		vmcore_unusable();
509 		return -EINVAL;
510 	}
511 
512 	*start = (unsigned long)__va(elfcorehdr_addr);
513 	*end = *start + length;
514 	return 0;
515 }
516 
517 #endif /* CONFIG_PROC_VMCORE */
518 
519 void __init
520 setup_arch (char **cmdline_p)
521 {
522 	unw_init();
523 
524 	ia64_patch_vtop((u64) __start___vtop_patchlist, (u64) __end___vtop_patchlist);
525 
526 	*cmdline_p = __va(ia64_boot_param->command_line);
527 	strlcpy(boot_command_line, *cmdline_p, COMMAND_LINE_SIZE);
528 
529 	efi_init();
530 	io_port_init();
531 
532 #ifdef CONFIG_IA64_GENERIC
533 	/* machvec needs to be parsed from the command line
534 	 * before parse_early_param() is called to ensure
535 	 * that ia64_mv is initialised before any command line
536 	 * settings may cause console setup to occur
537 	 */
538 	machvec_init_from_cmdline(*cmdline_p);
539 #endif
540 
541 	parse_early_param();
542 
543 	if (early_console_setup(*cmdline_p) == 0)
544 		mark_bsp_online();
545 
546 #ifdef CONFIG_ACPI
547 	/* Initialize the ACPI boot-time table parser */
548 	acpi_table_init();
549 	early_acpi_boot_init();
550 # ifdef CONFIG_ACPI_NUMA
551 	acpi_numa_init();
552 	acpi_numa_fixup();
553 #  ifdef CONFIG_ACPI_HOTPLUG_CPU
554 	prefill_possible_map();
555 #  endif
556 	per_cpu_scan_finalize((cpumask_weight(&early_cpu_possible_map) == 0 ?
557 		32 : cpumask_weight(&early_cpu_possible_map)),
558 		additional_cpus > 0 ? additional_cpus : 0);
559 # endif
560 #endif /* CONFIG_APCI_BOOT */
561 
562 #ifdef CONFIG_SMP
563 	smp_build_cpu_map();
564 #endif
565 	find_memory();
566 
567 	/* process SAL system table: */
568 	ia64_sal_init(__va(efi.sal_systab));
569 
570 #ifdef CONFIG_ITANIUM
571 	ia64_patch_rse((u64) __start___rse_patchlist, (u64) __end___rse_patchlist);
572 #else
573 	{
574 		unsigned long num_phys_stacked;
575 
576 		if (ia64_pal_rse_info(&num_phys_stacked, 0) == 0 && num_phys_stacked > 96)
577 			ia64_patch_rse((u64) __start___rse_patchlist, (u64) __end___rse_patchlist);
578 	}
579 #endif
580 
581 #ifdef CONFIG_SMP
582 	cpu_physical_id(0) = hard_smp_processor_id();
583 #endif
584 
585 	cpu_init();	/* initialize the bootstrap CPU */
586 	mmu_context_init();	/* initialize context_id bitmap */
587 
588 #ifdef CONFIG_VT
589 	if (!conswitchp) {
590 # if defined(CONFIG_DUMMY_CONSOLE)
591 		conswitchp = &dummy_con;
592 # endif
593 # if defined(CONFIG_VGA_CONSOLE)
594 		/*
595 		 * Non-legacy systems may route legacy VGA MMIO range to system
596 		 * memory.  vga_con probes the MMIO hole, so memory looks like
597 		 * a VGA device to it.  The EFI memory map can tell us if it's
598 		 * memory so we can avoid this problem.
599 		 */
600 		if (efi_mem_type(0xA0000) != EFI_CONVENTIONAL_MEMORY)
601 			conswitchp = &vga_con;
602 # endif
603 	}
604 #endif
605 
606 	/* enable IA-64 Machine Check Abort Handling unless disabled */
607 	if (!nomca)
608 		ia64_mca_init();
609 
610 	platform_setup(cmdline_p);
611 #ifndef CONFIG_IA64_HP_SIM
612 	check_sal_cache_flush();
613 #endif
614 	paging_init();
615 
616 	clear_sched_clock_stable();
617 }
618 
619 /*
620  * Display cpu info for all CPUs.
621  */
622 static int
623 show_cpuinfo (struct seq_file *m, void *v)
624 {
625 #ifdef CONFIG_SMP
626 #	define lpj	c->loops_per_jiffy
627 #	define cpunum	c->cpu
628 #else
629 #	define lpj	loops_per_jiffy
630 #	define cpunum	0
631 #endif
632 	static struct {
633 		unsigned long mask;
634 		const char *feature_name;
635 	} feature_bits[] = {
636 		{ 1UL << 0, "branchlong" },
637 		{ 1UL << 1, "spontaneous deferral"},
638 		{ 1UL << 2, "16-byte atomic ops" }
639 	};
640 	char features[128], *cp, *sep;
641 	struct cpuinfo_ia64 *c = v;
642 	unsigned long mask;
643 	unsigned long proc_freq;
644 	int i, size;
645 
646 	mask = c->features;
647 
648 	/* build the feature string: */
649 	memcpy(features, "standard", 9);
650 	cp = features;
651 	size = sizeof(features);
652 	sep = "";
653 	for (i = 0; i < ARRAY_SIZE(feature_bits) && size > 1; ++i) {
654 		if (mask & feature_bits[i].mask) {
655 			cp += snprintf(cp, size, "%s%s", sep,
656 				       feature_bits[i].feature_name),
657 			sep = ", ";
658 			mask &= ~feature_bits[i].mask;
659 			size = sizeof(features) - (cp - features);
660 		}
661 	}
662 	if (mask && size > 1) {
663 		/* print unknown features as a hex value */
664 		snprintf(cp, size, "%s0x%lx", sep, mask);
665 	}
666 
667 	proc_freq = cpufreq_quick_get(cpunum);
668 	if (!proc_freq)
669 		proc_freq = c->proc_freq / 1000;
670 
671 	seq_printf(m,
672 		   "processor  : %d\n"
673 		   "vendor     : %s\n"
674 		   "arch       : IA-64\n"
675 		   "family     : %u\n"
676 		   "model      : %u\n"
677 		   "model name : %s\n"
678 		   "revision   : %u\n"
679 		   "archrev    : %u\n"
680 		   "features   : %s\n"
681 		   "cpu number : %lu\n"
682 		   "cpu regs   : %u\n"
683 		   "cpu MHz    : %lu.%03lu\n"
684 		   "itc MHz    : %lu.%06lu\n"
685 		   "BogoMIPS   : %lu.%02lu\n",
686 		   cpunum, c->vendor, c->family, c->model,
687 		   c->model_name, c->revision, c->archrev,
688 		   features, c->ppn, c->number,
689 		   proc_freq / 1000, proc_freq % 1000,
690 		   c->itc_freq / 1000000, c->itc_freq % 1000000,
691 		   lpj*HZ/500000, (lpj*HZ/5000) % 100);
692 #ifdef CONFIG_SMP
693 	seq_printf(m, "siblings   : %u\n",
694 		   cpumask_weight(&cpu_core_map[cpunum]));
695 	if (c->socket_id != -1)
696 		seq_printf(m, "physical id: %u\n", c->socket_id);
697 	if (c->threads_per_core > 1 || c->cores_per_socket > 1)
698 		seq_printf(m,
699 			   "core id    : %u\n"
700 			   "thread id  : %u\n",
701 			   c->core_id, c->thread_id);
702 #endif
703 	seq_printf(m,"\n");
704 
705 	return 0;
706 }
707 
708 static void *
709 c_start (struct seq_file *m, loff_t *pos)
710 {
711 #ifdef CONFIG_SMP
712 	while (*pos < nr_cpu_ids && !cpu_online(*pos))
713 		++*pos;
714 #endif
715 	return *pos < nr_cpu_ids ? cpu_data(*pos) : NULL;
716 }
717 
718 static void *
719 c_next (struct seq_file *m, void *v, loff_t *pos)
720 {
721 	++*pos;
722 	return c_start(m, pos);
723 }
724 
725 static void
726 c_stop (struct seq_file *m, void *v)
727 {
728 }
729 
730 const struct seq_operations cpuinfo_op = {
731 	.start =	c_start,
732 	.next =		c_next,
733 	.stop =		c_stop,
734 	.show =		show_cpuinfo
735 };
736 
737 #define MAX_BRANDS	8
738 static char brandname[MAX_BRANDS][128];
739 
740 static char *
741 get_model_name(__u8 family, __u8 model)
742 {
743 	static int overflow;
744 	char brand[128];
745 	int i;
746 
747 	memcpy(brand, "Unknown", 8);
748 	if (ia64_pal_get_brand_info(brand)) {
749 		if (family == 0x7)
750 			memcpy(brand, "Merced", 7);
751 		else if (family == 0x1f) switch (model) {
752 			case 0: memcpy(brand, "McKinley", 9); break;
753 			case 1: memcpy(brand, "Madison", 8); break;
754 			case 2: memcpy(brand, "Madison up to 9M cache", 23); break;
755 		}
756 	}
757 	for (i = 0; i < MAX_BRANDS; i++)
758 		if (strcmp(brandname[i], brand) == 0)
759 			return brandname[i];
760 	for (i = 0; i < MAX_BRANDS; i++)
761 		if (brandname[i][0] == '\0')
762 			return strcpy(brandname[i], brand);
763 	if (overflow++ == 0)
764 		printk(KERN_ERR
765 		       "%s: Table overflow. Some processor model information will be missing\n",
766 		       __func__);
767 	return "Unknown";
768 }
769 
770 static void
771 identify_cpu (struct cpuinfo_ia64 *c)
772 {
773 	union {
774 		unsigned long bits[5];
775 		struct {
776 			/* id 0 & 1: */
777 			char vendor[16];
778 
779 			/* id 2 */
780 			u64 ppn;		/* processor serial number */
781 
782 			/* id 3: */
783 			unsigned number		:  8;
784 			unsigned revision	:  8;
785 			unsigned model		:  8;
786 			unsigned family		:  8;
787 			unsigned archrev	:  8;
788 			unsigned reserved	: 24;
789 
790 			/* id 4: */
791 			u64 features;
792 		} field;
793 	} cpuid;
794 	pal_vm_info_1_u_t vm1;
795 	pal_vm_info_2_u_t vm2;
796 	pal_status_t status;
797 	unsigned long impl_va_msb = 50, phys_addr_size = 44;	/* Itanium defaults */
798 	int i;
799 	for (i = 0; i < 5; ++i)
800 		cpuid.bits[i] = ia64_get_cpuid(i);
801 
802 	memcpy(c->vendor, cpuid.field.vendor, 16);
803 #ifdef CONFIG_SMP
804 	c->cpu = smp_processor_id();
805 
806 	/* below default values will be overwritten  by identify_siblings()
807 	 * for Multi-Threading/Multi-Core capable CPUs
808 	 */
809 	c->threads_per_core = c->cores_per_socket = c->num_log = 1;
810 	c->socket_id = -1;
811 
812 	identify_siblings(c);
813 
814 	if (c->threads_per_core > smp_num_siblings)
815 		smp_num_siblings = c->threads_per_core;
816 #endif
817 	c->ppn = cpuid.field.ppn;
818 	c->number = cpuid.field.number;
819 	c->revision = cpuid.field.revision;
820 	c->model = cpuid.field.model;
821 	c->family = cpuid.field.family;
822 	c->archrev = cpuid.field.archrev;
823 	c->features = cpuid.field.features;
824 	c->model_name = get_model_name(c->family, c->model);
825 
826 	status = ia64_pal_vm_summary(&vm1, &vm2);
827 	if (status == PAL_STATUS_SUCCESS) {
828 		impl_va_msb = vm2.pal_vm_info_2_s.impl_va_msb;
829 		phys_addr_size = vm1.pal_vm_info_1_s.phys_add_size;
830 	}
831 	c->unimpl_va_mask = ~((7L<<61) | ((1L << (impl_va_msb + 1)) - 1));
832 	c->unimpl_pa_mask = ~((1L<<63) | ((1L << phys_addr_size) - 1));
833 }
834 
835 /*
836  * Do the following calculations:
837  *
838  * 1. the max. cache line size.
839  * 2. the minimum of the i-cache stride sizes for "flush_icache_range()".
840  * 3. the minimum of the cache stride sizes for "clflush_cache_range()".
841  */
842 static void
843 get_cache_info(void)
844 {
845 	unsigned long line_size, max = 1;
846 	unsigned long l, levels, unique_caches;
847 	pal_cache_config_info_t cci;
848 	long status;
849 
850         status = ia64_pal_cache_summary(&levels, &unique_caches);
851         if (status != 0) {
852                 printk(KERN_ERR "%s: ia64_pal_cache_summary() failed (status=%ld)\n",
853                        __func__, status);
854                 max = SMP_CACHE_BYTES;
855 		/* Safest setup for "flush_icache_range()" */
856 		ia64_i_cache_stride_shift = I_CACHE_STRIDE_SHIFT;
857 		/* Safest setup for "clflush_cache_range()" */
858 		ia64_cache_stride_shift = CACHE_STRIDE_SHIFT;
859 		goto out;
860         }
861 
862 	for (l = 0; l < levels; ++l) {
863 		/* cache_type (data_or_unified)=2 */
864 		status = ia64_pal_cache_config_info(l, 2, &cci);
865 		if (status != 0) {
866 			printk(KERN_ERR "%s: ia64_pal_cache_config_info"
867 				"(l=%lu, 2) failed (status=%ld)\n",
868 				__func__, l, status);
869 			max = SMP_CACHE_BYTES;
870 			/* The safest setup for "flush_icache_range()" */
871 			cci.pcci_stride = I_CACHE_STRIDE_SHIFT;
872 			/* The safest setup for "clflush_cache_range()" */
873 			ia64_cache_stride_shift = CACHE_STRIDE_SHIFT;
874 			cci.pcci_unified = 1;
875 		} else {
876 			if (cci.pcci_stride < ia64_cache_stride_shift)
877 				ia64_cache_stride_shift = cci.pcci_stride;
878 
879 			line_size = 1 << cci.pcci_line_size;
880 			if (line_size > max)
881 				max = line_size;
882 		}
883 
884 		if (!cci.pcci_unified) {
885 			/* cache_type (instruction)=1*/
886 			status = ia64_pal_cache_config_info(l, 1, &cci);
887 			if (status != 0) {
888 				printk(KERN_ERR "%s: ia64_pal_cache_config_info"
889 					"(l=%lu, 1) failed (status=%ld)\n",
890 					__func__, l, status);
891 				/* The safest setup for flush_icache_range() */
892 				cci.pcci_stride = I_CACHE_STRIDE_SHIFT;
893 			}
894 		}
895 		if (cci.pcci_stride < ia64_i_cache_stride_shift)
896 			ia64_i_cache_stride_shift = cci.pcci_stride;
897 	}
898   out:
899 	if (max > ia64_max_cacheline_size)
900 		ia64_max_cacheline_size = max;
901 }
902 
903 /*
904  * cpu_init() initializes state that is per-CPU.  This function acts
905  * as a 'CPU state barrier', nothing should get across.
906  */
907 void
908 cpu_init (void)
909 {
910 	extern void ia64_mmu_init(void *);
911 	static unsigned long max_num_phys_stacked = IA64_NUM_PHYS_STACK_REG;
912 	unsigned long num_phys_stacked;
913 	pal_vm_info_2_u_t vmi;
914 	unsigned int max_ctx;
915 	struct cpuinfo_ia64 *cpu_info;
916 	void *cpu_data;
917 
918 	cpu_data = per_cpu_init();
919 #ifdef CONFIG_SMP
920 	/*
921 	 * insert boot cpu into sibling and core mapes
922 	 * (must be done after per_cpu area is setup)
923 	 */
924 	if (smp_processor_id() == 0) {
925 		cpumask_set_cpu(0, &per_cpu(cpu_sibling_map, 0));
926 		cpumask_set_cpu(0, &cpu_core_map[0]);
927 	} else {
928 		/*
929 		 * Set ar.k3 so that assembly code in MCA handler can compute
930 		 * physical addresses of per cpu variables with a simple:
931 		 *   phys = ar.k3 + &per_cpu_var
932 		 * and the alt-dtlb-miss handler can set per-cpu mapping into
933 		 * the TLB when needed. head.S already did this for cpu0.
934 		 */
935 		ia64_set_kr(IA64_KR_PER_CPU_DATA,
936 			    ia64_tpa(cpu_data) - (long) __per_cpu_start);
937 	}
938 #endif
939 
940 	get_cache_info();
941 
942 	/*
943 	 * We can't pass "local_cpu_data" to identify_cpu() because we haven't called
944 	 * ia64_mmu_init() yet.  And we can't call ia64_mmu_init() first because it
945 	 * depends on the data returned by identify_cpu().  We break the dependency by
946 	 * accessing cpu_data() through the canonical per-CPU address.
947 	 */
948 	cpu_info = cpu_data + ((char *) &__ia64_per_cpu_var(ia64_cpu_info) - __per_cpu_start);
949 	identify_cpu(cpu_info);
950 
951 #ifdef CONFIG_MCKINLEY
952 	{
953 #		define FEATURE_SET 16
954 		struct ia64_pal_retval iprv;
955 
956 		if (cpu_info->family == 0x1f) {
957 			PAL_CALL_PHYS(iprv, PAL_PROC_GET_FEATURES, 0, FEATURE_SET, 0);
958 			if ((iprv.status == 0) && (iprv.v0 & 0x80) && (iprv.v2 & 0x80))
959 				PAL_CALL_PHYS(iprv, PAL_PROC_SET_FEATURES,
960 				              (iprv.v1 | 0x80), FEATURE_SET, 0);
961 		}
962 	}
963 #endif
964 
965 	/* Clear the stack memory reserved for pt_regs: */
966 	memset(task_pt_regs(current), 0, sizeof(struct pt_regs));
967 
968 	ia64_set_kr(IA64_KR_FPU_OWNER, 0);
969 
970 	/*
971 	 * Initialize the page-table base register to a global
972 	 * directory with all zeroes.  This ensure that we can handle
973 	 * TLB-misses to user address-space even before we created the
974 	 * first user address-space.  This may happen, e.g., due to
975 	 * aggressive use of lfetch.fault.
976 	 */
977 	ia64_set_kr(IA64_KR_PT_BASE, __pa(ia64_imva(empty_zero_page)));
978 
979 	/*
980 	 * Initialize default control register to defer speculative faults except
981 	 * for those arising from TLB misses, which are not deferred.  The
982 	 * kernel MUST NOT depend on a particular setting of these bits (in other words,
983 	 * the kernel must have recovery code for all speculative accesses).  Turn on
984 	 * dcr.lc as per recommendation by the architecture team.  Most IA-32 apps
985 	 * shouldn't be affected by this (moral: keep your ia32 locks aligned and you'll
986 	 * be fine).
987 	 */
988 	ia64_setreg(_IA64_REG_CR_DCR,  (  IA64_DCR_DP | IA64_DCR_DK | IA64_DCR_DX | IA64_DCR_DR
989 					| IA64_DCR_DA | IA64_DCR_DD | IA64_DCR_LC));
990 	mmgrab(&init_mm);
991 	current->active_mm = &init_mm;
992 	BUG_ON(current->mm);
993 
994 	ia64_mmu_init(ia64_imva(cpu_data));
995 	ia64_mca_cpu_init(ia64_imva(cpu_data));
996 
997 	/* Clear ITC to eliminate sched_clock() overflows in human time.  */
998 	ia64_set_itc(0);
999 
1000 	/* disable all local interrupt sources: */
1001 	ia64_set_itv(1 << 16);
1002 	ia64_set_lrr0(1 << 16);
1003 	ia64_set_lrr1(1 << 16);
1004 	ia64_setreg(_IA64_REG_CR_PMV, 1 << 16);
1005 	ia64_setreg(_IA64_REG_CR_CMCV, 1 << 16);
1006 
1007 	/* clear TPR & XTP to enable all interrupt classes: */
1008 	ia64_setreg(_IA64_REG_CR_TPR, 0);
1009 
1010 	/* Clear any pending interrupts left by SAL/EFI */
1011 	while (ia64_get_ivr() != IA64_SPURIOUS_INT_VECTOR)
1012 		ia64_eoi();
1013 
1014 #ifdef CONFIG_SMP
1015 	normal_xtp();
1016 #endif
1017 
1018 	/* set ia64_ctx.max_rid to the maximum RID that is supported by all CPUs: */
1019 	if (ia64_pal_vm_summary(NULL, &vmi) == 0) {
1020 		max_ctx = (1U << (vmi.pal_vm_info_2_s.rid_size - 3)) - 1;
1021 		setup_ptcg_sem(vmi.pal_vm_info_2_s.max_purges, NPTCG_FROM_PAL);
1022 	} else {
1023 		printk(KERN_WARNING "cpu_init: PAL VM summary failed, assuming 18 RID bits\n");
1024 		max_ctx = (1U << 15) - 1;	/* use architected minimum */
1025 	}
1026 	while (max_ctx < ia64_ctx.max_ctx) {
1027 		unsigned int old = ia64_ctx.max_ctx;
1028 		if (cmpxchg(&ia64_ctx.max_ctx, old, max_ctx) == old)
1029 			break;
1030 	}
1031 
1032 	if (ia64_pal_rse_info(&num_phys_stacked, NULL) != 0) {
1033 		printk(KERN_WARNING "cpu_init: PAL RSE info failed; assuming 96 physical "
1034 		       "stacked regs\n");
1035 		num_phys_stacked = 96;
1036 	}
1037 	/* size of physical stacked register partition plus 8 bytes: */
1038 	if (num_phys_stacked > max_num_phys_stacked) {
1039 		ia64_patch_phys_stack_reg(num_phys_stacked*8 + 8);
1040 		max_num_phys_stacked = num_phys_stacked;
1041 	}
1042 	platform_cpu_init();
1043 }
1044 
1045 void __init
1046 check_bugs (void)
1047 {
1048 	ia64_patch_mckinley_e9((unsigned long) __start___mckinley_e9_bundles,
1049 			       (unsigned long) __end___mckinley_e9_bundles);
1050 }
1051 
1052 static int __init run_dmi_scan(void)
1053 {
1054 	dmi_setup();
1055 	return 0;
1056 }
1057 core_initcall(run_dmi_scan);
1058