xref: /openbmc/linux/arch/ia64/kernel/setup.c (revision 612a462a)
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/cpu.h>
33 #include <linux/kernel.h>
34 #include <linux/reboot.h>
35 #include <linux/sched/mm.h>
36 #include <linux/sched/clock.h>
37 #include <linux/sched/task_stack.h>
38 #include <linux/seq_file.h>
39 #include <linux/string.h>
40 #include <linux/threads.h>
41 #include <linux/screen_info.h>
42 #include <linux/dmi.h>
43 #include <linux/serial.h>
44 #include <linux/serial_core.h>
45 #include <linux/efi.h>
46 #include <linux/initrd.h>
47 #include <linux/pm.h>
48 #include <linux/cpufreq.h>
49 #include <linux/kexec.h>
50 #include <linux/crash_dump.h>
51 
52 #include <asm/machvec.h>
53 #include <asm/mca.h>
54 #include <asm/meminit.h>
55 #include <asm/page.h>
56 #include <asm/patch.h>
57 #include <asm/pgtable.h>
58 #include <asm/processor.h>
59 #include <asm/sal.h>
60 #include <asm/sections.h>
61 #include <asm/setup.h>
62 #include <asm/smp.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, ia64_cpu_info);
77 EXPORT_SYMBOL(ia64_cpu_info);
78 DEFINE_PER_CPU(unsigned long, local_per_cpu_offset);
79 #ifdef CONFIG_SMP
80 EXPORT_SYMBOL(local_per_cpu_offset);
81 #endif
82 unsigned long ia64_cycles_per_usec;
83 struct ia64_boot_param *ia64_boot_param;
84 struct screen_info screen_info;
85 unsigned long vga_console_iobase;
86 unsigned long vga_console_membase;
87 
88 static struct resource data_resource = {
89 	.name	= "Kernel data",
90 	.flags	= IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
91 };
92 
93 static struct resource code_resource = {
94 	.name	= "Kernel code",
95 	.flags	= IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
96 };
97 
98 static struct resource bss_resource = {
99 	.name	= "Kernel bss",
100 	.flags	= IORESOURCE_BUSY | IORESOURCE_SYSTEM_RAM
101 };
102 
103 unsigned long ia64_max_cacheline_size;
104 
105 unsigned long ia64_iobase;	/* virtual address for I/O accesses */
106 EXPORT_SYMBOL(ia64_iobase);
107 struct io_space io_space[MAX_IO_SPACES];
108 EXPORT_SYMBOL(io_space);
109 unsigned int num_io_spaces;
110 
111 /*
112  * "flush_icache_range()" needs to know what processor dependent stride size to use
113  * when it makes i-cache(s) coherent with d-caches.
114  */
115 #define	I_CACHE_STRIDE_SHIFT	5	/* Safest way to go: 32 bytes by 32 bytes */
116 unsigned long ia64_i_cache_stride_shift = ~0;
117 /*
118  * "clflush_cache_range()" needs to know what processor dependent stride size to
119  * use when it flushes cache lines including both d-cache and i-cache.
120  */
121 /* Safest way to go: 32 bytes by 32 bytes */
122 #define	CACHE_STRIDE_SHIFT	5
123 unsigned long ia64_cache_stride_shift = ~0;
124 
125 /*
126  * The merge_mask variable needs to be set to (max(iommu_page_size(iommu)) - 1).  This
127  * mask specifies a mask of address bits that must be 0 in order for two buffers to be
128  * mergeable by the I/O MMU (i.e., the end address of the first buffer and the start
129  * address of the second buffer must be aligned to (merge_mask+1) in order to be
130  * mergeable).  By default, we assume there is no I/O MMU which can merge physically
131  * discontiguous buffers, so we set the merge_mask to ~0UL, which corresponds to a iommu
132  * page-size of 2^64.
133  */
134 unsigned long ia64_max_iommu_merge_mask = ~0UL;
135 EXPORT_SYMBOL(ia64_max_iommu_merge_mask);
136 
137 /*
138  * We use a special marker for the end of memory and it uses the extra (+1) slot
139  */
140 struct rsvd_region rsvd_region[IA64_MAX_RSVD_REGIONS + 1] __initdata;
141 int num_rsvd_regions __initdata;
142 
143 
144 /*
145  * Filter incoming memory segments based on the primitive map created from the boot
146  * parameters. Segments contained in the map are removed from the memory ranges. A
147  * caller-specified function is called with the memory ranges that remain after filtering.
148  * This routine does not assume the incoming segments are sorted.
149  */
150 int __init
151 filter_rsvd_memory (u64 start, u64 end, void *arg)
152 {
153 	u64 range_start, range_end, prev_start;
154 	void (*func)(unsigned long, unsigned long, int);
155 	int i;
156 
157 #if IGNORE_PFN0
158 	if (start == PAGE_OFFSET) {
159 		printk(KERN_WARNING "warning: skipping physical page 0\n");
160 		start += PAGE_SIZE;
161 		if (start >= end) return 0;
162 	}
163 #endif
164 	/*
165 	 * lowest possible address(walker uses virtual)
166 	 */
167 	prev_start = PAGE_OFFSET;
168 	func = arg;
169 
170 	for (i = 0; i < num_rsvd_regions; ++i) {
171 		range_start = max(start, prev_start);
172 		range_end   = min(end, rsvd_region[i].start);
173 
174 		if (range_start < range_end)
175 			call_pernode_memory(__pa(range_start), range_end - range_start, func);
176 
177 		/* nothing more available in this segment */
178 		if (range_end == end) return 0;
179 
180 		prev_start = rsvd_region[i].end;
181 	}
182 	/* end of memory marker allows full processing inside loop body */
183 	return 0;
184 }
185 
186 /*
187  * Similar to "filter_rsvd_memory()", but the reserved memory ranges
188  * are not filtered out.
189  */
190 int __init
191 filter_memory(u64 start, u64 end, void *arg)
192 {
193 	void (*func)(unsigned long, unsigned long, int);
194 
195 #if IGNORE_PFN0
196 	if (start == PAGE_OFFSET) {
197 		printk(KERN_WARNING "warning: skipping physical page 0\n");
198 		start += PAGE_SIZE;
199 		if (start >= end)
200 			return 0;
201 	}
202 #endif
203 	func = arg;
204 	if (start < end)
205 		call_pernode_memory(__pa(start), end - start, func);
206 	return 0;
207 }
208 
209 static void __init
210 sort_regions (struct rsvd_region *rsvd_region, int max)
211 {
212 	int j;
213 
214 	/* simple bubble sorting */
215 	while (max--) {
216 		for (j = 0; j < max; ++j) {
217 			if (rsvd_region[j].start > rsvd_region[j+1].start) {
218 				struct rsvd_region tmp;
219 				tmp = rsvd_region[j];
220 				rsvd_region[j] = rsvd_region[j + 1];
221 				rsvd_region[j + 1] = tmp;
222 			}
223 		}
224 	}
225 }
226 
227 /* merge overlaps */
228 static int __init
229 merge_regions (struct rsvd_region *rsvd_region, int max)
230 {
231 	int i;
232 	for (i = 1; i < max; ++i) {
233 		if (rsvd_region[i].start >= rsvd_region[i-1].end)
234 			continue;
235 		if (rsvd_region[i].end > rsvd_region[i-1].end)
236 			rsvd_region[i-1].end = rsvd_region[i].end;
237 		--max;
238 		memmove(&rsvd_region[i], &rsvd_region[i+1],
239 			(max - i) * sizeof(struct rsvd_region));
240 	}
241 	return max;
242 }
243 
244 /*
245  * Request address space for all standard resources
246  */
247 static int __init register_memory(void)
248 {
249 	code_resource.start = ia64_tpa(_text);
250 	code_resource.end   = ia64_tpa(_etext) - 1;
251 	data_resource.start = ia64_tpa(_etext);
252 	data_resource.end   = ia64_tpa(_edata) - 1;
253 	bss_resource.start  = ia64_tpa(__bss_start);
254 	bss_resource.end    = ia64_tpa(_end) - 1;
255 	efi_initialize_iomem_resources(&code_resource, &data_resource,
256 			&bss_resource);
257 
258 	return 0;
259 }
260 
261 __initcall(register_memory);
262 
263 
264 #ifdef CONFIG_KEXEC
265 
266 /*
267  * This function checks if the reserved crashkernel is allowed on the specific
268  * IA64 machine flavour. Machines without an IO TLB use swiotlb and require
269  * some memory below 4 GB (i.e. in 32 bit area), see the implementation of
270  * lib/swiotlb.c. The hpzx1 architecture has an IO TLB but cannot use that
271  * in kdump case. See the comment in sba_init() in sba_iommu.c.
272  *
273  * So, the only machvec that really supports loading the kdump kernel
274  * over 4 GB is "sn2".
275  */
276 static int __init check_crashkernel_memory(unsigned long pbase, size_t size)
277 {
278 	if (ia64_platform_is("sn2") || ia64_platform_is("uv"))
279 		return 1;
280 	else
281 		return pbase < (1UL << 32);
282 }
283 
284 static void __init setup_crashkernel(unsigned long total, int *n)
285 {
286 	unsigned long long base = 0, size = 0;
287 	int ret;
288 
289 	ret = parse_crashkernel(boot_command_line, total,
290 			&size, &base);
291 	if (ret == 0 && size > 0) {
292 		if (!base) {
293 			sort_regions(rsvd_region, *n);
294 			*n = merge_regions(rsvd_region, *n);
295 			base = kdump_find_rsvd_region(size,
296 					rsvd_region, *n);
297 		}
298 
299 		if (!check_crashkernel_memory(base, size)) {
300 			pr_warning("crashkernel: There would be kdump memory "
301 				"at %ld GB but this is unusable because it "
302 				"must\nbe below 4 GB. Change the memory "
303 				"configuration of the machine.\n",
304 				(unsigned long)(base >> 30));
305 			return;
306 		}
307 
308 		if (base != ~0UL) {
309 			printk(KERN_INFO "Reserving %ldMB of memory at %ldMB "
310 					"for crashkernel (System RAM: %ldMB)\n",
311 					(unsigned long)(size >> 20),
312 					(unsigned long)(base >> 20),
313 					(unsigned long)(total >> 20));
314 			rsvd_region[*n].start =
315 				(unsigned long)__va(base);
316 			rsvd_region[*n].end =
317 				(unsigned long)__va(base + size);
318 			(*n)++;
319 			crashk_res.start = base;
320 			crashk_res.end = base + size - 1;
321 		}
322 	}
323 	efi_memmap_res.start = ia64_boot_param->efi_memmap;
324 	efi_memmap_res.end = efi_memmap_res.start +
325 		ia64_boot_param->efi_memmap_size;
326 	boot_param_res.start = __pa(ia64_boot_param);
327 	boot_param_res.end = boot_param_res.start +
328 		sizeof(*ia64_boot_param);
329 }
330 #else
331 static inline void __init setup_crashkernel(unsigned long total, int *n)
332 {}
333 #endif
334 
335 /**
336  * reserve_memory - setup reserved memory areas
337  *
338  * Setup the reserved memory areas set aside for the boot parameters,
339  * initrd, etc.  There are currently %IA64_MAX_RSVD_REGIONS defined,
340  * see arch/ia64/include/asm/meminit.h if you need to define more.
341  */
342 void __init
343 reserve_memory (void)
344 {
345 	int n = 0;
346 	unsigned long total_memory;
347 
348 	/*
349 	 * none of the entries in this table overlap
350 	 */
351 	rsvd_region[n].start = (unsigned long) ia64_boot_param;
352 	rsvd_region[n].end   = rsvd_region[n].start + sizeof(*ia64_boot_param);
353 	n++;
354 
355 	rsvd_region[n].start = (unsigned long) __va(ia64_boot_param->efi_memmap);
356 	rsvd_region[n].end   = rsvd_region[n].start + ia64_boot_param->efi_memmap_size;
357 	n++;
358 
359 	rsvd_region[n].start = (unsigned long) __va(ia64_boot_param->command_line);
360 	rsvd_region[n].end   = (rsvd_region[n].start
361 				+ strlen(__va(ia64_boot_param->command_line)) + 1);
362 	n++;
363 
364 	rsvd_region[n].start = (unsigned long) ia64_imva((void *)KERNEL_START);
365 	rsvd_region[n].end   = (unsigned long) ia64_imva(_end);
366 	n++;
367 
368 #ifdef CONFIG_BLK_DEV_INITRD
369 	if (ia64_boot_param->initrd_start) {
370 		rsvd_region[n].start = (unsigned long)__va(ia64_boot_param->initrd_start);
371 		rsvd_region[n].end   = rsvd_region[n].start + ia64_boot_param->initrd_size;
372 		n++;
373 	}
374 #endif
375 
376 #ifdef CONFIG_CRASH_DUMP
377 	if (reserve_elfcorehdr(&rsvd_region[n].start,
378 			       &rsvd_region[n].end) == 0)
379 		n++;
380 #endif
381 
382 	total_memory = efi_memmap_init(&rsvd_region[n].start, &rsvd_region[n].end);
383 	n++;
384 
385 	setup_crashkernel(total_memory, &n);
386 
387 	/* end of memory marker */
388 	rsvd_region[n].start = ~0UL;
389 	rsvd_region[n].end   = ~0UL;
390 	n++;
391 
392 	num_rsvd_regions = n;
393 	BUG_ON(IA64_MAX_RSVD_REGIONS + 1 < n);
394 
395 	sort_regions(rsvd_region, num_rsvd_regions);
396 	num_rsvd_regions = merge_regions(rsvd_region, num_rsvd_regions);
397 }
398 
399 
400 /**
401  * find_initrd - get initrd parameters from the boot parameter structure
402  *
403  * Grab the initrd start and end from the boot parameter struct given us by
404  * the boot loader.
405  */
406 void __init
407 find_initrd (void)
408 {
409 #ifdef CONFIG_BLK_DEV_INITRD
410 	if (ia64_boot_param->initrd_start) {
411 		initrd_start = (unsigned long)__va(ia64_boot_param->initrd_start);
412 		initrd_end   = initrd_start+ia64_boot_param->initrd_size;
413 
414 		printk(KERN_INFO "Initial ramdisk at: 0x%lx (%llu bytes)\n",
415 		       initrd_start, ia64_boot_param->initrd_size);
416 	}
417 #endif
418 }
419 
420 static void __init
421 io_port_init (void)
422 {
423 	unsigned long phys_iobase;
424 
425 	/*
426 	 * Set `iobase' based on the EFI memory map or, failing that, the
427 	 * value firmware left in ar.k0.
428 	 *
429 	 * Note that in ia32 mode, IN/OUT instructions use ar.k0 to compute
430 	 * the port's virtual address, so ia32_load_state() loads it with a
431 	 * user virtual address.  But in ia64 mode, glibc uses the
432 	 * *physical* address in ar.k0 to mmap the appropriate area from
433 	 * /dev/mem, and the inX()/outX() interfaces use MMIO.  In both
434 	 * cases, user-mode can only use the legacy 0-64K I/O port space.
435 	 *
436 	 * ar.k0 is not involved in kernel I/O port accesses, which can use
437 	 * any of the I/O port spaces and are done via MMIO using the
438 	 * virtual mmio_base from the appropriate io_space[].
439 	 */
440 	phys_iobase = efi_get_iobase();
441 	if (!phys_iobase) {
442 		phys_iobase = ia64_get_kr(IA64_KR_IO_BASE);
443 		printk(KERN_INFO "No I/O port range found in EFI memory map, "
444 			"falling back to AR.KR0 (0x%lx)\n", phys_iobase);
445 	}
446 	ia64_iobase = (unsigned long) ioremap(phys_iobase, 0);
447 	ia64_set_kr(IA64_KR_IO_BASE, __pa(ia64_iobase));
448 
449 	/* setup legacy IO port space */
450 	io_space[0].mmio_base = ia64_iobase;
451 	io_space[0].sparse = 1;
452 	num_io_spaces = 1;
453 }
454 
455 /**
456  * early_console_setup - setup debugging console
457  *
458  * Consoles started here require little enough setup that we can start using
459  * them very early in the boot process, either right after the machine
460  * vector initialization, or even before if the drivers can detect their hw.
461  *
462  * Returns non-zero if a console couldn't be setup.
463  */
464 static inline int __init
465 early_console_setup (char *cmdline)
466 {
467 	int earlycons = 0;
468 
469 #ifdef CONFIG_SERIAL_SGI_L1_CONSOLE
470 	{
471 		extern int sn_serial_console_early_setup(void);
472 		if (!sn_serial_console_early_setup())
473 			earlycons++;
474 	}
475 #endif
476 #ifdef CONFIG_EFI_PCDP
477 	if (!efi_setup_pcdp_console(cmdline))
478 		earlycons++;
479 #endif
480 	if (!simcons_register())
481 		earlycons++;
482 
483 	return (earlycons) ? 0 : -1;
484 }
485 
486 static inline void
487 mark_bsp_online (void)
488 {
489 #ifdef CONFIG_SMP
490 	/* If we register an early console, allow CPU 0 to printk */
491 	set_cpu_online(smp_processor_id(), true);
492 #endif
493 }
494 
495 static __initdata int nomca;
496 static __init int setup_nomca(char *s)
497 {
498 	nomca = 1;
499 	return 0;
500 }
501 early_param("nomca", setup_nomca);
502 
503 #ifdef CONFIG_CRASH_DUMP
504 int __init reserve_elfcorehdr(u64 *start, u64 *end)
505 {
506 	u64 length;
507 
508 	/* We get the address using the kernel command line,
509 	 * but the size is extracted from the EFI tables.
510 	 * Both address and size are required for reservation
511 	 * to work properly.
512 	 */
513 
514 	if (!is_vmcore_usable())
515 		return -EINVAL;
516 
517 	if ((length = vmcore_find_descriptor_size(elfcorehdr_addr)) == 0) {
518 		vmcore_unusable();
519 		return -EINVAL;
520 	}
521 
522 	*start = (unsigned long)__va(elfcorehdr_addr);
523 	*end = *start + length;
524 	return 0;
525 }
526 
527 #endif /* CONFIG_PROC_VMCORE */
528 
529 void __init
530 setup_arch (char **cmdline_p)
531 {
532 	unw_init();
533 
534 	ia64_patch_vtop((u64) __start___vtop_patchlist, (u64) __end___vtop_patchlist);
535 
536 	*cmdline_p = __va(ia64_boot_param->command_line);
537 	strlcpy(boot_command_line, *cmdline_p, COMMAND_LINE_SIZE);
538 
539 	efi_init();
540 	io_port_init();
541 
542 #ifdef CONFIG_IA64_GENERIC
543 	/* machvec needs to be parsed from the command line
544 	 * before parse_early_param() is called to ensure
545 	 * that ia64_mv is initialised before any command line
546 	 * settings may cause console setup to occur
547 	 */
548 	machvec_init_from_cmdline(*cmdline_p);
549 #endif
550 
551 	parse_early_param();
552 
553 	if (early_console_setup(*cmdline_p) == 0)
554 		mark_bsp_online();
555 
556 #ifdef CONFIG_ACPI
557 	/* Initialize the ACPI boot-time table parser */
558 	acpi_table_init();
559 	early_acpi_boot_init();
560 # ifdef CONFIG_ACPI_NUMA
561 	acpi_numa_init();
562 	acpi_numa_fixup();
563 #  ifdef CONFIG_ACPI_HOTPLUG_CPU
564 	prefill_possible_map();
565 #  endif
566 	per_cpu_scan_finalize((cpumask_weight(&early_cpu_possible_map) == 0 ?
567 		32 : cpumask_weight(&early_cpu_possible_map)),
568 		additional_cpus > 0 ? additional_cpus : 0);
569 # endif
570 #endif /* CONFIG_APCI_BOOT */
571 
572 #ifdef CONFIG_SMP
573 	smp_build_cpu_map();
574 #endif
575 	find_memory();
576 
577 	/* process SAL system table: */
578 	ia64_sal_init(__va(efi.sal_systab));
579 
580 #ifdef CONFIG_ITANIUM
581 	ia64_patch_rse((u64) __start___rse_patchlist, (u64) __end___rse_patchlist);
582 #else
583 	{
584 		unsigned long num_phys_stacked;
585 
586 		if (ia64_pal_rse_info(&num_phys_stacked, 0) == 0 && num_phys_stacked > 96)
587 			ia64_patch_rse((u64) __start___rse_patchlist, (u64) __end___rse_patchlist);
588 	}
589 #endif
590 
591 #ifdef CONFIG_SMP
592 	cpu_physical_id(0) = hard_smp_processor_id();
593 #endif
594 
595 	cpu_init();	/* initialize the bootstrap CPU */
596 	mmu_context_init();	/* initialize context_id bitmap */
597 
598 #ifdef CONFIG_VT
599 	if (!conswitchp) {
600 # if defined(CONFIG_DUMMY_CONSOLE)
601 		conswitchp = &dummy_con;
602 # endif
603 # if defined(CONFIG_VGA_CONSOLE)
604 		/*
605 		 * Non-legacy systems may route legacy VGA MMIO range to system
606 		 * memory.  vga_con probes the MMIO hole, so memory looks like
607 		 * a VGA device to it.  The EFI memory map can tell us if it's
608 		 * memory so we can avoid this problem.
609 		 */
610 		if (efi_mem_type(0xA0000) != EFI_CONVENTIONAL_MEMORY)
611 			conswitchp = &vga_con;
612 # endif
613 	}
614 #endif
615 
616 	/* enable IA-64 Machine Check Abort Handling unless disabled */
617 	if (!nomca)
618 		ia64_mca_init();
619 
620 	platform_setup(cmdline_p);
621 #ifndef CONFIG_IA64_HP_SIM
622 	check_sal_cache_flush();
623 #endif
624 	paging_init();
625 
626 	clear_sched_clock_stable();
627 }
628 
629 /*
630  * Display cpu info for all CPUs.
631  */
632 static int
633 show_cpuinfo (struct seq_file *m, void *v)
634 {
635 #ifdef CONFIG_SMP
636 #	define lpj	c->loops_per_jiffy
637 #	define cpunum	c->cpu
638 #else
639 #	define lpj	loops_per_jiffy
640 #	define cpunum	0
641 #endif
642 	static struct {
643 		unsigned long mask;
644 		const char *feature_name;
645 	} feature_bits[] = {
646 		{ 1UL << 0, "branchlong" },
647 		{ 1UL << 1, "spontaneous deferral"},
648 		{ 1UL << 2, "16-byte atomic ops" }
649 	};
650 	char features[128], *cp, *sep;
651 	struct cpuinfo_ia64 *c = v;
652 	unsigned long mask;
653 	unsigned long proc_freq;
654 	int i, size;
655 
656 	mask = c->features;
657 
658 	/* build the feature string: */
659 	memcpy(features, "standard", 9);
660 	cp = features;
661 	size = sizeof(features);
662 	sep = "";
663 	for (i = 0; i < ARRAY_SIZE(feature_bits) && size > 1; ++i) {
664 		if (mask & feature_bits[i].mask) {
665 			cp += snprintf(cp, size, "%s%s", sep,
666 				       feature_bits[i].feature_name),
667 			sep = ", ";
668 			mask &= ~feature_bits[i].mask;
669 			size = sizeof(features) - (cp - features);
670 		}
671 	}
672 	if (mask && size > 1) {
673 		/* print unknown features as a hex value */
674 		snprintf(cp, size, "%s0x%lx", sep, mask);
675 	}
676 
677 	proc_freq = cpufreq_quick_get(cpunum);
678 	if (!proc_freq)
679 		proc_freq = c->proc_freq / 1000;
680 
681 	seq_printf(m,
682 		   "processor  : %d\n"
683 		   "vendor     : %s\n"
684 		   "arch       : IA-64\n"
685 		   "family     : %u\n"
686 		   "model      : %u\n"
687 		   "model name : %s\n"
688 		   "revision   : %u\n"
689 		   "archrev    : %u\n"
690 		   "features   : %s\n"
691 		   "cpu number : %lu\n"
692 		   "cpu regs   : %u\n"
693 		   "cpu MHz    : %lu.%03lu\n"
694 		   "itc MHz    : %lu.%06lu\n"
695 		   "BogoMIPS   : %lu.%02lu\n",
696 		   cpunum, c->vendor, c->family, c->model,
697 		   c->model_name, c->revision, c->archrev,
698 		   features, c->ppn, c->number,
699 		   proc_freq / 1000, proc_freq % 1000,
700 		   c->itc_freq / 1000000, c->itc_freq % 1000000,
701 		   lpj*HZ/500000, (lpj*HZ/5000) % 100);
702 #ifdef CONFIG_SMP
703 	seq_printf(m, "siblings   : %u\n",
704 		   cpumask_weight(&cpu_core_map[cpunum]));
705 	if (c->socket_id != -1)
706 		seq_printf(m, "physical id: %u\n", c->socket_id);
707 	if (c->threads_per_core > 1 || c->cores_per_socket > 1)
708 		seq_printf(m,
709 			   "core id    : %u\n"
710 			   "thread id  : %u\n",
711 			   c->core_id, c->thread_id);
712 #endif
713 	seq_printf(m,"\n");
714 
715 	return 0;
716 }
717 
718 static void *
719 c_start (struct seq_file *m, loff_t *pos)
720 {
721 #ifdef CONFIG_SMP
722 	while (*pos < nr_cpu_ids && !cpu_online(*pos))
723 		++*pos;
724 #endif
725 	return *pos < nr_cpu_ids ? cpu_data(*pos) : NULL;
726 }
727 
728 static void *
729 c_next (struct seq_file *m, void *v, loff_t *pos)
730 {
731 	++*pos;
732 	return c_start(m, pos);
733 }
734 
735 static void
736 c_stop (struct seq_file *m, void *v)
737 {
738 }
739 
740 const struct seq_operations cpuinfo_op = {
741 	.start =	c_start,
742 	.next =		c_next,
743 	.stop =		c_stop,
744 	.show =		show_cpuinfo
745 };
746 
747 #define MAX_BRANDS	8
748 static char brandname[MAX_BRANDS][128];
749 
750 static char *
751 get_model_name(__u8 family, __u8 model)
752 {
753 	static int overflow;
754 	char brand[128];
755 	int i;
756 
757 	memcpy(brand, "Unknown", 8);
758 	if (ia64_pal_get_brand_info(brand)) {
759 		if (family == 0x7)
760 			memcpy(brand, "Merced", 7);
761 		else if (family == 0x1f) switch (model) {
762 			case 0: memcpy(brand, "McKinley", 9); break;
763 			case 1: memcpy(brand, "Madison", 8); break;
764 			case 2: memcpy(brand, "Madison up to 9M cache", 23); break;
765 		}
766 	}
767 	for (i = 0; i < MAX_BRANDS; i++)
768 		if (strcmp(brandname[i], brand) == 0)
769 			return brandname[i];
770 	for (i = 0; i < MAX_BRANDS; i++)
771 		if (brandname[i][0] == '\0')
772 			return strcpy(brandname[i], brand);
773 	if (overflow++ == 0)
774 		printk(KERN_ERR
775 		       "%s: Table overflow. Some processor model information will be missing\n",
776 		       __func__);
777 	return "Unknown";
778 }
779 
780 static void
781 identify_cpu (struct cpuinfo_ia64 *c)
782 {
783 	union {
784 		unsigned long bits[5];
785 		struct {
786 			/* id 0 & 1: */
787 			char vendor[16];
788 
789 			/* id 2 */
790 			u64 ppn;		/* processor serial number */
791 
792 			/* id 3: */
793 			unsigned number		:  8;
794 			unsigned revision	:  8;
795 			unsigned model		:  8;
796 			unsigned family		:  8;
797 			unsigned archrev	:  8;
798 			unsigned reserved	: 24;
799 
800 			/* id 4: */
801 			u64 features;
802 		} field;
803 	} cpuid;
804 	pal_vm_info_1_u_t vm1;
805 	pal_vm_info_2_u_t vm2;
806 	pal_status_t status;
807 	unsigned long impl_va_msb = 50, phys_addr_size = 44;	/* Itanium defaults */
808 	int i;
809 	for (i = 0; i < 5; ++i)
810 		cpuid.bits[i] = ia64_get_cpuid(i);
811 
812 	memcpy(c->vendor, cpuid.field.vendor, 16);
813 #ifdef CONFIG_SMP
814 	c->cpu = smp_processor_id();
815 
816 	/* below default values will be overwritten  by identify_siblings()
817 	 * for Multi-Threading/Multi-Core capable CPUs
818 	 */
819 	c->threads_per_core = c->cores_per_socket = c->num_log = 1;
820 	c->socket_id = -1;
821 
822 	identify_siblings(c);
823 
824 	if (c->threads_per_core > smp_num_siblings)
825 		smp_num_siblings = c->threads_per_core;
826 #endif
827 	c->ppn = cpuid.field.ppn;
828 	c->number = cpuid.field.number;
829 	c->revision = cpuid.field.revision;
830 	c->model = cpuid.field.model;
831 	c->family = cpuid.field.family;
832 	c->archrev = cpuid.field.archrev;
833 	c->features = cpuid.field.features;
834 	c->model_name = get_model_name(c->family, c->model);
835 
836 	status = ia64_pal_vm_summary(&vm1, &vm2);
837 	if (status == PAL_STATUS_SUCCESS) {
838 		impl_va_msb = vm2.pal_vm_info_2_s.impl_va_msb;
839 		phys_addr_size = vm1.pal_vm_info_1_s.phys_add_size;
840 	}
841 	c->unimpl_va_mask = ~((7L<<61) | ((1L << (impl_va_msb + 1)) - 1));
842 	c->unimpl_pa_mask = ~((1L<<63) | ((1L << phys_addr_size) - 1));
843 }
844 
845 /*
846  * Do the following calculations:
847  *
848  * 1. the max. cache line size.
849  * 2. the minimum of the i-cache stride sizes for "flush_icache_range()".
850  * 3. the minimum of the cache stride sizes for "clflush_cache_range()".
851  */
852 static void
853 get_cache_info(void)
854 {
855 	unsigned long line_size, max = 1;
856 	unsigned long l, levels, unique_caches;
857 	pal_cache_config_info_t cci;
858 	long status;
859 
860         status = ia64_pal_cache_summary(&levels, &unique_caches);
861         if (status != 0) {
862                 printk(KERN_ERR "%s: ia64_pal_cache_summary() failed (status=%ld)\n",
863                        __func__, status);
864                 max = SMP_CACHE_BYTES;
865 		/* Safest setup for "flush_icache_range()" */
866 		ia64_i_cache_stride_shift = I_CACHE_STRIDE_SHIFT;
867 		/* Safest setup for "clflush_cache_range()" */
868 		ia64_cache_stride_shift = CACHE_STRIDE_SHIFT;
869 		goto out;
870         }
871 
872 	for (l = 0; l < levels; ++l) {
873 		/* cache_type (data_or_unified)=2 */
874 		status = ia64_pal_cache_config_info(l, 2, &cci);
875 		if (status != 0) {
876 			printk(KERN_ERR "%s: ia64_pal_cache_config_info"
877 				"(l=%lu, 2) failed (status=%ld)\n",
878 				__func__, l, status);
879 			max = SMP_CACHE_BYTES;
880 			/* The safest setup for "flush_icache_range()" */
881 			cci.pcci_stride = I_CACHE_STRIDE_SHIFT;
882 			/* The safest setup for "clflush_cache_range()" */
883 			ia64_cache_stride_shift = CACHE_STRIDE_SHIFT;
884 			cci.pcci_unified = 1;
885 		} else {
886 			if (cci.pcci_stride < ia64_cache_stride_shift)
887 				ia64_cache_stride_shift = cci.pcci_stride;
888 
889 			line_size = 1 << cci.pcci_line_size;
890 			if (line_size > max)
891 				max = line_size;
892 		}
893 
894 		if (!cci.pcci_unified) {
895 			/* cache_type (instruction)=1*/
896 			status = ia64_pal_cache_config_info(l, 1, &cci);
897 			if (status != 0) {
898 				printk(KERN_ERR "%s: ia64_pal_cache_config_info"
899 					"(l=%lu, 1) failed (status=%ld)\n",
900 					__func__, l, status);
901 				/* The safest setup for flush_icache_range() */
902 				cci.pcci_stride = I_CACHE_STRIDE_SHIFT;
903 			}
904 		}
905 		if (cci.pcci_stride < ia64_i_cache_stride_shift)
906 			ia64_i_cache_stride_shift = cci.pcci_stride;
907 	}
908   out:
909 	if (max > ia64_max_cacheline_size)
910 		ia64_max_cacheline_size = max;
911 }
912 
913 /*
914  * cpu_init() initializes state that is per-CPU.  This function acts
915  * as a 'CPU state barrier', nothing should get across.
916  */
917 void
918 cpu_init (void)
919 {
920 	extern void ia64_mmu_init(void *);
921 	static unsigned long max_num_phys_stacked = IA64_NUM_PHYS_STACK_REG;
922 	unsigned long num_phys_stacked;
923 	pal_vm_info_2_u_t vmi;
924 	unsigned int max_ctx;
925 	struct cpuinfo_ia64 *cpu_info;
926 	void *cpu_data;
927 
928 	cpu_data = per_cpu_init();
929 #ifdef CONFIG_SMP
930 	/*
931 	 * insert boot cpu into sibling and core mapes
932 	 * (must be done after per_cpu area is setup)
933 	 */
934 	if (smp_processor_id() == 0) {
935 		cpumask_set_cpu(0, &per_cpu(cpu_sibling_map, 0));
936 		cpumask_set_cpu(0, &cpu_core_map[0]);
937 	} else {
938 		/*
939 		 * Set ar.k3 so that assembly code in MCA handler can compute
940 		 * physical addresses of per cpu variables with a simple:
941 		 *   phys = ar.k3 + &per_cpu_var
942 		 * and the alt-dtlb-miss handler can set per-cpu mapping into
943 		 * the TLB when needed. head.S already did this for cpu0.
944 		 */
945 		ia64_set_kr(IA64_KR_PER_CPU_DATA,
946 			    ia64_tpa(cpu_data) - (long) __per_cpu_start);
947 	}
948 #endif
949 
950 	get_cache_info();
951 
952 	/*
953 	 * We can't pass "local_cpu_data" to identify_cpu() because we haven't called
954 	 * ia64_mmu_init() yet.  And we can't call ia64_mmu_init() first because it
955 	 * depends on the data returned by identify_cpu().  We break the dependency by
956 	 * accessing cpu_data() through the canonical per-CPU address.
957 	 */
958 	cpu_info = cpu_data + ((char *) &__ia64_per_cpu_var(ia64_cpu_info) - __per_cpu_start);
959 	identify_cpu(cpu_info);
960 
961 #ifdef CONFIG_MCKINLEY
962 	{
963 #		define FEATURE_SET 16
964 		struct ia64_pal_retval iprv;
965 
966 		if (cpu_info->family == 0x1f) {
967 			PAL_CALL_PHYS(iprv, PAL_PROC_GET_FEATURES, 0, FEATURE_SET, 0);
968 			if ((iprv.status == 0) && (iprv.v0 & 0x80) && (iprv.v2 & 0x80))
969 				PAL_CALL_PHYS(iprv, PAL_PROC_SET_FEATURES,
970 				              (iprv.v1 | 0x80), FEATURE_SET, 0);
971 		}
972 	}
973 #endif
974 
975 	/* Clear the stack memory reserved for pt_regs: */
976 	memset(task_pt_regs(current), 0, sizeof(struct pt_regs));
977 
978 	ia64_set_kr(IA64_KR_FPU_OWNER, 0);
979 
980 	/*
981 	 * Initialize the page-table base register to a global
982 	 * directory with all zeroes.  This ensure that we can handle
983 	 * TLB-misses to user address-space even before we created the
984 	 * first user address-space.  This may happen, e.g., due to
985 	 * aggressive use of lfetch.fault.
986 	 */
987 	ia64_set_kr(IA64_KR_PT_BASE, __pa(ia64_imva(empty_zero_page)));
988 
989 	/*
990 	 * Initialize default control register to defer speculative faults except
991 	 * for those arising from TLB misses, which are not deferred.  The
992 	 * kernel MUST NOT depend on a particular setting of these bits (in other words,
993 	 * the kernel must have recovery code for all speculative accesses).  Turn on
994 	 * dcr.lc as per recommendation by the architecture team.  Most IA-32 apps
995 	 * shouldn't be affected by this (moral: keep your ia32 locks aligned and you'll
996 	 * be fine).
997 	 */
998 	ia64_setreg(_IA64_REG_CR_DCR,  (  IA64_DCR_DP | IA64_DCR_DK | IA64_DCR_DX | IA64_DCR_DR
999 					| IA64_DCR_DA | IA64_DCR_DD | IA64_DCR_LC));
1000 	mmgrab(&init_mm);
1001 	current->active_mm = &init_mm;
1002 	BUG_ON(current->mm);
1003 
1004 	ia64_mmu_init(ia64_imva(cpu_data));
1005 	ia64_mca_cpu_init(ia64_imva(cpu_data));
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 }
1054 
1055 void __init
1056 check_bugs (void)
1057 {
1058 	ia64_patch_mckinley_e9((unsigned long) __start___mckinley_e9_bundles,
1059 			       (unsigned long) __end___mckinley_e9_bundles);
1060 }
1061 
1062 static int __init run_dmi_scan(void)
1063 {
1064 	dmi_scan_machine();
1065 	dmi_memdev_walk();
1066 	dmi_set_dump_stack_arch_desc();
1067 	return 0;
1068 }
1069 core_initcall(run_dmi_scan);
1070